projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
raid5: need to set STRIPE_HANDLE for batch head
[linux-2.6-block.git] / drivers / md / raid5.c
CommitLineData
af1a8899 1// SPDX-License-Identifier: GPL-2.0-or-later
1da177e4
LT		 2/*
3 * raid5.c : Multiple Devices driver for Linux
4 * Copyright (C) 1996, 1997 Ingo Molnar, Miguel de Icaza, Gadi Oxman
5 * Copyright (C) 1999, 2000 Ingo Molnar
16a53ecc 6 * Copyright (C) 2002, 2003 H. Peter Anvin
1da177e4 7 *
16a53ecc
N		 8 * RAID-4/5/6 management functions.
9 * Thanks to Penguin Computing for making the RAID-6 development possible
10 * by donating a test server!
1da177e4
LT		 11 */
12
ae3c20cc
N		 13/*
14 * BITMAP UNPLUGGING:
15 *
16 * The sequencing for updating the bitmap reliably is a little
17 * subtle (and I got it wrong the first time) so it deserves some
18 * explanation.
19 *
20 * We group bitmap updates into batches. Each batch has a number.
21 * We may write out several batches at once, but that isn't very important.
7c13edc8
N		 22 * conf->seq_write is the number of the last batch successfully written.
23 * conf->seq_flush is the number of the last batch that was closed to
ae3c20cc
N		 24 * new additions.
25 * When we discover that we will need to write to any block in a stripe
26 * (in add_stripe_bio) we update the in-memory bitmap and record in sh->bm_seq
7c13edc8 27 * the number of the batch it will be in. This is seq_flush+1.
ae3c20cc
N		 28 * When we are ready to do a write, if that batch hasn't been written yet,
29 * we plug the array and queue the stripe for later.
30 * When an unplug happens, we increment bm_flush, thus closing the current
31 * batch.
32 * When we notice that bm_flush > bm_write, we write out all pending updates
33 * to the bitmap, and advance bm_write to where bm_flush was.
34 * This may occasionally write a bit out twice, but is sure never to
35 * miss any bits.
36 */
1da177e4 37
bff61975 38#include <linux/blkdev.h>
f6705578 39#include <linux/kthread.h>
f701d589 40#include <linux/raid/pq.h>
91c00924 41#include <linux/async_tx.h>
056075c7 42#include <linux/module.h>
07a3b417 43#include <linux/async.h>
bff61975 44#include <linux/seq_file.h>
36d1c647 45#include <linux/cpu.h>
5a0e3ad6 46#include <linux/slab.h>
8bda470e 47#include <linux/ratelimit.h>
851c30c9 48#include <linux/nodemask.h>
3f07c014 49
a9add5d9 50#include <trace/events/block.h>
aaf9f12e 51#include <linux/list_sort.h>
a9add5d9 52
43b2e5d8 53#include "md.h"
bff61975 54#include "raid5.h"
54071b38 55#include "raid0.h"
935fe098 56#include "md-bitmap.h"
ff875738 57#include "raid5-log.h"
72626685 58
394ed8e4
SL		 59#define UNSUPPORTED_MDDEV_FLAGS (1L << MD_FAILFAST_SUPPORTED)
60
851c30c9
SL		 61#define cpu_to_group(cpu) cpu_to_node(cpu)
62#define ANY_GROUP NUMA_NO_NODE
63
8e0e99ba
N		 64static bool devices_handle_discard_safely = false;
65module_param(devices_handle_discard_safely, bool, 0644);
66MODULE_PARM_DESC(devices_handle_discard_safely,
67 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
851c30c9 68static struct workqueue_struct *raid5_wq;
1da177e4 69
d1688a6d 70static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect)
db298e19
N		 71{
72 int hash = (sect >> STRIPE_SHIFT) & HASH_MASK;
73 return &conf->stripe_hashtbl[hash];
74}
1da177e4 75
566c09c5
SL		 76static inline int stripe_hash_locks_hash(sector_t sect)
77{
78 return (sect >> STRIPE_SHIFT) & STRIPE_HASH_LOCKS_MASK;
79}
80
81static inline void lock_device_hash_lock(struct r5conf *conf, int hash)
82{
83 spin_lock_irq(conf->hash_locks + hash);
84 spin_lock(&conf->device_lock);
85}
86
87static inline void unlock_device_hash_lock(struct r5conf *conf, int hash)
88{
89 spin_unlock(&conf->device_lock);
90 spin_unlock_irq(conf->hash_locks + hash);
91}
92
93static inline void lock_all_device_hash_locks_irq(struct r5conf *conf)
94{
95 int i;
3d05f3ae 96 spin_lock_irq(conf->hash_locks);
566c09c5
SL		 97 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
98 spin_lock_nest_lock(conf->hash_locks + i, conf->hash_locks);
99 spin_lock(&conf->device_lock);
100}
101
102static inline void unlock_all_device_hash_locks_irq(struct r5conf *conf)
103{
104 int i;
105 spin_unlock(&conf->device_lock);
3d05f3ae
JC		 106 for (i = NR_STRIPE_HASH_LOCKS - 1; i; i--)
107 spin_unlock(conf->hash_locks + i);
108 spin_unlock_irq(conf->hash_locks);
566c09c5
SL		 109}
110
d0dabf7e
N		 111/* Find first data disk in a raid6 stripe */
112static inline int raid6_d0(struct stripe_head *sh)
113{
67cc2b81
N		 114 if (sh->ddf_layout)
115 /* ddf always start from first device */
116 return 0;
117 /* md starts just after Q block */
d0dabf7e
N		 118 if (sh->qd_idx == sh->disks - 1)
119 return 0;
120 else
121 return sh->qd_idx + 1;
122}
16a53ecc
N		 123static inline int raid6_next_disk(int disk, int raid_disks)
124{
125 disk++;
126 return (disk < raid_disks) ? disk : 0;
127}
a4456856 128
d0dabf7e
N		 129/* When walking through the disks in a raid5, starting at raid6_d0,
130 * We need to map each disk to a 'slot', where the data disks are slot
131 * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk
132 * is raid_disks-1. This help does that mapping.
133 */
67cc2b81
N		 134static int raid6_idx_to_slot(int idx, struct stripe_head *sh,
135 int *count, int syndrome_disks)
d0dabf7e 136{
6629542e 137 int slot = *count;
67cc2b81 138
e4424fee 139 if (sh->ddf_layout)
6629542e 140 (*count)++;
d0dabf7e 141 if (idx == sh->pd_idx)
67cc2b81 142 return syndrome_disks;
d0dabf7e 143 if (idx == sh->qd_idx)
67cc2b81 144 return syndrome_disks + 1;
e4424fee 145 if (!sh->ddf_layout)
6629542e 146 (*count)++;
d0dabf7e
N		 147 return slot;
148}
149
d1688a6d 150static void print_raid5_conf (struct r5conf *conf);
1da177e4 151
600aa109
DW		 152static int stripe_operations_active(struct stripe_head *sh)
153{
154 return sh->check_state || sh->reconstruct_state ||
155 test_bit(STRIPE_BIOFILL_RUN, &sh->state) ||
156 test_bit(STRIPE_COMPUTE_RUN, &sh->state);
157}
158
535ae4eb
SL		 159static bool stripe_is_lowprio(struct stripe_head *sh)
160{
161 return (test_bit(STRIPE_R5C_FULL_STRIPE, &sh->state) ||
162 test_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state)) &&
163 !test_bit(STRIPE_R5C_CACHING, &sh->state);
164}
165
851c30c9
SL		 166static void raid5_wakeup_stripe_thread(struct stripe_head *sh)
167{
168 struct r5conf *conf = sh->raid_conf;
169 struct r5worker_group *group;
bfc90cb0 170 int thread_cnt;
851c30c9
SL		 171 int i, cpu = sh->cpu;
172
173 if (!cpu_online(cpu)) {
174 cpu = cpumask_any(cpu_online_mask);
175 sh->cpu = cpu;
176 }
177
178 if (list_empty(&sh->lru)) {
179 struct r5worker_group *group;
180 group = conf->worker_groups + cpu_to_group(cpu);
535ae4eb
SL		 181 if (stripe_is_lowprio(sh))
182 list_add_tail(&sh->lru, &group->loprio_list);
183 else
184 list_add_tail(&sh->lru, &group->handle_list);
bfc90cb0
SL		 185 group->stripes_cnt++;
186 sh->group = group;
851c30c9
SL		 187 }
188
189 if (conf->worker_cnt_per_group == 0) {
190 md_wakeup_thread(conf->mddev->thread);
191 return;
192 }
193
194 group = conf->worker_groups + cpu_to_group(sh->cpu);
195
bfc90cb0
SL		 196 group->workers[0].working = true;
197 /* at least one worker should run to avoid race */
198 queue_work_on(sh->cpu, raid5_wq, &group->workers[0].work);
199
200 thread_cnt = group->stripes_cnt / MAX_STRIPE_BATCH - 1;
201 /* wakeup more workers */
202 for (i = 1; i < conf->worker_cnt_per_group && thread_cnt > 0; i++) {
203 if (group->workers[i].working == false) {
204 group->workers[i].working = true;
205 queue_work_on(sh->cpu, raid5_wq,
206 &group->workers[i].work);
207 thread_cnt--;
208 }
209 }
851c30c9
SL		 210}
211
566c09c5
SL		 212static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh,
213 struct list_head *temp_inactive_list)
1da177e4 214{
1e6d690b
SL		 215 int i;
216 int injournal = 0; /* number of date pages with R5_InJournal */
217
4eb788df
SL		 218 BUG_ON(!list_empty(&sh->lru));
219 BUG_ON(atomic_read(&conf->active_stripes)==0);
1e6d690b
SL		 220
221 if (r5c_is_writeback(conf->log))
222 for (i = sh->disks; i--; )
223 if (test_bit(R5_InJournal, &sh->dev[i].flags))
224 injournal++;
a39f7afd 225 /*
5ddf0440
SL		 226 * In the following cases, the stripe cannot be released to cached
227 * lists. Therefore, we make the stripe write out and set
228 * STRIPE_HANDLE:
229 * 1. when quiesce in r5c write back;
230 * 2. when resync is requested fot the stripe.
a39f7afd 231 */
5ddf0440
SL		 232 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) ||
233 (conf->quiesce && r5c_is_writeback(conf->log) &&
234 !test_bit(STRIPE_HANDLE, &sh->state) && injournal != 0)) {
a39f7afd
SL		 235 if (test_bit(STRIPE_R5C_CACHING, &sh->state))
236 r5c_make_stripe_write_out(sh);
237 set_bit(STRIPE_HANDLE, &sh->state);
238 }
1e6d690b 239
4eb788df
SL		 240 if (test_bit(STRIPE_HANDLE, &sh->state)) {
241 if (test_bit(STRIPE_DELAYED, &sh->state) &&
ad3ab8b6 242 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4eb788df 243 list_add_tail(&sh->lru, &conf->delayed_list);
ad3ab8b6 244 else if (test_bit(STRIPE_BIT_DELAY, &sh->state) &&
4eb788df
SL		 245 sh->bm_seq - conf->seq_write > 0)
246 list_add_tail(&sh->lru, &conf->bitmap_list);
247 else {
248 clear_bit(STRIPE_DELAYED, &sh->state);
249 clear_bit(STRIPE_BIT_DELAY, &sh->state);
851c30c9 250 if (conf->worker_cnt_per_group == 0) {
535ae4eb
SL		 251 if (stripe_is_lowprio(sh))
252 list_add_tail(&sh->lru,
253 &conf->loprio_list);
254 else
255 list_add_tail(&sh->lru,
256 &conf->handle_list);
851c30c9
SL		 257 } else {
258 raid5_wakeup_stripe_thread(sh);
259 return;
260 }
4eb788df
SL		 261 }
262 md_wakeup_thread(conf->mddev->thread);
263 } else {
264 BUG_ON(stripe_operations_active(sh));
265 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
266 if (atomic_dec_return(&conf->preread_active_stripes)
267 < IO_THRESHOLD)
268 md_wakeup_thread(conf->mddev->thread);
269 atomic_dec(&conf->active_stripes);
1e6d690b
SL		 270 if (!test_bit(STRIPE_EXPANDING, &sh->state)) {
271 if (!r5c_is_writeback(conf->log))
272 list_add_tail(&sh->lru, temp_inactive_list);
273 else {
274 WARN_ON(test_bit(R5_InJournal, &sh->dev[sh->pd_idx].flags));
275 if (injournal == 0)
276 list_add_tail(&sh->lru, temp_inactive_list);
277 else if (injournal == conf->raid_disks - conf->max_degraded) {
278 /* full stripe */
279 if (!test_and_set_bit(STRIPE_R5C_FULL_STRIPE, &sh->state))
280 atomic_inc(&conf->r5c_cached_full_stripes);
281 if (test_and_clear_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state))
282 atomic_dec(&conf->r5c_cached_partial_stripes);
283 list_add_tail(&sh->lru, &conf->r5c_full_stripe_list);
a39f7afd 284 r5c_check_cached_full_stripe(conf);
03b047f4
SL		 285 } else
286 /*
287 * STRIPE_R5C_PARTIAL_STRIPE is set in
288 * r5c_try_caching_write(). No need to
289 * set it again.
290 */
1e6d690b 291 list_add_tail(&sh->lru, &conf->r5c_partial_stripe_list);
1e6d690b
SL		 292 }
293 }
1da177e4
LT		 294 }
295}
d0dabf7e 296
566c09c5
SL		 297static void __release_stripe(struct r5conf *conf, struct stripe_head *sh,
298 struct list_head *temp_inactive_list)
4eb788df
SL		 299{
300 if (atomic_dec_and_test(&sh->count))
566c09c5
SL		 301 do_release_stripe(conf, sh, temp_inactive_list);
302}
303
304/*
305 * @hash could be NR_STRIPE_HASH_LOCKS, then we have a list of inactive_list
306 *
307 * Be careful: Only one task can add/delete stripes from temp_inactive_list at
308 * given time. Adding stripes only takes device lock, while deleting stripes
309 * only takes hash lock.
310 */
311static void release_inactive_stripe_list(struct r5conf *conf,
312 struct list_head *temp_inactive_list,
313 int hash)
314{
315 int size;
6ab2a4b8 316 bool do_wakeup = false;
566c09c5
SL		 317 unsigned long flags;
318
319 if (hash == NR_STRIPE_HASH_LOCKS) {
320 size = NR_STRIPE_HASH_LOCKS;
321 hash = NR_STRIPE_HASH_LOCKS - 1;
322 } else
323 size = 1;
324 while (size) {
325 struct list_head *list = &temp_inactive_list[size - 1];
326
327 /*
6d036f7d 328 * We don't hold any lock here yet, raid5_get_active_stripe() might
566c09c5
SL		 329 * remove stripes from the list
330 */
331 if (!list_empty_careful(list)) {
332 spin_lock_irqsave(conf->hash_locks + hash, flags);
4bda556a
SL		 333 if (list_empty(conf->inactive_list + hash) &&
334 !list_empty(list))
335 atomic_dec(&conf->empty_inactive_list_nr);
566c09c5 336 list_splice_tail_init(list, conf->inactive_list + hash);
6ab2a4b8 337 do_wakeup = true;
566c09c5
SL		 338 spin_unlock_irqrestore(conf->hash_locks + hash, flags);
339 }
340 size--;
341 hash--;
342 }
343
344 if (do_wakeup) {
6ab2a4b8 345 wake_up(&conf->wait_for_stripe);
b1b46486
YL		 346 if (atomic_read(&conf->active_stripes) == 0)
347 wake_up(&conf->wait_for_quiescent);
566c09c5
SL		 348 if (conf->retry_read_aligned)
349 md_wakeup_thread(conf->mddev->thread);
350 }
4eb788df
SL		 351}
352
773ca82f 353/* should hold conf->device_lock already */
566c09c5
SL		 354static int release_stripe_list(struct r5conf *conf,
355 struct list_head *temp_inactive_list)
773ca82f 356{
eae8263f 357 struct stripe_head *sh, *t;
773ca82f
SL		 358 int count = 0;
359 struct llist_node *head;
360
361 head = llist_del_all(&conf->released_stripes);
d265d9dc 362 head = llist_reverse_order(head);
eae8263f 363 llist_for_each_entry_safe(sh, t, head, release_list) {
566c09c5
SL		 364 int hash;
365
773ca82f
SL		 366 /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */
367 smp_mb();
368 clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state);
369 /*
370 * Don't worry the bit is set here, because if the bit is set
371 * again, the count is always > 1. This is true for
372 * STRIPE_ON_UNPLUG_LIST bit too.
373 */
566c09c5
SL		 374 hash = sh->hash_lock_index;
375 __release_stripe(conf, sh, &temp_inactive_list[hash]);
773ca82f
SL		 376 count++;
377 }
378
379 return count;
380}
381
6d036f7d 382void raid5_release_stripe(struct stripe_head *sh)
1da177e4 383{
d1688a6d 384 struct r5conf *conf = sh->raid_conf;
1da177e4 385 unsigned long flags;
566c09c5
SL		 386 struct list_head list;
387 int hash;
773ca82f 388 bool wakeup;
16a53ecc 389
cf170f3f
ES		 390 /* Avoid release_list until the last reference.
391 */
392 if (atomic_add_unless(&sh->count, -1, 1))
393 return;
394
ad4068de 395 if (unlikely(!conf->mddev->thread) ||
396 test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state))
773ca82f
SL		 397 goto slow_path;
398 wakeup = llist_add(&sh->release_list, &conf->released_stripes);
399 if (wakeup)
400 md_wakeup_thread(conf->mddev->thread);
401 return;
402slow_path:
773ca82f 403 /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */
685dbcaa 404 if (atomic_dec_and_lock_irqsave(&sh->count, &conf->device_lock, flags)) {
566c09c5
SL		 405 INIT_LIST_HEAD(&list);
406 hash = sh->hash_lock_index;
407 do_release_stripe(conf, sh, &list);
08edaaa6 408 spin_unlock_irqrestore(&conf->device_lock, flags);
566c09c5 409 release_inactive_stripe_list(conf, &list, hash);
4eb788df 410 }
1da177e4
LT		 411}
412
fccddba0 413static inline void remove_hash(struct stripe_head *sh)
1da177e4 414{
45b4233c
DW		 415 pr_debug("remove_hash(), stripe %llu\n",
416 (unsigned long long)sh->sector);
1da177e4 417
fccddba0 418 hlist_del_init(&sh->hash);
1da177e4
LT		 419}
420
d1688a6d 421static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh)
1da177e4 422{
fccddba0 423 struct hlist_head *hp = stripe_hash(conf, sh->sector);
1da177e4 424
45b4233c
DW		 425 pr_debug("insert_hash(), stripe %llu\n",
426 (unsigned long long)sh->sector);
1da177e4 427
fccddba0 428 hlist_add_head(&sh->hash, hp);
1da177e4
LT		 429}
430
1da177e4 431/* find an idle stripe, make sure it is unhashed, and return it. */
566c09c5 432static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash)
1da177e4
LT		 433{
434 struct stripe_head *sh = NULL;
435 struct list_head *first;
436
566c09c5 437 if (list_empty(conf->inactive_list + hash))
1da177e4 438 goto out;
566c09c5 439 first = (conf->inactive_list + hash)->next;
1da177e4
LT		 440 sh = list_entry(first, struct stripe_head, lru);
441 list_del_init(first);
442 remove_hash(sh);
443 atomic_inc(&conf->active_stripes);
566c09c5 444 BUG_ON(hash != sh->hash_lock_index);
4bda556a
SL		 445 if (list_empty(conf->inactive_list + hash))
446 atomic_inc(&conf->empty_inactive_list_nr);
1da177e4
LT		 447out:
448 return sh;
449}
450
e4e11e38 451static void shrink_buffers(struct stripe_head *sh)
1da177e4
LT		 452{
453 struct page *p;
454 int i;
e4e11e38 455 int num = sh->raid_conf->pool_size;
1da177e4 456
e4e11e38 457 for (i = 0; i < num ; i++) {
d592a996 458 WARN_ON(sh->dev[i].page != sh->dev[i].orig_page);
1da177e4
LT		 459 p = sh->dev[i].page;
460 if (!p)
461 continue;
462 sh->dev[i].page = NULL;
2d1f3b5d 463 put_page(p);
1da177e4
LT		 464 }
465}
466
a9683a79 467static int grow_buffers(struct stripe_head *sh, gfp_t gfp)
1da177e4
LT		 468{
469 int i;
e4e11e38 470 int num = sh->raid_conf->pool_size;
1da177e4 471
e4e11e38 472 for (i = 0; i < num; i++) {
1da177e4
LT		 473 struct page *page;
474
a9683a79 475 if (!(page = alloc_page(gfp))) {
1da177e4
LT		 476 return 1;
477 }
478 sh->dev[i].page = page;
d592a996 479 sh->dev[i].orig_page = page;
1da177e4 480 }
3418d036 481
1da177e4
LT		 482 return 0;
483}
484
d1688a6d 485static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
911d4ee8 486 struct stripe_head *sh);
1da177e4 487
b5663ba4 488static void init_stripe(struct stripe_head *sh, sector_t sector, int previous)
1da177e4 489{
d1688a6d 490 struct r5conf *conf = sh->raid_conf;
566c09c5 491 int i, seq;
1da177e4 492
78bafebd
ES		 493 BUG_ON(atomic_read(&sh->count) != 0);
494 BUG_ON(test_bit(STRIPE_HANDLE, &sh->state));
600aa109 495 BUG_ON(stripe_operations_active(sh));
59fc630b 496 BUG_ON(sh->batch_head);
d84e0f10 497
45b4233c 498 pr_debug("init_stripe called, stripe %llu\n",
b8e6a15a 499 (unsigned long long)sector);
566c09c5
SL		 500retry:
501 seq = read_seqcount_begin(&conf->gen_lock);
86b42c71 502 sh->generation = conf->generation - previous;
b5663ba4 503 sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks;
1da177e4 504 sh->sector = sector;
911d4ee8 505 stripe_set_idx(sector, conf, previous, sh);
1da177e4
LT		 506 sh->state = 0;
507
7ecaa1e6 508 for (i = sh->disks; i--; ) {
1da177e4
LT		 509 struct r5dev *dev = &sh->dev[i];
510
d84e0f10 511 if (dev->toread || dev->read || dev->towrite || dev->written ||
1da177e4 512 test_bit(R5_LOCKED, &dev->flags)) {
cc6167b4 513 pr_err("sector=%llx i=%d %p %p %p %p %d\n",
1da177e4 514 (unsigned long long)sh->sector, i, dev->toread,
d84e0f10 515 dev->read, dev->towrite, dev->written,
1da177e4 516 test_bit(R5_LOCKED, &dev->flags));
8cfa7b0f 517 WARN_ON(1);
1da177e4
LT		 518 }
519 dev->flags = 0;
27a4ff8f 520 dev->sector = raid5_compute_blocknr(sh, i, previous);
1da177e4 521 }
566c09c5
SL		 522 if (read_seqcount_retry(&conf->gen_lock, seq))
523 goto retry;
7a87f434 524 sh->overwrite_disks = 0;
1da177e4 525 insert_hash(conf, sh);
851c30c9 526 sh->cpu = smp_processor_id();
da41ba65 527 set_bit(STRIPE_BATCH_READY, &sh->state);
1da177e4
LT		 528}
529
d1688a6d 530static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector,
86b42c71 531 short generation)
1da177e4
LT		 532{
533 struct stripe_head *sh;
534
45b4233c 535 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
b67bfe0d 536 hlist_for_each_entry(sh, stripe_hash(conf, sector), hash)
86b42c71 537 if (sh->sector == sector && sh->generation == generation)
1da177e4 538 return sh;
45b4233c 539 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
1da177e4
LT		 540 return NULL;
541}
542
674806d6
N		 543/*
544 * Need to check if array has failed when deciding whether to:
545 * - start an array
546 * - remove non-faulty devices
547 * - add a spare
548 * - allow a reshape
549 * This determination is simple when no reshape is happening.
550 * However if there is a reshape, we need to carefully check
551 * both the before and after sections.
552 * This is because some failed devices may only affect one
553 * of the two sections, and some non-in_sync devices may
554 * be insync in the section most affected by failed devices.
555 */
2e38a37f 556int raid5_calc_degraded(struct r5conf *conf)
674806d6 557{
908f4fbd 558 int degraded, degraded2;
674806d6 559 int i;
674806d6
N		 560
561 rcu_read_lock();
562 degraded = 0;
563 for (i = 0; i < conf->previous_raid_disks; i++) {
3cb03002 564 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
e5c86471
N		 565 if (rdev && test_bit(Faulty, &rdev->flags))
566 rdev = rcu_dereference(conf->disks[i].replacement);
674806d6
N		 567 if (!rdev || test_bit(Faulty, &rdev->flags))
568 degraded++;
569 else if (test_bit(In_sync, &rdev->flags))
570 ;
571 else
572 /* not in-sync or faulty.
573 * If the reshape increases the number of devices,
574 * this is being recovered by the reshape, so
575 * this 'previous' section is not in_sync.
576 * If the number of devices is being reduced however,
577 * the device can only be part of the array if
578 * we are reverting a reshape, so this section will
579 * be in-sync.
580 */
581 if (conf->raid_disks >= conf->previous_raid_disks)
582 degraded++;
583 }
584 rcu_read_unlock();
908f4fbd
N		 585 if (conf->raid_disks == conf->previous_raid_disks)
586 return degraded;
674806d6 587 rcu_read_lock();
908f4fbd 588 degraded2 = 0;
674806d6 589 for (i = 0; i < conf->raid_disks; i++) {
3cb03002 590 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
e5c86471
N		 591 if (rdev && test_bit(Faulty, &rdev->flags))
592 rdev = rcu_dereference(conf->disks[i].replacement);
674806d6 593 if (!rdev || test_bit(Faulty, &rdev->flags))
908f4fbd 594 degraded2++;
674806d6
N		 595 else if (test_bit(In_sync, &rdev->flags))
596 ;
597 else
598 /* not in-sync or faulty.
599 * If reshape increases the number of devices, this
600 * section has already been recovered, else it
601 * almost certainly hasn't.
602 */
603 if (conf->raid_disks <= conf->previous_raid_disks)
908f4fbd 604 degraded2++;
674806d6
N		 605 }
606 rcu_read_unlock();
908f4fbd
N		 607 if (degraded2 > degraded)
608 return degraded2;
609 return degraded;
610}
611
612static int has_failed(struct r5conf *conf)
613{
614 int degraded;
615
616 if (conf->mddev->reshape_position == MaxSector)
617 return conf->mddev->degraded > conf->max_degraded;
618
2e38a37f 619 degraded = raid5_calc_degraded(conf);
674806d6
N		 620 if (degraded > conf->max_degraded)
621 return 1;
622 return 0;
623}
624
6d036f7d
SL		 625struct stripe_head *
626raid5_get_active_stripe(struct r5conf *conf, sector_t sector,
627 int previous, int noblock, int noquiesce)
1da177e4
LT		 628{
629 struct stripe_head *sh;
566c09c5 630 int hash = stripe_hash_locks_hash(sector);
ff00d3b4 631 int inc_empty_inactive_list_flag;
1da177e4 632
45b4233c 633 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector);
1da177e4 634
566c09c5 635 spin_lock_irq(conf->hash_locks + hash);
1da177e4
LT		 636
637 do {
b1b46486 638 wait_event_lock_irq(conf->wait_for_quiescent,
a8c906ca 639 conf->quiesce == 0 || noquiesce,
566c09c5 640 *(conf->hash_locks + hash));
86b42c71 641 sh = __find_stripe(conf, sector, conf->generation - previous);
1da177e4 642 if (!sh) {
edbe83ab 643 if (!test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) {
566c09c5 644 sh = get_free_stripe(conf, hash);
713bc5c2
SL		 645 if (!sh && !test_bit(R5_DID_ALLOC,
646 &conf->cache_state))
edbe83ab
N		 647 set_bit(R5_ALLOC_MORE,
648 &conf->cache_state);
649 }
1da177e4
LT		 650 if (noblock && sh == NULL)
651 break;
a39f7afd
SL		 652
653 r5c_check_stripe_cache_usage(conf);
1da177e4 654 if (!sh) {
5423399a
N		 655 set_bit(R5_INACTIVE_BLOCKED,
656 &conf->cache_state);
a39f7afd 657 r5l_wake_reclaim(conf->log, 0);
6ab2a4b8
SL		 658 wait_event_lock_irq(
659 conf->wait_for_stripe,
566c09c5
SL		 660 !list_empty(conf->inactive_list + hash) &&
661 (atomic_read(&conf->active_stripes)
662 < (conf->max_nr_stripes * 3 / 4)
5423399a
N		 663 || !test_bit(R5_INACTIVE_BLOCKED,
664 &conf->cache_state)),
6ab2a4b8 665 *(conf->hash_locks + hash));
5423399a
N		 666 clear_bit(R5_INACTIVE_BLOCKED,
667 &conf->cache_state);
7da9d450 668 } else {
b5663ba4 669 init_stripe(sh, sector, previous);
7da9d450
N		 670 atomic_inc(&sh->count);
671 }
e240c183 672 } else if (!atomic_inc_not_zero(&sh->count)) {
6d183de4 673 spin_lock(&conf->device_lock);
e240c183 674 if (!atomic_read(&sh->count)) {
1da177e4
LT		 675 if (!test_bit(STRIPE_HANDLE, &sh->state))
676 atomic_inc(&conf->active_stripes);
5af9bef7
N		 677 BUG_ON(list_empty(&sh->lru) &&
678 !test_bit(STRIPE_EXPANDING, &sh->state));
ff00d3b4
ZL		 679 inc_empty_inactive_list_flag = 0;
680 if (!list_empty(conf->inactive_list + hash))
681 inc_empty_inactive_list_flag = 1;
16a53ecc 682 list_del_init(&sh->lru);
ff00d3b4
ZL		 683 if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag)
684 atomic_inc(&conf->empty_inactive_list_nr);
bfc90cb0
SL		 685 if (sh->group) {
686 sh->group->stripes_cnt--;
687 sh->group = NULL;
688 }
1da177e4 689 }
7da9d450 690 atomic_inc(&sh->count);
6d183de4 691 spin_unlock(&conf->device_lock);
1da177e4
LT		 692 }
693 } while (sh == NULL);
694
566c09c5 695 spin_unlock_irq(conf->hash_locks + hash);
1da177e4
LT		 696 return sh;
697}
698
7a87f434 699static bool is_full_stripe_write(struct stripe_head *sh)
700{
701 BUG_ON(sh->overwrite_disks > (sh->disks - sh->raid_conf->max_degraded));
702 return sh->overwrite_disks == (sh->disks - sh->raid_conf->max_degraded);
703}
704
59fc630b 705static void lock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2)
368ecade
CH		 706 __acquires(&sh1->stripe_lock)
707 __acquires(&sh2->stripe_lock)
59fc630b 708{
59fc630b 709 if (sh1 > sh2) {
3d05f3ae 710 spin_lock_irq(&sh2->stripe_lock);
59fc630b 711 spin_lock_nested(&sh1->stripe_lock, 1);
712 } else {
3d05f3ae 713 spin_lock_irq(&sh1->stripe_lock);
59fc630b 714 spin_lock_nested(&sh2->stripe_lock, 1);
715 }
716}
717
718static void unlock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2)
368ecade
CH		 719 __releases(&sh1->stripe_lock)
720 __releases(&sh2->stripe_lock)
59fc630b 721{
722 spin_unlock(&sh1->stripe_lock);
3d05f3ae 723 spin_unlock_irq(&sh2->stripe_lock);
59fc630b 724}
725
726/* Only freshly new full stripe normal write stripe can be added to a batch list */
727static bool stripe_can_batch(struct stripe_head *sh)
728{
9c3e333d
SL		 729 struct r5conf *conf = sh->raid_conf;
730
e254de6b 731 if (raid5_has_log(conf) || raid5_has_ppl(conf))
9c3e333d 732 return false;
59fc630b 733 return test_bit(STRIPE_BATCH_READY, &sh->state) &&
d0852df5 734 !test_bit(STRIPE_BITMAP_PENDING, &sh->state) &&
59fc630b 735 is_full_stripe_write(sh);
736}
737
738/* we only do back search */
739static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh)
740{
741 struct stripe_head *head;
742 sector_t head_sector, tmp_sec;
743 int hash;
744 int dd_idx;
ff00d3b4 745 int inc_empty_inactive_list_flag;
59fc630b 746
59fc630b 747 /* Don't cross chunks, so stripe pd_idx/qd_idx is the same */
748 tmp_sec = sh->sector;
749 if (!sector_div(tmp_sec, conf->chunk_sectors))
750 return;
751 head_sector = sh->sector - STRIPE_SECTORS;
752
753 hash = stripe_hash_locks_hash(head_sector);
754 spin_lock_irq(conf->hash_locks + hash);
755 head = __find_stripe(conf, head_sector, conf->generation);
756 if (head && !atomic_inc_not_zero(&head->count)) {
757 spin_lock(&conf->device_lock);
758 if (!atomic_read(&head->count)) {
759 if (!test_bit(STRIPE_HANDLE, &head->state))
760 atomic_inc(&conf->active_stripes);
761 BUG_ON(list_empty(&head->lru) &&
762 !test_bit(STRIPE_EXPANDING, &head->state));
ff00d3b4
ZL		 763 inc_empty_inactive_list_flag = 0;
764 if (!list_empty(conf->inactive_list + hash))
765 inc_empty_inactive_list_flag = 1;
59fc630b 766 list_del_init(&head->lru);
ff00d3b4
ZL		 767 if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag)
768 atomic_inc(&conf->empty_inactive_list_nr);
59fc630b 769 if (head->group) {
770 head->group->stripes_cnt--;
771 head->group = NULL;
772 }
773 }
774 atomic_inc(&head->count);
775 spin_unlock(&conf->device_lock);
776 }
777 spin_unlock_irq(conf->hash_locks + hash);
778
779 if (!head)
780 return;
781 if (!stripe_can_batch(head))
782 goto out;
783
784 lock_two_stripes(head, sh);
785 /* clear_batch_ready clear the flag */
786 if (!stripe_can_batch(head) || !stripe_can_batch(sh))
787 goto unlock_out;
788
789 if (sh->batch_head)
790 goto unlock_out;
791
792 dd_idx = 0;
793 while (dd_idx == sh->pd_idx || dd_idx == sh->qd_idx)
794 dd_idx++;
1eff9d32 795 if (head->dev[dd_idx].towrite->bi_opf != sh->dev[dd_idx].towrite->bi_opf ||
796a5cf0 796 bio_op(head->dev[dd_idx].towrite) != bio_op(sh->dev[dd_idx].towrite))
59fc630b 797 goto unlock_out;
798
799 if (head->batch_head) {
800 spin_lock(&head->batch_head->batch_lock);
801 /* This batch list is already running */
802 if (!stripe_can_batch(head)) {
803 spin_unlock(&head->batch_head->batch_lock);
804 goto unlock_out;
805 }
3664847d
SL		 806 /*
807 * We must assign batch_head of this stripe within the
808 * batch_lock, otherwise clear_batch_ready of batch head
809 * stripe could clear BATCH_READY bit of this stripe and
810 * this stripe->batch_head doesn't get assigned, which
811 * could confuse clear_batch_ready for this stripe
812 */
813 sh->batch_head = head->batch_head;
59fc630b 814
815 /*
816 * at this point, head's BATCH_READY could be cleared, but we
817 * can still add the stripe to batch list
818 */
819 list_add(&sh->batch_list, &head->batch_list);
820 spin_unlock(&head->batch_head->batch_lock);
59fc630b 821 } else {
822 head->batch_head = head;
823 sh->batch_head = head->batch_head;
824 spin_lock(&head->batch_lock);
825 list_add_tail(&sh->batch_list, &head->batch_list);
826 spin_unlock(&head->batch_lock);
827 }
828
829 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
830 if (atomic_dec_return(&conf->preread_active_stripes)
831 < IO_THRESHOLD)
832 md_wakeup_thread(conf->mddev->thread);
833
2b6b2457
N		 834 if (test_and_clear_bit(STRIPE_BIT_DELAY, &sh->state)) {
835 int seq = sh->bm_seq;
836 if (test_bit(STRIPE_BIT_DELAY, &sh->batch_head->state) &&
837 sh->batch_head->bm_seq > seq)
838 seq = sh->batch_head->bm_seq;
839 set_bit(STRIPE_BIT_DELAY, &sh->batch_head->state);
840 sh->batch_head->bm_seq = seq;
841 }
842
59fc630b 843 atomic_inc(&sh->count);
844unlock_out:
845 unlock_two_stripes(head, sh);
846out:
6d036f7d 847 raid5_release_stripe(head);
59fc630b 848}
849
05616be5
N		 850/* Determine if 'data_offset' or 'new_data_offset' should be used
851 * in this stripe_head.
852 */
853static int use_new_offset(struct r5conf *conf, struct stripe_head *sh)
854{
855 sector_t progress = conf->reshape_progress;
856 /* Need a memory barrier to make sure we see the value
857 * of conf->generation, or ->data_offset that was set before
858 * reshape_progress was updated.
859 */
860 smp_rmb();
861 if (progress == MaxSector)
862 return 0;
863 if (sh->generation == conf->generation - 1)
864 return 0;
865 /* We are in a reshape, and this is a new-generation stripe,
866 * so use new_data_offset.
867 */
868 return 1;
869}
870
aaf9f12e 871static void dispatch_bio_list(struct bio_list *tmp)
765d704d 872{
765d704d
SL		 873 struct bio *bio;
874
aaf9f12e
SL		 875 while ((bio = bio_list_pop(tmp)))
876 generic_make_request(bio);
877}
878
879static int cmp_stripe(void *priv, struct list_head *a, struct list_head *b)
880{
881 const struct r5pending_data *da = list_entry(a,
882 struct r5pending_data, sibling);
883 const struct r5pending_data *db = list_entry(b,
884 struct r5pending_data, sibling);
885 if (da->sector > db->sector)
886 return 1;
887 if (da->sector < db->sector)
888 return -1;
889 return 0;
890}
891
892static void dispatch_defer_bios(struct r5conf *conf, int target,
893 struct bio_list *list)
894{
895 struct r5pending_data *data;
896 struct list_head *first, *next = NULL;
897 int cnt = 0;
898
899 if (conf->pending_data_cnt == 0)
900 return;
901
902 list_sort(NULL, &conf->pending_list, cmp_stripe);
903
904 first = conf->pending_list.next;
905
906 /* temporarily move the head */
907 if (conf->next_pending_data)
908 list_move_tail(&conf->pending_list,
909 &conf->next_pending_data->sibling);
910
911 while (!list_empty(&conf->pending_list)) {
912 data = list_first_entry(&conf->pending_list,
913 struct r5pending_data, sibling);
914 if (&data->sibling == first)
915 first = data->sibling.next;
916 next = data->sibling.next;
917
918 bio_list_merge(list, &data->bios);
919 list_move(&data->sibling, &conf->free_list);
920 cnt++;
921 if (cnt >= target)
922 break;
923 }
924 conf->pending_data_cnt -= cnt;
925 BUG_ON(conf->pending_data_cnt < 0 || cnt < target);
926
927 if (next != &conf->pending_list)
928 conf->next_pending_data = list_entry(next,
929 struct r5pending_data, sibling);
930 else
931 conf->next_pending_data = NULL;
932 /* list isn't empty */
933 if (first != &conf->pending_list)
934 list_move_tail(&conf->pending_list, first);
935}
936
937static void flush_deferred_bios(struct r5conf *conf)
938{
939 struct bio_list tmp = BIO_EMPTY_LIST;
940
941 if (conf->pending_data_cnt == 0)
765d704d
SL		 942 return;
943
765d704d 944 spin_lock(&conf->pending_bios_lock);
aaf9f12e
SL		 945 dispatch_defer_bios(conf, conf->pending_data_cnt, &tmp);
946 BUG_ON(conf->pending_data_cnt != 0);
765d704d
SL		 947 spin_unlock(&conf->pending_bios_lock);
948
aaf9f12e 949 dispatch_bio_list(&tmp);
765d704d
SL		 950}
951
aaf9f12e
SL		 952static void defer_issue_bios(struct r5conf *conf, sector_t sector,
953 struct bio_list *bios)
765d704d 954{
aaf9f12e
SL		 955 struct bio_list tmp = BIO_EMPTY_LIST;
956 struct r5pending_data *ent;
957
765d704d 958 spin_lock(&conf->pending_bios_lock);
aaf9f12e
SL		 959 ent = list_first_entry(&conf->free_list, struct r5pending_data,
960 sibling);
961 list_move_tail(&ent->sibling, &conf->pending_list);
962 ent->sector = sector;
963 bio_list_init(&ent->bios);
964 bio_list_merge(&ent->bios, bios);
965 conf->pending_data_cnt++;
966 if (conf->pending_data_cnt >= PENDING_IO_MAX)
967 dispatch_defer_bios(conf, PENDING_IO_ONE_FLUSH, &tmp);
968
765d704d 969 spin_unlock(&conf->pending_bios_lock);
aaf9f12e
SL		 970
971 dispatch_bio_list(&tmp);
765d704d
SL		 972}
973
6712ecf8 974static void
4246a0b6 975raid5_end_read_request(struct bio *bi);
6712ecf8 976static void
4246a0b6 977raid5_end_write_request(struct bio *bi);
91c00924 978
c4e5ac0a 979static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
91c00924 980{
d1688a6d 981 struct r5conf *conf = sh->raid_conf;
91c00924 982 int i, disks = sh->disks;
59fc630b 983 struct stripe_head *head_sh = sh;
aaf9f12e
SL		 984 struct bio_list pending_bios = BIO_EMPTY_LIST;
985 bool should_defer;
91c00924
DW		 986
987 might_sleep();
988
ff875738
AP		 989 if (log_stripe(sh, s) == 0)
990 return;
1e6d690b 991
aaf9f12e 992 should_defer = conf->batch_bio_dispatch && conf->group_cnt;
1e6d690b 993
91c00924 994 for (i = disks; i--; ) {
796a5cf0 995 int op, op_flags = 0;
9a3e1101 996 int replace_only = 0;
977df362
N		 997 struct bio *bi, *rbi;
998 struct md_rdev *rdev, *rrdev = NULL;
59fc630b 999
1000 sh = head_sh;
e9c7469b 1001 if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) {
796a5cf0 1002 op = REQ_OP_WRITE;
e9c7469b 1003 if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags))
70fd7614 1004 op_flags = REQ_FUA;
9e444768 1005 if (test_bit(R5_Discard, &sh->dev[i].flags))
796a5cf0 1006 op = REQ_OP_DISCARD;
e9c7469b 1007 } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags))
796a5cf0 1008 op = REQ_OP_READ;
9a3e1101
N		 1009 else if (test_and_clear_bit(R5_WantReplace,
1010 &sh->dev[i].flags)) {
796a5cf0 1011 op = REQ_OP_WRITE;
9a3e1101
N		 1012 replace_only = 1;
1013 } else
91c00924 1014 continue;
bc0934f0 1015 if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags))
796a5cf0 1016 op_flags |= REQ_SYNC;
91c00924 1017
59fc630b 1018again:
91c00924 1019 bi = &sh->dev[i].req;
977df362 1020 rbi = &sh->dev[i].rreq; /* For writing to replacement */
91c00924 1021
91c00924 1022 rcu_read_lock();
9a3e1101 1023 rrdev = rcu_dereference(conf->disks[i].replacement);
dd054fce
N		 1024 smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */
1025 rdev = rcu_dereference(conf->disks[i].rdev);
1026 if (!rdev) {
1027 rdev = rrdev;
1028 rrdev = NULL;
1029 }
796a5cf0 1030 if (op_is_write(op)) {
9a3e1101
N		 1031 if (replace_only)
1032 rdev = NULL;
dd054fce
N		 1033 if (rdev == rrdev)
1034 /* We raced and saw duplicates */
1035 rrdev = NULL;
9a3e1101 1036 } else {
59fc630b 1037 if (test_bit(R5_ReadRepl, &head_sh->dev[i].flags) && rrdev)
9a3e1101
N		 1038 rdev = rrdev;
1039 rrdev = NULL;
1040 }
977df362 1041
91c00924
DW		 1042 if (rdev && test_bit(Faulty, &rdev->flags))
1043 rdev = NULL;
1044 if (rdev)
1045 atomic_inc(&rdev->nr_pending);
977df362
N		 1046 if (rrdev && test_bit(Faulty, &rrdev->flags))
1047 rrdev = NULL;
1048 if (rrdev)
1049 atomic_inc(&rrdev->nr_pending);
91c00924
DW		 1050 rcu_read_unlock();
1051
73e92e51 1052 /* We have already checked bad blocks for reads. Now
977df362
N		 1053 * need to check for writes. We never accept write errors
1054 * on the replacement, so we don't to check rrdev.
73e92e51 1055 */
796a5cf0 1056 while (op_is_write(op) && rdev &&
73e92e51
N		 1057 test_bit(WriteErrorSeen, &rdev->flags)) {
1058 sector_t first_bad;
1059 int bad_sectors;
1060 int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
1061 &first_bad, &bad_sectors);
1062 if (!bad)
1063 break;
1064
1065 if (bad < 0) {
1066 set_bit(BlockedBadBlocks, &rdev->flags);
1067 if (!conf->mddev->external &&
2953079c 1068 conf->mddev->sb_flags) {
73e92e51
N		 1069 /* It is very unlikely, but we might
1070 * still need to write out the
1071 * bad block log - better give it
1072 * a chance*/
1073 md_check_recovery(conf->mddev);
1074 }
1850753d 1075 /*
1076 * Because md_wait_for_blocked_rdev
1077 * will dec nr_pending, we must
1078 * increment it first.
