projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
MD: kick out journal disk if it's not fresh
[linux-2.6-block.git] / drivers / md / raid5.c
CommitLineData
1da177e4
LT		 1/*
2 * raid5.c : Multiple Devices driver for Linux
3 * Copyright (C) 1996, 1997 Ingo Molnar, Miguel de Icaza, Gadi Oxman
4 * Copyright (C) 1999, 2000 Ingo Molnar
16a53ecc 5 * Copyright (C) 2002, 2003 H. Peter Anvin
1da177e4 6 *
16a53ecc
N		 7 * RAID-4/5/6 management functions.
8 * Thanks to Penguin Computing for making the RAID-6 development possible
9 * by donating a test server!
1da177e4
LT		 10 *
11 * This program is free software; you can redistribute it and/or modify
12 * it under the terms of the GNU General Public License as published by
13 * the Free Software Foundation; either version 2, or (at your option)
14 * any later version.
15 *
16 * You should have received a copy of the GNU General Public License
17 * (for example /usr/src/linux/COPYING); if not, write to the Free
18 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
19 */
20
ae3c20cc
N		 21/*
22 * BITMAP UNPLUGGING:
23 *
24 * The sequencing for updating the bitmap reliably is a little
25 * subtle (and I got it wrong the first time) so it deserves some
26 * explanation.
27 *
28 * We group bitmap updates into batches. Each batch has a number.
29 * We may write out several batches at once, but that isn't very important.
7c13edc8
N		 30 * conf->seq_write is the number of the last batch successfully written.
31 * conf->seq_flush is the number of the last batch that was closed to
ae3c20cc
N		 32 * new additions.
33 * When we discover that we will need to write to any block in a stripe
34 * (in add_stripe_bio) we update the in-memory bitmap and record in sh->bm_seq
7c13edc8 35 * the number of the batch it will be in. This is seq_flush+1.
ae3c20cc
N		 36 * When we are ready to do a write, if that batch hasn't been written yet,
37 * we plug the array and queue the stripe for later.
38 * When an unplug happens, we increment bm_flush, thus closing the current
39 * batch.
40 * When we notice that bm_flush > bm_write, we write out all pending updates
41 * to the bitmap, and advance bm_write to where bm_flush was.
42 * This may occasionally write a bit out twice, but is sure never to
43 * miss any bits.
44 */
1da177e4 45
bff61975 46#include <linux/blkdev.h>
f6705578 47#include <linux/kthread.h>
f701d589 48#include <linux/raid/pq.h>
91c00924 49#include <linux/async_tx.h>
056075c7 50#include <linux/module.h>
07a3b417 51#include <linux/async.h>
bff61975 52#include <linux/seq_file.h>
36d1c647 53#include <linux/cpu.h>
5a0e3ad6 54#include <linux/slab.h>
8bda470e 55#include <linux/ratelimit.h>
851c30c9 56#include <linux/nodemask.h>
46d5b785 57#include <linux/flex_array.h>
a9add5d9
N		 58#include <trace/events/block.h>
59
43b2e5d8 60#include "md.h"
bff61975 61#include "raid5.h"
54071b38 62#include "raid0.h"
ef740c37 63#include "bitmap.h"
72626685 64
851c30c9
SL		 65#define cpu_to_group(cpu) cpu_to_node(cpu)
66#define ANY_GROUP NUMA_NO_NODE
67
8e0e99ba
N		 68static bool devices_handle_discard_safely = false;
69module_param(devices_handle_discard_safely, bool, 0644);
70MODULE_PARM_DESC(devices_handle_discard_safely,
71 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
851c30c9 72static struct workqueue_struct *raid5_wq;
1da177e4
LT		 73/*
74 * Stripe cache
75 */
76
77#define NR_STRIPES 256
78#define STRIPE_SIZE PAGE_SIZE
79#define STRIPE_SHIFT (PAGE_SHIFT - 9)
80#define STRIPE_SECTORS (STRIPE_SIZE>>9)
81#define IO_THRESHOLD 1
8b3e6cdc 82#define BYPASS_THRESHOLD 1
fccddba0 83#define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
1da177e4 84#define HASH_MASK (NR_HASH - 1)
bfc90cb0 85#define MAX_STRIPE_BATCH 8
1da177e4 86
d1688a6d 87static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect)
db298e19
N		 88{
89 int hash = (sect >> STRIPE_SHIFT) & HASH_MASK;
90 return &conf->stripe_hashtbl[hash];
91}
1da177e4 92
566c09c5
SL		 93static inline int stripe_hash_locks_hash(sector_t sect)
94{
95 return (sect >> STRIPE_SHIFT) & STRIPE_HASH_LOCKS_MASK;
96}
97
98static inline void lock_device_hash_lock(struct r5conf *conf, int hash)
99{
100 spin_lock_irq(conf->hash_locks + hash);
101 spin_lock(&conf->device_lock);
102}
103
104static inline void unlock_device_hash_lock(struct r5conf *conf, int hash)
105{
106 spin_unlock(&conf->device_lock);
107 spin_unlock_irq(conf->hash_locks + hash);
108}
109
110static inline void lock_all_device_hash_locks_irq(struct r5conf *conf)
111{
112 int i;
113 local_irq_disable();
114 spin_lock(conf->hash_locks);
115 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
116 spin_lock_nest_lock(conf->hash_locks + i, conf->hash_locks);
117 spin_lock(&conf->device_lock);
118}
119
120static inline void unlock_all_device_hash_locks_irq(struct r5conf *conf)
121{
122 int i;
123 spin_unlock(&conf->device_lock);
124 for (i = NR_STRIPE_HASH_LOCKS; i; i--)
125 spin_unlock(conf->hash_locks + i - 1);
126 local_irq_enable();
127}
128
1da177e4
LT		 129/* bio's attached to a stripe+device for I/O are linked together in bi_sector
130 * order without overlap. There may be several bio's per stripe+device, and
131 * a bio could span several devices.
132 * When walking this list for a particular stripe+device, we must never proceed
133 * beyond a bio that extends past this device, as the next bio might no longer
134 * be valid.
db298e19 135 * This function is used to determine the 'next' bio in the list, given the sector
1da177e4
LT		 136 * of the current stripe+device
137 */
db298e19
N		 138static inline struct bio *r5_next_bio(struct bio *bio, sector_t sector)
139{
aa8b57aa 140 int sectors = bio_sectors(bio);
4f024f37 141 if (bio->bi_iter.bi_sector + sectors < sector + STRIPE_SECTORS)
db298e19
N		 142 return bio->bi_next;
143 else
144 return NULL;
145}
1da177e4 146
960e739d 147/*
5b99c2ff
JA		 148 * We maintain a biased count of active stripes in the bottom 16 bits of
149 * bi_phys_segments, and a count of processed stripes in the upper 16 bits
960e739d 150 */
e7836bd6 151static inline int raid5_bi_processed_stripes(struct bio *bio)
960e739d 152{
e7836bd6
SL		 153 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
154 return (atomic_read(segments) >> 16) & 0xffff;
960e739d
JA		 155}
156
e7836bd6 157static inline int raid5_dec_bi_active_stripes(struct bio *bio)
960e739d 158{
e7836bd6
SL		 159 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
160 return atomic_sub_return(1, segments) & 0xffff;
960e739d
JA		 161}
162
e7836bd6 163static inline void raid5_inc_bi_active_stripes(struct bio *bio)
960e739d 164{
e7836bd6
SL		 165 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
166 atomic_inc(segments);
960e739d
JA		 167}
168
e7836bd6
SL		 169static inline void raid5_set_bi_processed_stripes(struct bio *bio,
170 unsigned int cnt)
960e739d 171{
e7836bd6
SL		 172 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
173 int old, new;
960e739d 174
e7836bd6
SL		 175 do {
176 old = atomic_read(segments);
177 new = (old & 0xffff) | (cnt << 16);
178 } while (atomic_cmpxchg(segments, old, new) != old);
960e739d
JA		 179}
180
e7836bd6 181static inline void raid5_set_bi_stripes(struct bio *bio, unsigned int cnt)
960e739d 182{
e7836bd6
SL		 183 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
184 atomic_set(segments, cnt);
960e739d
JA		 185}
186
d0dabf7e
N		 187/* Find first data disk in a raid6 stripe */
188static inline int raid6_d0(struct stripe_head *sh)
189{
67cc2b81
N		 190 if (sh->ddf_layout)
191 /* ddf always start from first device */
192 return 0;
193 /* md starts just after Q block */
d0dabf7e
N		 194 if (sh->qd_idx == sh->disks - 1)
195 return 0;
196 else
197 return sh->qd_idx + 1;
198}
16a53ecc
N		 199static inline int raid6_next_disk(int disk, int raid_disks)
200{
201 disk++;
202 return (disk < raid_disks) ? disk : 0;
203}
a4456856 204
d0dabf7e
N		 205/* When walking through the disks in a raid5, starting at raid6_d0,
206 * We need to map each disk to a 'slot', where the data disks are slot
207 * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk
208 * is raid_disks-1. This help does that mapping.
209 */
67cc2b81
N		 210static int raid6_idx_to_slot(int idx, struct stripe_head *sh,
211 int *count, int syndrome_disks)
d0dabf7e 212{
6629542e 213 int slot = *count;
67cc2b81 214
e4424fee 215 if (sh->ddf_layout)
6629542e 216 (*count)++;
d0dabf7e 217 if (idx == sh->pd_idx)
67cc2b81 218 return syndrome_disks;
d0dabf7e 219 if (idx == sh->qd_idx)
67cc2b81 220 return syndrome_disks + 1;
e4424fee 221 if (!sh->ddf_layout)
6629542e 222 (*count)++;
d0dabf7e
N		 223 return slot;
224}
225
34a6f80e 226static void return_io(struct bio_list *return_bi)
a4456856 227{
34a6f80e
N		 228 struct bio *bi;
229 while ((bi = bio_list_pop(return_bi)) != NULL) {
4f024f37 230 bi->bi_iter.bi_size = 0;
0a82a8d1
LT		 231 trace_block_bio_complete(bdev_get_queue(bi->bi_bdev),
232 bi, 0);
4246a0b6 233 bio_endio(bi);
a4456856
DW		 234 }
235}
236
d1688a6d 237static void print_raid5_conf (struct r5conf *conf);
1da177e4 238
600aa109
DW		 239static int stripe_operations_active(struct stripe_head *sh)
240{
241 return sh->check_state || sh->reconstruct_state ||
242 test_bit(STRIPE_BIOFILL_RUN, &sh->state) ||
243 test_bit(STRIPE_COMPUTE_RUN, &sh->state);
244}
245
851c30c9
SL		 246static void raid5_wakeup_stripe_thread(struct stripe_head *sh)
247{
248 struct r5conf *conf = sh->raid_conf;
249 struct r5worker_group *group;
bfc90cb0 250 int thread_cnt;
851c30c9
SL		 251 int i, cpu = sh->cpu;
252
253 if (!cpu_online(cpu)) {
254 cpu = cpumask_any(cpu_online_mask);
255 sh->cpu = cpu;
256 }
257
258 if (list_empty(&sh->lru)) {
259 struct r5worker_group *group;
260 group = conf->worker_groups + cpu_to_group(cpu);
261 list_add_tail(&sh->lru, &group->handle_list);
bfc90cb0
SL		 262 group->stripes_cnt++;
263 sh->group = group;
851c30c9
SL		 264 }
265
266 if (conf->worker_cnt_per_group == 0) {
267 md_wakeup_thread(conf->mddev->thread);
268 return;
269 }
270
271 group = conf->worker_groups + cpu_to_group(sh->cpu);
272
bfc90cb0
SL		 273 group->workers[0].working = true;
274 /* at least one worker should run to avoid race */
275 queue_work_on(sh->cpu, raid5_wq, &group->workers[0].work);
276
277 thread_cnt = group->stripes_cnt / MAX_STRIPE_BATCH - 1;
278 /* wakeup more workers */
279 for (i = 1; i < conf->worker_cnt_per_group && thread_cnt > 0; i++) {
280 if (group->workers[i].working == false) {
281 group->workers[i].working = true;
282 queue_work_on(sh->cpu, raid5_wq,
283 &group->workers[i].work);
284 thread_cnt--;
285 }
286 }
851c30c9
SL		 287}
288
566c09c5
SL		 289static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh,
290 struct list_head *temp_inactive_list)
1da177e4 291{
4eb788df
SL		 292 BUG_ON(!list_empty(&sh->lru));
293 BUG_ON(atomic_read(&conf->active_stripes)==0);
294 if (test_bit(STRIPE_HANDLE, &sh->state)) {
295 if (test_bit(STRIPE_DELAYED, &sh->state) &&
ad3ab8b6 296 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4eb788df 297 list_add_tail(&sh->lru, &conf->delayed_list);
ad3ab8b6 298 else if (test_bit(STRIPE_BIT_DELAY, &sh->state) &&
4eb788df
SL		 299 sh->bm_seq - conf->seq_write > 0)
300 list_add_tail(&sh->lru, &conf->bitmap_list);
301 else {
302 clear_bit(STRIPE_DELAYED, &sh->state);
303 clear_bit(STRIPE_BIT_DELAY, &sh->state);
851c30c9
SL		 304 if (conf->worker_cnt_per_group == 0) {
305 list_add_tail(&sh->lru, &conf->handle_list);
306 } else {
307 raid5_wakeup_stripe_thread(sh);
308 return;
309 }
4eb788df
SL		 310 }
311 md_wakeup_thread(conf->mddev->thread);
312 } else {
313 BUG_ON(stripe_operations_active(sh));
314 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
315 if (atomic_dec_return(&conf->preread_active_stripes)
316 < IO_THRESHOLD)
317 md_wakeup_thread(conf->mddev->thread);
318 atomic_dec(&conf->active_stripes);
566c09c5
SL		 319 if (!test_bit(STRIPE_EXPANDING, &sh->state))
320 list_add_tail(&sh->lru, temp_inactive_list);
1da177e4
LT		 321 }
322}
d0dabf7e 323
566c09c5
SL		 324static void __release_stripe(struct r5conf *conf, struct stripe_head *sh,
325 struct list_head *temp_inactive_list)
4eb788df
SL		 326{
327 if (atomic_dec_and_test(&sh->count))
566c09c5
SL		 328 do_release_stripe(conf, sh, temp_inactive_list);
329}
330
331/*
332 * @hash could be NR_STRIPE_HASH_LOCKS, then we have a list of inactive_list
333 *
334 * Be careful: Only one task can add/delete stripes from temp_inactive_list at
335 * given time. Adding stripes only takes device lock, while deleting stripes
336 * only takes hash lock.
337 */
338static void release_inactive_stripe_list(struct r5conf *conf,
339 struct list_head *temp_inactive_list,
340 int hash)
341{
342 int size;
e9e4c377
YL		 343 unsigned long do_wakeup = 0;
344 int i = 0;
566c09c5
SL		 345 unsigned long flags;
346
347 if (hash == NR_STRIPE_HASH_LOCKS) {
348 size = NR_STRIPE_HASH_LOCKS;
349 hash = NR_STRIPE_HASH_LOCKS - 1;
350 } else
351 size = 1;
352 while (size) {
353 struct list_head *list = &temp_inactive_list[size - 1];
354
355 /*
6d036f7d 356 * We don't hold any lock here yet, raid5_get_active_stripe() might
566c09c5
SL		 357 * remove stripes from the list
358 */
359 if (!list_empty_careful(list)) {
360 spin_lock_irqsave(conf->hash_locks + hash, flags);
4bda556a
SL		 361 if (list_empty(conf->inactive_list + hash) &&
362 !list_empty(list))
363 atomic_dec(&conf->empty_inactive_list_nr);
566c09c5 364 list_splice_tail_init(list, conf->inactive_list + hash);
e9e4c377 365 do_wakeup |= 1 << hash;
566c09c5
SL		 366 spin_unlock_irqrestore(conf->hash_locks + hash, flags);
367 }
368 size--;
369 hash--;
370 }
371
e9e4c377
YL		 372 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) {
373 if (do_wakeup & (1 << i))
374 wake_up(&conf->wait_for_stripe[i]);
375 }
376
566c09c5 377 if (do_wakeup) {
b1b46486
YL		 378 if (atomic_read(&conf->active_stripes) == 0)
379 wake_up(&conf->wait_for_quiescent);
566c09c5
SL		 380 if (conf->retry_read_aligned)
381 md_wakeup_thread(conf->mddev->thread);
382 }
4eb788df
SL		 383}
384
773ca82f 385/* should hold conf->device_lock already */
566c09c5
SL		 386static int release_stripe_list(struct r5conf *conf,
387 struct list_head *temp_inactive_list)
773ca82f
SL		 388{
389 struct stripe_head *sh;
390 int count = 0;
391 struct llist_node *head;
392
393 head = llist_del_all(&conf->released_stripes);
d265d9dc 394 head = llist_reverse_order(head);
773ca82f 395 while (head) {
566c09c5
SL		 396 int hash;
397
773ca82f
SL		 398 sh = llist_entry(head, struct stripe_head, release_list);
399 head = llist_next(head);
400 /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */
401 smp_mb();
402 clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state);
403 /*
404 * Don't worry the bit is set here, because if the bit is set
405 * again, the count is always > 1. This is true for
406 * STRIPE_ON_UNPLUG_LIST bit too.
407 */
566c09c5
SL		 408 hash = sh->hash_lock_index;
409 __release_stripe(conf, sh, &temp_inactive_list[hash]);
773ca82f
SL		 410 count++;
411 }
412
413 return count;
414}
415
6d036f7d 416void raid5_release_stripe(struct stripe_head *sh)
1da177e4 417{
d1688a6d 418 struct r5conf *conf = sh->raid_conf;
1da177e4 419 unsigned long flags;
566c09c5
SL		 420 struct list_head list;
421 int hash;
773ca82f 422 bool wakeup;
16a53ecc 423
cf170f3f
ES		 424 /* Avoid release_list until the last reference.
425 */
426 if (atomic_add_unless(&sh->count, -1, 1))
427 return;
428
ad4068de 429 if (unlikely(!conf->mddev->thread) ||
430 test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state))
773ca82f
SL		 431 goto slow_path;
432 wakeup = llist_add(&sh->release_list, &conf->released_stripes);
433 if (wakeup)
434 md_wakeup_thread(conf->mddev->thread);
435 return;
436slow_path:
4eb788df 437 local_irq_save(flags);
773ca82f 438 /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */
4eb788df 439 if (atomic_dec_and_lock(&sh->count, &conf->device_lock)) {
566c09c5
SL		 440 INIT_LIST_HEAD(&list);
441 hash = sh->hash_lock_index;
442 do_release_stripe(conf, sh, &list);
4eb788df 443 spin_unlock(&conf->device_lock);
566c09c5 444 release_inactive_stripe_list(conf, &list, hash);
4eb788df
SL		 445 }
446 local_irq_restore(flags);
1da177e4
LT		 447}
448
fccddba0 449static inline void remove_hash(struct stripe_head *sh)
1da177e4 450{
45b4233c
DW		 451 pr_debug("remove_hash(), stripe %llu\n",
452 (unsigned long long)sh->sector);
1da177e4 453
fccddba0 454 hlist_del_init(&sh->hash);
1da177e4
LT		 455}
456
d1688a6d 457static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh)
1da177e4 458{
fccddba0 459 struct hlist_head *hp = stripe_hash(conf, sh->sector);
1da177e4 460
45b4233c
DW		 461 pr_debug("insert_hash(), stripe %llu\n",
462 (unsigned long long)sh->sector);
1da177e4 463
fccddba0 464 hlist_add_head(&sh->hash, hp);
1da177e4
LT		 465}
466
1da177e4 467/* find an idle stripe, make sure it is unhashed, and return it. */
566c09c5 468static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash)
1da177e4
LT		 469{
470 struct stripe_head *sh = NULL;
471 struct list_head *first;
472
566c09c5 473 if (list_empty(conf->inactive_list + hash))
1da177e4 474 goto out;
566c09c5 475 first = (conf->inactive_list + hash)->next;
1da177e4
LT		 476 sh = list_entry(first, struct stripe_head, lru);
477 list_del_init(first);
478 remove_hash(sh);
479 atomic_inc(&conf->active_stripes);
566c09c5 480 BUG_ON(hash != sh->hash_lock_index);
4bda556a
SL		 481 if (list_empty(conf->inactive_list + hash))
482 atomic_inc(&conf->empty_inactive_list_nr);
1da177e4
LT		 483out:
484 return sh;
485}
486
e4e11e38 487static void shrink_buffers(struct stripe_head *sh)
1da177e4
LT		 488{
489 struct page *p;
490 int i;
e4e11e38 491 int num = sh->raid_conf->pool_size;
1da177e4 492
e4e11e38 493 for (i = 0; i < num ; i++) {
d592a996 494 WARN_ON(sh->dev[i].page != sh->dev[i].orig_page);
1da177e4
LT		 495 p = sh->dev[i].page;
496 if (!p)
497 continue;
498 sh->dev[i].page = NULL;
2d1f3b5d 499 put_page(p);
1da177e4
LT		 500 }
501}
502
a9683a79 503static int grow_buffers(struct stripe_head *sh, gfp_t gfp)
1da177e4
LT		 504{
505 int i;
e4e11e38 506 int num = sh->raid_conf->pool_size;
1da177e4 507
e4e11e38 508 for (i = 0; i < num; i++) {
1da177e4
LT		 509 struct page *page;
510
a9683a79 511 if (!(page = alloc_page(gfp))) {
1da177e4
LT		 512 return 1;
513 }
514 sh->dev[i].page = page;
d592a996 515 sh->dev[i].orig_page = page;
1da177e4
LT		 516 }
517 return 0;
518}
519
784052ec 520static void raid5_build_block(struct stripe_head *sh, int i, int previous);
d1688a6d 521static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
911d4ee8 522 struct stripe_head *sh);
1da177e4 523
b5663ba4 524static void init_stripe(struct stripe_head *sh, sector_t sector, int previous)
1da177e4 525{
d1688a6d 526 struct r5conf *conf = sh->raid_conf;
566c09c5 527 int i, seq;
1da177e4 528
78bafebd
ES		 529 BUG_ON(atomic_read(&sh->count) != 0);
530 BUG_ON(test_bit(STRIPE_HANDLE, &sh->state));
600aa109 531 BUG_ON(stripe_operations_active(sh));
59fc630b 532 BUG_ON(sh->batch_head);
d84e0f10 533
45b4233c 534 pr_debug("init_stripe called, stripe %llu\n",
b8e6a15a 535 (unsigned long long)sector);
566c09c5
SL		 536retry:
537 seq = read_seqcount_begin(&conf->gen_lock);
86b42c71 538 sh->generation = conf->generation - previous;
b5663ba4 539 sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks;
1da177e4 540 sh->sector = sector;
911d4ee8 541 stripe_set_idx(sector, conf, previous, sh);
1da177e4
LT		 542 sh->state = 0;
543
7ecaa1e6 544 for (i = sh->disks; i--; ) {
1da177e4
LT		 545 struct r5dev *dev = &sh->dev[i];
546
d84e0f10 547 if (dev->toread || dev->read || dev->towrite || dev->written ||
1da177e4 548 test_bit(R5_LOCKED, &dev->flags)) {
d84e0f10 549 printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n",
1da177e4 550 (unsigned long long)sh->sector, i, dev->toread,
d84e0f10 551 dev->read, dev->towrite, dev->written,
1da177e4 552 test_bit(R5_LOCKED, &dev->flags));
8cfa7b0f 553 WARN_ON(1);
1da177e4
LT		 554 }
555 dev->flags = 0;
784052ec 556 raid5_build_block(sh, i, previous);
1da177e4 557 }
566c09c5
SL		 558 if (read_seqcount_retry(&conf->gen_lock, seq))
559 goto retry;
7a87f434 560 sh->overwrite_disks = 0;
1da177e4 561 insert_hash(conf, sh);
851c30c9 562 sh->cpu = smp_processor_id();
da41ba65 563 set_bit(STRIPE_BATCH_READY, &sh->state);
1da177e4
LT		 564}
565
d1688a6d 566static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector,
86b42c71 567 short generation)
1da177e4
LT		 568{
569 struct stripe_head *sh;
570
45b4233c 571 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
b67bfe0d 572 hlist_for_each_entry(sh, stripe_hash(conf, sector), hash)
86b42c71 573 if (sh->sector == sector && sh->generation == generation)
1da177e4 574 return sh;
45b4233c 575 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
1da177e4
LT		 576 return NULL;
577}
578
674806d6
N		 579/*
580 * Need to check if array has failed when deciding whether to:
581 * - start an array
582 * - remove non-faulty devices
583 * - add a spare
584 * - allow a reshape
585 * This determination is simple when no reshape is happening.
586 * However if there is a reshape, we need to carefully check
587 * both the before and after sections.
588 * This is because some failed devices may only affect one
589 * of the two sections, and some non-in_sync devices may
590 * be insync in the section most affected by failed devices.
591 */
908f4fbd 592static int calc_degraded(struct r5conf *conf)
674806d6 593{
908f4fbd 594 int degraded, degraded2;
674806d6 595 int i;
674806d6
N		 596
597 rcu_read_lock();
598 degraded = 0;
599 for (i = 0; i < conf->previous_raid_disks; i++) {
3cb03002 600 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
e5c86471
N		 601 if (rdev && test_bit(Faulty, &rdev->flags))
602 rdev = rcu_dereference(conf->disks[i].replacement);
674806d6
N		 603 if (!rdev || test_bit(Faulty, &rdev->flags))
604 degraded++;
605 else if (test_bit(In_sync, &rdev->flags))
606 ;
607 else
608 /* not in-sync or faulty.
609 * If the reshape increases the number of devices,
610 * this is being recovered by the reshape, so
611 * this 'previous' section is not in_sync.
612 * If the number of devices is being reduced however,
613 * the device can only be part of the array if
614 * we are reverting a reshape, so this section will
615 * be in-sync.
616 */
617 if (conf->raid_disks >= conf->previous_raid_disks)
618 degraded++;
619 }
620 rcu_read_unlock();
908f4fbd
N		 621 if (conf->raid_disks == conf->previous_raid_disks)
622 return degraded;
674806d6 623 rcu_read_lock();
908f4fbd 624 degraded2 = 0;
674806d6 625 for (i = 0; i < conf->raid_disks; i++) {
3cb03002 626 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
e5c86471
N		 627 if (rdev && test_bit(Faulty, &rdev->flags))
628 rdev = rcu_dereference(conf->disks[i].replacement);
674806d6 629 if (!rdev || test_bit(Faulty, &rdev->flags))
908f4fbd 630 degraded2++;
674806d6
N		 631 else if (test_bit(In_sync, &rdev->flags))
632 ;
633 else
634 /* not in-sync or faulty.
635 * If reshape increases the number of devices, this
636 * section has already been recovered, else it
637 * almost certainly hasn't.
638 */
639 if (conf->raid_disks <= conf->previous_raid_disks)
908f4fbd 640 degraded2++;
674806d6
N		 641 }
642 rcu_read_unlock();
908f4fbd
N		 643 if (degraded2 > degraded)
644 return degraded2;
645 return degraded;
646}
647
648static int has_failed(struct r5conf *conf)
649{
650 int degraded;
651
652 if (conf->mddev->reshape_position == MaxSector)
653 return conf->mddev->degraded > conf->max_degraded;
654
655 degraded = calc_degraded(conf);
674806d6
N		 656 if (degraded > conf->max_degraded)
657 return 1;
658 return 0;
659}
660
6d036f7d
SL		 661struct stripe_head *
662raid5_get_active_stripe(struct r5conf *conf, sector_t sector,
663 int previous, int noblock, int noquiesce)
1da177e4
LT		 664{
665 struct stripe_head *sh;
566c09c5 666 int hash = stripe_hash_locks_hash(sector);
1da177e4 667
45b4233c 668 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector);
1da177e4 669
566c09c5 670 spin_lock_irq(conf->hash_locks + hash);
1da177e4
LT		 671
672 do {
b1b46486 673 wait_event_lock_irq(conf->wait_for_quiescent,
a8c906ca 674 conf->quiesce == 0 || noquiesce,
566c09c5 675 *(conf->hash_locks + hash));
86b42c71 676 sh = __find_stripe(conf, sector, conf->generation - previous);
1da177e4 677 if (!sh) {
edbe83ab 678 if (!test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) {
566c09c5 679 sh = get_free_stripe(conf, hash);
713bc5c2
SL		 680 if (!sh && !test_bit(R5_DID_ALLOC,
681 &conf->cache_state))
edbe83ab
N		 682 set_bit(R5_ALLOC_MORE,
683 &conf->cache_state);
684 }
1da177e4
LT		 685 if (noblock && sh == NULL)
686 break;
687 if (!sh) {
5423399a
N		 688 set_bit(R5_INACTIVE_BLOCKED,
689 &conf->cache_state);
e9e4c377
YL		 690 wait_event_exclusive_cmd(
691 conf->wait_for_stripe[hash],
566c09c5
SL		 692 !list_empty(conf->inactive_list + hash) &&
693 (atomic_read(&conf->active_stripes)
694 < (conf->max_nr_stripes * 3 / 4)
5423399a
N		 695 || !test_bit(R5_INACTIVE_BLOCKED,
696 &conf->cache_state)),
e9e4c377
YL		 697 spin_unlock_irq(conf->hash_locks + hash),
698 spin_lock_irq(conf->hash_locks + hash));
5423399a
N		 699 clear_bit(R5_INACTIVE_BLOCKED,
700 &conf->cache_state);
7da9d450 701 } else {
b5663ba4 702 init_stripe(sh, sector, previous);
7da9d450
N		 703 atomic_inc(&sh->count);
704 }
e240c183 705 } else if (!atomic_inc_not_zero(&sh->count)) {
6d183de4 706 spin_lock(&conf->device_lock);
e240c183 707 if (!atomic_read(&sh->count)) {
1da177e4
LT		 708 if (!test_bit(STRIPE_HANDLE, &sh->state))
709 atomic_inc(&conf->active_stripes);
5af9bef7
N		 710 BUG_ON(list_empty(&sh->lru) &&
711 !test_bit(STRIPE_EXPANDING, &sh->state));
16a53ecc 712 list_del_init(&sh->lru);
bfc90cb0
SL		 713 if (sh->group) {
714 sh->group->stripes_cnt--;
715 sh->group = NULL;
716 }
1da177e4 717 }
7da9d450 718 atomic_inc(&sh->count);
6d183de4 719 spin_unlock(&conf->device_lock);
1da177e4
LT		 720 }
721 } while (sh == NULL);
722
e9e4c377
YL		 723 if (!list_empty(conf->inactive_list + hash))
724 wake_up(&conf->wait_for_stripe[hash]);
725
566c09c5 726 spin_unlock_irq(conf->hash_locks + hash);
1da177e4
LT		 727 return sh;
728}
729
7a87f434 730static bool is_full_stripe_write(struct stripe_head *sh)
731{
732 BUG_ON(sh->overwrite_disks > (sh->disks - sh->raid_conf->max_degraded));
733 return sh->overwrite_disks == (sh->disks - sh->raid_conf->max_degraded);
734}
735
59fc630b 736static void lock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2)
737{
738 local_irq_disable();
739 if (sh1 > sh2) {
740 spin_lock(&sh2->stripe_lock);
741 spin_lock_nested(&sh1->stripe_lock, 1);
742 } else {
743 spin_lock(&sh1->stripe_lock);
744 spin_lock_nested(&sh2->stripe_lock, 1);
745 }
746}
747
748static void unlock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2)
749{
750 spin_unlock(&sh1->stripe_lock);
751 spin_unlock(&sh2->stripe_lock);
752 local_irq_enable();
753}
754
755/* Only freshly new full stripe normal write stripe can be added to a batch list */
756static bool stripe_can_batch(struct stripe_head *sh)
757{
9c3e333d
SL		 758 struct r5conf *conf = sh->raid_conf;
759
760 if (conf->log)
761 return false;
59fc630b 762 return test_bit(STRIPE_BATCH_READY, &sh->state) &&
d0852df5 763 !test_bit(STRIPE_BITMAP_PENDING, &sh->state) &&
59fc630b 764 is_full_stripe_write(sh);
765}
766
767/* we only do back search */
768static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh)
769{
770 struct stripe_head *head;
771 sector_t head_sector, tmp_sec;
772 int hash;
773 int dd_idx;
774
775 if (!stripe_can_batch(sh))
776 return;
777 /* Don't cross chunks, so stripe pd_idx/qd_idx is the same */
778 tmp_sec = sh->sector;
779 if (!sector_div(tmp_sec, conf->chunk_sectors))
780 return;
781 head_sector = sh->sector - STRIPE_SECTORS;
782
783 hash = stripe_hash_locks_hash(head_sector);
784 spin_lock_irq(conf->hash_locks + hash);
785 head = __find_stripe(conf, head_sector, conf->generation);
786 if (head && !atomic_inc_not_zero(&head->count)) {
787 spin_lock(&conf->device_lock);
788 if (!atomic_read(&head->count)) {
789 if (!test_bit(STRIPE_HANDLE, &head->state))
790 atomic_inc(&conf->active_stripes);
791 BUG_ON(list_empty(&head->lru) &&
792 !test_bit(STRIPE_EXPANDING, &head->state));
793 list_del_init(&head->lru);
794 if (head->group) {
795 head->group->stripes_cnt--;
796 head->group = NULL;
797 }
798 }
799 atomic_inc(&head->count);
800 spin_unlock(&conf->device_lock);
801 }
802 spin_unlock_irq(conf->hash_locks + hash);
803
804 if (!head)
805 return;
806 if (!stripe_can_batch(head))
807 goto out;
808
809 lock_two_stripes(head, sh);
810 /* clear_batch_ready clear the flag */
811 if (!stripe_can_batch(head) || !stripe_can_batch(sh))
812 goto unlock_out;
813
814 if (sh->batch_head)
815 goto unlock_out;
816
817 dd_idx = 0;
818 while (dd_idx == sh->pd_idx || dd_idx == sh->qd_idx)
819 dd_idx++;
820 if (head->dev[dd_idx].towrite->bi_rw != sh->dev[dd_idx].towrite->bi_rw)
821 goto unlock_out;
822
823 if (head->batch_head) {
824 spin_lock(&head->batch_head->batch_lock);
825 /* This batch list is already running */
826 if (!stripe_can_batch(head)) {
827 spin_unlock(&head->batch_head->batch_lock);
828 goto unlock_out;
829 }
830
831 /*
832 * at this point, head's BATCH_READY could be cleared, but we
833 * can still add the stripe to batch list
834 */
835 list_add(&sh->batch_list, &head->batch_list);
836 spin_unlock(&head->batch_head->batch_lock);
837
838 sh->batch_head = head->batch_head;
839 } else {
840 head->batch_head = head;
841 sh->batch_head = head->batch_head;
842 spin_lock(&head->batch_lock);
843 list_add_tail(&sh->batch_list, &head->batch_list);
844 spin_unlock(&head->batch_lock);
845 }
846
847 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
848 if (atomic_dec_return(&conf->preread_active_stripes)
849 < IO_THRESHOLD)
850 md_wakeup_thread(conf->mddev->thread);
851
2b6b2457
N		 852 if (test_and_clear_bit(STRIPE_BIT_DELAY, &sh->state)) {
853 int seq = sh->bm_seq;
854 if (test_bit(STRIPE_BIT_DELAY, &sh->batch_head->state) &&
855 sh->batch_head->bm_seq > seq)
856 seq = sh->batch_head->bm_seq;
857 set_bit(STRIPE_BIT_DELAY, &sh->batch_head->state);
858 sh->batch_head->bm_seq = seq;
859 }
860
59fc630b 861 atomic_inc(&sh->count);
862unlock_out:
863 unlock_two_stripes(head, sh);
864out:
6d036f7d 865 raid5_release_stripe(head);
59fc630b 866}
867
05616be5
N		 868/* Determine if 'data_offset' or 'new_data_offset' should be used
869 * in this stripe_head.
870 */
871static int use_new_offset(struct r5conf *conf, struct stripe_head *sh)
872{
873 sector_t progress = conf->reshape_progress;
874 /* Need a memory barrier to make sure we see the value
875 * of conf->generation, or ->data_offset that was set before
876 * reshape_progress was updated.
877 */
878 smp_rmb();
879 if (progress == MaxSector)
880 return 0;
881 if (sh->generation == conf->generation - 1)
882 return 0;
883 /* We are in a reshape, and this is a new-generation stripe,
884 * so use new_data_offset.
885 */
886 return 1;
887}
888
6712ecf8 889static void
4246a0b6 890raid5_end_read_request(struct bio *bi);
6712ecf8 891static void
4246a0b6 892raid5_end_write_request(struct bio *bi);
91c00924 893
c4e5ac0a 894static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
91c00924 895{
d1688a6d 896 struct r5conf *conf = sh->raid_conf;
91c00924 897 int i, disks = sh->disks;
59fc630b 898 struct stripe_head *head_sh = sh;
91c00924
DW		 899
900 might_sleep();
901
f6bed0ef
SL		 902 if (r5l_write_stripe(conf->log, sh) == 0)
903 return;
91c00924
DW		 904 for (i = disks; i--; ) {
905 int rw;
9a3e1101 906 int replace_only = 0;
977df362
N		 907 struct bio *bi, *rbi;
908 struct md_rdev *rdev, *rrdev = NULL;
59fc630b 909
910 sh = head_sh;
e9c7469b
TH		 911 if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) {
912 if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags))
913 rw = WRITE_FUA;
914 else
915 rw = WRITE;
9e444768 916 if (test_bit(R5_Discard, &sh->dev[i].flags))
620125f2 917 rw |= REQ_DISCARD;
e9c7469b 918 } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags))
91c00924 919 rw = READ;
9a3e1101
N		 920 else if (test_and_clear_bit(R5_WantReplace,
921 &sh->dev[i].flags)) {
922 rw = WRITE;
923 replace_only = 1;
924 } else
91c00924 925 continue;
bc0934f0
SL		 926 if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags))
927 rw |= REQ_SYNC;
91c00924 928
59fc630b 929again:
91c00924 930 bi = &sh->dev[i].req;
977df362 931 rbi = &sh->dev[i].rreq; /* For writing to replacement */
91c00924 932
91c00924 933 rcu_read_lock();
9a3e1101 934 rrdev = rcu_dereference(conf->disks[i].replacement);
dd054fce
N		 935 smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */
936 rdev = rcu_dereference(conf->disks[i].rdev);
937 if (!rdev) {
938 rdev = rrdev;
939 rrdev = NULL;
940 }
9a3e1101
N		 941 if (rw & WRITE) {
942 if (replace_only)
943 rdev = NULL;
dd054fce
N		 944 if (rdev == rrdev)
945 /* We raced and saw duplicates */
946 rrdev = NULL;
9a3e1101 947 } else {
59fc630b 948 if (test_bit(R5_ReadRepl, &head_sh->dev[i].flags) && rrdev)
9a3e1101
N		 949 rdev = rrdev;
950 rrdev = NULL;
951 }
977df362 952
91c00924
DW		 953 if (rdev && test_bit(Faulty, &rdev->flags))
954 rdev = NULL;
955 if (rdev)
956 atomic_inc(&rdev->nr_pending);
977df362
N		 957 if (rrdev && test_bit(Faulty, &rrdev->flags))
958 rrdev = NULL;
959 if (rrdev)
960 atomic_inc(&rrdev->nr_pending);
91c00924
DW		 961 rcu_read_unlock();
962
73e92e51 963 /* We have already checked bad blocks for reads. Now
977df362
N		 964 * need to check for writes. We never accept write errors
965 * on the replacement, so we don't to check rrdev.
73e92e51
N		 966 */
967 while ((rw & WRITE) && rdev &&
968 test_bit(WriteErrorSeen, &rdev->flags)) {
969 sector_t first_bad;
970 int bad_sectors;
971 int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
972 &first_bad, &bad_sectors);
973 if (!bad)
974 break;
975
976 if (bad < 0) {
977 set_bit(BlockedBadBlocks, &rdev->flags);
978 if (!conf->mddev->external &&
979 conf->mddev->flags) {
980 /* It is very unlikely, but we might
981 * still need to write out the
982 * bad block log - better give it
983 * a chance*/
984 md_check_recovery(conf->mddev);
985 }
1850753d 986 /*
987 * Because md_wait_for_blocked_rdev
988 * will dec nr_pending, we must
989 * increment it first.
