projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
raid5: update analysis state for failed stripe
[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 /*
356 * We don't hold any lock here yet, get_active_stripe() might
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
1da177e4
LT		 416static void release_stripe(struct stripe_head *sh)
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
b5663ba4 661static struct stripe_head *
d1688a6d 662get_active_stripe(struct r5conf *conf, sector_t sector,
a8c906ca 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{
758 return test_bit(STRIPE_BATCH_READY, &sh->state) &&
d0852df5 759 !test_bit(STRIPE_BITMAP_PENDING, &sh->state) &&
59fc630b 760 is_full_stripe_write(sh);
761}
762
763/* we only do back search */
764static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh)
765{
766 struct stripe_head *head;
767 sector_t head_sector, tmp_sec;
768 int hash;
769 int dd_idx;
770
771 if (!stripe_can_batch(sh))
772 return;
773 /* Don't cross chunks, so stripe pd_idx/qd_idx is the same */
774 tmp_sec = sh->sector;
775 if (!sector_div(tmp_sec, conf->chunk_sectors))
776 return;
777 head_sector = sh->sector - STRIPE_SECTORS;
778
779 hash = stripe_hash_locks_hash(head_sector);
780 spin_lock_irq(conf->hash_locks + hash);
781 head = __find_stripe(conf, head_sector, conf->generation);
782 if (head && !atomic_inc_not_zero(&head->count)) {
783 spin_lock(&conf->device_lock);
784 if (!atomic_read(&head->count)) {
785 if (!test_bit(STRIPE_HANDLE, &head->state))
786 atomic_inc(&conf->active_stripes);
787 BUG_ON(list_empty(&head->lru) &&
788 !test_bit(STRIPE_EXPANDING, &head->state));
789 list_del_init(&head->lru);
790 if (head->group) {
791 head->group->stripes_cnt--;
792 head->group = NULL;
793 }
794 }
795 atomic_inc(&head->count);
796 spin_unlock(&conf->device_lock);
797 }
798 spin_unlock_irq(conf->hash_locks + hash);
799
800 if (!head)
801 return;
802 if (!stripe_can_batch(head))
803 goto out;
804
805 lock_two_stripes(head, sh);
806 /* clear_batch_ready clear the flag */
807 if (!stripe_can_batch(head) || !stripe_can_batch(sh))
808 goto unlock_out;
809
810 if (sh->batch_head)
811 goto unlock_out;
812
813 dd_idx = 0;
814 while (dd_idx == sh->pd_idx || dd_idx == sh->qd_idx)
815 dd_idx++;
816 if (head->dev[dd_idx].towrite->bi_rw != sh->dev[dd_idx].towrite->bi_rw)
817 goto unlock_out;
818
819 if (head->batch_head) {
820 spin_lock(&head->batch_head->batch_lock);
821 /* This batch list is already running */
822 if (!stripe_can_batch(head)) {
823 spin_unlock(&head->batch_head->batch_lock);
824 goto unlock_out;
825 }
826
827 /*
828 * at this point, head's BATCH_READY could be cleared, but we
829 * can still add the stripe to batch list
830 */
831 list_add(&sh->batch_list, &head->batch_list);
832 spin_unlock(&head->batch_head->batch_lock);
833
834 sh->batch_head = head->batch_head;
835 } else {
836 head->batch_head = head;
837 sh->batch_head = head->batch_head;
838 spin_lock(&head->batch_lock);
839 list_add_tail(&sh->batch_list, &head->batch_list);
840 spin_unlock(&head->batch_lock);
841 }
842
843 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
844 if (atomic_dec_return(&conf->preread_active_stripes)
845 < IO_THRESHOLD)
846 md_wakeup_thread(conf->mddev->thread);
847
2b6b2457
N		 848 if (test_and_clear_bit(STRIPE_BIT_DELAY, &sh->state)) {
849 int seq = sh->bm_seq;
850 if (test_bit(STRIPE_BIT_DELAY, &sh->batch_head->state) &&
851 sh->batch_head->bm_seq > seq)
852 seq = sh->batch_head->bm_seq;
853 set_bit(STRIPE_BIT_DELAY, &sh->batch_head->state);
854 sh->batch_head->bm_seq = seq;
855 }
856
59fc630b 857 atomic_inc(&sh->count);
858unlock_out:
859 unlock_two_stripes(head, sh);
860out:
861 release_stripe(head);
862}
863
05616be5
N		 864/* Determine if 'data_offset' or 'new_data_offset' should be used
865 * in this stripe_head.
866 */
867static int use_new_offset(struct r5conf *conf, struct stripe_head *sh)
868{
869 sector_t progress = conf->reshape_progress;
870 /* Need a memory barrier to make sure we see the value
871 * of conf->generation, or ->data_offset that was set before
872 * reshape_progress was updated.
873 */
874 smp_rmb();
875 if (progress == MaxSector)
876 return 0;
877 if (sh->generation == conf->generation - 1)
878 return 0;
879 /* We are in a reshape, and this is a new-generation stripe,
880 * so use new_data_offset.
881 */
882 return 1;
883}
884
6712ecf8 885static void
4246a0b6 886raid5_end_read_request(struct bio *bi);
6712ecf8 887static void
4246a0b6 888raid5_end_write_request(struct bio *bi);
91c00924 889
c4e5ac0a 890static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
91c00924 891{
d1688a6d 892 struct r5conf *conf = sh->raid_conf;
91c00924 893 int i, disks = sh->disks;
59fc630b 894 struct stripe_head *head_sh = sh;
91c00924
DW		 895
896 might_sleep();
897
898 for (i = disks; i--; ) {
899 int rw;
9a3e1101 900 int replace_only = 0;
977df362
N		 901 struct bio *bi, *rbi;
902 struct md_rdev *rdev, *rrdev = NULL;
59fc630b 903
904 sh = head_sh;
e9c7469b
TH		 905 if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) {
906 if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags))
907 rw = WRITE_FUA;
908 else
909 rw = WRITE;
9e444768 910 if (test_bit(R5_Discard, &sh->dev[i].flags))
620125f2 911 rw |= REQ_DISCARD;
e9c7469b 912 } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags))
91c00924 913 rw = READ;
9a3e1101
N		 914 else if (test_and_clear_bit(R5_WantReplace,
915 &sh->dev[i].flags)) {
916 rw = WRITE;
917 replace_only = 1;
918 } else
91c00924 919 continue;
bc0934f0
SL		 920 if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags))
921 rw |= REQ_SYNC;
91c00924 922
59fc630b 923again:
91c00924 924 bi = &sh->dev[i].req;
977df362 925 rbi = &sh->dev[i].rreq; /* For writing to replacement */
91c00924 926
91c00924 927 rcu_read_lock();
9a3e1101 928 rrdev = rcu_dereference(conf->disks[i].replacement);
dd054fce
N		 929 smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */
930 rdev = rcu_dereference(conf->disks[i].rdev);
931 if (!rdev) {
932 rdev = rrdev;
933 rrdev = NULL;
934 }
9a3e1101
N		 935 if (rw & WRITE) {
936 if (replace_only)
937 rdev = NULL;
dd054fce
N		 938 if (rdev == rrdev)
939 /* We raced and saw duplicates */
940 rrdev = NULL;
9a3e1101 941 } else {
59fc630b 942 if (test_bit(R5_ReadRepl, &head_sh->dev[i].flags) && rrdev)
9a3e1101
N		 943 rdev = rrdev;
944 rrdev = NULL;
945 }
977df362 946
91c00924
DW		 947 if (rdev && test_bit(Faulty, &rdev->flags))
948 rdev = NULL;
949 if (rdev)
950 atomic_inc(&rdev->nr_pending);
977df362
N		 951 if (rrdev && test_bit(Faulty, &rrdev->flags))
952 rrdev = NULL;
953 if (rrdev)
954 atomic_inc(&rrdev->nr_pending);
91c00924
DW		 955 rcu_read_unlock();
956
73e92e51 957 /* We have already checked bad blocks for reads. Now
977df362
N		 958 * need to check for writes. We never accept write errors
959 * on the replacement, so we don't to check rrdev.
73e92e51
N		 960 */
961 while ((rw & WRITE) && rdev &&
962 test_bit(WriteErrorSeen, &rdev->flags)) {
963 sector_t first_bad;
964 int bad_sectors;
965 int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
966 &first_bad, &bad_sectors);
967 if (!bad)
968 break;
969
970 if (bad < 0) {
971 set_bit(BlockedBadBlocks, &rdev->flags);
972 if (!conf->mddev->external &&
973 conf->mddev->flags) {
974 /* It is very unlikely, but we might
975 * still need to write out the
976 * bad block log - better give it
977 * a chance*/
978 md_check_recovery(conf->mddev);
979 }
1850753d 980 /*
981 * Because md_wait_for_blocked_rdev
982 * will dec nr_pending, we must
983 * increment it first.
984 */
985 atomic_inc(&rdev->nr_pending);
73e92e51
N		 986 md_wait_for_blocked_rdev(rdev, conf->mddev);
987 } else {
988 /* Acknowledged bad block - skip the write */
989 rdev_dec_pending(rdev, conf->mddev);
990 rdev = NULL;
991 }
992 }
993
91c00924 994 if (rdev) {
9a3e1101
N		 995 if (s->syncing || s->expanding || s->expanded
996 || s->replacing)
91c00924
DW		 997 md_sync_acct(rdev->bdev, STRIPE_SECTORS);
998
2b7497f0
DW		 999 set_bit(STRIPE_IO_STARTED, &sh->state);
1000
2f6db2a7 1001 bio_reset(bi);
91c00924 1002 bi->bi_bdev = rdev->bdev;
2f6db2a7
KO		 1003 bi->bi_rw = rw;
1004 bi->bi_end_io = (rw & WRITE)
1005 ? raid5_end_write_request
1006 : raid5_end_read_request;
1007 bi->bi_private = sh;
1008
91c00924 1009 pr_debug("%s: for %llu schedule op %ld on disc %d\n",
e46b272b 1010 __func__, (unsigned long long)sh->sector,
91c00924
DW		 1011 bi->bi_rw, i);
1012 atomic_inc(&sh->count);
59fc630b 1013 if (sh != head_sh)
1014 atomic_inc(&head_sh->count);
05616be5 1015 if (use_new_offset(conf, sh))
4f024f37 1016 bi->bi_iter.bi_sector = (sh->sector
05616be5
N		 1017 + rdev->new_data_offset);
1018 else
4f024f37 1019 bi->bi_iter.bi_sector = (sh->sector
05616be5 1020 + rdev->data_offset);
59fc630b 1021 if (test_bit(R5_ReadNoMerge, &head_sh->dev[i].flags))
e59aa23f 1022 bi->bi_rw |= REQ_NOMERGE;
3f9e7c14 1023
d592a996
SL		 1024 if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
1025 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
1026 sh->dev[i].vec.bv_page = sh->dev[i].page;
4997b72e 1027 bi->bi_vcnt = 1;
91c00924
DW		 1028 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
1029 bi->bi_io_vec[0].bv_offset = 0;
4f024f37 1030 bi->bi_iter.bi_size = STRIPE_SIZE;
37c61ff3
SL		 1031 /*
1032 * If this is discard request, set bi_vcnt 0. We don't
1033 * want to confuse SCSI because SCSI will replace payload
1034 */
1035 if (rw & REQ_DISCARD)
1036 bi->bi_vcnt = 0;
977df362
N		 1037 if (rrdev)
1038 set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags);
e3620a3a
JB		 1039
1040 if (conf->mddev->gendisk)
1041 trace_block_bio_remap(bdev_get_queue(bi->bi_bdev),
1042 bi, disk_devt(conf->mddev->gendisk),
1043 sh->dev[i].sector);
91c00924 1044 generic_make_request(bi);
977df362
N		 1045 }
1046 if (rrdev) {
9a3e1101
N		 1047 if (s->syncing || s->expanding || s->expanded
1048 || s->replacing)
977df362
N		 1049 md_sync_acct(rrdev->bdev, STRIPE_SECTORS);
1050
1051 set_bit(STRIPE_IO_STARTED, &sh->state);
1052
2f6db2a7 1053 bio_reset(rbi);
977df362 1054 rbi->bi_bdev = rrdev->bdev;
2f6db2a7
KO		 1055 rbi->bi_rw = rw;
1056 BUG_ON(!(rw & WRITE));
1057 rbi->bi_end_io = raid5_end_write_request;
1058 rbi->bi_private = sh;
1059
977df362
N		 1060 pr_debug("%s: for %llu schedule op %ld on "
1061 "replacement disc %d\n",
1062 __func__, (unsigned long long)sh->sector,
1063 rbi->bi_rw, i);
1064 atomic_inc(&sh->count);
59fc630b 1065 if (sh != head_sh)
1066 atomic_inc(&head_sh->count);
05616be5 1067 if (use_new_offset(conf, sh))
4f024f37 1068 rbi->bi_iter.bi_sector = (sh->sector
05616be5
N		 1069 + rrdev->new_data_offset);
1070 else
4f024f37 1071 rbi->bi_iter.bi_sector = (sh->sector
05616be5 1072 + rrdev->data_offset);
d592a996
SL		 1073 if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
1074 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
1075 sh->dev[i].rvec.bv_page = sh->dev[i].page;
4997b72e 1076 rbi->bi_vcnt = 1;
977df362
N		 1077 rbi->bi_io_vec[0].bv_len = STRIPE_SIZE;
1078 rbi->bi_io_vec[0].bv_offset = 0;
4f024f37 1079 rbi->bi_iter.bi_size = STRIPE_SIZE;
37c61ff3
SL		 1080 /*
1081 * If this is discard request, set bi_vcnt 0. We don't
1082 * want to confuse SCSI because SCSI will replace payload
1083 */
1084 if (rw & REQ_DISCARD)
1085 rbi->bi_vcnt = 0;
e3620a3a
JB		 1086 if (conf->mddev->gendisk)
1087 trace_block_bio_remap(bdev_get_queue(rbi->bi_bdev),
1088 rbi, disk_devt(conf->mddev->gendisk),
1089 sh->dev[i].sector);
977df362
N		 1090 generic_make_request(rbi);
1091 }
1092 if (!rdev && !rrdev) {
b062962e 1093 if (rw & WRITE)
91c00924
DW		 1094 set_bit(STRIPE_DEGRADED, &sh->state);
1095 pr_debug("skip op %ld on disc %d for sector %llu\n",
1096 bi->bi_rw, i, (unsigned long long)sh->sector);
1097 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1098 set_bit(STRIPE_HANDLE, &sh->state);
1099 }
59fc630b 1100
1101 if (!head_sh->batch_head)
1102 continue;
1103 sh = list_first_entry(&sh->batch_list, struct stripe_head,
1104 batch_list);
1105 if (sh != head_sh)
1106 goto again;
91c00924
DW		 1107 }
1108}
1109
1110static struct dma_async_tx_descriptor *
d592a996
SL		 1111async_copy_data(int frombio, struct bio *bio, struct page **page,
1112 sector_t sector, struct dma_async_tx_descriptor *tx,
1113 struct stripe_head *sh)
91c00924 1114{
7988613b
KO		 1115 struct bio_vec bvl;
1116 struct bvec_iter iter;
91c00924 1117 struct page *bio_page;
91c00924 1118 int page_offset;
a08abd8c 1119 struct async_submit_ctl submit;
0403e382 1120 enum async_tx_flags flags = 0;
91c00924 1121
4f024f37
KO		 1122 if (bio->bi_iter.bi_sector >= sector)
1123 page_offset = (signed)(bio->bi_iter.bi_sector - sector) * 512;
91c00924 1124 else
4f024f37 1125 page_offset = (signed)(sector - bio->bi_iter.bi_sector) * -512;
a08abd8c 1126
0403e382
DW		 1127 if (frombio)
1128 flags |= ASYNC_TX_FENCE;
1129 init_async_submit(&submit, flags, tx, NULL, NULL, NULL);
1130
7988613b
KO		 1131 bio_for_each_segment(bvl, bio, iter) {
1132 int len = bvl.bv_len;
91c00924
DW		 1133 int clen;
1134 int b_offset = 0;
1135
1136 if (page_offset < 0) {
1137 b_offset = -page_offset;
1138 page_offset += b_offset;
1139 len -= b_offset;
1140 }
1141
1142 if (len > 0 && page_offset + len > STRIPE_SIZE)
1143 clen = STRIPE_SIZE - page_offset;
1144 else
1145 clen = len;
1146
1147 if (clen > 0) {
7988613b
KO		 1148 b_offset += bvl.bv_offset;
1149 bio_page = bvl.bv_page;
d592a996
SL		 1150 if (frombio) {
1151 if (sh->raid_conf->skip_copy &&
1152 b_offset == 0 && page_offset == 0 &&
1153 clen == STRIPE_SIZE)
1154 *page = bio_page;
1155 else
1156 tx = async_memcpy(*page, bio_page, page_offset,
a08abd8c 1157 b_offset, clen, &submit);
d592a996
SL		 1158 } else
1159 tx = async_memcpy(bio_page, *page, b_offset,
a08abd8c 1160 page_offset, clen, &submit);
91c00924 1161 }
a08abd8c
DW		 1162 /* chain the operations */
1163 submit.depend_tx = tx;
1164
91c00924
DW		 1165 if (clen < len) /* hit end of page */
1166 break;
1167 page_offset += len;
1168 }
1169
1170 return tx;
1171}
1172
1173static void ops_complete_biofill(void *stripe_head_ref)
1174{
1175 struct stripe_head *sh = stripe_head_ref;
34a6f80e 1176 struct bio_list return_bi = BIO_EMPTY_LIST;
e4d84909 1177 int i;
91c00924 1178
e46b272b 1179 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1180 (unsigned long long)sh->sector);
1181
1182 /* clear completed biofills */
1183 for (i = sh->disks; i--; ) {
1184 struct r5dev *dev = &sh->dev[i];
91c00924
DW		 1185
1186 /* acknowledge completion of a biofill operation */
e4d84909
DW		 1187 /* and check if we need to reply to a read request,
1188 * new R5_Wantfill requests are held off until
83de75cc 1189 * !STRIPE_BIOFILL_RUN
e4d84909
DW		 1190 */
1191 if (test_and_clear_bit(R5_Wantfill, &dev->flags)) {
91c00924 1192 struct bio *rbi, *rbi2;
91c00924 1193
91c00924
DW		 1194 BUG_ON(!dev->read);
1195 rbi = dev->read;
1196 dev->read = NULL;
4f024f37 1197 while (rbi && rbi->bi_iter.bi_sector <
91c00924
DW		 1198 dev->sector + STRIPE_SECTORS) {
1199 rbi2 = r5_next_bio(rbi, dev->sector);
34a6f80e
N		 1200 if (!raid5_dec_bi_active_stripes(rbi))
1201 bio_list_add(&return_bi, rbi);
91c00924
DW		 1202 rbi = rbi2;
1203 }
1204 }
1205 }
83de75cc 1206 clear_bit(STRIPE_BIOFILL_RUN, &sh->state);
91c00924 1207
34a6f80e 1208 return_io(&return_bi);
91c00924 1209
e4d84909 1210 set_bit(STRIPE_HANDLE, &sh->state);
91c00924
DW		 1211 release_stripe(sh);
1212}
1213
1214static void ops_run_biofill(struct stripe_head *sh)
1215{
1216 struct dma_async_tx_descriptor *tx = NULL;
a08abd8c 1217 struct async_submit_ctl submit;
91c00924
DW		 1218 int i;
1219
59fc630b 1220 BUG_ON(sh->batch_head);
e46b272b 1221 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1222 (unsigned long long)sh->sector);
1223
1224 for (i = sh->disks; i--; ) {
1225 struct r5dev *dev = &sh->dev[i];
1226 if (test_bit(R5_Wantfill, &dev->flags)) {
1227 struct bio *rbi;
b17459c0 1228 spin_lock_irq(&sh->stripe_lock);
91c00924
DW		 1229 dev->read = rbi = dev->toread;
1230 dev->toread = NULL;
b17459c0 1231 spin_unlock_irq(&sh->stripe_lock);
4f024f37 1232 while (rbi && rbi->bi_iter.bi_sector <
91c00924 1233 dev->sector + STRIPE_SECTORS) {
d592a996
SL		 1234 tx = async_copy_data(0, rbi, &dev->page,
1235 dev->sector, tx, sh);
91c00924
DW		 1236 rbi = r5_next_bio(rbi, dev->sector);
1237 }
1238 }
1239 }
1240
1241 atomic_inc(&sh->count);
a08abd8c
DW		 1242 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL);
1243 async_trigger_callback(&submit);
91c00924
DW		 1244}
1245
4e7d2c0a 1246static void mark_target_uptodate(struct stripe_head *sh, int target)
91c00924 1247{
4e7d2c0a 1248 struct r5dev *tgt;
91c00924 1249
4e7d2c0a
DW		 1250 if (target < 0)
1251 return;
91c00924 1252
4e7d2c0a 1253 tgt = &sh->dev[target];
91c00924
DW		 1254 set_bit(R5_UPTODATE, &tgt->flags);
1255 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1256 clear_bit(R5_Wantcompute, &tgt->flags);
4e7d2c0a
DW		 1257}
1258
ac6b53b6 1259static void ops_complete_compute(void *stripe_head_ref)
91c00924
DW		 1260{
1261 struct stripe_head *sh = stripe_head_ref;
91c00924 1262
e46b272b 1263 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1264 (unsigned long long)sh->sector);
1265
ac6b53b6 1266 /* mark the computed target(s) as uptodate */
4e7d2c0a 1267 mark_target_uptodate(sh, sh->ops.target);
ac6b53b6 1268 mark_target_uptodate(sh, sh->ops.target2);
4e7d2c0a 1269
ecc65c9b
DW		 1270 clear_bit(STRIPE_COMPUTE_RUN, &sh->state);
1271 if (sh->check_state == check_state_compute_run)
1272 sh->check_state = check_state_compute_result;
91c00924
DW		 1273 set_bit(STRIPE_HANDLE, &sh->state);
1274 release_stripe(sh);
1275}
1276
d6f38f31
DW		 1277/* return a pointer to the address conversion region of the scribble buffer */
1278static addr_conv_t *to_addr_conv(struct stripe_head *sh,
46d5b785 1279 struct raid5_percpu *percpu, int i)
d6f38f31 1280{
46d5b785 1281 void *addr;
1282
1283 addr = flex_array_get(percpu->scribble, i);
1284 return addr + sizeof(struct page *) * (sh->disks + 2);
1285}
1286
1287/* return a pointer to the address conversion region of the scribble buffer */
1288static struct page **to_addr_page(struct raid5_percpu *percpu, int i)
1289{
1290 void *addr;
1291
1292 addr = flex_array_get(percpu->scribble, i);
1293 return addr;
d6f38f31
DW		 1294}
1295
1296static struct dma_async_tx_descriptor *
1297ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 1298{
91c00924 1299 int disks = sh->disks;
46d5b785 1300 struct page **xor_srcs = to_addr_page(percpu, 0);
91c00924
DW		 1301 int target = sh->ops.target;
1302 struct r5dev *tgt = &sh->dev[target];
1303 struct page *xor_dest = tgt->page;
1304 int count = 0;
1305 struct dma_async_tx_descriptor *tx;
a08abd8c 1306 struct async_submit_ctl submit;
91c00924
DW		 1307 int i;
1308
59fc630b 1309 BUG_ON(sh->batch_head);
1310
91c00924 1311 pr_debug("%s: stripe %llu block: %d\n",
e46b272b 1312 __func__, (unsigned long long)sh->sector, target);
91c00924
DW		 1313 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1314
1315 for (i = disks; i--; )
1316 if (i != target)
1317 xor_srcs[count++] = sh->dev[i].page;
1318
1319 atomic_inc(&sh->count);
1320
0403e382 1321 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL,
46d5b785 1322 ops_complete_compute, sh, to_addr_conv(sh, percpu, 0));
91c00924 1323 if (unlikely(count == 1))
a08abd8c 1324 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
91c00924 1325 else
a08abd8c 1326 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924 1327
91c00924
DW		 1328 return tx;
1329}
1330
ac6b53b6
DW		 1331/* set_syndrome_sources - populate source buffers for gen_syndrome
1332 * @srcs - (struct page *) array of size sh->disks
1333 * @sh - stripe_head to parse
1334 *
1335 * Populates srcs in proper layout order for the stripe and returns the
1336 * 'count' of sources to be used in a call to async_gen_syndrome. The P
1337 * destination buffer is recorded in srcs[count] and the Q destination
1338 * is recorded in srcs[count+1]].
1339 */
584acdd4
MS		 1340static int set_syndrome_sources(struct page **srcs,
1341 struct stripe_head *sh,
1342 int srctype)
ac6b53b6
DW		 1343{
1344 int disks = sh->disks;
1345 int syndrome_disks = sh->ddf_layout ? disks : (disks - 2);
1346 int d0_idx = raid6_d0(sh);
1347 int count;
1348 int i;
1349
1350 for (i = 0; i < disks; i++)
5dd33c9a 1351 srcs[i] = NULL;
ac6b53b6
DW		 1352
1353 count = 0;
1354 i = d0_idx;
1355 do {
1356 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
584acdd4 1357 struct r5dev *dev = &sh->dev[i];
ac6b53b6 1358
584acdd4
MS		 1359 if (i == sh->qd_idx || i == sh->pd_idx ||
1360 (srctype == SYNDROME_SRC_ALL) ||
1361 (srctype == SYNDROME_SRC_WANT_DRAIN &&
1362 test_bit(R5_Wantdrain, &dev->flags)) ||
1363 (srctype == SYNDROME_SRC_WRITTEN &&
1364 dev->written))
1365 srcs[slot] = sh->dev[i].page;
ac6b53b6
DW		 1366 i = raid6_next_disk(i, disks);
1367 } while (i != d0_idx);
ac6b53b6 1368
e4424fee 1369 return syndrome_disks;
ac6b53b6
DW		 1370}
1371
1372static struct dma_async_tx_descriptor *
1373ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu)
1374{
1375 int disks = sh->disks;
46d5b785 1376 struct page **blocks = to_addr_page(percpu, 0);
ac6b53b6
DW		 1377 int target;
1378 int qd_idx = sh->qd_idx;
1379 struct dma_async_tx_descriptor *tx;
1380 struct async_submit_ctl submit;
1381 struct r5dev *tgt;
1382 struct page *dest;
1383 int i;
1384 int count;
1385
59fc630b 1386 BUG_ON(sh->batch_head);
ac6b53b6
DW		 1387 if (sh->ops.target < 0)
1388 target = sh->ops.target2;
1389 else if (sh->ops.target2 < 0)
1390 target = sh->ops.target;
91c00924 1391 else
ac6b53b6
DW		 1392 /* we should only have one valid target */
1393 BUG();
1394 BUG_ON(target < 0);
1395 pr_debug("%s: stripe %llu block: %d\n",
1396 __func__, (unsigned long long)sh->sector, target);
1397
1398 tgt = &sh->dev[target];
1399 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1400 dest = tgt->page;
1401
1402 atomic_inc(&sh->count);
1403
1404 if (target == qd_idx) {
584acdd4 1405 count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL);
ac6b53b6
DW		 1406 blocks[count] = NULL; /* regenerating p is not necessary */
1407 BUG_ON(blocks[count+1] != dest); /* q should already be set */
0403e382
DW		 1408 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1409 ops_complete_compute, sh,
46d5b785 1410 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1411 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
1412 } else {
1413 /* Compute any data- or p-drive using XOR */
1414 count = 0;
1415 for (i = disks; i-- ; ) {
1416 if (i == target || i == qd_idx)
1417 continue;
1418 blocks[count++] = sh->dev[i].page;
1419 }
1420
0403e382
DW		 1421 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1422 NULL, ops_complete_compute, sh,
46d5b785 1423 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1424 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit);
1425 }
91c00924 1426
91c00924
DW		 1427 return tx;
1428}
1429
ac6b53b6
DW		 1430static struct dma_async_tx_descriptor *
1431ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu)
1432{
1433 int i, count, disks = sh->disks;
1434 int syndrome_disks = sh->ddf_layout ? disks : disks-2;
1435 int d0_idx = raid6_d0(sh);
1436 int faila = -1, failb = -1;
1437 int target = sh->ops.target;
1438 int target2 = sh->ops.target2;
1439 struct r5dev *tgt = &sh->dev[target];
1440 struct r5dev *tgt2 = &sh->dev[target2];
1441 struct dma_async_tx_descriptor *tx;
46d5b785 1442 struct page **blocks = to_addr_page(percpu, 0);
ac6b53b6
DW		 1443 struct async_submit_ctl submit;
1444
59fc630b 1445 BUG_ON(sh->batch_head);
ac6b53b6
DW		 1446 pr_debug("%s: stripe %llu block1: %d block2: %d\n",
1447 __func__, (unsigned long long)sh->sector, target, target2);
1448 BUG_ON(target < 0 || target2 < 0);
1449 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1450 BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags));
1451
6c910a78 1452 /* we need to open-code set_syndrome_sources to handle the
ac6b53b6
DW		 1453 * slot number conversion for 'faila' and 'failb'
1454 */
1455 for (i = 0; i < disks ; i++)
5dd33c9a 1456 blocks[i] = NULL;
ac6b53b6
DW		 1457 count = 0;
1458 i = d0_idx;
1459 do {
1460 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
1461
1462 blocks[slot] = sh->dev[i].page;
1463
1464 if (i == target)
1465 faila = slot;
1466 if (i == target2)
1467 failb = slot;
1468 i = raid6_next_disk(i, disks);
1469 } while (i != d0_idx);
ac6b53b6
DW		 1470
1471 BUG_ON(faila == failb);
1472 if (failb < faila)
1473 swap(faila, failb);
1474 pr_debug("%s: stripe: %llu faila: %d failb: %d\n",
1475 __func__, (unsigned long long)sh->sector, faila, failb);
1476
1477 atomic_inc(&sh->count);
1478
1479 if (failb == syndrome_disks+1) {
1480 /* Q disk is one of the missing disks */
1481 if (faila == syndrome_disks) {
1482 /* Missing P+Q, just recompute */
0403e382
DW		 1483 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1484 ops_complete_compute, sh,
46d5b785 1485 to_addr_conv(sh, percpu, 0));
e4424fee 1486 return async_gen_syndrome(blocks, 0, syndrome_disks+2,
ac6b53b6
DW		 1487 STRIPE_SIZE, &submit);
1488 } else {
1489 struct page *dest;
1490 int data_target;
1491 int qd_idx = sh->qd_idx;
1492
1493 /* Missing D+Q: recompute D from P, then recompute Q */
1494 if (target == qd_idx)
1495 data_target = target2;
1496 else
1497 data_target = target;
1498
1499 count = 0;
1500 for (i = disks; i-- ; ) {
1501 if (i == data_target || i == qd_idx)
1502 continue;
1503 blocks[count++] = sh->dev[i].page;
1504 }
1505 dest = sh->dev[data_target].page;
0403e382
DW		 1506 init_async_submit(&submit,
1507 ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1508 NULL, NULL, NULL,
46d5b785 1509 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1510 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE,
1511 &submit);
1512
584acdd4 1513 count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL);
0403e382
DW		 1514 init_async_submit(&submit, ASYNC_TX_FENCE, tx,
1515 ops_complete_compute, sh,
46d5b785 1516 to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1517 return async_gen_syndrome(blocks, 0, count+2,
1518 STRIPE_SIZE, &submit);
1519 }
ac6b53b6 1520 } else {
6c910a78
DW		 1521 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1522 ops_complete_compute, sh,
46d5b785 1523 to_addr_conv(sh, percpu, 0));
6c910a78
DW		 1524 if (failb == syndrome_disks) {
1525 /* We're missing D+P. */
1526 return async_raid6_datap_recov(syndrome_disks+2,
1527 STRIPE_SIZE, faila,
1528 blocks, &submit);
1529 } else {
1530 /* We're missing D+D. */
1531 return async_raid6_2data_recov(syndrome_disks+2,
1532 STRIPE_SIZE, faila, failb,
1533 blocks, &submit);
1534 }
ac6b53b6
DW		 1535 }
1536}
1537
91c00924
DW		 1538static void ops_complete_prexor(void *stripe_head_ref)
1539{
1540 struct stripe_head *sh = stripe_head_ref;
1541
e46b272b 1542 pr_debug("%s: stripe %llu\n", __func__,
91c00924 1543 (unsigned long long)sh->sector);
91c00924
DW		 1544}
1545
1546static struct dma_async_tx_descriptor *
584acdd4
MS		 1547ops_run_prexor5(struct stripe_head *sh, struct raid5_percpu *percpu,
1548 struct dma_async_tx_descriptor *tx)
91c00924 1549{
91c00924 1550 int disks = sh->disks;
46d5b785 1551 struct page **xor_srcs = to_addr_page(percpu, 0);
91c00924 1552 int count = 0, pd_idx = sh->pd_idx, i;
a08abd8c 1553 struct async_submit_ctl submit;
91c00924
DW		 1554
1555 /* existing parity data subtracted */
1556 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1557
59fc630b 1558 BUG_ON(sh->batch_head);
e46b272b 1559 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1560 (unsigned long long)sh->sector);
1561
1562 for (i = disks; i--; ) {
1563 struct r5dev *dev = &sh->dev[i];
1564 /* Only process blocks that are known to be uptodate */
d8ee0728 1565 if (test_bit(R5_Wantdrain, &dev->flags))
91c00924
DW		 1566 xor_srcs[count++] = dev->page;
1567 }
1568
0403e382 1569 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
46d5b785 1570 ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0));
a08abd8c 1571 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924
DW		 1572
1573 return tx;
1574}
1575
584acdd4
MS		 1576static struct dma_async_tx_descriptor *
1577ops_run_prexor6(struct stripe_head *sh, struct raid5_percpu *percpu,
1578 struct dma_async_tx_descriptor *tx)
1579{
1580 struct page **blocks = to_addr_page(percpu, 0);
1581 int count;
1582 struct async_submit_ctl submit;
1583
1584 pr_debug("%s: stripe %llu\n", __func__,
1585 (unsigned long long)sh->sector);
1586
1587 count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_WANT_DRAIN);
1588
1589 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_PQ_XOR_DST, tx,
1590 ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0));
1591 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
1592
1593 return tx;
1594}
1595
91c00924 1596static struct dma_async_tx_descriptor *
d8ee0728 1597ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
91c00924
DW		 1598{
1599 int disks = sh->disks;
d8ee0728 1600 int i;
59fc630b 1601 struct stripe_head *head_sh = sh;
91c00924 1602
e46b272b 1603 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1604 (unsigned long long)sh->sector);
1605
1606 for (i = disks; i--; ) {
59fc630b 1607 struct r5dev *dev;
91c00924 1608 struct bio *chosen;
91c00924 1609
59fc630b 1610 sh = head_sh;
1611 if (test_and_clear_bit(R5_Wantdrain, &head_sh->dev[i].flags)) {
91c00924
DW		 1612 struct bio *wbi;
1613
59fc630b 1614again:
1615 dev = &sh->dev[i];
b17459c0 1616 spin_lock_irq(&sh->stripe_lock);
91c00924
DW		 1617 chosen = dev->towrite;
1618 dev->towrite = NULL;
7a87f434 1619 sh->overwrite_disks = 0;
91c00924
DW		 1620 BUG_ON(dev->written);
1621 wbi = dev->written = chosen;
b17459c0 1622 spin_unlock_irq(&sh->stripe_lock);
d592a996 1623 WARN_ON(dev->page != dev->orig_page);
91c00924 1624
4f024f37 1625 while (wbi && wbi->bi_iter.bi_sector <
91c00924 1626 dev->sector + STRIPE_SECTORS) {
e9c7469b
TH		 1627 if (wbi->bi_rw & REQ_FUA)
1628 set_bit(R5_WantFUA, &dev->flags);
bc0934f0
SL		 1629 if (wbi->bi_rw & REQ_SYNC)
1630 set_bit(R5_SyncIO, &dev->flags);
9e444768 1631 if (wbi->bi_rw & REQ_DISCARD)
620125f2 1632 set_bit(R5_Discard, &dev->flags);
d592a996
SL		 1633 else {
1634 tx = async_copy_data(1, wbi, &dev->page,
1635 dev->sector, tx, sh);
1636 if (dev->page != dev->orig_page) {
1637 set_bit(R5_SkipCopy, &dev->flags);
1638 clear_bit(R5_UPTODATE, &dev->flags);
1639 clear_bit(R5_OVERWRITE, &dev->flags);
1640 }
1641 }
91c00924
DW		 1642 wbi = r5_next_bio(wbi, dev->sector);
1643 }
59fc630b 1644
1645 if (head_sh->batch_head) {
1646 sh = list_first_entry(&sh->batch_list,
1647 struct stripe_head,
1648 batch_list);
1649 if (sh == head_sh)
1650 continue;
1651 goto again;
1652 }
91c00924
DW		 1653 }
1654 }
1655
1656 return tx;
1657}
1658
ac6b53b6 1659static void ops_complete_reconstruct(void *stripe_head_ref)
91c00924
DW		 1660{
1661 struct stripe_head *sh = stripe_head_ref;
ac6b53b6
DW		 1662 int disks = sh->disks;
1663 int pd_idx = sh->pd_idx;
1664 int qd_idx = sh->qd_idx;
1665 int i;
9e444768 1666 bool fua = false, sync = false, discard = false;
91c00924 1667
e46b272b 1668 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1669 (unsigned long long)sh->sector);
1670
bc0934f0 1671 for (i = disks; i--; ) {
e9c7469b 1672 fua |= test_bit(R5_WantFUA, &sh->dev[i].flags);
bc0934f0 1673 sync |= test_bit(R5_SyncIO, &sh->dev[i].flags);
