projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md/raid5-cache: do not need to set STRIPE_PREREAD_ACTIVE repeatedly
[linux-2.6-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))
796a5cf0 900 op_flags = WRITE_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 &&
964 conf->mddev->flags) {
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
2018 bio_init(&dev->req);
45c91d80
SL		 2019 dev->req.bi_io_vec = &dev->vec;
2020 dev->req.bi_max_vecs = 1;
2021
5f9d1fde 2022 bio_init(&dev->rreq);
45c91d80
SL		 2023 dev->rreq.bi_io_vec = &dev->rvec;
2024 dev->rreq.bi_max_vecs = 1;
5f9d1fde 2025 }
f18c1a35
N		 2026 }
2027 return sh;
2028}
486f0644 2029static int grow_one_stripe(struct r5conf *conf, gfp_t gfp)
1da177e4
LT		 2030{
2031 struct stripe_head *sh;
f18c1a35 2032
5f9d1fde 2033 sh = alloc_stripe(conf->slab_cache, gfp, conf->pool_size);
3f294f4f
N		 2034 if (!sh)
2035 return 0;
6ce32846 2036
3f294f4f 2037 sh->raid_conf = conf;
3f294f4f 2038
a9683a79 2039 if (grow_buffers(sh, gfp)) {
e4e11e38 2040 shrink_buffers(sh);
3f294f4f
N		 2041 kmem_cache_free(conf->slab_cache, sh);
2042 return 0;
2043 }
486f0644
N		 2044 sh->hash_lock_index =
2045 conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS;
3f294f4f 2046 /* we just created an active stripe so... */
3f294f4f 2047 atomic_inc(&conf->active_stripes);
59fc630b 2048
6d036f7d 2049 raid5_release_stripe(sh);
486f0644 2050 conf->max_nr_stripes++;
3f294f4f
N		 2051 return 1;
2052}
2053
d1688a6d 2054static int grow_stripes(struct r5conf *conf, int num)
3f294f4f 2055{
e18b890b 2056 struct kmem_cache *sc;
5e5e3e78 2057 int devs = max(conf->raid_disks, conf->previous_raid_disks);
1da177e4 2058
f4be6b43
N		 2059 if (conf->mddev->gendisk)
2060 sprintf(conf->cache_name[0],
2061 "raid%d-%s", conf->level, mdname(conf->mddev));
2062 else
2063 sprintf(conf->cache_name[0],
2064 "raid%d-%p", conf->level, conf->mddev);
2065 sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]);
2066
ad01c9e3
N		 2067 conf->active_name = 0;
2068 sc = kmem_cache_create(conf->cache_name[conf->active_name],
1da177e4 2069 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
20c2df83 2070 0, 0, NULL);
1da177e4
LT		 2071 if (!sc)
2072 return 1;
2073 conf->slab_cache = sc;
ad01c9e3 2074 conf->pool_size = devs;
486f0644
N		 2075 while (num--)
2076 if (!grow_one_stripe(conf, GFP_KERNEL))
1da177e4 2077 return 1;
486f0644 2078
1da177e4
LT		 2079 return 0;
2080}
29269553 2081
d6f38f31
DW		 2082/**
2083 * scribble_len - return the required size of the scribble region
2084 * @num - total number of disks in the array
2085 *
2086 * The size must be enough to contain:
2087 * 1/ a struct page pointer for each device in the array +2
2088 * 2/ room to convert each entry in (1) to its corresponding dma
2089 * (dma_map_page()) or page (page_address()) address.
2090 *
2091 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
2092 * calculate over all devices (not just the data blocks), using zeros in place
2093 * of the P and Q blocks.
2094 */
46d5b785 2095static struct flex_array *scribble_alloc(int num, int cnt, gfp_t flags)
d6f38f31 2096{
46d5b785 2097 struct flex_array *ret;
d6f38f31
DW		 2098 size_t len;
2099
2100 len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2);
46d5b785 2101 ret = flex_array_alloc(len, cnt, flags);
2102 if (!ret)
2103 return NULL;
2104 /* always prealloc all elements, so no locking is required */
2105 if (flex_array_prealloc(ret, 0, cnt, flags)) {
2106 flex_array_free(ret);
2107 return NULL;
2108 }
2109 return ret;
d6f38f31
DW		 2110}
2111
738a2738
N		 2112static int resize_chunks(struct r5conf *conf, int new_disks, int new_sectors)
2113{
2114 unsigned long cpu;
2115 int err = 0;
2116
27a353c0
SL		 2117 /*
2118 * Never shrink. And mddev_suspend() could deadlock if this is called
2119 * from raid5d. In that case, scribble_disks and scribble_sectors
2120 * should equal to new_disks and new_sectors
2121 */
2122 if (conf->scribble_disks >= new_disks &&
2123 conf->scribble_sectors >= new_sectors)
2124 return 0;
738a2738
N		 2125 mddev_suspend(conf->mddev);
2126 get_online_cpus();
2127 for_each_present_cpu(cpu) {
2128 struct raid5_percpu *percpu;
2129 struct flex_array *scribble;
2130
2131 percpu = per_cpu_ptr(conf->percpu, cpu);
2132 scribble = scribble_alloc(new_disks,
2133 new_sectors / STRIPE_SECTORS,
2134 GFP_NOIO);
2135
2136 if (scribble) {
2137 flex_array_free(percpu->scribble);
2138 percpu->scribble = scribble;
2139 } else {
2140 err = -ENOMEM;
2141 break;
2142 }
2143 }
2144 put_online_cpus();
2145 mddev_resume(conf->mddev);
27a353c0
SL		 2146 if (!err) {
2147 conf->scribble_disks = new_disks;
2148 conf->scribble_sectors = new_sectors;
2149 }
738a2738
N		 2150 return err;
2151}
2152
d1688a6d 2153static int resize_stripes(struct r5conf *conf, int newsize)
ad01c9e3
N		 2154{
2155 /* Make all the stripes able to hold 'newsize' devices.
2156 * New slots in each stripe get 'page' set to a new page.
2157 *
2158 * This happens in stages:
2159 * 1/ create a new kmem_cache and allocate the required number of
2160 * stripe_heads.
83f0d77a 2161 * 2/ gather all the old stripe_heads and transfer the pages across
ad01c9e3
N		 2162 * to the new stripe_heads. This will have the side effect of
2163 * freezing the array as once all stripe_heads have been collected,
2164 * no IO will be possible. Old stripe heads are freed once their
2165 * pages have been transferred over, and the old kmem_cache is
2166 * freed when all stripes are done.
2167 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
2168 * we simple return a failre status - no need to clean anything up.
2169 * 4/ allocate new pages for the new slots in the new stripe_heads.
2170 * If this fails, we don't bother trying the shrink the
2171 * stripe_heads down again, we just leave them as they are.
2172 * As each stripe_head is processed the new one is released into
2173 * active service.
2174 *
2175 * Once step2 is started, we cannot afford to wait for a write,
2176 * so we use GFP_NOIO allocations.
2177 */
2178 struct stripe_head *osh, *nsh;
2179 LIST_HEAD(newstripes);
2180 struct disk_info *ndisks;
b5470dc5 2181 int err;
e18b890b 2182 struct kmem_cache *sc;
ad01c9e3 2183 int i;
566c09c5 2184 int hash, cnt;
ad01c9e3
N		 2185
2186 if (newsize <= conf->pool_size)
2187 return 0; /* never bother to shrink */
2188
b5470dc5
DW		 2189 err = md_allow_write(conf->mddev);
2190 if (err)
2191 return err;
2a2275d6 2192
ad01c9e3
N		 2193 /* Step 1 */
2194 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
2195 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
20c2df83 2196 0, 0, NULL);
ad01c9e3
N		 2197 if (!sc)
2198 return -ENOMEM;
2199
2d5b569b
N		 2200 /* Need to ensure auto-resizing doesn't interfere */
2201 mutex_lock(&conf->cache_size_mutex);
2202
ad01c9e3 2203 for (i = conf->max_nr_stripes; i; i--) {
5f9d1fde 2204 nsh = alloc_stripe(sc, GFP_KERNEL, newsize);
ad01c9e3
N		 2205 if (!nsh)
2206 break;
2207
ad01c9e3 2208 nsh->raid_conf = conf;
ad01c9e3
N		 2209 list_add(&nsh->lru, &newstripes);
2210 }
2211 if (i) {
2212 /* didn't get enough, give up */
2213 while (!list_empty(&newstripes)) {
2214 nsh = list_entry(newstripes.next, struct stripe_head, lru);
2215 list_del(&nsh->lru);
2216 kmem_cache_free(sc, nsh);
2217 }
2218 kmem_cache_destroy(sc);
2d5b569b 2219 mutex_unlock(&conf->cache_size_mutex);
ad01c9e3
N		 2220 return -ENOMEM;
2221 }
2222 /* Step 2 - Must use GFP_NOIO now.
2223 * OK, we have enough stripes, start collecting inactive
2224 * stripes and copying them over
2225 */
566c09c5
SL		 2226 hash = 0;
2227 cnt = 0;
ad01c9e3 2228 list_for_each_entry(nsh, &newstripes, lru) {
566c09c5 2229 lock_device_hash_lock(conf, hash);
6ab2a4b8 2230 wait_event_cmd(conf->wait_for_stripe,
566c09c5
SL		 2231 !list_empty(conf->inactive_list + hash),
2232 unlock_device_hash_lock(conf, hash),
2233 lock_device_hash_lock(conf, hash));
2234 osh = get_free_stripe(conf, hash);
2235 unlock_device_hash_lock(conf, hash);
f18c1a35 2236
d592a996 2237 for(i=0; i<conf->pool_size; i++) {
ad01c9e3 2238 nsh->dev[i].page = osh->dev[i].page;
d592a996
SL		 2239 nsh->dev[i].orig_page = osh->dev[i].page;
2240 }
566c09c5 2241 nsh->hash_lock_index = hash;
ad01c9e3 2242 kmem_cache_free(conf->slab_cache, osh);
566c09c5
SL		 2243 cnt++;
2244 if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS +
2245 !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) {
2246 hash++;
2247 cnt = 0;
2248 }
ad01c9e3
N		 2249 }
2250 kmem_cache_destroy(conf->slab_cache);
2251
2252 /* Step 3.
2253 * At this point, we are holding all the stripes so the array
2254 * is completely stalled, so now is a good time to resize
d6f38f31 2255 * conf->disks and the scribble region
ad01c9e3
N		 2256 */
2257 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
2258 if (ndisks) {
d7bd398e 2259 for (i = 0; i < conf->pool_size; i++)
ad01c9e3 2260 ndisks[i] = conf->disks[i];
d7bd398e
SL		 2261
2262 for (i = conf->pool_size; i < newsize; i++) {
2263 ndisks[i].extra_page = alloc_page(GFP_NOIO);
2264 if (!ndisks[i].extra_page)
2265 err = -ENOMEM;
2266 }
2267
2268 if (err) {
2269 for (i = conf->pool_size; i < newsize; i++)
2270 if (ndisks[i].extra_page)
2271 put_page(ndisks[i].extra_page);
2272 kfree(ndisks);
2273 } else {
2274 kfree(conf->disks);
2275 conf->disks = ndisks;
2276 }
ad01c9e3
N		 2277 } else
2278 err = -ENOMEM;
2279
2d5b569b 2280 mutex_unlock(&conf->cache_size_mutex);
ad01c9e3
N		 2281 /* Step 4, return new stripes to service */
2282 while(!list_empty(&newstripes)) {
2283 nsh = list_entry(newstripes.next, struct stripe_head, lru);
2284 list_del_init(&nsh->lru);
d6f38f31 2285
ad01c9e3
N		 2286 for (i=conf->raid_disks; i < newsize; i++)
2287 if (nsh->dev[i].page == NULL) {
2288 struct page *p = alloc_page(GFP_NOIO);
2289 nsh->dev[i].page = p;
d592a996 2290 nsh->dev[i].orig_page = p;
ad01c9e3
N		 2291 if (!p)
2292 err = -ENOMEM;
2293 }
6d036f7d 2294 raid5_release_stripe(nsh);
ad01c9e3
N		 2295 }
2296 /* critical section pass, GFP_NOIO no longer needed */
2297
2298 conf->slab_cache = sc;
2299 conf->active_name = 1-conf->active_name;
6e9eac2d
N		 2300 if (!err)
2301 conf->pool_size = newsize;
ad01c9e3
N		 2302 return err;
2303}
1da177e4 2304
486f0644 2305static int drop_one_stripe(struct r5conf *conf)
1da177e4
LT		 2306{
2307 struct stripe_head *sh;
49895bcc 2308 int hash = (conf->max_nr_stripes - 1) & STRIPE_HASH_LOCKS_MASK;
1da177e4 2309
566c09c5
SL		 2310 spin_lock_irq(conf->hash_locks + hash);
2311 sh = get_free_stripe(conf, hash);
2312 spin_unlock_irq(conf->hash_locks + hash);
3f294f4f
N		 2313 if (!sh)
2314 return 0;
78bafebd 2315 BUG_ON(atomic_read(&sh->count));
e4e11e38 2316 shrink_buffers(sh);
3f294f4f
N		 2317 kmem_cache_free(conf->slab_cache, sh);
2318 atomic_dec(&conf->active_stripes);
486f0644 2319 conf->max_nr_stripes--;
3f294f4f
N		 2320 return 1;
2321}
2322
d1688a6d 2323static void shrink_stripes(struct r5conf *conf)
3f294f4f 2324{
486f0644
N		 2325 while (conf->max_nr_stripes &&
2326 drop_one_stripe(conf))
2327 ;
3f294f4f 2328
644df1a8 2329 kmem_cache_destroy(conf->slab_cache);
1da177e4
LT		 2330 conf->slab_cache = NULL;
2331}
2332
4246a0b6 2333static void raid5_end_read_request(struct bio * bi)
1da177e4 2334{
99c0fb5f 2335 struct stripe_head *sh = bi->bi_private;
d1688a6d 2336 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 2337 int disks = sh->disks, i;
d6950432 2338 char b[BDEVNAME_SIZE];
dd054fce 2339 struct md_rdev *rdev = NULL;
05616be5 2340 sector_t s;
1da177e4
LT		 2341
2342 for (i=0 ; i<disks; i++)
2343 if (bi == &sh->dev[i].req)
2344 break;
2345
4246a0b6 2346 pr_debug("end_read_request %llu/%d, count: %d, error %d.\n",
45b4233c 2347 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
4246a0b6 2348 bi->bi_error);
1da177e4 2349 if (i == disks) {
5f9d1fde 2350 bio_reset(bi);
1da177e4 2351 BUG();
6712ecf8 2352 return;
1da177e4 2353 }
14a75d3e 2354 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
dd054fce
N		 2355 /* If replacement finished while this request was outstanding,
2356 * 'replacement' might be NULL already.
2357 * In that case it moved down to 'rdev'.
2358 * rdev is not removed until all requests are finished.
2359 */
14a75d3e 2360 rdev = conf->disks[i].replacement;
dd054fce 2361 if (!rdev)
14a75d3e 2362 rdev = conf->disks[i].rdev;
1da177e4 2363
05616be5
N		 2364 if (use_new_offset(conf, sh))
2365 s = sh->sector + rdev->new_data_offset;
2366 else
2367 s = sh->sector + rdev->data_offset;
4246a0b6 2368 if (!bi->bi_error) {
1da177e4 2369 set_bit(R5_UPTODATE, &sh->dev[i].flags);
4e5314b5 2370 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
14a75d3e
N		 2371 /* Note that this cannot happen on a
2372 * replacement device. We just fail those on
2373 * any error
2374 */
cc6167b4
N		 2375 pr_info_ratelimited(
2376 "md/raid:%s: read error corrected (%lu sectors at %llu on %s)\n",
8bda470e 2377 mdname(conf->mddev), STRIPE_SECTORS,
05616be5 2378 (unsigned long long)s,
8bda470e 2379 bdevname(rdev->bdev, b));
ddd5115f 2380 atomic_add(STRIPE_SECTORS, &rdev->corrected_errors);
4e5314b5
N		 2381 clear_bit(R5_ReadError, &sh->dev[i].flags);
2382 clear_bit(R5_ReWrite, &sh->dev[i].flags);
3f9e7c14 2383 } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2384 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2385
14a75d3e
N		 2386 if (atomic_read(&rdev->read_errors))
2387 atomic_set(&rdev->read_errors, 0);
1da177e4 2388 } else {
14a75d3e 2389 const char *bdn = bdevname(rdev->bdev, b);
ba22dcbf 2390 int retry = 0;
2e8ac303 2391 int set_bad = 0;
d6950432 2392
1da177e4 2393 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
d6950432 2394 atomic_inc(&rdev->read_errors);
14a75d3e 2395 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
cc6167b4
N		 2396 pr_warn_ratelimited(
2397 "md/raid:%s: read error on replacement device (sector %llu on %s).\n",
14a75d3e 2398 mdname(conf->mddev),
05616be5 2399 (unsigned long long)s,
14a75d3e 2400 bdn);
2e8ac303 2401 else if (conf->mddev->degraded >= conf->max_degraded) {
2402 set_bad = 1;
cc6167b4
N		 2403 pr_warn_ratelimited(
2404 "md/raid:%s: read error not correctable (sector %llu on %s).\n",
8bda470e 2405 mdname(conf->mddev),
05616be5 2406 (unsigned long long)s,
8bda470e 2407 bdn);
2e8ac303 2408 } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) {
4e5314b5 2409 /* Oh, no!!! */
2e8ac303 2410 set_bad = 1;
cc6167b4
N		 2411 pr_warn_ratelimited(
2412 "md/raid:%s: read error NOT corrected!! (sector %llu on %s).\n",
8bda470e 2413 mdname(conf->mddev),
05616be5 2414 (unsigned long long)s,
8bda470e 2415 bdn);
2e8ac303 2416 } else if (atomic_read(&rdev->read_errors)
ba22dcbf 2417 > conf->max_nr_stripes)
cc6167b4 2418 pr_warn("md/raid:%s: Too many read errors, failing device %s.\n",
d6950432 2419 mdname(conf->mddev), bdn);
ba22dcbf
N		 2420 else
2421 retry = 1;
edfa1f65
BY		 2422 if (set_bad && test_bit(In_sync, &rdev->flags)
2423 && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2424 retry = 1;
ba22dcbf 2425 if (retry)
3f9e7c14 2426 if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) {
2427 set_bit(R5_ReadError, &sh->dev[i].flags);
2428 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2429 } else
2430 set_bit(R5_ReadNoMerge, &sh->dev[i].flags);
ba22dcbf 2431 else {
4e5314b5
N		 2432 clear_bit(R5_ReadError, &sh->dev[i].flags);
2433 clear_bit(R5_ReWrite, &sh->dev[i].flags);
2e8ac303 2434 if (!(set_bad
2435 && test_bit(In_sync, &rdev->flags)
2436 && rdev_set_badblocks(
2437 rdev, sh->sector, STRIPE_SECTORS, 0)))
2438 md_error(conf->mddev, rdev);
ba22dcbf 2439 }
1da177e4 2440 }
14a75d3e 2441 rdev_dec_pending(rdev, conf->mddev);
c9445555 2442 bio_reset(bi);
1da177e4
LT		 2443 clear_bit(R5_LOCKED, &sh->dev[i].flags);
2444 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 2445 raid5_release_stripe(sh);
1da177e4
LT		 2446}
2447
4246a0b6 2448static void raid5_end_write_request(struct bio *bi)
1da177e4 2449{
99c0fb5f 2450 struct stripe_head *sh = bi->bi_private;
d1688a6d 2451 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 2452 int disks = sh->disks, i;
977df362 2453 struct md_rdev *uninitialized_var(rdev);
b84db560
N		 2454 sector_t first_bad;
2455 int bad_sectors;
977df362 2456 int replacement = 0;
1da177e4 2457
977df362
N		 2458 for (i = 0 ; i < disks; i++) {
2459 if (bi == &sh->dev[i].req) {
2460 rdev = conf->disks[i].rdev;
1da177e4 2461 break;
977df362
N		 2462 }
2463 if (bi == &sh->dev[i].rreq) {
2464 rdev = conf->disks[i].replacement;
dd054fce
N		 2465 if (rdev)
2466 replacement = 1;
2467 else
2468 /* rdev was removed and 'replacement'
2469 * replaced it. rdev is not removed
2470 * until all requests are finished.
2471 */
2472 rdev = conf->disks[i].rdev;
977df362
N		 2473 break;
2474 }
2475 }
4246a0b6 2476 pr_debug("end_write_request %llu/%d, count %d, error: %d.\n",
1da177e4 2477 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
4246a0b6 2478 bi->bi_error);
1da177e4 2479 if (i == disks) {
5f9d1fde 2480 bio_reset(bi);
1da177e4 2481 BUG();
6712ecf8 2482 return;
1da177e4
LT		 2483 }
2484
977df362 2485 if (replacement) {
4246a0b6 2486 if (bi->bi_error)
977df362
N		 2487 md_error(conf->mddev, rdev);
2488 else if (is_badblock(rdev, sh->sector,
2489 STRIPE_SECTORS,
2490 &first_bad, &bad_sectors))
2491 set_bit(R5_MadeGoodRepl, &sh->dev[i].flags);
2492 } else {
4246a0b6 2493 if (bi->bi_error) {
9f97e4b1 2494 set_bit(STRIPE_DEGRADED, &sh->state);
977df362
N		 2495 set_bit(WriteErrorSeen, &rdev->flags);
2496 set_bit(R5_WriteError, &sh->dev[i].flags);
3a6de292
N		 2497 if (!test_and_set_bit(WantReplacement, &rdev->flags))
2498 set_bit(MD_RECOVERY_NEEDED,
2499 &rdev->mddev->recovery);
977df362
N		 2500 } else if (is_badblock(rdev, sh->sector,
2501 STRIPE_SECTORS,
c0b32972 2502 &first_bad, &bad_sectors)) {
977df362 2503 set_bit(R5_MadeGood, &sh->dev[i].flags);
c0b32972
N		 2504 if (test_bit(R5_ReadError, &sh->dev[i].flags))
2505 /* That was a successful write so make
2506 * sure it looks like we already did
2507 * a re-write.
2508 */
2509 set_bit(R5_ReWrite, &sh->dev[i].flags);
2510 }
977df362
N		 2511 }
2512 rdev_dec_pending(rdev, conf->mddev);
1da177e4 2513
4246a0b6 2514 if (sh->batch_head && bi->bi_error && !replacement)
72ac7330 2515 set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state);
2516
c9445555 2517 bio_reset(bi);
977df362
N		 2518 if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags))
2519 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1da177e4 2520 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 2521 raid5_release_stripe(sh);
59fc630b 2522
2523 if (sh->batch_head && sh != sh->batch_head)
6d036f7d 2524 raid5_release_stripe(sh->batch_head);
1da177e4
LT		 2525}
2526
784052ec 2527static void raid5_build_block(struct stripe_head *sh, int i, int previous)
1da177e4
LT		 2528{
2529 struct r5dev *dev = &sh->dev[i];
2530
1da177e4 2531 dev->flags = 0;
6d036f7d 2532 dev->sector = raid5_compute_blocknr(sh, i, previous);
1da177e4
LT		 2533}
2534
849674e4 2535static void raid5_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 2536{
2537 char b[BDEVNAME_SIZE];
d1688a6d 2538 struct r5conf *conf = mddev->private;
908f4fbd 2539 unsigned long flags;
0c55e022 2540 pr_debug("raid456: error called\n");
1da177e4 2541
908f4fbd
N		 2542 spin_lock_irqsave(&conf->device_lock, flags);
2543 clear_bit(In_sync, &rdev->flags);
2544 mddev->degraded = calc_degraded(conf);
2545 spin_unlock_irqrestore(&conf->device_lock, flags);
2546 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2547
de393cde 2548 set_bit(Blocked, &rdev->flags);
6f8d0c77 2549 set_bit(Faulty, &rdev->flags);
85ad1d13
GJ		 2550 set_mask_bits(&mddev->flags, 0,
2551 BIT(MD_CHANGE_DEVS) | BIT(MD_CHANGE_PENDING));
cc6167b4
N		 2552 pr_crit("md/raid:%s: Disk failure on %s, disabling device.\n"
2553 "md/raid:%s: Operation continuing on %d devices.\n",
2554 mdname(mddev),
2555 bdevname(rdev->bdev, b),
2556 mdname(mddev),
2557 conf->raid_disks - mddev->degraded);
16a53ecc 2558}
1da177e4
LT		 2559
2560/*
2561 * Input: a 'big' sector number,
2562 * Output: index of the data and parity disk, and the sector # in them.
2563 */
6d036f7d
SL		 2564sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector,
2565 int previous, int *dd_idx,
2566 struct stripe_head *sh)
1da177e4 2567{
6e3b96ed 2568 sector_t stripe, stripe2;
35f2a591 2569 sector_t chunk_number;
1da177e4 2570 unsigned int chunk_offset;
911d4ee8 2571 int pd_idx, qd_idx;
67cc2b81 2572 int ddf_layout = 0;
1da177e4 2573 sector_t new_sector;
e183eaed
N		 2574 int algorithm = previous ? conf->prev_algo
2575 : conf->algorithm;
09c9e5fa
AN		 2576 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2577 : conf->chunk_sectors;
112bf897
N		 2578 int raid_disks = previous ? conf->previous_raid_disks
2579 : conf->raid_disks;
2580 int data_disks = raid_disks - conf->max_degraded;
1da177e4
LT		 2581
2582 /* First compute the information on this sector */
2583
2584 /*
2585 * Compute the chunk number and the sector offset inside the chunk
2586 */
2587 chunk_offset = sector_div(r_sector, sectors_per_chunk);
2588 chunk_number = r_sector;
1da177e4
LT		 2589
2590 /*
2591 * Compute the stripe number
2592 */
35f2a591
N		 2593 stripe = chunk_number;
2594 *dd_idx = sector_div(stripe, data_disks);
6e3b96ed 2595 stripe2 = stripe;
1da177e4
LT		 2596 /*
2597 * Select the parity disk based on the user selected algorithm.
2598 */
84789554 2599 pd_idx = qd_idx = -1;
16a53ecc
N		 2600 switch(conf->level) {
2601 case 4:
911d4ee8 2602 pd_idx = data_disks;
16a53ecc
N		 2603 break;
2604 case 5:
e183eaed 2605 switch (algorithm) {
1da177e4 2606 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2607 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2608 if (*dd_idx >= pd_idx)
1da177e4
LT		 2609 (*dd_idx)++;
2610 break;
2611 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2612 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2613 if (*dd_idx >= pd_idx)
1da177e4
LT		 2614 (*dd_idx)++;
2615 break;
2616 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2617 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2618 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4
LT		 2619 break;
2620 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2621 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2622 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4 2623 break;
99c0fb5f
N		 2624 case ALGORITHM_PARITY_0:
2625 pd_idx = 0;
2626 (*dd_idx)++;
2627 break;
2628 case ALGORITHM_PARITY_N:
2629 pd_idx = data_disks;
2630 break;
1da177e4 2631 default:
99c0fb5f 2632 BUG();
16a53ecc
N		 2633 }
2634 break;
2635 case 6:
2636
e183eaed 2637 switch (algorithm) {
16a53ecc 2638 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2639 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2640 qd_idx = pd_idx + 1;
2641 if (pd_idx == raid_disks-1) {
99c0fb5f 2642 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2643 qd_idx = 0;
2644 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2645 (*dd_idx) += 2; /* D D P Q D */
2646 break;
2647 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2648 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2649 qd_idx = pd_idx + 1;
2650 if (pd_idx == raid_disks-1) {
99c0fb5f 2651 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2652 qd_idx = 0;
2653 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2654 (*dd_idx) += 2; /* D D P Q D */
2655 break;
2656 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2657 pd_idx = raid_disks - 1 - 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
N		 2660 break;
2661 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2662 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2663 qd_idx = (pd_idx + 1) % raid_disks;
2664 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc 2665 break;
99c0fb5f
N		 2666
2667 case ALGORITHM_PARITY_0:
2668 pd_idx = 0;
2669 qd_idx = 1;
2670 (*dd_idx) += 2;
2671 break;
2672 case ALGORITHM_PARITY_N:
2673 pd_idx = data_disks;
2674 qd_idx = data_disks + 1;
2675 break;
2676
2677 case ALGORITHM_ROTATING_ZERO_RESTART:
2678 /* Exactly the same as RIGHT_ASYMMETRIC, but or
2679 * of blocks for computing Q is different.
2680 */
6e3b96ed 2681 pd_idx = sector_div(stripe2, raid_disks);
99c0fb5f
N		 2682 qd_idx = pd_idx + 1;
2683 if (pd_idx == raid_disks-1) {
2684 (*dd_idx)++; /* Q D D D P */
2685 qd_idx = 0;
2686 } else if (*dd_idx >= pd_idx)
2687 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2688 ddf_layout = 1;
99c0fb5f
N		 2689 break;
2690
2691 case ALGORITHM_ROTATING_N_RESTART:
2692 /* Same a left_asymmetric, by first stripe is
2693 * D D D P Q rather than
2694 * Q D D D P
2695 */
6e3b96ed
N		 2696 stripe2 += 1;
2697 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2698 qd_idx = pd_idx + 1;
2699 if (pd_idx == raid_disks-1) {
2700 (*dd_idx)++; /* Q D D D P */
2701 qd_idx = 0;
2702 } else if (*dd_idx >= pd_idx)
2703 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2704 ddf_layout = 1;
99c0fb5f
N		 2705 break;
2706
2707 case ALGORITHM_ROTATING_N_CONTINUE:
2708 /* Same as left_symmetric but Q is before P */
6e3b96ed 2709 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2710 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
2711 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
67cc2b81 2712 ddf_layout = 1;
99c0fb5f
N		 2713 break;
2714
2715 case ALGORITHM_LEFT_ASYMMETRIC_6:
2716 /* RAID5 left_asymmetric, with Q on last device */
6e3b96ed 2717 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2718 if (*dd_idx >= pd_idx)
2719 (*dd_idx)++;
2720 qd_idx = raid_disks - 1;
2721 break;
2722
2723 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6e3b96ed 2724 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2725 if (*dd_idx >= pd_idx)
2726 (*dd_idx)++;
2727 qd_idx = raid_disks - 1;
2728 break;
2729
2730 case ALGORITHM_LEFT_SYMMETRIC_6:
6e3b96ed 2731 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2732 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2733 qd_idx = raid_disks - 1;
2734 break;
2735
2736 case ALGORITHM_RIGHT_SYMMETRIC_6:
6e3b96ed 2737 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2738 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2739 qd_idx = raid_disks - 1;
2740 break;
2741
2742 case ALGORITHM_PARITY_0_6:
2743 pd_idx = 0;
2744 (*dd_idx)++;
2745 qd_idx = raid_disks - 1;
2746 break;
2747
16a53ecc 2748 default:
99c0fb5f 2749 BUG();
16a53ecc
N		 2750 }
2751 break;
1da177e4
LT		 2752 }
2753
911d4ee8
N		 2754 if (sh) {
2755 sh->pd_idx = pd_idx;
2756 sh->qd_idx = qd_idx;
67cc2b81 2757 sh->ddf_layout = ddf_layout;
911d4ee8 2758 }
1da177e4
LT		 2759 /*
2760 * Finally, compute the new sector number
2761 */
2762 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
2763 return new_sector;
2764}
2765
6d036f7d 2766sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous)
1da177e4 2767{
d1688a6d 2768 struct r5conf *conf = sh->raid_conf;
b875e531
N		 2769 int raid_disks = sh->disks;
2770 int data_disks = raid_disks - conf->max_degraded;
1da177e4 2771 sector_t new_sector = sh->sector, check;
09c9e5fa
AN		 2772 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2773 : conf->chunk_sectors;
e183eaed
N		 2774 int algorithm = previous ? conf->prev_algo
2775 : conf->algorithm;
1da177e4
LT		 2776 sector_t stripe;
2777 int chunk_offset;
35f2a591
N		 2778 sector_t chunk_number;
2779 int dummy1, dd_idx = i;
1da177e4 2780 sector_t r_sector;
911d4ee8 2781 struct stripe_head sh2;
1da177e4
LT		 2782
2783 chunk_offset = sector_div(new_sector, sectors_per_chunk);
2784 stripe = new_sector;
1da177e4 2785
16a53ecc
N		 2786 if (i == sh->pd_idx)
2787 return 0;
2788 switch(conf->level) {
2789 case 4: break;
2790 case 5:
e183eaed 2791 switch (algorithm) {
1da177e4
LT		 2792 case ALGORITHM_LEFT_ASYMMETRIC:
2793 case ALGORITHM_RIGHT_ASYMMETRIC:
2794 if (i > sh->pd_idx)
2795 i--;
2796 break;
2797 case ALGORITHM_LEFT_SYMMETRIC:
2798 case ALGORITHM_RIGHT_SYMMETRIC:
2799 if (i < sh->pd_idx)
2800 i += raid_disks;
2801 i -= (sh->pd_idx + 1);
2802 break;
99c0fb5f
N		 2803 case ALGORITHM_PARITY_0:
2804 i -= 1;
2805 break;
2806 case ALGORITHM_PARITY_N:
2807 break;
1da177e4 2808 default:
99c0fb5f 2809 BUG();
16a53ecc
N		 2810 }
2811 break;
2812 case 6:
d0dabf7e 2813 if (i == sh->qd_idx)
16a53ecc 2814 return 0; /* It is the Q disk */
e183eaed 2815 switch (algorithm) {
16a53ecc
N		 2816 case ALGORITHM_LEFT_ASYMMETRIC:
2817 case ALGORITHM_RIGHT_ASYMMETRIC:
99c0fb5f
N		 2818 case ALGORITHM_ROTATING_ZERO_RESTART:
2819 case ALGORITHM_ROTATING_N_RESTART:
2820 if (sh->pd_idx == raid_disks-1)
2821 i--; /* Q D D D P */
16a53ecc
N		 2822 else if (i > sh->pd_idx)
2823 i -= 2; /* D D P Q D */
2824 break;
2825 case ALGORITHM_LEFT_SYMMETRIC:
2826 case ALGORITHM_RIGHT_SYMMETRIC:
2827 if (sh->pd_idx == raid_disks-1)
2828 i--; /* Q D D D P */
2829 else {
2830 /* D D P Q D */
2831 if (i < sh->pd_idx)
2832 i += raid_disks;
2833 i -= (sh->pd_idx + 2);
2834 }
2835 break;
99c0fb5f
N		 2836 case ALGORITHM_PARITY_0:
2837 i -= 2;
2838 break;
2839 case ALGORITHM_PARITY_N:
2840 break;
2841 case ALGORITHM_ROTATING_N_CONTINUE:
e4424fee 2842 /* Like left_symmetric, but P is before Q */
99c0fb5f
N		 2843 if (sh->pd_idx == 0)
2844 i--; /* P D D D Q */
e4424fee
N		 2845 else {
2846 /* D D Q P D */
2847 if (i < sh->pd_idx)
2848 i += raid_disks;
2849 i -= (sh->pd_idx + 1);
2850 }
99c0fb5f
N		 2851 break;
2852 case ALGORITHM_LEFT_ASYMMETRIC_6:
2853 case ALGORITHM_RIGHT_ASYMMETRIC_6:
2854 if (i > sh->pd_idx)
2855 i--;
2856 break;
2857 case ALGORITHM_LEFT_SYMMETRIC_6:
2858 case ALGORITHM_RIGHT_SYMMETRIC_6:
2859 if (i < sh->pd_idx)
2860 i += data_disks + 1;
2861 i -= (sh->pd_idx + 1);
2862 break;
2863 case ALGORITHM_PARITY_0_6:
2864 i -= 1;
2865 break;
16a53ecc 2866 default:
99c0fb5f 2867 BUG();
16a53ecc
N		 2868 }
2869 break;
1da177e4
LT		 2870 }
2871
2872 chunk_number = stripe * data_disks + i;
35f2a591 2873 r_sector = chunk_number * sectors_per_chunk + chunk_offset;
1da177e4 2874
112bf897 2875 check = raid5_compute_sector(conf, r_sector,
784052ec 2876 previous, &dummy1, &sh2);
911d4ee8
N		 2877 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
2878 || sh2.qd_idx != sh->qd_idx) {
cc6167b4
N		 2879 pr_warn("md/raid:%s: compute_blocknr: map not correct\n",
2880 mdname(conf->mddev));
1da177e4
LT		 2881 return 0;
2882 }
2883 return r_sector;
2884}
2885
600aa109 2886static void
c0f7bddb 2887schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
600aa109 2888 int rcw, int expand)
e33129d8 2889{
584acdd4 2890 int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx, disks = sh->disks;
d1688a6d 2891 struct r5conf *conf = sh->raid_conf;
c0f7bddb 2892 int level = conf->level;
e33129d8
DW		 2893
2894 if (rcw) {
1e6d690b
SL		 2895 /*
2896 * In some cases, handle_stripe_dirtying initially decided to
2897 * run rmw and allocates extra page for prexor. However, rcw is
2898 * cheaper later on. We need to free the extra page now,
2899 * because we won't be able to do that in ops_complete_prexor().