990 */
991 atomic_inc(&rdev->nr_pending);
73e92e51
N		 992 md_wait_for_blocked_rdev(rdev, conf->mddev);
993 } else {
994 /* Acknowledged bad block - skip the write */
995 rdev_dec_pending(rdev, conf->mddev);
996 rdev = NULL;
997 }
998 }
999
91c00924 1000 if (rdev) {
9a3e1101
N		 1001 if (s->syncing || s->expanding || s->expanded
1002 || s->replacing)
91c00924
DW		 1003 md_sync_acct(rdev->bdev, STRIPE_SECTORS);
1004
2b7497f0
DW		 1005 set_bit(STRIPE_IO_STARTED, &sh->state);
1006
2f6db2a7 1007 bio_reset(bi);
91c00924 1008 bi->bi_bdev = rdev->bdev;
2f6db2a7
KO		 1009 bi->bi_rw = rw;
1010 bi->bi_end_io = (rw & WRITE)
1011 ? raid5_end_write_request
1012 : raid5_end_read_request;
1013 bi->bi_private = sh;
1014
91c00924 1015 pr_debug("%s: for %llu schedule op %ld on disc %d\n",
e46b272b 1016 __func__, (unsigned long long)sh->sector,
91c00924
DW		 1017 bi->bi_rw, i);
1018 atomic_inc(&sh->count);
59fc630b 1019 if (sh != head_sh)
1020 atomic_inc(&head_sh->count);
05616be5 1021 if (use_new_offset(conf, sh))
4f024f37 1022 bi->bi_iter.bi_sector = (sh->sector
05616be5
N		 1023 + rdev->new_data_offset);
1024 else
4f024f37 1025 bi->bi_iter.bi_sector = (sh->sector
05616be5 1026 + rdev->data_offset);
59fc630b 1027 if (test_bit(R5_ReadNoMerge, &head_sh->dev[i].flags))
e59aa23f 1028 bi->bi_rw |= REQ_NOMERGE;
3f9e7c14 1029
d592a996
SL		 1030 if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
1031 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
1032 sh->dev[i].vec.bv_page = sh->dev[i].page;
4997b72e 1033 bi->bi_vcnt = 1;
91c00924
DW		 1034 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
1035 bi->bi_io_vec[0].bv_offset = 0;
4f024f37 1036 bi->bi_iter.bi_size = STRIPE_SIZE;
37c61ff3
SL		 1037 /*
1038 * If this is discard request, set bi_vcnt 0. We don't
1039 * want to confuse SCSI because SCSI will replace payload
1040 */
1041 if (rw & REQ_DISCARD)
1042 bi->bi_vcnt = 0;
977df362
N		 1043 if (rrdev)
1044 set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags);
e3620a3a
JB		 1045
1046 if (conf->mddev->gendisk)
1047 trace_block_bio_remap(bdev_get_queue(bi->bi_bdev),
1048 bi, disk_devt(conf->mddev->gendisk),
1049 sh->dev[i].sector);
91c00924 1050 generic_make_request(bi);
977df362
N		 1051 }
1052 if (rrdev) {
9a3e1101
N		 1053 if (s->syncing || s->expanding || s->expanded
1054 || s->replacing)
977df362
N		 1055 md_sync_acct(rrdev->bdev, STRIPE_SECTORS);
1056
1057 set_bit(STRIPE_IO_STARTED, &sh->state);
1058
2f6db2a7 1059 bio_reset(rbi);
977df362 1060 rbi->bi_bdev = rrdev->bdev;
2f6db2a7
KO		 1061 rbi->bi_rw = rw;
1062 BUG_ON(!(rw & WRITE));
1063 rbi->bi_end_io = raid5_end_write_request;
1064 rbi->bi_private = sh;
1065
977df362
N		 1066 pr_debug("%s: for %llu schedule op %ld on "
1067 "replacement disc %d\n",
1068 __func__, (unsigned long long)sh->sector,
1069 rbi->bi_rw, i);
1070 atomic_inc(&sh->count);
59fc630b 1071 if (sh != head_sh)
1072 atomic_inc(&head_sh->count);
05616be5 1073 if (use_new_offset(conf, sh))
4f024f37 1074 rbi->bi_iter.bi_sector = (sh->sector
05616be5
N		 1075 + rrdev->new_data_offset);
1076 else
4f024f37 1077 rbi->bi_iter.bi_sector = (sh->sector
05616be5 1078 + rrdev->data_offset);
d592a996
SL		 1079 if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
1080 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
1081 sh->dev[i].rvec.bv_page = sh->dev[i].page;
4997b72e 1082 rbi->bi_vcnt = 1;
977df362
N		 1083 rbi->bi_io_vec[0].bv_len = STRIPE_SIZE;
1084 rbi->bi_io_vec[0].bv_offset = 0;
4f024f37 1085 rbi->bi_iter.bi_size = STRIPE_SIZE;
37c61ff3
SL		 1086 /*
1087 * If this is discard request, set bi_vcnt 0. We don't
1088 * want to confuse SCSI because SCSI will replace payload
1089 */
1090 if (rw & REQ_DISCARD)
1091 rbi->bi_vcnt = 0;
e3620a3a
JB		 1092 if (conf->mddev->gendisk)
1093 trace_block_bio_remap(bdev_get_queue(rbi->bi_bdev),
1094 rbi, disk_devt(conf->mddev->gendisk),
1095 sh->dev[i].sector);
977df362
N		 1096 generic_make_request(rbi);
1097 }
1098 if (!rdev && !rrdev) {
b062962e 1099 if (rw & WRITE)
91c00924
DW		 1100 set_bit(STRIPE_DEGRADED, &sh->state);
1101 pr_debug("skip op %ld on disc %d for sector %llu\n",
1102 bi->bi_rw, i, (unsigned long long)sh->sector);
1103 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1104 set_bit(STRIPE_HANDLE, &sh->state);
1105 }
59fc630b 1106
1107 if (!head_sh->batch_head)
1108 continue;
1109 sh = list_first_entry(&sh->batch_list, struct stripe_head,
1110 batch_list);
1111 if (sh != head_sh)
1112 goto again;
91c00924
DW		 1113 }
1114}
1115
1116static struct dma_async_tx_descriptor *
d592a996
SL		 1117async_copy_data(int frombio, struct bio *bio, struct page **page,
1118 sector_t sector, struct dma_async_tx_descriptor *tx,
1119 struct stripe_head *sh)
91c00924 1120{
7988613b
KO		 1121 struct bio_vec bvl;
1122 struct bvec_iter iter;
91c00924 1123 struct page *bio_page;
91c00924 1124 int page_offset;
a08abd8c 1125 struct async_submit_ctl submit;
0403e382 1126 enum async_tx_flags flags = 0;
91c00924 1127
4f024f37
KO		 1128 if (bio->bi_iter.bi_sector >= sector)
1129 page_offset = (signed)(bio->bi_iter.bi_sector - sector) * 512;
91c00924 1130 else
4f024f37 1131 page_offset = (signed)(sector - bio->bi_iter.bi_sector) * -512;
a08abd8c 1132
0403e382
DW		 1133 if (frombio)
1134 flags |= ASYNC_TX_FENCE;
1135 init_async_submit(&submit, flags, tx, NULL, NULL, NULL);
1136
7988613b
KO		 1137 bio_for_each_segment(bvl, bio, iter) {
1138 int len = bvl.bv_len;
91c00924
DW		 1139 int clen;
1140 int b_offset = 0;
1141
1142 if (page_offset < 0) {
1143 b_offset = -page_offset;
1144 page_offset += b_offset;
1145 len -= b_offset;
1146 }
1147
1148 if (len > 0 && page_offset + len > STRIPE_SIZE)
1149 clen = STRIPE_SIZE - page_offset;
1150 else
1151 clen = len;
1152
1153 if (clen > 0) {
7988613b
KO		 1154 b_offset += bvl.bv_offset;
1155 bio_page = bvl.bv_page;
d592a996
SL		 1156 if (frombio) {
1157 if (sh->raid_conf->skip_copy &&
1158 b_offset == 0 && page_offset == 0 &&
1159 clen == STRIPE_SIZE)
1160 *page = bio_page;
1161 else
1162 tx = async_memcpy(*page, bio_page, page_offset,
a08abd8c 1163 b_offset, clen, &submit);
d592a996
SL		 1164 } else
1165 tx = async_memcpy(bio_page, *page, b_offset,
a08abd8c 1166 page_offset, clen, &submit);
91c00924 1167 }
a08abd8c
DW		 1168 /* chain the operations */
1169 submit.depend_tx = tx;
1170
91c00924
DW		 1171 if (clen < len) /* hit end of page */
1172 break;
1173 page_offset += len;
1174 }
1175
1176 return tx;
1177}
1178
1179static void ops_complete_biofill(void *stripe_head_ref)
1180{
1181 struct stripe_head *sh = stripe_head_ref;
34a6f80e 1182 struct bio_list return_bi = BIO_EMPTY_LIST;
e4d84909 1183 int i;
91c00924 1184
e46b272b 1185 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1186 (unsigned long long)sh->sector);
1187
1188 /* clear completed biofills */
1189 for (i = sh->disks; i--; ) {
1190 struct r5dev *dev = &sh->dev[i];
91c00924
DW		 1191
1192 /* acknowledge completion of a biofill operation */
e4d84909
DW		 1193 /* and check if we need to reply to a read request,
1194 * new R5_Wantfill requests are held off until
83de75cc 1195 * !STRIPE_BIOFILL_RUN
e4d84909
DW		 1196 */
1197 if (test_and_clear_bit(R5_Wantfill, &dev->flags)) {
91c00924 1198 struct bio *rbi, *rbi2;
91c00924 1199
91c00924
DW		 1200 BUG_ON(!dev->read);
1201 rbi = dev->read;
1202 dev->read = NULL;
4f024f37 1203 while (rbi && rbi->bi_iter.bi_sector <
91c00924
DW		 1204 dev->sector + STRIPE_SECTORS) {
1205 rbi2 = r5_next_bio(rbi, dev->sector);
34a6f80e
N		 1206 if (!raid5_dec_bi_active_stripes(rbi))
1207 bio_list_add(&return_bi, rbi);
91c00924
DW		 1208 rbi = rbi2;
1209 }
1210 }
1211 }
83de75cc 1212 clear_bit(STRIPE_BIOFILL_RUN, &sh->state);
91c00924 1213
34a6f80e 1214 return_io(&return_bi);
91c00924 1215
e4d84909 1216 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 1217 raid5_release_stripe(sh);
91c00924
DW		 1218}
1219
1220static void ops_run_biofill(struct stripe_head *sh)
1221{
1222 struct dma_async_tx_descriptor *tx = NULL;
a08abd8c 1223 struct async_submit_ctl submit;
91c00924
DW		 1224 int i;
1225
59fc630b 1226 BUG_ON(sh->batch_head);
e46b272b 1227 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1228 (unsigned long long)sh->sector);
1229
1230 for (i = sh->disks; i--; ) {
1231 struct r5dev *dev = &sh->dev[i];
1232 if (test_bit(R5_Wantfill, &dev->flags)) {
1233 struct bio *rbi;
b17459c0 1234 spin_lock_irq(&sh->stripe_lock);
91c00924
DW		 1235 dev->read = rbi = dev->toread;
1236 dev->toread = NULL;
b17459c0 1237 spin_unlock_irq(&sh->stripe_lock);
4f024f37 1238 while (rbi && rbi->bi_iter.bi_sector <
91c00924 1239 dev->sector + STRIPE_SECTORS) {
d592a996
SL		 1240 tx = async_copy_data(0, rbi, &dev->page,
1241 dev->sector, tx, sh);
91c00924
DW		 1242 rbi = r5_next_bio(rbi, dev->sector);
1243 }
1244 }
1245 }
1246
1247 atomic_inc(&sh->count);
a08abd8c
DW		 1248 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL);
1249 async_trigger_callback(&submit);
91c00924
DW		 1250}
1251
4e7d2c0a 1252static void mark_target_uptodate(struct stripe_head *sh, int target)
91c00924 1253{
4e7d2c0a 1254 struct r5dev *tgt;
91c00924 1255
4e7d2c0a
DW		 1256 if (target < 0)
1257 return;
91c00924 1258
4e7d2c0a 1259 tgt = &sh->dev[target];
91c00924
DW		 1260 set_bit(R5_UPTODATE, &tgt->flags);
1261 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1262 clear_bit(R5_Wantcompute, &tgt->flags);
4e7d2c0a
DW		 1263}
1264
ac6b53b6 1265static void ops_complete_compute(void *stripe_head_ref)
91c00924
DW		 1266{
1267 struct stripe_head *sh = stripe_head_ref;
91c00924 1268
e46b272b 1269 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1270 (unsigned long long)sh->sector);
1271
ac6b53b6 1272 /* mark the computed target(s) as uptodate */
4e7d2c0a 1273 mark_target_uptodate(sh, sh->ops.target);
ac6b53b6 1274 mark_target_uptodate(sh, sh->ops.target2);
4e7d2c0a 1275
ecc65c9b
DW		 1276 clear_bit(STRIPE_COMPUTE_RUN, &sh->state);
1277 if (sh->check_state == check_state_compute_run)
1278 sh->check_state = check_state_compute_result;
91c00924 1279 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 1280 raid5_release_stripe(sh);
91c00924
DW		 1281}
1282
d6f38f31
DW		 1283/* return a pointer to the address conversion region of the scribble buffer */
1284static addr_conv_t *to_addr_conv(struct stripe_head *sh,
46d5b785 1285 struct raid5_percpu *percpu, int i)
d6f38f31 1286{
46d5b785 1287 void *addr;
1288
1289 addr = flex_array_get(percpu->scribble, i);
1290 return addr + sizeof(struct page *) * (sh->disks + 2);
1291}
1292
1293/* return a pointer to the address conversion region of the scribble buffer */
1294static struct page **to_addr_page(struct raid5_percpu *percpu, int i)
1295{
1296 void *addr;
1297
1298 addr = flex_array_get(percpu->scribble, i);
1299 return addr;
d6f38f31
DW		 1300}
1301
1302static struct dma_async_tx_descriptor *
1303ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 1304{
91c00924 1305 int disks = sh->disks;
46d5b785 1306 struct page **xor_srcs = to_addr_page(percpu, 0);
91c00924
DW		 1307 int target = sh->ops.target;
1308 struct r5dev *tgt = &sh->dev[target];
1309 struct page *xor_dest = tgt->page;
1310 int count = 0;
1311 struct dma_async_tx_descriptor *tx;
a08abd8c 1312 struct async_submit_ctl submit;
91c00924
DW		 1313 int i;
1314
59fc630b 1315 BUG_ON(sh->batch_head);
1316
91c00924 1317 pr_debug("%s: stripe %llu block: %d\n",
e46b272b 1318 __func__, (unsigned long long)sh->sector, target);
91c00924
DW		 1319 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1320
1321 for (i = disks; i--; )
1322 if (i != target)
1323 xor_srcs[count++] = sh->dev[i].page;
1324
1325 atomic_inc(&sh->count);
1326
0403e382 1327 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL,
46d5b785 1328 ops_complete_compute, sh, to_addr_conv(sh, percpu, 0));
91c00924 1329 if (unlikely(count == 1))
a08abd8c 1330 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
91c00924 1331 else
a08abd8c 1332 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924 1333
91c00924
DW		 1334 return tx;
1335}
1336
ac6b53b6
DW		 1337/* set_syndrome_sources - populate source buffers for gen_syndrome
1338 * @srcs - (struct page *) array of size sh->disks
1339 * @sh - stripe_head to parse
1340 *
1341 * Populates srcs in proper layout order for the stripe and returns the
1342 * 'count' of sources to be used in a call to async_gen_syndrome. The P
1343 * destination buffer is recorded in srcs[count] and the Q destination
1344 * is recorded in srcs[count+1]].
1345 */
584acdd4
MS		 1346static int set_syndrome_sources(struct page **srcs,
1347 struct stripe_head *sh,
1348 int srctype)
ac6b53b6
DW		 1349{
1350 int disks = sh->disks;
1351 int syndrome_disks = sh->ddf_layout ? disks : (disks - 2);
1352 int d0_idx = raid6_d0(sh);
1353 int count;
1354 int i;
1355
1356 for (i = 0; i < disks; i++)
5dd33c9a 1357 srcs[i] = NULL;
ac6b53b6
DW		 1358
1359 count = 0;
1360 i = d0_idx;
1361 do {
1362 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
584acdd4 1363 struct r5dev *dev = &sh->dev[i];
ac6b53b6 1364
584acdd4
MS		 1365 if (i == sh->qd_idx || i == sh->pd_idx ||
1366 (srctype == SYNDROME_SRC_ALL) ||
1367 (srctype == SYNDROME_SRC_WANT_DRAIN &&
1368 test_bit(R5_Wantdrain, &dev->flags)) ||
1369 (srctype == SYNDROME_SRC_WRITTEN &&
1370 dev->written))
1371 srcs[slot] = sh->dev[i].page;
ac6b53b6
DW		 1372 i = raid6_next_disk(i, disks);
1373 } while (i != d0_idx);
ac6b53b6 1374
e4424fee 1375 return syndrome_disks;
ac6b53b6
DW		 1376}
1377
1378static struct dma_async_tx_descriptor *
1379ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu)
1380{
1381 int disks = sh->disks;
46d5b785 1382 struct page **blocks = to_addr_page(percpu, 0);
ac6b53b6
DW		 1383 int target;
1384 int qd_idx = sh->qd_idx;
1385 struct dma_async_tx_descriptor *tx;
1386 struct async_submit_ctl submit;
1387 struct r5dev *tgt;
1388 struct page *dest;
1389 int i;
1390 int count;
1391
59fc630b 1392 BUG_ON(sh->batch_head);
ac6b53b6
DW		 1393 if (sh->ops.target < 0)
1394 target = sh->ops.target2;
1395 else if (sh->ops.target2 < 0)
1396 target = sh->ops.target;
91c00924 1397 else
ac6b53b6
DW		 1398 /* we should only have one valid target */
1399 BUG();
1400 BUG_ON(target < 0);
1401 pr_debug("%s: stripe %llu block: %d\n",
1402 __func__, (unsigned long long)sh->sector, target);
1403
1404 tgt = &sh->dev[target];
1405 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1406 dest = tgt->page;
1407
1408 atomic_inc(&sh->count);
1409
1410 if (target == qd_idx) {
584acdd4 1411 count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL);
ac6b53b6
DW		 1412 blocks[count] = NULL; /* regenerating p is not necessary */
1413 BUG_ON(blocks[count+1] != dest); /* q should already be set */
0403e382
DW		 1414 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1415 ops_complete_compute, sh,
46d5b785 1416 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1417 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
1418 } else {
1419 /* Compute any data- or p-drive using XOR */
1420 count = 0;
1421 for (i = disks; i-- ; ) {
1422 if (i == target || i == qd_idx)
1423 continue;
1424 blocks[count++] = sh->dev[i].page;
1425 }
1426
0403e382
DW		 1427 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1428 NULL, ops_complete_compute, sh,
46d5b785 1429 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1430 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit);
1431 }
91c00924 1432
91c00924
DW		 1433 return tx;
1434}
1435
ac6b53b6
DW		 1436static struct dma_async_tx_descriptor *
1437ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu)
1438{
1439 int i, count, disks = sh->disks;
1440 int syndrome_disks = sh->ddf_layout ? disks : disks-2;
1441 int d0_idx = raid6_d0(sh);
1442 int faila = -1, failb = -1;
1443 int target = sh->ops.target;
1444 int target2 = sh->ops.target2;
1445 struct r5dev *tgt = &sh->dev[target];
1446 struct r5dev *tgt2 = &sh->dev[target2];
1447 struct dma_async_tx_descriptor *tx;
46d5b785 1448 struct page **blocks = to_addr_page(percpu, 0);
ac6b53b6
DW		 1449 struct async_submit_ctl submit;
1450
59fc630b 1451 BUG_ON(sh->batch_head);
ac6b53b6
DW		 1452 pr_debug("%s: stripe %llu block1: %d block2: %d\n",
1453 __func__, (unsigned long long)sh->sector, target, target2);
1454 BUG_ON(target < 0 || target2 < 0);
1455 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1456 BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags));
1457
6c910a78 1458 /* we need to open-code set_syndrome_sources to handle the
ac6b53b6
DW		 1459 * slot number conversion for 'faila' and 'failb'
1460 */
1461 for (i = 0; i < disks ; i++)
5dd33c9a 1462 blocks[i] = NULL;
ac6b53b6
DW		 1463 count = 0;
1464 i = d0_idx;
1465 do {
1466 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
1467
1468 blocks[slot] = sh->dev[i].page;
1469
1470 if (i == target)
1471 faila = slot;
1472 if (i == target2)
1473 failb = slot;
1474 i = raid6_next_disk(i, disks);
1475 } while (i != d0_idx);
ac6b53b6
DW		 1476
1477 BUG_ON(faila == failb);
1478 if (failb < faila)
1479 swap(faila, failb);
1480 pr_debug("%s: stripe: %llu faila: %d failb: %d\n",
1481 __func__, (unsigned long long)sh->sector, faila, failb);
1482
1483 atomic_inc(&sh->count);
1484
1485 if (failb == syndrome_disks+1) {
1486 /* Q disk is one of the missing disks */
1487 if (faila == syndrome_disks) {
1488 /* Missing P+Q, just recompute */
0403e382
DW		 1489 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1490 ops_complete_compute, sh,
46d5b785 1491 to_addr_conv(sh, percpu, 0));
e4424fee 1492 return async_gen_syndrome(blocks, 0, syndrome_disks+2,
ac6b53b6
DW		 1493 STRIPE_SIZE, &submit);
1494 } else {
1495 struct page *dest;
1496 int data_target;
1497 int qd_idx = sh->qd_idx;
1498
1499 /* Missing D+Q: recompute D from P, then recompute Q */
1500 if (target == qd_idx)
1501 data_target = target2;
1502 else
1503 data_target = target;
1504
1505 count = 0;
1506 for (i = disks; i-- ; ) {
1507 if (i == data_target || i == qd_idx)
1508 continue;
1509 blocks[count++] = sh->dev[i].page;
1510 }
1511 dest = sh->dev[data_target].page;
0403e382
DW		 1512 init_async_submit(&submit,
1513 ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1514 NULL, NULL, NULL,
46d5b785 1515 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1516 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE,
1517 &submit);
1518
584acdd4 1519 count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL);
0403e382
DW		 1520 init_async_submit(&submit, ASYNC_TX_FENCE, tx,
1521 ops_complete_compute, sh,
46d5b785 1522 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1523 return async_gen_syndrome(blocks, 0, count+2,
1524 STRIPE_SIZE, &submit);
1525 }
ac6b53b6 1526 } else {
6c910a78
DW		 1527 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1528 ops_complete_compute, sh,
46d5b785 1529 to_addr_conv(sh, percpu, 0));
6c910a78
DW		 1530 if (failb == syndrome_disks) {
1531 /* We're missing D+P. */
1532 return async_raid6_datap_recov(syndrome_disks+2,
1533 STRIPE_SIZE, faila,
1534 blocks, &submit);
1535 } else {
1536 /* We're missing D+D. */
1537 return async_raid6_2data_recov(syndrome_disks+2,
1538 STRIPE_SIZE, faila, failb,
1539 blocks, &submit);
1540 }
ac6b53b6
DW		 1541 }
1542}
1543
91c00924
DW		 1544static void ops_complete_prexor(void *stripe_head_ref)
1545{
1546 struct stripe_head *sh = stripe_head_ref;
1547
e46b272b 1548 pr_debug("%s: stripe %llu\n", __func__,
91c00924 1549 (unsigned long long)sh->sector);
91c00924
DW		 1550}
1551
1552static struct dma_async_tx_descriptor *
584acdd4
MS		 1553ops_run_prexor5(struct stripe_head *sh, struct raid5_percpu *percpu,
1554 struct dma_async_tx_descriptor *tx)
91c00924 1555{
91c00924 1556 int disks = sh->disks;
46d5b785 1557 struct page **xor_srcs = to_addr_page(percpu, 0);
91c00924 1558 int count = 0, pd_idx = sh->pd_idx, i;
a08abd8c 1559 struct async_submit_ctl submit;
91c00924
DW		 1560
1561 /* existing parity data subtracted */
1562 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1563
59fc630b 1564 BUG_ON(sh->batch_head);
e46b272b 1565 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1566 (unsigned long long)sh->sector);
1567
1568 for (i = disks; i--; ) {
1569 struct r5dev *dev = &sh->dev[i];
1570 /* Only process blocks that are known to be uptodate */
d8ee0728 1571 if (test_bit(R5_Wantdrain, &dev->flags))
91c00924
DW		 1572 xor_srcs[count++] = dev->page;
1573 }
1574
0403e382 1575 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
46d5b785 1576 ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0));
a08abd8c 1577 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924
DW		 1578
1579 return tx;
1580}
1581
584acdd4
MS		 1582static struct dma_async_tx_descriptor *
1583ops_run_prexor6(struct stripe_head *sh, struct raid5_percpu *percpu,
1584 struct dma_async_tx_descriptor *tx)
1585{
1586 struct page **blocks = to_addr_page(percpu, 0);
1587 int count;
1588 struct async_submit_ctl submit;
1589
1590 pr_debug("%s: stripe %llu\n", __func__,
1591 (unsigned long long)sh->sector);
1592
1593 count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_WANT_DRAIN);
1594
1595 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_PQ_XOR_DST, tx,
1596 ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0));
1597 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
1598
1599 return tx;
1600}
1601
91c00924 1602static struct dma_async_tx_descriptor *
d8ee0728 1603ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
91c00924
DW		 1604{
1605 int disks = sh->disks;
d8ee0728 1606 int i;
59fc630b 1607 struct stripe_head *head_sh = sh;
91c00924 1608
e46b272b 1609 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1610 (unsigned long long)sh->sector);
1611
1612 for (i = disks; i--; ) {
59fc630b 1613 struct r5dev *dev;
91c00924 1614 struct bio *chosen;
91c00924 1615
59fc630b 1616 sh = head_sh;
1617 if (test_and_clear_bit(R5_Wantdrain, &head_sh->dev[i].flags)) {
91c00924
DW		 1618 struct bio *wbi;
1619
59fc630b 1620again:
1621 dev = &sh->dev[i];
b17459c0 1622 spin_lock_irq(&sh->stripe_lock);
91c00924
DW		 1623 chosen = dev->towrite;
1624 dev->towrite = NULL;
7a87f434 1625 sh->overwrite_disks = 0;
91c00924
DW		 1626 BUG_ON(dev->written);
1627 wbi = dev->written = chosen;
b17459c0 1628 spin_unlock_irq(&sh->stripe_lock);
d592a996 1629 WARN_ON(dev->page != dev->orig_page);
91c00924 1630
4f024f37 1631 while (wbi && wbi->bi_iter.bi_sector <
91c00924 1632 dev->sector + STRIPE_SECTORS) {
e9c7469b
TH		 1633 if (wbi->bi_rw & REQ_FUA)
1634 set_bit(R5_WantFUA, &dev->flags);
bc0934f0
SL		 1635 if (wbi->bi_rw & REQ_SYNC)
1636 set_bit(R5_SyncIO, &dev->flags);
9e444768 1637 if (wbi->bi_rw & REQ_DISCARD)
620125f2 1638 set_bit(R5_Discard, &dev->flags);
d592a996
SL		 1639 else {
1640 tx = async_copy_data(1, wbi, &dev->page,
1641 dev->sector, tx, sh);
1642 if (dev->page != dev->orig_page) {
1643 set_bit(R5_SkipCopy, &dev->flags);
1644 clear_bit(R5_UPTODATE, &dev->flags);
1645 clear_bit(R5_OVERWRITE, &dev->flags);
1646 }
1647 }
91c00924
DW		 1648 wbi = r5_next_bio(wbi, dev->sector);
1649 }
59fc630b 1650
1651 if (head_sh->batch_head) {
1652 sh = list_first_entry(&sh->batch_list,
1653 struct stripe_head,
1654 batch_list);
1655 if (sh == head_sh)
1656 continue;
1657 goto again;
1658 }
91c00924
DW		 1659 }
1660 }
1661
1662 return tx;
1663}
1664
ac6b53b6 1665static void ops_complete_reconstruct(void *stripe_head_ref)
91c00924
DW		 1666{
1667 struct stripe_head *sh = stripe_head_ref;
ac6b53b6
DW		 1668 int disks = sh->disks;
1669 int pd_idx = sh->pd_idx;
1670 int qd_idx = sh->qd_idx;
1671 int i;
9e444768 1672 bool fua = false, sync = false, discard = false;
91c00924 1673
e46b272b 1674 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1675 (unsigned long long)sh->sector);
1676
bc0934f0 1677 for (i = disks; i--; ) {
e9c7469b 1678 fua |= test_bit(R5_WantFUA, &sh->dev[i].flags);
bc0934f0 1679 sync |= test_bit(R5_SyncIO, &sh->dev[i].flags);
9e444768 1680 discard |= test_bit(R5_Discard, &sh->dev[i].flags);
bc0934f0 1681 }
e9c7469b 1682
91c00924
DW		 1683 for (i = disks; i--; ) {
1684 struct r5dev *dev = &sh->dev[i];
ac6b53b6 1685
e9c7469b 1686 if (dev->written || i == pd_idx || i == qd_idx) {
d592a996 1687 if (!discard && !test_bit(R5_SkipCopy, &dev->flags))
9e444768 1688 set_bit(R5_UPTODATE, &dev->flags);
e9c7469b
TH		 1689 if (fua)
1690 set_bit(R5_WantFUA, &dev->flags);
bc0934f0
SL		 1691 if (sync)
1692 set_bit(R5_SyncIO, &dev->flags);
e9c7469b 1693 }
91c00924
DW		 1694 }
1695
d8ee0728
DW		 1696 if (sh->reconstruct_state == reconstruct_state_drain_run)
1697 sh->reconstruct_state = reconstruct_state_drain_result;
1698 else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run)
1699 sh->reconstruct_state = reconstruct_state_prexor_drain_result;
1700 else {
1701 BUG_ON(sh->reconstruct_state != reconstruct_state_run);
1702 sh->reconstruct_state = reconstruct_state_result;
1703 }
91c00924
DW		 1704
1705 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 1706 raid5_release_stripe(sh);
91c00924
DW		 1707}
1708
1709static void
ac6b53b6
DW		 1710ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu,
1711 struct dma_async_tx_descriptor *tx)
91c00924 1712{
91c00924 1713 int disks = sh->disks;
59fc630b 1714 struct page **xor_srcs;
a08abd8c 1715 struct async_submit_ctl submit;
59fc630b 1716 int count, pd_idx = sh->pd_idx, i;
91c00924 1717 struct page *xor_dest;
d8ee0728 1718 int prexor = 0;
91c00924 1719 unsigned long flags;
59fc630b 1720 int j = 0;
1721 struct stripe_head *head_sh = sh;
1722 int last_stripe;
91c00924 1723
e46b272b 1724 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1725 (unsigned long long)sh->sector);
1726
620125f2
SL		 1727 for (i = 0; i < sh->disks; i++) {
1728 if (pd_idx == i)
1729 continue;
1730 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1731 break;
1732 }
1733 if (i >= sh->disks) {
1734 atomic_inc(&sh->count);
620125f2
SL		 1735 set_bit(R5_Discard, &sh->dev[pd_idx].flags);
1736 ops_complete_reconstruct(sh);
1737 return;
1738 }
59fc630b 1739again:
1740 count = 0;
1741 xor_srcs = to_addr_page(percpu, j);
91c00924
DW		 1742 /* check if prexor is active which means only process blocks
1743 * that are part of a read-modify-write (written)
1744 */
59fc630b 1745 if (head_sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
d8ee0728 1746 prexor = 1;
91c00924
DW		 1747 xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1748 for (i = disks; i--; ) {
1749 struct r5dev *dev = &sh->dev[i];
59fc630b 1750 if (head_sh->dev[i].written)
91c00924
DW		 1751 xor_srcs[count++] = dev->page;
1752 }
1753 } else {
1754 xor_dest = sh->dev[pd_idx].page;
1755 for (i = disks; i--; ) {
1756 struct r5dev *dev = &sh->dev[i];
1757 if (i != pd_idx)
1758 xor_srcs[count++] = dev->page;
1759 }
1760 }
1761
91c00924
DW		 1762 /* 1/ if we prexor'd then the dest is reused as a source
1763 * 2/ if we did not prexor then we are redoing the parity
1764 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
1765 * for the synchronous xor case
1766 */
59fc630b 1767 last_stripe = !head_sh->batch_head ||
1768 list_first_entry(&sh->batch_list,
1769 struct stripe_head, batch_list) == head_sh;
1770 if (last_stripe) {
1771 flags = ASYNC_TX_ACK |
1772 (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
1773
1774 atomic_inc(&head_sh->count);
1775 init_async_submit(&submit, flags, tx, ops_complete_reconstruct, head_sh,
1776 to_addr_conv(sh, percpu, j));
1777 } else {
1778 flags = prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST;
1779 init_async_submit(&submit, flags, tx, NULL, NULL,
1780 to_addr_conv(sh, percpu, j));
1781 }
91c00924 1782
a08abd8c
DW		 1783 if (unlikely(count == 1))
1784 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
1785 else
1786 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
59fc630b 1787 if (!last_stripe) {
1788 j++;
1789 sh = list_first_entry(&sh->batch_list, struct stripe_head,
1790 batch_list);
1791 goto again;
1792 }
91c00924
DW		 1793}
1794
ac6b53b6
DW		 1795static void
1796ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu,
1797 struct dma_async_tx_descriptor *tx)
1798{
1799 struct async_submit_ctl submit;
59fc630b 1800 struct page **blocks;
1801 int count, i, j = 0;
1802 struct stripe_head *head_sh = sh;
1803 int last_stripe;
584acdd4
MS		 1804 int synflags;
1805 unsigned long txflags;
ac6b53b6
DW		 1806
1807 pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector);
1808
620125f2
SL		 1809 for (i = 0; i < sh->disks; i++) {
1810 if (sh->pd_idx == i || sh->qd_idx == i)
1811 continue;
1812 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1813 break;
1814 }
1815 if (i >= sh->disks) {
1816 atomic_inc(&sh->count);
620125f2
SL		 1817 set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
1818 set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
1819 ops_complete_reconstruct(sh);
1820 return;
1821 }
1822
59fc630b 1823again:
1824 blocks = to_addr_page(percpu, j);
584acdd4
MS		 1825
1826 if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
1827 synflags = SYNDROME_SRC_WRITTEN;
1828 txflags = ASYNC_TX_ACK | ASYNC_TX_PQ_XOR_DST;
1829 } else {
1830 synflags = SYNDROME_SRC_ALL;
1831 txflags = ASYNC_TX_ACK;
1832 }
1833
1834 count = set_syndrome_sources(blocks, sh, synflags);
59fc630b 1835 last_stripe = !head_sh->batch_head ||
1836 list_first_entry(&sh->batch_list,
1837 struct stripe_head, batch_list) == head_sh;
1838
1839 if (last_stripe) {
1840 atomic_inc(&head_sh->count);
584acdd4 1841 init_async_submit(&submit, txflags, tx, ops_complete_reconstruct,
59fc630b 1842 head_sh, to_addr_conv(sh, percpu, j));
1843 } else
1844 init_async_submit(&submit, 0, tx, NULL, NULL,
1845 to_addr_conv(sh, percpu, j));
48769695 1846 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
59fc630b 1847 if (!last_stripe) {
1848 j++;
1849 sh = list_first_entry(&sh->batch_list, struct stripe_head,
1850 batch_list);
1851 goto again;
1852 }
91c00924
DW		 1853}
1854
1855static void ops_complete_check(void *stripe_head_ref)
1856{
1857 struct stripe_head *sh = stripe_head_ref;
91c00924 1858
e46b272b 1859 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1860 (unsigned long long)sh->sector);
1861
ecc65c9b 1862 sh->check_state = check_state_check_result;
91c00924 1863 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 1864 raid5_release_stripe(sh);
91c00924
DW		 1865}
1866
ac6b53b6 1867static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 1868{
91c00924 1869 int disks = sh->disks;
ac6b53b6
DW		 1870 int pd_idx = sh->pd_idx;
1871 int qd_idx = sh->qd_idx;
1872 struct page *xor_dest;
46d5b785 1873 struct page **xor_srcs = to_addr_page(percpu, 0);
91c00924 1874 struct dma_async_tx_descriptor *tx;
a08abd8c 1875 struct async_submit_ctl submit;
ac6b53b6
DW		 1876 int count;
1877 int i;
91c00924 1878
e46b272b 1879 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1880 (unsigned long long)sh->sector);
1881
59fc630b 1882 BUG_ON(sh->batch_head);
ac6b53b6
DW		 1883 count = 0;
1884 xor_dest = sh->dev[pd_idx].page;
1885 xor_srcs[count++] = xor_dest;
91c00924 1886 for (i = disks; i--; ) {
ac6b53b6
DW		 1887 if (i == pd_idx || i == qd_idx)
1888 continue;
1889 xor_srcs[count++] = sh->dev[i].page;
91c00924
DW		 1890 }
1891
d6f38f31 1892 init_async_submit(&submit, 0, NULL, NULL, NULL,
46d5b785 1893 to_addr_conv(sh, percpu, 0));
099f53cb 1894 tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
a08abd8c 1895 &sh->ops.zero_sum_result, &submit);
91c00924 1896
91c00924 1897 atomic_inc(&sh->count);
a08abd8c
DW		 1898 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL);
1899 tx = async_trigger_callback(&submit);
91c00924
DW		 1900}
1901
ac6b53b6
DW		 1902static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp)
1903{
46d5b785 1904 struct page **srcs = to_addr_page(percpu, 0);
ac6b53b6
DW		 1905 struct async_submit_ctl submit;
1906 int count;
1907
1908 pr_debug("%s: stripe %llu checkp: %d\n", __func__,
1909 (unsigned long long)sh->sector, checkp);
1910
59fc630b 1911 BUG_ON(sh->batch_head);
584acdd4 1912 count = set_syndrome_sources(srcs, sh, SYNDROME_SRC_ALL);
ac6b53b6
DW		 1913 if (!checkp)
1914 srcs[count] = NULL;
91c00924 1915
91c00924 1916 atomic_inc(&sh->count);
ac6b53b6 1917 init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check,
46d5b785 1918 sh, to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1919 async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE,
1920 &sh->ops.zero_sum_result, percpu->spare_page, &submit);
91c00924
DW		 1921}
1922
51acbcec 1923static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
91c00924
DW		 1924{
1925 int overlap_clear = 0, i, disks = sh->disks;
1926 struct dma_async_tx_descriptor *tx = NULL;
d1688a6d 1927 struct r5conf *conf = sh->raid_conf;
ac6b53b6 1928 int level = conf->level;
d6f38f31
DW		 1929 struct raid5_percpu *percpu;
1930 unsigned long cpu;
91c00924 1931
d6f38f31
DW		 1932 cpu = get_cpu();
1933 percpu = per_cpu_ptr(conf->percpu, cpu);
83de75cc 1934 if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) {
91c00924
DW		 1935 ops_run_biofill(sh);
1936 overlap_clear++;
1937 }
1938
7b3a871e 1939 if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) {
ac6b53b6
DW		 1940 if (level < 6)
1941 tx = ops_run_compute5(sh, percpu);
1942 else {
1943 if (sh->ops.target2 < 0 || sh->ops.target < 0)
1944 tx = ops_run_compute6_1(sh, percpu);
1945 else
1946 tx = ops_run_compute6_2(sh, percpu);
1947 }
1948 /* terminate the chain if reconstruct is not set to be run */
1949 if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request))
7b3a871e
DW		 1950 async_tx_ack(tx);
1951 }
91c00924 1952
584acdd4
MS		 1953 if (test_bit(STRIPE_OP_PREXOR, &ops_request)) {
1954 if (level < 6)
1955 tx = ops_run_prexor5(sh, percpu, tx);
1956 else
1957 tx = ops_run_prexor6(sh, percpu, tx);
1958 }
91c00924 1959
600aa109 1960 if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) {
d8ee0728 1961 tx = ops_run_biodrain(sh, tx);
91c00924
DW		 1962 overlap_clear++;
1963 }
1964
ac6b53b6
DW		 1965 if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) {
1966 if (level < 6)
1967 ops_run_reconstruct5(sh, percpu, tx);
1968 else
1969 ops_run_reconstruct6(sh, percpu, tx);
1970 }
91c00924 1971
ac6b53b6
DW		 1972 if (test_bit(STRIPE_OP_CHECK, &ops_request)) {
1973 if (sh->check_state == check_state_run)
1974 ops_run_check_p(sh, percpu);
1975 else if (sh->check_state == check_state_run_q)
1976 ops_run_check_pq(sh, percpu, 0);
1977 else if (sh->check_state == check_state_run_pq)
1978 ops_run_check_pq(sh, percpu, 1);
1979 else
1980 BUG();
1981 }
91c00924 1982
59fc630b 1983 if (overlap_clear && !sh->batch_head)
91c00924
DW		 1984 for (i = disks; i--; ) {
1985 struct r5dev *dev = &sh->dev[i];
1986 if (test_and_clear_bit(R5_Overlap, &dev->flags))
1987 wake_up(&sh->raid_conf->wait_for_overlap);
1988 }
d6f38f31 1989 put_cpu();
91c00924
DW		 1990}
1991
f18c1a35
N		 1992static struct stripe_head *alloc_stripe(struct kmem_cache *sc, gfp_t gfp)
1993{
1994 struct stripe_head *sh;
1995
1996 sh = kmem_cache_zalloc(sc, gfp);
1997 if (sh) {
1998 spin_lock_init(&sh->stripe_lock);
1999 spin_lock_init(&sh->batch_lock);
2000 INIT_LIST_HEAD(&sh->batch_list);
2001 INIT_LIST_HEAD(&sh->lru);
2002 atomic_set(&sh->count, 1);
2003 }
2004 return sh;
2005}
486f0644 2006static int grow_one_stripe(struct r5conf *conf, gfp_t gfp)
1da177e4
LT		 2007{
2008 struct stripe_head *sh;
f18c1a35
N		 2009
2010 sh = alloc_stripe(conf->slab_cache, gfp);
3f294f4f
N		 2011 if (!sh)
2012 return 0;
6ce32846 2013
3f294f4f 2014 sh->raid_conf = conf;
3f294f4f 2015
a9683a79 2016 if (grow_buffers(sh, gfp)) {
e4e11e38 2017 shrink_buffers(sh);
3f294f4f
N		 2018 kmem_cache_free(conf->slab_cache, sh);
2019 return 0;
2020 }
486f0644
N		 2021 sh->hash_lock_index =
2022 conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS;
3f294f4f 2023 /* we just created an active stripe so... */
3f294f4f 2024 atomic_inc(&conf->active_stripes);
59fc630b 2025
6d036f7d 2026 raid5_release_stripe(sh);
486f0644 2027 conf->max_nr_stripes++;
3f294f4f
N		 2028 return 1;
2029}
2030
d1688a6d 2031static int grow_stripes(struct r5conf *conf, int num)
3f294f4f 2032{
e18b890b 2033 struct kmem_cache *sc;
5e5e3e78 2034 int devs = max(conf->raid_disks, conf->previous_raid_disks);
1da177e4 2035
f4be6b43
N		 2036 if (conf->mddev->gendisk)
2037 sprintf(conf->cache_name[0],
2038 "raid%d-%s", conf->level, mdname(conf->mddev));
2039 else
2040 sprintf(conf->cache_name[0],
2041 "raid%d-%p", conf->level, conf->mddev);
2042 sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]);
2043
ad01c9e3
N		 2044 conf->active_name = 0;
2045 sc = kmem_cache_create(conf->cache_name[conf->active_name],
1da177e4 2046 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
20c2df83 2047 0, 0, NULL);
1da177e4
LT		 2048 if (!sc)
2049 return 1;
2050 conf->slab_cache = sc;
ad01c9e3 2051 conf->pool_size = devs;
486f0644
N		 2052 while (num--)
2053 if (!grow_one_stripe(conf, GFP_KERNEL))
1da177e4 2054 return 1;
486f0644 2055
1da177e4
LT		 2056 return 0;
2057}
29269553 2058
d6f38f31
DW		 2059/**
2060 * scribble_len - return the required size of the scribble region
2061 * @num - total number of disks in the array
2062 *
2063 * The size must be enough to contain:
2064 * 1/ a struct page pointer for each device in the array +2
2065 * 2/ room to convert each entry in (1) to its corresponding dma
2066 * (dma_map_page()) or page (page_address()) address.
