projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md: raid10: remove VLAIS
[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>
3f07c014
IM		 58#include <linux/sched/signal.h>
59
a9add5d9 60#include <trace/events/block.h>
aaf9f12e 61#include <linux/list_sort.h>
a9add5d9 62
43b2e5d8 63#include "md.h"
bff61975 64#include "raid5.h"
54071b38 65#include "raid0.h"
ef740c37 66#include "bitmap.h"
ff875738 67#include "raid5-log.h"
72626685 68
394ed8e4
SL		 69#define UNSUPPORTED_MDDEV_FLAGS (1L << MD_FAILFAST_SUPPORTED)
70
851c30c9
SL		 71#define cpu_to_group(cpu) cpu_to_node(cpu)
72#define ANY_GROUP NUMA_NO_NODE
73
8e0e99ba
N		 74static bool devices_handle_discard_safely = false;
75module_param(devices_handle_discard_safely, bool, 0644);
76MODULE_PARM_DESC(devices_handle_discard_safely,
77 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
851c30c9 78static struct workqueue_struct *raid5_wq;
1da177e4 79
d1688a6d 80static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect)
db298e19
N		 81{
82 int hash = (sect >> STRIPE_SHIFT) & HASH_MASK;
83 return &conf->stripe_hashtbl[hash];
84}
1da177e4 85
566c09c5
SL		 86static inline int stripe_hash_locks_hash(sector_t sect)
87{
88 return (sect >> STRIPE_SHIFT) & STRIPE_HASH_LOCKS_MASK;
89}
90
91static inline void lock_device_hash_lock(struct r5conf *conf, int hash)
92{
93 spin_lock_irq(conf->hash_locks + hash);
94 spin_lock(&conf->device_lock);
95}
96
97static inline void unlock_device_hash_lock(struct r5conf *conf, int hash)
98{
99 spin_unlock(&conf->device_lock);
100 spin_unlock_irq(conf->hash_locks + hash);
101}
102
103static inline void lock_all_device_hash_locks_irq(struct r5conf *conf)
104{
105 int i;
3d05f3ae 106 spin_lock_irq(conf->hash_locks);
566c09c5
SL		 107 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
108 spin_lock_nest_lock(conf->hash_locks + i, conf->hash_locks);
109 spin_lock(&conf->device_lock);
110}
111
112static inline void unlock_all_device_hash_locks_irq(struct r5conf *conf)
113{
114 int i;
115 spin_unlock(&conf->device_lock);
3d05f3ae
JC		 116 for (i = NR_STRIPE_HASH_LOCKS - 1; i; i--)
117 spin_unlock(conf->hash_locks + i);
118 spin_unlock_irq(conf->hash_locks);
566c09c5
SL		 119}
120
d0dabf7e
N		 121/* Find first data disk in a raid6 stripe */
122static inline int raid6_d0(struct stripe_head *sh)
123{
67cc2b81
N		 124 if (sh->ddf_layout)
125 /* ddf always start from first device */
126 return 0;
127 /* md starts just after Q block */
d0dabf7e
N		 128 if (sh->qd_idx == sh->disks - 1)
129 return 0;
130 else
131 return sh->qd_idx + 1;
132}
16a53ecc
N		 133static inline int raid6_next_disk(int disk, int raid_disks)
134{
135 disk++;
136 return (disk < raid_disks) ? disk : 0;
137}
a4456856 138
d0dabf7e
N		 139/* When walking through the disks in a raid5, starting at raid6_d0,
140 * We need to map each disk to a 'slot', where the data disks are slot
141 * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk
142 * is raid_disks-1. This help does that mapping.
143 */
67cc2b81
N		 144static int raid6_idx_to_slot(int idx, struct stripe_head *sh,
145 int *count, int syndrome_disks)
d0dabf7e 146{
6629542e 147 int slot = *count;
67cc2b81 148
e4424fee 149 if (sh->ddf_layout)
6629542e 150 (*count)++;
d0dabf7e 151 if (idx == sh->pd_idx)
67cc2b81 152 return syndrome_disks;
d0dabf7e 153 if (idx == sh->qd_idx)
67cc2b81 154 return syndrome_disks + 1;
e4424fee 155 if (!sh->ddf_layout)
6629542e 156 (*count)++;
d0dabf7e
N		 157 return slot;
158}
159
d1688a6d 160static void print_raid5_conf (struct r5conf *conf);
1da177e4 161
600aa109
DW		 162static int stripe_operations_active(struct stripe_head *sh)
163{
164 return sh->check_state || sh->reconstruct_state ||
165 test_bit(STRIPE_BIOFILL_RUN, &sh->state) ||
166 test_bit(STRIPE_COMPUTE_RUN, &sh->state);
167}
168
535ae4eb
SL		 169static bool stripe_is_lowprio(struct stripe_head *sh)
170{
171 return (test_bit(STRIPE_R5C_FULL_STRIPE, &sh->state) ||
172 test_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state)) &&
173 !test_bit(STRIPE_R5C_CACHING, &sh->state);
174}
175
851c30c9
SL		 176static void raid5_wakeup_stripe_thread(struct stripe_head *sh)
177{
178 struct r5conf *conf = sh->raid_conf;
179 struct r5worker_group *group;
bfc90cb0 180 int thread_cnt;
851c30c9
SL		 181 int i, cpu = sh->cpu;
182
183 if (!cpu_online(cpu)) {
184 cpu = cpumask_any(cpu_online_mask);
185 sh->cpu = cpu;
186 }
187
188 if (list_empty(&sh->lru)) {
189 struct r5worker_group *group;
190 group = conf->worker_groups + cpu_to_group(cpu);
535ae4eb
SL		 191 if (stripe_is_lowprio(sh))
192 list_add_tail(&sh->lru, &group->loprio_list);
193 else
194 list_add_tail(&sh->lru, &group->handle_list);
bfc90cb0
SL		 195 group->stripes_cnt++;
196 sh->group = group;
851c30c9
SL		 197 }
198
199 if (conf->worker_cnt_per_group == 0) {
200 md_wakeup_thread(conf->mddev->thread);
201 return;
202 }
203
204 group = conf->worker_groups + cpu_to_group(sh->cpu);
205
bfc90cb0
SL		 206 group->workers[0].working = true;
207 /* at least one worker should run to avoid race */
208 queue_work_on(sh->cpu, raid5_wq, &group->workers[0].work);
209
210 thread_cnt = group->stripes_cnt / MAX_STRIPE_BATCH - 1;
211 /* wakeup more workers */
212 for (i = 1; i < conf->worker_cnt_per_group && thread_cnt > 0; i++) {
213 if (group->workers[i].working == false) {
214 group->workers[i].working = true;
215 queue_work_on(sh->cpu, raid5_wq,
216 &group->workers[i].work);
217 thread_cnt--;
218 }
219 }
851c30c9
SL		 220}
221
566c09c5
SL		 222static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh,
223 struct list_head *temp_inactive_list)
1da177e4 224{
1e6d690b
SL		 225 int i;
226 int injournal = 0; /* number of date pages with R5_InJournal */
227
4eb788df
SL		 228 BUG_ON(!list_empty(&sh->lru));
229 BUG_ON(atomic_read(&conf->active_stripes)==0);
1e6d690b
SL		 230
231 if (r5c_is_writeback(conf->log))
232 for (i = sh->disks; i--; )
233 if (test_bit(R5_InJournal, &sh->dev[i].flags))
234 injournal++;
a39f7afd 235 /*
5ddf0440
SL		 236 * In the following cases, the stripe cannot be released to cached
237 * lists. Therefore, we make the stripe write out and set
238 * STRIPE_HANDLE:
239 * 1. when quiesce in r5c write back;
240 * 2. when resync is requested fot the stripe.
a39f7afd 241 */
5ddf0440
SL		 242 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) ||
243 (conf->quiesce && r5c_is_writeback(conf->log) &&
244 !test_bit(STRIPE_HANDLE, &sh->state) && injournal != 0)) {
a39f7afd
SL		 245 if (test_bit(STRIPE_R5C_CACHING, &sh->state))
246 r5c_make_stripe_write_out(sh);
247 set_bit(STRIPE_HANDLE, &sh->state);
248 }
1e6d690b 249
4eb788df
SL		 250 if (test_bit(STRIPE_HANDLE, &sh->state)) {
251 if (test_bit(STRIPE_DELAYED, &sh->state) &&
ad3ab8b6 252 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4eb788df 253 list_add_tail(&sh->lru, &conf->delayed_list);
ad3ab8b6 254 else if (test_bit(STRIPE_BIT_DELAY, &sh->state) &&
4eb788df
SL		 255 sh->bm_seq - conf->seq_write > 0)
256 list_add_tail(&sh->lru, &conf->bitmap_list);
257 else {
258 clear_bit(STRIPE_DELAYED, &sh->state);
259 clear_bit(STRIPE_BIT_DELAY, &sh->state);
851c30c9 260 if (conf->worker_cnt_per_group == 0) {
535ae4eb
SL		 261 if (stripe_is_lowprio(sh))
262 list_add_tail(&sh->lru,
263 &conf->loprio_list);
264 else
265 list_add_tail(&sh->lru,
266 &conf->handle_list);
851c30c9
SL		 267 } else {
268 raid5_wakeup_stripe_thread(sh);
269 return;
270 }
4eb788df
SL		 271 }
272 md_wakeup_thread(conf->mddev->thread);
273 } else {
274 BUG_ON(stripe_operations_active(sh));
275 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
276 if (atomic_dec_return(&conf->preread_active_stripes)
277 < IO_THRESHOLD)
278 md_wakeup_thread(conf->mddev->thread);
279 atomic_dec(&conf->active_stripes);
1e6d690b
SL		 280 if (!test_bit(STRIPE_EXPANDING, &sh->state)) {
281 if (!r5c_is_writeback(conf->log))
282 list_add_tail(&sh->lru, temp_inactive_list);
283 else {
284 WARN_ON(test_bit(R5_InJournal, &sh->dev[sh->pd_idx].flags));
285 if (injournal == 0)
286 list_add_tail(&sh->lru, temp_inactive_list);
287 else if (injournal == conf->raid_disks - conf->max_degraded) {
288 /* full stripe */
289 if (!test_and_set_bit(STRIPE_R5C_FULL_STRIPE, &sh->state))
290 atomic_inc(&conf->r5c_cached_full_stripes);
291 if (test_and_clear_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state))
292 atomic_dec(&conf->r5c_cached_partial_stripes);
293 list_add_tail(&sh->lru, &conf->r5c_full_stripe_list);
a39f7afd 294 r5c_check_cached_full_stripe(conf);
03b047f4
SL		 295 } else
296 /*
297 * STRIPE_R5C_PARTIAL_STRIPE is set in
298 * r5c_try_caching_write(). No need to
299 * set it again.
300 */
1e6d690b 301 list_add_tail(&sh->lru, &conf->r5c_partial_stripe_list);
1e6d690b
SL		 302 }
303 }
1da177e4
LT		 304 }
305}
d0dabf7e 306
566c09c5
SL		 307static void __release_stripe(struct r5conf *conf, struct stripe_head *sh,
308 struct list_head *temp_inactive_list)
4eb788df
SL		 309{
310 if (atomic_dec_and_test(&sh->count))
566c09c5
SL		 311 do_release_stripe(conf, sh, temp_inactive_list);
312}
313
314/*
315 * @hash could be NR_STRIPE_HASH_LOCKS, then we have a list of inactive_list
316 *
317 * Be careful: Only one task can add/delete stripes from temp_inactive_list at
318 * given time. Adding stripes only takes device lock, while deleting stripes
319 * only takes hash lock.
320 */
321static void release_inactive_stripe_list(struct r5conf *conf,
322 struct list_head *temp_inactive_list,
323 int hash)
324{
325 int size;
6ab2a4b8 326 bool do_wakeup = false;
566c09c5
SL		 327 unsigned long flags;
328
329 if (hash == NR_STRIPE_HASH_LOCKS) {
330 size = NR_STRIPE_HASH_LOCKS;
331 hash = NR_STRIPE_HASH_LOCKS - 1;
332 } else
333 size = 1;
334 while (size) {
335 struct list_head *list = &temp_inactive_list[size - 1];
336
337 /*
6d036f7d 338 * We don't hold any lock here yet, raid5_get_active_stripe() might
566c09c5
SL		 339 * remove stripes from the list
340 */
341 if (!list_empty_careful(list)) {
342 spin_lock_irqsave(conf->hash_locks + hash, flags);
4bda556a
SL		 343 if (list_empty(conf->inactive_list + hash) &&
344 !list_empty(list))
345 atomic_dec(&conf->empty_inactive_list_nr);
566c09c5 346 list_splice_tail_init(list, conf->inactive_list + hash);
6ab2a4b8 347 do_wakeup = true;
566c09c5
SL		 348 spin_unlock_irqrestore(conf->hash_locks + hash, flags);
349 }
350 size--;
351 hash--;
352 }
353
354 if (do_wakeup) {
6ab2a4b8 355 wake_up(&conf->wait_for_stripe);
b1b46486
YL		 356 if (atomic_read(&conf->active_stripes) == 0)
357 wake_up(&conf->wait_for_quiescent);
566c09c5
SL		 358 if (conf->retry_read_aligned)
359 md_wakeup_thread(conf->mddev->thread);
360 }
4eb788df
SL		 361}
362
773ca82f 363/* should hold conf->device_lock already */
566c09c5
SL		 364static int release_stripe_list(struct r5conf *conf,
365 struct list_head *temp_inactive_list)
773ca82f 366{
eae8263f 367 struct stripe_head *sh, *t;
773ca82f
SL		 368 int count = 0;
369 struct llist_node *head;
370
371 head = llist_del_all(&conf->released_stripes);
d265d9dc 372 head = llist_reverse_order(head);
eae8263f 373 llist_for_each_entry_safe(sh, t, head, release_list) {
566c09c5
SL		 374 int hash;
375
773ca82f
SL		 376 /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */
377 smp_mb();
378 clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state);
379 /*
380 * Don't worry the bit is set here, because if the bit is set
381 * again, the count is always > 1. This is true for
382 * STRIPE_ON_UNPLUG_LIST bit too.
383 */
566c09c5
SL		 384 hash = sh->hash_lock_index;
385 __release_stripe(conf, sh, &temp_inactive_list[hash]);
773ca82f
SL		 386 count++;
387 }
388
389 return count;
390}
391
6d036f7d 392void raid5_release_stripe(struct stripe_head *sh)
1da177e4 393{
d1688a6d 394 struct r5conf *conf = sh->raid_conf;
1da177e4 395 unsigned long flags;
566c09c5
SL		 396 struct list_head list;
397 int hash;
773ca82f 398 bool wakeup;
16a53ecc 399
cf170f3f
ES		 400 /* Avoid release_list until the last reference.
401 */
402 if (atomic_add_unless(&sh->count, -1, 1))
403 return;
404
ad4068de 405 if (unlikely(!conf->mddev->thread) ||
406 test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state))
773ca82f
SL		 407 goto slow_path;
408 wakeup = llist_add(&sh->release_list, &conf->released_stripes);
409 if (wakeup)
410 md_wakeup_thread(conf->mddev->thread);
411 return;
412slow_path:
4eb788df 413 local_irq_save(flags);
773ca82f 414 /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */
4eb788df 415 if (atomic_dec_and_lock(&sh->count, &conf->device_lock)) {
566c09c5
SL		 416 INIT_LIST_HEAD(&list);
417 hash = sh->hash_lock_index;
418 do_release_stripe(conf, sh, &list);
4eb788df 419 spin_unlock(&conf->device_lock);
566c09c5 420 release_inactive_stripe_list(conf, &list, hash);
4eb788df
SL		 421 }
422 local_irq_restore(flags);
1da177e4
LT		 423}
424
fccddba0 425static inline void remove_hash(struct stripe_head *sh)
1da177e4 426{
45b4233c
DW		 427 pr_debug("remove_hash(), stripe %llu\n",
428 (unsigned long long)sh->sector);
1da177e4 429
fccddba0 430 hlist_del_init(&sh->hash);
1da177e4
LT		 431}
432
d1688a6d 433static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh)
1da177e4 434{
fccddba0 435 struct hlist_head *hp = stripe_hash(conf, sh->sector);
1da177e4 436
45b4233c
DW		 437 pr_debug("insert_hash(), stripe %llu\n",
438 (unsigned long long)sh->sector);
1da177e4 439
fccddba0 440 hlist_add_head(&sh->hash, hp);
1da177e4
LT		 441}
442
1da177e4 443/* find an idle stripe, make sure it is unhashed, and return it. */
566c09c5 444static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash)
1da177e4
LT		 445{
446 struct stripe_head *sh = NULL;
447 struct list_head *first;
448
566c09c5 449 if (list_empty(conf->inactive_list + hash))
1da177e4 450 goto out;
566c09c5 451 first = (conf->inactive_list + hash)->next;
1da177e4
LT		 452 sh = list_entry(first, struct stripe_head, lru);
453 list_del_init(first);
454 remove_hash(sh);
455 atomic_inc(&conf->active_stripes);
566c09c5 456 BUG_ON(hash != sh->hash_lock_index);
4bda556a
SL		 457 if (list_empty(conf->inactive_list + hash))
458 atomic_inc(&conf->empty_inactive_list_nr);
1da177e4
LT		 459out:
460 return sh;
461}
462
e4e11e38 463static void shrink_buffers(struct stripe_head *sh)
1da177e4
LT		 464{
465 struct page *p;
466 int i;
e4e11e38 467 int num = sh->raid_conf->pool_size;
1da177e4 468
e4e11e38 469 for (i = 0; i < num ; i++) {
d592a996 470 WARN_ON(sh->dev[i].page != sh->dev[i].orig_page);
1da177e4
LT		 471 p = sh->dev[i].page;
472 if (!p)
473 continue;
474 sh->dev[i].page = NULL;
2d1f3b5d 475 put_page(p);
1da177e4
LT		 476 }
477}
478
a9683a79 479static int grow_buffers(struct stripe_head *sh, gfp_t gfp)
1da177e4
LT		 480{
481 int i;
e4e11e38 482 int num = sh->raid_conf->pool_size;
1da177e4 483
e4e11e38 484 for (i = 0; i < num; i++) {
1da177e4
LT		 485 struct page *page;
486
a9683a79 487 if (!(page = alloc_page(gfp))) {
1da177e4
LT		 488 return 1;
489 }
490 sh->dev[i].page = page;
d592a996 491 sh->dev[i].orig_page = page;
1da177e4 492 }
3418d036 493
1da177e4
LT		 494 return 0;
495}
496
d1688a6d 497static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
911d4ee8 498 struct stripe_head *sh);
1da177e4 499
b5663ba4 500static void init_stripe(struct stripe_head *sh, sector_t sector, int previous)
1da177e4 501{
d1688a6d 502 struct r5conf *conf = sh->raid_conf;
566c09c5 503 int i, seq;
1da177e4 504
78bafebd
ES		 505 BUG_ON(atomic_read(&sh->count) != 0);
506 BUG_ON(test_bit(STRIPE_HANDLE, &sh->state));
600aa109 507 BUG_ON(stripe_operations_active(sh));
59fc630b 508 BUG_ON(sh->batch_head);
d84e0f10 509
45b4233c 510 pr_debug("init_stripe called, stripe %llu\n",
b8e6a15a 511 (unsigned long long)sector);
566c09c5
SL		 512retry:
513 seq = read_seqcount_begin(&conf->gen_lock);
86b42c71 514 sh->generation = conf->generation - previous;
b5663ba4 515 sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks;
1da177e4 516 sh->sector = sector;
911d4ee8 517 stripe_set_idx(sector, conf, previous, sh);
1da177e4
LT		 518 sh->state = 0;
519
7ecaa1e6 520 for (i = sh->disks; i--; ) {
1da177e4
LT		 521 struct r5dev *dev = &sh->dev[i];
522
d84e0f10 523 if (dev->toread || dev->read || dev->towrite || dev->written ||
1da177e4 524 test_bit(R5_LOCKED, &dev->flags)) {
cc6167b4 525 pr_err("sector=%llx i=%d %p %p %p %p %d\n",
1da177e4 526 (unsigned long long)sh->sector, i, dev->toread,
d84e0f10 527 dev->read, dev->towrite, dev->written,
1da177e4 528 test_bit(R5_LOCKED, &dev->flags));
8cfa7b0f 529 WARN_ON(1);
1da177e4
LT		 530 }
531 dev->flags = 0;
27a4ff8f 532 dev->sector = raid5_compute_blocknr(sh, i, previous);
1da177e4 533 }
566c09c5
SL		 534 if (read_seqcount_retry(&conf->gen_lock, seq))
535 goto retry;
7a87f434 536 sh->overwrite_disks = 0;
1da177e4 537 insert_hash(conf, sh);
851c30c9 538 sh->cpu = smp_processor_id();
da41ba65 539 set_bit(STRIPE_BATCH_READY, &sh->state);
1da177e4
LT		 540}
541
d1688a6d 542static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector,
86b42c71 543 short generation)
1da177e4
LT		 544{
545 struct stripe_head *sh;
546
45b4233c 547 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
b67bfe0d 548 hlist_for_each_entry(sh, stripe_hash(conf, sector), hash)
86b42c71 549 if (sh->sector == sector && sh->generation == generation)
1da177e4 550 return sh;
45b4233c 551 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
1da177e4
LT		 552 return NULL;
553}
554
674806d6
N		 555/*
556 * Need to check if array has failed when deciding whether to:
557 * - start an array
558 * - remove non-faulty devices
559 * - add a spare
560 * - allow a reshape
561 * This determination is simple when no reshape is happening.
562 * However if there is a reshape, we need to carefully check
563 * both the before and after sections.
564 * This is because some failed devices may only affect one
565 * of the two sections, and some non-in_sync devices may
566 * be insync in the section most affected by failed devices.
567 */
2e38a37f 568int raid5_calc_degraded(struct r5conf *conf)
674806d6 569{
908f4fbd 570 int degraded, degraded2;
674806d6 571 int i;
674806d6
N		 572
573 rcu_read_lock();
574 degraded = 0;
575 for (i = 0; i < conf->previous_raid_disks; i++) {
3cb03002 576 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
e5c86471
N		 577 if (rdev && test_bit(Faulty, &rdev->flags))
578 rdev = rcu_dereference(conf->disks[i].replacement);
674806d6
N		 579 if (!rdev || test_bit(Faulty, &rdev->flags))
580 degraded++;
581 else if (test_bit(In_sync, &rdev->flags))
582 ;
583 else
584 /* not in-sync or faulty.
585 * If the reshape increases the number of devices,
586 * this is being recovered by the reshape, so
587 * this 'previous' section is not in_sync.
588 * If the number of devices is being reduced however,
589 * the device can only be part of the array if
590 * we are reverting a reshape, so this section will
591 * be in-sync.
592 */
593 if (conf->raid_disks >= conf->previous_raid_disks)
594 degraded++;
595 }
596 rcu_read_unlock();
908f4fbd
N		 597 if (conf->raid_disks == conf->previous_raid_disks)
598 return degraded;
674806d6 599 rcu_read_lock();
908f4fbd 600 degraded2 = 0;
674806d6 601 for (i = 0; i < conf->raid_disks; i++) {
3cb03002 602 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
e5c86471
N		 603 if (rdev && test_bit(Faulty, &rdev->flags))
604 rdev = rcu_dereference(conf->disks[i].replacement);
674806d6 605 if (!rdev || test_bit(Faulty, &rdev->flags))
908f4fbd 606 degraded2++;
674806d6
N		 607 else if (test_bit(In_sync, &rdev->flags))
608 ;
609 else
610 /* not in-sync or faulty.
611 * If reshape increases the number of devices, this
612 * section has already been recovered, else it
613 * almost certainly hasn't.
614 */
615 if (conf->raid_disks <= conf->previous_raid_disks)
908f4fbd 616 degraded2++;
674806d6
N		 617 }
618 rcu_read_unlock();
908f4fbd
N		 619 if (degraded2 > degraded)
620 return degraded2;
621 return degraded;
622}
623
624static int has_failed(struct r5conf *conf)
625{
626 int degraded;
627
628 if (conf->mddev->reshape_position == MaxSector)
629 return conf->mddev->degraded > conf->max_degraded;
630
2e38a37f 631 degraded = raid5_calc_degraded(conf);
674806d6
N		 632 if (degraded > conf->max_degraded)
633 return 1;
634 return 0;
635}
636
6d036f7d
SL		 637struct stripe_head *
638raid5_get_active_stripe(struct r5conf *conf, sector_t sector,
639 int previous, int noblock, int noquiesce)
1da177e4
LT		 640{
641 struct stripe_head *sh;
566c09c5 642 int hash = stripe_hash_locks_hash(sector);
ff00d3b4 643 int inc_empty_inactive_list_flag;
1da177e4 644
45b4233c 645 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector);
1da177e4 646
566c09c5 647 spin_lock_irq(conf->hash_locks + hash);
1da177e4
LT		 648
649 do {
b1b46486 650 wait_event_lock_irq(conf->wait_for_quiescent,
a8c906ca 651 conf->quiesce == 0 || noquiesce,
566c09c5 652 *(conf->hash_locks + hash));
86b42c71 653 sh = __find_stripe(conf, sector, conf->generation - previous);
1da177e4 654 if (!sh) {
edbe83ab 655 if (!test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) {
566c09c5 656 sh = get_free_stripe(conf, hash);
713bc5c2
SL		 657 if (!sh && !test_bit(R5_DID_ALLOC,
658 &conf->cache_state))
edbe83ab
N		 659 set_bit(R5_ALLOC_MORE,
660 &conf->cache_state);
661 }
1da177e4
LT		 662 if (noblock && sh == NULL)
663 break;
a39f7afd
SL		 664
665 r5c_check_stripe_cache_usage(conf);
1da177e4 666 if (!sh) {
5423399a
N		 667 set_bit(R5_INACTIVE_BLOCKED,
668 &conf->cache_state);
a39f7afd 669 r5l_wake_reclaim(conf->log, 0);
6ab2a4b8
SL		 670 wait_event_lock_irq(
671 conf->wait_for_stripe,
566c09c5
SL		 672 !list_empty(conf->inactive_list + hash) &&
673 (atomic_read(&conf->active_stripes)
674 < (conf->max_nr_stripes * 3 / 4)
5423399a
N		 675 || !test_bit(R5_INACTIVE_BLOCKED,
676 &conf->cache_state)),
6ab2a4b8 677 *(conf->hash_locks + hash));
5423399a
N		 678 clear_bit(R5_INACTIVE_BLOCKED,
679 &conf->cache_state);
7da9d450 680 } else {
b5663ba4 681 init_stripe(sh, sector, previous);
7da9d450
N		 682 atomic_inc(&sh->count);
683 }
e240c183 684 } else if (!atomic_inc_not_zero(&sh->count)) {
6d183de4 685 spin_lock(&conf->device_lock);
e240c183 686 if (!atomic_read(&sh->count)) {
1da177e4
LT		 687 if (!test_bit(STRIPE_HANDLE, &sh->state))
688 atomic_inc(&conf->active_stripes);
5af9bef7
N		 689 BUG_ON(list_empty(&sh->lru) &&
690 !test_bit(STRIPE_EXPANDING, &sh->state));
ff00d3b4
ZL		 691 inc_empty_inactive_list_flag = 0;
692 if (!list_empty(conf->inactive_list + hash))
693 inc_empty_inactive_list_flag = 1;
16a53ecc 694 list_del_init(&sh->lru);
ff00d3b4
ZL		 695 if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag)
696 atomic_inc(&conf->empty_inactive_list_nr);
bfc90cb0
SL		 697 if (sh->group) {
698 sh->group->stripes_cnt--;
699 sh->group = NULL;
700 }
1da177e4 701 }
7da9d450 702 atomic_inc(&sh->count);
6d183de4 703 spin_unlock(&conf->device_lock);
1da177e4
LT		 704 }
705 } while (sh == NULL);
706
566c09c5 707 spin_unlock_irq(conf->hash_locks + hash);
1da177e4
LT		 708 return sh;
709}
710
7a87f434 711static bool is_full_stripe_write(struct stripe_head *sh)
712{
713 BUG_ON(sh->overwrite_disks > (sh->disks - sh->raid_conf->max_degraded));
714 return sh->overwrite_disks == (sh->disks - sh->raid_conf->max_degraded);
715}
716
59fc630b 717static void lock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2)
718{
59fc630b 719 if (sh1 > sh2) {
3d05f3ae 720 spin_lock_irq(&sh2->stripe_lock);
59fc630b 721 spin_lock_nested(&sh1->stripe_lock, 1);
722 } else {
3d05f3ae 723 spin_lock_irq(&sh1->stripe_lock);
59fc630b 724 spin_lock_nested(&sh2->stripe_lock, 1);
725 }
726}
727
728static void unlock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2)
729{
730 spin_unlock(&sh1->stripe_lock);
3d05f3ae 731 spin_unlock_irq(&sh2->stripe_lock);
59fc630b 732}
733
734/* Only freshly new full stripe normal write stripe can be added to a batch list */
735static bool stripe_can_batch(struct stripe_head *sh)
736{
9c3e333d
SL		 737 struct r5conf *conf = sh->raid_conf;
738
3418d036 739 if (conf->log || raid5_has_ppl(conf))
9c3e333d 740 return false;
59fc630b 741 return test_bit(STRIPE_BATCH_READY, &sh->state) &&
d0852df5 742 !test_bit(STRIPE_BITMAP_PENDING, &sh->state) &&
59fc630b 743 is_full_stripe_write(sh);
744}
745
746/* we only do back search */
747static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh)
748{
749 struct stripe_head *head;
750 sector_t head_sector, tmp_sec;
751 int hash;
752 int dd_idx;
ff00d3b4 753 int inc_empty_inactive_list_flag;
59fc630b 754
59fc630b 755 /* Don't cross chunks, so stripe pd_idx/qd_idx is the same */
756 tmp_sec = sh->sector;
757 if (!sector_div(tmp_sec, conf->chunk_sectors))
758 return;
759 head_sector = sh->sector - STRIPE_SECTORS;
760
761 hash = stripe_hash_locks_hash(head_sector);
762 spin_lock_irq(conf->hash_locks + hash);
763 head = __find_stripe(conf, head_sector, conf->generation);
764 if (head && !atomic_inc_not_zero(&head->count)) {
765 spin_lock(&conf->device_lock);
766 if (!atomic_read(&head->count)) {
767 if (!test_bit(STRIPE_HANDLE, &head->state))
768 atomic_inc(&conf->active_stripes);
769 BUG_ON(list_empty(&head->lru) &&
770 !test_bit(STRIPE_EXPANDING, &head->state));
ff00d3b4
ZL		 771 inc_empty_inactive_list_flag = 0;
772 if (!list_empty(conf->inactive_list + hash))
773 inc_empty_inactive_list_flag = 1;
59fc630b 774 list_del_init(&head->lru);
ff00d3b4
ZL		 775 if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag)
776 atomic_inc(&conf->empty_inactive_list_nr);
59fc630b 777 if (head->group) {
778 head->group->stripes_cnt--;
779 head->group = NULL;
780 }
781 }
782 atomic_inc(&head->count);
783 spin_unlock(&conf->device_lock);
784 }
785 spin_unlock_irq(conf->hash_locks + hash);
786
787 if (!head)
788 return;
789 if (!stripe_can_batch(head))
790 goto out;
791
792 lock_two_stripes(head, sh);
793 /* clear_batch_ready clear the flag */
794 if (!stripe_can_batch(head) || !stripe_can_batch(sh))
795 goto unlock_out;
796
797 if (sh->batch_head)
798 goto unlock_out;
799
800 dd_idx = 0;
801 while (dd_idx == sh->pd_idx || dd_idx == sh->qd_idx)
802 dd_idx++;
1eff9d32 803 if (head->dev[dd_idx].towrite->bi_opf != sh->dev[dd_idx].towrite->bi_opf ||
796a5cf0 804 bio_op(head->dev[dd_idx].towrite) != bio_op(sh->dev[dd_idx].towrite))
59fc630b 805 goto unlock_out;
806
807 if (head->batch_head) {
808 spin_lock(&head->batch_head->batch_lock);
809 /* This batch list is already running */
810 if (!stripe_can_batch(head)) {
811 spin_unlock(&head->batch_head->batch_lock);
812 goto unlock_out;
813 }
3664847d
SL		 814 /*
815 * We must assign batch_head of this stripe within the
816 * batch_lock, otherwise clear_batch_ready of batch head
817 * stripe could clear BATCH_READY bit of this stripe and
818 * this stripe->batch_head doesn't get assigned, which
819 * could confuse clear_batch_ready for this stripe
820 */
821 sh->batch_head = head->batch_head;
59fc630b 822
823 /*
824 * at this point, head's BATCH_READY could be cleared, but we
825 * can still add the stripe to batch list
826 */
827 list_add(&sh->batch_list, &head->batch_list);
828 spin_unlock(&head->batch_head->batch_lock);
59fc630b 829 } else {
830 head->batch_head = head;
831 sh->batch_head = head->batch_head;
832 spin_lock(&head->batch_lock);
833 list_add_tail(&sh->batch_list, &head->batch_list);
834 spin_unlock(&head->batch_lock);
835 }
836
837 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
838 if (atomic_dec_return(&conf->preread_active_stripes)
839 < IO_THRESHOLD)
840 md_wakeup_thread(conf->mddev->thread);
841
2b6b2457
N		 842 if (test_and_clear_bit(STRIPE_BIT_DELAY, &sh->state)) {
843 int seq = sh->bm_seq;
844 if (test_bit(STRIPE_BIT_DELAY, &sh->batch_head->state) &&
845 sh->batch_head->bm_seq > seq)
846 seq = sh->batch_head->bm_seq;
847 set_bit(STRIPE_BIT_DELAY, &sh->batch_head->state);
848 sh->batch_head->bm_seq = seq;
849 }
850
59fc630b 851 atomic_inc(&sh->count);
852unlock_out:
853 unlock_two_stripes(head, sh);
854out:
6d036f7d 855 raid5_release_stripe(head);
59fc630b 856}
857
05616be5
N		 858/* Determine if 'data_offset' or 'new_data_offset' should be used
859 * in this stripe_head.
860 */
861static int use_new_offset(struct r5conf *conf, struct stripe_head *sh)
862{
863 sector_t progress = conf->reshape_progress;
864 /* Need a memory barrier to make sure we see the value
865 * of conf->generation, or ->data_offset that was set before
866 * reshape_progress was updated.
867 */
868 smp_rmb();
869 if (progress == MaxSector)
870 return 0;
871 if (sh->generation == conf->generation - 1)
872 return 0;
873 /* We are in a reshape, and this is a new-generation stripe,
874 * so use new_data_offset.
875 */
876 return 1;
877}
878
aaf9f12e 879static void dispatch_bio_list(struct bio_list *tmp)
765d704d 880{
765d704d
SL		 881 struct bio *bio;
882
aaf9f12e
SL		 883 while ((bio = bio_list_pop(tmp)))
884 generic_make_request(bio);
885}
886
887static int cmp_stripe(void *priv, struct list_head *a, struct list_head *b)
888{
889 const struct r5pending_data *da = list_entry(a,
890 struct r5pending_data, sibling);
891 const struct r5pending_data *db = list_entry(b,
892 struct r5pending_data, sibling);
893 if (da->sector > db->sector)
894 return 1;
895 if (da->sector < db->sector)
896 return -1;
897 return 0;
898}
899
900static void dispatch_defer_bios(struct r5conf *conf, int target,
901 struct bio_list *list)
902{
903 struct r5pending_data *data;
904 struct list_head *first, *next = NULL;
905 int cnt = 0;
906
907 if (conf->pending_data_cnt == 0)
908 return;
909
910 list_sort(NULL, &conf->pending_list, cmp_stripe);
911
912 first = conf->pending_list.next;
913
914 /* temporarily move the head */
915 if (conf->next_pending_data)
916 list_move_tail(&conf->pending_list,
917 &conf->next_pending_data->sibling);
918
919 while (!list_empty(&conf->pending_list)) {
920 data = list_first_entry(&conf->pending_list,
921 struct r5pending_data, sibling);
922 if (&data->sibling == first)
923 first = data->sibling.next;
924 next = data->sibling.next;
925
926 bio_list_merge(list, &data->bios);
927 list_move(&data->sibling, &conf->free_list);
928 cnt++;
929 if (cnt >= target)
930 break;
931 }
932 conf->pending_data_cnt -= cnt;
933 BUG_ON(conf->pending_data_cnt < 0 || cnt < target);
934
935 if (next != &conf->pending_list)
936 conf->next_pending_data = list_entry(next,
937 struct r5pending_data, sibling);
938 else
939 conf->next_pending_data = NULL;
940 /* list isn't empty */
941 if (first != &conf->pending_list)
942 list_move_tail(&conf->pending_list, first);
943}
944
945static void flush_deferred_bios(struct r5conf *conf)
946{
947 struct bio_list tmp = BIO_EMPTY_LIST;
948
949 if (conf->pending_data_cnt == 0)
765d704d
SL		 950 return;
951
765d704d 952 spin_lock(&conf->pending_bios_lock);
aaf9f12e
SL		 953 dispatch_defer_bios(conf, conf->pending_data_cnt, &tmp);
954 BUG_ON(conf->pending_data_cnt != 0);
765d704d
SL		 955 spin_unlock(&conf->pending_bios_lock);
956
aaf9f12e 957 dispatch_bio_list(&tmp);
765d704d
SL		 958}
959
aaf9f12e
SL		 960static void defer_issue_bios(struct r5conf *conf, sector_t sector,
961 struct bio_list *bios)
765d704d 962{
aaf9f12e
SL		 963 struct bio_list tmp = BIO_EMPTY_LIST;
964 struct r5pending_data *ent;
965
765d704d 966 spin_lock(&conf->pending_bios_lock);
aaf9f12e
SL		 967 ent = list_first_entry(&conf->free_list, struct r5pending_data,
968 sibling);
969 list_move_tail(&ent->sibling, &conf->pending_list);
970 ent->sector = sector;
971 bio_list_init(&ent->bios);
972 bio_list_merge(&ent->bios, bios);
973 conf->pending_data_cnt++;
974 if (conf->pending_data_cnt >= PENDING_IO_MAX)
975 dispatch_defer_bios(conf, PENDING_IO_ONE_FLUSH, &tmp);
976
765d704d 977 spin_unlock(&conf->pending_bios_lock);
aaf9f12e
SL		 978
979 dispatch_bio_list(&tmp);
765d704d
SL		 980}
981
6712ecf8 982static void
4246a0b6 983raid5_end_read_request(struct bio *bi);
6712ecf8 984static void
4246a0b6 985raid5_end_write_request(struct bio *bi);
91c00924 986
c4e5ac0a 987static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
91c00924 988{
d1688a6d 989 struct r5conf *conf = sh->raid_conf;
91c00924 990 int i, disks = sh->disks;
59fc630b 991 struct stripe_head *head_sh = sh;
aaf9f12e
SL		 992 struct bio_list pending_bios = BIO_EMPTY_LIST;
993 bool should_defer;
91c00924
DW		 994
995 might_sleep();
996
ff875738
AP		 997 if (log_stripe(sh, s) == 0)
998 return;
1e6d690b 999
aaf9f12e 1000 should_defer = conf->batch_bio_dispatch && conf->group_cnt;
1e6d690b 1001
91c00924 1002 for (i = disks; i--; ) {
796a5cf0 1003 int op, op_flags = 0;
9a3e1101 1004 int replace_only = 0;
977df362
N		 1005 struct bio *bi, *rbi;
1006 struct md_rdev *rdev, *rrdev = NULL;
59fc630b 1007
1008 sh = head_sh;
e9c7469b 1009 if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) {
796a5cf0 1010 op = REQ_OP_WRITE;
e9c7469b 1011 if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags))
70fd7614 1012 op_flags = REQ_FUA;
9e444768 1013 if (test_bit(R5_Discard, &sh->dev[i].flags))
796a5cf0 1014 op = REQ_OP_DISCARD;
e9c7469b 1015 } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags))
796a5cf0 1016 op = REQ_OP_READ;
9a3e1101
N		 1017 else if (test_and_clear_bit(R5_WantReplace,
1018 &sh->dev[i].flags)) {
796a5cf0 1019 op = REQ_OP_WRITE;
9a3e1101
N		 1020 replace_only = 1;
1021 } else
91c00924 1022 continue;
bc0934f0 1023 if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags))
796a5cf0 1024 op_flags |= REQ_SYNC;
91c00924 1025
59fc630b 1026again:
91c00924 1027 bi = &sh->dev[i].req;
977df362 1028 rbi = &sh->dev[i].rreq; /* For writing to replacement */
91c00924 1029
91c00924 1030 rcu_read_lock();
9a3e1101 1031 rrdev = rcu_dereference(conf->disks[i].replacement);
dd054fce
N		 1032 smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */
1033 rdev = rcu_dereference(conf->disks[i].rdev);
1034 if (!rdev) {
1035 rdev = rrdev;
1036 rrdev = NULL;
1037 }
796a5cf0 1038 if (op_is_write(op)) {
9a3e1101
N		 1039 if (replace_only)
1040 rdev = NULL;
dd054fce
N		 1041 if (rdev == rrdev)
1042 /* We raced and saw duplicates */
1043 rrdev = NULL;
9a3e1101 1044 } else {
59fc630b 1045 if (test_bit(R5_ReadRepl, &head_sh->dev[i].flags) && rrdev)
9a3e1101
N		 1046 rdev = rrdev;
1047 rrdev = NULL;
1048 }
977df362 1049
91c00924
DW		 1050 if (rdev && test_bit(Faulty, &rdev->flags))
1051 rdev = NULL;
1052 if (rdev)
1053 atomic_inc(&rdev->nr_pending);
977df362
N		 1054 if (rrdev && test_bit(Faulty, &rrdev->flags))
1055 rrdev = NULL;
1056 if (rrdev)
1057 atomic_inc(&rrdev->nr_pending);
91c00924
DW		 1058 rcu_read_unlock();
1059
73e92e51 1060 /* We have already checked bad blocks for reads. Now
977df362
N		 1061 * need to check for writes. We never accept write errors
1062 * on the replacement, so we don't to check rrdev.
73e92e51 1063 */
796a5cf0 1064 while (op_is_write(op) && rdev &&
73e92e51
N		 1065 test_bit(WriteErrorSeen, &rdev->flags)) {
1066 sector_t first_bad;
1067 int bad_sectors;
1068 int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
1069 &first_bad, &bad_sectors);
1070 if (!bad)
1071 break;
1072
1073 if (bad < 0) {
1074 set_bit(BlockedBadBlocks, &rdev->flags);
1075 if (!conf->mddev->external &&
2953079c 1076 conf->mddev->sb_flags) {
73e92e51
N		 1077 /* It is very unlikely, but we might
1078 * still need to write out the
1079 * bad block log - better give it
1080 * a chance*/
1081 md_check_recovery(conf->mddev);
1082 }
1850753d 1083 /*
1084 * Because md_wait_for_blocked_rdev
1085 * will dec nr_pending, we must
1086 * increment it first.
