projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md/r5cache: fix spelling mistake on "recoverying"
[linux-block.git] / drivers / md / raid5.c
CommitLineData
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>
a9add5d9
N		 58#include <trace/events/block.h>
59
43b2e5d8 60#include "md.h"
bff61975 61#include "raid5.h"
54071b38 62#include "raid0.h"
ef740c37 63#include "bitmap.h"
72626685 64
851c30c9
SL		 65#define cpu_to_group(cpu) cpu_to_node(cpu)
66#define ANY_GROUP NUMA_NO_NODE
67
8e0e99ba
N		 68static bool devices_handle_discard_safely = false;
69module_param(devices_handle_discard_safely, bool, 0644);
70MODULE_PARM_DESC(devices_handle_discard_safely,
71 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
851c30c9 72static struct workqueue_struct *raid5_wq;
1da177e4 73
d1688a6d 74static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect)
db298e19
N		 75{
76 int hash = (sect >> STRIPE_SHIFT) & HASH_MASK;
77 return &conf->stripe_hashtbl[hash];
78}
1da177e4 79
566c09c5
SL		 80static inline int stripe_hash_locks_hash(sector_t sect)
81{
82 return (sect >> STRIPE_SHIFT) & STRIPE_HASH_LOCKS_MASK;
83}
84
85static inline void lock_device_hash_lock(struct r5conf *conf, int hash)
86{
87 spin_lock_irq(conf->hash_locks + hash);
88 spin_lock(&conf->device_lock);
89}
90
91static inline void unlock_device_hash_lock(struct r5conf *conf, int hash)
92{
93 spin_unlock(&conf->device_lock);
94 spin_unlock_irq(conf->hash_locks + hash);
95}
96
97static inline void lock_all_device_hash_locks_irq(struct r5conf *conf)
98{
99 int i;
100 local_irq_disable();
101 spin_lock(conf->hash_locks);
102 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
103 spin_lock_nest_lock(conf->hash_locks + i, conf->hash_locks);
104 spin_lock(&conf->device_lock);
105}
106
107static inline void unlock_all_device_hash_locks_irq(struct r5conf *conf)
108{
109 int i;
110 spin_unlock(&conf->device_lock);
111 for (i = NR_STRIPE_HASH_LOCKS; i; i--)
112 spin_unlock(conf->hash_locks + i - 1);
113 local_irq_enable();
114}
115
d0dabf7e
N		 116/* Find first data disk in a raid6 stripe */
117static inline int raid6_d0(struct stripe_head *sh)
118{
67cc2b81
N		 119 if (sh->ddf_layout)
120 /* ddf always start from first device */
121 return 0;
122 /* md starts just after Q block */
d0dabf7e
N		 123 if (sh->qd_idx == sh->disks - 1)
124 return 0;
125 else
126 return sh->qd_idx + 1;
127}
16a53ecc
N		 128static inline int raid6_next_disk(int disk, int raid_disks)
129{
130 disk++;
131 return (disk < raid_disks) ? disk : 0;
132}
a4456856 133
d0dabf7e
N		 134/* When walking through the disks in a raid5, starting at raid6_d0,
135 * We need to map each disk to a 'slot', where the data disks are slot
136 * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk
137 * is raid_disks-1. This help does that mapping.
138 */
67cc2b81
N		 139static int raid6_idx_to_slot(int idx, struct stripe_head *sh,
140 int *count, int syndrome_disks)
d0dabf7e 141{
6629542e 142 int slot = *count;
67cc2b81 143
e4424fee 144 if (sh->ddf_layout)
6629542e 145 (*count)++;
d0dabf7e 146 if (idx == sh->pd_idx)
67cc2b81 147 return syndrome_disks;
d0dabf7e 148 if (idx == sh->qd_idx)
67cc2b81 149 return syndrome_disks + 1;
e4424fee 150 if (!sh->ddf_layout)
6629542e 151 (*count)++;
d0dabf7e
N		 152 return slot;
153}
154
34a6f80e 155static void return_io(struct bio_list *return_bi)
a4456856 156{
34a6f80e
N		 157 struct bio *bi;
158 while ((bi = bio_list_pop(return_bi)) != NULL) {
4f024f37 159 bi->bi_iter.bi_size = 0;
0a82a8d1
LT		 160 trace_block_bio_complete(bdev_get_queue(bi->bi_bdev),
161 bi, 0);
4246a0b6 162 bio_endio(bi);
a4456856
DW		 163 }
164}
165
d1688a6d 166static void print_raid5_conf (struct r5conf *conf);
1da177e4 167
600aa109
DW		 168static int stripe_operations_active(struct stripe_head *sh)
169{
170 return sh->check_state || sh->reconstruct_state ||
171 test_bit(STRIPE_BIOFILL_RUN, &sh->state) ||
172 test_bit(STRIPE_COMPUTE_RUN, &sh->state);
173}
174
851c30c9
SL		 175static void raid5_wakeup_stripe_thread(struct stripe_head *sh)
176{
177 struct r5conf *conf = sh->raid_conf;
178 struct r5worker_group *group;
bfc90cb0 179 int thread_cnt;
851c30c9
SL		 180 int i, cpu = sh->cpu;
181
182 if (!cpu_online(cpu)) {
183 cpu = cpumask_any(cpu_online_mask);
184 sh->cpu = cpu;
185 }
186
187 if (list_empty(&sh->lru)) {
188 struct r5worker_group *group;
189 group = conf->worker_groups + cpu_to_group(cpu);
190 list_add_tail(&sh->lru, &group->handle_list);
bfc90cb0
SL		 191 group->stripes_cnt++;
192 sh->group = group;
851c30c9
SL		 193 }
194
195 if (conf->worker_cnt_per_group == 0) {
196 md_wakeup_thread(conf->mddev->thread);
197 return;
198 }
199
200 group = conf->worker_groups + cpu_to_group(sh->cpu);
201
bfc90cb0
SL		 202 group->workers[0].working = true;
203 /* at least one worker should run to avoid race */
204 queue_work_on(sh->cpu, raid5_wq, &group->workers[0].work);
205
206 thread_cnt = group->stripes_cnt / MAX_STRIPE_BATCH - 1;
207 /* wakeup more workers */
208 for (i = 1; i < conf->worker_cnt_per_group && thread_cnt > 0; i++) {
209 if (group->workers[i].working == false) {
210 group->workers[i].working = true;
211 queue_work_on(sh->cpu, raid5_wq,
212 &group->workers[i].work);
213 thread_cnt--;
214 }
215 }
851c30c9
SL		 216}
217
566c09c5
SL		 218static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh,
219 struct list_head *temp_inactive_list)
1da177e4 220{
1e6d690b
SL		 221 int i;
222 int injournal = 0; /* number of date pages with R5_InJournal */
223
4eb788df
SL		 224 BUG_ON(!list_empty(&sh->lru));
225 BUG_ON(atomic_read(&conf->active_stripes)==0);
1e6d690b
SL		 226
227 if (r5c_is_writeback(conf->log))
228 for (i = sh->disks; i--; )
229 if (test_bit(R5_InJournal, &sh->dev[i].flags))
230 injournal++;
a39f7afd
SL		 231 /*
232 * When quiesce in r5c write back, set STRIPE_HANDLE for stripes with
233 * data in journal, so they are not released to cached lists
234 */
235 if (conf->quiesce && r5c_is_writeback(conf->log) &&
236 !test_bit(STRIPE_HANDLE, &sh->state) && injournal != 0) {
237 if (test_bit(STRIPE_R5C_CACHING, &sh->state))
238 r5c_make_stripe_write_out(sh);
239 set_bit(STRIPE_HANDLE, &sh->state);
240 }
1e6d690b 241
4eb788df
SL		 242 if (test_bit(STRIPE_HANDLE, &sh->state)) {
243 if (test_bit(STRIPE_DELAYED, &sh->state) &&
ad3ab8b6 244 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4eb788df 245 list_add_tail(&sh->lru, &conf->delayed_list);
ad3ab8b6 246 else if (test_bit(STRIPE_BIT_DELAY, &sh->state) &&
4eb788df
SL		 247 sh->bm_seq - conf->seq_write > 0)
248 list_add_tail(&sh->lru, &conf->bitmap_list);
249 else {
250 clear_bit(STRIPE_DELAYED, &sh->state);
251 clear_bit(STRIPE_BIT_DELAY, &sh->state);
851c30c9
SL		 252 if (conf->worker_cnt_per_group == 0) {
253 list_add_tail(&sh->lru, &conf->handle_list);
254 } else {
255 raid5_wakeup_stripe_thread(sh);
256 return;
257 }
4eb788df
SL		 258 }
259 md_wakeup_thread(conf->mddev->thread);
260 } else {
261 BUG_ON(stripe_operations_active(sh));
262 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
263 if (atomic_dec_return(&conf->preread_active_stripes)
264 < IO_THRESHOLD)
265 md_wakeup_thread(conf->mddev->thread);
266 atomic_dec(&conf->active_stripes);
1e6d690b
SL		 267 if (!test_bit(STRIPE_EXPANDING, &sh->state)) {
268 if (!r5c_is_writeback(conf->log))
269 list_add_tail(&sh->lru, temp_inactive_list);
270 else {
271 WARN_ON(test_bit(R5_InJournal, &sh->dev[sh->pd_idx].flags));
272 if (injournal == 0)
273 list_add_tail(&sh->lru, temp_inactive_list);
274 else if (injournal == conf->raid_disks - conf->max_degraded) {
275 /* full stripe */
276 if (!test_and_set_bit(STRIPE_R5C_FULL_STRIPE, &sh->state))
277 atomic_inc(&conf->r5c_cached_full_stripes);
278 if (test_and_clear_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state))
279 atomic_dec(&conf->r5c_cached_partial_stripes);
280 list_add_tail(&sh->lru, &conf->r5c_full_stripe_list);
a39f7afd 281 r5c_check_cached_full_stripe(conf);
1e6d690b
SL		 282 } else {
283 /* partial stripe */
284 if (!test_and_set_bit(STRIPE_R5C_PARTIAL_STRIPE,
285 &sh->state))
286 atomic_inc(&conf->r5c_cached_partial_stripes);
287 list_add_tail(&sh->lru, &conf->r5c_partial_stripe_list);
288 }
289 }
290 }
1da177e4
LT		 291 }
292}
d0dabf7e 293
566c09c5
SL		 294static void __release_stripe(struct r5conf *conf, struct stripe_head *sh,
295 struct list_head *temp_inactive_list)
4eb788df
SL		 296{
297 if (atomic_dec_and_test(&sh->count))
566c09c5
SL		 298 do_release_stripe(conf, sh, temp_inactive_list);
299}
300
301/*
302 * @hash could be NR_STRIPE_HASH_LOCKS, then we have a list of inactive_list
303 *
304 * Be careful: Only one task can add/delete stripes from temp_inactive_list at
305 * given time. Adding stripes only takes device lock, while deleting stripes
306 * only takes hash lock.
307 */
308static void release_inactive_stripe_list(struct r5conf *conf,
309 struct list_head *temp_inactive_list,
310 int hash)
311{
312 int size;
6ab2a4b8 313 bool do_wakeup = false;
566c09c5
SL		 314 unsigned long flags;
315
316 if (hash == NR_STRIPE_HASH_LOCKS) {
317 size = NR_STRIPE_HASH_LOCKS;
318 hash = NR_STRIPE_HASH_LOCKS - 1;
319 } else
320 size = 1;
321 while (size) {
322 struct list_head *list = &temp_inactive_list[size - 1];
323
324 /*
6d036f7d 325 * We don't hold any lock here yet, raid5_get_active_stripe() might
566c09c5
SL		 326 * remove stripes from the list
327 */
328 if (!list_empty_careful(list)) {
329 spin_lock_irqsave(conf->hash_locks + hash, flags);
4bda556a
SL		 330 if (list_empty(conf->inactive_list + hash) &&
331 !list_empty(list))
332 atomic_dec(&conf->empty_inactive_list_nr);
566c09c5 333 list_splice_tail_init(list, conf->inactive_list + hash);
6ab2a4b8 334 do_wakeup = true;
566c09c5
SL		 335 spin_unlock_irqrestore(conf->hash_locks + hash, flags);
336 }
337 size--;
338 hash--;
339 }
340
341 if (do_wakeup) {
6ab2a4b8 342 wake_up(&conf->wait_for_stripe);
b1b46486
YL		 343 if (atomic_read(&conf->active_stripes) == 0)
344 wake_up(&conf->wait_for_quiescent);
566c09c5
SL		 345 if (conf->retry_read_aligned)
346 md_wakeup_thread(conf->mddev->thread);
347 }
4eb788df
SL		 348}
349
773ca82f 350/* should hold conf->device_lock already */
566c09c5
SL		 351static int release_stripe_list(struct r5conf *conf,
352 struct list_head *temp_inactive_list)
773ca82f
SL		 353{
354 struct stripe_head *sh;
355 int count = 0;
356 struct llist_node *head;
357
358 head = llist_del_all(&conf->released_stripes);
d265d9dc 359 head = llist_reverse_order(head);
773ca82f 360 while (head) {
566c09c5
SL		 361 int hash;
362
773ca82f
SL		 363 sh = llist_entry(head, struct stripe_head, release_list);
364 head = llist_next(head);
365 /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */
366 smp_mb();
367 clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state);
368 /*
369 * Don't worry the bit is set here, because if the bit is set
370 * again, the count is always > 1. This is true for
371 * STRIPE_ON_UNPLUG_LIST bit too.
372 */
566c09c5
SL		 373 hash = sh->hash_lock_index;
374 __release_stripe(conf, sh, &temp_inactive_list[hash]);
773ca82f
SL		 375 count++;
376 }
377
378 return count;
379}
380
6d036f7d 381void raid5_release_stripe(struct stripe_head *sh)
1da177e4 382{
d1688a6d 383 struct r5conf *conf = sh->raid_conf;
1da177e4 384 unsigned long flags;
566c09c5
SL		 385 struct list_head list;
386 int hash;
773ca82f 387 bool wakeup;
16a53ecc 388
cf170f3f
ES		 389 /* Avoid release_list until the last reference.
390 */
391 if (atomic_add_unless(&sh->count, -1, 1))
392 return;
393
ad4068de 394 if (unlikely(!conf->mddev->thread) ||
395 test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state))
773ca82f
SL		 396 goto slow_path;
397 wakeup = llist_add(&sh->release_list, &conf->released_stripes);
398 if (wakeup)
399 md_wakeup_thread(conf->mddev->thread);
400 return;
401slow_path:
4eb788df 402 local_irq_save(flags);
773ca82f 403 /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */
4eb788df 404 if (atomic_dec_and_lock(&sh->count, &conf->device_lock)) {
566c09c5
SL		 405 INIT_LIST_HEAD(&list);
406 hash = sh->hash_lock_index;
407 do_release_stripe(conf, sh, &list);
4eb788df 408 spin_unlock(&conf->device_lock);
566c09c5 409 release_inactive_stripe_list(conf, &list, hash);
4eb788df
SL		 410 }
411 local_irq_restore(flags);
1da177e4
LT		 412}
413
fccddba0 414static inline void remove_hash(struct stripe_head *sh)
1da177e4 415{
45b4233c
DW		 416 pr_debug("remove_hash(), stripe %llu\n",
417 (unsigned long long)sh->sector);
1da177e4 418
fccddba0 419 hlist_del_init(&sh->hash);
1da177e4
LT		 420}
421
d1688a6d 422static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh)
1da177e4 423{
fccddba0 424 struct hlist_head *hp = stripe_hash(conf, sh->sector);
1da177e4 425
45b4233c
DW		 426 pr_debug("insert_hash(), stripe %llu\n",
427 (unsigned long long)sh->sector);
1da177e4 428
fccddba0 429 hlist_add_head(&sh->hash, hp);
1da177e4
LT		 430}
431
1da177e4 432/* find an idle stripe, make sure it is unhashed, and return it. */
566c09c5 433static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash)
1da177e4
LT		 434{
435 struct stripe_head *sh = NULL;
436 struct list_head *first;
437
566c09c5 438 if (list_empty(conf->inactive_list + hash))
1da177e4 439 goto out;
566c09c5 440 first = (conf->inactive_list + hash)->next;
1da177e4
LT		 441 sh = list_entry(first, struct stripe_head, lru);
442 list_del_init(first);
443 remove_hash(sh);
444 atomic_inc(&conf->active_stripes);
566c09c5 445 BUG_ON(hash != sh->hash_lock_index);
4bda556a
SL		 446 if (list_empty(conf->inactive_list + hash))
447 atomic_inc(&conf->empty_inactive_list_nr);
1da177e4
LT		 448out:
449 return sh;
450}
451
e4e11e38 452static void shrink_buffers(struct stripe_head *sh)
1da177e4
LT		 453{
454 struct page *p;
455 int i;
e4e11e38 456 int num = sh->raid_conf->pool_size;
1da177e4 457
e4e11e38 458 for (i = 0; i < num ; i++) {
d592a996 459 WARN_ON(sh->dev[i].page != sh->dev[i].orig_page);
1da177e4
LT		 460 p = sh->dev[i].page;
461 if (!p)
462 continue;
463 sh->dev[i].page = NULL;
2d1f3b5d 464 put_page(p);
1da177e4
LT		 465 }
466}
467
a9683a79 468static int grow_buffers(struct stripe_head *sh, gfp_t gfp)
1da177e4
LT		 469{
470 int i;
e4e11e38 471 int num = sh->raid_conf->pool_size;
1da177e4 472
e4e11e38 473 for (i = 0; i < num; i++) {
1da177e4
LT		 474 struct page *page;
475
a9683a79 476 if (!(page = alloc_page(gfp))) {
1da177e4
LT		 477 return 1;
478 }
479 sh->dev[i].page = page;
d592a996 480 sh->dev[i].orig_page = page;
1da177e4
LT		 481 }
482 return 0;
483}
484
784052ec 485static void raid5_build_block(struct stripe_head *sh, int i, int previous);
d1688a6d 486static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
911d4ee8 487 struct stripe_head *sh);
1da177e4 488
b5663ba4 489static void init_stripe(struct stripe_head *sh, sector_t sector, int previous)
1da177e4 490{
d1688a6d 491 struct r5conf *conf = sh->raid_conf;
566c09c5 492 int i, seq;
1da177e4 493
78bafebd
ES		 494 BUG_ON(atomic_read(&sh->count) != 0);
495 BUG_ON(test_bit(STRIPE_HANDLE, &sh->state));
600aa109 496 BUG_ON(stripe_operations_active(sh));
59fc630b 497 BUG_ON(sh->batch_head);
d84e0f10 498
45b4233c 499 pr_debug("init_stripe called, stripe %llu\n",
b8e6a15a 500 (unsigned long long)sector);
566c09c5
SL		 501retry:
502 seq = read_seqcount_begin(&conf->gen_lock);
86b42c71 503 sh->generation = conf->generation - previous;
b5663ba4 504 sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks;
1da177e4 505 sh->sector = sector;
911d4ee8 506 stripe_set_idx(sector, conf, previous, sh);
1da177e4
LT		 507 sh->state = 0;
508
7ecaa1e6 509 for (i = sh->disks; i--; ) {
1da177e4
LT		 510 struct r5dev *dev = &sh->dev[i];
511
d84e0f10 512 if (dev->toread || dev->read || dev->towrite || dev->written ||
1da177e4 513 test_bit(R5_LOCKED, &dev->flags)) {
cc6167b4 514 pr_err("sector=%llx i=%d %p %p %p %p %d\n",
1da177e4 515 (unsigned long long)sh->sector, i, dev->toread,
d84e0f10 516 dev->read, dev->towrite, dev->written,
1da177e4 517 test_bit(R5_LOCKED, &dev->flags));
8cfa7b0f 518 WARN_ON(1);
1da177e4
LT		 519 }
520 dev->flags = 0;
784052ec 521 raid5_build_block(sh, i, previous);
1da177e4 522 }
566c09c5
SL		 523 if (read_seqcount_retry(&conf->gen_lock, seq))
524 goto retry;
7a87f434 525 sh->overwrite_disks = 0;
1da177e4 526 insert_hash(conf, sh);
851c30c9 527 sh->cpu = smp_processor_id();
da41ba65 528 set_bit(STRIPE_BATCH_READY, &sh->state);
1da177e4
LT		 529}
530
d1688a6d 531static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector,
86b42c71 532 short generation)
1da177e4
LT		 533{
534 struct stripe_head *sh;
535
45b4233c 536 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
b67bfe0d 537 hlist_for_each_entry(sh, stripe_hash(conf, sector), hash)
86b42c71 538 if (sh->sector == sector && sh->generation == generation)
1da177e4 539 return sh;
45b4233c 540 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
1da177e4
LT		 541 return NULL;
542}
543
674806d6
N		 544/*
545 * Need to check if array has failed when deciding whether to:
546 * - start an array
547 * - remove non-faulty devices
548 * - add a spare
549 * - allow a reshape
550 * This determination is simple when no reshape is happening.
551 * However if there is a reshape, we need to carefully check
552 * both the before and after sections.
553 * This is because some failed devices may only affect one
554 * of the two sections, and some non-in_sync devices may
555 * be insync in the section most affected by failed devices.
556 */
908f4fbd 557static int calc_degraded(struct r5conf *conf)
674806d6 558{
908f4fbd 559 int degraded, degraded2;
674806d6 560 int i;
674806d6
N		 561
562 rcu_read_lock();
563 degraded = 0;
564 for (i = 0; i < conf->previous_raid_disks; i++) {
3cb03002 565 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
e5c86471
N		 566 if (rdev && test_bit(Faulty, &rdev->flags))
567 rdev = rcu_dereference(conf->disks[i].replacement);
674806d6
N		 568 if (!rdev || test_bit(Faulty, &rdev->flags))
569 degraded++;
570 else if (test_bit(In_sync, &rdev->flags))
571 ;
572 else
573 /* not in-sync or faulty.
574 * If the reshape increases the number of devices,
575 * this is being recovered by the reshape, so
576 * this 'previous' section is not in_sync.
577 * If the number of devices is being reduced however,
578 * the device can only be part of the array if
579 * we are reverting a reshape, so this section will
580 * be in-sync.
581 */
582 if (conf->raid_disks >= conf->previous_raid_disks)
583 degraded++;
584 }
585 rcu_read_unlock();
908f4fbd
N		 586 if (conf->raid_disks == conf->previous_raid_disks)
587 return degraded;
674806d6 588 rcu_read_lock();
908f4fbd 589 degraded2 = 0;
674806d6 590 for (i = 0; i < conf->raid_disks; i++) {
3cb03002 591 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
e5c86471
N		 592 if (rdev && test_bit(Faulty, &rdev->flags))
593 rdev = rcu_dereference(conf->disks[i].replacement);
674806d6 594 if (!rdev || test_bit(Faulty, &rdev->flags))
908f4fbd 595 degraded2++;
674806d6
N		 596 else if (test_bit(In_sync, &rdev->flags))
597 ;
598 else
599 /* not in-sync or faulty.
600 * If reshape increases the number of devices, this
601 * section has already been recovered, else it
602 * almost certainly hasn't.
603 */
604 if (conf->raid_disks <= conf->previous_raid_disks)
908f4fbd 605 degraded2++;
674806d6
N		 606 }
607 rcu_read_unlock();
908f4fbd
N		 608 if (degraded2 > degraded)
609 return degraded2;
610 return degraded;
611}
612
613static int has_failed(struct r5conf *conf)
614{
615 int degraded;
616
617 if (conf->mddev->reshape_position == MaxSector)
618 return conf->mddev->degraded > conf->max_degraded;
619
620 degraded = calc_degraded(conf);
674806d6
N		 621 if (degraded > conf->max_degraded)
622 return 1;
623 return 0;
624}
625
6d036f7d
SL		 626struct stripe_head *
627raid5_get_active_stripe(struct r5conf *conf, sector_t sector,
628 int previous, int noblock, int noquiesce)
1da177e4
LT		 629{
630 struct stripe_head *sh;
566c09c5 631 int hash = stripe_hash_locks_hash(sector);
ff00d3b4 632 int inc_empty_inactive_list_flag;
1da177e4 633
45b4233c 634 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector);
1da177e4 635
566c09c5 636 spin_lock_irq(conf->hash_locks + hash);
1da177e4
LT		 637
638 do {
b1b46486 639 wait_event_lock_irq(conf->wait_for_quiescent,
a8c906ca 640 conf->quiesce == 0 || noquiesce,
566c09c5 641 *(conf->hash_locks + hash));
86b42c71 642 sh = __find_stripe(conf, sector, conf->generation - previous);
1da177e4 643 if (!sh) {
edbe83ab 644 if (!test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) {
566c09c5 645 sh = get_free_stripe(conf, hash);
713bc5c2
SL		 646 if (!sh && !test_bit(R5_DID_ALLOC,
647 &conf->cache_state))
edbe83ab
N		 648 set_bit(R5_ALLOC_MORE,
649 &conf->cache_state);
650 }
1da177e4
LT		 651 if (noblock && sh == NULL)
652 break;
a39f7afd
SL		 653
654 r5c_check_stripe_cache_usage(conf);
1da177e4 655 if (!sh) {
5423399a
N		 656 set_bit(R5_INACTIVE_BLOCKED,
657 &conf->cache_state);
a39f7afd 658 r5l_wake_reclaim(conf->log, 0);
6ab2a4b8
SL		 659 wait_event_lock_irq(
660 conf->wait_for_stripe,
566c09c5
SL		 661 !list_empty(conf->inactive_list + hash) &&
662 (atomic_read(&conf->active_stripes)
663 < (conf->max_nr_stripes * 3 / 4)
5423399a
N		 664 || !test_bit(R5_INACTIVE_BLOCKED,
665 &conf->cache_state)),
6ab2a4b8 666 *(conf->hash_locks + hash));
5423399a
N		 667 clear_bit(R5_INACTIVE_BLOCKED,
668 &conf->cache_state);
7da9d450 669 } else {
b5663ba4 670 init_stripe(sh, sector, previous);
7da9d450
N		 671 atomic_inc(&sh->count);
672 }
e240c183 673 } else if (!atomic_inc_not_zero(&sh->count)) {
6d183de4 674 spin_lock(&conf->device_lock);
e240c183 675 if (!atomic_read(&sh->count)) {
1da177e4
LT		 676 if (!test_bit(STRIPE_HANDLE, &sh->state))
677 atomic_inc(&conf->active_stripes);
5af9bef7
N		 678 BUG_ON(list_empty(&sh->lru) &&
679 !test_bit(STRIPE_EXPANDING, &sh->state));
ff00d3b4
ZL		 680 inc_empty_inactive_list_flag = 0;
681 if (!list_empty(conf->inactive_list + hash))
682 inc_empty_inactive_list_flag = 1;
16a53ecc 683 list_del_init(&sh->lru);
ff00d3b4
ZL		 684 if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag)
685 atomic_inc(&conf->empty_inactive_list_nr);
bfc90cb0
SL		 686 if (sh->group) {
687 sh->group->stripes_cnt--;
688 sh->group = NULL;
689 }
1da177e4 690 }
7da9d450 691 atomic_inc(&sh->count);
6d183de4 692 spin_unlock(&conf->device_lock);
1da177e4
LT		 693 }
694 } while (sh == NULL);
695
566c09c5 696 spin_unlock_irq(conf->hash_locks + hash);
1da177e4
LT		 697 return sh;
698}
699
7a87f434 700static bool is_full_stripe_write(struct stripe_head *sh)
701{
702 BUG_ON(sh->overwrite_disks > (sh->disks - sh->raid_conf->max_degraded));
703 return sh->overwrite_disks == (sh->disks - sh->raid_conf->max_degraded);
704}
705
59fc630b 706static void lock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2)
707{
708 local_irq_disable();
709 if (sh1 > sh2) {
710 spin_lock(&sh2->stripe_lock);
711 spin_lock_nested(&sh1->stripe_lock, 1);
712 } else {
713 spin_lock(&sh1->stripe_lock);
714 spin_lock_nested(&sh2->stripe_lock, 1);
715 }
716}
717
718static void unlock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2)
719{
720 spin_unlock(&sh1->stripe_lock);
721 spin_unlock(&sh2->stripe_lock);
722 local_irq_enable();
723}
724
725/* Only freshly new full stripe normal write stripe can be added to a batch list */
726static bool stripe_can_batch(struct stripe_head *sh)
727{
9c3e333d
SL		 728 struct r5conf *conf = sh->raid_conf;
729
730 if (conf->log)
731 return false;
59fc630b 732 return test_bit(STRIPE_BATCH_READY, &sh->state) &&
d0852df5 733 !test_bit(STRIPE_BITMAP_PENDING, &sh->state) &&
59fc630b 734 is_full_stripe_write(sh);
735}
736
737/* we only do back search */
738static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh)
739{
740 struct stripe_head *head;
741 sector_t head_sector, tmp_sec;
742 int hash;
743 int dd_idx;
ff00d3b4 744 int inc_empty_inactive_list_flag;
59fc630b 745
59fc630b 746 /* Don't cross chunks, so stripe pd_idx/qd_idx is the same */
747 tmp_sec = sh->sector;
748 if (!sector_div(tmp_sec, conf->chunk_sectors))
749 return;
750 head_sector = sh->sector - STRIPE_SECTORS;
751
752 hash = stripe_hash_locks_hash(head_sector);
753 spin_lock_irq(conf->hash_locks + hash);
754 head = __find_stripe(conf, head_sector, conf->generation);
755 if (head && !atomic_inc_not_zero(&head->count)) {
756 spin_lock(&conf->device_lock);
757 if (!atomic_read(&head->count)) {
758 if (!test_bit(STRIPE_HANDLE, &head->state))
759 atomic_inc(&conf->active_stripes);
760 BUG_ON(list_empty(&head->lru) &&
761 !test_bit(STRIPE_EXPANDING, &head->state));
ff00d3b4
ZL		 762 inc_empty_inactive_list_flag = 0;
763 if (!list_empty(conf->inactive_list + hash))
764 inc_empty_inactive_list_flag = 1;
59fc630b 765 list_del_init(&head->lru);
ff00d3b4
ZL		 766 if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag)
767 atomic_inc(&conf->empty_inactive_list_nr);
59fc630b 768 if (head->group) {
769 head->group->stripes_cnt--;
770 head->group = NULL;
771 }
772 }
773 atomic_inc(&head->count);
774 spin_unlock(&conf->device_lock);
775 }
776 spin_unlock_irq(conf->hash_locks + hash);
777
778 if (!head)
779 return;
780 if (!stripe_can_batch(head))
781 goto out;
782
783 lock_two_stripes(head, sh);
784 /* clear_batch_ready clear the flag */
785 if (!stripe_can_batch(head) || !stripe_can_batch(sh))
786 goto unlock_out;
787
788 if (sh->batch_head)
789 goto unlock_out;
790
791 dd_idx = 0;
792 while (dd_idx == sh->pd_idx || dd_idx == sh->qd_idx)
793 dd_idx++;
1eff9d32 794 if (head->dev[dd_idx].towrite->bi_opf != sh->dev[dd_idx].towrite->bi_opf ||
796a5cf0 795 bio_op(head->dev[dd_idx].towrite) != bio_op(sh->dev[dd_idx].towrite))
59fc630b 796 goto unlock_out;
797
798 if (head->batch_head) {
799 spin_lock(&head->batch_head->batch_lock);
800 /* This batch list is already running */
801 if (!stripe_can_batch(head)) {
802 spin_unlock(&head->batch_head->batch_lock);
803 goto unlock_out;
804 }
805
806 /*
807 * at this point, head's BATCH_READY could be cleared, but we
808 * can still add the stripe to batch list
809 */
810 list_add(&sh->batch_list, &head->batch_list);
811 spin_unlock(&head->batch_head->batch_lock);
812
813 sh->batch_head = head->batch_head;
814 } else {
815 head->batch_head = head;
816 sh->batch_head = head->batch_head;
817 spin_lock(&head->batch_lock);
818 list_add_tail(&sh->batch_list, &head->batch_list);
819 spin_unlock(&head->batch_lock);
820 }
821
822 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
823 if (atomic_dec_return(&conf->preread_active_stripes)
824 < IO_THRESHOLD)
825 md_wakeup_thread(conf->mddev->thread);
826
2b6b2457
N		 827 if (test_and_clear_bit(STRIPE_BIT_DELAY, &sh->state)) {
828 int seq = sh->bm_seq;
829 if (test_bit(STRIPE_BIT_DELAY, &sh->batch_head->state) &&
830 sh->batch_head->bm_seq > seq)
831 seq = sh->batch_head->bm_seq;
832 set_bit(STRIPE_BIT_DELAY, &sh->batch_head->state);
833 sh->batch_head->bm_seq = seq;
834 }
835
59fc630b 836 atomic_inc(&sh->count);
837unlock_out:
838 unlock_two_stripes(head, sh);
839out:
6d036f7d 840 raid5_release_stripe(head);
59fc630b 841}
842
05616be5
N		 843/* Determine if 'data_offset' or 'new_data_offset' should be used
844 * in this stripe_head.
845 */
846static int use_new_offset(struct r5conf *conf, struct stripe_head *sh)
847{
848 sector_t progress = conf->reshape_progress;
849 /* Need a memory barrier to make sure we see the value
850 * of conf->generation, or ->data_offset that was set before
851 * reshape_progress was updated.
852 */
853 smp_rmb();
854 if (progress == MaxSector)
855 return 0;
856 if (sh->generation == conf->generation - 1)
857 return 0;
858 /* We are in a reshape, and this is a new-generation stripe,
859 * so use new_data_offset.
860 */
861 return 1;
862}
863
6712ecf8 864static void
4246a0b6 865raid5_end_read_request(struct bio *bi);
6712ecf8 866static void
4246a0b6 867raid5_end_write_request(struct bio *bi);
91c00924 868
c4e5ac0a 869static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
91c00924 870{
d1688a6d 871 struct r5conf *conf = sh->raid_conf;
91c00924 872 int i, disks = sh->disks;
59fc630b 873 struct stripe_head *head_sh = sh;
91c00924
DW		 874
875 might_sleep();
876
1e6d690b
SL		 877 if (!test_bit(STRIPE_R5C_CACHING, &sh->state)) {
878 /* writing out phase */
d7bd398e
SL		 879 if (s->waiting_extra_page)
880 return;
1e6d690b
SL		 881 if (r5l_write_stripe(conf->log, sh) == 0)
882 return;
883 } else { /* caching phase */
884 if (test_bit(STRIPE_LOG_TRAPPED, &sh->state)) {
885 r5c_cache_data(conf->log, sh, s);
886 return;
887 }
888 }
889
91c00924 890 for (i = disks; i--; ) {
796a5cf0 891 int op, op_flags = 0;
9a3e1101 892 int replace_only = 0;
977df362
N		 893 struct bio *bi, *rbi;
894 struct md_rdev *rdev, *rrdev = NULL;
59fc630b 895
896 sh = head_sh;
e9c7469b 897 if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) {
796a5cf0 898 op = REQ_OP_WRITE;
e9c7469b 899 if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags))
70fd7614 900 op_flags = REQ_FUA;
9e444768 901 if (test_bit(R5_Discard, &sh->dev[i].flags))
796a5cf0 902 op = REQ_OP_DISCARD;
e9c7469b 903 } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags))
796a5cf0 904 op = REQ_OP_READ;
9a3e1101
N		 905 else if (test_and_clear_bit(R5_WantReplace,
906 &sh->dev[i].flags)) {
796a5cf0 907 op = REQ_OP_WRITE;
9a3e1101
N		 908 replace_only = 1;
909 } else
91c00924 910 continue;
bc0934f0 911 if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags))
796a5cf0 912 op_flags |= REQ_SYNC;
91c00924 913
59fc630b 914again:
91c00924 915 bi = &sh->dev[i].req;
977df362 916 rbi = &sh->dev[i].rreq; /* For writing to replacement */
91c00924 917
91c00924 918 rcu_read_lock();
9a3e1101 919 rrdev = rcu_dereference(conf->disks[i].replacement);
dd054fce
N		 920 smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */
921 rdev = rcu_dereference(conf->disks[i].rdev);
922 if (!rdev) {
923 rdev = rrdev;
924 rrdev = NULL;
925 }
796a5cf0 926 if (op_is_write(op)) {
9a3e1101
N		 927 if (replace_only)
928 rdev = NULL;
dd054fce
N		 929 if (rdev == rrdev)
930 /* We raced and saw duplicates */
931 rrdev = NULL;
9a3e1101 932 } else {
59fc630b 933 if (test_bit(R5_ReadRepl, &head_sh->dev[i].flags) && rrdev)
9a3e1101
N		 934 rdev = rrdev;
935 rrdev = NULL;
936 }
977df362 937
91c00924
DW		 938 if (rdev && test_bit(Faulty, &rdev->flags))
939 rdev = NULL;
940 if (rdev)
941 atomic_inc(&rdev->nr_pending);
977df362
N		 942 if (rrdev && test_bit(Faulty, &rrdev->flags))
943 rrdev = NULL;
944 if (rrdev)
945 atomic_inc(&rrdev->nr_pending);
91c00924
DW		 946 rcu_read_unlock();
947
73e92e51 948 /* We have already checked bad blocks for reads. Now
977df362
N		 949 * need to check for writes. We never accept write errors
950 * on the replacement, so we don't to check rrdev.
73e92e51 951 */
796a5cf0 952 while (op_is_write(op) && rdev &&
73e92e51
N		 953 test_bit(WriteErrorSeen, &rdev->flags)) {
954 sector_t first_bad;
955 int bad_sectors;
956 int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
957 &first_bad, &bad_sectors);
958 if (!bad)
959 break;
960
961 if (bad < 0) {
962 set_bit(BlockedBadBlocks, &rdev->flags);
963 if (!conf->mddev->external &&
2953079c 964 conf->mddev->sb_flags) {
73e92e51
N		 965 /* It is very unlikely, but we might
966 * still need to write out the
967 * bad block log - better give it
968 * a chance*/
969 md_check_recovery(conf->mddev);
970 }
1850753d 971 /*
972 * Because md_wait_for_blocked_rdev
973 * will dec nr_pending, we must
974 * increment it first.