2067 *
2068 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
2069 * calculate over all devices (not just the data blocks), using zeros in place
2070 * of the P and Q blocks.
2071 */
46d5b785 2072static struct flex_array *scribble_alloc(int num, int cnt, gfp_t flags)
d6f38f31 2073{
46d5b785 2074 struct flex_array *ret;
d6f38f31
DW		 2075 size_t len;
2076
2077 len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2);
46d5b785 2078 ret = flex_array_alloc(len, cnt, flags);
2079 if (!ret)
2080 return NULL;
2081 /* always prealloc all elements, so no locking is required */
2082 if (flex_array_prealloc(ret, 0, cnt, flags)) {
2083 flex_array_free(ret);
2084 return NULL;
2085 }
2086 return ret;
d6f38f31
DW		 2087}
2088
738a2738
N		 2089static int resize_chunks(struct r5conf *conf, int new_disks, int new_sectors)
2090{
2091 unsigned long cpu;
2092 int err = 0;
2093
2094 mddev_suspend(conf->mddev);
2095 get_online_cpus();
2096 for_each_present_cpu(cpu) {
2097 struct raid5_percpu *percpu;
2098 struct flex_array *scribble;
2099
2100 percpu = per_cpu_ptr(conf->percpu, cpu);
2101 scribble = scribble_alloc(new_disks,
2102 new_sectors / STRIPE_SECTORS,
2103 GFP_NOIO);
2104
2105 if (scribble) {
2106 flex_array_free(percpu->scribble);
2107 percpu->scribble = scribble;
2108 } else {
2109 err = -ENOMEM;
2110 break;
2111 }
2112 }
2113 put_online_cpus();
2114 mddev_resume(conf->mddev);
2115 return err;
2116}
2117
d1688a6d 2118static int resize_stripes(struct r5conf *conf, int newsize)
ad01c9e3
N		 2119{
2120 /* Make all the stripes able to hold 'newsize' devices.
2121 * New slots in each stripe get 'page' set to a new page.
2122 *
2123 * This happens in stages:
2124 * 1/ create a new kmem_cache and allocate the required number of
2125 * stripe_heads.
83f0d77a 2126 * 2/ gather all the old stripe_heads and transfer the pages across
ad01c9e3
N		 2127 * to the new stripe_heads. This will have the side effect of
2128 * freezing the array as once all stripe_heads have been collected,
2129 * no IO will be possible. Old stripe heads are freed once their
2130 * pages have been transferred over, and the old kmem_cache is
2131 * freed when all stripes are done.
2132 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
2133 * we simple return a failre status - no need to clean anything up.
2134 * 4/ allocate new pages for the new slots in the new stripe_heads.
2135 * If this fails, we don't bother trying the shrink the
2136 * stripe_heads down again, we just leave them as they are.
2137 * As each stripe_head is processed the new one is released into
2138 * active service.
2139 *
2140 * Once step2 is started, we cannot afford to wait for a write,
2141 * so we use GFP_NOIO allocations.
2142 */
2143 struct stripe_head *osh, *nsh;
2144 LIST_HEAD(newstripes);
2145 struct disk_info *ndisks;
b5470dc5 2146 int err;
e18b890b 2147 struct kmem_cache *sc;
ad01c9e3 2148 int i;
566c09c5 2149 int hash, cnt;
ad01c9e3
N		 2150
2151 if (newsize <= conf->pool_size)
2152 return 0; /* never bother to shrink */
2153
b5470dc5
DW		 2154 err = md_allow_write(conf->mddev);
2155 if (err)
2156 return err;
2a2275d6 2157
ad01c9e3
N		 2158 /* Step 1 */
2159 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
2160 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
20c2df83 2161 0, 0, NULL);
ad01c9e3
N		 2162 if (!sc)
2163 return -ENOMEM;
2164
2d5b569b
N		 2165 /* Need to ensure auto-resizing doesn't interfere */
2166 mutex_lock(&conf->cache_size_mutex);
2167
ad01c9e3 2168 for (i = conf->max_nr_stripes; i; i--) {
f18c1a35 2169 nsh = alloc_stripe(sc, GFP_KERNEL);
ad01c9e3
N		 2170 if (!nsh)
2171 break;
2172
ad01c9e3 2173 nsh->raid_conf = conf;
ad01c9e3
N		 2174 list_add(&nsh->lru, &newstripes);
2175 }
2176 if (i) {
2177 /* didn't get enough, give up */
2178 while (!list_empty(&newstripes)) {
2179 nsh = list_entry(newstripes.next, struct stripe_head, lru);
2180 list_del(&nsh->lru);
2181 kmem_cache_free(sc, nsh);
2182 }
2183 kmem_cache_destroy(sc);
2d5b569b 2184 mutex_unlock(&conf->cache_size_mutex);
ad01c9e3
N		 2185 return -ENOMEM;
2186 }
2187 /* Step 2 - Must use GFP_NOIO now.
2188 * OK, we have enough stripes, start collecting inactive
2189 * stripes and copying them over
2190 */
566c09c5
SL		 2191 hash = 0;
2192 cnt = 0;
ad01c9e3 2193 list_for_each_entry(nsh, &newstripes, lru) {
566c09c5 2194 lock_device_hash_lock(conf, hash);
e9e4c377 2195 wait_event_exclusive_cmd(conf->wait_for_stripe[hash],
566c09c5
SL		 2196 !list_empty(conf->inactive_list + hash),
2197 unlock_device_hash_lock(conf, hash),
2198 lock_device_hash_lock(conf, hash));
2199 osh = get_free_stripe(conf, hash);
2200 unlock_device_hash_lock(conf, hash);
f18c1a35 2201
d592a996 2202 for(i=0; i<conf->pool_size; i++) {
ad01c9e3 2203 nsh->dev[i].page = osh->dev[i].page;
d592a996
SL		 2204 nsh->dev[i].orig_page = osh->dev[i].page;
2205 }
566c09c5 2206 nsh->hash_lock_index = hash;
ad01c9e3 2207 kmem_cache_free(conf->slab_cache, osh);
566c09c5
SL		 2208 cnt++;
2209 if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS +
2210 !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) {
2211 hash++;
2212 cnt = 0;
2213 }
ad01c9e3
N		 2214 }
2215 kmem_cache_destroy(conf->slab_cache);
2216
2217 /* Step 3.
2218 * At this point, we are holding all the stripes so the array
2219 * is completely stalled, so now is a good time to resize
d6f38f31 2220 * conf->disks and the scribble region
ad01c9e3
N		 2221 */
2222 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
2223 if (ndisks) {
2224 for (i=0; i<conf->raid_disks; i++)
2225 ndisks[i] = conf->disks[i];
2226 kfree(conf->disks);
2227 conf->disks = ndisks;
2228 } else
2229 err = -ENOMEM;
2230
2d5b569b 2231 mutex_unlock(&conf->cache_size_mutex);
ad01c9e3
N		 2232 /* Step 4, return new stripes to service */
2233 while(!list_empty(&newstripes)) {
2234 nsh = list_entry(newstripes.next, struct stripe_head, lru);
2235 list_del_init(&nsh->lru);
d6f38f31 2236
ad01c9e3
N		 2237 for (i=conf->raid_disks; i < newsize; i++)
2238 if (nsh->dev[i].page == NULL) {
2239 struct page *p = alloc_page(GFP_NOIO);
2240 nsh->dev[i].page = p;
d592a996 2241 nsh->dev[i].orig_page = p;
ad01c9e3
N		 2242 if (!p)
2243 err = -ENOMEM;
2244 }
6d036f7d 2245 raid5_release_stripe(nsh);
ad01c9e3
N		 2246 }
2247 /* critical section pass, GFP_NOIO no longer needed */
2248
2249 conf->slab_cache = sc;
2250 conf->active_name = 1-conf->active_name;
6e9eac2d
N		 2251 if (!err)
2252 conf->pool_size = newsize;
ad01c9e3
N		 2253 return err;
2254}
1da177e4 2255
486f0644 2256static int drop_one_stripe(struct r5conf *conf)
1da177e4
LT		 2257{
2258 struct stripe_head *sh;
49895bcc 2259 int hash = (conf->max_nr_stripes - 1) & STRIPE_HASH_LOCKS_MASK;
1da177e4 2260
566c09c5
SL		 2261 spin_lock_irq(conf->hash_locks + hash);
2262 sh = get_free_stripe(conf, hash);
2263 spin_unlock_irq(conf->hash_locks + hash);
3f294f4f
N		 2264 if (!sh)
2265 return 0;
78bafebd 2266 BUG_ON(atomic_read(&sh->count));
e4e11e38 2267 shrink_buffers(sh);
3f294f4f
N		 2268 kmem_cache_free(conf->slab_cache, sh);
2269 atomic_dec(&conf->active_stripes);
486f0644 2270 conf->max_nr_stripes--;
3f294f4f
N		 2271 return 1;
2272}
2273
d1688a6d 2274static void shrink_stripes(struct r5conf *conf)
3f294f4f 2275{
486f0644
N		 2276 while (conf->max_nr_stripes &&
2277 drop_one_stripe(conf))
2278 ;
3f294f4f 2279
644df1a8 2280 kmem_cache_destroy(conf->slab_cache);
1da177e4
LT		 2281 conf->slab_cache = NULL;
2282}
2283
4246a0b6 2284static void raid5_end_read_request(struct bio * bi)
1da177e4 2285{
99c0fb5f 2286 struct stripe_head *sh = bi->bi_private;
d1688a6d 2287 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 2288 int disks = sh->disks, i;
d6950432 2289 char b[BDEVNAME_SIZE];
dd054fce 2290 struct md_rdev *rdev = NULL;
05616be5 2291 sector_t s;
1da177e4
LT		 2292
2293 for (i=0 ; i<disks; i++)
2294 if (bi == &sh->dev[i].req)
2295 break;
2296
4246a0b6 2297 pr_debug("end_read_request %llu/%d, count: %d, error %d.\n",
45b4233c 2298 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
4246a0b6 2299 bi->bi_error);
1da177e4
LT		 2300 if (i == disks) {
2301 BUG();
6712ecf8 2302 return;
1da177e4 2303 }
14a75d3e 2304 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
dd054fce
N		 2305 /* If replacement finished while this request was outstanding,
2306 * 'replacement' might be NULL already.
2307 * In that case it moved down to 'rdev'.
2308 * rdev is not removed until all requests are finished.
2309 */
14a75d3e 2310 rdev = conf->disks[i].replacement;
dd054fce 2311 if (!rdev)
14a75d3e 2312 rdev = conf->disks[i].rdev;
1da177e4 2313
05616be5
N		 2314 if (use_new_offset(conf, sh))
2315 s = sh->sector + rdev->new_data_offset;
2316 else
2317 s = sh->sector + rdev->data_offset;
4246a0b6 2318 if (!bi->bi_error) {
1da177e4 2319 set_bit(R5_UPTODATE, &sh->dev[i].flags);
4e5314b5 2320 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
14a75d3e
N		 2321 /* Note that this cannot happen on a
2322 * replacement device. We just fail those on
2323 * any error
2324 */
8bda470e
CD		 2325 printk_ratelimited(
2326 KERN_INFO
2327 "md/raid:%s: read error corrected"
2328 " (%lu sectors at %llu on %s)\n",
2329 mdname(conf->mddev), STRIPE_SECTORS,
05616be5 2330 (unsigned long long)s,
8bda470e 2331 bdevname(rdev->bdev, b));
ddd5115f 2332 atomic_add(STRIPE_SECTORS, &rdev->corrected_errors);
4e5314b5
N		 2333 clear_bit(R5_ReadError, &sh->dev[i].flags);
2334 clear_bit(R5_ReWrite, &sh->dev[i].flags);
3f9e7c14 2335 } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2336 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2337
14a75d3e
N		 2338 if (atomic_read(&rdev->read_errors))
2339 atomic_set(&rdev->read_errors, 0);
1da177e4 2340 } else {
14a75d3e 2341 const char *bdn = bdevname(rdev->bdev, b);
ba22dcbf 2342 int retry = 0;
2e8ac303 2343 int set_bad = 0;
d6950432 2344
1da177e4 2345 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
d6950432 2346 atomic_inc(&rdev->read_errors);
14a75d3e
N		 2347 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
2348 printk_ratelimited(
2349 KERN_WARNING
2350 "md/raid:%s: read error on replacement device "
2351 "(sector %llu on %s).\n",
2352 mdname(conf->mddev),
05616be5 2353 (unsigned long long)s,
14a75d3e 2354 bdn);
2e8ac303 2355 else if (conf->mddev->degraded >= conf->max_degraded) {
2356 set_bad = 1;
8bda470e
CD		 2357 printk_ratelimited(
2358 KERN_WARNING
2359 "md/raid:%s: read error not correctable "
2360 "(sector %llu on %s).\n",
2361 mdname(conf->mddev),
05616be5 2362 (unsigned long long)s,
8bda470e 2363 bdn);
2e8ac303 2364 } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) {
4e5314b5 2365 /* Oh, no!!! */
2e8ac303 2366 set_bad = 1;
8bda470e
CD		 2367 printk_ratelimited(
2368 KERN_WARNING
2369 "md/raid:%s: read error NOT corrected!! "
2370 "(sector %llu on %s).\n",
2371 mdname(conf->mddev),
05616be5 2372 (unsigned long long)s,
8bda470e 2373 bdn);
2e8ac303 2374 } else if (atomic_read(&rdev->read_errors)
ba22dcbf 2375 > conf->max_nr_stripes)
14f8d26b 2376 printk(KERN_WARNING
0c55e022 2377 "md/raid:%s: Too many read errors, failing device %s.\n",
d6950432 2378 mdname(conf->mddev), bdn);
ba22dcbf
N		 2379 else
2380 retry = 1;
edfa1f65
BY		 2381 if (set_bad && test_bit(In_sync, &rdev->flags)
2382 && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2383 retry = 1;
ba22dcbf 2384 if (retry)
3f9e7c14 2385 if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) {
2386 set_bit(R5_ReadError, &sh->dev[i].flags);
2387 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2388 } else
2389 set_bit(R5_ReadNoMerge, &sh->dev[i].flags);
ba22dcbf 2390 else {
4e5314b5
N		 2391 clear_bit(R5_ReadError, &sh->dev[i].flags);
2392 clear_bit(R5_ReWrite, &sh->dev[i].flags);
2e8ac303 2393 if (!(set_bad
2394 && test_bit(In_sync, &rdev->flags)
2395 && rdev_set_badblocks(
2396 rdev, sh->sector, STRIPE_SECTORS, 0)))
2397 md_error(conf->mddev, rdev);
ba22dcbf 2398 }
1da177e4 2399 }
14a75d3e 2400 rdev_dec_pending(rdev, conf->mddev);
1da177e4
LT		 2401 clear_bit(R5_LOCKED, &sh->dev[i].flags);
2402 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 2403 raid5_release_stripe(sh);
1da177e4
LT		 2404}
2405
4246a0b6 2406static void raid5_end_write_request(struct bio *bi)
1da177e4 2407{
99c0fb5f 2408 struct stripe_head *sh = bi->bi_private;
d1688a6d 2409 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 2410 int disks = sh->disks, i;
977df362 2411 struct md_rdev *uninitialized_var(rdev);
b84db560
N		 2412 sector_t first_bad;
2413 int bad_sectors;
977df362 2414 int replacement = 0;
1da177e4 2415
977df362
N		 2416 for (i = 0 ; i < disks; i++) {
2417 if (bi == &sh->dev[i].req) {
2418 rdev = conf->disks[i].rdev;
1da177e4 2419 break;
977df362
N		 2420 }
2421 if (bi == &sh->dev[i].rreq) {
2422 rdev = conf->disks[i].replacement;
dd054fce
N		 2423 if (rdev)
2424 replacement = 1;
2425 else
2426 /* rdev was removed and 'replacement'
2427 * replaced it. rdev is not removed
2428 * until all requests are finished.
2429 */
2430 rdev = conf->disks[i].rdev;
977df362
N		 2431 break;
2432 }
2433 }
4246a0b6 2434 pr_debug("end_write_request %llu/%d, count %d, error: %d.\n",
1da177e4 2435 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
4246a0b6 2436 bi->bi_error);
1da177e4
LT		 2437 if (i == disks) {
2438 BUG();
6712ecf8 2439 return;
1da177e4
LT		 2440 }
2441
977df362 2442 if (replacement) {
4246a0b6 2443 if (bi->bi_error)
977df362
N		 2444 md_error(conf->mddev, rdev);
2445 else if (is_badblock(rdev, sh->sector,
2446 STRIPE_SECTORS,
2447 &first_bad, &bad_sectors))
2448 set_bit(R5_MadeGoodRepl, &sh->dev[i].flags);
2449 } else {
4246a0b6 2450 if (bi->bi_error) {
9f97e4b1 2451 set_bit(STRIPE_DEGRADED, &sh->state);
977df362
N		 2452 set_bit(WriteErrorSeen, &rdev->flags);
2453 set_bit(R5_WriteError, &sh->dev[i].flags);
3a6de292
N		 2454 if (!test_and_set_bit(WantReplacement, &rdev->flags))
2455 set_bit(MD_RECOVERY_NEEDED,
2456 &rdev->mddev->recovery);
977df362
N		 2457 } else if (is_badblock(rdev, sh->sector,
2458 STRIPE_SECTORS,
c0b32972 2459 &first_bad, &bad_sectors)) {
977df362 2460 set_bit(R5_MadeGood, &sh->dev[i].flags);
c0b32972
N		 2461 if (test_bit(R5_ReadError, &sh->dev[i].flags))
2462 /* That was a successful write so make
2463 * sure it looks like we already did
2464 * a re-write.
2465 */
2466 set_bit(R5_ReWrite, &sh->dev[i].flags);
2467 }
977df362
N		 2468 }
2469 rdev_dec_pending(rdev, conf->mddev);
1da177e4 2470
4246a0b6 2471 if (sh->batch_head && bi->bi_error && !replacement)
72ac7330 2472 set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state);
2473
977df362
N		 2474 if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags))
2475 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1da177e4 2476 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 2477 raid5_release_stripe(sh);
59fc630b 2478
2479 if (sh->batch_head && sh != sh->batch_head)
6d036f7d 2480 raid5_release_stripe(sh->batch_head);
1da177e4
LT		 2481}
2482
784052ec 2483static void raid5_build_block(struct stripe_head *sh, int i, int previous)
1da177e4
LT		 2484{
2485 struct r5dev *dev = &sh->dev[i];
2486
2487 bio_init(&dev->req);
2488 dev->req.bi_io_vec = &dev->vec;
d592a996 2489 dev->req.bi_max_vecs = 1;
1da177e4
LT		 2490 dev->req.bi_private = sh;
2491
977df362
N		 2492 bio_init(&dev->rreq);
2493 dev->rreq.bi_io_vec = &dev->rvec;
d592a996 2494 dev->rreq.bi_max_vecs = 1;
977df362 2495 dev->rreq.bi_private = sh;
977df362 2496
1da177e4 2497 dev->flags = 0;
6d036f7d 2498 dev->sector = raid5_compute_blocknr(sh, i, previous);
1da177e4
LT		 2499}
2500
fd01b88c 2501static void error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 2502{
2503 char b[BDEVNAME_SIZE];
d1688a6d 2504 struct r5conf *conf = mddev->private;
908f4fbd 2505 unsigned long flags;
0c55e022 2506 pr_debug("raid456: error called\n");
1da177e4 2507
908f4fbd
N		 2508 spin_lock_irqsave(&conf->device_lock, flags);
2509 clear_bit(In_sync, &rdev->flags);
2510 mddev->degraded = calc_degraded(conf);
2511 spin_unlock_irqrestore(&conf->device_lock, flags);
2512 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2513
de393cde 2514 set_bit(Blocked, &rdev->flags);
6f8d0c77
N		 2515 set_bit(Faulty, &rdev->flags);
2516 set_bit(MD_CHANGE_DEVS, &mddev->flags);
c3cce6cd 2517 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6f8d0c77
N		 2518 printk(KERN_ALERT
2519 "md/raid:%s: Disk failure on %s, disabling device.\n"
2520 "md/raid:%s: Operation continuing on %d devices.\n",
2521 mdname(mddev),
2522 bdevname(rdev->bdev, b),
2523 mdname(mddev),
2524 conf->raid_disks - mddev->degraded);
16a53ecc 2525}
1da177e4
LT		 2526
2527/*
2528 * Input: a 'big' sector number,
2529 * Output: index of the data and parity disk, and the sector # in them.
2530 */
6d036f7d
SL		 2531sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector,
2532 int previous, int *dd_idx,
2533 struct stripe_head *sh)
1da177e4 2534{
6e3b96ed 2535 sector_t stripe, stripe2;
35f2a591 2536 sector_t chunk_number;
1da177e4 2537 unsigned int chunk_offset;
911d4ee8 2538 int pd_idx, qd_idx;
67cc2b81 2539 int ddf_layout = 0;
1da177e4 2540 sector_t new_sector;
e183eaed
N		 2541 int algorithm = previous ? conf->prev_algo
2542 : conf->algorithm;
09c9e5fa
AN		 2543 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2544 : conf->chunk_sectors;
112bf897
N		 2545 int raid_disks = previous ? conf->previous_raid_disks
2546 : conf->raid_disks;
2547 int data_disks = raid_disks - conf->max_degraded;
1da177e4
LT		 2548
2549 /* First compute the information on this sector */
2550
2551 /*
2552 * Compute the chunk number and the sector offset inside the chunk
2553 */
2554 chunk_offset = sector_div(r_sector, sectors_per_chunk);
2555 chunk_number = r_sector;
1da177e4
LT		 2556
2557 /*
2558 * Compute the stripe number
2559 */
35f2a591
N		 2560 stripe = chunk_number;
2561 *dd_idx = sector_div(stripe, data_disks);
6e3b96ed 2562 stripe2 = stripe;
1da177e4
LT		 2563 /*
2564 * Select the parity disk based on the user selected algorithm.
2565 */
84789554 2566 pd_idx = qd_idx = -1;
16a53ecc
N		 2567 switch(conf->level) {
2568 case 4:
911d4ee8 2569 pd_idx = data_disks;
16a53ecc
N		 2570 break;
2571 case 5:
e183eaed 2572 switch (algorithm) {
1da177e4 2573 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2574 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2575 if (*dd_idx >= pd_idx)
1da177e4
LT		 2576 (*dd_idx)++;
2577 break;
2578 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2579 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2580 if (*dd_idx >= pd_idx)
1da177e4
LT		 2581 (*dd_idx)++;
2582 break;
2583 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2584 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2585 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4
LT		 2586 break;
2587 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2588 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2589 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4 2590 break;
99c0fb5f
N		 2591 case ALGORITHM_PARITY_0:
2592 pd_idx = 0;
2593 (*dd_idx)++;
2594 break;
2595 case ALGORITHM_PARITY_N:
2596 pd_idx = data_disks;
2597 break;
1da177e4 2598 default:
99c0fb5f 2599 BUG();
16a53ecc
N		 2600 }
2601 break;
2602 case 6:
2603
e183eaed 2604 switch (algorithm) {
16a53ecc 2605 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2606 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2607 qd_idx = pd_idx + 1;
2608 if (pd_idx == raid_disks-1) {
99c0fb5f 2609 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2610 qd_idx = 0;
2611 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2612 (*dd_idx) += 2; /* D D P Q D */
2613 break;
2614 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2615 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2616 qd_idx = pd_idx + 1;
2617 if (pd_idx == raid_disks-1) {
99c0fb5f 2618 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2619 qd_idx = 0;
2620 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2621 (*dd_idx) += 2; /* D D P Q D */
2622 break;
2623 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2624 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2625 qd_idx = (pd_idx + 1) % raid_disks;
2626 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc
N		 2627 break;
2628 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2629 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2630 qd_idx = (pd_idx + 1) % raid_disks;
2631 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc 2632 break;
99c0fb5f
N		 2633
2634 case ALGORITHM_PARITY_0:
2635 pd_idx = 0;
2636 qd_idx = 1;
2637 (*dd_idx) += 2;
2638 break;
2639 case ALGORITHM_PARITY_N:
2640 pd_idx = data_disks;
2641 qd_idx = data_disks + 1;
2642 break;
2643
2644 case ALGORITHM_ROTATING_ZERO_RESTART:
2645 /* Exactly the same as RIGHT_ASYMMETRIC, but or
2646 * of blocks for computing Q is different.
2647 */
6e3b96ed 2648 pd_idx = sector_div(stripe2, raid_disks);
99c0fb5f
N		 2649 qd_idx = pd_idx + 1;
2650 if (pd_idx == raid_disks-1) {
2651 (*dd_idx)++; /* Q D D D P */
2652 qd_idx = 0;
2653 } else if (*dd_idx >= pd_idx)
2654 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2655 ddf_layout = 1;
99c0fb5f
N		 2656 break;
2657
2658 case ALGORITHM_ROTATING_N_RESTART:
2659 /* Same a left_asymmetric, by first stripe is
2660 * D D D P Q rather than
2661 * Q D D D P
2662 */
6e3b96ed
N		 2663 stripe2 += 1;
2664 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2665 qd_idx = pd_idx + 1;
2666 if (pd_idx == raid_disks-1) {
2667 (*dd_idx)++; /* Q D D D P */
2668 qd_idx = 0;
2669 } else if (*dd_idx >= pd_idx)
2670 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2671 ddf_layout = 1;
99c0fb5f
N		 2672 break;
2673
2674 case ALGORITHM_ROTATING_N_CONTINUE:
2675 /* Same as left_symmetric but Q is before P */
6e3b96ed 2676 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2677 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
2678 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
67cc2b81 2679 ddf_layout = 1;
99c0fb5f
N		 2680 break;
2681
2682 case ALGORITHM_LEFT_ASYMMETRIC_6:
2683 /* RAID5 left_asymmetric, with Q on last device */
6e3b96ed 2684 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2685 if (*dd_idx >= pd_idx)
2686 (*dd_idx)++;
2687 qd_idx = raid_disks - 1;
2688 break;
2689
2690 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6e3b96ed 2691 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2692 if (*dd_idx >= pd_idx)
2693 (*dd_idx)++;
2694 qd_idx = raid_disks - 1;
2695 break;
2696
2697 case ALGORITHM_LEFT_SYMMETRIC_6:
6e3b96ed 2698 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2699 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2700 qd_idx = raid_disks - 1;
2701 break;
2702
2703 case ALGORITHM_RIGHT_SYMMETRIC_6:
6e3b96ed 2704 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2705 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2706 qd_idx = raid_disks - 1;
2707 break;
2708
2709 case ALGORITHM_PARITY_0_6:
2710 pd_idx = 0;
2711 (*dd_idx)++;
2712 qd_idx = raid_disks - 1;
2713 break;
2714
16a53ecc 2715 default:
99c0fb5f 2716 BUG();
16a53ecc
N		 2717 }
2718 break;
1da177e4
LT		 2719 }
2720
911d4ee8
N		 2721 if (sh) {
2722 sh->pd_idx = pd_idx;
2723 sh->qd_idx = qd_idx;
67cc2b81 2724 sh->ddf_layout = ddf_layout;
911d4ee8 2725 }
1da177e4
LT		 2726 /*
2727 * Finally, compute the new sector number
2728 */
2729 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
2730 return new_sector;
2731}
2732
6d036f7d 2733sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous)
1da177e4 2734{
d1688a6d 2735 struct r5conf *conf = sh->raid_conf;
b875e531
N		 2736 int raid_disks = sh->disks;
2737 int data_disks = raid_disks - conf->max_degraded;
1da177e4 2738 sector_t new_sector = sh->sector, check;
09c9e5fa
AN		 2739 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2740 : conf->chunk_sectors;
e183eaed
N		 2741 int algorithm = previous ? conf->prev_algo
2742 : conf->algorithm;
1da177e4
LT		 2743 sector_t stripe;
2744 int chunk_offset;
35f2a591
N		 2745 sector_t chunk_number;
2746 int dummy1, dd_idx = i;
1da177e4 2747 sector_t r_sector;
911d4ee8 2748 struct stripe_head sh2;
1da177e4
LT		 2749
2750 chunk_offset = sector_div(new_sector, sectors_per_chunk);
2751 stripe = new_sector;
1da177e4 2752
16a53ecc
N		 2753 if (i == sh->pd_idx)
2754 return 0;
2755 switch(conf->level) {
2756 case 4: break;
2757 case 5:
e183eaed 2758 switch (algorithm) {
1da177e4
LT		 2759 case ALGORITHM_LEFT_ASYMMETRIC:
2760 case ALGORITHM_RIGHT_ASYMMETRIC:
2761 if (i > sh->pd_idx)
2762 i--;
2763 break;
2764 case ALGORITHM_LEFT_SYMMETRIC:
2765 case ALGORITHM_RIGHT_SYMMETRIC:
2766 if (i < sh->pd_idx)
2767 i += raid_disks;
2768 i -= (sh->pd_idx + 1);
2769 break;
99c0fb5f
N		 2770 case ALGORITHM_PARITY_0:
2771 i -= 1;
2772 break;
2773 case ALGORITHM_PARITY_N:
2774 break;
1da177e4 2775 default:
99c0fb5f 2776 BUG();
16a53ecc
N		 2777 }
2778 break;
2779 case 6:
d0dabf7e 2780 if (i == sh->qd_idx)
16a53ecc 2781 return 0; /* It is the Q disk */
e183eaed 2782 switch (algorithm) {
16a53ecc
N		 2783 case ALGORITHM_LEFT_ASYMMETRIC:
2784 case ALGORITHM_RIGHT_ASYMMETRIC:
99c0fb5f
N		 2785 case ALGORITHM_ROTATING_ZERO_RESTART:
2786 case ALGORITHM_ROTATING_N_RESTART:
2787 if (sh->pd_idx == raid_disks-1)
2788 i--; /* Q D D D P */
16a53ecc
N		 2789 else if (i > sh->pd_idx)
2790 i -= 2; /* D D P Q D */
2791 break;
2792 case ALGORITHM_LEFT_SYMMETRIC:
2793 case ALGORITHM_RIGHT_SYMMETRIC:
2794 if (sh->pd_idx == raid_disks-1)
2795 i--; /* Q D D D P */
2796 else {
2797 /* D D P Q D */
2798 if (i < sh->pd_idx)
2799 i += raid_disks;
2800 i -= (sh->pd_idx + 2);
2801 }
2802 break;
99c0fb5f
N		 2803 case ALGORITHM_PARITY_0:
2804 i -= 2;
2805 break;
2806 case ALGORITHM_PARITY_N:
2807 break;
2808 case ALGORITHM_ROTATING_N_CONTINUE:
e4424fee 2809 /* Like left_symmetric, but P is before Q */
99c0fb5f
N		 2810 if (sh->pd_idx == 0)
2811 i--; /* P D D D Q */
e4424fee
N		 2812 else {
2813 /* D D Q P D */
2814 if (i < sh->pd_idx)
2815 i += raid_disks;
2816 i -= (sh->pd_idx + 1);
2817 }
99c0fb5f
N		 2818 break;
2819 case ALGORITHM_LEFT_ASYMMETRIC_6:
2820 case ALGORITHM_RIGHT_ASYMMETRIC_6:
2821 if (i > sh->pd_idx)
2822 i--;
2823 break;
2824 case ALGORITHM_LEFT_SYMMETRIC_6:
2825 case ALGORITHM_RIGHT_SYMMETRIC_6:
2826 if (i < sh->pd_idx)
2827 i += data_disks + 1;
2828 i -= (sh->pd_idx + 1);
2829 break;
2830 case ALGORITHM_PARITY_0_6:
2831 i -= 1;
2832 break;
16a53ecc 2833 default:
99c0fb5f 2834 BUG();
16a53ecc
N		 2835 }
2836 break;
1da177e4
LT		 2837 }
2838
2839 chunk_number = stripe * data_disks + i;
35f2a591 2840 r_sector = chunk_number * sectors_per_chunk + chunk_offset;
1da177e4 2841
112bf897 2842 check = raid5_compute_sector(conf, r_sector,
784052ec 2843 previous, &dummy1, &sh2);
911d4ee8
N		 2844 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
2845 || sh2.qd_idx != sh->qd_idx) {
0c55e022
N		 2846 printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n",
2847 mdname(conf->mddev));
1da177e4
LT		 2848 return 0;
2849 }
2850 return r_sector;
2851}
2852
600aa109 2853static void
c0f7bddb 2854schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
600aa109 2855 int rcw, int expand)
e33129d8 2856{
584acdd4 2857 int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx, disks = sh->disks;
d1688a6d 2858 struct r5conf *conf = sh->raid_conf;
c0f7bddb 2859 int level = conf->level;
e33129d8
DW		 2860
2861 if (rcw) {
e33129d8
DW		 2862
2863 for (i = disks; i--; ) {
2864 struct r5dev *dev = &sh->dev[i];
2865
2866 if (dev->towrite) {
2867 set_bit(R5_LOCKED, &dev->flags);
d8ee0728 2868 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2869 if (!expand)
2870 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2871 s->locked++;
e33129d8
DW		 2872 }
2873 }
ce7d363a
N		 2874 /* if we are not expanding this is a proper write request, and
2875 * there will be bios with new data to be drained into the
2876 * stripe cache
2877 */
2878 if (!expand) {
2879 if (!s->locked)
2880 /* False alarm, nothing to do */
2881 return;
2882 sh->reconstruct_state = reconstruct_state_drain_run;
2883 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2884 } else
2885 sh->reconstruct_state = reconstruct_state_run;
2886
2887 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
2888
c0f7bddb 2889 if (s->locked + conf->max_degraded == disks)
8b3e6cdc 2890 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
c0f7bddb 2891 atomic_inc(&conf->pending_full_writes);
e33129d8
DW		 2892 } else {
2893 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
2894 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
584acdd4
MS		 2895 BUG_ON(level == 6 &&
2896 (!(test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags) ||
2897 test_bit(R5_Wantcompute, &sh->dev[qd_idx].flags))));
e33129d8 2898
e33129d8
DW		 2899 for (i = disks; i--; ) {
2900 struct r5dev *dev = &sh->dev[i];
584acdd4 2901 if (i == pd_idx || i == qd_idx)
e33129d8
DW		 2902 continue;
2903
e33129d8
DW		 2904 if (dev->towrite &&
2905 (test_bit(R5_UPTODATE, &dev->flags) ||
d8ee0728
DW		 2906 test_bit(R5_Wantcompute, &dev->flags))) {
2907 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2908 set_bit(R5_LOCKED, &dev->flags);
2909 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2910 s->locked++;
e33129d8
DW		 2911 }
2912 }
ce7d363a
N		 2913 if (!s->locked)
2914 /* False alarm - nothing to do */
2915 return;
2916 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
2917 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
2918 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2919 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
e33129d8
DW		 2920 }
2921
c0f7bddb 2922 /* keep the parity disk(s) locked while asynchronous operations
e33129d8
DW		 2923 * are in flight
2924 */
2925 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
2926 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
600aa109 2927 s->locked++;
e33129d8 2928
c0f7bddb
YT		 2929 if (level == 6) {
2930 int qd_idx = sh->qd_idx;
2931 struct r5dev *dev = &sh->dev[qd_idx];
2932
2933 set_bit(R5_LOCKED, &dev->flags);
2934 clear_bit(R5_UPTODATE, &dev->flags);
2935 s->locked++;
2936 }
2937
600aa109 2938 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
e46b272b 2939 __func__, (unsigned long long)sh->sector,
600aa109 2940 s->locked, s->ops_request);
e33129d8 2941}
16a53ecc 2942
1da177e4
LT		 2943/*
2944 * Each stripe/dev can have one or more bion attached.
16a53ecc 2945 * toread/towrite point to the first in a chain.
1da177e4
LT		 2946 * The bi_next chain must be in order.
2947 */
da41ba65 2948static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx,
2949 int forwrite, int previous)
1da177e4
LT		 2950{
2951 struct bio **bip;
d1688a6d 2952 struct r5conf *conf = sh->raid_conf;
72626685 2953 int firstwrite=0;
1da177e4 2954
cbe47ec5 2955 pr_debug("adding bi b#%llu to stripe s#%llu\n",
4f024f37 2956 (unsigned long long)bi->bi_iter.bi_sector,
1da177e4
LT		 2957 (unsigned long long)sh->sector);
2958
b17459c0
SL		 2959 /*
2960 * If several bio share a stripe. The bio bi_phys_segments acts as a
2961 * reference count to avoid race. The reference count should already be
2962 * increased before this function is called (for example, in
2963 * make_request()), so other bio sharing this stripe will not free the
2964 * stripe. If a stripe is owned by one stripe, the stripe lock will
2965 * protect it.
2966 */
2967 spin_lock_irq(&sh->stripe_lock);
59fc630b 2968 /* Don't allow new IO added to stripes in batch list */
2969 if (sh->batch_head)
2970 goto overlap;
72626685 2971 if (forwrite) {
1da177e4 2972 bip = &sh->dev[dd_idx].towrite;
7eaf7e8e 2973 if (*bip == NULL)
72626685
N		 2974 firstwrite = 1;
2975 } else
1da177e4 2976 bip = &sh->dev[dd_idx].toread;
4f024f37
KO		 2977 while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) {
2978 if (bio_end_sector(*bip) > bi->bi_iter.bi_sector)
1da177e4
LT		 2979 goto overlap;
2980 bip = & (*bip)->bi_next;
2981 }
4f024f37 2982 if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi))
1da177e4
LT		 2983 goto overlap;
2984
da41ba65 2985 if (!forwrite || previous)
2986 clear_bit(STRIPE_BATCH_READY, &sh->state);
2987
78bafebd 2988 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
1da177e4
LT		 2989 if (*bip)
2990 bi->bi_next = *bip;
2991 *bip = bi;
e7836bd6 2992 raid5_inc_bi_active_stripes(bi);
72626685 2993
1da177e4
LT		 2994 if (forwrite) {
2995 /* check if page is covered */
2996 sector_t sector = sh->dev[dd_idx].sector;
2997 for (bi=sh->dev[dd_idx].towrite;
2998 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
4f024f37 2999 bi && bi->bi_iter.bi_sector <= sector;
1da177e4 3000 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
f73a1c7d
KO		 3001 if (bio_end_sector(bi) >= sector)
3002 sector = bio_end_sector(bi);
1da177e4
LT		 3003 }
3004 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
7a87f434 3005 if (!test_and_set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags))
3006 sh->overwrite_disks++;
1da177e4 3007 }
cbe47ec5
N		 3008
3009 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
4f024f37 3010 (unsigned long long)(*bip)->bi_iter.bi_sector,
cbe47ec5
N		 3011 (unsigned long long)sh->sector, dd_idx);
3012
3013 if (conf->mddev->bitmap && firstwrite) {
d0852df5
N		 3014 /* Cannot hold spinlock over bitmap_startwrite,
3015 * but must ensure this isn't added to a batch until
3016 * we have added to the bitmap and set bm_seq.
3017 * So set STRIPE_BITMAP_PENDING to prevent
3018 * batching.
3019 * If multiple add_stripe_bio() calls race here they
3020 * much all set STRIPE_BITMAP_PENDING. So only the first one
3021 * to complete "bitmap_startwrite" gets to set
3022 * STRIPE_BIT_DELAY. This is important as once a stripe
3023 * is added to a batch, STRIPE_BIT_DELAY cannot be changed
3024 * any more.