1087 */
1088 atomic_inc(&rdev->nr_pending);
73e92e51
N		 1089 md_wait_for_blocked_rdev(rdev, conf->mddev);
1090 } else {
1091 /* Acknowledged bad block - skip the write */
1092 rdev_dec_pending(rdev, conf->mddev);
1093 rdev = NULL;
1094 }
1095 }
1096
91c00924 1097 if (rdev) {
9a3e1101
N		 1098 if (s->syncing || s->expanding || s->expanded
1099 || s->replacing)
91c00924
DW		 1100 md_sync_acct(rdev->bdev, STRIPE_SECTORS);
1101
2b7497f0
DW		 1102 set_bit(STRIPE_IO_STARTED, &sh->state);
1103
74d46992 1104 bio_set_dev(bi, rdev->bdev);
796a5cf0
MC		 1105 bio_set_op_attrs(bi, op, op_flags);
1106 bi->bi_end_io = op_is_write(op)
2f6db2a7
KO		 1107 ? raid5_end_write_request
1108 : raid5_end_read_request;
1109 bi->bi_private = sh;
1110
6296b960 1111 pr_debug("%s: for %llu schedule op %d on disc %d\n",
e46b272b 1112 __func__, (unsigned long long)sh->sector,
1eff9d32 1113 bi->bi_opf, i);
91c00924 1114 atomic_inc(&sh->count);
59fc630b 1115 if (sh != head_sh)
1116 atomic_inc(&head_sh->count);
05616be5 1117 if (use_new_offset(conf, sh))
4f024f37 1118 bi->bi_iter.bi_sector = (sh->sector
05616be5
N		 1119 + rdev->new_data_offset);
1120 else
4f024f37 1121 bi->bi_iter.bi_sector = (sh->sector
05616be5 1122 + rdev->data_offset);
59fc630b 1123 if (test_bit(R5_ReadNoMerge, &head_sh->dev[i].flags))
1eff9d32 1124 bi->bi_opf |= REQ_NOMERGE;
3f9e7c14 1125
d592a996
SL		 1126 if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
1127 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
86aa1397
SL		 1128
1129 if (!op_is_write(op) &&
1130 test_bit(R5_InJournal, &sh->dev[i].flags))
1131 /*
1132 * issuing read for a page in journal, this
1133 * must be preparing for prexor in rmw; read
1134 * the data into orig_page
1135 */
1136 sh->dev[i].vec.bv_page = sh->dev[i].orig_page;
1137 else
1138 sh->dev[i].vec.bv_page = sh->dev[i].page;
4997b72e 1139 bi->bi_vcnt = 1;
91c00924
DW		 1140 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
1141 bi->bi_io_vec[0].bv_offset = 0;
4f024f37 1142 bi->bi_iter.bi_size = STRIPE_SIZE;
37c61ff3
SL		 1143 /*
1144 * If this is discard request, set bi_vcnt 0. We don't
1145 * want to confuse SCSI because SCSI will replace payload
1146 */
796a5cf0 1147 if (op == REQ_OP_DISCARD)
37c61ff3 1148 bi->bi_vcnt = 0;
977df362
N		 1149 if (rrdev)
1150 set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags);
e3620a3a
JB		 1151
1152 if (conf->mddev->gendisk)
74d46992 1153 trace_block_bio_remap(bi->bi_disk->queue,
e3620a3a
JB		 1154 bi, disk_devt(conf->mddev->gendisk),
1155 sh->dev[i].sector);
aaf9f12e
SL		 1156 if (should_defer && op_is_write(op))
1157 bio_list_add(&pending_bios, bi);
1158 else
1159 generic_make_request(bi);
977df362
N		 1160 }
1161 if (rrdev) {
9a3e1101
N		 1162 if (s->syncing || s->expanding || s->expanded
1163 || s->replacing)
977df362
N		 1164 md_sync_acct(rrdev->bdev, STRIPE_SECTORS);
1165
1166 set_bit(STRIPE_IO_STARTED, &sh->state);
1167
74d46992 1168 bio_set_dev(rbi, rrdev->bdev);
796a5cf0
MC		 1169 bio_set_op_attrs(rbi, op, op_flags);
1170 BUG_ON(!op_is_write(op));
2f6db2a7
KO		 1171 rbi->bi_end_io = raid5_end_write_request;
1172 rbi->bi_private = sh;
1173
6296b960 1174 pr_debug("%s: for %llu schedule op %d on "
977df362
N		 1175 "replacement disc %d\n",
1176 __func__, (unsigned long long)sh->sector,
1eff9d32 1177 rbi->bi_opf, i);
977df362 1178 atomic_inc(&sh->count);
59fc630b 1179 if (sh != head_sh)
1180 atomic_inc(&head_sh->count);
05616be5 1181 if (use_new_offset(conf, sh))
4f024f37 1182 rbi->bi_iter.bi_sector = (sh->sector
05616be5
N		 1183 + rrdev->new_data_offset);
1184 else
4f024f37 1185 rbi->bi_iter.bi_sector = (sh->sector
05616be5 1186 + rrdev->data_offset);
d592a996
SL		 1187 if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
1188 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
1189 sh->dev[i].rvec.bv_page = sh->dev[i].page;
4997b72e 1190 rbi->bi_vcnt = 1;
977df362
N		 1191 rbi->bi_io_vec[0].bv_len = STRIPE_SIZE;
1192 rbi->bi_io_vec[0].bv_offset = 0;
4f024f37 1193 rbi->bi_iter.bi_size = STRIPE_SIZE;
37c61ff3
SL		 1194 /*
1195 * If this is discard request, set bi_vcnt 0. We don't
1196 * want to confuse SCSI because SCSI will replace payload
1197 */
796a5cf0 1198 if (op == REQ_OP_DISCARD)
37c61ff3 1199 rbi->bi_vcnt = 0;
e3620a3a 1200 if (conf->mddev->gendisk)
74d46992 1201 trace_block_bio_remap(rbi->bi_disk->queue,
e3620a3a
JB		 1202 rbi, disk_devt(conf->mddev->gendisk),
1203 sh->dev[i].sector);
aaf9f12e
SL		 1204 if (should_defer && op_is_write(op))
1205 bio_list_add(&pending_bios, rbi);
1206 else
1207 generic_make_request(rbi);
977df362
N		 1208 }
1209 if (!rdev && !rrdev) {
796a5cf0 1210 if (op_is_write(op))
91c00924 1211 set_bit(STRIPE_DEGRADED, &sh->state);
6296b960 1212 pr_debug("skip op %d on disc %d for sector %llu\n",
1eff9d32 1213 bi->bi_opf, i, (unsigned long long)sh->sector);
91c00924
DW		 1214 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1215 set_bit(STRIPE_HANDLE, &sh->state);
1216 }
59fc630b 1217
1218 if (!head_sh->batch_head)
1219 continue;
1220 sh = list_first_entry(&sh->batch_list, struct stripe_head,
1221 batch_list);
1222 if (sh != head_sh)
1223 goto again;
91c00924 1224 }
aaf9f12e
SL		 1225
1226 if (should_defer && !bio_list_empty(&pending_bios))
1227 defer_issue_bios(conf, head_sh->sector, &pending_bios);
91c00924
DW		 1228}
1229
1230static struct dma_async_tx_descriptor *
d592a996
SL		 1231async_copy_data(int frombio, struct bio *bio, struct page **page,
1232 sector_t sector, struct dma_async_tx_descriptor *tx,
1e6d690b 1233 struct stripe_head *sh, int no_skipcopy)
91c00924 1234{
7988613b
KO		 1235 struct bio_vec bvl;
1236 struct bvec_iter iter;
91c00924 1237 struct page *bio_page;
91c00924 1238 int page_offset;
a08abd8c 1239 struct async_submit_ctl submit;
0403e382 1240 enum async_tx_flags flags = 0;
91c00924 1241
4f024f37
KO		 1242 if (bio->bi_iter.bi_sector >= sector)
1243 page_offset = (signed)(bio->bi_iter.bi_sector - sector) * 512;
91c00924 1244 else
4f024f37 1245 page_offset = (signed)(sector - bio->bi_iter.bi_sector) * -512;
a08abd8c 1246
0403e382
DW		 1247 if (frombio)
1248 flags |= ASYNC_TX_FENCE;
1249 init_async_submit(&submit, flags, tx, NULL, NULL, NULL);
1250
7988613b
KO		 1251 bio_for_each_segment(bvl, bio, iter) {
1252 int len = bvl.bv_len;
91c00924
DW		 1253 int clen;
1254 int b_offset = 0;
1255
1256 if (page_offset < 0) {
1257 b_offset = -page_offset;
1258 page_offset += b_offset;
1259 len -= b_offset;
1260 }
1261
1262 if (len > 0 && page_offset + len > STRIPE_SIZE)
1263 clen = STRIPE_SIZE - page_offset;
1264 else
1265 clen = len;
1266
1267 if (clen > 0) {
7988613b
KO		 1268 b_offset += bvl.bv_offset;
1269 bio_page = bvl.bv_page;
d592a996
SL		 1270 if (frombio) {
1271 if (sh->raid_conf->skip_copy &&
1272 b_offset == 0 && page_offset == 0 &&
1e6d690b
SL		 1273 clen == STRIPE_SIZE &&
1274 !no_skipcopy)
d592a996
SL		 1275 *page = bio_page;
1276 else
1277 tx = async_memcpy(*page, bio_page, page_offset,
a08abd8c 1278 b_offset, clen, &submit);
d592a996
SL		 1279 } else
1280 tx = async_memcpy(bio_page, *page, b_offset,
a08abd8c 1281 page_offset, clen, &submit);
91c00924 1282 }
a08abd8c
DW		 1283 /* chain the operations */
1284 submit.depend_tx = tx;
1285
91c00924
DW		 1286 if (clen < len) /* hit end of page */
1287 break;
1288 page_offset += len;
1289 }
1290
1291 return tx;
1292}
1293
1294static void ops_complete_biofill(void *stripe_head_ref)
1295{
1296 struct stripe_head *sh = stripe_head_ref;
e4d84909 1297 int i;
91c00924 1298
e46b272b 1299 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1300 (unsigned long long)sh->sector);
1301
1302 /* clear completed biofills */
1303 for (i = sh->disks; i--; ) {
1304 struct r5dev *dev = &sh->dev[i];
91c00924
DW		 1305
1306 /* acknowledge completion of a biofill operation */
e4d84909
DW		 1307 /* and check if we need to reply to a read request,
1308 * new R5_Wantfill requests are held off until
83de75cc 1309 * !STRIPE_BIOFILL_RUN
e4d84909
DW		 1310 */
1311 if (test_and_clear_bit(R5_Wantfill, &dev->flags)) {
91c00924 1312 struct bio *rbi, *rbi2;
91c00924 1313
91c00924
DW		 1314 BUG_ON(!dev->read);
1315 rbi = dev->read;
1316 dev->read = NULL;
4f024f37 1317 while (rbi && rbi->bi_iter.bi_sector <
91c00924
DW		 1318 dev->sector + STRIPE_SECTORS) {
1319 rbi2 = r5_next_bio(rbi, dev->sector);
016c76ac 1320 bio_endio(rbi);
91c00924
DW		 1321 rbi = rbi2;
1322 }
1323 }
1324 }
83de75cc 1325 clear_bit(STRIPE_BIOFILL_RUN, &sh->state);
91c00924 1326
e4d84909 1327 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 1328 raid5_release_stripe(sh);
91c00924
DW		 1329}
1330
1331static void ops_run_biofill(struct stripe_head *sh)
1332{
1333 struct dma_async_tx_descriptor *tx = NULL;
a08abd8c 1334 struct async_submit_ctl submit;
91c00924
DW		 1335 int i;
1336
59fc630b 1337 BUG_ON(sh->batch_head);
e46b272b 1338 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1339 (unsigned long long)sh->sector);
1340
1341 for (i = sh->disks; i--; ) {
1342 struct r5dev *dev = &sh->dev[i];
1343 if (test_bit(R5_Wantfill, &dev->flags)) {
1344 struct bio *rbi;
b17459c0 1345 spin_lock_irq(&sh->stripe_lock);
91c00924
DW		 1346 dev->read = rbi = dev->toread;
1347 dev->toread = NULL;
b17459c0 1348 spin_unlock_irq(&sh->stripe_lock);
4f024f37 1349 while (rbi && rbi->bi_iter.bi_sector <
91c00924 1350 dev->sector + STRIPE_SECTORS) {
d592a996 1351 tx = async_copy_data(0, rbi, &dev->page,
1e6d690b 1352 dev->sector, tx, sh, 0);
91c00924
DW		 1353 rbi = r5_next_bio(rbi, dev->sector);
1354 }
1355 }
1356 }
1357
1358 atomic_inc(&sh->count);
a08abd8c
DW		 1359 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL);
1360 async_trigger_callback(&submit);
91c00924
DW		 1361}
1362
4e7d2c0a 1363static void mark_target_uptodate(struct stripe_head *sh, int target)
91c00924 1364{
4e7d2c0a 1365 struct r5dev *tgt;
91c00924 1366
4e7d2c0a
DW		 1367 if (target < 0)
1368 return;
91c00924 1369
4e7d2c0a 1370 tgt = &sh->dev[target];
91c00924
DW		 1371 set_bit(R5_UPTODATE, &tgt->flags);
1372 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1373 clear_bit(R5_Wantcompute, &tgt->flags);
4e7d2c0a
DW		 1374}
1375
ac6b53b6 1376static void ops_complete_compute(void *stripe_head_ref)
91c00924
DW		 1377{
1378 struct stripe_head *sh = stripe_head_ref;
91c00924 1379
e46b272b 1380 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1381 (unsigned long long)sh->sector);
1382
ac6b53b6 1383 /* mark the computed target(s) as uptodate */
4e7d2c0a 1384 mark_target_uptodate(sh, sh->ops.target);
ac6b53b6 1385 mark_target_uptodate(sh, sh->ops.target2);
4e7d2c0a 1386
ecc65c9b
DW		 1387 clear_bit(STRIPE_COMPUTE_RUN, &sh->state);
1388 if (sh->check_state == check_state_compute_run)
1389 sh->check_state = check_state_compute_result;
91c00924 1390 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 1391 raid5_release_stripe(sh);
91c00924
DW		 1392}
1393
d6f38f31
DW		 1394/* return a pointer to the address conversion region of the scribble buffer */
1395static addr_conv_t *to_addr_conv(struct stripe_head *sh,
46d5b785 1396 struct raid5_percpu *percpu, int i)
d6f38f31 1397{
46d5b785 1398 void *addr;
1399
1400 addr = flex_array_get(percpu->scribble, i);
1401 return addr + sizeof(struct page *) * (sh->disks + 2);
1402}
1403
1404/* return a pointer to the address conversion region of the scribble buffer */
1405static struct page **to_addr_page(struct raid5_percpu *percpu, int i)
1406{
1407 void *addr;
1408
1409 addr = flex_array_get(percpu->scribble, i);
1410 return addr;
d6f38f31
DW		 1411}
1412
1413static struct dma_async_tx_descriptor *
1414ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 1415{
91c00924 1416 int disks = sh->disks;
46d5b785 1417 struct page **xor_srcs = to_addr_page(percpu, 0);
91c00924
DW		 1418 int target = sh->ops.target;
1419 struct r5dev *tgt = &sh->dev[target];
1420 struct page *xor_dest = tgt->page;
1421 int count = 0;
1422 struct dma_async_tx_descriptor *tx;
a08abd8c 1423 struct async_submit_ctl submit;
91c00924
DW		 1424 int i;
1425
59fc630b 1426 BUG_ON(sh->batch_head);
1427
91c00924 1428 pr_debug("%s: stripe %llu block: %d\n",
e46b272b 1429 __func__, (unsigned long long)sh->sector, target);
91c00924
DW		 1430 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1431
1432 for (i = disks; i--; )
1433 if (i != target)
1434 xor_srcs[count++] = sh->dev[i].page;
1435
1436 atomic_inc(&sh->count);
1437
0403e382 1438 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL,
46d5b785 1439 ops_complete_compute, sh, to_addr_conv(sh, percpu, 0));
91c00924 1440 if (unlikely(count == 1))
a08abd8c 1441 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
91c00924 1442 else
a08abd8c 1443 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924 1444
91c00924
DW		 1445 return tx;
1446}
1447
ac6b53b6
DW		 1448/* set_syndrome_sources - populate source buffers for gen_syndrome
1449 * @srcs - (struct page *) array of size sh->disks
1450 * @sh - stripe_head to parse
1451 *
1452 * Populates srcs in proper layout order for the stripe and returns the
1453 * 'count' of sources to be used in a call to async_gen_syndrome. The P
1454 * destination buffer is recorded in srcs[count] and the Q destination
1455 * is recorded in srcs[count+1]].
1456 */
584acdd4
MS		 1457static int set_syndrome_sources(struct page **srcs,
1458 struct stripe_head *sh,
1459 int srctype)
ac6b53b6
DW		 1460{
1461 int disks = sh->disks;
1462 int syndrome_disks = sh->ddf_layout ? disks : (disks - 2);
1463 int d0_idx = raid6_d0(sh);
1464 int count;
1465 int i;
1466
1467 for (i = 0; i < disks; i++)
5dd33c9a 1468 srcs[i] = NULL;
ac6b53b6
DW		 1469
1470 count = 0;
1471 i = d0_idx;
1472 do {
1473 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
584acdd4 1474 struct r5dev *dev = &sh->dev[i];
ac6b53b6 1475
584acdd4
MS		 1476 if (i == sh->qd_idx || i == sh->pd_idx ||
1477 (srctype == SYNDROME_SRC_ALL) ||
1478 (srctype == SYNDROME_SRC_WANT_DRAIN &&
1e6d690b
SL		 1479 (test_bit(R5_Wantdrain, &dev->flags) ||
1480 test_bit(R5_InJournal, &dev->flags))) ||
584acdd4 1481 (srctype == SYNDROME_SRC_WRITTEN &&
0977762f
SL		 1482 (dev->written ||
1483 test_bit(R5_InJournal, &dev->flags)))) {
1e6d690b
SL		 1484 if (test_bit(R5_InJournal, &dev->flags))
1485 srcs[slot] = sh->dev[i].orig_page;
1486 else
1487 srcs[slot] = sh->dev[i].page;
1488 }
ac6b53b6
DW		 1489 i = raid6_next_disk(i, disks);
1490 } while (i != d0_idx);
ac6b53b6 1491
e4424fee 1492 return syndrome_disks;
ac6b53b6
DW		 1493}
1494
1495static struct dma_async_tx_descriptor *
1496ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu)
1497{
1498 int disks = sh->disks;
46d5b785 1499 struct page **blocks = to_addr_page(percpu, 0);
ac6b53b6
DW		 1500 int target;
1501 int qd_idx = sh->qd_idx;
1502 struct dma_async_tx_descriptor *tx;
1503 struct async_submit_ctl submit;
1504 struct r5dev *tgt;
1505 struct page *dest;
1506 int i;
1507 int count;
1508
59fc630b 1509 BUG_ON(sh->batch_head);
ac6b53b6
DW		 1510 if (sh->ops.target < 0)
1511 target = sh->ops.target2;
1512 else if (sh->ops.target2 < 0)
1513 target = sh->ops.target;
91c00924 1514 else
ac6b53b6
DW		 1515 /* we should only have one valid target */
1516 BUG();
1517 BUG_ON(target < 0);
1518 pr_debug("%s: stripe %llu block: %d\n",
1519 __func__, (unsigned long long)sh->sector, target);
1520
1521 tgt = &sh->dev[target];
1522 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1523 dest = tgt->page;
1524
1525 atomic_inc(&sh->count);
1526
1527 if (target == qd_idx) {
584acdd4 1528 count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL);
ac6b53b6
DW		 1529 blocks[count] = NULL; /* regenerating p is not necessary */
1530 BUG_ON(blocks[count+1] != dest); /* q should already be set */
0403e382
DW		 1531 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1532 ops_complete_compute, sh,
46d5b785 1533 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1534 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
1535 } else {
1536 /* Compute any data- or p-drive using XOR */
1537 count = 0;
1538 for (i = disks; i-- ; ) {
1539 if (i == target || i == qd_idx)
1540 continue;
1541 blocks[count++] = sh->dev[i].page;
1542 }
1543
0403e382
DW		 1544 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1545 NULL, ops_complete_compute, sh,
46d5b785 1546 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1547 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit);
1548 }
91c00924 1549
91c00924
DW		 1550 return tx;
1551}
1552
ac6b53b6
DW		 1553static struct dma_async_tx_descriptor *
1554ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu)
1555{
1556 int i, count, disks = sh->disks;
1557 int syndrome_disks = sh->ddf_layout ? disks : disks-2;
1558 int d0_idx = raid6_d0(sh);
1559 int faila = -1, failb = -1;
1560 int target = sh->ops.target;
1561 int target2 = sh->ops.target2;
1562 struct r5dev *tgt = &sh->dev[target];
1563 struct r5dev *tgt2 = &sh->dev[target2];
1564 struct dma_async_tx_descriptor *tx;
46d5b785 1565 struct page **blocks = to_addr_page(percpu, 0);
ac6b53b6
DW		 1566 struct async_submit_ctl submit;
1567
59fc630b 1568 BUG_ON(sh->batch_head);
ac6b53b6
DW		 1569 pr_debug("%s: stripe %llu block1: %d block2: %d\n",
1570 __func__, (unsigned long long)sh->sector, target, target2);
1571 BUG_ON(target < 0 || target2 < 0);
1572 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1573 BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags));
1574
6c910a78 1575 /* we need to open-code set_syndrome_sources to handle the
ac6b53b6
DW		 1576 * slot number conversion for 'faila' and 'failb'
1577 */
1578 for (i = 0; i < disks ; i++)
5dd33c9a 1579 blocks[i] = NULL;
ac6b53b6
DW		 1580 count = 0;
1581 i = d0_idx;
1582 do {
1583 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
1584
1585 blocks[slot] = sh->dev[i].page;
1586
1587 if (i == target)
1588 faila = slot;
1589 if (i == target2)
1590 failb = slot;
1591 i = raid6_next_disk(i, disks);
1592 } while (i != d0_idx);
ac6b53b6
DW		 1593
1594 BUG_ON(faila == failb);
1595 if (failb < faila)
1596 swap(faila, failb);
1597 pr_debug("%s: stripe: %llu faila: %d failb: %d\n",
1598 __func__, (unsigned long long)sh->sector, faila, failb);
1599
1600 atomic_inc(&sh->count);
1601
1602 if (failb == syndrome_disks+1) {
1603 /* Q disk is one of the missing disks */
1604 if (faila == syndrome_disks) {
1605 /* Missing P+Q, just recompute */
0403e382
DW		 1606 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1607 ops_complete_compute, sh,
46d5b785 1608 to_addr_conv(sh, percpu, 0));
e4424fee 1609 return async_gen_syndrome(blocks, 0, syndrome_disks+2,
ac6b53b6
DW		 1610 STRIPE_SIZE, &submit);
1611 } else {
1612 struct page *dest;
1613 int data_target;
1614 int qd_idx = sh->qd_idx;
1615
1616 /* Missing D+Q: recompute D from P, then recompute Q */
1617 if (target == qd_idx)
1618 data_target = target2;
1619 else
1620 data_target = target;
1621
1622 count = 0;
1623 for (i = disks; i-- ; ) {
1624 if (i == data_target || i == qd_idx)
1625 continue;
1626 blocks[count++] = sh->dev[i].page;
1627 }
1628 dest = sh->dev[data_target].page;
0403e382
DW		 1629 init_async_submit(&submit,
1630 ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1631 NULL, NULL, NULL,
46d5b785 1632 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1633 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE,
1634 &submit);
1635
584acdd4 1636 count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL);
0403e382
DW		 1637 init_async_submit(&submit, ASYNC_TX_FENCE, tx,
1638 ops_complete_compute, sh,
46d5b785 1639 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1640 return async_gen_syndrome(blocks, 0, count+2,
1641 STRIPE_SIZE, &submit);
1642 }
ac6b53b6 1643 } else {
6c910a78
DW		 1644 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1645 ops_complete_compute, sh,
46d5b785 1646 to_addr_conv(sh, percpu, 0));
6c910a78
DW		 1647 if (failb == syndrome_disks) {
1648 /* We're missing D+P. */
1649 return async_raid6_datap_recov(syndrome_disks+2,
1650 STRIPE_SIZE, faila,
1651 blocks, &submit);
1652 } else {
1653 /* We're missing D+D. */
1654 return async_raid6_2data_recov(syndrome_disks+2,
1655 STRIPE_SIZE, faila, failb,
1656 blocks, &submit);
1657 }
ac6b53b6
DW		 1658 }
1659}
1660
91c00924
DW		 1661static void ops_complete_prexor(void *stripe_head_ref)
1662{
1663 struct stripe_head *sh = stripe_head_ref;
1664
e46b272b 1665 pr_debug("%s: stripe %llu\n", __func__,
91c00924 1666 (unsigned long long)sh->sector);
1e6d690b
SL		 1667
1668 if (r5c_is_writeback(sh->raid_conf->log))
1669 /*
1670 * raid5-cache write back uses orig_page during prexor.
1671 * After prexor, it is time to free orig_page
1672 */
1673 r5c_release_extra_page(sh);
91c00924
DW		 1674}
1675
1676static struct dma_async_tx_descriptor *
584acdd4
MS		 1677ops_run_prexor5(struct stripe_head *sh, struct raid5_percpu *percpu,
1678 struct dma_async_tx_descriptor *tx)
91c00924 1679{
91c00924 1680 int disks = sh->disks;
46d5b785 1681 struct page **xor_srcs = to_addr_page(percpu, 0);
91c00924 1682 int count = 0, pd_idx = sh->pd_idx, i;
a08abd8c 1683 struct async_submit_ctl submit;
91c00924
DW		 1684
1685 /* existing parity data subtracted */
1686 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1687
59fc630b 1688 BUG_ON(sh->batch_head);
e46b272b 1689 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1690 (unsigned long long)sh->sector);
1691
1692 for (i = disks; i--; ) {
1693 struct r5dev *dev = &sh->dev[i];
1694 /* Only process blocks that are known to be uptodate */
1e6d690b
SL		 1695 if (test_bit(R5_InJournal, &dev->flags))
1696 xor_srcs[count++] = dev->orig_page;
1697 else if (test_bit(R5_Wantdrain, &dev->flags))
91c00924
DW		 1698 xor_srcs[count++] = dev->page;
1699 }
1700
0403e382 1701 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
46d5b785 1702 ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0));
a08abd8c 1703 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924
DW		 1704
1705 return tx;
1706}
1707
584acdd4
MS		 1708static struct dma_async_tx_descriptor *
1709ops_run_prexor6(struct stripe_head *sh, struct raid5_percpu *percpu,
1710 struct dma_async_tx_descriptor *tx)
1711{
1712 struct page **blocks = to_addr_page(percpu, 0);
1713 int count;
1714 struct async_submit_ctl submit;
1715
1716 pr_debug("%s: stripe %llu\n", __func__,
1717 (unsigned long long)sh->sector);
1718
1719 count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_WANT_DRAIN);
1720
1721 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_PQ_XOR_DST, tx,
1722 ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0));
1723 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
1724
1725 return tx;
1726}
1727
91c00924 1728static struct dma_async_tx_descriptor *
d8ee0728 1729ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
91c00924 1730{
1e6d690b 1731 struct r5conf *conf = sh->raid_conf;
91c00924 1732 int disks = sh->disks;
d8ee0728 1733 int i;
59fc630b 1734 struct stripe_head *head_sh = sh;
91c00924 1735
e46b272b 1736 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1737 (unsigned long long)sh->sector);
1738
1739 for (i = disks; i--; ) {
59fc630b 1740 struct r5dev *dev;
91c00924 1741 struct bio *chosen;
91c00924 1742
59fc630b 1743 sh = head_sh;
1744 if (test_and_clear_bit(R5_Wantdrain, &head_sh->dev[i].flags)) {
91c00924
DW		 1745 struct bio *wbi;
1746
59fc630b 1747again:
1748 dev = &sh->dev[i];
1e6d690b
SL		 1749 /*
1750 * clear R5_InJournal, so when rewriting a page in
1751 * journal, it is not skipped by r5l_log_stripe()
1752 */
1753 clear_bit(R5_InJournal, &dev->flags);
b17459c0 1754 spin_lock_irq(&sh->stripe_lock);
91c00924
DW		 1755 chosen = dev->towrite;
1756 dev->towrite = NULL;
7a87f434 1757 sh->overwrite_disks = 0;
91c00924
DW		 1758 BUG_ON(dev->written);
1759 wbi = dev->written = chosen;
b17459c0 1760 spin_unlock_irq(&sh->stripe_lock);
d592a996 1761 WARN_ON(dev->page != dev->orig_page);
91c00924 1762
4f024f37 1763 while (wbi && wbi->bi_iter.bi_sector <
91c00924 1764 dev->sector + STRIPE_SECTORS) {
1eff9d32 1765 if (wbi->bi_opf & REQ_FUA)
e9c7469b 1766 set_bit(R5_WantFUA, &dev->flags);
1eff9d32 1767 if (wbi->bi_opf & REQ_SYNC)
bc0934f0 1768 set_bit(R5_SyncIO, &dev->flags);
796a5cf0 1769 if (bio_op(wbi) == REQ_OP_DISCARD)
620125f2 1770 set_bit(R5_Discard, &dev->flags);
d592a996
SL		 1771 else {
1772 tx = async_copy_data(1, wbi, &dev->page,
1e6d690b
SL		 1773 dev->sector, tx, sh,
1774 r5c_is_writeback(conf->log));
1775 if (dev->page != dev->orig_page &&
1776 !r5c_is_writeback(conf->log)) {
d592a996
SL		 1777 set_bit(R5_SkipCopy, &dev->flags);
1778 clear_bit(R5_UPTODATE, &dev->flags);
1779 clear_bit(R5_OVERWRITE, &dev->flags);
1780 }
1781 }
91c00924
DW		 1782 wbi = r5_next_bio(wbi, dev->sector);
1783 }
59fc630b 1784
1785 if (head_sh->batch_head) {
1786 sh = list_first_entry(&sh->batch_list,
1787 struct stripe_head,
1788 batch_list);
1789 if (sh == head_sh)
1790 continue;
1791 goto again;
1792 }
91c00924
DW		 1793 }
1794 }
1795
1796 return tx;
1797}
1798
ac6b53b6 1799static void ops_complete_reconstruct(void *stripe_head_ref)
91c00924
DW		 1800{
1801 struct stripe_head *sh = stripe_head_ref;
ac6b53b6
DW		 1802 int disks = sh->disks;
1803 int pd_idx = sh->pd_idx;
1804 int qd_idx = sh->qd_idx;
1805 int i;
9e444768 1806 bool fua = false, sync = false, discard = false;
91c00924 1807
e46b272b 1808 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1809 (unsigned long long)sh->sector);
1810
bc0934f0 1811 for (i = disks; i--; ) {
e9c7469b 1812 fua |= test_bit(R5_WantFUA, &sh->dev[i].flags);
bc0934f0 1813 sync |= test_bit(R5_SyncIO, &sh->dev[i].flags);
9e444768 1814 discard |= test_bit(R5_Discard, &sh->dev[i].flags);
bc0934f0 1815 }
e9c7469b 1816
91c00924
DW		 1817 for (i = disks; i--; ) {
1818 struct r5dev *dev = &sh->dev[i];
ac6b53b6 1819
e9c7469b 1820 if (dev->written || i == pd_idx || i == qd_idx) {
d592a996 1821 if (!discard && !test_bit(R5_SkipCopy, &dev->flags))
9e444768 1822 set_bit(R5_UPTODATE, &dev->flags);
e9c7469b
TH		 1823 if (fua)
1824 set_bit(R5_WantFUA, &dev->flags);
bc0934f0
SL		 1825 if (sync)
1826 set_bit(R5_SyncIO, &dev->flags);
e9c7469b 1827 }
91c00924
DW		 1828 }
1829
d8ee0728
DW		 1830 if (sh->reconstruct_state == reconstruct_state_drain_run)
1831 sh->reconstruct_state = reconstruct_state_drain_result;
1832 else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run)
1833 sh->reconstruct_state = reconstruct_state_prexor_drain_result;
1834 else {
1835 BUG_ON(sh->reconstruct_state != reconstruct_state_run);
1836 sh->reconstruct_state = reconstruct_state_result;
1837 }
91c00924
DW		 1838
1839 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 1840 raid5_release_stripe(sh);
91c00924
DW		 1841}
1842
1843static void
ac6b53b6
DW		 1844ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu,
1845 struct dma_async_tx_descriptor *tx)
91c00924 1846{
91c00924 1847 int disks = sh->disks;
59fc630b 1848 struct page **xor_srcs;
a08abd8c 1849 struct async_submit_ctl submit;
59fc630b 1850 int count, pd_idx = sh->pd_idx, i;
91c00924 1851 struct page *xor_dest;
d8ee0728 1852 int prexor = 0;
91c00924 1853 unsigned long flags;
59fc630b 1854 int j = 0;
1855 struct stripe_head *head_sh = sh;
1856 int last_stripe;
91c00924 1857
e46b272b 1858 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1859 (unsigned long long)sh->sector);
1860
620125f2
SL		 1861 for (i = 0; i < sh->disks; i++) {
1862 if (pd_idx == i)
1863 continue;
1864 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1865 break;
1866 }
1867 if (i >= sh->disks) {
1868 atomic_inc(&sh->count);
620125f2
SL		 1869 set_bit(R5_Discard, &sh->dev[pd_idx].flags);
1870 ops_complete_reconstruct(sh);
1871 return;
1872 }
59fc630b 1873again:
1874 count = 0;
1875 xor_srcs = to_addr_page(percpu, j);
91c00924
DW		 1876 /* check if prexor is active which means only process blocks
1877 * that are part of a read-modify-write (written)
1878 */
59fc630b 1879 if (head_sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
d8ee0728 1880 prexor = 1;
91c00924
DW		 1881 xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1882 for (i = disks; i--; ) {
1883 struct r5dev *dev = &sh->dev[i];
1e6d690b
SL		 1884 if (head_sh->dev[i].written ||
1885 test_bit(R5_InJournal, &head_sh->dev[i].flags))
91c00924
DW		 1886 xor_srcs[count++] = dev->page;
1887 }
1888 } else {
1889 xor_dest = sh->dev[pd_idx].page;
1890 for (i = disks; i--; ) {
1891 struct r5dev *dev = &sh->dev[i];
1892 if (i != pd_idx)
1893 xor_srcs[count++] = dev->page;
1894 }
1895 }
1896
91c00924
DW		 1897 /* 1/ if we prexor'd then the dest is reused as a source
1898 * 2/ if we did not prexor then we are redoing the parity
1899 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
1900 * for the synchronous xor case
1901 */
59fc630b 1902 last_stripe = !head_sh->batch_head ||
1903 list_first_entry(&sh->batch_list,
1904 struct stripe_head, batch_list) == head_sh;
1905 if (last_stripe) {
1906 flags = ASYNC_TX_ACK |
1907 (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
1908
1909 atomic_inc(&head_sh->count);
1910 init_async_submit(&submit, flags, tx, ops_complete_reconstruct, head_sh,
1911 to_addr_conv(sh, percpu, j));
1912 } else {
1913 flags = prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST;
1914 init_async_submit(&submit, flags, tx, NULL, NULL,
1915 to_addr_conv(sh, percpu, j));
1916 }
91c00924 1917
a08abd8c
DW		 1918 if (unlikely(count == 1))
1919 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
1920 else
1921 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
59fc630b 1922 if (!last_stripe) {
1923 j++;
1924 sh = list_first_entry(&sh->batch_list, struct stripe_head,
1925 batch_list);
1926 goto again;
1927 }
91c00924
DW		 1928}
1929
ac6b53b6
DW		 1930static void
1931ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu,
1932 struct dma_async_tx_descriptor *tx)
1933{
1934 struct async_submit_ctl submit;
59fc630b 1935 struct page **blocks;
1936 int count, i, j = 0;
1937 struct stripe_head *head_sh = sh;
1938 int last_stripe;
584acdd4
MS		 1939 int synflags;
1940 unsigned long txflags;
ac6b53b6
DW		 1941
1942 pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector);
1943
620125f2
SL		 1944 for (i = 0; i < sh->disks; i++) {
1945 if (sh->pd_idx == i || sh->qd_idx == i)
1946 continue;
1947 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1948 break;
1949 }
1950 if (i >= sh->disks) {
1951 atomic_inc(&sh->count);
620125f2
SL		 1952 set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
1953 set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
1954 ops_complete_reconstruct(sh);
1955 return;
1956 }
1957
59fc630b 1958again:
1959 blocks = to_addr_page(percpu, j);
584acdd4
MS		 1960
1961 if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
1962 synflags = SYNDROME_SRC_WRITTEN;
1963 txflags = ASYNC_TX_ACK | ASYNC_TX_PQ_XOR_DST;
1964 } else {
1965 synflags = SYNDROME_SRC_ALL;
1966 txflags = ASYNC_TX_ACK;
1967 }
1968
1969 count = set_syndrome_sources(blocks, sh, synflags);
59fc630b 1970 last_stripe = !head_sh->batch_head ||
1971 list_first_entry(&sh->batch_list,
1972 struct stripe_head, batch_list) == head_sh;
1973
1974 if (last_stripe) {
1975 atomic_inc(&head_sh->count);
584acdd4 1976 init_async_submit(&submit, txflags, tx, ops_complete_reconstruct,
59fc630b 1977 head_sh, to_addr_conv(sh, percpu, j));
1978 } else
1979 init_async_submit(&submit, 0, tx, NULL, NULL,
1980 to_addr_conv(sh, percpu, j));
48769695 1981 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
59fc630b 1982 if (!last_stripe) {
1983 j++;
1984 sh = list_first_entry(&sh->batch_list, struct stripe_head,
1985 batch_list);
1986 goto again;
1987 }
91c00924
DW		 1988}
1989
1990static void ops_complete_check(void *stripe_head_ref)
1991{
1992 struct stripe_head *sh = stripe_head_ref;
91c00924 1993
e46b272b 1994 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1995 (unsigned long long)sh->sector);
1996
ecc65c9b 1997 sh->check_state = check_state_check_result;
91c00924 1998 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 1999 raid5_release_stripe(sh);
91c00924
DW		 2000}
2001
ac6b53b6 2002static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 2003{
91c00924 2004 int disks = sh->disks;
ac6b53b6
DW		 2005 int pd_idx = sh->pd_idx;
2006 int qd_idx = sh->qd_idx;
2007 struct page *xor_dest;
46d5b785 2008 struct page **xor_srcs = to_addr_page(percpu, 0);
91c00924 2009 struct dma_async_tx_descriptor *tx;
a08abd8c 2010 struct async_submit_ctl submit;
ac6b53b6
DW		 2011 int count;
2012 int i;
91c00924 2013
e46b272b 2014 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 2015 (unsigned long long)sh->sector);
2016
59fc630b 2017 BUG_ON(sh->batch_head);
ac6b53b6
DW		 2018 count = 0;
2019 xor_dest = sh->dev[pd_idx].page;
2020 xor_srcs[count++] = xor_dest;
91c00924 2021 for (i = disks; i--; ) {
ac6b53b6
DW		 2022 if (i == pd_idx || i == qd_idx)
2023 continue;
2024 xor_srcs[count++] = sh->dev[i].page;
91c00924
DW		 2025 }
2026
d6f38f31 2027 init_async_submit(&submit, 0, NULL, NULL, NULL,
46d5b785 2028 to_addr_conv(sh, percpu, 0));
099f53cb 2029 tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
a08abd8c 2030 &sh->ops.zero_sum_result, &submit);
91c00924 2031
91c00924 2032 atomic_inc(&sh->count);
a08abd8c
DW		 2033 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL);
2034 tx = async_trigger_callback(&submit);
91c00924
DW		 2035}
2036
ac6b53b6
DW		 2037static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp)
2038{
46d5b785 2039 struct page **srcs = to_addr_page(percpu, 0);
ac6b53b6
DW		 2040 struct async_submit_ctl submit;
2041 int count;
2042
2043 pr_debug("%s: stripe %llu checkp: %d\n", __func__,
2044 (unsigned long long)sh->sector, checkp);
2045
59fc630b 2046 BUG_ON(sh->batch_head);
584acdd4 2047 count = set_syndrome_sources(srcs, sh, SYNDROME_SRC_ALL);
ac6b53b6
DW		 2048 if (!checkp)
2049 srcs[count] = NULL;
91c00924 2050
91c00924 2051 atomic_inc(&sh->count);
ac6b53b6 2052 init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check,
46d5b785 2053 sh, to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 2054 async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE,
2055 &sh->ops.zero_sum_result, percpu->spare_page, &submit);
91c00924
DW		 2056}
2057
51acbcec 2058static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
91c00924
DW		 2059{
2060 int overlap_clear = 0, i, disks = sh->disks;
2061 struct dma_async_tx_descriptor *tx = NULL;
d1688a6d 2062 struct r5conf *conf = sh->raid_conf;
ac6b53b6 2063 int level = conf->level;
d6f38f31
DW		 2064 struct raid5_percpu *percpu;
2065 unsigned long cpu;
91c00924 2066
d6f38f31
DW		 2067 cpu = get_cpu();
2068 percpu = per_cpu_ptr(conf->percpu, cpu);
83de75cc 2069 if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) {
91c00924
DW		 2070 ops_run_biofill(sh);
2071 overlap_clear++;
2072 }
2073
7b3a871e 2074 if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) {
ac6b53b6
DW		 2075 if (level < 6)
2076 tx = ops_run_compute5(sh, percpu);
2077 else {
2078 if (sh->ops.target2 < 0 || sh->ops.target < 0)
2079 tx = ops_run_compute6_1(sh, percpu);
2080 else
2081 tx = ops_run_compute6_2(sh, percpu);
2082 }
2083 /* terminate the chain if reconstruct is not set to be run */
2084 if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request))
7b3a871e
DW		 2085 async_tx_ack(tx);
2086 }
91c00924 2087
584acdd4
MS		 2088 if (test_bit(STRIPE_OP_PREXOR, &ops_request)) {
2089 if (level < 6)
2090 tx = ops_run_prexor5(sh, percpu, tx);
2091 else
2092 tx = ops_run_prexor6(sh, percpu, tx);
2093 }
91c00924 2094
ae1713e2
AP		 2095 if (test_bit(STRIPE_OP_PARTIAL_PARITY, &ops_request))
2096 tx = ops_run_partial_parity(sh, percpu, tx);
2097
600aa109 2098 if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) {
d8ee0728 2099 tx = ops_run_biodrain(sh, tx);
91c00924
DW		 2100 overlap_clear++;
2101 }
2102
ac6b53b6
DW		 2103 if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) {
2104 if (level < 6)
2105 ops_run_reconstruct5(sh, percpu, tx);
2106 else
2107 ops_run_reconstruct6(sh, percpu, tx);
2108 }
91c00924 2109
ac6b53b6
DW		 2110 if (test_bit(STRIPE_OP_CHECK, &ops_request)) {
2111 if (sh->check_state == check_state_run)
2112 ops_run_check_p(sh, percpu);
2113 else if (sh->check_state == check_state_run_q)
2114 ops_run_check_pq(sh, percpu, 0);
2115 else if (sh->check_state == check_state_run_pq)
2116 ops_run_check_pq(sh, percpu, 1);
2117 else
2118 BUG();
2119 }
91c00924 2120
59fc630b 2121 if (overlap_clear && !sh->batch_head)
91c00924
DW		 2122 for (i = disks; i--; ) {
2123 struct r5dev *dev = &sh->dev[i];
2124 if (test_and_clear_bit(R5_Overlap, &dev->flags))
2125 wake_up(&sh->raid_conf->wait_for_overlap);
2126 }
d6f38f31 2127 put_cpu();
91c00924
DW		 2128}
2129
845b9e22
AP		 2130static void free_stripe(struct kmem_cache *sc, struct stripe_head *sh)
2131{
2132 if (sh->ppl_page)
2133 __free_page(sh->ppl_page);
2134 kmem_cache_free(sc, sh);
2135}
2136
5f9d1fde 2137static struct stripe_head *alloc_stripe(struct kmem_cache *sc, gfp_t gfp,
845b9e22 2138 int disks, struct r5conf *conf)
f18c1a35
N		 2139{
2140 struct stripe_head *sh;
5f9d1fde 2141 int i;
f18c1a35
N		 2142
2143 sh = kmem_cache_zalloc(sc, gfp);
2144 if (sh) {
2145 spin_lock_init(&sh->stripe_lock);
2146 spin_lock_init(&sh->batch_lock);
2147 INIT_LIST_HEAD(&sh->batch_list);
2148 INIT_LIST_HEAD(&sh->lru);
a39f7afd 2149 INIT_LIST_HEAD(&sh->r5c);
d7bd398e 2150 INIT_LIST_HEAD(&sh->log_list);
f18c1a35 2151 atomic_set(&sh->count, 1);
845b9e22 2152 sh->raid_conf = conf;
a39f7afd 2153 sh->log_start = MaxSector;
5f9d1fde
SL		 2154 for (i = 0; i < disks; i++) {
2155 struct r5dev *dev = &sh->dev[i];
2156
3a83f467
ML		 2157 bio_init(&dev->req, &dev->vec, 1);
2158 bio_init(&dev->rreq, &dev->rvec, 1);
5f9d1fde 2159 }
845b9e22
AP		 2160
2161 if (raid5_has_ppl(conf)) {
2162 sh->ppl_page = alloc_page(gfp);
2163 if (!sh->ppl_page) {
2164 free_stripe(sc, sh);
2165 sh = NULL;
2166 }
2167 }
f18c1a35
N		 2168 }
2169 return sh;
2170}
486f0644 2171static int grow_one_stripe(struct r5conf *conf, gfp_t gfp)
1da177e4
LT		 2172{
2173 struct stripe_head *sh;
f18c1a35 2174
845b9e22 2175 sh = alloc_stripe(conf->slab_cache, gfp, conf->pool_size, conf);
3f294f4f
N		 2176 if (!sh)
2177 return 0;
6ce32846 2178
a9683a79 2179 if (grow_buffers(sh, gfp)) {
e4e11e38 2180 shrink_buffers(sh);
845b9e22 2181 free_stripe(conf->slab_cache, sh);
3f294f4f
N		 2182 return 0;
2183 }
486f0644
N		 2184 sh->hash_lock_index =
2185 conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS;
3f294f4f 2186 /* we just created an active stripe so... */
3f294f4f 2187 atomic_inc(&conf->active_stripes);
59fc630b 2188
6d036f7d 2189 raid5_release_stripe(sh);
486f0644 2190 conf->max_nr_stripes++;
3f294f4f
N		 2191 return 1;
2192}
2193
d1688a6d 2194static int grow_stripes(struct r5conf *conf, int num)
3f294f4f 2195{
e18b890b 2196 struct kmem_cache *sc;
5e5e3e78 2197 int devs = max(conf->raid_disks, conf->previous_raid_disks);
1da177e4 2198
f4be6b43
N		 2199 if (conf->mddev->gendisk)
2200 sprintf(conf->cache_name[0],
2201 "raid%d-%s", conf->level, mdname(conf->mddev));
2202 else
2203 sprintf(conf->cache_name[0],
2204 "raid%d-%p", conf->level, conf->mddev);
2205 sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]);
2206
ad01c9e3
N		 2207 conf->active_name = 0;
2208 sc = kmem_cache_create(conf->cache_name[conf->active_name],
1da177e4 2209 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
20c2df83 2210 0, 0, NULL);
1da177e4
LT		 2211 if (!sc)
2212 return 1;
2213 conf->slab_cache = sc;
ad01c9e3 2214 conf->pool_size = devs;
486f0644
N		 2215 while (num--)
2216 if (!grow_one_stripe(conf, GFP_KERNEL))
1da177e4 2217 return 1;
486f0644 2218
1da177e4
LT		 2219 return 0;
2220}
29269553 2221
d6f38f31
DW		 2222/**
2223 * scribble_len - return the required size of the scribble region
2224 * @num - total number of disks in the array
2225 *
2226 * The size must be enough to contain:
2227 * 1/ a struct page pointer for each device in the array +2
2228 * 2/ room to convert each entry in (1) to its corresponding dma
2229 * (dma_map_page()) or page (page_address()) address.
2230 *
2231 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
2232 * calculate over all devices (not just the data blocks), using zeros in place
2233 * of the P and Q blocks.
