projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Linux 5.0-rc4
[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 58
a9add5d9 59#include <trace/events/block.h>
aaf9f12e 60#include <linux/list_sort.h>
a9add5d9 61
43b2e5d8 62#include "md.h"
bff61975 63#include "raid5.h"
54071b38 64#include "raid0.h"
935fe098 65#include "md-bitmap.h"
ff875738 66#include "raid5-log.h"
72626685 67
394ed8e4
SL		 68#define UNSUPPORTED_MDDEV_FLAGS (1L << MD_FAILFAST_SUPPORTED)
69
851c30c9
SL		 70#define cpu_to_group(cpu) cpu_to_node(cpu)
71#define ANY_GROUP NUMA_NO_NODE
72
8e0e99ba
N		 73static bool devices_handle_discard_safely = false;
74module_param(devices_handle_discard_safely, bool, 0644);
75MODULE_PARM_DESC(devices_handle_discard_safely,
76 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
851c30c9 77static struct workqueue_struct *raid5_wq;
1da177e4 78
d1688a6d 79static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect)
db298e19
N		 80{
81 int hash = (sect >> STRIPE_SHIFT) & HASH_MASK;
82 return &conf->stripe_hashtbl[hash];
83}
1da177e4 84
566c09c5
SL		 85static inline int stripe_hash_locks_hash(sector_t sect)
86{
87 return (sect >> STRIPE_SHIFT) & STRIPE_HASH_LOCKS_MASK;
88}
89
90static inline void lock_device_hash_lock(struct r5conf *conf, int hash)
91{
92 spin_lock_irq(conf->hash_locks + hash);
93 spin_lock(&conf->device_lock);
94}
95
96static inline void unlock_device_hash_lock(struct r5conf *conf, int hash)
97{
98 spin_unlock(&conf->device_lock);
99 spin_unlock_irq(conf->hash_locks + hash);
100}
101
102static inline void lock_all_device_hash_locks_irq(struct r5conf *conf)
103{
104 int i;
3d05f3ae 105 spin_lock_irq(conf->hash_locks);
566c09c5
SL		 106 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
107 spin_lock_nest_lock(conf->hash_locks + i, conf->hash_locks);
108 spin_lock(&conf->device_lock);
109}
110
111static inline void unlock_all_device_hash_locks_irq(struct r5conf *conf)
112{
113 int i;
114 spin_unlock(&conf->device_lock);
3d05f3ae
JC		 115 for (i = NR_STRIPE_HASH_LOCKS - 1; i; i--)
116 spin_unlock(conf->hash_locks + i);
117 spin_unlock_irq(conf->hash_locks);
566c09c5
SL		 118}
119
d0dabf7e
N		 120/* Find first data disk in a raid6 stripe */
121static inline int raid6_d0(struct stripe_head *sh)
122{
67cc2b81
N		 123 if (sh->ddf_layout)
124 /* ddf always start from first device */
125 return 0;
126 /* md starts just after Q block */
d0dabf7e
N		 127 if (sh->qd_idx == sh->disks - 1)
128 return 0;
129 else
130 return sh->qd_idx + 1;
131}
16a53ecc
N		 132static inline int raid6_next_disk(int disk, int raid_disks)
133{
134 disk++;
135 return (disk < raid_disks) ? disk : 0;
136}
a4456856 137
d0dabf7e
N		 138/* When walking through the disks in a raid5, starting at raid6_d0,
139 * We need to map each disk to a 'slot', where the data disks are slot
140 * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk
141 * is raid_disks-1. This help does that mapping.
142 */
67cc2b81
N		 143static int raid6_idx_to_slot(int idx, struct stripe_head *sh,
144 int *count, int syndrome_disks)
d0dabf7e 145{
6629542e 146 int slot = *count;
67cc2b81 147
e4424fee 148 if (sh->ddf_layout)
6629542e 149 (*count)++;
d0dabf7e 150 if (idx == sh->pd_idx)
67cc2b81 151 return syndrome_disks;
d0dabf7e 152 if (idx == sh->qd_idx)
67cc2b81 153 return syndrome_disks + 1;
e4424fee 154 if (!sh->ddf_layout)
6629542e 155 (*count)++;
d0dabf7e
N		 156 return slot;
157}
158
d1688a6d 159static void print_raid5_conf (struct r5conf *conf);
1da177e4 160
600aa109
DW		 161static int stripe_operations_active(struct stripe_head *sh)
162{
163 return sh->check_state || sh->reconstruct_state ||
164 test_bit(STRIPE_BIOFILL_RUN, &sh->state) ||
165 test_bit(STRIPE_COMPUTE_RUN, &sh->state);
166}
167
535ae4eb
SL		 168static bool stripe_is_lowprio(struct stripe_head *sh)
169{
170 return (test_bit(STRIPE_R5C_FULL_STRIPE, &sh->state) ||
171 test_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state)) &&
172 !test_bit(STRIPE_R5C_CACHING, &sh->state);
173}
174
851c30c9
SL		 175static void raid5_wakeup_stripe_thread(struct stripe_head *sh)
176{
177 struct r5conf *conf = sh->raid_conf;
178 struct r5worker_group *group;
bfc90cb0 179 int thread_cnt;
851c30c9
SL		 180 int i, cpu = sh->cpu;
181
182 if (!cpu_online(cpu)) {
183 cpu = cpumask_any(cpu_online_mask);
184 sh->cpu = cpu;
185 }
186
187 if (list_empty(&sh->lru)) {
188 struct r5worker_group *group;
189 group = conf->worker_groups + cpu_to_group(cpu);
535ae4eb
SL		 190 if (stripe_is_lowprio(sh))
191 list_add_tail(&sh->lru, &group->loprio_list);
192 else
193 list_add_tail(&sh->lru, &group->handle_list);
bfc90cb0
SL		 194 group->stripes_cnt++;
195 sh->group = group;
851c30c9
SL		 196 }
197
198 if (conf->worker_cnt_per_group == 0) {
199 md_wakeup_thread(conf->mddev->thread);
200 return;
201 }
202
203 group = conf->worker_groups + cpu_to_group(sh->cpu);
204
bfc90cb0
SL		 205 group->workers[0].working = true;
206 /* at least one worker should run to avoid race */
207 queue_work_on(sh->cpu, raid5_wq, &group->workers[0].work);
208
209 thread_cnt = group->stripes_cnt / MAX_STRIPE_BATCH - 1;
210 /* wakeup more workers */
211 for (i = 1; i < conf->worker_cnt_per_group && thread_cnt > 0; i++) {
212 if (group->workers[i].working == false) {
213 group->workers[i].working = true;
214 queue_work_on(sh->cpu, raid5_wq,
215 &group->workers[i].work);
216 thread_cnt--;
217 }
218 }
851c30c9
SL		 219}
220
566c09c5
SL		 221static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh,
222 struct list_head *temp_inactive_list)
1da177e4 223{
1e6d690b
SL		 224 int i;
225 int injournal = 0; /* number of date pages with R5_InJournal */
226
4eb788df
SL		 227 BUG_ON(!list_empty(&sh->lru));
228 BUG_ON(atomic_read(&conf->active_stripes)==0);
1e6d690b
SL		 229
230 if (r5c_is_writeback(conf->log))
231 for (i = sh->disks; i--; )
232 if (test_bit(R5_InJournal, &sh->dev[i].flags))
233 injournal++;
a39f7afd 234 /*
5ddf0440
SL		 235 * In the following cases, the stripe cannot be released to cached
236 * lists. Therefore, we make the stripe write out and set
237 * STRIPE_HANDLE:
238 * 1. when quiesce in r5c write back;
239 * 2. when resync is requested fot the stripe.
a39f7afd 240 */
5ddf0440
SL		 241 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) ||
242 (conf->quiesce && r5c_is_writeback(conf->log) &&
243 !test_bit(STRIPE_HANDLE, &sh->state) && injournal != 0)) {
a39f7afd
SL		 244 if (test_bit(STRIPE_R5C_CACHING, &sh->state))
245 r5c_make_stripe_write_out(sh);
246 set_bit(STRIPE_HANDLE, &sh->state);
247 }
1e6d690b 248
4eb788df
SL		 249 if (test_bit(STRIPE_HANDLE, &sh->state)) {
250 if (test_bit(STRIPE_DELAYED, &sh->state) &&
ad3ab8b6 251 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4eb788df 252 list_add_tail(&sh->lru, &conf->delayed_list);
ad3ab8b6 253 else if (test_bit(STRIPE_BIT_DELAY, &sh->state) &&
4eb788df
SL		 254 sh->bm_seq - conf->seq_write > 0)
255 list_add_tail(&sh->lru, &conf->bitmap_list);
256 else {
257 clear_bit(STRIPE_DELAYED, &sh->state);
258 clear_bit(STRIPE_BIT_DELAY, &sh->state);
851c30c9 259 if (conf->worker_cnt_per_group == 0) {
535ae4eb
SL		 260 if (stripe_is_lowprio(sh))
261 list_add_tail(&sh->lru,
262 &conf->loprio_list);
263 else
264 list_add_tail(&sh->lru,
265 &conf->handle_list);
851c30c9
SL		 266 } else {
267 raid5_wakeup_stripe_thread(sh);
268 return;
269 }
4eb788df
SL		 270 }
271 md_wakeup_thread(conf->mddev->thread);
272 } else {
273 BUG_ON(stripe_operations_active(sh));
274 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
275 if (atomic_dec_return(&conf->preread_active_stripes)
276 < IO_THRESHOLD)
277 md_wakeup_thread(conf->mddev->thread);
278 atomic_dec(&conf->active_stripes);
1e6d690b
SL		 279 if (!test_bit(STRIPE_EXPANDING, &sh->state)) {
280 if (!r5c_is_writeback(conf->log))
281 list_add_tail(&sh->lru, temp_inactive_list);
282 else {
283 WARN_ON(test_bit(R5_InJournal, &sh->dev[sh->pd_idx].flags));
284 if (injournal == 0)
285 list_add_tail(&sh->lru, temp_inactive_list);
286 else if (injournal == conf->raid_disks - conf->max_degraded) {
287 /* full stripe */
288 if (!test_and_set_bit(STRIPE_R5C_FULL_STRIPE, &sh->state))
289 atomic_inc(&conf->r5c_cached_full_stripes);
290 if (test_and_clear_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state))
291 atomic_dec(&conf->r5c_cached_partial_stripes);
292 list_add_tail(&sh->lru, &conf->r5c_full_stripe_list);
a39f7afd 293 r5c_check_cached_full_stripe(conf);
03b047f4
SL		 294 } else
295 /*
296 * STRIPE_R5C_PARTIAL_STRIPE is set in
297 * r5c_try_caching_write(). No need to
298 * set it again.
299 */
1e6d690b 300 list_add_tail(&sh->lru, &conf->r5c_partial_stripe_list);
1e6d690b
SL		 301 }
302 }
1da177e4
LT		 303 }
304}
d0dabf7e 305
566c09c5
SL		 306static void __release_stripe(struct r5conf *conf, struct stripe_head *sh,
307 struct list_head *temp_inactive_list)
4eb788df
SL		 308{
309 if (atomic_dec_and_test(&sh->count))
566c09c5
SL		 310 do_release_stripe(conf, sh, temp_inactive_list);
311}
312
313/*
314 * @hash could be NR_STRIPE_HASH_LOCKS, then we have a list of inactive_list
315 *
316 * Be careful: Only one task can add/delete stripes from temp_inactive_list at
317 * given time. Adding stripes only takes device lock, while deleting stripes
318 * only takes hash lock.
319 */
320static void release_inactive_stripe_list(struct r5conf *conf,
321 struct list_head *temp_inactive_list,
322 int hash)
323{
324 int size;
6ab2a4b8 325 bool do_wakeup = false;
566c09c5
SL		 326 unsigned long flags;
327
328 if (hash == NR_STRIPE_HASH_LOCKS) {
329 size = NR_STRIPE_HASH_LOCKS;
330 hash = NR_STRIPE_HASH_LOCKS - 1;
331 } else
332 size = 1;
333 while (size) {
334 struct list_head *list = &temp_inactive_list[size - 1];
335
336 /*
6d036f7d 337 * We don't hold any lock here yet, raid5_get_active_stripe() might
566c09c5
SL		 338 * remove stripes from the list
339 */
340 if (!list_empty_careful(list)) {
341 spin_lock_irqsave(conf->hash_locks + hash, flags);
4bda556a
SL		 342 if (list_empty(conf->inactive_list + hash) &&
343 !list_empty(list))
344 atomic_dec(&conf->empty_inactive_list_nr);
566c09c5 345 list_splice_tail_init(list, conf->inactive_list + hash);
6ab2a4b8 346 do_wakeup = true;
566c09c5
SL		 347 spin_unlock_irqrestore(conf->hash_locks + hash, flags);
348 }
349 size--;
350 hash--;
351 }
352
353 if (do_wakeup) {
6ab2a4b8 354 wake_up(&conf->wait_for_stripe);
b1b46486
YL		 355 if (atomic_read(&conf->active_stripes) == 0)
356 wake_up(&conf->wait_for_quiescent);
566c09c5
SL		 357 if (conf->retry_read_aligned)
358 md_wakeup_thread(conf->mddev->thread);
359 }
4eb788df
SL		 360}
361
773ca82f 362/* should hold conf->device_lock already */
566c09c5
SL		 363static int release_stripe_list(struct r5conf *conf,
364 struct list_head *temp_inactive_list)
773ca82f 365{
eae8263f 366 struct stripe_head *sh, *t;
773ca82f
SL		 367 int count = 0;
368 struct llist_node *head;
369
370 head = llist_del_all(&conf->released_stripes);
d265d9dc 371 head = llist_reverse_order(head);
eae8263f 372 llist_for_each_entry_safe(sh, t, head, release_list) {
566c09c5
SL		 373 int hash;
374
773ca82f
SL		 375 /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */
376 smp_mb();
377 clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state);
378 /*
379 * Don't worry the bit is set here, because if the bit is set
380 * again, the count is always > 1. This is true for
381 * STRIPE_ON_UNPLUG_LIST bit too.
382 */
566c09c5
SL		 383 hash = sh->hash_lock_index;
384 __release_stripe(conf, sh, &temp_inactive_list[hash]);
773ca82f
SL		 385 count++;
386 }
387
388 return count;
389}
390
6d036f7d 391void raid5_release_stripe(struct stripe_head *sh)
1da177e4 392{
d1688a6d 393 struct r5conf *conf = sh->raid_conf;
1da177e4 394 unsigned long flags;
566c09c5
SL		 395 struct list_head list;
396 int hash;
773ca82f 397 bool wakeup;
16a53ecc 398
cf170f3f
ES		 399 /* Avoid release_list until the last reference.
400 */
401 if (atomic_add_unless(&sh->count, -1, 1))
402 return;
403
ad4068de 404 if (unlikely(!conf->mddev->thread) ||
405 test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state))
773ca82f
SL		 406 goto slow_path;
407 wakeup = llist_add(&sh->release_list, &conf->released_stripes);
408 if (wakeup)
409 md_wakeup_thread(conf->mddev->thread);
410 return;
411slow_path:
773ca82f 412 /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */
685dbcaa 413 if (atomic_dec_and_lock_irqsave(&sh->count, &conf->device_lock, flags)) {
566c09c5
SL		 414 INIT_LIST_HEAD(&list);
415 hash = sh->hash_lock_index;
416 do_release_stripe(conf, sh, &list);
08edaaa6 417 spin_unlock_irqrestore(&conf->device_lock, flags);
566c09c5 418 release_inactive_stripe_list(conf, &list, hash);
4eb788df 419 }
1da177e4
LT		 420}
421
fccddba0 422static inline void remove_hash(struct stripe_head *sh)
1da177e4 423{
45b4233c
DW		 424 pr_debug("remove_hash(), stripe %llu\n",
425 (unsigned long long)sh->sector);
1da177e4 426
fccddba0 427 hlist_del_init(&sh->hash);
1da177e4
LT		 428}
429
d1688a6d 430static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh)
1da177e4 431{
fccddba0 432 struct hlist_head *hp = stripe_hash(conf, sh->sector);
1da177e4 433
45b4233c
DW		 434 pr_debug("insert_hash(), stripe %llu\n",
435 (unsigned long long)sh->sector);
1da177e4 436
fccddba0 437 hlist_add_head(&sh->hash, hp);
1da177e4
LT		 438}
439
1da177e4 440/* find an idle stripe, make sure it is unhashed, and return it. */
566c09c5 441static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash)
1da177e4
LT		 442{
443 struct stripe_head *sh = NULL;
444 struct list_head *first;
445
566c09c5 446 if (list_empty(conf->inactive_list + hash))
1da177e4 447 goto out;
566c09c5 448 first = (conf->inactive_list + hash)->next;
1da177e4
LT		 449 sh = list_entry(first, struct stripe_head, lru);
450 list_del_init(first);
451 remove_hash(sh);
452 atomic_inc(&conf->active_stripes);
566c09c5 453 BUG_ON(hash != sh->hash_lock_index);
4bda556a
SL		 454 if (list_empty(conf->inactive_list + hash))
455 atomic_inc(&conf->empty_inactive_list_nr);
1da177e4
LT		 456out:
457 return sh;
458}
459
e4e11e38 460static void shrink_buffers(struct stripe_head *sh)
1da177e4
LT		 461{
462 struct page *p;
463 int i;
e4e11e38 464 int num = sh->raid_conf->pool_size;
1da177e4 465
e4e11e38 466 for (i = 0; i < num ; i++) {
d592a996 467 WARN_ON(sh->dev[i].page != sh->dev[i].orig_page);
1da177e4
LT		 468 p = sh->dev[i].page;
469 if (!p)
470 continue;
471 sh->dev[i].page = NULL;
2d1f3b5d 472 put_page(p);
1da177e4
LT		 473 }
474}
475
a9683a79 476static int grow_buffers(struct stripe_head *sh, gfp_t gfp)
1da177e4
LT		 477{
478 int i;
e4e11e38 479 int num = sh->raid_conf->pool_size;
1da177e4 480
e4e11e38 481 for (i = 0; i < num; i++) {
1da177e4
LT		 482 struct page *page;
483
a9683a79 484 if (!(page = alloc_page(gfp))) {
1da177e4
LT		 485 return 1;
486 }
487 sh->dev[i].page = page;
d592a996 488 sh->dev[i].orig_page = page;
1da177e4 489 }
3418d036 490
1da177e4
LT		 491 return 0;
492}
493
d1688a6d 494static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
911d4ee8 495 struct stripe_head *sh);
1da177e4 496
b5663ba4 497static void init_stripe(struct stripe_head *sh, sector_t sector, int previous)
1da177e4 498{
d1688a6d 499 struct r5conf *conf = sh->raid_conf;
566c09c5 500 int i, seq;
1da177e4 501
78bafebd
ES		 502 BUG_ON(atomic_read(&sh->count) != 0);
503 BUG_ON(test_bit(STRIPE_HANDLE, &sh->state));
600aa109 504 BUG_ON(stripe_operations_active(sh));
59fc630b 505 BUG_ON(sh->batch_head);
d84e0f10 506
45b4233c 507 pr_debug("init_stripe called, stripe %llu\n",
b8e6a15a 508 (unsigned long long)sector);
566c09c5
SL		 509retry:
510 seq = read_seqcount_begin(&conf->gen_lock);
86b42c71 511 sh->generation = conf->generation - previous;
b5663ba4 512 sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks;
1da177e4 513 sh->sector = sector;
911d4ee8 514 stripe_set_idx(sector, conf, previous, sh);
1da177e4
LT		 515 sh->state = 0;
516
7ecaa1e6 517 for (i = sh->disks; i--; ) {
1da177e4
LT		 518 struct r5dev *dev = &sh->dev[i];
519
d84e0f10 520 if (dev->toread || dev->read || dev->towrite || dev->written ||
1da177e4 521 test_bit(R5_LOCKED, &dev->flags)) {
cc6167b4 522 pr_err("sector=%llx i=%d %p %p %p %p %d\n",
1da177e4 523 (unsigned long long)sh->sector, i, dev->toread,
d84e0f10 524 dev->read, dev->towrite, dev->written,
1da177e4 525 test_bit(R5_LOCKED, &dev->flags));
8cfa7b0f 526 WARN_ON(1);
1da177e4
LT		 527 }
528 dev->flags = 0;
27a4ff8f 529 dev->sector = raid5_compute_blocknr(sh, i, previous);
1da177e4 530 }
566c09c5
SL		 531 if (read_seqcount_retry(&conf->gen_lock, seq))
532 goto retry;
7a87f434 533 sh->overwrite_disks = 0;
1da177e4 534 insert_hash(conf, sh);
851c30c9 535 sh->cpu = smp_processor_id();
da41ba65 536 set_bit(STRIPE_BATCH_READY, &sh->state);
1da177e4
LT		 537}
538
d1688a6d 539static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector,
86b42c71 540 short generation)
1da177e4
LT		 541{
542 struct stripe_head *sh;
543
45b4233c 544 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
b67bfe0d 545 hlist_for_each_entry(sh, stripe_hash(conf, sector), hash)
86b42c71 546 if (sh->sector == sector && sh->generation == generation)
1da177e4 547 return sh;
45b4233c 548 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
1da177e4
LT		 549 return NULL;
550}
551
674806d6
N		 552/*
553 * Need to check if array has failed when deciding whether to:
554 * - start an array
555 * - remove non-faulty devices
556 * - add a spare
557 * - allow a reshape
558 * This determination is simple when no reshape is happening.
559 * However if there is a reshape, we need to carefully check
560 * both the before and after sections.
561 * This is because some failed devices may only affect one
562 * of the two sections, and some non-in_sync devices may
563 * be insync in the section most affected by failed devices.
564 */
2e38a37f 565int raid5_calc_degraded(struct r5conf *conf)
674806d6 566{
908f4fbd 567 int degraded, degraded2;
674806d6 568 int i;
674806d6
N		 569
570 rcu_read_lock();
571 degraded = 0;
572 for (i = 0; i < conf->previous_raid_disks; i++) {
3cb03002 573 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
e5c86471
N		 574 if (rdev && test_bit(Faulty, &rdev->flags))
575 rdev = rcu_dereference(conf->disks[i].replacement);
674806d6
N		 576 if (!rdev || test_bit(Faulty, &rdev->flags))
577 degraded++;
578 else if (test_bit(In_sync, &rdev->flags))
579 ;
580 else
581 /* not in-sync or faulty.
582 * If the reshape increases the number of devices,
583 * this is being recovered by the reshape, so
584 * this 'previous' section is not in_sync.
585 * If the number of devices is being reduced however,
586 * the device can only be part of the array if
587 * we are reverting a reshape, so this section will
588 * be in-sync.
589 */
590 if (conf->raid_disks >= conf->previous_raid_disks)
591 degraded++;
592 }
593 rcu_read_unlock();
908f4fbd
N		 594 if (conf->raid_disks == conf->previous_raid_disks)
595 return degraded;
674806d6 596 rcu_read_lock();
908f4fbd 597 degraded2 = 0;
674806d6 598 for (i = 0; i < conf->raid_disks; i++) {
3cb03002 599 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
e5c86471
N		 600 if (rdev && test_bit(Faulty, &rdev->flags))
601 rdev = rcu_dereference(conf->disks[i].replacement);
674806d6 602 if (!rdev || test_bit(Faulty, &rdev->flags))
908f4fbd 603 degraded2++;
674806d6
N		 604 else if (test_bit(In_sync, &rdev->flags))
605 ;
606 else
607 /* not in-sync or faulty.
608 * If reshape increases the number of devices, this
609 * section has already been recovered, else it
610 * almost certainly hasn't.
611 */
612 if (conf->raid_disks <= conf->previous_raid_disks)
908f4fbd 613 degraded2++;
674806d6
N		 614 }
615 rcu_read_unlock();
908f4fbd
N		 616 if (degraded2 > degraded)
617 return degraded2;
618 return degraded;
619}
620
621static int has_failed(struct r5conf *conf)
622{
623 int degraded;
624
625 if (conf->mddev->reshape_position == MaxSector)
626 return conf->mddev->degraded > conf->max_degraded;
627
2e38a37f 628 degraded = raid5_calc_degraded(conf);
674806d6
N		 629 if (degraded > conf->max_degraded)
630 return 1;
631 return 0;
632}
633
6d036f7d
SL		 634struct stripe_head *
635raid5_get_active_stripe(struct r5conf *conf, sector_t sector,
636 int previous, int noblock, int noquiesce)
1da177e4
LT		 637{
638 struct stripe_head *sh;
566c09c5 639 int hash = stripe_hash_locks_hash(sector);
ff00d3b4 640 int inc_empty_inactive_list_flag;
1da177e4 641
45b4233c 642 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector);
1da177e4 643
566c09c5 644 spin_lock_irq(conf->hash_locks + hash);
1da177e4
LT		 645
646 do {
b1b46486 647 wait_event_lock_irq(conf->wait_for_quiescent,
a8c906ca 648 conf->quiesce == 0 || noquiesce,
566c09c5 649 *(conf->hash_locks + hash));
86b42c71 650 sh = __find_stripe(conf, sector, conf->generation - previous);
1da177e4 651 if (!sh) {
edbe83ab 652 if (!test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) {
566c09c5 653 sh = get_free_stripe(conf, hash);
713bc5c2
SL		 654 if (!sh && !test_bit(R5_DID_ALLOC,
655 &conf->cache_state))
edbe83ab
N		 656 set_bit(R5_ALLOC_MORE,
657 &conf->cache_state);
658 }
1da177e4
LT		 659 if (noblock && sh == NULL)
660 break;
a39f7afd
SL		 661
662 r5c_check_stripe_cache_usage(conf);
1da177e4 663 if (!sh) {
5423399a
N		 664 set_bit(R5_INACTIVE_BLOCKED,
665 &conf->cache_state);
a39f7afd 666 r5l_wake_reclaim(conf->log, 0);
6ab2a4b8
SL		 667 wait_event_lock_irq(
668 conf->wait_for_stripe,
566c09c5
SL		 669 !list_empty(conf->inactive_list + hash) &&
670 (atomic_read(&conf->active_stripes)
671 < (conf->max_nr_stripes * 3 / 4)
5423399a
N		 672 || !test_bit(R5_INACTIVE_BLOCKED,
673 &conf->cache_state)),
6ab2a4b8 674 *(conf->hash_locks + hash));
5423399a
N		 675 clear_bit(R5_INACTIVE_BLOCKED,
676 &conf->cache_state);
7da9d450 677 } else {
b5663ba4 678 init_stripe(sh, sector, previous);
7da9d450
N		 679 atomic_inc(&sh->count);
680 }
e240c183 681 } else if (!atomic_inc_not_zero(&sh->count)) {
6d183de4 682 spin_lock(&conf->device_lock);
e240c183 683 if (!atomic_read(&sh->count)) {
1da177e4
LT		 684 if (!test_bit(STRIPE_HANDLE, &sh->state))
685 atomic_inc(&conf->active_stripes);
5af9bef7
N		 686 BUG_ON(list_empty(&sh->lru) &&
687 !test_bit(STRIPE_EXPANDING, &sh->state));
ff00d3b4
ZL		 688 inc_empty_inactive_list_flag = 0;
689 if (!list_empty(conf->inactive_list + hash))
690 inc_empty_inactive_list_flag = 1;
16a53ecc 691 list_del_init(&sh->lru);
ff00d3b4
ZL		 692 if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag)
693 atomic_inc(&conf->empty_inactive_list_nr);
bfc90cb0
SL		 694 if (sh->group) {
695 sh->group->stripes_cnt--;
696 sh->group = NULL;
697 }
1da177e4 698 }
7da9d450 699 atomic_inc(&sh->count);
6d183de4 700 spin_unlock(&conf->device_lock);
1da177e4
LT		 701 }
702 } while (sh == NULL);
703
566c09c5 704 spin_unlock_irq(conf->hash_locks + hash);
1da177e4
LT		 705 return sh;
706}
707
7a87f434 708static bool is_full_stripe_write(struct stripe_head *sh)
709{
710 BUG_ON(sh->overwrite_disks > (sh->disks - sh->raid_conf->max_degraded));
711 return sh->overwrite_disks == (sh->disks - sh->raid_conf->max_degraded);
712}
713
59fc630b 714static void lock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2)
715{
59fc630b 716 if (sh1 > sh2) {
3d05f3ae 717 spin_lock_irq(&sh2->stripe_lock);
59fc630b 718 spin_lock_nested(&sh1->stripe_lock, 1);
719 } else {
3d05f3ae 720 spin_lock_irq(&sh1->stripe_lock);
59fc630b 721 spin_lock_nested(&sh2->stripe_lock, 1);
722 }
723}
724
725static void unlock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2)
726{
727 spin_unlock(&sh1->stripe_lock);
3d05f3ae 728 spin_unlock_irq(&sh2->stripe_lock);
59fc630b 729}
730
731/* Only freshly new full stripe normal write stripe can be added to a batch list */
732static bool stripe_can_batch(struct stripe_head *sh)
733{
9c3e333d
SL		 734 struct r5conf *conf = sh->raid_conf;
735
e254de6b 736 if (raid5_has_log(conf) || raid5_has_ppl(conf))
9c3e333d 737 return false;
59fc630b 738 return test_bit(STRIPE_BATCH_READY, &sh->state) &&
d0852df5 739 !test_bit(STRIPE_BITMAP_PENDING, &sh->state) &&
59fc630b 740 is_full_stripe_write(sh);
741}
742
743/* we only do back search */
744static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh)
745{
746 struct stripe_head *head;
747 sector_t head_sector, tmp_sec;
748 int hash;
749 int dd_idx;
ff00d3b4 750 int inc_empty_inactive_list_flag;
59fc630b 751
59fc630b 752 /* Don't cross chunks, so stripe pd_idx/qd_idx is the same */
753 tmp_sec = sh->sector;
754 if (!sector_div(tmp_sec, conf->chunk_sectors))
755 return;
756 head_sector = sh->sector - STRIPE_SECTORS;
757
758 hash = stripe_hash_locks_hash(head_sector);
759 spin_lock_irq(conf->hash_locks + hash);
760 head = __find_stripe(conf, head_sector, conf->generation);
761 if (head && !atomic_inc_not_zero(&head->count)) {
762 spin_lock(&conf->device_lock);
763 if (!atomic_read(&head->count)) {
764 if (!test_bit(STRIPE_HANDLE, &head->state))
765 atomic_inc(&conf->active_stripes);
766 BUG_ON(list_empty(&head->lru) &&
767 !test_bit(STRIPE_EXPANDING, &head->state));
ff00d3b4
ZL		 768 inc_empty_inactive_list_flag = 0;
769 if (!list_empty(conf->inactive_list + hash))
770 inc_empty_inactive_list_flag = 1;
59fc630b 771 list_del_init(&head->lru);
ff00d3b4
ZL		 772 if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag)
773 atomic_inc(&conf->empty_inactive_list_nr);
59fc630b 774 if (head->group) {
775 head->group->stripes_cnt--;
776 head->group = NULL;
777 }
778 }
779 atomic_inc(&head->count);
780 spin_unlock(&conf->device_lock);
781 }
782 spin_unlock_irq(conf->hash_locks + hash);
783
784 if (!head)
785 return;
786 if (!stripe_can_batch(head))
787 goto out;
788
789 lock_two_stripes(head, sh);
790 /* clear_batch_ready clear the flag */
791 if (!stripe_can_batch(head) || !stripe_can_batch(sh))
792 goto unlock_out;
793
794 if (sh->batch_head)
795 goto unlock_out;
796
797 dd_idx = 0;
798 while (dd_idx == sh->pd_idx || dd_idx == sh->qd_idx)
799 dd_idx++;
1eff9d32 800 if (head->dev[dd_idx].towrite->bi_opf != sh->dev[dd_idx].towrite->bi_opf ||
796a5cf0 801 bio_op(head->dev[dd_idx].towrite) != bio_op(sh->dev[dd_idx].towrite))
59fc630b 802 goto unlock_out;
803
804 if (head->batch_head) {
805 spin_lock(&head->batch_head->batch_lock);
806 /* This batch list is already running */
807 if (!stripe_can_batch(head)) {
808 spin_unlock(&head->batch_head->batch_lock);
809 goto unlock_out;
810 }
3664847d
SL		 811 /*
812 * We must assign batch_head of this stripe within the
813 * batch_lock, otherwise clear_batch_ready of batch head
814 * stripe could clear BATCH_READY bit of this stripe and
815 * this stripe->batch_head doesn't get assigned, which
816 * could confuse clear_batch_ready for this stripe
817 */
818 sh->batch_head = head->batch_head;
59fc630b 819
820 /*
821 * at this point, head's BATCH_READY could be cleared, but we
822 * can still add the stripe to batch list
823 */
824 list_add(&sh->batch_list, &head->batch_list);
825 spin_unlock(&head->batch_head->batch_lock);
59fc630b 826 } else {
827 head->batch_head = head;
828 sh->batch_head = head->batch_head;
829 spin_lock(&head->batch_lock);
830 list_add_tail(&sh->batch_list, &head->batch_list);
831 spin_unlock(&head->batch_lock);
832 }
833
834 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
835 if (atomic_dec_return(&conf->preread_active_stripes)
836 < IO_THRESHOLD)
837 md_wakeup_thread(conf->mddev->thread);
838
2b6b2457
N		 839 if (test_and_clear_bit(STRIPE_BIT_DELAY, &sh->state)) {
840 int seq = sh->bm_seq;
841 if (test_bit(STRIPE_BIT_DELAY, &sh->batch_head->state) &&
842 sh->batch_head->bm_seq > seq)
843 seq = sh->batch_head->bm_seq;
844 set_bit(STRIPE_BIT_DELAY, &sh->batch_head->state);
845 sh->batch_head->bm_seq = seq;
846 }
847
59fc630b 848 atomic_inc(&sh->count);
849unlock_out:
850 unlock_two_stripes(head, sh);
851out:
6d036f7d 852 raid5_release_stripe(head);
59fc630b 853}
854
05616be5
N		 855/* Determine if 'data_offset' or 'new_data_offset' should be used
856 * in this stripe_head.
857 */
858static int use_new_offset(struct r5conf *conf, struct stripe_head *sh)
859{
860 sector_t progress = conf->reshape_progress;
861 /* Need a memory barrier to make sure we see the value
862 * of conf->generation, or ->data_offset that was set before
863 * reshape_progress was updated.
864 */
865 smp_rmb();
866 if (progress == MaxSector)
867 return 0;
868 if (sh->generation == conf->generation - 1)
869 return 0;
870 /* We are in a reshape, and this is a new-generation stripe,
871 * so use new_data_offset.
872 */
873 return 1;
874}
875
aaf9f12e 876static void dispatch_bio_list(struct bio_list *tmp)
765d704d 877{
765d704d
SL		 878 struct bio *bio;
879
aaf9f12e
SL		 880 while ((bio = bio_list_pop(tmp)))
881 generic_make_request(bio);
882}
883
884static int cmp_stripe(void *priv, struct list_head *a, struct list_head *b)
885{
886 const struct r5pending_data *da = list_entry(a,
887 struct r5pending_data, sibling);
888 const struct r5pending_data *db = list_entry(b,
889 struct r5pending_data, sibling);
890 if (da->sector > db->sector)
891 return 1;
892 if (da->sector < db->sector)
893 return -1;
894 return 0;
895}
896
897static void dispatch_defer_bios(struct r5conf *conf, int target,
898 struct bio_list *list)
899{
900 struct r5pending_data *data;
901 struct list_head *first, *next = NULL;
902 int cnt = 0;
903
904 if (conf->pending_data_cnt == 0)
905 return;
906
907 list_sort(NULL, &conf->pending_list, cmp_stripe);
908
909 first = conf->pending_list.next;
910
911 /* temporarily move the head */
912 if (conf->next_pending_data)
913 list_move_tail(&conf->pending_list,
914 &conf->next_pending_data->sibling);
915
916 while (!list_empty(&conf->pending_list)) {
917 data = list_first_entry(&conf->pending_list,
918 struct r5pending_data, sibling);
919 if (&data->sibling == first)
920 first = data->sibling.next;
921 next = data->sibling.next;
922
923 bio_list_merge(list, &data->bios);
924 list_move(&data->sibling, &conf->free_list);
925 cnt++;
926 if (cnt >= target)
927 break;
928 }
929 conf->pending_data_cnt -= cnt;
930 BUG_ON(conf->pending_data_cnt < 0 || cnt < target);
931
932 if (next != &conf->pending_list)
933 conf->next_pending_data = list_entry(next,
934 struct r5pending_data, sibling);
935 else
936 conf->next_pending_data = NULL;
937 /* list isn't empty */
938 if (first != &conf->pending_list)
939 list_move_tail(&conf->pending_list, first);
940}
941
942static void flush_deferred_bios(struct r5conf *conf)
943{
944 struct bio_list tmp = BIO_EMPTY_LIST;
945
946 if (conf->pending_data_cnt == 0)
765d704d
SL		 947 return;
948
765d704d 949 spin_lock(&conf->pending_bios_lock);
aaf9f12e
SL		 950 dispatch_defer_bios(conf, conf->pending_data_cnt, &tmp);
951 BUG_ON(conf->pending_data_cnt != 0);
765d704d
SL		 952 spin_unlock(&conf->pending_bios_lock);
953
aaf9f12e 954 dispatch_bio_list(&tmp);
765d704d
SL		 955}
956
aaf9f12e
SL		 957static void defer_issue_bios(struct r5conf *conf, sector_t sector,
958 struct bio_list *bios)
765d704d 959{
aaf9f12e
SL		 960 struct bio_list tmp = BIO_EMPTY_LIST;
961 struct r5pending_data *ent;
962
765d704d 963 spin_lock(&conf->pending_bios_lock);
aaf9f12e
SL		 964 ent = list_first_entry(&conf->free_list, struct r5pending_data,
965 sibling);
966 list_move_tail(&ent->sibling, &conf->pending_list);
967 ent->sector = sector;
968 bio_list_init(&ent->bios);
969 bio_list_merge(&ent->bios, bios);
970 conf->pending_data_cnt++;
971 if (conf->pending_data_cnt >= PENDING_IO_MAX)
972 dispatch_defer_bios(conf, PENDING_IO_ONE_FLUSH, &tmp);
973
765d704d 974 spin_unlock(&conf->pending_bios_lock);
aaf9f12e
SL		 975
976 dispatch_bio_list(&tmp);
765d704d
SL		 977}
978
6712ecf8 979static void
4246a0b6 980raid5_end_read_request(struct bio *bi);
6712ecf8 981static void
4246a0b6 982raid5_end_write_request(struct bio *bi);
91c00924 983
c4e5ac0a 984static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
91c00924 985{
d1688a6d 986 struct r5conf *conf = sh->raid_conf;
91c00924 987 int i, disks = sh->disks;
59fc630b 988 struct stripe_head *head_sh = sh;
aaf9f12e
SL		 989 struct bio_list pending_bios = BIO_EMPTY_LIST;
990 bool should_defer;
91c00924
DW		 991
992 might_sleep();
993
ff875738
AP		 994 if (log_stripe(sh, s) == 0)
995 return;
1e6d690b 996
aaf9f12e 997 should_defer = conf->batch_bio_dispatch && conf->group_cnt;
1e6d690b 998
91c00924 999 for (i = disks; i--; ) {
796a5cf0 1000 int op, op_flags = 0;
9a3e1101 1001 int replace_only = 0;
977df362
N		 1002 struct bio *bi, *rbi;
1003 struct md_rdev *rdev, *rrdev = NULL;
59fc630b 1004
1005 sh = head_sh;
e9c7469b 1006 if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) {
796a5cf0 1007 op = REQ_OP_WRITE;
e9c7469b 1008 if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags))
70fd7614 1009 op_flags = REQ_FUA;
9e444768 1010 if (test_bit(R5_Discard, &sh->dev[i].flags))
796a5cf0 1011 op = REQ_OP_DISCARD;
e9c7469b 1012 } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags))
796a5cf0 1013 op = REQ_OP_READ;
9a3e1101
N		 1014 else if (test_and_clear_bit(R5_WantReplace,
1015 &sh->dev[i].flags)) {
796a5cf0 1016 op = REQ_OP_WRITE;
9a3e1101
N		 1017 replace_only = 1;
1018 } else
91c00924 1019 continue;
bc0934f0 1020 if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags))
796a5cf0 1021 op_flags |= REQ_SYNC;
91c00924 1022
59fc630b 1023again:
91c00924 1024 bi = &sh->dev[i].req;
977df362 1025 rbi = &sh->dev[i].rreq; /* For writing to replacement */
91c00924 1026
91c00924 1027 rcu_read_lock();
9a3e1101 1028 rrdev = rcu_dereference(conf->disks[i].replacement);
dd054fce
N		 1029 smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */
1030 rdev = rcu_dereference(conf->disks[i].rdev);
1031 if (!rdev) {
1032 rdev = rrdev;
1033 rrdev = NULL;
1034 }
796a5cf0 1035 if (op_is_write(op)) {
9a3e1101
N		 1036 if (replace_only)
1037 rdev = NULL;
dd054fce
N		 1038 if (rdev == rrdev)
1039 /* We raced and saw duplicates */
1040 rrdev = NULL;
9a3e1101 1041 } else {
59fc630b 1042 if (test_bit(R5_ReadRepl, &head_sh->dev[i].flags) && rrdev)
9a3e1101
N		 1043 rdev = rrdev;
1044 rrdev = NULL;
1045 }
977df362 1046
91c00924
DW		 1047 if (rdev && test_bit(Faulty, &rdev->flags))
1048 rdev = NULL;
1049 if (rdev)
1050 atomic_inc(&rdev->nr_pending);
977df362
N		 1051 if (rrdev && test_bit(Faulty, &rrdev->flags))
1052 rrdev = NULL;
1053 if (rrdev)
1054 atomic_inc(&rrdev->nr_pending);
91c00924
DW		 1055 rcu_read_unlock();
1056
73e92e51 1057 /* We have already checked bad blocks for reads. Now
977df362
N		 1058 * need to check for writes. We never accept write errors
1059 * on the replacement, so we don't to check rrdev.
73e92e51 1060 */
796a5cf0 1061 while (op_is_write(op) && rdev &&
73e92e51
N		 1062 test_bit(WriteErrorSeen, &rdev->flags)) {
1063 sector_t first_bad;
1064 int bad_sectors;
1065 int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
1066 &first_bad, &bad_sectors);
1067 if (!bad)
1068 break;
1069
1070 if (bad < 0) {
1071 set_bit(BlockedBadBlocks, &rdev->flags);
1072 if (!conf->mddev->external &&
2953079c 1073 conf->mddev->sb_flags) {
73e92e51
N		 1074 /* It is very unlikely, but we might
1075 * still need to write out the
1076 * bad block log - better give it
1077 * a chance*/
1078 md_check_recovery(conf->mddev);
1079 }
1850753d 1080 /*
1081 * Because md_wait_for_blocked_rdev
1082 * will dec nr_pending, we must
1083 * increment it first.