3025 */
3026 set_bit(STRIPE_BITMAP_PENDING, &sh->state);
3027 spin_unlock_irq(&sh->stripe_lock);
cbe47ec5
N		 3028 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
3029 STRIPE_SECTORS, 0);
d0852df5
N		 3030 spin_lock_irq(&sh->stripe_lock);
3031 clear_bit(STRIPE_BITMAP_PENDING, &sh->state);
3032 if (!sh->batch_head) {
3033 sh->bm_seq = conf->seq_flush+1;
3034 set_bit(STRIPE_BIT_DELAY, &sh->state);
3035 }
cbe47ec5 3036 }
d0852df5 3037 spin_unlock_irq(&sh->stripe_lock);
59fc630b 3038
3039 if (stripe_can_batch(sh))
3040 stripe_add_to_batch_list(conf, sh);
1da177e4
LT		 3041 return 1;
3042
3043 overlap:
3044 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
b17459c0 3045 spin_unlock_irq(&sh->stripe_lock);
1da177e4
LT		 3046 return 0;
3047}
3048
d1688a6d 3049static void end_reshape(struct r5conf *conf);
29269553 3050
d1688a6d 3051static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
911d4ee8 3052 struct stripe_head *sh)
ccfcc3c1 3053{
784052ec 3054 int sectors_per_chunk =
09c9e5fa 3055 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
911d4ee8 3056 int dd_idx;
2d2063ce 3057 int chunk_offset = sector_div(stripe, sectors_per_chunk);
112bf897 3058 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
2d2063ce 3059
112bf897
N		 3060 raid5_compute_sector(conf,
3061 stripe * (disks - conf->max_degraded)
b875e531 3062 *sectors_per_chunk + chunk_offset,
112bf897 3063 previous,
911d4ee8 3064 &dd_idx, sh);
ccfcc3c1
N		 3065}
3066
a4456856 3067static void
d1688a6d 3068handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh,
a4456856 3069 struct stripe_head_state *s, int disks,
34a6f80e 3070 struct bio_list *return_bi)
a4456856
DW		 3071{
3072 int i;
59fc630b 3073 BUG_ON(sh->batch_head);
a4456856
DW		 3074 for (i = disks; i--; ) {
3075 struct bio *bi;
3076 int bitmap_end = 0;
3077
3078 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
3cb03002 3079 struct md_rdev *rdev;
a4456856
DW		 3080 rcu_read_lock();
3081 rdev = rcu_dereference(conf->disks[i].rdev);
3082 if (rdev && test_bit(In_sync, &rdev->flags))
7f0da59b
N		 3083 atomic_inc(&rdev->nr_pending);
3084 else
3085 rdev = NULL;
a4456856 3086 rcu_read_unlock();
7f0da59b
N		 3087 if (rdev) {
3088 if (!rdev_set_badblocks(
3089 rdev,
3090 sh->sector,
3091 STRIPE_SECTORS, 0))
3092 md_error(conf->mddev, rdev);
3093 rdev_dec_pending(rdev, conf->mddev);
3094 }
a4456856 3095 }
b17459c0 3096 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 3097 /* fail all writes first */
3098 bi = sh->dev[i].towrite;
3099 sh->dev[i].towrite = NULL;
7a87f434 3100 sh->overwrite_disks = 0;
b17459c0 3101 spin_unlock_irq(&sh->stripe_lock);
1ed850f3 3102 if (bi)
a4456856 3103 bitmap_end = 1;
a4456856 3104
0576b1c6
SL		 3105 r5l_stripe_write_finished(sh);
3106
a4456856
DW		 3107 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
3108 wake_up(&conf->wait_for_overlap);
3109
4f024f37 3110 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3111 sh->dev[i].sector + STRIPE_SECTORS) {
3112 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
4246a0b6
CH		 3113
3114 bi->bi_error = -EIO;
e7836bd6 3115 if (!raid5_dec_bi_active_stripes(bi)) {
a4456856 3116 md_write_end(conf->mddev);
34a6f80e 3117 bio_list_add(return_bi, bi);
a4456856
DW		 3118 }
3119 bi = nextbi;
3120 }
7eaf7e8e
SL		 3121 if (bitmap_end)
3122 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3123 STRIPE_SECTORS, 0, 0);
3124 bitmap_end = 0;
a4456856
DW		 3125 /* and fail all 'written' */
3126 bi = sh->dev[i].written;
3127 sh->dev[i].written = NULL;
d592a996
SL		 3128 if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) {
3129 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
3130 sh->dev[i].page = sh->dev[i].orig_page;
3131 }
3132
a4456856 3133 if (bi) bitmap_end = 1;
4f024f37 3134 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3135 sh->dev[i].sector + STRIPE_SECTORS) {
3136 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
4246a0b6
CH		 3137
3138 bi->bi_error = -EIO;
e7836bd6 3139 if (!raid5_dec_bi_active_stripes(bi)) {
a4456856 3140 md_write_end(conf->mddev);
34a6f80e 3141 bio_list_add(return_bi, bi);
a4456856
DW		 3142 }
3143 bi = bi2;
3144 }
3145
b5e98d65
DW		 3146 /* fail any reads if this device is non-operational and
3147 * the data has not reached the cache yet.
3148 */
3149 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
6e74a9cf 3150 s->failed > conf->max_degraded &&
b5e98d65
DW		 3151 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
3152 test_bit(R5_ReadError, &sh->dev[i].flags))) {
143c4d05 3153 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 3154 bi = sh->dev[i].toread;
3155 sh->dev[i].toread = NULL;
143c4d05 3156 spin_unlock_irq(&sh->stripe_lock);
a4456856
DW		 3157 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
3158 wake_up(&conf->wait_for_overlap);
ebda780b
SL		 3159 if (bi)
3160 s->to_read--;
4f024f37 3161 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3162 sh->dev[i].sector + STRIPE_SECTORS) {
3163 struct bio *nextbi =
3164 r5_next_bio(bi, sh->dev[i].sector);
4246a0b6
CH		 3165
3166 bi->bi_error = -EIO;
34a6f80e
N		 3167 if (!raid5_dec_bi_active_stripes(bi))
3168 bio_list_add(return_bi, bi);
a4456856
DW		 3169 bi = nextbi;
3170 }
3171 }
a4456856
DW		 3172 if (bitmap_end)
3173 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3174 STRIPE_SECTORS, 0, 0);
8cfa7b0f
N		 3175 /* If we were in the middle of a write the parity block might
3176 * still be locked - so just clear all R5_LOCKED flags
3177 */
3178 clear_bit(R5_LOCKED, &sh->dev[i].flags);
a4456856 3179 }
ebda780b
SL		 3180 s->to_write = 0;
3181 s->written = 0;
a4456856 3182
8b3e6cdc
DW		 3183 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3184 if (atomic_dec_and_test(&conf->pending_full_writes))
3185 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 3186}
3187
7f0da59b 3188static void
d1688a6d 3189handle_failed_sync(struct r5conf *conf, struct stripe_head *sh,
7f0da59b
N		 3190 struct stripe_head_state *s)
3191{
3192 int abort = 0;
3193 int i;
3194
59fc630b 3195 BUG_ON(sh->batch_head);
7f0da59b 3196 clear_bit(STRIPE_SYNCING, &sh->state);
f8dfcffd
N		 3197 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
3198 wake_up(&conf->wait_for_overlap);
7f0da59b 3199 s->syncing = 0;
9a3e1101 3200 s->replacing = 0;
7f0da59b 3201 /* There is nothing more to do for sync/check/repair.
18b9837e
N		 3202 * Don't even need to abort as that is handled elsewhere
3203 * if needed, and not always wanted e.g. if there is a known
3204 * bad block here.
9a3e1101 3205 * For recover/replace we need to record a bad block on all
7f0da59b
N		 3206 * non-sync devices, or abort the recovery
3207 */
18b9837e
N		 3208 if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) {
3209 /* During recovery devices cannot be removed, so
3210 * locking and refcounting of rdevs is not needed
3211 */
3212 for (i = 0; i < conf->raid_disks; i++) {
3213 struct md_rdev *rdev = conf->disks[i].rdev;
3214 if (rdev
3215 && !test_bit(Faulty, &rdev->flags)
3216 && !test_bit(In_sync, &rdev->flags)
3217 && !rdev_set_badblocks(rdev, sh->sector,
3218 STRIPE_SECTORS, 0))
3219 abort = 1;
3220 rdev = conf->disks[i].replacement;
3221 if (rdev
3222 && !test_bit(Faulty, &rdev->flags)
3223 && !test_bit(In_sync, &rdev->flags)
3224 && !rdev_set_badblocks(rdev, sh->sector,
3225 STRIPE_SECTORS, 0))
3226 abort = 1;
3227 }
3228 if (abort)
3229 conf->recovery_disabled =
3230 conf->mddev->recovery_disabled;
7f0da59b 3231 }
18b9837e 3232 md_done_sync(conf->mddev, STRIPE_SECTORS, !abort);
7f0da59b
N		 3233}
3234
9a3e1101
N		 3235static int want_replace(struct stripe_head *sh, int disk_idx)
3236{
3237 struct md_rdev *rdev;
3238 int rv = 0;
3239 /* Doing recovery so rcu locking not required */
3240 rdev = sh->raid_conf->disks[disk_idx].replacement;
3241 if (rdev
3242 && !test_bit(Faulty, &rdev->flags)
3243 && !test_bit(In_sync, &rdev->flags)
3244 && (rdev->recovery_offset <= sh->sector
3245 || rdev->mddev->recovery_cp <= sh->sector))
3246 rv = 1;
3247
3248 return rv;
3249}
3250
93b3dbce 3251/* fetch_block - checks the given member device to see if its data needs
1fe797e6
DW		 3252 * to be read or computed to satisfy a request.
3253 *
3254 * Returns 1 when no more member devices need to be checked, otherwise returns
93b3dbce 3255 * 0 to tell the loop in handle_stripe_fill to continue
f38e1219 3256 */
2c58f06e
N		 3257
3258static int need_this_block(struct stripe_head *sh, struct stripe_head_state *s,
3259 int disk_idx, int disks)
a4456856 3260{
5599becc 3261 struct r5dev *dev = &sh->dev[disk_idx];
f2b3b44d
N		 3262 struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]],
3263 &sh->dev[s->failed_num[1]] };
ea664c82 3264 int i;
5599becc 3265
a79cfe12
N		 3266
3267 if (test_bit(R5_LOCKED, &dev->flags) ||
3268 test_bit(R5_UPTODATE, &dev->flags))
3269 /* No point reading this as we already have it or have
3270 * decided to get it.
3271 */
3272 return 0;
3273
3274 if (dev->toread ||
3275 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)))
3276 /* We need this block to directly satisfy a request */
3277 return 1;
3278
3279 if (s->syncing || s->expanding ||
3280 (s->replacing && want_replace(sh, disk_idx)))
3281 /* When syncing, or expanding we read everything.
3282 * When replacing, we need the replaced block.
3283 */
3284 return 1;
3285
3286 if ((s->failed >= 1 && fdev[0]->toread) ||
3287 (s->failed >= 2 && fdev[1]->toread))
3288 /* If we want to read from a failed device, then
3289 * we need to actually read every other device.
3290 */
3291 return 1;
3292
a9d56950
N		 3293 /* Sometimes neither read-modify-write nor reconstruct-write
3294 * cycles can work. In those cases we read every block we
3295 * can. Then the parity-update is certain to have enough to
3296 * work with.
3297 * This can only be a problem when we need to write something,
3298 * and some device has failed. If either of those tests
3299 * fail we need look no further.
3300 */
3301 if (!s->failed || !s->to_write)
3302 return 0;
3303
3304 if (test_bit(R5_Insync, &dev->flags) &&
3305 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3306 /* Pre-reads at not permitted until after short delay
3307 * to gather multiple requests. However if this
3308 * device is no Insync, the block could only be be computed
3309 * and there is no need to delay that.
3310 */
3311 return 0;
ea664c82 3312
36707bb2 3313 for (i = 0; i < s->failed && i < 2; i++) {
ea664c82
N		 3314 if (fdev[i]->towrite &&
3315 !test_bit(R5_UPTODATE, &fdev[i]->flags) &&
3316 !test_bit(R5_OVERWRITE, &fdev[i]->flags))
3317 /* If we have a partial write to a failed
3318 * device, then we will need to reconstruct
3319 * the content of that device, so all other
3320 * devices must be read.
3321 */
3322 return 1;
3323 }
3324
3325 /* If we are forced to do a reconstruct-write, either because
3326 * the current RAID6 implementation only supports that, or
3327 * or because parity cannot be trusted and we are currently
3328 * recovering it, there is extra need to be careful.
3329 * If one of the devices that we would need to read, because
3330 * it is not being overwritten (and maybe not written at all)
3331 * is missing/faulty, then we need to read everything we can.
3332 */
3333 if (sh->raid_conf->level != 6 &&
3334 sh->sector < sh->raid_conf->mddev->recovery_cp)
3335 /* reconstruct-write isn't being forced */
3336 return 0;
36707bb2 3337 for (i = 0; i < s->failed && i < 2; i++) {
10d82c5f
N		 3338 if (s->failed_num[i] != sh->pd_idx &&
3339 s->failed_num[i] != sh->qd_idx &&
3340 !test_bit(R5_UPTODATE, &fdev[i]->flags) &&
ea664c82
N		 3341 !test_bit(R5_OVERWRITE, &fdev[i]->flags))
3342 return 1;
3343 }
3344
2c58f06e
N		 3345 return 0;
3346}
3347
3348static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s,
3349 int disk_idx, int disks)
3350{
3351 struct r5dev *dev = &sh->dev[disk_idx];
3352
3353 /* is the data in this block needed, and can we get it? */
3354 if (need_this_block(sh, s, disk_idx, disks)) {
5599becc
YT		 3355 /* we would like to get this block, possibly by computing it,
3356 * otherwise read it if the backing disk is insync
3357 */
3358 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
3359 BUG_ON(test_bit(R5_Wantread, &dev->flags));
b0c783b3 3360 BUG_ON(sh->batch_head);
5599becc 3361 if ((s->uptodate == disks - 1) &&
f2b3b44d
N		 3362 (s->failed && (disk_idx == s->failed_num[0] ||
3363 disk_idx == s->failed_num[1]))) {
5599becc
YT		 3364 /* have disk failed, and we're requested to fetch it;
3365 * do compute it
a4456856 3366 */
5599becc
YT		 3367 pr_debug("Computing stripe %llu block %d\n",
3368 (unsigned long long)sh->sector, disk_idx);
3369 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3370 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3371 set_bit(R5_Wantcompute, &dev->flags);
3372 sh->ops.target = disk_idx;
3373 sh->ops.target2 = -1; /* no 2nd target */
3374 s->req_compute = 1;
93b3dbce
N		 3375 /* Careful: from this point on 'uptodate' is in the eye
3376 * of raid_run_ops which services 'compute' operations
3377 * before writes. R5_Wantcompute flags a block that will
3378 * be R5_UPTODATE by the time it is needed for a
3379 * subsequent operation.
3380 */
5599becc
YT		 3381 s->uptodate++;
3382 return 1;
3383 } else if (s->uptodate == disks-2 && s->failed >= 2) {
3384 /* Computing 2-failure is *very* expensive; only
3385 * do it if failed >= 2
3386 */
3387 int other;
3388 for (other = disks; other--; ) {
3389 if (other == disk_idx)
3390 continue;
3391 if (!test_bit(R5_UPTODATE,
3392 &sh->dev[other].flags))
3393 break;
a4456856 3394 }
5599becc
YT		 3395 BUG_ON(other < 0);
3396 pr_debug("Computing stripe %llu blocks %d,%d\n",
3397 (unsigned long long)sh->sector,
3398 disk_idx, other);
3399 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3400 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3401 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
3402 set_bit(R5_Wantcompute, &sh->dev[other].flags);
3403 sh->ops.target = disk_idx;
3404 sh->ops.target2 = other;
3405 s->uptodate += 2;
3406 s->req_compute = 1;
3407 return 1;
3408 } else if (test_bit(R5_Insync, &dev->flags)) {
3409 set_bit(R5_LOCKED, &dev->flags);
3410 set_bit(R5_Wantread, &dev->flags);
3411 s->locked++;
3412 pr_debug("Reading block %d (sync=%d)\n",
3413 disk_idx, s->syncing);
a4456856
DW		 3414 }
3415 }
5599becc
YT		 3416
3417 return 0;
3418}
3419
3420/**
93b3dbce 3421 * handle_stripe_fill - read or compute data to satisfy pending requests.
5599becc 3422 */
93b3dbce
N		 3423static void handle_stripe_fill(struct stripe_head *sh,
3424 struct stripe_head_state *s,
3425 int disks)
5599becc
YT		 3426{
3427 int i;
3428
3429 /* look for blocks to read/compute, skip this if a compute
3430 * is already in flight, or if the stripe contents are in the
3431 * midst of changing due to a write
3432 */
3433 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
3434 !sh->reconstruct_state)
3435 for (i = disks; i--; )
93b3dbce 3436 if (fetch_block(sh, s, i, disks))
5599becc 3437 break;
a4456856
DW		 3438 set_bit(STRIPE_HANDLE, &sh->state);
3439}
3440
787b76fa
N		 3441static void break_stripe_batch_list(struct stripe_head *head_sh,
3442 unsigned long handle_flags);
1fe797e6 3443/* handle_stripe_clean_event
a4456856
DW		 3444 * any written block on an uptodate or failed drive can be returned.
3445 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
3446 * never LOCKED, so we don't need to test 'failed' directly.
3447 */
d1688a6d 3448static void handle_stripe_clean_event(struct r5conf *conf,
34a6f80e 3449 struct stripe_head *sh, int disks, struct bio_list *return_bi)
a4456856
DW		 3450{
3451 int i;
3452 struct r5dev *dev;
f8dfcffd 3453 int discard_pending = 0;
59fc630b 3454 struct stripe_head *head_sh = sh;
3455 bool do_endio = false;
a4456856
DW		 3456
3457 for (i = disks; i--; )
3458 if (sh->dev[i].written) {
3459 dev = &sh->dev[i];
3460 if (!test_bit(R5_LOCKED, &dev->flags) &&
9e444768 3461 (test_bit(R5_UPTODATE, &dev->flags) ||
d592a996
SL		 3462 test_bit(R5_Discard, &dev->flags) ||
3463 test_bit(R5_SkipCopy, &dev->flags))) {
a4456856
DW		 3464 /* We can return any write requests */
3465 struct bio *wbi, *wbi2;
45b4233c 3466 pr_debug("Return write for disc %d\n", i);
ca64cae9
N		 3467 if (test_and_clear_bit(R5_Discard, &dev->flags))
3468 clear_bit(R5_UPTODATE, &dev->flags);
d592a996
SL		 3469 if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) {
3470 WARN_ON(test_bit(R5_UPTODATE, &dev->flags));
d592a996 3471 }
59fc630b 3472 do_endio = true;
3473
3474returnbi:
3475 dev->page = dev->orig_page;
a4456856
DW		 3476 wbi = dev->written;
3477 dev->written = NULL;
4f024f37 3478 while (wbi && wbi->bi_iter.bi_sector <
a4456856
DW		 3479 dev->sector + STRIPE_SECTORS) {
3480 wbi2 = r5_next_bio(wbi, dev->sector);
e7836bd6 3481 if (!raid5_dec_bi_active_stripes(wbi)) {
a4456856 3482 md_write_end(conf->mddev);
34a6f80e 3483 bio_list_add(return_bi, wbi);
a4456856
DW		 3484 }
3485 wbi = wbi2;
3486 }
7eaf7e8e
SL		 3487 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3488 STRIPE_SECTORS,
a4456856 3489 !test_bit(STRIPE_DEGRADED, &sh->state),
7eaf7e8e 3490 0);
59fc630b 3491 if (head_sh->batch_head) {
3492 sh = list_first_entry(&sh->batch_list,
3493 struct stripe_head,
3494 batch_list);
3495 if (sh != head_sh) {
3496 dev = &sh->dev[i];
3497 goto returnbi;
3498 }
3499 }
3500 sh = head_sh;
3501 dev = &sh->dev[i];
f8dfcffd
N		 3502 } else if (test_bit(R5_Discard, &dev->flags))
3503 discard_pending = 1;
d592a996
SL		 3504 WARN_ON(test_bit(R5_SkipCopy, &dev->flags));
3505 WARN_ON(dev->page != dev->orig_page);
f8dfcffd 3506 }
f6bed0ef 3507
0576b1c6
SL		 3508 r5l_stripe_write_finished(sh);
3509
f8dfcffd
N		 3510 if (!discard_pending &&
3511 test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) {
3512 clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
3513 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
3514 if (sh->qd_idx >= 0) {
3515 clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
3516 clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags);
3517 }
3518 /* now that discard is done we can proceed with any sync */
3519 clear_bit(STRIPE_DISCARD, &sh->state);
d47648fc
SL		 3520 /*
3521 * SCSI discard will change some bio fields and the stripe has
3522 * no updated data, so remove it from hash list and the stripe
3523 * will be reinitialized
3524 */
3525 spin_lock_irq(&conf->device_lock);
59fc630b 3526unhash:
d47648fc 3527 remove_hash(sh);
59fc630b 3528 if (head_sh->batch_head) {
3529 sh = list_first_entry(&sh->batch_list,
3530 struct stripe_head, batch_list);
3531 if (sh != head_sh)
3532 goto unhash;
3533 }
d47648fc 3534 spin_unlock_irq(&conf->device_lock);
59fc630b 3535 sh = head_sh;
3536
f8dfcffd
N		 3537 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state))
3538 set_bit(STRIPE_HANDLE, &sh->state);
3539
3540 }
8b3e6cdc
DW		 3541
3542 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3543 if (atomic_dec_and_test(&conf->pending_full_writes))
3544 md_wakeup_thread(conf->mddev->thread);
59fc630b 3545
787b76fa
N		 3546 if (head_sh->batch_head && do_endio)
3547 break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS);
a4456856
DW		 3548}
3549
d1688a6d 3550static void handle_stripe_dirtying(struct r5conf *conf,
c8ac1803
N		 3551 struct stripe_head *sh,
3552 struct stripe_head_state *s,
3553 int disks)
a4456856
DW		 3554{
3555 int rmw = 0, rcw = 0, i;
a7854487
AL		 3556 sector_t recovery_cp = conf->mddev->recovery_cp;
3557
584acdd4 3558 /* Check whether resync is now happening or should start.
a7854487
AL		 3559 * If yes, then the array is dirty (after unclean shutdown or
3560 * initial creation), so parity in some stripes might be inconsistent.
3561 * In this case, we need to always do reconstruct-write, to ensure
3562 * that in case of drive failure or read-error correction, we
3563 * generate correct data from the parity.
3564 */
584acdd4 3565 if (conf->rmw_level == PARITY_DISABLE_RMW ||
26ac1073
N		 3566 (recovery_cp < MaxSector && sh->sector >= recovery_cp &&
3567 s->failed == 0)) {
a7854487 3568 /* Calculate the real rcw later - for now make it
c8ac1803
N		 3569 * look like rcw is cheaper
3570 */
3571 rcw = 1; rmw = 2;
584acdd4
MS		 3572 pr_debug("force RCW rmw_level=%u, recovery_cp=%llu sh->sector=%llu\n",
3573 conf->rmw_level, (unsigned long long)recovery_cp,
a7854487 3574 (unsigned long long)sh->sector);
c8ac1803 3575 } else for (i = disks; i--; ) {
a4456856
DW		 3576 /* would I have to read this buffer for read_modify_write */
3577 struct r5dev *dev = &sh->dev[i];
584acdd4 3578 if ((dev->towrite || i == sh->pd_idx || i == sh->qd_idx) &&
a4456856 3579 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 3580 !(test_bit(R5_UPTODATE, &dev->flags) ||
3581 test_bit(R5_Wantcompute, &dev->flags))) {
a4456856
DW		 3582 if (test_bit(R5_Insync, &dev->flags))
3583 rmw++;
3584 else
3585 rmw += 2*disks; /* cannot read it */
3586 }
3587 /* Would I have to read this buffer for reconstruct_write */
584acdd4
MS		 3588 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
3589 i != sh->pd_idx && i != sh->qd_idx &&
a4456856 3590 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 3591 !(test_bit(R5_UPTODATE, &dev->flags) ||
3592 test_bit(R5_Wantcompute, &dev->flags))) {
67f45548
N		 3593 if (test_bit(R5_Insync, &dev->flags))
3594 rcw++;
a4456856
DW		 3595 else
3596 rcw += 2*disks;
3597 }
3598 }
45b4233c 3599 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
a4456856
DW		 3600 (unsigned long long)sh->sector, rmw, rcw);
3601 set_bit(STRIPE_HANDLE, &sh->state);
584acdd4 3602 if ((rmw < rcw || (rmw == rcw && conf->rmw_level == PARITY_ENABLE_RMW)) && rmw > 0) {
a4456856 3603 /* prefer read-modify-write, but need to get some data */
e3620a3a
JB		 3604 if (conf->mddev->queue)
3605 blk_add_trace_msg(conf->mddev->queue,
3606 "raid5 rmw %llu %d",
3607 (unsigned long long)sh->sector, rmw);
a4456856
DW		 3608 for (i = disks; i--; ) {
3609 struct r5dev *dev = &sh->dev[i];
584acdd4 3610 if ((dev->towrite || i == sh->pd_idx || i == sh->qd_idx) &&
a4456856 3611 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 3612 !(test_bit(R5_UPTODATE, &dev->flags) ||
3613 test_bit(R5_Wantcompute, &dev->flags)) &&
a4456856 3614 test_bit(R5_Insync, &dev->flags)) {
67f45548
N		 3615 if (test_bit(STRIPE_PREREAD_ACTIVE,
3616 &sh->state)) {
3617 pr_debug("Read_old block %d for r-m-w\n",
3618 i);
a4456856
DW		 3619 set_bit(R5_LOCKED, &dev->flags);
3620 set_bit(R5_Wantread, &dev->flags);
3621 s->locked++;
3622 } else {
3623 set_bit(STRIPE_DELAYED, &sh->state);
3624 set_bit(STRIPE_HANDLE, &sh->state);
3625 }
3626 }
3627 }
a9add5d9 3628 }
584acdd4 3629 if ((rcw < rmw || (rcw == rmw && conf->rmw_level != PARITY_ENABLE_RMW)) && rcw > 0) {
a4456856 3630 /* want reconstruct write, but need to get some data */
a9add5d9 3631 int qread =0;
c8ac1803 3632 rcw = 0;
a4456856
DW		 3633 for (i = disks; i--; ) {
3634 struct r5dev *dev = &sh->dev[i];
3635 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
c8ac1803 3636 i != sh->pd_idx && i != sh->qd_idx &&
a4456856 3637 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219 3638 !(test_bit(R5_UPTODATE, &dev->flags) ||
c8ac1803
N		 3639 test_bit(R5_Wantcompute, &dev->flags))) {
3640 rcw++;
67f45548
N		 3641 if (test_bit(R5_Insync, &dev->flags) &&
3642 test_bit(STRIPE_PREREAD_ACTIVE,
3643 &sh->state)) {
45b4233c 3644 pr_debug("Read_old block "
a4456856
DW		 3645 "%d for Reconstruct\n", i);
3646 set_bit(R5_LOCKED, &dev->flags);
3647 set_bit(R5_Wantread, &dev->flags);
3648 s->locked++;
a9add5d9 3649 qread++;
a4456856
DW		 3650 } else {
3651 set_bit(STRIPE_DELAYED, &sh->state);
3652 set_bit(STRIPE_HANDLE, &sh->state);
3653 }
3654 }
3655 }
e3620a3a 3656 if (rcw && conf->mddev->queue)
a9add5d9
N		 3657 blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d",
3658 (unsigned long long)sh->sector,
3659 rcw, qread, test_bit(STRIPE_DELAYED, &sh->state));
c8ac1803 3660 }
b1b02fe9
N		 3661
3662 if (rcw > disks && rmw > disks &&
3663 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3664 set_bit(STRIPE_DELAYED, &sh->state);
3665
a4456856
DW		 3666 /* now if nothing is locked, and if we have enough data,
3667 * we can start a write request
3668 */
f38e1219
DW		 3669 /* since handle_stripe can be called at any time we need to handle the
3670 * case where a compute block operation has been submitted and then a
ac6b53b6
DW		 3671 * subsequent call wants to start a write request. raid_run_ops only
3672 * handles the case where compute block and reconstruct are requested
f38e1219
DW		 3673 * simultaneously. If this is not the case then new writes need to be
3674 * held off until the compute completes.
3675 */
976ea8d4
DW		 3676 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
3677 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
3678 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
c0f7bddb 3679 schedule_reconstruction(sh, s, rcw == 0, 0);
a4456856
DW		 3680}
3681
d1688a6d 3682static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh,
a4456856
DW		 3683 struct stripe_head_state *s, int disks)
3684{
ecc65c9b 3685 struct r5dev *dev = NULL;
bd2ab670 3686
59fc630b 3687 BUG_ON(sh->batch_head);
a4456856 3688 set_bit(STRIPE_HANDLE, &sh->state);
e89f8962 3689
ecc65c9b
DW		 3690 switch (sh->check_state) {
3691 case check_state_idle:
3692 /* start a new check operation if there are no failures */
bd2ab670 3693 if (s->failed == 0) {
bd2ab670 3694 BUG_ON(s->uptodate != disks);
ecc65c9b
DW		 3695 sh->check_state = check_state_run;
3696 set_bit(STRIPE_OP_CHECK, &s->ops_request);
bd2ab670 3697 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
bd2ab670 3698 s->uptodate--;
ecc65c9b 3699 break;
bd2ab670 3700 }
f2b3b44d 3701 dev = &sh->dev[s->failed_num[0]];
ecc65c9b
DW		 3702 /* fall through */
3703 case check_state_compute_result:
3704 sh->check_state = check_state_idle;
3705 if (!dev)
3706 dev = &sh->dev[sh->pd_idx];
3707
3708 /* check that a write has not made the stripe insync */
3709 if (test_bit(STRIPE_INSYNC, &sh->state))
3710 break;
c8894419 3711
a4456856 3712 /* either failed parity check, or recovery is happening */
a4456856
DW		 3713 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
3714 BUG_ON(s->uptodate != disks);
3715
3716 set_bit(R5_LOCKED, &dev->flags);
ecc65c9b 3717 s->locked++;
a4456856 3718 set_bit(R5_Wantwrite, &dev->flags);
830ea016 3719
a4456856 3720 clear_bit(STRIPE_DEGRADED, &sh->state);
a4456856 3721 set_bit(STRIPE_INSYNC, &sh->state);
ecc65c9b
DW		 3722 break;
3723 case check_state_run:
3724 break; /* we will be called again upon completion */
3725 case check_state_check_result:
3726 sh->check_state = check_state_idle;
3727
3728 /* if a failure occurred during the check operation, leave
3729 * STRIPE_INSYNC not set and let the stripe be handled again
3730 */
3731 if (s->failed)
3732 break;
3733
3734 /* handle a successful check operation, if parity is correct
3735 * we are done. Otherwise update the mismatch count and repair
3736 * parity if !MD_RECOVERY_CHECK
3737 */
ad283ea4 3738 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
ecc65c9b
DW		 3739 /* parity is correct (on disc,
3740 * not in buffer any more)
3741 */
3742 set_bit(STRIPE_INSYNC, &sh->state);
3743 else {
7f7583d4 3744 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
ecc65c9b
DW		 3745 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
3746 /* don't try to repair!! */
3747 set_bit(STRIPE_INSYNC, &sh->state);
3748 else {
3749 sh->check_state = check_state_compute_run;
976ea8d4 3750 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
ecc65c9b
DW		 3751 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3752 set_bit(R5_Wantcompute,
3753 &sh->dev[sh->pd_idx].flags);
3754 sh->ops.target = sh->pd_idx;
ac6b53b6 3755 sh->ops.target2 = -1;
ecc65c9b
DW		 3756 s->uptodate++;
3757 }
3758 }
3759 break;
3760 case check_state_compute_run:
3761 break;
3762 default:
3763 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
3764 __func__, sh->check_state,
3765 (unsigned long long) sh->sector);
3766 BUG();
a4456856
DW		 3767 }
3768}
3769
d1688a6d 3770static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh,
36d1c647 3771 struct stripe_head_state *s,
f2b3b44d 3772 int disks)
a4456856 3773{
a4456856 3774 int pd_idx = sh->pd_idx;
34e04e87 3775 int qd_idx = sh->qd_idx;
d82dfee0 3776 struct r5dev *dev;
a4456856 3777
59fc630b 3778 BUG_ON(sh->batch_head);
a4456856
DW		 3779 set_bit(STRIPE_HANDLE, &sh->state);
3780
3781 BUG_ON(s->failed > 2);
d82dfee0 3782
a4456856
DW		 3783 /* Want to check and possibly repair P and Q.
3784 * However there could be one 'failed' device, in which
3785 * case we can only check one of them, possibly using the
3786 * other to generate missing data
3787 */
3788
d82dfee0
DW		 3789 switch (sh->check_state) {
3790 case check_state_idle:
3791 /* start a new check operation if there are < 2 failures */
f2b3b44d 3792 if (s->failed == s->q_failed) {
d82dfee0 3793 /* The only possible failed device holds Q, so it
a4456856
DW		 3794 * makes sense to check P (If anything else were failed,
3795 * we would have used P to recreate it).
3796 */
d82dfee0 3797 sh->check_state = check_state_run;
a4456856 3798 }
f2b3b44d 3799 if (!s->q_failed && s->failed < 2) {
d82dfee0 3800 /* Q is not failed, and we didn't use it to generate
a4456856
DW		 3801 * anything, so it makes sense to check it
3802 */
d82dfee0
DW		 3803 if (sh->check_state == check_state_run)
3804 sh->check_state = check_state_run_pq;
3805 else
3806 sh->check_state = check_state_run_q;
a4456856 3807 }
a4456856 3808
d82dfee0
DW		 3809 /* discard potentially stale zero_sum_result */
3810 sh->ops.zero_sum_result = 0;
a4456856 3811
d82dfee0
DW		 3812 if (sh->check_state == check_state_run) {
3813 /* async_xor_zero_sum destroys the contents of P */
3814 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
3815 s->uptodate--;
a4456856 3816 }
d82dfee0
DW		 3817 if (sh->check_state >= check_state_run &&
3818 sh->check_state <= check_state_run_pq) {
3819 /* async_syndrome_zero_sum preserves P and Q, so
3820 * no need to mark them !uptodate here
3821 */
3822 set_bit(STRIPE_OP_CHECK, &s->ops_request);
3823 break;
a4456856
DW		 3824 }
3825
d82dfee0
DW		 3826 /* we have 2-disk failure */
3827 BUG_ON(s->failed != 2);
3828 /* fall through */
3829 case check_state_compute_result:
3830 sh->check_state = check_state_idle;
a4456856 3831
d82dfee0
DW		 3832 /* check that a write has not made the stripe insync */
3833 if (test_bit(STRIPE_INSYNC, &sh->state))
3834 break;
a4456856
DW		 3835
3836 /* now write out any block on a failed drive,
d82dfee0 3837 * or P or Q if they were recomputed
a4456856 3838 */
d82dfee0 3839 BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */
a4456856 3840 if (s->failed == 2) {
f2b3b44d 3841 dev = &sh->dev[s->failed_num[1]];
a4456856
DW		 3842 s->locked++;
3843 set_bit(R5_LOCKED, &dev->flags);
3844 set_bit(R5_Wantwrite, &dev->flags);
3845 }
3846 if (s->failed >= 1) {
f2b3b44d 3847 dev = &sh->dev[s->failed_num[0]];
a4456856
DW		 3848 s->locked++;
3849 set_bit(R5_LOCKED, &dev->flags);
3850 set_bit(R5_Wantwrite, &dev->flags);
3851 }
d82dfee0 3852 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
a4456856
DW		 3853 dev = &sh->dev[pd_idx];
3854 s->locked++;
3855 set_bit(R5_LOCKED, &dev->flags);
3856 set_bit(R5_Wantwrite, &dev->flags);
3857 }
d82dfee0 3858 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
a4456856
DW		 3859 dev = &sh->dev[qd_idx];
3860 s->locked++;
3861 set_bit(R5_LOCKED, &dev->flags);
3862 set_bit(R5_Wantwrite, &dev->flags);
3863 }
3864 clear_bit(STRIPE_DEGRADED, &sh->state);
3865
3866 set_bit(STRIPE_INSYNC, &sh->state);
d82dfee0
DW		 3867 break;
3868 case check_state_run:
3869 case check_state_run_q:
3870 case check_state_run_pq:
3871 break; /* we will be called again upon completion */
3872 case check_state_check_result:
3873 sh->check_state = check_state_idle;
3874
3875 /* handle a successful check operation, if parity is correct
3876 * we are done. Otherwise update the mismatch count and repair
3877 * parity if !MD_RECOVERY_CHECK
3878 */
3879 if (sh->ops.zero_sum_result == 0) {
3880 /* both parities are correct */
3881 if (!s->failed)
3882 set_bit(STRIPE_INSYNC, &sh->state);
3883 else {
3884 /* in contrast to the raid5 case we can validate
3885 * parity, but still have a failure to write
3886 * back
3887 */
3888 sh->check_state = check_state_compute_result;
3889 /* Returning at this point means that we may go
3890 * off and bring p and/or q uptodate again so
3891 * we make sure to check zero_sum_result again
3892 * to verify if p or q need writeback
3893 */
3894 }
3895 } else {
7f7583d4 3896 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
d82dfee0
DW		 3897 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
3898 /* don't try to repair!! */
3899 set_bit(STRIPE_INSYNC, &sh->state);
3900 else {
3901 int *target = &sh->ops.target;
3902
3903 sh->ops.target = -1;
3904 sh->ops.target2 = -1;
3905 sh->check_state = check_state_compute_run;
3906 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3907 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3908 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
3909 set_bit(R5_Wantcompute,
3910 &sh->dev[pd_idx].flags);
3911 *target = pd_idx;
3912 target = &sh->ops.target2;
3913 s->uptodate++;
3914 }
3915 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
3916 set_bit(R5_Wantcompute,
3917 &sh->dev[qd_idx].flags);
3918 *target = qd_idx;
3919 s->uptodate++;
3920 }
3921 }
3922 }
3923 break;
3924 case check_state_compute_run:
3925 break;
3926 default:
3927 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
3928 __func__, sh->check_state,
3929 (unsigned long long) sh->sector);
3930 BUG();
a4456856
DW		 3931 }
3932}
3933
d1688a6d 3934static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh)
a4456856
DW		 3935{
3936 int i;
3937
3938 /* We have read all the blocks in this stripe and now we need to
3939 * copy some of them into a target stripe for expand.
3940 */
f0a50d37 3941 struct dma_async_tx_descriptor *tx = NULL;
59fc630b 3942 BUG_ON(sh->batch_head);
a4456856
DW		 3943 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3944 for (i = 0; i < sh->disks; i++)
34e04e87 3945 if (i != sh->pd_idx && i != sh->qd_idx) {
911d4ee8 3946 int dd_idx, j;
a4456856 3947 struct stripe_head *sh2;
a08abd8c 3948 struct async_submit_ctl submit;
a4456856 3949
6d036f7d 3950 sector_t bn = raid5_compute_blocknr(sh, i, 1);
911d4ee8
N		 3951 sector_t s = raid5_compute_sector(conf, bn, 0,
3952 &dd_idx, NULL);
6d036f7d 3953 sh2 = raid5_get_active_stripe(conf, s, 0, 1, 1);
a4456856
DW		 3954 if (sh2 == NULL)
3955 /* so far only the early blocks of this stripe
3956 * have been requested. When later blocks
3957 * get requested, we will try again
3958 */
3959 continue;
3960 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
3961 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
3962 /* must have already done this block */
6d036f7d 3963 raid5_release_stripe(sh2);
a4456856
DW		 3964 continue;
3965 }
f0a50d37
DW		 3966
3967 /* place all the copies on one channel */
a08abd8c 3968 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
f0a50d37 3969 tx = async_memcpy(sh2->dev[dd_idx].page,
88ba2aa5 3970 sh->dev[i].page, 0, 0, STRIPE_SIZE,
a08abd8c 3971 &submit);
f0a50d37 3972
a4456856
DW		 3973 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
3974 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
3975 for (j = 0; j < conf->raid_disks; j++)
3976 if (j != sh2->pd_idx &&
86c374ba 3977 j != sh2->qd_idx &&
a4456856
DW		 3978 !test_bit(R5_Expanded, &sh2->dev[j].flags))
3979 break;
3980 if (j == conf->raid_disks) {
3981 set_bit(STRIPE_EXPAND_READY, &sh2->state);
3982 set_bit(STRIPE_HANDLE, &sh2->state);
3983 }
6d036f7d 3984 raid5_release_stripe(sh2);
f0a50d37 3985
a4456856 3986 }
a2e08551 3987 /* done submitting copies, wait for them to complete */
749586b7 3988 async_tx_quiesce(&tx);
a4456856 3989}
1da177e4
LT		 3990
3991/*
3992 * handle_stripe - do things to a stripe.
3993 *
9a3e1101
N		 3994 * We lock the stripe by setting STRIPE_ACTIVE and then examine the
3995 * state of various bits to see what needs to be done.
1da177e4 3996 * Possible results:
9a3e1101
N		 3997 * return some read requests which now have data
3998 * return some write requests which are safely on storage
1da177e4
LT		 3999 * schedule a read on some buffers
4000 * schedule a write of some buffers
4001 * return confirmation of parity correctness
4002 *
1da177e4 4003 */
a4456856 4004
acfe726b 4005static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s)
1da177e4 4006{
d1688a6d 4007 struct r5conf *conf = sh->raid_conf;
f416885e 4008 int disks = sh->disks;
474af965
N		 4009 struct r5dev *dev;
4010 int i;
9a3e1101 4011 int do_recovery = 0;
1da177e4 4012
acfe726b
N		 4013 memset(s, 0, sizeof(*s));
4014
dabc4ec6 4015 s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state) && !sh->batch_head;
4016 s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state) && !sh->batch_head;
acfe726b
N		 4017 s->failed_num[0] = -1;
4018 s->failed_num[1] = -1;
6e74a9cf 4019 s->log_failed = r5l_log_disk_error(conf);
1da177e4 4020
acfe726b 4021 /* Now to look around and see what can be done */
1da177e4 4022 rcu_read_lock();
16a53ecc 4023 for (i=disks; i--; ) {
3cb03002 4024 struct md_rdev *rdev;
31c176ec
N		 4025 sector_t first_bad;
4026 int bad_sectors;
4027 int is_bad = 0;
acfe726b 4028
16a53ecc 4029 dev = &sh->dev[i];
1da177e4 4030
45b4233c 4031 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
9a3e1101
N		 4032 i, dev->flags,
4033 dev->toread, dev->towrite, dev->written);
6c0069c0
YT		 4034 /* maybe we can reply to a read
4035 *
4036 * new wantfill requests are only permitted while
4037 * ops_complete_biofill is guaranteed to be inactive
4038 */
4039 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
4040 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
4041 set_bit(R5_Wantfill, &dev->flags);
1da177e4 4042
16a53ecc 4043 /* now count some things */
cc94015a
N		 4044 if (test_bit(R5_LOCKED, &dev->flags))
4045 s->locked++;
4046 if (test_bit(R5_UPTODATE, &dev->flags))
4047 s->uptodate++;
2d6e4ecc 4048 if (test_bit(R5_Wantcompute, &dev->flags)) {
cc94015a
N		 4049 s->compute++;
4050 BUG_ON(s->compute > 2);
2d6e4ecc 4051 }
1da177e4 4052
acfe726b 4053 if (test_bit(R5_Wantfill, &dev->flags))
cc94015a 4054 s->to_fill++;
acfe726b 4055 else if (dev->toread)
cc94015a 4056 s->to_read++;
16a53ecc 4057 if (dev->towrite) {
cc94015a 4058 s->to_write++;
16a53ecc 4059 if (!test_bit(R5_OVERWRITE, &dev->flags))
cc94015a 4060 s->non_overwrite++;
16a53ecc 4061 }
a4456856 4062 if (dev->written)
cc94015a 4063 s->written++;
14a75d3e
N		 4064 /* Prefer to use the replacement for reads, but only
4065 * if it is recovered enough and has no bad blocks.