2234 */
46d5b785 2235static struct flex_array *scribble_alloc(int num, int cnt, gfp_t flags)
d6f38f31 2236{
46d5b785 2237 struct flex_array *ret;
d6f38f31
DW		 2238 size_t len;
2239
2240 len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2);
46d5b785 2241 ret = flex_array_alloc(len, cnt, flags);
2242 if (!ret)
2243 return NULL;
2244 /* always prealloc all elements, so no locking is required */
2245 if (flex_array_prealloc(ret, 0, cnt, flags)) {
2246 flex_array_free(ret);
2247 return NULL;
2248 }
2249 return ret;
d6f38f31
DW		 2250}
2251
738a2738
N		 2252static int resize_chunks(struct r5conf *conf, int new_disks, int new_sectors)
2253{
2254 unsigned long cpu;
2255 int err = 0;
2256
27a353c0
SL		 2257 /*
2258 * Never shrink. And mddev_suspend() could deadlock if this is called
2259 * from raid5d. In that case, scribble_disks and scribble_sectors
2260 * should equal to new_disks and new_sectors
2261 */
2262 if (conf->scribble_disks >= new_disks &&
2263 conf->scribble_sectors >= new_sectors)
2264 return 0;
738a2738
N		 2265 mddev_suspend(conf->mddev);
2266 get_online_cpus();
2267 for_each_present_cpu(cpu) {
2268 struct raid5_percpu *percpu;
2269 struct flex_array *scribble;
2270
2271 percpu = per_cpu_ptr(conf->percpu, cpu);
2272 scribble = scribble_alloc(new_disks,
2273 new_sectors / STRIPE_SECTORS,
2274 GFP_NOIO);
2275
2276 if (scribble) {
2277 flex_array_free(percpu->scribble);
2278 percpu->scribble = scribble;
2279 } else {
2280 err = -ENOMEM;
2281 break;
2282 }
2283 }
2284 put_online_cpus();
2285 mddev_resume(conf->mddev);
27a353c0
SL		 2286 if (!err) {
2287 conf->scribble_disks = new_disks;
2288 conf->scribble_sectors = new_sectors;
2289 }
738a2738
N		 2290 return err;
2291}
2292
d1688a6d 2293static int resize_stripes(struct r5conf *conf, int newsize)
ad01c9e3
N		 2294{
2295 /* Make all the stripes able to hold 'newsize' devices.
2296 * New slots in each stripe get 'page' set to a new page.
2297 *
2298 * This happens in stages:
2299 * 1/ create a new kmem_cache and allocate the required number of
2300 * stripe_heads.
83f0d77a 2301 * 2/ gather all the old stripe_heads and transfer the pages across
ad01c9e3
N		 2302 * to the new stripe_heads. This will have the side effect of
2303 * freezing the array as once all stripe_heads have been collected,
2304 * no IO will be possible. Old stripe heads are freed once their
2305 * pages have been transferred over, and the old kmem_cache is
2306 * freed when all stripes are done.
2307 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
3560741e 2308 * we simple return a failure status - no need to clean anything up.
ad01c9e3
N		 2309 * 4/ allocate new pages for the new slots in the new stripe_heads.
2310 * If this fails, we don't bother trying the shrink the
2311 * stripe_heads down again, we just leave them as they are.
2312 * As each stripe_head is processed the new one is released into
2313 * active service.
2314 *
2315 * Once step2 is started, we cannot afford to wait for a write,
2316 * so we use GFP_NOIO allocations.
2317 */
2318 struct stripe_head *osh, *nsh;
2319 LIST_HEAD(newstripes);
2320 struct disk_info *ndisks;
2214c260 2321 int err = 0;
e18b890b 2322 struct kmem_cache *sc;
ad01c9e3 2323 int i;
566c09c5 2324 int hash, cnt;
ad01c9e3 2325
2214c260 2326 md_allow_write(conf->mddev);
2a2275d6 2327
ad01c9e3
N		 2328 /* Step 1 */
2329 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
2330 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
20c2df83 2331 0, 0, NULL);
ad01c9e3
N		 2332 if (!sc)
2333 return -ENOMEM;
2334
2d5b569b
N		 2335 /* Need to ensure auto-resizing doesn't interfere */
2336 mutex_lock(&conf->cache_size_mutex);
2337
ad01c9e3 2338 for (i = conf->max_nr_stripes; i; i--) {
845b9e22 2339 nsh = alloc_stripe(sc, GFP_KERNEL, newsize, conf);
ad01c9e3
N		 2340 if (!nsh)
2341 break;
2342
ad01c9e3
N		 2343 list_add(&nsh->lru, &newstripes);
2344 }
2345 if (i) {
2346 /* didn't get enough, give up */
2347 while (!list_empty(&newstripes)) {
2348 nsh = list_entry(newstripes.next, struct stripe_head, lru);
2349 list_del(&nsh->lru);
845b9e22 2350 free_stripe(sc, nsh);
ad01c9e3
N		 2351 }
2352 kmem_cache_destroy(sc);
2d5b569b 2353 mutex_unlock(&conf->cache_size_mutex);
ad01c9e3
N		 2354 return -ENOMEM;
2355 }
2356 /* Step 2 - Must use GFP_NOIO now.
2357 * OK, we have enough stripes, start collecting inactive
2358 * stripes and copying them over
2359 */
566c09c5
SL		 2360 hash = 0;
2361 cnt = 0;
ad01c9e3 2362 list_for_each_entry(nsh, &newstripes, lru) {
566c09c5 2363 lock_device_hash_lock(conf, hash);
6ab2a4b8 2364 wait_event_cmd(conf->wait_for_stripe,
566c09c5
SL		 2365 !list_empty(conf->inactive_list + hash),
2366 unlock_device_hash_lock(conf, hash),
2367 lock_device_hash_lock(conf, hash));
2368 osh = get_free_stripe(conf, hash);
2369 unlock_device_hash_lock(conf, hash);
f18c1a35 2370
d592a996 2371 for(i=0; i<conf->pool_size; i++) {
ad01c9e3 2372 nsh->dev[i].page = osh->dev[i].page;
d592a996
SL		 2373 nsh->dev[i].orig_page = osh->dev[i].page;
2374 }
566c09c5 2375 nsh->hash_lock_index = hash;
845b9e22 2376 free_stripe(conf->slab_cache, osh);
566c09c5
SL		 2377 cnt++;
2378 if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS +
2379 !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) {
2380 hash++;
2381 cnt = 0;
2382 }
ad01c9e3
N		 2383 }
2384 kmem_cache_destroy(conf->slab_cache);
2385
2386 /* Step 3.
2387 * At this point, we are holding all the stripes so the array
2388 * is completely stalled, so now is a good time to resize
d6f38f31 2389 * conf->disks and the scribble region
ad01c9e3
N		 2390 */
2391 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
2392 if (ndisks) {
d7bd398e 2393 for (i = 0; i < conf->pool_size; i++)
ad01c9e3 2394 ndisks[i] = conf->disks[i];
d7bd398e
SL		 2395
2396 for (i = conf->pool_size; i < newsize; i++) {
2397 ndisks[i].extra_page = alloc_page(GFP_NOIO);
2398 if (!ndisks[i].extra_page)
2399 err = -ENOMEM;
2400 }
2401
2402 if (err) {
2403 for (i = conf->pool_size; i < newsize; i++)
2404 if (ndisks[i].extra_page)
2405 put_page(ndisks[i].extra_page);
2406 kfree(ndisks);
2407 } else {
2408 kfree(conf->disks);
2409 conf->disks = ndisks;
2410 }
ad01c9e3
N		 2411 } else
2412 err = -ENOMEM;
2413
2d5b569b 2414 mutex_unlock(&conf->cache_size_mutex);
583da48e
DY		 2415
2416 conf->slab_cache = sc;
2417 conf->active_name = 1-conf->active_name;
2418
ad01c9e3
N		 2419 /* Step 4, return new stripes to service */
2420 while(!list_empty(&newstripes)) {
2421 nsh = list_entry(newstripes.next, struct stripe_head, lru);
2422 list_del_init(&nsh->lru);
d6f38f31 2423
ad01c9e3
N		 2424 for (i=conf->raid_disks; i < newsize; i++)
2425 if (nsh->dev[i].page == NULL) {
2426 struct page *p = alloc_page(GFP_NOIO);
2427 nsh->dev[i].page = p;
d592a996 2428 nsh->dev[i].orig_page = p;
ad01c9e3
N		 2429 if (!p)
2430 err = -ENOMEM;
2431 }
6d036f7d 2432 raid5_release_stripe(nsh);
ad01c9e3
N		 2433 }
2434 /* critical section pass, GFP_NOIO no longer needed */
2435
6e9eac2d
N		 2436 if (!err)
2437 conf->pool_size = newsize;
ad01c9e3
N		 2438 return err;
2439}
1da177e4 2440
486f0644 2441static int drop_one_stripe(struct r5conf *conf)
1da177e4
LT		 2442{
2443 struct stripe_head *sh;
49895bcc 2444 int hash = (conf->max_nr_stripes - 1) & STRIPE_HASH_LOCKS_MASK;
1da177e4 2445
566c09c5
SL		 2446 spin_lock_irq(conf->hash_locks + hash);
2447 sh = get_free_stripe(conf, hash);
2448 spin_unlock_irq(conf->hash_locks + hash);
3f294f4f
N		 2449 if (!sh)
2450 return 0;
78bafebd 2451 BUG_ON(atomic_read(&sh->count));
e4e11e38 2452 shrink_buffers(sh);
845b9e22 2453 free_stripe(conf->slab_cache, sh);
3f294f4f 2454 atomic_dec(&conf->active_stripes);
486f0644 2455 conf->max_nr_stripes--;
3f294f4f
N		 2456 return 1;
2457}
2458
d1688a6d 2459static void shrink_stripes(struct r5conf *conf)
3f294f4f 2460{
486f0644
N		 2461 while (conf->max_nr_stripes &&
2462 drop_one_stripe(conf))
2463 ;
3f294f4f 2464
644df1a8 2465 kmem_cache_destroy(conf->slab_cache);
1da177e4
LT		 2466 conf->slab_cache = NULL;
2467}
2468
4246a0b6 2469static void raid5_end_read_request(struct bio * bi)
1da177e4 2470{
99c0fb5f 2471 struct stripe_head *sh = bi->bi_private;
d1688a6d 2472 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 2473 int disks = sh->disks, i;
d6950432 2474 char b[BDEVNAME_SIZE];
dd054fce 2475 struct md_rdev *rdev = NULL;
05616be5 2476 sector_t s;
1da177e4
LT		 2477
2478 for (i=0 ; i<disks; i++)
2479 if (bi == &sh->dev[i].req)
2480 break;
2481
4246a0b6 2482 pr_debug("end_read_request %llu/%d, count: %d, error %d.\n",
45b4233c 2483 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
4e4cbee9 2484 bi->bi_status);
1da177e4 2485 if (i == disks) {
5f9d1fde 2486 bio_reset(bi);
1da177e4 2487 BUG();
6712ecf8 2488 return;
1da177e4 2489 }
14a75d3e 2490 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
dd054fce
N		 2491 /* If replacement finished while this request was outstanding,
2492 * 'replacement' might be NULL already.
2493 * In that case it moved down to 'rdev'.
2494 * rdev is not removed until all requests are finished.
2495 */
14a75d3e 2496 rdev = conf->disks[i].replacement;
dd054fce 2497 if (!rdev)
14a75d3e 2498 rdev = conf->disks[i].rdev;
1da177e4 2499
05616be5
N		 2500 if (use_new_offset(conf, sh))
2501 s = sh->sector + rdev->new_data_offset;
2502 else
2503 s = sh->sector + rdev->data_offset;
4e4cbee9 2504 if (!bi->bi_status) {
1da177e4 2505 set_bit(R5_UPTODATE, &sh->dev[i].flags);
4e5314b5 2506 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
14a75d3e
N		 2507 /* Note that this cannot happen on a
2508 * replacement device. We just fail those on
2509 * any error
2510 */
cc6167b4
N		 2511 pr_info_ratelimited(
2512 "md/raid:%s: read error corrected (%lu sectors at %llu on %s)\n",
8bda470e 2513 mdname(conf->mddev), STRIPE_SECTORS,
05616be5 2514 (unsigned long long)s,
8bda470e 2515 bdevname(rdev->bdev, b));
ddd5115f 2516 atomic_add(STRIPE_SECTORS, &rdev->corrected_errors);
4e5314b5
N		 2517 clear_bit(R5_ReadError, &sh->dev[i].flags);
2518 clear_bit(R5_ReWrite, &sh->dev[i].flags);
3f9e7c14 2519 } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2520 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2521
86aa1397
SL		 2522 if (test_bit(R5_InJournal, &sh->dev[i].flags))
2523 /*
2524 * end read for a page in journal, this
2525 * must be preparing for prexor in rmw
2526 */
2527 set_bit(R5_OrigPageUPTDODATE, &sh->dev[i].flags);
2528
14a75d3e
N		 2529 if (atomic_read(&rdev->read_errors))
2530 atomic_set(&rdev->read_errors, 0);
1da177e4 2531 } else {
14a75d3e 2532 const char *bdn = bdevname(rdev->bdev, b);
ba22dcbf 2533 int retry = 0;
2e8ac303 2534 int set_bad = 0;
d6950432 2535
1da177e4 2536 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
d6950432 2537 atomic_inc(&rdev->read_errors);
14a75d3e 2538 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
cc6167b4
N		 2539 pr_warn_ratelimited(
2540 "md/raid:%s: read error on replacement device (sector %llu on %s).\n",
14a75d3e 2541 mdname(conf->mddev),
05616be5 2542 (unsigned long long)s,
14a75d3e 2543 bdn);
2e8ac303 2544 else if (conf->mddev->degraded >= conf->max_degraded) {
2545 set_bad = 1;
cc6167b4
N		 2546 pr_warn_ratelimited(
2547 "md/raid:%s: read error not correctable (sector %llu on %s).\n",
8bda470e 2548 mdname(conf->mddev),
05616be5 2549 (unsigned long long)s,
8bda470e 2550 bdn);
2e8ac303 2551 } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) {
4e5314b5 2552 /* Oh, no!!! */
2e8ac303 2553 set_bad = 1;
cc6167b4
N		 2554 pr_warn_ratelimited(
2555 "md/raid:%s: read error NOT corrected!! (sector %llu on %s).\n",
8bda470e 2556 mdname(conf->mddev),
05616be5 2557 (unsigned long long)s,
8bda470e 2558 bdn);
2e8ac303 2559 } else if (atomic_read(&rdev->read_errors)
ba22dcbf 2560 > conf->max_nr_stripes)
cc6167b4 2561 pr_warn("md/raid:%s: Too many read errors, failing device %s.\n",
d6950432 2562 mdname(conf->mddev), bdn);
ba22dcbf
N		 2563 else
2564 retry = 1;
edfa1f65
BY		 2565 if (set_bad && test_bit(In_sync, &rdev->flags)
2566 && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2567 retry = 1;
ba22dcbf 2568 if (retry)
3f9e7c14 2569 if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) {
2570 set_bit(R5_ReadError, &sh->dev[i].flags);
2571 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2572 } else
2573 set_bit(R5_ReadNoMerge, &sh->dev[i].flags);
ba22dcbf 2574 else {
4e5314b5
N		 2575 clear_bit(R5_ReadError, &sh->dev[i].flags);
2576 clear_bit(R5_ReWrite, &sh->dev[i].flags);
2e8ac303 2577 if (!(set_bad
2578 && test_bit(In_sync, &rdev->flags)
2579 && rdev_set_badblocks(
2580 rdev, sh->sector, STRIPE_SECTORS, 0)))
2581 md_error(conf->mddev, rdev);
ba22dcbf 2582 }
1da177e4 2583 }
14a75d3e 2584 rdev_dec_pending(rdev, conf->mddev);
c9445555 2585 bio_reset(bi);
1da177e4
LT		 2586 clear_bit(R5_LOCKED, &sh->dev[i].flags);
2587 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 2588 raid5_release_stripe(sh);
1da177e4
LT		 2589}
2590
4246a0b6 2591static void raid5_end_write_request(struct bio *bi)
1da177e4 2592{
99c0fb5f 2593 struct stripe_head *sh = bi->bi_private;
d1688a6d 2594 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 2595 int disks = sh->disks, i;
977df362 2596 struct md_rdev *uninitialized_var(rdev);
b84db560
N		 2597 sector_t first_bad;
2598 int bad_sectors;
977df362 2599 int replacement = 0;
1da177e4 2600
977df362
N		 2601 for (i = 0 ; i < disks; i++) {
2602 if (bi == &sh->dev[i].req) {
2603 rdev = conf->disks[i].rdev;
1da177e4 2604 break;
977df362
N		 2605 }
2606 if (bi == &sh->dev[i].rreq) {
2607 rdev = conf->disks[i].replacement;
dd054fce
N		 2608 if (rdev)
2609 replacement = 1;
2610 else
2611 /* rdev was removed and 'replacement'
2612 * replaced it. rdev is not removed
2613 * until all requests are finished.
2614 */
2615 rdev = conf->disks[i].rdev;
977df362
N		 2616 break;
2617 }
2618 }
4246a0b6 2619 pr_debug("end_write_request %llu/%d, count %d, error: %d.\n",
1da177e4 2620 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
4e4cbee9 2621 bi->bi_status);
1da177e4 2622 if (i == disks) {
5f9d1fde 2623 bio_reset(bi);
1da177e4 2624 BUG();
6712ecf8 2625 return;
1da177e4
LT		 2626 }
2627
977df362 2628 if (replacement) {
4e4cbee9 2629 if (bi->bi_status)
977df362
N		 2630 md_error(conf->mddev, rdev);
2631 else if (is_badblock(rdev, sh->sector,
2632 STRIPE_SECTORS,
2633 &first_bad, &bad_sectors))
2634 set_bit(R5_MadeGoodRepl, &sh->dev[i].flags);
2635 } else {
4e4cbee9 2636 if (bi->bi_status) {
9f97e4b1 2637 set_bit(STRIPE_DEGRADED, &sh->state);
977df362
N		 2638 set_bit(WriteErrorSeen, &rdev->flags);
2639 set_bit(R5_WriteError, &sh->dev[i].flags);
3a6de292
N		 2640 if (!test_and_set_bit(WantReplacement, &rdev->flags))
2641 set_bit(MD_RECOVERY_NEEDED,
2642 &rdev->mddev->recovery);
977df362
N		 2643 } else if (is_badblock(rdev, sh->sector,
2644 STRIPE_SECTORS,
c0b32972 2645 &first_bad, &bad_sectors)) {
977df362 2646 set_bit(R5_MadeGood, &sh->dev[i].flags);
c0b32972
N		 2647 if (test_bit(R5_ReadError, &sh->dev[i].flags))
2648 /* That was a successful write so make
2649 * sure it looks like we already did
2650 * a re-write.
2651 */
2652 set_bit(R5_ReWrite, &sh->dev[i].flags);
2653 }
977df362
N		 2654 }
2655 rdev_dec_pending(rdev, conf->mddev);
1da177e4 2656
4e4cbee9 2657 if (sh->batch_head && bi->bi_status && !replacement)
72ac7330 2658 set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state);
2659
c9445555 2660 bio_reset(bi);
977df362
N		 2661 if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags))
2662 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1da177e4 2663 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 2664 raid5_release_stripe(sh);
59fc630b 2665
2666 if (sh->batch_head && sh != sh->batch_head)
6d036f7d 2667 raid5_release_stripe(sh->batch_head);
1da177e4
LT		 2668}
2669
849674e4 2670static void raid5_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 2671{
2672 char b[BDEVNAME_SIZE];
d1688a6d 2673 struct r5conf *conf = mddev->private;
908f4fbd 2674 unsigned long flags;
0c55e022 2675 pr_debug("raid456: error called\n");
1da177e4 2676
908f4fbd
N		 2677 spin_lock_irqsave(&conf->device_lock, flags);
2678 clear_bit(In_sync, &rdev->flags);
2e38a37f 2679 mddev->degraded = raid5_calc_degraded(conf);
908f4fbd
N		 2680 spin_unlock_irqrestore(&conf->device_lock, flags);
2681 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2682
de393cde 2683 set_bit(Blocked, &rdev->flags);
6f8d0c77 2684 set_bit(Faulty, &rdev->flags);
2953079c
SL		 2685 set_mask_bits(&mddev->sb_flags, 0,
2686 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING));
cc6167b4
N		 2687 pr_crit("md/raid:%s: Disk failure on %s, disabling device.\n"
2688 "md/raid:%s: Operation continuing on %d devices.\n",
2689 mdname(mddev),
2690 bdevname(rdev->bdev, b),
2691 mdname(mddev),
2692 conf->raid_disks - mddev->degraded);
70d466f7 2693 r5c_update_on_rdev_error(mddev, rdev);
16a53ecc 2694}
1da177e4
LT		 2695
2696/*
2697 * Input: a 'big' sector number,
2698 * Output: index of the data and parity disk, and the sector # in them.
2699 */
6d036f7d
SL		 2700sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector,
2701 int previous, int *dd_idx,
2702 struct stripe_head *sh)
1da177e4 2703{
6e3b96ed 2704 sector_t stripe, stripe2;
35f2a591 2705 sector_t chunk_number;
1da177e4 2706 unsigned int chunk_offset;
911d4ee8 2707 int pd_idx, qd_idx;
67cc2b81 2708 int ddf_layout = 0;
1da177e4 2709 sector_t new_sector;
e183eaed
N		 2710 int algorithm = previous ? conf->prev_algo
2711 : conf->algorithm;
09c9e5fa
AN		 2712 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2713 : conf->chunk_sectors;
112bf897
N		 2714 int raid_disks = previous ? conf->previous_raid_disks
2715 : conf->raid_disks;
2716 int data_disks = raid_disks - conf->max_degraded;
1da177e4
LT		 2717
2718 /* First compute the information on this sector */
2719
2720 /*
2721 * Compute the chunk number and the sector offset inside the chunk
2722 */
2723 chunk_offset = sector_div(r_sector, sectors_per_chunk);
2724 chunk_number = r_sector;
1da177e4
LT		 2725
2726 /*
2727 * Compute the stripe number
2728 */
35f2a591
N		 2729 stripe = chunk_number;
2730 *dd_idx = sector_div(stripe, data_disks);
6e3b96ed 2731 stripe2 = stripe;
1da177e4
LT		 2732 /*
2733 * Select the parity disk based on the user selected algorithm.
2734 */
84789554 2735 pd_idx = qd_idx = -1;
16a53ecc
N		 2736 switch(conf->level) {
2737 case 4:
911d4ee8 2738 pd_idx = data_disks;
16a53ecc
N		 2739 break;
2740 case 5:
e183eaed 2741 switch (algorithm) {
1da177e4 2742 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2743 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2744 if (*dd_idx >= pd_idx)
1da177e4
LT		 2745 (*dd_idx)++;
2746 break;
2747 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2748 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2749 if (*dd_idx >= pd_idx)
1da177e4
LT		 2750 (*dd_idx)++;
2751 break;
2752 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2753 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2754 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4
LT		 2755 break;
2756 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2757 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2758 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4 2759 break;
99c0fb5f
N		 2760 case ALGORITHM_PARITY_0:
2761 pd_idx = 0;
2762 (*dd_idx)++;
2763 break;
2764 case ALGORITHM_PARITY_N:
2765 pd_idx = data_disks;
2766 break;
1da177e4 2767 default:
99c0fb5f 2768 BUG();
16a53ecc
N		 2769 }
2770 break;
2771 case 6:
2772
e183eaed 2773 switch (algorithm) {
16a53ecc 2774 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2775 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2776 qd_idx = pd_idx + 1;
2777 if (pd_idx == raid_disks-1) {
99c0fb5f 2778 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2779 qd_idx = 0;
2780 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2781 (*dd_idx) += 2; /* D D P Q D */
2782 break;
2783 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2784 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2785 qd_idx = pd_idx + 1;
2786 if (pd_idx == raid_disks-1) {
99c0fb5f 2787 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2788 qd_idx = 0;
2789 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2790 (*dd_idx) += 2; /* D D P Q D */
2791 break;
2792 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2793 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2794 qd_idx = (pd_idx + 1) % raid_disks;
2795 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc
N		 2796 break;
2797 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2798 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2799 qd_idx = (pd_idx + 1) % raid_disks;
2800 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc 2801 break;
99c0fb5f
N		 2802
2803 case ALGORITHM_PARITY_0:
2804 pd_idx = 0;
2805 qd_idx = 1;
2806 (*dd_idx) += 2;
2807 break;
2808 case ALGORITHM_PARITY_N:
2809 pd_idx = data_disks;
2810 qd_idx = data_disks + 1;
2811 break;
2812
2813 case ALGORITHM_ROTATING_ZERO_RESTART:
2814 /* Exactly the same as RIGHT_ASYMMETRIC, but or
2815 * of blocks for computing Q is different.
2816 */
6e3b96ed 2817 pd_idx = sector_div(stripe2, raid_disks);
99c0fb5f
N		 2818 qd_idx = pd_idx + 1;
2819 if (pd_idx == raid_disks-1) {
2820 (*dd_idx)++; /* Q D D D P */
2821 qd_idx = 0;
2822 } else if (*dd_idx >= pd_idx)
2823 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2824 ddf_layout = 1;
99c0fb5f
N		 2825 break;
2826
2827 case ALGORITHM_ROTATING_N_RESTART:
2828 /* Same a left_asymmetric, by first stripe is
2829 * D D D P Q rather than
2830 * Q D D D P
2831 */
6e3b96ed
N		 2832 stripe2 += 1;
2833 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2834 qd_idx = pd_idx + 1;
2835 if (pd_idx == raid_disks-1) {
2836 (*dd_idx)++; /* Q D D D P */
2837 qd_idx = 0;
2838 } else if (*dd_idx >= pd_idx)
2839 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2840 ddf_layout = 1;
99c0fb5f
N		 2841 break;
2842
2843 case ALGORITHM_ROTATING_N_CONTINUE:
2844 /* Same as left_symmetric but Q is before P */
6e3b96ed 2845 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2846 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
2847 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
67cc2b81 2848 ddf_layout = 1;
99c0fb5f
N		 2849 break;
2850
2851 case ALGORITHM_LEFT_ASYMMETRIC_6:
2852 /* RAID5 left_asymmetric, with Q on last device */
6e3b96ed 2853 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2854 if (*dd_idx >= pd_idx)
2855 (*dd_idx)++;
2856 qd_idx = raid_disks - 1;
2857 break;
2858
2859 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6e3b96ed 2860 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2861 if (*dd_idx >= pd_idx)
2862 (*dd_idx)++;
2863 qd_idx = raid_disks - 1;
2864 break;
2865
2866 case ALGORITHM_LEFT_SYMMETRIC_6:
6e3b96ed 2867 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2868 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2869 qd_idx = raid_disks - 1;
2870 break;
2871
2872 case ALGORITHM_RIGHT_SYMMETRIC_6:
6e3b96ed 2873 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2874 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2875 qd_idx = raid_disks - 1;
2876 break;
2877
2878 case ALGORITHM_PARITY_0_6:
2879 pd_idx = 0;
2880 (*dd_idx)++;
2881 qd_idx = raid_disks - 1;
2882 break;
2883
16a53ecc 2884 default:
99c0fb5f 2885 BUG();
16a53ecc
N		 2886 }
2887 break;
1da177e4
LT		 2888 }
2889
911d4ee8
N		 2890 if (sh) {
2891 sh->pd_idx = pd_idx;
2892 sh->qd_idx = qd_idx;
67cc2b81 2893 sh->ddf_layout = ddf_layout;
911d4ee8 2894 }
1da177e4
LT		 2895 /*
2896 * Finally, compute the new sector number
2897 */
2898 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
2899 return new_sector;
2900}
2901
6d036f7d 2902sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous)
1da177e4 2903{
d1688a6d 2904 struct r5conf *conf = sh->raid_conf;
b875e531
N		 2905 int raid_disks = sh->disks;
2906 int data_disks = raid_disks - conf->max_degraded;
1da177e4 2907 sector_t new_sector = sh->sector, check;
09c9e5fa
AN		 2908 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2909 : conf->chunk_sectors;
e183eaed
N		 2910 int algorithm = previous ? conf->prev_algo
2911 : conf->algorithm;
1da177e4
LT		 2912 sector_t stripe;
2913 int chunk_offset;
35f2a591
N		 2914 sector_t chunk_number;
2915 int dummy1, dd_idx = i;
1da177e4 2916 sector_t r_sector;
911d4ee8 2917 struct stripe_head sh2;
1da177e4
LT		 2918
2919 chunk_offset = sector_div(new_sector, sectors_per_chunk);
2920 stripe = new_sector;
1da177e4 2921
16a53ecc
N		 2922 if (i == sh->pd_idx)
2923 return 0;
2924 switch(conf->level) {
2925 case 4: break;
2926 case 5:
e183eaed 2927 switch (algorithm) {
1da177e4
LT		 2928 case ALGORITHM_LEFT_ASYMMETRIC:
2929 case ALGORITHM_RIGHT_ASYMMETRIC:
2930 if (i > sh->pd_idx)
2931 i--;
2932 break;
2933 case ALGORITHM_LEFT_SYMMETRIC:
2934 case ALGORITHM_RIGHT_SYMMETRIC:
2935 if (i < sh->pd_idx)
2936 i += raid_disks;
2937 i -= (sh->pd_idx + 1);
2938 break;
99c0fb5f
N		 2939 case ALGORITHM_PARITY_0:
2940 i -= 1;
2941 break;
2942 case ALGORITHM_PARITY_N:
2943 break;
1da177e4 2944 default:
99c0fb5f 2945 BUG();
16a53ecc
N		 2946 }
2947 break;
2948 case 6:
d0dabf7e 2949 if (i == sh->qd_idx)
16a53ecc 2950 return 0; /* It is the Q disk */
e183eaed 2951 switch (algorithm) {
16a53ecc
N		 2952 case ALGORITHM_LEFT_ASYMMETRIC:
2953 case ALGORITHM_RIGHT_ASYMMETRIC:
99c0fb5f
N		 2954 case ALGORITHM_ROTATING_ZERO_RESTART:
2955 case ALGORITHM_ROTATING_N_RESTART:
2956 if (sh->pd_idx == raid_disks-1)
2957 i--; /* Q D D D P */
16a53ecc
N		 2958 else if (i > sh->pd_idx)
2959 i -= 2; /* D D P Q D */
2960 break;
2961 case ALGORITHM_LEFT_SYMMETRIC:
2962 case ALGORITHM_RIGHT_SYMMETRIC:
2963 if (sh->pd_idx == raid_disks-1)
2964 i--; /* Q D D D P */
2965 else {
2966 /* D D P Q D */
2967 if (i < sh->pd_idx)
2968 i += raid_disks;
2969 i -= (sh->pd_idx + 2);
2970 }
2971 break;
99c0fb5f
N		 2972 case ALGORITHM_PARITY_0:
2973 i -= 2;
2974 break;
2975 case ALGORITHM_PARITY_N:
2976 break;
2977 case ALGORITHM_ROTATING_N_CONTINUE:
e4424fee 2978 /* Like left_symmetric, but P is before Q */
99c0fb5f
N		 2979 if (sh->pd_idx == 0)
2980 i--; /* P D D D Q */
e4424fee
N		 2981 else {
2982 /* D D Q P D */
2983 if (i < sh->pd_idx)
2984 i += raid_disks;
2985 i -= (sh->pd_idx + 1);
2986 }
99c0fb5f
N		 2987 break;
2988 case ALGORITHM_LEFT_ASYMMETRIC_6:
2989 case ALGORITHM_RIGHT_ASYMMETRIC_6:
2990 if (i > sh->pd_idx)
2991 i--;
2992 break;
2993 case ALGORITHM_LEFT_SYMMETRIC_6:
2994 case ALGORITHM_RIGHT_SYMMETRIC_6:
2995 if (i < sh->pd_idx)
2996 i += data_disks + 1;
2997 i -= (sh->pd_idx + 1);
2998 break;
2999 case ALGORITHM_PARITY_0_6:
3000 i -= 1;
3001 break;
16a53ecc 3002 default:
99c0fb5f 3003 BUG();
16a53ecc
N		 3004 }
3005 break;
1da177e4
LT		 3006 }
3007
3008 chunk_number = stripe * data_disks + i;
35f2a591 3009 r_sector = chunk_number * sectors_per_chunk + chunk_offset;
1da177e4 3010
112bf897 3011 check = raid5_compute_sector(conf, r_sector,
784052ec 3012 previous, &dummy1, &sh2);
911d4ee8
N		 3013 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
3014 || sh2.qd_idx != sh->qd_idx) {
cc6167b4
N		 3015 pr_warn("md/raid:%s: compute_blocknr: map not correct\n",
3016 mdname(conf->mddev));
1da177e4
LT		 3017 return 0;
3018 }
3019 return r_sector;
3020}
3021
07e83364
SL		 3022/*
3023 * There are cases where we want handle_stripe_dirtying() and
3024 * schedule_reconstruction() to delay towrite to some dev of a stripe.
3025 *
3026 * This function checks whether we want to delay the towrite. Specifically,
3027 * we delay the towrite when:
3028 *
3029 * 1. degraded stripe has a non-overwrite to the missing dev, AND this
3030 * stripe has data in journal (for other devices).
3031 *
3032 * In this case, when reading data for the non-overwrite dev, it is
3033 * necessary to handle complex rmw of write back cache (prexor with
3034 * orig_page, and xor with page). To keep read path simple, we would
3035 * like to flush data in journal to RAID disks first, so complex rmw
3036 * is handled in the write patch (handle_stripe_dirtying).
3037 *
39b99586
SL		 3038 * 2. when journal space is critical (R5C_LOG_CRITICAL=1)
3039 *
3040 * It is important to be able to flush all stripes in raid5-cache.
3041 * Therefore, we need reserve some space on the journal device for
3042 * these flushes. If flush operation includes pending writes to the
3043 * stripe, we need to reserve (conf->raid_disk + 1) pages per stripe
3044 * for the flush out. If we exclude these pending writes from flush
3045 * operation, we only need (conf->max_degraded + 1) pages per stripe.
3046 * Therefore, excluding pending writes in these cases enables more
3047 * efficient use of the journal device.
3048 *
3049 * Note: To make sure the stripe makes progress, we only delay
3050 * towrite for stripes with data already in journal (injournal > 0).
3051 * When LOG_CRITICAL, stripes with injournal == 0 will be sent to
3052 * no_space_stripes list.
3053 *
70d466f7
SL		 3054 * 3. during journal failure
3055 * In journal failure, we try to flush all cached data to raid disks
3056 * based on data in stripe cache. The array is read-only to upper
3057 * layers, so we would skip all pending writes.
3058 *
07e83364 3059 */
39b99586
SL		 3060static inline bool delay_towrite(struct r5conf *conf,
3061 struct r5dev *dev,
3062 struct stripe_head_state *s)
07e83364 3063{
39b99586
SL		 3064 /* case 1 above */
3065 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
3066 !test_bit(R5_Insync, &dev->flags) && s->injournal)
3067 return true;
3068 /* case 2 above */
3069 if (test_bit(R5C_LOG_CRITICAL, &conf->cache_state) &&
3070 s->injournal > 0)
3071 return true;
70d466f7
SL		 3072 /* case 3 above */
3073 if (s->log_failed && s->injournal)
3074 return true;
39b99586 3075 return false;
07e83364
SL		 3076}
3077
600aa109 3078static void
c0f7bddb 3079schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
600aa109 3080 int rcw, int expand)
e33129d8 3081{
584acdd4 3082 int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx, disks = sh->disks;
d1688a6d 3083 struct r5conf *conf = sh->raid_conf;
c0f7bddb 3084 int level = conf->level;
e33129d8
DW		 3085
3086 if (rcw) {
1e6d690b
SL		 3087 /*
3088 * In some cases, handle_stripe_dirtying initially decided to
3089 * run rmw and allocates extra page for prexor. However, rcw is
3090 * cheaper later on. We need to free the extra page now,
3091 * because we won't be able to do that in ops_complete_prexor().
3092 */
3093 r5c_release_extra_page(sh);
e33129d8
DW		 3094
3095 for (i = disks; i--; ) {
3096 struct r5dev *dev = &sh->dev[i];
3097
39b99586 3098 if (dev->towrite && !delay_towrite(conf, dev, s)) {
e33129d8 3099 set_bit(R5_LOCKED, &dev->flags);
d8ee0728 3100 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 3101 if (!expand)
3102 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 3103 s->locked++;
1e6d690b
SL		 3104 } else if (test_bit(R5_InJournal, &dev->flags)) {
3105 set_bit(R5_LOCKED, &dev->flags);
3106 s->locked++;
e33129d8
DW		 3107 }
3108 }
ce7d363a
N		 3109 /* if we are not expanding this is a proper write request, and
3110 * there will be bios with new data to be drained into the
3111 * stripe cache
3112 */
3113 if (!expand) {
3114 if (!s->locked)
3115 /* False alarm, nothing to do */
3116 return;
3117 sh->reconstruct_state = reconstruct_state_drain_run;
3118 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
3119 } else
3120 sh->reconstruct_state = reconstruct_state_run;
3121
3122 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
3123
c0f7bddb 3124 if (s->locked + conf->max_degraded == disks)
8b3e6cdc 3125 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
c0f7bddb 3126 atomic_inc(&conf->pending_full_writes);
e33129d8
DW		 3127 } else {
3128 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
3129 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
584acdd4
MS		 3130 BUG_ON(level == 6 &&
3131 (!(test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags) ||
3132 test_bit(R5_Wantcompute, &sh->dev[qd_idx].flags))));
e33129d8 3133
e33129d8
DW		 3134 for (i = disks; i--; ) {
3135 struct r5dev *dev = &sh->dev[i];
584acdd4 3136 if (i == pd_idx || i == qd_idx)
e33129d8
DW		 3137 continue;
3138
e33129d8
DW		 3139 if (dev->towrite &&
3140 (test_bit(R5_UPTODATE, &dev->flags) ||
d8ee0728
DW		 3141 test_bit(R5_Wantcompute, &dev->flags))) {
3142 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 3143 set_bit(R5_LOCKED, &dev->flags);
3144 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 3145 s->locked++;
1e6d690b
SL		 3146 } else if (test_bit(R5_InJournal, &dev->flags)) {
3147 set_bit(R5_LOCKED, &dev->flags);
3148 s->locked++;
e33129d8
DW		 3149 }
3150 }
ce7d363a
N		 3151 if (!s->locked)
3152 /* False alarm - nothing to do */
3153 return;
3154 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
3155 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
3156 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
3157 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
e33129d8
DW		 3158 }
3159
c0f7bddb 3160 /* keep the parity disk(s) locked while asynchronous operations
e33129d8
DW		 3161 * are in flight
3162 */
3163 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
3164 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
600aa109 3165 s->locked++;
e33129d8 3166
c0f7bddb
YT		 3167 if (level == 6) {
3168 int qd_idx = sh->qd_idx;
3169 struct r5dev *dev = &sh->dev[qd_idx];
3170
3171 set_bit(R5_LOCKED, &dev->flags);
3172 clear_bit(R5_UPTODATE, &dev->flags);
3173 s->locked++;
3174 }
3175
845b9e22 3176 if (raid5_has_ppl(sh->raid_conf) && sh->ppl_page &&
3418d036
AP		 3177 test_bit(STRIPE_OP_BIODRAIN, &s->ops_request) &&
3178 !test_bit(STRIPE_FULL_WRITE, &sh->state) &&
3179 test_bit(R5_Insync, &sh->dev[pd_idx].flags))
3180 set_bit(STRIPE_OP_PARTIAL_PARITY, &s->ops_request);
3181
600aa109 3182 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
e46b272b 3183 __func__, (unsigned long long)sh->sector,
600aa109 3184 s->locked, s->ops_request);
e33129d8 3185}
16a53ecc 3186
1da177e4
LT		 3187/*
3188 * Each stripe/dev can have one or more bion attached.
16a53ecc 3189 * toread/towrite point to the first in a chain.
1da177e4
LT		 3190 * The bi_next chain must be in order.
3191 */
da41ba65 3192static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx,
3193 int forwrite, int previous)
1da177e4
LT		 3194{
3195 struct bio **bip;
d1688a6d 3196 struct r5conf *conf = sh->raid_conf;
72626685 3197 int firstwrite=0;
1da177e4 3198
cbe47ec5 3199 pr_debug("adding bi b#%llu to stripe s#%llu\n",
4f024f37 3200 (unsigned long long)bi->bi_iter.bi_sector,
1da177e4
LT		 3201 (unsigned long long)sh->sector);
3202
b17459c0 3203 spin_lock_irq(&sh->stripe_lock);
59fc630b 3204 /* Don't allow new IO added to stripes in batch list */
3205 if (sh->batch_head)
3206 goto overlap;
72626685 3207 if (forwrite) {
1da177e4 3208 bip = &sh->dev[dd_idx].towrite;
7eaf7e8e 3209 if (*bip == NULL)
72626685
N		 3210 firstwrite = 1;
3211 } else
1da177e4 3212 bip = &sh->dev[dd_idx].toread;
4f024f37
KO		 3213 while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) {
3214 if (bio_end_sector(*bip) > bi->bi_iter.bi_sector)
1da177e4
LT		 3215 goto overlap;
3216 bip = & (*bip)->bi_next;
3217 }
4f024f37 3218 if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi))
1da177e4
LT		 3219 goto overlap;
3220
3418d036
AP		 3221 if (forwrite && raid5_has_ppl(conf)) {
3222 /*
3223 * With PPL only writes to consecutive data chunks within a
3224 * stripe are allowed because for a single stripe_head we can
3225 * only have one PPL entry at a time, which describes one data
3226 * range. Not really an overlap, but wait_for_overlap can be
3227 * used to handle this.
3228 */
3229 sector_t sector;
3230 sector_t first = 0;
3231 sector_t last = 0;
3232 int count = 0;
3233 int i;
3234
3235 for (i = 0; i < sh->disks; i++) {
3236 if (i != sh->pd_idx &&
3237 (i == dd_idx || sh->dev[i].towrite)) {
3238 sector = sh->dev[i].sector;
3239 if (count == 0 || sector < first)
3240 first = sector;
3241 if (sector > last)
3242 last = sector;
3243 count++;
3244 }
3245 }
3246
3247 if (first + conf->chunk_sectors * (count - 1) != last)
3248 goto overlap;
3249 }
3250
da41ba65 3251 if (!forwrite || previous)
3252 clear_bit(STRIPE_BATCH_READY, &sh->state);
3253
78bafebd 3254 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
1da177e4
LT		 3255 if (*bip)
3256 bi->bi_next = *bip;
3257 *bip = bi;
016c76ac 3258 bio_inc_remaining(bi);
49728050 3259 md_write_inc(conf->mddev, bi);
72626685 3260
1da177e4
LT		 3261 if (forwrite) {
3262 /* check if page is covered */
3263 sector_t sector = sh->dev[dd_idx].sector;
3264 for (bi=sh->dev[dd_idx].towrite;
3265 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
4f024f37 3266 bi && bi->bi_iter.bi_sector <= sector;
1da177e4 3267 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
f73a1c7d
KO		 3268 if (bio_end_sector(bi) >= sector)
3269 sector = bio_end_sector(bi);
1da177e4
LT		 3270 }
3271 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
7a87f434 3272 if (!test_and_set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags))
3273 sh->overwrite_disks++;
1da177e4 3274 }
cbe47ec5
N		 3275
3276 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
4f024f37 3277 (unsigned long long)(*bip)->bi_iter.bi_sector,
cbe47ec5
N		 3278 (unsigned long long)sh->sector, dd_idx);
3279
3280 if (conf->mddev->bitmap && firstwrite) {
d0852df5
N		 3281 /* Cannot hold spinlock over bitmap_startwrite,
3282 * but must ensure this isn't added to a batch until
3283 * we have added to the bitmap and set bm_seq.
3284 * So set STRIPE_BITMAP_PENDING to prevent
3285 * batching.
3286 * If multiple add_stripe_bio() calls race here they
3287 * much all set STRIPE_BITMAP_PENDING. So only the first one
3288 * to complete "bitmap_startwrite" gets to set
3289 * STRIPE_BIT_DELAY. This is important as once a stripe
3290 * is added to a batch, STRIPE_BIT_DELAY cannot be changed
3291 * any more.