2900 */
2901 r5c_release_extra_page(sh);
e33129d8
DW		 2902
2903 for (i = disks; i--; ) {
2904 struct r5dev *dev = &sh->dev[i];
2905
2906 if (dev->towrite) {
2907 set_bit(R5_LOCKED, &dev->flags);
d8ee0728 2908 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2909 if (!expand)
2910 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2911 s->locked++;
1e6d690b
SL		 2912 } else if (test_bit(R5_InJournal, &dev->flags)) {
2913 set_bit(R5_LOCKED, &dev->flags);
2914 s->locked++;
e33129d8
DW		 2915 }
2916 }
ce7d363a
N		 2917 /* if we are not expanding this is a proper write request, and
2918 * there will be bios with new data to be drained into the
2919 * stripe cache
2920 */
2921 if (!expand) {
2922 if (!s->locked)
2923 /* False alarm, nothing to do */
2924 return;
2925 sh->reconstruct_state = reconstruct_state_drain_run;
2926 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2927 } else
2928 sh->reconstruct_state = reconstruct_state_run;
2929
2930 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
2931
c0f7bddb 2932 if (s->locked + conf->max_degraded == disks)
8b3e6cdc 2933 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
c0f7bddb 2934 atomic_inc(&conf->pending_full_writes);
e33129d8
DW		 2935 } else {
2936 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
2937 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
584acdd4
MS		 2938 BUG_ON(level == 6 &&
2939 (!(test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags) ||
2940 test_bit(R5_Wantcompute, &sh->dev[qd_idx].flags))));
e33129d8 2941
e33129d8
DW		 2942 for (i = disks; i--; ) {
2943 struct r5dev *dev = &sh->dev[i];
584acdd4 2944 if (i == pd_idx || i == qd_idx)
e33129d8
DW		 2945 continue;
2946
e33129d8
DW		 2947 if (dev->towrite &&
2948 (test_bit(R5_UPTODATE, &dev->flags) ||
d8ee0728
DW		 2949 test_bit(R5_Wantcompute, &dev->flags))) {
2950 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2951 set_bit(R5_LOCKED, &dev->flags);
2952 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2953 s->locked++;
1e6d690b
SL		 2954 } else if (test_bit(R5_InJournal, &dev->flags)) {
2955 set_bit(R5_LOCKED, &dev->flags);
2956 s->locked++;
e33129d8
DW		 2957 }
2958 }
ce7d363a
N		 2959 if (!s->locked)
2960 /* False alarm - nothing to do */
2961 return;
2962 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
2963 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
2964 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2965 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
e33129d8
DW		 2966 }
2967
c0f7bddb 2968 /* keep the parity disk(s) locked while asynchronous operations
e33129d8
DW		 2969 * are in flight
2970 */
2971 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
2972 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
600aa109 2973 s->locked++;
e33129d8 2974
c0f7bddb
YT		 2975 if (level == 6) {
2976 int qd_idx = sh->qd_idx;
2977 struct r5dev *dev = &sh->dev[qd_idx];
2978
2979 set_bit(R5_LOCKED, &dev->flags);
2980 clear_bit(R5_UPTODATE, &dev->flags);
2981 s->locked++;
2982 }
2983
600aa109 2984 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
e46b272b 2985 __func__, (unsigned long long)sh->sector,
600aa109 2986 s->locked, s->ops_request);
e33129d8 2987}
16a53ecc 2988
1da177e4
LT		 2989/*
2990 * Each stripe/dev can have one or more bion attached.
16a53ecc 2991 * toread/towrite point to the first in a chain.
1da177e4
LT		 2992 * The bi_next chain must be in order.
2993 */
da41ba65 2994static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx,
2995 int forwrite, int previous)
1da177e4
LT		 2996{
2997 struct bio **bip;
d1688a6d 2998 struct r5conf *conf = sh->raid_conf;
72626685 2999 int firstwrite=0;
1da177e4 3000
cbe47ec5 3001 pr_debug("adding bi b#%llu to stripe s#%llu\n",
4f024f37 3002 (unsigned long long)bi->bi_iter.bi_sector,
1da177e4
LT		 3003 (unsigned long long)sh->sector);
3004
b17459c0
SL		 3005 /*
3006 * If several bio share a stripe. The bio bi_phys_segments acts as a
3007 * reference count to avoid race. The reference count should already be
3008 * increased before this function is called (for example, in
849674e4 3009 * raid5_make_request()), so other bio sharing this stripe will not free the
b17459c0
SL		 3010 * stripe. If a stripe is owned by one stripe, the stripe lock will
3011 * protect it.
3012 */
3013 spin_lock_irq(&sh->stripe_lock);
59fc630b 3014 /* Don't allow new IO added to stripes in batch list */
3015 if (sh->batch_head)
3016 goto overlap;
72626685 3017 if (forwrite) {
1da177e4 3018 bip = &sh->dev[dd_idx].towrite;
7eaf7e8e 3019 if (*bip == NULL)
72626685
N		 3020 firstwrite = 1;
3021 } else
1da177e4 3022 bip = &sh->dev[dd_idx].toread;
4f024f37
KO		 3023 while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) {
3024 if (bio_end_sector(*bip) > bi->bi_iter.bi_sector)
1da177e4
LT		 3025 goto overlap;
3026 bip = & (*bip)->bi_next;
3027 }
4f024f37 3028 if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi))
1da177e4
LT		 3029 goto overlap;
3030
da41ba65 3031 if (!forwrite || previous)
3032 clear_bit(STRIPE_BATCH_READY, &sh->state);
3033
78bafebd 3034 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
1da177e4
LT		 3035 if (*bip)
3036 bi->bi_next = *bip;
3037 *bip = bi;
e7836bd6 3038 raid5_inc_bi_active_stripes(bi);
72626685 3039
1da177e4
LT		 3040 if (forwrite) {
3041 /* check if page is covered */
3042 sector_t sector = sh->dev[dd_idx].sector;
3043 for (bi=sh->dev[dd_idx].towrite;
3044 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
4f024f37 3045 bi && bi->bi_iter.bi_sector <= sector;
1da177e4 3046 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
f73a1c7d
KO		 3047 if (bio_end_sector(bi) >= sector)
3048 sector = bio_end_sector(bi);
1da177e4
LT		 3049 }
3050 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
7a87f434 3051 if (!test_and_set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags))
3052 sh->overwrite_disks++;
1da177e4 3053 }
cbe47ec5
N		 3054
3055 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
4f024f37 3056 (unsigned long long)(*bip)->bi_iter.bi_sector,
cbe47ec5
N		 3057 (unsigned long long)sh->sector, dd_idx);
3058
3059 if (conf->mddev->bitmap && firstwrite) {
d0852df5
N		 3060 /* Cannot hold spinlock over bitmap_startwrite,
3061 * but must ensure this isn't added to a batch until
3062 * we have added to the bitmap and set bm_seq.
3063 * So set STRIPE_BITMAP_PENDING to prevent
3064 * batching.
3065 * If multiple add_stripe_bio() calls race here they
3066 * much all set STRIPE_BITMAP_PENDING. So only the first one
3067 * to complete "bitmap_startwrite" gets to set
3068 * STRIPE_BIT_DELAY. This is important as once a stripe
3069 * is added to a batch, STRIPE_BIT_DELAY cannot be changed
3070 * any more.
3071 */
3072 set_bit(STRIPE_BITMAP_PENDING, &sh->state);
3073 spin_unlock_irq(&sh->stripe_lock);
cbe47ec5
N		 3074 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
3075 STRIPE_SECTORS, 0);
d0852df5
N		 3076 spin_lock_irq(&sh->stripe_lock);
3077 clear_bit(STRIPE_BITMAP_PENDING, &sh->state);
3078 if (!sh->batch_head) {
3079 sh->bm_seq = conf->seq_flush+1;
3080 set_bit(STRIPE_BIT_DELAY, &sh->state);
3081 }
cbe47ec5 3082 }
d0852df5 3083 spin_unlock_irq(&sh->stripe_lock);
59fc630b 3084
3085 if (stripe_can_batch(sh))
3086 stripe_add_to_batch_list(conf, sh);
1da177e4
LT		 3087 return 1;
3088
3089 overlap:
3090 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
b17459c0 3091 spin_unlock_irq(&sh->stripe_lock);
1da177e4
LT		 3092 return 0;
3093}
3094
d1688a6d 3095static void end_reshape(struct r5conf *conf);
29269553 3096
d1688a6d 3097static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
911d4ee8 3098 struct stripe_head *sh)
ccfcc3c1 3099{
784052ec 3100 int sectors_per_chunk =
09c9e5fa 3101 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
911d4ee8 3102 int dd_idx;
2d2063ce 3103 int chunk_offset = sector_div(stripe, sectors_per_chunk);
112bf897 3104 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
2d2063ce 3105
112bf897
N		 3106 raid5_compute_sector(conf,
3107 stripe * (disks - conf->max_degraded)
b875e531 3108 *sectors_per_chunk + chunk_offset,
112bf897 3109 previous,
911d4ee8 3110 &dd_idx, sh);
ccfcc3c1
N		 3111}
3112
a4456856 3113static void
d1688a6d 3114handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh,
a4456856 3115 struct stripe_head_state *s, int disks,
34a6f80e 3116 struct bio_list *return_bi)
a4456856
DW		 3117{
3118 int i;
59fc630b 3119 BUG_ON(sh->batch_head);
a4456856
DW		 3120 for (i = disks; i--; ) {
3121 struct bio *bi;
3122 int bitmap_end = 0;
3123
3124 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
3cb03002 3125 struct md_rdev *rdev;
a4456856
DW		 3126 rcu_read_lock();
3127 rdev = rcu_dereference(conf->disks[i].rdev);
f5b67ae8
N		 3128 if (rdev && test_bit(In_sync, &rdev->flags) &&
3129 !test_bit(Faulty, &rdev->flags))
7f0da59b
N		 3130 atomic_inc(&rdev->nr_pending);
3131 else
3132 rdev = NULL;
a4456856 3133 rcu_read_unlock();
7f0da59b
N		 3134 if (rdev) {
3135 if (!rdev_set_badblocks(
3136 rdev,
3137 sh->sector,
3138 STRIPE_SECTORS, 0))
3139 md_error(conf->mddev, rdev);
3140 rdev_dec_pending(rdev, conf->mddev);
3141 }
a4456856 3142 }
b17459c0 3143 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 3144 /* fail all writes first */
3145 bi = sh->dev[i].towrite;
3146 sh->dev[i].towrite = NULL;
7a87f434 3147 sh->overwrite_disks = 0;
b17459c0 3148 spin_unlock_irq(&sh->stripe_lock);
1ed850f3 3149 if (bi)
a4456856 3150 bitmap_end = 1;
a4456856 3151
0576b1c6
SL		 3152 r5l_stripe_write_finished(sh);
3153
a4456856
DW		 3154 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
3155 wake_up(&conf->wait_for_overlap);
3156
4f024f37 3157 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3158 sh->dev[i].sector + STRIPE_SECTORS) {
3159 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
4246a0b6
CH		 3160
3161 bi->bi_error = -EIO;
e7836bd6 3162 if (!raid5_dec_bi_active_stripes(bi)) {
a4456856 3163 md_write_end(conf->mddev);
34a6f80e 3164 bio_list_add(return_bi, bi);
a4456856
DW		 3165 }
3166 bi = nextbi;
3167 }
7eaf7e8e
SL		 3168 if (bitmap_end)
3169 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3170 STRIPE_SECTORS, 0, 0);
3171 bitmap_end = 0;
a4456856
DW		 3172 /* and fail all 'written' */
3173 bi = sh->dev[i].written;
3174 sh->dev[i].written = NULL;
d592a996
SL		 3175 if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) {
3176 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
3177 sh->dev[i].page = sh->dev[i].orig_page;
3178 }
3179
a4456856 3180 if (bi) bitmap_end = 1;
4f024f37 3181 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3182 sh->dev[i].sector + STRIPE_SECTORS) {
3183 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
4246a0b6
CH		 3184
3185 bi->bi_error = -EIO;
e7836bd6 3186 if (!raid5_dec_bi_active_stripes(bi)) {
a4456856 3187 md_write_end(conf->mddev);
34a6f80e 3188 bio_list_add(return_bi, bi);
a4456856
DW		 3189 }
3190 bi = bi2;
3191 }
3192
b5e98d65
DW		 3193 /* fail any reads if this device is non-operational and
3194 * the data has not reached the cache yet.
3195 */
3196 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
6e74a9cf 3197 s->failed > conf->max_degraded &&
b5e98d65
DW		 3198 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
3199 test_bit(R5_ReadError, &sh->dev[i].flags))) {
143c4d05 3200 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 3201 bi = sh->dev[i].toread;
3202 sh->dev[i].toread = NULL;
143c4d05 3203 spin_unlock_irq(&sh->stripe_lock);
a4456856
DW		 3204 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
3205 wake_up(&conf->wait_for_overlap);
ebda780b
SL		 3206 if (bi)
3207 s->to_read--;
4f024f37 3208 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3209 sh->dev[i].sector + STRIPE_SECTORS) {
3210 struct bio *nextbi =
3211 r5_next_bio(bi, sh->dev[i].sector);
4246a0b6
CH		 3212
3213 bi->bi_error = -EIO;
34a6f80e
N		 3214 if (!raid5_dec_bi_active_stripes(bi))
3215 bio_list_add(return_bi, bi);
a4456856
DW		 3216 bi = nextbi;
3217 }
3218 }
a4456856
DW		 3219 if (bitmap_end)
3220 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3221 STRIPE_SECTORS, 0, 0);
8cfa7b0f
N		 3222 /* If we were in the middle of a write the parity block might
3223 * still be locked - so just clear all R5_LOCKED flags
3224 */
3225 clear_bit(R5_LOCKED, &sh->dev[i].flags);
a4456856 3226 }
ebda780b
SL		 3227 s->to_write = 0;
3228 s->written = 0;
a4456856 3229
8b3e6cdc
DW		 3230 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3231 if (atomic_dec_and_test(&conf->pending_full_writes))
3232 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 3233}
3234
7f0da59b 3235static void
d1688a6d 3236handle_failed_sync(struct r5conf *conf, struct stripe_head *sh,
7f0da59b
N		 3237 struct stripe_head_state *s)
3238{
3239 int abort = 0;
3240 int i;
3241
59fc630b 3242 BUG_ON(sh->batch_head);
7f0da59b 3243 clear_bit(STRIPE_SYNCING, &sh->state);
f8dfcffd
N		 3244 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
3245 wake_up(&conf->wait_for_overlap);
7f0da59b 3246 s->syncing = 0;
9a3e1101 3247 s->replacing = 0;
7f0da59b 3248 /* There is nothing more to do for sync/check/repair.
18b9837e
N		 3249 * Don't even need to abort as that is handled elsewhere
3250 * if needed, and not always wanted e.g. if there is a known
3251 * bad block here.
9a3e1101 3252 * For recover/replace we need to record a bad block on all
7f0da59b
N		 3253 * non-sync devices, or abort the recovery
3254 */
18b9837e
N		 3255 if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) {
3256 /* During recovery devices cannot be removed, so
3257 * locking and refcounting of rdevs is not needed
3258 */
e50d3992 3259 rcu_read_lock();
18b9837e 3260 for (i = 0; i < conf->raid_disks; i++) {
e50d3992 3261 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
18b9837e
N		 3262 if (rdev
3263 && !test_bit(Faulty, &rdev->flags)
3264 && !test_bit(In_sync, &rdev->flags)
3265 && !rdev_set_badblocks(rdev, sh->sector,
3266 STRIPE_SECTORS, 0))
3267 abort = 1;
e50d3992 3268 rdev = rcu_dereference(conf->disks[i].replacement);
18b9837e
N		 3269 if (rdev
3270 && !test_bit(Faulty, &rdev->flags)
3271 && !test_bit(In_sync, &rdev->flags)
3272 && !rdev_set_badblocks(rdev, sh->sector,
3273 STRIPE_SECTORS, 0))
3274 abort = 1;
3275 }
e50d3992 3276 rcu_read_unlock();
18b9837e
N		 3277 if (abort)
3278 conf->recovery_disabled =
3279 conf->mddev->recovery_disabled;
7f0da59b 3280 }
18b9837e 3281 md_done_sync(conf->mddev, STRIPE_SECTORS, !abort);
7f0da59b
N		 3282}
3283
9a3e1101
N		 3284static int want_replace(struct stripe_head *sh, int disk_idx)
3285{
3286 struct md_rdev *rdev;
3287 int rv = 0;
3f232d6a
N		 3288
3289 rcu_read_lock();
3290 rdev = rcu_dereference(sh->raid_conf->disks[disk_idx].replacement);
9a3e1101
N		 3291 if (rdev
3292 && !test_bit(Faulty, &rdev->flags)
3293 && !test_bit(In_sync, &rdev->flags)
3294 && (rdev->recovery_offset <= sh->sector
3295 || rdev->mddev->recovery_cp <= sh->sector))
3296 rv = 1;
3f232d6a 3297 rcu_read_unlock();
9a3e1101
N		 3298 return rv;
3299}
3300
93b3dbce 3301/* fetch_block - checks the given member device to see if its data needs
1fe797e6
DW		 3302 * to be read or computed to satisfy a request.
3303 *
3304 * Returns 1 when no more member devices need to be checked, otherwise returns
93b3dbce 3305 * 0 to tell the loop in handle_stripe_fill to continue
f38e1219 3306 */
2c58f06e
N		 3307
3308static int need_this_block(struct stripe_head *sh, struct stripe_head_state *s,
3309 int disk_idx, int disks)
a4456856 3310{
5599becc 3311 struct r5dev *dev = &sh->dev[disk_idx];
f2b3b44d
N		 3312 struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]],
3313 &sh->dev[s->failed_num[1]] };
ea664c82 3314 int i;
5599becc 3315
a79cfe12
N		 3316
3317 if (test_bit(R5_LOCKED, &dev->flags) ||
3318 test_bit(R5_UPTODATE, &dev->flags))
3319 /* No point reading this as we already have it or have
3320 * decided to get it.
3321 */
3322 return 0;
3323
3324 if (dev->toread ||
3325 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)))
3326 /* We need this block to directly satisfy a request */
3327 return 1;
3328
3329 if (s->syncing || s->expanding ||
3330 (s->replacing && want_replace(sh, disk_idx)))
3331 /* When syncing, or expanding we read everything.
3332 * When replacing, we need the replaced block.
3333 */
3334 return 1;
3335
3336 if ((s->failed >= 1 && fdev[0]->toread) ||
3337 (s->failed >= 2 && fdev[1]->toread))
3338 /* If we want to read from a failed device, then
3339 * we need to actually read every other device.
3340 */
3341 return 1;
3342
a9d56950
N		 3343 /* Sometimes neither read-modify-write nor reconstruct-write
3344 * cycles can work. In those cases we read every block we
3345 * can. Then the parity-update is certain to have enough to
3346 * work with.
3347 * This can only be a problem when we need to write something,
3348 * and some device has failed. If either of those tests
3349 * fail we need look no further.
3350 */
3351 if (!s->failed || !s->to_write)
3352 return 0;
3353
3354 if (test_bit(R5_Insync, &dev->flags) &&
3355 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3356 /* Pre-reads at not permitted until after short delay
3357 * to gather multiple requests. However if this
3358 * device is no Insync, the block could only be be computed
3359 * and there is no need to delay that.
3360 */
3361 return 0;
ea664c82 3362
36707bb2 3363 for (i = 0; i < s->failed && i < 2; i++) {
ea664c82
N		 3364 if (fdev[i]->towrite &&
3365 !test_bit(R5_UPTODATE, &fdev[i]->flags) &&
3366 !test_bit(R5_OVERWRITE, &fdev[i]->flags))
3367 /* If we have a partial write to a failed
3368 * device, then we will need to reconstruct
3369 * the content of that device, so all other
3370 * devices must be read.
3371 */
3372 return 1;
3373 }
3374
3375 /* If we are forced to do a reconstruct-write, either because
3376 * the current RAID6 implementation only supports that, or
3377 * or because parity cannot be trusted and we are currently
3378 * recovering it, there is extra need to be careful.
3379 * If one of the devices that we would need to read, because
3380 * it is not being overwritten (and maybe not written at all)
3381 * is missing/faulty, then we need to read everything we can.
3382 */
3383 if (sh->raid_conf->level != 6 &&
3384 sh->sector < sh->raid_conf->mddev->recovery_cp)
3385 /* reconstruct-write isn't being forced */
3386 return 0;
36707bb2 3387 for (i = 0; i < s->failed && i < 2; i++) {
10d82c5f
N		 3388 if (s->failed_num[i] != sh->pd_idx &&
3389 s->failed_num[i] != sh->qd_idx &&
3390 !test_bit(R5_UPTODATE, &fdev[i]->flags) &&
ea664c82
N		 3391 !test_bit(R5_OVERWRITE, &fdev[i]->flags))
3392 return 1;
3393 }
3394
2c58f06e
N		 3395 return 0;
3396}
3397
3398static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s,
3399 int disk_idx, int disks)
3400{
3401 struct r5dev *dev = &sh->dev[disk_idx];
3402
3403 /* is the data in this block needed, and can we get it? */
3404 if (need_this_block(sh, s, disk_idx, disks)) {
5599becc
YT		 3405 /* we would like to get this block, possibly by computing it,
3406 * otherwise read it if the backing disk is insync
3407 */
3408 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
3409 BUG_ON(test_bit(R5_Wantread, &dev->flags));
b0c783b3 3410 BUG_ON(sh->batch_head);
5599becc 3411 if ((s->uptodate == disks - 1) &&
f2b3b44d
N		 3412 (s->failed && (disk_idx == s->failed_num[0] ||
3413 disk_idx == s->failed_num[1]))) {
5599becc
YT		 3414 /* have disk failed, and we're requested to fetch it;
3415 * do compute it
a4456856 3416 */
5599becc
YT		 3417 pr_debug("Computing stripe %llu block %d\n",
3418 (unsigned long long)sh->sector, disk_idx);
3419 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3420 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3421 set_bit(R5_Wantcompute, &dev->flags);
3422 sh->ops.target = disk_idx;
3423 sh->ops.target2 = -1; /* no 2nd target */
3424 s->req_compute = 1;
93b3dbce
N		 3425 /* Careful: from this point on 'uptodate' is in the eye
3426 * of raid_run_ops which services 'compute' operations
3427 * before writes. R5_Wantcompute flags a block that will
3428 * be R5_UPTODATE by the time it is needed for a
3429 * subsequent operation.
3430 */
5599becc
YT		 3431 s->uptodate++;
3432 return 1;
3433 } else if (s->uptodate == disks-2 && s->failed >= 2) {
3434 /* Computing 2-failure is *very* expensive; only
3435 * do it if failed >= 2
3436 */
3437 int other;
3438 for (other = disks; other--; ) {
3439 if (other == disk_idx)
3440 continue;
3441 if (!test_bit(R5_UPTODATE,
3442 &sh->dev[other].flags))
3443 break;
a4456856 3444 }
5599becc
YT		 3445 BUG_ON(other < 0);
3446 pr_debug("Computing stripe %llu blocks %d,%d\n",
3447 (unsigned long long)sh->sector,
3448 disk_idx, other);
3449 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3450 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3451 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
3452 set_bit(R5_Wantcompute, &sh->dev[other].flags);
3453 sh->ops.target = disk_idx;
3454 sh->ops.target2 = other;
3455 s->uptodate += 2;
3456 s->req_compute = 1;
3457 return 1;
3458 } else if (test_bit(R5_Insync, &dev->flags)) {
3459 set_bit(R5_LOCKED, &dev->flags);
3460 set_bit(R5_Wantread, &dev->flags);
3461 s->locked++;
3462 pr_debug("Reading block %d (sync=%d)\n",
3463 disk_idx, s->syncing);
a4456856
DW		 3464 }
3465 }
5599becc
YT		 3466
3467 return 0;
3468}
3469
3470/**
93b3dbce 3471 * handle_stripe_fill - read or compute data to satisfy pending requests.
5599becc 3472 */
93b3dbce
N		 3473static void handle_stripe_fill(struct stripe_head *sh,
3474 struct stripe_head_state *s,
3475 int disks)
5599becc
YT		 3476{
3477 int i;
3478
3479 /* look for blocks to read/compute, skip this if a compute
3480 * is already in flight, or if the stripe contents are in the
3481 * midst of changing due to a write
3482 */
3483 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
3484 !sh->reconstruct_state)
3485 for (i = disks; i--; )
93b3dbce 3486 if (fetch_block(sh, s, i, disks))
5599becc 3487 break;
a4456856
DW		 3488 set_bit(STRIPE_HANDLE, &sh->state);
3489}
3490
787b76fa
N		 3491static void break_stripe_batch_list(struct stripe_head *head_sh,
3492 unsigned long handle_flags);
1fe797e6 3493/* handle_stripe_clean_event
a4456856
DW		 3494 * any written block on an uptodate or failed drive can be returned.
3495 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
3496 * never LOCKED, so we don't need to test 'failed' directly.
3497 */
d1688a6d 3498static void handle_stripe_clean_event(struct r5conf *conf,
34a6f80e 3499 struct stripe_head *sh, int disks, struct bio_list *return_bi)
a4456856
DW		 3500{
3501 int i;
3502 struct r5dev *dev;
f8dfcffd 3503 int discard_pending = 0;
59fc630b 3504 struct stripe_head *head_sh = sh;
3505 bool do_endio = false;
a4456856
DW		 3506
3507 for (i = disks; i--; )
3508 if (sh->dev[i].written) {
3509 dev = &sh->dev[i];
3510 if (!test_bit(R5_LOCKED, &dev->flags) &&
9e444768 3511 (test_bit(R5_UPTODATE, &dev->flags) ||
d592a996
SL		 3512 test_bit(R5_Discard, &dev->flags) ||
3513 test_bit(R5_SkipCopy, &dev->flags))) {
a4456856
DW		 3514 /* We can return any write requests */
3515 struct bio *wbi, *wbi2;
45b4233c 3516 pr_debug("Return write for disc %d\n", i);
ca64cae9
N		 3517 if (test_and_clear_bit(R5_Discard, &dev->flags))
3518 clear_bit(R5_UPTODATE, &dev->flags);
d592a996
SL		 3519 if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) {
3520 WARN_ON(test_bit(R5_UPTODATE, &dev->flags));
d592a996 3521 }
59fc630b 3522 do_endio = true;
3523
3524returnbi:
3525 dev->page = dev->orig_page;
a4456856
DW		 3526 wbi = dev->written;
3527 dev->written = NULL;
4f024f37 3528 while (wbi && wbi->bi_iter.bi_sector <
a4456856
DW		 3529 dev->sector + STRIPE_SECTORS) {
3530 wbi2 = r5_next_bio(wbi, dev->sector);
e7836bd6 3531 if (!raid5_dec_bi_active_stripes(wbi)) {
a4456856 3532 md_write_end(conf->mddev);
34a6f80e 3533 bio_list_add(return_bi, wbi);
a4456856
DW		 3534 }
3535 wbi = wbi2;
3536 }
7eaf7e8e
SL		 3537 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3538 STRIPE_SECTORS,
a4456856 3539 !test_bit(STRIPE_DEGRADED, &sh->state),
7eaf7e8e 3540 0);
59fc630b 3541 if (head_sh->batch_head) {
3542 sh = list_first_entry(&sh->batch_list,
3543 struct stripe_head,
3544 batch_list);
3545 if (sh != head_sh) {
3546 dev = &sh->dev[i];
3547 goto returnbi;
3548 }
3549 }
3550 sh = head_sh;
3551 dev = &sh->dev[i];
f8dfcffd
N		 3552 } else if (test_bit(R5_Discard, &dev->flags))
3553 discard_pending = 1;
3554 }
f6bed0ef 3555
0576b1c6
SL		 3556 r5l_stripe_write_finished(sh);
3557
f8dfcffd
N		 3558 if (!discard_pending &&
3559 test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) {
b8a9d66d 3560 int hash;
f8dfcffd
N		 3561 clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
3562 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
3563 if (sh->qd_idx >= 0) {
3564 clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
3565 clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags);
3566 }
3567 /* now that discard is done we can proceed with any sync */
3568 clear_bit(STRIPE_DISCARD, &sh->state);
d47648fc
SL		 3569 /*
3570 * SCSI discard will change some bio fields and the stripe has
3571 * no updated data, so remove it from hash list and the stripe
3572 * will be reinitialized
3573 */
59fc630b 3574unhash:
b8a9d66d
RG		 3575 hash = sh->hash_lock_index;
3576 spin_lock_irq(conf->hash_locks + hash);
d47648fc 3577 remove_hash(sh);
b8a9d66d 3578 spin_unlock_irq(conf->hash_locks + hash);
59fc630b 3579 if (head_sh->batch_head) {
3580 sh = list_first_entry(&sh->batch_list,
3581 struct stripe_head, batch_list);
3582 if (sh != head_sh)
3583 goto unhash;
3584 }
59fc630b 3585 sh = head_sh;
3586
f8dfcffd
N		 3587 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state))
3588 set_bit(STRIPE_HANDLE, &sh->state);
3589
3590 }
8b3e6cdc
DW		 3591
3592 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3593 if (atomic_dec_and_test(&conf->pending_full_writes))
3594 md_wakeup_thread(conf->mddev->thread);
59fc630b 3595
787b76fa
N		 3596 if (head_sh->batch_head && do_endio)
3597 break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS);
a4456856
DW		 3598}
3599
d7bd398e
SL		 3600static int handle_stripe_dirtying(struct r5conf *conf,
3601 struct stripe_head *sh,
3602 struct stripe_head_state *s,
3603 int disks)
a4456856
DW		 3604{
3605 int rmw = 0, rcw = 0, i;
a7854487
AL		 3606 sector_t recovery_cp = conf->mddev->recovery_cp;
3607
584acdd4 3608 /* Check whether resync is now happening or should start.
a7854487
AL		 3609 * If yes, then the array is dirty (after unclean shutdown or
3610 * initial creation), so parity in some stripes might be inconsistent.