975 */
976 atomic_inc(&rdev->nr_pending);
73e92e51
N		 977 md_wait_for_blocked_rdev(rdev, conf->mddev);
978 } else {
979 /* Acknowledged bad block - skip the write */
980 rdev_dec_pending(rdev, conf->mddev);
981 rdev = NULL;
982 }
983 }
984
91c00924 985 if (rdev) {
9a3e1101
N		 986 if (s->syncing || s->expanding || s->expanded
987 || s->replacing)
91c00924
DW		 988 md_sync_acct(rdev->bdev, STRIPE_SECTORS);
989
2b7497f0
DW		 990 set_bit(STRIPE_IO_STARTED, &sh->state);
991
91c00924 992 bi->bi_bdev = rdev->bdev;
796a5cf0
MC		 993 bio_set_op_attrs(bi, op, op_flags);
994 bi->bi_end_io = op_is_write(op)
2f6db2a7
KO		 995 ? raid5_end_write_request
996 : raid5_end_read_request;
997 bi->bi_private = sh;
998
6296b960 999 pr_debug("%s: for %llu schedule op %d on disc %d\n",
e46b272b 1000 __func__, (unsigned long long)sh->sector,
1eff9d32 1001 bi->bi_opf, i);
91c00924 1002 atomic_inc(&sh->count);
59fc630b 1003 if (sh != head_sh)
1004 atomic_inc(&head_sh->count);
05616be5 1005 if (use_new_offset(conf, sh))
4f024f37 1006 bi->bi_iter.bi_sector = (sh->sector
05616be5
N		 1007 + rdev->new_data_offset);
1008 else
4f024f37 1009 bi->bi_iter.bi_sector = (sh->sector
05616be5 1010 + rdev->data_offset);
59fc630b 1011 if (test_bit(R5_ReadNoMerge, &head_sh->dev[i].flags))
1eff9d32 1012 bi->bi_opf |= REQ_NOMERGE;
3f9e7c14 1013
d592a996
SL		 1014 if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
1015 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
1016 sh->dev[i].vec.bv_page = sh->dev[i].page;
4997b72e 1017 bi->bi_vcnt = 1;
91c00924
DW		 1018 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
1019 bi->bi_io_vec[0].bv_offset = 0;
4f024f37 1020 bi->bi_iter.bi_size = STRIPE_SIZE;
37c61ff3
SL		 1021 /*
1022 * If this is discard request, set bi_vcnt 0. We don't
1023 * want to confuse SCSI because SCSI will replace payload
1024 */
796a5cf0 1025 if (op == REQ_OP_DISCARD)
37c61ff3 1026 bi->bi_vcnt = 0;
977df362
N		 1027 if (rrdev)
1028 set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags);
e3620a3a
JB		 1029
1030 if (conf->mddev->gendisk)
1031 trace_block_bio_remap(bdev_get_queue(bi->bi_bdev),
1032 bi, disk_devt(conf->mddev->gendisk),
1033 sh->dev[i].sector);
91c00924 1034 generic_make_request(bi);
977df362
N		 1035 }
1036 if (rrdev) {
9a3e1101
N		 1037 if (s->syncing || s->expanding || s->expanded
1038 || s->replacing)
977df362
N		 1039 md_sync_acct(rrdev->bdev, STRIPE_SECTORS);
1040
1041 set_bit(STRIPE_IO_STARTED, &sh->state);
1042
1043 rbi->bi_bdev = rrdev->bdev;
796a5cf0
MC		 1044 bio_set_op_attrs(rbi, op, op_flags);
1045 BUG_ON(!op_is_write(op));
2f6db2a7
KO		 1046 rbi->bi_end_io = raid5_end_write_request;
1047 rbi->bi_private = sh;
1048
6296b960 1049 pr_debug("%s: for %llu schedule op %d on "
977df362
N		 1050 "replacement disc %d\n",
1051 __func__, (unsigned long long)sh->sector,
1eff9d32 1052 rbi->bi_opf, i);
977df362 1053 atomic_inc(&sh->count);
59fc630b 1054 if (sh != head_sh)
1055 atomic_inc(&head_sh->count);
05616be5 1056 if (use_new_offset(conf, sh))
4f024f37 1057 rbi->bi_iter.bi_sector = (sh->sector
05616be5
N		 1058 + rrdev->new_data_offset);
1059 else
4f024f37 1060 rbi->bi_iter.bi_sector = (sh->sector
05616be5 1061 + rrdev->data_offset);
d592a996
SL		 1062 if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
1063 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
1064 sh->dev[i].rvec.bv_page = sh->dev[i].page;
4997b72e 1065 rbi->bi_vcnt = 1;
977df362
N		 1066 rbi->bi_io_vec[0].bv_len = STRIPE_SIZE;
1067 rbi->bi_io_vec[0].bv_offset = 0;
4f024f37 1068 rbi->bi_iter.bi_size = STRIPE_SIZE;
37c61ff3
SL		 1069 /*
1070 * If this is discard request, set bi_vcnt 0. We don't
1071 * want to confuse SCSI because SCSI will replace payload
1072 */
796a5cf0 1073 if (op == REQ_OP_DISCARD)
37c61ff3 1074 rbi->bi_vcnt = 0;
e3620a3a
JB		 1075 if (conf->mddev->gendisk)
1076 trace_block_bio_remap(bdev_get_queue(rbi->bi_bdev),
1077 rbi, disk_devt(conf->mddev->gendisk),
1078 sh->dev[i].sector);
977df362
N		 1079 generic_make_request(rbi);
1080 }
1081 if (!rdev && !rrdev) {
796a5cf0 1082 if (op_is_write(op))
91c00924 1083 set_bit(STRIPE_DEGRADED, &sh->state);
6296b960 1084 pr_debug("skip op %d on disc %d for sector %llu\n",
1eff9d32 1085 bi->bi_opf, i, (unsigned long long)sh->sector);
91c00924
DW		 1086 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1087 set_bit(STRIPE_HANDLE, &sh->state);
1088 }
59fc630b 1089
1090 if (!head_sh->batch_head)
1091 continue;
1092 sh = list_first_entry(&sh->batch_list, struct stripe_head,
1093 batch_list);
1094 if (sh != head_sh)
1095 goto again;
91c00924
DW		 1096 }
1097}
1098
1099static struct dma_async_tx_descriptor *
d592a996
SL		 1100async_copy_data(int frombio, struct bio *bio, struct page **page,
1101 sector_t sector, struct dma_async_tx_descriptor *tx,
1e6d690b 1102 struct stripe_head *sh, int no_skipcopy)
91c00924 1103{
7988613b
KO		 1104 struct bio_vec bvl;
1105 struct bvec_iter iter;
91c00924 1106 struct page *bio_page;
91c00924 1107 int page_offset;
a08abd8c 1108 struct async_submit_ctl submit;
0403e382 1109 enum async_tx_flags flags = 0;
91c00924 1110
4f024f37
KO		 1111 if (bio->bi_iter.bi_sector >= sector)
1112 page_offset = (signed)(bio->bi_iter.bi_sector - sector) * 512;
91c00924 1113 else
4f024f37 1114 page_offset = (signed)(sector - bio->bi_iter.bi_sector) * -512;
a08abd8c 1115
0403e382
DW		 1116 if (frombio)
1117 flags |= ASYNC_TX_FENCE;
1118 init_async_submit(&submit, flags, tx, NULL, NULL, NULL);
1119
7988613b
KO		 1120 bio_for_each_segment(bvl, bio, iter) {
1121 int len = bvl.bv_len;
91c00924
DW		 1122 int clen;
1123 int b_offset = 0;
1124
1125 if (page_offset < 0) {
1126 b_offset = -page_offset;
1127 page_offset += b_offset;
1128 len -= b_offset;
1129 }
1130
1131 if (len > 0 && page_offset + len > STRIPE_SIZE)
1132 clen = STRIPE_SIZE - page_offset;
1133 else
1134 clen = len;
1135
1136 if (clen > 0) {
7988613b
KO		 1137 b_offset += bvl.bv_offset;
1138 bio_page = bvl.bv_page;
d592a996
SL		 1139 if (frombio) {
1140 if (sh->raid_conf->skip_copy &&
1141 b_offset == 0 && page_offset == 0 &&
1e6d690b
SL		 1142 clen == STRIPE_SIZE &&
1143 !no_skipcopy)
d592a996
SL		 1144 *page = bio_page;
1145 else
1146 tx = async_memcpy(*page, bio_page, page_offset,
a08abd8c 1147 b_offset, clen, &submit);
d592a996
SL		 1148 } else
1149 tx = async_memcpy(bio_page, *page, b_offset,
a08abd8c 1150 page_offset, clen, &submit);
91c00924 1151 }
a08abd8c
DW		 1152 /* chain the operations */
1153 submit.depend_tx = tx;
1154
91c00924
DW		 1155 if (clen < len) /* hit end of page */
1156 break;
1157 page_offset += len;
1158 }
1159
1160 return tx;
1161}
1162
1163static void ops_complete_biofill(void *stripe_head_ref)
1164{
1165 struct stripe_head *sh = stripe_head_ref;
34a6f80e 1166 struct bio_list return_bi = BIO_EMPTY_LIST;
e4d84909 1167 int i;
91c00924 1168
e46b272b 1169 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1170 (unsigned long long)sh->sector);
1171
1172 /* clear completed biofills */
1173 for (i = sh->disks; i--; ) {
1174 struct r5dev *dev = &sh->dev[i];
91c00924
DW		 1175
1176 /* acknowledge completion of a biofill operation */
e4d84909
DW		 1177 /* and check if we need to reply to a read request,
1178 * new R5_Wantfill requests are held off until
83de75cc 1179 * !STRIPE_BIOFILL_RUN
e4d84909
DW		 1180 */
1181 if (test_and_clear_bit(R5_Wantfill, &dev->flags)) {
91c00924 1182 struct bio *rbi, *rbi2;
91c00924 1183
91c00924
DW		 1184 BUG_ON(!dev->read);
1185 rbi = dev->read;
1186 dev->read = NULL;
4f024f37 1187 while (rbi && rbi->bi_iter.bi_sector <
91c00924
DW		 1188 dev->sector + STRIPE_SECTORS) {
1189 rbi2 = r5_next_bio(rbi, dev->sector);
34a6f80e
N		 1190 if (!raid5_dec_bi_active_stripes(rbi))
1191 bio_list_add(&return_bi, rbi);
91c00924
DW		 1192 rbi = rbi2;
1193 }
1194 }
1195 }
83de75cc 1196 clear_bit(STRIPE_BIOFILL_RUN, &sh->state);
91c00924 1197
34a6f80e 1198 return_io(&return_bi);
91c00924 1199
e4d84909 1200 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 1201 raid5_release_stripe(sh);
91c00924
DW		 1202}
1203
1204static void ops_run_biofill(struct stripe_head *sh)
1205{
1206 struct dma_async_tx_descriptor *tx = NULL;
a08abd8c 1207 struct async_submit_ctl submit;
91c00924
DW		 1208 int i;
1209
59fc630b 1210 BUG_ON(sh->batch_head);
e46b272b 1211 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1212 (unsigned long long)sh->sector);
1213
1214 for (i = sh->disks; i--; ) {
1215 struct r5dev *dev = &sh->dev[i];
1216 if (test_bit(R5_Wantfill, &dev->flags)) {
1217 struct bio *rbi;
b17459c0 1218 spin_lock_irq(&sh->stripe_lock);
91c00924
DW		 1219 dev->read = rbi = dev->toread;
1220 dev->toread = NULL;
b17459c0 1221 spin_unlock_irq(&sh->stripe_lock);
4f024f37 1222 while (rbi && rbi->bi_iter.bi_sector <
91c00924 1223 dev->sector + STRIPE_SECTORS) {
d592a996 1224 tx = async_copy_data(0, rbi, &dev->page,
1e6d690b 1225 dev->sector, tx, sh, 0);
91c00924
DW		 1226 rbi = r5_next_bio(rbi, dev->sector);
1227 }
1228 }
1229 }
1230
1231 atomic_inc(&sh->count);
a08abd8c
DW		 1232 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL);
1233 async_trigger_callback(&submit);
91c00924
DW		 1234}
1235
4e7d2c0a 1236static void mark_target_uptodate(struct stripe_head *sh, int target)
91c00924 1237{
4e7d2c0a 1238 struct r5dev *tgt;
91c00924 1239
4e7d2c0a
DW		 1240 if (target < 0)
1241 return;
91c00924 1242
4e7d2c0a 1243 tgt = &sh->dev[target];
91c00924
DW		 1244 set_bit(R5_UPTODATE, &tgt->flags);
1245 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1246 clear_bit(R5_Wantcompute, &tgt->flags);
4e7d2c0a
DW		 1247}
1248
ac6b53b6 1249static void ops_complete_compute(void *stripe_head_ref)
91c00924
DW		 1250{
1251 struct stripe_head *sh = stripe_head_ref;
91c00924 1252
e46b272b 1253 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1254 (unsigned long long)sh->sector);
1255
ac6b53b6 1256 /* mark the computed target(s) as uptodate */
4e7d2c0a 1257 mark_target_uptodate(sh, sh->ops.target);
ac6b53b6 1258 mark_target_uptodate(sh, sh->ops.target2);
4e7d2c0a 1259
ecc65c9b
DW		 1260 clear_bit(STRIPE_COMPUTE_RUN, &sh->state);
1261 if (sh->check_state == check_state_compute_run)
1262 sh->check_state = check_state_compute_result;
91c00924 1263 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 1264 raid5_release_stripe(sh);
91c00924
DW		 1265}
1266
d6f38f31
DW		 1267/* return a pointer to the address conversion region of the scribble buffer */
1268static addr_conv_t *to_addr_conv(struct stripe_head *sh,
46d5b785 1269 struct raid5_percpu *percpu, int i)
d6f38f31 1270{
46d5b785 1271 void *addr;
1272
1273 addr = flex_array_get(percpu->scribble, i);
1274 return addr + sizeof(struct page *) * (sh->disks + 2);
1275}
1276
1277/* return a pointer to the address conversion region of the scribble buffer */
1278static struct page **to_addr_page(struct raid5_percpu *percpu, int i)
1279{
1280 void *addr;
1281
1282 addr = flex_array_get(percpu->scribble, i);
1283 return addr;
d6f38f31
DW		 1284}
1285
1286static struct dma_async_tx_descriptor *
1287ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 1288{
91c00924 1289 int disks = sh->disks;
46d5b785 1290 struct page **xor_srcs = to_addr_page(percpu, 0);
91c00924
DW		 1291 int target = sh->ops.target;
1292 struct r5dev *tgt = &sh->dev[target];
1293 struct page *xor_dest = tgt->page;
1294 int count = 0;
1295 struct dma_async_tx_descriptor *tx;
a08abd8c 1296 struct async_submit_ctl submit;
91c00924
DW		 1297 int i;
1298
59fc630b 1299 BUG_ON(sh->batch_head);
1300
91c00924 1301 pr_debug("%s: stripe %llu block: %d\n",
e46b272b 1302 __func__, (unsigned long long)sh->sector, target);
91c00924
DW		 1303 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1304
1305 for (i = disks; i--; )
1306 if (i != target)
1307 xor_srcs[count++] = sh->dev[i].page;
1308
1309 atomic_inc(&sh->count);
1310
0403e382 1311 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL,
46d5b785 1312 ops_complete_compute, sh, to_addr_conv(sh, percpu, 0));
91c00924 1313 if (unlikely(count == 1))
a08abd8c 1314 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
91c00924 1315 else
a08abd8c 1316 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924 1317
91c00924
DW		 1318 return tx;
1319}
1320
ac6b53b6
DW		 1321/* set_syndrome_sources - populate source buffers for gen_syndrome
1322 * @srcs - (struct page *) array of size sh->disks
1323 * @sh - stripe_head to parse
1324 *
1325 * Populates srcs in proper layout order for the stripe and returns the
1326 * 'count' of sources to be used in a call to async_gen_syndrome. The P
1327 * destination buffer is recorded in srcs[count] and the Q destination
1328 * is recorded in srcs[count+1]].
1329 */
584acdd4
MS		 1330static int set_syndrome_sources(struct page **srcs,
1331 struct stripe_head *sh,
1332 int srctype)
ac6b53b6
DW		 1333{
1334 int disks = sh->disks;
1335 int syndrome_disks = sh->ddf_layout ? disks : (disks - 2);
1336 int d0_idx = raid6_d0(sh);
1337 int count;
1338 int i;
1339
1340 for (i = 0; i < disks; i++)
5dd33c9a 1341 srcs[i] = NULL;
ac6b53b6
DW		 1342
1343 count = 0;
1344 i = d0_idx;
1345 do {
1346 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
584acdd4 1347 struct r5dev *dev = &sh->dev[i];
ac6b53b6 1348
584acdd4
MS		 1349 if (i == sh->qd_idx || i == sh->pd_idx ||
1350 (srctype == SYNDROME_SRC_ALL) ||
1351 (srctype == SYNDROME_SRC_WANT_DRAIN &&
1e6d690b
SL		 1352 (test_bit(R5_Wantdrain, &dev->flags) ||
1353 test_bit(R5_InJournal, &dev->flags))) ||
584acdd4 1354 (srctype == SYNDROME_SRC_WRITTEN &&
1e6d690b
SL		 1355 dev->written)) {
1356 if (test_bit(R5_InJournal, &dev->flags))
1357 srcs[slot] = sh->dev[i].orig_page;
1358 else
1359 srcs[slot] = sh->dev[i].page;
1360 }
ac6b53b6
DW		 1361 i = raid6_next_disk(i, disks);
1362 } while (i != d0_idx);
ac6b53b6 1363
e4424fee 1364 return syndrome_disks;
ac6b53b6
DW		 1365}
1366
1367static struct dma_async_tx_descriptor *
1368ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu)
1369{
1370 int disks = sh->disks;
46d5b785 1371 struct page **blocks = to_addr_page(percpu, 0);
ac6b53b6
DW		 1372 int target;
1373 int qd_idx = sh->qd_idx;
1374 struct dma_async_tx_descriptor *tx;
1375 struct async_submit_ctl submit;
1376 struct r5dev *tgt;
1377 struct page *dest;
1378 int i;
1379 int count;
1380
59fc630b 1381 BUG_ON(sh->batch_head);
ac6b53b6
DW		 1382 if (sh->ops.target < 0)
1383 target = sh->ops.target2;
1384 else if (sh->ops.target2 < 0)
1385 target = sh->ops.target;
91c00924 1386 else
ac6b53b6
DW		 1387 /* we should only have one valid target */
1388 BUG();
1389 BUG_ON(target < 0);
1390 pr_debug("%s: stripe %llu block: %d\n",
1391 __func__, (unsigned long long)sh->sector, target);
1392
1393 tgt = &sh->dev[target];
1394 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1395 dest = tgt->page;
1396
1397 atomic_inc(&sh->count);
1398
1399 if (target == qd_idx) {
584acdd4 1400 count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL);
ac6b53b6
DW		 1401 blocks[count] = NULL; /* regenerating p is not necessary */
1402 BUG_ON(blocks[count+1] != dest); /* q should already be set */
0403e382
DW		 1403 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1404 ops_complete_compute, sh,
46d5b785 1405 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1406 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
1407 } else {
1408 /* Compute any data- or p-drive using XOR */
1409 count = 0;
1410 for (i = disks; i-- ; ) {
1411 if (i == target || i == qd_idx)
1412 continue;
1413 blocks[count++] = sh->dev[i].page;
1414 }
1415
0403e382
DW		 1416 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1417 NULL, ops_complete_compute, sh,
46d5b785 1418 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1419 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit);
1420 }
91c00924 1421
91c00924
DW		 1422 return tx;
1423}
1424
ac6b53b6
DW		 1425static struct dma_async_tx_descriptor *
1426ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu)
1427{
1428 int i, count, disks = sh->disks;
1429 int syndrome_disks = sh->ddf_layout ? disks : disks-2;
1430 int d0_idx = raid6_d0(sh);
1431 int faila = -1, failb = -1;
1432 int target = sh->ops.target;
1433 int target2 = sh->ops.target2;
1434 struct r5dev *tgt = &sh->dev[target];
1435 struct r5dev *tgt2 = &sh->dev[target2];
1436 struct dma_async_tx_descriptor *tx;
46d5b785 1437 struct page **blocks = to_addr_page(percpu, 0);
ac6b53b6
DW		 1438 struct async_submit_ctl submit;
1439
59fc630b 1440 BUG_ON(sh->batch_head);
ac6b53b6
DW		 1441 pr_debug("%s: stripe %llu block1: %d block2: %d\n",
1442 __func__, (unsigned long long)sh->sector, target, target2);
1443 BUG_ON(target < 0 || target2 < 0);
1444 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1445 BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags));
1446
6c910a78 1447 /* we need to open-code set_syndrome_sources to handle the
ac6b53b6
DW		 1448 * slot number conversion for 'faila' and 'failb'
1449 */
1450 for (i = 0; i < disks ; i++)
5dd33c9a 1451 blocks[i] = NULL;
ac6b53b6
DW		 1452 count = 0;
1453 i = d0_idx;
1454 do {
1455 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
1456
1457 blocks[slot] = sh->dev[i].page;
1458
1459 if (i == target)
1460 faila = slot;
1461 if (i == target2)
1462 failb = slot;
1463 i = raid6_next_disk(i, disks);
1464 } while (i != d0_idx);
ac6b53b6
DW		 1465
1466 BUG_ON(faila == failb);
1467 if (failb < faila)
1468 swap(faila, failb);
1469 pr_debug("%s: stripe: %llu faila: %d failb: %d\n",
1470 __func__, (unsigned long long)sh->sector, faila, failb);
1471
1472 atomic_inc(&sh->count);
1473
1474 if (failb == syndrome_disks+1) {
1475 /* Q disk is one of the missing disks */
1476 if (faila == syndrome_disks) {
1477 /* Missing P+Q, just recompute */
0403e382
DW		 1478 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1479 ops_complete_compute, sh,
46d5b785 1480 to_addr_conv(sh, percpu, 0));
e4424fee 1481 return async_gen_syndrome(blocks, 0, syndrome_disks+2,
ac6b53b6
DW		 1482 STRIPE_SIZE, &submit);
1483 } else {
1484 struct page *dest;
1485 int data_target;
1486 int qd_idx = sh->qd_idx;
1487
1488 /* Missing D+Q: recompute D from P, then recompute Q */
1489 if (target == qd_idx)
1490 data_target = target2;
1491 else
1492 data_target = target;
1493
1494 count = 0;
1495 for (i = disks; i-- ; ) {
1496 if (i == data_target || i == qd_idx)
1497 continue;
1498 blocks[count++] = sh->dev[i].page;
1499 }
1500 dest = sh->dev[data_target].page;
0403e382
DW		 1501 init_async_submit(&submit,
1502 ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1503 NULL, NULL, NULL,
46d5b785 1504 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1505 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE,
1506 &submit);
1507
584acdd4 1508 count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL);
0403e382
DW		 1509 init_async_submit(&submit, ASYNC_TX_FENCE, tx,
1510 ops_complete_compute, sh,
46d5b785 1511 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1512 return async_gen_syndrome(blocks, 0, count+2,
1513 STRIPE_SIZE, &submit);
1514 }
ac6b53b6 1515 } else {
6c910a78
DW		 1516 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1517 ops_complete_compute, sh,
46d5b785 1518 to_addr_conv(sh, percpu, 0));
6c910a78
DW		 1519 if (failb == syndrome_disks) {
1520 /* We're missing D+P. */
1521 return async_raid6_datap_recov(syndrome_disks+2,
1522 STRIPE_SIZE, faila,
1523 blocks, &submit);
1524 } else {
1525 /* We're missing D+D. */
1526 return async_raid6_2data_recov(syndrome_disks+2,
1527 STRIPE_SIZE, faila, failb,
1528 blocks, &submit);
1529 }
ac6b53b6
DW		 1530 }
1531}
1532
91c00924
DW		 1533static void ops_complete_prexor(void *stripe_head_ref)
1534{
1535 struct stripe_head *sh = stripe_head_ref;
1536
e46b272b 1537 pr_debug("%s: stripe %llu\n", __func__,
91c00924 1538 (unsigned long long)sh->sector);
1e6d690b
SL		 1539
1540 if (r5c_is_writeback(sh->raid_conf->log))
1541 /*
1542 * raid5-cache write back uses orig_page during prexor.
1543 * After prexor, it is time to free orig_page
1544 */
1545 r5c_release_extra_page(sh);
91c00924
DW		 1546}
1547
1548static struct dma_async_tx_descriptor *
584acdd4
MS		 1549ops_run_prexor5(struct stripe_head *sh, struct raid5_percpu *percpu,
1550 struct dma_async_tx_descriptor *tx)
91c00924 1551{
91c00924 1552 int disks = sh->disks;
46d5b785 1553 struct page **xor_srcs = to_addr_page(percpu, 0);
91c00924 1554 int count = 0, pd_idx = sh->pd_idx, i;
a08abd8c 1555 struct async_submit_ctl submit;
91c00924
DW		 1556
1557 /* existing parity data subtracted */
1558 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1559
59fc630b 1560 BUG_ON(sh->batch_head);
e46b272b 1561 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1562 (unsigned long long)sh->sector);
1563
1564 for (i = disks; i--; ) {
1565 struct r5dev *dev = &sh->dev[i];
1566 /* Only process blocks that are known to be uptodate */
1e6d690b
SL		 1567 if (test_bit(R5_InJournal, &dev->flags))
1568 xor_srcs[count++] = dev->orig_page;
1569 else if (test_bit(R5_Wantdrain, &dev->flags))
91c00924
DW		 1570 xor_srcs[count++] = dev->page;
1571 }
1572
0403e382 1573 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
46d5b785 1574 ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0));
a08abd8c 1575 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924
DW		 1576
1577 return tx;
1578}
1579
584acdd4
MS		 1580static struct dma_async_tx_descriptor *
1581ops_run_prexor6(struct stripe_head *sh, struct raid5_percpu *percpu,
1582 struct dma_async_tx_descriptor *tx)
1583{
1584 struct page **blocks = to_addr_page(percpu, 0);
1585 int count;
1586 struct async_submit_ctl submit;
1587
1588 pr_debug("%s: stripe %llu\n", __func__,
1589 (unsigned long long)sh->sector);
1590
1591 count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_WANT_DRAIN);
1592
1593 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_PQ_XOR_DST, tx,
1594 ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0));
1595 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
1596
1597 return tx;
1598}
1599
91c00924 1600static struct dma_async_tx_descriptor *
d8ee0728 1601ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
91c00924 1602{
1e6d690b 1603 struct r5conf *conf = sh->raid_conf;
91c00924 1604 int disks = sh->disks;
d8ee0728 1605 int i;
59fc630b 1606 struct stripe_head *head_sh = sh;
91c00924 1607
e46b272b 1608 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1609 (unsigned long long)sh->sector);
1610
1611 for (i = disks; i--; ) {
59fc630b 1612 struct r5dev *dev;
91c00924 1613 struct bio *chosen;
91c00924 1614
59fc630b 1615 sh = head_sh;
1616 if (test_and_clear_bit(R5_Wantdrain, &head_sh->dev[i].flags)) {
91c00924
DW		 1617 struct bio *wbi;
1618
59fc630b 1619again:
1620 dev = &sh->dev[i];
1e6d690b
SL		 1621 /*
1622 * clear R5_InJournal, so when rewriting a page in
1623 * journal, it is not skipped by r5l_log_stripe()
1624 */
1625 clear_bit(R5_InJournal, &dev->flags);
b17459c0 1626 spin_lock_irq(&sh->stripe_lock);
91c00924
DW		 1627 chosen = dev->towrite;
1628 dev->towrite = NULL;
7a87f434 1629 sh->overwrite_disks = 0;
91c00924
DW		 1630 BUG_ON(dev->written);
1631 wbi = dev->written = chosen;
b17459c0 1632 spin_unlock_irq(&sh->stripe_lock);
d592a996 1633 WARN_ON(dev->page != dev->orig_page);
91c00924 1634
4f024f37 1635 while (wbi && wbi->bi_iter.bi_sector <
91c00924 1636 dev->sector + STRIPE_SECTORS) {
1eff9d32 1637 if (wbi->bi_opf & REQ_FUA)
e9c7469b 1638 set_bit(R5_WantFUA, &dev->flags);
1eff9d32 1639 if (wbi->bi_opf & REQ_SYNC)
bc0934f0 1640 set_bit(R5_SyncIO, &dev->flags);
796a5cf0 1641 if (bio_op(wbi) == REQ_OP_DISCARD)
620125f2 1642 set_bit(R5_Discard, &dev->flags);
d592a996
SL		 1643 else {
1644 tx = async_copy_data(1, wbi, &dev->page,
1e6d690b
SL		 1645 dev->sector, tx, sh,
1646 r5c_is_writeback(conf->log));
1647 if (dev->page != dev->orig_page &&
1648 !r5c_is_writeback(conf->log)) {
d592a996
SL		 1649 set_bit(R5_SkipCopy, &dev->flags);
1650 clear_bit(R5_UPTODATE, &dev->flags);
1651 clear_bit(R5_OVERWRITE, &dev->flags);
1652 }
1653 }
91c00924
DW		 1654 wbi = r5_next_bio(wbi, dev->sector);
1655 }
59fc630b 1656
1657 if (head_sh->batch_head) {
1658 sh = list_first_entry(&sh->batch_list,
1659 struct stripe_head,
1660 batch_list);
1661 if (sh == head_sh)
1662 continue;
1663 goto again;
1664 }
91c00924
DW		 1665 }
1666 }
1667
1668 return tx;
1669}
1670
ac6b53b6 1671static void ops_complete_reconstruct(void *stripe_head_ref)
91c00924
DW		 1672{
1673 struct stripe_head *sh = stripe_head_ref;
ac6b53b6
DW		 1674 int disks = sh->disks;
1675 int pd_idx = sh->pd_idx;
1676 int qd_idx = sh->qd_idx;
1677 int i;
9e444768 1678 bool fua = false, sync = false, discard = false;
91c00924 1679
e46b272b 1680 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1681 (unsigned long long)sh->sector);
1682
bc0934f0 1683 for (i = disks; i--; ) {
e9c7469b 1684 fua |= test_bit(R5_WantFUA, &sh->dev[i].flags);
bc0934f0 1685 sync |= test_bit(R5_SyncIO, &sh->dev[i].flags);
9e444768 1686 discard |= test_bit(R5_Discard, &sh->dev[i].flags);
bc0934f0 1687 }
e9c7469b 1688
91c00924
DW		 1689 for (i = disks; i--; ) {
1690 struct r5dev *dev = &sh->dev[i];
ac6b53b6 1691
e9c7469b 1692 if (dev->written || i == pd_idx || i == qd_idx) {
d592a996 1693 if (!discard && !test_bit(R5_SkipCopy, &dev->flags))
9e444768 1694 set_bit(R5_UPTODATE, &dev->flags);
e9c7469b
TH		 1695 if (fua)
1696 set_bit(R5_WantFUA, &dev->flags);
bc0934f0
SL		 1697 if (sync)
1698 set_bit(R5_SyncIO, &dev->flags);
e9c7469b 1699 }
91c00924
DW		 1700 }
1701
d8ee0728
DW		 1702 if (sh->reconstruct_state == reconstruct_state_drain_run)
1703 sh->reconstruct_state = reconstruct_state_drain_result;
1704 else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run)
1705 sh->reconstruct_state = reconstruct_state_prexor_drain_result;
1706 else {
1707 BUG_ON(sh->reconstruct_state != reconstruct_state_run);
1708 sh->reconstruct_state = reconstruct_state_result;
1709 }
91c00924
DW		 1710
1711 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 1712 raid5_release_stripe(sh);
91c00924
DW		 1713}
1714
1715static void
ac6b53b6
DW		 1716ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu,
1717 struct dma_async_tx_descriptor *tx)
91c00924 1718{
91c00924 1719 int disks = sh->disks;
59fc630b 1720 struct page **xor_srcs;
a08abd8c 1721 struct async_submit_ctl submit;
59fc630b 1722 int count, pd_idx = sh->pd_idx, i;
91c00924 1723 struct page *xor_dest;
d8ee0728 1724 int prexor = 0;
91c00924 1725 unsigned long flags;
59fc630b 1726 int j = 0;
1727 struct stripe_head *head_sh = sh;
1728 int last_stripe;
91c00924 1729
e46b272b 1730 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1731 (unsigned long long)sh->sector);
1732
620125f2
SL		 1733 for (i = 0; i < sh->disks; i++) {
1734 if (pd_idx == i)
1735 continue;
1736 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1737 break;
1738 }
1739 if (i >= sh->disks) {
1740 atomic_inc(&sh->count);
620125f2
SL		 1741 set_bit(R5_Discard, &sh->dev[pd_idx].flags);
1742 ops_complete_reconstruct(sh);
1743 return;
1744 }
59fc630b 1745again:
1746 count = 0;
1747 xor_srcs = to_addr_page(percpu, j);
91c00924
DW		 1748 /* check if prexor is active which means only process blocks
1749 * that are part of a read-modify-write (written)
1750 */
59fc630b 1751 if (head_sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
d8ee0728 1752 prexor = 1;
91c00924
DW		 1753 xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1754 for (i = disks; i--; ) {
1755 struct r5dev *dev = &sh->dev[i];
1e6d690b
SL		 1756 if (head_sh->dev[i].written ||
1757 test_bit(R5_InJournal, &head_sh->dev[i].flags))
91c00924
DW		 1758 xor_srcs[count++] = dev->page;
1759 }
1760 } else {
1761 xor_dest = sh->dev[pd_idx].page;
1762 for (i = disks; i--; ) {
1763 struct r5dev *dev = &sh->dev[i];
1764 if (i != pd_idx)
1765 xor_srcs[count++] = dev->page;
1766 }
1767 }
1768
91c00924
DW		 1769 /* 1/ if we prexor'd then the dest is reused as a source
1770 * 2/ if we did not prexor then we are redoing the parity
1771 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
1772 * for the synchronous xor case
1773 */
59fc630b 1774 last_stripe = !head_sh->batch_head ||
1775 list_first_entry(&sh->batch_list,
1776 struct stripe_head, batch_list) == head_sh;
1777 if (last_stripe) {
1778 flags = ASYNC_TX_ACK |
1779 (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
1780
1781 atomic_inc(&head_sh->count);
1782 init_async_submit(&submit, flags, tx, ops_complete_reconstruct, head_sh,
1783 to_addr_conv(sh, percpu, j));
1784 } else {
1785 flags = prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST;
1786 init_async_submit(&submit, flags, tx, NULL, NULL,
1787 to_addr_conv(sh, percpu, j));
1788 }
91c00924 1789
a08abd8c
DW		 1790 if (unlikely(count == 1))
1791 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
1792 else
1793 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
59fc630b 1794 if (!last_stripe) {
1795 j++;
1796 sh = list_first_entry(&sh->batch_list, struct stripe_head,
1797 batch_list);
1798 goto again;
1799 }
91c00924
DW		 1800}
1801
ac6b53b6
DW		 1802static void
1803ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu,
1804 struct dma_async_tx_descriptor *tx)
1805{
1806 struct async_submit_ctl submit;
59fc630b 1807 struct page **blocks;
1808 int count, i, j = 0;
1809 struct stripe_head *head_sh = sh;
1810 int last_stripe;
584acdd4
MS		 1811 int synflags;
1812 unsigned long txflags;
ac6b53b6
DW		 1813
1814 pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector);
1815
620125f2
SL		 1816 for (i = 0; i < sh->disks; i++) {
1817 if (sh->pd_idx == i || sh->qd_idx == i)
1818 continue;
1819 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1820 break;
1821 }
1822 if (i >= sh->disks) {
1823 atomic_inc(&sh->count);
620125f2
SL		 1824 set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
1825 set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
1826 ops_complete_reconstruct(sh);
1827 return;
1828 }
1829
59fc630b 1830again:
1831 blocks = to_addr_page(percpu, j);
584acdd4
MS		 1832
1833 if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
1834 synflags = SYNDROME_SRC_WRITTEN;
1835 txflags = ASYNC_TX_ACK | ASYNC_TX_PQ_XOR_DST;
1836 } else {
1837 synflags = SYNDROME_SRC_ALL;
1838 txflags = ASYNC_TX_ACK;
1839 }
1840
1841 count = set_syndrome_sources(blocks, sh, synflags);
59fc630b 1842 last_stripe = !head_sh->batch_head ||
1843 list_first_entry(&sh->batch_list,
1844 struct stripe_head, batch_list) == head_sh;
1845
1846 if (last_stripe) {
1847 atomic_inc(&head_sh->count);
584acdd4 1848 init_async_submit(&submit, txflags, tx, ops_complete_reconstruct,
59fc630b 1849 head_sh, to_addr_conv(sh, percpu, j));
1850 } else
1851 init_async_submit(&submit, 0, tx, NULL, NULL,
1852 to_addr_conv(sh, percpu, j));
48769695 1853 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
59fc630b 1854 if (!last_stripe) {
1855 j++;
1856 sh = list_first_entry(&sh->batch_list, struct stripe_head,
1857 batch_list);
1858 goto again;
1859 }
91c00924
DW		 1860}
1861
1862static void ops_complete_check(void *stripe_head_ref)
1863{
1864 struct stripe_head *sh = stripe_head_ref;
91c00924 1865
e46b272b 1866 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1867 (unsigned long long)sh->sector);
1868
ecc65c9b 1869 sh->check_state = check_state_check_result;
91c00924 1870 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 1871 raid5_release_stripe(sh);
91c00924
DW		 1872}
1873
ac6b53b6 1874static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 1875{
91c00924 1876 int disks = sh->disks;
ac6b53b6
DW		 1877 int pd_idx = sh->pd_idx;
1878 int qd_idx = sh->qd_idx;
1879 struct page *xor_dest;
46d5b785 1880 struct page **xor_srcs = to_addr_page(percpu, 0);
91c00924 1881 struct dma_async_tx_descriptor *tx;
a08abd8c 1882 struct async_submit_ctl submit;
ac6b53b6
DW		 1883 int count;
1884 int i;
91c00924 1885
e46b272b 1886 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1887 (unsigned long long)sh->sector);
1888
59fc630b 1889 BUG_ON(sh->batch_head);
ac6b53b6
DW		 1890 count = 0;
1891 xor_dest = sh->dev[pd_idx].page;
1892 xor_srcs[count++] = xor_dest;
91c00924 1893 for (i = disks; i--; ) {
ac6b53b6
DW		 1894 if (i == pd_idx || i == qd_idx)
1895 continue;
1896 xor_srcs[count++] = sh->dev[i].page;
91c00924
DW		 1897 }
1898
d6f38f31 1899 init_async_submit(&submit, 0, NULL, NULL, NULL,
46d5b785 1900 to_addr_conv(sh, percpu, 0));
099f53cb 1901 tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
a08abd8c 1902 &sh->ops.zero_sum_result, &submit);
91c00924 1903
91c00924 1904 atomic_inc(&sh->count);
a08abd8c
DW		 1905 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL);
1906 tx = async_trigger_callback(&submit);
91c00924
DW		 1907}
1908
ac6b53b6
DW		 1909static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp)
1910{
46d5b785 1911 struct page **srcs = to_addr_page(percpu, 0);
ac6b53b6
DW		 1912 struct async_submit_ctl submit;
1913 int count;
1914
1915 pr_debug("%s: stripe %llu checkp: %d\n", __func__,
1916 (unsigned long long)sh->sector, checkp);
1917
59fc630b 1918 BUG_ON(sh->batch_head);
584acdd4 1919 count = set_syndrome_sources(srcs, sh, SYNDROME_SRC_ALL);
ac6b53b6
DW		 1920 if (!checkp)
1921 srcs[count] = NULL;
91c00924 1922
91c00924 1923 atomic_inc(&sh->count);
ac6b53b6 1924 init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check,
46d5b785 1925 sh, to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1926 async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE,
1927 &sh->ops.zero_sum_result, percpu->spare_page, &submit);
91c00924
DW		 1928}
1929
51acbcec 1930static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
91c00924
DW		 1931{
1932 int overlap_clear = 0, i, disks = sh->disks;
1933 struct dma_async_tx_descriptor *tx = NULL;
d1688a6d 1934 struct r5conf *conf = sh->raid_conf;
ac6b53b6 1935 int level = conf->level;
d6f38f31
DW		 1936 struct raid5_percpu *percpu;
1937 unsigned long cpu;
91c00924 1938
d6f38f31
DW		 1939 cpu = get_cpu();
1940 percpu = per_cpu_ptr(conf->percpu, cpu);
83de75cc 1941 if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) {
91c00924
DW		 1942 ops_run_biofill(sh);
1943 overlap_clear++;
1944 }
1945
7b3a871e 1946 if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) {
ac6b53b6
DW		 1947 if (level < 6)
1948 tx = ops_run_compute5(sh, percpu);
1949 else {
1950 if (sh->ops.target2 < 0 || sh->ops.target < 0)
1951 tx = ops_run_compute6_1(sh, percpu);
1952 else
1953 tx = ops_run_compute6_2(sh, percpu);
1954 }
1955 /* terminate the chain if reconstruct is not set to be run */
1956 if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request))
7b3a871e
DW		 1957 async_tx_ack(tx);
1958 }
91c00924 1959
584acdd4
MS		 1960 if (test_bit(STRIPE_OP_PREXOR, &ops_request)) {
1961 if (level < 6)
1962 tx = ops_run_prexor5(sh, percpu, tx);
1963 else
1964 tx = ops_run_prexor6(sh, percpu, tx);
1965 }
91c00924 1966
600aa109 1967 if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) {
d8ee0728 1968 tx = ops_run_biodrain(sh, tx);
91c00924
DW		 1969 overlap_clear++;
1970 }
1971
ac6b53b6
DW		 1972 if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) {
1973 if (level < 6)
1974 ops_run_reconstruct5(sh, percpu, tx);
1975 else
1976 ops_run_reconstruct6(sh, percpu, tx);
1977 }
91c00924 1978
ac6b53b6
DW		 1979 if (test_bit(STRIPE_OP_CHECK, &ops_request)) {
1980 if (sh->check_state == check_state_run)
1981 ops_run_check_p(sh, percpu);
1982 else if (sh->check_state == check_state_run_q)
1983 ops_run_check_pq(sh, percpu, 0);
1984 else if (sh->check_state == check_state_run_pq)
1985 ops_run_check_pq(sh, percpu, 1);
1986 else
1987 BUG();
1988 }
91c00924 1989
59fc630b 1990 if (overlap_clear && !sh->batch_head)
91c00924
DW		 1991 for (i = disks; i--; ) {
1992 struct r5dev *dev = &sh->dev[i];
1993 if (test_and_clear_bit(R5_Overlap, &dev->flags))
1994 wake_up(&sh->raid_conf->wait_for_overlap);
1995 }
d6f38f31 1996 put_cpu();
91c00924
DW		 1997}
1998
5f9d1fde
SL		 1999static struct stripe_head *alloc_stripe(struct kmem_cache *sc, gfp_t gfp,
2000 int disks)
f18c1a35
N		 2001{
2002 struct stripe_head *sh;
5f9d1fde 2003 int i;
f18c1a35
N		 2004
2005 sh = kmem_cache_zalloc(sc, gfp);
2006 if (sh) {
2007 spin_lock_init(&sh->stripe_lock);
2008 spin_lock_init(&sh->batch_lock);
2009 INIT_LIST_HEAD(&sh->batch_list);
2010 INIT_LIST_HEAD(&sh->lru);
a39f7afd 2011 INIT_LIST_HEAD(&sh->r5c);
d7bd398e 2012 INIT_LIST_HEAD(&sh->log_list);
f18c1a35 2013 atomic_set(&sh->count, 1);
a39f7afd 2014 sh->log_start = MaxSector;
5f9d1fde
SL		 2015 for (i = 0; i < disks; i++) {
2016 struct r5dev *dev = &sh->dev[i];
2017
3a83f467
ML		 2018 bio_init(&dev->req, &dev->vec, 1);
2019 bio_init(&dev->rreq, &dev->rvec, 1);
5f9d1fde 2020 }
f18c1a35
N		 2021 }
2022 return sh;
2023}
486f0644 2024static int grow_one_stripe(struct r5conf *conf, gfp_t gfp)
1da177e4
LT		 2025{
2026 struct stripe_head *sh;
f18c1a35 2027
5f9d1fde 2028 sh = alloc_stripe(conf->slab_cache, gfp, conf->pool_size);
3f294f4f
N		 2029 if (!sh)
2030 return 0;
6ce32846 2031
3f294f4f 2032 sh->raid_conf = conf;
3f294f4f 2033
a9683a79 2034 if (grow_buffers(sh, gfp)) {
e4e11e38 2035 shrink_buffers(sh);
3f294f4f
N		 2036 kmem_cache_free(conf->slab_cache, sh);
2037 return 0;
2038 }
486f0644
N		 2039 sh->hash_lock_index =
2040 conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS;
3f294f4f 2041 /* we just created an active stripe so... */
3f294f4f 2042 atomic_inc(&conf->active_stripes);
59fc630b 2043
6d036f7d 2044 raid5_release_stripe(sh);
486f0644 2045 conf->max_nr_stripes++;
3f294f4f
N		 2046 return 1;
2047}
2048
d1688a6d 2049static int grow_stripes(struct r5conf *conf, int num)
3f294f4f 2050{
e18b890b 2051 struct kmem_cache *sc;
5e5e3e78 2052 int devs = max(conf->raid_disks, conf->previous_raid_disks);
1da177e4 2053
f4be6b43
N		 2054 if (conf->mddev->gendisk)
2055 sprintf(conf->cache_name[0],
2056 "raid%d-%s", conf->level, mdname(conf->mddev));
2057 else
2058 sprintf(conf->cache_name[0],
2059 "raid%d-%p", conf->level, conf->mddev);
2060 sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]);
2061
ad01c9e3
N		 2062 conf->active_name = 0;
2063 sc = kmem_cache_create(conf->cache_name[conf->active_name],
1da177e4 2064 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
20c2df83 2065 0, 0, NULL);
1da177e4
LT		 2066 if (!sc)
2067 return 1;
2068 conf->slab_cache = sc;
ad01c9e3 2069 conf->pool_size = devs;
486f0644
N		 2070 while (num--)
2071 if (!grow_one_stripe(conf, GFP_KERNEL))
1da177e4 2072 return 1;
486f0644 2073
1da177e4
LT		 2074 return 0;
2075}
29269553 2076
d6f38f31
DW		 2077/**
2078 * scribble_len - return the required size of the scribble region
2079 * @num - total number of disks in the array
2080 *
2081 * The size must be enough to contain:
2082 * 1/ a struct page pointer for each device in the array +2
2083 * 2/ room to convert each entry in (1) to its corresponding dma
2084 * (dma_map_page()) or page (page_address()) address.
2085 *
2086 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
2087 * calculate over all devices (not just the data blocks), using zeros in place
2088 * of the P and Q blocks.
2089 */
46d5b785 2090static struct flex_array *scribble_alloc(int num, int cnt, gfp_t flags)
d6f38f31 2091{
46d5b785 2092 struct flex_array *ret;
d6f38f31
DW		 2093 size_t len;
2094
2095 len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2);
46d5b785 2096 ret = flex_array_alloc(len, cnt, flags);
2097 if (!ret)
2098 return NULL;
2099 /* always prealloc all elements, so no locking is required */
2100 if (flex_array_prealloc(ret, 0, cnt, flags)) {
2101 flex_array_free(ret);
2102 return NULL;
2103 }
2104 return ret;
d6f38f31
DW		 2105}
2106
738a2738
N		 2107static int resize_chunks(struct r5conf *conf, int new_disks, int new_sectors)
2108{
2109 unsigned long cpu;
2110 int err = 0;
2111
27a353c0
SL		 2112 /*
2113 * Never shrink. And mddev_suspend() could deadlock if this is called
2114 * from raid5d. In that case, scribble_disks and scribble_sectors
2115 * should equal to new_disks and new_sectors
2116 */
2117 if (conf->scribble_disks >= new_disks &&
2118 conf->scribble_sectors >= new_sectors)
2119 return 0;
738a2738
N		 2120 mddev_suspend(conf->mddev);
2121 get_online_cpus();
2122 for_each_present_cpu(cpu) {
2123 struct raid5_percpu *percpu;
2124 struct flex_array *scribble;
2125
2126 percpu = per_cpu_ptr(conf->percpu, cpu);
2127 scribble = scribble_alloc(new_disks,
2128 new_sectors / STRIPE_SECTORS,
2129 GFP_NOIO);
2130
2131 if (scribble) {
2132 flex_array_free(percpu->scribble);
2133 percpu->scribble = scribble;
2134 } else {
2135 err = -ENOMEM;
2136 break;
2137 }
2138 }
2139 put_online_cpus();
2140 mddev_resume(conf->mddev);
27a353c0
SL		 2141 if (!err) {
2142 conf->scribble_disks = new_disks;
2143 conf->scribble_sectors = new_sectors;
2144 }
738a2738
N		 2145 return err;
2146}
2147
d1688a6d 2148static int resize_stripes(struct r5conf *conf, int newsize)
ad01c9e3
N		 2149{
2150 /* Make all the stripes able to hold 'newsize' devices.