3292 */
3293 set_bit(STRIPE_BITMAP_PENDING, &sh->state);
3294 spin_unlock_irq(&sh->stripe_lock);
cbe47ec5
N		 3295 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
3296 STRIPE_SECTORS, 0);
d0852df5
N		 3297 spin_lock_irq(&sh->stripe_lock);
3298 clear_bit(STRIPE_BITMAP_PENDING, &sh->state);
3299 if (!sh->batch_head) {
3300 sh->bm_seq = conf->seq_flush+1;
3301 set_bit(STRIPE_BIT_DELAY, &sh->state);
3302 }
cbe47ec5 3303 }
d0852df5 3304 spin_unlock_irq(&sh->stripe_lock);
59fc630b 3305
3306 if (stripe_can_batch(sh))
3307 stripe_add_to_batch_list(conf, sh);
1da177e4
LT		 3308 return 1;
3309
3310 overlap:
3311 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
b17459c0 3312 spin_unlock_irq(&sh->stripe_lock);
1da177e4
LT		 3313 return 0;
3314}
3315
d1688a6d 3316static void end_reshape(struct r5conf *conf);
29269553 3317
d1688a6d 3318static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
911d4ee8 3319 struct stripe_head *sh)
ccfcc3c1 3320{
784052ec 3321 int sectors_per_chunk =
09c9e5fa 3322 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
911d4ee8 3323 int dd_idx;
2d2063ce 3324 int chunk_offset = sector_div(stripe, sectors_per_chunk);
112bf897 3325 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
2d2063ce 3326
112bf897
N		 3327 raid5_compute_sector(conf,
3328 stripe * (disks - conf->max_degraded)
b875e531 3329 *sectors_per_chunk + chunk_offset,
112bf897 3330 previous,
911d4ee8 3331 &dd_idx, sh);
ccfcc3c1
N		 3332}
3333
a4456856 3334static void
d1688a6d 3335handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh,
bd83d0a2 3336 struct stripe_head_state *s, int disks)
a4456856
DW		 3337{
3338 int i;
59fc630b 3339 BUG_ON(sh->batch_head);
a4456856
DW		 3340 for (i = disks; i--; ) {
3341 struct bio *bi;
3342 int bitmap_end = 0;
3343
3344 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
3cb03002 3345 struct md_rdev *rdev;
a4456856
DW		 3346 rcu_read_lock();
3347 rdev = rcu_dereference(conf->disks[i].rdev);
f5b67ae8
N		 3348 if (rdev && test_bit(In_sync, &rdev->flags) &&
3349 !test_bit(Faulty, &rdev->flags))
7f0da59b
N		 3350 atomic_inc(&rdev->nr_pending);
3351 else
3352 rdev = NULL;
a4456856 3353 rcu_read_unlock();
7f0da59b
N		 3354 if (rdev) {
3355 if (!rdev_set_badblocks(
3356 rdev,
3357 sh->sector,
3358 STRIPE_SECTORS, 0))
3359 md_error(conf->mddev, rdev);
3360 rdev_dec_pending(rdev, conf->mddev);
3361 }
a4456856 3362 }
b17459c0 3363 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 3364 /* fail all writes first */
3365 bi = sh->dev[i].towrite;
3366 sh->dev[i].towrite = NULL;
7a87f434 3367 sh->overwrite_disks = 0;
b17459c0 3368 spin_unlock_irq(&sh->stripe_lock);
1ed850f3 3369 if (bi)
a4456856 3370 bitmap_end = 1;
a4456856 3371
ff875738 3372 log_stripe_write_finished(sh);
0576b1c6 3373
a4456856
DW		 3374 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
3375 wake_up(&conf->wait_for_overlap);
3376
4f024f37 3377 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3378 sh->dev[i].sector + STRIPE_SECTORS) {
3379 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
4246a0b6 3380
49728050 3381 md_write_end(conf->mddev);
6308d8e3 3382 bio_io_error(bi);
a4456856
DW		 3383 bi = nextbi;
3384 }
7eaf7e8e
SL		 3385 if (bitmap_end)
3386 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3387 STRIPE_SECTORS, 0, 0);
3388 bitmap_end = 0;
a4456856
DW		 3389 /* and fail all 'written' */
3390 bi = sh->dev[i].written;
3391 sh->dev[i].written = NULL;
d592a996
SL		 3392 if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) {
3393 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
3394 sh->dev[i].page = sh->dev[i].orig_page;
3395 }
3396
a4456856 3397 if (bi) bitmap_end = 1;
4f024f37 3398 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3399 sh->dev[i].sector + STRIPE_SECTORS) {
3400 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
4246a0b6 3401
49728050 3402 md_write_end(conf->mddev);
6308d8e3 3403 bio_io_error(bi);
a4456856
DW		 3404 bi = bi2;
3405 }
3406
b5e98d65
DW		 3407 /* fail any reads if this device is non-operational and
3408 * the data has not reached the cache yet.
3409 */
3410 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
6e74a9cf 3411 s->failed > conf->max_degraded &&
b5e98d65
DW		 3412 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
3413 test_bit(R5_ReadError, &sh->dev[i].flags))) {
143c4d05 3414 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 3415 bi = sh->dev[i].toread;
3416 sh->dev[i].toread = NULL;
143c4d05 3417 spin_unlock_irq(&sh->stripe_lock);
a4456856
DW		 3418 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
3419 wake_up(&conf->wait_for_overlap);
ebda780b
SL		 3420 if (bi)
3421 s->to_read--;
4f024f37 3422 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3423 sh->dev[i].sector + STRIPE_SECTORS) {
3424 struct bio *nextbi =
3425 r5_next_bio(bi, sh->dev[i].sector);
4246a0b6 3426
6308d8e3 3427 bio_io_error(bi);
a4456856
DW		 3428 bi = nextbi;
3429 }
3430 }
a4456856
DW		 3431 if (bitmap_end)
3432 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3433 STRIPE_SECTORS, 0, 0);
8cfa7b0f
N		 3434 /* If we were in the middle of a write the parity block might
3435 * still be locked - so just clear all R5_LOCKED flags
3436 */
3437 clear_bit(R5_LOCKED, &sh->dev[i].flags);
a4456856 3438 }
ebda780b
SL		 3439 s->to_write = 0;
3440 s->written = 0;
a4456856 3441
8b3e6cdc
DW		 3442 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3443 if (atomic_dec_and_test(&conf->pending_full_writes))
3444 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 3445}
3446
7f0da59b 3447static void
d1688a6d 3448handle_failed_sync(struct r5conf *conf, struct stripe_head *sh,
7f0da59b
N		 3449 struct stripe_head_state *s)
3450{
3451 int abort = 0;
3452 int i;
3453
59fc630b 3454 BUG_ON(sh->batch_head);
7f0da59b 3455 clear_bit(STRIPE_SYNCING, &sh->state);
f8dfcffd
N		 3456 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
3457 wake_up(&conf->wait_for_overlap);
7f0da59b 3458 s->syncing = 0;
9a3e1101 3459 s->replacing = 0;
7f0da59b 3460 /* There is nothing more to do for sync/check/repair.
18b9837e
N		 3461 * Don't even need to abort as that is handled elsewhere
3462 * if needed, and not always wanted e.g. if there is a known
3463 * bad block here.
9a3e1101 3464 * For recover/replace we need to record a bad block on all
7f0da59b
N		 3465 * non-sync devices, or abort the recovery
3466 */
18b9837e
N		 3467 if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) {
3468 /* During recovery devices cannot be removed, so
3469 * locking and refcounting of rdevs is not needed
3470 */
e50d3992 3471 rcu_read_lock();
18b9837e 3472 for (i = 0; i < conf->raid_disks; i++) {
e50d3992 3473 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
18b9837e
N		 3474 if (rdev
3475 && !test_bit(Faulty, &rdev->flags)
3476 && !test_bit(In_sync, &rdev->flags)
3477 && !rdev_set_badblocks(rdev, sh->sector,
3478 STRIPE_SECTORS, 0))
3479 abort = 1;
e50d3992 3480 rdev = rcu_dereference(conf->disks[i].replacement);
18b9837e
N		 3481 if (rdev
3482 && !test_bit(Faulty, &rdev->flags)
3483 && !test_bit(In_sync, &rdev->flags)
3484 && !rdev_set_badblocks(rdev, sh->sector,
3485 STRIPE_SECTORS, 0))
3486 abort = 1;
3487 }
e50d3992 3488 rcu_read_unlock();
18b9837e
N		 3489 if (abort)
3490 conf->recovery_disabled =
3491 conf->mddev->recovery_disabled;
7f0da59b 3492 }
18b9837e 3493 md_done_sync(conf->mddev, STRIPE_SECTORS, !abort);
7f0da59b
N		 3494}
3495
9a3e1101
N		 3496static int want_replace(struct stripe_head *sh, int disk_idx)
3497{
3498 struct md_rdev *rdev;
3499 int rv = 0;
3f232d6a
N		 3500
3501 rcu_read_lock();
3502 rdev = rcu_dereference(sh->raid_conf->disks[disk_idx].replacement);
9a3e1101
N		 3503 if (rdev
3504 && !test_bit(Faulty, &rdev->flags)
3505 && !test_bit(In_sync, &rdev->flags)
3506 && (rdev->recovery_offset <= sh->sector
3507 || rdev->mddev->recovery_cp <= sh->sector))
3508 rv = 1;
3f232d6a 3509 rcu_read_unlock();
9a3e1101
N		 3510 return rv;
3511}
3512
2c58f06e
N		 3513static int need_this_block(struct stripe_head *sh, struct stripe_head_state *s,
3514 int disk_idx, int disks)
a4456856 3515{
5599becc 3516 struct r5dev *dev = &sh->dev[disk_idx];
f2b3b44d
N		 3517 struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]],
3518 &sh->dev[s->failed_num[1]] };
ea664c82 3519 int i;
5599becc 3520
a79cfe12
N		 3521
3522 if (test_bit(R5_LOCKED, &dev->flags) ||
3523 test_bit(R5_UPTODATE, &dev->flags))
3524 /* No point reading this as we already have it or have
3525 * decided to get it.
3526 */
3527 return 0;
3528
3529 if (dev->toread ||
3530 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)))
3531 /* We need this block to directly satisfy a request */
3532 return 1;
3533
3534 if (s->syncing || s->expanding ||
3535 (s->replacing && want_replace(sh, disk_idx)))
3536 /* When syncing, or expanding we read everything.
3537 * When replacing, we need the replaced block.
3538 */
3539 return 1;
3540
3541 if ((s->failed >= 1 && fdev[0]->toread) ||
3542 (s->failed >= 2 && fdev[1]->toread))
3543 /* If we want to read from a failed device, then
3544 * we need to actually read every other device.
3545 */
3546 return 1;
3547
a9d56950
N		 3548 /* Sometimes neither read-modify-write nor reconstruct-write
3549 * cycles can work. In those cases we read every block we
3550 * can. Then the parity-update is certain to have enough to
3551 * work with.
3552 * This can only be a problem when we need to write something,
3553 * and some device has failed. If either of those tests
3554 * fail we need look no further.
3555 */
3556 if (!s->failed || !s->to_write)
3557 return 0;
3558
3559 if (test_bit(R5_Insync, &dev->flags) &&
3560 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3561 /* Pre-reads at not permitted until after short delay
3562 * to gather multiple requests. However if this
3560741e 3563 * device is no Insync, the block could only be computed
a9d56950
N		 3564 * and there is no need to delay that.
3565 */
3566 return 0;
ea664c82 3567
36707bb2 3568 for (i = 0; i < s->failed && i < 2; i++) {
ea664c82
N		 3569 if (fdev[i]->towrite &&
3570 !test_bit(R5_UPTODATE, &fdev[i]->flags) &&
3571 !test_bit(R5_OVERWRITE, &fdev[i]->flags))
3572 /* If we have a partial write to a failed
3573 * device, then we will need to reconstruct
3574 * the content of that device, so all other
3575 * devices must be read.
3576 */
3577 return 1;
3578 }
3579
3580 /* If we are forced to do a reconstruct-write, either because
3581 * the current RAID6 implementation only supports that, or
3560741e 3582 * because parity cannot be trusted and we are currently
ea664c82
N		 3583 * recovering it, there is extra need to be careful.
3584 * If one of the devices that we would need to read, because
3585 * it is not being overwritten (and maybe not written at all)
3586 * is missing/faulty, then we need to read everything we can.
3587 */
3588 if (sh->raid_conf->level != 6 &&
3589 sh->sector < sh->raid_conf->mddev->recovery_cp)
3590 /* reconstruct-write isn't being forced */
3591 return 0;
36707bb2 3592 for (i = 0; i < s->failed && i < 2; i++) {
10d82c5f
N		 3593 if (s->failed_num[i] != sh->pd_idx &&
3594 s->failed_num[i] != sh->qd_idx &&
3595 !test_bit(R5_UPTODATE, &fdev[i]->flags) &&
ea664c82
N		 3596 !test_bit(R5_OVERWRITE, &fdev[i]->flags))
3597 return 1;
3598 }
3599
2c58f06e
N		 3600 return 0;
3601}
3602
ba02684d
SL		 3603/* fetch_block - checks the given member device to see if its data needs
3604 * to be read or computed to satisfy a request.
3605 *
3606 * Returns 1 when no more member devices need to be checked, otherwise returns
3607 * 0 to tell the loop in handle_stripe_fill to continue
3608 */
2c58f06e
N		 3609static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s,
3610 int disk_idx, int disks)
3611{
3612 struct r5dev *dev = &sh->dev[disk_idx];
3613
3614 /* is the data in this block needed, and can we get it? */
3615 if (need_this_block(sh, s, disk_idx, disks)) {
5599becc
YT		 3616 /* we would like to get this block, possibly by computing it,
3617 * otherwise read it if the backing disk is insync
3618 */
3619 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
3620 BUG_ON(test_bit(R5_Wantread, &dev->flags));
b0c783b3 3621 BUG_ON(sh->batch_head);
7471fb77
N		 3622
3623 /*
3624 * In the raid6 case if the only non-uptodate disk is P
3625 * then we already trusted P to compute the other failed
3626 * drives. It is safe to compute rather than re-read P.
3627 * In other cases we only compute blocks from failed
3628 * devices, otherwise check/repair might fail to detect
3629 * a real inconsistency.
3630 */
3631
5599becc 3632 if ((s->uptodate == disks - 1) &&
7471fb77 3633 ((sh->qd_idx >= 0 && sh->pd_idx == disk_idx) ||
f2b3b44d 3634 (s->failed && (disk_idx == s->failed_num[0] ||
7471fb77 3635 disk_idx == s->failed_num[1])))) {
5599becc
YT		 3636 /* have disk failed, and we're requested to fetch it;
3637 * do compute it
a4456856 3638 */
5599becc
YT		 3639 pr_debug("Computing stripe %llu block %d\n",
3640 (unsigned long long)sh->sector, disk_idx);
3641 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3642 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3643 set_bit(R5_Wantcompute, &dev->flags);
3644 sh->ops.target = disk_idx;
3645 sh->ops.target2 = -1; /* no 2nd target */
3646 s->req_compute = 1;
93b3dbce
N		 3647 /* Careful: from this point on 'uptodate' is in the eye
3648 * of raid_run_ops which services 'compute' operations
3649 * before writes. R5_Wantcompute flags a block that will
3650 * be R5_UPTODATE by the time it is needed for a
3651 * subsequent operation.
3652 */
5599becc
YT		 3653 s->uptodate++;
3654 return 1;
3655 } else if (s->uptodate == disks-2 && s->failed >= 2) {
3656 /* Computing 2-failure is *very* expensive; only
3657 * do it if failed >= 2
3658 */
3659 int other;
3660 for (other = disks; other--; ) {
3661 if (other == disk_idx)
3662 continue;
3663 if (!test_bit(R5_UPTODATE,
3664 &sh->dev[other].flags))
3665 break;
a4456856 3666 }
5599becc
YT		 3667 BUG_ON(other < 0);
3668 pr_debug("Computing stripe %llu blocks %d,%d\n",
3669 (unsigned long long)sh->sector,
3670 disk_idx, other);
3671 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3672 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3673 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
3674 set_bit(R5_Wantcompute, &sh->dev[other].flags);
3675 sh->ops.target = disk_idx;
3676 sh->ops.target2 = other;
3677 s->uptodate += 2;
3678 s->req_compute = 1;
3679 return 1;
3680 } else if (test_bit(R5_Insync, &dev->flags)) {
3681 set_bit(R5_LOCKED, &dev->flags);
3682 set_bit(R5_Wantread, &dev->flags);
3683 s->locked++;
3684 pr_debug("Reading block %d (sync=%d)\n",
3685 disk_idx, s->syncing);
a4456856
DW		 3686 }
3687 }
5599becc
YT		 3688
3689 return 0;
3690}
3691
3692/**
93b3dbce 3693 * handle_stripe_fill - read or compute data to satisfy pending requests.
5599becc 3694 */
93b3dbce
N		 3695static void handle_stripe_fill(struct stripe_head *sh,
3696 struct stripe_head_state *s,
3697 int disks)
5599becc
YT		 3698{
3699 int i;
3700
3701 /* look for blocks to read/compute, skip this if a compute
3702 * is already in flight, or if the stripe contents are in the
3703 * midst of changing due to a write
3704 */
3705 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
07e83364
SL		 3706 !sh->reconstruct_state) {
3707
3708 /*
3709 * For degraded stripe with data in journal, do not handle
3710 * read requests yet, instead, flush the stripe to raid
3711 * disks first, this avoids handling complex rmw of write
3712 * back cache (prexor with orig_page, and then xor with
3713 * page) in the read path
3714 */
3715 if (s->injournal && s->failed) {
3716 if (test_bit(STRIPE_R5C_CACHING, &sh->state))
3717 r5c_make_stripe_write_out(sh);
3718 goto out;
3719 }
3720
5599becc 3721 for (i = disks; i--; )
93b3dbce 3722 if (fetch_block(sh, s, i, disks))
5599becc 3723 break;
07e83364
SL		 3724 }
3725out:
a4456856
DW		 3726 set_bit(STRIPE_HANDLE, &sh->state);
3727}
3728
787b76fa
N		 3729static void break_stripe_batch_list(struct stripe_head *head_sh,
3730 unsigned long handle_flags);
1fe797e6 3731/* handle_stripe_clean_event
a4456856
DW		 3732 * any written block on an uptodate or failed drive can be returned.
3733 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
3734 * never LOCKED, so we don't need to test 'failed' directly.
3735 */
d1688a6d 3736static void handle_stripe_clean_event(struct r5conf *conf,
bd83d0a2 3737 struct stripe_head *sh, int disks)
a4456856
DW		 3738{
3739 int i;
3740 struct r5dev *dev;
f8dfcffd 3741 int discard_pending = 0;
59fc630b 3742 struct stripe_head *head_sh = sh;
3743 bool do_endio = false;
a4456856
DW		 3744
3745 for (i = disks; i--; )
3746 if (sh->dev[i].written) {
3747 dev = &sh->dev[i];
3748 if (!test_bit(R5_LOCKED, &dev->flags) &&
9e444768 3749 (test_bit(R5_UPTODATE, &dev->flags) ||
d592a996
SL		 3750 test_bit(R5_Discard, &dev->flags) ||
3751 test_bit(R5_SkipCopy, &dev->flags))) {
a4456856
DW		 3752 /* We can return any write requests */
3753 struct bio *wbi, *wbi2;
45b4233c 3754 pr_debug("Return write for disc %d\n", i);
ca64cae9
N		 3755 if (test_and_clear_bit(R5_Discard, &dev->flags))
3756 clear_bit(R5_UPTODATE, &dev->flags);
d592a996
SL		 3757 if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) {
3758 WARN_ON(test_bit(R5_UPTODATE, &dev->flags));
d592a996 3759 }
59fc630b 3760 do_endio = true;
3761
3762returnbi:
3763 dev->page = dev->orig_page;
a4456856
DW		 3764 wbi = dev->written;
3765 dev->written = NULL;
4f024f37 3766 while (wbi && wbi->bi_iter.bi_sector <
a4456856
DW		 3767 dev->sector + STRIPE_SECTORS) {
3768 wbi2 = r5_next_bio(wbi, dev->sector);
49728050 3769 md_write_end(conf->mddev);
016c76ac 3770 bio_endio(wbi);
a4456856
DW		 3771 wbi = wbi2;
3772 }
7eaf7e8e
SL		 3773 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3774 STRIPE_SECTORS,
a4456856 3775 !test_bit(STRIPE_DEGRADED, &sh->state),
7eaf7e8e 3776 0);
59fc630b 3777 if (head_sh->batch_head) {
3778 sh = list_first_entry(&sh->batch_list,
3779 struct stripe_head,
3780 batch_list);
3781 if (sh != head_sh) {
3782 dev = &sh->dev[i];
3783 goto returnbi;
3784 }
3785 }
3786 sh = head_sh;
3787 dev = &sh->dev[i];
f8dfcffd
N		 3788 } else if (test_bit(R5_Discard, &dev->flags))
3789 discard_pending = 1;
3790 }
f6bed0ef 3791
ff875738 3792 log_stripe_write_finished(sh);
0576b1c6 3793
f8dfcffd
N		 3794 if (!discard_pending &&
3795 test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) {
b8a9d66d 3796 int hash;
f8dfcffd
N		 3797 clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
3798 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
3799 if (sh->qd_idx >= 0) {
3800 clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
3801 clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags);
3802 }
3803 /* now that discard is done we can proceed with any sync */
3804 clear_bit(STRIPE_DISCARD, &sh->state);
d47648fc
SL		 3805 /*
3806 * SCSI discard will change some bio fields and the stripe has
3807 * no updated data, so remove it from hash list and the stripe
3808 * will be reinitialized
3809 */
59fc630b 3810unhash:
b8a9d66d
RG		 3811 hash = sh->hash_lock_index;
3812 spin_lock_irq(conf->hash_locks + hash);
d47648fc 3813 remove_hash(sh);
b8a9d66d 3814 spin_unlock_irq(conf->hash_locks + hash);
59fc630b 3815 if (head_sh->batch_head) {
3816 sh = list_first_entry(&sh->batch_list,
3817 struct stripe_head, batch_list);
3818 if (sh != head_sh)
3819 goto unhash;
3820 }
59fc630b 3821 sh = head_sh;
3822
f8dfcffd
N		 3823 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state))
3824 set_bit(STRIPE_HANDLE, &sh->state);
3825
3826 }
8b3e6cdc
DW		 3827
3828 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3829 if (atomic_dec_and_test(&conf->pending_full_writes))
3830 md_wakeup_thread(conf->mddev->thread);
59fc630b 3831
787b76fa
N		 3832 if (head_sh->batch_head && do_endio)
3833 break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS);
a4456856
DW		 3834}
3835
86aa1397
SL		 3836/*
3837 * For RMW in write back cache, we need extra page in prexor to store the
3838 * old data. This page is stored in dev->orig_page.
3839 *
3840 * This function checks whether we have data for prexor. The exact logic
3841 * is:
3842 * R5_UPTODATE && (!R5_InJournal || R5_OrigPageUPTDODATE)
3843 */
3844static inline bool uptodate_for_rmw(struct r5dev *dev)
3845{
3846 return (test_bit(R5_UPTODATE, &dev->flags)) &&
3847 (!test_bit(R5_InJournal, &dev->flags) ||
3848 test_bit(R5_OrigPageUPTDODATE, &dev->flags));
3849}
3850
d7bd398e
SL		 3851static int handle_stripe_dirtying(struct r5conf *conf,
3852 struct stripe_head *sh,
3853 struct stripe_head_state *s,
3854 int disks)
a4456856
DW		 3855{
3856 int rmw = 0, rcw = 0, i;
a7854487
AL		 3857 sector_t recovery_cp = conf->mddev->recovery_cp;
3858
584acdd4 3859 /* Check whether resync is now happening or should start.
a7854487
AL		 3860 * If yes, then the array is dirty (after unclean shutdown or
3861 * initial creation), so parity in some stripes might be inconsistent.
3862 * In this case, we need to always do reconstruct-write, to ensure
3863 * that in case of drive failure or read-error correction, we
3864 * generate correct data from the parity.
3865 */
584acdd4 3866 if (conf->rmw_level == PARITY_DISABLE_RMW ||
26ac1073
N		 3867 (recovery_cp < MaxSector && sh->sector >= recovery_cp &&
3868 s->failed == 0)) {
a7854487 3869 /* Calculate the real rcw later - for now make it
c8ac1803
N		 3870 * look like rcw is cheaper
3871 */
3872 rcw = 1; rmw = 2;
584acdd4
MS		 3873 pr_debug("force RCW rmw_level=%u, recovery_cp=%llu sh->sector=%llu\n",
3874 conf->rmw_level, (unsigned long long)recovery_cp,
a7854487 3875 (unsigned long long)sh->sector);
c8ac1803 3876 } else for (i = disks; i--; ) {
a4456856
DW		 3877 /* would I have to read this buffer for read_modify_write */
3878 struct r5dev *dev = &sh->dev[i];
39b99586 3879 if (((dev->towrite && !delay_towrite(conf, dev, s)) ||
07e83364 3880 i == sh->pd_idx || i == sh->qd_idx ||
1e6d690b 3881 test_bit(R5_InJournal, &dev->flags)) &&
a4456856 3882 !test_bit(R5_LOCKED, &dev->flags) &&
86aa1397 3883 !(uptodate_for_rmw(dev) ||
f38e1219 3884 test_bit(R5_Wantcompute, &dev->flags))) {
a4456856
DW		 3885 if (test_bit(R5_Insync, &dev->flags))
3886 rmw++;
3887 else
3888 rmw += 2*disks; /* cannot read it */
3889 }
3890 /* Would I have to read this buffer for reconstruct_write */
584acdd4
MS		 3891 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
3892 i != sh->pd_idx && i != sh->qd_idx &&
a4456856 3893 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219 3894 !(test_bit(R5_UPTODATE, &dev->flags) ||
1e6d690b 3895 test_bit(R5_Wantcompute, &dev->flags))) {
67f45548
N		 3896 if (test_bit(R5_Insync, &dev->flags))
3897 rcw++;
a4456856
DW		 3898 else
3899 rcw += 2*disks;
3900 }
3901 }
1e6d690b 3902
39b99586
SL		 3903 pr_debug("for sector %llu state 0x%lx, rmw=%d rcw=%d\n",
3904 (unsigned long long)sh->sector, sh->state, rmw, rcw);
a4456856 3905 set_bit(STRIPE_HANDLE, &sh->state);
41257580 3906 if ((rmw < rcw || (rmw == rcw && conf->rmw_level == PARITY_PREFER_RMW)) && rmw > 0) {
a4456856 3907 /* prefer read-modify-write, but need to get some data */
e3620a3a
JB		 3908 if (conf->mddev->queue)
3909 blk_add_trace_msg(conf->mddev->queue,
3910 "raid5 rmw %llu %d",
3911 (unsigned long long)sh->sector, rmw);
a4456856
DW		 3912 for (i = disks; i--; ) {
3913 struct r5dev *dev = &sh->dev[i];
1e6d690b
SL		 3914 if (test_bit(R5_InJournal, &dev->flags) &&
3915 dev->page == dev->orig_page &&
3916 !test_bit(R5_LOCKED, &sh->dev[sh->pd_idx].flags)) {
3917 /* alloc page for prexor */
d7bd398e
SL		 3918 struct page *p = alloc_page(GFP_NOIO);
3919
3920 if (p) {
3921 dev->orig_page = p;
3922 continue;
3923 }
3924
3925 /*
3926 * alloc_page() failed, try use
3927 * disk_info->extra_page
3928 */
3929 if (!test_and_set_bit(R5C_EXTRA_PAGE_IN_USE,
3930 &conf->cache_state)) {
3931 r5c_use_extra_page(sh);
3932 break;
3933 }
1e6d690b 3934
d7bd398e
SL		 3935 /* extra_page in use, add to delayed_list */
3936 set_bit(STRIPE_DELAYED, &sh->state);
3937 s->waiting_extra_page = 1;
3938 return -EAGAIN;
1e6d690b 3939 }
d7bd398e 3940 }
1e6d690b 3941
d7bd398e
SL		 3942 for (i = disks; i--; ) {
3943 struct r5dev *dev = &sh->dev[i];
39b99586 3944 if (((dev->towrite && !delay_towrite(conf, dev, s)) ||
1e6d690b
SL		 3945 i == sh->pd_idx || i == sh->qd_idx ||
3946 test_bit(R5_InJournal, &dev->flags)) &&
a4456856 3947 !test_bit(R5_LOCKED, &dev->flags) &&
86aa1397 3948 !(uptodate_for_rmw(dev) ||
1e6d690b 3949 test_bit(R5_Wantcompute, &dev->flags)) &&
a4456856 3950 test_bit(R5_Insync, &dev->flags)) {
67f45548
N		 3951 if (test_bit(STRIPE_PREREAD_ACTIVE,
3952 &sh->state)) {
3953 pr_debug("Read_old block %d for r-m-w\n",
3954 i);
a4456856
DW		 3955 set_bit(R5_LOCKED, &dev->flags);
3956 set_bit(R5_Wantread, &dev->flags);
3957 s->locked++;
3958 } else {
3959 set_bit(STRIPE_DELAYED, &sh->state);
3960 set_bit(STRIPE_HANDLE, &sh->state);
3961 }
3962 }
3963 }
a9add5d9 3964 }
41257580 3965 if ((rcw < rmw || (rcw == rmw && conf->rmw_level != PARITY_PREFER_RMW)) && rcw > 0) {
a4456856 3966 /* want reconstruct write, but need to get some data */
a9add5d9 3967 int qread =0;
c8ac1803 3968 rcw = 0;
a4456856
DW		 3969 for (i = disks; i--; ) {
3970 struct r5dev *dev = &sh->dev[i];
3971 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
c8ac1803 3972 i != sh->pd_idx && i != sh->qd_idx &&
a4456856 3973 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219 3974 !(test_bit(R5_UPTODATE, &dev->flags) ||
c8ac1803
N		 3975 test_bit(R5_Wantcompute, &dev->flags))) {
3976 rcw++;
67f45548
N		 3977 if (test_bit(R5_Insync, &dev->flags) &&
3978 test_bit(STRIPE_PREREAD_ACTIVE,
3979 &sh->state)) {
45b4233c 3980 pr_debug("Read_old block "
a4456856
DW		 3981 "%d for Reconstruct\n", i);
3982 set_bit(R5_LOCKED, &dev->flags);
3983 set_bit(R5_Wantread, &dev->flags);
3984 s->locked++;
a9add5d9 3985 qread++;
a4456856
DW		 3986 } else {
3987 set_bit(STRIPE_DELAYED, &sh->state);
3988 set_bit(STRIPE_HANDLE, &sh->state);
3989 }
3990 }
3991 }
e3620a3a 3992 if (rcw && conf->mddev->queue)
a9add5d9
N		 3993 blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d",
3994 (unsigned long long)sh->sector,
3995 rcw, qread, test_bit(STRIPE_DELAYED, &sh->state));
c8ac1803 3996 }
b1b02fe9
N		 3997
3998 if (rcw > disks && rmw > disks &&
3999 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4000 set_bit(STRIPE_DELAYED, &sh->state);
4001
a4456856
DW		 4002 /* now if nothing is locked, and if we have enough data,
4003 * we can start a write request
4004 */
f38e1219
DW		 4005 /* since handle_stripe can be called at any time we need to handle the
4006 * case where a compute block operation has been submitted and then a
ac6b53b6
DW		 4007 * subsequent call wants to start a write request. raid_run_ops only
4008 * handles the case where compute block and reconstruct are requested
f38e1219
DW		 4009 * simultaneously. If this is not the case then new writes need to be
4010 * held off until the compute completes.
4011 */
976ea8d4
DW		 4012 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
4013 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
1e6d690b 4014 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
c0f7bddb 4015 schedule_reconstruction(sh, s, rcw == 0, 0);
d7bd398e 4016 return 0;
a4456856
DW		 4017}
4018
d1688a6d 4019static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh,
a4456856
DW		 4020 struct stripe_head_state *s, int disks)
4021{
ecc65c9b 4022 struct r5dev *dev = NULL;
bd2ab670 4023
59fc630b 4024 BUG_ON(sh->batch_head);
a4456856 4025 set_bit(STRIPE_HANDLE, &sh->state);
e89f8962 4026
ecc65c9b
DW		 4027 switch (sh->check_state) {
4028 case check_state_idle:
4029 /* start a new check operation if there are no failures */
bd2ab670 4030 if (s->failed == 0) {
bd2ab670 4031 BUG_ON(s->uptodate != disks);
ecc65c9b
DW		 4032 sh->check_state = check_state_run;
4033 set_bit(STRIPE_OP_CHECK, &s->ops_request);
bd2ab670 4034 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
bd2ab670 4035 s->uptodate--;
ecc65c9b 4036 break;
bd2ab670 4037 }
f2b3b44d 4038 dev = &sh->dev[s->failed_num[0]];
ecc65c9b
DW		 4039 /* fall through */
4040 case check_state_compute_result:
4041 sh->check_state = check_state_idle;
4042 if (!dev)
4043 dev = &sh->dev[sh->pd_idx];
4044
4045 /* check that a write has not made the stripe insync */
4046 if (test_bit(STRIPE_INSYNC, &sh->state))
4047 break;
c8894419 4048
a4456856 4049 /* either failed parity check, or recovery is happening */
a4456856
DW		 4050 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
4051 BUG_ON(s->uptodate != disks);
4052
4053 set_bit(R5_LOCKED, &dev->flags);
ecc65c9b 4054 s->locked++;
a4456856 4055 set_bit(R5_Wantwrite, &dev->flags);
830ea016 4056
a4456856 4057 clear_bit(STRIPE_DEGRADED, &sh->state);
a4456856 4058 set_bit(STRIPE_INSYNC, &sh->state);
ecc65c9b
DW		 4059 break;
4060 case check_state_run:
4061 break; /* we will be called again upon completion */
4062 case check_state_check_result:
4063 sh->check_state = check_state_idle;
4064
4065 /* if a failure occurred during the check operation, leave
4066 * STRIPE_INSYNC not set and let the stripe be handled again
4067 */
4068 if (s->failed)
4069 break;
4070
4071 /* handle a successful check operation, if parity is correct
4072 * we are done. Otherwise update the mismatch count and repair
4073 * parity if !MD_RECOVERY_CHECK
4074 */
ad283ea4 4075 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
ecc65c9b
DW		 4076 /* parity is correct (on disc,
4077 * not in buffer any more)
4078 */
4079 set_bit(STRIPE_INSYNC, &sh->state);
4080 else {
7f7583d4 4081 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
e1539036 4082 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) {
ecc65c9b
DW		 4083 /* don't try to repair!! */
4084 set_bit(STRIPE_INSYNC, &sh->state);
e1539036
N		 4085 pr_warn_ratelimited("%s: mismatch sector in range "
4086 "%llu-%llu\n", mdname(conf->mddev),
4087 (unsigned long long) sh->sector,
4088 (unsigned long long) sh->sector +
4089 STRIPE_SECTORS);
4090 } else {
ecc65c9b 4091 sh->check_state = check_state_compute_run;
976ea8d4 4092 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
ecc65c9b
DW		 4093 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
4094 set_bit(R5_Wantcompute,
4095 &sh->dev[sh->pd_idx].flags);
4096 sh->ops.target = sh->pd_idx;
ac6b53b6 4097 sh->ops.target2 = -1;
ecc65c9b
DW		 4098 s->uptodate++;
4099 }
4100 }
4101 break;
4102 case check_state_compute_run:
4103 break;
4104 default:
cc6167b4 4105 pr_err("%s: unknown check_state: %d sector: %llu\n",
ecc65c9b
DW		 4106 __func__, sh->check_state,
4107 (unsigned long long) sh->sector);
4108 BUG();
a4456856
DW		 4109 }
4110}
4111
d1688a6d 4112static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh,
36d1c647 4113 struct stripe_head_state *s,
f2b3b44d 4114 int disks)
a4456856 4115{
a4456856 4116 int pd_idx = sh->pd_idx;
34e04e87 4117 int qd_idx = sh->qd_idx;
d82dfee0 4118 struct r5dev *dev;
a4456856 4119
59fc630b 4120 BUG_ON(sh->batch_head);
a4456856
DW		 4121 set_bit(STRIPE_HANDLE, &sh->state);
4122
4123 BUG_ON(s->failed > 2);
d82dfee0 4124
a4456856
DW		 4125 /* Want to check and possibly repair P and Q.
4126 * However there could be one 'failed' device, in which
4127 * case we can only check one of them, possibly using the
4128 * other to generate missing data
4129 */
4130
d82dfee0
DW		 4131 switch (sh->check_state) {
4132 case check_state_idle:
4133 /* start a new check operation if there are < 2 failures */
f2b3b44d 4134 if (s->failed == s->q_failed) {
d82dfee0 4135 /* The only possible failed device holds Q, so it
a4456856
DW		 4136 * makes sense to check P (If anything else were failed,
4137 * we would have used P to recreate it).
4138 */
d82dfee0 4139 sh->check_state = check_state_run;
a4456856 4140 }
f2b3b44d 4141 if (!s->q_failed && s->failed < 2) {
d82dfee0 4142 /* Q is not failed, and we didn't use it to generate
a4456856
DW		 4143 * anything, so it makes sense to check it
4144 */
d82dfee0
DW		 4145 if (sh->check_state == check_state_run)
4146 sh->check_state = check_state_run_pq;
4147 else
4148 sh->check_state = check_state_run_q;
a4456856 4149 }
a4456856 4150
d82dfee0
DW		 4151 /* discard potentially stale zero_sum_result */
4152 sh->ops.zero_sum_result = 0;
a4456856 4153
d82dfee0
DW		 4154 if (sh->check_state == check_state_run) {
4155 /* async_xor_zero_sum destroys the contents of P */
4156 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
4157 s->uptodate--;
a4456856 4158 }
d82dfee0
DW		 4159 if (sh->check_state >= check_state_run &&
4160 sh->check_state <= check_state_run_pq) {
4161 /* async_syndrome_zero_sum preserves P and Q, so
4162 * no need to mark them !uptodate here
4163 */
4164 set_bit(STRIPE_OP_CHECK, &s->ops_request);
4165 break;
a4456856
DW		 4166 }
4167
d82dfee0
DW		 4168 /* we have 2-disk failure */
4169 BUG_ON(s->failed != 2);
4170 /* fall through */
4171 case check_state_compute_result:
4172 sh->check_state = check_state_idle;
a4456856 4173
d82dfee0
DW		 4174 /* check that a write has not made the stripe insync */
4175 if (test_bit(STRIPE_INSYNC, &sh->state))
4176 break;
a4456856
DW		 4177
4178 /* now write out any block on a failed drive,
d82dfee0 4179 * or P or Q if they were recomputed
a4456856 4180 */
d82dfee0 4181 BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */
a4456856 4182 if (s->failed == 2) {
f2b3b44d 4183 dev = &sh->dev[s->failed_num[1]];
a4456856
DW		 4184 s->locked++;
4185 set_bit(R5_LOCKED, &dev->flags);
4186 set_bit(R5_Wantwrite, &dev->flags);
4187 }
4188 if (s->failed >= 1) {
f2b3b44d 4189 dev = &sh->dev[s->failed_num[0]];
a4456856
DW		 4190 s->locked++;
4191 set_bit(R5_LOCKED, &dev->flags);
4192 set_bit(R5_Wantwrite, &dev->flags);
4193 }
d82dfee0 4194 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
a4456856
DW		 4195 dev = &sh->dev[pd_idx];
4196 s->locked++;
4197 set_bit(R5_LOCKED, &dev->flags);
4198 set_bit(R5_Wantwrite, &dev->flags);
4199 }
d82dfee0 4200 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
a4456856
DW		 4201 dev = &sh->dev[qd_idx];
4202 s->locked++;
4203 set_bit(R5_LOCKED, &dev->flags);
4204 set_bit(R5_Wantwrite, &dev->flags);
4205 }
4206 clear_bit(STRIPE_DEGRADED, &sh->state);
4207
4208 set_bit(STRIPE_INSYNC, &sh->state);
d82dfee0
DW		 4209 break;
4210 case check_state_run:
4211 case check_state_run_q:
4212 case check_state_run_pq:
4213 break; /* we will be called again upon completion */
4214 case check_state_check_result:
4215 sh->check_state = check_state_idle;
4216
4217 /* handle a successful check operation, if parity is correct
4218 * we are done. Otherwise update the mismatch count and repair
4219 * parity if !MD_RECOVERY_CHECK
4220 */
4221 if (sh->ops.zero_sum_result == 0) {
4222 /* both parities are correct */
4223 if (!s->failed)
4224 set_bit(STRIPE_INSYNC, &sh->state);
4225 else {
4226 /* in contrast to the raid5 case we can validate
4227 * parity, but still have a failure to write
4228 * back
4229 */
4230 sh->check_state = check_state_compute_result;
4231 /* Returning at this point means that we may go
4232 * off and bring p and/or q uptodate again so
4233 * we make sure to check zero_sum_result again
4234 * to verify if p or q need writeback
4235 */
4236 }
4237 } else {
7f7583d4 4238 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
e1539036 4239 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) {
d82dfee0
DW		 4240 /* don't try to repair!! */
4241 set_bit(STRIPE_INSYNC, &sh->state);
e1539036
N		 4242 pr_warn_ratelimited("%s: mismatch sector in range "
4243 "%llu-%llu\n", mdname(conf->mddev),
4244 (unsigned long long) sh->sector,
4245 (unsigned long long) sh->sector +
4246 STRIPE_SECTORS);
4247 } else {
d82dfee0
DW		 4248 int *target = &sh->ops.target;
4249
4250 sh->ops.target = -1;
4251 sh->ops.target2 = -1;
4252 sh->check_state = check_state_compute_run;
4253 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
4254 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
4255 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
4256 set_bit(R5_Wantcompute,
4257 &sh->dev[pd_idx].flags);
4258 *target = pd_idx;
4259 target = &sh->ops.target2;
4260 s->uptodate++;
4261 }
4262 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
4263 set_bit(R5_Wantcompute,
4264 &sh->dev[qd_idx].flags);
4265 *target = qd_idx;
4266 s->uptodate++;
4267 }
4268 }
4269 }
4270 break;
4271 case check_state_compute_run:
4272 break;
4273 default:
cc6167b4
N		 4274 pr_warn("%s: unknown check_state: %d sector: %llu\n",
4275 __func__, sh->check_state,
4276 (unsigned long long) sh->sector);
d82dfee0 4277 BUG();
a4456856
DW		 4278 }
4279}
4280
d1688a6d 4281static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh)
a4456856
DW		 4282{
4283 int i;
4284
4285 /* We have read all the blocks in this stripe and now we need to
4286 * copy some of them into a target stripe for expand.
4287 */
f0a50d37 4288 struct dma_async_tx_descriptor *tx = NULL;
59fc630b 4289 BUG_ON(sh->batch_head);
a4456856
DW		 4290 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
4291 for (i = 0; i < sh->disks; i++)
34e04e87 4292 if (i != sh->pd_idx && i != sh->qd_idx) {
911d4ee8 4293 int dd_idx, j;
a4456856 4294 struct stripe_head *sh2;
a08abd8c 4295 struct async_submit_ctl submit;
a4456856 4296
6d036f7d 4297 sector_t bn = raid5_compute_blocknr(sh, i, 1);
911d4ee8
N		 4298 sector_t s = raid5_compute_sector(conf, bn, 0,
4299 &dd_idx, NULL);
6d036f7d 4300 sh2 = raid5_get_active_stripe(conf, s, 0, 1, 1);
a4456856
DW		 4301 if (sh2 == NULL)
4302 /* so far only the early blocks of this stripe
4303 * have been requested. When later blocks
4304 * get requested, we will try again
4305 */
4306 continue;
4307 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
4308 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
4309 /* must have already done this block */
6d036f7d 4310 raid5_release_stripe(sh2);
a4456856
DW		 4311 continue;
4312 }
f0a50d37
DW		 4313
4314 /* place all the copies on one channel */
a08abd8c 4315 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
f0a50d37 4316 tx = async_memcpy(sh2->dev[dd_idx].page,
88ba2aa5 4317 sh->dev[i].page, 0, 0, STRIPE_SIZE,
a08abd8c 4318 &submit);
f0a50d37 4319
a4456856
DW		 4320 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
4321 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
4322 for (j = 0; j < conf->raid_disks; j++)
4323 if (j != sh2->pd_idx &&
86c374ba 4324 j != sh2->qd_idx &&
a4456856
DW		 4325 !test_bit(R5_Expanded, &sh2->dev[j].flags))
4326 break;
4327 if (j == conf->raid_disks) {
4328 set_bit(STRIPE_EXPAND_READY, &sh2->state);
4329 set_bit(STRIPE_HANDLE, &sh2->state);
4330 }
6d036f7d 4331 raid5_release_stripe(sh2);
f0a50d37 4332
a4456856 4333 }
a2e08551 4334 /* done submitting copies, wait for them to complete */
749586b7 4335 async_tx_quiesce(&tx);
a4456856 4336}
1da177e4
LT		 4337
4338/*
4339 * handle_stripe - do things to a stripe.
4340 *
9a3e1101
N		 4341 * We lock the stripe by setting STRIPE_ACTIVE and then examine the
4342 * state of various bits to see what needs to be done.
1da177e4 4343 * Possible results:
9a3e1101
N		 4344 * return some read requests which now have data
4345 * return some write requests which are safely on storage
1da177e4
LT		 4346 * schedule a read on some buffers
4347 * schedule a write of some buffers
4348 * return confirmation of parity correctness
4349 *
1da177e4 4350 */
a4456856 4351
acfe726b 4352static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s)
1da177e4 4353{
d1688a6d 4354 struct r5conf *conf = sh->raid_conf;
f416885e 4355 int disks = sh->disks;
474af965
N		 4356 struct r5dev *dev;
4357 int i;
9a3e1101 4358 int do_recovery = 0;
1da177e4 4359
acfe726b
N		 4360 memset(s, 0, sizeof(*s));
4361
dabc4ec6 4362 s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state) && !sh->batch_head;
4363 s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state) && !sh->batch_head;
acfe726b
N		 4364 s->failed_num[0] = -1;
4365 s->failed_num[1] = -1;
6e74a9cf 4366 s->log_failed = r5l_log_disk_error(conf);
1da177e4 4367
acfe726b 4368 /* Now to look around and see what can be done */
1da177e4 4369 rcu_read_lock();
16a53ecc 4370 for (i=disks; i--; ) {
3cb03002 4371 struct md_rdev *rdev;
31c176ec
N		 4372 sector_t first_bad;
4373 int bad_sectors;
4374 int is_bad = 0;
acfe726b 4375
16a53ecc 4376 dev = &sh->dev[i];
1da177e4 4377
45b4233c 4378 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
9a3e1101
N		 4379 i, dev->flags,
4380 dev->toread, dev->towrite, dev->written);
6c0069c0
YT		 4381 /* maybe we can reply to a read
4382 *
4383 * new wantfill requests are only permitted while
4384 * ops_complete_biofill is guaranteed to be inactive
4385 */
4386 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
4387 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
4388 set_bit(R5_Wantfill, &dev->flags);
1da177e4 4389
16a53ecc 4390 /* now count some things */
cc94015a
N		 4391 if (test_bit(R5_LOCKED, &dev->flags))
4392 s->locked++;
4393 if (test_bit(R5_UPTODATE, &dev->flags))
4394 s->uptodate++;
2d6e4ecc 4395 if (test_bit(R5_Wantcompute, &dev->flags)) {
cc94015a
N		 4396 s->compute++;
4397 BUG_ON(s->compute > 2);
2d6e4ecc 4398 }
1da177e4 4399
acfe726b 4400 if (test_bit(R5_Wantfill, &dev->flags))
cc94015a 4401 s->to_fill++;
acfe726b 4402 else if (dev->toread)
cc94015a 4403 s->to_read++;
16a53ecc 4404 if (dev->towrite) {
cc94015a 4405 s->to_write++;
16a53ecc 4406 if (!test_bit(R5_OVERWRITE, &dev->flags))
cc94015a 4407 s->non_overwrite++;
16a53ecc 4408 }
a4456856 4409 if (dev->written)
cc94015a 4410 s->written++;
14a75d3e
N		 4411 /* Prefer to use the replacement for reads, but only
4412 * if it is recovered enough and has no bad blocks.