1084 */
1085 atomic_inc(&rdev->nr_pending);
73e92e51
N		 1086 md_wait_for_blocked_rdev(rdev, conf->mddev);
1087 } else {
1088 /* Acknowledged bad block - skip the write */
1089 rdev_dec_pending(rdev, conf->mddev);
1090 rdev = NULL;
1091 }
1092 }
1093
91c00924 1094 if (rdev) {
9a3e1101
N		 1095 if (s->syncing || s->expanding || s->expanded
1096 || s->replacing)
91c00924
DW		 1097 md_sync_acct(rdev->bdev, STRIPE_SECTORS);
1098
2b7497f0
DW		 1099 set_bit(STRIPE_IO_STARTED, &sh->state);
1100
74d46992 1101 bio_set_dev(bi, rdev->bdev);
796a5cf0
MC		 1102 bio_set_op_attrs(bi, op, op_flags);
1103 bi->bi_end_io = op_is_write(op)
2f6db2a7
KO		 1104 ? raid5_end_write_request
1105 : raid5_end_read_request;
1106 bi->bi_private = sh;
1107
6296b960 1108 pr_debug("%s: for %llu schedule op %d on disc %d\n",
e46b272b 1109 __func__, (unsigned long long)sh->sector,
1eff9d32 1110 bi->bi_opf, i);
91c00924 1111 atomic_inc(&sh->count);
59fc630b 1112 if (sh != head_sh)
1113 atomic_inc(&head_sh->count);
05616be5 1114 if (use_new_offset(conf, sh))
4f024f37 1115 bi->bi_iter.bi_sector = (sh->sector
05616be5
N		 1116 + rdev->new_data_offset);
1117 else
4f024f37 1118 bi->bi_iter.bi_sector = (sh->sector
05616be5 1119 + rdev->data_offset);
59fc630b 1120 if (test_bit(R5_ReadNoMerge, &head_sh->dev[i].flags))
1eff9d32 1121 bi->bi_opf |= REQ_NOMERGE;
3f9e7c14 1122
d592a996
SL		 1123 if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
1124 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
86aa1397
SL		 1125
1126 if (!op_is_write(op) &&
1127 test_bit(R5_InJournal, &sh->dev[i].flags))
1128 /*
1129 * issuing read for a page in journal, this
1130 * must be preparing for prexor in rmw; read
1131 * the data into orig_page
1132 */
1133 sh->dev[i].vec.bv_page = sh->dev[i].orig_page;
1134 else
1135 sh->dev[i].vec.bv_page = sh->dev[i].page;
4997b72e 1136 bi->bi_vcnt = 1;
91c00924
DW		 1137 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
1138 bi->bi_io_vec[0].bv_offset = 0;
4f024f37 1139 bi->bi_iter.bi_size = STRIPE_SIZE;
2cd259a7
MD		 1140 bi->bi_write_hint = sh->dev[i].write_hint;
1141 if (!rrdev)
1142 sh->dev[i].write_hint = RWF_WRITE_LIFE_NOT_SET;
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;
2cd259a7
MD		 1194 rbi->bi_write_hint = sh->dev[i].write_hint;
1195 sh->dev[i].write_hint = RWF_WRITE_LIFE_NOT_SET;
37c61ff3
SL		 1196 /*
1197 * If this is discard request, set bi_vcnt 0. We don't
1198 * want to confuse SCSI because SCSI will replace payload
1199 */
796a5cf0 1200 if (op == REQ_OP_DISCARD)
37c61ff3 1201 rbi->bi_vcnt = 0;
e3620a3a 1202 if (conf->mddev->gendisk)
74d46992 1203 trace_block_bio_remap(rbi->bi_disk->queue,
e3620a3a
JB		 1204 rbi, disk_devt(conf->mddev->gendisk),
1205 sh->dev[i].sector);
aaf9f12e
SL		 1206 if (should_defer && op_is_write(op))
1207 bio_list_add(&pending_bios, rbi);
1208 else
1209 generic_make_request(rbi);
977df362
N		 1210 }
1211 if (!rdev && !rrdev) {
796a5cf0 1212 if (op_is_write(op))
91c00924 1213 set_bit(STRIPE_DEGRADED, &sh->state);
6296b960 1214 pr_debug("skip op %d on disc %d for sector %llu\n",
1eff9d32 1215 bi->bi_opf, i, (unsigned long long)sh->sector);
91c00924
DW		 1216 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1217 set_bit(STRIPE_HANDLE, &sh->state);
1218 }
59fc630b 1219
1220 if (!head_sh->batch_head)
1221 continue;
1222 sh = list_first_entry(&sh->batch_list, struct stripe_head,
1223 batch_list);
1224 if (sh != head_sh)
1225 goto again;
91c00924 1226 }
aaf9f12e
SL		 1227
1228 if (should_defer && !bio_list_empty(&pending_bios))
1229 defer_issue_bios(conf, head_sh->sector, &pending_bios);
91c00924
DW		 1230}
1231
1232static struct dma_async_tx_descriptor *
d592a996
SL		 1233async_copy_data(int frombio, struct bio *bio, struct page **page,
1234 sector_t sector, struct dma_async_tx_descriptor *tx,
1e6d690b 1235 struct stripe_head *sh, int no_skipcopy)
91c00924 1236{
7988613b
KO		 1237 struct bio_vec bvl;
1238 struct bvec_iter iter;
91c00924 1239 struct page *bio_page;
91c00924 1240 int page_offset;
a08abd8c 1241 struct async_submit_ctl submit;
0403e382 1242 enum async_tx_flags flags = 0;
91c00924 1243
4f024f37
KO		 1244 if (bio->bi_iter.bi_sector >= sector)
1245 page_offset = (signed)(bio->bi_iter.bi_sector - sector) * 512;
91c00924 1246 else
4f024f37 1247 page_offset = (signed)(sector - bio->bi_iter.bi_sector) * -512;
a08abd8c 1248
0403e382
DW		 1249 if (frombio)
1250 flags |= ASYNC_TX_FENCE;
1251 init_async_submit(&submit, flags, tx, NULL, NULL, NULL);
1252
7988613b
KO		 1253 bio_for_each_segment(bvl, bio, iter) {
1254 int len = bvl.bv_len;
91c00924
DW		 1255 int clen;
1256 int b_offset = 0;
1257
1258 if (page_offset < 0) {
1259 b_offset = -page_offset;
1260 page_offset += b_offset;
1261 len -= b_offset;
1262 }
1263
1264 if (len > 0 && page_offset + len > STRIPE_SIZE)
1265 clen = STRIPE_SIZE - page_offset;
1266 else
1267 clen = len;
1268
1269 if (clen > 0) {
7988613b
KO		 1270 b_offset += bvl.bv_offset;
1271 bio_page = bvl.bv_page;
d592a996
SL		 1272 if (frombio) {
1273 if (sh->raid_conf->skip_copy &&
1274 b_offset == 0 && page_offset == 0 &&
1e6d690b
SL		 1275 clen == STRIPE_SIZE &&
1276 !no_skipcopy)
d592a996
SL		 1277 *page = bio_page;
1278 else
1279 tx = async_memcpy(*page, bio_page, page_offset,
a08abd8c 1280 b_offset, clen, &submit);
d592a996
SL		 1281 } else
1282 tx = async_memcpy(bio_page, *page, b_offset,
a08abd8c 1283 page_offset, clen, &submit);
91c00924 1284 }
a08abd8c
DW		 1285 /* chain the operations */
1286 submit.depend_tx = tx;
1287
91c00924
DW		 1288 if (clen < len) /* hit end of page */
1289 break;
1290 page_offset += len;
1291 }
1292
1293 return tx;
1294}
1295
1296static void ops_complete_biofill(void *stripe_head_ref)
1297{
1298 struct stripe_head *sh = stripe_head_ref;
e4d84909 1299 int i;
91c00924 1300
e46b272b 1301 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1302 (unsigned long long)sh->sector);
1303
1304 /* clear completed biofills */
1305 for (i = sh->disks; i--; ) {
1306 struct r5dev *dev = &sh->dev[i];
91c00924
DW		 1307
1308 /* acknowledge completion of a biofill operation */
e4d84909
DW		 1309 /* and check if we need to reply to a read request,
1310 * new R5_Wantfill requests are held off until
83de75cc 1311 * !STRIPE_BIOFILL_RUN
e4d84909
DW		 1312 */
1313 if (test_and_clear_bit(R5_Wantfill, &dev->flags)) {
91c00924 1314 struct bio *rbi, *rbi2;
91c00924 1315
91c00924
DW		 1316 BUG_ON(!dev->read);
1317 rbi = dev->read;
1318 dev->read = NULL;
4f024f37 1319 while (rbi && rbi->bi_iter.bi_sector <
91c00924
DW		 1320 dev->sector + STRIPE_SECTORS) {
1321 rbi2 = r5_next_bio(rbi, dev->sector);
016c76ac 1322 bio_endio(rbi);
91c00924
DW		 1323 rbi = rbi2;
1324 }
1325 }
1326 }
83de75cc 1327 clear_bit(STRIPE_BIOFILL_RUN, &sh->state);
91c00924 1328
e4d84909 1329 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 1330 raid5_release_stripe(sh);
91c00924
DW		 1331}
1332
1333static void ops_run_biofill(struct stripe_head *sh)
1334{
1335 struct dma_async_tx_descriptor *tx = NULL;
a08abd8c 1336 struct async_submit_ctl submit;
91c00924
DW		 1337 int i;
1338
59fc630b 1339 BUG_ON(sh->batch_head);
e46b272b 1340 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1341 (unsigned long long)sh->sector);
1342
1343 for (i = sh->disks; i--; ) {
1344 struct r5dev *dev = &sh->dev[i];
1345 if (test_bit(R5_Wantfill, &dev->flags)) {
1346 struct bio *rbi;
b17459c0 1347 spin_lock_irq(&sh->stripe_lock);
91c00924
DW		 1348 dev->read = rbi = dev->toread;
1349 dev->toread = NULL;
b17459c0 1350 spin_unlock_irq(&sh->stripe_lock);
4f024f37 1351 while (rbi && rbi->bi_iter.bi_sector <
91c00924 1352 dev->sector + STRIPE_SECTORS) {
d592a996 1353 tx = async_copy_data(0, rbi, &dev->page,
1e6d690b 1354 dev->sector, tx, sh, 0);
91c00924
DW		 1355 rbi = r5_next_bio(rbi, dev->sector);
1356 }
1357 }
1358 }
1359
1360 atomic_inc(&sh->count);
a08abd8c
DW		 1361 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL);
1362 async_trigger_callback(&submit);
91c00924
DW		 1363}
1364
4e7d2c0a 1365static void mark_target_uptodate(struct stripe_head *sh, int target)
91c00924 1366{
4e7d2c0a 1367 struct r5dev *tgt;
91c00924 1368
4e7d2c0a
DW		 1369 if (target < 0)
1370 return;
91c00924 1371
4e7d2c0a 1372 tgt = &sh->dev[target];
91c00924
DW		 1373 set_bit(R5_UPTODATE, &tgt->flags);
1374 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1375 clear_bit(R5_Wantcompute, &tgt->flags);
4e7d2c0a
DW		 1376}
1377
ac6b53b6 1378static void ops_complete_compute(void *stripe_head_ref)
91c00924
DW		 1379{
1380 struct stripe_head *sh = stripe_head_ref;
91c00924 1381
e46b272b 1382 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1383 (unsigned long long)sh->sector);
1384
ac6b53b6 1385 /* mark the computed target(s) as uptodate */
4e7d2c0a 1386 mark_target_uptodate(sh, sh->ops.target);
ac6b53b6 1387 mark_target_uptodate(sh, sh->ops.target2);
4e7d2c0a 1388
ecc65c9b
DW		 1389 clear_bit(STRIPE_COMPUTE_RUN, &sh->state);
1390 if (sh->check_state == check_state_compute_run)
1391 sh->check_state = check_state_compute_result;
91c00924 1392 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 1393 raid5_release_stripe(sh);
91c00924
DW		 1394}
1395
d6f38f31
DW		 1396/* return a pointer to the address conversion region of the scribble buffer */
1397static addr_conv_t *to_addr_conv(struct stripe_head *sh,
46d5b785 1398 struct raid5_percpu *percpu, int i)
d6f38f31 1399{
46d5b785 1400 void *addr;
1401
1402 addr = flex_array_get(percpu->scribble, i);
1403 return addr + sizeof(struct page *) * (sh->disks + 2);
1404}
1405
1406/* return a pointer to the address conversion region of the scribble buffer */
1407static struct page **to_addr_page(struct raid5_percpu *percpu, int i)
1408{
1409 void *addr;
1410
1411 addr = flex_array_get(percpu->scribble, i);
1412 return addr;
d6f38f31
DW		 1413}
1414
1415static struct dma_async_tx_descriptor *
1416ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 1417{
91c00924 1418 int disks = sh->disks;
46d5b785 1419 struct page **xor_srcs = to_addr_page(percpu, 0);
91c00924
DW		 1420 int target = sh->ops.target;
1421 struct r5dev *tgt = &sh->dev[target];
1422 struct page *xor_dest = tgt->page;
1423 int count = 0;
1424 struct dma_async_tx_descriptor *tx;
a08abd8c 1425 struct async_submit_ctl submit;
91c00924
DW		 1426 int i;
1427
59fc630b 1428 BUG_ON(sh->batch_head);
1429
91c00924 1430 pr_debug("%s: stripe %llu block: %d\n",
e46b272b 1431 __func__, (unsigned long long)sh->sector, target);
91c00924
DW		 1432 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1433
1434 for (i = disks; i--; )
1435 if (i != target)
1436 xor_srcs[count++] = sh->dev[i].page;
1437
1438 atomic_inc(&sh->count);
1439
0403e382 1440 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL,
46d5b785 1441 ops_complete_compute, sh, to_addr_conv(sh, percpu, 0));
91c00924 1442 if (unlikely(count == 1))
a08abd8c 1443 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
91c00924 1444 else
a08abd8c 1445 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924 1446
91c00924
DW		 1447 return tx;
1448}
1449
ac6b53b6
DW		 1450/* set_syndrome_sources - populate source buffers for gen_syndrome
1451 * @srcs - (struct page *) array of size sh->disks
1452 * @sh - stripe_head to parse
1453 *
1454 * Populates srcs in proper layout order for the stripe and returns the
1455 * 'count' of sources to be used in a call to async_gen_syndrome. The P
1456 * destination buffer is recorded in srcs[count] and the Q destination
1457 * is recorded in srcs[count+1]].
1458 */
584acdd4
MS		 1459static int set_syndrome_sources(struct page **srcs,
1460 struct stripe_head *sh,
1461 int srctype)
ac6b53b6
DW		 1462{
1463 int disks = sh->disks;
1464 int syndrome_disks = sh->ddf_layout ? disks : (disks - 2);
1465 int d0_idx = raid6_d0(sh);
1466 int count;
1467 int i;
1468
1469 for (i = 0; i < disks; i++)
5dd33c9a 1470 srcs[i] = NULL;
ac6b53b6
DW		 1471
1472 count = 0;
1473 i = d0_idx;
1474 do {
1475 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
584acdd4 1476 struct r5dev *dev = &sh->dev[i];
ac6b53b6 1477
584acdd4
MS		 1478 if (i == sh->qd_idx || i == sh->pd_idx ||
1479 (srctype == SYNDROME_SRC_ALL) ||
1480 (srctype == SYNDROME_SRC_WANT_DRAIN &&
1e6d690b
SL		 1481 (test_bit(R5_Wantdrain, &dev->flags) ||
1482 test_bit(R5_InJournal, &dev->flags))) ||
584acdd4 1483 (srctype == SYNDROME_SRC_WRITTEN &&
0977762f
SL		 1484 (dev->written ||
1485 test_bit(R5_InJournal, &dev->flags)))) {
1e6d690b
SL		 1486 if (test_bit(R5_InJournal, &dev->flags))
1487 srcs[slot] = sh->dev[i].orig_page;
1488 else
1489 srcs[slot] = sh->dev[i].page;
1490 }
ac6b53b6
DW		 1491 i = raid6_next_disk(i, disks);
1492 } while (i != d0_idx);
ac6b53b6 1493
e4424fee 1494 return syndrome_disks;
ac6b53b6
DW		 1495}
1496
1497static struct dma_async_tx_descriptor *
1498ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu)
1499{
1500 int disks = sh->disks;
46d5b785 1501 struct page **blocks = to_addr_page(percpu, 0);
ac6b53b6
DW		 1502 int target;
1503 int qd_idx = sh->qd_idx;
1504 struct dma_async_tx_descriptor *tx;
1505 struct async_submit_ctl submit;
1506 struct r5dev *tgt;
1507 struct page *dest;
1508 int i;
1509 int count;
1510
59fc630b 1511 BUG_ON(sh->batch_head);
ac6b53b6
DW		 1512 if (sh->ops.target < 0)
1513 target = sh->ops.target2;
1514 else if (sh->ops.target2 < 0)
1515 target = sh->ops.target;
91c00924 1516 else
ac6b53b6
DW		 1517 /* we should only have one valid target */
1518 BUG();
1519 BUG_ON(target < 0);
1520 pr_debug("%s: stripe %llu block: %d\n",
1521 __func__, (unsigned long long)sh->sector, target);
1522
1523 tgt = &sh->dev[target];
1524 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1525 dest = tgt->page;
1526
1527 atomic_inc(&sh->count);
1528
1529 if (target == qd_idx) {
584acdd4 1530 count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL);
ac6b53b6
DW		 1531 blocks[count] = NULL; /* regenerating p is not necessary */
1532 BUG_ON(blocks[count+1] != dest); /* q should already be set */
0403e382
DW		 1533 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1534 ops_complete_compute, sh,
46d5b785 1535 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1536 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
1537 } else {
1538 /* Compute any data- or p-drive using XOR */
1539 count = 0;
1540 for (i = disks; i-- ; ) {
1541 if (i == target || i == qd_idx)
1542 continue;
1543 blocks[count++] = sh->dev[i].page;
1544 }
1545
0403e382
DW		 1546 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1547 NULL, ops_complete_compute, sh,
46d5b785 1548 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1549 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit);
1550 }
91c00924 1551
91c00924
DW		 1552 return tx;
1553}
1554
ac6b53b6
DW		 1555static struct dma_async_tx_descriptor *
1556ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu)
1557{
1558 int i, count, disks = sh->disks;
1559 int syndrome_disks = sh->ddf_layout ? disks : disks-2;
1560 int d0_idx = raid6_d0(sh);
1561 int faila = -1, failb = -1;
1562 int target = sh->ops.target;
1563 int target2 = sh->ops.target2;
1564 struct r5dev *tgt = &sh->dev[target];
1565 struct r5dev *tgt2 = &sh->dev[target2];
1566 struct dma_async_tx_descriptor *tx;
46d5b785 1567 struct page **blocks = to_addr_page(percpu, 0);
ac6b53b6
DW		 1568 struct async_submit_ctl submit;
1569
59fc630b 1570 BUG_ON(sh->batch_head);
ac6b53b6
DW		 1571 pr_debug("%s: stripe %llu block1: %d block2: %d\n",
1572 __func__, (unsigned long long)sh->sector, target, target2);
1573 BUG_ON(target < 0 || target2 < 0);
1574 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1575 BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags));
1576
6c910a78 1577 /* we need to open-code set_syndrome_sources to handle the
ac6b53b6
DW		 1578 * slot number conversion for 'faila' and 'failb'
1579 */
1580 for (i = 0; i < disks ; i++)
5dd33c9a 1581 blocks[i] = NULL;
ac6b53b6
DW		 1582 count = 0;
1583 i = d0_idx;
1584 do {
1585 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
1586
1587 blocks[slot] = sh->dev[i].page;
1588
1589 if (i == target)
1590 faila = slot;
1591 if (i == target2)
1592 failb = slot;
1593 i = raid6_next_disk(i, disks);
1594 } while (i != d0_idx);
ac6b53b6
DW		 1595
1596 BUG_ON(faila == failb);
1597 if (failb < faila)
1598 swap(faila, failb);
1599 pr_debug("%s: stripe: %llu faila: %d failb: %d\n",
1600 __func__, (unsigned long long)sh->sector, faila, failb);
1601
1602 atomic_inc(&sh->count);
1603
1604 if (failb == syndrome_disks+1) {
1605 /* Q disk is one of the missing disks */
1606 if (faila == syndrome_disks) {
1607 /* Missing P+Q, just recompute */
0403e382
DW		 1608 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1609 ops_complete_compute, sh,
46d5b785 1610 to_addr_conv(sh, percpu, 0));
e4424fee 1611 return async_gen_syndrome(blocks, 0, syndrome_disks+2,
ac6b53b6
DW		 1612 STRIPE_SIZE, &submit);
1613 } else {
1614 struct page *dest;
1615 int data_target;
1616 int qd_idx = sh->qd_idx;
1617
1618 /* Missing D+Q: recompute D from P, then recompute Q */
1619 if (target == qd_idx)
1620 data_target = target2;
1621 else
1622 data_target = target;
1623
1624 count = 0;
1625 for (i = disks; i-- ; ) {
1626 if (i == data_target || i == qd_idx)
1627 continue;
1628 blocks[count++] = sh->dev[i].page;
1629 }
1630 dest = sh->dev[data_target].page;
0403e382
DW		 1631 init_async_submit(&submit,
1632 ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1633 NULL, NULL, NULL,
46d5b785 1634 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1635 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE,
1636 &submit);
1637
584acdd4 1638 count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL);
0403e382
DW		 1639 init_async_submit(&submit, ASYNC_TX_FENCE, tx,
1640 ops_complete_compute, sh,
46d5b785 1641 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1642 return async_gen_syndrome(blocks, 0, count+2,
1643 STRIPE_SIZE, &submit);
1644 }
ac6b53b6 1645 } else {
6c910a78
DW		 1646 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1647 ops_complete_compute, sh,
46d5b785 1648 to_addr_conv(sh, percpu, 0));
6c910a78
DW		 1649 if (failb == syndrome_disks) {
1650 /* We're missing D+P. */
1651 return async_raid6_datap_recov(syndrome_disks+2,
1652 STRIPE_SIZE, faila,
1653 blocks, &submit);
1654 } else {
1655 /* We're missing D+D. */
1656 return async_raid6_2data_recov(syndrome_disks+2,
1657 STRIPE_SIZE, faila, failb,
1658 blocks, &submit);
1659 }
ac6b53b6
DW		 1660 }
1661}
1662
91c00924
DW		 1663static void ops_complete_prexor(void *stripe_head_ref)
1664{
1665 struct stripe_head *sh = stripe_head_ref;
1666
e46b272b 1667 pr_debug("%s: stripe %llu\n", __func__,
91c00924 1668 (unsigned long long)sh->sector);
1e6d690b
SL		 1669
1670 if (r5c_is_writeback(sh->raid_conf->log))
1671 /*
1672 * raid5-cache write back uses orig_page during prexor.
1673 * After prexor, it is time to free orig_page
1674 */
1675 r5c_release_extra_page(sh);
91c00924
DW		 1676}
1677
1678static struct dma_async_tx_descriptor *
584acdd4
MS		 1679ops_run_prexor5(struct stripe_head *sh, struct raid5_percpu *percpu,
1680 struct dma_async_tx_descriptor *tx)
91c00924 1681{
91c00924 1682 int disks = sh->disks;
46d5b785 1683 struct page **xor_srcs = to_addr_page(percpu, 0);
91c00924 1684 int count = 0, pd_idx = sh->pd_idx, i;
a08abd8c 1685 struct async_submit_ctl submit;
91c00924
DW		 1686
1687 /* existing parity data subtracted */
1688 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1689
59fc630b 1690 BUG_ON(sh->batch_head);
e46b272b 1691 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1692 (unsigned long long)sh->sector);
1693
1694 for (i = disks; i--; ) {
1695 struct r5dev *dev = &sh->dev[i];
1696 /* Only process blocks that are known to be uptodate */
1e6d690b
SL		 1697 if (test_bit(R5_InJournal, &dev->flags))
1698 xor_srcs[count++] = dev->orig_page;
1699 else if (test_bit(R5_Wantdrain, &dev->flags))
91c00924
DW		 1700 xor_srcs[count++] = dev->page;
1701 }
1702
0403e382 1703 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
46d5b785 1704 ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0));
a08abd8c 1705 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924
DW		 1706
1707 return tx;
1708}
1709
584acdd4
MS		 1710static struct dma_async_tx_descriptor *
1711ops_run_prexor6(struct stripe_head *sh, struct raid5_percpu *percpu,
1712 struct dma_async_tx_descriptor *tx)
1713{
1714 struct page **blocks = to_addr_page(percpu, 0);
1715 int count;
1716 struct async_submit_ctl submit;
1717
1718 pr_debug("%s: stripe %llu\n", __func__,
1719 (unsigned long long)sh->sector);
1720
1721 count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_WANT_DRAIN);
1722
1723 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_PQ_XOR_DST, tx,
1724 ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0));
1725 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
1726
1727 return tx;
1728}
1729
91c00924 1730static struct dma_async_tx_descriptor *
d8ee0728 1731ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
91c00924 1732{
1e6d690b 1733 struct r5conf *conf = sh->raid_conf;
91c00924 1734 int disks = sh->disks;
d8ee0728 1735 int i;
59fc630b 1736 struct stripe_head *head_sh = sh;
91c00924 1737
e46b272b 1738 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1739 (unsigned long long)sh->sector);
1740
1741 for (i = disks; i--; ) {
59fc630b 1742 struct r5dev *dev;
91c00924 1743 struct bio *chosen;
91c00924 1744
59fc630b 1745 sh = head_sh;
1746 if (test_and_clear_bit(R5_Wantdrain, &head_sh->dev[i].flags)) {
91c00924
DW		 1747 struct bio *wbi;
1748
59fc630b 1749again:
1750 dev = &sh->dev[i];
1e6d690b
SL		 1751 /*
1752 * clear R5_InJournal, so when rewriting a page in
1753 * journal, it is not skipped by r5l_log_stripe()
1754 */
1755 clear_bit(R5_InJournal, &dev->flags);
b17459c0 1756 spin_lock_irq(&sh->stripe_lock);
91c00924
DW		 1757 chosen = dev->towrite;
1758 dev->towrite = NULL;
7a87f434 1759 sh->overwrite_disks = 0;
91c00924
DW		 1760 BUG_ON(dev->written);
1761 wbi = dev->written = chosen;
b17459c0 1762 spin_unlock_irq(&sh->stripe_lock);
d592a996 1763 WARN_ON(dev->page != dev->orig_page);
91c00924 1764
4f024f37 1765 while (wbi && wbi->bi_iter.bi_sector <
91c00924 1766 dev->sector + STRIPE_SECTORS) {
1eff9d32 1767 if (wbi->bi_opf & REQ_FUA)
e9c7469b 1768 set_bit(R5_WantFUA, &dev->flags);
1eff9d32 1769 if (wbi->bi_opf & REQ_SYNC)
bc0934f0 1770 set_bit(R5_SyncIO, &dev->flags);
796a5cf0 1771 if (bio_op(wbi) == REQ_OP_DISCARD)
620125f2 1772 set_bit(R5_Discard, &dev->flags);
d592a996
SL		 1773 else {
1774 tx = async_copy_data(1, wbi, &dev->page,
1e6d690b
SL		 1775 dev->sector, tx, sh,
1776 r5c_is_writeback(conf->log));
1777 if (dev->page != dev->orig_page &&
1778 !r5c_is_writeback(conf->log)) {
d592a996
SL		 1779 set_bit(R5_SkipCopy, &dev->flags);
1780 clear_bit(R5_UPTODATE, &dev->flags);
1781 clear_bit(R5_OVERWRITE, &dev->flags);
1782 }
1783 }
91c00924
DW		 1784 wbi = r5_next_bio(wbi, dev->sector);
1785 }
59fc630b 1786
1787 if (head_sh->batch_head) {
1788 sh = list_first_entry(&sh->batch_list,
1789 struct stripe_head,
1790 batch_list);
1791 if (sh == head_sh)
1792 continue;
1793 goto again;
1794 }
91c00924
DW		 1795 }
1796 }
1797
1798 return tx;
1799}
1800
ac6b53b6 1801static void ops_complete_reconstruct(void *stripe_head_ref)
91c00924
DW		 1802{
1803 struct stripe_head *sh = stripe_head_ref;
ac6b53b6
DW		 1804 int disks = sh->disks;
1805 int pd_idx = sh->pd_idx;
1806 int qd_idx = sh->qd_idx;
1807 int i;
9e444768 1808 bool fua = false, sync = false, discard = false;
91c00924 1809
e46b272b 1810 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1811 (unsigned long long)sh->sector);
1812
bc0934f0 1813 for (i = disks; i--; ) {
e9c7469b 1814 fua |= test_bit(R5_WantFUA, &sh->dev[i].flags);
bc0934f0 1815 sync |= test_bit(R5_SyncIO, &sh->dev[i].flags);
9e444768 1816 discard |= test_bit(R5_Discard, &sh->dev[i].flags);
bc0934f0 1817 }
e9c7469b 1818
91c00924
DW		 1819 for (i = disks; i--; ) {
1820 struct r5dev *dev = &sh->dev[i];
ac6b53b6 1821
e9c7469b 1822 if (dev->written || i == pd_idx || i == qd_idx) {
235b6003 1823 if (!discard && !test_bit(R5_SkipCopy, &dev->flags)) {
9e444768 1824 set_bit(R5_UPTODATE, &dev->flags);
235b6003
N		 1825 if (test_bit(STRIPE_EXPAND_READY, &sh->state))
1826 set_bit(R5_Expanded, &dev->flags);
1827 }
e9c7469b
TH		 1828 if (fua)
1829 set_bit(R5_WantFUA, &dev->flags);
bc0934f0
SL		 1830 if (sync)
1831 set_bit(R5_SyncIO, &dev->flags);
e9c7469b 1832 }
91c00924
DW		 1833 }
1834
d8ee0728
DW		 1835 if (sh->reconstruct_state == reconstruct_state_drain_run)
1836 sh->reconstruct_state = reconstruct_state_drain_result;
1837 else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run)
1838 sh->reconstruct_state = reconstruct_state_prexor_drain_result;
1839 else {
1840 BUG_ON(sh->reconstruct_state != reconstruct_state_run);
1841 sh->reconstruct_state = reconstruct_state_result;
1842 }
91c00924
DW		 1843
1844 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 1845 raid5_release_stripe(sh);
91c00924
DW		 1846}
1847
1848static void
ac6b53b6
DW		 1849ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu,
1850 struct dma_async_tx_descriptor *tx)
91c00924 1851{
91c00924 1852 int disks = sh->disks;
59fc630b 1853 struct page **xor_srcs;
a08abd8c 1854 struct async_submit_ctl submit;
59fc630b 1855 int count, pd_idx = sh->pd_idx, i;
91c00924 1856 struct page *xor_dest;
d8ee0728 1857 int prexor = 0;
91c00924 1858 unsigned long flags;
59fc630b 1859 int j = 0;
1860 struct stripe_head *head_sh = sh;
1861 int last_stripe;
91c00924 1862
e46b272b 1863 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1864 (unsigned long long)sh->sector);
1865
620125f2
SL		 1866 for (i = 0; i < sh->disks; i++) {
1867 if (pd_idx == i)
1868 continue;
1869 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1870 break;
1871 }
1872 if (i >= sh->disks) {
1873 atomic_inc(&sh->count);
620125f2
SL		 1874 set_bit(R5_Discard, &sh->dev[pd_idx].flags);
1875 ops_complete_reconstruct(sh);
1876 return;
1877 }
59fc630b 1878again:
1879 count = 0;
1880 xor_srcs = to_addr_page(percpu, j);
91c00924
DW		 1881 /* check if prexor is active which means only process blocks
1882 * that are part of a read-modify-write (written)
1883 */
59fc630b 1884 if (head_sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
d8ee0728 1885 prexor = 1;
91c00924
DW		 1886 xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1887 for (i = disks; i--; ) {
1888 struct r5dev *dev = &sh->dev[i];
1e6d690b
SL		 1889 if (head_sh->dev[i].written ||
1890 test_bit(R5_InJournal, &head_sh->dev[i].flags))
91c00924
DW		 1891 xor_srcs[count++] = dev->page;
1892 }
1893 } else {
1894 xor_dest = sh->dev[pd_idx].page;
1895 for (i = disks; i--; ) {
1896 struct r5dev *dev = &sh->dev[i];
1897 if (i != pd_idx)
1898 xor_srcs[count++] = dev->page;
1899 }
1900 }
1901
91c00924
DW		 1902 /* 1/ if we prexor'd then the dest is reused as a source
1903 * 2/ if we did not prexor then we are redoing the parity
1904 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
1905 * for the synchronous xor case
1906 */
59fc630b 1907 last_stripe = !head_sh->batch_head ||
1908 list_first_entry(&sh->batch_list,
1909 struct stripe_head, batch_list) == head_sh;
1910 if (last_stripe) {
1911 flags = ASYNC_TX_ACK |
1912 (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
1913
1914 atomic_inc(&head_sh->count);
1915 init_async_submit(&submit, flags, tx, ops_complete_reconstruct, head_sh,
1916 to_addr_conv(sh, percpu, j));
1917 } else {
1918 flags = prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST;
1919 init_async_submit(&submit, flags, tx, NULL, NULL,
1920 to_addr_conv(sh, percpu, j));
1921 }
91c00924 1922
a08abd8c
DW		 1923 if (unlikely(count == 1))
1924 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
1925 else
1926 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
59fc630b 1927 if (!last_stripe) {
1928 j++;
1929 sh = list_first_entry(&sh->batch_list, struct stripe_head,
1930 batch_list);
1931 goto again;
1932 }
91c00924
DW		 1933}
1934
ac6b53b6
DW		 1935static void
1936ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu,
1937 struct dma_async_tx_descriptor *tx)
1938{
1939 struct async_submit_ctl submit;
59fc630b 1940 struct page **blocks;
1941 int count, i, j = 0;
1942 struct stripe_head *head_sh = sh;
1943 int last_stripe;
584acdd4
MS		 1944 int synflags;
1945 unsigned long txflags;
ac6b53b6
DW		 1946
1947 pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector);
1948
620125f2
SL		 1949 for (i = 0; i < sh->disks; i++) {
1950 if (sh->pd_idx == i || sh->qd_idx == i)
1951 continue;
1952 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1953 break;
1954 }
1955 if (i >= sh->disks) {
1956 atomic_inc(&sh->count);
620125f2
SL		 1957 set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
1958 set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
1959 ops_complete_reconstruct(sh);
1960 return;
1961 }
1962
59fc630b 1963again:
1964 blocks = to_addr_page(percpu, j);
584acdd4
MS		 1965
1966 if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
1967 synflags = SYNDROME_SRC_WRITTEN;
1968 txflags = ASYNC_TX_ACK | ASYNC_TX_PQ_XOR_DST;
1969 } else {
1970 synflags = SYNDROME_SRC_ALL;
1971 txflags = ASYNC_TX_ACK;
1972 }
1973
1974 count = set_syndrome_sources(blocks, sh, synflags);
59fc630b 1975 last_stripe = !head_sh->batch_head ||
1976 list_first_entry(&sh->batch_list,
1977 struct stripe_head, batch_list) == head_sh;
1978
1979 if (last_stripe) {
1980 atomic_inc(&head_sh->count);
584acdd4 1981 init_async_submit(&submit, txflags, tx, ops_complete_reconstruct,
59fc630b 1982 head_sh, to_addr_conv(sh, percpu, j));
1983 } else
1984 init_async_submit(&submit, 0, tx, NULL, NULL,
1985 to_addr_conv(sh, percpu, j));
48769695 1986 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
59fc630b 1987 if (!last_stripe) {
1988 j++;
1989 sh = list_first_entry(&sh->batch_list, struct stripe_head,
1990 batch_list);
1991 goto again;
1992 }
91c00924
DW		 1993}
1994
1995static void ops_complete_check(void *stripe_head_ref)
1996{
1997 struct stripe_head *sh = stripe_head_ref;
91c00924 1998
e46b272b 1999 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 2000 (unsigned long long)sh->sector);
2001
ecc65c9b 2002 sh->check_state = check_state_check_result;
91c00924 2003 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 2004 raid5_release_stripe(sh);
91c00924
DW		 2005}
2006
ac6b53b6 2007static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 2008{
91c00924 2009 int disks = sh->disks;
ac6b53b6
DW		 2010 int pd_idx = sh->pd_idx;
2011 int qd_idx = sh->qd_idx;
2012 struct page *xor_dest;
46d5b785 2013 struct page **xor_srcs = to_addr_page(percpu, 0);
91c00924 2014 struct dma_async_tx_descriptor *tx;
a08abd8c 2015 struct async_submit_ctl submit;
ac6b53b6
DW		 2016 int count;
2017 int i;
91c00924 2018
e46b272b 2019 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 2020 (unsigned long long)sh->sector);
2021
59fc630b 2022 BUG_ON(sh->batch_head);
ac6b53b6
DW		 2023 count = 0;
2024 xor_dest = sh->dev[pd_idx].page;
2025 xor_srcs[count++] = xor_dest;
91c00924 2026 for (i = disks; i--; ) {
ac6b53b6
DW		 2027 if (i == pd_idx || i == qd_idx)
2028 continue;
2029 xor_srcs[count++] = sh->dev[i].page;
91c00924
DW		 2030 }
2031
d6f38f31 2032 init_async_submit(&submit, 0, NULL, NULL, NULL,
46d5b785 2033 to_addr_conv(sh, percpu, 0));
099f53cb 2034 tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
a08abd8c 2035 &sh->ops.zero_sum_result, &submit);
91c00924 2036
91c00924 2037 atomic_inc(&sh->count);
a08abd8c
DW		 2038 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL);
2039 tx = async_trigger_callback(&submit);
91c00924
DW		 2040}
2041
ac6b53b6
DW		 2042static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp)
2043{
46d5b785 2044 struct page **srcs = to_addr_page(percpu, 0);
ac6b53b6
DW		 2045 struct async_submit_ctl submit;
2046 int count;
2047
2048 pr_debug("%s: stripe %llu checkp: %d\n", __func__,
2049 (unsigned long long)sh->sector, checkp);
2050
59fc630b 2051 BUG_ON(sh->batch_head);
584acdd4 2052 count = set_syndrome_sources(srcs, sh, SYNDROME_SRC_ALL);
ac6b53b6
DW		 2053 if (!checkp)
2054 srcs[count] = NULL;
91c00924 2055
91c00924 2056 atomic_inc(&sh->count);
ac6b53b6 2057 init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check,
46d5b785 2058 sh, to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 2059 async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE,
2060 &sh->ops.zero_sum_result, percpu->spare_page, &submit);
91c00924
DW		 2061}
2062
51acbcec 2063static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
91c00924
DW		 2064{
2065 int overlap_clear = 0, i, disks = sh->disks;
2066 struct dma_async_tx_descriptor *tx = NULL;
d1688a6d 2067 struct r5conf *conf = sh->raid_conf;
ac6b53b6 2068 int level = conf->level;
d6f38f31
DW		 2069 struct raid5_percpu *percpu;
2070 unsigned long cpu;
91c00924 2071
d6f38f31
DW		 2072 cpu = get_cpu();
2073 percpu = per_cpu_ptr(conf->percpu, cpu);
83de75cc 2074 if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) {
91c00924
DW		 2075 ops_run_biofill(sh);
2076 overlap_clear++;
2077 }
2078
7b3a871e 2079 if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) {
ac6b53b6
DW		 2080 if (level < 6)
2081 tx = ops_run_compute5(sh, percpu);
2082 else {
2083 if (sh->ops.target2 < 0 || sh->ops.target < 0)
2084 tx = ops_run_compute6_1(sh, percpu);
2085 else
2086 tx = ops_run_compute6_2(sh, percpu);
2087 }
2088 /* terminate the chain if reconstruct is not set to be run */
2089 if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request))
7b3a871e
DW		 2090 async_tx_ack(tx);
2091 }
91c00924 2092
584acdd4
MS		 2093 if (test_bit(STRIPE_OP_PREXOR, &ops_request)) {
2094 if (level < 6)
2095 tx = ops_run_prexor5(sh, percpu, tx);
2096 else
2097 tx = ops_run_prexor6(sh, percpu, tx);
2098 }
91c00924 2099
ae1713e2
AP		 2100 if (test_bit(STRIPE_OP_PARTIAL_PARITY, &ops_request))
2101 tx = ops_run_partial_parity(sh, percpu, tx);
2102
600aa109 2103 if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) {
d8ee0728 2104 tx = ops_run_biodrain(sh, tx);
91c00924
DW		 2105 overlap_clear++;
2106 }
2107
ac6b53b6
DW		 2108 if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) {
2109 if (level < 6)
2110 ops_run_reconstruct5(sh, percpu, tx);
2111 else
2112 ops_run_reconstruct6(sh, percpu, tx);
2113 }
91c00924 2114
ac6b53b6
DW		 2115 if (test_bit(STRIPE_OP_CHECK, &ops_request)) {
2116 if (sh->check_state == check_state_run)
2117 ops_run_check_p(sh, percpu);
2118 else if (sh->check_state == check_state_run_q)
2119 ops_run_check_pq(sh, percpu, 0);
2120 else if (sh->check_state == check_state_run_pq)
2121 ops_run_check_pq(sh, percpu, 1);
2122 else
2123 BUG();
2124 }
91c00924 2125
59fc630b 2126 if (overlap_clear && !sh->batch_head)
91c00924
DW		 2127 for (i = disks; i--; ) {
2128 struct r5dev *dev = &sh->dev[i];
2129 if (test_and_clear_bit(R5_Overlap, &dev->flags))
2130 wake_up(&sh->raid_conf->wait_for_overlap);
2131 }
d6f38f31 2132 put_cpu();
91c00924
DW		 2133}
2134
845b9e22
AP		 2135static void free_stripe(struct kmem_cache *sc, struct stripe_head *sh)
2136{
2137 if (sh->ppl_page)
2138 __free_page(sh->ppl_page);
2139 kmem_cache_free(sc, sh);
2140}
2141
5f9d1fde 2142static struct stripe_head *alloc_stripe(struct kmem_cache *sc, gfp_t gfp,
845b9e22 2143 int disks, struct r5conf *conf)
f18c1a35
N		 2144{
2145 struct stripe_head *sh;
5f9d1fde 2146 int i;
f18c1a35
N		 2147
2148 sh = kmem_cache_zalloc(sc, gfp);
2149 if (sh) {
2150 spin_lock_init(&sh->stripe_lock);
2151 spin_lock_init(&sh->batch_lock);
2152 INIT_LIST_HEAD(&sh->batch_list);
2153 INIT_LIST_HEAD(&sh->lru);
a39f7afd 2154 INIT_LIST_HEAD(&sh->r5c);
d7bd398e 2155 INIT_LIST_HEAD(&sh->log_list);
f18c1a35 2156 atomic_set(&sh->count, 1);
845b9e22 2157 sh->raid_conf = conf;
a39f7afd 2158 sh->log_start = MaxSector;
5f9d1fde
SL		 2159 for (i = 0; i < disks; i++) {
2160 struct r5dev *dev = &sh->dev[i];
2161
3a83f467
ML		 2162 bio_init(&dev->req, &dev->vec, 1);
2163 bio_init(&dev->rreq, &dev->rvec, 1);
5f9d1fde 2164 }
845b9e22
AP		 2165
2166 if (raid5_has_ppl(conf)) {
2167 sh->ppl_page = alloc_page(gfp);
2168 if (!sh->ppl_page) {
2169 free_stripe(sc, sh);
2170 sh = NULL;
2171 }
2172 }
f18c1a35
N		 2173 }
2174 return sh;
2175}
486f0644 2176static int grow_one_stripe(struct r5conf *conf, gfp_t gfp)
1da177e4
LT		 2177{
2178 struct stripe_head *sh;
f18c1a35 2179
845b9e22 2180 sh = alloc_stripe(conf->slab_cache, gfp, conf->pool_size, conf);
3f294f4f
N		 2181 if (!sh)
2182 return 0;
6ce32846 2183
a9683a79 2184 if (grow_buffers(sh, gfp)) {
e4e11e38 2185 shrink_buffers(sh);
845b9e22 2186 free_stripe(conf->slab_cache, sh);
3f294f4f
N		 2187 return 0;
2188 }
486f0644
N		 2189 sh->hash_lock_index =
2190 conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS;
3f294f4f 2191 /* we just created an active stripe so... */
3f294f4f 2192 atomic_inc(&conf->active_stripes);
59fc630b 2193
6d036f7d 2194 raid5_release_stripe(sh);
486f0644 2195 conf->max_nr_stripes++;
3f294f4f
N		 2196 return 1;
2197}
2198
d1688a6d 2199static int grow_stripes(struct r5conf *conf, int num)
3f294f4f 2200{
e18b890b 2201 struct kmem_cache *sc;
53b8d89d 2202 size_t namelen = sizeof(conf->cache_name[0]);
5e5e3e78 2203 int devs = max(conf->raid_disks, conf->previous_raid_disks);
1da177e4 2204
f4be6b43 2205 if (conf->mddev->gendisk)
53b8d89d 2206 snprintf(conf->cache_name[0], namelen,
f4be6b43
N		 2207 "raid%d-%s", conf->level, mdname(conf->mddev));
2208 else
53b8d89d 2209 snprintf(conf->cache_name[0], namelen,
f4be6b43 2210 "raid%d-%p", conf->level, conf->mddev);
53b8d89d 2211 snprintf(conf->cache_name[1], namelen, "%.27s-alt", conf->cache_name[0]);
f4be6b43 2212
ad01c9e3
N		 2213 conf->active_name = 0;
2214 sc = kmem_cache_create(conf->cache_name[conf->active_name],
1da177e4 2215 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
20c2df83 2216 0, 0, NULL);
1da177e4
LT		 2217 if (!sc)
2218 return 1;
2219 conf->slab_cache = sc;
ad01c9e3 2220 conf->pool_size = devs;
486f0644
N		 2221 while (num--)
2222 if (!grow_one_stripe(conf, GFP_KERNEL))
1da177e4 2223 return 1;
486f0644 2224
1da177e4
LT		 2225 return 0;
2226}
29269553 2227
d6f38f31
DW		 2228/**
2229 * scribble_len - return the required size of the scribble region
2230 * @num - total number of disks in the array
2231 *
2232 * The size must be enough to contain:
2233 * 1/ a struct page pointer for each device in the array +2
2234 * 2/ room to convert each entry in (1) to its corresponding dma
2235 * (dma_map_page()) or page (page_address()) address.
2236 *
2237 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
2238 * calculate over all devices (not just the data blocks), using zeros in place
2239 * of the P and Q blocks.
2240 */
46d5b785 2241static struct flex_array *scribble_alloc(int num, int cnt, gfp_t flags)
d6f38f31 2242{
46d5b785 2243 struct flex_array *ret;
d6f38f31
DW		 2244 size_t len;
2245
2246 len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2);
46d5b785 2247 ret = flex_array_alloc(len, cnt, flags);
2248 if (!ret)
2249 return NULL;
2250 /* always prealloc all elements, so no locking is required */
2251 if (flex_array_prealloc(ret, 0, cnt, flags)) {
2252 flex_array_free(ret);
2253 return NULL;
2254 }
2255 return ret;
d6f38f31
DW		 2256}
2257
738a2738
N		 2258static int resize_chunks(struct r5conf *conf, int new_disks, int new_sectors)
2259{
2260 unsigned long cpu;
2261 int err = 0;
2262
27a353c0
SL		 2263 /*
2264 * Never shrink. And mddev_suspend() could deadlock if this is called
2265 * from raid5d. In that case, scribble_disks and scribble_sectors
2266 * should equal to new_disks and new_sectors
2267 */
2268 if (conf->scribble_disks >= new_disks &&
2269 conf->scribble_sectors >= new_sectors)
2270 return 0;
738a2738
N		 2271 mddev_suspend(conf->mddev);
2272 get_online_cpus();
2273 for_each_present_cpu(cpu) {
2274 struct raid5_percpu *percpu;
2275 struct flex_array *scribble;
2276
2277 percpu = per_cpu_ptr(conf->percpu, cpu);
2278 scribble = scribble_alloc(new_disks,
2279 new_sectors / STRIPE_SECTORS,
2280 GFP_NOIO);
2281
2282 if (scribble) {
2283 flex_array_free(percpu->scribble);
2284 percpu->scribble = scribble;
2285 } else {
2286 err = -ENOMEM;
2287 break;
2288 }
2289 }
2290 put_online_cpus();
2291 mddev_resume(conf->mddev);
27a353c0
SL		 2292 if (!err) {
2293 conf->scribble_disks = new_disks;
2294 conf->scribble_sectors = new_sectors;
2295 }
738a2738
N		 2296 return err;
2297}
2298
d1688a6d 2299static int resize_stripes(struct r5conf *conf, int newsize)
ad01c9e3
N		 2300{
2301 /* Make all the stripes able to hold 'newsize' devices.