2151 * New slots in each stripe get 'page' set to a new page.
2152 *
2153 * This happens in stages:
2154 * 1/ create a new kmem_cache and allocate the required number of
2155 * stripe_heads.
83f0d77a 2156 * 2/ gather all the old stripe_heads and transfer the pages across
ad01c9e3
N		 2157 * to the new stripe_heads. This will have the side effect of
2158 * freezing the array as once all stripe_heads have been collected,
2159 * no IO will be possible. Old stripe heads are freed once their
2160 * pages have been transferred over, and the old kmem_cache is
2161 * freed when all stripes are done.
2162 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
2163 * we simple return a failre status - no need to clean anything up.
2164 * 4/ allocate new pages for the new slots in the new stripe_heads.
2165 * If this fails, we don't bother trying the shrink the
2166 * stripe_heads down again, we just leave them as they are.
2167 * As each stripe_head is processed the new one is released into
2168 * active service.
2169 *
2170 * Once step2 is started, we cannot afford to wait for a write,
2171 * so we use GFP_NOIO allocations.
2172 */
2173 struct stripe_head *osh, *nsh;
2174 LIST_HEAD(newstripes);
2175 struct disk_info *ndisks;
b5470dc5 2176 int err;
e18b890b 2177 struct kmem_cache *sc;
ad01c9e3 2178 int i;
566c09c5 2179 int hash, cnt;
ad01c9e3
N		 2180
2181 if (newsize <= conf->pool_size)
2182 return 0; /* never bother to shrink */
2183
b5470dc5
DW		 2184 err = md_allow_write(conf->mddev);
2185 if (err)
2186 return err;
2a2275d6 2187
ad01c9e3
N		 2188 /* Step 1 */
2189 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
2190 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
20c2df83 2191 0, 0, NULL);
ad01c9e3
N		 2192 if (!sc)
2193 return -ENOMEM;
2194
2d5b569b
N		 2195 /* Need to ensure auto-resizing doesn't interfere */
2196 mutex_lock(&conf->cache_size_mutex);
2197
ad01c9e3 2198 for (i = conf->max_nr_stripes; i; i--) {
5f9d1fde 2199 nsh = alloc_stripe(sc, GFP_KERNEL, newsize);
ad01c9e3
N		 2200 if (!nsh)
2201 break;
2202
ad01c9e3 2203 nsh->raid_conf = conf;
ad01c9e3
N		 2204 list_add(&nsh->lru, &newstripes);
2205 }
2206 if (i) {
2207 /* didn't get enough, give up */
2208 while (!list_empty(&newstripes)) {
2209 nsh = list_entry(newstripes.next, struct stripe_head, lru);
2210 list_del(&nsh->lru);
2211 kmem_cache_free(sc, nsh);
2212 }
2213 kmem_cache_destroy(sc);
2d5b569b 2214 mutex_unlock(&conf->cache_size_mutex);
ad01c9e3
N		 2215 return -ENOMEM;
2216 }
2217 /* Step 2 - Must use GFP_NOIO now.
2218 * OK, we have enough stripes, start collecting inactive
2219 * stripes and copying them over
2220 */
566c09c5
SL		 2221 hash = 0;
2222 cnt = 0;
ad01c9e3 2223 list_for_each_entry(nsh, &newstripes, lru) {
566c09c5 2224 lock_device_hash_lock(conf, hash);
6ab2a4b8 2225 wait_event_cmd(conf->wait_for_stripe,
566c09c5
SL		 2226 !list_empty(conf->inactive_list + hash),
2227 unlock_device_hash_lock(conf, hash),
2228 lock_device_hash_lock(conf, hash));
2229 osh = get_free_stripe(conf, hash);
2230 unlock_device_hash_lock(conf, hash);
f18c1a35 2231
d592a996 2232 for(i=0; i<conf->pool_size; i++) {
ad01c9e3 2233 nsh->dev[i].page = osh->dev[i].page;
d592a996
SL		 2234 nsh->dev[i].orig_page = osh->dev[i].page;
2235 }
566c09c5 2236 nsh->hash_lock_index = hash;
ad01c9e3 2237 kmem_cache_free(conf->slab_cache, osh);
566c09c5
SL		 2238 cnt++;
2239 if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS +
2240 !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) {
2241 hash++;
2242 cnt = 0;
2243 }
ad01c9e3
N		 2244 }
2245 kmem_cache_destroy(conf->slab_cache);
2246
2247 /* Step 3.
2248 * At this point, we are holding all the stripes so the array
2249 * is completely stalled, so now is a good time to resize
d6f38f31 2250 * conf->disks and the scribble region
ad01c9e3
N		 2251 */
2252 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
2253 if (ndisks) {
d7bd398e 2254 for (i = 0; i < conf->pool_size; i++)
ad01c9e3 2255 ndisks[i] = conf->disks[i];
d7bd398e
SL		 2256
2257 for (i = conf->pool_size; i < newsize; i++) {
2258 ndisks[i].extra_page = alloc_page(GFP_NOIO);
2259 if (!ndisks[i].extra_page)
2260 err = -ENOMEM;
2261 }
2262
2263 if (err) {
2264 for (i = conf->pool_size; i < newsize; i++)
2265 if (ndisks[i].extra_page)
2266 put_page(ndisks[i].extra_page);
2267 kfree(ndisks);
2268 } else {
2269 kfree(conf->disks);
2270 conf->disks = ndisks;
2271 }
ad01c9e3
N		 2272 } else
2273 err = -ENOMEM;
2274
2d5b569b 2275 mutex_unlock(&conf->cache_size_mutex);
ad01c9e3
N		 2276 /* Step 4, return new stripes to service */
2277 while(!list_empty(&newstripes)) {
2278 nsh = list_entry(newstripes.next, struct stripe_head, lru);
2279 list_del_init(&nsh->lru);
d6f38f31 2280
ad01c9e3
N		 2281 for (i=conf->raid_disks; i < newsize; i++)
2282 if (nsh->dev[i].page == NULL) {
2283 struct page *p = alloc_page(GFP_NOIO);
2284 nsh->dev[i].page = p;
d592a996 2285 nsh->dev[i].orig_page = p;
ad01c9e3
N		 2286 if (!p)
2287 err = -ENOMEM;
2288 }
6d036f7d 2289 raid5_release_stripe(nsh);
ad01c9e3
N		 2290 }
2291 /* critical section pass, GFP_NOIO no longer needed */
2292
2293 conf->slab_cache = sc;
2294 conf->active_name = 1-conf->active_name;
6e9eac2d
N		 2295 if (!err)
2296 conf->pool_size = newsize;
ad01c9e3
N		 2297 return err;
2298}
1da177e4 2299
486f0644 2300static int drop_one_stripe(struct r5conf *conf)
1da177e4
LT		 2301{
2302 struct stripe_head *sh;
49895bcc 2303 int hash = (conf->max_nr_stripes - 1) & STRIPE_HASH_LOCKS_MASK;
1da177e4 2304
566c09c5
SL		 2305 spin_lock_irq(conf->hash_locks + hash);
2306 sh = get_free_stripe(conf, hash);
2307 spin_unlock_irq(conf->hash_locks + hash);
3f294f4f
N		 2308 if (!sh)
2309 return 0;
78bafebd 2310 BUG_ON(atomic_read(&sh->count));
e4e11e38 2311 shrink_buffers(sh);
3f294f4f
N		 2312 kmem_cache_free(conf->slab_cache, sh);
2313 atomic_dec(&conf->active_stripes);
486f0644 2314 conf->max_nr_stripes--;
3f294f4f
N		 2315 return 1;
2316}
2317
d1688a6d 2318static void shrink_stripes(struct r5conf *conf)
3f294f4f 2319{
486f0644
N		 2320 while (conf->max_nr_stripes &&
2321 drop_one_stripe(conf))
2322 ;
3f294f4f 2323
644df1a8 2324 kmem_cache_destroy(conf->slab_cache);
1da177e4
LT		 2325 conf->slab_cache = NULL;
2326}
2327
4246a0b6 2328static void raid5_end_read_request(struct bio * bi)
1da177e4 2329{
99c0fb5f 2330 struct stripe_head *sh = bi->bi_private;
d1688a6d 2331 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 2332 int disks = sh->disks, i;
d6950432 2333 char b[BDEVNAME_SIZE];
dd054fce 2334 struct md_rdev *rdev = NULL;
05616be5 2335 sector_t s;
1da177e4
LT		 2336
2337 for (i=0 ; i<disks; i++)
2338 if (bi == &sh->dev[i].req)
2339 break;
2340
4246a0b6 2341 pr_debug("end_read_request %llu/%d, count: %d, error %d.\n",
45b4233c 2342 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
4246a0b6 2343 bi->bi_error);
1da177e4 2344 if (i == disks) {
5f9d1fde 2345 bio_reset(bi);
1da177e4 2346 BUG();
6712ecf8 2347 return;
1da177e4 2348 }
14a75d3e 2349 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
dd054fce
N		 2350 /* If replacement finished while this request was outstanding,
2351 * 'replacement' might be NULL already.
2352 * In that case it moved down to 'rdev'.
2353 * rdev is not removed until all requests are finished.
2354 */
14a75d3e 2355 rdev = conf->disks[i].replacement;
dd054fce 2356 if (!rdev)
14a75d3e 2357 rdev = conf->disks[i].rdev;
1da177e4 2358
05616be5
N		 2359 if (use_new_offset(conf, sh))
2360 s = sh->sector + rdev->new_data_offset;
2361 else
2362 s = sh->sector + rdev->data_offset;
4246a0b6 2363 if (!bi->bi_error) {
1da177e4 2364 set_bit(R5_UPTODATE, &sh->dev[i].flags);
4e5314b5 2365 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
14a75d3e
N		 2366 /* Note that this cannot happen on a
2367 * replacement device. We just fail those on
2368 * any error
2369 */
cc6167b4
N		 2370 pr_info_ratelimited(
2371 "md/raid:%s: read error corrected (%lu sectors at %llu on %s)\n",
8bda470e 2372 mdname(conf->mddev), STRIPE_SECTORS,
05616be5 2373 (unsigned long long)s,
8bda470e 2374 bdevname(rdev->bdev, b));
ddd5115f 2375 atomic_add(STRIPE_SECTORS, &rdev->corrected_errors);
4e5314b5
N		 2376 clear_bit(R5_ReadError, &sh->dev[i].flags);
2377 clear_bit(R5_ReWrite, &sh->dev[i].flags);
3f9e7c14 2378 } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2379 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2380
14a75d3e
N		 2381 if (atomic_read(&rdev->read_errors))
2382 atomic_set(&rdev->read_errors, 0);
1da177e4 2383 } else {
14a75d3e 2384 const char *bdn = bdevname(rdev->bdev, b);
ba22dcbf 2385 int retry = 0;
2e8ac303 2386 int set_bad = 0;
d6950432 2387
1da177e4 2388 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
d6950432 2389 atomic_inc(&rdev->read_errors);
14a75d3e 2390 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
cc6167b4
N		 2391 pr_warn_ratelimited(
2392 "md/raid:%s: read error on replacement device (sector %llu on %s).\n",
14a75d3e 2393 mdname(conf->mddev),
05616be5 2394 (unsigned long long)s,
14a75d3e 2395 bdn);
2e8ac303 2396 else if (conf->mddev->degraded >= conf->max_degraded) {
2397 set_bad = 1;
cc6167b4
N		 2398 pr_warn_ratelimited(
2399 "md/raid:%s: read error not correctable (sector %llu on %s).\n",
8bda470e 2400 mdname(conf->mddev),
05616be5 2401 (unsigned long long)s,
8bda470e 2402 bdn);
2e8ac303 2403 } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) {
4e5314b5 2404 /* Oh, no!!! */
2e8ac303 2405 set_bad = 1;
cc6167b4
N		 2406 pr_warn_ratelimited(
2407 "md/raid:%s: read error NOT corrected!! (sector %llu on %s).\n",
8bda470e 2408 mdname(conf->mddev),
05616be5 2409 (unsigned long long)s,
8bda470e 2410 bdn);
2e8ac303 2411 } else if (atomic_read(&rdev->read_errors)
ba22dcbf 2412 > conf->max_nr_stripes)
cc6167b4 2413 pr_warn("md/raid:%s: Too many read errors, failing device %s.\n",
d6950432 2414 mdname(conf->mddev), bdn);
ba22dcbf
N		 2415 else
2416 retry = 1;
edfa1f65
BY		 2417 if (set_bad && test_bit(In_sync, &rdev->flags)
2418 && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2419 retry = 1;
ba22dcbf 2420 if (retry)
3f9e7c14 2421 if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) {
2422 set_bit(R5_ReadError, &sh->dev[i].flags);
2423 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2424 } else
2425 set_bit(R5_ReadNoMerge, &sh->dev[i].flags);
ba22dcbf 2426 else {
4e5314b5
N		 2427 clear_bit(R5_ReadError, &sh->dev[i].flags);
2428 clear_bit(R5_ReWrite, &sh->dev[i].flags);
2e8ac303 2429 if (!(set_bad
2430 && test_bit(In_sync, &rdev->flags)
2431 && rdev_set_badblocks(
2432 rdev, sh->sector, STRIPE_SECTORS, 0)))
2433 md_error(conf->mddev, rdev);
ba22dcbf 2434 }
1da177e4 2435 }
14a75d3e 2436 rdev_dec_pending(rdev, conf->mddev);
c9445555 2437 bio_reset(bi);
1da177e4
LT		 2438 clear_bit(R5_LOCKED, &sh->dev[i].flags);
2439 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 2440 raid5_release_stripe(sh);
1da177e4
LT		 2441}
2442
4246a0b6 2443static void raid5_end_write_request(struct bio *bi)
1da177e4 2444{
99c0fb5f 2445 struct stripe_head *sh = bi->bi_private;
d1688a6d 2446 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 2447 int disks = sh->disks, i;
977df362 2448 struct md_rdev *uninitialized_var(rdev);
b84db560
N		 2449 sector_t first_bad;
2450 int bad_sectors;
977df362 2451 int replacement = 0;
1da177e4 2452
977df362
N		 2453 for (i = 0 ; i < disks; i++) {
2454 if (bi == &sh->dev[i].req) {
2455 rdev = conf->disks[i].rdev;
1da177e4 2456 break;
977df362
N		 2457 }
2458 if (bi == &sh->dev[i].rreq) {
2459 rdev = conf->disks[i].replacement;
dd054fce
N		 2460 if (rdev)
2461 replacement = 1;
2462 else
2463 /* rdev was removed and 'replacement'
2464 * replaced it. rdev is not removed
2465 * until all requests are finished.
2466 */
2467 rdev = conf->disks[i].rdev;
977df362
N		 2468 break;
2469 }
2470 }
4246a0b6 2471 pr_debug("end_write_request %llu/%d, count %d, error: %d.\n",
1da177e4 2472 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
4246a0b6 2473 bi->bi_error);
1da177e4 2474 if (i == disks) {
5f9d1fde 2475 bio_reset(bi);
1da177e4 2476 BUG();
6712ecf8 2477 return;
1da177e4
LT		 2478 }
2479
977df362 2480 if (replacement) {
4246a0b6 2481 if (bi->bi_error)
977df362
N		 2482 md_error(conf->mddev, rdev);
2483 else if (is_badblock(rdev, sh->sector,
2484 STRIPE_SECTORS,
2485 &first_bad, &bad_sectors))
2486 set_bit(R5_MadeGoodRepl, &sh->dev[i].flags);
2487 } else {
4246a0b6 2488 if (bi->bi_error) {
9f97e4b1 2489 set_bit(STRIPE_DEGRADED, &sh->state);
977df362
N		 2490 set_bit(WriteErrorSeen, &rdev->flags);
2491 set_bit(R5_WriteError, &sh->dev[i].flags);
3a6de292
N		 2492 if (!test_and_set_bit(WantReplacement, &rdev->flags))
2493 set_bit(MD_RECOVERY_NEEDED,
2494 &rdev->mddev->recovery);
977df362
N		 2495 } else if (is_badblock(rdev, sh->sector,
2496 STRIPE_SECTORS,
c0b32972 2497 &first_bad, &bad_sectors)) {
977df362 2498 set_bit(R5_MadeGood, &sh->dev[i].flags);
c0b32972
N		 2499 if (test_bit(R5_ReadError, &sh->dev[i].flags))
2500 /* That was a successful write so make
2501 * sure it looks like we already did
2502 * a re-write.
2503 */
2504 set_bit(R5_ReWrite, &sh->dev[i].flags);
2505 }
977df362
N		 2506 }
2507 rdev_dec_pending(rdev, conf->mddev);
1da177e4 2508
4246a0b6 2509 if (sh->batch_head && bi->bi_error && !replacement)
72ac7330 2510 set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state);
2511
c9445555 2512 bio_reset(bi);
977df362
N		 2513 if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags))
2514 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1da177e4 2515 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 2516 raid5_release_stripe(sh);
59fc630b 2517
2518 if (sh->batch_head && sh != sh->batch_head)
6d036f7d 2519 raid5_release_stripe(sh->batch_head);
1da177e4
LT		 2520}
2521
784052ec 2522static void raid5_build_block(struct stripe_head *sh, int i, int previous)
1da177e4
LT		 2523{
2524 struct r5dev *dev = &sh->dev[i];
2525
1da177e4 2526 dev->flags = 0;
6d036f7d 2527 dev->sector = raid5_compute_blocknr(sh, i, previous);
1da177e4
LT		 2528}
2529
849674e4 2530static void raid5_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 2531{
2532 char b[BDEVNAME_SIZE];
d1688a6d 2533 struct r5conf *conf = mddev->private;
908f4fbd 2534 unsigned long flags;
0c55e022 2535 pr_debug("raid456: error called\n");
1da177e4 2536
908f4fbd
N		 2537 spin_lock_irqsave(&conf->device_lock, flags);
2538 clear_bit(In_sync, &rdev->flags);
2539 mddev->degraded = calc_degraded(conf);
2540 spin_unlock_irqrestore(&conf->device_lock, flags);
2541 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2542
de393cde 2543 set_bit(Blocked, &rdev->flags);
6f8d0c77 2544 set_bit(Faulty, &rdev->flags);
2953079c
SL		 2545 set_mask_bits(&mddev->sb_flags, 0,
2546 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING));
cc6167b4
N		 2547 pr_crit("md/raid:%s: Disk failure on %s, disabling device.\n"
2548 "md/raid:%s: Operation continuing on %d devices.\n",
2549 mdname(mddev),
2550 bdevname(rdev->bdev, b),
2551 mdname(mddev),
2552 conf->raid_disks - mddev->degraded);
16a53ecc 2553}
1da177e4
LT		 2554
2555/*
2556 * Input: a 'big' sector number,
2557 * Output: index of the data and parity disk, and the sector # in them.
2558 */
6d036f7d
SL		 2559sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector,
2560 int previous, int *dd_idx,
2561 struct stripe_head *sh)
1da177e4 2562{
6e3b96ed 2563 sector_t stripe, stripe2;
35f2a591 2564 sector_t chunk_number;
1da177e4 2565 unsigned int chunk_offset;
911d4ee8 2566 int pd_idx, qd_idx;
67cc2b81 2567 int ddf_layout = 0;
1da177e4 2568 sector_t new_sector;
e183eaed
N		 2569 int algorithm = previous ? conf->prev_algo
2570 : conf->algorithm;
09c9e5fa
AN		 2571 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2572 : conf->chunk_sectors;
112bf897
N		 2573 int raid_disks = previous ? conf->previous_raid_disks
2574 : conf->raid_disks;
2575 int data_disks = raid_disks - conf->max_degraded;
1da177e4
LT		 2576
2577 /* First compute the information on this sector */
2578
2579 /*
2580 * Compute the chunk number and the sector offset inside the chunk
2581 */
2582 chunk_offset = sector_div(r_sector, sectors_per_chunk);
2583 chunk_number = r_sector;
1da177e4
LT		 2584
2585 /*
2586 * Compute the stripe number
2587 */
35f2a591
N		 2588 stripe = chunk_number;
2589 *dd_idx = sector_div(stripe, data_disks);
6e3b96ed 2590 stripe2 = stripe;
1da177e4
LT		 2591 /*
2592 * Select the parity disk based on the user selected algorithm.
2593 */
84789554 2594 pd_idx = qd_idx = -1;
16a53ecc
N		 2595 switch(conf->level) {
2596 case 4:
911d4ee8 2597 pd_idx = data_disks;
16a53ecc
N		 2598 break;
2599 case 5:
e183eaed 2600 switch (algorithm) {
1da177e4 2601 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2602 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2603 if (*dd_idx >= pd_idx)
1da177e4
LT		 2604 (*dd_idx)++;
2605 break;
2606 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2607 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2608 if (*dd_idx >= pd_idx)
1da177e4
LT		 2609 (*dd_idx)++;
2610 break;
2611 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2612 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2613 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4
LT		 2614 break;
2615 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2616 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2617 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4 2618 break;
99c0fb5f
N		 2619 case ALGORITHM_PARITY_0:
2620 pd_idx = 0;
2621 (*dd_idx)++;
2622 break;
2623 case ALGORITHM_PARITY_N:
2624 pd_idx = data_disks;
2625 break;
1da177e4 2626 default:
99c0fb5f 2627 BUG();
16a53ecc
N		 2628 }
2629 break;
2630 case 6:
2631
e183eaed 2632 switch (algorithm) {
16a53ecc 2633 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2634 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2635 qd_idx = pd_idx + 1;
2636 if (pd_idx == raid_disks-1) {
99c0fb5f 2637 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2638 qd_idx = 0;
2639 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2640 (*dd_idx) += 2; /* D D P Q D */
2641 break;
2642 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2643 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2644 qd_idx = pd_idx + 1;
2645 if (pd_idx == raid_disks-1) {
99c0fb5f 2646 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2647 qd_idx = 0;
2648 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2649 (*dd_idx) += 2; /* D D P Q D */
2650 break;
2651 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2652 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2653 qd_idx = (pd_idx + 1) % raid_disks;
2654 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc
N		 2655 break;
2656 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2657 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2658 qd_idx = (pd_idx + 1) % raid_disks;
2659 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc 2660 break;
99c0fb5f
N		 2661
2662 case ALGORITHM_PARITY_0:
2663 pd_idx = 0;
2664 qd_idx = 1;
2665 (*dd_idx) += 2;
2666 break;
2667 case ALGORITHM_PARITY_N:
2668 pd_idx = data_disks;
2669 qd_idx = data_disks + 1;
2670 break;
2671
2672 case ALGORITHM_ROTATING_ZERO_RESTART:
2673 /* Exactly the same as RIGHT_ASYMMETRIC, but or
2674 * of blocks for computing Q is different.
2675 */
6e3b96ed 2676 pd_idx = sector_div(stripe2, raid_disks);
99c0fb5f
N		 2677 qd_idx = pd_idx + 1;
2678 if (pd_idx == raid_disks-1) {
2679 (*dd_idx)++; /* Q D D D P */
2680 qd_idx = 0;
2681 } else if (*dd_idx >= pd_idx)
2682 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2683 ddf_layout = 1;
99c0fb5f
N		 2684 break;
2685
2686 case ALGORITHM_ROTATING_N_RESTART:
2687 /* Same a left_asymmetric, by first stripe is
2688 * D D D P Q rather than
2689 * Q D D D P
2690 */
6e3b96ed
N		 2691 stripe2 += 1;
2692 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2693 qd_idx = pd_idx + 1;
2694 if (pd_idx == raid_disks-1) {
2695 (*dd_idx)++; /* Q D D D P */
2696 qd_idx = 0;
2697 } else if (*dd_idx >= pd_idx)
2698 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2699 ddf_layout = 1;
99c0fb5f
N		 2700 break;
2701
2702 case ALGORITHM_ROTATING_N_CONTINUE:
2703 /* Same as left_symmetric but Q is before P */
6e3b96ed 2704 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2705 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
2706 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
67cc2b81 2707 ddf_layout = 1;
99c0fb5f
N		 2708 break;
2709
2710 case ALGORITHM_LEFT_ASYMMETRIC_6:
2711 /* RAID5 left_asymmetric, with Q on last device */
6e3b96ed 2712 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2713 if (*dd_idx >= pd_idx)
2714 (*dd_idx)++;
2715 qd_idx = raid_disks - 1;
2716 break;
2717
2718 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6e3b96ed 2719 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2720 if (*dd_idx >= pd_idx)
2721 (*dd_idx)++;
2722 qd_idx = raid_disks - 1;
2723 break;
2724
2725 case ALGORITHM_LEFT_SYMMETRIC_6:
6e3b96ed 2726 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2727 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2728 qd_idx = raid_disks - 1;
2729 break;
2730
2731 case ALGORITHM_RIGHT_SYMMETRIC_6:
6e3b96ed 2732 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2733 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2734 qd_idx = raid_disks - 1;
2735 break;
2736
2737 case ALGORITHM_PARITY_0_6:
2738 pd_idx = 0;
2739 (*dd_idx)++;
2740 qd_idx = raid_disks - 1;
2741 break;
2742
16a53ecc 2743 default:
99c0fb5f 2744 BUG();
16a53ecc
N		 2745 }
2746 break;
1da177e4
LT		 2747 }
2748
911d4ee8
N		 2749 if (sh) {
2750 sh->pd_idx = pd_idx;
2751 sh->qd_idx = qd_idx;
67cc2b81 2752 sh->ddf_layout = ddf_layout;
911d4ee8 2753 }
1da177e4
LT		 2754 /*
2755 * Finally, compute the new sector number
2756 */
2757 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
2758 return new_sector;
2759}
2760
6d036f7d 2761sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous)
1da177e4 2762{
d1688a6d 2763 struct r5conf *conf = sh->raid_conf;
b875e531
N		 2764 int raid_disks = sh->disks;
2765 int data_disks = raid_disks - conf->max_degraded;
1da177e4 2766 sector_t new_sector = sh->sector, check;
09c9e5fa
AN		 2767 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2768 : conf->chunk_sectors;
e183eaed
N		 2769 int algorithm = previous ? conf->prev_algo
2770 : conf->algorithm;
1da177e4
LT		 2771 sector_t stripe;
2772 int chunk_offset;
35f2a591
N		 2773 sector_t chunk_number;
2774 int dummy1, dd_idx = i;
1da177e4 2775 sector_t r_sector;
911d4ee8 2776 struct stripe_head sh2;
1da177e4
LT		 2777
2778 chunk_offset = sector_div(new_sector, sectors_per_chunk);
2779 stripe = new_sector;
1da177e4 2780
16a53ecc
N		 2781 if (i == sh->pd_idx)
2782 return 0;
2783 switch(conf->level) {
2784 case 4: break;
2785 case 5:
e183eaed 2786 switch (algorithm) {
1da177e4
LT		 2787 case ALGORITHM_LEFT_ASYMMETRIC:
2788 case ALGORITHM_RIGHT_ASYMMETRIC:
2789 if (i > sh->pd_idx)
2790 i--;
2791 break;
2792 case ALGORITHM_LEFT_SYMMETRIC:
2793 case ALGORITHM_RIGHT_SYMMETRIC:
2794 if (i < sh->pd_idx)
2795 i += raid_disks;
2796 i -= (sh->pd_idx + 1);
2797 break;
99c0fb5f
N		 2798 case ALGORITHM_PARITY_0:
2799 i -= 1;
2800 break;
2801 case ALGORITHM_PARITY_N:
2802 break;
1da177e4 2803 default:
99c0fb5f 2804 BUG();
16a53ecc
N		 2805 }
2806 break;
2807 case 6:
d0dabf7e 2808 if (i == sh->qd_idx)
16a53ecc 2809 return 0; /* It is the Q disk */
e183eaed 2810 switch (algorithm) {
16a53ecc
N		 2811 case ALGORITHM_LEFT_ASYMMETRIC:
2812 case ALGORITHM_RIGHT_ASYMMETRIC:
99c0fb5f
N		 2813 case ALGORITHM_ROTATING_ZERO_RESTART:
2814 case ALGORITHM_ROTATING_N_RESTART:
2815 if (sh->pd_idx == raid_disks-1)
2816 i--; /* Q D D D P */
16a53ecc
N		 2817 else if (i > sh->pd_idx)
2818 i -= 2; /* D D P Q D */
2819 break;
2820 case ALGORITHM_LEFT_SYMMETRIC:
2821 case ALGORITHM_RIGHT_SYMMETRIC:
2822 if (sh->pd_idx == raid_disks-1)
2823 i--; /* Q D D D P */
2824 else {
2825 /* D D P Q D */
2826 if (i < sh->pd_idx)
2827 i += raid_disks;
2828 i -= (sh->pd_idx + 2);
2829 }
2830 break;
99c0fb5f
N		 2831 case ALGORITHM_PARITY_0:
2832 i -= 2;
2833 break;
2834 case ALGORITHM_PARITY_N:
2835 break;
2836 case ALGORITHM_ROTATING_N_CONTINUE:
e4424fee 2837 /* Like left_symmetric, but P is before Q */
99c0fb5f
N		 2838 if (sh->pd_idx == 0)
2839 i--; /* P D D D Q */
e4424fee
N		 2840 else {
2841 /* D D Q P D */
2842 if (i < sh->pd_idx)
2843 i += raid_disks;
2844 i -= (sh->pd_idx + 1);
2845 }
99c0fb5f
N		 2846 break;
2847 case ALGORITHM_LEFT_ASYMMETRIC_6:
2848 case ALGORITHM_RIGHT_ASYMMETRIC_6:
2849 if (i > sh->pd_idx)
2850 i--;
2851 break;
2852 case ALGORITHM_LEFT_SYMMETRIC_6:
2853 case ALGORITHM_RIGHT_SYMMETRIC_6:
2854 if (i < sh->pd_idx)
2855 i += data_disks + 1;
2856 i -= (sh->pd_idx + 1);
2857 break;
2858 case ALGORITHM_PARITY_0_6:
2859 i -= 1;
2860 break;
16a53ecc 2861 default:
99c0fb5f 2862 BUG();
16a53ecc
N		 2863 }
2864 break;
1da177e4
LT		 2865 }
2866
2867 chunk_number = stripe * data_disks + i;
35f2a591 2868 r_sector = chunk_number * sectors_per_chunk + chunk_offset;
1da177e4 2869
112bf897 2870 check = raid5_compute_sector(conf, r_sector,
784052ec 2871 previous, &dummy1, &sh2);
911d4ee8
N		 2872 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
2873 || sh2.qd_idx != sh->qd_idx) {
cc6167b4
N		 2874 pr_warn("md/raid:%s: compute_blocknr: map not correct\n",
2875 mdname(conf->mddev));
1da177e4
LT		 2876 return 0;
2877 }
2878 return r_sector;
2879}
2880
600aa109 2881static void
c0f7bddb 2882schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
600aa109 2883 int rcw, int expand)
e33129d8 2884{
584acdd4 2885 int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx, disks = sh->disks;
d1688a6d 2886 struct r5conf *conf = sh->raid_conf;
c0f7bddb 2887 int level = conf->level;
e33129d8
DW		 2888
2889 if (rcw) {
1e6d690b
SL		 2890 /*
2891 * In some cases, handle_stripe_dirtying initially decided to
2892 * run rmw and allocates extra page for prexor. However, rcw is
2893 * cheaper later on. We need to free the extra page now,
2894 * because we won't be able to do that in ops_complete_prexor().
2895 */
2896 r5c_release_extra_page(sh);
e33129d8
DW		 2897
2898 for (i = disks; i--; ) {
2899 struct r5dev *dev = &sh->dev[i];
2900
2901 if (dev->towrite) {
2902 set_bit(R5_LOCKED, &dev->flags);
d8ee0728 2903 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2904 if (!expand)
2905 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2906 s->locked++;
1e6d690b
SL		 2907 } else if (test_bit(R5_InJournal, &dev->flags)) {
2908 set_bit(R5_LOCKED, &dev->flags);
2909 s->locked++;
e33129d8
DW		 2910 }
2911 }
ce7d363a
N		 2912 /* if we are not expanding this is a proper write request, and
2913 * there will be bios with new data to be drained into the
2914 * stripe cache
2915 */
2916 if (!expand) {
2917 if (!s->locked)
2918 /* False alarm, nothing to do */
2919 return;
2920 sh->reconstruct_state = reconstruct_state_drain_run;
2921 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2922 } else
2923 sh->reconstruct_state = reconstruct_state_run;
2924
2925 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
2926
c0f7bddb 2927 if (s->locked + conf->max_degraded == disks)
8b3e6cdc 2928 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
c0f7bddb 2929 atomic_inc(&conf->pending_full_writes);
e33129d8
DW		 2930 } else {
2931 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
2932 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
584acdd4
MS		 2933 BUG_ON(level == 6 &&
2934 (!(test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags) ||
2935 test_bit(R5_Wantcompute, &sh->dev[qd_idx].flags))));
e33129d8 2936
e33129d8
DW		 2937 for (i = disks; i--; ) {
2938 struct r5dev *dev = &sh->dev[i];
584acdd4 2939 if (i == pd_idx || i == qd_idx)
e33129d8
DW		 2940 continue;
2941
e33129d8
DW		 2942 if (dev->towrite &&
2943 (test_bit(R5_UPTODATE, &dev->flags) ||
d8ee0728
DW		 2944 test_bit(R5_Wantcompute, &dev->flags))) {
2945 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2946 set_bit(R5_LOCKED, &dev->flags);
2947 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2948 s->locked++;
1e6d690b
SL		 2949 } else if (test_bit(R5_InJournal, &dev->flags)) {
2950 set_bit(R5_LOCKED, &dev->flags);
2951 s->locked++;
e33129d8
DW		 2952 }
2953 }
ce7d363a
N		 2954 if (!s->locked)
2955 /* False alarm - nothing to do */
2956 return;
2957 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
2958 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
2959 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2960 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
e33129d8
DW		 2961 }
2962
c0f7bddb 2963 /* keep the parity disk(s) locked while asynchronous operations
e33129d8
DW		 2964 * are in flight
2965 */
2966 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
2967 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
600aa109 2968 s->locked++;
e33129d8 2969
c0f7bddb
YT		 2970 if (level == 6) {
2971 int qd_idx = sh->qd_idx;
2972 struct r5dev *dev = &sh->dev[qd_idx];
2973
2974 set_bit(R5_LOCKED, &dev->flags);
2975 clear_bit(R5_UPTODATE, &dev->flags);
2976 s->locked++;
2977 }
2978
600aa109 2979 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
e46b272b 2980 __func__, (unsigned long long)sh->sector,
600aa109 2981 s->locked, s->ops_request);
e33129d8 2982}
16a53ecc 2983
1da177e4
LT		 2984/*
2985 * Each stripe/dev can have one or more bion attached.
16a53ecc 2986 * toread/towrite point to the first in a chain.
1da177e4
LT		 2987 * The bi_next chain must be in order.
2988 */
da41ba65 2989static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx,
2990 int forwrite, int previous)
1da177e4
LT		 2991{
2992 struct bio **bip;
d1688a6d 2993 struct r5conf *conf = sh->raid_conf;
72626685 2994 int firstwrite=0;
1da177e4 2995
cbe47ec5 2996 pr_debug("adding bi b#%llu to stripe s#%llu\n",
4f024f37 2997 (unsigned long long)bi->bi_iter.bi_sector,
1da177e4
LT		 2998 (unsigned long long)sh->sector);
2999
b17459c0
SL		 3000 /*
3001 * If several bio share a stripe. The bio bi_phys_segments acts as a
3002 * reference count to avoid race. The reference count should already be
3003 * increased before this function is called (for example, in
849674e4 3004 * raid5_make_request()), so other bio sharing this stripe will not free the
b17459c0
SL		 3005 * stripe. If a stripe is owned by one stripe, the stripe lock will
3006 * protect it.
3007 */
3008 spin_lock_irq(&sh->stripe_lock);
59fc630b 3009 /* Don't allow new IO added to stripes in batch list */
3010 if (sh->batch_head)
3011 goto overlap;
72626685 3012 if (forwrite) {
1da177e4 3013 bip = &sh->dev[dd_idx].towrite;
7eaf7e8e 3014 if (*bip == NULL)
72626685
N		 3015 firstwrite = 1;
3016 } else
1da177e4 3017 bip = &sh->dev[dd_idx].toread;
4f024f37
KO		 3018 while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) {
3019 if (bio_end_sector(*bip) > bi->bi_iter.bi_sector)
1da177e4
LT		 3020 goto overlap;
3021 bip = & (*bip)->bi_next;
3022 }
4f024f37 3023 if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi))
1da177e4
LT		 3024 goto overlap;
3025
da41ba65 3026 if (!forwrite || previous)
3027 clear_bit(STRIPE_BATCH_READY, &sh->state);
3028
78bafebd 3029 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
1da177e4
LT		 3030 if (*bip)
3031 bi->bi_next = *bip;
3032 *bip = bi;
e7836bd6 3033 raid5_inc_bi_active_stripes(bi);
72626685 3034
1da177e4
LT		 3035 if (forwrite) {
3036 /* check if page is covered */
3037 sector_t sector = sh->dev[dd_idx].sector;
3038 for (bi=sh->dev[dd_idx].towrite;
3039 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
4f024f37 3040 bi && bi->bi_iter.bi_sector <= sector;
1da177e4 3041 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
f73a1c7d
KO		 3042 if (bio_end_sector(bi) >= sector)
3043 sector = bio_end_sector(bi);
1da177e4
LT		 3044 }
3045 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
7a87f434 3046 if (!test_and_set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags))
3047 sh->overwrite_disks++;
1da177e4 3048 }
cbe47ec5
N		 3049
3050 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
4f024f37 3051 (unsigned long long)(*bip)->bi_iter.bi_sector,
cbe47ec5
N		 3052 (unsigned long long)sh->sector, dd_idx);
3053
3054 if (conf->mddev->bitmap && firstwrite) {
d0852df5
N		 3055 /* Cannot hold spinlock over bitmap_startwrite,
3056 * but must ensure this isn't added to a batch until
3057 * we have added to the bitmap and set bm_seq.
3058 * So set STRIPE_BITMAP_PENDING to prevent
3059 * batching.
3060 * If multiple add_stripe_bio() calls race here they
3061 * much all set STRIPE_BITMAP_PENDING. So only the first one
3062 * to complete "bitmap_startwrite" gets to set
3063 * STRIPE_BIT_DELAY. This is important as once a stripe
3064 * is added to a batch, STRIPE_BIT_DELAY cannot be changed
3065 * any more.
3066 */
3067 set_bit(STRIPE_BITMAP_PENDING, &sh->state);
3068 spin_unlock_irq(&sh->stripe_lock);
cbe47ec5
N		 3069 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
3070 STRIPE_SECTORS, 0);
d0852df5
N		 3071 spin_lock_irq(&sh->stripe_lock);
3072 clear_bit(STRIPE_BITMAP_PENDING, &sh->state);
3073 if (!sh->batch_head) {
3074 sh->bm_seq = conf->seq_flush+1;
3075 set_bit(STRIPE_BIT_DELAY, &sh->state);
3076 }
cbe47ec5 3077 }
d0852df5 3078 spin_unlock_irq(&sh->stripe_lock);
59fc630b 3079
3080 if (stripe_can_batch(sh))
3081 stripe_add_to_batch_list(conf, sh);
1da177e4
LT		 3082 return 1;
3083
3084 overlap:
3085 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
b17459c0 3086 spin_unlock_irq(&sh->stripe_lock);
1da177e4
LT		 3087 return 0;
3088}
3089
d1688a6d 3090static void end_reshape(struct r5conf *conf);
29269553 3091
d1688a6d 3092static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
911d4ee8 3093 struct stripe_head *sh)
ccfcc3c1 3094{
784052ec 3095 int sectors_per_chunk =
09c9e5fa 3096 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
911d4ee8 3097 int dd_idx;
2d2063ce 3098 int chunk_offset = sector_div(stripe, sectors_per_chunk);
112bf897 3099 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
2d2063ce 3100
112bf897
N		 3101 raid5_compute_sector(conf,
3102 stripe * (disks - conf->max_degraded)
b875e531 3103 *sectors_per_chunk + chunk_offset,
112bf897 3104 previous,
911d4ee8 3105 &dd_idx, sh);
ccfcc3c1
N		 3106}
3107
a4456856 3108static void
d1688a6d 3109handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh,
a4456856 3110 struct stripe_head_state *s, int disks,
34a6f80e 3111 struct bio_list *return_bi)
a4456856
DW		 3112{
3113 int i;
59fc630b 3114 BUG_ON(sh->batch_head);
a4456856
DW		 3115 for (i = disks; i--; ) {
3116 struct bio *bi;
3117 int bitmap_end = 0;
3118
3119 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
3cb03002 3120 struct md_rdev *rdev;
a4456856
DW		 3121 rcu_read_lock();
3122 rdev = rcu_dereference(conf->disks[i].rdev);
f5b67ae8
N		 3123 if (rdev && test_bit(In_sync, &rdev->flags) &&
3124 !test_bit(Faulty, &rdev->flags))
7f0da59b
N		 3125 atomic_inc(&rdev->nr_pending);
3126 else
3127 rdev = NULL;
a4456856 3128 rcu_read_unlock();
7f0da59b
N		 3129 if (rdev) {
3130 if (!rdev_set_badblocks(
3131 rdev,
3132 sh->sector,
3133 STRIPE_SECTORS, 0))
3134 md_error(conf->mddev, rdev);
3135 rdev_dec_pending(rdev, conf->mddev);
3136 }
a4456856 3137 }
b17459c0 3138 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 3139 /* fail all writes first */
3140 bi = sh->dev[i].towrite;
3141 sh->dev[i].towrite = NULL;
7a87f434 3142 sh->overwrite_disks = 0;
b17459c0 3143 spin_unlock_irq(&sh->stripe_lock);
1ed850f3 3144 if (bi)
a4456856 3145 bitmap_end = 1;
a4456856 3146
0576b1c6
SL		 3147 r5l_stripe_write_finished(sh);
3148
a4456856
DW		 3149 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
3150 wake_up(&conf->wait_for_overlap);
3151
4f024f37 3152 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3153 sh->dev[i].sector + STRIPE_SECTORS) {
3154 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
4246a0b6
CH		 3155
3156 bi->bi_error = -EIO;
e7836bd6 3157 if (!raid5_dec_bi_active_stripes(bi)) {
a4456856 3158 md_write_end(conf->mddev);
34a6f80e 3159 bio_list_add(return_bi, bi);
a4456856
DW		 3160 }
3161 bi = nextbi;
3162 }
7eaf7e8e
SL		 3163 if (bitmap_end)
3164 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3165 STRIPE_SECTORS, 0, 0);
3166 bitmap_end = 0;
a4456856
DW		 3167 /* and fail all 'written' */
3168 bi = sh->dev[i].written;
3169 sh->dev[i].written = NULL;
d592a996
SL		 3170 if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) {
3171 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
3172 sh->dev[i].page = sh->dev[i].orig_page;
3173 }
3174
a4456856 3175 if (bi) bitmap_end = 1;
4f024f37 3176 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3177 sh->dev[i].sector + STRIPE_SECTORS) {
3178 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
4246a0b6
CH		 3179
3180 bi->bi_error = -EIO;
e7836bd6 3181 if (!raid5_dec_bi_active_stripes(bi)) {
a4456856 3182 md_write_end(conf->mddev);
34a6f80e 3183 bio_list_add(return_bi, bi);
a4456856
DW		 3184 }
3185 bi = bi2;
3186 }
3187
b5e98d65
DW		 3188 /* fail any reads if this device is non-operational and
3189 * the data has not reached the cache yet.
