projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
x86/raid6: correctly check for assembler capabilities
[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
8e0e99ba
N		 67static bool devices_handle_discard_safely = false;
68module_param(devices_handle_discard_safely, bool, 0644);
69MODULE_PARM_DESC(devices_handle_discard_safely,
70 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
851c30c9 71static struct workqueue_struct *raid5_wq;
1da177e4
LT		 72/*
73 * Stripe cache
74 */
75
76#define NR_STRIPES 256
77#define STRIPE_SIZE PAGE_SIZE
78#define STRIPE_SHIFT (PAGE_SHIFT - 9)
79#define STRIPE_SECTORS (STRIPE_SIZE>>9)
80#define IO_THRESHOLD 1
8b3e6cdc 81#define BYPASS_THRESHOLD 1
fccddba0 82#define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
1da177e4 83#define HASH_MASK (NR_HASH - 1)
bfc90cb0 84#define MAX_STRIPE_BATCH 8
1da177e4 85
d1688a6d 86static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect)
db298e19
N		 87{
88 int hash = (sect >> STRIPE_SHIFT) & HASH_MASK;
89 return &conf->stripe_hashtbl[hash];
90}
1da177e4 91
566c09c5
SL		 92static inline int stripe_hash_locks_hash(sector_t sect)
93{
94 return (sect >> STRIPE_SHIFT) & STRIPE_HASH_LOCKS_MASK;
95}
96
97static inline void lock_device_hash_lock(struct r5conf *conf, int hash)
98{
99 spin_lock_irq(conf->hash_locks + hash);
100 spin_lock(&conf->device_lock);
101}
102
103static inline void unlock_device_hash_lock(struct r5conf *conf, int hash)
104{
105 spin_unlock(&conf->device_lock);
106 spin_unlock_irq(conf->hash_locks + hash);
107}
108
109static inline void lock_all_device_hash_locks_irq(struct r5conf *conf)
110{
111 int i;
112 local_irq_disable();
113 spin_lock(conf->hash_locks);
114 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
115 spin_lock_nest_lock(conf->hash_locks + i, conf->hash_locks);
116 spin_lock(&conf->device_lock);
117}
118
119static inline void unlock_all_device_hash_locks_irq(struct r5conf *conf)
120{
121 int i;
122 spin_unlock(&conf->device_lock);
123 for (i = NR_STRIPE_HASH_LOCKS; i; i--)
124 spin_unlock(conf->hash_locks + i - 1);
125 local_irq_enable();
126}
127
1da177e4
LT		 128/* bio's attached to a stripe+device for I/O are linked together in bi_sector
129 * order without overlap. There may be several bio's per stripe+device, and
130 * a bio could span several devices.
131 * When walking this list for a particular stripe+device, we must never proceed
132 * beyond a bio that extends past this device, as the next bio might no longer
133 * be valid.
db298e19 134 * This function is used to determine the 'next' bio in the list, given the sector
1da177e4
LT		 135 * of the current stripe+device
136 */
db298e19
N		 137static inline struct bio *r5_next_bio(struct bio *bio, sector_t sector)
138{
aa8b57aa 139 int sectors = bio_sectors(bio);
4f024f37 140 if (bio->bi_iter.bi_sector + sectors < sector + STRIPE_SECTORS)
db298e19
N		 141 return bio->bi_next;
142 else
143 return NULL;
144}
1da177e4 145
960e739d 146/*
5b99c2ff
JA		 147 * We maintain a biased count of active stripes in the bottom 16 bits of
148 * bi_phys_segments, and a count of processed stripes in the upper 16 bits
960e739d 149 */
e7836bd6 150static inline int raid5_bi_processed_stripes(struct bio *bio)
960e739d 151{
e7836bd6
SL		 152 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
153 return (atomic_read(segments) >> 16) & 0xffff;
960e739d
JA		 154}
155
e7836bd6 156static inline int raid5_dec_bi_active_stripes(struct bio *bio)
960e739d 157{
e7836bd6
SL		 158 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
159 return atomic_sub_return(1, segments) & 0xffff;
960e739d
JA		 160}
161
e7836bd6 162static inline void raid5_inc_bi_active_stripes(struct bio *bio)
960e739d 163{
e7836bd6
SL		 164 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
165 atomic_inc(segments);
960e739d
JA		 166}
167
e7836bd6
SL		 168static inline void raid5_set_bi_processed_stripes(struct bio *bio,
169 unsigned int cnt)
960e739d 170{
e7836bd6
SL		 171 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
172 int old, new;
960e739d 173
e7836bd6
SL		 174 do {
175 old = atomic_read(segments);
176 new = (old & 0xffff) | (cnt << 16);
177 } while (atomic_cmpxchg(segments, old, new) != old);
960e739d
JA		 178}
179
e7836bd6 180static inline void raid5_set_bi_stripes(struct bio *bio, unsigned int cnt)
960e739d 181{
e7836bd6
SL		 182 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
183 atomic_set(segments, cnt);
960e739d
JA		 184}
185
d0dabf7e
N		 186/* Find first data disk in a raid6 stripe */
187static inline int raid6_d0(struct stripe_head *sh)
188{
67cc2b81
N		 189 if (sh->ddf_layout)
190 /* ddf always start from first device */
191 return 0;
192 /* md starts just after Q block */
d0dabf7e
N		 193 if (sh->qd_idx == sh->disks - 1)
194 return 0;
195 else
196 return sh->qd_idx + 1;
197}
16a53ecc
N		 198static inline int raid6_next_disk(int disk, int raid_disks)
199{
200 disk++;
201 return (disk < raid_disks) ? disk : 0;
202}
a4456856 203
d0dabf7e
N		 204/* When walking through the disks in a raid5, starting at raid6_d0,
205 * We need to map each disk to a 'slot', where the data disks are slot
206 * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk
207 * is raid_disks-1. This help does that mapping.
208 */
67cc2b81
N		 209static int raid6_idx_to_slot(int idx, struct stripe_head *sh,
210 int *count, int syndrome_disks)
d0dabf7e 211{
6629542e 212 int slot = *count;
67cc2b81 213
e4424fee 214 if (sh->ddf_layout)
6629542e 215 (*count)++;
d0dabf7e 216 if (idx == sh->pd_idx)
67cc2b81 217 return syndrome_disks;
d0dabf7e 218 if (idx == sh->qd_idx)
67cc2b81 219 return syndrome_disks + 1;
e4424fee 220 if (!sh->ddf_layout)
6629542e 221 (*count)++;
d0dabf7e
N		 222 return slot;
223}
224
a4456856
DW		 225static void return_io(struct bio *return_bi)
226{
227 struct bio *bi = return_bi;
228 while (bi) {
a4456856
DW		 229
230 return_bi = bi->bi_next;
231 bi->bi_next = NULL;
4f024f37 232 bi->bi_iter.bi_size = 0;
0a82a8d1
LT		 233 trace_block_bio_complete(bdev_get_queue(bi->bi_bdev),
234 bi, 0);
0e13fe23 235 bio_endio(bi, 0);
a4456856
DW		 236 bi = return_bi;
237 }
238}
239
d1688a6d 240static void print_raid5_conf (struct r5conf *conf);
1da177e4 241
600aa109
DW		 242static int stripe_operations_active(struct stripe_head *sh)
243{
244 return sh->check_state || sh->reconstruct_state ||
245 test_bit(STRIPE_BIOFILL_RUN, &sh->state) ||
246 test_bit(STRIPE_COMPUTE_RUN, &sh->state);
247}
248
851c30c9
SL		 249static void raid5_wakeup_stripe_thread(struct stripe_head *sh)
250{
251 struct r5conf *conf = sh->raid_conf;
252 struct r5worker_group *group;
bfc90cb0 253 int thread_cnt;
851c30c9
SL		 254 int i, cpu = sh->cpu;
255
256 if (!cpu_online(cpu)) {
257 cpu = cpumask_any(cpu_online_mask);
258 sh->cpu = cpu;
259 }
260
261 if (list_empty(&sh->lru)) {
262 struct r5worker_group *group;
263 group = conf->worker_groups + cpu_to_group(cpu);
264 list_add_tail(&sh->lru, &group->handle_list);
bfc90cb0
SL		 265 group->stripes_cnt++;
266 sh->group = group;
851c30c9
SL		 267 }
268
269 if (conf->worker_cnt_per_group == 0) {
270 md_wakeup_thread(conf->mddev->thread);
271 return;
272 }
273
274 group = conf->worker_groups + cpu_to_group(sh->cpu);
275
bfc90cb0
SL		 276 group->workers[0].working = true;
277 /* at least one worker should run to avoid race */
278 queue_work_on(sh->cpu, raid5_wq, &group->workers[0].work);
279
280 thread_cnt = group->stripes_cnt / MAX_STRIPE_BATCH - 1;
281 /* wakeup more workers */
282 for (i = 1; i < conf->worker_cnt_per_group && thread_cnt > 0; i++) {
283 if (group->workers[i].working == false) {
284 group->workers[i].working = true;
285 queue_work_on(sh->cpu, raid5_wq,
286 &group->workers[i].work);
287 thread_cnt--;
288 }
289 }
851c30c9
SL		 290}
291
566c09c5
SL		 292static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh,
293 struct list_head *temp_inactive_list)
1da177e4 294{
4eb788df
SL		 295 BUG_ON(!list_empty(&sh->lru));
296 BUG_ON(atomic_read(&conf->active_stripes)==0);
297 if (test_bit(STRIPE_HANDLE, &sh->state)) {
298 if (test_bit(STRIPE_DELAYED, &sh->state) &&
67f45548 299 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
4eb788df 300 list_add_tail(&sh->lru, &conf->delayed_list);
67f45548
N		 301 if (atomic_read(&conf->preread_active_stripes)
302 < IO_THRESHOLD)
303 md_wakeup_thread(conf->mddev->thread);
304 } else if (test_bit(STRIPE_BIT_DELAY, &sh->state) &&
4eb788df
SL		 305 sh->bm_seq - conf->seq_write > 0)
306 list_add_tail(&sh->lru, &conf->bitmap_list);
307 else {
308 clear_bit(STRIPE_DELAYED, &sh->state);
309 clear_bit(STRIPE_BIT_DELAY, &sh->state);
851c30c9
SL		 310 if (conf->worker_cnt_per_group == 0) {
311 list_add_tail(&sh->lru, &conf->handle_list);
312 } else {
313 raid5_wakeup_stripe_thread(sh);
314 return;
315 }
4eb788df
SL		 316 }
317 md_wakeup_thread(conf->mddev->thread);
318 } else {
319 BUG_ON(stripe_operations_active(sh));
320 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
321 if (atomic_dec_return(&conf->preread_active_stripes)
322 < IO_THRESHOLD)
323 md_wakeup_thread(conf->mddev->thread);
324 atomic_dec(&conf->active_stripes);
566c09c5
SL		 325 if (!test_bit(STRIPE_EXPANDING, &sh->state))
326 list_add_tail(&sh->lru, temp_inactive_list);
1da177e4
LT		 327 }
328}
d0dabf7e 329
566c09c5
SL		 330static void __release_stripe(struct r5conf *conf, struct stripe_head *sh,
331 struct list_head *temp_inactive_list)
4eb788df
SL		 332{
333 if (atomic_dec_and_test(&sh->count))
566c09c5
SL		 334 do_release_stripe(conf, sh, temp_inactive_list);
335}
336
337/*
338 * @hash could be NR_STRIPE_HASH_LOCKS, then we have a list of inactive_list
339 *
340 * Be careful: Only one task can add/delete stripes from temp_inactive_list at
341 * given time. Adding stripes only takes device lock, while deleting stripes
342 * only takes hash lock.
343 */
344static void release_inactive_stripe_list(struct r5conf *conf,
345 struct list_head *temp_inactive_list,
346 int hash)
347{
348 int size;
349 bool do_wakeup = false;
350 unsigned long flags;
351
352 if (hash == NR_STRIPE_HASH_LOCKS) {
353 size = NR_STRIPE_HASH_LOCKS;
354 hash = NR_STRIPE_HASH_LOCKS - 1;
355 } else
356 size = 1;
357 while (size) {
358 struct list_head *list = &temp_inactive_list[size - 1];
359
360 /*
361 * We don't hold any lock here yet, get_active_stripe() might
362 * remove stripes from the list
363 */
364 if (!list_empty_careful(list)) {
365 spin_lock_irqsave(conf->hash_locks + hash, flags);
4bda556a
SL		 366 if (list_empty(conf->inactive_list + hash) &&
367 !list_empty(list))
368 atomic_dec(&conf->empty_inactive_list_nr);
566c09c5
SL		 369 list_splice_tail_init(list, conf->inactive_list + hash);
370 do_wakeup = true;
371 spin_unlock_irqrestore(conf->hash_locks + hash, flags);
372 }
373 size--;
374 hash--;
375 }
376
377 if (do_wakeup) {
378 wake_up(&conf->wait_for_stripe);
379 if (conf->retry_read_aligned)
380 md_wakeup_thread(conf->mddev->thread);
381 }
4eb788df
SL		 382}
383
773ca82f 384/* should hold conf->device_lock already */
566c09c5
SL		 385static int release_stripe_list(struct r5conf *conf,
386 struct list_head *temp_inactive_list)
773ca82f
SL		 387{
388 struct stripe_head *sh;
389 int count = 0;
390 struct llist_node *head;
391
392 head = llist_del_all(&conf->released_stripes);
d265d9dc 393 head = llist_reverse_order(head);
773ca82f 394 while (head) {
566c09c5
SL		 395 int hash;
396
773ca82f
SL		 397 sh = llist_entry(head, struct stripe_head, release_list);
398 head = llist_next(head);
399 /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */
400 smp_mb();
401 clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state);
402 /*
403 * Don't worry the bit is set here, because if the bit is set
404 * again, the count is always > 1. This is true for
405 * STRIPE_ON_UNPLUG_LIST bit too.
406 */
566c09c5
SL		 407 hash = sh->hash_lock_index;
408 __release_stripe(conf, sh, &temp_inactive_list[hash]);
773ca82f
SL		 409 count++;
410 }
411
412 return count;
413}
414
1da177e4
LT		 415static void release_stripe(struct stripe_head *sh)
416{
d1688a6d 417 struct r5conf *conf = sh->raid_conf;
1da177e4 418 unsigned long flags;
566c09c5
SL		 419 struct list_head list;
420 int hash;
773ca82f 421 bool wakeup;
16a53ecc 422
cf170f3f
ES		 423 /* Avoid release_list until the last reference.