2302 * New slots in each stripe get 'page' set to a new page.
2303 *
2304 * This happens in stages:
2305 * 1/ create a new kmem_cache and allocate the required number of
2306 * stripe_heads.
83f0d77a 2307 * 2/ gather all the old stripe_heads and transfer the pages across
ad01c9e3
N		 2308 * to the new stripe_heads. This will have the side effect of
2309 * freezing the array as once all stripe_heads have been collected,
2310 * no IO will be possible. Old stripe heads are freed once their
2311 * pages have been transferred over, and the old kmem_cache is
2312 * freed when all stripes are done.
2313 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
3560741e 2314 * we simple return a failure status - no need to clean anything up.
ad01c9e3
N		 2315 * 4/ allocate new pages for the new slots in the new stripe_heads.
2316 * If this fails, we don't bother trying the shrink the
2317 * stripe_heads down again, we just leave them as they are.
2318 * As each stripe_head is processed the new one is released into
2319 * active service.
2320 *
2321 * Once step2 is started, we cannot afford to wait for a write,
2322 * so we use GFP_NOIO allocations.
2323 */
2324 struct stripe_head *osh, *nsh;
2325 LIST_HEAD(newstripes);
2326 struct disk_info *ndisks;
2214c260 2327 int err = 0;
e18b890b 2328 struct kmem_cache *sc;
ad01c9e3 2329 int i;
566c09c5 2330 int hash, cnt;
ad01c9e3 2331
2214c260 2332 md_allow_write(conf->mddev);
2a2275d6 2333
ad01c9e3
N		 2334 /* Step 1 */
2335 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
2336 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
20c2df83 2337 0, 0, NULL);
ad01c9e3
N		 2338 if (!sc)
2339 return -ENOMEM;
2340
2d5b569b
N		 2341 /* Need to ensure auto-resizing doesn't interfere */
2342 mutex_lock(&conf->cache_size_mutex);
2343
ad01c9e3 2344 for (i = conf->max_nr_stripes; i; i--) {
845b9e22 2345 nsh = alloc_stripe(sc, GFP_KERNEL, newsize, conf);
ad01c9e3
N		 2346 if (!nsh)
2347 break;
2348
ad01c9e3
N		 2349 list_add(&nsh->lru, &newstripes);
2350 }
2351 if (i) {
2352 /* didn't get enough, give up */
2353 while (!list_empty(&newstripes)) {
2354 nsh = list_entry(newstripes.next, struct stripe_head, lru);
2355 list_del(&nsh->lru);
845b9e22 2356 free_stripe(sc, nsh);
ad01c9e3
N		 2357 }
2358 kmem_cache_destroy(sc);
2d5b569b 2359 mutex_unlock(&conf->cache_size_mutex);
ad01c9e3
N		 2360 return -ENOMEM;
2361 }
2362 /* Step 2 - Must use GFP_NOIO now.
2363 * OK, we have enough stripes, start collecting inactive
2364 * stripes and copying them over
2365 */
566c09c5
SL		 2366 hash = 0;
2367 cnt = 0;
ad01c9e3 2368 list_for_each_entry(nsh, &newstripes, lru) {
566c09c5 2369 lock_device_hash_lock(conf, hash);
6ab2a4b8 2370 wait_event_cmd(conf->wait_for_stripe,
566c09c5
SL		 2371 !list_empty(conf->inactive_list + hash),
2372 unlock_device_hash_lock(conf, hash),
2373 lock_device_hash_lock(conf, hash));
2374 osh = get_free_stripe(conf, hash);
2375 unlock_device_hash_lock(conf, hash);
f18c1a35 2376
d592a996 2377 for(i=0; i<conf->pool_size; i++) {
ad01c9e3 2378 nsh->dev[i].page = osh->dev[i].page;
d592a996
SL		 2379 nsh->dev[i].orig_page = osh->dev[i].page;
2380 }
566c09c5 2381 nsh->hash_lock_index = hash;
845b9e22 2382 free_stripe(conf->slab_cache, osh);
566c09c5
SL		 2383 cnt++;
2384 if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS +
2385 !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) {
2386 hash++;
2387 cnt = 0;
2388 }
ad01c9e3
N		 2389 }
2390 kmem_cache_destroy(conf->slab_cache);
2391
2392 /* Step 3.
2393 * At this point, we are holding all the stripes so the array
2394 * is completely stalled, so now is a good time to resize
d6f38f31 2395 * conf->disks and the scribble region
ad01c9e3 2396 */
6396bb22 2397 ndisks = kcalloc(newsize, sizeof(struct disk_info), GFP_NOIO);
ad01c9e3 2398 if (ndisks) {
d7bd398e 2399 for (i = 0; i < conf->pool_size; i++)
ad01c9e3 2400 ndisks[i] = conf->disks[i];
d7bd398e
SL		 2401
2402 for (i = conf->pool_size; i < newsize; i++) {
2403 ndisks[i].extra_page = alloc_page(GFP_NOIO);
2404 if (!ndisks[i].extra_page)
2405 err = -ENOMEM;
2406 }
2407
2408 if (err) {
2409 for (i = conf->pool_size; i < newsize; i++)
2410 if (ndisks[i].extra_page)
2411 put_page(ndisks[i].extra_page);
2412 kfree(ndisks);
2413 } else {
2414 kfree(conf->disks);
2415 conf->disks = ndisks;
2416 }
ad01c9e3
N		 2417 } else
2418 err = -ENOMEM;
2419
2d5b569b 2420 mutex_unlock(&conf->cache_size_mutex);
583da48e
DY		 2421
2422 conf->slab_cache = sc;
2423 conf->active_name = 1-conf->active_name;
2424
ad01c9e3
N		 2425 /* Step 4, return new stripes to service */
2426 while(!list_empty(&newstripes)) {
2427 nsh = list_entry(newstripes.next, struct stripe_head, lru);
2428 list_del_init(&nsh->lru);
d6f38f31 2429
ad01c9e3
N		 2430 for (i=conf->raid_disks; i < newsize; i++)
2431 if (nsh->dev[i].page == NULL) {
2432 struct page *p = alloc_page(GFP_NOIO);
2433 nsh->dev[i].page = p;
d592a996 2434 nsh->dev[i].orig_page = p;
ad01c9e3
N		 2435 if (!p)
2436 err = -ENOMEM;
2437 }
6d036f7d 2438 raid5_release_stripe(nsh);
ad01c9e3
N		 2439 }
2440 /* critical section pass, GFP_NOIO no longer needed */
2441
6e9eac2d
N		 2442 if (!err)
2443 conf->pool_size = newsize;
ad01c9e3
N		 2444 return err;
2445}
1da177e4 2446
486f0644 2447static int drop_one_stripe(struct r5conf *conf)
1da177e4
LT		 2448{
2449 struct stripe_head *sh;
49895bcc 2450 int hash = (conf->max_nr_stripes - 1) & STRIPE_HASH_LOCKS_MASK;
1da177e4 2451
566c09c5
SL		 2452 spin_lock_irq(conf->hash_locks + hash);
2453 sh = get_free_stripe(conf, hash);
2454 spin_unlock_irq(conf->hash_locks + hash);
3f294f4f
N		 2455 if (!sh)
2456 return 0;
78bafebd 2457 BUG_ON(atomic_read(&sh->count));
e4e11e38 2458 shrink_buffers(sh);
845b9e22 2459 free_stripe(conf->slab_cache, sh);
3f294f4f 2460 atomic_dec(&conf->active_stripes);
486f0644 2461 conf->max_nr_stripes--;
3f294f4f
N		 2462 return 1;
2463}
2464
d1688a6d 2465static void shrink_stripes(struct r5conf *conf)
3f294f4f 2466{
486f0644
N		 2467 while (conf->max_nr_stripes &&
2468 drop_one_stripe(conf))
2469 ;
3f294f4f 2470
644df1a8 2471 kmem_cache_destroy(conf->slab_cache);
1da177e4
LT		 2472 conf->slab_cache = NULL;
2473}
2474
4246a0b6 2475static void raid5_end_read_request(struct bio * bi)
1da177e4 2476{
99c0fb5f 2477 struct stripe_head *sh = bi->bi_private;
d1688a6d 2478 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 2479 int disks = sh->disks, i;
d6950432 2480 char b[BDEVNAME_SIZE];
dd054fce 2481 struct md_rdev *rdev = NULL;
05616be5 2482 sector_t s;
1da177e4
LT		 2483
2484 for (i=0 ; i<disks; i++)
2485 if (bi == &sh->dev[i].req)
2486 break;
2487
4246a0b6 2488 pr_debug("end_read_request %llu/%d, count: %d, error %d.\n",
45b4233c 2489 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
4e4cbee9 2490 bi->bi_status);
1da177e4 2491 if (i == disks) {
5f9d1fde 2492 bio_reset(bi);
1da177e4 2493 BUG();
6712ecf8 2494 return;
1da177e4 2495 }
14a75d3e 2496 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
dd054fce
N		 2497 /* If replacement finished while this request was outstanding,
2498 * 'replacement' might be NULL already.
2499 * In that case it moved down to 'rdev'.
2500 * rdev is not removed until all requests are finished.
2501 */
14a75d3e 2502 rdev = conf->disks[i].replacement;
dd054fce 2503 if (!rdev)
14a75d3e 2504 rdev = conf->disks[i].rdev;
1da177e4 2505
05616be5
N		 2506 if (use_new_offset(conf, sh))
2507 s = sh->sector + rdev->new_data_offset;
2508 else
2509 s = sh->sector + rdev->data_offset;
4e4cbee9 2510 if (!bi->bi_status) {
1da177e4 2511 set_bit(R5_UPTODATE, &sh->dev[i].flags);
4e5314b5 2512 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
14a75d3e
N		 2513 /* Note that this cannot happen on a
2514 * replacement device. We just fail those on
2515 * any error
2516 */
cc6167b4
N		 2517 pr_info_ratelimited(
2518 "md/raid:%s: read error corrected (%lu sectors at %llu on %s)\n",
8bda470e 2519 mdname(conf->mddev), STRIPE_SECTORS,
05616be5 2520 (unsigned long long)s,
8bda470e 2521 bdevname(rdev->bdev, b));
ddd5115f 2522 atomic_add(STRIPE_SECTORS, &rdev->corrected_errors);
4e5314b5
N		 2523 clear_bit(R5_ReadError, &sh->dev[i].flags);
2524 clear_bit(R5_ReWrite, &sh->dev[i].flags);
3f9e7c14 2525 } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2526 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2527
86aa1397
SL		 2528 if (test_bit(R5_InJournal, &sh->dev[i].flags))
2529 /*
2530 * end read for a page in journal, this
2531 * must be preparing for prexor in rmw
2532 */
2533 set_bit(R5_OrigPageUPTDODATE, &sh->dev[i].flags);
2534
14a75d3e
N		 2535 if (atomic_read(&rdev->read_errors))
2536 atomic_set(&rdev->read_errors, 0);
1da177e4 2537 } else {
14a75d3e 2538 const char *bdn = bdevname(rdev->bdev, b);
ba22dcbf 2539 int retry = 0;
2e8ac303 2540 int set_bad = 0;
d6950432 2541
1da177e4 2542 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
d6950432 2543 atomic_inc(&rdev->read_errors);
14a75d3e 2544 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
cc6167b4
N		 2545 pr_warn_ratelimited(
2546 "md/raid:%s: read error on replacement device (sector %llu on %s).\n",
14a75d3e 2547 mdname(conf->mddev),
05616be5 2548 (unsigned long long)s,
14a75d3e 2549 bdn);
2e8ac303 2550 else if (conf->mddev->degraded >= conf->max_degraded) {
2551 set_bad = 1;
cc6167b4
N		 2552 pr_warn_ratelimited(
2553 "md/raid:%s: read error not correctable (sector %llu on %s).\n",
8bda470e 2554 mdname(conf->mddev),
05616be5 2555 (unsigned long long)s,
8bda470e 2556 bdn);
2e8ac303 2557 } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) {
4e5314b5 2558 /* Oh, no!!! */
2e8ac303 2559 set_bad = 1;
cc6167b4
N		 2560 pr_warn_ratelimited(
2561 "md/raid:%s: read error NOT corrected!! (sector %llu on %s).\n",
8bda470e 2562 mdname(conf->mddev),
05616be5 2563 (unsigned long long)s,
8bda470e 2564 bdn);
2e8ac303 2565 } else if (atomic_read(&rdev->read_errors)
ba22dcbf 2566 > conf->max_nr_stripes)
cc6167b4 2567 pr_warn("md/raid:%s: Too many read errors, failing device %s.\n",
d6950432 2568 mdname(conf->mddev), bdn);
ba22dcbf
N		 2569 else
2570 retry = 1;
edfa1f65
BY		 2571 if (set_bad && test_bit(In_sync, &rdev->flags)
2572 && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2573 retry = 1;
ba22dcbf 2574 if (retry)
3f9e7c14 2575 if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) {
2576 set_bit(R5_ReadError, &sh->dev[i].flags);
2577 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2578 } else
2579 set_bit(R5_ReadNoMerge, &sh->dev[i].flags);
ba22dcbf 2580 else {
4e5314b5
N		 2581 clear_bit(R5_ReadError, &sh->dev[i].flags);
2582 clear_bit(R5_ReWrite, &sh->dev[i].flags);
2e8ac303 2583 if (!(set_bad
2584 && test_bit(In_sync, &rdev->flags)
2585 && rdev_set_badblocks(
2586 rdev, sh->sector, STRIPE_SECTORS, 0)))
2587 md_error(conf->mddev, rdev);
ba22dcbf 2588 }
1da177e4 2589 }
14a75d3e 2590 rdev_dec_pending(rdev, conf->mddev);
c9445555 2591 bio_reset(bi);
1da177e4
LT		 2592 clear_bit(R5_LOCKED, &sh->dev[i].flags);
2593 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 2594 raid5_release_stripe(sh);
1da177e4
LT		 2595}
2596
4246a0b6 2597static void raid5_end_write_request(struct bio *bi)
1da177e4 2598{
99c0fb5f 2599 struct stripe_head *sh = bi->bi_private;
d1688a6d 2600 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 2601 int disks = sh->disks, i;
977df362 2602 struct md_rdev *uninitialized_var(rdev);
b84db560
N		 2603 sector_t first_bad;
2604 int bad_sectors;
977df362 2605 int replacement = 0;
1da177e4 2606
977df362
N		 2607 for (i = 0 ; i < disks; i++) {
2608 if (bi == &sh->dev[i].req) {
2609 rdev = conf->disks[i].rdev;
1da177e4 2610 break;
977df362
N		 2611 }
2612 if (bi == &sh->dev[i].rreq) {
2613 rdev = conf->disks[i].replacement;
dd054fce
N		 2614 if (rdev)
2615 replacement = 1;
2616 else
2617 /* rdev was removed and 'replacement'
2618 * replaced it. rdev is not removed
2619 * until all requests are finished.
2620 */
2621 rdev = conf->disks[i].rdev;
977df362
N		 2622 break;
2623 }
2624 }
4246a0b6 2625 pr_debug("end_write_request %llu/%d, count %d, error: %d.\n",
1da177e4 2626 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
4e4cbee9 2627 bi->bi_status);
1da177e4 2628 if (i == disks) {
5f9d1fde 2629 bio_reset(bi);
1da177e4 2630 BUG();
6712ecf8 2631 return;
1da177e4
LT		 2632 }
2633
977df362 2634 if (replacement) {
4e4cbee9 2635 if (bi->bi_status)
977df362
N		 2636 md_error(conf->mddev, rdev);
2637 else if (is_badblock(rdev, sh->sector,
2638 STRIPE_SECTORS,
2639 &first_bad, &bad_sectors))
2640 set_bit(R5_MadeGoodRepl, &sh->dev[i].flags);
2641 } else {
4e4cbee9 2642 if (bi->bi_status) {
9f97e4b1 2643 set_bit(STRIPE_DEGRADED, &sh->state);
977df362
N		 2644 set_bit(WriteErrorSeen, &rdev->flags);
2645 set_bit(R5_WriteError, &sh->dev[i].flags);
3a6de292
N		 2646 if (!test_and_set_bit(WantReplacement, &rdev->flags))
2647 set_bit(MD_RECOVERY_NEEDED,
2648 &rdev->mddev->recovery);
977df362
N		 2649 } else if (is_badblock(rdev, sh->sector,
2650 STRIPE_SECTORS,
c0b32972 2651 &first_bad, &bad_sectors)) {
977df362 2652 set_bit(R5_MadeGood, &sh->dev[i].flags);
c0b32972
N		 2653 if (test_bit(R5_ReadError, &sh->dev[i].flags))
2654 /* That was a successful write so make
2655 * sure it looks like we already did
2656 * a re-write.
2657 */
2658 set_bit(R5_ReWrite, &sh->dev[i].flags);
2659 }
977df362
N		 2660 }
2661 rdev_dec_pending(rdev, conf->mddev);
1da177e4 2662
4e4cbee9 2663 if (sh->batch_head && bi->bi_status && !replacement)
72ac7330 2664 set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state);
2665
c9445555 2666 bio_reset(bi);
977df362
N		 2667 if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags))
2668 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1da177e4 2669 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 2670 raid5_release_stripe(sh);
59fc630b 2671
2672 if (sh->batch_head && sh != sh->batch_head)
6d036f7d 2673 raid5_release_stripe(sh->batch_head);
1da177e4
LT		 2674}
2675
849674e4 2676static void raid5_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 2677{
2678 char b[BDEVNAME_SIZE];
d1688a6d 2679 struct r5conf *conf = mddev->private;
908f4fbd 2680 unsigned long flags;
0c55e022 2681 pr_debug("raid456: error called\n");
1da177e4 2682
908f4fbd 2683 spin_lock_irqsave(&conf->device_lock, flags);
fb73b357
MT		 2684
2685 if (test_bit(In_sync, &rdev->flags) &&
2686 mddev->degraded == conf->max_degraded) {
2687 /*
2688 * Don't allow to achieve failed state
2689 * Don't try to recover this device
2690 */
2691 conf->recovery_disabled = mddev->recovery_disabled;
2692 spin_unlock_irqrestore(&conf->device_lock, flags);
2693 return;
2694 }
2695
aff69d89 2696 set_bit(Faulty, &rdev->flags);
908f4fbd 2697 clear_bit(In_sync, &rdev->flags);
2e38a37f 2698 mddev->degraded = raid5_calc_degraded(conf);
908f4fbd
N		 2699 spin_unlock_irqrestore(&conf->device_lock, flags);
2700 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2701
de393cde 2702 set_bit(Blocked, &rdev->flags);
2953079c
SL		 2703 set_mask_bits(&mddev->sb_flags, 0,
2704 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING));
cc6167b4
N		 2705 pr_crit("md/raid:%s: Disk failure on %s, disabling device.\n"
2706 "md/raid:%s: Operation continuing on %d devices.\n",
2707 mdname(mddev),
2708 bdevname(rdev->bdev, b),
2709 mdname(mddev),
2710 conf->raid_disks - mddev->degraded);
70d466f7 2711 r5c_update_on_rdev_error(mddev, rdev);
16a53ecc 2712}
1da177e4
LT		 2713
2714/*
2715 * Input: a 'big' sector number,
2716 * Output: index of the data and parity disk, and the sector # in them.
2717 */
6d036f7d
SL		 2718sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector,
2719 int previous, int *dd_idx,
2720 struct stripe_head *sh)
1da177e4 2721{
6e3b96ed 2722 sector_t stripe, stripe2;
35f2a591 2723 sector_t chunk_number;
1da177e4 2724 unsigned int chunk_offset;
911d4ee8 2725 int pd_idx, qd_idx;
67cc2b81 2726 int ddf_layout = 0;
1da177e4 2727 sector_t new_sector;
e183eaed
N		 2728 int algorithm = previous ? conf->prev_algo
2729 : conf->algorithm;
09c9e5fa
AN		 2730 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2731 : conf->chunk_sectors;
112bf897
N		 2732 int raid_disks = previous ? conf->previous_raid_disks
2733 : conf->raid_disks;
2734 int data_disks = raid_disks - conf->max_degraded;
1da177e4
LT		 2735
2736 /* First compute the information on this sector */
2737
2738 /*
2739 * Compute the chunk number and the sector offset inside the chunk
2740 */
2741 chunk_offset = sector_div(r_sector, sectors_per_chunk);
2742 chunk_number = r_sector;
1da177e4
LT		 2743
2744 /*
2745 * Compute the stripe number
2746 */
35f2a591
N		 2747 stripe = chunk_number;
2748 *dd_idx = sector_div(stripe, data_disks);
6e3b96ed 2749 stripe2 = stripe;
1da177e4
LT		 2750 /*
2751 * Select the parity disk based on the user selected algorithm.
2752 */
84789554 2753 pd_idx = qd_idx = -1;
16a53ecc
N		 2754 switch(conf->level) {
2755 case 4:
911d4ee8 2756 pd_idx = data_disks;
16a53ecc
N		 2757 break;
2758 case 5:
e183eaed 2759 switch (algorithm) {
1da177e4 2760 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2761 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2762 if (*dd_idx >= pd_idx)
1da177e4
LT		 2763 (*dd_idx)++;
2764 break;
2765 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2766 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2767 if (*dd_idx >= pd_idx)
1da177e4
LT		 2768 (*dd_idx)++;
2769 break;
2770 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2771 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2772 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4
LT		 2773 break;
2774 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2775 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2776 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4 2777 break;
99c0fb5f
N		 2778 case ALGORITHM_PARITY_0:
2779 pd_idx = 0;
2780 (*dd_idx)++;
2781 break;
2782 case ALGORITHM_PARITY_N:
2783 pd_idx = data_disks;
2784 break;
1da177e4 2785 default:
99c0fb5f 2786 BUG();
16a53ecc
N		 2787 }
2788 break;
2789 case 6:
2790
e183eaed 2791 switch (algorithm) {
16a53ecc 2792 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2793 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2794 qd_idx = pd_idx + 1;
2795 if (pd_idx == raid_disks-1) {
99c0fb5f 2796 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2797 qd_idx = 0;
2798 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2799 (*dd_idx) += 2; /* D D P Q D */
2800 break;
2801 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2802 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2803 qd_idx = pd_idx + 1;
2804 if (pd_idx == raid_disks-1) {
99c0fb5f 2805 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2806 qd_idx = 0;
2807 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2808 (*dd_idx) += 2; /* D D P Q D */
2809 break;
2810 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2811 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2812 qd_idx = (pd_idx + 1) % raid_disks;
2813 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc
N		 2814 break;
2815 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2816 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2817 qd_idx = (pd_idx + 1) % raid_disks;
2818 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc 2819 break;
99c0fb5f
N		 2820
2821 case ALGORITHM_PARITY_0:
2822 pd_idx = 0;
2823 qd_idx = 1;
2824 (*dd_idx) += 2;
2825 break;
2826 case ALGORITHM_PARITY_N:
2827 pd_idx = data_disks;
2828 qd_idx = data_disks + 1;
2829 break;
2830
2831 case ALGORITHM_ROTATING_ZERO_RESTART:
2832 /* Exactly the same as RIGHT_ASYMMETRIC, but or
2833 * of blocks for computing Q is different.
2834 */
6e3b96ed 2835 pd_idx = sector_div(stripe2, raid_disks);
99c0fb5f
N		 2836 qd_idx = pd_idx + 1;
2837 if (pd_idx == raid_disks-1) {
2838 (*dd_idx)++; /* Q D D D P */
2839 qd_idx = 0;
2840 } else if (*dd_idx >= pd_idx)
2841 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2842 ddf_layout = 1;
99c0fb5f
N		 2843 break;
2844
2845 case ALGORITHM_ROTATING_N_RESTART:
2846 /* Same a left_asymmetric, by first stripe is
2847 * D D D P Q rather than
2848 * Q D D D P
2849 */
6e3b96ed
N		 2850 stripe2 += 1;
2851 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2852 qd_idx = pd_idx + 1;
2853 if (pd_idx == raid_disks-1) {
2854 (*dd_idx)++; /* Q D D D P */
2855 qd_idx = 0;
2856 } else if (*dd_idx >= pd_idx)
2857 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2858 ddf_layout = 1;
99c0fb5f
N		 2859 break;
2860
2861 case ALGORITHM_ROTATING_N_CONTINUE:
2862 /* Same as left_symmetric but Q is before P */
6e3b96ed 2863 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2864 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
2865 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
67cc2b81 2866 ddf_layout = 1;
99c0fb5f
N		 2867 break;
2868
2869 case ALGORITHM_LEFT_ASYMMETRIC_6:
2870 /* RAID5 left_asymmetric, with Q on last device */
6e3b96ed 2871 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2872 if (*dd_idx >= pd_idx)
2873 (*dd_idx)++;
2874 qd_idx = raid_disks - 1;
2875 break;
2876
2877 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6e3b96ed 2878 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2879 if (*dd_idx >= pd_idx)
2880 (*dd_idx)++;
2881 qd_idx = raid_disks - 1;
2882 break;
2883
2884 case ALGORITHM_LEFT_SYMMETRIC_6:
6e3b96ed 2885 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2886 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2887 qd_idx = raid_disks - 1;
2888 break;
2889
2890 case ALGORITHM_RIGHT_SYMMETRIC_6:
6e3b96ed 2891 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2892 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2893 qd_idx = raid_disks - 1;
2894 break;
2895
2896 case ALGORITHM_PARITY_0_6:
2897 pd_idx = 0;
2898 (*dd_idx)++;
2899 qd_idx = raid_disks - 1;
2900 break;
2901
16a53ecc 2902 default:
99c0fb5f 2903 BUG();
16a53ecc
N		 2904 }
2905 break;
1da177e4
LT		 2906 }
2907
911d4ee8
N		 2908 if (sh) {
2909 sh->pd_idx = pd_idx;
2910 sh->qd_idx = qd_idx;
67cc2b81 2911 sh->ddf_layout = ddf_layout;
911d4ee8 2912 }
1da177e4
LT		 2913 /*
2914 * Finally, compute the new sector number
2915 */
2916 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
2917 return new_sector;
2918}
2919
6d036f7d 2920sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous)
1da177e4 2921{
d1688a6d 2922 struct r5conf *conf = sh->raid_conf;
b875e531
N		 2923 int raid_disks = sh->disks;
2924 int data_disks = raid_disks - conf->max_degraded;
1da177e4 2925 sector_t new_sector = sh->sector, check;
09c9e5fa
AN		 2926 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2927 : conf->chunk_sectors;
e183eaed
N		 2928 int algorithm = previous ? conf->prev_algo
2929 : conf->algorithm;
1da177e4
LT		 2930 sector_t stripe;
2931 int chunk_offset;
35f2a591
N		 2932 sector_t chunk_number;
2933 int dummy1, dd_idx = i;
1da177e4 2934 sector_t r_sector;
911d4ee8 2935 struct stripe_head sh2;
1da177e4
LT		 2936
2937 chunk_offset = sector_div(new_sector, sectors_per_chunk);
2938 stripe = new_sector;
1da177e4 2939
16a53ecc
N		 2940 if (i == sh->pd_idx)
2941 return 0;
2942 switch(conf->level) {
2943 case 4: break;
2944 case 5:
e183eaed 2945 switch (algorithm) {
1da177e4
LT		 2946 case ALGORITHM_LEFT_ASYMMETRIC:
2947 case ALGORITHM_RIGHT_ASYMMETRIC:
2948 if (i > sh->pd_idx)
2949 i--;
2950 break;
2951 case ALGORITHM_LEFT_SYMMETRIC:
2952 case ALGORITHM_RIGHT_SYMMETRIC:
2953 if (i < sh->pd_idx)
2954 i += raid_disks;
2955 i -= (sh->pd_idx + 1);
2956 break;
99c0fb5f
N		 2957 case ALGORITHM_PARITY_0:
2958 i -= 1;
2959 break;
2960 case ALGORITHM_PARITY_N:
2961 break;
1da177e4 2962 default:
99c0fb5f 2963 BUG();
16a53ecc
N		 2964 }
2965 break;
2966 case 6:
d0dabf7e 2967 if (i == sh->qd_idx)
16a53ecc 2968 return 0; /* It is the Q disk */
e183eaed 2969 switch (algorithm) {
16a53ecc
N		 2970 case ALGORITHM_LEFT_ASYMMETRIC:
2971 case ALGORITHM_RIGHT_ASYMMETRIC:
99c0fb5f
N		 2972 case ALGORITHM_ROTATING_ZERO_RESTART:
2973 case ALGORITHM_ROTATING_N_RESTART:
2974 if (sh->pd_idx == raid_disks-1)
2975 i--; /* Q D D D P */
16a53ecc
N		 2976 else if (i > sh->pd_idx)
2977 i -= 2; /* D D P Q D */
2978 break;
2979 case ALGORITHM_LEFT_SYMMETRIC:
2980 case ALGORITHM_RIGHT_SYMMETRIC:
2981 if (sh->pd_idx == raid_disks-1)
2982 i--; /* Q D D D P */
2983 else {
2984 /* D D P Q D */
2985 if (i < sh->pd_idx)
2986 i += raid_disks;
2987 i -= (sh->pd_idx + 2);
2988 }
2989 break;
99c0fb5f
N		 2990 case ALGORITHM_PARITY_0:
2991 i -= 2;
2992 break;
2993 case ALGORITHM_PARITY_N:
2994 break;
2995 case ALGORITHM_ROTATING_N_CONTINUE:
e4424fee 2996 /* Like left_symmetric, but P is before Q */
99c0fb5f
N		 2997 if (sh->pd_idx == 0)
2998 i--; /* P D D D Q */
e4424fee
N		 2999 else {
3000 /* D D Q P D */
3001 if (i < sh->pd_idx)
3002 i += raid_disks;
3003 i -= (sh->pd_idx + 1);
3004 }
99c0fb5f
N		 3005 break;
3006 case ALGORITHM_LEFT_ASYMMETRIC_6:
3007 case ALGORITHM_RIGHT_ASYMMETRIC_6:
3008 if (i > sh->pd_idx)
3009 i--;
3010 break;
3011 case ALGORITHM_LEFT_SYMMETRIC_6:
3012 case ALGORITHM_RIGHT_SYMMETRIC_6:
3013 if (i < sh->pd_idx)
3014 i += data_disks + 1;
3015 i -= (sh->pd_idx + 1);
3016 break;
3017 case ALGORITHM_PARITY_0_6:
3018 i -= 1;
3019 break;
16a53ecc 3020 default:
99c0fb5f 3021 BUG();
16a53ecc
N		 3022 }
3023 break;
1da177e4
LT		 3024 }
3025
3026 chunk_number = stripe * data_disks + i;
35f2a591 3027 r_sector = chunk_number * sectors_per_chunk + chunk_offset;
1da177e4 3028
112bf897 3029 check = raid5_compute_sector(conf, r_sector,
784052ec 3030 previous, &dummy1, &sh2);
911d4ee8
N		 3031 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
3032 || sh2.qd_idx != sh->qd_idx) {
cc6167b4
N		 3033 pr_warn("md/raid:%s: compute_blocknr: map not correct\n",
3034 mdname(conf->mddev));
1da177e4
LT		 3035 return 0;
3036 }
3037 return r_sector;
3038}
3039
07e83364
SL		 3040/*
3041 * There are cases where we want handle_stripe_dirtying() and
3042 * schedule_reconstruction() to delay towrite to some dev of a stripe.
3043 *
3044 * This function checks whether we want to delay the towrite. Specifically,
3045 * we delay the towrite when:
3046 *
3047 * 1. degraded stripe has a non-overwrite to the missing dev, AND this
3048 * stripe has data in journal (for other devices).
3049 *
3050 * In this case, when reading data for the non-overwrite dev, it is
3051 * necessary to handle complex rmw of write back cache (prexor with
3052 * orig_page, and xor with page). To keep read path simple, we would
3053 * like to flush data in journal to RAID disks first, so complex rmw
3054 * is handled in the write patch (handle_stripe_dirtying).
3055 *
39b99586
SL		 3056 * 2. when journal space is critical (R5C_LOG_CRITICAL=1)
3057 *
3058 * It is important to be able to flush all stripes in raid5-cache.
3059 * Therefore, we need reserve some space on the journal device for
3060 * these flushes. If flush operation includes pending writes to the
3061 * stripe, we need to reserve (conf->raid_disk + 1) pages per stripe
3062 * for the flush out. If we exclude these pending writes from flush
3063 * operation, we only need (conf->max_degraded + 1) pages per stripe.
3064 * Therefore, excluding pending writes in these cases enables more
3065 * efficient use of the journal device.
3066 *
3067 * Note: To make sure the stripe makes progress, we only delay
3068 * towrite for stripes with data already in journal (injournal > 0).
3069 * When LOG_CRITICAL, stripes with injournal == 0 will be sent to
3070 * no_space_stripes list.
3071 *
70d466f7
SL		 3072 * 3. during journal failure
3073 * In journal failure, we try to flush all cached data to raid disks
3074 * based on data in stripe cache. The array is read-only to upper
3075 * layers, so we would skip all pending writes.
3076 *
07e83364 3077 */
39b99586
SL		 3078static inline bool delay_towrite(struct r5conf *conf,
3079 struct r5dev *dev,
3080 struct stripe_head_state *s)
07e83364 3081{
39b99586
SL		 3082 /* case 1 above */
3083 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
3084 !test_bit(R5_Insync, &dev->flags) && s->injournal)
3085 return true;
3086 /* case 2 above */
3087 if (test_bit(R5C_LOG_CRITICAL, &conf->cache_state) &&
3088 s->injournal > 0)
3089 return true;
70d466f7
SL		 3090 /* case 3 above */
3091 if (s->log_failed && s->injournal)
3092 return true;
39b99586 3093 return false;
07e83364
SL		 3094}
3095
600aa109 3096static void
c0f7bddb 3097schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
600aa109 3098 int rcw, int expand)
e33129d8 3099{
584acdd4 3100 int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx, disks = sh->disks;
d1688a6d 3101 struct r5conf *conf = sh->raid_conf;
c0f7bddb 3102 int level = conf->level;
e33129d8
DW		 3103
3104 if (rcw) {
1e6d690b
SL		 3105 /*
3106 * In some cases, handle_stripe_dirtying initially decided to
3107 * run rmw and allocates extra page for prexor. However, rcw is
3108 * cheaper later on. We need to free the extra page now,
3109 * because we won't be able to do that in ops_complete_prexor().
3110 */
3111 r5c_release_extra_page(sh);
e33129d8
DW		 3112
3113 for (i = disks; i--; ) {
3114 struct r5dev *dev = &sh->dev[i];
3115
39b99586 3116 if (dev->towrite && !delay_towrite(conf, dev, s)) {
e33129d8 3117 set_bit(R5_LOCKED, &dev->flags);
d8ee0728 3118 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 3119 if (!expand)
3120 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 3121 s->locked++;
1e6d690b
SL		 3122 } else if (test_bit(R5_InJournal, &dev->flags)) {
3123 set_bit(R5_LOCKED, &dev->flags);
3124 s->locked++;
e33129d8
DW		 3125 }
3126 }
ce7d363a
N		 3127 /* if we are not expanding this is a proper write request, and
3128 * there will be bios with new data to be drained into the
3129 * stripe cache
3130 */
3131 if (!expand) {
3132 if (!s->locked)
3133 /* False alarm, nothing to do */
3134 return;
3135 sh->reconstruct_state = reconstruct_state_drain_run;
3136 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
3137 } else
3138 sh->reconstruct_state = reconstruct_state_run;
3139
3140 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
3141
c0f7bddb 3142 if (s->locked + conf->max_degraded == disks)
8b3e6cdc 3143 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
c0f7bddb 3144 atomic_inc(&conf->pending_full_writes);
e33129d8
DW		 3145 } else {
3146 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
3147 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
584acdd4
MS		 3148 BUG_ON(level == 6 &&
3149 (!(test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags) ||
3150 test_bit(R5_Wantcompute, &sh->dev[qd_idx].flags))));
e33129d8 3151
e33129d8
DW		 3152 for (i = disks; i--; ) {
3153 struct r5dev *dev = &sh->dev[i];
584acdd4 3154 if (i == pd_idx || i == qd_idx)
e33129d8
DW		 3155 continue;
3156
e33129d8
DW		 3157 if (dev->towrite &&
3158 (test_bit(R5_UPTODATE, &dev->flags) ||
d8ee0728
DW		 3159 test_bit(R5_Wantcompute, &dev->flags))) {
3160 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 3161 set_bit(R5_LOCKED, &dev->flags);
3162 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 3163 s->locked++;
1e6d690b
SL		 3164 } else if (test_bit(R5_InJournal, &dev->flags)) {
3165 set_bit(R5_LOCKED, &dev->flags);
3166 s->locked++;
e33129d8
DW		 3167 }
3168 }
ce7d363a
N		 3169 if (!s->locked)
3170 /* False alarm - nothing to do */
3171 return;
3172 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
3173 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
3174 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
3175 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
e33129d8
DW		 3176 }
3177
c0f7bddb 3178 /* keep the parity disk(s) locked while asynchronous operations
e33129d8
DW		 3179 * are in flight
3180 */
3181 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
3182 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
600aa109 3183 s->locked++;
e33129d8 3184
c0f7bddb
YT		 3185 if (level == 6) {
3186 int qd_idx = sh->qd_idx;
3187 struct r5dev *dev = &sh->dev[qd_idx];
3188
3189 set_bit(R5_LOCKED, &dev->flags);
3190 clear_bit(R5_UPTODATE, &dev->flags);
3191 s->locked++;
3192 }
3193
845b9e22 3194 if (raid5_has_ppl(sh->raid_conf) && sh->ppl_page &&
3418d036
AP		 3195 test_bit(STRIPE_OP_BIODRAIN, &s->ops_request) &&
3196 !test_bit(STRIPE_FULL_WRITE, &sh->state) &&
3197 test_bit(R5_Insync, &sh->dev[pd_idx].flags))
3198 set_bit(STRIPE_OP_PARTIAL_PARITY, &s->ops_request);
3199
600aa109 3200 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
e46b272b 3201 __func__, (unsigned long long)sh->sector,
600aa109 3202 s->locked, s->ops_request);
e33129d8 3203}
16a53ecc 3204
1da177e4
LT		 3205/*
3206 * Each stripe/dev can have one or more bion attached.
16a53ecc 3207 * toread/towrite point to the first in a chain.
1da177e4
LT		 3208 * The bi_next chain must be in order.
3209 */
da41ba65 3210static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx,
3211 int forwrite, int previous)
1da177e4
LT		 3212{
3213 struct bio **bip;
d1688a6d 3214 struct r5conf *conf = sh->raid_conf;
72626685 3215 int firstwrite=0;
1da177e4 3216
cbe47ec5 3217 pr_debug("adding bi b#%llu to stripe s#%llu\n",
4f024f37 3218 (unsigned long long)bi->bi_iter.bi_sector,
1da177e4
LT		 3219 (unsigned long long)sh->sector);
3220
b17459c0 3221 spin_lock_irq(&sh->stripe_lock);
2cd259a7 3222 sh->dev[dd_idx].write_hint = bi->bi_write_hint;
59fc630b 3223 /* Don't allow new IO added to stripes in batch list */
3224 if (sh->batch_head)
3225 goto overlap;
72626685 3226 if (forwrite) {
1da177e4 3227 bip = &sh->dev[dd_idx].towrite;
7eaf7e8e 3228 if (*bip == NULL)
72626685
N		 3229 firstwrite = 1;
3230 } else
1da177e4 3231 bip = &sh->dev[dd_idx].toread;
4f024f37
KO		 3232 while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) {
3233 if (bio_end_sector(*bip) > bi->bi_iter.bi_sector)
1da177e4
LT		 3234 goto overlap;
3235 bip = & (*bip)->bi_next;
3236 }
4f024f37 3237 if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi))
1da177e4
LT		 3238 goto overlap;
3239
3418d036
AP		 3240 if (forwrite && raid5_has_ppl(conf)) {
3241 /*
3242 * With PPL only writes to consecutive data chunks within a
3243 * stripe are allowed because for a single stripe_head we can
3244 * only have one PPL entry at a time, which describes one data
3245 * range. Not really an overlap, but wait_for_overlap can be
3246 * used to handle this.
3247 */
3248 sector_t sector;
3249 sector_t first = 0;
3250 sector_t last = 0;
3251 int count = 0;
3252 int i;
3253
3254 for (i = 0; i < sh->disks; i++) {
3255 if (i != sh->pd_idx &&
3256 (i == dd_idx || sh->dev[i].towrite)) {
3257 sector = sh->dev[i].sector;
3258 if (count == 0 || sector < first)
3259 first = sector;
3260 if (sector > last)
3261 last = sector;
3262 count++;
3263 }
3264 }
3265
3266 if (first + conf->chunk_sectors * (count - 1) != last)
3267 goto overlap;
3268 }
3269
da41ba65 3270 if (!forwrite || previous)
3271 clear_bit(STRIPE_BATCH_READY, &sh->state);
3272
78bafebd 3273 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
1da177e4
LT		 3274 if (*bip)
3275 bi->bi_next = *bip;
3276 *bip = bi;
016c76ac 3277 bio_inc_remaining(bi);
49728050 3278 md_write_inc(conf->mddev, bi);
72626685 3279
1da177e4
LT		 3280 if (forwrite) {
3281 /* check if page is covered */
3282 sector_t sector = sh->dev[dd_idx].sector;
3283 for (bi=sh->dev[dd_idx].towrite;
3284 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
4f024f37 3285 bi && bi->bi_iter.bi_sector <= sector;
1da177e4 3286 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
f73a1c7d
KO		 3287 if (bio_end_sector(bi) >= sector)
3288 sector = bio_end_sector(bi);
1da177e4
LT		 3289 }
3290 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
7a87f434 3291 if (!test_and_set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags))
3292 sh->overwrite_disks++;
1da177e4 3293 }
cbe47ec5
N		 3294
3295 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
4f024f37 3296 (unsigned long long)(*bip)->bi_iter.bi_sector,
cbe47ec5
N		 3297 (unsigned long long)sh->sector, dd_idx);
3298
3299 if (conf->mddev->bitmap && firstwrite) {
d0852df5
N		 3300 /* Cannot hold spinlock over bitmap_startwrite,
3301 * but must ensure this isn't added to a batch until
3302 * we have added to the bitmap and set bm_seq.
3303 * So set STRIPE_BITMAP_PENDING to prevent
3304 * batching.
3305 * If multiple add_stripe_bio() calls race here they
3306 * much all set STRIPE_BITMAP_PENDING. So only the first one
3307 * to complete "bitmap_startwrite" gets to set
3308 * STRIPE_BIT_DELAY. This is important as once a stripe
3309 * is added to a batch, STRIPE_BIT_DELAY cannot be changed
3310 * any more.
3311 */
3312 set_bit(STRIPE_BITMAP_PENDING, &sh->state);
3313 spin_unlock_irq(&sh->stripe_lock);
e64e4018
AS		 3314 md_bitmap_startwrite(conf->mddev->bitmap, sh->sector,
3315 STRIPE_SECTORS, 0);
d0852df5
N		 3316 spin_lock_irq(&sh->stripe_lock);
3317 clear_bit(STRIPE_BITMAP_PENDING, &sh->state);
3318 if (!sh->batch_head) {
3319 sh->bm_seq = conf->seq_flush+1;
3320 set_bit(STRIPE_BIT_DELAY, &sh->state);
3321 }
cbe47ec5 3322 }
d0852df5 3323 spin_unlock_irq(&sh->stripe_lock);
59fc630b 3324
3325 if (stripe_can_batch(sh))
3326 stripe_add_to_batch_list(conf, sh);
1da177e4
LT		 3327 return 1;
3328
3329 overlap:
3330 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
b17459c0 3331 spin_unlock_irq(&sh->stripe_lock);
1da177e4
LT		 3332 return 0;
3333}
3334
d1688a6d 3335static void end_reshape(struct r5conf *conf);
29269553 3336
d1688a6d 3337static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
911d4ee8 3338 struct stripe_head *sh)
ccfcc3c1 3339{
784052ec 3340 int sectors_per_chunk =
09c9e5fa 3341 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
911d4ee8 3342 int dd_idx;
2d2063ce 3343 int chunk_offset = sector_div(stripe, sectors_per_chunk);
112bf897 3344 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
2d2063ce 3345
112bf897
N		 3346 raid5_compute_sector(conf,
3347 stripe * (disks - conf->max_degraded)
b875e531 3348 *sectors_per_chunk + chunk_offset,
112bf897 3349 previous,
911d4ee8 3350 &dd_idx, sh);
ccfcc3c1
N		 3351}
3352
a4456856 3353static void
d1688a6d 3354handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh,
bd83d0a2 3355 struct stripe_head_state *s, int disks)
a4456856
DW		 3356{
3357 int i;
59fc630b 3358 BUG_ON(sh->batch_head);
a4456856
DW		 3359 for (i = disks; i--; ) {
3360 struct bio *bi;
3361 int bitmap_end = 0;
3362
3363 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
3cb03002 3364 struct md_rdev *rdev;
a4456856
DW		 3365 rcu_read_lock();
3366 rdev = rcu_dereference(conf->disks[i].rdev);
f5b67ae8
N		 3367 if (rdev && test_bit(In_sync, &rdev->flags) &&
3368 !test_bit(Faulty, &rdev->flags))
7f0da59b
N		 3369 atomic_inc(&rdev->nr_pending);
3370 else
3371 rdev = NULL;
a4456856 3372 rcu_read_unlock();
7f0da59b
N		 3373 if (rdev) {
3374 if (!rdev_set_badblocks(
3375 rdev,
3376 sh->sector,
3377 STRIPE_SECTORS, 0))
3378 md_error(conf->mddev, rdev);
3379 rdev_dec_pending(rdev, conf->mddev);
3380 }
a4456856 3381 }
b17459c0 3382 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 3383 /* fail all writes first */
3384 bi = sh->dev[i].towrite;
3385 sh->dev[i].towrite = NULL;
7a87f434 3386 sh->overwrite_disks = 0;
b17459c0 3387 spin_unlock_irq(&sh->stripe_lock);
1ed850f3 3388 if (bi)
a4456856 3389 bitmap_end = 1;
a4456856 3390
ff875738 3391 log_stripe_write_finished(sh);
0576b1c6 3392
a4456856
DW		 3393 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
3394 wake_up(&conf->wait_for_overlap);
3395
4f024f37 3396 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3397 sh->dev[i].sector + STRIPE_SECTORS) {
3398 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
4246a0b6 3399
49728050 3400 md_write_end(conf->mddev);
6308d8e3 3401 bio_io_error(bi);
a4456856
DW		 3402 bi = nextbi;
3403 }
7eaf7e8e 3404 if (bitmap_end)
e64e4018
AS		 3405 md_bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3406 STRIPE_SECTORS, 0, 0);
7eaf7e8e 3407 bitmap_end = 0;
a4456856
DW		 3408 /* and fail all 'written' */
3409 bi = sh->dev[i].written;
3410 sh->dev[i].written = NULL;
d592a996
SL		 3411 if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) {
3412 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
3413 sh->dev[i].page = sh->dev[i].orig_page;
3414 }
3415
a4456856 3416 if (bi) bitmap_end = 1;
4f024f37 3417 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3418 sh->dev[i].sector + STRIPE_SECTORS) {
3419 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
4246a0b6 3420
49728050 3421 md_write_end(conf->mddev);
6308d8e3 3422 bio_io_error(bi);
a4456856
DW		 3423 bi = bi2;
3424 }
3425
b5e98d65
DW		 3426 /* fail any reads if this device is non-operational and
3427 * the data has not reached the cache yet.
