projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md/raid1: fail run raid1 array when active disk less than one
[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)
1137 sh->dev[i].write_hint = RWF_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
MD		 1189 rbi->bi_write_hint = sh->dev[i].write_hint;
1190 sh->dev[i].write_hint = RWF_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);
d6950432 2529 atomic_inc(&rdev->read_errors);
14a75d3e 2530 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
cc6167b4
N		 2531 pr_warn_ratelimited(
2532 "md/raid:%s: read error on replacement device (sector %llu on %s).\n",
14a75d3e 2533 mdname(conf->mddev),
05616be5 2534 (unsigned long long)s,
14a75d3e 2535 bdn);
2e8ac303 2536 else if (conf->mddev->degraded >= conf->max_degraded) {
2537 set_bad = 1;
cc6167b4
N		 2538 pr_warn_ratelimited(
2539 "md/raid:%s: read error not correctable (sector %llu on %s).\n",
8bda470e 2540 mdname(conf->mddev),
05616be5 2541 (unsigned long long)s,
8bda470e 2542 bdn);
2e8ac303 2543 } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) {
4e5314b5 2544 /* Oh, no!!! */
2e8ac303 2545 set_bad = 1;
cc6167b4
N		 2546 pr_warn_ratelimited(
2547 "md/raid:%s: read error NOT corrected!! (sector %llu on %s).\n",
8bda470e 2548 mdname(conf->mddev),
05616be5 2549 (unsigned long long)s,
8bda470e 2550 bdn);
2e8ac303 2551 } else if (atomic_read(&rdev->read_errors)
0009fad0
NC		 2552 > conf->max_nr_stripes) {
2553 if (!test_bit(Faulty, &rdev->flags)) {
2554 pr_warn("md/raid:%s: %d read_errors > %d stripes\n",
2555 mdname(conf->mddev),
2556 atomic_read(&rdev->read_errors),
2557 conf->max_nr_stripes);
2558 pr_warn("md/raid:%s: Too many read errors, failing device %s.\n",
2559 mdname(conf->mddev), bdn);
2560 }
2561 } else
ba22dcbf 2562 retry = 1;
edfa1f65
BY		 2563 if (set_bad && test_bit(In_sync, &rdev->flags)
2564 && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2565 retry = 1;
ba22dcbf 2566 if (retry)
143f6e73
XN		 2567 if (sh->qd_idx >= 0 && sh->pd_idx == i)
2568 set_bit(R5_ReadError, &sh->dev[i].flags);
2569 else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) {
3f9e7c14 2570 set_bit(R5_ReadError, &sh->dev[i].flags);
2571 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2572 } else
2573 set_bit(R5_ReadNoMerge, &sh->dev[i].flags);
ba22dcbf 2574 else {
4e5314b5
N		 2575 clear_bit(R5_ReadError, &sh->dev[i].flags);
2576 clear_bit(R5_ReWrite, &sh->dev[i].flags);
2e8ac303 2577 if (!(set_bad
2578 && test_bit(In_sync, &rdev->flags)
2579 && rdev_set_badblocks(
2580 rdev, sh->sector, STRIPE_SECTORS, 0)))
2581 md_error(conf->mddev, rdev);
ba22dcbf 2582 }
1da177e4 2583 }
14a75d3e 2584 rdev_dec_pending(rdev, conf->mddev);
c9445555 2585 bio_reset(bi);
1da177e4
LT		 2586 clear_bit(R5_LOCKED, &sh->dev[i].flags);
2587 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 2588 raid5_release_stripe(sh);
1da177e4
LT		 2589}
2590
4246a0b6 2591static void raid5_end_write_request(struct bio *bi)
1da177e4 2592{
99c0fb5f 2593 struct stripe_head *sh = bi->bi_private;
d1688a6d 2594 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 2595 int disks = sh->disks, i;
977df362 2596 struct md_rdev *uninitialized_var(rdev);
b84db560
N		 2597 sector_t first_bad;
2598 int bad_sectors;
977df362 2599 int replacement = 0;
1da177e4 2600
977df362
N		 2601 for (i = 0 ; i < disks; i++) {
2602 if (bi == &sh->dev[i].req) {
2603 rdev = conf->disks[i].rdev;
1da177e4 2604 break;
977df362
N		 2605 }
2606 if (bi == &sh->dev[i].rreq) {
2607 rdev = conf->disks[i].replacement;
dd054fce
N		 2608 if (rdev)
2609 replacement = 1;
2610 else
2611 /* rdev was removed and 'replacement'
2612 * replaced it. rdev is not removed
2613 * until all requests are finished.
2614 */
2615 rdev = conf->disks[i].rdev;
977df362
N		 2616 break;
2617 }
2618 }
4246a0b6 2619 pr_debug("end_write_request %llu/%d, count %d, error: %d.\n",
1da177e4 2620 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
4e4cbee9 2621 bi->bi_status);
1da177e4 2622 if (i == disks) {
5f9d1fde 2623 bio_reset(bi);
1da177e4 2624 BUG();
6712ecf8 2625 return;
1da177e4
LT		 2626 }
2627
977df362 2628 if (replacement) {
4e4cbee9 2629 if (bi->bi_status)
977df362
N		 2630 md_error(conf->mddev, rdev);
2631 else if (is_badblock(rdev, sh->sector,
2632 STRIPE_SECTORS,
2633 &first_bad, &bad_sectors))
2634 set_bit(R5_MadeGoodRepl, &sh->dev[i].flags);
2635 } else {
4e4cbee9 2636 if (bi->bi_status) {
9f97e4b1 2637 set_bit(STRIPE_DEGRADED, &sh->state);
977df362
N		 2638 set_bit(WriteErrorSeen, &rdev->flags);
2639 set_bit(R5_WriteError, &sh->dev[i].flags);
3a6de292
N		 2640 if (!test_and_set_bit(WantReplacement, &rdev->flags))
2641 set_bit(MD_RECOVERY_NEEDED,
2642 &rdev->mddev->recovery);
977df362
N		 2643 } else if (is_badblock(rdev, sh->sector,
2644 STRIPE_SECTORS,
c0b32972 2645 &first_bad, &bad_sectors)) {
977df362 2646 set_bit(R5_MadeGood, &sh->dev[i].flags);
c0b32972
N		 2647 if (test_bit(R5_ReadError, &sh->dev[i].flags))
2648 /* That was a successful write so make
2649 * sure it looks like we already did
2650 * a re-write.
2651 */
2652 set_bit(R5_ReWrite, &sh->dev[i].flags);
2653 }
977df362
N		 2654 }
2655 rdev_dec_pending(rdev, conf->mddev);
1da177e4 2656
4e4cbee9 2657 if (sh->batch_head && bi->bi_status && !replacement)
72ac7330 2658 set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state);
2659
c9445555 2660 bio_reset(bi);
977df362
N		 2661 if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags))
2662 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1da177e4 2663 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 2664 raid5_release_stripe(sh);
59fc630b 2665
2666 if (sh->batch_head && sh != sh->batch_head)
6d036f7d 2667 raid5_release_stripe(sh->batch_head);
1da177e4
LT		 2668}
2669
849674e4 2670static void raid5_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 2671{
2672 char b[BDEVNAME_SIZE];
d1688a6d 2673 struct r5conf *conf = mddev->private;
908f4fbd 2674 unsigned long flags;
0c55e022 2675 pr_debug("raid456: error called\n");
1da177e4 2676
908f4fbd 2677 spin_lock_irqsave(&conf->device_lock, flags);
fb73b357
MT		 2678
2679 if (test_bit(In_sync, &rdev->flags) &&
2680 mddev->degraded == conf->max_degraded) {
2681 /*
2682 * Don't allow to achieve failed state
2683 * Don't try to recover this device
2684 */
2685 conf->recovery_disabled = mddev->recovery_disabled;
2686 spin_unlock_irqrestore(&conf->device_lock, flags);
2687 return;
2688 }
2689
aff69d89 2690 set_bit(Faulty, &rdev->flags);
908f4fbd 2691 clear_bit(In_sync, &rdev->flags);
2e38a37f 2692 mddev->degraded = raid5_calc_degraded(conf);
908f4fbd
N		 2693 spin_unlock_irqrestore(&conf->device_lock, flags);
2694 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2695
de393cde 2696 set_bit(Blocked, &rdev->flags);
2953079c
SL		 2697 set_mask_bits(&mddev->sb_flags, 0,
2698 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING));
cc6167b4
N		 2699 pr_crit("md/raid:%s: Disk failure on %s, disabling device.\n"
2700 "md/raid:%s: Operation continuing on %d devices.\n",
2701 mdname(mddev),
2702 bdevname(rdev->bdev, b),
2703 mdname(mddev),
2704 conf->raid_disks - mddev->degraded);
70d466f7 2705 r5c_update_on_rdev_error(mddev, rdev);
16a53ecc 2706}
1da177e4
LT		 2707
2708/*
2709 * Input: a 'big' sector number,
2710 * Output: index of the data and parity disk, and the sector # in them.
2711 */
6d036f7d
SL		 2712sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector,
2713 int previous, int *dd_idx,
2714 struct stripe_head *sh)
1da177e4 2715{
6e3b96ed 2716 sector_t stripe, stripe2;
35f2a591 2717 sector_t chunk_number;
1da177e4 2718 unsigned int chunk_offset;
911d4ee8 2719 int pd_idx, qd_idx;
67cc2b81 2720 int ddf_layout = 0;
1da177e4 2721 sector_t new_sector;
e183eaed
N		 2722 int algorithm = previous ? conf->prev_algo
2723 : conf->algorithm;
09c9e5fa
AN		 2724 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2725 : conf->chunk_sectors;
112bf897
N		 2726 int raid_disks = previous ? conf->previous_raid_disks
2727 : conf->raid_disks;
2728 int data_disks = raid_disks - conf->max_degraded;
1da177e4
LT		 2729
2730 /* First compute the information on this sector */
2731
2732 /*
2733 * Compute the chunk number and the sector offset inside the chunk
2734 */
2735 chunk_offset = sector_div(r_sector, sectors_per_chunk);
2736 chunk_number = r_sector;
1da177e4
LT		 2737
2738 /*
2739 * Compute the stripe number
2740 */
35f2a591
N		 2741 stripe = chunk_number;
2742 *dd_idx = sector_div(stripe, data_disks);
6e3b96ed 2743 stripe2 = stripe;
1da177e4
LT		 2744 /*
2745 * Select the parity disk based on the user selected algorithm.
2746 */
84789554 2747 pd_idx = qd_idx = -1;
16a53ecc
N		 2748 switch(conf->level) {
2749 case 4:
911d4ee8 2750 pd_idx = data_disks;
16a53ecc
N		 2751 break;
2752 case 5:
e183eaed 2753 switch (algorithm) {
1da177e4 2754 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2755 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2756 if (*dd_idx >= pd_idx)
1da177e4
LT		 2757 (*dd_idx)++;
2758 break;
2759 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2760 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2761 if (*dd_idx >= pd_idx)
1da177e4
LT		 2762 (*dd_idx)++;
2763 break;
2764 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2765 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2766 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4
LT		 2767 break;
2768 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2769 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2770 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4 2771 break;
99c0fb5f
N		 2772 case ALGORITHM_PARITY_0:
2773 pd_idx = 0;
2774 (*dd_idx)++;
2775 break;
2776 case ALGORITHM_PARITY_N:
2777 pd_idx = data_disks;
2778 break;
1da177e4 2779 default:
99c0fb5f 2780 BUG();
16a53ecc
N		 2781 }
2782 break;
2783 case 6:
2784
e183eaed 2785 switch (algorithm) {
16a53ecc 2786 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2787 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2788 qd_idx = pd_idx + 1;
2789 if (pd_idx == raid_disks-1) {
99c0fb5f 2790 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2791 qd_idx = 0;
2792 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2793 (*dd_idx) += 2; /* D D P Q D */
2794 break;
2795 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2796 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2797 qd_idx = pd_idx + 1;
2798 if (pd_idx == raid_disks-1) {
99c0fb5f 2799 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2800 qd_idx = 0;
2801 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2802 (*dd_idx) += 2; /* D D P Q D */
2803 break;
2804 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2805 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2806 qd_idx = (pd_idx + 1) % raid_disks;
2807 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc
N		 2808 break;
2809 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2810 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2811 qd_idx = (pd_idx + 1) % raid_disks;
2812 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc 2813 break;
99c0fb5f
N		 2814
2815 case ALGORITHM_PARITY_0:
2816 pd_idx = 0;
2817 qd_idx = 1;
2818 (*dd_idx) += 2;
2819 break;
2820 case ALGORITHM_PARITY_N:
2821 pd_idx = data_disks;
2822 qd_idx = data_disks + 1;
2823 break;
2824
2825 case ALGORITHM_ROTATING_ZERO_RESTART:
2826 /* Exactly the same as RIGHT_ASYMMETRIC, but or
2827 * of blocks for computing Q is different.
2828 */
6e3b96ed 2829 pd_idx = sector_div(stripe2, raid_disks);
99c0fb5f
N		 2830 qd_idx = pd_idx + 1;
2831 if (pd_idx == raid_disks-1) {
2832 (*dd_idx)++; /* Q D D D P */
2833 qd_idx = 0;
2834 } else if (*dd_idx >= pd_idx)
2835 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2836 ddf_layout = 1;
99c0fb5f
N		 2837 break;
2838
2839 case ALGORITHM_ROTATING_N_RESTART:
2840 /* Same a left_asymmetric, by first stripe is
2841 * D D D P Q rather than
2842 * Q D D D P
2843 */
6e3b96ed
N		 2844 stripe2 += 1;
2845 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2846 qd_idx = pd_idx + 1;
2847 if (pd_idx == raid_disks-1) {
2848 (*dd_idx)++; /* Q D D D P */
2849 qd_idx = 0;
2850 } else if (*dd_idx >= pd_idx)
2851 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2852 ddf_layout = 1;
99c0fb5f
N		 2853 break;
2854
2855 case ALGORITHM_ROTATING_N_CONTINUE:
2856 /* Same as left_symmetric but Q is before P */
6e3b96ed 2857 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2858 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
2859 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
67cc2b81 2860 ddf_layout = 1;
99c0fb5f
N		 2861 break;
2862
2863 case ALGORITHM_LEFT_ASYMMETRIC_6:
2864 /* RAID5 left_asymmetric, with Q on last device */
6e3b96ed 2865 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2866 if (*dd_idx >= pd_idx)
2867 (*dd_idx)++;
2868 qd_idx = raid_disks - 1;
2869 break;
2870
2871 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6e3b96ed 2872 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2873 if (*dd_idx >= pd_idx)
2874 (*dd_idx)++;
2875 qd_idx = raid_disks - 1;
2876 break;
2877
2878 case ALGORITHM_LEFT_SYMMETRIC_6:
6e3b96ed 2879 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2880 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2881 qd_idx = raid_disks - 1;
2882 break;
2883
2884 case ALGORITHM_RIGHT_SYMMETRIC_6:
6e3b96ed 2885 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2886 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2887 qd_idx = raid_disks - 1;
2888 break;
2889
2890 case ALGORITHM_PARITY_0_6:
2891 pd_idx = 0;
2892 (*dd_idx)++;
2893 qd_idx = raid_disks - 1;
2894 break;
2895
16a53ecc 2896 default:
99c0fb5f 2897 BUG();
16a53ecc
N		 2898 }
2899 break;
1da177e4
LT		 2900 }
2901
911d4ee8
N		 2902 if (sh) {
2903 sh->pd_idx = pd_idx;
2904 sh->qd_idx = qd_idx;
67cc2b81 2905 sh->ddf_layout = ddf_layout;
911d4ee8 2906 }
1da177e4
LT		 2907 /*
2908 * Finally, compute the new sector number
2909 */
2910 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
2911 return new_sector;
2912}
2913
6d036f7d 2914sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous)
1da177e4 2915{
d1688a6d 2916 struct r5conf *conf = sh->raid_conf;
b875e531
N		 2917 int raid_disks = sh->disks;
2918 int data_disks = raid_disks - conf->max_degraded;
1da177e4 2919 sector_t new_sector = sh->sector, check;
09c9e5fa
AN		 2920 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2921 : conf->chunk_sectors;
e183eaed
N		 2922 int algorithm = previous ? conf->prev_algo
2923 : conf->algorithm;
1da177e4
LT		 2924 sector_t stripe;
2925 int chunk_offset;
35f2a591
N		 2926 sector_t chunk_number;
2927 int dummy1, dd_idx = i;
1da177e4 2928 sector_t r_sector;
911d4ee8 2929 struct stripe_head sh2;
1da177e4
LT		 2930
2931 chunk_offset = sector_div(new_sector, sectors_per_chunk);
2932 stripe = new_sector;
1da177e4 2933
16a53ecc
N		 2934 if (i == sh->pd_idx)
2935 return 0;
2936 switch(conf->level) {
2937 case 4: break;
2938 case 5:
e183eaed 2939 switch (algorithm) {
1da177e4
LT		 2940 case ALGORITHM_LEFT_ASYMMETRIC:
2941 case ALGORITHM_RIGHT_ASYMMETRIC:
2942 if (i > sh->pd_idx)
2943 i--;
2944 break;
2945 case ALGORITHM_LEFT_SYMMETRIC:
2946 case ALGORITHM_RIGHT_SYMMETRIC:
2947 if (i < sh->pd_idx)
2948 i += raid_disks;
2949 i -= (sh->pd_idx + 1);
2950 break;
99c0fb5f
N		 2951 case ALGORITHM_PARITY_0:
2952 i -= 1;
2953 break;
2954 case ALGORITHM_PARITY_N:
2955 break;
1da177e4 2956 default:
99c0fb5f 2957 BUG();
16a53ecc
N		 2958 }
2959 break;
2960 case 6:
d0dabf7e 2961 if (i == sh->qd_idx)
16a53ecc 2962 return 0; /* It is the Q disk */
e183eaed 2963 switch (algorithm) {
16a53ecc
N		 2964 case ALGORITHM_LEFT_ASYMMETRIC:
2965 case ALGORITHM_RIGHT_ASYMMETRIC:
99c0fb5f
N		 2966 case ALGORITHM_ROTATING_ZERO_RESTART:
2967 case ALGORITHM_ROTATING_N_RESTART:
2968 if (sh->pd_idx == raid_disks-1)
2969 i--; /* Q D D D P */
16a53ecc
N		 2970 else if (i > sh->pd_idx)
2971 i -= 2; /* D D P Q D */
2972 break;
2973 case ALGORITHM_LEFT_SYMMETRIC:
2974 case ALGORITHM_RIGHT_SYMMETRIC:
2975 if (sh->pd_idx == raid_disks-1)
2976 i--; /* Q D D D P */
2977 else {
2978 /* D D P Q D */
2979 if (i < sh->pd_idx)
2980 i += raid_disks;
2981 i -= (sh->pd_idx + 2);
2982 }
2983 break;
99c0fb5f
N		 2984 case ALGORITHM_PARITY_0:
2985 i -= 2;
2986 break;
2987 case ALGORITHM_PARITY_N:
2988 break;
2989 case ALGORITHM_ROTATING_N_CONTINUE:
e4424fee 2990 /* Like left_symmetric, but P is before Q */
99c0fb5f
N		 2991 if (sh->pd_idx == 0)
2992 i--; /* P D D D Q */
e4424fee
N		 2993 else {
2994 /* D D Q P D */
2995 if (i < sh->pd_idx)
2996 i += raid_disks;
2997 i -= (sh->pd_idx + 1);
2998 }
99c0fb5f
N		 2999 break;
3000 case ALGORITHM_LEFT_ASYMMETRIC_6:
3001 case ALGORITHM_RIGHT_ASYMMETRIC_6:
3002 if (i > sh->pd_idx)
3003 i--;
3004 break;
3005 case ALGORITHM_LEFT_SYMMETRIC_6:
3006 case ALGORITHM_RIGHT_SYMMETRIC_6:
3007 if (i < sh->pd_idx)
3008 i += data_disks + 1;
3009 i -= (sh->pd_idx + 1);
3010 break;
3011 case ALGORITHM_PARITY_0_6:
3012 i -= 1;
3013 break;
16a53ecc 3014 default:
99c0fb5f 3015 BUG();
16a53ecc
N		 3016 }
3017 break;
1da177e4
LT		 3018 }
3019
3020 chunk_number = stripe * data_disks + i;
35f2a591 3021 r_sector = chunk_number * sectors_per_chunk + chunk_offset;
1da177e4 3022
112bf897 3023 check = raid5_compute_sector(conf, r_sector,
784052ec 3024 previous, &dummy1, &sh2);
911d4ee8
N		 3025 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
3026 || sh2.qd_idx != sh->qd_idx) {
cc6167b4
N		 3027 pr_warn("md/raid:%s: compute_blocknr: map not correct\n",
3028 mdname(conf->mddev));
1da177e4
LT		 3029 return 0;
3030 }
3031 return r_sector;
3032}
3033
07e83364
SL		 3034/*
3035 * There are cases where we want handle_stripe_dirtying() and
3036 * schedule_reconstruction() to delay towrite to some dev of a stripe.
3037 *
3038 * This function checks whether we want to delay the towrite. Specifically,
3039 * we delay the towrite when:
3040 *
3041 * 1. degraded stripe has a non-overwrite to the missing dev, AND this
3042 * stripe has data in journal (for other devices).
3043 *
3044 * In this case, when reading data for the non-overwrite dev, it is
3045 * necessary to handle complex rmw of write back cache (prexor with
3046 * orig_page, and xor with page). To keep read path simple, we would
3047 * like to flush data in journal to RAID disks first, so complex rmw
3048 * is handled in the write patch (handle_stripe_dirtying).
3049 *
39b99586
SL		 3050 * 2. when journal space is critical (R5C_LOG_CRITICAL=1)
3051 *
3052 * It is important to be able to flush all stripes in raid5-cache.
3053 * Therefore, we need reserve some space on the journal device for
3054 * these flushes. If flush operation includes pending writes to the
3055 * stripe, we need to reserve (conf->raid_disk + 1) pages per stripe
3056 * for the flush out. If we exclude these pending writes from flush
3057 * operation, we only need (conf->max_degraded + 1) pages per stripe.
3058 * Therefore, excluding pending writes in these cases enables more
3059 * efficient use of the journal device.
3060 *
3061 * Note: To make sure the stripe makes progress, we only delay
3062 * towrite for stripes with data already in journal (injournal > 0).
3063 * When LOG_CRITICAL, stripes with injournal == 0 will be sent to
3064 * no_space_stripes list.
3065 *
70d466f7
SL		 3066 * 3. during journal failure
3067 * In journal failure, we try to flush all cached data to raid disks
3068 * based on data in stripe cache. The array is read-only to upper
3069 * layers, so we would skip all pending writes.
3070 *
07e83364 3071 */
39b99586
SL		 3072static inline bool delay_towrite(struct r5conf *conf,
3073 struct r5dev *dev,
3074 struct stripe_head_state *s)
07e83364 3075{
39b99586
SL		 3076 /* case 1 above */
3077 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
3078 !test_bit(R5_Insync, &dev->flags) && s->injournal)
3079 return true;
3080 /* case 2 above */
3081 if (test_bit(R5C_LOG_CRITICAL, &conf->cache_state) &&
3082 s->injournal > 0)
3083 return true;
70d466f7
SL		 3084 /* case 3 above */
3085 if (s->log_failed && s->injournal)
3086 return true;
39b99586 3087 return false;
07e83364
SL		 3088}
3089
600aa109 3090static void
c0f7bddb 3091schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
600aa109 3092 int rcw, int expand)
e33129d8 3093{
584acdd4 3094 int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx, disks = sh->disks;
d1688a6d 3095 struct r5conf *conf = sh->raid_conf;
c0f7bddb 3096 int level = conf->level;
e33129d8
DW		 3097
3098 if (rcw) {
1e6d690b
SL		 3099 /*
3100 * In some cases, handle_stripe_dirtying initially decided to
3101 * run rmw and allocates extra page for prexor. However, rcw is
3102 * cheaper later on. We need to free the extra page now,
3103 * because we won't be able to do that in ops_complete_prexor().
3104 */
3105 r5c_release_extra_page(sh);
e33129d8
DW		 3106
3107 for (i = disks; i--; ) {
3108 struct r5dev *dev = &sh->dev[i];
3109
39b99586 3110 if (dev->towrite && !delay_towrite(conf, dev, s)) {
e33129d8 3111 set_bit(R5_LOCKED, &dev->flags);
d8ee0728 3112 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 3113 if (!expand)
3114 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 3115 s->locked++;
1e6d690b
SL		 3116 } else if (test_bit(R5_InJournal, &dev->flags)) {
3117 set_bit(R5_LOCKED, &dev->flags);
3118 s->locked++;
e33129d8
DW		 3119 }
3120 }
ce7d363a
N		 3121 /* if we are not expanding this is a proper write request, and
3122 * there will be bios with new data to be drained into the
3123 * stripe cache
3124 */
3125 if (!expand) {
3126 if (!s->locked)
3127 /* False alarm, nothing to do */
3128 return;
3129 sh->reconstruct_state = reconstruct_state_drain_run;
3130 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
3131 } else
3132 sh->reconstruct_state = reconstruct_state_run;
3133
3134 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
3135
c0f7bddb 3136 if (s->locked + conf->max_degraded == disks)
8b3e6cdc 3137 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
c0f7bddb 3138 atomic_inc(&conf->pending_full_writes);
e33129d8
DW		 3139 } else {
3140 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
3141 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
584acdd4
MS		 3142 BUG_ON(level == 6 &&
3143 (!(test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags) ||
3144 test_bit(R5_Wantcompute, &sh->dev[qd_idx].flags))));
e33129d8 3145
e33129d8
DW		 3146 for (i = disks; i--; ) {
3147 struct r5dev *dev = &sh->dev[i];
584acdd4 3148 if (i == pd_idx || i == qd_idx)
e33129d8
DW		 3149 continue;
3150
e33129d8
DW		 3151 if (dev->towrite &&
3152 (test_bit(R5_UPTODATE, &dev->flags) ||
d8ee0728
DW		 3153 test_bit(R5_Wantcompute, &dev->flags))) {
3154 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 3155 set_bit(R5_LOCKED, &dev->flags);
3156 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 3157 s->locked++;
1e6d690b
SL		 3158 } else if (test_bit(R5_InJournal, &dev->flags)) {
3159 set_bit(R5_LOCKED, &dev->flags);
3160 s->locked++;
e33129d8
DW		 3161 }
3162 }
ce7d363a
N		 3163 if (!s->locked)
3164 /* False alarm - nothing to do */
3165 return;
3166 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
3167 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
3168 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
3169 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
e33129d8
DW		 3170 }
3171
c0f7bddb 3172 /* keep the parity disk(s) locked while asynchronous operations
e33129d8
DW		 3173 * are in flight
3174 */
3175 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
3176 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
600aa109 3177 s->locked++;
e33129d8 3178
c0f7bddb
YT		 3179 if (level == 6) {
3180 int qd_idx = sh->qd_idx;
3181 struct r5dev *dev = &sh->dev[qd_idx];
3182
3183 set_bit(R5_LOCKED, &dev->flags);
3184 clear_bit(R5_UPTODATE, &dev->flags);
3185 s->locked++;
3186 }
3187
845b9e22 3188 if (raid5_has_ppl(sh->raid_conf) && sh->ppl_page &&
3418d036
AP		 3189 test_bit(STRIPE_OP_BIODRAIN, &s->ops_request) &&
3190 !test_bit(STRIPE_FULL_WRITE, &sh->state) &&
3191 test_bit(R5_Insync, &sh->dev[pd_idx].flags))
3192 set_bit(STRIPE_OP_PARTIAL_PARITY, &s->ops_request);
3193
600aa109 3194 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
e46b272b 3195 __func__, (unsigned long long)sh->sector,
600aa109 3196 s->locked, s->ops_request);
e33129d8 3197}
16a53ecc 3198
1da177e4
LT		 3199/*
3200 * Each stripe/dev can have one or more bion attached.