9e444768 1674 discard |= test_bit(R5_Discard, &sh->dev[i].flags);
bc0934f0 1675 }
e9c7469b 1676
91c00924
DW		 1677 for (i = disks; i--; ) {
1678 struct r5dev *dev = &sh->dev[i];
ac6b53b6 1679
e9c7469b 1680 if (dev->written || i == pd_idx || i == qd_idx) {
d592a996 1681 if (!discard && !test_bit(R5_SkipCopy, &dev->flags))
9e444768 1682 set_bit(R5_UPTODATE, &dev->flags);
e9c7469b
TH		 1683 if (fua)
1684 set_bit(R5_WantFUA, &dev->flags);
bc0934f0
SL		 1685 if (sync)
1686 set_bit(R5_SyncIO, &dev->flags);
e9c7469b 1687 }
91c00924
DW		 1688 }
1689
d8ee0728
DW		 1690 if (sh->reconstruct_state == reconstruct_state_drain_run)
1691 sh->reconstruct_state = reconstruct_state_drain_result;
1692 else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run)
1693 sh->reconstruct_state = reconstruct_state_prexor_drain_result;
1694 else {
1695 BUG_ON(sh->reconstruct_state != reconstruct_state_run);
1696 sh->reconstruct_state = reconstruct_state_result;
1697 }
91c00924
DW		 1698
1699 set_bit(STRIPE_HANDLE, &sh->state);
1700 release_stripe(sh);
1701}
1702
1703static void
ac6b53b6
DW		 1704ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu,
1705 struct dma_async_tx_descriptor *tx)
91c00924 1706{
91c00924 1707 int disks = sh->disks;
59fc630b 1708 struct page **xor_srcs;
a08abd8c 1709 struct async_submit_ctl submit;
59fc630b 1710 int count, pd_idx = sh->pd_idx, i;
91c00924 1711 struct page *xor_dest;
d8ee0728 1712 int prexor = 0;
91c00924 1713 unsigned long flags;
59fc630b 1714 int j = 0;
1715 struct stripe_head *head_sh = sh;
1716 int last_stripe;
91c00924 1717
e46b272b 1718 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1719 (unsigned long long)sh->sector);
1720
620125f2
SL		 1721 for (i = 0; i < sh->disks; i++) {
1722 if (pd_idx == i)
1723 continue;
1724 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1725 break;
1726 }
1727 if (i >= sh->disks) {
1728 atomic_inc(&sh->count);
620125f2
SL		 1729 set_bit(R5_Discard, &sh->dev[pd_idx].flags);
1730 ops_complete_reconstruct(sh);
1731 return;
1732 }
59fc630b 1733again:
1734 count = 0;
1735 xor_srcs = to_addr_page(percpu, j);
91c00924
DW		 1736 /* check if prexor is active which means only process blocks
1737 * that are part of a read-modify-write (written)
1738 */
59fc630b 1739 if (head_sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
d8ee0728 1740 prexor = 1;
91c00924
DW		 1741 xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1742 for (i = disks; i--; ) {
1743 struct r5dev *dev = &sh->dev[i];
59fc630b 1744 if (head_sh->dev[i].written)
91c00924
DW		 1745 xor_srcs[count++] = dev->page;
1746 }
1747 } else {
1748 xor_dest = sh->dev[pd_idx].page;
1749 for (i = disks; i--; ) {
1750 struct r5dev *dev = &sh->dev[i];
1751 if (i != pd_idx)
1752 xor_srcs[count++] = dev->page;
1753 }
1754 }
1755
91c00924
DW		 1756 /* 1/ if we prexor'd then the dest is reused as a source
1757 * 2/ if we did not prexor then we are redoing the parity
1758 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
1759 * for the synchronous xor case
1760 */
59fc630b 1761 last_stripe = !head_sh->batch_head ||
1762 list_first_entry(&sh->batch_list,
1763 struct stripe_head, batch_list) == head_sh;
1764 if (last_stripe) {
1765 flags = ASYNC_TX_ACK |
1766 (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
1767
1768 atomic_inc(&head_sh->count);
1769 init_async_submit(&submit, flags, tx, ops_complete_reconstruct, head_sh,
1770 to_addr_conv(sh, percpu, j));
1771 } else {
1772 flags = prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST;
1773 init_async_submit(&submit, flags, tx, NULL, NULL,
1774 to_addr_conv(sh, percpu, j));
1775 }
91c00924 1776
a08abd8c
DW		 1777 if (unlikely(count == 1))
1778 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
1779 else
1780 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
59fc630b 1781 if (!last_stripe) {
1782 j++;
1783 sh = list_first_entry(&sh->batch_list, struct stripe_head,
1784 batch_list);
1785 goto again;
1786 }
91c00924
DW		 1787}
1788
ac6b53b6
DW		 1789static void
1790ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu,
1791 struct dma_async_tx_descriptor *tx)
1792{
1793 struct async_submit_ctl submit;
59fc630b 1794 struct page **blocks;
1795 int count, i, j = 0;
1796 struct stripe_head *head_sh = sh;
1797 int last_stripe;
584acdd4
MS		 1798 int synflags;
1799 unsigned long txflags;
ac6b53b6
DW		 1800
1801 pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector);
1802
620125f2
SL		 1803 for (i = 0; i < sh->disks; i++) {
1804 if (sh->pd_idx == i || sh->qd_idx == i)
1805 continue;
1806 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1807 break;
1808 }
1809 if (i >= sh->disks) {
1810 atomic_inc(&sh->count);
620125f2
SL		 1811 set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
1812 set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
1813 ops_complete_reconstruct(sh);
1814 return;
1815 }
1816
59fc630b 1817again:
1818 blocks = to_addr_page(percpu, j);
584acdd4
MS		 1819
1820 if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
1821 synflags = SYNDROME_SRC_WRITTEN;
1822 txflags = ASYNC_TX_ACK | ASYNC_TX_PQ_XOR_DST;
1823 } else {
1824 synflags = SYNDROME_SRC_ALL;
1825 txflags = ASYNC_TX_ACK;
1826 }
1827
1828 count = set_syndrome_sources(blocks, sh, synflags);
59fc630b 1829 last_stripe = !head_sh->batch_head ||
1830 list_first_entry(&sh->batch_list,
1831 struct stripe_head, batch_list) == head_sh;
1832
1833 if (last_stripe) {
1834 atomic_inc(&head_sh->count);
584acdd4 1835 init_async_submit(&submit, txflags, tx, ops_complete_reconstruct,
59fc630b 1836 head_sh, to_addr_conv(sh, percpu, j));
1837 } else
1838 init_async_submit(&submit, 0, tx, NULL, NULL,
1839 to_addr_conv(sh, percpu, j));
48769695 1840 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
59fc630b 1841 if (!last_stripe) {
1842 j++;
1843 sh = list_first_entry(&sh->batch_list, struct stripe_head,
1844 batch_list);
1845 goto again;
1846 }
91c00924
DW		 1847}
1848
1849static void ops_complete_check(void *stripe_head_ref)
1850{
1851 struct stripe_head *sh = stripe_head_ref;
91c00924 1852
e46b272b 1853 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1854 (unsigned long long)sh->sector);
1855
ecc65c9b 1856 sh->check_state = check_state_check_result;
91c00924
DW		 1857 set_bit(STRIPE_HANDLE, &sh->state);
1858 release_stripe(sh);
1859}
1860
ac6b53b6 1861static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 1862{
91c00924 1863 int disks = sh->disks;
ac6b53b6
DW		 1864 int pd_idx = sh->pd_idx;
1865 int qd_idx = sh->qd_idx;
1866 struct page *xor_dest;
46d5b785 1867 struct page **xor_srcs = to_addr_page(percpu, 0);
91c00924 1868 struct dma_async_tx_descriptor *tx;
a08abd8c 1869 struct async_submit_ctl submit;
ac6b53b6
DW		 1870 int count;
1871 int i;
91c00924 1872
e46b272b 1873 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1874 (unsigned long long)sh->sector);
1875
59fc630b 1876 BUG_ON(sh->batch_head);
ac6b53b6
DW		 1877 count = 0;
1878 xor_dest = sh->dev[pd_idx].page;
1879 xor_srcs[count++] = xor_dest;
91c00924 1880 for (i = disks; i--; ) {
ac6b53b6
DW		 1881 if (i == pd_idx || i == qd_idx)
1882 continue;
1883 xor_srcs[count++] = sh->dev[i].page;
91c00924
DW		 1884 }
1885
d6f38f31 1886 init_async_submit(&submit, 0, NULL, NULL, NULL,
46d5b785 1887 to_addr_conv(sh, percpu, 0));
099f53cb 1888 tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
a08abd8c 1889 &sh->ops.zero_sum_result, &submit);
91c00924 1890
91c00924 1891 atomic_inc(&sh->count);
a08abd8c
DW		 1892 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL);
1893 tx = async_trigger_callback(&submit);
91c00924
DW		 1894}
1895
ac6b53b6
DW		 1896static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp)
1897{
46d5b785 1898 struct page **srcs = to_addr_page(percpu, 0);
ac6b53b6
DW		 1899 struct async_submit_ctl submit;
1900 int count;
1901
1902 pr_debug("%s: stripe %llu checkp: %d\n", __func__,
1903 (unsigned long long)sh->sector, checkp);
1904
59fc630b 1905 BUG_ON(sh->batch_head);
584acdd4 1906 count = set_syndrome_sources(srcs, sh, SYNDROME_SRC_ALL);
ac6b53b6
DW		 1907 if (!checkp)
1908 srcs[count] = NULL;
91c00924 1909
91c00924 1910 atomic_inc(&sh->count);
ac6b53b6 1911 init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check,
46d5b785 1912 sh, to_addr_conv(sh, percpu, 0));
ac6b53b6
DW		 1913 async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE,
1914 &sh->ops.zero_sum_result, percpu->spare_page, &submit);
91c00924
DW		 1915}
1916
51acbcec 1917static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
91c00924
DW		 1918{
1919 int overlap_clear = 0, i, disks = sh->disks;
1920 struct dma_async_tx_descriptor *tx = NULL;
d1688a6d 1921 struct r5conf *conf = sh->raid_conf;
ac6b53b6 1922 int level = conf->level;
d6f38f31
DW		 1923 struct raid5_percpu *percpu;
1924 unsigned long cpu;
91c00924 1925
d6f38f31
DW		 1926 cpu = get_cpu();
1927 percpu = per_cpu_ptr(conf->percpu, cpu);
83de75cc 1928 if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) {
91c00924
DW		 1929 ops_run_biofill(sh);
1930 overlap_clear++;
1931 }
1932
7b3a871e 1933 if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) {
ac6b53b6
DW		 1934 if (level < 6)
1935 tx = ops_run_compute5(sh, percpu);
1936 else {
1937 if (sh->ops.target2 < 0 || sh->ops.target < 0)
1938 tx = ops_run_compute6_1(sh, percpu);
1939 else
1940 tx = ops_run_compute6_2(sh, percpu);
1941 }
1942 /* terminate the chain if reconstruct is not set to be run */
1943 if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request))
7b3a871e
DW		 1944 async_tx_ack(tx);
1945 }
91c00924 1946
584acdd4
MS		 1947 if (test_bit(STRIPE_OP_PREXOR, &ops_request)) {
1948 if (level < 6)
1949 tx = ops_run_prexor5(sh, percpu, tx);
1950 else
1951 tx = ops_run_prexor6(sh, percpu, tx);
1952 }
91c00924 1953
600aa109 1954 if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) {
d8ee0728 1955 tx = ops_run_biodrain(sh, tx);
91c00924
DW		 1956 overlap_clear++;
1957 }
1958
ac6b53b6
DW		 1959 if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) {
1960 if (level < 6)
1961 ops_run_reconstruct5(sh, percpu, tx);
1962 else
1963 ops_run_reconstruct6(sh, percpu, tx);
1964 }
91c00924 1965
ac6b53b6
DW		 1966 if (test_bit(STRIPE_OP_CHECK, &ops_request)) {
1967 if (sh->check_state == check_state_run)
1968 ops_run_check_p(sh, percpu);
1969 else if (sh->check_state == check_state_run_q)
1970 ops_run_check_pq(sh, percpu, 0);
1971 else if (sh->check_state == check_state_run_pq)
1972 ops_run_check_pq(sh, percpu, 1);
1973 else
1974 BUG();
1975 }
91c00924 1976
59fc630b 1977 if (overlap_clear && !sh->batch_head)
91c00924
DW		 1978 for (i = disks; i--; ) {
1979 struct r5dev *dev = &sh->dev[i];
1980 if (test_and_clear_bit(R5_Overlap, &dev->flags))
1981 wake_up(&sh->raid_conf->wait_for_overlap);
1982 }
d6f38f31 1983 put_cpu();
91c00924
DW		 1984}
1985
f18c1a35
N		 1986static struct stripe_head *alloc_stripe(struct kmem_cache *sc, gfp_t gfp)
1987{
1988 struct stripe_head *sh;
1989
1990 sh = kmem_cache_zalloc(sc, gfp);
1991 if (sh) {
1992 spin_lock_init(&sh->stripe_lock);
1993 spin_lock_init(&sh->batch_lock);
1994 INIT_LIST_HEAD(&sh->batch_list);
1995 INIT_LIST_HEAD(&sh->lru);
1996 atomic_set(&sh->count, 1);
1997 }
1998 return sh;
1999}
486f0644 2000static int grow_one_stripe(struct r5conf *conf, gfp_t gfp)
1da177e4
LT		 2001{
2002 struct stripe_head *sh;
f18c1a35
N		 2003
2004 sh = alloc_stripe(conf->slab_cache, gfp);
3f294f4f
N		 2005 if (!sh)
2006 return 0;
6ce32846 2007
3f294f4f 2008 sh->raid_conf = conf;
3f294f4f 2009
a9683a79 2010 if (grow_buffers(sh, gfp)) {
e4e11e38 2011 shrink_buffers(sh);
3f294f4f
N		 2012 kmem_cache_free(conf->slab_cache, sh);
2013 return 0;
2014 }
486f0644
N		 2015 sh->hash_lock_index =
2016 conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS;
3f294f4f 2017 /* we just created an active stripe so... */
3f294f4f 2018 atomic_inc(&conf->active_stripes);
59fc630b 2019
3f294f4f 2020 release_stripe(sh);
486f0644 2021 conf->max_nr_stripes++;
3f294f4f
N		 2022 return 1;
2023}
2024
d1688a6d 2025static int grow_stripes(struct r5conf *conf, int num)
3f294f4f 2026{
e18b890b 2027 struct kmem_cache *sc;
5e5e3e78 2028 int devs = max(conf->raid_disks, conf->previous_raid_disks);
1da177e4 2029
f4be6b43
N		 2030 if (conf->mddev->gendisk)
2031 sprintf(conf->cache_name[0],
2032 "raid%d-%s", conf->level, mdname(conf->mddev));
2033 else
2034 sprintf(conf->cache_name[0],
2035 "raid%d-%p", conf->level, conf->mddev);
2036 sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]);
2037
ad01c9e3
N		 2038 conf->active_name = 0;
2039 sc = kmem_cache_create(conf->cache_name[conf->active_name],
1da177e4 2040 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
20c2df83 2041 0, 0, NULL);
1da177e4
LT		 2042 if (!sc)
2043 return 1;
2044 conf->slab_cache = sc;
ad01c9e3 2045 conf->pool_size = devs;
486f0644
N		 2046 while (num--)
2047 if (!grow_one_stripe(conf, GFP_KERNEL))
1da177e4 2048 return 1;
486f0644 2049
1da177e4
LT		 2050 return 0;
2051}
29269553 2052
d6f38f31
DW		 2053/**
2054 * scribble_len - return the required size of the scribble region
2055 * @num - total number of disks in the array
2056 *
2057 * The size must be enough to contain:
2058 * 1/ a struct page pointer for each device in the array +2
2059 * 2/ room to convert each entry in (1) to its corresponding dma
2060 * (dma_map_page()) or page (page_address()) address.
2061 *
2062 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
2063 * calculate over all devices (not just the data blocks), using zeros in place
2064 * of the P and Q blocks.
2065 */
46d5b785 2066static struct flex_array *scribble_alloc(int num, int cnt, gfp_t flags)
d6f38f31 2067{
46d5b785 2068 struct flex_array *ret;
d6f38f31
DW		 2069 size_t len;
2070
2071 len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2);
46d5b785 2072 ret = flex_array_alloc(len, cnt, flags);
2073 if (!ret)
2074 return NULL;
2075 /* always prealloc all elements, so no locking is required */
2076 if (flex_array_prealloc(ret, 0, cnt, flags)) {
2077 flex_array_free(ret);
2078 return NULL;
2079 }
2080 return ret;
d6f38f31
DW		 2081}
2082
738a2738
N		 2083static int resize_chunks(struct r5conf *conf, int new_disks, int new_sectors)
2084{
2085 unsigned long cpu;
2086 int err = 0;
2087
2088 mddev_suspend(conf->mddev);
2089 get_online_cpus();
2090 for_each_present_cpu(cpu) {
2091 struct raid5_percpu *percpu;
2092 struct flex_array *scribble;
2093
2094 percpu = per_cpu_ptr(conf->percpu, cpu);
2095 scribble = scribble_alloc(new_disks,
2096 new_sectors / STRIPE_SECTORS,
2097 GFP_NOIO);
2098
2099 if (scribble) {
2100 flex_array_free(percpu->scribble);
2101 percpu->scribble = scribble;
2102 } else {
2103 err = -ENOMEM;
2104 break;
2105 }
2106 }
2107 put_online_cpus();
2108 mddev_resume(conf->mddev);
2109 return err;
2110}
2111
d1688a6d 2112static int resize_stripes(struct r5conf *conf, int newsize)
ad01c9e3
N		 2113{
2114 /* Make all the stripes able to hold 'newsize' devices.
2115 * New slots in each stripe get 'page' set to a new page.
2116 *
2117 * This happens in stages:
2118 * 1/ create a new kmem_cache and allocate the required number of
2119 * stripe_heads.
83f0d77a 2120 * 2/ gather all the old stripe_heads and transfer the pages across
ad01c9e3
N		 2121 * to the new stripe_heads. This will have the side effect of
2122 * freezing the array as once all stripe_heads have been collected,
2123 * no IO will be possible. Old stripe heads are freed once their
2124 * pages have been transferred over, and the old kmem_cache is
2125 * freed when all stripes are done.
2126 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
2127 * we simple return a failre status - no need to clean anything up.
2128 * 4/ allocate new pages for the new slots in the new stripe_heads.
2129 * If this fails, we don't bother trying the shrink the
2130 * stripe_heads down again, we just leave them as they are.
2131 * As each stripe_head is processed the new one is released into
2132 * active service.
2133 *
2134 * Once step2 is started, we cannot afford to wait for a write,
2135 * so we use GFP_NOIO allocations.
2136 */
2137 struct stripe_head *osh, *nsh;
2138 LIST_HEAD(newstripes);
2139 struct disk_info *ndisks;
b5470dc5 2140 int err;
e18b890b 2141 struct kmem_cache *sc;
ad01c9e3 2142 int i;
566c09c5 2143 int hash, cnt;
ad01c9e3
N		 2144
2145 if (newsize <= conf->pool_size)
2146 return 0; /* never bother to shrink */
2147
b5470dc5
DW		 2148 err = md_allow_write(conf->mddev);
2149 if (err)
2150 return err;
2a2275d6 2151
ad01c9e3
N		 2152 /* Step 1 */
2153 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
2154 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
20c2df83 2155 0, 0, NULL);
ad01c9e3
N		 2156 if (!sc)
2157 return -ENOMEM;
2158
2d5b569b
N		 2159 /* Need to ensure auto-resizing doesn't interfere */
2160 mutex_lock(&conf->cache_size_mutex);
2161
ad01c9e3 2162 for (i = conf->max_nr_stripes; i; i--) {
f18c1a35 2163 nsh = alloc_stripe(sc, GFP_KERNEL);
ad01c9e3
N		 2164 if (!nsh)
2165 break;
2166
ad01c9e3 2167 nsh->raid_conf = conf;
ad01c9e3
N		 2168 list_add(&nsh->lru, &newstripes);
2169 }
2170 if (i) {
2171 /* didn't get enough, give up */
2172 while (!list_empty(&newstripes)) {
2173 nsh = list_entry(newstripes.next, struct stripe_head, lru);
2174 list_del(&nsh->lru);
2175 kmem_cache_free(sc, nsh);
2176 }
2177 kmem_cache_destroy(sc);
2d5b569b 2178 mutex_unlock(&conf->cache_size_mutex);
ad01c9e3
N		 2179 return -ENOMEM;
2180 }
2181 /* Step 2 - Must use GFP_NOIO now.
2182 * OK, we have enough stripes, start collecting inactive
2183 * stripes and copying them over
2184 */
566c09c5
SL		 2185 hash = 0;
2186 cnt = 0;
ad01c9e3 2187 list_for_each_entry(nsh, &newstripes, lru) {
566c09c5 2188 lock_device_hash_lock(conf, hash);
e9e4c377 2189 wait_event_exclusive_cmd(conf->wait_for_stripe[hash],
566c09c5
SL		 2190 !list_empty(conf->inactive_list + hash),
2191 unlock_device_hash_lock(conf, hash),
2192 lock_device_hash_lock(conf, hash));
2193 osh = get_free_stripe(conf, hash);
2194 unlock_device_hash_lock(conf, hash);
f18c1a35 2195
d592a996 2196 for(i=0; i<conf->pool_size; i++) {
ad01c9e3 2197 nsh->dev[i].page = osh->dev[i].page;
d592a996
SL		 2198 nsh->dev[i].orig_page = osh->dev[i].page;
2199 }
566c09c5 2200 nsh->hash_lock_index = hash;
ad01c9e3 2201 kmem_cache_free(conf->slab_cache, osh);
566c09c5
SL		 2202 cnt++;
2203 if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS +
2204 !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) {
2205 hash++;
2206 cnt = 0;
2207 }
ad01c9e3
N		 2208 }
2209 kmem_cache_destroy(conf->slab_cache);
2210
2211 /* Step 3.
2212 * At this point, we are holding all the stripes so the array
2213 * is completely stalled, so now is a good time to resize
d6f38f31 2214 * conf->disks and the scribble region
ad01c9e3
N		 2215 */
2216 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
2217 if (ndisks) {
2218 for (i=0; i<conf->raid_disks; i++)
2219 ndisks[i] = conf->disks[i];
2220 kfree(conf->disks);
2221 conf->disks = ndisks;
2222 } else
2223 err = -ENOMEM;
2224
2d5b569b 2225 mutex_unlock(&conf->cache_size_mutex);
ad01c9e3
N		 2226 /* Step 4, return new stripes to service */
2227 while(!list_empty(&newstripes)) {
2228 nsh = list_entry(newstripes.next, struct stripe_head, lru);
2229 list_del_init(&nsh->lru);
d6f38f31 2230
ad01c9e3
N		 2231 for (i=conf->raid_disks; i < newsize; i++)
2232 if (nsh->dev[i].page == NULL) {
2233 struct page *p = alloc_page(GFP_NOIO);
2234 nsh->dev[i].page = p;
d592a996 2235 nsh->dev[i].orig_page = p;
ad01c9e3
N		 2236 if (!p)
2237 err = -ENOMEM;
2238 }
2239 release_stripe(nsh);
2240 }
2241 /* critical section pass, GFP_NOIO no longer needed */
2242
2243 conf->slab_cache = sc;
2244 conf->active_name = 1-conf->active_name;
6e9eac2d
N		 2245 if (!err)
2246 conf->pool_size = newsize;
ad01c9e3
N		 2247 return err;
2248}
1da177e4 2249
486f0644 2250static int drop_one_stripe(struct r5conf *conf)
1da177e4
LT		 2251{
2252 struct stripe_head *sh;
49895bcc 2253 int hash = (conf->max_nr_stripes - 1) & STRIPE_HASH_LOCKS_MASK;
1da177e4 2254
566c09c5
SL		 2255 spin_lock_irq(conf->hash_locks + hash);
2256 sh = get_free_stripe(conf, hash);
2257 spin_unlock_irq(conf->hash_locks + hash);
3f294f4f
N		 2258 if (!sh)
2259 return 0;
78bafebd 2260 BUG_ON(atomic_read(&sh->count));
e4e11e38 2261 shrink_buffers(sh);
3f294f4f
N		 2262 kmem_cache_free(conf->slab_cache, sh);
2263 atomic_dec(&conf->active_stripes);
486f0644 2264 conf->max_nr_stripes--;
3f294f4f
N		 2265 return 1;
2266}
2267
d1688a6d 2268static void shrink_stripes(struct r5conf *conf)
3f294f4f 2269{
486f0644
N		 2270 while (conf->max_nr_stripes &&
2271 drop_one_stripe(conf))
2272 ;
3f294f4f 2273
29fc7e3e
N		 2274 if (conf->slab_cache)
2275 kmem_cache_destroy(conf->slab_cache);
1da177e4
LT		 2276 conf->slab_cache = NULL;
2277}
2278
4246a0b6 2279static void raid5_end_read_request(struct bio * bi)
1da177e4 2280{
99c0fb5f 2281 struct stripe_head *sh = bi->bi_private;
d1688a6d 2282 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 2283 int disks = sh->disks, i;
d6950432 2284 char b[BDEVNAME_SIZE];
dd054fce 2285 struct md_rdev *rdev = NULL;
05616be5 2286 sector_t s;
1da177e4
LT		 2287
2288 for (i=0 ; i<disks; i++)
2289 if (bi == &sh->dev[i].req)
2290 break;
2291
4246a0b6 2292 pr_debug("end_read_request %llu/%d, count: %d, error %d.\n",
45b4233c 2293 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
4246a0b6 2294 bi->bi_error);
1da177e4
LT		 2295 if (i == disks) {
2296 BUG();
6712ecf8 2297 return;
1da177e4 2298 }
14a75d3e 2299 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
dd054fce
N		 2300 /* If replacement finished while this request was outstanding,
2301 * 'replacement' might be NULL already.
2302 * In that case it moved down to 'rdev'.
2303 * rdev is not removed until all requests are finished.
2304 */
14a75d3e 2305 rdev = conf->disks[i].replacement;
dd054fce 2306 if (!rdev)
14a75d3e 2307 rdev = conf->disks[i].rdev;
1da177e4 2308
05616be5
N		 2309 if (use_new_offset(conf, sh))
2310 s = sh->sector + rdev->new_data_offset;
2311 else
2312 s = sh->sector + rdev->data_offset;
4246a0b6 2313 if (!bi->bi_error) {
1da177e4 2314 set_bit(R5_UPTODATE, &sh->dev[i].flags);
4e5314b5 2315 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
14a75d3e
N		 2316 /* Note that this cannot happen on a
2317 * replacement device. We just fail those on
2318 * any error
2319 */
8bda470e
CD		 2320 printk_ratelimited(
2321 KERN_INFO
2322 "md/raid:%s: read error corrected"
2323 " (%lu sectors at %llu on %s)\n",
2324 mdname(conf->mddev), STRIPE_SECTORS,
05616be5 2325 (unsigned long long)s,
8bda470e 2326 bdevname(rdev->bdev, b));
ddd5115f 2327 atomic_add(STRIPE_SECTORS, &rdev->corrected_errors);
4e5314b5
N		 2328 clear_bit(R5_ReadError, &sh->dev[i].flags);
2329 clear_bit(R5_ReWrite, &sh->dev[i].flags);
3f9e7c14 2330 } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2331 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2332
14a75d3e
N		 2333 if (atomic_read(&rdev->read_errors))
2334 atomic_set(&rdev->read_errors, 0);
1da177e4 2335 } else {
14a75d3e 2336 const char *bdn = bdevname(rdev->bdev, b);
ba22dcbf 2337 int retry = 0;
2e8ac303 2338 int set_bad = 0;
d6950432 2339
1da177e4 2340 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
d6950432 2341 atomic_inc(&rdev->read_errors);
14a75d3e
N		 2342 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
2343 printk_ratelimited(
2344 KERN_WARNING
2345 "md/raid:%s: read error on replacement device "
2346 "(sector %llu on %s).\n",
2347 mdname(conf->mddev),
05616be5 2348 (unsigned long long)s,
14a75d3e 2349 bdn);
2e8ac303 2350 else if (conf->mddev->degraded >= conf->max_degraded) {
2351 set_bad = 1;
8bda470e
CD		 2352 printk_ratelimited(
2353 KERN_WARNING
2354 "md/raid:%s: read error not correctable "
2355 "(sector %llu on %s).\n",
2356 mdname(conf->mddev),
05616be5 2357 (unsigned long long)s,
8bda470e 2358 bdn);
2e8ac303 2359 } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) {
4e5314b5 2360 /* Oh, no!!! */
2e8ac303 2361 set_bad = 1;
8bda470e
CD		 2362 printk_ratelimited(
2363 KERN_WARNING
2364 "md/raid:%s: read error NOT corrected!! "
2365 "(sector %llu on %s).\n",
2366 mdname(conf->mddev),
05616be5 2367 (unsigned long long)s,
8bda470e 2368 bdn);
2e8ac303 2369 } else if (atomic_read(&rdev->read_errors)
ba22dcbf 2370 > conf->max_nr_stripes)
14f8d26b 2371 printk(KERN_WARNING
0c55e022 2372 "md/raid:%s: Too many read errors, failing device %s.\n",
d6950432 2373 mdname(conf->mddev), bdn);
ba22dcbf
N		 2374 else
2375 retry = 1;
edfa1f65
BY		 2376 if (set_bad && test_bit(In_sync, &rdev->flags)
2377 && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2378 retry = 1;
ba22dcbf 2379 if (retry)
3f9e7c14 2380 if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) {
2381 set_bit(R5_ReadError, &sh->dev[i].flags);
2382 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2383 } else
2384 set_bit(R5_ReadNoMerge, &sh->dev[i].flags);
ba22dcbf 2385 else {
4e5314b5
N		 2386 clear_bit(R5_ReadError, &sh->dev[i].flags);
2387 clear_bit(R5_ReWrite, &sh->dev[i].flags);
2e8ac303 2388 if (!(set_bad
2389 && test_bit(In_sync, &rdev->flags)
2390 && rdev_set_badblocks(
2391 rdev, sh->sector, STRIPE_SECTORS, 0)))
2392 md_error(conf->mddev, rdev);
ba22dcbf 2393 }
1da177e4 2394 }
14a75d3e 2395 rdev_dec_pending(rdev, conf->mddev);
1da177e4
LT		 2396 clear_bit(R5_LOCKED, &sh->dev[i].flags);
2397 set_bit(STRIPE_HANDLE, &sh->state);
2398 release_stripe(sh);
1da177e4
LT		 2399}
2400
4246a0b6 2401static void raid5_end_write_request(struct bio *bi)
1da177e4 2402{
99c0fb5f 2403 struct stripe_head *sh = bi->bi_private;
d1688a6d 2404 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 2405 int disks = sh->disks, i;
977df362 2406 struct md_rdev *uninitialized_var(rdev);
b84db560
N		 2407 sector_t first_bad;
2408 int bad_sectors;
977df362 2409 int replacement = 0;
1da177e4 2410
977df362
N		 2411 for (i = 0 ; i < disks; i++) {
2412 if (bi == &sh->dev[i].req) {
2413 rdev = conf->disks[i].rdev;
1da177e4 2414 break;
977df362
N		 2415 }
2416 if (bi == &sh->dev[i].rreq) {
2417 rdev = conf->disks[i].replacement;
dd054fce
N		 2418 if (rdev)
2419 replacement = 1;
2420 else
2421 /* rdev was removed and 'replacement'
2422 * replaced it. rdev is not removed
2423 * until all requests are finished.
2424 */
2425 rdev = conf->disks[i].rdev;
977df362
N		 2426 break;
2427 }
2428 }
4246a0b6 2429 pr_debug("end_write_request %llu/%d, count %d, error: %d.\n",
1da177e4 2430 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
4246a0b6 2431 bi->bi_error);
1da177e4
LT		 2432 if (i == disks) {
2433 BUG();
6712ecf8 2434 return;
1da177e4
LT		 2435 }
2436
977df362 2437 if (replacement) {
4246a0b6 2438 if (bi->bi_error)
977df362
N		 2439 md_error(conf->mddev, rdev);
2440 else if (is_badblock(rdev, sh->sector,
2441 STRIPE_SECTORS,
2442 &first_bad, &bad_sectors))
2443 set_bit(R5_MadeGoodRepl, &sh->dev[i].flags);
2444 } else {
4246a0b6 2445 if (bi->bi_error) {
9f97e4b1 2446 set_bit(STRIPE_DEGRADED, &sh->state);
977df362
N		 2447 set_bit(WriteErrorSeen, &rdev->flags);
2448 set_bit(R5_WriteError, &sh->dev[i].flags);
3a6de292
N		 2449 if (!test_and_set_bit(WantReplacement, &rdev->flags))
2450 set_bit(MD_RECOVERY_NEEDED,
2451 &rdev->mddev->recovery);
977df362
N		 2452 } else if (is_badblock(rdev, sh->sector,
2453 STRIPE_SECTORS,
c0b32972 2454 &first_bad, &bad_sectors)) {
977df362 2455 set_bit(R5_MadeGood, &sh->dev[i].flags);
c0b32972
N		 2456 if (test_bit(R5_ReadError, &sh->dev[i].flags))
2457 /* That was a successful write so make
2458 * sure it looks like we already did
2459 * a re-write.
2460 */
2461 set_bit(R5_ReWrite, &sh->dev[i].flags);
2462 }
977df362
N		 2463 }
2464 rdev_dec_pending(rdev, conf->mddev);
1da177e4 2465
4246a0b6 2466 if (sh->batch_head && bi->bi_error && !replacement)
72ac7330 2467 set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state);
2468
977df362
N		 2469 if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags))
2470 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1da177e4 2471 set_bit(STRIPE_HANDLE, &sh->state);
c04be0aa 2472 release_stripe(sh);
59fc630b 2473
2474 if (sh->batch_head && sh != sh->batch_head)
2475 release_stripe(sh->batch_head);
1da177e4
LT		 2476}
2477
784052ec 2478static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous);
d592a996 2479
784052ec 2480static void raid5_build_block(struct stripe_head *sh, int i, int previous)
1da177e4
LT		 2481{
2482 struct r5dev *dev = &sh->dev[i];
2483
2484 bio_init(&dev->req);
2485 dev->req.bi_io_vec = &dev->vec;
d592a996 2486 dev->req.bi_max_vecs = 1;
1da177e4
LT		 2487 dev->req.bi_private = sh;
2488
977df362
N		 2489 bio_init(&dev->rreq);
2490 dev->rreq.bi_io_vec = &dev->rvec;
d592a996 2491 dev->rreq.bi_max_vecs = 1;
977df362 2492 dev->rreq.bi_private = sh;
977df362 2493
1da177e4 2494 dev->flags = 0;
784052ec 2495 dev->sector = compute_blocknr(sh, i, previous);
1da177e4
LT		 2496}
2497
fd01b88c 2498static void error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 2499{
2500 char b[BDEVNAME_SIZE];
d1688a6d 2501 struct r5conf *conf = mddev->private;
908f4fbd 2502 unsigned long flags;
0c55e022 2503 pr_debug("raid456: error called\n");
1da177e4 2504
908f4fbd
N		 2505 spin_lock_irqsave(&conf->device_lock, flags);
2506 clear_bit(In_sync, &rdev->flags);
2507 mddev->degraded = calc_degraded(conf);
2508 spin_unlock_irqrestore(&conf->device_lock, flags);
2509 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2510
de393cde 2511 set_bit(Blocked, &rdev->flags);
6f8d0c77
N		 2512 set_bit(Faulty, &rdev->flags);
2513 set_bit(MD_CHANGE_DEVS, &mddev->flags);
c3cce6cd 2514 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6f8d0c77
N		 2515 printk(KERN_ALERT
2516 "md/raid:%s: Disk failure on %s, disabling device.\n"
2517 "md/raid:%s: Operation continuing on %d devices.\n",
2518 mdname(mddev),
2519 bdevname(rdev->bdev, b),
2520 mdname(mddev),
2521 conf->raid_disks - mddev->degraded);
16a53ecc 2522}
1da177e4
LT		 2523
2524/*
2525 * Input: a 'big' sector number,
2526 * Output: index of the data and parity disk, and the sector # in them.
2527 */
d1688a6d 2528static sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector,
911d4ee8
N		 2529 int previous, int *dd_idx,
2530 struct stripe_head *sh)
1da177e4 2531{
6e3b96ed 2532 sector_t stripe, stripe2;
35f2a591 2533 sector_t chunk_number;
1da177e4 2534 unsigned int chunk_offset;
911d4ee8 2535 int pd_idx, qd_idx;
67cc2b81 2536 int ddf_layout = 0;
1da177e4 2537 sector_t new_sector;
e183eaed
N		 2538 int algorithm = previous ? conf->prev_algo
2539 : conf->algorithm;
09c9e5fa
AN		 2540 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2541 : conf->chunk_sectors;
112bf897
N		 2542 int raid_disks = previous ? conf->previous_raid_disks
2543 : conf->raid_disks;
2544 int data_disks = raid_disks - conf->max_degraded;
1da177e4
LT		 2545
2546 /* First compute the information on this sector */
2547
2548 /*
2549 * Compute the chunk number and the sector offset inside the chunk
2550 */
2551 chunk_offset = sector_div(r_sector, sectors_per_chunk);
2552 chunk_number = r_sector;
1da177e4
LT		 2553
2554 /*
2555 * Compute the stripe number
2556 */
35f2a591
N		 2557 stripe = chunk_number;
2558 *dd_idx = sector_div(stripe, data_disks);
6e3b96ed 2559 stripe2 = stripe;
1da177e4
LT		 2560 /*
2561 * Select the parity disk based on the user selected algorithm.