1079 */
1080 atomic_inc(&rdev->nr_pending);
73e92e51
N		 1081 md_wait_for_blocked_rdev(rdev, conf->mddev);
1082 } else {
1083 /* Acknowledged bad block - skip the write */
1084 rdev_dec_pending(rdev, conf->mddev);
1085 rdev = NULL;
1086 }
1087 }
1088
91c00924 1089 if (rdev) {
9a3e1101
N		 1090 if (s->syncing || s->expanding || s->expanded
1091 || s->replacing)
91c00924
DW		 1092 md_sync_acct(rdev->bdev, STRIPE_SECTORS);
1093
2b7497f0
DW		 1094 set_bit(STRIPE_IO_STARTED, &sh->state);
1095
74d46992 1096 bio_set_dev(bi, rdev->bdev);
796a5cf0
MC		 1097 bio_set_op_attrs(bi, op, op_flags);
1098 bi->bi_end_io = op_is_write(op)
2f6db2a7
KO		 1099 ? raid5_end_write_request
1100 : raid5_end_read_request;
1101 bi->bi_private = sh;
1102
6296b960 1103 pr_debug("%s: for %llu schedule op %d on disc %d\n",
e46b272b 1104 __func__, (unsigned long long)sh->sector,
1eff9d32 1105 bi->bi_opf, i);
91c00924 1106 atomic_inc(&sh->count);
59fc630b 1107 if (sh != head_sh)
1108 atomic_inc(&head_sh->count);
05616be5 1109 if (use_new_offset(conf, sh))
4f024f37 1110 bi->bi_iter.bi_sector = (sh->sector
05616be5
N		 1111 + rdev->new_data_offset);
1112 else
4f024f37 1113 bi->bi_iter.bi_sector = (sh->sector
05616be5 1114 + rdev->data_offset);
59fc630b 1115 if (test_bit(R5_ReadNoMerge, &head_sh->dev[i].flags))
1eff9d32 1116 bi->bi_opf |= REQ_NOMERGE;
3f9e7c14 1117
d592a996
SL		 1118 if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
1119 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
86aa1397
SL		 1120
1121 if (!op_is_write(op) &&
1122 test_bit(R5_InJournal, &sh->dev[i].flags))
1123 /*
1124 * issuing read for a page in journal, this
1125 * must be preparing for prexor in rmw; read
1126 * the data into orig_page
1127 */
1128 sh->dev[i].vec.bv_page = sh->dev[i].orig_page;
1129 else
1130 sh->dev[i].vec.bv_page = sh->dev[i].page;
4997b72e 1131 bi->bi_vcnt = 1;
91c00924
DW		 1132 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
1133 bi->bi_io_vec[0].bv_offset = 0;
4f024f37 1134 bi->bi_iter.bi_size = STRIPE_SIZE;
2cd259a7
MD		 1135 bi->bi_write_hint = sh->dev[i].write_hint;
1136 if (!rrdev)
f1934892 1137 sh->dev[i].write_hint = RWH_WRITE_LIFE_NOT_SET;
37c61ff3
SL		 1138 /*
1139 * If this is discard request, set bi_vcnt 0. We don't
1140 * want to confuse SCSI because SCSI will replace payload
1141 */
796a5cf0 1142 if (op == REQ_OP_DISCARD)
37c61ff3 1143 bi->bi_vcnt = 0;
977df362
N		 1144 if (rrdev)
1145 set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags);
e3620a3a
JB		 1146
1147 if (conf->mddev->gendisk)
74d46992 1148 trace_block_bio_remap(bi->bi_disk->queue,
e3620a3a
JB		 1149 bi, disk_devt(conf->mddev->gendisk),
1150 sh->dev[i].sector);
aaf9f12e
SL		 1151 if (should_defer && op_is_write(op))
1152 bio_list_add(&pending_bios, bi);
1153 else
1154 generic_make_request(bi);
977df362
N		 1155 }
1156 if (rrdev) {
9a3e1101
N		 1157 if (s->syncing || s->expanding || s->expanded
1158 || s->replacing)
977df362
N		 1159 md_sync_acct(rrdev->bdev, STRIPE_SECTORS);
1160
1161 set_bit(STRIPE_IO_STARTED, &sh->state);
1162
74d46992 1163 bio_set_dev(rbi, rrdev->bdev);
796a5cf0
MC		 1164 bio_set_op_attrs(rbi, op, op_flags);
1165 BUG_ON(!op_is_write(op));
2f6db2a7
KO		 1166 rbi->bi_end_io = raid5_end_write_request;
1167 rbi->bi_private = sh;
1168
6296b960 1169 pr_debug("%s: for %llu schedule op %d on "
977df362
N		 1170 "replacement disc %d\n",
1171 __func__, (unsigned long long)sh->sector,
1eff9d32 1172 rbi->bi_opf, i);
977df362 1173 atomic_inc(&sh->count);
59fc630b 1174 if (sh != head_sh)
1175 atomic_inc(&head_sh->count);
05616be5 1176 if (use_new_offset(conf, sh))
4f024f37 1177 rbi->bi_iter.bi_sector = (sh->sector
05616be5
N		 1178 + rrdev->new_data_offset);
1179 else
4f024f37 1180 rbi->bi_iter.bi_sector = (sh->sector
05616be5 1181 + rrdev->data_offset);
d592a996
SL		 1182 if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
1183 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
1184 sh->dev[i].rvec.bv_page = sh->dev[i].page;
4997b72e 1185 rbi->bi_vcnt = 1;
977df362
N		 1186 rbi->bi_io_vec[0].bv_len = STRIPE_SIZE;
1187 rbi->bi_io_vec[0].bv_offset = 0;
4f024f37 1188 rbi->bi_iter.bi_size = STRIPE_SIZE;
2cd259a7 1189 rbi->bi_write_hint = sh->dev[i].write_hint;
f1934892 1190 sh->dev[i].write_hint = RWH_WRITE_LIFE_NOT_SET;
37c61ff3
SL		 1191 /*
1192 * If this is discard request, set bi_vcnt 0. We don't
1193 * want to confuse SCSI because SCSI will replace payload
1194 */
796a5cf0 1195 if (op == REQ_OP_DISCARD)
37c61ff3 1196 rbi->bi_vcnt = 0;
e3620a3a 1197 if (conf->mddev->gendisk)
74d46992 1198 trace_block_bio_remap(rbi->bi_disk->queue,
e3620a3a
JB		 1199 rbi, disk_devt(conf->mddev->gendisk),
1200 sh->dev[i].sector);
aaf9f12e
SL		 1201 if (should_defer && op_is_write(op))
1202 bio_list_add(&pending_bios, rbi);
1203 else
1204 generic_make_request(rbi);
977df362
N		 1205 }
1206 if (!rdev && !rrdev) {
796a5cf0 1207 if (op_is_write(op))
91c00924 1208 set_bit(STRIPE_DEGRADED, &sh->state);
6296b960 1209 pr_debug("skip op %d on disc %d for sector %llu\n",
1eff9d32 1210 bi->bi_opf, i, (unsigned long long)sh->sector);
91c00924
DW		 1211 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1212 set_bit(STRIPE_HANDLE, &sh->state);
1213 }
59fc630b 1214
1215 if (!head_sh->batch_head)
1216 continue;
1217 sh = list_first_entry(&sh->batch_list, struct stripe_head,
1218 batch_list);
1219 if (sh != head_sh)
1220 goto again;
91c00924 1221 }
aaf9f12e
SL		 1222
1223 if (should_defer && !bio_list_empty(&pending_bios))
1224 defer_issue_bios(conf, head_sh->sector, &pending_bios);
91c00924
DW		 1225}
1226
1227static struct dma_async_tx_descriptor *
d592a996
SL		 1228async_copy_data(int frombio, struct bio *bio, struct page **page,
1229 sector_t sector, struct dma_async_tx_descriptor *tx,
1e6d690b 1230 struct stripe_head *sh, int no_skipcopy)
91c00924 1231{
7988613b
KO		 1232 struct bio_vec bvl;
1233 struct bvec_iter iter;
91c00924 1234 struct page *bio_page;
91c00924 1235 int page_offset;
a08abd8c 1236 struct async_submit_ctl submit;
0403e382 1237 enum async_tx_flags flags = 0;
91c00924 1238
4f024f37
KO		 1239 if (bio->bi_iter.bi_sector >= sector)
1240 page_offset = (signed)(bio->bi_iter.bi_sector - sector) * 512;
91c00924 1241 else
4f024f37 1242 page_offset = (signed)(sector - bio->bi_iter.bi_sector) * -512;
a08abd8c 1243
0403e382
DW		 1244 if (frombio)
1245 flags |= ASYNC_TX_FENCE;
1246 init_async_submit(&submit, flags, tx, NULL, NULL, NULL);
1247
7988613b
KO		 1248 bio_for_each_segment(bvl, bio, iter) {
1249 int len = bvl.bv_len;
91c00924
DW		 1250 int clen;
1251 int b_offset = 0;
1252
1253 if (page_offset < 0) {
1254 b_offset = -page_offset;
1255 page_offset += b_offset;
1256 len -= b_offset;
1257 }
1258
1259 if (len > 0 && page_offset + len > STRIPE_SIZE)
1260 clen = STRIPE_SIZE - page_offset;
1261 else
1262 clen = len;
1263
1264 if (clen > 0) {
7988613b
KO		 1265 b_offset += bvl.bv_offset;
1266 bio_page = bvl.bv_page;
d592a996
SL		 1267 if (frombio) {
1268 if (sh->raid_conf->skip_copy &&
1269 b_offset == 0 && page_offset == 0 &&
1e6d690b
SL		 1270 clen == STRIPE_SIZE &&
1271 !no_skipcopy)
d592a996
SL		 1272 *page = bio_page;
1273 else
1274 tx = async_memcpy(*page, bio_page, page_offset,
a08abd8c 1275 b_offset, clen, &submit);
d592a996
SL		 1276 } else
1277 tx = async_memcpy(bio_page, *page, b_offset,
a08abd8c 1278 page_offset, clen, &submit);
91c00924 1279 }
a08abd8c
DW		 1280 /* chain the operations */
1281 submit.depend_tx = tx;
1282
91c00924
DW		 1283 if (clen < len) /* hit end of page */
1284 break;
1285 page_offset += len;
1286 }
1287
1288 return tx;
1289}
1290
1291static void ops_complete_biofill(void *stripe_head_ref)
1292{
1293 struct stripe_head *sh = stripe_head_ref;
e4d84909 1294 int i;
91c00924 1295
e46b272b 1296 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1297 (unsigned long long)sh->sector);
1298
1299 /* clear completed biofills */
1300 for (i = sh->disks; i--; ) {
1301 struct r5dev *dev = &sh->dev[i];
91c00924
DW		 1302
1303 /* acknowledge completion of a biofill operation */
e4d84909
DW		 1304 /* and check if we need to reply to a read request,
1305 * new R5_Wantfill requests are held off until
83de75cc 1306 * !STRIPE_BIOFILL_RUN
e4d84909
DW		 1307 */
1308 if (test_and_clear_bit(R5_Wantfill, &dev->flags)) {
91c00924 1309 struct bio *rbi, *rbi2;
91c00924 1310
91c00924
DW		 1311 BUG_ON(!dev->read);
1312 rbi = dev->read;
1313 dev->read = NULL;
4f024f37 1314 while (rbi && rbi->bi_iter.bi_sector <
91c00924
DW		 1315 dev->sector + STRIPE_SECTORS) {
1316 rbi2 = r5_next_bio(rbi, dev->sector);
016c76ac 1317 bio_endio(rbi);
91c00924
DW		 1318 rbi = rbi2;
1319 }
1320 }
1321 }
83de75cc 1322 clear_bit(STRIPE_BIOFILL_RUN, &sh->state);
91c00924 1323
e4d84909 1324 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 1325 raid5_release_stripe(sh);
91c00924
DW		 1326}
1327
1328static void ops_run_biofill(struct stripe_head *sh)
1329{
1330 struct dma_async_tx_descriptor *tx = NULL;
a08abd8c 1331 struct async_submit_ctl submit;
91c00924
DW		 1332 int i;
1333
59fc630b 1334 BUG_ON(sh->batch_head);
e46b272b 1335 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1336 (unsigned long long)sh->sector);
1337
1338 for (i = sh->disks; i--; ) {
1339 struct r5dev *dev = &sh->dev[i];
1340 if (test_bit(R5_Wantfill, &dev->flags)) {
1341 struct bio *rbi;
b17459c0 1342 spin_lock_irq(&sh->stripe_lock);
91c00924
DW		 1343 dev->read = rbi = dev->toread;
1344 dev->toread = NULL;
b17459c0 1345 spin_unlock_irq(&sh->stripe_lock);
4f024f37 1346 while (rbi && rbi->bi_iter.bi_sector <
91c00924 1347 dev->sector + STRIPE_SECTORS) {
d592a996 1348 tx = async_copy_data(0, rbi, &dev->page,
1e6d690b 1349 dev->sector, tx, sh, 0);
91c00924
DW		 1350 rbi = r5_next_bio(rbi, dev->sector);
1351 }
1352 }
1353 }
1354
1355 atomic_inc(&sh->count);
a08abd8c
DW		 1356 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL);
1357 async_trigger_callback(&submit);
91c00924
DW		 1358}
1359
4e7d2c0a 1360static void mark_target_uptodate(struct stripe_head *sh, int target)
91c00924 1361{
4e7d2c0a 1362 struct r5dev *tgt;
91c00924 1363
4e7d2c0a
DW		 1364 if (target < 0)
1365 return;
91c00924 1366
4e7d2c0a 1367 tgt = &sh->dev[target];
91c00924
DW		 1368 set_bit(R5_UPTODATE, &tgt->flags);
1369 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1370 clear_bit(R5_Wantcompute, &tgt->flags);
4e7d2c0a
DW		 1371}
1372
ac6b53b6 1373static void ops_complete_compute(void *stripe_head_ref)
91c00924
DW		 1374{
1375 struct stripe_head *sh = stripe_head_ref;
91c00924 1376
e46b272b 1377 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1378 (unsigned long long)sh->sector);
1379
ac6b53b6 1380 /* mark the computed target(s) as uptodate */
4e7d2c0a 1381 mark_target_uptodate(sh, sh->ops.target);
ac6b53b6 1382 mark_target_uptodate(sh, sh->ops.target2);
4e7d2c0a 1383
ecc65c9b
DW		 1384 clear_bit(STRIPE_COMPUTE_RUN, &sh->state);
1385 if (sh->check_state == check_state_compute_run)
1386 sh->check_state = check_state_compute_result;
91c00924 1387 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 1388 raid5_release_stripe(sh);
91c00924
DW		 1389}
1390
d6f38f31 1391/* return a pointer to the address conversion region of the scribble buffer */
b330e6a4 1392static struct page **to_addr_page(struct raid5_percpu *percpu, int i)
d6f38f31 1393{
b330e6a4 1394 return percpu->scribble + i * percpu->scribble_obj_size;
46d5b785 1395}
1396
1397/* return a pointer to the address conversion region of the scribble buffer */
b330e6a4
KO		 1398static addr_conv_t *to_addr_conv(struct stripe_head *sh,
1399 struct raid5_percpu *percpu, int i)
46d5b785 1400{
b330e6a4 1401 return (void *) (to_addr_page(percpu, i) + sh->disks + 2);
d6f38f31
DW		 1402}
1403
1404static struct dma_async_tx_descriptor *
1405ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 1406{
91c00924 1407 int disks = sh->disks;
46d5b785 1408 struct page **xor_srcs = to_addr_page(percpu, 0);
91c00924
DW		 1409 int target = sh->ops.target;
1410 struct r5dev *tgt = &sh->dev[target];
1411 struct page *xor_dest = tgt->page;
1412 int count = 0;
1413 struct dma_async_tx_descriptor *tx;
a08abd8c 1414 struct async_submit_ctl submit;
91c00924
DW		 1415 int i;
1416
59fc630b 1417 BUG_ON(sh->batch_head);
1418
91c00924 1419 pr_debug("%s: stripe %llu block: %d\n",
e46b272b 1420 __func__, (unsigned long long)sh->sector, target);
91c00924
DW		 1421 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1422
1423 for (i = disks; i--; )
1424 if (i != target)
1425 xor_srcs[count++] = sh->dev[i].page;
1426
1427 atomic_inc(&sh->count);
1428
0403e382 1429 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL,
46d5b785 1430 ops_complete_compute, sh, to_addr_conv(sh, percpu, 0));
91c00924 1431 if (unlikely(count == 1))
a08abd8c 1432 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
91c00924 1433 else
a08abd8c 1434 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924 1435
91c00924
DW		 1436 return tx;
1437}
1438
ac6b53b6
DW		 1439/* set_syndrome_sources - populate source buffers for gen_syndrome
1440 * @srcs - (struct page *) array of size sh->disks
1441 * @sh - stripe_head to parse
1442 *
1443 * Populates srcs in proper layout order for the stripe and returns the
1444 * 'count' of sources to be used in a call to async_gen_syndrome. The P
1445 * destination buffer is recorded in srcs[count] and the Q destination
1446 * is recorded in srcs[count+1]].
1447 */
584acdd4
MS		 1448static int set_syndrome_sources(struct page **srcs,
1449 struct stripe_head *sh,
1450 int srctype)
ac6b53b6
DW		 1451{
1452 int disks = sh->disks;
1453 int syndrome_disks = sh->ddf_layout ? disks : (disks - 2);
1454 int d0_idx = raid6_d0(sh);
1455 int count;
1456 int i;
1457
1458 for (i = 0; i < disks; i++)
5dd33c9a 1459 srcs[i] = NULL;
ac6b53b6
DW		 1460
1461 count = 0;
1462 i = d0_idx;
1463 do {
1464 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
584acdd4 1465 struct r5dev *dev = &sh->dev[i];
ac6b53b6 1466
584acdd4
MS		 1467 if (i == sh->qd_idx || i == sh->pd_idx ||
1468 (srctype == SYNDROME_SRC_ALL) ||
1469 (srctype == SYNDROME_SRC_WANT_DRAIN &&
1e6d690b
SL		 1470 (test_bit(R5_Wantdrain, &dev->flags) ||
1471 test_bit(R5_InJournal, &dev->flags))) ||
584acdd4 1472 (srctype == SYNDROME_SRC_WRITTEN &&
0977762f
SL		 1473 (dev->written ||
1474 test_bit(R5_InJournal, &dev->flags)))) {
1e6d690b
SL		 1475 if (test_bit(R5_InJournal, &dev->flags))
1476 srcs[slot] = sh->dev[i].orig_page;
1477 else
1478 srcs[slot] = sh->dev[i].page;
1479 }
ac6b53b6
DW		 1480 i = raid6_next_disk(i, disks);
1481 } while (i != d0_idx);
ac6b53b6 1482
e4424fee 1483 return syndrome_disks;
ac6b53b6
DW		 1484}
1485
1486static struct dma_async_tx_descriptor *
1487ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu)
1488{
1489 int disks = sh->disks;
46d5b785 1490 struct page **blocks = to_addr_page(percpu, 0);
ac6b53b6
DW		 1491 int target;
1492 int qd_idx = sh->qd_idx;
1493 struct dma_async_tx_descriptor *tx;
1494 struct async_submit_ctl submit;
1495 struct r5dev *tgt;
1496 struct page *dest;
1497 int i;
1498 int count;
1499
59fc630b 1500 BUG_ON(sh->batch_head);
ac6b53b6
DW		 1501 if (sh->ops.target < 0)
1502 target = sh->ops.target2;
1503 else if (sh->ops.target2 < 0)
1504 target = sh->ops.target;
91c00924 1505 else
ac6b53b6
DW		 1506 /* we should only have one valid target */
1507 BUG();
1508 BUG_ON(target < 0);
1509 pr_debug("%s: stripe %llu block: %d\n",
1510 __func__, (unsigned long long)sh->sector, target);
1511
1512 tgt = &sh->dev[target];
1513 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1514 dest = tgt->page;
1515
1516 atomic_inc(&sh->count);
1517
1518 if (target == qd_idx) {
584acdd4 1519 count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL);
ac6b53b6
DW		 1520 blocks[count] = NULL; /* regenerating p is not necessary */
1521 BUG_ON(blocks[count+1] != dest); /* q should already be set */
0403e382
DW		 1522 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1523 ops_complete_compute, sh,
46d5b785 1524 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1525 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
1526 } else {
1527 /* Compute any data- or p-drive using XOR */
1528 count = 0;
1529 for (i = disks; i-- ; ) {
1530 if (i == target || i == qd_idx)
1531 continue;
1532 blocks[count++] = sh->dev[i].page;
1533 }
1534
0403e382
DW		 1535 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1536 NULL, ops_complete_compute, sh,
46d5b785 1537 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1538 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit);
1539 }
91c00924 1540
91c00924
DW		 1541 return tx;
1542}
1543
ac6b53b6
DW		 1544static struct dma_async_tx_descriptor *
1545ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu)
1546{
1547 int i, count, disks = sh->disks;
1548 int syndrome_disks = sh->ddf_layout ? disks : disks-2;
1549 int d0_idx = raid6_d0(sh);
1550 int faila = -1, failb = -1;
1551 int target = sh->ops.target;
1552 int target2 = sh->ops.target2;
1553 struct r5dev *tgt = &sh->dev[target];
1554 struct r5dev *tgt2 = &sh->dev[target2];
1555 struct dma_async_tx_descriptor *tx;
46d5b785 1556 struct page **blocks = to_addr_page(percpu, 0);
ac6b53b6
DW		 1557 struct async_submit_ctl submit;
1558
59fc630b 1559 BUG_ON(sh->batch_head);
ac6b53b6
DW		 1560 pr_debug("%s: stripe %llu block1: %d block2: %d\n",
1561 __func__, (unsigned long long)sh->sector, target, target2);
1562 BUG_ON(target < 0 || target2 < 0);
1563 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1564 BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags));
1565
6c910a78 1566 /* we need to open-code set_syndrome_sources to handle the
ac6b53b6
DW		 1567 * slot number conversion for 'faila' and 'failb'
1568 */
1569 for (i = 0; i < disks ; i++)
5dd33c9a 1570 blocks[i] = NULL;
ac6b53b6
DW		 1571 count = 0;
1572 i = d0_idx;
1573 do {
1574 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
1575
1576 blocks[slot] = sh->dev[i].page;
1577
1578 if (i == target)
1579 faila = slot;
1580 if (i == target2)
1581 failb = slot;
1582 i = raid6_next_disk(i, disks);
1583 } while (i != d0_idx);
ac6b53b6
DW		 1584
1585 BUG_ON(faila == failb);
1586 if (failb < faila)
1587 swap(faila, failb);
1588 pr_debug("%s: stripe: %llu faila: %d failb: %d\n",
1589 __func__, (unsigned long long)sh->sector, faila, failb);
1590
1591 atomic_inc(&sh->count);
1592
1593 if (failb == syndrome_disks+1) {
1594 /* Q disk is one of the missing disks */
1595 if (faila == syndrome_disks) {
1596 /* Missing P+Q, just recompute */
0403e382
DW		 1597 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1598 ops_complete_compute, sh,
46d5b785 1599 to_addr_conv(sh, percpu, 0));
e4424fee 1600 return async_gen_syndrome(blocks, 0, syndrome_disks+2,
ac6b53b6
DW		 1601 STRIPE_SIZE, &submit);
1602 } else {
1603 struct page *dest;
1604 int data_target;
1605 int qd_idx = sh->qd_idx;
1606
1607 /* Missing D+Q: recompute D from P, then recompute Q */
1608 if (target == qd_idx)
1609 data_target = target2;
1610 else
1611 data_target = target;
1612
1613 count = 0;
1614 for (i = disks; i-- ; ) {
1615 if (i == data_target || i == qd_idx)
1616 continue;
1617 blocks[count++] = sh->dev[i].page;
1618 }
1619 dest = sh->dev[data_target].page;
0403e382
DW		 1620 init_async_submit(&submit,
1621 ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1622 NULL, NULL, NULL,
46d5b785 1623 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1624 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE,
1625 &submit);
1626
584acdd4 1627 count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL);
0403e382
DW		 1628 init_async_submit(&submit, ASYNC_TX_FENCE, tx,
1629 ops_complete_compute, sh,
46d5b785 1630 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1631 return async_gen_syndrome(blocks, 0, count+2,
1632 STRIPE_SIZE, &submit);
1633 }
ac6b53b6 1634 } else {
6c910a78
DW		 1635 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1636 ops_complete_compute, sh,
46d5b785 1637 to_addr_conv(sh, percpu, 0));
6c910a78
DW		 1638 if (failb == syndrome_disks) {
1639 /* We're missing D+P. */
1640 return async_raid6_datap_recov(syndrome_disks+2,
1641 STRIPE_SIZE, faila,
1642 blocks, &submit);
1643 } else {
1644 /* We're missing D+D. */
1645 return async_raid6_2data_recov(syndrome_disks+2,
1646 STRIPE_SIZE, faila, failb,
1647 blocks, &submit);
1648 }
ac6b53b6
DW		 1649 }
1650}
1651
91c00924
DW		 1652static void ops_complete_prexor(void *stripe_head_ref)
1653{
1654 struct stripe_head *sh = stripe_head_ref;
1655
e46b272b 1656 pr_debug("%s: stripe %llu\n", __func__,
91c00924 1657 (unsigned long long)sh->sector);
1e6d690b
SL		 1658
1659 if (r5c_is_writeback(sh->raid_conf->log))
1660 /*
1661 * raid5-cache write back uses orig_page during prexor.
1662 * After prexor, it is time to free orig_page
1663 */
1664 r5c_release_extra_page(sh);
91c00924
DW		 1665}
1666
1667static struct dma_async_tx_descriptor *
584acdd4
MS		 1668ops_run_prexor5(struct stripe_head *sh, struct raid5_percpu *percpu,
1669 struct dma_async_tx_descriptor *tx)
91c00924 1670{
91c00924 1671 int disks = sh->disks;
46d5b785 1672 struct page **xor_srcs = to_addr_page(percpu, 0);
91c00924 1673 int count = 0, pd_idx = sh->pd_idx, i;
a08abd8c 1674 struct async_submit_ctl submit;
91c00924
DW		 1675
1676 /* existing parity data subtracted */
1677 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1678
59fc630b 1679 BUG_ON(sh->batch_head);
e46b272b 1680 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1681 (unsigned long long)sh->sector);
1682
1683 for (i = disks; i--; ) {
1684 struct r5dev *dev = &sh->dev[i];
1685 /* Only process blocks that are known to be uptodate */
1e6d690b
SL		 1686 if (test_bit(R5_InJournal, &dev->flags))
1687 xor_srcs[count++] = dev->orig_page;
1688 else if (test_bit(R5_Wantdrain, &dev->flags))
91c00924
DW		 1689 xor_srcs[count++] = dev->page;
1690 }
1691
0403e382 1692 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
46d5b785 1693 ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0));
a08abd8c 1694 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924
DW		 1695
1696 return tx;
1697}
1698
584acdd4
MS		 1699static struct dma_async_tx_descriptor *
1700ops_run_prexor6(struct stripe_head *sh, struct raid5_percpu *percpu,
1701 struct dma_async_tx_descriptor *tx)
1702{
1703 struct page **blocks = to_addr_page(percpu, 0);
1704 int count;
1705 struct async_submit_ctl submit;
1706
1707 pr_debug("%s: stripe %llu\n", __func__,
1708 (unsigned long long)sh->sector);
1709
1710 count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_WANT_DRAIN);
1711
1712 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_PQ_XOR_DST, tx,
1713 ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0));
1714 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
1715
1716 return tx;
1717}
1718
91c00924 1719static struct dma_async_tx_descriptor *
d8ee0728 1720ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
91c00924 1721{
1e6d690b 1722 struct r5conf *conf = sh->raid_conf;
91c00924 1723 int disks = sh->disks;
d8ee0728 1724 int i;
59fc630b 1725 struct stripe_head *head_sh = sh;
91c00924 1726
e46b272b 1727 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1728 (unsigned long long)sh->sector);
1729
1730 for (i = disks; i--; ) {
59fc630b 1731 struct r5dev *dev;
91c00924 1732 struct bio *chosen;
91c00924 1733
59fc630b 1734 sh = head_sh;
1735 if (test_and_clear_bit(R5_Wantdrain, &head_sh->dev[i].flags)) {
91c00924
DW		 1736 struct bio *wbi;
1737
59fc630b 1738again:
1739 dev = &sh->dev[i];
1e6d690b
SL		 1740 /*
1741 * clear R5_InJournal, so when rewriting a page in
1742 * journal, it is not skipped by r5l_log_stripe()
1743 */
1744 clear_bit(R5_InJournal, &dev->flags);
b17459c0 1745 spin_lock_irq(&sh->stripe_lock);
91c00924
DW		 1746 chosen = dev->towrite;
1747 dev->towrite = NULL;
7a87f434 1748 sh->overwrite_disks = 0;
91c00924
DW		 1749 BUG_ON(dev->written);
1750 wbi = dev->written = chosen;
b17459c0 1751 spin_unlock_irq(&sh->stripe_lock);
d592a996 1752 WARN_ON(dev->page != dev->orig_page);
91c00924 1753
4f024f37 1754 while (wbi && wbi->bi_iter.bi_sector <
91c00924 1755 dev->sector + STRIPE_SECTORS) {
1eff9d32 1756 if (wbi->bi_opf & REQ_FUA)
e9c7469b 1757 set_bit(R5_WantFUA, &dev->flags);
1eff9d32 1758 if (wbi->bi_opf & REQ_SYNC)
bc0934f0 1759 set_bit(R5_SyncIO, &dev->flags);
796a5cf0 1760 if (bio_op(wbi) == REQ_OP_DISCARD)
620125f2 1761 set_bit(R5_Discard, &dev->flags);
d592a996
SL		 1762 else {
1763 tx = async_copy_data(1, wbi, &dev->page,
1e6d690b
SL		 1764 dev->sector, tx, sh,
1765 r5c_is_writeback(conf->log));
1766 if (dev->page != dev->orig_page &&
1767 !r5c_is_writeback(conf->log)) {
d592a996
SL		 1768 set_bit(R5_SkipCopy, &dev->flags);
1769 clear_bit(R5_UPTODATE, &dev->flags);
1770 clear_bit(R5_OVERWRITE, &dev->flags);
1771 }
1772 }
91c00924
DW		 1773 wbi = r5_next_bio(wbi, dev->sector);
1774 }
59fc630b 1775
1776 if (head_sh->batch_head) {
1777 sh = list_first_entry(&sh->batch_list,
1778 struct stripe_head,
1779 batch_list);
1780 if (sh == head_sh)
1781 continue;
1782 goto again;
1783 }
91c00924
DW		 1784 }
1785 }
1786
1787 return tx;
1788}
1789
ac6b53b6 1790static void ops_complete_reconstruct(void *stripe_head_ref)
91c00924
DW		 1791{
1792 struct stripe_head *sh = stripe_head_ref;
ac6b53b6
DW		 1793 int disks = sh->disks;
1794 int pd_idx = sh->pd_idx;
1795 int qd_idx = sh->qd_idx;
1796 int i;
9e444768 1797 bool fua = false, sync = false, discard = false;
91c00924 1798
e46b272b 1799 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1800 (unsigned long long)sh->sector);
1801
bc0934f0 1802 for (i = disks; i--; ) {
e9c7469b 1803 fua |= test_bit(R5_WantFUA, &sh->dev[i].flags);
bc0934f0 1804 sync |= test_bit(R5_SyncIO, &sh->dev[i].flags);
9e444768 1805 discard |= test_bit(R5_Discard, &sh->dev[i].flags);
bc0934f0 1806 }
e9c7469b 1807
91c00924
DW		 1808 for (i = disks; i--; ) {
1809 struct r5dev *dev = &sh->dev[i];
ac6b53b6 1810
e9c7469b 1811 if (dev->written || i == pd_idx || i == qd_idx) {
235b6003 1812 if (!discard && !test_bit(R5_SkipCopy, &dev->flags)) {
9e444768 1813 set_bit(R5_UPTODATE, &dev->flags);
235b6003
N		 1814 if (test_bit(STRIPE_EXPAND_READY, &sh->state))
1815 set_bit(R5_Expanded, &dev->flags);
1816 }
e9c7469b
TH		 1817 if (fua)
1818 set_bit(R5_WantFUA, &dev->flags);
bc0934f0
SL		 1819 if (sync)
1820 set_bit(R5_SyncIO, &dev->flags);
e9c7469b 1821 }
91c00924
DW		 1822 }
1823
d8ee0728
DW		 1824 if (sh->reconstruct_state == reconstruct_state_drain_run)
1825 sh->reconstruct_state = reconstruct_state_drain_result;
1826 else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run)
1827 sh->reconstruct_state = reconstruct_state_prexor_drain_result;
1828 else {
1829 BUG_ON(sh->reconstruct_state != reconstruct_state_run);
1830 sh->reconstruct_state = reconstruct_state_result;
1831 }
91c00924
DW		 1832
1833 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 1834 raid5_release_stripe(sh);
91c00924
DW		 1835}
1836
1837static void
ac6b53b6
DW		 1838ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu,
1839 struct dma_async_tx_descriptor *tx)
91c00924 1840{
91c00924 1841 int disks = sh->disks;
59fc630b 1842 struct page **xor_srcs;
a08abd8c 1843 struct async_submit_ctl submit;
59fc630b 1844 int count, pd_idx = sh->pd_idx, i;
91c00924 1845 struct page *xor_dest;
d8ee0728 1846 int prexor = 0;
91c00924 1847 unsigned long flags;
59fc630b 1848 int j = 0;
1849 struct stripe_head *head_sh = sh;
1850 int last_stripe;
91c00924 1851
e46b272b 1852 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1853 (unsigned long long)sh->sector);
1854
620125f2
SL		 1855 for (i = 0; i < sh->disks; i++) {
1856 if (pd_idx == i)
1857 continue;
1858 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1859 break;
1860 }
1861 if (i >= sh->disks) {
1862 atomic_inc(&sh->count);
620125f2
SL		 1863 set_bit(R5_Discard, &sh->dev[pd_idx].flags);
1864 ops_complete_reconstruct(sh);
1865 return;
1866 }
59fc630b 1867again:
1868 count = 0;
1869 xor_srcs = to_addr_page(percpu, j);
91c00924
DW		 1870 /* check if prexor is active which means only process blocks
1871 * that are part of a read-modify-write (written)
1872 */
59fc630b 1873 if (head_sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
d8ee0728 1874 prexor = 1;
91c00924
DW		 1875 xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1876 for (i = disks; i--; ) {
1877 struct r5dev *dev = &sh->dev[i];
1e6d690b
SL		 1878 if (head_sh->dev[i].written ||
1879 test_bit(R5_InJournal, &head_sh->dev[i].flags))
91c00924
DW		 1880 xor_srcs[count++] = dev->page;
1881 }
1882 } else {
1883 xor_dest = sh->dev[pd_idx].page;
1884 for (i = disks; i--; ) {
1885 struct r5dev *dev = &sh->dev[i];
1886 if (i != pd_idx)
1887 xor_srcs[count++] = dev->page;
1888 }
1889 }
1890
91c00924
DW		 1891 /* 1/ if we prexor'd then the dest is reused as a source
1892 * 2/ if we did not prexor then we are redoing the parity
1893 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
1894 * for the synchronous xor case
1895 */
59fc630b 1896 last_stripe = !head_sh->batch_head ||
1897 list_first_entry(&sh->batch_list,
1898 struct stripe_head, batch_list) == head_sh;
1899 if (last_stripe) {
1900 flags = ASYNC_TX_ACK |
1901 (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
1902
1903 atomic_inc(&head_sh->count);
1904 init_async_submit(&submit, flags, tx, ops_complete_reconstruct, head_sh,
1905 to_addr_conv(sh, percpu, j));
1906 } else {
1907 flags = prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST;
1908 init_async_submit(&submit, flags, tx, NULL, NULL,
1909 to_addr_conv(sh, percpu, j));
1910 }
91c00924 1911
a08abd8c
DW		 1912 if (unlikely(count == 1))
1913 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
1914 else
1915 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
59fc630b 1916 if (!last_stripe) {
1917 j++;
1918 sh = list_first_entry(&sh->batch_list, struct stripe_head,
1919 batch_list);
1920 goto again;
1921 }
91c00924
DW		 1922}
1923
ac6b53b6
DW		 1924static void
1925ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu,
1926 struct dma_async_tx_descriptor *tx)
1927{
1928 struct async_submit_ctl submit;
59fc630b 1929 struct page **blocks;
1930 int count, i, j = 0;
1931 struct stripe_head *head_sh = sh;
1932 int last_stripe;
584acdd4
MS		 1933 int synflags;
1934 unsigned long txflags;
ac6b53b6
DW		 1935
1936 pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector);
1937
620125f2
SL		 1938 for (i = 0; i < sh->disks; i++) {
1939 if (sh->pd_idx == i || sh->qd_idx == i)
1940 continue;
1941 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1942 break;
1943 }
1944 if (i >= sh->disks) {
1945 atomic_inc(&sh->count);
620125f2
SL		 1946 set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
1947 set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
1948 ops_complete_reconstruct(sh);
1949 return;
1950 }
1951
59fc630b 1952again:
1953 blocks = to_addr_page(percpu, j);
584acdd4
MS		 1954
1955 if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
1956 synflags = SYNDROME_SRC_WRITTEN;
1957 txflags = ASYNC_TX_ACK | ASYNC_TX_PQ_XOR_DST;
1958 } else {
1959 synflags = SYNDROME_SRC_ALL;
1960 txflags = ASYNC_TX_ACK;
1961 }
1962
1963 count = set_syndrome_sources(blocks, sh, synflags);
59fc630b 1964 last_stripe = !head_sh->batch_head ||
1965 list_first_entry(&sh->batch_list,
1966 struct stripe_head, batch_list) == head_sh;
1967
1968 if (last_stripe) {
1969 atomic_inc(&head_sh->count);
584acdd4 1970 init_async_submit(&submit, txflags, tx, ops_complete_reconstruct,
59fc630b 1971 head_sh, to_addr_conv(sh, percpu, j));
1972 } else
1973 init_async_submit(&submit, 0, tx, NULL, NULL,
1974 to_addr_conv(sh, percpu, j));
48769695 1975 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
59fc630b 1976 if (!last_stripe) {
1977 j++;
1978 sh = list_first_entry(&sh->batch_list, struct stripe_head,
1979 batch_list);
1980 goto again;
1981 }
91c00924
DW		 1982}
1983
1984static void ops_complete_check(void *stripe_head_ref)
1985{
1986 struct stripe_head *sh = stripe_head_ref;
91c00924 1987
e46b272b 1988 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1989 (unsigned long long)sh->sector);
1990
ecc65c9b 1991 sh->check_state = check_state_check_result;
91c00924 1992 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 1993 raid5_release_stripe(sh);
91c00924
DW		 1994}
1995
ac6b53b6 1996static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 1997{
91c00924 1998 int disks = sh->disks;
ac6b53b6
DW		 1999 int pd_idx = sh->pd_idx;
2000 int qd_idx = sh->qd_idx;
2001 struct page *xor_dest;
46d5b785 2002 struct page **xor_srcs = to_addr_page(percpu, 0);
91c00924 2003 struct dma_async_tx_descriptor *tx;
a08abd8c 2004 struct async_submit_ctl submit;
ac6b53b6
DW		 2005 int count;
2006 int i;
91c00924 2007
e46b272b 2008 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 2009 (unsigned long long)sh->sector);
2010
59fc630b 2011 BUG_ON(sh->batch_head);
ac6b53b6
DW		 2012 count = 0;
2013 xor_dest = sh->dev[pd_idx].page;
2014 xor_srcs[count++] = xor_dest;
91c00924 2015 for (i = disks; i--; ) {
ac6b53b6
DW		 2016 if (i == pd_idx || i == qd_idx)
2017 continue;
2018 xor_srcs[count++] = sh->dev[i].page;
91c00924
DW		 2019 }
2020
d6f38f31 2021 init_async_submit(&submit, 0, NULL, NULL, NULL,
46d5b785 2022 to_addr_conv(sh, percpu, 0));
099f53cb 2023 tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
a08abd8c 2024 &sh->ops.zero_sum_result, &submit);
91c00924 2025
91c00924 2026 atomic_inc(&sh->count);
a08abd8c
DW		 2027 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL);
2028 tx = async_trigger_callback(&submit);
91c00924
DW		 2029}
2030
ac6b53b6
DW		 2031static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp)
2032{
46d5b785 2033 struct page **srcs = to_addr_page(percpu, 0);
ac6b53b6
DW		 2034 struct async_submit_ctl submit;
2035 int count;
2036
2037 pr_debug("%s: stripe %llu checkp: %d\n", __func__,
2038 (unsigned long long)sh->sector, checkp);
2039
59fc630b 2040 BUG_ON(sh->batch_head);
584acdd4 2041 count = set_syndrome_sources(srcs, sh, SYNDROME_SRC_ALL);
ac6b53b6
DW		 2042 if (!checkp)
2043 srcs[count] = NULL;
91c00924 2044
91c00924 2045 atomic_inc(&sh->count);
ac6b53b6 2046 init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check,
46d5b785 2047 sh, to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 2048 async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE,
2049 &sh->ops.zero_sum_result, percpu->spare_page, &submit);
91c00924
DW		 2050}
2051
51acbcec 2052static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
91c00924
DW		 2053{
2054 int overlap_clear = 0, i, disks = sh->disks;
2055 struct dma_async_tx_descriptor *tx = NULL;
d1688a6d 2056 struct r5conf *conf = sh->raid_conf;
ac6b53b6 2057 int level = conf->level;
d6f38f31
DW		 2058 struct raid5_percpu *percpu;
2059 unsigned long cpu;
91c00924 2060
d6f38f31
DW		 2061 cpu = get_cpu();
2062 percpu = per_cpu_ptr(conf->percpu, cpu);
83de75cc 2063 if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) {
91c00924
DW		 2064 ops_run_biofill(sh);
2065 overlap_clear++;
2066 }
2067
7b3a871e 2068 if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) {
ac6b53b6
DW		 2069 if (level < 6)
2070 tx = ops_run_compute5(sh, percpu);
2071 else {
2072 if (sh->ops.target2 < 0 || sh->ops.target < 0)
2073 tx = ops_run_compute6_1(sh, percpu);
2074 else
2075 tx = ops_run_compute6_2(sh, percpu);
2076 }
2077 /* terminate the chain if reconstruct is not set to be run */
2078 if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request))
7b3a871e
DW		 2079 async_tx_ack(tx);
2080 }
91c00924 2081
584acdd4
MS		 2082 if (test_bit(STRIPE_OP_PREXOR, &ops_request)) {
2083 if (level < 6)
2084 tx = ops_run_prexor5(sh, percpu, tx);
2085 else
2086 tx = ops_run_prexor6(sh, percpu, tx);
2087 }
91c00924 2088
ae1713e2
AP		 2089 if (test_bit(STRIPE_OP_PARTIAL_PARITY, &ops_request))
2090 tx = ops_run_partial_parity(sh, percpu, tx);
2091
600aa109 2092 if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) {
d8ee0728 2093 tx = ops_run_biodrain(sh, tx);
91c00924
DW		 2094 overlap_clear++;
2095 }
2096
ac6b53b6
DW		 2097 if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) {
2098 if (level < 6)
2099 ops_run_reconstruct5(sh, percpu, tx);
2100 else
2101 ops_run_reconstruct6(sh, percpu, tx);
2102 }
91c00924 2103
ac6b53b6
DW		 2104 if (test_bit(STRIPE_OP_CHECK, &ops_request)) {
2105 if (sh->check_state == check_state_run)
2106 ops_run_check_p(sh, percpu);
2107 else if (sh->check_state == check_state_run_q)
2108 ops_run_check_pq(sh, percpu, 0);
2109 else if (sh->check_state == check_state_run_pq)
2110 ops_run_check_pq(sh, percpu, 1);
2111 else
2112 BUG();
2113 }
91c00924 2114
59fc630b 2115 if (overlap_clear && !sh->batch_head)
91c00924
DW		 2116 for (i = disks; i--; ) {
2117 struct r5dev *dev = &sh->dev[i];
2118 if (test_and_clear_bit(R5_Overlap, &dev->flags))
2119 wake_up(&sh->raid_conf->wait_for_overlap);
2120 }
d6f38f31 2121 put_cpu();
91c00924
DW		 2122}
2123
845b9e22
AP		 2124static void free_stripe(struct kmem_cache *sc, struct stripe_head *sh)
2125{
2126 if (sh->ppl_page)
2127 __free_page(sh->ppl_page);
2128 kmem_cache_free(sc, sh);
2129}
2130
5f9d1fde 2131static struct stripe_head *alloc_stripe(struct kmem_cache *sc, gfp_t gfp,
845b9e22 2132 int disks, struct r5conf *conf)
f18c1a35
N		 2133{
2134 struct stripe_head *sh;
5f9d1fde 2135 int i;
f18c1a35
N		 2136
2137 sh = kmem_cache_zalloc(sc, gfp);
2138 if (sh) {
2139 spin_lock_init(&sh->stripe_lock);
2140 spin_lock_init(&sh->batch_lock);
2141 INIT_LIST_HEAD(&sh->batch_list);
2142 INIT_LIST_HEAD(&sh->lru);
a39f7afd 2143 INIT_LIST_HEAD(&sh->r5c);
d7bd398e 2144 INIT_LIST_HEAD(&sh->log_list);
f18c1a35 2145 atomic_set(&sh->count, 1);
845b9e22 2146 sh->raid_conf = conf;
a39f7afd 2147 sh->log_start = MaxSector;
5f9d1fde
SL		 2148 for (i = 0; i < disks; i++) {
2149 struct r5dev *dev = &sh->dev[i];
2150
3a83f467
ML		 2151 bio_init(&dev->req, &dev->vec, 1);
2152 bio_init(&dev->rreq, &dev->rvec, 1);
5f9d1fde 2153 }
845b9e22
AP		 2154
2155 if (raid5_has_ppl(conf)) {
2156 sh->ppl_page = alloc_page(gfp);
2157 if (!sh->ppl_page) {
2158 free_stripe(sc, sh);
2159 sh = NULL;
2160 }
2161 }
f18c1a35
N		 2162 }
2163 return sh;
2164}
486f0644 2165static int grow_one_stripe(struct r5conf *conf, gfp_t gfp)
1da177e4
LT		 2166{
2167 struct stripe_head *sh;
f18c1a35 2168
845b9e22 2169 sh = alloc_stripe(conf->slab_cache, gfp, conf->pool_size, conf);
3f294f4f
N		 2170 if (!sh)
2171 return 0;
6ce32846 2172
a9683a79 2173 if (grow_buffers(sh, gfp)) {
e4e11e38 2174 shrink_buffers(sh);
845b9e22 2175 free_stripe(conf->slab_cache, sh);
3f294f4f
N		 2176 return 0;
2177 }
486f0644
N		 2178 sh->hash_lock_index =
2179 conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS;
3f294f4f 2180 /* we just created an active stripe so... */
3f294f4f 2181 atomic_inc(&conf->active_stripes);
59fc630b 2182
6d036f7d 2183 raid5_release_stripe(sh);
486f0644 2184 conf->max_nr_stripes++;
3f294f4f
N		 2185 return 1;
2186}
2187
d1688a6d 2188static int grow_stripes(struct r5conf *conf, int num)
3f294f4f 2189{
e18b890b 2190 struct kmem_cache *sc;
53b8d89d 2191 size_t namelen = sizeof(conf->cache_name[0]);
5e5e3e78 2192 int devs = max(conf->raid_disks, conf->previous_raid_disks);
1da177e4 2193
f4be6b43 2194 if (conf->mddev->gendisk)
53b8d89d 2195 snprintf(conf->cache_name[0], namelen,
f4be6b43
N		 2196 "raid%d-%s", conf->level, mdname(conf->mddev));
2197 else
53b8d89d 2198 snprintf(conf->cache_name[0], namelen,
f4be6b43 2199 "raid%d-%p", conf->level, conf->mddev);
53b8d89d 2200 snprintf(conf->cache_name[1], namelen, "%.27s-alt", conf->cache_name[0]);
f4be6b43 2201
ad01c9e3
N		 2202 conf->active_name = 0;
2203 sc = kmem_cache_create(conf->cache_name[conf->active_name],
1da177e4 2204 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
20c2df83 2205 0, 0, NULL);
1da177e4
LT		 2206 if (!sc)
2207 return 1;
2208 conf->slab_cache = sc;
ad01c9e3 2209 conf->pool_size = devs;
486f0644
N		 2210 while (num--)
2211 if (!grow_one_stripe(conf, GFP_KERNEL))
1da177e4 2212 return 1;
486f0644 2213
1da177e4
LT		 2214 return 0;
2215}
29269553 2216
d6f38f31
DW		 2217/**
2218 * scribble_len - return the required size of the scribble region
2219 * @num - total number of disks in the array
2220 *
2221 * The size must be enough to contain:
2222 * 1/ a struct page pointer for each device in the array +2
2223 * 2/ room to convert each entry in (1) to its corresponding dma
2224 * (dma_map_page()) or page (page_address()) address.