16a53ecc 3201 * toread/towrite point to the first in a chain.
1da177e4
LT		 3202 * The bi_next chain must be in order.
3203 */
da41ba65 3204static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx,
3205 int forwrite, int previous)
1da177e4
LT		 3206{
3207 struct bio **bip;
d1688a6d 3208 struct r5conf *conf = sh->raid_conf;
72626685 3209 int firstwrite=0;
1da177e4 3210
cbe47ec5 3211 pr_debug("adding bi b#%llu to stripe s#%llu\n",
4f024f37 3212 (unsigned long long)bi->bi_iter.bi_sector,
1da177e4
LT		 3213 (unsigned long long)sh->sector);
3214
b17459c0 3215 spin_lock_irq(&sh->stripe_lock);
2cd259a7 3216 sh->dev[dd_idx].write_hint = bi->bi_write_hint;
59fc630b 3217 /* Don't allow new IO added to stripes in batch list */
3218 if (sh->batch_head)
3219 goto overlap;
72626685 3220 if (forwrite) {
1da177e4 3221 bip = &sh->dev[dd_idx].towrite;
7eaf7e8e 3222 if (*bip == NULL)
72626685
N		 3223 firstwrite = 1;
3224 } else
1da177e4 3225 bip = &sh->dev[dd_idx].toread;
4f024f37
KO		 3226 while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) {
3227 if (bio_end_sector(*bip) > bi->bi_iter.bi_sector)
1da177e4
LT		 3228 goto overlap;
3229 bip = & (*bip)->bi_next;
3230 }
4f024f37 3231 if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi))
1da177e4
LT		 3232 goto overlap;
3233
3418d036
AP		 3234 if (forwrite && raid5_has_ppl(conf)) {
3235 /*
3236 * With PPL only writes to consecutive data chunks within a
3237 * stripe are allowed because for a single stripe_head we can
3238 * only have one PPL entry at a time, which describes one data
3239 * range. Not really an overlap, but wait_for_overlap can be
3240 * used to handle this.
3241 */
3242 sector_t sector;
3243 sector_t first = 0;
3244 sector_t last = 0;
3245 int count = 0;
3246 int i;
3247
3248 for (i = 0; i < sh->disks; i++) {
3249 if (i != sh->pd_idx &&
3250 (i == dd_idx || sh->dev[i].towrite)) {
3251 sector = sh->dev[i].sector;
3252 if (count == 0 || sector < first)
3253 first = sector;
3254 if (sector > last)
3255 last = sector;
3256 count++;
3257 }
3258 }
3259
3260 if (first + conf->chunk_sectors * (count - 1) != last)
3261 goto overlap;
3262 }
3263
da41ba65 3264 if (!forwrite || previous)
3265 clear_bit(STRIPE_BATCH_READY, &sh->state);
3266
78bafebd 3267 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
1da177e4
LT		 3268 if (*bip)
3269 bi->bi_next = *bip;
3270 *bip = bi;
016c76ac 3271 bio_inc_remaining(bi);
49728050 3272 md_write_inc(conf->mddev, bi);
72626685 3273
1da177e4
LT		 3274 if (forwrite) {
3275 /* check if page is covered */
3276 sector_t sector = sh->dev[dd_idx].sector;
3277 for (bi=sh->dev[dd_idx].towrite;
3278 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
4f024f37 3279 bi && bi->bi_iter.bi_sector <= sector;
1da177e4 3280 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
f73a1c7d
KO		 3281 if (bio_end_sector(bi) >= sector)
3282 sector = bio_end_sector(bi);
1da177e4
LT		 3283 }
3284 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
7a87f434 3285 if (!test_and_set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags))
3286 sh->overwrite_disks++;
1da177e4 3287 }
cbe47ec5
N		 3288
3289 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
4f024f37 3290 (unsigned long long)(*bip)->bi_iter.bi_sector,
cbe47ec5
N		 3291 (unsigned long long)sh->sector, dd_idx);
3292
3293 if (conf->mddev->bitmap && firstwrite) {
d0852df5
N		 3294 /* Cannot hold spinlock over bitmap_startwrite,
3295 * but must ensure this isn't added to a batch until
3296 * we have added to the bitmap and set bm_seq.
3297 * So set STRIPE_BITMAP_PENDING to prevent
3298 * batching.
3299 * If multiple add_stripe_bio() calls race here they
3300 * much all set STRIPE_BITMAP_PENDING. So only the first one
3301 * to complete "bitmap_startwrite" gets to set
3302 * STRIPE_BIT_DELAY. This is important as once a stripe
3303 * is added to a batch, STRIPE_BIT_DELAY cannot be changed
3304 * any more.
3305 */
3306 set_bit(STRIPE_BITMAP_PENDING, &sh->state);
3307 spin_unlock_irq(&sh->stripe_lock);
e64e4018
AS		 3308 md_bitmap_startwrite(conf->mddev->bitmap, sh->sector,
3309 STRIPE_SECTORS, 0);
d0852df5
N		 3310 spin_lock_irq(&sh->stripe_lock);
3311 clear_bit(STRIPE_BITMAP_PENDING, &sh->state);
3312 if (!sh->batch_head) {
3313 sh->bm_seq = conf->seq_flush+1;
3314 set_bit(STRIPE_BIT_DELAY, &sh->state);
3315 }
cbe47ec5 3316 }
d0852df5 3317 spin_unlock_irq(&sh->stripe_lock);
59fc630b 3318
3319 if (stripe_can_batch(sh))
3320 stripe_add_to_batch_list(conf, sh);
1da177e4
LT		 3321 return 1;
3322
3323 overlap:
3324 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
b17459c0 3325 spin_unlock_irq(&sh->stripe_lock);
1da177e4
LT		 3326 return 0;
3327}
3328
d1688a6d 3329static void end_reshape(struct r5conf *conf);
29269553 3330
d1688a6d 3331static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
911d4ee8 3332 struct stripe_head *sh)
ccfcc3c1 3333{
784052ec 3334 int sectors_per_chunk =
09c9e5fa 3335 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
911d4ee8 3336 int dd_idx;
2d2063ce 3337 int chunk_offset = sector_div(stripe, sectors_per_chunk);
112bf897 3338 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
2d2063ce 3339
112bf897
N		 3340 raid5_compute_sector(conf,
3341 stripe * (disks - conf->max_degraded)
b875e531 3342 *sectors_per_chunk + chunk_offset,
112bf897 3343 previous,
911d4ee8 3344 &dd_idx, sh);
ccfcc3c1
N		 3345}
3346
a4456856 3347static void
d1688a6d 3348handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh,
bd83d0a2 3349 struct stripe_head_state *s, int disks)
a4456856
DW		 3350{
3351 int i;
59fc630b 3352 BUG_ON(sh->batch_head);
a4456856
DW		 3353 for (i = disks; i--; ) {
3354 struct bio *bi;
3355 int bitmap_end = 0;
3356
3357 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
3cb03002 3358 struct md_rdev *rdev;
a4456856
DW		 3359 rcu_read_lock();
3360 rdev = rcu_dereference(conf->disks[i].rdev);
f5b67ae8
N		 3361 if (rdev && test_bit(In_sync, &rdev->flags) &&
3362 !test_bit(Faulty, &rdev->flags))
7f0da59b
N		 3363 atomic_inc(&rdev->nr_pending);
3364 else
3365 rdev = NULL;
a4456856 3366 rcu_read_unlock();
7f0da59b
N		 3367 if (rdev) {
3368 if (!rdev_set_badblocks(
3369 rdev,
3370 sh->sector,
3371 STRIPE_SECTORS, 0))
3372 md_error(conf->mddev, rdev);
3373 rdev_dec_pending(rdev, conf->mddev);
3374 }
a4456856 3375 }
b17459c0 3376 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 3377 /* fail all writes first */
3378 bi = sh->dev[i].towrite;
3379 sh->dev[i].towrite = NULL;
7a87f434 3380 sh->overwrite_disks = 0;
b17459c0 3381 spin_unlock_irq(&sh->stripe_lock);
1ed850f3 3382 if (bi)
a4456856 3383 bitmap_end = 1;
a4456856 3384
ff875738 3385 log_stripe_write_finished(sh);
0576b1c6 3386
a4456856
DW		 3387 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
3388 wake_up(&conf->wait_for_overlap);
3389
4f024f37 3390 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3391 sh->dev[i].sector + STRIPE_SECTORS) {
3392 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
4246a0b6 3393
49728050 3394 md_write_end(conf->mddev);
6308d8e3 3395 bio_io_error(bi);
a4456856
DW		 3396 bi = nextbi;
3397 }
7eaf7e8e 3398 if (bitmap_end)
e64e4018
AS		 3399 md_bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3400 STRIPE_SECTORS, 0, 0);
7eaf7e8e 3401 bitmap_end = 0;
a4456856
DW		 3402 /* and fail all 'written' */
3403 bi = sh->dev[i].written;
3404 sh->dev[i].written = NULL;
d592a996
SL		 3405 if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) {
3406 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
3407 sh->dev[i].page = sh->dev[i].orig_page;
3408 }
3409
a4456856 3410 if (bi) bitmap_end = 1;
4f024f37 3411 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3412 sh->dev[i].sector + STRIPE_SECTORS) {
3413 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
4246a0b6 3414
49728050 3415 md_write_end(conf->mddev);
6308d8e3 3416 bio_io_error(bi);
a4456856
DW		 3417 bi = bi2;
3418 }
3419
b5e98d65
DW		 3420 /* fail any reads if this device is non-operational and
3421 * the data has not reached the cache yet.
3422 */
3423 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
6e74a9cf 3424 s->failed > conf->max_degraded &&
b5e98d65
DW		 3425 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
3426 test_bit(R5_ReadError, &sh->dev[i].flags))) {
143c4d05 3427 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 3428 bi = sh->dev[i].toread;
3429 sh->dev[i].toread = NULL;
143c4d05 3430 spin_unlock_irq(&sh->stripe_lock);
a4456856
DW		 3431 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
3432 wake_up(&conf->wait_for_overlap);
ebda780b
SL		 3433 if (bi)
3434 s->to_read--;
4f024f37 3435 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3436 sh->dev[i].sector + STRIPE_SECTORS) {
3437 struct bio *nextbi =
3438 r5_next_bio(bi, sh->dev[i].sector);
4246a0b6 3439
6308d8e3 3440 bio_io_error(bi);
a4456856
DW		 3441 bi = nextbi;
3442 }
3443 }
a4456856 3444 if (bitmap_end)
e64e4018
AS		 3445 md_bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3446 STRIPE_SECTORS, 0, 0);
8cfa7b0f
N		 3447 /* If we were in the middle of a write the parity block might
3448 * still be locked - so just clear all R5_LOCKED flags
3449 */
3450 clear_bit(R5_LOCKED, &sh->dev[i].flags);
a4456856 3451 }
ebda780b
SL		 3452 s->to_write = 0;
3453 s->written = 0;
a4456856 3454
8b3e6cdc
DW		 3455 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3456 if (atomic_dec_and_test(&conf->pending_full_writes))
3457 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 3458}
3459
7f0da59b 3460static void
d1688a6d 3461handle_failed_sync(struct r5conf *conf, struct stripe_head *sh,
7f0da59b
N		 3462 struct stripe_head_state *s)
3463{
3464 int abort = 0;
3465 int i;
3466
59fc630b 3467 BUG_ON(sh->batch_head);
7f0da59b 3468 clear_bit(STRIPE_SYNCING, &sh->state);
f8dfcffd
N		 3469 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
3470 wake_up(&conf->wait_for_overlap);
7f0da59b 3471 s->syncing = 0;
9a3e1101 3472 s->replacing = 0;
7f0da59b 3473 /* There is nothing more to do for sync/check/repair.
18b9837e
N		 3474 * Don't even need to abort as that is handled elsewhere
3475 * if needed, and not always wanted e.g. if there is a known
3476 * bad block here.
9a3e1101 3477 * For recover/replace we need to record a bad block on all
7f0da59b
N		 3478 * non-sync devices, or abort the recovery
3479 */
18b9837e
N		 3480 if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) {
3481 /* During recovery devices cannot be removed, so
3482 * locking and refcounting of rdevs is not needed
3483 */
e50d3992 3484 rcu_read_lock();
18b9837e 3485 for (i = 0; i < conf->raid_disks; i++) {
e50d3992 3486 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
18b9837e
N		 3487 if (rdev
3488 && !test_bit(Faulty, &rdev->flags)
3489 && !test_bit(In_sync, &rdev->flags)
3490 && !rdev_set_badblocks(rdev, sh->sector,
3491 STRIPE_SECTORS, 0))
3492 abort = 1;
e50d3992 3493 rdev = rcu_dereference(conf->disks[i].replacement);
18b9837e
N		 3494 if (rdev
3495 && !test_bit(Faulty, &rdev->flags)
3496 && !test_bit(In_sync, &rdev->flags)
3497 && !rdev_set_badblocks(rdev, sh->sector,
3498 STRIPE_SECTORS, 0))
3499 abort = 1;
3500 }
e50d3992 3501 rcu_read_unlock();
18b9837e
N		 3502 if (abort)
3503 conf->recovery_disabled =
3504 conf->mddev->recovery_disabled;
7f0da59b 3505 }
18b9837e 3506 md_done_sync(conf->mddev, STRIPE_SECTORS, !abort);
7f0da59b
N		 3507}
3508
9a3e1101
N		 3509static int want_replace(struct stripe_head *sh, int disk_idx)
3510{
3511 struct md_rdev *rdev;
3512 int rv = 0;
3f232d6a
N		 3513
3514 rcu_read_lock();
3515 rdev = rcu_dereference(sh->raid_conf->disks[disk_idx].replacement);
9a3e1101
N		 3516 if (rdev
3517 && !test_bit(Faulty, &rdev->flags)
3518 && !test_bit(In_sync, &rdev->flags)
3519 && (rdev->recovery_offset <= sh->sector
3520 || rdev->mddev->recovery_cp <= sh->sector))
3521 rv = 1;
3f232d6a 3522 rcu_read_unlock();
9a3e1101
N		 3523 return rv;
3524}
3525
2c58f06e
N		 3526static int need_this_block(struct stripe_head *sh, struct stripe_head_state *s,
3527 int disk_idx, int disks)
a4456856 3528{
5599becc 3529 struct r5dev *dev = &sh->dev[disk_idx];
f2b3b44d
N		 3530 struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]],
3531 &sh->dev[s->failed_num[1]] };
ea664c82 3532 int i;
5599becc 3533
a79cfe12
N		 3534
3535 if (test_bit(R5_LOCKED, &dev->flags) ||
3536 test_bit(R5_UPTODATE, &dev->flags))
3537 /* No point reading this as we already have it or have
3538 * decided to get it.
3539 */
3540 return 0;
3541
3542 if (dev->toread ||
3543 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)))
3544 /* We need this block to directly satisfy a request */
3545 return 1;
3546
3547 if (s->syncing || s->expanding ||
3548 (s->replacing && want_replace(sh, disk_idx)))
3549 /* When syncing, or expanding we read everything.
3550 * When replacing, we need the replaced block.
3551 */
3552 return 1;
3553
3554 if ((s->failed >= 1 && fdev[0]->toread) ||
3555 (s->failed >= 2 && fdev[1]->toread))
3556 /* If we want to read from a failed device, then
3557 * we need to actually read every other device.
3558 */
3559 return 1;
3560
a9d56950
N		 3561 /* Sometimes neither read-modify-write nor reconstruct-write
3562 * cycles can work. In those cases we read every block we
3563 * can. Then the parity-update is certain to have enough to
3564 * work with.
3565 * This can only be a problem when we need to write something,
3566 * and some device has failed. If either of those tests
3567 * fail we need look no further.
3568 */
3569 if (!s->failed || !s->to_write)
3570 return 0;
3571
3572 if (test_bit(R5_Insync, &dev->flags) &&
3573 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3574 /* Pre-reads at not permitted until after short delay
3575 * to gather multiple requests. However if this
3560741e 3576 * device is no Insync, the block could only be computed
a9d56950
N		 3577 * and there is no need to delay that.
3578 */
3579 return 0;
ea664c82 3580
36707bb2 3581 for (i = 0; i < s->failed && i < 2; i++) {
ea664c82
N		 3582 if (fdev[i]->towrite &&
3583 !test_bit(R5_UPTODATE, &fdev[i]->flags) &&
3584 !test_bit(R5_OVERWRITE, &fdev[i]->flags))
3585 /* If we have a partial write to a failed
3586 * device, then we will need to reconstruct
3587 * the content of that device, so all other
3588 * devices must be read.
3589 */
3590 return 1;
3591 }
3592
3593 /* If we are forced to do a reconstruct-write, either because
3594 * the current RAID6 implementation only supports that, or
3560741e 3595 * because parity cannot be trusted and we are currently
ea664c82
N		 3596 * recovering it, there is extra need to be careful.
3597 * If one of the devices that we would need to read, because
3598 * it is not being overwritten (and maybe not written at all)
3599 * is missing/faulty, then we need to read everything we can.
3600 */
3601 if (sh->raid_conf->level != 6 &&
3602 sh->sector < sh->raid_conf->mddev->recovery_cp)
3603 /* reconstruct-write isn't being forced */
3604 return 0;
36707bb2 3605 for (i = 0; i < s->failed && i < 2; i++) {
10d82c5f
N		 3606 if (s->failed_num[i] != sh->pd_idx &&
3607 s->failed_num[i] != sh->qd_idx &&
3608 !test_bit(R5_UPTODATE, &fdev[i]->flags) &&
ea664c82
N		 3609 !test_bit(R5_OVERWRITE, &fdev[i]->flags))
3610 return 1;
3611 }
3612
2c58f06e
N		 3613 return 0;
3614}
3615
ba02684d
SL		 3616/* fetch_block - checks the given member device to see if its data needs
3617 * to be read or computed to satisfy a request.
3618 *
3619 * Returns 1 when no more member devices need to be checked, otherwise returns
3620 * 0 to tell the loop in handle_stripe_fill to continue
3621 */
2c58f06e
N		 3622static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s,
3623 int disk_idx, int disks)
3624{
3625 struct r5dev *dev = &sh->dev[disk_idx];
3626
3627 /* is the data in this block needed, and can we get it? */
3628 if (need_this_block(sh, s, disk_idx, disks)) {
5599becc
YT		 3629 /* we would like to get this block, possibly by computing it,
3630 * otherwise read it if the backing disk is insync
3631 */
3632 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
3633 BUG_ON(test_bit(R5_Wantread, &dev->flags));
b0c783b3 3634 BUG_ON(sh->batch_head);
7471fb77
N		 3635
3636 /*
3637 * In the raid6 case if the only non-uptodate disk is P
3638 * then we already trusted P to compute the other failed
3639 * drives. It is safe to compute rather than re-read P.
3640 * In other cases we only compute blocks from failed
3641 * devices, otherwise check/repair might fail to detect
3642 * a real inconsistency.
3643 */
3644
5599becc 3645 if ((s->uptodate == disks - 1) &&
7471fb77 3646 ((sh->qd_idx >= 0 && sh->pd_idx == disk_idx) ||
f2b3b44d 3647 (s->failed && (disk_idx == s->failed_num[0] ||
7471fb77 3648 disk_idx == s->failed_num[1])))) {
5599becc
YT		 3649 /* have disk failed, and we're requested to fetch it;
3650 * do compute it
a4456856 3651 */
5599becc
YT		 3652 pr_debug("Computing stripe %llu block %d\n",
3653 (unsigned long long)sh->sector, disk_idx);
3654 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3655 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3656 set_bit(R5_Wantcompute, &dev->flags);
3657 sh->ops.target = disk_idx;
3658 sh->ops.target2 = -1; /* no 2nd target */
3659 s->req_compute = 1;
93b3dbce
N		 3660 /* Careful: from this point on 'uptodate' is in the eye
3661 * of raid_run_ops which services 'compute' operations
3662 * before writes. R5_Wantcompute flags a block that will
3663 * be R5_UPTODATE by the time it is needed for a
3664 * subsequent operation.
3665 */
5599becc
YT		 3666 s->uptodate++;
3667 return 1;
3668 } else if (s->uptodate == disks-2 && s->failed >= 2) {
3669 /* Computing 2-failure is *very* expensive; only
3670 * do it if failed >= 2
3671 */
3672 int other;
3673 for (other = disks; other--; ) {
3674 if (other == disk_idx)
3675 continue;
3676 if (!test_bit(R5_UPTODATE,
3677 &sh->dev[other].flags))
3678 break;
a4456856 3679 }
5599becc
YT		 3680 BUG_ON(other < 0);
3681 pr_debug("Computing stripe %llu blocks %d,%d\n",
3682 (unsigned long long)sh->sector,
3683 disk_idx, other);
3684 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3685 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3686 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
3687 set_bit(R5_Wantcompute, &sh->dev[other].flags);
3688 sh->ops.target = disk_idx;
3689 sh->ops.target2 = other;
3690 s->uptodate += 2;
3691 s->req_compute = 1;
3692 return 1;
3693 } else if (test_bit(R5_Insync, &dev->flags)) {
3694 set_bit(R5_LOCKED, &dev->flags);
3695 set_bit(R5_Wantread, &dev->flags);
3696 s->locked++;
3697 pr_debug("Reading block %d (sync=%d)\n",
3698 disk_idx, s->syncing);
a4456856
DW		 3699 }
3700 }
5599becc
YT		 3701
3702 return 0;
3703}
3704
3705/**
93b3dbce 3706 * handle_stripe_fill - read or compute data to satisfy pending requests.
5599becc 3707 */
93b3dbce
N		 3708static void handle_stripe_fill(struct stripe_head *sh,
3709 struct stripe_head_state *s,
3710 int disks)
5599becc
YT		 3711{
3712 int i;
3713
3714 /* look for blocks to read/compute, skip this if a compute
3715 * is already in flight, or if the stripe contents are in the
3716 * midst of changing due to a write
3717 */
3718 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
07e83364
SL		 3719 !sh->reconstruct_state) {
3720
3721 /*
3722 * For degraded stripe with data in journal, do not handle
3723 * read requests yet, instead, flush the stripe to raid
3724 * disks first, this avoids handling complex rmw of write
3725 * back cache (prexor with orig_page, and then xor with
3726 * page) in the read path
3727 */
3728 if (s->injournal && s->failed) {
3729 if (test_bit(STRIPE_R5C_CACHING, &sh->state))
3730 r5c_make_stripe_write_out(sh);
3731 goto out;
3732 }
3733
5599becc 3734 for (i = disks; i--; )
93b3dbce 3735 if (fetch_block(sh, s, i, disks))
5599becc 3736 break;
07e83364
SL		 3737 }
3738out:
a4456856
DW		 3739 set_bit(STRIPE_HANDLE, &sh->state);
3740}
3741
787b76fa
N		 3742static void break_stripe_batch_list(struct stripe_head *head_sh,
3743 unsigned long handle_flags);
1fe797e6 3744/* handle_stripe_clean_event
a4456856
DW		 3745 * any written block on an uptodate or failed drive can be returned.
3746 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
3747 * never LOCKED, so we don't need to test 'failed' directly.
3748 */
d1688a6d 3749static void handle_stripe_clean_event(struct r5conf *conf,
bd83d0a2 3750 struct stripe_head *sh, int disks)
a4456856
DW		 3751{
3752 int i;
3753 struct r5dev *dev;
f8dfcffd 3754 int discard_pending = 0;
59fc630b 3755 struct stripe_head *head_sh = sh;
3756 bool do_endio = false;
a4456856
DW		 3757
3758 for (i = disks; i--; )
3759 if (sh->dev[i].written) {
3760 dev = &sh->dev[i];
3761 if (!test_bit(R5_LOCKED, &dev->flags) &&
9e444768 3762 (test_bit(R5_UPTODATE, &dev->flags) ||
d592a996
SL		 3763 test_bit(R5_Discard, &dev->flags) ||
3764 test_bit(R5_SkipCopy, &dev->flags))) {
a4456856
DW		 3765 /* We can return any write requests */
3766 struct bio *wbi, *wbi2;
45b4233c 3767 pr_debug("Return write for disc %d\n", i);
ca64cae9
N		 3768 if (test_and_clear_bit(R5_Discard, &dev->flags))
3769 clear_bit(R5_UPTODATE, &dev->flags);
d592a996
SL		 3770 if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) {
3771 WARN_ON(test_bit(R5_UPTODATE, &dev->flags));
d592a996 3772 }
59fc630b 3773 do_endio = true;
3774
3775returnbi:
3776 dev->page = dev->orig_page;
a4456856
DW		 3777 wbi = dev->written;
3778 dev->written = NULL;
4f024f37 3779 while (wbi && wbi->bi_iter.bi_sector <
a4456856
DW		 3780 dev->sector + STRIPE_SECTORS) {
3781 wbi2 = r5_next_bio(wbi, dev->sector);
49728050 3782 md_write_end(conf->mddev);
016c76ac 3783 bio_endio(wbi);
a4456856
DW		 3784 wbi = wbi2;
3785 }
e64e4018
AS		 3786 md_bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3787 STRIPE_SECTORS,
3788 !test_bit(STRIPE_DEGRADED, &sh->state),
3789 0);
59fc630b 3790 if (head_sh->batch_head) {
3791 sh = list_first_entry(&sh->batch_list,
3792 struct stripe_head,
3793 batch_list);
3794 if (sh != head_sh) {
3795 dev = &sh->dev[i];
3796 goto returnbi;
3797 }
3798 }
3799 sh = head_sh;
3800 dev = &sh->dev[i];
f8dfcffd
N		 3801 } else if (test_bit(R5_Discard, &dev->flags))
3802 discard_pending = 1;
3803 }
f6bed0ef 3804
ff875738 3805 log_stripe_write_finished(sh);
0576b1c6 3806
f8dfcffd
N		 3807 if (!discard_pending &&
3808 test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) {
b8a9d66d 3809 int hash;
f8dfcffd
N		 3810 clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
3811 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
3812 if (sh->qd_idx >= 0) {
3813 clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
3814 clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags);
3815 }
3816 /* now that discard is done we can proceed with any sync */
3817 clear_bit(STRIPE_DISCARD, &sh->state);
d47648fc
SL		 3818 /*
3819 * SCSI discard will change some bio fields and the stripe has
3820 * no updated data, so remove it from hash list and the stripe
3821 * will be reinitialized
3822 */
59fc630b 3823unhash:
b8a9d66d
RG		 3824 hash = sh->hash_lock_index;
3825 spin_lock_irq(conf->hash_locks + hash);
d47648fc 3826 remove_hash(sh);
b8a9d66d 3827 spin_unlock_irq(conf->hash_locks + hash);
59fc630b 3828 if (head_sh->batch_head) {
3829 sh = list_first_entry(&sh->batch_list,
3830 struct stripe_head, batch_list);
3831 if (sh != head_sh)
3832 goto unhash;
3833 }
59fc630b 3834 sh = head_sh;
3835
f8dfcffd
N		 3836 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state))
3837 set_bit(STRIPE_HANDLE, &sh->state);
3838
3839 }
8b3e6cdc
DW		 3840
3841 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3842 if (atomic_dec_and_test(&conf->pending_full_writes))
3843 md_wakeup_thread(conf->mddev->thread);
59fc630b 3844
787b76fa
N		 3845 if (head_sh->batch_head && do_endio)
3846 break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS);
a4456856
DW		 3847}
3848
86aa1397
SL		 3849/*
3850 * For RMW in write back cache, we need extra page in prexor to store the
3851 * old data. This page is stored in dev->orig_page.
