projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md: check command validity early in md_ioctl().
[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>
a9add5d9
N		 57#include <trace/events/block.h>
58
43b2e5d8 59#include "md.h"
bff61975 60#include "raid5.h"
54071b38 61#include "raid0.h"
ef740c37 62#include "bitmap.h"
72626685 63
851c30c9
SL		 64#define cpu_to_group(cpu) cpu_to_node(cpu)
65#define ANY_GROUP NUMA_NO_NODE
66
67static struct workqueue_struct *raid5_wq;
1da177e4
LT		 68/*
69 * Stripe cache
70 */
71
72#define NR_STRIPES 256
73#define STRIPE_SIZE PAGE_SIZE
74#define STRIPE_SHIFT (PAGE_SHIFT - 9)
75#define STRIPE_SECTORS (STRIPE_SIZE>>9)
76#define IO_THRESHOLD 1
8b3e6cdc 77#define BYPASS_THRESHOLD 1
fccddba0 78#define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
1da177e4 79#define HASH_MASK (NR_HASH - 1)
bfc90cb0 80#define MAX_STRIPE_BATCH 8
1da177e4 81
d1688a6d 82static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect)
db298e19
N		 83{
84 int hash = (sect >> STRIPE_SHIFT) & HASH_MASK;
85 return &conf->stripe_hashtbl[hash];
86}
1da177e4 87
566c09c5
SL		 88static inline int stripe_hash_locks_hash(sector_t sect)
89{
90 return (sect >> STRIPE_SHIFT) & STRIPE_HASH_LOCKS_MASK;
91}
92
93static inline void lock_device_hash_lock(struct r5conf *conf, int hash)
94{
95 spin_lock_irq(conf->hash_locks + hash);
96 spin_lock(&conf->device_lock);
97}
98
99static inline void unlock_device_hash_lock(struct r5conf *conf, int hash)
100{
101 spin_unlock(&conf->device_lock);
102 spin_unlock_irq(conf->hash_locks + hash);
103}
104
105static inline void lock_all_device_hash_locks_irq(struct r5conf *conf)
106{
107 int i;
108 local_irq_disable();
109 spin_lock(conf->hash_locks);
110 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
111 spin_lock_nest_lock(conf->hash_locks + i, conf->hash_locks);
112 spin_lock(&conf->device_lock);
113}
114
115static inline void unlock_all_device_hash_locks_irq(struct r5conf *conf)
116{
117 int i;
118 spin_unlock(&conf->device_lock);
119 for (i = NR_STRIPE_HASH_LOCKS; i; i--)
120 spin_unlock(conf->hash_locks + i - 1);
121 local_irq_enable();
122}
123
1da177e4
LT		 124/* bio's attached to a stripe+device for I/O are linked together in bi_sector
125 * order without overlap. There may be several bio's per stripe+device, and
126 * a bio could span several devices.
127 * When walking this list for a particular stripe+device, we must never proceed
128 * beyond a bio that extends past this device, as the next bio might no longer
129 * be valid.
db298e19 130 * This function is used to determine the 'next' bio in the list, given the sector
1da177e4
LT		 131 * of the current stripe+device
132 */
db298e19
N		 133static inline struct bio *r5_next_bio(struct bio *bio, sector_t sector)
134{
aa8b57aa 135 int sectors = bio_sectors(bio);
db298e19
N		 136 if (bio->bi_sector + sectors < sector + STRIPE_SECTORS)
137 return bio->bi_next;
138 else
139 return NULL;
140}
1da177e4 141
960e739d 142/*
5b99c2ff
JA		 143 * We maintain a biased count of active stripes in the bottom 16 bits of
144 * bi_phys_segments, and a count of processed stripes in the upper 16 bits
960e739d 145 */
e7836bd6 146static inline int raid5_bi_processed_stripes(struct bio *bio)
960e739d 147{
e7836bd6
SL		 148 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
149 return (atomic_read(segments) >> 16) & 0xffff;
960e739d
JA		 150}
151
e7836bd6 152static inline int raid5_dec_bi_active_stripes(struct bio *bio)
960e739d 153{
e7836bd6
SL		 154 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
155 return atomic_sub_return(1, segments) & 0xffff;
960e739d
JA		 156}
157
e7836bd6 158static inline void raid5_inc_bi_active_stripes(struct bio *bio)
960e739d 159{
e7836bd6
SL		 160 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
161 atomic_inc(segments);
960e739d
JA		 162}
163
e7836bd6
SL		 164static inline void raid5_set_bi_processed_stripes(struct bio *bio,
165 unsigned int cnt)
960e739d 166{
e7836bd6
SL		 167 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
168 int old, new;
960e739d 169
e7836bd6
SL		 170 do {
171 old = atomic_read(segments);
172 new = (old & 0xffff) | (cnt << 16);
173 } while (atomic_cmpxchg(segments, old, new) != old);
960e739d
JA		 174}
175
e7836bd6 176static inline void raid5_set_bi_stripes(struct bio *bio, unsigned int cnt)
960e739d 177{
e7836bd6
SL		 178 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
179 atomic_set(segments, cnt);
960e739d
JA		 180}
181
d0dabf7e
N		 182/* Find first data disk in a raid6 stripe */
183static inline int raid6_d0(struct stripe_head *sh)
184{
67cc2b81
N		 185 if (sh->ddf_layout)
186 /* ddf always start from first device */
187 return 0;
188 /* md starts just after Q block */
d0dabf7e
N		 189 if (sh->qd_idx == sh->disks - 1)
190 return 0;
191 else
192 return sh->qd_idx + 1;
193}
16a53ecc
N		 194static inline int raid6_next_disk(int disk, int raid_disks)
195{
196 disk++;
197 return (disk < raid_disks) ? disk : 0;
198}
a4456856 199
d0dabf7e
N		 200/* When walking through the disks in a raid5, starting at raid6_d0,
201 * We need to map each disk to a 'slot', where the data disks are slot
202 * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk
203 * is raid_disks-1. This help does that mapping.
204 */
67cc2b81
N		 205static int raid6_idx_to_slot(int idx, struct stripe_head *sh,
206 int *count, int syndrome_disks)
d0dabf7e 207{
6629542e 208 int slot = *count;
67cc2b81 209
e4424fee 210 if (sh->ddf_layout)
6629542e 211 (*count)++;
d0dabf7e 212 if (idx == sh->pd_idx)
67cc2b81 213 return syndrome_disks;
d0dabf7e 214 if (idx == sh->qd_idx)
67cc2b81 215 return syndrome_disks + 1;
e4424fee 216 if (!sh->ddf_layout)
6629542e 217 (*count)++;
d0dabf7e
N		 218 return slot;
219}
220
a4456856
DW		 221static void return_io(struct bio *return_bi)
222{
223 struct bio *bi = return_bi;
224 while (bi) {
a4456856
DW		 225
226 return_bi = bi->bi_next;
227 bi->bi_next = NULL;
228 bi->bi_size = 0;
0a82a8d1
LT		 229 trace_block_bio_complete(bdev_get_queue(bi->bi_bdev),
230 bi, 0);
0e13fe23 231 bio_endio(bi, 0);
a4456856
DW		 232 bi = return_bi;
233 }
234}
235
d1688a6d 236static void print_raid5_conf (struct r5conf *conf);
1da177e4 237
600aa109
DW		 238static int stripe_operations_active(struct stripe_head *sh)
239{
240 return sh->check_state || sh->reconstruct_state ||
241 test_bit(STRIPE_BIOFILL_RUN, &sh->state) ||
242 test_bit(STRIPE_COMPUTE_RUN, &sh->state);
243}
244
851c30c9
SL		 245static void raid5_wakeup_stripe_thread(struct stripe_head *sh)
246{
247 struct r5conf *conf = sh->raid_conf;
248 struct r5worker_group *group;
bfc90cb0 249 int thread_cnt;
851c30c9
SL		 250 int i, cpu = sh->cpu;
251
252 if (!cpu_online(cpu)) {
253 cpu = cpumask_any(cpu_online_mask);
254 sh->cpu = cpu;
255 }
256
257 if (list_empty(&sh->lru)) {
258 struct r5worker_group *group;
259 group = conf->worker_groups + cpu_to_group(cpu);
260 list_add_tail(&sh->lru, &group->handle_list);
bfc90cb0
SL		 261 group->stripes_cnt++;
262 sh->group = group;
851c30c9
SL		 263 }
264
265 if (conf->worker_cnt_per_group == 0) {
266 md_wakeup_thread(conf->mddev->thread);
267 return;
268 }
269
270 group = conf->worker_groups + cpu_to_group(sh->cpu);
271
bfc90cb0
SL		 272 group->workers[0].working = true;
273 /* at least one worker should run to avoid race */
274 queue_work_on(sh->cpu, raid5_wq, &group->workers[0].work);
275
276 thread_cnt = group->stripes_cnt / MAX_STRIPE_BATCH - 1;
277 /* wakeup more workers */
278 for (i = 1; i < conf->worker_cnt_per_group && thread_cnt > 0; i++) {
279 if (group->workers[i].working == false) {
280 group->workers[i].working = true;
281 queue_work_on(sh->cpu, raid5_wq,
282 &group->workers[i].work);
283 thread_cnt--;
284 }
285 }
851c30c9
SL		 286}
287
566c09c5
SL		 288static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh,
289 struct list_head *temp_inactive_list)
1da177e4 290{
4eb788df
SL		 291 BUG_ON(!list_empty(&sh->lru));
292 BUG_ON(atomic_read(&conf->active_stripes)==0);
293 if (test_bit(STRIPE_HANDLE, &sh->state)) {
294 if (test_bit(STRIPE_DELAYED, &sh->state) &&
295 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
296 list_add_tail(&sh->lru, &conf->delayed_list);
297 else if (test_bit(STRIPE_BIT_DELAY, &sh->state) &&
298 sh->bm_seq - conf->seq_write > 0)
299 list_add_tail(&sh->lru, &conf->bitmap_list);
300 else {
301 clear_bit(STRIPE_DELAYED, &sh->state);
302 clear_bit(STRIPE_BIT_DELAY, &sh->state);
851c30c9
SL		 303 if (conf->worker_cnt_per_group == 0) {
304 list_add_tail(&sh->lru, &conf->handle_list);
305 } else {
306 raid5_wakeup_stripe_thread(sh);
307 return;
308 }
4eb788df
SL		 309 }
310 md_wakeup_thread(conf->mddev->thread);
311 } else {
312 BUG_ON(stripe_operations_active(sh));
313 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
314 if (atomic_dec_return(&conf->preread_active_stripes)
315 < IO_THRESHOLD)
316 md_wakeup_thread(conf->mddev->thread);
317 atomic_dec(&conf->active_stripes);
566c09c5
SL		 318 if (!test_bit(STRIPE_EXPANDING, &sh->state))
319 list_add_tail(&sh->lru, temp_inactive_list);
1da177e4
LT		 320 }
321}
d0dabf7e 322
566c09c5
SL		 323static void __release_stripe(struct r5conf *conf, struct stripe_head *sh,
324 struct list_head *temp_inactive_list)
4eb788df
SL		 325{
326 if (atomic_dec_and_test(&sh->count))
566c09c5
SL		 327 do_release_stripe(conf, sh, temp_inactive_list);
328}
329
330/*
331 * @hash could be NR_STRIPE_HASH_LOCKS, then we have a list of inactive_list
332 *
333 * Be careful: Only one task can add/delete stripes from temp_inactive_list at
334 * given time. Adding stripes only takes device lock, while deleting stripes
335 * only takes hash lock.
336 */
337static void release_inactive_stripe_list(struct r5conf *conf,
338 struct list_head *temp_inactive_list,
339 int hash)
340{
341 int size;
342 bool do_wakeup = false;
343 unsigned long flags;
344
345 if (hash == NR_STRIPE_HASH_LOCKS) {
346 size = NR_STRIPE_HASH_LOCKS;
347 hash = NR_STRIPE_HASH_LOCKS - 1;
348 } else
349 size = 1;
350 while (size) {
351 struct list_head *list = &temp_inactive_list[size - 1];
352
353 /*
354 * We don't hold any lock here yet, get_active_stripe() might
355 * remove stripes from the list
356 */
357 if (!list_empty_careful(list)) {
358 spin_lock_irqsave(conf->hash_locks + hash, flags);
4bda556a
SL		 359 if (list_empty(conf->inactive_list + hash) &&
360 !list_empty(list))
361 atomic_dec(&conf->empty_inactive_list_nr);
566c09c5
SL		 362 list_splice_tail_init(list, conf->inactive_list + hash);
363 do_wakeup = true;
364 spin_unlock_irqrestore(conf->hash_locks + hash, flags);
365 }
366 size--;
367 hash--;
368 }
369
370 if (do_wakeup) {
371 wake_up(&conf->wait_for_stripe);
372 if (conf->retry_read_aligned)
373 md_wakeup_thread(conf->mddev->thread);
374 }
4eb788df
SL		 375}
376
773ca82f 377/* should hold conf->device_lock already */
566c09c5
SL		 378static int release_stripe_list(struct r5conf *conf,
379 struct list_head *temp_inactive_list)
773ca82f
SL		 380{
381 struct stripe_head *sh;
382 int count = 0;
383 struct llist_node *head;
384
385 head = llist_del_all(&conf->released_stripes);
d265d9dc 386 head = llist_reverse_order(head);
773ca82f 387 while (head) {
566c09c5
SL		 388 int hash;
389
773ca82f
SL		 390 sh = llist_entry(head, struct stripe_head, release_list);
391 head = llist_next(head);
392 /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */
393 smp_mb();
394 clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state);
395 /*
396 * Don't worry the bit is set here, because if the bit is set
397 * again, the count is always > 1. This is true for
398 * STRIPE_ON_UNPLUG_LIST bit too.
399 */
566c09c5
SL		 400 hash = sh->hash_lock_index;
401 __release_stripe(conf, sh, &temp_inactive_list[hash]);
773ca82f
SL		 402 count++;
403 }
404
405 return count;
406}
407
1da177e4
LT		 408static void release_stripe(struct stripe_head *sh)
409{
d1688a6d 410 struct r5conf *conf = sh->raid_conf;
1da177e4 411 unsigned long flags;
566c09c5
SL		 412 struct list_head list;
413 int hash;
773ca82f 414 bool wakeup;
16a53ecc 415
ad4068de 416 if (unlikely(!conf->mddev->thread) ||
417 test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state))
773ca82f
SL		 418 goto slow_path;
419 wakeup = llist_add(&sh->release_list, &conf->released_stripes);
420 if (wakeup)
421 md_wakeup_thread(conf->mddev->thread);
422 return;
423slow_path:
4eb788df 424 local_irq_save(flags);
773ca82f 425 /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */
4eb788df 426 if (atomic_dec_and_lock(&sh->count, &conf->device_lock)) {
566c09c5
SL		 427 INIT_LIST_HEAD(&list);
428 hash = sh->hash_lock_index;
429 do_release_stripe(conf, sh, &list);
4eb788df 430 spin_unlock(&conf->device_lock);
566c09c5 431 release_inactive_stripe_list(conf, &list, hash);
4eb788df
SL		 432 }
433 local_irq_restore(flags);
1da177e4
LT		 434}
435
fccddba0 436static inline void remove_hash(struct stripe_head *sh)
1da177e4 437{
45b4233c
DW		 438 pr_debug("remove_hash(), stripe %llu\n",
439 (unsigned long long)sh->sector);
1da177e4 440
fccddba0 441 hlist_del_init(&sh->hash);
1da177e4
LT		 442}
443
d1688a6d 444static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh)
1da177e4 445{
fccddba0 446 struct hlist_head *hp = stripe_hash(conf, sh->sector);
1da177e4 447
45b4233c
DW		 448 pr_debug("insert_hash(), stripe %llu\n",
449 (unsigned long long)sh->sector);
1da177e4 450
fccddba0 451 hlist_add_head(&sh->hash, hp);
1da177e4
LT		 452}
453
454
455/* find an idle stripe, make sure it is unhashed, and return it. */
566c09c5 456static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash)
1da177e4
LT		 457{
458 struct stripe_head *sh = NULL;
459 struct list_head *first;
460
566c09c5 461 if (list_empty(conf->inactive_list + hash))
1da177e4 462 goto out;
566c09c5 463 first = (conf->inactive_list + hash)->next;
1da177e4
LT		 464 sh = list_entry(first, struct stripe_head, lru);
465 list_del_init(first);
466 remove_hash(sh);
467 atomic_inc(&conf->active_stripes);
566c09c5 468 BUG_ON(hash != sh->hash_lock_index);
4bda556a
SL		 469 if (list_empty(conf->inactive_list + hash))
470 atomic_inc(&conf->empty_inactive_list_nr);
1da177e4
LT		 471out:
472 return sh;
473}
474
e4e11e38 475static void shrink_buffers(struct stripe_head *sh)
1da177e4
LT		 476{
477 struct page *p;
478 int i;
e4e11e38 479 int num = sh->raid_conf->pool_size;
1da177e4 480
e4e11e38 481 for (i = 0; i < num ; i++) {
1da177e4
LT		 482 p = sh->dev[i].page;
483 if (!p)
484 continue;
485 sh->dev[i].page = NULL;
2d1f3b5d 486 put_page(p);
1da177e4
LT		 487 }
488}
489
e4e11e38 490static int grow_buffers(struct stripe_head *sh)
1da177e4
LT		 491{
492 int i;
e4e11e38 493 int num = sh->raid_conf->pool_size;
1da177e4 494
e4e11e38 495 for (i = 0; i < num; i++) {
1da177e4
LT		 496 struct page *page;
497
498 if (!(page = alloc_page(GFP_KERNEL))) {
499 return 1;
500 }
501 sh->dev[i].page = page;
502 }
503 return 0;
504}
505
784052ec 506static void raid5_build_block(struct stripe_head *sh, int i, int previous);
d1688a6d 507static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
911d4ee8 508 struct stripe_head *sh);
1da177e4 509
b5663ba4 510static void init_stripe(struct stripe_head *sh, sector_t sector, int previous)
1da177e4 511{
d1688a6d 512 struct r5conf *conf = sh->raid_conf;
566c09c5 513 int i, seq;
1da177e4 514
78bafebd
ES		 515 BUG_ON(atomic_read(&sh->count) != 0);
516 BUG_ON(test_bit(STRIPE_HANDLE, &sh->state));
600aa109 517 BUG_ON(stripe_operations_active(sh));
d84e0f10 518
45b4233c 519 pr_debug("init_stripe called, stripe %llu\n",
1da177e4
LT		 520 (unsigned long long)sh->sector);
521
522 remove_hash(sh);
566c09c5
SL		 523retry:
524 seq = read_seqcount_begin(&conf->gen_lock);
86b42c71 525 sh->generation = conf->generation - previous;
b5663ba4 526 sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks;
1da177e4 527 sh->sector = sector;
911d4ee8 528 stripe_set_idx(sector, conf, previous, sh);
1da177e4
LT		 529 sh->state = 0;
530
7ecaa1e6
N		 531
532 for (i = sh->disks; i--; ) {
1da177e4
LT		 533 struct r5dev *dev = &sh->dev[i];
534
d84e0f10 535 if (dev->toread || dev->read || dev->towrite || dev->written ||
1da177e4 536 test_bit(R5_LOCKED, &dev->flags)) {
d84e0f10 537 printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n",
1da177e4 538 (unsigned long long)sh->sector, i, dev->toread,
d84e0f10 539 dev->read, dev->towrite, dev->written,
1da177e4 540 test_bit(R5_LOCKED, &dev->flags));
8cfa7b0f 541 WARN_ON(1);
1da177e4
LT		 542 }
543 dev->flags = 0;
784052ec 544 raid5_build_block(sh, i, previous);
1da177e4 545 }
566c09c5
SL		 546 if (read_seqcount_retry(&conf->gen_lock, seq))
547 goto retry;
1da177e4 548 insert_hash(conf, sh);
851c30c9 549 sh->cpu = smp_processor_id();
1da177e4
LT		 550}
551
d1688a6d 552static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector,
86b42c71 553 short generation)
1da177e4
LT		 554{
555 struct stripe_head *sh;
556
45b4233c 557 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
b67bfe0d 558 hlist_for_each_entry(sh, stripe_hash(conf, sector), hash)
86b42c71 559 if (sh->sector == sector && sh->generation == generation)
1da177e4 560 return sh;
45b4233c 561 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
1da177e4
LT		 562 return NULL;
563}
564
674806d6
N		 565/*
566 * Need to check if array has failed when deciding whether to:
567 * - start an array
568 * - remove non-faulty devices
569 * - add a spare
570 * - allow a reshape
571 * This determination is simple when no reshape is happening.
572 * However if there is a reshape, we need to carefully check
573 * both the before and after sections.
574 * This is because some failed devices may only affect one
575 * of the two sections, and some non-in_sync devices may
576 * be insync in the section most affected by failed devices.
577 */
908f4fbd 578static int calc_degraded(struct r5conf *conf)
674806d6 579{
908f4fbd 580 int degraded, degraded2;
674806d6 581 int i;
674806d6
N		 582
583 rcu_read_lock();
584 degraded = 0;
585 for (i = 0; i < conf->previous_raid_disks; i++) {
3cb03002 586 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
e5c86471
N		 587 if (rdev && test_bit(Faulty, &rdev->flags))
588 rdev = rcu_dereference(conf->disks[i].replacement);
674806d6
N		 589 if (!rdev || test_bit(Faulty, &rdev->flags))
590 degraded++;
591 else if (test_bit(In_sync, &rdev->flags))
592 ;
593 else
594 /* not in-sync or faulty.
595 * If the reshape increases the number of devices,
596 * this is being recovered by the reshape, so
597 * this 'previous' section is not in_sync.
598 * If the number of devices is being reduced however,
599 * the device can only be part of the array if
600 * we are reverting a reshape, so this section will
601 * be in-sync.
602 */
603 if (conf->raid_disks >= conf->previous_raid_disks)
604 degraded++;
605 }
606 rcu_read_unlock();
908f4fbd
N		 607 if (conf->raid_disks == conf->previous_raid_disks)
608 return degraded;
674806d6 609 rcu_read_lock();
908f4fbd 610 degraded2 = 0;
674806d6 611 for (i = 0; i < conf->raid_disks; i++) {
3cb03002 612 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
e5c86471
N		 613 if (rdev && test_bit(Faulty, &rdev->flags))
614 rdev = rcu_dereference(conf->disks[i].replacement);
674806d6 615 if (!rdev || test_bit(Faulty, &rdev->flags))
908f4fbd 616 degraded2++;
674806d6
N		 617 else if (test_bit(In_sync, &rdev->flags))
618 ;
619 else
620 /* not in-sync or faulty.
621 * If reshape increases the number of devices, this
622 * section has already been recovered, else it
623 * almost certainly hasn't.
624 */
625 if (conf->raid_disks <= conf->previous_raid_disks)
908f4fbd 626 degraded2++;
674806d6
N		 627 }
628 rcu_read_unlock();
908f4fbd
N		 629 if (degraded2 > degraded)
630 return degraded2;
631 return degraded;
632}
633
634static int has_failed(struct r5conf *conf)
635{
636 int degraded;
637
638 if (conf->mddev->reshape_position == MaxSector)
639 return conf->mddev->degraded > conf->max_degraded;
640
641 degraded = calc_degraded(conf);
674806d6
N		 642 if (degraded > conf->max_degraded)
643 return 1;
644 return 0;
645}
646
b5663ba4 647static struct stripe_head *
d1688a6d 648get_active_stripe(struct r5conf *conf, sector_t sector,
a8c906ca 649 int previous, int noblock, int noquiesce)
1da177e4
LT		 650{
651 struct stripe_head *sh;
566c09c5 652 int hash = stripe_hash_locks_hash(sector);
1da177e4 653
45b4233c 654 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector);
1da177e4 655
566c09c5 656 spin_lock_irq(conf->hash_locks + hash);
1da177e4
LT		 657
658 do {
72626685 659 wait_event_lock_irq(conf->wait_for_stripe,
a8c906ca 660 conf->quiesce == 0 || noquiesce,
566c09c5 661 *(conf->hash_locks + hash));
86b42c71 662 sh = __find_stripe(conf, sector, conf->generation - previous);
1da177e4
LT		 663 if (!sh) {
664 if (!conf->inactive_blocked)
566c09c5 665 sh = get_free_stripe(conf, hash);
1da177e4
LT		 666 if (noblock && sh == NULL)
667 break;
668 if (!sh) {
669 conf->inactive_blocked = 1;
566c09c5
SL		 670 wait_event_lock_irq(
671 conf->wait_for_stripe,
672 !list_empty(conf->inactive_list + hash) &&
673 (atomic_read(&conf->active_stripes)
674 < (conf->max_nr_stripes * 3 / 4)
675 || !conf->inactive_blocked),
676 *(conf->hash_locks + hash));
1da177e4
LT		 677 conf->inactive_blocked = 0;
678 } else
b5663ba4 679 init_stripe(sh, sector, previous);
1da177e4 680 } else {
6d183de4 681 spin_lock(&conf->device_lock);
1da177e4 682 if (atomic_read(&sh->count)) {
ab69ae12 683 BUG_ON(!list_empty(&sh->lru)
8811b596 684 && !test_bit(STRIPE_EXPANDING, &sh->state)
773ca82f 685 && !test_bit(STRIPE_ON_UNPLUG_LIST, &sh->state)
6d183de4 686 );
1da177e4
LT		 687 } else {
688 if (!test_bit(STRIPE_HANDLE, &sh->state))
689 atomic_inc(&conf->active_stripes);
5af9bef7
N		 690 BUG_ON(list_empty(&sh->lru) &&
691 !test_bit(STRIPE_EXPANDING, &sh->state));
16a53ecc 692 list_del_init(&sh->lru);
bfc90cb0
SL		 693 if (sh->group) {
694 sh->group->stripes_cnt--;
695 sh->group = NULL;
696 }
1da177e4 697 }
6d183de4 698 spin_unlock(&conf->device_lock);
1da177e4
LT		 699 }
700 } while (sh == NULL);
701
702 if (sh)
703 atomic_inc(&sh->count);
704
566c09c5 705 spin_unlock_irq(conf->hash_locks + hash);
1da177e4
LT		 706 return sh;
707}
708
05616be5
N		 709/* Determine if 'data_offset' or 'new_data_offset' should be used
710 * in this stripe_head.
711 */
712static int use_new_offset(struct r5conf *conf, struct stripe_head *sh)
713{
714 sector_t progress = conf->reshape_progress;
715 /* Need a memory barrier to make sure we see the value
716 * of conf->generation, or ->data_offset that was set before
717 * reshape_progress was updated.
718 */
719 smp_rmb();
720 if (progress == MaxSector)
721 return 0;
722 if (sh->generation == conf->generation - 1)
723 return 0;
724 /* We are in a reshape, and this is a new-generation stripe,
725 * so use new_data_offset.
726 */
727 return 1;
728}
729
6712ecf8
N		 730static void
731raid5_end_read_request(struct bio *bi, int error);
732static void
733raid5_end_write_request(struct bio *bi, int error);
91c00924 734
c4e5ac0a 735static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
91c00924 736{
d1688a6d 737 struct r5conf *conf = sh->raid_conf;
91c00924
DW		 738 int i, disks = sh->disks;
739
740 might_sleep();
741
742 for (i = disks; i--; ) {
743 int rw;
9a3e1101 744 int replace_only = 0;
977df362
N		 745 struct bio *bi, *rbi;
746 struct md_rdev *rdev, *rrdev = NULL;
e9c7469b
TH		 747 if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) {
748 if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags))
749 rw = WRITE_FUA;
750 else
751 rw = WRITE;
9e444768 752 if (test_bit(R5_Discard, &sh->dev[i].flags))
620125f2 753 rw |= REQ_DISCARD;
e9c7469b 754 } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags))
91c00924 755 rw = READ;
9a3e1101
N		 756 else if (test_and_clear_bit(R5_WantReplace,
757 &sh->dev[i].flags)) {
758 rw = WRITE;
759 replace_only = 1;
760 } else
91c00924 761 continue;
bc0934f0
SL		 762 if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags))
763 rw |= REQ_SYNC;
91c00924
DW		 764
765 bi = &sh->dev[i].req;
977df362 766 rbi = &sh->dev[i].rreq; /* For writing to replacement */
91c00924 767
91c00924 768 rcu_read_lock();
9a3e1101 769 rrdev = rcu_dereference(conf->disks[i].replacement);
dd054fce
N		 770 smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */
771 rdev = rcu_dereference(conf->disks[i].rdev);
772 if (!rdev) {
773 rdev = rrdev;
774 rrdev = NULL;
775 }
9a3e1101
N		 776 if (rw & WRITE) {
777 if (replace_only)
778 rdev = NULL;
dd054fce
N		 779 if (rdev == rrdev)
780 /* We raced and saw duplicates */
781 rrdev = NULL;
9a3e1101 782 } else {
dd054fce 783 if (test_bit(R5_ReadRepl, &sh->dev[i].flags) && rrdev)
9a3e1101
N		 784 rdev = rrdev;
785 rrdev = NULL;
786 }
977df362 787
91c00924
DW		 788 if (rdev && test_bit(Faulty, &rdev->flags))
789 rdev = NULL;
790 if (rdev)
791 atomic_inc(&rdev->nr_pending);
977df362
N		 792 if (rrdev && test_bit(Faulty, &rrdev->flags))
793 rrdev = NULL;
794 if (rrdev)
795 atomic_inc(&rrdev->nr_pending);
91c00924
DW		 796 rcu_read_unlock();
797
73e92e51 798 /* We have already checked bad blocks for reads. Now
977df362
N		 799 * need to check for writes. We never accept write errors
800 * on the replacement, so we don't to check rrdev.
73e92e51
N		 801 */
802 while ((rw & WRITE) && rdev &&
803 test_bit(WriteErrorSeen, &rdev->flags)) {
804 sector_t first_bad;
805 int bad_sectors;
806 int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
807 &first_bad, &bad_sectors);
808 if (!bad)
809 break;
810
811 if (bad < 0) {
812 set_bit(BlockedBadBlocks, &rdev->flags);
813 if (!conf->mddev->external &&
814 conf->mddev->flags) {
815 /* It is very unlikely, but we might
816 * still need to write out the
817 * bad block log - better give it
818 * a chance*/
819 md_check_recovery(conf->mddev);
820 }
1850753d 821 /*
822 * Because md_wait_for_blocked_rdev
823 * will dec nr_pending, we must
824 * increment it first.