2225 *
2226 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
2227 * calculate over all devices (not just the data blocks), using zeros in place
2228 * of the P and Q blocks.
2229 */
b330e6a4
KO		 2230static int scribble_alloc(struct raid5_percpu *percpu,
2231 int num, int cnt, gfp_t flags)
d6f38f31 2232{
b330e6a4
KO		 2233 size_t obj_size =
2234 sizeof(struct page *) * (num+2) +
2235 sizeof(addr_conv_t) * (num+2);
2236 void *scribble;
d6f38f31 2237
b330e6a4
KO		 2238 scribble = kvmalloc_array(cnt, obj_size, flags);
2239 if (!scribble)
2240 return -ENOMEM;
2241
2242 kvfree(percpu->scribble);
2243
2244 percpu->scribble = scribble;
2245 percpu->scribble_obj_size = obj_size;
2246 return 0;
d6f38f31
DW		 2247}
2248
738a2738
N		 2249static int resize_chunks(struct r5conf *conf, int new_disks, int new_sectors)
2250{
2251 unsigned long cpu;
2252 int err = 0;
2253
27a353c0
SL		 2254 /*
2255 * Never shrink. And mddev_suspend() could deadlock if this is called
2256 * from raid5d. In that case, scribble_disks and scribble_sectors
2257 * should equal to new_disks and new_sectors
2258 */
2259 if (conf->scribble_disks >= new_disks &&
2260 conf->scribble_sectors >= new_sectors)
2261 return 0;
738a2738
N		 2262 mddev_suspend(conf->mddev);
2263 get_online_cpus();
b330e6a4 2264
738a2738
N		 2265 for_each_present_cpu(cpu) {
2266 struct raid5_percpu *percpu;
738a2738
N		 2267
2268 percpu = per_cpu_ptr(conf->percpu, cpu);
b330e6a4
KO		 2269 err = scribble_alloc(percpu, new_disks,
2270 new_sectors / STRIPE_SECTORS,
2271 GFP_NOIO);
2272 if (err)
738a2738 2273 break;
738a2738 2274 }
b330e6a4 2275
738a2738
N		 2276 put_online_cpus();
2277 mddev_resume(conf->mddev);
27a353c0
SL		 2278 if (!err) {
2279 conf->scribble_disks = new_disks;
2280 conf->scribble_sectors = new_sectors;
2281 }
738a2738
N		 2282 return err;
2283}
2284
d1688a6d 2285static int resize_stripes(struct r5conf *conf, int newsize)
ad01c9e3
N		 2286{
2287 /* Make all the stripes able to hold 'newsize' devices.
2288 * New slots in each stripe get 'page' set to a new page.
2289 *
2290 * This happens in stages:
2291 * 1/ create a new kmem_cache and allocate the required number of
2292 * stripe_heads.
83f0d77a 2293 * 2/ gather all the old stripe_heads and transfer the pages across
ad01c9e3
N		 2294 * to the new stripe_heads. This will have the side effect of
2295 * freezing the array as once all stripe_heads have been collected,
2296 * no IO will be possible. Old stripe heads are freed once their
2297 * pages have been transferred over, and the old kmem_cache is
2298 * freed when all stripes are done.
2299 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
3560741e 2300 * we simple return a failure status - no need to clean anything up.
ad01c9e3
N		 2301 * 4/ allocate new pages for the new slots in the new stripe_heads.
2302 * If this fails, we don't bother trying the shrink the
2303 * stripe_heads down again, we just leave them as they are.
2304 * As each stripe_head is processed the new one is released into
2305 * active service.
2306 *
2307 * Once step2 is started, we cannot afford to wait for a write,
2308 * so we use GFP_NOIO allocations.
2309 */
2310 struct stripe_head *osh, *nsh;
2311 LIST_HEAD(newstripes);
2312 struct disk_info *ndisks;
2214c260 2313 int err = 0;
e18b890b 2314 struct kmem_cache *sc;
ad01c9e3 2315 int i;
566c09c5 2316 int hash, cnt;
ad01c9e3 2317
2214c260 2318 md_allow_write(conf->mddev);
2a2275d6 2319
ad01c9e3
N		 2320 /* Step 1 */
2321 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
2322 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
20c2df83 2323 0, 0, NULL);
ad01c9e3
N		 2324 if (!sc)
2325 return -ENOMEM;
2326
2d5b569b
N		 2327 /* Need to ensure auto-resizing doesn't interfere */
2328 mutex_lock(&conf->cache_size_mutex);
2329
ad01c9e3 2330 for (i = conf->max_nr_stripes; i; i--) {
845b9e22 2331 nsh = alloc_stripe(sc, GFP_KERNEL, newsize, conf);
ad01c9e3
N		 2332 if (!nsh)
2333 break;
2334
ad01c9e3
N		 2335 list_add(&nsh->lru, &newstripes);
2336 }
2337 if (i) {
2338 /* didn't get enough, give up */
2339 while (!list_empty(&newstripes)) {
2340 nsh = list_entry(newstripes.next, struct stripe_head, lru);
2341 list_del(&nsh->lru);
845b9e22 2342 free_stripe(sc, nsh);
ad01c9e3
N		 2343 }
2344 kmem_cache_destroy(sc);
2d5b569b 2345 mutex_unlock(&conf->cache_size_mutex);
ad01c9e3
N		 2346 return -ENOMEM;
2347 }
2348 /* Step 2 - Must use GFP_NOIO now.
2349 * OK, we have enough stripes, start collecting inactive
2350 * stripes and copying them over
2351 */
566c09c5
SL		 2352 hash = 0;
2353 cnt = 0;
ad01c9e3 2354 list_for_each_entry(nsh, &newstripes, lru) {
566c09c5 2355 lock_device_hash_lock(conf, hash);
6ab2a4b8 2356 wait_event_cmd(conf->wait_for_stripe,
566c09c5
SL		 2357 !list_empty(conf->inactive_list + hash),
2358 unlock_device_hash_lock(conf, hash),
2359 lock_device_hash_lock(conf, hash));
2360 osh = get_free_stripe(conf, hash);
2361 unlock_device_hash_lock(conf, hash);
f18c1a35 2362
d592a996 2363 for(i=0; i<conf->pool_size; i++) {
ad01c9e3 2364 nsh->dev[i].page = osh->dev[i].page;
d592a996
SL		 2365 nsh->dev[i].orig_page = osh->dev[i].page;
2366 }
566c09c5 2367 nsh->hash_lock_index = hash;
845b9e22 2368 free_stripe(conf->slab_cache, osh);
566c09c5
SL		 2369 cnt++;
2370 if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS +
2371 !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) {
2372 hash++;
2373 cnt = 0;
2374 }
ad01c9e3
N		 2375 }
2376 kmem_cache_destroy(conf->slab_cache);
2377
2378 /* Step 3.
2379 * At this point, we are holding all the stripes so the array
2380 * is completely stalled, so now is a good time to resize
d6f38f31 2381 * conf->disks and the scribble region
ad01c9e3 2382 */
6396bb22 2383 ndisks = kcalloc(newsize, sizeof(struct disk_info), GFP_NOIO);
ad01c9e3 2384 if (ndisks) {
d7bd398e 2385 for (i = 0; i < conf->pool_size; i++)
ad01c9e3 2386 ndisks[i] = conf->disks[i];
d7bd398e
SL		 2387
2388 for (i = conf->pool_size; i < newsize; i++) {
2389 ndisks[i].extra_page = alloc_page(GFP_NOIO);
2390 if (!ndisks[i].extra_page)
2391 err = -ENOMEM;
2392 }
2393
2394 if (err) {
2395 for (i = conf->pool_size; i < newsize; i++)
2396 if (ndisks[i].extra_page)
2397 put_page(ndisks[i].extra_page);
2398 kfree(ndisks);
2399 } else {
2400 kfree(conf->disks);
2401 conf->disks = ndisks;
2402 }
ad01c9e3
N		 2403 } else
2404 err = -ENOMEM;
2405
2d5b569b 2406 mutex_unlock(&conf->cache_size_mutex);
583da48e
DY		 2407
2408 conf->slab_cache = sc;
2409 conf->active_name = 1-conf->active_name;
2410
ad01c9e3
N		 2411 /* Step 4, return new stripes to service */
2412 while(!list_empty(&newstripes)) {
2413 nsh = list_entry(newstripes.next, struct stripe_head, lru);
2414 list_del_init(&nsh->lru);
d6f38f31 2415
ad01c9e3
N		 2416 for (i=conf->raid_disks; i < newsize; i++)
2417 if (nsh->dev[i].page == NULL) {
2418 struct page *p = alloc_page(GFP_NOIO);
2419 nsh->dev[i].page = p;
d592a996 2420 nsh->dev[i].orig_page = p;
ad01c9e3
N		 2421 if (!p)
2422 err = -ENOMEM;
2423 }
6d036f7d 2424 raid5_release_stripe(nsh);
ad01c9e3
N		 2425 }
2426 /* critical section pass, GFP_NOIO no longer needed */
2427
6e9eac2d
N		 2428 if (!err)
2429 conf->pool_size = newsize;
ad01c9e3
N		 2430 return err;
2431}
1da177e4 2432
486f0644 2433static int drop_one_stripe(struct r5conf *conf)
1da177e4
LT		 2434{
2435 struct stripe_head *sh;
49895bcc 2436 int hash = (conf->max_nr_stripes - 1) & STRIPE_HASH_LOCKS_MASK;
1da177e4 2437
566c09c5
SL		 2438 spin_lock_irq(conf->hash_locks + hash);
2439 sh = get_free_stripe(conf, hash);
2440 spin_unlock_irq(conf->hash_locks + hash);
3f294f4f
N		 2441 if (!sh)
2442 return 0;
78bafebd 2443 BUG_ON(atomic_read(&sh->count));
e4e11e38 2444 shrink_buffers(sh);
845b9e22 2445 free_stripe(conf->slab_cache, sh);
3f294f4f 2446 atomic_dec(&conf->active_stripes);
486f0644 2447 conf->max_nr_stripes--;
3f294f4f
N		 2448 return 1;
2449}
2450
d1688a6d 2451static void shrink_stripes(struct r5conf *conf)
3f294f4f 2452{
486f0644
N		 2453 while (conf->max_nr_stripes &&
2454 drop_one_stripe(conf))
2455 ;
3f294f4f 2456
644df1a8 2457 kmem_cache_destroy(conf->slab_cache);
1da177e4
LT		 2458 conf->slab_cache = NULL;
2459}
2460
4246a0b6 2461static void raid5_end_read_request(struct bio * bi)
1da177e4 2462{
99c0fb5f 2463 struct stripe_head *sh = bi->bi_private;
d1688a6d 2464 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 2465 int disks = sh->disks, i;
d6950432 2466 char b[BDEVNAME_SIZE];
dd054fce 2467 struct md_rdev *rdev = NULL;
05616be5 2468 sector_t s;
1da177e4
LT		 2469
2470 for (i=0 ; i<disks; i++)
2471 if (bi == &sh->dev[i].req)
2472 break;
2473
4246a0b6 2474 pr_debug("end_read_request %llu/%d, count: %d, error %d.\n",
45b4233c 2475 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
4e4cbee9 2476 bi->bi_status);
1da177e4 2477 if (i == disks) {
5f9d1fde 2478 bio_reset(bi);
1da177e4 2479 BUG();
6712ecf8 2480 return;
1da177e4 2481 }
14a75d3e 2482 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
dd054fce
N		 2483 /* If replacement finished while this request was outstanding,
2484 * 'replacement' might be NULL already.
2485 * In that case it moved down to 'rdev'.
2486 * rdev is not removed until all requests are finished.
2487 */
14a75d3e 2488 rdev = conf->disks[i].replacement;
dd054fce 2489 if (!rdev)
14a75d3e 2490 rdev = conf->disks[i].rdev;
1da177e4 2491
05616be5
N		 2492 if (use_new_offset(conf, sh))
2493 s = sh->sector + rdev->new_data_offset;
2494 else
2495 s = sh->sector + rdev->data_offset;
4e4cbee9 2496 if (!bi->bi_status) {
1da177e4 2497 set_bit(R5_UPTODATE, &sh->dev[i].flags);
4e5314b5 2498 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
14a75d3e
N		 2499 /* Note that this cannot happen on a
2500 * replacement device. We just fail those on
2501 * any error
2502 */
cc6167b4
N		 2503 pr_info_ratelimited(
2504 "md/raid:%s: read error corrected (%lu sectors at %llu on %s)\n",
8bda470e 2505 mdname(conf->mddev), STRIPE_SECTORS,
05616be5 2506 (unsigned long long)s,
8bda470e 2507 bdevname(rdev->bdev, b));
ddd5115f 2508 atomic_add(STRIPE_SECTORS, &rdev->corrected_errors);
4e5314b5
N		 2509 clear_bit(R5_ReadError, &sh->dev[i].flags);
2510 clear_bit(R5_ReWrite, &sh->dev[i].flags);
3f9e7c14 2511 } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2512 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2513
86aa1397
SL		 2514 if (test_bit(R5_InJournal, &sh->dev[i].flags))
2515 /*
2516 * end read for a page in journal, this
2517 * must be preparing for prexor in rmw
2518 */
2519 set_bit(R5_OrigPageUPTDODATE, &sh->dev[i].flags);
2520
14a75d3e
N		 2521 if (atomic_read(&rdev->read_errors))
2522 atomic_set(&rdev->read_errors, 0);
1da177e4 2523 } else {
14a75d3e 2524 const char *bdn = bdevname(rdev->bdev, b);
ba22dcbf 2525 int retry = 0;
2e8ac303 2526 int set_bad = 0;
d6950432 2527
1da177e4 2528 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
b76b4715
NC		 2529 if (!(bi->bi_status == BLK_STS_PROTECTION))
2530 atomic_inc(&rdev->read_errors);
14a75d3e 2531 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
cc6167b4
N		 2532 pr_warn_ratelimited(
2533 "md/raid:%s: read error on replacement device (sector %llu on %s).\n",
14a75d3e 2534 mdname(conf->mddev),
05616be5 2535 (unsigned long long)s,
14a75d3e 2536 bdn);
2e8ac303 2537 else if (conf->mddev->degraded >= conf->max_degraded) {
2538 set_bad = 1;
cc6167b4
N		 2539 pr_warn_ratelimited(
2540 "md/raid:%s: read error not correctable (sector %llu on %s).\n",
8bda470e 2541 mdname(conf->mddev),
05616be5 2542 (unsigned long long)s,
8bda470e 2543 bdn);
2e8ac303 2544 } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) {
4e5314b5 2545 /* Oh, no!!! */
2e8ac303 2546 set_bad = 1;
cc6167b4
N		 2547 pr_warn_ratelimited(
2548 "md/raid:%s: read error NOT corrected!! (sector %llu on %s).\n",
8bda470e 2549 mdname(conf->mddev),
05616be5 2550 (unsigned long long)s,
8bda470e 2551 bdn);
2e8ac303 2552 } else if (atomic_read(&rdev->read_errors)
0009fad0
NC		 2553 > conf->max_nr_stripes) {
2554 if (!test_bit(Faulty, &rdev->flags)) {
2555 pr_warn("md/raid:%s: %d read_errors > %d stripes\n",
2556 mdname(conf->mddev),
2557 atomic_read(&rdev->read_errors),
2558 conf->max_nr_stripes);
2559 pr_warn("md/raid:%s: Too many read errors, failing device %s.\n",
2560 mdname(conf->mddev), bdn);
2561 }
2562 } else
ba22dcbf 2563 retry = 1;
edfa1f65
BY		 2564 if (set_bad && test_bit(In_sync, &rdev->flags)
2565 && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2566 retry = 1;
ba22dcbf 2567 if (retry)
143f6e73
XN		 2568 if (sh->qd_idx >= 0 && sh->pd_idx == i)
2569 set_bit(R5_ReadError, &sh->dev[i].flags);
2570 else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) {
3f9e7c14 2571 set_bit(R5_ReadError, &sh->dev[i].flags);
2572 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2573 } else
2574 set_bit(R5_ReadNoMerge, &sh->dev[i].flags);
ba22dcbf 2575 else {
4e5314b5
N		 2576 clear_bit(R5_ReadError, &sh->dev[i].flags);
2577 clear_bit(R5_ReWrite, &sh->dev[i].flags);
2e8ac303 2578 if (!(set_bad
2579 && test_bit(In_sync, &rdev->flags)
2580 && rdev_set_badblocks(
2581 rdev, sh->sector, STRIPE_SECTORS, 0)))
2582 md_error(conf->mddev, rdev);
ba22dcbf 2583 }
1da177e4 2584 }
14a75d3e 2585 rdev_dec_pending(rdev, conf->mddev);
c9445555 2586 bio_reset(bi);
1da177e4
LT		 2587 clear_bit(R5_LOCKED, &sh->dev[i].flags);
2588 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 2589 raid5_release_stripe(sh);
1da177e4
LT		 2590}
2591
4246a0b6 2592static void raid5_end_write_request(struct bio *bi)
1da177e4 2593{
99c0fb5f 2594 struct stripe_head *sh = bi->bi_private;
d1688a6d 2595 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 2596 int disks = sh->disks, i;
977df362 2597 struct md_rdev *uninitialized_var(rdev);
b84db560
N		 2598 sector_t first_bad;
2599 int bad_sectors;
977df362 2600 int replacement = 0;
1da177e4 2601
977df362
N		 2602 for (i = 0 ; i < disks; i++) {
2603 if (bi == &sh->dev[i].req) {
2604 rdev = conf->disks[i].rdev;
1da177e4 2605 break;
977df362
N		 2606 }
2607 if (bi == &sh->dev[i].rreq) {
2608 rdev = conf->disks[i].replacement;
dd054fce
N		 2609 if (rdev)
2610 replacement = 1;
2611 else
2612 /* rdev was removed and 'replacement'
2613 * replaced it. rdev is not removed
2614 * until all requests are finished.
2615 */
2616 rdev = conf->disks[i].rdev;
977df362
N		 2617 break;
2618 }
2619 }
4246a0b6 2620 pr_debug("end_write_request %llu/%d, count %d, error: %d.\n",
1da177e4 2621 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
4e4cbee9 2622 bi->bi_status);
1da177e4 2623 if (i == disks) {
5f9d1fde 2624 bio_reset(bi);
1da177e4 2625 BUG();
6712ecf8 2626 return;
1da177e4
LT		 2627 }
2628
977df362 2629 if (replacement) {
4e4cbee9 2630 if (bi->bi_status)
977df362
N		 2631 md_error(conf->mddev, rdev);
2632 else if (is_badblock(rdev, sh->sector,
2633 STRIPE_SECTORS,
2634 &first_bad, &bad_sectors))
2635 set_bit(R5_MadeGoodRepl, &sh->dev[i].flags);
2636 } else {
4e4cbee9 2637 if (bi->bi_status) {
9f97e4b1 2638 set_bit(STRIPE_DEGRADED, &sh->state);
977df362
N		 2639 set_bit(WriteErrorSeen, &rdev->flags);
2640 set_bit(R5_WriteError, &sh->dev[i].flags);
3a6de292
N		 2641 if (!test_and_set_bit(WantReplacement, &rdev->flags))
2642 set_bit(MD_RECOVERY_NEEDED,
2643 &rdev->mddev->recovery);
977df362
N		 2644 } else if (is_badblock(rdev, sh->sector,
2645 STRIPE_SECTORS,
c0b32972 2646 &first_bad, &bad_sectors)) {
977df362 2647 set_bit(R5_MadeGood, &sh->dev[i].flags);
c0b32972
N		 2648 if (test_bit(R5_ReadError, &sh->dev[i].flags))
2649 /* That was a successful write so make
2650 * sure it looks like we already did
2651 * a re-write.
2652 */
2653 set_bit(R5_ReWrite, &sh->dev[i].flags);
2654 }
977df362
N		 2655 }
2656 rdev_dec_pending(rdev, conf->mddev);
1da177e4 2657
4e4cbee9 2658 if (sh->batch_head && bi->bi_status && !replacement)
72ac7330 2659 set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state);
2660
c9445555 2661 bio_reset(bi);
977df362
N		 2662 if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags))
2663 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1da177e4 2664 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 2665 raid5_release_stripe(sh);
59fc630b 2666
2667 if (sh->batch_head && sh != sh->batch_head)
6d036f7d 2668 raid5_release_stripe(sh->batch_head);
1da177e4
LT		 2669}
2670
849674e4 2671static void raid5_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 2672{
2673 char b[BDEVNAME_SIZE];
d1688a6d 2674 struct r5conf *conf = mddev->private;
908f4fbd 2675 unsigned long flags;
0c55e022 2676 pr_debug("raid456: error called\n");
1da177e4 2677
908f4fbd 2678 spin_lock_irqsave(&conf->device_lock, flags);
fb73b357
MT		 2679
2680 if (test_bit(In_sync, &rdev->flags) &&
2681 mddev->degraded == conf->max_degraded) {
2682 /*
2683 * Don't allow to achieve failed state
2684 * Don't try to recover this device
2685 */
2686 conf->recovery_disabled = mddev->recovery_disabled;
2687 spin_unlock_irqrestore(&conf->device_lock, flags);
2688 return;
2689 }
2690
aff69d89 2691 set_bit(Faulty, &rdev->flags);
908f4fbd 2692 clear_bit(In_sync, &rdev->flags);
2e38a37f 2693 mddev->degraded = raid5_calc_degraded(conf);
908f4fbd
N		 2694 spin_unlock_irqrestore(&conf->device_lock, flags);
2695 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2696
de393cde 2697 set_bit(Blocked, &rdev->flags);
2953079c
SL		 2698 set_mask_bits(&mddev->sb_flags, 0,
2699 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING));
cc6167b4
N		 2700 pr_crit("md/raid:%s: Disk failure on %s, disabling device.\n"
2701 "md/raid:%s: Operation continuing on %d devices.\n",
2702 mdname(mddev),
2703 bdevname(rdev->bdev, b),
2704 mdname(mddev),
2705 conf->raid_disks - mddev->degraded);
70d466f7 2706 r5c_update_on_rdev_error(mddev, rdev);
16a53ecc 2707}
1da177e4
LT		 2708
2709/*
2710 * Input: a 'big' sector number,
2711 * Output: index of the data and parity disk, and the sector # in them.
2712 */
6d036f7d
SL		 2713sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector,
2714 int previous, int *dd_idx,
2715 struct stripe_head *sh)
1da177e4 2716{
6e3b96ed 2717 sector_t stripe, stripe2;
35f2a591 2718 sector_t chunk_number;
1da177e4 2719 unsigned int chunk_offset;
911d4ee8 2720 int pd_idx, qd_idx;
67cc2b81 2721 int ddf_layout = 0;
1da177e4 2722 sector_t new_sector;
e183eaed
N		 2723 int algorithm = previous ? conf->prev_algo
2724 : conf->algorithm;
09c9e5fa
AN		 2725 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2726 : conf->chunk_sectors;
112bf897
N		 2727 int raid_disks = previous ? conf->previous_raid_disks
2728 : conf->raid_disks;
2729 int data_disks = raid_disks - conf->max_degraded;
1da177e4
LT		 2730
2731 /* First compute the information on this sector */
2732
2733 /*
2734 * Compute the chunk number and the sector offset inside the chunk
2735 */
2736 chunk_offset = sector_div(r_sector, sectors_per_chunk);
2737 chunk_number = r_sector;
1da177e4
LT		 2738
2739 /*
2740 * Compute the stripe number
2741 */
35f2a591
N		 2742 stripe = chunk_number;
2743 *dd_idx = sector_div(stripe, data_disks);
6e3b96ed 2744 stripe2 = stripe;
1da177e4
LT		 2745 /*
2746 * Select the parity disk based on the user selected algorithm.
2747 */
84789554 2748 pd_idx = qd_idx = -1;
16a53ecc
N		 2749 switch(conf->level) {
2750 case 4:
911d4ee8 2751 pd_idx = data_disks;
16a53ecc
N		 2752 break;
2753 case 5:
e183eaed 2754 switch (algorithm) {
1da177e4 2755 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2756 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2757 if (*dd_idx >= pd_idx)
1da177e4
LT		 2758 (*dd_idx)++;
2759 break;
2760 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2761 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2762 if (*dd_idx >= pd_idx)
1da177e4
LT		 2763 (*dd_idx)++;
2764 break;
2765 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2766 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2767 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4
LT		 2768 break;
2769 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2770 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2771 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4 2772 break;
99c0fb5f
N		 2773 case ALGORITHM_PARITY_0:
2774 pd_idx = 0;
2775 (*dd_idx)++;
2776 break;
2777 case ALGORITHM_PARITY_N:
2778 pd_idx = data_disks;
2779 break;
1da177e4 2780 default:
99c0fb5f 2781 BUG();
16a53ecc
N		 2782 }
2783 break;
2784 case 6:
2785
e183eaed 2786 switch (algorithm) {
16a53ecc 2787 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2788 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2789 qd_idx = pd_idx + 1;
2790 if (pd_idx == raid_disks-1) {
99c0fb5f 2791 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2792 qd_idx = 0;
2793 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2794 (*dd_idx) += 2; /* D D P Q D */
2795 break;
2796 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2797 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2798 qd_idx = pd_idx + 1;
2799 if (pd_idx == raid_disks-1) {
99c0fb5f 2800 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2801 qd_idx = 0;
2802 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2803 (*dd_idx) += 2; /* D D P Q D */
2804 break;
2805 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2806 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2807 qd_idx = (pd_idx + 1) % raid_disks;
2808 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc
N		 2809 break;
2810 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2811 pd_idx = 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 2814 break;
99c0fb5f
N		 2815
2816 case ALGORITHM_PARITY_0:
2817 pd_idx = 0;
2818 qd_idx = 1;
2819 (*dd_idx) += 2;
2820 break;
2821 case ALGORITHM_PARITY_N:
2822 pd_idx = data_disks;
2823 qd_idx = data_disks + 1;
2824 break;
2825
2826 case ALGORITHM_ROTATING_ZERO_RESTART:
2827 /* Exactly the same as RIGHT_ASYMMETRIC, but or
2828 * of blocks for computing Q is different.
2829 */
6e3b96ed 2830 pd_idx = sector_div(stripe2, raid_disks);
99c0fb5f
N		 2831 qd_idx = pd_idx + 1;
2832 if (pd_idx == raid_disks-1) {
2833 (*dd_idx)++; /* Q D D D P */
2834 qd_idx = 0;
2835 } else if (*dd_idx >= pd_idx)
2836 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2837 ddf_layout = 1;
99c0fb5f
N		 2838 break;
2839
2840 case ALGORITHM_ROTATING_N_RESTART:
2841 /* Same a left_asymmetric, by first stripe is
2842 * D D D P Q rather than
2843 * Q D D D P
2844 */
6e3b96ed
N		 2845 stripe2 += 1;
2846 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2847 qd_idx = pd_idx + 1;
2848 if (pd_idx == raid_disks-1) {
2849 (*dd_idx)++; /* Q D D D P */
2850 qd_idx = 0;
2851 } else if (*dd_idx >= pd_idx)
2852 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2853 ddf_layout = 1;
99c0fb5f
N		 2854 break;
2855
2856 case ALGORITHM_ROTATING_N_CONTINUE:
2857 /* Same as left_symmetric but Q is before P */
6e3b96ed 2858 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2859 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
2860 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
67cc2b81 2861 ddf_layout = 1;
99c0fb5f
N		 2862 break;
2863
2864 case ALGORITHM_LEFT_ASYMMETRIC_6:
2865 /* RAID5 left_asymmetric, with Q on last device */
6e3b96ed 2866 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2867 if (*dd_idx >= pd_idx)
2868 (*dd_idx)++;
2869 qd_idx = raid_disks - 1;
2870 break;
2871
2872 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6e3b96ed 2873 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2874 if (*dd_idx >= pd_idx)
2875 (*dd_idx)++;
2876 qd_idx = raid_disks - 1;
2877 break;
2878
2879 case ALGORITHM_LEFT_SYMMETRIC_6:
6e3b96ed 2880 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2881 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2882 qd_idx = raid_disks - 1;
2883 break;
2884
2885 case ALGORITHM_RIGHT_SYMMETRIC_6:
6e3b96ed 2886 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2887 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2888 qd_idx = raid_disks - 1;
2889 break;
2890
2891 case ALGORITHM_PARITY_0_6:
2892 pd_idx = 0;
2893 (*dd_idx)++;
2894 qd_idx = raid_disks - 1;
2895 break;
2896
16a53ecc 2897 default:
99c0fb5f 2898 BUG();
16a53ecc
N		 2899 }
2900 break;
1da177e4
LT		 2901 }
2902
911d4ee8
N		 2903 if (sh) {
2904 sh->pd_idx = pd_idx;
2905 sh->qd_idx = qd_idx;
67cc2b81 2906 sh->ddf_layout = ddf_layout;
911d4ee8 2907 }
1da177e4
LT		 2908 /*
2909 * Finally, compute the new sector number
2910 */
2911 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
2912 return new_sector;
2913}
2914
6d036f7d 2915sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous)
1da177e4 2916{
d1688a6d 2917 struct r5conf *conf = sh->raid_conf;
b875e531
N		 2918 int raid_disks = sh->disks;
2919 int data_disks = raid_disks - conf->max_degraded;
1da177e4 2920 sector_t new_sector = sh->sector, check;
09c9e5fa
AN		 2921 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2922 : conf->chunk_sectors;
e183eaed
N		 2923 int algorithm = previous ? conf->prev_algo
2924 : conf->algorithm;
1da177e4
LT		 2925 sector_t stripe;
2926 int chunk_offset;
35f2a591
N		 2927 sector_t chunk_number;
2928 int dummy1, dd_idx = i;
1da177e4 2929 sector_t r_sector;
911d4ee8 2930 struct stripe_head sh2;
1da177e4
LT		 2931
2932 chunk_offset = sector_div(new_sector, sectors_per_chunk);
2933 stripe = new_sector;
1da177e4 2934
16a53ecc
N		 2935 if (i == sh->pd_idx)
2936 return 0;
2937 switch(conf->level) {
2938 case 4: break;
2939 case 5:
e183eaed 2940 switch (algorithm) {
1da177e4
LT		 2941 case ALGORITHM_LEFT_ASYMMETRIC:
2942 case ALGORITHM_RIGHT_ASYMMETRIC:
2943 if (i > sh->pd_idx)
2944 i--;
2945 break;
2946 case ALGORITHM_LEFT_SYMMETRIC:
2947 case ALGORITHM_RIGHT_SYMMETRIC:
2948 if (i < sh->pd_idx)
2949 i += raid_disks;
2950 i -= (sh->pd_idx + 1);
2951 break;
99c0fb5f
N		 2952 case ALGORITHM_PARITY_0:
2953 i -= 1;
2954 break;
2955 case ALGORITHM_PARITY_N:
2956 break;
1da177e4 2957 default:
99c0fb5f 2958 BUG();
16a53ecc
N		 2959 }
2960 break;
2961 case 6:
d0dabf7e 2962 if (i == sh->qd_idx)
16a53ecc 2963 return 0; /* It is the Q disk */
e183eaed 2964 switch (algorithm) {
16a53ecc
N		 2965 case ALGORITHM_LEFT_ASYMMETRIC:
2966 case ALGORITHM_RIGHT_ASYMMETRIC:
99c0fb5f
N		 2967 case ALGORITHM_ROTATING_ZERO_RESTART:
2968 case ALGORITHM_ROTATING_N_RESTART:
2969 if (sh->pd_idx == raid_disks-1)
2970 i--; /* Q D D D P */
16a53ecc
N		 2971 else if (i > sh->pd_idx)
2972 i -= 2; /* D D P Q D */
2973 break;
2974 case ALGORITHM_LEFT_SYMMETRIC:
2975 case ALGORITHM_RIGHT_SYMMETRIC:
2976 if (sh->pd_idx == raid_disks-1)
2977 i--; /* Q D D D P */
2978 else {
2979 /* D D P Q D */
2980 if (i < sh->pd_idx)
2981 i += raid_disks;
2982 i -= (sh->pd_idx + 2);
2983 }
2984 break;
99c0fb5f
N		 2985 case ALGORITHM_PARITY_0:
2986 i -= 2;
2987 break;
2988 case ALGORITHM_PARITY_N:
2989 break;
2990 case ALGORITHM_ROTATING_N_CONTINUE:
e4424fee 2991 /* Like left_symmetric, but P is before Q */
99c0fb5f
N		 2992 if (sh->pd_idx == 0)
2993 i--; /* P D D D Q */
e4424fee
N		 2994 else {
2995 /* D D Q P D */
2996 if (i < sh->pd_idx)
2997 i += raid_disks;
2998 i -= (sh->pd_idx + 1);
2999 }
99c0fb5f
N		 3000 break;
3001 case ALGORITHM_LEFT_ASYMMETRIC_6:
3002 case ALGORITHM_RIGHT_ASYMMETRIC_6:
3003 if (i > sh->pd_idx)
3004 i--;
3005 break;
3006 case ALGORITHM_LEFT_SYMMETRIC_6:
3007 case ALGORITHM_RIGHT_SYMMETRIC_6:
3008 if (i < sh->pd_idx)
3009 i += data_disks + 1;
3010 i -= (sh->pd_idx + 1);
3011 break;
3012 case ALGORITHM_PARITY_0_6:
3013 i -= 1;
3014 break;
16a53ecc 3015 default:
99c0fb5f 3016 BUG();
16a53ecc
N		 3017 }
3018 break;
1da177e4
LT		 3019 }
3020
3021 chunk_number = stripe * data_disks + i;
35f2a591 3022 r_sector = chunk_number * sectors_per_chunk + chunk_offset;
1da177e4 3023
112bf897 3024 check = raid5_compute_sector(conf, r_sector,
784052ec 3025 previous, &dummy1, &sh2);
911d4ee8
N		 3026 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
3027 || sh2.qd_idx != sh->qd_idx) {
cc6167b4
N		 3028 pr_warn("md/raid:%s: compute_blocknr: map not correct\n",
3029 mdname(conf->mddev));
1da177e4
LT		 3030 return 0;
3031 }
3032 return r_sector;
3033}
3034
07e83364
SL		 3035/*
3036 * There are cases where we want handle_stripe_dirtying() and
3037 * schedule_reconstruction() to delay towrite to some dev of a stripe.
3038 *
3039 * This function checks whether we want to delay the towrite. Specifically,
3040 * we delay the towrite when:
3041 *
3042 * 1. degraded stripe has a non-overwrite to the missing dev, AND this
3043 * stripe has data in journal (for other devices).
3044 *
3045 * In this case, when reading data for the non-overwrite dev, it is
3046 * necessary to handle complex rmw of write back cache (prexor with
3047 * orig_page, and xor with page). To keep read path simple, we would
3048 * like to flush data in journal to RAID disks first, so complex rmw
3049 * is handled in the write patch (handle_stripe_dirtying).
3050 *
39b99586
SL		 3051 * 2. when journal space is critical (R5C_LOG_CRITICAL=1)
3052 *
3053 * It is important to be able to flush all stripes in raid5-cache.
3054 * Therefore, we need reserve some space on the journal device for
3055 * these flushes. If flush operation includes pending writes to the
3056 * stripe, we need to reserve (conf->raid_disk + 1) pages per stripe
3057 * for the flush out. If we exclude these pending writes from flush
3058 * operation, we only need (conf->max_degraded + 1) pages per stripe.
3059 * Therefore, excluding pending writes in these cases enables more
3060 * efficient use of the journal device.
3061 *
3062 * Note: To make sure the stripe makes progress, we only delay
3063 * towrite for stripes with data already in journal (injournal > 0).
3064 * When LOG_CRITICAL, stripes with injournal == 0 will be sent to
3065 * no_space_stripes list.
3066 *
70d466f7
SL		 3067 * 3. during journal failure
3068 * In journal failure, we try to flush all cached data to raid disks
3069 * based on data in stripe cache. The array is read-only to upper
3070 * layers, so we would skip all pending writes.
3071 *
07e83364 3072 */
39b99586
SL		 3073static inline bool delay_towrite(struct r5conf *conf,
3074 struct r5dev *dev,
3075 struct stripe_head_state *s)
07e83364 3076{
39b99586
SL		 3077 /* case 1 above */
3078 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
3079 !test_bit(R5_Insync, &dev->flags) && s->injournal)
3080 return true;
3081 /* case 2 above */
3082 if (test_bit(R5C_LOG_CRITICAL, &conf->cache_state) &&
3083 s->injournal > 0)
3084 return true;
70d466f7
SL		 3085 /* case 3 above */
3086 if (s->log_failed && s->injournal)
3087 return true;
39b99586 3088 return false;
07e83364
SL		 3089}
3090
600aa109 3091static void
c0f7bddb 3092schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
600aa109 3093 int rcw, int expand)
e33129d8 3094{
584acdd4 3095 int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx, disks = sh->disks;
d1688a6d 3096 struct r5conf *conf = sh->raid_conf;
c0f7bddb 3097 int level = conf->level;
e33129d8
DW		 3098
3099 if (rcw) {
1e6d690b
SL		 3100 /*
3101 * In some cases, handle_stripe_dirtying initially decided to
3102 * run rmw and allocates extra page for prexor. However, rcw is
3103 * cheaper later on. We need to free the extra page now,
3104 * because we won't be able to do that in ops_complete_prexor().
3105 */
3106 r5c_release_extra_page(sh);
e33129d8
DW		 3107
3108 for (i = disks; i--; ) {
3109 struct r5dev *dev = &sh->dev[i];
3110
39b99586 3111 if (dev->towrite && !delay_towrite(conf, dev, s)) {
e33129d8 3112 set_bit(R5_LOCKED, &dev->flags);
d8ee0728 3113 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 3114 if (!expand)
3115 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 3116 s->locked++;
1e6d690b
SL		 3117 } else if (test_bit(R5_InJournal, &dev->flags)) {
3118 set_bit(R5_LOCKED, &dev->flags);
3119 s->locked++;
e33129d8
DW		 3120 }
3121 }
ce7d363a
N		 3122 /* if we are not expanding this is a proper write request, and
3123 * there will be bios with new data to be drained into the
3124 * stripe cache
3125 */
3126 if (!expand) {
3127 if (!s->locked)
3128 /* False alarm, nothing to do */
3129 return;
3130 sh->reconstruct_state = reconstruct_state_drain_run;
3131 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
3132 } else
3133 sh->reconstruct_state = reconstruct_state_run;
3134
3135 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
3136
c0f7bddb 3137 if (s->locked + conf->max_degraded == disks)
8b3e6cdc 3138 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
c0f7bddb 3139 atomic_inc(&conf->pending_full_writes);
e33129d8
DW		 3140 } else {
3141 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
3142 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
584acdd4
MS		 3143 BUG_ON(level == 6 &&
3144 (!(test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags) ||
3145 test_bit(R5_Wantcompute, &sh->dev[qd_idx].flags))));
e33129d8 3146
e33129d8
DW		 3147 for (i = disks; i--; ) {
3148 struct r5dev *dev = &sh->dev[i];
584acdd4 3149 if (i == pd_idx || i == qd_idx)
e33129d8
DW		 3150 continue;
3151
e33129d8
DW		 3152 if (dev->towrite &&
3153 (test_bit(R5_UPTODATE, &dev->flags) ||
d8ee0728
DW		 3154 test_bit(R5_Wantcompute, &dev->flags))) {
3155 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 3156 set_bit(R5_LOCKED, &dev->flags);
3157 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 3158 s->locked++;
1e6d690b
SL		 3159 } else if (test_bit(R5_InJournal, &dev->flags)) {
3160 set_bit(R5_LOCKED, &dev->flags);
3161 s->locked++;
e33129d8
DW		 3162 }
3163 }
ce7d363a
N		 3164 if (!s->locked)
3165 /* False alarm - nothing to do */
3166 return;
3167 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
3168 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
3169 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
3170 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
e33129d8
DW		 3171 }
3172
c0f7bddb 3173 /* keep the parity disk(s) locked while asynchronous operations
e33129d8
DW		 3174 * are in flight
3175 */
3176 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
3177 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
600aa109 3178 s->locked++;
e33129d8 3179
c0f7bddb
YT		 3180 if (level == 6) {
3181 int qd_idx = sh->qd_idx;
3182 struct r5dev *dev = &sh->dev[qd_idx];
3183
3184 set_bit(R5_LOCKED, &dev->flags);
3185 clear_bit(R5_UPTODATE, &dev->flags);
3186 s->locked++;
3187 }
3188
845b9e22 3189 if (raid5_has_ppl(sh->raid_conf) && sh->ppl_page &&
3418d036
AP		 3190 test_bit(STRIPE_OP_BIODRAIN, &s->ops_request) &&
3191 !test_bit(STRIPE_FULL_WRITE, &sh->state) &&
3192 test_bit(R5_Insync, &sh->dev[pd_idx].flags))
3193 set_bit(STRIPE_OP_PARTIAL_PARITY, &s->ops_request);
3194
600aa109 3195 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
e46b272b 3196 __func__, (unsigned long long)sh->sector,
600aa109 3197 s->locked, s->ops_request);
e33129d8 3198}
16a53ecc 3199
1da177e4
LT		 3200/*
3201 * Each stripe/dev can have one or more bion attached.
16a53ecc 3202 * toread/towrite point to the first in a chain.
1da177e4
LT		 3203 * The bi_next chain must be in order.
3204 */
da41ba65 3205static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx,
3206 int forwrite, int previous)
1da177e4
LT		 3207{
3208 struct bio **bip;
d1688a6d 3209 struct r5conf *conf = sh->raid_conf;
72626685 3210 int firstwrite=0;
1da177e4 3211
cbe47ec5 3212 pr_debug("adding bi b#%llu to stripe s#%llu\n",
4f024f37 3213 (unsigned long long)bi->bi_iter.bi_sector,
1da177e4
LT		 3214 (unsigned long long)sh->sector);
3215
b17459c0 3216 spin_lock_irq(&sh->stripe_lock);
2cd259a7 3217 sh->dev[dd_idx].write_hint = bi->bi_write_hint;
59fc630b 3218 /* Don't allow new IO added to stripes in batch list */
3219 if (sh->batch_head)
3220 goto overlap;
72626685 3221 if (forwrite) {
1da177e4 3222 bip = &sh->dev[dd_idx].towrite;
7eaf7e8e 3223 if (*bip == NULL)
72626685
N		 3224 firstwrite = 1;
3225 } else
1da177e4 3226 bip = &sh->dev[dd_idx].toread;
4f024f37
KO		 3227 while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) {
3228 if (bio_end_sector(*bip) > bi->bi_iter.bi_sector)
1da177e4
LT		 3229 goto overlap;
3230 bip = & (*bip)->bi_next;
3231 }
4f024f37 3232 if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi))
1da177e4
LT		 3233 goto overlap;
3234
3418d036
AP		 3235 if (forwrite && raid5_has_ppl(conf)) {
3236 /*
3237 * With PPL only writes to consecutive data chunks within a
3238 * stripe are allowed because for a single stripe_head we can
3239 * only have one PPL entry at a time, which describes one data
3240 * range. Not really an overlap, but wait_for_overlap can be
3241 * used to handle this.
3242 */
3243 sector_t sector;
3244 sector_t first = 0;
3245 sector_t last = 0;
3246 int count = 0;
3247 int i;
3248
3249 for (i = 0; i < sh->disks; i++) {
3250 if (i != sh->pd_idx &&
3251 (i == dd_idx || sh->dev[i].towrite)) {
3252 sector = sh->dev[i].sector;
3253 if (count == 0 || sector < first)
3254 first = sector;
3255 if (sector > last)
3256 last = sector;
3257 count++;
3258 }
3259 }
3260
3261 if (first + conf->chunk_sectors * (count - 1) != last)
3262 goto overlap;
3263 }
3264
da41ba65 3265 if (!forwrite || previous)
3266 clear_bit(STRIPE_BATCH_READY, &sh->state);
3267
78bafebd 3268 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
1da177e4
LT		 3269 if (*bip)
3270 bi->bi_next = *bip;
3271 *bip = bi;
016c76ac 3272 bio_inc_remaining(bi);
49728050 3273 md_write_inc(conf->mddev, bi);
72626685 3274
1da177e4
LT		 3275 if (forwrite) {
3276 /* check if page is covered */
3277 sector_t sector = sh->dev[dd_idx].sector;
3278 for (bi=sh->dev[dd_idx].towrite;
3279 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
4f024f37 3280 bi && bi->bi_iter.bi_sector <= sector;
1da177e4 3281 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
f73a1c7d
KO		 3282 if (bio_end_sector(bi) >= sector)
3283 sector = bio_end_sector(bi);
1da177e4
LT		 3284 }
3285 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
7a87f434 3286 if (!test_and_set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags))
3287 sh->overwrite_disks++;
1da177e4 3288 }
cbe47ec5
N		 3289
3290 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
4f024f37 3291 (unsigned long long)(*bip)->bi_iter.bi_sector,
cbe47ec5
N		 3292 (unsigned long long)sh->sector, dd_idx);
3293
3294 if (conf->mddev->bitmap && firstwrite) {
d0852df5
N		 3295 /* Cannot hold spinlock over bitmap_startwrite,
3296 * but must ensure this isn't added to a batch until
3297 * we have added to the bitmap and set bm_seq.