2562 */
84789554 2563 pd_idx = qd_idx = -1;
16a53ecc
N		 2564 switch(conf->level) {
2565 case 4:
911d4ee8 2566 pd_idx = data_disks;
16a53ecc
N		 2567 break;
2568 case 5:
e183eaed 2569 switch (algorithm) {
1da177e4 2570 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2571 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2572 if (*dd_idx >= pd_idx)
1da177e4
LT		 2573 (*dd_idx)++;
2574 break;
2575 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2576 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2577 if (*dd_idx >= pd_idx)
1da177e4
LT		 2578 (*dd_idx)++;
2579 break;
2580 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2581 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2582 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4
LT		 2583 break;
2584 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2585 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2586 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4 2587 break;
99c0fb5f
N		 2588 case ALGORITHM_PARITY_0:
2589 pd_idx = 0;
2590 (*dd_idx)++;
2591 break;
2592 case ALGORITHM_PARITY_N:
2593 pd_idx = data_disks;
2594 break;
1da177e4 2595 default:
99c0fb5f 2596 BUG();
16a53ecc
N		 2597 }
2598 break;
2599 case 6:
2600
e183eaed 2601 switch (algorithm) {
16a53ecc 2602 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2603 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2604 qd_idx = pd_idx + 1;
2605 if (pd_idx == raid_disks-1) {
99c0fb5f 2606 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2607 qd_idx = 0;
2608 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2609 (*dd_idx) += 2; /* D D P Q D */
2610 break;
2611 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2612 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2613 qd_idx = pd_idx + 1;
2614 if (pd_idx == raid_disks-1) {
99c0fb5f 2615 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2616 qd_idx = 0;
2617 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2618 (*dd_idx) += 2; /* D D P Q D */
2619 break;
2620 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2621 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2622 qd_idx = (pd_idx + 1) % raid_disks;
2623 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc
N		 2624 break;
2625 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2626 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2627 qd_idx = (pd_idx + 1) % raid_disks;
2628 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc 2629 break;
99c0fb5f
N		 2630
2631 case ALGORITHM_PARITY_0:
2632 pd_idx = 0;
2633 qd_idx = 1;
2634 (*dd_idx) += 2;
2635 break;
2636 case ALGORITHM_PARITY_N:
2637 pd_idx = data_disks;
2638 qd_idx = data_disks + 1;
2639 break;
2640
2641 case ALGORITHM_ROTATING_ZERO_RESTART:
2642 /* Exactly the same as RIGHT_ASYMMETRIC, but or
2643 * of blocks for computing Q is different.
2644 */
6e3b96ed 2645 pd_idx = sector_div(stripe2, raid_disks);
99c0fb5f
N		 2646 qd_idx = pd_idx + 1;
2647 if (pd_idx == raid_disks-1) {
2648 (*dd_idx)++; /* Q D D D P */
2649 qd_idx = 0;
2650 } else if (*dd_idx >= pd_idx)
2651 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2652 ddf_layout = 1;
99c0fb5f
N		 2653 break;
2654
2655 case ALGORITHM_ROTATING_N_RESTART:
2656 /* Same a left_asymmetric, by first stripe is
2657 * D D D P Q rather than
2658 * Q D D D P
2659 */
6e3b96ed
N		 2660 stripe2 += 1;
2661 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2662 qd_idx = pd_idx + 1;
2663 if (pd_idx == raid_disks-1) {
2664 (*dd_idx)++; /* Q D D D P */
2665 qd_idx = 0;
2666 } else if (*dd_idx >= pd_idx)
2667 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2668 ddf_layout = 1;
99c0fb5f
N		 2669 break;
2670
2671 case ALGORITHM_ROTATING_N_CONTINUE:
2672 /* Same as left_symmetric but Q is before P */
6e3b96ed 2673 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2674 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
2675 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
67cc2b81 2676 ddf_layout = 1;
99c0fb5f
N		 2677 break;
2678
2679 case ALGORITHM_LEFT_ASYMMETRIC_6:
2680 /* RAID5 left_asymmetric, with Q on last device */
6e3b96ed 2681 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2682 if (*dd_idx >= pd_idx)
2683 (*dd_idx)++;
2684 qd_idx = raid_disks - 1;
2685 break;
2686
2687 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6e3b96ed 2688 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2689 if (*dd_idx >= pd_idx)
2690 (*dd_idx)++;
2691 qd_idx = raid_disks - 1;
2692 break;
2693
2694 case ALGORITHM_LEFT_SYMMETRIC_6:
6e3b96ed 2695 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2696 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2697 qd_idx = raid_disks - 1;
2698 break;
2699
2700 case ALGORITHM_RIGHT_SYMMETRIC_6:
6e3b96ed 2701 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2702 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2703 qd_idx = raid_disks - 1;
2704 break;
2705
2706 case ALGORITHM_PARITY_0_6:
2707 pd_idx = 0;
2708 (*dd_idx)++;
2709 qd_idx = raid_disks - 1;
2710 break;
2711
16a53ecc 2712 default:
99c0fb5f 2713 BUG();
16a53ecc
N		 2714 }
2715 break;
1da177e4
LT		 2716 }
2717
911d4ee8
N		 2718 if (sh) {
2719 sh->pd_idx = pd_idx;
2720 sh->qd_idx = qd_idx;
67cc2b81 2721 sh->ddf_layout = ddf_layout;
911d4ee8 2722 }
1da177e4
LT		 2723 /*
2724 * Finally, compute the new sector number
2725 */
2726 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
2727 return new_sector;
2728}
2729
784052ec 2730static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous)
1da177e4 2731{
d1688a6d 2732 struct r5conf *conf = sh->raid_conf;
b875e531
N		 2733 int raid_disks = sh->disks;
2734 int data_disks = raid_disks - conf->max_degraded;
1da177e4 2735 sector_t new_sector = sh->sector, check;
09c9e5fa
AN		 2736 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2737 : conf->chunk_sectors;
e183eaed
N		 2738 int algorithm = previous ? conf->prev_algo
2739 : conf->algorithm;
1da177e4
LT		 2740 sector_t stripe;
2741 int chunk_offset;
35f2a591
N		 2742 sector_t chunk_number;
2743 int dummy1, dd_idx = i;
1da177e4 2744 sector_t r_sector;
911d4ee8 2745 struct stripe_head sh2;
1da177e4
LT		 2746
2747 chunk_offset = sector_div(new_sector, sectors_per_chunk);
2748 stripe = new_sector;
1da177e4 2749
16a53ecc
N		 2750 if (i == sh->pd_idx)
2751 return 0;
2752 switch(conf->level) {
2753 case 4: break;
2754 case 5:
e183eaed 2755 switch (algorithm) {
1da177e4
LT		 2756 case ALGORITHM_LEFT_ASYMMETRIC:
2757 case ALGORITHM_RIGHT_ASYMMETRIC:
2758 if (i > sh->pd_idx)
2759 i--;
2760 break;
2761 case ALGORITHM_LEFT_SYMMETRIC:
2762 case ALGORITHM_RIGHT_SYMMETRIC:
2763 if (i < sh->pd_idx)
2764 i += raid_disks;
2765 i -= (sh->pd_idx + 1);
2766 break;
99c0fb5f
N		 2767 case ALGORITHM_PARITY_0:
2768 i -= 1;
2769 break;
2770 case ALGORITHM_PARITY_N:
2771 break;
1da177e4 2772 default:
99c0fb5f 2773 BUG();
16a53ecc
N		 2774 }
2775 break;
2776 case 6:
d0dabf7e 2777 if (i == sh->qd_idx)
16a53ecc 2778 return 0; /* It is the Q disk */
e183eaed 2779 switch (algorithm) {
16a53ecc
N		 2780 case ALGORITHM_LEFT_ASYMMETRIC:
2781 case ALGORITHM_RIGHT_ASYMMETRIC:
99c0fb5f
N		 2782 case ALGORITHM_ROTATING_ZERO_RESTART:
2783 case ALGORITHM_ROTATING_N_RESTART:
2784 if (sh->pd_idx == raid_disks-1)
2785 i--; /* Q D D D P */
16a53ecc
N		 2786 else if (i > sh->pd_idx)
2787 i -= 2; /* D D P Q D */
2788 break;
2789 case ALGORITHM_LEFT_SYMMETRIC:
2790 case ALGORITHM_RIGHT_SYMMETRIC:
2791 if (sh->pd_idx == raid_disks-1)
2792 i--; /* Q D D D P */
2793 else {
2794 /* D D P Q D */
2795 if (i < sh->pd_idx)
2796 i += raid_disks;
2797 i -= (sh->pd_idx + 2);
2798 }
2799 break;
99c0fb5f
N		 2800 case ALGORITHM_PARITY_0:
2801 i -= 2;
2802 break;
2803 case ALGORITHM_PARITY_N:
2804 break;
2805 case ALGORITHM_ROTATING_N_CONTINUE:
e4424fee 2806 /* Like left_symmetric, but P is before Q */
99c0fb5f
N		 2807 if (sh->pd_idx == 0)
2808 i--; /* P D D D Q */
e4424fee
N		 2809 else {
2810 /* D D Q P D */
2811 if (i < sh->pd_idx)
2812 i += raid_disks;
2813 i -= (sh->pd_idx + 1);
2814 }
99c0fb5f
N		 2815 break;
2816 case ALGORITHM_LEFT_ASYMMETRIC_6:
2817 case ALGORITHM_RIGHT_ASYMMETRIC_6:
2818 if (i > sh->pd_idx)
2819 i--;
2820 break;
2821 case ALGORITHM_LEFT_SYMMETRIC_6:
2822 case ALGORITHM_RIGHT_SYMMETRIC_6:
2823 if (i < sh->pd_idx)
2824 i += data_disks + 1;
2825 i -= (sh->pd_idx + 1);
2826 break;
2827 case ALGORITHM_PARITY_0_6:
2828 i -= 1;
2829 break;
16a53ecc 2830 default:
99c0fb5f 2831 BUG();
16a53ecc
N		 2832 }
2833 break;
1da177e4
LT		 2834 }
2835
2836 chunk_number = stripe * data_disks + i;
35f2a591 2837 r_sector = chunk_number * sectors_per_chunk + chunk_offset;
1da177e4 2838
112bf897 2839 check = raid5_compute_sector(conf, r_sector,
784052ec 2840 previous, &dummy1, &sh2);
911d4ee8
N		 2841 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
2842 || sh2.qd_idx != sh->qd_idx) {
0c55e022
N		 2843 printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n",
2844 mdname(conf->mddev));
1da177e4
LT		 2845 return 0;
2846 }
2847 return r_sector;
2848}
2849
600aa109 2850static void
c0f7bddb 2851schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
600aa109 2852 int rcw, int expand)
e33129d8 2853{
584acdd4 2854 int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx, disks = sh->disks;
d1688a6d 2855 struct r5conf *conf = sh->raid_conf;
c0f7bddb 2856 int level = conf->level;
e33129d8
DW		 2857
2858 if (rcw) {
e33129d8
DW		 2859
2860 for (i = disks; i--; ) {
2861 struct r5dev *dev = &sh->dev[i];
2862
2863 if (dev->towrite) {
2864 set_bit(R5_LOCKED, &dev->flags);
d8ee0728 2865 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2866 if (!expand)
2867 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2868 s->locked++;
e33129d8
DW		 2869 }
2870 }
ce7d363a
N		 2871 /* if we are not expanding this is a proper write request, and
2872 * there will be bios with new data to be drained into the
2873 * stripe cache
2874 */
2875 if (!expand) {
2876 if (!s->locked)
2877 /* False alarm, nothing to do */
2878 return;
2879 sh->reconstruct_state = reconstruct_state_drain_run;
2880 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2881 } else
2882 sh->reconstruct_state = reconstruct_state_run;
2883
2884 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
2885
c0f7bddb 2886 if (s->locked + conf->max_degraded == disks)
8b3e6cdc 2887 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
c0f7bddb 2888 atomic_inc(&conf->pending_full_writes);
e33129d8
DW		 2889 } else {
2890 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
2891 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
584acdd4
MS		 2892 BUG_ON(level == 6 &&
2893 (!(test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags) ||
2894 test_bit(R5_Wantcompute, &sh->dev[qd_idx].flags))));
e33129d8 2895
e33129d8
DW		 2896 for (i = disks; i--; ) {
2897 struct r5dev *dev = &sh->dev[i];
584acdd4 2898 if (i == pd_idx || i == qd_idx)
e33129d8
DW		 2899 continue;
2900
e33129d8
DW		 2901 if (dev->towrite &&
2902 (test_bit(R5_UPTODATE, &dev->flags) ||
d8ee0728
DW		 2903 test_bit(R5_Wantcompute, &dev->flags))) {
2904 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2905 set_bit(R5_LOCKED, &dev->flags);
2906 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2907 s->locked++;
e33129d8
DW		 2908 }
2909 }
ce7d363a
N		 2910 if (!s->locked)
2911 /* False alarm - nothing to do */
2912 return;
2913 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
2914 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
2915 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2916 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
e33129d8
DW		 2917 }
2918
c0f7bddb 2919 /* keep the parity disk(s) locked while asynchronous operations
e33129d8
DW		 2920 * are in flight
2921 */
2922 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
2923 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
600aa109 2924 s->locked++;
e33129d8 2925
c0f7bddb
YT		 2926 if (level == 6) {
2927 int qd_idx = sh->qd_idx;
2928 struct r5dev *dev = &sh->dev[qd_idx];
2929
2930 set_bit(R5_LOCKED, &dev->flags);
2931 clear_bit(R5_UPTODATE, &dev->flags);
2932 s->locked++;
2933 }
2934
600aa109 2935 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
e46b272b 2936 __func__, (unsigned long long)sh->sector,
600aa109 2937 s->locked, s->ops_request);
e33129d8 2938}
16a53ecc 2939
1da177e4
LT		 2940/*
2941 * Each stripe/dev can have one or more bion attached.
16a53ecc 2942 * toread/towrite point to the first in a chain.
1da177e4
LT		 2943 * The bi_next chain must be in order.
2944 */
da41ba65 2945static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx,
2946 int forwrite, int previous)
1da177e4
LT		 2947{
2948 struct bio **bip;
d1688a6d 2949 struct r5conf *conf = sh->raid_conf;
72626685 2950 int firstwrite=0;
1da177e4 2951
cbe47ec5 2952 pr_debug("adding bi b#%llu to stripe s#%llu\n",
4f024f37 2953 (unsigned long long)bi->bi_iter.bi_sector,
1da177e4
LT		 2954 (unsigned long long)sh->sector);
2955
b17459c0
SL		 2956 /*
2957 * If several bio share a stripe. The bio bi_phys_segments acts as a
2958 * reference count to avoid race. The reference count should already be
2959 * increased before this function is called (for example, in
2960 * make_request()), so other bio sharing this stripe will not free the
2961 * stripe. If a stripe is owned by one stripe, the stripe lock will
2962 * protect it.
2963 */
2964 spin_lock_irq(&sh->stripe_lock);
59fc630b 2965 /* Don't allow new IO added to stripes in batch list */
2966 if (sh->batch_head)
2967 goto overlap;
72626685 2968 if (forwrite) {
1da177e4 2969 bip = &sh->dev[dd_idx].towrite;
7eaf7e8e 2970 if (*bip == NULL)
72626685
N		 2971 firstwrite = 1;
2972 } else
1da177e4 2973 bip = &sh->dev[dd_idx].toread;
4f024f37
KO		 2974 while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) {
2975 if (bio_end_sector(*bip) > bi->bi_iter.bi_sector)
1da177e4
LT		 2976 goto overlap;
2977 bip = & (*bip)->bi_next;
2978 }
4f024f37 2979 if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi))
1da177e4
LT		 2980 goto overlap;
2981
da41ba65 2982 if (!forwrite || previous)
2983 clear_bit(STRIPE_BATCH_READY, &sh->state);
2984
78bafebd 2985 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
1da177e4
LT		 2986 if (*bip)
2987 bi->bi_next = *bip;
2988 *bip = bi;
e7836bd6 2989 raid5_inc_bi_active_stripes(bi);
72626685 2990
1da177e4
LT		 2991 if (forwrite) {
2992 /* check if page is covered */
2993 sector_t sector = sh->dev[dd_idx].sector;
2994 for (bi=sh->dev[dd_idx].towrite;
2995 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
4f024f37 2996 bi && bi->bi_iter.bi_sector <= sector;
1da177e4 2997 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
f73a1c7d
KO		 2998 if (bio_end_sector(bi) >= sector)
2999 sector = bio_end_sector(bi);
1da177e4
LT		 3000 }
3001 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
7a87f434 3002 if (!test_and_set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags))
3003 sh->overwrite_disks++;
1da177e4 3004 }
cbe47ec5
N		 3005
3006 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
4f024f37 3007 (unsigned long long)(*bip)->bi_iter.bi_sector,
cbe47ec5
N		 3008 (unsigned long long)sh->sector, dd_idx);
3009
3010 if (conf->mddev->bitmap && firstwrite) {
d0852df5
N		 3011 /* Cannot hold spinlock over bitmap_startwrite,
3012 * but must ensure this isn't added to a batch until
3013 * we have added to the bitmap and set bm_seq.
3014 * So set STRIPE_BITMAP_PENDING to prevent
3015 * batching.
3016 * If multiple add_stripe_bio() calls race here they
3017 * much all set STRIPE_BITMAP_PENDING. So only the first one
3018 * to complete "bitmap_startwrite" gets to set
3019 * STRIPE_BIT_DELAY. This is important as once a stripe
3020 * is added to a batch, STRIPE_BIT_DELAY cannot be changed
3021 * any more.
3022 */
3023 set_bit(STRIPE_BITMAP_PENDING, &sh->state);
3024 spin_unlock_irq(&sh->stripe_lock);
cbe47ec5
N		 3025 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
3026 STRIPE_SECTORS, 0);
d0852df5
N		 3027 spin_lock_irq(&sh->stripe_lock);
3028 clear_bit(STRIPE_BITMAP_PENDING, &sh->state);
3029 if (!sh->batch_head) {
3030 sh->bm_seq = conf->seq_flush+1;
3031 set_bit(STRIPE_BIT_DELAY, &sh->state);
3032 }
cbe47ec5 3033 }
d0852df5 3034 spin_unlock_irq(&sh->stripe_lock);
59fc630b 3035
3036 if (stripe_can_batch(sh))
3037 stripe_add_to_batch_list(conf, sh);
1da177e4
LT		 3038 return 1;
3039
3040 overlap:
3041 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
b17459c0 3042 spin_unlock_irq(&sh->stripe_lock);
1da177e4
LT		 3043 return 0;
3044}
3045
d1688a6d 3046static void end_reshape(struct r5conf *conf);
29269553 3047
d1688a6d 3048static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
911d4ee8 3049 struct stripe_head *sh)
ccfcc3c1 3050{
784052ec 3051 int sectors_per_chunk =
09c9e5fa 3052 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
911d4ee8 3053 int dd_idx;
2d2063ce 3054 int chunk_offset = sector_div(stripe, sectors_per_chunk);
112bf897 3055 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
2d2063ce 3056
112bf897
N		 3057 raid5_compute_sector(conf,
3058 stripe * (disks - conf->max_degraded)
b875e531 3059 *sectors_per_chunk + chunk_offset,
112bf897 3060 previous,
911d4ee8 3061 &dd_idx, sh);
ccfcc3c1
N		 3062}
3063
a4456856 3064static void
d1688a6d 3065handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh,
a4456856 3066 struct stripe_head_state *s, int disks,
34a6f80e 3067 struct bio_list *return_bi)
a4456856
DW		 3068{
3069 int i;
59fc630b 3070 BUG_ON(sh->batch_head);
a4456856
DW		 3071 for (i = disks; i--; ) {
3072 struct bio *bi;
3073 int bitmap_end = 0;
3074
3075 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
3cb03002 3076 struct md_rdev *rdev;
a4456856
DW		 3077 rcu_read_lock();
3078 rdev = rcu_dereference(conf->disks[i].rdev);
3079 if (rdev && test_bit(In_sync, &rdev->flags))
7f0da59b
N		 3080 atomic_inc(&rdev->nr_pending);
3081 else
3082 rdev = NULL;
a4456856 3083 rcu_read_unlock();
7f0da59b
N		 3084 if (rdev) {
3085 if (!rdev_set_badblocks(
3086 rdev,
3087 sh->sector,
3088 STRIPE_SECTORS, 0))
3089 md_error(conf->mddev, rdev);
3090 rdev_dec_pending(rdev, conf->mddev);
3091 }
a4456856 3092 }
b17459c0 3093 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 3094 /* fail all writes first */
3095 bi = sh->dev[i].towrite;
3096 sh->dev[i].towrite = NULL;
7a87f434 3097 sh->overwrite_disks = 0;
b17459c0 3098 spin_unlock_irq(&sh->stripe_lock);
1ed850f3 3099 if (bi)
a4456856 3100 bitmap_end = 1;
a4456856
DW		 3101
3102 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
3103 wake_up(&conf->wait_for_overlap);
3104
4f024f37 3105 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3106 sh->dev[i].sector + STRIPE_SECTORS) {
3107 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
4246a0b6
CH		 3108
3109 bi->bi_error = -EIO;
e7836bd6 3110 if (!raid5_dec_bi_active_stripes(bi)) {
a4456856 3111 md_write_end(conf->mddev);
34a6f80e 3112 bio_list_add(return_bi, bi);
a4456856
DW		 3113 }
3114 bi = nextbi;
3115 }
7eaf7e8e
SL		 3116 if (bitmap_end)
3117 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3118 STRIPE_SECTORS, 0, 0);
3119 bitmap_end = 0;
a4456856
DW		 3120 /* and fail all 'written' */
3121 bi = sh->dev[i].written;
3122 sh->dev[i].written = NULL;
d592a996
SL		 3123 if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) {
3124 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
3125 sh->dev[i].page = sh->dev[i].orig_page;
3126 }
3127
a4456856 3128 if (bi) bitmap_end = 1;
4f024f37 3129 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3130 sh->dev[i].sector + STRIPE_SECTORS) {
3131 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
4246a0b6
CH		 3132
3133 bi->bi_error = -EIO;
e7836bd6 3134 if (!raid5_dec_bi_active_stripes(bi)) {
a4456856 3135 md_write_end(conf->mddev);
34a6f80e 3136 bio_list_add(return_bi, bi);
a4456856
DW		 3137 }
3138 bi = bi2;
3139 }
3140
b5e98d65
DW		 3141 /* fail any reads if this device is non-operational and
3142 * the data has not reached the cache yet.
3143 */
3144 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
3145 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
3146 test_bit(R5_ReadError, &sh->dev[i].flags))) {
143c4d05 3147 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 3148 bi = sh->dev[i].toread;
3149 sh->dev[i].toread = NULL;
143c4d05 3150 spin_unlock_irq(&sh->stripe_lock);
a4456856
DW		 3151 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
3152 wake_up(&conf->wait_for_overlap);
ebda780b
SL		 3153 if (bi)
3154 s->to_read--;
4f024f37 3155 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 3156 sh->dev[i].sector + STRIPE_SECTORS) {
3157 struct bio *nextbi =
3158 r5_next_bio(bi, sh->dev[i].sector);
4246a0b6
CH		 3159
3160 bi->bi_error = -EIO;
34a6f80e
N		 3161 if (!raid5_dec_bi_active_stripes(bi))
3162 bio_list_add(return_bi, bi);
a4456856
DW		 3163 bi = nextbi;
3164 }
3165 }
a4456856
DW		 3166 if (bitmap_end)
3167 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3168 STRIPE_SECTORS, 0, 0);
8cfa7b0f
N		 3169 /* If we were in the middle of a write the parity block might
3170 * still be locked - so just clear all R5_LOCKED flags
3171 */
3172 clear_bit(R5_LOCKED, &sh->dev[i].flags);
a4456856 3173 }
ebda780b
SL		 3174 s->to_write = 0;
3175 s->written = 0;
a4456856 3176
8b3e6cdc
DW		 3177 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3178 if (atomic_dec_and_test(&conf->pending_full_writes))
3179 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 3180}
3181
7f0da59b 3182static void
d1688a6d 3183handle_failed_sync(struct r5conf *conf, struct stripe_head *sh,
7f0da59b
N		 3184 struct stripe_head_state *s)
3185{
3186 int abort = 0;
3187 int i;
3188
59fc630b 3189 BUG_ON(sh->batch_head);
7f0da59b 3190 clear_bit(STRIPE_SYNCING, &sh->state);
f8dfcffd
N		 3191 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
3192 wake_up(&conf->wait_for_overlap);
7f0da59b 3193 s->syncing = 0;
9a3e1101 3194 s->replacing = 0;
7f0da59b 3195 /* There is nothing more to do for sync/check/repair.
18b9837e
N		 3196 * Don't even need to abort as that is handled elsewhere
3197 * if needed, and not always wanted e.g. if there is a known
3198 * bad block here.
9a3e1101 3199 * For recover/replace we need to record a bad block on all
7f0da59b
N		 3200 * non-sync devices, or abort the recovery
3201 */
18b9837e
N		 3202 if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) {
3203 /* During recovery devices cannot be removed, so
3204 * locking and refcounting of rdevs is not needed
3205 */
3206 for (i = 0; i < conf->raid_disks; i++) {
3207 struct md_rdev *rdev = conf->disks[i].rdev;
3208 if (rdev
3209 && !test_bit(Faulty, &rdev->flags)
3210 && !test_bit(In_sync, &rdev->flags)
3211 && !rdev_set_badblocks(rdev, sh->sector,
3212 STRIPE_SECTORS, 0))
3213 abort = 1;
3214 rdev = conf->disks[i].replacement;
3215 if (rdev
3216 && !test_bit(Faulty, &rdev->flags)
3217 && !test_bit(In_sync, &rdev->flags)
3218 && !rdev_set_badblocks(rdev, sh->sector,
3219 STRIPE_SECTORS, 0))
3220 abort = 1;
3221 }
3222 if (abort)
3223 conf->recovery_disabled =
3224 conf->mddev->recovery_disabled;
7f0da59b 3225 }
18b9837e 3226 md_done_sync(conf->mddev, STRIPE_SECTORS, !abort);
7f0da59b
N		 3227}
3228
9a3e1101
N		 3229static int want_replace(struct stripe_head *sh, int disk_idx)
3230{
3231 struct md_rdev *rdev;
3232 int rv = 0;
3233 /* Doing recovery so rcu locking not required */
3234 rdev = sh->raid_conf->disks[disk_idx].replacement;
3235 if (rdev
3236 && !test_bit(Faulty, &rdev->flags)
3237 && !test_bit(In_sync, &rdev->flags)
3238 && (rdev->recovery_offset <= sh->sector
3239 || rdev->mddev->recovery_cp <= sh->sector))
3240 rv = 1;
3241
3242 return rv;
3243}
3244
93b3dbce 3245/* fetch_block - checks the given member device to see if its data needs
1fe797e6
DW		 3246 * to be read or computed to satisfy a request.
3247 *
3248 * Returns 1 when no more member devices need to be checked, otherwise returns
93b3dbce 3249 * 0 to tell the loop in handle_stripe_fill to continue
f38e1219 3250 */
2c58f06e
N		 3251
3252static int need_this_block(struct stripe_head *sh, struct stripe_head_state *s,
3253 int disk_idx, int disks)
a4456856 3254{
5599becc 3255 struct r5dev *dev = &sh->dev[disk_idx];
f2b3b44d
N		 3256 struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]],
3257 &sh->dev[s->failed_num[1]] };
ea664c82 3258 int i;
5599becc 3259
a79cfe12
N		 3260
3261 if (test_bit(R5_LOCKED, &dev->flags) ||
3262 test_bit(R5_UPTODATE, &dev->flags))
3263 /* No point reading this as we already have it or have
3264 * decided to get it.
3265 */
3266 return 0;
3267
3268 if (dev->toread ||
3269 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)))
3270 /* We need this block to directly satisfy a request */
3271 return 1;
3272
3273 if (s->syncing || s->expanding ||
3274 (s->replacing && want_replace(sh, disk_idx)))
3275 /* When syncing, or expanding we read everything.
3276 * When replacing, we need the replaced block.
3277 */
3278 return 1;
3279
3280 if ((s->failed >= 1 && fdev[0]->toread) ||
3281 (s->failed >= 2 && fdev[1]->toread))
3282 /* If we want to read from a failed device, then
3283 * we need to actually read every other device.
3284 */
3285 return 1;
3286
a9d56950
N		 3287 /* Sometimes neither read-modify-write nor reconstruct-write
3288 * cycles can work. In those cases we read every block we
3289 * can. Then the parity-update is certain to have enough to
3290 * work with.
3291 * This can only be a problem when we need to write something,
3292 * and some device has failed. If either of those tests
3293 * fail we need look no further.
3294 */
3295 if (!s->failed || !s->to_write)
3296 return 0;
3297
3298 if (test_bit(R5_Insync, &dev->flags) &&
3299 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3300 /* Pre-reads at not permitted until after short delay
3301 * to gather multiple requests. However if this
3302 * device is no Insync, the block could only be be computed
3303 * and there is no need to delay that.
3304 */
3305 return 0;
ea664c82
N		 3306
3307 for (i = 0; i < s->failed; i++) {
3308 if (fdev[i]->towrite &&
3309 !test_bit(R5_UPTODATE, &fdev[i]->flags) &&
3310 !test_bit(R5_OVERWRITE, &fdev[i]->flags))
3311 /* If we have a partial write to a failed
3312 * device, then we will need to reconstruct
3313 * the content of that device, so all other
3314 * devices must be read.
3315 */
3316 return 1;
3317 }
3318
3319 /* If we are forced to do a reconstruct-write, either because
3320 * the current RAID6 implementation only supports that, or
3321 * or because parity cannot be trusted and we are currently
3322 * recovering it, there is extra need to be careful.
3323 * If one of the devices that we would need to read, because
3324 * it is not being overwritten (and maybe not written at all)
3325 * is missing/faulty, then we need to read everything we can.
3326 */
3327 if (sh->raid_conf->level != 6 &&
3328 sh->sector < sh->raid_conf->mddev->recovery_cp)
3329 /* reconstruct-write isn't being forced */
3330 return 0;
3331 for (i = 0; i < s->failed; i++) {
10d82c5f
N		 3332 if (s->failed_num[i] != sh->pd_idx &&
3333 s->failed_num[i] != sh->qd_idx &&
3334 !test_bit(R5_UPTODATE, &fdev[i]->flags) &&
ea664c82
N		 3335 !test_bit(R5_OVERWRITE, &fdev[i]->flags))
3336 return 1;
3337 }
3338
2c58f06e
N		 3339 return 0;
3340}
3341
3342static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s,
3343 int disk_idx, int disks)
3344{
3345 struct r5dev *dev = &sh->dev[disk_idx];
3346
3347 /* is the data in this block needed, and can we get it? */
3348 if (need_this_block(sh, s, disk_idx, disks)) {
5599becc
YT		 3349 /* we would like to get this block, possibly by computing it,
3350 * otherwise read it if the backing disk is insync
3351 */
3352 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
3353 BUG_ON(test_bit(R5_Wantread, &dev->flags));
b0c783b3 3354 BUG_ON(sh->batch_head);
5599becc 3355 if ((s->uptodate == disks - 1) &&
f2b3b44d
N		 3356 (s->failed && (disk_idx == s->failed_num[0] ||
3357 disk_idx == s->failed_num[1]))) {
5599becc
YT		 3358 /* have disk failed, and we're requested to fetch it;
3359 * do compute it
a4456856 3360 */
5599becc
YT		 3361 pr_debug("Computing stripe %llu block %d\n",
3362 (unsigned long long)sh->sector, disk_idx);
3363 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3364 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3365 set_bit(R5_Wantcompute, &dev->flags);
3366 sh->ops.target = disk_idx;
3367 sh->ops.target2 = -1; /* no 2nd target */
3368 s->req_compute = 1;
93b3dbce
N		 3369 /* Careful: from this point on 'uptodate' is in the eye
3370 * of raid_run_ops which services 'compute' operations
3371 * before writes. R5_Wantcompute flags a block that will
3372 * be R5_UPTODATE by the time it is needed for a
3373 * subsequent operation.
3374 */
5599becc
YT		 3375 s->uptodate++;
3376 return 1;
3377 } else if (s->uptodate == disks-2 && s->failed >= 2) {
3378 /* Computing 2-failure is *very* expensive; only
3379 * do it if failed >= 2
3380 */
3381 int other;
3382 for (other = disks; other--; ) {
3383 if (other == disk_idx)
3384 continue;
3385 if (!test_bit(R5_UPTODATE,
3386 &sh->dev[other].flags))
3387 break;
a4456856 3388 }
5599becc
YT		 3389 BUG_ON(other < 0);
3390 pr_debug("Computing stripe %llu blocks %d,%d\n",
3391 (unsigned long long)sh->sector,
3392 disk_idx, other);
3393 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3394 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3395 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
3396 set_bit(R5_Wantcompute, &sh->dev[other].flags);
3397 sh->ops.target = disk_idx;
3398 sh->ops.target2 = other;
3399 s->uptodate += 2;
3400 s->req_compute = 1;
3401 return 1;
3402 } else if (test_bit(R5_Insync, &dev->flags)) {
3403 set_bit(R5_LOCKED, &dev->flags);
3404 set_bit(R5_Wantread, &dev->flags);
3405 s->locked++;
3406 pr_debug("Reading block %d (sync=%d)\n",
3407 disk_idx, s->syncing);
a4456856
DW		 3408 }
3409 }
5599becc
YT		 3410
3411 return 0;
3412}
3413
3414/**
93b3dbce 3415 * handle_stripe_fill - read or compute data to satisfy pending requests.
5599becc 3416 */
93b3dbce
N		 3417static void handle_stripe_fill(struct stripe_head *sh,
3418 struct stripe_head_state *s,
3419 int disks)
5599becc
YT		 3420{
3421 int i;
3422
3423 /* look for blocks to read/compute, skip this if a compute
3424 * is already in flight, or if the stripe contents are in the
3425 * midst of changing due to a write
3426 */
3427 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
3428 !sh->reconstruct_state)
3429 for (i = disks; i--; )
93b3dbce 3430 if (fetch_block(sh, s, i, disks))
5599becc 3431 break;
a4456856
DW		 3432 set_bit(STRIPE_HANDLE, &sh->state);
3433}
3434
787b76fa
N		 3435static void break_stripe_batch_list(struct stripe_head *head_sh,
3436 unsigned long handle_flags);
1fe797e6 3437/* handle_stripe_clean_event
a4456856
DW		 3438 * any written block on an uptodate or failed drive can be returned.
3439 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
3440 * never LOCKED, so we don't need to test 'failed' directly.
3441 */
d1688a6d 3442static void handle_stripe_clean_event(struct r5conf *conf,
34a6f80e 3443 struct stripe_head *sh, int disks, struct bio_list *return_bi)
a4456856
DW		 3444{
3445 int i;
3446 struct r5dev *dev;
f8dfcffd 3447 int discard_pending = 0;
59fc630b 3448 struct stripe_head *head_sh = sh;
3449 bool do_endio = false;
a4456856
DW		 3450
3451 for (i = disks; i--; )
3452 if (sh->dev[i].written) {
3453 dev = &sh->dev[i];
3454 if (!test_bit(R5_LOCKED, &dev->flags) &&
9e444768 3455 (test_bit(R5_UPTODATE, &dev->flags) ||
d592a996
SL		 3456 test_bit(R5_Discard, &dev->flags) ||
3457 test_bit(R5_SkipCopy, &dev->flags))) {
a4456856
DW		 3458 /* We can return any write requests */
3459 struct bio *wbi, *wbi2;
45b4233c 3460 pr_debug("Return write for disc %d\n", i);
ca64cae9
N		 3461 if (test_and_clear_bit(R5_Discard, &dev->flags))
3462 clear_bit(R5_UPTODATE, &dev->flags);
d592a996
SL		 3463 if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) {
3464 WARN_ON(test_bit(R5_UPTODATE, &dev->flags));
d592a996 3465 }
59fc630b 3466 do_endio = true;
3467
3468returnbi:
3469 dev->page = dev->orig_page;
a4456856
DW		 3470 wbi = dev->written;
3471 dev->written = NULL;
4f024f37 3472 while (wbi && wbi->bi_iter.bi_sector <
a4456856
DW		 3473 dev->sector + STRIPE_SECTORS) {
3474 wbi2 = r5_next_bio(wbi, dev->sector);
e7836bd6 3475 if (!raid5_dec_bi_active_stripes(wbi)) {
a4456856 3476 md_write_end(conf->mddev);
34a6f80e 3477 bio_list_add(return_bi, wbi);
a4456856
DW		 3478 }
3479 wbi = wbi2;
3480 }
7eaf7e8e
SL		 3481 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3482 STRIPE_SECTORS,
a4456856 3483 !test_bit(STRIPE_DEGRADED, &sh->state),
7eaf7e8e 3484 0);
59fc630b 3485 if (head_sh->batch_head) {
3486 sh = list_first_entry(&sh->batch_list,
3487 struct stripe_head,
3488 batch_list);
3489 if (sh != head_sh) {
3490 dev = &sh->dev[i];
3491 goto returnbi;
3492 }
3493 }
3494 sh = head_sh;
3495 dev = &sh->dev[i];
f8dfcffd
N		 3496 } else if (test_bit(R5_Discard, &dev->flags))
3497 discard_pending = 1;
d592a996
SL		 3498 WARN_ON(test_bit(R5_SkipCopy, &dev->flags));
3499 WARN_ON(dev->page != dev->orig_page);
f8dfcffd
N		 3500 }
3501 if (!discard_pending &&
3502 test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) {
3503 clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
3504 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
3505 if (sh->qd_idx >= 0) {
3506 clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
3507 clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags);
3508 }
3509 /* now that discard is done we can proceed with any sync */
3510 clear_bit(STRIPE_DISCARD, &sh->state);
d47648fc
SL		 3511 /*
3512 * SCSI discard will change some bio fields and the stripe has
3513 * no updated data, so remove it from hash list and the stripe
3514 * will be reinitialized
3515 */
3516 spin_lock_irq(&conf->device_lock);
59fc630b 3517unhash:
d47648fc 3518 remove_hash(sh);
59fc630b 3519 if (head_sh->batch_head) {
3520 sh = list_first_entry(&sh->batch_list,
3521 struct stripe_head, batch_list);
3522 if (sh != head_sh)
3523 goto unhash;
3524 }
d47648fc 3525 spin_unlock_irq(&conf->device_lock);
59fc630b 3526 sh = head_sh;
3527
f8dfcffd
N		 3528 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state))
3529 set_bit(STRIPE_HANDLE, &sh->state);
3530
3531 }
8b3e6cdc
DW		 3532
3533 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3534 if (atomic_dec_and_test(&conf->pending_full_writes))
3535 md_wakeup_thread(conf->mddev->thread);
59fc630b 3536
787b76fa
N		 3537 if (head_sh->batch_head && do_endio)
3538 break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS);
a4456856
DW		 3539}
3540
d1688a6d 3541static void handle_stripe_dirtying(struct r5conf *conf,
c8ac1803
N		 3542 struct stripe_head *sh,
3543 struct stripe_head_state *s,
3544 int disks)
a4456856
DW		 3545{
3546 int rmw = 0, rcw = 0, i;
a7854487
AL		 3547 sector_t recovery_cp = conf->mddev->recovery_cp;
3548
584acdd4 3549 /* Check whether resync is now happening or should start.
a7854487
AL		 3550 * If yes, then the array is dirty (after unclean shutdown or
3551 * initial creation), so parity in some stripes might be inconsistent.