3190 */
3191 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
6e74a9cf 3192 s->failed > conf->max_degraded &&
b5e98d65
DW		 3193 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
3194 test_bit(R5_ReadError, &sh->dev[i].flags))) {
143c4d05 3195 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 3196 bi = sh->dev[i].toread;
3197 sh->dev[i].toread = NULL;
143c4d05 3198 spin_unlock_irq(&sh->stripe_lock);
a4456856
DW		 3199 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
3200 wake_up(&conf->wait_for_overlap);
ebda780b
SL		 3201 if (bi)
3202 s->to_read--;
4f024f37 3203 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3204 sh->dev[i].sector + STRIPE_SECTORS) {
3205 struct bio *nextbi =
3206 r5_next_bio(bi, sh->dev[i].sector);
4246a0b6
CH		 3207
3208 bi->bi_error = -EIO;
34a6f80e
N		 3209 if (!raid5_dec_bi_active_stripes(bi))
3210 bio_list_add(return_bi, bi);
a4456856
DW		 3211 bi = nextbi;
3212 }
3213 }
a4456856
DW		 3214 if (bitmap_end)
3215 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3216 STRIPE_SECTORS, 0, 0);
8cfa7b0f
N		 3217 /* If we were in the middle of a write the parity block might
3218 * still be locked - so just clear all R5_LOCKED flags
3219 */
3220 clear_bit(R5_LOCKED, &sh->dev[i].flags);
a4456856 3221 }
ebda780b
SL		 3222 s->to_write = 0;
3223 s->written = 0;
a4456856 3224
8b3e6cdc
DW		 3225 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3226 if (atomic_dec_and_test(&conf->pending_full_writes))
3227 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 3228}
3229
7f0da59b 3230static void
d1688a6d 3231handle_failed_sync(struct r5conf *conf, struct stripe_head *sh,
7f0da59b
N		 3232 struct stripe_head_state *s)
3233{
3234 int abort = 0;
3235 int i;
3236
59fc630b 3237 BUG_ON(sh->batch_head);
7f0da59b 3238 clear_bit(STRIPE_SYNCING, &sh->state);
f8dfcffd
N		 3239 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
3240 wake_up(&conf->wait_for_overlap);
7f0da59b 3241 s->syncing = 0;
9a3e1101 3242 s->replacing = 0;
7f0da59b 3243 /* There is nothing more to do for sync/check/repair.
18b9837e
N		 3244 * Don't even need to abort as that is handled elsewhere
3245 * if needed, and not always wanted e.g. if there is a known
3246 * bad block here.
9a3e1101 3247 * For recover/replace we need to record a bad block on all
7f0da59b
N		 3248 * non-sync devices, or abort the recovery
3249 */
18b9837e
N		 3250 if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) {
3251 /* During recovery devices cannot be removed, so
3252 * locking and refcounting of rdevs is not needed
3253 */
e50d3992 3254 rcu_read_lock();
18b9837e 3255 for (i = 0; i < conf->raid_disks; i++) {
e50d3992 3256 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
18b9837e
N		 3257 if (rdev
3258 && !test_bit(Faulty, &rdev->flags)
3259 && !test_bit(In_sync, &rdev->flags)
3260 && !rdev_set_badblocks(rdev, sh->sector,
3261 STRIPE_SECTORS, 0))
3262 abort = 1;
e50d3992 3263 rdev = rcu_dereference(conf->disks[i].replacement);
18b9837e
N		 3264 if (rdev
3265 && !test_bit(Faulty, &rdev->flags)
3266 && !test_bit(In_sync, &rdev->flags)
3267 && !rdev_set_badblocks(rdev, sh->sector,
3268 STRIPE_SECTORS, 0))
3269 abort = 1;
3270 }
e50d3992 3271 rcu_read_unlock();
18b9837e
N		 3272 if (abort)
3273 conf->recovery_disabled =
3274 conf->mddev->recovery_disabled;
7f0da59b 3275 }
18b9837e 3276 md_done_sync(conf->mddev, STRIPE_SECTORS, !abort);
7f0da59b
N		 3277}
3278
9a3e1101
N		 3279static int want_replace(struct stripe_head *sh, int disk_idx)
3280{
3281 struct md_rdev *rdev;
3282 int rv = 0;
3f232d6a
N		 3283
3284 rcu_read_lock();
3285 rdev = rcu_dereference(sh->raid_conf->disks[disk_idx].replacement);
9a3e1101
N		 3286 if (rdev
3287 && !test_bit(Faulty, &rdev->flags)
3288 && !test_bit(In_sync, &rdev->flags)
3289 && (rdev->recovery_offset <= sh->sector
3290 || rdev->mddev->recovery_cp <= sh->sector))
3291 rv = 1;
3f232d6a 3292 rcu_read_unlock();
9a3e1101
N		 3293 return rv;
3294}
3295
93b3dbce 3296/* fetch_block - checks the given member device to see if its data needs
1fe797e6
DW		 3297 * to be read or computed to satisfy a request.
3298 *
3299 * Returns 1 when no more member devices need to be checked, otherwise returns
93b3dbce 3300 * 0 to tell the loop in handle_stripe_fill to continue
f38e1219 3301 */
2c58f06e
N		 3302
3303static int need_this_block(struct stripe_head *sh, struct stripe_head_state *s,
3304 int disk_idx, int disks)
a4456856 3305{
5599becc 3306 struct r5dev *dev = &sh->dev[disk_idx];
f2b3b44d
N		 3307 struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]],
3308 &sh->dev[s->failed_num[1]] };
ea664c82 3309 int i;
5599becc 3310
a79cfe12
N		 3311
3312 if (test_bit(R5_LOCKED, &dev->flags) ||
3313 test_bit(R5_UPTODATE, &dev->flags))
3314 /* No point reading this as we already have it or have
3315 * decided to get it.
3316 */
3317 return 0;
3318
3319 if (dev->toread ||
3320 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)))
3321 /* We need this block to directly satisfy a request */
3322 return 1;
3323
3324 if (s->syncing || s->expanding ||
3325 (s->replacing && want_replace(sh, disk_idx)))
3326 /* When syncing, or expanding we read everything.
3327 * When replacing, we need the replaced block.
3328 */
3329 return 1;
3330
3331 if ((s->failed >= 1 && fdev[0]->toread) ||
3332 (s->failed >= 2 && fdev[1]->toread))
3333 /* If we want to read from a failed device, then
3334 * we need to actually read every other device.
3335 */
3336 return 1;
3337
a9d56950
N		 3338 /* Sometimes neither read-modify-write nor reconstruct-write
3339 * cycles can work. In those cases we read every block we
3340 * can. Then the parity-update is certain to have enough to
3341 * work with.
3342 * This can only be a problem when we need to write something,
3343 * and some device has failed. If either of those tests
3344 * fail we need look no further.
3345 */
3346 if (!s->failed || !s->to_write)
3347 return 0;
3348
3349 if (test_bit(R5_Insync, &dev->flags) &&
3350 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3351 /* Pre-reads at not permitted until after short delay
3352 * to gather multiple requests. However if this
3353 * device is no Insync, the block could only be be computed
3354 * and there is no need to delay that.
3355 */
3356 return 0;
ea664c82 3357
36707bb2 3358 for (i = 0; i < s->failed && i < 2; i++) {
ea664c82
N		 3359 if (fdev[i]->towrite &&
3360 !test_bit(R5_UPTODATE, &fdev[i]->flags) &&
3361 !test_bit(R5_OVERWRITE, &fdev[i]->flags))
3362 /* If we have a partial write to a failed
3363 * device, then we will need to reconstruct
3364 * the content of that device, so all other
3365 * devices must be read.
3366 */
3367 return 1;
3368 }
3369
3370 /* If we are forced to do a reconstruct-write, either because
3371 * the current RAID6 implementation only supports that, or
3372 * or because parity cannot be trusted and we are currently
3373 * recovering it, there is extra need to be careful.
3374 * If one of the devices that we would need to read, because
3375 * it is not being overwritten (and maybe not written at all)
3376 * is missing/faulty, then we need to read everything we can.
3377 */
3378 if (sh->raid_conf->level != 6 &&
3379 sh->sector < sh->raid_conf->mddev->recovery_cp)
3380 /* reconstruct-write isn't being forced */
3381 return 0;
36707bb2 3382 for (i = 0; i < s->failed && i < 2; i++) {
10d82c5f
N		 3383 if (s->failed_num[i] != sh->pd_idx &&
3384 s->failed_num[i] != sh->qd_idx &&
3385 !test_bit(R5_UPTODATE, &fdev[i]->flags) &&
ea664c82
N		 3386 !test_bit(R5_OVERWRITE, &fdev[i]->flags))
3387 return 1;
3388 }
3389
2c58f06e
N		 3390 return 0;
3391}
3392
3393static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s,
3394 int disk_idx, int disks)
3395{
3396 struct r5dev *dev = &sh->dev[disk_idx];
3397
3398 /* is the data in this block needed, and can we get it? */
3399 if (need_this_block(sh, s, disk_idx, disks)) {
5599becc
YT		 3400 /* we would like to get this block, possibly by computing it,
3401 * otherwise read it if the backing disk is insync
3402 */
3403 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
3404 BUG_ON(test_bit(R5_Wantread, &dev->flags));
b0c783b3 3405 BUG_ON(sh->batch_head);
5599becc 3406 if ((s->uptodate == disks - 1) &&
f2b3b44d
N		 3407 (s->failed && (disk_idx == s->failed_num[0] ||
3408 disk_idx == s->failed_num[1]))) {
5599becc
YT		 3409 /* have disk failed, and we're requested to fetch it;
3410 * do compute it
a4456856 3411 */
5599becc
YT		 3412 pr_debug("Computing stripe %llu block %d\n",
3413 (unsigned long long)sh->sector, disk_idx);
3414 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3415 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3416 set_bit(R5_Wantcompute, &dev->flags);
3417 sh->ops.target = disk_idx;
3418 sh->ops.target2 = -1; /* no 2nd target */
3419 s->req_compute = 1;
93b3dbce
N		 3420 /* Careful: from this point on 'uptodate' is in the eye
3421 * of raid_run_ops which services 'compute' operations
3422 * before writes. R5_Wantcompute flags a block that will
3423 * be R5_UPTODATE by the time it is needed for a
3424 * subsequent operation.
3425 */
5599becc
YT		 3426 s->uptodate++;
3427 return 1;
3428 } else if (s->uptodate == disks-2 && s->failed >= 2) {
3429 /* Computing 2-failure is *very* expensive; only
3430 * do it if failed >= 2
3431 */
3432 int other;
3433 for (other = disks; other--; ) {
3434 if (other == disk_idx)
3435 continue;
3436 if (!test_bit(R5_UPTODATE,
3437 &sh->dev[other].flags))
3438 break;
a4456856 3439 }
5599becc
YT		 3440 BUG_ON(other < 0);
3441 pr_debug("Computing stripe %llu blocks %d,%d\n",
3442 (unsigned long long)sh->sector,
3443 disk_idx, other);
3444 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3445 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3446 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
3447 set_bit(R5_Wantcompute, &sh->dev[other].flags);
3448 sh->ops.target = disk_idx;
3449 sh->ops.target2 = other;
3450 s->uptodate += 2;
3451 s->req_compute = 1;
3452 return 1;
3453 } else if (test_bit(R5_Insync, &dev->flags)) {
3454 set_bit(R5_LOCKED, &dev->flags);
3455 set_bit(R5_Wantread, &dev->flags);
3456 s->locked++;
3457 pr_debug("Reading block %d (sync=%d)\n",
3458 disk_idx, s->syncing);
a4456856
DW		 3459 }
3460 }
5599becc
YT		 3461
3462 return 0;
3463}
3464
3465/**
93b3dbce 3466 * handle_stripe_fill - read or compute data to satisfy pending requests.
5599becc 3467 */
93b3dbce
N		 3468static void handle_stripe_fill(struct stripe_head *sh,
3469 struct stripe_head_state *s,
3470 int disks)
5599becc
YT		 3471{
3472 int i;
3473
3474 /* look for blocks to read/compute, skip this if a compute
3475 * is already in flight, or if the stripe contents are in the
3476 * midst of changing due to a write
3477 */
3478 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
3479 !sh->reconstruct_state)
3480 for (i = disks; i--; )
93b3dbce 3481 if (fetch_block(sh, s, i, disks))
5599becc 3482 break;
a4456856
DW		 3483 set_bit(STRIPE_HANDLE, &sh->state);
3484}
3485
787b76fa
N		 3486static void break_stripe_batch_list(struct stripe_head *head_sh,
3487 unsigned long handle_flags);
1fe797e6 3488/* handle_stripe_clean_event
a4456856
DW		 3489 * any written block on an uptodate or failed drive can be returned.
3490 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
3491 * never LOCKED, so we don't need to test 'failed' directly.
3492 */
d1688a6d 3493static void handle_stripe_clean_event(struct r5conf *conf,
34a6f80e 3494 struct stripe_head *sh, int disks, struct bio_list *return_bi)
a4456856
DW		 3495{
3496 int i;
3497 struct r5dev *dev;
f8dfcffd 3498 int discard_pending = 0;
59fc630b 3499 struct stripe_head *head_sh = sh;
3500 bool do_endio = false;
a4456856
DW		 3501
3502 for (i = disks; i--; )
3503 if (sh->dev[i].written) {
3504 dev = &sh->dev[i];
3505 if (!test_bit(R5_LOCKED, &dev->flags) &&
9e444768 3506 (test_bit(R5_UPTODATE, &dev->flags) ||
d592a996
SL		 3507 test_bit(R5_Discard, &dev->flags) ||
3508 test_bit(R5_SkipCopy, &dev->flags))) {
a4456856
DW		 3509 /* We can return any write requests */
3510 struct bio *wbi, *wbi2;
45b4233c 3511 pr_debug("Return write for disc %d\n", i);
ca64cae9
N		 3512 if (test_and_clear_bit(R5_Discard, &dev->flags))
3513 clear_bit(R5_UPTODATE, &dev->flags);
d592a996
SL		 3514 if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) {
3515 WARN_ON(test_bit(R5_UPTODATE, &dev->flags));
d592a996 3516 }
59fc630b 3517 do_endio = true;
3518
3519returnbi:
3520 dev->page = dev->orig_page;
a4456856
DW		 3521 wbi = dev->written;
3522 dev->written = NULL;
4f024f37 3523 while (wbi && wbi->bi_iter.bi_sector <
a4456856
DW		 3524 dev->sector + STRIPE_SECTORS) {
3525 wbi2 = r5_next_bio(wbi, dev->sector);
e7836bd6 3526 if (!raid5_dec_bi_active_stripes(wbi)) {
a4456856 3527 md_write_end(conf->mddev);
34a6f80e 3528 bio_list_add(return_bi, wbi);
a4456856
DW		 3529 }
3530 wbi = wbi2;
3531 }
7eaf7e8e
SL		 3532 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3533 STRIPE_SECTORS,
a4456856 3534 !test_bit(STRIPE_DEGRADED, &sh->state),
7eaf7e8e 3535 0);
59fc630b 3536 if (head_sh->batch_head) {
3537 sh = list_first_entry(&sh->batch_list,
3538 struct stripe_head,
3539 batch_list);
3540 if (sh != head_sh) {
3541 dev = &sh->dev[i];
3542 goto returnbi;
3543 }
3544 }
3545 sh = head_sh;
3546 dev = &sh->dev[i];
f8dfcffd
N		 3547 } else if (test_bit(R5_Discard, &dev->flags))
3548 discard_pending = 1;
3549 }
f6bed0ef 3550
0576b1c6
SL		 3551 r5l_stripe_write_finished(sh);
3552
f8dfcffd
N		 3553 if (!discard_pending &&
3554 test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) {
b8a9d66d 3555 int hash;
f8dfcffd
N		 3556 clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
3557 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
3558 if (sh->qd_idx >= 0) {
3559 clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
3560 clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags);
3561 }
3562 /* now that discard is done we can proceed with any sync */
3563 clear_bit(STRIPE_DISCARD, &sh->state);
d47648fc
SL		 3564 /*
3565 * SCSI discard will change some bio fields and the stripe has
3566 * no updated data, so remove it from hash list and the stripe
3567 * will be reinitialized
3568 */
59fc630b 3569unhash:
b8a9d66d
RG		 3570 hash = sh->hash_lock_index;
3571 spin_lock_irq(conf->hash_locks + hash);
d47648fc 3572 remove_hash(sh);
b8a9d66d 3573 spin_unlock_irq(conf->hash_locks + hash);
59fc630b 3574 if (head_sh->batch_head) {
3575 sh = list_first_entry(&sh->batch_list,
3576 struct stripe_head, batch_list);
3577 if (sh != head_sh)
3578 goto unhash;
3579 }
59fc630b 3580 sh = head_sh;
3581
f8dfcffd
N		 3582 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state))
3583 set_bit(STRIPE_HANDLE, &sh->state);
3584
3585 }
8b3e6cdc
DW		 3586
3587 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3588 if (atomic_dec_and_test(&conf->pending_full_writes))
3589 md_wakeup_thread(conf->mddev->thread);
59fc630b 3590
787b76fa
N		 3591 if (head_sh->batch_head && do_endio)
3592 break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS);
a4456856
DW		 3593}
3594
d7bd398e
SL		 3595static int handle_stripe_dirtying(struct r5conf *conf,
3596 struct stripe_head *sh,
3597 struct stripe_head_state *s,
3598 int disks)
a4456856
DW		 3599{
3600 int rmw = 0, rcw = 0, i;
a7854487
AL		 3601 sector_t recovery_cp = conf->mddev->recovery_cp;
3602
584acdd4 3603 /* Check whether resync is now happening or should start.
a7854487
AL		 3604 * If yes, then the array is dirty (after unclean shutdown or
3605 * initial creation), so parity in some stripes might be inconsistent.
3606 * In this case, we need to always do reconstruct-write, to ensure
3607 * that in case of drive failure or read-error correction, we
3608 * generate correct data from the parity.
3609 */
584acdd4 3610 if (conf->rmw_level == PARITY_DISABLE_RMW ||
26ac1073
N		 3611 (recovery_cp < MaxSector && sh->sector >= recovery_cp &&
3612 s->failed == 0)) {
a7854487 3613 /* Calculate the real rcw later - for now make it
c8ac1803
N		 3614 * look like rcw is cheaper
3615 */
3616 rcw = 1; rmw = 2;
584acdd4
MS		 3617 pr_debug("force RCW rmw_level=%u, recovery_cp=%llu sh->sector=%llu\n",
3618 conf->rmw_level, (unsigned long long)recovery_cp,
a7854487 3619 (unsigned long long)sh->sector);
c8ac1803 3620 } else for (i = disks; i--; ) {
a4456856
DW		 3621 /* would I have to read this buffer for read_modify_write */
3622 struct r5dev *dev = &sh->dev[i];
1e6d690b
SL		 3623 if ((dev->towrite || i == sh->pd_idx || i == sh->qd_idx ||
3624 test_bit(R5_InJournal, &dev->flags)) &&
a4456856 3625 !test_bit(R5_LOCKED, &dev->flags) &&
1e6d690b
SL		 3626 !((test_bit(R5_UPTODATE, &dev->flags) &&
3627 (!test_bit(R5_InJournal, &dev->flags) ||
3628 dev->page != dev->orig_page)) ||
f38e1219 3629 test_bit(R5_Wantcompute, &dev->flags))) {
a4456856
DW		 3630 if (test_bit(R5_Insync, &dev->flags))
3631 rmw++;
3632 else
3633 rmw += 2*disks; /* cannot read it */
3634 }
3635 /* Would I have to read this buffer for reconstruct_write */
584acdd4
MS		 3636 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
3637 i != sh->pd_idx && i != sh->qd_idx &&
a4456856 3638 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219 3639 !(test_bit(R5_UPTODATE, &dev->flags) ||
1e6d690b
SL		 3640 test_bit(R5_InJournal, &dev->flags) ||
3641 test_bit(R5_Wantcompute, &dev->flags))) {
67f45548
N		 3642 if (test_bit(R5_Insync, &dev->flags))
3643 rcw++;
a4456856
DW		 3644 else
3645 rcw += 2*disks;
3646 }
3647 }
1e6d690b 3648
45b4233c 3649 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
a4456856
DW		 3650 (unsigned long long)sh->sector, rmw, rcw);
3651 set_bit(STRIPE_HANDLE, &sh->state);
41257580 3652 if ((rmw < rcw || (rmw == rcw && conf->rmw_level == PARITY_PREFER_RMW)) && rmw > 0) {
a4456856 3653 /* prefer read-modify-write, but need to get some data */
e3620a3a
JB		 3654 if (conf->mddev->queue)
3655 blk_add_trace_msg(conf->mddev->queue,
3656 "raid5 rmw %llu %d",
3657 (unsigned long long)sh->sector, rmw);
a4456856
DW		 3658 for (i = disks; i--; ) {
3659 struct r5dev *dev = &sh->dev[i];
1e6d690b
SL		 3660 if (test_bit(R5_InJournal, &dev->flags) &&
3661 dev->page == dev->orig_page &&
3662 !test_bit(R5_LOCKED, &sh->dev[sh->pd_idx].flags)) {
3663 /* alloc page for prexor */
d7bd398e
SL		 3664 struct page *p = alloc_page(GFP_NOIO);
3665
3666 if (p) {
3667 dev->orig_page = p;
3668 continue;
3669 }
3670
3671 /*
3672 * alloc_page() failed, try use
3673 * disk_info->extra_page
3674 */
3675 if (!test_and_set_bit(R5C_EXTRA_PAGE_IN_USE,
3676 &conf->cache_state)) {
3677 r5c_use_extra_page(sh);
3678 break;
3679 }
1e6d690b 3680
d7bd398e
SL		 3681 /* extra_page in use, add to delayed_list */
3682 set_bit(STRIPE_DELAYED, &sh->state);
3683 s->waiting_extra_page = 1;
3684 return -EAGAIN;
1e6d690b 3685 }
d7bd398e 3686 }
1e6d690b 3687
d7bd398e
SL		 3688 for (i = disks; i--; ) {
3689 struct r5dev *dev = &sh->dev[i];
1e6d690b
SL		 3690 if ((dev->towrite ||
3691 i == sh->pd_idx || i == sh->qd_idx ||
3692 test_bit(R5_InJournal, &dev->flags)) &&
a4456856 3693 !test_bit(R5_LOCKED, &dev->flags) &&
1e6d690b
SL		 3694 !((test_bit(R5_UPTODATE, &dev->flags) &&
3695 (!test_bit(R5_InJournal, &dev->flags) ||
3696 dev->page != dev->orig_page)) ||
3697 test_bit(R5_Wantcompute, &dev->flags)) &&
a4456856 3698 test_bit(R5_Insync, &dev->flags)) {
67f45548
N		 3699 if (test_bit(STRIPE_PREREAD_ACTIVE,
3700 &sh->state)) {
3701 pr_debug("Read_old block %d for r-m-w\n",
3702 i);
a4456856
DW		 3703 set_bit(R5_LOCKED, &dev->flags);
3704 set_bit(R5_Wantread, &dev->flags);
3705 s->locked++;
3706 } else {
3707 set_bit(STRIPE_DELAYED, &sh->state);
3708 set_bit(STRIPE_HANDLE, &sh->state);
3709 }
3710 }
3711 }
a9add5d9 3712 }
41257580 3713 if ((rcw < rmw || (rcw == rmw && conf->rmw_level != PARITY_PREFER_RMW)) && rcw > 0) {
a4456856 3714 /* want reconstruct write, but need to get some data */
a9add5d9 3715 int qread =0;
c8ac1803 3716 rcw = 0;
a4456856
DW		 3717 for (i = disks; i--; ) {
3718 struct r5dev *dev = &sh->dev[i];
3719 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
c8ac1803 3720 i != sh->pd_idx && i != sh->qd_idx &&
a4456856 3721 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219 3722 !(test_bit(R5_UPTODATE, &dev->flags) ||
1e6d690b 3723 test_bit(R5_InJournal, &dev->flags) ||
c8ac1803
N		 3724 test_bit(R5_Wantcompute, &dev->flags))) {
3725 rcw++;
67f45548
N		 3726 if (test_bit(R5_Insync, &dev->flags) &&
3727 test_bit(STRIPE_PREREAD_ACTIVE,
3728 &sh->state)) {
45b4233c 3729 pr_debug("Read_old block "
a4456856
DW		 3730 "%d for Reconstruct\n", i);
3731 set_bit(R5_LOCKED, &dev->flags);
3732 set_bit(R5_Wantread, &dev->flags);
3733 s->locked++;
a9add5d9 3734 qread++;
a4456856
DW		 3735 } else {
3736 set_bit(STRIPE_DELAYED, &sh->state);
3737 set_bit(STRIPE_HANDLE, &sh->state);
3738 }
3739 }
3740 }
e3620a3a 3741 if (rcw && conf->mddev->queue)
a9add5d9
N		 3742 blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d",
3743 (unsigned long long)sh->sector,
3744 rcw, qread, test_bit(STRIPE_DELAYED, &sh->state));
c8ac1803 3745 }
b1b02fe9
N		 3746
3747 if (rcw > disks && rmw > disks &&
3748 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3749 set_bit(STRIPE_DELAYED, &sh->state);
3750
a4456856
DW		 3751 /* now if nothing is locked, and if we have enough data,
3752 * we can start a write request
3753 */
f38e1219
DW		 3754 /* since handle_stripe can be called at any time we need to handle the
3755 * case where a compute block operation has been submitted and then a
ac6b53b6
DW		 3756 * subsequent call wants to start a write request. raid_run_ops only
3757 * handles the case where compute block and reconstruct are requested
f38e1219
DW		 3758 * simultaneously. If this is not the case then new writes need to be
3759 * held off until the compute completes.
3760 */
976ea8d4
DW		 3761 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
3762 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
1e6d690b 3763 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
c0f7bddb 3764 schedule_reconstruction(sh, s, rcw == 0, 0);
d7bd398e 3765 return 0;
a4456856
DW		 3766}
3767
d1688a6d 3768static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh,
a4456856
DW		 3769 struct stripe_head_state *s, int disks)
3770{
ecc65c9b 3771 struct r5dev *dev = NULL;
bd2ab670 3772
59fc630b 3773 BUG_ON(sh->batch_head);
a4456856 3774 set_bit(STRIPE_HANDLE, &sh->state);
e89f8962 3775
ecc65c9b
DW		 3776 switch (sh->check_state) {
3777 case check_state_idle:
3778 /* start a new check operation if there are no failures */
bd2ab670 3779 if (s->failed == 0) {
bd2ab670 3780 BUG_ON(s->uptodate != disks);
ecc65c9b
DW		 3781 sh->check_state = check_state_run;
3782 set_bit(STRIPE_OP_CHECK, &s->ops_request);
bd2ab670 3783 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
bd2ab670 3784 s->uptodate--;
ecc65c9b 3785 break;
bd2ab670 3786 }
f2b3b44d 3787 dev = &sh->dev[s->failed_num[0]];
ecc65c9b
DW		 3788 /* fall through */
3789 case check_state_compute_result:
3790 sh->check_state = check_state_idle;
3791 if (!dev)
3792 dev = &sh->dev[sh->pd_idx];
3793
3794 /* check that a write has not made the stripe insync */
3795 if (test_bit(STRIPE_INSYNC, &sh->state))
3796 break;
c8894419 3797
a4456856 3798 /* either failed parity check, or recovery is happening */
a4456856
DW		 3799 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
3800 BUG_ON(s->uptodate != disks);
3801
3802 set_bit(R5_LOCKED, &dev->flags);
ecc65c9b 3803 s->locked++;
a4456856 3804 set_bit(R5_Wantwrite, &dev->flags);
830ea016 3805
a4456856 3806 clear_bit(STRIPE_DEGRADED, &sh->state);
a4456856 3807 set_bit(STRIPE_INSYNC, &sh->state);
ecc65c9b
DW		 3808 break;
3809 case check_state_run:
3810 break; /* we will be called again upon completion */
3811 case check_state_check_result:
3812 sh->check_state = check_state_idle;
3813
3814 /* if a failure occurred during the check operation, leave
3815 * STRIPE_INSYNC not set and let the stripe be handled again
3816 */
3817 if (s->failed)
3818 break;
3819
3820 /* handle a successful check operation, if parity is correct
3821 * we are done. Otherwise update the mismatch count and repair
3822 * parity if !MD_RECOVERY_CHECK
3823 */
ad283ea4 3824 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
ecc65c9b
DW		 3825 /* parity is correct (on disc,
3826 * not in buffer any more)
3827 */
3828 set_bit(STRIPE_INSYNC, &sh->state);
3829 else {
7f7583d4 3830 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
ecc65c9b
DW		 3831 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
3832 /* don't try to repair!! */
3833 set_bit(STRIPE_INSYNC, &sh->state);
3834 else {
3835 sh->check_state = check_state_compute_run;
976ea8d4 3836 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
ecc65c9b
DW		 3837 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3838 set_bit(R5_Wantcompute,
3839 &sh->dev[sh->pd_idx].flags);
3840 sh->ops.target = sh->pd_idx;
ac6b53b6 3841 sh->ops.target2 = -1;
ecc65c9b
DW		 3842 s->uptodate++;
3843 }
3844 }
3845 break;
3846 case check_state_compute_run:
3847 break;
3848 default:
cc6167b4 3849 pr_err("%s: unknown check_state: %d sector: %llu\n",
ecc65c9b
DW		 3850 __func__, sh->check_state,
3851 (unsigned long long) sh->sector);
3852 BUG();
a4456856
DW		 3853 }
3854}
3855
d1688a6d 3856static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh,
36d1c647 3857 struct stripe_head_state *s,
f2b3b44d 3858 int disks)
a4456856 3859{
a4456856 3860 int pd_idx = sh->pd_idx;
34e04e87 3861 int qd_idx = sh->qd_idx;
d82dfee0 3862 struct r5dev *dev;
a4456856 3863
59fc630b 3864 BUG_ON(sh->batch_head);
a4456856
DW		 3865 set_bit(STRIPE_HANDLE, &sh->state);
3866
3867 BUG_ON(s->failed > 2);
d82dfee0 3868
a4456856
DW		 3869 /* Want to check and possibly repair P and Q.
3870 * However there could be one 'failed' device, in which
3871 * case we can only check one of them, possibly using the
3872 * other to generate missing data
3873 */
3874
d82dfee0
DW		 3875 switch (sh->check_state) {
3876 case check_state_idle:
3877 /* start a new check operation if there are < 2 failures */
f2b3b44d 3878 if (s->failed == s->q_failed) {
d82dfee0 3879 /* The only possible failed device holds Q, so it
a4456856
DW		 3880 * makes sense to check P (If anything else were failed,
3881 * we would have used P to recreate it).
3882 */
d82dfee0 3883 sh->check_state = check_state_run;
a4456856 3884 }
f2b3b44d 3885 if (!s->q_failed && s->failed < 2) {
d82dfee0 3886 /* Q is not failed, and we didn't use it to generate
a4456856
DW		 3887 * anything, so it makes sense to check it
3888 */
d82dfee0
DW		 3889 if (sh->check_state == check_state_run)
3890 sh->check_state = check_state_run_pq;
3891 else
3892 sh->check_state = check_state_run_q;
a4456856 3893 }
a4456856 3894
d82dfee0
DW		 3895 /* discard potentially stale zero_sum_result */
3896 sh->ops.zero_sum_result = 0;
a4456856 3897
d82dfee0
DW		 3898 if (sh->check_state == check_state_run) {
3899 /* async_xor_zero_sum destroys the contents of P */
3900 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
3901 s->uptodate--;
a4456856 3902 }
d82dfee0
DW		 3903 if (sh->check_state >= check_state_run &&
3904 sh->check_state <= check_state_run_pq) {
3905 /* async_syndrome_zero_sum preserves P and Q, so
3906 * no need to mark them !uptodate here
3907 */
3908 set_bit(STRIPE_OP_CHECK, &s->ops_request);
3909 break;
a4456856
DW		 3910 }
3911
d82dfee0
DW		 3912 /* we have 2-disk failure */
3913 BUG_ON(s->failed != 2);
3914 /* fall through */
3915 case check_state_compute_result:
3916 sh->check_state = check_state_idle;
a4456856 3917
d82dfee0
DW		 3918 /* check that a write has not made the stripe insync */
3919 if (test_bit(STRIPE_INSYNC, &sh->state))
3920 break;
a4456856
DW		 3921
3922 /* now write out any block on a failed drive,
d82dfee0 3923 * or P or Q if they were recomputed
a4456856 3924 */
d82dfee0 3925 BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */
a4456856 3926 if (s->failed == 2) {
f2b3b44d 3927 dev = &sh->dev[s->failed_num[1]];
a4456856
DW		 3928 s->locked++;
3929 set_bit(R5_LOCKED, &dev->flags);
3930 set_bit(R5_Wantwrite, &dev->flags);
3931 }
3932 if (s->failed >= 1) {
f2b3b44d 3933 dev = &sh->dev[s->failed_num[0]];
a4456856
DW		 3934 s->locked++;
3935 set_bit(R5_LOCKED, &dev->flags);
3936 set_bit(R5_Wantwrite, &dev->flags);
3937 }
d82dfee0 3938 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
a4456856
DW		 3939 dev = &sh->dev[pd_idx];
3940 s->locked++;
3941 set_bit(R5_LOCKED, &dev->flags);
3942 set_bit(R5_Wantwrite, &dev->flags);
3943 }
d82dfee0 3944 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
a4456856
DW		 3945 dev = &sh->dev[qd_idx];
3946 s->locked++;
3947 set_bit(R5_LOCKED, &dev->flags);
3948 set_bit(R5_Wantwrite, &dev->flags);
3949 }
3950 clear_bit(STRIPE_DEGRADED, &sh->state);
3951
3952 set_bit(STRIPE_INSYNC, &sh->state);
d82dfee0
DW		 3953 break;
3954 case check_state_run:
3955 case check_state_run_q:
3956 case check_state_run_pq:
3957 break; /* we will be called again upon completion */
3958 case check_state_check_result:
3959 sh->check_state = check_state_idle;
3960
3961 /* handle a successful check operation, if parity is correct
3962 * we are done. Otherwise update the mismatch count and repair
3963 * parity if !MD_RECOVERY_CHECK
3964 */
3965 if (sh->ops.zero_sum_result == 0) {
3966 /* both parities are correct */
3967 if (!s->failed)
3968 set_bit(STRIPE_INSYNC, &sh->state);
3969 else {
3970 /* in contrast to the raid5 case we can validate
3971 * parity, but still have a failure to write
3972 * back
3973 */
3974 sh->check_state = check_state_compute_result;
3975 /* Returning at this point means that we may go
3976 * off and bring p and/or q uptodate again so
3977 * we make sure to check zero_sum_result again
3978 * to verify if p or q need writeback
3979 */
3980 }
3981 } else {
7f7583d4 3982 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
d82dfee0
DW		 3983 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
3984 /* don't try to repair!! */
3985 set_bit(STRIPE_INSYNC, &sh->state);
3986 else {
3987 int *target = &sh->ops.target;
3988
3989 sh->ops.target = -1;
3990 sh->ops.target2 = -1;
3991 sh->check_state = check_state_compute_run;
3992 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3993 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3994 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
3995 set_bit(R5_Wantcompute,
3996 &sh->dev[pd_idx].flags);
3997 *target = pd_idx;
3998 target = &sh->ops.target2;
3999 s->uptodate++;
4000 }
4001 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
4002 set_bit(R5_Wantcompute,
4003 &sh->dev[qd_idx].flags);
4004 *target = qd_idx;
4005 s->uptodate++;
4006 }
4007 }
4008 }
4009 break;
4010 case check_state_compute_run:
4011 break;
4012 default:
cc6167b4
N		 4013 pr_warn("%s: unknown check_state: %d sector: %llu\n",
4014 __func__, sh->check_state,
4015 (unsigned long long) sh->sector);
d82dfee0 4016 BUG();
a4456856
DW		 4017 }
4018}
4019
d1688a6d 4020static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh)
a4456856
DW		 4021{
4022 int i;
4023
4024 /* We have read all the blocks in this stripe and now we need to
4025 * copy some of them into a target stripe for expand.
4026 */
f0a50d37 4027 struct dma_async_tx_descriptor *tx = NULL;
59fc630b 4028 BUG_ON(sh->batch_head);
a4456856
DW		 4029 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
4030 for (i = 0; i < sh->disks; i++)
34e04e87 4031 if (i != sh->pd_idx && i != sh->qd_idx) {
911d4ee8 4032 int dd_idx, j;
a4456856 4033 struct stripe_head *sh2;
a08abd8c 4034 struct async_submit_ctl submit;
a4456856 4035
6d036f7d 4036 sector_t bn = raid5_compute_blocknr(sh, i, 1);
911d4ee8
N		 4037 sector_t s = raid5_compute_sector(conf, bn, 0,
4038 &dd_idx, NULL);
6d036f7d 4039 sh2 = raid5_get_active_stripe(conf, s, 0, 1, 1);
a4456856
DW		 4040 if (sh2 == NULL)
4041 /* so far only the early blocks of this stripe
4042 * have been requested. When later blocks
4043 * get requested, we will try again
4044 */
4045 continue;
4046 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
4047 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
4048 /* must have already done this block */
6d036f7d 4049 raid5_release_stripe(sh2);
a4456856
DW		 4050 continue;
4051 }
f0a50d37
DW		 4052
4053 /* place all the copies on one channel */
a08abd8c 4054 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
f0a50d37 4055 tx = async_memcpy(sh2->dev[dd_idx].page,
88ba2aa5 4056 sh->dev[i].page, 0, 0, STRIPE_SIZE,
a08abd8c 4057 &submit);
f0a50d37 4058
a4456856
DW		 4059 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
4060 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
4061 for (j = 0; j < conf->raid_disks; j++)
4062 if (j != sh2->pd_idx &&
86c374ba 4063 j != sh2->qd_idx &&
a4456856
DW		 4064 !test_bit(R5_Expanded, &sh2->dev[j].flags))
4065 break;
4066 if (j == conf->raid_disks) {
4067 set_bit(STRIPE_EXPAND_READY, &sh2->state);
4068 set_bit(STRIPE_HANDLE, &sh2->state);
4069 }
6d036f7d 4070 raid5_release_stripe(sh2);
f0a50d37 4071
a4456856 4072 }
a2e08551 4073 /* done submitting copies, wait for them to complete */
749586b7 4074 async_tx_quiesce(&tx);
a4456856 4075}
1da177e4
LT		 4076
4077/*
4078 * handle_stripe - do things to a stripe.
4079 *
9a3e1101
N		 4080 * We lock the stripe by setting STRIPE_ACTIVE and then examine the
4081 * state of various bits to see what needs to be done.
1da177e4 4082 * Possible results:
9a3e1101
N		 4083 * return some read requests which now have data
4084 * return some write requests which are safely on storage
1da177e4
LT		 4085 * schedule a read on some buffers
4086 * schedule a write of some buffers
4087 * return confirmation of parity correctness
4088 *
1da177e4 4089 */
a4456856 4090
acfe726b 4091static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s)
1da177e4 4092{
d1688a6d 4093 struct r5conf *conf = sh->raid_conf;
f416885e 4094 int disks = sh->disks;
474af965
N		 4095 struct r5dev *dev;
4096 int i;
9a3e1101 4097 int do_recovery = 0;
1da177e4 4098
acfe726b
N		 4099 memset(s, 0, sizeof(*s));
4100
dabc4ec6 4101 s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state) && !sh->batch_head;
4102 s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state) && !sh->batch_head;
acfe726b
N		 4103 s->failed_num[0] = -1;
4104 s->failed_num[1] = -1;
6e74a9cf 4105 s->log_failed = r5l_log_disk_error(conf);
1da177e4 4106
acfe726b 4107 /* Now to look around and see what can be done */
1da177e4 4108 rcu_read_lock();
16a53ecc 4109 for (i=disks; i--; ) {
3cb03002 4110 struct md_rdev *rdev;
31c176ec
N		 4111 sector_t first_bad;
4112 int bad_sectors;
4113 int is_bad = 0;
acfe726b 4114
16a53ecc 4115 dev = &sh->dev[i];
1da177e4 4116
45b4233c 4117 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
9a3e1101
N		 4118 i, dev->flags,
4119 dev->toread, dev->towrite, dev->written);
6c0069c0
YT		 4120 /* maybe we can reply to a read
4121 *
4122 * new wantfill requests are only permitted while
4123 * ops_complete_biofill is guaranteed to be inactive
4124 */
4125 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
4126 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
4127 set_bit(R5_Wantfill, &dev->flags);
1da177e4 4128
16a53ecc 4129 /* now count some things */
cc94015a
N		 4130 if (test_bit(R5_LOCKED, &dev->flags))
4131 s->locked++;
4132 if (test_bit(R5_UPTODATE, &dev->flags))
4133 s->uptodate++;
2d6e4ecc 4134 if (test_bit(R5_Wantcompute, &dev->flags)) {
cc94015a
N		 4135 s->compute++;
4136 BUG_ON(s->compute > 2);
2d6e4ecc 4137 }
1da177e4 4138
acfe726b 4139 if (test_bit(R5_Wantfill, &dev->flags))
cc94015a 4140 s->to_fill++;
acfe726b 4141 else if (dev->toread)
cc94015a 4142 s->to_read++;
16a53ecc 4143 if (dev->towrite) {
cc94015a 4144 s->to_write++;
16a53ecc 4145 if (!test_bit(R5_OVERWRITE, &dev->flags))
cc94015a 4146 s->non_overwrite++;
16a53ecc 4147 }
a4456856 4148 if (dev->written)
cc94015a 4149 s->written++;
14a75d3e
N		 4150 /* Prefer to use the replacement for reads, but only
4151 * if it is recovered enough and has no bad blocks.
4152 */
4153 rdev = rcu_dereference(conf->disks[i].replacement);
4154 if (rdev && !test_bit(Faulty, &rdev->flags) &&
4155 rdev->recovery_offset >= sh->sector + STRIPE_SECTORS &&
4156 !is_badblock(rdev, sh->sector, STRIPE_SECTORS,
4157 &first_bad, &bad_sectors))
4158 set_bit(R5_ReadRepl, &dev->flags);
4159 else {
e6030cb0 4160 if (rdev && !test_bit(Faulty, &rdev->flags))
9a3e1101 4161 set_bit(R5_NeedReplace, &dev->flags);
e6030cb0
N		 4162 else
4163 clear_bit(R5_NeedReplace, &dev->flags);
14a75d3e
N		 4164 rdev = rcu_dereference(conf->disks[i].rdev);
4165 clear_bit(R5_ReadRepl, &dev->flags);
4166 }
9283d8c5
N		 4167 if (rdev && test_bit(Faulty, &rdev->flags))
4168 rdev = NULL;
31c176ec
N		 4169 if (rdev) {
4170 is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
4171 &first_bad, &bad_sectors);
4172 if (s->blocked_rdev == NULL
4173 && (test_bit(Blocked, &rdev->flags)
4174 || is_bad < 0)) {
4175 if (is_bad < 0)
4176 set_bit(BlockedBadBlocks,
4177 &rdev->flags);
4178 s->blocked_rdev = rdev;
4179 atomic_inc(&rdev->nr_pending);
4180 }
6bfe0b49 4181 }
415e72d0
N		 4182 clear_bit(R5_Insync, &dev->flags);
4183 if (!rdev)
4184 /* Not in-sync */;
31c176ec
N		 4185 else if (is_bad) {
4186 /* also not in-sync */
18b9837e
N		 4187 if (!test_bit(WriteErrorSeen, &rdev->flags) &&
4188 test_bit(R5_UPTODATE, &dev->flags)) {
31c176ec
N		 4189 /* treat as in-sync, but with a read error
4190 * which we can now try to correct
4191 */
4192 set_bit(R5_Insync, &dev->flags);
4193 set_bit(R5_ReadError, &dev->flags);
4194 }
4195 } else if (test_bit(In_sync, &rdev->flags))
415e72d0 4196 set_bit(R5_Insync, &dev->flags);
30d7a483 4197 else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
415e72d0 4198 /* in sync if before recovery_offset */
30d7a483
N		 4199 set_bit(R5_Insync, &dev->flags);
4200 else if (test_bit(R5_UPTODATE, &dev->flags) &&
4201 test_bit(R5_Expanded, &dev->flags))
4202 /* If we've reshaped into here, we assume it is Insync.