3298 * So set STRIPE_BITMAP_PENDING to prevent
3299 * batching.
3300 * If multiple add_stripe_bio() calls race here they
3301 * much all set STRIPE_BITMAP_PENDING. So only the first one
3302 * to complete "bitmap_startwrite" gets to set
3303 * STRIPE_BIT_DELAY. This is important as once a stripe
3304 * is added to a batch, STRIPE_BIT_DELAY cannot be changed
3305 * any more.
3306 */
3307 set_bit(STRIPE_BITMAP_PENDING, &sh->state);
3308 spin_unlock_irq(&sh->stripe_lock);
e64e4018
AS		 3309 md_bitmap_startwrite(conf->mddev->bitmap, sh->sector,
3310 STRIPE_SECTORS, 0);
d0852df5
N		 3311 spin_lock_irq(&sh->stripe_lock);
3312 clear_bit(STRIPE_BITMAP_PENDING, &sh->state);
3313 if (!sh->batch_head) {
3314 sh->bm_seq = conf->seq_flush+1;
3315 set_bit(STRIPE_BIT_DELAY, &sh->state);
3316 }
cbe47ec5 3317 }
d0852df5 3318 spin_unlock_irq(&sh->stripe_lock);
59fc630b 3319
3320 if (stripe_can_batch(sh))
3321 stripe_add_to_batch_list(conf, sh);
1da177e4
LT		 3322 return 1;
3323
3324 overlap:
3325 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
b17459c0 3326 spin_unlock_irq(&sh->stripe_lock);
1da177e4
LT		 3327 return 0;
3328}
3329
d1688a6d 3330static void end_reshape(struct r5conf *conf);
29269553 3331
d1688a6d 3332static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
911d4ee8 3333 struct stripe_head *sh)
ccfcc3c1 3334{
784052ec 3335 int sectors_per_chunk =
09c9e5fa 3336 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
911d4ee8 3337 int dd_idx;
2d2063ce 3338 int chunk_offset = sector_div(stripe, sectors_per_chunk);
112bf897 3339 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
2d2063ce 3340
112bf897
N		 3341 raid5_compute_sector(conf,
3342 stripe * (disks - conf->max_degraded)
b875e531 3343 *sectors_per_chunk + chunk_offset,
112bf897 3344 previous,
911d4ee8 3345 &dd_idx, sh);
ccfcc3c1
N		 3346}
3347
a4456856 3348static void
d1688a6d 3349handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh,
bd83d0a2 3350 struct stripe_head_state *s, int disks)
a4456856
DW		 3351{
3352 int i;
59fc630b 3353 BUG_ON(sh->batch_head);
a4456856
DW		 3354 for (i = disks; i--; ) {
3355 struct bio *bi;
3356 int bitmap_end = 0;
3357
3358 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
3cb03002 3359 struct md_rdev *rdev;
a4456856
DW		 3360 rcu_read_lock();
3361 rdev = rcu_dereference(conf->disks[i].rdev);
f5b67ae8
N		 3362 if (rdev && test_bit(In_sync, &rdev->flags) &&
3363 !test_bit(Faulty, &rdev->flags))
7f0da59b
N		 3364 atomic_inc(&rdev->nr_pending);
3365 else
3366 rdev = NULL;
a4456856 3367 rcu_read_unlock();
7f0da59b
N		 3368 if (rdev) {
3369 if (!rdev_set_badblocks(
3370 rdev,
3371 sh->sector,
3372 STRIPE_SECTORS, 0))
3373 md_error(conf->mddev, rdev);
3374 rdev_dec_pending(rdev, conf->mddev);
3375 }
a4456856 3376 }
b17459c0 3377 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 3378 /* fail all writes first */
3379 bi = sh->dev[i].towrite;
3380 sh->dev[i].towrite = NULL;
7a87f434 3381 sh->overwrite_disks = 0;
b17459c0 3382 spin_unlock_irq(&sh->stripe_lock);
1ed850f3 3383 if (bi)
a4456856 3384 bitmap_end = 1;
a4456856 3385
ff875738 3386 log_stripe_write_finished(sh);
0576b1c6 3387
a4456856
DW		 3388 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
3389 wake_up(&conf->wait_for_overlap);
3390
4f024f37 3391 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3392 sh->dev[i].sector + STRIPE_SECTORS) {
3393 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
4246a0b6 3394
49728050 3395 md_write_end(conf->mddev);
6308d8e3 3396 bio_io_error(bi);
a4456856
DW		 3397 bi = nextbi;
3398 }
7eaf7e8e 3399 if (bitmap_end)
e64e4018
AS		 3400 md_bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3401 STRIPE_SECTORS, 0, 0);
7eaf7e8e 3402 bitmap_end = 0;
a4456856
DW		 3403 /* and fail all 'written' */
3404 bi = sh->dev[i].written;
3405 sh->dev[i].written = NULL;
d592a996
SL		 3406 if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) {
3407 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
3408 sh->dev[i].page = sh->dev[i].orig_page;
3409 }
3410
a4456856 3411 if (bi) bitmap_end = 1;
4f024f37 3412 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3413 sh->dev[i].sector + STRIPE_SECTORS) {
3414 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
4246a0b6 3415
49728050 3416 md_write_end(conf->mddev);
6308d8e3 3417 bio_io_error(bi);
a4456856
DW		 3418 bi = bi2;
3419 }
3420
b5e98d65
DW		 3421 /* fail any reads if this device is non-operational and
3422 * the data has not reached the cache yet.
3423 */
3424 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
6e74a9cf 3425 s->failed > conf->max_degraded &&
b5e98d65
DW		 3426 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
3427 test_bit(R5_ReadError, &sh->dev[i].flags))) {
143c4d05 3428 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 3429 bi = sh->dev[i].toread;
3430 sh->dev[i].toread = NULL;
143c4d05 3431 spin_unlock_irq(&sh->stripe_lock);
a4456856
DW		 3432 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
3433 wake_up(&conf->wait_for_overlap);
ebda780b
SL		 3434 if (bi)
3435 s->to_read--;
4f024f37 3436 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3437 sh->dev[i].sector + STRIPE_SECTORS) {
3438 struct bio *nextbi =
3439 r5_next_bio(bi, sh->dev[i].sector);
4246a0b6 3440
6308d8e3 3441 bio_io_error(bi);
a4456856
DW		 3442 bi = nextbi;
3443 }
3444 }
a4456856 3445 if (bitmap_end)
e64e4018
AS		 3446 md_bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3447 STRIPE_SECTORS, 0, 0);
8cfa7b0f
N		 3448 /* If we were in the middle of a write the parity block might
3449 * still be locked - so just clear all R5_LOCKED flags
3450 */
3451 clear_bit(R5_LOCKED, &sh->dev[i].flags);
a4456856 3452 }
ebda780b
SL		 3453 s->to_write = 0;
3454 s->written = 0;
a4456856 3455
8b3e6cdc
DW		 3456 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3457 if (atomic_dec_and_test(&conf->pending_full_writes))
3458 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 3459}
3460
7f0da59b 3461static void
d1688a6d 3462handle_failed_sync(struct r5conf *conf, struct stripe_head *sh,
7f0da59b
N		 3463 struct stripe_head_state *s)
3464{
3465 int abort = 0;
3466 int i;
3467
59fc630b 3468 BUG_ON(sh->batch_head);
7f0da59b 3469 clear_bit(STRIPE_SYNCING, &sh->state);
f8dfcffd
N		 3470 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
3471 wake_up(&conf->wait_for_overlap);
7f0da59b 3472 s->syncing = 0;
9a3e1101 3473 s->replacing = 0;
7f0da59b 3474 /* There is nothing more to do for sync/check/repair.
18b9837e
N		 3475 * Don't even need to abort as that is handled elsewhere
3476 * if needed, and not always wanted e.g. if there is a known
3477 * bad block here.
9a3e1101 3478 * For recover/replace we need to record a bad block on all
7f0da59b
N		 3479 * non-sync devices, or abort the recovery
3480 */
18b9837e
N		 3481 if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) {
3482 /* During recovery devices cannot be removed, so
3483 * locking and refcounting of rdevs is not needed
3484 */
e50d3992 3485 rcu_read_lock();
18b9837e 3486 for (i = 0; i < conf->raid_disks; i++) {
e50d3992 3487 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
18b9837e
N		 3488 if (rdev
3489 && !test_bit(Faulty, &rdev->flags)
3490 && !test_bit(In_sync, &rdev->flags)
3491 && !rdev_set_badblocks(rdev, sh->sector,
3492 STRIPE_SECTORS, 0))
3493 abort = 1;
e50d3992 3494 rdev = rcu_dereference(conf->disks[i].replacement);
18b9837e
N		 3495 if (rdev
3496 && !test_bit(Faulty, &rdev->flags)
3497 && !test_bit(In_sync, &rdev->flags)
3498 && !rdev_set_badblocks(rdev, sh->sector,
3499 STRIPE_SECTORS, 0))
3500 abort = 1;
3501 }
e50d3992 3502 rcu_read_unlock();
18b9837e
N		 3503 if (abort)
3504 conf->recovery_disabled =
3505 conf->mddev->recovery_disabled;
7f0da59b 3506 }
18b9837e 3507 md_done_sync(conf->mddev, STRIPE_SECTORS, !abort);
7f0da59b
N		 3508}
3509
9a3e1101
N		 3510static int want_replace(struct stripe_head *sh, int disk_idx)
3511{
3512 struct md_rdev *rdev;
3513 int rv = 0;
3f232d6a
N		 3514
3515 rcu_read_lock();
3516 rdev = rcu_dereference(sh->raid_conf->disks[disk_idx].replacement);
9a3e1101
N		 3517 if (rdev
3518 && !test_bit(Faulty, &rdev->flags)
3519 && !test_bit(In_sync, &rdev->flags)
3520 && (rdev->recovery_offset <= sh->sector
3521 || rdev->mddev->recovery_cp <= sh->sector))
3522 rv = 1;
3f232d6a 3523 rcu_read_unlock();
9a3e1101
N		 3524 return rv;
3525}
3526
2c58f06e
N		 3527static int need_this_block(struct stripe_head *sh, struct stripe_head_state *s,
3528 int disk_idx, int disks)
a4456856 3529{
5599becc 3530 struct r5dev *dev = &sh->dev[disk_idx];
f2b3b44d
N		 3531 struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]],
3532 &sh->dev[s->failed_num[1]] };
ea664c82 3533 int i;
5599becc 3534
a79cfe12
N		 3535
3536 if (test_bit(R5_LOCKED, &dev->flags) ||
3537 test_bit(R5_UPTODATE, &dev->flags))
3538 /* No point reading this as we already have it or have
3539 * decided to get it.
3540 */
3541 return 0;
3542
3543 if (dev->toread ||
3544 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)))
3545 /* We need this block to directly satisfy a request */
3546 return 1;
3547
3548 if (s->syncing || s->expanding ||
3549 (s->replacing && want_replace(sh, disk_idx)))
3550 /* When syncing, or expanding we read everything.
3551 * When replacing, we need the replaced block.
3552 */
3553 return 1;
3554
3555 if ((s->failed >= 1 && fdev[0]->toread) ||
3556 (s->failed >= 2 && fdev[1]->toread))
3557 /* If we want to read from a failed device, then
3558 * we need to actually read every other device.
3559 */
3560 return 1;
3561
a9d56950
N		 3562 /* Sometimes neither read-modify-write nor reconstruct-write
3563 * cycles can work. In those cases we read every block we
3564 * can. Then the parity-update is certain to have enough to
3565 * work with.
3566 * This can only be a problem when we need to write something,
3567 * and some device has failed. If either of those tests
3568 * fail we need look no further.
3569 */
3570 if (!s->failed || !s->to_write)
3571 return 0;
3572
3573 if (test_bit(R5_Insync, &dev->flags) &&
3574 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3575 /* Pre-reads at not permitted until after short delay
3576 * to gather multiple requests. However if this
3560741e 3577 * device is no Insync, the block could only be computed
a9d56950
N		 3578 * and there is no need to delay that.
3579 */
3580 return 0;
ea664c82 3581
36707bb2 3582 for (i = 0; i < s->failed && i < 2; i++) {
ea664c82
N		 3583 if (fdev[i]->towrite &&
3584 !test_bit(R5_UPTODATE, &fdev[i]->flags) &&
3585 !test_bit(R5_OVERWRITE, &fdev[i]->flags))
3586 /* If we have a partial write to a failed
3587 * device, then we will need to reconstruct
3588 * the content of that device, so all other
3589 * devices must be read.
3590 */
3591 return 1;
3592 }
3593
3594 /* If we are forced to do a reconstruct-write, either because
3595 * the current RAID6 implementation only supports that, or
3560741e 3596 * because parity cannot be trusted and we are currently
ea664c82
N		 3597 * recovering it, there is extra need to be careful.
3598 * If one of the devices that we would need to read, because
3599 * it is not being overwritten (and maybe not written at all)
3600 * is missing/faulty, then we need to read everything we can.
3601 */
3602 if (sh->raid_conf->level != 6 &&
3603 sh->sector < sh->raid_conf->mddev->recovery_cp)
3604 /* reconstruct-write isn't being forced */
3605 return 0;
36707bb2 3606 for (i = 0; i < s->failed && i < 2; i++) {
10d82c5f
N		 3607 if (s->failed_num[i] != sh->pd_idx &&
3608 s->failed_num[i] != sh->qd_idx &&
3609 !test_bit(R5_UPTODATE, &fdev[i]->flags) &&
ea664c82
N		 3610 !test_bit(R5_OVERWRITE, &fdev[i]->flags))
3611 return 1;
3612 }
3613
2c58f06e
N		 3614 return 0;
3615}
3616
ba02684d
SL		 3617/* fetch_block - checks the given member device to see if its data needs
3618 * to be read or computed to satisfy a request.
3619 *
3620 * Returns 1 when no more member devices need to be checked, otherwise returns
3621 * 0 to tell the loop in handle_stripe_fill to continue
3622 */
2c58f06e
N		 3623static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s,
3624 int disk_idx, int disks)
3625{
3626 struct r5dev *dev = &sh->dev[disk_idx];
3627
3628 /* is the data in this block needed, and can we get it? */
3629 if (need_this_block(sh, s, disk_idx, disks)) {
5599becc
YT		 3630 /* we would like to get this block, possibly by computing it,
3631 * otherwise read it if the backing disk is insync
3632 */
3633 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
3634 BUG_ON(test_bit(R5_Wantread, &dev->flags));
b0c783b3 3635 BUG_ON(sh->batch_head);
7471fb77
N		 3636
3637 /*
3638 * In the raid6 case if the only non-uptodate disk is P
3639 * then we already trusted P to compute the other failed
3640 * drives. It is safe to compute rather than re-read P.
3641 * In other cases we only compute blocks from failed
3642 * devices, otherwise check/repair might fail to detect
3643 * a real inconsistency.
3644 */
3645
5599becc 3646 if ((s->uptodate == disks - 1) &&
7471fb77 3647 ((sh->qd_idx >= 0 && sh->pd_idx == disk_idx) ||
f2b3b44d 3648 (s->failed && (disk_idx == s->failed_num[0] ||
7471fb77 3649 disk_idx == s->failed_num[1])))) {
5599becc
YT		 3650 /* have disk failed, and we're requested to fetch it;
3651 * do compute it
a4456856 3652 */
5599becc
YT		 3653 pr_debug("Computing stripe %llu block %d\n",
3654 (unsigned long long)sh->sector, disk_idx);
3655 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3656 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3657 set_bit(R5_Wantcompute, &dev->flags);
3658 sh->ops.target = disk_idx;
3659 sh->ops.target2 = -1; /* no 2nd target */
3660 s->req_compute = 1;
93b3dbce
N		 3661 /* Careful: from this point on 'uptodate' is in the eye
3662 * of raid_run_ops which services 'compute' operations
3663 * before writes. R5_Wantcompute flags a block that will
3664 * be R5_UPTODATE by the time it is needed for a
3665 * subsequent operation.
3666 */
5599becc
YT		 3667 s->uptodate++;
3668 return 1;
3669 } else if (s->uptodate == disks-2 && s->failed >= 2) {
3670 /* Computing 2-failure is *very* expensive; only
3671 * do it if failed >= 2
3672 */
3673 int other;
3674 for (other = disks; other--; ) {
3675 if (other == disk_idx)
3676 continue;
3677 if (!test_bit(R5_UPTODATE,
3678 &sh->dev[other].flags))
3679 break;
a4456856 3680 }
5599becc
YT		 3681 BUG_ON(other < 0);
3682 pr_debug("Computing stripe %llu blocks %d,%d\n",
3683 (unsigned long long)sh->sector,
3684 disk_idx, other);
3685 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3686 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3687 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
3688 set_bit(R5_Wantcompute, &sh->dev[other].flags);
3689 sh->ops.target = disk_idx;
3690 sh->ops.target2 = other;
3691 s->uptodate += 2;
3692 s->req_compute = 1;
3693 return 1;
3694 } else if (test_bit(R5_Insync, &dev->flags)) {
3695 set_bit(R5_LOCKED, &dev->flags);
3696 set_bit(R5_Wantread, &dev->flags);
3697 s->locked++;
3698 pr_debug("Reading block %d (sync=%d)\n",
3699 disk_idx, s->syncing);
a4456856
DW		 3700 }
3701 }
5599becc
YT		 3702
3703 return 0;
3704}
3705
3706/**
93b3dbce 3707 * handle_stripe_fill - read or compute data to satisfy pending requests.
5599becc 3708 */
93b3dbce
N		 3709static void handle_stripe_fill(struct stripe_head *sh,
3710 struct stripe_head_state *s,
3711 int disks)
5599becc
YT		 3712{
3713 int i;
3714
3715 /* look for blocks to read/compute, skip this if a compute
3716 * is already in flight, or if the stripe contents are in the
3717 * midst of changing due to a write
3718 */
3719 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
07e83364
SL		 3720 !sh->reconstruct_state) {
3721
3722 /*
3723 * For degraded stripe with data in journal, do not handle
3724 * read requests yet, instead, flush the stripe to raid
3725 * disks first, this avoids handling complex rmw of write
3726 * back cache (prexor with orig_page, and then xor with
3727 * page) in the read path
3728 */
3729 if (s->injournal && s->failed) {
3730 if (test_bit(STRIPE_R5C_CACHING, &sh->state))
3731 r5c_make_stripe_write_out(sh);
3732 goto out;
3733 }
3734
5599becc 3735 for (i = disks; i--; )
93b3dbce 3736 if (fetch_block(sh, s, i, disks))
5599becc 3737 break;
07e83364
SL		 3738 }
3739out:
a4456856
DW		 3740 set_bit(STRIPE_HANDLE, &sh->state);
3741}
3742
787b76fa
N		 3743static void break_stripe_batch_list(struct stripe_head *head_sh,
3744 unsigned long handle_flags);
1fe797e6 3745/* handle_stripe_clean_event
a4456856
DW		 3746 * any written block on an uptodate or failed drive can be returned.
3747 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
3748 * never LOCKED, so we don't need to test 'failed' directly.
3749 */
d1688a6d 3750static void handle_stripe_clean_event(struct r5conf *conf,
bd83d0a2 3751 struct stripe_head *sh, int disks)
a4456856
DW		 3752{
3753 int i;
3754 struct r5dev *dev;
f8dfcffd 3755 int discard_pending = 0;
59fc630b 3756 struct stripe_head *head_sh = sh;
3757 bool do_endio = false;
a4456856
DW		 3758
3759 for (i = disks; i--; )
3760 if (sh->dev[i].written) {
3761 dev = &sh->dev[i];
3762 if (!test_bit(R5_LOCKED, &dev->flags) &&
9e444768 3763 (test_bit(R5_UPTODATE, &dev->flags) ||
d592a996
SL		 3764 test_bit(R5_Discard, &dev->flags) ||
3765 test_bit(R5_SkipCopy, &dev->flags))) {
a4456856
DW		 3766 /* We can return any write requests */
3767 struct bio *wbi, *wbi2;
45b4233c 3768 pr_debug("Return write for disc %d\n", i);
ca64cae9
N		 3769 if (test_and_clear_bit(R5_Discard, &dev->flags))
3770 clear_bit(R5_UPTODATE, &dev->flags);
d592a996
SL		 3771 if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) {
3772 WARN_ON(test_bit(R5_UPTODATE, &dev->flags));
d592a996 3773 }
59fc630b 3774 do_endio = true;
3775
3776returnbi:
3777 dev->page = dev->orig_page;
a4456856
DW		 3778 wbi = dev->written;
3779 dev->written = NULL;
4f024f37 3780 while (wbi && wbi->bi_iter.bi_sector <
a4456856
DW		 3781 dev->sector + STRIPE_SECTORS) {
3782 wbi2 = r5_next_bio(wbi, dev->sector);
49728050 3783 md_write_end(conf->mddev);
016c76ac 3784 bio_endio(wbi);
a4456856
DW		 3785 wbi = wbi2;
3786 }
e64e4018
AS		 3787 md_bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3788 STRIPE_SECTORS,
3789 !test_bit(STRIPE_DEGRADED, &sh->state),
3790 0);
59fc630b 3791 if (head_sh->batch_head) {
3792 sh = list_first_entry(&sh->batch_list,
3793 struct stripe_head,
3794 batch_list);
3795 if (sh != head_sh) {
3796 dev = &sh->dev[i];
3797 goto returnbi;
3798 }
3799 }
3800 sh = head_sh;
3801 dev = &sh->dev[i];
f8dfcffd
N		 3802 } else if (test_bit(R5_Discard, &dev->flags))
3803 discard_pending = 1;
3804 }
f6bed0ef 3805
ff875738 3806 log_stripe_write_finished(sh);
0576b1c6 3807
f8dfcffd
N		 3808 if (!discard_pending &&
3809 test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) {
b8a9d66d 3810 int hash;
f8dfcffd
N		 3811 clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
3812 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
3813 if (sh->qd_idx >= 0) {
3814 clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
3815 clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags);
3816 }
3817 /* now that discard is done we can proceed with any sync */
3818 clear_bit(STRIPE_DISCARD, &sh->state);
d47648fc
SL		 3819 /*
3820 * SCSI discard will change some bio fields and the stripe has
3821 * no updated data, so remove it from hash list and the stripe
3822 * will be reinitialized
3823 */
59fc630b 3824unhash:
b8a9d66d
RG		 3825 hash = sh->hash_lock_index;
3826 spin_lock_irq(conf->hash_locks + hash);
d47648fc 3827 remove_hash(sh);
b8a9d66d 3828 spin_unlock_irq(conf->hash_locks + hash);
59fc630b 3829 if (head_sh->batch_head) {
3830 sh = list_first_entry(&sh->batch_list,
3831 struct stripe_head, batch_list);
3832 if (sh != head_sh)
3833 goto unhash;
3834 }
59fc630b 3835 sh = head_sh;
3836
f8dfcffd
N		 3837 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state))
3838 set_bit(STRIPE_HANDLE, &sh->state);
3839
3840 }
8b3e6cdc
DW		 3841
3842 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3843 if (atomic_dec_and_test(&conf->pending_full_writes))
3844 md_wakeup_thread(conf->mddev->thread);
59fc630b 3845
787b76fa
N		 3846 if (head_sh->batch_head && do_endio)
3847 break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS);
a4456856
DW		 3848}
3849
86aa1397
SL		 3850/*
3851 * For RMW in write back cache, we need extra page in prexor to store the
3852 * old data. This page is stored in dev->orig_page.
3853 *
3854 * This function checks whether we have data for prexor. The exact logic
3855 * is:
3856 * R5_UPTODATE && (!R5_InJournal || R5_OrigPageUPTDODATE)
3857 */
3858static inline bool uptodate_for_rmw(struct r5dev *dev)
3859{
3860 return (test_bit(R5_UPTODATE, &dev->flags)) &&
3861 (!test_bit(R5_InJournal, &dev->flags) ||
3862 test_bit(R5_OrigPageUPTDODATE, &dev->flags));
3863}
3864
d7bd398e
SL		 3865static int handle_stripe_dirtying(struct r5conf *conf,
3866 struct stripe_head *sh,
3867 struct stripe_head_state *s,
3868 int disks)
a4456856
DW		 3869{
3870 int rmw = 0, rcw = 0, i;
a7854487
AL		 3871 sector_t recovery_cp = conf->mddev->recovery_cp;
3872
584acdd4 3873 /* Check whether resync is now happening or should start.
a7854487
AL		 3874 * If yes, then the array is dirty (after unclean shutdown or
3875 * initial creation), so parity in some stripes might be inconsistent.
3876 * In this case, we need to always do reconstruct-write, to ensure
3877 * that in case of drive failure or read-error correction, we
3878 * generate correct data from the parity.
3879 */
584acdd4 3880 if (conf->rmw_level == PARITY_DISABLE_RMW ||
26ac1073
N		 3881 (recovery_cp < MaxSector && sh->sector >= recovery_cp &&
3882 s->failed == 0)) {
a7854487 3883 /* Calculate the real rcw later - for now make it
c8ac1803
N		 3884 * look like rcw is cheaper
3885 */
3886 rcw = 1; rmw = 2;
584acdd4
MS		 3887 pr_debug("force RCW rmw_level=%u, recovery_cp=%llu sh->sector=%llu\n",
3888 conf->rmw_level, (unsigned long long)recovery_cp,
a7854487 3889 (unsigned long long)sh->sector);
c8ac1803 3890 } else for (i = disks; i--; ) {
a4456856
DW		 3891 /* would I have to read this buffer for read_modify_write */
3892 struct r5dev *dev = &sh->dev[i];
39b99586 3893 if (((dev->towrite && !delay_towrite(conf, dev, s)) ||
07e83364 3894 i == sh->pd_idx || i == sh->qd_idx ||
1e6d690b 3895 test_bit(R5_InJournal, &dev->flags)) &&
a4456856 3896 !test_bit(R5_LOCKED, &dev->flags) &&
86aa1397 3897 !(uptodate_for_rmw(dev) ||
f38e1219 3898 test_bit(R5_Wantcompute, &dev->flags))) {
a4456856
DW		 3899 if (test_bit(R5_Insync, &dev->flags))
3900 rmw++;
3901 else
3902 rmw += 2*disks; /* cannot read it */
3903 }
3904 /* Would I have to read this buffer for reconstruct_write */
584acdd4
MS		 3905 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
3906 i != sh->pd_idx && i != sh->qd_idx &&
a4456856 3907 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219 3908 !(test_bit(R5_UPTODATE, &dev->flags) ||
1e6d690b 3909 test_bit(R5_Wantcompute, &dev->flags))) {
67f45548
N		 3910 if (test_bit(R5_Insync, &dev->flags))
3911 rcw++;
a4456856
DW		 3912 else
3913 rcw += 2*disks;
3914 }
3915 }
1e6d690b 3916
39b99586
SL		 3917 pr_debug("for sector %llu state 0x%lx, rmw=%d rcw=%d\n",
3918 (unsigned long long)sh->sector, sh->state, rmw, rcw);
a4456856 3919 set_bit(STRIPE_HANDLE, &sh->state);
41257580 3920 if ((rmw < rcw || (rmw == rcw && conf->rmw_level == PARITY_PREFER_RMW)) && rmw > 0) {
a4456856 3921 /* prefer read-modify-write, but need to get some data */
e3620a3a
JB		 3922 if (conf->mddev->queue)
3923 blk_add_trace_msg(conf->mddev->queue,
3924 "raid5 rmw %llu %d",
3925 (unsigned long long)sh->sector, rmw);
a4456856
DW		 3926 for (i = disks; i--; ) {
3927 struct r5dev *dev = &sh->dev[i];
1e6d690b
SL		 3928 if (test_bit(R5_InJournal, &dev->flags) &&
3929 dev->page == dev->orig_page &&
3930 !test_bit(R5_LOCKED, &sh->dev[sh->pd_idx].flags)) {
3931 /* alloc page for prexor */
d7bd398e
SL		 3932 struct page *p = alloc_page(GFP_NOIO);
3933
3934 if (p) {
3935 dev->orig_page = p;
3936 continue;
3937 }
3938
3939 /*
3940 * alloc_page() failed, try use
3941 * disk_info->extra_page
3942 */
3943 if (!test_and_set_bit(R5C_EXTRA_PAGE_IN_USE,
3944 &conf->cache_state)) {
3945 r5c_use_extra_page(sh);
3946 break;
3947 }
1e6d690b 3948
d7bd398e
SL		 3949 /* extra_page in use, add to delayed_list */
3950 set_bit(STRIPE_DELAYED, &sh->state);
3951 s->waiting_extra_page = 1;
3952 return -EAGAIN;
1e6d690b 3953 }
d7bd398e 3954 }
1e6d690b 3955
d7bd398e
SL		 3956 for (i = disks; i--; ) {
3957 struct r5dev *dev = &sh->dev[i];
39b99586 3958 if (((dev->towrite && !delay_towrite(conf, dev, s)) ||
1e6d690b
SL		 3959 i == sh->pd_idx || i == sh->qd_idx ||
3960 test_bit(R5_InJournal, &dev->flags)) &&
a4456856 3961 !test_bit(R5_LOCKED, &dev->flags) &&
86aa1397 3962 !(uptodate_for_rmw(dev) ||
1e6d690b 3963 test_bit(R5_Wantcompute, &dev->flags)) &&
a4456856 3964 test_bit(R5_Insync, &dev->flags)) {
67f45548
N		 3965 if (test_bit(STRIPE_PREREAD_ACTIVE,
3966 &sh->state)) {
3967 pr_debug("Read_old block %d for r-m-w\n",
3968 i);
a4456856
DW		 3969 set_bit(R5_LOCKED, &dev->flags);
3970 set_bit(R5_Wantread, &dev->flags);
3971 s->locked++;
3972 } else {
3973 set_bit(STRIPE_DELAYED, &sh->state);
3974 set_bit(STRIPE_HANDLE, &sh->state);
3975 }
3976 }
3977 }
a9add5d9 3978 }
41257580 3979 if ((rcw < rmw || (rcw == rmw && conf->rmw_level != PARITY_PREFER_RMW)) && rcw > 0) {
a4456856 3980 /* want reconstruct write, but need to get some data */
a9add5d9 3981 int qread =0;
c8ac1803 3982 rcw = 0;
a4456856
DW		 3983 for (i = disks; i--; ) {
3984 struct r5dev *dev = &sh->dev[i];
3985 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
c8ac1803 3986 i != sh->pd_idx && i != sh->qd_idx &&
a4456856 3987 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219 3988 !(test_bit(R5_UPTODATE, &dev->flags) ||
c8ac1803
N		 3989 test_bit(R5_Wantcompute, &dev->flags))) {
3990 rcw++;
67f45548
N		 3991 if (test_bit(R5_Insync, &dev->flags) &&
3992 test_bit(STRIPE_PREREAD_ACTIVE,
3993 &sh->state)) {
45b4233c 3994 pr_debug("Read_old block "
a4456856
DW		 3995 "%d for Reconstruct\n", i);
3996 set_bit(R5_LOCKED, &dev->flags);
3997 set_bit(R5_Wantread, &dev->flags);
3998 s->locked++;
a9add5d9 3999 qread++;
a4456856
DW		 4000 } else {
4001 set_bit(STRIPE_DELAYED, &sh->state);
4002 set_bit(STRIPE_HANDLE, &sh->state);
4003 }
4004 }
4005 }
e3620a3a 4006 if (rcw && conf->mddev->queue)
a9add5d9
N		 4007 blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d",
4008 (unsigned long long)sh->sector,
4009 rcw, qread, test_bit(STRIPE_DELAYED, &sh->state));
c8ac1803 4010 }
b1b02fe9
N		 4011
4012 if (rcw > disks && rmw > disks &&
4013 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4014 set_bit(STRIPE_DELAYED, &sh->state);
4015
a4456856
DW		 4016 /* now if nothing is locked, and if we have enough data,
4017 * we can start a write request
4018 */
f38e1219
DW		 4019 /* since handle_stripe can be called at any time we need to handle the
4020 * case where a compute block operation has been submitted and then a
ac6b53b6
DW		 4021 * subsequent call wants to start a write request. raid_run_ops only
4022 * handles the case where compute block and reconstruct are requested
f38e1219
DW		 4023 * simultaneously. If this is not the case then new writes need to be
4024 * held off until the compute completes.
4025 */
976ea8d4
DW		 4026 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
4027 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
1e6d690b 4028 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
c0f7bddb 4029 schedule_reconstruction(sh, s, rcw == 0, 0);
d7bd398e 4030 return 0;
a4456856
DW		 4031}
4032
d1688a6d 4033static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh,
a4456856
DW		 4034 struct stripe_head_state *s, int disks)
4035{
ecc65c9b 4036 struct r5dev *dev = NULL;
bd2ab670 4037
59fc630b 4038 BUG_ON(sh->batch_head);
a4456856 4039 set_bit(STRIPE_HANDLE, &sh->state);
e89f8962 4040
ecc65c9b
DW		 4041 switch (sh->check_state) {
4042 case check_state_idle:
4043 /* start a new check operation if there are no failures */
bd2ab670 4044 if (s->failed == 0) {
bd2ab670 4045 BUG_ON(s->uptodate != disks);
ecc65c9b
DW		 4046 sh->check_state = check_state_run;
4047 set_bit(STRIPE_OP_CHECK, &s->ops_request);
bd2ab670 4048 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
bd2ab670 4049 s->uptodate--;
ecc65c9b 4050 break;
bd2ab670 4051 }
f2b3b44d 4052 dev = &sh->dev[s->failed_num[0]];
ecc65c9b
DW		 4053 /* fall through */
4054 case check_state_compute_result:
4055 sh->check_state = check_state_idle;
4056 if (!dev)
4057 dev = &sh->dev[sh->pd_idx];
4058
4059 /* check that a write has not made the stripe insync */
4060 if (test_bit(STRIPE_INSYNC, &sh->state))
4061 break;
c8894419 4062
a4456856 4063 /* either failed parity check, or recovery is happening */
a4456856
DW		 4064 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
4065 BUG_ON(s->uptodate != disks);
4066
4067 set_bit(R5_LOCKED, &dev->flags);
ecc65c9b 4068 s->locked++;
a4456856 4069 set_bit(R5_Wantwrite, &dev->flags);
830ea016 4070
a4456856 4071 clear_bit(STRIPE_DEGRADED, &sh->state);
a4456856 4072 set_bit(STRIPE_INSYNC, &sh->state);
ecc65c9b
DW		 4073 break;
4074 case check_state_run:
4075 break; /* we will be called again upon completion */
4076 case check_state_check_result:
4077 sh->check_state = check_state_idle;
4078
4079 /* if a failure occurred during the check operation, leave
4080 * STRIPE_INSYNC not set and let the stripe be handled again
4081 */
4082 if (s->failed)
4083 break;
4084
4085 /* handle a successful check operation, if parity is correct
4086 * we are done. Otherwise update the mismatch count and repair
4087 * parity if !MD_RECOVERY_CHECK
4088 */
ad283ea4 4089 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
ecc65c9b
DW		 4090 /* parity is correct (on disc,
4091 * not in buffer any more)
4092 */
4093 set_bit(STRIPE_INSYNC, &sh->state);
4094 else {
7f7583d4 4095 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
e1539036 4096 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) {
ecc65c9b
DW		 4097 /* don't try to repair!! */
4098 set_bit(STRIPE_INSYNC, &sh->state);
e1539036
N		 4099 pr_warn_ratelimited("%s: mismatch sector in range "
4100 "%llu-%llu\n", mdname(conf->mddev),
4101 (unsigned long long) sh->sector,
4102 (unsigned long long) sh->sector +
4103 STRIPE_SECTORS);
4104 } else {
ecc65c9b 4105 sh->check_state = check_state_compute_run;
976ea8d4 4106 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
ecc65c9b
DW		 4107 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
4108 set_bit(R5_Wantcompute,
4109 &sh->dev[sh->pd_idx].flags);
4110 sh->ops.target = sh->pd_idx;
ac6b53b6 4111 sh->ops.target2 = -1;
ecc65c9b
DW		 4112 s->uptodate++;
4113 }
4114 }
4115 break;
4116 case check_state_compute_run:
4117 break;
4118 default:
cc6167b4 4119 pr_err("%s: unknown check_state: %d sector: %llu\n",
ecc65c9b
DW		 4120 __func__, sh->check_state,
4121 (unsigned long long) sh->sector);
4122 BUG();
a4456856
DW		 4123 }
4124}
4125
d1688a6d 4126static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh,
36d1c647 4127 struct stripe_head_state *s,
f2b3b44d 4128 int disks)
a4456856 4129{
a4456856 4130 int pd_idx = sh->pd_idx;
34e04e87 4131 int qd_idx = sh->qd_idx;
d82dfee0 4132 struct r5dev *dev;
a4456856 4133
59fc630b 4134 BUG_ON(sh->batch_head);
a4456856
DW		 4135 set_bit(STRIPE_HANDLE, &sh->state);
4136
4137 BUG_ON(s->failed > 2);
d82dfee0 4138
a4456856
DW		 4139 /* Want to check and possibly repair P and Q.
4140 * However there could be one 'failed' device, in which
4141 * case we can only check one of them, possibly using the
4142 * other to generate missing data
4143 */
4144
d82dfee0
DW		 4145 switch (sh->check_state) {
4146 case check_state_idle:
4147 /* start a new check operation if there are < 2 failures */
f2b3b44d 4148 if (s->failed == s->q_failed) {
d82dfee0 4149 /* The only possible failed device holds Q, so it
a4456856
DW		 4150 * makes sense to check P (If anything else were failed,
4151 * we would have used P to recreate it).
4152 */
d82dfee0 4153 sh->check_state = check_state_run;
a4456856 4154 }
f2b3b44d 4155 if (!s->q_failed && s->failed < 2) {
d82dfee0 4156 /* Q is not failed, and we didn't use it to generate
a4456856
DW		 4157 * anything, so it makes sense to check it
4158 */
d82dfee0
DW		 4159 if (sh->check_state == check_state_run)
4160 sh->check_state = check_state_run_pq;
4161 else
4162 sh->check_state = check_state_run_q;
a4456856 4163 }
a4456856 4164
d82dfee0
DW		 4165 /* discard potentially stale zero_sum_result */
4166 sh->ops.zero_sum_result = 0;
a4456856 4167
d82dfee0
DW		 4168 if (sh->check_state == check_state_run) {
4169 /* async_xor_zero_sum destroys the contents of P */
4170 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
4171 s->uptodate--;
a4456856 4172 }
d82dfee0
DW		 4173 if (sh->check_state >= check_state_run &&
4174 sh->check_state <= check_state_run_pq) {
4175 /* async_syndrome_zero_sum preserves P and Q, so
4176 * no need to mark them !uptodate here
4177 */
4178 set_bit(STRIPE_OP_CHECK, &s->ops_request);
4179 break;
a4456856
DW		 4180 }
4181
d82dfee0
DW		 4182 /* we have 2-disk failure */
4183 BUG_ON(s->failed != 2);
4184 /* fall through */
4185 case check_state_compute_result:
4186 sh->check_state = check_state_idle;
a4456856 4187
d82dfee0
DW		 4188 /* check that a write has not made the stripe insync */
4189 if (test_bit(STRIPE_INSYNC, &sh->state))
4190 break;
a4456856
DW		 4191
4192 /* now write out any block on a failed drive,
d82dfee0 4193 * or P or Q if they were recomputed
a4456856 4194 */
b2176a1d 4195 dev = NULL;
a4456856 4196 if (s->failed == 2) {
f2b3b44d 4197 dev = &sh->dev[s->failed_num[1]];
a4456856
DW		 4198 s->locked++;
4199 set_bit(R5_LOCKED, &dev->flags);
4200 set_bit(R5_Wantwrite, &dev->flags);
4201 }
4202 if (s->failed >= 1) {
f2b3b44d 4203 dev = &sh->dev[s->failed_num[0]];
a4456856
DW		 4204 s->locked++;
4205 set_bit(R5_LOCKED, &dev->flags);
4206 set_bit(R5_Wantwrite, &dev->flags);
4207 }
d82dfee0 4208 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
a4456856
DW		 4209 dev = &sh->dev[pd_idx];
4210 s->locked++;
4211 set_bit(R5_LOCKED, &dev->flags);
4212 set_bit(R5_Wantwrite, &dev->flags);
4213 }
d82dfee0 4214 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
a4456856
DW		 4215 dev = &sh->dev[qd_idx];
4216 s->locked++;
4217 set_bit(R5_LOCKED, &dev->flags);
4218 set_bit(R5_Wantwrite, &dev->flags);
4219 }
b2176a1d
NC		 4220 if (WARN_ONCE(dev && !test_bit(R5_UPTODATE, &dev->flags),
4221 "%s: disk%td not up to date\n",
4222 mdname(conf->mddev),
4223 dev - (struct r5dev *) &sh->dev)) {
4224 clear_bit(R5_LOCKED, &dev->flags);
4225 clear_bit(R5_Wantwrite, &dev->flags);
4226 s->locked--;
4227 }
a4456856
DW		 4228 clear_bit(STRIPE_DEGRADED, &sh->state);
4229
4230 set_bit(STRIPE_INSYNC, &sh->state);
d82dfee0
DW		 4231 break;
4232 case check_state_run:
4233 case check_state_run_q:
4234 case check_state_run_pq:
4235 break; /* we will be called again upon completion */
4236 case check_state_check_result:
4237 sh->check_state = check_state_idle;
4238
4239 /* handle a successful check operation, if parity is correct
4240 * we are done. Otherwise update the mismatch count and repair
4241 * parity if !MD_RECOVERY_CHECK
4242 */
4243 if (sh->ops.zero_sum_result == 0) {
a25d8c32
SL		 4244 /* both parities are correct */
4245 if (!s->failed)
4246 set_bit(STRIPE_INSYNC, &sh->state);
4247 else {
4248 /* in contrast to the raid5 case we can validate
4249 * parity, but still have a failure to write
4250 * back
4251 */
4252 sh->check_state = check_state_compute_result;
4253 /* Returning at this point means that we may go
4254 * off and bring p and/or q uptodate again so
4255 * we make sure to check zero_sum_result again
4256 * to verify if p or q need writeback
4257 */
4258 }
d82dfee0 4259 } else {
7f7583d4 4260 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
e1539036 4261 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) {
d82dfee0
DW		 4262 /* don't try to repair!! */
4263 set_bit(STRIPE_INSYNC, &sh->state);
e1539036
N		 4264 pr_warn_ratelimited("%s: mismatch sector in range "
4265 "%llu-%llu\n", mdname(conf->mddev),
4266 (unsigned long long) sh->sector,
4267 (unsigned long long) sh->sector +
4268 STRIPE_SECTORS);
4269 } else {
d82dfee0
DW		 4270 int *target = &sh->ops.target;
4271
4272 sh->ops.target = -1;
4273 sh->ops.target2 = -1;
4274 sh->check_state = check_state_compute_run;
4275 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
4276 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
4277 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
4278 set_bit(R5_Wantcompute,
4279 &sh->dev[pd_idx].flags);
4280 *target = pd_idx;
4281 target = &sh->ops.target2;
4282 s->uptodate++;
4283 }
4284 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
4285 set_bit(R5_Wantcompute,
4286 &sh->dev[qd_idx].flags);
4287 *target = qd_idx;
4288 s->uptodate++;
4289 }
4290 }
4291 }
4292 break;
4293 case check_state_compute_run:
4294 break;
4295 default:
cc6167b4
N		 4296 pr_warn("%s: unknown check_state: %d sector: %llu\n",
4297 __func__, sh->check_state,
4298 (unsigned long long) sh->sector);
d82dfee0 4299 BUG();
a4456856
DW		 4300 }
4301}
4302
d1688a6d 4303static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh)
a4456856
DW		 4304{
4305 int i;
4306
4307 /* We have read all the blocks in this stripe and now we need to
4308 * copy some of them into a target stripe for expand.
4309 */
f0a50d37 4310 struct dma_async_tx_descriptor *tx = NULL;
59fc630b 4311 BUG_ON(sh->batch_head);
a4456856
DW		 4312 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
4313 for (i = 0; i < sh->disks; i++)
34e04e87 4314 if (i != sh->pd_idx && i != sh->qd_idx) {
911d4ee8 4315 int dd_idx, j;
a4456856 4316 struct stripe_head *sh2;
a08abd8c 4317 struct async_submit_ctl submit;
a4456856 4318
6d036f7d 4319 sector_t bn = raid5_compute_blocknr(sh, i, 1);
911d4ee8
N		 4320 sector_t s = raid5_compute_sector(conf, bn, 0,
4321 &dd_idx, NULL);
6d036f7d 4322 sh2 = raid5_get_active_stripe(conf, s, 0, 1, 1);
a4456856
DW		 4323 if (sh2 == NULL)
4324 /* so far only the early blocks of this stripe
4325 * have been requested. When later blocks
4326 * get requested, we will try again
4327 */
4328 continue;
4329 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
4330 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
4331 /* must have already done this block */
6d036f7d 4332 raid5_release_stripe(sh2);
a4456856
DW		 4333 continue;
4334 }
f0a50d37
DW		 4335
4336 /* place all the copies on one channel */
a08abd8c 4337 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
f0a50d37 4338 tx = async_memcpy(sh2->dev[dd_idx].page,
88ba2aa5 4339 sh->dev[i].page, 0, 0, STRIPE_SIZE,
a08abd8c 4340 &submit);
f0a50d37 4341
a4456856
DW		 4342 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
4343 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
4344 for (j = 0; j < conf->raid_disks; j++)
4345 if (j != sh2->pd_idx &&
86c374ba 4346 j != sh2->qd_idx &&
a4456856
DW		 4347 !test_bit(R5_Expanded, &sh2->dev[j].flags))
4348 break;
4349 if (j == conf->raid_disks) {
4350 set_bit(STRIPE_EXPAND_READY, &sh2->state);
4351 set_bit(STRIPE_HANDLE, &sh2->state);
4352 }
6d036f7d 4353 raid5_release_stripe(sh2);
f0a50d37 4354
a4456856 4355 }
a2e08551 4356 /* done submitting copies, wait for them to complete */
749586b7 4357 async_tx_quiesce(&tx);
a4456856 4358}
1da177e4
LT		 4359
4360/*
4361 * handle_stripe - do things to a stripe.
4362 *
9a3e1101
N		 4363 * We lock the stripe by setting STRIPE_ACTIVE and then examine the
4364 * state of various bits to see what needs to be done.