4413 */
4414 rdev = rcu_dereference(conf->disks[i].replacement);
4415 if (rdev && !test_bit(Faulty, &rdev->flags) &&
4416 rdev->recovery_offset >= sh->sector + STRIPE_SECTORS &&
4417 !is_badblock(rdev, sh->sector, STRIPE_SECTORS,
4418 &first_bad, &bad_sectors))
4419 set_bit(R5_ReadRepl, &dev->flags);
4420 else {
e6030cb0 4421 if (rdev && !test_bit(Faulty, &rdev->flags))
9a3e1101 4422 set_bit(R5_NeedReplace, &dev->flags);
e6030cb0
N		 4423 else
4424 clear_bit(R5_NeedReplace, &dev->flags);
14a75d3e
N		 4425 rdev = rcu_dereference(conf->disks[i].rdev);
4426 clear_bit(R5_ReadRepl, &dev->flags);
4427 }
9283d8c5
N		 4428 if (rdev && test_bit(Faulty, &rdev->flags))
4429 rdev = NULL;
31c176ec
N		 4430 if (rdev) {
4431 is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
4432 &first_bad, &bad_sectors);
4433 if (s->blocked_rdev == NULL
4434 && (test_bit(Blocked, &rdev->flags)
4435 || is_bad < 0)) {
4436 if (is_bad < 0)
4437 set_bit(BlockedBadBlocks,
4438 &rdev->flags);
4439 s->blocked_rdev = rdev;
4440 atomic_inc(&rdev->nr_pending);
4441 }
6bfe0b49 4442 }
415e72d0
N		 4443 clear_bit(R5_Insync, &dev->flags);
4444 if (!rdev)
4445 /* Not in-sync */;
31c176ec
N		 4446 else if (is_bad) {
4447 /* also not in-sync */
18b9837e
N		 4448 if (!test_bit(WriteErrorSeen, &rdev->flags) &&
4449 test_bit(R5_UPTODATE, &dev->flags)) {
31c176ec
N		 4450 /* treat as in-sync, but with a read error
4451 * which we can now try to correct
4452 */
4453 set_bit(R5_Insync, &dev->flags);
4454 set_bit(R5_ReadError, &dev->flags);
4455 }
4456 } else if (test_bit(In_sync, &rdev->flags))
415e72d0 4457 set_bit(R5_Insync, &dev->flags);
30d7a483 4458 else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
415e72d0 4459 /* in sync if before recovery_offset */
30d7a483
N		 4460 set_bit(R5_Insync, &dev->flags);
4461 else if (test_bit(R5_UPTODATE, &dev->flags) &&
4462 test_bit(R5_Expanded, &dev->flags))
4463 /* If we've reshaped into here, we assume it is Insync.
4464 * We will shortly update recovery_offset to make
4465 * it official.
4466 */
4467 set_bit(R5_Insync, &dev->flags);
4468
1cc03eb9 4469 if (test_bit(R5_WriteError, &dev->flags)) {
14a75d3e
N		 4470 /* This flag does not apply to '.replacement'
4471 * only to .rdev, so make sure to check that*/
4472 struct md_rdev *rdev2 = rcu_dereference(
4473 conf->disks[i].rdev);
4474 if (rdev2 == rdev)
4475 clear_bit(R5_Insync, &dev->flags);
4476 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
bc2607f3 4477 s->handle_bad_blocks = 1;
14a75d3e 4478 atomic_inc(&rdev2->nr_pending);
bc2607f3
N		 4479 } else
4480 clear_bit(R5_WriteError, &dev->flags);
4481 }
1cc03eb9 4482 if (test_bit(R5_MadeGood, &dev->flags)) {
14a75d3e
N		 4483 /* This flag does not apply to '.replacement'
4484 * only to .rdev, so make sure to check that*/
4485 struct md_rdev *rdev2 = rcu_dereference(
4486 conf->disks[i].rdev);
4487 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
b84db560 4488 s->handle_bad_blocks = 1;
14a75d3e 4489 atomic_inc(&rdev2->nr_pending);
b84db560
N		 4490 } else
4491 clear_bit(R5_MadeGood, &dev->flags);
4492 }
977df362
N		 4493 if (test_bit(R5_MadeGoodRepl, &dev->flags)) {
4494 struct md_rdev *rdev2 = rcu_dereference(
4495 conf->disks[i].replacement);
4496 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
4497 s->handle_bad_blocks = 1;
4498 atomic_inc(&rdev2->nr_pending);
4499 } else
4500 clear_bit(R5_MadeGoodRepl, &dev->flags);
4501 }
415e72d0 4502 if (!test_bit(R5_Insync, &dev->flags)) {
16a53ecc
N		 4503 /* The ReadError flag will just be confusing now */
4504 clear_bit(R5_ReadError, &dev->flags);
4505 clear_bit(R5_ReWrite, &dev->flags);
1da177e4 4506 }
415e72d0
N		 4507 if (test_bit(R5_ReadError, &dev->flags))
4508 clear_bit(R5_Insync, &dev->flags);
4509 if (!test_bit(R5_Insync, &dev->flags)) {
cc94015a
N		 4510 if (s->failed < 2)
4511 s->failed_num[s->failed] = i;
4512 s->failed++;
9a3e1101
N		 4513 if (rdev && !test_bit(Faulty, &rdev->flags))
4514 do_recovery = 1;
415e72d0 4515 }
2ded3703
SL		 4516
4517 if (test_bit(R5_InJournal, &dev->flags))
4518 s->injournal++;
1e6d690b
SL		 4519 if (test_bit(R5_InJournal, &dev->flags) && dev->written)
4520 s->just_cached++;
1da177e4 4521 }
9a3e1101
N		 4522 if (test_bit(STRIPE_SYNCING, &sh->state)) {
4523 /* If there is a failed device being replaced,
4524 * we must be recovering.
4525 * else if we are after recovery_cp, we must be syncing
c6d2e084 4526 * else if MD_RECOVERY_REQUESTED is set, we also are syncing.
9a3e1101
N		 4527 * else we can only be replacing
4528 * sync and recovery both need to read all devices, and so
4529 * use the same flag.
4530 */
4531 if (do_recovery ||
c6d2e084 4532 sh->sector >= conf->mddev->recovery_cp ||
4533 test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery)))
9a3e1101
N		 4534 s->syncing = 1;
4535 else
4536 s->replacing = 1;
4537 }
1da177e4 4538 rcu_read_unlock();
cc94015a
N		 4539}
4540
59fc630b 4541static int clear_batch_ready(struct stripe_head *sh)
4542{
b15a9dbd
N		 4543 /* Return '1' if this is a member of batch, or
4544 * '0' if it is a lone stripe or a head which can now be
4545 * handled.
4546 */
59fc630b 4547 struct stripe_head *tmp;
4548 if (!test_and_clear_bit(STRIPE_BATCH_READY, &sh->state))
b15a9dbd 4549 return (sh->batch_head && sh->batch_head != sh);
59fc630b 4550 spin_lock(&sh->stripe_lock);
4551 if (!sh->batch_head) {
4552 spin_unlock(&sh->stripe_lock);
4553 return 0;
4554 }
4555
4556 /*
4557 * this stripe could be added to a batch list before we check
4558 * BATCH_READY, skips it
4559 */
4560 if (sh->batch_head != sh) {
4561 spin_unlock(&sh->stripe_lock);
4562 return 1;
4563 }
4564 spin_lock(&sh->batch_lock);
4565 list_for_each_entry(tmp, &sh->batch_list, batch_list)
4566 clear_bit(STRIPE_BATCH_READY, &tmp->state);
4567 spin_unlock(&sh->batch_lock);
4568 spin_unlock(&sh->stripe_lock);
4569
4570 /*
4571 * BATCH_READY is cleared, no new stripes can be added.
4572 * batch_list can be accessed without lock
4573 */
4574 return 0;
4575}
4576
3960ce79
N		 4577static void break_stripe_batch_list(struct stripe_head *head_sh,
4578 unsigned long handle_flags)
72ac7330 4579{
4e3d62ff 4580 struct stripe_head *sh, *next;
72ac7330 4581 int i;
fb642b92 4582 int do_wakeup = 0;
72ac7330 4583
bb27051f
N		 4584 list_for_each_entry_safe(sh, next, &head_sh->batch_list, batch_list) {
4585
72ac7330 4586 list_del_init(&sh->batch_list);
4587
fb3229d5 4588 WARN_ONCE(sh->state & ((1 << STRIPE_ACTIVE) |
1b956f7a
N		 4589 (1 << STRIPE_SYNCING) |
4590 (1 << STRIPE_REPLACED) |
1b956f7a
N		 4591 (1 << STRIPE_DELAYED) |
4592 (1 << STRIPE_BIT_DELAY) |
4593 (1 << STRIPE_FULL_WRITE) |
4594 (1 << STRIPE_BIOFILL_RUN) |
4595 (1 << STRIPE_COMPUTE_RUN) |
4596 (1 << STRIPE_OPS_REQ_PENDING) |
4597 (1 << STRIPE_DISCARD) |
4598 (1 << STRIPE_BATCH_READY) |
4599 (1 << STRIPE_BATCH_ERR) |
fb3229d5
SL		 4600 (1 << STRIPE_BITMAP_PENDING)),
4601 "stripe state: %lx\n", sh->state);
4602 WARN_ONCE(head_sh->state & ((1 << STRIPE_DISCARD) |
4603 (1 << STRIPE_REPLACED)),
4604 "head stripe state: %lx\n", head_sh->state);
1b956f7a
N		 4605
4606 set_mask_bits(&sh->state, ~(STRIPE_EXPAND_SYNC_FLAGS |
550da24f 4607 (1 << STRIPE_PREREAD_ACTIVE) |
184a09eb
DY		 4608 (1 << STRIPE_DEGRADED) |
4609 (1 << STRIPE_ON_UNPLUG_LIST)),
1b956f7a
N		 4610 head_sh->state & (1 << STRIPE_INSYNC));
4611
72ac7330 4612 sh->check_state = head_sh->check_state;
4613 sh->reconstruct_state = head_sh->reconstruct_state;
fb642b92
N		 4614 for (i = 0; i < sh->disks; i++) {
4615 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
4616 do_wakeup = 1;
72ac7330 4617 sh->dev[i].flags = head_sh->dev[i].flags &
4618 (~((1 << R5_WriteError) | (1 << R5_Overlap)));
fb642b92 4619 }
72ac7330 4620 spin_lock_irq(&sh->stripe_lock);
4621 sh->batch_head = NULL;
4622 spin_unlock_irq(&sh->stripe_lock);
3960ce79
N		 4623 if (handle_flags == 0 ||
4624 sh->state & handle_flags)
4625 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 4626 raid5_release_stripe(sh);
72ac7330 4627 }
fb642b92
N		 4628 spin_lock_irq(&head_sh->stripe_lock);
4629 head_sh->batch_head = NULL;
4630 spin_unlock_irq(&head_sh->stripe_lock);
4631 for (i = 0; i < head_sh->disks; i++)
4632 if (test_and_clear_bit(R5_Overlap, &head_sh->dev[i].flags))
4633 do_wakeup = 1;
3960ce79
N		 4634 if (head_sh->state & handle_flags)
4635 set_bit(STRIPE_HANDLE, &head_sh->state);
fb642b92
N		 4636
4637 if (do_wakeup)
4638 wake_up(&head_sh->raid_conf->wait_for_overlap);
72ac7330 4639}
4640
cc94015a
N		 4641static void handle_stripe(struct stripe_head *sh)
4642{
4643 struct stripe_head_state s;
d1688a6d 4644 struct r5conf *conf = sh->raid_conf;
3687c061 4645 int i;
84789554
N		 4646 int prexor;
4647 int disks = sh->disks;
474af965 4648 struct r5dev *pdev, *qdev;
cc94015a
N		 4649
4650 clear_bit(STRIPE_HANDLE, &sh->state);
257a4b42 4651 if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) {
cc94015a
N		 4652 /* already being handled, ensure it gets handled
4653 * again when current action finishes */
4654 set_bit(STRIPE_HANDLE, &sh->state);
4655 return;
4656 }
4657
59fc630b 4658 if (clear_batch_ready(sh) ) {
4659 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
4660 return;
4661 }
4662
4e3d62ff 4663 if (test_and_clear_bit(STRIPE_BATCH_ERR, &sh->state))
3960ce79 4664 break_stripe_batch_list(sh, 0);
72ac7330 4665
dabc4ec6 4666 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) && !sh->batch_head) {
f8dfcffd 4667 spin_lock(&sh->stripe_lock);
5ddf0440
SL		 4668 /*
4669 * Cannot process 'sync' concurrently with 'discard'.
4670 * Flush data in r5cache before 'sync'.
4671 */
4672 if (!test_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state) &&
4673 !test_bit(STRIPE_R5C_FULL_STRIPE, &sh->state) &&
4674 !test_bit(STRIPE_DISCARD, &sh->state) &&
f8dfcffd
N		 4675 test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
4676 set_bit(STRIPE_SYNCING, &sh->state);
4677 clear_bit(STRIPE_INSYNC, &sh->state);
f94c0b66 4678 clear_bit(STRIPE_REPLACED, &sh->state);
f8dfcffd
N		 4679 }
4680 spin_unlock(&sh->stripe_lock);
cc94015a
N		 4681 }
4682 clear_bit(STRIPE_DELAYED, &sh->state);
4683
4684 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
4685 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
4686 (unsigned long long)sh->sector, sh->state,
4687 atomic_read(&sh->count), sh->pd_idx, sh->qd_idx,
4688 sh->check_state, sh->reconstruct_state);
3687c061 4689
acfe726b 4690 analyse_stripe(sh, &s);
c5a31000 4691
b70abcb2
SL		 4692 if (test_bit(STRIPE_LOG_TRAPPED, &sh->state))
4693 goto finish;
4694
16d997b7
N		 4695 if (s.handle_bad_blocks ||
4696 test_bit(MD_SB_CHANGE_PENDING, &conf->mddev->sb_flags)) {
bc2607f3
N		 4697 set_bit(STRIPE_HANDLE, &sh->state);
4698 goto finish;
4699 }
4700
474af965
N		 4701 if (unlikely(s.blocked_rdev)) {
4702 if (s.syncing || s.expanding || s.expanded ||
9a3e1101 4703 s.replacing || s.to_write || s.written) {
474af965
N		 4704 set_bit(STRIPE_HANDLE, &sh->state);
4705 goto finish;
4706 }
4707 /* There is nothing for the blocked_rdev to block */
4708 rdev_dec_pending(s.blocked_rdev, conf->mddev);
4709 s.blocked_rdev = NULL;
4710 }
4711
4712 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
4713 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
4714 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
4715 }
4716
4717 pr_debug("locked=%d uptodate=%d to_read=%d"
4718 " to_write=%d failed=%d failed_num=%d,%d\n",
4719 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
4720 s.failed_num[0], s.failed_num[1]);
70d466f7
SL		 4721 /*
4722 * check if the array has lost more than max_degraded devices and,
474af965 4723 * if so, some requests might need to be failed.
70d466f7
SL		 4724 *
4725 * When journal device failed (log_failed), we will only process
4726 * the stripe if there is data need write to raid disks
474af965 4727 */
70d466f7
SL		 4728 if (s.failed > conf->max_degraded ||
4729 (s.log_failed && s.injournal == 0)) {
9a3f530f
N		 4730 sh->check_state = 0;
4731 sh->reconstruct_state = 0;
626f2092 4732 break_stripe_batch_list(sh, 0);
9a3f530f 4733 if (s.to_read+s.to_write+s.written)
bd83d0a2 4734 handle_failed_stripe(conf, sh, &s, disks);
9a3e1101 4735 if (s.syncing + s.replacing)
9a3f530f
N		 4736 handle_failed_sync(conf, sh, &s);
4737 }
474af965 4738
84789554
N		 4739 /* Now we check to see if any write operations have recently
4740 * completed
4741 */
4742 prexor = 0;
4743 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
4744 prexor = 1;
4745 if (sh->reconstruct_state == reconstruct_state_drain_result ||
4746 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
4747 sh->reconstruct_state = reconstruct_state_idle;
4748
4749 /* All the 'written' buffers and the parity block are ready to
4750 * be written back to disk
4751 */
9e444768
SL		 4752 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) &&
4753 !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags));
84789554 4754 BUG_ON(sh->qd_idx >= 0 &&
9e444768
SL		 4755 !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) &&
4756 !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags));
84789554
N		 4757 for (i = disks; i--; ) {
4758 struct r5dev *dev = &sh->dev[i];
4759 if (test_bit(R5_LOCKED, &dev->flags) &&
4760 (i == sh->pd_idx || i == sh->qd_idx ||
1e6d690b
SL		 4761 dev->written || test_bit(R5_InJournal,
4762 &dev->flags))) {
84789554
N		 4763 pr_debug("Writing block %d\n", i);
4764 set_bit(R5_Wantwrite, &dev->flags);
4765 if (prexor)
4766 continue;
9c4bdf69
N		 4767 if (s.failed > 1)
4768 continue;
84789554
N		 4769 if (!test_bit(R5_Insync, &dev->flags) ||
4770 ((i == sh->pd_idx || i == sh->qd_idx) &&
4771 s.failed == 0))
4772 set_bit(STRIPE_INSYNC, &sh->state);
4773 }
4774 }
4775 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4776 s.dec_preread_active = 1;
4777 }
4778
ef5b7c69
N		 4779 /*
4780 * might be able to return some write requests if the parity blocks
4781 * are safe, or on a failed drive
4782 */
4783 pdev = &sh->dev[sh->pd_idx];
4784 s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx)
4785 || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx);
4786 qdev = &sh->dev[sh->qd_idx];
4787 s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx)
4788 || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx)
4789 || conf->level < 6;
4790
4791 if (s.written &&
4792 (s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
4793 && !test_bit(R5_LOCKED, &pdev->flags)
4794 && (test_bit(R5_UPTODATE, &pdev->flags) ||
4795 test_bit(R5_Discard, &pdev->flags))))) &&
4796 (s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
4797 && !test_bit(R5_LOCKED, &qdev->flags)
4798 && (test_bit(R5_UPTODATE, &qdev->flags) ||
4799 test_bit(R5_Discard, &qdev->flags))))))
bd83d0a2 4800 handle_stripe_clean_event(conf, sh, disks);
ef5b7c69 4801
1e6d690b 4802 if (s.just_cached)
bd83d0a2 4803 r5c_handle_cached_data_endio(conf, sh, disks);
ff875738 4804 log_stripe_write_finished(sh);
1e6d690b 4805
ef5b7c69
N		 4806 /* Now we might consider reading some blocks, either to check/generate
4807 * parity, or to satisfy requests
4808 * or to load a block that is being partially written.
4809 */
4810 if (s.to_read || s.non_overwrite
4811 || (conf->level == 6 && s.to_write && s.failed)
4812 || (s.syncing && (s.uptodate + s.compute < disks))
4813 || s.replacing
4814 || s.expanding)
4815 handle_stripe_fill(sh, &s, disks);
4816
2ded3703
SL		 4817 /*
4818 * When the stripe finishes full journal write cycle (write to journal
4819 * and raid disk), this is the clean up procedure so it is ready for
4820 * next operation.
4821 */
4822 r5c_finish_stripe_write_out(conf, sh, &s);
4823
4824 /*
4825 * Now to consider new write requests, cache write back and what else,
4826 * if anything should be read. We do not handle new writes when:
84789554
N		 4827 * 1/ A 'write' operation (copy+xor) is already in flight.
4828 * 2/ A 'check' operation is in flight, as it may clobber the parity
4829 * block.
2ded3703 4830 * 3/ A r5c cache log write is in flight.
84789554 4831 */
2ded3703
SL		 4832
4833 if (!sh->reconstruct_state && !sh->check_state && !sh->log_io) {
4834 if (!r5c_is_writeback(conf->log)) {
4835 if (s.to_write)
4836 handle_stripe_dirtying(conf, sh, &s, disks);
4837 } else { /* write back cache */
4838 int ret = 0;
4839
4840 /* First, try handle writes in caching phase */
4841 if (s.to_write)
4842 ret = r5c_try_caching_write(conf, sh, &s,
4843 disks);
4844 /*
4845 * If caching phase failed: ret == -EAGAIN
4846 * OR
4847 * stripe under reclaim: !caching && injournal
4848 *
4849 * fall back to handle_stripe_dirtying()
4850 */
4851 if (ret == -EAGAIN ||
4852 /* stripe under reclaim: !caching && injournal */
4853 (!test_bit(STRIPE_R5C_CACHING, &sh->state) &&
d7bd398e
SL		 4854 s.injournal > 0)) {
4855 ret = handle_stripe_dirtying(conf, sh, &s,
4856 disks);
4857 if (ret == -EAGAIN)
4858 goto finish;
4859 }
2ded3703
SL		 4860 }
4861 }
84789554
N		 4862
4863 /* maybe we need to check and possibly fix the parity for this stripe
4864 * Any reads will already have been scheduled, so we just see if enough
4865 * data is available. The parity check is held off while parity
4866 * dependent operations are in flight.
4867 */
4868 if (sh->check_state ||
4869 (s.syncing && s.locked == 0 &&
4870 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
4871 !test_bit(STRIPE_INSYNC, &sh->state))) {
4872 if (conf->level == 6)
4873 handle_parity_checks6(conf, sh, &s, disks);
4874 else
4875 handle_parity_checks5(conf, sh, &s, disks);
4876 }
c5a31000 4877
f94c0b66
N		 4878 if ((s.replacing || s.syncing) && s.locked == 0
4879 && !test_bit(STRIPE_COMPUTE_RUN, &sh->state)
4880 && !test_bit(STRIPE_REPLACED, &sh->state)) {
9a3e1101
N		 4881 /* Write out to replacement devices where possible */
4882 for (i = 0; i < conf->raid_disks; i++)
f94c0b66
N		 4883 if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) {
4884 WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags));
9a3e1101
N		 4885 set_bit(R5_WantReplace, &sh->dev[i].flags);
4886 set_bit(R5_LOCKED, &sh->dev[i].flags);
4887 s.locked++;
4888 }
f94c0b66
N		 4889 if (s.replacing)
4890 set_bit(STRIPE_INSYNC, &sh->state);
4891 set_bit(STRIPE_REPLACED, &sh->state);
9a3e1101
N		 4892 }
4893 if ((s.syncing || s.replacing) && s.locked == 0 &&
f94c0b66 4894 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
9a3e1101 4895 test_bit(STRIPE_INSYNC, &sh->state)) {
c5a31000
N		 4896 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
4897 clear_bit(STRIPE_SYNCING, &sh->state);
f8dfcffd
N		 4898 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
4899 wake_up(&conf->wait_for_overlap);
c5a31000
N		 4900 }
4901
4902 /* If the failed drives are just a ReadError, then we might need
4903 * to progress the repair/check process
4904 */
4905 if (s.failed <= conf->max_degraded && !conf->mddev->ro)
4906 for (i = 0; i < s.failed; i++) {
4907 struct r5dev *dev = &sh->dev[s.failed_num[i]];
4908 if (test_bit(R5_ReadError, &dev->flags)
4909 && !test_bit(R5_LOCKED, &dev->flags)
4910 && test_bit(R5_UPTODATE, &dev->flags)
4911 ) {
4912 if (!test_bit(R5_ReWrite, &dev->flags)) {
4913 set_bit(R5_Wantwrite, &dev->flags);
4914 set_bit(R5_ReWrite, &dev->flags);
4915 set_bit(R5_LOCKED, &dev->flags);
4916 s.locked++;
4917 } else {
4918 /* let's read it back */
4919 set_bit(R5_Wantread, &dev->flags);
4920 set_bit(R5_LOCKED, &dev->flags);
4921 s.locked++;
4922 }
4923 }
4924 }
4925
3687c061
N		 4926 /* Finish reconstruct operations initiated by the expansion process */
4927 if (sh->reconstruct_state == reconstruct_state_result) {
4928 struct stripe_head *sh_src
6d036f7d 4929 = raid5_get_active_stripe(conf, sh->sector, 1, 1, 1);
3687c061
N		 4930 if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) {
4931 /* sh cannot be written until sh_src has been read.
4932 * so arrange for sh to be delayed a little
4933 */
4934 set_bit(STRIPE_DELAYED, &sh->state);
4935 set_bit(STRIPE_HANDLE, &sh->state);
4936 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
4937 &sh_src->state))
4938 atomic_inc(&conf->preread_active_stripes);
6d036f7d 4939 raid5_release_stripe(sh_src);
3687c061
N		 4940 goto finish;
4941 }
4942 if (sh_src)
6d036f7d 4943 raid5_release_stripe(sh_src);
3687c061
N		 4944
4945 sh->reconstruct_state = reconstruct_state_idle;
4946 clear_bit(STRIPE_EXPANDING, &sh->state);
4947 for (i = conf->raid_disks; i--; ) {
4948 set_bit(R5_Wantwrite, &sh->dev[i].flags);
4949 set_bit(R5_LOCKED, &sh->dev[i].flags);
4950 s.locked++;
4951 }
4952 }
f416885e 4953
3687c061
N		 4954 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
4955 !sh->reconstruct_state) {
4956 /* Need to write out all blocks after computing parity */
4957 sh->disks = conf->raid_disks;
4958 stripe_set_idx(sh->sector, conf, 0, sh);
4959 schedule_reconstruction(sh, &s, 1, 1);
4960 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
4961 clear_bit(STRIPE_EXPAND_READY, &sh->state);
4962 atomic_dec(&conf->reshape_stripes);
4963 wake_up(&conf->wait_for_overlap);
4964 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
4965 }
4966
4967 if (s.expanding && s.locked == 0 &&
4968 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
4969 handle_stripe_expansion(conf, sh);
16a53ecc 4970
3687c061 4971finish:
6bfe0b49 4972 /* wait for this device to become unblocked */
5f066c63
N		 4973 if (unlikely(s.blocked_rdev)) {
4974 if (conf->mddev->external)
4975 md_wait_for_blocked_rdev(s.blocked_rdev,
4976 conf->mddev);
4977 else
4978 /* Internal metadata will immediately
4979 * be written by raid5d, so we don't
4980 * need to wait here.
4981 */
4982 rdev_dec_pending(s.blocked_rdev,
4983 conf->mddev);
4984 }
6bfe0b49 4985
bc2607f3
N		 4986 if (s.handle_bad_blocks)
4987 for (i = disks; i--; ) {
3cb03002 4988 struct md_rdev *rdev;
bc2607f3
N		 4989 struct r5dev *dev = &sh->dev[i];
4990 if (test_and_clear_bit(R5_WriteError, &dev->flags)) {
4991 /* We own a safe reference to the rdev */
4992 rdev = conf->disks[i].rdev;
4993 if (!rdev_set_badblocks(rdev, sh->sector,
4994 STRIPE_SECTORS, 0))
4995 md_error(conf->mddev, rdev);
4996 rdev_dec_pending(rdev, conf->mddev);
4997 }
b84db560
N		 4998 if (test_and_clear_bit(R5_MadeGood, &dev->flags)) {
4999 rdev = conf->disks[i].rdev;
5000 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 5001 STRIPE_SECTORS, 0);
b84db560
N		 5002 rdev_dec_pending(rdev, conf->mddev);
5003 }
977df362
N		 5004 if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) {
5005 rdev = conf->disks[i].replacement;
dd054fce
N		 5006 if (!rdev)
5007 /* rdev have been moved down */
5008 rdev = conf->disks[i].rdev;
977df362 5009 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 5010 STRIPE_SECTORS, 0);
977df362
N		 5011 rdev_dec_pending(rdev, conf->mddev);
5012 }
bc2607f3
N		 5013 }
5014
6c0069c0
YT		 5015 if (s.ops_request)
5016 raid_run_ops(sh, s.ops_request);
5017
f0e43bcd 5018 ops_run_io(sh, &s);
16a53ecc 5019
c5709ef6 5020 if (s.dec_preread_active) {
729a1866 5021 /* We delay this until after ops_run_io so that if make_request
e9c7469b 5022 * is waiting on a flush, it won't continue until the writes
729a1866
N		 5023 * have actually been submitted.
5024 */
5025 atomic_dec(&conf->preread_active_stripes);
5026 if (atomic_read(&conf->preread_active_stripes) <
5027 IO_THRESHOLD)
5028 md_wakeup_thread(conf->mddev->thread);
5029 }
5030
257a4b42 5031 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
16a53ecc
N		 5032}
5033
d1688a6d 5034static void raid5_activate_delayed(struct r5conf *conf)
16a53ecc
N		 5035{
5036 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
5037 while (!list_empty(&conf->delayed_list)) {
5038 struct list_head *l = conf->delayed_list.next;
5039 struct stripe_head *sh;
5040 sh = list_entry(l, struct stripe_head, lru);
5041 list_del_init(l);
5042 clear_bit(STRIPE_DELAYED, &sh->state);
5043 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
5044 atomic_inc(&conf->preread_active_stripes);
8b3e6cdc 5045 list_add_tail(&sh->lru, &conf->hold_list);
851c30c9 5046 raid5_wakeup_stripe_thread(sh);
16a53ecc 5047 }
482c0834 5048 }
16a53ecc
N		 5049}
5050
566c09c5
SL		 5051static void activate_bit_delay(struct r5conf *conf,
5052 struct list_head *temp_inactive_list)
16a53ecc
N		 5053{
5054 /* device_lock is held */
5055 struct list_head head;
5056 list_add(&head, &conf->bitmap_list);
5057 list_del_init(&conf->bitmap_list);
5058 while (!list_empty(&head)) {
5059 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
566c09c5 5060 int hash;
16a53ecc
N		 5061 list_del_init(&sh->lru);
5062 atomic_inc(&sh->count);
566c09c5
SL		 5063 hash = sh->hash_lock_index;
5064 __release_stripe(conf, sh, &temp_inactive_list[hash]);
16a53ecc
N		 5065 }
5066}
5067
5c675f83 5068static int raid5_congested(struct mddev *mddev, int bits)
f022b2fd 5069{
d1688a6d 5070 struct r5conf *conf = mddev->private;
f022b2fd
N		 5071
5072 /* No difference between reads and writes. Just check
5073 * how busy the stripe_cache is
5074 */
3fa841d7 5075
5423399a 5076 if (test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state))
f022b2fd 5077 return 1;
a39f7afd
SL		 5078
5079 /* Also checks whether there is pressure on r5cache log space */
5080 if (test_bit(R5C_LOG_TIGHT, &conf->cache_state))
5081 return 1;
f022b2fd
N		 5082 if (conf->quiesce)
5083 return 1;
4bda556a 5084 if (atomic_read(&conf->empty_inactive_list_nr))
f022b2fd
N		 5085 return 1;
5086
5087 return 0;
5088}
5089
fd01b88c 5090static int in_chunk_boundary(struct mddev *mddev, struct bio *bio)
f679623f 5091{
3cb5edf4 5092 struct r5conf *conf = mddev->private;
10433d04 5093 sector_t sector = bio->bi_iter.bi_sector;
3cb5edf4 5094 unsigned int chunk_sectors;
aa8b57aa 5095 unsigned int bio_sectors = bio_sectors(bio);
f679623f 5096
10433d04
CH		 5097 WARN_ON_ONCE(bio->bi_partno);
5098
3cb5edf4 5099 chunk_sectors = min(conf->chunk_sectors, conf->prev_chunk_sectors);
f679623f
RBJ		 5100 return chunk_sectors >=
5101 ((sector & (chunk_sectors - 1)) + bio_sectors);
5102}
5103
46031f9a
RBJ		 5104/*
5105 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
5106 * later sampled by raid5d.
5107 */
d1688a6d 5108static void add_bio_to_retry(struct bio *bi,struct r5conf *conf)
46031f9a
RBJ		 5109{
5110 unsigned long flags;
5111
5112 spin_lock_irqsave(&conf->device_lock, flags);
5113
5114 bi->bi_next = conf->retry_read_aligned_list;
5115 conf->retry_read_aligned_list = bi;
5116
5117 spin_unlock_irqrestore(&conf->device_lock, flags);
5118 md_wakeup_thread(conf->mddev->thread);
5119}
5120
0472a42b
N		 5121static struct bio *remove_bio_from_retry(struct r5conf *conf,
5122 unsigned int *offset)
46031f9a
RBJ		 5123{
5124 struct bio *bi;
5125
5126 bi = conf->retry_read_aligned;
5127 if (bi) {
0472a42b 5128 *offset = conf->retry_read_offset;
46031f9a
RBJ		 5129 conf->retry_read_aligned = NULL;
5130 return bi;
5131 }
5132 bi = conf->retry_read_aligned_list;
5133 if(bi) {
387bb173 5134 conf->retry_read_aligned_list = bi->bi_next;
46031f9a 5135 bi->bi_next = NULL;
0472a42b 5136 *offset = 0;
46031f9a
RBJ		 5137 }
5138
5139 return bi;
5140}
5141
f679623f
RBJ		 5142/*
5143 * The "raid5_align_endio" should check if the read succeeded and if it
5144 * did, call bio_endio on the original bio (having bio_put the new bio
5145 * first).
5146 * If the read failed..
5147 */
4246a0b6 5148static void raid5_align_endio(struct bio *bi)
f679623f
RBJ		 5149{
5150 struct bio* raid_bi = bi->bi_private;
fd01b88c 5151 struct mddev *mddev;
d1688a6d 5152 struct r5conf *conf;
3cb03002 5153 struct md_rdev *rdev;
4e4cbee9 5154 blk_status_t error = bi->bi_status;
46031f9a 5155
f679623f 5156 bio_put(bi);
46031f9a 5157
46031f9a
RBJ		 5158 rdev = (void*)raid_bi->bi_next;
5159 raid_bi->bi_next = NULL;
2b7f2228
N		 5160 mddev = rdev->mddev;
5161 conf = mddev->private;
46031f9a
RBJ		 5162
5163 rdev_dec_pending(rdev, conf->mddev);
5164
9b81c842 5165 if (!error) {
4246a0b6 5166 bio_endio(raid_bi);
46031f9a 5167 if (atomic_dec_and_test(&conf->active_aligned_reads))
b1b46486 5168 wake_up(&conf->wait_for_quiescent);
6712ecf8 5169 return;
46031f9a
RBJ		 5170 }
5171
45b4233c 5172 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
46031f9a
RBJ		 5173
5174 add_bio_to_retry(raid_bi, conf);
f679623f
RBJ		 5175}
5176
7ef6b12a 5177static int raid5_read_one_chunk(struct mddev *mddev, struct bio *raid_bio)
f679623f 5178{
d1688a6d 5179 struct r5conf *conf = mddev->private;
8553fe7e 5180 int dd_idx;
f679623f 5181 struct bio* align_bi;
3cb03002 5182 struct md_rdev *rdev;
671488cc 5183 sector_t end_sector;
f679623f
RBJ		 5184
5185 if (!in_chunk_boundary(mddev, raid_bio)) {
7ef6b12a 5186 pr_debug("%s: non aligned\n", __func__);
f679623f
RBJ		 5187 return 0;
5188 }
5189 /*
d7a10308 5190 * use bio_clone_fast to make a copy of the bio
f679623f 5191 */
d7a10308 5192 align_bi = bio_clone_fast(raid_bio, GFP_NOIO, mddev->bio_set);
f679623f
RBJ		 5193 if (!align_bi)
5194 return 0;
5195 /*
5196 * set bi_end_io to a new function, and set bi_private to the
5197 * original bio.
5198 */
5199 align_bi->bi_end_io = raid5_align_endio;
5200 align_bi->bi_private = raid_bio;
5201 /*
5202 * compute position
5203 */
4f024f37
KO		 5204 align_bi->bi_iter.bi_sector =
5205 raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector,
5206 0, &dd_idx, NULL);
f679623f 5207
f73a1c7d 5208 end_sector = bio_end_sector(align_bi);
f679623f 5209 rcu_read_lock();
671488cc
N		 5210 rdev = rcu_dereference(conf->disks[dd_idx].replacement);
5211 if (!rdev || test_bit(Faulty, &rdev->flags) ||
5212 rdev->recovery_offset < end_sector) {
5213 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
5214 if (rdev &&
5215 (test_bit(Faulty, &rdev->flags) ||
5216 !(test_bit(In_sync, &rdev->flags) ||
5217 rdev->recovery_offset >= end_sector)))
5218 rdev = NULL;
5219 }
03b047f4
SL		 5220
5221 if (r5c_big_stripe_cached(conf, align_bi->bi_iter.bi_sector)) {
5222 rcu_read_unlock();
5223 bio_put(align_bi);
5224 return 0;
5225 }
5226
671488cc 5227 if (rdev) {
31c176ec
N		 5228 sector_t first_bad;
5229 int bad_sectors;
5230
f679623f
RBJ		 5231 atomic_inc(&rdev->nr_pending);
5232 rcu_read_unlock();
46031f9a 5233 raid_bio->bi_next = (void*)rdev;
74d46992 5234 bio_set_dev(align_bi, rdev->bdev);
b7c44ed9 5235 bio_clear_flag(align_bi, BIO_SEG_VALID);
46031f9a 5236
7140aafc 5237 if (is_badblock(rdev, align_bi->bi_iter.bi_sector,
4f024f37 5238 bio_sectors(align_bi),
31c176ec 5239 &first_bad, &bad_sectors)) {
387bb173
NB		 5240 bio_put(align_bi);
5241 rdev_dec_pending(rdev, mddev);
5242 return 0;
5243 }
5244
6c0544e2 5245 /* No reshape active, so we can trust rdev->data_offset */
4f024f37 5246 align_bi->bi_iter.bi_sector += rdev->data_offset;
6c0544e2 5247
46031f9a 5248 spin_lock_irq(&conf->device_lock);
b1b46486 5249 wait_event_lock_irq(conf->wait_for_quiescent,
46031f9a 5250 conf->quiesce == 0,
eed8c02e 5251 conf->device_lock);
46031f9a
RBJ		 5252 atomic_inc(&conf->active_aligned_reads);
5253 spin_unlock_irq(&conf->device_lock);
5254
e3620a3a 5255 if (mddev->gendisk)
74d46992 5256 trace_block_bio_remap(align_bi->bi_disk->queue,
e3620a3a 5257 align_bi, disk_devt(mddev->gendisk),
4f024f37 5258 raid_bio->bi_iter.bi_sector);
f679623f
RBJ		 5259 generic_make_request(align_bi);
5260 return 1;
5261 } else {
5262 rcu_read_unlock();
46031f9a 5263 bio_put(align_bi);
f679623f
RBJ		 5264 return 0;
5265 }
5266}
5267
7ef6b12a
ML		 5268static struct bio *chunk_aligned_read(struct mddev *mddev, struct bio *raid_bio)
5269{
5270 struct bio *split;
dd7a8f5d
N		 5271 sector_t sector = raid_bio->bi_iter.bi_sector;
5272 unsigned chunk_sects = mddev->chunk_sectors;
5273 unsigned sectors = chunk_sects - (sector & (chunk_sects-1));
7ef6b12a 5274
dd7a8f5d
N		 5275 if (sectors < bio_sectors(raid_bio)) {
5276 struct r5conf *conf = mddev->private;
5277 split = bio_split(raid_bio, sectors, GFP_NOIO, conf->bio_split);
5278 bio_chain(split, raid_bio);
5279 generic_make_request(raid_bio);
5280 raid_bio = split;
5281 }
7ef6b12a 5282
dd7a8f5d
N		 5283 if (!raid5_read_one_chunk(mddev, raid_bio))
5284 return raid_bio;
7ef6b12a
ML		 5285
5286 return NULL;
5287}
5288
8b3e6cdc
DW		 5289/* __get_priority_stripe - get the next stripe to process
5290 *
5291 * Full stripe writes are allowed to pass preread active stripes up until
5292 * the bypass_threshold is exceeded. In general the bypass_count
5293 * increments when the handle_list is handled before the hold_list; however, it
5294 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
5295 * stripe with in flight i/o. The bypass_count will be reset when the
5296 * head of the hold_list has changed, i.e. the head was promoted to the
5297 * handle_list.