3552 * In this case, we need to always do reconstruct-write, to ensure
3553 * that in case of drive failure or read-error correction, we
3554 * generate correct data from the parity.
3555 */
584acdd4 3556 if (conf->rmw_level == PARITY_DISABLE_RMW ||
26ac1073
N		 3557 (recovery_cp < MaxSector && sh->sector >= recovery_cp &&
3558 s->failed == 0)) {
a7854487 3559 /* Calculate the real rcw later - for now make it
c8ac1803
N		 3560 * look like rcw is cheaper
3561 */
3562 rcw = 1; rmw = 2;
584acdd4
MS		 3563 pr_debug("force RCW rmw_level=%u, recovery_cp=%llu sh->sector=%llu\n",
3564 conf->rmw_level, (unsigned long long)recovery_cp,
a7854487 3565 (unsigned long long)sh->sector);
c8ac1803 3566 } else for (i = disks; i--; ) {
a4456856
DW		 3567 /* would I have to read this buffer for read_modify_write */
3568 struct r5dev *dev = &sh->dev[i];
584acdd4 3569 if ((dev->towrite || i == sh->pd_idx || i == sh->qd_idx) &&
a4456856 3570 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 3571 !(test_bit(R5_UPTODATE, &dev->flags) ||
3572 test_bit(R5_Wantcompute, &dev->flags))) {
a4456856
DW		 3573 if (test_bit(R5_Insync, &dev->flags))
3574 rmw++;
3575 else
3576 rmw += 2*disks; /* cannot read it */
3577 }
3578 /* Would I have to read this buffer for reconstruct_write */
584acdd4
MS		 3579 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
3580 i != sh->pd_idx && i != sh->qd_idx &&
a4456856 3581 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 3582 !(test_bit(R5_UPTODATE, &dev->flags) ||
3583 test_bit(R5_Wantcompute, &dev->flags))) {
67f45548
N		 3584 if (test_bit(R5_Insync, &dev->flags))
3585 rcw++;
a4456856
DW		 3586 else
3587 rcw += 2*disks;
3588 }
3589 }
45b4233c 3590 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
a4456856
DW		 3591 (unsigned long long)sh->sector, rmw, rcw);
3592 set_bit(STRIPE_HANDLE, &sh->state);
584acdd4 3593 if ((rmw < rcw || (rmw == rcw && conf->rmw_level == PARITY_ENABLE_RMW)) && rmw > 0) {
a4456856 3594 /* prefer read-modify-write, but need to get some data */
e3620a3a
JB		 3595 if (conf->mddev->queue)
3596 blk_add_trace_msg(conf->mddev->queue,
3597 "raid5 rmw %llu %d",
3598 (unsigned long long)sh->sector, rmw);
a4456856
DW		 3599 for (i = disks; i--; ) {
3600 struct r5dev *dev = &sh->dev[i];
584acdd4 3601 if ((dev->towrite || i == sh->pd_idx || i == sh->qd_idx) &&
a4456856 3602 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 3603 !(test_bit(R5_UPTODATE, &dev->flags) ||
3604 test_bit(R5_Wantcompute, &dev->flags)) &&
a4456856 3605 test_bit(R5_Insync, &dev->flags)) {
67f45548
N		 3606 if (test_bit(STRIPE_PREREAD_ACTIVE,
3607 &sh->state)) {
3608 pr_debug("Read_old block %d for r-m-w\n",
3609 i);
a4456856
DW		 3610 set_bit(R5_LOCKED, &dev->flags);
3611 set_bit(R5_Wantread, &dev->flags);
3612 s->locked++;
3613 } else {
3614 set_bit(STRIPE_DELAYED, &sh->state);
3615 set_bit(STRIPE_HANDLE, &sh->state);
3616 }
3617 }
3618 }
a9add5d9 3619 }
584acdd4 3620 if ((rcw < rmw || (rcw == rmw && conf->rmw_level != PARITY_ENABLE_RMW)) && rcw > 0) {
a4456856 3621 /* want reconstruct write, but need to get some data */
a9add5d9 3622 int qread =0;
c8ac1803 3623 rcw = 0;
a4456856
DW		 3624 for (i = disks; i--; ) {
3625 struct r5dev *dev = &sh->dev[i];
3626 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
c8ac1803 3627 i != sh->pd_idx && i != sh->qd_idx &&
a4456856 3628 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219 3629 !(test_bit(R5_UPTODATE, &dev->flags) ||
c8ac1803
N		 3630 test_bit(R5_Wantcompute, &dev->flags))) {
3631 rcw++;
67f45548
N		 3632 if (test_bit(R5_Insync, &dev->flags) &&
3633 test_bit(STRIPE_PREREAD_ACTIVE,
3634 &sh->state)) {
45b4233c 3635 pr_debug("Read_old block "
a4456856
DW		 3636 "%d for Reconstruct\n", i);
3637 set_bit(R5_LOCKED, &dev->flags);
3638 set_bit(R5_Wantread, &dev->flags);
3639 s->locked++;
a9add5d9 3640 qread++;
a4456856
DW		 3641 } else {
3642 set_bit(STRIPE_DELAYED, &sh->state);
3643 set_bit(STRIPE_HANDLE, &sh->state);
3644 }
3645 }
3646 }
e3620a3a 3647 if (rcw && conf->mddev->queue)
a9add5d9
N		 3648 blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d",
3649 (unsigned long long)sh->sector,
3650 rcw, qread, test_bit(STRIPE_DELAYED, &sh->state));
c8ac1803 3651 }
b1b02fe9
N		 3652
3653 if (rcw > disks && rmw > disks &&
3654 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3655 set_bit(STRIPE_DELAYED, &sh->state);
3656
a4456856
DW		 3657 /* now if nothing is locked, and if we have enough data,
3658 * we can start a write request
3659 */
f38e1219
DW		 3660 /* since handle_stripe can be called at any time we need to handle the
3661 * case where a compute block operation has been submitted and then a
ac6b53b6
DW		 3662 * subsequent call wants to start a write request. raid_run_ops only
3663 * handles the case where compute block and reconstruct are requested
f38e1219
DW		 3664 * simultaneously. If this is not the case then new writes need to be
3665 * held off until the compute completes.
3666 */
976ea8d4
DW		 3667 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
3668 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
3669 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
c0f7bddb 3670 schedule_reconstruction(sh, s, rcw == 0, 0);
a4456856
DW		 3671}
3672
d1688a6d 3673static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh,
a4456856
DW		 3674 struct stripe_head_state *s, int disks)
3675{
ecc65c9b 3676 struct r5dev *dev = NULL;
bd2ab670 3677
59fc630b 3678 BUG_ON(sh->batch_head);
a4456856 3679 set_bit(STRIPE_HANDLE, &sh->state);
e89f8962 3680
ecc65c9b
DW		 3681 switch (sh->check_state) {
3682 case check_state_idle:
3683 /* start a new check operation if there are no failures */
bd2ab670 3684 if (s->failed == 0) {
bd2ab670 3685 BUG_ON(s->uptodate != disks);
ecc65c9b
DW		 3686 sh->check_state = check_state_run;
3687 set_bit(STRIPE_OP_CHECK, &s->ops_request);
bd2ab670 3688 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
bd2ab670 3689 s->uptodate--;
ecc65c9b 3690 break;
bd2ab670 3691 }
f2b3b44d 3692 dev = &sh->dev[s->failed_num[0]];
ecc65c9b
DW		 3693 /* fall through */
3694 case check_state_compute_result:
3695 sh->check_state = check_state_idle;
3696 if (!dev)
3697 dev = &sh->dev[sh->pd_idx];
3698
3699 /* check that a write has not made the stripe insync */
3700 if (test_bit(STRIPE_INSYNC, &sh->state))
3701 break;
c8894419 3702
a4456856 3703 /* either failed parity check, or recovery is happening */
a4456856
DW		 3704 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
3705 BUG_ON(s->uptodate != disks);
3706
3707 set_bit(R5_LOCKED, &dev->flags);
ecc65c9b 3708 s->locked++;
a4456856 3709 set_bit(R5_Wantwrite, &dev->flags);
830ea016 3710
a4456856 3711 clear_bit(STRIPE_DEGRADED, &sh->state);
a4456856 3712 set_bit(STRIPE_INSYNC, &sh->state);
ecc65c9b
DW		 3713 break;
3714 case check_state_run:
3715 break; /* we will be called again upon completion */
3716 case check_state_check_result:
3717 sh->check_state = check_state_idle;
3718
3719 /* if a failure occurred during the check operation, leave
3720 * STRIPE_INSYNC not set and let the stripe be handled again
3721 */
3722 if (s->failed)
3723 break;
3724
3725 /* handle a successful check operation, if parity is correct
3726 * we are done. Otherwise update the mismatch count and repair
3727 * parity if !MD_RECOVERY_CHECK
3728 */
ad283ea4 3729 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
ecc65c9b
DW		 3730 /* parity is correct (on disc,
3731 * not in buffer any more)
3732 */
3733 set_bit(STRIPE_INSYNC, &sh->state);
3734 else {
7f7583d4 3735 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
ecc65c9b
DW		 3736 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
3737 /* don't try to repair!! */
3738 set_bit(STRIPE_INSYNC, &sh->state);
3739 else {
3740 sh->check_state = check_state_compute_run;
976ea8d4 3741 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
ecc65c9b
DW		 3742 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3743 set_bit(R5_Wantcompute,
3744 &sh->dev[sh->pd_idx].flags);
3745 sh->ops.target = sh->pd_idx;
ac6b53b6 3746 sh->ops.target2 = -1;
ecc65c9b
DW		 3747 s->uptodate++;
3748 }
3749 }
3750 break;
3751 case check_state_compute_run:
3752 break;
3753 default:
3754 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
3755 __func__, sh->check_state,
3756 (unsigned long long) sh->sector);
3757 BUG();
a4456856
DW		 3758 }
3759}
3760
d1688a6d 3761static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh,
36d1c647 3762 struct stripe_head_state *s,
f2b3b44d 3763 int disks)
a4456856 3764{
a4456856 3765 int pd_idx = sh->pd_idx;
34e04e87 3766 int qd_idx = sh->qd_idx;
d82dfee0 3767 struct r5dev *dev;
a4456856 3768
59fc630b 3769 BUG_ON(sh->batch_head);
a4456856
DW		 3770 set_bit(STRIPE_HANDLE, &sh->state);
3771
3772 BUG_ON(s->failed > 2);
d82dfee0 3773
a4456856
DW		 3774 /* Want to check and possibly repair P and Q.
3775 * However there could be one 'failed' device, in which
3776 * case we can only check one of them, possibly using the
3777 * other to generate missing data
3778 */
3779
d82dfee0
DW		 3780 switch (sh->check_state) {
3781 case check_state_idle:
3782 /* start a new check operation if there are < 2 failures */
f2b3b44d 3783 if (s->failed == s->q_failed) {
d82dfee0 3784 /* The only possible failed device holds Q, so it
a4456856
DW		 3785 * makes sense to check P (If anything else were failed,
3786 * we would have used P to recreate it).
3787 */
d82dfee0 3788 sh->check_state = check_state_run;
a4456856 3789 }
f2b3b44d 3790 if (!s->q_failed && s->failed < 2) {
d82dfee0 3791 /* Q is not failed, and we didn't use it to generate
a4456856
DW		 3792 * anything, so it makes sense to check it
3793 */
d82dfee0
DW		 3794 if (sh->check_state == check_state_run)
3795 sh->check_state = check_state_run_pq;
3796 else
3797 sh->check_state = check_state_run_q;
a4456856 3798 }
a4456856 3799
d82dfee0
DW		 3800 /* discard potentially stale zero_sum_result */
3801 sh->ops.zero_sum_result = 0;
a4456856 3802
d82dfee0
DW		 3803 if (sh->check_state == check_state_run) {
3804 /* async_xor_zero_sum destroys the contents of P */
3805 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
3806 s->uptodate--;
a4456856 3807 }
d82dfee0
DW		 3808 if (sh->check_state >= check_state_run &&
3809 sh->check_state <= check_state_run_pq) {
3810 /* async_syndrome_zero_sum preserves P and Q, so
3811 * no need to mark them !uptodate here
3812 */
3813 set_bit(STRIPE_OP_CHECK, &s->ops_request);
3814 break;
a4456856
DW		 3815 }
3816
d82dfee0
DW		 3817 /* we have 2-disk failure */
3818 BUG_ON(s->failed != 2);
3819 /* fall through */
3820 case check_state_compute_result:
3821 sh->check_state = check_state_idle;
a4456856 3822
d82dfee0
DW		 3823 /* check that a write has not made the stripe insync */
3824 if (test_bit(STRIPE_INSYNC, &sh->state))
3825 break;
a4456856
DW		 3826
3827 /* now write out any block on a failed drive,
d82dfee0 3828 * or P or Q if they were recomputed
a4456856 3829 */
d82dfee0 3830 BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */
a4456856 3831 if (s->failed == 2) {
f2b3b44d 3832 dev = &sh->dev[s->failed_num[1]];
a4456856
DW		 3833 s->locked++;
3834 set_bit(R5_LOCKED, &dev->flags);
3835 set_bit(R5_Wantwrite, &dev->flags);
3836 }
3837 if (s->failed >= 1) {
f2b3b44d 3838 dev = &sh->dev[s->failed_num[0]];
a4456856
DW		 3839 s->locked++;
3840 set_bit(R5_LOCKED, &dev->flags);
3841 set_bit(R5_Wantwrite, &dev->flags);
3842 }
d82dfee0 3843 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
a4456856
DW		 3844 dev = &sh->dev[pd_idx];
3845 s->locked++;
3846 set_bit(R5_LOCKED, &dev->flags);
3847 set_bit(R5_Wantwrite, &dev->flags);
3848 }
d82dfee0 3849 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
a4456856
DW		 3850 dev = &sh->dev[qd_idx];
3851 s->locked++;
3852 set_bit(R5_LOCKED, &dev->flags);
3853 set_bit(R5_Wantwrite, &dev->flags);
3854 }
3855 clear_bit(STRIPE_DEGRADED, &sh->state);
3856
3857 set_bit(STRIPE_INSYNC, &sh->state);
d82dfee0
DW		 3858 break;
3859 case check_state_run:
3860 case check_state_run_q:
3861 case check_state_run_pq:
3862 break; /* we will be called again upon completion */
3863 case check_state_check_result:
3864 sh->check_state = check_state_idle;
3865
3866 /* handle a successful check operation, if parity is correct
3867 * we are done. Otherwise update the mismatch count and repair
3868 * parity if !MD_RECOVERY_CHECK
3869 */
3870 if (sh->ops.zero_sum_result == 0) {
3871 /* both parities are correct */
3872 if (!s->failed)
3873 set_bit(STRIPE_INSYNC, &sh->state);
3874 else {
3875 /* in contrast to the raid5 case we can validate
3876 * parity, but still have a failure to write
3877 * back
3878 */
3879 sh->check_state = check_state_compute_result;
3880 /* Returning at this point means that we may go
3881 * off and bring p and/or q uptodate again so
3882 * we make sure to check zero_sum_result again
3883 * to verify if p or q need writeback
3884 */
3885 }
3886 } else {
7f7583d4 3887 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
d82dfee0
DW		 3888 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
3889 /* don't try to repair!! */
3890 set_bit(STRIPE_INSYNC, &sh->state);
3891 else {
3892 int *target = &sh->ops.target;
3893
3894 sh->ops.target = -1;
3895 sh->ops.target2 = -1;
3896 sh->check_state = check_state_compute_run;
3897 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3898 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3899 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
3900 set_bit(R5_Wantcompute,
3901 &sh->dev[pd_idx].flags);
3902 *target = pd_idx;
3903 target = &sh->ops.target2;
3904 s->uptodate++;
3905 }
3906 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
3907 set_bit(R5_Wantcompute,
3908 &sh->dev[qd_idx].flags);
3909 *target = qd_idx;
3910 s->uptodate++;
3911 }
3912 }
3913 }
3914 break;
3915 case check_state_compute_run:
3916 break;
3917 default:
3918 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
3919 __func__, sh->check_state,
3920 (unsigned long long) sh->sector);
3921 BUG();
a4456856
DW		 3922 }
3923}
3924
d1688a6d 3925static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh)
a4456856
DW		 3926{
3927 int i;
3928
3929 /* We have read all the blocks in this stripe and now we need to
3930 * copy some of them into a target stripe for expand.
3931 */
f0a50d37 3932 struct dma_async_tx_descriptor *tx = NULL;
59fc630b 3933 BUG_ON(sh->batch_head);
a4456856
DW		 3934 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3935 for (i = 0; i < sh->disks; i++)
34e04e87 3936 if (i != sh->pd_idx && i != sh->qd_idx) {
911d4ee8 3937 int dd_idx, j;
a4456856 3938 struct stripe_head *sh2;
a08abd8c 3939 struct async_submit_ctl submit;
a4456856 3940
784052ec 3941 sector_t bn = compute_blocknr(sh, i, 1);
911d4ee8
N		 3942 sector_t s = raid5_compute_sector(conf, bn, 0,
3943 &dd_idx, NULL);
a8c906ca 3944 sh2 = get_active_stripe(conf, s, 0, 1, 1);
a4456856
DW		 3945 if (sh2 == NULL)
3946 /* so far only the early blocks of this stripe
3947 * have been requested. When later blocks
3948 * get requested, we will try again
3949 */
3950 continue;
3951 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
3952 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
3953 /* must have already done this block */
3954 release_stripe(sh2);
3955 continue;
3956 }
f0a50d37
DW		 3957
3958 /* place all the copies on one channel */
a08abd8c 3959 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
f0a50d37 3960 tx = async_memcpy(sh2->dev[dd_idx].page,
88ba2aa5 3961 sh->dev[i].page, 0, 0, STRIPE_SIZE,
a08abd8c 3962 &submit);
f0a50d37 3963
a4456856
DW		 3964 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
3965 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
3966 for (j = 0; j < conf->raid_disks; j++)
3967 if (j != sh2->pd_idx &&
86c374ba 3968 j != sh2->qd_idx &&
a4456856
DW		 3969 !test_bit(R5_Expanded, &sh2->dev[j].flags))
3970 break;
3971 if (j == conf->raid_disks) {
3972 set_bit(STRIPE_EXPAND_READY, &sh2->state);
3973 set_bit(STRIPE_HANDLE, &sh2->state);
3974 }
3975 release_stripe(sh2);
f0a50d37 3976
a4456856 3977 }
a2e08551 3978 /* done submitting copies, wait for them to complete */
749586b7 3979 async_tx_quiesce(&tx);
a4456856 3980}
1da177e4
LT		 3981
3982/*
3983 * handle_stripe - do things to a stripe.
3984 *
9a3e1101
N		 3985 * We lock the stripe by setting STRIPE_ACTIVE and then examine the
3986 * state of various bits to see what needs to be done.
1da177e4 3987 * Possible results:
9a3e1101
N		 3988 * return some read requests which now have data
3989 * return some write requests which are safely on storage
1da177e4
LT		 3990 * schedule a read on some buffers
3991 * schedule a write of some buffers
3992 * return confirmation of parity correctness
3993 *
1da177e4 3994 */
a4456856 3995
acfe726b 3996static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s)
1da177e4 3997{
d1688a6d 3998 struct r5conf *conf = sh->raid_conf;
f416885e 3999 int disks = sh->disks;
474af965
N		 4000 struct r5dev *dev;
4001 int i;
9a3e1101 4002 int do_recovery = 0;
1da177e4 4003
acfe726b
N		 4004 memset(s, 0, sizeof(*s));
4005
dabc4ec6 4006 s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state) && !sh->batch_head;
4007 s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state) && !sh->batch_head;
acfe726b
N		 4008 s->failed_num[0] = -1;
4009 s->failed_num[1] = -1;
1da177e4 4010
acfe726b 4011 /* Now to look around and see what can be done */
1da177e4 4012 rcu_read_lock();
16a53ecc 4013 for (i=disks; i--; ) {
3cb03002 4014 struct md_rdev *rdev;
31c176ec
N		 4015 sector_t first_bad;
4016 int bad_sectors;
4017 int is_bad = 0;
acfe726b 4018
16a53ecc 4019 dev = &sh->dev[i];
1da177e4 4020
45b4233c 4021 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
9a3e1101
N		 4022 i, dev->flags,
4023 dev->toread, dev->towrite, dev->written);
6c0069c0
YT		 4024 /* maybe we can reply to a read
4025 *
4026 * new wantfill requests are only permitted while
4027 * ops_complete_biofill is guaranteed to be inactive
4028 */
4029 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
4030 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
4031 set_bit(R5_Wantfill, &dev->flags);
1da177e4 4032
16a53ecc 4033 /* now count some things */
cc94015a
N		 4034 if (test_bit(R5_LOCKED, &dev->flags))
4035 s->locked++;
4036 if (test_bit(R5_UPTODATE, &dev->flags))
4037 s->uptodate++;
2d6e4ecc 4038 if (test_bit(R5_Wantcompute, &dev->flags)) {
cc94015a
N		 4039 s->compute++;
4040 BUG_ON(s->compute > 2);
2d6e4ecc 4041 }
1da177e4 4042
acfe726b 4043 if (test_bit(R5_Wantfill, &dev->flags))
cc94015a 4044 s->to_fill++;
acfe726b 4045 else if (dev->toread)
cc94015a 4046 s->to_read++;
16a53ecc 4047 if (dev->towrite) {
cc94015a 4048 s->to_write++;
16a53ecc 4049 if (!test_bit(R5_OVERWRITE, &dev->flags))
cc94015a 4050 s->non_overwrite++;
16a53ecc 4051 }
a4456856 4052 if (dev->written)
cc94015a 4053 s->written++;
14a75d3e
N		 4054 /* Prefer to use the replacement for reads, but only
4055 * if it is recovered enough and has no bad blocks.
4056 */
4057 rdev = rcu_dereference(conf->disks[i].replacement);
4058 if (rdev && !test_bit(Faulty, &rdev->flags) &&
4059 rdev->recovery_offset >= sh->sector + STRIPE_SECTORS &&
4060 !is_badblock(rdev, sh->sector, STRIPE_SECTORS,
4061 &first_bad, &bad_sectors))
4062 set_bit(R5_ReadRepl, &dev->flags);
4063 else {
e6030cb0 4064 if (rdev && !test_bit(Faulty, &rdev->flags))
9a3e1101 4065 set_bit(R5_NeedReplace, &dev->flags);
e6030cb0
N		 4066 else
4067 clear_bit(R5_NeedReplace, &dev->flags);
14a75d3e
N		 4068 rdev = rcu_dereference(conf->disks[i].rdev);
4069 clear_bit(R5_ReadRepl, &dev->flags);
4070 }
9283d8c5
N		 4071 if (rdev && test_bit(Faulty, &rdev->flags))
4072 rdev = NULL;
31c176ec
N		 4073 if (rdev) {
4074 is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
4075 &first_bad, &bad_sectors);
4076 if (s->blocked_rdev == NULL
4077 && (test_bit(Blocked, &rdev->flags)
4078 || is_bad < 0)) {
4079 if (is_bad < 0)
4080 set_bit(BlockedBadBlocks,
4081 &rdev->flags);
4082 s->blocked_rdev = rdev;
4083 atomic_inc(&rdev->nr_pending);
4084 }
6bfe0b49 4085 }
415e72d0
N		 4086 clear_bit(R5_Insync, &dev->flags);
4087 if (!rdev)
4088 /* Not in-sync */;
31c176ec
N		 4089 else if (is_bad) {
4090 /* also not in-sync */
18b9837e
N		 4091 if (!test_bit(WriteErrorSeen, &rdev->flags) &&
4092 test_bit(R5_UPTODATE, &dev->flags)) {
31c176ec
N		 4093 /* treat as in-sync, but with a read error
4094 * which we can now try to correct
4095 */
4096 set_bit(R5_Insync, &dev->flags);
4097 set_bit(R5_ReadError, &dev->flags);
4098 }
4099 } else if (test_bit(In_sync, &rdev->flags))
415e72d0 4100 set_bit(R5_Insync, &dev->flags);
30d7a483 4101 else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
415e72d0 4102 /* in sync if before recovery_offset */
30d7a483
N		 4103 set_bit(R5_Insync, &dev->flags);
4104 else if (test_bit(R5_UPTODATE, &dev->flags) &&
4105 test_bit(R5_Expanded, &dev->flags))
4106 /* If we've reshaped into here, we assume it is Insync.
4107 * We will shortly update recovery_offset to make
4108 * it official.
4109 */
4110 set_bit(R5_Insync, &dev->flags);
4111
1cc03eb9 4112 if (test_bit(R5_WriteError, &dev->flags)) {
14a75d3e
N		 4113 /* This flag does not apply to '.replacement'
4114 * only to .rdev, so make sure to check that*/
4115 struct md_rdev *rdev2 = rcu_dereference(
4116 conf->disks[i].rdev);
4117 if (rdev2 == rdev)
4118 clear_bit(R5_Insync, &dev->flags);
4119 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
bc2607f3 4120 s->handle_bad_blocks = 1;
14a75d3e 4121 atomic_inc(&rdev2->nr_pending);
bc2607f3
N		 4122 } else
4123 clear_bit(R5_WriteError, &dev->flags);
4124 }
1cc03eb9 4125 if (test_bit(R5_MadeGood, &dev->flags)) {
14a75d3e
N		 4126 /* This flag does not apply to '.replacement'
4127 * only to .rdev, so make sure to check that*/
4128 struct md_rdev *rdev2 = rcu_dereference(
4129 conf->disks[i].rdev);
4130 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
b84db560 4131 s->handle_bad_blocks = 1;
14a75d3e 4132 atomic_inc(&rdev2->nr_pending);
b84db560
N		 4133 } else
4134 clear_bit(R5_MadeGood, &dev->flags);
4135 }
977df362
N		 4136 if (test_bit(R5_MadeGoodRepl, &dev->flags)) {
4137 struct md_rdev *rdev2 = rcu_dereference(
4138 conf->disks[i].replacement);
4139 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
4140 s->handle_bad_blocks = 1;
4141 atomic_inc(&rdev2->nr_pending);
4142 } else
4143 clear_bit(R5_MadeGoodRepl, &dev->flags);
4144 }
415e72d0 4145 if (!test_bit(R5_Insync, &dev->flags)) {
16a53ecc
N		 4146 /* The ReadError flag will just be confusing now */
4147 clear_bit(R5_ReadError, &dev->flags);
4148 clear_bit(R5_ReWrite, &dev->flags);
1da177e4 4149 }
415e72d0
N		 4150 if (test_bit(R5_ReadError, &dev->flags))
4151 clear_bit(R5_Insync, &dev->flags);
4152 if (!test_bit(R5_Insync, &dev->flags)) {
cc94015a
N		 4153 if (s->failed < 2)
4154 s->failed_num[s->failed] = i;
4155 s->failed++;
9a3e1101
N		 4156 if (rdev && !test_bit(Faulty, &rdev->flags))
4157 do_recovery = 1;
415e72d0 4158 }
1da177e4 4159 }
9a3e1101
N		 4160 if (test_bit(STRIPE_SYNCING, &sh->state)) {
4161 /* If there is a failed device being replaced,
4162 * we must be recovering.
4163 * else if we are after recovery_cp, we must be syncing
c6d2e084 4164 * else if MD_RECOVERY_REQUESTED is set, we also are syncing.
9a3e1101
N		 4165 * else we can only be replacing
4166 * sync and recovery both need to read all devices, and so
4167 * use the same flag.
4168 */
4169 if (do_recovery ||
c6d2e084 4170 sh->sector >= conf->mddev->recovery_cp ||
4171 test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery)))
9a3e1101
N		 4172 s->syncing = 1;
4173 else
4174 s->replacing = 1;
4175 }
1da177e4 4176 rcu_read_unlock();
cc94015a
N		 4177}
4178
59fc630b 4179static int clear_batch_ready(struct stripe_head *sh)
4180{
b15a9dbd
N		 4181 /* Return '1' if this is a member of batch, or
4182 * '0' if it is a lone stripe or a head which can now be
4183 * handled.
4184 */
59fc630b 4185 struct stripe_head *tmp;
4186 if (!test_and_clear_bit(STRIPE_BATCH_READY, &sh->state))
b15a9dbd 4187 return (sh->batch_head && sh->batch_head != sh);
59fc630b 4188 spin_lock(&sh->stripe_lock);
4189 if (!sh->batch_head) {
4190 spin_unlock(&sh->stripe_lock);
4191 return 0;
4192 }
4193
4194 /*
4195 * this stripe could be added to a batch list before we check
4196 * BATCH_READY, skips it
4197 */
4198 if (sh->batch_head != sh) {
4199 spin_unlock(&sh->stripe_lock);
4200 return 1;
4201 }
4202 spin_lock(&sh->batch_lock);
4203 list_for_each_entry(tmp, &sh->batch_list, batch_list)
4204 clear_bit(STRIPE_BATCH_READY, &tmp->state);
4205 spin_unlock(&sh->batch_lock);
4206 spin_unlock(&sh->stripe_lock);
4207
4208 /*
4209 * BATCH_READY is cleared, no new stripes can be added.
4210 * batch_list can be accessed without lock
4211 */
4212 return 0;
4213}
4214
3960ce79
N		 4215static void break_stripe_batch_list(struct stripe_head *head_sh,
4216 unsigned long handle_flags)
72ac7330 4217{
4e3d62ff 4218 struct stripe_head *sh, *next;
72ac7330 4219 int i;
fb642b92 4220 int do_wakeup = 0;
72ac7330 4221
bb27051f
N		 4222 list_for_each_entry_safe(sh, next, &head_sh->batch_list, batch_list) {
4223
72ac7330 4224 list_del_init(&sh->batch_list);
4225
1b956f7a
N		 4226 WARN_ON_ONCE(sh->state & ((1 << STRIPE_ACTIVE) |
4227 (1 << STRIPE_SYNCING) |
4228 (1 << STRIPE_REPLACED) |
4229 (1 << STRIPE_PREREAD_ACTIVE) |
4230 (1 << STRIPE_DELAYED) |
4231 (1 << STRIPE_BIT_DELAY) |
4232 (1 << STRIPE_FULL_WRITE) |
4233 (1 << STRIPE_BIOFILL_RUN) |
4234 (1 << STRIPE_COMPUTE_RUN) |
4235 (1 << STRIPE_OPS_REQ_PENDING) |
4236 (1 << STRIPE_DISCARD) |
4237 (1 << STRIPE_BATCH_READY) |
4238 (1 << STRIPE_BATCH_ERR) |
4239 (1 << STRIPE_BITMAP_PENDING)));
4240 WARN_ON_ONCE(head_sh->state & ((1 << STRIPE_DISCARD) |
4241 (1 << STRIPE_REPLACED)));
4242
4243 set_mask_bits(&sh->state, ~(STRIPE_EXPAND_SYNC_FLAGS |
4244 (1 << STRIPE_DEGRADED)),
4245 head_sh->state & (1 << STRIPE_INSYNC));
4246
72ac7330 4247 sh->check_state = head_sh->check_state;
4248 sh->reconstruct_state = head_sh->reconstruct_state;
fb642b92
N		 4249 for (i = 0; i < sh->disks; i++) {
4250 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
4251 do_wakeup = 1;
72ac7330 4252 sh->dev[i].flags = head_sh->dev[i].flags &
4253 (~((1 << R5_WriteError) | (1 << R5_Overlap)));
fb642b92 4254 }
72ac7330 4255 spin_lock_irq(&sh->stripe_lock);
4256 sh->batch_head = NULL;
4257 spin_unlock_irq(&sh->stripe_lock);
3960ce79
N		 4258 if (handle_flags == 0 ||
4259 sh->state & handle_flags)
4260 set_bit(STRIPE_HANDLE, &sh->state);
72ac7330 4261 release_stripe(sh);
72ac7330 4262 }
fb642b92
N		 4263 spin_lock_irq(&head_sh->stripe_lock);
4264 head_sh->batch_head = NULL;
4265 spin_unlock_irq(&head_sh->stripe_lock);
4266 for (i = 0; i < head_sh->disks; i++)
4267 if (test_and_clear_bit(R5_Overlap, &head_sh->dev[i].flags))
4268 do_wakeup = 1;
3960ce79
N		 4269 if (head_sh->state & handle_flags)
4270 set_bit(STRIPE_HANDLE, &head_sh->state);
fb642b92
N		 4271
4272 if (do_wakeup)
4273 wake_up(&head_sh->raid_conf->wait_for_overlap);
72ac7330 4274}
4275
cc94015a
N		 4276static void handle_stripe(struct stripe_head *sh)
4277{
4278 struct stripe_head_state s;
d1688a6d 4279 struct r5conf *conf = sh->raid_conf;
3687c061 4280 int i;
84789554
N		 4281 int prexor;
4282 int disks = sh->disks;
474af965 4283 struct r5dev *pdev, *qdev;
cc94015a
N		 4284
4285 clear_bit(STRIPE_HANDLE, &sh->state);
257a4b42 4286 if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) {
cc94015a
N		 4287 /* already being handled, ensure it gets handled
4288 * again when current action finishes */
4289 set_bit(STRIPE_HANDLE, &sh->state);
4290 return;
4291 }
4292
59fc630b 4293 if (clear_batch_ready(sh) ) {
4294 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
4295 return;
4296 }
4297
4e3d62ff 4298 if (test_and_clear_bit(STRIPE_BATCH_ERR, &sh->state))
3960ce79 4299 break_stripe_batch_list(sh, 0);
72ac7330 4300
dabc4ec6 4301 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) && !sh->batch_head) {
f8dfcffd
N		 4302 spin_lock(&sh->stripe_lock);
4303 /* Cannot process 'sync' concurrently with 'discard' */
4304 if (!test_bit(STRIPE_DISCARD, &sh->state) &&
4305 test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
4306 set_bit(STRIPE_SYNCING, &sh->state);
4307 clear_bit(STRIPE_INSYNC, &sh->state);
f94c0b66 4308 clear_bit(STRIPE_REPLACED, &sh->state);
f8dfcffd
N		 4309 }
4310 spin_unlock(&sh->stripe_lock);
cc94015a
N		 4311 }
4312 clear_bit(STRIPE_DELAYED, &sh->state);
4313
4314 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
4315 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
4316 (unsigned long long)sh->sector, sh->state,
4317 atomic_read(&sh->count), sh->pd_idx, sh->qd_idx,
4318 sh->check_state, sh->reconstruct_state);
3687c061 4319
acfe726b 4320 analyse_stripe(sh, &s);
c5a31000 4321
bc2607f3
N		 4322 if (s.handle_bad_blocks) {
4323 set_bit(STRIPE_HANDLE, &sh->state);
4324 goto finish;
4325 }
4326
474af965
N		 4327 if (unlikely(s.blocked_rdev)) {
4328 if (s.syncing || s.expanding || s.expanded ||
9a3e1101 4329 s.replacing || s.to_write || s.written) {
474af965
N		 4330 set_bit(STRIPE_HANDLE, &sh->state);
4331 goto finish;
4332 }
4333 /* There is nothing for the blocked_rdev to block */
4334 rdev_dec_pending(s.blocked_rdev, conf->mddev);
4335 s.blocked_rdev = NULL;
4336 }
4337
4338 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
4339 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
4340 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
4341 }
4342
4343 pr_debug("locked=%d uptodate=%d to_read=%d"
4344 " to_write=%d failed=%d failed_num=%d,%d\n",
4345 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
4346 s.failed_num[0], s.failed_num[1]);
4347 /* check if the array has lost more than max_degraded devices and,
4348 * if so, some requests might need to be failed.