1da177e4 4365 * Possible results:
9a3e1101
N		 4366 * return some read requests which now have data
4367 * return some write requests which are safely on storage
1da177e4
LT		 4368 * schedule a read on some buffers
4369 * schedule a write of some buffers
4370 * return confirmation of parity correctness
4371 *
1da177e4 4372 */
a4456856 4373
acfe726b 4374static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s)
1da177e4 4375{
d1688a6d 4376 struct r5conf *conf = sh->raid_conf;
f416885e 4377 int disks = sh->disks;
474af965
N		 4378 struct r5dev *dev;
4379 int i;
9a3e1101 4380 int do_recovery = 0;
1da177e4 4381
acfe726b
N		 4382 memset(s, 0, sizeof(*s));
4383
dabc4ec6 4384 s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state) && !sh->batch_head;
4385 s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state) && !sh->batch_head;
acfe726b
N		 4386 s->failed_num[0] = -1;
4387 s->failed_num[1] = -1;
6e74a9cf 4388 s->log_failed = r5l_log_disk_error(conf);
1da177e4 4389
acfe726b 4390 /* Now to look around and see what can be done */
1da177e4 4391 rcu_read_lock();
16a53ecc 4392 for (i=disks; i--; ) {
3cb03002 4393 struct md_rdev *rdev;
31c176ec
N		 4394 sector_t first_bad;
4395 int bad_sectors;
4396 int is_bad = 0;
acfe726b 4397
16a53ecc 4398 dev = &sh->dev[i];
1da177e4 4399
45b4233c 4400 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
9a3e1101
N		 4401 i, dev->flags,
4402 dev->toread, dev->towrite, dev->written);
6c0069c0
YT		 4403 /* maybe we can reply to a read
4404 *
4405 * new wantfill requests are only permitted while
4406 * ops_complete_biofill is guaranteed to be inactive
4407 */
4408 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
4409 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
4410 set_bit(R5_Wantfill, &dev->flags);
1da177e4 4411
16a53ecc 4412 /* now count some things */
cc94015a
N		 4413 if (test_bit(R5_LOCKED, &dev->flags))
4414 s->locked++;
4415 if (test_bit(R5_UPTODATE, &dev->flags))
4416 s->uptodate++;
2d6e4ecc 4417 if (test_bit(R5_Wantcompute, &dev->flags)) {
cc94015a
N		 4418 s->compute++;
4419 BUG_ON(s->compute > 2);
2d6e4ecc 4420 }
1da177e4 4421
acfe726b 4422 if (test_bit(R5_Wantfill, &dev->flags))
cc94015a 4423 s->to_fill++;
acfe726b 4424 else if (dev->toread)
cc94015a 4425 s->to_read++;
16a53ecc 4426 if (dev->towrite) {
cc94015a 4427 s->to_write++;
16a53ecc 4428 if (!test_bit(R5_OVERWRITE, &dev->flags))
cc94015a 4429 s->non_overwrite++;
16a53ecc 4430 }
a4456856 4431 if (dev->written)
cc94015a 4432 s->written++;
14a75d3e
N		 4433 /* Prefer to use the replacement for reads, but only
4434 * if it is recovered enough and has no bad blocks.
4435 */
4436 rdev = rcu_dereference(conf->disks[i].replacement);
4437 if (rdev && !test_bit(Faulty, &rdev->flags) &&
4438 rdev->recovery_offset >= sh->sector + STRIPE_SECTORS &&
4439 !is_badblock(rdev, sh->sector, STRIPE_SECTORS,
4440 &first_bad, &bad_sectors))
4441 set_bit(R5_ReadRepl, &dev->flags);
4442 else {
e6030cb0 4443 if (rdev && !test_bit(Faulty, &rdev->flags))
9a3e1101 4444 set_bit(R5_NeedReplace, &dev->flags);
e6030cb0
N		 4445 else
4446 clear_bit(R5_NeedReplace, &dev->flags);
14a75d3e
N		 4447 rdev = rcu_dereference(conf->disks[i].rdev);
4448 clear_bit(R5_ReadRepl, &dev->flags);
4449 }
9283d8c5
N		 4450 if (rdev && test_bit(Faulty, &rdev->flags))
4451 rdev = NULL;
31c176ec
N		 4452 if (rdev) {
4453 is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
4454 &first_bad, &bad_sectors);
4455 if (s->blocked_rdev == NULL
4456 && (test_bit(Blocked, &rdev->flags)
4457 || is_bad < 0)) {
4458 if (is_bad < 0)
4459 set_bit(BlockedBadBlocks,
4460 &rdev->flags);
4461 s->blocked_rdev = rdev;
4462 atomic_inc(&rdev->nr_pending);
4463 }
6bfe0b49 4464 }
415e72d0
N		 4465 clear_bit(R5_Insync, &dev->flags);
4466 if (!rdev)
4467 /* Not in-sync */;
31c176ec
N		 4468 else if (is_bad) {
4469 /* also not in-sync */
18b9837e
N		 4470 if (!test_bit(WriteErrorSeen, &rdev->flags) &&
4471 test_bit(R5_UPTODATE, &dev->flags)) {
31c176ec
N		 4472 /* treat as in-sync, but with a read error
4473 * which we can now try to correct
4474 */
4475 set_bit(R5_Insync, &dev->flags);
4476 set_bit(R5_ReadError, &dev->flags);
4477 }
4478 } else if (test_bit(In_sync, &rdev->flags))
415e72d0 4479 set_bit(R5_Insync, &dev->flags);
30d7a483 4480 else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
415e72d0 4481 /* in sync if before recovery_offset */
30d7a483
N		 4482 set_bit(R5_Insync, &dev->flags);
4483 else if (test_bit(R5_UPTODATE, &dev->flags) &&
4484 test_bit(R5_Expanded, &dev->flags))
4485 /* If we've reshaped into here, we assume it is Insync.
4486 * We will shortly update recovery_offset to make
4487 * it official.
4488 */
4489 set_bit(R5_Insync, &dev->flags);
4490
1cc03eb9 4491 if (test_bit(R5_WriteError, &dev->flags)) {
14a75d3e
N		 4492 /* This flag does not apply to '.replacement'
4493 * only to .rdev, so make sure to check that*/
4494 struct md_rdev *rdev2 = rcu_dereference(
4495 conf->disks[i].rdev);
4496 if (rdev2 == rdev)
4497 clear_bit(R5_Insync, &dev->flags);
4498 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
bc2607f3 4499 s->handle_bad_blocks = 1;
14a75d3e 4500 atomic_inc(&rdev2->nr_pending);
bc2607f3
N		 4501 } else
4502 clear_bit(R5_WriteError, &dev->flags);
4503 }
1cc03eb9 4504 if (test_bit(R5_MadeGood, &dev->flags)) {
14a75d3e
N		 4505 /* This flag does not apply to '.replacement'
4506 * only to .rdev, so make sure to check that*/
4507 struct md_rdev *rdev2 = rcu_dereference(
4508 conf->disks[i].rdev);
4509 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
b84db560 4510 s->handle_bad_blocks = 1;
14a75d3e 4511 atomic_inc(&rdev2->nr_pending);
b84db560
N		 4512 } else
4513 clear_bit(R5_MadeGood, &dev->flags);
4514 }
977df362
N		 4515 if (test_bit(R5_MadeGoodRepl, &dev->flags)) {
4516 struct md_rdev *rdev2 = rcu_dereference(
4517 conf->disks[i].replacement);
4518 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
4519 s->handle_bad_blocks = 1;
4520 atomic_inc(&rdev2->nr_pending);
4521 } else
4522 clear_bit(R5_MadeGoodRepl, &dev->flags);
4523 }
415e72d0 4524 if (!test_bit(R5_Insync, &dev->flags)) {
16a53ecc
N		 4525 /* The ReadError flag will just be confusing now */
4526 clear_bit(R5_ReadError, &dev->flags);
4527 clear_bit(R5_ReWrite, &dev->flags);
1da177e4 4528 }
415e72d0
N		 4529 if (test_bit(R5_ReadError, &dev->flags))
4530 clear_bit(R5_Insync, &dev->flags);
4531 if (!test_bit(R5_Insync, &dev->flags)) {
cc94015a
N		 4532 if (s->failed < 2)
4533 s->failed_num[s->failed] = i;
4534 s->failed++;
9a3e1101
N		 4535 if (rdev && !test_bit(Faulty, &rdev->flags))
4536 do_recovery = 1;
d63e2fc8
BC		 4537 else if (!rdev) {
4538 rdev = rcu_dereference(
4539 conf->disks[i].replacement);
4540 if (rdev && !test_bit(Faulty, &rdev->flags))
4541 do_recovery = 1;
4542 }
415e72d0 4543 }
2ded3703
SL		 4544
4545 if (test_bit(R5_InJournal, &dev->flags))
4546 s->injournal++;
1e6d690b
SL		 4547 if (test_bit(R5_InJournal, &dev->flags) && dev->written)
4548 s->just_cached++;
1da177e4 4549 }
9a3e1101
N		 4550 if (test_bit(STRIPE_SYNCING, &sh->state)) {
4551 /* If there is a failed device being replaced,
4552 * we must be recovering.
4553 * else if we are after recovery_cp, we must be syncing
c6d2e084 4554 * else if MD_RECOVERY_REQUESTED is set, we also are syncing.
9a3e1101
N		 4555 * else we can only be replacing
4556 * sync and recovery both need to read all devices, and so
4557 * use the same flag.
4558 */
4559 if (do_recovery ||
c6d2e084 4560 sh->sector >= conf->mddev->recovery_cp ||
4561 test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery)))
9a3e1101
N		 4562 s->syncing = 1;
4563 else
4564 s->replacing = 1;
4565 }
1da177e4 4566 rcu_read_unlock();
cc94015a
N		 4567}
4568
59fc630b 4569static int clear_batch_ready(struct stripe_head *sh)
4570{
b15a9dbd
N		 4571 /* Return '1' if this is a member of batch, or
4572 * '0' if it is a lone stripe or a head which can now be
4573 * handled.
4574 */
59fc630b 4575 struct stripe_head *tmp;
4576 if (!test_and_clear_bit(STRIPE_BATCH_READY, &sh->state))
b15a9dbd 4577 return (sh->batch_head && sh->batch_head != sh);
59fc630b 4578 spin_lock(&sh->stripe_lock);
4579 if (!sh->batch_head) {
4580 spin_unlock(&sh->stripe_lock);
4581 return 0;
4582 }
4583
4584 /*
4585 * this stripe could be added to a batch list before we check
4586 * BATCH_READY, skips it
4587 */
4588 if (sh->batch_head != sh) {
4589 spin_unlock(&sh->stripe_lock);
4590 return 1;
4591 }
4592 spin_lock(&sh->batch_lock);
4593 list_for_each_entry(tmp, &sh->batch_list, batch_list)
4594 clear_bit(STRIPE_BATCH_READY, &tmp->state);
4595 spin_unlock(&sh->batch_lock);
4596 spin_unlock(&sh->stripe_lock);
4597
4598 /*
4599 * BATCH_READY is cleared, no new stripes can be added.
4600 * batch_list can be accessed without lock
4601 */
4602 return 0;
4603}
4604
3960ce79
N		 4605static void break_stripe_batch_list(struct stripe_head *head_sh,
4606 unsigned long handle_flags)
72ac7330 4607{
4e3d62ff 4608 struct stripe_head *sh, *next;
72ac7330 4609 int i;
fb642b92 4610 int do_wakeup = 0;
72ac7330 4611
bb27051f
N		 4612 list_for_each_entry_safe(sh, next, &head_sh->batch_list, batch_list) {
4613
72ac7330 4614 list_del_init(&sh->batch_list);
4615
fb3229d5 4616 WARN_ONCE(sh->state & ((1 << STRIPE_ACTIVE) |
1b956f7a
N		 4617 (1 << STRIPE_SYNCING) |
4618 (1 << STRIPE_REPLACED) |
1b956f7a
N		 4619 (1 << STRIPE_DELAYED) |
4620 (1 << STRIPE_BIT_DELAY) |
4621 (1 << STRIPE_FULL_WRITE) |
4622 (1 << STRIPE_BIOFILL_RUN) |
4623 (1 << STRIPE_COMPUTE_RUN) |
1b956f7a
N		 4624 (1 << STRIPE_DISCARD) |
4625 (1 << STRIPE_BATCH_READY) |
4626 (1 << STRIPE_BATCH_ERR) |
fb3229d5
SL		 4627 (1 << STRIPE_BITMAP_PENDING)),
4628 "stripe state: %lx\n", sh->state);
4629 WARN_ONCE(head_sh->state & ((1 << STRIPE_DISCARD) |
4630 (1 << STRIPE_REPLACED)),
4631 "head stripe state: %lx\n", head_sh->state);
1b956f7a
N		 4632
4633 set_mask_bits(&sh->state, ~(STRIPE_EXPAND_SYNC_FLAGS |
550da24f 4634 (1 << STRIPE_PREREAD_ACTIVE) |
184a09eb
DY		 4635 (1 << STRIPE_DEGRADED) |
4636 (1 << STRIPE_ON_UNPLUG_LIST)),
1b956f7a
N		 4637 head_sh->state & (1 << STRIPE_INSYNC));
4638
72ac7330 4639 sh->check_state = head_sh->check_state;
4640 sh->reconstruct_state = head_sh->reconstruct_state;
448ec638
AC		 4641 spin_lock_irq(&sh->stripe_lock);
4642 sh->batch_head = NULL;
4643 spin_unlock_irq(&sh->stripe_lock);
fb642b92
N		 4644 for (i = 0; i < sh->disks; i++) {
4645 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
4646 do_wakeup = 1;
72ac7330 4647 sh->dev[i].flags = head_sh->dev[i].flags &
4648 (~((1 << R5_WriteError) | (1 << R5_Overlap)));
fb642b92 4649 }
3960ce79
N		 4650 if (handle_flags == 0 ||
4651 sh->state & handle_flags)
4652 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 4653 raid5_release_stripe(sh);
72ac7330 4654 }
fb642b92
N		 4655 spin_lock_irq(&head_sh->stripe_lock);
4656 head_sh->batch_head = NULL;
4657 spin_unlock_irq(&head_sh->stripe_lock);
4658 for (i = 0; i < head_sh->disks; i++)
4659 if (test_and_clear_bit(R5_Overlap, &head_sh->dev[i].flags))
4660 do_wakeup = 1;
3960ce79
N		 4661 if (head_sh->state & handle_flags)
4662 set_bit(STRIPE_HANDLE, &head_sh->state);
fb642b92
N		 4663
4664 if (do_wakeup)
4665 wake_up(&head_sh->raid_conf->wait_for_overlap);
72ac7330 4666}
4667
cc94015a
N		 4668static void handle_stripe(struct stripe_head *sh)
4669{
4670 struct stripe_head_state s;
d1688a6d 4671 struct r5conf *conf = sh->raid_conf;
3687c061 4672 int i;
84789554
N		 4673 int prexor;
4674 int disks = sh->disks;
474af965 4675 struct r5dev *pdev, *qdev;
cc94015a
N		 4676
4677 clear_bit(STRIPE_HANDLE, &sh->state);
257a4b42 4678 if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) {
cc94015a
N		 4679 /* already being handled, ensure it gets handled
4680 * again when current action finishes */
4681 set_bit(STRIPE_HANDLE, &sh->state);
4682 return;
4683 }
4684
59fc630b 4685 if (clear_batch_ready(sh) ) {
4686 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
4687 return;
4688 }
4689
4e3d62ff 4690 if (test_and_clear_bit(STRIPE_BATCH_ERR, &sh->state))
3960ce79 4691 break_stripe_batch_list(sh, 0);
72ac7330 4692
dabc4ec6 4693 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) && !sh->batch_head) {
f8dfcffd 4694 spin_lock(&sh->stripe_lock);
5ddf0440
SL		 4695 /*
4696 * Cannot process 'sync' concurrently with 'discard'.
4697 * Flush data in r5cache before 'sync'.
4698 */
4699 if (!test_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state) &&
4700 !test_bit(STRIPE_R5C_FULL_STRIPE, &sh->state) &&
4701 !test_bit(STRIPE_DISCARD, &sh->state) &&
f8dfcffd
N		 4702 test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
4703 set_bit(STRIPE_SYNCING, &sh->state);
4704 clear_bit(STRIPE_INSYNC, &sh->state);
f94c0b66 4705 clear_bit(STRIPE_REPLACED, &sh->state);
f8dfcffd
N		 4706 }
4707 spin_unlock(&sh->stripe_lock);
cc94015a
N		 4708 }
4709 clear_bit(STRIPE_DELAYED, &sh->state);
4710
4711 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
4712 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
4713 (unsigned long long)sh->sector, sh->state,
4714 atomic_read(&sh->count), sh->pd_idx, sh->qd_idx,
4715 sh->check_state, sh->reconstruct_state);
3687c061 4716
acfe726b 4717 analyse_stripe(sh, &s);
c5a31000 4718
b70abcb2
SL		 4719 if (test_bit(STRIPE_LOG_TRAPPED, &sh->state))
4720 goto finish;
4721
16d997b7
N		 4722 if (s.handle_bad_blocks ||
4723 test_bit(MD_SB_CHANGE_PENDING, &conf->mddev->sb_flags)) {
bc2607f3
N		 4724 set_bit(STRIPE_HANDLE, &sh->state);
4725 goto finish;
4726 }
4727
474af965
N		 4728 if (unlikely(s.blocked_rdev)) {
4729 if (s.syncing || s.expanding || s.expanded ||
9a3e1101 4730 s.replacing || s.to_write || s.written) {
474af965
N		 4731 set_bit(STRIPE_HANDLE, &sh->state);
4732 goto finish;
4733 }
4734 /* There is nothing for the blocked_rdev to block */
4735 rdev_dec_pending(s.blocked_rdev, conf->mddev);
4736 s.blocked_rdev = NULL;
4737 }
4738
4739 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
4740 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
4741 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
4742 }
4743
4744 pr_debug("locked=%d uptodate=%d to_read=%d"
4745 " to_write=%d failed=%d failed_num=%d,%d\n",
4746 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
4747 s.failed_num[0], s.failed_num[1]);
70d466f7
SL		 4748 /*
4749 * check if the array has lost more than max_degraded devices and,
474af965 4750 * if so, some requests might need to be failed.
70d466f7
SL		 4751 *
4752 * When journal device failed (log_failed), we will only process
4753 * the stripe if there is data need write to raid disks
474af965 4754 */
70d466f7
SL		 4755 if (s.failed > conf->max_degraded ||
4756 (s.log_failed && s.injournal == 0)) {
9a3f530f
N		 4757 sh->check_state = 0;
4758 sh->reconstruct_state = 0;
626f2092 4759 break_stripe_batch_list(sh, 0);
9a3f530f 4760 if (s.to_read+s.to_write+s.written)
bd83d0a2 4761 handle_failed_stripe(conf, sh, &s, disks);
9a3e1101 4762 if (s.syncing + s.replacing)
9a3f530f
N		 4763 handle_failed_sync(conf, sh, &s);
4764 }
474af965 4765
84789554
N		 4766 /* Now we check to see if any write operations have recently
4767 * completed
4768 */
4769 prexor = 0;
4770 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
4771 prexor = 1;
4772 if (sh->reconstruct_state == reconstruct_state_drain_result ||
4773 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
4774 sh->reconstruct_state = reconstruct_state_idle;
4775
4776 /* All the 'written' buffers and the parity block are ready to
4777 * be written back to disk
4778 */
9e444768
SL		 4779 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) &&
4780 !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags));
84789554 4781 BUG_ON(sh->qd_idx >= 0 &&
9e444768
SL		 4782 !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) &&
4783 !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags));
84789554
N		 4784 for (i = disks; i--; ) {
4785 struct r5dev *dev = &sh->dev[i];
4786 if (test_bit(R5_LOCKED, &dev->flags) &&
4787 (i == sh->pd_idx || i == sh->qd_idx ||
1e6d690b
SL		 4788 dev->written || test_bit(R5_InJournal,
4789 &dev->flags))) {
84789554
N		 4790 pr_debug("Writing block %d\n", i);
4791 set_bit(R5_Wantwrite, &dev->flags);
4792 if (prexor)
4793 continue;
9c4bdf69
N		 4794 if (s.failed > 1)
4795 continue;
84789554
N		 4796 if (!test_bit(R5_Insync, &dev->flags) ||
4797 ((i == sh->pd_idx || i == sh->qd_idx) &&
4798 s.failed == 0))
4799 set_bit(STRIPE_INSYNC, &sh->state);
4800 }
4801 }
4802 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4803 s.dec_preread_active = 1;
4804 }
4805
ef5b7c69
N		 4806 /*
4807 * might be able to return some write requests if the parity blocks
4808 * are safe, or on a failed drive
4809 */
4810 pdev = &sh->dev[sh->pd_idx];
4811 s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx)
4812 || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx);
4813 qdev = &sh->dev[sh->qd_idx];
4814 s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx)
4815 || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx)
4816 || conf->level < 6;
4817
4818 if (s.written &&
4819 (s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
4820 && !test_bit(R5_LOCKED, &pdev->flags)
4821 && (test_bit(R5_UPTODATE, &pdev->flags) ||
4822 test_bit(R5_Discard, &pdev->flags))))) &&
4823 (s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
4824 && !test_bit(R5_LOCKED, &qdev->flags)
4825 && (test_bit(R5_UPTODATE, &qdev->flags) ||
4826 test_bit(R5_Discard, &qdev->flags))))))
bd83d0a2 4827 handle_stripe_clean_event(conf, sh, disks);
ef5b7c69 4828
1e6d690b 4829 if (s.just_cached)
bd83d0a2 4830 r5c_handle_cached_data_endio(conf, sh, disks);
ff875738 4831 log_stripe_write_finished(sh);
1e6d690b 4832
ef5b7c69
N		 4833 /* Now we might consider reading some blocks, either to check/generate
4834 * parity, or to satisfy requests
4835 * or to load a block that is being partially written.
4836 */
4837 if (s.to_read || s.non_overwrite
4838 || (conf->level == 6 && s.to_write && s.failed)
4839 || (s.syncing && (s.uptodate + s.compute < disks))
4840 || s.replacing
4841 || s.expanding)
4842 handle_stripe_fill(sh, &s, disks);
4843
2ded3703
SL		 4844 /*
4845 * When the stripe finishes full journal write cycle (write to journal
4846 * and raid disk), this is the clean up procedure so it is ready for
4847 * next operation.
4848 */
4849 r5c_finish_stripe_write_out(conf, sh, &s);
4850
4851 /*
4852 * Now to consider new write requests, cache write back and what else,
4853 * if anything should be read. We do not handle new writes when:
84789554
N		 4854 * 1/ A 'write' operation (copy+xor) is already in flight.
4855 * 2/ A 'check' operation is in flight, as it may clobber the parity
4856 * block.
2ded3703 4857 * 3/ A r5c cache log write is in flight.
84789554 4858 */
2ded3703
SL		 4859
4860 if (!sh->reconstruct_state && !sh->check_state && !sh->log_io) {
4861 if (!r5c_is_writeback(conf->log)) {
4862 if (s.to_write)
4863 handle_stripe_dirtying(conf, sh, &s, disks);
4864 } else { /* write back cache */
4865 int ret = 0;
4866
4867 /* First, try handle writes in caching phase */
4868 if (s.to_write)
4869 ret = r5c_try_caching_write(conf, sh, &s,
4870 disks);
4871 /*
4872 * If caching phase failed: ret == -EAGAIN
4873 * OR
4874 * stripe under reclaim: !caching && injournal
4875 *
4876 * fall back to handle_stripe_dirtying()
4877 */
4878 if (ret == -EAGAIN ||
4879 /* stripe under reclaim: !caching && injournal */
4880 (!test_bit(STRIPE_R5C_CACHING, &sh->state) &&
d7bd398e
SL		 4881 s.injournal > 0)) {
4882 ret = handle_stripe_dirtying(conf, sh, &s,
4883 disks);
4884 if (ret == -EAGAIN)
4885 goto finish;
4886 }
2ded3703
SL		 4887 }
4888 }
84789554
N		 4889
4890 /* maybe we need to check and possibly fix the parity for this stripe
4891 * Any reads will already have been scheduled, so we just see if enough
4892 * data is available. The parity check is held off while parity
4893 * dependent operations are in flight.
4894 */
4895 if (sh->check_state ||
4896 (s.syncing && s.locked == 0 &&
4897 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
4898 !test_bit(STRIPE_INSYNC, &sh->state))) {
4899 if (conf->level == 6)
4900 handle_parity_checks6(conf, sh, &s, disks);
4901 else
4902 handle_parity_checks5(conf, sh, &s, disks);
4903 }
c5a31000 4904
f94c0b66
N		 4905 if ((s.replacing || s.syncing) && s.locked == 0
4906 && !test_bit(STRIPE_COMPUTE_RUN, &sh->state)
4907 && !test_bit(STRIPE_REPLACED, &sh->state)) {
9a3e1101
N		 4908 /* Write out to replacement devices where possible */
4909 for (i = 0; i < conf->raid_disks; i++)
f94c0b66
N		 4910 if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) {
4911 WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags));
9a3e1101
N		 4912 set_bit(R5_WantReplace, &sh->dev[i].flags);
4913 set_bit(R5_LOCKED, &sh->dev[i].flags);
4914 s.locked++;
4915 }
f94c0b66
N		 4916 if (s.replacing)
4917 set_bit(STRIPE_INSYNC, &sh->state);
4918 set_bit(STRIPE_REPLACED, &sh->state);
9a3e1101
N		 4919 }
4920 if ((s.syncing || s.replacing) && s.locked == 0 &&
f94c0b66 4921 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
9a3e1101 4922 test_bit(STRIPE_INSYNC, &sh->state)) {
c5a31000
N		 4923 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
4924 clear_bit(STRIPE_SYNCING, &sh->state);
f8dfcffd
N		 4925 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
4926 wake_up(&conf->wait_for_overlap);
c5a31000
N		 4927 }
4928
4929 /* If the failed drives are just a ReadError, then we might need
4930 * to progress the repair/check process
4931 */
4932 if (s.failed <= conf->max_degraded && !conf->mddev->ro)
4933 for (i = 0; i < s.failed; i++) {
4934 struct r5dev *dev = &sh->dev[s.failed_num[i]];
4935 if (test_bit(R5_ReadError, &dev->flags)
4936 && !test_bit(R5_LOCKED, &dev->flags)
4937 && test_bit(R5_UPTODATE, &dev->flags)
4938 ) {
4939 if (!test_bit(R5_ReWrite, &dev->flags)) {
4940 set_bit(R5_Wantwrite, &dev->flags);
4941 set_bit(R5_ReWrite, &dev->flags);
4942 set_bit(R5_LOCKED, &dev->flags);
4943 s.locked++;
4944 } else {
4945 /* let's read it back */
4946 set_bit(R5_Wantread, &dev->flags);
4947 set_bit(R5_LOCKED, &dev->flags);
4948 s.locked++;
4949 }
4950 }
4951 }
4952
3687c061
N		 4953 /* Finish reconstruct operations initiated by the expansion process */
4954 if (sh->reconstruct_state == reconstruct_state_result) {
4955 struct stripe_head *sh_src
6d036f7d 4956 = raid5_get_active_stripe(conf, sh->sector, 1, 1, 1);
3687c061
N		 4957 if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) {
4958 /* sh cannot be written until sh_src has been read.
4959 * so arrange for sh to be delayed a little
4960 */
4961 set_bit(STRIPE_DELAYED, &sh->state);
4962 set_bit(STRIPE_HANDLE, &sh->state);
4963 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
4964 &sh_src->state))
4965 atomic_inc(&conf->preread_active_stripes);
6d036f7d 4966 raid5_release_stripe(sh_src);
3687c061
N		 4967 goto finish;
4968 }
4969 if (sh_src)
6d036f7d 4970 raid5_release_stripe(sh_src);
3687c061
N		 4971
4972 sh->reconstruct_state = reconstruct_state_idle;
4973 clear_bit(STRIPE_EXPANDING, &sh->state);
4974 for (i = conf->raid_disks; i--; ) {
4975 set_bit(R5_Wantwrite, &sh->dev[i].flags);
4976 set_bit(R5_LOCKED, &sh->dev[i].flags);
4977 s.locked++;
4978 }
4979 }
f416885e 4980
3687c061
N		 4981 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
4982 !sh->reconstruct_state) {
4983 /* Need to write out all blocks after computing parity */
4984 sh->disks = conf->raid_disks;
4985 stripe_set_idx(sh->sector, conf, 0, sh);
4986 schedule_reconstruction(sh, &s, 1, 1);
4987 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
4988 clear_bit(STRIPE_EXPAND_READY, &sh->state);
4989 atomic_dec(&conf->reshape_stripes);
4990 wake_up(&conf->wait_for_overlap);
4991 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
4992 }
4993
4994 if (s.expanding && s.locked == 0 &&
4995 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
4996 handle_stripe_expansion(conf, sh);
16a53ecc 4997
3687c061 4998finish:
6bfe0b49 4999 /* wait for this device to become unblocked */
5f066c63
N		 5000 if (unlikely(s.blocked_rdev)) {
5001 if (conf->mddev->external)
5002 md_wait_for_blocked_rdev(s.blocked_rdev,
5003 conf->mddev);
5004 else
5005 /* Internal metadata will immediately
5006 * be written by raid5d, so we don't
5007 * need to wait here.
5008 */
5009 rdev_dec_pending(s.blocked_rdev,
5010 conf->mddev);
5011 }
6bfe0b49 5012
bc2607f3
N		 5013 if (s.handle_bad_blocks)
5014 for (i = disks; i--; ) {
3cb03002 5015 struct md_rdev *rdev;
bc2607f3
N		 5016 struct r5dev *dev = &sh->dev[i];
5017 if (test_and_clear_bit(R5_WriteError, &dev->flags)) {
5018 /* We own a safe reference to the rdev */
5019 rdev = conf->disks[i].rdev;
5020 if (!rdev_set_badblocks(rdev, sh->sector,
5021 STRIPE_SECTORS, 0))
5022 md_error(conf->mddev, rdev);
5023 rdev_dec_pending(rdev, conf->mddev);
5024 }
b84db560
N		 5025 if (test_and_clear_bit(R5_MadeGood, &dev->flags)) {
5026 rdev = conf->disks[i].rdev;
5027 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 5028 STRIPE_SECTORS, 0);
b84db560
N		 5029 rdev_dec_pending(rdev, conf->mddev);
5030 }
977df362
N		 5031 if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) {
5032 rdev = conf->disks[i].replacement;
dd054fce
N		 5033 if (!rdev)
5034 /* rdev have been moved down */
5035 rdev = conf->disks[i].rdev;
977df362 5036 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 5037 STRIPE_SECTORS, 0);
977df362
N		 5038 rdev_dec_pending(rdev, conf->mddev);
5039 }
bc2607f3
N		 5040 }
5041
6c0069c0
YT		 5042 if (s.ops_request)
5043 raid_run_ops(sh, s.ops_request);
5044
f0e43bcd 5045 ops_run_io(sh, &s);
16a53ecc 5046
c5709ef6 5047 if (s.dec_preread_active) {
729a1866 5048 /* We delay this until after ops_run_io so that if make_request
e9c7469b 5049 * is waiting on a flush, it won't continue until the writes
729a1866
N		 5050 * have actually been submitted.
5051 */
5052 atomic_dec(&conf->preread_active_stripes);
5053 if (atomic_read(&conf->preread_active_stripes) <
5054 IO_THRESHOLD)
5055 md_wakeup_thread(conf->mddev->thread);
5056 }
5057
257a4b42 5058 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
16a53ecc
N		 5059}
5060
d1688a6d 5061static void raid5_activate_delayed(struct r5conf *conf)
16a53ecc
N		 5062{
5063 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
5064 while (!list_empty(&conf->delayed_list)) {
5065 struct list_head *l = conf->delayed_list.next;
5066 struct stripe_head *sh;
5067 sh = list_entry(l, struct stripe_head, lru);
5068 list_del_init(l);
5069 clear_bit(STRIPE_DELAYED, &sh->state);
5070 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
5071 atomic_inc(&conf->preread_active_stripes);
8b3e6cdc 5072 list_add_tail(&sh->lru, &conf->hold_list);
851c30c9 5073 raid5_wakeup_stripe_thread(sh);
16a53ecc 5074 }
482c0834 5075 }
16a53ecc
N		 5076}
5077
566c09c5
SL		 5078static void activate_bit_delay(struct r5conf *conf,
5079 struct list_head *temp_inactive_list)
16a53ecc
N		 5080{
5081 /* device_lock is held */
5082 struct list_head head;
5083 list_add(&head, &conf->bitmap_list);
5084 list_del_init(&conf->bitmap_list);
5085 while (!list_empty(&head)) {
5086 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
566c09c5 5087 int hash;
16a53ecc
N		 5088 list_del_init(&sh->lru);
5089 atomic_inc(&sh->count);
566c09c5
SL		 5090 hash = sh->hash_lock_index;
5091 __release_stripe(conf, sh, &temp_inactive_list[hash]);
16a53ecc
N		 5092 }
5093}
5094
5c675f83 5095static int raid5_congested(struct mddev *mddev, int bits)
f022b2fd 5096{
d1688a6d 5097 struct r5conf *conf = mddev->private;
f022b2fd
N		 5098
5099 /* No difference between reads and writes. Just check
5100 * how busy the stripe_cache is
5101 */
3fa841d7 5102
5423399a 5103 if (test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state))
f022b2fd 5104 return 1;
a39f7afd
SL		 5105
5106 /* Also checks whether there is pressure on r5cache log space */
5107 if (test_bit(R5C_LOG_TIGHT, &conf->cache_state))
5108 return 1;
f022b2fd
N		 5109 if (conf->quiesce)
5110 return 1;
4bda556a 5111 if (atomic_read(&conf->empty_inactive_list_nr))
f022b2fd
N		 5112 return 1;
5113
5114 return 0;
5115}
5116
fd01b88c 5117static int in_chunk_boundary(struct mddev *mddev, struct bio *bio)
f679623f 5118{
3cb5edf4 5119 struct r5conf *conf = mddev->private;
10433d04 5120 sector_t sector = bio->bi_iter.bi_sector;
3cb5edf4 5121 unsigned int chunk_sectors;
aa8b57aa 5122 unsigned int bio_sectors = bio_sectors(bio);
f679623f 5123
10433d04
CH		 5124 WARN_ON_ONCE(bio->bi_partno);
5125
3cb5edf4 5126 chunk_sectors = min(conf->chunk_sectors, conf->prev_chunk_sectors);
f679623f
RBJ		 5127 return chunk_sectors >=
5128 ((sector & (chunk_sectors - 1)) + bio_sectors);
5129}
5130
46031f9a
RBJ		 5131/*
5132 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
5133 * later sampled by raid5d.
5134 */
d1688a6d 5135static void add_bio_to_retry(struct bio *bi,struct r5conf *conf)
46031f9a
RBJ		 5136{
5137 unsigned long flags;
5138
5139 spin_lock_irqsave(&conf->device_lock, flags);
5140
5141 bi->bi_next = conf->retry_read_aligned_list;
5142 conf->retry_read_aligned_list = bi;
5143
5144 spin_unlock_irqrestore(&conf->device_lock, flags);
5145 md_wakeup_thread(conf->mddev->thread);
5146}
5147
0472a42b
N		 5148static struct bio *remove_bio_from_retry(struct r5conf *conf,
5149 unsigned int *offset)
46031f9a
RBJ		 5150{
5151 struct bio *bi;
5152
5153 bi = conf->retry_read_aligned;
5154 if (bi) {
0472a42b 5155 *offset = conf->retry_read_offset;
46031f9a
RBJ		 5156 conf->retry_read_aligned = NULL;
5157 return bi;
5158 }
5159 bi = conf->retry_read_aligned_list;
5160 if(bi) {
387bb173 5161 conf->retry_read_aligned_list = bi->bi_next;
46031f9a 5162 bi->bi_next = NULL;
0472a42b 5163 *offset = 0;
46031f9a
RBJ		 5164 }
5165
5166 return bi;
5167}
5168
f679623f
RBJ		 5169/*
5170 * The "raid5_align_endio" should check if the read succeeded and if it
5171 * did, call bio_endio on the original bio (having bio_put the new bio
5172 * first).
5173 * If the read failed..
5174 */
4246a0b6 5175static void raid5_align_endio(struct bio *bi)
f679623f
RBJ		 5176{
5177 struct bio* raid_bi = bi->bi_private;
fd01b88c 5178 struct mddev *mddev;
d1688a6d 5179 struct r5conf *conf;
3cb03002 5180 struct md_rdev *rdev;
4e4cbee9 5181 blk_status_t error = bi->bi_status;
46031f9a 5182
f679623f 5183 bio_put(bi);
46031f9a 5184
46031f9a
RBJ		 5185 rdev = (void*)raid_bi->bi_next;
5186 raid_bi->bi_next = NULL;
2b7f2228
N		 5187 mddev = rdev->mddev;
5188 conf = mddev->private;
46031f9a
RBJ		 5189
5190 rdev_dec_pending(rdev, conf->mddev);
5191
9b81c842 5192 if (!error) {
4246a0b6 5193 bio_endio(raid_bi);
46031f9a 5194 if (atomic_dec_and_test(&conf->active_aligned_reads))
b1b46486 5195 wake_up(&conf->wait_for_quiescent);
6712ecf8 5196 return;
46031f9a
RBJ		 5197 }
5198
45b4233c 5199 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
46031f9a
RBJ		 5200
5201 add_bio_to_retry(raid_bi, conf);
f679623f
RBJ		 5202}
5203
7ef6b12a 5204static int raid5_read_one_chunk(struct mddev *mddev, struct bio *raid_bio)
f679623f 5205{
d1688a6d 5206 struct r5conf *conf = mddev->private;
8553fe7e 5207 int dd_idx;
f679623f 5208 struct bio* align_bi;
3cb03002 5209 struct md_rdev *rdev;
671488cc 5210 sector_t end_sector;
f679623f
RBJ		 5211
5212 if (!in_chunk_boundary(mddev, raid_bio)) {
7ef6b12a 5213 pr_debug("%s: non aligned\n", __func__);
f679623f
RBJ		 5214 return 0;
5215 }
5216 /*
d7a10308 5217 * use bio_clone_fast to make a copy of the bio
f679623f 5218 */
afeee514 5219 align_bi = bio_clone_fast(raid_bio, GFP_NOIO, &mddev->bio_set);
f679623f
RBJ		 5220 if (!align_bi)
5221 return 0;
5222 /*
5223 * set bi_end_io to a new function, and set bi_private to the
5224 * original bio.
5225 */
5226 align_bi->bi_end_io = raid5_align_endio;
5227 align_bi->bi_private = raid_bio;
5228 /*
5229 * compute position
5230 */
4f024f37
KO		 5231 align_bi->bi_iter.bi_sector =
5232 raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector,
5233 0, &dd_idx, NULL);
f679623f 5234
f73a1c7d 5235 end_sector = bio_end_sector(align_bi);
f679623f 5236 rcu_read_lock();
671488cc
N		 5237 rdev = rcu_dereference(conf->disks[dd_idx].replacement);
5238 if (!rdev || test_bit(Faulty, &rdev->flags) ||
5239 rdev->recovery_offset < end_sector) {
5240 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
5241 if (rdev &&
5242 (test_bit(Faulty, &rdev->flags) ||
5243 !(test_bit(In_sync, &rdev->flags) ||
5244 rdev->recovery_offset >= end_sector)))
5245 rdev = NULL;
5246 }
03b047f4
SL		 5247
5248 if (r5c_big_stripe_cached(conf, align_bi->bi_iter.bi_sector)) {
5249 rcu_read_unlock();
5250 bio_put(align_bi);
5251 return 0;
5252 }
5253
671488cc 5254 if (rdev) {
31c176ec
N		 5255 sector_t first_bad;
5256 int bad_sectors;
5257
f679623f
RBJ		 5258 atomic_inc(&rdev->nr_pending);
5259 rcu_read_unlock();
46031f9a 5260 raid_bio->bi_next = (void*)rdev;
74d46992 5261 bio_set_dev(align_bi, rdev->bdev);
46031f9a 5262
7140aafc 5263 if (is_badblock(rdev, align_bi->bi_iter.bi_sector,
4f024f37 5264 bio_sectors(align_bi),
31c176ec 5265 &first_bad, &bad_sectors)) {
387bb173
NB		 5266 bio_put(align_bi);
5267 rdev_dec_pending(rdev, mddev);
5268 return 0;
5269 }
5270
6c0544e2 5271 /* No reshape active, so we can trust rdev->data_offset */
4f024f37 5272 align_bi->bi_iter.bi_sector += rdev->data_offset;
6c0544e2 5273
46031f9a 5274 spin_lock_irq(&conf->device_lock);
b1b46486 5275 wait_event_lock_irq(conf->wait_for_quiescent,
46031f9a 5276 conf->quiesce == 0,
eed8c02e 5277 conf->device_lock);
46031f9a
RBJ		 5278 atomic_inc(&conf->active_aligned_reads);
5279 spin_unlock_irq(&conf->device_lock);
5280
e3620a3a 5281 if (mddev->gendisk)
74d46992 5282 trace_block_bio_remap(align_bi->bi_disk->queue,
e3620a3a 5283 align_bi, disk_devt(mddev->gendisk),
4f024f37 5284 raid_bio->bi_iter.bi_sector);
f679623f
RBJ		 5285 generic_make_request(align_bi);
5286 return 1;
5287 } else {
5288 rcu_read_unlock();
46031f9a 5289 bio_put(align_bi);
f679623f
RBJ		 5290 return 0;
5291 }
5292}
5293
7ef6b12a
ML		 5294static struct bio *chunk_aligned_read(struct mddev *mddev, struct bio *raid_bio)
5295{
5296 struct bio *split;
dd7a8f5d
N		 5297 sector_t sector = raid_bio->bi_iter.bi_sector;
5298 unsigned chunk_sects = mddev->chunk_sectors;
5299 unsigned sectors = chunk_sects - (sector & (chunk_sects-1));
7ef6b12a 5300
dd7a8f5d
N		 5301 if (sectors < bio_sectors(raid_bio)) {
5302 struct r5conf *conf = mddev->private;
afeee514 5303 split = bio_split(raid_bio, sectors, GFP_NOIO, &conf->bio_split);
dd7a8f5d
N		 5304 bio_chain(split, raid_bio);
5305 generic_make_request(raid_bio);
5306 raid_bio = split;
5307 }
7ef6b12a 5308
dd7a8f5d
N		 5309 if (!raid5_read_one_chunk(mddev, raid_bio))
5310 return raid_bio;
7ef6b12a
ML		 5311
5312 return NULL;
5313}
5314
8b3e6cdc
DW		 5315/* __get_priority_stripe - get the next stripe to process
5316 *
5317 * Full stripe writes are allowed to pass preread active stripes up until
5318 * the bypass_threshold is exceeded. In general the bypass_count
5319 * increments when the handle_list is handled before the hold_list; however, it
5320 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
5321 * stripe with in flight i/o. The bypass_count will be reset when the
5322 * head of the hold_list has changed, i.e. the head was promoted to the
5323 * handle_list.