424 */
425 if (atomic_add_unless(&sh->count, -1, 1))
426 return;
427
ad4068de 428 if (unlikely(!conf->mddev->thread) ||
429 test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state))
773ca82f
SL		 430 goto slow_path;
431 wakeup = llist_add(&sh->release_list, &conf->released_stripes);
432 if (wakeup)
433 md_wakeup_thread(conf->mddev->thread);
434 return;
435slow_path:
4eb788df 436 local_irq_save(flags);
773ca82f 437 /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */
4eb788df 438 if (atomic_dec_and_lock(&sh->count, &conf->device_lock)) {
566c09c5
SL		 439 INIT_LIST_HEAD(&list);
440 hash = sh->hash_lock_index;
441 do_release_stripe(conf, sh, &list);
4eb788df 442 spin_unlock(&conf->device_lock);
566c09c5 443 release_inactive_stripe_list(conf, &list, hash);
4eb788df
SL		 444 }
445 local_irq_restore(flags);
1da177e4
LT		 446}
447
fccddba0 448static inline void remove_hash(struct stripe_head *sh)
1da177e4 449{
45b4233c
DW		 450 pr_debug("remove_hash(), stripe %llu\n",
451 (unsigned long long)sh->sector);
1da177e4 452
fccddba0 453 hlist_del_init(&sh->hash);
1da177e4
LT		 454}
455
d1688a6d 456static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh)
1da177e4 457{
fccddba0 458 struct hlist_head *hp = stripe_hash(conf, sh->sector);
1da177e4 459
45b4233c
DW		 460 pr_debug("insert_hash(), stripe %llu\n",
461 (unsigned long long)sh->sector);
1da177e4 462
fccddba0 463 hlist_add_head(&sh->hash, hp);
1da177e4
LT		 464}
465
1da177e4 466/* find an idle stripe, make sure it is unhashed, and return it. */
566c09c5 467static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash)
1da177e4
LT		 468{
469 struct stripe_head *sh = NULL;
470 struct list_head *first;
471
566c09c5 472 if (list_empty(conf->inactive_list + hash))
1da177e4 473 goto out;
566c09c5 474 first = (conf->inactive_list + hash)->next;
1da177e4
LT		 475 sh = list_entry(first, struct stripe_head, lru);
476 list_del_init(first);
477 remove_hash(sh);
478 atomic_inc(&conf->active_stripes);
566c09c5 479 BUG_ON(hash != sh->hash_lock_index);
4bda556a
SL		 480 if (list_empty(conf->inactive_list + hash))
481 atomic_inc(&conf->empty_inactive_list_nr);
1da177e4
LT		 482out:
483 return sh;
484}
485
e4e11e38 486static void shrink_buffers(struct stripe_head *sh)
1da177e4
LT		 487{
488 struct page *p;
489 int i;
e4e11e38 490 int num = sh->raid_conf->pool_size;
1da177e4 491
e4e11e38 492 for (i = 0; i < num ; i++) {
d592a996 493 WARN_ON(sh->dev[i].page != sh->dev[i].orig_page);
1da177e4
LT		 494 p = sh->dev[i].page;
495 if (!p)
496 continue;
497 sh->dev[i].page = NULL;
2d1f3b5d 498 put_page(p);
1da177e4
LT		 499 }
500}
501
e4e11e38 502static int grow_buffers(struct stripe_head *sh)
1da177e4
LT		 503{
504 int i;
e4e11e38 505 int num = sh->raid_conf->pool_size;
1da177e4 506
e4e11e38 507 for (i = 0; i < num; i++) {
1da177e4
LT		 508 struct page *page;
509
510 if (!(page = alloc_page(GFP_KERNEL))) {
511 return 1;
512 }
513 sh->dev[i].page = page;
d592a996 514 sh->dev[i].orig_page = page;
1da177e4
LT		 515 }
516 return 0;
517}
518
784052ec 519static void raid5_build_block(struct stripe_head *sh, int i, int previous);
d1688a6d 520static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
911d4ee8 521 struct stripe_head *sh);
1da177e4 522
b5663ba4 523static void init_stripe(struct stripe_head *sh, sector_t sector, int previous)
1da177e4 524{
d1688a6d 525 struct r5conf *conf = sh->raid_conf;
566c09c5 526 int i, seq;
1da177e4 527
78bafebd
ES		 528 BUG_ON(atomic_read(&sh->count) != 0);
529 BUG_ON(test_bit(STRIPE_HANDLE, &sh->state));
600aa109 530 BUG_ON(stripe_operations_active(sh));
d84e0f10 531
45b4233c 532 pr_debug("init_stripe called, stripe %llu\n",
b8e6a15a 533 (unsigned long long)sector);
566c09c5
SL		 534retry:
535 seq = read_seqcount_begin(&conf->gen_lock);
86b42c71 536 sh->generation = conf->generation - previous;
b5663ba4 537 sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks;
1da177e4 538 sh->sector = sector;
911d4ee8 539 stripe_set_idx(sector, conf, previous, sh);
1da177e4
LT		 540 sh->state = 0;
541
7ecaa1e6 542 for (i = sh->disks; i--; ) {
1da177e4
LT		 543 struct r5dev *dev = &sh->dev[i];
544
d84e0f10 545 if (dev->toread || dev->read || dev->towrite || dev->written ||
1da177e4 546 test_bit(R5_LOCKED, &dev->flags)) {
d84e0f10 547 printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n",
1da177e4 548 (unsigned long long)sh->sector, i, dev->toread,
d84e0f10 549 dev->read, dev->towrite, dev->written,
1da177e4 550 test_bit(R5_LOCKED, &dev->flags));
8cfa7b0f 551 WARN_ON(1);
1da177e4
LT		 552 }
553 dev->flags = 0;
784052ec 554 raid5_build_block(sh, i, previous);
1da177e4 555 }
566c09c5
SL		 556 if (read_seqcount_retry(&conf->gen_lock, seq))
557 goto retry;
1da177e4 558 insert_hash(conf, sh);
851c30c9 559 sh->cpu = smp_processor_id();
1da177e4
LT		 560}
561
d1688a6d 562static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector,
86b42c71 563 short generation)
1da177e4
LT		 564{
565 struct stripe_head *sh;
566
45b4233c 567 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
b67bfe0d 568 hlist_for_each_entry(sh, stripe_hash(conf, sector), hash)
86b42c71 569 if (sh->sector == sector && sh->generation == generation)
1da177e4 570 return sh;
45b4233c 571 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
1da177e4
LT		 572 return NULL;
573}
574
674806d6
N		 575/*
576 * Need to check if array has failed when deciding whether to:
577 * - start an array
578 * - remove non-faulty devices
579 * - add a spare
580 * - allow a reshape
581 * This determination is simple when no reshape is happening.
582 * However if there is a reshape, we need to carefully check
583 * both the before and after sections.
584 * This is because some failed devices may only affect one
585 * of the two sections, and some non-in_sync devices may
586 * be insync in the section most affected by failed devices.
587 */
908f4fbd 588static int calc_degraded(struct r5conf *conf)
674806d6 589{
908f4fbd 590 int degraded, degraded2;
674806d6 591 int i;
674806d6
N		 592
593 rcu_read_lock();
594 degraded = 0;
595 for (i = 0; i < conf->previous_raid_disks; i++) {
3cb03002 596 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
e5c86471
N		 597 if (rdev && test_bit(Faulty, &rdev->flags))
598 rdev = rcu_dereference(conf->disks[i].replacement);
674806d6
N		 599 if (!rdev || test_bit(Faulty, &rdev->flags))
600 degraded++;
601 else if (test_bit(In_sync, &rdev->flags))
602 ;
603 else
604 /* not in-sync or faulty.
605 * If the reshape increases the number of devices,
606 * this is being recovered by the reshape, so
607 * this 'previous' section is not in_sync.
608 * If the number of devices is being reduced however,
609 * the device can only be part of the array if
610 * we are reverting a reshape, so this section will
611 * be in-sync.
612 */
613 if (conf->raid_disks >= conf->previous_raid_disks)
614 degraded++;
615 }
616 rcu_read_unlock();
908f4fbd
N		 617 if (conf->raid_disks == conf->previous_raid_disks)
618 return degraded;
674806d6 619 rcu_read_lock();
908f4fbd 620 degraded2 = 0;
674806d6 621 for (i = 0; i < conf->raid_disks; i++) {
3cb03002 622 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
e5c86471
N		 623 if (rdev && test_bit(Faulty, &rdev->flags))
624 rdev = rcu_dereference(conf->disks[i].replacement);
674806d6 625 if (!rdev || test_bit(Faulty, &rdev->flags))
908f4fbd 626 degraded2++;
674806d6
N		 627 else if (test_bit(In_sync, &rdev->flags))
628 ;
629 else
630 /* not in-sync or faulty.
631 * If reshape increases the number of devices, this
632 * section has already been recovered, else it
633 * almost certainly hasn't.
634 */
635 if (conf->raid_disks <= conf->previous_raid_disks)
908f4fbd 636 degraded2++;
674806d6
N		 637 }
638 rcu_read_unlock();
908f4fbd
N		 639 if (degraded2 > degraded)
640 return degraded2;
641 return degraded;
642}
643
644static int has_failed(struct r5conf *conf)
645{
646 int degraded;
647
648 if (conf->mddev->reshape_position == MaxSector)
649 return conf->mddev->degraded > conf->max_degraded;
650
651 degraded = calc_degraded(conf);
674806d6
N		 652 if (degraded > conf->max_degraded)
653 return 1;
654 return 0;
655}
656
b5663ba4 657static struct stripe_head *
d1688a6d 658get_active_stripe(struct r5conf *conf, sector_t sector,
a8c906ca 659 int previous, int noblock, int noquiesce)
1da177e4
LT		 660{
661 struct stripe_head *sh;
566c09c5 662 int hash = stripe_hash_locks_hash(sector);
1da177e4 663
45b4233c 664 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector);
1da177e4 665
566c09c5 666 spin_lock_irq(conf->hash_locks + hash);
1da177e4
LT		 667
668 do {
72626685 669 wait_event_lock_irq(conf->wait_for_stripe,
a8c906ca 670 conf->quiesce == 0 || noquiesce,
566c09c5 671 *(conf->hash_locks + hash));
86b42c71 672 sh = __find_stripe(conf, sector, conf->generation - previous);
1da177e4
LT		 673 if (!sh) {
674 if (!conf->inactive_blocked)
566c09c5 675 sh = get_free_stripe(conf, hash);
1da177e4
LT		 676 if (noblock && sh == NULL)
677 break;
678 if (!sh) {
679 conf->inactive_blocked = 1;
566c09c5
SL		 680 wait_event_lock_irq(
681 conf->wait_for_stripe,
682 !list_empty(conf->inactive_list + hash) &&
683 (atomic_read(&conf->active_stripes)
684 < (conf->max_nr_stripes * 3 / 4)
685 || !conf->inactive_blocked),
686 *(conf->hash_locks + hash));
1da177e4 687 conf->inactive_blocked = 0;
7da9d450 688 } else {
b5663ba4 689 init_stripe(sh, sector, previous);
7da9d450
N		 690 atomic_inc(&sh->count);
691 }
e240c183 692 } else if (!atomic_inc_not_zero(&sh->count)) {
6d183de4 693 spin_lock(&conf->device_lock);
e240c183 694 if (!atomic_read(&sh->count)) {
1da177e4
LT		 695 if (!test_bit(STRIPE_HANDLE, &sh->state))
696 atomic_inc(&conf->active_stripes);
5af9bef7
N		 697 BUG_ON(list_empty(&sh->lru) &&
698 !test_bit(STRIPE_EXPANDING, &sh->state));
16a53ecc 699 list_del_init(&sh->lru);
bfc90cb0
SL		 700 if (sh->group) {
701 sh->group->stripes_cnt--;
702 sh->group = NULL;
703 }
1da177e4 704 }
7da9d450 705 atomic_inc(&sh->count);
6d183de4 706 spin_unlock(&conf->device_lock);
1da177e4
LT		 707 }
708 } while (sh == NULL);
709
566c09c5 710 spin_unlock_irq(conf->hash_locks + hash);
1da177e4
LT		 711 return sh;
712}
713
05616be5
N		 714/* Determine if 'data_offset' or 'new_data_offset' should be used
715 * in this stripe_head.
716 */
717static int use_new_offset(struct r5conf *conf, struct stripe_head *sh)
718{
719 sector_t progress = conf->reshape_progress;
720 /* Need a memory barrier to make sure we see the value
721 * of conf->generation, or ->data_offset that was set before
722 * reshape_progress was updated.
723 */
724 smp_rmb();
725 if (progress == MaxSector)
726 return 0;
727 if (sh->generation == conf->generation - 1)
728 return 0;
729 /* We are in a reshape, and this is a new-generation stripe,
730 * so use new_data_offset.
731 */
732 return 1;
733}
734
6712ecf8
N		 735static void
736raid5_end_read_request(struct bio *bi, int error);
737static void
738raid5_end_write_request(struct bio *bi, int error);
91c00924 739
c4e5ac0a 740static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
91c00924 741{
d1688a6d 742 struct r5conf *conf = sh->raid_conf;
91c00924
DW		 743 int i, disks = sh->disks;
744
745 might_sleep();
746
747 for (i = disks; i--; ) {
748 int rw;
9a3e1101 749 int replace_only = 0;
977df362
N		 750 struct bio *bi, *rbi;
751 struct md_rdev *rdev, *rrdev = NULL;
e9c7469b
TH		 752 if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) {
753 if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags))
754 rw = WRITE_FUA;
755 else
756 rw = WRITE;
9e444768 757 if (test_bit(R5_Discard, &sh->dev[i].flags))
620125f2 758 rw |= REQ_DISCARD;
e9c7469b 759 } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags))
91c00924 760 rw = READ;
9a3e1101
N		 761 else if (test_and_clear_bit(R5_WantReplace,
762 &sh->dev[i].flags)) {
763 rw = WRITE;
764 replace_only = 1;
765 } else
91c00924 766 continue;
bc0934f0
SL		 767 if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags))
768 rw |= REQ_SYNC;
91c00924
DW		 769
770 bi = &sh->dev[i].req;
977df362 771 rbi = &sh->dev[i].rreq; /* For writing to replacement */
91c00924 772
91c00924 773 rcu_read_lock();
9a3e1101 774 rrdev = rcu_dereference(conf->disks[i].replacement);
dd054fce
N		 775 smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */
776 rdev = rcu_dereference(conf->disks[i].rdev);
777 if (!rdev) {
778 rdev = rrdev;
779 rrdev = NULL;
780 }
9a3e1101
N		 781 if (rw & WRITE) {
782 if (replace_only)
783 rdev = NULL;
dd054fce
N		 784 if (rdev == rrdev)
785 /* We raced and saw duplicates */
786 rrdev = NULL;
9a3e1101 787 } else {
dd054fce 788 if (test_bit(R5_ReadRepl, &sh->dev[i].flags) && rrdev)
9a3e1101
N		 789 rdev = rrdev;
790 rrdev = NULL;
791 }
977df362 792
91c00924
DW		 793 if (rdev && test_bit(Faulty, &rdev->flags))
794 rdev = NULL;
795 if (rdev)
796 atomic_inc(&rdev->nr_pending);
977df362
N		 797 if (rrdev && test_bit(Faulty, &rrdev->flags))
798 rrdev = NULL;
799 if (rrdev)
800 atomic_inc(&rrdev->nr_pending);
91c00924
DW		 801 rcu_read_unlock();
802
73e92e51 803 /* We have already checked bad blocks for reads. Now
977df362
N		 804 * need to check for writes. We never accept write errors
805 * on the replacement, so we don't to check rrdev.
73e92e51
N		 806 */
807 while ((rw & WRITE) && rdev &&
808 test_bit(WriteErrorSeen, &rdev->flags)) {
809 sector_t first_bad;
810 int bad_sectors;
811 int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
812 &first_bad, &bad_sectors);
813 if (!bad)
814 break;
815
816 if (bad < 0) {
817 set_bit(BlockedBadBlocks, &rdev->flags);
818 if (!conf->mddev->external &&
819 conf->mddev->flags) {
820 /* It is very unlikely, but we might
821 * still need to write out the
822 * bad block log - better give it
823 * a chance*/
824 md_check_recovery(conf->mddev);
825 }
1850753d 826 /*
827 * Because md_wait_for_blocked_rdev
828 * will dec nr_pending, we must
829 * increment it first.