3852 *
3853 * This function checks whether we have data for prexor. The exact logic
3854 * is:
3855 * R5_UPTODATE && (!R5_InJournal || R5_OrigPageUPTDODATE)
3856 */
3857static inline bool uptodate_for_rmw(struct r5dev *dev)
3858{
3859 return (test_bit(R5_UPTODATE, &dev->flags)) &&
3860 (!test_bit(R5_InJournal, &dev->flags) ||
3861 test_bit(R5_OrigPageUPTDODATE, &dev->flags));
3862}
3863
d7bd398e
SL		 3864static int handle_stripe_dirtying(struct r5conf *conf,
3865 struct stripe_head *sh,
3866 struct stripe_head_state *s,
3867 int disks)
a4456856
DW		 3868{
3869 int rmw = 0, rcw = 0, i;
a7854487
AL		 3870 sector_t recovery_cp = conf->mddev->recovery_cp;
3871
584acdd4 3872 /* Check whether resync is now happening or should start.
a7854487
AL		 3873 * If yes, then the array is dirty (after unclean shutdown or
3874 * initial creation), so parity in some stripes might be inconsistent.
3875 * In this case, we need to always do reconstruct-write, to ensure
3876 * that in case of drive failure or read-error correction, we
3877 * generate correct data from the parity.
3878 */
584acdd4 3879 if (conf->rmw_level == PARITY_DISABLE_RMW ||
26ac1073
N		 3880 (recovery_cp < MaxSector && sh->sector >= recovery_cp &&
3881 s->failed == 0)) {
a7854487 3882 /* Calculate the real rcw later - for now make it
c8ac1803
N		 3883 * look like rcw is cheaper
3884 */
3885 rcw = 1; rmw = 2;
584acdd4
MS		 3886 pr_debug("force RCW rmw_level=%u, recovery_cp=%llu sh->sector=%llu\n",
3887 conf->rmw_level, (unsigned long long)recovery_cp,
a7854487 3888 (unsigned long long)sh->sector);
c8ac1803 3889 } else for (i = disks; i--; ) {
a4456856
DW		 3890 /* would I have to read this buffer for read_modify_write */
3891 struct r5dev *dev = &sh->dev[i];
39b99586 3892 if (((dev->towrite && !delay_towrite(conf, dev, s)) ||
07e83364 3893 i == sh->pd_idx || i == sh->qd_idx ||
1e6d690b 3894 test_bit(R5_InJournal, &dev->flags)) &&
a4456856 3895 !test_bit(R5_LOCKED, &dev->flags) &&
86aa1397 3896 !(uptodate_for_rmw(dev) ||
f38e1219 3897 test_bit(R5_Wantcompute, &dev->flags))) {
a4456856
DW		 3898 if (test_bit(R5_Insync, &dev->flags))
3899 rmw++;
3900 else
3901 rmw += 2*disks; /* cannot read it */
3902 }
3903 /* Would I have to read this buffer for reconstruct_write */
584acdd4
MS		 3904 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
3905 i != sh->pd_idx && i != sh->qd_idx &&
a4456856 3906 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219 3907 !(test_bit(R5_UPTODATE, &dev->flags) ||
1e6d690b 3908 test_bit(R5_Wantcompute, &dev->flags))) {
67f45548
N		 3909 if (test_bit(R5_Insync, &dev->flags))
3910 rcw++;
a4456856
DW		 3911 else
3912 rcw += 2*disks;
3913 }
3914 }
1e6d690b 3915
39b99586
SL		 3916 pr_debug("for sector %llu state 0x%lx, rmw=%d rcw=%d\n",
3917 (unsigned long long)sh->sector, sh->state, rmw, rcw);
a4456856 3918 set_bit(STRIPE_HANDLE, &sh->state);
41257580 3919 if ((rmw < rcw || (rmw == rcw && conf->rmw_level == PARITY_PREFER_RMW)) && rmw > 0) {
a4456856 3920 /* prefer read-modify-write, but need to get some data */
e3620a3a
JB		 3921 if (conf->mddev->queue)
3922 blk_add_trace_msg(conf->mddev->queue,
3923 "raid5 rmw %llu %d",
3924 (unsigned long long)sh->sector, rmw);
a4456856
DW		 3925 for (i = disks; i--; ) {
3926 struct r5dev *dev = &sh->dev[i];
1e6d690b
SL		 3927 if (test_bit(R5_InJournal, &dev->flags) &&
3928 dev->page == dev->orig_page &&
3929 !test_bit(R5_LOCKED, &sh->dev[sh->pd_idx].flags)) {
3930 /* alloc page for prexor */
d7bd398e
SL		 3931 struct page *p = alloc_page(GFP_NOIO);
3932
3933 if (p) {
3934 dev->orig_page = p;
3935 continue;
3936 }
3937
3938 /*
3939 * alloc_page() failed, try use
3940 * disk_info->extra_page
3941 */
3942 if (!test_and_set_bit(R5C_EXTRA_PAGE_IN_USE,
3943 &conf->cache_state)) {
3944 r5c_use_extra_page(sh);
3945 break;
3946 }
1e6d690b 3947
d7bd398e
SL		 3948 /* extra_page in use, add to delayed_list */
3949 set_bit(STRIPE_DELAYED, &sh->state);
3950 s->waiting_extra_page = 1;
3951 return -EAGAIN;
1e6d690b 3952 }
d7bd398e 3953 }
1e6d690b 3954
d7bd398e
SL		 3955 for (i = disks; i--; ) {
3956 struct r5dev *dev = &sh->dev[i];
39b99586 3957 if (((dev->towrite && !delay_towrite(conf, dev, s)) ||
1e6d690b
SL		 3958 i == sh->pd_idx || i == sh->qd_idx ||
3959 test_bit(R5_InJournal, &dev->flags)) &&
a4456856 3960 !test_bit(R5_LOCKED, &dev->flags) &&
86aa1397 3961 !(uptodate_for_rmw(dev) ||
1e6d690b 3962 test_bit(R5_Wantcompute, &dev->flags)) &&
a4456856 3963 test_bit(R5_Insync, &dev->flags)) {
67f45548
N		 3964 if (test_bit(STRIPE_PREREAD_ACTIVE,
3965 &sh->state)) {
3966 pr_debug("Read_old block %d for r-m-w\n",
3967 i);
a4456856
DW		 3968 set_bit(R5_LOCKED, &dev->flags);
3969 set_bit(R5_Wantread, &dev->flags);
3970 s->locked++;
3971 } else {
3972 set_bit(STRIPE_DELAYED, &sh->state);
3973 set_bit(STRIPE_HANDLE, &sh->state);
3974 }
3975 }
3976 }
a9add5d9 3977 }
41257580 3978 if ((rcw < rmw || (rcw == rmw && conf->rmw_level != PARITY_PREFER_RMW)) && rcw > 0) {
a4456856 3979 /* want reconstruct write, but need to get some data */
a9add5d9 3980 int qread =0;
c8ac1803 3981 rcw = 0;
a4456856
DW		 3982 for (i = disks; i--; ) {
3983 struct r5dev *dev = &sh->dev[i];
3984 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
c8ac1803 3985 i != sh->pd_idx && i != sh->qd_idx &&
a4456856 3986 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219 3987 !(test_bit(R5_UPTODATE, &dev->flags) ||
c8ac1803
N		 3988 test_bit(R5_Wantcompute, &dev->flags))) {
3989 rcw++;
67f45548
N		 3990 if (test_bit(R5_Insync, &dev->flags) &&
3991 test_bit(STRIPE_PREREAD_ACTIVE,
3992 &sh->state)) {
45b4233c 3993 pr_debug("Read_old block "
a4456856
DW		 3994 "%d for Reconstruct\n", i);
3995 set_bit(R5_LOCKED, &dev->flags);
3996 set_bit(R5_Wantread, &dev->flags);
3997 s->locked++;
a9add5d9 3998 qread++;
a4456856
DW		 3999 } else {
4000 set_bit(STRIPE_DELAYED, &sh->state);
4001 set_bit(STRIPE_HANDLE, &sh->state);
4002 }
4003 }
4004 }
e3620a3a 4005 if (rcw && conf->mddev->queue)
a9add5d9
N		 4006 blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d",
4007 (unsigned long long)sh->sector,
4008 rcw, qread, test_bit(STRIPE_DELAYED, &sh->state));
c8ac1803 4009 }
b1b02fe9
N		 4010
4011 if (rcw > disks && rmw > disks &&
4012 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4013 set_bit(STRIPE_DELAYED, &sh->state);
4014
a4456856
DW		 4015 /* now if nothing is locked, and if we have enough data,
4016 * we can start a write request
4017 */
f38e1219
DW		 4018 /* since handle_stripe can be called at any time we need to handle the
4019 * case where a compute block operation has been submitted and then a
ac6b53b6
DW		 4020 * subsequent call wants to start a write request. raid_run_ops only
4021 * handles the case where compute block and reconstruct are requested
f38e1219
DW		 4022 * simultaneously. If this is not the case then new writes need to be
4023 * held off until the compute completes.
4024 */
976ea8d4
DW		 4025 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
4026 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
1e6d690b 4027 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
c0f7bddb 4028 schedule_reconstruction(sh, s, rcw == 0, 0);
d7bd398e 4029 return 0;
a4456856
DW		 4030}
4031
d1688a6d 4032static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh,
a4456856
DW		 4033 struct stripe_head_state *s, int disks)
4034{
ecc65c9b 4035 struct r5dev *dev = NULL;
bd2ab670 4036
59fc630b 4037 BUG_ON(sh->batch_head);
a4456856 4038 set_bit(STRIPE_HANDLE, &sh->state);
e89f8962 4039
ecc65c9b
DW		 4040 switch (sh->check_state) {
4041 case check_state_idle:
4042 /* start a new check operation if there are no failures */
bd2ab670 4043 if (s->failed == 0) {
bd2ab670 4044 BUG_ON(s->uptodate != disks);
ecc65c9b
DW		 4045 sh->check_state = check_state_run;
4046 set_bit(STRIPE_OP_CHECK, &s->ops_request);
bd2ab670 4047 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
bd2ab670 4048 s->uptodate--;
ecc65c9b 4049 break;
bd2ab670 4050 }
f2b3b44d 4051 dev = &sh->dev[s->failed_num[0]];
ecc65c9b
DW		 4052 /* fall through */
4053 case check_state_compute_result:
4054 sh->check_state = check_state_idle;
4055 if (!dev)
4056 dev = &sh->dev[sh->pd_idx];
4057
4058 /* check that a write has not made the stripe insync */
4059 if (test_bit(STRIPE_INSYNC, &sh->state))
4060 break;
c8894419 4061
a4456856 4062 /* either failed parity check, or recovery is happening */
a4456856
DW		 4063 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
4064 BUG_ON(s->uptodate != disks);
4065
4066 set_bit(R5_LOCKED, &dev->flags);
ecc65c9b 4067 s->locked++;
a4456856 4068 set_bit(R5_Wantwrite, &dev->flags);
830ea016 4069
a4456856 4070 clear_bit(STRIPE_DEGRADED, &sh->state);
a4456856 4071 set_bit(STRIPE_INSYNC, &sh->state);
ecc65c9b
DW		 4072 break;
4073 case check_state_run:
4074 break; /* we will be called again upon completion */
4075 case check_state_check_result:
4076 sh->check_state = check_state_idle;
4077
4078 /* if a failure occurred during the check operation, leave
4079 * STRIPE_INSYNC not set and let the stripe be handled again
4080 */
4081 if (s->failed)
4082 break;
4083
4084 /* handle a successful check operation, if parity is correct
4085 * we are done. Otherwise update the mismatch count and repair
4086 * parity if !MD_RECOVERY_CHECK
4087 */
ad283ea4 4088 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
ecc65c9b
DW		 4089 /* parity is correct (on disc,
4090 * not in buffer any more)
4091 */
4092 set_bit(STRIPE_INSYNC, &sh->state);
4093 else {
7f7583d4 4094 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
e1539036 4095 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) {
ecc65c9b
DW		 4096 /* don't try to repair!! */
4097 set_bit(STRIPE_INSYNC, &sh->state);
e1539036
N		 4098 pr_warn_ratelimited("%s: mismatch sector in range "
4099 "%llu-%llu\n", mdname(conf->mddev),
4100 (unsigned long long) sh->sector,
4101 (unsigned long long) sh->sector +
4102 STRIPE_SECTORS);
4103 } else {
ecc65c9b 4104 sh->check_state = check_state_compute_run;
976ea8d4 4105 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
ecc65c9b
DW		 4106 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
4107 set_bit(R5_Wantcompute,
4108 &sh->dev[sh->pd_idx].flags);
4109 sh->ops.target = sh->pd_idx;
ac6b53b6 4110 sh->ops.target2 = -1;
ecc65c9b
DW		 4111 s->uptodate++;
4112 }
4113 }
4114 break;
4115 case check_state_compute_run:
4116 break;
4117 default:
cc6167b4 4118 pr_err("%s: unknown check_state: %d sector: %llu\n",
ecc65c9b
DW		 4119 __func__, sh->check_state,
4120 (unsigned long long) sh->sector);
4121 BUG();
a4456856
DW		 4122 }
4123}
4124
d1688a6d 4125static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh,
36d1c647 4126 struct stripe_head_state *s,
f2b3b44d 4127 int disks)
a4456856 4128{
a4456856 4129 int pd_idx = sh->pd_idx;
34e04e87 4130 int qd_idx = sh->qd_idx;
d82dfee0 4131 struct r5dev *dev;
a4456856 4132
59fc630b 4133 BUG_ON(sh->batch_head);
a4456856
DW		 4134 set_bit(STRIPE_HANDLE, &sh->state);
4135
4136 BUG_ON(s->failed > 2);
d82dfee0 4137
a4456856
DW		 4138 /* Want to check and possibly repair P and Q.
4139 * However there could be one 'failed' device, in which
4140 * case we can only check one of them, possibly using the
4141 * other to generate missing data
4142 */
4143
d82dfee0
DW		 4144 switch (sh->check_state) {
4145 case check_state_idle:
4146 /* start a new check operation if there are < 2 failures */
f2b3b44d 4147 if (s->failed == s->q_failed) {
d82dfee0 4148 /* The only possible failed device holds Q, so it
a4456856
DW		 4149 * makes sense to check P (If anything else were failed,
4150 * we would have used P to recreate it).
4151 */
d82dfee0 4152 sh->check_state = check_state_run;
a4456856 4153 }
f2b3b44d 4154 if (!s->q_failed && s->failed < 2) {
d82dfee0 4155 /* Q is not failed, and we didn't use it to generate
a4456856
DW		 4156 * anything, so it makes sense to check it
4157 */
d82dfee0
DW		 4158 if (sh->check_state == check_state_run)
4159 sh->check_state = check_state_run_pq;
4160 else
4161 sh->check_state = check_state_run_q;
a4456856 4162 }
a4456856 4163
d82dfee0
DW		 4164 /* discard potentially stale zero_sum_result */
4165 sh->ops.zero_sum_result = 0;
a4456856 4166
d82dfee0
DW		 4167 if (sh->check_state == check_state_run) {
4168 /* async_xor_zero_sum destroys the contents of P */
4169 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
4170 s->uptodate--;
a4456856 4171 }
d82dfee0
DW		 4172 if (sh->check_state >= check_state_run &&
4173 sh->check_state <= check_state_run_pq) {
4174 /* async_syndrome_zero_sum preserves P and Q, so
4175 * no need to mark them !uptodate here
4176 */
4177 set_bit(STRIPE_OP_CHECK, &s->ops_request);
4178 break;
a4456856
DW		 4179 }
4180
d82dfee0
DW		 4181 /* we have 2-disk failure */
4182 BUG_ON(s->failed != 2);
4183 /* fall through */
4184 case check_state_compute_result:
4185 sh->check_state = check_state_idle;
a4456856 4186
d82dfee0
DW		 4187 /* check that a write has not made the stripe insync */
4188 if (test_bit(STRIPE_INSYNC, &sh->state))
4189 break;
a4456856
DW		 4190
4191 /* now write out any block on a failed drive,
d82dfee0 4192 * or P or Q if they were recomputed
a4456856 4193 */
b2176a1d 4194 dev = NULL;
a4456856 4195 if (s->failed == 2) {
f2b3b44d 4196 dev = &sh->dev[s->failed_num[1]];
a4456856
DW		 4197 s->locked++;
4198 set_bit(R5_LOCKED, &dev->flags);
4199 set_bit(R5_Wantwrite, &dev->flags);
4200 }
4201 if (s->failed >= 1) {
f2b3b44d 4202 dev = &sh->dev[s->failed_num[0]];
a4456856
DW		 4203 s->locked++;
4204 set_bit(R5_LOCKED, &dev->flags);
4205 set_bit(R5_Wantwrite, &dev->flags);
4206 }
d82dfee0 4207 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
a4456856
DW		 4208 dev = &sh->dev[pd_idx];
4209 s->locked++;
4210 set_bit(R5_LOCKED, &dev->flags);
4211 set_bit(R5_Wantwrite, &dev->flags);
4212 }
d82dfee0 4213 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
a4456856
DW		 4214 dev = &sh->dev[qd_idx];
4215 s->locked++;
4216 set_bit(R5_LOCKED, &dev->flags);
4217 set_bit(R5_Wantwrite, &dev->flags);
4218 }
b2176a1d
NC		 4219 if (WARN_ONCE(dev && !test_bit(R5_UPTODATE, &dev->flags),
4220 "%s: disk%td not up to date\n",
4221 mdname(conf->mddev),
4222 dev - (struct r5dev *) &sh->dev)) {
4223 clear_bit(R5_LOCKED, &dev->flags);
4224 clear_bit(R5_Wantwrite, &dev->flags);
4225 s->locked--;
4226 }
a4456856
DW		 4227 clear_bit(STRIPE_DEGRADED, &sh->state);
4228
4229 set_bit(STRIPE_INSYNC, &sh->state);
d82dfee0
DW		 4230 break;
4231 case check_state_run:
4232 case check_state_run_q:
4233 case check_state_run_pq:
4234 break; /* we will be called again upon completion */
4235 case check_state_check_result:
4236 sh->check_state = check_state_idle;
4237
4238 /* handle a successful check operation, if parity is correct
4239 * we are done. Otherwise update the mismatch count and repair
4240 * parity if !MD_RECOVERY_CHECK
4241 */
4242 if (sh->ops.zero_sum_result == 0) {
a25d8c32
SL		 4243 /* both parities are correct */
4244 if (!s->failed)
4245 set_bit(STRIPE_INSYNC, &sh->state);
4246 else {
4247 /* in contrast to the raid5 case we can validate
4248 * parity, but still have a failure to write
4249 * back
4250 */
4251 sh->check_state = check_state_compute_result;
4252 /* Returning at this point means that we may go
4253 * off and bring p and/or q uptodate again so
4254 * we make sure to check zero_sum_result again
4255 * to verify if p or q need writeback
4256 */
4257 }
d82dfee0 4258 } else {
7f7583d4 4259 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
e1539036 4260 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) {
d82dfee0
DW		 4261 /* don't try to repair!! */
4262 set_bit(STRIPE_INSYNC, &sh->state);
e1539036
N		 4263 pr_warn_ratelimited("%s: mismatch sector in range "
4264 "%llu-%llu\n", mdname(conf->mddev),
4265 (unsigned long long) sh->sector,
4266 (unsigned long long) sh->sector +
4267 STRIPE_SECTORS);
4268 } else {
d82dfee0
DW		 4269 int *target = &sh->ops.target;
4270
4271 sh->ops.target = -1;
4272 sh->ops.target2 = -1;
4273 sh->check_state = check_state_compute_run;
4274 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
4275 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
4276 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
4277 set_bit(R5_Wantcompute,
4278 &sh->dev[pd_idx].flags);
4279 *target = pd_idx;
4280 target = &sh->ops.target2;
4281 s->uptodate++;
4282 }
4283 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
4284 set_bit(R5_Wantcompute,
4285 &sh->dev[qd_idx].flags);
4286 *target = qd_idx;
4287 s->uptodate++;
4288 }
4289 }
4290 }
4291 break;
4292 case check_state_compute_run:
4293 break;
4294 default:
cc6167b4
N		 4295 pr_warn("%s: unknown check_state: %d sector: %llu\n",
4296 __func__, sh->check_state,
4297 (unsigned long long) sh->sector);
d82dfee0 4298 BUG();
a4456856
DW		 4299 }
4300}
4301
d1688a6d 4302static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh)
a4456856
DW		 4303{
4304 int i;
4305
4306 /* We have read all the blocks in this stripe and now we need to
4307 * copy some of them into a target stripe for expand.
4308 */
f0a50d37 4309 struct dma_async_tx_descriptor *tx = NULL;
59fc630b 4310 BUG_ON(sh->batch_head);
a4456856
DW		 4311 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
4312 for (i = 0; i < sh->disks; i++)
34e04e87 4313 if (i != sh->pd_idx && i != sh->qd_idx) {
911d4ee8 4314 int dd_idx, j;
a4456856 4315 struct stripe_head *sh2;
a08abd8c 4316 struct async_submit_ctl submit;
a4456856 4317
6d036f7d 4318 sector_t bn = raid5_compute_blocknr(sh, i, 1);
911d4ee8
N		 4319 sector_t s = raid5_compute_sector(conf, bn, 0,
4320 &dd_idx, NULL);
6d036f7d 4321 sh2 = raid5_get_active_stripe(conf, s, 0, 1, 1);
a4456856
DW		 4322 if (sh2 == NULL)
4323 /* so far only the early blocks of this stripe
4324 * have been requested. When later blocks
4325 * get requested, we will try again
4326 */
4327 continue;
4328 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
4329 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
4330 /* must have already done this block */
6d036f7d 4331 raid5_release_stripe(sh2);
a4456856
DW		 4332 continue;
4333 }
f0a50d37
DW		 4334
4335 /* place all the copies on one channel */
a08abd8c 4336 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
f0a50d37 4337 tx = async_memcpy(sh2->dev[dd_idx].page,
88ba2aa5 4338 sh->dev[i].page, 0, 0, STRIPE_SIZE,
a08abd8c 4339 &submit);
f0a50d37 4340
a4456856
DW		 4341 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
4342 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
4343 for (j = 0; j < conf->raid_disks; j++)
4344 if (j != sh2->pd_idx &&
86c374ba 4345 j != sh2->qd_idx &&
a4456856
DW		 4346 !test_bit(R5_Expanded, &sh2->dev[j].flags))
4347 break;
4348 if (j == conf->raid_disks) {
4349 set_bit(STRIPE_EXPAND_READY, &sh2->state);
4350 set_bit(STRIPE_HANDLE, &sh2->state);
4351 }
6d036f7d 4352 raid5_release_stripe(sh2);
f0a50d37 4353
a4456856 4354 }
a2e08551 4355 /* done submitting copies, wait for them to complete */
749586b7 4356 async_tx_quiesce(&tx);
a4456856 4357}
1da177e4
LT		 4358
4359/*
4360 * handle_stripe - do things to a stripe.
4361 *
9a3e1101
N		 4362 * We lock the stripe by setting STRIPE_ACTIVE and then examine the
4363 * state of various bits to see what needs to be done.
1da177e4 4364 * Possible results:
9a3e1101
N		 4365 * return some read requests which now have data
4366 * return some write requests which are safely on storage
1da177e4
LT		 4367 * schedule a read on some buffers
4368 * schedule a write of some buffers
4369 * return confirmation of parity correctness
4370 *
1da177e4 4371 */
a4456856 4372
acfe726b 4373static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s)
1da177e4 4374{
d1688a6d 4375 struct r5conf *conf = sh->raid_conf;
f416885e 4376 int disks = sh->disks;
474af965
N		 4377 struct r5dev *dev;
4378 int i;
9a3e1101 4379 int do_recovery = 0;
1da177e4 4380
acfe726b
N		 4381 memset(s, 0, sizeof(*s));
4382
dabc4ec6 4383 s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state) && !sh->batch_head;
4384 s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state) && !sh->batch_head;
acfe726b
N		 4385 s->failed_num[0] = -1;
4386 s->failed_num[1] = -1;
6e74a9cf 4387 s->log_failed = r5l_log_disk_error(conf);
1da177e4 4388
acfe726b 4389 /* Now to look around and see what can be done */
1da177e4 4390 rcu_read_lock();
16a53ecc 4391 for (i=disks; i--; ) {
3cb03002 4392 struct md_rdev *rdev;
31c176ec
N		 4393 sector_t first_bad;
4394 int bad_sectors;
4395 int is_bad = 0;
acfe726b 4396
16a53ecc 4397 dev = &sh->dev[i];
1da177e4 4398
45b4233c 4399 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
9a3e1101
N		 4400 i, dev->flags,
4401 dev->toread, dev->towrite, dev->written);
6c0069c0
YT		 4402 /* maybe we can reply to a read
4403 *
4404 * new wantfill requests are only permitted while
4405 * ops_complete_biofill is guaranteed to be inactive
4406 */
4407 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
4408 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
4409 set_bit(R5_Wantfill, &dev->flags);
1da177e4 4410
16a53ecc 4411 /* now count some things */
cc94015a
N		 4412 if (test_bit(R5_LOCKED, &dev->flags))
4413 s->locked++;
4414 if (test_bit(R5_UPTODATE, &dev->flags))
4415 s->uptodate++;
2d6e4ecc 4416 if (test_bit(R5_Wantcompute, &dev->flags)) {
cc94015a
N		 4417 s->compute++;
4418 BUG_ON(s->compute > 2);
2d6e4ecc 4419 }
1da177e4 4420
acfe726b 4421 if (test_bit(R5_Wantfill, &dev->flags))
cc94015a 4422 s->to_fill++;
acfe726b 4423 else if (dev->toread)
cc94015a 4424 s->to_read++;
16a53ecc 4425 if (dev->towrite) {
cc94015a 4426 s->to_write++;
16a53ecc 4427 if (!test_bit(R5_OVERWRITE, &dev->flags))
cc94015a 4428 s->non_overwrite++;
16a53ecc 4429 }
a4456856 4430 if (dev->written)
cc94015a 4431 s->written++;
14a75d3e
N		 4432 /* Prefer to use the replacement for reads, but only
4433 * if it is recovered enough and has no bad blocks.