825 */
826 atomic_inc(&rdev->nr_pending);
73e92e51
N		 827 md_wait_for_blocked_rdev(rdev, conf->mddev);
828 } else {
829 /* Acknowledged bad block - skip the write */
830 rdev_dec_pending(rdev, conf->mddev);
831 rdev = NULL;
832 }
833 }
834
91c00924 835 if (rdev) {
9a3e1101
N		 836 if (s->syncing || s->expanding || s->expanded
837 || s->replacing)
91c00924
DW		 838 md_sync_acct(rdev->bdev, STRIPE_SECTORS);
839
2b7497f0
DW		 840 set_bit(STRIPE_IO_STARTED, &sh->state);
841
2f6db2a7 842 bio_reset(bi);
91c00924 843 bi->bi_bdev = rdev->bdev;
2f6db2a7
KO		 844 bi->bi_rw = rw;
845 bi->bi_end_io = (rw & WRITE)
846 ? raid5_end_write_request
847 : raid5_end_read_request;
848 bi->bi_private = sh;
849
91c00924 850 pr_debug("%s: for %llu schedule op %ld on disc %d\n",
e46b272b 851 __func__, (unsigned long long)sh->sector,
91c00924
DW		 852 bi->bi_rw, i);
853 atomic_inc(&sh->count);
05616be5
N		 854 if (use_new_offset(conf, sh))
855 bi->bi_sector = (sh->sector
856 + rdev->new_data_offset);
857 else
858 bi->bi_sector = (sh->sector
859 + rdev->data_offset);
3f9e7c14 860 if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
e59aa23f 861 bi->bi_rw |= REQ_NOMERGE;
3f9e7c14 862
4997b72e 863 bi->bi_vcnt = 1;
91c00924
DW		 864 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
865 bi->bi_io_vec[0].bv_offset = 0;
866 bi->bi_size = STRIPE_SIZE;
37c61ff3
SL		 867 /*
868 * If this is discard request, set bi_vcnt 0. We don't
869 * want to confuse SCSI because SCSI will replace payload
870 */
871 if (rw & REQ_DISCARD)
872 bi->bi_vcnt = 0;
977df362
N		 873 if (rrdev)
874 set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags);
e3620a3a
JB		 875
876 if (conf->mddev->gendisk)
877 trace_block_bio_remap(bdev_get_queue(bi->bi_bdev),
878 bi, disk_devt(conf->mddev->gendisk),
879 sh->dev[i].sector);
91c00924 880 generic_make_request(bi);
977df362
N		 881 }
882 if (rrdev) {
9a3e1101
N		 883 if (s->syncing || s->expanding || s->expanded
884 || s->replacing)
977df362
N		 885 md_sync_acct(rrdev->bdev, STRIPE_SECTORS);
886
887 set_bit(STRIPE_IO_STARTED, &sh->state);
888
2f6db2a7 889 bio_reset(rbi);
977df362 890 rbi->bi_bdev = rrdev->bdev;
2f6db2a7
KO		 891 rbi->bi_rw = rw;
892 BUG_ON(!(rw & WRITE));
893 rbi->bi_end_io = raid5_end_write_request;
894 rbi->bi_private = sh;
895
977df362
N		 896 pr_debug("%s: for %llu schedule op %ld on "
897 "replacement disc %d\n",
898 __func__, (unsigned long long)sh->sector,
899 rbi->bi_rw, i);
900 atomic_inc(&sh->count);
05616be5
N		 901 if (use_new_offset(conf, sh))
902 rbi->bi_sector = (sh->sector
903 + rrdev->new_data_offset);
904 else
905 rbi->bi_sector = (sh->sector
906 + rrdev->data_offset);
4997b72e 907 rbi->bi_vcnt = 1;
977df362
N		 908 rbi->bi_io_vec[0].bv_len = STRIPE_SIZE;
909 rbi->bi_io_vec[0].bv_offset = 0;
910 rbi->bi_size = STRIPE_SIZE;
37c61ff3
SL		 911 /*
912 * If this is discard request, set bi_vcnt 0. We don't
913 * want to confuse SCSI because SCSI will replace payload
914 */
915 if (rw & REQ_DISCARD)
916 rbi->bi_vcnt = 0;
e3620a3a
JB		 917 if (conf->mddev->gendisk)
918 trace_block_bio_remap(bdev_get_queue(rbi->bi_bdev),
919 rbi, disk_devt(conf->mddev->gendisk),
920 sh->dev[i].sector);
977df362
N		 921 generic_make_request(rbi);
922 }
923 if (!rdev && !rrdev) {
b062962e 924 if (rw & WRITE)
91c00924
DW		 925 set_bit(STRIPE_DEGRADED, &sh->state);
926 pr_debug("skip op %ld on disc %d for sector %llu\n",
927 bi->bi_rw, i, (unsigned long long)sh->sector);
928 clear_bit(R5_LOCKED, &sh->dev[i].flags);
929 set_bit(STRIPE_HANDLE, &sh->state);
930 }
931 }
932}
933
934static struct dma_async_tx_descriptor *
935async_copy_data(int frombio, struct bio *bio, struct page *page,
936 sector_t sector, struct dma_async_tx_descriptor *tx)
937{
938 struct bio_vec *bvl;
939 struct page *bio_page;
940 int i;
941 int page_offset;
a08abd8c 942 struct async_submit_ctl submit;
0403e382 943 enum async_tx_flags flags = 0;
91c00924
DW		 944
945 if (bio->bi_sector >= sector)
946 page_offset = (signed)(bio->bi_sector - sector) * 512;
947 else
948 page_offset = (signed)(sector - bio->bi_sector) * -512;
a08abd8c 949
0403e382
DW		 950 if (frombio)
951 flags |= ASYNC_TX_FENCE;
952 init_async_submit(&submit, flags, tx, NULL, NULL, NULL);
953
91c00924 954 bio_for_each_segment(bvl, bio, i) {
fcde9075 955 int len = bvl->bv_len;
91c00924
DW		 956 int clen;
957 int b_offset = 0;
958
959 if (page_offset < 0) {
960 b_offset = -page_offset;
961 page_offset += b_offset;
962 len -= b_offset;
963 }
964
965 if (len > 0 && page_offset + len > STRIPE_SIZE)
966 clen = STRIPE_SIZE - page_offset;
967 else
968 clen = len;
969
970 if (clen > 0) {
fcde9075
NK		 971 b_offset += bvl->bv_offset;
972 bio_page = bvl->bv_page;
91c00924
DW		 973 if (frombio)
974 tx = async_memcpy(page, bio_page, page_offset,
a08abd8c 975 b_offset, clen, &submit);
91c00924
DW		 976 else
977 tx = async_memcpy(bio_page, page, b_offset,
a08abd8c 978 page_offset, clen, &submit);
91c00924 979 }
a08abd8c
DW		 980 /* chain the operations */
981 submit.depend_tx = tx;
982
91c00924
DW		 983 if (clen < len) /* hit end of page */
984 break;
985 page_offset += len;
986 }
987
988 return tx;
989}
990
991static void ops_complete_biofill(void *stripe_head_ref)
992{
993 struct stripe_head *sh = stripe_head_ref;
994 struct bio *return_bi = NULL;
e4d84909 995 int i;
91c00924 996
e46b272b 997 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 998 (unsigned long long)sh->sector);
999
1000 /* clear completed biofills */
1001 for (i = sh->disks; i--; ) {
1002 struct r5dev *dev = &sh->dev[i];
91c00924
DW		 1003
1004 /* acknowledge completion of a biofill operation */
e4d84909
DW		 1005 /* and check if we need to reply to a read request,
1006 * new R5_Wantfill requests are held off until
83de75cc 1007 * !STRIPE_BIOFILL_RUN
e4d84909
DW		 1008 */
1009 if (test_and_clear_bit(R5_Wantfill, &dev->flags)) {
91c00924 1010 struct bio *rbi, *rbi2;
91c00924 1011
91c00924
DW		 1012 BUG_ON(!dev->read);
1013 rbi = dev->read;
1014 dev->read = NULL;
1015 while (rbi && rbi->bi_sector <
1016 dev->sector + STRIPE_SECTORS) {
1017 rbi2 = r5_next_bio(rbi, dev->sector);
e7836bd6 1018 if (!raid5_dec_bi_active_stripes(rbi)) {
91c00924
DW		 1019 rbi->bi_next = return_bi;
1020 return_bi = rbi;
1021 }
91c00924
DW		 1022 rbi = rbi2;
1023 }
1024 }
1025 }
83de75cc 1026 clear_bit(STRIPE_BIOFILL_RUN, &sh->state);
91c00924
DW		 1027
1028 return_io(return_bi);
1029
e4d84909 1030 set_bit(STRIPE_HANDLE, &sh->state);
91c00924
DW		 1031 release_stripe(sh);
1032}
1033
1034static void ops_run_biofill(struct stripe_head *sh)
1035{
1036 struct dma_async_tx_descriptor *tx = NULL;
a08abd8c 1037 struct async_submit_ctl submit;
91c00924
DW		 1038 int i;
1039
e46b272b 1040 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1041 (unsigned long long)sh->sector);
1042
1043 for (i = sh->disks; i--; ) {
1044 struct r5dev *dev = &sh->dev[i];
1045 if (test_bit(R5_Wantfill, &dev->flags)) {
1046 struct bio *rbi;
b17459c0 1047 spin_lock_irq(&sh->stripe_lock);
91c00924
DW		 1048 dev->read = rbi = dev->toread;
1049 dev->toread = NULL;
b17459c0 1050 spin_unlock_irq(&sh->stripe_lock);
91c00924
DW		 1051 while (rbi && rbi->bi_sector <
1052 dev->sector + STRIPE_SECTORS) {
1053 tx = async_copy_data(0, rbi, dev->page,
1054 dev->sector, tx);
1055 rbi = r5_next_bio(rbi, dev->sector);
1056 }
1057 }
1058 }
1059
1060 atomic_inc(&sh->count);
a08abd8c
DW		 1061 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL);
1062 async_trigger_callback(&submit);
91c00924
DW		 1063}
1064
4e7d2c0a 1065static void mark_target_uptodate(struct stripe_head *sh, int target)
91c00924 1066{
4e7d2c0a 1067 struct r5dev *tgt;
91c00924 1068
4e7d2c0a
DW		 1069 if (target < 0)
1070 return;
91c00924 1071
4e7d2c0a 1072 tgt = &sh->dev[target];
91c00924
DW		 1073 set_bit(R5_UPTODATE, &tgt->flags);
1074 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1075 clear_bit(R5_Wantcompute, &tgt->flags);
4e7d2c0a
DW		 1076}
1077
ac6b53b6 1078static void ops_complete_compute(void *stripe_head_ref)
91c00924
DW		 1079{
1080 struct stripe_head *sh = stripe_head_ref;
91c00924 1081
e46b272b 1082 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1083 (unsigned long long)sh->sector);
1084
ac6b53b6 1085 /* mark the computed target(s) as uptodate */
4e7d2c0a 1086 mark_target_uptodate(sh, sh->ops.target);
ac6b53b6 1087 mark_target_uptodate(sh, sh->ops.target2);
4e7d2c0a 1088
ecc65c9b
DW		 1089 clear_bit(STRIPE_COMPUTE_RUN, &sh->state);
1090 if (sh->check_state == check_state_compute_run)
1091 sh->check_state = check_state_compute_result;
91c00924
DW		 1092 set_bit(STRIPE_HANDLE, &sh->state);
1093 release_stripe(sh);
1094}
1095
d6f38f31
DW		 1096/* return a pointer to the address conversion region of the scribble buffer */
1097static addr_conv_t *to_addr_conv(struct stripe_head *sh,
1098 struct raid5_percpu *percpu)
1099{
1100 return percpu->scribble + sizeof(struct page *) * (sh->disks + 2);
1101}
1102
1103static struct dma_async_tx_descriptor *
1104ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 1105{
91c00924 1106 int disks = sh->disks;
d6f38f31 1107 struct page **xor_srcs = percpu->scribble;
91c00924
DW		 1108 int target = sh->ops.target;
1109 struct r5dev *tgt = &sh->dev[target];
1110 struct page *xor_dest = tgt->page;
1111 int count = 0;
1112 struct dma_async_tx_descriptor *tx;
a08abd8c 1113 struct async_submit_ctl submit;
91c00924
DW		 1114 int i;
1115
1116 pr_debug("%s: stripe %llu block: %d\n",
e46b272b 1117 __func__, (unsigned long long)sh->sector, target);
91c00924
DW		 1118 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1119
1120 for (i = disks; i--; )
1121 if (i != target)
1122 xor_srcs[count++] = sh->dev[i].page;
1123
1124 atomic_inc(&sh->count);
1125
0403e382 1126 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL,
ac6b53b6 1127 ops_complete_compute, sh, to_addr_conv(sh, percpu));
91c00924 1128 if (unlikely(count == 1))
a08abd8c 1129 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
91c00924 1130 else
a08abd8c 1131 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924 1132
91c00924
DW		 1133 return tx;
1134}
1135
ac6b53b6
DW		 1136/* set_syndrome_sources - populate source buffers for gen_syndrome
1137 * @srcs - (struct page *) array of size sh->disks
1138 * @sh - stripe_head to parse
1139 *
1140 * Populates srcs in proper layout order for the stripe and returns the
1141 * 'count' of sources to be used in a call to async_gen_syndrome. The P
1142 * destination buffer is recorded in srcs[count] and the Q destination
1143 * is recorded in srcs[count+1]].
1144 */
1145static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh)
1146{
1147 int disks = sh->disks;
1148 int syndrome_disks = sh->ddf_layout ? disks : (disks - 2);
1149 int d0_idx = raid6_d0(sh);
1150 int count;
1151 int i;
1152
1153 for (i = 0; i < disks; i++)
5dd33c9a 1154 srcs[i] = NULL;
ac6b53b6
DW		 1155
1156 count = 0;
1157 i = d0_idx;
1158 do {
1159 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
1160
1161 srcs[slot] = sh->dev[i].page;
1162 i = raid6_next_disk(i, disks);
1163 } while (i != d0_idx);
ac6b53b6 1164
e4424fee 1165 return syndrome_disks;
ac6b53b6
DW		 1166}
1167
1168static struct dma_async_tx_descriptor *
1169ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu)
1170{
1171 int disks = sh->disks;
1172 struct page **blocks = percpu->scribble;
1173 int target;
1174 int qd_idx = sh->qd_idx;
1175 struct dma_async_tx_descriptor *tx;
1176 struct async_submit_ctl submit;
1177 struct r5dev *tgt;
1178 struct page *dest;
1179 int i;
1180 int count;
1181
1182 if (sh->ops.target < 0)
1183 target = sh->ops.target2;
1184 else if (sh->ops.target2 < 0)
1185 target = sh->ops.target;
91c00924 1186 else
ac6b53b6
DW		 1187 /* we should only have one valid target */
1188 BUG();
1189 BUG_ON(target < 0);
1190 pr_debug("%s: stripe %llu block: %d\n",
1191 __func__, (unsigned long long)sh->sector, target);
1192
1193 tgt = &sh->dev[target];
1194 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1195 dest = tgt->page;
1196
1197 atomic_inc(&sh->count);
1198
1199 if (target == qd_idx) {
1200 count = set_syndrome_sources(blocks, sh);
1201 blocks[count] = NULL; /* regenerating p is not necessary */
1202 BUG_ON(blocks[count+1] != dest); /* q should already be set */
0403e382
DW		 1203 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1204 ops_complete_compute, sh,
ac6b53b6
DW		 1205 to_addr_conv(sh, percpu));
1206 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
1207 } else {
1208 /* Compute any data- or p-drive using XOR */
1209 count = 0;
1210 for (i = disks; i-- ; ) {
1211 if (i == target || i == qd_idx)
1212 continue;
1213 blocks[count++] = sh->dev[i].page;
1214 }
1215
0403e382
DW		 1216 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1217 NULL, ops_complete_compute, sh,
ac6b53b6
DW		 1218 to_addr_conv(sh, percpu));
1219 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit);
1220 }
91c00924 1221
91c00924
DW		 1222 return tx;
1223}
1224
ac6b53b6
DW		 1225static struct dma_async_tx_descriptor *
1226ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu)
1227{
1228 int i, count, disks = sh->disks;
1229 int syndrome_disks = sh->ddf_layout ? disks : disks-2;
1230 int d0_idx = raid6_d0(sh);
1231 int faila = -1, failb = -1;
1232 int target = sh->ops.target;
1233 int target2 = sh->ops.target2;
1234 struct r5dev *tgt = &sh->dev[target];
1235 struct r5dev *tgt2 = &sh->dev[target2];
1236 struct dma_async_tx_descriptor *tx;
1237 struct page **blocks = percpu->scribble;
1238 struct async_submit_ctl submit;
1239
1240 pr_debug("%s: stripe %llu block1: %d block2: %d\n",
1241 __func__, (unsigned long long)sh->sector, target, target2);
1242 BUG_ON(target < 0 || target2 < 0);
1243 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1244 BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags));
1245
6c910a78 1246 /* we need to open-code set_syndrome_sources to handle the
ac6b53b6
DW		 1247 * slot number conversion for 'faila' and 'failb'
1248 */
1249 for (i = 0; i < disks ; i++)
5dd33c9a 1250 blocks[i] = NULL;
ac6b53b6
DW		 1251 count = 0;
1252 i = d0_idx;
1253 do {
1254 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
1255
1256 blocks[slot] = sh->dev[i].page;
1257
1258 if (i == target)
1259 faila = slot;
1260 if (i == target2)
1261 failb = slot;
1262 i = raid6_next_disk(i, disks);
1263 } while (i != d0_idx);
ac6b53b6
DW		 1264
1265 BUG_ON(faila == failb);
1266 if (failb < faila)
1267 swap(faila, failb);
1268 pr_debug("%s: stripe: %llu faila: %d failb: %d\n",
1269 __func__, (unsigned long long)sh->sector, faila, failb);
1270
1271 atomic_inc(&sh->count);
1272
1273 if (failb == syndrome_disks+1) {
1274 /* Q disk is one of the missing disks */
1275 if (faila == syndrome_disks) {
1276 /* Missing P+Q, just recompute */
0403e382
DW		 1277 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1278 ops_complete_compute, sh,
1279 to_addr_conv(sh, percpu));
e4424fee 1280 return async_gen_syndrome(blocks, 0, syndrome_disks+2,
ac6b53b6
DW		 1281 STRIPE_SIZE, &submit);
1282 } else {
1283 struct page *dest;
1284 int data_target;
1285 int qd_idx = sh->qd_idx;
1286
1287 /* Missing D+Q: recompute D from P, then recompute Q */
1288 if (target == qd_idx)
1289 data_target = target2;
1290 else
1291 data_target = target;
1292
1293 count = 0;
1294 for (i = disks; i-- ; ) {
1295 if (i == data_target || i == qd_idx)
1296 continue;
1297 blocks[count++] = sh->dev[i].page;
1298 }
1299 dest = sh->dev[data_target].page;
0403e382
DW		 1300 init_async_submit(&submit,
1301 ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1302 NULL, NULL, NULL,
1303 to_addr_conv(sh, percpu));
ac6b53b6
DW		 1304 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE,
1305 &submit);
1306
1307 count = set_syndrome_sources(blocks, sh);
0403e382
DW		 1308 init_async_submit(&submit, ASYNC_TX_FENCE, tx,
1309 ops_complete_compute, sh,
1310 to_addr_conv(sh, percpu));
ac6b53b6
DW		 1311 return async_gen_syndrome(blocks, 0, count+2,
1312 STRIPE_SIZE, &submit);
1313 }
ac6b53b6 1314 } else {
6c910a78
DW		 1315 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1316 ops_complete_compute, sh,
1317 to_addr_conv(sh, percpu));
1318 if (failb == syndrome_disks) {
1319 /* We're missing D+P. */
1320 return async_raid6_datap_recov(syndrome_disks+2,
1321 STRIPE_SIZE, faila,
1322 blocks, &submit);
1323 } else {
1324 /* We're missing D+D. */
1325 return async_raid6_2data_recov(syndrome_disks+2,
1326 STRIPE_SIZE, faila, failb,
1327 blocks, &submit);
1328 }
ac6b53b6
DW		 1329 }
1330}
1331
1332
91c00924
DW		 1333static void ops_complete_prexor(void *stripe_head_ref)
1334{
1335 struct stripe_head *sh = stripe_head_ref;
1336
e46b272b 1337 pr_debug("%s: stripe %llu\n", __func__,
91c00924 1338 (unsigned long long)sh->sector);
91c00924
DW		 1339}
1340
1341static struct dma_async_tx_descriptor *
d6f38f31
DW		 1342ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu,
1343 struct dma_async_tx_descriptor *tx)
91c00924 1344{
91c00924 1345 int disks = sh->disks;
d6f38f31 1346 struct page **xor_srcs = percpu->scribble;
91c00924 1347 int count = 0, pd_idx = sh->pd_idx, i;
a08abd8c 1348 struct async_submit_ctl submit;
91c00924
DW		 1349
1350 /* existing parity data subtracted */
1351 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1352
e46b272b 1353 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1354 (unsigned long long)sh->sector);
1355
1356 for (i = disks; i--; ) {
1357 struct r5dev *dev = &sh->dev[i];
1358 /* Only process blocks that are known to be uptodate */
d8ee0728 1359 if (test_bit(R5_Wantdrain, &dev->flags))
91c00924
DW		 1360 xor_srcs[count++] = dev->page;
1361 }
1362
0403e382 1363 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
d6f38f31 1364 ops_complete_prexor, sh, to_addr_conv(sh, percpu));
a08abd8c 1365 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924
DW		 1366
1367 return tx;
1368}
1369
1370static struct dma_async_tx_descriptor *
d8ee0728 1371ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
91c00924
DW		 1372{
1373 int disks = sh->disks;
d8ee0728 1374 int i;
91c00924 1375
e46b272b 1376 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1377 (unsigned long long)sh->sector);
1378
1379 for (i = disks; i--; ) {
1380 struct r5dev *dev = &sh->dev[i];
1381 struct bio *chosen;
91c00924 1382
d8ee0728 1383 if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) {
91c00924
DW		 1384 struct bio *wbi;
1385
b17459c0 1386 spin_lock_irq(&sh->stripe_lock);
91c00924
DW		 1387 chosen = dev->towrite;
1388 dev->towrite = NULL;
1389 BUG_ON(dev->written);
1390 wbi = dev->written = chosen;
b17459c0 1391 spin_unlock_irq(&sh->stripe_lock);
91c00924
DW		 1392
1393 while (wbi && wbi->bi_sector <
1394 dev->sector + STRIPE_SECTORS) {
e9c7469b
TH		 1395 if (wbi->bi_rw & REQ_FUA)
1396 set_bit(R5_WantFUA, &dev->flags);
bc0934f0
SL		 1397 if (wbi->bi_rw & REQ_SYNC)
1398 set_bit(R5_SyncIO, &dev->flags);
9e444768 1399 if (wbi->bi_rw & REQ_DISCARD)
620125f2 1400 set_bit(R5_Discard, &dev->flags);
9e444768 1401 else
620125f2
SL		 1402 tx = async_copy_data(1, wbi, dev->page,
1403 dev->sector, tx);
91c00924
DW		 1404 wbi = r5_next_bio(wbi, dev->sector);
1405 }
1406 }
1407 }
1408
1409 return tx;
1410}
1411
ac6b53b6 1412static void ops_complete_reconstruct(void *stripe_head_ref)
91c00924
DW		 1413{
1414 struct stripe_head *sh = stripe_head_ref;
ac6b53b6
DW		 1415 int disks = sh->disks;
1416 int pd_idx = sh->pd_idx;
1417 int qd_idx = sh->qd_idx;
1418 int i;
9e444768 1419 bool fua = false, sync = false, discard = false;
91c00924 1420
e46b272b 1421 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1422 (unsigned long long)sh->sector);
1423
bc0934f0 1424 for (i = disks; i--; ) {
e9c7469b 1425 fua |= test_bit(R5_WantFUA, &sh->dev[i].flags);
bc0934f0 1426 sync |= test_bit(R5_SyncIO, &sh->dev[i].flags);
9e444768 1427 discard |= test_bit(R5_Discard, &sh->dev[i].flags);
bc0934f0 1428 }
e9c7469b 1429
91c00924
DW		 1430 for (i = disks; i--; ) {
1431 struct r5dev *dev = &sh->dev[i];
ac6b53b6 1432
e9c7469b 1433 if (dev->written || i == pd_idx || i == qd_idx) {
9e444768
SL		 1434 if (!discard)
1435 set_bit(R5_UPTODATE, &dev->flags);
e9c7469b
TH		 1436 if (fua)
1437 set_bit(R5_WantFUA, &dev->flags);
bc0934f0
SL		 1438 if (sync)
1439 set_bit(R5_SyncIO, &dev->flags);
e9c7469b 1440 }
91c00924
DW		 1441 }
1442
d8ee0728
DW		 1443 if (sh->reconstruct_state == reconstruct_state_drain_run)
1444 sh->reconstruct_state = reconstruct_state_drain_result;
1445 else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run)
1446 sh->reconstruct_state = reconstruct_state_prexor_drain_result;
1447 else {
1448 BUG_ON(sh->reconstruct_state != reconstruct_state_run);
1449 sh->reconstruct_state = reconstruct_state_result;
1450 }
91c00924
DW		 1451
1452 set_bit(STRIPE_HANDLE, &sh->state);
1453 release_stripe(sh);
1454}
1455
1456static void
ac6b53b6
DW		 1457ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu,
1458 struct dma_async_tx_descriptor *tx)
91c00924 1459{
91c00924 1460 int disks = sh->disks;
d6f38f31 1461 struct page **xor_srcs = percpu->scribble;
a08abd8c 1462 struct async_submit_ctl submit;
91c00924
DW		 1463 int count = 0, pd_idx = sh->pd_idx, i;
1464 struct page *xor_dest;
d8ee0728 1465 int prexor = 0;
91c00924 1466 unsigned long flags;
91c00924 1467
e46b272b 1468 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1469 (unsigned long long)sh->sector);
1470
620125f2
SL		 1471 for (i = 0; i < sh->disks; i++) {
1472 if (pd_idx == i)
1473 continue;
1474 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1475 break;
1476 }
1477 if (i >= sh->disks) {
1478 atomic_inc(&sh->count);
620125f2
SL		 1479 set_bit(R5_Discard, &sh->dev[pd_idx].flags);
1480 ops_complete_reconstruct(sh);
1481 return;
1482 }
91c00924
DW		 1483 /* check if prexor is active which means only process blocks
1484 * that are part of a read-modify-write (written)
1485 */
d8ee0728
DW		 1486 if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
1487 prexor = 1;
91c00924
DW		 1488 xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1489 for (i = disks; i--; ) {
1490 struct r5dev *dev = &sh->dev[i];
1491 if (dev->written)
1492 xor_srcs[count++] = dev->page;
1493 }
1494 } else {
1495 xor_dest = sh->dev[pd_idx].page;
1496 for (i = disks; i--; ) {
1497 struct r5dev *dev = &sh->dev[i];
1498 if (i != pd_idx)
1499 xor_srcs[count++] = dev->page;
1500 }
1501 }
1502
91c00924
DW		 1503 /* 1/ if we prexor'd then the dest is reused as a source
1504 * 2/ if we did not prexor then we are redoing the parity
1505 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
1506 * for the synchronous xor case
1507 */
88ba2aa5 1508 flags = ASYNC_TX_ACK |
91c00924
DW		 1509 (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
1510
1511 atomic_inc(&sh->count);
1512
ac6b53b6 1513 init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh,
d6f38f31 1514 to_addr_conv(sh, percpu));
a08abd8c
DW		 1515 if (unlikely(count == 1))
1516 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
1517 else
1518 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924
DW		 1519}
1520
ac6b53b6
DW		 1521static void
1522ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu,
1523 struct dma_async_tx_descriptor *tx)
1524{
1525 struct async_submit_ctl submit;
1526 struct page **blocks = percpu->scribble;
620125f2 1527 int count, i;
ac6b53b6
DW		 1528
1529 pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector);
1530
620125f2
SL		 1531 for (i = 0; i < sh->disks; i++) {
1532 if (sh->pd_idx == i || sh->qd_idx == i)
1533 continue;
1534 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1535 break;
1536 }
1537 if (i >= sh->disks) {
1538 atomic_inc(&sh->count);
620125f2
SL		 1539 set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
1540 set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
1541 ops_complete_reconstruct(sh);
1542 return;
1543 }
1544
ac6b53b6
DW		 1545 count = set_syndrome_sources(blocks, sh);
1546
1547 atomic_inc(&sh->count);
1548
1549 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct,
1550 sh, to_addr_conv(sh, percpu));
1551 async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
91c00924
DW		 1552}
1553
1554static void ops_complete_check(void *stripe_head_ref)
1555{
1556 struct stripe_head *sh = stripe_head_ref;
91c00924 1557
e46b272b 1558 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1559 (unsigned long long)sh->sector);
1560
ecc65c9b 1561 sh->check_state = check_state_check_result;
91c00924
DW		 1562 set_bit(STRIPE_HANDLE, &sh->state);
1563 release_stripe(sh);
1564}
1565
ac6b53b6 1566static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 1567{
91c00924 1568 int disks = sh->disks;
ac6b53b6
DW		 1569 int pd_idx = sh->pd_idx;
1570 int qd_idx = sh->qd_idx;
1571 struct page *xor_dest;
d6f38f31 1572 struct page **xor_srcs = percpu->scribble;
91c00924 1573 struct dma_async_tx_descriptor *tx;
a08abd8c 1574 struct async_submit_ctl submit;
ac6b53b6
DW		 1575 int count;
1576 int i;
91c00924 1577
e46b272b 1578 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1579 (unsigned long long)sh->sector);
1580
ac6b53b6
DW		 1581 count = 0;
1582 xor_dest = sh->dev[pd_idx].page;
1583 xor_srcs[count++] = xor_dest;
91c00924 1584 for (i = disks; i--; ) {
ac6b53b6
DW		 1585 if (i == pd_idx || i == qd_idx)
1586 continue;
1587 xor_srcs[count++] = sh->dev[i].page;
91c00924
DW		 1588 }
1589
d6f38f31
DW		 1590 init_async_submit(&submit, 0, NULL, NULL, NULL,
1591 to_addr_conv(sh, percpu));
099f53cb 1592 tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
a08abd8c 1593 &sh->ops.zero_sum_result, &submit);
91c00924 1594
91c00924 1595 atomic_inc(&sh->count);
a08abd8c
DW		 1596 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL);
1597 tx = async_trigger_callback(&submit);
91c00924
DW		 1598}
1599
ac6b53b6
DW		 1600static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp)
1601{
1602 struct page **srcs = percpu->scribble;
1603 struct async_submit_ctl submit;
1604 int count;
1605
1606 pr_debug("%s: stripe %llu checkp: %d\n", __func__,
1607 (unsigned long long)sh->sector, checkp);
1608
1609 count = set_syndrome_sources(srcs, sh);
1610 if (!checkp)
1611 srcs[count] = NULL;
91c00924 1612
91c00924 1613 atomic_inc(&sh->count);
ac6b53b6
DW		 1614 init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check,
1615 sh, to_addr_conv(sh, percpu));
1616 async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE,
1617 &sh->ops.zero_sum_result, percpu->spare_page, &submit);
91c00924
DW		 1618}
1619
51acbcec 1620static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
91c00924
DW		 1621{
1622 int overlap_clear = 0, i, disks = sh->disks;
1623 struct dma_async_tx_descriptor *tx = NULL;
d1688a6d 1624 struct r5conf *conf = sh->raid_conf;
ac6b53b6 1625 int level = conf->level;
d6f38f31
DW		 1626 struct raid5_percpu *percpu;
1627 unsigned long cpu;
91c00924 1628
d6f38f31
DW		 1629 cpu = get_cpu();
1630 percpu = per_cpu_ptr(conf->percpu, cpu);
83de75cc 1631 if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) {
91c00924
DW		 1632 ops_run_biofill(sh);
1633 overlap_clear++;
1634 }
1635
7b3a871e 1636 if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) {
ac6b53b6
DW		 1637 if (level < 6)
1638 tx = ops_run_compute5(sh, percpu);
1639 else {
1640 if (sh->ops.target2 < 0 || sh->ops.target < 0)
1641 tx = ops_run_compute6_1(sh, percpu);
1642 else
1643 tx = ops_run_compute6_2(sh, percpu);
1644 }
1645 /* terminate the chain if reconstruct is not set to be run */
1646 if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request))
7b3a871e
DW		 1647 async_tx_ack(tx);
1648 }
91c00924 1649
600aa109 1650 if (test_bit(STRIPE_OP_PREXOR, &ops_request))
d6f38f31 1651 tx = ops_run_prexor(sh, percpu, tx);
91c00924 1652
600aa109 1653 if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) {
d8ee0728 1654 tx = ops_run_biodrain(sh, tx);
91c00924
DW		 1655 overlap_clear++;
1656 }
1657
ac6b53b6
DW		 1658 if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) {
1659 if (level < 6)
1660 ops_run_reconstruct5(sh, percpu, tx);
1661 else
1662 ops_run_reconstruct6(sh, percpu, tx);
1663 }
91c00924 1664
ac6b53b6
DW		 1665 if (test_bit(STRIPE_OP_CHECK, &ops_request)) {
1666 if (sh->check_state == check_state_run)
1667 ops_run_check_p(sh, percpu);
1668 else if (sh->check_state == check_state_run_q)
1669 ops_run_check_pq(sh, percpu, 0);
1670 else if (sh->check_state == check_state_run_pq)
1671 ops_run_check_pq(sh, percpu, 1);
1672 else
1673 BUG();
1674 }
91c00924 1675
91c00924
DW		 1676 if (overlap_clear)
1677 for (i = disks; i--; ) {
1678 struct r5dev *dev = &sh->dev[i];
1679 if (test_and_clear_bit(R5_Overlap, &dev->flags))
1680 wake_up(&sh->raid_conf->wait_for_overlap);
1681 }
d6f38f31 1682 put_cpu();
91c00924
DW		 1683}
1684
566c09c5 1685static int grow_one_stripe(struct r5conf *conf, int hash)
1da177e4
LT		 1686{
1687 struct stripe_head *sh;
6ce32846 1688 sh = kmem_cache_zalloc(conf->slab_cache, GFP_KERNEL);
3f294f4f
N		 1689 if (!sh)
1690 return 0;
6ce32846 1691
3f294f4f 1692 sh->raid_conf = conf;
3f294f4f 1693
b17459c0
SL		 1694 spin_lock_init(&sh->stripe_lock);
1695
e4e11e38
N		 1696 if (grow_buffers(sh)) {
1697 shrink_buffers(sh);
3f294f4f
N		 1698 kmem_cache_free(conf->slab_cache, sh);
1699 return 0;
1700 }
566c09c5 1701 sh->hash_lock_index = hash;
3f294f4f
N		 1702 /* we just created an active stripe so... */
1703 atomic_set(&sh->count, 1);
1704 atomic_inc(&conf->active_stripes);
1705 INIT_LIST_HEAD(&sh->lru);
1706 release_stripe(sh);
1707 return 1;
1708}
1709
d1688a6d 1710static int grow_stripes(struct r5conf *conf, int num)
3f294f4f 1711{
e18b890b 1712 struct kmem_cache *sc;
5e5e3e78 1713 int devs = max(conf->raid_disks, conf->previous_raid_disks);
566c09c5 1714 int hash;
1da177e4 1715
f4be6b43
N		 1716 if (conf->mddev->gendisk)
1717 sprintf(conf->cache_name[0],
1718 "raid%d-%s", conf->level, mdname(conf->mddev));
1719 else
1720 sprintf(conf->cache_name[0],
1721 "raid%d-%p", conf->level, conf->mddev);
1722 sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]);
1723
ad01c9e3
N		 1724 conf->active_name = 0;
1725 sc = kmem_cache_create(conf->cache_name[conf->active_name],
1da177e4 1726 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
20c2df83 1727 0, 0, NULL);
1da177e4
LT		 1728 if (!sc)
1729 return 1;
1730 conf->slab_cache = sc;
ad01c9e3 1731 conf->pool_size = devs;
566c09c5
SL		 1732 hash = conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS;
1733 while (num--) {
1734 if (!grow_one_stripe(conf, hash))
1da177e4 1735 return 1;
566c09c5
SL		 1736 conf->max_nr_stripes++;
1737 hash = (hash + 1) % NR_STRIPE_HASH_LOCKS;
1738 }
1da177e4
LT		 1739 return 0;
1740}
29269553 1741
d6f38f31
DW		 1742/**
1743 * scribble_len - return the required size of the scribble region
1744 * @num - total number of disks in the array
1745 *
1746 * The size must be enough to contain:
1747 * 1/ a struct page pointer for each device in the array +2
1748 * 2/ room to convert each entry in (1) to its corresponding dma
1749 * (dma_map_page()) or page (page_address()) address.
1750 *
1751 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
1752 * calculate over all devices (not just the data blocks), using zeros in place
1753 * of the P and Q blocks.
1754 */
1755static size_t scribble_len(int num)
1756{
1757 size_t len;
1758
1759 len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2);
1760
1761 return len;
1762}
1763
d1688a6d 1764static int resize_stripes(struct r5conf *conf, int newsize)
ad01c9e3
N		 1765{
1766 /* Make all the stripes able to hold 'newsize' devices.
1767 * New slots in each stripe get 'page' set to a new page.
1768 *
1769 * This happens in stages:
1770 * 1/ create a new kmem_cache and allocate the required number of
1771 * stripe_heads.
83f0d77a 1772 * 2/ gather all the old stripe_heads and transfer the pages across
ad01c9e3
N		 1773 * to the new stripe_heads. This will have the side effect of
1774 * freezing the array as once all stripe_heads have been collected,
1775 * no IO will be possible. Old stripe heads are freed once their
1776 * pages have been transferred over, and the old kmem_cache is
1777 * freed when all stripes are done.
1778 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
1779 * we simple return a failre status - no need to clean anything up.
1780 * 4/ allocate new pages for the new slots in the new stripe_heads.
1781 * If this fails, we don't bother trying the shrink the
1782 * stripe_heads down again, we just leave them as they are.
1783 * As each stripe_head is processed the new one is released into
1784 * active service.
1785 *
1786 * Once step2 is started, we cannot afford to wait for a write,
1787 * so we use GFP_NOIO allocations.
1788 */
1789 struct stripe_head *osh, *nsh;
1790 LIST_HEAD(newstripes);
1791 struct disk_info *ndisks;
d6f38f31 1792 unsigned long cpu;
b5470dc5 1793 int err;
e18b890b 1794 struct kmem_cache *sc;
ad01c9e3 1795 int i;
566c09c5 1796 int hash, cnt;
ad01c9e3
N		 1797
1798 if (newsize <= conf->pool_size)
1799 return 0; /* never bother to shrink */
1800
b5470dc5
DW		 1801 err = md_allow_write(conf->mddev);
1802 if (err)
1803 return err;
2a2275d6 1804
ad01c9e3
N		 1805 /* Step 1 */
1806 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
1807 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
20c2df83 1808 0, 0, NULL);
ad01c9e3
N		 1809 if (!sc)
1810 return -ENOMEM;
1811
1812 for (i = conf->max_nr_stripes; i; i--) {
6ce32846 1813 nsh = kmem_cache_zalloc(sc, GFP_KERNEL);
ad01c9e3
N		 1814 if (!nsh)
1815 break;
1816
ad01c9e3 1817 nsh->raid_conf = conf;
cb13ff69 1818 spin_lock_init(&nsh->stripe_lock);
ad01c9e3
N		 1819
1820 list_add(&nsh->lru, &newstripes);
1821 }
1822 if (i) {
1823 /* didn't get enough, give up */
1824 while (!list_empty(&newstripes)) {
1825 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1826 list_del(&nsh->lru);
1827 kmem_cache_free(sc, nsh);
1828 }
1829 kmem_cache_destroy(sc);
1830 return -ENOMEM;
1831 }
1832 /* Step 2 - Must use GFP_NOIO now.
1833 * OK, we have enough stripes, start collecting inactive
1834 * stripes and copying them over
1835 */
566c09c5
SL		 1836 hash = 0;
1837 cnt = 0;
ad01c9e3 1838 list_for_each_entry(nsh, &newstripes, lru) {
566c09c5
SL		 1839 lock_device_hash_lock(conf, hash);
1840 wait_event_cmd(conf->wait_for_stripe,
1841 !list_empty(conf->inactive_list + hash),
1842 unlock_device_hash_lock(conf, hash),
1843 lock_device_hash_lock(conf, hash));
1844 osh = get_free_stripe(conf, hash);
1845 unlock_device_hash_lock(conf, hash);
ad01c9e3
N		 1846 atomic_set(&nsh->count, 1);
1847 for(i=0; i<conf->pool_size; i++)
1848 nsh->dev[i].page = osh->dev[i].page;
1849 for( ; i<newsize; i++)
1850 nsh->dev[i].page = NULL;
566c09c5 1851 nsh->hash_lock_index = hash;
ad01c9e3 1852 kmem_cache_free(conf->slab_cache, osh);
566c09c5
SL		 1853 cnt++;
1854 if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS +
1855 !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) {
1856 hash++;
1857 cnt = 0;
1858 }
ad01c9e3
N		 1859 }
1860 kmem_cache_destroy(conf->slab_cache);
1861
1862 /* Step 3.