830 */
831 atomic_inc(&rdev->nr_pending);
73e92e51
N		 832 md_wait_for_blocked_rdev(rdev, conf->mddev);
833 } else {
834 /* Acknowledged bad block - skip the write */
835 rdev_dec_pending(rdev, conf->mddev);
836 rdev = NULL;
837 }
838 }
839
91c00924 840 if (rdev) {
9a3e1101
N		 841 if (s->syncing || s->expanding || s->expanded
842 || s->replacing)
91c00924
DW		 843 md_sync_acct(rdev->bdev, STRIPE_SECTORS);
844
2b7497f0
DW		 845 set_bit(STRIPE_IO_STARTED, &sh->state);
846
2f6db2a7 847 bio_reset(bi);
91c00924 848 bi->bi_bdev = rdev->bdev;
2f6db2a7
KO		 849 bi->bi_rw = rw;
850 bi->bi_end_io = (rw & WRITE)
851 ? raid5_end_write_request
852 : raid5_end_read_request;
853 bi->bi_private = sh;
854
91c00924 855 pr_debug("%s: for %llu schedule op %ld on disc %d\n",
e46b272b 856 __func__, (unsigned long long)sh->sector,
91c00924
DW		 857 bi->bi_rw, i);
858 atomic_inc(&sh->count);
05616be5 859 if (use_new_offset(conf, sh))
4f024f37 860 bi->bi_iter.bi_sector = (sh->sector
05616be5
N		 861 + rdev->new_data_offset);
862 else
4f024f37 863 bi->bi_iter.bi_sector = (sh->sector
05616be5 864 + rdev->data_offset);
3f9e7c14 865 if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
e59aa23f 866 bi->bi_rw |= REQ_NOMERGE;
3f9e7c14 867
d592a996
SL		 868 if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
869 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
870 sh->dev[i].vec.bv_page = sh->dev[i].page;
4997b72e 871 bi->bi_vcnt = 1;
91c00924
DW		 872 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
873 bi->bi_io_vec[0].bv_offset = 0;
4f024f37 874 bi->bi_iter.bi_size = STRIPE_SIZE;
37c61ff3
SL		 875 /*
876 * If this is discard request, set bi_vcnt 0. We don't
877 * want to confuse SCSI because SCSI will replace payload
878 */
879 if (rw & REQ_DISCARD)
880 bi->bi_vcnt = 0;
977df362
N		 881 if (rrdev)
882 set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags);
e3620a3a
JB		 883
884 if (conf->mddev->gendisk)
885 trace_block_bio_remap(bdev_get_queue(bi->bi_bdev),
886 bi, disk_devt(conf->mddev->gendisk),
887 sh->dev[i].sector);
91c00924 888 generic_make_request(bi);
977df362
N		 889 }
890 if (rrdev) {
9a3e1101
N		 891 if (s->syncing || s->expanding || s->expanded
892 || s->replacing)
977df362
N		 893 md_sync_acct(rrdev->bdev, STRIPE_SECTORS);
894
895 set_bit(STRIPE_IO_STARTED, &sh->state);
896
2f6db2a7 897 bio_reset(rbi);
977df362 898 rbi->bi_bdev = rrdev->bdev;
2f6db2a7
KO		 899 rbi->bi_rw = rw;
900 BUG_ON(!(rw & WRITE));
901 rbi->bi_end_io = raid5_end_write_request;
902 rbi->bi_private = sh;
903
977df362
N		 904 pr_debug("%s: for %llu schedule op %ld on "
905 "replacement disc %d\n",
906 __func__, (unsigned long long)sh->sector,
907 rbi->bi_rw, i);
908 atomic_inc(&sh->count);
05616be5 909 if (use_new_offset(conf, sh))
4f024f37 910 rbi->bi_iter.bi_sector = (sh->sector
05616be5
N		 911 + rrdev->new_data_offset);
912 else
4f024f37 913 rbi->bi_iter.bi_sector = (sh->sector
05616be5 914 + rrdev->data_offset);
d592a996
SL		 915 if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
916 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
917 sh->dev[i].rvec.bv_page = sh->dev[i].page;
4997b72e 918 rbi->bi_vcnt = 1;
977df362
N		 919 rbi->bi_io_vec[0].bv_len = STRIPE_SIZE;
920 rbi->bi_io_vec[0].bv_offset = 0;
4f024f37 921 rbi->bi_iter.bi_size = STRIPE_SIZE;
37c61ff3
SL		 922 /*
923 * If this is discard request, set bi_vcnt 0. We don't
924 * want to confuse SCSI because SCSI will replace payload
925 */
926 if (rw & REQ_DISCARD)
927 rbi->bi_vcnt = 0;
e3620a3a
JB		 928 if (conf->mddev->gendisk)
929 trace_block_bio_remap(bdev_get_queue(rbi->bi_bdev),
930 rbi, disk_devt(conf->mddev->gendisk),
931 sh->dev[i].sector);
977df362
N		 932 generic_make_request(rbi);
933 }
934 if (!rdev && !rrdev) {
b062962e 935 if (rw & WRITE)
91c00924
DW		 936 set_bit(STRIPE_DEGRADED, &sh->state);
937 pr_debug("skip op %ld on disc %d for sector %llu\n",
938 bi->bi_rw, i, (unsigned long long)sh->sector);
939 clear_bit(R5_LOCKED, &sh->dev[i].flags);
940 set_bit(STRIPE_HANDLE, &sh->state);
941 }
942 }
943}
944
945static struct dma_async_tx_descriptor *
d592a996
SL		 946async_copy_data(int frombio, struct bio *bio, struct page **page,
947 sector_t sector, struct dma_async_tx_descriptor *tx,
948 struct stripe_head *sh)
91c00924 949{
7988613b
KO		 950 struct bio_vec bvl;
951 struct bvec_iter iter;
91c00924 952 struct page *bio_page;
91c00924 953 int page_offset;
a08abd8c 954 struct async_submit_ctl submit;
0403e382 955 enum async_tx_flags flags = 0;
91c00924 956
4f024f37
KO		 957 if (bio->bi_iter.bi_sector >= sector)
958 page_offset = (signed)(bio->bi_iter.bi_sector - sector) * 512;
91c00924 959 else
4f024f37 960 page_offset = (signed)(sector - bio->bi_iter.bi_sector) * -512;
a08abd8c 961
0403e382
DW		 962 if (frombio)
963 flags |= ASYNC_TX_FENCE;
964 init_async_submit(&submit, flags, tx, NULL, NULL, NULL);
965
7988613b
KO		 966 bio_for_each_segment(bvl, bio, iter) {
967 int len = bvl.bv_len;
91c00924
DW		 968 int clen;
969 int b_offset = 0;
970
971 if (page_offset < 0) {
972 b_offset = -page_offset;
973 page_offset += b_offset;
974 len -= b_offset;
975 }
976
977 if (len > 0 && page_offset + len > STRIPE_SIZE)
978 clen = STRIPE_SIZE - page_offset;
979 else
980 clen = len;
981
982 if (clen > 0) {
7988613b
KO		 983 b_offset += bvl.bv_offset;
984 bio_page = bvl.bv_page;
d592a996
SL		 985 if (frombio) {
986 if (sh->raid_conf->skip_copy &&
987 b_offset == 0 && page_offset == 0 &&
988 clen == STRIPE_SIZE)
989 *page = bio_page;
990 else
991 tx = async_memcpy(*page, bio_page, page_offset,
a08abd8c 992 b_offset, clen, &submit);
d592a996
SL		 993 } else
994 tx = async_memcpy(bio_page, *page, b_offset,
a08abd8c 995 page_offset, clen, &submit);
91c00924 996 }
a08abd8c
DW		 997 /* chain the operations */
998 submit.depend_tx = tx;
999
91c00924
DW		 1000 if (clen < len) /* hit end of page */
1001 break;
1002 page_offset += len;
1003 }
1004
1005 return tx;
1006}
1007
1008static void ops_complete_biofill(void *stripe_head_ref)
1009{
1010 struct stripe_head *sh = stripe_head_ref;
1011 struct bio *return_bi = NULL;
e4d84909 1012 int i;
91c00924 1013
e46b272b 1014 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1015 (unsigned long long)sh->sector);
1016
1017 /* clear completed biofills */
1018 for (i = sh->disks; i--; ) {
1019 struct r5dev *dev = &sh->dev[i];
91c00924
DW		 1020
1021 /* acknowledge completion of a biofill operation */
e4d84909
DW		 1022 /* and check if we need to reply to a read request,
1023 * new R5_Wantfill requests are held off until
83de75cc 1024 * !STRIPE_BIOFILL_RUN
e4d84909
DW		 1025 */
1026 if (test_and_clear_bit(R5_Wantfill, &dev->flags)) {
91c00924 1027 struct bio *rbi, *rbi2;
91c00924 1028
91c00924
DW		 1029 BUG_ON(!dev->read);
1030 rbi = dev->read;
1031 dev->read = NULL;
4f024f37 1032 while (rbi && rbi->bi_iter.bi_sector <
91c00924
DW		 1033 dev->sector + STRIPE_SECTORS) {
1034 rbi2 = r5_next_bio(rbi, dev->sector);
e7836bd6 1035 if (!raid5_dec_bi_active_stripes(rbi)) {
91c00924
DW		 1036 rbi->bi_next = return_bi;
1037 return_bi = rbi;
1038 }
91c00924
DW		 1039 rbi = rbi2;
1040 }
1041 }
1042 }
83de75cc 1043 clear_bit(STRIPE_BIOFILL_RUN, &sh->state);
91c00924
DW		 1044
1045 return_io(return_bi);
1046
e4d84909 1047 set_bit(STRIPE_HANDLE, &sh->state);
91c00924
DW		 1048 release_stripe(sh);
1049}
1050
1051static void ops_run_biofill(struct stripe_head *sh)
1052{
1053 struct dma_async_tx_descriptor *tx = NULL;
a08abd8c 1054 struct async_submit_ctl submit;
91c00924
DW		 1055 int i;
1056
e46b272b 1057 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1058 (unsigned long long)sh->sector);
1059
1060 for (i = sh->disks; i--; ) {
1061 struct r5dev *dev = &sh->dev[i];
1062 if (test_bit(R5_Wantfill, &dev->flags)) {
1063 struct bio *rbi;
b17459c0 1064 spin_lock_irq(&sh->stripe_lock);
91c00924
DW		 1065 dev->read = rbi = dev->toread;
1066 dev->toread = NULL;
b17459c0 1067 spin_unlock_irq(&sh->stripe_lock);
4f024f37 1068 while (rbi && rbi->bi_iter.bi_sector <
91c00924 1069 dev->sector + STRIPE_SECTORS) {
d592a996
SL		 1070 tx = async_copy_data(0, rbi, &dev->page,
1071 dev->sector, tx, sh);
91c00924
DW		 1072 rbi = r5_next_bio(rbi, dev->sector);
1073 }
1074 }
1075 }
1076
1077 atomic_inc(&sh->count);
a08abd8c
DW		 1078 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL);
1079 async_trigger_callback(&submit);
91c00924
DW		 1080}
1081
4e7d2c0a 1082static void mark_target_uptodate(struct stripe_head *sh, int target)
91c00924 1083{
4e7d2c0a 1084 struct r5dev *tgt;
91c00924 1085
4e7d2c0a
DW		 1086 if (target < 0)
1087 return;
91c00924 1088
4e7d2c0a 1089 tgt = &sh->dev[target];
91c00924
DW		 1090 set_bit(R5_UPTODATE, &tgt->flags);
1091 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1092 clear_bit(R5_Wantcompute, &tgt->flags);
4e7d2c0a
DW		 1093}
1094
ac6b53b6 1095static void ops_complete_compute(void *stripe_head_ref)
91c00924
DW		 1096{
1097 struct stripe_head *sh = stripe_head_ref;
91c00924 1098
e46b272b 1099 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1100 (unsigned long long)sh->sector);
1101
ac6b53b6 1102 /* mark the computed target(s) as uptodate */
4e7d2c0a 1103 mark_target_uptodate(sh, sh->ops.target);
ac6b53b6 1104 mark_target_uptodate(sh, sh->ops.target2);
4e7d2c0a 1105
ecc65c9b
DW		 1106 clear_bit(STRIPE_COMPUTE_RUN, &sh->state);
1107 if (sh->check_state == check_state_compute_run)
1108 sh->check_state = check_state_compute_result;
91c00924
DW		 1109 set_bit(STRIPE_HANDLE, &sh->state);
1110 release_stripe(sh);
1111}
1112
d6f38f31
DW		 1113/* return a pointer to the address conversion region of the scribble buffer */
1114static addr_conv_t *to_addr_conv(struct stripe_head *sh,
1115 struct raid5_percpu *percpu)
1116{
1117 return percpu->scribble + sizeof(struct page *) * (sh->disks + 2);
1118}
1119
1120static struct dma_async_tx_descriptor *
1121ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 1122{
91c00924 1123 int disks = sh->disks;
d6f38f31 1124 struct page **xor_srcs = percpu->scribble;
91c00924
DW		 1125 int target = sh->ops.target;
1126 struct r5dev *tgt = &sh->dev[target];
1127 struct page *xor_dest = tgt->page;
1128 int count = 0;
1129 struct dma_async_tx_descriptor *tx;
a08abd8c 1130 struct async_submit_ctl submit;
91c00924
DW		 1131 int i;
1132
1133 pr_debug("%s: stripe %llu block: %d\n",
e46b272b 1134 __func__, (unsigned long long)sh->sector, target);
91c00924
DW		 1135 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1136
1137 for (i = disks; i--; )
1138 if (i != target)
1139 xor_srcs[count++] = sh->dev[i].page;
1140
1141 atomic_inc(&sh->count);
1142
0403e382 1143 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL,
ac6b53b6 1144 ops_complete_compute, sh, to_addr_conv(sh, percpu));
91c00924 1145 if (unlikely(count == 1))
a08abd8c 1146 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
91c00924 1147 else
a08abd8c 1148 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924 1149
91c00924
DW		 1150 return tx;
1151}
1152
ac6b53b6
DW		 1153/* set_syndrome_sources - populate source buffers for gen_syndrome
1154 * @srcs - (struct page *) array of size sh->disks
1155 * @sh - stripe_head to parse
1156 *
1157 * Populates srcs in proper layout order for the stripe and returns the
1158 * 'count' of sources to be used in a call to async_gen_syndrome. The P
1159 * destination buffer is recorded in srcs[count] and the Q destination
1160 * is recorded in srcs[count+1]].