4434 */
4435 rdev = rcu_dereference(conf->disks[i].replacement);
4436 if (rdev && !test_bit(Faulty, &rdev->flags) &&
4437 rdev->recovery_offset >= sh->sector + STRIPE_SECTORS &&
4438 !is_badblock(rdev, sh->sector, STRIPE_SECTORS,
4439 &first_bad, &bad_sectors))
4440 set_bit(R5_ReadRepl, &dev->flags);
4441 else {
e6030cb0 4442 if (rdev && !test_bit(Faulty, &rdev->flags))
9a3e1101 4443 set_bit(R5_NeedReplace, &dev->flags);
e6030cb0
N		 4444 else
4445 clear_bit(R5_NeedReplace, &dev->flags);
14a75d3e
N		 4446 rdev = rcu_dereference(conf->disks[i].rdev);
4447 clear_bit(R5_ReadRepl, &dev->flags);
4448 }
9283d8c5
N		 4449 if (rdev && test_bit(Faulty, &rdev->flags))
4450 rdev = NULL;
31c176ec
N		 4451 if (rdev) {
4452 is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
4453 &first_bad, &bad_sectors);
4454 if (s->blocked_rdev == NULL
4455 && (test_bit(Blocked, &rdev->flags)
4456 || is_bad < 0)) {
4457 if (is_bad < 0)
4458 set_bit(BlockedBadBlocks,
4459 &rdev->flags);
4460 s->blocked_rdev = rdev;
4461 atomic_inc(&rdev->nr_pending);
4462 }
6bfe0b49 4463 }
415e72d0
N		 4464 clear_bit(R5_Insync, &dev->flags);
4465 if (!rdev)
4466 /* Not in-sync */;
31c176ec
N		 4467 else if (is_bad) {
4468 /* also not in-sync */
18b9837e
N		 4469 if (!test_bit(WriteErrorSeen, &rdev->flags) &&
4470 test_bit(R5_UPTODATE, &dev->flags)) {
31c176ec
N		 4471 /* treat as in-sync, but with a read error
4472 * which we can now try to correct
4473 */
4474 set_bit(R5_Insync, &dev->flags);
4475 set_bit(R5_ReadError, &dev->flags);
4476 }
4477 } else if (test_bit(In_sync, &rdev->flags))
415e72d0 4478 set_bit(R5_Insync, &dev->flags);
30d7a483 4479 else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
415e72d0 4480 /* in sync if before recovery_offset */
30d7a483
N		 4481 set_bit(R5_Insync, &dev->flags);
4482 else if (test_bit(R5_UPTODATE, &dev->flags) &&
4483 test_bit(R5_Expanded, &dev->flags))
4484 /* If we've reshaped into here, we assume it is Insync.
4485 * We will shortly update recovery_offset to make
4486 * it official.
4487 */
4488 set_bit(R5_Insync, &dev->flags);
4489
1cc03eb9 4490 if (test_bit(R5_WriteError, &dev->flags)) {
14a75d3e
N		 4491 /* This flag does not apply to '.replacement'
4492 * only to .rdev, so make sure to check that*/
4493 struct md_rdev *rdev2 = rcu_dereference(
4494 conf->disks[i].rdev);
4495 if (rdev2 == rdev)
4496 clear_bit(R5_Insync, &dev->flags);
4497 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
bc2607f3 4498 s->handle_bad_blocks = 1;
14a75d3e 4499 atomic_inc(&rdev2->nr_pending);
bc2607f3
N		 4500 } else
4501 clear_bit(R5_WriteError, &dev->flags);
4502 }
1cc03eb9 4503 if (test_bit(R5_MadeGood, &dev->flags)) {
14a75d3e
N		 4504 /* This flag does not apply to '.replacement'
4505 * only to .rdev, so make sure to check that*/
4506 struct md_rdev *rdev2 = rcu_dereference(
4507 conf->disks[i].rdev);
4508 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
b84db560 4509 s->handle_bad_blocks = 1;
14a75d3e 4510 atomic_inc(&rdev2->nr_pending);
b84db560
N		 4511 } else
4512 clear_bit(R5_MadeGood, &dev->flags);
4513 }
977df362
N		 4514 if (test_bit(R5_MadeGoodRepl, &dev->flags)) {
4515 struct md_rdev *rdev2 = rcu_dereference(
4516 conf->disks[i].replacement);
4517 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
4518 s->handle_bad_blocks = 1;
4519 atomic_inc(&rdev2->nr_pending);
4520 } else
4521 clear_bit(R5_MadeGoodRepl, &dev->flags);
4522 }
415e72d0 4523 if (!test_bit(R5_Insync, &dev->flags)) {
16a53ecc
N		 4524 /* The ReadError flag will just be confusing now */
4525 clear_bit(R5_ReadError, &dev->flags);
4526 clear_bit(R5_ReWrite, &dev->flags);
1da177e4 4527 }
415e72d0
N		 4528 if (test_bit(R5_ReadError, &dev->flags))
4529 clear_bit(R5_Insync, &dev->flags);
4530 if (!test_bit(R5_Insync, &dev->flags)) {
cc94015a
N		 4531 if (s->failed < 2)
4532 s->failed_num[s->failed] = i;
4533 s->failed++;
9a3e1101
N		 4534 if (rdev && !test_bit(Faulty, &rdev->flags))
4535 do_recovery = 1;
d63e2fc8
BC		 4536 else if (!rdev) {
4537 rdev = rcu_dereference(
4538 conf->disks[i].replacement);
4539 if (rdev && !test_bit(Faulty, &rdev->flags))
4540 do_recovery = 1;
4541 }
415e72d0 4542 }
2ded3703
SL		 4543
4544 if (test_bit(R5_InJournal, &dev->flags))
4545 s->injournal++;
1e6d690b
SL		 4546 if (test_bit(R5_InJournal, &dev->flags) && dev->written)
4547 s->just_cached++;
1da177e4 4548 }
9a3e1101
N		 4549 if (test_bit(STRIPE_SYNCING, &sh->state)) {
4550 /* If there is a failed device being replaced,
4551 * we must be recovering.
4552 * else if we are after recovery_cp, we must be syncing
c6d2e084 4553 * else if MD_RECOVERY_REQUESTED is set, we also are syncing.
9a3e1101
N		 4554 * else we can only be replacing
4555 * sync and recovery both need to read all devices, and so
4556 * use the same flag.
4557 */
4558 if (do_recovery ||
c6d2e084 4559 sh->sector >= conf->mddev->recovery_cp ||
4560 test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery)))
9a3e1101
N		 4561 s->syncing = 1;
4562 else
4563 s->replacing = 1;
4564 }
1da177e4 4565 rcu_read_unlock();
cc94015a
N		 4566}
4567
59fc630b 4568static int clear_batch_ready(struct stripe_head *sh)
4569{
b15a9dbd
N		 4570 /* Return '1' if this is a member of batch, or
4571 * '0' if it is a lone stripe or a head which can now be
4572 * handled.
4573 */
59fc630b 4574 struct stripe_head *tmp;
4575 if (!test_and_clear_bit(STRIPE_BATCH_READY, &sh->state))
b15a9dbd 4576 return (sh->batch_head && sh->batch_head != sh);
59fc630b 4577 spin_lock(&sh->stripe_lock);
4578 if (!sh->batch_head) {
4579 spin_unlock(&sh->stripe_lock);
4580 return 0;
4581 }
4582
4583 /*
4584 * this stripe could be added to a batch list before we check
4585 * BATCH_READY, skips it
4586 */
4587 if (sh->batch_head != sh) {
4588 spin_unlock(&sh->stripe_lock);
4589 return 1;
4590 }
4591 spin_lock(&sh->batch_lock);
4592 list_for_each_entry(tmp, &sh->batch_list, batch_list)
4593 clear_bit(STRIPE_BATCH_READY, &tmp->state);
4594 spin_unlock(&sh->batch_lock);
4595 spin_unlock(&sh->stripe_lock);
4596
4597 /*
4598 * BATCH_READY is cleared, no new stripes can be added.
4599 * batch_list can be accessed without lock
4600 */
4601 return 0;
4602}
4603
3960ce79
N		 4604static void break_stripe_batch_list(struct stripe_head *head_sh,
4605 unsigned long handle_flags)
72ac7330 4606{
4e3d62ff 4607 struct stripe_head *sh, *next;
72ac7330 4608 int i;
fb642b92 4609 int do_wakeup = 0;
72ac7330 4610
bb27051f
N		 4611 list_for_each_entry_safe(sh, next, &head_sh->batch_list, batch_list) {
4612
72ac7330 4613 list_del_init(&sh->batch_list);
4614
fb3229d5 4615 WARN_ONCE(sh->state & ((1 << STRIPE_ACTIVE) |
1b956f7a
N		 4616 (1 << STRIPE_SYNCING) |
4617 (1 << STRIPE_REPLACED) |
1b956f7a
N		 4618 (1 << STRIPE_DELAYED) |
4619 (1 << STRIPE_BIT_DELAY) |
4620 (1 << STRIPE_FULL_WRITE) |
4621 (1 << STRIPE_BIOFILL_RUN) |
4622 (1 << STRIPE_COMPUTE_RUN) |
4623 (1 << STRIPE_OPS_REQ_PENDING) |
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
KO		 5501 logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1);
5502 last_sector = bi->bi_iter.bi_sector + (bi->bi_iter.bi_size>>9);
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
SL		 5594 if (ret == -ENODEV) {
5595 md_flush_request(mddev, bi);
cc27b0c7 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
6ed3003c
N		 5729 set_bit(STRIPE_HANDLE, &sh->state);
5730 clear_bit(STRIPE_DELAYED, &sh->state);
59fc630b 5731 if ((!sh->batch_head || sh == sh->batch_head) &&
1eff9d32 5732 (bi->bi_opf & REQ_SYNC) &&
729a1866
N		 5733 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
5734 atomic_inc(&conf->preread_active_stripes);
8811b596 5735 release_stripe_plug(mddev, sh);
1da177e4
LT		 5736 } else {
5737 /* cannot get stripe for read-ahead, just give-up */
4e4cbee9 5738 bi->bi_status = BLK_STS_IOERR;
1da177e4
LT		 5739 break;
5740 }
1da177e4 5741 }
27c0f68f 5742 finish_wait(&conf->wait_for_overlap, &w);
7c13edc8 5743
49728050
N		 5744 if (rw == WRITE)
5745 md_write_end(mddev);
016c76ac 5746 bio_endio(bi);
cc27b0c7 5747 return true;
1da177e4
LT		 5748}
5749
fd01b88c 5750static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks);
b522adcd 5751
fd01b88c 5752static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped)
1da177e4 5753{
52c03291
N		 5754 /* reshaping is quite different to recovery/resync so it is
5755 * handled quite separately ... here.
5756 *
5757 * On each call to sync_request, we gather one chunk worth of
5758 * destination stripes and flag them as expanding.
5759 * Then we find all the source stripes and request reads.
5760 * As the reads complete, handle_stripe will copy the data
5761 * into the destination stripe and release that stripe.
5762 */
d1688a6d 5763 struct r5conf *conf = mddev->private;
1da177e4 5764 struct stripe_head *sh;
db0505d3 5765 struct md_rdev *rdev;
ccfcc3c1 5766 sector_t first_sector, last_sector;
f416885e
N		 5767 int raid_disks = conf->previous_raid_disks;
5768 int data_disks = raid_disks - conf->max_degraded;
5769 int new_data_disks = conf->raid_disks - conf->max_degraded;
52c03291
N		 5770 int i;
5771 int dd_idx;
c8f517c4 5772 sector_t writepos, readpos, safepos;
ec32a2bd 5773 sector_t stripe_addr;
7a661381 5774 int reshape_sectors;
ab69ae12 5775 struct list_head stripes;
92140480 5776 sector_t retn;
52c03291 5777
fef9c61f
N		 5778 if (sector_nr == 0) {
5779 /* If restarting in the middle, skip the initial sectors */
2c810cdd 5780 if (mddev->reshape_backwards &&
fef9c61f
N		 5781 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
5782 sector_nr = raid5_size(mddev, 0, 0)
5783 - conf->reshape_progress;
6cbd8148
N		 5784 } else if (mddev->reshape_backwards &&
5785 conf->reshape_progress == MaxSector) {
5786 /* shouldn't happen, but just in case, finish up.*/
5787 sector_nr = MaxSector;
2c810cdd 5788 } else if (!mddev->reshape_backwards &&
fef9c61f
N		 5789 conf->reshape_progress > 0)
5790 sector_nr = conf->reshape_progress;
f416885e 5791 sector_div(sector_nr, new_data_disks);
fef9c61f 5792 if (sector_nr) {
8dee7211
N		 5793 mddev->curr_resync_completed = sector_nr;
5794 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
fef9c61f 5795 *skipped = 1;
92140480
N		 5796 retn = sector_nr;
5797 goto finish;
fef9c61f 5798 }
52c03291
N		 5799 }
5800
7a661381
N		 5801 /* We need to process a full chunk at a time.
5802 * If old and new chunk sizes differ, we need to process the
5803 * largest of these
5804 */
3cb5edf4
N		 5805
5806 reshape_sectors = max(conf->chunk_sectors, conf->prev_chunk_sectors);
7a661381 5807
b5254dd5
N		 5808 /* We update the metadata at least every 10 seconds, or when
5809 * the data about to be copied would over-write the source of
5810 * the data at the front of the range. i.e. one new_stripe
5811 * along from reshape_progress new_maps to after where
5812 * reshape_safe old_maps to
52c03291 5813 */
fef9c61f 5814 writepos = conf->reshape_progress;
f416885e 5815 sector_div(writepos, new_data_disks);
c8f517c4
N		 5816 readpos = conf->reshape_progress;
5817 sector_div(readpos, data_disks);
fef9c61f 5818 safepos = conf->reshape_safe;
f416885e 5819 sector_div(safepos, data_disks);
2c810cdd 5820 if (mddev->reshape_backwards) {
c74c0d76
N		 5821 BUG_ON(writepos < reshape_sectors);
5822 writepos -= reshape_sectors;
c8f517c4 5823 readpos += reshape_sectors;
7a661381 5824 safepos += reshape_sectors;
fef9c61f 5825 } else {
7a661381 5826 writepos += reshape_sectors;
c74c0d76
N		 5827 /* readpos and safepos are worst-case calculations.
5828 * A negative number is overly pessimistic, and causes
5829 * obvious problems for unsigned storage. So clip to 0.
5830 */
ed37d83e
N		 5831 readpos -= min_t(sector_t, reshape_sectors, readpos);
5832 safepos -= min_t(sector_t, reshape_sectors, safepos);
fef9c61f 5833 }
52c03291 5834
b5254dd5
N		 5835 /* Having calculated the 'writepos' possibly use it
5836 * to set 'stripe_addr' which is where we will write to.
5837 */
5838 if (mddev->reshape_backwards) {
5839 BUG_ON(conf->reshape_progress == 0);
5840 stripe_addr = writepos;
5841 BUG_ON((mddev->dev_sectors &
5842 ~((sector_t)reshape_sectors - 1))
5843 - reshape_sectors - stripe_addr
5844 != sector_nr);
5845 } else {
5846 BUG_ON(writepos != sector_nr + reshape_sectors);
5847 stripe_addr = sector_nr;
5848 }
5849
c8f517c4
N		 5850 /* 'writepos' is the most advanced device address we might write.
5851 * 'readpos' is the least advanced device address we might read.
5852 * 'safepos' is the least address recorded in the metadata as having
5853 * been reshaped.
b5254dd5
N		 5854 * If there is a min_offset_diff, these are adjusted either by
5855 * increasing the safepos/readpos if diff is negative, or
5856 * increasing writepos if diff is positive.
5857 * If 'readpos' is then behind 'writepos', there is no way that we can
c8f517c4
N		 5858 * ensure safety in the face of a crash - that must be done by userspace
5859 * making a backup of the data. So in that case there is no particular
5860 * rush to update metadata.
5861 * Otherwise if 'safepos' is behind 'writepos', then we really need to
5862 * update the metadata to advance 'safepos' to match 'readpos' so that
5863 * we can be safe in the event of a crash.
5864 * So we insist on updating metadata if safepos is behind writepos and
5865 * readpos is beyond writepos.
5866 * In any case, update the metadata every 10 seconds.
5867 * Maybe that number should be configurable, but I'm not sure it is
5868 * worth it.... maybe it could be a multiple of safemode_delay???
5869 */
b5254dd5
N		 5870 if (conf->min_offset_diff < 0) {
5871 safepos += -conf->min_offset_diff;
5872 readpos += -conf->min_offset_diff;
5873 } else
5874 writepos += conf->min_offset_diff;
5875
2c810cdd 5876 if ((mddev->reshape_backwards
c8f517c4
N		 5877 ? (safepos > writepos && readpos < writepos)
5878 : (safepos < writepos && readpos > writepos)) ||
5879 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
52c03291
N		 5880 /* Cannot proceed until we've updated the superblock... */
5881 wait_event(conf->wait_for_overlap,
c91abf5a
N		 5882 atomic_read(&conf->reshape_stripes)==0
5883 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5884 if (atomic_read(&conf->reshape_stripes) != 0)
5885 return 0;
fef9c61f 5886 mddev->reshape_position = conf->reshape_progress;
75d3da43 5887 mddev->curr_resync_completed = sector_nr;
db0505d3
N		 5888 if (!mddev->reshape_backwards)
5889 /* Can update recovery_offset */
5890 rdev_for_each(rdev, mddev)
5891 if (rdev->raid_disk >= 0 &&
5892 !test_bit(Journal, &rdev->flags) &&
5893 !test_bit(In_sync, &rdev->flags) &&
5894 rdev->recovery_offset < sector_nr)
5895 rdev->recovery_offset = sector_nr;
5896
c8f517c4 5897 conf->reshape_checkpoint = jiffies;
2953079c 5898 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
52c03291 5899 md_wakeup_thread(mddev->thread);
2953079c 5900 wait_event(mddev->sb_wait, mddev->sb_flags == 0 ||
c91abf5a
N		 5901 test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5902 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5903 return 0;
52c03291 5904 spin_lock_irq(&conf->device_lock);
fef9c61f 5905 conf->reshape_safe = mddev->reshape_position;
52c03291
N		 5906 spin_unlock_irq(&conf->device_lock);
5907 wake_up(&conf->wait_for_overlap);
acb180b0 5908 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
52c03291
N		 5909 }
5910
ab69ae12 5911 INIT_LIST_HEAD(&stripes);
7a661381 5912 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
52c03291 5913 int j;
a9f326eb 5914 int skipped_disk = 0;
6d036f7d 5915 sh = raid5_get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
52c03291
N		 5916 set_bit(STRIPE_EXPANDING, &sh->state);
5917 atomic_inc(&conf->reshape_stripes);
5918 /* If any of this stripe is beyond the end of the old
5919 * array, then we need to zero those blocks
5920 */
5921 for (j=sh->disks; j--;) {
5922 sector_t s;
5923 if (j == sh->pd_idx)
5924 continue;
f416885e 5925 if (conf->level == 6 &&
d0dabf7e 5926 j == sh->qd_idx)
f416885e 5927 continue;
6d036f7d 5928 s = raid5_compute_blocknr(sh, j, 0);
b522adcd 5929 if (s < raid5_size(mddev, 0, 0)) {
a9f326eb 5930 skipped_disk = 1;
52c03291
N		 5931 continue;
5932 }
5933 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
5934 set_bit(R5_Expanded, &sh->dev[j].flags);
5935 set_bit(R5_UPTODATE, &sh->dev[j].flags);
5936 }
a9f326eb 5937 if (!skipped_disk) {
52c03291
N		 5938 set_bit(STRIPE_EXPAND_READY, &sh->state);
5939 set_bit(STRIPE_HANDLE, &sh->state);
5940 }
ab69ae12 5941 list_add(&sh->lru, &stripes);
52c03291
N		 5942 }
5943 spin_lock_irq(&conf->device_lock);
2c810cdd 5944 if (mddev->reshape_backwards)
7a661381 5945 conf->reshape_progress -= reshape_sectors * new_data_disks;
fef9c61f 5946 else
7a661381 5947 conf->reshape_progress += reshape_sectors * new_data_disks;
52c03291
N		 5948 spin_unlock_irq(&conf->device_lock);
5949 /* Ok, those stripe are ready. We can start scheduling
5950 * reads on the source stripes.
5951 * The source stripes are determined by mapping the first and last
5952 * block on the destination stripes.
5953 */
52c03291 5954 first_sector =
ec32a2bd 5955 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
911d4ee8 5956 1, &dd_idx, NULL);
52c03291 5957 last_sector =
0e6e0271 5958 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
09c9e5fa 5959 * new_data_disks - 1),
911d4ee8 5960 1, &dd_idx, NULL);
58c0fed4
AN		 5961 if (last_sector >= mddev->dev_sectors)
5962 last_sector = mddev->dev_sectors - 1;
52c03291 5963 while (first_sector <= last_sector) {
6d036f7d 5964 sh = raid5_get_active_stripe(conf, first_sector, 1, 0, 1);
52c03291
N		 5965 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
5966 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 5967 raid5_release_stripe(sh);
52c03291
N		 5968 first_sector += STRIPE_SECTORS;
5969 }
ab69ae12
N		 5970 /* Now that the sources are clearly marked, we can release
5971 * the destination stripes
5972 */
5973 while (!list_empty(&stripes)) {
5974 sh = list_entry(stripes.next, struct stripe_head, lru);
5975 list_del_init(&sh->lru);
6d036f7d 5976 raid5_release_stripe(sh);
ab69ae12 5977 }
c6207277
N		 5978 /* If this takes us to the resync_max point where we have to pause,
5979 * then we need to write out the superblock.
5980 */
7a661381 5981 sector_nr += reshape_sectors;
92140480
N		 5982 retn = reshape_sectors;
5983finish:
c5e19d90
N		 5984 if (mddev->curr_resync_completed > mddev->resync_max ||
5985 (sector_nr - mddev->curr_resync_completed) * 2
c03f6a19 5986 >= mddev->resync_max - mddev->curr_resync_completed) {
c6207277
N		 5987 /* Cannot proceed until we've updated the superblock... */
5988 wait_event(conf->wait_for_overlap,
c91abf5a
N		 5989 atomic_read(&conf->reshape_stripes) == 0
5990 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5991 if (atomic_read(&conf->reshape_stripes) != 0)
5992 goto ret;
fef9c61f 5993 mddev->reshape_position = conf->reshape_progress;
75d3da43 5994 mddev->curr_resync_completed = sector_nr;
db0505d3
N		 5995 if (!mddev->reshape_backwards)
5996 /* Can update recovery_offset */
5997 rdev_for_each(rdev, mddev)
5998 if (rdev->raid_disk >= 0 &&
5999 !test_bit(Journal, &rdev->flags) &&
6000 !test_bit(In_sync, &rdev->flags) &&
6001 rdev->recovery_offset < sector_nr)
6002 rdev->recovery_offset = sector_nr;
c8f517c4 6003 conf->reshape_checkpoint = jiffies;
2953079c 6004 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
c6207277
N		 6005 md_wakeup_thread(mddev->thread);
6006 wait_event(mddev->sb_wait,
2953079c 6007 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)
c91abf5a
N		 6008 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
6009 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6010 goto ret;
c6207277 6011 spin_lock_irq(&conf->device_lock);
fef9c61f 6012 conf->reshape_safe = mddev->reshape_position;
c6207277
N		 6013 spin_unlock_irq(&conf->device_lock);
6014 wake_up(&conf->wait_for_overlap);
acb180b0 6015 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
c6207277 6016 }
c91abf5a 6017ret:
92140480 6018 return retn;
52c03291
N		 6019}
6020
849674e4
SL		 6021static inline sector_t raid5_sync_request(struct mddev *mddev, sector_t sector_nr,
6022 int *skipped)
52c03291 6023{
d1688a6d 6024 struct r5conf *conf = mddev->private;
52c03291 6025 struct stripe_head *sh;
58c0fed4 6026 sector_t max_sector = mddev->dev_sectors;
57dab0bd 6027 sector_t sync_blocks;
16a53ecc
N		 6028 int still_degraded = 0;
6029 int i;
1da177e4 6030
72626685 6031 if (sector_nr >= max_sector) {
1da177e4 6032 /* just being told to finish up .. nothing much to do */
cea9c228 6033
29269553
N		 6034 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
6035 end_reshape(conf);
6036 return 0;
6037 }
72626685
N		 6038
6039 if (mddev->curr_resync < max_sector) /* aborted */
e64e4018
AS		 6040 md_bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
6041 &sync_blocks, 1);
16a53ecc 6042 else /* completed sync */
72626685 6043 conf->fullsync = 0;
e64e4018 6044 md_bitmap_close_sync(mddev->bitmap);
72626685 6045
1da177e4
LT		 6046 return 0;
6047 }
ccfcc3c1 6048
64bd660b
N		 6049 /* Allow raid5_quiesce to complete */
6050 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
6051
52c03291
N		 6052 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6053 return reshape_request(mddev, sector_nr, skipped);
f6705578 6054
c6207277
N		 6055 /* No need to check resync_max as we never do more than one
6056 * stripe, and as resync_max will always be on a chunk boundary,
6057 * if the check in md_do_sync didn't fire, there is no chance
6058 * of overstepping resync_max here
6059 */
6060
16a53ecc 6061 /* if there is too many failed drives and we are trying
1da177e4
LT		 6062 * to resync, then assert that we are finished, because there is
6063 * nothing we can do.
6064 */
3285edf1 6065 if (mddev->degraded >= conf->max_degraded &&
16a53ecc 6066 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
58c0fed4 6067 sector_t rv = mddev->dev_sectors - sector_nr;
57afd89f 6068 *skipped = 1;
1da177e4
LT		 6069 return rv;
6070 }
6f608040 6071 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
6072 !conf->fullsync &&
e64e4018 6073 !md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
6f608040 6074 sync_blocks >= STRIPE_SECTORS) {
72626685
N		 6075 /* we can skip this block, and probably more */
6076 sync_blocks /= STRIPE_SECTORS;
6077 *skipped = 1;
6078 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
6079 }
1da177e4 6080
e64e4018 6081 md_bitmap_cond_end_sync(mddev->bitmap, sector_nr, false);
b47490c9 6082
6d036f7d 6083 sh = raid5_get_active_stripe(conf, sector_nr, 0, 1, 0);
1da177e4 6084 if (sh == NULL) {
6d036f7d 6085 sh = raid5_get_active_stripe(conf, sector_nr, 0, 0, 0);
1da177e4 6086 /* make sure we don't swamp the stripe cache if someone else
16a53ecc 6087 * is trying to get access
1da177e4 6088 */
66c006a5 6089 schedule_timeout_uninterruptible(1);
1da177e4 6090 }
16a53ecc 6091 /* Need to check if array will still be degraded after recovery/resync
16d9cfab
EM		 6092 * Note in case of > 1 drive failures it's possible we're rebuilding
6093 * one drive while leaving another faulty drive in array.