1863 * At this point, we are holding all the stripes so the array
1864 * is completely stalled, so now is a good time to resize
d6f38f31 1865 * conf->disks and the scribble region
ad01c9e3
N		 1866 */
1867 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
1868 if (ndisks) {
1869 for (i=0; i<conf->raid_disks; i++)
1870 ndisks[i] = conf->disks[i];
1871 kfree(conf->disks);
1872 conf->disks = ndisks;
1873 } else
1874 err = -ENOMEM;
1875
d6f38f31
DW		 1876 get_online_cpus();
1877 conf->scribble_len = scribble_len(newsize);
1878 for_each_present_cpu(cpu) {
1879 struct raid5_percpu *percpu;
1880 void *scribble;
1881
1882 percpu = per_cpu_ptr(conf->percpu, cpu);
1883 scribble = kmalloc(conf->scribble_len, GFP_NOIO);
1884
1885 if (scribble) {
1886 kfree(percpu->scribble);
1887 percpu->scribble = scribble;
1888 } else {
1889 err = -ENOMEM;
1890 break;
1891 }
1892 }
1893 put_online_cpus();
1894
ad01c9e3
N		 1895 /* Step 4, return new stripes to service */
1896 while(!list_empty(&newstripes)) {
1897 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1898 list_del_init(&nsh->lru);
d6f38f31 1899
ad01c9e3
N		 1900 for (i=conf->raid_disks; i < newsize; i++)
1901 if (nsh->dev[i].page == NULL) {
1902 struct page *p = alloc_page(GFP_NOIO);
1903 nsh->dev[i].page = p;
1904 if (!p)
1905 err = -ENOMEM;
1906 }
1907 release_stripe(nsh);
1908 }
1909 /* critical section pass, GFP_NOIO no longer needed */
1910
1911 conf->slab_cache = sc;
1912 conf->active_name = 1-conf->active_name;
1913 conf->pool_size = newsize;
1914 return err;
1915}
1da177e4 1916
566c09c5 1917static int drop_one_stripe(struct r5conf *conf, int hash)
1da177e4
LT		 1918{
1919 struct stripe_head *sh;
1920
566c09c5
SL		 1921 spin_lock_irq(conf->hash_locks + hash);
1922 sh = get_free_stripe(conf, hash);
1923 spin_unlock_irq(conf->hash_locks + hash);
3f294f4f
N		 1924 if (!sh)
1925 return 0;
78bafebd 1926 BUG_ON(atomic_read(&sh->count));
e4e11e38 1927 shrink_buffers(sh);
3f294f4f
N		 1928 kmem_cache_free(conf->slab_cache, sh);
1929 atomic_dec(&conf->active_stripes);
1930 return 1;
1931}
1932
d1688a6d 1933static void shrink_stripes(struct r5conf *conf)
3f294f4f 1934{
566c09c5
SL		 1935 int hash;
1936 for (hash = 0; hash < NR_STRIPE_HASH_LOCKS; hash++)
1937 while (drop_one_stripe(conf, hash))
1938 ;
3f294f4f 1939
29fc7e3e
N		 1940 if (conf->slab_cache)
1941 kmem_cache_destroy(conf->slab_cache);
1da177e4
LT		 1942 conf->slab_cache = NULL;
1943}
1944
6712ecf8 1945static void raid5_end_read_request(struct bio * bi, int error)
1da177e4 1946{
99c0fb5f 1947 struct stripe_head *sh = bi->bi_private;
d1688a6d 1948 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 1949 int disks = sh->disks, i;
1da177e4 1950 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
d6950432 1951 char b[BDEVNAME_SIZE];
dd054fce 1952 struct md_rdev *rdev = NULL;
05616be5 1953 sector_t s;
1da177e4
LT		 1954
1955 for (i=0 ; i<disks; i++)
1956 if (bi == &sh->dev[i].req)
1957 break;
1958
45b4233c
DW		 1959 pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n",
1960 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
1da177e4
LT		 1961 uptodate);
1962 if (i == disks) {
1963 BUG();
6712ecf8 1964 return;
1da177e4 1965 }
14a75d3e 1966 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
dd054fce
N		 1967 /* If replacement finished while this request was outstanding,
1968 * 'replacement' might be NULL already.
1969 * In that case it moved down to 'rdev'.
1970 * rdev is not removed until all requests are finished.
1971 */
14a75d3e 1972 rdev = conf->disks[i].replacement;
dd054fce 1973 if (!rdev)
14a75d3e 1974 rdev = conf->disks[i].rdev;
1da177e4 1975
05616be5
N		 1976 if (use_new_offset(conf, sh))
1977 s = sh->sector + rdev->new_data_offset;
1978 else
1979 s = sh->sector + rdev->data_offset;
1da177e4 1980 if (uptodate) {
1da177e4 1981 set_bit(R5_UPTODATE, &sh->dev[i].flags);
4e5314b5 1982 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
14a75d3e
N		 1983 /* Note that this cannot happen on a
1984 * replacement device. We just fail those on
1985 * any error
1986 */
8bda470e
CD		 1987 printk_ratelimited(
1988 KERN_INFO
1989 "md/raid:%s: read error corrected"
1990 " (%lu sectors at %llu on %s)\n",
1991 mdname(conf->mddev), STRIPE_SECTORS,
05616be5 1992 (unsigned long long)s,
8bda470e 1993 bdevname(rdev->bdev, b));
ddd5115f 1994 atomic_add(STRIPE_SECTORS, &rdev->corrected_errors);
4e5314b5
N		 1995 clear_bit(R5_ReadError, &sh->dev[i].flags);
1996 clear_bit(R5_ReWrite, &sh->dev[i].flags);
3f9e7c14 1997 } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
1998 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
1999
14a75d3e
N		 2000 if (atomic_read(&rdev->read_errors))
2001 atomic_set(&rdev->read_errors, 0);
1da177e4 2002 } else {
14a75d3e 2003 const char *bdn = bdevname(rdev->bdev, b);
ba22dcbf 2004 int retry = 0;
2e8ac303 2005 int set_bad = 0;
d6950432 2006
1da177e4 2007 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
d6950432 2008 atomic_inc(&rdev->read_errors);
14a75d3e
N		 2009 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
2010 printk_ratelimited(
2011 KERN_WARNING
2012 "md/raid:%s: read error on replacement device "
2013 "(sector %llu on %s).\n",
2014 mdname(conf->mddev),
05616be5 2015 (unsigned long long)s,
14a75d3e 2016 bdn);
2e8ac303 2017 else if (conf->mddev->degraded >= conf->max_degraded) {
2018 set_bad = 1;
8bda470e
CD		 2019 printk_ratelimited(
2020 KERN_WARNING
2021 "md/raid:%s: read error not correctable "
2022 "(sector %llu on %s).\n",
2023 mdname(conf->mddev),
05616be5 2024 (unsigned long long)s,
8bda470e 2025 bdn);
2e8ac303 2026 } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) {
4e5314b5 2027 /* Oh, no!!! */
2e8ac303 2028 set_bad = 1;
8bda470e
CD		 2029 printk_ratelimited(
2030 KERN_WARNING
2031 "md/raid:%s: read error NOT corrected!! "
2032 "(sector %llu on %s).\n",
2033 mdname(conf->mddev),
05616be5 2034 (unsigned long long)s,
8bda470e 2035 bdn);
2e8ac303 2036 } else if (atomic_read(&rdev->read_errors)
ba22dcbf 2037 > conf->max_nr_stripes)
14f8d26b 2038 printk(KERN_WARNING
0c55e022 2039 "md/raid:%s: Too many read errors, failing device %s.\n",
d6950432 2040 mdname(conf->mddev), bdn);
ba22dcbf
N		 2041 else
2042 retry = 1;
edfa1f65
BY		 2043 if (set_bad && test_bit(In_sync, &rdev->flags)
2044 && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2045 retry = 1;
ba22dcbf 2046 if (retry)
3f9e7c14 2047 if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) {
2048 set_bit(R5_ReadError, &sh->dev[i].flags);
2049 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2050 } else
2051 set_bit(R5_ReadNoMerge, &sh->dev[i].flags);
ba22dcbf 2052 else {
4e5314b5
N		 2053 clear_bit(R5_ReadError, &sh->dev[i].flags);
2054 clear_bit(R5_ReWrite, &sh->dev[i].flags);
2e8ac303 2055 if (!(set_bad
2056 && test_bit(In_sync, &rdev->flags)
2057 && rdev_set_badblocks(
2058 rdev, sh->sector, STRIPE_SECTORS, 0)))
2059 md_error(conf->mddev, rdev);
ba22dcbf 2060 }
1da177e4 2061 }
14a75d3e 2062 rdev_dec_pending(rdev, conf->mddev);
1da177e4
LT		 2063 clear_bit(R5_LOCKED, &sh->dev[i].flags);
2064 set_bit(STRIPE_HANDLE, &sh->state);
2065 release_stripe(sh);
1da177e4
LT		 2066}
2067
d710e138 2068static void raid5_end_write_request(struct bio *bi, int error)
1da177e4 2069{
99c0fb5f 2070 struct stripe_head *sh = bi->bi_private;
d1688a6d 2071 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 2072 int disks = sh->disks, i;
977df362 2073 struct md_rdev *uninitialized_var(rdev);
1da177e4 2074 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
b84db560
N		 2075 sector_t first_bad;
2076 int bad_sectors;
977df362 2077 int replacement = 0;
1da177e4 2078
977df362
N		 2079 for (i = 0 ; i < disks; i++) {
2080 if (bi == &sh->dev[i].req) {
2081 rdev = conf->disks[i].rdev;
1da177e4 2082 break;
977df362
N		 2083 }
2084 if (bi == &sh->dev[i].rreq) {
2085 rdev = conf->disks[i].replacement;
dd054fce
N		 2086 if (rdev)
2087 replacement = 1;
2088 else
2089 /* rdev was removed and 'replacement'
2090 * replaced it. rdev is not removed
2091 * until all requests are finished.
2092 */
2093 rdev = conf->disks[i].rdev;
977df362
N		 2094 break;
2095 }
2096 }
45b4233c 2097 pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n",
1da177e4
LT		 2098 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
2099 uptodate);
2100 if (i == disks) {
2101 BUG();
6712ecf8 2102 return;
1da177e4
LT		 2103 }
2104
977df362
N		 2105 if (replacement) {
2106 if (!uptodate)
2107 md_error(conf->mddev, rdev);
2108 else if (is_badblock(rdev, sh->sector,
2109 STRIPE_SECTORS,
2110 &first_bad, &bad_sectors))
2111 set_bit(R5_MadeGoodRepl, &sh->dev[i].flags);
2112 } else {
2113 if (!uptodate) {
2114 set_bit(WriteErrorSeen, &rdev->flags);
2115 set_bit(R5_WriteError, &sh->dev[i].flags);
3a6de292
N		 2116 if (!test_and_set_bit(WantReplacement, &rdev->flags))
2117 set_bit(MD_RECOVERY_NEEDED,
2118 &rdev->mddev->recovery);
977df362
N		 2119 } else if (is_badblock(rdev, sh->sector,
2120 STRIPE_SECTORS,
c0b32972 2121 &first_bad, &bad_sectors)) {
977df362 2122 set_bit(R5_MadeGood, &sh->dev[i].flags);
c0b32972
N		 2123 if (test_bit(R5_ReadError, &sh->dev[i].flags))
2124 /* That was a successful write so make
2125 * sure it looks like we already did
2126 * a re-write.
2127 */
2128 set_bit(R5_ReWrite, &sh->dev[i].flags);
2129 }
977df362
N		 2130 }
2131 rdev_dec_pending(rdev, conf->mddev);
1da177e4 2132
977df362
N		 2133 if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags))
2134 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1da177e4 2135 set_bit(STRIPE_HANDLE, &sh->state);
c04be0aa 2136 release_stripe(sh);
1da177e4
LT		 2137}
2138
784052ec 2139static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous);
1da177e4 2140
784052ec 2141static void raid5_build_block(struct stripe_head *sh, int i, int previous)
1da177e4
LT		 2142{
2143 struct r5dev *dev = &sh->dev[i];
2144
2145 bio_init(&dev->req);
2146 dev->req.bi_io_vec = &dev->vec;
2147 dev->req.bi_vcnt++;
2148 dev->req.bi_max_vecs++;
1da177e4 2149 dev->req.bi_private = sh;
995c4275 2150 dev->vec.bv_page = dev->page;
1da177e4 2151
977df362
N		 2152 bio_init(&dev->rreq);
2153 dev->rreq.bi_io_vec = &dev->rvec;
2154 dev->rreq.bi_vcnt++;
2155 dev->rreq.bi_max_vecs++;
2156 dev->rreq.bi_private = sh;
2157 dev->rvec.bv_page = dev->page;
2158
1da177e4 2159 dev->flags = 0;
784052ec 2160 dev->sector = compute_blocknr(sh, i, previous);
1da177e4
LT		 2161}
2162
fd01b88c 2163static void error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 2164{
2165 char b[BDEVNAME_SIZE];
d1688a6d 2166 struct r5conf *conf = mddev->private;
908f4fbd 2167 unsigned long flags;
0c55e022 2168 pr_debug("raid456: error called\n");
1da177e4 2169
908f4fbd
N		 2170 spin_lock_irqsave(&conf->device_lock, flags);
2171 clear_bit(In_sync, &rdev->flags);
2172 mddev->degraded = calc_degraded(conf);
2173 spin_unlock_irqrestore(&conf->device_lock, flags);
2174 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2175
de393cde 2176 set_bit(Blocked, &rdev->flags);
6f8d0c77
N		 2177 set_bit(Faulty, &rdev->flags);
2178 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2179 printk(KERN_ALERT
2180 "md/raid:%s: Disk failure on %s, disabling device.\n"
2181 "md/raid:%s: Operation continuing on %d devices.\n",
2182 mdname(mddev),
2183 bdevname(rdev->bdev, b),
2184 mdname(mddev),
2185 conf->raid_disks - mddev->degraded);
16a53ecc 2186}
1da177e4
LT		 2187
2188/*
2189 * Input: a 'big' sector number,
2190 * Output: index of the data and parity disk, and the sector # in them.
2191 */
d1688a6d 2192static sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector,
911d4ee8
N		 2193 int previous, int *dd_idx,
2194 struct stripe_head *sh)
1da177e4 2195{
6e3b96ed 2196 sector_t stripe, stripe2;
35f2a591 2197 sector_t chunk_number;
1da177e4 2198 unsigned int chunk_offset;
911d4ee8 2199 int pd_idx, qd_idx;
67cc2b81 2200 int ddf_layout = 0;
1da177e4 2201 sector_t new_sector;
e183eaed
N		 2202 int algorithm = previous ? conf->prev_algo
2203 : conf->algorithm;
09c9e5fa
AN		 2204 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2205 : conf->chunk_sectors;
112bf897
N		 2206 int raid_disks = previous ? conf->previous_raid_disks
2207 : conf->raid_disks;
2208 int data_disks = raid_disks - conf->max_degraded;
1da177e4
LT		 2209
2210 /* First compute the information on this sector */
2211
2212 /*
2213 * Compute the chunk number and the sector offset inside the chunk
2214 */
2215 chunk_offset = sector_div(r_sector, sectors_per_chunk);
2216 chunk_number = r_sector;
1da177e4
LT		 2217
2218 /*
2219 * Compute the stripe number
2220 */
35f2a591
N		 2221 stripe = chunk_number;
2222 *dd_idx = sector_div(stripe, data_disks);
6e3b96ed 2223 stripe2 = stripe;
1da177e4
LT		 2224 /*
2225 * Select the parity disk based on the user selected algorithm.
2226 */
84789554 2227 pd_idx = qd_idx = -1;
16a53ecc
N		 2228 switch(conf->level) {
2229 case 4:
911d4ee8 2230 pd_idx = data_disks;
16a53ecc
N		 2231 break;
2232 case 5:
e183eaed 2233 switch (algorithm) {
1da177e4 2234 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2235 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2236 if (*dd_idx >= pd_idx)
1da177e4
LT		 2237 (*dd_idx)++;
2238 break;
2239 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2240 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2241 if (*dd_idx >= pd_idx)
1da177e4
LT		 2242 (*dd_idx)++;
2243 break;
2244 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2245 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2246 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4
LT		 2247 break;
2248 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2249 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2250 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4 2251 break;
99c0fb5f
N		 2252 case ALGORITHM_PARITY_0:
2253 pd_idx = 0;
2254 (*dd_idx)++;
2255 break;
2256 case ALGORITHM_PARITY_N:
2257 pd_idx = data_disks;
2258 break;
1da177e4 2259 default:
99c0fb5f 2260 BUG();
16a53ecc
N		 2261 }
2262 break;
2263 case 6:
2264
e183eaed 2265 switch (algorithm) {
16a53ecc 2266 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2267 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2268 qd_idx = pd_idx + 1;
2269 if (pd_idx == raid_disks-1) {
99c0fb5f 2270 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2271 qd_idx = 0;
2272 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2273 (*dd_idx) += 2; /* D D P Q D */
2274 break;
2275 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2276 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2277 qd_idx = pd_idx + 1;
2278 if (pd_idx == raid_disks-1) {
99c0fb5f 2279 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2280 qd_idx = 0;
2281 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2282 (*dd_idx) += 2; /* D D P Q D */
2283 break;
2284 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2285 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2286 qd_idx = (pd_idx + 1) % raid_disks;
2287 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc
N		 2288 break;
2289 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2290 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2291 qd_idx = (pd_idx + 1) % raid_disks;
2292 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc 2293 break;
99c0fb5f
N		 2294
2295 case ALGORITHM_PARITY_0:
2296 pd_idx = 0;
2297 qd_idx = 1;
2298 (*dd_idx) += 2;
2299 break;
2300 case ALGORITHM_PARITY_N:
2301 pd_idx = data_disks;
2302 qd_idx = data_disks + 1;
2303 break;
2304
2305 case ALGORITHM_ROTATING_ZERO_RESTART:
2306 /* Exactly the same as RIGHT_ASYMMETRIC, but or
2307 * of blocks for computing Q is different.
2308 */
6e3b96ed 2309 pd_idx = sector_div(stripe2, raid_disks);
99c0fb5f
N		 2310 qd_idx = pd_idx + 1;
2311 if (pd_idx == raid_disks-1) {
2312 (*dd_idx)++; /* Q D D D P */
2313 qd_idx = 0;
2314 } else if (*dd_idx >= pd_idx)
2315 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2316 ddf_layout = 1;
99c0fb5f
N		 2317 break;
2318
2319 case ALGORITHM_ROTATING_N_RESTART:
2320 /* Same a left_asymmetric, by first stripe is
2321 * D D D P Q rather than
2322 * Q D D D P
2323 */
6e3b96ed
N		 2324 stripe2 += 1;
2325 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2326 qd_idx = pd_idx + 1;
2327 if (pd_idx == raid_disks-1) {
2328 (*dd_idx)++; /* Q D D D P */
2329 qd_idx = 0;
2330 } else if (*dd_idx >= pd_idx)
2331 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2332 ddf_layout = 1;
99c0fb5f
N		 2333 break;
2334
2335 case ALGORITHM_ROTATING_N_CONTINUE:
2336 /* Same as left_symmetric but Q is before P */
6e3b96ed 2337 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2338 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
2339 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
67cc2b81 2340 ddf_layout = 1;
99c0fb5f
N		 2341 break;
2342
2343 case ALGORITHM_LEFT_ASYMMETRIC_6:
2344 /* RAID5 left_asymmetric, with Q on last device */
6e3b96ed 2345 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2346 if (*dd_idx >= pd_idx)
2347 (*dd_idx)++;
2348 qd_idx = raid_disks - 1;
2349 break;
2350
2351 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6e3b96ed 2352 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2353 if (*dd_idx >= pd_idx)
2354 (*dd_idx)++;
2355 qd_idx = raid_disks - 1;
2356 break;
2357
2358 case ALGORITHM_LEFT_SYMMETRIC_6:
6e3b96ed 2359 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2360 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2361 qd_idx = raid_disks - 1;
2362 break;
2363
2364 case ALGORITHM_RIGHT_SYMMETRIC_6:
6e3b96ed 2365 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2366 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2367 qd_idx = raid_disks - 1;
2368 break;
2369
2370 case ALGORITHM_PARITY_0_6:
2371 pd_idx = 0;
2372 (*dd_idx)++;
2373 qd_idx = raid_disks - 1;
2374 break;
2375
16a53ecc 2376 default:
99c0fb5f 2377 BUG();
16a53ecc
N		 2378 }
2379 break;
1da177e4
LT		 2380 }
2381
911d4ee8
N		 2382 if (sh) {
2383 sh->pd_idx = pd_idx;
2384 sh->qd_idx = qd_idx;
67cc2b81 2385 sh->ddf_layout = ddf_layout;
911d4ee8 2386 }
1da177e4
LT		 2387 /*
2388 * Finally, compute the new sector number
2389 */
2390 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
2391 return new_sector;
2392}
2393
2394
784052ec 2395static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous)
1da177e4 2396{
d1688a6d 2397 struct r5conf *conf = sh->raid_conf;
b875e531
N		 2398 int raid_disks = sh->disks;
2399 int data_disks = raid_disks - conf->max_degraded;
1da177e4 2400 sector_t new_sector = sh->sector, check;
09c9e5fa
AN		 2401 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2402 : conf->chunk_sectors;
e183eaed
N		 2403 int algorithm = previous ? conf->prev_algo
2404 : conf->algorithm;
1da177e4
LT		 2405 sector_t stripe;
2406 int chunk_offset;
35f2a591
N		 2407 sector_t chunk_number;
2408 int dummy1, dd_idx = i;
1da177e4 2409 sector_t r_sector;
911d4ee8 2410 struct stripe_head sh2;
1da177e4 2411
16a53ecc 2412
1da177e4
LT		 2413 chunk_offset = sector_div(new_sector, sectors_per_chunk);
2414 stripe = new_sector;
1da177e4 2415
16a53ecc
N		 2416 if (i == sh->pd_idx)
2417 return 0;
2418 switch(conf->level) {
2419 case 4: break;
2420 case 5:
e183eaed 2421 switch (algorithm) {
1da177e4
LT		 2422 case ALGORITHM_LEFT_ASYMMETRIC:
2423 case ALGORITHM_RIGHT_ASYMMETRIC:
2424 if (i > sh->pd_idx)
2425 i--;
2426 break;
2427 case ALGORITHM_LEFT_SYMMETRIC:
2428 case ALGORITHM_RIGHT_SYMMETRIC:
2429 if (i < sh->pd_idx)
2430 i += raid_disks;
2431 i -= (sh->pd_idx + 1);
2432 break;
99c0fb5f
N		 2433 case ALGORITHM_PARITY_0:
2434 i -= 1;
2435 break;
2436 case ALGORITHM_PARITY_N:
2437 break;
1da177e4 2438 default:
99c0fb5f 2439 BUG();
16a53ecc
N		 2440 }
2441 break;
2442 case 6:
d0dabf7e 2443 if (i == sh->qd_idx)
16a53ecc 2444 return 0; /* It is the Q disk */
e183eaed 2445 switch (algorithm) {
16a53ecc
N		 2446 case ALGORITHM_LEFT_ASYMMETRIC:
2447 case ALGORITHM_RIGHT_ASYMMETRIC:
99c0fb5f
N		 2448 case ALGORITHM_ROTATING_ZERO_RESTART:
2449 case ALGORITHM_ROTATING_N_RESTART:
2450 if (sh->pd_idx == raid_disks-1)
2451 i--; /* Q D D D P */
16a53ecc
N		 2452 else if (i > sh->pd_idx)
2453 i -= 2; /* D D P Q D */
2454 break;
2455 case ALGORITHM_LEFT_SYMMETRIC:
2456 case ALGORITHM_RIGHT_SYMMETRIC:
2457 if (sh->pd_idx == raid_disks-1)
2458 i--; /* Q D D D P */
2459 else {
2460 /* D D P Q D */
2461 if (i < sh->pd_idx)
2462 i += raid_disks;
2463 i -= (sh->pd_idx + 2);
2464 }
2465 break;
99c0fb5f
N		 2466 case ALGORITHM_PARITY_0:
2467 i -= 2;
2468 break;
2469 case ALGORITHM_PARITY_N:
2470 break;
2471 case ALGORITHM_ROTATING_N_CONTINUE:
e4424fee 2472 /* Like left_symmetric, but P is before Q */
99c0fb5f
N		 2473 if (sh->pd_idx == 0)
2474 i--; /* P D D D Q */
e4424fee
N		 2475 else {
2476 /* D D Q P D */
2477 if (i < sh->pd_idx)
2478 i += raid_disks;
2479 i -= (sh->pd_idx + 1);
2480 }
99c0fb5f
N		 2481 break;
2482 case ALGORITHM_LEFT_ASYMMETRIC_6:
2483 case ALGORITHM_RIGHT_ASYMMETRIC_6:
2484 if (i > sh->pd_idx)
2485 i--;
2486 break;
2487 case ALGORITHM_LEFT_SYMMETRIC_6:
2488 case ALGORITHM_RIGHT_SYMMETRIC_6:
2489 if (i < sh->pd_idx)
2490 i += data_disks + 1;
2491 i -= (sh->pd_idx + 1);
2492 break;
2493 case ALGORITHM_PARITY_0_6:
2494 i -= 1;
2495 break;
16a53ecc 2496 default:
99c0fb5f 2497 BUG();
16a53ecc
N		 2498 }
2499 break;
1da177e4
LT		 2500 }
2501
2502 chunk_number = stripe * data_disks + i;
35f2a591 2503 r_sector = chunk_number * sectors_per_chunk + chunk_offset;
1da177e4 2504
112bf897 2505 check = raid5_compute_sector(conf, r_sector,
784052ec 2506 previous, &dummy1, &sh2);
911d4ee8
N		 2507 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
2508 || sh2.qd_idx != sh->qd_idx) {
0c55e022
N		 2509 printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n",
2510 mdname(conf->mddev));
1da177e4
LT		 2511 return 0;
2512 }
2513 return r_sector;
2514}
2515
2516
600aa109 2517static void
c0f7bddb 2518schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
600aa109 2519 int rcw, int expand)
e33129d8
DW		 2520{
2521 int i, pd_idx = sh->pd_idx, disks = sh->disks;
d1688a6d 2522 struct r5conf *conf = sh->raid_conf;
c0f7bddb 2523 int level = conf->level;
e33129d8
DW		 2524
2525 if (rcw) {
e33129d8
DW		 2526
2527 for (i = disks; i--; ) {
2528 struct r5dev *dev = &sh->dev[i];
2529
2530 if (dev->towrite) {
2531 set_bit(R5_LOCKED, &dev->flags);
d8ee0728 2532 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2533 if (!expand)
2534 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2535 s->locked++;
e33129d8
DW		 2536 }
2537 }
ce7d363a
N		 2538 /* if we are not expanding this is a proper write request, and
2539 * there will be bios with new data to be drained into the
2540 * stripe cache
2541 */
2542 if (!expand) {
2543 if (!s->locked)
2544 /* False alarm, nothing to do */
2545 return;
2546 sh->reconstruct_state = reconstruct_state_drain_run;
2547 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2548 } else
2549 sh->reconstruct_state = reconstruct_state_run;
2550
2551 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
2552
c0f7bddb 2553 if (s->locked + conf->max_degraded == disks)
8b3e6cdc 2554 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
c0f7bddb 2555 atomic_inc(&conf->pending_full_writes);
e33129d8 2556 } else {
c0f7bddb 2557 BUG_ON(level == 6);
e33129d8
DW		 2558 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
2559 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
2560
e33129d8
DW		 2561 for (i = disks; i--; ) {
2562 struct r5dev *dev = &sh->dev[i];
2563 if (i == pd_idx)
2564 continue;
2565
e33129d8
DW		 2566 if (dev->towrite &&
2567 (test_bit(R5_UPTODATE, &dev->flags) ||
d8ee0728
DW		 2568 test_bit(R5_Wantcompute, &dev->flags))) {
2569 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2570 set_bit(R5_LOCKED, &dev->flags);
2571 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2572 s->locked++;
e33129d8
DW		 2573 }
2574 }
ce7d363a
N		 2575 if (!s->locked)
2576 /* False alarm - nothing to do */
2577 return;
2578 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
2579 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
2580 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2581 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
e33129d8
DW		 2582 }
2583
c0f7bddb 2584 /* keep the parity disk(s) locked while asynchronous operations
e33129d8
DW		 2585 * are in flight
2586 */
2587 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
2588 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
600aa109 2589 s->locked++;
e33129d8 2590
c0f7bddb
YT		 2591 if (level == 6) {
2592 int qd_idx = sh->qd_idx;
2593 struct r5dev *dev = &sh->dev[qd_idx];
2594
2595 set_bit(R5_LOCKED, &dev->flags);
2596 clear_bit(R5_UPTODATE, &dev->flags);
2597 s->locked++;
2598 }
2599
600aa109 2600 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
e46b272b 2601 __func__, (unsigned long long)sh->sector,
600aa109 2602 s->locked, s->ops_request);
e33129d8 2603}
16a53ecc 2604
1da177e4
LT		 2605/*
2606 * Each stripe/dev can have one or more bion attached.
16a53ecc 2607 * toread/towrite point to the first in a chain.
1da177e4
LT		 2608 * The bi_next chain must be in order.
2609 */
2610static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite)
2611{
2612 struct bio **bip;
d1688a6d 2613 struct r5conf *conf = sh->raid_conf;
72626685 2614 int firstwrite=0;
1da177e4 2615
cbe47ec5 2616 pr_debug("adding bi b#%llu to stripe s#%llu\n",
1da177e4
LT		 2617 (unsigned long long)bi->bi_sector,
2618 (unsigned long long)sh->sector);
2619
b17459c0
SL		 2620 /*
2621 * If several bio share a stripe. The bio bi_phys_segments acts as a
2622 * reference count to avoid race. The reference count should already be
2623 * increased before this function is called (for example, in
2624 * make_request()), so other bio sharing this stripe will not free the
2625 * stripe. If a stripe is owned by one stripe, the stripe lock will
2626 * protect it.
2627 */
2628 spin_lock_irq(&sh->stripe_lock);
72626685 2629 if (forwrite) {
1da177e4 2630 bip = &sh->dev[dd_idx].towrite;
7eaf7e8e 2631 if (*bip == NULL)
72626685
N		 2632 firstwrite = 1;
2633 } else
1da177e4
LT		 2634 bip = &sh->dev[dd_idx].toread;
2635 while (*bip && (*bip)->bi_sector < bi->bi_sector) {
f73a1c7d 2636 if (bio_end_sector(*bip) > bi->bi_sector)
1da177e4
LT		 2637 goto overlap;
2638 bip = & (*bip)->bi_next;
2639 }
f73a1c7d 2640 if (*bip && (*bip)->bi_sector < bio_end_sector(bi))
1da177e4
LT		 2641 goto overlap;
2642
78bafebd 2643 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
1da177e4
LT		 2644 if (*bip)
2645 bi->bi_next = *bip;
2646 *bip = bi;
e7836bd6 2647 raid5_inc_bi_active_stripes(bi);
72626685 2648
1da177e4
LT		 2649 if (forwrite) {
2650 /* check if page is covered */
2651 sector_t sector = sh->dev[dd_idx].sector;
2652 for (bi=sh->dev[dd_idx].towrite;
2653 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
2654 bi && bi->bi_sector <= sector;
2655 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
f73a1c7d
KO		 2656 if (bio_end_sector(bi) >= sector)
2657 sector = bio_end_sector(bi);
1da177e4
LT		 2658 }
2659 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
2660 set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags);
2661 }
cbe47ec5
N		 2662
2663 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
2664 (unsigned long long)(*bip)->bi_sector,
2665 (unsigned long long)sh->sector, dd_idx);
b97390ae 2666 spin_unlock_irq(&sh->stripe_lock);
cbe47ec5
N		 2667
2668 if (conf->mddev->bitmap && firstwrite) {
2669 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
2670 STRIPE_SECTORS, 0);
2671 sh->bm_seq = conf->seq_flush+1;
2672 set_bit(STRIPE_BIT_DELAY, &sh->state);
2673 }
1da177e4
LT		 2674 return 1;
2675
2676 overlap:
2677 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
b17459c0 2678 spin_unlock_irq(&sh->stripe_lock);
1da177e4
LT		 2679 return 0;
2680}
2681
d1688a6d 2682static void end_reshape(struct r5conf *conf);
29269553 2683
d1688a6d 2684static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
911d4ee8 2685 struct stripe_head *sh)
ccfcc3c1 2686{
784052ec 2687 int sectors_per_chunk =
09c9e5fa 2688 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
911d4ee8 2689 int dd_idx;
2d2063ce 2690 int chunk_offset = sector_div(stripe, sectors_per_chunk);
112bf897 2691 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
2d2063ce 2692
112bf897
N		 2693 raid5_compute_sector(conf,
2694 stripe * (disks - conf->max_degraded)
b875e531 2695 *sectors_per_chunk + chunk_offset,
112bf897 2696 previous,
911d4ee8 2697 &dd_idx, sh);
ccfcc3c1
N		 2698}
2699
a4456856 2700static void
d1688a6d 2701handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh,
a4456856
DW		 2702 struct stripe_head_state *s, int disks,
2703 struct bio **return_bi)
2704{
2705 int i;
2706 for (i = disks; i--; ) {
2707 struct bio *bi;
2708 int bitmap_end = 0;
2709
2710 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
3cb03002 2711 struct md_rdev *rdev;
a4456856
DW		 2712 rcu_read_lock();
2713 rdev = rcu_dereference(conf->disks[i].rdev);
2714 if (rdev && test_bit(In_sync, &rdev->flags))
7f0da59b
N		 2715 atomic_inc(&rdev->nr_pending);
2716 else
2717 rdev = NULL;
a4456856 2718 rcu_read_unlock();
7f0da59b
N		 2719 if (rdev) {
2720 if (!rdev_set_badblocks(
2721 rdev,
2722 sh->sector,
2723 STRIPE_SECTORS, 0))
2724 md_error(conf->mddev, rdev);
2725 rdev_dec_pending(rdev, conf->mddev);
2726 }
a4456856 2727 }
b17459c0 2728 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 2729 /* fail all writes first */
2730 bi = sh->dev[i].towrite;
2731 sh->dev[i].towrite = NULL;
b17459c0 2732 spin_unlock_irq(&sh->stripe_lock);
1ed850f3 2733 if (bi)
a4456856 2734 bitmap_end = 1;
a4456856
DW		 2735
2736 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2737 wake_up(&conf->wait_for_overlap);
2738
2739 while (bi && bi->bi_sector <
2740 sh->dev[i].sector + STRIPE_SECTORS) {
2741 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
2742 clear_bit(BIO_UPTODATE, &bi->bi_flags);
e7836bd6 2743 if (!raid5_dec_bi_active_stripes(bi)) {
a4456856
DW		 2744 md_write_end(conf->mddev);
2745 bi->bi_next = *return_bi;
2746 *return_bi = bi;
2747 }
2748 bi = nextbi;
2749 }
7eaf7e8e
SL		 2750 if (bitmap_end)
2751 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
2752 STRIPE_SECTORS, 0, 0);
2753 bitmap_end = 0;
a4456856
DW		 2754 /* and fail all 'written' */
2755 bi = sh->dev[i].written;
2756 sh->dev[i].written = NULL;
2757 if (bi) bitmap_end = 1;
2758 while (bi && bi->bi_sector <
2759 sh->dev[i].sector + STRIPE_SECTORS) {
2760 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
2761 clear_bit(BIO_UPTODATE, &bi->bi_flags);
e7836bd6 2762 if (!raid5_dec_bi_active_stripes(bi)) {
a4456856
DW		 2763 md_write_end(conf->mddev);
2764 bi->bi_next = *return_bi;
2765 *return_bi = bi;
2766 }
2767 bi = bi2;
2768 }
2769
b5e98d65
DW		 2770 /* fail any reads if this device is non-operational and
2771 * the data has not reached the cache yet.
2772 */
2773 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
2774 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
2775 test_bit(R5_ReadError, &sh->dev[i].flags))) {
143c4d05 2776 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 2777 bi = sh->dev[i].toread;
2778 sh->dev[i].toread = NULL;
143c4d05 2779 spin_unlock_irq(&sh->stripe_lock);
a4456856
DW		 2780 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2781 wake_up(&conf->wait_for_overlap);
a4456856
DW		 2782 while (bi && bi->bi_sector <
2783 sh->dev[i].sector + STRIPE_SECTORS) {
2784 struct bio *nextbi =
2785 r5_next_bio(bi, sh->dev[i].sector);
2786 clear_bit(BIO_UPTODATE, &bi->bi_flags);
e7836bd6 2787 if (!raid5_dec_bi_active_stripes(bi)) {
a4456856
DW		 2788 bi->bi_next = *return_bi;
2789 *return_bi = bi;
2790 }
2791 bi = nextbi;
2792 }
2793 }
a4456856
DW		 2794 if (bitmap_end)
2795 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
2796 STRIPE_SECTORS, 0, 0);
8cfa7b0f
N		 2797 /* If we were in the middle of a write the parity block might
2798 * still be locked - so just clear all R5_LOCKED flags
2799 */
2800 clear_bit(R5_LOCKED, &sh->dev[i].flags);
a4456856
DW		 2801 }
2802
8b3e6cdc
DW		 2803 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2804 if (atomic_dec_and_test(&conf->pending_full_writes))
2805 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 2806}
2807
7f0da59b 2808static void
d1688a6d 2809handle_failed_sync(struct r5conf *conf, struct stripe_head *sh,
7f0da59b
N		 2810 struct stripe_head_state *s)
2811{
2812 int abort = 0;
2813 int i;
2814
7f0da59b 2815 clear_bit(STRIPE_SYNCING, &sh->state);
f8dfcffd
N		 2816 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
2817 wake_up(&conf->wait_for_overlap);
7f0da59b 2818 s->syncing = 0;
9a3e1101 2819 s->replacing = 0;
7f0da59b 2820 /* There is nothing more to do for sync/check/repair.