3428 */
3429 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
6e74a9cf 3430 s->failed > conf->max_degraded &&
b5e98d65
DW		 3431 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
3432 test_bit(R5_ReadError, &sh->dev[i].flags))) {
143c4d05 3433 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 3434 bi = sh->dev[i].toread;
3435 sh->dev[i].toread = NULL;
143c4d05 3436 spin_unlock_irq(&sh->stripe_lock);
a4456856
DW		 3437 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
3438 wake_up(&conf->wait_for_overlap);
ebda780b
SL		 3439 if (bi)
3440 s->to_read--;
4f024f37 3441 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3442 sh->dev[i].sector + STRIPE_SECTORS) {
3443 struct bio *nextbi =
3444 r5_next_bio(bi, sh->dev[i].sector);
4246a0b6 3445
6308d8e3 3446 bio_io_error(bi);
a4456856
DW		 3447 bi = nextbi;
3448 }
3449 }
a4456856 3450 if (bitmap_end)
e64e4018
AS		 3451 md_bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3452 STRIPE_SECTORS, 0, 0);
8cfa7b0f
N		 3453 /* If we were in the middle of a write the parity block might
3454 * still be locked - so just clear all R5_LOCKED flags
3455 */
3456 clear_bit(R5_LOCKED, &sh->dev[i].flags);
a4456856 3457 }
ebda780b
SL		 3458 s->to_write = 0;
3459 s->written = 0;
a4456856 3460
8b3e6cdc
DW		 3461 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3462 if (atomic_dec_and_test(&conf->pending_full_writes))
3463 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 3464}
3465
7f0da59b 3466static void
d1688a6d 3467handle_failed_sync(struct r5conf *conf, struct stripe_head *sh,
7f0da59b
N		 3468 struct stripe_head_state *s)
3469{
3470 int abort = 0;
3471 int i;
3472
59fc630b 3473 BUG_ON(sh->batch_head);
7f0da59b 3474 clear_bit(STRIPE_SYNCING, &sh->state);
f8dfcffd
N		 3475 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
3476 wake_up(&conf->wait_for_overlap);
7f0da59b 3477 s->syncing = 0;
9a3e1101 3478 s->replacing = 0;
7f0da59b 3479 /* There is nothing more to do for sync/check/repair.
18b9837e
N		 3480 * Don't even need to abort as that is handled elsewhere
3481 * if needed, and not always wanted e.g. if there is a known
3482 * bad block here.
9a3e1101 3483 * For recover/replace we need to record a bad block on all
7f0da59b
N		 3484 * non-sync devices, or abort the recovery
3485 */
18b9837e
N		 3486 if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) {
3487 /* During recovery devices cannot be removed, so
3488 * locking and refcounting of rdevs is not needed
3489 */
e50d3992 3490 rcu_read_lock();
18b9837e 3491 for (i = 0; i < conf->raid_disks; i++) {
e50d3992 3492 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
18b9837e
N		 3493 if (rdev
3494 && !test_bit(Faulty, &rdev->flags)
3495 && !test_bit(In_sync, &rdev->flags)
3496 && !rdev_set_badblocks(rdev, sh->sector,
3497 STRIPE_SECTORS, 0))
3498 abort = 1;
e50d3992 3499 rdev = rcu_dereference(conf->disks[i].replacement);
18b9837e
N		 3500 if (rdev
3501 && !test_bit(Faulty, &rdev->flags)
3502 && !test_bit(In_sync, &rdev->flags)
3503 && !rdev_set_badblocks(rdev, sh->sector,
3504 STRIPE_SECTORS, 0))
3505 abort = 1;
3506 }
e50d3992 3507 rcu_read_unlock();
18b9837e
N		 3508 if (abort)
3509 conf->recovery_disabled =
3510 conf->mddev->recovery_disabled;
7f0da59b 3511 }
18b9837e 3512 md_done_sync(conf->mddev, STRIPE_SECTORS, !abort);
7f0da59b
N		 3513}
3514
9a3e1101
N		 3515static int want_replace(struct stripe_head *sh, int disk_idx)
3516{
3517 struct md_rdev *rdev;
3518 int rv = 0;
3f232d6a
N		 3519
3520 rcu_read_lock();
3521 rdev = rcu_dereference(sh->raid_conf->disks[disk_idx].replacement);
9a3e1101
N		 3522 if (rdev
3523 && !test_bit(Faulty, &rdev->flags)
3524 && !test_bit(In_sync, &rdev->flags)
3525 && (rdev->recovery_offset <= sh->sector
3526 || rdev->mddev->recovery_cp <= sh->sector))
3527 rv = 1;
3f232d6a 3528 rcu_read_unlock();
9a3e1101
N		 3529 return rv;
3530}
3531
2c58f06e
N		 3532static int need_this_block(struct stripe_head *sh, struct stripe_head_state *s,
3533 int disk_idx, int disks)
a4456856 3534{
5599becc 3535 struct r5dev *dev = &sh->dev[disk_idx];
f2b3b44d
N		 3536 struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]],
3537 &sh->dev[s->failed_num[1]] };
ea664c82 3538 int i;
5599becc 3539
a79cfe12
N		 3540
3541 if (test_bit(R5_LOCKED, &dev->flags) ||
3542 test_bit(R5_UPTODATE, &dev->flags))
3543 /* No point reading this as we already have it or have
3544 * decided to get it.
3545 */
3546 return 0;
3547
3548 if (dev->toread ||
3549 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)))
3550 /* We need this block to directly satisfy a request */
3551 return 1;
3552
3553 if (s->syncing || s->expanding ||
3554 (s->replacing && want_replace(sh, disk_idx)))
3555 /* When syncing, or expanding we read everything.
3556 * When replacing, we need the replaced block.
3557 */
3558 return 1;
3559
3560 if ((s->failed >= 1 && fdev[0]->toread) ||
3561 (s->failed >= 2 && fdev[1]->toread))
3562 /* If we want to read from a failed device, then
3563 * we need to actually read every other device.
3564 */
3565 return 1;
3566
a9d56950
N		 3567 /* Sometimes neither read-modify-write nor reconstruct-write
3568 * cycles can work. In those cases we read every block we
3569 * can. Then the parity-update is certain to have enough to
3570 * work with.
3571 * This can only be a problem when we need to write something,
3572 * and some device has failed. If either of those tests
3573 * fail we need look no further.
3574 */
3575 if (!s->failed || !s->to_write)
3576 return 0;
3577
3578 if (test_bit(R5_Insync, &dev->flags) &&
3579 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3580 /* Pre-reads at not permitted until after short delay
3581 * to gather multiple requests. However if this
3560741e 3582 * device is no Insync, the block could only be computed
a9d56950
N		 3583 * and there is no need to delay that.
3584 */
3585 return 0;
ea664c82 3586
36707bb2 3587 for (i = 0; i < s->failed && i < 2; i++) {
ea664c82
N		 3588 if (fdev[i]->towrite &&
3589 !test_bit(R5_UPTODATE, &fdev[i]->flags) &&
3590 !test_bit(R5_OVERWRITE, &fdev[i]->flags))
3591 /* If we have a partial write to a failed
3592 * device, then we will need to reconstruct
3593 * the content of that device, so all other
3594 * devices must be read.
3595 */
3596 return 1;
3597 }
3598
3599 /* If we are forced to do a reconstruct-write, either because
3600 * the current RAID6 implementation only supports that, or
3560741e 3601 * because parity cannot be trusted and we are currently
ea664c82
N		 3602 * recovering it, there is extra need to be careful.
3603 * If one of the devices that we would need to read, because
3604 * it is not being overwritten (and maybe not written at all)
3605 * is missing/faulty, then we need to read everything we can.
3606 */
3607 if (sh->raid_conf->level != 6 &&
3608 sh->sector < sh->raid_conf->mddev->recovery_cp)
3609 /* reconstruct-write isn't being forced */
3610 return 0;
36707bb2 3611 for (i = 0; i < s->failed && i < 2; i++) {
10d82c5f
N		 3612 if (s->failed_num[i] != sh->pd_idx &&
3613 s->failed_num[i] != sh->qd_idx &&
3614 !test_bit(R5_UPTODATE, &fdev[i]->flags) &&
ea664c82
N		 3615 !test_bit(R5_OVERWRITE, &fdev[i]->flags))
3616 return 1;
3617 }
3618
2c58f06e
N		 3619 return 0;
3620}
3621
ba02684d
SL		 3622/* fetch_block - checks the given member device to see if its data needs
3623 * to be read or computed to satisfy a request.
3624 *
3625 * Returns 1 when no more member devices need to be checked, otherwise returns
3626 * 0 to tell the loop in handle_stripe_fill to continue
3627 */
2c58f06e
N		 3628static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s,
3629 int disk_idx, int disks)
3630{
3631 struct r5dev *dev = &sh->dev[disk_idx];
3632
3633 /* is the data in this block needed, and can we get it? */
3634 if (need_this_block(sh, s, disk_idx, disks)) {
5599becc
YT		 3635 /* we would like to get this block, possibly by computing it,
3636 * otherwise read it if the backing disk is insync
3637 */
3638 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
3639 BUG_ON(test_bit(R5_Wantread, &dev->flags));
b0c783b3 3640 BUG_ON(sh->batch_head);
7471fb77
N		 3641
3642 /*
3643 * In the raid6 case if the only non-uptodate disk is P
3644 * then we already trusted P to compute the other failed
3645 * drives. It is safe to compute rather than re-read P.
3646 * In other cases we only compute blocks from failed
3647 * devices, otherwise check/repair might fail to detect
3648 * a real inconsistency.
3649 */
3650
5599becc 3651 if ((s->uptodate == disks - 1) &&
7471fb77 3652 ((sh->qd_idx >= 0 && sh->pd_idx == disk_idx) ||
f2b3b44d 3653 (s->failed && (disk_idx == s->failed_num[0] ||
7471fb77 3654 disk_idx == s->failed_num[1])))) {
5599becc
YT		 3655 /* have disk failed, and we're requested to fetch it;
3656 * do compute it
a4456856 3657 */
5599becc
YT		 3658 pr_debug("Computing stripe %llu block %d\n",
3659 (unsigned long long)sh->sector, disk_idx);
3660 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3661 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3662 set_bit(R5_Wantcompute, &dev->flags);
3663 sh->ops.target = disk_idx;
3664 sh->ops.target2 = -1; /* no 2nd target */
3665 s->req_compute = 1;
93b3dbce
N		 3666 /* Careful: from this point on 'uptodate' is in the eye
3667 * of raid_run_ops which services 'compute' operations
3668 * before writes. R5_Wantcompute flags a block that will
3669 * be R5_UPTODATE by the time it is needed for a
3670 * subsequent operation.
3671 */
5599becc
YT		 3672 s->uptodate++;
3673 return 1;
3674 } else if (s->uptodate == disks-2 && s->failed >= 2) {
3675 /* Computing 2-failure is *very* expensive; only
3676 * do it if failed >= 2
3677 */
3678 int other;
3679 for (other = disks; other--; ) {
3680 if (other == disk_idx)
3681 continue;
3682 if (!test_bit(R5_UPTODATE,
3683 &sh->dev[other].flags))
3684 break;
a4456856 3685 }
5599becc
YT		 3686 BUG_ON(other < 0);
3687 pr_debug("Computing stripe %llu blocks %d,%d\n",
3688 (unsigned long long)sh->sector,
3689 disk_idx, other);
3690 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3691 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3692 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
3693 set_bit(R5_Wantcompute, &sh->dev[other].flags);
3694 sh->ops.target = disk_idx;
3695 sh->ops.target2 = other;
3696 s->uptodate += 2;
3697 s->req_compute = 1;
3698 return 1;
3699 } else if (test_bit(R5_Insync, &dev->flags)) {
3700 set_bit(R5_LOCKED, &dev->flags);
3701 set_bit(R5_Wantread, &dev->flags);
3702 s->locked++;
3703 pr_debug("Reading block %d (sync=%d)\n",
3704 disk_idx, s->syncing);
a4456856
DW		 3705 }
3706 }
5599becc
YT		 3707
3708 return 0;
3709}
3710
3711/**
93b3dbce 3712 * handle_stripe_fill - read or compute data to satisfy pending requests.
5599becc 3713 */
93b3dbce
N		 3714static void handle_stripe_fill(struct stripe_head *sh,
3715 struct stripe_head_state *s,
3716 int disks)
5599becc
YT		 3717{
3718 int i;
3719
3720 /* look for blocks to read/compute, skip this if a compute
3721 * is already in flight, or if the stripe contents are in the
3722 * midst of changing due to a write
3723 */
3724 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
07e83364
SL		 3725 !sh->reconstruct_state) {
3726
3727 /*
3728 * For degraded stripe with data in journal, do not handle
3729 * read requests yet, instead, flush the stripe to raid
3730 * disks first, this avoids handling complex rmw of write
3731 * back cache (prexor with orig_page, and then xor with
3732 * page) in the read path
3733 */
3734 if (s->injournal && s->failed) {
3735 if (test_bit(STRIPE_R5C_CACHING, &sh->state))
3736 r5c_make_stripe_write_out(sh);
3737 goto out;
3738 }
3739
5599becc 3740 for (i = disks; i--; )
93b3dbce 3741 if (fetch_block(sh, s, i, disks))
5599becc 3742 break;
07e83364
SL		 3743 }
3744out:
a4456856
DW		 3745 set_bit(STRIPE_HANDLE, &sh->state);
3746}
3747
787b76fa
N		 3748static void break_stripe_batch_list(struct stripe_head *head_sh,
3749 unsigned long handle_flags);
1fe797e6 3750/* handle_stripe_clean_event
a4456856
DW		 3751 * any written block on an uptodate or failed drive can be returned.
3752 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
3753 * never LOCKED, so we don't need to test 'failed' directly.
3754 */
d1688a6d 3755static void handle_stripe_clean_event(struct r5conf *conf,
bd83d0a2 3756 struct stripe_head *sh, int disks)
a4456856
DW		 3757{
3758 int i;
3759 struct r5dev *dev;
f8dfcffd 3760 int discard_pending = 0;
59fc630b 3761 struct stripe_head *head_sh = sh;
3762 bool do_endio = false;
a4456856
DW		 3763
3764 for (i = disks; i--; )
3765 if (sh->dev[i].written) {
3766 dev = &sh->dev[i];
3767 if (!test_bit(R5_LOCKED, &dev->flags) &&
9e444768 3768 (test_bit(R5_UPTODATE, &dev->flags) ||
d592a996
SL		 3769 test_bit(R5_Discard, &dev->flags) ||
3770 test_bit(R5_SkipCopy, &dev->flags))) {
a4456856
DW		 3771 /* We can return any write requests */
3772 struct bio *wbi, *wbi2;
45b4233c 3773 pr_debug("Return write for disc %d\n", i);
ca64cae9
N		 3774 if (test_and_clear_bit(R5_Discard, &dev->flags))
3775 clear_bit(R5_UPTODATE, &dev->flags);
d592a996
SL		 3776 if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) {
3777 WARN_ON(test_bit(R5_UPTODATE, &dev->flags));
d592a996 3778 }
59fc630b 3779 do_endio = true;
3780
3781returnbi:
3782 dev->page = dev->orig_page;
a4456856
DW		 3783 wbi = dev->written;
3784 dev->written = NULL;
4f024f37 3785 while (wbi && wbi->bi_iter.bi_sector <
a4456856
DW		 3786 dev->sector + STRIPE_SECTORS) {
3787 wbi2 = r5_next_bio(wbi, dev->sector);
49728050 3788 md_write_end(conf->mddev);
016c76ac 3789 bio_endio(wbi);
a4456856
DW		 3790 wbi = wbi2;
3791 }
e64e4018
AS		 3792 md_bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3793 STRIPE_SECTORS,
3794 !test_bit(STRIPE_DEGRADED, &sh->state),
3795 0);
59fc630b 3796 if (head_sh->batch_head) {
3797 sh = list_first_entry(&sh->batch_list,
3798 struct stripe_head,
3799 batch_list);
3800 if (sh != head_sh) {
3801 dev = &sh->dev[i];
3802 goto returnbi;
3803 }
3804 }
3805 sh = head_sh;
3806 dev = &sh->dev[i];
f8dfcffd
N		 3807 } else if (test_bit(R5_Discard, &dev->flags))
3808 discard_pending = 1;
3809 }
f6bed0ef 3810
ff875738 3811 log_stripe_write_finished(sh);
0576b1c6 3812
f8dfcffd
N		 3813 if (!discard_pending &&
3814 test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) {
b8a9d66d 3815 int hash;
f8dfcffd
N		 3816 clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
3817 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
3818 if (sh->qd_idx >= 0) {
3819 clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
3820 clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags);
3821 }
3822 /* now that discard is done we can proceed with any sync */
3823 clear_bit(STRIPE_DISCARD, &sh->state);
d47648fc
SL		 3824 /*
3825 * SCSI discard will change some bio fields and the stripe has
3826 * no updated data, so remove it from hash list and the stripe
3827 * will be reinitialized
3828 */
59fc630b 3829unhash:
b8a9d66d
RG		 3830 hash = sh->hash_lock_index;
3831 spin_lock_irq(conf->hash_locks + hash);
d47648fc 3832 remove_hash(sh);
b8a9d66d 3833 spin_unlock_irq(conf->hash_locks + hash);
59fc630b 3834 if (head_sh->batch_head) {
3835 sh = list_first_entry(&sh->batch_list,
3836 struct stripe_head, batch_list);
3837 if (sh != head_sh)
3838 goto unhash;
3839 }
59fc630b 3840 sh = head_sh;
3841
f8dfcffd
N		 3842 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state))
3843 set_bit(STRIPE_HANDLE, &sh->state);
3844
3845 }
8b3e6cdc
DW		 3846
3847 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3848 if (atomic_dec_and_test(&conf->pending_full_writes))
3849 md_wakeup_thread(conf->mddev->thread);
59fc630b 3850
787b76fa
N		 3851 if (head_sh->batch_head && do_endio)
3852 break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS);
a4456856
DW		 3853}
3854
86aa1397
SL		 3855/*
3856 * For RMW in write back cache, we need extra page in prexor to store the
3857 * old data. This page is stored in dev->orig_page.
3858 *
3859 * This function checks whether we have data for prexor. The exact logic
3860 * is:
3861 * R5_UPTODATE && (!R5_InJournal || R5_OrigPageUPTDODATE)
3862 */
3863static inline bool uptodate_for_rmw(struct r5dev *dev)
3864{
3865 return (test_bit(R5_UPTODATE, &dev->flags)) &&
3866 (!test_bit(R5_InJournal, &dev->flags) ||
3867 test_bit(R5_OrigPageUPTDODATE, &dev->flags));
3868}
3869
d7bd398e
SL		 3870static int handle_stripe_dirtying(struct r5conf *conf,
3871 struct stripe_head *sh,
3872 struct stripe_head_state *s,
3873 int disks)
a4456856
DW		 3874{
3875 int rmw = 0, rcw = 0, i;
a7854487
AL		 3876 sector_t recovery_cp = conf->mddev->recovery_cp;
3877
584acdd4 3878 /* Check whether resync is now happening or should start.
a7854487
AL		 3879 * If yes, then the array is dirty (after unclean shutdown or
3880 * initial creation), so parity in some stripes might be inconsistent.
3881 * In this case, we need to always do reconstruct-write, to ensure
3882 * that in case of drive failure or read-error correction, we
3883 * generate correct data from the parity.
3884 */
584acdd4 3885 if (conf->rmw_level == PARITY_DISABLE_RMW ||
26ac1073
N		 3886 (recovery_cp < MaxSector && sh->sector >= recovery_cp &&
3887 s->failed == 0)) {
a7854487 3888 /* Calculate the real rcw later - for now make it
c8ac1803
N		 3889 * look like rcw is cheaper
3890 */
3891 rcw = 1; rmw = 2;
584acdd4
MS		 3892 pr_debug("force RCW rmw_level=%u, recovery_cp=%llu sh->sector=%llu\n",
3893 conf->rmw_level, (unsigned long long)recovery_cp,
a7854487 3894 (unsigned long long)sh->sector);
c8ac1803 3895 } else for (i = disks; i--; ) {
a4456856
DW		 3896 /* would I have to read this buffer for read_modify_write */
3897 struct r5dev *dev = &sh->dev[i];
39b99586 3898 if (((dev->towrite && !delay_towrite(conf, dev, s)) ||
07e83364 3899 i == sh->pd_idx || i == sh->qd_idx ||
1e6d690b 3900 test_bit(R5_InJournal, &dev->flags)) &&
a4456856 3901 !test_bit(R5_LOCKED, &dev->flags) &&
86aa1397 3902 !(uptodate_for_rmw(dev) ||
f38e1219 3903 test_bit(R5_Wantcompute, &dev->flags))) {
a4456856
DW		 3904 if (test_bit(R5_Insync, &dev->flags))
3905 rmw++;
3906 else
3907 rmw += 2*disks; /* cannot read it */
3908 }
3909 /* Would I have to read this buffer for reconstruct_write */
584acdd4
MS		 3910 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
3911 i != sh->pd_idx && i != sh->qd_idx &&
a4456856 3912 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219 3913 !(test_bit(R5_UPTODATE, &dev->flags) ||
1e6d690b 3914 test_bit(R5_Wantcompute, &dev->flags))) {
67f45548
N		 3915 if (test_bit(R5_Insync, &dev->flags))
3916 rcw++;
a4456856
DW		 3917 else
3918 rcw += 2*disks;
3919 }
3920 }
1e6d690b 3921
39b99586
SL		 3922 pr_debug("for sector %llu state 0x%lx, rmw=%d rcw=%d\n",
3923 (unsigned long long)sh->sector, sh->state, rmw, rcw);
a4456856 3924 set_bit(STRIPE_HANDLE, &sh->state);
41257580 3925 if ((rmw < rcw || (rmw == rcw && conf->rmw_level == PARITY_PREFER_RMW)) && rmw > 0) {
a4456856 3926 /* prefer read-modify-write, but need to get some data */
e3620a3a
JB		 3927 if (conf->mddev->queue)
3928 blk_add_trace_msg(conf->mddev->queue,
3929 "raid5 rmw %llu %d",
3930 (unsigned long long)sh->sector, rmw);
a4456856
DW		 3931 for (i = disks; i--; ) {
3932 struct r5dev *dev = &sh->dev[i];
1e6d690b
SL		 3933 if (test_bit(R5_InJournal, &dev->flags) &&
3934 dev->page == dev->orig_page &&
3935 !test_bit(R5_LOCKED, &sh->dev[sh->pd_idx].flags)) {
3936 /* alloc page for prexor */
d7bd398e
SL		 3937 struct page *p = alloc_page(GFP_NOIO);
3938
3939 if (p) {
3940 dev->orig_page = p;
3941 continue;
3942 }
3943
3944 /*
3945 * alloc_page() failed, try use
3946 * disk_info->extra_page
3947 */
3948 if (!test_and_set_bit(R5C_EXTRA_PAGE_IN_USE,
3949 &conf->cache_state)) {
3950 r5c_use_extra_page(sh);
3951 break;
3952 }
1e6d690b 3953
d7bd398e
SL		 3954 /* extra_page in use, add to delayed_list */
3955 set_bit(STRIPE_DELAYED, &sh->state);
3956 s->waiting_extra_page = 1;
3957 return -EAGAIN;
1e6d690b 3958 }
d7bd398e 3959 }
1e6d690b 3960
d7bd398e
SL		 3961 for (i = disks; i--; ) {
3962 struct r5dev *dev = &sh->dev[i];
39b99586 3963 if (((dev->towrite && !delay_towrite(conf, dev, s)) ||
1e6d690b
SL		 3964 i == sh->pd_idx || i == sh->qd_idx ||
3965 test_bit(R5_InJournal, &dev->flags)) &&
a4456856 3966 !test_bit(R5_LOCKED, &dev->flags) &&
86aa1397 3967 !(uptodate_for_rmw(dev) ||
1e6d690b 3968 test_bit(R5_Wantcompute, &dev->flags)) &&
a4456856 3969 test_bit(R5_Insync, &dev->flags)) {
67f45548
N		 3970 if (test_bit(STRIPE_PREREAD_ACTIVE,
3971 &sh->state)) {
3972 pr_debug("Read_old block %d for r-m-w\n",
3973 i);
a4456856
DW		 3974 set_bit(R5_LOCKED, &dev->flags);
3975 set_bit(R5_Wantread, &dev->flags);
3976 s->locked++;
3977 } else {
3978 set_bit(STRIPE_DELAYED, &sh->state);
3979 set_bit(STRIPE_HANDLE, &sh->state);
3980 }
3981 }
3982 }
a9add5d9 3983 }
41257580 3984 if ((rcw < rmw || (rcw == rmw && conf->rmw_level != PARITY_PREFER_RMW)) && rcw > 0) {
a4456856 3985 /* want reconstruct write, but need to get some data */
a9add5d9 3986 int qread =0;
c8ac1803 3987 rcw = 0;
a4456856
DW		 3988 for (i = disks; i--; ) {
3989 struct r5dev *dev = &sh->dev[i];
3990 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
c8ac1803 3991 i != sh->pd_idx && i != sh->qd_idx &&
a4456856 3992 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219 3993 !(test_bit(R5_UPTODATE, &dev->flags) ||
c8ac1803
N		 3994 test_bit(R5_Wantcompute, &dev->flags))) {
3995 rcw++;
67f45548
N		 3996 if (test_bit(R5_Insync, &dev->flags) &&
3997 test_bit(STRIPE_PREREAD_ACTIVE,
3998 &sh->state)) {
45b4233c 3999 pr_debug("Read_old block "
a4456856
DW		 4000 "%d for Reconstruct\n", i);
4001 set_bit(R5_LOCKED, &dev->flags);
4002 set_bit(R5_Wantread, &dev->flags);
4003 s->locked++;
a9add5d9 4004 qread++;
a4456856
DW		 4005 } else {
4006 set_bit(STRIPE_DELAYED, &sh->state);
4007 set_bit(STRIPE_HANDLE, &sh->state);
4008 }
4009 }
4010 }
e3620a3a 4011 if (rcw && conf->mddev->queue)
a9add5d9
N		 4012 blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d",
4013 (unsigned long long)sh->sector,
4014 rcw, qread, test_bit(STRIPE_DELAYED, &sh->state));
c8ac1803 4015 }
b1b02fe9
N		 4016
4017 if (rcw > disks && rmw > disks &&
4018 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4019 set_bit(STRIPE_DELAYED, &sh->state);
4020
a4456856
DW		 4021 /* now if nothing is locked, and if we have enough data,
4022 * we can start a write request
4023 */
f38e1219
DW		 4024 /* since handle_stripe can be called at any time we need to handle the
4025 * case where a compute block operation has been submitted and then a
ac6b53b6
DW		 4026 * subsequent call wants to start a write request. raid_run_ops only
4027 * handles the case where compute block and reconstruct are requested
f38e1219
DW		 4028 * simultaneously. If this is not the case then new writes need to be
4029 * held off until the compute completes.
4030 */
976ea8d4
DW		 4031 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
4032 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
1e6d690b 4033 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
c0f7bddb 4034 schedule_reconstruction(sh, s, rcw == 0, 0);
d7bd398e 4035 return 0;
a4456856
DW		 4036}
4037
d1688a6d 4038static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh,
a4456856
DW		 4039 struct stripe_head_state *s, int disks)
4040{
ecc65c9b 4041 struct r5dev *dev = NULL;
bd2ab670 4042
59fc630b 4043 BUG_ON(sh->batch_head);
a4456856 4044 set_bit(STRIPE_HANDLE, &sh->state);
e89f8962 4045
ecc65c9b
DW		 4046 switch (sh->check_state) {
4047 case check_state_idle:
4048 /* start a new check operation if there are no failures */
bd2ab670 4049 if (s->failed == 0) {
bd2ab670 4050 BUG_ON(s->uptodate != disks);
ecc65c9b
DW		 4051 sh->check_state = check_state_run;
4052 set_bit(STRIPE_OP_CHECK, &s->ops_request);
bd2ab670 4053 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
bd2ab670 4054 s->uptodate--;
ecc65c9b 4055 break;
bd2ab670 4056 }
f2b3b44d 4057 dev = &sh->dev[s->failed_num[0]];
ecc65c9b
DW		 4058 /* fall through */
4059 case check_state_compute_result:
4060 sh->check_state = check_state_idle;
4061 if (!dev)
4062 dev = &sh->dev[sh->pd_idx];
4063
4064 /* check that a write has not made the stripe insync */
4065 if (test_bit(STRIPE_INSYNC, &sh->state))
4066 break;
c8894419 4067
a4456856 4068 /* either failed parity check, or recovery is happening */
a4456856
DW		 4069 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
4070 BUG_ON(s->uptodate != disks);
4071
4072 set_bit(R5_LOCKED, &dev->flags);
ecc65c9b 4073 s->locked++;
a4456856 4074 set_bit(R5_Wantwrite, &dev->flags);
830ea016 4075
a4456856 4076 clear_bit(STRIPE_DEGRADED, &sh->state);
a4456856 4077 set_bit(STRIPE_INSYNC, &sh->state);
ecc65c9b
DW		 4078 break;
4079 case check_state_run:
4080 break; /* we will be called again upon completion */
4081 case check_state_check_result:
4082 sh->check_state = check_state_idle;
4083
4084 /* if a failure occurred during the check operation, leave
4085 * STRIPE_INSYNC not set and let the stripe be handled again
4086 */
4087 if (s->failed)
4088 break;
4089
4090 /* handle a successful check operation, if parity is correct
4091 * we are done. Otherwise update the mismatch count and repair
4092 * parity if !MD_RECOVERY_CHECK
4093 */
ad283ea4 4094 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
ecc65c9b
DW		 4095 /* parity is correct (on disc,
4096 * not in buffer any more)
4097 */
4098 set_bit(STRIPE_INSYNC, &sh->state);
4099 else {
7f7583d4 4100 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
e1539036 4101 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) {
ecc65c9b
DW		 4102 /* don't try to repair!! */
4103 set_bit(STRIPE_INSYNC, &sh->state);
e1539036
N		 4104 pr_warn_ratelimited("%s: mismatch sector in range "
4105 "%llu-%llu\n", mdname(conf->mddev),
4106 (unsigned long long) sh->sector,
4107 (unsigned long long) sh->sector +
4108 STRIPE_SECTORS);
4109 } else {
ecc65c9b 4110 sh->check_state = check_state_compute_run;
976ea8d4 4111 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
ecc65c9b
DW		 4112 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
4113 set_bit(R5_Wantcompute,
4114 &sh->dev[sh->pd_idx].flags);
4115 sh->ops.target = sh->pd_idx;
ac6b53b6 4116 sh->ops.target2 = -1;
ecc65c9b
DW		 4117 s->uptodate++;
4118 }
4119 }
4120 break;
4121 case check_state_compute_run:
4122 break;
4123 default:
cc6167b4 4124 pr_err("%s: unknown check_state: %d sector: %llu\n",
ecc65c9b
DW		 4125 __func__, sh->check_state,
4126 (unsigned long long) sh->sector);
4127 BUG();
a4456856
DW		 4128 }
4129}
4130
d1688a6d 4131static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh,
36d1c647 4132 struct stripe_head_state *s,
f2b3b44d 4133 int disks)
a4456856 4134{
a4456856 4135 int pd_idx = sh->pd_idx;
34e04e87 4136 int qd_idx = sh->qd_idx;
d82dfee0 4137 struct r5dev *dev;
a4456856 4138
59fc630b 4139 BUG_ON(sh->batch_head);
a4456856
DW		 4140 set_bit(STRIPE_HANDLE, &sh->state);
4141
4142 BUG_ON(s->failed > 2);
d82dfee0 4143
a4456856
DW		 4144 /* Want to check and possibly repair P and Q.
4145 * However there could be one 'failed' device, in which
4146 * case we can only check one of them, possibly using the
4147 * other to generate missing data
4148 */
4149
d82dfee0
DW		 4150 switch (sh->check_state) {
4151 case check_state_idle:
4152 /* start a new check operation if there are < 2 failures */
f2b3b44d 4153 if (s->failed == s->q_failed) {
d82dfee0 4154 /* The only possible failed device holds Q, so it
a4456856
DW		 4155 * makes sense to check P (If anything else were failed,
4156 * we would have used P to recreate it).
4157 */
d82dfee0 4158 sh->check_state = check_state_run;
a4456856 4159 }
f2b3b44d 4160 if (!s->q_failed && s->failed < 2) {
d82dfee0 4161 /* Q is not failed, and we didn't use it to generate
a4456856
DW		 4162 * anything, so it makes sense to check it
4163 */
d82dfee0
DW		 4164 if (sh->check_state == check_state_run)
4165 sh->check_state = check_state_run_pq;
4166 else
4167 sh->check_state = check_state_run_q;
a4456856 4168 }
a4456856 4169
d82dfee0
DW		 4170 /* discard potentially stale zero_sum_result */
4171 sh->ops.zero_sum_result = 0;
a4456856 4172
d82dfee0
DW		 4173 if (sh->check_state == check_state_run) {
4174 /* async_xor_zero_sum destroys the contents of P */
4175 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
4176 s->uptodate--;
a4456856 4177 }
d82dfee0
DW		 4178 if (sh->check_state >= check_state_run &&
4179 sh->check_state <= check_state_run_pq) {
4180 /* async_syndrome_zero_sum preserves P and Q, so
4181 * no need to mark them !uptodate here
4182 */
4183 set_bit(STRIPE_OP_CHECK, &s->ops_request);
4184 break;
a4456856
DW		 4185 }
4186
d82dfee0
DW		 4187 /* we have 2-disk failure */
4188 BUG_ON(s->failed != 2);
4189 /* fall through */
4190 case check_state_compute_result:
4191 sh->check_state = check_state_idle;
a4456856 4192
d82dfee0
DW		 4193 /* check that a write has not made the stripe insync */
4194 if (test_bit(STRIPE_INSYNC, &sh->state))
4195 break;
a4456856
DW		 4196
4197 /* now write out any block on a failed drive,
d82dfee0 4198 * or P or Q if they were recomputed
a4456856 4199 */
d82dfee0 4200 BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */
a4456856 4201 if (s->failed == 2) {
f2b3b44d 4202 dev = &sh->dev[s->failed_num[1]];
a4456856
DW		 4203 s->locked++;
4204 set_bit(R5_LOCKED, &dev->flags);
4205 set_bit(R5_Wantwrite, &dev->flags);
4206 }
4207 if (s->failed >= 1) {
f2b3b44d 4208 dev = &sh->dev[s->failed_num[0]];
a4456856
DW		 4209 s->locked++;
4210 set_bit(R5_LOCKED, &dev->flags);
4211 set_bit(R5_Wantwrite, &dev->flags);
4212 }
d82dfee0 4213 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
a4456856
DW		 4214 dev = &sh->dev[pd_idx];
4215 s->locked++;
4216 set_bit(R5_LOCKED, &dev->flags);
4217 set_bit(R5_Wantwrite, &dev->flags);
4218 }
d82dfee0 4219 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
a4456856
DW		 4220 dev = &sh->dev[qd_idx];
4221 s->locked++;
4222 set_bit(R5_LOCKED, &dev->flags);
4223 set_bit(R5_Wantwrite, &dev->flags);
4224 }
4225 clear_bit(STRIPE_DEGRADED, &sh->state);
4226
4227 set_bit(STRIPE_INSYNC, &sh->state);
d82dfee0
DW		 4228 break;
4229 case check_state_run:
4230 case check_state_run_q:
4231 case check_state_run_pq:
4232 break; /* we will be called again upon completion */
4233 case check_state_check_result:
4234 sh->check_state = check_state_idle;
4235
4236 /* handle a successful check operation, if parity is correct
4237 * we are done. Otherwise update the mismatch count and repair
4238 * parity if !MD_RECOVERY_CHECK
4239 */
4240 if (sh->ops.zero_sum_result == 0) {
4241 /* both parities are correct */
4242 if (!s->failed)
4243 set_bit(STRIPE_INSYNC, &sh->state);
4244 else {
4245 /* in contrast to the raid5 case we can validate
4246 * parity, but still have a failure to write
4247 * back
4248 */
4249 sh->check_state = check_state_compute_result;
4250 /* Returning at this point means that we may go
4251 * off and bring p and/or q uptodate again so
4252 * we make sure to check zero_sum_result again
4253 * to verify if p or q need writeback
4254 */
4255 }
4256 } else {
7f7583d4 4257 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
e1539036 4258 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) {
d82dfee0
DW		 4259 /* don't try to repair!! */
4260 set_bit(STRIPE_INSYNC, &sh->state);
e1539036
N		 4261 pr_warn_ratelimited("%s: mismatch sector in range "
4262 "%llu-%llu\n", mdname(conf->mddev),
4263 (unsigned long long) sh->sector,
4264 (unsigned long long) sh->sector +
4265 STRIPE_SECTORS);
4266 } else {
d82dfee0
DW		 4267 int *target = &sh->ops.target;
4268
4269 sh->ops.target = -1;
4270 sh->ops.target2 = -1;
4271 sh->check_state = check_state_compute_run;
4272 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
4273 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
4274 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
4275 set_bit(R5_Wantcompute,
4276 &sh->dev[pd_idx].flags);
4277 *target = pd_idx;
4278 target = &sh->ops.target2;
4279 s->uptodate++;
4280 }
4281 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
4282 set_bit(R5_Wantcompute,
4283 &sh->dev[qd_idx].flags);
4284 *target = qd_idx;
4285 s->uptodate++;
4286 }
4287 }
4288 }
4289 break;
4290 case check_state_compute_run:
4291 break;
4292 default:
cc6167b4
N		 4293 pr_warn("%s: unknown check_state: %d sector: %llu\n",
4294 __func__, sh->check_state,
4295 (unsigned long long) sh->sector);
d82dfee0 4296 BUG();
a4456856
DW		 4297 }
4298}
4299
d1688a6d 4300static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh)
a4456856
DW		 4301{
4302 int i;
4303
4304 /* We have read all the blocks in this stripe and now we need to
4305 * copy some of them into a target stripe for expand.
4306 */
f0a50d37 4307 struct dma_async_tx_descriptor *tx = NULL;
59fc630b 4308 BUG_ON(sh->batch_head);
a4456856
DW		 4309 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
4310 for (i = 0; i < sh->disks; i++)
34e04e87 4311 if (i != sh->pd_idx && i != sh->qd_idx) {
911d4ee8 4312 int dd_idx, j;
a4456856 4313 struct stripe_head *sh2;
a08abd8c 4314 struct async_submit_ctl submit;
a4456856 4315
6d036f7d 4316 sector_t bn = raid5_compute_blocknr(sh, i, 1);
911d4ee8
N		 4317 sector_t s = raid5_compute_sector(conf, bn, 0,
4318 &dd_idx, NULL);
6d036f7d 4319 sh2 = raid5_get_active_stripe(conf, s, 0, 1, 1);
a4456856
DW		 4320 if (sh2 == NULL)
4321 /* so far only the early blocks of this stripe
4322 * have been requested. When later blocks
4323 * get requested, we will try again
4324 */
4325 continue;
4326 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
4327 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
4328 /* must have already done this block */
6d036f7d 4329 raid5_release_stripe(sh2);
a4456856
DW		 4330 continue;
4331 }
f0a50d37
DW		 4332
4333 /* place all the copies on one channel */
a08abd8c 4334 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
f0a50d37 4335 tx = async_memcpy(sh2->dev[dd_idx].page,
88ba2aa5 4336 sh->dev[i].page, 0, 0, STRIPE_SIZE,
a08abd8c 4337 &submit);
f0a50d37 4338
a4456856
DW		 4339 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
4340 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
4341 for (j = 0; j < conf->raid_disks; j++)
4342 if (j != sh2->pd_idx &&
86c374ba 4343 j != sh2->qd_idx &&
a4456856
DW		 4344 !test_bit(R5_Expanded, &sh2->dev[j].flags))
4345 break;
4346 if (j == conf->raid_disks) {
4347 set_bit(STRIPE_EXPAND_READY, &sh2->state);
4348 set_bit(STRIPE_HANDLE, &sh2->state);
4349 }
6d036f7d 4350 raid5_release_stripe(sh2);
f0a50d37 4351
a4456856 4352 }
a2e08551 4353 /* done submitting copies, wait for them to complete */
749586b7 4354 async_tx_quiesce(&tx);
a4456856 4355}
1da177e4
LT		 4356
4357/*
4358 * handle_stripe - do things to a stripe.
4359 *
9a3e1101
N		 4360 * We lock the stripe by setting STRIPE_ACTIVE and then examine the
4361 * state of various bits to see what needs to be done.
1da177e4 4362 * Possible results:
9a3e1101
N		 4363 * return some read requests which now have data
4364 * return some write requests which are safely on storage
1da177e4
LT		 4365 * schedule a read on some buffers
4366 * schedule a write of some buffers
4367 * return confirmation of parity correctness
4368 *
1da177e4 4369 */
a4456856 4370
acfe726b 4371static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s)
1da177e4 4372{
d1688a6d 4373 struct r5conf *conf = sh->raid_conf;
f416885e 4374 int disks = sh->disks;
474af965
N		 4375 struct r5dev *dev;
4376 int i;
9a3e1101 4377 int do_recovery = 0;
1da177e4 4378
acfe726b
N		 4379 memset(s, 0, sizeof(*s));
4380
dabc4ec6 4381 s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state) && !sh->batch_head;
4382 s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state) && !sh->batch_head;
acfe726b
N		 4383 s->failed_num[0] = -1;
4384 s->failed_num[1] = -1;
6e74a9cf 4385 s->log_failed = r5l_log_disk_error(conf);
1da177e4 4386
acfe726b 4387 /* Now to look around and see what can be done */
1da177e4 4388 rcu_read_lock();
16a53ecc 4389 for (i=disks; i--; ) {
3cb03002 4390 struct md_rdev *rdev;
31c176ec
N		 4391 sector_t first_bad;
4392 int bad_sectors;
4393 int is_bad = 0;
acfe726b 4394
16a53ecc 4395 dev = &sh->dev[i];
1da177e4 4396
45b4233c 4397 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
9a3e1101
N		 4398 i, dev->flags,
4399 dev->toread, dev->towrite, dev->written);
6c0069c0
YT		 4400 /* maybe we can reply to a read
4401 *
4402 * new wantfill requests are only permitted while
4403 * ops_complete_biofill is guaranteed to be inactive
4404 */
4405 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
4406 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
4407 set_bit(R5_Wantfill, &dev->flags);
1da177e4 4408
16a53ecc 4409 /* now count some things */
cc94015a
N		 4410 if (test_bit(R5_LOCKED, &dev->flags))
4411 s->locked++;
4412 if (test_bit(R5_UPTODATE, &dev->flags))
4413 s->uptodate++;
2d6e4ecc 4414 if (test_bit(R5_Wantcompute, &dev->flags)) {
cc94015a
N		 4415 s->compute++;
4416 BUG_ON(s->compute > 2);
2d6e4ecc 4417 }
1da177e4 4418
acfe726b 4419 if (test_bit(R5_Wantfill, &dev->flags))
cc94015a 4420 s->to_fill++;
acfe726b 4421 else if (dev->toread)
cc94015a 4422 s->to_read++;
16a53ecc 4423 if (dev->towrite) {
cc94015a 4424 s->to_write++;
16a53ecc 4425 if (!test_bit(R5_OVERWRITE, &dev->flags))
cc94015a 4426 s->non_overwrite++;
16a53ecc 4427 }
a4456856 4428 if (dev->written)
cc94015a 4429 s->written++;
14a75d3e
N		 4430 /* Prefer to use the replacement for reads, but only
4431 * if it is recovered enough and has no bad blocks.