4066 */
4067 rdev = rcu_dereference(conf->disks[i].replacement);
4068 if (rdev && !test_bit(Faulty, &rdev->flags) &&
4069 rdev->recovery_offset >= sh->sector + STRIPE_SECTORS &&
4070 !is_badblock(rdev, sh->sector, STRIPE_SECTORS,
4071 &first_bad, &bad_sectors))
4072 set_bit(R5_ReadRepl, &dev->flags);
4073 else {
e6030cb0 4074 if (rdev && !test_bit(Faulty, &rdev->flags))
9a3e1101 4075 set_bit(R5_NeedReplace, &dev->flags);
e6030cb0
N		 4076 else
4077 clear_bit(R5_NeedReplace, &dev->flags);
14a75d3e
N		 4078 rdev = rcu_dereference(conf->disks[i].rdev);
4079 clear_bit(R5_ReadRepl, &dev->flags);
4080 }
9283d8c5
N		 4081 if (rdev && test_bit(Faulty, &rdev->flags))
4082 rdev = NULL;
31c176ec
N		 4083 if (rdev) {
4084 is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
4085 &first_bad, &bad_sectors);
4086 if (s->blocked_rdev == NULL
4087 && (test_bit(Blocked, &rdev->flags)
4088 || is_bad < 0)) {
4089 if (is_bad < 0)
4090 set_bit(BlockedBadBlocks,
4091 &rdev->flags);
4092 s->blocked_rdev = rdev;
4093 atomic_inc(&rdev->nr_pending);
4094 }
6bfe0b49 4095 }
415e72d0
N		 4096 clear_bit(R5_Insync, &dev->flags);
4097 if (!rdev)
4098 /* Not in-sync */;
31c176ec
N		 4099 else if (is_bad) {
4100 /* also not in-sync */
18b9837e
N		 4101 if (!test_bit(WriteErrorSeen, &rdev->flags) &&
4102 test_bit(R5_UPTODATE, &dev->flags)) {
31c176ec
N		 4103 /* treat as in-sync, but with a read error
4104 * which we can now try to correct
4105 */
4106 set_bit(R5_Insync, &dev->flags);
4107 set_bit(R5_ReadError, &dev->flags);
4108 }
4109 } else if (test_bit(In_sync, &rdev->flags))
415e72d0 4110 set_bit(R5_Insync, &dev->flags);
30d7a483 4111 else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
415e72d0 4112 /* in sync if before recovery_offset */
30d7a483
N		 4113 set_bit(R5_Insync, &dev->flags);
4114 else if (test_bit(R5_UPTODATE, &dev->flags) &&
4115 test_bit(R5_Expanded, &dev->flags))
4116 /* If we've reshaped into here, we assume it is Insync.
4117 * We will shortly update recovery_offset to make
4118 * it official.
4119 */
4120 set_bit(R5_Insync, &dev->flags);
4121
1cc03eb9 4122 if (test_bit(R5_WriteError, &dev->flags)) {
14a75d3e
N		 4123 /* This flag does not apply to '.replacement'
4124 * only to .rdev, so make sure to check that*/
4125 struct md_rdev *rdev2 = rcu_dereference(
4126 conf->disks[i].rdev);
4127 if (rdev2 == rdev)
4128 clear_bit(R5_Insync, &dev->flags);
4129 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
bc2607f3 4130 s->handle_bad_blocks = 1;
14a75d3e 4131 atomic_inc(&rdev2->nr_pending);
bc2607f3
N		 4132 } else
4133 clear_bit(R5_WriteError, &dev->flags);
4134 }
1cc03eb9 4135 if (test_bit(R5_MadeGood, &dev->flags)) {
14a75d3e
N		 4136 /* This flag does not apply to '.replacement'
4137 * only to .rdev, so make sure to check that*/
4138 struct md_rdev *rdev2 = rcu_dereference(
4139 conf->disks[i].rdev);
4140 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
b84db560 4141 s->handle_bad_blocks = 1;
14a75d3e 4142 atomic_inc(&rdev2->nr_pending);
b84db560
N		 4143 } else
4144 clear_bit(R5_MadeGood, &dev->flags);
4145 }
977df362
N		 4146 if (test_bit(R5_MadeGoodRepl, &dev->flags)) {
4147 struct md_rdev *rdev2 = rcu_dereference(
4148 conf->disks[i].replacement);
4149 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
4150 s->handle_bad_blocks = 1;
4151 atomic_inc(&rdev2->nr_pending);
4152 } else
4153 clear_bit(R5_MadeGoodRepl, &dev->flags);
4154 }
415e72d0 4155 if (!test_bit(R5_Insync, &dev->flags)) {
16a53ecc
N		 4156 /* The ReadError flag will just be confusing now */
4157 clear_bit(R5_ReadError, &dev->flags);
4158 clear_bit(R5_ReWrite, &dev->flags);
1da177e4 4159 }
415e72d0
N		 4160 if (test_bit(R5_ReadError, &dev->flags))
4161 clear_bit(R5_Insync, &dev->flags);
4162 if (!test_bit(R5_Insync, &dev->flags)) {
cc94015a
N		 4163 if (s->failed < 2)
4164 s->failed_num[s->failed] = i;
4165 s->failed++;
9a3e1101
N		 4166 if (rdev && !test_bit(Faulty, &rdev->flags))
4167 do_recovery = 1;
415e72d0 4168 }
1da177e4 4169 }
9a3e1101
N		 4170 if (test_bit(STRIPE_SYNCING, &sh->state)) {
4171 /* If there is a failed device being replaced,
4172 * we must be recovering.
4173 * else if we are after recovery_cp, we must be syncing
c6d2e084 4174 * else if MD_RECOVERY_REQUESTED is set, we also are syncing.
9a3e1101
N		 4175 * else we can only be replacing
4176 * sync and recovery both need to read all devices, and so
4177 * use the same flag.
4178 */
4179 if (do_recovery ||
c6d2e084 4180 sh->sector >= conf->mddev->recovery_cp ||
4181 test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery)))
9a3e1101
N		 4182 s->syncing = 1;
4183 else
4184 s->replacing = 1;
4185 }
1da177e4 4186 rcu_read_unlock();
cc94015a
N		 4187}
4188
59fc630b 4189static int clear_batch_ready(struct stripe_head *sh)
4190{
b15a9dbd
N		 4191 /* Return '1' if this is a member of batch, or
4192 * '0' if it is a lone stripe or a head which can now be
4193 * handled.
4194 */
59fc630b 4195 struct stripe_head *tmp;
4196 if (!test_and_clear_bit(STRIPE_BATCH_READY, &sh->state))
b15a9dbd 4197 return (sh->batch_head && sh->batch_head != sh);
59fc630b 4198 spin_lock(&sh->stripe_lock);
4199 if (!sh->batch_head) {
4200 spin_unlock(&sh->stripe_lock);
4201 return 0;
4202 }
4203
4204 /*
4205 * this stripe could be added to a batch list before we check
4206 * BATCH_READY, skips it
4207 */
4208 if (sh->batch_head != sh) {
4209 spin_unlock(&sh->stripe_lock);
4210 return 1;
4211 }
4212 spin_lock(&sh->batch_lock);
4213 list_for_each_entry(tmp, &sh->batch_list, batch_list)
4214 clear_bit(STRIPE_BATCH_READY, &tmp->state);
4215 spin_unlock(&sh->batch_lock);
4216 spin_unlock(&sh->stripe_lock);
4217
4218 /*
4219 * BATCH_READY is cleared, no new stripes can be added.
4220 * batch_list can be accessed without lock
4221 */
4222 return 0;
4223}
4224
3960ce79
N		 4225static void break_stripe_batch_list(struct stripe_head *head_sh,
4226 unsigned long handle_flags)
72ac7330 4227{
4e3d62ff 4228 struct stripe_head *sh, *next;
72ac7330 4229 int i;
fb642b92 4230 int do_wakeup = 0;
72ac7330 4231
bb27051f
N		 4232 list_for_each_entry_safe(sh, next, &head_sh->batch_list, batch_list) {
4233
72ac7330 4234 list_del_init(&sh->batch_list);
4235
1b956f7a
N		 4236 WARN_ON_ONCE(sh->state & ((1 << STRIPE_ACTIVE) |
4237 (1 << STRIPE_SYNCING) |
4238 (1 << STRIPE_REPLACED) |
4239 (1 << STRIPE_PREREAD_ACTIVE) |
4240 (1 << STRIPE_DELAYED) |
4241 (1 << STRIPE_BIT_DELAY) |
4242 (1 << STRIPE_FULL_WRITE) |
4243 (1 << STRIPE_BIOFILL_RUN) |
4244 (1 << STRIPE_COMPUTE_RUN) |
4245 (1 << STRIPE_OPS_REQ_PENDING) |
4246 (1 << STRIPE_DISCARD) |
4247 (1 << STRIPE_BATCH_READY) |
4248 (1 << STRIPE_BATCH_ERR) |
4249 (1 << STRIPE_BITMAP_PENDING)));
4250 WARN_ON_ONCE(head_sh->state & ((1 << STRIPE_DISCARD) |
4251 (1 << STRIPE_REPLACED)));
4252
4253 set_mask_bits(&sh->state, ~(STRIPE_EXPAND_SYNC_FLAGS |
4254 (1 << STRIPE_DEGRADED)),
4255 head_sh->state & (1 << STRIPE_INSYNC));
4256
72ac7330 4257 sh->check_state = head_sh->check_state;
4258 sh->reconstruct_state = head_sh->reconstruct_state;
fb642b92
N		 4259 for (i = 0; i < sh->disks; i++) {
4260 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
4261 do_wakeup = 1;
72ac7330 4262 sh->dev[i].flags = head_sh->dev[i].flags &
4263 (~((1 << R5_WriteError) | (1 << R5_Overlap)));
fb642b92 4264 }
72ac7330 4265 spin_lock_irq(&sh->stripe_lock);
4266 sh->batch_head = NULL;
4267 spin_unlock_irq(&sh->stripe_lock);
3960ce79
N		 4268 if (handle_flags == 0 ||
4269 sh->state & handle_flags)
4270 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 4271 raid5_release_stripe(sh);
72ac7330 4272 }
fb642b92
N		 4273 spin_lock_irq(&head_sh->stripe_lock);
4274 head_sh->batch_head = NULL;
4275 spin_unlock_irq(&head_sh->stripe_lock);
4276 for (i = 0; i < head_sh->disks; i++)
4277 if (test_and_clear_bit(R5_Overlap, &head_sh->dev[i].flags))
4278 do_wakeup = 1;
3960ce79
N		 4279 if (head_sh->state & handle_flags)
4280 set_bit(STRIPE_HANDLE, &head_sh->state);
fb642b92
N		 4281
4282 if (do_wakeup)
4283 wake_up(&head_sh->raid_conf->wait_for_overlap);
72ac7330 4284}
4285
cc94015a
N		 4286static void handle_stripe(struct stripe_head *sh)
4287{
4288 struct stripe_head_state s;
d1688a6d 4289 struct r5conf *conf = sh->raid_conf;
3687c061 4290 int i;
84789554
N		 4291 int prexor;
4292 int disks = sh->disks;
474af965 4293 struct r5dev *pdev, *qdev;
cc94015a
N		 4294
4295 clear_bit(STRIPE_HANDLE, &sh->state);
257a4b42 4296 if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) {
cc94015a
N		 4297 /* already being handled, ensure it gets handled
4298 * again when current action finishes */
4299 set_bit(STRIPE_HANDLE, &sh->state);
4300 return;
4301 }
4302
59fc630b 4303 if (clear_batch_ready(sh) ) {
4304 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
4305 return;
4306 }
4307
4e3d62ff 4308 if (test_and_clear_bit(STRIPE_BATCH_ERR, &sh->state))
3960ce79 4309 break_stripe_batch_list(sh, 0);
72ac7330 4310
dabc4ec6 4311 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) && !sh->batch_head) {
f8dfcffd
N		 4312 spin_lock(&sh->stripe_lock);
4313 /* Cannot process 'sync' concurrently with 'discard' */
4314 if (!test_bit(STRIPE_DISCARD, &sh->state) &&
4315 test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
4316 set_bit(STRIPE_SYNCING, &sh->state);
4317 clear_bit(STRIPE_INSYNC, &sh->state);
f94c0b66 4318 clear_bit(STRIPE_REPLACED, &sh->state);
f8dfcffd
N		 4319 }
4320 spin_unlock(&sh->stripe_lock);
cc94015a
N		 4321 }
4322 clear_bit(STRIPE_DELAYED, &sh->state);
4323
4324 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
4325 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
4326 (unsigned long long)sh->sector, sh->state,
4327 atomic_read(&sh->count), sh->pd_idx, sh->qd_idx,
4328 sh->check_state, sh->reconstruct_state);
3687c061 4329
acfe726b 4330 analyse_stripe(sh, &s);
c5a31000 4331
b70abcb2
SL		 4332 if (test_bit(STRIPE_LOG_TRAPPED, &sh->state))
4333 goto finish;
4334
bc2607f3
N		 4335 if (s.handle_bad_blocks) {
4336 set_bit(STRIPE_HANDLE, &sh->state);
4337 goto finish;
4338 }
4339
474af965
N		 4340 if (unlikely(s.blocked_rdev)) {
4341 if (s.syncing || s.expanding || s.expanded ||
9a3e1101 4342 s.replacing || s.to_write || s.written) {
474af965
N		 4343 set_bit(STRIPE_HANDLE, &sh->state);
4344 goto finish;
4345 }
4346 /* There is nothing for the blocked_rdev to block */
4347 rdev_dec_pending(s.blocked_rdev, conf->mddev);
4348 s.blocked_rdev = NULL;
4349 }
4350
4351 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
4352 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
4353 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
4354 }
4355
4356 pr_debug("locked=%d uptodate=%d to_read=%d"
4357 " to_write=%d failed=%d failed_num=%d,%d\n",
4358 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
4359 s.failed_num[0], s.failed_num[1]);
4360 /* check if the array has lost more than max_degraded devices and,
4361 * if so, some requests might need to be failed.
4362 */
6e74a9cf 4363 if (s.failed > conf->max_degraded || s.log_failed) {
9a3f530f
N		 4364 sh->check_state = 0;
4365 sh->reconstruct_state = 0;
626f2092 4366 break_stripe_batch_list(sh, 0);
9a3f530f
N		 4367 if (s.to_read+s.to_write+s.written)
4368 handle_failed_stripe(conf, sh, &s, disks, &s.return_bi);
9a3e1101 4369 if (s.syncing + s.replacing)
9a3f530f
N		 4370 handle_failed_sync(conf, sh, &s);
4371 }
474af965 4372
84789554
N		 4373 /* Now we check to see if any write operations have recently
4374 * completed
4375 */
4376 prexor = 0;
4377 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
4378 prexor = 1;
4379 if (sh->reconstruct_state == reconstruct_state_drain_result ||
4380 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
4381 sh->reconstruct_state = reconstruct_state_idle;
4382
4383 /* All the 'written' buffers and the parity block are ready to
4384 * be written back to disk
4385 */
9e444768
SL		 4386 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) &&
4387 !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags));
84789554 4388 BUG_ON(sh->qd_idx >= 0 &&
9e444768
SL		 4389 !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) &&
4390 !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags));
84789554
N		 4391 for (i = disks; i--; ) {
4392 struct r5dev *dev = &sh->dev[i];
4393 if (test_bit(R5_LOCKED, &dev->flags) &&
4394 (i == sh->pd_idx || i == sh->qd_idx ||
4395 dev->written)) {
4396 pr_debug("Writing block %d\n", i);
4397 set_bit(R5_Wantwrite, &dev->flags);
4398 if (prexor)
4399 continue;
9c4bdf69
N		 4400 if (s.failed > 1)
4401 continue;
84789554
N		 4402 if (!test_bit(R5_Insync, &dev->flags) ||
4403 ((i == sh->pd_idx || i == sh->qd_idx) &&
4404 s.failed == 0))
4405 set_bit(STRIPE_INSYNC, &sh->state);
4406 }
4407 }
4408 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4409 s.dec_preread_active = 1;
4410 }
4411
ef5b7c69
N		 4412 /*
4413 * might be able to return some write requests if the parity blocks
4414 * are safe, or on a failed drive
4415 */
4416 pdev = &sh->dev[sh->pd_idx];
4417 s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx)
4418 || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx);
4419 qdev = &sh->dev[sh->qd_idx];
4420 s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx)
4421 || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx)
4422 || conf->level < 6;
4423
4424 if (s.written &&
4425 (s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
4426 && !test_bit(R5_LOCKED, &pdev->flags)
4427 && (test_bit(R5_UPTODATE, &pdev->flags) ||
4428 test_bit(R5_Discard, &pdev->flags))))) &&
4429 (s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
4430 && !test_bit(R5_LOCKED, &qdev->flags)
4431 && (test_bit(R5_UPTODATE, &qdev->flags) ||
4432 test_bit(R5_Discard, &qdev->flags))))))
4433 handle_stripe_clean_event(conf, sh, disks, &s.return_bi);
4434
4435 /* Now we might consider reading some blocks, either to check/generate
4436 * parity, or to satisfy requests
4437 * or to load a block that is being partially written.
4438 */
4439 if (s.to_read || s.non_overwrite
4440 || (conf->level == 6 && s.to_write && s.failed)
4441 || (s.syncing && (s.uptodate + s.compute < disks))
4442 || s.replacing
4443 || s.expanding)
4444 handle_stripe_fill(sh, &s, disks);
4445
84789554
N		 4446 /* Now to consider new write requests and what else, if anything
4447 * should be read. We do not handle new writes when:
4448 * 1/ A 'write' operation (copy+xor) is already in flight.
4449 * 2/ A 'check' operation is in flight, as it may clobber the parity
4450 * block.
4451 */
4452 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
4453 handle_stripe_dirtying(conf, sh, &s, disks);
4454
4455 /* maybe we need to check and possibly fix the parity for this stripe
4456 * Any reads will already have been scheduled, so we just see if enough
4457 * data is available. The parity check is held off while parity
4458 * dependent operations are in flight.
4459 */
4460 if (sh->check_state ||
4461 (s.syncing && s.locked == 0 &&
4462 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
4463 !test_bit(STRIPE_INSYNC, &sh->state))) {
4464 if (conf->level == 6)
4465 handle_parity_checks6(conf, sh, &s, disks);
4466 else
4467 handle_parity_checks5(conf, sh, &s, disks);
4468 }
c5a31000 4469
f94c0b66
N		 4470 if ((s.replacing || s.syncing) && s.locked == 0
4471 && !test_bit(STRIPE_COMPUTE_RUN, &sh->state)
4472 && !test_bit(STRIPE_REPLACED, &sh->state)) {
9a3e1101
N		 4473 /* Write out to replacement devices where possible */
4474 for (i = 0; i < conf->raid_disks; i++)
f94c0b66
N		 4475 if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) {
4476 WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags));
9a3e1101
N		 4477 set_bit(R5_WantReplace, &sh->dev[i].flags);
4478 set_bit(R5_LOCKED, &sh->dev[i].flags);
4479 s.locked++;
4480 }
f94c0b66
N		 4481 if (s.replacing)
4482 set_bit(STRIPE_INSYNC, &sh->state);
4483 set_bit(STRIPE_REPLACED, &sh->state);
9a3e1101
N		 4484 }
4485 if ((s.syncing || s.replacing) && s.locked == 0 &&
f94c0b66 4486 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
9a3e1101 4487 test_bit(STRIPE_INSYNC, &sh->state)) {
c5a31000
N		 4488 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
4489 clear_bit(STRIPE_SYNCING, &sh->state);
f8dfcffd
N		 4490 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
4491 wake_up(&conf->wait_for_overlap);
c5a31000
N		 4492 }
4493
4494 /* If the failed drives are just a ReadError, then we might need
4495 * to progress the repair/check process
4496 */
4497 if (s.failed <= conf->max_degraded && !conf->mddev->ro)
4498 for (i = 0; i < s.failed; i++) {
4499 struct r5dev *dev = &sh->dev[s.failed_num[i]];
4500 if (test_bit(R5_ReadError, &dev->flags)
4501 && !test_bit(R5_LOCKED, &dev->flags)
4502 && test_bit(R5_UPTODATE, &dev->flags)
4503 ) {
4504 if (!test_bit(R5_ReWrite, &dev->flags)) {
4505 set_bit(R5_Wantwrite, &dev->flags);
4506 set_bit(R5_ReWrite, &dev->flags);
4507 set_bit(R5_LOCKED, &dev->flags);
4508 s.locked++;
4509 } else {
4510 /* let's read it back */
4511 set_bit(R5_Wantread, &dev->flags);
4512 set_bit(R5_LOCKED, &dev->flags);
4513 s.locked++;
4514 }
4515 }
4516 }
4517
3687c061
N		 4518 /* Finish reconstruct operations initiated by the expansion process */
4519 if (sh->reconstruct_state == reconstruct_state_result) {
4520 struct stripe_head *sh_src
6d036f7d 4521 = raid5_get_active_stripe(conf, sh->sector, 1, 1, 1);
3687c061
N		 4522 if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) {
4523 /* sh cannot be written until sh_src has been read.
4524 * so arrange for sh to be delayed a little
4525 */
4526 set_bit(STRIPE_DELAYED, &sh->state);
4527 set_bit(STRIPE_HANDLE, &sh->state);
4528 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
4529 &sh_src->state))
4530 atomic_inc(&conf->preread_active_stripes);
6d036f7d 4531 raid5_release_stripe(sh_src);
3687c061
N		 4532 goto finish;
4533 }
4534 if (sh_src)
6d036f7d 4535 raid5_release_stripe(sh_src);
3687c061
N		 4536
4537 sh->reconstruct_state = reconstruct_state_idle;
4538 clear_bit(STRIPE_EXPANDING, &sh->state);
4539 for (i = conf->raid_disks; i--; ) {
4540 set_bit(R5_Wantwrite, &sh->dev[i].flags);
4541 set_bit(R5_LOCKED, &sh->dev[i].flags);
4542 s.locked++;
4543 }
4544 }
f416885e 4545
3687c061
N		 4546 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
4547 !sh->reconstruct_state) {
4548 /* Need to write out all blocks after computing parity */
4549 sh->disks = conf->raid_disks;
4550 stripe_set_idx(sh->sector, conf, 0, sh);
4551 schedule_reconstruction(sh, &s, 1, 1);
4552 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
4553 clear_bit(STRIPE_EXPAND_READY, &sh->state);
4554 atomic_dec(&conf->reshape_stripes);
4555 wake_up(&conf->wait_for_overlap);
4556 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
4557 }
4558
4559 if (s.expanding && s.locked == 0 &&
4560 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
4561 handle_stripe_expansion(conf, sh);
16a53ecc 4562
3687c061 4563finish:
6bfe0b49 4564 /* wait for this device to become unblocked */
5f066c63
N		 4565 if (unlikely(s.blocked_rdev)) {
4566 if (conf->mddev->external)
4567 md_wait_for_blocked_rdev(s.blocked_rdev,
4568 conf->mddev);
4569 else
4570 /* Internal metadata will immediately
4571 * be written by raid5d, so we don't
4572 * need to wait here.
4573 */
4574 rdev_dec_pending(s.blocked_rdev,
4575 conf->mddev);
4576 }
6bfe0b49 4577
bc2607f3
N		 4578 if (s.handle_bad_blocks)
4579 for (i = disks; i--; ) {
3cb03002 4580 struct md_rdev *rdev;
bc2607f3
N		 4581 struct r5dev *dev = &sh->dev[i];
4582 if (test_and_clear_bit(R5_WriteError, &dev->flags)) {
4583 /* We own a safe reference to the rdev */
4584 rdev = conf->disks[i].rdev;
4585 if (!rdev_set_badblocks(rdev, sh->sector,
4586 STRIPE_SECTORS, 0))
4587 md_error(conf->mddev, rdev);
4588 rdev_dec_pending(rdev, conf->mddev);
4589 }
b84db560
N		 4590 if (test_and_clear_bit(R5_MadeGood, &dev->flags)) {
4591 rdev = conf->disks[i].rdev;
4592 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 4593 STRIPE_SECTORS, 0);
b84db560
N		 4594 rdev_dec_pending(rdev, conf->mddev);
4595 }
977df362
N		 4596 if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) {
4597 rdev = conf->disks[i].replacement;
dd054fce
N		 4598 if (!rdev)
4599 /* rdev have been moved down */
4600 rdev = conf->disks[i].rdev;
977df362 4601 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 4602 STRIPE_SECTORS, 0);
977df362
N		 4603 rdev_dec_pending(rdev, conf->mddev);
4604 }
bc2607f3
N		 4605 }
4606
6c0069c0
YT		 4607 if (s.ops_request)
4608 raid_run_ops(sh, s.ops_request);
4609
f0e43bcd 4610 ops_run_io(sh, &s);
16a53ecc 4611
c5709ef6 4612 if (s.dec_preread_active) {
729a1866 4613 /* We delay this until after ops_run_io so that if make_request
e9c7469b 4614 * is waiting on a flush, it won't continue until the writes
729a1866
N		 4615 * have actually been submitted.
4616 */
4617 atomic_dec(&conf->preread_active_stripes);
4618 if (atomic_read(&conf->preread_active_stripes) <
4619 IO_THRESHOLD)
4620 md_wakeup_thread(conf->mddev->thread);
4621 }
4622
c3cce6cd
N		 4623 if (!bio_list_empty(&s.return_bi)) {
4624 if (test_bit(MD_CHANGE_PENDING, &conf->mddev->flags)) {
4625 spin_lock_irq(&conf->device_lock);
4626 bio_list_merge(&conf->return_bi, &s.return_bi);
4627 spin_unlock_irq(&conf->device_lock);
4628 md_wakeup_thread(conf->mddev->thread);
4629 } else
4630 return_io(&s.return_bi);
4631 }
16a53ecc 4632
257a4b42 4633 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
16a53ecc
N		 4634}
4635
d1688a6d 4636static void raid5_activate_delayed(struct r5conf *conf)
16a53ecc
N		 4637{
4638 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
4639 while (!list_empty(&conf->delayed_list)) {
4640 struct list_head *l = conf->delayed_list.next;
4641 struct stripe_head *sh;
4642 sh = list_entry(l, struct stripe_head, lru);
4643 list_del_init(l);
4644 clear_bit(STRIPE_DELAYED, &sh->state);
4645 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4646 atomic_inc(&conf->preread_active_stripes);
8b3e6cdc 4647 list_add_tail(&sh->lru, &conf->hold_list);
851c30c9 4648 raid5_wakeup_stripe_thread(sh);
16a53ecc 4649 }
482c0834 4650 }
16a53ecc
N		 4651}
4652
566c09c5
SL		 4653static void activate_bit_delay(struct r5conf *conf,
4654 struct list_head *temp_inactive_list)
16a53ecc
N		 4655{
4656 /* device_lock is held */
4657 struct list_head head;
4658 list_add(&head, &conf->bitmap_list);
4659 list_del_init(&conf->bitmap_list);
4660 while (!list_empty(&head)) {
4661 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
566c09c5 4662 int hash;
16a53ecc
N		 4663 list_del_init(&sh->lru);
4664 atomic_inc(&sh->count);
566c09c5
SL		 4665 hash = sh->hash_lock_index;
4666 __release_stripe(conf, sh, &temp_inactive_list[hash]);
16a53ecc
N		 4667 }
4668}
4669
5c675f83 4670static int raid5_congested(struct mddev *mddev, int bits)
f022b2fd 4671{
d1688a6d 4672 struct r5conf *conf = mddev->private;
f022b2fd
N		 4673
4674 /* No difference between reads and writes. Just check
4675 * how busy the stripe_cache is
4676 */
3fa841d7 4677
5423399a 4678 if (test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state))
f022b2fd
N		 4679 return 1;
4680 if (conf->quiesce)
4681 return 1;
4bda556a 4682 if (atomic_read(&conf->empty_inactive_list_nr))
f022b2fd
N		 4683 return 1;
4684
4685 return 0;
4686}
4687
fd01b88c 4688static int in_chunk_boundary(struct mddev *mddev, struct bio *bio)
f679623f 4689{
3cb5edf4 4690 struct r5conf *conf = mddev->private;
4f024f37 4691 sector_t sector = bio->bi_iter.bi_sector + get_start_sect(bio->bi_bdev);
3cb5edf4 4692 unsigned int chunk_sectors;
aa8b57aa 4693 unsigned int bio_sectors = bio_sectors(bio);
f679623f 4694
3cb5edf4 4695 chunk_sectors = min(conf->chunk_sectors, conf->prev_chunk_sectors);
f679623f
RBJ		 4696 return chunk_sectors >=
4697 ((sector & (chunk_sectors - 1)) + bio_sectors);
4698}
4699
46031f9a
RBJ		 4700/*
4701 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
4702 * later sampled by raid5d.
4703 */
d1688a6d 4704static void add_bio_to_retry(struct bio *bi,struct r5conf *conf)
46031f9a
RBJ		 4705{
4706 unsigned long flags;
4707
4708 spin_lock_irqsave(&conf->device_lock, flags);
4709
4710 bi->bi_next = conf->retry_read_aligned_list;
4711 conf->retry_read_aligned_list = bi;
4712
4713 spin_unlock_irqrestore(&conf->device_lock, flags);
4714 md_wakeup_thread(conf->mddev->thread);
4715}
4716
d1688a6d 4717static struct bio *remove_bio_from_retry(struct r5conf *conf)
46031f9a
RBJ		 4718{
4719 struct bio *bi;
4720
4721 bi = conf->retry_read_aligned;
4722 if (bi) {
4723 conf->retry_read_aligned = NULL;
4724 return bi;
4725 }
4726 bi = conf->retry_read_aligned_list;
4727 if(bi) {
387bb173 4728 conf->retry_read_aligned_list = bi->bi_next;
46031f9a 4729 bi->bi_next = NULL;
960e739d
JA		 4730 /*
4731 * this sets the active strip count to 1 and the processed
4732 * strip count to zero (upper 8 bits)
4733 */
e7836bd6 4734 raid5_set_bi_stripes(bi, 1); /* biased count of active stripes */
46031f9a
RBJ		 4735 }
4736
4737 return bi;
4738}
4739
f679623f
RBJ		 4740/*
4741 * The "raid5_align_endio" should check if the read succeeded and if it
4742 * did, call bio_endio on the original bio (having bio_put the new bio
4743 * first).
4744 * If the read failed..
4745 */
4246a0b6 4746static void raid5_align_endio(struct bio *bi)
f679623f
RBJ		 4747{
4748 struct bio* raid_bi = bi->bi_private;
fd01b88c 4749 struct mddev *mddev;
d1688a6d 4750 struct r5conf *conf;
3cb03002 4751 struct md_rdev *rdev;
9b81c842 4752 int error = bi->bi_error;
46031f9a 4753
f679623f 4754 bio_put(bi);
46031f9a 4755
46031f9a
RBJ		 4756 rdev = (void*)raid_bi->bi_next;
4757 raid_bi->bi_next = NULL;
2b7f2228
N		 4758 mddev = rdev->mddev;
4759 conf = mddev->private;
46031f9a
RBJ		 4760
4761 rdev_dec_pending(rdev, conf->mddev);
4762
9b81c842 4763 if (!error) {
0a82a8d1
LT		 4764 trace_block_bio_complete(bdev_get_queue(raid_bi->bi_bdev),
4765 raid_bi, 0);
4246a0b6 4766 bio_endio(raid_bi);
46031f9a 4767 if (atomic_dec_and_test(&conf->active_aligned_reads))
b1b46486 4768 wake_up(&conf->wait_for_quiescent);
6712ecf8 4769 return;
46031f9a
RBJ		 4770 }
4771
45b4233c 4772 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
46031f9a
RBJ		 4773
4774 add_bio_to_retry(raid_bi, conf);
f679623f
RBJ		 4775}
4776
7ef6b12a 4777static int raid5_read_one_chunk(struct mddev *mddev, struct bio *raid_bio)
f679623f 4778{
d1688a6d 4779 struct r5conf *conf = mddev->private;
8553fe7e 4780 int dd_idx;
f679623f 4781 struct bio* align_bi;
3cb03002 4782 struct md_rdev *rdev;
671488cc 4783 sector_t end_sector;
f679623f
RBJ		 4784
4785 if (!in_chunk_boundary(mddev, raid_bio)) {
7ef6b12a 4786 pr_debug("%s: non aligned\n", __func__);
f679623f
RBJ		 4787 return 0;
4788 }
4789 /*
a167f663 4790 * use bio_clone_mddev to make a copy of the bio
f679623f 4791 */
a167f663 4792 align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev);
f679623f
RBJ		 4793 if (!align_bi)
4794 return 0;
4795 /*
4796 * set bi_end_io to a new function, and set bi_private to the
4797 * original bio.
4798 */
4799 align_bi->bi_end_io = raid5_align_endio;
4800 align_bi->bi_private = raid_bio;
4801 /*
4802 * compute position
4803 */
4f024f37
KO		 4804 align_bi->bi_iter.bi_sector =
4805 raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector,
4806 0, &dd_idx, NULL);
f679623f 4807
f73a1c7d 4808 end_sector = bio_end_sector(align_bi);
f679623f 4809 rcu_read_lock();
671488cc
N		 4810 rdev = rcu_dereference(conf->disks[dd_idx].replacement);
4811 if (!rdev || test_bit(Faulty, &rdev->flags) ||
4812 rdev->recovery_offset < end_sector) {
4813 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
4814 if (rdev &&
4815 (test_bit(Faulty, &rdev->flags) ||
4816 !(test_bit(In_sync, &rdev->flags) ||
4817 rdev->recovery_offset >= end_sector)))
4818 rdev = NULL;
4819 }
4820 if (rdev) {
31c176ec
N		 4821 sector_t first_bad;
4822 int bad_sectors;
4823
f679623f
RBJ		 4824 atomic_inc(&rdev->nr_pending);
4825 rcu_read_unlock();
46031f9a
RBJ		 4826 raid_bio->bi_next = (void*)rdev;
4827 align_bi->bi_bdev = rdev->bdev;
b7c44ed9 4828 bio_clear_flag(align_bi, BIO_SEG_VALID);
46031f9a 4829
7140aafc 4830 if (is_badblock(rdev, align_bi->bi_iter.bi_sector,
4f024f37 4831 bio_sectors(align_bi),
31c176ec 4832 &first_bad, &bad_sectors)) {
387bb173
NB		 4833 bio_put(align_bi);
4834 rdev_dec_pending(rdev, mddev);
4835 return 0;
4836 }
4837
6c0544e2 4838 /* No reshape active, so we can trust rdev->data_offset */
4f024f37 4839 align_bi->bi_iter.bi_sector += rdev->data_offset;
6c0544e2 4840
46031f9a 4841 spin_lock_irq(&conf->device_lock);
b1b46486 4842 wait_event_lock_irq(conf->wait_for_quiescent,
46031f9a 4843 conf->quiesce == 0,
eed8c02e 4844 conf->device_lock);
46031f9a
RBJ		 4845 atomic_inc(&conf->active_aligned_reads);
4846 spin_unlock_irq(&conf->device_lock);
4847
e3620a3a
JB		 4848 if (mddev->gendisk)
4849 trace_block_bio_remap(bdev_get_queue(align_bi->bi_bdev),
4850 align_bi, disk_devt(mddev->gendisk),
4f024f37 4851 raid_bio->bi_iter.bi_sector);
f679623f
RBJ		 4852 generic_make_request(align_bi);
4853 return 1;
4854 } else {
4855 rcu_read_unlock();
46031f9a 4856 bio_put(align_bi);
f679623f
RBJ		 4857 return 0;
4858 }
4859}
4860
7ef6b12a
ML		 4861static struct bio *chunk_aligned_read(struct mddev *mddev, struct bio *raid_bio)
4862{
4863 struct bio *split;
4864
4865 do {
4866 sector_t sector = raid_bio->bi_iter.bi_sector;
4867 unsigned chunk_sects = mddev->chunk_sectors;
4868 unsigned sectors = chunk_sects - (sector & (chunk_sects-1));
4869
4870 if (sectors < bio_sectors(raid_bio)) {
4871 split = bio_split(raid_bio, sectors, GFP_NOIO, fs_bio_set);
4872 bio_chain(split, raid_bio);
4873 } else
4874 split = raid_bio;
4875
4876 if (!raid5_read_one_chunk(mddev, split)) {
4877 if (split != raid_bio)
4878 generic_make_request(raid_bio);
4879 return split;
4880 }
4881 } while (split != raid_bio);
4882
4883 return NULL;
4884}
4885
8b3e6cdc
DW		 4886/* __get_priority_stripe - get the next stripe to process
4887 *
4888 * Full stripe writes are allowed to pass preread active stripes up until
4889 * the bypass_threshold is exceeded. In general the bypass_count
4890 * increments when the handle_list is handled before the hold_list; however, it
4891 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
4892 * stripe with in flight i/o. The bypass_count will be reset when the
4893 * head of the hold_list has changed, i.e. the head was promoted to the
4894 * handle_list.