4349 */
9a3f530f
N		 4350 if (s.failed > conf->max_degraded) {
4351 sh->check_state = 0;
4352 sh->reconstruct_state = 0;
626f2092 4353 break_stripe_batch_list(sh, 0);
9a3f530f
N		 4354 if (s.to_read+s.to_write+s.written)
4355 handle_failed_stripe(conf, sh, &s, disks, &s.return_bi);
9a3e1101 4356 if (s.syncing + s.replacing)
9a3f530f
N		 4357 handle_failed_sync(conf, sh, &s);
4358 }
474af965 4359
84789554
N		 4360 /* Now we check to see if any write operations have recently
4361 * completed
4362 */
4363 prexor = 0;
4364 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
4365 prexor = 1;
4366 if (sh->reconstruct_state == reconstruct_state_drain_result ||
4367 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
4368 sh->reconstruct_state = reconstruct_state_idle;
4369
4370 /* All the 'written' buffers and the parity block are ready to
4371 * be written back to disk
4372 */
9e444768
SL		 4373 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) &&
4374 !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags));
84789554 4375 BUG_ON(sh->qd_idx >= 0 &&
9e444768
SL		 4376 !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) &&
4377 !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags));
84789554
N		 4378 for (i = disks; i--; ) {
4379 struct r5dev *dev = &sh->dev[i];
4380 if (test_bit(R5_LOCKED, &dev->flags) &&
4381 (i == sh->pd_idx || i == sh->qd_idx ||
4382 dev->written)) {
4383 pr_debug("Writing block %d\n", i);
4384 set_bit(R5_Wantwrite, &dev->flags);
4385 if (prexor)
4386 continue;
9c4bdf69
N		 4387 if (s.failed > 1)
4388 continue;
84789554
N		 4389 if (!test_bit(R5_Insync, &dev->flags) ||
4390 ((i == sh->pd_idx || i == sh->qd_idx) &&
4391 s.failed == 0))
4392 set_bit(STRIPE_INSYNC, &sh->state);
4393 }
4394 }
4395 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4396 s.dec_preread_active = 1;
4397 }
4398
ef5b7c69
N		 4399 /*
4400 * might be able to return some write requests if the parity blocks
4401 * are safe, or on a failed drive
4402 */
4403 pdev = &sh->dev[sh->pd_idx];
4404 s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx)
4405 || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx);
4406 qdev = &sh->dev[sh->qd_idx];
4407 s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx)
4408 || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx)
4409 || conf->level < 6;
4410
4411 if (s.written &&
4412 (s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
4413 && !test_bit(R5_LOCKED, &pdev->flags)
4414 && (test_bit(R5_UPTODATE, &pdev->flags) ||
4415 test_bit(R5_Discard, &pdev->flags))))) &&
4416 (s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
4417 && !test_bit(R5_LOCKED, &qdev->flags)
4418 && (test_bit(R5_UPTODATE, &qdev->flags) ||
4419 test_bit(R5_Discard, &qdev->flags))))))
4420 handle_stripe_clean_event(conf, sh, disks, &s.return_bi);
4421
4422 /* Now we might consider reading some blocks, either to check/generate
4423 * parity, or to satisfy requests
4424 * or to load a block that is being partially written.
4425 */
4426 if (s.to_read || s.non_overwrite
4427 || (conf->level == 6 && s.to_write && s.failed)
4428 || (s.syncing && (s.uptodate + s.compute < disks))
4429 || s.replacing
4430 || s.expanding)
4431 handle_stripe_fill(sh, &s, disks);
4432
84789554
N		 4433 /* Now to consider new write requests and what else, if anything
4434 * should be read. We do not handle new writes when:
4435 * 1/ A 'write' operation (copy+xor) is already in flight.
4436 * 2/ A 'check' operation is in flight, as it may clobber the parity
4437 * block.
4438 */
4439 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
4440 handle_stripe_dirtying(conf, sh, &s, disks);
4441
4442 /* maybe we need to check and possibly fix the parity for this stripe
4443 * Any reads will already have been scheduled, so we just see if enough
4444 * data is available. The parity check is held off while parity
4445 * dependent operations are in flight.
4446 */
4447 if (sh->check_state ||
4448 (s.syncing && s.locked == 0 &&
4449 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
4450 !test_bit(STRIPE_INSYNC, &sh->state))) {
4451 if (conf->level == 6)
4452 handle_parity_checks6(conf, sh, &s, disks);
4453 else
4454 handle_parity_checks5(conf, sh, &s, disks);
4455 }
c5a31000 4456
f94c0b66
N		 4457 if ((s.replacing || s.syncing) && s.locked == 0
4458 && !test_bit(STRIPE_COMPUTE_RUN, &sh->state)
4459 && !test_bit(STRIPE_REPLACED, &sh->state)) {
9a3e1101
N		 4460 /* Write out to replacement devices where possible */
4461 for (i = 0; i < conf->raid_disks; i++)
f94c0b66
N		 4462 if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) {
4463 WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags));
9a3e1101
N		 4464 set_bit(R5_WantReplace, &sh->dev[i].flags);
4465 set_bit(R5_LOCKED, &sh->dev[i].flags);
4466 s.locked++;
4467 }
f94c0b66
N		 4468 if (s.replacing)
4469 set_bit(STRIPE_INSYNC, &sh->state);
4470 set_bit(STRIPE_REPLACED, &sh->state);
9a3e1101
N		 4471 }
4472 if ((s.syncing || s.replacing) && s.locked == 0 &&
f94c0b66 4473 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
9a3e1101 4474 test_bit(STRIPE_INSYNC, &sh->state)) {
c5a31000
N		 4475 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
4476 clear_bit(STRIPE_SYNCING, &sh->state);
f8dfcffd
N		 4477 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
4478 wake_up(&conf->wait_for_overlap);
c5a31000
N		 4479 }
4480
4481 /* If the failed drives are just a ReadError, then we might need
4482 * to progress the repair/check process
4483 */
4484 if (s.failed <= conf->max_degraded && !conf->mddev->ro)
4485 for (i = 0; i < s.failed; i++) {
4486 struct r5dev *dev = &sh->dev[s.failed_num[i]];
4487 if (test_bit(R5_ReadError, &dev->flags)
4488 && !test_bit(R5_LOCKED, &dev->flags)
4489 && test_bit(R5_UPTODATE, &dev->flags)
4490 ) {
4491 if (!test_bit(R5_ReWrite, &dev->flags)) {
4492 set_bit(R5_Wantwrite, &dev->flags);
4493 set_bit(R5_ReWrite, &dev->flags);
4494 set_bit(R5_LOCKED, &dev->flags);
4495 s.locked++;
4496 } else {
4497 /* let's read it back */
4498 set_bit(R5_Wantread, &dev->flags);
4499 set_bit(R5_LOCKED, &dev->flags);
4500 s.locked++;
4501 }
4502 }
4503 }
4504
3687c061
N		 4505 /* Finish reconstruct operations initiated by the expansion process */
4506 if (sh->reconstruct_state == reconstruct_state_result) {
4507 struct stripe_head *sh_src
4508 = get_active_stripe(conf, sh->sector, 1, 1, 1);
4509 if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) {
4510 /* sh cannot be written until sh_src has been read.
4511 * so arrange for sh to be delayed a little
4512 */
4513 set_bit(STRIPE_DELAYED, &sh->state);
4514 set_bit(STRIPE_HANDLE, &sh->state);
4515 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
4516 &sh_src->state))
4517 atomic_inc(&conf->preread_active_stripes);
4518 release_stripe(sh_src);
4519 goto finish;
4520 }
4521 if (sh_src)
4522 release_stripe(sh_src);
4523
4524 sh->reconstruct_state = reconstruct_state_idle;
4525 clear_bit(STRIPE_EXPANDING, &sh->state);
4526 for (i = conf->raid_disks; i--; ) {
4527 set_bit(R5_Wantwrite, &sh->dev[i].flags);
4528 set_bit(R5_LOCKED, &sh->dev[i].flags);
4529 s.locked++;
4530 }
4531 }
f416885e 4532
3687c061
N		 4533 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
4534 !sh->reconstruct_state) {
4535 /* Need to write out all blocks after computing parity */
4536 sh->disks = conf->raid_disks;
4537 stripe_set_idx(sh->sector, conf, 0, sh);
4538 schedule_reconstruction(sh, &s, 1, 1);
4539 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
4540 clear_bit(STRIPE_EXPAND_READY, &sh->state);
4541 atomic_dec(&conf->reshape_stripes);
4542 wake_up(&conf->wait_for_overlap);
4543 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
4544 }
4545
4546 if (s.expanding && s.locked == 0 &&
4547 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
4548 handle_stripe_expansion(conf, sh);
16a53ecc 4549
3687c061 4550finish:
6bfe0b49 4551 /* wait for this device to become unblocked */
5f066c63
N		 4552 if (unlikely(s.blocked_rdev)) {
4553 if (conf->mddev->external)
4554 md_wait_for_blocked_rdev(s.blocked_rdev,
4555 conf->mddev);
4556 else
4557 /* Internal metadata will immediately
4558 * be written by raid5d, so we don't
4559 * need to wait here.
4560 */
4561 rdev_dec_pending(s.blocked_rdev,
4562 conf->mddev);
4563 }
6bfe0b49 4564
bc2607f3
N		 4565 if (s.handle_bad_blocks)
4566 for (i = disks; i--; ) {
3cb03002 4567 struct md_rdev *rdev;
bc2607f3
N		 4568 struct r5dev *dev = &sh->dev[i];
4569 if (test_and_clear_bit(R5_WriteError, &dev->flags)) {
4570 /* We own a safe reference to the rdev */
4571 rdev = conf->disks[i].rdev;
4572 if (!rdev_set_badblocks(rdev, sh->sector,
4573 STRIPE_SECTORS, 0))
4574 md_error(conf->mddev, rdev);
4575 rdev_dec_pending(rdev, conf->mddev);
4576 }
b84db560
N		 4577 if (test_and_clear_bit(R5_MadeGood, &dev->flags)) {
4578 rdev = conf->disks[i].rdev;
4579 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 4580 STRIPE_SECTORS, 0);
b84db560
N		 4581 rdev_dec_pending(rdev, conf->mddev);
4582 }
977df362
N		 4583 if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) {
4584 rdev = conf->disks[i].replacement;
dd054fce
N		 4585 if (!rdev)
4586 /* rdev have been moved down */
4587 rdev = conf->disks[i].rdev;
977df362 4588 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 4589 STRIPE_SECTORS, 0);
977df362
N		 4590 rdev_dec_pending(rdev, conf->mddev);
4591 }
bc2607f3
N		 4592 }
4593
6c0069c0
YT		 4594 if (s.ops_request)
4595 raid_run_ops(sh, s.ops_request);
4596
f0e43bcd 4597 ops_run_io(sh, &s);
16a53ecc 4598
c5709ef6 4599 if (s.dec_preread_active) {
729a1866 4600 /* We delay this until after ops_run_io so that if make_request
e9c7469b 4601 * is waiting on a flush, it won't continue until the writes
729a1866
N		 4602 * have actually been submitted.
4603 */
4604 atomic_dec(&conf->preread_active_stripes);
4605 if (atomic_read(&conf->preread_active_stripes) <
4606 IO_THRESHOLD)
4607 md_wakeup_thread(conf->mddev->thread);
4608 }
4609
c3cce6cd
N		 4610 if (!bio_list_empty(&s.return_bi)) {
4611 if (test_bit(MD_CHANGE_PENDING, &conf->mddev->flags)) {
4612 spin_lock_irq(&conf->device_lock);
4613 bio_list_merge(&conf->return_bi, &s.return_bi);
4614 spin_unlock_irq(&conf->device_lock);
4615 md_wakeup_thread(conf->mddev->thread);
4616 } else
4617 return_io(&s.return_bi);
4618 }
16a53ecc 4619
257a4b42 4620 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
16a53ecc
N		 4621}
4622
d1688a6d 4623static void raid5_activate_delayed(struct r5conf *conf)
16a53ecc
N		 4624{
4625 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
4626 while (!list_empty(&conf->delayed_list)) {
4627 struct list_head *l = conf->delayed_list.next;
4628 struct stripe_head *sh;
4629 sh = list_entry(l, struct stripe_head, lru);
4630 list_del_init(l);
4631 clear_bit(STRIPE_DELAYED, &sh->state);
4632 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4633 atomic_inc(&conf->preread_active_stripes);
8b3e6cdc 4634 list_add_tail(&sh->lru, &conf->hold_list);
851c30c9 4635 raid5_wakeup_stripe_thread(sh);
16a53ecc 4636 }
482c0834 4637 }
16a53ecc
N		 4638}
4639
566c09c5
SL		 4640static void activate_bit_delay(struct r5conf *conf,
4641 struct list_head *temp_inactive_list)
16a53ecc
N		 4642{
4643 /* device_lock is held */
4644 struct list_head head;
4645 list_add(&head, &conf->bitmap_list);
4646 list_del_init(&conf->bitmap_list);
4647 while (!list_empty(&head)) {
4648 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
566c09c5 4649 int hash;
16a53ecc
N		 4650 list_del_init(&sh->lru);
4651 atomic_inc(&sh->count);
566c09c5
SL		 4652 hash = sh->hash_lock_index;
4653 __release_stripe(conf, sh, &temp_inactive_list[hash]);
16a53ecc
N		 4654 }
4655}
4656
5c675f83 4657static int raid5_congested(struct mddev *mddev, int bits)
f022b2fd 4658{
d1688a6d 4659 struct r5conf *conf = mddev->private;
f022b2fd
N		 4660
4661 /* No difference between reads and writes. Just check
4662 * how busy the stripe_cache is
4663 */
3fa841d7 4664
5423399a 4665 if (test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state))
f022b2fd
N		 4666 return 1;
4667 if (conf->quiesce)
4668 return 1;
4bda556a 4669 if (atomic_read(&conf->empty_inactive_list_nr))
f022b2fd
N		 4670 return 1;
4671
4672 return 0;
4673}
4674
fd01b88c 4675static int in_chunk_boundary(struct mddev *mddev, struct bio *bio)
f679623f 4676{
3cb5edf4 4677 struct r5conf *conf = mddev->private;
4f024f37 4678 sector_t sector = bio->bi_iter.bi_sector + get_start_sect(bio->bi_bdev);
3cb5edf4 4679 unsigned int chunk_sectors;
aa8b57aa 4680 unsigned int bio_sectors = bio_sectors(bio);
f679623f 4681
3cb5edf4 4682 chunk_sectors = min(conf->chunk_sectors, conf->prev_chunk_sectors);
f679623f
RBJ		 4683 return chunk_sectors >=
4684 ((sector & (chunk_sectors - 1)) + bio_sectors);
4685}
4686
46031f9a
RBJ		 4687/*
4688 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
4689 * later sampled by raid5d.
4690 */
d1688a6d 4691static void add_bio_to_retry(struct bio *bi,struct r5conf *conf)
46031f9a
RBJ		 4692{
4693 unsigned long flags;
4694
4695 spin_lock_irqsave(&conf->device_lock, flags);
4696
4697 bi->bi_next = conf->retry_read_aligned_list;
4698 conf->retry_read_aligned_list = bi;
4699
4700 spin_unlock_irqrestore(&conf->device_lock, flags);
4701 md_wakeup_thread(conf->mddev->thread);
4702}
4703
d1688a6d 4704static struct bio *remove_bio_from_retry(struct r5conf *conf)
46031f9a
RBJ		 4705{
4706 struct bio *bi;
4707
4708 bi = conf->retry_read_aligned;
4709 if (bi) {
4710 conf->retry_read_aligned = NULL;
4711 return bi;
4712 }
4713 bi = conf->retry_read_aligned_list;
4714 if(bi) {
387bb173 4715 conf->retry_read_aligned_list = bi->bi_next;
46031f9a 4716 bi->bi_next = NULL;
960e739d
JA		 4717 /*
4718 * this sets the active strip count to 1 and the processed
4719 * strip count to zero (upper 8 bits)
4720 */
e7836bd6 4721 raid5_set_bi_stripes(bi, 1); /* biased count of active stripes */
46031f9a
RBJ		 4722 }
4723
4724 return bi;
4725}
4726
f679623f
RBJ		 4727/*
4728 * The "raid5_align_endio" should check if the read succeeded and if it
4729 * did, call bio_endio on the original bio (having bio_put the new bio
4730 * first).
4731 * If the read failed..
4732 */
4246a0b6 4733static void raid5_align_endio(struct bio *bi)
f679623f
RBJ		 4734{
4735 struct bio* raid_bi = bi->bi_private;
fd01b88c 4736 struct mddev *mddev;
d1688a6d 4737 struct r5conf *conf;
3cb03002 4738 struct md_rdev *rdev;
9b81c842 4739 int error = bi->bi_error;
46031f9a 4740
f679623f 4741 bio_put(bi);
46031f9a 4742
46031f9a
RBJ		 4743 rdev = (void*)raid_bi->bi_next;
4744 raid_bi->bi_next = NULL;
2b7f2228
N		 4745 mddev = rdev->mddev;
4746 conf = mddev->private;
46031f9a
RBJ		 4747
4748 rdev_dec_pending(rdev, conf->mddev);
4749
9b81c842 4750 if (!error) {
0a82a8d1
LT		 4751 trace_block_bio_complete(bdev_get_queue(raid_bi->bi_bdev),
4752 raid_bi, 0);
4246a0b6 4753 bio_endio(raid_bi);
46031f9a 4754 if (atomic_dec_and_test(&conf->active_aligned_reads))
b1b46486 4755 wake_up(&conf->wait_for_quiescent);
6712ecf8 4756 return;
46031f9a
RBJ		 4757 }
4758
45b4233c 4759 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
46031f9a
RBJ		 4760
4761 add_bio_to_retry(raid_bi, conf);
f679623f
RBJ		 4762}
4763
7ef6b12a 4764static int raid5_read_one_chunk(struct mddev *mddev, struct bio *raid_bio)
f679623f 4765{
d1688a6d 4766 struct r5conf *conf = mddev->private;
8553fe7e 4767 int dd_idx;
f679623f 4768 struct bio* align_bi;
3cb03002 4769 struct md_rdev *rdev;
671488cc 4770 sector_t end_sector;
f679623f
RBJ		 4771
4772 if (!in_chunk_boundary(mddev, raid_bio)) {
7ef6b12a 4773 pr_debug("%s: non aligned\n", __func__);
f679623f
RBJ		 4774 return 0;
4775 }
4776 /*
a167f663 4777 * use bio_clone_mddev to make a copy of the bio
f679623f 4778 */
a167f663 4779 align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev);
f679623f
RBJ		 4780 if (!align_bi)
4781 return 0;
4782 /*
4783 * set bi_end_io to a new function, and set bi_private to the
4784 * original bio.
4785 */
4786 align_bi->bi_end_io = raid5_align_endio;
4787 align_bi->bi_private = raid_bio;
4788 /*
4789 * compute position
4790 */
4f024f37
KO		 4791 align_bi->bi_iter.bi_sector =
4792 raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector,
4793 0, &dd_idx, NULL);
f679623f 4794
f73a1c7d 4795 end_sector = bio_end_sector(align_bi);
f679623f 4796 rcu_read_lock();
671488cc
N		 4797 rdev = rcu_dereference(conf->disks[dd_idx].replacement);
4798 if (!rdev || test_bit(Faulty, &rdev->flags) ||
4799 rdev->recovery_offset < end_sector) {
4800 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
4801 if (rdev &&
4802 (test_bit(Faulty, &rdev->flags) ||
4803 !(test_bit(In_sync, &rdev->flags) ||
4804 rdev->recovery_offset >= end_sector)))
4805 rdev = NULL;
4806 }
4807 if (rdev) {
31c176ec
N		 4808 sector_t first_bad;
4809 int bad_sectors;
4810
f679623f
RBJ		 4811 atomic_inc(&rdev->nr_pending);
4812 rcu_read_unlock();
46031f9a
RBJ		 4813 raid_bio->bi_next = (void*)rdev;
4814 align_bi->bi_bdev = rdev->bdev;
b7c44ed9 4815 bio_clear_flag(align_bi, BIO_SEG_VALID);
46031f9a 4816
7140aafc 4817 if (is_badblock(rdev, align_bi->bi_iter.bi_sector,
4f024f37 4818 bio_sectors(align_bi),
31c176ec 4819 &first_bad, &bad_sectors)) {
387bb173
NB		 4820 bio_put(align_bi);
4821 rdev_dec_pending(rdev, mddev);
4822 return 0;
4823 }
4824
6c0544e2 4825 /* No reshape active, so we can trust rdev->data_offset */
4f024f37 4826 align_bi->bi_iter.bi_sector += rdev->data_offset;
6c0544e2 4827
46031f9a 4828 spin_lock_irq(&conf->device_lock);
b1b46486 4829 wait_event_lock_irq(conf->wait_for_quiescent,
46031f9a 4830 conf->quiesce == 0,
eed8c02e 4831 conf->device_lock);
46031f9a
RBJ		 4832 atomic_inc(&conf->active_aligned_reads);
4833 spin_unlock_irq(&conf->device_lock);
4834
e3620a3a
JB		 4835 if (mddev->gendisk)
4836 trace_block_bio_remap(bdev_get_queue(align_bi->bi_bdev),
4837 align_bi, disk_devt(mddev->gendisk),
4f024f37 4838 raid_bio->bi_iter.bi_sector);
f679623f
RBJ		 4839 generic_make_request(align_bi);
4840 return 1;
4841 } else {
4842 rcu_read_unlock();
46031f9a 4843 bio_put(align_bi);
f679623f
RBJ		 4844 return 0;
4845 }
4846}
4847
7ef6b12a
ML		 4848static struct bio *chunk_aligned_read(struct mddev *mddev, struct bio *raid_bio)
4849{
4850 struct bio *split;
4851
4852 do {
4853 sector_t sector = raid_bio->bi_iter.bi_sector;
4854 unsigned chunk_sects = mddev->chunk_sectors;
4855 unsigned sectors = chunk_sects - (sector & (chunk_sects-1));
4856
4857 if (sectors < bio_sectors(raid_bio)) {
4858 split = bio_split(raid_bio, sectors, GFP_NOIO, fs_bio_set);
4859 bio_chain(split, raid_bio);
4860 } else
4861 split = raid_bio;
4862
4863 if (!raid5_read_one_chunk(mddev, split)) {
4864 if (split != raid_bio)
4865 generic_make_request(raid_bio);
4866 return split;
4867 }
4868 } while (split != raid_bio);
4869
4870 return NULL;
4871}
4872
8b3e6cdc
DW		 4873/* __get_priority_stripe - get the next stripe to process
4874 *
4875 * Full stripe writes are allowed to pass preread active stripes up until
4876 * the bypass_threshold is exceeded. In general the bypass_count
4877 * increments when the handle_list is handled before the hold_list; however, it
4878 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
4879 * stripe with in flight i/o. The bypass_count will be reset when the
4880 * head of the hold_list has changed, i.e. the head was promoted to the
4881 * handle_list.
4882 */
851c30c9 4883static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group)
8b3e6cdc 4884{
851c30c9
SL		 4885 struct stripe_head *sh = NULL, *tmp;
4886 struct list_head *handle_list = NULL;
bfc90cb0 4887 struct r5worker_group *wg = NULL;
851c30c9
SL		 4888
4889 if (conf->worker_cnt_per_group == 0) {
4890 handle_list = &conf->handle_list;
4891 } else if (group != ANY_GROUP) {
4892 handle_list = &conf->worker_groups[group].handle_list;
bfc90cb0 4893 wg = &conf->worker_groups[group];
851c30c9
SL		 4894 } else {
4895 int i;
4896 for (i = 0; i < conf->group_cnt; i++) {
4897 handle_list = &conf->worker_groups[i].handle_list;
bfc90cb0 4898 wg = &conf->worker_groups[i];
851c30c9
SL		 4899 if (!list_empty(handle_list))
4900 break;
4901 }
4902 }
8b3e6cdc
DW		 4903
4904 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
4905 __func__,
851c30c9 4906 list_empty(handle_list) ? "empty" : "busy",
8b3e6cdc
DW		 4907 list_empty(&conf->hold_list) ? "empty" : "busy",
4908 atomic_read(&conf->pending_full_writes), conf->bypass_count);
4909
851c30c9
SL		 4910 if (!list_empty(handle_list)) {
4911 sh = list_entry(handle_list->next, typeof(*sh), lru);
8b3e6cdc
DW		 4912
4913 if (list_empty(&conf->hold_list))
4914 conf->bypass_count = 0;
4915 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
4916 if (conf->hold_list.next == conf->last_hold)
4917 conf->bypass_count++;
4918 else {
4919 conf->last_hold = conf->hold_list.next;
4920 conf->bypass_count -= conf->bypass_threshold;
4921 if (conf->bypass_count < 0)
4922 conf->bypass_count = 0;
4923 }
4924 }
4925 } else if (!list_empty(&conf->hold_list) &&
4926 ((conf->bypass_threshold &&
4927 conf->bypass_count > conf->bypass_threshold) ||
4928 atomic_read(&conf->pending_full_writes) == 0)) {
851c30c9
SL		 4929
4930 list_for_each_entry(tmp, &conf->hold_list, lru) {
4931 if (conf->worker_cnt_per_group == 0 ||
4932 group == ANY_GROUP ||
4933 !cpu_online(tmp->cpu) ||
4934 cpu_to_group(tmp->cpu) == group) {
4935 sh = tmp;
4936 break;
4937 }
4938 }
4939
4940 if (sh) {
4941 conf->bypass_count -= conf->bypass_threshold;
4942 if (conf->bypass_count < 0)
4943 conf->bypass_count = 0;
4944 }
bfc90cb0 4945 wg = NULL;
851c30c9
SL		 4946 }
4947
4948 if (!sh)
8b3e6cdc
DW		 4949 return NULL;
4950
bfc90cb0
SL		 4951 if (wg) {
4952 wg->stripes_cnt--;
4953 sh->group = NULL;
4954 }
8b3e6cdc 4955 list_del_init(&sh->lru);
c7a6d35e 4956 BUG_ON(atomic_inc_return(&sh->count) != 1);
8b3e6cdc
DW		 4957 return sh;
4958}
f679623f 4959
8811b596
SL		 4960struct raid5_plug_cb {
4961 struct blk_plug_cb cb;
4962 struct list_head list;
566c09c5 4963 struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
8811b596
SL		 4964};
4965
4966static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule)
4967{
4968 struct raid5_plug_cb *cb = container_of(
4969 blk_cb, struct raid5_plug_cb, cb);
4970 struct stripe_head *sh;
4971 struct mddev *mddev = cb->cb.data;
4972 struct r5conf *conf = mddev->private;
a9add5d9 4973 int cnt = 0;
566c09c5 4974 int hash;
8811b596
SL		 4975
4976 if (cb->list.next && !list_empty(&cb->list)) {
4977 spin_lock_irq(&conf->device_lock);
4978 while (!list_empty(&cb->list)) {
4979 sh = list_first_entry(&cb->list, struct stripe_head, lru);
4980 list_del_init(&sh->lru);
4981 /*
4982 * avoid race release_stripe_plug() sees
4983 * STRIPE_ON_UNPLUG_LIST clear but the stripe
4984 * is still in our list
4985 */
4e857c58 4986 smp_mb__before_atomic();
8811b596 4987 clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state);
773ca82f
SL		 4988 /*
4989 * STRIPE_ON_RELEASE_LIST could be set here. In that
4990 * case, the count is always > 1 here
4991 */
566c09c5
SL		 4992 hash = sh->hash_lock_index;
4993 __release_stripe(conf, sh, &cb->temp_inactive_list[hash]);
a9add5d9 4994 cnt++;
8811b596
SL		 4995 }
4996 spin_unlock_irq(&conf->device_lock);
4997 }
566c09c5
SL		 4998 release_inactive_stripe_list(conf, cb->temp_inactive_list,
4999 NR_STRIPE_HASH_LOCKS);
e3620a3a
JB		 5000 if (mddev->queue)
5001 trace_block_unplug(mddev->queue, cnt, !from_schedule);
8811b596
SL		 5002 kfree(cb);
5003}
5004
5005static void release_stripe_plug(struct mddev *mddev,
5006 struct stripe_head *sh)
5007{
5008 struct blk_plug_cb *blk_cb = blk_check_plugged(
5009 raid5_unplug, mddev,
5010 sizeof(struct raid5_plug_cb));
5011 struct raid5_plug_cb *cb;
5012
5013 if (!blk_cb) {
5014 release_stripe(sh);
5015 return;
5016 }
5017
5018 cb = container_of(blk_cb, struct raid5_plug_cb, cb);
5019
566c09c5
SL		 5020 if (cb->list.next == NULL) {
5021 int i;
8811b596 5022 INIT_LIST_HEAD(&cb->list);
566c09c5
SL		 5023 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5024 INIT_LIST_HEAD(cb->temp_inactive_list + i);
5025 }
8811b596
SL		 5026
5027 if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state))
5028 list_add_tail(&sh->lru, &cb->list);
5029 else
5030 release_stripe(sh);
5031}
5032
620125f2
SL		 5033static void make_discard_request(struct mddev *mddev, struct bio *bi)
5034{
5035 struct r5conf *conf = mddev->private;
5036 sector_t logical_sector, last_sector;
5037 struct stripe_head *sh;
5038 int remaining;
5039 int stripe_sectors;
5040
5041 if (mddev->reshape_position != MaxSector)
5042 /* Skip discard while reshape is happening */
5043 return;
5044
4f024f37
KO		 5045 logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1);
5046 last_sector = bi->bi_iter.bi_sector + (bi->bi_iter.bi_size>>9);
620125f2
SL		 5047
5048 bi->bi_next = NULL;
5049 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
5050
5051 stripe_sectors = conf->chunk_sectors *
5052 (conf->raid_disks - conf->max_degraded);
5053 logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector,
5054 stripe_sectors);
5055 sector_div(last_sector, stripe_sectors);
5056
5057 logical_sector *= conf->chunk_sectors;
5058 last_sector *= conf->chunk_sectors;
5059
5060 for (; logical_sector < last_sector;
5061 logical_sector += STRIPE_SECTORS) {
5062 DEFINE_WAIT(w);
5063 int d;
5064 again:
5065 sh = get_active_stripe(conf, logical_sector, 0, 0, 0);
5066 prepare_to_wait(&conf->wait_for_overlap, &w,
5067 TASK_UNINTERRUPTIBLE);
f8dfcffd
N		 5068 set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
5069 if (test_bit(STRIPE_SYNCING, &sh->state)) {
5070 release_stripe(sh);
5071 schedule();
5072 goto again;
5073 }
5074 clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
620125f2
SL		 5075 spin_lock_irq(&sh->stripe_lock);
5076 for (d = 0; d < conf->raid_disks; d++) {
5077 if (d == sh->pd_idx || d == sh->qd_idx)
5078 continue;
5079 if (sh->dev[d].towrite || sh->dev[d].toread) {
5080 set_bit(R5_Overlap, &sh->dev[d].flags);
5081 spin_unlock_irq(&sh->stripe_lock);
5082 release_stripe(sh);
5083 schedule();
5084 goto again;
5085 }
5086 }
f8dfcffd 5087 set_bit(STRIPE_DISCARD, &sh->state);
620125f2 5088 finish_wait(&conf->wait_for_overlap, &w);
7a87f434 5089 sh->overwrite_disks = 0;
620125f2
SL		 5090 for (d = 0; d < conf->raid_disks; d++) {
5091 if (d == sh->pd_idx || d == sh->qd_idx)
5092 continue;
5093 sh->dev[d].towrite = bi;
5094 set_bit(R5_OVERWRITE, &sh->dev[d].flags);
5095 raid5_inc_bi_active_stripes(bi);
7a87f434 5096 sh->overwrite_disks++;
620125f2
SL		 5097 }
5098 spin_unlock_irq(&sh->stripe_lock);
5099 if (conf->mddev->bitmap) {
5100 for (d = 0;
5101 d < conf->raid_disks - conf->max_degraded;
5102 d++)
5103 bitmap_startwrite(mddev->bitmap,
5104 sh->sector,
5105 STRIPE_SECTORS,
5106 0);
5107 sh->bm_seq = conf->seq_flush + 1;
5108 set_bit(STRIPE_BIT_DELAY, &sh->state);
5109 }
5110
5111 set_bit(STRIPE_HANDLE, &sh->state);
5112 clear_bit(STRIPE_DELAYED, &sh->state);
5113 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
5114 atomic_inc(&conf->preread_active_stripes);
5115 release_stripe_plug(mddev, sh);
5116 }
5117
5118 remaining = raid5_dec_bi_active_stripes(bi);
5119 if (remaining == 0) {
5120 md_write_end(mddev);
4246a0b6 5121 bio_endio(bi);
620125f2
SL		 5122 }
5123}
5124
b4fdcb02 5125static void make_request(struct mddev *mddev, struct bio * bi)
1da177e4 5126{
d1688a6d 5127 struct r5conf *conf = mddev->private;
911d4ee8 5128 int dd_idx;
1da177e4
LT		 5129 sector_t new_sector;
5130 sector_t logical_sector, last_sector;
5131 struct stripe_head *sh;
a362357b 5132 const int rw = bio_data_dir(bi);
49077326 5133 int remaining;
27c0f68f
SL		 5134 DEFINE_WAIT(w);
5135 bool do_prepare;
1da177e4 5136
e9c7469b
TH		 5137 if (unlikely(bi->bi_rw & REQ_FLUSH)) {
5138 md_flush_request(mddev, bi);
5a7bbad2 5139 return;
e5dcdd80
N		 5140 }
5141
3d310eb7 5142 md_write_start(mddev, bi);
06d91a5f 5143
9ffc8f7c
EM		 5144 /*
5145 * If array is degraded, better not do chunk aligned read because
5146 * later we might have to read it again in order to reconstruct
5147 * data on failed drives.
5148 */
5149 if (rw == READ && mddev->degraded == 0 &&
7ef6b12a
ML		 5150 mddev->reshape_position == MaxSector) {
5151 bi = chunk_aligned_read(mddev, bi);
5152 if (!bi)
5153 return;
5154 }
52488615 5155
620125f2
SL		 5156 if (unlikely(bi->bi_rw & REQ_DISCARD)) {
5157 make_discard_request(mddev, bi);
5158 return;
5159 }
5160
4f024f37 5161 logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1);
f73a1c7d 5162 last_sector = bio_end_sector(bi);
1da177e4
LT		 5163 bi->bi_next = NULL;
5164 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
06d91a5f 5165
27c0f68f 5166 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
1da177e4 5167 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
b5663ba4 5168 int previous;
c46501b2 5169 int seq;
b578d55f 5170
27c0f68f 5171 do_prepare = false;
7ecaa1e6 5172 retry:
c46501b2 5173 seq = read_seqcount_begin(&conf->gen_lock);
b5663ba4 5174 previous = 0;
27c0f68f
SL		 5175 if (do_prepare)
5176 prepare_to_wait(&conf->wait_for_overlap, &w,
5177 TASK_UNINTERRUPTIBLE);
b0f9ec04 5178 if (unlikely(conf->reshape_progress != MaxSector)) {
fef9c61f 5179 /* spinlock is needed as reshape_progress may be
df8e7f76
N		 5180 * 64bit on a 32bit platform, and so it might be
5181 * possible to see a half-updated value
aeb878b0 5182 * Of course reshape_progress could change after
df8e7f76
N		 5183 * the lock is dropped, so once we get a reference
5184 * to the stripe that we think it is, we will have
5185 * to check again.
5186 */
7ecaa1e6 5187 spin_lock_irq(&conf->device_lock);
2c810cdd 5188 if (mddev->reshape_backwards
fef9c61f
N		 5189 ? logical_sector < conf->reshape_progress
5190 : logical_sector >= conf->reshape_progress) {
b5663ba4
N		 5191 previous = 1;
5192 } else {
2c810cdd 5193 if (mddev->reshape_backwards
fef9c61f
N		 5194 ? logical_sector < conf->reshape_safe
5195 : logical_sector >= conf->reshape_safe) {
b578d55f
N		 5196 spin_unlock_irq(&conf->device_lock);
5197 schedule();
27c0f68f 5198 do_prepare = true;
b578d55f
N		 5199 goto retry;
5200 }
5201 }
7ecaa1e6
N		 5202 spin_unlock_irq(&conf->device_lock);
5203 }
16a53ecc 5204
112bf897
N		 5205 new_sector = raid5_compute_sector(conf, logical_sector,
5206 previous,
911d4ee8 5207 &dd_idx, NULL);
0c55e022 5208 pr_debug("raid456: make_request, sector %llu logical %llu\n",
c46501b2 5209 (unsigned long long)new_sector,
1da177e4
LT		 5210 (unsigned long long)logical_sector);
5211
b5663ba4 5212 sh = get_active_stripe(conf, new_sector, previous,
a8c906ca 5213 (bi->bi_rw&RWA_MASK), 0);
1da177e4 5214 if (sh) {
b0f9ec04 5215 if (unlikely(previous)) {
7ecaa1e6 5216 /* expansion might have moved on while waiting for a
df8e7f76
N		 5217 * stripe, so we must do the range check again.
5218 * Expansion could still move past after this
5219 * test, but as we are holding a reference to
5220 * 'sh', we know that if that happens,
5221 * STRIPE_EXPANDING will get set and the expansion
5222 * won't proceed until we finish with the stripe.
7ecaa1e6
N		 5223 */
5224 int must_retry = 0;
5225 spin_lock_irq(&conf->device_lock);
2c810cdd 5226 if (mddev->reshape_backwards
b0f9ec04
N		 5227 ? logical_sector >= conf->reshape_progress
5228 : logical_sector < conf->reshape_progress)
7ecaa1e6
N		 5229 /* mismatch, need to try again */
5230 must_retry = 1;
5231 spin_unlock_irq(&conf->device_lock);
5232 if (must_retry) {
5233 release_stripe(sh);
7a3ab908 5234 schedule();
27c0f68f 5235 do_prepare = true;
7ecaa1e6
N		 5236 goto retry;
5237 }
5238 }
c46501b2
N		 5239 if (read_seqcount_retry(&conf->gen_lock, seq)) {
5240 /* Might have got the wrong stripe_head
5241 * by accident
5242 */
5243 release_stripe(sh);
5244 goto retry;
5245 }
e62e58a5 5246
ffd96e35 5247 if (rw == WRITE &&
a5c308d4 5248 logical_sector >= mddev->suspend_lo &&
e464eafd
N		 5249 logical_sector < mddev->suspend_hi) {
5250 release_stripe(sh);
e62e58a5
N		 5251 /* As the suspend_* range is controlled by
5252 * userspace, we want an interruptible
5253 * wait.