4432 */
4433 rdev = rcu_dereference(conf->disks[i].replacement);
4434 if (rdev && !test_bit(Faulty, &rdev->flags) &&
4435 rdev->recovery_offset >= sh->sector + STRIPE_SECTORS &&
4436 !is_badblock(rdev, sh->sector, STRIPE_SECTORS,
4437 &first_bad, &bad_sectors))
4438 set_bit(R5_ReadRepl, &dev->flags);
4439 else {
e6030cb0 4440 if (rdev && !test_bit(Faulty, &rdev->flags))
9a3e1101 4441 set_bit(R5_NeedReplace, &dev->flags);
e6030cb0
N		 4442 else
4443 clear_bit(R5_NeedReplace, &dev->flags);
14a75d3e
N		 4444 rdev = rcu_dereference(conf->disks[i].rdev);
4445 clear_bit(R5_ReadRepl, &dev->flags);
4446 }
9283d8c5
N		 4447 if (rdev && test_bit(Faulty, &rdev->flags))
4448 rdev = NULL;
31c176ec
N		 4449 if (rdev) {
4450 is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
4451 &first_bad, &bad_sectors);
4452 if (s->blocked_rdev == NULL
4453 && (test_bit(Blocked, &rdev->flags)
4454 || is_bad < 0)) {
4455 if (is_bad < 0)
4456 set_bit(BlockedBadBlocks,
4457 &rdev->flags);
4458 s->blocked_rdev = rdev;
4459 atomic_inc(&rdev->nr_pending);
4460 }
6bfe0b49 4461 }
415e72d0
N		 4462 clear_bit(R5_Insync, &dev->flags);
4463 if (!rdev)
4464 /* Not in-sync */;
31c176ec
N		 4465 else if (is_bad) {
4466 /* also not in-sync */
18b9837e
N		 4467 if (!test_bit(WriteErrorSeen, &rdev->flags) &&
4468 test_bit(R5_UPTODATE, &dev->flags)) {
31c176ec
N		 4469 /* treat as in-sync, but with a read error
4470 * which we can now try to correct
4471 */
4472 set_bit(R5_Insync, &dev->flags);
4473 set_bit(R5_ReadError, &dev->flags);
4474 }
4475 } else if (test_bit(In_sync, &rdev->flags))
415e72d0 4476 set_bit(R5_Insync, &dev->flags);
30d7a483 4477 else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
415e72d0 4478 /* in sync if before recovery_offset */
30d7a483
N		 4479 set_bit(R5_Insync, &dev->flags);
4480 else if (test_bit(R5_UPTODATE, &dev->flags) &&
4481 test_bit(R5_Expanded, &dev->flags))
4482 /* If we've reshaped into here, we assume it is Insync.
4483 * We will shortly update recovery_offset to make
4484 * it official.
4485 */
4486 set_bit(R5_Insync, &dev->flags);
4487
1cc03eb9 4488 if (test_bit(R5_WriteError, &dev->flags)) {
14a75d3e
N		 4489 /* This flag does not apply to '.replacement'
4490 * only to .rdev, so make sure to check that*/
4491 struct md_rdev *rdev2 = rcu_dereference(
4492 conf->disks[i].rdev);
4493 if (rdev2 == rdev)
4494 clear_bit(R5_Insync, &dev->flags);
4495 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
bc2607f3 4496 s->handle_bad_blocks = 1;
14a75d3e 4497 atomic_inc(&rdev2->nr_pending);
bc2607f3
N		 4498 } else
4499 clear_bit(R5_WriteError, &dev->flags);
4500 }
1cc03eb9 4501 if (test_bit(R5_MadeGood, &dev->flags)) {
14a75d3e
N		 4502 /* This flag does not apply to '.replacement'
4503 * only to .rdev, so make sure to check that*/
4504 struct md_rdev *rdev2 = rcu_dereference(
4505 conf->disks[i].rdev);
4506 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
b84db560 4507 s->handle_bad_blocks = 1;
14a75d3e 4508 atomic_inc(&rdev2->nr_pending);
b84db560
N		 4509 } else
4510 clear_bit(R5_MadeGood, &dev->flags);
4511 }
977df362
N		 4512 if (test_bit(R5_MadeGoodRepl, &dev->flags)) {
4513 struct md_rdev *rdev2 = rcu_dereference(
4514 conf->disks[i].replacement);
4515 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
4516 s->handle_bad_blocks = 1;
4517 atomic_inc(&rdev2->nr_pending);
4518 } else
4519 clear_bit(R5_MadeGoodRepl, &dev->flags);
4520 }
415e72d0 4521 if (!test_bit(R5_Insync, &dev->flags)) {
16a53ecc
N		 4522 /* The ReadError flag will just be confusing now */
4523 clear_bit(R5_ReadError, &dev->flags);
4524 clear_bit(R5_ReWrite, &dev->flags);
1da177e4 4525 }
415e72d0
N		 4526 if (test_bit(R5_ReadError, &dev->flags))
4527 clear_bit(R5_Insync, &dev->flags);
4528 if (!test_bit(R5_Insync, &dev->flags)) {
cc94015a
N		 4529 if (s->failed < 2)
4530 s->failed_num[s->failed] = i;
4531 s->failed++;
9a3e1101
N		 4532 if (rdev && !test_bit(Faulty, &rdev->flags))
4533 do_recovery = 1;
d63e2fc8
BC		 4534 else if (!rdev) {
4535 rdev = rcu_dereference(
4536 conf->disks[i].replacement);
4537 if (rdev && !test_bit(Faulty, &rdev->flags))
4538 do_recovery = 1;
4539 }
415e72d0 4540 }
2ded3703
SL		 4541
4542 if (test_bit(R5_InJournal, &dev->flags))
4543 s->injournal++;
1e6d690b
SL		 4544 if (test_bit(R5_InJournal, &dev->flags) && dev->written)
4545 s->just_cached++;
1da177e4 4546 }
9a3e1101
N		 4547 if (test_bit(STRIPE_SYNCING, &sh->state)) {
4548 /* If there is a failed device being replaced,
4549 * we must be recovering.
4550 * else if we are after recovery_cp, we must be syncing
c6d2e084 4551 * else if MD_RECOVERY_REQUESTED is set, we also are syncing.
9a3e1101
N		 4552 * else we can only be replacing
4553 * sync and recovery both need to read all devices, and so
4554 * use the same flag.
4555 */
4556 if (do_recovery ||
c6d2e084 4557 sh->sector >= conf->mddev->recovery_cp ||
4558 test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery)))
9a3e1101
N		 4559 s->syncing = 1;
4560 else
4561 s->replacing = 1;
4562 }
1da177e4 4563 rcu_read_unlock();
cc94015a
N		 4564}
4565
59fc630b 4566static int clear_batch_ready(struct stripe_head *sh)
4567{
b15a9dbd
N		 4568 /* Return '1' if this is a member of batch, or
4569 * '0' if it is a lone stripe or a head which can now be
4570 * handled.
4571 */
59fc630b 4572 struct stripe_head *tmp;
4573 if (!test_and_clear_bit(STRIPE_BATCH_READY, &sh->state))
b15a9dbd 4574 return (sh->batch_head && sh->batch_head != sh);
59fc630b 4575 spin_lock(&sh->stripe_lock);
4576 if (!sh->batch_head) {
4577 spin_unlock(&sh->stripe_lock);
4578 return 0;
4579 }
4580
4581 /*
4582 * this stripe could be added to a batch list before we check
4583 * BATCH_READY, skips it
4584 */
4585 if (sh->batch_head != sh) {
4586 spin_unlock(&sh->stripe_lock);
4587 return 1;
4588 }
4589 spin_lock(&sh->batch_lock);
4590 list_for_each_entry(tmp, &sh->batch_list, batch_list)
4591 clear_bit(STRIPE_BATCH_READY, &tmp->state);
4592 spin_unlock(&sh->batch_lock);
4593 spin_unlock(&sh->stripe_lock);
4594
4595 /*
4596 * BATCH_READY is cleared, no new stripes can be added.
4597 * batch_list can be accessed without lock
4598 */
4599 return 0;
4600}
4601
3960ce79
N		 4602static void break_stripe_batch_list(struct stripe_head *head_sh,
4603 unsigned long handle_flags)
72ac7330 4604{
4e3d62ff 4605 struct stripe_head *sh, *next;
72ac7330 4606 int i;
fb642b92 4607 int do_wakeup = 0;
72ac7330 4608
bb27051f
N		 4609 list_for_each_entry_safe(sh, next, &head_sh->batch_list, batch_list) {
4610
72ac7330 4611 list_del_init(&sh->batch_list);
4612
fb3229d5 4613 WARN_ONCE(sh->state & ((1 << STRIPE_ACTIVE) |
1b956f7a
N		 4614 (1 << STRIPE_SYNCING) |
4615 (1 << STRIPE_REPLACED) |
1b956f7a
N		 4616 (1 << STRIPE_DELAYED) |
4617 (1 << STRIPE_BIT_DELAY) |
4618 (1 << STRIPE_FULL_WRITE) |
4619 (1 << STRIPE_BIOFILL_RUN) |
4620 (1 << STRIPE_COMPUTE_RUN) |
4621 (1 << STRIPE_OPS_REQ_PENDING) |
4622 (1 << STRIPE_DISCARD) |
4623 (1 << STRIPE_BATCH_READY) |
4624 (1 << STRIPE_BATCH_ERR) |
fb3229d5
SL		 4625 (1 << STRIPE_BITMAP_PENDING)),
4626 "stripe state: %lx\n", sh->state);
4627 WARN_ONCE(head_sh->state & ((1 << STRIPE_DISCARD) |
4628 (1 << STRIPE_REPLACED)),
4629 "head stripe state: %lx\n", head_sh->state);
1b956f7a
N		 4630
4631 set_mask_bits(&sh->state, ~(STRIPE_EXPAND_SYNC_FLAGS |
550da24f 4632 (1 << STRIPE_PREREAD_ACTIVE) |
184a09eb
DY		 4633 (1 << STRIPE_DEGRADED) |
4634 (1 << STRIPE_ON_UNPLUG_LIST)),
1b956f7a
N		 4635 head_sh->state & (1 << STRIPE_INSYNC));
4636
72ac7330 4637 sh->check_state = head_sh->check_state;
4638 sh->reconstruct_state = head_sh->reconstruct_state;
448ec638
AC		 4639 spin_lock_irq(&sh->stripe_lock);
4640 sh->batch_head = NULL;
4641 spin_unlock_irq(&sh->stripe_lock);
fb642b92
N		 4642 for (i = 0; i < sh->disks; i++) {
4643 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
4644 do_wakeup = 1;
72ac7330 4645 sh->dev[i].flags = head_sh->dev[i].flags &
4646 (~((1 << R5_WriteError) | (1 << R5_Overlap)));
fb642b92 4647 }
3960ce79
N		 4648 if (handle_flags == 0 ||
4649 sh->state & handle_flags)
4650 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 4651 raid5_release_stripe(sh);
72ac7330 4652 }
fb642b92
N		 4653 spin_lock_irq(&head_sh->stripe_lock);
4654 head_sh->batch_head = NULL;
4655 spin_unlock_irq(&head_sh->stripe_lock);
4656 for (i = 0; i < head_sh->disks; i++)
4657 if (test_and_clear_bit(R5_Overlap, &head_sh->dev[i].flags))
4658 do_wakeup = 1;
3960ce79
N		 4659 if (head_sh->state & handle_flags)
4660 set_bit(STRIPE_HANDLE, &head_sh->state);
fb642b92
N		 4661
4662 if (do_wakeup)
4663 wake_up(&head_sh->raid_conf->wait_for_overlap);
72ac7330 4664}
4665
cc94015a
N		 4666static void handle_stripe(struct stripe_head *sh)
4667{
4668 struct stripe_head_state s;
d1688a6d 4669 struct r5conf *conf = sh->raid_conf;
3687c061 4670 int i;
84789554
N		 4671 int prexor;
4672 int disks = sh->disks;
474af965 4673 struct r5dev *pdev, *qdev;
cc94015a
N		 4674
4675 clear_bit(STRIPE_HANDLE, &sh->state);
257a4b42 4676 if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) {
cc94015a
N		 4677 /* already being handled, ensure it gets handled
4678 * again when current action finishes */
4679 set_bit(STRIPE_HANDLE, &sh->state);
4680 return;
4681 }
4682
59fc630b 4683 if (clear_batch_ready(sh) ) {
4684 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
4685 return;
4686 }
4687
4e3d62ff 4688 if (test_and_clear_bit(STRIPE_BATCH_ERR, &sh->state))
3960ce79 4689 break_stripe_batch_list(sh, 0);
72ac7330 4690
dabc4ec6 4691 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) && !sh->batch_head) {
f8dfcffd 4692 spin_lock(&sh->stripe_lock);
5ddf0440
SL		 4693 /*
4694 * Cannot process 'sync' concurrently with 'discard'.
4695 * Flush data in r5cache before 'sync'.
4696 */
4697 if (!test_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state) &&
4698 !test_bit(STRIPE_R5C_FULL_STRIPE, &sh->state) &&
4699 !test_bit(STRIPE_DISCARD, &sh->state) &&
f8dfcffd
N		 4700 test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
4701 set_bit(STRIPE_SYNCING, &sh->state);
4702 clear_bit(STRIPE_INSYNC, &sh->state);
f94c0b66 4703 clear_bit(STRIPE_REPLACED, &sh->state);
f8dfcffd
N		 4704 }
4705 spin_unlock(&sh->stripe_lock);
cc94015a
N		 4706 }
4707 clear_bit(STRIPE_DELAYED, &sh->state);
4708
4709 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
4710 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
4711 (unsigned long long)sh->sector, sh->state,
4712 atomic_read(&sh->count), sh->pd_idx, sh->qd_idx,
4713 sh->check_state, sh->reconstruct_state);
3687c061 4714
acfe726b 4715 analyse_stripe(sh, &s);
c5a31000 4716
b70abcb2
SL		 4717 if (test_bit(STRIPE_LOG_TRAPPED, &sh->state))
4718 goto finish;
4719
16d997b7
N		 4720 if (s.handle_bad_blocks ||
4721 test_bit(MD_SB_CHANGE_PENDING, &conf->mddev->sb_flags)) {
bc2607f3
N		 4722 set_bit(STRIPE_HANDLE, &sh->state);
4723 goto finish;
4724 }
4725
474af965
N		 4726 if (unlikely(s.blocked_rdev)) {
4727 if (s.syncing || s.expanding || s.expanded ||
9a3e1101 4728 s.replacing || s.to_write || s.written) {
474af965
N		 4729 set_bit(STRIPE_HANDLE, &sh->state);
4730 goto finish;
4731 }
4732 /* There is nothing for the blocked_rdev to block */
4733 rdev_dec_pending(s.blocked_rdev, conf->mddev);
4734 s.blocked_rdev = NULL;
4735 }
4736
4737 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
4738 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
4739 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
4740 }
4741
4742 pr_debug("locked=%d uptodate=%d to_read=%d"
4743 " to_write=%d failed=%d failed_num=%d,%d\n",
4744 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
4745 s.failed_num[0], s.failed_num[1]);
70d466f7
SL		 4746 /*
4747 * check if the array has lost more than max_degraded devices and,
474af965 4748 * if so, some requests might need to be failed.
70d466f7
SL		 4749 *
4750 * When journal device failed (log_failed), we will only process
4751 * the stripe if there is data need write to raid disks
474af965 4752 */
70d466f7
SL		 4753 if (s.failed > conf->max_degraded ||
4754 (s.log_failed && s.injournal == 0)) {
9a3f530f
N		 4755 sh->check_state = 0;
4756 sh->reconstruct_state = 0;
626f2092 4757 break_stripe_batch_list(sh, 0);
9a3f530f 4758 if (s.to_read+s.to_write+s.written)
bd83d0a2 4759 handle_failed_stripe(conf, sh, &s, disks);
9a3e1101 4760 if (s.syncing + s.replacing)
9a3f530f
N		 4761 handle_failed_sync(conf, sh, &s);
4762 }
474af965 4763
84789554
N		 4764 /* Now we check to see if any write operations have recently
4765 * completed
4766 */
4767 prexor = 0;
4768 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
4769 prexor = 1;
4770 if (sh->reconstruct_state == reconstruct_state_drain_result ||
4771 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
4772 sh->reconstruct_state = reconstruct_state_idle;
4773
4774 /* All the 'written' buffers and the parity block are ready to
4775 * be written back to disk
4776 */
9e444768
SL		 4777 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) &&
4778 !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags));
84789554 4779 BUG_ON(sh->qd_idx >= 0 &&
9e444768
SL		 4780 !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) &&
4781 !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags));
84789554
N		 4782 for (i = disks; i--; ) {
4783 struct r5dev *dev = &sh->dev[i];
4784 if (test_bit(R5_LOCKED, &dev->flags) &&
4785 (i == sh->pd_idx || i == sh->qd_idx ||
1e6d690b
SL		 4786 dev->written || test_bit(R5_InJournal,
4787 &dev->flags))) {
84789554
N		 4788 pr_debug("Writing block %d\n", i);
4789 set_bit(R5_Wantwrite, &dev->flags);
4790 if (prexor)
4791 continue;
9c4bdf69
N		 4792 if (s.failed > 1)
4793 continue;
84789554
N		 4794 if (!test_bit(R5_Insync, &dev->flags) ||
4795 ((i == sh->pd_idx || i == sh->qd_idx) &&
4796 s.failed == 0))
4797 set_bit(STRIPE_INSYNC, &sh->state);
4798 }
4799 }
4800 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4801 s.dec_preread_active = 1;
4802 }
4803
ef5b7c69
N		 4804 /*
4805 * might be able to return some write requests if the parity blocks
4806 * are safe, or on a failed drive
4807 */
4808 pdev = &sh->dev[sh->pd_idx];
4809 s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx)
4810 || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx);
4811 qdev = &sh->dev[sh->qd_idx];
4812 s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx)
4813 || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx)
4814 || conf->level < 6;
4815
4816 if (s.written &&
4817 (s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
4818 && !test_bit(R5_LOCKED, &pdev->flags)
4819 && (test_bit(R5_UPTODATE, &pdev->flags) ||
4820 test_bit(R5_Discard, &pdev->flags))))) &&
4821 (s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
4822 && !test_bit(R5_LOCKED, &qdev->flags)
4823 && (test_bit(R5_UPTODATE, &qdev->flags) ||
4824 test_bit(R5_Discard, &qdev->flags))))))
bd83d0a2 4825 handle_stripe_clean_event(conf, sh, disks);
ef5b7c69 4826
1e6d690b 4827 if (s.just_cached)
bd83d0a2 4828 r5c_handle_cached_data_endio(conf, sh, disks);
ff875738 4829 log_stripe_write_finished(sh);
1e6d690b 4830
ef5b7c69
N		 4831 /* Now we might consider reading some blocks, either to check/generate
4832 * parity, or to satisfy requests
4833 * or to load a block that is being partially written.
4834 */
4835 if (s.to_read || s.non_overwrite
4836 || (conf->level == 6 && s.to_write && s.failed)
4837 || (s.syncing && (s.uptodate + s.compute < disks))
4838 || s.replacing
4839 || s.expanding)
4840 handle_stripe_fill(sh, &s, disks);
4841
2ded3703
SL		 4842 /*
4843 * When the stripe finishes full journal write cycle (write to journal
4844 * and raid disk), this is the clean up procedure so it is ready for
4845 * next operation.
4846 */
4847 r5c_finish_stripe_write_out(conf, sh, &s);
4848
4849 /*
4850 * Now to consider new write requests, cache write back and what else,
4851 * if anything should be read. We do not handle new writes when:
84789554
N		 4852 * 1/ A 'write' operation (copy+xor) is already in flight.
4853 * 2/ A 'check' operation is in flight, as it may clobber the parity
4854 * block.
2ded3703 4855 * 3/ A r5c cache log write is in flight.
84789554 4856 */
2ded3703
SL		 4857
4858 if (!sh->reconstruct_state && !sh->check_state && !sh->log_io) {
4859 if (!r5c_is_writeback(conf->log)) {
4860 if (s.to_write)
4861 handle_stripe_dirtying(conf, sh, &s, disks);
4862 } else { /* write back cache */
4863 int ret = 0;
4864
4865 /* First, try handle writes in caching phase */
4866 if (s.to_write)
4867 ret = r5c_try_caching_write(conf, sh, &s,
4868 disks);
4869 /*
4870 * If caching phase failed: ret == -EAGAIN
4871 * OR
4872 * stripe under reclaim: !caching && injournal
4873 *
4874 * fall back to handle_stripe_dirtying()
4875 */
4876 if (ret == -EAGAIN ||
4877 /* stripe under reclaim: !caching && injournal */
4878 (!test_bit(STRIPE_R5C_CACHING, &sh->state) &&
d7bd398e
SL		 4879 s.injournal > 0)) {
4880 ret = handle_stripe_dirtying(conf, sh, &s,
4881 disks);
4882 if (ret == -EAGAIN)
4883 goto finish;
4884 }
2ded3703
SL		 4885 }
4886 }
84789554
N		 4887
4888 /* maybe we need to check and possibly fix the parity for this stripe
4889 * Any reads will already have been scheduled, so we just see if enough
4890 * data is available. The parity check is held off while parity
4891 * dependent operations are in flight.
4892 */
4893 if (sh->check_state ||
4894 (s.syncing && s.locked == 0 &&
4895 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
4896 !test_bit(STRIPE_INSYNC, &sh->state))) {
4897 if (conf->level == 6)
4898 handle_parity_checks6(conf, sh, &s, disks);
4899 else
4900 handle_parity_checks5(conf, sh, &s, disks);
4901 }
c5a31000 4902
f94c0b66
N		 4903 if ((s.replacing || s.syncing) && s.locked == 0
4904 && !test_bit(STRIPE_COMPUTE_RUN, &sh->state)
4905 && !test_bit(STRIPE_REPLACED, &sh->state)) {
9a3e1101
N		 4906 /* Write out to replacement devices where possible */
4907 for (i = 0; i < conf->raid_disks; i++)
f94c0b66
N		 4908 if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) {
4909 WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags));
9a3e1101
N		 4910 set_bit(R5_WantReplace, &sh->dev[i].flags);
4911 set_bit(R5_LOCKED, &sh->dev[i].flags);
4912 s.locked++;
4913 }
f94c0b66
N		 4914 if (s.replacing)
4915 set_bit(STRIPE_INSYNC, &sh->state);
4916 set_bit(STRIPE_REPLACED, &sh->state);
9a3e1101
N		 4917 }
4918 if ((s.syncing || s.replacing) && s.locked == 0 &&
f94c0b66 4919 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
9a3e1101 4920 test_bit(STRIPE_INSYNC, &sh->state)) {
c5a31000
N		 4921 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
4922 clear_bit(STRIPE_SYNCING, &sh->state);
f8dfcffd
N		 4923 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
4924 wake_up(&conf->wait_for_overlap);
c5a31000
N		 4925 }
4926
4927 /* If the failed drives are just a ReadError, then we might need
4928 * to progress the repair/check process
4929 */
4930 if (s.failed <= conf->max_degraded && !conf->mddev->ro)
4931 for (i = 0; i < s.failed; i++) {
4932 struct r5dev *dev = &sh->dev[s.failed_num[i]];
4933 if (test_bit(R5_ReadError, &dev->flags)
4934 && !test_bit(R5_LOCKED, &dev->flags)
4935 && test_bit(R5_UPTODATE, &dev->flags)
4936 ) {
4937 if (!test_bit(R5_ReWrite, &dev->flags)) {
4938 set_bit(R5_Wantwrite, &dev->flags);
4939 set_bit(R5_ReWrite, &dev->flags);
4940 set_bit(R5_LOCKED, &dev->flags);
4941 s.locked++;
4942 } else {
4943 /* let's read it back */
4944 set_bit(R5_Wantread, &dev->flags);
4945 set_bit(R5_LOCKED, &dev->flags);
4946 s.locked++;
4947 }
4948 }
4949 }
4950
3687c061
N		 4951 /* Finish reconstruct operations initiated by the expansion process */
4952 if (sh->reconstruct_state == reconstruct_state_result) {
4953 struct stripe_head *sh_src
6d036f7d 4954 = raid5_get_active_stripe(conf, sh->sector, 1, 1, 1);
3687c061
N		 4955 if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) {
4956 /* sh cannot be written until sh_src has been read.
4957 * so arrange for sh to be delayed a little
4958 */
4959 set_bit(STRIPE_DELAYED, &sh->state);
4960 set_bit(STRIPE_HANDLE, &sh->state);
4961 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
4962 &sh_src->state))
4963 atomic_inc(&conf->preread_active_stripes);
6d036f7d 4964 raid5_release_stripe(sh_src);
3687c061
N		 4965 goto finish;
4966 }
4967 if (sh_src)
6d036f7d 4968 raid5_release_stripe(sh_src);
3687c061
N		 4969
4970 sh->reconstruct_state = reconstruct_state_idle;
4971 clear_bit(STRIPE_EXPANDING, &sh->state);
4972 for (i = conf->raid_disks; i--; ) {
4973 set_bit(R5_Wantwrite, &sh->dev[i].flags);
4974 set_bit(R5_LOCKED, &sh->dev[i].flags);
4975 s.locked++;
4976 }
4977 }
f416885e 4978
3687c061
N		 4979 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
4980 !sh->reconstruct_state) {
4981 /* Need to write out all blocks after computing parity */
4982 sh->disks = conf->raid_disks;
4983 stripe_set_idx(sh->sector, conf, 0, sh);
4984 schedule_reconstruction(sh, &s, 1, 1);
4985 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
4986 clear_bit(STRIPE_EXPAND_READY, &sh->state);
4987 atomic_dec(&conf->reshape_stripes);
4988 wake_up(&conf->wait_for_overlap);
4989 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
4990 }
4991
4992 if (s.expanding && s.locked == 0 &&
4993 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
4994 handle_stripe_expansion(conf, sh);
16a53ecc 4995
3687c061 4996finish:
6bfe0b49 4997 /* wait for this device to become unblocked */
5f066c63
N		 4998 if (unlikely(s.blocked_rdev)) {
4999 if (conf->mddev->external)
5000 md_wait_for_blocked_rdev(s.blocked_rdev,
5001 conf->mddev);
5002 else
5003 /* Internal metadata will immediately
5004 * be written by raid5d, so we don't
5005 * need to wait here.
5006 */
5007 rdev_dec_pending(s.blocked_rdev,
5008 conf->mddev);
5009 }
6bfe0b49 5010
bc2607f3
N		 5011 if (s.handle_bad_blocks)
5012 for (i = disks; i--; ) {
3cb03002 5013 struct md_rdev *rdev;
bc2607f3
N		 5014 struct r5dev *dev = &sh->dev[i];
5015 if (test_and_clear_bit(R5_WriteError, &dev->flags)) {
5016 /* We own a safe reference to the rdev */
5017 rdev = conf->disks[i].rdev;
5018 if (!rdev_set_badblocks(rdev, sh->sector,
5019 STRIPE_SECTORS, 0))
5020 md_error(conf->mddev, rdev);
5021 rdev_dec_pending(rdev, conf->mddev);
5022 }
b84db560
N		 5023 if (test_and_clear_bit(R5_MadeGood, &dev->flags)) {
5024 rdev = conf->disks[i].rdev;
5025 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 5026 STRIPE_SECTORS, 0);
b84db560
N		 5027 rdev_dec_pending(rdev, conf->mddev);
5028 }
977df362
N		 5029 if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) {
5030 rdev = conf->disks[i].replacement;
dd054fce
N		 5031 if (!rdev)
5032 /* rdev have been moved down */
5033 rdev = conf->disks[i].rdev;
977df362 5034 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 5035 STRIPE_SECTORS, 0);
977df362
N		 5036 rdev_dec_pending(rdev, conf->mddev);
5037 }
bc2607f3
N		 5038 }
5039
6c0069c0
YT		 5040 if (s.ops_request)
5041 raid_run_ops(sh, s.ops_request);
5042
f0e43bcd 5043 ops_run_io(sh, &s);
16a53ecc 5044
c5709ef6 5045 if (s.dec_preread_active) {
729a1866 5046 /* We delay this until after ops_run_io so that if make_request
e9c7469b 5047 * is waiting on a flush, it won't continue until the writes
729a1866
N		 5048 * have actually been submitted.
5049 */
5050 atomic_dec(&conf->preread_active_stripes);
5051 if (atomic_read(&conf->preread_active_stripes) <
5052 IO_THRESHOLD)
5053 md_wakeup_thread(conf->mddev->thread);
5054 }
5055
257a4b42 5056 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
16a53ecc
N		 5057}
5058
d1688a6d 5059static void raid5_activate_delayed(struct r5conf *conf)
16a53ecc
N		 5060{
5061 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
5062 while (!list_empty(&conf->delayed_list)) {
5063 struct list_head *l = conf->delayed_list.next;
5064 struct stripe_head *sh;
5065 sh = list_entry(l, struct stripe_head, lru);
5066 list_del_init(l);
5067 clear_bit(STRIPE_DELAYED, &sh->state);
5068 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
5069 atomic_inc(&conf->preread_active_stripes);
8b3e6cdc 5070 list_add_tail(&sh->lru, &conf->hold_list);
851c30c9 5071 raid5_wakeup_stripe_thread(sh);
16a53ecc 5072 }
482c0834 5073 }
16a53ecc
N		 5074}
5075
566c09c5
SL		 5076static void activate_bit_delay(struct r5conf *conf,
5077 struct list_head *temp_inactive_list)
16a53ecc
N		 5078{
5079 /* device_lock is held */
5080 struct list_head head;
5081 list_add(&head, &conf->bitmap_list);
5082 list_del_init(&conf->bitmap_list);
5083 while (!list_empty(&head)) {
5084 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
566c09c5 5085 int hash;
16a53ecc
N		 5086 list_del_init(&sh->lru);
5087 atomic_inc(&sh->count);
566c09c5
SL		 5088 hash = sh->hash_lock_index;
5089 __release_stripe(conf, sh, &temp_inactive_list[hash]);
16a53ecc
N		 5090 }
5091}
5092
5c675f83 5093static int raid5_congested(struct mddev *mddev, int bits)
f022b2fd 5094{
d1688a6d 5095 struct r5conf *conf = mddev->private;
f022b2fd
N		 5096
5097 /* No difference between reads and writes. Just check
5098 * how busy the stripe_cache is
5099 */
3fa841d7 5100
5423399a 5101 if (test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state))
f022b2fd 5102 return 1;
a39f7afd
SL		 5103
5104 /* Also checks whether there is pressure on r5cache log space */
5105 if (test_bit(R5C_LOG_TIGHT, &conf->cache_state))
5106 return 1;
f022b2fd
N		 5107 if (conf->quiesce)
5108 return 1;
4bda556a 5109 if (atomic_read(&conf->empty_inactive_list_nr))
f022b2fd
N		 5110 return 1;
5111
5112 return 0;
5113}
5114
fd01b88c 5115static int in_chunk_boundary(struct mddev *mddev, struct bio *bio)
f679623f 5116{
3cb5edf4 5117 struct r5conf *conf = mddev->private;
10433d04 5118 sector_t sector = bio->bi_iter.bi_sector;
3cb5edf4 5119 unsigned int chunk_sectors;
aa8b57aa 5120 unsigned int bio_sectors = bio_sectors(bio);
f679623f 5121
10433d04
CH		 5122 WARN_ON_ONCE(bio->bi_partno);
5123
3cb5edf4 5124 chunk_sectors = min(conf->chunk_sectors, conf->prev_chunk_sectors);
f679623f
RBJ		 5125 return chunk_sectors >=
5126 ((sector & (chunk_sectors - 1)) + bio_sectors);
5127}
5128
46031f9a
RBJ		 5129/*
5130 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
5131 * later sampled by raid5d.
5132 */
d1688a6d 5133static void add_bio_to_retry(struct bio *bi,struct r5conf *conf)
46031f9a
RBJ		 5134{
5135 unsigned long flags;
5136
5137 spin_lock_irqsave(&conf->device_lock, flags);
5138
5139 bi->bi_next = conf->retry_read_aligned_list;
5140 conf->retry_read_aligned_list = bi;
5141
5142 spin_unlock_irqrestore(&conf->device_lock, flags);
5143 md_wakeup_thread(conf->mddev->thread);
5144}
5145
0472a42b
N		 5146static struct bio *remove_bio_from_retry(struct r5conf *conf,
5147 unsigned int *offset)
46031f9a
RBJ		 5148{
5149 struct bio *bi;
5150
5151 bi = conf->retry_read_aligned;
5152 if (bi) {
0472a42b 5153 *offset = conf->retry_read_offset;
46031f9a
RBJ		 5154 conf->retry_read_aligned = NULL;
5155 return bi;
5156 }
5157 bi = conf->retry_read_aligned_list;
5158 if(bi) {
387bb173 5159 conf->retry_read_aligned_list = bi->bi_next;
46031f9a 5160 bi->bi_next = NULL;
0472a42b 5161 *offset = 0;
46031f9a
RBJ		 5162 }
5163
5164 return bi;
5165}
5166
f679623f
RBJ		 5167/*
5168 * The "raid5_align_endio" should check if the read succeeded and if it
5169 * did, call bio_endio on the original bio (having bio_put the new bio
5170 * first).
5171 * If the read failed..
5172 */
4246a0b6 5173static void raid5_align_endio(struct bio *bi)
f679623f
RBJ		 5174{
5175 struct bio* raid_bi = bi->bi_private;
fd01b88c 5176 struct mddev *mddev;
d1688a6d 5177 struct r5conf *conf;
3cb03002 5178 struct md_rdev *rdev;
4e4cbee9 5179 blk_status_t error = bi->bi_status;
46031f9a 5180
f679623f 5181 bio_put(bi);
46031f9a 5182
46031f9a
RBJ		 5183 rdev = (void*)raid_bi->bi_next;
5184 raid_bi->bi_next = NULL;
2b7f2228
N		 5185 mddev = rdev->mddev;
5186 conf = mddev->private;
46031f9a
RBJ		 5187
5188 rdev_dec_pending(rdev, conf->mddev);
5189
9b81c842 5190 if (!error) {
4246a0b6 5191 bio_endio(raid_bi);
46031f9a 5192 if (atomic_dec_and_test(&conf->active_aligned_reads))
b1b46486 5193 wake_up(&conf->wait_for_quiescent);
6712ecf8 5194 return;
46031f9a
RBJ		 5195 }
5196
45b4233c 5197 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
46031f9a
RBJ		 5198
5199 add_bio_to_retry(raid_bi, conf);
f679623f
RBJ		 5200}
5201
7ef6b12a 5202static int raid5_read_one_chunk(struct mddev *mddev, struct bio *raid_bio)
f679623f 5203{
d1688a6d 5204 struct r5conf *conf = mddev->private;
8553fe7e 5205 int dd_idx;
f679623f 5206 struct bio* align_bi;
3cb03002 5207 struct md_rdev *rdev;
671488cc 5208 sector_t end_sector;
f679623f
RBJ		 5209
5210 if (!in_chunk_boundary(mddev, raid_bio)) {
7ef6b12a 5211 pr_debug("%s: non aligned\n", __func__);
f679623f
RBJ		 5212 return 0;
5213 }
5214 /*
d7a10308 5215 * use bio_clone_fast to make a copy of the bio
f679623f 5216 */
afeee514 5217 align_bi = bio_clone_fast(raid_bio, GFP_NOIO, &mddev->bio_set);
f679623f
RBJ		 5218 if (!align_bi)
5219 return 0;
5220 /*
5221 * set bi_end_io to a new function, and set bi_private to the
5222 * original bio.
5223 */
5224 align_bi->bi_end_io = raid5_align_endio;
5225 align_bi->bi_private = raid_bio;
5226 /*
5227 * compute position
5228 */
4f024f37
KO		 5229 align_bi->bi_iter.bi_sector =
5230 raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector,
5231 0, &dd_idx, NULL);
f679623f 5232
f73a1c7d 5233 end_sector = bio_end_sector(align_bi);
f679623f 5234 rcu_read_lock();
671488cc
N		 5235 rdev = rcu_dereference(conf->disks[dd_idx].replacement);
5236 if (!rdev || test_bit(Faulty, &rdev->flags) ||
5237 rdev->recovery_offset < end_sector) {
5238 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
5239 if (rdev &&
5240 (test_bit(Faulty, &rdev->flags) ||
5241 !(test_bit(In_sync, &rdev->flags) ||
5242 rdev->recovery_offset >= end_sector)))
5243 rdev = NULL;
5244 }
03b047f4
SL		 5245
5246 if (r5c_big_stripe_cached(conf, align_bi->bi_iter.bi_sector)) {
5247 rcu_read_unlock();
5248 bio_put(align_bi);
5249 return 0;
5250 }
5251
671488cc 5252 if (rdev) {
31c176ec
N		 5253 sector_t first_bad;
5254 int bad_sectors;
5255
f679623f
RBJ		 5256 atomic_inc(&rdev->nr_pending);
5257 rcu_read_unlock();
46031f9a 5258 raid_bio->bi_next = (void*)rdev;
74d46992 5259 bio_set_dev(align_bi, rdev->bdev);
b7c44ed9 5260 bio_clear_flag(align_bi, BIO_SEG_VALID);
46031f9a 5261
7140aafc 5262 if (is_badblock(rdev, align_bi->bi_iter.bi_sector,
4f024f37 5263 bio_sectors(align_bi),
31c176ec 5264 &first_bad, &bad_sectors)) {
387bb173
NB		 5265 bio_put(align_bi);
5266 rdev_dec_pending(rdev, mddev);
5267 return 0;
5268 }
5269
6c0544e2 5270 /* No reshape active, so we can trust rdev->data_offset */
4f024f37 5271 align_bi->bi_iter.bi_sector += rdev->data_offset;
6c0544e2 5272
46031f9a 5273 spin_lock_irq(&conf->device_lock);
b1b46486 5274 wait_event_lock_irq(conf->wait_for_quiescent,
46031f9a 5275 conf->quiesce == 0,
eed8c02e 5276 conf->device_lock);
46031f9a
RBJ		 5277 atomic_inc(&conf->active_aligned_reads);
5278 spin_unlock_irq(&conf->device_lock);
5279
e3620a3a 5280 if (mddev->gendisk)
74d46992 5281 trace_block_bio_remap(align_bi->bi_disk->queue,
e3620a3a 5282 align_bi, disk_devt(mddev->gendisk),
4f024f37 5283 raid_bio->bi_iter.bi_sector);
f679623f
RBJ		 5284 generic_make_request(align_bi);
5285 return 1;
5286 } else {
5287 rcu_read_unlock();
46031f9a 5288 bio_put(align_bi);
f679623f
RBJ		 5289 return 0;
5290 }
5291}
5292
7ef6b12a
ML		 5293static struct bio *chunk_aligned_read(struct mddev *mddev, struct bio *raid_bio)
5294{
5295 struct bio *split;
dd7a8f5d
N		 5296 sector_t sector = raid_bio->bi_iter.bi_sector;
5297 unsigned chunk_sects = mddev->chunk_sectors;
5298 unsigned sectors = chunk_sects - (sector & (chunk_sects-1));
7ef6b12a 5299
dd7a8f5d
N		 5300 if (sectors < bio_sectors(raid_bio)) {
5301 struct r5conf *conf = mddev->private;
afeee514 5302 split = bio_split(raid_bio, sectors, GFP_NOIO, &conf->bio_split);
dd7a8f5d
N		 5303 bio_chain(split, raid_bio);
5304 generic_make_request(raid_bio);
5305 raid_bio = split;
5306 }
7ef6b12a 5307
dd7a8f5d
N		 5308 if (!raid5_read_one_chunk(mddev, raid_bio))
5309 return raid_bio;
7ef6b12a
ML		 5310
5311 return NULL;
5312}
5313
8b3e6cdc
DW		 5314/* __get_priority_stripe - get the next stripe to process
5315 *
5316 * Full stripe writes are allowed to pass preread active stripes up until
5317 * the bypass_threshold is exceeded. In general the bypass_count
5318 * increments when the handle_list is handled before the hold_list; however, it
5319 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
5320 * stripe with in flight i/o. The bypass_count will be reset when the
5321 * head of the hold_list has changed, i.e. the head was promoted to the
5322 * handle_list.