3611 * In this case, we need to always do reconstruct-write, to ensure
3612 * that in case of drive failure or read-error correction, we
3613 * generate correct data from the parity.
3614 */
584acdd4 3615 if (conf->rmw_level == PARITY_DISABLE_RMW ||
26ac1073
N		 3616 (recovery_cp < MaxSector && sh->sector >= recovery_cp &&
3617 s->failed == 0)) {
a7854487 3618 /* Calculate the real rcw later - for now make it
c8ac1803
N		 3619 * look like rcw is cheaper
3620 */
3621 rcw = 1; rmw = 2;
584acdd4
MS		 3622 pr_debug("force RCW rmw_level=%u, recovery_cp=%llu sh->sector=%llu\n",
3623 conf->rmw_level, (unsigned long long)recovery_cp,
a7854487 3624 (unsigned long long)sh->sector);
c8ac1803 3625 } else for (i = disks; i--; ) {
a4456856
DW		 3626 /* would I have to read this buffer for read_modify_write */
3627 struct r5dev *dev = &sh->dev[i];
1e6d690b
SL		 3628 if ((dev->towrite || i == sh->pd_idx || i == sh->qd_idx ||
3629 test_bit(R5_InJournal, &dev->flags)) &&
a4456856 3630 !test_bit(R5_LOCKED, &dev->flags) &&
1e6d690b
SL		 3631 !((test_bit(R5_UPTODATE, &dev->flags) &&
3632 (!test_bit(R5_InJournal, &dev->flags) ||
3633 dev->page != dev->orig_page)) ||
f38e1219 3634 test_bit(R5_Wantcompute, &dev->flags))) {
a4456856
DW		 3635 if (test_bit(R5_Insync, &dev->flags))
3636 rmw++;
3637 else
3638 rmw += 2*disks; /* cannot read it */
3639 }
3640 /* Would I have to read this buffer for reconstruct_write */
584acdd4
MS		 3641 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
3642 i != sh->pd_idx && i != sh->qd_idx &&
a4456856 3643 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219 3644 !(test_bit(R5_UPTODATE, &dev->flags) ||
1e6d690b
SL		 3645 test_bit(R5_InJournal, &dev->flags) ||
3646 test_bit(R5_Wantcompute, &dev->flags))) {
67f45548
N		 3647 if (test_bit(R5_Insync, &dev->flags))
3648 rcw++;
a4456856
DW		 3649 else
3650 rcw += 2*disks;
3651 }
3652 }
1e6d690b 3653
45b4233c 3654 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
a4456856
DW		 3655 (unsigned long long)sh->sector, rmw, rcw);
3656 set_bit(STRIPE_HANDLE, &sh->state);
41257580 3657 if ((rmw < rcw || (rmw == rcw && conf->rmw_level == PARITY_PREFER_RMW)) && rmw > 0) {
a4456856 3658 /* prefer read-modify-write, but need to get some data */
e3620a3a
JB		 3659 if (conf->mddev->queue)
3660 blk_add_trace_msg(conf->mddev->queue,
3661 "raid5 rmw %llu %d",
3662 (unsigned long long)sh->sector, rmw);
a4456856
DW		 3663 for (i = disks; i--; ) {
3664 struct r5dev *dev = &sh->dev[i];
1e6d690b
SL		 3665 if (test_bit(R5_InJournal, &dev->flags) &&
3666 dev->page == dev->orig_page &&
3667 !test_bit(R5_LOCKED, &sh->dev[sh->pd_idx].flags)) {
3668 /* alloc page for prexor */
d7bd398e
SL		 3669 struct page *p = alloc_page(GFP_NOIO);
3670
3671 if (p) {
3672 dev->orig_page = p;
3673 continue;
3674 }
3675
3676 /*
3677 * alloc_page() failed, try use
3678 * disk_info->extra_page
3679 */
3680 if (!test_and_set_bit(R5C_EXTRA_PAGE_IN_USE,
3681 &conf->cache_state)) {
3682 r5c_use_extra_page(sh);
3683 break;
3684 }
1e6d690b 3685
d7bd398e
SL		 3686 /* extra_page in use, add to delayed_list */
3687 set_bit(STRIPE_DELAYED, &sh->state);
3688 s->waiting_extra_page = 1;
3689 return -EAGAIN;
1e6d690b 3690 }
d7bd398e 3691 }
1e6d690b 3692
d7bd398e
SL		 3693 for (i = disks; i--; ) {
3694 struct r5dev *dev = &sh->dev[i];
1e6d690b
SL		 3695 if ((dev->towrite ||
3696 i == sh->pd_idx || i == sh->qd_idx ||
3697 test_bit(R5_InJournal, &dev->flags)) &&
a4456856 3698 !test_bit(R5_LOCKED, &dev->flags) &&
1e6d690b
SL		 3699 !((test_bit(R5_UPTODATE, &dev->flags) &&
3700 (!test_bit(R5_InJournal, &dev->flags) ||
3701 dev->page != dev->orig_page)) ||
3702 test_bit(R5_Wantcompute, &dev->flags)) &&
a4456856 3703 test_bit(R5_Insync, &dev->flags)) {
67f45548
N		 3704 if (test_bit(STRIPE_PREREAD_ACTIVE,
3705 &sh->state)) {
3706 pr_debug("Read_old block %d for r-m-w\n",
3707 i);
a4456856
DW		 3708 set_bit(R5_LOCKED, &dev->flags);
3709 set_bit(R5_Wantread, &dev->flags);
3710 s->locked++;
3711 } else {
3712 set_bit(STRIPE_DELAYED, &sh->state);
3713 set_bit(STRIPE_HANDLE, &sh->state);
3714 }
3715 }
3716 }
a9add5d9 3717 }
41257580 3718 if ((rcw < rmw || (rcw == rmw && conf->rmw_level != PARITY_PREFER_RMW)) && rcw > 0) {
a4456856 3719 /* want reconstruct write, but need to get some data */
a9add5d9 3720 int qread =0;
c8ac1803 3721 rcw = 0;
a4456856
DW		 3722 for (i = disks; i--; ) {
3723 struct r5dev *dev = &sh->dev[i];
3724 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
c8ac1803 3725 i != sh->pd_idx && i != sh->qd_idx &&
a4456856 3726 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219 3727 !(test_bit(R5_UPTODATE, &dev->flags) ||
1e6d690b 3728 test_bit(R5_InJournal, &dev->flags) ||
c8ac1803
N		 3729 test_bit(R5_Wantcompute, &dev->flags))) {
3730 rcw++;
67f45548
N		 3731 if (test_bit(R5_Insync, &dev->flags) &&
3732 test_bit(STRIPE_PREREAD_ACTIVE,
3733 &sh->state)) {
45b4233c 3734 pr_debug("Read_old block "
a4456856
DW		 3735 "%d for Reconstruct\n", i);
3736 set_bit(R5_LOCKED, &dev->flags);
3737 set_bit(R5_Wantread, &dev->flags);
3738 s->locked++;
a9add5d9 3739 qread++;
a4456856
DW		 3740 } else {
3741 set_bit(STRIPE_DELAYED, &sh->state);
3742 set_bit(STRIPE_HANDLE, &sh->state);
3743 }
3744 }
3745 }
e3620a3a 3746 if (rcw && conf->mddev->queue)
a9add5d9
N		 3747 blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d",
3748 (unsigned long long)sh->sector,
3749 rcw, qread, test_bit(STRIPE_DELAYED, &sh->state));
c8ac1803 3750 }
b1b02fe9
N		 3751
3752 if (rcw > disks && rmw > disks &&
3753 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3754 set_bit(STRIPE_DELAYED, &sh->state);
3755
a4456856
DW		 3756 /* now if nothing is locked, and if we have enough data,
3757 * we can start a write request
3758 */
f38e1219
DW		 3759 /* since handle_stripe can be called at any time we need to handle the
3760 * case where a compute block operation has been submitted and then a
ac6b53b6
DW		 3761 * subsequent call wants to start a write request. raid_run_ops only
3762 * handles the case where compute block and reconstruct are requested
f38e1219
DW		 3763 * simultaneously. If this is not the case then new writes need to be
3764 * held off until the compute completes.
3765 */
976ea8d4
DW		 3766 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
3767 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
1e6d690b 3768 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
c0f7bddb 3769 schedule_reconstruction(sh, s, rcw == 0, 0);
d7bd398e 3770 return 0;
a4456856
DW		 3771}
3772
d1688a6d 3773static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh,
a4456856
DW		 3774 struct stripe_head_state *s, int disks)
3775{
ecc65c9b 3776 struct r5dev *dev = NULL;
bd2ab670 3777
59fc630b 3778 BUG_ON(sh->batch_head);
a4456856 3779 set_bit(STRIPE_HANDLE, &sh->state);
e89f8962 3780
ecc65c9b
DW		 3781 switch (sh->check_state) {
3782 case check_state_idle:
3783 /* start a new check operation if there are no failures */
bd2ab670 3784 if (s->failed == 0) {
bd2ab670 3785 BUG_ON(s->uptodate != disks);
ecc65c9b
DW		 3786 sh->check_state = check_state_run;
3787 set_bit(STRIPE_OP_CHECK, &s->ops_request);
bd2ab670 3788 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
bd2ab670 3789 s->uptodate--;
ecc65c9b 3790 break;
bd2ab670 3791 }
f2b3b44d 3792 dev = &sh->dev[s->failed_num[0]];
ecc65c9b
DW		 3793 /* fall through */
3794 case check_state_compute_result:
3795 sh->check_state = check_state_idle;
3796 if (!dev)
3797 dev = &sh->dev[sh->pd_idx];
3798
3799 /* check that a write has not made the stripe insync */
3800 if (test_bit(STRIPE_INSYNC, &sh->state))
3801 break;
c8894419 3802
a4456856 3803 /* either failed parity check, or recovery is happening */
a4456856
DW		 3804 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
3805 BUG_ON(s->uptodate != disks);
3806
3807 set_bit(R5_LOCKED, &dev->flags);
ecc65c9b 3808 s->locked++;
a4456856 3809 set_bit(R5_Wantwrite, &dev->flags);
830ea016 3810
a4456856 3811 clear_bit(STRIPE_DEGRADED, &sh->state);
a4456856 3812 set_bit(STRIPE_INSYNC, &sh->state);
ecc65c9b
DW		 3813 break;
3814 case check_state_run:
3815 break; /* we will be called again upon completion */
3816 case check_state_check_result:
3817 sh->check_state = check_state_idle;
3818
3819 /* if a failure occurred during the check operation, leave
3820 * STRIPE_INSYNC not set and let the stripe be handled again
3821 */
3822 if (s->failed)
3823 break;
3824
3825 /* handle a successful check operation, if parity is correct
3826 * we are done. Otherwise update the mismatch count and repair
3827 * parity if !MD_RECOVERY_CHECK
3828 */
ad283ea4 3829 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
ecc65c9b
DW		 3830 /* parity is correct (on disc,
3831 * not in buffer any more)
3832 */
3833 set_bit(STRIPE_INSYNC, &sh->state);
3834 else {
7f7583d4 3835 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
ecc65c9b
DW		 3836 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
3837 /* don't try to repair!! */
3838 set_bit(STRIPE_INSYNC, &sh->state);
3839 else {
3840 sh->check_state = check_state_compute_run;
976ea8d4 3841 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
ecc65c9b
DW		 3842 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3843 set_bit(R5_Wantcompute,
3844 &sh->dev[sh->pd_idx].flags);
3845 sh->ops.target = sh->pd_idx;
ac6b53b6 3846 sh->ops.target2 = -1;
ecc65c9b
DW		 3847 s->uptodate++;
3848 }
3849 }
3850 break;
3851 case check_state_compute_run:
3852 break;
3853 default:
cc6167b4 3854 pr_err("%s: unknown check_state: %d sector: %llu\n",
ecc65c9b
DW		 3855 __func__, sh->check_state,
3856 (unsigned long long) sh->sector);
3857 BUG();
a4456856
DW		 3858 }
3859}
3860
d1688a6d 3861static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh,
36d1c647 3862 struct stripe_head_state *s,
f2b3b44d 3863 int disks)
a4456856 3864{
a4456856 3865 int pd_idx = sh->pd_idx;
34e04e87 3866 int qd_idx = sh->qd_idx;
d82dfee0 3867 struct r5dev *dev;
a4456856 3868
59fc630b 3869 BUG_ON(sh->batch_head);
a4456856
DW		 3870 set_bit(STRIPE_HANDLE, &sh->state);
3871
3872 BUG_ON(s->failed > 2);
d82dfee0 3873
a4456856
DW		 3874 /* Want to check and possibly repair P and Q.
3875 * However there could be one 'failed' device, in which
3876 * case we can only check one of them, possibly using the
3877 * other to generate missing data
3878 */
3879
d82dfee0
DW		 3880 switch (sh->check_state) {
3881 case check_state_idle:
3882 /* start a new check operation if there are < 2 failures */
f2b3b44d 3883 if (s->failed == s->q_failed) {
d82dfee0 3884 /* The only possible failed device holds Q, so it
a4456856
DW		 3885 * makes sense to check P (If anything else were failed,
3886 * we would have used P to recreate it).
3887 */
d82dfee0 3888 sh->check_state = check_state_run;
a4456856 3889 }
f2b3b44d 3890 if (!s->q_failed && s->failed < 2) {
d82dfee0 3891 /* Q is not failed, and we didn't use it to generate
a4456856
DW		 3892 * anything, so it makes sense to check it
3893 */
d82dfee0
DW		 3894 if (sh->check_state == check_state_run)
3895 sh->check_state = check_state_run_pq;
3896 else
3897 sh->check_state = check_state_run_q;
a4456856 3898 }
a4456856 3899
d82dfee0
DW		 3900 /* discard potentially stale zero_sum_result */
3901 sh->ops.zero_sum_result = 0;
a4456856 3902
d82dfee0
DW		 3903 if (sh->check_state == check_state_run) {
3904 /* async_xor_zero_sum destroys the contents of P */
3905 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
3906 s->uptodate--;
a4456856 3907 }
d82dfee0
DW		 3908 if (sh->check_state >= check_state_run &&
3909 sh->check_state <= check_state_run_pq) {
3910 /* async_syndrome_zero_sum preserves P and Q, so
3911 * no need to mark them !uptodate here
3912 */
3913 set_bit(STRIPE_OP_CHECK, &s->ops_request);
3914 break;
a4456856
DW		 3915 }
3916
d82dfee0
DW		 3917 /* we have 2-disk failure */
3918 BUG_ON(s->failed != 2);
3919 /* fall through */
3920 case check_state_compute_result:
3921 sh->check_state = check_state_idle;
a4456856 3922
d82dfee0
DW		 3923 /* check that a write has not made the stripe insync */
3924 if (test_bit(STRIPE_INSYNC, &sh->state))
3925 break;
a4456856
DW		 3926
3927 /* now write out any block on a failed drive,
d82dfee0 3928 * or P or Q if they were recomputed
a4456856 3929 */
d82dfee0 3930 BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */
a4456856 3931 if (s->failed == 2) {
f2b3b44d 3932 dev = &sh->dev[s->failed_num[1]];
a4456856
DW		 3933 s->locked++;
3934 set_bit(R5_LOCKED, &dev->flags);
3935 set_bit(R5_Wantwrite, &dev->flags);
3936 }
3937 if (s->failed >= 1) {
f2b3b44d 3938 dev = &sh->dev[s->failed_num[0]];
a4456856
DW		 3939 s->locked++;
3940 set_bit(R5_LOCKED, &dev->flags);
3941 set_bit(R5_Wantwrite, &dev->flags);
3942 }
d82dfee0 3943 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
a4456856
DW		 3944 dev = &sh->dev[pd_idx];
3945 s->locked++;
3946 set_bit(R5_LOCKED, &dev->flags);
3947 set_bit(R5_Wantwrite, &dev->flags);
3948 }
d82dfee0 3949 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
a4456856
DW		 3950 dev = &sh->dev[qd_idx];
3951 s->locked++;
3952 set_bit(R5_LOCKED, &dev->flags);
3953 set_bit(R5_Wantwrite, &dev->flags);
3954 }
3955 clear_bit(STRIPE_DEGRADED, &sh->state);
3956
3957 set_bit(STRIPE_INSYNC, &sh->state);
d82dfee0
DW		 3958 break;
3959 case check_state_run:
3960 case check_state_run_q:
3961 case check_state_run_pq:
3962 break; /* we will be called again upon completion */
3963 case check_state_check_result:
3964 sh->check_state = check_state_idle;
3965
3966 /* handle a successful check operation, if parity is correct
3967 * we are done. Otherwise update the mismatch count and repair
3968 * parity if !MD_RECOVERY_CHECK
3969 */
3970 if (sh->ops.zero_sum_result == 0) {
3971 /* both parities are correct */
3972 if (!s->failed)
3973 set_bit(STRIPE_INSYNC, &sh->state);
3974 else {
3975 /* in contrast to the raid5 case we can validate
3976 * parity, but still have a failure to write
3977 * back
3978 */
3979 sh->check_state = check_state_compute_result;
3980 /* Returning at this point means that we may go
3981 * off and bring p and/or q uptodate again so
3982 * we make sure to check zero_sum_result again
3983 * to verify if p or q need writeback
3984 */
3985 }
3986 } else {
7f7583d4 3987 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
d82dfee0
DW		 3988 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
3989 /* don't try to repair!! */
3990 set_bit(STRIPE_INSYNC, &sh->state);
3991 else {
3992 int *target = &sh->ops.target;
3993
3994 sh->ops.target = -1;
3995 sh->ops.target2 = -1;
3996 sh->check_state = check_state_compute_run;
3997 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3998 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3999 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
4000 set_bit(R5_Wantcompute,
4001 &sh->dev[pd_idx].flags);
4002 *target = pd_idx;
4003 target = &sh->ops.target2;
4004 s->uptodate++;
4005 }
4006 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
4007 set_bit(R5_Wantcompute,
4008 &sh->dev[qd_idx].flags);
4009 *target = qd_idx;
4010 s->uptodate++;
4011 }
4012 }
4013 }
4014 break;
4015 case check_state_compute_run:
4016 break;
4017 default:
cc6167b4
N		 4018 pr_warn("%s: unknown check_state: %d sector: %llu\n",
4019 __func__, sh->check_state,
4020 (unsigned long long) sh->sector);
d82dfee0 4021 BUG();
a4456856
DW		 4022 }
4023}
4024
d1688a6d 4025static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh)
a4456856
DW		 4026{
4027 int i;
4028
4029 /* We have read all the blocks in this stripe and now we need to
4030 * copy some of them into a target stripe for expand.
4031 */
f0a50d37 4032 struct dma_async_tx_descriptor *tx = NULL;
59fc630b 4033 BUG_ON(sh->batch_head);
a4456856
DW		 4034 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
4035 for (i = 0; i < sh->disks; i++)
34e04e87 4036 if (i != sh->pd_idx && i != sh->qd_idx) {
911d4ee8 4037 int dd_idx, j;
a4456856 4038 struct stripe_head *sh2;
a08abd8c 4039 struct async_submit_ctl submit;
a4456856 4040
6d036f7d 4041 sector_t bn = raid5_compute_blocknr(sh, i, 1);
911d4ee8
N		 4042 sector_t s = raid5_compute_sector(conf, bn, 0,
4043 &dd_idx, NULL);
6d036f7d 4044 sh2 = raid5_get_active_stripe(conf, s, 0, 1, 1);
a4456856
DW		 4045 if (sh2 == NULL)
4046 /* so far only the early blocks of this stripe
4047 * have been requested. When later blocks
4048 * get requested, we will try again
4049 */
4050 continue;
4051 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
4052 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
4053 /* must have already done this block */
6d036f7d 4054 raid5_release_stripe(sh2);
a4456856
DW		 4055 continue;
4056 }
f0a50d37
DW		 4057
4058 /* place all the copies on one channel */
a08abd8c 4059 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
f0a50d37 4060 tx = async_memcpy(sh2->dev[dd_idx].page,
88ba2aa5 4061 sh->dev[i].page, 0, 0, STRIPE_SIZE,
a08abd8c 4062 &submit);
f0a50d37 4063
a4456856
DW		 4064 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
4065 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
4066 for (j = 0; j < conf->raid_disks; j++)
4067 if (j != sh2->pd_idx &&
86c374ba 4068 j != sh2->qd_idx &&
a4456856
DW		 4069 !test_bit(R5_Expanded, &sh2->dev[j].flags))
4070 break;
4071 if (j == conf->raid_disks) {
4072 set_bit(STRIPE_EXPAND_READY, &sh2->state);
4073 set_bit(STRIPE_HANDLE, &sh2->state);
4074 }
6d036f7d 4075 raid5_release_stripe(sh2);
f0a50d37 4076
a4456856 4077 }
a2e08551 4078 /* done submitting copies, wait for them to complete */
749586b7 4079 async_tx_quiesce(&tx);
a4456856 4080}
1da177e4
LT		 4081
4082/*
4083 * handle_stripe - do things to a stripe.
4084 *
9a3e1101
N		 4085 * We lock the stripe by setting STRIPE_ACTIVE and then examine the
4086 * state of various bits to see what needs to be done.
1da177e4 4087 * Possible results:
9a3e1101
N		 4088 * return some read requests which now have data
4089 * return some write requests which are safely on storage
1da177e4
LT		 4090 * schedule a read on some buffers
4091 * schedule a write of some buffers
4092 * return confirmation of parity correctness
4093 *
1da177e4 4094 */
a4456856 4095
acfe726b 4096static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s)
1da177e4 4097{
d1688a6d 4098 struct r5conf *conf = sh->raid_conf;
f416885e 4099 int disks = sh->disks;
474af965
N		 4100 struct r5dev *dev;
4101 int i;
9a3e1101 4102 int do_recovery = 0;
1da177e4 4103
acfe726b
N		 4104 memset(s, 0, sizeof(*s));
4105
dabc4ec6 4106 s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state) && !sh->batch_head;
4107 s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state) && !sh->batch_head;
acfe726b
N		 4108 s->failed_num[0] = -1;
4109 s->failed_num[1] = -1;
6e74a9cf 4110 s->log_failed = r5l_log_disk_error(conf);
1da177e4 4111
acfe726b 4112 /* Now to look around and see what can be done */
1da177e4 4113 rcu_read_lock();
16a53ecc 4114 for (i=disks; i--; ) {
3cb03002 4115 struct md_rdev *rdev;
31c176ec
N		 4116 sector_t first_bad;
4117 int bad_sectors;
4118 int is_bad = 0;
acfe726b 4119
16a53ecc 4120 dev = &sh->dev[i];
1da177e4 4121
45b4233c 4122 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
9a3e1101
N		 4123 i, dev->flags,
4124 dev->toread, dev->towrite, dev->written);
6c0069c0
YT		 4125 /* maybe we can reply to a read
4126 *
4127 * new wantfill requests are only permitted while
4128 * ops_complete_biofill is guaranteed to be inactive
4129 */
4130 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
4131 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
4132 set_bit(R5_Wantfill, &dev->flags);
1da177e4 4133
16a53ecc 4134 /* now count some things */
cc94015a
N		 4135 if (test_bit(R5_LOCKED, &dev->flags))
4136 s->locked++;
4137 if (test_bit(R5_UPTODATE, &dev->flags))
4138 s->uptodate++;
2d6e4ecc 4139 if (test_bit(R5_Wantcompute, &dev->flags)) {
cc94015a
N		 4140 s->compute++;
4141 BUG_ON(s->compute > 2);
2d6e4ecc 4142 }
1da177e4 4143
acfe726b 4144 if (test_bit(R5_Wantfill, &dev->flags))
cc94015a 4145 s->to_fill++;
acfe726b 4146 else if (dev->toread)
cc94015a 4147 s->to_read++;
16a53ecc 4148 if (dev->towrite) {
cc94015a 4149 s->to_write++;
16a53ecc 4150 if (!test_bit(R5_OVERWRITE, &dev->flags))
cc94015a 4151 s->non_overwrite++;
16a53ecc 4152 }
a4456856 4153 if (dev->written)
cc94015a 4154 s->written++;
14a75d3e
N		 4155 /* Prefer to use the replacement for reads, but only
4156 * if it is recovered enough and has no bad blocks.
4157 */
4158 rdev = rcu_dereference(conf->disks[i].replacement);
4159 if (rdev && !test_bit(Faulty, &rdev->flags) &&
4160 rdev->recovery_offset >= sh->sector + STRIPE_SECTORS &&
4161 !is_badblock(rdev, sh->sector, STRIPE_SECTORS,
4162 &first_bad, &bad_sectors))
4163 set_bit(R5_ReadRepl, &dev->flags);
4164 else {
e6030cb0 4165 if (rdev && !test_bit(Faulty, &rdev->flags))
9a3e1101 4166 set_bit(R5_NeedReplace, &dev->flags);
e6030cb0
N		 4167 else
4168 clear_bit(R5_NeedReplace, &dev->flags);
14a75d3e
N		 4169 rdev = rcu_dereference(conf->disks[i].rdev);
4170 clear_bit(R5_ReadRepl, &dev->flags);
4171 }
9283d8c5
N		 4172 if (rdev && test_bit(Faulty, &rdev->flags))
4173 rdev = NULL;
31c176ec
N		 4174 if (rdev) {
4175 is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
4176 &first_bad, &bad_sectors);
4177 if (s->blocked_rdev == NULL
4178 && (test_bit(Blocked, &rdev->flags)
4179 || is_bad < 0)) {
4180 if (is_bad < 0)
4181 set_bit(BlockedBadBlocks,
4182 &rdev->flags);
4183 s->blocked_rdev = rdev;
4184 atomic_inc(&rdev->nr_pending);
4185 }
6bfe0b49 4186 }
415e72d0
N		 4187 clear_bit(R5_Insync, &dev->flags);
4188 if (!rdev)
4189 /* Not in-sync */;
31c176ec
N		 4190 else if (is_bad) {
4191 /* also not in-sync */
18b9837e
N		 4192 if (!test_bit(WriteErrorSeen, &rdev->flags) &&
4193 test_bit(R5_UPTODATE, &dev->flags)) {
31c176ec
N		 4194 /* treat as in-sync, but with a read error
4195 * which we can now try to correct
4196 */
4197 set_bit(R5_Insync, &dev->flags);
4198 set_bit(R5_ReadError, &dev->flags);
4199 }
4200 } else if (test_bit(In_sync, &rdev->flags))
415e72d0 4201 set_bit(R5_Insync, &dev->flags);
30d7a483 4202 else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
415e72d0 4203 /* in sync if before recovery_offset */
30d7a483
N		 4204 set_bit(R5_Insync, &dev->flags);
4205 else if (test_bit(R5_UPTODATE, &dev->flags) &&
4206 test_bit(R5_Expanded, &dev->flags))
4207 /* If we've reshaped into here, we assume it is Insync.
4208 * We will shortly update recovery_offset to make
4209 * it official.
4210 */
4211 set_bit(R5_Insync, &dev->flags);
4212
1cc03eb9 4213 if (test_bit(R5_WriteError, &dev->flags)) {
14a75d3e
N		 4214 /* This flag does not apply to '.replacement'
4215 * only to .rdev, so make sure to check that*/
4216 struct md_rdev *rdev2 = rcu_dereference(
4217 conf->disks[i].rdev);
4218 if (rdev2 == rdev)
4219 clear_bit(R5_Insync, &dev->flags);
4220 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
bc2607f3 4221 s->handle_bad_blocks = 1;
14a75d3e 4222 atomic_inc(&rdev2->nr_pending);
bc2607f3
N		 4223 } else
4224 clear_bit(R5_WriteError, &dev->flags);
4225 }
1cc03eb9 4226 if (test_bit(R5_MadeGood, &dev->flags)) {
14a75d3e
N		 4227 /* This flag does not apply to '.replacement'
4228 * only to .rdev, so make sure to check that*/
4229 struct md_rdev *rdev2 = rcu_dereference(
4230 conf->disks[i].rdev);
4231 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
b84db560 4232 s->handle_bad_blocks = 1;
14a75d3e 4233 atomic_inc(&rdev2->nr_pending);
b84db560
N		 4234 } else
4235 clear_bit(R5_MadeGood, &dev->flags);
4236 }
977df362
N		 4237 if (test_bit(R5_MadeGoodRepl, &dev->flags)) {
4238 struct md_rdev *rdev2 = rcu_dereference(
4239 conf->disks[i].replacement);
4240 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
4241 s->handle_bad_blocks = 1;
4242 atomic_inc(&rdev2->nr_pending);
4243 } else
4244 clear_bit(R5_MadeGoodRepl, &dev->flags);
4245 }
415e72d0 4246 if (!test_bit(R5_Insync, &dev->flags)) {
16a53ecc
N		 4247 /* The ReadError flag will just be confusing now */
4248 clear_bit(R5_ReadError, &dev->flags);
4249 clear_bit(R5_ReWrite, &dev->flags);
1da177e4 4250 }
415e72d0
N		 4251 if (test_bit(R5_ReadError, &dev->flags))
4252 clear_bit(R5_Insync, &dev->flags);
4253 if (!test_bit(R5_Insync, &dev->flags)) {
cc94015a
N		 4254 if (s->failed < 2)
4255 s->failed_num[s->failed] = i;
4256 s->failed++;
9a3e1101
N		 4257 if (rdev && !test_bit(Faulty, &rdev->flags))
4258 do_recovery = 1;
415e72d0 4259 }
2ded3703
SL		 4260
4261 if (test_bit(R5_InJournal, &dev->flags))
4262 s->injournal++;
1e6d690b
SL		 4263 if (test_bit(R5_InJournal, &dev->flags) && dev->written)
4264 s->just_cached++;
1da177e4 4265 }
9a3e1101
N		 4266 if (test_bit(STRIPE_SYNCING, &sh->state)) {
4267 /* If there is a failed device being replaced,
4268 * we must be recovering.
4269 * else if we are after recovery_cp, we must be syncing
c6d2e084 4270 * else if MD_RECOVERY_REQUESTED is set, we also are syncing.
9a3e1101
N		 4271 * else we can only be replacing
4272 * sync and recovery both need to read all devices, and so
4273 * use the same flag.
4274 */
4275 if (do_recovery ||
c6d2e084 4276 sh->sector >= conf->mddev->recovery_cp ||
4277 test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery)))
9a3e1101
N		 4278 s->syncing = 1;
4279 else
4280 s->replacing = 1;
4281 }
1da177e4 4282 rcu_read_unlock();
cc94015a
N		 4283}
4284
59fc630b 4285static int clear_batch_ready(struct stripe_head *sh)
4286{
b15a9dbd
N		 4287 /* Return '1' if this is a member of batch, or
4288 * '0' if it is a lone stripe or a head which can now be
4289 * handled.
4290 */
59fc630b 4291 struct stripe_head *tmp;
4292 if (!test_and_clear_bit(STRIPE_BATCH_READY, &sh->state))
b15a9dbd 4293 return (sh->batch_head && sh->batch_head != sh);
59fc630b 4294 spin_lock(&sh->stripe_lock);
4295 if (!sh->batch_head) {
4296 spin_unlock(&sh->stripe_lock);
4297 return 0;
4298 }
4299
4300 /*
4301 * this stripe could be added to a batch list before we check
4302 * BATCH_READY, skips it
4303 */
4304 if (sh->batch_head != sh) {
4305 spin_unlock(&sh->stripe_lock);
4306 return 1;
4307 }
4308 spin_lock(&sh->batch_lock);
4309 list_for_each_entry(tmp, &sh->batch_list, batch_list)
4310 clear_bit(STRIPE_BATCH_READY, &tmp->state);
4311 spin_unlock(&sh->batch_lock);
4312 spin_unlock(&sh->stripe_lock);
4313
4314 /*
4315 * BATCH_READY is cleared, no new stripes can be added.