4203 * We will shortly update recovery_offset to make
4204 * it official.
4205 */
4206 set_bit(R5_Insync, &dev->flags);
4207
1cc03eb9 4208 if (test_bit(R5_WriteError, &dev->flags)) {
14a75d3e
N		 4209 /* This flag does not apply to '.replacement'
4210 * only to .rdev, so make sure to check that*/
4211 struct md_rdev *rdev2 = rcu_dereference(
4212 conf->disks[i].rdev);
4213 if (rdev2 == rdev)
4214 clear_bit(R5_Insync, &dev->flags);
4215 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
bc2607f3 4216 s->handle_bad_blocks = 1;
14a75d3e 4217 atomic_inc(&rdev2->nr_pending);
bc2607f3
N		 4218 } else
4219 clear_bit(R5_WriteError, &dev->flags);
4220 }
1cc03eb9 4221 if (test_bit(R5_MadeGood, &dev->flags)) {
14a75d3e
N		 4222 /* This flag does not apply to '.replacement'
4223 * only to .rdev, so make sure to check that*/
4224 struct md_rdev *rdev2 = rcu_dereference(
4225 conf->disks[i].rdev);
4226 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
b84db560 4227 s->handle_bad_blocks = 1;
14a75d3e 4228 atomic_inc(&rdev2->nr_pending);
b84db560
N		 4229 } else
4230 clear_bit(R5_MadeGood, &dev->flags);
4231 }
977df362
N		 4232 if (test_bit(R5_MadeGoodRepl, &dev->flags)) {
4233 struct md_rdev *rdev2 = rcu_dereference(
4234 conf->disks[i].replacement);
4235 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
4236 s->handle_bad_blocks = 1;
4237 atomic_inc(&rdev2->nr_pending);
4238 } else
4239 clear_bit(R5_MadeGoodRepl, &dev->flags);
4240 }
415e72d0 4241 if (!test_bit(R5_Insync, &dev->flags)) {
16a53ecc
N		 4242 /* The ReadError flag will just be confusing now */
4243 clear_bit(R5_ReadError, &dev->flags);
4244 clear_bit(R5_ReWrite, &dev->flags);
1da177e4 4245 }
415e72d0
N		 4246 if (test_bit(R5_ReadError, &dev->flags))
4247 clear_bit(R5_Insync, &dev->flags);
4248 if (!test_bit(R5_Insync, &dev->flags)) {
cc94015a
N		 4249 if (s->failed < 2)
4250 s->failed_num[s->failed] = i;
4251 s->failed++;
9a3e1101
N		 4252 if (rdev && !test_bit(Faulty, &rdev->flags))
4253 do_recovery = 1;
415e72d0 4254 }
2ded3703
SL		 4255
4256 if (test_bit(R5_InJournal, &dev->flags))
4257 s->injournal++;
1e6d690b
SL		 4258 if (test_bit(R5_InJournal, &dev->flags) && dev->written)
4259 s->just_cached++;
1da177e4 4260 }
9a3e1101
N		 4261 if (test_bit(STRIPE_SYNCING, &sh->state)) {
4262 /* If there is a failed device being replaced,
4263 * we must be recovering.
4264 * else if we are after recovery_cp, we must be syncing
c6d2e084 4265 * else if MD_RECOVERY_REQUESTED is set, we also are syncing.
9a3e1101
N		 4266 * else we can only be replacing
4267 * sync and recovery both need to read all devices, and so
4268 * use the same flag.
4269 */
4270 if (do_recovery ||
c6d2e084 4271 sh->sector >= conf->mddev->recovery_cp ||
4272 test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery)))
9a3e1101
N		 4273 s->syncing = 1;
4274 else
4275 s->replacing = 1;
4276 }
1da177e4 4277 rcu_read_unlock();
cc94015a
N		 4278}
4279
59fc630b 4280static int clear_batch_ready(struct stripe_head *sh)
4281{
b15a9dbd
N		 4282 /* Return '1' if this is a member of batch, or
4283 * '0' if it is a lone stripe or a head which can now be
4284 * handled.
4285 */
59fc630b 4286 struct stripe_head *tmp;
4287 if (!test_and_clear_bit(STRIPE_BATCH_READY, &sh->state))
b15a9dbd 4288 return (sh->batch_head && sh->batch_head != sh);
59fc630b 4289 spin_lock(&sh->stripe_lock);
4290 if (!sh->batch_head) {
4291 spin_unlock(&sh->stripe_lock);
4292 return 0;
4293 }
4294
4295 /*
4296 * this stripe could be added to a batch list before we check
4297 * BATCH_READY, skips it
4298 */
4299 if (sh->batch_head != sh) {
4300 spin_unlock(&sh->stripe_lock);
4301 return 1;
4302 }
4303 spin_lock(&sh->batch_lock);
4304 list_for_each_entry(tmp, &sh->batch_list, batch_list)
4305 clear_bit(STRIPE_BATCH_READY, &tmp->state);
4306 spin_unlock(&sh->batch_lock);
4307 spin_unlock(&sh->stripe_lock);
4308
4309 /*
4310 * BATCH_READY is cleared, no new stripes can be added.
4311 * batch_list can be accessed without lock
4312 */
4313 return 0;
4314}
4315
3960ce79
N		 4316static void break_stripe_batch_list(struct stripe_head *head_sh,
4317 unsigned long handle_flags)
72ac7330 4318{
4e3d62ff 4319 struct stripe_head *sh, *next;
72ac7330 4320 int i;
fb642b92 4321 int do_wakeup = 0;
72ac7330 4322
bb27051f
N		 4323 list_for_each_entry_safe(sh, next, &head_sh->batch_list, batch_list) {
4324
72ac7330 4325 list_del_init(&sh->batch_list);
4326
fb3229d5 4327 WARN_ONCE(sh->state & ((1 << STRIPE_ACTIVE) |
1b956f7a
N		 4328 (1 << STRIPE_SYNCING) |
4329 (1 << STRIPE_REPLACED) |
1b956f7a
N		 4330 (1 << STRIPE_DELAYED) |
4331 (1 << STRIPE_BIT_DELAY) |
4332 (1 << STRIPE_FULL_WRITE) |
4333 (1 << STRIPE_BIOFILL_RUN) |
4334 (1 << STRIPE_COMPUTE_RUN) |
4335 (1 << STRIPE_OPS_REQ_PENDING) |
4336 (1 << STRIPE_DISCARD) |
4337 (1 << STRIPE_BATCH_READY) |
4338 (1 << STRIPE_BATCH_ERR) |
fb3229d5
SL		 4339 (1 << STRIPE_BITMAP_PENDING)),
4340 "stripe state: %lx\n", sh->state);
4341 WARN_ONCE(head_sh->state & ((1 << STRIPE_DISCARD) |
4342 (1 << STRIPE_REPLACED)),
4343 "head stripe state: %lx\n", head_sh->state);
1b956f7a
N		 4344
4345 set_mask_bits(&sh->state, ~(STRIPE_EXPAND_SYNC_FLAGS |
550da24f 4346 (1 << STRIPE_PREREAD_ACTIVE) |
1b956f7a
N		 4347 (1 << STRIPE_DEGRADED)),
4348 head_sh->state & (1 << STRIPE_INSYNC));
4349
72ac7330 4350 sh->check_state = head_sh->check_state;
4351 sh->reconstruct_state = head_sh->reconstruct_state;
fb642b92
N		 4352 for (i = 0; i < sh->disks; i++) {
4353 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
4354 do_wakeup = 1;
72ac7330 4355 sh->dev[i].flags = head_sh->dev[i].flags &
4356 (~((1 << R5_WriteError) | (1 << R5_Overlap)));
fb642b92 4357 }
72ac7330 4358 spin_lock_irq(&sh->stripe_lock);
4359 sh->batch_head = NULL;
4360 spin_unlock_irq(&sh->stripe_lock);
3960ce79
N		 4361 if (handle_flags == 0 ||
4362 sh->state & handle_flags)
4363 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 4364 raid5_release_stripe(sh);
72ac7330 4365 }
fb642b92
N		 4366 spin_lock_irq(&head_sh->stripe_lock);
4367 head_sh->batch_head = NULL;
4368 spin_unlock_irq(&head_sh->stripe_lock);
4369 for (i = 0; i < head_sh->disks; i++)
4370 if (test_and_clear_bit(R5_Overlap, &head_sh->dev[i].flags))
4371 do_wakeup = 1;
3960ce79
N		 4372 if (head_sh->state & handle_flags)
4373 set_bit(STRIPE_HANDLE, &head_sh->state);
fb642b92
N		 4374
4375 if (do_wakeup)
4376 wake_up(&head_sh->raid_conf->wait_for_overlap);
72ac7330 4377}
4378
cc94015a
N		 4379static void handle_stripe(struct stripe_head *sh)
4380{
4381 struct stripe_head_state s;
d1688a6d 4382 struct r5conf *conf = sh->raid_conf;
3687c061 4383 int i;
84789554
N		 4384 int prexor;
4385 int disks = sh->disks;
474af965 4386 struct r5dev *pdev, *qdev;
cc94015a
N		 4387
4388 clear_bit(STRIPE_HANDLE, &sh->state);
257a4b42 4389 if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) {
cc94015a
N		 4390 /* already being handled, ensure it gets handled
4391 * again when current action finishes */
4392 set_bit(STRIPE_HANDLE, &sh->state);
4393 return;
4394 }
4395
59fc630b 4396 if (clear_batch_ready(sh) ) {
4397 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
4398 return;
4399 }
4400
4e3d62ff 4401 if (test_and_clear_bit(STRIPE_BATCH_ERR, &sh->state))
3960ce79 4402 break_stripe_batch_list(sh, 0);
72ac7330 4403
dabc4ec6 4404 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) && !sh->batch_head) {
f8dfcffd
N		 4405 spin_lock(&sh->stripe_lock);
4406 /* Cannot process 'sync' concurrently with 'discard' */
4407 if (!test_bit(STRIPE_DISCARD, &sh->state) &&
4408 test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
4409 set_bit(STRIPE_SYNCING, &sh->state);
4410 clear_bit(STRIPE_INSYNC, &sh->state);
f94c0b66 4411 clear_bit(STRIPE_REPLACED, &sh->state);
f8dfcffd
N		 4412 }
4413 spin_unlock(&sh->stripe_lock);
cc94015a
N		 4414 }
4415 clear_bit(STRIPE_DELAYED, &sh->state);
4416
4417 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
4418 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
4419 (unsigned long long)sh->sector, sh->state,
4420 atomic_read(&sh->count), sh->pd_idx, sh->qd_idx,
4421 sh->check_state, sh->reconstruct_state);
3687c061 4422
acfe726b 4423 analyse_stripe(sh, &s);
c5a31000 4424
b70abcb2
SL		 4425 if (test_bit(STRIPE_LOG_TRAPPED, &sh->state))
4426 goto finish;
4427
bc2607f3
N		 4428 if (s.handle_bad_blocks) {
4429 set_bit(STRIPE_HANDLE, &sh->state);
4430 goto finish;
4431 }
4432
474af965
N		 4433 if (unlikely(s.blocked_rdev)) {
4434 if (s.syncing || s.expanding || s.expanded ||
9a3e1101 4435 s.replacing || s.to_write || s.written) {
474af965
N		 4436 set_bit(STRIPE_HANDLE, &sh->state);
4437 goto finish;
4438 }
4439 /* There is nothing for the blocked_rdev to block */
4440 rdev_dec_pending(s.blocked_rdev, conf->mddev);
4441 s.blocked_rdev = NULL;
4442 }
4443
4444 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
4445 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
4446 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
4447 }
4448
4449 pr_debug("locked=%d uptodate=%d to_read=%d"
4450 " to_write=%d failed=%d failed_num=%d,%d\n",
4451 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
4452 s.failed_num[0], s.failed_num[1]);
4453 /* check if the array has lost more than max_degraded devices and,
4454 * if so, some requests might need to be failed.
4455 */
6e74a9cf 4456 if (s.failed > conf->max_degraded || s.log_failed) {
9a3f530f
N		 4457 sh->check_state = 0;
4458 sh->reconstruct_state = 0;
626f2092 4459 break_stripe_batch_list(sh, 0);
9a3f530f
N		 4460 if (s.to_read+s.to_write+s.written)
4461 handle_failed_stripe(conf, sh, &s, disks, &s.return_bi);
9a3e1101 4462 if (s.syncing + s.replacing)
9a3f530f
N		 4463 handle_failed_sync(conf, sh, &s);
4464 }
474af965 4465
84789554
N		 4466 /* Now we check to see if any write operations have recently
4467 * completed
4468 */
4469 prexor = 0;
4470 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
4471 prexor = 1;
4472 if (sh->reconstruct_state == reconstruct_state_drain_result ||
4473 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
4474 sh->reconstruct_state = reconstruct_state_idle;
4475
4476 /* All the 'written' buffers and the parity block are ready to
4477 * be written back to disk
4478 */
9e444768
SL		 4479 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) &&
4480 !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags));
84789554 4481 BUG_ON(sh->qd_idx >= 0 &&
9e444768
SL		 4482 !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) &&
4483 !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags));
84789554
N		 4484 for (i = disks; i--; ) {
4485 struct r5dev *dev = &sh->dev[i];
4486 if (test_bit(R5_LOCKED, &dev->flags) &&
4487 (i == sh->pd_idx || i == sh->qd_idx ||
1e6d690b
SL		 4488 dev->written || test_bit(R5_InJournal,
4489 &dev->flags))) {
84789554
N		 4490 pr_debug("Writing block %d\n", i);
4491 set_bit(R5_Wantwrite, &dev->flags);
4492 if (prexor)
4493 continue;
9c4bdf69
N		 4494 if (s.failed > 1)
4495 continue;
84789554
N		 4496 if (!test_bit(R5_Insync, &dev->flags) ||
4497 ((i == sh->pd_idx || i == sh->qd_idx) &&
4498 s.failed == 0))
4499 set_bit(STRIPE_INSYNC, &sh->state);
4500 }
4501 }
4502 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4503 s.dec_preread_active = 1;
4504 }
4505
ef5b7c69
N		 4506 /*
4507 * might be able to return some write requests if the parity blocks
4508 * are safe, or on a failed drive
4509 */
4510 pdev = &sh->dev[sh->pd_idx];
4511 s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx)
4512 || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx);
4513 qdev = &sh->dev[sh->qd_idx];
4514 s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx)
4515 || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx)
4516 || conf->level < 6;
4517
4518 if (s.written &&
4519 (s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
4520 && !test_bit(R5_LOCKED, &pdev->flags)
4521 && (test_bit(R5_UPTODATE, &pdev->flags) ||
4522 test_bit(R5_Discard, &pdev->flags))))) &&
4523 (s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
4524 && !test_bit(R5_LOCKED, &qdev->flags)
4525 && (test_bit(R5_UPTODATE, &qdev->flags) ||
4526 test_bit(R5_Discard, &qdev->flags))))))
4527 handle_stripe_clean_event(conf, sh, disks, &s.return_bi);
4528
1e6d690b
SL		 4529 if (s.just_cached)
4530 r5c_handle_cached_data_endio(conf, sh, disks, &s.return_bi);
4531 r5l_stripe_write_finished(sh);
4532
ef5b7c69
N		 4533 /* Now we might consider reading some blocks, either to check/generate
4534 * parity, or to satisfy requests
4535 * or to load a block that is being partially written.
4536 */
4537 if (s.to_read || s.non_overwrite
4538 || (conf->level == 6 && s.to_write && s.failed)
4539 || (s.syncing && (s.uptodate + s.compute < disks))
4540 || s.replacing
4541 || s.expanding)
4542 handle_stripe_fill(sh, &s, disks);
4543
2ded3703
SL		 4544 /*
4545 * When the stripe finishes full journal write cycle (write to journal
4546 * and raid disk), this is the clean up procedure so it is ready for
4547 * next operation.
4548 */
4549 r5c_finish_stripe_write_out(conf, sh, &s);
4550
4551 /*
4552 * Now to consider new write requests, cache write back and what else,
4553 * if anything should be read. We do not handle new writes when:
84789554
N		 4554 * 1/ A 'write' operation (copy+xor) is already in flight.
4555 * 2/ A 'check' operation is in flight, as it may clobber the parity
4556 * block.
2ded3703 4557 * 3/ A r5c cache log write is in flight.
84789554 4558 */
2ded3703
SL		 4559
4560 if (!sh->reconstruct_state && !sh->check_state && !sh->log_io) {
4561 if (!r5c_is_writeback(conf->log)) {
4562 if (s.to_write)
4563 handle_stripe_dirtying(conf, sh, &s, disks);
4564 } else { /* write back cache */
4565 int ret = 0;
4566
4567 /* First, try handle writes in caching phase */
4568 if (s.to_write)
4569 ret = r5c_try_caching_write(conf, sh, &s,
4570 disks);
4571 /*
4572 * If caching phase failed: ret == -EAGAIN
4573 * OR
4574 * stripe under reclaim: !caching && injournal
4575 *
4576 * fall back to handle_stripe_dirtying()
4577 */
4578 if (ret == -EAGAIN ||
4579 /* stripe under reclaim: !caching && injournal */
4580 (!test_bit(STRIPE_R5C_CACHING, &sh->state) &&
d7bd398e
SL		 4581 s.injournal > 0)) {
4582 ret = handle_stripe_dirtying(conf, sh, &s,
4583 disks);
4584 if (ret == -EAGAIN)
4585 goto finish;
4586 }
2ded3703
SL		 4587 }
4588 }
84789554
N		 4589
4590 /* maybe we need to check and possibly fix the parity for this stripe
4591 * Any reads will already have been scheduled, so we just see if enough
4592 * data is available. The parity check is held off while parity
4593 * dependent operations are in flight.
4594 */
4595 if (sh->check_state ||
4596 (s.syncing && s.locked == 0 &&
4597 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
4598 !test_bit(STRIPE_INSYNC, &sh->state))) {
4599 if (conf->level == 6)
4600 handle_parity_checks6(conf, sh, &s, disks);
4601 else
4602 handle_parity_checks5(conf, sh, &s, disks);
4603 }
c5a31000 4604
f94c0b66
N		 4605 if ((s.replacing || s.syncing) && s.locked == 0
4606 && !test_bit(STRIPE_COMPUTE_RUN, &sh->state)
4607 && !test_bit(STRIPE_REPLACED, &sh->state)) {
9a3e1101
N		 4608 /* Write out to replacement devices where possible */
4609 for (i = 0; i < conf->raid_disks; i++)
f94c0b66
N		 4610 if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) {
4611 WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags));
9a3e1101
N		 4612 set_bit(R5_WantReplace, &sh->dev[i].flags);
4613 set_bit(R5_LOCKED, &sh->dev[i].flags);
4614 s.locked++;
4615 }
f94c0b66
N		 4616 if (s.replacing)
4617 set_bit(STRIPE_INSYNC, &sh->state);
4618 set_bit(STRIPE_REPLACED, &sh->state);
9a3e1101
N		 4619 }
4620 if ((s.syncing || s.replacing) && s.locked == 0 &&
f94c0b66 4621 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
9a3e1101 4622 test_bit(STRIPE_INSYNC, &sh->state)) {
c5a31000
N		 4623 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
4624 clear_bit(STRIPE_SYNCING, &sh->state);
f8dfcffd
N		 4625 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
4626 wake_up(&conf->wait_for_overlap);
c5a31000
N		 4627 }
4628
4629 /* If the failed drives are just a ReadError, then we might need
4630 * to progress the repair/check process
4631 */
4632 if (s.failed <= conf->max_degraded && !conf->mddev->ro)
4633 for (i = 0; i < s.failed; i++) {
4634 struct r5dev *dev = &sh->dev[s.failed_num[i]];
4635 if (test_bit(R5_ReadError, &dev->flags)
4636 && !test_bit(R5_LOCKED, &dev->flags)
4637 && test_bit(R5_UPTODATE, &dev->flags)
4638 ) {
4639 if (!test_bit(R5_ReWrite, &dev->flags)) {
4640 set_bit(R5_Wantwrite, &dev->flags);
4641 set_bit(R5_ReWrite, &dev->flags);
4642 set_bit(R5_LOCKED, &dev->flags);
4643 s.locked++;
4644 } else {
4645 /* let's read it back */
4646 set_bit(R5_Wantread, &dev->flags);
4647 set_bit(R5_LOCKED, &dev->flags);
4648 s.locked++;
4649 }
4650 }
4651 }
4652
3687c061
N		 4653 /* Finish reconstruct operations initiated by the expansion process */
4654 if (sh->reconstruct_state == reconstruct_state_result) {
4655 struct stripe_head *sh_src
6d036f7d 4656 = raid5_get_active_stripe(conf, sh->sector, 1, 1, 1);
3687c061
N		 4657 if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) {
4658 /* sh cannot be written until sh_src has been read.
4659 * so arrange for sh to be delayed a little
4660 */
4661 set_bit(STRIPE_DELAYED, &sh->state);
4662 set_bit(STRIPE_HANDLE, &sh->state);
4663 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
4664 &sh_src->state))
4665 atomic_inc(&conf->preread_active_stripes);
6d036f7d 4666 raid5_release_stripe(sh_src);
3687c061
N		 4667 goto finish;
4668 }
4669 if (sh_src)
6d036f7d 4670 raid5_release_stripe(sh_src);
3687c061
N		 4671
4672 sh->reconstruct_state = reconstruct_state_idle;
4673 clear_bit(STRIPE_EXPANDING, &sh->state);
4674 for (i = conf->raid_disks; i--; ) {
4675 set_bit(R5_Wantwrite, &sh->dev[i].flags);
4676 set_bit(R5_LOCKED, &sh->dev[i].flags);
4677 s.locked++;
4678 }
4679 }
f416885e 4680
3687c061
N		 4681 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
4682 !sh->reconstruct_state) {
4683 /* Need to write out all blocks after computing parity */
4684 sh->disks = conf->raid_disks;
4685 stripe_set_idx(sh->sector, conf, 0, sh);
4686 schedule_reconstruction(sh, &s, 1, 1);
4687 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
4688 clear_bit(STRIPE_EXPAND_READY, &sh->state);
4689 atomic_dec(&conf->reshape_stripes);
4690 wake_up(&conf->wait_for_overlap);
4691 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
4692 }
4693
4694 if (s.expanding && s.locked == 0 &&
4695 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
4696 handle_stripe_expansion(conf, sh);
16a53ecc 4697
3687c061 4698finish:
6bfe0b49 4699 /* wait for this device to become unblocked */
5f066c63
N		 4700 if (unlikely(s.blocked_rdev)) {
4701 if (conf->mddev->external)
4702 md_wait_for_blocked_rdev(s.blocked_rdev,
4703 conf->mddev);
4704 else
4705 /* Internal metadata will immediately
4706 * be written by raid5d, so we don't
4707 * need to wait here.
4708 */
4709 rdev_dec_pending(s.blocked_rdev,
4710 conf->mddev);
4711 }
6bfe0b49 4712
bc2607f3
N		 4713 if (s.handle_bad_blocks)
4714 for (i = disks; i--; ) {
3cb03002 4715 struct md_rdev *rdev;
bc2607f3
N		 4716 struct r5dev *dev = &sh->dev[i];
4717 if (test_and_clear_bit(R5_WriteError, &dev->flags)) {
4718 /* We own a safe reference to the rdev */
4719 rdev = conf->disks[i].rdev;
4720 if (!rdev_set_badblocks(rdev, sh->sector,
4721 STRIPE_SECTORS, 0))
4722 md_error(conf->mddev, rdev);
4723 rdev_dec_pending(rdev, conf->mddev);
4724 }
b84db560
N		 4725 if (test_and_clear_bit(R5_MadeGood, &dev->flags)) {
4726 rdev = conf->disks[i].rdev;
4727 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 4728 STRIPE_SECTORS, 0);
b84db560
N		 4729 rdev_dec_pending(rdev, conf->mddev);
4730 }
977df362
N		 4731 if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) {
4732 rdev = conf->disks[i].replacement;
dd054fce
N		 4733 if (!rdev)
4734 /* rdev have been moved down */
4735 rdev = conf->disks[i].rdev;
977df362 4736 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 4737 STRIPE_SECTORS, 0);
977df362
N		 4738 rdev_dec_pending(rdev, conf->mddev);
4739 }
bc2607f3
N		 4740 }
4741
6c0069c0
YT		 4742 if (s.ops_request)
4743 raid_run_ops(sh, s.ops_request);
4744
f0e43bcd 4745 ops_run_io(sh, &s);
16a53ecc 4746
c5709ef6 4747 if (s.dec_preread_active) {
729a1866 4748 /* We delay this until after ops_run_io so that if make_request
e9c7469b 4749 * is waiting on a flush, it won't continue until the writes
729a1866
N		 4750 * have actually been submitted.
4751 */
4752 atomic_dec(&conf->preread_active_stripes);
4753 if (atomic_read(&conf->preread_active_stripes) <
4754 IO_THRESHOLD)
4755 md_wakeup_thread(conf->mddev->thread);
4756 }
4757
c3cce6cd 4758 if (!bio_list_empty(&s.return_bi)) {
2953079c 4759 if (test_bit(MD_SB_CHANGE_PENDING, &conf->mddev->sb_flags)) {
c3cce6cd
N		 4760 spin_lock_irq(&conf->device_lock);
4761 bio_list_merge(&conf->return_bi, &s.return_bi);
4762 spin_unlock_irq(&conf->device_lock);
4763 md_wakeup_thread(conf->mddev->thread);
4764 } else
4765 return_io(&s.return_bi);
4766 }
16a53ecc 4767
257a4b42 4768 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
16a53ecc
N		 4769}
4770
d1688a6d 4771static void raid5_activate_delayed(struct r5conf *conf)
16a53ecc
N		 4772{
4773 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
4774 while (!list_empty(&conf->delayed_list)) {
4775 struct list_head *l = conf->delayed_list.next;
4776 struct stripe_head *sh;
4777 sh = list_entry(l, struct stripe_head, lru);
4778 list_del_init(l);
4779 clear_bit(STRIPE_DELAYED, &sh->state);
4780 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4781 atomic_inc(&conf->preread_active_stripes);
8b3e6cdc 4782 list_add_tail(&sh->lru, &conf->hold_list);
851c30c9 4783 raid5_wakeup_stripe_thread(sh);
16a53ecc 4784 }
482c0834 4785 }
16a53ecc
N		 4786}
4787
566c09c5
SL		 4788static void activate_bit_delay(struct r5conf *conf,
4789 struct list_head *temp_inactive_list)
16a53ecc
N		 4790{
4791 /* device_lock is held */
4792 struct list_head head;
4793 list_add(&head, &conf->bitmap_list);
4794 list_del_init(&conf->bitmap_list);
4795 while (!list_empty(&head)) {
4796 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
566c09c5 4797 int hash;
16a53ecc
N		 4798 list_del_init(&sh->lru);
4799 atomic_inc(&sh->count);
566c09c5
SL		 4800 hash = sh->hash_lock_index;
4801 __release_stripe(conf, sh, &temp_inactive_list[hash]);
16a53ecc
N		 4802 }
4803}
4804
5c675f83 4805static int raid5_congested(struct mddev *mddev, int bits)
f022b2fd 4806{
d1688a6d 4807 struct r5conf *conf = mddev->private;
f022b2fd
N		 4808
4809 /* No difference between reads and writes. Just check
4810 * how busy the stripe_cache is
4811 */
3fa841d7 4812
5423399a 4813 if (test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state))
f022b2fd 4814 return 1;
a39f7afd
SL		 4815
4816 /* Also checks whether there is pressure on r5cache log space */
4817 if (test_bit(R5C_LOG_TIGHT, &conf->cache_state))
4818 return 1;
f022b2fd
N		 4819 if (conf->quiesce)
4820 return 1;
4bda556a 4821 if (atomic_read(&conf->empty_inactive_list_nr))
f022b2fd
N		 4822 return 1;
4823
4824 return 0;
4825}
4826
fd01b88c 4827static int in_chunk_boundary(struct mddev *mddev, struct bio *bio)
f679623f 4828{
3cb5edf4 4829 struct r5conf *conf = mddev->private;
4f024f37 4830 sector_t sector = bio->bi_iter.bi_sector + get_start_sect(bio->bi_bdev);
3cb5edf4 4831 unsigned int chunk_sectors;
aa8b57aa 4832 unsigned int bio_sectors = bio_sectors(bio);
f679623f 4833
3cb5edf4 4834 chunk_sectors = min(conf->chunk_sectors, conf->prev_chunk_sectors);
f679623f
RBJ		 4835 return chunk_sectors >=
4836 ((sector & (chunk_sectors - 1)) + bio_sectors);
4837}
4838
46031f9a
RBJ		 4839/*
4840 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
4841 * later sampled by raid5d.
4842 */
d1688a6d 4843static void add_bio_to_retry(struct bio *bi,struct r5conf *conf)
46031f9a
RBJ		 4844{
4845 unsigned long flags;
4846
4847 spin_lock_irqsave(&conf->device_lock, flags);
4848
4849 bi->bi_next = conf->retry_read_aligned_list;
4850 conf->retry_read_aligned_list = bi;
4851
4852 spin_unlock_irqrestore(&conf->device_lock, flags);
4853 md_wakeup_thread(conf->mddev->thread);
4854}
4855
d1688a6d 4856static struct bio *remove_bio_from_retry(struct r5conf *conf)
46031f9a
RBJ		 4857{
4858 struct bio *bi;
4859
4860 bi = conf->retry_read_aligned;
4861 if (bi) {
4862 conf->retry_read_aligned = NULL;
4863 return bi;
4864 }
4865 bi = conf->retry_read_aligned_list;
4866 if(bi) {
387bb173 4867 conf->retry_read_aligned_list = bi->bi_next;
46031f9a 4868 bi->bi_next = NULL;
960e739d
JA		 4869 /*
4870 * this sets the active strip count to 1 and the processed
4871 * strip count to zero (upper 8 bits)
4872 */
e7836bd6 4873 raid5_set_bi_stripes(bi, 1); /* biased count of active stripes */
46031f9a
RBJ		 4874 }
4875
4876 return bi;
4877}
4878
f679623f
RBJ		 4879/*
4880 * The "raid5_align_endio" should check if the read succeeded and if it
4881 * did, call bio_endio on the original bio (having bio_put the new bio
4882 * first).
4883 * If the read failed..
4884 */
4246a0b6 4885static void raid5_align_endio(struct bio *bi)
f679623f
RBJ		 4886{
4887 struct bio* raid_bi = bi->bi_private;
fd01b88c 4888 struct mddev *mddev;
d1688a6d 4889 struct r5conf *conf;
3cb03002 4890 struct md_rdev *rdev;
9b81c842 4891 int error = bi->bi_error;
46031f9a 4892
f679623f 4893 bio_put(bi);
46031f9a 4894
46031f9a
RBJ		 4895 rdev = (void*)raid_bi->bi_next;
4896 raid_bi->bi_next = NULL;
2b7f2228
N		 4897 mddev = rdev->mddev;
4898 conf = mddev->private;
46031f9a
RBJ		 4899
4900 rdev_dec_pending(rdev, conf->mddev);
4901
9b81c842 4902 if (!error) {
0a82a8d1
LT		 4903 trace_block_bio_complete(bdev_get_queue(raid_bi->bi_bdev),
4904 raid_bi, 0);
4246a0b6 4905 bio_endio(raid_bi);
46031f9a 4906 if (atomic_dec_and_test(&conf->active_aligned_reads))
b1b46486 4907 wake_up(&conf->wait_for_quiescent);
6712ecf8 4908 return;
46031f9a
RBJ		 4909 }
4910
45b4233c 4911 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
46031f9a
RBJ		 4912
4913 add_bio_to_retry(raid_bi, conf);
f679623f
RBJ		 4914}
4915
7ef6b12a 4916static int raid5_read_one_chunk(struct mddev *mddev, struct bio *raid_bio)
f679623f 4917{
d1688a6d 4918 struct r5conf *conf = mddev->private;
8553fe7e 4919 int dd_idx;
f679623f 4920 struct bio* align_bi;
3cb03002 4921 struct md_rdev *rdev;
671488cc 4922 sector_t end_sector;
f679623f
RBJ		 4923
4924 if (!in_chunk_boundary(mddev, raid_bio)) {
7ef6b12a 4925 pr_debug("%s: non aligned\n", __func__);
f679623f
RBJ		 4926 return 0;
4927 }
4928 /*
a167f663 4929 * use bio_clone_mddev to make a copy of the bio
f679623f 4930 */
a167f663 4931 align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev);
f679623f
RBJ		 4932 if (!align_bi)
4933 return 0;
4934 /*
4935 * set bi_end_io to a new function, and set bi_private to the
4936 * original bio.
4937 */
4938 align_bi->bi_end_io = raid5_align_endio;
4939 align_bi->bi_private = raid_bio;
4940 /*
4941 * compute position
4942 */
4f024f37
KO		 4943 align_bi->bi_iter.bi_sector =
4944 raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector,
4945 0, &dd_idx, NULL);
f679623f 4946
f73a1c7d 4947 end_sector = bio_end_sector(align_bi);
f679623f 4948 rcu_read_lock();
671488cc
N		 4949 rdev = rcu_dereference(conf->disks[dd_idx].replacement);
4950 if (!rdev || test_bit(Faulty, &rdev->flags) ||
4951 rdev->recovery_offset < end_sector) {
4952 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
4953 if (rdev &&
4954 (test_bit(Faulty, &rdev->flags) ||
4955 !(test_bit(In_sync, &rdev->flags) ||
4956 rdev->recovery_offset >= end_sector)))
4957 rdev = NULL;
4958 }
4959 if (rdev) {
31c176ec
N		 4960 sector_t first_bad;
4961 int bad_sectors;
4962
f679623f
RBJ		 4963 atomic_inc(&rdev->nr_pending);
4964 rcu_read_unlock();
46031f9a
RBJ		 4965 raid_bio->bi_next = (void*)rdev;
4966 align_bi->bi_bdev = rdev->bdev;
b7c44ed9 4967 bio_clear_flag(align_bi, BIO_SEG_VALID);
46031f9a 4968
7140aafc 4969 if (is_badblock(rdev, align_bi->bi_iter.bi_sector,
4f024f37 4970 bio_sectors(align_bi),
31c176ec 4971 &first_bad, &bad_sectors)) {
387bb173
NB		 4972 bio_put(align_bi);
4973 rdev_dec_pending(rdev, mddev);
4974 return 0;
4975 }
4976
6c0544e2 4977 /* No reshape active, so we can trust rdev->data_offset */
4f024f37 4978 align_bi->bi_iter.bi_sector += rdev->data_offset;
6c0544e2 4979
46031f9a 4980 spin_lock_irq(&conf->device_lock);
b1b46486 4981 wait_event_lock_irq(conf->wait_for_quiescent,
46031f9a 4982 conf->quiesce == 0,
eed8c02e 4983 conf->device_lock);
46031f9a
RBJ		 4984 atomic_inc(&conf->active_aligned_reads);
4985 spin_unlock_irq(&conf->device_lock);
4986
e3620a3a
JB		 4987 if (mddev->gendisk)
4988 trace_block_bio_remap(bdev_get_queue(align_bi->bi_bdev),
4989 align_bi, disk_devt(mddev->gendisk),
4f024f37 4990 raid_bio->bi_iter.bi_sector);
f679623f
RBJ		 4991 generic_make_request(align_bi);
4992 return 1;
4993 } else {
4994 rcu_read_unlock();
46031f9a 4995 bio_put(align_bi);
f679623f
RBJ		 4996 return 0;
4997 }
4998}
4999
7ef6b12a
ML		 5000static struct bio *chunk_aligned_read(struct mddev *mddev, struct bio *raid_bio)
5001{
5002 struct bio *split;
5003
5004 do {
5005 sector_t sector = raid_bio->bi_iter.bi_sector;
5006 unsigned chunk_sects = mddev->chunk_sectors;
5007 unsigned sectors = chunk_sects - (sector & (chunk_sects-1));
5008
5009 if (sectors < bio_sectors(raid_bio)) {
5010 split = bio_split(raid_bio, sectors, GFP_NOIO, fs_bio_set);
5011 bio_chain(split, raid_bio);
5012 } else
5013 split = raid_bio;
5014
5015 if (!raid5_read_one_chunk(mddev, split)) {
5016 if (split != raid_bio)
5017 generic_make_request(raid_bio);
5018 return split;
5019 }
5020 } while (split != raid_bio);
5021
5022 return NULL;
5023}
5024
8b3e6cdc
DW		 5025/* __get_priority_stripe - get the next stripe to process
5026 *
5027 * Full stripe writes are allowed to pass preread active stripes up until
5028 * the bypass_threshold is exceeded. In general the bypass_count
5029 * increments when the handle_list is handled before the hold_list; however, it
5030 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
5031 * stripe with in flight i/o. The bypass_count will be reset when the
5032 * head of the hold_list has changed, i.e. the head was promoted to the
5033 * handle_list.
5034 */
851c30c9 5035static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group)
8b3e6cdc 5036{
851c30c9
SL		 5037 struct stripe_head *sh = NULL, *tmp;
5038 struct list_head *handle_list = NULL;
bfc90cb0 5039 struct r5worker_group *wg = NULL;
851c30c9
SL		 5040
5041 if (conf->worker_cnt_per_group == 0) {
5042 handle_list = &conf->handle_list;
5043 } else if (group != ANY_GROUP) {
5044 handle_list = &conf->worker_groups[group].handle_list;
bfc90cb0 5045 wg = &conf->worker_groups[group];
851c30c9
SL		 5046 } else {
5047 int i;
5048 for (i = 0; i < conf->group_cnt; i++) {
5049 handle_list = &conf->worker_groups[i].handle_list;
bfc90cb0 5050 wg = &conf->worker_groups[i];
851c30c9
SL		 5051 if (!list_empty(handle_list))
5052 break;
5053 }
5054 }
8b3e6cdc
DW		 5055
5056 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
5057 __func__,
851c30c9 5058 list_empty(handle_list) ? "empty" : "busy",
8b3e6cdc
DW		 5059 list_empty(&conf->hold_list) ? "empty" : "busy",
5060 atomic_read(&conf->pending_full_writes), conf->bypass_count);
5061
851c30c9
SL		 5062 if (!list_empty(handle_list)) {
5063 sh = list_entry(handle_list->next, typeof(*sh), lru);
8b3e6cdc
DW		 5064
5065 if (list_empty(&conf->hold_list))
5066 conf->bypass_count = 0;
5067 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
5068 if (conf->hold_list.next == conf->last_hold)
5069 conf->bypass_count++;
5070 else {
5071 conf->last_hold = conf->hold_list.next;
5072 conf->bypass_count -= conf->bypass_threshold;
5073 if (conf->bypass_count < 0)
5074 conf->bypass_count = 0;
5075 }
5076 }
5077 } else if (!list_empty(&conf->hold_list) &&
5078 ((conf->bypass_threshold &&
5079 conf->bypass_count > conf->bypass_threshold) ||
5080 atomic_read(&conf->pending_full_writes) == 0)) {
851c30c9
SL		 5081
5082 list_for_each_entry(tmp, &conf->hold_list, lru) {
5083 if (conf->worker_cnt_per_group == 0 ||
5084 group == ANY_GROUP ||
5085 !cpu_online(tmp->cpu) ||
5086 cpu_to_group(tmp->cpu) == group) {
5087 sh = tmp;
5088 break;
5089 }
5090 }
5091
5092 if (sh) {
5093 conf->bypass_count -= conf->bypass_threshold;
5094 if (conf->bypass_count < 0)
5095 conf->bypass_count = 0;
5096 }
bfc90cb0 5097 wg = NULL;
851c30c9
SL		 5098 }
5099
5100 if (!sh)
8b3e6cdc
DW		 5101 return NULL;
5102
bfc90cb0
SL		 5103 if (wg) {
5104 wg->stripes_cnt--;
5105 sh->group = NULL;
5106 }
8b3e6cdc 5107 list_del_init(&sh->lru);
c7a6d35e 5108 BUG_ON(atomic_inc_return(&sh->count) != 1);
8b3e6cdc
DW		 5109 return sh;
5110}
f679623f 5111
8811b596
SL		 5112struct raid5_plug_cb {
5113 struct blk_plug_cb cb;
5114 struct list_head list;
566c09c5 5115 struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
8811b596
SL		 5116};
5117
5118static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule)
5119{
5120 struct raid5_plug_cb *cb = container_of(
5121 blk_cb, struct raid5_plug_cb, cb);
5122 struct stripe_head *sh;
5123 struct mddev *mddev = cb->cb.data;
5124 struct r5conf *conf = mddev->private;
a9add5d9 5125 int cnt = 0;
566c09c5 5126 int hash;
8811b596
SL		 5127
5128 if (cb->list.next && !list_empty(&cb->list)) {
5129 spin_lock_irq(&conf->device_lock);
5130 while (!list_empty(&cb->list)) {
5131 sh = list_first_entry(&cb->list, struct stripe_head, lru);
5132 list_del_init(&sh->lru);
5133 /*
5134 * avoid race release_stripe_plug() sees
5135 * STRIPE_ON_UNPLUG_LIST clear but the stripe
5136 * is still in our list
5137 */
4e857c58 5138 smp_mb__before_atomic();
8811b596 5139 clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state);
773ca82f
SL		 5140 /*
5141 * STRIPE_ON_RELEASE_LIST could be set here. In that
5142 * case, the count is always > 1 here
5143 */
566c09c5
SL		 5144 hash = sh->hash_lock_index;
5145 __release_stripe(conf, sh, &cb->temp_inactive_list[hash]);
a9add5d9 5146 cnt++;
8811b596
SL		 5147 }
5148 spin_unlock_irq(&conf->device_lock);
5149 }
566c09c5
SL		 5150 release_inactive_stripe_list(conf, cb->temp_inactive_list,
5151 NR_STRIPE_HASH_LOCKS);
e3620a3a
JB		 5152 if (mddev->queue)
5153 trace_block_unplug(mddev->queue, cnt, !from_schedule);
8811b596
SL		 5154 kfree(cb);
5155}
5156
5157static void release_stripe_plug(struct mddev *mddev,
5158 struct stripe_head *sh)
5159{
5160 struct blk_plug_cb *blk_cb = blk_check_plugged(
5161 raid5_unplug, mddev,
5162 sizeof(struct raid5_plug_cb));
5163 struct raid5_plug_cb *cb;
5164
5165 if (!blk_cb) {
6d036f7d 5166 raid5_release_stripe(sh);
8811b596
SL		 5167 return;
5168 }
5169
5170 cb = container_of(blk_cb, struct raid5_plug_cb, cb);
5171
566c09c5
SL		 5172 if (cb->list.next == NULL) {
5173 int i;
8811b596 5174 INIT_LIST_HEAD(&cb->list);
566c09c5
SL		 5175 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5176 INIT_LIST_HEAD(cb->temp_inactive_list + i);
5177 }
8811b596
SL		 5178
5179 if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state))
5180 list_add_tail(&sh->lru, &cb->list);
5181 else
6d036f7d 5182 raid5_release_stripe(sh);
8811b596
SL		 5183}
5184
620125f2
SL		 5185static void make_discard_request(struct mddev *mddev, struct bio *bi)
5186{
5187 struct r5conf *conf = mddev->private;
5188 sector_t logical_sector, last_sector;
5189 struct stripe_head *sh;
5190 int remaining;
5191 int stripe_sectors;
5192
5193 if (mddev->reshape_position != MaxSector)
5194 /* Skip discard while reshape is happening */
5195 return;
5196
4f024f37
KO		 5197 logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1);
5198 last_sector = bi->bi_iter.bi_sector + (bi->bi_iter.bi_size>>9);
620125f2
SL		 5199
5200 bi->bi_next = NULL;
5201 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
5202
5203 stripe_sectors = conf->chunk_sectors *
5204 (conf->raid_disks - conf->max_degraded);
5205 logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector,
5206 stripe_sectors);
5207 sector_div(last_sector, stripe_sectors);
5208
5209 logical_sector *= conf->chunk_sectors;
5210 last_sector *= conf->chunk_sectors;
5211
5212 for (; logical_sector < last_sector;
5213 logical_sector += STRIPE_SECTORS) {
5214 DEFINE_WAIT(w);
5215 int d;
5216 again:
6d036f7d 5217 sh = raid5_get_active_stripe(conf, logical_sector, 0, 0, 0);
620125f2
SL		 5218 prepare_to_wait(&conf->wait_for_overlap, &w,
5219 TASK_UNINTERRUPTIBLE);
f8dfcffd
N		 5220 set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
5221 if (test_bit(STRIPE_SYNCING, &sh->state)) {
6d036f7d 5222 raid5_release_stripe(sh);
f8dfcffd
N		 5223 schedule();
5224 goto again;
5225 }
5226 clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
620125f2
SL		 5227 spin_lock_irq(&sh->stripe_lock);
5228 for (d = 0; d < conf->raid_disks; d++) {
5229 if (d == sh->pd_idx || d == sh->qd_idx)
5230 continue;
5231 if (sh->dev[d].towrite || sh->dev[d].toread) {
5232 set_bit(R5_Overlap, &sh->dev[d].flags);
5233 spin_unlock_irq(&sh->stripe_lock);
6d036f7d 5234 raid5_release_stripe(sh);
620125f2
SL		 5235 schedule();
5236 goto again;
5237 }
5238 }
f8dfcffd 5239 set_bit(STRIPE_DISCARD, &sh->state);
620125f2 5240 finish_wait(&conf->wait_for_overlap, &w);
7a87f434 5241 sh->overwrite_disks = 0;
620125f2
SL		 5242 for (d = 0; d < conf->raid_disks; d++) {
5243 if (d == sh->pd_idx || d == sh->qd_idx)
5244 continue;
5245 sh->dev[d].towrite = bi;
5246 set_bit(R5_OVERWRITE, &sh->dev[d].flags);
5247 raid5_inc_bi_active_stripes(bi);
7a87f434 5248 sh->overwrite_disks++;
620125f2
SL		 5249 }
5250 spin_unlock_irq(&sh->stripe_lock);
5251 if (conf->mddev->bitmap) {
5252 for (d = 0;
5253 d < conf->raid_disks - conf->max_degraded;
5254 d++)
5255 bitmap_startwrite(mddev->bitmap,
5256 sh->sector,
5257 STRIPE_SECTORS,
5258 0);
5259 sh->bm_seq = conf->seq_flush + 1;
5260 set_bit(STRIPE_BIT_DELAY, &sh->state);
5261 }
5262
5263 set_bit(STRIPE_HANDLE, &sh->state);
5264 clear_bit(STRIPE_DELAYED, &sh->state);
5265 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
5266 atomic_inc(&conf->preread_active_stripes);
5267 release_stripe_plug(mddev, sh);
5268 }
5269
5270 remaining = raid5_dec_bi_active_stripes(bi);
5271 if (remaining == 0) {
5272 md_write_end(mddev);
4246a0b6 5273 bio_endio(bi);
620125f2
SL		 5274 }
5275}
5276
849674e4 5277static void raid5_make_request(struct mddev *mddev, struct bio * bi)
1da177e4 5278{
d1688a6d 5279 struct r5conf *conf = mddev->private;
911d4ee8 5280 int dd_idx;
1da177e4
LT		 5281 sector_t new_sector;
5282 sector_t logical_sector, last_sector;
5283 struct stripe_head *sh;
a362357b 5284 const int rw = bio_data_dir(bi);
49077326 5285 int remaining;
27c0f68f
SL		 5286 DEFINE_WAIT(w);
5287 bool do_prepare;
3bddb7f8 5288 bool do_flush = false;
1da177e4 5289
1eff9d32 5290 if (unlikely(bi->bi_opf & REQ_PREFLUSH)) {
828cbe98
SL		 5291 int ret = r5l_handle_flush_request(conf->log, bi);
5292
5293 if (ret == 0)
5294 return;
5295 if (ret == -ENODEV) {
5296 md_flush_request(mddev, bi);
5297 return;
5298 }
5299 /* ret == -EAGAIN, fallback */
3bddb7f8
SL		 5300 /*
5301 * if r5l_handle_flush_request() didn't clear REQ_PREFLUSH,
5302 * we need to flush journal device
5303 */
5304 do_flush = bi->bi_opf & REQ_PREFLUSH;
e5dcdd80
N		 5305 }
5306
3d310eb7 5307 md_write_start(mddev, bi);
06d91a5f 5308
9ffc8f7c
EM		 5309 /*
5310 * If array is degraded, better not do chunk aligned read because
5311 * later we might have to read it again in order to reconstruct
5312 * data on failed drives.