5323 */
851c30c9 5324static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group)
8b3e6cdc 5325{
535ae4eb 5326 struct stripe_head *sh, *tmp;
851c30c9 5327 struct list_head *handle_list = NULL;
535ae4eb 5328 struct r5worker_group *wg;
70d466f7
SL		 5329 bool second_try = !r5c_is_writeback(conf->log) &&
5330 !r5l_log_disk_error(conf);
5331 bool try_loprio = test_bit(R5C_LOG_TIGHT, &conf->cache_state) ||
5332 r5l_log_disk_error(conf);
851c30c9 5333
535ae4eb
SL		 5334again:
5335 wg = NULL;
5336 sh = NULL;
851c30c9 5337 if (conf->worker_cnt_per_group == 0) {
535ae4eb
SL		 5338 handle_list = try_loprio ? &conf->loprio_list :
5339 &conf->handle_list;
851c30c9 5340 } else if (group != ANY_GROUP) {
535ae4eb
SL		 5341 handle_list = try_loprio ? &conf->worker_groups[group].loprio_list :
5342 &conf->worker_groups[group].handle_list;
bfc90cb0 5343 wg = &conf->worker_groups[group];
851c30c9
SL		 5344 } else {
5345 int i;
5346 for (i = 0; i < conf->group_cnt; i++) {
535ae4eb
SL		 5347 handle_list = try_loprio ? &conf->worker_groups[i].loprio_list :
5348 &conf->worker_groups[i].handle_list;
bfc90cb0 5349 wg = &conf->worker_groups[i];
851c30c9
SL		 5350 if (!list_empty(handle_list))
5351 break;
5352 }
5353 }
8b3e6cdc
DW		 5354
5355 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
5356 __func__,
851c30c9 5357 list_empty(handle_list) ? "empty" : "busy",
8b3e6cdc
DW		 5358 list_empty(&conf->hold_list) ? "empty" : "busy",
5359 atomic_read(&conf->pending_full_writes), conf->bypass_count);
5360
851c30c9
SL		 5361 if (!list_empty(handle_list)) {
5362 sh = list_entry(handle_list->next, typeof(*sh), lru);
8b3e6cdc
DW		 5363
5364 if (list_empty(&conf->hold_list))
5365 conf->bypass_count = 0;
5366 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
5367 if (conf->hold_list.next == conf->last_hold)
5368 conf->bypass_count++;
5369 else {
5370 conf->last_hold = conf->hold_list.next;
5371 conf->bypass_count -= conf->bypass_threshold;
5372 if (conf->bypass_count < 0)
5373 conf->bypass_count = 0;
5374 }
5375 }
5376 } else if (!list_empty(&conf->hold_list) &&
5377 ((conf->bypass_threshold &&
5378 conf->bypass_count > conf->bypass_threshold) ||
5379 atomic_read(&conf->pending_full_writes) == 0)) {
851c30c9
SL		 5380
5381 list_for_each_entry(tmp, &conf->hold_list, lru) {
5382 if (conf->worker_cnt_per_group == 0 ||
5383 group == ANY_GROUP ||
5384 !cpu_online(tmp->cpu) ||
5385 cpu_to_group(tmp->cpu) == group) {
5386 sh = tmp;
5387 break;
5388 }
5389 }
5390
5391 if (sh) {
5392 conf->bypass_count -= conf->bypass_threshold;
5393 if (conf->bypass_count < 0)
5394 conf->bypass_count = 0;
5395 }
bfc90cb0 5396 wg = NULL;
851c30c9
SL		 5397 }
5398
535ae4eb
SL		 5399 if (!sh) {
5400 if (second_try)
5401 return NULL;
5402 second_try = true;
5403 try_loprio = !try_loprio;
5404 goto again;
5405 }
8b3e6cdc 5406
bfc90cb0
SL		 5407 if (wg) {
5408 wg->stripes_cnt--;
5409 sh->group = NULL;
5410 }
8b3e6cdc 5411 list_del_init(&sh->lru);
c7a6d35e 5412 BUG_ON(atomic_inc_return(&sh->count) != 1);
8b3e6cdc
DW		 5413 return sh;
5414}
f679623f 5415
8811b596
SL		 5416struct raid5_plug_cb {
5417 struct blk_plug_cb cb;
5418 struct list_head list;
566c09c5 5419 struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
8811b596
SL		 5420};
5421
5422static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule)
5423{
5424 struct raid5_plug_cb *cb = container_of(
5425 blk_cb, struct raid5_plug_cb, cb);
5426 struct stripe_head *sh;
5427 struct mddev *mddev = cb->cb.data;
5428 struct r5conf *conf = mddev->private;
a9add5d9 5429 int cnt = 0;
566c09c5 5430 int hash;
8811b596
SL		 5431
5432 if (cb->list.next && !list_empty(&cb->list)) {
5433 spin_lock_irq(&conf->device_lock);
5434 while (!list_empty(&cb->list)) {
5435 sh = list_first_entry(&cb->list, struct stripe_head, lru);
5436 list_del_init(&sh->lru);
5437 /*
5438 * avoid race release_stripe_plug() sees
5439 * STRIPE_ON_UNPLUG_LIST clear but the stripe
5440 * is still in our list
5441 */
4e857c58 5442 smp_mb__before_atomic();
8811b596 5443 clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state);
773ca82f
SL		 5444 /*
5445 * STRIPE_ON_RELEASE_LIST could be set here. In that
5446 * case, the count is always > 1 here
5447 */
566c09c5
SL		 5448 hash = sh->hash_lock_index;
5449 __release_stripe(conf, sh, &cb->temp_inactive_list[hash]);
a9add5d9 5450 cnt++;
8811b596
SL		 5451 }
5452 spin_unlock_irq(&conf->device_lock);
5453 }
566c09c5
SL		 5454 release_inactive_stripe_list(conf, cb->temp_inactive_list,
5455 NR_STRIPE_HASH_LOCKS);
e3620a3a
JB		 5456 if (mddev->queue)
5457 trace_block_unplug(mddev->queue, cnt, !from_schedule);
8811b596
SL		 5458 kfree(cb);
5459}
5460
5461static void release_stripe_plug(struct mddev *mddev,
5462 struct stripe_head *sh)
5463{
5464 struct blk_plug_cb *blk_cb = blk_check_plugged(
5465 raid5_unplug, mddev,
5466 sizeof(struct raid5_plug_cb));
5467 struct raid5_plug_cb *cb;
5468
5469 if (!blk_cb) {
6d036f7d 5470 raid5_release_stripe(sh);
8811b596
SL		 5471 return;
5472 }
5473
5474 cb = container_of(blk_cb, struct raid5_plug_cb, cb);
5475
566c09c5
SL		 5476 if (cb->list.next == NULL) {
5477 int i;
8811b596 5478 INIT_LIST_HEAD(&cb->list);
566c09c5
SL		 5479 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5480 INIT_LIST_HEAD(cb->temp_inactive_list + i);
5481 }
8811b596
SL		 5482
5483 if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state))
5484 list_add_tail(&sh->lru, &cb->list);
5485 else
6d036f7d 5486 raid5_release_stripe(sh);
8811b596
SL		 5487}
5488
620125f2
SL		 5489static void make_discard_request(struct mddev *mddev, struct bio *bi)
5490{
5491 struct r5conf *conf = mddev->private;
5492 sector_t logical_sector, last_sector;
5493 struct stripe_head *sh;
620125f2
SL		 5494 int stripe_sectors;
5495
5496 if (mddev->reshape_position != MaxSector)
5497 /* Skip discard while reshape is happening */
5498 return;
5499
4f024f37
KO		 5500 logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1);
5501 last_sector = bi->bi_iter.bi_sector + (bi->bi_iter.bi_size>>9);
620125f2
SL		 5502
5503 bi->bi_next = NULL;
620125f2
SL		 5504
5505 stripe_sectors = conf->chunk_sectors *
5506 (conf->raid_disks - conf->max_degraded);
5507 logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector,
5508 stripe_sectors);
5509 sector_div(last_sector, stripe_sectors);
5510
5511 logical_sector *= conf->chunk_sectors;
5512 last_sector *= conf->chunk_sectors;
5513
5514 for (; logical_sector < last_sector;
5515 logical_sector += STRIPE_SECTORS) {
5516 DEFINE_WAIT(w);
5517 int d;
5518 again:
6d036f7d 5519 sh = raid5_get_active_stripe(conf, logical_sector, 0, 0, 0);
620125f2
SL		 5520 prepare_to_wait(&conf->wait_for_overlap, &w,
5521 TASK_UNINTERRUPTIBLE);
f8dfcffd
N		 5522 set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
5523 if (test_bit(STRIPE_SYNCING, &sh->state)) {
6d036f7d 5524 raid5_release_stripe(sh);
f8dfcffd
N		 5525 schedule();
5526 goto again;
5527 }
5528 clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
620125f2
SL		 5529 spin_lock_irq(&sh->stripe_lock);
5530 for (d = 0; d < conf->raid_disks; d++) {
5531 if (d == sh->pd_idx || d == sh->qd_idx)
5532 continue;
5533 if (sh->dev[d].towrite || sh->dev[d].toread) {
5534 set_bit(R5_Overlap, &sh->dev[d].flags);
5535 spin_unlock_irq(&sh->stripe_lock);
6d036f7d 5536 raid5_release_stripe(sh);
620125f2
SL		 5537 schedule();
5538 goto again;
5539 }
5540 }
f8dfcffd 5541 set_bit(STRIPE_DISCARD, &sh->state);
620125f2 5542 finish_wait(&conf->wait_for_overlap, &w);
7a87f434 5543 sh->overwrite_disks = 0;
620125f2
SL		 5544 for (d = 0; d < conf->raid_disks; d++) {
5545 if (d == sh->pd_idx || d == sh->qd_idx)
5546 continue;
5547 sh->dev[d].towrite = bi;
5548 set_bit(R5_OVERWRITE, &sh->dev[d].flags);
016c76ac 5549 bio_inc_remaining(bi);
49728050 5550 md_write_inc(mddev, bi);
7a87f434 5551 sh->overwrite_disks++;
620125f2
SL		 5552 }
5553 spin_unlock_irq(&sh->stripe_lock);
5554 if (conf->mddev->bitmap) {
5555 for (d = 0;
5556 d < conf->raid_disks - conf->max_degraded;
5557 d++)
e64e4018
AS		 5558 md_bitmap_startwrite(mddev->bitmap,
5559 sh->sector,
5560 STRIPE_SECTORS,
5561 0);
620125f2
SL		 5562 sh->bm_seq = conf->seq_flush + 1;
5563 set_bit(STRIPE_BIT_DELAY, &sh->state);
5564 }
5565
5566 set_bit(STRIPE_HANDLE, &sh->state);
5567 clear_bit(STRIPE_DELAYED, &sh->state);
5568 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
5569 atomic_inc(&conf->preread_active_stripes);
5570 release_stripe_plug(mddev, sh);
5571 }
5572
016c76ac 5573 bio_endio(bi);
620125f2
SL		 5574}
5575
cc27b0c7 5576static bool raid5_make_request(struct mddev *mddev, struct bio * bi)
1da177e4 5577{
d1688a6d 5578 struct r5conf *conf = mddev->private;
911d4ee8 5579 int dd_idx;
1da177e4
LT		 5580 sector_t new_sector;
5581 sector_t logical_sector, last_sector;
5582 struct stripe_head *sh;
a362357b 5583 const int rw = bio_data_dir(bi);
27c0f68f
SL		 5584 DEFINE_WAIT(w);
5585 bool do_prepare;
3bddb7f8 5586 bool do_flush = false;
1da177e4 5587
1eff9d32 5588 if (unlikely(bi->bi_opf & REQ_PREFLUSH)) {
1532d9e8 5589 int ret = log_handle_flush_request(conf, bi);
828cbe98
SL		 5590
5591 if (ret == 0)
cc27b0c7 5592 return true;
828cbe98
SL		 5593 if (ret == -ENODEV) {
5594 md_flush_request(mddev, bi);
cc27b0c7 5595 return true;
828cbe98
SL		 5596 }
5597 /* ret == -EAGAIN, fallback */
3bddb7f8
SL		 5598 /*
5599 * if r5l_handle_flush_request() didn't clear REQ_PREFLUSH,
5600 * we need to flush journal device
5601 */
5602 do_flush = bi->bi_opf & REQ_PREFLUSH;
e5dcdd80
N		 5603 }
5604
cc27b0c7
N		 5605 if (!md_write_start(mddev, bi))
5606 return false;
9ffc8f7c
EM		 5607 /*
5608 * If array is degraded, better not do chunk aligned read because
5609 * later we might have to read it again in order to reconstruct
5610 * data on failed drives.
5611 */
5612 if (rw == READ && mddev->degraded == 0 &&
7ef6b12a
ML		 5613 mddev->reshape_position == MaxSector) {
5614 bi = chunk_aligned_read(mddev, bi);
5615 if (!bi)
cc27b0c7 5616 return true;
7ef6b12a 5617 }
52488615 5618
796a5cf0 5619 if (unlikely(bio_op(bi) == REQ_OP_DISCARD)) {
620125f2 5620 make_discard_request(mddev, bi);
cc27b0c7
N		 5621 md_write_end(mddev);
5622 return true;
620125f2
SL		 5623 }
5624
4f024f37 5625 logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1);
f73a1c7d 5626 last_sector = bio_end_sector(bi);
1da177e4 5627 bi->bi_next = NULL;
06d91a5f 5628
27c0f68f 5629 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
1da177e4 5630 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
b5663ba4 5631 int previous;
c46501b2 5632 int seq;
b578d55f 5633
27c0f68f 5634 do_prepare = false;
7ecaa1e6 5635 retry:
c46501b2 5636 seq = read_seqcount_begin(&conf->gen_lock);
b5663ba4 5637 previous = 0;
27c0f68f
SL		 5638 if (do_prepare)
5639 prepare_to_wait(&conf->wait_for_overlap, &w,
5640 TASK_UNINTERRUPTIBLE);
b0f9ec04 5641 if (unlikely(conf->reshape_progress != MaxSector)) {
fef9c61f 5642 /* spinlock is needed as reshape_progress may be
df8e7f76
N		 5643 * 64bit on a 32bit platform, and so it might be
5644 * possible to see a half-updated value
aeb878b0 5645 * Of course reshape_progress could change after
df8e7f76
N		 5646 * the lock is dropped, so once we get a reference
5647 * to the stripe that we think it is, we will have
5648 * to check again.
5649 */
7ecaa1e6 5650 spin_lock_irq(&conf->device_lock);
2c810cdd 5651 if (mddev->reshape_backwards
fef9c61f
N		 5652 ? logical_sector < conf->reshape_progress
5653 : logical_sector >= conf->reshape_progress) {
b5663ba4
N		 5654 previous = 1;
5655 } else {
2c810cdd 5656 if (mddev->reshape_backwards
fef9c61f
N		 5657 ? logical_sector < conf->reshape_safe
5658 : logical_sector >= conf->reshape_safe) {
b578d55f
N		 5659 spin_unlock_irq(&conf->device_lock);
5660 schedule();
27c0f68f 5661 do_prepare = true;
b578d55f
N		 5662 goto retry;
5663 }
5664 }
7ecaa1e6
N		 5665 spin_unlock_irq(&conf->device_lock);
5666 }
16a53ecc 5667
112bf897
N		 5668 new_sector = raid5_compute_sector(conf, logical_sector,
5669 previous,
911d4ee8 5670 &dd_idx, NULL);
849674e4 5671 pr_debug("raid456: raid5_make_request, sector %llu logical %llu\n",
c46501b2 5672 (unsigned long long)new_sector,
1da177e4
LT		 5673 (unsigned long long)logical_sector);
5674
6d036f7d 5675 sh = raid5_get_active_stripe(conf, new_sector, previous,
1eff9d32 5676 (bi->bi_opf & REQ_RAHEAD), 0);
1da177e4 5677 if (sh) {
b0f9ec04 5678 if (unlikely(previous)) {
7ecaa1e6 5679 /* expansion might have moved on while waiting for a
df8e7f76
N		 5680 * stripe, so we must do the range check again.
5681 * Expansion could still move past after this
5682 * test, but as we are holding a reference to
5683 * 'sh', we know that if that happens,
5684 * STRIPE_EXPANDING will get set and the expansion
5685 * won't proceed until we finish with the stripe.
7ecaa1e6
N		 5686 */
5687 int must_retry = 0;
5688 spin_lock_irq(&conf->device_lock);
2c810cdd 5689 if (mddev->reshape_backwards
b0f9ec04
N		 5690 ? logical_sector >= conf->reshape_progress
5691 : logical_sector < conf->reshape_progress)
7ecaa1e6
N		 5692 /* mismatch, need to try again */
5693 must_retry = 1;
5694 spin_unlock_irq(&conf->device_lock);
5695 if (must_retry) {
6d036f7d 5696 raid5_release_stripe(sh);
7a3ab908 5697 schedule();
27c0f68f 5698 do_prepare = true;
7ecaa1e6
N		 5699 goto retry;
5700 }
5701 }
c46501b2
N		 5702 if (read_seqcount_retry(&conf->gen_lock, seq)) {
5703 /* Might have got the wrong stripe_head
5704 * by accident
5705 */
6d036f7d 5706 raid5_release_stripe(sh);
c46501b2
N		 5707 goto retry;
5708 }
e62e58a5 5709
7ecaa1e6 5710 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
da41ba65 5711 !add_stripe_bio(sh, bi, dd_idx, rw, previous)) {
7ecaa1e6
N		 5712 /* Stripe is busy expanding or
5713 * add failed due to overlap. Flush everything
1da177e4
LT		 5714 * and wait a while
5715 */
482c0834 5716 md_wakeup_thread(mddev->thread);
6d036f7d 5717 raid5_release_stripe(sh);
1da177e4 5718 schedule();
27c0f68f 5719 do_prepare = true;
1da177e4
LT		 5720 goto retry;
5721 }
3bddb7f8
SL		 5722 if (do_flush) {
5723 set_bit(STRIPE_R5C_PREFLUSH, &sh->state);
5724 /* we only need flush for one stripe */
5725 do_flush = false;
5726 }
5727
6ed3003c
N		 5728 set_bit(STRIPE_HANDLE, &sh->state);
5729 clear_bit(STRIPE_DELAYED, &sh->state);
59fc630b 5730 if ((!sh->batch_head || sh == sh->batch_head) &&
1eff9d32 5731 (bi->bi_opf & REQ_SYNC) &&
729a1866
N		 5732 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
5733 atomic_inc(&conf->preread_active_stripes);
8811b596 5734 release_stripe_plug(mddev, sh);
1da177e4
LT		 5735 } else {
5736 /* cannot get stripe for read-ahead, just give-up */
4e4cbee9 5737 bi->bi_status = BLK_STS_IOERR;
1da177e4
LT		 5738 break;
5739 }
1da177e4 5740 }
27c0f68f 5741 finish_wait(&conf->wait_for_overlap, &w);
7c13edc8 5742
49728050
N		 5743 if (rw == WRITE)
5744 md_write_end(mddev);
016c76ac 5745 bio_endio(bi);
cc27b0c7 5746 return true;
1da177e4
LT		 5747}
5748
fd01b88c 5749static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks);
b522adcd 5750
fd01b88c 5751static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped)
1da177e4 5752{
52c03291
N		 5753 /* reshaping is quite different to recovery/resync so it is
5754 * handled quite separately ... here.
5755 *
5756 * On each call to sync_request, we gather one chunk worth of
5757 * destination stripes and flag them as expanding.
5758 * Then we find all the source stripes and request reads.
5759 * As the reads complete, handle_stripe will copy the data
5760 * into the destination stripe and release that stripe.
5761 */
d1688a6d 5762 struct r5conf *conf = mddev->private;
1da177e4 5763 struct stripe_head *sh;
db0505d3 5764 struct md_rdev *rdev;
ccfcc3c1 5765 sector_t first_sector, last_sector;
f416885e
N		 5766 int raid_disks = conf->previous_raid_disks;
5767 int data_disks = raid_disks - conf->max_degraded;
5768 int new_data_disks = conf->raid_disks - conf->max_degraded;
52c03291
N		 5769 int i;
5770 int dd_idx;
c8f517c4 5771 sector_t writepos, readpos, safepos;
ec32a2bd 5772 sector_t stripe_addr;
7a661381 5773 int reshape_sectors;
ab69ae12 5774 struct list_head stripes;
92140480 5775 sector_t retn;
52c03291 5776
fef9c61f
N		 5777 if (sector_nr == 0) {
5778 /* If restarting in the middle, skip the initial sectors */
2c810cdd 5779 if (mddev->reshape_backwards &&
fef9c61f
N		 5780 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
5781 sector_nr = raid5_size(mddev, 0, 0)
5782 - conf->reshape_progress;
6cbd8148
N		 5783 } else if (mddev->reshape_backwards &&
5784 conf->reshape_progress == MaxSector) {
5785 /* shouldn't happen, but just in case, finish up.*/
5786 sector_nr = MaxSector;
2c810cdd 5787 } else if (!mddev->reshape_backwards &&
fef9c61f
N		 5788 conf->reshape_progress > 0)
5789 sector_nr = conf->reshape_progress;
f416885e 5790 sector_div(sector_nr, new_data_disks);
fef9c61f 5791 if (sector_nr) {
8dee7211
N		 5792 mddev->curr_resync_completed = sector_nr;
5793 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
fef9c61f 5794 *skipped = 1;
92140480
N		 5795 retn = sector_nr;
5796 goto finish;
fef9c61f 5797 }
52c03291
N		 5798 }
5799
7a661381
N		 5800 /* We need to process a full chunk at a time.
5801 * If old and new chunk sizes differ, we need to process the
5802 * largest of these
5803 */
3cb5edf4
N		 5804
5805 reshape_sectors = max(conf->chunk_sectors, conf->prev_chunk_sectors);
7a661381 5806
b5254dd5
N		 5807 /* We update the metadata at least every 10 seconds, or when
5808 * the data about to be copied would over-write the source of
5809 * the data at the front of the range. i.e. one new_stripe
5810 * along from reshape_progress new_maps to after where
5811 * reshape_safe old_maps to
52c03291 5812 */
fef9c61f 5813 writepos = conf->reshape_progress;
f416885e 5814 sector_div(writepos, new_data_disks);
c8f517c4
N		 5815 readpos = conf->reshape_progress;
5816 sector_div(readpos, data_disks);
fef9c61f 5817 safepos = conf->reshape_safe;
f416885e 5818 sector_div(safepos, data_disks);
2c810cdd 5819 if (mddev->reshape_backwards) {
c74c0d76
N		 5820 BUG_ON(writepos < reshape_sectors);
5821 writepos -= reshape_sectors;
c8f517c4 5822 readpos += reshape_sectors;
7a661381 5823 safepos += reshape_sectors;
fef9c61f 5824 } else {
7a661381 5825 writepos += reshape_sectors;
c74c0d76
N		 5826 /* readpos and safepos are worst-case calculations.
5827 * A negative number is overly pessimistic, and causes
5828 * obvious problems for unsigned storage. So clip to 0.
5829 */
ed37d83e
N		 5830 readpos -= min_t(sector_t, reshape_sectors, readpos);
5831 safepos -= min_t(sector_t, reshape_sectors, safepos);
fef9c61f 5832 }
52c03291 5833
b5254dd5
N		 5834 /* Having calculated the 'writepos' possibly use it
5835 * to set 'stripe_addr' which is where we will write to.
5836 */
5837 if (mddev->reshape_backwards) {
5838 BUG_ON(conf->reshape_progress == 0);
5839 stripe_addr = writepos;
5840 BUG_ON((mddev->dev_sectors &
5841 ~((sector_t)reshape_sectors - 1))
5842 - reshape_sectors - stripe_addr
5843 != sector_nr);
5844 } else {
5845 BUG_ON(writepos != sector_nr + reshape_sectors);
5846 stripe_addr = sector_nr;
5847 }
5848
c8f517c4
N		 5849 /* 'writepos' is the most advanced device address we might write.
5850 * 'readpos' is the least advanced device address we might read.
5851 * 'safepos' is the least address recorded in the metadata as having
5852 * been reshaped.
b5254dd5
N		 5853 * If there is a min_offset_diff, these are adjusted either by
5854 * increasing the safepos/readpos if diff is negative, or
5855 * increasing writepos if diff is positive.
5856 * If 'readpos' is then behind 'writepos', there is no way that we can
c8f517c4
N		 5857 * ensure safety in the face of a crash - that must be done by userspace
5858 * making a backup of the data. So in that case there is no particular
5859 * rush to update metadata.
5860 * Otherwise if 'safepos' is behind 'writepos', then we really need to
5861 * update the metadata to advance 'safepos' to match 'readpos' so that
5862 * we can be safe in the event of a crash.
5863 * So we insist on updating metadata if safepos is behind writepos and
5864 * readpos is beyond writepos.
5865 * In any case, update the metadata every 10 seconds.
5866 * Maybe that number should be configurable, but I'm not sure it is
5867 * worth it.... maybe it could be a multiple of safemode_delay???
5868 */
b5254dd5
N		 5869 if (conf->min_offset_diff < 0) {
5870 safepos += -conf->min_offset_diff;
5871 readpos += -conf->min_offset_diff;
5872 } else
5873 writepos += conf->min_offset_diff;
5874
2c810cdd 5875 if ((mddev->reshape_backwards
c8f517c4
N		 5876 ? (safepos > writepos && readpos < writepos)
5877 : (safepos < writepos && readpos > writepos)) ||
5878 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
52c03291
N		 5879 /* Cannot proceed until we've updated the superblock... */
5880 wait_event(conf->wait_for_overlap,
c91abf5a
N		 5881 atomic_read(&conf->reshape_stripes)==0
5882 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5883 if (atomic_read(&conf->reshape_stripes) != 0)
5884 return 0;
fef9c61f 5885 mddev->reshape_position = conf->reshape_progress;
75d3da43 5886 mddev->curr_resync_completed = sector_nr;
db0505d3
N		 5887 if (!mddev->reshape_backwards)
5888 /* Can update recovery_offset */
5889 rdev_for_each(rdev, mddev)
5890 if (rdev->raid_disk >= 0 &&
5891 !test_bit(Journal, &rdev->flags) &&
5892 !test_bit(In_sync, &rdev->flags) &&
5893 rdev->recovery_offset < sector_nr)
5894 rdev->recovery_offset = sector_nr;
5895
c8f517c4 5896 conf->reshape_checkpoint = jiffies;
2953079c 5897 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
52c03291 5898 md_wakeup_thread(mddev->thread);
2953079c 5899 wait_event(mddev->sb_wait, mddev->sb_flags == 0 ||
c91abf5a
N		 5900 test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5901 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5902 return 0;
52c03291 5903 spin_lock_irq(&conf->device_lock);
fef9c61f 5904 conf->reshape_safe = mddev->reshape_position;
52c03291
N		 5905 spin_unlock_irq(&conf->device_lock);
5906 wake_up(&conf->wait_for_overlap);
acb180b0 5907 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
52c03291
N		 5908 }
5909
ab69ae12 5910 INIT_LIST_HEAD(&stripes);
7a661381 5911 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
52c03291 5912 int j;
a9f326eb 5913 int skipped_disk = 0;
6d036f7d 5914 sh = raid5_get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
52c03291
N		 5915 set_bit(STRIPE_EXPANDING, &sh->state);
5916 atomic_inc(&conf->reshape_stripes);
5917 /* If any of this stripe is beyond the end of the old
5918 * array, then we need to zero those blocks
5919 */
5920 for (j=sh->disks; j--;) {
5921 sector_t s;
5922 if (j == sh->pd_idx)
5923 continue;
f416885e 5924 if (conf->level == 6 &&
d0dabf7e 5925 j == sh->qd_idx)
f416885e 5926 continue;
6d036f7d 5927 s = raid5_compute_blocknr(sh, j, 0);
b522adcd 5928 if (s < raid5_size(mddev, 0, 0)) {
a9f326eb 5929 skipped_disk = 1;
52c03291
N		 5930 continue;
5931 }
5932 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
5933 set_bit(R5_Expanded, &sh->dev[j].flags);
5934 set_bit(R5_UPTODATE, &sh->dev[j].flags);
5935 }
a9f326eb 5936 if (!skipped_disk) {
52c03291
N		 5937 set_bit(STRIPE_EXPAND_READY, &sh->state);
5938 set_bit(STRIPE_HANDLE, &sh->state);
5939 }
ab69ae12 5940 list_add(&sh->lru, &stripes);
52c03291
N		 5941 }
5942 spin_lock_irq(&conf->device_lock);
2c810cdd 5943 if (mddev->reshape_backwards)
7a661381 5944 conf->reshape_progress -= reshape_sectors * new_data_disks;
fef9c61f 5945 else
7a661381 5946 conf->reshape_progress += reshape_sectors * new_data_disks;
52c03291
N		 5947 spin_unlock_irq(&conf->device_lock);
5948 /* Ok, those stripe are ready. We can start scheduling
5949 * reads on the source stripes.
5950 * The source stripes are determined by mapping the first and last
5951 * block on the destination stripes.
5952 */
52c03291 5953 first_sector =
ec32a2bd 5954 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
911d4ee8 5955 1, &dd_idx, NULL);
52c03291 5956 last_sector =
0e6e0271 5957 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
09c9e5fa 5958 * new_data_disks - 1),
911d4ee8 5959 1, &dd_idx, NULL);
58c0fed4
AN		 5960 if (last_sector >= mddev->dev_sectors)
5961 last_sector = mddev->dev_sectors - 1;
52c03291 5962 while (first_sector <= last_sector) {
6d036f7d 5963 sh = raid5_get_active_stripe(conf, first_sector, 1, 0, 1);
52c03291
N		 5964 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
5965 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 5966 raid5_release_stripe(sh);
52c03291
N		 5967 first_sector += STRIPE_SECTORS;
5968 }
ab69ae12
N		 5969 /* Now that the sources are clearly marked, we can release
5970 * the destination stripes
5971 */
5972 while (!list_empty(&stripes)) {
5973 sh = list_entry(stripes.next, struct stripe_head, lru);
5974 list_del_init(&sh->lru);
6d036f7d 5975 raid5_release_stripe(sh);
ab69ae12 5976 }
c6207277
N		 5977 /* If this takes us to the resync_max point where we have to pause,
5978 * then we need to write out the superblock.
5979 */
7a661381 5980 sector_nr += reshape_sectors;
92140480
N		 5981 retn = reshape_sectors;
5982finish:
c5e19d90
N		 5983 if (mddev->curr_resync_completed > mddev->resync_max ||
5984 (sector_nr - mddev->curr_resync_completed) * 2
c03f6a19 5985 >= mddev->resync_max - mddev->curr_resync_completed) {
c6207277
N		 5986 /* Cannot proceed until we've updated the superblock... */
5987 wait_event(conf->wait_for_overlap,
c91abf5a
N		 5988 atomic_read(&conf->reshape_stripes) == 0
5989 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5990 if (atomic_read(&conf->reshape_stripes) != 0)
5991 goto ret;
fef9c61f 5992 mddev->reshape_position = conf->reshape_progress;
75d3da43 5993 mddev->curr_resync_completed = sector_nr;
db0505d3
N		 5994 if (!mddev->reshape_backwards)
5995 /* Can update recovery_offset */
5996 rdev_for_each(rdev, mddev)
5997 if (rdev->raid_disk >= 0 &&
5998 !test_bit(Journal, &rdev->flags) &&
5999 !test_bit(In_sync, &rdev->flags) &&
6000 rdev->recovery_offset < sector_nr)
6001 rdev->recovery_offset = sector_nr;
c8f517c4 6002 conf->reshape_checkpoint = jiffies;
2953079c 6003 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
c6207277
N		 6004 md_wakeup_thread(mddev->thread);
6005 wait_event(mddev->sb_wait,
2953079c 6006 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)
c91abf5a
N		 6007 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
6008 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6009 goto ret;
c6207277 6010 spin_lock_irq(&conf->device_lock);
fef9c61f 6011 conf->reshape_safe = mddev->reshape_position;
c6207277
N		 6012 spin_unlock_irq(&conf->device_lock);
6013 wake_up(&conf->wait_for_overlap);
acb180b0 6014 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
c6207277 6015 }
c91abf5a 6016ret:
92140480 6017 return retn;
52c03291
N		 6018}
6019
849674e4
SL		 6020static inline sector_t raid5_sync_request(struct mddev *mddev, sector_t sector_nr,
6021 int *skipped)
52c03291 6022{
d1688a6d 6023 struct r5conf *conf = mddev->private;
52c03291 6024 struct stripe_head *sh;
58c0fed4 6025 sector_t max_sector = mddev->dev_sectors;
57dab0bd 6026 sector_t sync_blocks;
16a53ecc
N		 6027 int still_degraded = 0;
6028 int i;
1da177e4 6029
72626685 6030 if (sector_nr >= max_sector) {
1da177e4 6031 /* just being told to finish up .. nothing much to do */
cea9c228 6032
29269553
N		 6033 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
6034 end_reshape(conf);
6035 return 0;
6036 }
72626685
N		 6037
6038 if (mddev->curr_resync < max_sector) /* aborted */
e64e4018
AS		 6039 md_bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
6040 &sync_blocks, 1);
16a53ecc 6041 else /* completed sync */
72626685 6042 conf->fullsync = 0;
e64e4018 6043 md_bitmap_close_sync(mddev->bitmap);
72626685 6044
1da177e4
LT		 6045 return 0;
6046 }
ccfcc3c1 6047
64bd660b
N		 6048 /* Allow raid5_quiesce to complete */
6049 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
6050
52c03291
N		 6051 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6052 return reshape_request(mddev, sector_nr, skipped);
f6705578 6053
c6207277
N		 6054 /* No need to check resync_max as we never do more than one
6055 * stripe, and as resync_max will always be on a chunk boundary,
6056 * if the check in md_do_sync didn't fire, there is no chance
6057 * of overstepping resync_max here
6058 */
6059
16a53ecc 6060 /* if there is too many failed drives and we are trying
1da177e4
LT		 6061 * to resync, then assert that we are finished, because there is
6062 * nothing we can do.
6063 */
3285edf1 6064 if (mddev->degraded >= conf->max_degraded &&
16a53ecc 6065 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
58c0fed4 6066 sector_t rv = mddev->dev_sectors - sector_nr;
57afd89f 6067 *skipped = 1;
1da177e4
LT		 6068 return rv;
6069 }
6f608040 6070 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
6071 !conf->fullsync &&
e64e4018 6072 !md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
6f608040 6073 sync_blocks >= STRIPE_SECTORS) {
72626685
N		 6074 /* we can skip this block, and probably more */
6075 sync_blocks /= STRIPE_SECTORS;
6076 *skipped = 1;
6077 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
6078 }
1da177e4 6079
e64e4018 6080 md_bitmap_cond_end_sync(mddev->bitmap, sector_nr, false);
b47490c9 6081
6d036f7d 6082 sh = raid5_get_active_stripe(conf, sector_nr, 0, 1, 0);
1da177e4 6083 if (sh == NULL) {
6d036f7d 6084 sh = raid5_get_active_stripe(conf, sector_nr, 0, 0, 0);
1da177e4 6085 /* make sure we don't swamp the stripe cache if someone else
16a53ecc 6086 * is trying to get access
1da177e4 6087 */
66c006a5 6088 schedule_timeout_uninterruptible(1);
1da177e4 6089 }
16a53ecc 6090 /* Need to check if array will still be degraded after recovery/resync
16d9cfab
EM		 6091 * Note in case of > 1 drive failures it's possible we're rebuilding
6092 * one drive while leaving another faulty drive in array.
16a53ecc 6093 */
16d9cfab
EM		 6094 rcu_read_lock();
6095 for (i = 0; i < conf->raid_disks; i++) {
6aa7de05 6096 struct md_rdev *rdev = READ_ONCE(conf->disks[i].rdev);
16d9cfab
EM		 6097
6098 if (rdev == NULL || test_bit(Faulty, &rdev->flags))
16a53ecc 6099 still_degraded = 1;
16d9cfab
EM		 6100 }
6101 rcu_read_unlock();
16a53ecc 6102
e64e4018 6103 md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
16a53ecc 6104
83206d66 6105 set_bit(STRIPE_SYNC_REQUESTED, &sh->state);
053f5b65 6106 set_bit(STRIPE_HANDLE, &sh->state);
1da177e4 6107
6d036f7d 6108 raid5_release_stripe(sh);
1da177e4
LT		 6109
6110 return STRIPE_SECTORS;
6111}
6112
0472a42b
N		 6113static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio,
6114 unsigned int offset)
46031f9a
RBJ		 6115{
6116 /* We may not be able to submit a whole bio at once as there
6117 * may not be enough stripe_heads available.
6118 * We cannot pre-allocate enough stripe_heads as we may need
6119 * more than exist in the cache (if we allow ever large chunks).
6120 * So we do one stripe head at a time and record in
6121 * ->bi_hw_segments how many have been done.
6122 *
6123 * We *know* that this entire raid_bio is in one chunk, so
6124 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
6125 */
6126 struct stripe_head *sh;
911d4ee8 6127 int dd_idx;
46031f9a
RBJ		 6128 sector_t sector, logical_sector, last_sector;
6129 int scnt = 0;
46031f9a
RBJ		 6130 int handled = 0;
6131
4f024f37
KO		 6132 logical_sector = raid_bio->bi_iter.bi_sector &
6133 ~((sector_t)STRIPE_SECTORS-1);
112bf897 6134 sector = raid5_compute_sector(conf, logical_sector,
911d4ee8 6135 0, &dd_idx, NULL);
f73a1c7d 6136 last_sector = bio_end_sector(raid_bio);
46031f9a
RBJ		 6137
6138 for (; logical_sector < last_sector;
387bb173
NB		 6139 logical_sector += STRIPE_SECTORS,
6140 sector += STRIPE_SECTORS,
6141 scnt++) {
46031f9a 6142
0472a42b 6143 if (scnt < offset)
46031f9a
RBJ		 6144 /* already done this stripe */
6145 continue;
6146
6d036f7d 6147 sh = raid5_get_active_stripe(conf, sector, 0, 1, 1);
46031f9a
RBJ		 6148
6149 if (!sh) {
6150 /* failed to get a stripe - must wait */
46031f9a 6151 conf->retry_read_aligned = raid_bio;
0472a42b 6152 conf->retry_read_offset = scnt;
46031f9a
RBJ		 6153 return handled;
6154 }
6155
da41ba65 6156 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) {
6d036f7d 6157 raid5_release_stripe(sh);
387bb173 6158 conf->retry_read_aligned = raid_bio;
0472a42b 6159 conf->retry_read_offset = scnt;
387bb173
NB		 6160 return handled;
6161 }
6162
3f9e7c14 6163 set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags);
36d1c647 6164 handle_stripe(sh);
6d036f7d 6165 raid5_release_stripe(sh);
46031f9a
RBJ		 6166 handled++;
6167 }
016c76ac
N		 6168
6169 bio_endio(raid_bio);
6170
46031f9a 6171 if (atomic_dec_and_test(&conf->active_aligned_reads))
b1b46486 6172 wake_up(&conf->wait_for_quiescent);
46031f9a
RBJ		 6173 return handled;
6174}
6175
bfc90cb0 6176static int handle_active_stripes(struct r5conf *conf, int group,
566c09c5
SL		 6177 struct r5worker *worker,
6178 struct list_head *temp_inactive_list)
46a06401
SL		 6179{
6180 struct stripe_head *batch[MAX_STRIPE_BATCH], *sh;
566c09c5
SL		 6181 int i, batch_size = 0, hash;
6182 bool release_inactive = false;
46a06401
SL		 6183
6184 while (batch_size < MAX_STRIPE_BATCH &&
851c30c9 6185 (sh = __get_priority_stripe(conf, group)) != NULL)
46a06401
SL		 6186 batch[batch_size++] = sh;
6187
566c09c5
SL		 6188 if (batch_size == 0) {
6189 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
6190 if (!list_empty(temp_inactive_list + i))
6191 break;
a8c34f91
SL		 6192 if (i == NR_STRIPE_HASH_LOCKS) {
6193 spin_unlock_irq(&conf->device_lock);
1532d9e8 6194 log_flush_stripe_to_raid(conf);
a8c34f91 6195 spin_lock_irq(&conf->device_lock);
566c09c5 6196 return batch_size;
a8c34f91 6197 }
566c09c5
SL		 6198 release_inactive = true;
6199 }
46a06401
SL		 6200 spin_unlock_irq(&conf->device_lock);
6201
566c09c5
SL		 6202 release_inactive_stripe_list(conf, temp_inactive_list,
6203 NR_STRIPE_HASH_LOCKS);
6204
a8c34f91 6205 r5l_flush_stripe_to_raid(conf->log);
566c09c5
SL		 6206 if (release_inactive) {
6207 spin_lock_irq(&conf->device_lock);
6208 return 0;
6209 }
6210
46a06401
SL		 6211 for (i = 0; i < batch_size; i++)
6212 handle_stripe(batch[i]);
ff875738 6213 log_write_stripe_run(conf);
46a06401
SL		 6214
6215 cond_resched();
6216
6217 spin_lock_irq(&conf->device_lock);
566c09c5
SL		 6218 for (i = 0; i < batch_size; i++) {
6219 hash = batch[i]->hash_lock_index;
6220 __release_stripe(conf, batch[i], &temp_inactive_list[hash]);
6221 }
46a06401
SL		 6222 return batch_size;
6223}
46031f9a 6224
851c30c9
SL		 6225static void raid5_do_work(struct work_struct *work)
6226{
6227 struct r5worker *worker = container_of(work, struct r5worker, work);
6228 struct r5worker_group *group = worker->group;
6229 struct r5conf *conf = group->conf;
16d997b7 6230 struct mddev *mddev = conf->mddev;
851c30c9
SL		 6231 int group_id = group - conf->worker_groups;
6232 int handled;
6233 struct blk_plug plug;
6234
6235 pr_debug("+++ raid5worker active\n");
6236
6237 blk_start_plug(&plug);
6238 handled = 0;
6239 spin_lock_irq(&conf->device_lock);
6240 while (1) {
6241 int batch_size, released;
6242
566c09c5 6243 released = release_stripe_list(conf, worker->temp_inactive_list);
851c30c9 6244
566c09c5
SL		 6245 batch_size = handle_active_stripes(conf, group_id, worker,
6246 worker->temp_inactive_list);
bfc90cb0 6247 worker->working = false;
851c30c9
SL		 6248 if (!batch_size && !released)
6249 break;
6250 handled += batch_size;
16d997b7
N		 6251 wait_event_lock_irq(mddev->sb_wait,
6252 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags),
6253 conf->device_lock);
851c30c9
SL		 6254 }
6255 pr_debug("%d stripes handled\n", handled);
6256
6257 spin_unlock_irq(&conf->device_lock);
7e96d559 6258
9c72a18e
SL		 6259 flush_deferred_bios(conf);
6260
6261 r5l_flush_stripe_to_raid(conf->log);
6262
7e96d559 6263 async_tx_issue_pending_all();
851c30c9
SL		 6264 blk_finish_plug(&plug);
6265
6266 pr_debug("--- raid5worker inactive\n");
6267}
6268
1da177e4
LT		 6269/*
6270 * This is our raid5 kernel thread.
6271 *
6272 * We scan the hash table for stripes which can be handled now.
6273 * During the scan, completed stripes are saved for us by the interrupt
6274 * handler, so that they will not have to wait for our next wakeup.