16a53ecc 6094 */
16d9cfab
EM		 6095 rcu_read_lock();
6096 for (i = 0; i < conf->raid_disks; i++) {
6aa7de05 6097 struct md_rdev *rdev = READ_ONCE(conf->disks[i].rdev);
16d9cfab
EM		 6098
6099 if (rdev == NULL || test_bit(Faulty, &rdev->flags))
16a53ecc 6100 still_degraded = 1;
16d9cfab
EM		 6101 }
6102 rcu_read_unlock();
16a53ecc 6103
e64e4018 6104 md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
16a53ecc 6105
83206d66 6106 set_bit(STRIPE_SYNC_REQUESTED, &sh->state);
053f5b65 6107 set_bit(STRIPE_HANDLE, &sh->state);
1da177e4 6108
6d036f7d 6109 raid5_release_stripe(sh);
1da177e4
LT		 6110
6111 return STRIPE_SECTORS;
6112}
6113
0472a42b
N		 6114static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio,
6115 unsigned int offset)
46031f9a
RBJ		 6116{
6117 /* We may not be able to submit a whole bio at once as there
6118 * may not be enough stripe_heads available.
6119 * We cannot pre-allocate enough stripe_heads as we may need
6120 * more than exist in the cache (if we allow ever large chunks).
6121 * So we do one stripe head at a time and record in
6122 * ->bi_hw_segments how many have been done.
6123 *
6124 * We *know* that this entire raid_bio is in one chunk, so
6125 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
6126 */
6127 struct stripe_head *sh;
911d4ee8 6128 int dd_idx;
46031f9a
RBJ		 6129 sector_t sector, logical_sector, last_sector;
6130 int scnt = 0;
46031f9a
RBJ		 6131 int handled = 0;
6132
4f024f37
KO		 6133 logical_sector = raid_bio->bi_iter.bi_sector &
6134 ~((sector_t)STRIPE_SECTORS-1);
112bf897 6135 sector = raid5_compute_sector(conf, logical_sector,
911d4ee8 6136 0, &dd_idx, NULL);
f73a1c7d 6137 last_sector = bio_end_sector(raid_bio);
46031f9a
RBJ		 6138
6139 for (; logical_sector < last_sector;
387bb173
NB		 6140 logical_sector += STRIPE_SECTORS,
6141 sector += STRIPE_SECTORS,
6142 scnt++) {
46031f9a 6143
0472a42b 6144 if (scnt < offset)
46031f9a
RBJ		 6145 /* already done this stripe */
6146 continue;
6147
6d036f7d 6148 sh = raid5_get_active_stripe(conf, sector, 0, 1, 1);
46031f9a
RBJ		 6149
6150 if (!sh) {
6151 /* failed to get a stripe - must wait */
46031f9a 6152 conf->retry_read_aligned = raid_bio;
0472a42b 6153 conf->retry_read_offset = scnt;
46031f9a
RBJ		 6154 return handled;
6155 }
6156
da41ba65 6157 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) {
6d036f7d 6158 raid5_release_stripe(sh);
387bb173 6159 conf->retry_read_aligned = raid_bio;
0472a42b 6160 conf->retry_read_offset = scnt;
387bb173
NB		 6161 return handled;
6162 }
6163
3f9e7c14 6164 set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags);
36d1c647 6165 handle_stripe(sh);
6d036f7d 6166 raid5_release_stripe(sh);
46031f9a
RBJ		 6167 handled++;
6168 }
016c76ac
N		 6169
6170 bio_endio(raid_bio);
6171
46031f9a 6172 if (atomic_dec_and_test(&conf->active_aligned_reads))
b1b46486 6173 wake_up(&conf->wait_for_quiescent);
46031f9a
RBJ		 6174 return handled;
6175}
6176
bfc90cb0 6177static int handle_active_stripes(struct r5conf *conf, int group,
566c09c5
SL		 6178 struct r5worker *worker,
6179 struct list_head *temp_inactive_list)
efcd487c
CH		 6180 __releases(&conf->device_lock)
6181 __acquires(&conf->device_lock)
46a06401
SL		 6182{
6183 struct stripe_head *batch[MAX_STRIPE_BATCH], *sh;
566c09c5
SL		 6184 int i, batch_size = 0, hash;
6185 bool release_inactive = false;
46a06401
SL		 6186
6187 while (batch_size < MAX_STRIPE_BATCH &&
851c30c9 6188 (sh = __get_priority_stripe(conf, group)) != NULL)
46a06401
SL		 6189 batch[batch_size++] = sh;
6190
566c09c5
SL		 6191 if (batch_size == 0) {
6192 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
6193 if (!list_empty(temp_inactive_list + i))
6194 break;
a8c34f91
SL		 6195 if (i == NR_STRIPE_HASH_LOCKS) {
6196 spin_unlock_irq(&conf->device_lock);
1532d9e8 6197 log_flush_stripe_to_raid(conf);
a8c34f91 6198 spin_lock_irq(&conf->device_lock);
566c09c5 6199 return batch_size;
a8c34f91 6200 }
566c09c5
SL		 6201 release_inactive = true;
6202 }
46a06401
SL		 6203 spin_unlock_irq(&conf->device_lock);
6204
566c09c5
SL		 6205 release_inactive_stripe_list(conf, temp_inactive_list,
6206 NR_STRIPE_HASH_LOCKS);
6207
a8c34f91 6208 r5l_flush_stripe_to_raid(conf->log);
566c09c5
SL		 6209 if (release_inactive) {
6210 spin_lock_irq(&conf->device_lock);
6211 return 0;
6212 }
6213
46a06401
SL		 6214 for (i = 0; i < batch_size; i++)
6215 handle_stripe(batch[i]);
ff875738 6216 log_write_stripe_run(conf);
46a06401
SL		 6217
6218 cond_resched();
6219
6220 spin_lock_irq(&conf->device_lock);
566c09c5
SL		 6221 for (i = 0; i < batch_size; i++) {
6222 hash = batch[i]->hash_lock_index;
6223 __release_stripe(conf, batch[i], &temp_inactive_list[hash]);
6224 }
46a06401
SL		 6225 return batch_size;
6226}
46031f9a 6227
851c30c9
SL		 6228static void raid5_do_work(struct work_struct *work)
6229{
6230 struct r5worker *worker = container_of(work, struct r5worker, work);
6231 struct r5worker_group *group = worker->group;
6232 struct r5conf *conf = group->conf;
16d997b7 6233 struct mddev *mddev = conf->mddev;
851c30c9
SL		 6234 int group_id = group - conf->worker_groups;
6235 int handled;
6236 struct blk_plug plug;
6237
6238 pr_debug("+++ raid5worker active\n");
6239
6240 blk_start_plug(&plug);
6241 handled = 0;
6242 spin_lock_irq(&conf->device_lock);
6243 while (1) {
6244 int batch_size, released;
6245
566c09c5 6246 released = release_stripe_list(conf, worker->temp_inactive_list);
851c30c9 6247
566c09c5
SL		 6248 batch_size = handle_active_stripes(conf, group_id, worker,
6249 worker->temp_inactive_list);
bfc90cb0 6250 worker->working = false;
851c30c9
SL		 6251 if (!batch_size && !released)
6252 break;
6253 handled += batch_size;
16d997b7
N		 6254 wait_event_lock_irq(mddev->sb_wait,
6255 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags),
6256 conf->device_lock);
851c30c9
SL		 6257 }
6258 pr_debug("%d stripes handled\n", handled);
6259
6260 spin_unlock_irq(&conf->device_lock);
7e96d559 6261
9c72a18e
SL		 6262 flush_deferred_bios(conf);
6263
6264 r5l_flush_stripe_to_raid(conf->log);
6265
7e96d559 6266 async_tx_issue_pending_all();
851c30c9
SL		 6267 blk_finish_plug(&plug);
6268
6269 pr_debug("--- raid5worker inactive\n");
6270}
6271
1da177e4
LT		 6272/*
6273 * This is our raid5 kernel thread.
6274 *
6275 * We scan the hash table for stripes which can be handled now.
6276 * During the scan, completed stripes are saved for us by the interrupt
6277 * handler, so that they will not have to wait for our next wakeup.
6278 */
4ed8731d 6279static void raid5d(struct md_thread *thread)
1da177e4 6280{
4ed8731d 6281 struct mddev *mddev = thread->mddev;
d1688a6d 6282 struct r5conf *conf = mddev->private;
1da177e4 6283 int handled;
e1dfa0a2 6284 struct blk_plug plug;
1da177e4 6285
45b4233c 6286 pr_debug("+++ raid5d active\n");
1da177e4
LT		 6287
6288 md_check_recovery(mddev);
1da177e4 6289
e1dfa0a2 6290 blk_start_plug(&plug);
1da177e4
LT		 6291 handled = 0;
6292 spin_lock_irq(&conf->device_lock);
6293 while (1) {
46031f9a 6294 struct bio *bio;
773ca82f 6295 int batch_size, released;
0472a42b 6296 unsigned int offset;
773ca82f 6297
566c09c5 6298 released = release_stripe_list(conf, conf->temp_inactive_list);
edbe83ab
N		 6299 if (released)
6300 clear_bit(R5_DID_ALLOC, &conf->cache_state);
1da177e4 6301
0021b7bc 6302 if (
7c13edc8
N		 6303 !list_empty(&conf->bitmap_list)) {
6304 /* Now is a good time to flush some bitmap updates */
6305 conf->seq_flush++;
700e432d 6306 spin_unlock_irq(&conf->device_lock);
e64e4018 6307 md_bitmap_unplug(mddev->bitmap);
700e432d 6308 spin_lock_irq(&conf->device_lock);
7c13edc8 6309 conf->seq_write = conf->seq_flush;
566c09c5 6310 activate_bit_delay(conf, conf->temp_inactive_list);
72626685 6311 }
0021b7bc 6312 raid5_activate_delayed(conf);
72626685 6313
0472a42b 6314 while ((bio = remove_bio_from_retry(conf, &offset))) {
46031f9a
RBJ		 6315 int ok;
6316 spin_unlock_irq(&conf->device_lock);
0472a42b 6317 ok = retry_aligned_read(conf, bio, offset);
46031f9a
RBJ		 6318 spin_lock_irq(&conf->device_lock);
6319 if (!ok)
6320 break;
6321 handled++;
6322 }
6323
566c09c5
SL		 6324 batch_size = handle_active_stripes(conf, ANY_GROUP, NULL,
6325 conf->temp_inactive_list);
773ca82f 6326 if (!batch_size && !released)
1da177e4 6327 break;
46a06401 6328 handled += batch_size;
1da177e4 6329
2953079c 6330 if (mddev->sb_flags & ~(1 << MD_SB_CHANGE_PENDING)) {
46a06401 6331 spin_unlock_irq(&conf->device_lock);
de393cde 6332 md_check_recovery(mddev);
46a06401
SL		 6333 spin_lock_irq(&conf->device_lock);
6334 }
1da177e4 6335 }
45b4233c 6336 pr_debug("%d stripes handled\n", handled);
1da177e4
LT		 6337
6338 spin_unlock_irq(&conf->device_lock);
2d5b569b
N		 6339 if (test_and_clear_bit(R5_ALLOC_MORE, &conf->cache_state) &&
6340 mutex_trylock(&conf->cache_size_mutex)) {
edbe83ab
N		 6341 grow_one_stripe(conf, __GFP_NOWARN);
6342 /* Set flag even if allocation failed. This helps
6343 * slow down allocation requests when mem is short
6344 */
6345 set_bit(R5_DID_ALLOC, &conf->cache_state);
2d5b569b 6346 mutex_unlock(&conf->cache_size_mutex);
edbe83ab 6347 }
1da177e4 6348
765d704d
SL		 6349 flush_deferred_bios(conf);
6350
0576b1c6
SL		 6351 r5l_flush_stripe_to_raid(conf->log);
6352
c9f21aaf 6353 async_tx_issue_pending_all();
e1dfa0a2 6354 blk_finish_plug(&plug);
1da177e4 6355
45b4233c 6356 pr_debug("--- raid5d inactive\n");
1da177e4
LT		 6357}
6358
3f294f4f 6359static ssize_t
fd01b88c 6360raid5_show_stripe_cache_size(struct mddev *mddev, char *page)
3f294f4f 6361{
7b1485ba
N		 6362 struct r5conf *conf;
6363 int ret = 0;
6364 spin_lock(&mddev->lock);
6365 conf = mddev->private;
96de1e66 6366 if (conf)
edbe83ab 6367 ret = sprintf(page, "%d\n", conf->min_nr_stripes);
7b1485ba
N		 6368 spin_unlock(&mddev->lock);
6369 return ret;
3f294f4f
N		 6370}
6371
c41d4ac4 6372int
fd01b88c 6373raid5_set_cache_size(struct mddev *mddev, int size)
3f294f4f 6374{
483cbbed 6375 int result = 0;
d1688a6d 6376 struct r5conf *conf = mddev->private;
b5470dc5 6377
c41d4ac4 6378 if (size <= 16 || size > 32768)
3f294f4f 6379 return -EINVAL;
486f0644 6380
edbe83ab 6381 conf->min_nr_stripes = size;
2d5b569b 6382 mutex_lock(&conf->cache_size_mutex);
486f0644
N		 6383 while (size < conf->max_nr_stripes &&
6384 drop_one_stripe(conf))
6385 ;
2d5b569b 6386 mutex_unlock(&conf->cache_size_mutex);
486f0644 6387
2214c260 6388 md_allow_write(mddev);
486f0644 6389
2d5b569b 6390 mutex_lock(&conf->cache_size_mutex);
486f0644 6391 while (size > conf->max_nr_stripes)
483cbbed
AN		 6392 if (!grow_one_stripe(conf, GFP_KERNEL)) {
6393 conf->min_nr_stripes = conf->max_nr_stripes;
6394 result = -ENOMEM;
486f0644 6395 break;
483cbbed 6396 }
2d5b569b 6397 mutex_unlock(&conf->cache_size_mutex);
486f0644 6398
483cbbed 6399 return result;
c41d4ac4
N		 6400}
6401EXPORT_SYMBOL(raid5_set_cache_size);
6402
6403static ssize_t
fd01b88c 6404raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len)
c41d4ac4 6405{
6791875e 6406 struct r5conf *conf;
c41d4ac4
N		 6407 unsigned long new;
6408 int err;
6409
6410 if (len >= PAGE_SIZE)
6411 return -EINVAL;
b29bebd6 6412 if (kstrtoul(page, 10, &new))
c41d4ac4 6413 return -EINVAL;
6791875e 6414 err = mddev_lock(mddev);
c41d4ac4
N		 6415 if (err)
6416 return err;
6791875e
N		 6417 conf = mddev->private;
6418 if (!conf)
6419 err = -ENODEV;
6420 else
6421 err = raid5_set_cache_size(mddev, new);
6422 mddev_unlock(mddev);
6423
6424 return err ?: len;
3f294f4f 6425}
007583c9 6426
96de1e66
N		 6427static struct md_sysfs_entry
6428raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
6429 raid5_show_stripe_cache_size,
6430 raid5_store_stripe_cache_size);
3f294f4f 6431
d06f191f
MS		 6432static ssize_t
6433raid5_show_rmw_level(struct mddev *mddev, char *page)
6434{
6435 struct r5conf *conf = mddev->private;
6436 if (conf)
6437 return sprintf(page, "%d\n", conf->rmw_level);
6438 else
6439 return 0;
6440}
6441
6442static ssize_t
6443raid5_store_rmw_level(struct mddev *mddev, const char *page, size_t len)
6444{
6445 struct r5conf *conf = mddev->private;
6446 unsigned long new;
6447
6448 if (!conf)
6449 return -ENODEV;
6450
6451 if (len >= PAGE_SIZE)
6452 return -EINVAL;
6453
6454 if (kstrtoul(page, 10, &new))
6455 return -EINVAL;
6456
6457 if (new != PARITY_DISABLE_RMW && !raid6_call.xor_syndrome)
6458 return -EINVAL;
6459
6460 if (new != PARITY_DISABLE_RMW &&
6461 new != PARITY_ENABLE_RMW &&
6462 new != PARITY_PREFER_RMW)
6463 return -EINVAL;
6464
6465 conf->rmw_level = new;
6466 return len;
6467}
6468
6469static struct md_sysfs_entry
6470raid5_rmw_level = __ATTR(rmw_level, S_IRUGO | S_IWUSR,
6471 raid5_show_rmw_level,
6472 raid5_store_rmw_level);
6473
6474
8b3e6cdc 6475static ssize_t
fd01b88c 6476raid5_show_preread_threshold(struct mddev *mddev, char *page)
8b3e6cdc 6477{
7b1485ba
N		 6478 struct r5conf *conf;
6479 int ret = 0;
6480 spin_lock(&mddev->lock);
6481 conf = mddev->private;
8b3e6cdc 6482 if (conf)
7b1485ba
N		 6483 ret = sprintf(page, "%d\n", conf->bypass_threshold);
6484 spin_unlock(&mddev->lock);
6485 return ret;
8b3e6cdc
DW		 6486}
6487
6488static ssize_t
fd01b88c 6489raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len)
8b3e6cdc 6490{
6791875e 6491 struct r5conf *conf;
4ef197d8 6492 unsigned long new;
6791875e
N		 6493 int err;
6494
8b3e6cdc
DW		 6495 if (len >= PAGE_SIZE)
6496 return -EINVAL;
b29bebd6 6497 if (kstrtoul(page, 10, &new))
8b3e6cdc 6498 return -EINVAL;
6791875e
N		 6499
6500 err = mddev_lock(mddev);
6501 if (err)
6502 return err;
6503 conf = mddev->private;
6504 if (!conf)
6505 err = -ENODEV;
edbe83ab 6506 else if (new > conf->min_nr_stripes)
6791875e
N		 6507 err = -EINVAL;
6508 else
6509 conf->bypass_threshold = new;
6510 mddev_unlock(mddev);
6511 return err ?: len;
8b3e6cdc
DW		 6512}
6513
6514static struct md_sysfs_entry
6515raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
6516 S_IRUGO | S_IWUSR,
6517 raid5_show_preread_threshold,
6518 raid5_store_preread_threshold);
6519
d592a996
SL		 6520static ssize_t
6521raid5_show_skip_copy(struct mddev *mddev, char *page)
6522{
7b1485ba
N		 6523 struct r5conf *conf;
6524 int ret = 0;
6525 spin_lock(&mddev->lock);
6526 conf = mddev->private;
d592a996 6527 if (conf)
7b1485ba
N		 6528 ret = sprintf(page, "%d\n", conf->skip_copy);
6529 spin_unlock(&mddev->lock);
6530 return ret;
d592a996
SL		 6531}
6532
6533static ssize_t
6534raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len)
6535{
6791875e 6536 struct r5conf *conf;
d592a996 6537 unsigned long new;
6791875e
N		 6538 int err;
6539
d592a996
SL		 6540 if (len >= PAGE_SIZE)
6541 return -EINVAL;
d592a996
SL		 6542 if (kstrtoul(page, 10, &new))
6543 return -EINVAL;
6544 new = !!new;
6791875e
N		 6545
6546 err = mddev_lock(mddev);
6547 if (err)
6548 return err;
6549 conf = mddev->private;
6550 if (!conf)
6551 err = -ENODEV;
6552 else if (new != conf->skip_copy) {
6553 mddev_suspend(mddev);
6554 conf->skip_copy = new;
6555 if (new)
dc3b17cc 6556 mddev->queue->backing_dev_info->capabilities |=
6791875e
N		 6557 BDI_CAP_STABLE_WRITES;
6558 else
dc3b17cc 6559 mddev->queue->backing_dev_info->capabilities &=
6791875e
N		 6560 ~BDI_CAP_STABLE_WRITES;
6561 mddev_resume(mddev);
6562 }
6563 mddev_unlock(mddev);
6564 return err ?: len;
d592a996
SL		 6565}
6566
6567static struct md_sysfs_entry
6568raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR,
6569 raid5_show_skip_copy,
6570 raid5_store_skip_copy);
6571
3f294f4f 6572static ssize_t
fd01b88c 6573stripe_cache_active_show(struct mddev *mddev, char *page)
3f294f4f 6574{
d1688a6d 6575 struct r5conf *conf = mddev->private;
96de1e66
N		 6576 if (conf)
6577 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
6578 else
6579 return 0;
3f294f4f
N		 6580}
6581
96de1e66
N		 6582static struct md_sysfs_entry
6583raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
3f294f4f 6584
b721420e
SL		 6585static ssize_t
6586raid5_show_group_thread_cnt(struct mddev *mddev, char *page)
6587{
7b1485ba
N		 6588 struct r5conf *conf;
6589 int ret = 0;
6590 spin_lock(&mddev->lock);
6591 conf = mddev->private;
b721420e 6592 if (conf)
7b1485ba
N		 6593 ret = sprintf(page, "%d\n", conf->worker_cnt_per_group);
6594 spin_unlock(&mddev->lock);
6595 return ret;
b721420e
SL		 6596}
6597
60aaf933 6598static int alloc_thread_groups(struct r5conf *conf, int cnt,
6599 int *group_cnt,
6600 int *worker_cnt_per_group,
6601 struct r5worker_group **worker_groups);
b721420e
SL		 6602static ssize_t
6603raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len)
6604{
6791875e 6605 struct r5conf *conf;
7d5d7b50 6606 unsigned int new;
b721420e 6607 int err;
60aaf933 6608 struct r5worker_group *new_groups, *old_groups;
6609 int group_cnt, worker_cnt_per_group;
b721420e
SL		 6610
6611 if (len >= PAGE_SIZE)
6612 return -EINVAL;
7d5d7b50
SL		 6613 if (kstrtouint(page, 10, &new))
6614 return -EINVAL;
6615 /* 8192 should be big enough */
6616 if (new > 8192)
b721420e
SL		 6617 return -EINVAL;
6618
6791875e
N		 6619 err = mddev_lock(mddev);
6620 if (err)
6621 return err;
6622 conf = mddev->private;
6623 if (!conf)
6624 err = -ENODEV;
6625 else if (new != conf->worker_cnt_per_group) {
6626 mddev_suspend(mddev);
b721420e 6627
6791875e
N		 6628 old_groups = conf->worker_groups;
6629 if (old_groups)
6630 flush_workqueue(raid5_wq);
d206dcfa 6631
6791875e
N		 6632 err = alloc_thread_groups(conf, new,
6633 &group_cnt, &worker_cnt_per_group,
6634 &new_groups);
6635 if (!err) {
6636 spin_lock_irq(&conf->device_lock);
6637 conf->group_cnt = group_cnt;
6638 conf->worker_cnt_per_group = worker_cnt_per_group;
6639 conf->worker_groups = new_groups;
6640 spin_unlock_irq(&conf->device_lock);
b721420e 6641
6791875e
N		 6642 if (old_groups)
6643 kfree(old_groups[0].workers);
6644 kfree(old_groups);
6645 }
6646 mddev_resume(mddev);
b721420e 6647 }
6791875e 6648 mddev_unlock(mddev);
b721420e 6649
6791875e 6650 return err ?: len;
b721420e
SL		 6651}
6652
6653static struct md_sysfs_entry
6654raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR,
6655 raid5_show_group_thread_cnt,
6656 raid5_store_group_thread_cnt);
6657
007583c9 6658static struct attribute *raid5_attrs[] = {
3f294f4f
N		 6659 &raid5_stripecache_size.attr,
6660 &raid5_stripecache_active.attr,
8b3e6cdc 6661 &raid5_preread_bypass_threshold.attr,
b721420e 6662 &raid5_group_thread_cnt.attr,
d592a996 6663 &raid5_skip_copy.attr,
d06f191f 6664 &raid5_rmw_level.attr,
2c7da14b 6665 &r5c_journal_mode.attr,
a596d086 6666 &ppl_write_hint.attr,
3f294f4f
N		 6667 NULL,
6668};
007583c9
N		 6669static struct attribute_group raid5_attrs_group = {
6670 .name = NULL,
6671 .attrs = raid5_attrs,
3f294f4f
N		 6672};
6673
60aaf933 6674static int alloc_thread_groups(struct r5conf *conf, int cnt,
6675 int *group_cnt,
6676 int *worker_cnt_per_group,
6677 struct r5worker_group **worker_groups)
851c30c9 6678{
566c09c5 6679 int i, j, k;
851c30c9
SL		 6680 ssize_t size;
6681 struct r5worker *workers;
6682
60aaf933 6683 *worker_cnt_per_group = cnt;
851c30c9 6684 if (cnt == 0) {
60aaf933 6685 *group_cnt = 0;
6686 *worker_groups = NULL;
851c30c9
SL		 6687 return 0;
6688 }
60aaf933 6689 *group_cnt = num_possible_nodes();
851c30c9 6690 size = sizeof(struct r5worker) * cnt;
6396bb22
KC		 6691 workers = kcalloc(size, *group_cnt, GFP_NOIO);
6692 *worker_groups = kcalloc(*group_cnt, sizeof(struct r5worker_group),
6693 GFP_NOIO);
60aaf933 6694 if (!*worker_groups || !workers) {
851c30c9 6695 kfree(workers);
60aaf933 6696 kfree(*worker_groups);
851c30c9
SL		 6697 return -ENOMEM;
6698 }
6699
60aaf933 6700 for (i = 0; i < *group_cnt; i++) {
851c30c9
SL		 6701 struct r5worker_group *group;
6702
0c775d52 6703 group = &(*worker_groups)[i];
851c30c9 6704 INIT_LIST_HEAD(&group->handle_list);
535ae4eb 6705 INIT_LIST_HEAD(&group->loprio_list);
851c30c9
SL		 6706 group->conf = conf;
6707 group->workers = workers + i * cnt;
6708
6709 for (j = 0; j < cnt; j++) {
566c09c5
SL		 6710 struct r5worker *worker = group->workers + j;
6711 worker->group = group;
6712 INIT_WORK(&worker->work, raid5_do_work);
6713
6714 for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++)
6715 INIT_LIST_HEAD(worker->temp_inactive_list + k);
851c30c9
SL		 6716 }
6717 }
6718
6719 return 0;
6720}
6721
6722static void free_thread_groups(struct r5conf *conf)
6723{
6724 if (conf->worker_groups)
6725 kfree(conf->worker_groups[0].workers);
6726 kfree(conf->worker_groups);
6727 conf->worker_groups = NULL;
6728}
6729
80c3a6ce 6730static sector_t
fd01b88c 6731raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks)
80c3a6ce 6732{
d1688a6d 6733 struct r5conf *conf = mddev->private;
80c3a6ce
DW		 6734
6735 if (!sectors)
6736 sectors = mddev->dev_sectors;
5e5e3e78 6737 if (!raid_disks)
7ec05478 6738 /* size is defined by the smallest of previous and new size */
5e5e3e78 6739 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
80c3a6ce 6740
3cb5edf4
N		 6741 sectors &= ~((sector_t)conf->chunk_sectors - 1);
6742 sectors &= ~((sector_t)conf->prev_chunk_sectors - 1);
80c3a6ce
DW		 6743 return sectors * (raid_disks - conf->max_degraded);
6744}
6745
789b5e03
ON		 6746static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
6747{
6748 safe_put_page(percpu->spare_page);
789b5e03 6749 percpu->spare_page = NULL;
b330e6a4 6750 kvfree(percpu->scribble);
789b5e03
ON		 6751 percpu->scribble = NULL;
6752}
6753
6754static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
6755{
b330e6a4 6756 if (conf->level == 6 && !percpu->spare_page) {
789b5e03 6757 percpu->spare_page = alloc_page(GFP_KERNEL);
b330e6a4
KO		 6758 if (!percpu->spare_page)
6759 return -ENOMEM;
6760 }
6761
6762 if (scribble_alloc(percpu,
6763 max(conf->raid_disks,
6764 conf->previous_raid_disks),
6765 max(conf->chunk_sectors,
6766 conf->prev_chunk_sectors)
6767 / STRIPE_SECTORS,
6768 GFP_KERNEL)) {
789b5e03