5313 */
5314 if (rw == READ && mddev->degraded == 0 &&
2ded3703 5315 !r5c_is_writeback(conf->log) &&
7ef6b12a
ML		 5316 mddev->reshape_position == MaxSector) {
5317 bi = chunk_aligned_read(mddev, bi);
5318 if (!bi)
5319 return;
5320 }
52488615 5321
796a5cf0 5322 if (unlikely(bio_op(bi) == REQ_OP_DISCARD)) {
620125f2
SL		 5323 make_discard_request(mddev, bi);
5324 return;
5325 }
5326
4f024f37 5327 logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1);
f73a1c7d 5328 last_sector = bio_end_sector(bi);
1da177e4
LT		 5329 bi->bi_next = NULL;
5330 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
06d91a5f 5331
27c0f68f 5332 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
1da177e4 5333 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
b5663ba4 5334 int previous;
c46501b2 5335 int seq;
b578d55f 5336
27c0f68f 5337 do_prepare = false;
7ecaa1e6 5338 retry:
c46501b2 5339 seq = read_seqcount_begin(&conf->gen_lock);
b5663ba4 5340 previous = 0;
27c0f68f
SL		 5341 if (do_prepare)
5342 prepare_to_wait(&conf->wait_for_overlap, &w,
5343 TASK_UNINTERRUPTIBLE);
b0f9ec04 5344 if (unlikely(conf->reshape_progress != MaxSector)) {
fef9c61f 5345 /* spinlock is needed as reshape_progress may be
df8e7f76
N		 5346 * 64bit on a 32bit platform, and so it might be
5347 * possible to see a half-updated value
aeb878b0 5348 * Of course reshape_progress could change after
df8e7f76
N		 5349 * the lock is dropped, so once we get a reference
5350 * to the stripe that we think it is, we will have
5351 * to check again.
5352 */
7ecaa1e6 5353 spin_lock_irq(&conf->device_lock);
2c810cdd 5354 if (mddev->reshape_backwards
fef9c61f
N		 5355 ? logical_sector < conf->reshape_progress
5356 : logical_sector >= conf->reshape_progress) {
b5663ba4
N		 5357 previous = 1;
5358 } else {
2c810cdd 5359 if (mddev->reshape_backwards
fef9c61f
N		 5360 ? logical_sector < conf->reshape_safe
5361 : logical_sector >= conf->reshape_safe) {
b578d55f
N		 5362 spin_unlock_irq(&conf->device_lock);
5363 schedule();
27c0f68f 5364 do_prepare = true;
b578d55f
N		 5365 goto retry;
5366 }
5367 }
7ecaa1e6
N		 5368 spin_unlock_irq(&conf->device_lock);
5369 }
16a53ecc 5370
112bf897
N		 5371 new_sector = raid5_compute_sector(conf, logical_sector,
5372 previous,
911d4ee8 5373 &dd_idx, NULL);
849674e4 5374 pr_debug("raid456: raid5_make_request, sector %llu logical %llu\n",
c46501b2 5375 (unsigned long long)new_sector,
1da177e4
LT		 5376 (unsigned long long)logical_sector);
5377
6d036f7d 5378 sh = raid5_get_active_stripe(conf, new_sector, previous,
1eff9d32 5379 (bi->bi_opf & REQ_RAHEAD), 0);
1da177e4 5380 if (sh) {
b0f9ec04 5381 if (unlikely(previous)) {
7ecaa1e6 5382 /* expansion might have moved on while waiting for a
df8e7f76
N		 5383 * stripe, so we must do the range check again.
5384 * Expansion could still move past after this
5385 * test, but as we are holding a reference to
5386 * 'sh', we know that if that happens,
5387 * STRIPE_EXPANDING will get set and the expansion
5388 * won't proceed until we finish with the stripe.
7ecaa1e6
N		 5389 */
5390 int must_retry = 0;
5391 spin_lock_irq(&conf->device_lock);
2c810cdd 5392 if (mddev->reshape_backwards
b0f9ec04
N		 5393 ? logical_sector >= conf->reshape_progress
5394 : logical_sector < conf->reshape_progress)
7ecaa1e6
N		 5395 /* mismatch, need to try again */
5396 must_retry = 1;
5397 spin_unlock_irq(&conf->device_lock);
5398 if (must_retry) {
6d036f7d 5399 raid5_release_stripe(sh);
7a3ab908 5400 schedule();
27c0f68f 5401 do_prepare = true;
7ecaa1e6
N		 5402 goto retry;
5403 }
5404 }
c46501b2
N		 5405 if (read_seqcount_retry(&conf->gen_lock, seq)) {
5406 /* Might have got the wrong stripe_head
5407 * by accident
5408 */
6d036f7d 5409 raid5_release_stripe(sh);
c46501b2
N		 5410 goto retry;
5411 }
e62e58a5 5412
ffd96e35 5413 if (rw == WRITE &&
a5c308d4 5414 logical_sector >= mddev->suspend_lo &&
e464eafd 5415 logical_sector < mddev->suspend_hi) {
6d036f7d 5416 raid5_release_stripe(sh);
e62e58a5
N		 5417 /* As the suspend_* range is controlled by
5418 * userspace, we want an interruptible
5419 * wait.
5420 */
5421 flush_signals(current);
5422 prepare_to_wait(&conf->wait_for_overlap,
5423 &w, TASK_INTERRUPTIBLE);
5424 if (logical_sector >= mddev->suspend_lo &&
27c0f68f 5425 logical_sector < mddev->suspend_hi) {
e62e58a5 5426 schedule();
27c0f68f
SL		 5427 do_prepare = true;
5428 }
e464eafd
N		 5429 goto retry;
5430 }
7ecaa1e6
N		 5431
5432 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
da41ba65 5433 !add_stripe_bio(sh, bi, dd_idx, rw, previous)) {
7ecaa1e6
N		 5434 /* Stripe is busy expanding or
5435 * add failed due to overlap. Flush everything
1da177e4
LT		 5436 * and wait a while
5437 */
482c0834 5438 md_wakeup_thread(mddev->thread);
6d036f7d 5439 raid5_release_stripe(sh);
1da177e4 5440 schedule();
27c0f68f 5441 do_prepare = true;
1da177e4
LT		 5442 goto retry;
5443 }
3bddb7f8
SL		 5444 if (do_flush) {
5445 set_bit(STRIPE_R5C_PREFLUSH, &sh->state);
5446 /* we only need flush for one stripe */
5447 do_flush = false;
5448 }
5449
6ed3003c
N		 5450 set_bit(STRIPE_HANDLE, &sh->state);
5451 clear_bit(STRIPE_DELAYED, &sh->state);
59fc630b 5452 if ((!sh->batch_head || sh == sh->batch_head) &&
1eff9d32 5453 (bi->bi_opf & REQ_SYNC) &&
729a1866
N		 5454 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
5455 atomic_inc(&conf->preread_active_stripes);
8811b596 5456 release_stripe_plug(mddev, sh);
1da177e4
LT		 5457 } else {
5458 /* cannot get stripe for read-ahead, just give-up */
4246a0b6 5459 bi->bi_error = -EIO;
1da177e4
LT		 5460 break;
5461 }
1da177e4 5462 }
27c0f68f 5463 finish_wait(&conf->wait_for_overlap, &w);
7c13edc8 5464
e7836bd6 5465 remaining = raid5_dec_bi_active_stripes(bi);
f6344757 5466 if (remaining == 0) {
1da177e4 5467
16a53ecc 5468 if ( rw == WRITE )
1da177e4 5469 md_write_end(mddev);
6712ecf8 5470
0a82a8d1
LT		 5471 trace_block_bio_complete(bdev_get_queue(bi->bi_bdev),
5472 bi, 0);
4246a0b6 5473 bio_endio(bi);
1da177e4 5474 }
1da177e4
LT		 5475}
5476
fd01b88c 5477static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks);
b522adcd 5478
fd01b88c 5479static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped)
1da177e4 5480{
52c03291
N		 5481 /* reshaping is quite different to recovery/resync so it is
5482 * handled quite separately ... here.
5483 *
5484 * On each call to sync_request, we gather one chunk worth of
5485 * destination stripes and flag them as expanding.
5486 * Then we find all the source stripes and request reads.
5487 * As the reads complete, handle_stripe will copy the data
5488 * into the destination stripe and release that stripe.
5489 */
d1688a6d 5490 struct r5conf *conf = mddev->private;
1da177e4 5491 struct stripe_head *sh;
ccfcc3c1 5492 sector_t first_sector, last_sector;
f416885e
N		 5493 int raid_disks = conf->previous_raid_disks;
5494 int data_disks = raid_disks - conf->max_degraded;
5495 int new_data_disks = conf->raid_disks - conf->max_degraded;
52c03291
N		 5496 int i;
5497 int dd_idx;
c8f517c4 5498 sector_t writepos, readpos, safepos;
ec32a2bd 5499 sector_t stripe_addr;
7a661381 5500 int reshape_sectors;
ab69ae12 5501 struct list_head stripes;
92140480 5502 sector_t retn;
52c03291 5503
fef9c61f
N		 5504 if (sector_nr == 0) {
5505 /* If restarting in the middle, skip the initial sectors */
2c810cdd 5506 if (mddev->reshape_backwards &&
fef9c61f
N		 5507 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
5508 sector_nr = raid5_size(mddev, 0, 0)
5509 - conf->reshape_progress;
6cbd8148
N		 5510 } else if (mddev->reshape_backwards &&
5511 conf->reshape_progress == MaxSector) {
5512 /* shouldn't happen, but just in case, finish up.*/
5513 sector_nr = MaxSector;
2c810cdd 5514 } else if (!mddev->reshape_backwards &&
fef9c61f
N		 5515 conf->reshape_progress > 0)
5516 sector_nr = conf->reshape_progress;
f416885e 5517 sector_div(sector_nr, new_data_disks);
fef9c61f 5518 if (sector_nr) {
8dee7211
N		 5519 mddev->curr_resync_completed = sector_nr;
5520 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
fef9c61f 5521 *skipped = 1;
92140480
N		 5522 retn = sector_nr;
5523 goto finish;
fef9c61f 5524 }
52c03291
N		 5525 }
5526
7a661381
N		 5527 /* We need to process a full chunk at a time.
5528 * If old and new chunk sizes differ, we need to process the
5529 * largest of these
5530 */
3cb5edf4
N		 5531
5532 reshape_sectors = max(conf->chunk_sectors, conf->prev_chunk_sectors);
7a661381 5533
b5254dd5
N		 5534 /* We update the metadata at least every 10 seconds, or when
5535 * the data about to be copied would over-write the source of
5536 * the data at the front of the range. i.e. one new_stripe
5537 * along from reshape_progress new_maps to after where
5538 * reshape_safe old_maps to
52c03291 5539 */
fef9c61f 5540 writepos = conf->reshape_progress;
f416885e 5541 sector_div(writepos, new_data_disks);
c8f517c4
N		 5542 readpos = conf->reshape_progress;
5543 sector_div(readpos, data_disks);
fef9c61f 5544 safepos = conf->reshape_safe;
f416885e 5545 sector_div(safepos, data_disks);
2c810cdd 5546 if (mddev->reshape_backwards) {
c74c0d76
N		 5547 BUG_ON(writepos < reshape_sectors);
5548 writepos -= reshape_sectors;
c8f517c4 5549 readpos += reshape_sectors;
7a661381 5550 safepos += reshape_sectors;
fef9c61f 5551 } else {
7a661381 5552 writepos += reshape_sectors;
c74c0d76
N		 5553 /* readpos and safepos are worst-case calculations.
5554 * A negative number is overly pessimistic, and causes
5555 * obvious problems for unsigned storage. So clip to 0.
5556 */
ed37d83e
N		 5557 readpos -= min_t(sector_t, reshape_sectors, readpos);
5558 safepos -= min_t(sector_t, reshape_sectors, safepos);
fef9c61f 5559 }
52c03291 5560
b5254dd5
N		 5561 /* Having calculated the 'writepos' possibly use it
5562 * to set 'stripe_addr' which is where we will write to.
5563 */
5564 if (mddev->reshape_backwards) {
5565 BUG_ON(conf->reshape_progress == 0);
5566 stripe_addr = writepos;
5567 BUG_ON((mddev->dev_sectors &
5568 ~((sector_t)reshape_sectors - 1))
5569 - reshape_sectors - stripe_addr
5570 != sector_nr);
5571 } else {
5572 BUG_ON(writepos != sector_nr + reshape_sectors);
5573 stripe_addr = sector_nr;
5574 }
5575
c8f517c4
N		 5576 /* 'writepos' is the most advanced device address we might write.
5577 * 'readpos' is the least advanced device address we might read.
5578 * 'safepos' is the least address recorded in the metadata as having
5579 * been reshaped.
b5254dd5
N		 5580 * If there is a min_offset_diff, these are adjusted either by
5581 * increasing the safepos/readpos if diff is negative, or
5582 * increasing writepos if diff is positive.
5583 * If 'readpos' is then behind 'writepos', there is no way that we can
c8f517c4
N		 5584 * ensure safety in the face of a crash - that must be done by userspace
5585 * making a backup of the data. So in that case there is no particular
5586 * rush to update metadata.
5587 * Otherwise if 'safepos' is behind 'writepos', then we really need to
5588 * update the metadata to advance 'safepos' to match 'readpos' so that
5589 * we can be safe in the event of a crash.
5590 * So we insist on updating metadata if safepos is behind writepos and
5591 * readpos is beyond writepos.
5592 * In any case, update the metadata every 10 seconds.
5593 * Maybe that number should be configurable, but I'm not sure it is
5594 * worth it.... maybe it could be a multiple of safemode_delay???
5595 */
b5254dd5
N		 5596 if (conf->min_offset_diff < 0) {
5597 safepos += -conf->min_offset_diff;
5598 readpos += -conf->min_offset_diff;
5599 } else
5600 writepos += conf->min_offset_diff;
5601
2c810cdd 5602 if ((mddev->reshape_backwards
c8f517c4
N		 5603 ? (safepos > writepos && readpos < writepos)
5604 : (safepos < writepos && readpos > writepos)) ||
5605 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
52c03291
N		 5606 /* Cannot proceed until we've updated the superblock... */
5607 wait_event(conf->wait_for_overlap,
c91abf5a
N		 5608 atomic_read(&conf->reshape_stripes)==0
5609 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5610 if (atomic_read(&conf->reshape_stripes) != 0)
5611 return 0;
fef9c61f 5612 mddev->reshape_position = conf->reshape_progress;
75d3da43 5613 mddev->curr_resync_completed = sector_nr;
c8f517c4 5614 conf->reshape_checkpoint = jiffies;
2953079c 5615 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
52c03291 5616 md_wakeup_thread(mddev->thread);
2953079c 5617 wait_event(mddev->sb_wait, mddev->sb_flags == 0 ||
c91abf5a
N		 5618 test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5619 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5620 return 0;
52c03291 5621 spin_lock_irq(&conf->device_lock);
fef9c61f 5622 conf->reshape_safe = mddev->reshape_position;
52c03291
N		 5623 spin_unlock_irq(&conf->device_lock);
5624 wake_up(&conf->wait_for_overlap);
acb180b0 5625 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
52c03291
N		 5626 }
5627
ab69ae12 5628 INIT_LIST_HEAD(&stripes);
7a661381 5629 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
52c03291 5630 int j;
a9f326eb 5631 int skipped_disk = 0;
6d036f7d 5632 sh = raid5_get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
52c03291
N		 5633 set_bit(STRIPE_EXPANDING, &sh->state);
5634 atomic_inc(&conf->reshape_stripes);
5635 /* If any of this stripe is beyond the end of the old
5636 * array, then we need to zero those blocks
5637 */
5638 for (j=sh->disks; j--;) {
5639 sector_t s;
5640 if (j == sh->pd_idx)
5641 continue;
f416885e 5642 if (conf->level == 6 &&
d0dabf7e 5643 j == sh->qd_idx)
f416885e 5644 continue;
6d036f7d 5645 s = raid5_compute_blocknr(sh, j, 0);
b522adcd 5646 if (s < raid5_size(mddev, 0, 0)) {
a9f326eb 5647 skipped_disk = 1;
52c03291
N		 5648 continue;
5649 }
5650 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
5651 set_bit(R5_Expanded, &sh->dev[j].flags);
5652 set_bit(R5_UPTODATE, &sh->dev[j].flags);
5653 }
a9f326eb 5654 if (!skipped_disk) {
52c03291
N		 5655 set_bit(STRIPE_EXPAND_READY, &sh->state);
5656 set_bit(STRIPE_HANDLE, &sh->state);
5657 }
ab69ae12 5658 list_add(&sh->lru, &stripes);
52c03291
N		 5659 }
5660 spin_lock_irq(&conf->device_lock);
2c810cdd 5661 if (mddev->reshape_backwards)
7a661381 5662 conf->reshape_progress -= reshape_sectors * new_data_disks;
fef9c61f 5663 else
7a661381 5664 conf->reshape_progress += reshape_sectors * new_data_disks;
52c03291
N		 5665 spin_unlock_irq(&conf->device_lock);
5666 /* Ok, those stripe are ready. We can start scheduling
5667 * reads on the source stripes.
5668 * The source stripes are determined by mapping the first and last
5669 * block on the destination stripes.
5670 */
52c03291 5671 first_sector =
ec32a2bd 5672 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
911d4ee8 5673 1, &dd_idx, NULL);
52c03291 5674 last_sector =
0e6e0271 5675 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
09c9e5fa 5676 * new_data_disks - 1),
911d4ee8 5677 1, &dd_idx, NULL);
58c0fed4
AN		 5678 if (last_sector >= mddev->dev_sectors)
5679 last_sector = mddev->dev_sectors - 1;
52c03291 5680 while (first_sector <= last_sector) {
6d036f7d 5681 sh = raid5_get_active_stripe(conf, first_sector, 1, 0, 1);
52c03291
N		 5682 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
5683 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 5684 raid5_release_stripe(sh);
52c03291
N		 5685 first_sector += STRIPE_SECTORS;
5686 }
ab69ae12
N		 5687 /* Now that the sources are clearly marked, we can release
5688 * the destination stripes
5689 */
5690 while (!list_empty(&stripes)) {
5691 sh = list_entry(stripes.next, struct stripe_head, lru);
5692 list_del_init(&sh->lru);
6d036f7d 5693 raid5_release_stripe(sh);
ab69ae12 5694 }
c6207277
N		 5695 /* If this takes us to the resync_max point where we have to pause,
5696 * then we need to write out the superblock.
5697 */
7a661381 5698 sector_nr += reshape_sectors;
92140480
N		 5699 retn = reshape_sectors;
5700finish:
c5e19d90
N		 5701 if (mddev->curr_resync_completed > mddev->resync_max ||
5702 (sector_nr - mddev->curr_resync_completed) * 2
c03f6a19 5703 >= mddev->resync_max - mddev->curr_resync_completed) {
c6207277
N		 5704 /* Cannot proceed until we've updated the superblock... */
5705 wait_event(conf->wait_for_overlap,
c91abf5a
N		 5706 atomic_read(&conf->reshape_stripes) == 0
5707 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5708 if (atomic_read(&conf->reshape_stripes) != 0)
5709 goto ret;
fef9c61f 5710 mddev->reshape_position = conf->reshape_progress;
75d3da43 5711 mddev->curr_resync_completed = sector_nr;
c8f517c4 5712 conf->reshape_checkpoint = jiffies;
2953079c 5713 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
c6207277
N		 5714 md_wakeup_thread(mddev->thread);
5715 wait_event(mddev->sb_wait,
2953079c 5716 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)
c91abf5a
N		 5717 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5718 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5719 goto ret;
c6207277 5720 spin_lock_irq(&conf->device_lock);
fef9c61f 5721 conf->reshape_safe = mddev->reshape_position;
c6207277
N		 5722 spin_unlock_irq(&conf->device_lock);
5723 wake_up(&conf->wait_for_overlap);
acb180b0 5724 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
c6207277 5725 }
c91abf5a 5726ret:
92140480 5727 return retn;
52c03291
N		 5728}
5729
849674e4
SL		 5730static inline sector_t raid5_sync_request(struct mddev *mddev, sector_t sector_nr,
5731 int *skipped)
52c03291 5732{
d1688a6d 5733 struct r5conf *conf = mddev->private;
52c03291 5734 struct stripe_head *sh;
58c0fed4 5735 sector_t max_sector = mddev->dev_sectors;
57dab0bd 5736 sector_t sync_blocks;
16a53ecc
N		 5737 int still_degraded = 0;
5738 int i;
1da177e4 5739
72626685 5740 if (sector_nr >= max_sector) {
1da177e4 5741 /* just being told to finish up .. nothing much to do */
cea9c228 5742
29269553
N		 5743 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
5744 end_reshape(conf);
5745 return 0;
5746 }
72626685
N		 5747
5748 if (mddev->curr_resync < max_sector) /* aborted */
5749 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
5750 &sync_blocks, 1);
16a53ecc 5751 else /* completed sync */
72626685
N		 5752 conf->fullsync = 0;
5753 bitmap_close_sync(mddev->bitmap);
5754
1da177e4
LT		 5755 return 0;
5756 }
ccfcc3c1 5757
64bd660b
N		 5758 /* Allow raid5_quiesce to complete */
5759 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
5760
52c03291
N		 5761 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5762 return reshape_request(mddev, sector_nr, skipped);
f6705578 5763
c6207277
N		 5764 /* No need to check resync_max as we never do more than one
5765 * stripe, and as resync_max will always be on a chunk boundary,
5766 * if the check in md_do_sync didn't fire, there is no chance
5767 * of overstepping resync_max here
5768 */
5769
16a53ecc 5770 /* if there is too many failed drives and we are trying
1da177e4
LT		 5771 * to resync, then assert that we are finished, because there is
5772 * nothing we can do.
5773 */
3285edf1 5774 if (mddev->degraded >= conf->max_degraded &&
16a53ecc 5775 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
58c0fed4 5776 sector_t rv = mddev->dev_sectors - sector_nr;
57afd89f 5777 *skipped = 1;
1da177e4
LT		 5778 return rv;
5779 }
6f608040 5780 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
5781 !conf->fullsync &&
5782 !bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
5783 sync_blocks >= STRIPE_SECTORS) {
72626685
N		 5784 /* we can skip this block, and probably more */
5785 sync_blocks /= STRIPE_SECTORS;
5786 *skipped = 1;
5787 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
5788 }
1da177e4 5789
c40f341f 5790 bitmap_cond_end_sync(mddev->bitmap, sector_nr, false);
b47490c9 5791
6d036f7d 5792 sh = raid5_get_active_stripe(conf, sector_nr, 0, 1, 0);
1da177e4 5793 if (sh == NULL) {
6d036f7d 5794 sh = raid5_get_active_stripe(conf, sector_nr, 0, 0, 0);
1da177e4 5795 /* make sure we don't swamp the stripe cache if someone else
16a53ecc 5796 * is trying to get access
1da177e4 5797 */
66c006a5 5798 schedule_timeout_uninterruptible(1);
1da177e4 5799 }
16a53ecc 5800 /* Need to check if array will still be degraded after recovery/resync
16d9cfab
EM		 5801 * Note in case of > 1 drive failures it's possible we're rebuilding
5802 * one drive while leaving another faulty drive in array.
16a53ecc 5803 */
16d9cfab
EM		 5804 rcu_read_lock();
5805 for (i = 0; i < conf->raid_disks; i++) {
5806 struct md_rdev *rdev = ACCESS_ONCE(conf->disks[i].rdev);
5807
5808 if (rdev == NULL || test_bit(Faulty, &rdev->flags))
16a53ecc 5809 still_degraded = 1;
16d9cfab
EM		 5810 }
5811 rcu_read_unlock();
16a53ecc
N		 5812
5813 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
5814
83206d66 5815 set_bit(STRIPE_SYNC_REQUESTED, &sh->state);
053f5b65 5816 set_bit(STRIPE_HANDLE, &sh->state);
1da177e4 5817
6d036f7d 5818 raid5_release_stripe(sh);
1da177e4
LT		 5819
5820 return STRIPE_SECTORS;
5821}
5822
d1688a6d 5823static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio)
46031f9a
RBJ		 5824{
5825 /* We may not be able to submit a whole bio at once as there
5826 * may not be enough stripe_heads available.
5827 * We cannot pre-allocate enough stripe_heads as we may need
5828 * more than exist in the cache (if we allow ever large chunks).
5829 * So we do one stripe head at a time and record in
5830 * ->bi_hw_segments how many have been done.
5831 *
5832 * We *know* that this entire raid_bio is in one chunk, so
5833 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
5834 */
5835 struct stripe_head *sh;
911d4ee8 5836 int dd_idx;
46031f9a
RBJ		 5837 sector_t sector, logical_sector, last_sector;
5838 int scnt = 0;
5839 int remaining;
5840 int handled = 0;
5841
4f024f37
KO		 5842 logical_sector = raid_bio->bi_iter.bi_sector &
5843 ~((sector_t)STRIPE_SECTORS-1);
112bf897 5844 sector = raid5_compute_sector(conf, logical_sector,
911d4ee8 5845 0, &dd_idx, NULL);
f73a1c7d 5846 last_sector = bio_end_sector(raid_bio);
46031f9a
RBJ		 5847
5848 for (; logical_sector < last_sector;
387bb173
NB		 5849 logical_sector += STRIPE_SECTORS,
5850 sector += STRIPE_SECTORS,
5851 scnt++) {
46031f9a 5852
e7836bd6 5853 if (scnt < raid5_bi_processed_stripes(raid_bio))
46031f9a
RBJ		 5854 /* already done this stripe */
5855 continue;
5856
6d036f7d 5857 sh = raid5_get_active_stripe(conf, sector, 0, 1, 1);
46031f9a
RBJ		 5858
5859 if (!sh) {
5860 /* failed to get a stripe - must wait */
e7836bd6 5861 raid5_set_bi_processed_stripes(raid_bio, scnt);
46031f9a
RBJ		 5862 conf->retry_read_aligned = raid_bio;
5863 return handled;
5864 }
5865
da41ba65 5866 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) {
6d036f7d 5867 raid5_release_stripe(sh);
e7836bd6 5868 raid5_set_bi_processed_stripes(raid_bio, scnt);
387bb173
NB		 5869 conf->retry_read_aligned = raid_bio;
5870 return handled;
5871 }
5872
3f9e7c14 5873 set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags);
36d1c647 5874 handle_stripe(sh);
6d036f7d 5875 raid5_release_stripe(sh);
46031f9a
RBJ		 5876 handled++;
5877 }
e7836bd6 5878 remaining = raid5_dec_bi_active_stripes(raid_bio);
0a82a8d1
LT		 5879 if (remaining == 0) {
5880 trace_block_bio_complete(bdev_get_queue(raid_bio->bi_bdev),
5881 raid_bio, 0);
4246a0b6 5882 bio_endio(raid_bio);
0a82a8d1 5883 }
46031f9a 5884 if (atomic_dec_and_test(&conf->active_aligned_reads))
b1b46486 5885 wake_up(&conf->wait_for_quiescent);
46031f9a
RBJ		 5886 return handled;
5887}
5888
bfc90cb0 5889static int handle_active_stripes(struct r5conf *conf, int group,
566c09c5
SL		 5890 struct r5worker *worker,
5891 struct list_head *temp_inactive_list)
46a06401
SL		 5892{
5893 struct stripe_head *batch[MAX_STRIPE_BATCH], *sh;
566c09c5
SL		 5894 int i, batch_size = 0, hash;
5895 bool release_inactive = false;
46a06401
SL		 5896
5897 while (batch_size < MAX_STRIPE_BATCH &&
851c30c9 5898 (sh = __get_priority_stripe(conf, group)) != NULL)
46a06401
SL		 5899 batch[batch_size++] = sh;
5900
566c09c5
SL		 5901 if (batch_size == 0) {
5902 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5903 if (!list_empty(temp_inactive_list + i))
5904 break;
a8c34f91
SL		 5905 if (i == NR_STRIPE_HASH_LOCKS) {
5906 spin_unlock_irq(&conf->device_lock);
5907 r5l_flush_stripe_to_raid(conf->log);
5908 spin_lock_irq(&conf->device_lock);
566c09c5 5909 return batch_size;
a8c34f91 5910 }
566c09c5
SL		 5911 release_inactive = true;
5912 }
46a06401
SL		 5913 spin_unlock_irq(&conf->device_lock);
5914
566c09c5
SL		 5915 release_inactive_stripe_list(conf, temp_inactive_list,
5916 NR_STRIPE_HASH_LOCKS);
5917
a8c34f91 5918 r5l_flush_stripe_to_raid(conf->log);
566c09c5
SL		 5919 if (release_inactive) {
5920 spin_lock_irq(&conf->device_lock);
5921 return 0;
5922 }
5923
46a06401
SL		 5924 for (i = 0; i < batch_size; i++)
5925 handle_stripe(batch[i]);
f6bed0ef 5926 r5l_write_stripe_run(conf->log);
46a06401
SL		 5927
5928 cond_resched();
5929
5930 spin_lock_irq(&conf->device_lock);
566c09c5
SL		 5931 for (i = 0; i < batch_size; i++) {
5932 hash = batch[i]->hash_lock_index;
5933 __release_stripe(conf, batch[i], &temp_inactive_list[hash]);
5934 }
46a06401
SL		 5935 return batch_size;
5936}
46031f9a 5937
851c30c9
SL		 5938static void raid5_do_work(struct work_struct *work)
5939{
5940 struct r5worker *worker = container_of(work, struct r5worker, work);
5941 struct r5worker_group *group = worker->group;
5942 struct r5conf *conf = group->conf;
5943 int group_id = group - conf->worker_groups;
5944 int handled;
5945 struct blk_plug plug;
5946
5947 pr_debug("+++ raid5worker active\n");
5948
5949 blk_start_plug(&plug);
5950 handled = 0;
5951 spin_lock_irq(&conf->device_lock);
5952 while (1) {
5953 int batch_size, released;
5954
566c09c5 5955 released = release_stripe_list(conf, worker->temp_inactive_list);
851c30c9 5956
566c09c5
SL		 5957 batch_size = handle_active_stripes(conf, group_id, worker,
5958 worker->temp_inactive_list);
bfc90cb0 5959 worker->working = false;
851c30c9
SL		 5960 if (!batch_size && !released)
5961 break;
5962 handled += batch_size;
5963 }
5964 pr_debug("%d stripes handled\n", handled);
5965
5966 spin_unlock_irq(&conf->device_lock);
5967 blk_finish_plug(&plug);
5968
5969 pr_debug("--- raid5worker inactive\n");
5970}
5971
1da177e4
LT		 5972/*
5973 * This is our raid5 kernel thread.
5974 *
5975 * We scan the hash table for stripes which can be handled now.
5976 * During the scan, completed stripes are saved for us by the interrupt
5977 * handler, so that they will not have to wait for our next wakeup.