6275 */
4ed8731d 6276static void raid5d(struct md_thread *thread)
1da177e4 6277{
4ed8731d 6278 struct mddev *mddev = thread->mddev;
d1688a6d 6279 struct r5conf *conf = mddev->private;
1da177e4 6280 int handled;
e1dfa0a2 6281 struct blk_plug plug;
1da177e4 6282
45b4233c 6283 pr_debug("+++ raid5d active\n");
1da177e4
LT		 6284
6285 md_check_recovery(mddev);
1da177e4 6286
e1dfa0a2 6287 blk_start_plug(&plug);
1da177e4
LT		 6288 handled = 0;
6289 spin_lock_irq(&conf->device_lock);
6290 while (1) {
46031f9a 6291 struct bio *bio;
773ca82f 6292 int batch_size, released;
0472a42b 6293 unsigned int offset;
773ca82f 6294
566c09c5 6295 released = release_stripe_list(conf, conf->temp_inactive_list);
edbe83ab
N		 6296 if (released)
6297 clear_bit(R5_DID_ALLOC, &conf->cache_state);
1da177e4 6298
0021b7bc 6299 if (
7c13edc8
N		 6300 !list_empty(&conf->bitmap_list)) {
6301 /* Now is a good time to flush some bitmap updates */
6302 conf->seq_flush++;
700e432d 6303 spin_unlock_irq(&conf->device_lock);
e64e4018 6304 md_bitmap_unplug(mddev->bitmap);
700e432d 6305 spin_lock_irq(&conf->device_lock);
7c13edc8 6306 conf->seq_write = conf->seq_flush;
566c09c5 6307 activate_bit_delay(conf, conf->temp_inactive_list);
72626685 6308 }
0021b7bc 6309 raid5_activate_delayed(conf);
72626685 6310
0472a42b 6311 while ((bio = remove_bio_from_retry(conf, &offset))) {
46031f9a
RBJ		 6312 int ok;
6313 spin_unlock_irq(&conf->device_lock);
0472a42b 6314 ok = retry_aligned_read(conf, bio, offset);
46031f9a
RBJ		 6315 spin_lock_irq(&conf->device_lock);
6316 if (!ok)
6317 break;
6318 handled++;
6319 }
6320
566c09c5
SL		 6321 batch_size = handle_active_stripes(conf, ANY_GROUP, NULL,
6322 conf->temp_inactive_list);
773ca82f 6323 if (!batch_size && !released)
1da177e4 6324 break;
46a06401 6325 handled += batch_size;
1da177e4 6326
2953079c 6327 if (mddev->sb_flags & ~(1 << MD_SB_CHANGE_PENDING)) {
46a06401 6328 spin_unlock_irq(&conf->device_lock);
de393cde 6329 md_check_recovery(mddev);
46a06401
SL		 6330 spin_lock_irq(&conf->device_lock);
6331 }
1da177e4 6332 }
45b4233c 6333 pr_debug("%d stripes handled\n", handled);
1da177e4
LT		 6334
6335 spin_unlock_irq(&conf->device_lock);
2d5b569b
N		 6336 if (test_and_clear_bit(R5_ALLOC_MORE, &conf->cache_state) &&
6337 mutex_trylock(&conf->cache_size_mutex)) {
edbe83ab
N		 6338 grow_one_stripe(conf, __GFP_NOWARN);
6339 /* Set flag even if allocation failed. This helps
6340 * slow down allocation requests when mem is short
6341 */
6342 set_bit(R5_DID_ALLOC, &conf->cache_state);
2d5b569b 6343 mutex_unlock(&conf->cache_size_mutex);
edbe83ab 6344 }
1da177e4 6345
765d704d
SL		 6346 flush_deferred_bios(conf);
6347
0576b1c6
SL		 6348 r5l_flush_stripe_to_raid(conf->log);
6349
c9f21aaf 6350 async_tx_issue_pending_all();
e1dfa0a2 6351 blk_finish_plug(&plug);
1da177e4 6352
45b4233c 6353 pr_debug("--- raid5d inactive\n");
1da177e4
LT		 6354}
6355
3f294f4f 6356static ssize_t
fd01b88c 6357raid5_show_stripe_cache_size(struct mddev *mddev, char *page)
3f294f4f 6358{
7b1485ba
N		 6359 struct r5conf *conf;
6360 int ret = 0;
6361 spin_lock(&mddev->lock);
6362 conf = mddev->private;
96de1e66 6363 if (conf)
edbe83ab 6364 ret = sprintf(page, "%d\n", conf->min_nr_stripes);
7b1485ba
N		 6365 spin_unlock(&mddev->lock);
6366 return ret;
3f294f4f
N		 6367}
6368
c41d4ac4 6369int
fd01b88c 6370raid5_set_cache_size(struct mddev *mddev, int size)
3f294f4f 6371{
d1688a6d 6372 struct r5conf *conf = mddev->private;
b5470dc5 6373
c41d4ac4 6374 if (size <= 16 || size > 32768)
3f294f4f 6375 return -EINVAL;
486f0644 6376
edbe83ab 6377 conf->min_nr_stripes = size;
2d5b569b 6378 mutex_lock(&conf->cache_size_mutex);
486f0644
N		 6379 while (size < conf->max_nr_stripes &&
6380 drop_one_stripe(conf))
6381 ;
2d5b569b 6382 mutex_unlock(&conf->cache_size_mutex);
486f0644 6383
2214c260 6384 md_allow_write(mddev);
486f0644 6385
2d5b569b 6386 mutex_lock(&conf->cache_size_mutex);
486f0644
N		 6387 while (size > conf->max_nr_stripes)
6388 if (!grow_one_stripe(conf, GFP_KERNEL))
6389 break;
2d5b569b 6390 mutex_unlock(&conf->cache_size_mutex);
486f0644 6391
c41d4ac4
N		 6392 return 0;
6393}
6394EXPORT_SYMBOL(raid5_set_cache_size);
6395
6396static ssize_t
fd01b88c 6397raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len)
c41d4ac4 6398{
6791875e 6399 struct r5conf *conf;
c41d4ac4
N		 6400 unsigned long new;
6401 int err;
6402
6403 if (len >= PAGE_SIZE)
6404 return -EINVAL;
b29bebd6 6405 if (kstrtoul(page, 10, &new))
c41d4ac4 6406 return -EINVAL;
6791875e 6407 err = mddev_lock(mddev);
c41d4ac4
N		 6408 if (err)
6409 return err;
6791875e
N		 6410 conf = mddev->private;
6411 if (!conf)
6412 err = -ENODEV;
6413 else
6414 err = raid5_set_cache_size(mddev, new);
6415 mddev_unlock(mddev);
6416
6417 return err ?: len;
3f294f4f 6418}
007583c9 6419
96de1e66
N		 6420static struct md_sysfs_entry
6421raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
6422 raid5_show_stripe_cache_size,
6423 raid5_store_stripe_cache_size);
3f294f4f 6424
d06f191f
MS		 6425static ssize_t
6426raid5_show_rmw_level(struct mddev *mddev, char *page)
6427{
6428 struct r5conf *conf = mddev->private;
6429 if (conf)
6430 return sprintf(page, "%d\n", conf->rmw_level);
6431 else
6432 return 0;
6433}
6434
6435static ssize_t
6436raid5_store_rmw_level(struct mddev *mddev, const char *page, size_t len)
6437{
6438 struct r5conf *conf = mddev->private;
6439 unsigned long new;
6440
6441 if (!conf)
6442 return -ENODEV;
6443
6444 if (len >= PAGE_SIZE)
6445 return -EINVAL;
6446
6447 if (kstrtoul(page, 10, &new))
6448 return -EINVAL;
6449
6450 if (new != PARITY_DISABLE_RMW && !raid6_call.xor_syndrome)
6451 return -EINVAL;
6452
6453 if (new != PARITY_DISABLE_RMW &&
6454 new != PARITY_ENABLE_RMW &&
6455 new != PARITY_PREFER_RMW)
6456 return -EINVAL;
6457
6458 conf->rmw_level = new;
6459 return len;
6460}
6461
6462static struct md_sysfs_entry
6463raid5_rmw_level = __ATTR(rmw_level, S_IRUGO | S_IWUSR,
6464 raid5_show_rmw_level,
6465 raid5_store_rmw_level);
6466
6467
8b3e6cdc 6468static ssize_t
fd01b88c 6469raid5_show_preread_threshold(struct mddev *mddev, char *page)
8b3e6cdc 6470{
7b1485ba
N		 6471 struct r5conf *conf;
6472 int ret = 0;
6473 spin_lock(&mddev->lock);
6474 conf = mddev->private;
8b3e6cdc 6475 if (conf)
7b1485ba
N		 6476 ret = sprintf(page, "%d\n", conf->bypass_threshold);
6477 spin_unlock(&mddev->lock);
6478 return ret;
8b3e6cdc
DW		 6479}
6480
6481static ssize_t
fd01b88c 6482raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len)
8b3e6cdc 6483{
6791875e 6484 struct r5conf *conf;
4ef197d8 6485 unsigned long new;
6791875e
N		 6486 int err;
6487
8b3e6cdc
DW		 6488 if (len >= PAGE_SIZE)
6489 return -EINVAL;
b29bebd6 6490 if (kstrtoul(page, 10, &new))
8b3e6cdc 6491 return -EINVAL;
6791875e
N		 6492
6493 err = mddev_lock(mddev);
6494 if (err)
6495 return err;
6496 conf = mddev->private;
6497 if (!conf)
6498 err = -ENODEV;
edbe83ab 6499 else if (new > conf->min_nr_stripes)
6791875e
N		 6500 err = -EINVAL;
6501 else
6502 conf->bypass_threshold = new;
6503 mddev_unlock(mddev);
6504 return err ?: len;
8b3e6cdc
DW		 6505}
6506
6507static struct md_sysfs_entry
6508raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
6509 S_IRUGO | S_IWUSR,
6510 raid5_show_preread_threshold,
6511 raid5_store_preread_threshold);
6512
d592a996
SL		 6513static ssize_t
6514raid5_show_skip_copy(struct mddev *mddev, char *page)
6515{
7b1485ba
N		 6516 struct r5conf *conf;
6517 int ret = 0;
6518 spin_lock(&mddev->lock);
6519 conf = mddev->private;
d592a996 6520 if (conf)
7b1485ba
N		 6521 ret = sprintf(page, "%d\n", conf->skip_copy);
6522 spin_unlock(&mddev->lock);
6523 return ret;
d592a996
SL		 6524}
6525
6526static ssize_t
6527raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len)
6528{
6791875e 6529 struct r5conf *conf;
d592a996 6530 unsigned long new;
6791875e
N		 6531 int err;
6532
d592a996
SL		 6533 if (len >= PAGE_SIZE)
6534 return -EINVAL;
d592a996
SL		 6535 if (kstrtoul(page, 10, &new))
6536 return -EINVAL;
6537 new = !!new;
6791875e
N		 6538
6539 err = mddev_lock(mddev);
6540 if (err)
6541 return err;
6542 conf = mddev->private;
6543 if (!conf)
6544 err = -ENODEV;
6545 else if (new != conf->skip_copy) {
6546 mddev_suspend(mddev);
6547 conf->skip_copy = new;
6548 if (new)
dc3b17cc 6549 mddev->queue->backing_dev_info->capabilities |=
6791875e
N		 6550 BDI_CAP_STABLE_WRITES;
6551 else
dc3b17cc 6552 mddev->queue->backing_dev_info->capabilities &=
6791875e
N		 6553 ~BDI_CAP_STABLE_WRITES;
6554 mddev_resume(mddev);
6555 }
6556 mddev_unlock(mddev);
6557 return err ?: len;
d592a996
SL		 6558}
6559
6560static struct md_sysfs_entry
6561raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR,
6562 raid5_show_skip_copy,
6563 raid5_store_skip_copy);
6564
3f294f4f 6565static ssize_t
fd01b88c 6566stripe_cache_active_show(struct mddev *mddev, char *page)
3f294f4f 6567{
d1688a6d 6568 struct r5conf *conf = mddev->private;
96de1e66
N		 6569 if (conf)
6570 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
6571 else
6572 return 0;
3f294f4f
N		 6573}
6574
96de1e66
N		 6575static struct md_sysfs_entry
6576raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
3f294f4f 6577
b721420e
SL		 6578static ssize_t
6579raid5_show_group_thread_cnt(struct mddev *mddev, char *page)
6580{
7b1485ba
N		 6581 struct r5conf *conf;
6582 int ret = 0;
6583 spin_lock(&mddev->lock);
6584 conf = mddev->private;
b721420e 6585 if (conf)
7b1485ba
N		 6586 ret = sprintf(page, "%d\n", conf->worker_cnt_per_group);
6587 spin_unlock(&mddev->lock);
6588 return ret;
b721420e
SL		 6589}
6590
60aaf933 6591static int alloc_thread_groups(struct r5conf *conf, int cnt,
6592 int *group_cnt,
6593 int *worker_cnt_per_group,
6594 struct r5worker_group **worker_groups);
b721420e
SL		 6595static ssize_t
6596raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len)
6597{
6791875e 6598 struct r5conf *conf;
7d5d7b50 6599 unsigned int new;
b721420e 6600 int err;
60aaf933 6601 struct r5worker_group *new_groups, *old_groups;
6602 int group_cnt, worker_cnt_per_group;
b721420e
SL		 6603
6604 if (len >= PAGE_SIZE)
6605 return -EINVAL;
7d5d7b50
SL		 6606 if (kstrtouint(page, 10, &new))
6607 return -EINVAL;
6608 /* 8192 should be big enough */
6609 if (new > 8192)
b721420e
SL		 6610 return -EINVAL;
6611
6791875e
N		 6612 err = mddev_lock(mddev);
6613 if (err)
6614 return err;
6615 conf = mddev->private;
6616 if (!conf)
6617 err = -ENODEV;
6618 else if (new != conf->worker_cnt_per_group) {
6619 mddev_suspend(mddev);
b721420e 6620
6791875e
N		 6621 old_groups = conf->worker_groups;
6622 if (old_groups)
6623 flush_workqueue(raid5_wq);
d206dcfa 6624
6791875e
N		 6625 err = alloc_thread_groups(conf, new,
6626 &group_cnt, &worker_cnt_per_group,
6627 &new_groups);
6628 if (!err) {
6629 spin_lock_irq(&conf->device_lock);
6630 conf->group_cnt = group_cnt;
6631 conf->worker_cnt_per_group = worker_cnt_per_group;
6632 conf->worker_groups = new_groups;
6633 spin_unlock_irq(&conf->device_lock);
b721420e 6634
6791875e
N		 6635 if (old_groups)
6636 kfree(old_groups[0].workers);
6637 kfree(old_groups);
6638 }
6639 mddev_resume(mddev);
b721420e 6640 }
6791875e 6641 mddev_unlock(mddev);
b721420e 6642
6791875e 6643 return err ?: len;
b721420e
SL		 6644}
6645
6646static struct md_sysfs_entry
6647raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR,
6648 raid5_show_group_thread_cnt,
6649 raid5_store_group_thread_cnt);
6650
007583c9 6651static struct attribute *raid5_attrs[] = {
3f294f4f
N		 6652 &raid5_stripecache_size.attr,
6653 &raid5_stripecache_active.attr,
8b3e6cdc 6654 &raid5_preread_bypass_threshold.attr,
b721420e 6655 &raid5_group_thread_cnt.attr,
d592a996 6656 &raid5_skip_copy.attr,
d06f191f 6657 &raid5_rmw_level.attr,
2c7da14b 6658 &r5c_journal_mode.attr,
3f294f4f
N		 6659 NULL,
6660};
007583c9
N		 6661static struct attribute_group raid5_attrs_group = {
6662 .name = NULL,
6663 .attrs = raid5_attrs,
3f294f4f
N		 6664};
6665
60aaf933 6666static int alloc_thread_groups(struct r5conf *conf, int cnt,
6667 int *group_cnt,
6668 int *worker_cnt_per_group,
6669 struct r5worker_group **worker_groups)
851c30c9 6670{
566c09c5 6671 int i, j, k;
851c30c9
SL		 6672 ssize_t size;
6673 struct r5worker *workers;
6674
60aaf933 6675 *worker_cnt_per_group = cnt;
851c30c9 6676 if (cnt == 0) {
60aaf933 6677 *group_cnt = 0;
6678 *worker_groups = NULL;
851c30c9
SL		 6679 return 0;
6680 }
60aaf933 6681 *group_cnt = num_possible_nodes();
851c30c9 6682 size = sizeof(struct r5worker) * cnt;
6396bb22
KC		 6683 workers = kcalloc(size, *group_cnt, GFP_NOIO);
6684 *worker_groups = kcalloc(*group_cnt, sizeof(struct r5worker_group),
6685 GFP_NOIO);
60aaf933 6686 if (!*worker_groups || !workers) {
851c30c9 6687 kfree(workers);
60aaf933 6688 kfree(*worker_groups);
851c30c9
SL		 6689 return -ENOMEM;
6690 }
6691
60aaf933 6692 for (i = 0; i < *group_cnt; i++) {
851c30c9
SL		 6693 struct r5worker_group *group;
6694
0c775d52 6695 group = &(*worker_groups)[i];
851c30c9 6696 INIT_LIST_HEAD(&group->handle_list);
535ae4eb 6697 INIT_LIST_HEAD(&group->loprio_list);
851c30c9
SL		 6698 group->conf = conf;
6699 group->workers = workers + i * cnt;
6700
6701 for (j = 0; j < cnt; j++) {
566c09c5
SL		 6702 struct r5worker *worker = group->workers + j;
6703 worker->group = group;
6704 INIT_WORK(&worker->work, raid5_do_work);
6705
6706 for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++)
6707 INIT_LIST_HEAD(worker->temp_inactive_list + k);
851c30c9
SL		 6708 }
6709 }
6710
6711 return 0;
6712}
6713
6714static void free_thread_groups(struct r5conf *conf)
6715{
6716 if (conf->worker_groups)
6717 kfree(conf->worker_groups[0].workers);
6718 kfree(conf->worker_groups);
6719 conf->worker_groups = NULL;
6720}
6721
80c3a6ce 6722static sector_t
fd01b88c 6723raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks)
80c3a6ce 6724{
d1688a6d 6725 struct r5conf *conf = mddev->private;
80c3a6ce
DW		 6726
6727 if (!sectors)
6728 sectors = mddev->dev_sectors;
5e5e3e78 6729 if (!raid_disks)
7ec05478 6730 /* size is defined by the smallest of previous and new size */
5e5e3e78 6731 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
80c3a6ce 6732
3cb5edf4
N		 6733 sectors &= ~((sector_t)conf->chunk_sectors - 1);
6734 sectors &= ~((sector_t)conf->prev_chunk_sectors - 1);
80c3a6ce
DW		 6735 return sectors * (raid_disks - conf->max_degraded);
6736}
6737
789b5e03
ON		 6738static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
6739{
6740 safe_put_page(percpu->spare_page);
46d5b785 6741 if (percpu->scribble)
6742 flex_array_free(percpu->scribble);
789b5e03
ON		 6743 percpu->spare_page = NULL;
6744 percpu->scribble = NULL;
6745}
6746
6747static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
6748{
6749 if (conf->level == 6 && !percpu->spare_page)
6750 percpu->spare_page = alloc_page(GFP_KERNEL);
6751 if (!percpu->scribble)
46d5b785 6752 percpu->scribble = scribble_alloc(max(conf->raid_disks,
738a2738
N		 6753 conf->previous_raid_disks),
6754 max(conf->chunk_sectors,
6755 conf->prev_chunk_sectors)
6756 / STRIPE_SECTORS,
6757 GFP_KERNEL);
789b5e03
ON		 6758
6759 if (!percpu->scribble || (conf->level == 6 && !percpu->spare_page)) {
6760 free_scratch_buffer(conf, percpu);
6761 return -ENOMEM;
6762 }
6763
6764 return 0;
6765}
6766
29c6d1bb 6767static int raid456_cpu_dead(unsigned int cpu, struct hlist_node *node)
36d1c647 6768{
29c6d1bb
SAS		 6769 struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node);
6770
6771 free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
6772 return 0;
6773}
36d1c647 6774
29c6d1bb
SAS		 6775static void raid5_free_percpu(struct r5conf *conf)
6776{
36d1c647
DW		 6777 if (!conf->percpu)
6778 return;
6779
29c6d1bb 6780 cpuhp_state_remove_instance(CPUHP_MD_RAID5_PREPARE, &conf->node);
36d1c647
DW		 6781 free_percpu(conf->percpu);
6782}
6783
d1688a6d 6784static void free_conf(struct r5conf *conf)
95fc17aa 6785{
d7bd398e
SL		 6786 int i;
6787
ff875738
AP		 6788 log_exit(conf);
6789
565e0450 6790 unregister_shrinker(&conf->shrinker);
851c30c9 6791 free_thread_groups(conf);
95fc17aa 6792 shrink_stripes(conf);
36d1c647 6793 raid5_free_percpu(conf);
d7bd398e
SL		 6794 for (i = 0; i < conf->pool_size; i++)
6795 if (conf->disks[i].extra_page)
6796 put_page(conf->disks[i].extra_page);
95fc17aa 6797 kfree(conf->disks);
afeee514 6798 bioset_exit(&conf->bio_split);
95fc17aa 6799 kfree(conf->stripe_hashtbl);
aaf9f12e 6800 kfree(conf->pending_data);
95fc17aa
DW		 6801 kfree(conf);
6802}
6803
29c6d1bb 6804static int raid456_cpu_up_prepare(unsigned int cpu, struct hlist_node *node)
36d1c647 6805{
29c6d1bb 6806 struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node);
36d1c647
DW		 6807 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
6808
29c6d1bb 6809 if (alloc_scratch_buffer(conf, percpu)) {
cc6167b4
N		 6810 pr_warn("%s: failed memory allocation for cpu%u\n",
6811 __func__, cpu);
29c6d1bb 6812 return -ENOMEM;
36d1c647 6813 }
29c6d1bb 6814 return 0;
36d1c647 6815}
36d1c647 6816
d1688a6d 6817static int raid5_alloc_percpu(struct r5conf *conf)
36d1c647 6818{
789b5e03 6819 int err = 0;
36d1c647 6820
789b5e03
ON		 6821 conf->percpu = alloc_percpu(struct raid5_percpu);
6822 if (!conf->percpu)
36d1c647 6823 return -ENOMEM;
789b5e03 6824
29c6d1bb 6825 err = cpuhp_state_add_instance(CPUHP_MD_RAID5_PREPARE, &conf->node);
27a353c0
SL		 6826 if (!err) {
6827 conf->scribble_disks = max(conf->raid_disks,
6828 conf->previous_raid_disks);
6829 conf->scribble_sectors = max(conf->chunk_sectors,
6830 conf->prev_chunk_sectors);
6831 }
36d1c647
DW		 6832 return err;
6833}
6834
edbe83ab
N		 6835static unsigned long raid5_cache_scan(struct shrinker *shrink,
6836 struct shrink_control *sc)
6837{
6838 struct r5conf *conf = container_of(shrink, struct r5conf, shrinker);
2d5b569b
N		 6839 unsigned long ret = SHRINK_STOP;
6840
6841 if (mutex_trylock(&conf->cache_size_mutex)) {
6842 ret= 0;
49895bcc
N		 6843 while (ret < sc->nr_to_scan &&
6844 conf->max_nr_stripes > conf->min_nr_stripes) {
2d5b569b
N		 6845 if (drop_one_stripe(conf) == 0) {
6846 ret = SHRINK_STOP;
6847 break;
6848 }
6849 ret++;
6850 }
6851 mutex_unlock(&conf->cache_size_mutex);
edbe83ab
N		 6852 }
6853 return ret;
6854}
6855
6856static unsigned long raid5_cache_count(struct shrinker *shrink,
6857 struct shrink_control *sc)
6858{
6859 struct r5conf *conf = container_of(shrink, struct r5conf, shrinker);
6860
6861 if (conf->max_nr_stripes < conf->min_nr_stripes)
6862 /* unlikely, but not impossible */
6863 return 0;
6864 return conf->max_nr_stripes - conf->min_nr_stripes;
6865}
6866
d1688a6d 6867static struct r5conf *setup_conf(struct mddev *mddev)
1da177e4 6868{
d1688a6d 6869 struct r5conf *conf;
5e5e3e78 6870 int raid_disk, memory, max_disks;
3cb03002 6871 struct md_rdev *rdev;
1da177e4 6872 struct disk_info *disk;
0232605d 6873 char pers_name[6];
566c09c5 6874 int i;
60aaf933 6875 int group_cnt, worker_cnt_per_group;
6876 struct r5worker_group *new_group;
afeee514 6877 int ret;
1da177e4 6878
91adb564
N		 6879 if (mddev->new_level != 5
6880 && mddev->new_level != 4
6881 && mddev->new_level != 6) {
cc6167b4
N		 6882 pr_warn("md/raid:%s: raid level not set to 4/5/6 (%d)\n",
6883 mdname(mddev), mddev->new_level);
91adb564 6884 return ERR_PTR(-EIO);
1da177e4 6885 }
91adb564
N		 6886 if ((mddev->new_level == 5
6887 && !algorithm_valid_raid5(mddev->new_layout)) ||
6888 (mddev->new_level == 6
6889 && !algorithm_valid_raid6(mddev->new_layout))) {
cc6167b4
N		 6890 pr_warn("md/raid:%s: layout %d not supported\n",
6891 mdname(mddev), mddev->new_layout);
91adb564 6892 return ERR_PTR(-EIO);
99c0fb5f 6893 }
91adb564 6894 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
cc6167b4
N		 6895 pr_warn("md/raid:%s: not enough configured devices (%d, minimum 4)\n",
6896 mdname(mddev), mddev->raid_disks);
91adb564 6897 return ERR_PTR(-EINVAL);
4bbf3771
N		 6898 }
6899
664e7c41
AN		 6900 if (!mddev->new_chunk_sectors ||
6901 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
6902 !is_power_of_2(mddev->new_chunk_sectors)) {
cc6167b4
N		 6903 pr_warn("md/raid:%s: invalid chunk size %d\n",
6904 mdname(mddev), mddev->new_chunk_sectors << 9);
91adb564 6905 return ERR_PTR(-EINVAL);
f6705578
N		 6906 }
6907
d1688a6d 6908 conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL);
91adb564 6909 if (conf == NULL)
1da177e4 6910 goto abort;
aaf9f12e
SL		 6911 INIT_LIST_HEAD(&conf->free_list);
6912 INIT_LIST_HEAD(&conf->pending_list);
6396bb22
KC		 6913 conf->pending_data = kcalloc(PENDING_IO_MAX,
6914 sizeof(struct r5pending_data),
6915 GFP_KERNEL);
aaf9f12e
SL		 6916 if (!conf->pending_data)
6917 goto abort;
6918 for (i = 0; i < PENDING_IO_MAX; i++)
6919 list_add(&conf->pending_data[i].sibling, &conf->free_list);
851c30c9 6920 /* Don't enable multi-threading by default*/
60aaf933 6921 if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group,
6922 &new_group)) {
6923 conf->group_cnt = group_cnt;
6924 conf->worker_cnt_per_group = worker_cnt_per_group;
6925 conf->worker_groups = new_group;
6926 } else
851c30c9 6927 goto abort;
f5efd45a 6928 spin_lock_init(&conf->device_lock);
c46501b2 6929 seqcount_init(&conf->gen_lock);
2d5b569b 6930 mutex_init(&conf->cache_size_mutex);
b1b46486 6931 init_waitqueue_head(&conf->wait_for_quiescent);
6ab2a4b8 6932 init_waitqueue_head(&conf->wait_for_stripe);
f5efd45a
DW		 6933 init_waitqueue_head(&conf->wait_for_overlap);
6934 INIT_LIST_HEAD(&conf->handle_list);
535ae4eb 6935 INIT_LIST_HEAD(&conf->loprio_list);
f5efd45a
DW		 6936 INIT_LIST_HEAD(&conf->hold_list);
6937 INIT_LIST_HEAD(&conf->delayed_list);
6938 INIT_LIST_HEAD(&conf->bitmap_list);
773ca82f 6939 init_llist_head(&conf->released_stripes);
f5efd45a
DW		 6940 atomic_set(&conf->active_stripes, 0);
6941 atomic_set(&conf->preread_active_stripes, 0);
6942 atomic_set(&conf->active_aligned_reads, 0);
765d704d
SL		 6943 spin_lock_init(&conf->pending_bios_lock);
6944 conf->batch_bio_dispatch = true;
6945 rdev_for_each(rdev, mddev) {
6946 if (test_bit(Journal, &rdev->flags))
6947 continue;
6948 if (blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
6949 conf->batch_bio_dispatch = false;
6950 break;
6951 }
6952 }
6953
f5efd45a 6954 conf->bypass_threshold = BYPASS_THRESHOLD;
d890fa2b 6955 conf->recovery_disabled = mddev->recovery_disabled - 1;
91adb564
N		 6956
6957 conf->raid_disks = mddev->raid_disks;
6958 if (mddev->reshape_position == MaxSector)
6959 conf->previous_raid_disks = mddev->raid_disks;
6960 else
f6705578 6961 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
5e5e3e78 6962 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
f6705578 6963
6396bb22 6964 conf->disks = kcalloc(max_disks, sizeof(struct disk_info),
b55e6bfc 6965 GFP_KERNEL);
d7bd398e 6966
b55e6bfc
N		 6967 if (!conf->disks)
6968 goto abort;
9ffae0cf 6969
d7bd398e
SL		 6970 for (i = 0; i < max_disks; i++) {
6971 conf->disks[i].extra_page = alloc_page(GFP_KERNEL);
6972 if (!conf->disks[i].extra_page)
6973 goto abort;
6974 }
6975
afeee514
KO		 6976 ret = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0);
6977 if (ret)
dd7a8f5d 6978 goto abort;
1da177e4
LT		 6979 conf->mddev = mddev;
6980
fccddba0 6981 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
1da177e4 6982 goto abort;
1da177e4 6983
566c09c5
SL		 6984 /* We init hash_locks[0] separately to that it can be used
6985 * as the reference lock in the spin_lock_nest_lock() call
6986 * in lock_all_device_hash_locks_irq in order to convince
6987 * lockdep that we know what we are doing.
6988 */
6989 spin_lock_init(conf->hash_locks);
6990 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
6991 spin_lock_init(conf->hash_locks + i);
6992
6993 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
6994 INIT_LIST_HEAD(conf->inactive_list + i);
6995
6996 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
6997 INIT_LIST_HEAD(conf->temp_inactive_list + i);
6998
1e6d690b
SL		 6999 atomic_set(&conf->r5c_cached_full_stripes, 0);
7000 INIT_LIST_HEAD(&conf->r5c_full_stripe_list);
7001 atomic_set(&conf->r5c_cached_partial_stripes, 0);
7002 INIT_LIST_HEAD(&conf->r5c_partial_stripe_list);
e33fbb9c
SL		 7003 atomic_set(&conf->r5c_flushing_full_stripes, 0);
7004 atomic_set(&conf->r5c_flushing_partial_stripes, 0);
1e6d690b 7005
36d1c647 7006 conf->level = mddev->new_level;
46d5b785 7007 conf->chunk_sectors = mddev->new_chunk_sectors;
36d1c647
DW		 7008 if (raid5_alloc_percpu(conf) != 0)
7009 goto abort;
7010
0c55e022 7011 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
1da177e4 7012
dafb20fa 7013 rdev_for_each(rdev, mddev) {
1da177e4 7014 raid_disk = rdev->raid_disk;
5e5e3e78 7015 if (raid_disk >= max_disks
f2076e7d 7016 || raid_disk < 0 || test_bit(Journal, &rdev->flags))
1da177e4
LT		 7017 continue;
7018 disk = conf->disks + raid_disk;
7019
17045f52
N		 7020 if (test_bit(Replacement, &rdev->flags)) {
7021 if (disk->replacement)
7022 goto abort;
7023 disk->replacement = rdev;
7024 } else {
7025 if (disk->rdev)
7026 goto abort;
7027 disk->rdev = rdev;
7028 }
1da177e4 7029
b2d444d7 7030 if (test_bit(In_sync, &rdev->flags)) {
1da177e4 7031 char b[BDEVNAME_SIZE];
cc6167b4
N		 7032 pr_info("md/raid:%s: device %s operational as raid disk %d\n",
7033 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
d6b212f4 7034 } else if (rdev->saved_raid_disk != raid_disk)
8c2e870a
NB		 7035 /* Cannot rely on bitmap to complete recovery */
7036 conf->fullsync = 1;
1da177e4
LT		 7037 }
7038
91adb564 7039 conf->level = mddev->new_level;
584acdd4 7040 if (conf->level == 6) {
16a53ecc 7041 conf->max_degraded = 2;
584acdd4
MS		 7042 if (raid6_call.xor_syndrome)
7043 conf->rmw_level = PARITY_ENABLE_RMW;
7044 else
7045 conf->rmw_level = PARITY_DISABLE_RMW;
7046 } else {
16a53ecc 7047 conf->max_degraded = 1;
584acdd4
MS		 7048 conf->rmw_level = PARITY_ENABLE_RMW;
7049 }
91adb564 7050 conf->algorithm = mddev->new_layout;
fef9c61f 7051 conf->reshape_progress = mddev->reshape_position;
e183eaed 7052 if (conf->reshape_progress != MaxSector) {
09c9e5fa 7053 conf->prev_chunk_sectors = mddev->chunk_sectors;
e183eaed 7054 conf->prev_algo = mddev->layout;
5cac6bcb
N		 7055 } else {
7056 conf->prev_chunk_sectors = conf->chunk_sectors;
7057 conf->prev_algo = conf->algorithm;
e183eaed 7058 }
1da177e4 7059
edbe83ab 7060 conf->min_nr_stripes = NR_STRIPES;
ad5b0f76
SL		 7061 if (mddev->reshape_position != MaxSector) {
7062 int stripes = max_t(int,
7063 ((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4,
7064 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4);
7065 conf->min_nr_stripes = max(NR_STRIPES, stripes);
7066 if (conf->min_nr_stripes != NR_STRIPES)
cc6167b4 7067 pr_info("md/raid:%s: force stripe size %d for reshape\n",
ad5b0f76
SL		 7068 mdname(mddev), conf->min_nr_stripes);
7069 }
edbe83ab 7070 memory = conf->min_nr_stripes * (sizeof(struct stripe_head) +
5e5e3e78 7071 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
4bda556a 7072 atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS);
edbe83ab 7073 if (grow_stripes(conf, conf->min_nr_stripes)) {
cc6167b4
N		 7074 pr_warn("md/raid:%s: couldn't allocate %dkB for buffers\n",
7075 mdname(mddev), memory);
91adb564
N		 7076 goto abort;
7077 } else
cc6167b4 7078 pr_debug("md/raid:%s: allocated %dkB\n", mdname(mddev), memory);
edbe83ab
N		 7079 /*
7080 * Losing a stripe head costs more than the time to refill it,
7081 * it reduces the queue depth and so can hurt throughput.
7082 * So set it rather large, scaled by number of devices.
7083 */
7084 conf->shrinker.seeks = DEFAULT_SEEKS * conf->raid_disks * 4;
7085 conf->shrinker.scan_objects = raid5_cache_scan;
7086 conf->shrinker.count_objects = raid5_cache_count;
7087 conf->shrinker.batch = 128;
7088 conf->shrinker.flags = 0;
6a0f53ff 7089 if (register_shrinker(&conf->shrinker)) {
cc6167b4
N		 7090 pr_warn("md/raid:%s: couldn't register shrinker.\n",
7091 mdname(mddev));
6a0f53ff
CY		 7092 goto abort;
7093 }
1da177e4 7094
0232605d
N		 7095 sprintf(pers_name, "raid%d", mddev->new_level);
7096 conf->thread = md_register_thread(raid5d, mddev, pers_name);
91adb564 7097 if (!conf->thread) {
cc6167b4
N		 7098 pr_warn("md/raid:%s: couldn't allocate thread.\n",
7099 mdname(mddev));
16a53ecc
N		 7100 goto abort;
7101 }
91adb564
N		 7102
7103 return conf;
7104
7105 abort:
7106 if (conf) {
95fc17aa 7107 free_conf(conf);
91adb564
N		 7108 return ERR_PTR(-EIO);
7109 } else
7110 return ERR_PTR(-ENOMEM);
7111}
7112
c148ffdc
N		 7113static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
7114{
7115 switch (algo) {
7116 case ALGORITHM_PARITY_0:
7117 if (raid_disk < max_degraded)
7118 return 1;
7119 break;
7120 case ALGORITHM_PARITY_N:
7121 if (raid_disk >= raid_disks - max_degraded)
7122 return 1;
7123 break;
7124 case ALGORITHM_PARITY_0_6:
f72ffdd6 7125 if (raid_disk == 0 ||
c148ffdc
N		 7126 raid_disk == raid_disks - 1)
7127 return 1;
7128 break;
7129 case ALGORITHM_LEFT_ASYMMETRIC_6:
7130 case ALGORITHM_RIGHT_ASYMMETRIC_6:
7131 case ALGORITHM_LEFT_SYMMETRIC_6:
7132 case ALGORITHM_RIGHT_SYMMETRIC_6:
7133 if (raid_disk == raid_disks - 1)
7134 return 1;
7135 }
7136 return 0;
7137}
7138
849674e4 7139static int raid5_run(struct mddev *mddev)
91adb564 7140{
d1688a6d 7141 struct r5conf *conf;
9f7c2220 7142 int working_disks = 0;
c148ffdc 7143 int dirty_parity_disks = 0;
3cb03002 7144 struct md_rdev *rdev;
713cf5a6 7145 struct md_rdev *journal_dev = NULL;
c148ffdc 7146 sector_t reshape_offset = 0;
17045f52 7147 int i;
b5254dd5
N		 7148 long long min_offset_diff = 0;
7149 int first = 1;
91adb564 7150
a415c0f1
N		 7151 if (mddev_init_writes_pending(mddev) < 0)
7152 return -ENOMEM;
7153
8c6ac868 7154 if (mddev->recovery_cp != MaxSector)
cc6167b4
N		 7155 pr_notice("md/raid:%s: not clean -- starting background reconstruction\n",
7156 mdname(mddev));
b5254dd5
N		 7157
7158 rdev_for_each(rdev, mddev) {
7159 long long diff;
713cf5a6 7160
f2076e7d 7161 if (test_bit(Journal, &rdev->flags)) {
713cf5a6 7162 journal_dev = rdev;
f2076e7d
SL		 7163 continue;
7164 }
b5254dd5
N		 7165 if (rdev->raid_disk < 0)
7166 continue;
7167 diff = (rdev->new_data_offset - rdev->data_offset);
7168 if (first) {
7169 min_offset_diff = diff;
7170 first = 0;
7171 } else if (mddev->reshape_backwards &&
7172 diff < min_offset_diff)
7173 min_offset_diff = diff;
7174 else if (!mddev->reshape_backwards &&
7175 diff > min_offset_diff)
7176 min_offset_diff = diff;
7177 }
7178
230b55fa
N		 7179 if ((test_bit(MD_HAS_JOURNAL, &mddev->flags) || journal_dev) &&
7180 (mddev->bitmap_info.offset || mddev->bitmap_info.file)) {
7181 pr_notice("md/raid:%s: array cannot have both journal and bitmap\n",
7182 mdname(mddev));
7183 return -EINVAL;
7184 }
7185
91adb564
N		 7186 if (mddev->reshape_position != MaxSector) {
7187 /* Check that we can continue the reshape.
b5254dd5
N		 7188 * Difficulties arise if the stripe we would write to
7189 * next is at or after the stripe we would read from next.
7190 * For a reshape that changes the number of devices, this
7191 * is only possible for a very short time, and mdadm makes
7192 * sure that time appears to have past before assembling
7193 * the array. So we fail if that time hasn't passed.
7194 * For a reshape that keeps the number of devices the same
7195 * mdadm must be monitoring the reshape can keeping the
7196 * critical areas read-only and backed up. It will start
7197 * the array in read-only mode, so we check for that.
91adb564
N		 7198 */
7199 sector_t here_new, here_old;
7200 int old_disks;
18b00334 7201 int max_degraded = (mddev->level == 6 ? 2 : 1);
05256d98
N		 7202 int chunk_sectors;
7203 int new_data_disks;
91adb564 7204
713cf5a6 7205 if (journal_dev) {
cc6167b4
N		 7206 pr_warn("md/raid:%s: don't support reshape with journal - aborting.\n",
7207 mdname(mddev));
713cf5a6
SL		 7208 return -EINVAL;
7209 }
7210
88ce4930 7211 if (mddev->new_level != mddev->level) {
cc6167b4
N		 7212 pr_warn("md/raid:%s: unsupported reshape required - aborting.\n",
7213 mdname(mddev));
91adb564
N		 7214 return -EINVAL;
7215 }
91adb564
N		 7216 old_disks = mddev->raid_disks - mddev->delta_disks;
7217 /* reshape_position must be on a new-stripe boundary, and one
7218 * further up in new geometry must map after here in old
7219 * geometry.
05256d98
N		 7220 * If the chunk sizes are different, then as we perform reshape
7221 * in units of the largest of the two, reshape_position needs
7222 * be a multiple of the largest chunk size times new data disks.
91adb564
N		 7223 */
7224 here_new = mddev->reshape_position;
05256d98
N		 7225 chunk_sectors = max(mddev->chunk_sectors, mddev->new_chunk_sectors);
7226 new_data_disks = mddev->raid_disks - max_degraded;
7227 if (sector_div(here_new, chunk_sectors * new_data_disks)) {
cc6167b4
N		 7228 pr_warn("md/raid:%s: reshape_position not on a stripe boundary\n",
7229 mdname(mddev));
91adb564
N		 7230 return -EINVAL;
7231 }
05256d98 7232 reshape_offset = here_new * chunk_sectors;
91adb564
N		 7233 /* here_new is the stripe we will write to */
7234 here_old = mddev->reshape_position;
05256d98 7235 sector_div(here_old, chunk_sectors * (old_disks-max_degraded));
91adb564
N		 7236 /* here_old is the first stripe that we might need to read
7237 * from */
67ac6011
N		 7238 if (mddev->delta_disks == 0) {
7239 /* We cannot be sure it is safe to start an in-place
b5254dd5 7240 * reshape. It is only safe if user-space is monitoring
67ac6011
N		 7241 * and taking constant backups.
7242 * mdadm always starts a situation like this in
7243 * readonly mode so it can take control before
7244 * allowing any writes. So just check for that.
7245 */
b5254dd5
N		 7246 if (abs(min_offset_diff) >= mddev->chunk_sectors &&
7247 abs(min_offset_diff) >= mddev->new_chunk_sectors)
7248 /* not really in-place - so OK */;
7249 else if (mddev->ro == 0) {
cc6167b4
N		 7250 pr_warn("md/raid:%s: in-place reshape must be started in read-only mode - aborting\n",
7251 mdname(mddev));
67ac6011
N		 7252 return -EINVAL;
7253 }
2c810cdd 7254 } else if (mddev->reshape_backwards
05256d98
N		 7255 ? (here_new * chunk_sectors + min_offset_diff <=
7256 here_old * chunk_sectors)
7257 : (here_new * chunk_sectors >=
7258 here_old * chunk_sectors + (-min_offset_diff))) {
91adb564 7259 /* Reading from the same stripe as writing to - bad */
cc6167b4
N		 7260 pr_warn("md/raid:%s: reshape_position too early for auto-recovery - aborting.\n",
7261 mdname(mddev));
91adb564
N		 7262 return -EINVAL;
7263 }
cc6167b4 7264 pr_debug("md/raid:%s: reshape will continue\n", mdname(mddev));
91adb564
N		 7265 /* OK, we should be able to continue; */
7266 } else {
7267 BUG_ON(mddev->level != mddev->new_level);
7268 BUG_ON(mddev->layout != mddev->new_layout);
664e7c41 7269 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
91adb564 7270 BUG_ON(mddev->delta_disks != 0);
1da177e4 7271 }
91adb564 7272
3418d036
AP		 7273 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) &&
7274 test_bit(MD_HAS_PPL, &mddev->flags)) {
7275 pr_warn("md/raid:%s: using journal device and PPL not allowed - disabling PPL\n",
7276 mdname(mddev));
7277 clear_bit(MD_HAS_PPL, &mddev->flags);
ddc08823 7278 clear_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags);
3418d036
AP		 7279 }
7280
245f46c2
N		 7281 if (mddev->private == NULL)
7282 conf = setup_conf(mddev);
7283 else
7284 conf = mddev->private;
7285
91adb564
N		 7286 if (IS_ERR(conf))
7287 return PTR_ERR(conf);
7288
486b0f7b
SL		 7289 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
7290 if (!journal_dev) {
cc6167b4
N		 7291 pr_warn("md/raid:%s: journal disk is missing, force array readonly\n",
7292 mdname(mddev));
486b0f7b
SL		 7293 mddev->ro = 1;
7294 set_disk_ro(mddev->gendisk, 1);
7295 } else if (mddev->recovery_cp == MaxSector)
7296 set_bit(MD_JOURNAL_CLEAN, &mddev->flags);
7dde2ad3
SL		 7297 }
7298
b5254dd5 7299 conf->min_offset_diff = min_offset_diff;
91adb564
N		 7300 mddev->thread = conf->thread;
7301 conf->thread = NULL;
7302 mddev->private = conf;
7303
17045f52
N		 7304 for (i = 0; i < conf->raid_disks && conf->previous_raid_disks;
7305 i++) {
7306 rdev = conf->disks[i].rdev;
7307 if (!rdev && conf->disks[i].replacement) {
7308 /* The replacement is all we have yet */
7309 rdev = conf->disks[i].replacement;
7310 conf->disks[i].replacement = NULL;
7311 clear_bit(Replacement, &rdev->flags);
7312 conf->disks[i].rdev = rdev;
7313 }
7314 if (!rdev)
c148ffdc 7315 continue;
17045f52
N		 7316 if (conf->disks[i].replacement &&
7317 conf->reshape_progress != MaxSector) {
7318 /* replacements and reshape simply do not mix. */
cc6167b4 7319 pr_warn("md: cannot handle concurrent replacement and reshape.\n");
17045f52
N		 7320 goto abort;
7321 }
2f115882 7322 if (test_bit(In_sync, &rdev->flags)) {
91adb564 7323 working_disks++;
2f115882
N		 7324 continue;
7325 }
c148ffdc
N		 7326 /* This disc is not fully in-sync. However if it
7327 * just stored parity (beyond the recovery_offset),
7328 * when we don't need to be concerned about the
7329 * array being dirty.
7330 * When reshape goes 'backwards', we never have
7331 * partially completed devices, so we only need
7332 * to worry about reshape going forwards.
7333 */
7334 /* Hack because v0.91 doesn't store recovery_offset properly. */
7335 if (mddev->major_version == 0 &&
7336 mddev->minor_version > 90)
7337 rdev->recovery_offset = reshape_offset;
5026d7a9 7338
c148ffdc
N		 7339 if (rdev->recovery_offset < reshape_offset) {
7340 /* We need to check old and new layout */
7341 if (!only_parity(rdev->raid_disk,
7342 conf->algorithm,
7343 conf->raid_disks,
7344 conf->max_degraded))
7345 continue;
7346 }
7347 if (!only_parity(rdev->raid_disk,
7348 conf->prev_algo,
7349 conf->previous_raid_disks,
7350 conf->max_degraded))
7351 continue;
7352 dirty_parity_disks++;
7353 }
91adb564 7354
17045f52
N		 7355 /*
7356 * 0 for a fully functional array, 1 or 2 for a degraded array.
7357 */
2e38a37f 7358 mddev->degraded = raid5_calc_degraded(conf);
91adb564 7359
674806d6 7360 if (has_failed(conf)) {
cc6167b4 7361 pr_crit("md/raid:%s: not enough operational devices (%d/%d failed)\n",
02c2de8c 7362 mdname(mddev), mddev->degraded, conf->raid_disks);
1da177e4
LT		 7363 goto abort;
7364 }
7365
91adb564 7366 /* device size must be a multiple of chunk size */
9d8f0363 7367 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
91adb564
N		 7368 mddev->resync_max_sectors = mddev->dev_sectors;
7369
c148ffdc 7370 if (mddev->degraded > dirty_parity_disks &&
1da177e4 7371 mddev->recovery_cp != MaxSector) {
4536bf9b
AP		 7372 if (test_bit(MD_HAS_PPL, &mddev->flags))
7373 pr_crit("md/raid:%s: starting dirty degraded array with PPL.\n",
7374 mdname(mddev));
7375 else if (mddev->ok_start_degraded)
cc6167b4
N		 7376 pr_crit("md/raid:%s: starting dirty degraded array - data corruption possible.\n",
7377 mdname(mddev));
6ff8d8ec 7378 else {
cc6167b4
N		 7379 pr_crit("md/raid:%s: cannot start dirty degraded array.\n",
7380 mdname(mddev));
6ff8d8ec
N		 7381 goto abort;
7382 }
1da177e4
LT		 7383 }
7384
cc6167b4
N		 7385 pr_info("md/raid:%s: raid level %d active with %d out of %d devices, algorithm %d\n",
7386 mdname(mddev), conf->level,
7387 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
7388 mddev->new_layout);
1da177e4
LT		 7389
7390 print_raid5_conf(conf);
7391
fef9c61f 7392 if (conf->reshape_progress != MaxSector) {
fef9c61f 7393 conf->reshape_safe = conf->reshape_progress;
f6705578
N		 7394 atomic_set(&conf->reshape_stripes, 0);
7395 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7396 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7397 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7398 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7399 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 7400 "reshape");
f6705578
N		 7401 }
7402
1da177e4 7403 /* Ok, everything is just fine now */
a64c876f
N		 7404 if (mddev->to_remove == &raid5_attrs_group)
7405 mddev->to_remove = NULL;
00bcb4ac
N		 7406 else if (mddev->kobj.sd &&
7407 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
cc6167b4
N		 7408 pr_warn("raid5: failed to create sysfs attributes for %s\n",
7409 mdname(mddev));
4a5add49 7410 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
7a5febe9 7411
4a5add49 7412 if (mddev->queue) {
9f7c2220 7413 int chunk_size;
4a5add49
N		 7414 /* read-ahead size must cover two whole stripes, which
7415 * is 2 * (datadisks) * chunksize where 'n' is the
7416 * number of raid devices
7417 */
7418 int data_disks = conf->previous_raid_disks - conf->max_degraded;
7419 int stripe = data_disks *
7420 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
dc3b17cc
JK		 7421 if (mddev->queue->backing_dev_info->ra_pages < 2 * stripe)
7422 mddev->queue->backing_dev_info->ra_pages = 2 * stripe;
91adb564 7423
9f7c2220
N		 7424 chunk_size = mddev->chunk_sectors << 9;
7425 blk_queue_io_min(mddev->queue, chunk_size);
7426 blk_queue_io_opt(mddev->queue, chunk_size *
7427 (conf->raid_disks - conf->max_degraded));
c78afc62 7428 mddev->queue->limits.raid_partial_stripes_expensive = 1;
620125f2
SL		 7429 /*
7430 * We can only discard a whole stripe. It doesn't make sense to
7431 * discard data disk but write parity disk
7432 */
7433 stripe = stripe * PAGE_SIZE;
4ac6875e
N		 7434 /* Round up to power of 2, as discard handling
7435 * currently assumes that */
7436 while ((stripe-1) & stripe)
7437 stripe = (stripe | (stripe-1)) + 1;
620125f2
SL		 7438 mddev->queue->limits.discard_alignment = stripe;
7439 mddev->queue->limits.discard_granularity = stripe;
e8d7c332 7440
5026d7a9 7441 blk_queue_max_write_same_sectors(mddev->queue, 0);
3deff1a7 7442 blk_queue_max_write_zeroes_sectors(mddev->queue, 0);
5026d7a9 7443
05616be5 7444 rdev_for_each(rdev, mddev) {
9f7c2220
N		 7445 disk_stack_limits(mddev->gendisk, rdev->bdev,
7446 rdev->data_offset << 9);
05616be5
N		 7447 disk_stack_limits(mddev->gendisk, rdev->bdev,
7448 rdev->new_data_offset << 9);
7449 }
620125f2 7450
48920ff2
CH		 7451 /*
7452 * zeroing is required, otherwise data
7453 * could be lost. Consider a scenario: discard a stripe
7454 * (the stripe could be inconsistent if
7455 * discard_zeroes_data is 0); write one disk of the
7456 * stripe (the stripe could be inconsistent again
7457 * depending on which disks are used to calculate
7458 * parity); the disk is broken; The stripe data of this
7459 * disk is lost.
7460 *
7461 * We only allow DISCARD if the sysadmin has confirmed that
7462 * only safe devices are in use by setting a module parameter.
7463 * A better idea might be to turn DISCARD into WRITE_ZEROES
7464 * requests, as that is required to be safe.