1161 */
1162static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh)
1163{
1164 int disks = sh->disks;
1165 int syndrome_disks = sh->ddf_layout ? disks : (disks - 2);
1166 int d0_idx = raid6_d0(sh);
1167 int count;
1168 int i;
1169
1170 for (i = 0; i < disks; i++)
5dd33c9a 1171 srcs[i] = NULL;
ac6b53b6
DW		 1172
1173 count = 0;
1174 i = d0_idx;
1175 do {
1176 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
1177
1178 srcs[slot] = sh->dev[i].page;
1179 i = raid6_next_disk(i, disks);
1180 } while (i != d0_idx);
ac6b53b6 1181
e4424fee 1182 return syndrome_disks;
ac6b53b6
DW		 1183}
1184
1185static struct dma_async_tx_descriptor *
1186ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu)
1187{
1188 int disks = sh->disks;
1189 struct page **blocks = percpu->scribble;
1190 int target;
1191 int qd_idx = sh->qd_idx;
1192 struct dma_async_tx_descriptor *tx;
1193 struct async_submit_ctl submit;
1194 struct r5dev *tgt;
1195 struct page *dest;
1196 int i;
1197 int count;
1198
1199 if (sh->ops.target < 0)
1200 target = sh->ops.target2;
1201 else if (sh->ops.target2 < 0)
1202 target = sh->ops.target;
91c00924 1203 else
ac6b53b6
DW		 1204 /* we should only have one valid target */
1205 BUG();
1206 BUG_ON(target < 0);
1207 pr_debug("%s: stripe %llu block: %d\n",
1208 __func__, (unsigned long long)sh->sector, target);
1209
1210 tgt = &sh->dev[target];
1211 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1212 dest = tgt->page;
1213
1214 atomic_inc(&sh->count);
1215
1216 if (target == qd_idx) {
1217 count = set_syndrome_sources(blocks, sh);
1218 blocks[count] = NULL; /* regenerating p is not necessary */
1219 BUG_ON(blocks[count+1] != dest); /* q should already be set */
0403e382
DW		 1220 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1221 ops_complete_compute, sh,
ac6b53b6
DW		 1222 to_addr_conv(sh, percpu));
1223 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
1224 } else {
1225 /* Compute any data- or p-drive using XOR */
1226 count = 0;
1227 for (i = disks; i-- ; ) {
1228 if (i == target || i == qd_idx)
1229 continue;
1230 blocks[count++] = sh->dev[i].page;
1231 }
1232
0403e382
DW		 1233 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1234 NULL, ops_complete_compute, sh,
ac6b53b6
DW		 1235 to_addr_conv(sh, percpu));
1236 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit);
1237 }
91c00924 1238
91c00924
DW		 1239 return tx;
1240}
1241
ac6b53b6
DW		 1242static struct dma_async_tx_descriptor *
1243ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu)
1244{
1245 int i, count, disks = sh->disks;
1246 int syndrome_disks = sh->ddf_layout ? disks : disks-2;
1247 int d0_idx = raid6_d0(sh);
1248 int faila = -1, failb = -1;
1249 int target = sh->ops.target;
1250 int target2 = sh->ops.target2;
1251 struct r5dev *tgt = &sh->dev[target];
1252 struct r5dev *tgt2 = &sh->dev[target2];
1253 struct dma_async_tx_descriptor *tx;
1254 struct page **blocks = percpu->scribble;
1255 struct async_submit_ctl submit;
1256
1257 pr_debug("%s: stripe %llu block1: %d block2: %d\n",
1258 __func__, (unsigned long long)sh->sector, target, target2);
1259 BUG_ON(target < 0 || target2 < 0);
1260 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1261 BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags));
1262
6c910a78 1263 /* we need to open-code set_syndrome_sources to handle the
ac6b53b6
DW		 1264 * slot number conversion for 'faila' and 'failb'
1265 */
1266 for (i = 0; i < disks ; i++)
5dd33c9a 1267 blocks[i] = NULL;
ac6b53b6
DW		 1268 count = 0;
1269 i = d0_idx;
1270 do {
1271 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
1272
1273 blocks[slot] = sh->dev[i].page;
1274
1275 if (i == target)
1276 faila = slot;
1277 if (i == target2)
1278 failb = slot;
1279 i = raid6_next_disk(i, disks);
1280 } while (i != d0_idx);
ac6b53b6
DW		 1281
1282 BUG_ON(faila == failb);
1283 if (failb < faila)
1284 swap(faila, failb);
1285 pr_debug("%s: stripe: %llu faila: %d failb: %d\n",
1286 __func__, (unsigned long long)sh->sector, faila, failb);
1287
1288 atomic_inc(&sh->count);
1289
1290 if (failb == syndrome_disks+1) {
1291 /* Q disk is one of the missing disks */
1292 if (faila == syndrome_disks) {
1293 /* Missing P+Q, just recompute */
0403e382
DW		 1294 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1295 ops_complete_compute, sh,
1296 to_addr_conv(sh, percpu));
e4424fee 1297 return async_gen_syndrome(blocks, 0, syndrome_disks+2,
ac6b53b6
DW		 1298 STRIPE_SIZE, &submit);
1299 } else {
1300 struct page *dest;
1301 int data_target;
1302 int qd_idx = sh->qd_idx;
1303
1304 /* Missing D+Q: recompute D from P, then recompute Q */
1305 if (target == qd_idx)
1306 data_target = target2;
1307 else
1308 data_target = target;
1309
1310 count = 0;
1311 for (i = disks; i-- ; ) {
1312 if (i == data_target || i == qd_idx)
1313 continue;
1314 blocks[count++] = sh->dev[i].page;
1315 }
1316 dest = sh->dev[data_target].page;
0403e382
DW		 1317 init_async_submit(&submit,
1318 ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1319 NULL, NULL, NULL,
1320 to_addr_conv(sh, percpu));
ac6b53b6
DW		 1321 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE,
1322 &submit);
1323
1324 count = set_syndrome_sources(blocks, sh);
0403e382
DW		 1325 init_async_submit(&submit, ASYNC_TX_FENCE, tx,
1326 ops_complete_compute, sh,
1327 to_addr_conv(sh, percpu));
ac6b53b6
DW		 1328 return async_gen_syndrome(blocks, 0, count+2,
1329 STRIPE_SIZE, &submit);
1330 }
ac6b53b6 1331 } else {
6c910a78
DW		 1332 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1333 ops_complete_compute, sh,
1334 to_addr_conv(sh, percpu));
1335 if (failb == syndrome_disks) {
1336 /* We're missing D+P. */
1337 return async_raid6_datap_recov(syndrome_disks+2,
1338 STRIPE_SIZE, faila,
1339 blocks, &submit);
1340 } else {
1341 /* We're missing D+D. */
1342 return async_raid6_2data_recov(syndrome_disks+2,
1343 STRIPE_SIZE, faila, failb,
1344 blocks, &submit);
1345 }
ac6b53b6
DW		 1346 }
1347}
1348
91c00924
DW		 1349static void ops_complete_prexor(void *stripe_head_ref)
1350{
1351 struct stripe_head *sh = stripe_head_ref;
1352
e46b272b 1353 pr_debug("%s: stripe %llu\n", __func__,
91c00924 1354 (unsigned long long)sh->sector);
91c00924
DW		 1355}
1356
1357static struct dma_async_tx_descriptor *
d6f38f31
DW		 1358ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu,
1359 struct dma_async_tx_descriptor *tx)
91c00924 1360{
91c00924 1361 int disks = sh->disks;
d6f38f31 1362 struct page **xor_srcs = percpu->scribble;
91c00924 1363 int count = 0, pd_idx = sh->pd_idx, i;
a08abd8c 1364 struct async_submit_ctl submit;
91c00924
DW		 1365
1366 /* existing parity data subtracted */
1367 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1368
e46b272b 1369 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1370 (unsigned long long)sh->sector);
1371
1372 for (i = disks; i--; ) {
1373 struct r5dev *dev = &sh->dev[i];
1374 /* Only process blocks that are known to be uptodate */
d8ee0728 1375 if (test_bit(R5_Wantdrain, &dev->flags))
91c00924
DW		 1376 xor_srcs[count++] = dev->page;
1377 }
1378
0403e382 1379 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
d6f38f31 1380 ops_complete_prexor, sh, to_addr_conv(sh, percpu));
a08abd8c 1381 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924
DW		 1382
1383 return tx;
1384}
1385
1386static struct dma_async_tx_descriptor *
d8ee0728 1387ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
91c00924
DW		 1388{
1389 int disks = sh->disks;
d8ee0728 1390 int i;
91c00924 1391
e46b272b 1392 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1393 (unsigned long long)sh->sector);
1394
1395 for (i = disks; i--; ) {
1396 struct r5dev *dev = &sh->dev[i];
1397 struct bio *chosen;
91c00924 1398
d8ee0728 1399 if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) {
91c00924
DW		 1400 struct bio *wbi;
1401
b17459c0 1402 spin_lock_irq(&sh->stripe_lock);
91c00924
DW		 1403 chosen = dev->towrite;
1404 dev->towrite = NULL;
1405 BUG_ON(dev->written);
1406 wbi = dev->written = chosen;
b17459c0 1407 spin_unlock_irq(&sh->stripe_lock);
d592a996 1408 WARN_ON(dev->page != dev->orig_page);
91c00924 1409
4f024f37 1410 while (wbi && wbi->bi_iter.bi_sector <
91c00924 1411 dev->sector + STRIPE_SECTORS) {
e9c7469b
TH		 1412 if (wbi->bi_rw & REQ_FUA)
1413 set_bit(R5_WantFUA, &dev->flags);
bc0934f0
SL		 1414 if (wbi->bi_rw & REQ_SYNC)
1415 set_bit(R5_SyncIO, &dev->flags);
9e444768 1416 if (wbi->bi_rw & REQ_DISCARD)
620125f2 1417 set_bit(R5_Discard, &dev->flags);
d592a996
SL		 1418 else {
1419 tx = async_copy_data(1, wbi, &dev->page,
1420 dev->sector, tx, sh);
1421 if (dev->page != dev->orig_page) {
1422 set_bit(R5_SkipCopy, &dev->flags);
1423 clear_bit(R5_UPTODATE, &dev->flags);
1424 clear_bit(R5_OVERWRITE, &dev->flags);
1425 }
1426 }
91c00924
DW		 1427 wbi = r5_next_bio(wbi, dev->sector);
1428 }
1429 }
1430 }
1431
1432 return tx;
1433}
1434
ac6b53b6 1435static void ops_complete_reconstruct(void *stripe_head_ref)
91c00924
DW		 1436{
1437 struct stripe_head *sh = stripe_head_ref;
ac6b53b6
DW		 1438 int disks = sh->disks;
1439 int pd_idx = sh->pd_idx;
1440 int qd_idx = sh->qd_idx;
1441 int i;
9e444768 1442 bool fua = false, sync = false, discard = false;
91c00924 1443
e46b272b 1444 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1445 (unsigned long long)sh->sector);
1446
bc0934f0 1447 for (i = disks; i--; ) {
e9c7469b 1448 fua |= test_bit(R5_WantFUA, &sh->dev[i].flags);
bc0934f0 1449 sync |= test_bit(R5_SyncIO, &sh->dev[i].flags);
9e444768 1450 discard |= test_bit(R5_Discard, &sh->dev[i].flags);
bc0934f0 1451 }
e9c7469b 1452
91c00924
DW		 1453 for (i = disks; i--; ) {
1454 struct r5dev *dev = &sh->dev[i];
ac6b53b6 1455
e9c7469b 1456 if (dev->written || i == pd_idx || i == qd_idx) {
d592a996 1457 if (!discard && !test_bit(R5_SkipCopy, &dev->flags))
9e444768 1458 set_bit(R5_UPTODATE, &dev->flags);
e9c7469b
TH		 1459 if (fua)
1460 set_bit(R5_WantFUA, &dev->flags);
bc0934f0
SL		 1461 if (sync)
1462 set_bit(R5_SyncIO, &dev->flags);
e9c7469b 1463 }
91c00924
DW		 1464 }
1465
d8ee0728
DW		 1466 if (sh->reconstruct_state == reconstruct_state_drain_run)
1467 sh->reconstruct_state = reconstruct_state_drain_result;
1468 else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run)
1469 sh->reconstruct_state = reconstruct_state_prexor_drain_result;
1470 else {
1471 BUG_ON(sh->reconstruct_state != reconstruct_state_run);
1472 sh->reconstruct_state = reconstruct_state_result;
1473 }
91c00924
DW		 1474
1475 set_bit(STRIPE_HANDLE, &sh->state);
1476 release_stripe(sh);
1477}
1478
1479static void
ac6b53b6
DW		 1480ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu,
1481 struct dma_async_tx_descriptor *tx)
91c00924 1482{
91c00924 1483 int disks = sh->disks;
d6f38f31 1484 struct page **xor_srcs = percpu->scribble;
a08abd8c 1485 struct async_submit_ctl submit;
91c00924
DW		 1486 int count = 0, pd_idx = sh->pd_idx, i;
1487 struct page *xor_dest;
d8ee0728 1488 int prexor = 0;
91c00924 1489 unsigned long flags;
91c00924 1490
e46b272b 1491 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1492 (unsigned long long)sh->sector);
1493
620125f2
SL		 1494 for (i = 0; i < sh->disks; i++) {
1495 if (pd_idx == i)
1496 continue;
1497 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1498 break;
1499 }
1500 if (i >= sh->disks) {
1501 atomic_inc(&sh->count);
620125f2
SL		 1502 set_bit(R5_Discard, &sh->dev[pd_idx].flags);
1503 ops_complete_reconstruct(sh);
1504 return;
1505 }
91c00924
DW		 1506 /* check if prexor is active which means only process blocks
1507 * that are part of a read-modify-write (written)
1508 */
d8ee0728
DW		 1509 if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
1510 prexor = 1;
91c00924
DW		 1511 xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1512 for (i = disks; i--; ) {
1513 struct r5dev *dev = &sh->dev[i];
1514 if (dev->written)
1515 xor_srcs[count++] = dev->page;
1516 }
1517 } else {
1518 xor_dest = sh->dev[pd_idx].page;
1519 for (i = disks; i--; ) {
1520 struct r5dev *dev = &sh->dev[i];
1521 if (i != pd_idx)
1522 xor_srcs[count++] = dev->page;
1523 }
1524 }
1525
91c00924
DW		 1526 /* 1/ if we prexor'd then the dest is reused as a source
1527 * 2/ if we did not prexor then we are redoing the parity
1528 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
1529 * for the synchronous xor case
1530 */
88ba2aa5 1531 flags = ASYNC_TX_ACK |
91c00924
DW		 1532 (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
1533
1534 atomic_inc(&sh->count);
1535
ac6b53b6 1536 init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh,
d6f38f31 1537 to_addr_conv(sh, percpu));
a08abd8c
DW		 1538 if (unlikely(count == 1))
1539 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
1540 else
1541 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924
DW		 1542}
1543
ac6b53b6
DW		 1544static void
1545ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu,
1546 struct dma_async_tx_descriptor *tx)
1547{
1548 struct async_submit_ctl submit;
1549 struct page **blocks = percpu->scribble;
620125f2 1550 int count, i;
ac6b53b6
DW		 1551
1552 pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector);
1553
620125f2
SL		 1554 for (i = 0; i < sh->disks; i++) {
1555 if (sh->pd_idx == i || sh->qd_idx == i)
1556 continue;
1557 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1558 break;
1559 }
1560 if (i >= sh->disks) {
1561 atomic_inc(&sh->count);
620125f2
SL		 1562 set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
1563 set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
1564 ops_complete_reconstruct(sh);
1565 return;
1566 }
1567
ac6b53b6
DW		 1568 count = set_syndrome_sources(blocks, sh);
1569
1570 atomic_inc(&sh->count);
1571
1572 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct,
1573 sh, to_addr_conv(sh, percpu));
1574 async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
91c00924
DW		 1575}
1576
1577static void ops_complete_check(void *stripe_head_ref)
1578{
1579 struct stripe_head *sh = stripe_head_ref;
91c00924 1580
e46b272b 1581 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1582 (unsigned long long)sh->sector);
1583
ecc65c9b 1584 sh->check_state = check_state_check_result;
91c00924
DW		 1585 set_bit(STRIPE_HANDLE, &sh->state);
1586 release_stripe(sh);
1587}
1588
ac6b53b6 1589static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 1590{
91c00924 1591 int disks = sh->disks;
ac6b53b6
DW		 1592 int pd_idx = sh->pd_idx;
1593 int qd_idx = sh->qd_idx;
1594 struct page *xor_dest;
d6f38f31 1595 struct page **xor_srcs = percpu->scribble;
91c00924 1596 struct dma_async_tx_descriptor *tx;
a08abd8c 1597 struct async_submit_ctl submit;
ac6b53b6
DW		 1598 int count;
1599 int i;
91c00924 1600
e46b272b 1601 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1602 (unsigned long long)sh->sector);
1603
ac6b53b6
DW		 1604 count = 0;
1605 xor_dest = sh->dev[pd_idx].page;
1606 xor_srcs[count++] = xor_dest;
91c00924 1607 for (i = disks; i--; ) {
ac6b53b6
DW		 1608 if (i == pd_idx || i == qd_idx)
1609 continue;
1610 xor_srcs[count++] = sh->dev[i].page;
91c00924
DW		 1611 }
1612
d6f38f31
DW		 1613 init_async_submit(&submit, 0, NULL, NULL, NULL,
1614 to_addr_conv(sh, percpu));
099f53cb 1615 tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
a08abd8c 1616 &sh->ops.zero_sum_result, &submit);
91c00924 1617
91c00924 1618 atomic_inc(&sh->count);
a08abd8c
DW		 1619 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL);
1620 tx = async_trigger_callback(&submit);
91c00924
DW		 1621}
1622
ac6b53b6
DW		 1623static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp)
1624{
1625 struct page **srcs = percpu->scribble;
1626 struct async_submit_ctl submit;
1627 int count;
1628
1629 pr_debug("%s: stripe %llu checkp: %d\n", __func__,
1630 (unsigned long long)sh->sector, checkp);
1631
1632 count = set_syndrome_sources(srcs, sh);
1633 if (!checkp)
1634 srcs[count] = NULL;
91c00924 1635
91c00924 1636 atomic_inc(&sh->count);
ac6b53b6
DW		 1637 init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check,
1638 sh, to_addr_conv(sh, percpu));
1639 async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE,
1640 &sh->ops.zero_sum_result, percpu->spare_page, &submit);
91c00924
DW		 1641}
1642
51acbcec 1643static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
91c00924
DW		 1644{
1645 int overlap_clear = 0, i, disks = sh->disks;
1646 struct dma_async_tx_descriptor *tx = NULL;
d1688a6d 1647 struct r5conf *conf = sh->raid_conf;
ac6b53b6 1648 int level = conf->level;
d6f38f31
DW		 1649 struct raid5_percpu *percpu;
1650 unsigned long cpu;
91c00924 1651
d6f38f31
DW		 1652 cpu = get_cpu();
1653 percpu = per_cpu_ptr(conf->percpu, cpu);
83de75cc 1654 if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) {
91c00924
DW		 1655 ops_run_biofill(sh);
1656 overlap_clear++;
1657 }
1658
7b3a871e 1659 if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) {
ac6b53b6
DW		 1660 if (level < 6)
1661 tx = ops_run_compute5(sh, percpu);
1662 else {
1663 if (sh->ops.target2 < 0 || sh->ops.target < 0)
1664 tx = ops_run_compute6_1(sh, percpu);
1665 else
1666 tx = ops_run_compute6_2(sh, percpu);
1667 }
1668 /* terminate the chain if reconstruct is not set to be run */
1669 if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request))
7b3a871e
DW		 1670 async_tx_ack(tx);
1671 }
91c00924 1672
600aa109 1673 if (test_bit(STRIPE_OP_PREXOR, &ops_request))
d6f38f31 1674 tx = ops_run_prexor(sh, percpu, tx);
91c00924 1675
600aa109 1676 if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) {
d8ee0728 1677 tx = ops_run_biodrain(sh, tx);
91c00924
DW		 1678 overlap_clear++;
1679 }
1680
ac6b53b6
DW		 1681 if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) {
1682 if (level < 6)
1683 ops_run_reconstruct5(sh, percpu, tx);
1684 else
1685 ops_run_reconstruct6(sh, percpu, tx);
1686 }
91c00924 1687
ac6b53b6
DW		 1688 if (test_bit(STRIPE_OP_CHECK, &ops_request)) {
1689 if (sh->check_state == check_state_run)
1690 ops_run_check_p(sh, percpu);
1691 else if (sh->check_state == check_state_run_q)
1692 ops_run_check_pq(sh, percpu, 0);
1693 else if (sh->check_state == check_state_run_pq)
1694 ops_run_check_pq(sh, percpu, 1);
1695 else
1696 BUG();
1697 }
91c00924 1698
91c00924
DW		 1699 if (overlap_clear)
1700 for (i = disks; i--; ) {
1701 struct r5dev *dev = &sh->dev[i];
1702 if (test_and_clear_bit(R5_Overlap, &dev->flags))
1703 wake_up(&sh->raid_conf->wait_for_overlap);
1704 }
d6f38f31 1705 put_cpu();
91c00924
DW		 1706}
1707
566c09c5 1708static int grow_one_stripe(struct r5conf *conf, int hash)
1da177e4
LT		 1709{
1710 struct stripe_head *sh;
6ce32846 1711 sh = kmem_cache_zalloc(conf->slab_cache, GFP_KERNEL);
3f294f4f
N		 1712 if (!sh)
1713 return 0;
6ce32846 1714
3f294f4f 1715 sh->raid_conf = conf;
3f294f4f 1716
b17459c0
SL		 1717 spin_lock_init(&sh->stripe_lock);
1718
e4e11e38
N		 1719 if (grow_buffers(sh)) {
1720 shrink_buffers(sh);
3f294f4f
N		 1721 kmem_cache_free(conf->slab_cache, sh);
1722 return 0;
1723 }
566c09c5 1724 sh->hash_lock_index = hash;
3f294f4f
N		 1725 /* we just created an active stripe so... */
1726 atomic_set(&sh->count, 1);
1727 atomic_inc(&conf->active_stripes);
1728 INIT_LIST_HEAD(&sh->lru);
1729 release_stripe(sh);
1730 return 1;
1731}
1732
d1688a6d 1733static int grow_stripes(struct r5conf *conf, int num)
3f294f4f 1734{
e18b890b 1735 struct kmem_cache *sc;
5e5e3e78 1736 int devs = max(conf->raid_disks, conf->previous_raid_disks);
566c09c5 1737 int hash;
1da177e4 1738
f4be6b43
N		 1739 if (conf->mddev->gendisk)
1740 sprintf(conf->cache_name[0],
1741 "raid%d-%s", conf->level, mdname(conf->mddev));
1742 else
1743 sprintf(conf->cache_name[0],
1744 "raid%d-%p", conf->level, conf->mddev);
1745 sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]);
1746
ad01c9e3
N		 1747 conf->active_name = 0;
1748 sc = kmem_cache_create(conf->cache_name[conf->active_name],
1da177e4 1749 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
20c2df83 1750 0, 0, NULL);
1da177e4
LT		 1751 if (!sc)
1752 return 1;
1753 conf->slab_cache = sc;
ad01c9e3 1754 conf->pool_size = devs;
566c09c5
SL		 1755 hash = conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS;
1756 while (num--) {
1757 if (!grow_one_stripe(conf, hash))
1da177e4 1758 return 1;
566c09c5
SL		 1759 conf->max_nr_stripes++;
1760 hash = (hash + 1) % NR_STRIPE_HASH_LOCKS;
1761 }
1da177e4
LT		 1762 return 0;
1763}
29269553 1764
d6f38f31
DW		 1765/**
1766 * scribble_len - return the required size of the scribble region
1767 * @num - total number of disks in the array
1768 *
1769 * The size must be enough to contain:
1770 * 1/ a struct page pointer for each device in the array +2
1771 * 2/ room to convert each entry in (1) to its corresponding dma
1772 * (dma_map_page()) or page (page_address()) address.