4316 * batch_list can be accessed without lock
4317 */
4318 return 0;
4319}
4320
3960ce79
N		 4321static void break_stripe_batch_list(struct stripe_head *head_sh,
4322 unsigned long handle_flags)
72ac7330 4323{
4e3d62ff 4324 struct stripe_head *sh, *next;
72ac7330 4325 int i;
fb642b92 4326 int do_wakeup = 0;
72ac7330 4327
bb27051f
N		 4328 list_for_each_entry_safe(sh, next, &head_sh->batch_list, batch_list) {
4329
72ac7330 4330 list_del_init(&sh->batch_list);
4331
fb3229d5 4332 WARN_ONCE(sh->state & ((1 << STRIPE_ACTIVE) |
1b956f7a
N		 4333 (1 << STRIPE_SYNCING) |
4334 (1 << STRIPE_REPLACED) |
1b956f7a
N		 4335 (1 << STRIPE_DELAYED) |
4336 (1 << STRIPE_BIT_DELAY) |
4337 (1 << STRIPE_FULL_WRITE) |
4338 (1 << STRIPE_BIOFILL_RUN) |
4339 (1 << STRIPE_COMPUTE_RUN) |
4340 (1 << STRIPE_OPS_REQ_PENDING) |
4341 (1 << STRIPE_DISCARD) |
4342 (1 << STRIPE_BATCH_READY) |
4343 (1 << STRIPE_BATCH_ERR) |
fb3229d5
SL		 4344 (1 << STRIPE_BITMAP_PENDING)),
4345 "stripe state: %lx\n", sh->state);
4346 WARN_ONCE(head_sh->state & ((1 << STRIPE_DISCARD) |
4347 (1 << STRIPE_REPLACED)),
4348 "head stripe state: %lx\n", head_sh->state);
1b956f7a
N		 4349
4350 set_mask_bits(&sh->state, ~(STRIPE_EXPAND_SYNC_FLAGS |
550da24f 4351 (1 << STRIPE_PREREAD_ACTIVE) |
1b956f7a
N		 4352 (1 << STRIPE_DEGRADED)),
4353 head_sh->state & (1 << STRIPE_INSYNC));
4354
72ac7330 4355 sh->check_state = head_sh->check_state;
4356 sh->reconstruct_state = head_sh->reconstruct_state;
fb642b92
N		 4357 for (i = 0; i < sh->disks; i++) {
4358 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
4359 do_wakeup = 1;
72ac7330 4360 sh->dev[i].flags = head_sh->dev[i].flags &
4361 (~((1 << R5_WriteError) | (1 << R5_Overlap)));
fb642b92 4362 }
72ac7330 4363 spin_lock_irq(&sh->stripe_lock);
4364 sh->batch_head = NULL;
4365 spin_unlock_irq(&sh->stripe_lock);
3960ce79
N		 4366 if (handle_flags == 0 ||
4367 sh->state & handle_flags)
4368 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 4369 raid5_release_stripe(sh);
72ac7330 4370 }
fb642b92
N		 4371 spin_lock_irq(&head_sh->stripe_lock);
4372 head_sh->batch_head = NULL;
4373 spin_unlock_irq(&head_sh->stripe_lock);
4374 for (i = 0; i < head_sh->disks; i++)
4375 if (test_and_clear_bit(R5_Overlap, &head_sh->dev[i].flags))
4376 do_wakeup = 1;
3960ce79
N		 4377 if (head_sh->state & handle_flags)
4378 set_bit(STRIPE_HANDLE, &head_sh->state);
fb642b92
N		 4379
4380 if (do_wakeup)
4381 wake_up(&head_sh->raid_conf->wait_for_overlap);
72ac7330 4382}
4383
cc94015a
N		 4384static void handle_stripe(struct stripe_head *sh)
4385{
4386 struct stripe_head_state s;
d1688a6d 4387 struct r5conf *conf = sh->raid_conf;
3687c061 4388 int i;
84789554
N		 4389 int prexor;
4390 int disks = sh->disks;
474af965 4391 struct r5dev *pdev, *qdev;
cc94015a
N		 4392
4393 clear_bit(STRIPE_HANDLE, &sh->state);
257a4b42 4394 if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) {
cc94015a
N		 4395 /* already being handled, ensure it gets handled
4396 * again when current action finishes */
4397 set_bit(STRIPE_HANDLE, &sh->state);
4398 return;
4399 }
4400
59fc630b 4401 if (clear_batch_ready(sh) ) {
4402 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
4403 return;
4404 }
4405
4e3d62ff 4406 if (test_and_clear_bit(STRIPE_BATCH_ERR, &sh->state))
3960ce79 4407 break_stripe_batch_list(sh, 0);
72ac7330 4408
dabc4ec6 4409 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) && !sh->batch_head) {
f8dfcffd
N		 4410 spin_lock(&sh->stripe_lock);
4411 /* Cannot process 'sync' concurrently with 'discard' */
4412 if (!test_bit(STRIPE_DISCARD, &sh->state) &&
4413 test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
4414 set_bit(STRIPE_SYNCING, &sh->state);
4415 clear_bit(STRIPE_INSYNC, &sh->state);
f94c0b66 4416 clear_bit(STRIPE_REPLACED, &sh->state);
f8dfcffd
N		 4417 }
4418 spin_unlock(&sh->stripe_lock);
cc94015a
N		 4419 }
4420 clear_bit(STRIPE_DELAYED, &sh->state);
4421
4422 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
4423 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
4424 (unsigned long long)sh->sector, sh->state,
4425 atomic_read(&sh->count), sh->pd_idx, sh->qd_idx,
4426 sh->check_state, sh->reconstruct_state);
3687c061 4427
acfe726b 4428 analyse_stripe(sh, &s);
c5a31000 4429
b70abcb2
SL		 4430 if (test_bit(STRIPE_LOG_TRAPPED, &sh->state))
4431 goto finish;
4432
bc2607f3
N		 4433 if (s.handle_bad_blocks) {
4434 set_bit(STRIPE_HANDLE, &sh->state);
4435 goto finish;
4436 }
4437
474af965
N		 4438 if (unlikely(s.blocked_rdev)) {
4439 if (s.syncing || s.expanding || s.expanded ||
9a3e1101 4440 s.replacing || s.to_write || s.written) {
474af965
N		 4441 set_bit(STRIPE_HANDLE, &sh->state);
4442 goto finish;
4443 }
4444 /* There is nothing for the blocked_rdev to block */
4445 rdev_dec_pending(s.blocked_rdev, conf->mddev);
4446 s.blocked_rdev = NULL;
4447 }
4448
4449 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
4450 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
4451 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
4452 }
4453
4454 pr_debug("locked=%d uptodate=%d to_read=%d"
4455 " to_write=%d failed=%d failed_num=%d,%d\n",
4456 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
4457 s.failed_num[0], s.failed_num[1]);
4458 /* check if the array has lost more than max_degraded devices and,
4459 * if so, some requests might need to be failed.
4460 */
6e74a9cf 4461 if (s.failed > conf->max_degraded || s.log_failed) {
9a3f530f
N		 4462 sh->check_state = 0;
4463 sh->reconstruct_state = 0;
626f2092 4464 break_stripe_batch_list(sh, 0);
9a3f530f
N		 4465 if (s.to_read+s.to_write+s.written)
4466 handle_failed_stripe(conf, sh, &s, disks, &s.return_bi);
9a3e1101 4467 if (s.syncing + s.replacing)
9a3f530f
N		 4468 handle_failed_sync(conf, sh, &s);
4469 }
474af965 4470
84789554
N		 4471 /* Now we check to see if any write operations have recently
4472 * completed
4473 */
4474 prexor = 0;
4475 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
4476 prexor = 1;
4477 if (sh->reconstruct_state == reconstruct_state_drain_result ||
4478 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
4479 sh->reconstruct_state = reconstruct_state_idle;
4480
4481 /* All the 'written' buffers and the parity block are ready to
4482 * be written back to disk
4483 */
9e444768
SL		 4484 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) &&
4485 !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags));
84789554 4486 BUG_ON(sh->qd_idx >= 0 &&
9e444768
SL		 4487 !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) &&
4488 !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags));
84789554
N		 4489 for (i = disks; i--; ) {
4490 struct r5dev *dev = &sh->dev[i];
4491 if (test_bit(R5_LOCKED, &dev->flags) &&
4492 (i == sh->pd_idx || i == sh->qd_idx ||
1e6d690b
SL		 4493 dev->written || test_bit(R5_InJournal,
4494 &dev->flags))) {
84789554
N		 4495 pr_debug("Writing block %d\n", i);
4496 set_bit(R5_Wantwrite, &dev->flags);
4497 if (prexor)
4498 continue;
9c4bdf69
N		 4499 if (s.failed > 1)
4500 continue;
84789554
N		 4501 if (!test_bit(R5_Insync, &dev->flags) ||
4502 ((i == sh->pd_idx || i == sh->qd_idx) &&
4503 s.failed == 0))
4504 set_bit(STRIPE_INSYNC, &sh->state);
4505 }
4506 }
4507 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4508 s.dec_preread_active = 1;
4509 }
4510
ef5b7c69
N		 4511 /*
4512 * might be able to return some write requests if the parity blocks
4513 * are safe, or on a failed drive
4514 */
4515 pdev = &sh->dev[sh->pd_idx];
4516 s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx)
4517 || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx);
4518 qdev = &sh->dev[sh->qd_idx];
4519 s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx)
4520 || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx)
4521 || conf->level < 6;
4522
4523 if (s.written &&
4524 (s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
4525 && !test_bit(R5_LOCKED, &pdev->flags)
4526 && (test_bit(R5_UPTODATE, &pdev->flags) ||
4527 test_bit(R5_Discard, &pdev->flags))))) &&
4528 (s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
4529 && !test_bit(R5_LOCKED, &qdev->flags)
4530 && (test_bit(R5_UPTODATE, &qdev->flags) ||
4531 test_bit(R5_Discard, &qdev->flags))))))
4532 handle_stripe_clean_event(conf, sh, disks, &s.return_bi);
4533
1e6d690b
SL		 4534 if (s.just_cached)
4535 r5c_handle_cached_data_endio(conf, sh, disks, &s.return_bi);
4536 r5l_stripe_write_finished(sh);
4537
ef5b7c69
N		 4538 /* Now we might consider reading some blocks, either to check/generate
4539 * parity, or to satisfy requests
4540 * or to load a block that is being partially written.
4541 */
4542 if (s.to_read || s.non_overwrite
4543 || (conf->level == 6 && s.to_write && s.failed)
4544 || (s.syncing && (s.uptodate + s.compute < disks))
4545 || s.replacing
4546 || s.expanding)
4547 handle_stripe_fill(sh, &s, disks);
4548
2ded3703
SL		 4549 /*
4550 * When the stripe finishes full journal write cycle (write to journal
4551 * and raid disk), this is the clean up procedure so it is ready for
4552 * next operation.
4553 */
4554 r5c_finish_stripe_write_out(conf, sh, &s);
4555
4556 /*
4557 * Now to consider new write requests, cache write back and what else,
4558 * if anything should be read. We do not handle new writes when:
84789554
N		 4559 * 1/ A 'write' operation (copy+xor) is already in flight.
4560 * 2/ A 'check' operation is in flight, as it may clobber the parity
4561 * block.
2ded3703 4562 * 3/ A r5c cache log write is in flight.
84789554 4563 */
2ded3703
SL		 4564
4565 if (!sh->reconstruct_state && !sh->check_state && !sh->log_io) {
4566 if (!r5c_is_writeback(conf->log)) {
4567 if (s.to_write)
4568 handle_stripe_dirtying(conf, sh, &s, disks);
4569 } else { /* write back cache */
4570 int ret = 0;
4571
4572 /* First, try handle writes in caching phase */
4573 if (s.to_write)
4574 ret = r5c_try_caching_write(conf, sh, &s,
4575 disks);
4576 /*
4577 * If caching phase failed: ret == -EAGAIN
4578 * OR
4579 * stripe under reclaim: !caching && injournal
4580 *
4581 * fall back to handle_stripe_dirtying()
4582 */
4583 if (ret == -EAGAIN ||
4584 /* stripe under reclaim: !caching && injournal */
4585 (!test_bit(STRIPE_R5C_CACHING, &sh->state) &&
d7bd398e
SL		 4586 s.injournal > 0)) {
4587 ret = handle_stripe_dirtying(conf, sh, &s,
4588 disks);
4589 if (ret == -EAGAIN)
4590 goto finish;
4591 }
2ded3703
SL		 4592 }
4593 }
84789554
N		 4594
4595 /* maybe we need to check and possibly fix the parity for this stripe
4596 * Any reads will already have been scheduled, so we just see if enough
4597 * data is available. The parity check is held off while parity
4598 * dependent operations are in flight.
4599 */
4600 if (sh->check_state ||
4601 (s.syncing && s.locked == 0 &&
4602 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
4603 !test_bit(STRIPE_INSYNC, &sh->state))) {
4604 if (conf->level == 6)
4605 handle_parity_checks6(conf, sh, &s, disks);
4606 else
4607 handle_parity_checks5(conf, sh, &s, disks);
4608 }
c5a31000 4609
f94c0b66
N		 4610 if ((s.replacing || s.syncing) && s.locked == 0
4611 && !test_bit(STRIPE_COMPUTE_RUN, &sh->state)
4612 && !test_bit(STRIPE_REPLACED, &sh->state)) {
9a3e1101
N		 4613 /* Write out to replacement devices where possible */
4614 for (i = 0; i < conf->raid_disks; i++)
f94c0b66
N		 4615 if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) {
4616 WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags));
9a3e1101
N		 4617 set_bit(R5_WantReplace, &sh->dev[i].flags);
4618 set_bit(R5_LOCKED, &sh->dev[i].flags);
4619 s.locked++;
4620 }
f94c0b66
N		 4621 if (s.replacing)
4622 set_bit(STRIPE_INSYNC, &sh->state);
4623 set_bit(STRIPE_REPLACED, &sh->state);
9a3e1101
N		 4624 }
4625 if ((s.syncing || s.replacing) && s.locked == 0 &&
f94c0b66 4626 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
9a3e1101 4627 test_bit(STRIPE_INSYNC, &sh->state)) {
c5a31000
N		 4628 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
4629 clear_bit(STRIPE_SYNCING, &sh->state);
f8dfcffd
N		 4630 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
4631 wake_up(&conf->wait_for_overlap);
c5a31000
N		 4632 }
4633
4634 /* If the failed drives are just a ReadError, then we might need
4635 * to progress the repair/check process
4636 */
4637 if (s.failed <= conf->max_degraded && !conf->mddev->ro)
4638 for (i = 0; i < s.failed; i++) {
4639 struct r5dev *dev = &sh->dev[s.failed_num[i]];
4640 if (test_bit(R5_ReadError, &dev->flags)
4641 && !test_bit(R5_LOCKED, &dev->flags)
4642 && test_bit(R5_UPTODATE, &dev->flags)
4643 ) {
4644 if (!test_bit(R5_ReWrite, &dev->flags)) {
4645 set_bit(R5_Wantwrite, &dev->flags);
4646 set_bit(R5_ReWrite, &dev->flags);
4647 set_bit(R5_LOCKED, &dev->flags);
4648 s.locked++;
4649 } else {
4650 /* let's read it back */
4651 set_bit(R5_Wantread, &dev->flags);
4652 set_bit(R5_LOCKED, &dev->flags);
4653 s.locked++;
4654 }
4655 }
4656 }
4657
3687c061
N		 4658 /* Finish reconstruct operations initiated by the expansion process */
4659 if (sh->reconstruct_state == reconstruct_state_result) {
4660 struct stripe_head *sh_src
6d036f7d 4661 = raid5_get_active_stripe(conf, sh->sector, 1, 1, 1);
3687c061
N		 4662 if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) {
4663 /* sh cannot be written until sh_src has been read.
4664 * so arrange for sh to be delayed a little
4665 */
4666 set_bit(STRIPE_DELAYED, &sh->state);
4667 set_bit(STRIPE_HANDLE, &sh->state);
4668 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
4669 &sh_src->state))
4670 atomic_inc(&conf->preread_active_stripes);
6d036f7d 4671 raid5_release_stripe(sh_src);
3687c061
N		 4672 goto finish;
4673 }
4674 if (sh_src)
6d036f7d 4675 raid5_release_stripe(sh_src);
3687c061
N		 4676
4677 sh->reconstruct_state = reconstruct_state_idle;
4678 clear_bit(STRIPE_EXPANDING, &sh->state);
4679 for (i = conf->raid_disks; i--; ) {
4680 set_bit(R5_Wantwrite, &sh->dev[i].flags);
4681 set_bit(R5_LOCKED, &sh->dev[i].flags);
4682 s.locked++;
4683 }
4684 }
f416885e 4685
3687c061
N		 4686 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
4687 !sh->reconstruct_state) {
4688 /* Need to write out all blocks after computing parity */
4689 sh->disks = conf->raid_disks;
4690 stripe_set_idx(sh->sector, conf, 0, sh);
4691 schedule_reconstruction(sh, &s, 1, 1);
4692 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
4693 clear_bit(STRIPE_EXPAND_READY, &sh->state);
4694 atomic_dec(&conf->reshape_stripes);
4695 wake_up(&conf->wait_for_overlap);
4696 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
4697 }
4698
4699 if (s.expanding && s.locked == 0 &&
4700 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
4701 handle_stripe_expansion(conf, sh);
16a53ecc 4702
3687c061 4703finish:
6bfe0b49 4704 /* wait for this device to become unblocked */
5f066c63
N		 4705 if (unlikely(s.blocked_rdev)) {
4706 if (conf->mddev->external)
4707 md_wait_for_blocked_rdev(s.blocked_rdev,
4708 conf->mddev);
4709 else
4710 /* Internal metadata will immediately
4711 * be written by raid5d, so we don't
4712 * need to wait here.
4713 */
4714 rdev_dec_pending(s.blocked_rdev,
4715 conf->mddev);
4716 }
6bfe0b49 4717
bc2607f3
N		 4718 if (s.handle_bad_blocks)
4719 for (i = disks; i--; ) {
3cb03002 4720 struct md_rdev *rdev;
bc2607f3
N		 4721 struct r5dev *dev = &sh->dev[i];
4722 if (test_and_clear_bit(R5_WriteError, &dev->flags)) {
4723 /* We own a safe reference to the rdev */
4724 rdev = conf->disks[i].rdev;
4725 if (!rdev_set_badblocks(rdev, sh->sector,
4726 STRIPE_SECTORS, 0))
4727 md_error(conf->mddev, rdev);
4728 rdev_dec_pending(rdev, conf->mddev);
4729 }
b84db560
N		 4730 if (test_and_clear_bit(R5_MadeGood, &dev->flags)) {
4731 rdev = conf->disks[i].rdev;
4732 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 4733 STRIPE_SECTORS, 0);
b84db560
N		 4734 rdev_dec_pending(rdev, conf->mddev);
4735 }
977df362
N		 4736 if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) {
4737 rdev = conf->disks[i].replacement;
dd054fce
N		 4738 if (!rdev)
4739 /* rdev have been moved down */
4740 rdev = conf->disks[i].rdev;
977df362 4741 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 4742 STRIPE_SECTORS, 0);
977df362
N		 4743 rdev_dec_pending(rdev, conf->mddev);
4744 }
bc2607f3
N		 4745 }
4746
6c0069c0
YT		 4747 if (s.ops_request)
4748 raid_run_ops(sh, s.ops_request);
4749
f0e43bcd 4750 ops_run_io(sh, &s);
16a53ecc 4751
c5709ef6 4752 if (s.dec_preread_active) {
729a1866 4753 /* We delay this until after ops_run_io so that if make_request
e9c7469b 4754 * is waiting on a flush, it won't continue until the writes
729a1866
N		 4755 * have actually been submitted.
4756 */
4757 atomic_dec(&conf->preread_active_stripes);
4758 if (atomic_read(&conf->preread_active_stripes) <
4759 IO_THRESHOLD)
4760 md_wakeup_thread(conf->mddev->thread);
4761 }
4762
c3cce6cd 4763 if (!bio_list_empty(&s.return_bi)) {
7adb072c 4764 if (test_bit(MD_CHANGE_PENDING, &conf->mddev->flags)) {
c3cce6cd
N		 4765 spin_lock_irq(&conf->device_lock);
4766 bio_list_merge(&conf->return_bi, &s.return_bi);
4767 spin_unlock_irq(&conf->device_lock);
4768 md_wakeup_thread(conf->mddev->thread);
4769 } else
4770 return_io(&s.return_bi);
4771 }
16a53ecc 4772
257a4b42 4773 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
16a53ecc
N		 4774}
4775
d1688a6d 4776static void raid5_activate_delayed(struct r5conf *conf)
16a53ecc
N		 4777{
4778 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
4779 while (!list_empty(&conf->delayed_list)) {
4780 struct list_head *l = conf->delayed_list.next;
4781 struct stripe_head *sh;
4782 sh = list_entry(l, struct stripe_head, lru);
4783 list_del_init(l);
4784 clear_bit(STRIPE_DELAYED, &sh->state);
4785 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4786 atomic_inc(&conf->preread_active_stripes);
8b3e6cdc 4787 list_add_tail(&sh->lru, &conf->hold_list);
851c30c9 4788 raid5_wakeup_stripe_thread(sh);
16a53ecc 4789 }
482c0834 4790 }
16a53ecc
N		 4791}
4792
566c09c5
SL		 4793static void activate_bit_delay(struct r5conf *conf,
4794 struct list_head *temp_inactive_list)
16a53ecc
N		 4795{
4796 /* device_lock is held */
4797 struct list_head head;
4798 list_add(&head, &conf->bitmap_list);
4799 list_del_init(&conf->bitmap_list);
4800 while (!list_empty(&head)) {
4801 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
566c09c5 4802 int hash;
16a53ecc
N		 4803 list_del_init(&sh->lru);
4804 atomic_inc(&sh->count);
566c09c5
SL		 4805 hash = sh->hash_lock_index;
4806 __release_stripe(conf, sh, &temp_inactive_list[hash]);
16a53ecc
N		 4807 }
4808}
4809
5c675f83 4810static int raid5_congested(struct mddev *mddev, int bits)
f022b2fd 4811{
d1688a6d 4812 struct r5conf *conf = mddev->private;
f022b2fd
N		 4813
4814 /* No difference between reads and writes. Just check
4815 * how busy the stripe_cache is
4816 */
3fa841d7 4817
5423399a 4818 if (test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state))
f022b2fd 4819 return 1;
a39f7afd
SL		 4820
4821 /* Also checks whether there is pressure on r5cache log space */
4822 if (test_bit(R5C_LOG_TIGHT, &conf->cache_state))
4823 return 1;
f022b2fd
N		 4824 if (conf->quiesce)
4825 return 1;
4bda556a 4826 if (atomic_read(&conf->empty_inactive_list_nr))
f022b2fd
N		 4827 return 1;
4828
4829 return 0;
4830}
4831
fd01b88c 4832static int in_chunk_boundary(struct mddev *mddev, struct bio *bio)
f679623f 4833{
3cb5edf4 4834 struct r5conf *conf = mddev->private;
4f024f37 4835 sector_t sector = bio->bi_iter.bi_sector + get_start_sect(bio->bi_bdev);
3cb5edf4 4836 unsigned int chunk_sectors;
aa8b57aa 4837 unsigned int bio_sectors = bio_sectors(bio);
f679623f 4838
3cb5edf4 4839 chunk_sectors = min(conf->chunk_sectors, conf->prev_chunk_sectors);
f679623f
RBJ		 4840 return chunk_sectors >=
4841 ((sector & (chunk_sectors - 1)) + bio_sectors);
4842}
4843
46031f9a
RBJ		 4844/*
4845 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
4846 * later sampled by raid5d.
4847 */
d1688a6d 4848static void add_bio_to_retry(struct bio *bi,struct r5conf *conf)
46031f9a
RBJ		 4849{
4850 unsigned long flags;
4851
4852 spin_lock_irqsave(&conf->device_lock, flags);
4853
4854 bi->bi_next = conf->retry_read_aligned_list;
4855 conf->retry_read_aligned_list = bi;
4856
4857 spin_unlock_irqrestore(&conf->device_lock, flags);
4858 md_wakeup_thread(conf->mddev->thread);
4859}
4860
d1688a6d 4861static struct bio *remove_bio_from_retry(struct r5conf *conf)
46031f9a
RBJ		 4862{
4863 struct bio *bi;
4864
4865 bi = conf->retry_read_aligned;
4866 if (bi) {
4867 conf->retry_read_aligned = NULL;
4868 return bi;
4869 }
4870 bi = conf->retry_read_aligned_list;
4871 if(bi) {
387bb173 4872 conf->retry_read_aligned_list = bi->bi_next;
46031f9a 4873 bi->bi_next = NULL;
960e739d
JA		 4874 /*
4875 * this sets the active strip count to 1 and the processed
4876 * strip count to zero (upper 8 bits)
4877 */
e7836bd6 4878 raid5_set_bi_stripes(bi, 1); /* biased count of active stripes */
46031f9a
RBJ		 4879 }
4880
4881 return bi;
4882}
4883
f679623f
RBJ		 4884/*
4885 * The "raid5_align_endio" should check if the read succeeded and if it
4886 * did, call bio_endio on the original bio (having bio_put the new bio
4887 * first).
4888 * If the read failed..
4889 */
4246a0b6 4890static void raid5_align_endio(struct bio *bi)
f679623f
RBJ		 4891{
4892 struct bio* raid_bi = bi->bi_private;
fd01b88c 4893 struct mddev *mddev;
d1688a6d 4894 struct r5conf *conf;
3cb03002 4895 struct md_rdev *rdev;
9b81c842 4896 int error = bi->bi_error;
46031f9a 4897
f679623f 4898 bio_put(bi);
46031f9a 4899
46031f9a
RBJ		 4900 rdev = (void*)raid_bi->bi_next;
4901 raid_bi->bi_next = NULL;
2b7f2228
N		 4902 mddev = rdev->mddev;
4903 conf = mddev->private;
46031f9a
RBJ		 4904
4905 rdev_dec_pending(rdev, conf->mddev);
4906
9b81c842 4907 if (!error) {
0a82a8d1
LT		 4908 trace_block_bio_complete(bdev_get_queue(raid_bi->bi_bdev),
4909 raid_bi, 0);
4246a0b6 4910 bio_endio(raid_bi);
46031f9a 4911 if (atomic_dec_and_test(&conf->active_aligned_reads))
b1b46486 4912 wake_up(&conf->wait_for_quiescent);
6712ecf8 4913 return;
46031f9a
RBJ		 4914 }
4915
45b4233c 4916 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
46031f9a
RBJ		 4917
4918 add_bio_to_retry(raid_bi, conf);
f679623f
RBJ		 4919}
4920
7ef6b12a 4921static int raid5_read_one_chunk(struct mddev *mddev, struct bio *raid_bio)
f679623f 4922{
d1688a6d 4923 struct r5conf *conf = mddev->private;
8553fe7e 4924 int dd_idx;
f679623f 4925 struct bio* align_bi;
3cb03002 4926 struct md_rdev *rdev;
671488cc 4927 sector_t end_sector;
f679623f
RBJ		 4928
4929 if (!in_chunk_boundary(mddev, raid_bio)) {
7ef6b12a 4930 pr_debug("%s: non aligned\n", __func__);
f679623f
RBJ		 4931 return 0;
4932 }
4933 /*
a167f663 4934 * use bio_clone_mddev to make a copy of the bio
f679623f 4935 */
a167f663 4936 align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev);
f679623f
RBJ		 4937 if (!align_bi)
4938 return 0;
4939 /*
4940 * set bi_end_io to a new function, and set bi_private to the
4941 * original bio.
4942 */
4943 align_bi->bi_end_io = raid5_align_endio;
4944 align_bi->bi_private = raid_bio;
4945 /*
4946 * compute position
4947 */
4f024f37
KO		 4948 align_bi->bi_iter.bi_sector =
4949 raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector,
4950 0, &dd_idx, NULL);
f679623f 4951
f73a1c7d 4952 end_sector = bio_end_sector(align_bi);
f679623f 4953 rcu_read_lock();
671488cc
N		 4954 rdev = rcu_dereference(conf->disks[dd_idx].replacement);
4955 if (!rdev || test_bit(Faulty, &rdev->flags) ||
4956 rdev->recovery_offset < end_sector) {
4957 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
4958 if (rdev &&
4959 (test_bit(Faulty, &rdev->flags) ||
4960 !(test_bit(In_sync, &rdev->flags) ||
4961 rdev->recovery_offset >= end_sector)))
4962 rdev = NULL;
4963 }
4964 if (rdev) {
31c176ec
N		 4965 sector_t first_bad;
4966 int bad_sectors;
4967
f679623f
RBJ		 4968 atomic_inc(&rdev->nr_pending);
4969 rcu_read_unlock();
46031f9a
RBJ		 4970 raid_bio->bi_next = (void*)rdev;
4971 align_bi->bi_bdev = rdev->bdev;
b7c44ed9 4972 bio_clear_flag(align_bi, BIO_SEG_VALID);
46031f9a 4973
7140aafc 4974 if (is_badblock(rdev, align_bi->bi_iter.bi_sector,
4f024f37 4975 bio_sectors(align_bi),
31c176ec 4976 &first_bad, &bad_sectors)) {
387bb173
NB		 4977 bio_put(align_bi);
4978 rdev_dec_pending(rdev, mddev);
4979 return 0;
4980 }
4981
6c0544e2 4982 /* No reshape active, so we can trust rdev->data_offset */
4f024f37 4983 align_bi->bi_iter.bi_sector += rdev->data_offset;
6c0544e2 4984
46031f9a 4985 spin_lock_irq(&conf->device_lock);
b1b46486 4986 wait_event_lock_irq(conf->wait_for_quiescent,
46031f9a 4987 conf->quiesce == 0,
eed8c02e 4988 conf->device_lock);
46031f9a
RBJ		 4989 atomic_inc(&conf->active_aligned_reads);
4990 spin_unlock_irq(&conf->device_lock);
4991
e3620a3a
JB		 4992 if (mddev->gendisk)
4993 trace_block_bio_remap(bdev_get_queue(align_bi->bi_bdev),
4994 align_bi, disk_devt(mddev->gendisk),
4f024f37 4995 raid_bio->bi_iter.bi_sector);
f679623f
RBJ		 4996 generic_make_request(align_bi);
4997 return 1;
4998 } else {
4999 rcu_read_unlock();
46031f9a 5000 bio_put(align_bi);
f679623f
RBJ		 5001 return 0;
5002 }
5003}
5004
7ef6b12a
ML		 5005static struct bio *chunk_aligned_read(struct mddev *mddev, struct bio *raid_bio)
5006{
5007 struct bio *split;
5008
5009 do {
5010 sector_t sector = raid_bio->bi_iter.bi_sector;
5011 unsigned chunk_sects = mddev->chunk_sectors;
5012 unsigned sectors = chunk_sects - (sector & (chunk_sects-1));
5013
5014 if (sectors < bio_sectors(raid_bio)) {
5015 split = bio_split(raid_bio, sectors, GFP_NOIO, fs_bio_set);
5016 bio_chain(split, raid_bio);
5017 } else
5018 split = raid_bio;
5019
5020 if (!raid5_read_one_chunk(mddev, split)) {
5021 if (split != raid_bio)
5022 generic_make_request(raid_bio);
5023 return split;
5024 }
5025 } while (split != raid_bio);
5026
5027 return NULL;
5028}
5029
8b3e6cdc
DW		 5030/* __get_priority_stripe - get the next stripe to process
5031 *
5032 * Full stripe writes are allowed to pass preread active stripes up until
5033 * the bypass_threshold is exceeded. In general the bypass_count
5034 * increments when the handle_list is handled before the hold_list; however, it
5035 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
5036 * stripe with in flight i/o. The bypass_count will be reset when the
5037 * head of the hold_list has changed, i.e. the head was promoted to the
5038 * handle_list.