4895 */
851c30c9 4896static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group)
8b3e6cdc 4897{
851c30c9
SL		 4898 struct stripe_head *sh = NULL, *tmp;
4899 struct list_head *handle_list = NULL;
bfc90cb0 4900 struct r5worker_group *wg = NULL;
851c30c9
SL		 4901
4902 if (conf->worker_cnt_per_group == 0) {
4903 handle_list = &conf->handle_list;
4904 } else if (group != ANY_GROUP) {
4905 handle_list = &conf->worker_groups[group].handle_list;
bfc90cb0 4906 wg = &conf->worker_groups[group];
851c30c9
SL		 4907 } else {
4908 int i;
4909 for (i = 0; i < conf->group_cnt; i++) {
4910 handle_list = &conf->worker_groups[i].handle_list;
bfc90cb0 4911 wg = &conf->worker_groups[i];
851c30c9
SL		 4912 if (!list_empty(handle_list))
4913 break;
4914 }
4915 }
8b3e6cdc
DW		 4916
4917 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
4918 __func__,
851c30c9 4919 list_empty(handle_list) ? "empty" : "busy",
8b3e6cdc
DW		 4920 list_empty(&conf->hold_list) ? "empty" : "busy",
4921 atomic_read(&conf->pending_full_writes), conf->bypass_count);
4922
851c30c9
SL		 4923 if (!list_empty(handle_list)) {
4924 sh = list_entry(handle_list->next, typeof(*sh), lru);
8b3e6cdc
DW		 4925
4926 if (list_empty(&conf->hold_list))
4927 conf->bypass_count = 0;
4928 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
4929 if (conf->hold_list.next == conf->last_hold)
4930 conf->bypass_count++;
4931 else {
4932 conf->last_hold = conf->hold_list.next;
4933 conf->bypass_count -= conf->bypass_threshold;
4934 if (conf->bypass_count < 0)
4935 conf->bypass_count = 0;
4936 }
4937 }
4938 } else if (!list_empty(&conf->hold_list) &&
4939 ((conf->bypass_threshold &&
4940 conf->bypass_count > conf->bypass_threshold) ||
4941 atomic_read(&conf->pending_full_writes) == 0)) {
851c30c9
SL		 4942
4943 list_for_each_entry(tmp, &conf->hold_list, lru) {
4944 if (conf->worker_cnt_per_group == 0 ||
4945 group == ANY_GROUP ||
4946 !cpu_online(tmp->cpu) ||
4947 cpu_to_group(tmp->cpu) == group) {
4948 sh = tmp;
4949 break;
4950 }
4951 }
4952
4953 if (sh) {
4954 conf->bypass_count -= conf->bypass_threshold;
4955 if (conf->bypass_count < 0)
4956 conf->bypass_count = 0;
4957 }
bfc90cb0 4958 wg = NULL;
851c30c9
SL		 4959 }
4960
4961 if (!sh)
8b3e6cdc
DW		 4962 return NULL;
4963
bfc90cb0
SL		 4964 if (wg) {
4965 wg->stripes_cnt--;
4966 sh->group = NULL;
4967 }
8b3e6cdc 4968 list_del_init(&sh->lru);
c7a6d35e 4969 BUG_ON(atomic_inc_return(&sh->count) != 1);
8b3e6cdc
DW		 4970 return sh;
4971}
f679623f 4972
8811b596
SL		 4973struct raid5_plug_cb {
4974 struct blk_plug_cb cb;
4975 struct list_head list;
566c09c5 4976 struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
8811b596
SL		 4977};
4978
4979static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule)
4980{
4981 struct raid5_plug_cb *cb = container_of(
4982 blk_cb, struct raid5_plug_cb, cb);
4983 struct stripe_head *sh;
4984 struct mddev *mddev = cb->cb.data;
4985 struct r5conf *conf = mddev->private;
a9add5d9 4986 int cnt = 0;
566c09c5 4987 int hash;
8811b596
SL		 4988
4989 if (cb->list.next && !list_empty(&cb->list)) {
4990 spin_lock_irq(&conf->device_lock);
4991 while (!list_empty(&cb->list)) {
4992 sh = list_first_entry(&cb->list, struct stripe_head, lru);
4993 list_del_init(&sh->lru);
4994 /*
4995 * avoid race release_stripe_plug() sees
4996 * STRIPE_ON_UNPLUG_LIST clear but the stripe
4997 * is still in our list
4998 */
4e857c58 4999 smp_mb__before_atomic();
8811b596 5000 clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state);
773ca82f
SL		 5001 /*
5002 * STRIPE_ON_RELEASE_LIST could be set here. In that
5003 * case, the count is always > 1 here
5004 */
566c09c5
SL		 5005 hash = sh->hash_lock_index;
5006 __release_stripe(conf, sh, &cb->temp_inactive_list[hash]);
a9add5d9 5007 cnt++;
8811b596
SL		 5008 }
5009 spin_unlock_irq(&conf->device_lock);
5010 }
566c09c5
SL		 5011 release_inactive_stripe_list(conf, cb->temp_inactive_list,
5012 NR_STRIPE_HASH_LOCKS);
e3620a3a
JB		 5013 if (mddev->queue)
5014 trace_block_unplug(mddev->queue, cnt, !from_schedule);
8811b596
SL		 5015 kfree(cb);
5016}
5017
5018static void release_stripe_plug(struct mddev *mddev,
5019 struct stripe_head *sh)
5020{
5021 struct blk_plug_cb *blk_cb = blk_check_plugged(
5022 raid5_unplug, mddev,
5023 sizeof(struct raid5_plug_cb));
5024 struct raid5_plug_cb *cb;
5025
5026 if (!blk_cb) {
6d036f7d 5027 raid5_release_stripe(sh);
8811b596
SL		 5028 return;
5029 }
5030
5031 cb = container_of(blk_cb, struct raid5_plug_cb, cb);
5032
566c09c5
SL		 5033 if (cb->list.next == NULL) {
5034 int i;
8811b596 5035 INIT_LIST_HEAD(&cb->list);
566c09c5
SL		 5036 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5037 INIT_LIST_HEAD(cb->temp_inactive_list + i);
5038 }
8811b596
SL		 5039
5040 if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state))
5041 list_add_tail(&sh->lru, &cb->list);
5042 else
6d036f7d 5043 raid5_release_stripe(sh);
8811b596
SL		 5044}
5045
620125f2
SL		 5046static void make_discard_request(struct mddev *mddev, struct bio *bi)
5047{
5048 struct r5conf *conf = mddev->private;
5049 sector_t logical_sector, last_sector;
5050 struct stripe_head *sh;
5051 int remaining;
5052 int stripe_sectors;
5053
5054 if (mddev->reshape_position != MaxSector)
5055 /* Skip discard while reshape is happening */
5056 return;
5057
4f024f37
KO		 5058 logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1);
5059 last_sector = bi->bi_iter.bi_sector + (bi->bi_iter.bi_size>>9);
620125f2
SL		 5060
5061 bi->bi_next = NULL;
5062 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
5063
5064 stripe_sectors = conf->chunk_sectors *
5065 (conf->raid_disks - conf->max_degraded);
5066 logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector,
5067 stripe_sectors);
5068 sector_div(last_sector, stripe_sectors);
5069
5070 logical_sector *= conf->chunk_sectors;
5071 last_sector *= conf->chunk_sectors;
5072
5073 for (; logical_sector < last_sector;
5074 logical_sector += STRIPE_SECTORS) {
5075 DEFINE_WAIT(w);
5076 int d;
5077 again:
6d036f7d 5078 sh = raid5_get_active_stripe(conf, logical_sector, 0, 0, 0);
620125f2
SL		 5079 prepare_to_wait(&conf->wait_for_overlap, &w,
5080 TASK_UNINTERRUPTIBLE);
f8dfcffd
N		 5081 set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
5082 if (test_bit(STRIPE_SYNCING, &sh->state)) {
6d036f7d 5083 raid5_release_stripe(sh);
f8dfcffd
N		 5084 schedule();
5085 goto again;
5086 }
5087 clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
620125f2
SL		 5088 spin_lock_irq(&sh->stripe_lock);
5089 for (d = 0; d < conf->raid_disks; d++) {
5090 if (d == sh->pd_idx || d == sh->qd_idx)
5091 continue;
5092 if (sh->dev[d].towrite || sh->dev[d].toread) {
5093 set_bit(R5_Overlap, &sh->dev[d].flags);
5094 spin_unlock_irq(&sh->stripe_lock);
6d036f7d 5095 raid5_release_stripe(sh);
620125f2
SL		 5096 schedule();
5097 goto again;
5098 }
5099 }
f8dfcffd 5100 set_bit(STRIPE_DISCARD, &sh->state);
620125f2 5101 finish_wait(&conf->wait_for_overlap, &w);
7a87f434 5102 sh->overwrite_disks = 0;
620125f2
SL		 5103 for (d = 0; d < conf->raid_disks; d++) {
5104 if (d == sh->pd_idx || d == sh->qd_idx)
5105 continue;
5106 sh->dev[d].towrite = bi;
5107 set_bit(R5_OVERWRITE, &sh->dev[d].flags);
5108 raid5_inc_bi_active_stripes(bi);
7a87f434 5109 sh->overwrite_disks++;
620125f2
SL		 5110 }
5111 spin_unlock_irq(&sh->stripe_lock);
5112 if (conf->mddev->bitmap) {
5113 for (d = 0;
5114 d < conf->raid_disks - conf->max_degraded;
5115 d++)
5116 bitmap_startwrite(mddev->bitmap,
5117 sh->sector,
5118 STRIPE_SECTORS,
5119 0);
5120 sh->bm_seq = conf->seq_flush + 1;
5121 set_bit(STRIPE_BIT_DELAY, &sh->state);
5122 }
5123
5124 set_bit(STRIPE_HANDLE, &sh->state);
5125 clear_bit(STRIPE_DELAYED, &sh->state);
5126 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
5127 atomic_inc(&conf->preread_active_stripes);
5128 release_stripe_plug(mddev, sh);
5129 }
5130
5131 remaining = raid5_dec_bi_active_stripes(bi);
5132 if (remaining == 0) {
5133 md_write_end(mddev);
4246a0b6 5134 bio_endio(bi);
620125f2
SL		 5135 }
5136}
5137
b4fdcb02 5138static void make_request(struct mddev *mddev, struct bio * bi)
1da177e4 5139{
d1688a6d 5140 struct r5conf *conf = mddev->private;
911d4ee8 5141 int dd_idx;
1da177e4
LT		 5142 sector_t new_sector;
5143 sector_t logical_sector, last_sector;
5144 struct stripe_head *sh;
a362357b 5145 const int rw = bio_data_dir(bi);
49077326 5146 int remaining;
27c0f68f
SL		 5147 DEFINE_WAIT(w);
5148 bool do_prepare;
1da177e4 5149
e9c7469b 5150 if (unlikely(bi->bi_rw & REQ_FLUSH)) {
828cbe98
SL		 5151 int ret = r5l_handle_flush_request(conf->log, bi);
5152
5153 if (ret == 0)
5154 return;
5155 if (ret == -ENODEV) {
5156 md_flush_request(mddev, bi);
5157 return;
5158 }
5159 /* ret == -EAGAIN, fallback */
e5dcdd80
N		 5160 }
5161
3d310eb7 5162 md_write_start(mddev, bi);
06d91a5f 5163
9ffc8f7c
EM		 5164 /*
5165 * If array is degraded, better not do chunk aligned read because
5166 * later we might have to read it again in order to reconstruct
5167 * data on failed drives.
5168 */
5169 if (rw == READ && mddev->degraded == 0 &&
7ef6b12a
ML		 5170 mddev->reshape_position == MaxSector) {
5171 bi = chunk_aligned_read(mddev, bi);
5172 if (!bi)
5173 return;
5174 }
52488615 5175
620125f2
SL		 5176 if (unlikely(bi->bi_rw & REQ_DISCARD)) {
5177 make_discard_request(mddev, bi);
5178 return;
5179 }
5180
4f024f37 5181 logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1);
f73a1c7d 5182 last_sector = bio_end_sector(bi);
1da177e4
LT		 5183 bi->bi_next = NULL;
5184 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
06d91a5f 5185
27c0f68f 5186 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
1da177e4 5187 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
b5663ba4 5188 int previous;
c46501b2 5189 int seq;
b578d55f 5190
27c0f68f 5191 do_prepare = false;
7ecaa1e6 5192 retry:
c46501b2 5193 seq = read_seqcount_begin(&conf->gen_lock);
b5663ba4 5194 previous = 0;
27c0f68f
SL		 5195 if (do_prepare)
5196 prepare_to_wait(&conf->wait_for_overlap, &w,
5197 TASK_UNINTERRUPTIBLE);
b0f9ec04 5198 if (unlikely(conf->reshape_progress != MaxSector)) {
fef9c61f 5199 /* spinlock is needed as reshape_progress may be
df8e7f76
N		 5200 * 64bit on a 32bit platform, and so it might be
5201 * possible to see a half-updated value
aeb878b0 5202 * Of course reshape_progress could change after
df8e7f76
N		 5203 * the lock is dropped, so once we get a reference
5204 * to the stripe that we think it is, we will have
5205 * to check again.
5206 */
7ecaa1e6 5207 spin_lock_irq(&conf->device_lock);
2c810cdd 5208 if (mddev->reshape_backwards
fef9c61f
N		 5209 ? logical_sector < conf->reshape_progress
5210 : logical_sector >= conf->reshape_progress) {
b5663ba4
N		 5211 previous = 1;
5212 } else {
2c810cdd 5213 if (mddev->reshape_backwards
fef9c61f
N		 5214 ? logical_sector < conf->reshape_safe
5215 : logical_sector >= conf->reshape_safe) {
b578d55f
N		 5216 spin_unlock_irq(&conf->device_lock);
5217 schedule();
27c0f68f 5218 do_prepare = true;
b578d55f
N		 5219 goto retry;
5220 }
5221 }
7ecaa1e6
N		 5222 spin_unlock_irq(&conf->device_lock);
5223 }
16a53ecc 5224
112bf897
N		 5225 new_sector = raid5_compute_sector(conf, logical_sector,
5226 previous,
911d4ee8 5227 &dd_idx, NULL);
0c55e022 5228 pr_debug("raid456: make_request, sector %llu logical %llu\n",
c46501b2 5229 (unsigned long long)new_sector,
1da177e4
LT		 5230 (unsigned long long)logical_sector);
5231
6d036f7d 5232 sh = raid5_get_active_stripe(conf, new_sector, previous,
a8c906ca 5233 (bi->bi_rw&RWA_MASK), 0);
1da177e4 5234 if (sh) {
b0f9ec04 5235 if (unlikely(previous)) {
7ecaa1e6 5236 /* expansion might have moved on while waiting for a
df8e7f76
N		 5237 * stripe, so we must do the range check again.
5238 * Expansion could still move past after this
5239 * test, but as we are holding a reference to
5240 * 'sh', we know that if that happens,
5241 * STRIPE_EXPANDING will get set and the expansion
5242 * won't proceed until we finish with the stripe.
7ecaa1e6
N		 5243 */
5244 int must_retry = 0;
5245 spin_lock_irq(&conf->device_lock);
2c810cdd 5246 if (mddev->reshape_backwards
b0f9ec04
N		 5247 ? logical_sector >= conf->reshape_progress
5248 : logical_sector < conf->reshape_progress)
7ecaa1e6
N		 5249 /* mismatch, need to try again */
5250 must_retry = 1;
5251 spin_unlock_irq(&conf->device_lock);
5252 if (must_retry) {
6d036f7d 5253 raid5_release_stripe(sh);
7a3ab908 5254 schedule();
27c0f68f 5255 do_prepare = true;
7ecaa1e6
N		 5256 goto retry;
5257 }
5258 }
c46501b2
N		 5259 if (read_seqcount_retry(&conf->gen_lock, seq)) {
5260 /* Might have got the wrong stripe_head
5261 * by accident
5262 */
6d036f7d 5263 raid5_release_stripe(sh);
c46501b2
N		 5264 goto retry;
5265 }
e62e58a5 5266
ffd96e35 5267 if (rw == WRITE &&
a5c308d4 5268 logical_sector >= mddev->suspend_lo &&
e464eafd 5269 logical_sector < mddev->suspend_hi) {
6d036f7d 5270 raid5_release_stripe(sh);
e62e58a5
N		 5271 /* As the suspend_* range is controlled by
5272 * userspace, we want an interruptible
5273 * wait.
5274 */
5275 flush_signals(current);
5276 prepare_to_wait(&conf->wait_for_overlap,
5277 &w, TASK_INTERRUPTIBLE);
5278 if (logical_sector >= mddev->suspend_lo &&
27c0f68f 5279 logical_sector < mddev->suspend_hi) {
e62e58a5 5280 schedule();
27c0f68f
SL		 5281 do_prepare = true;
5282 }
e464eafd
N		 5283 goto retry;
5284 }
7ecaa1e6
N		 5285
5286 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
da41ba65 5287 !add_stripe_bio(sh, bi, dd_idx, rw, previous)) {
7ecaa1e6
N		 5288 /* Stripe is busy expanding or
5289 * add failed due to overlap. Flush everything
1da177e4
LT		 5290 * and wait a while
5291 */
482c0834 5292 md_wakeup_thread(mddev->thread);
6d036f7d 5293 raid5_release_stripe(sh);
1da177e4 5294 schedule();
27c0f68f 5295 do_prepare = true;
1da177e4
LT		 5296 goto retry;
5297 }
6ed3003c
N		 5298 set_bit(STRIPE_HANDLE, &sh->state);
5299 clear_bit(STRIPE_DELAYED, &sh->state);
59fc630b 5300 if ((!sh->batch_head || sh == sh->batch_head) &&
5301 (bi->bi_rw & REQ_SYNC) &&
729a1866
N		 5302 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
5303 atomic_inc(&conf->preread_active_stripes);
8811b596 5304 release_stripe_plug(mddev, sh);
1da177e4
LT		 5305 } else {
5306 /* cannot get stripe for read-ahead, just give-up */
4246a0b6 5307 bi->bi_error = -EIO;
1da177e4
LT		 5308 break;
5309 }
1da177e4 5310 }
27c0f68f 5311 finish_wait(&conf->wait_for_overlap, &w);
7c13edc8 5312
e7836bd6 5313 remaining = raid5_dec_bi_active_stripes(bi);
f6344757 5314 if (remaining == 0) {
1da177e4 5315
16a53ecc 5316 if ( rw == WRITE )
1da177e4 5317 md_write_end(mddev);
6712ecf8 5318
0a82a8d1
LT		 5319 trace_block_bio_complete(bdev_get_queue(bi->bi_bdev),
5320 bi, 0);
4246a0b6 5321 bio_endio(bi);
1da177e4 5322 }
1da177e4
LT		 5323}
5324
fd01b88c 5325static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks);
b522adcd 5326
fd01b88c 5327static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped)
1da177e4 5328{
52c03291
N		 5329 /* reshaping is quite different to recovery/resync so it is
5330 * handled quite separately ... here.
5331 *
5332 * On each call to sync_request, we gather one chunk worth of
5333 * destination stripes and flag them as expanding.
5334 * Then we find all the source stripes and request reads.
5335 * As the reads complete, handle_stripe will copy the data
5336 * into the destination stripe and release that stripe.
5337 */
d1688a6d 5338 struct r5conf *conf = mddev->private;
1da177e4 5339 struct stripe_head *sh;
ccfcc3c1 5340 sector_t first_sector, last_sector;
f416885e
N		 5341 int raid_disks = conf->previous_raid_disks;
5342 int data_disks = raid_disks - conf->max_degraded;
5343 int new_data_disks = conf->raid_disks - conf->max_degraded;
52c03291
N		 5344 int i;
5345 int dd_idx;
c8f517c4 5346 sector_t writepos, readpos, safepos;
ec32a2bd 5347 sector_t stripe_addr;
7a661381 5348 int reshape_sectors;
ab69ae12 5349 struct list_head stripes;
92140480 5350 sector_t retn;
52c03291 5351
fef9c61f
N		 5352 if (sector_nr == 0) {
5353 /* If restarting in the middle, skip the initial sectors */
2c810cdd 5354 if (mddev->reshape_backwards &&
fef9c61f
N		 5355 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
5356 sector_nr = raid5_size(mddev, 0, 0)
5357 - conf->reshape_progress;
6cbd8148
N		 5358 } else if (mddev->reshape_backwards &&
5359 conf->reshape_progress == MaxSector) {
5360 /* shouldn't happen, but just in case, finish up.*/
5361 sector_nr = MaxSector;
2c810cdd 5362 } else if (!mddev->reshape_backwards &&
fef9c61f
N		 5363 conf->reshape_progress > 0)
5364 sector_nr = conf->reshape_progress;
f416885e 5365 sector_div(sector_nr, new_data_disks);
fef9c61f 5366 if (sector_nr) {
8dee7211
N		 5367 mddev->curr_resync_completed = sector_nr;
5368 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
fef9c61f 5369 *skipped = 1;
92140480
N		 5370 retn = sector_nr;
5371 goto finish;
fef9c61f 5372 }
52c03291
N		 5373 }
5374
7a661381
N		 5375 /* We need to process a full chunk at a time.
5376 * If old and new chunk sizes differ, we need to process the
5377 * largest of these
5378 */
3cb5edf4
N		 5379
5380 reshape_sectors = max(conf->chunk_sectors, conf->prev_chunk_sectors);
7a661381 5381
b5254dd5
N		 5382 /* We update the metadata at least every 10 seconds, or when
5383 * the data about to be copied would over-write the source of
5384 * the data at the front of the range. i.e. one new_stripe
5385 * along from reshape_progress new_maps to after where
5386 * reshape_safe old_maps to
52c03291 5387 */
fef9c61f 5388 writepos = conf->reshape_progress;
f416885e 5389 sector_div(writepos, new_data_disks);
c8f517c4
N		 5390 readpos = conf->reshape_progress;
5391 sector_div(readpos, data_disks);
fef9c61f 5392 safepos = conf->reshape_safe;
f416885e 5393 sector_div(safepos, data_disks);
2c810cdd 5394 if (mddev->reshape_backwards) {
c74c0d76
N		 5395 BUG_ON(writepos < reshape_sectors);
5396 writepos -= reshape_sectors;
c8f517c4 5397 readpos += reshape_sectors;
7a661381 5398 safepos += reshape_sectors;
fef9c61f 5399 } else {
7a661381 5400 writepos += reshape_sectors;
c74c0d76
N		 5401 /* readpos and safepos are worst-case calculations.
5402 * A negative number is overly pessimistic, and causes
5403 * obvious problems for unsigned storage. So clip to 0.
5404 */
ed37d83e
N		 5405 readpos -= min_t(sector_t, reshape_sectors, readpos);
5406 safepos -= min_t(sector_t, reshape_sectors, safepos);
fef9c61f 5407 }
52c03291 5408
b5254dd5
N		 5409 /* Having calculated the 'writepos' possibly use it
5410 * to set 'stripe_addr' which is where we will write to.
5411 */
5412 if (mddev->reshape_backwards) {
5413 BUG_ON(conf->reshape_progress == 0);
5414 stripe_addr = writepos;
5415 BUG_ON((mddev->dev_sectors &
5416 ~((sector_t)reshape_sectors - 1))
5417 - reshape_sectors - stripe_addr
5418 != sector_nr);
5419 } else {
5420 BUG_ON(writepos != sector_nr + reshape_sectors);
5421 stripe_addr = sector_nr;
5422 }
5423
c8f517c4
N		 5424 /* 'writepos' is the most advanced device address we might write.
5425 * 'readpos' is the least advanced device address we might read.
5426 * 'safepos' is the least address recorded in the metadata as having
5427 * been reshaped.
b5254dd5
N		 5428 * If there is a min_offset_diff, these are adjusted either by
5429 * increasing the safepos/readpos if diff is negative, or
5430 * increasing writepos if diff is positive.
5431 * If 'readpos' is then behind 'writepos', there is no way that we can
c8f517c4
N		 5432 * ensure safety in the face of a crash - that must be done by userspace
5433 * making a backup of the data. So in that case there is no particular
5434 * rush to update metadata.
5435 * Otherwise if 'safepos' is behind 'writepos', then we really need to
5436 * update the metadata to advance 'safepos' to match 'readpos' so that
5437 * we can be safe in the event of a crash.
5438 * So we insist on updating metadata if safepos is behind writepos and
5439 * readpos is beyond writepos.
5440 * In any case, update the metadata every 10 seconds.
5441 * Maybe that number should be configurable, but I'm not sure it is
5442 * worth it.... maybe it could be a multiple of safemode_delay???
5443 */
b5254dd5
N		 5444 if (conf->min_offset_diff < 0) {
5445 safepos += -conf->min_offset_diff;
5446 readpos += -conf->min_offset_diff;
5447 } else
5448 writepos += conf->min_offset_diff;
5449
2c810cdd 5450 if ((mddev->reshape_backwards
c8f517c4
N		 5451 ? (safepos > writepos && readpos < writepos)
5452 : (safepos < writepos && readpos > writepos)) ||
5453 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
52c03291
N		 5454 /* Cannot proceed until we've updated the superblock... */
5455 wait_event(conf->wait_for_overlap,
c91abf5a
N		 5456 atomic_read(&conf->reshape_stripes)==0
5457 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5458 if (atomic_read(&conf->reshape_stripes) != 0)
5459 return 0;
fef9c61f 5460 mddev->reshape_position = conf->reshape_progress;
75d3da43 5461 mddev->curr_resync_completed = sector_nr;
c8f517c4 5462 conf->reshape_checkpoint = jiffies;
850b2b42 5463 set_bit(MD_CHANGE_DEVS, &mddev->flags);
52c03291 5464 md_wakeup_thread(mddev->thread);
850b2b42 5465 wait_event(mddev->sb_wait, mddev->flags == 0 ||
c91abf5a
N		 5466 test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5467 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5468 return 0;
52c03291 5469 spin_lock_irq(&conf->device_lock);
fef9c61f 5470 conf->reshape_safe = mddev->reshape_position;
52c03291
N		 5471 spin_unlock_irq(&conf->device_lock);
5472 wake_up(&conf->wait_for_overlap);
acb180b0 5473 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
52c03291
N		 5474 }
5475
ab69ae12 5476 INIT_LIST_HEAD(&stripes);
7a661381 5477 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
52c03291 5478 int j;
a9f326eb 5479 int skipped_disk = 0;
6d036f7d 5480 sh = raid5_get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
52c03291
N		 5481 set_bit(STRIPE_EXPANDING, &sh->state);
5482 atomic_inc(&conf->reshape_stripes);
5483 /* If any of this stripe is beyond the end of the old
5484 * array, then we need to zero those blocks
5485 */
5486 for (j=sh->disks; j--;) {
5487 sector_t s;
5488 if (j == sh->pd_idx)
5489 continue;
f416885e 5490 if (conf->level == 6 &&
d0dabf7e 5491 j == sh->qd_idx)
f416885e 5492 continue;
6d036f7d 5493 s = raid5_compute_blocknr(sh, j, 0);
b522adcd 5494 if (s < raid5_size(mddev, 0, 0)) {
a9f326eb 5495 skipped_disk = 1;
52c03291
N		 5496 continue;
5497 }
5498 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
5499 set_bit(R5_Expanded, &sh->dev[j].flags);
5500 set_bit(R5_UPTODATE, &sh->dev[j].flags);
5501 }
a9f326eb 5502 if (!skipped_disk) {
52c03291
N		 5503 set_bit(STRIPE_EXPAND_READY, &sh->state);
5504 set_bit(STRIPE_HANDLE, &sh->state);
5505 }
ab69ae12 5506 list_add(&sh->lru, &stripes);
52c03291
N		 5507 }
5508 spin_lock_irq(&conf->device_lock);
2c810cdd 5509 if (mddev->reshape_backwards)
7a661381 5510 conf->reshape_progress -= reshape_sectors * new_data_disks;
fef9c61f 5511 else
7a661381 5512 conf->reshape_progress += reshape_sectors * new_data_disks;
52c03291
N		 5513 spin_unlock_irq(&conf->device_lock);
5514 /* Ok, those stripe are ready. We can start scheduling
5515 * reads on the source stripes.
5516 * The source stripes are determined by mapping the first and last
5517 * block on the destination stripes.
5518 */
52c03291 5519 first_sector =
ec32a2bd 5520 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
911d4ee8 5521 1, &dd_idx, NULL);
52c03291 5522 last_sector =
0e6e0271 5523 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
09c9e5fa 5524 * new_data_disks - 1),
911d4ee8 5525 1, &dd_idx, NULL);
58c0fed4
AN		 5526 if (last_sector >= mddev->dev_sectors)
5527 last_sector = mddev->dev_sectors - 1;
52c03291 5528 while (first_sector <= last_sector) {
6d036f7d 5529 sh = raid5_get_active_stripe(conf, first_sector, 1, 0, 1);
52c03291
N		 5530 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
5531 set_bit(STRIPE_HANDLE, &sh->state);
6d036f7d 5532 raid5_release_stripe(sh);
52c03291
N		 5533 first_sector += STRIPE_SECTORS;
5534 }
ab69ae12
N		 5535 /* Now that the sources are clearly marked, we can release
5536 * the destination stripes
5537 */
5538 while (!list_empty(&stripes)) {
5539 sh = list_entry(stripes.next, struct stripe_head, lru);
5540 list_del_init(&sh->lru);
6d036f7d 5541 raid5_release_stripe(sh);
ab69ae12 5542 }
c6207277
N		 5543 /* If this takes us to the resync_max point where we have to pause,
5544 * then we need to write out the superblock.
5545 */
7a661381 5546 sector_nr += reshape_sectors;
92140480
N		 5547 retn = reshape_sectors;
5548finish:
c5e19d90
N		 5549 if (mddev->curr_resync_completed > mddev->resync_max ||
5550 (sector_nr - mddev->curr_resync_completed) * 2
c03f6a19 5551 >= mddev->resync_max - mddev->curr_resync_completed) {
c6207277
N		 5552 /* Cannot proceed until we've updated the superblock... */
5553 wait_event(conf->wait_for_overlap,
c91abf5a
N		 5554 atomic_read(&conf->reshape_stripes) == 0
5555 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5556 if (atomic_read(&conf->reshape_stripes) != 0)
5557 goto ret;
fef9c61f 5558 mddev->reshape_position = conf->reshape_progress;
75d3da43 5559 mddev->curr_resync_completed = sector_nr;
c8f517c4 5560 conf->reshape_checkpoint = jiffies;
c6207277
N		 5561 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5562 md_wakeup_thread(mddev->thread);
5563 wait_event(mddev->sb_wait,
5564 !test_bit(MD_CHANGE_DEVS, &mddev->flags)
c91abf5a
N		 5565 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5566 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5567 goto ret;
c6207277 5568 spin_lock_irq(&conf->device_lock);
fef9c61f 5569 conf->reshape_safe = mddev->reshape_position;
c6207277
N		 5570 spin_unlock_irq(&conf->device_lock);
5571 wake_up(&conf->wait_for_overlap);
acb180b0 5572 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
c6207277 5573 }
c91abf5a 5574ret:
92140480 5575 return retn;
52c03291
N		 5576}
5577
09314799 5578static inline sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped)
52c03291 5579{
d1688a6d 5580 struct r5conf *conf = mddev->private;
52c03291 5581 struct stripe_head *sh;
58c0fed4 5582 sector_t max_sector = mddev->dev_sectors;
57dab0bd 5583 sector_t sync_blocks;
16a53ecc
N		 5584 int still_degraded = 0;
5585 int i;
1da177e4 5586
72626685 5587 if (sector_nr >= max_sector) {
1da177e4 5588 /* just being told to finish up .. nothing much to do */
cea9c228 5589
29269553
N		 5590 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
5591 end_reshape(conf);
5592 return 0;
5593 }
72626685
N		 5594
5595 if (mddev->curr_resync < max_sector) /* aborted */
5596 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
5597 &sync_blocks, 1);
16a53ecc 5598 else /* completed sync */
72626685
N		 5599 conf->fullsync = 0;
5600 bitmap_close_sync(mddev->bitmap);
5601
1da177e4
LT		 5602 return 0;
5603 }
ccfcc3c1 5604
64bd660b
N		 5605 /* Allow raid5_quiesce to complete */
5606 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
5607
52c03291
N		 5608 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5609 return reshape_request(mddev, sector_nr, skipped);
f6705578 5610
c6207277
N		 5611 /* No need to check resync_max as we never do more than one
5612 * stripe, and as resync_max will always be on a chunk boundary,
5613 * if the check in md_do_sync didn't fire, there is no chance
5614 * of overstepping resync_max here
5615 */
5616
16a53ecc 5617 /* if there is too many failed drives and we are trying
1da177e4
LT		 5618 * to resync, then assert that we are finished, because there is
5619 * nothing we can do.
5620 */
3285edf1 5621 if (mddev->degraded >= conf->max_degraded &&
16a53ecc 5622 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
58c0fed4 5623 sector_t rv = mddev->dev_sectors - sector_nr;
57afd89f 5624 *skipped = 1;
1da177e4
LT		 5625 return rv;
5626 }
6f608040 5627 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
5628 !conf->fullsync &&
5629 !bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
5630 sync_blocks >= STRIPE_SECTORS) {
72626685
N		 5631 /* we can skip this block, and probably more */
5632 sync_blocks /= STRIPE_SECTORS;
5633 *skipped = 1;
5634 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
5635 }
1da177e4 5636
c40f341f 5637 bitmap_cond_end_sync(mddev->bitmap, sector_nr, false);
b47490c9 5638
6d036f7d 5639 sh = raid5_get_active_stripe(conf, sector_nr, 0, 1, 0);
1da177e4 5640 if (sh == NULL) {
6d036f7d 5641 sh = raid5_get_active_stripe(conf, sector_nr, 0, 0, 0);
1da177e4 5642 /* make sure we don't swamp the stripe cache if someone else
16a53ecc 5643 * is trying to get access
1da177e4 5644 */
66c006a5 5645 schedule_timeout_uninterruptible(1);
1da177e4 5646 }
16a53ecc 5647 /* Need to check if array will still be degraded after recovery/resync
16d9cfab
EM		 5648 * Note in case of > 1 drive failures it's possible we're rebuilding
5649 * one drive while leaving another faulty drive in array.
16a53ecc 5650 */
16d9cfab
EM		 5651 rcu_read_lock();
5652 for (i = 0; i < conf->raid_disks; i++) {
5653 struct md_rdev *rdev = ACCESS_ONCE(conf->disks[i].rdev);
5654
5655 if (rdev == NULL || test_bit(Faulty, &rdev->flags))
16a53ecc 5656 still_degraded = 1;
16d9cfab
EM		 5657 }
5658 rcu_read_unlock();
16a53ecc
N		 5659
5660 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
5661
83206d66 5662 set_bit(STRIPE_SYNC_REQUESTED, &sh->state);
053f5b65 5663 set_bit(STRIPE_HANDLE, &sh->state);
1da177e4 5664
6d036f7d 5665 raid5_release_stripe(sh);
1da177e4
LT		 5666
5667 return STRIPE_SECTORS;
5668}
5669
d1688a6d 5670static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio)
46031f9a
RBJ		 5671{
5672 /* We may not be able to submit a whole bio at once as there
5673 * may not be enough stripe_heads available.
5674 * We cannot pre-allocate enough stripe_heads as we may need
5675 * more than exist in the cache (if we allow ever large chunks).
5676 * So we do one stripe head at a time and record in
5677 * ->bi_hw_segments how many have been done.
5678 *
5679 * We *know* that this entire raid_bio is in one chunk, so
5680 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
5681 */
5682 struct stripe_head *sh;
911d4ee8 5683 int dd_idx;
46031f9a
RBJ		 5684 sector_t sector, logical_sector, last_sector;
5685 int scnt = 0;
5686 int remaining;
5687 int handled = 0;
5688
4f024f37
KO		 5689 logical_sector = raid_bio->bi_iter.bi_sector &
5690 ~((sector_t)STRIPE_SECTORS-1);
112bf897 5691 sector = raid5_compute_sector(conf, logical_sector,
911d4ee8 5692 0, &dd_idx, NULL);
f73a1c7d 5693 last_sector = bio_end_sector(raid_bio);
46031f9a
RBJ		 5694
5695 for (; logical_sector < last_sector;
387bb173
NB		 5696 logical_sector += STRIPE_SECTORS,
5697 sector += STRIPE_SECTORS,
5698 scnt++) {
46031f9a 5699
e7836bd6 5700 if (scnt < raid5_bi_processed_stripes(raid_bio))
46031f9a
RBJ		 5701 /* already done this stripe */
5702 continue;
5703
6d036f7d 5704 sh = raid5_get_active_stripe(conf, sector, 0, 1, 1);
46031f9a
RBJ		 5705
5706 if (!sh) {
5707 /* failed to get a stripe - must wait */
e7836bd6 5708 raid5_set_bi_processed_stripes(raid_bio, scnt);
46031f9a
RBJ		 5709 conf->retry_read_aligned = raid_bio;
5710 return handled;
5711 }
5712
da41ba65 5713 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) {
6d036f7d 5714 raid5_release_stripe(sh);
e7836bd6 5715 raid5_set_bi_processed_stripes(raid_bio, scnt);
387bb173
NB		 5716 conf->retry_read_aligned = raid_bio;
5717 return handled;
5718 }
5719
3f9e7c14 5720 set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags);
36d1c647 5721 handle_stripe(sh);
6d036f7d 5722 raid5_release_stripe(sh);
46031f9a
RBJ		 5723 handled++;
5724 }
e7836bd6 5725 remaining = raid5_dec_bi_active_stripes(raid_bio);
0a82a8d1
LT		 5726 if (remaining == 0) {
5727 trace_block_bio_complete(bdev_get_queue(raid_bio->bi_bdev),
5728 raid_bio, 0);
4246a0b6 5729 bio_endio(raid_bio);
0a82a8d1 5730 }
46031f9a 5731 if (atomic_dec_and_test(&conf->active_aligned_reads))
b1b46486 5732 wake_up(&conf->wait_for_quiescent);
46031f9a
RBJ		 5733 return handled;
5734}
5735
bfc90cb0 5736static int handle_active_stripes(struct r5conf *conf, int group,
566c09c5
SL		 5737 struct r5worker *worker,
5738 struct list_head *temp_inactive_list)
46a06401
SL		 5739{
5740 struct stripe_head *batch[MAX_STRIPE_BATCH], *sh;
566c09c5
SL		 5741 int i, batch_size = 0, hash;
5742 bool release_inactive = false;
46a06401
SL		 5743
5744 while (batch_size < MAX_STRIPE_BATCH &&
851c30c9 5745 (sh = __get_priority_stripe(conf, group)) != NULL)
46a06401
SL		 5746 batch[batch_size++] = sh;
5747
566c09c5
SL		 5748 if (batch_size == 0) {
5749 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5750 if (!list_empty(temp_inactive_list + i))
5751 break;
a8c34f91
SL		 5752 if (i == NR_STRIPE_HASH_LOCKS) {
5753 spin_unlock_irq(&conf->device_lock);
5754 r5l_flush_stripe_to_raid(conf->log);
5755 spin_lock_irq(&conf->device_lock);
566c09c5 5756 return batch_size;
a8c34f91 5757 }
566c09c5
SL		 5758 release_inactive = true;
5759 }
46a06401
SL		 5760 spin_unlock_irq(&conf->device_lock);
5761
566c09c5
SL		 5762 release_inactive_stripe_list(conf, temp_inactive_list,
5763 NR_STRIPE_HASH_LOCKS);
5764
a8c34f91 5765 r5l_flush_stripe_to_raid(conf->log);
566c09c5
SL		 5766 if (release_inactive) {
5767 spin_lock_irq(&conf->device_lock);
5768 return 0;
5769 }
5770
46a06401
SL		 5771 for (i = 0; i < batch_size; i++)
5772 handle_stripe(batch[i]);
f6bed0ef 5773 r5l_write_stripe_run(conf->log);
46a06401
SL		 5774
5775 cond_resched();
5776
5777 spin_lock_irq(&conf->device_lock);
566c09c5
SL		 5778 for (i = 0; i < batch_size; i++) {
5779 hash = batch[i]->hash_lock_index;
5780 __release_stripe(conf, batch[i], &temp_inactive_list[hash]);
5781 }
46a06401
SL		 5782 return batch_size;
5783}
46031f9a 5784
851c30c9
SL		 5785static void raid5_do_work(struct work_struct *work)
5786{
5787 struct r5worker *worker = container_of(work, struct r5worker, work);
5788 struct r5worker_group *group = worker->group;
5789 struct r5conf *conf = group->conf;
5790 int group_id = group - conf->worker_groups;
5791 int handled;
5792 struct blk_plug plug;
5793
5794 pr_debug("+++ raid5worker active\n");
5795
5796 blk_start_plug(&plug);
5797 handled = 0;
5798 spin_lock_irq(&conf->device_lock);
5799 while (1) {
5800 int batch_size, released;
5801
566c09c5 5802 released = release_stripe_list(conf, worker->temp_inactive_list);
851c30c9 5803
566c09c5
SL		 5804 batch_size = handle_active_stripes(conf, group_id, worker,
5805 worker->temp_inactive_list);
bfc90cb0 5806 worker->working = false;
851c30c9
SL		 5807 if (!batch_size && !released)
5808 break;
5809 handled += batch_size;
5810 }
5811 pr_debug("%d stripes handled\n", handled);
5812
5813 spin_unlock_irq(&conf->device_lock);
5814 blk_finish_plug(&plug);
5815
5816 pr_debug("--- raid5worker inactive\n");
5817}
5818
1da177e4
LT		 5819/*
5820 * This is our raid5 kernel thread.
5821 *
5822 * We scan the hash table for stripes which can be handled now.
5823 * During the scan, completed stripes are saved for us by the interrupt
5824 * handler, so that they will not have to wait for our next wakeup.