1773 *
1774 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
1775 * calculate over all devices (not just the data blocks), using zeros in place
1776 * of the P and Q blocks.
1777 */
1778static size_t scribble_len(int num)
1779{
1780 size_t len;
1781
1782 len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2);
1783
1784 return len;
1785}
1786
d1688a6d 1787static int resize_stripes(struct r5conf *conf, int newsize)
ad01c9e3
N		 1788{
1789 /* Make all the stripes able to hold 'newsize' devices.
1790 * New slots in each stripe get 'page' set to a new page.
1791 *
1792 * This happens in stages:
1793 * 1/ create a new kmem_cache and allocate the required number of
1794 * stripe_heads.
83f0d77a 1795 * 2/ gather all the old stripe_heads and transfer the pages across
ad01c9e3
N		 1796 * to the new stripe_heads. This will have the side effect of
1797 * freezing the array as once all stripe_heads have been collected,
1798 * no IO will be possible. Old stripe heads are freed once their
1799 * pages have been transferred over, and the old kmem_cache is
1800 * freed when all stripes are done.
1801 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
1802 * we simple return a failre status - no need to clean anything up.
1803 * 4/ allocate new pages for the new slots in the new stripe_heads.
1804 * If this fails, we don't bother trying the shrink the
1805 * stripe_heads down again, we just leave them as they are.
1806 * As each stripe_head is processed the new one is released into
1807 * active service.
1808 *
1809 * Once step2 is started, we cannot afford to wait for a write,
1810 * so we use GFP_NOIO allocations.
1811 */
1812 struct stripe_head *osh, *nsh;
1813 LIST_HEAD(newstripes);
1814 struct disk_info *ndisks;
d6f38f31 1815 unsigned long cpu;
b5470dc5 1816 int err;
e18b890b 1817 struct kmem_cache *sc;
ad01c9e3 1818 int i;
566c09c5 1819 int hash, cnt;
ad01c9e3
N		 1820
1821 if (newsize <= conf->pool_size)
1822 return 0; /* never bother to shrink */
1823
b5470dc5
DW		 1824 err = md_allow_write(conf->mddev);
1825 if (err)
1826 return err;
2a2275d6 1827
ad01c9e3
N		 1828 /* Step 1 */
1829 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
1830 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
20c2df83 1831 0, 0, NULL);
ad01c9e3
N		 1832 if (!sc)
1833 return -ENOMEM;
1834
1835 for (i = conf->max_nr_stripes; i; i--) {
6ce32846 1836 nsh = kmem_cache_zalloc(sc, GFP_KERNEL);
ad01c9e3
N		 1837 if (!nsh)
1838 break;
1839
ad01c9e3 1840 nsh->raid_conf = conf;
cb13ff69 1841 spin_lock_init(&nsh->stripe_lock);
ad01c9e3
N		 1842
1843 list_add(&nsh->lru, &newstripes);
1844 }
1845 if (i) {
1846 /* didn't get enough, give up */
1847 while (!list_empty(&newstripes)) {
1848 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1849 list_del(&nsh->lru);
1850 kmem_cache_free(sc, nsh);
1851 }
1852 kmem_cache_destroy(sc);
1853 return -ENOMEM;
1854 }
1855 /* Step 2 - Must use GFP_NOIO now.
1856 * OK, we have enough stripes, start collecting inactive
1857 * stripes and copying them over
1858 */
566c09c5
SL		 1859 hash = 0;
1860 cnt = 0;
ad01c9e3 1861 list_for_each_entry(nsh, &newstripes, lru) {
566c09c5
SL		 1862 lock_device_hash_lock(conf, hash);
1863 wait_event_cmd(conf->wait_for_stripe,
1864 !list_empty(conf->inactive_list + hash),
1865 unlock_device_hash_lock(conf, hash),
1866 lock_device_hash_lock(conf, hash));
1867 osh = get_free_stripe(conf, hash);
1868 unlock_device_hash_lock(conf, hash);
ad01c9e3 1869 atomic_set(&nsh->count, 1);
d592a996 1870 for(i=0; i<conf->pool_size; i++) {
ad01c9e3 1871 nsh->dev[i].page = osh->dev[i].page;
d592a996
SL		 1872 nsh->dev[i].orig_page = osh->dev[i].page;
1873 }
ad01c9e3
N		 1874 for( ; i<newsize; i++)
1875 nsh->dev[i].page = NULL;
566c09c5 1876 nsh->hash_lock_index = hash;
ad01c9e3 1877 kmem_cache_free(conf->slab_cache, osh);
566c09c5
SL		 1878 cnt++;
1879 if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS +
1880 !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) {
1881 hash++;
1882 cnt = 0;
1883 }
ad01c9e3
N		 1884 }
1885 kmem_cache_destroy(conf->slab_cache);
1886
1887 /* Step 3.
1888 * At this point, we are holding all the stripes so the array
1889 * is completely stalled, so now is a good time to resize
d6f38f31 1890 * conf->disks and the scribble region
ad01c9e3
N		 1891 */
1892 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
1893 if (ndisks) {
1894 for (i=0; i<conf->raid_disks; i++)
1895 ndisks[i] = conf->disks[i];
1896 kfree(conf->disks);
1897 conf->disks = ndisks;
1898 } else
1899 err = -ENOMEM;
1900
d6f38f31
DW		 1901 get_online_cpus();
1902 conf->scribble_len = scribble_len(newsize);
1903 for_each_present_cpu(cpu) {
1904 struct raid5_percpu *percpu;
1905 void *scribble;
1906
1907 percpu = per_cpu_ptr(conf->percpu, cpu);
1908 scribble = kmalloc(conf->scribble_len, GFP_NOIO);
1909
1910 if (scribble) {
1911 kfree(percpu->scribble);
1912 percpu->scribble = scribble;
1913 } else {
1914 err = -ENOMEM;
1915 break;
1916 }
1917 }
1918 put_online_cpus();
1919
ad01c9e3
N		 1920 /* Step 4, return new stripes to service */
1921 while(!list_empty(&newstripes)) {
1922 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1923 list_del_init(&nsh->lru);
d6f38f31 1924
ad01c9e3
N		 1925 for (i=conf->raid_disks; i < newsize; i++)
1926 if (nsh->dev[i].page == NULL) {
1927 struct page *p = alloc_page(GFP_NOIO);
1928 nsh->dev[i].page = p;
d592a996 1929 nsh->dev[i].orig_page = p;
ad01c9e3
N		 1930 if (!p)
1931 err = -ENOMEM;
1932 }
1933 release_stripe(nsh);
1934 }
1935 /* critical section pass, GFP_NOIO no longer needed */
1936
1937 conf->slab_cache = sc;
1938 conf->active_name = 1-conf->active_name;
1939 conf->pool_size = newsize;
1940 return err;
1941}
1da177e4 1942
566c09c5 1943static int drop_one_stripe(struct r5conf *conf, int hash)
1da177e4
LT		 1944{
1945 struct stripe_head *sh;
1946
566c09c5
SL		 1947 spin_lock_irq(conf->hash_locks + hash);
1948 sh = get_free_stripe(conf, hash);
1949 spin_unlock_irq(conf->hash_locks + hash);
3f294f4f
N		 1950 if (!sh)
1951 return 0;
78bafebd 1952 BUG_ON(atomic_read(&sh->count));
e4e11e38 1953 shrink_buffers(sh);
3f294f4f
N		 1954 kmem_cache_free(conf->slab_cache, sh);
1955 atomic_dec(&conf->active_stripes);
1956 return 1;
1957}
1958
d1688a6d 1959static void shrink_stripes(struct r5conf *conf)
3f294f4f 1960{
566c09c5
SL		 1961 int hash;
1962 for (hash = 0; hash < NR_STRIPE_HASH_LOCKS; hash++)
1963 while (drop_one_stripe(conf, hash))
1964 ;
3f294f4f 1965
29fc7e3e
N		 1966 if (conf->slab_cache)
1967 kmem_cache_destroy(conf->slab_cache);
1da177e4
LT		 1968 conf->slab_cache = NULL;
1969}
1970
6712ecf8 1971static void raid5_end_read_request(struct bio * bi, int error)
1da177e4 1972{
99c0fb5f 1973 struct stripe_head *sh = bi->bi_private;
d1688a6d 1974 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 1975 int disks = sh->disks, i;
1da177e4 1976 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
d6950432 1977 char b[BDEVNAME_SIZE];
dd054fce 1978 struct md_rdev *rdev = NULL;
05616be5 1979 sector_t s;
1da177e4
LT		 1980
1981 for (i=0 ; i<disks; i++)
1982 if (bi == &sh->dev[i].req)
1983 break;
1984
45b4233c
DW		 1985 pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n",
1986 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
1da177e4
LT		 1987 uptodate);
1988 if (i == disks) {
1989 BUG();
6712ecf8 1990 return;
1da177e4 1991 }
14a75d3e 1992 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
dd054fce
N		 1993 /* If replacement finished while this request was outstanding,
1994 * 'replacement' might be NULL already.
1995 * In that case it moved down to 'rdev'.
1996 * rdev is not removed until all requests are finished.
1997 */
14a75d3e 1998 rdev = conf->disks[i].replacement;
dd054fce 1999 if (!rdev)
14a75d3e 2000 rdev = conf->disks[i].rdev;
1da177e4 2001
05616be5
N		 2002 if (use_new_offset(conf, sh))
2003 s = sh->sector + rdev->new_data_offset;
2004 else
2005 s = sh->sector + rdev->data_offset;
1da177e4 2006 if (uptodate) {
1da177e4 2007 set_bit(R5_UPTODATE, &sh->dev[i].flags);
4e5314b5 2008 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
14a75d3e
N		 2009 /* Note that this cannot happen on a
2010 * replacement device. We just fail those on
2011 * any error
2012 */
8bda470e
CD		 2013 printk_ratelimited(
2014 KERN_INFO
2015 "md/raid:%s: read error corrected"
2016 " (%lu sectors at %llu on %s)\n",
2017 mdname(conf->mddev), STRIPE_SECTORS,
05616be5 2018 (unsigned long long)s,
8bda470e 2019 bdevname(rdev->bdev, b));
ddd5115f 2020 atomic_add(STRIPE_SECTORS, &rdev->corrected_errors);
4e5314b5
N		 2021 clear_bit(R5_ReadError, &sh->dev[i].flags);
2022 clear_bit(R5_ReWrite, &sh->dev[i].flags);
3f9e7c14 2023 } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2024 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2025
14a75d3e
N		 2026 if (atomic_read(&rdev->read_errors))
2027 atomic_set(&rdev->read_errors, 0);
1da177e4 2028 } else {
14a75d3e 2029 const char *bdn = bdevname(rdev->bdev, b);
ba22dcbf 2030 int retry = 0;
2e8ac303 2031 int set_bad = 0;
d6950432 2032
1da177e4 2033 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
d6950432 2034 atomic_inc(&rdev->read_errors);
14a75d3e
N		 2035 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
2036 printk_ratelimited(
2037 KERN_WARNING
2038 "md/raid:%s: read error on replacement device "
2039 "(sector %llu on %s).\n",
2040 mdname(conf->mddev),
05616be5 2041 (unsigned long long)s,
14a75d3e 2042 bdn);
2e8ac303 2043 else if (conf->mddev->degraded >= conf->max_degraded) {
2044 set_bad = 1;
8bda470e
CD		 2045 printk_ratelimited(
2046 KERN_WARNING
2047 "md/raid:%s: read error not correctable "
2048 "(sector %llu on %s).\n",
2049 mdname(conf->mddev),
05616be5 2050 (unsigned long long)s,
8bda470e 2051 bdn);
2e8ac303 2052 } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) {
4e5314b5 2053 /* Oh, no!!! */
2e8ac303 2054 set_bad = 1;
8bda470e
CD		 2055 printk_ratelimited(
2056 KERN_WARNING
2057 "md/raid:%s: read error NOT corrected!! "
2058 "(sector %llu on %s).\n",
2059 mdname(conf->mddev),
05616be5 2060 (unsigned long long)s,
8bda470e 2061 bdn);
2e8ac303 2062 } else if (atomic_read(&rdev->read_errors)
ba22dcbf 2063 > conf->max_nr_stripes)
14f8d26b 2064 printk(KERN_WARNING
0c55e022 2065 "md/raid:%s: Too many read errors, failing device %s.\n",
d6950432 2066 mdname(conf->mddev), bdn);
ba22dcbf
N		 2067 else
2068 retry = 1;
edfa1f65
BY		 2069 if (set_bad && test_bit(In_sync, &rdev->flags)
2070 && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2071 retry = 1;
ba22dcbf 2072 if (retry)
3f9e7c14 2073 if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) {
2074 set_bit(R5_ReadError, &sh->dev[i].flags);
2075 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2076 } else
2077 set_bit(R5_ReadNoMerge, &sh->dev[i].flags);
ba22dcbf 2078 else {
4e5314b5
N		 2079 clear_bit(R5_ReadError, &sh->dev[i].flags);
2080 clear_bit(R5_ReWrite, &sh->dev[i].flags);
2e8ac303 2081 if (!(set_bad
2082 && test_bit(In_sync, &rdev->flags)
2083 && rdev_set_badblocks(
2084 rdev, sh->sector, STRIPE_SECTORS, 0)))
2085 md_error(conf->mddev, rdev);
ba22dcbf 2086 }
1da177e4 2087 }
14a75d3e 2088 rdev_dec_pending(rdev, conf->mddev);
1da177e4
LT		 2089 clear_bit(R5_LOCKED, &sh->dev[i].flags);
2090 set_bit(STRIPE_HANDLE, &sh->state);
2091 release_stripe(sh);
1da177e4
LT		 2092}
2093
d710e138 2094static void raid5_end_write_request(struct bio *bi, int error)
1da177e4 2095{
99c0fb5f 2096 struct stripe_head *sh = bi->bi_private;
d1688a6d 2097 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 2098 int disks = sh->disks, i;
977df362 2099 struct md_rdev *uninitialized_var(rdev);
1da177e4 2100 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
b84db560
N		 2101 sector_t first_bad;
2102 int bad_sectors;
977df362 2103 int replacement = 0;
1da177e4 2104
977df362
N		 2105 for (i = 0 ; i < disks; i++) {
2106 if (bi == &sh->dev[i].req) {
2107 rdev = conf->disks[i].rdev;
1da177e4 2108 break;
977df362
N		 2109 }
2110 if (bi == &sh->dev[i].rreq) {
2111 rdev = conf->disks[i].replacement;
dd054fce
N		 2112 if (rdev)
2113 replacement = 1;
2114 else
2115 /* rdev was removed and 'replacement'
2116 * replaced it. rdev is not removed
2117 * until all requests are finished.