5039 */
851c30c9 5040static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group)
8b3e6cdc 5041{
851c30c9
SL		 5042 struct stripe_head *sh = NULL, *tmp;
5043 struct list_head *handle_list = NULL;
bfc90cb0 5044 struct r5worker_group *wg = NULL;
851c30c9
SL		 5045
5046 if (conf->worker_cnt_per_group == 0) {
5047 handle_list = &conf->handle_list;
5048 } else if (group != ANY_GROUP) {
5049 handle_list = &conf->worker_groups[group].handle_list;
bfc90cb0 5050 wg = &conf->worker_groups[group];
851c30c9
SL		 5051 } else {
5052 int i;
5053 for (i = 0; i < conf->group_cnt; i++) {
5054 handle_list = &conf->worker_groups[i].handle_list;
bfc90cb0 5055 wg = &conf->worker_groups[i];
851c30c9
SL		 5056 if (!list_empty(handle_list))
5057 break;
5058 }
5059 }
8b3e6cdc
DW		 5060
5061 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
5062 __func__,
851c30c9 5063 list_empty(handle_list) ? "empty" : "busy",
8b3e6cdc
DW		 5064 list_empty(&conf->hold_list) ? "empty" : "busy",
5065 atomic_read(&conf->pending_full_writes), conf->bypass_count);
5066
851c30c9
SL		 5067 if (!list_empty(handle_list)) {
5068 sh = list_entry(handle_list->next, typeof(*sh), lru);
8b3e6cdc
DW		 5069
5070 if (list_empty(&conf->hold_list))
5071 conf->bypass_count = 0;
5072 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
5073 if (conf->hold_list.next == conf->last_hold)
5074 conf->bypass_count++;
5075 else {
5076 conf->last_hold = conf->hold_list.next;
5077 conf->bypass_count -= conf->bypass_threshold;
5078 if (conf->bypass_count < 0)
5079 conf->bypass_count = 0;
5080 }
5081 }
5082 } else if (!list_empty(&conf->hold_list) &&
5083 ((conf->bypass_threshold &&
5084 conf->bypass_count > conf->bypass_threshold) ||
5085 atomic_read(&conf->pending_full_writes) == 0)) {
851c30c9
SL		 5086
5087 list_for_each_entry(tmp, &conf->hold_list, lru) {
5088 if (conf->worker_cnt_per_group == 0 ||
5089 group == ANY_GROUP ||
5090 !cpu_online(tmp->cpu) ||
5091 cpu_to_group(tmp->cpu) == group) {
5092 sh = tmp;
5093 break;
5094 }
5095 }
5096
5097 if (sh) {
5098 conf->bypass_count -= conf->bypass_threshold;
5099 if (conf->bypass_count < 0)
5100 conf->bypass_count = 0;
5101 }
bfc90cb0 5102 wg = NULL;
851c30c9
SL		 5103 }
5104
5105 if (!sh)
8b3e6cdc
DW		 5106 return NULL;
5107
bfc90cb0
SL		 5108 if (wg) {
5109 wg->stripes_cnt--;
5110 sh->group = NULL;
5111 }
8b3e6cdc 5112 list_del_init(&sh->lru);
c7a6d35e 5113 BUG_ON(atomic_inc_return(&sh->count) != 1);
8b3e6cdc
DW		 5114 return sh;
5115}
f679623f 5116
8811b596
SL		 5117struct raid5_plug_cb {
5118 struct blk_plug_cb cb;
5119 struct list_head list;
566c09c5 5120 struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
8811b596
SL		 5121};
5122
5123static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule)
5124{
5125 struct raid5_plug_cb *cb = container_of(
5126 blk_cb, struct raid5_plug_cb, cb);
5127 struct stripe_head *sh;
5128 struct mddev *mddev = cb->cb.data;
5129 struct r5conf *conf = mddev->private;
a9add5d9 5130 int cnt = 0;
566c09c5 5131 int hash;
8811b596
SL		 5132
5133 if (cb->list.next && !list_empty(&cb->list)) {
5134 spin_lock_irq(&conf->device_lock);
5135 while (!list_empty(&cb->list)) {
5136 sh = list_first_entry(&cb->list, struct stripe_head, lru);
5137 list_del_init(&sh->lru);
5138 /*
5139 * avoid race release_stripe_plug() sees
5140 * STRIPE_ON_UNPLUG_LIST clear but the stripe
5141 * is still in our list
5142 */
4e857c58 5143 smp_mb__before_atomic();
8811b596 5144 clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state);
773ca82f
SL		 5145 /*
5146 * STRIPE_ON_RELEASE_LIST could be set here. In that
5147 * case, the count is always > 1 here
5148 */
566c09c5
SL		 5149 hash = sh->hash_lock_index;
5150 __release_stripe(conf, sh, &cb->temp_inactive_list[hash]);
a9add5d9 5151 cnt++;
8811b596
SL		 5152 }
5153 spin_unlock_irq(&conf->device_lock);
5154 }
566c09c5
SL		 5155 release_inactive_stripe_list(conf, cb->temp_inactive_list,
5156 NR_STRIPE_HASH_LOCKS);
e3620a3a
JB		 5157 if (mddev->queue)
5158 trace_block_unplug(mddev->queue, cnt, !from_schedule);
8811b596
SL		 5159 kfree(cb);
5160}
5161
5162static void release_stripe_plug(struct mddev *mddev,
5163 struct stripe_head *sh)
5164{
5165 struct blk_plug_cb *blk_cb = blk_check_plugged(
5166 raid5_unplug, mddev,
5167 sizeof(struct raid5_plug_cb));
5168 struct raid5_plug_cb *cb;
5169
5170 if (!blk_cb) {
6d036f7d 5171 raid5_release_stripe(sh);
8811b596
SL		 5172 return;
5173 }
5174
5175 cb = container_of(blk_cb, struct raid5_plug_cb, cb);
5176
566c09c5
SL		 5177 if (cb->list.next == NULL) {
5178 int i;
8811b596 5179 INIT_LIST_HEAD(&cb->list);
566c09c5
SL		 5180 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5181 INIT_LIST_HEAD(cb->temp_inactive_list + i);
5182 }
8811b596
SL		 5183
5184 if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state))
5185 list_add_tail(&sh->lru, &cb->list);
5186 else
6d036f7d 5187 raid5_release_stripe(sh);
8811b596
SL		 5188}
5189
620125f2
SL		 5190static void make_discard_request(struct mddev *mddev, struct bio *bi)
5191{
5192 struct r5conf *conf = mddev->private;
5193 sector_t logical_sector, last_sector;
5194 struct stripe_head *sh;
5195 int remaining;
5196 int stripe_sectors;
5197
5198 if (mddev->reshape_position != MaxSector)
5199 /* Skip discard while reshape is happening */
5200 return;
5201
4f024f37
KO		 5202 logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1);
5203 last_sector = bi->bi_iter.bi_sector + (bi->bi_iter.bi_size>>9);
620125f2
SL		 5204
5205 bi->bi_next = NULL;
5206 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
5207
5208 stripe_sectors = conf->chunk_sectors *
5209 (conf->raid_disks - conf->max_degraded);
5210 logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector,
5211 stripe_sectors);
5212 sector_div(last_sector, stripe_sectors);
5213
5214 logical_sector *= conf->chunk_sectors;
5215 last_sector *= conf->chunk_sectors;
5216
5217 for (; logical_sector < last_sector;
5218 logical_sector += STRIPE_SECTORS) {
5219 DEFINE_WAIT(w);
5220 int d;
5221 again:
6d036f7d 5222 sh = raid5_get_active_stripe(conf, logical_sector, 0, 0, 0);
620125f2
SL		 5223 prepare_to_wait(&conf->wait_for_overlap, &w,
5224 TASK_UNINTERRUPTIBLE);
f8dfcffd
N		 5225 set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
5226 if (test_bit(STRIPE_SYNCING, &sh->state)) {
6d036f7d 5227 raid5_release_stripe(sh);
f8dfcffd
N		 5228 schedule();
5229 goto again;
5230 }
5231 clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
620125f2
SL		 5232 spin_lock_irq(&sh->stripe_lock);
5233 for (d = 0; d < conf->raid_disks; d++) {
5234 if (d == sh->pd_idx || d == sh->qd_idx)
5235 continue;
5236 if (sh->dev[d].towrite || sh->dev[d].toread) {
5237 set_bit(R5_Overlap, &sh->dev[d].flags);
5238 spin_unlock_irq(&sh->stripe_lock);
6d036f7d 5239 raid5_release_stripe(sh);
620125f2
SL		 5240 schedule();
5241 goto again;
5242 }
5243 }
f8dfcffd 5244 set_bit(STRIPE_DISCARD, &sh->state);
620125f2 5245 finish_wait(&conf->wait_for_overlap, &w);
7a87f434 5246 sh->overwrite_disks = 0;
620125f2
SL		 5247 for (d = 0; d < conf->raid_disks; d++) {
5248 if (d == sh->pd_idx || d == sh->qd_idx)
5249 continue;
5250 sh->dev[d].towrite = bi;
5251 set_bit(R5_OVERWRITE, &sh->dev[d].flags);
5252 raid5_inc_bi_active_stripes(bi);
7a87f434 5253 sh->overwrite_disks++;
620125f2
SL		 5254 }
5255 spin_unlock_irq(&sh->stripe_lock);
5256 if (conf->mddev->bitmap) {
5257 for (d = 0;
5258 d < conf->raid_disks - conf->max_degraded;
5259 d++)
5260 bitmap_startwrite(mddev->bitmap,
5261 sh->sector,
5262 STRIPE_SECTORS,
5263 0);
5264 sh->bm_seq = conf->seq_flush + 1;
5265 set_bit(STRIPE_BIT_DELAY, &sh->state);
5266 }
5267
5268 set_bit(STRIPE_HANDLE, &sh->state);
5269 clear_bit(STRIPE_DELAYED, &sh->state);
5270 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
5271 atomic_inc(&conf->preread_active_stripes);
5272 release_stripe_plug(mddev, sh);
5273 }
5274
5275 remaining = raid5_dec_bi_active_stripes(bi);
5276 if (remaining == 0) {
5277 md_write_end(mddev);
4246a0b6 5278 bio_endio(bi);
620125f2
SL		 5279 }
5280}
5281
849674e4 5282static void raid5_make_request(struct mddev *mddev, struct bio * bi)
1da177e4 5283{
d1688a6d 5284 struct r5conf *conf = mddev->private;
911d4ee8 5285 int dd_idx;
1da177e4
LT		 5286 sector_t new_sector;
5287 sector_t logical_sector, last_sector;
5288 struct stripe_head *sh;
a362357b 5289 const int rw = bio_data_dir(bi);
49077326 5290 int remaining;
27c0f68f
SL		 5291 DEFINE_WAIT(w);
5292 bool do_prepare;
3bddb7f8 5293 bool do_flush = false;
1da177e4 5294
1eff9d32 5295 if (unlikely(bi->bi_opf & REQ_PREFLUSH)) {
828cbe98
SL		 5296 int ret = r5l_handle_flush_request(conf->log, bi);
5297
5298 if (ret == 0)
5299 return;
5300 if (ret == -ENODEV) {
5301 md_flush_request(mddev, bi);
5302 return;
5303 }
5304 /* ret == -EAGAIN, fallback */
3bddb7f8
SL		 5305 /*
5306 * if r5l_handle_flush_request() didn't clear REQ_PREFLUSH,
5307 * we need to flush journal device
5308 */
5309 do_flush = bi->bi_opf & REQ_PREFLUSH;
e5dcdd80
N		 5310 }
5311
3d310eb7 5312 md_write_start(mddev, bi);
06d91a5f 5313
9ffc8f7c
EM		 5314 /*
5315 * If array is degraded, better not do chunk aligned read because
5316 * later we might have to read it again in order to reconstruct
5317 * data on failed drives.
5318 */
5319 if (rw == READ && mddev->degraded == 0 &&
2ded3703 5320 !r5c_is_writeback(conf->log) &&
7ef6b12a
ML		 5321 mddev->reshape_position == MaxSector) {
5322 bi = chunk_aligned_read(mddev, bi);
5323 if (!bi)
5324 return;
5325 }
52488615 5326
796a5cf0 5327 if (unlikely(bio_op(bi) == REQ_OP_DISCARD)) {
620125f2
SL		 5328 make_discard_request(mddev, bi);
5329 return;
5330 }
5331
4f024f37 5332 logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1);
f73a1c7d 5333 last_sector = bio_end_sector(bi);
1da177e4
LT		 5334 bi->bi_next = NULL;
5335 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
06d91a5f 5336
27c0f68f 5337 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
1da177e4 5338 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
b5663ba4 5339 int previous;
c46501b2 5340 int seq;
b578d55f 5341
27c0f68f 5342 do_prepare = false;
7ecaa1e6 5343 retry:
c46501b2 5344 seq = read_seqcount_begin(&conf->gen_lock);
b5663ba4 5345 previous = 0;
27c0f68f
SL		 5346 if (do_prepare)
5347 prepare_to_wait(&conf->wait_for_overlap, &w,
5348 TASK_UNINTERRUPTIBLE);
b0f9ec04 5349 if (unlikely(conf->reshape_progress != MaxSector)) {
fef9c61f 5350 /* spinlock is needed as reshape_progress may be
df8e7f76
N		 5351 * 64bit on a 32bit platform, and so it might be
5352 * possible to see a half-updated value
aeb878b0 5353 * Of course reshape_progress could change after
df8e7f76
N		 5354 * the lock is dropped, so once we get a reference
5355 * to the stripe that we think it is, we will have
5356 * to check again.
5357 */
7ecaa1e6 5358 spin_lock_irq(&conf->device_lock);
2c810cdd 5359 if (mddev->reshape_backwards
fef9c61f
N		 5360 ? logical_sector < conf->reshape_progress
5361 : logical_sector >= conf->reshape_progress) {
b5663ba4
N		 5362 previous = 1;
5363 } else {
2c810cdd 5364 if (mddev->reshape_backwards
fef9c61f
N		 5365 ? logical_sector < conf->reshape_safe
5366 : logical_sector >= conf->reshape_safe) {
b578d55f
N		 5367 spin_unlock_irq(&conf->device_lock);
5368 schedule();
27c0f68f 5369 do_prepare = true;
b578d55f
N		 5370 goto retry;
5371 }
5372 }
7ecaa1e6
N		 5373 spin_unlock_irq(&conf->device_lock);
5374 }
16a53ecc 5375
112bf897
N		 5376 new_sector = raid5_compute_sector(conf, logical_sector,
5377 previous,
911d4ee8 5378 &dd_idx, NULL);
849674e4 5379 pr_debug("raid456: raid5_make_request, sector %llu logical %llu\n",
c46501b2 5380 (unsigned long long)new_sector,
1da177e4
LT		 5381 (unsigned long long)logical_sector);
5382
6d036f7d 5383 sh = raid5_get_active_stripe(conf, new_sector, previous,
1eff9d32 5384 (bi->bi_opf & REQ_RAHEAD), 0);
1da177e4 5385 if (sh) {
b0f9ec04 5386 if (unlikely(previous)) {
7ecaa1e6 5387 /* expansion might have moved on while waiting for a
df8e7f76
N		 5388 * stripe, so we must do the range check again.
5389 * Expansion could still move past after this
5390 * test, but as we are holding a reference to
5391 * 'sh', we know that if that happens,
5392 * STRIPE_EXPANDING will get set and the expansion
5393 * won't proceed until we finish with the stripe.
7ecaa1e6
N		 5394 */
5395 int must_retry = 0;
5396 spin_lock_irq(&conf->device_lock);
2c810cdd 5397 if (mddev->reshape_backwards
b0f9ec04
N		 5398 ? logical_sector >= conf->reshape_progress
5399 : logical_sector < conf->reshape_progress)
7ecaa1e6
N		 5400 /* mismatch, need to try again */
5401 must_retry = 1;
5402 spin_unlock_irq(&conf->device_lock);
5403 if (must_retry) {
6d036f7d 5404 raid5_release_stripe(sh);
7a3ab908 5405 schedule();
27c0f68f 5406 do_prepare = true;
7ecaa1e6
N		 5407 goto retry;
5408 }
5409 }
c46501b2
N		 5410 if (read_seqcount_retry(&conf->gen_lock, seq)) {
5411 /* Might have got the wrong stripe_head
5412 * by accident
5413 */
6d036f7d 5414 raid5_release_stripe(sh);
c46501b2
N		 5415 goto retry;
5416 }
e62e58a5 5417
ffd96e35 5418 if (rw == WRITE &&
a5c308d4 5419 logical_sector >= mddev->suspend_lo &&
e464eafd 5420 logical_sector < mddev->suspend_hi) {
6d036f7d 5421 raid5_release_stripe(sh);
e62e58a5
N		 5422 /* As the suspend_* range is controlled by
5423 * userspace, we want an interruptible
5424 * wait.
5425 */
5426 flush_signals(current);
5427 prepare_to_wait(&conf->wait_for_overlap,
5428 &w, TASK_INTERRUPTIBLE);
5429 if (logical_sector >= mddev->suspend_lo &&
27c0f68f 5430 logical_sector < mddev->suspend_hi) {
e62e58a5 5431 schedule();
27c0f68f
SL		 5432 do_prepare = true;
5433 }
e464eafd
N		 5434 goto retry;
5435 }
7ecaa1e6
N		 5436
5437 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
da41ba65 5438 !add_stripe_bio(sh, bi, dd_idx, rw, previous)) {
7ecaa1e6
N		 5439 /* Stripe is busy expanding or
5440 * add failed due to overlap. Flush everything
1da177e4
LT		 5441 * and wait a while
5442 */
482c0834 5443 md_wakeup_thread(mddev->thread);
6d036f7d 5444 raid5_release_stripe(sh);
1da177e4 5445 schedule();
27c0f68f 5446 do_prepare = true;
1da177e4
LT		 5447 goto retry;
5448 }
3bddb7f8
SL		 5449 if (do_flush) {
5450 set_bit(STRIPE_R5C_PREFLUSH, &sh->state);
5451 /* we only need flush for one stripe */
5452 do_flush = false;
5453 }
5454
6ed3003c
N		 5455 set_bit(STRIPE_HANDLE, &sh->state);
5456 clear_bit(STRIPE_DELAYED, &sh->state);
59fc630b 5457 if ((!sh->batch_head || sh == sh->batch_head) &&
1eff9d32 5458 (bi->bi_opf & REQ_SYNC) &&
729a1866
N		 5459 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
5460 atomic_inc(&conf->preread_active_stripes);
8811b596 5461 release_stripe_plug(mddev, sh);
1da177e4
LT		 5462 } else {
5463 /* cannot get stripe for read-ahead, just give-up */
4246a0b6 5464 bi->bi_error = -EIO;
1da177e4
LT		 5465 break;
5466 }
1da177e4 5467 }
27c0f68f 5468 finish_wait(&conf->wait_for_overlap, &w);
7c13edc8 5469
e7836bd6 5470 remaining = raid5_dec_bi_active_stripes(bi);
f6344757 5471 if (remaining == 0) {
1da177e4 5472
16a53ecc 5473 if ( rw == WRITE )
1da177e4 5474 md_write_end(mddev);
6712ecf8 5475
0a82a8d1
LT		 5476 trace_block_bio_complete(bdev_get_queue(bi->bi_bdev),
5477 bi, 0);
4246a0b6 5478 bio_endio(bi);
1da177e4 5479 }
1da177e4
LT		 5480}
5481
fd01b88c 5482static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks);
b522adcd 5483
fd01b88c 5484static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped)
1da177e4 5485{
52c03291
N		 5486 /* reshaping is quite different to recovery/resync so it is
5487 * handled quite separately ... here.
5488 *
5489 * On each call to sync_request, we gather one chunk worth of
5490 * destination stripes and flag them as expanding.
5491 * Then we find all the source stripes and request reads.
5492 * As the reads complete, handle_stripe will copy the data
5493 * into the destination stripe and release that stripe.
5494 */
d1688a6d 5495 struct r5conf *conf = mddev->private;
1da177e4 5496 struct stripe_head *sh;
ccfcc3c1 5497 sector_t first_sector, last_sector;
f416885e
N		 5498 int raid_disks = conf->previous_raid_disks;
5499 int data_disks = raid_disks - conf->max_degraded;
5500 int new_data_disks = conf->raid_disks - conf->max_degraded;
52c03291
N		 5501 int i;
5502 int dd_idx;
c8f517c4 5503 sector_t writepos, readpos, safepos;
ec32a2bd 5504 sector_t stripe_addr;
7a661381 5505 int reshape_sectors;
ab69ae12 5506 struct list_head stripes;
92140480 5507 sector_t retn;
52c03291 5508
fef9c61f
N		 5509 if (sector_nr == 0) {
5510 /* If restarting in the middle, skip the initial sectors */
2c810cdd 5511 if (mddev->reshape_backwards &&
fef9c61f
N		 5512 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
5513 sector_nr = raid5_size(mddev, 0, 0)
5514 - conf->reshape_progress;
6cbd8148
N		 5515 } else if (mddev->reshape_backwards &&
5516 conf->reshape_progress == MaxSector) {
5517 /* shouldn't happen, but just in case, finish up.*/
5518 sector_nr = MaxSector;
2c810cdd 5519 } else if (!mddev->reshape_backwards &&
fef9c61f
N		 5520 conf->reshape_progress > 0)
5521 sector_nr = conf->reshape_progress;
f416885e 5522 sector_div(sector_nr, new_data_disks);
fef9c61f 5523 if (sector_nr) {
8dee7211
N		 5524 mddev->curr_resync_completed = sector_nr;
5525 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
fef9c61f 5526 *skipped = 1;
92140480
N		 5527 retn = sector_nr;
5528 goto finish;
fef9c61f 5529 }
52c03291
N		 5530 }
5531
7a661381
N		 5532 /* We need to process a full chunk at a time.
5533 * If old and new chunk sizes differ, we need to process the
5534 * largest of these
5535 */
3cb5edf4
N		 5536
5537 reshape_sectors = max(conf->chunk_sectors, conf->prev_chunk_sectors);
7a661381 5538
b5254dd5
N		 5539 /* We update the metadata at least every 10 seconds, or when
5540 * the data about to be copied would over-write the source of
5541 * the data at the front of the range. i.e. one new_stripe
5542 * along from reshape_progress new_maps to after where
5543 * reshape_safe old_maps to
52c03291 5544 */
fef9c61f 5545 writepos = conf->reshape_progress;
f416885e 5546 sector_div(writepos, new_data_disks);
c8f517c4
N		 5547 readpos = conf->reshape_progress;
5548 sector_div(readpos, data_disks);
fef9c61f 5549 safepos = conf->reshape_safe;
f416885e 5550 sector_div(safepos, data_disks);
2c810cdd 5551 if (mddev->reshape_backwards) {
c74c0d76
N		 5552 BUG_ON(writepos < reshape_sectors);
5553 writepos -= reshape_sectors;
c8f517c4 5554 readpos += reshape_sectors;
7a661381 5555 safepos += reshape_sectors;
fef9c61f 5556 } else {
7a661381 5557 writepos += reshape_sectors;
c74c0d76
N		 5558 /* readpos and safepos are worst-case calculations.
5559 * A negative number is overly pessimistic, and causes
5560 * obvious problems for unsigned storage. So clip to 0.
5561 */
ed37d83e
N		 5562 readpos -= min_t(sector_t, reshape_sectors, readpos);
5563 safepos -= min_t(sector_t, reshape_sectors, safepos);
fef9c61f 5564 }
52c03291 5565
b5254dd5
N		 5566 /* Having calculated the 'writepos' possibly use it
5567 * to set 'stripe_addr' which is where we will write to.
5568 */
5569 if (mddev->reshape_backwards) {
5570 BUG_ON(conf->reshape_progress == 0);
5571 stripe_addr = writepos;
5572 BUG_ON((mddev->dev_sectors &
5573 ~((sector_t)reshape_sectors - 1))
5574 - reshape_sectors - stripe_addr
5575 != sector_nr);
5576 } else {
5577 BUG_ON(writepos != sector_nr + reshape_sectors);
5578 stripe_addr = sector_nr;
5579 }
5580
c8f517c4
N		 5581 /* 'writepos' is the most advanced device address we might write.
5582 * 'readpos' is the least advanced device address we might read.
5583 * 'safepos' is the least address recorded in the metadata as having
5584 * been reshaped.
b5254dd5
N		 5585 * If there is a min_offset_diff, these are adjusted either by
5586 * increasing the safepos/readpos if diff is negative, or
5587 * increasing writepos if diff is positive.
5588 * If 'readpos' is then behind 'writepos', there is no way that we can
c8f517c4
N		 5589 * ensure safety in the face of a crash - that must be done by userspace
5590 * making a backup of the data. So in that case there is no particular
5591 * rush to update metadata.
5592 * Otherwise if 'safepos' is behind 'writepos', then we really need to
5593 * update the metadata to advance 'safepos' to match 'readpos' so that
5594 * we can be safe in the event of a crash.
5595 * So we insist on updating metadata if safepos is behind writepos and
5596 * readpos is beyond writepos.
5597 * In any case, update the metadata every 10 seconds.
5598 * Maybe that number should be configurable, but I'm not sure it is
5599 * worth it.... maybe it could be a multiple of safemode_delay???
5600 */
b5254dd5
N		 5601 if (conf->min_offset_diff < 0) {
5602 safepos += -conf->min_offset_diff;
5603 readpos += -conf->min_offset_diff;
5604 } else
5605 writepos += conf->min_offset_diff;
5606
2c810cdd 5607 if ((mddev->reshape_backwards
c8f517c4
N		 5608 ? (safepos > writepos && readpos < writepos)
5609 : (safepos < writepos && readpos > writepos)) ||
5610 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
52c03291
N		 5611 /* Cannot proceed until we've updated the superblock... */
5612 wait_event(conf->wait_for_overlap,
c91abf5a
N		 5613 atomic_read(&conf->reshape_stripes)==0
5614 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5615 if (atomic_read(&conf->reshape_stripes) != 0)
5616 return 0;
fef9c61f 5617 mddev->reshape_position = conf->reshape_progress;
75d3da43 5618 mddev->curr_resync_completed = sector_nr;
c8f517c4 5619 conf->reshape_checkpoint = jiffies;
850b2b42 5620 set_bit(MD_CHANGE_DEVS, &mddev->flags);
52c03291 5621 md_wakeup_thread(mddev->thread);
850b2b42 5622 wait_event(mddev->sb_wait, mddev->flags == 0 ||
c91abf5a
N		 5623 test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5624 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5625 return 0;
52c03291 5626 spin_lock_irq(&conf->device_lock);
fef9c61f 5627 conf->reshape_safe = mddev->reshape_position;
52c03291
N		 5628 spin_unlock_irq(&conf->device_lock);
5629 wake_up(&conf->wait_for_overlap);
acb180b0 5630 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
52c03291
N		 5631 }
5632
ab69ae12 5633 INIT_LIST_HEAD(&stripes);
7a661381 5634 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
52c03291 5635 int j;
a9f326eb 5636 int skipped_disk = 0;
6d036f7d 5637 sh = raid5_get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
52c03291
N		 5638 set_bit(STRIPE_EXPANDING, &sh->state);
5639 atomic_inc(&conf->reshape_stripes);
5640 /* If any of this stripe is beyond the end of the old
5641 * array, then we need to zero those blocks
5642 */
5643 for (j=sh->disks; j--;) {
5644 sector_t s;
5645 if (j == sh->pd_idx)
5646 continue;
f416885e 5647 if (conf->level == 6 &&
d0dabf7e 5648 j == sh->qd_idx)
f416885e 5649 continue;
6d036f7d 5650 s = raid5_compute_blocknr(sh, j, 0);
b522adcd 5651 if (s < raid5_size(mddev, 0, 0)) {
a9f326eb 5652 skipped_disk = 1;
52c03291
N		 5653 continue;
5654 }
5655 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
5656 set_bit(R5_Expanded, &sh->dev[j].flags);
5657 set_bit(R5_UPTODATE, &sh->dev[j].flags);
5658 }
a9f326eb 5659 if (!skipped_disk) {
52c03291
N		 5660 set_bit(STRIPE_EXPAND_READY, &sh->state);
5661 set_bit(STRIPE_HANDLE, &sh->state);
5662 }
ab69ae12 5663 list_add(&sh->lru, &stripes);
52c03291
N		 5664 }
5665 spin_lock_irq(&conf->device_lock);
2c810cdd 5666 if (mddev->reshape_backwards)
7a661381 5667 conf->reshape_progress -= reshape_sectors * new_data_disks;
fef9c61f 5668 else
7a661381 5669 conf->reshape_progress += reshape_sectors * new_data_disks;
52c03291
N		 5670 spin_unlock_irq(&conf->device_lock);
5671 /* Ok, those stripe are ready. We can start scheduling
5672 * reads on the source stripes.
5673 * The source stripes are determined by mapping the first and last
5674 * block on the destination stripes.
5675 */
52c03291 5676 first_sector =
ec32a2bd 5677 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
911d4ee8 5678 1, &dd_idx, NULL);
52c03291 5679 last_sector =
0e6e0271 5680 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
09c9e5fa 5681 * new_data_disks - 1),
911d4ee8 5682 1, &dd_idx, NULL);
58c0fed4
AN		 5683 if (last_sector >= mddev->dev_sectors)
5684 last_sector = mddev->dev_sectors - 1;
52c03291 5685 while (first_sector <= last_sector) {
6d036f7d 5686 sh = raid5_get_active_stripe(conf, first_sector, 1, 0, 1);
52c03291
N		 5687 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
5688 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 5689 raid5_release_stripe(sh);
52c03291
N		 5690 first_sector += STRIPE_SECTORS;
5691 }
ab69ae12
N		 5692 /* Now that the sources are clearly marked, we can release
5693 * the destination stripes
5694 */
5695 while (!list_empty(&stripes)) {
5696 sh = list_entry(stripes.next, struct stripe_head, lru);
5697 list_del_init(&sh->lru);
6d036f7d 5698 raid5_release_stripe(sh);
ab69ae12 5699 }
c6207277
N		 5700 /* If this takes us to the resync_max point where we have to pause,
5701 * then we need to write out the superblock.
5702 */
7a661381 5703 sector_nr += reshape_sectors;
92140480
N		 5704 retn = reshape_sectors;
5705finish:
c5e19d90
N		 5706 if (mddev->curr_resync_completed > mddev->resync_max ||
5707 (sector_nr - mddev->curr_resync_completed) * 2
c03f6a19 5708 >= mddev->resync_max - mddev->curr_resync_completed) {
c6207277
N		 5709 /* Cannot proceed until we've updated the superblock... */
5710 wait_event(conf->wait_for_overlap,
c91abf5a
N		 5711 atomic_read(&conf->reshape_stripes) == 0
5712 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5713 if (atomic_read(&conf->reshape_stripes) != 0)
5714 goto ret;
fef9c61f 5715 mddev->reshape_position = conf->reshape_progress;
75d3da43 5716 mddev->curr_resync_completed = sector_nr;
c8f517c4 5717 conf->reshape_checkpoint = jiffies;
c6207277
N		 5718 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5719 md_wakeup_thread(mddev->thread);
5720 wait_event(mddev->sb_wait,
5721 !test_bit(MD_CHANGE_DEVS, &mddev->flags)
c91abf5a
N		 5722 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5723 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5724 goto ret;
c6207277 5725 spin_lock_irq(&conf->device_lock);
fef9c61f 5726 conf->reshape_safe = mddev->reshape_position;
c6207277
N		 5727 spin_unlock_irq(&conf->device_lock);
5728 wake_up(&conf->wait_for_overlap);
acb180b0 5729 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
c6207277 5730 }
c91abf5a 5731ret:
92140480 5732 return retn;
52c03291
N		 5733}
5734
849674e4
SL		 5735static inline sector_t raid5_sync_request(struct mddev *mddev, sector_t sector_nr,
5736 int *skipped)
52c03291 5737{
d1688a6d 5738 struct r5conf *conf = mddev->private;
52c03291 5739 struct stripe_head *sh;
58c0fed4 5740 sector_t max_sector = mddev->dev_sectors;
57dab0bd 5741 sector_t sync_blocks;
16a53ecc
N		 5742 int still_degraded = 0;
5743 int i;
1da177e4 5744
72626685 5745 if (sector_nr >= max_sector) {
1da177e4 5746 /* just being told to finish up .. nothing much to do */
cea9c228 5747
29269553
N		 5748 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
5749 end_reshape(conf);
5750 return 0;
5751 }
72626685
N		 5752
5753 if (mddev->curr_resync < max_sector) /* aborted */
5754 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
5755 &sync_blocks, 1);
16a53ecc 5756 else /* completed sync */
72626685
N		 5757 conf->fullsync = 0;
5758 bitmap_close_sync(mddev->bitmap);
5759
1da177e4
LT		 5760 return 0;
5761 }
ccfcc3c1 5762
64bd660b
N		 5763 /* Allow raid5_quiesce to complete */
5764 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
5765
52c03291
N		 5766 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5767 return reshape_request(mddev, sector_nr, skipped);
f6705578 5768
c6207277
N		 5769 /* No need to check resync_max as we never do more than one
5770 * stripe, and as resync_max will always be on a chunk boundary,
5771 * if the check in md_do_sync didn't fire, there is no chance
5772 * of overstepping resync_max here
5773 */
5774
16a53ecc 5775 /* if there is too many failed drives and we are trying
1da177e4
LT		 5776 * to resync, then assert that we are finished, because there is
5777 * nothing we can do.
5778 */
3285edf1 5779 if (mddev->degraded >= conf->max_degraded &&
16a53ecc 5780 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
58c0fed4 5781 sector_t rv = mddev->dev_sectors - sector_nr;
57afd89f 5782 *skipped = 1;
1da177e4
LT		 5783 return rv;
5784 }
6f608040 5785 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
5786 !conf->fullsync &&
5787 !bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
5788 sync_blocks >= STRIPE_SECTORS) {
72626685
N		 5789 /* we can skip this block, and probably more */
5790 sync_blocks /= STRIPE_SECTORS;
5791 *skipped = 1;
5792 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
5793 }
1da177e4 5794
c40f341f 5795 bitmap_cond_end_sync(mddev->bitmap, sector_nr, false);
b47490c9 5796
6d036f7d 5797 sh = raid5_get_active_stripe(conf, sector_nr, 0, 1, 0);
1da177e4 5798 if (sh == NULL) {
6d036f7d 5799 sh = raid5_get_active_stripe(conf, sector_nr, 0, 0, 0);
1da177e4 5800 /* make sure we don't swamp the stripe cache if someone else
16a53ecc 5801 * is trying to get access
1da177e4 5802 */
66c006a5 5803 schedule_timeout_uninterruptible(1);
1da177e4 5804 }
16a53ecc 5805 /* Need to check if array will still be degraded after recovery/resync
16d9cfab
EM		 5806 * Note in case of > 1 drive failures it's possible we're rebuilding
5807 * one drive while leaving another faulty drive in array.
16a53ecc 5808 */
16d9cfab
EM		 5809 rcu_read_lock();
5810 for (i = 0; i < conf->raid_disks; i++) {
5811 struct md_rdev *rdev = ACCESS_ONCE(conf->disks[i].rdev);
5812
5813 if (rdev == NULL || test_bit(Faulty, &rdev->flags))
16a53ecc 5814 still_degraded = 1;
16d9cfab
EM		 5815 }
5816 rcu_read_unlock();
16a53ecc
N		 5817
5818 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
5819
83206d66 5820 set_bit(STRIPE_SYNC_REQUESTED, &sh->state);
053f5b65 5821 set_bit(STRIPE_HANDLE, &sh->state);
1da177e4 5822
6d036f7d 5823 raid5_release_stripe(sh);
1da177e4
LT		 5824
5825 return STRIPE_SECTORS;
5826}
5827
d1688a6d 5828static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio)
46031f9a
RBJ		 5829{
5830 /* We may not be able to submit a whole bio at once as there
5831 * may not be enough stripe_heads available.
5832 * We cannot pre-allocate enough stripe_heads as we may need
5833 * more than exist in the cache (if we allow ever large chunks).
5834 * So we do one stripe head at a time and record in
5835 * ->bi_hw_segments how many have been done.