18b9837e
N		 2821 * Don't even need to abort as that is handled elsewhere
2822 * if needed, and not always wanted e.g. if there is a known
2823 * bad block here.
9a3e1101 2824 * For recover/replace we need to record a bad block on all
7f0da59b
N		 2825 * non-sync devices, or abort the recovery
2826 */
18b9837e
N		 2827 if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) {
2828 /* During recovery devices cannot be removed, so
2829 * locking and refcounting of rdevs is not needed
2830 */
2831 for (i = 0; i < conf->raid_disks; i++) {
2832 struct md_rdev *rdev = conf->disks[i].rdev;
2833 if (rdev
2834 && !test_bit(Faulty, &rdev->flags)
2835 && !test_bit(In_sync, &rdev->flags)
2836 && !rdev_set_badblocks(rdev, sh->sector,
2837 STRIPE_SECTORS, 0))
2838 abort = 1;
2839 rdev = conf->disks[i].replacement;
2840 if (rdev
2841 && !test_bit(Faulty, &rdev->flags)
2842 && !test_bit(In_sync, &rdev->flags)
2843 && !rdev_set_badblocks(rdev, sh->sector,
2844 STRIPE_SECTORS, 0))
2845 abort = 1;
2846 }
2847 if (abort)
2848 conf->recovery_disabled =
2849 conf->mddev->recovery_disabled;
7f0da59b 2850 }
18b9837e 2851 md_done_sync(conf->mddev, STRIPE_SECTORS, !abort);
7f0da59b
N		 2852}
2853
9a3e1101
N		 2854static int want_replace(struct stripe_head *sh, int disk_idx)
2855{
2856 struct md_rdev *rdev;
2857 int rv = 0;
2858 /* Doing recovery so rcu locking not required */
2859 rdev = sh->raid_conf->disks[disk_idx].replacement;
2860 if (rdev
2861 && !test_bit(Faulty, &rdev->flags)
2862 && !test_bit(In_sync, &rdev->flags)
2863 && (rdev->recovery_offset <= sh->sector
2864 || rdev->mddev->recovery_cp <= sh->sector))
2865 rv = 1;
2866
2867 return rv;
2868}
2869
93b3dbce 2870/* fetch_block - checks the given member device to see if its data needs
1fe797e6
DW		 2871 * to be read or computed to satisfy a request.
2872 *
2873 * Returns 1 when no more member devices need to be checked, otherwise returns
93b3dbce 2874 * 0 to tell the loop in handle_stripe_fill to continue
f38e1219 2875 */
93b3dbce
N		 2876static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s,
2877 int disk_idx, int disks)
a4456856 2878{
5599becc 2879 struct r5dev *dev = &sh->dev[disk_idx];
f2b3b44d
N		 2880 struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]],
2881 &sh->dev[s->failed_num[1]] };
5599becc 2882
93b3dbce 2883 /* is the data in this block needed, and can we get it? */
5599becc
YT		 2884 if (!test_bit(R5_LOCKED, &dev->flags) &&
2885 !test_bit(R5_UPTODATE, &dev->flags) &&
2886 (dev->toread ||
2887 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
2888 s->syncing || s->expanding ||
9a3e1101 2889 (s->replacing && want_replace(sh, disk_idx)) ||
5d35e09c
N		 2890 (s->failed >= 1 && fdev[0]->toread) ||
2891 (s->failed >= 2 && fdev[1]->toread) ||
93b3dbce
N		 2892 (sh->raid_conf->level <= 5 && s->failed && fdev[0]->towrite &&
2893 !test_bit(R5_OVERWRITE, &fdev[0]->flags)) ||
2894 (sh->raid_conf->level == 6 && s->failed && s->to_write))) {
5599becc
YT		 2895 /* we would like to get this block, possibly by computing it,
2896 * otherwise read it if the backing disk is insync
2897 */
2898 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
2899 BUG_ON(test_bit(R5_Wantread, &dev->flags));
2900 if ((s->uptodate == disks - 1) &&
f2b3b44d
N		 2901 (s->failed && (disk_idx == s->failed_num[0] ||
2902 disk_idx == s->failed_num[1]))) {
5599becc
YT		 2903 /* have disk failed, and we're requested to fetch it;
2904 * do compute it
a4456856 2905 */
5599becc
YT		 2906 pr_debug("Computing stripe %llu block %d\n",
2907 (unsigned long long)sh->sector, disk_idx);
2908 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2909 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2910 set_bit(R5_Wantcompute, &dev->flags);
2911 sh->ops.target = disk_idx;
2912 sh->ops.target2 = -1; /* no 2nd target */
2913 s->req_compute = 1;
93b3dbce
N		 2914 /* Careful: from this point on 'uptodate' is in the eye
2915 * of raid_run_ops which services 'compute' operations
2916 * before writes. R5_Wantcompute flags a block that will
2917 * be R5_UPTODATE by the time it is needed for a
2918 * subsequent operation.
2919 */
5599becc
YT		 2920 s->uptodate++;
2921 return 1;
2922 } else if (s->uptodate == disks-2 && s->failed >= 2) {
2923 /* Computing 2-failure is *very* expensive; only
2924 * do it if failed >= 2
2925 */
2926 int other;
2927 for (other = disks; other--; ) {
2928 if (other == disk_idx)
2929 continue;
2930 if (!test_bit(R5_UPTODATE,
2931 &sh->dev[other].flags))
2932 break;
a4456856 2933 }
5599becc
YT		 2934 BUG_ON(other < 0);
2935 pr_debug("Computing stripe %llu blocks %d,%d\n",
2936 (unsigned long long)sh->sector,
2937 disk_idx, other);
2938 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2939 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2940 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
2941 set_bit(R5_Wantcompute, &sh->dev[other].flags);
2942 sh->ops.target = disk_idx;
2943 sh->ops.target2 = other;
2944 s->uptodate += 2;
2945 s->req_compute = 1;
2946 return 1;
2947 } else if (test_bit(R5_Insync, &dev->flags)) {
2948 set_bit(R5_LOCKED, &dev->flags);
2949 set_bit(R5_Wantread, &dev->flags);
2950 s->locked++;
2951 pr_debug("Reading block %d (sync=%d)\n",
2952 disk_idx, s->syncing);
a4456856
DW		 2953 }
2954 }
5599becc
YT		 2955
2956 return 0;
2957}
2958
2959/**
93b3dbce 2960 * handle_stripe_fill - read or compute data to satisfy pending requests.
5599becc 2961 */
93b3dbce
N		 2962static void handle_stripe_fill(struct stripe_head *sh,
2963 struct stripe_head_state *s,
2964 int disks)
5599becc
YT		 2965{
2966 int i;
2967
2968 /* look for blocks to read/compute, skip this if a compute
2969 * is already in flight, or if the stripe contents are in the
2970 * midst of changing due to a write
2971 */
2972 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
2973 !sh->reconstruct_state)
2974 for (i = disks; i--; )
93b3dbce 2975 if (fetch_block(sh, s, i, disks))
5599becc 2976 break;
a4456856
DW		 2977 set_bit(STRIPE_HANDLE, &sh->state);
2978}
2979
2980
1fe797e6 2981/* handle_stripe_clean_event
a4456856
DW		 2982 * any written block on an uptodate or failed drive can be returned.
2983 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
2984 * never LOCKED, so we don't need to test 'failed' directly.
2985 */
d1688a6d 2986static void handle_stripe_clean_event(struct r5conf *conf,
a4456856
DW		 2987 struct stripe_head *sh, int disks, struct bio **return_bi)
2988{
2989 int i;
2990 struct r5dev *dev;
f8dfcffd 2991 int discard_pending = 0;
a4456856
DW		 2992
2993 for (i = disks; i--; )
2994 if (sh->dev[i].written) {
2995 dev = &sh->dev[i];
2996 if (!test_bit(R5_LOCKED, &dev->flags) &&
9e444768 2997 (test_bit(R5_UPTODATE, &dev->flags) ||
ca64cae9 2998 test_bit(R5_Discard, &dev->flags))) {
a4456856
DW		 2999 /* We can return any write requests */
3000 struct bio *wbi, *wbi2;
45b4233c 3001 pr_debug("Return write for disc %d\n", i);
ca64cae9
N		 3002 if (test_and_clear_bit(R5_Discard, &dev->flags))
3003 clear_bit(R5_UPTODATE, &dev->flags);
a4456856
DW		 3004 wbi = dev->written;
3005 dev->written = NULL;
3006 while (wbi && wbi->bi_sector <
3007 dev->sector + STRIPE_SECTORS) {
3008 wbi2 = r5_next_bio(wbi, dev->sector);
e7836bd6 3009 if (!raid5_dec_bi_active_stripes(wbi)) {
a4456856
DW		 3010 md_write_end(conf->mddev);
3011 wbi->bi_next = *return_bi;
3012 *return_bi = wbi;
3013 }
3014 wbi = wbi2;
3015 }
7eaf7e8e
SL		 3016 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3017 STRIPE_SECTORS,
a4456856 3018 !test_bit(STRIPE_DEGRADED, &sh->state),
7eaf7e8e 3019 0);
f8dfcffd
N		 3020 } else if (test_bit(R5_Discard, &dev->flags))
3021 discard_pending = 1;
3022 }
3023 if (!discard_pending &&
3024 test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) {
3025 clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
3026 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
3027 if (sh->qd_idx >= 0) {
3028 clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
3029 clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags);
3030 }
3031 /* now that discard is done we can proceed with any sync */
3032 clear_bit(STRIPE_DISCARD, &sh->state);
d47648fc
SL		 3033 /*
3034 * SCSI discard will change some bio fields and the stripe has
3035 * no updated data, so remove it from hash list and the stripe
3036 * will be reinitialized
3037 */
3038 spin_lock_irq(&conf->device_lock);
3039 remove_hash(sh);
3040 spin_unlock_irq(&conf->device_lock);
f8dfcffd
N		 3041 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state))
3042 set_bit(STRIPE_HANDLE, &sh->state);
3043
3044 }
8b3e6cdc
DW		 3045
3046 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3047 if (atomic_dec_and_test(&conf->pending_full_writes))
3048 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 3049}
3050
d1688a6d 3051static void handle_stripe_dirtying(struct r5conf *conf,
c8ac1803
N		 3052 struct stripe_head *sh,
3053 struct stripe_head_state *s,
3054 int disks)
a4456856
DW		 3055{
3056 int rmw = 0, rcw = 0, i;
a7854487
AL		 3057 sector_t recovery_cp = conf->mddev->recovery_cp;
3058
3059 /* RAID6 requires 'rcw' in current implementation.
3060 * Otherwise, check whether resync is now happening or should start.
3061 * If yes, then the array is dirty (after unclean shutdown or
3062 * initial creation), so parity in some stripes might be inconsistent.
3063 * In this case, we need to always do reconstruct-write, to ensure
3064 * that in case of drive failure or read-error correction, we
3065 * generate correct data from the parity.
3066 */
3067 if (conf->max_degraded == 2 ||
3068 (recovery_cp < MaxSector && sh->sector >= recovery_cp)) {
3069 /* Calculate the real rcw later - for now make it
c8ac1803
N		 3070 * look like rcw is cheaper
3071 */
3072 rcw = 1; rmw = 2;
a7854487
AL		 3073 pr_debug("force RCW max_degraded=%u, recovery_cp=%llu sh->sector=%llu\n",
3074 conf->max_degraded, (unsigned long long)recovery_cp,
3075 (unsigned long long)sh->sector);
c8ac1803 3076 } else for (i = disks; i--; ) {
a4456856
DW		 3077 /* would I have to read this buffer for read_modify_write */
3078 struct r5dev *dev = &sh->dev[i];
3079 if ((dev->towrite || i == sh->pd_idx) &&
3080 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 3081 !(test_bit(R5_UPTODATE, &dev->flags) ||
3082 test_bit(R5_Wantcompute, &dev->flags))) {
a4456856
DW		 3083 if (test_bit(R5_Insync, &dev->flags))
3084 rmw++;
3085 else
3086 rmw += 2*disks; /* cannot read it */
3087 }
3088 /* Would I have to read this buffer for reconstruct_write */
3089 if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx &&
3090 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 3091 !(test_bit(R5_UPTODATE, &dev->flags) ||
3092 test_bit(R5_Wantcompute, &dev->flags))) {
3093 if (test_bit(R5_Insync, &dev->flags)) rcw++;
a4456856
DW		 3094 else
3095 rcw += 2*disks;
3096 }
3097 }
45b4233c 3098 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
a4456856
DW		 3099 (unsigned long long)sh->sector, rmw, rcw);
3100 set_bit(STRIPE_HANDLE, &sh->state);
a9add5d9 3101 if (rmw < rcw && rmw > 0) {
a4456856 3102 /* prefer read-modify-write, but need to get some data */
e3620a3a
JB		 3103 if (conf->mddev->queue)
3104 blk_add_trace_msg(conf->mddev->queue,
3105 "raid5 rmw %llu %d",
3106 (unsigned long long)sh->sector, rmw);
a4456856
DW		 3107 for (i = disks; i--; ) {
3108 struct r5dev *dev = &sh->dev[i];
3109 if ((dev->towrite || i == sh->pd_idx) &&
3110 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 3111 !(test_bit(R5_UPTODATE, &dev->flags) ||
3112 test_bit(R5_Wantcompute, &dev->flags)) &&
a4456856
DW		 3113 test_bit(R5_Insync, &dev->flags)) {
3114 if (
3115 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
45b4233c 3116 pr_debug("Read_old block "
a9add5d9 3117 "%d for r-m-w\n", i);
a4456856
DW		 3118 set_bit(R5_LOCKED, &dev->flags);
3119 set_bit(R5_Wantread, &dev->flags);
3120 s->locked++;
3121 } else {
3122 set_bit(STRIPE_DELAYED, &sh->state);
3123 set_bit(STRIPE_HANDLE, &sh->state);
3124 }
3125 }
3126 }
a9add5d9 3127 }
c8ac1803 3128 if (rcw <= rmw && rcw > 0) {
a4456856 3129 /* want reconstruct write, but need to get some data */
a9add5d9 3130 int qread =0;
c8ac1803 3131 rcw = 0;
a4456856
DW		 3132 for (i = disks; i--; ) {
3133 struct r5dev *dev = &sh->dev[i];
3134 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
c8ac1803 3135 i != sh->pd_idx && i != sh->qd_idx &&
a4456856 3136 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219 3137 !(test_bit(R5_UPTODATE, &dev->flags) ||
c8ac1803
N		 3138 test_bit(R5_Wantcompute, &dev->flags))) {
3139 rcw++;
3140 if (!test_bit(R5_Insync, &dev->flags))
3141 continue; /* it's a failed drive */
a4456856
DW		 3142 if (
3143 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
45b4233c 3144 pr_debug("Read_old block "
a4456856
DW		 3145 "%d for Reconstruct\n", i);
3146 set_bit(R5_LOCKED, &dev->flags);
3147 set_bit(R5_Wantread, &dev->flags);
3148 s->locked++;
a9add5d9 3149 qread++;
a4456856
DW		 3150 } else {
3151 set_bit(STRIPE_DELAYED, &sh->state);
3152 set_bit(STRIPE_HANDLE, &sh->state);
3153 }
3154 }
3155 }
e3620a3a 3156 if (rcw && conf->mddev->queue)
a9add5d9
N		 3157 blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d",
3158 (unsigned long long)sh->sector,
3159 rcw, qread, test_bit(STRIPE_DELAYED, &sh->state));
c8ac1803 3160 }
a4456856
DW		 3161 /* now if nothing is locked, and if we have enough data,
3162 * we can start a write request
3163 */
f38e1219
DW		 3164 /* since handle_stripe can be called at any time we need to handle the
3165 * case where a compute block operation has been submitted and then a
ac6b53b6
DW		 3166 * subsequent call wants to start a write request. raid_run_ops only
3167 * handles the case where compute block and reconstruct are requested
f38e1219
DW		 3168 * simultaneously. If this is not the case then new writes need to be
3169 * held off until the compute completes.
3170 */
976ea8d4
DW		 3171 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
3172 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
3173 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
c0f7bddb 3174 schedule_reconstruction(sh, s, rcw == 0, 0);
a4456856
DW		 3175}
3176
d1688a6d 3177static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh,
a4456856
DW		 3178 struct stripe_head_state *s, int disks)
3179{
ecc65c9b 3180 struct r5dev *dev = NULL;
bd2ab670 3181
a4456856 3182 set_bit(STRIPE_HANDLE, &sh->state);
e89f8962 3183
ecc65c9b
DW		 3184 switch (sh->check_state) {
3185 case check_state_idle:
3186 /* start a new check operation if there are no failures */
bd2ab670 3187 if (s->failed == 0) {
bd2ab670 3188 BUG_ON(s->uptodate != disks);
ecc65c9b
DW		 3189 sh->check_state = check_state_run;
3190 set_bit(STRIPE_OP_CHECK, &s->ops_request);
bd2ab670 3191 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
bd2ab670 3192 s->uptodate--;
ecc65c9b 3193 break;
bd2ab670 3194 }
f2b3b44d 3195 dev = &sh->dev[s->failed_num[0]];
ecc65c9b
DW		 3196 /* fall through */
3197 case check_state_compute_result:
3198 sh->check_state = check_state_idle;
3199 if (!dev)
3200 dev = &sh->dev[sh->pd_idx];
3201
3202 /* check that a write has not made the stripe insync */
3203 if (test_bit(STRIPE_INSYNC, &sh->state))
3204 break;
c8894419 3205
a4456856 3206 /* either failed parity check, or recovery is happening */
a4456856
DW		 3207 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
3208 BUG_ON(s->uptodate != disks);
3209
3210 set_bit(R5_LOCKED, &dev->flags);
ecc65c9b 3211 s->locked++;
a4456856 3212 set_bit(R5_Wantwrite, &dev->flags);
830ea016 3213
a4456856 3214 clear_bit(STRIPE_DEGRADED, &sh->state);
a4456856 3215 set_bit(STRIPE_INSYNC, &sh->state);
ecc65c9b
DW		 3216 break;
3217 case check_state_run:
3218 break; /* we will be called again upon completion */
3219 case check_state_check_result:
3220 sh->check_state = check_state_idle;
3221
3222 /* if a failure occurred during the check operation, leave
3223 * STRIPE_INSYNC not set and let the stripe be handled again
3224 */
3225 if (s->failed)
3226 break;
3227
3228 /* handle a successful check operation, if parity is correct
3229 * we are done. Otherwise update the mismatch count and repair
3230 * parity if !MD_RECOVERY_CHECK
3231 */
ad283ea4 3232 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
ecc65c9b
DW		 3233 /* parity is correct (on disc,
3234 * not in buffer any more)
3235 */
3236 set_bit(STRIPE_INSYNC, &sh->state);
3237 else {
7f7583d4 3238 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
ecc65c9b
DW		 3239 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
3240 /* don't try to repair!! */
3241 set_bit(STRIPE_INSYNC, &sh->state);
3242 else {
3243 sh->check_state = check_state_compute_run;
976ea8d4 3244 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
ecc65c9b
DW		 3245 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3246 set_bit(R5_Wantcompute,
3247 &sh->dev[sh->pd_idx].flags);
3248 sh->ops.target = sh->pd_idx;
ac6b53b6 3249 sh->ops.target2 = -1;
ecc65c9b
DW		 3250 s->uptodate++;
3251 }
3252 }
3253 break;
3254 case check_state_compute_run:
3255 break;
3256 default:
3257 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
3258 __func__, sh->check_state,
3259 (unsigned long long) sh->sector);
3260 BUG();
a4456856
DW		 3261 }
3262}
3263
3264
d1688a6d 3265static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh,
36d1c647 3266 struct stripe_head_state *s,
f2b3b44d 3267 int disks)
a4456856 3268{
a4456856 3269 int pd_idx = sh->pd_idx;
34e04e87 3270 int qd_idx = sh->qd_idx;
d82dfee0 3271 struct r5dev *dev;
a4456856
DW		 3272
3273 set_bit(STRIPE_HANDLE, &sh->state);
3274
3275 BUG_ON(s->failed > 2);
d82dfee0 3276
a4456856
DW		 3277 /* Want to check and possibly repair P and Q.
3278 * However there could be one 'failed' device, in which
3279 * case we can only check one of them, possibly using the
3280 * other to generate missing data
3281 */
3282
d82dfee0
DW		 3283 switch (sh->check_state) {
3284 case check_state_idle:
3285 /* start a new check operation if there are < 2 failures */
f2b3b44d 3286 if (s->failed == s->q_failed) {
d82dfee0 3287 /* The only possible failed device holds Q, so it
a4456856
DW		 3288 * makes sense to check P (If anything else were failed,
3289 * we would have used P to recreate it).
3290 */
d82dfee0 3291 sh->check_state = check_state_run;
a4456856 3292 }
f2b3b44d 3293 if (!s->q_failed && s->failed < 2) {
d82dfee0 3294 /* Q is not failed, and we didn't use it to generate
a4456856
DW		 3295 * anything, so it makes sense to check it
3296 */
d82dfee0
DW		 3297 if (sh->check_state == check_state_run)
3298 sh->check_state = check_state_run_pq;
3299 else
3300 sh->check_state = check_state_run_q;
a4456856 3301 }
a4456856 3302
d82dfee0
DW		 3303 /* discard potentially stale zero_sum_result */
3304 sh->ops.zero_sum_result = 0;
a4456856 3305
d82dfee0
DW		 3306 if (sh->check_state == check_state_run) {
3307 /* async_xor_zero_sum destroys the contents of P */
3308 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
3309 s->uptodate--;
a4456856 3310 }
d82dfee0
DW		 3311 if (sh->check_state >= check_state_run &&
3312 sh->check_state <= check_state_run_pq) {
3313 /* async_syndrome_zero_sum preserves P and Q, so
3314 * no need to mark them !uptodate here
3315 */
3316 set_bit(STRIPE_OP_CHECK, &s->ops_request);
3317 break;
a4456856
DW		 3318 }
3319
d82dfee0
DW		 3320 /* we have 2-disk failure */
3321 BUG_ON(s->failed != 2);
3322 /* fall through */
3323 case check_state_compute_result:
3324 sh->check_state = check_state_idle;
a4456856 3325
d82dfee0
DW		 3326 /* check that a write has not made the stripe insync */
3327 if (test_bit(STRIPE_INSYNC, &sh->state))
3328 break;
a4456856
DW		 3329
3330 /* now write out any block on a failed drive,
d82dfee0 3331 * or P or Q if they were recomputed
a4456856 3332 */
d82dfee0 3333 BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */
a4456856 3334 if (s->failed == 2) {
f2b3b44d 3335 dev = &sh->dev[s->failed_num[1]];
a4456856
DW		 3336 s->locked++;
3337 set_bit(R5_LOCKED, &dev->flags);
3338 set_bit(R5_Wantwrite, &dev->flags);
3339 }
3340 if (s->failed >= 1) {
f2b3b44d 3341 dev = &sh->dev[s->failed_num[0]];
a4456856
DW		 3342 s->locked++;
3343 set_bit(R5_LOCKED, &dev->flags);
3344 set_bit(R5_Wantwrite, &dev->flags);
3345 }
d82dfee0 3346 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
a4456856
DW		 3347 dev = &sh->dev[pd_idx];
3348 s->locked++;
3349 set_bit(R5_LOCKED, &dev->flags);
3350 set_bit(R5_Wantwrite, &dev->flags);
3351 }
d82dfee0 3352 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
a4456856
DW		 3353 dev = &sh->dev[qd_idx];
3354 s->locked++;
3355 set_bit(R5_LOCKED, &dev->flags);
3356 set_bit(R5_Wantwrite, &dev->flags);
3357 }
3358 clear_bit(STRIPE_DEGRADED, &sh->state);
3359
3360 set_bit(STRIPE_INSYNC, &sh->state);
d82dfee0
DW		 3361 break;
3362 case check_state_run:
3363 case check_state_run_q:
3364 case check_state_run_pq:
3365 break; /* we will be called again upon completion */
3366 case check_state_check_result:
3367 sh->check_state = check_state_idle;
3368
3369 /* handle a successful check operation, if parity is correct
3370 * we are done. Otherwise update the mismatch count and repair
3371 * parity if !MD_RECOVERY_CHECK
3372 */
3373 if (sh->ops.zero_sum_result == 0) {
3374 /* both parities are correct */
3375 if (!s->failed)
3376 set_bit(STRIPE_INSYNC, &sh->state);
3377 else {
3378 /* in contrast to the raid5 case we can validate
3379 * parity, but still have a failure to write
3380 * back
3381 */
3382 sh->check_state = check_state_compute_result;
3383 /* Returning at this point means that we may go
3384 * off and bring p and/or q uptodate again so
3385 * we make sure to check zero_sum_result again
3386 * to verify if p or q need writeback
3387 */
3388 }
3389 } else {
7f7583d4 3390 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
d82dfee0
DW		 3391 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
3392 /* don't try to repair!! */
3393 set_bit(STRIPE_INSYNC, &sh->state);
3394 else {
3395 int *target = &sh->ops.target;
3396
3397 sh->ops.target = -1;
3398 sh->ops.target2 = -1;
3399 sh->check_state = check_state_compute_run;
3400 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3401 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3402 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
3403 set_bit(R5_Wantcompute,
3404 &sh->dev[pd_idx].flags);
3405 *target = pd_idx;
3406 target = &sh->ops.target2;
3407 s->uptodate++;
3408 }
3409 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
3410 set_bit(R5_Wantcompute,
3411 &sh->dev[qd_idx].flags);
3412 *target = qd_idx;
3413 s->uptodate++;
3414 }
3415 }
3416 }
3417 break;
3418 case check_state_compute_run:
3419 break;
3420 default:
3421 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
3422 __func__, sh->check_state,
3423 (unsigned long long) sh->sector);
3424 BUG();
a4456856
DW		 3425 }
3426}
3427
d1688a6d 3428static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh)
a4456856
DW		 3429{
3430 int i;
3431
3432 /* We have read all the blocks in this stripe and now we need to
3433 * copy some of them into a target stripe for expand.
3434 */
f0a50d37 3435 struct dma_async_tx_descriptor *tx = NULL;
a4456856
DW		 3436 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3437 for (i = 0; i < sh->disks; i++)
34e04e87 3438 if (i != sh->pd_idx && i != sh->qd_idx) {
911d4ee8 3439 int dd_idx, j;
a4456856 3440 struct stripe_head *sh2;
a08abd8c 3441 struct async_submit_ctl submit;
a4456856 3442
784052ec 3443 sector_t bn = compute_blocknr(sh, i, 1);
911d4ee8
N		 3444 sector_t s = raid5_compute_sector(conf, bn, 0,
3445 &dd_idx, NULL);
a8c906ca 3446 sh2 = get_active_stripe(conf, s, 0, 1, 1);
a4456856
DW		 3447 if (sh2 == NULL)
3448 /* so far only the early blocks of this stripe
3449 * have been requested. When later blocks
3450 * get requested, we will try again
3451 */
3452 continue;
3453 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
3454 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
3455 /* must have already done this block */
3456 release_stripe(sh2);
3457 continue;
3458 }
f0a50d37
DW		 3459
3460 /* place all the copies on one channel */
a08abd8c 3461 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
f0a50d37 3462 tx = async_memcpy(sh2->dev[dd_idx].page,
88ba2aa5 3463 sh->dev[i].page, 0, 0, STRIPE_SIZE,
a08abd8c 3464 &submit);
f0a50d37 3465
a4456856
DW		 3466 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
3467 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
3468 for (j = 0; j < conf->raid_disks; j++)
3469 if (j != sh2->pd_idx &&
86c374ba 3470 j != sh2->qd_idx &&
a4456856
DW		 3471 !test_bit(R5_Expanded, &sh2->dev[j].flags))
3472 break;
3473 if (j == conf->raid_disks) {
3474 set_bit(STRIPE_EXPAND_READY, &sh2->state);
3475 set_bit(STRIPE_HANDLE, &sh2->state);
3476 }
3477 release_stripe(sh2);
f0a50d37 3478
a4456856 3479 }
a2e08551 3480 /* done submitting copies, wait for them to complete */
749586b7 3481 async_tx_quiesce(&tx);
a4456856 3482}
1da177e4
LT		 3483
3484/*
3485 * handle_stripe - do things to a stripe.
3486 *
9a3e1101
N		 3487 * We lock the stripe by setting STRIPE_ACTIVE and then examine the
3488 * state of various bits to see what needs to be done.
1da177e4 3489 * Possible results:
9a3e1101
N		 3490 * return some read requests which now have data
3491 * return some write requests which are safely on storage
1da177e4
LT		 3492 * schedule a read on some buffers
3493 * schedule a write of some buffers
3494 * return confirmation of parity correctness
3495 *
1da177e4 3496 */
a4456856 3497
acfe726b 3498static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s)
1da177e4 3499{
d1688a6d 3500 struct r5conf *conf = sh->raid_conf;
f416885e 3501 int disks = sh->disks;
474af965
N		 3502 struct r5dev *dev;
3503 int i;
9a3e1101 3504 int do_recovery = 0;
1da177e4 3505
acfe726b
N		 3506 memset(s, 0, sizeof(*s));
3507
acfe726b
N		 3508 s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3509 s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
3510 s->failed_num[0] = -1;
3511 s->failed_num[1] = -1;
1da177e4 3512
acfe726b 3513 /* Now to look around and see what can be done */
1da177e4 3514 rcu_read_lock();
16a53ecc 3515 for (i=disks; i--; ) {
3cb03002 3516 struct md_rdev *rdev;
31c176ec
N		 3517 sector_t first_bad;
3518 int bad_sectors;
3519 int is_bad = 0;
acfe726b 3520
16a53ecc 3521 dev = &sh->dev[i];
1da177e4 3522
45b4233c 3523 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
9a3e1101
N		 3524 i, dev->flags,
3525 dev->toread, dev->towrite, dev->written);
6c0069c0
YT		 3526 /* maybe we can reply to a read
3527 *
3528 * new wantfill requests are only permitted while
3529 * ops_complete_biofill is guaranteed to be inactive
3530 */
3531 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
3532 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
3533 set_bit(R5_Wantfill, &dev->flags);
1da177e4 3534
16a53ecc 3535 /* now count some things */
cc94015a
N		 3536 if (test_bit(R5_LOCKED, &dev->flags))
3537 s->locked++;
3538 if (test_bit(R5_UPTODATE, &dev->flags))
3539 s->uptodate++;
2d6e4ecc 3540 if (test_bit(R5_Wantcompute, &dev->flags)) {
cc94015a
N		 3541 s->compute++;
3542 BUG_ON(s->compute > 2);
2d6e4ecc 3543 }
1da177e4 3544
acfe726b 3545 if (test_bit(R5_Wantfill, &dev->flags))
cc94015a 3546 s->to_fill++;
acfe726b 3547 else if (dev->toread)
cc94015a 3548 s->to_read++;
16a53ecc 3549 if (dev->towrite) {
cc94015a 3550 s->to_write++;
16a53ecc 3551 if (!test_bit(R5_OVERWRITE, &dev->flags))
cc94015a 3552 s->non_overwrite++;
16a53ecc 3553 }
a4456856 3554 if (dev->written)
cc94015a 3555 s->written++;
14a75d3e
N		 3556 /* Prefer to use the replacement for reads, but only
3557 * if it is recovered enough and has no bad blocks.
3558 */
3559 rdev = rcu_dereference(conf->disks[i].replacement);
3560 if (rdev && !test_bit(Faulty, &rdev->flags) &&
3561 rdev->recovery_offset >= sh->sector + STRIPE_SECTORS &&
3562 !is_badblock(rdev, sh->sector, STRIPE_SECTORS,
3563 &first_bad, &bad_sectors))
3564 set_bit(R5_ReadRepl, &dev->flags);
3565 else {
9a3e1101
N		 3566 if (rdev)
3567 set_bit(R5_NeedReplace, &dev->flags);
14a75d3e
N		 3568 rdev = rcu_dereference(conf->disks[i].rdev);
3569 clear_bit(R5_ReadRepl, &dev->flags);
3570 }
9283d8c5
N		 3571 if (rdev && test_bit(Faulty, &rdev->flags))
3572 rdev = NULL;
31c176ec
N		 3573 if (rdev) {
3574 is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
3575 &first_bad, &bad_sectors);
3576 if (s->blocked_rdev == NULL
3577 && (test_bit(Blocked, &rdev->flags)
3578 || is_bad < 0)) {
3579 if (is_bad < 0)
3580 set_bit(BlockedBadBlocks,
3581 &rdev->flags);
3582 s->blocked_rdev = rdev;
3583 atomic_inc(&rdev->nr_pending);
3584 }
6bfe0b49 3585 }
415e72d0
N		 3586 clear_bit(R5_Insync, &dev->flags);
3587 if (!rdev)
3588 /* Not in-sync */;
31c176ec
N		 3589 else if (is_bad) {
3590 /* also not in-sync */
18b9837e
N		 3591 if (!test_bit(WriteErrorSeen, &rdev->flags) &&
3592 test_bit(R5_UPTODATE, &dev->flags)) {
31c176ec
N		 3593 /* treat as in-sync, but with a read error
3594 * which we can now try to correct
3595 */
3596 set_bit(R5_Insync, &dev->flags);
3597 set_bit(R5_ReadError, &dev->flags);
3598 }
3599 } else if (test_bit(In_sync, &rdev->flags))
415e72d0 3600 set_bit(R5_Insync, &dev->flags);
30d7a483 3601 else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
415e72d0 3602 /* in sync if before recovery_offset */
30d7a483
N		 3603 set_bit(R5_Insync, &dev->flags);
3604 else if (test_bit(R5_UPTODATE, &dev->flags) &&
3605 test_bit(R5_Expanded, &dev->flags))
3606 /* If we've reshaped into here, we assume it is Insync.
3607 * We will shortly update recovery_offset to make
3608 * it official.
3609 */
3610 set_bit(R5_Insync, &dev->flags);
3611
1cc03eb9 3612 if (test_bit(R5_WriteError, &dev->flags)) {
14a75d3e
N		 3613 /* This flag does not apply to '.replacement'
3614 * only to .rdev, so make sure to check that*/
3615 struct md_rdev *rdev2 = rcu_dereference(
3616 conf->disks[i].rdev);
3617 if (rdev2 == rdev)
3618 clear_bit(R5_Insync, &dev->flags);
3619 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
bc2607f3 3620 s->handle_bad_blocks = 1;
14a75d3e 3621 atomic_inc(&rdev2->nr_pending);
bc2607f3
N		 3622 } else
3623 clear_bit(R5_WriteError, &dev->flags);
3624 }
1cc03eb9 3625 if (test_bit(R5_MadeGood, &dev->flags)) {
14a75d3e
N		 3626 /* This flag does not apply to '.replacement'
3627 * only to .rdev, so make sure to check that*/
3628 struct md_rdev *rdev2 = rcu_dereference(
3629 conf->disks[i].rdev);
3630 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
b84db560 3631 s->handle_bad_blocks = 1;
14a75d3e 3632 atomic_inc(&rdev2->nr_pending);
b84db560
N		 3633 } else
3634 clear_bit(R5_MadeGood, &dev->flags);
3635 }
977df362
N		 3636 if (test_bit(R5_MadeGoodRepl, &dev->flags)) {
3637 struct md_rdev *rdev2 = rcu_dereference(
3638 conf->disks[i].replacement);
3639 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
3640 s->handle_bad_blocks = 1;
3641 atomic_inc(&rdev2->nr_pending);
3642 } else
3643 clear_bit(R5_MadeGoodRepl, &dev->flags);
3644 }
415e72d0 3645 if (!test_bit(R5_Insync, &dev->flags)) {
16a53ecc
N		 3646 /* The ReadError flag will just be confusing now */
3647 clear_bit(R5_ReadError, &dev->flags);
3648 clear_bit(R5_ReWrite, &dev->flags);
1da177e4 3649 }
415e72d0
N		 3650 if (test_bit(R5_ReadError, &dev->flags))
3651 clear_bit(R5_Insync, &dev->flags);
3652 if (!test_bit(R5_Insync, &dev->flags)) {
cc94015a
N		 3653 if (s->failed < 2)
3654 s->failed_num[s->failed] = i;
3655 s->failed++;
9a3e1101
N		 3656 if (rdev && !test_bit(Faulty, &rdev->flags))
3657 do_recovery = 1;
415e72d0 3658 }
1da177e4 3659 }
9a3e1101
N		 3660 if (test_bit(STRIPE_SYNCING, &sh->state)) {
3661 /* If there is a failed device being replaced,
3662 * we must be recovering.