ON		 6769 free_scratch_buffer(conf, percpu);
6770 return -ENOMEM;
6771 }
6772
6773 return 0;
6774}
6775
29c6d1bb 6776static int raid456_cpu_dead(unsigned int cpu, struct hlist_node *node)
36d1c647 6777{
29c6d1bb
SAS		 6778 struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node);
6779
6780 free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
6781 return 0;
6782}
36d1c647 6783
29c6d1bb
SAS		 6784static void raid5_free_percpu(struct r5conf *conf)
6785{
36d1c647
DW		 6786 if (!conf->percpu)
6787 return;
6788
29c6d1bb 6789 cpuhp_state_remove_instance(CPUHP_MD_RAID5_PREPARE, &conf->node);
36d1c647
DW		 6790 free_percpu(conf->percpu);
6791}
6792
d1688a6d 6793static void free_conf(struct r5conf *conf)
95fc17aa 6794{
d7bd398e
SL		 6795 int i;
6796
ff875738
AP		 6797 log_exit(conf);
6798
565e0450 6799 unregister_shrinker(&conf->shrinker);
851c30c9 6800 free_thread_groups(conf);
95fc17aa 6801 shrink_stripes(conf);
36d1c647 6802 raid5_free_percpu(conf);
d7bd398e
SL		 6803 for (i = 0; i < conf->pool_size; i++)
6804 if (conf->disks[i].extra_page)
6805 put_page(conf->disks[i].extra_page);
95fc17aa 6806 kfree(conf->disks);
afeee514 6807 bioset_exit(&conf->bio_split);
95fc17aa 6808 kfree(conf->stripe_hashtbl);
aaf9f12e 6809 kfree(conf->pending_data);
95fc17aa
DW		 6810 kfree(conf);
6811}
6812
29c6d1bb 6813static int raid456_cpu_up_prepare(unsigned int cpu, struct hlist_node *node)
36d1c647 6814{
29c6d1bb 6815 struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node);
36d1c647
DW		 6816 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
6817
29c6d1bb 6818 if (alloc_scratch_buffer(conf, percpu)) {
cc6167b4
N		 6819 pr_warn("%s: failed memory allocation for cpu%u\n",
6820 __func__, cpu);
29c6d1bb 6821 return -ENOMEM;
36d1c647 6822 }
29c6d1bb 6823 return 0;
36d1c647 6824}
36d1c647 6825
d1688a6d 6826static int raid5_alloc_percpu(struct r5conf *conf)
36d1c647 6827{
789b5e03 6828 int err = 0;
36d1c647 6829
789b5e03
ON		 6830 conf->percpu = alloc_percpu(struct raid5_percpu);
6831 if (!conf->percpu)
36d1c647 6832 return -ENOMEM;
789b5e03 6833
29c6d1bb 6834 err = cpuhp_state_add_instance(CPUHP_MD_RAID5_PREPARE, &conf->node);
27a353c0
SL		 6835 if (!err) {
6836 conf->scribble_disks = max(conf->raid_disks,
6837 conf->previous_raid_disks);
6838 conf->scribble_sectors = max(conf->chunk_sectors,
6839 conf->prev_chunk_sectors);
6840 }
36d1c647
DW		 6841 return err;
6842}
6843
edbe83ab
N		 6844static unsigned long raid5_cache_scan(struct shrinker *shrink,
6845 struct shrink_control *sc)
6846{
6847 struct r5conf *conf = container_of(shrink, struct r5conf, shrinker);
2d5b569b
N		 6848 unsigned long ret = SHRINK_STOP;
6849
6850 if (mutex_trylock(&conf->cache_size_mutex)) {
6851 ret= 0;
49895bcc
N		 6852 while (ret < sc->nr_to_scan &&
6853 conf->max_nr_stripes > conf->min_nr_stripes) {
2d5b569b
N		 6854 if (drop_one_stripe(conf) == 0) {
6855 ret = SHRINK_STOP;
6856 break;
6857 }
6858 ret++;
6859 }
6860 mutex_unlock(&conf->cache_size_mutex);
edbe83ab
N		 6861 }
6862 return ret;
6863}
6864
6865static unsigned long raid5_cache_count(struct shrinker *shrink,
6866 struct shrink_control *sc)
6867{
6868 struct r5conf *conf = container_of(shrink, struct r5conf, shrinker);
6869
6870 if (conf->max_nr_stripes < conf->min_nr_stripes)
6871 /* unlikely, but not impossible */
6872 return 0;
6873 return conf->max_nr_stripes - conf->min_nr_stripes;
6874}
6875
d1688a6d 6876static struct r5conf *setup_conf(struct mddev *mddev)
1da177e4 6877{
d1688a6d 6878 struct r5conf *conf;
5e5e3e78 6879 int raid_disk, memory, max_disks;
3cb03002 6880 struct md_rdev *rdev;
1da177e4 6881 struct disk_info *disk;
0232605d 6882 char pers_name[6];
566c09c5 6883 int i;
60aaf933 6884 int group_cnt, worker_cnt_per_group;
6885 struct r5worker_group *new_group;
afeee514 6886 int ret;
1da177e4 6887
91adb564
N		 6888 if (mddev->new_level != 5
6889 && mddev->new_level != 4
6890 && mddev->new_level != 6) {
cc6167b4
N		 6891 pr_warn("md/raid:%s: raid level not set to 4/5/6 (%d)\n",
6892 mdname(mddev), mddev->new_level);
91adb564 6893 return ERR_PTR(-EIO);
1da177e4 6894 }
91adb564
N		 6895 if ((mddev->new_level == 5
6896 && !algorithm_valid_raid5(mddev->new_layout)) ||
6897 (mddev->new_level == 6
6898 && !algorithm_valid_raid6(mddev->new_layout))) {
cc6167b4
N		 6899 pr_warn("md/raid:%s: layout %d not supported\n",
6900 mdname(mddev), mddev->new_layout);
91adb564 6901 return ERR_PTR(-EIO);
99c0fb5f 6902 }
91adb564 6903 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
cc6167b4
N		 6904 pr_warn("md/raid:%s: not enough configured devices (%d, minimum 4)\n",
6905 mdname(mddev), mddev->raid_disks);
91adb564 6906 return ERR_PTR(-EINVAL);
4bbf3771
N		 6907 }
6908
664e7c41
AN		 6909 if (!mddev->new_chunk_sectors ||
6910 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
6911 !is_power_of_2(mddev->new_chunk_sectors)) {
cc6167b4
N		 6912 pr_warn("md/raid:%s: invalid chunk size %d\n",
6913 mdname(mddev), mddev->new_chunk_sectors << 9);
91adb564 6914 return ERR_PTR(-EINVAL);
f6705578
N		 6915 }
6916
d1688a6d 6917 conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL);
91adb564 6918 if (conf == NULL)
1da177e4 6919 goto abort;
aaf9f12e
SL		 6920 INIT_LIST_HEAD(&conf->free_list);
6921 INIT_LIST_HEAD(&conf->pending_list);
6396bb22
KC		 6922 conf->pending_data = kcalloc(PENDING_IO_MAX,
6923 sizeof(struct r5pending_data),
6924 GFP_KERNEL);
aaf9f12e
SL		 6925 if (!conf->pending_data)
6926 goto abort;
6927 for (i = 0; i < PENDING_IO_MAX; i++)
6928 list_add(&conf->pending_data[i].sibling, &conf->free_list);
851c30c9 6929 /* Don't enable multi-threading by default*/
60aaf933 6930 if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group,
6931 &new_group)) {
6932 conf->group_cnt = group_cnt;
6933 conf->worker_cnt_per_group = worker_cnt_per_group;
6934 conf->worker_groups = new_group;
6935 } else
851c30c9 6936 goto abort;
f5efd45a 6937 spin_lock_init(&conf->device_lock);
c46501b2 6938 seqcount_init(&conf->gen_lock);
2d5b569b 6939 mutex_init(&conf->cache_size_mutex);
b1b46486 6940 init_waitqueue_head(&conf->wait_for_quiescent);
6ab2a4b8 6941 init_waitqueue_head(&conf->wait_for_stripe);
f5efd45a
DW		 6942 init_waitqueue_head(&conf->wait_for_overlap);
6943 INIT_LIST_HEAD(&conf->handle_list);
535ae4eb 6944 INIT_LIST_HEAD(&conf->loprio_list);
f5efd45a
DW		 6945 INIT_LIST_HEAD(&conf->hold_list);
6946 INIT_LIST_HEAD(&conf->delayed_list);
6947 INIT_LIST_HEAD(&conf->bitmap_list);
773ca82f 6948 init_llist_head(&conf->released_stripes);
f5efd45a
DW		 6949 atomic_set(&conf->active_stripes, 0);
6950 atomic_set(&conf->preread_active_stripes, 0);
6951 atomic_set(&conf->active_aligned_reads, 0);
765d704d
SL		 6952 spin_lock_init(&conf->pending_bios_lock);
6953 conf->batch_bio_dispatch = true;
6954 rdev_for_each(rdev, mddev) {
6955 if (test_bit(Journal, &rdev->flags))
6956 continue;
6957 if (blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
6958 conf->batch_bio_dispatch = false;
6959 break;
6960 }
6961 }
6962
f5efd45a 6963 conf->bypass_threshold = BYPASS_THRESHOLD;
d890fa2b 6964 conf->recovery_disabled = mddev->recovery_disabled - 1;
91adb564
N		 6965
6966 conf->raid_disks = mddev->raid_disks;
6967 if (mddev->reshape_position == MaxSector)
6968 conf->previous_raid_disks = mddev->raid_disks;
6969 else
f6705578 6970 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
5e5e3e78 6971 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
f6705578 6972
6396bb22 6973 conf->disks = kcalloc(max_disks, sizeof(struct disk_info),
b55e6bfc 6974 GFP_KERNEL);
d7bd398e 6975
b55e6bfc
N		 6976 if (!conf->disks)
6977 goto abort;
9ffae0cf 6978
d7bd398e
SL		 6979 for (i = 0; i < max_disks; i++) {
6980 conf->disks[i].extra_page = alloc_page(GFP_KERNEL);
6981 if (!conf->disks[i].extra_page)
6982 goto abort;
6983 }
6984
afeee514
KO		 6985 ret = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0);
6986 if (ret)
dd7a8f5d 6987 goto abort;
1da177e4
LT		 6988 conf->mddev = mddev;
6989
fccddba0 6990 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
1da177e4 6991 goto abort;
1da177e4 6992
566c09c5
SL		 6993 /* We init hash_locks[0] separately to that it can be used
6994 * as the reference lock in the spin_lock_nest_lock() call
6995 * in lock_all_device_hash_locks_irq in order to convince
6996 * lockdep that we know what we are doing.
6997 */
6998 spin_lock_init(conf->hash_locks);
6999 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
7000 spin_lock_init(conf->hash_locks + i);
7001
7002 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
7003 INIT_LIST_HEAD(conf->inactive_list + i);
7004
7005 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
7006 INIT_LIST_HEAD(conf->temp_inactive_list + i);
7007
1e6d690b
SL		 7008 atomic_set(&conf->r5c_cached_full_stripes, 0);
7009 INIT_LIST_HEAD(&conf->r5c_full_stripe_list);
7010 atomic_set(&conf->r5c_cached_partial_stripes, 0);
7011 INIT_LIST_HEAD(&conf->r5c_partial_stripe_list);
e33fbb9c
SL		 7012 atomic_set(&conf->r5c_flushing_full_stripes, 0);
7013 atomic_set(&conf->r5c_flushing_partial_stripes, 0);
1e6d690b 7014
36d1c647 7015 conf->level = mddev->new_level;
46d5b785 7016 conf->chunk_sectors = mddev->new_chunk_sectors;
36d1c647
DW		 7017 if (raid5_alloc_percpu(conf) != 0)
7018 goto abort;
7019
0c55e022 7020 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
1da177e4 7021
dafb20fa 7022 rdev_for_each(rdev, mddev) {
1da177e4 7023 raid_disk = rdev->raid_disk;
5e5e3e78 7024 if (raid_disk >= max_disks
f2076e7d 7025 || raid_disk < 0 || test_bit(Journal, &rdev->flags))
1da177e4
LT		 7026 continue;
7027 disk = conf->disks + raid_disk;
7028
17045f52
N		 7029 if (test_bit(Replacement, &rdev->flags)) {
7030 if (disk->replacement)
7031 goto abort;
7032 disk->replacement = rdev;
7033 } else {
7034 if (disk->rdev)
7035 goto abort;
7036 disk->rdev = rdev;
7037 }
1da177e4 7038
b2d444d7 7039 if (test_bit(In_sync, &rdev->flags)) {
1da177e4 7040 char b[BDEVNAME_SIZE];
cc6167b4
N		 7041 pr_info("md/raid:%s: device %s operational as raid disk %d\n",
7042 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
d6b212f4 7043 } else if (rdev->saved_raid_disk != raid_disk)
8c2e870a
NB		 7044 /* Cannot rely on bitmap to complete recovery */
7045 conf->fullsync = 1;
1da177e4
LT		 7046 }
7047
91adb564 7048 conf->level = mddev->new_level;
584acdd4 7049 if (conf->level == 6) {
16a53ecc 7050 conf->max_degraded = 2;
584acdd4
MS		 7051 if (raid6_call.xor_syndrome)
7052 conf->rmw_level = PARITY_ENABLE_RMW;
7053 else
7054 conf->rmw_level = PARITY_DISABLE_RMW;
7055 } else {
16a53ecc 7056 conf->max_degraded = 1;
584acdd4
MS		 7057 conf->rmw_level = PARITY_ENABLE_RMW;
7058 }
91adb564 7059 conf->algorithm = mddev->new_layout;
fef9c61f 7060 conf->reshape_progress = mddev->reshape_position;
e183eaed 7061 if (conf->reshape_progress != MaxSector) {
09c9e5fa 7062 conf->prev_chunk_sectors = mddev->chunk_sectors;
e183eaed 7063 conf->prev_algo = mddev->layout;
5cac6bcb
N		 7064 } else {
7065 conf->prev_chunk_sectors = conf->chunk_sectors;
7066 conf->prev_algo = conf->algorithm;
e183eaed 7067 }
1da177e4 7068
edbe83ab 7069 conf->min_nr_stripes = NR_STRIPES;
ad5b0f76
SL		 7070 if (mddev->reshape_position != MaxSector) {
7071 int stripes = max_t(int,
7072 ((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4,
7073 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4);
7074 conf->min_nr_stripes = max(NR_STRIPES, stripes);
7075 if (conf->min_nr_stripes != NR_STRIPES)
cc6167b4 7076 pr_info("md/raid:%s: force stripe size %d for reshape\n",
ad5b0f76
SL		 7077 mdname(mddev), conf->min_nr_stripes);
7078 }
edbe83ab 7079 memory = conf->min_nr_stripes * (sizeof(struct stripe_head) +
5e5e3e78 7080 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
4bda556a 7081 atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS);
edbe83ab 7082 if (grow_stripes(conf, conf->min_nr_stripes)) {
cc6167b4
N		 7083 pr_warn("md/raid:%s: couldn't allocate %dkB for buffers\n",
7084 mdname(mddev), memory);
91adb564
N		 7085 goto abort;
7086 } else
cc6167b4 7087 pr_debug("md/raid:%s: allocated %dkB\n", mdname(mddev), memory);
edbe83ab
N		 7088 /*
7089 * Losing a stripe head costs more than the time to refill it,
7090 * it reduces the queue depth and so can hurt throughput.
7091 * So set it rather large, scaled by number of devices.
7092 */
7093 conf->shrinker.seeks = DEFAULT_SEEKS * conf->raid_disks * 4;
7094 conf->shrinker.scan_objects = raid5_cache_scan;
7095 conf->shrinker.count_objects = raid5_cache_count;
7096 conf->shrinker.batch = 128;
7097 conf->shrinker.flags = 0;
6a0f53ff 7098 if (register_shrinker(&conf->shrinker)) {
cc6167b4
N		 7099 pr_warn("md/raid:%s: couldn't register shrinker.\n",
7100 mdname(mddev));
6a0f53ff
CY		 7101 goto abort;
7102 }
1da177e4 7103
0232605d
N		 7104 sprintf(pers_name, "raid%d", mddev->new_level);
7105 conf->thread = md_register_thread(raid5d, mddev, pers_name);
91adb564 7106 if (!conf->thread) {
cc6167b4
N		 7107 pr_warn("md/raid:%s: couldn't allocate thread.\n",
7108 mdname(mddev));
16a53ecc
N		 7109 goto abort;
7110 }
91adb564
N		 7111
7112 return conf;
7113
7114 abort:
7115 if (conf) {
95fc17aa 7116 free_conf(conf);
91adb564
N		 7117 return ERR_PTR(-EIO);
7118 } else
7119 return ERR_PTR(-ENOMEM);
7120}
7121
c148ffdc
N		 7122static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
7123{
7124 switch (algo) {
7125 case ALGORITHM_PARITY_0:
7126 if (raid_disk < max_degraded)
7127 return 1;
7128 break;
7129 case ALGORITHM_PARITY_N:
7130 if (raid_disk >= raid_disks - max_degraded)
7131 return 1;
7132 break;
7133 case ALGORITHM_PARITY_0_6:
f72ffdd6 7134 if (raid_disk == 0 ||
c148ffdc
N		 7135 raid_disk == raid_disks - 1)
7136 return 1;
7137 break;
7138 case ALGORITHM_LEFT_ASYMMETRIC_6:
7139 case ALGORITHM_RIGHT_ASYMMETRIC_6:
7140 case ALGORITHM_LEFT_SYMMETRIC_6:
7141 case ALGORITHM_RIGHT_SYMMETRIC_6:
7142 if (raid_disk == raid_disks - 1)
7143 return 1;
7144 }
7145 return 0;
7146}
7147
849674e4 7148static int raid5_run(struct mddev *mddev)
91adb564 7149{
d1688a6d 7150 struct r5conf *conf;
9f7c2220 7151 int working_disks = 0;
c148ffdc 7152 int dirty_parity_disks = 0;
3cb03002 7153 struct md_rdev *rdev;
713cf5a6 7154 struct md_rdev *journal_dev = NULL;
c148ffdc 7155 sector_t reshape_offset = 0;
17045f52 7156 int i;
b5254dd5
N		 7157 long long min_offset_diff = 0;
7158 int first = 1;
91adb564 7159
a415c0f1
N		 7160 if (mddev_init_writes_pending(mddev) < 0)
7161 return -ENOMEM;
7162
8c6ac868 7163 if (mddev->recovery_cp != MaxSector)
cc6167b4
N		 7164 pr_notice("md/raid:%s: not clean -- starting background reconstruction\n",
7165 mdname(mddev));
b5254dd5
N		 7166
7167 rdev_for_each(rdev, mddev) {
7168 long long diff;
713cf5a6 7169
f2076e7d 7170 if (test_bit(Journal, &rdev->flags)) {
713cf5a6 7171 journal_dev = rdev;
f2076e7d
SL		 7172 continue;
7173 }
b5254dd5
N		 7174 if (rdev->raid_disk < 0)
7175 continue;
7176 diff = (rdev->new_data_offset - rdev->data_offset);
7177 if (first) {
7178 min_offset_diff = diff;
7179 first = 0;
7180 } else if (mddev->reshape_backwards &&
7181 diff < min_offset_diff)
7182 min_offset_diff = diff;
7183 else if (!mddev->reshape_backwards &&
7184 diff > min_offset_diff)
7185 min_offset_diff = diff;
7186 }
7187
230b55fa
N		 7188 if ((test_bit(MD_HAS_JOURNAL, &mddev->flags) || journal_dev) &&
7189 (mddev->bitmap_info.offset || mddev->bitmap_info.file)) {
7190 pr_notice("md/raid:%s: array cannot have both journal and bitmap\n",
7191 mdname(mddev));
7192 return -EINVAL;
7193 }
7194
91adb564
N		 7195 if (mddev->reshape_position != MaxSector) {
7196 /* Check that we can continue the reshape.
b5254dd5
N		 7197 * Difficulties arise if the stripe we would write to
7198 * next is at or after the stripe we would read from next.
7199 * For a reshape that changes the number of devices, this
7200 * is only possible for a very short time, and mdadm makes
7201 * sure that time appears to have past before assembling
7202 * the array. So we fail if that time hasn't passed.
7203 * For a reshape that keeps the number of devices the same
7204 * mdadm must be monitoring the reshape can keeping the
7205 * critical areas read-only and backed up. It will start
7206 * the array in read-only mode, so we check for that.
91adb564
N		 7207 */
7208 sector_t here_new, here_old;
7209 int old_disks;
18b00334 7210 int max_degraded = (mddev->level == 6 ? 2 : 1);
05256d98
N		 7211 int chunk_sectors;
7212 int new_data_disks;
91adb564 7213
713cf5a6 7214 if (journal_dev) {
cc6167b4
N		 7215 pr_warn("md/raid:%s: don't support reshape with journal - aborting.\n",
7216 mdname(mddev));
713cf5a6
SL		 7217 return -EINVAL;
7218 }
7219
88ce4930 7220 if (mddev->new_level != mddev->level) {
cc6167b4
N		 7221 pr_warn("md/raid:%s: unsupported reshape required - aborting.\n",
7222 mdname(mddev));
91adb564
N		 7223 return -EINVAL;
7224 }
91adb564
N		 7225 old_disks = mddev->raid_disks - mddev->delta_disks;
7226 /* reshape_position must be on a new-stripe boundary, and one
7227 * further up in new geometry must map after here in old
7228 * geometry.
05256d98
N		 7229 * If the chunk sizes are different, then as we perform reshape
7230 * in units of the largest of the two, reshape_position needs
7231 * be a multiple of the largest chunk size times new data disks.
91adb564
N		 7232 */
7233 here_new = mddev->reshape_position;
05256d98
N		 7234 chunk_sectors = max(mddev->chunk_sectors, mddev->new_chunk_sectors);
7235 new_data_disks = mddev->raid_disks - max_degraded;
7236 if (sector_div(here_new, chunk_sectors * new_data_disks)) {
cc6167b4
N		 7237 pr_warn("md/raid:%s: reshape_position not on a stripe boundary\n",
7238 mdname(mddev));
91adb564
N		 7239 return -EINVAL;
7240 }
05256d98 7241 reshape_offset = here_new * chunk_sectors;
91adb564
N		 7242 /* here_new is the stripe we will write to */
7243 here_old = mddev->reshape_position;
05256d98 7244 sector_div(here_old, chunk_sectors * (old_disks-max_degraded));
91adb564
N		 7245 /* here_old is the first stripe that we might need to read
7246 * from */
67ac6011
N		 7247 if (mddev->delta_disks == 0) {
7248 /* We cannot be sure it is safe to start an in-place
b5254dd5 7249 * reshape. It is only safe if user-space is monitoring
67ac6011
N		 7250 * and taking constant backups.
7251 * mdadm always starts a situation like this in
7252 * readonly mode so it can take control before
7253 * allowing any writes. So just check for that.
7254 */
b5254dd5
N		 7255 if (abs(min_offset_diff) >= mddev->chunk_sectors &&
7256 abs(min_offset_diff) >= mddev->new_chunk_sectors)
7257 /* not really in-place - so OK */;
7258 else if (mddev->ro == 0) {
cc6167b4
N		 7259 pr_warn("md/raid:%s: in-place reshape must be started in read-only mode - aborting\n",
7260 mdname(mddev));
67ac6011
N		 7261 return -EINVAL;
7262 }
2c810cdd 7263 } else if (mddev->reshape_backwards
05256d98
N		 7264 ? (here_new * chunk_sectors + min_offset_diff <=
7265 here_old * chunk_sectors)
7266 : (here_new * chunk_sectors >=
7267 here_old * chunk_sectors + (-min_offset_diff))) {
91adb564 7268 /* Reading from the same stripe as writing to - bad */
cc6167b4
N		 7269 pr_warn("md/raid:%s: reshape_position too early for auto-recovery - aborting.\n",
7270 mdname(mddev));
91adb564
N		 7271 return -EINVAL;
7272 }
cc6167b4 7273 pr_debug("md/raid:%s: reshape will continue\n", mdname(mddev));
91adb564
N		 7274 /* OK, we should be able to continue; */
7275 } else {
7276 BUG_ON(mddev->level != mddev->new_level);
7277 BUG_ON(mddev->layout != mddev->new_layout);
664e7c41 7278 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
91adb564 7279 BUG_ON(mddev->delta_disks != 0);
1da177e4 7280 }
91adb564 7281
3418d036
AP		 7282 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) &&
7283 test_bit(MD_HAS_PPL, &mddev->flags)) {
7284 pr_warn("md/raid:%s: using journal device and PPL not allowed - disabling PPL\n",
7285 mdname(mddev));
7286 clear_bit(MD_HAS_PPL, &mddev->flags);
ddc08823 7287 clear_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags);
3418d036
AP		 7288 }
7289
245f46c2
N		 7290 if (mddev->private == NULL)
7291 conf = setup_conf(mddev);
7292 else
7293 conf = mddev->private;
7294
91adb564
N		 7295 if (IS_ERR(conf))
7296 return PTR_ERR(conf);
7297
486b0f7b
SL		 7298 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
7299 if (!journal_dev) {
cc6167b4
N		 7300 pr_warn("md/raid:%s: journal disk is missing, force array readonly\n",
7301 mdname(mddev));
486b0f7b
SL		 7302 mddev->ro = 1;
7303 set_disk_ro(mddev->gendisk, 1);
7304 } else if (mddev->recovery_cp == MaxSector)
7305 set_bit(MD_JOURNAL_CLEAN, &mddev->flags);
7dde2ad3
SL		 7306 }
7307
b5254dd5 7308 conf->min_offset_diff = min_offset_diff;
91adb564
N		 7309 mddev->thread = conf->thread;
7310 conf->thread = NULL;
7311 mddev->private = conf;
7312
17045f52
N		 7313 for (i = 0; i < conf->raid_disks && conf->previous_raid_disks;
7314 i++) {
7315 rdev = conf->disks[i].rdev;
7316 if (!rdev && conf->disks[i].replacement) {
7317 /* The replacement is all we have yet */
7318 rdev = conf->disks[i].replacement;
7319 conf->disks[i].replacement = NULL;
7320 clear_bit(Replacement, &rdev->flags);
7321 conf->disks[i].rdev = rdev;
7322 }
7323 if (!rdev)
c148ffdc 7324 continue;
17045f52
N		 7325 if (conf->disks[i].replacement &&
7326 conf->reshape_progress != MaxSector) {
7327 /* replacements and reshape simply do not mix. */
cc6167b4 7328 pr_warn("md: cannot handle concurrent replacement and reshape.\n");
17045f52
N		 7329 goto abort;
7330 }
2f115882 7331 if (test_bit(In_sync, &rdev->flags)) {
91adb564 7332 working_disks++;
2f115882
N		 7333 continue;
7334 }
c148ffdc
N		 7335 /* This disc is not fully in-sync. However if it
7336 * just stored parity (beyond the recovery_offset),
7337 * when we don't need to be concerned about the
7338 * array being dirty.