5298 */
851c30c9 5299static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group)
8b3e6cdc 5300{
535ae4eb 5301 struct stripe_head *sh, *tmp;
851c30c9 5302 struct list_head *handle_list = NULL;
535ae4eb 5303 struct r5worker_group *wg;
70d466f7
SL		 5304 bool second_try = !r5c_is_writeback(conf->log) &&
5305 !r5l_log_disk_error(conf);
5306 bool try_loprio = test_bit(R5C_LOG_TIGHT, &conf->cache_state) ||
5307 r5l_log_disk_error(conf);
851c30c9 5308
535ae4eb
SL		 5309again:
5310 wg = NULL;
5311 sh = NULL;
851c30c9 5312 if (conf->worker_cnt_per_group == 0) {
535ae4eb
SL		 5313 handle_list = try_loprio ? &conf->loprio_list :
5314 &conf->handle_list;
851c30c9 5315 } else if (group != ANY_GROUP) {
535ae4eb
SL		 5316 handle_list = try_loprio ? &conf->worker_groups[group].loprio_list :
5317 &conf->worker_groups[group].handle_list;
bfc90cb0 5318 wg = &conf->worker_groups[group];
851c30c9
SL		 5319 } else {
5320 int i;
5321 for (i = 0; i < conf->group_cnt; i++) {
535ae4eb
SL		 5322 handle_list = try_loprio ? &conf->worker_groups[i].loprio_list :
5323 &conf->worker_groups[i].handle_list;
bfc90cb0 5324 wg = &conf->worker_groups[i];
851c30c9
SL		 5325 if (!list_empty(handle_list))
5326 break;
5327 }
5328 }
8b3e6cdc
DW		 5329
5330 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
5331 __func__,
851c30c9 5332 list_empty(handle_list) ? "empty" : "busy",
8b3e6cdc
DW		 5333 list_empty(&conf->hold_list) ? "empty" : "busy",
5334 atomic_read(&conf->pending_full_writes), conf->bypass_count);
5335
851c30c9
SL		 5336 if (!list_empty(handle_list)) {
5337 sh = list_entry(handle_list->next, typeof(*sh), lru);
8b3e6cdc
DW		 5338
5339 if (list_empty(&conf->hold_list))
5340 conf->bypass_count = 0;
5341 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
5342 if (conf->hold_list.next == conf->last_hold)
5343 conf->bypass_count++;
5344 else {
5345 conf->last_hold = conf->hold_list.next;
5346 conf->bypass_count -= conf->bypass_threshold;
5347 if (conf->bypass_count < 0)
5348 conf->bypass_count = 0;
5349 }
5350 }
5351 } else if (!list_empty(&conf->hold_list) &&
5352 ((conf->bypass_threshold &&
5353 conf->bypass_count > conf->bypass_threshold) ||
5354 atomic_read(&conf->pending_full_writes) == 0)) {
851c30c9
SL		 5355
5356 list_for_each_entry(tmp, &conf->hold_list, lru) {
5357 if (conf->worker_cnt_per_group == 0 ||
5358 group == ANY_GROUP ||
5359 !cpu_online(tmp->cpu) ||
5360 cpu_to_group(tmp->cpu) == group) {
5361 sh = tmp;
5362 break;
5363 }
5364 }
5365
5366 if (sh) {
5367 conf->bypass_count -= conf->bypass_threshold;
5368 if (conf->bypass_count < 0)
5369 conf->bypass_count = 0;
5370 }
bfc90cb0 5371 wg = NULL;
851c30c9
SL		 5372 }
5373
535ae4eb
SL		 5374 if (!sh) {
5375 if (second_try)
5376 return NULL;
5377 second_try = true;
5378 try_loprio = !try_loprio;
5379 goto again;
5380 }
8b3e6cdc 5381
bfc90cb0
SL		 5382 if (wg) {
5383 wg->stripes_cnt--;
5384 sh->group = NULL;
5385 }
8b3e6cdc 5386 list_del_init(&sh->lru);
c7a6d35e 5387 BUG_ON(atomic_inc_return(&sh->count) != 1);
8b3e6cdc
DW		 5388 return sh;
5389}
f679623f 5390
8811b596
SL		 5391struct raid5_plug_cb {
5392 struct blk_plug_cb cb;
5393 struct list_head list;
566c09c5 5394 struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
8811b596
SL		 5395};
5396
5397static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule)
5398{
5399 struct raid5_plug_cb *cb = container_of(
5400 blk_cb, struct raid5_plug_cb, cb);
5401 struct stripe_head *sh;
5402 struct mddev *mddev = cb->cb.data;
5403 struct r5conf *conf = mddev->private;
a9add5d9 5404 int cnt = 0;
566c09c5 5405 int hash;
8811b596
SL		 5406
5407 if (cb->list.next && !list_empty(&cb->list)) {
5408 spin_lock_irq(&conf->device_lock);
5409 while (!list_empty(&cb->list)) {
5410 sh = list_first_entry(&cb->list, struct stripe_head, lru);
5411 list_del_init(&sh->lru);
5412 /*
5413 * avoid race release_stripe_plug() sees
5414 * STRIPE_ON_UNPLUG_LIST clear but the stripe
5415 * is still in our list
5416 */
4e857c58 5417 smp_mb__before_atomic();
8811b596 5418 clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state);
773ca82f
SL		 5419 /*
5420 * STRIPE_ON_RELEASE_LIST could be set here. In that
5421 * case, the count is always > 1 here
5422 */
566c09c5
SL		 5423 hash = sh->hash_lock_index;
5424 __release_stripe(conf, sh, &cb->temp_inactive_list[hash]);
a9add5d9 5425 cnt++;
8811b596
SL		 5426 }
5427 spin_unlock_irq(&conf->device_lock);
5428 }
566c09c5
SL		 5429 release_inactive_stripe_list(conf, cb->temp_inactive_list,
5430 NR_STRIPE_HASH_LOCKS);
e3620a3a
JB		 5431 if (mddev->queue)
5432 trace_block_unplug(mddev->queue, cnt, !from_schedule);
8811b596
SL		 5433 kfree(cb);
5434}
5435
5436static void release_stripe_plug(struct mddev *mddev,
5437 struct stripe_head *sh)
5438{
5439 struct blk_plug_cb *blk_cb = blk_check_plugged(
5440 raid5_unplug, mddev,
5441 sizeof(struct raid5_plug_cb));
5442 struct raid5_plug_cb *cb;
5443
5444 if (!blk_cb) {
6d036f7d 5445 raid5_release_stripe(sh);
8811b596
SL		 5446 return;
5447 }
5448
5449 cb = container_of(blk_cb, struct raid5_plug_cb, cb);
5450
566c09c5
SL		 5451 if (cb->list.next == NULL) {
5452 int i;
8811b596 5453 INIT_LIST_HEAD(&cb->list);
566c09c5
SL		 5454 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5455 INIT_LIST_HEAD(cb->temp_inactive_list + i);
5456 }
8811b596
SL		 5457
5458 if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state))
5459 list_add_tail(&sh->lru, &cb->list);
5460 else
6d036f7d 5461 raid5_release_stripe(sh);
8811b596
SL		 5462}
5463
620125f2
SL		 5464static void make_discard_request(struct mddev *mddev, struct bio *bi)
5465{
5466 struct r5conf *conf = mddev->private;
5467 sector_t logical_sector, last_sector;
5468 struct stripe_head *sh;
620125f2
SL		 5469 int stripe_sectors;
5470
5471 if (mddev->reshape_position != MaxSector)
5472 /* Skip discard while reshape is happening */
5473 return;
5474
4f024f37
KO		 5475 logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1);
5476 last_sector = bi->bi_iter.bi_sector + (bi->bi_iter.bi_size>>9);
620125f2
SL		 5477
5478 bi->bi_next = NULL;
620125f2
SL		 5479
5480 stripe_sectors = conf->chunk_sectors *
5481 (conf->raid_disks - conf->max_degraded);
5482 logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector,
5483 stripe_sectors);
5484 sector_div(last_sector, stripe_sectors);
5485
5486 logical_sector *= conf->chunk_sectors;
5487 last_sector *= conf->chunk_sectors;
5488
5489 for (; logical_sector < last_sector;
5490 logical_sector += STRIPE_SECTORS) {
5491 DEFINE_WAIT(w);
5492 int d;
5493 again:
6d036f7d 5494 sh = raid5_get_active_stripe(conf, logical_sector, 0, 0, 0);
620125f2
SL		 5495 prepare_to_wait(&conf->wait_for_overlap, &w,
5496 TASK_UNINTERRUPTIBLE);
f8dfcffd
N		 5497 set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
5498 if (test_bit(STRIPE_SYNCING, &sh->state)) {
6d036f7d 5499 raid5_release_stripe(sh);
f8dfcffd
N		 5500 schedule();
5501 goto again;
5502 }
5503 clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
620125f2
SL		 5504 spin_lock_irq(&sh->stripe_lock);
5505 for (d = 0; d < conf->raid_disks; d++) {
5506 if (d == sh->pd_idx || d == sh->qd_idx)
5507 continue;
5508 if (sh->dev[d].towrite || sh->dev[d].toread) {
5509 set_bit(R5_Overlap, &sh->dev[d].flags);
5510 spin_unlock_irq(&sh->stripe_lock);
6d036f7d 5511 raid5_release_stripe(sh);
620125f2
SL		 5512 schedule();
5513 goto again;
5514 }
5515 }
f8dfcffd 5516 set_bit(STRIPE_DISCARD, &sh->state);
620125f2 5517 finish_wait(&conf->wait_for_overlap, &w);
7a87f434 5518 sh->overwrite_disks = 0;
620125f2
SL		 5519 for (d = 0; d < conf->raid_disks; d++) {
5520 if (d == sh->pd_idx || d == sh->qd_idx)
5521 continue;
5522 sh->dev[d].towrite = bi;
5523 set_bit(R5_OVERWRITE, &sh->dev[d].flags);
016c76ac 5524 bio_inc_remaining(bi);
49728050 5525 md_write_inc(mddev, bi);
7a87f434 5526 sh->overwrite_disks++;
620125f2
SL		 5527 }
5528 spin_unlock_irq(&sh->stripe_lock);
5529 if (conf->mddev->bitmap) {
5530 for (d = 0;
5531 d < conf->raid_disks - conf->max_degraded;
5532 d++)
5533 bitmap_startwrite(mddev->bitmap,
5534 sh->sector,
5535 STRIPE_SECTORS,
5536 0);
5537 sh->bm_seq = conf->seq_flush + 1;
5538 set_bit(STRIPE_BIT_DELAY, &sh->state);
5539 }
5540
5541 set_bit(STRIPE_HANDLE, &sh->state);
5542 clear_bit(STRIPE_DELAYED, &sh->state);
5543 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
5544 atomic_inc(&conf->preread_active_stripes);
5545 release_stripe_plug(mddev, sh);
5546 }
5547
016c76ac 5548 bio_endio(bi);
620125f2
SL		 5549}
5550
cc27b0c7 5551static bool raid5_make_request(struct mddev *mddev, struct bio * bi)
1da177e4 5552{
d1688a6d 5553 struct r5conf *conf = mddev->private;
911d4ee8 5554 int dd_idx;
1da177e4
LT		 5555 sector_t new_sector;
5556 sector_t logical_sector, last_sector;
5557 struct stripe_head *sh;
a362357b 5558 const int rw = bio_data_dir(bi);
27c0f68f
SL		 5559 DEFINE_WAIT(w);
5560 bool do_prepare;
3bddb7f8 5561 bool do_flush = false;
1da177e4 5562
1eff9d32 5563 if (unlikely(bi->bi_opf & REQ_PREFLUSH)) {
828cbe98
SL		 5564 int ret = r5l_handle_flush_request(conf->log, bi);
5565
5566 if (ret == 0)
cc27b0c7 5567 return true;
828cbe98
SL		 5568 if (ret == -ENODEV) {
5569 md_flush_request(mddev, bi);
cc27b0c7 5570 return true;
828cbe98
SL		 5571 }
5572 /* ret == -EAGAIN, fallback */
3bddb7f8
SL		 5573 /*
5574 * if r5l_handle_flush_request() didn't clear REQ_PREFLUSH,
5575 * we need to flush journal device
5576 */
5577 do_flush = bi->bi_opf & REQ_PREFLUSH;
e5dcdd80
N		 5578 }
5579
cc27b0c7
N		 5580 if (!md_write_start(mddev, bi))
5581 return false;
9ffc8f7c
EM		 5582 /*
5583 * If array is degraded, better not do chunk aligned read because
5584 * later we might have to read it again in order to reconstruct
5585 * data on failed drives.
5586 */
5587 if (rw == READ && mddev->degraded == 0 &&
7ef6b12a
ML		 5588 mddev->reshape_position == MaxSector) {
5589 bi = chunk_aligned_read(mddev, bi);
5590 if (!bi)
cc27b0c7 5591 return true;
7ef6b12a 5592 }
52488615 5593
796a5cf0 5594 if (unlikely(bio_op(bi) == REQ_OP_DISCARD)) {
620125f2 5595 make_discard_request(mddev, bi);
cc27b0c7
N		 5596 md_write_end(mddev);
5597 return true;
620125f2
SL		 5598 }
5599
4f024f37 5600 logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1);
f73a1c7d 5601 last_sector = bio_end_sector(bi);
1da177e4 5602 bi->bi_next = NULL;
06d91a5f 5603
27c0f68f 5604 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
1da177e4 5605 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
b5663ba4 5606 int previous;
c46501b2 5607 int seq;
b578d55f 5608
27c0f68f 5609 do_prepare = false;
7ecaa1e6 5610 retry:
c46501b2 5611 seq = read_seqcount_begin(&conf->gen_lock);
b5663ba4 5612 previous = 0;
27c0f68f
SL		 5613 if (do_prepare)
5614 prepare_to_wait(&conf->wait_for_overlap, &w,
5615 TASK_UNINTERRUPTIBLE);
b0f9ec04 5616 if (unlikely(conf->reshape_progress != MaxSector)) {
fef9c61f 5617 /* spinlock is needed as reshape_progress may be
df8e7f76
N		 5618 * 64bit on a 32bit platform, and so it might be
5619 * possible to see a half-updated value
aeb878b0 5620 * Of course reshape_progress could change after
df8e7f76
N		 5621 * the lock is dropped, so once we get a reference
5622 * to the stripe that we think it is, we will have
5623 * to check again.
5624 */
7ecaa1e6 5625 spin_lock_irq(&conf->device_lock);
2c810cdd 5626 if (mddev->reshape_backwards
fef9c61f
N		 5627 ? logical_sector < conf->reshape_progress
5628 : logical_sector >= conf->reshape_progress) {
b5663ba4
N		 5629 previous = 1;
5630 } else {
2c810cdd 5631 if (mddev->reshape_backwards
fef9c61f
N		 5632 ? logical_sector < conf->reshape_safe
5633 : logical_sector >= conf->reshape_safe) {
b578d55f
N		 5634 spin_unlock_irq(&conf->device_lock);
5635 schedule();
27c0f68f 5636 do_prepare = true;
b578d55f
N		 5637 goto retry;
5638 }
5639 }
7ecaa1e6
N		 5640 spin_unlock_irq(&conf->device_lock);
5641 }
16a53ecc 5642
112bf897
N		 5643 new_sector = raid5_compute_sector(conf, logical_sector,
5644 previous,
911d4ee8 5645 &dd_idx, NULL);
849674e4 5646 pr_debug("raid456: raid5_make_request, sector %llu logical %llu\n",
c46501b2 5647 (unsigned long long)new_sector,
1da177e4
LT		 5648 (unsigned long long)logical_sector);
5649
6d036f7d 5650 sh = raid5_get_active_stripe(conf, new_sector, previous,
1eff9d32 5651 (bi->bi_opf & REQ_RAHEAD), 0);
1da177e4 5652 if (sh) {
b0f9ec04 5653 if (unlikely(previous)) {
7ecaa1e6 5654 /* expansion might have moved on while waiting for a
df8e7f76
N		 5655 * stripe, so we must do the range check again.
5656 * Expansion could still move past after this
5657 * test, but as we are holding a reference to
5658 * 'sh', we know that if that happens,
5659 * STRIPE_EXPANDING will get set and the expansion
5660 * won't proceed until we finish with the stripe.
7ecaa1e6
N		 5661 */
5662 int must_retry = 0;
5663 spin_lock_irq(&conf->device_lock);
2c810cdd 5664 if (mddev->reshape_backwards
b0f9ec04
N		 5665 ? logical_sector >= conf->reshape_progress
5666 : logical_sector < conf->reshape_progress)
7ecaa1e6
N		 5667 /* mismatch, need to try again */
5668 must_retry = 1;
5669 spin_unlock_irq(&conf->device_lock);
5670 if (must_retry) {
6d036f7d 5671 raid5_release_stripe(sh);
7a3ab908 5672 schedule();
27c0f68f 5673 do_prepare = true;
7ecaa1e6
N		 5674 goto retry;
5675 }
5676 }
c46501b2
N		 5677 if (read_seqcount_retry(&conf->gen_lock, seq)) {
5678 /* Might have got the wrong stripe_head
5679 * by accident
5680 */
6d036f7d 5681 raid5_release_stripe(sh);
c46501b2
N		 5682 goto retry;
5683 }
e62e58a5 5684
ffd96e35 5685 if (rw == WRITE &&
a5c308d4 5686 logical_sector >= mddev->suspend_lo &&
e464eafd 5687 logical_sector < mddev->suspend_hi) {
6d036f7d 5688 raid5_release_stripe(sh);
e62e58a5
N		 5689 /* As the suspend_* range is controlled by
5690 * userspace, we want an interruptible
5691 * wait.
5692 */
e62e58a5
N		 5693 prepare_to_wait(&conf->wait_for_overlap,
5694 &w, TASK_INTERRUPTIBLE);
5695 if (logical_sector >= mddev->suspend_lo &&
27c0f68f 5696 logical_sector < mddev->suspend_hi) {
f9c79bc0
MP		 5697 sigset_t full, old;
5698 sigfillset(&full);
5699 sigprocmask(SIG_BLOCK, &full, &old);
e62e58a5 5700 schedule();
f9c79bc0 5701 sigprocmask(SIG_SETMASK, &old, NULL);
27c0f68f
SL		 5702 do_prepare = true;
5703 }
e464eafd
N		 5704 goto retry;
5705 }
7ecaa1e6
N		 5706
5707 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
da41ba65 5708 !add_stripe_bio(sh, bi, dd_idx, rw, previous)) {
7ecaa1e6
N		 5709 /* Stripe is busy expanding or
5710 * add failed due to overlap. Flush everything
1da177e4
LT		 5711 * and wait a while
5712 */
482c0834 5713 md_wakeup_thread(mddev->thread);
6d036f7d 5714 raid5_release_stripe(sh);
1da177e4 5715 schedule();
27c0f68f 5716 do_prepare = true;
1da177e4
LT		 5717 goto retry;
5718 }
3bddb7f8
SL		 5719 if (do_flush) {
5720 set_bit(STRIPE_R5C_PREFLUSH, &sh->state);
5721 /* we only need flush for one stripe */
5722 do_flush = false;
5723 }
5724
6ed3003c
N		 5725 set_bit(STRIPE_HANDLE, &sh->state);
5726 clear_bit(STRIPE_DELAYED, &sh->state);
59fc630b 5727 if ((!sh->batch_head || sh == sh->batch_head) &&
1eff9d32 5728 (bi->bi_opf & REQ_SYNC) &&
729a1866
N		 5729 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
5730 atomic_inc(&conf->preread_active_stripes);
8811b596 5731 release_stripe_plug(mddev, sh);
1da177e4
LT		 5732 } else {
5733 /* cannot get stripe for read-ahead, just give-up */
4e4cbee9 5734 bi->bi_status = BLK_STS_IOERR;
1da177e4
LT		 5735 break;
5736 }
1da177e4 5737 }
27c0f68f 5738 finish_wait(&conf->wait_for_overlap, &w);
7c13edc8 5739
49728050
N		 5740 if (rw == WRITE)
5741 md_write_end(mddev);
016c76ac 5742 bio_endio(bi);
cc27b0c7 5743 return true;
1da177e4
LT		 5744}
5745
fd01b88c 5746static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks);
b522adcd 5747
fd01b88c 5748static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped)
1da177e4 5749{
52c03291
N		 5750 /* reshaping is quite different to recovery/resync so it is
5751 * handled quite separately ... here.
5752 *
5753 * On each call to sync_request, we gather one chunk worth of
5754 * destination stripes and flag them as expanding.
5755 * Then we find all the source stripes and request reads.
5756 * As the reads complete, handle_stripe will copy the data
5757 * into the destination stripe and release that stripe.
5758 */
d1688a6d 5759 struct r5conf *conf = mddev->private;
1da177e4 5760 struct stripe_head *sh;
ccfcc3c1 5761 sector_t first_sector, last_sector;
f416885e
N		 5762 int raid_disks = conf->previous_raid_disks;
5763 int data_disks = raid_disks - conf->max_degraded;
5764 int new_data_disks = conf->raid_disks - conf->max_degraded;
52c03291
N		 5765 int i;
5766 int dd_idx;
c8f517c4 5767 sector_t writepos, readpos, safepos;
ec32a2bd 5768 sector_t stripe_addr;
7a661381 5769 int reshape_sectors;
ab69ae12 5770 struct list_head stripes;
92140480 5771 sector_t retn;
52c03291 5772
fef9c61f
N		 5773 if (sector_nr == 0) {
5774 /* If restarting in the middle, skip the initial sectors */
2c810cdd 5775 if (mddev->reshape_backwards &&
fef9c61f
N		 5776 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
5777 sector_nr = raid5_size(mddev, 0, 0)
5778 - conf->reshape_progress;
6cbd8148
N		 5779 } else if (mddev->reshape_backwards &&
5780 conf->reshape_progress == MaxSector) {
5781 /* shouldn't happen, but just in case, finish up.*/
5782 sector_nr = MaxSector;
2c810cdd 5783 } else if (!mddev->reshape_backwards &&
fef9c61f
N		 5784 conf->reshape_progress > 0)
5785 sector_nr = conf->reshape_progress;
f416885e 5786 sector_div(sector_nr, new_data_disks);
fef9c61f 5787 if (sector_nr) {
8dee7211
N		 5788 mddev->curr_resync_completed = sector_nr;
5789 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
fef9c61f 5790 *skipped = 1;
92140480
N		 5791 retn = sector_nr;
5792 goto finish;
fef9c61f 5793 }
52c03291
N		 5794 }
5795
7a661381
N		 5796 /* We need to process a full chunk at a time.
5797 * If old and new chunk sizes differ, we need to process the
5798 * largest of these
5799 */
3cb5edf4
N		 5800
5801 reshape_sectors = max(conf->chunk_sectors, conf->prev_chunk_sectors);
7a661381 5802
b5254dd5
N		 5803 /* We update the metadata at least every 10 seconds, or when
5804 * the data about to be copied would over-write the source of
5805 * the data at the front of the range. i.e. one new_stripe
5806 * along from reshape_progress new_maps to after where
5807 * reshape_safe old_maps to
52c03291 5808 */
fef9c61f 5809 writepos = conf->reshape_progress;
f416885e 5810 sector_div(writepos, new_data_disks);
c8f517c4
N		 5811 readpos = conf->reshape_progress;
5812 sector_div(readpos, data_disks);
fef9c61f 5813 safepos = conf->reshape_safe;
f416885e 5814 sector_div(safepos, data_disks);
2c810cdd 5815 if (mddev->reshape_backwards) {
c74c0d76
N		 5816 BUG_ON(writepos < reshape_sectors);
5817 writepos -= reshape_sectors;
c8f517c4 5818 readpos += reshape_sectors;
7a661381 5819 safepos += reshape_sectors;
fef9c61f 5820 } else {
7a661381 5821 writepos += reshape_sectors;
c74c0d76
N		 5822 /* readpos and safepos are worst-case calculations.
5823 * A negative number is overly pessimistic, and causes
5824 * obvious problems for unsigned storage. So clip to 0.
5825 */
ed37d83e
N		 5826 readpos -= min_t(sector_t, reshape_sectors, readpos);
5827 safepos -= min_t(sector_t, reshape_sectors, safepos);
fef9c61f 5828 }
52c03291 5829
b5254dd5
N		 5830 /* Having calculated the 'writepos' possibly use it
5831 * to set 'stripe_addr' which is where we will write to.
5832 */
5833 if (mddev->reshape_backwards) {
5834 BUG_ON(conf->reshape_progress == 0);
5835 stripe_addr = writepos;
5836 BUG_ON((mddev->dev_sectors &
5837 ~((sector_t)reshape_sectors - 1))
5838 - reshape_sectors - stripe_addr
5839 != sector_nr);
5840 } else {
5841 BUG_ON(writepos != sector_nr + reshape_sectors);
5842 stripe_addr = sector_nr;
5843 }
5844
c8f517c4
N		 5845 /* 'writepos' is the most advanced device address we might write.
5846 * 'readpos' is the least advanced device address we might read.
5847 * 'safepos' is the least address recorded in the metadata as having
5848 * been reshaped.
b5254dd5
N		 5849 * If there is a min_offset_diff, these are adjusted either by
5850 * increasing the safepos/readpos if diff is negative, or
5851 * increasing writepos if diff is positive.
5852 * If 'readpos' is then behind 'writepos', there is no way that we can
c8f517c4
N		 5853 * ensure safety in the face of a crash - that must be done by userspace
5854 * making a backup of the data. So in that case there is no particular
5855 * rush to update metadata.
5856 * Otherwise if 'safepos' is behind 'writepos', then we really need to
5857 * update the metadata to advance 'safepos' to match 'readpos' so that
5858 * we can be safe in the event of a crash.
5859 * So we insist on updating metadata if safepos is behind writepos and
5860 * readpos is beyond writepos.
5861 * In any case, update the metadata every 10 seconds.
5862 * Maybe that number should be configurable, but I'm not sure it is
5863 * worth it.... maybe it could be a multiple of safemode_delay???
5864 */
b5254dd5
N		 5865 if (conf->min_offset_diff < 0) {
5866 safepos += -conf->min_offset_diff;
5867 readpos += -conf->min_offset_diff;
5868 } else
5869 writepos += conf->min_offset_diff;
5870
2c810cdd 5871 if ((mddev->reshape_backwards
c8f517c4
N		 5872 ? (safepos > writepos && readpos < writepos)
5873 : (safepos < writepos && readpos > writepos)) ||
5874 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
52c03291
N		 5875 /* Cannot proceed until we've updated the superblock... */
5876 wait_event(conf->wait_for_overlap,
c91abf5a
N		 5877 atomic_read(&conf->reshape_stripes)==0
5878 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5879 if (atomic_read(&conf->reshape_stripes) != 0)
5880 return 0;
fef9c61f 5881 mddev->reshape_position = conf->reshape_progress;
75d3da43 5882 mddev->curr_resync_completed = sector_nr;
c8f517c4 5883 conf->reshape_checkpoint = jiffies;
2953079c 5884 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
52c03291 5885 md_wakeup_thread(mddev->thread);
2953079c 5886 wait_event(mddev->sb_wait, mddev->sb_flags == 0 ||
c91abf5a
N		 5887 test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5888 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5889 return 0;
52c03291 5890 spin_lock_irq(&conf->device_lock);
fef9c61f 5891 conf->reshape_safe = mddev->reshape_position;
52c03291
N		 5892 spin_unlock_irq(&conf->device_lock);
5893 wake_up(&conf->wait_for_overlap);
acb180b0 5894 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
52c03291
N		 5895 }
5896
ab69ae12 5897 INIT_LIST_HEAD(&stripes);
7a661381 5898 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
52c03291 5899 int j;
a9f326eb 5900 int skipped_disk = 0;
6d036f7d 5901 sh = raid5_get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
52c03291
N		 5902 set_bit(STRIPE_EXPANDING, &sh->state);
5903 atomic_inc(&conf->reshape_stripes);
5904 /* If any of this stripe is beyond the end of the old
5905 * array, then we need to zero those blocks
5906 */
5907 for (j=sh->disks; j--;) {
5908 sector_t s;
5909 if (j == sh->pd_idx)
5910 continue;
f416885e 5911 if (conf->level == 6 &&
d0dabf7e 5912 j == sh->qd_idx)
f416885e 5913 continue;
6d036f7d 5914 s = raid5_compute_blocknr(sh, j, 0);
b522adcd 5915 if (s < raid5_size(mddev, 0, 0)) {
a9f326eb 5916 skipped_disk = 1;
52c03291
N		 5917 continue;
5918 }
5919 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
5920 set_bit(R5_Expanded, &sh->dev[j].flags);
5921 set_bit(R5_UPTODATE, &sh->dev[j].flags);
5922 }
a9f326eb 5923 if (!skipped_disk) {
52c03291
N		 5924 set_bit(STRIPE_EXPAND_READY, &sh->state);
5925 set_bit(STRIPE_HANDLE, &sh->state);
5926 }
ab69ae12 5927 list_add(&sh->lru, &stripes);
52c03291
N		 5928 }
5929 spin_lock_irq(&conf->device_lock);
2c810cdd 5930 if (mddev->reshape_backwards)
7a661381 5931 conf->reshape_progress -= reshape_sectors * new_data_disks;
fef9c61f 5932 else
7a661381 5933 conf->reshape_progress += reshape_sectors * new_data_disks;
52c03291
N		 5934 spin_unlock_irq(&conf->device_lock);
5935 /* Ok, those stripe are ready. We can start scheduling
5936 * reads on the source stripes.
5937 * The source stripes are determined by mapping the first and last
5938 * block on the destination stripes.
5939 */
52c03291 5940 first_sector =
ec32a2bd 5941 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
911d4ee8 5942 1, &dd_idx, NULL);
52c03291 5943 last_sector =
0e6e0271 5944 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
09c9e5fa 5945 * new_data_disks - 1),
911d4ee8 5946 1, &dd_idx, NULL);
58c0fed4
AN		 5947 if (last_sector >= mddev->dev_sectors)
5948 last_sector = mddev->dev_sectors - 1;
52c03291 5949 while (first_sector <= last_sector) {
6d036f7d 5950 sh = raid5_get_active_stripe(conf, first_sector, 1, 0, 1);
52c03291
N		 5951 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
5952 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 5953 raid5_release_stripe(sh);
52c03291
N		 5954 first_sector += STRIPE_SECTORS;
5955 }
ab69ae12
N		 5956 /* Now that the sources are clearly marked, we can release
5957 * the destination stripes
5958 */
5959 while (!list_empty(&stripes)) {
5960 sh = list_entry(stripes.next, struct stripe_head, lru);
5961 list_del_init(&sh->lru);
6d036f7d 5962 raid5_release_stripe(sh);
ab69ae12 5963 }
c6207277
N		 5964 /* If this takes us to the resync_max point where we have to pause,
5965 * then we need to write out the superblock.
5966 */
7a661381 5967 sector_nr += reshape_sectors;
92140480
N		 5968 retn = reshape_sectors;
5969finish:
c5e19d90
N		 5970 if (mddev->curr_resync_completed > mddev->resync_max ||
5971 (sector_nr - mddev->curr_resync_completed) * 2
c03f6a19 5972 >= mddev->resync_max - mddev->curr_resync_completed) {
c6207277
N		 5973 /* Cannot proceed until we've updated the superblock... */
5974 wait_event(conf->wait_for_overlap,
c91abf5a
N		 5975 atomic_read(&conf->reshape_stripes) == 0
5976 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5977 if (atomic_read(&conf->reshape_stripes) != 0)
5978 goto ret;
fef9c61f 5979 mddev->reshape_position = conf->reshape_progress;
75d3da43 5980 mddev->curr_resync_completed = sector_nr;
c8f517c4 5981 conf->reshape_checkpoint = jiffies;
2953079c 5982 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
c6207277
N		 5983 md_wakeup_thread(mddev->thread);
5984 wait_event(mddev->sb_wait,
2953079c 5985 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)
c91abf5a
N		 5986 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5987 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5988 goto ret;
c6207277 5989 spin_lock_irq(&conf->device_lock);
fef9c61f 5990 conf->reshape_safe = mddev->reshape_position;
c6207277
N		 5991 spin_unlock_irq(&conf->device_lock);
5992 wake_up(&conf->wait_for_overlap);
acb180b0 5993 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
c6207277 5994 }
c91abf5a 5995ret:
92140480 5996 return retn;
52c03291
N		 5997}
5998
849674e4
SL		 5999static inline sector_t raid5_sync_request(struct mddev *mddev, sector_t sector_nr,
6000 int *skipped)
52c03291 6001{
d1688a6d 6002 struct r5conf *conf = mddev->private;
52c03291 6003 struct stripe_head *sh;
58c0fed4 6004 sector_t max_sector = mddev->dev_sectors;
57dab0bd 6005 sector_t sync_blocks;
16a53ecc
N		 6006 int still_degraded = 0;
6007 int i;
1da177e4 6008
72626685 6009 if (sector_nr >= max_sector) {
1da177e4 6010 /* just being told to finish up .. nothing much to do */
cea9c228 6011
29269553
N		 6012 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
6013 end_reshape(conf);
6014 return 0;
6015 }
72626685
N		 6016
6017 if (mddev->curr_resync < max_sector) /* aborted */
6018 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
6019 &sync_blocks, 1);
16a53ecc 6020 else /* completed sync */
72626685
N		 6021 conf->fullsync = 0;
6022 bitmap_close_sync(mddev->bitmap);
6023
1da177e4
LT		 6024 return 0;
6025 }
ccfcc3c1 6026
64bd660b
N		 6027 /* Allow raid5_quiesce to complete */
6028 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
6029
52c03291
N		 6030 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6031 return reshape_request(mddev, sector_nr, skipped);
f6705578 6032
c6207277
N		 6033 /* No need to check resync_max as we never do more than one
6034 * stripe, and as resync_max will always be on a chunk boundary,
6035 * if the check in md_do_sync didn't fire, there is no chance
6036 * of overstepping resync_max here
6037 */
6038
16a53ecc 6039 /* if there is too many failed drives and we are trying
1da177e4
LT		 6040 * to resync, then assert that we are finished, because there is
6041 * nothing we can do.
6042 */
3285edf1 6043 if (mddev->degraded >= conf->max_degraded &&
16a53ecc 6044 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
58c0fed4 6045 sector_t rv = mddev->dev_sectors - sector_nr;
57afd89f 6046 *skipped = 1;
1da177e4
LT		 6047 return rv;
6048 }
6f608040 6049 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
6050 !conf->fullsync &&
6051 !bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
6052 sync_blocks >= STRIPE_SECTORS) {
72626685
N		 6053 /* we can skip this block, and probably more */
6054 sync_blocks /= STRIPE_SECTORS;
6055 *skipped = 1;
6056 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
6057 }
1da177e4 6058
c40f341f 6059 bitmap_cond_end_sync(mddev->bitmap, sector_nr, false);
b47490c9 6060
6d036f7d 6061 sh = raid5_get_active_stripe(conf, sector_nr, 0, 1, 0);
1da177e4 6062 if (sh == NULL) {
6d036f7d 6063 sh = raid5_get_active_stripe(conf, sector_nr, 0, 0, 0);
1da177e4 6064 /* make sure we don't swamp the stripe cache if someone else
16a53ecc 6065 * is trying to get access
1da177e4 6066 */
66c006a5 6067 schedule_timeout_uninterruptible(1);
1da177e4 6068 }
16a53ecc 6069 /* Need to check if array will still be degraded after recovery/resync
16d9cfab
EM		 6070 * Note in case of > 1 drive failures it's possible we're rebuilding
6071 * one drive while leaving another faulty drive in array.
16a53ecc 6072 */
16d9cfab
EM		 6073 rcu_read_lock();
6074 for (i = 0; i < conf->raid_disks; i++) {
6075 struct md_rdev *rdev = ACCESS_ONCE(conf->disks[i].rdev);
6076
6077 if (rdev == NULL || test_bit(Faulty, &rdev->flags))
16a53ecc 6078 still_degraded = 1;
16d9cfab
EM		 6079 }
6080 rcu_read_unlock();
16a53ecc
N		 6081
6082 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
6083
83206d66 6084 set_bit(STRIPE_SYNC_REQUESTED, &sh->state);
053f5b65 6085 set_bit(STRIPE_HANDLE, &sh->state);
1da177e4 6086
6d036f7d 6087 raid5_release_stripe(sh);
1da177e4
LT		 6088
6089 return STRIPE_SECTORS;
6090}
6091
0472a42b
N		 6092static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio,
6093 unsigned int offset)
46031f9a
RBJ		 6094{
6095 /* We may not be able to submit a whole bio at once as there
6096 * may not be enough stripe_heads available.
6097 * We cannot pre-allocate enough stripe_heads as we may need
6098 * more than exist in the cache (if we allow ever large chunks).
6099 * So we do one stripe head at a time and record in
6100 * ->bi_hw_segments how many have been done.
6101 *
6102 * We *know* that this entire raid_bio is in one chunk, so
6103 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
6104 */
6105 struct stripe_head *sh;
911d4ee8 6106 int dd_idx;
46031f9a
RBJ		 6107 sector_t sector, logical_sector, last_sector;
6108 int scnt = 0;
46031f9a
RBJ		 6109 int handled = 0;
6110
4f024f37
KO		 6111 logical_sector = raid_bio->bi_iter.bi_sector &
6112 ~((sector_t)STRIPE_SECTORS-1);
112bf897 6113 sector = raid5_compute_sector(conf, logical_sector,
911d4ee8 6114 0, &dd_idx, NULL);
f73a1c7d 6115 last_sector = bio_end_sector(raid_bio);
46031f9a
RBJ		 6116
6117 for (; logical_sector < last_sector;
387bb173
NB		 6118 logical_sector += STRIPE_SECTORS,
6119 sector += STRIPE_SECTORS,
6120 scnt++) {
46031f9a 6121
0472a42b 6122 if (scnt < offset)
46031f9a
RBJ		 6123 /* already done this stripe */
6124 continue;
6125
6d036f7d 6126 sh = raid5_get_active_stripe(conf, sector, 0, 1, 1);
46031f9a
RBJ		 6127
6128 if (!sh) {
6129 /* failed to get a stripe - must wait */
46031f9a 6130 conf->retry_read_aligned = raid_bio;
0472a42b 6131 conf->retry_read_offset = scnt;
46031f9a
RBJ		 6132 return handled;
6133 }
6134
da41ba65 6135 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) {
6d036f7d 6136 raid5_release_stripe(sh);
387bb173 6137 conf->retry_read_aligned = raid_bio;
0472a42b 6138 conf->retry_read_offset = scnt;
387bb173
NB		 6139 return handled;
6140 }
6141
3f9e7c14 6142 set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags);
36d1c647 6143 handle_stripe(sh);
6d036f7d 6144 raid5_release_stripe(sh);
46031f9a
RBJ		 6145 handled++;
6146 }
016c76ac
N		 6147
6148 bio_endio(raid_bio);
6149
46031f9a 6150 if (atomic_dec_and_test(&conf->active_aligned_reads))
b1b46486 6151 wake_up(&conf->wait_for_quiescent);
46031f9a
RBJ		 6152 return handled;
6153}
6154
bfc90cb0 6155static int handle_active_stripes(struct r5conf *conf, int group,
566c09c5
SL		 6156 struct r5worker *worker,
6157 struct list_head *temp_inactive_list)
46a06401
SL		 6158{
6159 struct stripe_head *batch[MAX_STRIPE_BATCH], *sh;
566c09c5
SL		 6160 int i, batch_size = 0, hash;
6161 bool release_inactive = false;
46a06401
SL		 6162
6163 while (batch_size < MAX_STRIPE_BATCH &&
851c30c9 6164 (sh = __get_priority_stripe(conf, group)) != NULL)
46a06401
SL		 6165 batch[batch_size++] = sh;
6166
566c09c5
SL		 6167 if (batch_size == 0) {
6168 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
6169 if (!list_empty(temp_inactive_list + i))
6170 break;
a8c34f91
SL		 6171 if (i == NR_STRIPE_HASH_LOCKS) {
6172 spin_unlock_irq(&conf->device_lock);
6173 r5l_flush_stripe_to_raid(conf->log);
6174 spin_lock_irq(&conf->device_lock);
566c09c5 6175 return batch_size;
a8c34f91 6176 }
566c09c5
SL		 6177 release_inactive = true;
6178 }
46a06401
SL		 6179 spin_unlock_irq(&conf->device_lock);
6180
566c09c5
SL		 6181 release_inactive_stripe_list(conf, temp_inactive_list,
6182 NR_STRIPE_HASH_LOCKS);
6183
a8c34f91 6184 r5l_flush_stripe_to_raid(conf->log);
566c09c5
SL		 6185 if (release_inactive) {
6186 spin_lock_irq(&conf->device_lock);
6187 return 0;
6188 }
6189
46a06401
SL		 6190 for (i = 0; i < batch_size; i++)
6191 handle_stripe(batch[i]);
ff875738 6192 log_write_stripe_run(conf);
46a06401
SL		 6193
6194 cond_resched();
6195
6196 spin_lock_irq(&conf->device_lock);
566c09c5
SL		 6197 for (i = 0; i < batch_size; i++) {
6198 hash = batch[i]->hash_lock_index;
6199 __release_stripe(conf, batch[i], &temp_inactive_list[hash]);
6200 }
46a06401
SL		 6201 return batch_size;
6202}
46031f9a 6203
851c30c9
SL		 6204static void raid5_do_work(struct work_struct *work)
6205{
6206 struct r5worker *worker = container_of(work, struct r5worker, work);
6207 struct r5worker_group *group = worker->group;
6208 struct r5conf *conf = group->conf;
16d997b7 6209 struct mddev *mddev = conf->mddev;
851c30c9
SL		 6210 int group_id = group - conf->worker_groups;
6211 int handled;
6212 struct blk_plug plug;
6213
6214 pr_debug("+++ raid5worker active\n");
6215
6216 blk_start_plug(&plug);
6217 handled = 0;
6218 spin_lock_irq(&conf->device_lock);
6219 while (1) {
6220 int batch_size, released;
6221
566c09c5 6222 released = release_stripe_list(conf, worker->temp_inactive_list);
851c30c9 6223
566c09c5
SL		 6224 batch_size = handle_active_stripes(conf, group_id, worker,
6225 worker->temp_inactive_list);
bfc90cb0 6226 worker->working = false;
851c30c9
SL		 6227 if (!batch_size && !released)
6228 break;
6229 handled += batch_size;
16d997b7
N		 6230 wait_event_lock_irq(mddev->sb_wait,
6231 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags),
6232 conf->device_lock);
851c30c9
SL		 6233 }
6234 pr_debug("%d stripes handled\n", handled);
6235
6236 spin_unlock_irq(&conf->device_lock);
7e96d559 6237
9c72a18e
SL		 6238 flush_deferred_bios(conf);
6239
6240 r5l_flush_stripe_to_raid(conf->log);
6241
7e96d559 6242 async_tx_issue_pending_all();
851c30c9
SL		 6243 blk_finish_plug(&plug);
6244
6245 pr_debug("--- raid5worker inactive\n");
6246}
6247
1da177e4
LT		 6248/*
6249 * This is our raid5 kernel thread.
6250 *
6251 * We scan the hash table for stripes which can be handled now.
6252 * During the scan, completed stripes are saved for us by the interrupt
6253 * handler, so that they will not have to wait for our next wakeup.