5836 *
5837 * We *know* that this entire raid_bio is in one chunk, so
5838 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
5839 */
5840 struct stripe_head *sh;
911d4ee8 5841 int dd_idx;
46031f9a
RBJ		 5842 sector_t sector, logical_sector, last_sector;
5843 int scnt = 0;
5844 int remaining;
5845 int handled = 0;
5846
4f024f37
KO		 5847 logical_sector = raid_bio->bi_iter.bi_sector &
5848 ~((sector_t)STRIPE_SECTORS-1);
112bf897 5849 sector = raid5_compute_sector(conf, logical_sector,
911d4ee8 5850 0, &dd_idx, NULL);
f73a1c7d 5851 last_sector = bio_end_sector(raid_bio);
46031f9a
RBJ		 5852
5853 for (; logical_sector < last_sector;
387bb173
NB		 5854 logical_sector += STRIPE_SECTORS,
5855 sector += STRIPE_SECTORS,
5856 scnt++) {
46031f9a 5857
e7836bd6 5858 if (scnt < raid5_bi_processed_stripes(raid_bio))
46031f9a
RBJ		 5859 /* already done this stripe */
5860 continue;
5861
6d036f7d 5862 sh = raid5_get_active_stripe(conf, sector, 0, 1, 1);
46031f9a
RBJ		 5863
5864 if (!sh) {
5865 /* failed to get a stripe - must wait */
e7836bd6 5866 raid5_set_bi_processed_stripes(raid_bio, scnt);
46031f9a
RBJ		 5867 conf->retry_read_aligned = raid_bio;
5868 return handled;
5869 }
5870
da41ba65 5871 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) {
6d036f7d 5872 raid5_release_stripe(sh);
e7836bd6 5873 raid5_set_bi_processed_stripes(raid_bio, scnt);
387bb173
NB		 5874 conf->retry_read_aligned = raid_bio;
5875 return handled;
5876 }
5877
3f9e7c14 5878 set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags);
36d1c647 5879 handle_stripe(sh);
6d036f7d 5880 raid5_release_stripe(sh);
46031f9a
RBJ		 5881 handled++;
5882 }
e7836bd6 5883 remaining = raid5_dec_bi_active_stripes(raid_bio);
0a82a8d1
LT		 5884 if (remaining == 0) {
5885 trace_block_bio_complete(bdev_get_queue(raid_bio->bi_bdev),
5886 raid_bio, 0);
4246a0b6 5887 bio_endio(raid_bio);
0a82a8d1 5888 }
46031f9a 5889 if (atomic_dec_and_test(&conf->active_aligned_reads))
b1b46486 5890 wake_up(&conf->wait_for_quiescent);
46031f9a
RBJ		 5891 return handled;
5892}
5893
bfc90cb0 5894static int handle_active_stripes(struct r5conf *conf, int group,
566c09c5
SL		 5895 struct r5worker *worker,
5896 struct list_head *temp_inactive_list)
46a06401
SL		 5897{
5898 struct stripe_head *batch[MAX_STRIPE_BATCH], *sh;
566c09c5
SL		 5899 int i, batch_size = 0, hash;
5900 bool release_inactive = false;
46a06401
SL		 5901
5902 while (batch_size < MAX_STRIPE_BATCH &&
851c30c9 5903 (sh = __get_priority_stripe(conf, group)) != NULL)
46a06401
SL		 5904 batch[batch_size++] = sh;
5905
566c09c5
SL		 5906 if (batch_size == 0) {
5907 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5908 if (!list_empty(temp_inactive_list + i))
5909 break;
a8c34f91
SL		 5910 if (i == NR_STRIPE_HASH_LOCKS) {
5911 spin_unlock_irq(&conf->device_lock);
5912 r5l_flush_stripe_to_raid(conf->log);
5913 spin_lock_irq(&conf->device_lock);
566c09c5 5914 return batch_size;
a8c34f91 5915 }
566c09c5
SL		 5916 release_inactive = true;
5917 }
46a06401
SL		 5918 spin_unlock_irq(&conf->device_lock);
5919
566c09c5
SL		 5920 release_inactive_stripe_list(conf, temp_inactive_list,
5921 NR_STRIPE_HASH_LOCKS);
5922
a8c34f91 5923 r5l_flush_stripe_to_raid(conf->log);
566c09c5
SL		 5924 if (release_inactive) {
5925 spin_lock_irq(&conf->device_lock);
5926 return 0;
5927 }
5928
46a06401
SL		 5929 for (i = 0; i < batch_size; i++)
5930 handle_stripe(batch[i]);
f6bed0ef 5931 r5l_write_stripe_run(conf->log);
46a06401
SL		 5932
5933 cond_resched();
5934
5935 spin_lock_irq(&conf->device_lock);
566c09c5
SL		 5936 for (i = 0; i < batch_size; i++) {
5937 hash = batch[i]->hash_lock_index;
5938 __release_stripe(conf, batch[i], &temp_inactive_list[hash]);
5939 }
46a06401
SL		 5940 return batch_size;
5941}
46031f9a 5942
851c30c9
SL		 5943static void raid5_do_work(struct work_struct *work)
5944{
5945 struct r5worker *worker = container_of(work, struct r5worker, work);
5946 struct r5worker_group *group = worker->group;
5947 struct r5conf *conf = group->conf;
5948 int group_id = group - conf->worker_groups;
5949 int handled;
5950 struct blk_plug plug;
5951
5952 pr_debug("+++ raid5worker active\n");
5953
5954 blk_start_plug(&plug);
5955 handled = 0;
5956 spin_lock_irq(&conf->device_lock);
5957 while (1) {
5958 int batch_size, released;
5959
566c09c5 5960 released = release_stripe_list(conf, worker->temp_inactive_list);
851c30c9 5961
566c09c5
SL		 5962 batch_size = handle_active_stripes(conf, group_id, worker,
5963 worker->temp_inactive_list);
bfc90cb0 5964 worker->working = false;
851c30c9
SL		 5965 if (!batch_size && !released)
5966 break;
5967 handled += batch_size;
5968 }
5969 pr_debug("%d stripes handled\n", handled);
5970
5971 spin_unlock_irq(&conf->device_lock);
5972 blk_finish_plug(&plug);
5973
5974 pr_debug("--- raid5worker inactive\n");
5975}
5976
1da177e4
LT		 5977/*
5978 * This is our raid5 kernel thread.
5979 *
5980 * We scan the hash table for stripes which can be handled now.
5981 * During the scan, completed stripes are saved for us by the interrupt
5982 * handler, so that they will not have to wait for our next wakeup.
5983 */
4ed8731d 5984static void raid5d(struct md_thread *thread)
1da177e4 5985{
4ed8731d 5986 struct mddev *mddev = thread->mddev;
d1688a6d 5987 struct r5conf *conf = mddev->private;
1da177e4 5988 int handled;
e1dfa0a2 5989 struct blk_plug plug;
1da177e4 5990
45b4233c 5991 pr_debug("+++ raid5d active\n");
1da177e4
LT		 5992
5993 md_check_recovery(mddev);
1da177e4 5994
c3cce6cd
N		 5995 if (!bio_list_empty(&conf->return_bi) &&
5996 !test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
5997 struct bio_list tmp = BIO_EMPTY_LIST;
5998 spin_lock_irq(&conf->device_lock);
5999 if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
6000 bio_list_merge(&tmp, &conf->return_bi);
6001 bio_list_init(&conf->return_bi);
6002 }
6003 spin_unlock_irq(&conf->device_lock);
6004 return_io(&tmp);
6005 }
6006
e1dfa0a2 6007 blk_start_plug(&plug);
1da177e4
LT		 6008 handled = 0;
6009 spin_lock_irq(&conf->device_lock);
6010 while (1) {
46031f9a 6011 struct bio *bio;
773ca82f
SL		 6012 int batch_size, released;
6013
566c09c5 6014 released = release_stripe_list(conf, conf->temp_inactive_list);
edbe83ab
N		 6015 if (released)
6016 clear_bit(R5_DID_ALLOC, &conf->cache_state);
1da177e4 6017
0021b7bc 6018 if (
7c13edc8
N		 6019 !list_empty(&conf->bitmap_list)) {
6020 /* Now is a good time to flush some bitmap updates */
6021 conf->seq_flush++;
700e432d 6022 spin_unlock_irq(&conf->device_lock);
72626685 6023 bitmap_unplug(mddev->bitmap);
700e432d 6024 spin_lock_irq(&conf->device_lock);
7c13edc8 6025 conf->seq_write = conf->seq_flush;
566c09c5 6026 activate_bit_delay(conf, conf->temp_inactive_list);
72626685 6027 }
0021b7bc 6028 raid5_activate_delayed(conf);
72626685 6029
46031f9a
RBJ		 6030 while ((bio = remove_bio_from_retry(conf))) {
6031 int ok;
6032 spin_unlock_irq(&conf->device_lock);
6033 ok = retry_aligned_read(conf, bio);
6034 spin_lock_irq(&conf->device_lock);
6035 if (!ok)
6036 break;
6037 handled++;
6038 }
6039
566c09c5
SL		 6040 batch_size = handle_active_stripes(conf, ANY_GROUP, NULL,
6041 conf->temp_inactive_list);
773ca82f 6042 if (!batch_size && !released)
1da177e4 6043 break;
46a06401 6044 handled += batch_size;
1da177e4 6045
46a06401
SL		 6046 if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) {
6047 spin_unlock_irq(&conf->device_lock);
de393cde 6048 md_check_recovery(mddev);
46a06401
SL		 6049 spin_lock_irq(&conf->device_lock);
6050 }
1da177e4 6051 }
45b4233c 6052 pr_debug("%d stripes handled\n", handled);
1da177e4
LT		 6053
6054 spin_unlock_irq(&conf->device_lock);
2d5b569b
N		 6055 if (test_and_clear_bit(R5_ALLOC_MORE, &conf->cache_state) &&
6056 mutex_trylock(&conf->cache_size_mutex)) {
edbe83ab
N		 6057 grow_one_stripe(conf, __GFP_NOWARN);
6058 /* Set flag even if allocation failed. This helps
6059 * slow down allocation requests when mem is short
6060 */
6061 set_bit(R5_DID_ALLOC, &conf->cache_state);
2d5b569b 6062 mutex_unlock(&conf->cache_size_mutex);
edbe83ab 6063 }
1da177e4 6064
0576b1c6
SL		 6065 r5l_flush_stripe_to_raid(conf->log);
6066
c9f21aaf 6067 async_tx_issue_pending_all();
e1dfa0a2 6068 blk_finish_plug(&plug);
1da177e4 6069
45b4233c 6070 pr_debug("--- raid5d inactive\n");
1da177e4
LT		 6071}
6072
3f294f4f 6073static ssize_t
fd01b88c 6074raid5_show_stripe_cache_size(struct mddev *mddev, char *page)
3f294f4f 6075{
7b1485ba
N		 6076 struct r5conf *conf;
6077 int ret = 0;
6078 spin_lock(&mddev->lock);
6079 conf = mddev->private;
96de1e66 6080 if (conf)
edbe83ab 6081 ret = sprintf(page, "%d\n", conf->min_nr_stripes);
7b1485ba
N		 6082 spin_unlock(&mddev->lock);
6083 return ret;
3f294f4f
N		 6084}
6085
c41d4ac4 6086int
fd01b88c 6087raid5_set_cache_size(struct mddev *mddev, int size)
3f294f4f 6088{
d1688a6d 6089 struct r5conf *conf = mddev->private;
b5470dc5
DW		 6090 int err;
6091
c41d4ac4 6092 if (size <= 16 || size > 32768)
3f294f4f 6093 return -EINVAL;
486f0644 6094
edbe83ab 6095 conf->min_nr_stripes = size;
2d5b569b 6096 mutex_lock(&conf->cache_size_mutex);
486f0644
N		 6097 while (size < conf->max_nr_stripes &&
6098 drop_one_stripe(conf))
6099 ;
2d5b569b 6100 mutex_unlock(&conf->cache_size_mutex);
486f0644 6101
edbe83ab 6102
b5470dc5
DW		 6103 err = md_allow_write(mddev);
6104 if (err)
6105 return err;
486f0644 6106
2d5b569b 6107 mutex_lock(&conf->cache_size_mutex);
486f0644
N		 6108 while (size > conf->max_nr_stripes)
6109 if (!grow_one_stripe(conf, GFP_KERNEL))
6110 break;
2d5b569b 6111 mutex_unlock(&conf->cache_size_mutex);
486f0644 6112
c41d4ac4
N		 6113 return 0;
6114}
6115EXPORT_SYMBOL(raid5_set_cache_size);
6116
6117static ssize_t
fd01b88c 6118raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len)
c41d4ac4 6119{
6791875e 6120 struct r5conf *conf;
c41d4ac4
N		 6121 unsigned long new;
6122 int err;
6123
6124 if (len >= PAGE_SIZE)
6125 return -EINVAL;
b29bebd6 6126 if (kstrtoul(page, 10, &new))
c41d4ac4 6127 return -EINVAL;
6791875e 6128 err = mddev_lock(mddev);
c41d4ac4
N		 6129 if (err)
6130 return err;
6791875e
N		 6131 conf = mddev->private;
6132 if (!conf)
6133 err = -ENODEV;
6134 else
6135 err = raid5_set_cache_size(mddev, new);
6136 mddev_unlock(mddev);
6137
6138 return err ?: len;
3f294f4f 6139}
007583c9 6140
96de1e66
N		 6141static struct md_sysfs_entry
6142raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
6143 raid5_show_stripe_cache_size,
6144 raid5_store_stripe_cache_size);
3f294f4f 6145
d06f191f
MS		 6146static ssize_t
6147raid5_show_rmw_level(struct mddev *mddev, char *page)
6148{
6149 struct r5conf *conf = mddev->private;
6150 if (conf)
6151 return sprintf(page, "%d\n", conf->rmw_level);
6152 else
6153 return 0;
6154}
6155
6156static ssize_t
6157raid5_store_rmw_level(struct mddev *mddev, const char *page, size_t len)
6158{
6159 struct r5conf *conf = mddev->private;
6160 unsigned long new;
6161
6162 if (!conf)
6163 return -ENODEV;
6164
6165 if (len >= PAGE_SIZE)
6166 return -EINVAL;
6167
6168 if (kstrtoul(page, 10, &new))
6169 return -EINVAL;
6170
6171 if (new != PARITY_DISABLE_RMW && !raid6_call.xor_syndrome)
6172 return -EINVAL;
6173
6174 if (new != PARITY_DISABLE_RMW &&
6175 new != PARITY_ENABLE_RMW &&
6176 new != PARITY_PREFER_RMW)
6177 return -EINVAL;
6178
6179 conf->rmw_level = new;
6180 return len;
6181}
6182
6183static struct md_sysfs_entry
6184raid5_rmw_level = __ATTR(rmw_level, S_IRUGO | S_IWUSR,
6185 raid5_show_rmw_level,
6186 raid5_store_rmw_level);
6187
6188
8b3e6cdc 6189static ssize_t
fd01b88c 6190raid5_show_preread_threshold(struct mddev *mddev, char *page)
8b3e6cdc 6191{
7b1485ba
N		 6192 struct r5conf *conf;
6193 int ret = 0;
6194 spin_lock(&mddev->lock);
6195 conf = mddev->private;
8b3e6cdc 6196 if (conf)
7b1485ba
N		 6197 ret = sprintf(page, "%d\n", conf->bypass_threshold);
6198 spin_unlock(&mddev->lock);
6199 return ret;
8b3e6cdc
DW		 6200}
6201
6202static ssize_t
fd01b88c 6203raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len)
8b3e6cdc 6204{
6791875e 6205 struct r5conf *conf;
4ef197d8 6206 unsigned long new;
6791875e
N		 6207 int err;
6208
8b3e6cdc
DW		 6209 if (len >= PAGE_SIZE)
6210 return -EINVAL;
b29bebd6 6211 if (kstrtoul(page, 10, &new))
8b3e6cdc 6212 return -EINVAL;
6791875e
N		 6213
6214 err = mddev_lock(mddev);
6215 if (err)
6216 return err;
6217 conf = mddev->private;
6218 if (!conf)
6219 err = -ENODEV;
edbe83ab 6220 else if (new > conf->min_nr_stripes)
6791875e
N		 6221 err = -EINVAL;
6222 else
6223 conf->bypass_threshold = new;
6224 mddev_unlock(mddev);
6225 return err ?: len;
8b3e6cdc
DW		 6226}
6227
6228static struct md_sysfs_entry
6229raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
6230 S_IRUGO | S_IWUSR,
6231 raid5_show_preread_threshold,
6232 raid5_store_preread_threshold);
6233
d592a996
SL		 6234static ssize_t
6235raid5_show_skip_copy(struct mddev *mddev, char *page)
6236{
7b1485ba
N		 6237 struct r5conf *conf;
6238 int ret = 0;
6239 spin_lock(&mddev->lock);
6240 conf = mddev->private;
d592a996 6241 if (conf)
7b1485ba
N		 6242 ret = sprintf(page, "%d\n", conf->skip_copy);
6243 spin_unlock(&mddev->lock);
6244 return ret;
d592a996
SL		 6245}
6246
6247static ssize_t
6248raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len)
6249{
6791875e 6250 struct r5conf *conf;
d592a996 6251 unsigned long new;
6791875e
N		 6252 int err;
6253
d592a996
SL		 6254 if (len >= PAGE_SIZE)
6255 return -EINVAL;
d592a996
SL		 6256 if (kstrtoul(page, 10, &new))
6257 return -EINVAL;
6258 new = !!new;
6791875e
N		 6259
6260 err = mddev_lock(mddev);
6261 if (err)
6262 return err;
6263 conf = mddev->private;
6264 if (!conf)
6265 err = -ENODEV;
6266 else if (new != conf->skip_copy) {
6267 mddev_suspend(mddev);
6268 conf->skip_copy = new;
6269 if (new)
6270 mddev->queue->backing_dev_info.capabilities |=
6271 BDI_CAP_STABLE_WRITES;
6272 else
6273 mddev->queue->backing_dev_info.capabilities &=
6274 ~BDI_CAP_STABLE_WRITES;
6275 mddev_resume(mddev);
6276 }
6277 mddev_unlock(mddev);
6278 return err ?: len;
d592a996
SL		 6279}
6280
6281static struct md_sysfs_entry
6282raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR,
6283 raid5_show_skip_copy,
6284 raid5_store_skip_copy);
6285
3f294f4f 6286static ssize_t
fd01b88c 6287stripe_cache_active_show(struct mddev *mddev, char *page)
3f294f4f 6288{
d1688a6d 6289 struct r5conf *conf = mddev->private;
96de1e66
N		 6290 if (conf)
6291 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
6292 else
6293 return 0;
3f294f4f
N		 6294}
6295
96de1e66
N		 6296static struct md_sysfs_entry
6297raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
3f294f4f 6298
b721420e
SL		 6299static ssize_t
6300raid5_show_group_thread_cnt(struct mddev *mddev, char *page)
6301{
7b1485ba
N		 6302 struct r5conf *conf;
6303 int ret = 0;
6304 spin_lock(&mddev->lock);
6305 conf = mddev->private;
b721420e 6306 if (conf)
7b1485ba
N		 6307 ret = sprintf(page, "%d\n", conf->worker_cnt_per_group);
6308 spin_unlock(&mddev->lock);
6309 return ret;
b721420e
SL		 6310}
6311
60aaf933 6312static int alloc_thread_groups(struct r5conf *conf, int cnt,
6313 int *group_cnt,
6314 int *worker_cnt_per_group,
6315 struct r5worker_group **worker_groups);
b721420e
SL		 6316static ssize_t
6317raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len)
6318{
6791875e 6319 struct r5conf *conf;
b721420e
SL		 6320 unsigned long new;
6321 int err;
60aaf933 6322 struct r5worker_group *new_groups, *old_groups;
6323 int group_cnt, worker_cnt_per_group;
b721420e
SL		 6324
6325 if (len >= PAGE_SIZE)
6326 return -EINVAL;
b721420e
SL		 6327 if (kstrtoul(page, 10, &new))
6328 return -EINVAL;
6329
6791875e
N		 6330 err = mddev_lock(mddev);
6331 if (err)
6332 return err;
6333 conf = mddev->private;
6334 if (!conf)
6335 err = -ENODEV;
6336 else if (new != conf->worker_cnt_per_group) {
6337 mddev_suspend(mddev);
b721420e 6338
6791875e
N		 6339 old_groups = conf->worker_groups;
6340 if (old_groups)
6341 flush_workqueue(raid5_wq);
d206dcfa 6342
6791875e
N		 6343 err = alloc_thread_groups(conf, new,
6344 &group_cnt, &worker_cnt_per_group,
6345 &new_groups);
6346 if (!err) {
6347 spin_lock_irq(&conf->device_lock);
6348 conf->group_cnt = group_cnt;
6349 conf->worker_cnt_per_group = worker_cnt_per_group;
6350 conf->worker_groups = new_groups;
6351 spin_unlock_irq(&conf->device_lock);
b721420e 6352
6791875e
N		 6353 if (old_groups)
6354 kfree(old_groups[0].workers);
6355 kfree(old_groups);
6356 }
6357 mddev_resume(mddev);
b721420e 6358 }
6791875e 6359 mddev_unlock(mddev);
b721420e 6360
6791875e 6361 return err ?: len;
b721420e
SL		 6362}
6363
6364static struct md_sysfs_entry
6365raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR,
6366 raid5_show_group_thread_cnt,
6367 raid5_store_group_thread_cnt);
6368
007583c9 6369static struct attribute *raid5_attrs[] = {
3f294f4f
N		 6370 &raid5_stripecache_size.attr,
6371 &raid5_stripecache_active.attr,
8b3e6cdc 6372 &raid5_preread_bypass_threshold.attr,
b721420e 6373 &raid5_group_thread_cnt.attr,
d592a996 6374 &raid5_skip_copy.attr,
d06f191f 6375 &raid5_rmw_level.attr,
2c7da14b 6376 &r5c_journal_mode.attr,
3f294f4f
N		 6377 NULL,
6378};
007583c9
N		 6379static struct attribute_group raid5_attrs_group = {
6380 .name = NULL,
6381 .attrs = raid5_attrs,
3f294f4f
N		 6382};
6383
60aaf933 6384static int alloc_thread_groups(struct r5conf *conf, int cnt,
6385 int *group_cnt,
6386 int *worker_cnt_per_group,
6387 struct r5worker_group **worker_groups)
851c30c9 6388{
566c09c5 6389 int i, j, k;
851c30c9
SL		 6390 ssize_t size;
6391 struct r5worker *workers;
6392
60aaf933 6393 *worker_cnt_per_group = cnt;
851c30c9 6394 if (cnt == 0) {
60aaf933 6395 *group_cnt = 0;
6396 *worker_groups = NULL;
851c30c9
SL		 6397 return 0;
6398 }
60aaf933 6399 *group_cnt = num_possible_nodes();
851c30c9 6400 size = sizeof(struct r5worker) * cnt;
60aaf933 6401 workers = kzalloc(size * *group_cnt, GFP_NOIO);
6402 *worker_groups = kzalloc(sizeof(struct r5worker_group) *
6403 *group_cnt, GFP_NOIO);
6404 if (!*worker_groups || !workers) {
851c30c9 6405 kfree(workers);
60aaf933 6406 kfree(*worker_groups);
851c30c9
SL		 6407 return -ENOMEM;
6408 }
6409
60aaf933 6410 for (i = 0; i < *group_cnt; i++) {
851c30c9
SL		 6411 struct r5worker_group *group;
6412
0c775d52 6413 group = &(*worker_groups)[i];
851c30c9
SL		 6414 INIT_LIST_HEAD(&group->handle_list);
6415 group->conf = conf;
6416 group->workers = workers + i * cnt;
6417
6418 for (j = 0; j < cnt; j++) {
566c09c5
SL		 6419 struct r5worker *worker = group->workers + j;
6420 worker->group = group;
6421 INIT_WORK(&worker->work, raid5_do_work);
6422
6423 for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++)
6424 INIT_LIST_HEAD(worker->temp_inactive_list + k);
851c30c9
SL		 6425 }
6426 }
6427
6428 return 0;
6429}
6430
6431static void free_thread_groups(struct r5conf *conf)
6432{
6433 if (conf->worker_groups)
6434 kfree(conf->worker_groups[0].workers);
6435 kfree(conf->worker_groups);
6436 conf->worker_groups = NULL;
6437}
6438
80c3a6ce 6439static sector_t
fd01b88c 6440raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks)
80c3a6ce 6441{
d1688a6d 6442 struct r5conf *conf = mddev->private;
80c3a6ce
DW		 6443
6444 if (!sectors)
6445 sectors = mddev->dev_sectors;
5e5e3e78 6446 if (!raid_disks)
7ec05478 6447 /* size is defined by the smallest of previous and new size */
5e5e3e78 6448 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
80c3a6ce 6449
3cb5edf4
N		 6450 sectors &= ~((sector_t)conf->chunk_sectors - 1);
6451 sectors &= ~((sector_t)conf->prev_chunk_sectors - 1);
80c3a6ce
DW		 6452 return sectors * (raid_disks - conf->max_degraded);
6453}
6454
789b5e03
ON		 6455static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
6456{
6457 safe_put_page(percpu->spare_page);
46d5b785 6458 if (percpu->scribble)
6459 flex_array_free(percpu->scribble);
789b5e03
ON		 6460 percpu->spare_page = NULL;
6461 percpu->scribble = NULL;
6462}
6463
6464static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
6465{
6466 if (conf->level == 6 && !percpu->spare_page)
6467 percpu->spare_page = alloc_page(GFP_KERNEL);
6468 if (!percpu->scribble)
46d5b785 6469 percpu->scribble = scribble_alloc(max(conf->raid_disks,
738a2738
N		 6470 conf->previous_raid_disks),
6471 max(conf->chunk_sectors,
6472 conf->prev_chunk_sectors)
6473 / STRIPE_SECTORS,
6474 GFP_KERNEL);
789b5e03
ON		 6475
6476 if (!percpu->scribble || (conf->level == 6 && !percpu->spare_page)) {
6477 free_scratch_buffer(conf, percpu);
6478 return -ENOMEM;
6479 }
6480
6481 return 0;
6482}
6483
29c6d1bb 6484static int raid456_cpu_dead(unsigned int cpu, struct hlist_node *node)
36d1c647 6485{
29c6d1bb
SAS		 6486 struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node);
6487
6488 free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
6489 return 0;
6490}
36d1c647 6491
29c6d1bb
SAS		 6492static void raid5_free_percpu(struct r5conf *conf)
6493{
36d1c647
DW		 6494 if (!conf->percpu)
6495 return;
6496
29c6d1bb 6497 cpuhp_state_remove_instance(CPUHP_MD_RAID5_PREPARE, &conf->node);
36d1c647
DW		 6498 free_percpu(conf->percpu);
6499}
6500
d1688a6d 6501static void free_conf(struct r5conf *conf)
95fc17aa 6502{
d7bd398e
SL		 6503 int i;
6504
5c7e81c3
SL		 6505 if (conf->log)
6506 r5l_exit_log(conf->log);
30c89465 6507 if (conf->shrinker.nr_deferred)
edbe83ab 6508 unregister_shrinker(&conf->shrinker);
5c7e81c3 6509
851c30c9 6510 free_thread_groups(conf);
95fc17aa 6511 shrink_stripes(conf);
36d1c647 6512 raid5_free_percpu(conf);
d7bd398e
SL		 6513 for (i = 0; i < conf->pool_size; i++)
6514 if (conf->disks[i].extra_page)
6515 put_page(conf->disks[i].extra_page);
95fc17aa
DW		 6516 kfree(conf->disks);
6517 kfree(conf->stripe_hashtbl);
6518 kfree(conf);
6519}
6520
29c6d1bb 6521static int raid456_cpu_up_prepare(unsigned int cpu, struct hlist_node *node)
36d1c647 6522{
29c6d1bb 6523 struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node);
36d1c647
DW		 6524 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
6525
29c6d1bb 6526 if (alloc_scratch_buffer(conf, percpu)) {
cc6167b4
N		 6527 pr_warn("%s: failed memory allocation for cpu%u\n",
6528 __func__, cpu);
29c6d1bb 6529 return -ENOMEM;
36d1c647 6530 }
29c6d1bb 6531 return 0;
36d1c647 6532}
36d1c647 6533
d1688a6d 6534static int raid5_alloc_percpu(struct r5conf *conf)
36d1c647 6535{
789b5e03 6536 int err = 0;
36d1c647 6537
789b5e03
ON		 6538 conf->percpu = alloc_percpu(struct raid5_percpu);
6539 if (!conf->percpu)
36d1c647 6540 return -ENOMEM;
789b5e03 6541
29c6d1bb 6542 err = cpuhp_state_add_instance(CPUHP_MD_RAID5_PREPARE, &conf->node);
27a353c0
SL		 6543 if (!err) {
6544 conf->scribble_disks = max(conf->raid_disks,
6545 conf->previous_raid_disks);
6546 conf->scribble_sectors = max(conf->chunk_sectors,
6547 conf->prev_chunk_sectors);
6548 }
36d1c647
DW		 6549 return err;
6550}
6551
edbe83ab
N		 6552static unsigned long raid5_cache_scan(struct shrinker *shrink,
6553 struct shrink_control *sc)
6554{
6555 struct r5conf *conf = container_of(shrink, struct r5conf, shrinker);
2d5b569b
N		 6556 unsigned long ret = SHRINK_STOP;
6557
6558 if (mutex_trylock(&conf->cache_size_mutex)) {
6559 ret= 0;
49895bcc
N		 6560 while (ret < sc->nr_to_scan &&
6561 conf->max_nr_stripes > conf->min_nr_stripes) {
2d5b569b
N		 6562 if (drop_one_stripe(conf) == 0) {
6563 ret = SHRINK_STOP;
6564 break;
6565 }
6566 ret++;
6567 }
6568 mutex_unlock(&conf->cache_size_mutex);
edbe83ab
N		 6569 }
6570 return ret;
6571}
6572
6573static unsigned long raid5_cache_count(struct shrinker *shrink,
6574 struct shrink_control *sc)
6575{
6576 struct r5conf *conf = container_of(shrink, struct r5conf, shrinker);
6577
6578 if (conf->max_nr_stripes < conf->min_nr_stripes)
6579 /* unlikely, but not impossible */
6580 return 0;
6581 return conf->max_nr_stripes - conf->min_nr_stripes;
6582}
6583
d1688a6d 6584static struct r5conf *setup_conf(struct mddev *mddev)
1da177e4 6585{
d1688a6d 6586 struct r5conf *conf;
5e5e3e78 6587 int raid_disk, memory, max_disks;
3cb03002 6588 struct md_rdev *rdev;
1da177e4 6589 struct disk_info *disk;
0232605d 6590 char pers_name[6];
566c09c5 6591 int i;
60aaf933 6592 int group_cnt, worker_cnt_per_group;
6593 struct r5worker_group *new_group;
1da177e4 6594
91adb564
N		 6595 if (mddev->new_level != 5
6596 && mddev->new_level != 4
6597 && mddev->new_level != 6) {
cc6167b4
N		 6598 pr_warn("md/raid:%s: raid level not set to 4/5/6 (%d)\n",
6599 mdname(mddev), mddev->new_level);
91adb564 6600 return ERR_PTR(-EIO);
1da177e4 6601 }
91adb564
N		 6602 if ((mddev->new_level == 5
6603 && !algorithm_valid_raid5(mddev->new_layout)) ||
6604 (mddev->new_level == 6
6605 && !algorithm_valid_raid6(mddev->new_layout))) {
cc6167b4
N		 6606 pr_warn("md/raid:%s: layout %d not supported\n",
6607 mdname(mddev), mddev->new_layout);
91adb564 6608 return ERR_PTR(-EIO);
99c0fb5f 6609 }
91adb564 6610 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
cc6167b4
N		 6611 pr_warn("md/raid:%s: not enough configured devices (%d, minimum 4)\n",
6612 mdname(mddev), mddev->raid_disks);
91adb564 6613 return ERR_PTR(-EINVAL);
4bbf3771
N		 6614 }
6615
664e7c41
AN		 6616 if (!mddev->new_chunk_sectors ||
6617 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
6618 !is_power_of_2(mddev->new_chunk_sectors)) {
cc6167b4
N		 6619 pr_warn("md/raid:%s: invalid chunk size %d\n",
6620 mdname(mddev), mddev->new_chunk_sectors << 9);
91adb564 6621 return ERR_PTR(-EINVAL);
f6705578
N		 6622 }
6623
d1688a6d 6624 conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL);
91adb564 6625 if (conf == NULL)
1da177e4 6626 goto abort;
851c30c9 6627 /* Don't enable multi-threading by default*/
60aaf933 6628 if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group,
6629 &new_group)) {
6630 conf->group_cnt = group_cnt;
6631 conf->worker_cnt_per_group = worker_cnt_per_group;
6632 conf->worker_groups = new_group;
6633 } else
851c30c9 6634 goto abort;
f5efd45a 6635 spin_lock_init(&conf->device_lock);
c46501b2 6636 seqcount_init(&conf->gen_lock);
2d5b569b 6637 mutex_init(&conf->cache_size_mutex);
b1b46486 6638 init_waitqueue_head(&conf->wait_for_quiescent);
6ab2a4b8 6639 init_waitqueue_head(&conf->wait_for_stripe);
f5efd45a
DW		 6640 init_waitqueue_head(&conf->wait_for_overlap);
6641 INIT_LIST_HEAD(&conf->handle_list);
6642 INIT_LIST_HEAD(&conf->hold_list);
6643 INIT_LIST_HEAD(&conf->delayed_list);
6644 INIT_LIST_HEAD(&conf->bitmap_list);
c3cce6cd 6645 bio_list_init(&conf->return_bi);
773ca82f 6646 init_llist_head(&conf->released_stripes);
f5efd45a
DW		 6647 atomic_set(&conf->active_stripes, 0);
6648 atomic_set(&conf->preread_active_stripes, 0);
6649 atomic_set(&conf->active_aligned_reads, 0);
6650 conf->bypass_threshold = BYPASS_THRESHOLD;
d890fa2b 6651 conf->recovery_disabled = mddev->recovery_disabled - 1;
91adb564
N		 6652
6653 conf->raid_disks = mddev->raid_disks;
6654 if (mddev->reshape_position == MaxSector)
6655 conf->previous_raid_disks = mddev->raid_disks;
6656 else
f6705578 6657 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
5e5e3e78 6658 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
f6705578 6659
5e5e3e78 6660 conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
b55e6bfc 6661 GFP_KERNEL);
d7bd398e 6662
b55e6bfc
N		 6663 if (!conf->disks)
6664 goto abort;
9ffae0cf 6665
d7bd398e
SL		 6666 for (i = 0; i < max_disks; i++) {
6667 conf->disks[i].extra_page = alloc_page(GFP_KERNEL);
6668 if (!conf->disks[i].extra_page)
6669 goto abort;
6670 }
6671
1da177e4
LT		 6672 conf->mddev = mddev;
6673
fccddba0 6674 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
1da177e4 6675 goto abort;
1da177e4 6676
566c09c5
SL		 6677 /* We init hash_locks[0] separately to that it can be used
6678 * as the reference lock in the spin_lock_nest_lock() call
6679 * in lock_all_device_hash_locks_irq in order to convince
6680 * lockdep that we know what we are doing.