7339 * When reshape goes 'backwards', we never have
7340 * partially completed devices, so we only need
7341 * to worry about reshape going forwards.
7342 */
7343 /* Hack because v0.91 doesn't store recovery_offset properly. */
7344 if (mddev->major_version == 0 &&
7345 mddev->minor_version > 90)
7346 rdev->recovery_offset = reshape_offset;
5026d7a9 7347
c148ffdc
N		 7348 if (rdev->recovery_offset < reshape_offset) {
7349 /* We need to check old and new layout */
7350 if (!only_parity(rdev->raid_disk,
7351 conf->algorithm,
7352 conf->raid_disks,
7353 conf->max_degraded))
7354 continue;
7355 }
7356 if (!only_parity(rdev->raid_disk,
7357 conf->prev_algo,
7358 conf->previous_raid_disks,
7359 conf->max_degraded))
7360 continue;
7361 dirty_parity_disks++;
7362 }
91adb564 7363
17045f52
N		 7364 /*
7365 * 0 for a fully functional array, 1 or 2 for a degraded array.
7366 */
2e38a37f 7367 mddev->degraded = raid5_calc_degraded(conf);
91adb564 7368
674806d6 7369 if (has_failed(conf)) {
cc6167b4 7370 pr_crit("md/raid:%s: not enough operational devices (%d/%d failed)\n",
02c2de8c 7371 mdname(mddev), mddev->degraded, conf->raid_disks);
1da177e4
LT		 7372 goto abort;
7373 }
7374
91adb564 7375 /* device size must be a multiple of chunk size */
9d8f0363 7376 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
91adb564
N		 7377 mddev->resync_max_sectors = mddev->dev_sectors;
7378
c148ffdc 7379 if (mddev->degraded > dirty_parity_disks &&
1da177e4 7380 mddev->recovery_cp != MaxSector) {
4536bf9b
AP		 7381 if (test_bit(MD_HAS_PPL, &mddev->flags))
7382 pr_crit("md/raid:%s: starting dirty degraded array with PPL.\n",
7383 mdname(mddev));
7384 else if (mddev->ok_start_degraded)
cc6167b4
N		 7385 pr_crit("md/raid:%s: starting dirty degraded array - data corruption possible.\n",
7386 mdname(mddev));
6ff8d8ec 7387 else {
cc6167b4
N		 7388 pr_crit("md/raid:%s: cannot start dirty degraded array.\n",
7389 mdname(mddev));
6ff8d8ec
N		 7390 goto abort;
7391 }
1da177e4
LT		 7392 }
7393
cc6167b4
N		 7394 pr_info("md/raid:%s: raid level %d active with %d out of %d devices, algorithm %d\n",
7395 mdname(mddev), conf->level,
7396 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
7397 mddev->new_layout);
1da177e4
LT		 7398
7399 print_raid5_conf(conf);
7400
fef9c61f 7401 if (conf->reshape_progress != MaxSector) {
fef9c61f 7402 conf->reshape_safe = conf->reshape_progress;
f6705578
N		 7403 atomic_set(&conf->reshape_stripes, 0);
7404 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7405 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7406 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7407 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7408 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 7409 "reshape");
e406f12d
AP		 7410 if (!mddev->sync_thread)
7411 goto abort;
f6705578
N		 7412 }
7413
1da177e4 7414 /* Ok, everything is just fine now */
a64c876f
N		 7415 if (mddev->to_remove == &raid5_attrs_group)
7416 mddev->to_remove = NULL;
00bcb4ac
N		 7417 else if (mddev->kobj.sd &&
7418 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
cc6167b4
N		 7419 pr_warn("raid5: failed to create sysfs attributes for %s\n",
7420 mdname(mddev));
4a5add49 7421 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
7a5febe9 7422
4a5add49 7423 if (mddev->queue) {
9f7c2220 7424 int chunk_size;
4a5add49
N		 7425 /* read-ahead size must cover two whole stripes, which
7426 * is 2 * (datadisks) * chunksize where 'n' is the
7427 * number of raid devices
7428 */
7429 int data_disks = conf->previous_raid_disks - conf->max_degraded;
7430 int stripe = data_disks *
7431 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
dc3b17cc
JK		 7432 if (mddev->queue->backing_dev_info->ra_pages < 2 * stripe)
7433 mddev->queue->backing_dev_info->ra_pages = 2 * stripe;
91adb564 7434
9f7c2220
N		 7435 chunk_size = mddev->chunk_sectors << 9;
7436 blk_queue_io_min(mddev->queue, chunk_size);
7437 blk_queue_io_opt(mddev->queue, chunk_size *
7438 (conf->raid_disks - conf->max_degraded));
c78afc62 7439 mddev->queue->limits.raid_partial_stripes_expensive = 1;
620125f2
SL		 7440 /*
7441 * We can only discard a whole stripe. It doesn't make sense to
7442 * discard data disk but write parity disk
7443 */
7444 stripe = stripe * PAGE_SIZE;
4ac6875e
N		 7445 /* Round up to power of 2, as discard handling
7446 * currently assumes that */
7447 while ((stripe-1) & stripe)
7448 stripe = (stripe | (stripe-1)) + 1;
620125f2
SL		 7449 mddev->queue->limits.discard_alignment = stripe;
7450 mddev->queue->limits.discard_granularity = stripe;
e8d7c332 7451
5026d7a9 7452 blk_queue_max_write_same_sectors(mddev->queue, 0);
3deff1a7 7453 blk_queue_max_write_zeroes_sectors(mddev->queue, 0);
5026d7a9 7454
05616be5 7455 rdev_for_each(rdev, mddev) {
9f7c2220
N		 7456 disk_stack_limits(mddev->gendisk, rdev->bdev,
7457 rdev->data_offset << 9);
05616be5
N		 7458 disk_stack_limits(mddev->gendisk, rdev->bdev,
7459 rdev->new_data_offset << 9);
7460 }
620125f2 7461
48920ff2
CH		 7462 /*
7463 * zeroing is required, otherwise data
7464 * could be lost. Consider a scenario: discard a stripe
7465 * (the stripe could be inconsistent if
7466 * discard_zeroes_data is 0); write one disk of the
7467 * stripe (the stripe could be inconsistent again
7468 * depending on which disks are used to calculate
7469 * parity); the disk is broken; The stripe data of this
7470 * disk is lost.
7471 *
7472 * We only allow DISCARD if the sysadmin has confirmed that
7473 * only safe devices are in use by setting a module parameter.
7474 * A better idea might be to turn DISCARD into WRITE_ZEROES
7475 * requests, as that is required to be safe.
7476 */
7477 if (devices_handle_discard_safely &&
e7597e69
JS		 7478 mddev->queue->limits.max_discard_sectors >= (stripe >> 9) &&
7479 mddev->queue->limits.discard_granularity >= stripe)
8b904b5b 7480 blk_queue_flag_set(QUEUE_FLAG_DISCARD,
620125f2
SL		 7481 mddev->queue);
7482 else
8b904b5b 7483 blk_queue_flag_clear(QUEUE_FLAG_DISCARD,
620125f2 7484 mddev->queue);
1dffdddd
SL		 7485
7486 blk_queue_max_hw_sectors(mddev->queue, UINT_MAX);
9f7c2220 7487 }
23032a0e 7488
845b9e22 7489 if (log_init(conf, journal_dev, raid5_has_ppl(conf)))
ff875738 7490 goto abort;
5c7e81c3 7491
1da177e4
LT		 7492 return 0;
7493abort:
01f96c0a 7494 md_unregister_thread(&mddev->thread);
e4f869d9
N		 7495 print_raid5_conf(conf);
7496 free_conf(conf);
1da177e4 7497 mddev->private = NULL;
cc6167b4 7498 pr_warn("md/raid:%s: failed to run raid set.\n", mdname(mddev));
1da177e4
LT		 7499 return -EIO;
7500}
7501
afa0f557 7502static void raid5_free(struct mddev *mddev, void *priv)
1da177e4 7503{
afa0f557 7504 struct r5conf *conf = priv;
1da177e4 7505
95fc17aa 7506 free_conf(conf);
a64c876f 7507 mddev->to_remove = &raid5_attrs_group;
1da177e4
LT		 7508}
7509
849674e4 7510static void raid5_status(struct seq_file *seq, struct mddev *mddev)
1da177e4 7511{
d1688a6d 7512 struct r5conf *conf = mddev->private;
1da177e4
LT		 7513 int i;
7514
9d8f0363 7515 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
3cb5edf4 7516 conf->chunk_sectors / 2, mddev->layout);
02c2de8c 7517 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
5fd13351
N		 7518 rcu_read_lock();
7519 for (i = 0; i < conf->raid_disks; i++) {
7520 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
7521 seq_printf (seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
7522 }
7523 rcu_read_unlock();
1da177e4 7524 seq_printf (seq, "]");
1da177e4
LT		 7525}
7526
d1688a6d 7527static void print_raid5_conf (struct r5conf *conf)
1da177e4
LT		 7528{
7529 int i;
7530 struct disk_info *tmp;
7531
cc6167b4 7532 pr_debug("RAID conf printout:\n");
1da177e4 7533 if (!conf) {
cc6167b4 7534 pr_debug("(conf==NULL)\n");
1da177e4
LT		 7535 return;
7536 }
cc6167b4 7537 pr_debug(" --- level:%d rd:%d wd:%d\n", conf->level,
0c55e022
N		 7538 conf->raid_disks,
7539 conf->raid_disks - conf->mddev->degraded);
1da177e4
LT		 7540
7541 for (i = 0; i < conf->raid_disks; i++) {
7542 char b[BDEVNAME_SIZE];
7543 tmp = conf->disks + i;
7544 if (tmp->rdev)
cc6167b4 7545 pr_debug(" disk %d, o:%d, dev:%s\n",
0c55e022
N		 7546 i, !test_bit(Faulty, &tmp->rdev->flags),
7547 bdevname(tmp->rdev->bdev, b));
1da177e4
LT		 7548 }
7549}
7550
fd01b88c 7551static int raid5_spare_active(struct mddev *mddev)
1da177e4
LT		 7552{
7553 int i;
d1688a6d 7554 struct r5conf *conf = mddev->private;
1da177e4 7555 struct disk_info *tmp;
6b965620
N		 7556 int count = 0;
7557 unsigned long flags;
1da177e4
LT		 7558
7559 for (i = 0; i < conf->raid_disks; i++) {
7560 tmp = conf->disks + i;
dd054fce
N		 7561 if (tmp->replacement
7562 && tmp->replacement->recovery_offset == MaxSector
7563 && !test_bit(Faulty, &tmp->replacement->flags)
7564 && !test_and_set_bit(In_sync, &tmp->replacement->flags)) {
7565 /* Replacement has just become active. */
7566 if (!tmp->rdev
7567 || !test_and_clear_bit(In_sync, &tmp->rdev->flags))
7568 count++;
7569 if (tmp->rdev) {
7570 /* Replaced device not technically faulty,
7571 * but we need to be sure it gets removed
7572 * and never re-added.
7573 */
7574 set_bit(Faulty, &tmp->rdev->flags);
7575 sysfs_notify_dirent_safe(
7576 tmp->rdev->sysfs_state);
7577 }
7578 sysfs_notify_dirent_safe(tmp->replacement->sysfs_state);
7579 } else if (tmp->rdev
70fffd0b 7580 && tmp->rdev->recovery_offset == MaxSector
b2d444d7 7581 && !test_bit(Faulty, &tmp->rdev->flags)
c04be0aa 7582 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
6b965620 7583 count++;
43c73ca4 7584 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
1da177e4
LT		 7585 }
7586 }
6b965620 7587 spin_lock_irqsave(&conf->device_lock, flags);
2e38a37f 7588 mddev->degraded = raid5_calc_degraded(conf);
6b965620 7589 spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4 7590 print_raid5_conf(conf);
6b965620 7591 return count;
1da177e4
LT		 7592}
7593
b8321b68 7594static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7595{
d1688a6d 7596 struct r5conf *conf = mddev->private;
1da177e4 7597 int err = 0;
b8321b68 7598 int number = rdev->raid_disk;
657e3e4d 7599 struct md_rdev **rdevp;
1da177e4
LT		 7600 struct disk_info *p = conf->disks + number;
7601
7602 print_raid5_conf(conf);
f6b6ec5c 7603 if (test_bit(Journal, &rdev->flags) && conf->log) {
c2bb6242 7604 /*
f6b6ec5c
SL		 7605 * we can't wait pending write here, as this is called in
7606 * raid5d, wait will deadlock.
84dd97a6
N		 7607 * neilb: there is no locking about new writes here,
7608 * so this cannot be safe.
c2bb6242 7609 */
70d466f7
SL		 7610 if (atomic_read(&conf->active_stripes) ||
7611 atomic_read(&conf->r5c_cached_full_stripes) ||
7612 atomic_read(&conf->r5c_cached_partial_stripes)) {
f6b6ec5c 7613 return -EBUSY;
84dd97a6 7614 }
ff875738 7615 log_exit(conf);
f6b6ec5c 7616 return 0;
c2bb6242 7617 }
657e3e4d
N		 7618 if (rdev == p->rdev)
7619 rdevp = &p->rdev;
7620 else if (rdev == p->replacement)
7621 rdevp = &p->replacement;
7622 else
7623 return 0;
7624
7625 if (number >= conf->raid_disks &&
7626 conf->reshape_progress == MaxSector)
7627 clear_bit(In_sync, &rdev->flags);
7628
7629 if (test_bit(In_sync, &rdev->flags) ||
7630 atomic_read(&rdev->nr_pending)) {
7631 err = -EBUSY;
7632 goto abort;
7633 }
7634 /* Only remove non-faulty devices if recovery
7635 * isn't possible.
7636 */
7637 if (!test_bit(Faulty, &rdev->flags) &&
7638 mddev->recovery_disabled != conf->recovery_disabled &&
7639 !has_failed(conf) &&
dd054fce 7640 (!p->replacement || p->replacement == rdev) &&
657e3e4d
N		 7641 number < conf->raid_disks) {
7642 err = -EBUSY;
7643 goto abort;
7644 }
7645 *rdevp = NULL;
d787be40
N		 7646 if (!test_bit(RemoveSynchronized, &rdev->flags)) {
7647 synchronize_rcu();
7648 if (atomic_read(&rdev->nr_pending)) {
7649 /* lost the race, try later */
7650 err = -EBUSY;
7651 *rdevp = rdev;
7652 }
7653 }
6358c239
AP		 7654 if (!err) {
7655 err = log_modify(conf, rdev, false);
7656 if (err)
7657 goto abort;
7658 }
d787be40 7659 if (p->replacement) {
dd054fce
N		 7660 /* We must have just cleared 'rdev' */
7661 p->rdev = p->replacement;
7662 clear_bit(Replacement, &p->replacement->flags);
7663 smp_mb(); /* Make sure other CPUs may see both as identical
7664 * but will never see neither - if they are careful
7665 */
7666 p->replacement = NULL;
6358c239
AP		 7667
7668 if (!err)
7669 err = log_modify(conf, p->rdev, true);
e5bc9c3c
GJ		 7670 }
7671
7672 clear_bit(WantReplacement, &rdev->flags);
1da177e4
LT		 7673abort:
7674
7675 print_raid5_conf(conf);
7676 return err;
7677}
7678
fd01b88c 7679static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7680{
d1688a6d 7681 struct r5conf *conf = mddev->private;
d9771f5e 7682 int ret, err = -EEXIST;
1da177e4
LT		 7683 int disk;
7684 struct disk_info *p;
6c2fce2e
NB		 7685 int first = 0;
7686 int last = conf->raid_disks - 1;
1da177e4 7687
f6b6ec5c 7688 if (test_bit(Journal, &rdev->flags)) {
f6b6ec5c
SL		 7689 if (conf->log)
7690 return -EBUSY;
7691
7692 rdev->raid_disk = 0;
7693 /*
7694 * The array is in readonly mode if journal is missing, so no
7695 * write requests running. We should be safe
7696 */
d9771f5e
XN		 7697 ret = log_init(conf, rdev, false);
7698 if (ret)
7699 return ret;
7700
7701 ret = r5l_start(conf->log);
7702 if (ret)
7703 return ret;
7704
f6b6ec5c
SL		 7705 return 0;
7706 }
7f0da59b
N		 7707 if (mddev->recovery_disabled == conf->recovery_disabled)
7708 return -EBUSY;
7709
dc10c643 7710 if (rdev->saved_raid_disk < 0 && has_failed(conf))
1da177e4 7711 /* no point adding a device */
199050ea 7712 return -EINVAL;
1da177e4 7713
6c2fce2e
NB		 7714 if (rdev->raid_disk >= 0)
7715 first = last = rdev->raid_disk;
1da177e4
LT		 7716
7717 /*
16a53ecc
N		 7718 * find the disk ... but prefer rdev->saved_raid_disk
7719 * if possible.
1da177e4 7720 */
16a53ecc 7721 if (rdev->saved_raid_disk >= 0 &&
6c2fce2e 7722 rdev->saved_raid_disk >= first &&
16a53ecc 7723 conf->disks[rdev->saved_raid_disk].rdev == NULL)
5cfb22a1
N		 7724 first = rdev->saved_raid_disk;
7725
7726 for (disk = first; disk <= last; disk++) {
7bfec5f3
N		 7727 p = conf->disks + disk;
7728 if (p->rdev == NULL) {
b2d444d7 7729 clear_bit(In_sync, &rdev->flags);
1da177e4 7730 rdev->raid_disk = disk;
72626685
N		 7731 if (rdev->saved_raid_disk != disk)
7732 conf->fullsync = 1;
d6065f7b 7733 rcu_assign_pointer(p->rdev, rdev);
6358c239
AP		 7734
7735 err = log_modify(conf, rdev, true);
7736
5cfb22a1 7737 goto out;
1da177e4 7738 }
5cfb22a1
N		 7739 }
7740 for (disk = first; disk <= last; disk++) {
7741 p = conf->disks + disk;
7bfec5f3
N		 7742 if (test_bit(WantReplacement, &p->rdev->flags) &&
7743 p->replacement == NULL) {
7744 clear_bit(In_sync, &rdev->flags);
7745 set_bit(Replacement, &rdev->flags);
7746 rdev->raid_disk = disk;
7747 err = 0;
7748 conf->fullsync = 1;
7749 rcu_assign_pointer(p->replacement, rdev);
7750 break;
7751 }
7752 }
5cfb22a1 7753out:
1da177e4 7754 print_raid5_conf(conf);
199050ea 7755 return err;
1da177e4
LT		 7756}
7757
fd01b88c 7758static int raid5_resize(struct mddev *mddev, sector_t sectors)
1da177e4
LT		 7759{
7760 /* no resync is happening, and there is enough space
7761 * on all devices, so we can resize.
7762 * We need to make sure resync covers any new space.
7763 * If the array is shrinking we should possibly wait until
7764 * any io in the removed space completes, but it hardly seems
7765 * worth it.
7766 */
a4a6125a 7767 sector_t newsize;
3cb5edf4
N		 7768 struct r5conf *conf = mddev->private;
7769
e254de6b 7770 if (raid5_has_log(conf) || raid5_has_ppl(conf))
713cf5a6 7771 return -EINVAL;
3cb5edf4 7772 sectors &= ~((sector_t)conf->chunk_sectors - 1);
a4a6125a
N		 7773 newsize = raid5_size(mddev, sectors, mddev->raid_disks);
7774 if (mddev->external_size &&
7775 mddev->array_sectors > newsize)
b522adcd 7776 return -EINVAL;
a4a6125a 7777 if (mddev->bitmap) {
e64e4018 7778 int ret = md_bitmap_resize(mddev->bitmap, sectors, 0, 0);
a4a6125a
N		 7779 if (ret)
7780 return ret;
7781 }
7782 md_set_array_sectors(mddev, newsize);
b098636c
N		 7783 if (sectors > mddev->dev_sectors &&
7784 mddev->recovery_cp > mddev->dev_sectors) {
58c0fed4 7785 mddev->recovery_cp = mddev->dev_sectors;
1da177e4
LT		 7786 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7787 }
58c0fed4 7788 mddev->dev_sectors = sectors;
4b5c7ae8 7789 mddev->resync_max_sectors = sectors;
1da177e4
LT		 7790 return 0;
7791}
7792
fd01b88c 7793static int check_stripe_cache(struct mddev *mddev)
01ee22b4
N		 7794{
7795 /* Can only proceed if there are plenty of stripe_heads.
7796 * We need a minimum of one full stripe,, and for sensible progress
7797 * it is best to have about 4 times that.
7798 * If we require 4 times, then the default 256 4K stripe_heads will
7799 * allow for chunk sizes up to 256K, which is probably OK.
7800 * If the chunk size is greater, user-space should request more
7801 * stripe_heads first.
7802 */
d1688a6d 7803 struct r5conf *conf = mddev->private;
01ee22b4 7804 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
edbe83ab 7805 > conf->min_nr_stripes ||
01ee22b4 7806 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
edbe83ab 7807 > conf->min_nr_stripes) {
cc6167b4
N		 7808 pr_warn("md/raid:%s: reshape: not enough stripes. Needed %lu\n",
7809 mdname(mddev),
7810 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
7811 / STRIPE_SIZE)*4);
01ee22b4
N		 7812 return 0;
7813 }
7814 return 1;
7815}
7816
fd01b88c 7817static int check_reshape(struct mddev *mddev)
29269553 7818{
d1688a6d 7819 struct r5conf *conf = mddev->private;
29269553 7820
e254de6b 7821 if (raid5_has_log(conf) || raid5_has_ppl(conf))
713cf5a6 7822 return -EINVAL;
88ce4930
N		 7823 if (mddev->delta_disks == 0 &&
7824 mddev->new_layout == mddev->layout &&
664e7c41 7825 mddev->new_chunk_sectors == mddev->chunk_sectors)
50ac168a 7826 return 0; /* nothing to do */
674806d6 7827 if (has_failed(conf))
ec32a2bd 7828 return -EINVAL;
fdcfbbb6 7829 if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) {
ec32a2bd
N		 7830 /* We might be able to shrink, but the devices must
7831 * be made bigger first.
7832 * For raid6, 4 is the minimum size.
7833 * Otherwise 2 is the minimum
7834 */
7835 int min = 2;
7836 if (mddev->level == 6)
7837 min = 4;
7838 if (mddev->raid_disks + mddev->delta_disks < min)
7839 return -EINVAL;
7840 }
29269553 7841
01ee22b4 7842 if (!check_stripe_cache(mddev))
29269553 7843 return -ENOSPC;
29269553 7844
738a2738
N		 7845 if (mddev->new_chunk_sectors > mddev->chunk_sectors ||
7846 mddev->delta_disks > 0)
7847 if (resize_chunks(conf,
7848 conf->previous_raid_disks
7849 + max(0, mddev->delta_disks),
7850 max(mddev->new_chunk_sectors,
7851 mddev->chunk_sectors)
7852 ) < 0)
7853 return -ENOMEM;
845b9e22
AP		 7854
7855 if (conf->previous_raid_disks + mddev->delta_disks <= conf->pool_size)
7856 return 0; /* never bother to shrink */
e56108d6
N		 7857 return resize_stripes(conf, (conf->previous_raid_disks
7858 + mddev->delta_disks));
63c70c4f
N		 7859}
7860
fd01b88c 7861static int raid5_start_reshape(struct mddev *mddev)
63c70c4f 7862{
d1688a6d 7863 struct r5conf *conf = mddev->private;
3cb03002 7864 struct md_rdev *rdev;
63c70c4f 7865 int spares = 0;
c04be0aa 7866 unsigned long flags;
63c70c4f 7867
f416885e 7868 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
63c70c4f
N		 7869 return -EBUSY;
7870
01ee22b4
N		 7871 if (!check_stripe_cache(mddev))
7872 return -ENOSPC;
7873
30b67645
N		 7874 if (has_failed(conf))
7875 return -EINVAL;
7876
c6563a8c 7877 rdev_for_each(rdev, mddev) {
469518a3
N		 7878 if (!test_bit(In_sync, &rdev->flags)
7879 && !test_bit(Faulty, &rdev->flags))
29269553 7880 spares++;
c6563a8c 7881 }
63c70c4f 7882
f416885e 7883 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
29269553
N		 7884 /* Not enough devices even to make a degraded array
7885 * of that size
7886 */
7887 return -EINVAL;
7888
ec32a2bd
N		 7889 /* Refuse to reduce size of the array. Any reductions in
7890 * array size must be through explicit setting of array_size
7891 * attribute.
7892 */
7893 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
7894 < mddev->array_sectors) {
cc6167b4
N		 7895 pr_warn("md/raid:%s: array size must be reduced before number of disks\n",
7896 mdname(mddev));
ec32a2bd
N		 7897 return -EINVAL;
7898 }
7899
f6705578 7900 atomic_set(&conf->reshape_stripes, 0);
29269553 7901 spin_lock_irq(&conf->device_lock);
c46501b2 7902 write_seqcount_begin(&conf->gen_lock);
29269553 7903 conf->previous_raid_disks = conf->raid_disks;
63c70c4f 7904 conf->raid_disks += mddev->delta_disks;
09c9e5fa
AN		 7905 conf->prev_chunk_sectors = conf->chunk_sectors;
7906 conf->chunk_sectors = mddev->new_chunk_sectors;
88ce4930
N		 7907 conf->prev_algo = conf->algorithm;
7908 conf->algorithm = mddev->new_layout;
05616be5
N		 7909 conf->generation++;
7910 /* Code that selects data_offset needs to see the generation update
7911 * if reshape_progress has been set - so a memory barrier needed.
7912 */
7913 smp_mb();
2c810cdd 7914 if (mddev->reshape_backwards)
fef9c61f
N		 7915 conf->reshape_progress = raid5_size(mddev, 0, 0);
7916 else
7917 conf->reshape_progress = 0;
7918 conf->reshape_safe = conf->reshape_progress;
c46501b2 7919 write_seqcount_end(&conf->gen_lock);
29269553
N		 7920 spin_unlock_irq(&conf->device_lock);
7921
4d77e3ba
N		 7922 /* Now make sure any requests that proceeded on the assumption
7923 * the reshape wasn't running - like Discard or Read - have
7924 * completed.