5324 */
851c30c9 5325static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group)
8b3e6cdc 5326{
535ae4eb 5327 struct stripe_head *sh, *tmp;
851c30c9 5328 struct list_head *handle_list = NULL;
535ae4eb 5329 struct r5worker_group *wg;
70d466f7
SL		 5330 bool second_try = !r5c_is_writeback(conf->log) &&
5331 !r5l_log_disk_error(conf);
5332 bool try_loprio = test_bit(R5C_LOG_TIGHT, &conf->cache_state) ||
5333 r5l_log_disk_error(conf);
851c30c9 5334
535ae4eb
SL		 5335again:
5336 wg = NULL;
5337 sh = NULL;
851c30c9 5338 if (conf->worker_cnt_per_group == 0) {
535ae4eb
SL		 5339 handle_list = try_loprio ? &conf->loprio_list :
5340 &conf->handle_list;
851c30c9 5341 } else if (group != ANY_GROUP) {
535ae4eb
SL		 5342 handle_list = try_loprio ? &conf->worker_groups[group].loprio_list :
5343 &conf->worker_groups[group].handle_list;
bfc90cb0 5344 wg = &conf->worker_groups[group];
851c30c9
SL		 5345 } else {
5346 int i;
5347 for (i = 0; i < conf->group_cnt; i++) {
535ae4eb
SL		 5348 handle_list = try_loprio ? &conf->worker_groups[i].loprio_list :
5349 &conf->worker_groups[i].handle_list;
bfc90cb0 5350 wg = &conf->worker_groups[i];
851c30c9
SL		 5351 if (!list_empty(handle_list))
5352 break;
5353 }
5354 }
8b3e6cdc
DW		 5355
5356 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
5357 __func__,
851c30c9 5358 list_empty(handle_list) ? "empty" : "busy",
8b3e6cdc
DW		 5359 list_empty(&conf->hold_list) ? "empty" : "busy",
5360 atomic_read(&conf->pending_full_writes), conf->bypass_count);
5361
851c30c9
SL		 5362 if (!list_empty(handle_list)) {
5363 sh = list_entry(handle_list->next, typeof(*sh), lru);
8b3e6cdc
DW		 5364
5365 if (list_empty(&conf->hold_list))
5366 conf->bypass_count = 0;
5367 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
5368 if (conf->hold_list.next == conf->last_hold)
5369 conf->bypass_count++;
5370 else {
5371 conf->last_hold = conf->hold_list.next;
5372 conf->bypass_count -= conf->bypass_threshold;
5373 if (conf->bypass_count < 0)
5374 conf->bypass_count = 0;
5375 }
5376 }
5377 } else if (!list_empty(&conf->hold_list) &&
5378 ((conf->bypass_threshold &&
5379 conf->bypass_count > conf->bypass_threshold) ||
5380 atomic_read(&conf->pending_full_writes) == 0)) {
851c30c9
SL		 5381
5382 list_for_each_entry(tmp, &conf->hold_list, lru) {
5383 if (conf->worker_cnt_per_group == 0 ||
5384 group == ANY_GROUP ||
5385 !cpu_online(tmp->cpu) ||
5386 cpu_to_group(tmp->cpu) == group) {
5387 sh = tmp;
5388 break;
5389 }
5390 }
5391
5392 if (sh) {
5393 conf->bypass_count -= conf->bypass_threshold;
5394 if (conf->bypass_count < 0)
5395 conf->bypass_count = 0;
5396 }
bfc90cb0 5397 wg = NULL;
851c30c9
SL		 5398 }
5399
535ae4eb
SL		 5400 if (!sh) {
5401 if (second_try)
5402 return NULL;
5403 second_try = true;
5404 try_loprio = !try_loprio;
5405 goto again;
5406 }
8b3e6cdc 5407
bfc90cb0
SL		 5408 if (wg) {
5409 wg->stripes_cnt--;
5410 sh->group = NULL;
5411 }
8b3e6cdc 5412 list_del_init(&sh->lru);
c7a6d35e 5413 BUG_ON(atomic_inc_return(&sh->count) != 1);
8b3e6cdc
DW		 5414 return sh;
5415}
f679623f 5416
8811b596
SL		 5417struct raid5_plug_cb {
5418 struct blk_plug_cb cb;
5419 struct list_head list;
566c09c5 5420 struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
8811b596
SL		 5421};
5422
5423static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule)
5424{
5425 struct raid5_plug_cb *cb = container_of(
5426 blk_cb, struct raid5_plug_cb, cb);
5427 struct stripe_head *sh;
5428 struct mddev *mddev = cb->cb.data;
5429 struct r5conf *conf = mddev->private;
a9add5d9 5430 int cnt = 0;
566c09c5 5431 int hash;
8811b596
SL		 5432
5433 if (cb->list.next && !list_empty(&cb->list)) {
5434 spin_lock_irq(&conf->device_lock);
5435 while (!list_empty(&cb->list)) {
5436 sh = list_first_entry(&cb->list, struct stripe_head, lru);
5437 list_del_init(&sh->lru);
5438 /*
5439 * avoid race release_stripe_plug() sees
5440 * STRIPE_ON_UNPLUG_LIST clear but the stripe
5441 * is still in our list
5442 */
4e857c58 5443 smp_mb__before_atomic();
8811b596 5444 clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state);
773ca82f
SL		 5445 /*
5446 * STRIPE_ON_RELEASE_LIST could be set here. In that
5447 * case, the count is always > 1 here
5448 */
566c09c5
SL		 5449 hash = sh->hash_lock_index;
5450 __release_stripe(conf, sh, &cb->temp_inactive_list[hash]);
a9add5d9 5451 cnt++;
8811b596
SL		 5452 }
5453 spin_unlock_irq(&conf->device_lock);
5454 }
566c09c5
SL		 5455 release_inactive_stripe_list(conf, cb->temp_inactive_list,
5456 NR_STRIPE_HASH_LOCKS);
e3620a3a
JB		 5457 if (mddev->queue)
5458 trace_block_unplug(mddev->queue, cnt, !from_schedule);
8811b596
SL		 5459 kfree(cb);
5460}
5461
5462static void release_stripe_plug(struct mddev *mddev,
5463 struct stripe_head *sh)
5464{
5465 struct blk_plug_cb *blk_cb = blk_check_plugged(
5466 raid5_unplug, mddev,
5467 sizeof(struct raid5_plug_cb));
5468 struct raid5_plug_cb *cb;
5469
5470 if (!blk_cb) {
6d036f7d 5471 raid5_release_stripe(sh);
8811b596
SL		 5472 return;
5473 }
5474
5475 cb = container_of(blk_cb, struct raid5_plug_cb, cb);
5476
566c09c5
SL		 5477 if (cb->list.next == NULL) {
5478 int i;
8811b596 5479 INIT_LIST_HEAD(&cb->list);
566c09c5
SL		 5480 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5481 INIT_LIST_HEAD(cb->temp_inactive_list + i);
5482 }
8811b596
SL		 5483
5484 if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state))
5485 list_add_tail(&sh->lru, &cb->list);
5486 else
6d036f7d 5487 raid5_release_stripe(sh);
8811b596
SL		 5488}
5489
620125f2
SL		 5490static void make_discard_request(struct mddev *mddev, struct bio *bi)
5491{
5492 struct r5conf *conf = mddev->private;
5493 sector_t logical_sector, last_sector;
5494 struct stripe_head *sh;
620125f2
SL		 5495 int stripe_sectors;
5496
5497 if (mddev->reshape_position != MaxSector)
5498 /* Skip discard while reshape is happening */
5499 return;
5500
4f024f37 5501 logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1);
b0f01ecf 5502 last_sector = bio_end_sector(bi);
620125f2
SL		 5503
5504 bi->bi_next = NULL;
620125f2
SL		 5505
5506 stripe_sectors = conf->chunk_sectors *
5507 (conf->raid_disks - conf->max_degraded);
5508 logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector,
5509 stripe_sectors);
5510 sector_div(last_sector, stripe_sectors);
5511
5512 logical_sector *= conf->chunk_sectors;
5513 last_sector *= conf->chunk_sectors;
5514
5515 for (; logical_sector < last_sector;
5516 logical_sector += STRIPE_SECTORS) {
5517 DEFINE_WAIT(w);
5518 int d;
5519 again:
6d036f7d 5520 sh = raid5_get_active_stripe(conf, logical_sector, 0, 0, 0);
620125f2
SL		 5521 prepare_to_wait(&conf->wait_for_overlap, &w,
5522 TASK_UNINTERRUPTIBLE);
f8dfcffd
N		 5523 set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
5524 if (test_bit(STRIPE_SYNCING, &sh->state)) {
6d036f7d 5525 raid5_release_stripe(sh);
f8dfcffd
N		 5526 schedule();
5527 goto again;
5528 }
5529 clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
620125f2
SL		 5530 spin_lock_irq(&sh->stripe_lock);
5531 for (d = 0; d < conf->raid_disks; d++) {
5532 if (d == sh->pd_idx || d == sh->qd_idx)
5533 continue;
5534 if (sh->dev[d].towrite || sh->dev[d].toread) {
5535 set_bit(R5_Overlap, &sh->dev[d].flags);
5536 spin_unlock_irq(&sh->stripe_lock);
6d036f7d 5537 raid5_release_stripe(sh);
620125f2
SL		 5538 schedule();
5539 goto again;
5540 }
5541 }
f8dfcffd 5542 set_bit(STRIPE_DISCARD, &sh->state);
620125f2 5543 finish_wait(&conf->wait_for_overlap, &w);
7a87f434 5544 sh->overwrite_disks = 0;
620125f2
SL		 5545 for (d = 0; d < conf->raid_disks; d++) {
5546 if (d == sh->pd_idx || d == sh->qd_idx)
5547 continue;
5548 sh->dev[d].towrite = bi;
5549 set_bit(R5_OVERWRITE, &sh->dev[d].flags);
016c76ac 5550 bio_inc_remaining(bi);
49728050 5551 md_write_inc(mddev, bi);
7a87f434 5552 sh->overwrite_disks++;
620125f2
SL		 5553 }
5554 spin_unlock_irq(&sh->stripe_lock);
5555 if (conf->mddev->bitmap) {
5556 for (d = 0;
5557 d < conf->raid_disks - conf->max_degraded;
5558 d++)
e64e4018
AS		 5559 md_bitmap_startwrite(mddev->bitmap,
5560 sh->sector,
5561 STRIPE_SECTORS,
5562 0);
620125f2
SL		 5563 sh->bm_seq = conf->seq_flush + 1;
5564 set_bit(STRIPE_BIT_DELAY, &sh->state);
5565 }
5566
5567 set_bit(STRIPE_HANDLE, &sh->state);
5568 clear_bit(STRIPE_DELAYED, &sh->state);
5569 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
5570 atomic_inc(&conf->preread_active_stripes);
5571 release_stripe_plug(mddev, sh);
5572 }
5573
016c76ac 5574 bio_endio(bi);
620125f2
SL		 5575}
5576
cc27b0c7 5577static bool raid5_make_request(struct mddev *mddev, struct bio * bi)
1da177e4 5578{
d1688a6d 5579 struct r5conf *conf = mddev->private;
911d4ee8 5580 int dd_idx;
1da177e4
LT		 5581 sector_t new_sector;
5582 sector_t logical_sector, last_sector;
5583 struct stripe_head *sh;
a362357b 5584 const int rw = bio_data_dir(bi);
27c0f68f
SL		 5585 DEFINE_WAIT(w);
5586 bool do_prepare;
3bddb7f8 5587 bool do_flush = false;
1da177e4 5588
1eff9d32 5589 if (unlikely(bi->bi_opf & REQ_PREFLUSH)) {
1532d9e8 5590 int ret = log_handle_flush_request(conf, bi);
828cbe98
SL		 5591
5592 if (ret == 0)
cc27b0c7 5593 return true;
828cbe98 5594 if (ret == -ENODEV) {
775d7831
DJ		 5595 if (md_flush_request(mddev, bi))
5596 return true;
828cbe98
SL		 5597 }
5598 /* ret == -EAGAIN, fallback */
3bddb7f8
SL		 5599 /*
5600 * if r5l_handle_flush_request() didn't clear REQ_PREFLUSH,
5601 * we need to flush journal device
5602 */
5603 do_flush = bi->bi_opf & REQ_PREFLUSH;
e5dcdd80
N		 5604 }
5605
cc27b0c7
N		 5606 if (!md_write_start(mddev, bi))
5607 return false;
9ffc8f7c
EM		 5608 /*
5609 * If array is degraded, better not do chunk aligned read because
5610 * later we might have to read it again in order to reconstruct
5611 * data on failed drives.
5612 */
5613 if (rw == READ && mddev->degraded == 0 &&
7ef6b12a
ML		 5614 mddev->reshape_position == MaxSector) {
5615 bi = chunk_aligned_read(mddev, bi);
5616 if (!bi)
cc27b0c7 5617 return true;
7ef6b12a 5618 }
52488615 5619
796a5cf0 5620 if (unlikely(bio_op(bi) == REQ_OP_DISCARD)) {
620125f2 5621 make_discard_request(mddev, bi);
cc27b0c7
N		 5622 md_write_end(mddev);
5623 return true;
620125f2
SL		 5624 }
5625
4f024f37 5626 logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1);
f73a1c7d 5627 last_sector = bio_end_sector(bi);
1da177e4 5628 bi->bi_next = NULL;
06d91a5f 5629
27c0f68f 5630 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
1da177e4 5631 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
b5663ba4 5632 int previous;
c46501b2 5633 int seq;
b578d55f 5634
27c0f68f 5635 do_prepare = false;
7ecaa1e6 5636 retry:
c46501b2 5637 seq = read_seqcount_begin(&conf->gen_lock);
b5663ba4 5638 previous = 0;
27c0f68f
SL		 5639 if (do_prepare)
5640 prepare_to_wait(&conf->wait_for_overlap, &w,
5641 TASK_UNINTERRUPTIBLE);
b0f9ec04 5642 if (unlikely(conf->reshape_progress != MaxSector)) {
fef9c61f 5643 /* spinlock is needed as reshape_progress may be
df8e7f76
N		 5644 * 64bit on a 32bit platform, and so it might be
5645 * possible to see a half-updated value
aeb878b0 5646 * Of course reshape_progress could change after
df8e7f76
N		 5647 * the lock is dropped, so once we get a reference
5648 * to the stripe that we think it is, we will have
5649 * to check again.
5650 */
7ecaa1e6 5651 spin_lock_irq(&conf->device_lock);
2c810cdd 5652 if (mddev->reshape_backwards
fef9c61f
N		 5653 ? logical_sector < conf->reshape_progress
5654 : logical_sector >= conf->reshape_progress) {
b5663ba4
N		 5655 previous = 1;
5656 } else {
2c810cdd 5657 if (mddev->reshape_backwards
fef9c61f
N		 5658 ? logical_sector < conf->reshape_safe
5659 : logical_sector >= conf->reshape_safe) {
b578d55f
N		 5660 spin_unlock_irq(&conf->device_lock);
5661 schedule();
27c0f68f 5662 do_prepare = true;
b578d55f
N		 5663 goto retry;
5664 }
5665 }
7ecaa1e6
N		 5666 spin_unlock_irq(&conf->device_lock);
5667 }
16a53ecc 5668
112bf897
N		 5669 new_sector = raid5_compute_sector(conf, logical_sector,
5670 previous,
911d4ee8 5671 &dd_idx, NULL);
849674e4 5672 pr_debug("raid456: raid5_make_request, sector %llu logical %llu\n",
c46501b2 5673 (unsigned long long)new_sector,
1da177e4
LT		 5674 (unsigned long long)logical_sector);
5675
6d036f7d 5676 sh = raid5_get_active_stripe(conf, new_sector, previous,
1eff9d32 5677 (bi->bi_opf & REQ_RAHEAD), 0);
1da177e4 5678 if (sh) {
b0f9ec04 5679 if (unlikely(previous)) {
7ecaa1e6 5680 /* expansion might have moved on while waiting for a
df8e7f76
N		 5681 * stripe, so we must do the range check again.
5682 * Expansion could still move past after this
5683 * test, but as we are holding a reference to
5684 * 'sh', we know that if that happens,
5685 * STRIPE_EXPANDING will get set and the expansion
5686 * won't proceed until we finish with the stripe.
7ecaa1e6
N		 5687 */
5688 int must_retry = 0;
5689 spin_lock_irq(&conf->device_lock);
2c810cdd 5690 if (mddev->reshape_backwards
b0f9ec04
N		 5691 ? logical_sector >= conf->reshape_progress
5692 : logical_sector < conf->reshape_progress)
7ecaa1e6
N		 5693 /* mismatch, need to try again */
5694 must_retry = 1;
5695 spin_unlock_irq(&conf->device_lock);
5696 if (must_retry) {
6d036f7d 5697 raid5_release_stripe(sh);
7a3ab908 5698 schedule();
27c0f68f 5699 do_prepare = true;
7ecaa1e6
N		 5700 goto retry;
5701 }
5702 }
c46501b2
N		 5703 if (read_seqcount_retry(&conf->gen_lock, seq)) {
5704 /* Might have got the wrong stripe_head
5705 * by accident
5706 */
6d036f7d 5707 raid5_release_stripe(sh);
c46501b2
N		 5708 goto retry;
5709 }
e62e58a5 5710
7ecaa1e6 5711 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
da41ba65 5712 !add_stripe_bio(sh, bi, dd_idx, rw, previous)) {
7ecaa1e6
N		 5713 /* Stripe is busy expanding or
5714 * add failed due to overlap. Flush everything
1da177e4
LT		 5715 * and wait a while
5716 */
482c0834 5717 md_wakeup_thread(mddev->thread);
6d036f7d 5718 raid5_release_stripe(sh);
1da177e4 5719 schedule();
27c0f68f 5720 do_prepare = true;
1da177e4
LT		 5721 goto retry;
5722 }
3bddb7f8
SL		 5723 if (do_flush) {
5724 set_bit(STRIPE_R5C_PREFLUSH, &sh->state);
5725 /* we only need flush for one stripe */
5726 do_flush = false;
5727 }
5728
a7ede3d1 5729 if (!sh->batch_head || sh == sh->batch_head)
6ce220dd 5730 set_bit(STRIPE_HANDLE, &sh->state);
6ed3003c 5731 clear_bit(STRIPE_DELAYED, &sh->state);
59fc630b 5732 if ((!sh->batch_head || sh == sh->batch_head) &&
1eff9d32 5733 (bi->bi_opf & REQ_SYNC) &&
729a1866
N		 5734 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
5735 atomic_inc(&conf->preread_active_stripes);
8811b596 5736 release_stripe_plug(mddev, sh);
1da177e4
LT		 5737 } else {
5738 /* cannot get stripe for read-ahead, just give-up */
4e4cbee9 5739 bi->bi_status = BLK_STS_IOERR;
1da177e4
LT		 5740 break;
5741 }
1da177e4 5742 }
27c0f68f 5743 finish_wait(&conf->wait_for_overlap, &w);
7c13edc8 5744
49728050
N		 5745 if (rw == WRITE)
5746 md_write_end(mddev);
016c76ac 5747 bio_endio(bi);
cc27b0c7 5748 return true;
1da177e4
LT		 5749}
5750
fd01b88c 5751static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks);
b522adcd 5752
fd01b88c 5753static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped)
1da177e4 5754{
52c03291
N		 5755 /* reshaping is quite different to recovery/resync so it is
5756 * handled quite separately ... here.
5757 *
5758 * On each call to sync_request, we gather one chunk worth of
5759 * destination stripes and flag them as expanding.
5760 * Then we find all the source stripes and request reads.
5761 * As the reads complete, handle_stripe will copy the data
5762 * into the destination stripe and release that stripe.
5763 */
d1688a6d 5764 struct r5conf *conf = mddev->private;
1da177e4 5765 struct stripe_head *sh;
db0505d3 5766 struct md_rdev *rdev;
ccfcc3c1 5767 sector_t first_sector, last_sector;
f416885e
N		 5768 int raid_disks = conf->previous_raid_disks;
5769 int data_disks = raid_disks - conf->max_degraded;
5770 int new_data_disks = conf->raid_disks - conf->max_degraded;
52c03291
N		 5771 int i;
5772 int dd_idx;
c8f517c4 5773 sector_t writepos, readpos, safepos;
ec32a2bd 5774 sector_t stripe_addr;
7a661381 5775 int reshape_sectors;
ab69ae12 5776 struct list_head stripes;
92140480 5777 sector_t retn;
52c03291 5778
fef9c61f
N		 5779 if (sector_nr == 0) {
5780 /* If restarting in the middle, skip the initial sectors */
2c810cdd 5781 if (mddev->reshape_backwards &&
fef9c61f
N		 5782 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
5783 sector_nr = raid5_size(mddev, 0, 0)
5784 - conf->reshape_progress;
6cbd8148
N		 5785 } else if (mddev->reshape_backwards &&
5786 conf->reshape_progress == MaxSector) {
5787 /* shouldn't happen, but just in case, finish up.*/
5788 sector_nr = MaxSector;
2c810cdd 5789 } else if (!mddev->reshape_backwards &&
fef9c61f
N		 5790 conf->reshape_progress > 0)
5791 sector_nr = conf->reshape_progress;
f416885e 5792 sector_div(sector_nr, new_data_disks);
fef9c61f 5793 if (sector_nr) {
8dee7211
N		 5794 mddev->curr_resync_completed = sector_nr;
5795 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
fef9c61f 5796 *skipped = 1;
92140480
N		 5797 retn = sector_nr;
5798 goto finish;
fef9c61f 5799 }
52c03291
N		 5800 }
5801
7a661381
N		 5802 /* We need to process a full chunk at a time.
5803 * If old and new chunk sizes differ, we need to process the
5804 * largest of these
5805 */
3cb5edf4
N		 5806
5807 reshape_sectors = max(conf->chunk_sectors, conf->prev_chunk_sectors);
7a661381 5808
b5254dd5
N		 5809 /* We update the metadata at least every 10 seconds, or when
5810 * the data about to be copied would over-write the source of
5811 * the data at the front of the range. i.e. one new_stripe
5812 * along from reshape_progress new_maps to after where
5813 * reshape_safe old_maps to
52c03291 5814 */
fef9c61f 5815 writepos = conf->reshape_progress;
f416885e 5816 sector_div(writepos, new_data_disks);
c8f517c4
N		 5817 readpos = conf->reshape_progress;
5818 sector_div(readpos, data_disks);
fef9c61f 5819 safepos = conf->reshape_safe;
f416885e 5820 sector_div(safepos, data_disks);
2c810cdd 5821 if (mddev->reshape_backwards) {
c74c0d76
N		 5822 BUG_ON(writepos < reshape_sectors);
5823 writepos -= reshape_sectors;
c8f517c4 5824 readpos += reshape_sectors;
7a661381 5825 safepos += reshape_sectors;
fef9c61f 5826 } else {
7a661381 5827 writepos += reshape_sectors;
c74c0d76
N		 5828 /* readpos and safepos are worst-case calculations.
5829 * A negative number is overly pessimistic, and causes
5830 * obvious problems for unsigned storage. So clip to 0.
5831 */
ed37d83e
N		 5832 readpos -= min_t(sector_t, reshape_sectors, readpos);
5833 safepos -= min_t(sector_t, reshape_sectors, safepos);
fef9c61f 5834 }
52c03291 5835
b5254dd5
N		 5836 /* Having calculated the 'writepos' possibly use it
5837 * to set 'stripe_addr' which is where we will write to.
5838 */
5839 if (mddev->reshape_backwards) {
5840 BUG_ON(conf->reshape_progress == 0);
5841 stripe_addr = writepos;
5842 BUG_ON((mddev->dev_sectors &
5843 ~((sector_t)reshape_sectors - 1))
5844 - reshape_sectors - stripe_addr
5845 != sector_nr);
5846 } else {
5847 BUG_ON(writepos != sector_nr + reshape_sectors);
5848 stripe_addr = sector_nr;
5849 }
5850
c8f517c4
N		 5851 /* 'writepos' is the most advanced device address we might write.
5852 * 'readpos' is the least advanced device address we might read.
5853 * 'safepos' is the least address recorded in the metadata as having
5854 * been reshaped.
b5254dd5
N		 5855 * If there is a min_offset_diff, these are adjusted either by
5856 * increasing the safepos/readpos if diff is negative, or
5857 * increasing writepos if diff is positive.
5858 * If 'readpos' is then behind 'writepos', there is no way that we can
c8f517c4
N		 5859 * ensure safety in the face of a crash - that must be done by userspace
5860 * making a backup of the data. So in that case there is no particular
5861 * rush to update metadata.
5862 * Otherwise if 'safepos' is behind 'writepos', then we really need to
5863 * update the metadata to advance 'safepos' to match 'readpos' so that
5864 * we can be safe in the event of a crash.
5865 * So we insist on updating metadata if safepos is behind writepos and
5866 * readpos is beyond writepos.
5867 * In any case, update the metadata every 10 seconds.
5868 * Maybe that number should be configurable, but I'm not sure it is
5869 * worth it.... maybe it could be a multiple of safemode_delay???
5870 */
b5254dd5
N		 5871 if (conf->min_offset_diff < 0) {
5872 safepos += -conf->min_offset_diff;
5873 readpos += -conf->min_offset_diff;
5874 } else
5875 writepos += conf->min_offset_diff;
5876
2c810cdd 5877 if ((mddev->reshape_backwards
c8f517c4
N		 5878 ? (safepos > writepos && readpos < writepos)
5879 : (safepos < writepos && readpos > writepos)) ||
5880 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
52c03291
N		 5881 /* Cannot proceed until we've updated the superblock... */
5882 wait_event(conf->wait_for_overlap,
c91abf5a
N		 5883 atomic_read(&conf->reshape_stripes)==0
5884 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5885 if (atomic_read(&conf->reshape_stripes) != 0)
5886 return 0;
fef9c61f 5887 mddev->reshape_position = conf->reshape_progress;
75d3da43 5888 mddev->curr_resync_completed = sector_nr;
db0505d3
N		 5889 if (!mddev->reshape_backwards)
5890 /* Can update recovery_offset */
5891 rdev_for_each(rdev, mddev)
5892 if (rdev->raid_disk >= 0 &&
5893 !test_bit(Journal, &rdev->flags) &&
5894 !test_bit(In_sync, &rdev->flags) &&
5895 rdev->recovery_offset < sector_nr)
5896 rdev->recovery_offset = sector_nr;
5897
c8f517c4 5898 conf->reshape_checkpoint = jiffies;
2953079c 5899 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
52c03291 5900 md_wakeup_thread(mddev->thread);
2953079c 5901 wait_event(mddev->sb_wait, mddev->sb_flags == 0 ||
c91abf5a
N		 5902 test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5903 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5904 return 0;
52c03291 5905 spin_lock_irq(&conf->device_lock);
fef9c61f 5906 conf->reshape_safe = mddev->reshape_position;
52c03291
N		 5907 spin_unlock_irq(&conf->device_lock);
5908 wake_up(&conf->wait_for_overlap);
acb180b0 5909 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
52c03291
N		 5910 }
5911
ab69ae12 5912 INIT_LIST_HEAD(&stripes);
7a661381 5913 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
52c03291 5914 int j;
a9f326eb 5915 int skipped_disk = 0;
6d036f7d 5916 sh = raid5_get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
52c03291
N		 5917 set_bit(STRIPE_EXPANDING, &sh->state);
5918 atomic_inc(&conf->reshape_stripes);
5919 /* If any of this stripe is beyond the end of the old
5920 * array, then we need to zero those blocks
5921 */
5922 for (j=sh->disks; j--;) {
5923 sector_t s;
5924 if (j == sh->pd_idx)
5925 continue;
f416885e 5926 if (conf->level == 6 &&
d0dabf7e 5927 j == sh->qd_idx)
f416885e 5928 continue;
6d036f7d 5929 s = raid5_compute_blocknr(sh, j, 0);
b522adcd 5930 if (s < raid5_size(mddev, 0, 0)) {
a9f326eb 5931 skipped_disk = 1;
52c03291
N		 5932 continue;
5933 }
5934 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
5935 set_bit(R5_Expanded, &sh->dev[j].flags);
5936 set_bit(R5_UPTODATE, &sh->dev[j].flags);
5937 }
a9f326eb 5938 if (!skipped_disk) {
52c03291
N		 5939 set_bit(STRIPE_EXPAND_READY, &sh->state);
5940 set_bit(STRIPE_HANDLE, &sh->state);
5941 }
ab69ae12 5942 list_add(&sh->lru, &stripes);
52c03291
N		 5943 }
5944 spin_lock_irq(&conf->device_lock);
2c810cdd 5945 if (mddev->reshape_backwards)
7a661381 5946 conf->reshape_progress -= reshape_sectors * new_data_disks;
fef9c61f 5947 else
7a661381 5948 conf->reshape_progress += reshape_sectors * new_data_disks;
52c03291
N		 5949 spin_unlock_irq(&conf->device_lock);
5950 /* Ok, those stripe are ready. We can start scheduling
5951 * reads on the source stripes.
5952 * The source stripes are determined by mapping the first and last
5953 * block on the destination stripes.
5954 */
52c03291 5955 first_sector =
ec32a2bd 5956 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
911d4ee8 5957 1, &dd_idx, NULL);
52c03291 5958 last_sector =
0e6e0271 5959 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
09c9e5fa 5960 * new_data_disks - 1),
911d4ee8 5961 1, &dd_idx, NULL);
58c0fed4
AN		 5962 if (last_sector >= mddev->dev_sectors)
5963 last_sector = mddev->dev_sectors - 1;
52c03291 5964 while (first_sector <= last_sector) {
6d036f7d 5965 sh = raid5_get_active_stripe(conf, first_sector, 1, 0, 1);
52c03291
N		 5966 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
5967 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 5968 raid5_release_stripe(sh);
52c03291
N		 5969 first_sector += STRIPE_SECTORS;
5970 }
ab69ae12
N		 5971 /* Now that the sources are clearly marked, we can release
5972 * the destination stripes
5973 */
5974 while (!list_empty(&stripes)) {
5975 sh = list_entry(stripes.next, struct stripe_head, lru);
5976 list_del_init(&sh->lru);
6d036f7d 5977 raid5_release_stripe(sh);
ab69ae12 5978 }
c6207277
N		 5979 /* If this takes us to the resync_max point where we have to pause,
5980 * then we need to write out the superblock.
5981 */
7a661381 5982 sector_nr += reshape_sectors;
92140480
N		 5983 retn = reshape_sectors;
5984finish:
c5e19d90
N		 5985 if (mddev->curr_resync_completed > mddev->resync_max ||
5986 (sector_nr - mddev->curr_resync_completed) * 2
c03f6a19 5987 >= mddev->resync_max - mddev->curr_resync_completed) {
c6207277
N		 5988 /* Cannot proceed until we've updated the superblock... */
5989 wait_event(conf->wait_for_overlap,
c91abf5a
N		 5990 atomic_read(&conf->reshape_stripes) == 0
5991 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5992 if (atomic_read(&conf->reshape_stripes) != 0)
5993 goto ret;
fef9c61f 5994 mddev->reshape_position = conf->reshape_progress;
75d3da43 5995 mddev->curr_resync_completed = sector_nr;
db0505d3
N		 5996 if (!mddev->reshape_backwards)
5997 /* Can update recovery_offset */
5998 rdev_for_each(rdev, mddev)
5999 if (rdev->raid_disk >= 0 &&
6000 !test_bit(Journal, &rdev->flags) &&
6001 !test_bit(In_sync, &rdev->flags) &&
6002 rdev->recovery_offset < sector_nr)
6003 rdev->recovery_offset = sector_nr;
c8f517c4 6004 conf->reshape_checkpoint = jiffies;
2953079c 6005 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
c6207277
N		 6006 md_wakeup_thread(mddev->thread);
6007 wait_event(mddev->sb_wait,
2953079c 6008 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)
c91abf5a
N		 6009 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
6010 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6011 goto ret;
c6207277 6012 spin_lock_irq(&conf->device_lock);
fef9c61f 6013 conf->reshape_safe = mddev->reshape_position;
c6207277
N		 6014 spin_unlock_irq(&conf->device_lock);
6015 wake_up(&conf->wait_for_overlap);
acb180b0 6016 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
c6207277 6017 }
c91abf5a 6018ret:
92140480 6019 return retn;
52c03291
N		 6020}
6021
849674e4
SL		 6022static inline sector_t raid5_sync_request(struct mddev *mddev, sector_t sector_nr,
6023 int *skipped)
52c03291 6024{
d1688a6d 6025 struct r5conf *conf = mddev->private;
52c03291 6026 struct stripe_head *sh;
58c0fed4 6027 sector_t max_sector = mddev->dev_sectors;
57dab0bd 6028 sector_t sync_blocks;
16a53ecc
N		 6029 int still_degraded = 0;
6030 int i;
1da177e4 6031
72626685 6032 if (sector_nr >= max_sector) {
1da177e4 6033 /* just being told to finish up .. nothing much to do */
cea9c228 6034
29269553
N		 6035 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
6036 end_reshape(conf);
6037 return 0;
6038 }
72626685
N		 6039
6040 if (mddev->curr_resync < max_sector) /* aborted */
e64e4018
AS		 6041 md_bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
6042 &sync_blocks, 1);
16a53ecc 6043 else /* completed sync */
72626685 6044 conf->fullsync = 0;
e64e4018 6045 md_bitmap_close_sync(mddev->bitmap);
72626685 6046
1da177e4
LT		 6047 return 0;
6048 }
ccfcc3c1 6049
64bd660b
N		 6050 /* Allow raid5_quiesce to complete */
6051 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
6052
52c03291
N		 6053 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6054 return reshape_request(mddev, sector_nr, skipped);
f6705578 6055
c6207277
N		 6056 /* No need to check resync_max as we never do more than one
6057 * stripe, and as resync_max will always be on a chunk boundary,
6058 * if the check in md_do_sync didn't fire, there is no chance
6059 * of overstepping resync_max here
6060 */
6061
16a53ecc 6062 /* if there is too many failed drives and we are trying
1da177e4
LT		 6063 * to resync, then assert that we are finished, because there is
6064 * nothing we can do.
6065 */
3285edf1 6066 if (mddev->degraded >= conf->max_degraded &&
16a53ecc 6067 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
58c0fed4 6068 sector_t rv = mddev->dev_sectors - sector_nr;
57afd89f 6069 *skipped = 1;
1da177e4
LT		 6070 return rv;
6071 }
6f608040 6072 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
6073 !conf->fullsync &&
e64e4018 6074 !md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
6f608040 6075 sync_blocks >= STRIPE_SECTORS) {
72626685
N		 6076 /* we can skip this block, and probably more */
6077 sync_blocks /= STRIPE_SECTORS;
6078 *skipped = 1;
6079 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
6080 }
1da177e4 6081
e64e4018 6082 md_bitmap_cond_end_sync(mddev->bitmap, sector_nr, false);
b47490c9 6083
6d036f7d 6084 sh = raid5_get_active_stripe(conf, sector_nr, 0, 1, 0);
1da177e4 6085 if (sh == NULL) {
6d036f7d 6086 sh = raid5_get_active_stripe(conf, sector_nr, 0, 0, 0);
1da177e4 6087 /* make sure we don't swamp the stripe cache if someone else
16a53ecc 6088 * is trying to get access
1da177e4 6089 */
66c006a5 6090 schedule_timeout_uninterruptible(1);
1da177e4 6091 }
16a53ecc 6092 /* Need to check if array will still be degraded after recovery/resync
16d9cfab
EM		 6093 * Note in case of > 1 drive failures it's possible we're rebuilding
6094 * one drive while leaving another faulty drive in array.
16a53ecc 6095 */
16d9cfab
EM		 6096 rcu_read_lock();
6097 for (i = 0; i < conf->raid_disks; i++) {
6aa7de05 6098 struct md_rdev *rdev = READ_ONCE(conf->disks[i].rdev);
16d9cfab
EM		 6099
6100 if (rdev == NULL || test_bit(Faulty, &rdev->flags))
16a53ecc 6101 still_degraded = 1;
16d9cfab
EM		 6102 }
6103 rcu_read_unlock();
16a53ecc 6104
e64e4018 6105 md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
16a53ecc 6106
83206d66 6107 set_bit(STRIPE_SYNC_REQUESTED, &sh->state);
053f5b65 6108 set_bit(STRIPE_HANDLE, &sh->state);
1da177e4 6109
6d036f7d 6110 raid5_release_stripe(sh);
1da177e4
LT		 6111
6112 return STRIPE_SECTORS;
6113}
6114
0472a42b
N		 6115static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio,
6116 unsigned int offset)
46031f9a
RBJ		 6117{
6118 /* We may not be able to submit a whole bio at once as there
6119 * may not be enough stripe_heads available.
6120 * We cannot pre-allocate enough stripe_heads as we may need
6121 * more than exist in the cache (if we allow ever large chunks).
6122 * So we do one stripe head at a time and record in
6123 * ->bi_hw_segments how many have been done.
6124 *
6125 * We *know* that this entire raid_bio is in one chunk, so
6126 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
6127 */
6128 struct stripe_head *sh;
911d4ee8 6129 int dd_idx;
46031f9a
RBJ		 6130 sector_t sector, logical_sector, last_sector;
6131 int scnt = 0;
46031f9a
RBJ		 6132 int handled = 0;
6133
4f024f37
KO		 6134 logical_sector = raid_bio->bi_iter.bi_sector &
6135 ~((sector_t)STRIPE_SECTORS-1);
112bf897 6136 sector = raid5_compute_sector(conf, logical_sector,
911d4ee8 6137 0, &dd_idx, NULL);
f73a1c7d 6138 last_sector = bio_end_sector(raid_bio);
46031f9a
RBJ		 6139
6140 for (; logical_sector < last_sector;
387bb173
NB		 6141 logical_sector += STRIPE_SECTORS,
6142 sector += STRIPE_SECTORS,
6143 scnt++) {
46031f9a 6144
0472a42b 6145 if (scnt < offset)
46031f9a
RBJ		 6146 /* already done this stripe */
6147 continue;
6148
6d036f7d 6149 sh = raid5_get_active_stripe(conf, sector, 0, 1, 1);
46031f9a
RBJ		 6150
6151 if (!sh) {
6152 /* failed to get a stripe - must wait */
46031f9a 6153 conf->retry_read_aligned = raid_bio;
0472a42b 6154 conf->retry_read_offset = scnt;
46031f9a
RBJ		 6155 return handled;
6156 }
6157
da41ba65 6158 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) {
6d036f7d 6159 raid5_release_stripe(sh);
387bb173 6160 conf->retry_read_aligned = raid_bio;
0472a42b 6161 conf->retry_read_offset = scnt;
387bb173
NB		 6162 return handled;
6163 }
6164
3f9e7c14 6165 set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags);
36d1c647 6166 handle_stripe(sh);
6d036f7d 6167 raid5_release_stripe(sh);
46031f9a
RBJ		 6168 handled++;
6169 }
016c76ac
N		 6170
6171 bio_endio(raid_bio);
6172
46031f9a 6173 if (atomic_dec_and_test(&conf->active_aligned_reads))
b1b46486 6174 wake_up(&conf->wait_for_quiescent);
46031f9a
RBJ		 6175 return handled;
6176}
6177
bfc90cb0 6178static int handle_active_stripes(struct r5conf *conf, int group,
566c09c5
SL		 6179 struct r5worker *worker,
6180 struct list_head *temp_inactive_list)
efcd487c
CH		 6181 __releases(&conf->device_lock)
6182 __acquires(&conf->device_lock)
46a06401
SL		 6183{
6184 struct stripe_head *batch[MAX_STRIPE_BATCH], *sh;
566c09c5
SL		 6185 int i, batch_size = 0, hash;
6186 bool release_inactive = false;
46a06401
SL		 6187
6188 while (batch_size < MAX_STRIPE_BATCH &&
851c30c9 6189 (sh = __get_priority_stripe(conf, group)) != NULL)
46a06401
SL		 6190 batch[batch_size++] = sh;
6191
566c09c5
SL		 6192 if (batch_size == 0) {
6193 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
6194 if (!list_empty(temp_inactive_list + i))
6195 break;
a8c34f91
SL		 6196 if (i == NR_STRIPE_HASH_LOCKS) {
6197 spin_unlock_irq(&conf->device_lock);
1532d9e8 6198 log_flush_stripe_to_raid(conf);
a8c34f91 6199 spin_lock_irq(&conf->device_lock);
566c09c5 6200 return batch_size;
a8c34f91 6201 }
566c09c5
SL		 6202 release_inactive = true;
6203 }
46a06401
SL		 6204 spin_unlock_irq(&conf->device_lock);
6205
566c09c5
SL		 6206 release_inactive_stripe_list(conf, temp_inactive_list,
6207 NR_STRIPE_HASH_LOCKS);
6208
a8c34f91 6209 r5l_flush_stripe_to_raid(conf->log);
566c09c5
SL		 6210 if (release_inactive) {
6211 spin_lock_irq(&conf->device_lock);
6212 return 0;
6213 }
6214
46a06401
SL		 6215 for (i = 0; i < batch_size; i++)
6216 handle_stripe(batch[i]);
ff875738 6217 log_write_stripe_run(conf);
46a06401
SL		 6218
6219 cond_resched();
6220
6221 spin_lock_irq(&conf->device_lock);
566c09c5
SL		 6222 for (i = 0; i < batch_size; i++) {
6223 hash = batch[i]->hash_lock_index;
6224 __release_stripe(conf, batch[i], &temp_inactive_list[hash]);
6225 }
46a06401
SL		 6226 return batch_size;
6227}
46031f9a 6228
851c30c9
SL		 6229static void raid5_do_work(struct work_struct *work)
6230{
6231 struct r5worker *worker = container_of(work, struct r5worker, work);
6232 struct r5worker_group *group = worker->group;
6233 struct r5conf *conf = group->conf;
16d997b7 6234 struct mddev *mddev = conf->mddev;
851c30c9
SL		 6235 int group_id = group - conf->worker_groups;
6236 int handled;
6237 struct blk_plug plug;
6238
6239 pr_debug("+++ raid5worker active\n");
6240
6241 blk_start_plug(&plug);
6242 handled = 0;
6243 spin_lock_irq(&conf->device_lock);
6244 while (1) {
6245 int batch_size, released;
6246
566c09c5 6247 released = release_stripe_list(conf, worker->temp_inactive_list);
851c30c9 6248
566c09c5
SL		 6249 batch_size = handle_active_stripes(conf, group_id, worker,
6250 worker->temp_inactive_list);
bfc90cb0 6251 worker->working = false;
851c30c9
SL		 6252 if (!batch_size && !released)
6253 break;
6254 handled += batch_size;
16d997b7
N		 6255 wait_event_lock_irq(mddev->sb_wait,
6256 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags),
6257 conf->device_lock);
851c30c9
SL		 6258 }
6259 pr_debug("%d stripes handled\n", handled);
6260
6261 spin_unlock_irq(&conf->device_lock);
7e96d559 6262
9c72a18e
SL		 6263 flush_deferred_bios(conf);
6264
6265 r5l_flush_stripe_to_raid(conf->log);
6266
7e96d559 6267 async_tx_issue_pending_all();
851c30c9
SL		 6268 blk_finish_plug(&plug);
6269
6270 pr_debug("--- raid5worker inactive\n");
6271}
6272
1da177e4
LT		 6273/*
6274 * This is our raid5 kernel thread.
6275 *
6276 * We scan the hash table for stripes which can be handled now.
6277 * During the scan, completed stripes are saved for us by the interrupt
6278 * handler, so that they will not have to wait for our next wakeup.