3663 * else if we are after recovery_cp, we must be syncing
c6d2e084 3664 * else if MD_RECOVERY_REQUESTED is set, we also are syncing.
9a3e1101
N		 3665 * else we can only be replacing
3666 * sync and recovery both need to read all devices, and so
3667 * use the same flag.
3668 */
3669 if (do_recovery ||
c6d2e084 3670 sh->sector >= conf->mddev->recovery_cp ||
3671 test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery)))
9a3e1101
N		 3672 s->syncing = 1;
3673 else
3674 s->replacing = 1;
3675 }
1da177e4 3676 rcu_read_unlock();
cc94015a
N		 3677}
3678
3679static void handle_stripe(struct stripe_head *sh)
3680{
3681 struct stripe_head_state s;
d1688a6d 3682 struct r5conf *conf = sh->raid_conf;
3687c061 3683 int i;
84789554
N		 3684 int prexor;
3685 int disks = sh->disks;
474af965 3686 struct r5dev *pdev, *qdev;
cc94015a
N		 3687
3688 clear_bit(STRIPE_HANDLE, &sh->state);
257a4b42 3689 if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) {
cc94015a
N		 3690 /* already being handled, ensure it gets handled
3691 * again when current action finishes */
3692 set_bit(STRIPE_HANDLE, &sh->state);
3693 return;
3694 }
3695
f8dfcffd
N		 3696 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
3697 spin_lock(&sh->stripe_lock);
3698 /* Cannot process 'sync' concurrently with 'discard' */
3699 if (!test_bit(STRIPE_DISCARD, &sh->state) &&
3700 test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
3701 set_bit(STRIPE_SYNCING, &sh->state);
3702 clear_bit(STRIPE_INSYNC, &sh->state);
f94c0b66 3703 clear_bit(STRIPE_REPLACED, &sh->state);
f8dfcffd
N		 3704 }
3705 spin_unlock(&sh->stripe_lock);
cc94015a
N		 3706 }
3707 clear_bit(STRIPE_DELAYED, &sh->state);
3708
3709 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
3710 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
3711 (unsigned long long)sh->sector, sh->state,
3712 atomic_read(&sh->count), sh->pd_idx, sh->qd_idx,
3713 sh->check_state, sh->reconstruct_state);
3687c061 3714
acfe726b 3715 analyse_stripe(sh, &s);
c5a31000 3716
bc2607f3
N		 3717 if (s.handle_bad_blocks) {
3718 set_bit(STRIPE_HANDLE, &sh->state);
3719 goto finish;
3720 }
3721
474af965
N		 3722 if (unlikely(s.blocked_rdev)) {
3723 if (s.syncing || s.expanding || s.expanded ||
9a3e1101 3724 s.replacing || s.to_write || s.written) {
474af965
N		 3725 set_bit(STRIPE_HANDLE, &sh->state);
3726 goto finish;
3727 }
3728 /* There is nothing for the blocked_rdev to block */
3729 rdev_dec_pending(s.blocked_rdev, conf->mddev);
3730 s.blocked_rdev = NULL;
3731 }
3732
3733 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
3734 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
3735 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
3736 }
3737
3738 pr_debug("locked=%d uptodate=%d to_read=%d"
3739 " to_write=%d failed=%d failed_num=%d,%d\n",
3740 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
3741 s.failed_num[0], s.failed_num[1]);
3742 /* check if the array has lost more than max_degraded devices and,
3743 * if so, some requests might need to be failed.
3744 */
9a3f530f
N		 3745 if (s.failed > conf->max_degraded) {
3746 sh->check_state = 0;
3747 sh->reconstruct_state = 0;
3748 if (s.to_read+s.to_write+s.written)
3749 handle_failed_stripe(conf, sh, &s, disks, &s.return_bi);
9a3e1101 3750 if (s.syncing + s.replacing)
9a3f530f
N		 3751 handle_failed_sync(conf, sh, &s);
3752 }
474af965 3753
84789554
N		 3754 /* Now we check to see if any write operations have recently
3755 * completed
3756 */
3757 prexor = 0;
3758 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
3759 prexor = 1;
3760 if (sh->reconstruct_state == reconstruct_state_drain_result ||
3761 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
3762 sh->reconstruct_state = reconstruct_state_idle;
3763
3764 /* All the 'written' buffers and the parity block are ready to
3765 * be written back to disk
3766 */
9e444768
SL		 3767 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) &&
3768 !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags));
84789554 3769 BUG_ON(sh->qd_idx >= 0 &&
9e444768
SL		 3770 !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) &&
3771 !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags));
84789554
N		 3772 for (i = disks; i--; ) {
3773 struct r5dev *dev = &sh->dev[i];
3774 if (test_bit(R5_LOCKED, &dev->flags) &&
3775 (i == sh->pd_idx || i == sh->qd_idx ||
3776 dev->written)) {
3777 pr_debug("Writing block %d\n", i);
3778 set_bit(R5_Wantwrite, &dev->flags);
3779 if (prexor)
3780 continue;
3781 if (!test_bit(R5_Insync, &dev->flags) ||
3782 ((i == sh->pd_idx || i == sh->qd_idx) &&
3783 s.failed == 0))
3784 set_bit(STRIPE_INSYNC, &sh->state);
3785 }
3786 }
3787 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3788 s.dec_preread_active = 1;
3789 }
3790
ef5b7c69
N		 3791 /*
3792 * might be able to return some write requests if the parity blocks
3793 * are safe, or on a failed drive
3794 */
3795 pdev = &sh->dev[sh->pd_idx];
3796 s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx)
3797 || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx);
3798 qdev = &sh->dev[sh->qd_idx];
3799 s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx)
3800 || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx)
3801 || conf->level < 6;
3802
3803 if (s.written &&
3804 (s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
3805 && !test_bit(R5_LOCKED, &pdev->flags)
3806 && (test_bit(R5_UPTODATE, &pdev->flags) ||
3807 test_bit(R5_Discard, &pdev->flags))))) &&
3808 (s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
3809 && !test_bit(R5_LOCKED, &qdev->flags)
3810 && (test_bit(R5_UPTODATE, &qdev->flags) ||
3811 test_bit(R5_Discard, &qdev->flags))))))
3812 handle_stripe_clean_event(conf, sh, disks, &s.return_bi);
3813
3814 /* Now we might consider reading some blocks, either to check/generate
3815 * parity, or to satisfy requests
3816 * or to load a block that is being partially written.
3817 */
3818 if (s.to_read || s.non_overwrite
3819 || (conf->level == 6 && s.to_write && s.failed)
3820 || (s.syncing && (s.uptodate + s.compute < disks))
3821 || s.replacing
3822 || s.expanding)
3823 handle_stripe_fill(sh, &s, disks);
3824
84789554
N		 3825 /* Now to consider new write requests and what else, if anything
3826 * should be read. We do not handle new writes when:
3827 * 1/ A 'write' operation (copy+xor) is already in flight.
3828 * 2/ A 'check' operation is in flight, as it may clobber the parity
3829 * block.
3830 */
3831 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
3832 handle_stripe_dirtying(conf, sh, &s, disks);
3833
3834 /* maybe we need to check and possibly fix the parity for this stripe
3835 * Any reads will already have been scheduled, so we just see if enough
3836 * data is available. The parity check is held off while parity
3837 * dependent operations are in flight.
3838 */
3839 if (sh->check_state ||
3840 (s.syncing && s.locked == 0 &&
3841 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
3842 !test_bit(STRIPE_INSYNC, &sh->state))) {
3843 if (conf->level == 6)
3844 handle_parity_checks6(conf, sh, &s, disks);
3845 else
3846 handle_parity_checks5(conf, sh, &s, disks);
3847 }
c5a31000 3848
f94c0b66
N		 3849 if ((s.replacing || s.syncing) && s.locked == 0
3850 && !test_bit(STRIPE_COMPUTE_RUN, &sh->state)
3851 && !test_bit(STRIPE_REPLACED, &sh->state)) {
9a3e1101
N		 3852 /* Write out to replacement devices where possible */
3853 for (i = 0; i < conf->raid_disks; i++)
f94c0b66
N		 3854 if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) {
3855 WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags));
9a3e1101
N		 3856 set_bit(R5_WantReplace, &sh->dev[i].flags);
3857 set_bit(R5_LOCKED, &sh->dev[i].flags);
3858 s.locked++;
3859 }
f94c0b66
N		 3860 if (s.replacing)
3861 set_bit(STRIPE_INSYNC, &sh->state);
3862 set_bit(STRIPE_REPLACED, &sh->state);
9a3e1101
N		 3863 }
3864 if ((s.syncing || s.replacing) && s.locked == 0 &&
f94c0b66 3865 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
9a3e1101 3866 test_bit(STRIPE_INSYNC, &sh->state)) {
c5a31000
N		 3867 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3868 clear_bit(STRIPE_SYNCING, &sh->state);
f8dfcffd
N		 3869 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
3870 wake_up(&conf->wait_for_overlap);
c5a31000
N		 3871 }
3872
3873 /* If the failed drives are just a ReadError, then we might need
3874 * to progress the repair/check process
3875 */
3876 if (s.failed <= conf->max_degraded && !conf->mddev->ro)
3877 for (i = 0; i < s.failed; i++) {
3878 struct r5dev *dev = &sh->dev[s.failed_num[i]];
3879 if (test_bit(R5_ReadError, &dev->flags)
3880 && !test_bit(R5_LOCKED, &dev->flags)
3881 && test_bit(R5_UPTODATE, &dev->flags)
3882 ) {
3883 if (!test_bit(R5_ReWrite, &dev->flags)) {
3884 set_bit(R5_Wantwrite, &dev->flags);
3885 set_bit(R5_ReWrite, &dev->flags);
3886 set_bit(R5_LOCKED, &dev->flags);
3887 s.locked++;
3888 } else {
3889 /* let's read it back */
3890 set_bit(R5_Wantread, &dev->flags);
3891 set_bit(R5_LOCKED, &dev->flags);
3892 s.locked++;
3893 }
3894 }
3895 }
3896
3897
3687c061
N		 3898 /* Finish reconstruct operations initiated by the expansion process */
3899 if (sh->reconstruct_state == reconstruct_state_result) {
3900 struct stripe_head *sh_src
3901 = get_active_stripe(conf, sh->sector, 1, 1, 1);
3902 if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) {
3903 /* sh cannot be written until sh_src has been read.
3904 * so arrange for sh to be delayed a little
3905 */
3906 set_bit(STRIPE_DELAYED, &sh->state);
3907 set_bit(STRIPE_HANDLE, &sh->state);
3908 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
3909 &sh_src->state))
3910 atomic_inc(&conf->preread_active_stripes);
3911 release_stripe(sh_src);
3912 goto finish;
3913 }
3914 if (sh_src)
3915 release_stripe(sh_src);
3916
3917 sh->reconstruct_state = reconstruct_state_idle;
3918 clear_bit(STRIPE_EXPANDING, &sh->state);
3919 for (i = conf->raid_disks; i--; ) {
3920 set_bit(R5_Wantwrite, &sh->dev[i].flags);
3921 set_bit(R5_LOCKED, &sh->dev[i].flags);
3922 s.locked++;
3923 }
3924 }
f416885e 3925
3687c061
N		 3926 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
3927 !sh->reconstruct_state) {
3928 /* Need to write out all blocks after computing parity */
3929 sh->disks = conf->raid_disks;
3930 stripe_set_idx(sh->sector, conf, 0, sh);
3931 schedule_reconstruction(sh, &s, 1, 1);
3932 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
3933 clear_bit(STRIPE_EXPAND_READY, &sh->state);
3934 atomic_dec(&conf->reshape_stripes);
3935 wake_up(&conf->wait_for_overlap);
3936 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3937 }
3938
3939 if (s.expanding && s.locked == 0 &&
3940 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
3941 handle_stripe_expansion(conf, sh);
16a53ecc 3942
3687c061 3943finish:
6bfe0b49 3944 /* wait for this device to become unblocked */
5f066c63
N		 3945 if (unlikely(s.blocked_rdev)) {
3946 if (conf->mddev->external)
3947 md_wait_for_blocked_rdev(s.blocked_rdev,
3948 conf->mddev);
3949 else
3950 /* Internal metadata will immediately
3951 * be written by raid5d, so we don't
3952 * need to wait here.
3953 */
3954 rdev_dec_pending(s.blocked_rdev,
3955 conf->mddev);
3956 }
6bfe0b49 3957
bc2607f3
N		 3958 if (s.handle_bad_blocks)
3959 for (i = disks; i--; ) {
3cb03002 3960 struct md_rdev *rdev;
bc2607f3
N		 3961 struct r5dev *dev = &sh->dev[i];
3962 if (test_and_clear_bit(R5_WriteError, &dev->flags)) {
3963 /* We own a safe reference to the rdev */
3964 rdev = conf->disks[i].rdev;
3965 if (!rdev_set_badblocks(rdev, sh->sector,
3966 STRIPE_SECTORS, 0))
3967 md_error(conf->mddev, rdev);
3968 rdev_dec_pending(rdev, conf->mddev);
3969 }
b84db560
N		 3970 if (test_and_clear_bit(R5_MadeGood, &dev->flags)) {
3971 rdev = conf->disks[i].rdev;
3972 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 3973 STRIPE_SECTORS, 0);
b84db560
N		 3974 rdev_dec_pending(rdev, conf->mddev);
3975 }
977df362
N		 3976 if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) {
3977 rdev = conf->disks[i].replacement;
dd054fce
N		 3978 if (!rdev)
3979 /* rdev have been moved down */
3980 rdev = conf->disks[i].rdev;
977df362 3981 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 3982 STRIPE_SECTORS, 0);
977df362
N		 3983 rdev_dec_pending(rdev, conf->mddev);
3984 }
bc2607f3
N		 3985 }
3986
6c0069c0
YT		 3987 if (s.ops_request)
3988 raid_run_ops(sh, s.ops_request);
3989
f0e43bcd 3990 ops_run_io(sh, &s);
16a53ecc 3991
c5709ef6 3992 if (s.dec_preread_active) {
729a1866 3993 /* We delay this until after ops_run_io so that if make_request
e9c7469b 3994 * is waiting on a flush, it won't continue until the writes
729a1866
N		 3995 * have actually been submitted.
3996 */
3997 atomic_dec(&conf->preread_active_stripes);
3998 if (atomic_read(&conf->preread_active_stripes) <
3999 IO_THRESHOLD)
4000 md_wakeup_thread(conf->mddev->thread);
4001 }
4002
c5709ef6 4003 return_io(s.return_bi);
16a53ecc 4004
257a4b42 4005 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
16a53ecc
N		 4006}
4007
d1688a6d 4008static void raid5_activate_delayed(struct r5conf *conf)
16a53ecc
N		 4009{
4010 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
4011 while (!list_empty(&conf->delayed_list)) {
4012 struct list_head *l = conf->delayed_list.next;
4013 struct stripe_head *sh;
4014 sh = list_entry(l, struct stripe_head, lru);
4015 list_del_init(l);
4016 clear_bit(STRIPE_DELAYED, &sh->state);
4017 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4018 atomic_inc(&conf->preread_active_stripes);
8b3e6cdc 4019 list_add_tail(&sh->lru, &conf->hold_list);
851c30c9 4020 raid5_wakeup_stripe_thread(sh);
16a53ecc 4021 }
482c0834 4022 }
16a53ecc
N		 4023}
4024
566c09c5
SL		 4025static void activate_bit_delay(struct r5conf *conf,
4026 struct list_head *temp_inactive_list)
16a53ecc
N		 4027{
4028 /* device_lock is held */
4029 struct list_head head;
4030 list_add(&head, &conf->bitmap_list);
4031 list_del_init(&conf->bitmap_list);
4032 while (!list_empty(&head)) {
4033 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
566c09c5 4034 int hash;
16a53ecc
N		 4035 list_del_init(&sh->lru);
4036 atomic_inc(&sh->count);
566c09c5
SL		 4037 hash = sh->hash_lock_index;
4038 __release_stripe(conf, sh, &temp_inactive_list[hash]);
16a53ecc
N		 4039 }
4040}
4041
fd01b88c 4042int md_raid5_congested(struct mddev *mddev, int bits)
f022b2fd 4043{
d1688a6d 4044 struct r5conf *conf = mddev->private;
f022b2fd
N		 4045
4046 /* No difference between reads and writes. Just check
4047 * how busy the stripe_cache is
4048 */
3fa841d7 4049
f022b2fd
N		 4050 if (conf->inactive_blocked)
4051 return 1;
4052 if (conf->quiesce)
4053 return 1;
4bda556a 4054 if (atomic_read(&conf->empty_inactive_list_nr))
f022b2fd
N		 4055 return 1;
4056
4057 return 0;
4058}
11d8a6e3
N		 4059EXPORT_SYMBOL_GPL(md_raid5_congested);
4060
4061static int raid5_congested(void *data, int bits)
4062{
fd01b88c 4063 struct mddev *mddev = data;
11d8a6e3
N		 4064
4065 return mddev_congested(mddev, bits) ||
4066 md_raid5_congested(mddev, bits);
4067}
f022b2fd 4068
23032a0e
RBJ		 4069/* We want read requests to align with chunks where possible,
4070 * but write requests don't need to.
4071 */
cc371e66
AK		 4072static int raid5_mergeable_bvec(struct request_queue *q,
4073 struct bvec_merge_data *bvm,
4074 struct bio_vec *biovec)
23032a0e 4075{
fd01b88c 4076 struct mddev *mddev = q->queuedata;
cc371e66 4077 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
23032a0e 4078 int max;
9d8f0363 4079 unsigned int chunk_sectors = mddev->chunk_sectors;
cc371e66 4080 unsigned int bio_sectors = bvm->bi_size >> 9;
23032a0e 4081
cc371e66 4082 if ((bvm->bi_rw & 1) == WRITE)
23032a0e
RBJ		 4083 return biovec->bv_len; /* always allow writes to be mergeable */
4084
664e7c41
AN		 4085 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
4086 chunk_sectors = mddev->new_chunk_sectors;
23032a0e
RBJ		 4087 max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
4088 if (max < 0) max = 0;
4089 if (max <= biovec->bv_len && bio_sectors == 0)
4090 return biovec->bv_len;
4091 else
4092 return max;
4093}
4094
f679623f 4095
fd01b88c 4096static int in_chunk_boundary(struct mddev *mddev, struct bio *bio)
f679623f
RBJ		 4097{
4098 sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
9d8f0363 4099 unsigned int chunk_sectors = mddev->chunk_sectors;
aa8b57aa 4100 unsigned int bio_sectors = bio_sectors(bio);
f679623f 4101
664e7c41
AN		 4102 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
4103 chunk_sectors = mddev->new_chunk_sectors;
f679623f
RBJ		 4104 return chunk_sectors >=
4105 ((sector & (chunk_sectors - 1)) + bio_sectors);
4106}
4107
46031f9a
RBJ		 4108/*
4109 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
4110 * later sampled by raid5d.
4111 */
d1688a6d 4112static void add_bio_to_retry(struct bio *bi,struct r5conf *conf)
46031f9a
RBJ		 4113{
4114 unsigned long flags;
4115
4116 spin_lock_irqsave(&conf->device_lock, flags);
4117
4118 bi->bi_next = conf->retry_read_aligned_list;
4119 conf->retry_read_aligned_list = bi;
4120
4121 spin_unlock_irqrestore(&conf->device_lock, flags);
4122 md_wakeup_thread(conf->mddev->thread);
4123}
4124
4125
d1688a6d 4126static struct bio *remove_bio_from_retry(struct r5conf *conf)
46031f9a
RBJ		 4127{
4128 struct bio *bi;
4129
4130 bi = conf->retry_read_aligned;
4131 if (bi) {
4132 conf->retry_read_aligned = NULL;
4133 return bi;
4134 }
4135 bi = conf->retry_read_aligned_list;
4136 if(bi) {
387bb173 4137 conf->retry_read_aligned_list = bi->bi_next;
46031f9a 4138 bi->bi_next = NULL;
960e739d
JA		 4139 /*
4140 * this sets the active strip count to 1 and the processed
4141 * strip count to zero (upper 8 bits)
4142 */
e7836bd6 4143 raid5_set_bi_stripes(bi, 1); /* biased count of active stripes */
46031f9a
RBJ		 4144 }
4145
4146 return bi;
4147}
4148
4149
f679623f
RBJ		 4150/*
4151 * The "raid5_align_endio" should check if the read succeeded and if it
4152 * did, call bio_endio on the original bio (having bio_put the new bio
4153 * first).
4154 * If the read failed..
4155 */
6712ecf8 4156static void raid5_align_endio(struct bio *bi, int error)
f679623f
RBJ		 4157{
4158 struct bio* raid_bi = bi->bi_private;
fd01b88c 4159 struct mddev *mddev;
d1688a6d 4160 struct r5conf *conf;
46031f9a 4161 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
3cb03002 4162 struct md_rdev *rdev;
46031f9a 4163
f679623f 4164 bio_put(bi);
46031f9a 4165
46031f9a
RBJ		 4166 rdev = (void*)raid_bi->bi_next;
4167 raid_bi->bi_next = NULL;
2b7f2228
N		 4168 mddev = rdev->mddev;
4169 conf = mddev->private;
46031f9a
RBJ		 4170
4171 rdev_dec_pending(rdev, conf->mddev);
4172
4173 if (!error && uptodate) {
0a82a8d1
LT		 4174 trace_block_bio_complete(bdev_get_queue(raid_bi->bi_bdev),
4175 raid_bi, 0);
6712ecf8 4176 bio_endio(raid_bi, 0);
46031f9a
RBJ		 4177 if (atomic_dec_and_test(&conf->active_aligned_reads))
4178 wake_up(&conf->wait_for_stripe);
6712ecf8 4179 return;
46031f9a
RBJ		 4180 }
4181
4182
45b4233c 4183 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
46031f9a
RBJ		 4184
4185 add_bio_to_retry(raid_bi, conf);
f679623f
RBJ		 4186}
4187
387bb173
NB		 4188static int bio_fits_rdev(struct bio *bi)
4189{
165125e1 4190 struct request_queue *q = bdev_get_queue(bi->bi_bdev);
387bb173 4191
aa8b57aa 4192 if (bio_sectors(bi) > queue_max_sectors(q))
387bb173
NB		 4193 return 0;
4194 blk_recount_segments(q, bi);
8a78362c 4195 if (bi->bi_phys_segments > queue_max_segments(q))
387bb173
NB		 4196 return 0;
4197
4198 if (q->merge_bvec_fn)
4199 /* it's too hard to apply the merge_bvec_fn at this stage,
4200 * just just give up
4201 */
4202 return 0;
4203
4204 return 1;
4205}
4206
4207
fd01b88c 4208static int chunk_aligned_read(struct mddev *mddev, struct bio * raid_bio)
f679623f 4209{
d1688a6d 4210 struct r5conf *conf = mddev->private;
8553fe7e 4211 int dd_idx;
f679623f 4212 struct bio* align_bi;
3cb03002 4213 struct md_rdev *rdev;
671488cc 4214 sector_t end_sector;
f679623f
RBJ		 4215
4216 if (!in_chunk_boundary(mddev, raid_bio)) {
45b4233c 4217 pr_debug("chunk_aligned_read : non aligned\n");
f679623f
RBJ		 4218 return 0;
4219 }
4220 /*
a167f663 4221 * use bio_clone_mddev to make a copy of the bio
f679623f 4222 */
a167f663 4223 align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev);
f679623f
RBJ		 4224 if (!align_bi)
4225 return 0;
4226 /*
4227 * set bi_end_io to a new function, and set bi_private to the
4228 * original bio.
4229 */
4230 align_bi->bi_end_io = raid5_align_endio;
4231 align_bi->bi_private = raid_bio;
4232 /*
4233 * compute position
4234 */
112bf897
N		 4235 align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector,
4236 0,
911d4ee8 4237 &dd_idx, NULL);
f679623f 4238
f73a1c7d 4239 end_sector = bio_end_sector(align_bi);
f679623f 4240 rcu_read_lock();
671488cc
N		 4241 rdev = rcu_dereference(conf->disks[dd_idx].replacement);
4242 if (!rdev || test_bit(Faulty, &rdev->flags) ||
4243 rdev->recovery_offset < end_sector) {
4244 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
4245 if (rdev &&
4246 (test_bit(Faulty, &rdev->flags) ||
4247 !(test_bit(In_sync, &rdev->flags) ||
4248 rdev->recovery_offset >= end_sector)))
4249 rdev = NULL;
4250 }
4251 if (rdev) {
31c176ec
N		 4252 sector_t first_bad;
4253 int bad_sectors;
4254
f679623f
RBJ		 4255 atomic_inc(&rdev->nr_pending);
4256 rcu_read_unlock();
46031f9a
RBJ		 4257 raid_bio->bi_next = (void*)rdev;
4258 align_bi->bi_bdev = rdev->bdev;
4259 align_bi->bi_flags &= ~(1 << BIO_SEG_VALID);
46031f9a 4260
31c176ec 4261 if (!bio_fits_rdev(align_bi) ||
aa8b57aa 4262 is_badblock(rdev, align_bi->bi_sector, bio_sectors(align_bi),
31c176ec
N		 4263 &first_bad, &bad_sectors)) {
4264 /* too big in some way, or has a known bad block */
387bb173
NB		 4265 bio_put(align_bi);
4266 rdev_dec_pending(rdev, mddev);
4267 return 0;
4268 }
4269
6c0544e2 4270 /* No reshape active, so we can trust rdev->data_offset */
4271 align_bi->bi_sector += rdev->data_offset;
4272
46031f9a
RBJ		 4273 spin_lock_irq(&conf->device_lock);
4274 wait_event_lock_irq(conf->wait_for_stripe,
4275 conf->quiesce == 0,
eed8c02e 4276 conf->device_lock);
46031f9a
RBJ		 4277 atomic_inc(&conf->active_aligned_reads);
4278 spin_unlock_irq(&conf->device_lock);
4279
e3620a3a
JB		 4280 if (mddev->gendisk)
4281 trace_block_bio_remap(bdev_get_queue(align_bi->bi_bdev),
4282 align_bi, disk_devt(mddev->gendisk),
4283 raid_bio->bi_sector);
f679623f
RBJ		 4284 generic_make_request(align_bi);
4285 return 1;
4286 } else {
4287 rcu_read_unlock();
46031f9a 4288 bio_put(align_bi);
f679623f
RBJ		 4289 return 0;
4290 }
4291}
4292
8b3e6cdc
DW		 4293/* __get_priority_stripe - get the next stripe to process
4294 *
4295 * Full stripe writes are allowed to pass preread active stripes up until
4296 * the bypass_threshold is exceeded. In general the bypass_count
4297 * increments when the handle_list is handled before the hold_list; however, it
4298 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
4299 * stripe with in flight i/o. The bypass_count will be reset when the
4300 * head of the hold_list has changed, i.e. the head was promoted to the
4301 * handle_list.
4302 */
851c30c9 4303static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group)
8b3e6cdc 4304{
851c30c9
SL		 4305 struct stripe_head *sh = NULL, *tmp;
4306 struct list_head *handle_list = NULL;
bfc90cb0 4307 struct r5worker_group *wg = NULL;
851c30c9
SL		 4308
4309 if (conf->worker_cnt_per_group == 0) {
4310 handle_list = &conf->handle_list;
4311 } else if (group != ANY_GROUP) {
4312 handle_list = &conf->worker_groups[group].handle_list;
bfc90cb0 4313 wg = &conf->worker_groups[group];
851c30c9
SL		 4314 } else {
4315 int i;
4316 for (i = 0; i < conf->group_cnt; i++) {
4317 handle_list = &conf->worker_groups[i].handle_list;
bfc90cb0 4318 wg = &conf->worker_groups[i];
851c30c9
SL		 4319 if (!list_empty(handle_list))
4320 break;
4321 }
4322 }
8b3e6cdc
DW		 4323
4324 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
4325 __func__,
851c30c9 4326 list_empty(handle_list) ? "empty" : "busy",
8b3e6cdc
DW		 4327 list_empty(&conf->hold_list) ? "empty" : "busy",
4328 atomic_read(&conf->pending_full_writes), conf->bypass_count);
4329
851c30c9
SL		 4330 if (!list_empty(handle_list)) {
4331 sh = list_entry(handle_list->next, typeof(*sh), lru);
8b3e6cdc
DW		 4332
4333 if (list_empty(&conf->hold_list))
4334 conf->bypass_count = 0;
4335 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
4336 if (conf->hold_list.next == conf->last_hold)
4337 conf->bypass_count++;
4338 else {
4339 conf->last_hold = conf->hold_list.next;
4340 conf->bypass_count -= conf->bypass_threshold;
4341 if (conf->bypass_count < 0)
4342 conf->bypass_count = 0;
4343 }
4344 }
4345 } else if (!list_empty(&conf->hold_list) &&
4346 ((conf->bypass_threshold &&
4347 conf->bypass_count > conf->bypass_threshold) ||
4348 atomic_read(&conf->pending_full_writes) == 0)) {
851c30c9
SL		 4349
4350 list_for_each_entry(tmp, &conf->hold_list, lru) {
4351 if (conf->worker_cnt_per_group == 0 ||
4352 group == ANY_GROUP ||
4353 !cpu_online(tmp->cpu) ||
4354 cpu_to_group(tmp->cpu) == group) {
4355 sh = tmp;
4356 break;
4357 }
4358 }
4359
4360 if (sh) {
4361 conf->bypass_count -= conf->bypass_threshold;
4362 if (conf->bypass_count < 0)
4363 conf->bypass_count = 0;
4364 }
bfc90cb0 4365 wg = NULL;
851c30c9
SL		 4366 }
4367
4368 if (!sh)
8b3e6cdc
DW		 4369 return NULL;
4370
bfc90cb0
SL		 4371 if (wg) {
4372 wg->stripes_cnt--;
4373 sh->group = NULL;
4374 }
8b3e6cdc
DW		 4375 list_del_init(&sh->lru);
4376 atomic_inc(&sh->count);
4377 BUG_ON(atomic_read(&sh->count) != 1);
4378 return sh;
4379}
f679623f 4380
8811b596
SL		 4381struct raid5_plug_cb {
4382 struct blk_plug_cb cb;
4383 struct list_head list;
566c09c5 4384 struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
8811b596
SL		 4385};
4386
4387static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule)
4388{
4389 struct raid5_plug_cb *cb = container_of(
4390 blk_cb, struct raid5_plug_cb, cb);
4391 struct stripe_head *sh;
4392 struct mddev *mddev = cb->cb.data;
4393 struct r5conf *conf = mddev->private;
a9add5d9 4394 int cnt = 0;
566c09c5 4395 int hash;
8811b596
SL		 4396
4397 if (cb->list.next && !list_empty(&cb->list)) {
4398 spin_lock_irq(&conf->device_lock);
4399 while (!list_empty(&cb->list)) {
4400 sh = list_first_entry(&cb->list, struct stripe_head, lru);
4401 list_del_init(&sh->lru);
4402 /*
4403 * avoid race release_stripe_plug() sees
4404 * STRIPE_ON_UNPLUG_LIST clear but the stripe
4405 * is still in our list
4406 */
4407 smp_mb__before_clear_bit();
4408 clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state);
773ca82f
SL		 4409 /*
4410 * STRIPE_ON_RELEASE_LIST could be set here. In that
4411 * case, the count is always > 1 here
4412 */
566c09c5
SL		 4413 hash = sh->hash_lock_index;
4414 __release_stripe(conf, sh, &cb->temp_inactive_list[hash]);
a9add5d9 4415 cnt++;
8811b596
SL		 4416 }
4417 spin_unlock_irq(&conf->device_lock);
4418 }
566c09c5
SL		 4419 release_inactive_stripe_list(conf, cb->temp_inactive_list,
4420 NR_STRIPE_HASH_LOCKS);
e3620a3a
JB		 4421 if (mddev->queue)
4422 trace_block_unplug(mddev->queue, cnt, !from_schedule);
8811b596
SL		 4423 kfree(cb);
4424}
4425
4426static void release_stripe_plug(struct mddev *mddev,
4427 struct stripe_head *sh)
4428{
4429 struct blk_plug_cb *blk_cb = blk_check_plugged(
4430 raid5_unplug, mddev,
4431 sizeof(struct raid5_plug_cb));
4432 struct raid5_plug_cb *cb;
4433
4434 if (!blk_cb) {
4435 release_stripe(sh);
4436 return;
4437 }
4438
4439 cb = container_of(blk_cb, struct raid5_plug_cb, cb);
4440
566c09c5
SL		 4441 if (cb->list.next == NULL) {
4442 int i;
8811b596 4443 INIT_LIST_HEAD(&cb->list);
566c09c5
SL		 4444 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
4445 INIT_LIST_HEAD(cb->temp_inactive_list + i);
4446 }
8811b596
SL		 4447
4448 if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state))
4449 list_add_tail(&sh->lru, &cb->list);
4450 else
4451 release_stripe(sh);
4452}
4453
620125f2
SL		 4454static void make_discard_request(struct mddev *mddev, struct bio *bi)
4455{
4456 struct r5conf *conf = mddev->private;
4457 sector_t logical_sector, last_sector;
4458 struct stripe_head *sh;
4459 int remaining;
4460 int stripe_sectors;
4461
4462 if (mddev->reshape_position != MaxSector)
4463 /* Skip discard while reshape is happening */
4464 return;
4465
4466 logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
4467 last_sector = bi->bi_sector + (bi->bi_size>>9);
4468
4469 bi->bi_next = NULL;
4470 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
4471
4472 stripe_sectors = conf->chunk_sectors *
4473 (conf->raid_disks - conf->max_degraded);
4474 logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector,
4475 stripe_sectors);
4476 sector_div(last_sector, stripe_sectors);
4477
4478 logical_sector *= conf->chunk_sectors;
4479 last_sector *= conf->chunk_sectors;
4480
4481 for (; logical_sector < last_sector;
4482 logical_sector += STRIPE_SECTORS) {
4483 DEFINE_WAIT(w);
4484 int d;
4485 again:
4486 sh = get_active_stripe(conf, logical_sector, 0, 0, 0);
4487 prepare_to_wait(&conf->wait_for_overlap, &w,
4488 TASK_UNINTERRUPTIBLE);
f8dfcffd
N		 4489 set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
4490 if (test_bit(STRIPE_SYNCING, &sh->state)) {
4491 release_stripe(sh);
4492 schedule();
4493 goto again;
4494 }
4495 clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
620125f2
SL		 4496 spin_lock_irq(&sh->stripe_lock);
4497 for (d = 0; d < conf->raid_disks; d++) {
4498 if (d == sh->pd_idx || d == sh->qd_idx)
4499 continue;
4500 if (sh->dev[d].towrite || sh->dev[d].toread) {
4501 set_bit(R5_Overlap, &sh->dev[d].flags);
4502 spin_unlock_irq(&sh->stripe_lock);
4503 release_stripe(sh);
4504 schedule();
4505 goto again;
4506 }
4507 }
f8dfcffd 4508 set_bit(STRIPE_DISCARD, &sh->state);
620125f2
SL		 4509 finish_wait(&conf->wait_for_overlap, &w);
4510 for (d = 0; d < conf->raid_disks; d++) {
4511 if (d == sh->pd_idx || d == sh->qd_idx)
4512 continue;
4513 sh->dev[d].towrite = bi;
4514 set_bit(R5_OVERWRITE, &sh->dev[d].flags);
4515 raid5_inc_bi_active_stripes(bi);
4516 }
4517 spin_unlock_irq(&sh->stripe_lock);
4518 if (conf->mddev->bitmap) {
4519 for (d = 0;
4520 d < conf->raid_disks - conf->max_degraded;
4521 d++)
4522 bitmap_startwrite(mddev->bitmap,
4523 sh->sector,
4524 STRIPE_SECTORS,
4525 0);
4526 sh->bm_seq = conf->seq_flush + 1;
4527 set_bit(STRIPE_BIT_DELAY, &sh->state);
4528 }
4529
4530 set_bit(STRIPE_HANDLE, &sh->state);
4531 clear_bit(STRIPE_DELAYED, &sh->state);
4532 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4533 atomic_inc(&conf->preread_active_stripes);
4534 release_stripe_plug(mddev, sh);
4535 }
4536
4537 remaining = raid5_dec_bi_active_stripes(bi);
4538 if (remaining == 0) {
4539 md_write_end(mddev);
4540 bio_endio(bi, 0);
4541 }
4542}
4543
b4fdcb02 4544static void make_request(struct mddev *mddev, struct bio * bi)
1da177e4 4545{
d1688a6d 4546 struct r5conf *conf = mddev->private;
911d4ee8 4547 int dd_idx;
1da177e4
LT		 4548 sector_t new_sector;
4549 sector_t logical_sector, last_sector;
4550 struct stripe_head *sh;
a362357b 4551 const int rw = bio_data_dir(bi);
49077326 4552 int remaining;
1da177e4 4553
e9c7469b
TH		 4554 if (unlikely(bi->bi_rw & REQ_FLUSH)) {
4555 md_flush_request(mddev, bi);
5a7bbad2 4556 return;
e5dcdd80
N		 4557 }
4558
3d310eb7 4559 md_write_start(mddev, bi);
06d91a5f 4560
802ba064 4561 if (rw == READ &&
52488615 4562 mddev->reshape_position == MaxSector &&
21a52c6d 4563 chunk_aligned_read(mddev,bi))
5a7bbad2 4564 return;
52488615 4565
620125f2
SL		 4566 if (unlikely(bi->bi_rw & REQ_DISCARD)) {
4567 make_discard_request(mddev, bi);
4568 return;
4569 }
4570
1da177e4 4571 logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
f73a1c7d 4572 last_sector = bio_end_sector(bi);
1da177e4
LT		 4573 bi->bi_next = NULL;
4574 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
06d91a5f 4575
1da177e4
LT		 4576 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
4577 DEFINE_WAIT(w);
b5663ba4 4578 int previous;
c46501b2 4579 int seq;
b578d55f 4580
7ecaa1e6 4581 retry:
c46501b2 4582 seq = read_seqcount_begin(&conf->gen_lock);
b5663ba4 4583 previous = 0;
b578d55f 4584 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
b0f9ec04 4585 if (unlikely(conf->reshape_progress != MaxSector)) {
fef9c61f 4586 /* spinlock is needed as reshape_progress may be
df8e7f76
N		 4587 * 64bit on a 32bit platform, and so it might be
4588 * possible to see a half-updated value
aeb878b0 4589 * Of course reshape_progress could change after
df8e7f76
N		 4590 * the lock is dropped, so once we get a reference
4591 * to the stripe that we think it is, we will have
4592 * to check again.