7465 */
7466 if (devices_handle_discard_safely &&
e7597e69
JS		 7467 mddev->queue->limits.max_discard_sectors >= (stripe >> 9) &&
7468 mddev->queue->limits.discard_granularity >= stripe)
8b904b5b 7469 blk_queue_flag_set(QUEUE_FLAG_DISCARD,
620125f2
SL		 7470 mddev->queue);
7471 else
8b904b5b 7472 blk_queue_flag_clear(QUEUE_FLAG_DISCARD,
620125f2 7473 mddev->queue);
1dffdddd
SL		 7474
7475 blk_queue_max_hw_sectors(mddev->queue, UINT_MAX);
9f7c2220 7476 }
23032a0e 7477
845b9e22 7478 if (log_init(conf, journal_dev, raid5_has_ppl(conf)))
ff875738 7479 goto abort;
5c7e81c3 7480
1da177e4
LT		 7481 return 0;
7482abort:
01f96c0a 7483 md_unregister_thread(&mddev->thread);
e4f869d9
N		 7484 print_raid5_conf(conf);
7485 free_conf(conf);
1da177e4 7486 mddev->private = NULL;
cc6167b4 7487 pr_warn("md/raid:%s: failed to run raid set.\n", mdname(mddev));
1da177e4
LT		 7488 return -EIO;
7489}
7490
afa0f557 7491static void raid5_free(struct mddev *mddev, void *priv)
1da177e4 7492{
afa0f557 7493 struct r5conf *conf = priv;
1da177e4 7494
95fc17aa 7495 free_conf(conf);
a64c876f 7496 mddev->to_remove = &raid5_attrs_group;
1da177e4
LT		 7497}
7498
849674e4 7499static void raid5_status(struct seq_file *seq, struct mddev *mddev)
1da177e4 7500{
d1688a6d 7501 struct r5conf *conf = mddev->private;
1da177e4
LT		 7502 int i;
7503
9d8f0363 7504 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
3cb5edf4 7505 conf->chunk_sectors / 2, mddev->layout);
02c2de8c 7506 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
5fd13351
N		 7507 rcu_read_lock();
7508 for (i = 0; i < conf->raid_disks; i++) {
7509 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
7510 seq_printf (seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
7511 }
7512 rcu_read_unlock();
1da177e4 7513 seq_printf (seq, "]");
1da177e4
LT		 7514}
7515
d1688a6d 7516static void print_raid5_conf (struct r5conf *conf)
1da177e4
LT		 7517{
7518 int i;
7519 struct disk_info *tmp;
7520
cc6167b4 7521 pr_debug("RAID conf printout:\n");
1da177e4 7522 if (!conf) {
cc6167b4 7523 pr_debug("(conf==NULL)\n");
1da177e4
LT		 7524 return;
7525 }
cc6167b4 7526 pr_debug(" --- level:%d rd:%d wd:%d\n", conf->level,
0c55e022
N		 7527 conf->raid_disks,
7528 conf->raid_disks - conf->mddev->degraded);
1da177e4
LT		 7529
7530 for (i = 0; i < conf->raid_disks; i++) {
7531 char b[BDEVNAME_SIZE];
7532 tmp = conf->disks + i;
7533 if (tmp->rdev)
cc6167b4 7534 pr_debug(" disk %d, o:%d, dev:%s\n",
0c55e022
N		 7535 i, !test_bit(Faulty, &tmp->rdev->flags),
7536 bdevname(tmp->rdev->bdev, b));
1da177e4
LT		 7537 }
7538}
7539
fd01b88c 7540static int raid5_spare_active(struct mddev *mddev)
1da177e4
LT		 7541{
7542 int i;
d1688a6d 7543 struct r5conf *conf = mddev->private;
1da177e4 7544 struct disk_info *tmp;
6b965620
N		 7545 int count = 0;
7546 unsigned long flags;
1da177e4
LT		 7547
7548 for (i = 0; i < conf->raid_disks; i++) {
7549 tmp = conf->disks + i;
dd054fce
N		 7550 if (tmp->replacement
7551 && tmp->replacement->recovery_offset == MaxSector
7552 && !test_bit(Faulty, &tmp->replacement->flags)
7553 && !test_and_set_bit(In_sync, &tmp->replacement->flags)) {
7554 /* Replacement has just become active. */
7555 if (!tmp->rdev
7556 || !test_and_clear_bit(In_sync, &tmp->rdev->flags))
7557 count++;
7558 if (tmp->rdev) {
7559 /* Replaced device not technically faulty,
7560 * but we need to be sure it gets removed
7561 * and never re-added.
7562 */
7563 set_bit(Faulty, &tmp->rdev->flags);
7564 sysfs_notify_dirent_safe(
7565 tmp->rdev->sysfs_state);
7566 }
7567 sysfs_notify_dirent_safe(tmp->replacement->sysfs_state);
7568 } else if (tmp->rdev
70fffd0b 7569 && tmp->rdev->recovery_offset == MaxSector
b2d444d7 7570 && !test_bit(Faulty, &tmp->rdev->flags)
c04be0aa 7571 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
6b965620 7572 count++;
43c73ca4 7573 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
1da177e4
LT		 7574 }
7575 }
6b965620 7576 spin_lock_irqsave(&conf->device_lock, flags);
2e38a37f 7577 mddev->degraded = raid5_calc_degraded(conf);
6b965620 7578 spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4 7579 print_raid5_conf(conf);
6b965620 7580 return count;
1da177e4
LT		 7581}
7582
b8321b68 7583static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7584{
d1688a6d 7585 struct r5conf *conf = mddev->private;
1da177e4 7586 int err = 0;
b8321b68 7587 int number = rdev->raid_disk;
657e3e4d 7588 struct md_rdev **rdevp;
1da177e4
LT		 7589 struct disk_info *p = conf->disks + number;
7590
7591 print_raid5_conf(conf);
f6b6ec5c 7592 if (test_bit(Journal, &rdev->flags) && conf->log) {
c2bb6242 7593 /*
f6b6ec5c
SL		 7594 * we can't wait pending write here, as this is called in
7595 * raid5d, wait will deadlock.
84dd97a6
N		 7596 * neilb: there is no locking about new writes here,
7597 * so this cannot be safe.
c2bb6242 7598 */
70d466f7
SL		 7599 if (atomic_read(&conf->active_stripes) ||
7600 atomic_read(&conf->r5c_cached_full_stripes) ||
7601 atomic_read(&conf->r5c_cached_partial_stripes)) {
f6b6ec5c 7602 return -EBUSY;
84dd97a6 7603 }
ff875738 7604 log_exit(conf);
f6b6ec5c 7605 return 0;
c2bb6242 7606 }
657e3e4d
N		 7607 if (rdev == p->rdev)
7608 rdevp = &p->rdev;
7609 else if (rdev == p->replacement)
7610 rdevp = &p->replacement;
7611 else
7612 return 0;
7613
7614 if (number >= conf->raid_disks &&
7615 conf->reshape_progress == MaxSector)
7616 clear_bit(In_sync, &rdev->flags);
7617
7618 if (test_bit(In_sync, &rdev->flags) ||
7619 atomic_read(&rdev->nr_pending)) {
7620 err = -EBUSY;
7621 goto abort;
7622 }
7623 /* Only remove non-faulty devices if recovery
7624 * isn't possible.
7625 */
7626 if (!test_bit(Faulty, &rdev->flags) &&
7627 mddev->recovery_disabled != conf->recovery_disabled &&
7628 !has_failed(conf) &&
dd054fce 7629 (!p->replacement || p->replacement == rdev) &&
657e3e4d
N		 7630 number < conf->raid_disks) {
7631 err = -EBUSY;
7632 goto abort;
7633 }
7634 *rdevp = NULL;
d787be40
N		 7635 if (!test_bit(RemoveSynchronized, &rdev->flags)) {
7636 synchronize_rcu();
7637 if (atomic_read(&rdev->nr_pending)) {
7638 /* lost the race, try later */
7639 err = -EBUSY;
7640 *rdevp = rdev;
7641 }
7642 }
6358c239
AP		 7643 if (!err) {
7644 err = log_modify(conf, rdev, false);
7645 if (err)
7646 goto abort;
7647 }
d787be40 7648 if (p->replacement) {
dd054fce
N		 7649 /* We must have just cleared 'rdev' */
7650 p->rdev = p->replacement;
7651 clear_bit(Replacement, &p->replacement->flags);
7652 smp_mb(); /* Make sure other CPUs may see both as identical
7653 * but will never see neither - if they are careful
7654 */
7655 p->replacement = NULL;
6358c239
AP		 7656
7657 if (!err)
7658 err = log_modify(conf, p->rdev, true);
e5bc9c3c
GJ		 7659 }
7660
7661 clear_bit(WantReplacement, &rdev->flags);
1da177e4
LT		 7662abort:
7663
7664 print_raid5_conf(conf);
7665 return err;
7666}
7667
fd01b88c 7668static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7669{
d1688a6d 7670 struct r5conf *conf = mddev->private;
199050ea 7671 int err = -EEXIST;
1da177e4
LT		 7672 int disk;
7673 struct disk_info *p;
6c2fce2e
NB		 7674 int first = 0;
7675 int last = conf->raid_disks - 1;
1da177e4 7676
f6b6ec5c 7677 if (test_bit(Journal, &rdev->flags)) {
f6b6ec5c
SL		 7678 if (conf->log)
7679 return -EBUSY;
7680
7681 rdev->raid_disk = 0;
7682 /*
7683 * The array is in readonly mode if journal is missing, so no
7684 * write requests running. We should be safe
7685 */
845b9e22 7686 log_init(conf, rdev, false);
f6b6ec5c
SL		 7687 return 0;
7688 }
7f0da59b
N		 7689 if (mddev->recovery_disabled == conf->recovery_disabled)
7690 return -EBUSY;
7691
dc10c643 7692 if (rdev->saved_raid_disk < 0 && has_failed(conf))
1da177e4 7693 /* no point adding a device */
199050ea 7694 return -EINVAL;
1da177e4 7695
6c2fce2e
NB		 7696 if (rdev->raid_disk >= 0)
7697 first = last = rdev->raid_disk;
1da177e4
LT		 7698
7699 /*
16a53ecc
N		 7700 * find the disk ... but prefer rdev->saved_raid_disk
7701 * if possible.
1da177e4 7702 */
16a53ecc 7703 if (rdev->saved_raid_disk >= 0 &&
6c2fce2e 7704 rdev->saved_raid_disk >= first &&
16a53ecc 7705 conf->disks[rdev->saved_raid_disk].rdev == NULL)
5cfb22a1
N		 7706 first = rdev->saved_raid_disk;
7707
7708 for (disk = first; disk <= last; disk++) {
7bfec5f3
N		 7709 p = conf->disks + disk;
7710 if (p->rdev == NULL) {
b2d444d7 7711 clear_bit(In_sync, &rdev->flags);
1da177e4 7712 rdev->raid_disk = disk;
72626685
N		 7713 if (rdev->saved_raid_disk != disk)
7714 conf->fullsync = 1;
d6065f7b 7715 rcu_assign_pointer(p->rdev, rdev);
6358c239
AP		 7716
7717 err = log_modify(conf, rdev, true);
7718
5cfb22a1 7719 goto out;
1da177e4 7720 }
5cfb22a1
N		 7721 }
7722 for (disk = first; disk <= last; disk++) {
7723 p = conf->disks + disk;
7bfec5f3
N		 7724 if (test_bit(WantReplacement, &p->rdev->flags) &&
7725 p->replacement == NULL) {
7726 clear_bit(In_sync, &rdev->flags);
7727 set_bit(Replacement, &rdev->flags);
7728 rdev->raid_disk = disk;
7729 err = 0;
7730 conf->fullsync = 1;
7731 rcu_assign_pointer(p->replacement, rdev);
7732 break;
7733 }
7734 }
5cfb22a1 7735out:
1da177e4 7736 print_raid5_conf(conf);
199050ea 7737 return err;
1da177e4
LT		 7738}
7739
fd01b88c 7740static int raid5_resize(struct mddev *mddev, sector_t sectors)
1da177e4
LT		 7741{
7742 /* no resync is happening, and there is enough space
7743 * on all devices, so we can resize.
7744 * We need to make sure resync covers any new space.
7745 * If the array is shrinking we should possibly wait until
7746 * any io in the removed space completes, but it hardly seems
7747 * worth it.
7748 */
a4a6125a 7749 sector_t newsize;
3cb5edf4
N		 7750 struct r5conf *conf = mddev->private;
7751
e254de6b 7752 if (raid5_has_log(conf) || raid5_has_ppl(conf))
713cf5a6 7753 return -EINVAL;
3cb5edf4 7754 sectors &= ~((sector_t)conf->chunk_sectors - 1);
a4a6125a
N		 7755 newsize = raid5_size(mddev, sectors, mddev->raid_disks);
7756 if (mddev->external_size &&
7757 mddev->array_sectors > newsize)
b522adcd 7758 return -EINVAL;
a4a6125a 7759 if (mddev->bitmap) {
e64e4018 7760 int ret = md_bitmap_resize(mddev->bitmap, sectors, 0, 0);
a4a6125a
N		 7761 if (ret)
7762 return ret;
7763 }
7764 md_set_array_sectors(mddev, newsize);
b098636c
N		 7765 if (sectors > mddev->dev_sectors &&
7766 mddev->recovery_cp > mddev->dev_sectors) {
58c0fed4 7767 mddev->recovery_cp = mddev->dev_sectors;
1da177e4
LT		 7768 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7769 }
58c0fed4 7770 mddev->dev_sectors = sectors;
4b5c7ae8 7771 mddev->resync_max_sectors = sectors;
1da177e4
LT		 7772 return 0;
7773}
7774
fd01b88c 7775static int check_stripe_cache(struct mddev *mddev)
01ee22b4
N		 7776{
7777 /* Can only proceed if there are plenty of stripe_heads.
7778 * We need a minimum of one full stripe,, and for sensible progress
7779 * it is best to have about 4 times that.
7780 * If we require 4 times, then the default 256 4K stripe_heads will
7781 * allow for chunk sizes up to 256K, which is probably OK.
7782 * If the chunk size is greater, user-space should request more
7783 * stripe_heads first.
7784 */
d1688a6d 7785 struct r5conf *conf = mddev->private;
01ee22b4 7786 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
edbe83ab 7787 > conf->min_nr_stripes ||
01ee22b4 7788 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
edbe83ab 7789 > conf->min_nr_stripes) {
cc6167b4
N		 7790 pr_warn("md/raid:%s: reshape: not enough stripes. Needed %lu\n",
7791 mdname(mddev),
7792 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
7793 / STRIPE_SIZE)*4);
01ee22b4
N		 7794 return 0;
7795 }
7796 return 1;
7797}
7798
fd01b88c 7799static int check_reshape(struct mddev *mddev)
29269553 7800{
d1688a6d 7801 struct r5conf *conf = mddev->private;
29269553 7802
e254de6b 7803 if (raid5_has_log(conf) || raid5_has_ppl(conf))
713cf5a6 7804 return -EINVAL;
88ce4930
N		 7805 if (mddev->delta_disks == 0 &&
7806 mddev->new_layout == mddev->layout &&
664e7c41 7807 mddev->new_chunk_sectors == mddev->chunk_sectors)
50ac168a 7808 return 0; /* nothing to do */
674806d6 7809 if (has_failed(conf))
ec32a2bd 7810 return -EINVAL;
fdcfbbb6 7811 if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) {
ec32a2bd
N		 7812 /* We might be able to shrink, but the devices must
7813 * be made bigger first.
7814 * For raid6, 4 is the minimum size.
7815 * Otherwise 2 is the minimum
7816 */
7817 int min = 2;
7818 if (mddev->level == 6)
7819 min = 4;
7820 if (mddev->raid_disks + mddev->delta_disks < min)
7821 return -EINVAL;
7822 }
29269553 7823
01ee22b4 7824 if (!check_stripe_cache(mddev))
29269553 7825 return -ENOSPC;
29269553 7826
738a2738
N		 7827 if (mddev->new_chunk_sectors > mddev->chunk_sectors ||
7828 mddev->delta_disks > 0)
7829 if (resize_chunks(conf,
7830 conf->previous_raid_disks
7831 + max(0, mddev->delta_disks),
7832 max(mddev->new_chunk_sectors,
7833 mddev->chunk_sectors)
7834 ) < 0)
7835 return -ENOMEM;
845b9e22
AP		 7836
7837 if (conf->previous_raid_disks + mddev->delta_disks <= conf->pool_size)
7838 return 0; /* never bother to shrink */
e56108d6
N		 7839 return resize_stripes(conf, (conf->previous_raid_disks
7840 + mddev->delta_disks));
63c70c4f
N		 7841}
7842
fd01b88c 7843static int raid5_start_reshape(struct mddev *mddev)
63c70c4f 7844{
d1688a6d 7845 struct r5conf *conf = mddev->private;
3cb03002 7846 struct md_rdev *rdev;
63c70c4f 7847 int spares = 0;
c04be0aa 7848 unsigned long flags;
63c70c4f 7849
f416885e 7850 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
63c70c4f
N		 7851 return -EBUSY;
7852
01ee22b4
N		 7853 if (!check_stripe_cache(mddev))
7854 return -ENOSPC;
7855
30b67645
N		 7856 if (has_failed(conf))
7857 return -EINVAL;
7858
c6563a8c 7859 rdev_for_each(rdev, mddev) {
469518a3
N		 7860 if (!test_bit(In_sync, &rdev->flags)
7861 && !test_bit(Faulty, &rdev->flags))
29269553 7862 spares++;
c6563a8c 7863 }
63c70c4f 7864
f416885e 7865 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
29269553
N		 7866 /* Not enough devices even to make a degraded array
7867 * of that size
7868 */
7869 return -EINVAL;
7870
ec32a2bd
N		 7871 /* Refuse to reduce size of the array. Any reductions in
7872 * array size must be through explicit setting of array_size
7873 * attribute.
7874 */
7875 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
7876 < mddev->array_sectors) {
cc6167b4
N		 7877 pr_warn("md/raid:%s: array size must be reduced before number of disks\n",
7878 mdname(mddev));
ec32a2bd
N		 7879 return -EINVAL;
7880 }
7881
f6705578 7882 atomic_set(&conf->reshape_stripes, 0);
29269553 7883 spin_lock_irq(&conf->device_lock);
c46501b2 7884 write_seqcount_begin(&conf->gen_lock);
29269553 7885 conf->previous_raid_disks = conf->raid_disks;
63c70c4f 7886 conf->raid_disks += mddev->delta_disks;
09c9e5fa
AN		 7887 conf->prev_chunk_sectors = conf->chunk_sectors;
7888 conf->chunk_sectors = mddev->new_chunk_sectors;
88ce4930
N		 7889 conf->prev_algo = conf->algorithm;
7890 conf->algorithm = mddev->new_layout;
05616be5
N		 7891 conf->generation++;
7892 /* Code that selects data_offset needs to see the generation update
7893 * if reshape_progress has been set - so a memory barrier needed.
7894 */
7895 smp_mb();
2c810cdd 7896 if (mddev->reshape_backwards)
fef9c61f
N		 7897 conf->reshape_progress = raid5_size(mddev, 0, 0);
7898 else
7899 conf->reshape_progress = 0;
7900 conf->reshape_safe = conf->reshape_progress;
c46501b2 7901 write_seqcount_end(&conf->gen_lock);
29269553
N		 7902 spin_unlock_irq(&conf->device_lock);
7903
4d77e3ba
N		 7904 /* Now make sure any requests that proceeded on the assumption
7905 * the reshape wasn't running - like Discard or Read - have
7906 * completed.
7907 */
7908 mddev_suspend(mddev);
7909 mddev_resume(mddev);
7910
29269553
N		 7911 /* Add some new drives, as many as will fit.
7912 * We know there are enough to make the newly sized array work.
3424bf6a
N		 7913 * Don't add devices if we are reducing the number of
7914 * devices in the array. This is because it is not possible
7915 * to correctly record the "partially reconstructed" state of
7916 * such devices during the reshape and confusion could result.
29269553 7917 */
87a8dec9 7918 if (mddev->delta_disks >= 0) {
dafb20fa 7919 rdev_for_each(rdev, mddev)
87a8dec9
N		 7920 if (rdev->raid_disk < 0 &&
7921 !test_bit(Faulty, &rdev->flags)) {
7922 if (raid5_add_disk(mddev, rdev) == 0) {
87a8dec9 7923 if (rdev->raid_disk
9d4c7d87 7924 >= conf->previous_raid_disks)
87a8dec9 7925 set_bit(In_sync, &rdev->flags);
9d4c7d87 7926 else
87a8dec9 7927 rdev->recovery_offset = 0;
36fad858
NK		 7928
7929 if (sysfs_link_rdev(mddev, rdev))
87a8dec9 7930 /* Failure here is OK */;
50da0840 7931 }
87a8dec9
N		 7932 } else if (rdev->raid_disk >= conf->previous_raid_disks
7933 && !test_bit(Faulty, &rdev->flags)) {
7934 /* This is a spare that was manually added */
7935 set_bit(In_sync, &rdev->flags);
87a8dec9 7936 }
29269553 7937
87a8dec9
N		 7938 /* When a reshape changes the number of devices,
7939 * ->degraded is measured against the larger of the
7940 * pre and post number of devices.
7941 */
ec32a2bd 7942 spin_lock_irqsave(&conf->device_lock, flags);
2e38a37f 7943 mddev->degraded = raid5_calc_degraded(conf);
ec32a2bd
N		 7944 spin_unlock_irqrestore(&conf->device_lock, flags);
7945 }
63c70c4f 7946 mddev->raid_disks = conf->raid_disks;
e516402c 7947 mddev->reshape_position = conf->reshape_progress;
2953079c 7948 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
f6705578 7949
29269553
N		 7950 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7951 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
ea358cd0 7952 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
29269553
N		 7953 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7954 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7955 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 7956 "reshape");
29269553
N		 7957 if (!mddev->sync_thread) {
7958 mddev->recovery = 0;
7959 spin_lock_irq(&conf->device_lock);
ba8805b9 7960 write_seqcount_begin(&conf->gen_lock);
29269553 7961 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
ba8805b9
N		 7962 mddev->new_chunk_sectors =
7963 conf->chunk_sectors = conf->prev_chunk_sectors;
7964 mddev->new_layout = conf->algorithm = conf->prev_algo;
05616be5
N		 7965 rdev_for_each(rdev, mddev)
7966 rdev->new_data_offset = rdev->data_offset;
7967 smp_wmb();
ba8805b9 7968 conf->generation --;
fef9c61f 7969 conf->reshape_progress = MaxSector;
1e3fa9bd 7970 mddev->reshape_position = MaxSector;
ba8805b9 7971 write_seqcount_end(&conf->gen_lock);
29269553
N		 7972 spin_unlock_irq(&conf->device_lock);
7973 return -EAGAIN;
7974 }
c8f517c4 7975 conf->reshape_checkpoint = jiffies;
29269553
N		 7976 md_wakeup_thread(mddev->sync_thread);
7977 md_new_event(mddev);
7978 return 0;
7979}
29269553 7980
ec32a2bd
N		 7981/* This is called from the reshape thread and should make any
7982 * changes needed in 'conf'
7983 */
d1688a6d 7984static void end_reshape(struct r5conf *conf)
29269553 7985{
29269553 7986
f6705578 7987 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
db0505d3 7988 struct md_rdev *rdev;
f6705578 7989
f6705578 7990 spin_lock_irq(&conf->device_lock);
cea9c228 7991 conf->previous_raid_disks = conf->raid_disks;
b5d27718 7992 md_finish_reshape(conf->mddev);
05616be5 7993 smp_wmb();
fef9c61f 7994 conf->reshape_progress = MaxSector;
6cbd8148 7995 conf->mddev->reshape_position = MaxSector;
db0505d3
N		 7996 rdev_for_each(rdev, conf->mddev)
7997 if (rdev->raid_disk >= 0 &&
7998 !test_bit(Journal, &rdev->flags) &&
7999 !test_bit(In_sync, &rdev->flags))
8000 rdev->recovery_offset = MaxSector;
f6705578 8001 spin_unlock_irq(&conf->device_lock);
b0f9ec04 8002 wake_up(&conf->wait_for_overlap);
16a53ecc
N		 8003
8004 /* read-ahead size must cover two whole stripes, which is
8005 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
8006 */
4a5add49 8007 if (conf->mddev->queue) {
cea9c228 8008 int data_disks = conf->raid_disks - conf->max_degraded;
09c9e5fa 8009 int stripe = data_disks * ((conf->chunk_sectors << 9)
cea9c228 8010 / PAGE_SIZE);
dc3b17cc
JK		 8011 if (conf->mddev->queue->backing_dev_info->ra_pages < 2 * stripe)
8012 conf->mddev->queue->backing_dev_info->ra_pages = 2 * stripe;
16a53ecc 8013 }
29269553 8014 }
29269553
N		 8015}
8016
ec32a2bd
N		 8017/* This is called from the raid5d thread with mddev_lock held.
8018 * It makes config changes to the device.
8019 */
fd01b88c 8020static void raid5_finish_reshape(struct mddev *mddev)
cea9c228 8021{
d1688a6d 8022 struct r5conf *conf = mddev->private;
cea9c228
N		 8023
8024 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8025
8876391e 8026 if (mddev->delta_disks <= 0) {
ec32a2bd 8027 int d;
908f4fbd 8028 spin_lock_irq(&conf->device_lock);
2e38a37f 8029 mddev->degraded = raid5_calc_degraded(conf);
908f4fbd 8030 spin_unlock_irq(&conf->device_lock);
ec32a2bd
N		 8031 for (d = conf->raid_disks ;
8032 d < conf->raid_disks - mddev->delta_disks;
1a67dde0 8033 d++) {
3cb03002 8034 struct md_rdev *rdev = conf->disks[d].rdev;
da7613b8
N		 8035 if (rdev)
8036 clear_bit(In_sync, &rdev->flags);
8037 rdev = conf->disks[d].replacement;
8038 if (rdev)
8039 clear_bit(In_sync, &rdev->flags);
1a67dde0 8040 }
cea9c228 8041 }
88ce4930 8042 mddev->layout = conf->algorithm;
09c9e5fa 8043 mddev->chunk_sectors = conf->chunk_sectors;
ec32a2bd
N		 8044 mddev->reshape_position = MaxSector;
8045 mddev->delta_disks = 0;
2c810cdd 8046 mddev->reshape_backwards = 0;
cea9c228
N		 8047 }
8048}
8049
b03e0ccb 8050static void raid5_quiesce(struct mddev *mddev, int quiesce)
72626685 8051{
d1688a6d 8052 struct r5conf *conf = mddev->private;
72626685 8053
b03e0ccb
N		 8054 if (quiesce) {
8055 /* stop all writes */
566c09c5 8056 lock_all_device_hash_locks_irq(conf);
64bd660b
N		 8057 /* '2' tells resync/reshape to pause so that all
8058 * active stripes can drain
8059 */
a39f7afd 8060 r5c_flush_cache(conf, INT_MAX);
64bd660b 8061 conf->quiesce = 2;
b1b46486 8062 wait_event_cmd(conf->wait_for_quiescent,
46031f9a
RBJ		 8063 atomic_read(&conf->active_stripes) == 0 &&
8064 atomic_read(&conf->active_aligned_reads) == 0,
566c09c5
SL		 8065 unlock_all_device_hash_locks_irq(conf),
8066 lock_all_device_hash_locks_irq(conf));
64bd660b 8067 conf->quiesce = 1;
566c09c5 8068 unlock_all_device_hash_locks_irq(conf);
64bd660b
N		 8069 /* allow reshape to continue */
8070 wake_up(&conf->wait_for_overlap);
b03e0ccb
N		 8071 } else {
8072 /* re-enable writes */
566c09c5 8073 lock_all_device_hash_locks_irq(conf);
72626685 8074 conf->quiesce = 0;
b1b46486 8075 wake_up(&conf->wait_for_quiescent);
e464eafd 8076 wake_up(&conf->wait_for_overlap);
566c09c5 8077 unlock_all_device_hash_locks_irq(conf);
72626685 8078 }
1532d9e8 8079 log_quiesce(conf, quiesce);
72626685 8080}
b15c2e57 8081
fd01b88c 8082static void *raid45_takeover_raid0(struct mddev *mddev, int level)
54071b38 8083{
e373ab10 8084 struct r0conf *raid0_conf = mddev->private;
d76c8420 8085 sector_t sectors;
54071b38 8086
f1b29bca 8087 /* for raid0 takeover only one zone is supported */
e373ab10 8088 if (raid0_conf->nr_strip_zones > 1) {
cc6167b4
N		 8089 pr_warn("md/raid:%s: cannot takeover raid0 with more than one zone.\n",
8090 mdname(mddev));
f1b29bca
DW		 8091 return ERR_PTR(-EINVAL);
8092 }
8093
e373ab10
N		 8094 sectors = raid0_conf->strip_zone[0].zone_end;
8095 sector_div(sectors, raid0_conf->strip_zone[0].nb_dev);
3b71bd93 8096 mddev->dev_sectors = sectors;
f1b29bca 8097 mddev->new_level = level;
54071b38
TM		 8098 mddev->new_layout = ALGORITHM_PARITY_N;
8099 mddev->new_chunk_sectors = mddev->chunk_sectors;
8100 mddev->raid_disks += 1;
8101 mddev->delta_disks = 1;
8102 /* make sure it will be not marked as dirty */
8103 mddev->recovery_cp = MaxSector;
8104
8105 return setup_conf(mddev);
8106}
8107
fd01b88c 8108static void *raid5_takeover_raid1(struct mddev *mddev)
d562b0c4
N		 8109{
8110 int chunksect;
6995f0b2 8111 void *ret;
d562b0c4
N		 8112
8113 if (mddev->raid_disks != 2 ||
8114 mddev->degraded > 1)
8115 return ERR_PTR(-EINVAL);
8116
8117 /* Should check if there are write-behind devices? */
8118
8119 chunksect = 64*2; /* 64K by default */
8120
8121 /* The array must be an exact multiple of chunksize */
8122 while (chunksect && (mddev->array_sectors & (chunksect-1)))
8123 chunksect >>= 1;
8124
8125 if ((chunksect<<9) < STRIPE_SIZE)
8126 /* array size does not allow a suitable chunk size */
8127 return ERR_PTR(-EINVAL);
8128
8129 mddev->new_level = 5;
8130 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
664e7c41 8131 mddev->new_chunk_sectors = chunksect;
d562b0c4 8132
6995f0b2 8133 ret = setup_conf(mddev);
32cd7cbb 8134 if (!IS_ERR(ret))
394ed8e4
SL		 8135 mddev_clear_unsupported_flags(mddev,
8136 UNSUPPORTED_MDDEV_FLAGS);
6995f0b2 8137 return ret;
d562b0c4
N		 8138}
8139
fd01b88c 8140static void *raid5_takeover_raid6(struct mddev *mddev)
fc9739c6
N		 8141{
8142 int new_layout;
8143
8144 switch (mddev->layout) {
8145 case ALGORITHM_LEFT_ASYMMETRIC_6:
8146 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
8147 break;
8148 case ALGORITHM_RIGHT_ASYMMETRIC_6:
8149 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
8150 break;
8151 case ALGORITHM_LEFT_SYMMETRIC_6:
8152 new_layout = ALGORITHM_LEFT_SYMMETRIC;
8153 break;
8154 case ALGORITHM_RIGHT_SYMMETRIC_6:
8155 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
8156 break;
8157 case ALGORITHM_PARITY_0_6:
8158 new_layout = ALGORITHM_PARITY_0;
8159 break;
8160 case ALGORITHM_PARITY_N:
8161 new_layout = ALGORITHM_PARITY_N;
8162 break;
8163 default:
8164 return ERR_PTR(-EINVAL);
8165 }
8166 mddev->new_level = 5;
8167 mddev->new_layout = new_layout;
8168 mddev->delta_disks = -1;
8169 mddev->raid_disks -= 1;
8170 return setup_conf(mddev);
8171}
8172
fd01b88c 8173static int raid5_check_reshape(struct mddev *mddev)
b3546035 8174{
88ce4930
N		 8175 /* For a 2-drive array, the layout and chunk size can be changed
8176 * immediately as not restriping is needed.
8177 * For larger arrays we record the new value - after validation
8178 * to be used by a reshape pass.
b3546035 8179 */
d1688a6d 8180 struct r5conf *conf = mddev->private;
597a711b 8181 int new_chunk = mddev->new_chunk_sectors;
b3546035 8182
597a711b 8183 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
b3546035
N		 8184 return -EINVAL;
8185 if (new_chunk > 0) {
0ba459d2 8186 if (!is_power_of_2(new_chunk))
b3546035 8187 return -EINVAL;
597a711b 8188 if (new_chunk < (PAGE_SIZE>>9))
b3546035 8189 return -EINVAL;
597a711b 8190 if (mddev->array_sectors & (new_chunk-1))
b3546035
N		 8191 /* not factor of array size */
8192 return -EINVAL;
8193 }
8194
8195 /* They look valid */
8196
88ce4930 8197 if (mddev->raid_disks == 2) {
597a711b
N		 8198 /* can make the change immediately */
8199 if (mddev->new_layout >= 0) {
8200 conf->algorithm = mddev->new_layout;
8201 mddev->layout = mddev->new_layout;
88ce4930
N		 8202 }
8203 if (new_chunk > 0) {
597a711b
N		 8204 conf->chunk_sectors = new_chunk ;
8205 mddev->chunk_sectors = new_chunk;
88ce4930 8206 }
2953079c 8207 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
88ce4930 8208 md_wakeup_thread(mddev->thread);
b3546035 8209 }
50ac168a 8210 return check_reshape(mddev);
88ce4930
N		 8211}
8212
fd01b88c 8213static int raid6_check_reshape(struct mddev *mddev)
88ce4930 8214{
597a711b 8215 int new_chunk = mddev->new_chunk_sectors;
50ac168a 8216
597a711b 8217 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
88ce4930 8218 return -EINVAL;
b3546035 8219 if (new_chunk > 0) {
0ba459d2 8220 if (!is_power_of_2(new_chunk))
88ce4930 8221 return -EINVAL;
597a711b 8222 if (new_chunk < (PAGE_SIZE >> 9))
88ce4930 8223 return -EINVAL;
597a711b 8224 if (mddev->array_sectors & (new_chunk-1))
88ce4930
N		 8225 /* not factor of array size */
8226 return -EINVAL;
b3546035 8227 }
88ce4930
N		 8228
8229 /* They look valid */
50ac168a 8230 return check_reshape(mddev);
b3546035
N		 8231}
8232
fd01b88c 8233static void *raid5_takeover(struct mddev *mddev)
d562b0c4
N		 8234{
8235 /* raid5 can take over:
f1b29bca 8236 * raid0 - if there is only one strip zone - make it a raid4 layout
d562b0c4
N		 8237 * raid1 - if there are two drives. We need to know the chunk size
8238 * raid4 - trivial - just use a raid4 layout.
8239 * raid6 - Providing it is a *_6 layout
d562b0c4 8240 */
f1b29bca
DW		 8241 if (mddev->level == 0)
8242 return raid45_takeover_raid0(mddev, 5);
d562b0c4
N		 8243 if (mddev->level == 1)
8244 return raid5_takeover_raid1(mddev);
e9d4758f
N		 8245 if (mddev->level == 4) {
8246 mddev->new_layout = ALGORITHM_PARITY_N;
8247 mddev->new_level = 5;
8248 return setup_conf(mddev);
8249 }
fc9739c6
N		 8250 if (mddev->level == 6)
8251 return raid5_takeover_raid6(mddev);
d562b0c4
N		 8252
8253 return ERR_PTR(-EINVAL);
8254}
8255
fd01b88c 8256static void *raid4_takeover(struct mddev *mddev)
a78d38a1 8257{
f1b29bca
DW		 8258 /* raid4 can take over:
8259 * raid0 - if there is only one strip zone
8260 * raid5 - if layout is right
a78d38a1 8261 */
f1b29bca
DW		 8262 if (mddev->level == 0)
8263 return raid45_takeover_raid0(mddev, 4);
a78d38a1
N		 8264 if (mddev->level == 5 &&
8265 mddev->layout == ALGORITHM_PARITY_N) {
8266 mddev->new_layout = 0;
8267 mddev->new_level = 4;
8268 return setup_conf(mddev);
8269 }
8270 return ERR_PTR(-EINVAL);
8271}
d562b0c4 8272
84fc4b56 8273static struct md_personality raid5_personality;
245f46c2 8274
fd01b88c 8275static void *raid6_takeover(struct mddev *mddev)
245f46c2
N		 8276{
8277 /* Currently can only take over a raid5. We map the
8278 * personality to an equivalent raid6 personality
8279 * with the Q block at the end.
8280 */
8281 int new_layout;
8282
8283 if (mddev->pers != &raid5_personality)
8284 return ERR_PTR(-EINVAL);
8285 if (mddev->degraded > 1)
8286 return ERR_PTR(-EINVAL);
8287 if (mddev->raid_disks > 253)
8288 return ERR_PTR(-EINVAL);
8289 if (mddev->raid_disks < 3)
8290 return ERR_PTR(-EINVAL);
8291
8292 switch (mddev->layout) {
8293 case ALGORITHM_LEFT_ASYMMETRIC:
8294 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
8295 break;
8296 case ALGORITHM_RIGHT_ASYMMETRIC:
8297 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
8298 break;
8299 case ALGORITHM_LEFT_SYMMETRIC:
8300 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
8301 break;
8302 case ALGORITHM_RIGHT_SYMMETRIC:
8303 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
8304 break;
8305 case ALGORITHM_PARITY_0:
8306 new_layout = ALGORITHM_PARITY_0_6;
8307 break;
8308 case ALGORITHM_PARITY_N:
8309 new_layout = ALGORITHM_PARITY_N;
8310 break;
8311 default:
8312 return ERR_PTR(-EINVAL);
8313 }
8314 mddev->new_level = 6;
8315 mddev->new_layout = new_layout;
8316 mddev->delta_disks = 1;
8317 mddev->raid_disks += 1;
8318 return setup_conf(mddev);
8319}
8320
ba903a3e
AP		 8321static int raid5_change_consistency_policy(struct mddev *mddev, const char *buf)
8322{
8323 struct r5conf *conf;
8324 int err;
8325
8326 err = mddev_lock(mddev);
8327 if (err)
8328 return err;
8329 conf = mddev->private;
8330 if (!conf) {
8331 mddev_unlock(mddev);
8332 return -ENODEV;
8333 }
8334
845b9e22 8335 if (strncmp(buf, "ppl", 3) == 0) {
0bb0c105 8336 /* ppl only works with RAID 5 */
845b9e22
AP		 8337 if (!raid5_has_ppl(conf) && conf->level == 5) {
8338 err = log_init(conf, NULL, true);
8339 if (!err) {
8340 err = resize_stripes(conf, conf->pool_size);
8341 if (err)
8342 log_exit(conf);
8343 }
0bb0c105
SL		 8344 } else
8345 err = -EINVAL;
8346 } else if (strncmp(buf, "resync", 6) == 0) {
8347 if (raid5_has_ppl(conf)) {
8348 mddev_suspend(mddev);
8349 log_exit(conf);
0bb0c105 8350 mddev_resume(mddev);
845b9e22 8351 err = resize_stripes(conf, conf->pool_size);
0bb0c105
SL		 8352 } else if (test_bit(MD_HAS_JOURNAL, &conf->mddev->flags) &&
8353 r5l_log_disk_error(conf)) {
8354 bool journal_dev_exists = false;
8355 struct md_rdev *rdev;
8356
8357 rdev_for_each(rdev, mddev)
8358 if (test_bit(Journal, &rdev->flags)) {
8359 journal_dev_exists = true;
8360 break;
8361 }
8362
8363 if (!journal_dev_exists) {
8364 mddev_suspend(mddev);
8365 clear_bit(MD_HAS_JOURNAL, &mddev->flags);
8366 mddev_resume(mddev);
8367 } else /* need remove journal device first */
8368 err = -EBUSY;
8369 } else
8370 err = -EINVAL;
ba903a3e
AP		 8371 } else {
8372 err = -EINVAL;
8373 }
8374
8375 if (!err)
8376 md_update_sb(mddev, 1);
8377
8378 mddev_unlock(mddev);
8379
8380 return err;
8381}
8382
d5d885fd
SL		 8383static int raid5_start(struct mddev *mddev)
8384{
8385 struct r5conf *conf = mddev->private;
8386
8387 return r5l_start(conf->log);
8388}
8389
84fc4b56 8390static struct md_personality raid6_personality =
16a53ecc
N		 8391{
8392 .name = "raid6",
8393 .level = 6,
8394 .owner = THIS_MODULE,
849674e4
SL		 8395 .make_request = raid5_make_request,
8396 .run = raid5_run,
d5d885fd 8397 .start = raid5_start,
afa0f557 8398 .free = raid5_free,
849674e4
SL		 8399 .status = raid5_status,
8400 .error_handler = raid5_error,
16a53ecc
N		 8401 .hot_add_disk = raid5_add_disk,
8402 .hot_remove_disk= raid5_remove_disk,
8403 .spare_active = raid5_spare_active,
849674e4 8404 .sync_request = raid5_sync_request,
16a53ecc 8405 .resize = raid5_resize,
80c3a6ce 8406 .size = raid5_size,
50ac168a 8407 .check_reshape = raid6_check_reshape,
f416885e 8408 .start_reshape = raid5_start_reshape,
cea9c228 8409 .finish_reshape = raid5_finish_reshape,
16a53ecc 8410 .quiesce = raid5_quiesce,
245f46c2 8411 .takeover = raid6_takeover,
5c675f83 8412 .congested = raid5_congested,
0bb0c105 8413 .change_consistency_policy = raid5_change_consistency_policy,
16a53ecc 8414};
84fc4b56 8415static struct md_personality raid5_personality =
1da177e4
LT		 8416{
8417 .name = "raid5",
2604b703 8418 .level = 5,
1da177e4 8419 .owner = THIS_MODULE,
849674e4
SL		 8420 .make_request = raid5_make_request,
8421 .run = raid5_run,
d5d885fd 8422 .start = raid5_start,
afa0f557 8423 .free = raid5_free,
849674e4
SL		 8424 .status = raid5_status,
8425 .error_handler = raid5_error,
1da177e4
LT		 8426 .hot_add_disk = raid5_add_disk,
8427 .hot_remove_disk= raid5_remove_disk,
8428 .spare_active = raid5_spare_active,
849674e4 8429 .sync_request = raid5_sync_request,
1da177e4 8430 .resize = raid5_resize,
80c3a6ce 8431 .size = raid5_size,
63c70c4f
N		 8432 .check_reshape = raid5_check_reshape,
8433 .start_reshape = raid5_start_reshape,
cea9c228 8434 .finish_reshape = raid5_finish_reshape,
72626685 8435 .quiesce = raid5_quiesce,
d562b0c4 8436 .takeover = raid5_takeover,
5c675f83 8437 .congested = raid5_congested,
ba903a3e 8438 .change_consistency_policy = raid5_change_consistency_policy,
1da177e4
LT		 8439};
8440
84fc4b56 8441static struct md_personality raid4_personality =
1da177e4 8442{
2604b703
N		 8443 .name = "raid4",
8444 .level = 4,
8445 .owner = THIS_MODULE,
849674e4
SL		 8446 .make_request = raid5_make_request,
8447 .run = raid5_run,
d5d885fd 8448 .start = raid5_start,
afa0f557 8449 .free = raid5_free,
849674e4
SL		 8450 .status = raid5_status,
8451 .error_handler = raid5_error,
2604b703
N		 8452 .hot_add_disk = raid5_add_disk,
8453 .hot_remove_disk= raid5_remove_disk,
8454 .spare_active = raid5_spare_active,
849674e4 8455 .sync_request = raid5_sync_request,
2604b703 8456 .resize = raid5_resize,
80c3a6ce 8457 .size = raid5_size,
3d37890b
N		 8458 .check_reshape = raid5_check_reshape,
8459 .start_reshape = raid5_start_reshape,
cea9c228 8460 .finish_reshape = raid5_finish_reshape,
2604b703 8461 .quiesce = raid5_quiesce,
a78d38a1 8462 .takeover = raid4_takeover,
5c675f83 8463 .congested = raid5_congested,
0bb0c105 8464 .change_consistency_policy = raid5_change_consistency_policy,
2604b703
N		 8465};
8466
8467static int __init raid5_init(void)
8468{
29c6d1bb
SAS		 8469 int ret;
8470
851c30c9
SL		 8471 raid5_wq = alloc_workqueue("raid5wq",
8472 WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0);
8473 if (!raid5_wq)
8474 return -ENOMEM;
29c6d1bb
SAS		 8475
8476 ret = cpuhp_setup_state_multi(CPUHP_MD_RAID5_PREPARE,
8477 "md/raid5:prepare",
8478 raid456_cpu_up_prepare,
8479 raid456_cpu_dead);
8480 if (ret) {
8481 destroy_workqueue(raid5_wq);
8482 return ret;
8483 }
16a53ecc 8484 register_md_personality(&raid6_personality);
2604b703
N		 8485 register_md_personality(&raid5_personality);
8486 register_md_personality(&raid4_personality);
8487 return 0;
1da177e4
LT		 8488}
8489
2604b703 8490static void raid5_exit(void)
1da177e4 8491{
16a53ecc 8492 unregister_md_personality(&raid6_personality);
2604b703
N		 8493 unregister_md_personality(&raid5_personality);
8494 unregister_md_personality(&raid4_personality);
29c6d1bb 8495 cpuhp_remove_multi_state(CPUHP_MD_RAID5_PREPARE);
851c30c9 8496 destroy_workqueue(raid5_wq);
1da177e4
LT		 8497}
8498
8499module_init(raid5_init);
8500module_exit(raid5_exit);
8501MODULE_LICENSE("GPL");
0efb9e61 8502MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
1da177e4 8503MODULE_ALIAS("md-personality-4"); /* RAID5 */
d9d166c2
N		 8504MODULE_ALIAS("md-raid5");
8505MODULE_ALIAS("md-raid4");
2604b703
N		 8506MODULE_ALIAS("md-level-5");
8507MODULE_ALIAS("md-level-4");
16a53ecc
N		 8508MODULE_ALIAS("md-personality-8"); /* RAID6 */
8509MODULE_ALIAS("md-raid6");
8510MODULE_ALIAS("md-level-6");
8511
8512/* This used to be two separate modules, they were: */
8513MODULE_ALIAS("raid5");
8514MODULE_ALIAS("raid6");
Linux 6.x block layer and io_uring tree(s)