6681 */
6682 spin_lock_init(conf->hash_locks);
6683 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
6684 spin_lock_init(conf->hash_locks + i);
6685
6686 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
6687 INIT_LIST_HEAD(conf->inactive_list + i);
6688
6689 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
6690 INIT_LIST_HEAD(conf->temp_inactive_list + i);
6691
1e6d690b
SL		 6692 atomic_set(&conf->r5c_cached_full_stripes, 0);
6693 INIT_LIST_HEAD(&conf->r5c_full_stripe_list);
6694 atomic_set(&conf->r5c_cached_partial_stripes, 0);
6695 INIT_LIST_HEAD(&conf->r5c_partial_stripe_list);
6696
36d1c647 6697 conf->level = mddev->new_level;
46d5b785 6698 conf->chunk_sectors = mddev->new_chunk_sectors;
36d1c647
DW		 6699 if (raid5_alloc_percpu(conf) != 0)
6700 goto abort;
6701
0c55e022 6702 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
1da177e4 6703
dafb20fa 6704 rdev_for_each(rdev, mddev) {
1da177e4 6705 raid_disk = rdev->raid_disk;
5e5e3e78 6706 if (raid_disk >= max_disks
f2076e7d 6707 || raid_disk < 0 || test_bit(Journal, &rdev->flags))
1da177e4
LT		 6708 continue;
6709 disk = conf->disks + raid_disk;
6710
17045f52
N		 6711 if (test_bit(Replacement, &rdev->flags)) {
6712 if (disk->replacement)
6713 goto abort;
6714 disk->replacement = rdev;
6715 } else {
6716 if (disk->rdev)
6717 goto abort;
6718 disk->rdev = rdev;
6719 }
1da177e4 6720
b2d444d7 6721 if (test_bit(In_sync, &rdev->flags)) {
1da177e4 6722 char b[BDEVNAME_SIZE];
cc6167b4
N		 6723 pr_info("md/raid:%s: device %s operational as raid disk %d\n",
6724 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
d6b212f4 6725 } else if (rdev->saved_raid_disk != raid_disk)
8c2e870a
NB		 6726 /* Cannot rely on bitmap to complete recovery */
6727 conf->fullsync = 1;
1da177e4
LT		 6728 }
6729
91adb564 6730 conf->level = mddev->new_level;
584acdd4 6731 if (conf->level == 6) {
16a53ecc 6732 conf->max_degraded = 2;
584acdd4
MS		 6733 if (raid6_call.xor_syndrome)
6734 conf->rmw_level = PARITY_ENABLE_RMW;
6735 else
6736 conf->rmw_level = PARITY_DISABLE_RMW;
6737 } else {
16a53ecc 6738 conf->max_degraded = 1;
584acdd4
MS		 6739 conf->rmw_level = PARITY_ENABLE_RMW;
6740 }
91adb564 6741 conf->algorithm = mddev->new_layout;
fef9c61f 6742 conf->reshape_progress = mddev->reshape_position;
e183eaed 6743 if (conf->reshape_progress != MaxSector) {
09c9e5fa 6744 conf->prev_chunk_sectors = mddev->chunk_sectors;
e183eaed 6745 conf->prev_algo = mddev->layout;
5cac6bcb
N		 6746 } else {
6747 conf->prev_chunk_sectors = conf->chunk_sectors;
6748 conf->prev_algo = conf->algorithm;
e183eaed 6749 }
1da177e4 6750
edbe83ab 6751 conf->min_nr_stripes = NR_STRIPES;
ad5b0f76
SL		 6752 if (mddev->reshape_position != MaxSector) {
6753 int stripes = max_t(int,
6754 ((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4,
6755 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4);
6756 conf->min_nr_stripes = max(NR_STRIPES, stripes);
6757 if (conf->min_nr_stripes != NR_STRIPES)
cc6167b4 6758 pr_info("md/raid:%s: force stripe size %d for reshape\n",
ad5b0f76
SL		 6759 mdname(mddev), conf->min_nr_stripes);
6760 }
edbe83ab 6761 memory = conf->min_nr_stripes * (sizeof(struct stripe_head) +
5e5e3e78 6762 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
4bda556a 6763 atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS);
edbe83ab 6764 if (grow_stripes(conf, conf->min_nr_stripes)) {
cc6167b4
N		 6765 pr_warn("md/raid:%s: couldn't allocate %dkB for buffers\n",
6766 mdname(mddev), memory);
91adb564
N		 6767 goto abort;
6768 } else
cc6167b4 6769 pr_debug("md/raid:%s: allocated %dkB\n", mdname(mddev), memory);
edbe83ab
N		 6770 /*
6771 * Losing a stripe head costs more than the time to refill it,
6772 * it reduces the queue depth and so can hurt throughput.
6773 * So set it rather large, scaled by number of devices.
6774 */
6775 conf->shrinker.seeks = DEFAULT_SEEKS * conf->raid_disks * 4;
6776 conf->shrinker.scan_objects = raid5_cache_scan;
6777 conf->shrinker.count_objects = raid5_cache_count;
6778 conf->shrinker.batch = 128;
6779 conf->shrinker.flags = 0;
6a0f53ff 6780 if (register_shrinker(&conf->shrinker)) {
cc6167b4
N		 6781 pr_warn("md/raid:%s: couldn't register shrinker.\n",
6782 mdname(mddev));
6a0f53ff
CY		 6783 goto abort;
6784 }
1da177e4 6785
0232605d
N		 6786 sprintf(pers_name, "raid%d", mddev->new_level);
6787 conf->thread = md_register_thread(raid5d, mddev, pers_name);
91adb564 6788 if (!conf->thread) {
cc6167b4
N		 6789 pr_warn("md/raid:%s: couldn't allocate thread.\n",
6790 mdname(mddev));
16a53ecc
N		 6791 goto abort;
6792 }
91adb564
N		 6793
6794 return conf;
6795
6796 abort:
6797 if (conf) {
95fc17aa 6798 free_conf(conf);
91adb564
N		 6799 return ERR_PTR(-EIO);
6800 } else
6801 return ERR_PTR(-ENOMEM);
6802}
6803
c148ffdc
N		 6804static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
6805{
6806 switch (algo) {
6807 case ALGORITHM_PARITY_0:
6808 if (raid_disk < max_degraded)
6809 return 1;
6810 break;
6811 case ALGORITHM_PARITY_N:
6812 if (raid_disk >= raid_disks - max_degraded)
6813 return 1;
6814 break;
6815 case ALGORITHM_PARITY_0_6:
f72ffdd6 6816 if (raid_disk == 0 ||
c148ffdc
N		 6817 raid_disk == raid_disks - 1)
6818 return 1;
6819 break;
6820 case ALGORITHM_LEFT_ASYMMETRIC_6:
6821 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6822 case ALGORITHM_LEFT_SYMMETRIC_6:
6823 case ALGORITHM_RIGHT_SYMMETRIC_6:
6824 if (raid_disk == raid_disks - 1)
6825 return 1;
6826 }
6827 return 0;
6828}
6829
849674e4 6830static int raid5_run(struct mddev *mddev)
91adb564 6831{
d1688a6d 6832 struct r5conf *conf;
9f7c2220 6833 int working_disks = 0;
c148ffdc 6834 int dirty_parity_disks = 0;
3cb03002 6835 struct md_rdev *rdev;
713cf5a6 6836 struct md_rdev *journal_dev = NULL;
c148ffdc 6837 sector_t reshape_offset = 0;
17045f52 6838 int i;
b5254dd5
N		 6839 long long min_offset_diff = 0;
6840 int first = 1;
91adb564 6841
8c6ac868 6842 if (mddev->recovery_cp != MaxSector)
cc6167b4
N		 6843 pr_notice("md/raid:%s: not clean -- starting background reconstruction\n",
6844 mdname(mddev));
b5254dd5
N		 6845
6846 rdev_for_each(rdev, mddev) {
6847 long long diff;
713cf5a6 6848
f2076e7d 6849 if (test_bit(Journal, &rdev->flags)) {
713cf5a6 6850 journal_dev = rdev;
f2076e7d
SL		 6851 continue;
6852 }
b5254dd5
N		 6853 if (rdev->raid_disk < 0)
6854 continue;
6855 diff = (rdev->new_data_offset - rdev->data_offset);
6856 if (first) {
6857 min_offset_diff = diff;
6858 first = 0;
6859 } else if (mddev->reshape_backwards &&
6860 diff < min_offset_diff)
6861 min_offset_diff = diff;
6862 else if (!mddev->reshape_backwards &&
6863 diff > min_offset_diff)
6864 min_offset_diff = diff;
6865 }
6866
91adb564
N		 6867 if (mddev->reshape_position != MaxSector) {
6868 /* Check that we can continue the reshape.
b5254dd5
N		 6869 * Difficulties arise if the stripe we would write to
6870 * next is at or after the stripe we would read from next.
6871 * For a reshape that changes the number of devices, this
6872 * is only possible for a very short time, and mdadm makes
6873 * sure that time appears to have past before assembling
6874 * the array. So we fail if that time hasn't passed.
6875 * For a reshape that keeps the number of devices the same
6876 * mdadm must be monitoring the reshape can keeping the
6877 * critical areas read-only and backed up. It will start
6878 * the array in read-only mode, so we check for that.
91adb564
N		 6879 */
6880 sector_t here_new, here_old;
6881 int old_disks;
18b00334 6882 int max_degraded = (mddev->level == 6 ? 2 : 1);
05256d98
N		 6883 int chunk_sectors;
6884 int new_data_disks;
91adb564 6885
713cf5a6 6886 if (journal_dev) {
cc6167b4
N		 6887 pr_warn("md/raid:%s: don't support reshape with journal - aborting.\n",
6888 mdname(mddev));
713cf5a6
SL		 6889 return -EINVAL;
6890 }
6891
88ce4930 6892 if (mddev->new_level != mddev->level) {
cc6167b4
N		 6893 pr_warn("md/raid:%s: unsupported reshape required - aborting.\n",
6894 mdname(mddev));
91adb564
N		 6895 return -EINVAL;
6896 }
91adb564
N		 6897 old_disks = mddev->raid_disks - mddev->delta_disks;
6898 /* reshape_position must be on a new-stripe boundary, and one
6899 * further up in new geometry must map after here in old
6900 * geometry.
05256d98
N		 6901 * If the chunk sizes are different, then as we perform reshape
6902 * in units of the largest of the two, reshape_position needs
6903 * be a multiple of the largest chunk size times new data disks.
91adb564
N		 6904 */
6905 here_new = mddev->reshape_position;
05256d98
N		 6906 chunk_sectors = max(mddev->chunk_sectors, mddev->new_chunk_sectors);
6907 new_data_disks = mddev->raid_disks - max_degraded;
6908 if (sector_div(here_new, chunk_sectors * new_data_disks)) {
cc6167b4
N		 6909 pr_warn("md/raid:%s: reshape_position not on a stripe boundary\n",
6910 mdname(mddev));
91adb564
N		 6911 return -EINVAL;
6912 }
05256d98 6913 reshape_offset = here_new * chunk_sectors;
91adb564
N		 6914 /* here_new is the stripe we will write to */
6915 here_old = mddev->reshape_position;
05256d98 6916 sector_div(here_old, chunk_sectors * (old_disks-max_degraded));
91adb564
N		 6917 /* here_old is the first stripe that we might need to read
6918 * from */
67ac6011
N		 6919 if (mddev->delta_disks == 0) {
6920 /* We cannot be sure it is safe to start an in-place
b5254dd5 6921 * reshape. It is only safe if user-space is monitoring
67ac6011
N		 6922 * and taking constant backups.
6923 * mdadm always starts a situation like this in
6924 * readonly mode so it can take control before
6925 * allowing any writes. So just check for that.
6926 */
b5254dd5
N		 6927 if (abs(min_offset_diff) >= mddev->chunk_sectors &&
6928 abs(min_offset_diff) >= mddev->new_chunk_sectors)
6929 /* not really in-place - so OK */;
6930 else if (mddev->ro == 0) {
cc6167b4
N		 6931 pr_warn("md/raid:%s: in-place reshape must be started in read-only mode - aborting\n",
6932 mdname(mddev));
67ac6011
N		 6933 return -EINVAL;
6934 }
2c810cdd 6935 } else if (mddev->reshape_backwards
05256d98
N		 6936 ? (here_new * chunk_sectors + min_offset_diff <=
6937 here_old * chunk_sectors)
6938 : (here_new * chunk_sectors >=
6939 here_old * chunk_sectors + (-min_offset_diff))) {
91adb564 6940 /* Reading from the same stripe as writing to - bad */
cc6167b4
N		 6941 pr_warn("md/raid:%s: reshape_position too early for auto-recovery - aborting.\n",
6942 mdname(mddev));
91adb564
N		 6943 return -EINVAL;
6944 }
cc6167b4 6945 pr_debug("md/raid:%s: reshape will continue\n", mdname(mddev));
91adb564
N		 6946 /* OK, we should be able to continue; */
6947 } else {
6948 BUG_ON(mddev->level != mddev->new_level);
6949 BUG_ON(mddev->layout != mddev->new_layout);
664e7c41 6950 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
91adb564 6951 BUG_ON(mddev->delta_disks != 0);
1da177e4 6952 }
91adb564 6953
245f46c2
N		 6954 if (mddev->private == NULL)
6955 conf = setup_conf(mddev);
6956 else
6957 conf = mddev->private;
6958
91adb564
N		 6959 if (IS_ERR(conf))
6960 return PTR_ERR(conf);
6961
486b0f7b
SL		 6962 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
6963 if (!journal_dev) {
cc6167b4
N		 6964 pr_warn("md/raid:%s: journal disk is missing, force array readonly\n",
6965 mdname(mddev));
486b0f7b
SL		 6966 mddev->ro = 1;
6967 set_disk_ro(mddev->gendisk, 1);
6968 } else if (mddev->recovery_cp == MaxSector)
6969 set_bit(MD_JOURNAL_CLEAN, &mddev->flags);
7dde2ad3
SL		 6970 }
6971
b5254dd5 6972 conf->min_offset_diff = min_offset_diff;
91adb564
N		 6973 mddev->thread = conf->thread;
6974 conf->thread = NULL;
6975 mddev->private = conf;
6976
17045f52
N		 6977 for (i = 0; i < conf->raid_disks && conf->previous_raid_disks;
6978 i++) {
6979 rdev = conf->disks[i].rdev;
6980 if (!rdev && conf->disks[i].replacement) {
6981 /* The replacement is all we have yet */
6982 rdev = conf->disks[i].replacement;
6983 conf->disks[i].replacement = NULL;
6984 clear_bit(Replacement, &rdev->flags);
6985 conf->disks[i].rdev = rdev;
6986 }
6987 if (!rdev)
c148ffdc 6988 continue;
17045f52
N		 6989 if (conf->disks[i].replacement &&
6990 conf->reshape_progress != MaxSector) {
6991 /* replacements and reshape simply do not mix. */
cc6167b4 6992 pr_warn("md: cannot handle concurrent replacement and reshape.\n");
17045f52
N		 6993 goto abort;
6994 }
2f115882 6995 if (test_bit(In_sync, &rdev->flags)) {
91adb564 6996 working_disks++;
2f115882
N		 6997 continue;
6998 }
c148ffdc
N		 6999 /* This disc is not fully in-sync. However if it
7000 * just stored parity (beyond the recovery_offset),
7001 * when we don't need to be concerned about the
7002 * array being dirty.
7003 * When reshape goes 'backwards', we never have
7004 * partially completed devices, so we only need
7005 * to worry about reshape going forwards.
7006 */
7007 /* Hack because v0.91 doesn't store recovery_offset properly. */
7008 if (mddev->major_version == 0 &&
7009 mddev->minor_version > 90)
7010 rdev->recovery_offset = reshape_offset;
5026d7a9 7011
c148ffdc
N		 7012 if (rdev->recovery_offset < reshape_offset) {
7013 /* We need to check old and new layout */
7014 if (!only_parity(rdev->raid_disk,
7015 conf->algorithm,
7016 conf->raid_disks,
7017 conf->max_degraded))
7018 continue;
7019 }
7020 if (!only_parity(rdev->raid_disk,
7021 conf->prev_algo,
7022 conf->previous_raid_disks,
7023 conf->max_degraded))
7024 continue;
7025 dirty_parity_disks++;
7026 }
91adb564 7027
17045f52
N		 7028 /*
7029 * 0 for a fully functional array, 1 or 2 for a degraded array.
7030 */
908f4fbd 7031 mddev->degraded = calc_degraded(conf);
91adb564 7032
674806d6 7033 if (has_failed(conf)) {
cc6167b4 7034 pr_crit("md/raid:%s: not enough operational devices (%d/%d failed)\n",
02c2de8c 7035 mdname(mddev), mddev->degraded, conf->raid_disks);
1da177e4
LT		 7036 goto abort;
7037 }
7038
91adb564 7039 /* device size must be a multiple of chunk size */
9d8f0363 7040 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
91adb564
N		 7041 mddev->resync_max_sectors = mddev->dev_sectors;
7042
c148ffdc 7043 if (mddev->degraded > dirty_parity_disks &&
1da177e4 7044 mddev->recovery_cp != MaxSector) {
6ff8d8ec 7045 if (mddev->ok_start_degraded)
cc6167b4
N		 7046 pr_crit("md/raid:%s: starting dirty degraded array - data corruption possible.\n",
7047 mdname(mddev));
6ff8d8ec 7048 else {
cc6167b4
N		 7049 pr_crit("md/raid:%s: cannot start dirty degraded array.\n",
7050 mdname(mddev));
6ff8d8ec
N		 7051 goto abort;
7052 }
1da177e4
LT		 7053 }
7054
cc6167b4
N		 7055 pr_info("md/raid:%s: raid level %d active with %d out of %d devices, algorithm %d\n",
7056 mdname(mddev), conf->level,
7057 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
7058 mddev->new_layout);
1da177e4
LT		 7059
7060 print_raid5_conf(conf);
7061
fef9c61f 7062 if (conf->reshape_progress != MaxSector) {
fef9c61f 7063 conf->reshape_safe = conf->reshape_progress;
f6705578
N		 7064 atomic_set(&conf->reshape_stripes, 0);
7065 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7066 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7067 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7068 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7069 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 7070 "reshape");
f6705578
N		 7071 }
7072
1da177e4 7073 /* Ok, everything is just fine now */
a64c876f
N		 7074 if (mddev->to_remove == &raid5_attrs_group)
7075 mddev->to_remove = NULL;
00bcb4ac
N		 7076 else if (mddev->kobj.sd &&
7077 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
cc6167b4
N		 7078 pr_warn("raid5: failed to create sysfs attributes for %s\n",
7079 mdname(mddev));
4a5add49 7080 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
7a5febe9 7081
4a5add49 7082 if (mddev->queue) {
9f7c2220 7083 int chunk_size;
620125f2 7084 bool discard_supported = true;
4a5add49
N		 7085 /* read-ahead size must cover two whole stripes, which
7086 * is 2 * (datadisks) * chunksize where 'n' is the
7087 * number of raid devices
7088 */
7089 int data_disks = conf->previous_raid_disks - conf->max_degraded;
7090 int stripe = data_disks *
7091 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
7092 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
7093 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
91adb564 7094
9f7c2220
N		 7095 chunk_size = mddev->chunk_sectors << 9;
7096 blk_queue_io_min(mddev->queue, chunk_size);
7097 blk_queue_io_opt(mddev->queue, chunk_size *
7098 (conf->raid_disks - conf->max_degraded));
c78afc62 7099 mddev->queue->limits.raid_partial_stripes_expensive = 1;
620125f2
SL		 7100 /*
7101 * We can only discard a whole stripe. It doesn't make sense to
7102 * discard data disk but write parity disk
7103 */
7104 stripe = stripe * PAGE_SIZE;
4ac6875e
N		 7105 /* Round up to power of 2, as discard handling
7106 * currently assumes that */
7107 while ((stripe-1) & stripe)
7108 stripe = (stripe | (stripe-1)) + 1;
620125f2
SL		 7109 mddev->queue->limits.discard_alignment = stripe;
7110 mddev->queue->limits.discard_granularity = stripe;
7111 /*
7112 * unaligned part of discard request will be ignored, so can't
8e0e99ba 7113 * guarantee discard_zeroes_data
620125f2
SL		 7114 */
7115 mddev->queue->limits.discard_zeroes_data = 0;
8f6c2e4b 7116
5026d7a9
PA		 7117 blk_queue_max_write_same_sectors(mddev->queue, 0);
7118
05616be5 7119 rdev_for_each(rdev, mddev) {
9f7c2220
N		 7120 disk_stack_limits(mddev->gendisk, rdev->bdev,
7121 rdev->data_offset << 9);
05616be5
N		 7122 disk_stack_limits(mddev->gendisk, rdev->bdev,
7123 rdev->new_data_offset << 9);
620125f2
SL		 7124 /*
7125 * discard_zeroes_data is required, otherwise data
7126 * could be lost. Consider a scenario: discard a stripe
7127 * (the stripe could be inconsistent if
7128 * discard_zeroes_data is 0); write one disk of the
7129 * stripe (the stripe could be inconsistent again
7130 * depending on which disks are used to calculate
7131 * parity); the disk is broken; The stripe data of this
7132 * disk is lost.
7133 */
7134 if (!blk_queue_discard(bdev_get_queue(rdev->bdev)) ||
7135 !bdev_get_queue(rdev->bdev)->
7136 limits.discard_zeroes_data)
7137 discard_supported = false;
8e0e99ba
N		 7138 /* Unfortunately, discard_zeroes_data is not currently
7139 * a guarantee - just a hint. So we only allow DISCARD
7140 * if the sysadmin has confirmed that only safe devices
7141 * are in use by setting a module parameter.
7142 */
7143 if (!devices_handle_discard_safely) {
7144 if (discard_supported) {
7145 pr_info("md/raid456: discard support disabled due to uncertainty.\n");
7146 pr_info("Set raid456.devices_handle_discard_safely=Y to override.\n");
7147 }
7148 discard_supported = false;
7149 }
05616be5 7150 }
620125f2
SL		 7151
7152 if (discard_supported &&
e7597e69
JS		 7153 mddev->queue->limits.max_discard_sectors >= (stripe >> 9) &&
7154 mddev->queue->limits.discard_granularity >= stripe)
620125f2
SL		 7155 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
7156 mddev->queue);
7157 else
7158 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
7159 mddev->queue);
1dffdddd
SL		 7160
7161 blk_queue_max_hw_sectors(mddev->queue, UINT_MAX);
9f7c2220 7162 }
23032a0e 7163
5c7e81c3
SL		 7164 if (journal_dev) {
7165 char b[BDEVNAME_SIZE];
7166
cc6167b4
N		 7167 pr_debug("md/raid:%s: using device %s as journal\n",
7168 mdname(mddev), bdevname(journal_dev->bdev, b));
5aabf7c4
SL		 7169 if (r5l_init_log(conf, journal_dev))
7170 goto abort;
5c7e81c3
SL		 7171 }
7172
1da177e4
LT		 7173 return 0;
7174abort:
01f96c0a 7175 md_unregister_thread(&mddev->thread);
e4f869d9
N		 7176 print_raid5_conf(conf);
7177 free_conf(conf);
1da177e4 7178 mddev->private = NULL;
cc6167b4 7179 pr_warn("md/raid:%s: failed to run raid set.\n", mdname(mddev));
1da177e4
LT		 7180 return -EIO;
7181}
7182
afa0f557 7183static void raid5_free(struct mddev *mddev, void *priv)
1da177e4 7184{
afa0f557 7185 struct r5conf *conf = priv;
1da177e4 7186
95fc17aa 7187 free_conf(conf);
a64c876f 7188 mddev->to_remove = &raid5_attrs_group;
1da177e4
LT		 7189}
7190
849674e4 7191static void raid5_status(struct seq_file *seq, struct mddev *mddev)
1da177e4 7192{
d1688a6d 7193 struct r5conf *conf = mddev->private;
1da177e4
LT		 7194 int i;
7195
9d8f0363 7196 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
3cb5edf4 7197 conf->chunk_sectors / 2, mddev->layout);
02c2de8c 7198 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
5fd13351
N		 7199 rcu_read_lock();
7200 for (i = 0; i < conf->raid_disks; i++) {
7201 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
7202 seq_printf (seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
7203 }
7204 rcu_read_unlock();
1da177e4 7205 seq_printf (seq, "]");
1da177e4
LT		 7206}
7207
d1688a6d 7208static void print_raid5_conf (struct r5conf *conf)
1da177e4
LT		 7209{
7210 int i;
7211 struct disk_info *tmp;
7212
cc6167b4 7213 pr_debug("RAID conf printout:\n");
1da177e4 7214 if (!conf) {
cc6167b4 7215 pr_debug("(conf==NULL)\n");
1da177e4
LT		 7216 return;
7217 }
cc6167b4 7218 pr_debug(" --- level:%d rd:%d wd:%d\n", conf->level,
0c55e022
N		 7219 conf->raid_disks,
7220 conf->raid_disks - conf->mddev->degraded);
1da177e4
LT		 7221
7222 for (i = 0; i < conf->raid_disks; i++) {
7223 char b[BDEVNAME_SIZE];
7224 tmp = conf->disks + i;
7225 if (tmp->rdev)
cc6167b4 7226 pr_debug(" disk %d, o:%d, dev:%s\n",
0c55e022
N		 7227 i, !test_bit(Faulty, &tmp->rdev->flags),
7228 bdevname(tmp->rdev->bdev, b));
1da177e4
LT		 7229 }
7230}
7231
fd01b88c 7232static int raid5_spare_active(struct mddev *mddev)
1da177e4
LT		 7233{
7234 int i;
d1688a6d 7235 struct r5conf *conf = mddev->private;
1da177e4 7236 struct disk_info *tmp;
6b965620
N		 7237 int count = 0;
7238 unsigned long flags;
1da177e4
LT		 7239
7240 for (i = 0; i < conf->raid_disks; i++) {
7241 tmp = conf->disks + i;
dd054fce
N		 7242 if (tmp->replacement
7243 && tmp->replacement->recovery_offset == MaxSector
7244 && !test_bit(Faulty, &tmp->replacement->flags)
7245 && !test_and_set_bit(In_sync, &tmp->replacement->flags)) {
7246 /* Replacement has just become active. */
7247 if (!tmp->rdev
7248 || !test_and_clear_bit(In_sync, &tmp->rdev->flags))
7249 count++;
7250 if (tmp->rdev) {
7251 /* Replaced device not technically faulty,
7252 * but we need to be sure it gets removed
7253 * and never re-added.
7254 */
7255 set_bit(Faulty, &tmp->rdev->flags);
7256 sysfs_notify_dirent_safe(
7257 tmp->rdev->sysfs_state);
7258 }
7259 sysfs_notify_dirent_safe(tmp->replacement->sysfs_state);
7260 } else if (tmp->rdev
70fffd0b 7261 && tmp->rdev->recovery_offset == MaxSector
b2d444d7 7262 && !test_bit(Faulty, &tmp->rdev->flags)
c04be0aa 7263 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
6b965620 7264 count++;
43c73ca4 7265 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
1da177e4
LT		 7266 }
7267 }
6b965620 7268 spin_lock_irqsave(&conf->device_lock, flags);
908f4fbd 7269 mddev->degraded = calc_degraded(conf);
6b965620 7270 spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4 7271 print_raid5_conf(conf);
6b965620 7272 return count;
1da177e4
LT		 7273}
7274
b8321b68 7275static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7276{
d1688a6d 7277 struct r5conf *conf = mddev->private;
1da177e4 7278 int err = 0;
b8321b68 7279 int number = rdev->raid_disk;
657e3e4d 7280 struct md_rdev **rdevp;
1da177e4
LT		 7281 struct disk_info *p = conf->disks + number;
7282
7283 print_raid5_conf(conf);
f6b6ec5c
SL		 7284 if (test_bit(Journal, &rdev->flags) && conf->log) {
7285 struct r5l_log *log;
c2bb6242 7286 /*
f6b6ec5c
SL		 7287 * we can't wait pending write here, as this is called in
7288 * raid5d, wait will deadlock.
c2bb6242 7289 */
f6b6ec5c
SL		 7290 if (atomic_read(&mddev->writes_pending))
7291 return -EBUSY;
7292 log = conf->log;
7293 conf->log = NULL;
7294 synchronize_rcu();
7295 r5l_exit_log(log);
7296 return 0;
c2bb6242 7297 }
657e3e4d
N		 7298 if (rdev == p->rdev)
7299 rdevp = &p->rdev;
7300 else if (rdev == p->replacement)
7301 rdevp = &p->replacement;
7302 else
7303 return 0;
7304
7305 if (number >= conf->raid_disks &&
7306 conf->reshape_progress == MaxSector)
7307 clear_bit(In_sync, &rdev->flags);
7308
7309 if (test_bit(In_sync, &rdev->flags) ||
7310 atomic_read(&rdev->nr_pending)) {
7311 err = -EBUSY;
7312 goto abort;
7313 }
7314 /* Only remove non-faulty devices if recovery
7315 * isn't possible.
7316 */
7317 if (!test_bit(Faulty, &rdev->flags) &&
7318 mddev->recovery_disabled != conf->recovery_disabled &&
7319 !has_failed(conf) &&
dd054fce 7320 (!p->replacement || p->replacement == rdev) &&
657e3e4d
N		 7321 number < conf->raid_disks) {
7322 err = -EBUSY;
7323 goto abort;
7324 }
7325 *rdevp = NULL;
d787be40
N		 7326 if (!test_bit(RemoveSynchronized, &rdev->flags)) {
7327 synchronize_rcu();
7328 if (atomic_read(&rdev->nr_pending)) {
7329 /* lost the race, try later */
7330 err = -EBUSY;
7331 *rdevp = rdev;
7332 }
7333 }
7334 if (p->replacement) {
dd054fce
N		 7335 /* We must have just cleared 'rdev' */
7336 p->rdev = p->replacement;
7337 clear_bit(Replacement, &p->replacement->flags);
7338 smp_mb(); /* Make sure other CPUs may see both as identical
7339 * but will never see neither - if they are careful
7340 */
7341 p->replacement = NULL;
7342 clear_bit(WantReplacement, &rdev->flags);
7343 } else
7344 /* We might have just removed the Replacement as faulty-
7345 * clear the bit just in case
7346 */
7347 clear_bit(WantReplacement, &rdev->flags);
1da177e4
LT		 7348abort:
7349
7350 print_raid5_conf(conf);
7351 return err;
7352}
7353
fd01b88c 7354static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7355{
d1688a6d 7356 struct r5conf *conf = mddev->private;
199050ea 7357 int err = -EEXIST;
1da177e4
LT		 7358 int disk;
7359 struct disk_info *p;
6c2fce2e
NB		 7360 int first = 0;
7361 int last = conf->raid_disks - 1;
1da177e4 7362
f6b6ec5c
SL		 7363 if (test_bit(Journal, &rdev->flags)) {
7364 char b[BDEVNAME_SIZE];
7365 if (conf->log)
7366 return -EBUSY;
7367
7368 rdev->raid_disk = 0;
7369 /*
7370 * The array is in readonly mode if journal is missing, so no
7371 * write requests running. We should be safe
7372 */
7373 r5l_init_log(conf, rdev);
cc6167b4
N		 7374 pr_debug("md/raid:%s: using device %s as journal\n",
7375 mdname(mddev), bdevname(rdev->bdev, b));
f6b6ec5c
SL		 7376 return 0;
7377 }
7f0da59b
N		 7378 if (mddev->recovery_disabled == conf->recovery_disabled)
7379 return -EBUSY;
7380
dc10c643 7381 if (rdev->saved_raid_disk < 0 && has_failed(conf))
1da177e4 7382 /* no point adding a device */
199050ea 7383 return -EINVAL;
1da177e4 7384
6c2fce2e
NB		 7385 if (rdev->raid_disk >= 0)
7386 first = last = rdev->raid_disk;
1da177e4
LT		 7387
7388 /*
16a53ecc
N		 7389 * find the disk ... but prefer rdev->saved_raid_disk
7390 * if possible.