5254 */
5255 flush_signals(current);
5256 prepare_to_wait(&conf->wait_for_overlap,
5257 &w, TASK_INTERRUPTIBLE);
5258 if (logical_sector >= mddev->suspend_lo &&
27c0f68f 5259 logical_sector < mddev->suspend_hi) {
e62e58a5 5260 schedule();
27c0f68f
SL		 5261 do_prepare = true;
5262 }
e464eafd
N		 5263 goto retry;
5264 }
7ecaa1e6
N		 5265
5266 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
da41ba65 5267 !add_stripe_bio(sh, bi, dd_idx, rw, previous)) {
7ecaa1e6
N		 5268 /* Stripe is busy expanding or
5269 * add failed due to overlap. Flush everything
1da177e4
LT		 5270 * and wait a while
5271 */
482c0834 5272 md_wakeup_thread(mddev->thread);
1da177e4
LT		 5273 release_stripe(sh);
5274 schedule();
27c0f68f 5275 do_prepare = true;
1da177e4
LT		 5276 goto retry;
5277 }
6ed3003c
N		 5278 set_bit(STRIPE_HANDLE, &sh->state);
5279 clear_bit(STRIPE_DELAYED, &sh->state);
59fc630b 5280 if ((!sh->batch_head || sh == sh->batch_head) &&
5281 (bi->bi_rw & REQ_SYNC) &&
729a1866
N		 5282 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
5283 atomic_inc(&conf->preread_active_stripes);
8811b596 5284 release_stripe_plug(mddev, sh);
1da177e4
LT		 5285 } else {
5286 /* cannot get stripe for read-ahead, just give-up */
4246a0b6 5287 bi->bi_error = -EIO;
1da177e4
LT		 5288 break;
5289 }
1da177e4 5290 }
27c0f68f 5291 finish_wait(&conf->wait_for_overlap, &w);
7c13edc8 5292
e7836bd6 5293 remaining = raid5_dec_bi_active_stripes(bi);
f6344757 5294 if (remaining == 0) {
1da177e4 5295
16a53ecc 5296 if ( rw == WRITE )
1da177e4 5297 md_write_end(mddev);
6712ecf8 5298
0a82a8d1
LT		 5299 trace_block_bio_complete(bdev_get_queue(bi->bi_bdev),
5300 bi, 0);
4246a0b6 5301 bio_endio(bi);
1da177e4 5302 }
1da177e4
LT		 5303}
5304
fd01b88c 5305static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks);
b522adcd 5306
fd01b88c 5307static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped)
1da177e4 5308{
52c03291
N		 5309 /* reshaping is quite different to recovery/resync so it is
5310 * handled quite separately ... here.
5311 *
5312 * On each call to sync_request, we gather one chunk worth of
5313 * destination stripes and flag them as expanding.
5314 * Then we find all the source stripes and request reads.
5315 * As the reads complete, handle_stripe will copy the data
5316 * into the destination stripe and release that stripe.
5317 */
d1688a6d 5318 struct r5conf *conf = mddev->private;
1da177e4 5319 struct stripe_head *sh;
ccfcc3c1 5320 sector_t first_sector, last_sector;
f416885e
N		 5321 int raid_disks = conf->previous_raid_disks;
5322 int data_disks = raid_disks - conf->max_degraded;
5323 int new_data_disks = conf->raid_disks - conf->max_degraded;
52c03291
N		 5324 int i;
5325 int dd_idx;
c8f517c4 5326 sector_t writepos, readpos, safepos;
ec32a2bd 5327 sector_t stripe_addr;
7a661381 5328 int reshape_sectors;
ab69ae12 5329 struct list_head stripes;
92140480 5330 sector_t retn;
52c03291 5331
fef9c61f
N		 5332 if (sector_nr == 0) {
5333 /* If restarting in the middle, skip the initial sectors */
2c810cdd 5334 if (mddev->reshape_backwards &&
fef9c61f
N		 5335 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
5336 sector_nr = raid5_size(mddev, 0, 0)
5337 - conf->reshape_progress;
6cbd8148
N		 5338 } else if (mddev->reshape_backwards &&
5339 conf->reshape_progress == MaxSector) {
5340 /* shouldn't happen, but just in case, finish up.*/
5341 sector_nr = MaxSector;
2c810cdd 5342 } else if (!mddev->reshape_backwards &&
fef9c61f
N		 5343 conf->reshape_progress > 0)
5344 sector_nr = conf->reshape_progress;
f416885e 5345 sector_div(sector_nr, new_data_disks);
fef9c61f 5346 if (sector_nr) {
8dee7211
N		 5347 mddev->curr_resync_completed = sector_nr;
5348 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
fef9c61f 5349 *skipped = 1;
92140480
N		 5350 retn = sector_nr;
5351 goto finish;
fef9c61f 5352 }
52c03291
N		 5353 }
5354
7a661381
N		 5355 /* We need to process a full chunk at a time.
5356 * If old and new chunk sizes differ, we need to process the
5357 * largest of these
5358 */
3cb5edf4
N		 5359
5360 reshape_sectors = max(conf->chunk_sectors, conf->prev_chunk_sectors);
7a661381 5361
b5254dd5
N		 5362 /* We update the metadata at least every 10 seconds, or when
5363 * the data about to be copied would over-write the source of
5364 * the data at the front of the range. i.e. one new_stripe
5365 * along from reshape_progress new_maps to after where
5366 * reshape_safe old_maps to
52c03291 5367 */
fef9c61f 5368 writepos = conf->reshape_progress;
f416885e 5369 sector_div(writepos, new_data_disks);
c8f517c4
N		 5370 readpos = conf->reshape_progress;
5371 sector_div(readpos, data_disks);
fef9c61f 5372 safepos = conf->reshape_safe;
f416885e 5373 sector_div(safepos, data_disks);
2c810cdd 5374 if (mddev->reshape_backwards) {
c74c0d76
N		 5375 BUG_ON(writepos < reshape_sectors);
5376 writepos -= reshape_sectors;
c8f517c4 5377 readpos += reshape_sectors;
7a661381 5378 safepos += reshape_sectors;
fef9c61f 5379 } else {
7a661381 5380 writepos += reshape_sectors;
c74c0d76
N		 5381 /* readpos and safepos are worst-case calculations.
5382 * A negative number is overly pessimistic, and causes
5383 * obvious problems for unsigned storage. So clip to 0.
5384 */
ed37d83e
N		 5385 readpos -= min_t(sector_t, reshape_sectors, readpos);
5386 safepos -= min_t(sector_t, reshape_sectors, safepos);
fef9c61f 5387 }
52c03291 5388
b5254dd5
N		 5389 /* Having calculated the 'writepos' possibly use it
5390 * to set 'stripe_addr' which is where we will write to.
5391 */
5392 if (mddev->reshape_backwards) {
5393 BUG_ON(conf->reshape_progress == 0);
5394 stripe_addr = writepos;
5395 BUG_ON((mddev->dev_sectors &
5396 ~((sector_t)reshape_sectors - 1))
5397 - reshape_sectors - stripe_addr
5398 != sector_nr);
5399 } else {
5400 BUG_ON(writepos != sector_nr + reshape_sectors);
5401 stripe_addr = sector_nr;
5402 }
5403
c8f517c4
N		 5404 /* 'writepos' is the most advanced device address we might write.
5405 * 'readpos' is the least advanced device address we might read.
5406 * 'safepos' is the least address recorded in the metadata as having
5407 * been reshaped.
b5254dd5
N		 5408 * If there is a min_offset_diff, these are adjusted either by
5409 * increasing the safepos/readpos if diff is negative, or
5410 * increasing writepos if diff is positive.
5411 * If 'readpos' is then behind 'writepos', there is no way that we can
c8f517c4
N		 5412 * ensure safety in the face of a crash - that must be done by userspace
5413 * making a backup of the data. So in that case there is no particular
5414 * rush to update metadata.
5415 * Otherwise if 'safepos' is behind 'writepos', then we really need to
5416 * update the metadata to advance 'safepos' to match 'readpos' so that
5417 * we can be safe in the event of a crash.
5418 * So we insist on updating metadata if safepos is behind writepos and
5419 * readpos is beyond writepos.
5420 * In any case, update the metadata every 10 seconds.
5421 * Maybe that number should be configurable, but I'm not sure it is
5422 * worth it.... maybe it could be a multiple of safemode_delay???
5423 */
b5254dd5
N		 5424 if (conf->min_offset_diff < 0) {
5425 safepos += -conf->min_offset_diff;
5426 readpos += -conf->min_offset_diff;
5427 } else
5428 writepos += conf->min_offset_diff;
5429
2c810cdd 5430 if ((mddev->reshape_backwards
c8f517c4
N		 5431 ? (safepos > writepos && readpos < writepos)
5432 : (safepos < writepos && readpos > writepos)) ||
5433 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
52c03291
N		 5434 /* Cannot proceed until we've updated the superblock... */
5435 wait_event(conf->wait_for_overlap,
c91abf5a
N		 5436 atomic_read(&conf->reshape_stripes)==0
5437 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5438 if (atomic_read(&conf->reshape_stripes) != 0)
5439 return 0;
fef9c61f 5440 mddev->reshape_position = conf->reshape_progress;
75d3da43 5441 mddev->curr_resync_completed = sector_nr;
c8f517c4 5442 conf->reshape_checkpoint = jiffies;
850b2b42 5443 set_bit(MD_CHANGE_DEVS, &mddev->flags);
52c03291 5444 md_wakeup_thread(mddev->thread);
850b2b42 5445 wait_event(mddev->sb_wait, mddev->flags == 0 ||
c91abf5a
N		 5446 test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5447 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5448 return 0;
52c03291 5449 spin_lock_irq(&conf->device_lock);
fef9c61f 5450 conf->reshape_safe = mddev->reshape_position;
52c03291
N		 5451 spin_unlock_irq(&conf->device_lock);
5452 wake_up(&conf->wait_for_overlap);
acb180b0 5453 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
52c03291
N		 5454 }
5455
ab69ae12 5456 INIT_LIST_HEAD(&stripes);
7a661381 5457 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
52c03291 5458 int j;
a9f326eb 5459 int skipped_disk = 0;
a8c906ca 5460 sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
52c03291
N		 5461 set_bit(STRIPE_EXPANDING, &sh->state);
5462 atomic_inc(&conf->reshape_stripes);
5463 /* If any of this stripe is beyond the end of the old
5464 * array, then we need to zero those blocks
5465 */
5466 for (j=sh->disks; j--;) {
5467 sector_t s;
5468 if (j == sh->pd_idx)
5469 continue;
f416885e 5470 if (conf->level == 6 &&
d0dabf7e 5471 j == sh->qd_idx)
f416885e 5472 continue;
784052ec 5473 s = compute_blocknr(sh, j, 0);
b522adcd 5474 if (s < raid5_size(mddev, 0, 0)) {
a9f326eb 5475 skipped_disk = 1;
52c03291
N		 5476 continue;
5477 }
5478 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
5479 set_bit(R5_Expanded, &sh->dev[j].flags);
5480 set_bit(R5_UPTODATE, &sh->dev[j].flags);
5481 }
a9f326eb 5482 if (!skipped_disk) {
52c03291
N		 5483 set_bit(STRIPE_EXPAND_READY, &sh->state);
5484 set_bit(STRIPE_HANDLE, &sh->state);
5485 }
ab69ae12 5486 list_add(&sh->lru, &stripes);
52c03291
N		 5487 }
5488 spin_lock_irq(&conf->device_lock);
2c810cdd 5489 if (mddev->reshape_backwards)
7a661381 5490 conf->reshape_progress -= reshape_sectors * new_data_disks;
fef9c61f 5491 else
7a661381 5492 conf->reshape_progress += reshape_sectors * new_data_disks;
52c03291
N		 5493 spin_unlock_irq(&conf->device_lock);
5494 /* Ok, those stripe are ready. We can start scheduling
5495 * reads on the source stripes.
5496 * The source stripes are determined by mapping the first and last
5497 * block on the destination stripes.
5498 */
52c03291 5499 first_sector =
ec32a2bd 5500 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
911d4ee8 5501 1, &dd_idx, NULL);
52c03291 5502 last_sector =
0e6e0271 5503 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
09c9e5fa 5504 * new_data_disks - 1),
911d4ee8 5505 1, &dd_idx, NULL);
58c0fed4
AN		 5506 if (last_sector >= mddev->dev_sectors)
5507 last_sector = mddev->dev_sectors - 1;
52c03291 5508 while (first_sector <= last_sector) {
a8c906ca 5509 sh = get_active_stripe(conf, first_sector, 1, 0, 1);
52c03291
N		 5510 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
5511 set_bit(STRIPE_HANDLE, &sh->state);
5512 release_stripe(sh);
5513 first_sector += STRIPE_SECTORS;
5514 }
ab69ae12
N		 5515 /* Now that the sources are clearly marked, we can release
5516 * the destination stripes
5517 */
5518 while (!list_empty(&stripes)) {
5519 sh = list_entry(stripes.next, struct stripe_head, lru);
5520 list_del_init(&sh->lru);
5521 release_stripe(sh);
5522 }
c6207277
N		 5523 /* If this takes us to the resync_max point where we have to pause,
5524 * then we need to write out the superblock.
5525 */
7a661381 5526 sector_nr += reshape_sectors;
92140480
N		 5527 retn = reshape_sectors;
5528finish:
c5e19d90
N		 5529 if (mddev->curr_resync_completed > mddev->resync_max ||
5530 (sector_nr - mddev->curr_resync_completed) * 2
c03f6a19 5531 >= mddev->resync_max - mddev->curr_resync_completed) {
c6207277
N		 5532 /* Cannot proceed until we've updated the superblock... */
5533 wait_event(conf->wait_for_overlap,
c91abf5a
N		 5534 atomic_read(&conf->reshape_stripes) == 0
5535 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5536 if (atomic_read(&conf->reshape_stripes) != 0)
5537 goto ret;
fef9c61f 5538 mddev->reshape_position = conf->reshape_progress;
75d3da43 5539 mddev->curr_resync_completed = sector_nr;
c8f517c4 5540 conf->reshape_checkpoint = jiffies;
c6207277
N		 5541 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5542 md_wakeup_thread(mddev->thread);
5543 wait_event(mddev->sb_wait,
5544 !test_bit(MD_CHANGE_DEVS, &mddev->flags)
c91abf5a
N		 5545 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5546 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5547 goto ret;
c6207277 5548 spin_lock_irq(&conf->device_lock);
fef9c61f 5549 conf->reshape_safe = mddev->reshape_position;
c6207277
N		 5550 spin_unlock_irq(&conf->device_lock);
5551 wake_up(&conf->wait_for_overlap);
acb180b0 5552 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
c6207277 5553 }
c91abf5a 5554ret:
92140480 5555 return retn;
52c03291
N		 5556}
5557
09314799 5558static inline sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped)
52c03291 5559{
d1688a6d 5560 struct r5conf *conf = mddev->private;
52c03291 5561 struct stripe_head *sh;
58c0fed4 5562 sector_t max_sector = mddev->dev_sectors;
57dab0bd 5563 sector_t sync_blocks;
16a53ecc
N		 5564 int still_degraded = 0;
5565 int i;
1da177e4 5566
72626685 5567 if (sector_nr >= max_sector) {
1da177e4 5568 /* just being told to finish up .. nothing much to do */
cea9c228 5569
29269553
N		 5570 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
5571 end_reshape(conf);
5572 return 0;
5573 }
72626685
N		 5574
5575 if (mddev->curr_resync < max_sector) /* aborted */
5576 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
5577 &sync_blocks, 1);
16a53ecc 5578 else /* completed sync */
72626685
N		 5579 conf->fullsync = 0;
5580 bitmap_close_sync(mddev->bitmap);
5581
1da177e4
LT		 5582 return 0;
5583 }
ccfcc3c1 5584
64bd660b
N		 5585 /* Allow raid5_quiesce to complete */
5586 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
5587
52c03291
N		 5588 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5589 return reshape_request(mddev, sector_nr, skipped);
f6705578 5590
c6207277
N		 5591 /* No need to check resync_max as we never do more than one
5592 * stripe, and as resync_max will always be on a chunk boundary,
5593 * if the check in md_do_sync didn't fire, there is no chance
5594 * of overstepping resync_max here
5595 */
5596
16a53ecc 5597 /* if there is too many failed drives and we are trying
1da177e4
LT		 5598 * to resync, then assert that we are finished, because there is
5599 * nothing we can do.
5600 */
3285edf1 5601 if (mddev->degraded >= conf->max_degraded &&
16a53ecc 5602 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
58c0fed4 5603 sector_t rv = mddev->dev_sectors - sector_nr;
57afd89f 5604 *skipped = 1;
1da177e4
LT		 5605 return rv;
5606 }
6f608040 5607 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
5608 !conf->fullsync &&
5609 !bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
5610 sync_blocks >= STRIPE_SECTORS) {
72626685
N		 5611 /* we can skip this block, and probably more */
5612 sync_blocks /= STRIPE_SECTORS;
5613 *skipped = 1;
5614 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
5615 }
1da177e4 5616
b47490c9
N		 5617 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
5618
a8c906ca 5619 sh = get_active_stripe(conf, sector_nr, 0, 1, 0);
1da177e4 5620 if (sh == NULL) {
a8c906ca 5621 sh = get_active_stripe(conf, sector_nr, 0, 0, 0);
1da177e4 5622 /* make sure we don't swamp the stripe cache if someone else
16a53ecc 5623 * is trying to get access
1da177e4 5624 */
66c006a5 5625 schedule_timeout_uninterruptible(1);
1da177e4 5626 }
16a53ecc 5627 /* Need to check if array will still be degraded after recovery/resync
16d9cfab
EM		 5628 * Note in case of > 1 drive failures it's possible we're rebuilding
5629 * one drive while leaving another faulty drive in array.
16a53ecc 5630 */
16d9cfab
EM		 5631 rcu_read_lock();
5632 for (i = 0; i < conf->raid_disks; i++) {
5633 struct md_rdev *rdev = ACCESS_ONCE(conf->disks[i].rdev);
5634
5635 if (rdev == NULL || test_bit(Faulty, &rdev->flags))
16a53ecc 5636 still_degraded = 1;
16d9cfab
EM		 5637 }
5638 rcu_read_unlock();
16a53ecc
N		 5639
5640 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
5641
83206d66 5642 set_bit(STRIPE_SYNC_REQUESTED, &sh->state);
053f5b65 5643 set_bit(STRIPE_HANDLE, &sh->state);
1da177e4 5644
1da177e4
LT		 5645 release_stripe(sh);
5646
5647 return STRIPE_SECTORS;
5648}
5649
d1688a6d 5650static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio)
46031f9a
RBJ		 5651{
5652 /* We may not be able to submit a whole bio at once as there
5653 * may not be enough stripe_heads available.
5654 * We cannot pre-allocate enough stripe_heads as we may need
5655 * more than exist in the cache (if we allow ever large chunks).
5656 * So we do one stripe head at a time and record in
5657 * ->bi_hw_segments how many have been done.
5658 *
5659 * We *know* that this entire raid_bio is in one chunk, so
5660 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
5661 */
5662 struct stripe_head *sh;
911d4ee8 5663 int dd_idx;
46031f9a
RBJ		 5664 sector_t sector, logical_sector, last_sector;
5665 int scnt = 0;
5666 int remaining;
5667 int handled = 0;
5668
4f024f37
KO		 5669 logical_sector = raid_bio->bi_iter.bi_sector &
5670 ~((sector_t)STRIPE_SECTORS-1);
112bf897 5671 sector = raid5_compute_sector(conf, logical_sector,
911d4ee8 5672 0, &dd_idx, NULL);
f73a1c7d 5673 last_sector = bio_end_sector(raid_bio);
46031f9a
RBJ		 5674
5675 for (; logical_sector < last_sector;
387bb173
NB		 5676 logical_sector += STRIPE_SECTORS,
5677 sector += STRIPE_SECTORS,
5678 scnt++) {
46031f9a 5679
e7836bd6 5680 if (scnt < raid5_bi_processed_stripes(raid_bio))
46031f9a
RBJ		 5681 /* already done this stripe */
5682 continue;
5683
2844dc32 5684 sh = get_active_stripe(conf, sector, 0, 1, 1);
46031f9a
RBJ		 5685
5686 if (!sh) {
5687 /* failed to get a stripe - must wait */
e7836bd6 5688 raid5_set_bi_processed_stripes(raid_bio, scnt);
46031f9a
RBJ		 5689 conf->retry_read_aligned = raid_bio;
5690 return handled;
5691 }
5692
da41ba65 5693 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) {
387bb173 5694 release_stripe(sh);
e7836bd6 5695 raid5_set_bi_processed_stripes(raid_bio, scnt);
387bb173
NB		 5696 conf->retry_read_aligned = raid_bio;
5697 return handled;
5698 }
5699
3f9e7c14 5700 set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags);
36d1c647 5701 handle_stripe(sh);
46031f9a
RBJ		 5702 release_stripe(sh);
5703 handled++;
5704 }
e7836bd6 5705 remaining = raid5_dec_bi_active_stripes(raid_bio);
0a82a8d1
LT		 5706 if (remaining == 0) {
5707 trace_block_bio_complete(bdev_get_queue(raid_bio->bi_bdev),
5708 raid_bio, 0);
4246a0b6 5709 bio_endio(raid_bio);
0a82a8d1 5710 }
46031f9a 5711 if (atomic_dec_and_test(&conf->active_aligned_reads))
b1b46486 5712 wake_up(&conf->wait_for_quiescent);
46031f9a
RBJ		 5713 return handled;
5714}
5715
bfc90cb0 5716static int handle_active_stripes(struct r5conf *conf, int group,
566c09c5
SL		 5717 struct r5worker *worker,
5718 struct list_head *temp_inactive_list)
46a06401
SL		 5719{
5720 struct stripe_head *batch[MAX_STRIPE_BATCH], *sh;
566c09c5
SL		 5721 int i, batch_size = 0, hash;
5722 bool release_inactive = false;
46a06401
SL		 5723
5724 while (batch_size < MAX_STRIPE_BATCH &&
851c30c9 5725 (sh = __get_priority_stripe(conf, group)) != NULL)
46a06401
SL		 5726 batch[batch_size++] = sh;
5727
566c09c5
SL		 5728 if (batch_size == 0) {
5729 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5730 if (!list_empty(temp_inactive_list + i))
5731 break;
5732 if (i == NR_STRIPE_HASH_LOCKS)
5733 return batch_size;
5734 release_inactive = true;
5735 }
46a06401
SL		 5736 spin_unlock_irq(&conf->device_lock);
5737
566c09c5
SL		 5738 release_inactive_stripe_list(conf, temp_inactive_list,
5739 NR_STRIPE_HASH_LOCKS);
5740
5741 if (release_inactive) {
5742 spin_lock_irq(&conf->device_lock);
5743 return 0;
5744 }
5745
46a06401
SL		 5746 for (i = 0; i < batch_size; i++)
5747 handle_stripe(batch[i]);
5748
5749 cond_resched();
5750
5751 spin_lock_irq(&conf->device_lock);
566c09c5
SL		 5752 for (i = 0; i < batch_size; i++) {
5753 hash = batch[i]->hash_lock_index;
5754 __release_stripe(conf, batch[i], &temp_inactive_list[hash]);
5755 }
46a06401
SL		 5756 return batch_size;
5757}
46031f9a 5758
851c30c9
SL		 5759static void raid5_do_work(struct work_struct *work)
5760{
5761 struct r5worker *worker = container_of(work, struct r5worker, work);
5762 struct r5worker_group *group = worker->group;
5763 struct r5conf *conf = group->conf;
5764 int group_id = group - conf->worker_groups;
5765 int handled;
5766 struct blk_plug plug;
5767
5768 pr_debug("+++ raid5worker active\n");
5769
5770 blk_start_plug(&plug);
5771 handled = 0;
5772 spin_lock_irq(&conf->device_lock);
5773 while (1) {
5774 int batch_size, released;
5775
566c09c5 5776 released = release_stripe_list(conf, worker->temp_inactive_list);
851c30c9 5777
566c09c5
SL		 5778 batch_size = handle_active_stripes(conf, group_id, worker,
5779 worker->temp_inactive_list);
bfc90cb0 5780 worker->working = false;
851c30c9
SL		 5781 if (!batch_size && !released)
5782 break;
5783 handled += batch_size;
5784 }
5785 pr_debug("%d stripes handled\n", handled);
5786
5787 spin_unlock_irq(&conf->device_lock);
5788 blk_finish_plug(&plug);
5789
5790 pr_debug("--- raid5worker inactive\n");
5791}
5792
1da177e4
LT		 5793/*
5794 * This is our raid5 kernel thread.
5795 *
5796 * We scan the hash table for stripes which can be handled now.
5797 * During the scan, completed stripes are saved for us by the interrupt
5798 * handler, so that they will not have to wait for our next wakeup.
5799 */
4ed8731d 5800static void raid5d(struct md_thread *thread)
1da177e4 5801{
4ed8731d 5802 struct mddev *mddev = thread->mddev;
d1688a6d 5803 struct r5conf *conf = mddev->private;
1da177e4 5804 int handled;
e1dfa0a2 5805 struct blk_plug plug;
1da177e4 5806
45b4233c 5807 pr_debug("+++ raid5d active\n");
1da177e4
LT		 5808
5809 md_check_recovery(mddev);
1da177e4 5810
c3cce6cd
N		 5811 if (!bio_list_empty(&conf->return_bi) &&
5812 !test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
5813 struct bio_list tmp = BIO_EMPTY_LIST;
5814 spin_lock_irq(&conf->device_lock);
5815 if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
5816 bio_list_merge(&tmp, &conf->return_bi);
5817 bio_list_init(&conf->return_bi);
5818 }
5819 spin_unlock_irq(&conf->device_lock);
5820 return_io(&tmp);
5821 }
5822
e1dfa0a2 5823 blk_start_plug(&plug);
1da177e4
LT		 5824 handled = 0;
5825 spin_lock_irq(&conf->device_lock);
5826 while (1) {
46031f9a 5827 struct bio *bio;
773ca82f
SL		 5828 int batch_size, released;
5829
566c09c5 5830 released = release_stripe_list(conf, conf->temp_inactive_list);
edbe83ab
N		 5831 if (released)
5832 clear_bit(R5_DID_ALLOC, &conf->cache_state);
1da177e4 5833
0021b7bc 5834 if (
7c13edc8
N		 5835 !list_empty(&conf->bitmap_list)) {
5836 /* Now is a good time to flush some bitmap updates */
5837 conf->seq_flush++;
700e432d 5838 spin_unlock_irq(&conf->device_lock);
72626685 5839 bitmap_unplug(mddev->bitmap);
700e432d 5840 spin_lock_irq(&conf->device_lock);
7c13edc8 5841 conf->seq_write = conf->seq_flush;
566c09c5 5842 activate_bit_delay(conf, conf->temp_inactive_list);
72626685 5843 }
0021b7bc 5844 raid5_activate_delayed(conf);
72626685 5845
46031f9a
RBJ		 5846 while ((bio = remove_bio_from_retry(conf))) {
5847 int ok;
5848 spin_unlock_irq(&conf->device_lock);
5849 ok = retry_aligned_read(conf, bio);
5850 spin_lock_irq(&conf->device_lock);
5851 if (!ok)
5852 break;
5853 handled++;
5854 }
5855
566c09c5
SL		 5856 batch_size = handle_active_stripes(conf, ANY_GROUP, NULL,
5857 conf->temp_inactive_list);
773ca82f 5858 if (!batch_size && !released)
1da177e4 5859 break;
46a06401 5860 handled += batch_size;
1da177e4 5861
46a06401
SL		 5862 if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) {
5863 spin_unlock_irq(&conf->device_lock);
de393cde 5864 md_check_recovery(mddev);
46a06401
SL		 5865 spin_lock_irq(&conf->device_lock);
5866 }
1da177e4 5867 }
45b4233c 5868 pr_debug("%d stripes handled\n", handled);
1da177e4
LT		 5869
5870 spin_unlock_irq(&conf->device_lock);
2d5b569b
N		 5871 if (test_and_clear_bit(R5_ALLOC_MORE, &conf->cache_state) &&
5872 mutex_trylock(&conf->cache_size_mutex)) {
edbe83ab
N		 5873 grow_one_stripe(conf, __GFP_NOWARN);
5874 /* Set flag even if allocation failed. This helps
5875 * slow down allocation requests when mem is short
5876 */
5877 set_bit(R5_DID_ALLOC, &conf->cache_state);
2d5b569b 5878 mutex_unlock(&conf->cache_size_mutex);
edbe83ab 5879 }
1da177e4 5880
c9f21aaf 5881 async_tx_issue_pending_all();
e1dfa0a2 5882 blk_finish_plug(&plug);
1da177e4 5883
45b4233c 5884 pr_debug("--- raid5d inactive\n");
1da177e4
LT		 5885}
5886
3f294f4f 5887static ssize_t
fd01b88c 5888raid5_show_stripe_cache_size(struct mddev *mddev, char *page)
3f294f4f 5889{
7b1485ba
N		 5890 struct r5conf *conf;
5891 int ret = 0;
5892 spin_lock(&mddev->lock);
5893 conf = mddev->private;
96de1e66 5894 if (conf)
edbe83ab 5895 ret = sprintf(page, "%d\n", conf->min_nr_stripes);
7b1485ba
N		 5896 spin_unlock(&mddev->lock);
5897 return ret;
3f294f4f
N		 5898}
5899
c41d4ac4 5900int
fd01b88c 5901raid5_set_cache_size(struct mddev *mddev, int size)
3f294f4f 5902{
d1688a6d 5903 struct r5conf *conf = mddev->private;
b5470dc5
DW		 5904 int err;
5905
c41d4ac4 5906 if (size <= 16 || size > 32768)
3f294f4f 5907 return -EINVAL;
486f0644 5908
edbe83ab 5909 conf->min_nr_stripes = size;
2d5b569b 5910 mutex_lock(&conf->cache_size_mutex);
486f0644
N		 5911 while (size < conf->max_nr_stripes &&
5912 drop_one_stripe(conf))
5913 ;
2d5b569b 5914 mutex_unlock(&conf->cache_size_mutex);
486f0644 5915
edbe83ab 5916
b5470dc5
DW		 5917 err = md_allow_write(mddev);
5918 if (err)
5919 return err;
486f0644 5920
2d5b569b 5921 mutex_lock(&conf->cache_size_mutex);
486f0644
N		 5922 while (size > conf->max_nr_stripes)
5923 if (!grow_one_stripe(conf, GFP_KERNEL))
5924 break;
2d5b569b 5925 mutex_unlock(&conf->cache_size_mutex);
486f0644 5926
c41d4ac4
N		 5927 return 0;
5928}
5929EXPORT_SYMBOL(raid5_set_cache_size);
5930
5931static ssize_t
fd01b88c 5932raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len)
c41d4ac4 5933{
6791875e 5934 struct r5conf *conf;
c41d4ac4
N		 5935 unsigned long new;
5936 int err;
5937
5938 if (len >= PAGE_SIZE)
5939 return -EINVAL;
b29bebd6 5940 if (kstrtoul(page, 10, &new))
c41d4ac4 5941 return -EINVAL;
6791875e 5942 err = mddev_lock(mddev);
c41d4ac4
N		 5943 if (err)
5944 return err;
6791875e
N		 5945 conf = mddev->private;
5946 if (!conf)
5947 err = -ENODEV;
5948 else
5949 err = raid5_set_cache_size(mddev, new);
5950 mddev_unlock(mddev);
5951
5952 return err ?: len;
3f294f4f 5953}
007583c9 5954
96de1e66
N		 5955static struct md_sysfs_entry
5956raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
5957 raid5_show_stripe_cache_size,
5958 raid5_store_stripe_cache_size);
3f294f4f 5959
d06f191f
MS		 5960static ssize_t
5961raid5_show_rmw_level(struct mddev *mddev, char *page)
5962{
5963 struct r5conf *conf = mddev->private;
5964 if (conf)
5965 return sprintf(page, "%d\n", conf->rmw_level);
5966 else
5967 return 0;
5968}
5969
5970static ssize_t
5971raid5_store_rmw_level(struct mddev *mddev, const char *page, size_t len)
5972{
5973 struct r5conf *conf = mddev->private;
5974 unsigned long new;
5975
5976 if (!conf)
5977 return -ENODEV;
5978
5979 if (len >= PAGE_SIZE)
5980 return -EINVAL;
5981
5982 if (kstrtoul(page, 10, &new))
5983 return -EINVAL;
5984
5985 if (new != PARITY_DISABLE_RMW && !raid6_call.xor_syndrome)
5986 return -EINVAL;
5987
5988 if (new != PARITY_DISABLE_RMW &&
5989 new != PARITY_ENABLE_RMW &&
5990 new != PARITY_PREFER_RMW)
5991 return -EINVAL;
5992
5993 conf->rmw_level = new;
5994 return len;
5995}
5996
5997static struct md_sysfs_entry
5998raid5_rmw_level = __ATTR(rmw_level, S_IRUGO | S_IWUSR,
5999 raid5_show_rmw_level,
6000 raid5_store_rmw_level);
6001
6002
8b3e6cdc 6003static ssize_t
fd01b88c 6004raid5_show_preread_threshold(struct mddev *mddev, char *page)
8b3e6cdc 6005{
7b1485ba
N		 6006 struct r5conf *conf;
6007 int ret = 0;
6008 spin_lock(&mddev->lock);
6009 conf = mddev->private;
8b3e6cdc 6010 if (conf)
7b1485ba
N		 6011 ret = sprintf(page, "%d\n", conf->bypass_threshold);
6012 spin_unlock(&mddev->lock);
6013 return ret;
8b3e6cdc
DW		 6014}
6015
6016static ssize_t
fd01b88c 6017raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len)
8b3e6cdc 6018{
6791875e 6019 struct r5conf *conf;
4ef197d8 6020 unsigned long new;
6791875e
N		 6021 int err;
6022
8b3e6cdc
DW		 6023 if (len >= PAGE_SIZE)
6024 return -EINVAL;
b29bebd6 6025 if (kstrtoul(page, 10, &new))
8b3e6cdc 6026 return -EINVAL;
6791875e
N		 6027
6028 err = mddev_lock(mddev);
6029 if (err)
6030 return err;
6031 conf = mddev->private;
6032 if (!conf)
6033 err = -ENODEV;
edbe83ab 6034 else if (new > conf->min_nr_stripes)
6791875e
N		 6035 err = -EINVAL;
6036 else
6037 conf->bypass_threshold = new;
6038 mddev_unlock(mddev);
6039 return err ?: len;
8b3e6cdc
DW		 6040}
6041
6042static struct md_sysfs_entry
6043raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
6044 S_IRUGO | S_IWUSR,
6045 raid5_show_preread_threshold,
6046 raid5_store_preread_threshold);
6047
d592a996
SL		 6048static ssize_t
6049raid5_show_skip_copy(struct mddev *mddev, char *page)
6050{
7b1485ba
N		 6051 struct r5conf *conf;
6052 int ret = 0;
6053 spin_lock(&mddev->lock);
6054 conf = mddev->private;
d592a996 6055 if (conf)
7b1485ba
N		 6056 ret = sprintf(page, "%d\n", conf->skip_copy);
6057 spin_unlock(&mddev->lock);
6058 return ret;
d592a996
SL		 6059}
6060
6061static ssize_t
6062raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len)
6063{
6791875e 6064 struct r5conf *conf;
d592a996 6065 unsigned long new;
6791875e
N		 6066 int err;
6067
d592a996
SL		 6068 if (len >= PAGE_SIZE)
6069 return -EINVAL;
d592a996
SL		 6070 if (kstrtoul(page, 10, &new))
6071 return -EINVAL;
6072 new = !!new;
6791875e
N		 6073
6074 err = mddev_lock(mddev);
6075 if (err)
6076 return err;
6077 conf = mddev->private;
6078 if (!conf)
6079 err = -ENODEV;
6080 else if (new != conf->skip_copy) {
6081 mddev_suspend(mddev);
6082 conf->skip_copy = new;
6083 if (new)
6084 mddev->queue->backing_dev_info.capabilities |=
6085 BDI_CAP_STABLE_WRITES;
6086 else
6087 mddev->queue->backing_dev_info.capabilities &=
6088 ~BDI_CAP_STABLE_WRITES;
6089 mddev_resume(mddev);
6090 }
6091 mddev_unlock(mddev);
6092 return err ?: len;
d592a996
SL		 6093}
6094
6095static struct md_sysfs_entry
6096raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR,
6097 raid5_show_skip_copy,
6098 raid5_store_skip_copy);
6099
3f294f4f 6100static ssize_t
fd01b88c 6101stripe_cache_active_show(struct mddev *mddev, char *page)
3f294f4f 6102{
d1688a6d 6103 struct r5conf *conf = mddev->private;
96de1e66
N		 6104 if (conf)