4593 */
7ecaa1e6 4594 spin_lock_irq(&conf->device_lock);
2c810cdd 4595 if (mddev->reshape_backwards
fef9c61f
N		 4596 ? logical_sector < conf->reshape_progress
4597 : logical_sector >= conf->reshape_progress) {
b5663ba4
N		 4598 previous = 1;
4599 } else {
2c810cdd 4600 if (mddev->reshape_backwards
fef9c61f
N		 4601 ? logical_sector < conf->reshape_safe
4602 : logical_sector >= conf->reshape_safe) {
b578d55f
N		 4603 spin_unlock_irq(&conf->device_lock);
4604 schedule();
4605 goto retry;
4606 }
4607 }
7ecaa1e6
N		 4608 spin_unlock_irq(&conf->device_lock);
4609 }
16a53ecc 4610
112bf897
N		 4611 new_sector = raid5_compute_sector(conf, logical_sector,
4612 previous,
911d4ee8 4613 &dd_idx, NULL);
0c55e022 4614 pr_debug("raid456: make_request, sector %llu logical %llu\n",
c46501b2 4615 (unsigned long long)new_sector,
1da177e4
LT		 4616 (unsigned long long)logical_sector);
4617
b5663ba4 4618 sh = get_active_stripe(conf, new_sector, previous,
a8c906ca 4619 (bi->bi_rw&RWA_MASK), 0);
1da177e4 4620 if (sh) {
b0f9ec04 4621 if (unlikely(previous)) {
7ecaa1e6 4622 /* expansion might have moved on while waiting for a
df8e7f76
N		 4623 * stripe, so we must do the range check again.
4624 * Expansion could still move past after this
4625 * test, but as we are holding a reference to
4626 * 'sh', we know that if that happens,
4627 * STRIPE_EXPANDING will get set and the expansion
4628 * won't proceed until we finish with the stripe.
7ecaa1e6
N		 4629 */
4630 int must_retry = 0;
4631 spin_lock_irq(&conf->device_lock);
2c810cdd 4632 if (mddev->reshape_backwards
b0f9ec04
N		 4633 ? logical_sector >= conf->reshape_progress
4634 : logical_sector < conf->reshape_progress)
7ecaa1e6
N		 4635 /* mismatch, need to try again */
4636 must_retry = 1;
4637 spin_unlock_irq(&conf->device_lock);
4638 if (must_retry) {
4639 release_stripe(sh);
7a3ab908 4640 schedule();
7ecaa1e6
N		 4641 goto retry;
4642 }
4643 }
c46501b2
N		 4644 if (read_seqcount_retry(&conf->gen_lock, seq)) {
4645 /* Might have got the wrong stripe_head
4646 * by accident
4647 */
4648 release_stripe(sh);
4649 goto retry;
4650 }
e62e58a5 4651
ffd96e35 4652 if (rw == WRITE &&
a5c308d4 4653 logical_sector >= mddev->suspend_lo &&
e464eafd
N		 4654 logical_sector < mddev->suspend_hi) {
4655 release_stripe(sh);
e62e58a5
N		 4656 /* As the suspend_* range is controlled by
4657 * userspace, we want an interruptible
4658 * wait.
4659 */
4660 flush_signals(current);
4661 prepare_to_wait(&conf->wait_for_overlap,
4662 &w, TASK_INTERRUPTIBLE);
4663 if (logical_sector >= mddev->suspend_lo &&
4664 logical_sector < mddev->suspend_hi)
4665 schedule();
e464eafd
N		 4666 goto retry;
4667 }
7ecaa1e6
N		 4668
4669 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
ffd96e35 4670 !add_stripe_bio(sh, bi, dd_idx, rw)) {
7ecaa1e6
N		 4671 /* Stripe is busy expanding or
4672 * add failed due to overlap. Flush everything
1da177e4
LT		 4673 * and wait a while
4674 */
482c0834 4675 md_wakeup_thread(mddev->thread);
1da177e4
LT		 4676 release_stripe(sh);
4677 schedule();
4678 goto retry;
4679 }
4680 finish_wait(&conf->wait_for_overlap, &w);
6ed3003c
N		 4681 set_bit(STRIPE_HANDLE, &sh->state);
4682 clear_bit(STRIPE_DELAYED, &sh->state);
a852d7b8 4683 if ((bi->bi_rw & REQ_SYNC) &&
729a1866
N		 4684 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4685 atomic_inc(&conf->preread_active_stripes);
8811b596 4686 release_stripe_plug(mddev, sh);
1da177e4
LT		 4687 } else {
4688 /* cannot get stripe for read-ahead, just give-up */
4689 clear_bit(BIO_UPTODATE, &bi->bi_flags);
4690 finish_wait(&conf->wait_for_overlap, &w);
4691 break;
4692 }
1da177e4 4693 }
7c13edc8 4694
e7836bd6 4695 remaining = raid5_dec_bi_active_stripes(bi);
f6344757 4696 if (remaining == 0) {
1da177e4 4697
16a53ecc 4698 if ( rw == WRITE )
1da177e4 4699 md_write_end(mddev);
6712ecf8 4700
0a82a8d1
LT		 4701 trace_block_bio_complete(bdev_get_queue(bi->bi_bdev),
4702 bi, 0);
0e13fe23 4703 bio_endio(bi, 0);
1da177e4 4704 }
1da177e4
LT		 4705}
4706
fd01b88c 4707static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks);
b522adcd 4708
fd01b88c 4709static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped)
1da177e4 4710{
52c03291
N		 4711 /* reshaping is quite different to recovery/resync so it is
4712 * handled quite separately ... here.
4713 *
4714 * On each call to sync_request, we gather one chunk worth of
4715 * destination stripes and flag them as expanding.
4716 * Then we find all the source stripes and request reads.
4717 * As the reads complete, handle_stripe will copy the data
4718 * into the destination stripe and release that stripe.
4719 */
d1688a6d 4720 struct r5conf *conf = mddev->private;
1da177e4 4721 struct stripe_head *sh;
ccfcc3c1 4722 sector_t first_sector, last_sector;
f416885e
N		 4723 int raid_disks = conf->previous_raid_disks;
4724 int data_disks = raid_disks - conf->max_degraded;
4725 int new_data_disks = conf->raid_disks - conf->max_degraded;
52c03291
N		 4726 int i;
4727 int dd_idx;
c8f517c4 4728 sector_t writepos, readpos, safepos;
ec32a2bd 4729 sector_t stripe_addr;
7a661381 4730 int reshape_sectors;
ab69ae12 4731 struct list_head stripes;
52c03291 4732
fef9c61f
N		 4733 if (sector_nr == 0) {
4734 /* If restarting in the middle, skip the initial sectors */
2c810cdd 4735 if (mddev->reshape_backwards &&
fef9c61f
N		 4736 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
4737 sector_nr = raid5_size(mddev, 0, 0)
4738 - conf->reshape_progress;
2c810cdd 4739 } else if (!mddev->reshape_backwards &&
fef9c61f
N		 4740 conf->reshape_progress > 0)
4741 sector_nr = conf->reshape_progress;
f416885e 4742 sector_div(sector_nr, new_data_disks);
fef9c61f 4743 if (sector_nr) {
8dee7211
N		 4744 mddev->curr_resync_completed = sector_nr;
4745 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
fef9c61f
N		 4746 *skipped = 1;
4747 return sector_nr;
4748 }
52c03291
N		 4749 }
4750
7a661381
N		 4751 /* We need to process a full chunk at a time.
4752 * If old and new chunk sizes differ, we need to process the
4753 * largest of these
4754 */
664e7c41
AN		 4755 if (mddev->new_chunk_sectors > mddev->chunk_sectors)
4756 reshape_sectors = mddev->new_chunk_sectors;
7a661381 4757 else
9d8f0363 4758 reshape_sectors = mddev->chunk_sectors;
7a661381 4759
b5254dd5
N		 4760 /* We update the metadata at least every 10 seconds, or when
4761 * the data about to be copied would over-write the source of
4762 * the data at the front of the range. i.e. one new_stripe
4763 * along from reshape_progress new_maps to after where
4764 * reshape_safe old_maps to
52c03291 4765 */
fef9c61f 4766 writepos = conf->reshape_progress;
f416885e 4767 sector_div(writepos, new_data_disks);
c8f517c4
N		 4768 readpos = conf->reshape_progress;
4769 sector_div(readpos, data_disks);
fef9c61f 4770 safepos = conf->reshape_safe;
f416885e 4771 sector_div(safepos, data_disks);
2c810cdd 4772 if (mddev->reshape_backwards) {
ed37d83e 4773 writepos -= min_t(sector_t, reshape_sectors, writepos);
c8f517c4 4774 readpos += reshape_sectors;
7a661381 4775 safepos += reshape_sectors;
fef9c61f 4776 } else {
7a661381 4777 writepos += reshape_sectors;
ed37d83e
N		 4778 readpos -= min_t(sector_t, reshape_sectors, readpos);
4779 safepos -= min_t(sector_t, reshape_sectors, safepos);
fef9c61f 4780 }
52c03291 4781
b5254dd5
N		 4782 /* Having calculated the 'writepos' possibly use it
4783 * to set 'stripe_addr' which is where we will write to.
4784 */
4785 if (mddev->reshape_backwards) {
4786 BUG_ON(conf->reshape_progress == 0);
4787 stripe_addr = writepos;
4788 BUG_ON((mddev->dev_sectors &
4789 ~((sector_t)reshape_sectors - 1))
4790 - reshape_sectors - stripe_addr
4791 != sector_nr);
4792 } else {
4793 BUG_ON(writepos != sector_nr + reshape_sectors);
4794 stripe_addr = sector_nr;
4795 }
4796
c8f517c4
N		 4797 /* 'writepos' is the most advanced device address we might write.
4798 * 'readpos' is the least advanced device address we might read.
4799 * 'safepos' is the least address recorded in the metadata as having
4800 * been reshaped.
b5254dd5
N		 4801 * If there is a min_offset_diff, these are adjusted either by
4802 * increasing the safepos/readpos if diff is negative, or
4803 * increasing writepos if diff is positive.
4804 * If 'readpos' is then behind 'writepos', there is no way that we can
c8f517c4
N		 4805 * ensure safety in the face of a crash - that must be done by userspace
4806 * making a backup of the data. So in that case there is no particular
4807 * rush to update metadata.
4808 * Otherwise if 'safepos' is behind 'writepos', then we really need to
4809 * update the metadata to advance 'safepos' to match 'readpos' so that
4810 * we can be safe in the event of a crash.
4811 * So we insist on updating metadata if safepos is behind writepos and
4812 * readpos is beyond writepos.
4813 * In any case, update the metadata every 10 seconds.
4814 * Maybe that number should be configurable, but I'm not sure it is
4815 * worth it.... maybe it could be a multiple of safemode_delay???
4816 */
b5254dd5
N		 4817 if (conf->min_offset_diff < 0) {
4818 safepos += -conf->min_offset_diff;
4819 readpos += -conf->min_offset_diff;
4820 } else
4821 writepos += conf->min_offset_diff;
4822
2c810cdd 4823 if ((mddev->reshape_backwards
c8f517c4
N		 4824 ? (safepos > writepos && readpos < writepos)
4825 : (safepos < writepos && readpos > writepos)) ||
4826 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
52c03291
N		 4827 /* Cannot proceed until we've updated the superblock... */
4828 wait_event(conf->wait_for_overlap,
c91abf5a
N		 4829 atomic_read(&conf->reshape_stripes)==0
4830 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
4831 if (atomic_read(&conf->reshape_stripes) != 0)
4832 return 0;
fef9c61f 4833 mddev->reshape_position = conf->reshape_progress;
75d3da43 4834 mddev->curr_resync_completed = sector_nr;
c8f517c4 4835 conf->reshape_checkpoint = jiffies;
850b2b42 4836 set_bit(MD_CHANGE_DEVS, &mddev->flags);
52c03291 4837 md_wakeup_thread(mddev->thread);
850b2b42 4838 wait_event(mddev->sb_wait, mddev->flags == 0 ||
c91abf5a
N		 4839 test_bit(MD_RECOVERY_INTR, &mddev->recovery));
4840 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
4841 return 0;
52c03291 4842 spin_lock_irq(&conf->device_lock);
fef9c61f 4843 conf->reshape_safe = mddev->reshape_position;
52c03291
N		 4844 spin_unlock_irq(&conf->device_lock);
4845 wake_up(&conf->wait_for_overlap);
acb180b0 4846 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
52c03291
N		 4847 }
4848
ab69ae12 4849 INIT_LIST_HEAD(&stripes);
7a661381 4850 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
52c03291 4851 int j;
a9f326eb 4852 int skipped_disk = 0;
a8c906ca 4853 sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
52c03291
N		 4854 set_bit(STRIPE_EXPANDING, &sh->state);
4855 atomic_inc(&conf->reshape_stripes);
4856 /* If any of this stripe is beyond the end of the old
4857 * array, then we need to zero those blocks
4858 */
4859 for (j=sh->disks; j--;) {
4860 sector_t s;
4861 if (j == sh->pd_idx)
4862 continue;
f416885e 4863 if (conf->level == 6 &&
d0dabf7e 4864 j == sh->qd_idx)
f416885e 4865 continue;
784052ec 4866 s = compute_blocknr(sh, j, 0);
b522adcd 4867 if (s < raid5_size(mddev, 0, 0)) {
a9f326eb 4868 skipped_disk = 1;
52c03291
N		 4869 continue;
4870 }
4871 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
4872 set_bit(R5_Expanded, &sh->dev[j].flags);
4873 set_bit(R5_UPTODATE, &sh->dev[j].flags);
4874 }
a9f326eb 4875 if (!skipped_disk) {
52c03291
N		 4876 set_bit(STRIPE_EXPAND_READY, &sh->state);
4877 set_bit(STRIPE_HANDLE, &sh->state);
4878 }
ab69ae12 4879 list_add(&sh->lru, &stripes);
52c03291
N		 4880 }
4881 spin_lock_irq(&conf->device_lock);
2c810cdd 4882 if (mddev->reshape_backwards)
7a661381 4883 conf->reshape_progress -= reshape_sectors * new_data_disks;
fef9c61f 4884 else
7a661381 4885 conf->reshape_progress += reshape_sectors * new_data_disks;
52c03291
N		 4886 spin_unlock_irq(&conf->device_lock);
4887 /* Ok, those stripe are ready. We can start scheduling
4888 * reads on the source stripes.
4889 * The source stripes are determined by mapping the first and last
4890 * block on the destination stripes.
4891 */
52c03291 4892 first_sector =
ec32a2bd 4893 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
911d4ee8 4894 1, &dd_idx, NULL);
52c03291 4895 last_sector =
0e6e0271 4896 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
09c9e5fa 4897 * new_data_disks - 1),
911d4ee8 4898 1, &dd_idx, NULL);
58c0fed4
AN		 4899 if (last_sector >= mddev->dev_sectors)
4900 last_sector = mddev->dev_sectors - 1;
52c03291 4901 while (first_sector <= last_sector) {
a8c906ca 4902 sh = get_active_stripe(conf, first_sector, 1, 0, 1);
52c03291
N		 4903 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
4904 set_bit(STRIPE_HANDLE, &sh->state);
4905 release_stripe(sh);
4906 first_sector += STRIPE_SECTORS;
4907 }
ab69ae12
N		 4908 /* Now that the sources are clearly marked, we can release
4909 * the destination stripes
4910 */
4911 while (!list_empty(&stripes)) {
4912 sh = list_entry(stripes.next, struct stripe_head, lru);
4913 list_del_init(&sh->lru);
4914 release_stripe(sh);
4915 }
c6207277
N		 4916 /* If this takes us to the resync_max point where we have to pause,
4917 * then we need to write out the superblock.
4918 */
7a661381 4919 sector_nr += reshape_sectors;
c03f6a19
N		 4920 if ((sector_nr - mddev->curr_resync_completed) * 2
4921 >= mddev->resync_max - mddev->curr_resync_completed) {
c6207277
N		 4922 /* Cannot proceed until we've updated the superblock... */
4923 wait_event(conf->wait_for_overlap,
c91abf5a
N		 4924 atomic_read(&conf->reshape_stripes) == 0
4925 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
4926 if (atomic_read(&conf->reshape_stripes) != 0)
4927 goto ret;
fef9c61f 4928 mddev->reshape_position = conf->reshape_progress;
75d3da43 4929 mddev->curr_resync_completed = sector_nr;
c8f517c4 4930 conf->reshape_checkpoint = jiffies;
c6207277
N		 4931 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4932 md_wakeup_thread(mddev->thread);
4933 wait_event(mddev->sb_wait,
4934 !test_bit(MD_CHANGE_DEVS, &mddev->flags)
c91abf5a
N		 4935 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
4936 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
4937 goto ret;
c6207277 4938 spin_lock_irq(&conf->device_lock);
fef9c61f 4939 conf->reshape_safe = mddev->reshape_position;
c6207277
N		 4940 spin_unlock_irq(&conf->device_lock);
4941 wake_up(&conf->wait_for_overlap);
acb180b0 4942 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
c6207277 4943 }
c91abf5a 4944ret:
7a661381 4945 return reshape_sectors;
52c03291
N		 4946}
4947
4948/* FIXME go_faster isn't used */
fd01b88c 4949static inline sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster)
52c03291 4950{
d1688a6d 4951 struct r5conf *conf = mddev->private;
52c03291 4952 struct stripe_head *sh;
58c0fed4 4953 sector_t max_sector = mddev->dev_sectors;
57dab0bd 4954 sector_t sync_blocks;
16a53ecc
N		 4955 int still_degraded = 0;
4956 int i;
1da177e4 4957
72626685 4958 if (sector_nr >= max_sector) {
1da177e4 4959 /* just being told to finish up .. nothing much to do */
cea9c228 4960
29269553
N		 4961 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
4962 end_reshape(conf);
4963 return 0;
4964 }
72626685
N		 4965
4966 if (mddev->curr_resync < max_sector) /* aborted */
4967 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
4968 &sync_blocks, 1);
16a53ecc 4969 else /* completed sync */
72626685
N		 4970 conf->fullsync = 0;
4971 bitmap_close_sync(mddev->bitmap);
4972
1da177e4
LT		 4973 return 0;
4974 }
ccfcc3c1 4975
64bd660b
N		 4976 /* Allow raid5_quiesce to complete */
4977 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
4978
52c03291
N		 4979 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4980 return reshape_request(mddev, sector_nr, skipped);
f6705578 4981
c6207277
N		 4982 /* No need to check resync_max as we never do more than one
4983 * stripe, and as resync_max will always be on a chunk boundary,
4984 * if the check in md_do_sync didn't fire, there is no chance
4985 * of overstepping resync_max here
4986 */
4987
16a53ecc 4988 /* if there is too many failed drives and we are trying
1da177e4
LT		 4989 * to resync, then assert that we are finished, because there is
4990 * nothing we can do.
4991 */
3285edf1 4992 if (mddev->degraded >= conf->max_degraded &&
16a53ecc 4993 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
58c0fed4 4994 sector_t rv = mddev->dev_sectors - sector_nr;
57afd89f 4995 *skipped = 1;
1da177e4
LT		 4996 return rv;
4997 }
6f608040 4998 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
4999 !conf->fullsync &&
5000 !bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
5001 sync_blocks >= STRIPE_SECTORS) {
72626685
N		 5002 /* we can skip this block, and probably more */
5003 sync_blocks /= STRIPE_SECTORS;
5004 *skipped = 1;
5005 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
5006 }
1da177e4 5007
b47490c9
N		 5008 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
5009
a8c906ca 5010 sh = get_active_stripe(conf, sector_nr, 0, 1, 0);
1da177e4 5011 if (sh == NULL) {
a8c906ca 5012 sh = get_active_stripe(conf, sector_nr, 0, 0, 0);
1da177e4 5013 /* make sure we don't swamp the stripe cache if someone else
16a53ecc 5014 * is trying to get access
1da177e4 5015 */
66c006a5 5016 schedule_timeout_uninterruptible(1);
1da177e4 5017 }
16a53ecc
N		 5018 /* Need to check if array will still be degraded after recovery/resync
5019 * We don't need to check the 'failed' flag as when that gets set,
5020 * recovery aborts.
5021 */
f001a70c 5022 for (i = 0; i < conf->raid_disks; i++)
16a53ecc
N		 5023 if (conf->disks[i].rdev == NULL)
5024 still_degraded = 1;
5025
5026 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
5027
83206d66 5028 set_bit(STRIPE_SYNC_REQUESTED, &sh->state);
1da177e4 5029
1442577b 5030 handle_stripe(sh);
1da177e4
LT		 5031 release_stripe(sh);
5032
5033 return STRIPE_SECTORS;
5034}
5035
d1688a6d 5036static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio)
46031f9a
RBJ		 5037{
5038 /* We may not be able to submit a whole bio at once as there
5039 * may not be enough stripe_heads available.
5040 * We cannot pre-allocate enough stripe_heads as we may need
5041 * more than exist in the cache (if we allow ever large chunks).
5042 * So we do one stripe head at a time and record in
5043 * ->bi_hw_segments how many have been done.
5044 *
5045 * We *know* that this entire raid_bio is in one chunk, so
5046 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
5047 */
5048 struct stripe_head *sh;
911d4ee8 5049 int dd_idx;
46031f9a
RBJ		 5050 sector_t sector, logical_sector, last_sector;
5051 int scnt = 0;
5052 int remaining;
5053 int handled = 0;
5054
5055 logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
112bf897 5056 sector = raid5_compute_sector(conf, logical_sector,
911d4ee8 5057 0, &dd_idx, NULL);
f73a1c7d 5058 last_sector = bio_end_sector(raid_bio);
46031f9a
RBJ		 5059
5060 for (; logical_sector < last_sector;
387bb173
NB		 5061 logical_sector += STRIPE_SECTORS,
5062 sector += STRIPE_SECTORS,
5063 scnt++) {
46031f9a 5064
e7836bd6 5065 if (scnt < raid5_bi_processed_stripes(raid_bio))
46031f9a
RBJ		 5066 /* already done this stripe */
5067 continue;
5068
a8c906ca 5069 sh = get_active_stripe(conf, sector, 0, 1, 0);
46031f9a
RBJ		 5070
5071 if (!sh) {
5072 /* failed to get a stripe - must wait */
e7836bd6 5073 raid5_set_bi_processed_stripes(raid_bio, scnt);
46031f9a
RBJ		 5074 conf->retry_read_aligned = raid_bio;
5075 return handled;
5076 }
5077
387bb173
NB		 5078 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) {
5079 release_stripe(sh);
e7836bd6 5080 raid5_set_bi_processed_stripes(raid_bio, scnt);
387bb173
NB		 5081 conf->retry_read_aligned = raid_bio;
5082 return handled;
5083 }
5084
3f9e7c14 5085 set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags);
36d1c647 5086 handle_stripe(sh);
46031f9a
RBJ		 5087 release_stripe(sh);
5088 handled++;
5089 }
e7836bd6 5090 remaining = raid5_dec_bi_active_stripes(raid_bio);
0a82a8d1
LT		 5091 if (remaining == 0) {
5092 trace_block_bio_complete(bdev_get_queue(raid_bio->bi_bdev),
5093 raid_bio, 0);
0e13fe23 5094 bio_endio(raid_bio, 0);
0a82a8d1 5095 }
46031f9a
RBJ		 5096 if (atomic_dec_and_test(&conf->active_aligned_reads))
5097 wake_up(&conf->wait_for_stripe);
5098 return handled;
5099}
5100
bfc90cb0 5101static int handle_active_stripes(struct r5conf *conf, int group,
566c09c5
SL		 5102 struct r5worker *worker,
5103 struct list_head *temp_inactive_list)
46a06401
SL		 5104{
5105 struct stripe_head *batch[MAX_STRIPE_BATCH], *sh;
566c09c5
SL		 5106 int i, batch_size = 0, hash;
5107 bool release_inactive = false;
46a06401
SL		 5108
5109 while (batch_size < MAX_STRIPE_BATCH &&
851c30c9 5110 (sh = __get_priority_stripe(conf, group)) != NULL)
46a06401
SL		 5111 batch[batch_size++] = sh;
5112
566c09c5
SL		 5113 if (batch_size == 0) {
5114 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5115 if (!list_empty(temp_inactive_list + i))
5116 break;
5117 if (i == NR_STRIPE_HASH_LOCKS)
5118 return batch_size;
5119 release_inactive = true;
5120 }
46a06401
SL		 5121 spin_unlock_irq(&conf->device_lock);
5122
566c09c5
SL		 5123 release_inactive_stripe_list(conf, temp_inactive_list,
5124 NR_STRIPE_HASH_LOCKS);
5125
5126 if (release_inactive) {
5127 spin_lock_irq(&conf->device_lock);
5128 return 0;
5129 }
5130
46a06401
SL		 5131 for (i = 0; i < batch_size; i++)
5132 handle_stripe(batch[i]);
5133
5134 cond_resched();
5135
5136 spin_lock_irq(&conf->device_lock);
566c09c5
SL		 5137 for (i = 0; i < batch_size; i++) {
5138 hash = batch[i]->hash_lock_index;
5139 __release_stripe(conf, batch[i], &temp_inactive_list[hash]);
5140 }
46a06401
SL		 5141 return batch_size;
5142}
46031f9a 5143
851c30c9
SL		 5144static void raid5_do_work(struct work_struct *work)
5145{
5146 struct r5worker *worker = container_of(work, struct r5worker, work);
5147 struct r5worker_group *group = worker->group;
5148 struct r5conf *conf = group->conf;
5149 int group_id = group - conf->worker_groups;
5150 int handled;
5151 struct blk_plug plug;
5152
5153 pr_debug("+++ raid5worker active\n");
5154
5155 blk_start_plug(&plug);
5156 handled = 0;
5157 spin_lock_irq(&conf->device_lock);
5158 while (1) {
5159 int batch_size, released;
5160
566c09c5 5161 released = release_stripe_list(conf, worker->temp_inactive_list);
851c30c9 5162
566c09c5
SL		 5163 batch_size = handle_active_stripes(conf, group_id, worker,
5164 worker->temp_inactive_list);
bfc90cb0 5165 worker->working = false;
851c30c9
SL		 5166 if (!batch_size && !released)
5167 break;
5168 handled += batch_size;
5169 }
5170 pr_debug("%d stripes handled\n", handled);
5171
5172 spin_unlock_irq(&conf->device_lock);
5173 blk_finish_plug(&plug);
5174
5175 pr_debug("--- raid5worker inactive\n");
5176}
5177
1da177e4
LT		 5178/*
5179 * This is our raid5 kernel thread.
5180 *
5181 * We scan the hash table for stripes which can be handled now.
5182 * During the scan, completed stripes are saved for us by the interrupt
5183 * handler, so that they will not have to wait for our next wakeup.