2118 */
2119 rdev = conf->disks[i].rdev;
977df362
N		 2120 break;
2121 }
2122 }
45b4233c 2123 pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n",
1da177e4
LT		 2124 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
2125 uptodate);
2126 if (i == disks) {
2127 BUG();
6712ecf8 2128 return;
1da177e4
LT		 2129 }
2130
977df362
N		 2131 if (replacement) {
2132 if (!uptodate)
2133 md_error(conf->mddev, rdev);
2134 else if (is_badblock(rdev, sh->sector,
2135 STRIPE_SECTORS,
2136 &first_bad, &bad_sectors))
2137 set_bit(R5_MadeGoodRepl, &sh->dev[i].flags);
2138 } else {
2139 if (!uptodate) {
9f97e4b1 2140 set_bit(STRIPE_DEGRADED, &sh->state);
977df362
N		 2141 set_bit(WriteErrorSeen, &rdev->flags);
2142 set_bit(R5_WriteError, &sh->dev[i].flags);
3a6de292
N		 2143 if (!test_and_set_bit(WantReplacement, &rdev->flags))
2144 set_bit(MD_RECOVERY_NEEDED,
2145 &rdev->mddev->recovery);
977df362
N		 2146 } else if (is_badblock(rdev, sh->sector,
2147 STRIPE_SECTORS,
c0b32972 2148 &first_bad, &bad_sectors)) {
977df362 2149 set_bit(R5_MadeGood, &sh->dev[i].flags);
c0b32972
N		 2150 if (test_bit(R5_ReadError, &sh->dev[i].flags))
2151 /* That was a successful write so make
2152 * sure it looks like we already did
2153 * a re-write.
2154 */
2155 set_bit(R5_ReWrite, &sh->dev[i].flags);
2156 }
977df362
N		 2157 }
2158 rdev_dec_pending(rdev, conf->mddev);
1da177e4 2159
977df362
N		 2160 if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags))
2161 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1da177e4 2162 set_bit(STRIPE_HANDLE, &sh->state);
c04be0aa 2163 release_stripe(sh);
1da177e4
LT		 2164}
2165
784052ec 2166static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous);
d592a996 2167
784052ec 2168static void raid5_build_block(struct stripe_head *sh, int i, int previous)
1da177e4
LT		 2169{
2170 struct r5dev *dev = &sh->dev[i];
2171
2172 bio_init(&dev->req);
2173 dev->req.bi_io_vec = &dev->vec;
d592a996 2174 dev->req.bi_max_vecs = 1;
1da177e4
LT		 2175 dev->req.bi_private = sh;
2176
977df362
N		 2177 bio_init(&dev->rreq);
2178 dev->rreq.bi_io_vec = &dev->rvec;
d592a996 2179 dev->rreq.bi_max_vecs = 1;
977df362 2180 dev->rreq.bi_private = sh;
977df362 2181
1da177e4 2182 dev->flags = 0;
784052ec 2183 dev->sector = compute_blocknr(sh, i, previous);
1da177e4
LT		 2184}
2185
fd01b88c 2186static void error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 2187{
2188 char b[BDEVNAME_SIZE];
d1688a6d 2189 struct r5conf *conf = mddev->private;
908f4fbd 2190 unsigned long flags;
0c55e022 2191 pr_debug("raid456: error called\n");
1da177e4 2192
908f4fbd
N		 2193 spin_lock_irqsave(&conf->device_lock, flags);
2194 clear_bit(In_sync, &rdev->flags);
2195 mddev->degraded = calc_degraded(conf);
2196 spin_unlock_irqrestore(&conf->device_lock, flags);
2197 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2198
de393cde 2199 set_bit(Blocked, &rdev->flags);
6f8d0c77
N		 2200 set_bit(Faulty, &rdev->flags);
2201 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2202 printk(KERN_ALERT
2203 "md/raid:%s: Disk failure on %s, disabling device.\n"
2204 "md/raid:%s: Operation continuing on %d devices.\n",
2205 mdname(mddev),
2206 bdevname(rdev->bdev, b),
2207 mdname(mddev),
2208 conf->raid_disks - mddev->degraded);
16a53ecc 2209}
1da177e4
LT		 2210
2211/*
2212 * Input: a 'big' sector number,
2213 * Output: index of the data and parity disk, and the sector # in them.
2214 */
d1688a6d 2215static sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector,
911d4ee8
N		 2216 int previous, int *dd_idx,
2217 struct stripe_head *sh)
1da177e4 2218{
6e3b96ed 2219 sector_t stripe, stripe2;
35f2a591 2220 sector_t chunk_number;
1da177e4 2221 unsigned int chunk_offset;
911d4ee8 2222 int pd_idx, qd_idx;
67cc2b81 2223 int ddf_layout = 0;
1da177e4 2224 sector_t new_sector;
e183eaed
N		 2225 int algorithm = previous ? conf->prev_algo
2226 : conf->algorithm;
09c9e5fa
AN		 2227 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2228 : conf->chunk_sectors;
112bf897
N		 2229 int raid_disks = previous ? conf->previous_raid_disks
2230 : conf->raid_disks;
2231 int data_disks = raid_disks - conf->max_degraded;
1da177e4
LT		 2232
2233 /* First compute the information on this sector */
2234
2235 /*
2236 * Compute the chunk number and the sector offset inside the chunk
2237 */
2238 chunk_offset = sector_div(r_sector, sectors_per_chunk);
2239 chunk_number = r_sector;
1da177e4
LT		 2240
2241 /*
2242 * Compute the stripe number
2243 */
35f2a591
N		 2244 stripe = chunk_number;
2245 *dd_idx = sector_div(stripe, data_disks);
6e3b96ed 2246 stripe2 = stripe;
1da177e4
LT		 2247 /*
2248 * Select the parity disk based on the user selected algorithm.
2249 */
84789554 2250 pd_idx = qd_idx = -1;
16a53ecc
N		 2251 switch(conf->level) {
2252 case 4:
911d4ee8 2253 pd_idx = data_disks;
16a53ecc
N		 2254 break;
2255 case 5:
e183eaed 2256 switch (algorithm) {
1da177e4 2257 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2258 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2259 if (*dd_idx >= pd_idx)
1da177e4
LT		 2260 (*dd_idx)++;
2261 break;
2262 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2263 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2264 if (*dd_idx >= pd_idx)
1da177e4
LT		 2265 (*dd_idx)++;
2266 break;
2267 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2268 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2269 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4
LT		 2270 break;
2271 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2272 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2273 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4 2274 break;
99c0fb5f
N		 2275 case ALGORITHM_PARITY_0:
2276 pd_idx = 0;
2277 (*dd_idx)++;
2278 break;
2279 case ALGORITHM_PARITY_N:
2280 pd_idx = data_disks;
2281 break;
1da177e4 2282 default:
99c0fb5f 2283 BUG();
16a53ecc
N		 2284 }
2285 break;
2286 case 6:
2287
e183eaed 2288 switch (algorithm) {
16a53ecc 2289 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2290 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2291 qd_idx = pd_idx + 1;
2292 if (pd_idx == raid_disks-1) {
99c0fb5f 2293 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2294 qd_idx = 0;
2295 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2296 (*dd_idx) += 2; /* D D P Q D */
2297 break;
2298 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2299 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2300 qd_idx = pd_idx + 1;
2301 if (pd_idx == raid_disks-1) {
99c0fb5f 2302 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2303 qd_idx = 0;
2304 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2305 (*dd_idx) += 2; /* D D P Q D */
2306 break;
2307 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2308 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2309 qd_idx = (pd_idx + 1) % raid_disks;
2310 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc
N		 2311 break;
2312 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2313 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2314 qd_idx = (pd_idx + 1) % raid_disks;
2315 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc 2316 break;
99c0fb5f
N		 2317
2318 case ALGORITHM_PARITY_0:
2319 pd_idx = 0;
2320 qd_idx = 1;
2321 (*dd_idx) += 2;
2322 break;
2323 case ALGORITHM_PARITY_N:
2324 pd_idx = data_disks;
2325 qd_idx = data_disks + 1;
2326 break;
2327
2328 case ALGORITHM_ROTATING_ZERO_RESTART:
2329 /* Exactly the same as RIGHT_ASYMMETRIC, but or
2330 * of blocks for computing Q is different.
2331 */
6e3b96ed 2332 pd_idx = sector_div(stripe2, raid_disks);
99c0fb5f
N		 2333 qd_idx = pd_idx + 1;
2334 if (pd_idx == raid_disks-1) {
2335 (*dd_idx)++; /* Q D D D P */
2336 qd_idx = 0;
2337 } else if (*dd_idx >= pd_idx)
2338 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2339 ddf_layout = 1;
99c0fb5f
N		 2340 break;
2341
2342 case ALGORITHM_ROTATING_N_RESTART:
2343 /* Same a left_asymmetric, by first stripe is
2344 * D D D P Q rather than
2345 * Q D D D P
2346 */
6e3b96ed
N		 2347 stripe2 += 1;
2348 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2349 qd_idx = pd_idx + 1;
2350 if (pd_idx == raid_disks-1) {
2351 (*dd_idx)++; /* Q D D D P */
2352 qd_idx = 0;
2353 } else if (*dd_idx >= pd_idx)
2354 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2355 ddf_layout = 1;
99c0fb5f
N		 2356 break;
2357
2358 case ALGORITHM_ROTATING_N_CONTINUE:
2359 /* Same as left_symmetric but Q is before P */
6e3b96ed 2360 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2361 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
2362 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
67cc2b81 2363 ddf_layout = 1;
99c0fb5f
N		 2364 break;
2365
2366 case ALGORITHM_LEFT_ASYMMETRIC_6:
2367 /* RAID5 left_asymmetric, with Q on last device */
6e3b96ed 2368 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2369 if (*dd_idx >= pd_idx)
2370 (*dd_idx)++;
2371 qd_idx = raid_disks - 1;
2372 break;
2373
2374 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6e3b96ed 2375 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2376 if (*dd_idx >= pd_idx)
2377 (*dd_idx)++;
2378 qd_idx = raid_disks - 1;
2379 break;
2380
2381 case ALGORITHM_LEFT_SYMMETRIC_6:
6e3b96ed 2382 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2383 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2384 qd_idx = raid_disks - 1;
2385 break;
2386
2387 case ALGORITHM_RIGHT_SYMMETRIC_6:
6e3b96ed 2388 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2389 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2390 qd_idx = raid_disks - 1;
2391 break;
2392
2393 case ALGORITHM_PARITY_0_6:
2394 pd_idx = 0;
2395 (*dd_idx)++;
2396 qd_idx = raid_disks - 1;
2397 break;
2398
16a53ecc 2399 default:
99c0fb5f 2400 BUG();
16a53ecc
N		 2401 }
2402 break;
1da177e4
LT		 2403 }
2404
911d4ee8
N		 2405 if (sh) {
2406 sh->pd_idx = pd_idx;
2407 sh->qd_idx = qd_idx;
67cc2b81 2408 sh->ddf_layout = ddf_layout;
911d4ee8 2409 }
1da177e4
LT		 2410 /*
2411 * Finally, compute the new sector number
2412 */
2413 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
2414 return new_sector;
2415}
2416
784052ec 2417static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous)
1da177e4 2418{
d1688a6d 2419 struct r5conf *conf = sh->raid_conf;
b875e531
N		 2420 int raid_disks = sh->disks;
2421 int data_disks = raid_disks - conf->max_degraded;
1da177e4 2422 sector_t new_sector = sh->sector, check;
09c9e5fa
AN		 2423 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2424 : conf->chunk_sectors;
e183eaed
N		 2425 int algorithm = previous ? conf->prev_algo
2426 : conf->algorithm;
1da177e4
LT		 2427 sector_t stripe;
2428 int chunk_offset;
35f2a591
N		 2429 sector_t chunk_number;
2430 int dummy1, dd_idx = i;
1da177e4 2431 sector_t r_sector;
911d4ee8 2432 struct stripe_head sh2;
1da177e4
LT		 2433
2434 chunk_offset = sector_div(new_sector, sectors_per_chunk);
2435 stripe = new_sector;
1da177e4 2436
16a53ecc
N		 2437 if (i == sh->pd_idx)
2438 return 0;
2439 switch(conf->level) {
2440 case 4: break;
2441 case 5:
e183eaed 2442 switch (algorithm) {
1da177e4
LT		 2443 case ALGORITHM_LEFT_ASYMMETRIC:
2444 case ALGORITHM_RIGHT_ASYMMETRIC:
2445 if (i > sh->pd_idx)
2446 i--;
2447 break;
2448 case ALGORITHM_LEFT_SYMMETRIC:
2449 case ALGORITHM_RIGHT_SYMMETRIC:
2450 if (i < sh->pd_idx)
2451 i += raid_disks;
2452 i -= (sh->pd_idx + 1);
2453 break;
99c0fb5f
N		 2454 case ALGORITHM_PARITY_0:
2455 i -= 1;
2456 break;
2457 case ALGORITHM_PARITY_N:
2458 break;
1da177e4 2459 default:
99c0fb5f 2460 BUG();
16a53ecc
N		 2461 }
2462 break;
2463 case 6:
d0dabf7e 2464 if (i == sh->qd_idx)
16a53ecc 2465 return 0; /* It is the Q disk */
e183eaed 2466 switch (algorithm) {
16a53ecc
N		 2467 case ALGORITHM_LEFT_ASYMMETRIC:
2468 case ALGORITHM_RIGHT_ASYMMETRIC:
99c0fb5f
N		 2469 case ALGORITHM_ROTATING_ZERO_RESTART:
2470 case ALGORITHM_ROTATING_N_RESTART:
2471 if (sh->pd_idx == raid_disks-1)
2472 i--; /* Q D D D P */
16a53ecc
N		 2473 else if (i > sh->pd_idx)
2474 i -= 2; /* D D P Q D */
2475 break;
2476 case ALGORITHM_LEFT_SYMMETRIC:
2477 case ALGORITHM_RIGHT_SYMMETRIC:
2478 if (sh->pd_idx == raid_disks-1)
2479 i--; /* Q D D D P */
2480 else {
2481 /* D D P Q D */
2482 if (i < sh->pd_idx)
2483 i += raid_disks;
2484 i -= (sh->pd_idx + 2);
2485 }
2486 break;
99c0fb5f
N		 2487 case ALGORITHM_PARITY_0:
2488 i -= 2;
2489 break;
2490 case ALGORITHM_PARITY_N:
2491 break;
2492 case ALGORITHM_ROTATING_N_CONTINUE:
e4424fee 2493 /* Like left_symmetric, but P is before Q */
99c0fb5f
N		 2494 if (sh->pd_idx == 0)
2495 i--; /* P D D D Q */
e4424fee
N		 2496 else {
2497 /* D D Q P D */
2498 if (i < sh->pd_idx)
2499 i += raid_disks;
2500 i -= (sh->pd_idx + 1);
2501 }
99c0fb5f
N		 2502 break;
2503 case ALGORITHM_LEFT_ASYMMETRIC_6:
2504 case ALGORITHM_RIGHT_ASYMMETRIC_6:
2505 if (i > sh->pd_idx)
2506 i--;
2507 break;
2508 case ALGORITHM_LEFT_SYMMETRIC_6:
2509 case ALGORITHM_RIGHT_SYMMETRIC_6:
2510 if (i < sh->pd_idx)
2511 i += data_disks + 1;
2512 i -= (sh->pd_idx + 1);
2513 break;
2514 case ALGORITHM_PARITY_0_6:
2515 i -= 1;
2516 break;
16a53ecc 2517 default:
99c0fb5f 2518 BUG();
16a53ecc
N		 2519 }
2520 break;
1da177e4
LT		 2521 }
2522
2523 chunk_number = stripe * data_disks + i;
35f2a591 2524 r_sector = chunk_number * sectors_per_chunk + chunk_offset;
1da177e4 2525
112bf897 2526 check = raid5_compute_sector(conf, r_sector,
784052ec 2527 previous, &dummy1, &sh2);
911d4ee8
N		 2528 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
2529 || sh2.qd_idx != sh->qd_idx) {
0c55e022
N		 2530 printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n",
2531 mdname(conf->mddev));
1da177e4
LT		 2532 return 0;
2533 }
2534 return r_sector;
2535}
2536
600aa109 2537static void
c0f7bddb 2538schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
600aa109 2539 int rcw, int expand)
e33129d8
DW		 2540{
2541 int i, pd_idx = sh->pd_idx, disks = sh->disks;
d1688a6d 2542 struct r5conf *conf = sh->raid_conf;
c0f7bddb 2543 int level = conf->level;
e33129d8
DW		 2544
2545 if (rcw) {
e33129d8
DW		 2546
2547 for (i = disks; i--; ) {
2548 struct r5dev *dev = &sh->dev[i];
2549
2550 if (dev->towrite) {
2551 set_bit(R5_LOCKED, &dev->flags);
d8ee0728 2552 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2553 if (!expand)
2554 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2555 s->locked++;
e33129d8
DW		 2556 }
2557 }
ce7d363a
N		 2558 /* if we are not expanding this is a proper write request, and
2559 * there will be bios with new data to be drained into the
2560 * stripe cache
2561 */
2562 if (!expand) {
2563 if (!s->locked)
2564 /* False alarm, nothing to do */
2565 return;
2566 sh->reconstruct_state = reconstruct_state_drain_run;
2567 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2568 } else
2569 sh->reconstruct_state = reconstruct_state_run;
2570
2571 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
2572
c0f7bddb 2573 if (s->locked + conf->max_degraded == disks)
8b3e6cdc 2574 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
c0f7bddb 2575 atomic_inc(&conf->pending_full_writes);
e33129d8 2576 } else {
c0f7bddb 2577 BUG_ON(level == 6);
e33129d8
DW		 2578 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
2579 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
2580
e33129d8
DW		 2581 for (i = disks; i--; ) {
2582 struct r5dev *dev = &sh->dev[i];
2583 if (i == pd_idx)
2584 continue;
2585
e33129d8
DW		 2586 if (dev->towrite &&
2587 (test_bit(R5_UPTODATE, &dev->flags) ||
d8ee0728
DW		 2588 test_bit(R5_Wantcompute, &dev->flags))) {
2589 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2590 set_bit(R5_LOCKED, &dev->flags);
2591 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2592 s->locked++;
e33129d8
DW		 2593 }
2594 }
ce7d363a
N		 2595 if (!s->locked)
2596 /* False alarm - nothing to do */
2597 return;
2598 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
2599 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
2600 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2601 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
e33129d8
DW		 2602 }
2603
c0f7bddb 2604 /* keep the parity disk(s) locked while asynchronous operations
e33129d8
DW		 2605 * are in flight
2606 */
2607 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
2608 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
600aa109 2609 s->locked++;
e33129d8 2610
c0f7bddb
YT		 2611 if (level == 6) {
2612 int qd_idx = sh->qd_idx;
2613 struct r5dev *dev = &sh->dev[qd_idx];
2614
2615 set_bit(R5_LOCKED, &dev->flags);
2616 clear_bit(R5_UPTODATE, &dev->flags);
2617 s->locked++;
2618 }
2619
600aa109 2620 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
e46b272b 2621 __func__, (unsigned long long)sh->sector,
600aa109 2622 s->locked, s->ops_request);
e33129d8 2623}
16a53ecc 2624
1da177e4
LT		 2625/*
2626 * Each stripe/dev can have one or more bion attached.