6105 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
6106 else
6107 return 0;
3f294f4f
N		 6108}
6109
96de1e66
N		 6110static struct md_sysfs_entry
6111raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
3f294f4f 6112
b721420e
SL		 6113static ssize_t
6114raid5_show_group_thread_cnt(struct mddev *mddev, char *page)
6115{
7b1485ba
N		 6116 struct r5conf *conf;
6117 int ret = 0;
6118 spin_lock(&mddev->lock);
6119 conf = mddev->private;
b721420e 6120 if (conf)
7b1485ba
N		 6121 ret = sprintf(page, "%d\n", conf->worker_cnt_per_group);
6122 spin_unlock(&mddev->lock);
6123 return ret;
b721420e
SL		 6124}
6125
60aaf933 6126static int alloc_thread_groups(struct r5conf *conf, int cnt,
6127 int *group_cnt,
6128 int *worker_cnt_per_group,
6129 struct r5worker_group **worker_groups);
b721420e
SL		 6130static ssize_t
6131raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len)
6132{
6791875e 6133 struct r5conf *conf;
b721420e
SL		 6134 unsigned long new;
6135 int err;
60aaf933 6136 struct r5worker_group *new_groups, *old_groups;
6137 int group_cnt, worker_cnt_per_group;
b721420e
SL		 6138
6139 if (len >= PAGE_SIZE)
6140 return -EINVAL;
b721420e
SL		 6141 if (kstrtoul(page, 10, &new))
6142 return -EINVAL;
6143
6791875e
N		 6144 err = mddev_lock(mddev);
6145 if (err)
6146 return err;
6147 conf = mddev->private;
6148 if (!conf)
6149 err = -ENODEV;
6150 else if (new != conf->worker_cnt_per_group) {
6151 mddev_suspend(mddev);
b721420e 6152
6791875e
N		 6153 old_groups = conf->worker_groups;
6154 if (old_groups)
6155 flush_workqueue(raid5_wq);
d206dcfa 6156
6791875e
N		 6157 err = alloc_thread_groups(conf, new,
6158 &group_cnt, &worker_cnt_per_group,
6159 &new_groups);
6160 if (!err) {
6161 spin_lock_irq(&conf->device_lock);
6162 conf->group_cnt = group_cnt;
6163 conf->worker_cnt_per_group = worker_cnt_per_group;
6164 conf->worker_groups = new_groups;
6165 spin_unlock_irq(&conf->device_lock);
b721420e 6166
6791875e
N		 6167 if (old_groups)
6168 kfree(old_groups[0].workers);
6169 kfree(old_groups);
6170 }
6171 mddev_resume(mddev);
b721420e 6172 }
6791875e 6173 mddev_unlock(mddev);
b721420e 6174
6791875e 6175 return err ?: len;
b721420e
SL		 6176}
6177
6178static struct md_sysfs_entry
6179raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR,
6180 raid5_show_group_thread_cnt,
6181 raid5_store_group_thread_cnt);
6182
007583c9 6183static struct attribute *raid5_attrs[] = {
3f294f4f
N		 6184 &raid5_stripecache_size.attr,
6185 &raid5_stripecache_active.attr,
8b3e6cdc 6186 &raid5_preread_bypass_threshold.attr,
b721420e 6187 &raid5_group_thread_cnt.attr,
d592a996 6188 &raid5_skip_copy.attr,
d06f191f 6189 &raid5_rmw_level.attr,
3f294f4f
N		 6190 NULL,
6191};
007583c9
N		 6192static struct attribute_group raid5_attrs_group = {
6193 .name = NULL,
6194 .attrs = raid5_attrs,
3f294f4f
N		 6195};
6196
60aaf933 6197static int alloc_thread_groups(struct r5conf *conf, int cnt,
6198 int *group_cnt,
6199 int *worker_cnt_per_group,
6200 struct r5worker_group **worker_groups)
851c30c9 6201{
566c09c5 6202 int i, j, k;
851c30c9
SL		 6203 ssize_t size;
6204 struct r5worker *workers;
6205
60aaf933 6206 *worker_cnt_per_group = cnt;
851c30c9 6207 if (cnt == 0) {
60aaf933 6208 *group_cnt = 0;
6209 *worker_groups = NULL;
851c30c9
SL		 6210 return 0;
6211 }
60aaf933 6212 *group_cnt = num_possible_nodes();
851c30c9 6213 size = sizeof(struct r5worker) * cnt;
60aaf933 6214 workers = kzalloc(size * *group_cnt, GFP_NOIO);
6215 *worker_groups = kzalloc(sizeof(struct r5worker_group) *
6216 *group_cnt, GFP_NOIO);
6217 if (!*worker_groups || !workers) {
851c30c9 6218 kfree(workers);
60aaf933 6219 kfree(*worker_groups);
851c30c9
SL		 6220 return -ENOMEM;
6221 }
6222
60aaf933 6223 for (i = 0; i < *group_cnt; i++) {
851c30c9
SL		 6224 struct r5worker_group *group;
6225
0c775d52 6226 group = &(*worker_groups)[i];
851c30c9
SL		 6227 INIT_LIST_HEAD(&group->handle_list);
6228 group->conf = conf;
6229 group->workers = workers + i * cnt;
6230
6231 for (j = 0; j < cnt; j++) {
566c09c5
SL		 6232 struct r5worker *worker = group->workers + j;
6233 worker->group = group;
6234 INIT_WORK(&worker->work, raid5_do_work);
6235
6236 for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++)
6237 INIT_LIST_HEAD(worker->temp_inactive_list + k);
851c30c9
SL		 6238 }
6239 }
6240
6241 return 0;
6242}
6243
6244static void free_thread_groups(struct r5conf *conf)
6245{
6246 if (conf->worker_groups)
6247 kfree(conf->worker_groups[0].workers);
6248 kfree(conf->worker_groups);
6249 conf->worker_groups = NULL;
6250}
6251
80c3a6ce 6252static sector_t
fd01b88c 6253raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks)
80c3a6ce 6254{
d1688a6d 6255 struct r5conf *conf = mddev->private;
80c3a6ce
DW		 6256
6257 if (!sectors)
6258 sectors = mddev->dev_sectors;
5e5e3e78 6259 if (!raid_disks)
7ec05478 6260 /* size is defined by the smallest of previous and new size */
5e5e3e78 6261 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
80c3a6ce 6262
3cb5edf4
N		 6263 sectors &= ~((sector_t)conf->chunk_sectors - 1);
6264 sectors &= ~((sector_t)conf->prev_chunk_sectors - 1);
80c3a6ce
DW		 6265 return sectors * (raid_disks - conf->max_degraded);
6266}
6267
789b5e03
ON		 6268static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
6269{
6270 safe_put_page(percpu->spare_page);
46d5b785 6271 if (percpu->scribble)
6272 flex_array_free(percpu->scribble);
789b5e03
ON		 6273 percpu->spare_page = NULL;
6274 percpu->scribble = NULL;
6275}
6276
6277static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
6278{
6279 if (conf->level == 6 && !percpu->spare_page)
6280 percpu->spare_page = alloc_page(GFP_KERNEL);
6281 if (!percpu->scribble)
46d5b785 6282 percpu->scribble = scribble_alloc(max(conf->raid_disks,
738a2738
N		 6283 conf->previous_raid_disks),
6284 max(conf->chunk_sectors,
6285 conf->prev_chunk_sectors)
6286 / STRIPE_SECTORS,
6287 GFP_KERNEL);
789b5e03
ON		 6288
6289 if (!percpu->scribble || (conf->level == 6 && !percpu->spare_page)) {
6290 free_scratch_buffer(conf, percpu);
6291 return -ENOMEM;
6292 }
6293
6294 return 0;
6295}
6296
d1688a6d 6297static void raid5_free_percpu(struct r5conf *conf)
36d1c647 6298{
36d1c647
DW		 6299 unsigned long cpu;
6300
6301 if (!conf->percpu)
6302 return;
6303
36d1c647
DW		 6304#ifdef CONFIG_HOTPLUG_CPU
6305 unregister_cpu_notifier(&conf->cpu_notify);
6306#endif
789b5e03
ON		 6307
6308 get_online_cpus();
6309 for_each_possible_cpu(cpu)
6310 free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
36d1c647
DW		 6311 put_online_cpus();
6312
6313 free_percpu(conf->percpu);
6314}
6315
d1688a6d 6316static void free_conf(struct r5conf *conf)
95fc17aa 6317{
edbe83ab
N		 6318 if (conf->shrinker.seeks)
6319 unregister_shrinker(&conf->shrinker);
851c30c9 6320 free_thread_groups(conf);
95fc17aa 6321 shrink_stripes(conf);
36d1c647 6322 raid5_free_percpu(conf);
95fc17aa
DW		 6323 kfree(conf->disks);
6324 kfree(conf->stripe_hashtbl);
6325 kfree(conf);
6326}
6327
36d1c647
DW		 6328#ifdef CONFIG_HOTPLUG_CPU
6329static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action,
6330 void *hcpu)
6331{
d1688a6d 6332 struct r5conf *conf = container_of(nfb, struct r5conf, cpu_notify);
36d1c647
DW		 6333 long cpu = (long)hcpu;
6334 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
6335
6336 switch (action) {
6337 case CPU_UP_PREPARE:
6338 case CPU_UP_PREPARE_FROZEN:
789b5e03 6339 if (alloc_scratch_buffer(conf, percpu)) {
36d1c647
DW		 6340 pr_err("%s: failed memory allocation for cpu%ld\n",
6341 __func__, cpu);
55af6bb5 6342 return notifier_from_errno(-ENOMEM);
36d1c647
DW		 6343 }
6344 break;
6345 case CPU_DEAD:
6346 case CPU_DEAD_FROZEN:
789b5e03 6347 free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
36d1c647
DW		 6348 break;
6349 default:
6350 break;
6351 }
6352 return NOTIFY_OK;
6353}
6354#endif
6355
d1688a6d 6356static int raid5_alloc_percpu(struct r5conf *conf)
36d1c647
DW		 6357{
6358 unsigned long cpu;
789b5e03 6359 int err = 0;
36d1c647 6360
789b5e03
ON		 6361 conf->percpu = alloc_percpu(struct raid5_percpu);
6362 if (!conf->percpu)
36d1c647 6363 return -ENOMEM;
789b5e03
ON		 6364
6365#ifdef CONFIG_HOTPLUG_CPU
6366 conf->cpu_notify.notifier_call = raid456_cpu_notify;
6367 conf->cpu_notify.priority = 0;
6368 err = register_cpu_notifier(&conf->cpu_notify);
6369 if (err)
6370 return err;
6371#endif
36d1c647
DW		 6372
6373 get_online_cpus();
36d1c647 6374 for_each_present_cpu(cpu) {
789b5e03
ON		 6375 err = alloc_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
6376 if (err) {
6377 pr_err("%s: failed memory allocation for cpu%ld\n",
6378 __func__, cpu);
36d1c647
DW		 6379 break;
6380 }
36d1c647 6381 }
36d1c647
DW		 6382 put_online_cpus();
6383
6384 return err;
6385}
6386
edbe83ab
N		 6387static unsigned long raid5_cache_scan(struct shrinker *shrink,
6388 struct shrink_control *sc)
6389{
6390 struct r5conf *conf = container_of(shrink, struct r5conf, shrinker);
2d5b569b
N		 6391 unsigned long ret = SHRINK_STOP;
6392
6393 if (mutex_trylock(&conf->cache_size_mutex)) {
6394 ret= 0;
49895bcc
N		 6395 while (ret < sc->nr_to_scan &&
6396 conf->max_nr_stripes > conf->min_nr_stripes) {
2d5b569b
N		 6397 if (drop_one_stripe(conf) == 0) {
6398 ret = SHRINK_STOP;
6399 break;
6400 }
6401 ret++;
6402 }
6403 mutex_unlock(&conf->cache_size_mutex);
edbe83ab
N		 6404 }
6405 return ret;
6406}
6407
6408static unsigned long raid5_cache_count(struct shrinker *shrink,
6409 struct shrink_control *sc)
6410{
6411 struct r5conf *conf = container_of(shrink, struct r5conf, shrinker);
6412
6413 if (conf->max_nr_stripes < conf->min_nr_stripes)
6414 /* unlikely, but not impossible */
6415 return 0;
6416 return conf->max_nr_stripes - conf->min_nr_stripes;
6417}
6418
d1688a6d 6419static struct r5conf *setup_conf(struct mddev *mddev)
1da177e4 6420{
d1688a6d 6421 struct r5conf *conf;
5e5e3e78 6422 int raid_disk, memory, max_disks;
3cb03002 6423 struct md_rdev *rdev;
1da177e4 6424 struct disk_info *disk;
0232605d 6425 char pers_name[6];
566c09c5 6426 int i;
60aaf933 6427 int group_cnt, worker_cnt_per_group;
6428 struct r5worker_group *new_group;
1da177e4 6429
91adb564
N		 6430 if (mddev->new_level != 5
6431 && mddev->new_level != 4
6432 && mddev->new_level != 6) {
0c55e022 6433 printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n",
91adb564
N		 6434 mdname(mddev), mddev->new_level);
6435 return ERR_PTR(-EIO);
1da177e4 6436 }
91adb564
N		 6437 if ((mddev->new_level == 5
6438 && !algorithm_valid_raid5(mddev->new_layout)) ||
6439 (mddev->new_level == 6
6440 && !algorithm_valid_raid6(mddev->new_layout))) {
0c55e022 6441 printk(KERN_ERR "md/raid:%s: layout %d not supported\n",
91adb564
N		 6442 mdname(mddev), mddev->new_layout);
6443 return ERR_PTR(-EIO);
99c0fb5f 6444 }
91adb564 6445 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
0c55e022 6446 printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n",
91adb564
N		 6447 mdname(mddev), mddev->raid_disks);
6448 return ERR_PTR(-EINVAL);
4bbf3771
N		 6449 }
6450
664e7c41
AN		 6451 if (!mddev->new_chunk_sectors ||
6452 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
6453 !is_power_of_2(mddev->new_chunk_sectors)) {
0c55e022
N		 6454 printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n",
6455 mdname(mddev), mddev->new_chunk_sectors << 9);
91adb564 6456 return ERR_PTR(-EINVAL);
f6705578
N		 6457 }
6458
d1688a6d 6459 conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL);
91adb564 6460 if (conf == NULL)
1da177e4 6461 goto abort;
851c30c9 6462 /* Don't enable multi-threading by default*/
60aaf933 6463 if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group,
6464 &new_group)) {
6465 conf->group_cnt = group_cnt;
6466 conf->worker_cnt_per_group = worker_cnt_per_group;
6467 conf->worker_groups = new_group;
6468 } else
851c30c9 6469 goto abort;
f5efd45a 6470 spin_lock_init(&conf->device_lock);
c46501b2 6471 seqcount_init(&conf->gen_lock);
2d5b569b 6472 mutex_init(&conf->cache_size_mutex);
b1b46486 6473 init_waitqueue_head(&conf->wait_for_quiescent);
e9e4c377
YL		 6474 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) {
6475 init_waitqueue_head(&conf->wait_for_stripe[i]);
6476 }
f5efd45a
DW		 6477 init_waitqueue_head(&conf->wait_for_overlap);
6478 INIT_LIST_HEAD(&conf->handle_list);
6479 INIT_LIST_HEAD(&conf->hold_list);
6480 INIT_LIST_HEAD(&conf->delayed_list);
6481 INIT_LIST_HEAD(&conf->bitmap_list);
c3cce6cd 6482 bio_list_init(&conf->return_bi);
773ca82f 6483 init_llist_head(&conf->released_stripes);
f5efd45a
DW		 6484 atomic_set(&conf->active_stripes, 0);
6485 atomic_set(&conf->preread_active_stripes, 0);
6486 atomic_set(&conf->active_aligned_reads, 0);
6487 conf->bypass_threshold = BYPASS_THRESHOLD;
d890fa2b 6488 conf->recovery_disabled = mddev->recovery_disabled - 1;
91adb564
N		 6489
6490 conf->raid_disks = mddev->raid_disks;
6491 if (mddev->reshape_position == MaxSector)
6492 conf->previous_raid_disks = mddev->raid_disks;
6493 else
f6705578 6494 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
5e5e3e78 6495 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
f6705578 6496
5e5e3e78 6497 conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
b55e6bfc
N		 6498 GFP_KERNEL);
6499 if (!conf->disks)
6500 goto abort;
9ffae0cf 6501
1da177e4
LT		 6502 conf->mddev = mddev;
6503
fccddba0 6504 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
1da177e4 6505 goto abort;
1da177e4 6506
566c09c5
SL		 6507 /* We init hash_locks[0] separately to that it can be used
6508 * as the reference lock in the spin_lock_nest_lock() call
6509 * in lock_all_device_hash_locks_irq in order to convince
6510 * lockdep that we know what we are doing.
6511 */
6512 spin_lock_init(conf->hash_locks);
6513 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
6514 spin_lock_init(conf->hash_locks + i);
6515
6516 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
6517 INIT_LIST_HEAD(conf->inactive_list + i);
6518
6519 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
6520 INIT_LIST_HEAD(conf->temp_inactive_list + i);
6521
36d1c647 6522 conf->level = mddev->new_level;
46d5b785 6523 conf->chunk_sectors = mddev->new_chunk_sectors;
36d1c647
DW		 6524 if (raid5_alloc_percpu(conf) != 0)
6525 goto abort;
6526
0c55e022 6527 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
1da177e4 6528
dafb20fa 6529 rdev_for_each(rdev, mddev) {
1da177e4 6530 raid_disk = rdev->raid_disk;
5e5e3e78 6531 if (raid_disk >= max_disks
1da177e4
LT		 6532 || raid_disk < 0)
6533 continue;
6534 disk = conf->disks + raid_disk;
6535
17045f52
N		 6536 if (test_bit(Replacement, &rdev->flags)) {
6537 if (disk->replacement)
6538 goto abort;
6539 disk->replacement = rdev;
6540 } else {
6541 if (disk->rdev)
6542 goto abort;
6543 disk->rdev = rdev;
6544 }
1da177e4 6545
b2d444d7 6546 if (test_bit(In_sync, &rdev->flags)) {
1da177e4 6547 char b[BDEVNAME_SIZE];
0c55e022
N		 6548 printk(KERN_INFO "md/raid:%s: device %s operational as raid"
6549 " disk %d\n",
6550 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
d6b212f4 6551 } else if (rdev->saved_raid_disk != raid_disk)
8c2e870a
NB		 6552 /* Cannot rely on bitmap to complete recovery */
6553 conf->fullsync = 1;
1da177e4
LT		 6554 }
6555
91adb564 6556 conf->level = mddev->new_level;
584acdd4 6557 if (conf->level == 6) {
16a53ecc 6558 conf->max_degraded = 2;
584acdd4
MS		 6559 if (raid6_call.xor_syndrome)
6560 conf->rmw_level = PARITY_ENABLE_RMW;
6561 else
6562 conf->rmw_level = PARITY_DISABLE_RMW;
6563 } else {
16a53ecc 6564 conf->max_degraded = 1;
584acdd4
MS		 6565 conf->rmw_level = PARITY_ENABLE_RMW;
6566 }
91adb564 6567 conf->algorithm = mddev->new_layout;
fef9c61f 6568 conf->reshape_progress = mddev->reshape_position;
e183eaed 6569 if (conf->reshape_progress != MaxSector) {
09c9e5fa 6570 conf->prev_chunk_sectors = mddev->chunk_sectors;
e183eaed 6571 conf->prev_algo = mddev->layout;
5cac6bcb
N		 6572 } else {
6573 conf->prev_chunk_sectors = conf->chunk_sectors;
6574 conf->prev_algo = conf->algorithm;
e183eaed 6575 }
1da177e4 6576
edbe83ab
N		 6577 conf->min_nr_stripes = NR_STRIPES;
6578 memory = conf->min_nr_stripes * (sizeof(struct stripe_head) +
5e5e3e78 6579 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
4bda556a 6580 atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS);
edbe83ab 6581 if (grow_stripes(conf, conf->min_nr_stripes)) {
91adb564 6582 printk(KERN_ERR
0c55e022
N		 6583 "md/raid:%s: couldn't allocate %dkB for buffers\n",
6584 mdname(mddev), memory);
91adb564
N		 6585 goto abort;
6586 } else
0c55e022
N		 6587 printk(KERN_INFO "md/raid:%s: allocated %dkB\n",
6588 mdname(mddev), memory);
edbe83ab
N		 6589 /*
6590 * Losing a stripe head costs more than the time to refill it,
6591 * it reduces the queue depth and so can hurt throughput.
6592 * So set it rather large, scaled by number of devices.
6593 */
6594 conf->shrinker.seeks = DEFAULT_SEEKS * conf->raid_disks * 4;
6595 conf->shrinker.scan_objects = raid5_cache_scan;
6596 conf->shrinker.count_objects = raid5_cache_count;
6597 conf->shrinker.batch = 128;
6598 conf->shrinker.flags = 0;
6599 register_shrinker(&conf->shrinker);
1da177e4 6600
0232605d
N		 6601 sprintf(pers_name, "raid%d", mddev->new_level);
6602 conf->thread = md_register_thread(raid5d, mddev, pers_name);
91adb564
N		 6603 if (!conf->thread) {
6604 printk(KERN_ERR
0c55e022 6605 "md/raid:%s: couldn't allocate thread.\n",
91adb564 6606 mdname(mddev));
16a53ecc
N		 6607 goto abort;
6608 }
91adb564
N		 6609
6610 return conf;
6611
6612 abort:
6613 if (conf) {
95fc17aa 6614 free_conf(conf);
91adb564
N		 6615 return ERR_PTR(-EIO);
6616 } else
6617 return ERR_PTR(-ENOMEM);
6618}
6619
c148ffdc
N		 6620static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
6621{
6622 switch (algo) {
6623 case ALGORITHM_PARITY_0:
6624 if (raid_disk < max_degraded)
6625 return 1;
6626 break;
6627 case ALGORITHM_PARITY_N:
6628 if (raid_disk >= raid_disks - max_degraded)
6629 return 1;
6630 break;
6631 case ALGORITHM_PARITY_0_6:
f72ffdd6 6632 if (raid_disk == 0 ||
c148ffdc
N		 6633 raid_disk == raid_disks - 1)
6634 return 1;
6635 break;
6636 case ALGORITHM_LEFT_ASYMMETRIC_6:
6637 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6638 case ALGORITHM_LEFT_SYMMETRIC_6:
6639 case ALGORITHM_RIGHT_SYMMETRIC_6:
6640 if (raid_disk == raid_disks - 1)
6641 return 1;
6642 }
6643 return 0;
6644}
6645
fd01b88c 6646static int run(struct mddev *mddev)
91adb564 6647{
d1688a6d 6648 struct r5conf *conf;
9f7c2220 6649 int working_disks = 0;
c148ffdc 6650 int dirty_parity_disks = 0;
3cb03002 6651 struct md_rdev *rdev;
c148ffdc 6652 sector_t reshape_offset = 0;
17045f52 6653 int i;
b5254dd5
N		 6654 long long min_offset_diff = 0;
6655 int first = 1;
91adb564 6656
8c6ac868 6657 if (mddev->recovery_cp != MaxSector)
0c55e022 6658 printk(KERN_NOTICE "md/raid:%s: not clean"
8c6ac868
AN		 6659 " -- starting background reconstruction\n",
6660 mdname(mddev));
b5254dd5
N		 6661
6662 rdev_for_each(rdev, mddev) {
6663 long long diff;
6664 if (rdev->raid_disk < 0)
6665 continue;
6666 diff = (rdev->new_data_offset - rdev->data_offset);
6667 if (first) {
6668 min_offset_diff = diff;
6669 first = 0;
6670 } else if (mddev->reshape_backwards &&
6671 diff < min_offset_diff)
6672 min_offset_diff = diff;
6673 else if (!mddev->reshape_backwards &&
6674 diff > min_offset_diff)
6675 min_offset_diff = diff;
6676 }
6677
91adb564
N		 6678 if (mddev->reshape_position != MaxSector) {
6679 /* Check that we can continue the reshape.
b5254dd5
N		 6680 * Difficulties arise if the stripe we would write to
6681 * next is at or after the stripe we would read from next.
6682 * For a reshape that changes the number of devices, this
6683 * is only possible for a very short time, and mdadm makes
6684 * sure that time appears to have past before assembling
6685 * the array. So we fail if that time hasn't passed.
6686 * For a reshape that keeps the number of devices the same
6687 * mdadm must be monitoring the reshape can keeping the
6688 * critical areas read-only and backed up. It will start
6689 * the array in read-only mode, so we check for that.
91adb564
N		 6690 */
6691 sector_t here_new, here_old;
6692 int old_disks;
18b00334 6693 int max_degraded = (mddev->level == 6 ? 2 : 1);
05256d98
N		 6694 int chunk_sectors;
6695 int new_data_disks;
91adb564 6696
88ce4930 6697 if (mddev->new_level != mddev->level) {
0c55e022 6698 printk(KERN_ERR "md/raid:%s: unsupported reshape "
91adb564
N		 6699 "required - aborting.\n",
6700 mdname(mddev));
6701 return -EINVAL;
6702 }
91adb564
N		 6703 old_disks = mddev->raid_disks - mddev->delta_disks;
6704 /* reshape_position must be on a new-stripe boundary, and one
6705 * further up in new geometry must map after here in old
6706 * geometry.
05256d98
N		 6707 * If the chunk sizes are different, then as we perform reshape
6708 * in units of the largest of the two, reshape_position needs
6709 * be a multiple of the largest chunk size times new data disks.
91adb564
N		 6710 */
6711 here_new = mddev->reshape_position;
05256d98
N		 6712 chunk_sectors = max(mddev->chunk_sectors, mddev->new_chunk_sectors);
6713 new_data_disks = mddev->raid_disks - max_degraded;
6714 if (sector_div(here_new, chunk_sectors * new_data_disks)) {
0c55e022
N		 6715 printk(KERN_ERR "md/raid:%s: reshape_position not "
6716 "on a stripe boundary\n", mdname(mddev));
91adb564
N		 6717 return -EINVAL;
6718 }
05256d98 6719 reshape_offset = here_new * chunk_sectors;
91adb564
N		 6720 /* here_new is the stripe we will write to */
6721 here_old = mddev->reshape_position;
05256d98 6722 sector_div(here_old, chunk_sectors * (old_disks-max_degraded));
91adb564
N		 6723 /* here_old is the first stripe that we might need to read
6724 * from */
67ac6011
N		 6725 if (mddev->delta_disks == 0) {
6726 /* We cannot be sure it is safe to start an in-place
b5254dd5 6727 * reshape. It is only safe if user-space is monitoring
67ac6011
N		 6728 * and taking constant backups.
6729 * mdadm always starts a situation like this in
6730 * readonly mode so it can take control before
6731 * allowing any writes. So just check for that.
6732 */
b5254dd5
N		 6733 if (abs(min_offset_diff) >= mddev->chunk_sectors &&
6734 abs(min_offset_diff) >= mddev->new_chunk_sectors)
6735 /* not really in-place - so OK */;
6736 else if (mddev->ro == 0) {
6737 printk(KERN_ERR "md/raid:%s: in-place reshape "
6738 "must be started in read-only mode "
6739 "- aborting\n",
0c55e022 6740 mdname(mddev));
67ac6011
N		 6741 return -EINVAL;
6742 }
2c810cdd 6743 } else if (mddev->reshape_backwards
05256d98
N		 6744 ? (here_new * chunk_sectors + min_offset_diff <=
6745 here_old * chunk_sectors)
6746 : (here_new * chunk_sectors >=
6747 here_old * chunk_sectors + (-min_offset_diff))) {
91adb564 6748 /* Reading from the same stripe as writing to - bad */
0c55e022
N		 6749 printk(KERN_ERR "md/raid:%s: reshape_position too early for "
6750 "auto-recovery - aborting.\n",
6751 mdname(mddev));
91adb564
N		 6752 return -EINVAL;
6753 }
0c55e022
N		 6754 printk(KERN_INFO "md/raid:%s: reshape will continue\n",
6755 mdname(mddev));
91adb564
N		 6756 /* OK, we should be able to continue; */
6757 } else {
6758 BUG_ON(mddev->level != mddev->new_level);
6759 BUG_ON(mddev->layout != mddev->new_layout);
664e7c41 6760 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
91adb564 6761 BUG_ON(mddev->delta_disks != 0);
1da177e4 6762 }
91adb564 6763
245f46c2
N		 6764 if (mddev->private == NULL)
6765 conf = setup_conf(mddev);
6766 else
6767 conf = mddev->private;
6768
91adb564
N		 6769 if (IS_ERR(conf))
6770 return PTR_ERR(conf);
6771
b5254dd5 6772 conf->min_offset_diff = min_offset_diff;
91adb564
N		 6773 mddev->thread = conf->thread;
6774 conf->thread = NULL;
6775 mddev->private = conf;
6776
17045f52
N		 6777 for (i = 0; i < conf->raid_disks && conf->previous_raid_disks;
6778 i++) {
6779 rdev = conf->disks[i].rdev;
6780 if (!rdev && conf->disks[i].replacement) {
6781 /* The replacement is all we have yet */
6782 rdev = conf->disks[i].replacement;
6783 conf->disks[i].replacement = NULL;
6784 clear_bit(Replacement, &rdev->flags);
6785 conf->disks[i].rdev = rdev;
6786 }
6787 if (!rdev)
c148ffdc 6788 continue;
17045f52
N		 6789 if (conf->disks[i].replacement &&
6790 conf->reshape_progress != MaxSector) {
6791 /* replacements and reshape simply do not mix. */
6792 printk(KERN_ERR "md: cannot handle concurrent "
6793 "replacement and reshape.\n");
6794 goto abort;
6795 }
2f115882 6796 if (test_bit(In_sync, &rdev->flags)) {
91adb564 6797 working_disks++;
2f115882
N		 6798 continue;
6799 }
c148ffdc
N		 6800 /* This disc is not fully in-sync. However if it
6801 * just stored parity (beyond the recovery_offset),
6802 * when we don't need to be concerned about the
6803 * array being dirty.
6804 * When reshape goes 'backwards', we never have
6805 * partially completed devices, so we only need
6806 * to worry about reshape going forwards.
6807 */
6808 /* Hack because v0.91 doesn't store recovery_offset properly. */
6809 if (mddev->major_version == 0 &&
6810 mddev->minor_version > 90)
6811 rdev->recovery_offset = reshape_offset;
5026d7a9 6812
c148ffdc
N		 6813 if (rdev->recovery_offset < reshape_offset) {
6814 /* We need to check old and new layout */
6815 if (!only_parity(rdev->raid_disk,
6816 conf->algorithm,
6817 conf->raid_disks,
6818 conf->max_degraded))
6819 continue;
6820 }
6821 if (!only_parity(rdev->raid_disk,
6822 conf->prev_algo,
6823 conf->previous_raid_disks,
6824 conf->max_degraded))
6825 continue;
6826 dirty_parity_disks++;
6827 }
91adb564 6828
17045f52
N		 6829 /*
6830 * 0 for a fully functional array, 1 or 2 for a degraded array.
6831 */
908f4fbd 6832 mddev->degraded = calc_degraded(conf);
91adb564 6833
674806d6 6834 if (has_failed(conf)) {
0c55e022 6835 printk(KERN_ERR "md/raid:%s: not enough operational devices"
1da177e4 6836 " (%d/%d failed)\n",
02c2de8c 6837 mdname(mddev), mddev->degraded, conf->raid_disks);
1da177e4
LT		 6838 goto abort;
6839 }
6840
91adb564 6841 /* device size must be a multiple of chunk size */
9d8f0363 6842 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
91adb564
N		 6843 mddev->resync_max_sectors = mddev->dev_sectors;
6844
c148ffdc 6845 if (mddev->degraded > dirty_parity_disks &&
1da177e4 6846 mddev->recovery_cp != MaxSector) {
6ff8d8ec
N		 6847 if (mddev->ok_start_degraded)
6848 printk(KERN_WARNING
0c55e022
N		 6849 "md/raid:%s: starting dirty degraded array"
6850 " - data corruption possible.\n",
6ff8d8ec
N		 6851 mdname(mddev));
6852 else {
6853 printk(KERN_ERR
0c55e022 6854 "md/raid:%s: cannot start dirty degraded array.\n",
6ff8d8ec
N		 6855 mdname(mddev));
6856 goto abort;
6857 }
1da177e4
LT		 6858 }
6859
1da177e4 6860 if (mddev->degraded == 0)
0c55e022
N		 6861 printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d"
6862 " devices, algorithm %d\n", mdname(mddev), conf->level,
e183eaed
N		 6863 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
6864 mddev->new_layout);
1da177e4 6865 else
0c55e022
N		 6866 printk(KERN_ALERT "md/raid:%s: raid level %d active with %d"
6867 " out of %d devices, algorithm %d\n",
6868 mdname(mddev), conf->level,
6869 mddev->raid_disks - mddev->degraded,
6870 mddev->raid_disks, mddev->new_layout);
1da177e4
LT		 6871
6872 print_raid5_conf(conf);
6873
fef9c61f 6874 if (conf->reshape_progress != MaxSector) {
fef9c61f 6875 conf->reshape_safe = conf->reshape_progress;
f6705578
N		 6876 atomic_set(&conf->reshape_stripes, 0);
6877 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6878 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6879 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6880 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6881 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 6882 "reshape");
f6705578
N		 6883 }
6884
1da177e4 6885 /* Ok, everything is just fine now */
a64c876f
N		 6886 if (mddev->to_remove == &raid5_attrs_group)
6887 mddev->to_remove = NULL;
00bcb4ac
N		 6888 else if (mddev->kobj.sd &&
6889 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
5e55e2f5 6890 printk(KERN_WARNING
4a5add49 6891 "raid5: failed to create sysfs attributes for %s\n",
5e55e2f5 6892 mdname(mddev));
4a5add49 6893 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
7a5febe9 6894
4a5add49 6895 if (mddev->queue) {
9f7c2220 6896 int chunk_size;
620125f2 6897 bool discard_supported = true;
4a5add49
N		 6898 /* read-ahead size must cover two whole stripes, which
6899 * is 2 * (datadisks) * chunksize where 'n' is the
6900 * number of raid devices
6901 */
6902 int data_disks = conf->previous_raid_disks - conf->max_degraded;
6903 int stripe = data_disks *
6904 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
6905 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
6906 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
91adb564 6907
9f7c2220
N		 6908 chunk_size = mddev->chunk_sectors << 9;
6909 blk_queue_io_min(mddev->queue, chunk_size);
6910 blk_queue_io_opt(mddev->queue, chunk_size *
6911 (conf->raid_disks - conf->max_degraded));
c78afc62 6912 mddev->queue->limits.raid_partial_stripes_expensive = 1;
620125f2
SL		 6913 /*
6914 * We can only discard a whole stripe. It doesn't make sense to
6915 * discard data disk but write parity disk
6916 */
6917 stripe = stripe * PAGE_SIZE;
4ac6875e
N		 6918 /* Round up to power of 2, as discard handling
6919 * currently assumes that */
6920 while ((stripe-1) & stripe)
6921 stripe = (stripe | (stripe-1)) + 1;
620125f2
SL		 6922 mddev->queue->limits.discard_alignment = stripe;
6923 mddev->queue->limits.discard_granularity = stripe;
6924 /*
6925 * unaligned part of discard request will be ignored, so can't
8e0e99ba 6926 * guarantee discard_zeroes_data
620125f2
SL		 6927 */
6928 mddev->queue->limits.discard_zeroes_data = 0;
8f6c2e4b 6929
5026d7a9
PA		 6930 blk_queue_max_write_same_sectors(mddev->queue, 0);
6931
05616be5 6932 rdev_for_each(rdev, mddev) {
9f7c2220
N		 6933 disk_stack_limits(mddev->gendisk, rdev->bdev,
6934 rdev->data_offset << 9);
05616be5
N		 6935 disk_stack_limits(mddev->gendisk, rdev->bdev,
6936 rdev->new_data_offset << 9);
620125f2
SL		 6937 /*
6938 * discard_zeroes_data is required, otherwise data
6939 * could be lost. Consider a scenario: discard a stripe
6940 * (the stripe could be inconsistent if
6941 * discard_zeroes_data is 0); write one disk of the
6942 * stripe (the stripe could be inconsistent again
6943 * depending on which disks are used to calculate
6944 * parity); the disk is broken; The stripe data of this
6945 * disk is lost.
6946 */
6947 if (!blk_queue_discard(bdev_get_queue(rdev->bdev)) ||
6948 !bdev_get_queue(rdev->bdev)->
6949 limits.discard_zeroes_data)
6950 discard_supported = false;
8e0e99ba
N		 6951 /* Unfortunately, discard_zeroes_data is not currently
6952 * a guarantee - just a hint. So we only allow DISCARD
6953 * if the sysadmin has confirmed that only safe devices
6954 * are in use by setting a module parameter.