7925 */
7926 mddev_suspend(mddev);
7927 mddev_resume(mddev);
7928
29269553
N		 7929 /* Add some new drives, as many as will fit.
7930 * We know there are enough to make the newly sized array work.
3424bf6a
N		 7931 * Don't add devices if we are reducing the number of
7932 * devices in the array. This is because it is not possible
7933 * to correctly record the "partially reconstructed" state of
7934 * such devices during the reshape and confusion could result.
29269553 7935 */
87a8dec9 7936 if (mddev->delta_disks >= 0) {
dafb20fa 7937 rdev_for_each(rdev, mddev)
87a8dec9
N		 7938 if (rdev->raid_disk < 0 &&
7939 !test_bit(Faulty, &rdev->flags)) {
7940 if (raid5_add_disk(mddev, rdev) == 0) {
87a8dec9 7941 if (rdev->raid_disk
9d4c7d87 7942 >= conf->previous_raid_disks)
87a8dec9 7943 set_bit(In_sync, &rdev->flags);
9d4c7d87 7944 else
87a8dec9 7945 rdev->recovery_offset = 0;
36fad858
NK		 7946
7947 if (sysfs_link_rdev(mddev, rdev))
87a8dec9 7948 /* Failure here is OK */;
50da0840 7949 }
87a8dec9
N		 7950 } else if (rdev->raid_disk >= conf->previous_raid_disks
7951 && !test_bit(Faulty, &rdev->flags)) {
7952 /* This is a spare that was manually added */
7953 set_bit(In_sync, &rdev->flags);
87a8dec9 7954 }
29269553 7955
87a8dec9
N		 7956 /* When a reshape changes the number of devices,
7957 * ->degraded is measured against the larger of the
7958 * pre and post number of devices.
7959 */
ec32a2bd 7960 spin_lock_irqsave(&conf->device_lock, flags);
2e38a37f 7961 mddev->degraded = raid5_calc_degraded(conf);
ec32a2bd
N		 7962 spin_unlock_irqrestore(&conf->device_lock, flags);
7963 }
63c70c4f 7964 mddev->raid_disks = conf->raid_disks;
e516402c 7965 mddev->reshape_position = conf->reshape_progress;
2953079c 7966 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
f6705578 7967
29269553
N		 7968 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7969 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
ea358cd0 7970 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
29269553
N		 7971 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7972 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7973 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 7974 "reshape");
29269553
N		 7975 if (!mddev->sync_thread) {
7976 mddev->recovery = 0;
7977 spin_lock_irq(&conf->device_lock);
ba8805b9 7978 write_seqcount_begin(&conf->gen_lock);
29269553 7979 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
ba8805b9
N		 7980 mddev->new_chunk_sectors =
7981 conf->chunk_sectors = conf->prev_chunk_sectors;
7982 mddev->new_layout = conf->algorithm = conf->prev_algo;
05616be5
N		 7983 rdev_for_each(rdev, mddev)
7984 rdev->new_data_offset = rdev->data_offset;
7985 smp_wmb();
ba8805b9 7986 conf->generation --;
fef9c61f 7987 conf->reshape_progress = MaxSector;
1e3fa9bd 7988 mddev->reshape_position = MaxSector;
ba8805b9 7989 write_seqcount_end(&conf->gen_lock);
29269553
N		 7990 spin_unlock_irq(&conf->device_lock);
7991 return -EAGAIN;
7992 }
c8f517c4 7993 conf->reshape_checkpoint = jiffies;
29269553
N		 7994 md_wakeup_thread(mddev->sync_thread);
7995 md_new_event(mddev);
7996 return 0;
7997}
29269553 7998
ec32a2bd
N		 7999/* This is called from the reshape thread and should make any
8000 * changes needed in 'conf'
8001 */
d1688a6d 8002static void end_reshape(struct r5conf *conf)
29269553 8003{
29269553 8004
f6705578 8005 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
db0505d3 8006 struct md_rdev *rdev;
f6705578 8007
f6705578 8008 spin_lock_irq(&conf->device_lock);
cea9c228 8009 conf->previous_raid_disks = conf->raid_disks;
b5d27718 8010 md_finish_reshape(conf->mddev);
05616be5 8011 smp_wmb();
fef9c61f 8012 conf->reshape_progress = MaxSector;
6cbd8148 8013 conf->mddev->reshape_position = MaxSector;
db0505d3
N		 8014 rdev_for_each(rdev, conf->mddev)
8015 if (rdev->raid_disk >= 0 &&
8016 !test_bit(Journal, &rdev->flags) &&
8017 !test_bit(In_sync, &rdev->flags))
8018 rdev->recovery_offset = MaxSector;
f6705578 8019 spin_unlock_irq(&conf->device_lock);
b0f9ec04 8020 wake_up(&conf->wait_for_overlap);
16a53ecc
N		 8021
8022 /* read-ahead size must cover two whole stripes, which is
8023 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
8024 */
4a5add49 8025 if (conf->mddev->queue) {
cea9c228 8026 int data_disks = conf->raid_disks - conf->max_degraded;
09c9e5fa 8027 int stripe = data_disks * ((conf->chunk_sectors << 9)
cea9c228 8028 / PAGE_SIZE);
dc3b17cc
JK		 8029 if (conf->mddev->queue->backing_dev_info->ra_pages < 2 * stripe)
8030 conf->mddev->queue->backing_dev_info->ra_pages = 2 * stripe;
16a53ecc 8031 }
29269553 8032 }
29269553
N		 8033}
8034
ec32a2bd
N		 8035/* This is called from the raid5d thread with mddev_lock held.
8036 * It makes config changes to the device.
8037 */
fd01b88c 8038static void raid5_finish_reshape(struct mddev *mddev)
cea9c228 8039{
d1688a6d 8040 struct r5conf *conf = mddev->private;
cea9c228
N		 8041
8042 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8043
8876391e 8044 if (mddev->delta_disks <= 0) {
ec32a2bd 8045 int d;
908f4fbd 8046 spin_lock_irq(&conf->device_lock);
2e38a37f 8047 mddev->degraded = raid5_calc_degraded(conf);
908f4fbd 8048 spin_unlock_irq(&conf->device_lock);
ec32a2bd
N		 8049 for (d = conf->raid_disks ;
8050 d < conf->raid_disks - mddev->delta_disks;
1a67dde0 8051 d++) {
3cb03002 8052 struct md_rdev *rdev = conf->disks[d].rdev;
da7613b8
N		 8053 if (rdev)
8054 clear_bit(In_sync, &rdev->flags);
8055 rdev = conf->disks[d].replacement;
8056 if (rdev)
8057 clear_bit(In_sync, &rdev->flags);
1a67dde0 8058 }
cea9c228 8059 }
88ce4930 8060 mddev->layout = conf->algorithm;
09c9e5fa 8061 mddev->chunk_sectors = conf->chunk_sectors;
ec32a2bd
N		 8062 mddev->reshape_position = MaxSector;
8063 mddev->delta_disks = 0;
2c810cdd 8064 mddev->reshape_backwards = 0;
cea9c228
N		 8065 }
8066}
8067
b03e0ccb 8068static void raid5_quiesce(struct mddev *mddev, int quiesce)
72626685 8069{
d1688a6d 8070 struct r5conf *conf = mddev->private;
72626685 8071
b03e0ccb
N		 8072 if (quiesce) {
8073 /* stop all writes */
566c09c5 8074 lock_all_device_hash_locks_irq(conf);
64bd660b
N		 8075 /* '2' tells resync/reshape to pause so that all
8076 * active stripes can drain
8077 */
a39f7afd 8078 r5c_flush_cache(conf, INT_MAX);
64bd660b 8079 conf->quiesce = 2;
b1b46486 8080 wait_event_cmd(conf->wait_for_quiescent,
46031f9a
RBJ		 8081 atomic_read(&conf->active_stripes) == 0 &&
8082 atomic_read(&conf->active_aligned_reads) == 0,
566c09c5
SL		 8083 unlock_all_device_hash_locks_irq(conf),
8084 lock_all_device_hash_locks_irq(conf));
64bd660b 8085 conf->quiesce = 1;
566c09c5 8086 unlock_all_device_hash_locks_irq(conf);
64bd660b
N		 8087 /* allow reshape to continue */
8088 wake_up(&conf->wait_for_overlap);
b03e0ccb
N		 8089 } else {
8090 /* re-enable writes */
566c09c5 8091 lock_all_device_hash_locks_irq(conf);
72626685 8092 conf->quiesce = 0;
b1b46486 8093 wake_up(&conf->wait_for_quiescent);
e464eafd 8094 wake_up(&conf->wait_for_overlap);
566c09c5 8095 unlock_all_device_hash_locks_irq(conf);
72626685 8096 }
1532d9e8 8097 log_quiesce(conf, quiesce);
72626685 8098}
b15c2e57 8099
fd01b88c 8100static void *raid45_takeover_raid0(struct mddev *mddev, int level)
54071b38 8101{
e373ab10 8102 struct r0conf *raid0_conf = mddev->private;
d76c8420 8103 sector_t sectors;
54071b38 8104
f1b29bca 8105 /* for raid0 takeover only one zone is supported */
e373ab10 8106 if (raid0_conf->nr_strip_zones > 1) {
cc6167b4
N		 8107 pr_warn("md/raid:%s: cannot takeover raid0 with more than one zone.\n",
8108 mdname(mddev));
f1b29bca
DW		 8109 return ERR_PTR(-EINVAL);
8110 }
8111
e373ab10
N		 8112 sectors = raid0_conf->strip_zone[0].zone_end;
8113 sector_div(sectors, raid0_conf->strip_zone[0].nb_dev);
3b71bd93 8114 mddev->dev_sectors = sectors;
f1b29bca 8115 mddev->new_level = level;
54071b38
TM		 8116 mddev->new_layout = ALGORITHM_PARITY_N;
8117 mddev->new_chunk_sectors = mddev->chunk_sectors;
8118 mddev->raid_disks += 1;
8119 mddev->delta_disks = 1;
8120 /* make sure it will be not marked as dirty */
8121 mddev->recovery_cp = MaxSector;
8122
8123 return setup_conf(mddev);
8124}
8125
fd01b88c 8126static void *raid5_takeover_raid1(struct mddev *mddev)
d562b0c4
N		 8127{
8128 int chunksect;
6995f0b2 8129 void *ret;
d562b0c4
N		 8130
8131 if (mddev->raid_disks != 2 ||
8132 mddev->degraded > 1)
8133 return ERR_PTR(-EINVAL);
8134
8135 /* Should check if there are write-behind devices? */
8136
8137 chunksect = 64*2; /* 64K by default */
8138
8139 /* The array must be an exact multiple of chunksize */
8140 while (chunksect && (mddev->array_sectors & (chunksect-1)))
8141 chunksect >>= 1;
8142
8143 if ((chunksect<<9) < STRIPE_SIZE)
8144 /* array size does not allow a suitable chunk size */
8145 return ERR_PTR(-EINVAL);
8146
8147 mddev->new_level = 5;
8148 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
664e7c41 8149 mddev->new_chunk_sectors = chunksect;
d562b0c4 8150
6995f0b2 8151 ret = setup_conf(mddev);
32cd7cbb 8152 if (!IS_ERR(ret))
394ed8e4
SL		 8153 mddev_clear_unsupported_flags(mddev,
8154 UNSUPPORTED_MDDEV_FLAGS);
6995f0b2 8155 return ret;
d562b0c4
N		 8156}
8157
fd01b88c 8158static void *raid5_takeover_raid6(struct mddev *mddev)
fc9739c6
N		 8159{
8160 int new_layout;
8161
8162 switch (mddev->layout) {
8163 case ALGORITHM_LEFT_ASYMMETRIC_6:
8164 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
8165 break;
8166 case ALGORITHM_RIGHT_ASYMMETRIC_6:
8167 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
8168 break;
8169 case ALGORITHM_LEFT_SYMMETRIC_6:
8170 new_layout = ALGORITHM_LEFT_SYMMETRIC;
8171 break;
8172 case ALGORITHM_RIGHT_SYMMETRIC_6:
8173 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
8174 break;
8175 case ALGORITHM_PARITY_0_6:
8176 new_layout = ALGORITHM_PARITY_0;
8177 break;
8178 case ALGORITHM_PARITY_N:
8179 new_layout = ALGORITHM_PARITY_N;
8180 break;
8181 default:
8182 return ERR_PTR(-EINVAL);
8183 }
8184 mddev->new_level = 5;
8185 mddev->new_layout = new_layout;
8186 mddev->delta_disks = -1;
8187 mddev->raid_disks -= 1;
8188 return setup_conf(mddev);
8189}
8190
fd01b88c 8191static int raid5_check_reshape(struct mddev *mddev)
b3546035 8192{
88ce4930
N		 8193 /* For a 2-drive array, the layout and chunk size can be changed
8194 * immediately as not restriping is needed.
8195 * For larger arrays we record the new value - after validation
8196 * to be used by a reshape pass.
b3546035 8197 */
d1688a6d 8198 struct r5conf *conf = mddev->private;
597a711b 8199 int new_chunk = mddev->new_chunk_sectors;
b3546035 8200
597a711b 8201 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
b3546035
N		 8202 return -EINVAL;
8203 if (new_chunk > 0) {
0ba459d2 8204 if (!is_power_of_2(new_chunk))
b3546035 8205 return -EINVAL;
597a711b 8206 if (new_chunk < (PAGE_SIZE>>9))
b3546035 8207 return -EINVAL;
597a711b 8208 if (mddev->array_sectors & (new_chunk-1))
b3546035
N		 8209 /* not factor of array size */
8210 return -EINVAL;
8211 }
8212
8213 /* They look valid */
8214
88ce4930 8215 if (mddev->raid_disks == 2) {
597a711b
N		 8216 /* can make the change immediately */
8217 if (mddev->new_layout >= 0) {
8218 conf->algorithm = mddev->new_layout;
8219 mddev->layout = mddev->new_layout;
88ce4930
N		 8220 }
8221 if (new_chunk > 0) {
597a711b
N		 8222 conf->chunk_sectors = new_chunk ;
8223 mddev->chunk_sectors = new_chunk;
88ce4930 8224 }
2953079c 8225 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
88ce4930 8226 md_wakeup_thread(mddev->thread);
b3546035 8227 }
50ac168a 8228 return check_reshape(mddev);
88ce4930
N		 8229}
8230
fd01b88c 8231static int raid6_check_reshape(struct mddev *mddev)
88ce4930 8232{
597a711b 8233 int new_chunk = mddev->new_chunk_sectors;
50ac168a 8234
597a711b 8235 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
88ce4930 8236 return -EINVAL;
b3546035 8237 if (new_chunk > 0) {
0ba459d2 8238 if (!is_power_of_2(new_chunk))
88ce4930 8239 return -EINVAL;
597a711b 8240 if (new_chunk < (PAGE_SIZE >> 9))
88ce4930 8241 return -EINVAL;
597a711b 8242 if (mddev->array_sectors & (new_chunk-1))
88ce4930
N		 8243 /* not factor of array size */
8244 return -EINVAL;
b3546035 8245 }
88ce4930
N		 8246
8247 /* They look valid */
50ac168a 8248 return check_reshape(mddev);
b3546035
N		 8249}
8250
fd01b88c 8251static void *raid5_takeover(struct mddev *mddev)
d562b0c4
N		 8252{
8253 /* raid5 can take over:
f1b29bca 8254 * raid0 - if there is only one strip zone - make it a raid4 layout
d562b0c4
N		 8255 * raid1 - if there are two drives. We need to know the chunk size
8256 * raid4 - trivial - just use a raid4 layout.
8257 * raid6 - Providing it is a *_6 layout
d562b0c4 8258 */
f1b29bca
DW		 8259 if (mddev->level == 0)
8260 return raid45_takeover_raid0(mddev, 5);
d562b0c4
N		 8261 if (mddev->level == 1)
8262 return raid5_takeover_raid1(mddev);
e9d4758f
N		 8263 if (mddev->level == 4) {
8264 mddev->new_layout = ALGORITHM_PARITY_N;
8265 mddev->new_level = 5;
8266 return setup_conf(mddev);
8267 }
fc9739c6
N		 8268 if (mddev->level == 6)
8269 return raid5_takeover_raid6(mddev);
d562b0c4
N		 8270
8271 return ERR_PTR(-EINVAL);
8272}
8273
fd01b88c 8274static void *raid4_takeover(struct mddev *mddev)
a78d38a1 8275{
f1b29bca
DW		 8276 /* raid4 can take over:
8277 * raid0 - if there is only one strip zone
8278 * raid5 - if layout is right
a78d38a1 8279 */
f1b29bca
DW		 8280 if (mddev->level == 0)
8281 return raid45_takeover_raid0(mddev, 4);
a78d38a1
N		 8282 if (mddev->level == 5 &&
8283 mddev->layout == ALGORITHM_PARITY_N) {
8284 mddev->new_layout = 0;
8285 mddev->new_level = 4;
8286 return setup_conf(mddev);
8287 }
8288 return ERR_PTR(-EINVAL);
8289}
d562b0c4 8290
84fc4b56 8291static struct md_personality raid5_personality;
245f46c2 8292
fd01b88c 8293static void *raid6_takeover(struct mddev *mddev)
245f46c2
N		 8294{
8295 /* Currently can only take over a raid5. We map the
8296 * personality to an equivalent raid6 personality
8297 * with the Q block at the end.
8298 */
8299 int new_layout;
8300
8301 if (mddev->pers != &raid5_personality)
8302 return ERR_PTR(-EINVAL);
8303 if (mddev->degraded > 1)
8304 return ERR_PTR(-EINVAL);
8305 if (mddev->raid_disks > 253)
8306 return ERR_PTR(-EINVAL);
8307 if (mddev->raid_disks < 3)
8308 return ERR_PTR(-EINVAL);
8309
8310 switch (mddev->layout) {
8311 case ALGORITHM_LEFT_ASYMMETRIC:
8312 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
8313 break;
8314 case ALGORITHM_RIGHT_ASYMMETRIC:
8315 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
8316 break;
8317 case ALGORITHM_LEFT_SYMMETRIC:
8318 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
8319 break;
8320 case ALGORITHM_RIGHT_SYMMETRIC:
8321 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
8322 break;
8323 case ALGORITHM_PARITY_0:
8324 new_layout = ALGORITHM_PARITY_0_6;
8325 break;
8326 case ALGORITHM_PARITY_N:
8327 new_layout = ALGORITHM_PARITY_N;
8328 break;
8329 default:
8330 return ERR_PTR(-EINVAL);
8331 }
8332 mddev->new_level = 6;
8333 mddev->new_layout = new_layout;
8334 mddev->delta_disks = 1;
8335 mddev->raid_disks += 1;
8336 return setup_conf(mddev);
8337}
8338
ba903a3e
AP		 8339static int raid5_change_consistency_policy(struct mddev *mddev, const char *buf)
8340{
8341 struct r5conf *conf;
8342 int err;
8343
8344 err = mddev_lock(mddev);
8345 if (err)
8346 return err;
8347 conf = mddev->private;
8348 if (!conf) {
8349 mddev_unlock(mddev);
8350 return -ENODEV;
8351 }
8352
845b9e22 8353 if (strncmp(buf, "ppl", 3) == 0) {
0bb0c105 8354 /* ppl only works with RAID 5 */
845b9e22
AP		 8355 if (!raid5_has_ppl(conf) && conf->level == 5) {
8356 err = log_init(conf, NULL, true);
8357 if (!err) {
8358 err = resize_stripes(conf, conf->pool_size);
8359 if (err)
8360 log_exit(conf);
8361 }
0bb0c105
SL		 8362 } else
8363 err = -EINVAL;
8364 } else if (strncmp(buf, "resync", 6) == 0) {
8365 if (raid5_has_ppl(conf)) {
8366 mddev_suspend(mddev);
8367 log_exit(conf);
0bb0c105 8368 mddev_resume(mddev);
845b9e22 8369 err = resize_stripes(conf, conf->pool_size);
0bb0c105
SL		 8370 } else if (test_bit(MD_HAS_JOURNAL, &conf->mddev->flags) &&
8371 r5l_log_disk_error(conf)) {
8372 bool journal_dev_exists = false;
8373 struct md_rdev *rdev;
8374
8375 rdev_for_each(rdev, mddev)
8376 if (test_bit(Journal, &rdev->flags)) {
8377 journal_dev_exists = true;
8378 break;
8379 }
8380
8381 if (!journal_dev_exists) {
8382 mddev_suspend(mddev);
8383 clear_bit(MD_HAS_JOURNAL, &mddev->flags);
8384 mddev_resume(mddev);
8385 } else /* need remove journal device first */
8386 err = -EBUSY;
8387 } else
8388 err = -EINVAL;
ba903a3e
AP		 8389 } else {
8390 err = -EINVAL;
8391 }
8392
8393 if (!err)
8394 md_update_sb(mddev, 1);
8395
8396 mddev_unlock(mddev);
8397
8398 return err;
8399}
8400
d5d885fd
SL		 8401static int raid5_start(struct mddev *mddev)
8402{
8403 struct r5conf *conf = mddev->private;
8404
8405 return r5l_start(conf->log);
8406}
8407
84fc4b56 8408static struct md_personality raid6_personality =
16a53ecc
N		 8409{
8410 .name = "raid6",
8411 .level = 6,
8412 .owner = THIS_MODULE,
849674e4
SL		 8413 .make_request = raid5_make_request,
8414 .run = raid5_run,
d5d885fd 8415 .start = raid5_start,
afa0f557 8416 .free = raid5_free,
849674e4
SL		 8417 .status = raid5_status,
8418 .error_handler = raid5_error,
16a53ecc
N		 8419 .hot_add_disk = raid5_add_disk,
8420 .hot_remove_disk= raid5_remove_disk,
8421 .spare_active = raid5_spare_active,
849674e4 8422 .sync_request = raid5_sync_request,
16a53ecc 8423 .resize = raid5_resize,
80c3a6ce 8424 .size = raid5_size,
50ac168a 8425 .check_reshape = raid6_check_reshape,
f416885e 8426 .start_reshape = raid5_start_reshape,
cea9c228 8427 .finish_reshape = raid5_finish_reshape,
16a53ecc 8428 .quiesce = raid5_quiesce,
245f46c2 8429 .takeover = raid6_takeover,
5c675f83 8430 .congested = raid5_congested,
0bb0c105 8431 .change_consistency_policy = raid5_change_consistency_policy,
16a53ecc 8432};
84fc4b56 8433static struct md_personality raid5_personality =
1da177e4
LT		 8434{
8435 .name = "raid5",
2604b703 8436 .level = 5,
1da177e4 8437 .owner = THIS_MODULE,
849674e4
SL		 8438 .make_request = raid5_make_request,
8439 .run = raid5_run,
d5d885fd 8440 .start = raid5_start,
afa0f557 8441 .free = raid5_free,
849674e4
SL		 8442 .status = raid5_status,
8443 .error_handler = raid5_error,
1da177e4
LT		 8444 .hot_add_disk = raid5_add_disk,
8445 .hot_remove_disk= raid5_remove_disk,
8446 .spare_active = raid5_spare_active,
849674e4 8447 .sync_request = raid5_sync_request,
1da177e4 8448 .resize = raid5_resize,
80c3a6ce 8449 .size = raid5_size,
63c70c4f
N		 8450 .check_reshape = raid5_check_reshape,
8451 .start_reshape = raid5_start_reshape,
cea9c228 8452 .finish_reshape = raid5_finish_reshape,
72626685 8453 .quiesce = raid5_quiesce,
d562b0c4 8454 .takeover = raid5_takeover,
5c675f83 8455 .congested = raid5_congested,
ba903a3e 8456 .change_consistency_policy = raid5_change_consistency_policy,
1da177e4
LT		 8457};
8458
84fc4b56 8459static struct md_personality raid4_personality =
1da177e4 8460{
2604b703
N		 8461 .name = "raid4",
8462 .level = 4,
8463 .owner = THIS_MODULE,
849674e4
SL		 8464 .make_request = raid5_make_request,
8465 .run = raid5_run,
d5d885fd 8466 .start = raid5_start,
afa0f557 8467 .free = raid5_free,
849674e4
SL		 8468 .status = raid5_status,
8469 .error_handler = raid5_error,
2604b703
N		 8470 .hot_add_disk = raid5_add_disk,
8471 .hot_remove_disk= raid5_remove_disk,
8472 .spare_active = raid5_spare_active,
849674e4 8473 .sync_request = raid5_sync_request,
2604b703 8474 .resize = raid5_resize,
80c3a6ce 8475 .size = raid5_size,
3d37890b
N		 8476 .check_reshape = raid5_check_reshape,
8477 .start_reshape = raid5_start_reshape,
cea9c228 8478 .finish_reshape = raid5_finish_reshape,
2604b703 8479 .quiesce = raid5_quiesce,
a78d38a1 8480 .takeover = raid4_takeover,
5c675f83 8481 .congested = raid5_congested,
0bb0c105 8482 .change_consistency_policy = raid5_change_consistency_policy,
2604b703
N		 8483};
8484
8485static int __init raid5_init(void)
8486{
29c6d1bb
SAS		 8487 int ret;
8488
851c30c9
SL		 8489 raid5_wq = alloc_workqueue("raid5wq",
8490 WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0);
8491 if (!raid5_wq)
8492 return -ENOMEM;
29c6d1bb
SAS		 8493
8494 ret = cpuhp_setup_state_multi(CPUHP_MD_RAID5_PREPARE,
8495 "md/raid5:prepare",
8496 raid456_cpu_up_prepare,
8497 raid456_cpu_dead);
8498 if (ret) {
8499 destroy_workqueue(raid5_wq);
8500 return ret;
8501 }
16a53ecc 8502 register_md_personality(&raid6_personality);
2604b703
N		 8503 register_md_personality(&raid5_personality);
8504 register_md_personality(&raid4_personality);
8505 return 0;
1da177e4
LT		 8506}
8507
2604b703 8508static void raid5_exit(void)
1da177e4 8509{
16a53ecc 8510 unregister_md_personality(&raid6_personality);
2604b703
N		 8511 unregister_md_personality(&raid5_personality);
8512 unregister_md_personality(&raid4_personality);
29c6d1bb 8513 cpuhp_remove_multi_state(CPUHP_MD_RAID5_PREPARE);
851c30c9 8514 destroy_workqueue(raid5_wq);
1da177e4
LT		 8515}
8516
8517module_init(raid5_init);
8518module_exit(raid5_exit);
8519MODULE_LICENSE("GPL");
0efb9e61 8520MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
1da177e4 8521MODULE_ALIAS("md-personality-4"); /* RAID5 */
d9d166c2
N		 8522MODULE_ALIAS("md-raid5");
8523MODULE_ALIAS("md-raid4");
2604b703
N		 8524MODULE_ALIAS("md-level-5");
8525MODULE_ALIAS("md-level-4");
16a53ecc
N		 8526MODULE_ALIAS("md-personality-8"); /* RAID6 */
8527MODULE_ALIAS("md-raid6");
8528MODULE_ALIAS("md-level-6");
8529
8530/* This used to be two separate modules, they were: */
8531MODULE_ALIAS("raid5");
8532MODULE_ALIAS("raid6");
Linux 6.x block layer and io_uring tree(s)