6279 */
4ed8731d 6280static void raid5d(struct md_thread *thread)
1da177e4 6281{
4ed8731d 6282 struct mddev *mddev = thread->mddev;
d1688a6d 6283 struct r5conf *conf = mddev->private;
1da177e4 6284 int handled;
e1dfa0a2 6285 struct blk_plug plug;
1da177e4 6286
45b4233c 6287 pr_debug("+++ raid5d active\n");
1da177e4
LT		 6288
6289 md_check_recovery(mddev);
1da177e4 6290
e1dfa0a2 6291 blk_start_plug(&plug);
1da177e4
LT		 6292 handled = 0;
6293 spin_lock_irq(&conf->device_lock);
6294 while (1) {
46031f9a 6295 struct bio *bio;
773ca82f 6296 int batch_size, released;
0472a42b 6297 unsigned int offset;
773ca82f 6298
566c09c5 6299 released = release_stripe_list(conf, conf->temp_inactive_list);
edbe83ab
N		 6300 if (released)
6301 clear_bit(R5_DID_ALLOC, &conf->cache_state);
1da177e4 6302
0021b7bc 6303 if (
7c13edc8
N		 6304 !list_empty(&conf->bitmap_list)) {
6305 /* Now is a good time to flush some bitmap updates */
6306 conf->seq_flush++;
700e432d 6307 spin_unlock_irq(&conf->device_lock);
e64e4018 6308 md_bitmap_unplug(mddev->bitmap);
700e432d 6309 spin_lock_irq(&conf->device_lock);
7c13edc8 6310 conf->seq_write = conf->seq_flush;
566c09c5 6311 activate_bit_delay(conf, conf->temp_inactive_list);
72626685 6312 }
0021b7bc 6313 raid5_activate_delayed(conf);
72626685 6314
0472a42b 6315 while ((bio = remove_bio_from_retry(conf, &offset))) {
46031f9a
RBJ		 6316 int ok;
6317 spin_unlock_irq(&conf->device_lock);
0472a42b 6318 ok = retry_aligned_read(conf, bio, offset);
46031f9a
RBJ		 6319 spin_lock_irq(&conf->device_lock);
6320 if (!ok)
6321 break;
6322 handled++;
6323 }
6324
566c09c5
SL		 6325 batch_size = handle_active_stripes(conf, ANY_GROUP, NULL,
6326 conf->temp_inactive_list);
773ca82f 6327 if (!batch_size && !released)
1da177e4 6328 break;
46a06401 6329 handled += batch_size;
1da177e4 6330
2953079c 6331 if (mddev->sb_flags & ~(1 << MD_SB_CHANGE_PENDING)) {
46a06401 6332 spin_unlock_irq(&conf->device_lock);
de393cde 6333 md_check_recovery(mddev);
46a06401
SL		 6334 spin_lock_irq(&conf->device_lock);
6335 }
1da177e4 6336 }
45b4233c 6337 pr_debug("%d stripes handled\n", handled);
1da177e4
LT		 6338
6339 spin_unlock_irq(&conf->device_lock);
2d5b569b
N		 6340 if (test_and_clear_bit(R5_ALLOC_MORE, &conf->cache_state) &&
6341 mutex_trylock(&conf->cache_size_mutex)) {
edbe83ab
N		 6342 grow_one_stripe(conf, __GFP_NOWARN);
6343 /* Set flag even if allocation failed. This helps
6344 * slow down allocation requests when mem is short
6345 */
6346 set_bit(R5_DID_ALLOC, &conf->cache_state);
2d5b569b 6347 mutex_unlock(&conf->cache_size_mutex);
edbe83ab 6348 }
1da177e4 6349
765d704d
SL		 6350 flush_deferred_bios(conf);
6351
0576b1c6
SL		 6352 r5l_flush_stripe_to_raid(conf->log);
6353
c9f21aaf 6354 async_tx_issue_pending_all();
e1dfa0a2 6355 blk_finish_plug(&plug);
1da177e4 6356
45b4233c 6357 pr_debug("--- raid5d inactive\n");
1da177e4
LT		 6358}
6359
3f294f4f 6360static ssize_t
fd01b88c 6361raid5_show_stripe_cache_size(struct mddev *mddev, char *page)
3f294f4f 6362{
7b1485ba
N		 6363 struct r5conf *conf;
6364 int ret = 0;
6365 spin_lock(&mddev->lock);
6366 conf = mddev->private;
96de1e66 6367 if (conf)
edbe83ab 6368 ret = sprintf(page, "%d\n", conf->min_nr_stripes);
7b1485ba
N		 6369 spin_unlock(&mddev->lock);
6370 return ret;
3f294f4f
N		 6371}
6372
c41d4ac4 6373int
fd01b88c 6374raid5_set_cache_size(struct mddev *mddev, int size)
3f294f4f 6375{
483cbbed 6376 int result = 0;
d1688a6d 6377 struct r5conf *conf = mddev->private;
b5470dc5 6378
c41d4ac4 6379 if (size <= 16 || size > 32768)
3f294f4f 6380 return -EINVAL;
486f0644 6381
edbe83ab 6382 conf->min_nr_stripes = size;
2d5b569b 6383 mutex_lock(&conf->cache_size_mutex);
486f0644
N		 6384 while (size < conf->max_nr_stripes &&
6385 drop_one_stripe(conf))
6386 ;
2d5b569b 6387 mutex_unlock(&conf->cache_size_mutex);
486f0644 6388
2214c260 6389 md_allow_write(mddev);
486f0644 6390
2d5b569b 6391 mutex_lock(&conf->cache_size_mutex);
486f0644 6392 while (size > conf->max_nr_stripes)
483cbbed
AN		 6393 if (!grow_one_stripe(conf, GFP_KERNEL)) {
6394 conf->min_nr_stripes = conf->max_nr_stripes;
6395 result = -ENOMEM;
486f0644 6396 break;
483cbbed 6397 }
2d5b569b 6398 mutex_unlock(&conf->cache_size_mutex);
486f0644 6399
483cbbed 6400 return result;
c41d4ac4
N		 6401}
6402EXPORT_SYMBOL(raid5_set_cache_size);
6403
6404static ssize_t
fd01b88c 6405raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len)
c41d4ac4 6406{
6791875e 6407 struct r5conf *conf;
c41d4ac4
N		 6408 unsigned long new;
6409 int err;
6410
6411 if (len >= PAGE_SIZE)
6412 return -EINVAL;
b29bebd6 6413 if (kstrtoul(page, 10, &new))
c41d4ac4 6414 return -EINVAL;
6791875e 6415 err = mddev_lock(mddev);
c41d4ac4
N		 6416 if (err)
6417 return err;
6791875e
N		 6418 conf = mddev->private;
6419 if (!conf)
6420 err = -ENODEV;
6421 else
6422 err = raid5_set_cache_size(mddev, new);
6423 mddev_unlock(mddev);
6424
6425 return err ?: len;
3f294f4f 6426}
007583c9 6427
96de1e66
N		 6428static struct md_sysfs_entry
6429raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
6430 raid5_show_stripe_cache_size,
6431 raid5_store_stripe_cache_size);
3f294f4f 6432
d06f191f
MS		 6433static ssize_t
6434raid5_show_rmw_level(struct mddev *mddev, char *page)
6435{
6436 struct r5conf *conf = mddev->private;
6437 if (conf)
6438 return sprintf(page, "%d\n", conf->rmw_level);
6439 else
6440 return 0;
6441}
6442
6443static ssize_t
6444raid5_store_rmw_level(struct mddev *mddev, const char *page, size_t len)
6445{
6446 struct r5conf *conf = mddev->private;
6447 unsigned long new;
6448
6449 if (!conf)
6450 return -ENODEV;
6451
6452 if (len >= PAGE_SIZE)
6453 return -EINVAL;
6454
6455 if (kstrtoul(page, 10, &new))
6456 return -EINVAL;
6457
6458 if (new != PARITY_DISABLE_RMW && !raid6_call.xor_syndrome)
6459 return -EINVAL;
6460
6461 if (new != PARITY_DISABLE_RMW &&
6462 new != PARITY_ENABLE_RMW &&
6463 new != PARITY_PREFER_RMW)
6464 return -EINVAL;
6465
6466 conf->rmw_level = new;
6467 return len;
6468}
6469
6470static struct md_sysfs_entry
6471raid5_rmw_level = __ATTR(rmw_level, S_IRUGO | S_IWUSR,
6472 raid5_show_rmw_level,
6473 raid5_store_rmw_level);
6474
6475
8b3e6cdc 6476static ssize_t
fd01b88c 6477raid5_show_preread_threshold(struct mddev *mddev, char *page)
8b3e6cdc 6478{
7b1485ba
N		 6479 struct r5conf *conf;
6480 int ret = 0;
6481 spin_lock(&mddev->lock);
6482 conf = mddev->private;
8b3e6cdc 6483 if (conf)
7b1485ba
N		 6484 ret = sprintf(page, "%d\n", conf->bypass_threshold);
6485 spin_unlock(&mddev->lock);
6486 return ret;
8b3e6cdc
DW		 6487}
6488
6489static ssize_t
fd01b88c 6490raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len)
8b3e6cdc 6491{
6791875e 6492 struct r5conf *conf;
4ef197d8 6493 unsigned long new;
6791875e
N		 6494 int err;
6495
8b3e6cdc
DW		 6496 if (len >= PAGE_SIZE)
6497 return -EINVAL;
b29bebd6 6498 if (kstrtoul(page, 10, &new))
8b3e6cdc 6499 return -EINVAL;
6791875e
N		 6500
6501 err = mddev_lock(mddev);
6502 if (err)
6503 return err;
6504 conf = mddev->private;
6505 if (!conf)
6506 err = -ENODEV;
edbe83ab 6507 else if (new > conf->min_nr_stripes)
6791875e
N		 6508 err = -EINVAL;
6509 else
6510 conf->bypass_threshold = new;
6511 mddev_unlock(mddev);
6512 return err ?: len;
8b3e6cdc
DW		 6513}
6514
6515static struct md_sysfs_entry
6516raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
6517 S_IRUGO | S_IWUSR,
6518 raid5_show_preread_threshold,
6519 raid5_store_preread_threshold);
6520
d592a996
SL		 6521static ssize_t
6522raid5_show_skip_copy(struct mddev *mddev, char *page)
6523{
7b1485ba
N		 6524 struct r5conf *conf;
6525 int ret = 0;
6526 spin_lock(&mddev->lock);
6527 conf = mddev->private;
d592a996 6528 if (conf)
7b1485ba
N		 6529 ret = sprintf(page, "%d\n", conf->skip_copy);
6530 spin_unlock(&mddev->lock);
6531 return ret;
d592a996
SL		 6532}
6533
6534static ssize_t
6535raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len)
6536{
6791875e 6537 struct r5conf *conf;
d592a996 6538 unsigned long new;
6791875e
N		 6539 int err;
6540
d592a996
SL		 6541 if (len >= PAGE_SIZE)
6542 return -EINVAL;
d592a996
SL		 6543 if (kstrtoul(page, 10, &new))
6544 return -EINVAL;
6545 new = !!new;
6791875e
N		 6546
6547 err = mddev_lock(mddev);
6548 if (err)
6549 return err;
6550 conf = mddev->private;
6551 if (!conf)
6552 err = -ENODEV;
6553 else if (new != conf->skip_copy) {
6554 mddev_suspend(mddev);
6555 conf->skip_copy = new;
6556 if (new)
dc3b17cc 6557 mddev->queue->backing_dev_info->capabilities |=
6791875e
N		 6558 BDI_CAP_STABLE_WRITES;
6559 else
dc3b17cc 6560 mddev->queue->backing_dev_info->capabilities &=
6791875e
N		 6561 ~BDI_CAP_STABLE_WRITES;
6562 mddev_resume(mddev);
6563 }
6564 mddev_unlock(mddev);
6565 return err ?: len;
d592a996
SL		 6566}
6567
6568static struct md_sysfs_entry
6569raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR,
6570 raid5_show_skip_copy,
6571 raid5_store_skip_copy);
6572
3f294f4f 6573static ssize_t
fd01b88c 6574stripe_cache_active_show(struct mddev *mddev, char *page)
3f294f4f 6575{
d1688a6d 6576 struct r5conf *conf = mddev->private;
96de1e66
N		 6577 if (conf)
6578 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
6579 else
6580 return 0;
3f294f4f
N		 6581}
6582
96de1e66
N		 6583static struct md_sysfs_entry
6584raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
3f294f4f 6585
b721420e
SL		 6586static ssize_t
6587raid5_show_group_thread_cnt(struct mddev *mddev, char *page)
6588{
7b1485ba
N		 6589 struct r5conf *conf;
6590 int ret = 0;
6591 spin_lock(&mddev->lock);
6592 conf = mddev->private;
b721420e 6593 if (conf)
7b1485ba
N		 6594 ret = sprintf(page, "%d\n", conf->worker_cnt_per_group);
6595 spin_unlock(&mddev->lock);
6596 return ret;
b721420e
SL		 6597}
6598
60aaf933 6599static int alloc_thread_groups(struct r5conf *conf, int cnt,
6600 int *group_cnt,
6601 int *worker_cnt_per_group,
6602 struct r5worker_group **worker_groups);
b721420e
SL		 6603static ssize_t
6604raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len)
6605{
6791875e 6606 struct r5conf *conf;
7d5d7b50 6607 unsigned int new;
b721420e 6608 int err;
60aaf933 6609 struct r5worker_group *new_groups, *old_groups;
6610 int group_cnt, worker_cnt_per_group;
b721420e
SL		 6611
6612 if (len >= PAGE_SIZE)
6613 return -EINVAL;
7d5d7b50
SL		 6614 if (kstrtouint(page, 10, &new))
6615 return -EINVAL;
6616 /* 8192 should be big enough */
6617 if (new > 8192)
b721420e
SL		 6618 return -EINVAL;
6619
6791875e
N		 6620 err = mddev_lock(mddev);
6621 if (err)
6622 return err;
6623 conf = mddev->private;
6624 if (!conf)
6625 err = -ENODEV;
6626 else if (new != conf->worker_cnt_per_group) {
6627 mddev_suspend(mddev);
b721420e 6628
6791875e
N		 6629 old_groups = conf->worker_groups;
6630 if (old_groups)
6631 flush_workqueue(raid5_wq);
d206dcfa 6632
6791875e
N		 6633 err = alloc_thread_groups(conf, new,
6634 &group_cnt, &worker_cnt_per_group,
6635 &new_groups);
6636 if (!err) {
6637 spin_lock_irq(&conf->device_lock);
6638 conf->group_cnt = group_cnt;
6639 conf->worker_cnt_per_group = worker_cnt_per_group;
6640 conf->worker_groups = new_groups;
6641 spin_unlock_irq(&conf->device_lock);
b721420e 6642
6791875e
N		 6643 if (old_groups)
6644 kfree(old_groups[0].workers);
6645 kfree(old_groups);
6646 }
6647 mddev_resume(mddev);
b721420e 6648 }
6791875e 6649 mddev_unlock(mddev);
b721420e 6650
6791875e 6651 return err ?: len;
b721420e
SL		 6652}
6653
6654static struct md_sysfs_entry
6655raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR,
6656 raid5_show_group_thread_cnt,
6657 raid5_store_group_thread_cnt);
6658
007583c9 6659static struct attribute *raid5_attrs[] = {
3f294f4f
N		 6660 &raid5_stripecache_size.attr,
6661 &raid5_stripecache_active.attr,
8b3e6cdc 6662 &raid5_preread_bypass_threshold.attr,
b721420e 6663 &raid5_group_thread_cnt.attr,
d592a996 6664 &raid5_skip_copy.attr,
d06f191f 6665 &raid5_rmw_level.attr,
2c7da14b 6666 &r5c_journal_mode.attr,
a596d086 6667 &ppl_write_hint.attr,
3f294f4f
N		 6668 NULL,
6669};
007583c9
N		 6670static struct attribute_group raid5_attrs_group = {
6671 .name = NULL,
6672 .attrs = raid5_attrs,
3f294f4f
N		 6673};
6674
60aaf933 6675static int alloc_thread_groups(struct r5conf *conf, int cnt,
6676 int *group_cnt,
6677 int *worker_cnt_per_group,
6678 struct r5worker_group **worker_groups)
851c30c9 6679{
566c09c5 6680 int i, j, k;
851c30c9
SL		 6681 ssize_t size;
6682 struct r5worker *workers;
6683
60aaf933 6684 *worker_cnt_per_group = cnt;
851c30c9 6685 if (cnt == 0) {
60aaf933 6686 *group_cnt = 0;
6687 *worker_groups = NULL;
851c30c9
SL		 6688 return 0;
6689 }
60aaf933 6690 *group_cnt = num_possible_nodes();
851c30c9 6691 size = sizeof(struct r5worker) * cnt;
6396bb22
KC		 6692 workers = kcalloc(size, *group_cnt, GFP_NOIO);
6693 *worker_groups = kcalloc(*group_cnt, sizeof(struct r5worker_group),
6694 GFP_NOIO);
60aaf933 6695 if (!*worker_groups || !workers) {
851c30c9 6696 kfree(workers);
60aaf933 6697 kfree(*worker_groups);
851c30c9
SL		 6698 return -ENOMEM;
6699 }
6700
60aaf933 6701 for (i = 0; i < *group_cnt; i++) {
851c30c9
SL		 6702 struct r5worker_group *group;
6703
0c775d52 6704 group = &(*worker_groups)[i];
851c30c9 6705 INIT_LIST_HEAD(&group->handle_list);
535ae4eb 6706 INIT_LIST_HEAD(&group->loprio_list);
851c30c9
SL		 6707 group->conf = conf;
6708 group->workers = workers + i * cnt;
6709
6710 for (j = 0; j < cnt; j++) {
566c09c5
SL		 6711 struct r5worker *worker = group->workers + j;
6712 worker->group = group;
6713 INIT_WORK(&worker->work, raid5_do_work);
6714
6715 for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++)
6716 INIT_LIST_HEAD(worker->temp_inactive_list + k);
851c30c9
SL		 6717 }
6718 }
6719
6720 return 0;
6721}
6722
6723static void free_thread_groups(struct r5conf *conf)
6724{
6725 if (conf->worker_groups)
6726 kfree(conf->worker_groups[0].workers);
6727 kfree(conf->worker_groups);
6728 conf->worker_groups = NULL;
6729}
6730
80c3a6ce 6731static sector_t
fd01b88c 6732raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks)
80c3a6ce 6733{
d1688a6d 6734 struct r5conf *conf = mddev->private;
80c3a6ce
DW		 6735
6736 if (!sectors)
6737 sectors = mddev->dev_sectors;
5e5e3e78 6738 if (!raid_disks)
7ec05478 6739 /* size is defined by the smallest of previous and new size */
5e5e3e78 6740 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
80c3a6ce 6741
3cb5edf4
N		 6742 sectors &= ~((sector_t)conf->chunk_sectors - 1);
6743 sectors &= ~((sector_t)conf->prev_chunk_sectors - 1);
80c3a6ce
DW		 6744 return sectors * (raid_disks - conf->max_degraded);
6745}
6746
789b5e03
ON		 6747static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
6748{
6749 safe_put_page(percpu->spare_page);
789b5e03 6750 percpu->spare_page = NULL;
b330e6a4 6751 kvfree(percpu->scribble);
789b5e03
ON		 6752 percpu->scribble = NULL;
6753}
6754
6755static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
6756{
b330e6a4 6757 if (conf->level == 6 && !percpu->spare_page) {
789b5e03 6758 percpu->spare_page = alloc_page(GFP_KERNEL);
b330e6a4
KO		 6759 if (!percpu->spare_page)
6760 return -ENOMEM;
6761 }
6762
6763 if (scribble_alloc(percpu,
6764 max(conf->raid_disks,
6765 conf->previous_raid_disks),
6766 max(conf->chunk_sectors,
6767 conf->prev_chunk_sectors)
6768 / STRIPE_SECTORS,
6769 GFP_KERNEL)) {
789b5e03
ON		 6770 free_scratch_buffer(conf, percpu);
6771 return -ENOMEM;
6772 }
6773
6774 return 0;
6775}
6776
29c6d1bb 6777static int raid456_cpu_dead(unsigned int cpu, struct hlist_node *node)
36d1c647 6778{
29c6d1bb
SAS		 6779 struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node);
6780
6781 free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
6782 return 0;
6783}
36d1c647 6784
29c6d1bb
SAS		 6785static void raid5_free_percpu(struct r5conf *conf)
6786{
36d1c647
DW		 6787 if (!conf->percpu)
6788 return;
6789
29c6d1bb 6790 cpuhp_state_remove_instance(CPUHP_MD_RAID5_PREPARE, &conf->node);
36d1c647
DW		 6791 free_percpu(conf->percpu);
6792}
6793
d1688a6d 6794static void free_conf(struct r5conf *conf)
95fc17aa 6795{
d7bd398e
SL		 6796 int i;
6797
ff875738
AP		 6798 log_exit(conf);
6799
565e0450 6800 unregister_shrinker(&conf->shrinker);
851c30c9 6801 free_thread_groups(conf);
95fc17aa 6802 shrink_stripes(conf);
36d1c647 6803 raid5_free_percpu(conf);
d7bd398e
SL		 6804 for (i = 0; i < conf->pool_size; i++)
6805 if (conf->disks[i].extra_page)
6806 put_page(conf->disks[i].extra_page);
95fc17aa 6807 kfree(conf->disks);
afeee514 6808 bioset_exit(&conf->bio_split);
95fc17aa 6809 kfree(conf->stripe_hashtbl);
aaf9f12e 6810 kfree(conf->pending_data);
95fc17aa
DW		 6811 kfree(conf);
6812}
6813
29c6d1bb 6814static int raid456_cpu_up_prepare(unsigned int cpu, struct hlist_node *node)
36d1c647 6815{
29c6d1bb 6816 struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node);
36d1c647
DW		 6817 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
6818
29c6d1bb 6819 if (alloc_scratch_buffer(conf, percpu)) {
cc6167b4
N		 6820 pr_warn("%s: failed memory allocation for cpu%u\n",
6821 __func__, cpu);
29c6d1bb 6822 return -ENOMEM;
36d1c647 6823 }
29c6d1bb 6824 return 0;
36d1c647 6825}
36d1c647 6826
d1688a6d 6827static int raid5_alloc_percpu(struct r5conf *conf)
36d1c647 6828{
789b5e03 6829 int err = 0;
36d1c647 6830
789b5e03
ON		 6831 conf->percpu = alloc_percpu(struct raid5_percpu);
6832 if (!conf->percpu)
36d1c647 6833 return -ENOMEM;
789b5e03 6834
29c6d1bb 6835 err = cpuhp_state_add_instance(CPUHP_MD_RAID5_PREPARE, &conf->node);
27a353c0
SL		 6836 if (!err) {
6837 conf->scribble_disks = max(conf->raid_disks,
6838 conf->previous_raid_disks);
6839 conf->scribble_sectors = max(conf->chunk_sectors,
6840 conf->prev_chunk_sectors);
6841 }
36d1c647
DW		 6842 return err;
6843}
6844
edbe83ab
N		 6845static unsigned long raid5_cache_scan(struct shrinker *shrink,
6846 struct shrink_control *sc)
6847{
6848 struct r5conf *conf = container_of(shrink, struct r5conf, shrinker);
2d5b569b
N		 6849 unsigned long ret = SHRINK_STOP;
6850
6851 if (mutex_trylock(&conf->cache_size_mutex)) {
6852 ret= 0;
49895bcc
N		 6853 while (ret < sc->nr_to_scan &&
6854 conf->max_nr_stripes > conf->min_nr_stripes) {
2d5b569b
N		 6855 if (drop_one_stripe(conf) == 0) {
6856 ret = SHRINK_STOP;
6857 break;
6858 }
6859 ret++;
6860 }
6861 mutex_unlock(&conf->cache_size_mutex);
edbe83ab
N		 6862 }
6863 return ret;
6864}
6865
6866static unsigned long raid5_cache_count(struct shrinker *shrink,
6867 struct shrink_control *sc)
6868{
6869 struct r5conf *conf = container_of(shrink, struct r5conf, shrinker);
6870
6871 if (conf->max_nr_stripes < conf->min_nr_stripes)
6872 /* unlikely, but not impossible */
6873 return 0;
6874 return conf->max_nr_stripes - conf->min_nr_stripes;
6875}
6876
d1688a6d 6877static struct r5conf *setup_conf(struct mddev *mddev)
1da177e4 6878{
d1688a6d 6879 struct r5conf *conf;
5e5e3e78 6880 int raid_disk, memory, max_disks;
3cb03002 6881 struct md_rdev *rdev;
1da177e4 6882 struct disk_info *disk;
0232605d 6883 char pers_name[6];
566c09c5 6884 int i;
60aaf933 6885 int group_cnt, worker_cnt_per_group;
6886 struct r5worker_group *new_group;
afeee514 6887 int ret;
1da177e4 6888
91adb564
N		 6889 if (mddev->new_level != 5
6890 && mddev->new_level != 4
6891 && mddev->new_level != 6) {
cc6167b4
N		 6892 pr_warn("md/raid:%s: raid level not set to 4/5/6 (%d)\n",
6893 mdname(mddev), mddev->new_level);
91adb564 6894 return ERR_PTR(-EIO);
1da177e4 6895 }
91adb564
N		 6896 if ((mddev->new_level == 5
6897 && !algorithm_valid_raid5(mddev->new_layout)) ||
6898 (mddev->new_level == 6
6899 && !algorithm_valid_raid6(mddev->new_layout))) {
cc6167b4
N		 6900 pr_warn("md/raid:%s: layout %d not supported\n",
6901 mdname(mddev), mddev->new_layout);
91adb564 6902 return ERR_PTR(-EIO);
99c0fb5f 6903 }
91adb564 6904 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
cc6167b4
N		 6905 pr_warn("md/raid:%s: not enough configured devices (%d, minimum 4)\n",
6906 mdname(mddev), mddev->raid_disks);
91adb564 6907 return ERR_PTR(-EINVAL);
4bbf3771
N		 6908 }
6909
664e7c41
AN		 6910 if (!mddev->new_chunk_sectors ||
6911 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
6912 !is_power_of_2(mddev->new_chunk_sectors)) {
cc6167b4
N		 6913 pr_warn("md/raid:%s: invalid chunk size %d\n",
6914 mdname(mddev), mddev->new_chunk_sectors << 9);
91adb564 6915 return ERR_PTR(-EINVAL);
f6705578
N		 6916 }
6917
d1688a6d 6918 conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL);
91adb564 6919 if (conf == NULL)
1da177e4 6920 goto abort;
aaf9f12e
SL		 6921 INIT_LIST_HEAD(&conf->free_list);
6922 INIT_LIST_HEAD(&conf->pending_list);
6396bb22
KC		 6923 conf->pending_data = kcalloc(PENDING_IO_MAX,
6924 sizeof(struct r5pending_data),
6925 GFP_KERNEL);
aaf9f12e
SL		 6926 if (!conf->pending_data)
6927 goto abort;
6928 for (i = 0; i < PENDING_IO_MAX; i++)
6929 list_add(&conf->pending_data[i].sibling, &conf->free_list);
851c30c9 6930 /* Don't enable multi-threading by default*/
60aaf933 6931 if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group,
6932 &new_group)) {
6933 conf->group_cnt = group_cnt;
6934 conf->worker_cnt_per_group = worker_cnt_per_group;
6935 conf->worker_groups = new_group;
6936 } else
851c30c9 6937 goto abort;
f5efd45a 6938 spin_lock_init(&conf->device_lock);
c46501b2 6939 seqcount_init(&conf->gen_lock);
2d5b569b 6940 mutex_init(&conf->cache_size_mutex);
b1b46486 6941 init_waitqueue_head(&conf->wait_for_quiescent);
6ab2a4b8 6942 init_waitqueue_head(&conf->wait_for_stripe);
f5efd45a
DW		 6943 init_waitqueue_head(&conf->wait_for_overlap);
6944 INIT_LIST_HEAD(&conf->handle_list);
535ae4eb 6945 INIT_LIST_HEAD(&conf->loprio_list);
f5efd45a
DW		 6946 INIT_LIST_HEAD(&conf->hold_list);
6947 INIT_LIST_HEAD(&conf->delayed_list);
6948 INIT_LIST_HEAD(&conf->bitmap_list);
773ca82f 6949 init_llist_head(&conf->released_stripes);
f5efd45a
DW		 6950 atomic_set(&conf->active_stripes, 0);
6951 atomic_set(&conf->preread_active_stripes, 0);
6952 atomic_set(&conf->active_aligned_reads, 0);
765d704d
SL		 6953 spin_lock_init(&conf->pending_bios_lock);
6954 conf->batch_bio_dispatch = true;
6955 rdev_for_each(rdev, mddev) {
6956 if (test_bit(Journal, &rdev->flags))
6957 continue;
6958 if (blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
6959 conf->batch_bio_dispatch = false;
6960 break;
6961 }
6962 }
6963
f5efd45a 6964 conf->bypass_threshold = BYPASS_THRESHOLD;
d890fa2b 6965 conf->recovery_disabled = mddev->recovery_disabled - 1;
91adb564
N		 6966
6967 conf->raid_disks = mddev->raid_disks;
6968 if (mddev->reshape_position == MaxSector)
6969 conf->previous_raid_disks = mddev->raid_disks;
6970 else
f6705578 6971 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
5e5e3e78 6972 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
f6705578 6973
6396bb22 6974 conf->disks = kcalloc(max_disks, sizeof(struct disk_info),
b55e6bfc 6975 GFP_KERNEL);
d7bd398e 6976
b55e6bfc
N		 6977 if (!conf->disks)
6978 goto abort;
9ffae0cf 6979
d7bd398e
SL		 6980 for (i = 0; i < max_disks; i++) {
6981 conf->disks[i].extra_page = alloc_page(GFP_KERNEL);
6982 if (!conf->disks[i].extra_page)
6983 goto abort;
6984 }
6985
afeee514
KO		 6986 ret = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0);
6987 if (ret)
dd7a8f5d 6988 goto abort;
1da177e4
LT		 6989 conf->mddev = mddev;
6990
fccddba0 6991 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
1da177e4 6992 goto abort;
1da177e4 6993
566c09c5
SL		 6994 /* We init hash_locks[0] separately to that it can be used
6995 * as the reference lock in the spin_lock_nest_lock() call
6996 * in lock_all_device_hash_locks_irq in order to convince
6997 * lockdep that we know what we are doing.
6998 */
6999 spin_lock_init(conf->hash_locks);
7000 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
7001 spin_lock_init(conf->hash_locks + i);
7002
7003 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
7004 INIT_LIST_HEAD(conf->inactive_list + i);
7005
7006 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
7007 INIT_LIST_HEAD(conf->temp_inactive_list + i);
7008
1e6d690b
SL		 7009 atomic_set(&conf->r5c_cached_full_stripes, 0);
7010 INIT_LIST_HEAD(&conf->r5c_full_stripe_list);
7011 atomic_set(&conf->r5c_cached_partial_stripes, 0);
7012 INIT_LIST_HEAD(&conf->r5c_partial_stripe_list);
e33fbb9c
SL		 7013 atomic_set(&conf->r5c_flushing_full_stripes, 0);
7014 atomic_set(&conf->r5c_flushing_partial_stripes, 0);
1e6d690b 7015
36d1c647 7016 conf->level = mddev->new_level;
46d5b785 7017 conf->chunk_sectors = mddev->new_chunk_sectors;
36d1c647
DW		 7018 if (raid5_alloc_percpu(conf) != 0)
7019 goto abort;
7020
0c55e022 7021 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
1da177e4 7022
dafb20fa 7023 rdev_for_each(rdev, mddev) {
1da177e4 7024 raid_disk = rdev->raid_disk;
5e5e3e78 7025 if (raid_disk >= max_disks
f2076e7d 7026 || raid_disk < 0 || test_bit(Journal, &rdev->flags))
1da177e4
LT		 7027 continue;
7028 disk = conf->disks + raid_disk;
7029
17045f52
N		 7030 if (test_bit(Replacement, &rdev->flags)) {
7031 if (disk->replacement)
7032 goto abort;
7033 disk->replacement = rdev;
7034 } else {
7035 if (disk->rdev)
7036 goto abort;
7037 disk->rdev = rdev;
7038 }
1da177e4 7039
b2d444d7 7040 if (test_bit(In_sync, &rdev->flags)) {
1da177e4 7041 char b[BDEVNAME_SIZE];
cc6167b4
N		 7042 pr_info("md/raid:%s: device %s operational as raid disk %d\n",
7043 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
d6b212f4 7044 } else if (rdev->saved_raid_disk != raid_disk)
8c2e870a
NB		 7045 /* Cannot rely on bitmap to complete recovery */
7046 conf->fullsync = 1;
1da177e4
LT		 7047 }
7048
91adb564 7049 conf->level = mddev->new_level;
584acdd4 7050 if (conf->level == 6) {
16a53ecc 7051 conf->max_degraded = 2;
584acdd4
MS		 7052 if (raid6_call.xor_syndrome)
7053 conf->rmw_level = PARITY_ENABLE_RMW;
7054 else
7055 conf->rmw_level = PARITY_DISABLE_RMW;
7056 } else {
16a53ecc 7057 conf->max_degraded = 1;
584acdd4
MS		 7058 conf->rmw_level = PARITY_ENABLE_RMW;
7059 }
91adb564 7060 conf->algorithm = mddev->new_layout;
fef9c61f 7061 conf->reshape_progress = mddev->reshape_position;
e183eaed 7062 if (conf->reshape_progress != MaxSector) {
09c9e5fa 7063 conf->prev_chunk_sectors = mddev->chunk_sectors;
e183eaed 7064 conf->prev_algo = mddev->layout;
5cac6bcb
N		 7065 } else {
7066 conf->prev_chunk_sectors = conf->chunk_sectors;
7067 conf->prev_algo = conf->algorithm;
e183eaed 7068 }
1da177e4 7069
edbe83ab 7070 conf->min_nr_stripes = NR_STRIPES;
ad5b0f76
SL		 7071 if (mddev->reshape_position != MaxSector) {
7072 int stripes = max_t(int,
7073 ((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4,
7074 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4);
7075 conf->min_nr_stripes = max(NR_STRIPES, stripes);
7076 if (conf->min_nr_stripes != NR_STRIPES)
cc6167b4 7077 pr_info("md/raid:%s: force stripe size %d for reshape\n",
ad5b0f76
SL		 7078 mdname(mddev), conf->min_nr_stripes);
7079 }
edbe83ab 7080 memory = conf->min_nr_stripes * (sizeof(struct stripe_head) +
5e5e3e78 7081 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
4bda556a 7082 atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS);
edbe83ab 7083 if (grow_stripes(conf, conf->min_nr_stripes)) {
cc6167b4
N		 7084 pr_warn("md/raid:%s: couldn't allocate %dkB for buffers\n",
7085 mdname(mddev), memory);
91adb564
N		 7086 goto abort;
7087 } else
cc6167b4 7088 pr_debug("md/raid:%s: allocated %dkB\n", mdname(mddev), memory);
edbe83ab
N		 7089 /*
7090 * Losing a stripe head costs more than the time to refill it,
7091 * it reduces the queue depth and so can hurt throughput.
7092 * So set it rather large, scaled by number of devices.
7093 */
7094 conf->shrinker.seeks = DEFAULT_SEEKS * conf->raid_disks * 4;
7095 conf->shrinker.scan_objects = raid5_cache_scan;
7096 conf->shrinker.count_objects = raid5_cache_count;
7097 conf->shrinker.batch = 128;
7098 conf->shrinker.flags = 0;
6a0f53ff 7099 if (register_shrinker(&conf->shrinker)) {
cc6167b4
N		 7100 pr_warn("md/raid:%s: couldn't register shrinker.\n",
7101 mdname(mddev));
6a0f53ff
CY		 7102 goto abort;
7103 }
1da177e4 7104
0232605d
N		 7105 sprintf(pers_name, "raid%d", mddev->new_level);
7106 conf->thread = md_register_thread(raid5d, mddev, pers_name);
91adb564 7107 if (!conf->thread) {
cc6167b4
N		 7108 pr_warn("md/raid:%s: couldn't allocate thread.\n",
7109 mdname(mddev));
16a53ecc
N		 7110 goto abort;
7111 }
91adb564
N		 7112
7113 return conf;
7114
7115 abort:
7116 if (conf) {
95fc17aa 7117 free_conf(conf);
91adb564
N		 7118 return ERR_PTR(-EIO);
7119 } else
7120 return ERR_PTR(-ENOMEM);
7121}
7122
c148ffdc
N		 7123static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
7124{
7125 switch (algo) {
7126 case ALGORITHM_PARITY_0:
7127 if (raid_disk < max_degraded)
7128 return 1;
7129 break;
7130 case ALGORITHM_PARITY_N:
7131 if (raid_disk >= raid_disks - max_degraded)
7132 return 1;
7133 break;
7134 case ALGORITHM_PARITY_0_6:
f72ffdd6 7135 if (raid_disk == 0 ||
c148ffdc
N		 7136 raid_disk == raid_disks - 1)
7137 return 1;
7138 break;
7139 case ALGORITHM_LEFT_ASYMMETRIC_6:
7140 case ALGORITHM_RIGHT_ASYMMETRIC_6:
7141 case ALGORITHM_LEFT_SYMMETRIC_6:
7142 case ALGORITHM_RIGHT_SYMMETRIC_6:
7143 if (raid_disk == raid_disks - 1)
7144 return 1;
7145 }
7146 return 0;
7147}
7148
849674e4 7149static int raid5_run(struct mddev *mddev)
91adb564 7150{
d1688a6d 7151 struct r5conf *conf;
9f7c2220 7152 int working_disks = 0;
c148ffdc 7153 int dirty_parity_disks = 0;
3cb03002 7154 struct md_rdev *rdev;
713cf5a6 7155 struct md_rdev *journal_dev = NULL;
c148ffdc 7156 sector_t reshape_offset = 0;
17045f52 7157 int i;
b5254dd5
N		 7158 long long min_offset_diff = 0;
7159 int first = 1;
91adb564 7160
a415c0f1
N		 7161 if (mddev_init_writes_pending(mddev) < 0)
7162 return -ENOMEM;
7163
8c6ac868 7164 if (mddev->recovery_cp != MaxSector)
cc6167b4
N		 7165 pr_notice("md/raid:%s: not clean -- starting background reconstruction\n",
7166 mdname(mddev));
b5254dd5
N		 7167
7168 rdev_for_each(rdev, mddev) {
7169 long long diff;
713cf5a6 7170
f2076e7d 7171 if (test_bit(Journal, &rdev->flags)) {
713cf5a6 7172 journal_dev = rdev;
f2076e7d
SL		 7173 continue;
7174 }
b5254dd5
N		 7175 if (rdev->raid_disk < 0)
7176 continue;
7177 diff = (rdev->new_data_offset - rdev->data_offset);
7178 if (first) {
7179 min_offset_diff = diff;
7180 first = 0;
7181 } else if (mddev->reshape_backwards &&
7182 diff < min_offset_diff)
7183 min_offset_diff = diff;
7184 else if (!mddev->reshape_backwards &&
7185 diff > min_offset_diff)
7186 min_offset_diff = diff;
7187 }
7188
230b55fa
N		 7189 if ((test_bit(MD_HAS_JOURNAL, &mddev->flags) || journal_dev) &&
7190 (mddev->bitmap_info.offset || mddev->bitmap_info.file)) {
7191 pr_notice("md/raid:%s: array cannot have both journal and bitmap\n",
7192 mdname(mddev));
7193 return -EINVAL;
7194 }
7195
91adb564
N		 7196 if (mddev->reshape_position != MaxSector) {
7197 /* Check that we can continue the reshape.
b5254dd5
N		 7198 * Difficulties arise if the stripe we would write to
7199 * next is at or after the stripe we would read from next.
7200 * For a reshape that changes the number of devices, this
7201 * is only possible for a very short time, and mdadm makes
7202 * sure that time appears to have past before assembling
7203 * the array. So we fail if that time hasn't passed.
7204 * For a reshape that keeps the number of devices the same
7205 * mdadm must be monitoring the reshape can keeping the
7206 * critical areas read-only and backed up. It will start
7207 * the array in read-only mode, so we check for that.
91adb564
N		 7208 */
7209 sector_t here_new, here_old;
7210 int old_disks;
18b00334 7211 int max_degraded = (mddev->level == 6 ? 2 : 1);
05256d98
N		 7212 int chunk_sectors;
7213 int new_data_disks;
91adb564 7214
713cf5a6 7215 if (journal_dev) {
cc6167b4
N		 7216 pr_warn("md/raid:%s: don't support reshape with journal - aborting.\n",
7217 mdname(mddev));
713cf5a6
SL		 7218 return -EINVAL;
7219 }
7220
88ce4930 7221 if (mddev->new_level != mddev->level) {
cc6167b4
N		 7222 pr_warn("md/raid:%s: unsupported reshape required - aborting.\n",
7223 mdname(mddev));
91adb564
N		 7224 return -EINVAL;
7225 }
91adb564
N		 7226 old_disks = mddev->raid_disks - mddev->delta_disks;
7227 /* reshape_position must be on a new-stripe boundary, and one
7228 * further up in new geometry must map after here in old
7229 * geometry.
05256d98
N		 7230 * If the chunk sizes are different, then as we perform reshape
7231 * in units of the largest of the two, reshape_position needs
7232 * be a multiple of the largest chunk size times new data disks.
91adb564
N		 7233 */
7234 here_new = mddev->reshape_position;
05256d98
N		 7235 chunk_sectors = max(mddev->chunk_sectors, mddev->new_chunk_sectors);
7236 new_data_disks = mddev->raid_disks - max_degraded;
7237 if (sector_div(here_new, chunk_sectors * new_data_disks)) {
cc6167b4
N		 7238 pr_warn("md/raid:%s: reshape_position not on a stripe boundary\n",
7239 mdname(mddev));
91adb564
N		 7240 return -EINVAL;
7241 }
05256d98 7242 reshape_offset = here_new * chunk_sectors;
91adb564
N		 7243 /* here_new is the stripe we will write to */
7244 here_old = mddev->reshape_position;
05256d98 7245 sector_div(here_old, chunk_sectors * (old_disks-max_degraded));
91adb564
N		 7246 /* here_old is the first stripe that we might need to read
7247 * from */
67ac6011
N		 7248 if (mddev->delta_disks == 0) {
7249 /* We cannot be sure it is safe to start an in-place
b5254dd5 7250 * reshape. It is only safe if user-space is monitoring
67ac6011
N		 7251 * and taking constant backups.
7252 * mdadm always starts a situation like this in
7253 * readonly mode so it can take control before
7254 * allowing any writes. So just check for that.
7255 */
b5254dd5
N		 7256 if (abs(min_offset_diff) >= mddev->chunk_sectors &&
7257 abs(min_offset_diff) >= mddev->new_chunk_sectors)
7258 /* not really in-place - so OK */;
7259 else if (mddev->ro == 0) {
cc6167b4
N		 7260 pr_warn("md/raid:%s: in-place reshape must be started in read-only mode - aborting\n",
7261 mdname(mddev));
67ac6011
N		 7262 return -EINVAL;
7263 }
2c810cdd 7264 } else if (mddev->reshape_backwards
05256d98
N		 7265 ? (here_new * chunk_sectors + min_offset_diff <=
7266 here_old * chunk_sectors)
7267 : (here_new * chunk_sectors >=
7268 here_old * chunk_sectors + (-min_offset_diff))) {
91adb564 7269 /* Reading from the same stripe as writing to - bad */
cc6167b4
N		 7270 pr_warn("md/raid:%s: reshape_position too early for auto-recovery - aborting.\n",
7271 mdname(mddev));
91adb564
N		 7272 return -EINVAL;
7273 }
cc6167b4 7274 pr_debug("md/raid:%s: reshape will continue\n", mdname(mddev));
91adb564
N		 7275 /* OK, we should be able to continue; */
7276 } else {
7277 BUG_ON(mddev->level != mddev->new_level);
7278 BUG_ON(mddev->layout != mddev->new_layout);
664e7c41 7279 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
91adb564 7280 BUG_ON(mddev->delta_disks != 0);
1da177e4 7281 }
91adb564 7282
3418d036
AP		 7283 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) &&
7284 test_bit(MD_HAS_PPL, &mddev->flags)) {
7285 pr_warn("md/raid:%s: using journal device and PPL not allowed - disabling PPL\n",
7286 mdname(mddev));
7287 clear_bit(MD_HAS_PPL, &mddev->flags);
ddc08823 7288 clear_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags);
3418d036
AP		 7289 }
7290
245f46c2
N		 7291 if (mddev->private == NULL)
7292 conf = setup_conf(mddev);
7293 else
7294 conf = mddev->private;
7295
91adb564
N		 7296 if (IS_ERR(conf))
7297 return PTR_ERR(conf);
7298
486b0f7b
SL		 7299 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
7300 if (!journal_dev) {
cc6167b4
N		 7301 pr_warn("md/raid:%s: journal disk is missing, force array readonly\n",
7302 mdname(mddev));
486b0f7b
SL		 7303 mddev->ro = 1;
7304 set_disk_ro(mddev->gendisk, 1);
7305 } else if (mddev->recovery_cp == MaxSector)
7306 set_bit(MD_JOURNAL_CLEAN, &mddev->flags);
7dde2ad3
SL		 7307 }
7308
b5254dd5 7309 conf->min_offset_diff = min_offset_diff;
91adb564
N		 7310 mddev->thread = conf->thread;
7311 conf->thread = NULL;
7312 mddev->private = conf;
7313
17045f52
N		 7314 for (i = 0; i < conf->raid_disks && conf->previous_raid_disks;
7315 i++) {
7316 rdev = conf->disks[i].rdev;
7317 if (!rdev && conf->disks[i].replacement) {
7318 /* The replacement is all we have yet */
7319 rdev = conf->disks[i].replacement;
7320 conf->disks[i].replacement = NULL;
7321 clear_bit(Replacement, &rdev->flags);
7322 conf->disks[i].rdev = rdev;
7323 }
7324 if (!rdev)
c148ffdc 7325 continue;
17045f52
N		 7326 if (conf->disks[i].replacement &&
7327 conf->reshape_progress != MaxSector) {
7328 /* replacements and reshape simply do not mix. */
cc6167b4 7329 pr_warn("md: cannot handle concurrent replacement and reshape.\n");
17045f52
N		 7330 goto abort;
7331 }
2f115882 7332 if (test_bit(In_sync, &rdev->flags)) {
91adb564 7333 working_disks++;
2f115882
N		 7334 continue;
7335 }
c148ffdc
N		 7336 /* This disc is not fully in-sync. However if it
7337 * just stored parity (beyond the recovery_offset),
7338 * when we don't need to be concerned about the
7339 * array being dirty.
7340 * When reshape goes 'backwards', we never have
7341 * partially completed devices, so we only need
7342 * to worry about reshape going forwards.
7343 */
7344 /* Hack because v0.91 doesn't store recovery_offset properly. */
7345 if (mddev->major_version == 0 &&
7346 mddev->minor_version > 90)
7347 rdev->recovery_offset = reshape_offset;
5026d7a9 7348
c148ffdc
N		 7349 if (rdev->recovery_offset < reshape_offset) {
7350 /* We need to check old and new layout */
7351 if (!only_parity(rdev->raid_disk,
7352 conf->algorithm,
7353 conf->raid_disks,
7354 conf->max_degraded))
7355 continue;
7356 }
7357 if (!only_parity(rdev->raid_disk,
7358 conf->prev_algo,
7359 conf->previous_raid_disks,
7360 conf->max_degraded))
7361 continue;
7362 dirty_parity_disks++;
7363 }
91adb564 7364
17045f52
N		 7365 /*
7366 * 0 for a fully functional array, 1 or 2 for a degraded array.
7367 */
2e38a37f 7368 mddev->degraded = raid5_calc_degraded(conf);
91adb564 7369
674806d6 7370 if (has_failed(conf)) {
cc6167b4 7371 pr_crit("md/raid:%s: not enough operational devices (%d/%d failed)\n",
02c2de8c 7372 mdname(mddev), mddev->degraded, conf->raid_disks);
1da177e4
LT		 7373 goto abort;
7374 }
7375
91adb564 7376 /* device size must be a multiple of chunk size */
9d8f0363 7377 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
91adb564
N		 7378 mddev->resync_max_sectors = mddev->dev_sectors;
7379
c148ffdc 7380 if (mddev->degraded > dirty_parity_disks &&
1da177e4 7381 mddev->recovery_cp != MaxSector) {
4536bf9b
AP		 7382 if (test_bit(MD_HAS_PPL, &mddev->flags))
7383 pr_crit("md/raid:%s: starting dirty degraded array with PPL.\n",
7384 mdname(mddev));
7385 else if (mddev->ok_start_degraded)
cc6167b4
N		 7386 pr_crit("md/raid:%s: starting dirty degraded array - data corruption possible.\n",
7387 mdname(mddev));
6ff8d8ec 7388 else {
cc6167b4
N		 7389 pr_crit("md/raid:%s: cannot start dirty degraded array.\n",
7390 mdname(mddev));
6ff8d8ec
N		 7391 goto abort;
7392 }
1da177e4
LT		 7393 }
7394
cc6167b4
N		 7395 pr_info("md/raid:%s: raid level %d active with %d out of %d devices, algorithm %d\n",
7396 mdname(mddev), conf->level,
7397 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
7398 mddev->new_layout);
1da177e4
LT		 7399
7400 print_raid5_conf(conf);
7401
fef9c61f 7402 if (conf->reshape_progress != MaxSector) {
fef9c61f 7403 conf->reshape_safe = conf->reshape_progress;
f6705578
N		 7404 atomic_set(&conf->reshape_stripes, 0);
7405 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7406 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7407 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7408 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7409 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 7410 "reshape");
e406f12d
AP		 7411 if (!mddev->sync_thread)
7412 goto abort;
f6705578
N		 7413 }
7414
1da177e4 7415 /* Ok, everything is just fine now */
a64c876f
N		 7416 if (mddev->to_remove == &raid5_attrs_group)
7417 mddev->to_remove = NULL;
00bcb4ac
N		 7418 else if (mddev->kobj.sd &&
7419 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
cc6167b4
N		 7420 pr_warn("raid5: failed to create sysfs attributes for %s\n",
7421 mdname(mddev));
4a5add49 7422 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
7a5febe9 7423
4a5add49 7424 if (mddev->queue) {
9f7c2220 7425 int chunk_size;
4a5add49
N		 7426 /* read-ahead size must cover two whole stripes, which
7427 * is 2 * (datadisks) * chunksize where 'n' is the
7428 * number of raid devices
7429 */
7430 int data_disks = conf->previous_raid_disks - conf->max_degraded;
7431 int stripe = data_disks *
7432 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
dc3b17cc
JK		 7433 if (mddev->queue->backing_dev_info->ra_pages < 2 * stripe)
7434 mddev->queue->backing_dev_info->ra_pages = 2 * stripe;
91adb564 7435
9f7c2220
N		 7436 chunk_size = mddev->chunk_sectors << 9;
7437 blk_queue_io_min(mddev->queue, chunk_size);
7438 blk_queue_io_opt(mddev->queue, chunk_size *
7439 (conf->raid_disks - conf->max_degraded));
c78afc62 7440 mddev->queue->limits.raid_partial_stripes_expensive = 1;
620125f2
SL		 7441 /*
7442 * We can only discard a whole stripe. It doesn't make sense to
7443 * discard data disk but write parity disk
7444 */
7445 stripe = stripe * PAGE_SIZE;
4ac6875e
N		 7446 /* Round up to power of 2, as discard handling
7447 * currently assumes that */
7448 while ((stripe-1) & stripe)
7449 stripe = (stripe | (stripe-1)) + 1;
620125f2
SL		 7450 mddev->queue->limits.discard_alignment = stripe;
7451 mddev->queue->limits.discard_granularity = stripe;
e8d7c332 7452
5026d7a9 7453 blk_queue_max_write_same_sectors(mddev->queue, 0);
3deff1a7 7454 blk_queue_max_write_zeroes_sectors(mddev->queue, 0);
5026d7a9 7455
05616be5 7456 rdev_for_each(rdev, mddev) {
9f7c2220
N		 7457 disk_stack_limits(mddev->gendisk, rdev->bdev,
7458 rdev->data_offset << 9);
05616be5
N		 7459 disk_stack_limits(mddev->gendisk, rdev->bdev,
7460 rdev->new_data_offset << 9);
7461 }
620125f2 7462
48920ff2
CH		 7463 /*
7464 * zeroing is required, otherwise data
7465 * could be lost. Consider a scenario: discard a stripe
7466 * (the stripe could be inconsistent if
7467 * discard_zeroes_data is 0); write one disk of the
7468 * stripe (the stripe could be inconsistent again
7469 * depending on which disks are used to calculate
7470 * parity); the disk is broken; The stripe data of this
7471 * disk is lost.
7472 *
7473 * We only allow DISCARD if the sysadmin has confirmed that
7474 * only safe devices are in use by setting a module parameter.
7475 * A better idea might be to turn DISCARD into WRITE_ZEROES
7476 * requests, as that is required to be safe.