6955 */
6956 if (!devices_handle_discard_safely) {
6957 if (discard_supported) {
6958 pr_info("md/raid456: discard support disabled due to uncertainty.\n");
6959 pr_info("Set raid456.devices_handle_discard_safely=Y to override.\n");
6960 }
6961 discard_supported = false;
6962 }
05616be5 6963 }
620125f2
SL		 6964
6965 if (discard_supported &&
6966 mddev->queue->limits.max_discard_sectors >= stripe &&
6967 mddev->queue->limits.discard_granularity >= stripe)
6968 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
6969 mddev->queue);
6970 else
6971 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
6972 mddev->queue);
9f7c2220 6973 }
23032a0e 6974
1da177e4
LT		 6975 return 0;
6976abort:
01f96c0a 6977 md_unregister_thread(&mddev->thread);
e4f869d9
N		 6978 print_raid5_conf(conf);
6979 free_conf(conf);
1da177e4 6980 mddev->private = NULL;
0c55e022 6981 printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev));
1da177e4
LT		 6982 return -EIO;
6983}
6984
afa0f557 6985static void raid5_free(struct mddev *mddev, void *priv)
1da177e4 6986{
afa0f557 6987 struct r5conf *conf = priv;
1da177e4 6988
95fc17aa 6989 free_conf(conf);
a64c876f 6990 mddev->to_remove = &raid5_attrs_group;
1da177e4
LT		 6991}
6992
fd01b88c 6993static void status(struct seq_file *seq, struct mddev *mddev)
1da177e4 6994{
d1688a6d 6995 struct r5conf *conf = mddev->private;
1da177e4
LT		 6996 int i;
6997
9d8f0363 6998 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
3cb5edf4 6999 conf->chunk_sectors / 2, mddev->layout);
02c2de8c 7000 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
1da177e4
LT		 7001 for (i = 0; i < conf->raid_disks; i++)
7002 seq_printf (seq, "%s",
7003 conf->disks[i].rdev &&
b2d444d7 7004 test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_");
1da177e4 7005 seq_printf (seq, "]");
1da177e4
LT		 7006}
7007
d1688a6d 7008static void print_raid5_conf (struct r5conf *conf)
1da177e4
LT		 7009{
7010 int i;
7011 struct disk_info *tmp;
7012
0c55e022 7013 printk(KERN_DEBUG "RAID conf printout:\n");
1da177e4
LT		 7014 if (!conf) {
7015 printk("(conf==NULL)\n");
7016 return;
7017 }
0c55e022
N		 7018 printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level,
7019 conf->raid_disks,
7020 conf->raid_disks - conf->mddev->degraded);
1da177e4
LT		 7021
7022 for (i = 0; i < conf->raid_disks; i++) {
7023 char b[BDEVNAME_SIZE];
7024 tmp = conf->disks + i;
7025 if (tmp->rdev)
0c55e022
N		 7026 printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n",
7027 i, !test_bit(Faulty, &tmp->rdev->flags),
7028 bdevname(tmp->rdev->bdev, b));
1da177e4
LT		 7029 }
7030}
7031
fd01b88c 7032static int raid5_spare_active(struct mddev *mddev)
1da177e4
LT		 7033{
7034 int i;
d1688a6d 7035 struct r5conf *conf = mddev->private;
1da177e4 7036 struct disk_info *tmp;
6b965620
N		 7037 int count = 0;
7038 unsigned long flags;
1da177e4
LT		 7039
7040 for (i = 0; i < conf->raid_disks; i++) {
7041 tmp = conf->disks + i;
dd054fce
N		 7042 if (tmp->replacement
7043 && tmp->replacement->recovery_offset == MaxSector
7044 && !test_bit(Faulty, &tmp->replacement->flags)
7045 && !test_and_set_bit(In_sync, &tmp->replacement->flags)) {
7046 /* Replacement has just become active. */
7047 if (!tmp->rdev
7048 || !test_and_clear_bit(In_sync, &tmp->rdev->flags))
7049 count++;
7050 if (tmp->rdev) {
7051 /* Replaced device not technically faulty,
7052 * but we need to be sure it gets removed
7053 * and never re-added.
7054 */
7055 set_bit(Faulty, &tmp->rdev->flags);
7056 sysfs_notify_dirent_safe(
7057 tmp->rdev->sysfs_state);
7058 }
7059 sysfs_notify_dirent_safe(tmp->replacement->sysfs_state);
7060 } else if (tmp->rdev
70fffd0b 7061 && tmp->rdev->recovery_offset == MaxSector
b2d444d7 7062 && !test_bit(Faulty, &tmp->rdev->flags)
c04be0aa 7063 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
6b965620 7064 count++;
43c73ca4 7065 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
1da177e4
LT		 7066 }
7067 }
6b965620 7068 spin_lock_irqsave(&conf->device_lock, flags);
908f4fbd 7069 mddev->degraded = calc_degraded(conf);
6b965620 7070 spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4 7071 print_raid5_conf(conf);
6b965620 7072 return count;
1da177e4
LT		 7073}
7074
b8321b68 7075static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7076{
d1688a6d 7077 struct r5conf *conf = mddev->private;
1da177e4 7078 int err = 0;
b8321b68 7079 int number = rdev->raid_disk;
657e3e4d 7080 struct md_rdev **rdevp;
1da177e4
LT		 7081 struct disk_info *p = conf->disks + number;
7082
7083 print_raid5_conf(conf);
657e3e4d
N		 7084 if (rdev == p->rdev)
7085 rdevp = &p->rdev;
7086 else if (rdev == p->replacement)
7087 rdevp = &p->replacement;
7088 else
7089 return 0;
7090
7091 if (number >= conf->raid_disks &&
7092 conf->reshape_progress == MaxSector)
7093 clear_bit(In_sync, &rdev->flags);
7094
7095 if (test_bit(In_sync, &rdev->flags) ||
7096 atomic_read(&rdev->nr_pending)) {
7097 err = -EBUSY;
7098 goto abort;
7099 }
7100 /* Only remove non-faulty devices if recovery
7101 * isn't possible.
7102 */
7103 if (!test_bit(Faulty, &rdev->flags) &&
7104 mddev->recovery_disabled != conf->recovery_disabled &&
7105 !has_failed(conf) &&
dd054fce 7106 (!p->replacement || p->replacement == rdev) &&
657e3e4d
N		 7107 number < conf->raid_disks) {
7108 err = -EBUSY;
7109 goto abort;
7110 }
7111 *rdevp = NULL;
7112 synchronize_rcu();
7113 if (atomic_read(&rdev->nr_pending)) {
7114 /* lost the race, try later */
7115 err = -EBUSY;
7116 *rdevp = rdev;
dd054fce
N		 7117 } else if (p->replacement) {
7118 /* We must have just cleared 'rdev' */
7119 p->rdev = p->replacement;
7120 clear_bit(Replacement, &p->replacement->flags);
7121 smp_mb(); /* Make sure other CPUs may see both as identical
7122 * but will never see neither - if they are careful
7123 */
7124 p->replacement = NULL;
7125 clear_bit(WantReplacement, &rdev->flags);
7126 } else
7127 /* We might have just removed the Replacement as faulty-
7128 * clear the bit just in case
7129 */
7130 clear_bit(WantReplacement, &rdev->flags);
1da177e4
LT		 7131abort:
7132
7133 print_raid5_conf(conf);
7134 return err;
7135}
7136
fd01b88c 7137static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7138{
d1688a6d 7139 struct r5conf *conf = mddev->private;
199050ea 7140 int err = -EEXIST;
1da177e4
LT		 7141 int disk;
7142 struct disk_info *p;
6c2fce2e
NB		 7143 int first = 0;
7144 int last = conf->raid_disks - 1;
1da177e4 7145
7f0da59b
N		 7146 if (mddev->recovery_disabled == conf->recovery_disabled)
7147 return -EBUSY;
7148
dc10c643 7149 if (rdev->saved_raid_disk < 0 && has_failed(conf))
1da177e4 7150 /* no point adding a device */
199050ea 7151 return -EINVAL;
1da177e4 7152
6c2fce2e
NB		 7153 if (rdev->raid_disk >= 0)
7154 first = last = rdev->raid_disk;
1da177e4
LT		 7155
7156 /*
16a53ecc
N		 7157 * find the disk ... but prefer rdev->saved_raid_disk
7158 * if possible.
1da177e4 7159 */
16a53ecc 7160 if (rdev->saved_raid_disk >= 0 &&
6c2fce2e 7161 rdev->saved_raid_disk >= first &&
16a53ecc 7162 conf->disks[rdev->saved_raid_disk].rdev == NULL)
5cfb22a1
N		 7163 first = rdev->saved_raid_disk;
7164
7165 for (disk = first; disk <= last; disk++) {
7bfec5f3
N		 7166 p = conf->disks + disk;
7167 if (p->rdev == NULL) {
b2d444d7 7168 clear_bit(In_sync, &rdev->flags);
1da177e4 7169 rdev->raid_disk = disk;
199050ea 7170 err = 0;
72626685
N		 7171 if (rdev->saved_raid_disk != disk)
7172 conf->fullsync = 1;
d6065f7b 7173 rcu_assign_pointer(p->rdev, rdev);
5cfb22a1 7174 goto out;
1da177e4 7175 }
5cfb22a1
N		 7176 }
7177 for (disk = first; disk <= last; disk++) {
7178 p = conf->disks + disk;
7bfec5f3
N		 7179 if (test_bit(WantReplacement, &p->rdev->flags) &&
7180 p->replacement == NULL) {
7181 clear_bit(In_sync, &rdev->flags);
7182 set_bit(Replacement, &rdev->flags);
7183 rdev->raid_disk = disk;
7184 err = 0;
7185 conf->fullsync = 1;
7186 rcu_assign_pointer(p->replacement, rdev);
7187 break;
7188 }
7189 }
5cfb22a1 7190out:
1da177e4 7191 print_raid5_conf(conf);
199050ea 7192 return err;
1da177e4
LT		 7193}
7194
fd01b88c 7195static int raid5_resize(struct mddev *mddev, sector_t sectors)
1da177e4
LT		 7196{
7197 /* no resync is happening, and there is enough space
7198 * on all devices, so we can resize.
7199 * We need to make sure resync covers any new space.
7200 * If the array is shrinking we should possibly wait until
7201 * any io in the removed space completes, but it hardly seems
7202 * worth it.
7203 */
a4a6125a 7204 sector_t newsize;
3cb5edf4
N		 7205 struct r5conf *conf = mddev->private;
7206
7207 sectors &= ~((sector_t)conf->chunk_sectors - 1);
a4a6125a
N		 7208 newsize = raid5_size(mddev, sectors, mddev->raid_disks);
7209 if (mddev->external_size &&
7210 mddev->array_sectors > newsize)
b522adcd 7211 return -EINVAL;
a4a6125a
N		 7212 if (mddev->bitmap) {
7213 int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0);
7214 if (ret)
7215 return ret;
7216 }
7217 md_set_array_sectors(mddev, newsize);
f233ea5c 7218 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 7219 revalidate_disk(mddev->gendisk);
b098636c
N		 7220 if (sectors > mddev->dev_sectors &&
7221 mddev->recovery_cp > mddev->dev_sectors) {
58c0fed4 7222 mddev->recovery_cp = mddev->dev_sectors;
1da177e4
LT		 7223 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7224 }
58c0fed4 7225 mddev->dev_sectors = sectors;
4b5c7ae8 7226 mddev->resync_max_sectors = sectors;
1da177e4
LT		 7227 return 0;
7228}
7229
fd01b88c 7230static int check_stripe_cache(struct mddev *mddev)
01ee22b4
N		 7231{
7232 /* Can only proceed if there are plenty of stripe_heads.
7233 * We need a minimum of one full stripe,, and for sensible progress
7234 * it is best to have about 4 times that.
7235 * If we require 4 times, then the default 256 4K stripe_heads will
7236 * allow for chunk sizes up to 256K, which is probably OK.
7237 * If the chunk size is greater, user-space should request more
7238 * stripe_heads first.
7239 */
d1688a6d 7240 struct r5conf *conf = mddev->private;
01ee22b4 7241 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
edbe83ab 7242 > conf->min_nr_stripes ||
01ee22b4 7243 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
edbe83ab 7244 > conf->min_nr_stripes) {
0c55e022
N		 7245 printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n",
7246 mdname(mddev),
01ee22b4
N		 7247 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
7248 / STRIPE_SIZE)*4);
7249 return 0;
7250 }
7251 return 1;
7252}
7253
fd01b88c 7254static int check_reshape(struct mddev *mddev)
29269553 7255{
d1688a6d 7256 struct r5conf *conf = mddev->private;
29269553 7257
88ce4930
N		 7258 if (mddev->delta_disks == 0 &&
7259 mddev->new_layout == mddev->layout &&
664e7c41 7260 mddev->new_chunk_sectors == mddev->chunk_sectors)
50ac168a 7261 return 0; /* nothing to do */
674806d6 7262 if (has_failed(conf))
ec32a2bd 7263 return -EINVAL;
fdcfbbb6 7264 if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) {
ec32a2bd
N		 7265 /* We might be able to shrink, but the devices must
7266 * be made bigger first.
7267 * For raid6, 4 is the minimum size.
7268 * Otherwise 2 is the minimum
7269 */
7270 int min = 2;
7271 if (mddev->level == 6)
7272 min = 4;
7273 if (mddev->raid_disks + mddev->delta_disks < min)
7274 return -EINVAL;
7275 }
29269553 7276
01ee22b4 7277 if (!check_stripe_cache(mddev))
29269553 7278 return -ENOSPC;
29269553 7279
738a2738
N		 7280 if (mddev->new_chunk_sectors > mddev->chunk_sectors ||
7281 mddev->delta_disks > 0)
7282 if (resize_chunks(conf,
7283 conf->previous_raid_disks
7284 + max(0, mddev->delta_disks),
7285 max(mddev->new_chunk_sectors,
7286 mddev->chunk_sectors)
7287 ) < 0)
7288 return -ENOMEM;
e56108d6
N		 7289 return resize_stripes(conf, (conf->previous_raid_disks
7290 + mddev->delta_disks));
63c70c4f
N		 7291}
7292
fd01b88c 7293static int raid5_start_reshape(struct mddev *mddev)
63c70c4f 7294{
d1688a6d 7295 struct r5conf *conf = mddev->private;
3cb03002 7296 struct md_rdev *rdev;
63c70c4f 7297 int spares = 0;
c04be0aa 7298 unsigned long flags;
63c70c4f 7299
f416885e 7300 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
63c70c4f
N		 7301 return -EBUSY;
7302
01ee22b4
N		 7303 if (!check_stripe_cache(mddev))
7304 return -ENOSPC;
7305
30b67645
N		 7306 if (has_failed(conf))
7307 return -EINVAL;
7308
c6563a8c 7309 rdev_for_each(rdev, mddev) {
469518a3
N		 7310 if (!test_bit(In_sync, &rdev->flags)
7311 && !test_bit(Faulty, &rdev->flags))
29269553 7312 spares++;
c6563a8c 7313 }
63c70c4f 7314
f416885e 7315 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
29269553
N		 7316 /* Not enough devices even to make a degraded array
7317 * of that size
7318 */
7319 return -EINVAL;
7320
ec32a2bd
N		 7321 /* Refuse to reduce size of the array. Any reductions in
7322 * array size must be through explicit setting of array_size
7323 * attribute.
7324 */
7325 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
7326 < mddev->array_sectors) {
0c55e022 7327 printk(KERN_ERR "md/raid:%s: array size must be reduced "
ec32a2bd
N		 7328 "before number of disks\n", mdname(mddev));
7329 return -EINVAL;
7330 }
7331
f6705578 7332 atomic_set(&conf->reshape_stripes, 0);
29269553 7333 spin_lock_irq(&conf->device_lock);
c46501b2 7334 write_seqcount_begin(&conf->gen_lock);
29269553 7335 conf->previous_raid_disks = conf->raid_disks;
63c70c4f 7336 conf->raid_disks += mddev->delta_disks;
09c9e5fa
AN		 7337 conf->prev_chunk_sectors = conf->chunk_sectors;
7338 conf->chunk_sectors = mddev->new_chunk_sectors;
88ce4930
N		 7339 conf->prev_algo = conf->algorithm;
7340 conf->algorithm = mddev->new_layout;
05616be5
N		 7341 conf->generation++;
7342 /* Code that selects data_offset needs to see the generation update
7343 * if reshape_progress has been set - so a memory barrier needed.
7344 */
7345 smp_mb();
2c810cdd 7346 if (mddev->reshape_backwards)
fef9c61f
N		 7347 conf->reshape_progress = raid5_size(mddev, 0, 0);
7348 else
7349 conf->reshape_progress = 0;
7350 conf->reshape_safe = conf->reshape_progress;
c46501b2 7351 write_seqcount_end(&conf->gen_lock);
29269553
N		 7352 spin_unlock_irq(&conf->device_lock);
7353
4d77e3ba
N		 7354 /* Now make sure any requests that proceeded on the assumption
7355 * the reshape wasn't running - like Discard or Read - have
7356 * completed.
7357 */
7358 mddev_suspend(mddev);
7359 mddev_resume(mddev);
7360
29269553
N		 7361 /* Add some new drives, as many as will fit.
7362 * We know there are enough to make the newly sized array work.
3424bf6a
N		 7363 * Don't add devices if we are reducing the number of
7364 * devices in the array. This is because it is not possible
7365 * to correctly record the "partially reconstructed" state of
7366 * such devices during the reshape and confusion could result.
29269553 7367 */
87a8dec9 7368 if (mddev->delta_disks >= 0) {
dafb20fa 7369 rdev_for_each(rdev, mddev)
87a8dec9
N		 7370 if (rdev->raid_disk < 0 &&
7371 !test_bit(Faulty, &rdev->flags)) {
7372 if (raid5_add_disk(mddev, rdev) == 0) {
87a8dec9 7373 if (rdev->raid_disk
9d4c7d87 7374 >= conf->previous_raid_disks)
87a8dec9 7375 set_bit(In_sync, &rdev->flags);
9d4c7d87 7376 else
87a8dec9 7377 rdev->recovery_offset = 0;
36fad858
NK		 7378
7379 if (sysfs_link_rdev(mddev, rdev))
87a8dec9 7380 /* Failure here is OK */;
50da0840 7381 }
87a8dec9
N		 7382 } else if (rdev->raid_disk >= conf->previous_raid_disks
7383 && !test_bit(Faulty, &rdev->flags)) {
7384 /* This is a spare that was manually added */
7385 set_bit(In_sync, &rdev->flags);
87a8dec9 7386 }
29269553 7387
87a8dec9
N		 7388 /* When a reshape changes the number of devices,
7389 * ->degraded is measured against the larger of the
7390 * pre and post number of devices.
7391 */
ec32a2bd 7392 spin_lock_irqsave(&conf->device_lock, flags);
908f4fbd 7393 mddev->degraded = calc_degraded(conf);
ec32a2bd
N		 7394 spin_unlock_irqrestore(&conf->device_lock, flags);
7395 }
63c70c4f 7396 mddev->raid_disks = conf->raid_disks;
e516402c 7397 mddev->reshape_position = conf->reshape_progress;
850b2b42 7398 set_bit(MD_CHANGE_DEVS, &mddev->flags);
f6705578 7399
29269553
N		 7400 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7401 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
ea358cd0 7402 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
29269553
N		 7403 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7404 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7405 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 7406 "reshape");
29269553
N		 7407 if (!mddev->sync_thread) {
7408 mddev->recovery = 0;
7409 spin_lock_irq(&conf->device_lock);
ba8805b9 7410 write_seqcount_begin(&conf->gen_lock);
29269553 7411 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
ba8805b9
N		 7412 mddev->new_chunk_sectors =
7413 conf->chunk_sectors = conf->prev_chunk_sectors;
7414 mddev->new_layout = conf->algorithm = conf->prev_algo;
05616be5
N		 7415 rdev_for_each(rdev, mddev)
7416 rdev->new_data_offset = rdev->data_offset;
7417 smp_wmb();
ba8805b9 7418 conf->generation --;
fef9c61f 7419 conf->reshape_progress = MaxSector;
1e3fa9bd 7420 mddev->reshape_position = MaxSector;
ba8805b9 7421 write_seqcount_end(&conf->gen_lock);
29269553
N		 7422 spin_unlock_irq(&conf->device_lock);
7423 return -EAGAIN;
7424 }
c8f517c4 7425 conf->reshape_checkpoint = jiffies;
29269553
N		 7426 md_wakeup_thread(mddev->sync_thread);
7427 md_new_event(mddev);
7428 return 0;
7429}
29269553 7430
ec32a2bd
N		 7431/* This is called from the reshape thread and should make any
7432 * changes needed in 'conf'
7433 */
d1688a6d 7434static void end_reshape(struct r5conf *conf)
29269553 7435{
29269553 7436
f6705578 7437 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
05616be5 7438 struct md_rdev *rdev;
f6705578 7439
f6705578 7440 spin_lock_irq(&conf->device_lock);
cea9c228 7441 conf->previous_raid_disks = conf->raid_disks;
05616be5
N		 7442 rdev_for_each(rdev, conf->mddev)
7443 rdev->data_offset = rdev->new_data_offset;
7444 smp_wmb();
fef9c61f 7445 conf->reshape_progress = MaxSector;
6cbd8148 7446 conf->mddev->reshape_position = MaxSector;
f6705578 7447 spin_unlock_irq(&conf->device_lock);
b0f9ec04 7448 wake_up(&conf->wait_for_overlap);
16a53ecc
N		 7449
7450 /* read-ahead size must cover two whole stripes, which is
7451 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
7452 */
4a5add49 7453 if (conf->mddev->queue) {
cea9c228 7454 int data_disks = conf->raid_disks - conf->max_degraded;
09c9e5fa 7455 int stripe = data_disks * ((conf->chunk_sectors << 9)
cea9c228 7456 / PAGE_SIZE);
16a53ecc
N		 7457 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
7458 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
7459 }
29269553 7460 }
29269553
N		 7461}
7462
ec32a2bd
N		 7463/* This is called from the raid5d thread with mddev_lock held.
7464 * It makes config changes to the device.
7465 */
fd01b88c 7466static void raid5_finish_reshape(struct mddev *mddev)
cea9c228 7467{
d1688a6d 7468 struct r5conf *conf = mddev->private;
cea9c228
N		 7469
7470 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7471
ec32a2bd
N		 7472 if (mddev->delta_disks > 0) {
7473 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
7474 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 7475 revalidate_disk(mddev->gendisk);
ec32a2bd
N		 7476 } else {
7477 int d;
908f4fbd
N		 7478 spin_lock_irq(&conf->device_lock);
7479 mddev->degraded = calc_degraded(conf);
7480 spin_unlock_irq(&conf->device_lock);
ec32a2bd
N		 7481 for (d = conf->raid_disks ;
7482 d < conf->raid_disks - mddev->delta_disks;
1a67dde0 7483 d++) {
3cb03002 7484 struct md_rdev *rdev = conf->disks[d].rdev;
da7613b8
N		 7485 if (rdev)
7486 clear_bit(In_sync, &rdev->flags);
7487 rdev = conf->disks[d].replacement;
7488 if (rdev)
7489 clear_bit(In_sync, &rdev->flags);
1a67dde0 7490 }
cea9c228 7491 }
88ce4930 7492 mddev->layout = conf->algorithm;
09c9e5fa 7493 mddev->chunk_sectors = conf->chunk_sectors;
ec32a2bd
N		 7494 mddev->reshape_position = MaxSector;
7495 mddev->delta_disks = 0;
2c810cdd 7496 mddev->reshape_backwards = 0;
cea9c228
N		 7497 }
7498}
7499
fd01b88c 7500static void raid5_quiesce(struct mddev *mddev, int state)
72626685 7501{
d1688a6d 7502 struct r5conf *conf = mddev->private;
72626685
N		 7503
7504 switch(state) {
e464eafd
N		 7505 case 2: /* resume for a suspend */
7506 wake_up(&conf->wait_for_overlap);
7507 break;
7508
72626685 7509 case 1: /* stop all writes */
566c09c5 7510 lock_all_device_hash_locks_irq(conf);
64bd660b
N		 7511 /* '2' tells resync/reshape to pause so that all
7512 * active stripes can drain
7513 */
7514 conf->quiesce = 2;
b1b46486 7515 wait_event_cmd(conf->wait_for_quiescent,
46031f9a
RBJ		 7516 atomic_read(&conf->active_stripes) == 0 &&
7517 atomic_read(&conf->active_aligned_reads) == 0,
566c09c5
SL		 7518 unlock_all_device_hash_locks_irq(conf),
7519 lock_all_device_hash_locks_irq(conf));
64bd660b 7520 conf->quiesce = 1;
566c09c5 7521 unlock_all_device_hash_locks_irq(conf);
64bd660b
N		 7522 /* allow reshape to continue */
7523 wake_up(&conf->wait_for_overlap);
72626685
N		 7524 break;
7525
7526 case 0: /* re-enable writes */
566c09c5 7527 lock_all_device_hash_locks_irq(conf);
72626685 7528 conf->quiesce = 0;
b1b46486 7529 wake_up(&conf->wait_for_quiescent);
e464eafd 7530 wake_up(&conf->wait_for_overlap);
566c09c5 7531 unlock_all_device_hash_locks_irq(conf);
72626685
N		 7532 break;
7533 }
72626685 7534}
b15c2e57 7535
fd01b88c 7536static void *raid45_takeover_raid0(struct mddev *mddev, int level)
54071b38 7537{
e373ab10 7538 struct r0conf *raid0_conf = mddev->private;
d76c8420 7539 sector_t sectors;
54071b38 7540
f1b29bca 7541 /* for raid0 takeover only one zone is supported */
e373ab10 7542 if (raid0_conf->nr_strip_zones > 1) {
0c55e022
N		 7543 printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n",
7544 mdname(mddev));
f1b29bca
DW		 7545 return ERR_PTR(-EINVAL);
7546 }
7547
e373ab10
N		 7548 sectors = raid0_conf->strip_zone[0].zone_end;
7549 sector_div(sectors, raid0_conf->strip_zone[0].nb_dev);
3b71bd93 7550 mddev->dev_sectors = sectors;
f1b29bca 7551 mddev->new_level = level;
54071b38
TM		 7552 mddev->new_layout = ALGORITHM_PARITY_N;
7553 mddev->new_chunk_sectors = mddev->chunk_sectors;
7554 mddev->raid_disks += 1;
7555 mddev->delta_disks = 1;
7556 /* make sure it will be not marked as dirty */
7557 mddev->recovery_cp = MaxSector;
7558
7559 return setup_conf(mddev);
7560}
7561
fd01b88c 7562static void *raid5_takeover_raid1(struct mddev *mddev)
d562b0c4
N		 7563{
7564 int chunksect;
7565
7566 if (mddev->raid_disks != 2 ||
7567 mddev->degraded > 1)
7568 return ERR_PTR(-EINVAL);
7569
7570 /* Should check if there are write-behind devices? */
7571
7572 chunksect = 64*2; /* 64K by default */
7573
7574 /* The array must be an exact multiple of chunksize */
7575 while (chunksect && (mddev->array_sectors & (chunksect-1)))
7576 chunksect >>= 1;
7577
7578 if ((chunksect<<9) < STRIPE_SIZE)
7579 /* array size does not allow a suitable chunk size */
7580 return ERR_PTR(-EINVAL);
7581
7582 mddev->new_level = 5;
7583 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
664e7c41 7584 mddev->new_chunk_sectors = chunksect;
d562b0c4
N		 7585
7586 return setup_conf(mddev);
7587}
7588
fd01b88c 7589static void *raid5_takeover_raid6(struct mddev *mddev)
fc9739c6
N		 7590{
7591 int new_layout;
7592
7593 switch (mddev->layout) {
7594 case ALGORITHM_LEFT_ASYMMETRIC_6:
7595 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
7596 break;
7597 case ALGORITHM_RIGHT_ASYMMETRIC_6:
7598 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
7599 break;
7600 case ALGORITHM_LEFT_SYMMETRIC_6:
7601 new_layout = ALGORITHM_LEFT_SYMMETRIC;
7602 break;
7603 case ALGORITHM_RIGHT_SYMMETRIC_6:
7604 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
7605 break;
7606 case ALGORITHM_PARITY_0_6:
7607 new_layout = ALGORITHM_PARITY_0;
7608 break;
7609 case ALGORITHM_PARITY_N:
7610 new_layout = ALGORITHM_PARITY_N;
7611 break;
7612 default:
7613 return ERR_PTR(-EINVAL);
7614 }
7615 mddev->new_level = 5;
7616 mddev->new_layout = new_layout;
7617 mddev->delta_disks = -1;
7618 mddev->raid_disks -= 1;
7619 return setup_conf(mddev);
7620}
7621
fd01b88c 7622static int raid5_check_reshape(struct mddev *mddev)
b3546035 7623{
88ce4930
N		 7624 /* For a 2-drive array, the layout and chunk size can be changed
7625 * immediately as not restriping is needed.
7626 * For larger arrays we record the new value - after validation
7627 * to be used by a reshape pass.
b3546035 7628 */
d1688a6d 7629 struct r5conf *conf = mddev->private;
597a711b 7630 int new_chunk = mddev->new_chunk_sectors;
b3546035 7631
597a711b 7632 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
b3546035
N		 7633 return -EINVAL;
7634 if (new_chunk > 0) {
0ba459d2 7635 if (!is_power_of_2(new_chunk))
b3546035 7636 return -EINVAL;
597a711b 7637 if (new_chunk < (PAGE_SIZE>>9))
b3546035 7638 return -EINVAL;
597a711b 7639 if (mddev->array_sectors & (new_chunk-1))
b3546035
N		 7640 /* not factor of array size */
7641 return -EINVAL;
7642 }
7643
7644 /* They look valid */
7645
88ce4930 7646 if (mddev->raid_disks == 2) {
597a711b
N		 7647 /* can make the change immediately */
7648 if (mddev->new_layout >= 0) {
7649 conf->algorithm = mddev->new_layout;
7650 mddev->layout = mddev->new_layout;
88ce4930
N		 7651 }
7652 if (new_chunk > 0) {
597a711b
N		 7653 conf->chunk_sectors = new_chunk ;
7654 mddev->chunk_sectors = new_chunk;
88ce4930
N		 7655 }
7656 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7657 md_wakeup_thread(mddev->thread);
b3546035 7658 }
50ac168a 7659 return check_reshape(mddev);
88ce4930
N		 7660}
7661
fd01b88c 7662static int raid6_check_reshape(struct mddev *mddev)
88ce4930 7663{
597a711b 7664 int new_chunk = mddev->new_chunk_sectors;
50ac168a 7665
597a711b 7666 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
88ce4930 7667 return -EINVAL;
b3546035 7668 if (new_chunk > 0) {
0ba459d2 7669 if (!is_power_of_2(new_chunk))
88ce4930 7670 return -EINVAL;
597a711b 7671 if (new_chunk < (PAGE_SIZE >> 9))
88ce4930 7672 return -EINVAL;
597a711b 7673 if (mddev->array_sectors & (new_chunk-1))
88ce4930
N		 7674 /* not factor of array size */
7675 return -EINVAL;
b3546035 7676 }
88ce4930
N		 7677
7678 /* They look valid */
50ac168a 7679 return check_reshape(mddev);
b3546035
N		 7680}
7681
fd01b88c 7682static void *raid5_takeover(struct mddev *mddev)
d562b0c4
N		 7683{
7684 /* raid5 can take over:
f1b29bca 7685 * raid0 - if there is only one strip zone - make it a raid4 layout
d562b0c4
N		 7686 * raid1 - if there are two drives. We need to know the chunk size
7687 * raid4 - trivial - just use a raid4 layout.
7688 * raid6 - Providing it is a *_6 layout
d562b0c4 7689 */
f1b29bca
DW		 7690 if (mddev->level == 0)
7691 return raid45_takeover_raid0(mddev, 5);
d562b0c4
N		 7692 if (mddev->level == 1)
7693 return raid5_takeover_raid1(mddev);
e9d4758f
N		 7694 if (mddev->level == 4) {
7695 mddev->new_layout = ALGORITHM_PARITY_N;
7696 mddev->new_level = 5;
7697 return setup_conf(mddev);
7698 }
fc9739c6
N		 7699 if (mddev->level == 6)
7700 return raid5_takeover_raid6(mddev);
d562b0c4
N		 7701
7702 return ERR_PTR(-EINVAL);
7703}
7704
fd01b88c 7705static void *raid4_takeover(struct mddev *mddev)
a78d38a1 7706{
f1b29bca
DW		 7707 /* raid4 can take over:
7708 * raid0 - if there is only one strip zone
7709 * raid5 - if layout is right
a78d38a1 7710 */
f1b29bca
DW		 7711 if (mddev->level == 0)
7712 return raid45_takeover_raid0(mddev, 4);
a78d38a1
N		 7713 if (mddev->level == 5 &&
7714 mddev->layout == ALGORITHM_PARITY_N) {
7715 mddev->new_layout = 0;
7716 mddev->new_level = 4;
7717 return setup_conf(mddev);
7718 }
7719 return ERR_PTR(-EINVAL);
7720}
d562b0c4 7721
84fc4b56 7722static struct md_personality raid5_personality;
245f46c2 7723
fd01b88c 7724static void *raid6_takeover(struct mddev *mddev)
245f46c2
N		 7725{
7726 /* Currently can only take over a raid5. We map the
7727 * personality to an equivalent raid6 personality
7728 * with the Q block at the end.
7729 */
7730 int new_layout;
7731
7732 if (mddev->pers != &raid5_personality)
7733 return ERR_PTR(-EINVAL);
7734 if (mddev->degraded > 1)
7735 return ERR_PTR(-EINVAL);
7736 if (mddev->raid_disks > 253)
7737 return ERR_PTR(-EINVAL);
7738 if (mddev->raid_disks < 3)
7739 return ERR_PTR(-EINVAL);
7740
7741 switch (mddev->layout) {
7742 case ALGORITHM_LEFT_ASYMMETRIC:
7743 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
7744 break;
7745 case ALGORITHM_RIGHT_ASYMMETRIC:
7746 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
7747 break;
7748 case ALGORITHM_LEFT_SYMMETRIC:
7749 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
7750 break;
7751 case ALGORITHM_RIGHT_SYMMETRIC:
7752 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
7753 break;
7754 case ALGORITHM_PARITY_0:
7755 new_layout = ALGORITHM_PARITY_0_6;
7756 break;
7757 case ALGORITHM_PARITY_N:
7758 new_layout = ALGORITHM_PARITY_N;
7759 break;
7760 default:
7761 return ERR_PTR(-EINVAL);
7762 }
7763 mddev->new_level = 6;
7764 mddev->new_layout = new_layout;
7765 mddev->delta_disks = 1;
7766 mddev->raid_disks += 1;
7767 return setup_conf(mddev);
7768}
7769
84fc4b56 7770static struct md_personality raid6_personality =
16a53ecc
N		 7771{
7772 .name = "raid6",
7773 .level = 6,
7774 .owner = THIS_MODULE,
7775 .make_request = make_request,
7776 .run = run,
afa0f557 7777 .free = raid5_free,
16a53ecc
N		 7778 .status = status,
7779 .error_handler = error,
7780 .hot_add_disk = raid5_add_disk,
7781 .hot_remove_disk= raid5_remove_disk,
7782 .spare_active = raid5_spare_active,
7783 .sync_request = sync_request,
7784 .resize = raid5_resize,
80c3a6ce 7785 .size = raid5_size,
50ac168a 7786 .check_reshape = raid6_check_reshape,
f416885e 7787 .start_reshape = raid5_start_reshape,
cea9c228 7788 .finish_reshape = raid5_finish_reshape,
16a53ecc 7789 .quiesce = raid5_quiesce,
245f46c2 7790 .takeover = raid6_takeover,
5c675f83 7791 .congested = raid5_congested,
16a53ecc 7792};
84fc4b56 7793static struct md_personality raid5_personality =
1da177e4
LT		 7794{
7795 .name = "raid5",
2604b703 7796 .level = 5,
1da177e4
LT		 7797 .owner = THIS_MODULE,
7798 .make_request = make_request,
7799 .run = run,
afa0f557 7800 .free = raid5_free,
1da177e4
LT		 7801 .status = status,
7802 .error_handler = error,
7803 .hot_add_disk = raid5_add_disk,
7804 .hot_remove_disk= raid5_remove_disk,
7805 .spare_active = raid5_spare_active,
7806 .sync_request = sync_request,
7807 .resize = raid5_resize,
80c3a6ce 7808 .size = raid5_size,
63c70c4f
N		 7809 .check_reshape = raid5_check_reshape,
7810 .start_reshape = raid5_start_reshape,
cea9c228 7811 .finish_reshape = raid5_finish_reshape,
72626685 7812 .quiesce = raid5_quiesce,
d562b0c4 7813 .takeover = raid5_takeover,
5c675f83 7814 .congested = raid5_congested,
1da177e4
LT		 7815};
7816
84fc4b56 7817static struct md_personality raid4_personality =
1da177e4 7818{
2604b703
N		 7819 .name = "raid4",
7820 .level = 4,
7821 .owner = THIS_MODULE,
7822 .make_request = make_request,
7823 .run = run,
afa0f557 7824 .free = raid5_free,
2604b703
N		 7825 .status = status,
7826 .error_handler = error,
7827 .hot_add_disk = raid5_add_disk,
7828 .hot_remove_disk= raid5_remove_disk,
7829 .spare_active = raid5_spare_active,
7830 .sync_request = sync_request,
7831 .resize = raid5_resize,
80c3a6ce 7832 .size = raid5_size,
3d37890b
N		 7833 .check_reshape = raid5_check_reshape,
7834 .start_reshape = raid5_start_reshape,
cea9c228 7835 .finish_reshape = raid5_finish_reshape,
2604b703 7836 .quiesce = raid5_quiesce,
a78d38a1 7837 .takeover = raid4_takeover,
5c675f83 7838 .congested = raid5_congested,
2604b703
N		 7839};
7840
7841static int __init raid5_init(void)
7842{
851c30c9
SL		 7843 raid5_wq = alloc_workqueue("raid5wq",
7844 WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0);
7845 if (!raid5_wq)
7846 return -ENOMEM;
16a53ecc 7847 register_md_personality(&raid6_personality);
2604b703
N		 7848 register_md_personality(&raid5_personality);
7849 register_md_personality(&raid4_personality);
7850 return 0;
1da177e4
LT		 7851}
7852
2604b703 7853static void raid5_exit(void)
1da177e4 7854{
16a53ecc 7855 unregister_md_personality(&raid6_personality);
2604b703
N		 7856 unregister_md_personality(&raid5_personality);
7857 unregister_md_personality(&raid4_personality);
851c30c9 7858 destroy_workqueue(raid5_wq);
1da177e4
LT		 7859}
7860
7861module_init(raid5_init);
7862module_exit(raid5_exit);
7863MODULE_LICENSE("GPL");
0efb9e61 7864MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
1da177e4 7865MODULE_ALIAS("md-personality-4"); /* RAID5 */
d9d166c2
N		 7866MODULE_ALIAS("md-raid5");
7867MODULE_ALIAS("md-raid4");
2604b703
N		 7868MODULE_ALIAS("md-level-5");
7869MODULE_ALIAS("md-level-4");
16a53ecc
N		 7870MODULE_ALIAS("md-personality-8"); /* RAID6 */
7871MODULE_ALIAS("md-raid6");
7872MODULE_ALIAS("md-level-6");
7873
7874/* This used to be two separate modules, they were: */
7875MODULE_ALIAS("raid5");
7876MODULE_ALIAS("raid6");
Linux 6.x block layer and io_uring tree(s)