1da177e4 7391 */
16a53ecc 7392 if (rdev->saved_raid_disk >= 0 &&
6c2fce2e 7393 rdev->saved_raid_disk >= first &&
16a53ecc 7394 conf->disks[rdev->saved_raid_disk].rdev == NULL)
5cfb22a1
N		 7395 first = rdev->saved_raid_disk;
7396
7397 for (disk = first; disk <= last; disk++) {
7bfec5f3
N		 7398 p = conf->disks + disk;
7399 if (p->rdev == NULL) {
b2d444d7 7400 clear_bit(In_sync, &rdev->flags);
1da177e4 7401 rdev->raid_disk = disk;
199050ea 7402 err = 0;
72626685
N		 7403 if (rdev->saved_raid_disk != disk)
7404 conf->fullsync = 1;
d6065f7b 7405 rcu_assign_pointer(p->rdev, rdev);
5cfb22a1 7406 goto out;
1da177e4 7407 }
5cfb22a1
N		 7408 }
7409 for (disk = first; disk <= last; disk++) {
7410 p = conf->disks + disk;
7bfec5f3
N		 7411 if (test_bit(WantReplacement, &p->rdev->flags) &&
7412 p->replacement == NULL) {
7413 clear_bit(In_sync, &rdev->flags);
7414 set_bit(Replacement, &rdev->flags);
7415 rdev->raid_disk = disk;
7416 err = 0;
7417 conf->fullsync = 1;
7418 rcu_assign_pointer(p->replacement, rdev);
7419 break;
7420 }
7421 }
5cfb22a1 7422out:
1da177e4 7423 print_raid5_conf(conf);
199050ea 7424 return err;
1da177e4
LT		 7425}
7426
fd01b88c 7427static int raid5_resize(struct mddev *mddev, sector_t sectors)
1da177e4
LT		 7428{
7429 /* no resync is happening, and there is enough space
7430 * on all devices, so we can resize.
7431 * We need to make sure resync covers any new space.
7432 * If the array is shrinking we should possibly wait until
7433 * any io in the removed space completes, but it hardly seems
7434 * worth it.
7435 */
a4a6125a 7436 sector_t newsize;
3cb5edf4
N		 7437 struct r5conf *conf = mddev->private;
7438
713cf5a6
SL		 7439 if (conf->log)
7440 return -EINVAL;
3cb5edf4 7441 sectors &= ~((sector_t)conf->chunk_sectors - 1);
a4a6125a
N		 7442 newsize = raid5_size(mddev, sectors, mddev->raid_disks);
7443 if (mddev->external_size &&
7444 mddev->array_sectors > newsize)
b522adcd 7445 return -EINVAL;
a4a6125a
N		 7446 if (mddev->bitmap) {
7447 int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0);
7448 if (ret)
7449 return ret;
7450 }
7451 md_set_array_sectors(mddev, newsize);
f233ea5c 7452 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 7453 revalidate_disk(mddev->gendisk);
b098636c
N		 7454 if (sectors > mddev->dev_sectors &&
7455 mddev->recovery_cp > mddev->dev_sectors) {
58c0fed4 7456 mddev->recovery_cp = mddev->dev_sectors;
1da177e4
LT		 7457 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7458 }
58c0fed4 7459 mddev->dev_sectors = sectors;
4b5c7ae8 7460 mddev->resync_max_sectors = sectors;
1da177e4
LT		 7461 return 0;
7462}
7463
fd01b88c 7464static int check_stripe_cache(struct mddev *mddev)
01ee22b4
N		 7465{
7466 /* Can only proceed if there are plenty of stripe_heads.
7467 * We need a minimum of one full stripe,, and for sensible progress
7468 * it is best to have about 4 times that.
7469 * If we require 4 times, then the default 256 4K stripe_heads will
7470 * allow for chunk sizes up to 256K, which is probably OK.
7471 * If the chunk size is greater, user-space should request more
7472 * stripe_heads first.
7473 */
d1688a6d 7474 struct r5conf *conf = mddev->private;
01ee22b4 7475 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
edbe83ab 7476 > conf->min_nr_stripes ||
01ee22b4 7477 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
edbe83ab 7478 > conf->min_nr_stripes) {
cc6167b4
N		 7479 pr_warn("md/raid:%s: reshape: not enough stripes. Needed %lu\n",
7480 mdname(mddev),
7481 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
7482 / STRIPE_SIZE)*4);
01ee22b4
N		 7483 return 0;
7484 }
7485 return 1;
7486}
7487
fd01b88c 7488static int check_reshape(struct mddev *mddev)
29269553 7489{
d1688a6d 7490 struct r5conf *conf = mddev->private;
29269553 7491
713cf5a6
SL		 7492 if (conf->log)
7493 return -EINVAL;
88ce4930
N		 7494 if (mddev->delta_disks == 0 &&
7495 mddev->new_layout == mddev->layout &&
664e7c41 7496 mddev->new_chunk_sectors == mddev->chunk_sectors)
50ac168a 7497 return 0; /* nothing to do */
674806d6 7498 if (has_failed(conf))
ec32a2bd 7499 return -EINVAL;
fdcfbbb6 7500 if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) {
ec32a2bd
N		 7501 /* We might be able to shrink, but the devices must
7502 * be made bigger first.
7503 * For raid6, 4 is the minimum size.
7504 * Otherwise 2 is the minimum
7505 */
7506 int min = 2;
7507 if (mddev->level == 6)
7508 min = 4;
7509 if (mddev->raid_disks + mddev->delta_disks < min)
7510 return -EINVAL;
7511 }
29269553 7512
01ee22b4 7513 if (!check_stripe_cache(mddev))
29269553 7514 return -ENOSPC;
29269553 7515
738a2738
N		 7516 if (mddev->new_chunk_sectors > mddev->chunk_sectors ||
7517 mddev->delta_disks > 0)
7518 if (resize_chunks(conf,
7519 conf->previous_raid_disks
7520 + max(0, mddev->delta_disks),
7521 max(mddev->new_chunk_sectors,
7522 mddev->chunk_sectors)
7523 ) < 0)
7524 return -ENOMEM;
e56108d6
N		 7525 return resize_stripes(conf, (conf->previous_raid_disks
7526 + mddev->delta_disks));
63c70c4f
N		 7527}
7528
fd01b88c 7529static int raid5_start_reshape(struct mddev *mddev)
63c70c4f 7530{
d1688a6d 7531 struct r5conf *conf = mddev->private;
3cb03002 7532 struct md_rdev *rdev;
63c70c4f 7533 int spares = 0;
c04be0aa 7534 unsigned long flags;
63c70c4f 7535
f416885e 7536 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
63c70c4f
N		 7537 return -EBUSY;
7538
01ee22b4
N		 7539 if (!check_stripe_cache(mddev))
7540 return -ENOSPC;
7541
30b67645
N		 7542 if (has_failed(conf))
7543 return -EINVAL;
7544
c6563a8c 7545 rdev_for_each(rdev, mddev) {
469518a3
N		 7546 if (!test_bit(In_sync, &rdev->flags)
7547 && !test_bit(Faulty, &rdev->flags))
29269553 7548 spares++;
c6563a8c 7549 }
63c70c4f 7550
f416885e 7551 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
29269553
N		 7552 /* Not enough devices even to make a degraded array
7553 * of that size
7554 */
7555 return -EINVAL;
7556
ec32a2bd
N		 7557 /* Refuse to reduce size of the array. Any reductions in
7558 * array size must be through explicit setting of array_size
7559 * attribute.
7560 */
7561 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
7562 < mddev->array_sectors) {
cc6167b4
N		 7563 pr_warn("md/raid:%s: array size must be reduced before number of disks\n",
7564 mdname(mddev));
ec32a2bd
N		 7565 return -EINVAL;
7566 }
7567
f6705578 7568 atomic_set(&conf->reshape_stripes, 0);
29269553 7569 spin_lock_irq(&conf->device_lock);
c46501b2 7570 write_seqcount_begin(&conf->gen_lock);
29269553 7571 conf->previous_raid_disks = conf->raid_disks;
63c70c4f 7572 conf->raid_disks += mddev->delta_disks;
09c9e5fa
AN		 7573 conf->prev_chunk_sectors = conf->chunk_sectors;
7574 conf->chunk_sectors = mddev->new_chunk_sectors;
88ce4930
N		 7575 conf->prev_algo = conf->algorithm;
7576 conf->algorithm = mddev->new_layout;
05616be5
N		 7577 conf->generation++;
7578 /* Code that selects data_offset needs to see the generation update
7579 * if reshape_progress has been set - so a memory barrier needed.
7580 */
7581 smp_mb();
2c810cdd 7582 if (mddev->reshape_backwards)
fef9c61f
N		 7583 conf->reshape_progress = raid5_size(mddev, 0, 0);
7584 else
7585 conf->reshape_progress = 0;
7586 conf->reshape_safe = conf->reshape_progress;
c46501b2 7587 write_seqcount_end(&conf->gen_lock);
29269553
N		 7588 spin_unlock_irq(&conf->device_lock);
7589
4d77e3ba
N		 7590 /* Now make sure any requests that proceeded on the assumption
7591 * the reshape wasn't running - like Discard or Read - have
7592 * completed.
7593 */
7594 mddev_suspend(mddev);
7595 mddev_resume(mddev);
7596
29269553
N		 7597 /* Add some new drives, as many as will fit.
7598 * We know there are enough to make the newly sized array work.
3424bf6a
N		 7599 * Don't add devices if we are reducing the number of
7600 * devices in the array. This is because it is not possible
7601 * to correctly record the "partially reconstructed" state of
7602 * such devices during the reshape and confusion could result.
29269553 7603 */
87a8dec9 7604 if (mddev->delta_disks >= 0) {
dafb20fa 7605 rdev_for_each(rdev, mddev)
87a8dec9
N		 7606 if (rdev->raid_disk < 0 &&
7607 !test_bit(Faulty, &rdev->flags)) {
7608 if (raid5_add_disk(mddev, rdev) == 0) {
87a8dec9 7609 if (rdev->raid_disk
9d4c7d87 7610 >= conf->previous_raid_disks)
87a8dec9 7611 set_bit(In_sync, &rdev->flags);
9d4c7d87 7612 else
87a8dec9 7613 rdev->recovery_offset = 0;
36fad858
NK		 7614
7615 if (sysfs_link_rdev(mddev, rdev))
87a8dec9 7616 /* Failure here is OK */;
50da0840 7617 }
87a8dec9
N		 7618 } else if (rdev->raid_disk >= conf->previous_raid_disks
7619 && !test_bit(Faulty, &rdev->flags)) {
7620 /* This is a spare that was manually added */
7621 set_bit(In_sync, &rdev->flags);
87a8dec9 7622 }
29269553 7623
87a8dec9
N		 7624 /* When a reshape changes the number of devices,
7625 * ->degraded is measured against the larger of the
7626 * pre and post number of devices.
7627 */
ec32a2bd 7628 spin_lock_irqsave(&conf->device_lock, flags);
908f4fbd 7629 mddev->degraded = calc_degraded(conf);
ec32a2bd
N		 7630 spin_unlock_irqrestore(&conf->device_lock, flags);
7631 }
63c70c4f 7632 mddev->raid_disks = conf->raid_disks;
e516402c 7633 mddev->reshape_position = conf->reshape_progress;
850b2b42 7634 set_bit(MD_CHANGE_DEVS, &mddev->flags);
f6705578 7635
29269553
N		 7636 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7637 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
ea358cd0 7638 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
29269553
N		 7639 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7640 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7641 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 7642 "reshape");
29269553
N		 7643 if (!mddev->sync_thread) {
7644 mddev->recovery = 0;
7645 spin_lock_irq(&conf->device_lock);
ba8805b9 7646 write_seqcount_begin(&conf->gen_lock);
29269553 7647 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
ba8805b9
N		 7648 mddev->new_chunk_sectors =
7649 conf->chunk_sectors = conf->prev_chunk_sectors;
7650 mddev->new_layout = conf->algorithm = conf->prev_algo;
05616be5
N		 7651 rdev_for_each(rdev, mddev)
7652 rdev->new_data_offset = rdev->data_offset;
7653 smp_wmb();
ba8805b9 7654 conf->generation --;
fef9c61f 7655 conf->reshape_progress = MaxSector;
1e3fa9bd 7656 mddev->reshape_position = MaxSector;
ba8805b9 7657 write_seqcount_end(&conf->gen_lock);
29269553
N		 7658 spin_unlock_irq(&conf->device_lock);
7659 return -EAGAIN;
7660 }
c8f517c4 7661 conf->reshape_checkpoint = jiffies;
29269553
N		 7662 md_wakeup_thread(mddev->sync_thread);
7663 md_new_event(mddev);
7664 return 0;
7665}
29269553 7666
ec32a2bd
N		 7667/* This is called from the reshape thread and should make any
7668 * changes needed in 'conf'
7669 */
d1688a6d 7670static void end_reshape(struct r5conf *conf)
29269553 7671{
29269553 7672
f6705578 7673 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
05616be5 7674 struct md_rdev *rdev;
f6705578 7675
f6705578 7676 spin_lock_irq(&conf->device_lock);
cea9c228 7677 conf->previous_raid_disks = conf->raid_disks;
05616be5
N		 7678 rdev_for_each(rdev, conf->mddev)
7679 rdev->data_offset = rdev->new_data_offset;
7680 smp_wmb();
fef9c61f 7681 conf->reshape_progress = MaxSector;
6cbd8148 7682 conf->mddev->reshape_position = MaxSector;
f6705578 7683 spin_unlock_irq(&conf->device_lock);
b0f9ec04 7684 wake_up(&conf->wait_for_overlap);
16a53ecc
N		 7685
7686 /* read-ahead size must cover two whole stripes, which is
7687 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
7688 */
4a5add49 7689 if (conf->mddev->queue) {
cea9c228 7690 int data_disks = conf->raid_disks - conf->max_degraded;
09c9e5fa 7691 int stripe = data_disks * ((conf->chunk_sectors << 9)
cea9c228 7692 / PAGE_SIZE);
16a53ecc
N		 7693 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
7694 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
7695 }
29269553 7696 }
29269553
N		 7697}
7698
ec32a2bd
N		 7699/* This is called from the raid5d thread with mddev_lock held.
7700 * It makes config changes to the device.
7701 */
fd01b88c 7702static void raid5_finish_reshape(struct mddev *mddev)
cea9c228 7703{
d1688a6d 7704 struct r5conf *conf = mddev->private;
cea9c228
N		 7705
7706 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7707
ec32a2bd
N		 7708 if (mddev->delta_disks > 0) {
7709 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
fe67d19a
HM		 7710 if (mddev->queue) {
7711 set_capacity(mddev->gendisk, mddev->array_sectors);
7712 revalidate_disk(mddev->gendisk);
7713 }
ec32a2bd
N		 7714 } else {
7715 int d;
908f4fbd
N		 7716 spin_lock_irq(&conf->device_lock);
7717 mddev->degraded = calc_degraded(conf);
7718 spin_unlock_irq(&conf->device_lock);
ec32a2bd
N		 7719 for (d = conf->raid_disks ;
7720 d < conf->raid_disks - mddev->delta_disks;
1a67dde0 7721 d++) {
3cb03002 7722 struct md_rdev *rdev = conf->disks[d].rdev;
da7613b8
N		 7723 if (rdev)
7724 clear_bit(In_sync, &rdev->flags);
7725 rdev = conf->disks[d].replacement;
7726 if (rdev)
7727 clear_bit(In_sync, &rdev->flags);
1a67dde0 7728 }
cea9c228 7729 }
88ce4930 7730 mddev->layout = conf->algorithm;
09c9e5fa 7731 mddev->chunk_sectors = conf->chunk_sectors;
ec32a2bd
N		 7732 mddev->reshape_position = MaxSector;
7733 mddev->delta_disks = 0;
2c810cdd 7734 mddev->reshape_backwards = 0;
cea9c228
N		 7735 }
7736}
7737
fd01b88c 7738static void raid5_quiesce(struct mddev *mddev, int state)
72626685 7739{
d1688a6d 7740 struct r5conf *conf = mddev->private;
72626685
N		 7741
7742 switch(state) {
e464eafd
N		 7743 case 2: /* resume for a suspend */
7744 wake_up(&conf->wait_for_overlap);
7745 break;
7746
72626685 7747 case 1: /* stop all writes */
566c09c5 7748 lock_all_device_hash_locks_irq(conf);
64bd660b
N		 7749 /* '2' tells resync/reshape to pause so that all
7750 * active stripes can drain
7751 */
a39f7afd 7752 r5c_flush_cache(conf, INT_MAX);
64bd660b 7753 conf->quiesce = 2;
b1b46486 7754 wait_event_cmd(conf->wait_for_quiescent,
46031f9a
RBJ		 7755 atomic_read(&conf->active_stripes) == 0 &&
7756 atomic_read(&conf->active_aligned_reads) == 0,
566c09c5
SL		 7757 unlock_all_device_hash_locks_irq(conf),
7758 lock_all_device_hash_locks_irq(conf));
64bd660b 7759 conf->quiesce = 1;
566c09c5 7760 unlock_all_device_hash_locks_irq(conf);
64bd660b
N		 7761 /* allow reshape to continue */
7762 wake_up(&conf->wait_for_overlap);
72626685
N		 7763 break;
7764
7765 case 0: /* re-enable writes */
566c09c5 7766 lock_all_device_hash_locks_irq(conf);
72626685 7767 conf->quiesce = 0;
b1b46486 7768 wake_up(&conf->wait_for_quiescent);
e464eafd 7769 wake_up(&conf->wait_for_overlap);
566c09c5 7770 unlock_all_device_hash_locks_irq(conf);
72626685
N		 7771 break;
7772 }
e6c033f7 7773 r5l_quiesce(conf->log, state);
72626685 7774}
b15c2e57 7775
fd01b88c 7776static void *raid45_takeover_raid0(struct mddev *mddev, int level)
54071b38 7777{
e373ab10 7778 struct r0conf *raid0_conf = mddev->private;
d76c8420 7779 sector_t sectors;
54071b38 7780
f1b29bca 7781 /* for raid0 takeover only one zone is supported */
e373ab10 7782 if (raid0_conf->nr_strip_zones > 1) {
cc6167b4
N		 7783 pr_warn("md/raid:%s: cannot takeover raid0 with more than one zone.\n",
7784 mdname(mddev));
f1b29bca
DW		 7785 return ERR_PTR(-EINVAL);
7786 }
7787
e373ab10
N		 7788 sectors = raid0_conf->strip_zone[0].zone_end;
7789 sector_div(sectors, raid0_conf->strip_zone[0].nb_dev);
3b71bd93 7790 mddev->dev_sectors = sectors;
f1b29bca 7791 mddev->new_level = level;
54071b38
TM		 7792 mddev->new_layout = ALGORITHM_PARITY_N;
7793 mddev->new_chunk_sectors = mddev->chunk_sectors;
7794 mddev->raid_disks += 1;
7795 mddev->delta_disks = 1;
7796 /* make sure it will be not marked as dirty */
7797 mddev->recovery_cp = MaxSector;
7798
7799 return setup_conf(mddev);
7800}
7801
fd01b88c 7802static void *raid5_takeover_raid1(struct mddev *mddev)
d562b0c4
N		 7803{
7804 int chunksect;
7805
7806 if (mddev->raid_disks != 2 ||
7807 mddev->degraded > 1)
7808 return ERR_PTR(-EINVAL);
7809
7810 /* Should check if there are write-behind devices? */
7811
7812 chunksect = 64*2; /* 64K by default */
7813
7814 /* The array must be an exact multiple of chunksize */
7815 while (chunksect && (mddev->array_sectors & (chunksect-1)))
7816 chunksect >>= 1;
7817
7818 if ((chunksect<<9) < STRIPE_SIZE)
7819 /* array size does not allow a suitable chunk size */
7820 return ERR_PTR(-EINVAL);
7821
7822 mddev->new_level = 5;
7823 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
664e7c41 7824 mddev->new_chunk_sectors = chunksect;
d562b0c4
N		 7825
7826 return setup_conf(mddev);
7827}
7828
fd01b88c 7829static void *raid5_takeover_raid6(struct mddev *mddev)
fc9739c6
N		 7830{
7831 int new_layout;
7832
7833 switch (mddev->layout) {
7834 case ALGORITHM_LEFT_ASYMMETRIC_6:
7835 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
7836 break;
7837 case ALGORITHM_RIGHT_ASYMMETRIC_6:
7838 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
7839 break;
7840 case ALGORITHM_LEFT_SYMMETRIC_6:
7841 new_layout = ALGORITHM_LEFT_SYMMETRIC;
7842 break;
7843 case ALGORITHM_RIGHT_SYMMETRIC_6:
7844 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
7845 break;
7846 case ALGORITHM_PARITY_0_6:
7847 new_layout = ALGORITHM_PARITY_0;
7848 break;
7849 case ALGORITHM_PARITY_N:
7850 new_layout = ALGORITHM_PARITY_N;
7851 break;
7852 default:
7853 return ERR_PTR(-EINVAL);
7854 }
7855 mddev->new_level = 5;
7856 mddev->new_layout = new_layout;
7857 mddev->delta_disks = -1;
7858 mddev->raid_disks -= 1;
7859 return setup_conf(mddev);
7860}
7861
fd01b88c 7862static int raid5_check_reshape(struct mddev *mddev)
b3546035 7863{
88ce4930
N		 7864 /* For a 2-drive array, the layout and chunk size can be changed
7865 * immediately as not restriping is needed.
7866 * For larger arrays we record the new value - after validation
7867 * to be used by a reshape pass.
b3546035 7868 */
d1688a6d 7869 struct r5conf *conf = mddev->private;
597a711b 7870 int new_chunk = mddev->new_chunk_sectors;
b3546035 7871
597a711b 7872 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
b3546035
N		 7873 return -EINVAL;
7874 if (new_chunk > 0) {
0ba459d2 7875 if (!is_power_of_2(new_chunk))
b3546035 7876 return -EINVAL;
597a711b 7877 if (new_chunk < (PAGE_SIZE>>9))
b3546035 7878 return -EINVAL;
597a711b 7879 if (mddev->array_sectors & (new_chunk-1))
b3546035
N		 7880 /* not factor of array size */
7881 return -EINVAL;
7882 }
7883
7884 /* They look valid */
7885
88ce4930 7886 if (mddev->raid_disks == 2) {
597a711b
N		 7887 /* can make the change immediately */
7888 if (mddev->new_layout >= 0) {
7889 conf->algorithm = mddev->new_layout;
7890 mddev->layout = mddev->new_layout;
88ce4930
N		 7891 }
7892 if (new_chunk > 0) {
597a711b
N		 7893 conf->chunk_sectors = new_chunk ;
7894 mddev->chunk_sectors = new_chunk;
88ce4930
N		 7895 }
7896 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7897 md_wakeup_thread(mddev->thread);
b3546035 7898 }
50ac168a 7899 return check_reshape(mddev);
88ce4930
N		 7900}
7901
fd01b88c 7902static int raid6_check_reshape(struct mddev *mddev)
88ce4930 7903{
597a711b 7904 int new_chunk = mddev->new_chunk_sectors;
50ac168a 7905
597a711b 7906 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
88ce4930 7907 return -EINVAL;
b3546035 7908 if (new_chunk > 0) {
0ba459d2 7909 if (!is_power_of_2(new_chunk))
88ce4930 7910 return -EINVAL;
597a711b 7911 if (new_chunk < (PAGE_SIZE >> 9))
88ce4930 7912 return -EINVAL;
597a711b 7913 if (mddev->array_sectors & (new_chunk-1))
88ce4930
N		 7914 /* not factor of array size */
7915 return -EINVAL;
b3546035 7916 }
88ce4930
N		 7917
7918 /* They look valid */
50ac168a 7919 return check_reshape(mddev);
b3546035
N		 7920}
7921
fd01b88c 7922static void *raid5_takeover(struct mddev *mddev)
d562b0c4
N		 7923{
7924 /* raid5 can take over:
f1b29bca 7925 * raid0 - if there is only one strip zone - make it a raid4 layout
d562b0c4
N		 7926 * raid1 - if there are two drives. We need to know the chunk size
7927 * raid4 - trivial - just use a raid4 layout.
7928 * raid6 - Providing it is a *_6 layout
d562b0c4 7929 */
f1b29bca
DW		 7930 if (mddev->level == 0)
7931 return raid45_takeover_raid0(mddev, 5);
d562b0c4
N		 7932 if (mddev->level == 1)
7933 return raid5_takeover_raid1(mddev);
e9d4758f
N		 7934 if (mddev->level == 4) {
7935 mddev->new_layout = ALGORITHM_PARITY_N;
7936 mddev->new_level = 5;
7937 return setup_conf(mddev);
7938 }
fc9739c6
N		 7939 if (mddev->level == 6)
7940 return raid5_takeover_raid6(mddev);
d562b0c4
N		 7941
7942 return ERR_PTR(-EINVAL);
7943}
7944
fd01b88c 7945static void *raid4_takeover(struct mddev *mddev)
a78d38a1 7946{
f1b29bca
DW		 7947 /* raid4 can take over:
7948 * raid0 - if there is only one strip zone
7949 * raid5 - if layout is right
a78d38a1 7950 */
f1b29bca
DW		 7951 if (mddev->level == 0)
7952 return raid45_takeover_raid0(mddev, 4);
a78d38a1
N		 7953 if (mddev->level == 5 &&
7954 mddev->layout == ALGORITHM_PARITY_N) {
7955 mddev->new_layout = 0;
7956 mddev->new_level = 4;
7957 return setup_conf(mddev);
7958 }
7959 return ERR_PTR(-EINVAL);
7960}
d562b0c4 7961
84fc4b56 7962static struct md_personality raid5_personality;
245f46c2 7963
fd01b88c 7964static void *raid6_takeover(struct mddev *mddev)
245f46c2
N		 7965{
7966 /* Currently can only take over a raid5. We map the
7967 * personality to an equivalent raid6 personality
7968 * with the Q block at the end.
7969 */
7970 int new_layout;
7971
7972 if (mddev->pers != &raid5_personality)
7973 return ERR_PTR(-EINVAL);
7974 if (mddev->degraded > 1)
7975 return ERR_PTR(-EINVAL);
7976 if (mddev->raid_disks > 253)
7977 return ERR_PTR(-EINVAL);
7978 if (mddev->raid_disks < 3)
7979 return ERR_PTR(-EINVAL);
7980
7981 switch (mddev->layout) {
7982 case ALGORITHM_LEFT_ASYMMETRIC:
7983 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
7984 break;
7985 case ALGORITHM_RIGHT_ASYMMETRIC:
7986 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
7987 break;
7988 case ALGORITHM_LEFT_SYMMETRIC:
7989 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
7990 break;
7991 case ALGORITHM_RIGHT_SYMMETRIC:
7992 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
7993 break;
7994 case ALGORITHM_PARITY_0:
7995 new_layout = ALGORITHM_PARITY_0_6;
7996 break;
7997 case ALGORITHM_PARITY_N:
7998 new_layout = ALGORITHM_PARITY_N;
7999 break;
8000 default:
8001 return ERR_PTR(-EINVAL);
8002 }
8003 mddev->new_level = 6;
8004 mddev->new_layout = new_layout;
8005 mddev->delta_disks = 1;
8006 mddev->raid_disks += 1;
8007 return setup_conf(mddev);
8008}
8009
84fc4b56 8010static struct md_personality raid6_personality =
16a53ecc
N		 8011{
8012 .name = "raid6",
8013 .level = 6,
8014 .owner = THIS_MODULE,
849674e4
SL		 8015 .make_request = raid5_make_request,
8016 .run = raid5_run,
afa0f557 8017 .free = raid5_free,
849674e4
SL		 8018 .status = raid5_status,
8019 .error_handler = raid5_error,
16a53ecc
N		 8020 .hot_add_disk = raid5_add_disk,
8021 .hot_remove_disk= raid5_remove_disk,
8022 .spare_active = raid5_spare_active,
849674e4 8023 .sync_request = raid5_sync_request,
16a53ecc 8024 .resize = raid5_resize,
80c3a6ce 8025 .size = raid5_size,
50ac168a 8026 .check_reshape = raid6_check_reshape,
f416885e 8027 .start_reshape = raid5_start_reshape,
cea9c228 8028 .finish_reshape = raid5_finish_reshape,
16a53ecc 8029 .quiesce = raid5_quiesce,
245f46c2 8030 .takeover = raid6_takeover,
5c675f83 8031 .congested = raid5_congested,
16a53ecc 8032};
84fc4b56 8033static struct md_personality raid5_personality =
1da177e4
LT		 8034{
8035 .name = "raid5",
2604b703 8036 .level = 5,
1da177e4 8037 .owner = THIS_MODULE,
849674e4
SL		 8038 .make_request = raid5_make_request,
8039 .run = raid5_run,
afa0f557 8040 .free = raid5_free,
849674e4
SL		 8041 .status = raid5_status,
8042 .error_handler = raid5_error,
1da177e4
LT		 8043 .hot_add_disk = raid5_add_disk,
8044 .hot_remove_disk= raid5_remove_disk,
8045 .spare_active = raid5_spare_active,
849674e4 8046 .sync_request = raid5_sync_request,
1da177e4 8047 .resize = raid5_resize,
80c3a6ce 8048 .size = raid5_size,
63c70c4f
N		 8049 .check_reshape = raid5_check_reshape,
8050 .start_reshape = raid5_start_reshape,
cea9c228 8051 .finish_reshape = raid5_finish_reshape,
72626685 8052 .quiesce = raid5_quiesce,
d562b0c4 8053 .takeover = raid5_takeover,
5c675f83 8054 .congested = raid5_congested,
1da177e4
LT		 8055};
8056
84fc4b56 8057static struct md_personality raid4_personality =
1da177e4 8058{
2604b703
N		 8059 .name = "raid4",
8060 .level = 4,
8061 .owner = THIS_MODULE,
849674e4
SL		 8062 .make_request = raid5_make_request,
8063 .run = raid5_run,
afa0f557 8064 .free = raid5_free,
849674e4
SL		 8065 .status = raid5_status,
8066 .error_handler = raid5_error,
2604b703
N		 8067 .hot_add_disk = raid5_add_disk,
8068 .hot_remove_disk= raid5_remove_disk,
8069 .spare_active = raid5_spare_active,
849674e4 8070 .sync_request = raid5_sync_request,
2604b703 8071 .resize = raid5_resize,
80c3a6ce 8072 .size = raid5_size,
3d37890b
N		 8073 .check_reshape = raid5_check_reshape,
8074 .start_reshape = raid5_start_reshape,
cea9c228 8075 .finish_reshape = raid5_finish_reshape,
2604b703 8076 .quiesce = raid5_quiesce,
a78d38a1 8077 .takeover = raid4_takeover,
5c675f83 8078 .congested = raid5_congested,
2604b703
N		 8079};
8080
8081static int __init raid5_init(void)
8082{
29c6d1bb
SAS		 8083 int ret;
8084
851c30c9
SL		 8085 raid5_wq = alloc_workqueue("raid5wq",
8086 WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0);
8087 if (!raid5_wq)
8088 return -ENOMEM;
29c6d1bb
SAS		 8089
8090 ret = cpuhp_setup_state_multi(CPUHP_MD_RAID5_PREPARE,
8091 "md/raid5:prepare",
8092 raid456_cpu_up_prepare,
8093 raid456_cpu_dead);
8094 if (ret) {
8095 destroy_workqueue(raid5_wq);
8096 return ret;
8097 }
16a53ecc 8098 register_md_personality(&raid6_personality);
2604b703
N		 8099 register_md_personality(&raid5_personality);
8100 register_md_personality(&raid4_personality);
8101 return 0;
1da177e4
LT		 8102}
8103
2604b703 8104static void raid5_exit(void)
1da177e4 8105{
16a53ecc 8106 unregister_md_personality(&raid6_personality);
2604b703
N		 8107 unregister_md_personality(&raid5_personality);
8108 unregister_md_personality(&raid4_personality);
29c6d1bb 8109 cpuhp_remove_multi_state(CPUHP_MD_RAID5_PREPARE);
851c30c9 8110 destroy_workqueue(raid5_wq);
1da177e4
LT		 8111}
8112
8113module_init(raid5_init);
8114module_exit(raid5_exit);
8115MODULE_LICENSE("GPL");
0efb9e61 8116MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
1da177e4 8117MODULE_ALIAS("md-personality-4"); /* RAID5 */
d9d166c2
N		 8118MODULE_ALIAS("md-raid5");
8119MODULE_ALIAS("md-raid4");
2604b703
N		 8120MODULE_ALIAS("md-level-5");
8121MODULE_ALIAS("md-level-4");
16a53ecc
N		 8122MODULE_ALIAS("md-personality-8"); /* RAID6 */
8123MODULE_ALIAS("md-raid6");
8124MODULE_ALIAS("md-level-6");
8125
8126/* This used to be two separate modules, they were: */
8127MODULE_ALIAS("raid5");
8128MODULE_ALIAS("raid6");
Linux 6.x block layer and io_uring tree(s)