7477 */
7478 if (devices_handle_discard_safely &&
e7597e69
JS		 7479 mddev->queue->limits.max_discard_sectors >= (stripe >> 9) &&
7480 mddev->queue->limits.discard_granularity >= stripe)
8b904b5b 7481 blk_queue_flag_set(QUEUE_FLAG_DISCARD,
620125f2
SL		 7482 mddev->queue);
7483 else
8b904b5b 7484 blk_queue_flag_clear(QUEUE_FLAG_DISCARD,
620125f2 7485 mddev->queue);
1dffdddd
SL		 7486
7487 blk_queue_max_hw_sectors(mddev->queue, UINT_MAX);
9f7c2220 7488 }
23032a0e 7489
845b9e22 7490 if (log_init(conf, journal_dev, raid5_has_ppl(conf)))
ff875738 7491 goto abort;
5c7e81c3 7492
1da177e4
LT		 7493 return 0;
7494abort:
01f96c0a 7495 md_unregister_thread(&mddev->thread);
e4f869d9
N		 7496 print_raid5_conf(conf);
7497 free_conf(conf);
1da177e4 7498 mddev->private = NULL;
cc6167b4 7499 pr_warn("md/raid:%s: failed to run raid set.\n", mdname(mddev));
1da177e4
LT		 7500 return -EIO;
7501}
7502
afa0f557 7503static void raid5_free(struct mddev *mddev, void *priv)
1da177e4 7504{
afa0f557 7505 struct r5conf *conf = priv;
1da177e4 7506
95fc17aa 7507 free_conf(conf);
a64c876f 7508 mddev->to_remove = &raid5_attrs_group;
1da177e4
LT		 7509}
7510
849674e4 7511static void raid5_status(struct seq_file *seq, struct mddev *mddev)
1da177e4 7512{
d1688a6d 7513 struct r5conf *conf = mddev->private;
1da177e4
LT		 7514 int i;
7515
9d8f0363 7516 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
3cb5edf4 7517 conf->chunk_sectors / 2, mddev->layout);
02c2de8c 7518 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
5fd13351
N		 7519 rcu_read_lock();
7520 for (i = 0; i < conf->raid_disks; i++) {
7521 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
7522 seq_printf (seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
7523 }
7524 rcu_read_unlock();
1da177e4 7525 seq_printf (seq, "]");
1da177e4
LT		 7526}
7527
d1688a6d 7528static void print_raid5_conf (struct r5conf *conf)
1da177e4
LT		 7529{
7530 int i;
7531 struct disk_info *tmp;
7532
cc6167b4 7533 pr_debug("RAID conf printout:\n");
1da177e4 7534 if (!conf) {
cc6167b4 7535 pr_debug("(conf==NULL)\n");
1da177e4
LT		 7536 return;
7537 }
cc6167b4 7538 pr_debug(" --- level:%d rd:%d wd:%d\n", conf->level,
0c55e022
N		 7539 conf->raid_disks,
7540 conf->raid_disks - conf->mddev->degraded);
1da177e4
LT		 7541
7542 for (i = 0; i < conf->raid_disks; i++) {
7543 char b[BDEVNAME_SIZE];
7544 tmp = conf->disks + i;
7545 if (tmp->rdev)
cc6167b4 7546 pr_debug(" disk %d, o:%d, dev:%s\n",
0c55e022
N		 7547 i, !test_bit(Faulty, &tmp->rdev->flags),
7548 bdevname(tmp->rdev->bdev, b));
1da177e4
LT		 7549 }
7550}
7551
fd01b88c 7552static int raid5_spare_active(struct mddev *mddev)
1da177e4
LT		 7553{
7554 int i;
d1688a6d 7555 struct r5conf *conf = mddev->private;
1da177e4 7556 struct disk_info *tmp;
6b965620
N		 7557 int count = 0;
7558 unsigned long flags;
1da177e4
LT		 7559
7560 for (i = 0; i < conf->raid_disks; i++) {
7561 tmp = conf->disks + i;
dd054fce
N		 7562 if (tmp->replacement
7563 && tmp->replacement->recovery_offset == MaxSector
7564 && !test_bit(Faulty, &tmp->replacement->flags)
7565 && !test_and_set_bit(In_sync, &tmp->replacement->flags)) {
7566 /* Replacement has just become active. */
7567 if (!tmp->rdev
7568 || !test_and_clear_bit(In_sync, &tmp->rdev->flags))
7569 count++;
7570 if (tmp->rdev) {
7571 /* Replaced device not technically faulty,
7572 * but we need to be sure it gets removed
7573 * and never re-added.
7574 */
7575 set_bit(Faulty, &tmp->rdev->flags);
7576 sysfs_notify_dirent_safe(
7577 tmp->rdev->sysfs_state);
7578 }
7579 sysfs_notify_dirent_safe(tmp->replacement->sysfs_state);
7580 } else if (tmp->rdev
70fffd0b 7581 && tmp->rdev->recovery_offset == MaxSector
b2d444d7 7582 && !test_bit(Faulty, &tmp->rdev->flags)
c04be0aa 7583 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
6b965620 7584 count++;
43c73ca4 7585 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
1da177e4
LT		 7586 }
7587 }
6b965620 7588 spin_lock_irqsave(&conf->device_lock, flags);
2e38a37f 7589 mddev->degraded = raid5_calc_degraded(conf);
6b965620 7590 spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4 7591 print_raid5_conf(conf);
6b965620 7592 return count;
1da177e4
LT		 7593}
7594
b8321b68 7595static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7596{
d1688a6d 7597 struct r5conf *conf = mddev->private;
1da177e4 7598 int err = 0;
b8321b68 7599 int number = rdev->raid_disk;
657e3e4d 7600 struct md_rdev **rdevp;
1da177e4
LT		 7601 struct disk_info *p = conf->disks + number;
7602
7603 print_raid5_conf(conf);
f6b6ec5c 7604 if (test_bit(Journal, &rdev->flags) && conf->log) {
c2bb6242 7605 /*
f6b6ec5c
SL		 7606 * we can't wait pending write here, as this is called in
7607 * raid5d, wait will deadlock.
84dd97a6
N		 7608 * neilb: there is no locking about new writes here,
7609 * so this cannot be safe.
c2bb6242 7610 */
70d466f7
SL		 7611 if (atomic_read(&conf->active_stripes) ||
7612 atomic_read(&conf->r5c_cached_full_stripes) ||
7613 atomic_read(&conf->r5c_cached_partial_stripes)) {
f6b6ec5c 7614 return -EBUSY;
84dd97a6 7615 }
ff875738 7616 log_exit(conf);
f6b6ec5c 7617 return 0;
c2bb6242 7618 }
657e3e4d
N		 7619 if (rdev == p->rdev)
7620 rdevp = &p->rdev;
7621 else if (rdev == p->replacement)
7622 rdevp = &p->replacement;
7623 else
7624 return 0;
7625
7626 if (number >= conf->raid_disks &&
7627 conf->reshape_progress == MaxSector)
7628 clear_bit(In_sync, &rdev->flags);
7629
7630 if (test_bit(In_sync, &rdev->flags) ||
7631 atomic_read(&rdev->nr_pending)) {
7632 err = -EBUSY;
7633 goto abort;
7634 }
7635 /* Only remove non-faulty devices if recovery
7636 * isn't possible.
7637 */
7638 if (!test_bit(Faulty, &rdev->flags) &&
7639 mddev->recovery_disabled != conf->recovery_disabled &&
7640 !has_failed(conf) &&
dd054fce 7641 (!p->replacement || p->replacement == rdev) &&
657e3e4d
N		 7642 number < conf->raid_disks) {
7643 err = -EBUSY;
7644 goto abort;
7645 }
7646 *rdevp = NULL;
d787be40
N		 7647 if (!test_bit(RemoveSynchronized, &rdev->flags)) {
7648 synchronize_rcu();
7649 if (atomic_read(&rdev->nr_pending)) {
7650 /* lost the race, try later */
7651 err = -EBUSY;
7652 *rdevp = rdev;
7653 }
7654 }
6358c239
AP		 7655 if (!err) {
7656 err = log_modify(conf, rdev, false);
7657 if (err)
7658 goto abort;
7659 }
d787be40 7660 if (p->replacement) {
dd054fce
N		 7661 /* We must have just cleared 'rdev' */
7662 p->rdev = p->replacement;
7663 clear_bit(Replacement, &p->replacement->flags);
7664 smp_mb(); /* Make sure other CPUs may see both as identical
7665 * but will never see neither - if they are careful
7666 */
7667 p->replacement = NULL;
6358c239
AP		 7668
7669 if (!err)
7670 err = log_modify(conf, p->rdev, true);
e5bc9c3c
GJ		 7671 }
7672
7673 clear_bit(WantReplacement, &rdev->flags);
1da177e4
LT		 7674abort:
7675
7676 print_raid5_conf(conf);
7677 return err;
7678}
7679
fd01b88c 7680static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7681{
d1688a6d 7682 struct r5conf *conf = mddev->private;
d9771f5e 7683 int ret, err = -EEXIST;
1da177e4
LT		 7684 int disk;
7685 struct disk_info *p;
6c2fce2e
NB		 7686 int first = 0;
7687 int last = conf->raid_disks - 1;
1da177e4 7688
f6b6ec5c 7689 if (test_bit(Journal, &rdev->flags)) {
f6b6ec5c
SL		 7690 if (conf->log)
7691 return -EBUSY;
7692
7693 rdev->raid_disk = 0;
7694 /*
7695 * The array is in readonly mode if journal is missing, so no
7696 * write requests running. We should be safe
7697 */
d9771f5e
XN		 7698 ret = log_init(conf, rdev, false);
7699 if (ret)
7700 return ret;
7701
7702 ret = r5l_start(conf->log);
7703 if (ret)
7704 return ret;
7705
f6b6ec5c
SL		 7706 return 0;
7707 }
7f0da59b
N		 7708 if (mddev->recovery_disabled == conf->recovery_disabled)
7709 return -EBUSY;
7710
dc10c643 7711 if (rdev->saved_raid_disk < 0 && has_failed(conf))
1da177e4 7712 /* no point adding a device */
199050ea 7713 return -EINVAL;
1da177e4 7714
6c2fce2e
NB		 7715 if (rdev->raid_disk >= 0)
7716 first = last = rdev->raid_disk;
1da177e4
LT		 7717
7718 /*
16a53ecc
N		 7719 * find the disk ... but prefer rdev->saved_raid_disk
7720 * if possible.
1da177e4 7721 */
16a53ecc 7722 if (rdev->saved_raid_disk >= 0 &&
6c2fce2e 7723 rdev->saved_raid_disk >= first &&
16a53ecc 7724 conf->disks[rdev->saved_raid_disk].rdev == NULL)
5cfb22a1
N		 7725 first = rdev->saved_raid_disk;
7726
7727 for (disk = first; disk <= last; disk++) {
7bfec5f3
N		 7728 p = conf->disks + disk;
7729 if (p->rdev == NULL) {
b2d444d7 7730 clear_bit(In_sync, &rdev->flags);
1da177e4 7731 rdev->raid_disk = disk;
72626685
N		 7732 if (rdev->saved_raid_disk != disk)
7733 conf->fullsync = 1;
d6065f7b 7734 rcu_assign_pointer(p->rdev, rdev);
6358c239
AP		 7735
7736 err = log_modify(conf, rdev, true);
7737
5cfb22a1 7738 goto out;
1da177e4 7739 }
5cfb22a1
N		 7740 }
7741 for (disk = first; disk <= last; disk++) {
7742 p = conf->disks + disk;
7bfec5f3
N		 7743 if (test_bit(WantReplacement, &p->rdev->flags) &&
7744 p->replacement == NULL) {
7745 clear_bit(In_sync, &rdev->flags);
7746 set_bit(Replacement, &rdev->flags);
7747 rdev->raid_disk = disk;
7748 err = 0;
7749 conf->fullsync = 1;
7750 rcu_assign_pointer(p->replacement, rdev);
7751 break;
7752 }
7753 }
5cfb22a1 7754out:
1da177e4 7755 print_raid5_conf(conf);
199050ea 7756 return err;
1da177e4
LT		 7757}
7758
fd01b88c 7759static int raid5_resize(struct mddev *mddev, sector_t sectors)
1da177e4
LT		 7760{
7761 /* no resync is happening, and there is enough space
7762 * on all devices, so we can resize.
7763 * We need to make sure resync covers any new space.
7764 * If the array is shrinking we should possibly wait until
7765 * any io in the removed space completes, but it hardly seems
7766 * worth it.
7767 */
a4a6125a 7768 sector_t newsize;
3cb5edf4
N		 7769 struct r5conf *conf = mddev->private;
7770
e254de6b 7771 if (raid5_has_log(conf) || raid5_has_ppl(conf))
713cf5a6 7772 return -EINVAL;
3cb5edf4 7773 sectors &= ~((sector_t)conf->chunk_sectors - 1);
a4a6125a
N		 7774 newsize = raid5_size(mddev, sectors, mddev->raid_disks);
7775 if (mddev->external_size &&
7776 mddev->array_sectors > newsize)
b522adcd 7777 return -EINVAL;
a4a6125a 7778 if (mddev->bitmap) {
e64e4018 7779 int ret = md_bitmap_resize(mddev->bitmap, sectors, 0, 0);
a4a6125a
N		 7780 if (ret)
7781 return ret;
7782 }
7783 md_set_array_sectors(mddev, newsize);
b098636c
N		 7784 if (sectors > mddev->dev_sectors &&
7785 mddev->recovery_cp > mddev->dev_sectors) {
58c0fed4 7786 mddev->recovery_cp = mddev->dev_sectors;
1da177e4
LT		 7787 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7788 }
58c0fed4 7789 mddev->dev_sectors = sectors;
4b5c7ae8 7790 mddev->resync_max_sectors = sectors;
1da177e4
LT		 7791 return 0;
7792}
7793
fd01b88c 7794static int check_stripe_cache(struct mddev *mddev)
01ee22b4
N		 7795{
7796 /* Can only proceed if there are plenty of stripe_heads.
7797 * We need a minimum of one full stripe,, and for sensible progress
7798 * it is best to have about 4 times that.
7799 * If we require 4 times, then the default 256 4K stripe_heads will
7800 * allow for chunk sizes up to 256K, which is probably OK.
7801 * If the chunk size is greater, user-space should request more
7802 * stripe_heads first.
7803 */
d1688a6d 7804 struct r5conf *conf = mddev->private;
01ee22b4 7805 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
edbe83ab 7806 > conf->min_nr_stripes ||
01ee22b4 7807 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
edbe83ab 7808 > conf->min_nr_stripes) {
cc6167b4
N		 7809 pr_warn("md/raid:%s: reshape: not enough stripes. Needed %lu\n",
7810 mdname(mddev),
7811 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
7812 / STRIPE_SIZE)*4);
01ee22b4
N		 7813 return 0;
7814 }
7815 return 1;
7816}
7817
fd01b88c 7818static int check_reshape(struct mddev *mddev)
29269553 7819{
d1688a6d 7820 struct r5conf *conf = mddev->private;
29269553 7821
e254de6b 7822 if (raid5_has_log(conf) || raid5_has_ppl(conf))
713cf5a6 7823 return -EINVAL;
88ce4930
N		 7824 if (mddev->delta_disks == 0 &&
7825 mddev->new_layout == mddev->layout &&
664e7c41 7826 mddev->new_chunk_sectors == mddev->chunk_sectors)
50ac168a 7827 return 0; /* nothing to do */
674806d6 7828 if (has_failed(conf))
ec32a2bd 7829 return -EINVAL;
fdcfbbb6 7830 if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) {
ec32a2bd
N		 7831 /* We might be able to shrink, but the devices must
7832 * be made bigger first.
7833 * For raid6, 4 is the minimum size.
7834 * Otherwise 2 is the minimum
7835 */
7836 int min = 2;
7837 if (mddev->level == 6)
7838 min = 4;
7839 if (mddev->raid_disks + mddev->delta_disks < min)
7840 return -EINVAL;
7841 }
29269553 7842
01ee22b4 7843 if (!check_stripe_cache(mddev))
29269553 7844 return -ENOSPC;
29269553 7845
738a2738
N		 7846 if (mddev->new_chunk_sectors > mddev->chunk_sectors ||
7847 mddev->delta_disks > 0)
7848 if (resize_chunks(conf,
7849 conf->previous_raid_disks
7850 + max(0, mddev->delta_disks),
7851 max(mddev->new_chunk_sectors,
7852 mddev->chunk_sectors)
7853 ) < 0)
7854 return -ENOMEM;
845b9e22
AP		 7855
7856 if (conf->previous_raid_disks + mddev->delta_disks <= conf->pool_size)
7857 return 0; /* never bother to shrink */
e56108d6
N		 7858 return resize_stripes(conf, (conf->previous_raid_disks
7859 + mddev->delta_disks));
63c70c4f
N		 7860}
7861
fd01b88c 7862static int raid5_start_reshape(struct mddev *mddev)
63c70c4f 7863{
d1688a6d 7864 struct r5conf *conf = mddev->private;
3cb03002 7865 struct md_rdev *rdev;
63c70c4f 7866 int spares = 0;
c04be0aa 7867 unsigned long flags;
63c70c4f 7868
f416885e 7869 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
63c70c4f
N		 7870 return -EBUSY;
7871
01ee22b4
N		 7872 if (!check_stripe_cache(mddev))
7873 return -ENOSPC;
7874
30b67645
N		 7875 if (has_failed(conf))
7876 return -EINVAL;
7877
c6563a8c 7878 rdev_for_each(rdev, mddev) {
469518a3
N		 7879 if (!test_bit(In_sync, &rdev->flags)
7880 && !test_bit(Faulty, &rdev->flags))
29269553 7881 spares++;
c6563a8c 7882 }
63c70c4f 7883
f416885e 7884 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
29269553
N		 7885 /* Not enough devices even to make a degraded array
7886 * of that size
7887 */
7888 return -EINVAL;
7889
ec32a2bd
N		 7890 /* Refuse to reduce size of the array. Any reductions in
7891 * array size must be through explicit setting of array_size
7892 * attribute.
7893 */
7894 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
7895 < mddev->array_sectors) {
cc6167b4
N		 7896 pr_warn("md/raid:%s: array size must be reduced before number of disks\n",
7897 mdname(mddev));
ec32a2bd
N		 7898 return -EINVAL;
7899 }
7900
f6705578 7901 atomic_set(&conf->reshape_stripes, 0);
29269553 7902 spin_lock_irq(&conf->device_lock);
c46501b2 7903 write_seqcount_begin(&conf->gen_lock);
29269553 7904 conf->previous_raid_disks = conf->raid_disks;
63c70c4f 7905 conf->raid_disks += mddev->delta_disks;
09c9e5fa
AN		 7906 conf->prev_chunk_sectors = conf->chunk_sectors;
7907 conf->chunk_sectors = mddev->new_chunk_sectors;
88ce4930
N		 7908 conf->prev_algo = conf->algorithm;
7909 conf->algorithm = mddev->new_layout;
05616be5
N		 7910 conf->generation++;
7911 /* Code that selects data_offset needs to see the generation update
7912 * if reshape_progress has been set - so a memory barrier needed.
7913 */
7914 smp_mb();
2c810cdd 7915 if (mddev->reshape_backwards)
fef9c61f
N		 7916 conf->reshape_progress = raid5_size(mddev, 0, 0);
7917 else
7918 conf->reshape_progress = 0;
7919 conf->reshape_safe = conf->reshape_progress;
c46501b2 7920 write_seqcount_end(&conf->gen_lock);
29269553
N		 7921 spin_unlock_irq(&conf->device_lock);
7922
4d77e3ba
N		 7923 /* Now make sure any requests that proceeded on the assumption
7924 * the reshape wasn't running - like Discard or Read - have
7925 * completed.
7926 */
7927 mddev_suspend(mddev);
7928 mddev_resume(mddev);
7929
29269553
N		 7930 /* Add some new drives, as many as will fit.
7931 * We know there are enough to make the newly sized array work.
3424bf6a
N		 7932 * Don't add devices if we are reducing the number of
7933 * devices in the array. This is because it is not possible
7934 * to correctly record the "partially reconstructed" state of
7935 * such devices during the reshape and confusion could result.
29269553 7936 */
87a8dec9 7937 if (mddev->delta_disks >= 0) {
dafb20fa 7938 rdev_for_each(rdev, mddev)
87a8dec9
N		 7939 if (rdev->raid_disk < 0 &&
7940 !test_bit(Faulty, &rdev->flags)) {
7941 if (raid5_add_disk(mddev, rdev) == 0) {
87a8dec9 7942 if (rdev->raid_disk
9d4c7d87 7943 >= conf->previous_raid_disks)
87a8dec9 7944 set_bit(In_sync, &rdev->flags);
9d4c7d87 7945 else
87a8dec9 7946 rdev->recovery_offset = 0;
36fad858
NK		 7947
7948 if (sysfs_link_rdev(mddev, rdev))
87a8dec9 7949 /* Failure here is OK */;
50da0840 7950 }
87a8dec9
N		 7951 } else if (rdev->raid_disk >= conf->previous_raid_disks
7952 && !test_bit(Faulty, &rdev->flags)) {
7953 /* This is a spare that was manually added */
7954 set_bit(In_sync, &rdev->flags);
87a8dec9 7955 }
29269553 7956
87a8dec9
N		 7957 /* When a reshape changes the number of devices,
7958 * ->degraded is measured against the larger of the
7959 * pre and post number of devices.
7960 */
ec32a2bd 7961 spin_lock_irqsave(&conf->device_lock, flags);
2e38a37f 7962 mddev->degraded = raid5_calc_degraded(conf);
ec32a2bd
N		 7963 spin_unlock_irqrestore(&conf->device_lock, flags);
7964 }
63c70c4f 7965 mddev->raid_disks = conf->raid_disks;
e516402c 7966 mddev->reshape_position = conf->reshape_progress;
2953079c 7967 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
f6705578 7968
29269553
N		 7969 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7970 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
ea358cd0 7971 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
29269553
N		 7972 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7973 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7974 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 7975 "reshape");
29269553
N		 7976 if (!mddev->sync_thread) {
7977 mddev->recovery = 0;
7978 spin_lock_irq(&conf->device_lock);
ba8805b9 7979 write_seqcount_begin(&conf->gen_lock);
29269553 7980 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
ba8805b9
N		 7981 mddev->new_chunk_sectors =
7982 conf->chunk_sectors = conf->prev_chunk_sectors;
7983 mddev->new_layout = conf->algorithm = conf->prev_algo;
05616be5
N		 7984 rdev_for_each(rdev, mddev)
7985 rdev->new_data_offset = rdev->data_offset;
7986 smp_wmb();
ba8805b9 7987 conf->generation --;
fef9c61f 7988 conf->reshape_progress = MaxSector;
1e3fa9bd 7989 mddev->reshape_position = MaxSector;
ba8805b9 7990 write_seqcount_end(&conf->gen_lock);
29269553
N		 7991 spin_unlock_irq(&conf->device_lock);
7992 return -EAGAIN;
7993 }
c8f517c4 7994 conf->reshape_checkpoint = jiffies;
29269553
N		 7995 md_wakeup_thread(mddev->sync_thread);
7996 md_new_event(mddev);
7997 return 0;
7998}
29269553 7999
ec32a2bd
N		 8000/* This is called from the reshape thread and should make any
8001 * changes needed in 'conf'
8002 */
d1688a6d 8003static void end_reshape(struct r5conf *conf)
29269553 8004{
29269553 8005
f6705578 8006 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
db0505d3 8007 struct md_rdev *rdev;
f6705578 8008
f6705578 8009 spin_lock_irq(&conf->device_lock);
cea9c228 8010 conf->previous_raid_disks = conf->raid_disks;
b5d27718 8011 md_finish_reshape(conf->mddev);
05616be5 8012 smp_wmb();
fef9c61f 8013 conf->reshape_progress = MaxSector;
6cbd8148 8014 conf->mddev->reshape_position = MaxSector;
db0505d3
N		 8015 rdev_for_each(rdev, conf->mddev)
8016 if (rdev->raid_disk >= 0 &&
8017 !test_bit(Journal, &rdev->flags) &&
8018 !test_bit(In_sync, &rdev->flags))
8019 rdev->recovery_offset = MaxSector;
f6705578 8020 spin_unlock_irq(&conf->device_lock);
b0f9ec04 8021 wake_up(&conf->wait_for_overlap);
16a53ecc
N		 8022
8023 /* read-ahead size must cover two whole stripes, which is
8024 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
8025 */
4a5add49 8026 if (conf->mddev->queue) {
cea9c228 8027 int data_disks = conf->raid_disks - conf->max_degraded;
09c9e5fa 8028 int stripe = data_disks * ((conf->chunk_sectors << 9)
cea9c228 8029 / PAGE_SIZE);
dc3b17cc
JK		 8030 if (conf->mddev->queue->backing_dev_info->ra_pages < 2 * stripe)
8031 conf->mddev->queue->backing_dev_info->ra_pages = 2 * stripe;
16a53ecc 8032 }
29269553 8033 }
29269553
N		 8034}
8035
ec32a2bd
N		 8036/* This is called from the raid5d thread with mddev_lock held.
8037 * It makes config changes to the device.
8038 */
fd01b88c 8039static void raid5_finish_reshape(struct mddev *mddev)
cea9c228 8040{
d1688a6d 8041 struct r5conf *conf = mddev->private;
cea9c228
N		 8042
8043 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8044
8876391e 8045 if (mddev->delta_disks <= 0) {
ec32a2bd 8046 int d;
908f4fbd 8047 spin_lock_irq(&conf->device_lock);
2e38a37f 8048 mddev->degraded = raid5_calc_degraded(conf);
908f4fbd 8049 spin_unlock_irq(&conf->device_lock);
ec32a2bd
N		 8050 for (d = conf->raid_disks ;
8051 d < conf->raid_disks - mddev->delta_disks;
1a67dde0 8052 d++) {
3cb03002 8053 struct md_rdev *rdev = conf->disks[d].rdev;
da7613b8
N		 8054 if (rdev)
8055 clear_bit(In_sync, &rdev->flags);
8056 rdev = conf->disks[d].replacement;
8057 if (rdev)
8058 clear_bit(In_sync, &rdev->flags);
1a67dde0 8059 }
cea9c228 8060 }
88ce4930 8061 mddev->layout = conf->algorithm;
09c9e5fa 8062 mddev->chunk_sectors = conf->chunk_sectors;
ec32a2bd
N		 8063 mddev->reshape_position = MaxSector;
8064 mddev->delta_disks = 0;
2c810cdd 8065 mddev->reshape_backwards = 0;
cea9c228
N		 8066 }
8067}
8068
b03e0ccb 8069static void raid5_quiesce(struct mddev *mddev, int quiesce)
72626685 8070{
d1688a6d 8071 struct r5conf *conf = mddev->private;
72626685 8072
b03e0ccb
N		 8073 if (quiesce) {
8074 /* stop all writes */
566c09c5 8075 lock_all_device_hash_locks_irq(conf);
64bd660b
N		 8076 /* '2' tells resync/reshape to pause so that all
8077 * active stripes can drain
8078 */
a39f7afd 8079 r5c_flush_cache(conf, INT_MAX);
64bd660b 8080 conf->quiesce = 2;
b1b46486 8081 wait_event_cmd(conf->wait_for_quiescent,
46031f9a
RBJ		 8082 atomic_read(&conf->active_stripes) == 0 &&
8083 atomic_read(&conf->active_aligned_reads) == 0,
566c09c5
SL		 8084 unlock_all_device_hash_locks_irq(conf),
8085 lock_all_device_hash_locks_irq(conf));
64bd660b 8086 conf->quiesce = 1;
566c09c5 8087 unlock_all_device_hash_locks_irq(conf);
64bd660b
N		 8088 /* allow reshape to continue */
8089 wake_up(&conf->wait_for_overlap);
b03e0ccb
N		 8090 } else {
8091 /* re-enable writes */
566c09c5 8092 lock_all_device_hash_locks_irq(conf);
72626685 8093 conf->quiesce = 0;
b1b46486 8094 wake_up(&conf->wait_for_quiescent);
e464eafd 8095 wake_up(&conf->wait_for_overlap);
566c09c5 8096 unlock_all_device_hash_locks_irq(conf);
72626685 8097 }
1532d9e8 8098 log_quiesce(conf, quiesce);
72626685 8099}
b15c2e57 8100
fd01b88c 8101static void *raid45_takeover_raid0(struct mddev *mddev, int level)
54071b38 8102{
e373ab10 8103 struct r0conf *raid0_conf = mddev->private;
d76c8420 8104 sector_t sectors;
54071b38 8105
f1b29bca 8106 /* for raid0 takeover only one zone is supported */
e373ab10 8107 if (raid0_conf->nr_strip_zones > 1) {
cc6167b4
N		 8108 pr_warn("md/raid:%s: cannot takeover raid0 with more than one zone.\n",
8109 mdname(mddev));
f1b29bca
DW		 8110 return ERR_PTR(-EINVAL);
8111 }
8112
e373ab10
N		 8113 sectors = raid0_conf->strip_zone[0].zone_end;
8114 sector_div(sectors, raid0_conf->strip_zone[0].nb_dev);
3b71bd93 8115 mddev->dev_sectors = sectors;
f1b29bca 8116 mddev->new_level = level;
54071b38
TM		 8117 mddev->new_layout = ALGORITHM_PARITY_N;
8118 mddev->new_chunk_sectors = mddev->chunk_sectors;
8119 mddev->raid_disks += 1;
8120 mddev->delta_disks = 1;
8121 /* make sure it will be not marked as dirty */
8122 mddev->recovery_cp = MaxSector;
8123
8124 return setup_conf(mddev);
8125}
8126
fd01b88c 8127static void *raid5_takeover_raid1(struct mddev *mddev)
d562b0c4
N		 8128{
8129 int chunksect;
6995f0b2 8130 void *ret;
d562b0c4
N		 8131
8132 if (mddev->raid_disks != 2 ||
8133 mddev->degraded > 1)
8134 return ERR_PTR(-EINVAL);
8135
8136 /* Should check if there are write-behind devices? */
8137
8138 chunksect = 64*2; /* 64K by default */
8139
8140 /* The array must be an exact multiple of chunksize */
8141 while (chunksect && (mddev->array_sectors & (chunksect-1)))
8142 chunksect >>= 1;
8143
8144 if ((chunksect<<9) < STRIPE_SIZE)
8145 /* array size does not allow a suitable chunk size */
8146 return ERR_PTR(-EINVAL);
8147
8148 mddev->new_level = 5;
8149 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
664e7c41 8150 mddev->new_chunk_sectors = chunksect;
d562b0c4 8151
6995f0b2 8152 ret = setup_conf(mddev);
32cd7cbb 8153 if (!IS_ERR(ret))
394ed8e4
SL		 8154 mddev_clear_unsupported_flags(mddev,
8155 UNSUPPORTED_MDDEV_FLAGS);
6995f0b2 8156 return ret;
d562b0c4
N		 8157}
8158
fd01b88c 8159static void *raid5_takeover_raid6(struct mddev *mddev)
fc9739c6
N		 8160{
8161 int new_layout;
8162
8163 switch (mddev->layout) {
8164 case ALGORITHM_LEFT_ASYMMETRIC_6:
8165 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
8166 break;
8167 case ALGORITHM_RIGHT_ASYMMETRIC_6:
8168 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
8169 break;
8170 case ALGORITHM_LEFT_SYMMETRIC_6:
8171 new_layout = ALGORITHM_LEFT_SYMMETRIC;
8172 break;
8173 case ALGORITHM_RIGHT_SYMMETRIC_6:
8174 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
8175 break;
8176 case ALGORITHM_PARITY_0_6:
8177 new_layout = ALGORITHM_PARITY_0;
8178 break;
8179 case ALGORITHM_PARITY_N:
8180 new_layout = ALGORITHM_PARITY_N;
8181 break;
8182 default:
8183 return ERR_PTR(-EINVAL);
8184 }
8185 mddev->new_level = 5;
8186 mddev->new_layout = new_layout;
8187 mddev->delta_disks = -1;
8188 mddev->raid_disks -= 1;
8189 return setup_conf(mddev);
8190}
8191
fd01b88c 8192static int raid5_check_reshape(struct mddev *mddev)
b3546035 8193{
88ce4930
N		 8194 /* For a 2-drive array, the layout and chunk size can be changed
8195 * immediately as not restriping is needed.
8196 * For larger arrays we record the new value - after validation
8197 * to be used by a reshape pass.
b3546035 8198 */
d1688a6d 8199 struct r5conf *conf = mddev->private;
597a711b 8200 int new_chunk = mddev->new_chunk_sectors;
b3546035 8201
597a711b 8202 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
b3546035
N		 8203 return -EINVAL;
8204 if (new_chunk > 0) {
0ba459d2 8205 if (!is_power_of_2(new_chunk))
b3546035 8206 return -EINVAL;
597a711b 8207 if (new_chunk < (PAGE_SIZE>>9))
b3546035 8208 return -EINVAL;
597a711b 8209 if (mddev->array_sectors & (new_chunk-1))
b3546035
N		 8210 /* not factor of array size */
8211 return -EINVAL;
8212 }
8213
8214 /* They look valid */
8215
88ce4930 8216 if (mddev->raid_disks == 2) {
597a711b
N		 8217 /* can make the change immediately */
8218 if (mddev->new_layout >= 0) {
8219 conf->algorithm = mddev->new_layout;
8220 mddev->layout = mddev->new_layout;
88ce4930
N		 8221 }
8222 if (new_chunk > 0) {
597a711b
N		 8223 conf->chunk_sectors = new_chunk ;
8224 mddev->chunk_sectors = new_chunk;
88ce4930 8225 }
2953079c 8226 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
88ce4930 8227 md_wakeup_thread(mddev->thread);
b3546035 8228 }
50ac168a 8229 return check_reshape(mddev);
88ce4930
N		 8230}
8231
fd01b88c 8232static int raid6_check_reshape(struct mddev *mddev)
88ce4930 8233{
597a711b 8234 int new_chunk = mddev->new_chunk_sectors;
50ac168a 8235
597a711b 8236 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
88ce4930 8237 return -EINVAL;
b3546035 8238 if (new_chunk > 0) {
0ba459d2 8239 if (!is_power_of_2(new_chunk))
88ce4930 8240 return -EINVAL;
597a711b 8241 if (new_chunk < (PAGE_SIZE >> 9))
88ce4930 8242 return -EINVAL;
597a711b 8243 if (mddev->array_sectors & (new_chunk-1))
88ce4930
N		 8244 /* not factor of array size */
8245 return -EINVAL;
b3546035 8246 }
88ce4930
N		 8247
8248 /* They look valid */
50ac168a 8249 return check_reshape(mddev);
b3546035
N		 8250}
8251
fd01b88c 8252static void *raid5_takeover(struct mddev *mddev)
d562b0c4
N		 8253{
8254 /* raid5 can take over:
f1b29bca 8255 * raid0 - if there is only one strip zone - make it a raid4 layout
d562b0c4
N		 8256 * raid1 - if there are two drives. We need to know the chunk size
8257 * raid4 - trivial - just use a raid4 layout.
8258 * raid6 - Providing it is a *_6 layout
d562b0c4 8259 */
f1b29bca
DW		 8260 if (mddev->level == 0)
8261 return raid45_takeover_raid0(mddev, 5);
d562b0c4
N		 8262 if (mddev->level == 1)
8263 return raid5_takeover_raid1(mddev);
e9d4758f
N		 8264 if (mddev->level == 4) {
8265 mddev->new_layout = ALGORITHM_PARITY_N;
8266 mddev->new_level = 5;
8267 return setup_conf(mddev);
8268 }
fc9739c6
N		 8269 if (mddev->level == 6)
8270 return raid5_takeover_raid6(mddev);
d562b0c4
N		 8271
8272 return ERR_PTR(-EINVAL);
8273}
8274
fd01b88c 8275static void *raid4_takeover(struct mddev *mddev)
a78d38a1 8276{
f1b29bca
DW		 8277 /* raid4 can take over:
8278 * raid0 - if there is only one strip zone
8279 * raid5 - if layout is right
a78d38a1 8280 */
f1b29bca
DW		 8281 if (mddev->level == 0)
8282 return raid45_takeover_raid0(mddev, 4);
a78d38a1
N		 8283 if (mddev->level == 5 &&
8284 mddev->layout == ALGORITHM_PARITY_N) {
8285 mddev->new_layout = 0;
8286 mddev->new_level = 4;
8287 return setup_conf(mddev);
8288 }
8289 return ERR_PTR(-EINVAL);
8290}
d562b0c4 8291
84fc4b56 8292static struct md_personality raid5_personality;
245f46c2 8293
fd01b88c 8294static void *raid6_takeover(struct mddev *mddev)
245f46c2
N		 8295{
8296 /* Currently can only take over a raid5. We map the
8297 * personality to an equivalent raid6 personality
8298 * with the Q block at the end.
8299 */
8300 int new_layout;
8301
8302 if (mddev->pers != &raid5_personality)
8303 return ERR_PTR(-EINVAL);
8304 if (mddev->degraded > 1)
8305 return ERR_PTR(-EINVAL);
8306 if (mddev->raid_disks > 253)
8307 return ERR_PTR(-EINVAL);
8308 if (mddev->raid_disks < 3)
8309 return ERR_PTR(-EINVAL);
8310
8311 switch (mddev->layout) {
8312 case ALGORITHM_LEFT_ASYMMETRIC:
8313 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
8314 break;
8315 case ALGORITHM_RIGHT_ASYMMETRIC:
8316 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
8317 break;
8318 case ALGORITHM_LEFT_SYMMETRIC:
8319 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
8320 break;
8321 case ALGORITHM_RIGHT_SYMMETRIC:
8322 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
8323 break;
8324 case ALGORITHM_PARITY_0:
8325 new_layout = ALGORITHM_PARITY_0_6;
8326 break;
8327 case ALGORITHM_PARITY_N:
8328 new_layout = ALGORITHM_PARITY_N;
8329 break;
8330 default:
8331 return ERR_PTR(-EINVAL);
8332 }
8333 mddev->new_level = 6;
8334 mddev->new_layout = new_layout;
8335 mddev->delta_disks = 1;
8336 mddev->raid_disks += 1;
8337 return setup_conf(mddev);
8338}
8339
ba903a3e
AP		 8340static int raid5_change_consistency_policy(struct mddev *mddev, const char *buf)
8341{
8342 struct r5conf *conf;
8343 int err;
8344
8345 err = mddev_lock(mddev);
8346 if (err)
8347 return err;
8348 conf = mddev->private;
8349 if (!conf) {
8350 mddev_unlock(mddev);
8351 return -ENODEV;
8352 }
8353
845b9e22 8354 if (strncmp(buf, "ppl", 3) == 0) {
0bb0c105 8355 /* ppl only works with RAID 5 */
845b9e22
AP		 8356 if (!raid5_has_ppl(conf) && conf->level == 5) {
8357 err = log_init(conf, NULL, true);
8358 if (!err) {
8359 err = resize_stripes(conf, conf->pool_size);
8360 if (err)
8361 log_exit(conf);
8362 }
0bb0c105
SL		 8363 } else
8364 err = -EINVAL;
8365 } else if (strncmp(buf, "resync", 6) == 0) {
8366 if (raid5_has_ppl(conf)) {
8367 mddev_suspend(mddev);
8368 log_exit(conf);
0bb0c105 8369 mddev_resume(mddev);
845b9e22 8370 err = resize_stripes(conf, conf->pool_size);
0bb0c105
SL		 8371 } else if (test_bit(MD_HAS_JOURNAL, &conf->mddev->flags) &&
8372 r5l_log_disk_error(conf)) {
8373 bool journal_dev_exists = false;
8374 struct md_rdev *rdev;
8375
8376 rdev_for_each(rdev, mddev)
8377 if (test_bit(Journal, &rdev->flags)) {
8378 journal_dev_exists = true;
8379 break;
8380 }
8381
8382 if (!journal_dev_exists) {
8383 mddev_suspend(mddev);
8384 clear_bit(MD_HAS_JOURNAL, &mddev->flags);
8385 mddev_resume(mddev);
8386 } else /* need remove journal device first */
8387 err = -EBUSY;
8388 } else
8389 err = -EINVAL;
ba903a3e
AP		 8390 } else {
8391 err = -EINVAL;
8392 }
8393
8394 if (!err)
8395 md_update_sb(mddev, 1);
8396
8397 mddev_unlock(mddev);
8398
8399 return err;
8400}
8401
d5d885fd
SL		 8402static int raid5_start(struct mddev *mddev)
8403{
8404 struct r5conf *conf = mddev->private;
8405
8406 return r5l_start(conf->log);
8407}
8408
84fc4b56 8409static struct md_personality raid6_personality =
16a53ecc
N		 8410{
8411 .name = "raid6",
8412 .level = 6,
8413 .owner = THIS_MODULE,
849674e4
SL		 8414 .make_request = raid5_make_request,
8415 .run = raid5_run,
d5d885fd 8416 .start = raid5_start,
afa0f557 8417 .free = raid5_free,
849674e4
SL		 8418 .status = raid5_status,
8419 .error_handler = raid5_error,
16a53ecc
N		 8420 .hot_add_disk = raid5_add_disk,
8421 .hot_remove_disk= raid5_remove_disk,
8422 .spare_active = raid5_spare_active,
849674e4 8423 .sync_request = raid5_sync_request,
16a53ecc 8424 .resize = raid5_resize,
80c3a6ce 8425 .size = raid5_size,
50ac168a 8426 .check_reshape = raid6_check_reshape,
f416885e 8427 .start_reshape = raid5_start_reshape,
cea9c228 8428 .finish_reshape = raid5_finish_reshape,
16a53ecc 8429 .quiesce = raid5_quiesce,
245f46c2 8430 .takeover = raid6_takeover,
5c675f83 8431 .congested = raid5_congested,
0bb0c105 8432 .change_consistency_policy = raid5_change_consistency_policy,
16a53ecc 8433};
84fc4b56 8434static struct md_personality raid5_personality =
1da177e4
LT		 8435{
8436 .name = "raid5",
2604b703 8437 .level = 5,
1da177e4 8438 .owner = THIS_MODULE,
849674e4
SL		 8439 .make_request = raid5_make_request,
8440 .run = raid5_run,
d5d885fd 8441 .start = raid5_start,
afa0f557 8442 .free = raid5_free,
849674e4
SL		 8443 .status = raid5_status,
8444 .error_handler = raid5_error,
1da177e4
LT		 8445 .hot_add_disk = raid5_add_disk,
8446 .hot_remove_disk= raid5_remove_disk,
8447 .spare_active = raid5_spare_active,
849674e4 8448 .sync_request = raid5_sync_request,
1da177e4 8449 .resize = raid5_resize,
80c3a6ce 8450 .size = raid5_size,
63c70c4f
N		 8451 .check_reshape = raid5_check_reshape,
8452 .start_reshape = raid5_start_reshape,
cea9c228 8453 .finish_reshape = raid5_finish_reshape,
72626685 8454 .quiesce = raid5_quiesce,
d562b0c4 8455 .takeover = raid5_takeover,
5c675f83 8456 .congested = raid5_congested,
ba903a3e 8457 .change_consistency_policy = raid5_change_consistency_policy,
1da177e4
LT		 8458};
8459
84fc4b56 8460static struct md_personality raid4_personality =
1da177e4 8461{
2604b703
N		 8462 .name = "raid4",
8463 .level = 4,
8464 .owner = THIS_MODULE,
849674e4
SL		 8465 .make_request = raid5_make_request,
8466 .run = raid5_run,
d5d885fd 8467 .start = raid5_start,
afa0f557 8468 .free = raid5_free,
849674e4
SL		 8469 .status = raid5_status,
8470 .error_handler = raid5_error,
2604b703
N		 8471 .hot_add_disk = raid5_add_disk,
8472 .hot_remove_disk= raid5_remove_disk,
8473 .spare_active = raid5_spare_active,
849674e4 8474 .sync_request = raid5_sync_request,
2604b703 8475 .resize = raid5_resize,
80c3a6ce 8476 .size = raid5_size,
3d37890b
N		 8477 .check_reshape = raid5_check_reshape,
8478 .start_reshape = raid5_start_reshape,
cea9c228 8479 .finish_reshape = raid5_finish_reshape,
2604b703 8480 .quiesce = raid5_quiesce,
a78d38a1 8481 .takeover = raid4_takeover,
5c675f83 8482 .congested = raid5_congested,
0bb0c105 8483 .change_consistency_policy = raid5_change_consistency_policy,
2604b703
N		 8484};
8485
8486static int __init raid5_init(void)
8487{
29c6d1bb
SAS		 8488 int ret;
8489
851c30c9
SL		 8490 raid5_wq = alloc_workqueue("raid5wq",
8491 WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0);
8492 if (!raid5_wq)
8493 return -ENOMEM;
29c6d1bb
SAS		 8494
8495 ret = cpuhp_setup_state_multi(CPUHP_MD_RAID5_PREPARE,
8496 "md/raid5:prepare",
8497 raid456_cpu_up_prepare,
8498 raid456_cpu_dead);
8499 if (ret) {
8500 destroy_workqueue(raid5_wq);
8501 return ret;
8502 }
16a53ecc 8503 register_md_personality(&raid6_personality);
2604b703
N		 8504 register_md_personality(&raid5_personality);
8505 register_md_personality(&raid4_personality);
8506 return 0;
1da177e4
LT		 8507}
8508
2604b703 8509static void raid5_exit(void)
1da177e4 8510{
16a53ecc 8511 unregister_md_personality(&raid6_personality);
2604b703
N		 8512 unregister_md_personality(&raid5_personality);
8513 unregister_md_personality(&raid4_personality);
29c6d1bb 8514 cpuhp_remove_multi_state(CPUHP_MD_RAID5_PREPARE);
851c30c9 8515 destroy_workqueue(raid5_wq);
1da177e4
LT		 8516}
8517
8518module_init(raid5_init);
8519module_exit(raid5_exit);
8520MODULE_LICENSE("GPL");
0efb9e61 8521MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
1da177e4 8522MODULE_ALIAS("md-personality-4"); /* RAID5 */
d9d166c2
N		 8523MODULE_ALIAS("md-raid5");
8524MODULE_ALIAS("md-raid4");
2604b703
N		 8525MODULE_ALIAS("md-level-5");
8526MODULE_ALIAS("md-level-4");
16a53ecc
N		 8527MODULE_ALIAS("md-personality-8"); /* RAID6 */
8528MODULE_ALIAS("md-raid6");
8529MODULE_ALIAS("md-level-6");
8530
8531/* This used to be two separate modules, they were: */
8532MODULE_ALIAS("raid5");
8533MODULE_ALIAS("raid6");
Linux 6.x block layer and io_uring tree(s)