5184 */
4ed8731d 5185static void raid5d(struct md_thread *thread)
1da177e4 5186{
4ed8731d 5187 struct mddev *mddev = thread->mddev;
d1688a6d 5188 struct r5conf *conf = mddev->private;
1da177e4 5189 int handled;
e1dfa0a2 5190 struct blk_plug plug;
1da177e4 5191
45b4233c 5192 pr_debug("+++ raid5d active\n");
1da177e4
LT		 5193
5194 md_check_recovery(mddev);
1da177e4 5195
e1dfa0a2 5196 blk_start_plug(&plug);
1da177e4
LT		 5197 handled = 0;
5198 spin_lock_irq(&conf->device_lock);
5199 while (1) {
46031f9a 5200 struct bio *bio;
773ca82f
SL		 5201 int batch_size, released;
5202
566c09c5 5203 released = release_stripe_list(conf, conf->temp_inactive_list);
1da177e4 5204
0021b7bc 5205 if (
7c13edc8
N		 5206 !list_empty(&conf->bitmap_list)) {
5207 /* Now is a good time to flush some bitmap updates */
5208 conf->seq_flush++;
700e432d 5209 spin_unlock_irq(&conf->device_lock);
72626685 5210 bitmap_unplug(mddev->bitmap);
700e432d 5211 spin_lock_irq(&conf->device_lock);
7c13edc8 5212 conf->seq_write = conf->seq_flush;
566c09c5 5213 activate_bit_delay(conf, conf->temp_inactive_list);
72626685 5214 }
0021b7bc 5215 raid5_activate_delayed(conf);
72626685 5216
46031f9a
RBJ		 5217 while ((bio = remove_bio_from_retry(conf))) {
5218 int ok;
5219 spin_unlock_irq(&conf->device_lock);
5220 ok = retry_aligned_read(conf, bio);
5221 spin_lock_irq(&conf->device_lock);
5222 if (!ok)
5223 break;
5224 handled++;
5225 }
5226
566c09c5
SL		 5227 batch_size = handle_active_stripes(conf, ANY_GROUP, NULL,
5228 conf->temp_inactive_list);
773ca82f 5229 if (!batch_size && !released)
1da177e4 5230 break;
46a06401 5231 handled += batch_size;
1da177e4 5232
46a06401
SL		 5233 if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) {
5234 spin_unlock_irq(&conf->device_lock);
de393cde 5235 md_check_recovery(mddev);
46a06401
SL		 5236 spin_lock_irq(&conf->device_lock);
5237 }
1da177e4 5238 }
45b4233c 5239 pr_debug("%d stripes handled\n", handled);
1da177e4
LT		 5240
5241 spin_unlock_irq(&conf->device_lock);
5242
c9f21aaf 5243 async_tx_issue_pending_all();
e1dfa0a2 5244 blk_finish_plug(&plug);
1da177e4 5245
45b4233c 5246 pr_debug("--- raid5d inactive\n");
1da177e4
LT		 5247}
5248
3f294f4f 5249static ssize_t
fd01b88c 5250raid5_show_stripe_cache_size(struct mddev *mddev, char *page)
3f294f4f 5251{
d1688a6d 5252 struct r5conf *conf = mddev->private;
96de1e66
N		 5253 if (conf)
5254 return sprintf(page, "%d\n", conf->max_nr_stripes);
5255 else
5256 return 0;
3f294f4f
N		 5257}
5258
c41d4ac4 5259int
fd01b88c 5260raid5_set_cache_size(struct mddev *mddev, int size)
3f294f4f 5261{
d1688a6d 5262 struct r5conf *conf = mddev->private;
b5470dc5 5263 int err;
566c09c5 5264 int hash;
b5470dc5 5265
c41d4ac4 5266 if (size <= 16 || size > 32768)
3f294f4f 5267 return -EINVAL;
566c09c5 5268 hash = (conf->max_nr_stripes - 1) % NR_STRIPE_HASH_LOCKS;
c41d4ac4 5269 while (size < conf->max_nr_stripes) {
566c09c5 5270 if (drop_one_stripe(conf, hash))
3f294f4f
N		 5271 conf->max_nr_stripes--;
5272 else
5273 break;
566c09c5
SL		 5274 hash--;
5275 if (hash < 0)
5276 hash = NR_STRIPE_HASH_LOCKS - 1;
3f294f4f 5277 }
b5470dc5
DW		 5278 err = md_allow_write(mddev);
5279 if (err)
5280 return err;
566c09c5 5281 hash = conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS;
c41d4ac4 5282 while (size > conf->max_nr_stripes) {
566c09c5 5283 if (grow_one_stripe(conf, hash))
3f294f4f
N		 5284 conf->max_nr_stripes++;
5285 else break;
566c09c5 5286 hash = (hash + 1) % NR_STRIPE_HASH_LOCKS;
3f294f4f 5287 }
c41d4ac4
N		 5288 return 0;
5289}
5290EXPORT_SYMBOL(raid5_set_cache_size);
5291
5292static ssize_t
fd01b88c 5293raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len)
c41d4ac4 5294{
d1688a6d 5295 struct r5conf *conf = mddev->private;
c41d4ac4
N		 5296 unsigned long new;
5297 int err;
5298
5299 if (len >= PAGE_SIZE)
5300 return -EINVAL;
5301 if (!conf)
5302 return -ENODEV;
5303
b29bebd6 5304 if (kstrtoul(page, 10, &new))
c41d4ac4
N		 5305 return -EINVAL;
5306 err = raid5_set_cache_size(mddev, new);
5307 if (err)
5308 return err;
3f294f4f
N		 5309 return len;
5310}
007583c9 5311
96de1e66
N		 5312static struct md_sysfs_entry
5313raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
5314 raid5_show_stripe_cache_size,
5315 raid5_store_stripe_cache_size);
3f294f4f 5316
8b3e6cdc 5317static ssize_t
fd01b88c 5318raid5_show_preread_threshold(struct mddev *mddev, char *page)
8b3e6cdc 5319{
d1688a6d 5320 struct r5conf *conf = mddev->private;
8b3e6cdc
DW		 5321 if (conf)
5322 return sprintf(page, "%d\n", conf->bypass_threshold);
5323 else
5324 return 0;
5325}
5326
5327static ssize_t
fd01b88c 5328raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len)
8b3e6cdc 5329{
d1688a6d 5330 struct r5conf *conf = mddev->private;
4ef197d8 5331 unsigned long new;
8b3e6cdc
DW		 5332 if (len >= PAGE_SIZE)
5333 return -EINVAL;
5334 if (!conf)
5335 return -ENODEV;
5336
b29bebd6 5337 if (kstrtoul(page, 10, &new))
8b3e6cdc 5338 return -EINVAL;
4ef197d8 5339 if (new > conf->max_nr_stripes)
8b3e6cdc
DW		 5340 return -EINVAL;
5341 conf->bypass_threshold = new;
5342 return len;
5343}
5344
5345static struct md_sysfs_entry
5346raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
5347 S_IRUGO | S_IWUSR,
5348 raid5_show_preread_threshold,
5349 raid5_store_preread_threshold);
5350
3f294f4f 5351static ssize_t
fd01b88c 5352stripe_cache_active_show(struct mddev *mddev, char *page)
3f294f4f 5353{
d1688a6d 5354 struct r5conf *conf = mddev->private;
96de1e66
N		 5355 if (conf)
5356 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
5357 else
5358 return 0;
3f294f4f
N		 5359}
5360
96de1e66
N		 5361static struct md_sysfs_entry
5362raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
3f294f4f 5363
b721420e
SL		 5364static ssize_t
5365raid5_show_group_thread_cnt(struct mddev *mddev, char *page)
5366{
5367 struct r5conf *conf = mddev->private;
5368 if (conf)
5369 return sprintf(page, "%d\n", conf->worker_cnt_per_group);
5370 else
5371 return 0;
5372}
5373
60aaf933 5374static int alloc_thread_groups(struct r5conf *conf, int cnt,
5375 int *group_cnt,
5376 int *worker_cnt_per_group,
5377 struct r5worker_group **worker_groups);
b721420e
SL		 5378static ssize_t
5379raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len)
5380{
5381 struct r5conf *conf = mddev->private;
5382 unsigned long new;
5383 int err;
60aaf933 5384 struct r5worker_group *new_groups, *old_groups;
5385 int group_cnt, worker_cnt_per_group;
b721420e
SL		 5386
5387 if (len >= PAGE_SIZE)
5388 return -EINVAL;
5389 if (!conf)
5390 return -ENODEV;
5391
5392 if (kstrtoul(page, 10, &new))
5393 return -EINVAL;
5394
5395 if (new == conf->worker_cnt_per_group)
5396 return len;
5397
5398 mddev_suspend(mddev);
5399
5400 old_groups = conf->worker_groups;
d206dcfa 5401 if (old_groups)
5402 flush_workqueue(raid5_wq);
5403
60aaf933 5404 err = alloc_thread_groups(conf, new,
5405 &group_cnt, &worker_cnt_per_group,
5406 &new_groups);
5407 if (!err) {
5408 spin_lock_irq(&conf->device_lock);
5409 conf->group_cnt = group_cnt;
5410 conf->worker_cnt_per_group = worker_cnt_per_group;
5411 conf->worker_groups = new_groups;
5412 spin_unlock_irq(&conf->device_lock);
b721420e 5413
b721420e
SL		 5414 if (old_groups)
5415 kfree(old_groups[0].workers);
5416 kfree(old_groups);
5417 }
5418
5419 mddev_resume(mddev);
5420
5421 if (err)
5422 return err;
5423 return len;
5424}
5425
5426static struct md_sysfs_entry
5427raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR,
5428 raid5_show_group_thread_cnt,
5429 raid5_store_group_thread_cnt);
5430
007583c9 5431static struct attribute *raid5_attrs[] = {
3f294f4f
N		 5432 &raid5_stripecache_size.attr,
5433 &raid5_stripecache_active.attr,
8b3e6cdc 5434 &raid5_preread_bypass_threshold.attr,
b721420e 5435 &raid5_group_thread_cnt.attr,
3f294f4f
N		 5436 NULL,
5437};
007583c9
N		 5438static struct attribute_group raid5_attrs_group = {
5439 .name = NULL,
5440 .attrs = raid5_attrs,
3f294f4f
N		 5441};
5442
60aaf933 5443static int alloc_thread_groups(struct r5conf *conf, int cnt,
5444 int *group_cnt,
5445 int *worker_cnt_per_group,
5446 struct r5worker_group **worker_groups)
851c30c9 5447{
566c09c5 5448 int i, j, k;
851c30c9
SL		 5449 ssize_t size;
5450 struct r5worker *workers;
5451
60aaf933 5452 *worker_cnt_per_group = cnt;
851c30c9 5453 if (cnt == 0) {
60aaf933 5454 *group_cnt = 0;
5455 *worker_groups = NULL;
851c30c9
SL		 5456 return 0;
5457 }
60aaf933 5458 *group_cnt = num_possible_nodes();
851c30c9 5459 size = sizeof(struct r5worker) * cnt;
60aaf933 5460 workers = kzalloc(size * *group_cnt, GFP_NOIO);
5461 *worker_groups = kzalloc(sizeof(struct r5worker_group) *
5462 *group_cnt, GFP_NOIO);
5463 if (!*worker_groups || !workers) {
851c30c9 5464 kfree(workers);
60aaf933 5465 kfree(*worker_groups);
851c30c9
SL		 5466 return -ENOMEM;
5467 }
5468
60aaf933 5469 for (i = 0; i < *group_cnt; i++) {
851c30c9
SL		 5470 struct r5worker_group *group;
5471
0c775d52 5472 group = &(*worker_groups)[i];
851c30c9
SL		 5473 INIT_LIST_HEAD(&group->handle_list);
5474 group->conf = conf;
5475 group->workers = workers + i * cnt;
5476
5477 for (j = 0; j < cnt; j++) {
566c09c5
SL		 5478 struct r5worker *worker = group->workers + j;
5479 worker->group = group;
5480 INIT_WORK(&worker->work, raid5_do_work);
5481
5482 for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++)
5483 INIT_LIST_HEAD(worker->temp_inactive_list + k);
851c30c9
SL		 5484 }
5485 }
5486
5487 return 0;
5488}
5489
5490static void free_thread_groups(struct r5conf *conf)
5491{
5492 if (conf->worker_groups)
5493 kfree(conf->worker_groups[0].workers);
5494 kfree(conf->worker_groups);
5495 conf->worker_groups = NULL;
5496}
5497
80c3a6ce 5498static sector_t
fd01b88c 5499raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks)
80c3a6ce 5500{
d1688a6d 5501 struct r5conf *conf = mddev->private;
80c3a6ce
DW		 5502
5503 if (!sectors)
5504 sectors = mddev->dev_sectors;
5e5e3e78 5505 if (!raid_disks)
7ec05478 5506 /* size is defined by the smallest of previous and new size */
5e5e3e78 5507 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
80c3a6ce 5508
9d8f0363 5509 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
664e7c41 5510 sectors &= ~((sector_t)mddev->new_chunk_sectors - 1);
80c3a6ce
DW		 5511 return sectors * (raid_disks - conf->max_degraded);
5512}
5513
d1688a6d 5514static void raid5_free_percpu(struct r5conf *conf)
36d1c647
DW		 5515{
5516 struct raid5_percpu *percpu;
5517 unsigned long cpu;
5518
5519 if (!conf->percpu)
5520 return;
5521
5522 get_online_cpus();
5523 for_each_possible_cpu(cpu) {
5524 percpu = per_cpu_ptr(conf->percpu, cpu);
5525 safe_put_page(percpu->spare_page);
d6f38f31 5526 kfree(percpu->scribble);
36d1c647
DW		 5527 }
5528#ifdef CONFIG_HOTPLUG_CPU
5529 unregister_cpu_notifier(&conf->cpu_notify);
5530#endif
5531 put_online_cpus();
5532
5533 free_percpu(conf->percpu);
5534}
5535
d1688a6d 5536static void free_conf(struct r5conf *conf)
95fc17aa 5537{
851c30c9 5538 free_thread_groups(conf);
95fc17aa 5539 shrink_stripes(conf);
36d1c647 5540 raid5_free_percpu(conf);
95fc17aa
DW		 5541 kfree(conf->disks);
5542 kfree(conf->stripe_hashtbl);
5543 kfree(conf);
5544}
5545
36d1c647
DW		 5546#ifdef CONFIG_HOTPLUG_CPU
5547static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action,
5548 void *hcpu)
5549{
d1688a6d 5550 struct r5conf *conf = container_of(nfb, struct r5conf, cpu_notify);
36d1c647
DW		 5551 long cpu = (long)hcpu;
5552 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
5553
5554 switch (action) {
5555 case CPU_UP_PREPARE:
5556 case CPU_UP_PREPARE_FROZEN:
d6f38f31 5557 if (conf->level == 6 && !percpu->spare_page)
36d1c647 5558 percpu->spare_page = alloc_page(GFP_KERNEL);
d6f38f31
DW		 5559 if (!percpu->scribble)
5560 percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
5561
5562 if (!percpu->scribble ||
5563 (conf->level == 6 && !percpu->spare_page)) {
5564 safe_put_page(percpu->spare_page);
5565 kfree(percpu->scribble);
36d1c647
DW		 5566 pr_err("%s: failed memory allocation for cpu%ld\n",
5567 __func__, cpu);
55af6bb5 5568 return notifier_from_errno(-ENOMEM);
36d1c647
DW		 5569 }
5570 break;
5571 case CPU_DEAD:
5572 case CPU_DEAD_FROZEN:
5573 safe_put_page(percpu->spare_page);
d6f38f31 5574 kfree(percpu->scribble);
36d1c647 5575 percpu->spare_page = NULL;
d6f38f31 5576 percpu->scribble = NULL;
36d1c647
DW		 5577 break;
5578 default:
5579 break;
5580 }
5581 return NOTIFY_OK;
5582}
5583#endif
5584
d1688a6d 5585static int raid5_alloc_percpu(struct r5conf *conf)
36d1c647
DW		 5586{
5587 unsigned long cpu;
5588 struct page *spare_page;
a29d8b8e 5589 struct raid5_percpu __percpu *allcpus;
d6f38f31 5590 void *scribble;
36d1c647
DW		 5591 int err;
5592
36d1c647
DW		 5593 allcpus = alloc_percpu(struct raid5_percpu);
5594 if (!allcpus)
5595 return -ENOMEM;
5596 conf->percpu = allcpus;
5597
5598 get_online_cpus();
5599 err = 0;
5600 for_each_present_cpu(cpu) {
d6f38f31
DW		 5601 if (conf->level == 6) {
5602 spare_page = alloc_page(GFP_KERNEL);
5603 if (!spare_page) {
5604 err = -ENOMEM;
5605 break;
5606 }
5607 per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page;
5608 }
5e5e3e78 5609 scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
d6f38f31 5610 if (!scribble) {
36d1c647
DW		 5611 err = -ENOMEM;
5612 break;
5613 }
d6f38f31 5614 per_cpu_ptr(conf->percpu, cpu)->scribble = scribble;
36d1c647
DW		 5615 }
5616#ifdef CONFIG_HOTPLUG_CPU
5617 conf->cpu_notify.notifier_call = raid456_cpu_notify;
5618 conf->cpu_notify.priority = 0;
5619 if (err == 0)
5620 err = register_cpu_notifier(&conf->cpu_notify);
5621#endif
5622 put_online_cpus();
5623
5624 return err;
5625}
5626
d1688a6d 5627static struct r5conf *setup_conf(struct mddev *mddev)
1da177e4 5628{
d1688a6d 5629 struct r5conf *conf;
5e5e3e78 5630 int raid_disk, memory, max_disks;
3cb03002 5631 struct md_rdev *rdev;
1da177e4 5632 struct disk_info *disk;
0232605d 5633 char pers_name[6];
566c09c5 5634 int i;
60aaf933 5635 int group_cnt, worker_cnt_per_group;
5636 struct r5worker_group *new_group;
1da177e4 5637
91adb564
N		 5638 if (mddev->new_level != 5
5639 && mddev->new_level != 4
5640 && mddev->new_level != 6) {
0c55e022 5641 printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n",
91adb564
N		 5642 mdname(mddev), mddev->new_level);
5643 return ERR_PTR(-EIO);
1da177e4 5644 }
91adb564
N		 5645 if ((mddev->new_level == 5
5646 && !algorithm_valid_raid5(mddev->new_layout)) ||
5647 (mddev->new_level == 6
5648 && !algorithm_valid_raid6(mddev->new_layout))) {
0c55e022 5649 printk(KERN_ERR "md/raid:%s: layout %d not supported\n",
91adb564
N		 5650 mdname(mddev), mddev->new_layout);
5651 return ERR_PTR(-EIO);
99c0fb5f 5652 }
91adb564 5653 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
0c55e022 5654 printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n",
91adb564
N		 5655 mdname(mddev), mddev->raid_disks);
5656 return ERR_PTR(-EINVAL);
4bbf3771
N		 5657 }
5658
664e7c41
AN		 5659 if (!mddev->new_chunk_sectors ||
5660 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
5661 !is_power_of_2(mddev->new_chunk_sectors)) {
0c55e022
N		 5662 printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n",
5663 mdname(mddev), mddev->new_chunk_sectors << 9);
91adb564 5664 return ERR_PTR(-EINVAL);
f6705578
N		 5665 }
5666
d1688a6d 5667 conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL);
91adb564 5668 if (conf == NULL)
1da177e4 5669 goto abort;
851c30c9 5670 /* Don't enable multi-threading by default*/
60aaf933 5671 if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group,
5672 &new_group)) {
5673 conf->group_cnt = group_cnt;
5674 conf->worker_cnt_per_group = worker_cnt_per_group;
5675 conf->worker_groups = new_group;
5676 } else
851c30c9 5677 goto abort;
f5efd45a 5678 spin_lock_init(&conf->device_lock);
c46501b2 5679 seqcount_init(&conf->gen_lock);
f5efd45a
DW		 5680 init_waitqueue_head(&conf->wait_for_stripe);
5681 init_waitqueue_head(&conf->wait_for_overlap);
5682 INIT_LIST_HEAD(&conf->handle_list);
5683 INIT_LIST_HEAD(&conf->hold_list);
5684 INIT_LIST_HEAD(&conf->delayed_list);
5685 INIT_LIST_HEAD(&conf->bitmap_list);
773ca82f 5686 init_llist_head(&conf->released_stripes);
f5efd45a
DW		 5687 atomic_set(&conf->active_stripes, 0);
5688 atomic_set(&conf->preread_active_stripes, 0);
5689 atomic_set(&conf->active_aligned_reads, 0);
5690 conf->bypass_threshold = BYPASS_THRESHOLD;
d890fa2b 5691 conf->recovery_disabled = mddev->recovery_disabled - 1;
91adb564
N		 5692
5693 conf->raid_disks = mddev->raid_disks;
5694 if (mddev->reshape_position == MaxSector)
5695 conf->previous_raid_disks = mddev->raid_disks;
5696 else
f6705578 5697 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
5e5e3e78
N		 5698 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
5699 conf->scribble_len = scribble_len(max_disks);
f6705578 5700
5e5e3e78 5701 conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
b55e6bfc
N		 5702 GFP_KERNEL);
5703 if (!conf->disks)
5704 goto abort;
9ffae0cf 5705
1da177e4
LT		 5706 conf->mddev = mddev;
5707
fccddba0 5708 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
1da177e4 5709 goto abort;
1da177e4 5710
566c09c5
SL		 5711 /* We init hash_locks[0] separately to that it can be used
5712 * as the reference lock in the spin_lock_nest_lock() call
5713 * in lock_all_device_hash_locks_irq in order to convince
5714 * lockdep that we know what we are doing.
5715 */
5716 spin_lock_init(conf->hash_locks);
5717 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
5718 spin_lock_init(conf->hash_locks + i);
5719
5720 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5721 INIT_LIST_HEAD(conf->inactive_list + i);
5722
5723 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5724 INIT_LIST_HEAD(conf->temp_inactive_list + i);
5725
36d1c647
DW		 5726 conf->level = mddev->new_level;
5727 if (raid5_alloc_percpu(conf) != 0)
5728 goto abort;
5729
0c55e022 5730 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
1da177e4 5731
dafb20fa 5732 rdev_for_each(rdev, mddev) {
1da177e4 5733 raid_disk = rdev->raid_disk;
5e5e3e78 5734 if (raid_disk >= max_disks
1da177e4
LT		 5735 || raid_disk < 0)
5736 continue;
5737 disk = conf->disks + raid_disk;
5738
17045f52
N		 5739 if (test_bit(Replacement, &rdev->flags)) {
5740 if (disk->replacement)
5741 goto abort;
5742 disk->replacement = rdev;
5743 } else {
5744 if (disk->rdev)
5745 goto abort;
5746 disk->rdev = rdev;
5747 }
1da177e4 5748
b2d444d7 5749 if (test_bit(In_sync, &rdev->flags)) {
1da177e4 5750 char b[BDEVNAME_SIZE];
0c55e022
N		 5751 printk(KERN_INFO "md/raid:%s: device %s operational as raid"
5752 " disk %d\n",
5753 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
d6b212f4 5754 } else if (rdev->saved_raid_disk != raid_disk)
8c2e870a
NB		 5755 /* Cannot rely on bitmap to complete recovery */
5756 conf->fullsync = 1;
1da177e4
LT		 5757 }
5758
09c9e5fa 5759 conf->chunk_sectors = mddev->new_chunk_sectors;
91adb564 5760 conf->level = mddev->new_level;
16a53ecc
N		 5761 if (conf->level == 6)
5762 conf->max_degraded = 2;
5763 else
5764 conf->max_degraded = 1;
91adb564 5765 conf->algorithm = mddev->new_layout;
fef9c61f 5766 conf->reshape_progress = mddev->reshape_position;
e183eaed 5767 if (conf->reshape_progress != MaxSector) {
09c9e5fa 5768 conf->prev_chunk_sectors = mddev->chunk_sectors;
e183eaed
N		 5769 conf->prev_algo = mddev->layout;
5770 }
1da177e4 5771
91adb564 5772 memory = conf->max_nr_stripes * (sizeof(struct stripe_head) +
5e5e3e78 5773 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
4bda556a 5774 atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS);
566c09c5 5775 if (grow_stripes(conf, NR_STRIPES)) {
91adb564 5776 printk(KERN_ERR
0c55e022
N		 5777 "md/raid:%s: couldn't allocate %dkB for buffers\n",
5778 mdname(mddev), memory);
91adb564
N		 5779 goto abort;
5780 } else
0c55e022
N		 5781 printk(KERN_INFO "md/raid:%s: allocated %dkB\n",
5782 mdname(mddev), memory);
1da177e4 5783
0232605d
N		 5784 sprintf(pers_name, "raid%d", mddev->new_level);
5785 conf->thread = md_register_thread(raid5d, mddev, pers_name);
91adb564
N		 5786 if (!conf->thread) {
5787 printk(KERN_ERR
0c55e022 5788 "md/raid:%s: couldn't allocate thread.\n",
91adb564 5789 mdname(mddev));
16a53ecc
N		 5790 goto abort;
5791 }
91adb564
N		 5792
5793 return conf;
5794
5795 abort:
5796 if (conf) {
95fc17aa 5797 free_conf(conf);
91adb564
N		 5798 return ERR_PTR(-EIO);
5799 } else
5800 return ERR_PTR(-ENOMEM);
5801}
5802
c148ffdc
N		 5803
5804static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
5805{
5806 switch (algo) {
5807 case ALGORITHM_PARITY_0:
5808 if (raid_disk < max_degraded)
5809 return 1;
5810 break;
5811 case ALGORITHM_PARITY_N:
5812 if (raid_disk >= raid_disks - max_degraded)
5813 return 1;
5814 break;
5815 case ALGORITHM_PARITY_0_6:
5816 if (raid_disk == 0 ||
5817 raid_disk == raid_disks - 1)
5818 return 1;
5819 break;
5820 case ALGORITHM_LEFT_ASYMMETRIC_6:
5821 case ALGORITHM_RIGHT_ASYMMETRIC_6:
5822 case ALGORITHM_LEFT_SYMMETRIC_6:
5823 case ALGORITHM_RIGHT_SYMMETRIC_6:
5824 if (raid_disk == raid_disks - 1)
5825 return 1;
5826 }
5827 return 0;
5828}
5829
fd01b88c 5830static int run(struct mddev *mddev)
91adb564 5831{
d1688a6d 5832 struct r5conf *conf;
9f7c2220 5833 int working_disks = 0;
c148ffdc 5834 int dirty_parity_disks = 0;
3cb03002 5835 struct md_rdev *rdev;
c148ffdc 5836 sector_t reshape_offset = 0;
17045f52 5837 int i;
b5254dd5
N		 5838 long long min_offset_diff = 0;
5839 int first = 1;
91adb564 5840
8c6ac868 5841 if (mddev->recovery_cp != MaxSector)
0c55e022 5842 printk(KERN_NOTICE "md/raid:%s: not clean"
8c6ac868
AN		 5843 " -- starting background reconstruction\n",
5844 mdname(mddev));
b5254dd5
N		 5845
5846 rdev_for_each(rdev, mddev) {
5847 long long diff;
5848 if (rdev->raid_disk < 0)
5849 continue;
5850 diff = (rdev->new_data_offset - rdev->data_offset);
5851 if (first) {
5852 min_offset_diff = diff;
5853 first = 0;
5854 } else if (mddev->reshape_backwards &&
5855 diff < min_offset_diff)
5856 min_offset_diff = diff;
5857 else if (!mddev->reshape_backwards &&
5858 diff > min_offset_diff)
5859 min_offset_diff = diff;
5860 }
5861
91adb564
N		 5862 if (mddev->reshape_position != MaxSector) {
5863 /* Check that we can continue the reshape.
b5254dd5
N		 5864 * Difficulties arise if the stripe we would write to
5865 * next is at or after the stripe we would read from next.
5866 * For a reshape that changes the number of devices, this
5867 * is only possible for a very short time, and mdadm makes
5868 * sure that time appears to have past before assembling
5869 * the array. So we fail if that time hasn't passed.
5870 * For a reshape that keeps the number of devices the same
5871 * mdadm must be monitoring the reshape can keeping the
5872 * critical areas read-only and backed up. It will start
5873 * the array in read-only mode, so we check for that.
91adb564
N		 5874 */
5875 sector_t here_new, here_old;
5876 int old_disks;
18b00334 5877 int max_degraded = (mddev->level == 6 ? 2 : 1);
91adb564 5878
88ce4930 5879 if (mddev->new_level != mddev->level) {
0c55e022 5880 printk(KERN_ERR "md/raid:%s: unsupported reshape "
91adb564
N		 5881 "required - aborting.\n",
5882 mdname(mddev));
5883 return -EINVAL;
5884 }
91adb564
N		 5885 old_disks = mddev->raid_disks - mddev->delta_disks;
5886 /* reshape_position must be on a new-stripe boundary, and one
5887 * further up in new geometry must map after here in old
5888 * geometry.
5889 */
5890 here_new = mddev->reshape_position;
664e7c41 5891 if (sector_div(here_new, mddev->new_chunk_sectors *
91adb564 5892 (mddev->raid_disks - max_degraded))) {
0c55e022
N		 5893 printk(KERN_ERR "md/raid:%s: reshape_position not "
5894 "on a stripe boundary\n", mdname(mddev));
91adb564
N		 5895 return -EINVAL;
5896 }
c148ffdc 5897 reshape_offset = here_new * mddev->new_chunk_sectors;
91adb564
N		 5898 /* here_new is the stripe we will write to */
5899 here_old = mddev->reshape_position;
9d8f0363 5900 sector_div(here_old, mddev->chunk_sectors *
91adb564
N		 5901 (old_disks-max_degraded));
5902 /* here_old is the first stripe that we might need to read
5903 * from */
67ac6011 5904 if (mddev->delta_disks == 0) {
b5254dd5
N		 5905 if ((here_new * mddev->new_chunk_sectors !=
5906 here_old * mddev->chunk_sectors)) {
5907 printk(KERN_ERR "md/raid:%s: reshape position is"
5908 " confused - aborting\n", mdname(mddev));
5909 return -EINVAL;
5910 }
67ac6011 5911 /* We cannot be sure it is safe to start an in-place
b5254dd5 5912 * reshape. It is only safe if user-space is monitoring
67ac6011
N		 5913 * and taking constant backups.
5914 * mdadm always starts a situation like this in
5915 * readonly mode so it can take control before
5916 * allowing any writes. So just check for that.
5917 */
b5254dd5
N		 5918 if (abs(min_offset_diff) >= mddev->chunk_sectors &&
5919 abs(min_offset_diff) >= mddev->new_chunk_sectors)
5920 /* not really in-place - so OK */;
5921 else if (mddev->ro == 0) {
5922 printk(KERN_ERR "md/raid:%s: in-place reshape "
5923 "must be started in read-only mode "
5924 "- aborting\n",
0c55e022 5925 mdname(mddev));
67ac6011
N		 5926 return -EINVAL;
5927 }
2c810cdd 5928 } else if (mddev->reshape_backwards
b5254dd5 5929 ? (here_new * mddev->new_chunk_sectors + min_offset_diff <=
67ac6011
N		 5930 here_old * mddev->chunk_sectors)
5931 : (here_new * mddev->new_chunk_sectors >=
b5254dd5 5932 here_old * mddev->chunk_sectors + (-min_offset_diff))) {
91adb564 5933 /* Reading from the same stripe as writing to - bad */
0c55e022
N		 5934 printk(KERN_ERR "md/raid:%s: reshape_position too early for "
5935 "auto-recovery - aborting.\n",
5936 mdname(mddev));
91adb564
N		 5937 return -EINVAL;
5938 }
0c55e022
N		 5939 printk(KERN_INFO "md/raid:%s: reshape will continue\n",
5940 mdname(mddev));
91adb564
N		 5941 /* OK, we should be able to continue; */
5942 } else {
5943 BUG_ON(mddev->level != mddev->new_level);
5944 BUG_ON(mddev->layout != mddev->new_layout);
664e7c41 5945 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
91adb564 5946 BUG_ON(mddev->delta_disks != 0);
1da177e4 5947 }
91adb564 5948
245f46c2
N		 5949 if (mddev->private == NULL)
5950 conf = setup_conf(mddev);
5951 else
5952 conf = mddev->private;
5953
91adb564
N		 5954 if (IS_ERR(conf))
5955 return PTR_ERR(conf);
5956
b5254dd5 5957 conf->min_offset_diff = min_offset_diff;
91adb564
N		 5958 mddev->thread = conf->thread;
5959 conf->thread = NULL;
5960 mddev->private = conf;
5961
17045f52
N		 5962 for (i = 0; i < conf->raid_disks && conf->previous_raid_disks;
5963 i++) {
5964 rdev = conf->disks[i].rdev;
5965 if (!rdev && conf->disks[i].replacement) {
5966 /* The replacement is all we have yet */
5967 rdev = conf->disks[i].replacement;
5968 conf->disks[i].replacement = NULL;
5969 clear_bit(Replacement, &rdev->flags);
5970 conf->disks[i].rdev = rdev;
5971 }
5972 if (!rdev)
c148ffdc 5973 continue;
17045f52
N		 5974 if (conf->disks[i].replacement &&
5975 conf->reshape_progress != MaxSector) {
5976 /* replacements and reshape simply do not mix. */
5977 printk(KERN_ERR "md: cannot handle concurrent "
5978 "replacement and reshape.\n");
5979 goto abort;
5980 }
2f115882 5981 if (test_bit(In_sync, &rdev->flags)) {
91adb564 5982 working_disks++;
2f115882
N		 5983 continue;
5984 }
c148ffdc
N		 5985 /* This disc is not fully in-sync. However if it
5986 * just stored parity (beyond the recovery_offset),
5987 * when we don't need to be concerned about the
5988 * array being dirty.
5989 * When reshape goes 'backwards', we never have
5990 * partially completed devices, so we only need
5991 * to worry about reshape going forwards.
5992 */
5993 /* Hack because v0.91 doesn't store recovery_offset properly. */
5994 if (mddev->major_version == 0 &&
5995 mddev->minor_version > 90)
5996 rdev->recovery_offset = reshape_offset;
5026d7a9 5997
c148ffdc
N		 5998 if (rdev->recovery_offset < reshape_offset) {
5999 /* We need to check old and new layout */
6000 if (!only_parity(rdev->raid_disk,
6001 conf->algorithm,
6002 conf->raid_disks,
6003 conf->max_degraded))
6004 continue;
6005 }
6006 if (!only_parity(rdev->raid_disk,
6007 conf->prev_algo,
6008 conf->previous_raid_disks,
6009 conf->max_degraded))
6010 continue;
6011 dirty_parity_disks++;
6012 }
91adb564 6013
17045f52
N		 6014 /*
6015 * 0 for a fully functional array, 1 or 2 for a degraded array.
6016 */
908f4fbd 6017 mddev->degraded = calc_degraded(conf);
91adb564 6018
674806d6 6019 if (has_failed(conf)) {
0c55e022 6020 printk(KERN_ERR "md/raid:%s: not enough operational devices"
1da177e4 6021 " (%d/%d failed)\n",
02c2de8c 6022 mdname(mddev), mddev->degraded, conf->raid_disks);
1da177e4
LT		 6023 goto abort;
6024 }
6025
91adb564 6026 /* device size must be a multiple of chunk size */
9d8f0363 6027 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
91adb564
N		 6028 mddev->resync_max_sectors = mddev->dev_sectors;
6029
c148ffdc 6030 if (mddev->degraded > dirty_parity_disks &&
1da177e4 6031 mddev->recovery_cp != MaxSector) {
6ff8d8ec
N		 6032 if (mddev->ok_start_degraded)
6033 printk(KERN_WARNING
0c55e022
N		 6034 "md/raid:%s: starting dirty degraded array"
6035 " - data corruption possible.\n",
6ff8d8ec
N		 6036 mdname(mddev));
6037 else {
6038 printk(KERN_ERR
0c55e022 6039 "md/raid:%s: cannot start dirty degraded array.\n",
6ff8d8ec
N		 6040 mdname(mddev));
6041 goto abort;
6042 }
1da177e4
LT		 6043 }
6044
1da177e4 6045 if (mddev->degraded == 0)
0c55e022
N		 6046 printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d"
6047 " devices, algorithm %d\n", mdname(mddev), conf->level,
e183eaed
N		 6048 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
6049 mddev->new_layout);
1da177e4 6050 else
0c55e022
N		 6051 printk(KERN_ALERT "md/raid:%s: raid level %d active with %d"
6052 " out of %d devices, algorithm %d\n",
6053 mdname(mddev), conf->level,
6054 mddev->raid_disks - mddev->degraded,
6055 mddev->raid_disks, mddev->new_layout);
1da177e4
LT		 6056
6057 print_raid5_conf(conf);
6058
fef9c61f 6059 if (conf->reshape_progress != MaxSector) {
fef9c61f 6060 conf->reshape_safe = conf->reshape_progress;
f6705578
N		 6061 atomic_set(&conf->reshape_stripes, 0);
6062 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6063 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6064 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6065 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6066 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 6067 "reshape");
f6705578
N		 6068 }
6069
1da177e4
LT		 6070
6071 /* Ok, everything is just fine now */
a64c876f
N		 6072 if (mddev->to_remove == &raid5_attrs_group)
6073 mddev->to_remove = NULL;
00bcb4ac
N		 6074 else if (mddev->kobj.sd &&
6075 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
5e55e2f5 6076 printk(KERN_WARNING
4a5add49 6077 "raid5: failed to create sysfs attributes for %s\n",
5e55e2f5 6078 mdname(mddev));
4a5add49 6079 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
7a5febe9 6080
4a5add49 6081 if (mddev->queue) {
9f7c2220 6082 int chunk_size;
620125f2 6083 bool discard_supported = true;
4a5add49
N		 6084 /* read-ahead size must cover two whole stripes, which
6085 * is 2 * (datadisks) * chunksize where 'n' is the
6086 * number of raid devices
6087 */
6088 int data_disks = conf->previous_raid_disks - conf->max_degraded;
6089 int stripe = data_disks *
6090 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
6091 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
6092 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
91adb564 6093
4a5add49 6094 blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec);
f022b2fd 6095
11d8a6e3
N		 6096 mddev->queue->backing_dev_info.congested_data = mddev;
6097 mddev->queue->backing_dev_info.congested_fn = raid5_congested;
7a5febe9 6098
9f7c2220
N		 6099 chunk_size = mddev->chunk_sectors << 9;
6100 blk_queue_io_min(mddev->queue, chunk_size);
6101 blk_queue_io_opt(mddev->queue, chunk_size *
6102 (conf->raid_disks - conf->max_degraded));
620125f2
SL		 6103 /*
6104 * We can only discard a whole stripe. It doesn't make sense to
6105 * discard data disk but write parity disk
6106 */
6107 stripe = stripe * PAGE_SIZE;
4ac6875e
N		 6108 /* Round up to power of 2, as discard handling
6109 * currently assumes that */
6110 while ((stripe-1) & stripe)
6111 stripe = (stripe | (stripe-1)) + 1;
620125f2
SL		 6112 mddev->queue->limits.discard_alignment = stripe;
6113 mddev->queue->limits.discard_granularity = stripe;
6114 /*
6115 * unaligned part of discard request will be ignored, so can't
6116 * guarantee discard_zerors_data
6117 */
6118 mddev->queue->limits.discard_zeroes_data = 0;
8f6c2e4b 6119
5026d7a9
PA		 6120 blk_queue_max_write_same_sectors(mddev->queue, 0);
6121
05616be5 6122 rdev_for_each(rdev, mddev) {
9f7c2220
N		 6123 disk_stack_limits(mddev->gendisk, rdev->bdev,
6124 rdev->data_offset << 9);
05616be5
N		 6125 disk_stack_limits(mddev->gendisk, rdev->bdev,
6126 rdev->new_data_offset << 9);
620125f2
SL		 6127 /*
6128 * discard_zeroes_data is required, otherwise data
6129 * could be lost. Consider a scenario: discard a stripe
6130 * (the stripe could be inconsistent if
6131 * discard_zeroes_data is 0); write one disk of the
6132 * stripe (the stripe could be inconsistent again
6133 * depending on which disks are used to calculate
6134 * parity); the disk is broken; The stripe data of this
6135 * disk is lost.
6136 */
6137 if (!blk_queue_discard(bdev_get_queue(rdev->bdev)) ||
6138 !bdev_get_queue(rdev->bdev)->
6139 limits.discard_zeroes_data)
6140 discard_supported = false;
05616be5 6141 }
620125f2
SL		 6142
6143 if (discard_supported &&
6144 mddev->queue->limits.max_discard_sectors >= stripe &&
6145 mddev->queue->limits.discard_granularity >= stripe)
6146 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
6147 mddev->queue);
6148 else
6149 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
6150 mddev->queue);
9f7c2220 6151 }
23032a0e 6152
1da177e4
LT		 6153 return 0;
6154abort:
01f96c0a 6155 md_unregister_thread(&mddev->thread);
e4f869d9
N		 6156 print_raid5_conf(conf);
6157 free_conf(conf);
1da177e4 6158 mddev->private = NULL;
0c55e022 6159 printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev));
1da177e4
LT		 6160 return -EIO;
6161}
6162
fd01b88c 6163static int stop(struct mddev *mddev)
1da177e4 6164{
d1688a6d 6165 struct r5conf *conf = mddev->private;
1da177e4 6166
01f96c0a 6167 md_unregister_thread(&mddev->thread);
11d8a6e3
N		 6168 if (mddev->queue)
6169 mddev->queue->backing_dev_info.congested_fn = NULL;
95fc17aa 6170 free_conf(conf);
a64c876f
N		 6171 mddev->private = NULL;
6172 mddev->to_remove = &raid5_attrs_group;
1da177e4
LT		 6173 return 0;
6174}
6175
fd01b88c 6176static void status(struct seq_file *seq, struct mddev *mddev)
1da177e4 6177{
d1688a6d 6178 struct r5conf *conf = mddev->private;
1da177e4
LT		 6179 int i;
6180
9d8f0363
AN		 6181 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
6182 mddev->chunk_sectors / 2, mddev->layout);
02c2de8c 6183 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
1da177e4
LT		 6184 for (i = 0; i < conf->raid_disks; i++)
6185 seq_printf (seq, "%s",
6186 conf->disks[i].rdev &&
b2d444d7 6187 test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_");
1da177e4 6188 seq_printf (seq, "]");
1da177e4
LT		 6189}
6190
d1688a6d 6191static void print_raid5_conf (struct r5conf *conf)
1da177e4
LT		 6192{
6193 int i;
6194 struct disk_info *tmp;
6195
0c55e022 6196 printk(KERN_DEBUG "RAID conf printout:\n");
1da177e4
LT		 6197 if (!conf) {
6198 printk("(conf==NULL)\n");
6199 return;
6200 }
0c55e022
N		 6201 printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level,
6202 conf->raid_disks,
6203 conf->raid_disks - conf->mddev->degraded);
1da177e4
LT		 6204
6205 for (i = 0; i < conf->raid_disks; i++) {
6206 char b[BDEVNAME_SIZE];
6207 tmp = conf->disks + i;
6208 if (tmp->rdev)
0c55e022
N		 6209 printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n",
6210 i, !test_bit(Faulty, &tmp->rdev->flags),
6211 bdevname(tmp->rdev->bdev, b));
1da177e4
LT		 6212 }
6213}
6214
fd01b88c 6215static int raid5_spare_active(struct mddev *mddev)
1da177e4
LT		 6216{
6217 int i;
d1688a6d 6218 struct r5conf *conf = mddev->private;
1da177e4 6219 struct disk_info *tmp;
6b965620
N		 6220 int count = 0;
6221 unsigned long flags;
1da177e4
LT		 6222
6223 for (i = 0; i < conf->raid_disks; i++) {
6224 tmp = conf->disks + i;
dd054fce
N		 6225 if (tmp->replacement
6226 && tmp->replacement->recovery_offset == MaxSector
6227 && !test_bit(Faulty, &tmp->replacement->flags)
6228 && !test_and_set_bit(In_sync, &tmp->replacement->flags)) {
6229 /* Replacement has just become active. */
6230 if (!tmp->rdev
6231 || !test_and_clear_bit(In_sync, &tmp->rdev->flags))
6232 count++;
6233 if (tmp->rdev) {
6234 /* Replaced device not technically faulty,
6235 * but we need to be sure it gets removed
6236 * and never re-added.