5978 */
4ed8731d 5979static void raid5d(struct md_thread *thread)
1da177e4 5980{
4ed8731d 5981 struct mddev *mddev = thread->mddev;
d1688a6d 5982 struct r5conf *conf = mddev->private;
1da177e4 5983 int handled;
e1dfa0a2 5984 struct blk_plug plug;
1da177e4 5985
45b4233c 5986 pr_debug("+++ raid5d active\n");
1da177e4
LT		 5987
5988 md_check_recovery(mddev);
1da177e4 5989
c3cce6cd 5990 if (!bio_list_empty(&conf->return_bi) &&
2953079c 5991 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
c3cce6cd
N		 5992 struct bio_list tmp = BIO_EMPTY_LIST;
5993 spin_lock_irq(&conf->device_lock);
2953079c 5994 if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
c3cce6cd
N		 5995 bio_list_merge(&tmp, &conf->return_bi);
5996 bio_list_init(&conf->return_bi);
5997 }
5998 spin_unlock_irq(&conf->device_lock);
5999 return_io(&tmp);
6000 }
6001
e1dfa0a2 6002 blk_start_plug(&plug);
1da177e4
LT		 6003 handled = 0;
6004 spin_lock_irq(&conf->device_lock);
6005 while (1) {
46031f9a 6006 struct bio *bio;
773ca82f
SL		 6007 int batch_size, released;
6008
566c09c5 6009 released = release_stripe_list(conf, conf->temp_inactive_list);
edbe83ab
N		 6010 if (released)
6011 clear_bit(R5_DID_ALLOC, &conf->cache_state);
1da177e4 6012
0021b7bc 6013 if (
7c13edc8
N		 6014 !list_empty(&conf->bitmap_list)) {
6015 /* Now is a good time to flush some bitmap updates */
6016 conf->seq_flush++;
700e432d 6017 spin_unlock_irq(&conf->device_lock);
72626685 6018 bitmap_unplug(mddev->bitmap);
700e432d 6019 spin_lock_irq(&conf->device_lock);
7c13edc8 6020 conf->seq_write = conf->seq_flush;
566c09c5 6021 activate_bit_delay(conf, conf->temp_inactive_list);
72626685 6022 }
0021b7bc 6023 raid5_activate_delayed(conf);
72626685 6024
46031f9a
RBJ		 6025 while ((bio = remove_bio_from_retry(conf))) {
6026 int ok;
6027 spin_unlock_irq(&conf->device_lock);
6028 ok = retry_aligned_read(conf, bio);
6029 spin_lock_irq(&conf->device_lock);
6030 if (!ok)
6031 break;
6032 handled++;
6033 }
6034
566c09c5
SL		 6035 batch_size = handle_active_stripes(conf, ANY_GROUP, NULL,
6036 conf->temp_inactive_list);
773ca82f 6037 if (!batch_size && !released)
1da177e4 6038 break;
46a06401 6039 handled += batch_size;
1da177e4 6040
2953079c 6041 if (mddev->sb_flags & ~(1 << MD_SB_CHANGE_PENDING)) {
46a06401 6042 spin_unlock_irq(&conf->device_lock);
de393cde 6043 md_check_recovery(mddev);
46a06401
SL		 6044 spin_lock_irq(&conf->device_lock);
6045 }
1da177e4 6046 }
45b4233c 6047 pr_debug("%d stripes handled\n", handled);
1da177e4
LT		 6048
6049 spin_unlock_irq(&conf->device_lock);
2d5b569b
N		 6050 if (test_and_clear_bit(R5_ALLOC_MORE, &conf->cache_state) &&
6051 mutex_trylock(&conf->cache_size_mutex)) {
edbe83ab
N		 6052 grow_one_stripe(conf, __GFP_NOWARN);
6053 /* Set flag even if allocation failed. This helps
6054 * slow down allocation requests when mem is short
6055 */
6056 set_bit(R5_DID_ALLOC, &conf->cache_state);
2d5b569b 6057 mutex_unlock(&conf->cache_size_mutex);
edbe83ab 6058 }
1da177e4 6059
0576b1c6
SL		 6060 r5l_flush_stripe_to_raid(conf->log);
6061
c9f21aaf 6062 async_tx_issue_pending_all();
e1dfa0a2 6063 blk_finish_plug(&plug);
1da177e4 6064
45b4233c 6065 pr_debug("--- raid5d inactive\n");
1da177e4
LT		 6066}
6067
3f294f4f 6068static ssize_t
fd01b88c 6069raid5_show_stripe_cache_size(struct mddev *mddev, char *page)
3f294f4f 6070{
7b1485ba
N		 6071 struct r5conf *conf;
6072 int ret = 0;
6073 spin_lock(&mddev->lock);
6074 conf = mddev->private;
96de1e66 6075 if (conf)
edbe83ab 6076 ret = sprintf(page, "%d\n", conf->min_nr_stripes);
7b1485ba
N		 6077 spin_unlock(&mddev->lock);
6078 return ret;
3f294f4f
N		 6079}
6080
c41d4ac4 6081int
fd01b88c 6082raid5_set_cache_size(struct mddev *mddev, int size)
3f294f4f 6083{
d1688a6d 6084 struct r5conf *conf = mddev->private;
b5470dc5
DW		 6085 int err;
6086
c41d4ac4 6087 if (size <= 16 || size > 32768)
3f294f4f 6088 return -EINVAL;
486f0644 6089
edbe83ab 6090 conf->min_nr_stripes = size;
2d5b569b 6091 mutex_lock(&conf->cache_size_mutex);
486f0644
N		 6092 while (size < conf->max_nr_stripes &&
6093 drop_one_stripe(conf))
6094 ;
2d5b569b 6095 mutex_unlock(&conf->cache_size_mutex);
486f0644 6096
edbe83ab 6097
b5470dc5
DW		 6098 err = md_allow_write(mddev);
6099 if (err)
6100 return err;
486f0644 6101
2d5b569b 6102 mutex_lock(&conf->cache_size_mutex);
486f0644
N		 6103 while (size > conf->max_nr_stripes)
6104 if (!grow_one_stripe(conf, GFP_KERNEL))
6105 break;
2d5b569b 6106 mutex_unlock(&conf->cache_size_mutex);
486f0644 6107
c41d4ac4
N		 6108 return 0;
6109}
6110EXPORT_SYMBOL(raid5_set_cache_size);
6111
6112static ssize_t
fd01b88c 6113raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len)
c41d4ac4 6114{
6791875e 6115 struct r5conf *conf;
c41d4ac4
N		 6116 unsigned long new;
6117 int err;
6118
6119 if (len >= PAGE_SIZE)
6120 return -EINVAL;
b29bebd6 6121 if (kstrtoul(page, 10, &new))
c41d4ac4 6122 return -EINVAL;
6791875e 6123 err = mddev_lock(mddev);
c41d4ac4
N		 6124 if (err)
6125 return err;
6791875e
N		 6126 conf = mddev->private;
6127 if (!conf)
6128 err = -ENODEV;
6129 else
6130 err = raid5_set_cache_size(mddev, new);
6131 mddev_unlock(mddev);
6132
6133 return err ?: len;
3f294f4f 6134}
007583c9 6135
96de1e66
N		 6136static struct md_sysfs_entry
6137raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
6138 raid5_show_stripe_cache_size,
6139 raid5_store_stripe_cache_size);
3f294f4f 6140
d06f191f
MS		 6141static ssize_t
6142raid5_show_rmw_level(struct mddev *mddev, char *page)
6143{
6144 struct r5conf *conf = mddev->private;
6145 if (conf)
6146 return sprintf(page, "%d\n", conf->rmw_level);
6147 else
6148 return 0;
6149}
6150
6151static ssize_t
6152raid5_store_rmw_level(struct mddev *mddev, const char *page, size_t len)
6153{
6154 struct r5conf *conf = mddev->private;
6155 unsigned long new;
6156
6157 if (!conf)
6158 return -ENODEV;
6159
6160 if (len >= PAGE_SIZE)
6161 return -EINVAL;
6162
6163 if (kstrtoul(page, 10, &new))
6164 return -EINVAL;
6165
6166 if (new != PARITY_DISABLE_RMW && !raid6_call.xor_syndrome)
6167 return -EINVAL;
6168
6169 if (new != PARITY_DISABLE_RMW &&
6170 new != PARITY_ENABLE_RMW &&
6171 new != PARITY_PREFER_RMW)
6172 return -EINVAL;
6173
6174 conf->rmw_level = new;
6175 return len;
6176}
6177
6178static struct md_sysfs_entry
6179raid5_rmw_level = __ATTR(rmw_level, S_IRUGO | S_IWUSR,
6180 raid5_show_rmw_level,
6181 raid5_store_rmw_level);
6182
6183
8b3e6cdc 6184static ssize_t
fd01b88c 6185raid5_show_preread_threshold(struct mddev *mddev, char *page)
8b3e6cdc 6186{
7b1485ba
N		 6187 struct r5conf *conf;
6188 int ret = 0;
6189 spin_lock(&mddev->lock);
6190 conf = mddev->private;
8b3e6cdc 6191 if (conf)
7b1485ba
N		 6192 ret = sprintf(page, "%d\n", conf->bypass_threshold);
6193 spin_unlock(&mddev->lock);
6194 return ret;
8b3e6cdc
DW		 6195}
6196
6197static ssize_t
fd01b88c 6198raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len)
8b3e6cdc 6199{
6791875e 6200 struct r5conf *conf;
4ef197d8 6201 unsigned long new;
6791875e
N		 6202 int err;
6203
8b3e6cdc
DW		 6204 if (len >= PAGE_SIZE)
6205 return -EINVAL;
b29bebd6 6206 if (kstrtoul(page, 10, &new))
8b3e6cdc 6207 return -EINVAL;
6791875e
N		 6208
6209 err = mddev_lock(mddev);
6210 if (err)
6211 return err;
6212 conf = mddev->private;
6213 if (!conf)
6214 err = -ENODEV;
edbe83ab 6215 else if (new > conf->min_nr_stripes)
6791875e
N		 6216 err = -EINVAL;
6217 else
6218 conf->bypass_threshold = new;
6219 mddev_unlock(mddev);
6220 return err ?: len;
8b3e6cdc
DW		 6221}
6222
6223static struct md_sysfs_entry
6224raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
6225 S_IRUGO | S_IWUSR,
6226 raid5_show_preread_threshold,
6227 raid5_store_preread_threshold);
6228
d592a996
SL		 6229static ssize_t
6230raid5_show_skip_copy(struct mddev *mddev, char *page)
6231{
7b1485ba
N		 6232 struct r5conf *conf;
6233 int ret = 0;
6234 spin_lock(&mddev->lock);
6235 conf = mddev->private;
d592a996 6236 if (conf)
7b1485ba
N		 6237 ret = sprintf(page, "%d\n", conf->skip_copy);
6238 spin_unlock(&mddev->lock);
6239 return ret;
d592a996
SL		 6240}
6241
6242static ssize_t
6243raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len)
6244{
6791875e 6245 struct r5conf *conf;
d592a996 6246 unsigned long new;
6791875e
N		 6247 int err;
6248
d592a996
SL		 6249 if (len >= PAGE_SIZE)
6250 return -EINVAL;
d592a996
SL		 6251 if (kstrtoul(page, 10, &new))
6252 return -EINVAL;
6253 new = !!new;
6791875e
N		 6254
6255 err = mddev_lock(mddev);
6256 if (err)
6257 return err;
6258 conf = mddev->private;
6259 if (!conf)
6260 err = -ENODEV;
6261 else if (new != conf->skip_copy) {
6262 mddev_suspend(mddev);
6263 conf->skip_copy = new;
6264 if (new)
6265 mddev->queue->backing_dev_info.capabilities |=
6266 BDI_CAP_STABLE_WRITES;
6267 else
6268 mddev->queue->backing_dev_info.capabilities &=
6269 ~BDI_CAP_STABLE_WRITES;
6270 mddev_resume(mddev);
6271 }
6272 mddev_unlock(mddev);
6273 return err ?: len;
d592a996
SL		 6274}
6275
6276static struct md_sysfs_entry
6277raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR,
6278 raid5_show_skip_copy,
6279 raid5_store_skip_copy);
6280
3f294f4f 6281static ssize_t
fd01b88c 6282stripe_cache_active_show(struct mddev *mddev, char *page)
3f294f4f 6283{
d1688a6d 6284 struct r5conf *conf = mddev->private;
96de1e66
N		 6285 if (conf)
6286 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
6287 else
6288 return 0;
3f294f4f
N		 6289}
6290
96de1e66
N		 6291static struct md_sysfs_entry
6292raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
3f294f4f 6293
b721420e
SL		 6294static ssize_t
6295raid5_show_group_thread_cnt(struct mddev *mddev, char *page)
6296{
7b1485ba
N		 6297 struct r5conf *conf;
6298 int ret = 0;
6299 spin_lock(&mddev->lock);
6300 conf = mddev->private;
b721420e 6301 if (conf)
7b1485ba
N		 6302 ret = sprintf(page, "%d\n", conf->worker_cnt_per_group);
6303 spin_unlock(&mddev->lock);
6304 return ret;
b721420e
SL		 6305}
6306
60aaf933 6307static int alloc_thread_groups(struct r5conf *conf, int cnt,
6308 int *group_cnt,
6309 int *worker_cnt_per_group,
6310 struct r5worker_group **worker_groups);
b721420e
SL		 6311static ssize_t
6312raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len)
6313{
6791875e 6314 struct r5conf *conf;
b721420e
SL		 6315 unsigned long new;
6316 int err;
60aaf933 6317 struct r5worker_group *new_groups, *old_groups;
6318 int group_cnt, worker_cnt_per_group;
b721420e
SL		 6319
6320 if (len >= PAGE_SIZE)
6321 return -EINVAL;
b721420e
SL		 6322 if (kstrtoul(page, 10, &new))
6323 return -EINVAL;
6324
6791875e
N		 6325 err = mddev_lock(mddev);
6326 if (err)
6327 return err;
6328 conf = mddev->private;
6329 if (!conf)
6330 err = -ENODEV;
6331 else if (new != conf->worker_cnt_per_group) {
6332 mddev_suspend(mddev);
b721420e 6333
6791875e
N		 6334 old_groups = conf->worker_groups;
6335 if (old_groups)
6336 flush_workqueue(raid5_wq);
d206dcfa 6337
6791875e
N		 6338 err = alloc_thread_groups(conf, new,
6339 &group_cnt, &worker_cnt_per_group,
6340 &new_groups);
6341 if (!err) {
6342 spin_lock_irq(&conf->device_lock);
6343 conf->group_cnt = group_cnt;
6344 conf->worker_cnt_per_group = worker_cnt_per_group;
6345 conf->worker_groups = new_groups;
6346 spin_unlock_irq(&conf->device_lock);
b721420e 6347
6791875e
N		 6348 if (old_groups)
6349 kfree(old_groups[0].workers);
6350 kfree(old_groups);
6351 }
6352 mddev_resume(mddev);
b721420e 6353 }
6791875e 6354 mddev_unlock(mddev);
b721420e 6355
6791875e 6356 return err ?: len;
b721420e
SL		 6357}
6358
6359static struct md_sysfs_entry
6360raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR,
6361 raid5_show_group_thread_cnt,
6362 raid5_store_group_thread_cnt);
6363
007583c9 6364static struct attribute *raid5_attrs[] = {
3f294f4f
N		 6365 &raid5_stripecache_size.attr,
6366 &raid5_stripecache_active.attr,
8b3e6cdc 6367 &raid5_preread_bypass_threshold.attr,
b721420e 6368 &raid5_group_thread_cnt.attr,
d592a996 6369 &raid5_skip_copy.attr,
d06f191f 6370 &raid5_rmw_level.attr,
2c7da14b 6371 &r5c_journal_mode.attr,
3f294f4f
N		 6372 NULL,
6373};
007583c9
N		 6374static struct attribute_group raid5_attrs_group = {
6375 .name = NULL,
6376 .attrs = raid5_attrs,
3f294f4f
N		 6377};
6378
60aaf933 6379static int alloc_thread_groups(struct r5conf *conf, int cnt,
6380 int *group_cnt,
6381 int *worker_cnt_per_group,
6382 struct r5worker_group **worker_groups)
851c30c9 6383{
566c09c5 6384 int i, j, k;
851c30c9
SL		 6385 ssize_t size;
6386 struct r5worker *workers;
6387
60aaf933 6388 *worker_cnt_per_group = cnt;
851c30c9 6389 if (cnt == 0) {
60aaf933 6390 *group_cnt = 0;
6391 *worker_groups = NULL;
851c30c9
SL		 6392 return 0;
6393 }
60aaf933 6394 *group_cnt = num_possible_nodes();
851c30c9 6395 size = sizeof(struct r5worker) * cnt;
60aaf933 6396 workers = kzalloc(size * *group_cnt, GFP_NOIO);
6397 *worker_groups = kzalloc(sizeof(struct r5worker_group) *
6398 *group_cnt, GFP_NOIO);
6399 if (!*worker_groups || !workers) {
851c30c9 6400 kfree(workers);
60aaf933 6401 kfree(*worker_groups);
851c30c9
SL		 6402 return -ENOMEM;
6403 }
6404
60aaf933 6405 for (i = 0; i < *group_cnt; i++) {
851c30c9
SL		 6406 struct r5worker_group *group;
6407
0c775d52 6408 group = &(*worker_groups)[i];
851c30c9
SL		 6409 INIT_LIST_HEAD(&group->handle_list);
6410 group->conf = conf;
6411 group->workers = workers + i * cnt;
6412
6413 for (j = 0; j < cnt; j++) {
566c09c5
SL		 6414 struct r5worker *worker = group->workers + j;
6415 worker->group = group;
6416 INIT_WORK(&worker->work, raid5_do_work);
6417
6418 for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++)
6419 INIT_LIST_HEAD(worker->temp_inactive_list + k);
851c30c9
SL		 6420 }
6421 }
6422
6423 return 0;
6424}
6425
6426static void free_thread_groups(struct r5conf *conf)
6427{
6428 if (conf->worker_groups)
6429 kfree(conf->worker_groups[0].workers);
6430 kfree(conf->worker_groups);
6431 conf->worker_groups = NULL;
6432}
6433
80c3a6ce 6434static sector_t
fd01b88c 6435raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks)
80c3a6ce 6436{
d1688a6d 6437 struct r5conf *conf = mddev->private;
80c3a6ce
DW		 6438
6439 if (!sectors)
6440 sectors = mddev->dev_sectors;
5e5e3e78 6441 if (!raid_disks)
7ec05478 6442 /* size is defined by the smallest of previous and new size */
5e5e3e78 6443 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
80c3a6ce 6444
3cb5edf4
N		 6445 sectors &= ~((sector_t)conf->chunk_sectors - 1);
6446 sectors &= ~((sector_t)conf->prev_chunk_sectors - 1);
80c3a6ce
DW		 6447 return sectors * (raid_disks - conf->max_degraded);
6448}
6449
789b5e03
ON		 6450static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
6451{
6452 safe_put_page(percpu->spare_page);
46d5b785 6453 if (percpu->scribble)
6454 flex_array_free(percpu->scribble);
789b5e03
ON		 6455 percpu->spare_page = NULL;
6456 percpu->scribble = NULL;
6457}
6458
6459static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
6460{
6461 if (conf->level == 6 && !percpu->spare_page)
6462 percpu->spare_page = alloc_page(GFP_KERNEL);
6463 if (!percpu->scribble)
46d5b785 6464 percpu->scribble = scribble_alloc(max(conf->raid_disks,
738a2738
N		 6465 conf->previous_raid_disks),
6466 max(conf->chunk_sectors,
6467 conf->prev_chunk_sectors)
6468 / STRIPE_SECTORS,
6469 GFP_KERNEL);
789b5e03
ON		 6470
6471 if (!percpu->scribble || (conf->level == 6 && !percpu->spare_page)) {
6472 free_scratch_buffer(conf, percpu);
6473 return -ENOMEM;
6474 }
6475
6476 return 0;
6477}
6478
29c6d1bb 6479static int raid456_cpu_dead(unsigned int cpu, struct hlist_node *node)
36d1c647 6480{
29c6d1bb
SAS		 6481 struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node);
6482
6483 free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
6484 return 0;
6485}
36d1c647 6486
29c6d1bb
SAS		 6487static void raid5_free_percpu(struct r5conf *conf)
6488{
36d1c647
DW		 6489 if (!conf->percpu)
6490 return;
6491
29c6d1bb 6492 cpuhp_state_remove_instance(CPUHP_MD_RAID5_PREPARE, &conf->node);
36d1c647
DW		 6493 free_percpu(conf->percpu);
6494}
6495
d1688a6d 6496static void free_conf(struct r5conf *conf)
95fc17aa 6497{
d7bd398e
SL		 6498 int i;
6499
5c7e81c3
SL		 6500 if (conf->log)
6501 r5l_exit_log(conf->log);
30c89465 6502 if (conf->shrinker.nr_deferred)
edbe83ab 6503 unregister_shrinker(&conf->shrinker);
5c7e81c3 6504
851c30c9 6505 free_thread_groups(conf);
95fc17aa 6506 shrink_stripes(conf);
36d1c647 6507 raid5_free_percpu(conf);
d7bd398e
SL		 6508 for (i = 0; i < conf->pool_size; i++)
6509 if (conf->disks[i].extra_page)
6510 put_page(conf->disks[i].extra_page);
95fc17aa
DW		 6511 kfree(conf->disks);
6512 kfree(conf->stripe_hashtbl);
6513 kfree(conf);
6514}
6515
29c6d1bb 6516static int raid456_cpu_up_prepare(unsigned int cpu, struct hlist_node *node)
36d1c647 6517{
29c6d1bb 6518 struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node);
36d1c647
DW		 6519 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
6520
29c6d1bb 6521 if (alloc_scratch_buffer(conf, percpu)) {
cc6167b4
N		 6522 pr_warn("%s: failed memory allocation for cpu%u\n",
6523 __func__, cpu);
29c6d1bb 6524 return -ENOMEM;
36d1c647 6525 }
29c6d1bb 6526 return 0;
36d1c647 6527}
36d1c647 6528
d1688a6d 6529static int raid5_alloc_percpu(struct r5conf *conf)
36d1c647 6530{
789b5e03 6531 int err = 0;
36d1c647 6532
789b5e03
ON		 6533 conf->percpu = alloc_percpu(struct raid5_percpu);
6534 if (!conf->percpu)
36d1c647 6535 return -ENOMEM;
789b5e03 6536
29c6d1bb 6537 err = cpuhp_state_add_instance(CPUHP_MD_RAID5_PREPARE, &conf->node);
27a353c0
SL		 6538 if (!err) {
6539 conf->scribble_disks = max(conf->raid_disks,
6540 conf->previous_raid_disks);
6541 conf->scribble_sectors = max(conf->chunk_sectors,
6542 conf->prev_chunk_sectors);
6543 }
36d1c647
DW		 6544 return err;
6545}
6546
edbe83ab
N		 6547static unsigned long raid5_cache_scan(struct shrinker *shrink,
6548 struct shrink_control *sc)
6549{
6550 struct r5conf *conf = container_of(shrink, struct r5conf, shrinker);
2d5b569b
N		 6551 unsigned long ret = SHRINK_STOP;
6552
6553 if (mutex_trylock(&conf->cache_size_mutex)) {
6554 ret= 0;
49895bcc
N		 6555 while (ret < sc->nr_to_scan &&
6556 conf->max_nr_stripes > conf->min_nr_stripes) {
2d5b569b
N		 6557 if (drop_one_stripe(conf) == 0) {
6558 ret = SHRINK_STOP;
6559 break;
6560 }
6561 ret++;
6562 }
6563 mutex_unlock(&conf->cache_size_mutex);
edbe83ab
N		 6564 }
6565 return ret;
6566}
6567
6568static unsigned long raid5_cache_count(struct shrinker *shrink,
6569 struct shrink_control *sc)
6570{
6571 struct r5conf *conf = container_of(shrink, struct r5conf, shrinker);
6572
6573 if (conf->max_nr_stripes < conf->min_nr_stripes)
6574 /* unlikely, but not impossible */
6575 return 0;
6576 return conf->max_nr_stripes - conf->min_nr_stripes;
6577}
6578
d1688a6d 6579static struct r5conf *setup_conf(struct mddev *mddev)
1da177e4 6580{
d1688a6d 6581 struct r5conf *conf;
5e5e3e78 6582 int raid_disk, memory, max_disks;
3cb03002 6583 struct md_rdev *rdev;
1da177e4 6584 struct disk_info *disk;
0232605d 6585 char pers_name[6];
566c09c5 6586 int i;
60aaf933 6587 int group_cnt, worker_cnt_per_group;
6588 struct r5worker_group *new_group;
1da177e4 6589
91adb564
N		 6590 if (mddev->new_level != 5
6591 && mddev->new_level != 4
6592 && mddev->new_level != 6) {
cc6167b4
N		 6593 pr_warn("md/raid:%s: raid level not set to 4/5/6 (%d)\n",
6594 mdname(mddev), mddev->new_level);
91adb564 6595 return ERR_PTR(-EIO);
1da177e4 6596 }
91adb564
N		 6597 if ((mddev->new_level == 5
6598 && !algorithm_valid_raid5(mddev->new_layout)) ||
6599 (mddev->new_level == 6
6600 && !algorithm_valid_raid6(mddev->new_layout))) {
cc6167b4
N		 6601 pr_warn("md/raid:%s: layout %d not supported\n",
6602 mdname(mddev), mddev->new_layout);
91adb564 6603 return ERR_PTR(-EIO);
99c0fb5f 6604 }
91adb564 6605 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
cc6167b4
N		 6606 pr_warn("md/raid:%s: not enough configured devices (%d, minimum 4)\n",
6607 mdname(mddev), mddev->raid_disks);
91adb564 6608 return ERR_PTR(-EINVAL);
4bbf3771
N		 6609 }
6610
664e7c41
AN		 6611 if (!mddev->new_chunk_sectors ||
6612 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
6613 !is_power_of_2(mddev->new_chunk_sectors)) {
cc6167b4
N		 6614 pr_warn("md/raid:%s: invalid chunk size %d\n",
6615 mdname(mddev), mddev->new_chunk_sectors << 9);
91adb564 6616 return ERR_PTR(-EINVAL);
f6705578
N		 6617 }
6618
d1688a6d 6619 conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL);
91adb564 6620 if (conf == NULL)
1da177e4 6621 goto abort;
851c30c9 6622 /* Don't enable multi-threading by default*/
60aaf933 6623 if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group,
6624 &new_group)) {
6625 conf->group_cnt = group_cnt;
6626 conf->worker_cnt_per_group = worker_cnt_per_group;
6627 conf->worker_groups = new_group;
6628 } else
851c30c9 6629 goto abort;
f5efd45a 6630 spin_lock_init(&conf->device_lock);
c46501b2 6631 seqcount_init(&conf->gen_lock);
2d5b569b 6632 mutex_init(&conf->cache_size_mutex);
b1b46486 6633 init_waitqueue_head(&conf->wait_for_quiescent);
6ab2a4b8 6634 init_waitqueue_head(&conf->wait_for_stripe);
f5efd45a
DW		 6635 init_waitqueue_head(&conf->wait_for_overlap);
6636 INIT_LIST_HEAD(&conf->handle_list);
6637 INIT_LIST_HEAD(&conf->hold_list);
6638 INIT_LIST_HEAD(&conf->delayed_list);
6639 INIT_LIST_HEAD(&conf->bitmap_list);
c3cce6cd 6640 bio_list_init(&conf->return_bi);
773ca82f 6641 init_llist_head(&conf->released_stripes);
f5efd45a
DW		 6642 atomic_set(&conf->active_stripes, 0);
6643 atomic_set(&conf->preread_active_stripes, 0);
6644 atomic_set(&conf->active_aligned_reads, 0);
6645 conf->bypass_threshold = BYPASS_THRESHOLD;
d890fa2b 6646 conf->recovery_disabled = mddev->recovery_disabled - 1;
91adb564
N		 6647
6648 conf->raid_disks = mddev->raid_disks;
6649 if (mddev->reshape_position == MaxSector)
6650 conf->previous_raid_disks = mddev->raid_disks;
6651 else
f6705578 6652 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
5e5e3e78 6653 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
f6705578 6654
5e5e3e78 6655 conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
b55e6bfc 6656 GFP_KERNEL);
d7bd398e 6657
b55e6bfc
N		 6658 if (!conf->disks)
6659 goto abort;
9ffae0cf 6660
d7bd398e
SL		 6661 for (i = 0; i < max_disks; i++) {
6662 conf->disks[i].extra_page = alloc_page(GFP_KERNEL);
6663 if (!conf->disks[i].extra_page)
6664 goto abort;
6665 }
6666
1da177e4
LT		 6667 conf->mddev = mddev;
6668
fccddba0 6669 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
1da177e4 6670 goto abort;
1da177e4 6671
566c09c5
SL		 6672 /* We init hash_locks[0] separately to that it can be used
6673 * as the reference lock in the spin_lock_nest_lock() call
6674 * in lock_all_device_hash_locks_irq in order to convince
6675 * lockdep that we know what we are doing.
6676 */
6677 spin_lock_init(conf->hash_locks);
6678 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
6679 spin_lock_init(conf->hash_locks + i);
6680
6681 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
6682 INIT_LIST_HEAD(conf->inactive_list + i);
6683
6684 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
6685 INIT_LIST_HEAD(conf->temp_inactive_list + i);
6686
1e6d690b
SL		 6687 atomic_set(&conf->r5c_cached_full_stripes, 0);
6688 INIT_LIST_HEAD(&conf->r5c_full_stripe_list);
6689 atomic_set(&conf->r5c_cached_partial_stripes, 0);
6690 INIT_LIST_HEAD(&conf->r5c_partial_stripe_list);
6691
36d1c647 6692 conf->level = mddev->new_level;
46d5b785 6693 conf->chunk_sectors = mddev->new_chunk_sectors;
36d1c647
DW		 6694 if (raid5_alloc_percpu(conf) != 0)
6695 goto abort;
6696
0c55e022 6697 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
1da177e4 6698
dafb20fa 6699 rdev_for_each(rdev, mddev) {
1da177e4 6700 raid_disk = rdev->raid_disk;
5e5e3e78 6701 if (raid_disk >= max_disks
f2076e7d 6702 || raid_disk < 0 || test_bit(Journal, &rdev->flags))
1da177e4
LT		 6703 continue;
6704 disk = conf->disks + raid_disk;
6705
17045f52
N		 6706 if (test_bit(Replacement, &rdev->flags)) {
6707 if (disk->replacement)
6708 goto abort;
6709 disk->replacement = rdev;
6710 } else {
6711 if (disk->rdev)
6712 goto abort;
6713 disk->rdev = rdev;
6714 }
1da177e4 6715
b2d444d7 6716 if (test_bit(In_sync, &rdev->flags)) {
1da177e4 6717 char b[BDEVNAME_SIZE];
cc6167b4
N		 6718 pr_info("md/raid:%s: device %s operational as raid disk %d\n",
6719 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
d6b212f4 6720 } else if (rdev->saved_raid_disk != raid_disk)
8c2e870a
NB		 6721 /* Cannot rely on bitmap to complete recovery */
6722 conf->fullsync = 1;
1da177e4
LT		 6723 }
6724
91adb564 6725 conf->level = mddev->new_level;
584acdd4 6726 if (conf->level == 6) {
16a53ecc 6727 conf->max_degraded = 2;
584acdd4
MS		 6728 if (raid6_call.xor_syndrome)
6729 conf->rmw_level = PARITY_ENABLE_RMW;
6730 else
6731 conf->rmw_level = PARITY_DISABLE_RMW;
6732 } else {
16a53ecc 6733 conf->max_degraded = 1;
584acdd4
MS		 6734 conf->rmw_level = PARITY_ENABLE_RMW;
6735 }
91adb564 6736 conf->algorithm = mddev->new_layout;
fef9c61f 6737 conf->reshape_progress = mddev->reshape_position;
e183eaed 6738 if (conf->reshape_progress != MaxSector) {
09c9e5fa 6739 conf->prev_chunk_sectors = mddev->chunk_sectors;
e183eaed 6740 conf->prev_algo = mddev->layout;
5cac6bcb
N		 6741 } else {
6742 conf->prev_chunk_sectors = conf->chunk_sectors;
6743 conf->prev_algo = conf->algorithm;
e183eaed 6744 }
1da177e4 6745
edbe83ab 6746 conf->min_nr_stripes = NR_STRIPES;
ad5b0f76
SL		 6747 if (mddev->reshape_position != MaxSector) {
6748 int stripes = max_t(int,
6749 ((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4,
6750 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4);
6751 conf->min_nr_stripes = max(NR_STRIPES, stripes);
6752 if (conf->min_nr_stripes != NR_STRIPES)
cc6167b4 6753 pr_info("md/raid:%s: force stripe size %d for reshape\n",
ad5b0f76
SL		 6754 mdname(mddev), conf->min_nr_stripes);
6755 }
edbe83ab 6756 memory = conf->min_nr_stripes * (sizeof(struct stripe_head) +
5e5e3e78 6757 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
4bda556a 6758 atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS);
edbe83ab 6759 if (grow_stripes(conf, conf->min_nr_stripes)) {
cc6167b4
N		 6760 pr_warn("md/raid:%s: couldn't allocate %dkB for buffers\n",
6761 mdname(mddev), memory);
91adb564
N		 6762 goto abort;
6763 } else
cc6167b4 6764 pr_debug("md/raid:%s: allocated %dkB\n", mdname(mddev), memory);
edbe83ab
N		 6765 /*
6766 * Losing a stripe head costs more than the time to refill it,
6767 * it reduces the queue depth and so can hurt throughput.
6768 * So set it rather large, scaled by number of devices.
6769 */
6770 conf->shrinker.seeks = DEFAULT_SEEKS * conf->raid_disks * 4;
6771 conf->shrinker.scan_objects = raid5_cache_scan;
6772 conf->shrinker.count_objects = raid5_cache_count;
6773 conf->shrinker.batch = 128;
6774 conf->shrinker.flags = 0;
6a0f53ff 6775 if (register_shrinker(&conf->shrinker)) {
cc6167b4
N		 6776 pr_warn("md/raid:%s: couldn't register shrinker.\n",
6777 mdname(mddev));
6a0f53ff
CY		 6778 goto abort;
6779 }
1da177e4 6780
0232605d
N		 6781 sprintf(pers_name, "raid%d", mddev->new_level);
6782 conf->thread = md_register_thread(raid5d, mddev, pers_name);
91adb564 6783 if (!conf->thread) {
cc6167b4
N		 6784 pr_warn("md/raid:%s: couldn't allocate thread.\n",
6785 mdname(mddev));
16a53ecc
N		 6786 goto abort;
6787 }
91adb564
N		 6788
6789 return conf;
6790
6791 abort:
6792 if (conf) {
95fc17aa 6793 free_conf(conf);
91adb564
N		 6794 return ERR_PTR(-EIO);
6795 } else
6796 return ERR_PTR(-ENOMEM);
6797}
6798
c148ffdc
N		 6799static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
6800{
6801 switch (algo) {
6802 case ALGORITHM_PARITY_0:
6803 if (raid_disk < max_degraded)
6804 return 1;
6805 break;
6806 case ALGORITHM_PARITY_N:
6807 if (raid_disk >= raid_disks - max_degraded)
6808 return 1;
6809 break;
6810 case ALGORITHM_PARITY_0_6:
f72ffdd6 6811 if (raid_disk == 0 ||
c148ffdc
N		 6812 raid_disk == raid_disks - 1)
6813 return 1;
6814 break;
6815 case ALGORITHM_LEFT_ASYMMETRIC_6:
6816 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6817 case ALGORITHM_LEFT_SYMMETRIC_6:
6818 case ALGORITHM_RIGHT_SYMMETRIC_6:
6819 if (raid_disk == raid_disks - 1)
6820 return 1;
6821 }
6822 return 0;
6823}
6824
849674e4 6825static int raid5_run(struct mddev *mddev)
91adb564 6826{
d1688a6d 6827 struct r5conf *conf;
9f7c2220 6828 int working_disks = 0;
c148ffdc 6829 int dirty_parity_disks = 0;
3cb03002 6830 struct md_rdev *rdev;
713cf5a6 6831 struct md_rdev *journal_dev = NULL;
c148ffdc 6832 sector_t reshape_offset = 0;
17045f52 6833 int i;
b5254dd5
N		 6834 long long min_offset_diff = 0;
6835 int first = 1;
91adb564 6836
8c6ac868 6837 if (mddev->recovery_cp != MaxSector)
cc6167b4
N		 6838 pr_notice("md/raid:%s: not clean -- starting background reconstruction\n",
6839 mdname(mddev));
b5254dd5
N		 6840
6841 rdev_for_each(rdev, mddev) {
6842 long long diff;
713cf5a6 6843
f2076e7d 6844 if (test_bit(Journal, &rdev->flags)) {
713cf5a6 6845 journal_dev = rdev;
f2076e7d
SL		 6846 continue;
6847 }
b5254dd5
N		 6848 if (rdev->raid_disk < 0)
6849 continue;
6850 diff = (rdev->new_data_offset - rdev->data_offset);
6851 if (first) {
6852 min_offset_diff = diff;
6853 first = 0;
6854 } else if (mddev->reshape_backwards &&
6855 diff < min_offset_diff)
6856 min_offset_diff = diff;
6857 else if (!mddev->reshape_backwards &&
6858 diff > min_offset_diff)
6859 min_offset_diff = diff;
6860 }
6861
91adb564
N		 6862 if (mddev->reshape_position != MaxSector) {
6863 /* Check that we can continue the reshape.
b5254dd5
N		 6864 * Difficulties arise if the stripe we would write to
6865 * next is at or after the stripe we would read from next.
6866 * For a reshape that changes the number of devices, this
6867 * is only possible for a very short time, and mdadm makes
6868 * sure that time appears to have past before assembling
6869 * the array. So we fail if that time hasn't passed.
6870 * For a reshape that keeps the number of devices the same
6871 * mdadm must be monitoring the reshape can keeping the
6872 * critical areas read-only and backed up. It will start
6873 * the array in read-only mode, so we check for that.
91adb564
N		 6874 */
6875 sector_t here_new, here_old;
6876 int old_disks;
18b00334 6877 int max_degraded = (mddev->level == 6 ? 2 : 1);
05256d98
N		 6878 int chunk_sectors;
6879 int new_data_disks;
91adb564 6880
713cf5a6 6881 if (journal_dev) {
cc6167b4
N		 6882 pr_warn("md/raid:%s: don't support reshape with journal - aborting.\n",
6883 mdname(mddev));
713cf5a6
SL		 6884 return -EINVAL;
6885 }
6886
88ce4930 6887 if (mddev->new_level != mddev->level) {
cc6167b4
N		 6888 pr_warn("md/raid:%s: unsupported reshape required - aborting.\n",
6889 mdname(mddev));
91adb564
N		 6890 return -EINVAL;
6891 }
91adb564
N		 6892 old_disks = mddev->raid_disks - mddev->delta_disks;
6893 /* reshape_position must be on a new-stripe boundary, and one
6894 * further up in new geometry must map after here in old
6895 * geometry.
05256d98
N		 6896 * If the chunk sizes are different, then as we perform reshape
6897 * in units of the largest of the two, reshape_position needs
6898 * be a multiple of the largest chunk size times new data disks.
91adb564
N		 6899 */
6900 here_new = mddev->reshape_position;
05256d98
N		 6901 chunk_sectors = max(mddev->chunk_sectors, mddev->new_chunk_sectors);
6902 new_data_disks = mddev->raid_disks - max_degraded;
6903 if (sector_div(here_new, chunk_sectors * new_data_disks)) {
cc6167b4
N		 6904 pr_warn("md/raid:%s: reshape_position not on a stripe boundary\n",
6905 mdname(mddev));
91adb564
N		 6906 return -EINVAL;
6907 }
05256d98 6908 reshape_offset = here_new * chunk_sectors;
91adb564
N		 6909 /* here_new is the stripe we will write to */
6910 here_old = mddev->reshape_position;
05256d98 6911 sector_div(here_old, chunk_sectors * (old_disks-max_degraded));
91adb564
N		 6912 /* here_old is the first stripe that we might need to read
6913 * from */
67ac6011
N		 6914 if (mddev->delta_disks == 0) {
6915 /* We cannot be sure it is safe to start an in-place
b5254dd5 6916 * reshape. It is only safe if user-space is monitoring
67ac6011
N		 6917 * and taking constant backups.
6918 * mdadm always starts a situation like this in
6919 * readonly mode so it can take control before
6920 * allowing any writes. So just check for that.
6921 */
b5254dd5
N		 6922 if (abs(min_offset_diff) >= mddev->chunk_sectors &&
6923 abs(min_offset_diff) >= mddev->new_chunk_sectors)
6924 /* not really in-place - so OK */;
6925 else if (mddev->ro == 0) {
cc6167b4
N		 6926 pr_warn("md/raid:%s: in-place reshape must be started in read-only mode - aborting\n",
6927 mdname(mddev));
67ac6011
N		 6928 return -EINVAL;
6929 }
2c810cdd 6930 } else if (mddev->reshape_backwards
05256d98
N		 6931 ? (here_new * chunk_sectors + min_offset_diff <=
6932 here_old * chunk_sectors)
6933 : (here_new * chunk_sectors >=
6934 here_old * chunk_sectors + (-min_offset_diff))) {
91adb564 6935 /* Reading from the same stripe as writing to - bad */
cc6167b4
N		 6936 pr_warn("md/raid:%s: reshape_position too early for auto-recovery - aborting.\n",
6937 mdname(mddev));
91adb564
N		 6938 return -EINVAL;
6939 }
cc6167b4 6940 pr_debug("md/raid:%s: reshape will continue\n", mdname(mddev));
91adb564
N		 6941 /* OK, we should be able to continue; */
6942 } else {
6943 BUG_ON(mddev->level != mddev->new_level);
6944 BUG_ON(mddev->layout != mddev->new_layout);
664e7c41 6945 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
91adb564 6946 BUG_ON(mddev->delta_disks != 0);
1da177e4 6947 }
91adb564 6948
245f46c2
N		 6949 if (mddev->private == NULL)
6950 conf = setup_conf(mddev);
6951 else
6952 conf = mddev->private;
6953
91adb564
N		 6954 if (IS_ERR(conf))
6955 return PTR_ERR(conf);
6956
486b0f7b
SL		 6957 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
6958 if (!journal_dev) {
cc6167b4
N		 6959 pr_warn("md/raid:%s: journal disk is missing, force array readonly\n",
6960 mdname(mddev));
486b0f7b
SL		 6961 mddev->ro = 1;
6962 set_disk_ro(mddev->gendisk, 1);
6963 } else if (mddev->recovery_cp == MaxSector)
6964 set_bit(MD_JOURNAL_CLEAN, &mddev->flags);
7dde2ad3
SL		 6965 }
6966
b5254dd5 6967 conf->min_offset_diff = min_offset_diff;
91adb564
N		 6968 mddev->thread = conf->thread;
6969 conf->thread = NULL;
6970 mddev->private = conf;
6971
17045f52
N		 6972 for (i = 0; i < conf->raid_disks && conf->previous_raid_disks;
6973 i++) {
6974 rdev = conf->disks[i].rdev;
6975 if (!rdev && conf->disks[i].replacement) {
6976 /* The replacement is all we have yet */
6977 rdev = conf->disks[i].replacement;
6978 conf->disks[i].replacement = NULL;
6979 clear_bit(Replacement, &rdev->flags);
6980 conf->disks[i].rdev = rdev;
6981 }
6982 if (!rdev)
c148ffdc 6983 continue;
17045f52
N		 6984 if (conf->disks[i].replacement &&
6985 conf->reshape_progress != MaxSector) {
6986 /* replacements and reshape simply do not mix. */
cc6167b4 6987 pr_warn("md: cannot handle concurrent replacement and reshape.\n");
17045f52
N		 6988 goto abort;
6989 }
2f115882 6990 if (test_bit(In_sync, &rdev->flags)) {
91adb564 6991 working_disks++;
2f115882
N		 6992 continue;
6993 }
c148ffdc
N		 6994 /* This disc is not fully in-sync. However if it
6995 * just stored parity (beyond the recovery_offset),
6996 * when we don't need to be concerned about the
6997 * array being dirty.
6998 * When reshape goes 'backwards', we never have
6999 * partially completed devices, so we only need
7000 * to worry about reshape going forwards.
7001 */
7002 /* Hack because v0.91 doesn't store recovery_offset properly. */
7003 if (mddev->major_version == 0 &&
7004 mddev->minor_version > 90)
7005 rdev->recovery_offset = reshape_offset;
5026d7a9 7006
c148ffdc
N		 7007 if (rdev->recovery_offset < reshape_offset) {
7008 /* We need to check old and new layout */
7009 if (!only_parity(rdev->raid_disk,
7010 conf->algorithm,
7011 conf->raid_disks,
7012 conf->max_degraded))
7013 continue;
7014 }
7015 if (!only_parity(rdev->raid_disk,
7016 conf->prev_algo,
7017 conf->previous_raid_disks,
7018 conf->max_degraded))
7019 continue;
7020 dirty_parity_disks++;
7021 }
91adb564 7022
17045f52
N		 7023 /*
7024 * 0 for a fully functional array, 1 or 2 for a degraded array.
7025 */
908f4fbd 7026 mddev->degraded = calc_degraded(conf);
91adb564 7027
674806d6 7028 if (has_failed(conf)) {
cc6167b4 7029 pr_crit("md/raid:%s: not enough operational devices (%d/%d failed)\n",
02c2de8c 7030 mdname(mddev), mddev->degraded, conf->raid_disks);
1da177e4
LT		 7031 goto abort;
7032 }
7033
91adb564 7034 /* device size must be a multiple of chunk size */
9d8f0363 7035 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
91adb564
N		 7036 mddev->resync_max_sectors = mddev->dev_sectors;
7037
c148ffdc 7038 if (mddev->degraded > dirty_parity_disks &&
1da177e4 7039 mddev->recovery_cp != MaxSector) {
6ff8d8ec 7040 if (mddev->ok_start_degraded)
cc6167b4
N		 7041 pr_crit("md/raid:%s: starting dirty degraded array - data corruption possible.\n",
7042 mdname(mddev));
6ff8d8ec 7043 else {
cc6167b4
N		 7044 pr_crit("md/raid:%s: cannot start dirty degraded array.\n",
7045 mdname(mddev));
6ff8d8ec
N		 7046 goto abort;
7047 }
1da177e4
LT		 7048 }
7049
cc6167b4
N		 7050 pr_info("md/raid:%s: raid level %d active with %d out of %d devices, algorithm %d\n",
7051 mdname(mddev), conf->level,
7052 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
7053 mddev->new_layout);
1da177e4
LT		 7054
7055 print_raid5_conf(conf);
7056
fef9c61f 7057 if (conf->reshape_progress != MaxSector) {
fef9c61f 7058 conf->reshape_safe = conf->reshape_progress;
f6705578
N		 7059 atomic_set(&conf->reshape_stripes, 0);
7060 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7061 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7062 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7063 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7064 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 7065 "reshape");
f6705578
N		 7066 }
7067
1da177e4 7068 /* Ok, everything is just fine now */
a64c876f
N		 7069 if (mddev->to_remove == &raid5_attrs_group)
7070 mddev->to_remove = NULL;
00bcb4ac
N		 7071 else if (mddev->kobj.sd &&
7072 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
cc6167b4
N		 7073 pr_warn("raid5: failed to create sysfs attributes for %s\n",
7074 mdname(mddev));
4a5add49 7075 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
7a5febe9 7076
4a5add49 7077 if (mddev->queue) {
9f7c2220 7078 int chunk_size;
620125f2 7079 bool discard_supported = true;
4a5add49
N		 7080 /* read-ahead size must cover two whole stripes, which
7081 * is 2 * (datadisks) * chunksize where 'n' is the
7082 * number of raid devices
7083 */
7084 int data_disks = conf->previous_raid_disks - conf->max_degraded;
7085 int stripe = data_disks *
7086 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
7087 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
7088 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
91adb564 7089
9f7c2220
N		 7090 chunk_size = mddev->chunk_sectors << 9;
7091 blk_queue_io_min(mddev->queue, chunk_size);
7092 blk_queue_io_opt(mddev->queue, chunk_size *
7093 (conf->raid_disks - conf->max_degraded));
c78afc62 7094 mddev->queue->limits.raid_partial_stripes_expensive = 1;
620125f2
SL		 7095 /*
7096 * We can only discard a whole stripe. It doesn't make sense to
7097 * discard data disk but write parity disk
7098 */
7099 stripe = stripe * PAGE_SIZE;
4ac6875e
N		 7100 /* Round up to power of 2, as discard handling
7101 * currently assumes that */
7102 while ((stripe-1) & stripe)
7103 stripe = (stripe | (stripe-1)) + 1;
620125f2
SL		 7104 mddev->queue->limits.discard_alignment = stripe;
7105 mddev->queue->limits.discard_granularity = stripe;
e8d7c332
KK		 7106
7107 /*
7108 * We use 16-bit counter of active stripes in bi_phys_segments
7109 * (minus one for over-loaded initialization)
7110 */
7111 blk_queue_max_hw_sectors(mddev->queue, 0xfffe * STRIPE_SECTORS);
7112 blk_queue_max_discard_sectors(mddev->queue,
7113 0xfffe * STRIPE_SECTORS);
7114
620125f2
SL		 7115 /*
7116 * unaligned part of discard request will be ignored, so can't
8e0e99ba 7117 * guarantee discard_zeroes_data
620125f2
SL		 7118 */
7119 mddev->queue->limits.discard_zeroes_data = 0;
8f6c2e4b 7120
5026d7a9
PA		 7121 blk_queue_max_write_same_sectors(mddev->queue, 0);
7122
05616be5 7123 rdev_for_each(rdev, mddev) {
9f7c2220
N		 7124 disk_stack_limits(mddev->gendisk, rdev->bdev,
7125 rdev->data_offset << 9);
05616be5
N		 7126 disk_stack_limits(mddev->gendisk, rdev->bdev,
7127 rdev->new_data_offset << 9);
620125f2
SL		 7128 /*
7129 * discard_zeroes_data is required, otherwise data
7130 * could be lost. Consider a scenario: discard a stripe
7131 * (the stripe could be inconsistent if
7132 * discard_zeroes_data is 0); write one disk of the
7133 * stripe (the stripe could be inconsistent again
7134 * depending on which disks are used to calculate
7135 * parity); the disk is broken; The stripe data of this
7136 * disk is lost.
7137 */
7138 if (!blk_queue_discard(bdev_get_queue(rdev->bdev)) ||
7139 !bdev_get_queue(rdev->bdev)->
7140 limits.discard_zeroes_data)
7141 discard_supported = false;
8e0e99ba
N		 7142 /* Unfortunately, discard_zeroes_data is not currently
7143 * a guarantee - just a hint. So we only allow DISCARD
7144 * if the sysadmin has confirmed that only safe devices
7145 * are in use by setting a module parameter.