6254 */
4ed8731d 6255static void raid5d(struct md_thread *thread)
1da177e4 6256{
4ed8731d 6257 struct mddev *mddev = thread->mddev;
d1688a6d 6258 struct r5conf *conf = mddev->private;
1da177e4 6259 int handled;
e1dfa0a2 6260 struct blk_plug plug;
1da177e4 6261
45b4233c 6262 pr_debug("+++ raid5d active\n");
1da177e4
LT		 6263
6264 md_check_recovery(mddev);
1da177e4 6265
e1dfa0a2 6266 blk_start_plug(&plug);
1da177e4
LT		 6267 handled = 0;
6268 spin_lock_irq(&conf->device_lock);
6269 while (1) {
46031f9a 6270 struct bio *bio;
773ca82f 6271 int batch_size, released;
0472a42b 6272 unsigned int offset;
773ca82f 6273
566c09c5 6274 released = release_stripe_list(conf, conf->temp_inactive_list);
edbe83ab
N		 6275 if (released)
6276 clear_bit(R5_DID_ALLOC, &conf->cache_state);
1da177e4 6277
0021b7bc 6278 if (
7c13edc8
N		 6279 !list_empty(&conf->bitmap_list)) {
6280 /* Now is a good time to flush some bitmap updates */
6281 conf->seq_flush++;
700e432d 6282 spin_unlock_irq(&conf->device_lock);
72626685 6283 bitmap_unplug(mddev->bitmap);
700e432d 6284 spin_lock_irq(&conf->device_lock);
7c13edc8 6285 conf->seq_write = conf->seq_flush;
566c09c5 6286 activate_bit_delay(conf, conf->temp_inactive_list);
72626685 6287 }
0021b7bc 6288 raid5_activate_delayed(conf);
72626685 6289
0472a42b 6290 while ((bio = remove_bio_from_retry(conf, &offset))) {
46031f9a
RBJ		 6291 int ok;
6292 spin_unlock_irq(&conf->device_lock);
0472a42b 6293 ok = retry_aligned_read(conf, bio, offset);
46031f9a
RBJ		 6294 spin_lock_irq(&conf->device_lock);
6295 if (!ok)
6296 break;
6297 handled++;
6298 }
6299
566c09c5
SL		 6300 batch_size = handle_active_stripes(conf, ANY_GROUP, NULL,
6301 conf->temp_inactive_list);
773ca82f 6302 if (!batch_size && !released)
1da177e4 6303 break;
46a06401 6304 handled += batch_size;
1da177e4 6305
2953079c 6306 if (mddev->sb_flags & ~(1 << MD_SB_CHANGE_PENDING)) {
46a06401 6307 spin_unlock_irq(&conf->device_lock);
de393cde 6308 md_check_recovery(mddev);
46a06401
SL		 6309 spin_lock_irq(&conf->device_lock);
6310 }
1da177e4 6311 }
45b4233c 6312 pr_debug("%d stripes handled\n", handled);
1da177e4
LT		 6313
6314 spin_unlock_irq(&conf->device_lock);
2d5b569b
N		 6315 if (test_and_clear_bit(R5_ALLOC_MORE, &conf->cache_state) &&
6316 mutex_trylock(&conf->cache_size_mutex)) {
edbe83ab
N		 6317 grow_one_stripe(conf, __GFP_NOWARN);
6318 /* Set flag even if allocation failed. This helps
6319 * slow down allocation requests when mem is short
6320 */
6321 set_bit(R5_DID_ALLOC, &conf->cache_state);
2d5b569b 6322 mutex_unlock(&conf->cache_size_mutex);
edbe83ab 6323 }
1da177e4 6324
765d704d
SL		 6325 flush_deferred_bios(conf);
6326
0576b1c6
SL		 6327 r5l_flush_stripe_to_raid(conf->log);
6328
c9f21aaf 6329 async_tx_issue_pending_all();
e1dfa0a2 6330 blk_finish_plug(&plug);
1da177e4 6331
45b4233c 6332 pr_debug("--- raid5d inactive\n");
1da177e4
LT		 6333}
6334
3f294f4f 6335static ssize_t
fd01b88c 6336raid5_show_stripe_cache_size(struct mddev *mddev, char *page)
3f294f4f 6337{
7b1485ba
N		 6338 struct r5conf *conf;
6339 int ret = 0;
6340 spin_lock(&mddev->lock);
6341 conf = mddev->private;
96de1e66 6342 if (conf)
edbe83ab 6343 ret = sprintf(page, "%d\n", conf->min_nr_stripes);
7b1485ba
N		 6344 spin_unlock(&mddev->lock);
6345 return ret;
3f294f4f
N		 6346}
6347
c41d4ac4 6348int
fd01b88c 6349raid5_set_cache_size(struct mddev *mddev, int size)
3f294f4f 6350{
d1688a6d 6351 struct r5conf *conf = mddev->private;
b5470dc5 6352
c41d4ac4 6353 if (size <= 16 || size > 32768)
3f294f4f 6354 return -EINVAL;
486f0644 6355
edbe83ab 6356 conf->min_nr_stripes = size;
2d5b569b 6357 mutex_lock(&conf->cache_size_mutex);
486f0644
N		 6358 while (size < conf->max_nr_stripes &&
6359 drop_one_stripe(conf))
6360 ;
2d5b569b 6361 mutex_unlock(&conf->cache_size_mutex);
486f0644 6362
2214c260 6363 md_allow_write(mddev);
486f0644 6364
2d5b569b 6365 mutex_lock(&conf->cache_size_mutex);
486f0644
N		 6366 while (size > conf->max_nr_stripes)
6367 if (!grow_one_stripe(conf, GFP_KERNEL))
6368 break;
2d5b569b 6369 mutex_unlock(&conf->cache_size_mutex);
486f0644 6370
c41d4ac4
N		 6371 return 0;
6372}
6373EXPORT_SYMBOL(raid5_set_cache_size);
6374
6375static ssize_t
fd01b88c 6376raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len)
c41d4ac4 6377{
6791875e 6378 struct r5conf *conf;
c41d4ac4
N		 6379 unsigned long new;
6380 int err;
6381
6382 if (len >= PAGE_SIZE)
6383 return -EINVAL;
b29bebd6 6384 if (kstrtoul(page, 10, &new))
c41d4ac4 6385 return -EINVAL;
6791875e 6386 err = mddev_lock(mddev);
c41d4ac4
N		 6387 if (err)
6388 return err;
6791875e
N		 6389 conf = mddev->private;
6390 if (!conf)
6391 err = -ENODEV;
6392 else
6393 err = raid5_set_cache_size(mddev, new);
6394 mddev_unlock(mddev);
6395
6396 return err ?: len;
3f294f4f 6397}
007583c9 6398
96de1e66
N		 6399static struct md_sysfs_entry
6400raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
6401 raid5_show_stripe_cache_size,
6402 raid5_store_stripe_cache_size);
3f294f4f 6403
d06f191f
MS		 6404static ssize_t
6405raid5_show_rmw_level(struct mddev *mddev, char *page)
6406{
6407 struct r5conf *conf = mddev->private;
6408 if (conf)
6409 return sprintf(page, "%d\n", conf->rmw_level);
6410 else
6411 return 0;
6412}
6413
6414static ssize_t
6415raid5_store_rmw_level(struct mddev *mddev, const char *page, size_t len)
6416{
6417 struct r5conf *conf = mddev->private;
6418 unsigned long new;
6419
6420 if (!conf)
6421 return -ENODEV;
6422
6423 if (len >= PAGE_SIZE)
6424 return -EINVAL;
6425
6426 if (kstrtoul(page, 10, &new))
6427 return -EINVAL;
6428
6429 if (new != PARITY_DISABLE_RMW && !raid6_call.xor_syndrome)
6430 return -EINVAL;
6431
6432 if (new != PARITY_DISABLE_RMW &&
6433 new != PARITY_ENABLE_RMW &&
6434 new != PARITY_PREFER_RMW)
6435 return -EINVAL;
6436
6437 conf->rmw_level = new;
6438 return len;
6439}
6440
6441static struct md_sysfs_entry
6442raid5_rmw_level = __ATTR(rmw_level, S_IRUGO | S_IWUSR,
6443 raid5_show_rmw_level,
6444 raid5_store_rmw_level);
6445
6446
8b3e6cdc 6447static ssize_t
fd01b88c 6448raid5_show_preread_threshold(struct mddev *mddev, char *page)
8b3e6cdc 6449{
7b1485ba
N		 6450 struct r5conf *conf;
6451 int ret = 0;
6452 spin_lock(&mddev->lock);
6453 conf = mddev->private;
8b3e6cdc 6454 if (conf)
7b1485ba
N		 6455 ret = sprintf(page, "%d\n", conf->bypass_threshold);
6456 spin_unlock(&mddev->lock);
6457 return ret;
8b3e6cdc
DW		 6458}
6459
6460static ssize_t
fd01b88c 6461raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len)
8b3e6cdc 6462{
6791875e 6463 struct r5conf *conf;
4ef197d8 6464 unsigned long new;
6791875e
N		 6465 int err;
6466
8b3e6cdc
DW		 6467 if (len >= PAGE_SIZE)
6468 return -EINVAL;
b29bebd6 6469 if (kstrtoul(page, 10, &new))
8b3e6cdc 6470 return -EINVAL;
6791875e
N		 6471
6472 err = mddev_lock(mddev);
6473 if (err)
6474 return err;
6475 conf = mddev->private;
6476 if (!conf)
6477 err = -ENODEV;
edbe83ab 6478 else if (new > conf->min_nr_stripes)
6791875e
N		 6479 err = -EINVAL;
6480 else
6481 conf->bypass_threshold = new;
6482 mddev_unlock(mddev);
6483 return err ?: len;
8b3e6cdc
DW		 6484}
6485
6486static struct md_sysfs_entry
6487raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
6488 S_IRUGO | S_IWUSR,
6489 raid5_show_preread_threshold,
6490 raid5_store_preread_threshold);
6491
d592a996
SL		 6492static ssize_t
6493raid5_show_skip_copy(struct mddev *mddev, char *page)
6494{
7b1485ba
N		 6495 struct r5conf *conf;
6496 int ret = 0;
6497 spin_lock(&mddev->lock);
6498 conf = mddev->private;
d592a996 6499 if (conf)
7b1485ba
N		 6500 ret = sprintf(page, "%d\n", conf->skip_copy);
6501 spin_unlock(&mddev->lock);
6502 return ret;
d592a996
SL		 6503}
6504
6505static ssize_t
6506raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len)
6507{
6791875e 6508 struct r5conf *conf;
d592a996 6509 unsigned long new;
6791875e
N		 6510 int err;
6511
d592a996
SL		 6512 if (len >= PAGE_SIZE)
6513 return -EINVAL;
d592a996
SL		 6514 if (kstrtoul(page, 10, &new))
6515 return -EINVAL;
6516 new = !!new;
6791875e
N		 6517
6518 err = mddev_lock(mddev);
6519 if (err)
6520 return err;
6521 conf = mddev->private;
6522 if (!conf)
6523 err = -ENODEV;
6524 else if (new != conf->skip_copy) {
6525 mddev_suspend(mddev);
6526 conf->skip_copy = new;
6527 if (new)
dc3b17cc 6528 mddev->queue->backing_dev_info->capabilities |=
6791875e
N		 6529 BDI_CAP_STABLE_WRITES;
6530 else
dc3b17cc 6531 mddev->queue->backing_dev_info->capabilities &=
6791875e
N		 6532 ~BDI_CAP_STABLE_WRITES;
6533 mddev_resume(mddev);
6534 }
6535 mddev_unlock(mddev);
6536 return err ?: len;
d592a996
SL		 6537}
6538
6539static struct md_sysfs_entry
6540raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR,
6541 raid5_show_skip_copy,
6542 raid5_store_skip_copy);
6543
3f294f4f 6544static ssize_t
fd01b88c 6545stripe_cache_active_show(struct mddev *mddev, char *page)
3f294f4f 6546{
d1688a6d 6547 struct r5conf *conf = mddev->private;
96de1e66
N		 6548 if (conf)
6549 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
6550 else
6551 return 0;
3f294f4f
N		 6552}
6553
96de1e66
N		 6554static struct md_sysfs_entry
6555raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
3f294f4f 6556
b721420e
SL		 6557static ssize_t
6558raid5_show_group_thread_cnt(struct mddev *mddev, char *page)
6559{
7b1485ba
N		 6560 struct r5conf *conf;
6561 int ret = 0;
6562 spin_lock(&mddev->lock);
6563 conf = mddev->private;
b721420e 6564 if (conf)
7b1485ba
N		 6565 ret = sprintf(page, "%d\n", conf->worker_cnt_per_group);
6566 spin_unlock(&mddev->lock);
6567 return ret;
b721420e
SL		 6568}
6569
60aaf933 6570static int alloc_thread_groups(struct r5conf *conf, int cnt,
6571 int *group_cnt,
6572 int *worker_cnt_per_group,
6573 struct r5worker_group **worker_groups);
b721420e
SL		 6574static ssize_t
6575raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len)
6576{
6791875e 6577 struct r5conf *conf;
7d5d7b50 6578 unsigned int new;
b721420e 6579 int err;
60aaf933 6580 struct r5worker_group *new_groups, *old_groups;
6581 int group_cnt, worker_cnt_per_group;
b721420e
SL		 6582
6583 if (len >= PAGE_SIZE)
6584 return -EINVAL;
7d5d7b50
SL		 6585 if (kstrtouint(page, 10, &new))
6586 return -EINVAL;
6587 /* 8192 should be big enough */
6588 if (new > 8192)
b721420e
SL		 6589 return -EINVAL;
6590
6791875e
N		 6591 err = mddev_lock(mddev);
6592 if (err)
6593 return err;
6594 conf = mddev->private;
6595 if (!conf)
6596 err = -ENODEV;
6597 else if (new != conf->worker_cnt_per_group) {
6598 mddev_suspend(mddev);
b721420e 6599
6791875e
N		 6600 old_groups = conf->worker_groups;
6601 if (old_groups)
6602 flush_workqueue(raid5_wq);
d206dcfa 6603
6791875e
N		 6604 err = alloc_thread_groups(conf, new,
6605 &group_cnt, &worker_cnt_per_group,
6606 &new_groups);
6607 if (!err) {
6608 spin_lock_irq(&conf->device_lock);
6609 conf->group_cnt = group_cnt;
6610 conf->worker_cnt_per_group = worker_cnt_per_group;
6611 conf->worker_groups = new_groups;
6612 spin_unlock_irq(&conf->device_lock);
b721420e 6613
6791875e
N		 6614 if (old_groups)
6615 kfree(old_groups[0].workers);
6616 kfree(old_groups);
6617 }
6618 mddev_resume(mddev);
b721420e 6619 }
6791875e 6620 mddev_unlock(mddev);
b721420e 6621
6791875e 6622 return err ?: len;
b721420e
SL		 6623}
6624
6625static struct md_sysfs_entry
6626raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR,
6627 raid5_show_group_thread_cnt,
6628 raid5_store_group_thread_cnt);
6629
007583c9 6630static struct attribute *raid5_attrs[] = {
3f294f4f
N		 6631 &raid5_stripecache_size.attr,
6632 &raid5_stripecache_active.attr,
8b3e6cdc 6633 &raid5_preread_bypass_threshold.attr,
b721420e 6634 &raid5_group_thread_cnt.attr,
d592a996 6635 &raid5_skip_copy.attr,
d06f191f 6636 &raid5_rmw_level.attr,
2c7da14b 6637 &r5c_journal_mode.attr,
3f294f4f
N		 6638 NULL,
6639};
007583c9
N		 6640static struct attribute_group raid5_attrs_group = {
6641 .name = NULL,
6642 .attrs = raid5_attrs,
3f294f4f
N		 6643};
6644
60aaf933 6645static int alloc_thread_groups(struct r5conf *conf, int cnt,
6646 int *group_cnt,
6647 int *worker_cnt_per_group,
6648 struct r5worker_group **worker_groups)
851c30c9 6649{
566c09c5 6650 int i, j, k;
851c30c9
SL		 6651 ssize_t size;
6652 struct r5worker *workers;
6653
60aaf933 6654 *worker_cnt_per_group = cnt;
851c30c9 6655 if (cnt == 0) {
60aaf933 6656 *group_cnt = 0;
6657 *worker_groups = NULL;
851c30c9
SL		 6658 return 0;
6659 }
60aaf933 6660 *group_cnt = num_possible_nodes();
851c30c9 6661 size = sizeof(struct r5worker) * cnt;
60aaf933 6662 workers = kzalloc(size * *group_cnt, GFP_NOIO);
6663 *worker_groups = kzalloc(sizeof(struct r5worker_group) *
6664 *group_cnt, GFP_NOIO);
6665 if (!*worker_groups || !workers) {
851c30c9 6666 kfree(workers);
60aaf933 6667 kfree(*worker_groups);
851c30c9
SL		 6668 return -ENOMEM;
6669 }
6670
60aaf933 6671 for (i = 0; i < *group_cnt; i++) {
851c30c9
SL		 6672 struct r5worker_group *group;
6673
0c775d52 6674 group = &(*worker_groups)[i];
851c30c9 6675 INIT_LIST_HEAD(&group->handle_list);
535ae4eb 6676 INIT_LIST_HEAD(&group->loprio_list);
851c30c9
SL		 6677 group->conf = conf;
6678 group->workers = workers + i * cnt;
6679
6680 for (j = 0; j < cnt; j++) {
566c09c5
SL		 6681 struct r5worker *worker = group->workers + j;
6682 worker->group = group;
6683 INIT_WORK(&worker->work, raid5_do_work);
6684
6685 for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++)
6686 INIT_LIST_HEAD(worker->temp_inactive_list + k);
851c30c9
SL		 6687 }
6688 }
6689
6690 return 0;
6691}
6692
6693static void free_thread_groups(struct r5conf *conf)
6694{
6695 if (conf->worker_groups)
6696 kfree(conf->worker_groups[0].workers);
6697 kfree(conf->worker_groups);
6698 conf->worker_groups = NULL;
6699}
6700
80c3a6ce 6701static sector_t
fd01b88c 6702raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks)
80c3a6ce 6703{
d1688a6d 6704 struct r5conf *conf = mddev->private;
80c3a6ce
DW		 6705
6706 if (!sectors)
6707 sectors = mddev->dev_sectors;
5e5e3e78 6708 if (!raid_disks)
7ec05478 6709 /* size is defined by the smallest of previous and new size */
5e5e3e78 6710 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
80c3a6ce 6711
3cb5edf4
N		 6712 sectors &= ~((sector_t)conf->chunk_sectors - 1);
6713 sectors &= ~((sector_t)conf->prev_chunk_sectors - 1);
80c3a6ce
DW		 6714 return sectors * (raid_disks - conf->max_degraded);
6715}
6716
789b5e03
ON		 6717static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
6718{
6719 safe_put_page(percpu->spare_page);
46d5b785 6720 if (percpu->scribble)
6721 flex_array_free(percpu->scribble);
789b5e03
ON		 6722 percpu->spare_page = NULL;
6723 percpu->scribble = NULL;
6724}
6725
6726static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
6727{
6728 if (conf->level == 6 && !percpu->spare_page)
6729 percpu->spare_page = alloc_page(GFP_KERNEL);
6730 if (!percpu->scribble)
46d5b785 6731 percpu->scribble = scribble_alloc(max(conf->raid_disks,
738a2738
N		 6732 conf->previous_raid_disks),
6733 max(conf->chunk_sectors,
6734 conf->prev_chunk_sectors)
6735 / STRIPE_SECTORS,
6736 GFP_KERNEL);
789b5e03
ON		 6737
6738 if (!percpu->scribble || (conf->level == 6 && !percpu->spare_page)) {
6739 free_scratch_buffer(conf, percpu);
6740 return -ENOMEM;
6741 }
6742
6743 return 0;
6744}
6745
29c6d1bb 6746static int raid456_cpu_dead(unsigned int cpu, struct hlist_node *node)
36d1c647 6747{
29c6d1bb
SAS		 6748 struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node);
6749
6750 free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
6751 return 0;
6752}
36d1c647 6753
29c6d1bb
SAS		 6754static void raid5_free_percpu(struct r5conf *conf)
6755{
36d1c647
DW		 6756 if (!conf->percpu)
6757 return;
6758
29c6d1bb 6759 cpuhp_state_remove_instance(CPUHP_MD_RAID5_PREPARE, &conf->node);
36d1c647
DW		 6760 free_percpu(conf->percpu);
6761}
6762
d1688a6d 6763static void free_conf(struct r5conf *conf)
95fc17aa 6764{
d7bd398e
SL		 6765 int i;
6766
ff875738
AP		 6767 log_exit(conf);
6768
30c89465 6769 if (conf->shrinker.nr_deferred)
edbe83ab 6770 unregister_shrinker(&conf->shrinker);
5c7e81c3 6771
851c30c9 6772 free_thread_groups(conf);
95fc17aa 6773 shrink_stripes(conf);
36d1c647 6774 raid5_free_percpu(conf);
d7bd398e
SL		 6775 for (i = 0; i < conf->pool_size; i++)
6776 if (conf->disks[i].extra_page)
6777 put_page(conf->disks[i].extra_page);
95fc17aa 6778 kfree(conf->disks);
dd7a8f5d
N		 6779 if (conf->bio_split)
6780 bioset_free(conf->bio_split);
95fc17aa 6781 kfree(conf->stripe_hashtbl);
aaf9f12e 6782 kfree(conf->pending_data);
95fc17aa
DW		 6783 kfree(conf);
6784}
6785
29c6d1bb 6786static int raid456_cpu_up_prepare(unsigned int cpu, struct hlist_node *node)
36d1c647 6787{
29c6d1bb 6788 struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node);
36d1c647
DW		 6789 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
6790
29c6d1bb 6791 if (alloc_scratch_buffer(conf, percpu)) {
cc6167b4
N		 6792 pr_warn("%s: failed memory allocation for cpu%u\n",
6793 __func__, cpu);
29c6d1bb 6794 return -ENOMEM;
36d1c647 6795 }
29c6d1bb 6796 return 0;
36d1c647 6797}
36d1c647 6798
d1688a6d 6799static int raid5_alloc_percpu(struct r5conf *conf)
36d1c647 6800{
789b5e03 6801 int err = 0;
36d1c647 6802
789b5e03
ON		 6803 conf->percpu = alloc_percpu(struct raid5_percpu);
6804 if (!conf->percpu)
36d1c647 6805 return -ENOMEM;
789b5e03 6806
29c6d1bb 6807 err = cpuhp_state_add_instance(CPUHP_MD_RAID5_PREPARE, &conf->node);
27a353c0
SL		 6808 if (!err) {
6809 conf->scribble_disks = max(conf->raid_disks,
6810 conf->previous_raid_disks);
6811 conf->scribble_sectors = max(conf->chunk_sectors,
6812 conf->prev_chunk_sectors);
6813 }
36d1c647
DW		 6814 return err;
6815}
6816
edbe83ab
N		 6817static unsigned long raid5_cache_scan(struct shrinker *shrink,
6818 struct shrink_control *sc)
6819{
6820 struct r5conf *conf = container_of(shrink, struct r5conf, shrinker);
2d5b569b
N		 6821 unsigned long ret = SHRINK_STOP;
6822
6823 if (mutex_trylock(&conf->cache_size_mutex)) {
6824 ret= 0;
49895bcc
N		 6825 while (ret < sc->nr_to_scan &&
6826 conf->max_nr_stripes > conf->min_nr_stripes) {
2d5b569b
N		 6827 if (drop_one_stripe(conf) == 0) {
6828 ret = SHRINK_STOP;
6829 break;
6830 }
6831 ret++;
6832 }
6833 mutex_unlock(&conf->cache_size_mutex);
edbe83ab
N		 6834 }
6835 return ret;
6836}
6837
6838static unsigned long raid5_cache_count(struct shrinker *shrink,
6839 struct shrink_control *sc)
6840{
6841 struct r5conf *conf = container_of(shrink, struct r5conf, shrinker);
6842
6843 if (conf->max_nr_stripes < conf->min_nr_stripes)
6844 /* unlikely, but not impossible */
6845 return 0;
6846 return conf->max_nr_stripes - conf->min_nr_stripes;
6847}
6848
d1688a6d 6849static struct r5conf *setup_conf(struct mddev *mddev)
1da177e4 6850{
d1688a6d 6851 struct r5conf *conf;
5e5e3e78 6852 int raid_disk, memory, max_disks;
3cb03002 6853 struct md_rdev *rdev;
1da177e4 6854 struct disk_info *disk;
0232605d 6855 char pers_name[6];
566c09c5 6856 int i;
60aaf933 6857 int group_cnt, worker_cnt_per_group;
6858 struct r5worker_group *new_group;
1da177e4 6859
91adb564
N		 6860 if (mddev->new_level != 5
6861 && mddev->new_level != 4
6862 && mddev->new_level != 6) {
cc6167b4
N		 6863 pr_warn("md/raid:%s: raid level not set to 4/5/6 (%d)\n",
6864 mdname(mddev), mddev->new_level);
91adb564 6865 return ERR_PTR(-EIO);
1da177e4 6866 }
91adb564
N		 6867 if ((mddev->new_level == 5
6868 && !algorithm_valid_raid5(mddev->new_layout)) ||
6869 (mddev->new_level == 6
6870 && !algorithm_valid_raid6(mddev->new_layout))) {
cc6167b4
N		 6871 pr_warn("md/raid:%s: layout %d not supported\n",
6872 mdname(mddev), mddev->new_layout);
91adb564 6873 return ERR_PTR(-EIO);
99c0fb5f 6874 }
91adb564 6875 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
cc6167b4
N		 6876 pr_warn("md/raid:%s: not enough configured devices (%d, minimum 4)\n",
6877 mdname(mddev), mddev->raid_disks);
91adb564 6878 return ERR_PTR(-EINVAL);
4bbf3771
N		 6879 }
6880
664e7c41
AN		 6881 if (!mddev->new_chunk_sectors ||
6882 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
6883 !is_power_of_2(mddev->new_chunk_sectors)) {
cc6167b4
N		 6884 pr_warn("md/raid:%s: invalid chunk size %d\n",
6885 mdname(mddev), mddev->new_chunk_sectors << 9);
91adb564 6886 return ERR_PTR(-EINVAL);
f6705578
N		 6887 }
6888
d1688a6d 6889 conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL);
91adb564 6890 if (conf == NULL)
1da177e4 6891 goto abort;
aaf9f12e
SL		 6892 INIT_LIST_HEAD(&conf->free_list);
6893 INIT_LIST_HEAD(&conf->pending_list);
6894 conf->pending_data = kzalloc(sizeof(struct r5pending_data) *
6895 PENDING_IO_MAX, GFP_KERNEL);
6896 if (!conf->pending_data)
6897 goto abort;
6898 for (i = 0; i < PENDING_IO_MAX; i++)
6899 list_add(&conf->pending_data[i].sibling, &conf->free_list);
851c30c9 6900 /* Don't enable multi-threading by default*/
60aaf933 6901 if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group,
6902 &new_group)) {
6903 conf->group_cnt = group_cnt;
6904 conf->worker_cnt_per_group = worker_cnt_per_group;
6905 conf->worker_groups = new_group;
6906 } else
851c30c9 6907 goto abort;
f5efd45a 6908 spin_lock_init(&conf->device_lock);
c46501b2 6909 seqcount_init(&conf->gen_lock);
2d5b569b 6910 mutex_init(&conf->cache_size_mutex);
b1b46486 6911 init_waitqueue_head(&conf->wait_for_quiescent);
6ab2a4b8 6912 init_waitqueue_head(&conf->wait_for_stripe);
f5efd45a
DW		 6913 init_waitqueue_head(&conf->wait_for_overlap);
6914 INIT_LIST_HEAD(&conf->handle_list);
535ae4eb 6915 INIT_LIST_HEAD(&conf->loprio_list);
f5efd45a
DW		 6916 INIT_LIST_HEAD(&conf->hold_list);
6917 INIT_LIST_HEAD(&conf->delayed_list);
6918 INIT_LIST_HEAD(&conf->bitmap_list);
773ca82f 6919 init_llist_head(&conf->released_stripes);
f5efd45a
DW		 6920 atomic_set(&conf->active_stripes, 0);
6921 atomic_set(&conf->preread_active_stripes, 0);
6922 atomic_set(&conf->active_aligned_reads, 0);
765d704d
SL		 6923 spin_lock_init(&conf->pending_bios_lock);
6924 conf->batch_bio_dispatch = true;
6925 rdev_for_each(rdev, mddev) {
6926 if (test_bit(Journal, &rdev->flags))
6927 continue;
6928 if (blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
6929 conf->batch_bio_dispatch = false;
6930 break;
6931 }
6932 }
6933
f5efd45a 6934 conf->bypass_threshold = BYPASS_THRESHOLD;
d890fa2b 6935 conf->recovery_disabled = mddev->recovery_disabled - 1;
91adb564
N		 6936
6937 conf->raid_disks = mddev->raid_disks;
6938 if (mddev->reshape_position == MaxSector)
6939 conf->previous_raid_disks = mddev->raid_disks;
6940 else
f6705578 6941 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
5e5e3e78 6942 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
f6705578 6943
5e5e3e78 6944 conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
b55e6bfc 6945 GFP_KERNEL);
d7bd398e 6946
b55e6bfc
N		 6947 if (!conf->disks)
6948 goto abort;
9ffae0cf 6949
d7bd398e
SL		 6950 for (i = 0; i < max_disks; i++) {
6951 conf->disks[i].extra_page = alloc_page(GFP_KERNEL);
6952 if (!conf->disks[i].extra_page)
6953 goto abort;
6954 }
6955
011067b0 6956 conf->bio_split = bioset_create(BIO_POOL_SIZE, 0, 0);
dd7a8f5d
N		 6957 if (!conf->bio_split)
6958 goto abort;
1da177e4
LT		 6959 conf->mddev = mddev;
6960
fccddba0 6961 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
1da177e4 6962 goto abort;
1da177e4 6963
566c09c5
SL		 6964 /* We init hash_locks[0] separately to that it can be used
6965 * as the reference lock in the spin_lock_nest_lock() call
6966 * in lock_all_device_hash_locks_irq in order to convince
6967 * lockdep that we know what we are doing.
6968 */
6969 spin_lock_init(conf->hash_locks);
6970 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
6971 spin_lock_init(conf->hash_locks + i);
6972
6973 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
6974 INIT_LIST_HEAD(conf->inactive_list + i);
6975
6976 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
6977 INIT_LIST_HEAD(conf->temp_inactive_list + i);
6978
1e6d690b
SL		 6979 atomic_set(&conf->r5c_cached_full_stripes, 0);
6980 INIT_LIST_HEAD(&conf->r5c_full_stripe_list);
6981 atomic_set(&conf->r5c_cached_partial_stripes, 0);
6982 INIT_LIST_HEAD(&conf->r5c_partial_stripe_list);
e33fbb9c
SL		 6983 atomic_set(&conf->r5c_flushing_full_stripes, 0);
6984 atomic_set(&conf->r5c_flushing_partial_stripes, 0);
1e6d690b 6985
36d1c647 6986 conf->level = mddev->new_level;
46d5b785 6987 conf->chunk_sectors = mddev->new_chunk_sectors;
36d1c647
DW		 6988 if (raid5_alloc_percpu(conf) != 0)
6989 goto abort;
6990
0c55e022 6991 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
1da177e4 6992
dafb20fa 6993 rdev_for_each(rdev, mddev) {
1da177e4 6994 raid_disk = rdev->raid_disk;
5e5e3e78 6995 if (raid_disk >= max_disks
f2076e7d 6996 || raid_disk < 0 || test_bit(Journal, &rdev->flags))
1da177e4
LT		 6997 continue;
6998 disk = conf->disks + raid_disk;
6999
17045f52
N		 7000 if (test_bit(Replacement, &rdev->flags)) {
7001 if (disk->replacement)
7002 goto abort;
7003 disk->replacement = rdev;
7004 } else {
7005 if (disk->rdev)
7006 goto abort;
7007 disk->rdev = rdev;
7008 }
1da177e4 7009
b2d444d7 7010 if (test_bit(In_sync, &rdev->flags)) {
1da177e4 7011 char b[BDEVNAME_SIZE];
cc6167b4
N		 7012 pr_info("md/raid:%s: device %s operational as raid disk %d\n",
7013 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
d6b212f4 7014 } else if (rdev->saved_raid_disk != raid_disk)
8c2e870a
NB		 7015 /* Cannot rely on bitmap to complete recovery */
7016 conf->fullsync = 1;
1da177e4
LT		 7017 }
7018
91adb564 7019 conf->level = mddev->new_level;
584acdd4 7020 if (conf->level == 6) {
16a53ecc 7021 conf->max_degraded = 2;
584acdd4
MS		 7022 if (raid6_call.xor_syndrome)
7023 conf->rmw_level = PARITY_ENABLE_RMW;
7024 else
7025 conf->rmw_level = PARITY_DISABLE_RMW;
7026 } else {
16a53ecc 7027 conf->max_degraded = 1;
584acdd4
MS		 7028 conf->rmw_level = PARITY_ENABLE_RMW;
7029 }
91adb564 7030 conf->algorithm = mddev->new_layout;
fef9c61f 7031 conf->reshape_progress = mddev->reshape_position;
e183eaed 7032 if (conf->reshape_progress != MaxSector) {
09c9e5fa 7033 conf->prev_chunk_sectors = mddev->chunk_sectors;
e183eaed 7034 conf->prev_algo = mddev->layout;
5cac6bcb
N		 7035 } else {
7036 conf->prev_chunk_sectors = conf->chunk_sectors;
7037 conf->prev_algo = conf->algorithm;
e183eaed 7038 }
1da177e4 7039
edbe83ab 7040 conf->min_nr_stripes = NR_STRIPES;
ad5b0f76
SL		 7041 if (mddev->reshape_position != MaxSector) {
7042 int stripes = max_t(int,
7043 ((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4,
7044 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4);
7045 conf->min_nr_stripes = max(NR_STRIPES, stripes);
7046 if (conf->min_nr_stripes != NR_STRIPES)
cc6167b4 7047 pr_info("md/raid:%s: force stripe size %d for reshape\n",
ad5b0f76
SL		 7048 mdname(mddev), conf->min_nr_stripes);
7049 }
edbe83ab 7050 memory = conf->min_nr_stripes * (sizeof(struct stripe_head) +
5e5e3e78 7051 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
4bda556a 7052 atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS);
edbe83ab 7053 if (grow_stripes(conf, conf->min_nr_stripes)) {
cc6167b4
N		 7054 pr_warn("md/raid:%s: couldn't allocate %dkB for buffers\n",
7055 mdname(mddev), memory);
91adb564
N		 7056 goto abort;
7057 } else
cc6167b4 7058 pr_debug("md/raid:%s: allocated %dkB\n", mdname(mddev), memory);
edbe83ab
N		 7059 /*
7060 * Losing a stripe head costs more than the time to refill it,
7061 * it reduces the queue depth and so can hurt throughput.
7062 * So set it rather large, scaled by number of devices.
7063 */
7064 conf->shrinker.seeks = DEFAULT_SEEKS * conf->raid_disks * 4;
7065 conf->shrinker.scan_objects = raid5_cache_scan;
7066 conf->shrinker.count_objects = raid5_cache_count;
7067 conf->shrinker.batch = 128;
7068 conf->shrinker.flags = 0;
6a0f53ff 7069 if (register_shrinker(&conf->shrinker)) {
cc6167b4
N		 7070 pr_warn("md/raid:%s: couldn't register shrinker.\n",
7071 mdname(mddev));
6a0f53ff
CY		 7072 goto abort;
7073 }
1da177e4 7074
0232605d
N		 7075 sprintf(pers_name, "raid%d", mddev->new_level);
7076 conf->thread = md_register_thread(raid5d, mddev, pers_name);
91adb564 7077 if (!conf->thread) {
cc6167b4
N		 7078 pr_warn("md/raid:%s: couldn't allocate thread.\n",
7079 mdname(mddev));
16a53ecc
N		 7080 goto abort;
7081 }
91adb564
N		 7082
7083 return conf;
7084
7085 abort:
7086 if (conf) {
95fc17aa 7087 free_conf(conf);
91adb564
N		 7088 return ERR_PTR(-EIO);
7089 } else
7090 return ERR_PTR(-ENOMEM);
7091}
7092
c148ffdc
N		 7093static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
7094{
7095 switch (algo) {
7096 case ALGORITHM_PARITY_0:
7097 if (raid_disk < max_degraded)
7098 return 1;
7099 break;
7100 case ALGORITHM_PARITY_N:
7101 if (raid_disk >= raid_disks - max_degraded)
7102 return 1;
7103 break;
7104 case ALGORITHM_PARITY_0_6:
f72ffdd6 7105 if (raid_disk == 0 ||
c148ffdc
N		 7106 raid_disk == raid_disks - 1)
7107 return 1;
7108 break;
7109 case ALGORITHM_LEFT_ASYMMETRIC_6:
7110 case ALGORITHM_RIGHT_ASYMMETRIC_6:
7111 case ALGORITHM_LEFT_SYMMETRIC_6:
7112 case ALGORITHM_RIGHT_SYMMETRIC_6:
7113 if (raid_disk == raid_disks - 1)
7114 return 1;
7115 }
7116 return 0;
7117}
7118
849674e4 7119static int raid5_run(struct mddev *mddev)
91adb564 7120{
d1688a6d 7121 struct r5conf *conf;
9f7c2220 7122 int working_disks = 0;
c148ffdc 7123 int dirty_parity_disks = 0;
3cb03002 7124 struct md_rdev *rdev;
713cf5a6 7125 struct md_rdev *journal_dev = NULL;
c148ffdc 7126 sector_t reshape_offset = 0;
17045f52 7127 int i;
b5254dd5
N		 7128 long long min_offset_diff = 0;
7129 int first = 1;
91adb564 7130
a415c0f1
N		 7131 if (mddev_init_writes_pending(mddev) < 0)
7132 return -ENOMEM;
7133
8c6ac868 7134 if (mddev->recovery_cp != MaxSector)
cc6167b4
N		 7135 pr_notice("md/raid:%s: not clean -- starting background reconstruction\n",
7136 mdname(mddev));
b5254dd5
N		 7137
7138 rdev_for_each(rdev, mddev) {
7139 long long diff;
713cf5a6 7140
f2076e7d 7141 if (test_bit(Journal, &rdev->flags)) {
713cf5a6 7142 journal_dev = rdev;
f2076e7d
SL		 7143 continue;
7144 }
b5254dd5
N		 7145 if (rdev->raid_disk < 0)
7146 continue;
7147 diff = (rdev->new_data_offset - rdev->data_offset);
7148 if (first) {
7149 min_offset_diff = diff;
7150 first = 0;
7151 } else if (mddev->reshape_backwards &&
7152 diff < min_offset_diff)
7153 min_offset_diff = diff;
7154 else if (!mddev->reshape_backwards &&
7155 diff > min_offset_diff)
7156 min_offset_diff = diff;
7157 }
7158
91adb564
N		 7159 if (mddev->reshape_position != MaxSector) {
7160 /* Check that we can continue the reshape.
b5254dd5
N		 7161 * Difficulties arise if the stripe we would write to
7162 * next is at or after the stripe we would read from next.
7163 * For a reshape that changes the number of devices, this
7164 * is only possible for a very short time, and mdadm makes
7165 * sure that time appears to have past before assembling
7166 * the array. So we fail if that time hasn't passed.
7167 * For a reshape that keeps the number of devices the same
7168 * mdadm must be monitoring the reshape can keeping the
7169 * critical areas read-only and backed up. It will start
7170 * the array in read-only mode, so we check for that.
91adb564
N		 7171 */
7172 sector_t here_new, here_old;
7173 int old_disks;
18b00334 7174 int max_degraded = (mddev->level == 6 ? 2 : 1);
05256d98
N		 7175 int chunk_sectors;
7176 int new_data_disks;
91adb564 7177
713cf5a6 7178 if (journal_dev) {
cc6167b4
N		 7179 pr_warn("md/raid:%s: don't support reshape with journal - aborting.\n",
7180 mdname(mddev));
713cf5a6
SL		 7181 return -EINVAL;
7182 }
7183
88ce4930 7184 if (mddev->new_level != mddev->level) {
cc6167b4
N		 7185 pr_warn("md/raid:%s: unsupported reshape required - aborting.\n",
7186 mdname(mddev));
91adb564
N		 7187 return -EINVAL;
7188 }
91adb564
N		 7189 old_disks = mddev->raid_disks - mddev->delta_disks;
7190 /* reshape_position must be on a new-stripe boundary, and one
7191 * further up in new geometry must map after here in old
7192 * geometry.
05256d98
N		 7193 * If the chunk sizes are different, then as we perform reshape
7194 * in units of the largest of the two, reshape_position needs
7195 * be a multiple of the largest chunk size times new data disks.
91adb564
N		 7196 */
7197 here_new = mddev->reshape_position;
05256d98
N		 7198 chunk_sectors = max(mddev->chunk_sectors, mddev->new_chunk_sectors);
7199 new_data_disks = mddev->raid_disks - max_degraded;
7200 if (sector_div(here_new, chunk_sectors * new_data_disks)) {
cc6167b4
N		 7201 pr_warn("md/raid:%s: reshape_position not on a stripe boundary\n",
7202 mdname(mddev));
91adb564
N		 7203 return -EINVAL;
7204 }
05256d98 7205 reshape_offset = here_new * chunk_sectors;
91adb564
N		 7206 /* here_new is the stripe we will write to */
7207 here_old = mddev->reshape_position;
05256d98 7208 sector_div(here_old, chunk_sectors * (old_disks-max_degraded));
91adb564
N		 7209 /* here_old is the first stripe that we might need to read
7210 * from */
67ac6011
N		 7211 if (mddev->delta_disks == 0) {
7212 /* We cannot be sure it is safe to start an in-place
b5254dd5 7213 * reshape. It is only safe if user-space is monitoring
67ac6011
N		 7214 * and taking constant backups.
7215 * mdadm always starts a situation like this in
7216 * readonly mode so it can take control before
7217 * allowing any writes. So just check for that.
7218 */
b5254dd5
N		 7219 if (abs(min_offset_diff) >= mddev->chunk_sectors &&
7220 abs(min_offset_diff) >= mddev->new_chunk_sectors)
7221 /* not really in-place - so OK */;
7222 else if (mddev->ro == 0) {
cc6167b4
N		 7223 pr_warn("md/raid:%s: in-place reshape must be started in read-only mode - aborting\n",
7224 mdname(mddev));
67ac6011
N		 7225 return -EINVAL;
7226 }
2c810cdd 7227 } else if (mddev->reshape_backwards
05256d98
N		 7228 ? (here_new * chunk_sectors + min_offset_diff <=
7229 here_old * chunk_sectors)
7230 : (here_new * chunk_sectors >=
7231 here_old * chunk_sectors + (-min_offset_diff))) {
91adb564 7232 /* Reading from the same stripe as writing to - bad */
cc6167b4
N		 7233 pr_warn("md/raid:%s: reshape_position too early for auto-recovery - aborting.\n",
7234 mdname(mddev));
91adb564
N		 7235 return -EINVAL;
7236 }
cc6167b4 7237 pr_debug("md/raid:%s: reshape will continue\n", mdname(mddev));
91adb564
N		 7238 /* OK, we should be able to continue; */
7239 } else {
7240 BUG_ON(mddev->level != mddev->new_level);
7241 BUG_ON(mddev->layout != mddev->new_layout);
664e7c41 7242 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
91adb564 7243 BUG_ON(mddev->delta_disks != 0);
1da177e4 7244 }
91adb564 7245
3418d036
AP		 7246 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) &&
7247 test_bit(MD_HAS_PPL, &mddev->flags)) {
7248 pr_warn("md/raid:%s: using journal device and PPL not allowed - disabling PPL\n",
7249 mdname(mddev));
7250 clear_bit(MD_HAS_PPL, &mddev->flags);
ddc08823 7251 clear_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags);
3418d036
AP		 7252 }
7253
245f46c2
N		 7254 if (mddev->private == NULL)
7255 conf = setup_conf(mddev);
7256 else
7257 conf = mddev->private;
7258
91adb564
N		 7259 if (IS_ERR(conf))
7260 return PTR_ERR(conf);
7261
486b0f7b
SL		 7262 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
7263 if (!journal_dev) {
cc6167b4
N		 7264 pr_warn("md/raid:%s: journal disk is missing, force array readonly\n",
7265 mdname(mddev));
486b0f7b
SL		 7266 mddev->ro = 1;
7267 set_disk_ro(mddev->gendisk, 1);
7268 } else if (mddev->recovery_cp == MaxSector)
7269 set_bit(MD_JOURNAL_CLEAN, &mddev->flags);
7dde2ad3
SL		 7270 }
7271
b5254dd5 7272 conf->min_offset_diff = min_offset_diff;
91adb564
N		 7273 mddev->thread = conf->thread;
7274 conf->thread = NULL;
7275 mddev->private = conf;
7276
17045f52
N		 7277 for (i = 0; i < conf->raid_disks && conf->previous_raid_disks;
7278 i++) {
7279 rdev = conf->disks[i].rdev;
7280 if (!rdev && conf->disks[i].replacement) {
7281 /* The replacement is all we have yet */
7282 rdev = conf->disks[i].replacement;
7283 conf->disks[i].replacement = NULL;
7284 clear_bit(Replacement, &rdev->flags);
7285 conf->disks[i].rdev = rdev;
7286 }
7287 if (!rdev)
c148ffdc 7288 continue;
17045f52
N		 7289 if (conf->disks[i].replacement &&
7290 conf->reshape_progress != MaxSector) {
7291 /* replacements and reshape simply do not mix. */
cc6167b4 7292 pr_warn("md: cannot handle concurrent replacement and reshape.\n");
17045f52
N		 7293 goto abort;
7294 }
2f115882 7295 if (test_bit(In_sync, &rdev->flags)) {
91adb564 7296 working_disks++;
2f115882
N		 7297 continue;
7298 }
c148ffdc
N		 7299 /* This disc is not fully in-sync. However if it
7300 * just stored parity (beyond the recovery_offset),
7301 * when we don't need to be concerned about the
7302 * array being dirty.
7303 * When reshape goes 'backwards', we never have
7304 * partially completed devices, so we only need
7305 * to worry about reshape going forwards.
7306 */
7307 /* Hack because v0.91 doesn't store recovery_offset properly. */
7308 if (mddev->major_version == 0 &&
7309 mddev->minor_version > 90)
7310 rdev->recovery_offset = reshape_offset;
5026d7a9 7311
c148ffdc
N		 7312 if (rdev->recovery_offset < reshape_offset) {
7313 /* We need to check old and new layout */
7314 if (!only_parity(rdev->raid_disk,
7315 conf->algorithm,
7316 conf->raid_disks,
7317 conf->max_degraded))
7318 continue;
7319 }
7320 if (!only_parity(rdev->raid_disk,
7321 conf->prev_algo,
7322 conf->previous_raid_disks,
7323 conf->max_degraded))
7324 continue;
7325 dirty_parity_disks++;
7326 }
91adb564 7327
17045f52
N		 7328 /*
7329 * 0 for a fully functional array, 1 or 2 for a degraded array.
7330 */
2e38a37f 7331 mddev->degraded = raid5_calc_degraded(conf);
91adb564 7332
674806d6 7333 if (has_failed(conf)) {
cc6167b4 7334 pr_crit("md/raid:%s: not enough operational devices (%d/%d failed)\n",
02c2de8c 7335 mdname(mddev), mddev->degraded, conf->raid_disks);
1da177e4
LT		 7336 goto abort;
7337 }
7338
91adb564 7339 /* device size must be a multiple of chunk size */
9d8f0363 7340 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
91adb564
N		 7341 mddev->resync_max_sectors = mddev->dev_sectors;
7342
c148ffdc 7343 if (mddev->degraded > dirty_parity_disks &&
1da177e4 7344 mddev->recovery_cp != MaxSector) {
4536bf9b
AP		 7345 if (test_bit(MD_HAS_PPL, &mddev->flags))
7346 pr_crit("md/raid:%s: starting dirty degraded array with PPL.\n",
7347 mdname(mddev));
7348 else if (mddev->ok_start_degraded)
cc6167b4
N		 7349 pr_crit("md/raid:%s: starting dirty degraded array - data corruption possible.\n",
7350 mdname(mddev));
6ff8d8ec 7351 else {
cc6167b4
N		 7352 pr_crit("md/raid:%s: cannot start dirty degraded array.\n",
7353 mdname(mddev));
6ff8d8ec
N		 7354 goto abort;
7355 }
1da177e4
LT		 7356 }
7357
cc6167b4
N		 7358 pr_info("md/raid:%s: raid level %d active with %d out of %d devices, algorithm %d\n",
7359 mdname(mddev), conf->level,
7360 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
7361 mddev->new_layout);
1da177e4
LT		 7362
7363 print_raid5_conf(conf);
7364
fef9c61f 7365 if (conf->reshape_progress != MaxSector) {
fef9c61f 7366 conf->reshape_safe = conf->reshape_progress;
f6705578
N		 7367 atomic_set(&conf->reshape_stripes, 0);
7368 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7369 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7370 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7371 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7372 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 7373 "reshape");
f6705578
N		 7374 }
7375
1da177e4 7376 /* Ok, everything is just fine now */
a64c876f
N		 7377 if (mddev->to_remove == &raid5_attrs_group)
7378 mddev->to_remove = NULL;
00bcb4ac
N		 7379 else if (mddev->kobj.sd &&
7380 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
cc6167b4
N		 7381 pr_warn("raid5: failed to create sysfs attributes for %s\n",
7382 mdname(mddev));
4a5add49 7383 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
7a5febe9 7384
4a5add49 7385 if (mddev->queue) {
9f7c2220 7386 int chunk_size;
4a5add49
N		 7387 /* read-ahead size must cover two whole stripes, which
7388 * is 2 * (datadisks) * chunksize where 'n' is the
7389 * number of raid devices
7390 */
7391 int data_disks = conf->previous_raid_disks - conf->max_degraded;
7392 int stripe = data_disks *
7393 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
dc3b17cc
JK		 7394 if (mddev->queue->backing_dev_info->ra_pages < 2 * stripe)
7395 mddev->queue->backing_dev_info->ra_pages = 2 * stripe;
91adb564 7396
9f7c2220
N		 7397 chunk_size = mddev->chunk_sectors << 9;
7398 blk_queue_io_min(mddev->queue, chunk_size);
7399 blk_queue_io_opt(mddev->queue, chunk_size *
7400 (conf->raid_disks - conf->max_degraded));
c78afc62 7401 mddev->queue->limits.raid_partial_stripes_expensive = 1;
620125f2
SL		 7402 /*
7403 * We can only discard a whole stripe. It doesn't make sense to
7404 * discard data disk but write parity disk
7405 */
7406 stripe = stripe * PAGE_SIZE;
4ac6875e
N		 7407 /* Round up to power of 2, as discard handling
7408 * currently assumes that */
7409 while ((stripe-1) & stripe)
7410 stripe = (stripe | (stripe-1)) + 1;
620125f2
SL		 7411 mddev->queue->limits.discard_alignment = stripe;
7412 mddev->queue->limits.discard_granularity = stripe;
e8d7c332 7413
5026d7a9 7414 blk_queue_max_write_same_sectors(mddev->queue, 0);
3deff1a7 7415 blk_queue_max_write_zeroes_sectors(mddev->queue, 0);
5026d7a9 7416
05616be5 7417 rdev_for_each(rdev, mddev) {
9f7c2220
N		 7418 disk_stack_limits(mddev->gendisk, rdev->bdev,
7419 rdev->data_offset << 9);
05616be5
N		 7420 disk_stack_limits(mddev->gendisk, rdev->bdev,
7421 rdev->new_data_offset << 9);
7422 }
620125f2 7423
48920ff2
CH		 7424 /*
7425 * zeroing is required, otherwise data
7426 * could be lost. Consider a scenario: discard a stripe
7427 * (the stripe could be inconsistent if
7428 * discard_zeroes_data is 0); write one disk of the
7429 * stripe (the stripe could be inconsistent again
7430 * depending on which disks are used to calculate
7431 * parity); the disk is broken; The stripe data of this
7432 * disk is lost.
7433 *
7434 * We only allow DISCARD if the sysadmin has confirmed that
7435 * only safe devices are in use by setting a module parameter.
7436 * A better idea might be to turn DISCARD into WRITE_ZEROES
7437 * requests, as that is required to be safe.