6237 */
6238 set_bit(Faulty, &tmp->rdev->flags);
6239 sysfs_notify_dirent_safe(
6240 tmp->rdev->sysfs_state);
6241 }
6242 sysfs_notify_dirent_safe(tmp->replacement->sysfs_state);
6243 } else if (tmp->rdev
70fffd0b 6244 && tmp->rdev->recovery_offset == MaxSector
b2d444d7 6245 && !test_bit(Faulty, &tmp->rdev->flags)
c04be0aa 6246 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
6b965620 6247 count++;
43c73ca4 6248 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
1da177e4
LT		 6249 }
6250 }
6b965620 6251 spin_lock_irqsave(&conf->device_lock, flags);
908f4fbd 6252 mddev->degraded = calc_degraded(conf);
6b965620 6253 spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4 6254 print_raid5_conf(conf);
6b965620 6255 return count;
1da177e4
LT		 6256}
6257
b8321b68 6258static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 6259{
d1688a6d 6260 struct r5conf *conf = mddev->private;
1da177e4 6261 int err = 0;
b8321b68 6262 int number = rdev->raid_disk;
657e3e4d 6263 struct md_rdev **rdevp;
1da177e4
LT		 6264 struct disk_info *p = conf->disks + number;
6265
6266 print_raid5_conf(conf);
657e3e4d
N		 6267 if (rdev == p->rdev)
6268 rdevp = &p->rdev;
6269 else if (rdev == p->replacement)
6270 rdevp = &p->replacement;
6271 else
6272 return 0;
6273
6274 if (number >= conf->raid_disks &&
6275 conf->reshape_progress == MaxSector)
6276 clear_bit(In_sync, &rdev->flags);
6277
6278 if (test_bit(In_sync, &rdev->flags) ||
6279 atomic_read(&rdev->nr_pending)) {
6280 err = -EBUSY;
6281 goto abort;
6282 }
6283 /* Only remove non-faulty devices if recovery
6284 * isn't possible.
6285 */
6286 if (!test_bit(Faulty, &rdev->flags) &&
6287 mddev->recovery_disabled != conf->recovery_disabled &&
6288 !has_failed(conf) &&
dd054fce 6289 (!p->replacement || p->replacement == rdev) &&
657e3e4d
N		 6290 number < conf->raid_disks) {
6291 err = -EBUSY;
6292 goto abort;
6293 }
6294 *rdevp = NULL;
6295 synchronize_rcu();
6296 if (atomic_read(&rdev->nr_pending)) {
6297 /* lost the race, try later */
6298 err = -EBUSY;
6299 *rdevp = rdev;
dd054fce
N		 6300 } else if (p->replacement) {
6301 /* We must have just cleared 'rdev' */
6302 p->rdev = p->replacement;
6303 clear_bit(Replacement, &p->replacement->flags);
6304 smp_mb(); /* Make sure other CPUs may see both as identical
6305 * but will never see neither - if they are careful
6306 */
6307 p->replacement = NULL;
6308 clear_bit(WantReplacement, &rdev->flags);
6309 } else
6310 /* We might have just removed the Replacement as faulty-
6311 * clear the bit just in case
6312 */
6313 clear_bit(WantReplacement, &rdev->flags);
1da177e4
LT		 6314abort:
6315
6316 print_raid5_conf(conf);
6317 return err;
6318}
6319
fd01b88c 6320static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 6321{
d1688a6d 6322 struct r5conf *conf = mddev->private;
199050ea 6323 int err = -EEXIST;
1da177e4
LT		 6324 int disk;
6325 struct disk_info *p;
6c2fce2e
NB		 6326 int first = 0;
6327 int last = conf->raid_disks - 1;
1da177e4 6328
7f0da59b
N		 6329 if (mddev->recovery_disabled == conf->recovery_disabled)
6330 return -EBUSY;
6331
dc10c643 6332 if (rdev->saved_raid_disk < 0 && has_failed(conf))
1da177e4 6333 /* no point adding a device */
199050ea 6334 return -EINVAL;
1da177e4 6335
6c2fce2e
NB		 6336 if (rdev->raid_disk >= 0)
6337 first = last = rdev->raid_disk;
1da177e4
LT		 6338
6339 /*
16a53ecc
N		 6340 * find the disk ... but prefer rdev->saved_raid_disk
6341 * if possible.
1da177e4 6342 */
16a53ecc 6343 if (rdev->saved_raid_disk >= 0 &&
6c2fce2e 6344 rdev->saved_raid_disk >= first &&
16a53ecc 6345 conf->disks[rdev->saved_raid_disk].rdev == NULL)
5cfb22a1
N		 6346 first = rdev->saved_raid_disk;
6347
6348 for (disk = first; disk <= last; disk++) {
7bfec5f3
N		 6349 p = conf->disks + disk;
6350 if (p->rdev == NULL) {
b2d444d7 6351 clear_bit(In_sync, &rdev->flags);
1da177e4 6352 rdev->raid_disk = disk;
199050ea 6353 err = 0;
72626685
N		 6354 if (rdev->saved_raid_disk != disk)
6355 conf->fullsync = 1;
d6065f7b 6356 rcu_assign_pointer(p->rdev, rdev);
5cfb22a1 6357 goto out;
1da177e4 6358 }
5cfb22a1
N		 6359 }
6360 for (disk = first; disk <= last; disk++) {
6361 p = conf->disks + disk;
7bfec5f3
N		 6362 if (test_bit(WantReplacement, &p->rdev->flags) &&
6363 p->replacement == NULL) {
6364 clear_bit(In_sync, &rdev->flags);
6365 set_bit(Replacement, &rdev->flags);
6366 rdev->raid_disk = disk;
6367 err = 0;
6368 conf->fullsync = 1;
6369 rcu_assign_pointer(p->replacement, rdev);
6370 break;
6371 }
6372 }
5cfb22a1 6373out:
1da177e4 6374 print_raid5_conf(conf);
199050ea 6375 return err;
1da177e4
LT		 6376}
6377
fd01b88c 6378static int raid5_resize(struct mddev *mddev, sector_t sectors)
1da177e4
LT		 6379{
6380 /* no resync is happening, and there is enough space
6381 * on all devices, so we can resize.
6382 * We need to make sure resync covers any new space.
6383 * If the array is shrinking we should possibly wait until
6384 * any io in the removed space completes, but it hardly seems
6385 * worth it.
6386 */
a4a6125a 6387 sector_t newsize;
9d8f0363 6388 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
a4a6125a
N		 6389 newsize = raid5_size(mddev, sectors, mddev->raid_disks);
6390 if (mddev->external_size &&
6391 mddev->array_sectors > newsize)
b522adcd 6392 return -EINVAL;
a4a6125a
N		 6393 if (mddev->bitmap) {
6394 int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0);
6395 if (ret)
6396 return ret;
6397 }
6398 md_set_array_sectors(mddev, newsize);
f233ea5c 6399 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 6400 revalidate_disk(mddev->gendisk);
b098636c
N		 6401 if (sectors > mddev->dev_sectors &&
6402 mddev->recovery_cp > mddev->dev_sectors) {
58c0fed4 6403 mddev->recovery_cp = mddev->dev_sectors;
1da177e4
LT		 6404 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6405 }
58c0fed4 6406 mddev->dev_sectors = sectors;
4b5c7ae8 6407 mddev->resync_max_sectors = sectors;
1da177e4
LT		 6408 return 0;
6409}
6410
fd01b88c 6411static int check_stripe_cache(struct mddev *mddev)
01ee22b4
N		 6412{
6413 /* Can only proceed if there are plenty of stripe_heads.
6414 * We need a minimum of one full stripe,, and for sensible progress
6415 * it is best to have about 4 times that.
6416 * If we require 4 times, then the default 256 4K stripe_heads will
6417 * allow for chunk sizes up to 256K, which is probably OK.
6418 * If the chunk size is greater, user-space should request more
6419 * stripe_heads first.
6420 */
d1688a6d 6421 struct r5conf *conf = mddev->private;
01ee22b4
N		 6422 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
6423 > conf->max_nr_stripes ||
6424 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
6425 > conf->max_nr_stripes) {
0c55e022
N		 6426 printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n",
6427 mdname(mddev),
01ee22b4
N		 6428 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
6429 / STRIPE_SIZE)*4);
6430 return 0;
6431 }
6432 return 1;
6433}
6434
fd01b88c 6435static int check_reshape(struct mddev *mddev)
29269553 6436{
d1688a6d 6437 struct r5conf *conf = mddev->private;
29269553 6438
88ce4930
N		 6439 if (mddev->delta_disks == 0 &&
6440 mddev->new_layout == mddev->layout &&
664e7c41 6441 mddev->new_chunk_sectors == mddev->chunk_sectors)
50ac168a 6442 return 0; /* nothing to do */
674806d6 6443 if (has_failed(conf))
ec32a2bd 6444 return -EINVAL;
fdcfbbb6 6445 if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) {
ec32a2bd
N		 6446 /* We might be able to shrink, but the devices must
6447 * be made bigger first.
6448 * For raid6, 4 is the minimum size.
6449 * Otherwise 2 is the minimum
6450 */
6451 int min = 2;
6452 if (mddev->level == 6)
6453 min = 4;
6454 if (mddev->raid_disks + mddev->delta_disks < min)
6455 return -EINVAL;
6456 }
29269553 6457
01ee22b4 6458 if (!check_stripe_cache(mddev))
29269553 6459 return -ENOSPC;
29269553 6460
e56108d6
N		 6461 return resize_stripes(conf, (conf->previous_raid_disks
6462 + mddev->delta_disks));
63c70c4f
N		 6463}
6464
fd01b88c 6465static int raid5_start_reshape(struct mddev *mddev)
63c70c4f 6466{
d1688a6d 6467 struct r5conf *conf = mddev->private;
3cb03002 6468 struct md_rdev *rdev;
63c70c4f 6469 int spares = 0;
c04be0aa 6470 unsigned long flags;
63c70c4f 6471
f416885e 6472 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
63c70c4f
N		 6473 return -EBUSY;
6474
01ee22b4
N		 6475 if (!check_stripe_cache(mddev))
6476 return -ENOSPC;
6477
30b67645
N		 6478 if (has_failed(conf))
6479 return -EINVAL;
6480
c6563a8c 6481 rdev_for_each(rdev, mddev) {
469518a3
N		 6482 if (!test_bit(In_sync, &rdev->flags)
6483 && !test_bit(Faulty, &rdev->flags))
29269553 6484 spares++;
c6563a8c 6485 }
63c70c4f 6486
f416885e 6487 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
29269553
N		 6488 /* Not enough devices even to make a degraded array
6489 * of that size
6490 */
6491 return -EINVAL;
6492
ec32a2bd
N		 6493 /* Refuse to reduce size of the array. Any reductions in
6494 * array size must be through explicit setting of array_size
6495 * attribute.
6496 */
6497 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
6498 < mddev->array_sectors) {
0c55e022 6499 printk(KERN_ERR "md/raid:%s: array size must be reduced "
ec32a2bd
N		 6500 "before number of disks\n", mdname(mddev));
6501 return -EINVAL;
6502 }
6503
f6705578 6504 atomic_set(&conf->reshape_stripes, 0);
29269553 6505 spin_lock_irq(&conf->device_lock);
c46501b2 6506 write_seqcount_begin(&conf->gen_lock);
29269553 6507 conf->previous_raid_disks = conf->raid_disks;
63c70c4f 6508 conf->raid_disks += mddev->delta_disks;
09c9e5fa
AN		 6509 conf->prev_chunk_sectors = conf->chunk_sectors;
6510 conf->chunk_sectors = mddev->new_chunk_sectors;
88ce4930
N		 6511 conf->prev_algo = conf->algorithm;
6512 conf->algorithm = mddev->new_layout;
05616be5
N		 6513 conf->generation++;
6514 /* Code that selects data_offset needs to see the generation update
6515 * if reshape_progress has been set - so a memory barrier needed.
6516 */
6517 smp_mb();
2c810cdd 6518 if (mddev->reshape_backwards)
fef9c61f
N		 6519 conf->reshape_progress = raid5_size(mddev, 0, 0);
6520 else
6521 conf->reshape_progress = 0;
6522 conf->reshape_safe = conf->reshape_progress;
c46501b2 6523 write_seqcount_end(&conf->gen_lock);
29269553
N		 6524 spin_unlock_irq(&conf->device_lock);
6525
4d77e3ba
N		 6526 /* Now make sure any requests that proceeded on the assumption
6527 * the reshape wasn't running - like Discard or Read - have
6528 * completed.
6529 */
6530 mddev_suspend(mddev);
6531 mddev_resume(mddev);
6532
29269553
N		 6533 /* Add some new drives, as many as will fit.
6534 * We know there are enough to make the newly sized array work.
3424bf6a
N		 6535 * Don't add devices if we are reducing the number of
6536 * devices in the array. This is because it is not possible
6537 * to correctly record the "partially reconstructed" state of
6538 * such devices during the reshape and confusion could result.
29269553 6539 */
87a8dec9 6540 if (mddev->delta_disks >= 0) {
dafb20fa 6541 rdev_for_each(rdev, mddev)
87a8dec9
N		 6542 if (rdev->raid_disk < 0 &&
6543 !test_bit(Faulty, &rdev->flags)) {
6544 if (raid5_add_disk(mddev, rdev) == 0) {
87a8dec9 6545 if (rdev->raid_disk
9d4c7d87 6546 >= conf->previous_raid_disks)
87a8dec9 6547 set_bit(In_sync, &rdev->flags);
9d4c7d87 6548 else
87a8dec9 6549 rdev->recovery_offset = 0;
36fad858
NK		 6550
6551 if (sysfs_link_rdev(mddev, rdev))
87a8dec9 6552 /* Failure here is OK */;
50da0840 6553 }
87a8dec9
N		 6554 } else if (rdev->raid_disk >= conf->previous_raid_disks
6555 && !test_bit(Faulty, &rdev->flags)) {
6556 /* This is a spare that was manually added */
6557 set_bit(In_sync, &rdev->flags);
87a8dec9 6558 }
29269553 6559
87a8dec9
N		 6560 /* When a reshape changes the number of devices,
6561 * ->degraded is measured against the larger of the
6562 * pre and post number of devices.
6563 */
ec32a2bd 6564 spin_lock_irqsave(&conf->device_lock, flags);
908f4fbd 6565 mddev->degraded = calc_degraded(conf);
ec32a2bd
N		 6566 spin_unlock_irqrestore(&conf->device_lock, flags);
6567 }
63c70c4f 6568 mddev->raid_disks = conf->raid_disks;
e516402c 6569 mddev->reshape_position = conf->reshape_progress;
850b2b42 6570 set_bit(MD_CHANGE_DEVS, &mddev->flags);
f6705578 6571
29269553
N		 6572 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6573 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6574 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6575 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6576 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 6577 "reshape");
29269553
N		 6578 if (!mddev->sync_thread) {
6579 mddev->recovery = 0;
6580 spin_lock_irq(&conf->device_lock);
ba8805b9 6581 write_seqcount_begin(&conf->gen_lock);
29269553 6582 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
ba8805b9
N		 6583 mddev->new_chunk_sectors =
6584 conf->chunk_sectors = conf->prev_chunk_sectors;
6585 mddev->new_layout = conf->algorithm = conf->prev_algo;
05616be5
N		 6586 rdev_for_each(rdev, mddev)
6587 rdev->new_data_offset = rdev->data_offset;
6588 smp_wmb();
ba8805b9 6589 conf->generation --;
fef9c61f 6590 conf->reshape_progress = MaxSector;
1e3fa9bd 6591 mddev->reshape_position = MaxSector;
ba8805b9 6592 write_seqcount_end(&conf->gen_lock);
29269553
N		 6593 spin_unlock_irq(&conf->device_lock);
6594 return -EAGAIN;
6595 }
c8f517c4 6596 conf->reshape_checkpoint = jiffies;
29269553
N		 6597 md_wakeup_thread(mddev->sync_thread);
6598 md_new_event(mddev);
6599 return 0;
6600}
29269553 6601
ec32a2bd
N		 6602/* This is called from the reshape thread and should make any
6603 * changes needed in 'conf'
6604 */
d1688a6d 6605static void end_reshape(struct r5conf *conf)
29269553 6606{
29269553 6607
f6705578 6608 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
05616be5 6609 struct md_rdev *rdev;
f6705578 6610
f6705578 6611 spin_lock_irq(&conf->device_lock);
cea9c228 6612 conf->previous_raid_disks = conf->raid_disks;
05616be5
N		 6613 rdev_for_each(rdev, conf->mddev)
6614 rdev->data_offset = rdev->new_data_offset;
6615 smp_wmb();
fef9c61f 6616 conf->reshape_progress = MaxSector;
f6705578 6617 spin_unlock_irq(&conf->device_lock);
b0f9ec04 6618 wake_up(&conf->wait_for_overlap);
16a53ecc
N		 6619
6620 /* read-ahead size must cover two whole stripes, which is
6621 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
6622 */
4a5add49 6623 if (conf->mddev->queue) {
cea9c228 6624 int data_disks = conf->raid_disks - conf->max_degraded;
09c9e5fa 6625 int stripe = data_disks * ((conf->chunk_sectors << 9)
cea9c228 6626 / PAGE_SIZE);
16a53ecc
N		 6627 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
6628 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
6629 }
29269553 6630 }
29269553
N		 6631}
6632
ec32a2bd
N		 6633/* This is called from the raid5d thread with mddev_lock held.
6634 * It makes config changes to the device.
6635 */
fd01b88c 6636static void raid5_finish_reshape(struct mddev *mddev)
cea9c228 6637{
d1688a6d 6638 struct r5conf *conf = mddev->private;
cea9c228
N		 6639
6640 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6641
ec32a2bd
N		 6642 if (mddev->delta_disks > 0) {
6643 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
6644 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 6645 revalidate_disk(mddev->gendisk);
ec32a2bd
N		 6646 } else {
6647 int d;
908f4fbd
N		 6648 spin_lock_irq(&conf->device_lock);
6649 mddev->degraded = calc_degraded(conf);
6650 spin_unlock_irq(&conf->device_lock);
ec32a2bd
N		 6651 for (d = conf->raid_disks ;
6652 d < conf->raid_disks - mddev->delta_disks;
1a67dde0 6653 d++) {
3cb03002 6654 struct md_rdev *rdev = conf->disks[d].rdev;
da7613b8
N		 6655 if (rdev)
6656 clear_bit(In_sync, &rdev->flags);
6657 rdev = conf->disks[d].replacement;
6658 if (rdev)
6659 clear_bit(In_sync, &rdev->flags);
1a67dde0 6660 }
cea9c228 6661 }
88ce4930 6662 mddev->layout = conf->algorithm;
09c9e5fa 6663 mddev->chunk_sectors = conf->chunk_sectors;
ec32a2bd
N		 6664 mddev->reshape_position = MaxSector;
6665 mddev->delta_disks = 0;
2c810cdd 6666 mddev->reshape_backwards = 0;
cea9c228
N		 6667 }
6668}
6669
fd01b88c 6670static void raid5_quiesce(struct mddev *mddev, int state)
72626685 6671{
d1688a6d 6672 struct r5conf *conf = mddev->private;
72626685
N		 6673
6674 switch(state) {
e464eafd
N		 6675 case 2: /* resume for a suspend */
6676 wake_up(&conf->wait_for_overlap);
6677 break;
6678
72626685 6679 case 1: /* stop all writes */
566c09c5 6680 lock_all_device_hash_locks_irq(conf);
64bd660b
N		 6681 /* '2' tells resync/reshape to pause so that all
6682 * active stripes can drain
6683 */
6684 conf->quiesce = 2;
566c09c5 6685 wait_event_cmd(conf->wait_for_stripe,
46031f9a
RBJ		 6686 atomic_read(&conf->active_stripes) == 0 &&
6687 atomic_read(&conf->active_aligned_reads) == 0,
566c09c5
SL		 6688 unlock_all_device_hash_locks_irq(conf),
6689 lock_all_device_hash_locks_irq(conf));
64bd660b 6690 conf->quiesce = 1;
566c09c5 6691 unlock_all_device_hash_locks_irq(conf);
64bd660b
N		 6692 /* allow reshape to continue */
6693 wake_up(&conf->wait_for_overlap);
72626685
N		 6694 break;
6695
6696 case 0: /* re-enable writes */
566c09c5 6697 lock_all_device_hash_locks_irq(conf);
72626685
N		 6698 conf->quiesce = 0;
6699 wake_up(&conf->wait_for_stripe);
e464eafd 6700 wake_up(&conf->wait_for_overlap);
566c09c5 6701 unlock_all_device_hash_locks_irq(conf);
72626685
N		 6702 break;
6703 }
72626685 6704}
b15c2e57 6705
d562b0c4 6706
fd01b88c 6707static void *raid45_takeover_raid0(struct mddev *mddev, int level)
54071b38 6708{
e373ab10 6709 struct r0conf *raid0_conf = mddev->private;
d76c8420 6710 sector_t sectors;
54071b38 6711
f1b29bca 6712 /* for raid0 takeover only one zone is supported */
e373ab10 6713 if (raid0_conf->nr_strip_zones > 1) {
0c55e022
N		 6714 printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n",
6715 mdname(mddev));
f1b29bca
DW		 6716 return ERR_PTR(-EINVAL);
6717 }
6718
e373ab10
N		 6719 sectors = raid0_conf->strip_zone[0].zone_end;
6720 sector_div(sectors, raid0_conf->strip_zone[0].nb_dev);
3b71bd93 6721 mddev->dev_sectors = sectors;
f1b29bca 6722 mddev->new_level = level;
54071b38
TM		 6723 mddev->new_layout = ALGORITHM_PARITY_N;
6724 mddev->new_chunk_sectors = mddev->chunk_sectors;
6725 mddev->raid_disks += 1;
6726 mddev->delta_disks = 1;
6727 /* make sure it will be not marked as dirty */
6728 mddev->recovery_cp = MaxSector;
6729
6730 return setup_conf(mddev);
6731}
6732
6733
fd01b88c 6734static void *raid5_takeover_raid1(struct mddev *mddev)
d562b0c4
N		 6735{
6736 int chunksect;
6737
6738 if (mddev->raid_disks != 2 ||
6739 mddev->degraded > 1)
6740 return ERR_PTR(-EINVAL);
6741
6742 /* Should check if there are write-behind devices? */
6743
6744 chunksect = 64*2; /* 64K by default */
6745
6746 /* The array must be an exact multiple of chunksize */
6747 while (chunksect && (mddev->array_sectors & (chunksect-1)))
6748 chunksect >>= 1;
6749
6750 if ((chunksect<<9) < STRIPE_SIZE)
6751 /* array size does not allow a suitable chunk size */
6752 return ERR_PTR(-EINVAL);
6753
6754 mddev->new_level = 5;
6755 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
664e7c41 6756 mddev->new_chunk_sectors = chunksect;
d562b0c4
N		 6757
6758 return setup_conf(mddev);
6759}
6760
fd01b88c 6761static void *raid5_takeover_raid6(struct mddev *mddev)
fc9739c6
N		 6762{
6763 int new_layout;
6764
6765 switch (mddev->layout) {
6766 case ALGORITHM_LEFT_ASYMMETRIC_6:
6767 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
6768 break;
6769 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6770 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
6771 break;
6772 case ALGORITHM_LEFT_SYMMETRIC_6:
6773 new_layout = ALGORITHM_LEFT_SYMMETRIC;
6774 break;
6775 case ALGORITHM_RIGHT_SYMMETRIC_6:
6776 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
6777 break;
6778 case ALGORITHM_PARITY_0_6:
6779 new_layout = ALGORITHM_PARITY_0;
6780 break;
6781 case ALGORITHM_PARITY_N:
6782 new_layout = ALGORITHM_PARITY_N;
6783 break;
6784 default:
6785 return ERR_PTR(-EINVAL);
6786 }
6787 mddev->new_level = 5;
6788 mddev->new_layout = new_layout;
6789 mddev->delta_disks = -1;
6790 mddev->raid_disks -= 1;
6791 return setup_conf(mddev);
6792}
6793
d562b0c4 6794
fd01b88c 6795static int raid5_check_reshape(struct mddev *mddev)
b3546035 6796{
88ce4930
N		 6797 /* For a 2-drive array, the layout and chunk size can be changed
6798 * immediately as not restriping is needed.
6799 * For larger arrays we record the new value - after validation
6800 * to be used by a reshape pass.
b3546035 6801 */
d1688a6d 6802 struct r5conf *conf = mddev->private;
597a711b 6803 int new_chunk = mddev->new_chunk_sectors;
b3546035 6804
597a711b 6805 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
b3546035
N		 6806 return -EINVAL;
6807 if (new_chunk > 0) {
0ba459d2 6808 if (!is_power_of_2(new_chunk))
b3546035 6809 return -EINVAL;
597a711b 6810 if (new_chunk < (PAGE_SIZE>>9))
b3546035 6811 return -EINVAL;
597a711b 6812 if (mddev->array_sectors & (new_chunk-1))
b3546035
N		 6813 /* not factor of array size */
6814 return -EINVAL;
6815 }
6816
6817 /* They look valid */
6818
88ce4930 6819 if (mddev->raid_disks == 2) {
597a711b
N		 6820 /* can make the change immediately */
6821 if (mddev->new_layout >= 0) {
6822 conf->algorithm = mddev->new_layout;
6823 mddev->layout = mddev->new_layout;
88ce4930
N		 6824 }
6825 if (new_chunk > 0) {
597a711b
N		 6826 conf->chunk_sectors = new_chunk ;
6827 mddev->chunk_sectors = new_chunk;
88ce4930
N		 6828 }
6829 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6830 md_wakeup_thread(mddev->thread);
b3546035 6831 }
50ac168a 6832 return check_reshape(mddev);
88ce4930
N		 6833}
6834
fd01b88c 6835static int raid6_check_reshape(struct mddev *mddev)
88ce4930 6836{
597a711b 6837 int new_chunk = mddev->new_chunk_sectors;
50ac168a 6838
597a711b 6839 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
88ce4930 6840 return -EINVAL;
b3546035 6841 if (new_chunk > 0) {
0ba459d2 6842 if (!is_power_of_2(new_chunk))
88ce4930 6843 return -EINVAL;
597a711b 6844 if (new_chunk < (PAGE_SIZE >> 9))
88ce4930 6845 return -EINVAL;
597a711b 6846 if (mddev->array_sectors & (new_chunk-1))
88ce4930
N		 6847 /* not factor of array size */
6848 return -EINVAL;
b3546035 6849 }
88ce4930
N		 6850
6851 /* They look valid */
50ac168a 6852 return check_reshape(mddev);
b3546035
N		 6853}
6854
fd01b88c 6855static void *raid5_takeover(struct mddev *mddev)
d562b0c4
N		 6856{
6857 /* raid5 can take over:
f1b29bca 6858 * raid0 - if there is only one strip zone - make it a raid4 layout
d562b0c4
N		 6859 * raid1 - if there are two drives. We need to know the chunk size
6860 * raid4 - trivial - just use a raid4 layout.
6861 * raid6 - Providing it is a *_6 layout
d562b0c4 6862 */
f1b29bca
DW		 6863 if (mddev->level == 0)
6864 return raid45_takeover_raid0(mddev, 5);
d562b0c4
N		 6865 if (mddev->level == 1)
6866 return raid5_takeover_raid1(mddev);
e9d4758f
N		 6867 if (mddev->level == 4) {
6868 mddev->new_layout = ALGORITHM_PARITY_N;
6869 mddev->new_level = 5;
6870 return setup_conf(mddev);
6871 }
fc9739c6
N		 6872 if (mddev->level == 6)
6873 return raid5_takeover_raid6(mddev);
d562b0c4
N		 6874
6875 return ERR_PTR(-EINVAL);
6876}
6877
fd01b88c 6878static void *raid4_takeover(struct mddev *mddev)
a78d38a1 6879{
f1b29bca
DW		 6880 /* raid4 can take over:
6881 * raid0 - if there is only one strip zone
6882 * raid5 - if layout is right
a78d38a1 6883 */
f1b29bca
DW		 6884 if (mddev->level == 0)
6885 return raid45_takeover_raid0(mddev, 4);
a78d38a1
N		 6886 if (mddev->level == 5 &&
6887 mddev->layout == ALGORITHM_PARITY_N) {
6888 mddev->new_layout = 0;
6889 mddev->new_level = 4;
6890 return setup_conf(mddev);
6891 }
6892 return ERR_PTR(-EINVAL);
6893}
d562b0c4 6894
84fc4b56 6895static struct md_personality raid5_personality;
245f46c2 6896
fd01b88c 6897static void *raid6_takeover(struct mddev *mddev)
245f46c2
N		 6898{
6899 /* Currently can only take over a raid5. We map the
6900 * personality to an equivalent raid6 personality
6901 * with the Q block at the end.
6902 */
6903 int new_layout;
6904
6905 if (mddev->pers != &raid5_personality)
6906 return ERR_PTR(-EINVAL);
6907 if (mddev->degraded > 1)
6908 return ERR_PTR(-EINVAL);
6909 if (mddev->raid_disks > 253)
6910 return ERR_PTR(-EINVAL);
6911 if (mddev->raid_disks < 3)
6912 return ERR_PTR(-EINVAL);
6913
6914 switch (mddev->layout) {
6915 case ALGORITHM_LEFT_ASYMMETRIC:
6916 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
6917 break;
6918 case ALGORITHM_RIGHT_ASYMMETRIC:
6919 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
6920 break;
6921 case ALGORITHM_LEFT_SYMMETRIC:
6922 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
6923 break;
6924 case ALGORITHM_RIGHT_SYMMETRIC:
6925 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
6926 break;
6927 case ALGORITHM_PARITY_0:
6928 new_layout = ALGORITHM_PARITY_0_6;
6929 break;
6930 case ALGORITHM_PARITY_N:
6931 new_layout = ALGORITHM_PARITY_N;
6932 break;
6933 default:
6934 return ERR_PTR(-EINVAL);
6935 }
6936 mddev->new_level = 6;
6937 mddev->new_layout = new_layout;
6938 mddev->delta_disks = 1;
6939 mddev->raid_disks += 1;
6940 return setup_conf(mddev);
6941}
6942
6943
84fc4b56 6944static struct md_personality raid6_personality =
16a53ecc
N		 6945{
6946 .name = "raid6",
6947 .level = 6,
6948 .owner = THIS_MODULE,
6949 .make_request = make_request,
6950 .run = run,
6951 .stop = stop,
6952 .status = status,
6953 .error_handler = error,
6954 .hot_add_disk = raid5_add_disk,
6955 .hot_remove_disk= raid5_remove_disk,
6956 .spare_active = raid5_spare_active,
6957 .sync_request = sync_request,
6958 .resize = raid5_resize,
80c3a6ce 6959 .size = raid5_size,
50ac168a 6960 .check_reshape = raid6_check_reshape,
f416885e 6961 .start_reshape = raid5_start_reshape,
cea9c228 6962 .finish_reshape = raid5_finish_reshape,
16a53ecc 6963 .quiesce = raid5_quiesce,
245f46c2 6964 .takeover = raid6_takeover,
16a53ecc 6965};
84fc4b56 6966static struct md_personality raid5_personality =
1da177e4
LT		 6967{
6968 .name = "raid5",
2604b703 6969 .level = 5,
1da177e4
LT		 6970 .owner = THIS_MODULE,
6971 .make_request = make_request,
6972 .run = run,
6973 .stop = stop,
6974 .status = status,
6975 .error_handler = error,
6976 .hot_add_disk = raid5_add_disk,
6977 .hot_remove_disk= raid5_remove_disk,
6978 .spare_active = raid5_spare_active,
6979 .sync_request = sync_request,
6980 .resize = raid5_resize,
80c3a6ce 6981 .size = raid5_size,
63c70c4f
N		 6982 .check_reshape = raid5_check_reshape,
6983 .start_reshape = raid5_start_reshape,
cea9c228 6984 .finish_reshape = raid5_finish_reshape,
72626685 6985 .quiesce = raid5_quiesce,
d562b0c4 6986 .takeover = raid5_takeover,
1da177e4
LT		 6987};
6988
84fc4b56 6989static struct md_personality raid4_personality =
1da177e4 6990{
2604b703
N		 6991 .name = "raid4",
6992 .level = 4,
6993 .owner = THIS_MODULE,
6994 .make_request = make_request,
6995 .run = run,
6996 .stop = stop,
6997 .status = status,
6998 .error_handler = error,
6999 .hot_add_disk = raid5_add_disk,
7000 .hot_remove_disk= raid5_remove_disk,
7001 .spare_active = raid5_spare_active,
7002 .sync_request = sync_request,
7003 .resize = raid5_resize,
80c3a6ce 7004 .size = raid5_size,
3d37890b
N		 7005 .check_reshape = raid5_check_reshape,
7006 .start_reshape = raid5_start_reshape,
cea9c228 7007 .finish_reshape = raid5_finish_reshape,
2604b703 7008 .quiesce = raid5_quiesce,
a78d38a1 7009 .takeover = raid4_takeover,
2604b703
N		 7010};
7011
7012static int __init raid5_init(void)
7013{
851c30c9
SL		 7014 raid5_wq = alloc_workqueue("raid5wq",
7015 WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0);
7016 if (!raid5_wq)
7017 return -ENOMEM;
16a53ecc 7018 register_md_personality(&raid6_personality);
2604b703
N		 7019 register_md_personality(&raid5_personality);
7020 register_md_personality(&raid4_personality);
7021 return 0;
1da177e4
LT		 7022}
7023
2604b703 7024static void raid5_exit(void)
1da177e4 7025{
16a53ecc 7026 unregister_md_personality(&raid6_personality);
2604b703
N		 7027 unregister_md_personality(&raid5_personality);
7028 unregister_md_personality(&raid4_personality);
851c30c9 7029 destroy_workqueue(raid5_wq);
1da177e4
LT		 7030}
7031
7032module_init(raid5_init);
7033module_exit(raid5_exit);
7034MODULE_LICENSE("GPL");
0efb9e61 7035MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
1da177e4 7036MODULE_ALIAS("md-personality-4"); /* RAID5 */
d9d166c2
N		 7037MODULE_ALIAS("md-raid5");
7038MODULE_ALIAS("md-raid4");
2604b703
N		 7039MODULE_ALIAS("md-level-5");
7040MODULE_ALIAS("md-level-4");
16a53ecc
N		 7041MODULE_ALIAS("md-personality-8"); /* RAID6 */
7042MODULE_ALIAS("md-raid6");
7043MODULE_ALIAS("md-level-6");
7044
7045/* This used to be two separate modules, they were: */
7046MODULE_ALIAS("raid5");
7047MODULE_ALIAS("raid6");
Linux 6.x block layer and io_uring tree(s)