7146 */
7147 if (!devices_handle_discard_safely) {
7148 if (discard_supported) {
7149 pr_info("md/raid456: discard support disabled due to uncertainty.\n");
7150 pr_info("Set raid456.devices_handle_discard_safely=Y to override.\n");
7151 }
7152 discard_supported = false;
7153 }
05616be5 7154 }
620125f2
SL		 7155
7156 if (discard_supported &&
e7597e69
JS		 7157 mddev->queue->limits.max_discard_sectors >= (stripe >> 9) &&
7158 mddev->queue->limits.discard_granularity >= stripe)
620125f2
SL		 7159 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
7160 mddev->queue);
7161 else
7162 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
7163 mddev->queue);
1dffdddd
SL		 7164
7165 blk_queue_max_hw_sectors(mddev->queue, UINT_MAX);
9f7c2220 7166 }
23032a0e 7167
5c7e81c3
SL		 7168 if (journal_dev) {
7169 char b[BDEVNAME_SIZE];
7170
cc6167b4
N		 7171 pr_debug("md/raid:%s: using device %s as journal\n",
7172 mdname(mddev), bdevname(journal_dev->bdev, b));
5aabf7c4
SL		 7173 if (r5l_init_log(conf, journal_dev))
7174 goto abort;
5c7e81c3
SL		 7175 }
7176
1da177e4
LT		 7177 return 0;
7178abort:
01f96c0a 7179 md_unregister_thread(&mddev->thread);
e4f869d9
N		 7180 print_raid5_conf(conf);
7181 free_conf(conf);
1da177e4 7182 mddev->private = NULL;
cc6167b4 7183 pr_warn("md/raid:%s: failed to run raid set.\n", mdname(mddev));
1da177e4
LT		 7184 return -EIO;
7185}
7186
afa0f557 7187static void raid5_free(struct mddev *mddev, void *priv)
1da177e4 7188{
afa0f557 7189 struct r5conf *conf = priv;
1da177e4 7190
95fc17aa 7191 free_conf(conf);
a64c876f 7192 mddev->to_remove = &raid5_attrs_group;
1da177e4
LT		 7193}
7194
849674e4 7195static void raid5_status(struct seq_file *seq, struct mddev *mddev)
1da177e4 7196{
d1688a6d 7197 struct r5conf *conf = mddev->private;
1da177e4
LT		 7198 int i;
7199
9d8f0363 7200 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
3cb5edf4 7201 conf->chunk_sectors / 2, mddev->layout);
02c2de8c 7202 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
5fd13351
N		 7203 rcu_read_lock();
7204 for (i = 0; i < conf->raid_disks; i++) {
7205 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
7206 seq_printf (seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
7207 }
7208 rcu_read_unlock();
1da177e4 7209 seq_printf (seq, "]");
1da177e4
LT		 7210}
7211
d1688a6d 7212static void print_raid5_conf (struct r5conf *conf)
1da177e4
LT		 7213{
7214 int i;
7215 struct disk_info *tmp;
7216
cc6167b4 7217 pr_debug("RAID conf printout:\n");
1da177e4 7218 if (!conf) {
cc6167b4 7219 pr_debug("(conf==NULL)\n");
1da177e4
LT		 7220 return;
7221 }
cc6167b4 7222 pr_debug(" --- level:%d rd:%d wd:%d\n", conf->level,
0c55e022
N		 7223 conf->raid_disks,
7224 conf->raid_disks - conf->mddev->degraded);
1da177e4
LT		 7225
7226 for (i = 0; i < conf->raid_disks; i++) {
7227 char b[BDEVNAME_SIZE];
7228 tmp = conf->disks + i;
7229 if (tmp->rdev)
cc6167b4 7230 pr_debug(" disk %d, o:%d, dev:%s\n",
0c55e022
N		 7231 i, !test_bit(Faulty, &tmp->rdev->flags),
7232 bdevname(tmp->rdev->bdev, b));
1da177e4
LT		 7233 }
7234}
7235
fd01b88c 7236static int raid5_spare_active(struct mddev *mddev)
1da177e4
LT		 7237{
7238 int i;
d1688a6d 7239 struct r5conf *conf = mddev->private;
1da177e4 7240 struct disk_info *tmp;
6b965620
N		 7241 int count = 0;
7242 unsigned long flags;
1da177e4
LT		 7243
7244 for (i = 0; i < conf->raid_disks; i++) {
7245 tmp = conf->disks + i;
dd054fce
N		 7246 if (tmp->replacement
7247 && tmp->replacement->recovery_offset == MaxSector
7248 && !test_bit(Faulty, &tmp->replacement->flags)
7249 && !test_and_set_bit(In_sync, &tmp->replacement->flags)) {
7250 /* Replacement has just become active. */
7251 if (!tmp->rdev
7252 || !test_and_clear_bit(In_sync, &tmp->rdev->flags))
7253 count++;
7254 if (tmp->rdev) {
7255 /* Replaced device not technically faulty,
7256 * but we need to be sure it gets removed
7257 * and never re-added.
7258 */
7259 set_bit(Faulty, &tmp->rdev->flags);
7260 sysfs_notify_dirent_safe(
7261 tmp->rdev->sysfs_state);
7262 }
7263 sysfs_notify_dirent_safe(tmp->replacement->sysfs_state);
7264 } else if (tmp->rdev
70fffd0b 7265 && tmp->rdev->recovery_offset == MaxSector
b2d444d7 7266 && !test_bit(Faulty, &tmp->rdev->flags)
c04be0aa 7267 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
6b965620 7268 count++;
43c73ca4 7269 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
1da177e4
LT		 7270 }
7271 }
6b965620 7272 spin_lock_irqsave(&conf->device_lock, flags);
908f4fbd 7273 mddev->degraded = calc_degraded(conf);
6b965620 7274 spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4 7275 print_raid5_conf(conf);
6b965620 7276 return count;
1da177e4
LT		 7277}
7278
b8321b68 7279static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7280{
d1688a6d 7281 struct r5conf *conf = mddev->private;
1da177e4 7282 int err = 0;
b8321b68 7283 int number = rdev->raid_disk;
657e3e4d 7284 struct md_rdev **rdevp;
1da177e4
LT		 7285 struct disk_info *p = conf->disks + number;
7286
7287 print_raid5_conf(conf);
f6b6ec5c
SL		 7288 if (test_bit(Journal, &rdev->flags) && conf->log) {
7289 struct r5l_log *log;
c2bb6242 7290 /*
f6b6ec5c
SL		 7291 * we can't wait pending write here, as this is called in
7292 * raid5d, wait will deadlock.
c2bb6242 7293 */
f6b6ec5c
SL		 7294 if (atomic_read(&mddev->writes_pending))
7295 return -EBUSY;
7296 log = conf->log;
7297 conf->log = NULL;
7298 synchronize_rcu();
7299 r5l_exit_log(log);
7300 return 0;
c2bb6242 7301 }
657e3e4d
N		 7302 if (rdev == p->rdev)
7303 rdevp = &p->rdev;
7304 else if (rdev == p->replacement)
7305 rdevp = &p->replacement;
7306 else
7307 return 0;
7308
7309 if (number >= conf->raid_disks &&
7310 conf->reshape_progress == MaxSector)
7311 clear_bit(In_sync, &rdev->flags);
7312
7313 if (test_bit(In_sync, &rdev->flags) ||
7314 atomic_read(&rdev->nr_pending)) {
7315 err = -EBUSY;
7316 goto abort;
7317 }
7318 /* Only remove non-faulty devices if recovery
7319 * isn't possible.
7320 */
7321 if (!test_bit(Faulty, &rdev->flags) &&
7322 mddev->recovery_disabled != conf->recovery_disabled &&
7323 !has_failed(conf) &&
dd054fce 7324 (!p->replacement || p->replacement == rdev) &&
657e3e4d
N		 7325 number < conf->raid_disks) {
7326 err = -EBUSY;
7327 goto abort;
7328 }
7329 *rdevp = NULL;
d787be40
N		 7330 if (!test_bit(RemoveSynchronized, &rdev->flags)) {
7331 synchronize_rcu();
7332 if (atomic_read(&rdev->nr_pending)) {
7333 /* lost the race, try later */
7334 err = -EBUSY;
7335 *rdevp = rdev;
7336 }
7337 }
7338 if (p->replacement) {
dd054fce
N		 7339 /* We must have just cleared 'rdev' */
7340 p->rdev = p->replacement;
7341 clear_bit(Replacement, &p->replacement->flags);
7342 smp_mb(); /* Make sure other CPUs may see both as identical
7343 * but will never see neither - if they are careful
7344 */
7345 p->replacement = NULL;
7346 clear_bit(WantReplacement, &rdev->flags);
7347 } else
7348 /* We might have just removed the Replacement as faulty-
7349 * clear the bit just in case
7350 */
7351 clear_bit(WantReplacement, &rdev->flags);
1da177e4
LT		 7352abort:
7353
7354 print_raid5_conf(conf);
7355 return err;
7356}
7357
fd01b88c 7358static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7359{
d1688a6d 7360 struct r5conf *conf = mddev->private;
199050ea 7361 int err = -EEXIST;
1da177e4
LT		 7362 int disk;
7363 struct disk_info *p;
6c2fce2e
NB		 7364 int first = 0;
7365 int last = conf->raid_disks - 1;
1da177e4 7366
f6b6ec5c
SL		 7367 if (test_bit(Journal, &rdev->flags)) {
7368 char b[BDEVNAME_SIZE];
7369 if (conf->log)
7370 return -EBUSY;
7371
7372 rdev->raid_disk = 0;
7373 /*
7374 * The array is in readonly mode if journal is missing, so no
7375 * write requests running. We should be safe
7376 */
7377 r5l_init_log(conf, rdev);
cc6167b4
N		 7378 pr_debug("md/raid:%s: using device %s as journal\n",
7379 mdname(mddev), bdevname(rdev->bdev, b));
f6b6ec5c
SL		 7380 return 0;
7381 }
7f0da59b
N		 7382 if (mddev->recovery_disabled == conf->recovery_disabled)
7383 return -EBUSY;
7384
dc10c643 7385 if (rdev->saved_raid_disk < 0 && has_failed(conf))
1da177e4 7386 /* no point adding a device */
199050ea 7387 return -EINVAL;
1da177e4 7388
6c2fce2e
NB		 7389 if (rdev->raid_disk >= 0)
7390 first = last = rdev->raid_disk;
1da177e4
LT		 7391
7392 /*
16a53ecc
N		 7393 * find the disk ... but prefer rdev->saved_raid_disk
7394 * if possible.
1da177e4 7395 */
16a53ecc 7396 if (rdev->saved_raid_disk >= 0 &&
6c2fce2e 7397 rdev->saved_raid_disk >= first &&
16a53ecc 7398 conf->disks[rdev->saved_raid_disk].rdev == NULL)
5cfb22a1
N		 7399 first = rdev->saved_raid_disk;
7400
7401 for (disk = first; disk <= last; disk++) {
7bfec5f3
N		 7402 p = conf->disks + disk;
7403 if (p->rdev == NULL) {
b2d444d7 7404 clear_bit(In_sync, &rdev->flags);
1da177e4 7405 rdev->raid_disk = disk;
199050ea 7406 err = 0;
72626685
N		 7407 if (rdev->saved_raid_disk != disk)
7408 conf->fullsync = 1;
d6065f7b 7409 rcu_assign_pointer(p->rdev, rdev);
5cfb22a1 7410 goto out;
1da177e4 7411 }
5cfb22a1
N		 7412 }
7413 for (disk = first; disk <= last; disk++) {
7414 p = conf->disks + disk;
7bfec5f3
N		 7415 if (test_bit(WantReplacement, &p->rdev->flags) &&
7416 p->replacement == NULL) {
7417 clear_bit(In_sync, &rdev->flags);
7418 set_bit(Replacement, &rdev->flags);
7419 rdev->raid_disk = disk;
7420 err = 0;
7421 conf->fullsync = 1;
7422 rcu_assign_pointer(p->replacement, rdev);
7423 break;
7424 }
7425 }
5cfb22a1 7426out:
1da177e4 7427 print_raid5_conf(conf);
199050ea 7428 return err;
1da177e4
LT		 7429}
7430
fd01b88c 7431static int raid5_resize(struct mddev *mddev, sector_t sectors)
1da177e4
LT		 7432{
7433 /* no resync is happening, and there is enough space
7434 * on all devices, so we can resize.
7435 * We need to make sure resync covers any new space.
7436 * If the array is shrinking we should possibly wait until
7437 * any io in the removed space completes, but it hardly seems
7438 * worth it.
7439 */
a4a6125a 7440 sector_t newsize;
3cb5edf4
N		 7441 struct r5conf *conf = mddev->private;
7442
713cf5a6
SL		 7443 if (conf->log)
7444 return -EINVAL;
3cb5edf4 7445 sectors &= ~((sector_t)conf->chunk_sectors - 1);
a4a6125a
N		 7446 newsize = raid5_size(mddev, sectors, mddev->raid_disks);
7447 if (mddev->external_size &&
7448 mddev->array_sectors > newsize)
b522adcd 7449 return -EINVAL;
a4a6125a
N		 7450 if (mddev->bitmap) {
7451 int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0);
7452 if (ret)
7453 return ret;
7454 }
7455 md_set_array_sectors(mddev, newsize);
f233ea5c 7456 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 7457 revalidate_disk(mddev->gendisk);
b098636c
N		 7458 if (sectors > mddev->dev_sectors &&
7459 mddev->recovery_cp > mddev->dev_sectors) {
58c0fed4 7460 mddev->recovery_cp = mddev->dev_sectors;
1da177e4
LT		 7461 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7462 }
58c0fed4 7463 mddev->dev_sectors = sectors;
4b5c7ae8 7464 mddev->resync_max_sectors = sectors;
1da177e4
LT		 7465 return 0;
7466}
7467
fd01b88c 7468static int check_stripe_cache(struct mddev *mddev)
01ee22b4
N		 7469{
7470 /* Can only proceed if there are plenty of stripe_heads.
7471 * We need a minimum of one full stripe,, and for sensible progress
7472 * it is best to have about 4 times that.
7473 * If we require 4 times, then the default 256 4K stripe_heads will
7474 * allow for chunk sizes up to 256K, which is probably OK.
7475 * If the chunk size is greater, user-space should request more
7476 * stripe_heads first.
7477 */
d1688a6d 7478 struct r5conf *conf = mddev->private;
01ee22b4 7479 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
edbe83ab 7480 > conf->min_nr_stripes ||
01ee22b4 7481 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
edbe83ab 7482 > conf->min_nr_stripes) {
cc6167b4
N		 7483 pr_warn("md/raid:%s: reshape: not enough stripes. Needed %lu\n",
7484 mdname(mddev),
7485 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
7486 / STRIPE_SIZE)*4);
01ee22b4
N		 7487 return 0;
7488 }
7489 return 1;
7490}
7491
fd01b88c 7492static int check_reshape(struct mddev *mddev)
29269553 7493{
d1688a6d 7494 struct r5conf *conf = mddev->private;
29269553 7495
713cf5a6
SL		 7496 if (conf->log)
7497 return -EINVAL;
88ce4930
N		 7498 if (mddev->delta_disks == 0 &&
7499 mddev->new_layout == mddev->layout &&
664e7c41 7500 mddev->new_chunk_sectors == mddev->chunk_sectors)
50ac168a 7501 return 0; /* nothing to do */
674806d6 7502 if (has_failed(conf))
ec32a2bd 7503 return -EINVAL;
fdcfbbb6 7504 if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) {
ec32a2bd
N		 7505 /* We might be able to shrink, but the devices must
7506 * be made bigger first.
7507 * For raid6, 4 is the minimum size.
7508 * Otherwise 2 is the minimum
7509 */
7510 int min = 2;
7511 if (mddev->level == 6)
7512 min = 4;
7513 if (mddev->raid_disks + mddev->delta_disks < min)
7514 return -EINVAL;
7515 }
29269553 7516
01ee22b4 7517 if (!check_stripe_cache(mddev))
29269553 7518 return -ENOSPC;
29269553 7519
738a2738
N		 7520 if (mddev->new_chunk_sectors > mddev->chunk_sectors ||
7521 mddev->delta_disks > 0)
7522 if (resize_chunks(conf,
7523 conf->previous_raid_disks
7524 + max(0, mddev->delta_disks),
7525 max(mddev->new_chunk_sectors,
7526 mddev->chunk_sectors)
7527 ) < 0)
7528 return -ENOMEM;
e56108d6
N		 7529 return resize_stripes(conf, (conf->previous_raid_disks
7530 + mddev->delta_disks));
63c70c4f
N		 7531}
7532
fd01b88c 7533static int raid5_start_reshape(struct mddev *mddev)
63c70c4f 7534{
d1688a6d 7535 struct r5conf *conf = mddev->private;
3cb03002 7536 struct md_rdev *rdev;
63c70c4f 7537 int spares = 0;
c04be0aa 7538 unsigned long flags;
63c70c4f 7539
f416885e 7540 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
63c70c4f
N		 7541 return -EBUSY;
7542
01ee22b4
N		 7543 if (!check_stripe_cache(mddev))
7544 return -ENOSPC;
7545
30b67645
N		 7546 if (has_failed(conf))
7547 return -EINVAL;
7548
c6563a8c 7549 rdev_for_each(rdev, mddev) {
469518a3
N		 7550 if (!test_bit(In_sync, &rdev->flags)
7551 && !test_bit(Faulty, &rdev->flags))
29269553 7552 spares++;
c6563a8c 7553 }
63c70c4f 7554
f416885e 7555 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
29269553
N		 7556 /* Not enough devices even to make a degraded array
7557 * of that size
7558 */
7559 return -EINVAL;
7560
ec32a2bd
N		 7561 /* Refuse to reduce size of the array. Any reductions in
7562 * array size must be through explicit setting of array_size
7563 * attribute.
7564 */
7565 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
7566 < mddev->array_sectors) {
cc6167b4
N		 7567 pr_warn("md/raid:%s: array size must be reduced before number of disks\n",
7568 mdname(mddev));
ec32a2bd
N		 7569 return -EINVAL;
7570 }
7571
f6705578 7572 atomic_set(&conf->reshape_stripes, 0);
29269553 7573 spin_lock_irq(&conf->device_lock);
c46501b2 7574 write_seqcount_begin(&conf->gen_lock);
29269553 7575 conf->previous_raid_disks = conf->raid_disks;
63c70c4f 7576 conf->raid_disks += mddev->delta_disks;
09c9e5fa
AN		 7577 conf->prev_chunk_sectors = conf->chunk_sectors;
7578 conf->chunk_sectors = mddev->new_chunk_sectors;
88ce4930
N		 7579 conf->prev_algo = conf->algorithm;
7580 conf->algorithm = mddev->new_layout;
05616be5
N		 7581 conf->generation++;
7582 /* Code that selects data_offset needs to see the generation update
7583 * if reshape_progress has been set - so a memory barrier needed.
7584 */
7585 smp_mb();
2c810cdd 7586 if (mddev->reshape_backwards)
fef9c61f
N		 7587 conf->reshape_progress = raid5_size(mddev, 0, 0);
7588 else
7589 conf->reshape_progress = 0;
7590 conf->reshape_safe = conf->reshape_progress;
c46501b2 7591 write_seqcount_end(&conf->gen_lock);
29269553
N		 7592 spin_unlock_irq(&conf->device_lock);
7593
4d77e3ba
N		 7594 /* Now make sure any requests that proceeded on the assumption
7595 * the reshape wasn't running - like Discard or Read - have
7596 * completed.
7597 */
7598 mddev_suspend(mddev);
7599 mddev_resume(mddev);
7600
29269553
N		 7601 /* Add some new drives, as many as will fit.
7602 * We know there are enough to make the newly sized array work.
3424bf6a
N		 7603 * Don't add devices if we are reducing the number of
7604 * devices in the array. This is because it is not possible
7605 * to correctly record the "partially reconstructed" state of
7606 * such devices during the reshape and confusion could result.
29269553 7607 */
87a8dec9 7608 if (mddev->delta_disks >= 0) {
dafb20fa 7609 rdev_for_each(rdev, mddev)
87a8dec9
N		 7610 if (rdev->raid_disk < 0 &&
7611 !test_bit(Faulty, &rdev->flags)) {
7612 if (raid5_add_disk(mddev, rdev) == 0) {
87a8dec9 7613 if (rdev->raid_disk
9d4c7d87 7614 >= conf->previous_raid_disks)
87a8dec9 7615 set_bit(In_sync, &rdev->flags);
9d4c7d87 7616 else
87a8dec9 7617 rdev->recovery_offset = 0;
36fad858
NK		 7618
7619 if (sysfs_link_rdev(mddev, rdev))
87a8dec9 7620 /* Failure here is OK */;
50da0840 7621 }
87a8dec9
N		 7622 } else if (rdev->raid_disk >= conf->previous_raid_disks
7623 && !test_bit(Faulty, &rdev->flags)) {
7624 /* This is a spare that was manually added */
7625 set_bit(In_sync, &rdev->flags);
87a8dec9 7626 }
29269553 7627
87a8dec9
N		 7628 /* When a reshape changes the number of devices,
7629 * ->degraded is measured against the larger of the
7630 * pre and post number of devices.
7631 */
ec32a2bd 7632 spin_lock_irqsave(&conf->device_lock, flags);
908f4fbd 7633 mddev->degraded = calc_degraded(conf);
ec32a2bd
N		 7634 spin_unlock_irqrestore(&conf->device_lock, flags);
7635 }
63c70c4f 7636 mddev->raid_disks = conf->raid_disks;
e516402c 7637 mddev->reshape_position = conf->reshape_progress;
2953079c 7638 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
f6705578 7639
29269553
N		 7640 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7641 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
ea358cd0 7642 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
29269553
N		 7643 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7644 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7645 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 7646 "reshape");
29269553
N		 7647 if (!mddev->sync_thread) {
7648 mddev->recovery = 0;
7649 spin_lock_irq(&conf->device_lock);
ba8805b9 7650 write_seqcount_begin(&conf->gen_lock);
29269553 7651 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
ba8805b9
N		 7652 mddev->new_chunk_sectors =
7653 conf->chunk_sectors = conf->prev_chunk_sectors;
7654 mddev->new_layout = conf->algorithm = conf->prev_algo;
05616be5
N		 7655 rdev_for_each(rdev, mddev)
7656 rdev->new_data_offset = rdev->data_offset;
7657 smp_wmb();
ba8805b9 7658 conf->generation --;
fef9c61f 7659 conf->reshape_progress = MaxSector;
1e3fa9bd 7660 mddev->reshape_position = MaxSector;
ba8805b9 7661 write_seqcount_end(&conf->gen_lock);
29269553
N		 7662 spin_unlock_irq(&conf->device_lock);
7663 return -EAGAIN;
7664 }
c8f517c4 7665 conf->reshape_checkpoint = jiffies;
29269553
N		 7666 md_wakeup_thread(mddev->sync_thread);
7667 md_new_event(mddev);
7668 return 0;
7669}
29269553 7670
ec32a2bd
N		 7671/* This is called from the reshape thread and should make any
7672 * changes needed in 'conf'
7673 */
d1688a6d 7674static void end_reshape(struct r5conf *conf)
29269553 7675{
29269553 7676
f6705578 7677 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
05616be5 7678 struct md_rdev *rdev;
f6705578 7679
f6705578 7680 spin_lock_irq(&conf->device_lock);
cea9c228 7681 conf->previous_raid_disks = conf->raid_disks;
05616be5
N		 7682 rdev_for_each(rdev, conf->mddev)
7683 rdev->data_offset = rdev->new_data_offset;
7684 smp_wmb();
fef9c61f 7685 conf->reshape_progress = MaxSector;
6cbd8148 7686 conf->mddev->reshape_position = MaxSector;
f6705578 7687 spin_unlock_irq(&conf->device_lock);
b0f9ec04 7688 wake_up(&conf->wait_for_overlap);
16a53ecc
N		 7689
7690 /* read-ahead size must cover two whole stripes, which is
7691 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
7692 */
4a5add49 7693 if (conf->mddev->queue) {
cea9c228 7694 int data_disks = conf->raid_disks - conf->max_degraded;
09c9e5fa 7695 int stripe = data_disks * ((conf->chunk_sectors << 9)
cea9c228 7696 / PAGE_SIZE);
16a53ecc
N		 7697 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
7698 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
7699 }
29269553 7700 }
29269553
N		 7701}
7702
ec32a2bd
N		 7703/* This is called from the raid5d thread with mddev_lock held.
7704 * It makes config changes to the device.
7705 */
fd01b88c 7706static void raid5_finish_reshape(struct mddev *mddev)
cea9c228 7707{
d1688a6d 7708 struct r5conf *conf = mddev->private;
cea9c228
N		 7709
7710 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7711
ec32a2bd
N		 7712 if (mddev->delta_disks > 0) {
7713 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
fe67d19a
HM		 7714 if (mddev->queue) {
7715 set_capacity(mddev->gendisk, mddev->array_sectors);
7716 revalidate_disk(mddev->gendisk);
7717 }
ec32a2bd
N		 7718 } else {
7719 int d;
908f4fbd
N		 7720 spin_lock_irq(&conf->device_lock);
7721 mddev->degraded = calc_degraded(conf);
7722 spin_unlock_irq(&conf->device_lock);
ec32a2bd
N		 7723 for (d = conf->raid_disks ;
7724 d < conf->raid_disks - mddev->delta_disks;
1a67dde0 7725 d++) {
3cb03002 7726 struct md_rdev *rdev = conf->disks[d].rdev;
da7613b8
N		 7727 if (rdev)
7728 clear_bit(In_sync, &rdev->flags);
7729 rdev = conf->disks[d].replacement;
7730 if (rdev)
7731 clear_bit(In_sync, &rdev->flags);
1a67dde0 7732 }
cea9c228 7733 }
88ce4930 7734 mddev->layout = conf->algorithm;
09c9e5fa 7735 mddev->chunk_sectors = conf->chunk_sectors;
ec32a2bd
N		 7736 mddev->reshape_position = MaxSector;
7737 mddev->delta_disks = 0;
2c810cdd 7738 mddev->reshape_backwards = 0;
cea9c228
N		 7739 }
7740}
7741
fd01b88c 7742static void raid5_quiesce(struct mddev *mddev, int state)
72626685 7743{
d1688a6d 7744 struct r5conf *conf = mddev->private;
72626685
N		 7745
7746 switch(state) {
e464eafd
N		 7747 case 2: /* resume for a suspend */
7748 wake_up(&conf->wait_for_overlap);
7749 break;
7750
72626685 7751 case 1: /* stop all writes */
566c09c5 7752 lock_all_device_hash_locks_irq(conf);
64bd660b
N		 7753 /* '2' tells resync/reshape to pause so that all
7754 * active stripes can drain
7755 */
a39f7afd 7756 r5c_flush_cache(conf, INT_MAX);
64bd660b 7757 conf->quiesce = 2;
b1b46486 7758 wait_event_cmd(conf->wait_for_quiescent,
46031f9a
RBJ		 7759 atomic_read(&conf->active_stripes) == 0 &&
7760 atomic_read(&conf->active_aligned_reads) == 0,
566c09c5
SL		 7761 unlock_all_device_hash_locks_irq(conf),
7762 lock_all_device_hash_locks_irq(conf));
64bd660b 7763 conf->quiesce = 1;
566c09c5 7764 unlock_all_device_hash_locks_irq(conf);
64bd660b
N		 7765 /* allow reshape to continue */
7766 wake_up(&conf->wait_for_overlap);
72626685
N		 7767 break;
7768
7769 case 0: /* re-enable writes */
566c09c5 7770 lock_all_device_hash_locks_irq(conf);
72626685 7771 conf->quiesce = 0;
b1b46486 7772 wake_up(&conf->wait_for_quiescent);
e464eafd 7773 wake_up(&conf->wait_for_overlap);
566c09c5 7774 unlock_all_device_hash_locks_irq(conf);
72626685
N		 7775 break;
7776 }
e6c033f7 7777 r5l_quiesce(conf->log, state);
72626685 7778}
b15c2e57 7779
fd01b88c 7780static void *raid45_takeover_raid0(struct mddev *mddev, int level)
54071b38 7781{
e373ab10 7782 struct r0conf *raid0_conf = mddev->private;
d76c8420 7783 sector_t sectors;
54071b38 7784
f1b29bca 7785 /* for raid0 takeover only one zone is supported */
e373ab10 7786 if (raid0_conf->nr_strip_zones > 1) {
cc6167b4
N		 7787 pr_warn("md/raid:%s: cannot takeover raid0 with more than one zone.\n",
7788 mdname(mddev));
f1b29bca
DW		 7789 return ERR_PTR(-EINVAL);
7790 }
7791
e373ab10
N		 7792 sectors = raid0_conf->strip_zone[0].zone_end;
7793 sector_div(sectors, raid0_conf->strip_zone[0].nb_dev);
3b71bd93 7794 mddev->dev_sectors = sectors;
f1b29bca 7795 mddev->new_level = level;
54071b38
TM		 7796 mddev->new_layout = ALGORITHM_PARITY_N;
7797 mddev->new_chunk_sectors = mddev->chunk_sectors;
7798 mddev->raid_disks += 1;
7799 mddev->delta_disks = 1;
7800 /* make sure it will be not marked as dirty */
7801 mddev->recovery_cp = MaxSector;
7802
7803 return setup_conf(mddev);
7804}
7805
fd01b88c 7806static void *raid5_takeover_raid1(struct mddev *mddev)
d562b0c4
N		 7807{
7808 int chunksect;
6995f0b2 7809 void *ret;
d562b0c4
N		 7810
7811 if (mddev->raid_disks != 2 ||
7812 mddev->degraded > 1)
7813 return ERR_PTR(-EINVAL);
7814
7815 /* Should check if there are write-behind devices? */
7816
7817 chunksect = 64*2; /* 64K by default */
7818
7819 /* The array must be an exact multiple of chunksize */
7820 while (chunksect && (mddev->array_sectors & (chunksect-1)))
7821 chunksect >>= 1;
7822
7823 if ((chunksect<<9) < STRIPE_SIZE)
7824 /* array size does not allow a suitable chunk size */
7825 return ERR_PTR(-EINVAL);
7826
7827 mddev->new_level = 5;
7828 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
664e7c41 7829 mddev->new_chunk_sectors = chunksect;
d562b0c4 7830
6995f0b2
SL		 7831 ret = setup_conf(mddev);
7832 if (!IS_ERR_VALUE(ret))
7833 clear_bit(MD_FAILFAST_SUPPORTED, &mddev->flags);
7834 return ret;
d562b0c4
N		 7835}
7836
fd01b88c 7837static void *raid5_takeover_raid6(struct mddev *mddev)
fc9739c6
N		 7838{
7839 int new_layout;
7840
7841 switch (mddev->layout) {
7842 case ALGORITHM_LEFT_ASYMMETRIC_6:
7843 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
7844 break;
7845 case ALGORITHM_RIGHT_ASYMMETRIC_6:
7846 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
7847 break;
7848 case ALGORITHM_LEFT_SYMMETRIC_6:
7849 new_layout = ALGORITHM_LEFT_SYMMETRIC;
7850 break;
7851 case ALGORITHM_RIGHT_SYMMETRIC_6:
7852 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
7853 break;
7854 case ALGORITHM_PARITY_0_6:
7855 new_layout = ALGORITHM_PARITY_0;
7856 break;
7857 case ALGORITHM_PARITY_N:
7858 new_layout = ALGORITHM_PARITY_N;
7859 break;
7860 default:
7861 return ERR_PTR(-EINVAL);
7862 }
7863 mddev->new_level = 5;
7864 mddev->new_layout = new_layout;
7865 mddev->delta_disks = -1;
7866 mddev->raid_disks -= 1;
7867 return setup_conf(mddev);
7868}
7869
fd01b88c 7870static int raid5_check_reshape(struct mddev *mddev)
b3546035 7871{
88ce4930
N		 7872 /* For a 2-drive array, the layout and chunk size can be changed
7873 * immediately as not restriping is needed.
7874 * For larger arrays we record the new value - after validation
7875 * to be used by a reshape pass.
b3546035 7876 */
d1688a6d 7877 struct r5conf *conf = mddev->private;
597a711b 7878 int new_chunk = mddev->new_chunk_sectors;
b3546035 7879
597a711b 7880 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
b3546035
N		 7881 return -EINVAL;
7882 if (new_chunk > 0) {
0ba459d2 7883 if (!is_power_of_2(new_chunk))
b3546035 7884 return -EINVAL;
597a711b 7885 if (new_chunk < (PAGE_SIZE>>9))
b3546035 7886 return -EINVAL;
597a711b 7887 if (mddev->array_sectors & (new_chunk-1))
b3546035
N		 7888 /* not factor of array size */
7889 return -EINVAL;
7890 }
7891
7892 /* They look valid */
7893
88ce4930 7894 if (mddev->raid_disks == 2) {
597a711b
N		 7895 /* can make the change immediately */
7896 if (mddev->new_layout >= 0) {
7897 conf->algorithm = mddev->new_layout;
7898 mddev->layout = mddev->new_layout;
88ce4930
N		 7899 }
7900 if (new_chunk > 0) {
597a711b
N		 7901 conf->chunk_sectors = new_chunk ;
7902 mddev->chunk_sectors = new_chunk;
88ce4930 7903 }
2953079c 7904 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
88ce4930 7905 md_wakeup_thread(mddev->thread);
b3546035 7906 }
50ac168a 7907 return check_reshape(mddev);
88ce4930
N		 7908}
7909
fd01b88c 7910static int raid6_check_reshape(struct mddev *mddev)
88ce4930 7911{
597a711b 7912 int new_chunk = mddev->new_chunk_sectors;
50ac168a 7913
597a711b 7914 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
88ce4930 7915 return -EINVAL;
b3546035 7916 if (new_chunk > 0) {
0ba459d2 7917 if (!is_power_of_2(new_chunk))
88ce4930 7918 return -EINVAL;
597a711b 7919 if (new_chunk < (PAGE_SIZE >> 9))
88ce4930 7920 return -EINVAL;
597a711b 7921 if (mddev->array_sectors & (new_chunk-1))
88ce4930
N		 7922 /* not factor of array size */
7923 return -EINVAL;
b3546035 7924 }
88ce4930
N		 7925
7926 /* They look valid */
50ac168a 7927 return check_reshape(mddev);
b3546035
N		 7928}
7929
fd01b88c 7930static void *raid5_takeover(struct mddev *mddev)
d562b0c4
N		 7931{
7932 /* raid5 can take over:
f1b29bca 7933 * raid0 - if there is only one strip zone - make it a raid4 layout
d562b0c4
N		 7934 * raid1 - if there are two drives. We need to know the chunk size
7935 * raid4 - trivial - just use a raid4 layout.
7936 * raid6 - Providing it is a *_6 layout
d562b0c4 7937 */
f1b29bca
DW		 7938 if (mddev->level == 0)
7939 return raid45_takeover_raid0(mddev, 5);
d562b0c4
N		 7940 if (mddev->level == 1)
7941 return raid5_takeover_raid1(mddev);
e9d4758f
N		 7942 if (mddev->level == 4) {
7943 mddev->new_layout = ALGORITHM_PARITY_N;
7944 mddev->new_level = 5;
7945 return setup_conf(mddev);
7946 }
fc9739c6
N		 7947 if (mddev->level == 6)
7948 return raid5_takeover_raid6(mddev);
d562b0c4
N		 7949
7950 return ERR_PTR(-EINVAL);
7951}
7952
fd01b88c 7953static void *raid4_takeover(struct mddev *mddev)
a78d38a1 7954{
f1b29bca
DW		 7955 /* raid4 can take over:
7956 * raid0 - if there is only one strip zone
7957 * raid5 - if layout is right
a78d38a1 7958 */
f1b29bca
DW		 7959 if (mddev->level == 0)
7960 return raid45_takeover_raid0(mddev, 4);
a78d38a1
N		 7961 if (mddev->level == 5 &&
7962 mddev->layout == ALGORITHM_PARITY_N) {
7963 mddev->new_layout = 0;
7964 mddev->new_level = 4;
7965 return setup_conf(mddev);
7966 }
7967 return ERR_PTR(-EINVAL);
7968}
d562b0c4 7969
84fc4b56 7970static struct md_personality raid5_personality;
245f46c2 7971
fd01b88c 7972static void *raid6_takeover(struct mddev *mddev)
245f46c2
N		 7973{
7974 /* Currently can only take over a raid5. We map the
7975 * personality to an equivalent raid6 personality
7976 * with the Q block at the end.
7977 */
7978 int new_layout;
7979
7980 if (mddev->pers != &raid5_personality)
7981 return ERR_PTR(-EINVAL);
7982 if (mddev->degraded > 1)
7983 return ERR_PTR(-EINVAL);
7984 if (mddev->raid_disks > 253)
7985 return ERR_PTR(-EINVAL);
7986 if (mddev->raid_disks < 3)
7987 return ERR_PTR(-EINVAL);
7988
7989 switch (mddev->layout) {
7990 case ALGORITHM_LEFT_ASYMMETRIC:
7991 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
7992 break;
7993 case ALGORITHM_RIGHT_ASYMMETRIC:
7994 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
7995 break;
7996 case ALGORITHM_LEFT_SYMMETRIC:
7997 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
7998 break;
7999 case ALGORITHM_RIGHT_SYMMETRIC:
8000 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
8001 break;
8002 case ALGORITHM_PARITY_0:
8003 new_layout = ALGORITHM_PARITY_0_6;
8004 break;
8005 case ALGORITHM_PARITY_N:
8006 new_layout = ALGORITHM_PARITY_N;
8007 break;
8008 default:
8009 return ERR_PTR(-EINVAL);
8010 }
8011 mddev->new_level = 6;
8012 mddev->new_layout = new_layout;
8013 mddev->delta_disks = 1;
8014 mddev->raid_disks += 1;
8015 return setup_conf(mddev);
8016}
8017
84fc4b56 8018static struct md_personality raid6_personality =
16a53ecc
N		 8019{
8020 .name = "raid6",
8021 .level = 6,
8022 .owner = THIS_MODULE,
849674e4
SL		 8023 .make_request = raid5_make_request,
8024 .run = raid5_run,
afa0f557 8025 .free = raid5_free,
849674e4
SL		 8026 .status = raid5_status,
8027 .error_handler = raid5_error,
16a53ecc
N		 8028 .hot_add_disk = raid5_add_disk,
8029 .hot_remove_disk= raid5_remove_disk,
8030 .spare_active = raid5_spare_active,
849674e4 8031 .sync_request = raid5_sync_request,
16a53ecc 8032 .resize = raid5_resize,
80c3a6ce 8033 .size = raid5_size,
50ac168a 8034 .check_reshape = raid6_check_reshape,
f416885e 8035 .start_reshape = raid5_start_reshape,
cea9c228 8036 .finish_reshape = raid5_finish_reshape,
16a53ecc 8037 .quiesce = raid5_quiesce,
245f46c2 8038 .takeover = raid6_takeover,
5c675f83 8039 .congested = raid5_congested,
16a53ecc 8040};
84fc4b56 8041static struct md_personality raid5_personality =
1da177e4
LT		 8042{
8043 .name = "raid5",
2604b703 8044 .level = 5,
1da177e4 8045 .owner = THIS_MODULE,
849674e4
SL		 8046 .make_request = raid5_make_request,
8047 .run = raid5_run,
afa0f557 8048 .free = raid5_free,
849674e4
SL		 8049 .status = raid5_status,
8050 .error_handler = raid5_error,
1da177e4
LT		 8051 .hot_add_disk = raid5_add_disk,
8052 .hot_remove_disk= raid5_remove_disk,
8053 .spare_active = raid5_spare_active,
849674e4 8054 .sync_request = raid5_sync_request,
1da177e4 8055 .resize = raid5_resize,
80c3a6ce 8056 .size = raid5_size,
63c70c4f
N		 8057 .check_reshape = raid5_check_reshape,
8058 .start_reshape = raid5_start_reshape,
cea9c228 8059 .finish_reshape = raid5_finish_reshape,
72626685 8060 .quiesce = raid5_quiesce,
d562b0c4 8061 .takeover = raid5_takeover,
5c675f83 8062 .congested = raid5_congested,
1da177e4
LT		 8063};
8064
84fc4b56 8065static struct md_personality raid4_personality =
1da177e4 8066{
2604b703
N		 8067 .name = "raid4",
8068 .level = 4,
8069 .owner = THIS_MODULE,
849674e4
SL		 8070 .make_request = raid5_make_request,
8071 .run = raid5_run,
afa0f557 8072 .free = raid5_free,
849674e4
SL		 8073 .status = raid5_status,
8074 .error_handler = raid5_error,
2604b703
N		 8075 .hot_add_disk = raid5_add_disk,
8076 .hot_remove_disk= raid5_remove_disk,
8077 .spare_active = raid5_spare_active,
849674e4 8078 .sync_request = raid5_sync_request,
2604b703 8079 .resize = raid5_resize,
80c3a6ce 8080 .size = raid5_size,
3d37890b
N		 8081 .check_reshape = raid5_check_reshape,
8082 .start_reshape = raid5_start_reshape,
cea9c228 8083 .finish_reshape = raid5_finish_reshape,
2604b703 8084 .quiesce = raid5_quiesce,
a78d38a1 8085 .takeover = raid4_takeover,
5c675f83 8086 .congested = raid5_congested,
2604b703
N		 8087};
8088
8089static int __init raid5_init(void)
8090{
29c6d1bb
SAS		 8091 int ret;
8092
851c30c9
SL		 8093 raid5_wq = alloc_workqueue("raid5wq",
8094 WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0);
8095 if (!raid5_wq)
8096 return -ENOMEM;
29c6d1bb
SAS		 8097
8098 ret = cpuhp_setup_state_multi(CPUHP_MD_RAID5_PREPARE,
8099 "md/raid5:prepare",
8100 raid456_cpu_up_prepare,
8101 raid456_cpu_dead);
8102 if (ret) {
8103 destroy_workqueue(raid5_wq);
8104 return ret;
8105 }
16a53ecc 8106 register_md_personality(&raid6_personality);
2604b703
N		 8107 register_md_personality(&raid5_personality);
8108 register_md_personality(&raid4_personality);
8109 return 0;
1da177e4
LT		 8110}
8111
2604b703 8112static void raid5_exit(void)
1da177e4 8113{
16a53ecc 8114 unregister_md_personality(&raid6_personality);
2604b703
N		 8115 unregister_md_personality(&raid5_personality);
8116 unregister_md_personality(&raid4_personality);
29c6d1bb 8117 cpuhp_remove_multi_state(CPUHP_MD_RAID5_PREPARE);
851c30c9 8118 destroy_workqueue(raid5_wq);
1da177e4
LT		 8119}
8120
8121module_init(raid5_init);
8122module_exit(raid5_exit);
8123MODULE_LICENSE("GPL");
0efb9e61 8124MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
1da177e4 8125MODULE_ALIAS("md-personality-4"); /* RAID5 */
d9d166c2
N		 8126MODULE_ALIAS("md-raid5");
8127MODULE_ALIAS("md-raid4");
2604b703
N		 8128MODULE_ALIAS("md-level-5");
8129MODULE_ALIAS("md-level-4");
16a53ecc
N		 8130MODULE_ALIAS("md-personality-8"); /* RAID6 */
8131MODULE_ALIAS("md-raid6");
8132MODULE_ALIAS("md-level-6");
8133
8134/* This used to be two separate modules, they were: */
8135MODULE_ALIAS("raid5");
8136MODULE_ALIAS("raid6");
Linux 6.x block layer and io_uring tree(s)