5825 */
4ed8731d 5826static void raid5d(struct md_thread *thread)
1da177e4 5827{
4ed8731d 5828 struct mddev *mddev = thread->mddev;
d1688a6d 5829 struct r5conf *conf = mddev->private;
1da177e4 5830 int handled;
e1dfa0a2 5831 struct blk_plug plug;
1da177e4 5832
45b4233c 5833 pr_debug("+++ raid5d active\n");
1da177e4
LT		 5834
5835 md_check_recovery(mddev);
1da177e4 5836
c3cce6cd
N		 5837 if (!bio_list_empty(&conf->return_bi) &&
5838 !test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
5839 struct bio_list tmp = BIO_EMPTY_LIST;
5840 spin_lock_irq(&conf->device_lock);
5841 if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
5842 bio_list_merge(&tmp, &conf->return_bi);
5843 bio_list_init(&conf->return_bi);
5844 }
5845 spin_unlock_irq(&conf->device_lock);
5846 return_io(&tmp);
5847 }
5848
e1dfa0a2 5849 blk_start_plug(&plug);
1da177e4
LT		 5850 handled = 0;
5851 spin_lock_irq(&conf->device_lock);
5852 while (1) {
46031f9a 5853 struct bio *bio;
773ca82f
SL		 5854 int batch_size, released;
5855
566c09c5 5856 released = release_stripe_list(conf, conf->temp_inactive_list);
edbe83ab
N		 5857 if (released)
5858 clear_bit(R5_DID_ALLOC, &conf->cache_state);
1da177e4 5859
0021b7bc 5860 if (
7c13edc8
N		 5861 !list_empty(&conf->bitmap_list)) {
5862 /* Now is a good time to flush some bitmap updates */
5863 conf->seq_flush++;
700e432d 5864 spin_unlock_irq(&conf->device_lock);
72626685 5865 bitmap_unplug(mddev->bitmap);
700e432d 5866 spin_lock_irq(&conf->device_lock);
7c13edc8 5867 conf->seq_write = conf->seq_flush;
566c09c5 5868 activate_bit_delay(conf, conf->temp_inactive_list);
72626685 5869 }
0021b7bc 5870 raid5_activate_delayed(conf);
72626685 5871
46031f9a
RBJ		 5872 while ((bio = remove_bio_from_retry(conf))) {
5873 int ok;
5874 spin_unlock_irq(&conf->device_lock);
5875 ok = retry_aligned_read(conf, bio);
5876 spin_lock_irq(&conf->device_lock);
5877 if (!ok)
5878 break;
5879 handled++;
5880 }
5881
566c09c5
SL		 5882 batch_size = handle_active_stripes(conf, ANY_GROUP, NULL,
5883 conf->temp_inactive_list);
773ca82f 5884 if (!batch_size && !released)
1da177e4 5885 break;
46a06401 5886 handled += batch_size;
1da177e4 5887
46a06401
SL		 5888 if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) {
5889 spin_unlock_irq(&conf->device_lock);
de393cde 5890 md_check_recovery(mddev);
46a06401
SL		 5891 spin_lock_irq(&conf->device_lock);
5892 }
1da177e4 5893 }
45b4233c 5894 pr_debug("%d stripes handled\n", handled);
1da177e4
LT		 5895
5896 spin_unlock_irq(&conf->device_lock);
2d5b569b
N		 5897 if (test_and_clear_bit(R5_ALLOC_MORE, &conf->cache_state) &&
5898 mutex_trylock(&conf->cache_size_mutex)) {
edbe83ab
N		 5899 grow_one_stripe(conf, __GFP_NOWARN);
5900 /* Set flag even if allocation failed. This helps
5901 * slow down allocation requests when mem is short
5902 */
5903 set_bit(R5_DID_ALLOC, &conf->cache_state);
2d5b569b 5904 mutex_unlock(&conf->cache_size_mutex);
edbe83ab 5905 }
1da177e4 5906
0576b1c6
SL		 5907 r5l_flush_stripe_to_raid(conf->log);
5908
c9f21aaf 5909 async_tx_issue_pending_all();
e1dfa0a2 5910 blk_finish_plug(&plug);
1da177e4 5911
45b4233c 5912 pr_debug("--- raid5d inactive\n");
1da177e4
LT		 5913}
5914
3f294f4f 5915static ssize_t
fd01b88c 5916raid5_show_stripe_cache_size(struct mddev *mddev, char *page)
3f294f4f 5917{
7b1485ba
N		 5918 struct r5conf *conf;
5919 int ret = 0;
5920 spin_lock(&mddev->lock);
5921 conf = mddev->private;
96de1e66 5922 if (conf)
edbe83ab 5923 ret = sprintf(page, "%d\n", conf->min_nr_stripes);
7b1485ba
N		 5924 spin_unlock(&mddev->lock);
5925 return ret;
3f294f4f
N		 5926}
5927
c41d4ac4 5928int
fd01b88c 5929raid5_set_cache_size(struct mddev *mddev, int size)
3f294f4f 5930{
d1688a6d 5931 struct r5conf *conf = mddev->private;
b5470dc5
DW		 5932 int err;
5933
c41d4ac4 5934 if (size <= 16 || size > 32768)
3f294f4f 5935 return -EINVAL;
486f0644 5936
edbe83ab 5937 conf->min_nr_stripes = size;
2d5b569b 5938 mutex_lock(&conf->cache_size_mutex);
486f0644
N		 5939 while (size < conf->max_nr_stripes &&
5940 drop_one_stripe(conf))
5941 ;
2d5b569b 5942 mutex_unlock(&conf->cache_size_mutex);
486f0644 5943
edbe83ab 5944
b5470dc5
DW		 5945 err = md_allow_write(mddev);
5946 if (err)
5947 return err;
486f0644 5948
2d5b569b 5949 mutex_lock(&conf->cache_size_mutex);
486f0644
N		 5950 while (size > conf->max_nr_stripes)
5951 if (!grow_one_stripe(conf, GFP_KERNEL))
5952 break;
2d5b569b 5953 mutex_unlock(&conf->cache_size_mutex);
486f0644 5954
c41d4ac4
N		 5955 return 0;
5956}
5957EXPORT_SYMBOL(raid5_set_cache_size);
5958
5959static ssize_t
fd01b88c 5960raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len)
c41d4ac4 5961{
6791875e 5962 struct r5conf *conf;
c41d4ac4
N		 5963 unsigned long new;
5964 int err;
5965
5966 if (len >= PAGE_SIZE)
5967 return -EINVAL;
b29bebd6 5968 if (kstrtoul(page, 10, &new))
c41d4ac4 5969 return -EINVAL;
6791875e 5970 err = mddev_lock(mddev);
c41d4ac4
N		 5971 if (err)
5972 return err;
6791875e
N		 5973 conf = mddev->private;
5974 if (!conf)
5975 err = -ENODEV;
5976 else
5977 err = raid5_set_cache_size(mddev, new);
5978 mddev_unlock(mddev);
5979
5980 return err ?: len;
3f294f4f 5981}
007583c9 5982
96de1e66
N		 5983static struct md_sysfs_entry
5984raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
5985 raid5_show_stripe_cache_size,
5986 raid5_store_stripe_cache_size);
3f294f4f 5987
d06f191f
MS		 5988static ssize_t
5989raid5_show_rmw_level(struct mddev *mddev, char *page)
5990{
5991 struct r5conf *conf = mddev->private;
5992 if (conf)
5993 return sprintf(page, "%d\n", conf->rmw_level);
5994 else
5995 return 0;
5996}
5997
5998static ssize_t
5999raid5_store_rmw_level(struct mddev *mddev, const char *page, size_t len)
6000{
6001 struct r5conf *conf = mddev->private;
6002 unsigned long new;
6003
6004 if (!conf)
6005 return -ENODEV;
6006
6007 if (len >= PAGE_SIZE)
6008 return -EINVAL;
6009
6010 if (kstrtoul(page, 10, &new))
6011 return -EINVAL;
6012
6013 if (new != PARITY_DISABLE_RMW && !raid6_call.xor_syndrome)
6014 return -EINVAL;
6015
6016 if (new != PARITY_DISABLE_RMW &&
6017 new != PARITY_ENABLE_RMW &&
6018 new != PARITY_PREFER_RMW)
6019 return -EINVAL;
6020
6021 conf->rmw_level = new;
6022 return len;
6023}
6024
6025static struct md_sysfs_entry
6026raid5_rmw_level = __ATTR(rmw_level, S_IRUGO | S_IWUSR,
6027 raid5_show_rmw_level,
6028 raid5_store_rmw_level);
6029
6030
8b3e6cdc 6031static ssize_t
fd01b88c 6032raid5_show_preread_threshold(struct mddev *mddev, char *page)
8b3e6cdc 6033{
7b1485ba
N		 6034 struct r5conf *conf;
6035 int ret = 0;
6036 spin_lock(&mddev->lock);
6037 conf = mddev->private;
8b3e6cdc 6038 if (conf)
7b1485ba
N		 6039 ret = sprintf(page, "%d\n", conf->bypass_threshold);
6040 spin_unlock(&mddev->lock);
6041 return ret;
8b3e6cdc
DW		 6042}
6043
6044static ssize_t
fd01b88c 6045raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len)
8b3e6cdc 6046{
6791875e 6047 struct r5conf *conf;
4ef197d8 6048 unsigned long new;
6791875e
N		 6049 int err;
6050
8b3e6cdc
DW		 6051 if (len >= PAGE_SIZE)
6052 return -EINVAL;
b29bebd6 6053 if (kstrtoul(page, 10, &new))
8b3e6cdc 6054 return -EINVAL;
6791875e
N		 6055
6056 err = mddev_lock(mddev);
6057 if (err)
6058 return err;
6059 conf = mddev->private;
6060 if (!conf)
6061 err = -ENODEV;
edbe83ab 6062 else if (new > conf->min_nr_stripes)
6791875e
N		 6063 err = -EINVAL;
6064 else
6065 conf->bypass_threshold = new;
6066 mddev_unlock(mddev);
6067 return err ?: len;
8b3e6cdc
DW		 6068}
6069
6070static struct md_sysfs_entry
6071raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
6072 S_IRUGO | S_IWUSR,
6073 raid5_show_preread_threshold,
6074 raid5_store_preread_threshold);
6075
d592a996
SL		 6076static ssize_t
6077raid5_show_skip_copy(struct mddev *mddev, char *page)
6078{
7b1485ba
N		 6079 struct r5conf *conf;
6080 int ret = 0;
6081 spin_lock(&mddev->lock);
6082 conf = mddev->private;
d592a996 6083 if (conf)
7b1485ba
N		 6084 ret = sprintf(page, "%d\n", conf->skip_copy);
6085 spin_unlock(&mddev->lock);
6086 return ret;
d592a996
SL		 6087}
6088
6089static ssize_t
6090raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len)
6091{
6791875e 6092 struct r5conf *conf;
d592a996 6093 unsigned long new;
6791875e
N		 6094 int err;
6095
d592a996
SL		 6096 if (len >= PAGE_SIZE)
6097 return -EINVAL;
d592a996
SL		 6098 if (kstrtoul(page, 10, &new))
6099 return -EINVAL;
6100 new = !!new;
6791875e
N		 6101
6102 err = mddev_lock(mddev);
6103 if (err)
6104 return err;
6105 conf = mddev->private;
6106 if (!conf)
6107 err = -ENODEV;
6108 else if (new != conf->skip_copy) {
6109 mddev_suspend(mddev);
6110 conf->skip_copy = new;
6111 if (new)
6112 mddev->queue->backing_dev_info.capabilities |=
6113 BDI_CAP_STABLE_WRITES;
6114 else
6115 mddev->queue->backing_dev_info.capabilities &=
6116 ~BDI_CAP_STABLE_WRITES;
6117 mddev_resume(mddev);
6118 }
6119 mddev_unlock(mddev);
6120 return err ?: len;
d592a996
SL		 6121}
6122
6123static struct md_sysfs_entry
6124raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR,
6125 raid5_show_skip_copy,
6126 raid5_store_skip_copy);
6127
3f294f4f 6128static ssize_t
fd01b88c 6129stripe_cache_active_show(struct mddev *mddev, char *page)
3f294f4f 6130{
d1688a6d 6131 struct r5conf *conf = mddev->private;
96de1e66
N		 6132 if (conf)
6133 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
6134 else
6135 return 0;
3f294f4f
N		 6136}
6137
96de1e66
N		 6138static struct md_sysfs_entry
6139raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
3f294f4f 6140
b721420e
SL		 6141static ssize_t
6142raid5_show_group_thread_cnt(struct mddev *mddev, char *page)
6143{
7b1485ba
N		 6144 struct r5conf *conf;
6145 int ret = 0;
6146 spin_lock(&mddev->lock);
6147 conf = mddev->private;
b721420e 6148 if (conf)
7b1485ba
N		 6149 ret = sprintf(page, "%d\n", conf->worker_cnt_per_group);
6150 spin_unlock(&mddev->lock);
6151 return ret;
b721420e
SL		 6152}
6153
60aaf933 6154static int alloc_thread_groups(struct r5conf *conf, int cnt,
6155 int *group_cnt,
6156 int *worker_cnt_per_group,
6157 struct r5worker_group **worker_groups);
b721420e
SL		 6158static ssize_t
6159raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len)
6160{
6791875e 6161 struct r5conf *conf;
b721420e
SL		 6162 unsigned long new;
6163 int err;
60aaf933 6164 struct r5worker_group *new_groups, *old_groups;
6165 int group_cnt, worker_cnt_per_group;
b721420e
SL		 6166
6167 if (len >= PAGE_SIZE)
6168 return -EINVAL;
b721420e
SL		 6169 if (kstrtoul(page, 10, &new))
6170 return -EINVAL;
6171
6791875e
N		 6172 err = mddev_lock(mddev);
6173 if (err)
6174 return err;
6175 conf = mddev->private;
6176 if (!conf)
6177 err = -ENODEV;
6178 else if (new != conf->worker_cnt_per_group) {
6179 mddev_suspend(mddev);
b721420e 6180
6791875e
N		 6181 old_groups = conf->worker_groups;
6182 if (old_groups)
6183 flush_workqueue(raid5_wq);
d206dcfa 6184
6791875e
N		 6185 err = alloc_thread_groups(conf, new,
6186 &group_cnt, &worker_cnt_per_group,
6187 &new_groups);
6188 if (!err) {
6189 spin_lock_irq(&conf->device_lock);
6190 conf->group_cnt = group_cnt;
6191 conf->worker_cnt_per_group = worker_cnt_per_group;
6192 conf->worker_groups = new_groups;
6193 spin_unlock_irq(&conf->device_lock);
b721420e 6194
6791875e
N		 6195 if (old_groups)
6196 kfree(old_groups[0].workers);
6197 kfree(old_groups);
6198 }
6199 mddev_resume(mddev);
b721420e 6200 }
6791875e 6201 mddev_unlock(mddev);
b721420e 6202
6791875e 6203 return err ?: len;
b721420e
SL		 6204}
6205
6206static struct md_sysfs_entry
6207raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR,
6208 raid5_show_group_thread_cnt,
6209 raid5_store_group_thread_cnt);
6210
007583c9 6211static struct attribute *raid5_attrs[] = {
3f294f4f
N		 6212 &raid5_stripecache_size.attr,
6213 &raid5_stripecache_active.attr,
8b3e6cdc 6214 &raid5_preread_bypass_threshold.attr,
b721420e 6215 &raid5_group_thread_cnt.attr,
d592a996 6216 &raid5_skip_copy.attr,
d06f191f 6217 &raid5_rmw_level.attr,
3f294f4f
N		 6218 NULL,
6219};
007583c9
N		 6220static struct attribute_group raid5_attrs_group = {
6221 .name = NULL,
6222 .attrs = raid5_attrs,
3f294f4f
N		 6223};
6224
60aaf933 6225static int alloc_thread_groups(struct r5conf *conf, int cnt,
6226 int *group_cnt,
6227 int *worker_cnt_per_group,
6228 struct r5worker_group **worker_groups)
851c30c9 6229{
566c09c5 6230 int i, j, k;
851c30c9
SL		 6231 ssize_t size;
6232 struct r5worker *workers;
6233
60aaf933 6234 *worker_cnt_per_group = cnt;
851c30c9 6235 if (cnt == 0) {
60aaf933 6236 *group_cnt = 0;
6237 *worker_groups = NULL;
851c30c9
SL		 6238 return 0;
6239 }
60aaf933 6240 *group_cnt = num_possible_nodes();
851c30c9 6241 size = sizeof(struct r5worker) * cnt;
60aaf933 6242 workers = kzalloc(size * *group_cnt, GFP_NOIO);
6243 *worker_groups = kzalloc(sizeof(struct r5worker_group) *
6244 *group_cnt, GFP_NOIO);
6245 if (!*worker_groups || !workers) {
851c30c9 6246 kfree(workers);
60aaf933 6247 kfree(*worker_groups);
851c30c9
SL		 6248 return -ENOMEM;
6249 }
6250
60aaf933 6251 for (i = 0; i < *group_cnt; i++) {
851c30c9
SL		 6252 struct r5worker_group *group;
6253
0c775d52 6254 group = &(*worker_groups)[i];
851c30c9
SL		 6255 INIT_LIST_HEAD(&group->handle_list);
6256 group->conf = conf;
6257 group->workers = workers + i * cnt;
6258
6259 for (j = 0; j < cnt; j++) {
566c09c5
SL		 6260 struct r5worker *worker = group->workers + j;
6261 worker->group = group;
6262 INIT_WORK(&worker->work, raid5_do_work);
6263
6264 for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++)
6265 INIT_LIST_HEAD(worker->temp_inactive_list + k);
851c30c9
SL		 6266 }
6267 }
6268
6269 return 0;
6270}
6271
6272static void free_thread_groups(struct r5conf *conf)
6273{
6274 if (conf->worker_groups)
6275 kfree(conf->worker_groups[0].workers);
6276 kfree(conf->worker_groups);
6277 conf->worker_groups = NULL;
6278}
6279
80c3a6ce 6280static sector_t
fd01b88c 6281raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks)
80c3a6ce 6282{
d1688a6d 6283 struct r5conf *conf = mddev->private;
80c3a6ce
DW		 6284
6285 if (!sectors)
6286 sectors = mddev->dev_sectors;
5e5e3e78 6287 if (!raid_disks)
7ec05478 6288 /* size is defined by the smallest of previous and new size */
5e5e3e78 6289 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
80c3a6ce 6290
3cb5edf4
N		 6291 sectors &= ~((sector_t)conf->chunk_sectors - 1);
6292 sectors &= ~((sector_t)conf->prev_chunk_sectors - 1);
80c3a6ce
DW		 6293 return sectors * (raid_disks - conf->max_degraded);
6294}
6295
789b5e03
ON		 6296static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
6297{
6298 safe_put_page(percpu->spare_page);
46d5b785 6299 if (percpu->scribble)
6300 flex_array_free(percpu->scribble);
789b5e03
ON		 6301 percpu->spare_page = NULL;
6302 percpu->scribble = NULL;
6303}
6304
6305static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
6306{
6307 if (conf->level == 6 && !percpu->spare_page)
6308 percpu->spare_page = alloc_page(GFP_KERNEL);
6309 if (!percpu->scribble)
46d5b785 6310 percpu->scribble = scribble_alloc(max(conf->raid_disks,
738a2738
N		 6311 conf->previous_raid_disks),
6312 max(conf->chunk_sectors,
6313 conf->prev_chunk_sectors)
6314 / STRIPE_SECTORS,
6315 GFP_KERNEL);
789b5e03
ON		 6316
6317 if (!percpu->scribble || (conf->level == 6 && !percpu->spare_page)) {
6318 free_scratch_buffer(conf, percpu);
6319 return -ENOMEM;
6320 }
6321
6322 return 0;
6323}
6324
d1688a6d 6325static void raid5_free_percpu(struct r5conf *conf)
36d1c647 6326{
36d1c647
DW		 6327 unsigned long cpu;
6328
6329 if (!conf->percpu)
6330 return;
6331
36d1c647
DW		 6332#ifdef CONFIG_HOTPLUG_CPU
6333 unregister_cpu_notifier(&conf->cpu_notify);
6334#endif
789b5e03
ON		 6335
6336 get_online_cpus();
6337 for_each_possible_cpu(cpu)
6338 free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
36d1c647
DW		 6339 put_online_cpus();
6340
6341 free_percpu(conf->percpu);
6342}
6343
d1688a6d 6344static void free_conf(struct r5conf *conf)
95fc17aa 6345{
5c7e81c3
SL		 6346 if (conf->log)
6347 r5l_exit_log(conf->log);
edbe83ab
N		 6348 if (conf->shrinker.seeks)
6349 unregister_shrinker(&conf->shrinker);
5c7e81c3 6350
851c30c9 6351 free_thread_groups(conf);
95fc17aa 6352 shrink_stripes(conf);
36d1c647 6353 raid5_free_percpu(conf);
95fc17aa
DW		 6354 kfree(conf->disks);
6355 kfree(conf->stripe_hashtbl);
6356 kfree(conf);
6357}
6358
36d1c647
DW		 6359#ifdef CONFIG_HOTPLUG_CPU
6360static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action,
6361 void *hcpu)
6362{
d1688a6d 6363 struct r5conf *conf = container_of(nfb, struct r5conf, cpu_notify);
36d1c647
DW		 6364 long cpu = (long)hcpu;
6365 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
6366
6367 switch (action) {
6368 case CPU_UP_PREPARE:
6369 case CPU_UP_PREPARE_FROZEN:
789b5e03 6370 if (alloc_scratch_buffer(conf, percpu)) {
36d1c647
DW		 6371 pr_err("%s: failed memory allocation for cpu%ld\n",
6372 __func__, cpu);
55af6bb5 6373 return notifier_from_errno(-ENOMEM);
36d1c647
DW		 6374 }
6375 break;
6376 case CPU_DEAD:
6377 case CPU_DEAD_FROZEN:
789b5e03 6378 free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
36d1c647
DW		 6379 break;
6380 default:
6381 break;
6382 }
6383 return NOTIFY_OK;
6384}
6385#endif
6386
d1688a6d 6387static int raid5_alloc_percpu(struct r5conf *conf)
36d1c647
DW		 6388{
6389 unsigned long cpu;
789b5e03 6390 int err = 0;
36d1c647 6391
789b5e03
ON		 6392 conf->percpu = alloc_percpu(struct raid5_percpu);
6393 if (!conf->percpu)
36d1c647 6394 return -ENOMEM;
789b5e03
ON		 6395
6396#ifdef CONFIG_HOTPLUG_CPU
6397 conf->cpu_notify.notifier_call = raid456_cpu_notify;
6398 conf->cpu_notify.priority = 0;
6399 err = register_cpu_notifier(&conf->cpu_notify);
6400 if (err)
6401 return err;
6402#endif
36d1c647
DW		 6403
6404 get_online_cpus();
36d1c647 6405 for_each_present_cpu(cpu) {
789b5e03
ON		 6406 err = alloc_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
6407 if (err) {
6408 pr_err("%s: failed memory allocation for cpu%ld\n",
6409 __func__, cpu);
36d1c647
DW		 6410 break;
6411 }
36d1c647 6412 }
36d1c647
DW		 6413 put_online_cpus();
6414
6415 return err;
6416}
6417
edbe83ab
N		 6418static unsigned long raid5_cache_scan(struct shrinker *shrink,
6419 struct shrink_control *sc)
6420{
6421 struct r5conf *conf = container_of(shrink, struct r5conf, shrinker);
2d5b569b
N		 6422 unsigned long ret = SHRINK_STOP;
6423
6424 if (mutex_trylock(&conf->cache_size_mutex)) {
6425 ret= 0;
49895bcc
N		 6426 while (ret < sc->nr_to_scan &&
6427 conf->max_nr_stripes > conf->min_nr_stripes) {
2d5b569b
N		 6428 if (drop_one_stripe(conf) == 0) {
6429 ret = SHRINK_STOP;
6430 break;
6431 }
6432 ret++;
6433 }
6434 mutex_unlock(&conf->cache_size_mutex);
edbe83ab
N		 6435 }
6436 return ret;
6437}
6438
6439static unsigned long raid5_cache_count(struct shrinker *shrink,
6440 struct shrink_control *sc)
6441{
6442 struct r5conf *conf = container_of(shrink, struct r5conf, shrinker);
6443
6444 if (conf->max_nr_stripes < conf->min_nr_stripes)
6445 /* unlikely, but not impossible */
6446 return 0;
6447 return conf->max_nr_stripes - conf->min_nr_stripes;
6448}
6449
d1688a6d 6450static struct r5conf *setup_conf(struct mddev *mddev)
1da177e4 6451{
d1688a6d 6452 struct r5conf *conf;
5e5e3e78 6453 int raid_disk, memory, max_disks;
3cb03002 6454 struct md_rdev *rdev;
1da177e4 6455 struct disk_info *disk;
0232605d 6456 char pers_name[6];
566c09c5 6457 int i;
60aaf933 6458 int group_cnt, worker_cnt_per_group;
6459 struct r5worker_group *new_group;
1da177e4 6460
91adb564
N		 6461 if (mddev->new_level != 5
6462 && mddev->new_level != 4
6463 && mddev->new_level != 6) {
0c55e022 6464 printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n",
91adb564
N		 6465 mdname(mddev), mddev->new_level);
6466 return ERR_PTR(-EIO);
1da177e4 6467 }
91adb564
N		 6468 if ((mddev->new_level == 5
6469 && !algorithm_valid_raid5(mddev->new_layout)) ||
6470 (mddev->new_level == 6
6471 && !algorithm_valid_raid6(mddev->new_layout))) {
0c55e022 6472 printk(KERN_ERR "md/raid:%s: layout %d not supported\n",
91adb564
N		 6473 mdname(mddev), mddev->new_layout);
6474 return ERR_PTR(-EIO);
99c0fb5f 6475 }
91adb564 6476 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
0c55e022 6477 printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n",
91adb564
N		 6478 mdname(mddev), mddev->raid_disks);
6479 return ERR_PTR(-EINVAL);
4bbf3771
N		 6480 }
6481
664e7c41
AN		 6482 if (!mddev->new_chunk_sectors ||
6483 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
6484 !is_power_of_2(mddev->new_chunk_sectors)) {
0c55e022
N		 6485 printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n",
6486 mdname(mddev), mddev->new_chunk_sectors << 9);
91adb564 6487 return ERR_PTR(-EINVAL);
f6705578
N		 6488 }
6489
d1688a6d 6490 conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL);
91adb564 6491 if (conf == NULL)
1da177e4 6492 goto abort;
851c30c9 6493 /* Don't enable multi-threading by default*/
60aaf933 6494 if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group,
6495 &new_group)) {
6496 conf->group_cnt = group_cnt;
6497 conf->worker_cnt_per_group = worker_cnt_per_group;
6498 conf->worker_groups = new_group;
6499 } else
851c30c9 6500 goto abort;
f5efd45a 6501 spin_lock_init(&conf->device_lock);
c46501b2 6502 seqcount_init(&conf->gen_lock);
2d5b569b 6503 mutex_init(&conf->cache_size_mutex);
b1b46486 6504 init_waitqueue_head(&conf->wait_for_quiescent);
e9e4c377
YL		 6505 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) {
6506 init_waitqueue_head(&conf->wait_for_stripe[i]);
6507 }
f5efd45a
DW		 6508 init_waitqueue_head(&conf->wait_for_overlap);
6509 INIT_LIST_HEAD(&conf->handle_list);
6510 INIT_LIST_HEAD(&conf->hold_list);
6511 INIT_LIST_HEAD(&conf->delayed_list);
6512 INIT_LIST_HEAD(&conf->bitmap_list);
c3cce6cd 6513 bio_list_init(&conf->return_bi);
773ca82f 6514 init_llist_head(&conf->released_stripes);
f5efd45a
DW		 6515 atomic_set(&conf->active_stripes, 0);
6516 atomic_set(&conf->preread_active_stripes, 0);
6517 atomic_set(&conf->active_aligned_reads, 0);
6518 conf->bypass_threshold = BYPASS_THRESHOLD;
d890fa2b 6519 conf->recovery_disabled = mddev->recovery_disabled - 1;
91adb564
N		 6520
6521 conf->raid_disks = mddev->raid_disks;
6522 if (mddev->reshape_position == MaxSector)
6523 conf->previous_raid_disks = mddev->raid_disks;
6524 else
f6705578 6525 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
5e5e3e78 6526 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
f6705578 6527
5e5e3e78 6528 conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
b55e6bfc
N		 6529 GFP_KERNEL);
6530 if (!conf->disks)
6531 goto abort;
9ffae0cf 6532
1da177e4
LT		 6533 conf->mddev = mddev;
6534
fccddba0 6535 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
1da177e4 6536 goto abort;
1da177e4 6537
566c09c5
SL		 6538 /* We init hash_locks[0] separately to that it can be used
6539 * as the reference lock in the spin_lock_nest_lock() call
6540 * in lock_all_device_hash_locks_irq in order to convince
6541 * lockdep that we know what we are doing.
6542 */
6543 spin_lock_init(conf->hash_locks);
6544 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
6545 spin_lock_init(conf->hash_locks + i);
6546
6547 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
6548 INIT_LIST_HEAD(conf->inactive_list + i);
6549
6550 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
6551 INIT_LIST_HEAD(conf->temp_inactive_list + i);
6552
36d1c647 6553 conf->level = mddev->new_level;
46d5b785 6554 conf->chunk_sectors = mddev->new_chunk_sectors;
36d1c647
DW		 6555 if (raid5_alloc_percpu(conf) != 0)
6556 goto abort;
6557
0c55e022 6558 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
1da177e4 6559
dafb20fa 6560 rdev_for_each(rdev, mddev) {
1da177e4 6561 raid_disk = rdev->raid_disk;
5e5e3e78 6562 if (raid_disk >= max_disks
1da177e4
LT		 6563 || raid_disk < 0)
6564 continue;
6565 disk = conf->disks + raid_disk;
6566
17045f52
N		 6567 if (test_bit(Replacement, &rdev->flags)) {
6568 if (disk->replacement)
6569 goto abort;
6570 disk->replacement = rdev;
6571 } else {
6572 if (disk->rdev)
6573 goto abort;
6574 disk->rdev = rdev;
6575 }
1da177e4 6576
b2d444d7 6577 if (test_bit(In_sync, &rdev->flags)) {
1da177e4 6578 char b[BDEVNAME_SIZE];
0c55e022
N		 6579 printk(KERN_INFO "md/raid:%s: device %s operational as raid"
6580 " disk %d\n",
6581 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
d6b212f4 6582 } else if (rdev->saved_raid_disk != raid_disk)
8c2e870a
NB		 6583 /* Cannot rely on bitmap to complete recovery */
6584 conf->fullsync = 1;
1da177e4
LT		 6585 }
6586
91adb564 6587 conf->level = mddev->new_level;
584acdd4 6588 if (conf->level == 6) {
16a53ecc 6589 conf->max_degraded = 2;
584acdd4
MS		 6590 if (raid6_call.xor_syndrome)
6591 conf->rmw_level = PARITY_ENABLE_RMW;
6592 else
6593 conf->rmw_level = PARITY_DISABLE_RMW;
6594 } else {
16a53ecc 6595 conf->max_degraded = 1;
584acdd4
MS		 6596 conf->rmw_level = PARITY_ENABLE_RMW;
6597 }
91adb564 6598 conf->algorithm = mddev->new_layout;
fef9c61f 6599 conf->reshape_progress = mddev->reshape_position;
e183eaed 6600 if (conf->reshape_progress != MaxSector) {
09c9e5fa 6601 conf->prev_chunk_sectors = mddev->chunk_sectors;
e183eaed 6602 conf->prev_algo = mddev->layout;
5cac6bcb
N		 6603 } else {
6604 conf->prev_chunk_sectors = conf->chunk_sectors;
6605 conf->prev_algo = conf->algorithm;
e183eaed 6606 }
1da177e4 6607
edbe83ab
N		 6608 conf->min_nr_stripes = NR_STRIPES;
6609 memory = conf->min_nr_stripes * (sizeof(struct stripe_head) +
5e5e3e78 6610 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
4bda556a 6611 atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS);
edbe83ab 6612 if (grow_stripes(conf, conf->min_nr_stripes)) {
91adb564 6613 printk(KERN_ERR
0c55e022
N		 6614 "md/raid:%s: couldn't allocate %dkB for buffers\n",
6615 mdname(mddev), memory);
91adb564
N		 6616 goto abort;
6617 } else
0c55e022
N		 6618 printk(KERN_INFO "md/raid:%s: allocated %dkB\n",
6619 mdname(mddev), memory);
edbe83ab
N		 6620 /*
6621 * Losing a stripe head costs more than the time to refill it,
6622 * it reduces the queue depth and so can hurt throughput.
6623 * So set it rather large, scaled by number of devices.
6624 */
6625 conf->shrinker.seeks = DEFAULT_SEEKS * conf->raid_disks * 4;
6626 conf->shrinker.scan_objects = raid5_cache_scan;
6627 conf->shrinker.count_objects = raid5_cache_count;
6628 conf->shrinker.batch = 128;
6629 conf->shrinker.flags = 0;
6630 register_shrinker(&conf->shrinker);
1da177e4 6631
0232605d
N		 6632 sprintf(pers_name, "raid%d", mddev->new_level);
6633 conf->thread = md_register_thread(raid5d, mddev, pers_name);
91adb564
N		 6634 if (!conf->thread) {
6635 printk(KERN_ERR
0c55e022 6636 "md/raid:%s: couldn't allocate thread.\n",
91adb564 6637 mdname(mddev));
16a53ecc
N		 6638 goto abort;
6639 }
91adb564
N		 6640
6641 return conf;
6642
6643 abort:
6644 if (conf) {
95fc17aa 6645 free_conf(conf);
91adb564
N		 6646 return ERR_PTR(-EIO);
6647 } else
6648 return ERR_PTR(-ENOMEM);
6649}
6650
c148ffdc
N		 6651static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
6652{
6653 switch (algo) {
6654 case ALGORITHM_PARITY_0:
6655 if (raid_disk < max_degraded)
6656 return 1;
6657 break;
6658 case ALGORITHM_PARITY_N:
6659 if (raid_disk >= raid_disks - max_degraded)
6660 return 1;
6661 break;
6662 case ALGORITHM_PARITY_0_6:
f72ffdd6 6663 if (raid_disk == 0 ||
c148ffdc
N		 6664 raid_disk == raid_disks - 1)
6665 return 1;
6666 break;
6667 case ALGORITHM_LEFT_ASYMMETRIC_6:
6668 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6669 case ALGORITHM_LEFT_SYMMETRIC_6:
6670 case ALGORITHM_RIGHT_SYMMETRIC_6:
6671 if (raid_disk == raid_disks - 1)
6672 return 1;
6673 }
6674 return 0;
6675}
6676
fd01b88c 6677static int run(struct mddev *mddev)
91adb564 6678{
d1688a6d 6679 struct r5conf *conf;
9f7c2220 6680 int working_disks = 0;
c148ffdc 6681 int dirty_parity_disks = 0;
3cb03002 6682 struct md_rdev *rdev;
713cf5a6 6683 struct md_rdev *journal_dev = NULL;
c148ffdc 6684 sector_t reshape_offset = 0;
17045f52 6685 int i;
b5254dd5
N		 6686 long long min_offset_diff = 0;
6687 int first = 1;
91adb564 6688
8c6ac868 6689 if (mddev->recovery_cp != MaxSector)
0c55e022 6690 printk(KERN_NOTICE "md/raid:%s: not clean"
8c6ac868
AN		 6691 " -- starting background reconstruction\n",
6692 mdname(mddev));
b5254dd5
N		 6693
6694 rdev_for_each(rdev, mddev) {
6695 long long diff;
713cf5a6
SL		 6696
6697 if (test_bit(Journal, &rdev->flags))
6698 journal_dev = rdev;
b5254dd5
N		 6699 if (rdev->raid_disk < 0)
6700 continue;
6701 diff = (rdev->new_data_offset - rdev->data_offset);
6702 if (first) {
6703 min_offset_diff = diff;
6704 first = 0;
6705 } else if (mddev->reshape_backwards &&
6706 diff < min_offset_diff)
6707 min_offset_diff = diff;
6708 else if (!mddev->reshape_backwards &&
6709 diff > min_offset_diff)
6710 min_offset_diff = diff;
6711 }
6712
91adb564
N		 6713 if (mddev->reshape_position != MaxSector) {
6714 /* Check that we can continue the reshape.
b5254dd5
N		 6715 * Difficulties arise if the stripe we would write to
6716 * next is at or after the stripe we would read from next.
6717 * For a reshape that changes the number of devices, this
6718 * is only possible for a very short time, and mdadm makes
6719 * sure that time appears to have past before assembling
6720 * the array. So we fail if that time hasn't passed.
6721 * For a reshape that keeps the number of devices the same
6722 * mdadm must be monitoring the reshape can keeping the
6723 * critical areas read-only and backed up. It will start
6724 * the array in read-only mode, so we check for that.
91adb564
N		 6725 */
6726 sector_t here_new, here_old;
6727 int old_disks;
18b00334 6728 int max_degraded = (mddev->level == 6 ? 2 : 1);
05256d98
N		 6729 int chunk_sectors;
6730 int new_data_disks;
91adb564 6731
713cf5a6
SL		 6732 if (journal_dev) {
6733 printk(KERN_ERR "md/raid:%s: don't support reshape with journal - aborting.\n",
6734 mdname(mddev));
6735 return -EINVAL;
6736 }
6737
88ce4930 6738 if (mddev->new_level != mddev->level) {
0c55e022 6739 printk(KERN_ERR "md/raid:%s: unsupported reshape "
91adb564
N		 6740 "required - aborting.\n",
6741 mdname(mddev));
6742 return -EINVAL;
6743 }
91adb564
N		 6744 old_disks = mddev->raid_disks - mddev->delta_disks;
6745 /* reshape_position must be on a new-stripe boundary, and one
6746 * further up in new geometry must map after here in old
6747 * geometry.
05256d98
N		 6748 * If the chunk sizes are different, then as we perform reshape
6749 * in units of the largest of the two, reshape_position needs
6750 * be a multiple of the largest chunk size times new data disks.
91adb564
N		 6751 */
6752 here_new = mddev->reshape_position;
05256d98
N		 6753 chunk_sectors = max(mddev->chunk_sectors, mddev->new_chunk_sectors);
6754 new_data_disks = mddev->raid_disks - max_degraded;
6755 if (sector_div(here_new, chunk_sectors * new_data_disks)) {
0c55e022
N		 6756 printk(KERN_ERR "md/raid:%s: reshape_position not "
6757 "on a stripe boundary\n", mdname(mddev));
91adb564
N		 6758 return -EINVAL;
6759 }
05256d98 6760 reshape_offset = here_new * chunk_sectors;
91adb564
N		 6761 /* here_new is the stripe we will write to */
6762 here_old = mddev->reshape_position;
05256d98 6763 sector_div(here_old, chunk_sectors * (old_disks-max_degraded));
91adb564
N		 6764 /* here_old is the first stripe that we might need to read
6765 * from */
67ac6011
N		 6766 if (mddev->delta_disks == 0) {
6767 /* We cannot be sure it is safe to start an in-place
b5254dd5 6768 * reshape. It is only safe if user-space is monitoring
67ac6011
N		 6769 * and taking constant backups.
6770 * mdadm always starts a situation like this in
6771 * readonly mode so it can take control before
6772 * allowing any writes. So just check for that.
6773 */
b5254dd5
N		 6774 if (abs(min_offset_diff) >= mddev->chunk_sectors &&
6775 abs(min_offset_diff) >= mddev->new_chunk_sectors)
6776 /* not really in-place - so OK */;
6777 else if (mddev->ro == 0) {
6778 printk(KERN_ERR "md/raid:%s: in-place reshape "
6779 "must be started in read-only mode "
6780 "- aborting\n",
0c55e022 6781 mdname(mddev));
67ac6011
N		 6782 return -EINVAL;
6783 }
2c810cdd 6784 } else if (mddev->reshape_backwards
05256d98
N		 6785 ? (here_new * chunk_sectors + min_offset_diff <=
6786 here_old * chunk_sectors)
6787 : (here_new * chunk_sectors >=
6788 here_old * chunk_sectors + (-min_offset_diff))) {
91adb564 6789 /* Reading from the same stripe as writing to - bad */
0c55e022
N		 6790 printk(KERN_ERR "md/raid:%s: reshape_position too early for "
6791 "auto-recovery - aborting.\n",
6792 mdname(mddev));
91adb564
N		 6793 return -EINVAL;
6794 }
0c55e022
N		 6795 printk(KERN_INFO "md/raid:%s: reshape will continue\n",
6796 mdname(mddev));
91adb564
N		 6797 /* OK, we should be able to continue; */
6798 } else {
6799 BUG_ON(mddev->level != mddev->new_level);
6800 BUG_ON(mddev->layout != mddev->new_layout);
664e7c41 6801 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
91adb564 6802 BUG_ON(mddev->delta_disks != 0);
1da177e4 6803 }
91adb564 6804
245f46c2
N		 6805 if (mddev->private == NULL)
6806 conf = setup_conf(mddev);
6807 else
6808 conf = mddev->private;
6809
91adb564
N		 6810 if (IS_ERR(conf))
6811 return PTR_ERR(conf);
6812
7dde2ad3
SL		 6813 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !journal_dev) {
6814 printk(KERN_ERR "md/raid:%s: journal disk is missing, force array readonly\n",
6815 mdname(mddev));
6816 mddev->ro = 1;
6817 set_disk_ro(mddev->gendisk, 1);
6818 }
6819
b5254dd5 6820 conf->min_offset_diff = min_offset_diff;
91adb564
N		 6821 mddev->thread = conf->thread;
6822 conf->thread = NULL;
6823 mddev->private = conf;
6824
17045f52
N		 6825 for (i = 0; i < conf->raid_disks && conf->previous_raid_disks;
6826 i++) {
6827 rdev = conf->disks[i].rdev;
6828 if (!rdev && conf->disks[i].replacement) {
6829 /* The replacement is all we have yet */
6830 rdev = conf->disks[i].replacement;
6831 conf->disks[i].replacement = NULL;
6832 clear_bit(Replacement, &rdev->flags);
6833 conf->disks[i].rdev = rdev;
6834 }
6835 if (!rdev)
c148ffdc 6836 continue;
17045f52
N		 6837 if (conf->disks[i].replacement &&
6838 conf->reshape_progress != MaxSector) {
6839 /* replacements and reshape simply do not mix. */
6840 printk(KERN_ERR "md: cannot handle concurrent "
6841 "replacement and reshape.\n");
6842 goto abort;
6843 }
2f115882 6844 if (test_bit(In_sync, &rdev->flags)) {
91adb564 6845 working_disks++;
2f115882
N		 6846 continue;
6847 }
c148ffdc
N		 6848 /* This disc is not fully in-sync. However if it
6849 * just stored parity (beyond the recovery_offset),
6850 * when we don't need to be concerned about the
6851 * array being dirty.
6852 * When reshape goes 'backwards', we never have
6853 * partially completed devices, so we only need
6854 * to worry about reshape going forwards.
6855 */
6856 /* Hack because v0.91 doesn't store recovery_offset properly. */
6857 if (mddev->major_version == 0 &&
6858 mddev->minor_version > 90)
6859 rdev->recovery_offset = reshape_offset;
5026d7a9 6860
c148ffdc
N		 6861 if (rdev->recovery_offset < reshape_offset) {
6862 /* We need to check old and new layout */
6863 if (!only_parity(rdev->raid_disk,
6864 conf->algorithm,
6865 conf->raid_disks,
6866 conf->max_degraded))
6867 continue;
6868 }
6869 if (!only_parity(rdev->raid_disk,
6870 conf->prev_algo,
6871 conf->previous_raid_disks,
6872 conf->max_degraded))
6873 continue;
6874 dirty_parity_disks++;
6875 }
91adb564 6876
17045f52
N		 6877 /*
6878 * 0 for a fully functional array, 1 or 2 for a degraded array.
6879 */
908f4fbd 6880 mddev->degraded = calc_degraded(conf);
91adb564 6881
674806d6 6882 if (has_failed(conf)) {
0c55e022 6883 printk(KERN_ERR "md/raid:%s: not enough operational devices"
1da177e4 6884 " (%d/%d failed)\n",
02c2de8c 6885 mdname(mddev), mddev->degraded, conf->raid_disks);
1da177e4
LT		 6886 goto abort;
6887 }
6888
91adb564 6889 /* device size must be a multiple of chunk size */
9d8f0363 6890 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
91adb564
N		 6891 mddev->resync_max_sectors = mddev->dev_sectors;
6892
c148ffdc 6893 if (mddev->degraded > dirty_parity_disks &&
1da177e4 6894 mddev->recovery_cp != MaxSector) {
6ff8d8ec
N		 6895 if (mddev->ok_start_degraded)
6896 printk(KERN_WARNING
0c55e022
N		 6897 "md/raid:%s: starting dirty degraded array"
6898 " - data corruption possible.\n",
6ff8d8ec
N		 6899 mdname(mddev));
6900 else {
6901 printk(KERN_ERR
0c55e022 6902 "md/raid:%s: cannot start dirty degraded array.\n",
6ff8d8ec
N		 6903 mdname(mddev));
6904 goto abort;
6905 }
1da177e4
LT		 6906 }
6907
1da177e4 6908 if (mddev->degraded == 0)
0c55e022
N		 6909 printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d"
6910 " devices, algorithm %d\n", mdname(mddev), conf->level,
e183eaed
N		 6911 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
6912 mddev->new_layout);
1da177e4 6913 else
0c55e022
N		 6914 printk(KERN_ALERT "md/raid:%s: raid level %d active with %d"
6915 " out of %d devices, algorithm %d\n",
6916 mdname(mddev), conf->level,
6917 mddev->raid_disks - mddev->degraded,
6918 mddev->raid_disks, mddev->new_layout);
1da177e4
LT		 6919
6920 print_raid5_conf(conf);
6921
fef9c61f 6922 if (conf->reshape_progress != MaxSector) {
fef9c61f 6923 conf->reshape_safe = conf->reshape_progress;
f6705578
N		 6924 atomic_set(&conf->reshape_stripes, 0);
6925 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6926 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6927 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6928 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6929 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 6930 "reshape");
f6705578
N		 6931 }
6932
1da177e4 6933 /* Ok, everything is just fine now */
a64c876f
N		 6934 if (mddev->to_remove == &raid5_attrs_group)
6935 mddev->to_remove = NULL;
00bcb4ac
N		 6936 else if (mddev->kobj.sd &&
6937 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
5e55e2f5 6938 printk(KERN_WARNING
4a5add49 6939 "raid5: failed to create sysfs attributes for %s\n",
5e55e2f5 6940 mdname(mddev));
4a5add49 6941 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
7a5febe9 6942
4a5add49 6943 if (mddev->queue) {
9f7c2220 6944 int chunk_size;
620125f2 6945 bool discard_supported = true;
4a5add49
N		 6946 /* read-ahead size must cover two whole stripes, which
6947 * is 2 * (datadisks) * chunksize where 'n' is the
6948 * number of raid devices
6949 */
6950 int data_disks = conf->previous_raid_disks - conf->max_degraded;
6951 int stripe = data_disks *
6952 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
6953 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
6954 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
91adb564 6955
9f7c2220
N		 6956 chunk_size = mddev->chunk_sectors << 9;
6957 blk_queue_io_min(mddev->queue, chunk_size);
6958 blk_queue_io_opt(mddev->queue, chunk_size *
6959 (conf->raid_disks - conf->max_degraded));
c78afc62 6960 mddev->queue->limits.raid_partial_stripes_expensive = 1;
620125f2
SL		 6961 /*
6962 * We can only discard a whole stripe. It doesn't make sense to
6963 * discard data disk but write parity disk
6964 */
6965 stripe = stripe * PAGE_SIZE;
4ac6875e
N		 6966 /* Round up to power of 2, as discard handling
6967 * currently assumes that */
6968 while ((stripe-1) & stripe)
6969 stripe = (stripe | (stripe-1)) + 1;
620125f2
SL		 6970 mddev->queue->limits.discard_alignment = stripe;
6971 mddev->queue->limits.discard_granularity = stripe;
6972 /*
6973 * unaligned part of discard request will be ignored, so can't
8e0e99ba 6974 * guarantee discard_zeroes_data
620125f2
SL		 6975 */
6976 mddev->queue->limits.discard_zeroes_data = 0;
8f6c2e4b 6977
5026d7a9
PA		 6978 blk_queue_max_write_same_sectors(mddev->queue, 0);
6979
05616be5 6980 rdev_for_each(rdev, mddev) {
9f7c2220
N		 6981 disk_stack_limits(mddev->gendisk, rdev->bdev,
6982 rdev->data_offset << 9);
05616be5
N		 6983 disk_stack_limits(mddev->gendisk, rdev->bdev,
6984 rdev->new_data_offset << 9);
620125f2
SL		 6985 /*
6986 * discard_zeroes_data is required, otherwise data
6987 * could be lost. Consider a scenario: discard a stripe
6988 * (the stripe could be inconsistent if
6989 * discard_zeroes_data is 0); write one disk of the
6990 * stripe (the stripe could be inconsistent again
6991 * depending on which disks are used to calculate
6992 * parity); the disk is broken; The stripe data of this
6993 * disk is lost.