16a53ecc 2627 * toread/towrite point to the first in a chain.
1da177e4
LT		 2628 * The bi_next chain must be in order.
2629 */
2630static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite)
2631{
2632 struct bio **bip;
d1688a6d 2633 struct r5conf *conf = sh->raid_conf;
72626685 2634 int firstwrite=0;
1da177e4 2635
cbe47ec5 2636 pr_debug("adding bi b#%llu to stripe s#%llu\n",
4f024f37 2637 (unsigned long long)bi->bi_iter.bi_sector,
1da177e4
LT		 2638 (unsigned long long)sh->sector);
2639
b17459c0
SL		 2640 /*
2641 * If several bio share a stripe. The bio bi_phys_segments acts as a
2642 * reference count to avoid race. The reference count should already be
2643 * increased before this function is called (for example, in
2644 * make_request()), so other bio sharing this stripe will not free the
2645 * stripe. If a stripe is owned by one stripe, the stripe lock will
2646 * protect it.
2647 */
2648 spin_lock_irq(&sh->stripe_lock);
72626685 2649 if (forwrite) {
1da177e4 2650 bip = &sh->dev[dd_idx].towrite;
7eaf7e8e 2651 if (*bip == NULL)
72626685
N		 2652 firstwrite = 1;
2653 } else
1da177e4 2654 bip = &sh->dev[dd_idx].toread;
4f024f37
KO		 2655 while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) {
2656 if (bio_end_sector(*bip) > bi->bi_iter.bi_sector)
1da177e4
LT		 2657 goto overlap;
2658 bip = & (*bip)->bi_next;
2659 }
4f024f37 2660 if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi))
1da177e4
LT		 2661 goto overlap;
2662
78bafebd 2663 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
1da177e4
LT		 2664 if (*bip)
2665 bi->bi_next = *bip;
2666 *bip = bi;
e7836bd6 2667 raid5_inc_bi_active_stripes(bi);
72626685 2668
1da177e4
LT		 2669 if (forwrite) {
2670 /* check if page is covered */
2671 sector_t sector = sh->dev[dd_idx].sector;
2672 for (bi=sh->dev[dd_idx].towrite;
2673 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
4f024f37 2674 bi && bi->bi_iter.bi_sector <= sector;
1da177e4 2675 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
f73a1c7d
KO		 2676 if (bio_end_sector(bi) >= sector)
2677 sector = bio_end_sector(bi);
1da177e4
LT		 2678 }
2679 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
2680 set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags);
2681 }
cbe47ec5
N		 2682
2683 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
4f024f37 2684 (unsigned long long)(*bip)->bi_iter.bi_sector,
cbe47ec5 2685 (unsigned long long)sh->sector, dd_idx);
b97390ae 2686 spin_unlock_irq(&sh->stripe_lock);
cbe47ec5
N		 2687
2688 if (conf->mddev->bitmap && firstwrite) {
2689 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
2690 STRIPE_SECTORS, 0);
2691 sh->bm_seq = conf->seq_flush+1;
2692 set_bit(STRIPE_BIT_DELAY, &sh->state);
2693 }
1da177e4
LT		 2694 return 1;
2695
2696 overlap:
2697 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
b17459c0 2698 spin_unlock_irq(&sh->stripe_lock);
1da177e4
LT		 2699 return 0;
2700}
2701
d1688a6d 2702static void end_reshape(struct r5conf *conf);
29269553 2703
d1688a6d 2704static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
911d4ee8 2705 struct stripe_head *sh)
ccfcc3c1 2706{
784052ec 2707 int sectors_per_chunk =
09c9e5fa 2708 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
911d4ee8 2709 int dd_idx;
2d2063ce 2710 int chunk_offset = sector_div(stripe, sectors_per_chunk);
112bf897 2711 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
2d2063ce 2712
112bf897
N		 2713 raid5_compute_sector(conf,
2714 stripe * (disks - conf->max_degraded)
b875e531 2715 *sectors_per_chunk + chunk_offset,
112bf897 2716 previous,
911d4ee8 2717 &dd_idx, sh);
ccfcc3c1
N		 2718}
2719
a4456856 2720static void
d1688a6d 2721handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh,
a4456856
DW		 2722 struct stripe_head_state *s, int disks,
2723 struct bio **return_bi)
2724{
2725 int i;
2726 for (i = disks; i--; ) {
2727 struct bio *bi;
2728 int bitmap_end = 0;
2729
2730 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
3cb03002 2731 struct md_rdev *rdev;
a4456856
DW		 2732 rcu_read_lock();
2733 rdev = rcu_dereference(conf->disks[i].rdev);
2734 if (rdev && test_bit(In_sync, &rdev->flags))
7f0da59b
N		 2735 atomic_inc(&rdev->nr_pending);
2736 else
2737 rdev = NULL;
a4456856 2738 rcu_read_unlock();
7f0da59b
N		 2739 if (rdev) {
2740 if (!rdev_set_badblocks(
2741 rdev,
2742 sh->sector,
2743 STRIPE_SECTORS, 0))
2744 md_error(conf->mddev, rdev);
2745 rdev_dec_pending(rdev, conf->mddev);
2746 }
a4456856 2747 }
b17459c0 2748 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 2749 /* fail all writes first */
2750 bi = sh->dev[i].towrite;
2751 sh->dev[i].towrite = NULL;
b17459c0 2752 spin_unlock_irq(&sh->stripe_lock);
1ed850f3 2753 if (bi)
a4456856 2754 bitmap_end = 1;
a4456856
DW		 2755
2756 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2757 wake_up(&conf->wait_for_overlap);
2758
4f024f37 2759 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 2760 sh->dev[i].sector + STRIPE_SECTORS) {
2761 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
2762 clear_bit(BIO_UPTODATE, &bi->bi_flags);
e7836bd6 2763 if (!raid5_dec_bi_active_stripes(bi)) {
a4456856
DW		 2764 md_write_end(conf->mddev);
2765 bi->bi_next = *return_bi;
2766 *return_bi = bi;
2767 }
2768 bi = nextbi;
2769 }
7eaf7e8e
SL		 2770 if (bitmap_end)
2771 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
2772 STRIPE_SECTORS, 0, 0);
2773 bitmap_end = 0;
a4456856
DW		 2774 /* and fail all 'written' */
2775 bi = sh->dev[i].written;
2776 sh->dev[i].written = NULL;
d592a996
SL		 2777 if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) {
2778 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
2779 sh->dev[i].page = sh->dev[i].orig_page;
2780 }
2781
a4456856 2782 if (bi) bitmap_end = 1;
4f024f37 2783 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 2784 sh->dev[i].sector + STRIPE_SECTORS) {
2785 struct bio *bi2 = 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 md_write_end(conf->mddev);
2789 bi->bi_next = *return_bi;
2790 *return_bi = bi;
2791 }
2792 bi = bi2;
2793 }
2794
b5e98d65
DW		 2795 /* fail any reads if this device is non-operational and
2796 * the data has not reached the cache yet.
2797 */
2798 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
2799 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
2800 test_bit(R5_ReadError, &sh->dev[i].flags))) {
143c4d05 2801 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 2802 bi = sh->dev[i].toread;
2803 sh->dev[i].toread = NULL;
143c4d05 2804 spin_unlock_irq(&sh->stripe_lock);
a4456856
DW		 2805 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2806 wake_up(&conf->wait_for_overlap);
4f024f37 2807 while (bi && bi->bi_iter.bi_sector <
a4456856
DW		 2808 sh->dev[i].sector + STRIPE_SECTORS) {
2809 struct bio *nextbi =
2810 r5_next_bio(bi, sh->dev[i].sector);
2811 clear_bit(BIO_UPTODATE, &bi->bi_flags);
e7836bd6 2812 if (!raid5_dec_bi_active_stripes(bi)) {
a4456856
DW		 2813 bi->bi_next = *return_bi;
2814 *return_bi = bi;
2815 }
2816 bi = nextbi;
2817 }
2818 }
a4456856
DW		 2819 if (bitmap_end)
2820 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
2821 STRIPE_SECTORS, 0, 0);
8cfa7b0f
N		 2822 /* If we were in the middle of a write the parity block might
2823 * still be locked - so just clear all R5_LOCKED flags
2824 */
2825 clear_bit(R5_LOCKED, &sh->dev[i].flags);
a4456856
DW		 2826 }
2827
8b3e6cdc
DW		 2828 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2829 if (atomic_dec_and_test(&conf->pending_full_writes))
2830 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 2831}
2832
7f0da59b 2833static void
d1688a6d 2834handle_failed_sync(struct r5conf *conf, struct stripe_head *sh,
7f0da59b
N		 2835 struct stripe_head_state *s)
2836{
2837 int abort = 0;
2838 int i;
2839
7f0da59b 2840 clear_bit(STRIPE_SYNCING, &sh->state);
f8dfcffd
N		 2841 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
2842 wake_up(&conf->wait_for_overlap);
7f0da59b 2843 s->syncing = 0;
9a3e1101 2844 s->replacing = 0;
7f0da59b 2845 /* There is nothing more to do for sync/check/repair.
18b9837e
N		 2846 * Don't even need to abort as that is handled elsewhere
2847 * if needed, and not always wanted e.g. if there is a known
2848 * bad block here.
9a3e1101 2849 * For recover/replace we need to record a bad block on all
7f0da59b
N		 2850 * non-sync devices, or abort the recovery
2851 */
18b9837e
N		 2852 if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) {
2853 /* During recovery devices cannot be removed, so
2854 * locking and refcounting of rdevs is not needed
2855 */
2856 for (i = 0; i < conf->raid_disks; i++) {
2857 struct md_rdev *rdev = conf->disks[i].rdev;
2858 if (rdev
2859 && !test_bit(Faulty, &rdev->flags)
2860 && !test_bit(In_sync, &rdev->flags)
2861 && !rdev_set_badblocks(rdev, sh->sector,
2862 STRIPE_SECTORS, 0))
2863 abort = 1;
2864 rdev = conf->disks[i].replacement;
2865 if (rdev
2866 && !test_bit(Faulty, &rdev->flags)
2867 && !test_bit(In_sync, &rdev->flags)
2868 && !rdev_set_badblocks(rdev, sh->sector,
2869 STRIPE_SECTORS, 0))
2870 abort = 1;
2871 }
2872 if (abort)
2873 conf->recovery_disabled =
2874 conf->mddev->recovery_disabled;
7f0da59b 2875 }
18b9837e 2876 md_done_sync(conf->mddev, STRIPE_SECTORS, !abort);
7f0da59b
N		 2877}
2878
9a3e1101
N		 2879static int want_replace(struct stripe_head *sh, int disk_idx)
2880{
2881 struct md_rdev *rdev;
2882 int rv = 0;
2883 /* Doing recovery so rcu locking not required */
2884 rdev = sh->raid_conf->disks[disk_idx].replacement;
2885 if (rdev
2886 && !test_bit(Faulty, &rdev->flags)
2887 && !test_bit(In_sync, &rdev->flags)
2888 && (rdev->recovery_offset <= sh->sector
2889 || rdev->mddev->recovery_cp <= sh->sector))
2890 rv = 1;
2891
2892 return rv;
2893}
2894
93b3dbce 2895/* fetch_block - checks the given member device to see if its data needs
1fe797e6
DW		 2896 * to be read or computed to satisfy a request.