7438 */
7439 if (devices_handle_discard_safely &&
e7597e69
JS		 7440 mddev->queue->limits.max_discard_sectors >= (stripe >> 9) &&
7441 mddev->queue->limits.discard_granularity >= stripe)
620125f2
SL		 7442 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
7443 mddev->queue);
7444 else
7445 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
7446 mddev->queue);
1dffdddd
SL		 7447
7448 blk_queue_max_hw_sectors(mddev->queue, UINT_MAX);
9f7c2220 7449 }
23032a0e 7450
845b9e22 7451 if (log_init(conf, journal_dev, raid5_has_ppl(conf)))
ff875738 7452 goto abort;
5c7e81c3 7453
1da177e4
LT		 7454 return 0;
7455abort:
01f96c0a 7456 md_unregister_thread(&mddev->thread);
e4f869d9
N		 7457 print_raid5_conf(conf);
7458 free_conf(conf);
1da177e4 7459 mddev->private = NULL;
cc6167b4 7460 pr_warn("md/raid:%s: failed to run raid set.\n", mdname(mddev));
1da177e4
LT		 7461 return -EIO;
7462}
7463
afa0f557 7464static void raid5_free(struct mddev *mddev, void *priv)
1da177e4 7465{
afa0f557 7466 struct r5conf *conf = priv;
1da177e4 7467
95fc17aa 7468 free_conf(conf);
a64c876f 7469 mddev->to_remove = &raid5_attrs_group;
1da177e4
LT		 7470}
7471
849674e4 7472static void raid5_status(struct seq_file *seq, struct mddev *mddev)
1da177e4 7473{
d1688a6d 7474 struct r5conf *conf = mddev->private;
1da177e4
LT		 7475 int i;
7476
9d8f0363 7477 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
3cb5edf4 7478 conf->chunk_sectors / 2, mddev->layout);
02c2de8c 7479 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
5fd13351
N		 7480 rcu_read_lock();
7481 for (i = 0; i < conf->raid_disks; i++) {
7482 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
7483 seq_printf (seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
7484 }
7485 rcu_read_unlock();
1da177e4 7486 seq_printf (seq, "]");
1da177e4
LT		 7487}
7488
d1688a6d 7489static void print_raid5_conf (struct r5conf *conf)
1da177e4
LT		 7490{
7491 int i;
7492 struct disk_info *tmp;
7493
cc6167b4 7494 pr_debug("RAID conf printout:\n");
1da177e4 7495 if (!conf) {
cc6167b4 7496 pr_debug("(conf==NULL)\n");
1da177e4
LT		 7497 return;
7498 }
cc6167b4 7499 pr_debug(" --- level:%d rd:%d wd:%d\n", conf->level,
0c55e022
N		 7500 conf->raid_disks,
7501 conf->raid_disks - conf->mddev->degraded);
1da177e4
LT		 7502
7503 for (i = 0; i < conf->raid_disks; i++) {
7504 char b[BDEVNAME_SIZE];
7505 tmp = conf->disks + i;
7506 if (tmp->rdev)
cc6167b4 7507 pr_debug(" disk %d, o:%d, dev:%s\n",
0c55e022
N		 7508 i, !test_bit(Faulty, &tmp->rdev->flags),
7509 bdevname(tmp->rdev->bdev, b));
1da177e4
LT		 7510 }
7511}
7512
fd01b88c 7513static int raid5_spare_active(struct mddev *mddev)
1da177e4
LT		 7514{
7515 int i;
d1688a6d 7516 struct r5conf *conf = mddev->private;
1da177e4 7517 struct disk_info *tmp;
6b965620
N		 7518 int count = 0;
7519 unsigned long flags;
1da177e4
LT		 7520
7521 for (i = 0; i < conf->raid_disks; i++) {
7522 tmp = conf->disks + i;
dd054fce
N		 7523 if (tmp->replacement
7524 && tmp->replacement->recovery_offset == MaxSector
7525 && !test_bit(Faulty, &tmp->replacement->flags)
7526 && !test_and_set_bit(In_sync, &tmp->replacement->flags)) {
7527 /* Replacement has just become active. */
7528 if (!tmp->rdev
7529 || !test_and_clear_bit(In_sync, &tmp->rdev->flags))
7530 count++;
7531 if (tmp->rdev) {
7532 /* Replaced device not technically faulty,
7533 * but we need to be sure it gets removed
7534 * and never re-added.
7535 */
7536 set_bit(Faulty, &tmp->rdev->flags);
7537 sysfs_notify_dirent_safe(
7538 tmp->rdev->sysfs_state);
7539 }
7540 sysfs_notify_dirent_safe(tmp->replacement->sysfs_state);
7541 } else if (tmp->rdev
70fffd0b 7542 && tmp->rdev->recovery_offset == MaxSector
b2d444d7 7543 && !test_bit(Faulty, &tmp->rdev->flags)
c04be0aa 7544 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
6b965620 7545 count++;
43c73ca4 7546 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
1da177e4
LT		 7547 }
7548 }
6b965620 7549 spin_lock_irqsave(&conf->device_lock, flags);
2e38a37f 7550 mddev->degraded = raid5_calc_degraded(conf);
6b965620 7551 spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4 7552 print_raid5_conf(conf);
6b965620 7553 return count;
1da177e4
LT		 7554}
7555
b8321b68 7556static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7557{
d1688a6d 7558 struct r5conf *conf = mddev->private;
1da177e4 7559 int err = 0;
b8321b68 7560 int number = rdev->raid_disk;
657e3e4d 7561 struct md_rdev **rdevp;
1da177e4
LT		 7562 struct disk_info *p = conf->disks + number;
7563
7564 print_raid5_conf(conf);
f6b6ec5c 7565 if (test_bit(Journal, &rdev->flags) && conf->log) {
c2bb6242 7566 /*
f6b6ec5c
SL		 7567 * we can't wait pending write here, as this is called in
7568 * raid5d, wait will deadlock.
84dd97a6
N		 7569 * neilb: there is no locking about new writes here,
7570 * so this cannot be safe.
c2bb6242 7571 */
70d466f7
SL		 7572 if (atomic_read(&conf->active_stripes) ||
7573 atomic_read(&conf->r5c_cached_full_stripes) ||
7574 atomic_read(&conf->r5c_cached_partial_stripes)) {
f6b6ec5c 7575 return -EBUSY;
84dd97a6 7576 }
ff875738 7577 log_exit(conf);
f6b6ec5c 7578 return 0;
c2bb6242 7579 }
657e3e4d
N		 7580 if (rdev == p->rdev)
7581 rdevp = &p->rdev;
7582 else if (rdev == p->replacement)
7583 rdevp = &p->replacement;
7584 else
7585 return 0;
7586
7587 if (number >= conf->raid_disks &&
7588 conf->reshape_progress == MaxSector)
7589 clear_bit(In_sync, &rdev->flags);
7590
7591 if (test_bit(In_sync, &rdev->flags) ||
7592 atomic_read(&rdev->nr_pending)) {
7593 err = -EBUSY;
7594 goto abort;
7595 }
7596 /* Only remove non-faulty devices if recovery
7597 * isn't possible.
7598 */
7599 if (!test_bit(Faulty, &rdev->flags) &&
7600 mddev->recovery_disabled != conf->recovery_disabled &&
7601 !has_failed(conf) &&
dd054fce 7602 (!p->replacement || p->replacement == rdev) &&
657e3e4d
N		 7603 number < conf->raid_disks) {
7604 err = -EBUSY;
7605 goto abort;
7606 }
7607 *rdevp = NULL;
d787be40
N		 7608 if (!test_bit(RemoveSynchronized, &rdev->flags)) {
7609 synchronize_rcu();
7610 if (atomic_read(&rdev->nr_pending)) {
7611 /* lost the race, try later */
7612 err = -EBUSY;
7613 *rdevp = rdev;
7614 }
7615 }
6358c239
AP		 7616 if (!err) {
7617 err = log_modify(conf, rdev, false);
7618 if (err)
7619 goto abort;
7620 }
d787be40 7621 if (p->replacement) {
dd054fce
N		 7622 /* We must have just cleared 'rdev' */
7623 p->rdev = p->replacement;
7624 clear_bit(Replacement, &p->replacement->flags);
7625 smp_mb(); /* Make sure other CPUs may see both as identical
7626 * but will never see neither - if they are careful
7627 */
7628 p->replacement = NULL;
6358c239
AP		 7629
7630 if (!err)
7631 err = log_modify(conf, p->rdev, true);
e5bc9c3c
GJ		 7632 }
7633
7634 clear_bit(WantReplacement, &rdev->flags);
1da177e4
LT		 7635abort:
7636
7637 print_raid5_conf(conf);
7638 return err;
7639}
7640
fd01b88c 7641static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7642{
d1688a6d 7643 struct r5conf *conf = mddev->private;
199050ea 7644 int err = -EEXIST;
1da177e4
LT		 7645 int disk;
7646 struct disk_info *p;
6c2fce2e
NB		 7647 int first = 0;
7648 int last = conf->raid_disks - 1;
1da177e4 7649
f6b6ec5c 7650 if (test_bit(Journal, &rdev->flags)) {
f6b6ec5c
SL		 7651 if (conf->log)
7652 return -EBUSY;
7653
7654 rdev->raid_disk = 0;
7655 /*
7656 * The array is in readonly mode if journal is missing, so no
7657 * write requests running. We should be safe
7658 */
845b9e22 7659 log_init(conf, rdev, false);
f6b6ec5c
SL		 7660 return 0;
7661 }
7f0da59b
N		 7662 if (mddev->recovery_disabled == conf->recovery_disabled)
7663 return -EBUSY;
7664
dc10c643 7665 if (rdev->saved_raid_disk < 0 && has_failed(conf))
1da177e4 7666 /* no point adding a device */
199050ea 7667 return -EINVAL;
1da177e4 7668
6c2fce2e
NB		 7669 if (rdev->raid_disk >= 0)
7670 first = last = rdev->raid_disk;
1da177e4
LT		 7671
7672 /*
16a53ecc
N		 7673 * find the disk ... but prefer rdev->saved_raid_disk
7674 * if possible.
1da177e4 7675 */
16a53ecc 7676 if (rdev->saved_raid_disk >= 0 &&
6c2fce2e 7677 rdev->saved_raid_disk >= first &&
16a53ecc 7678 conf->disks[rdev->saved_raid_disk].rdev == NULL)
5cfb22a1
N		 7679 first = rdev->saved_raid_disk;
7680
7681 for (disk = first; disk <= last; disk++) {
7bfec5f3
N		 7682 p = conf->disks + disk;
7683 if (p->rdev == NULL) {
b2d444d7 7684 clear_bit(In_sync, &rdev->flags);
1da177e4 7685 rdev->raid_disk = disk;
72626685
N		 7686 if (rdev->saved_raid_disk != disk)
7687 conf->fullsync = 1;
d6065f7b 7688 rcu_assign_pointer(p->rdev, rdev);
6358c239
AP		 7689
7690 err = log_modify(conf, rdev, true);
7691
5cfb22a1 7692 goto out;
1da177e4 7693 }
5cfb22a1
N		 7694 }
7695 for (disk = first; disk <= last; disk++) {
7696 p = conf->disks + disk;
7bfec5f3
N		 7697 if (test_bit(WantReplacement, &p->rdev->flags) &&
7698 p->replacement == NULL) {
7699 clear_bit(In_sync, &rdev->flags);
7700 set_bit(Replacement, &rdev->flags);
7701 rdev->raid_disk = disk;
7702 err = 0;
7703 conf->fullsync = 1;
7704 rcu_assign_pointer(p->replacement, rdev);
7705 break;
7706 }
7707 }
5cfb22a1 7708out:
1da177e4 7709 print_raid5_conf(conf);
199050ea 7710 return err;
1da177e4
LT		 7711}
7712
fd01b88c 7713static int raid5_resize(struct mddev *mddev, sector_t sectors)
1da177e4
LT		 7714{
7715 /* no resync is happening, and there is enough space
7716 * on all devices, so we can resize.
7717 * We need to make sure resync covers any new space.
7718 * If the array is shrinking we should possibly wait until
7719 * any io in the removed space completes, but it hardly seems
7720 * worth it.
7721 */
a4a6125a 7722 sector_t newsize;
3cb5edf4
N		 7723 struct r5conf *conf = mddev->private;
7724
3418d036 7725 if (conf->log || raid5_has_ppl(conf))
713cf5a6 7726 return -EINVAL;
3cb5edf4 7727 sectors &= ~((sector_t)conf->chunk_sectors - 1);
a4a6125a
N		 7728 newsize = raid5_size(mddev, sectors, mddev->raid_disks);
7729 if (mddev->external_size &&
7730 mddev->array_sectors > newsize)
b522adcd 7731 return -EINVAL;
a4a6125a
N		 7732 if (mddev->bitmap) {
7733 int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0);
7734 if (ret)
7735 return ret;
7736 }
7737 md_set_array_sectors(mddev, newsize);
b098636c
N		 7738 if (sectors > mddev->dev_sectors &&
7739 mddev->recovery_cp > mddev->dev_sectors) {
58c0fed4 7740 mddev->recovery_cp = mddev->dev_sectors;
1da177e4
LT		 7741 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7742 }
58c0fed4 7743 mddev->dev_sectors = sectors;
4b5c7ae8 7744 mddev->resync_max_sectors = sectors;
1da177e4
LT		 7745 return 0;
7746}
7747
fd01b88c 7748static int check_stripe_cache(struct mddev *mddev)
01ee22b4
N		 7749{
7750 /* Can only proceed if there are plenty of stripe_heads.
7751 * We need a minimum of one full stripe,, and for sensible progress
7752 * it is best to have about 4 times that.
7753 * If we require 4 times, then the default 256 4K stripe_heads will
7754 * allow for chunk sizes up to 256K, which is probably OK.
7755 * If the chunk size is greater, user-space should request more
7756 * stripe_heads first.
7757 */
d1688a6d 7758 struct r5conf *conf = mddev->private;
01ee22b4 7759 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
edbe83ab 7760 > conf->min_nr_stripes ||
01ee22b4 7761 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
edbe83ab 7762 > conf->min_nr_stripes) {
cc6167b4
N		 7763 pr_warn("md/raid:%s: reshape: not enough stripes. Needed %lu\n",
7764 mdname(mddev),
7765 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
7766 / STRIPE_SIZE)*4);
01ee22b4
N		 7767 return 0;
7768 }
7769 return 1;
7770}
7771
fd01b88c 7772static int check_reshape(struct mddev *mddev)
29269553 7773{
d1688a6d 7774 struct r5conf *conf = mddev->private;
29269553 7775
3418d036 7776 if (conf->log || raid5_has_ppl(conf))
713cf5a6 7777 return -EINVAL;
88ce4930
N		 7778 if (mddev->delta_disks == 0 &&
7779 mddev->new_layout == mddev->layout &&
664e7c41 7780 mddev->new_chunk_sectors == mddev->chunk_sectors)
50ac168a 7781 return 0; /* nothing to do */
674806d6 7782 if (has_failed(conf))
ec32a2bd 7783 return -EINVAL;
fdcfbbb6 7784 if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) {
ec32a2bd
N		 7785 /* We might be able to shrink, but the devices must
7786 * be made bigger first.
7787 * For raid6, 4 is the minimum size.
7788 * Otherwise 2 is the minimum
7789 */
7790 int min = 2;
7791 if (mddev->level == 6)
7792 min = 4;
7793 if (mddev->raid_disks + mddev->delta_disks < min)
7794 return -EINVAL;
7795 }
29269553 7796
01ee22b4 7797 if (!check_stripe_cache(mddev))
29269553 7798 return -ENOSPC;
29269553 7799
738a2738
N		 7800 if (mddev->new_chunk_sectors > mddev->chunk_sectors ||
7801 mddev->delta_disks > 0)
7802 if (resize_chunks(conf,
7803 conf->previous_raid_disks
7804 + max(0, mddev->delta_disks),
7805 max(mddev->new_chunk_sectors,
7806 mddev->chunk_sectors)
7807 ) < 0)
7808 return -ENOMEM;
845b9e22
AP		 7809
7810 if (conf->previous_raid_disks + mddev->delta_disks <= conf->pool_size)
7811 return 0; /* never bother to shrink */
e56108d6
N		 7812 return resize_stripes(conf, (conf->previous_raid_disks
7813 + mddev->delta_disks));
63c70c4f
N		 7814}
7815
fd01b88c 7816static int raid5_start_reshape(struct mddev *mddev)
63c70c4f 7817{
d1688a6d 7818 struct r5conf *conf = mddev->private;
3cb03002 7819 struct md_rdev *rdev;
63c70c4f 7820 int spares = 0;
c04be0aa 7821 unsigned long flags;
63c70c4f 7822
f416885e 7823 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
63c70c4f
N		 7824 return -EBUSY;
7825
01ee22b4
N		 7826 if (!check_stripe_cache(mddev))
7827 return -ENOSPC;
7828
30b67645
N		 7829 if (has_failed(conf))
7830 return -EINVAL;
7831
c6563a8c 7832 rdev_for_each(rdev, mddev) {
469518a3
N		 7833 if (!test_bit(In_sync, &rdev->flags)
7834 && !test_bit(Faulty, &rdev->flags))
29269553 7835 spares++;
c6563a8c 7836 }
63c70c4f 7837
f416885e 7838 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
29269553
N		 7839 /* Not enough devices even to make a degraded array
7840 * of that size
7841 */
7842 return -EINVAL;
7843
ec32a2bd
N		 7844 /* Refuse to reduce size of the array. Any reductions in
7845 * array size must be through explicit setting of array_size
7846 * attribute.
7847 */
7848 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
7849 < mddev->array_sectors) {
cc6167b4
N		 7850 pr_warn("md/raid:%s: array size must be reduced before number of disks\n",
7851 mdname(mddev));
ec32a2bd
N		 7852 return -EINVAL;
7853 }
7854
f6705578 7855 atomic_set(&conf->reshape_stripes, 0);
29269553 7856 spin_lock_irq(&conf->device_lock);
c46501b2 7857 write_seqcount_begin(&conf->gen_lock);
29269553 7858 conf->previous_raid_disks = conf->raid_disks;
63c70c4f 7859 conf->raid_disks += mddev->delta_disks;
09c9e5fa
AN		 7860 conf->prev_chunk_sectors = conf->chunk_sectors;
7861 conf->chunk_sectors = mddev->new_chunk_sectors;
88ce4930
N		 7862 conf->prev_algo = conf->algorithm;
7863 conf->algorithm = mddev->new_layout;
05616be5
N		 7864 conf->generation++;
7865 /* Code that selects data_offset needs to see the generation update
7866 * if reshape_progress has been set - so a memory barrier needed.
7867 */
7868 smp_mb();
2c810cdd 7869 if (mddev->reshape_backwards)
fef9c61f
N		 7870 conf->reshape_progress = raid5_size(mddev, 0, 0);
7871 else
7872 conf->reshape_progress = 0;
7873 conf->reshape_safe = conf->reshape_progress;
c46501b2 7874 write_seqcount_end(&conf->gen_lock);
29269553
N		 7875 spin_unlock_irq(&conf->device_lock);
7876
4d77e3ba
N		 7877 /* Now make sure any requests that proceeded on the assumption
7878 * the reshape wasn't running - like Discard or Read - have
7879 * completed.
7880 */
7881 mddev_suspend(mddev);
7882 mddev_resume(mddev);
7883
29269553
N		 7884 /* Add some new drives, as many as will fit.
7885 * We know there are enough to make the newly sized array work.
3424bf6a
N		 7886 * Don't add devices if we are reducing the number of
7887 * devices in the array. This is because it is not possible
7888 * to correctly record the "partially reconstructed" state of
7889 * such devices during the reshape and confusion could result.
29269553 7890 */
87a8dec9 7891 if (mddev->delta_disks >= 0) {
dafb20fa 7892 rdev_for_each(rdev, mddev)
87a8dec9
N		 7893 if (rdev->raid_disk < 0 &&
7894 !test_bit(Faulty, &rdev->flags)) {
7895 if (raid5_add_disk(mddev, rdev) == 0) {
87a8dec9 7896 if (rdev->raid_disk
9d4c7d87 7897 >= conf->previous_raid_disks)
87a8dec9 7898 set_bit(In_sync, &rdev->flags);
9d4c7d87 7899 else
87a8dec9 7900 rdev->recovery_offset = 0;
36fad858
NK		 7901
7902 if (sysfs_link_rdev(mddev, rdev))
87a8dec9 7903 /* Failure here is OK */;
50da0840 7904 }
87a8dec9
N		 7905 } else if (rdev->raid_disk >= conf->previous_raid_disks
7906 && !test_bit(Faulty, &rdev->flags)) {
7907 /* This is a spare that was manually added */
7908 set_bit(In_sync, &rdev->flags);
87a8dec9 7909 }
29269553 7910
87a8dec9
N		 7911 /* When a reshape changes the number of devices,
7912 * ->degraded is measured against the larger of the
7913 * pre and post number of devices.
7914 */
ec32a2bd 7915 spin_lock_irqsave(&conf->device_lock, flags);
2e38a37f 7916 mddev->degraded = raid5_calc_degraded(conf);
ec32a2bd
N		 7917 spin_unlock_irqrestore(&conf->device_lock, flags);
7918 }
63c70c4f 7919 mddev->raid_disks = conf->raid_disks;
e516402c 7920 mddev->reshape_position = conf->reshape_progress;
2953079c 7921 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
f6705578 7922
29269553
N		 7923 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7924 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
ea358cd0 7925 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
29269553
N		 7926 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7927 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7928 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 7929 "reshape");
29269553
N		 7930 if (!mddev->sync_thread) {
7931 mddev->recovery = 0;
7932 spin_lock_irq(&conf->device_lock);
ba8805b9 7933 write_seqcount_begin(&conf->gen_lock);
29269553 7934 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
ba8805b9
N		 7935 mddev->new_chunk_sectors =
7936 conf->chunk_sectors = conf->prev_chunk_sectors;
7937 mddev->new_layout = conf->algorithm = conf->prev_algo;
05616be5
N		 7938 rdev_for_each(rdev, mddev)
7939 rdev->new_data_offset = rdev->data_offset;
7940 smp_wmb();
ba8805b9 7941 conf->generation --;
fef9c61f 7942 conf->reshape_progress = MaxSector;
1e3fa9bd 7943 mddev->reshape_position = MaxSector;
ba8805b9 7944 write_seqcount_end(&conf->gen_lock);
29269553
N		 7945 spin_unlock_irq(&conf->device_lock);
7946 return -EAGAIN;
7947 }
c8f517c4 7948 conf->reshape_checkpoint = jiffies;
29269553
N		 7949 md_wakeup_thread(mddev->sync_thread);
7950 md_new_event(mddev);
7951 return 0;
7952}
29269553 7953
ec32a2bd
N		 7954/* This is called from the reshape thread and should make any
7955 * changes needed in 'conf'
7956 */
d1688a6d 7957static void end_reshape(struct r5conf *conf)
29269553 7958{
29269553 7959
f6705578 7960 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
f6705578 7961
f6705578 7962 spin_lock_irq(&conf->device_lock);
cea9c228 7963 conf->previous_raid_disks = conf->raid_disks;
b5d27718 7964 md_finish_reshape(conf->mddev);
05616be5 7965 smp_wmb();
fef9c61f 7966 conf->reshape_progress = MaxSector;
6cbd8148 7967 conf->mddev->reshape_position = MaxSector;
f6705578 7968 spin_unlock_irq(&conf->device_lock);
b0f9ec04 7969 wake_up(&conf->wait_for_overlap);
16a53ecc
N		 7970
7971 /* read-ahead size must cover two whole stripes, which is
7972 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
7973 */
4a5add49 7974 if (conf->mddev->queue) {
cea9c228 7975 int data_disks = conf->raid_disks - conf->max_degraded;
09c9e5fa 7976 int stripe = data_disks * ((conf->chunk_sectors << 9)
cea9c228 7977 / PAGE_SIZE);
dc3b17cc
JK		 7978 if (conf->mddev->queue->backing_dev_info->ra_pages < 2 * stripe)
7979 conf->mddev->queue->backing_dev_info->ra_pages = 2 * stripe;
16a53ecc 7980 }
29269553 7981 }
29269553
N		 7982}
7983
ec32a2bd
N		 7984/* This is called from the raid5d thread with mddev_lock held.
7985 * It makes config changes to the device.
7986 */
fd01b88c 7987static void raid5_finish_reshape(struct mddev *mddev)
cea9c228 7988{
d1688a6d 7989 struct r5conf *conf = mddev->private;
cea9c228
N		 7990
7991 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7992
ec32a2bd
N		 7993 if (mddev->delta_disks > 0) {
7994 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
fe67d19a
HM		 7995 if (mddev->queue) {
7996 set_capacity(mddev->gendisk, mddev->array_sectors);
7997 revalidate_disk(mddev->gendisk);
7998 }
ec32a2bd
N		 7999 } else {
8000 int d;
908f4fbd 8001 spin_lock_irq(&conf->device_lock);
2e38a37f 8002 mddev->degraded = raid5_calc_degraded(conf);
908f4fbd 8003 spin_unlock_irq(&conf->device_lock);
ec32a2bd
N		 8004 for (d = conf->raid_disks ;
8005 d < conf->raid_disks - mddev->delta_disks;
1a67dde0 8006 d++) {
3cb03002 8007 struct md_rdev *rdev = conf->disks[d].rdev;
da7613b8
N		 8008 if (rdev)
8009 clear_bit(In_sync, &rdev->flags);
8010 rdev = conf->disks[d].replacement;
8011 if (rdev)
8012 clear_bit(In_sync, &rdev->flags);
1a67dde0 8013 }
cea9c228 8014 }
88ce4930 8015 mddev->layout = conf->algorithm;
09c9e5fa 8016 mddev->chunk_sectors = conf->chunk_sectors;
ec32a2bd
N		 8017 mddev->reshape_position = MaxSector;
8018 mddev->delta_disks = 0;
2c810cdd 8019 mddev->reshape_backwards = 0;
cea9c228
N		 8020 }
8021}
8022
fd01b88c 8023static void raid5_quiesce(struct mddev *mddev, int state)
72626685 8024{
d1688a6d 8025 struct r5conf *conf = mddev->private;
72626685
N		 8026
8027 switch(state) {
e464eafd
N		 8028 case 2: /* resume for a suspend */
8029 wake_up(&conf->wait_for_overlap);
8030 break;
8031
72626685 8032 case 1: /* stop all writes */
566c09c5 8033 lock_all_device_hash_locks_irq(conf);
64bd660b
N		 8034 /* '2' tells resync/reshape to pause so that all
8035 * active stripes can drain
8036 */
a39f7afd 8037 r5c_flush_cache(conf, INT_MAX);
64bd660b 8038 conf->quiesce = 2;
b1b46486 8039 wait_event_cmd(conf->wait_for_quiescent,
46031f9a
RBJ		 8040 atomic_read(&conf->active_stripes) == 0 &&
8041 atomic_read(&conf->active_aligned_reads) == 0,
566c09c5
SL		 8042 unlock_all_device_hash_locks_irq(conf),
8043 lock_all_device_hash_locks_irq(conf));
64bd660b 8044 conf->quiesce = 1;
566c09c5 8045 unlock_all_device_hash_locks_irq(conf);
64bd660b
N		 8046 /* allow reshape to continue */
8047 wake_up(&conf->wait_for_overlap);
72626685
N		 8048 break;
8049
8050 case 0: /* re-enable writes */
566c09c5 8051 lock_all_device_hash_locks_irq(conf);
72626685 8052 conf->quiesce = 0;
b1b46486 8053 wake_up(&conf->wait_for_quiescent);
e464eafd 8054 wake_up(&conf->wait_for_overlap);
566c09c5 8055 unlock_all_device_hash_locks_irq(conf);
72626685
N		 8056 break;
8057 }
e6c033f7 8058 r5l_quiesce(conf->log, state);
72626685 8059}
b15c2e57 8060
fd01b88c 8061static void *raid45_takeover_raid0(struct mddev *mddev, int level)
54071b38 8062{
e373ab10 8063 struct r0conf *raid0_conf = mddev->private;
d76c8420 8064 sector_t sectors;
54071b38 8065
f1b29bca 8066 /* for raid0 takeover only one zone is supported */
e373ab10 8067 if (raid0_conf->nr_strip_zones > 1) {
cc6167b4
N		 8068 pr_warn("md/raid:%s: cannot takeover raid0 with more than one zone.\n",
8069 mdname(mddev));
f1b29bca
DW		 8070 return ERR_PTR(-EINVAL);
8071 }
8072
e373ab10
N		 8073 sectors = raid0_conf->strip_zone[0].zone_end;
8074 sector_div(sectors, raid0_conf->strip_zone[0].nb_dev);
3b71bd93 8075 mddev->dev_sectors = sectors;
f1b29bca 8076 mddev->new_level = level;
54071b38
TM		 8077 mddev->new_layout = ALGORITHM_PARITY_N;
8078 mddev->new_chunk_sectors = mddev->chunk_sectors;
8079 mddev->raid_disks += 1;
8080 mddev->delta_disks = 1;
8081 /* make sure it will be not marked as dirty */
8082 mddev->recovery_cp = MaxSector;
8083
8084 return setup_conf(mddev);
8085}
8086
fd01b88c 8087static void *raid5_takeover_raid1(struct mddev *mddev)
d562b0c4
N		 8088{
8089 int chunksect;
6995f0b2 8090 void *ret;
d562b0c4
N		 8091
8092 if (mddev->raid_disks != 2 ||
8093 mddev->degraded > 1)
8094 return ERR_PTR(-EINVAL);
8095
8096 /* Should check if there are write-behind devices? */
8097
8098 chunksect = 64*2; /* 64K by default */
8099
8100 /* The array must be an exact multiple of chunksize */
8101 while (chunksect && (mddev->array_sectors & (chunksect-1)))
8102 chunksect >>= 1;
8103
8104 if ((chunksect<<9) < STRIPE_SIZE)
8105 /* array size does not allow a suitable chunk size */
8106 return ERR_PTR(-EINVAL);
8107
8108 mddev->new_level = 5;
8109 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
664e7c41 8110 mddev->new_chunk_sectors = chunksect;
d562b0c4 8111
6995f0b2 8112 ret = setup_conf(mddev);
32cd7cbb 8113 if (!IS_ERR(ret))
394ed8e4
SL		 8114 mddev_clear_unsupported_flags(mddev,
8115 UNSUPPORTED_MDDEV_FLAGS);
6995f0b2 8116 return ret;
d562b0c4
N		 8117}
8118
fd01b88c 8119static void *raid5_takeover_raid6(struct mddev *mddev)
fc9739c6
N		 8120{
8121 int new_layout;
8122
8123 switch (mddev->layout) {
8124 case ALGORITHM_LEFT_ASYMMETRIC_6:
8125 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
8126 break;
8127 case ALGORITHM_RIGHT_ASYMMETRIC_6:
8128 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
8129 break;
8130 case ALGORITHM_LEFT_SYMMETRIC_6:
8131 new_layout = ALGORITHM_LEFT_SYMMETRIC;
8132 break;
8133 case ALGORITHM_RIGHT_SYMMETRIC_6:
8134 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
8135 break;
8136 case ALGORITHM_PARITY_0_6:
8137 new_layout = ALGORITHM_PARITY_0;
8138 break;
8139 case ALGORITHM_PARITY_N:
8140 new_layout = ALGORITHM_PARITY_N;
8141 break;
8142 default:
8143 return ERR_PTR(-EINVAL);
8144 }
8145 mddev->new_level = 5;
8146 mddev->new_layout = new_layout;
8147 mddev->delta_disks = -1;
8148 mddev->raid_disks -= 1;
8149 return setup_conf(mddev);
8150}
8151
fd01b88c 8152static int raid5_check_reshape(struct mddev *mddev)
b3546035 8153{
88ce4930
N		 8154 /* For a 2-drive array, the layout and chunk size can be changed
8155 * immediately as not restriping is needed.
8156 * For larger arrays we record the new value - after validation
8157 * to be used by a reshape pass.
b3546035 8158 */
d1688a6d 8159 struct r5conf *conf = mddev->private;
597a711b 8160 int new_chunk = mddev->new_chunk_sectors;
b3546035 8161
597a711b 8162 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
b3546035
N		 8163 return -EINVAL;
8164 if (new_chunk > 0) {
0ba459d2 8165 if (!is_power_of_2(new_chunk))
b3546035 8166 return -EINVAL;
597a711b 8167 if (new_chunk < (PAGE_SIZE>>9))
b3546035 8168 return -EINVAL;
597a711b 8169 if (mddev->array_sectors & (new_chunk-1))
b3546035
N		 8170 /* not factor of array size */
8171 return -EINVAL;
8172 }
8173
8174 /* They look valid */
8175
88ce4930 8176 if (mddev->raid_disks == 2) {
597a711b
N		 8177 /* can make the change immediately */
8178 if (mddev->new_layout >= 0) {
8179 conf->algorithm = mddev->new_layout;
8180 mddev->layout = mddev->new_layout;
88ce4930
N		 8181 }
8182 if (new_chunk > 0) {
597a711b
N		 8183 conf->chunk_sectors = new_chunk ;
8184 mddev->chunk_sectors = new_chunk;
88ce4930 8185 }
2953079c 8186 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
88ce4930 8187 md_wakeup_thread(mddev->thread);
b3546035 8188 }
50ac168a 8189 return check_reshape(mddev);
88ce4930
N		 8190}
8191
fd01b88c 8192static int raid6_check_reshape(struct mddev *mddev)
88ce4930 8193{
597a711b 8194 int new_chunk = mddev->new_chunk_sectors;
50ac168a 8195
597a711b 8196 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
88ce4930 8197 return -EINVAL;
b3546035 8198 if (new_chunk > 0) {
0ba459d2 8199 if (!is_power_of_2(new_chunk))
88ce4930 8200 return -EINVAL;
597a711b 8201 if (new_chunk < (PAGE_SIZE >> 9))
88ce4930 8202 return -EINVAL;
597a711b 8203 if (mddev->array_sectors & (new_chunk-1))
88ce4930
N		 8204 /* not factor of array size */
8205 return -EINVAL;
b3546035 8206 }
88ce4930
N		 8207
8208 /* They look valid */
50ac168a 8209 return check_reshape(mddev);
b3546035
N		 8210}
8211
fd01b88c 8212static void *raid5_takeover(struct mddev *mddev)
d562b0c4
N		 8213{
8214 /* raid5 can take over:
f1b29bca 8215 * raid0 - if there is only one strip zone - make it a raid4 layout
d562b0c4
N		 8216 * raid1 - if there are two drives. We need to know the chunk size
8217 * raid4 - trivial - just use a raid4 layout.
8218 * raid6 - Providing it is a *_6 layout
d562b0c4 8219 */
f1b29bca
DW		 8220 if (mddev->level == 0)
8221 return raid45_takeover_raid0(mddev, 5);
d562b0c4
N		 8222 if (mddev->level == 1)
8223 return raid5_takeover_raid1(mddev);
e9d4758f
N		 8224 if (mddev->level == 4) {
8225 mddev->new_layout = ALGORITHM_PARITY_N;
8226 mddev->new_level = 5;
8227 return setup_conf(mddev);
8228 }
fc9739c6
N		 8229 if (mddev->level == 6)
8230 return raid5_takeover_raid6(mddev);
d562b0c4
N		 8231
8232 return ERR_PTR(-EINVAL);
8233}
8234
fd01b88c 8235static void *raid4_takeover(struct mddev *mddev)
a78d38a1 8236{
f1b29bca
DW		 8237 /* raid4 can take over:
8238 * raid0 - if there is only one strip zone
8239 * raid5 - if layout is right
a78d38a1 8240 */
f1b29bca
DW		 8241 if (mddev->level == 0)
8242 return raid45_takeover_raid0(mddev, 4);
a78d38a1
N		 8243 if (mddev->level == 5 &&
8244 mddev->layout == ALGORITHM_PARITY_N) {
8245 mddev->new_layout = 0;
8246 mddev->new_level = 4;
8247 return setup_conf(mddev);
8248 }
8249 return ERR_PTR(-EINVAL);
8250}
d562b0c4 8251
84fc4b56 8252static struct md_personality raid5_personality;
245f46c2 8253
fd01b88c 8254static void *raid6_takeover(struct mddev *mddev)
245f46c2
N		 8255{
8256 /* Currently can only take over a raid5. We map the
8257 * personality to an equivalent raid6 personality
8258 * with the Q block at the end.
8259 */
8260 int new_layout;
8261
8262 if (mddev->pers != &raid5_personality)
8263 return ERR_PTR(-EINVAL);
8264 if (mddev->degraded > 1)
8265 return ERR_PTR(-EINVAL);
8266 if (mddev->raid_disks > 253)
8267 return ERR_PTR(-EINVAL);
8268 if (mddev->raid_disks < 3)
8269 return ERR_PTR(-EINVAL);
8270
8271 switch (mddev->layout) {
8272 case ALGORITHM_LEFT_ASYMMETRIC:
8273 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
8274 break;
8275 case ALGORITHM_RIGHT_ASYMMETRIC:
8276 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
8277 break;
8278 case ALGORITHM_LEFT_SYMMETRIC:
8279 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
8280 break;
8281 case ALGORITHM_RIGHT_SYMMETRIC:
8282 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
8283 break;
8284 case ALGORITHM_PARITY_0:
8285 new_layout = ALGORITHM_PARITY_0_6;
8286 break;
8287 case ALGORITHM_PARITY_N:
8288 new_layout = ALGORITHM_PARITY_N;
8289 break;
8290 default:
8291 return ERR_PTR(-EINVAL);
8292 }
8293 mddev->new_level = 6;
8294 mddev->new_layout = new_layout;
8295 mddev->delta_disks = 1;
8296 mddev->raid_disks += 1;
8297 return setup_conf(mddev);
8298}
8299
ba903a3e
AP		 8300static int raid5_change_consistency_policy(struct mddev *mddev, const char *buf)
8301{
8302 struct r5conf *conf;
8303 int err;
8304
8305 err = mddev_lock(mddev);
8306 if (err)
8307 return err;
8308 conf = mddev->private;
8309 if (!conf) {
8310 mddev_unlock(mddev);
8311 return -ENODEV;
8312 }
8313
845b9e22 8314 if (strncmp(buf, "ppl", 3) == 0) {
0bb0c105 8315 /* ppl only works with RAID 5 */
845b9e22
AP		 8316 if (!raid5_has_ppl(conf) && conf->level == 5) {
8317 err = log_init(conf, NULL, true);
8318 if (!err) {
8319 err = resize_stripes(conf, conf->pool_size);
8320 if (err)
8321 log_exit(conf);
8322 }
0bb0c105
SL		 8323 } else
8324 err = -EINVAL;
8325 } else if (strncmp(buf, "resync", 6) == 0) {
8326 if (raid5_has_ppl(conf)) {
8327 mddev_suspend(mddev);
8328 log_exit(conf);
0bb0c105 8329 mddev_resume(mddev);
845b9e22 8330 err = resize_stripes(conf, conf->pool_size);
0bb0c105
SL		 8331 } else if (test_bit(MD_HAS_JOURNAL, &conf->mddev->flags) &&
8332 r5l_log_disk_error(conf)) {
8333 bool journal_dev_exists = false;
8334 struct md_rdev *rdev;
8335
8336 rdev_for_each(rdev, mddev)
8337 if (test_bit(Journal, &rdev->flags)) {
8338 journal_dev_exists = true;
8339 break;
8340 }
8341
8342 if (!journal_dev_exists) {
8343 mddev_suspend(mddev);
8344 clear_bit(MD_HAS_JOURNAL, &mddev->flags);
8345 mddev_resume(mddev);
8346 } else /* need remove journal device first */
8347 err = -EBUSY;
8348 } else
8349 err = -EINVAL;
ba903a3e
AP		 8350 } else {
8351 err = -EINVAL;
8352 }
8353
8354 if (!err)
8355 md_update_sb(mddev, 1);
8356
8357 mddev_unlock(mddev);
8358
8359 return err;
8360}
8361
84fc4b56 8362static struct md_personality raid6_personality =
16a53ecc
N		 8363{
8364 .name = "raid6",
8365 .level = 6,
8366 .owner = THIS_MODULE,
849674e4
SL		 8367 .make_request = raid5_make_request,
8368 .run = raid5_run,
afa0f557 8369 .free = raid5_free,
849674e4
SL		 8370 .status = raid5_status,
8371 .error_handler = raid5_error,
16a53ecc
N		 8372 .hot_add_disk = raid5_add_disk,
8373 .hot_remove_disk= raid5_remove_disk,
8374 .spare_active = raid5_spare_active,
849674e4 8375 .sync_request = raid5_sync_request,
16a53ecc 8376 .resize = raid5_resize,
80c3a6ce 8377 .size = raid5_size,
50ac168a 8378 .check_reshape = raid6_check_reshape,
f416885e 8379 .start_reshape = raid5_start_reshape,
cea9c228 8380 .finish_reshape = raid5_finish_reshape,
16a53ecc 8381 .quiesce = raid5_quiesce,
245f46c2 8382 .takeover = raid6_takeover,
5c675f83 8383 .congested = raid5_congested,
0bb0c105 8384 .change_consistency_policy = raid5_change_consistency_policy,
16a53ecc 8385};
84fc4b56 8386static struct md_personality raid5_personality =
1da177e4
LT		 8387{
8388 .name = "raid5",
2604b703 8389 .level = 5,
1da177e4 8390 .owner = THIS_MODULE,
849674e4
SL		 8391 .make_request = raid5_make_request,
8392 .run = raid5_run,
afa0f557 8393 .free = raid5_free,
849674e4
SL		 8394 .status = raid5_status,
8395 .error_handler = raid5_error,
1da177e4
LT		 8396 .hot_add_disk = raid5_add_disk,
8397 .hot_remove_disk= raid5_remove_disk,
8398 .spare_active = raid5_spare_active,
849674e4 8399 .sync_request = raid5_sync_request,
1da177e4 8400 .resize = raid5_resize,
80c3a6ce 8401 .size = raid5_size,
63c70c4f
N		 8402 .check_reshape = raid5_check_reshape,
8403 .start_reshape = raid5_start_reshape,
cea9c228 8404 .finish_reshape = raid5_finish_reshape,
72626685 8405 .quiesce = raid5_quiesce,
d562b0c4 8406 .takeover = raid5_takeover,
5c675f83 8407 .congested = raid5_congested,
ba903a3e 8408 .change_consistency_policy = raid5_change_consistency_policy,
1da177e4
LT		 8409};
8410
84fc4b56 8411static struct md_personality raid4_personality =
1da177e4 8412{
2604b703
N		 8413 .name = "raid4",
8414 .level = 4,
8415 .owner = THIS_MODULE,
849674e4
SL		 8416 .make_request = raid5_make_request,
8417 .run = raid5_run,
afa0f557 8418 .free = raid5_free,
849674e4
SL		 8419 .status = raid5_status,
8420 .error_handler = raid5_error,
2604b703
N		 8421 .hot_add_disk = raid5_add_disk,
8422 .hot_remove_disk= raid5_remove_disk,
8423 .spare_active = raid5_spare_active,
849674e4 8424 .sync_request = raid5_sync_request,
2604b703 8425 .resize = raid5_resize,
80c3a6ce 8426 .size = raid5_size,
3d37890b
N		 8427 .check_reshape = raid5_check_reshape,
8428 .start_reshape = raid5_start_reshape,
cea9c228 8429 .finish_reshape = raid5_finish_reshape,
2604b703 8430 .quiesce = raid5_quiesce,
a78d38a1 8431 .takeover = raid4_takeover,
5c675f83 8432 .congested = raid5_congested,
0bb0c105 8433 .change_consistency_policy = raid5_change_consistency_policy,
2604b703
N		 8434};
8435
8436static int __init raid5_init(void)
8437{
29c6d1bb
SAS		 8438 int ret;
8439
851c30c9
SL		 8440 raid5_wq = alloc_workqueue("raid5wq",
8441 WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0);
8442 if (!raid5_wq)
8443 return -ENOMEM;
29c6d1bb
SAS		 8444
8445 ret = cpuhp_setup_state_multi(CPUHP_MD_RAID5_PREPARE,
8446 "md/raid5:prepare",
8447 raid456_cpu_up_prepare,
8448 raid456_cpu_dead);
8449 if (ret) {
8450 destroy_workqueue(raid5_wq);
8451 return ret;
8452 }
16a53ecc 8453 register_md_personality(&raid6_personality);
2604b703
N		 8454 register_md_personality(&raid5_personality);
8455 register_md_personality(&raid4_personality);
8456 return 0;
1da177e4
LT		 8457}
8458
2604b703 8459static void raid5_exit(void)
1da177e4 8460{
16a53ecc 8461 unregister_md_personality(&raid6_personality);
2604b703
N		 8462 unregister_md_personality(&raid5_personality);
8463 unregister_md_personality(&raid4_personality);
29c6d1bb 8464 cpuhp_remove_multi_state(CPUHP_MD_RAID5_PREPARE);
851c30c9 8465 destroy_workqueue(raid5_wq);
1da177e4
LT		 8466}
8467
8468module_init(raid5_init);
8469module_exit(raid5_exit);
8470MODULE_LICENSE("GPL");
0efb9e61 8471MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
1da177e4 8472MODULE_ALIAS("md-personality-4"); /* RAID5 */
d9d166c2
N		 8473MODULE_ALIAS("md-raid5");
8474MODULE_ALIAS("md-raid4");
2604b703
N		 8475MODULE_ALIAS("md-level-5");
8476MODULE_ALIAS("md-level-4");
16a53ecc
N		 8477MODULE_ALIAS("md-personality-8"); /* RAID6 */
8478MODULE_ALIAS("md-raid6");
8479MODULE_ALIAS("md-level-6");
8480
8481/* This used to be two separate modules, they were: */
8482MODULE_ALIAS("raid5");
8483MODULE_ALIAS("raid6");
Linux 6.x block layer and io_uring tree(s)