6994 */
6995 if (!blk_queue_discard(bdev_get_queue(rdev->bdev)) ||
6996 !bdev_get_queue(rdev->bdev)->
6997 limits.discard_zeroes_data)
6998 discard_supported = false;
8e0e99ba
N		 6999 /* Unfortunately, discard_zeroes_data is not currently
7000 * a guarantee - just a hint. So we only allow DISCARD
7001 * if the sysadmin has confirmed that only safe devices
7002 * are in use by setting a module parameter.
7003 */
7004 if (!devices_handle_discard_safely) {
7005 if (discard_supported) {
7006 pr_info("md/raid456: discard support disabled due to uncertainty.\n");
7007 pr_info("Set raid456.devices_handle_discard_safely=Y to override.\n");
7008 }
7009 discard_supported = false;
7010 }
05616be5 7011 }
620125f2
SL		 7012
7013 if (discard_supported &&
7014 mddev->queue->limits.max_discard_sectors >= stripe &&
7015 mddev->queue->limits.discard_granularity >= stripe)
7016 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
7017 mddev->queue);
7018 else
7019 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
7020 mddev->queue);
9f7c2220 7021 }
23032a0e 7022
5c7e81c3
SL		 7023 if (journal_dev) {
7024 char b[BDEVNAME_SIZE];
7025
7026 printk(KERN_INFO"md/raid:%s: using device %s as journal\n",
7027 mdname(mddev), bdevname(journal_dev->bdev, b));
7028 r5l_init_log(conf, journal_dev);
7029 }
7030
1da177e4
LT		 7031 return 0;
7032abort:
01f96c0a 7033 md_unregister_thread(&mddev->thread);
e4f869d9
N		 7034 print_raid5_conf(conf);
7035 free_conf(conf);
1da177e4 7036 mddev->private = NULL;
0c55e022 7037 printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev));
1da177e4
LT		 7038 return -EIO;
7039}
7040
afa0f557 7041static void raid5_free(struct mddev *mddev, void *priv)
1da177e4 7042{
afa0f557 7043 struct r5conf *conf = priv;
1da177e4 7044
95fc17aa 7045 free_conf(conf);
a64c876f 7046 mddev->to_remove = &raid5_attrs_group;
1da177e4
LT		 7047}
7048
fd01b88c 7049static void status(struct seq_file *seq, struct mddev *mddev)
1da177e4 7050{
d1688a6d 7051 struct r5conf *conf = mddev->private;
1da177e4
LT		 7052 int i;
7053
9d8f0363 7054 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
3cb5edf4 7055 conf->chunk_sectors / 2, mddev->layout);
02c2de8c 7056 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
1da177e4
LT		 7057 for (i = 0; i < conf->raid_disks; i++)
7058 seq_printf (seq, "%s",
7059 conf->disks[i].rdev &&
b2d444d7 7060 test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_");
1da177e4 7061 seq_printf (seq, "]");
1da177e4
LT		 7062}
7063
d1688a6d 7064static void print_raid5_conf (struct r5conf *conf)
1da177e4
LT		 7065{
7066 int i;
7067 struct disk_info *tmp;
7068
0c55e022 7069 printk(KERN_DEBUG "RAID conf printout:\n");
1da177e4
LT		 7070 if (!conf) {
7071 printk("(conf==NULL)\n");
7072 return;
7073 }
0c55e022
N		 7074 printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level,
7075 conf->raid_disks,
7076 conf->raid_disks - conf->mddev->degraded);
1da177e4
LT		 7077
7078 for (i = 0; i < conf->raid_disks; i++) {
7079 char b[BDEVNAME_SIZE];
7080 tmp = conf->disks + i;
7081 if (tmp->rdev)
0c55e022
N		 7082 printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n",
7083 i, !test_bit(Faulty, &tmp->rdev->flags),
7084 bdevname(tmp->rdev->bdev, b));
1da177e4
LT		 7085 }
7086}
7087
fd01b88c 7088static int raid5_spare_active(struct mddev *mddev)
1da177e4
LT		 7089{
7090 int i;
d1688a6d 7091 struct r5conf *conf = mddev->private;
1da177e4 7092 struct disk_info *tmp;
6b965620
N		 7093 int count = 0;
7094 unsigned long flags;
1da177e4
LT		 7095
7096 for (i = 0; i < conf->raid_disks; i++) {
7097 tmp = conf->disks + i;
dd054fce
N		 7098 if (tmp->replacement
7099 && tmp->replacement->recovery_offset == MaxSector
7100 && !test_bit(Faulty, &tmp->replacement->flags)
7101 && !test_and_set_bit(In_sync, &tmp->replacement->flags)) {
7102 /* Replacement has just become active. */
7103 if (!tmp->rdev
7104 || !test_and_clear_bit(In_sync, &tmp->rdev->flags))
7105 count++;
7106 if (tmp->rdev) {
7107 /* Replaced device not technically faulty,
7108 * but we need to be sure it gets removed
7109 * and never re-added.
7110 */
7111 set_bit(Faulty, &tmp->rdev->flags);
7112 sysfs_notify_dirent_safe(
7113 tmp->rdev->sysfs_state);
7114 }
7115 sysfs_notify_dirent_safe(tmp->replacement->sysfs_state);
7116 } else if (tmp->rdev
70fffd0b 7117 && tmp->rdev->recovery_offset == MaxSector
b2d444d7 7118 && !test_bit(Faulty, &tmp->rdev->flags)
c04be0aa 7119 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
6b965620 7120 count++;
43c73ca4 7121 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
1da177e4
LT		 7122 }
7123 }
6b965620 7124 spin_lock_irqsave(&conf->device_lock, flags);
908f4fbd 7125 mddev->degraded = calc_degraded(conf);
6b965620 7126 spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4 7127 print_raid5_conf(conf);
6b965620 7128 return count;
1da177e4
LT		 7129}
7130
b8321b68 7131static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7132{
d1688a6d 7133 struct r5conf *conf = mddev->private;
1da177e4 7134 int err = 0;
b8321b68 7135 int number = rdev->raid_disk;
657e3e4d 7136 struct md_rdev **rdevp;
1da177e4
LT		 7137 struct disk_info *p = conf->disks + number;
7138
7139 print_raid5_conf(conf);
c2bb6242
SL		 7140 if (test_bit(Journal, &rdev->flags)) {
7141 /*
7142 * journal disk is not removable, but we need give a chance to
7143 * update superblock of other disks. Otherwise journal disk
7144 * will be considered as 'fresh'
7145 */
7146 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7147 return -EINVAL;
7148 }
657e3e4d
N		 7149 if (rdev == p->rdev)
7150 rdevp = &p->rdev;
7151 else if (rdev == p->replacement)
7152 rdevp = &p->replacement;
7153 else
7154 return 0;
7155
7156 if (number >= conf->raid_disks &&
7157 conf->reshape_progress == MaxSector)
7158 clear_bit(In_sync, &rdev->flags);
7159
7160 if (test_bit(In_sync, &rdev->flags) ||
7161 atomic_read(&rdev->nr_pending)) {
7162 err = -EBUSY;
7163 goto abort;
7164 }
7165 /* Only remove non-faulty devices if recovery
7166 * isn't possible.
7167 */
7168 if (!test_bit(Faulty, &rdev->flags) &&
7169 mddev->recovery_disabled != conf->recovery_disabled &&
7170 !has_failed(conf) &&
dd054fce 7171 (!p->replacement || p->replacement == rdev) &&
657e3e4d
N		 7172 number < conf->raid_disks) {
7173 err = -EBUSY;
7174 goto abort;
7175 }
7176 *rdevp = NULL;
7177 synchronize_rcu();
7178 if (atomic_read(&rdev->nr_pending)) {
7179 /* lost the race, try later */
7180 err = -EBUSY;
7181 *rdevp = rdev;
dd054fce
N		 7182 } else if (p->replacement) {
7183 /* We must have just cleared 'rdev' */
7184 p->rdev = p->replacement;
7185 clear_bit(Replacement, &p->replacement->flags);
7186 smp_mb(); /* Make sure other CPUs may see both as identical
7187 * but will never see neither - if they are careful
7188 */
7189 p->replacement = NULL;
7190 clear_bit(WantReplacement, &rdev->flags);
7191 } else
7192 /* We might have just removed the Replacement as faulty-
7193 * clear the bit just in case
7194 */
7195 clear_bit(WantReplacement, &rdev->flags);
1da177e4
LT		 7196abort:
7197
7198 print_raid5_conf(conf);
7199 return err;
7200}
7201
fd01b88c 7202static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7203{
d1688a6d 7204 struct r5conf *conf = mddev->private;
199050ea 7205 int err = -EEXIST;
1da177e4
LT		 7206 int disk;
7207 struct disk_info *p;
6c2fce2e
NB		 7208 int first = 0;
7209 int last = conf->raid_disks - 1;
1da177e4 7210
c2bb6242
SL		 7211 if (test_bit(Journal, &rdev->flags))
7212 return -EINVAL;
7f0da59b
N		 7213 if (mddev->recovery_disabled == conf->recovery_disabled)
7214 return -EBUSY;
7215
dc10c643 7216 if (rdev->saved_raid_disk < 0 && has_failed(conf))
1da177e4 7217 /* no point adding a device */
199050ea 7218 return -EINVAL;
1da177e4 7219
6c2fce2e
NB		 7220 if (rdev->raid_disk >= 0)
7221 first = last = rdev->raid_disk;
1da177e4
LT		 7222
7223 /*
16a53ecc
N		 7224 * find the disk ... but prefer rdev->saved_raid_disk
7225 * if possible.
1da177e4 7226 */
16a53ecc 7227 if (rdev->saved_raid_disk >= 0 &&
6c2fce2e 7228 rdev->saved_raid_disk >= first &&
16a53ecc 7229 conf->disks[rdev->saved_raid_disk].rdev == NULL)
5cfb22a1
N		 7230 first = rdev->saved_raid_disk;
7231
7232 for (disk = first; disk <= last; disk++) {
7bfec5f3
N		 7233 p = conf->disks + disk;
7234 if (p->rdev == NULL) {
b2d444d7 7235 clear_bit(In_sync, &rdev->flags);
1da177e4 7236 rdev->raid_disk = disk;
199050ea 7237 err = 0;
72626685
N		 7238 if (rdev->saved_raid_disk != disk)
7239 conf->fullsync = 1;
d6065f7b 7240 rcu_assign_pointer(p->rdev, rdev);
5cfb22a1 7241 goto out;
1da177e4 7242 }
5cfb22a1
N		 7243 }
7244 for (disk = first; disk <= last; disk++) {
7245 p = conf->disks + disk;
7bfec5f3
N		 7246 if (test_bit(WantReplacement, &p->rdev->flags) &&
7247 p->replacement == NULL) {
7248 clear_bit(In_sync, &rdev->flags);
7249 set_bit(Replacement, &rdev->flags);
7250 rdev->raid_disk = disk;
7251 err = 0;
7252 conf->fullsync = 1;
7253 rcu_assign_pointer(p->replacement, rdev);
7254 break;
7255 }
7256 }
5cfb22a1 7257out:
1da177e4 7258 print_raid5_conf(conf);
199050ea 7259 return err;
1da177e4
LT		 7260}
7261
fd01b88c 7262static int raid5_resize(struct mddev *mddev, sector_t sectors)
1da177e4
LT		 7263{
7264 /* no resync is happening, and there is enough space
7265 * on all devices, so we can resize.
7266 * We need to make sure resync covers any new space.
7267 * If the array is shrinking we should possibly wait until
7268 * any io in the removed space completes, but it hardly seems
7269 * worth it.
7270 */
a4a6125a 7271 sector_t newsize;
3cb5edf4
N		 7272 struct r5conf *conf = mddev->private;
7273
713cf5a6
SL		 7274 if (conf->log)
7275 return -EINVAL;
3cb5edf4 7276 sectors &= ~((sector_t)conf->chunk_sectors - 1);
a4a6125a
N		 7277 newsize = raid5_size(mddev, sectors, mddev->raid_disks);
7278 if (mddev->external_size &&
7279 mddev->array_sectors > newsize)
b522adcd 7280 return -EINVAL;
a4a6125a
N		 7281 if (mddev->bitmap) {
7282 int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0);
7283 if (ret)
7284 return ret;
7285 }
7286 md_set_array_sectors(mddev, newsize);
f233ea5c 7287 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 7288 revalidate_disk(mddev->gendisk);
b098636c
N		 7289 if (sectors > mddev->dev_sectors &&
7290 mddev->recovery_cp > mddev->dev_sectors) {
58c0fed4 7291 mddev->recovery_cp = mddev->dev_sectors;
1da177e4
LT		 7292 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7293 }
58c0fed4 7294 mddev->dev_sectors = sectors;
4b5c7ae8 7295 mddev->resync_max_sectors = sectors;
1da177e4
LT		 7296 return 0;
7297}
7298
fd01b88c 7299static int check_stripe_cache(struct mddev *mddev)
01ee22b4
N		 7300{
7301 /* Can only proceed if there are plenty of stripe_heads.
7302 * We need a minimum of one full stripe,, and for sensible progress
7303 * it is best to have about 4 times that.
7304 * If we require 4 times, then the default 256 4K stripe_heads will
7305 * allow for chunk sizes up to 256K, which is probably OK.
7306 * If the chunk size is greater, user-space should request more
7307 * stripe_heads first.
7308 */
d1688a6d 7309 struct r5conf *conf = mddev->private;
01ee22b4 7310 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
edbe83ab 7311 > conf->min_nr_stripes ||
01ee22b4 7312 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
edbe83ab 7313 > conf->min_nr_stripes) {
0c55e022
N		 7314 printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n",
7315 mdname(mddev),
01ee22b4
N		 7316 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
7317 / STRIPE_SIZE)*4);
7318 return 0;
7319 }
7320 return 1;
7321}
7322
fd01b88c 7323static int check_reshape(struct mddev *mddev)
29269553 7324{
d1688a6d 7325 struct r5conf *conf = mddev->private;
29269553 7326
713cf5a6
SL		 7327 if (conf->log)
7328 return -EINVAL;
88ce4930
N		 7329 if (mddev->delta_disks == 0 &&
7330 mddev->new_layout == mddev->layout &&
664e7c41 7331 mddev->new_chunk_sectors == mddev->chunk_sectors)
50ac168a 7332 return 0; /* nothing to do */
674806d6 7333 if (has_failed(conf))
ec32a2bd 7334 return -EINVAL;
fdcfbbb6 7335 if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) {
ec32a2bd
N		 7336 /* We might be able to shrink, but the devices must
7337 * be made bigger first.
7338 * For raid6, 4 is the minimum size.
7339 * Otherwise 2 is the minimum
7340 */
7341 int min = 2;
7342 if (mddev->level == 6)
7343 min = 4;
7344 if (mddev->raid_disks + mddev->delta_disks < min)
7345 return -EINVAL;
7346 }
29269553 7347
01ee22b4 7348 if (!check_stripe_cache(mddev))
29269553 7349 return -ENOSPC;
29269553 7350
738a2738
N		 7351 if (mddev->new_chunk_sectors > mddev->chunk_sectors ||
7352 mddev->delta_disks > 0)
7353 if (resize_chunks(conf,
7354 conf->previous_raid_disks
7355 + max(0, mddev->delta_disks),
7356 max(mddev->new_chunk_sectors,
7357 mddev->chunk_sectors)
7358 ) < 0)
7359 return -ENOMEM;
e56108d6
N		 7360 return resize_stripes(conf, (conf->previous_raid_disks
7361 + mddev->delta_disks));
63c70c4f
N		 7362}
7363
fd01b88c 7364static int raid5_start_reshape(struct mddev *mddev)
63c70c4f 7365{
d1688a6d 7366 struct r5conf *conf = mddev->private;
3cb03002 7367 struct md_rdev *rdev;
63c70c4f 7368 int spares = 0;
c04be0aa 7369 unsigned long flags;
63c70c4f 7370
f416885e 7371 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
63c70c4f
N		 7372 return -EBUSY;
7373
01ee22b4
N		 7374 if (!check_stripe_cache(mddev))
7375 return -ENOSPC;
7376
30b67645
N		 7377 if (has_failed(conf))
7378 return -EINVAL;
7379
c6563a8c 7380 rdev_for_each(rdev, mddev) {
469518a3
N		 7381 if (!test_bit(In_sync, &rdev->flags)
7382 && !test_bit(Faulty, &rdev->flags))
29269553 7383 spares++;
c6563a8c 7384 }
63c70c4f 7385
f416885e 7386 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
29269553
N		 7387 /* Not enough devices even to make a degraded array
7388 * of that size
7389 */
7390 return -EINVAL;
7391
ec32a2bd
N		 7392 /* Refuse to reduce size of the array. Any reductions in
7393 * array size must be through explicit setting of array_size
7394 * attribute.
7395 */
7396 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
7397 < mddev->array_sectors) {
0c55e022 7398 printk(KERN_ERR "md/raid:%s: array size must be reduced "
ec32a2bd
N		 7399 "before number of disks\n", mdname(mddev));
7400 return -EINVAL;
7401 }
7402
f6705578 7403 atomic_set(&conf->reshape_stripes, 0);
29269553 7404 spin_lock_irq(&conf->device_lock);
c46501b2 7405 write_seqcount_begin(&conf->gen_lock);
29269553 7406 conf->previous_raid_disks = conf->raid_disks;
63c70c4f 7407 conf->raid_disks += mddev->delta_disks;
09c9e5fa
AN		 7408 conf->prev_chunk_sectors = conf->chunk_sectors;
7409 conf->chunk_sectors = mddev->new_chunk_sectors;
88ce4930
N		 7410 conf->prev_algo = conf->algorithm;
7411 conf->algorithm = mddev->new_layout;
05616be5
N		 7412 conf->generation++;
7413 /* Code that selects data_offset needs to see the generation update
7414 * if reshape_progress has been set - so a memory barrier needed.
7415 */
7416 smp_mb();
2c810cdd 7417 if (mddev->reshape_backwards)
fef9c61f
N		 7418 conf->reshape_progress = raid5_size(mddev, 0, 0);
7419 else
7420 conf->reshape_progress = 0;
7421 conf->reshape_safe = conf->reshape_progress;
c46501b2 7422 write_seqcount_end(&conf->gen_lock);
29269553
N		 7423 spin_unlock_irq(&conf->device_lock);
7424
4d77e3ba
N		 7425 /* Now make sure any requests that proceeded on the assumption
7426 * the reshape wasn't running - like Discard or Read - have
7427 * completed.
7428 */
7429 mddev_suspend(mddev);
7430 mddev_resume(mddev);
7431
29269553
N		 7432 /* Add some new drives, as many as will fit.
7433 * We know there are enough to make the newly sized array work.
3424bf6a
N		 7434 * Don't add devices if we are reducing the number of
7435 * devices in the array. This is because it is not possible
7436 * to correctly record the "partially reconstructed" state of
7437 * such devices during the reshape and confusion could result.
29269553 7438 */
87a8dec9 7439 if (mddev->delta_disks >= 0) {
dafb20fa 7440 rdev_for_each(rdev, mddev)
87a8dec9
N		 7441 if (rdev->raid_disk < 0 &&
7442 !test_bit(Faulty, &rdev->flags)) {
7443 if (raid5_add_disk(mddev, rdev) == 0) {
87a8dec9 7444 if (rdev->raid_disk
9d4c7d87 7445 >= conf->previous_raid_disks)
87a8dec9 7446 set_bit(In_sync, &rdev->flags);
9d4c7d87 7447 else
87a8dec9 7448 rdev->recovery_offset = 0;
36fad858
NK		 7449
7450 if (sysfs_link_rdev(mddev, rdev))
87a8dec9 7451 /* Failure here is OK */;
50da0840 7452 }
87a8dec9
N		 7453 } else if (rdev->raid_disk >= conf->previous_raid_disks
7454 && !test_bit(Faulty, &rdev->flags)) {
7455 /* This is a spare that was manually added */
7456 set_bit(In_sync, &rdev->flags);
87a8dec9 7457 }
29269553 7458
87a8dec9
N		 7459 /* When a reshape changes the number of devices,
7460 * ->degraded is measured against the larger of the
7461 * pre and post number of devices.
7462 */
ec32a2bd 7463 spin_lock_irqsave(&conf->device_lock, flags);
908f4fbd 7464 mddev->degraded = calc_degraded(conf);
ec32a2bd
N		 7465 spin_unlock_irqrestore(&conf->device_lock, flags);
7466 }
63c70c4f 7467 mddev->raid_disks = conf->raid_disks;
e516402c 7468 mddev->reshape_position = conf->reshape_progress;
850b2b42 7469 set_bit(MD_CHANGE_DEVS, &mddev->flags);
f6705578 7470
29269553
N		 7471 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7472 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
ea358cd0 7473 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
29269553
N		 7474 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7475 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7476 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 7477 "reshape");
29269553
N		 7478 if (!mddev->sync_thread) {
7479 mddev->recovery = 0;
7480 spin_lock_irq(&conf->device_lock);
ba8805b9 7481 write_seqcount_begin(&conf->gen_lock);
29269553 7482 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
ba8805b9
N		 7483 mddev->new_chunk_sectors =
7484 conf->chunk_sectors = conf->prev_chunk_sectors;
7485 mddev->new_layout = conf->algorithm = conf->prev_algo;
05616be5
N		 7486 rdev_for_each(rdev, mddev)
7487 rdev->new_data_offset = rdev->data_offset;
7488 smp_wmb();
ba8805b9 7489 conf->generation --;
fef9c61f 7490 conf->reshape_progress = MaxSector;
1e3fa9bd 7491 mddev->reshape_position = MaxSector;
ba8805b9 7492 write_seqcount_end(&conf->gen_lock);
29269553
N		 7493 spin_unlock_irq(&conf->device_lock);
7494 return -EAGAIN;
7495 }
c8f517c4 7496 conf->reshape_checkpoint = jiffies;
29269553
N		 7497 md_wakeup_thread(mddev->sync_thread);
7498 md_new_event(mddev);
7499 return 0;
7500}
29269553 7501
ec32a2bd
N		 7502/* This is called from the reshape thread and should make any
7503 * changes needed in 'conf'
7504 */
d1688a6d 7505static void end_reshape(struct r5conf *conf)
29269553 7506{
29269553 7507
f6705578 7508 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
05616be5 7509 struct md_rdev *rdev;
f6705578 7510
f6705578 7511 spin_lock_irq(&conf->device_lock);
cea9c228 7512 conf->previous_raid_disks = conf->raid_disks;
05616be5
N		 7513 rdev_for_each(rdev, conf->mddev)
7514 rdev->data_offset = rdev->new_data_offset;
7515 smp_wmb();
fef9c61f 7516 conf->reshape_progress = MaxSector;
6cbd8148 7517 conf->mddev->reshape_position = MaxSector;
f6705578 7518 spin_unlock_irq(&conf->device_lock);
b0f9ec04 7519 wake_up(&conf->wait_for_overlap);
16a53ecc
N		 7520
7521 /* read-ahead size must cover two whole stripes, which is
7522 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
7523 */
4a5add49 7524 if (conf->mddev->queue) {
cea9c228 7525 int data_disks = conf->raid_disks - conf->max_degraded;
09c9e5fa 7526 int stripe = data_disks * ((conf->chunk_sectors << 9)
cea9c228 7527 / PAGE_SIZE);
16a53ecc
N		 7528 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
7529 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
7530 }
29269553 7531 }
29269553
N		 7532}
7533
ec32a2bd
N		 7534/* This is called from the raid5d thread with mddev_lock held.
7535 * It makes config changes to the device.
7536 */
fd01b88c 7537static void raid5_finish_reshape(struct mddev *mddev)
cea9c228 7538{
d1688a6d 7539 struct r5conf *conf = mddev->private;
cea9c228
N		 7540
7541 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7542
ec32a2bd
N		 7543 if (mddev->delta_disks > 0) {
7544 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
7545 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 7546 revalidate_disk(mddev->gendisk);
ec32a2bd
N		 7547 } else {
7548 int d;
908f4fbd
N		 7549 spin_lock_irq(&conf->device_lock);
7550 mddev->degraded = calc_degraded(conf);
7551 spin_unlock_irq(&conf->device_lock);
ec32a2bd
N		 7552 for (d = conf->raid_disks ;
7553 d < conf->raid_disks - mddev->delta_disks;
1a67dde0 7554 d++) {
3cb03002 7555 struct md_rdev *rdev = conf->disks[d].rdev;
da7613b8
N		 7556 if (rdev)
7557 clear_bit(In_sync, &rdev->flags);
7558 rdev = conf->disks[d].replacement;
7559 if (rdev)
7560 clear_bit(In_sync, &rdev->flags);
1a67dde0 7561 }
cea9c228 7562 }
88ce4930 7563 mddev->layout = conf->algorithm;
09c9e5fa 7564 mddev->chunk_sectors = conf->chunk_sectors;
ec32a2bd
N		 7565 mddev->reshape_position = MaxSector;
7566 mddev->delta_disks = 0;
2c810cdd 7567 mddev->reshape_backwards = 0;
cea9c228
N		 7568 }
7569}
7570
fd01b88c 7571static void raid5_quiesce(struct mddev *mddev, int state)
72626685 7572{
d1688a6d 7573 struct r5conf *conf = mddev->private;
72626685
N		 7574
7575 switch(state) {
e464eafd
N		 7576 case 2: /* resume for a suspend */
7577 wake_up(&conf->wait_for_overlap);
7578 break;
7579
72626685 7580 case 1: /* stop all writes */
566c09c5 7581 lock_all_device_hash_locks_irq(conf);
64bd660b
N		 7582 /* '2' tells resync/reshape to pause so that all
7583 * active stripes can drain
7584 */
7585 conf->quiesce = 2;
b1b46486 7586 wait_event_cmd(conf->wait_for_quiescent,
46031f9a
RBJ		 7587 atomic_read(&conf->active_stripes) == 0 &&
7588 atomic_read(&conf->active_aligned_reads) == 0,
566c09c5
SL		 7589 unlock_all_device_hash_locks_irq(conf),
7590 lock_all_device_hash_locks_irq(conf));
64bd660b 7591 conf->quiesce = 1;
566c09c5 7592 unlock_all_device_hash_locks_irq(conf);
64bd660b
N		 7593 /* allow reshape to continue */
7594 wake_up(&conf->wait_for_overlap);
72626685
N		 7595 break;
7596
7597 case 0: /* re-enable writes */
566c09c5 7598 lock_all_device_hash_locks_irq(conf);
72626685 7599 conf->quiesce = 0;
b1b46486 7600 wake_up(&conf->wait_for_quiescent);
e464eafd 7601 wake_up(&conf->wait_for_overlap);
566c09c5 7602 unlock_all_device_hash_locks_irq(conf);
72626685
N		 7603 break;
7604 }
e6c033f7 7605 r5l_quiesce(conf->log, state);
72626685 7606}
b15c2e57 7607
fd01b88c 7608static void *raid45_takeover_raid0(struct mddev *mddev, int level)
54071b38 7609{
e373ab10 7610 struct r0conf *raid0_conf = mddev->private;
d76c8420 7611 sector_t sectors;
54071b38 7612
f1b29bca 7613 /* for raid0 takeover only one zone is supported */
e373ab10 7614 if (raid0_conf->nr_strip_zones > 1) {
0c55e022
N		 7615 printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n",
7616 mdname(mddev));
f1b29bca
DW		 7617 return ERR_PTR(-EINVAL);
7618 }
7619
e373ab10
N		 7620 sectors = raid0_conf->strip_zone[0].zone_end;
7621 sector_div(sectors, raid0_conf->strip_zone[0].nb_dev);
3b71bd93 7622 mddev->dev_sectors = sectors;
f1b29bca 7623 mddev->new_level = level;
54071b38
TM		 7624 mddev->new_layout = ALGORITHM_PARITY_N;
7625 mddev->new_chunk_sectors = mddev->chunk_sectors;
7626 mddev->raid_disks += 1;
7627 mddev->delta_disks = 1;
7628 /* make sure it will be not marked as dirty */
7629 mddev->recovery_cp = MaxSector;
7630
7631 return setup_conf(mddev);
7632}
7633
fd01b88c 7634static void *raid5_takeover_raid1(struct mddev *mddev)
d562b0c4
N		 7635{
7636 int chunksect;
7637
7638 if (mddev->raid_disks != 2 ||
7639 mddev->degraded > 1)
7640 return ERR_PTR(-EINVAL);
7641
7642 /* Should check if there are write-behind devices? */
7643
7644 chunksect = 64*2; /* 64K by default */
7645
7646 /* The array must be an exact multiple of chunksize */
7647 while (chunksect && (mddev->array_sectors & (chunksect-1)))
7648 chunksect >>= 1;
7649
7650 if ((chunksect<<9) < STRIPE_SIZE)
7651 /* array size does not allow a suitable chunk size */
7652 return ERR_PTR(-EINVAL);
7653
7654 mddev->new_level = 5;
7655 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
664e7c41 7656 mddev->new_chunk_sectors = chunksect;
d562b0c4
N		 7657
7658 return setup_conf(mddev);
7659}
7660
fd01b88c 7661static void *raid5_takeover_raid6(struct mddev *mddev)
fc9739c6
N		 7662{
7663 int new_layout;
7664
7665 switch (mddev->layout) {
7666 case ALGORITHM_LEFT_ASYMMETRIC_6:
7667 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
7668 break;
7669 case ALGORITHM_RIGHT_ASYMMETRIC_6:
7670 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
7671 break;
7672 case ALGORITHM_LEFT_SYMMETRIC_6:
7673 new_layout = ALGORITHM_LEFT_SYMMETRIC;
7674 break;
7675 case ALGORITHM_RIGHT_SYMMETRIC_6:
7676 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
7677 break;
7678 case ALGORITHM_PARITY_0_6:
7679 new_layout = ALGORITHM_PARITY_0;
7680 break;
7681 case ALGORITHM_PARITY_N:
7682 new_layout = ALGORITHM_PARITY_N;
7683 break;
7684 default:
7685 return ERR_PTR(-EINVAL);
7686 }
7687 mddev->new_level = 5;
7688 mddev->new_layout = new_layout;
7689 mddev->delta_disks = -1;
7690 mddev->raid_disks -= 1;
7691 return setup_conf(mddev);
7692}
7693
fd01b88c 7694static int raid5_check_reshape(struct mddev *mddev)
b3546035 7695{
88ce4930
N		 7696 /* For a 2-drive array, the layout and chunk size can be changed
7697 * immediately as not restriping is needed.
7698 * For larger arrays we record the new value - after validation
7699 * to be used by a reshape pass.
b3546035 7700 */
d1688a6d 7701 struct r5conf *conf = mddev->private;
597a711b 7702 int new_chunk = mddev->new_chunk_sectors;
b3546035 7703
597a711b 7704 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
b3546035
N		 7705 return -EINVAL;
7706 if (new_chunk > 0) {
0ba459d2 7707 if (!is_power_of_2(new_chunk))
b3546035 7708 return -EINVAL;
597a711b 7709 if (new_chunk < (PAGE_SIZE>>9))
b3546035 7710 return -EINVAL;
597a711b 7711 if (mddev->array_sectors & (new_chunk-1))
b3546035
N		 7712 /* not factor of array size */
7713 return -EINVAL;
7714 }
7715
7716 /* They look valid */
7717
88ce4930 7718 if (mddev->raid_disks == 2) {
597a711b
N		 7719 /* can make the change immediately */
7720 if (mddev->new_layout >= 0) {
7721 conf->algorithm = mddev->new_layout;
7722 mddev->layout = mddev->new_layout;
88ce4930
N		 7723 }
7724 if (new_chunk > 0) {
597a711b
N		 7725 conf->chunk_sectors = new_chunk ;
7726 mddev->chunk_sectors = new_chunk;
88ce4930
N		 7727 }
7728 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7729 md_wakeup_thread(mddev->thread);
b3546035 7730 }
50ac168a 7731 return check_reshape(mddev);
88ce4930
N		 7732}
7733
fd01b88c 7734static int raid6_check_reshape(struct mddev *mddev)
88ce4930 7735{
597a711b 7736 int new_chunk = mddev->new_chunk_sectors;
50ac168a 7737
597a711b 7738 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
88ce4930 7739 return -EINVAL;
b3546035 7740 if (new_chunk > 0) {
0ba459d2 7741 if (!is_power_of_2(new_chunk))
88ce4930 7742 return -EINVAL;
597a711b 7743 if (new_chunk < (PAGE_SIZE >> 9))
88ce4930 7744 return -EINVAL;
597a711b 7745 if (mddev->array_sectors & (new_chunk-1))
88ce4930
N		 7746 /* not factor of array size */
7747 return -EINVAL;
b3546035 7748 }
88ce4930
N		 7749
7750 /* They look valid */
50ac168a 7751 return check_reshape(mddev);
b3546035
N		 7752}
7753
fd01b88c 7754static void *raid5_takeover(struct mddev *mddev)
d562b0c4
N		 7755{
7756 /* raid5 can take over:
f1b29bca 7757 * raid0 - if there is only one strip zone - make it a raid4 layout
d562b0c4
N		 7758 * raid1 - if there are two drives. We need to know the chunk size
7759 * raid4 - trivial - just use a raid4 layout.
7760 * raid6 - Providing it is a *_6 layout
d562b0c4 7761 */
f1b29bca
DW		 7762 if (mddev->level == 0)
7763 return raid45_takeover_raid0(mddev, 5);
d562b0c4
N		 7764 if (mddev->level == 1)
7765 return raid5_takeover_raid1(mddev);
e9d4758f
N		 7766 if (mddev->level == 4) {
7767 mddev->new_layout = ALGORITHM_PARITY_N;
7768 mddev->new_level = 5;
7769 return setup_conf(mddev);
7770 }
fc9739c6
N		 7771 if (mddev->level == 6)
7772 return raid5_takeover_raid6(mddev);
d562b0c4
N		 7773
7774 return ERR_PTR(-EINVAL);
7775}
7776
fd01b88c 7777static void *raid4_takeover(struct mddev *mddev)
a78d38a1 7778{
f1b29bca
DW		 7779 /* raid4 can take over:
7780 * raid0 - if there is only one strip zone
7781 * raid5 - if layout is right
a78d38a1 7782 */
f1b29bca
DW		 7783 if (mddev->level == 0)
7784 return raid45_takeover_raid0(mddev, 4);
a78d38a1
N		 7785 if (mddev->level == 5 &&
7786 mddev->layout == ALGORITHM_PARITY_N) {
7787 mddev->new_layout = 0;
7788 mddev->new_level = 4;
7789 return setup_conf(mddev);
7790 }
7791 return ERR_PTR(-EINVAL);
7792}
d562b0c4 7793
84fc4b56 7794static struct md_personality raid5_personality;
245f46c2 7795
fd01b88c 7796static void *raid6_takeover(struct mddev *mddev)
245f46c2
N		 7797{
7798 /* Currently can only take over a raid5. We map the
7799 * personality to an equivalent raid6 personality
7800 * with the Q block at the end.
7801 */
7802 int new_layout;
7803
7804 if (mddev->pers != &raid5_personality)
7805 return ERR_PTR(-EINVAL);
7806 if (mddev->degraded > 1)
7807 return ERR_PTR(-EINVAL);
7808 if (mddev->raid_disks > 253)
7809 return ERR_PTR(-EINVAL);
7810 if (mddev->raid_disks < 3)
7811 return ERR_PTR(-EINVAL);
7812
7813 switch (mddev->layout) {
7814 case ALGORITHM_LEFT_ASYMMETRIC:
7815 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
7816 break;
7817 case ALGORITHM_RIGHT_ASYMMETRIC:
7818 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
7819 break;
7820 case ALGORITHM_LEFT_SYMMETRIC:
7821 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
7822 break;
7823 case ALGORITHM_RIGHT_SYMMETRIC:
7824 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
7825 break;
7826 case ALGORITHM_PARITY_0:
7827 new_layout = ALGORITHM_PARITY_0_6;
7828 break;
7829 case ALGORITHM_PARITY_N:
7830 new_layout = ALGORITHM_PARITY_N;
7831 break;
7832 default:
7833 return ERR_PTR(-EINVAL);
7834 }
7835 mddev->new_level = 6;
7836 mddev->new_layout = new_layout;
7837 mddev->delta_disks = 1;
7838 mddev->raid_disks += 1;
7839 return setup_conf(mddev);
7840}
7841
84fc4b56 7842static struct md_personality raid6_personality =
16a53ecc
N		 7843{
7844 .name = "raid6",
7845 .level = 6,
7846 .owner = THIS_MODULE,
7847 .make_request = make_request,
7848 .run = run,
afa0f557 7849 .free = raid5_free,
16a53ecc
N		 7850 .status = status,
7851 .error_handler = error,
7852 .hot_add_disk = raid5_add_disk,
7853 .hot_remove_disk= raid5_remove_disk,
7854 .spare_active = raid5_spare_active,
7855 .sync_request = sync_request,
7856 .resize = raid5_resize,
80c3a6ce 7857 .size = raid5_size,
50ac168a 7858 .check_reshape = raid6_check_reshape,
f416885e 7859 .start_reshape = raid5_start_reshape,
cea9c228 7860 .finish_reshape = raid5_finish_reshape,
16a53ecc 7861 .quiesce = raid5_quiesce,
245f46c2 7862 .takeover = raid6_takeover,
5c675f83 7863 .congested = raid5_congested,
16a53ecc 7864};
84fc4b56 7865static struct md_personality raid5_personality =
1da177e4
LT		 7866{
7867 .name = "raid5",
2604b703 7868 .level = 5,
1da177e4
LT		 7869 .owner = THIS_MODULE,
7870 .make_request = make_request,
7871 .run = run,
afa0f557 7872 .free = raid5_free,
1da177e4
LT		 7873 .status = status,
7874 .error_handler = error,
7875 .hot_add_disk = raid5_add_disk,
7876 .hot_remove_disk= raid5_remove_disk,
7877 .spare_active = raid5_spare_active,
7878 .sync_request = sync_request,
7879 .resize = raid5_resize,
80c3a6ce 7880 .size = raid5_size,
63c70c4f
N		 7881 .check_reshape = raid5_check_reshape,
7882 .start_reshape = raid5_start_reshape,
cea9c228 7883 .finish_reshape = raid5_finish_reshape,
72626685 7884 .quiesce = raid5_quiesce,
d562b0c4 7885 .takeover = raid5_takeover,
5c675f83 7886 .congested = raid5_congested,
1da177e4
LT		 7887};
7888
84fc4b56 7889static struct md_personality raid4_personality =
1da177e4 7890{
2604b703
N		 7891 .name = "raid4",
7892 .level = 4,
7893 .owner = THIS_MODULE,
7894 .make_request = make_request,
7895 .run = run,
afa0f557 7896 .free = raid5_free,
2604b703
N		 7897 .status = status,
7898 .error_handler = error,
7899 .hot_add_disk = raid5_add_disk,
7900 .hot_remove_disk= raid5_remove_disk,
7901 .spare_active = raid5_spare_active,
7902 .sync_request = sync_request,
7903 .resize = raid5_resize,
80c3a6ce 7904 .size = raid5_size,
3d37890b
N		 7905 .check_reshape = raid5_check_reshape,
7906 .start_reshape = raid5_start_reshape,
cea9c228 7907 .finish_reshape = raid5_finish_reshape,
2604b703 7908 .quiesce = raid5_quiesce,
a78d38a1 7909 .takeover = raid4_takeover,
5c675f83 7910 .congested = raid5_congested,
2604b703
N		 7911};
7912
7913static int __init raid5_init(void)
7914{
851c30c9
SL		 7915 raid5_wq = alloc_workqueue("raid5wq",
7916 WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0);
7917 if (!raid5_wq)
7918 return -ENOMEM;
16a53ecc 7919 register_md_personality(&raid6_personality);
2604b703
N		 7920 register_md_personality(&raid5_personality);
7921 register_md_personality(&raid4_personality);
7922 return 0;
1da177e4
LT		 7923}
7924
2604b703 7925static void raid5_exit(void)
1da177e4 7926{
16a53ecc 7927 unregister_md_personality(&raid6_personality);
2604b703
N		 7928 unregister_md_personality(&raid5_personality);
7929 unregister_md_personality(&raid4_personality);
851c30c9 7930 destroy_workqueue(raid5_wq);
1da177e4
LT		 7931}
7932
7933module_init(raid5_init);
7934module_exit(raid5_exit);
7935MODULE_LICENSE("GPL");
0efb9e61 7936MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
1da177e4 7937MODULE_ALIAS("md-personality-4"); /* RAID5 */
d9d166c2
N		 7938MODULE_ALIAS("md-raid5");
7939MODULE_ALIAS("md-raid4");
2604b703
N		 7940MODULE_ALIAS("md-level-5");
7941MODULE_ALIAS("md-level-4");
16a53ecc
N		 7942MODULE_ALIAS("md-personality-8"); /* RAID6 */
7943MODULE_ALIAS("md-raid6");
7944MODULE_ALIAS("md-level-6");
7945
7946/* This used to be two separate modules, they were: */
7947MODULE_ALIAS("raid5");
7948MODULE_ALIAS("raid6");
Linux 6.x block layer and io_uring tree(s)