2897 *
2898 * Returns 1 when no more member devices need to be checked, otherwise returns
93b3dbce 2899 * 0 to tell the loop in handle_stripe_fill to continue
f38e1219 2900 */
93b3dbce
N		 2901static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s,
2902 int disk_idx, int disks)
a4456856 2903{
5599becc 2904 struct r5dev *dev = &sh->dev[disk_idx];
f2b3b44d
N		 2905 struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]],
2906 &sh->dev[s->failed_num[1]] };
5599becc 2907
93b3dbce 2908 /* is the data in this block needed, and can we get it? */
5599becc
YT		 2909 if (!test_bit(R5_LOCKED, &dev->flags) &&
2910 !test_bit(R5_UPTODATE, &dev->flags) &&
2911 (dev->toread ||
2912 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
2913 s->syncing || s->expanding ||
9a3e1101 2914 (s->replacing && want_replace(sh, disk_idx)) ||
5d35e09c
N		 2915 (s->failed >= 1 && fdev[0]->toread) ||
2916 (s->failed >= 2 && fdev[1]->toread) ||
93b3dbce 2917 (sh->raid_conf->level <= 5 && s->failed && fdev[0]->towrite &&
67f45548 2918 (!test_bit(R5_Insync, &dev->flags) || test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) &&
93b3dbce 2919 !test_bit(R5_OVERWRITE, &fdev[0]->flags)) ||
108cef3a
N		 2920 ((sh->raid_conf->level == 6 ||
2921 sh->sector >= sh->raid_conf->mddev->recovery_cp)
2922 && s->failed && s->to_write &&
2923 (s->to_write - s->non_overwrite <
2924 sh->raid_conf->raid_disks - sh->raid_conf->max_degraded) &&
67f45548 2925 (!test_bit(R5_Insync, &dev->flags) || test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))))) {
5599becc
YT		 2926 /* we would like to get this block, possibly by computing it,
2927 * otherwise read it if the backing disk is insync
2928 */
2929 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
2930 BUG_ON(test_bit(R5_Wantread, &dev->flags));
2931 if ((s->uptodate == disks - 1) &&
f2b3b44d
N		 2932 (s->failed && (disk_idx == s->failed_num[0] ||
2933 disk_idx == s->failed_num[1]))) {
5599becc
YT		 2934 /* have disk failed, and we're requested to fetch it;
2935 * do compute it
a4456856 2936 */
5599becc
YT		 2937 pr_debug("Computing stripe %llu block %d\n",
2938 (unsigned long long)sh->sector, disk_idx);
2939 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2940 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2941 set_bit(R5_Wantcompute, &dev->flags);
2942 sh->ops.target = disk_idx;
2943 sh->ops.target2 = -1; /* no 2nd target */
2944 s->req_compute = 1;
93b3dbce
N		 2945 /* Careful: from this point on 'uptodate' is in the eye
2946 * of raid_run_ops which services 'compute' operations
2947 * before writes. R5_Wantcompute flags a block that will
2948 * be R5_UPTODATE by the time it is needed for a
2949 * subsequent operation.
2950 */
5599becc
YT		 2951 s->uptodate++;
2952 return 1;
2953 } else if (s->uptodate == disks-2 && s->failed >= 2) {
2954 /* Computing 2-failure is *very* expensive; only
2955 * do it if failed >= 2
2956 */
2957 int other;
2958 for (other = disks; other--; ) {
2959 if (other == disk_idx)
2960 continue;
2961 if (!test_bit(R5_UPTODATE,
2962 &sh->dev[other].flags))
2963 break;
a4456856 2964 }
5599becc
YT		 2965 BUG_ON(other < 0);
2966 pr_debug("Computing stripe %llu blocks %d,%d\n",
2967 (unsigned long long)sh->sector,
2968 disk_idx, other);
2969 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2970 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2971 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
2972 set_bit(R5_Wantcompute, &sh->dev[other].flags);
2973 sh->ops.target = disk_idx;
2974 sh->ops.target2 = other;
2975 s->uptodate += 2;
2976 s->req_compute = 1;
2977 return 1;
2978 } else if (test_bit(R5_Insync, &dev->flags)) {
2979 set_bit(R5_LOCKED, &dev->flags);
2980 set_bit(R5_Wantread, &dev->flags);
2981 s->locked++;
2982 pr_debug("Reading block %d (sync=%d)\n",
2983 disk_idx, s->syncing);
a4456856
DW		 2984 }
2985 }
5599becc
YT		 2986
2987 return 0;
2988}
2989
2990/**
93b3dbce 2991 * handle_stripe_fill - read or compute data to satisfy pending requests.
5599becc 2992 */
93b3dbce
N		 2993static void handle_stripe_fill(struct stripe_head *sh,
2994 struct stripe_head_state *s,
2995 int disks)
5599becc
YT		 2996{
2997 int i;
2998
2999 /* look for blocks to read/compute, skip this if a compute
3000 * is already in flight, or if the stripe contents are in the
3001 * midst of changing due to a write
3002 */
3003 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
3004 !sh->reconstruct_state)
3005 for (i = disks; i--; )
93b3dbce 3006 if (fetch_block(sh, s, i, disks))
5599becc 3007 break;
a4456856
DW		 3008 set_bit(STRIPE_HANDLE, &sh->state);
3009}
3010
1fe797e6 3011/* handle_stripe_clean_event
a4456856
DW		 3012 * any written block on an uptodate or failed drive can be returned.
3013 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
3014 * never LOCKED, so we don't need to test 'failed' directly.
3015 */
d1688a6d 3016static void handle_stripe_clean_event(struct r5conf *conf,
a4456856
DW		 3017 struct stripe_head *sh, int disks, struct bio **return_bi)
3018{
3019 int i;
3020 struct r5dev *dev;
f8dfcffd 3021 int discard_pending = 0;
a4456856
DW		 3022
3023 for (i = disks; i--; )
3024 if (sh->dev[i].written) {
3025 dev = &sh->dev[i];
3026 if (!test_bit(R5_LOCKED, &dev->flags) &&
9e444768 3027 (test_bit(R5_UPTODATE, &dev->flags) ||
d592a996
SL		 3028 test_bit(R5_Discard, &dev->flags) ||
3029 test_bit(R5_SkipCopy, &dev->flags))) {
a4456856
DW		 3030 /* We can return any write requests */
3031 struct bio *wbi, *wbi2;
45b4233c 3032 pr_debug("Return write for disc %d\n", i);
ca64cae9
N		 3033 if (test_and_clear_bit(R5_Discard, &dev->flags))
3034 clear_bit(R5_UPTODATE, &dev->flags);
d592a996
SL		 3035 if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) {
3036 WARN_ON(test_bit(R5_UPTODATE, &dev->flags));
3037 dev->page = dev->orig_page;
3038 }
a4456856
DW		 3039 wbi = dev->written;
3040 dev->written = NULL;
4f024f37 3041 while (wbi && wbi->bi_iter.bi_sector <
a4456856
DW		 3042 dev->sector + STRIPE_SECTORS) {
3043 wbi2 = r5_next_bio(wbi, dev->sector);
e7836bd6 3044 if (!raid5_dec_bi_active_stripes(wbi)) {
a4456856
DW		 3045 md_write_end(conf->mddev);
3046 wbi->bi_next = *return_bi;
3047 *return_bi = wbi;
3048 }
3049 wbi = wbi2;
3050 }
7eaf7e8e
SL		 3051 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3052 STRIPE_SECTORS,
a4456856 3053 !test_bit(STRIPE_DEGRADED, &sh->state),
7eaf7e8e 3054 0);
f8dfcffd
N		 3055 } else if (test_bit(R5_Discard, &dev->flags))
3056 discard_pending = 1;
d592a996
SL		 3057 WARN_ON(test_bit(R5_SkipCopy, &dev->flags));
3058 WARN_ON(dev->page != dev->orig_page);
f8dfcffd
N		 3059 }
3060 if (!discard_pending &&
3061 test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) {
3062 clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
3063 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
3064 if (sh->qd_idx >= 0) {
3065 clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
3066 clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags);
3067 }
3068 /* now that discard is done we can proceed with any sync */
3069 clear_bit(STRIPE_DISCARD, &sh->state);
d47648fc
SL		 3070 /*
3071 * SCSI discard will change some bio fields and the stripe has
3072 * no updated data, so remove it from hash list and the stripe
3073 * will be reinitialized
3074 */
3075 spin_lock_irq(&conf->device_lock);
3076 remove_hash(sh);
3077 spin_unlock_irq(&conf->device_lock);
f8dfcffd
N		 3078 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state))
3079 set_bit(STRIPE_HANDLE, &sh->state);
3080
3081 }
8b3e6cdc
DW		 3082
3083 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3084 if (atomic_dec_and_test(&conf->pending_full_writes))
3085 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 3086}
3087
d1688a6d 3088static void handle_stripe_dirtying(struct r5conf *conf,
c8ac1803
N		 3089 struct stripe_head *sh,
3090 struct stripe_head_state *s,
3091 int disks)
a4456856
DW		 3092{
3093 int rmw = 0, rcw = 0, i;
a7854487
AL		 3094 sector_t recovery_cp = conf->mddev->recovery_cp;
3095
3096 /* RAID6 requires 'rcw' in current implementation.
3097 * Otherwise, check whether resync is now happening or should start.
3098 * If yes, then the array is dirty (after unclean shutdown or
3099 * initial creation), so parity in some stripes might be inconsistent.
3100 * In this case, we need to always do reconstruct-write, to ensure
3101 * that in case of drive failure or read-error correction, we
3102 * generate correct data from the parity.
3103 */
3104 if (conf->max_degraded == 2 ||
3105 (recovery_cp < MaxSector && sh->sector >= recovery_cp)) {
3106 /* Calculate the real rcw later - for now make it
c8ac1803
N		 3107 * look like rcw is cheaper
3108 */
3109 rcw = 1; rmw = 2;
a7854487
AL		 3110 pr_debug("force RCW max_degraded=%u, recovery_cp=%llu sh->sector=%llu\n",
3111 conf->max_degraded, (unsigned long long)recovery_cp,
3112 (unsigned long long)sh->sector);
c8ac1803 3113 } else for (i = disks; i--; ) {
a4456856
DW		 3114 /* would I have to read this buffer for read_modify_write */
3115 struct r5dev *dev = &sh->dev[i];
3116 if ((dev->towrite || i == sh->pd_idx) &&
3117 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 3118 !(test_bit(R5_UPTODATE, &dev->flags) ||
3119 test_bit(R5_Wantcompute, &dev->flags))) {
a4456856
DW		 3120 if (test_bit(R5_Insync, &dev->flags))
3121 rmw++;
3122 else
3123 rmw += 2*disks; /* cannot read it */
3124 }
3125 /* Would I have to read this buffer for reconstruct_write */
3126 if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx &&
3127 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 3128 !(test_bit(R5_UPTODATE, &dev->flags) ||
3129 test_bit(R5_Wantcompute, &dev->flags))) {
67f45548
N		 3130 if (test_bit(R5_Insync, &dev->flags))
3131 rcw++;
a4456856
DW		 3132 else
3133 rcw += 2*disks;
3134 }
3135 }
45b4233c 3136 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
a4456856
DW		 3137 (unsigned long long)sh->sector, rmw, rcw);
3138 set_bit(STRIPE_HANDLE, &sh->state);
a9add5d9 3139 if (rmw < rcw && rmw > 0) {
a4456856 3140 /* prefer read-modify-write, but need to get some data */
e3620a3a
JB		 3141 if (conf->mddev->queue)
3142 blk_add_trace_msg(conf->mddev->queue,
3143 "raid5 rmw %llu %d",
3144 (unsigned long long)sh->sector, rmw);
a4456856
DW		 3145 for (i = disks; i--; ) {
3146 struct r5dev *dev = &sh->dev[i];
3147 if ((dev->towrite || i == sh->pd_idx) &&
3148 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 3149 !(test_bit(R5_UPTODATE, &dev->flags) ||
3150 test_bit(R5_Wantcompute, &dev->flags)) &&
a4456856 3151 test_bit(R5_Insync, &dev->flags)) {
67f45548
N		 3152 if (test_bit(STRIPE_PREREAD_ACTIVE,
3153 &sh->state)) {
3154 pr_debug("Read_old block %d for r-m-w\n",
3155 i);
a4456856
DW		 3156 set_bit(R5_LOCKED, &dev->flags);
3157 set_bit(R5_Wantread, &dev->flags);
3158 s->locked++;
3159 } else {
3160 set_bit(STRIPE_DELAYED, &sh->state);
3161 set_bit(STRIPE_HANDLE, &sh->state);
3162 }
3163 }
3164 }
a9add5d9 3165 }
c8ac1803 3166 if (rcw <= rmw && rcw > 0) {
a4456856 3167 /* want reconstruct write, but need to get some data */
a9add5d9 3168 int qread =0;
c8ac1803 3169 rcw = 0;
a4456856
DW		 3170 for (i = disks; i--; ) {
3171 struct r5dev *dev = &sh->dev[i];
3172 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
c8ac1803 3173 i != sh->pd_idx && i != sh->qd_idx &&
a4456856 3174 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219 3175 !(test_bit(R5_UPTODATE, &dev->flags) ||
c8ac1803
N		 3176 test_bit(R5_Wantcompute, &dev->flags))) {
3177 rcw++;
67f45548
N		 3178 if (test_bit(R5_Insync, &dev->flags) &&
3179 test_bit(STRIPE_PREREAD_ACTIVE,
3180 &sh->state)) {
45b4233c 3181 pr_debug("Read_old block "
a4456856
DW		 3182 "%d for Reconstruct\n", i);
3183 set_bit(R5_LOCKED, &dev->flags);
3184 set_bit(R5_Wantread, &dev->flags);
3185 s->locked++;
a9add5d9 3186 qread++;
a4456856
DW		 3187 } else {
3188 set_bit(STRIPE_DELAYED, &sh->state);
3189 set_bit(STRIPE_HANDLE, &sh->state);
3190 }
3191 }
3192 }
e3620a3a 3193 if (rcw && conf->mddev->queue)
a9add5d9
N		 3194 blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d",
3195 (unsigned long long)sh->sector,
3196 rcw, qread, test_bit(STRIPE_DELAYED, &sh->state));
c8ac1803 3197 }
b1b02fe9
N		 3198
3199 if (rcw > disks && rmw > disks &&
3200 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3201 set_bit(STRIPE_DELAYED, &sh->state);
3202
a4456856
DW		 3203 /* now if nothing is locked, and if we have enough data,
3204 * we can start a write request
3205 */
f38e1219
DW		 3206 /* since handle_stripe can be called at any time we need to handle the
3207 * case where a compute block operation has been submitted and then a
ac6b53b6
DW		 3208 * subsequent call wants to start a write request. raid_run_ops only
3209 * handles the case where compute block and reconstruct are requested
f38e1219
DW		 3210 * simultaneously. If this is not the case then new writes need to be
3211 * held off until the compute completes.
3212 */
976ea8d4
DW		 3213 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
3214 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
3215 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
c0f7bddb 3216 schedule_reconstruction(sh, s, rcw == 0, 0);
a4456856
DW		 3217}
3218
d1688a6d 3219static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh,
a4456856
DW		 3220 struct stripe_head_state *s, int disks)
3221{
ecc65c9b 3222 struct r5dev *dev = NULL;
bd2ab670 3223
a4456856 3224 set_bit(STRIPE_HANDLE, &sh->state);
e89f8962 3225
ecc65c9b
DW		 3226 switch (sh->check_state) {
3227 case check_state_idle:
3228 /* start a new check operation if there are no failures */
bd2ab670 3229 if (s->failed == 0) {
bd2ab670 3230 BUG_ON(s->uptodate != disks);
ecc65c9b
DW		 3231 sh->check_state = check_state_run;
3232 set_bit(STRIPE_OP_CHECK, &s->ops_request);
bd2ab670 3233 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
bd2ab670 3234 s->uptodate--;
ecc65c9b 3235 break;
bd2ab670 3236 }
f2b3b44d 3237 dev = &sh->dev[s->failed_num[0]];
ecc65c9b
DW		 3238 /* fall through */
3239 case check_state_compute_result:
3240 sh->check_state = check_state_idle;
3241 if (!dev)
3242 dev = &sh->dev[sh->pd_idx];
3243
3244 /* check that a write has not made the stripe insync */
3245 if (test_bit(STRIPE_INSYNC, &sh->state))
3246 break;
c8894419 3247
a4456856 3248 /* either failed parity check, or recovery is happening */
a4456856
DW		 3249 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
3250 BUG_ON(s->uptodate != disks);
3251
3252 set_bit(R5_LOCKED, &dev->flags);
ecc65c9b 3253 s->locked++;
a4456856 3254 set_bit(R5_Wantwrite, &dev->flags);
830ea016 3255
a4456856 3256 clear_bit(STRIPE_DEGRADED, &sh->state);
a4456856 3257 set_bit(STRIPE_INSYNC, &sh->state);
ecc65c9b
DW		 3258 break;
3259 case check_state_run:
3260 break; /* we will be called again upon completion */
3261 case check_state_check_result:
3262 sh->check_state = check_state_idle;
3263
3264 /* if a failure occurred during the check operation, leave
3265 * STRIPE_INSYNC not set and let the stripe be handled again
3266 */
3267 if (s->failed)
3268 break;
3269
3270 /* handle a successful check operation, if parity is correct
3271 * we are done. Otherwise update the mismatch count and repair
3272 * parity if !MD_RECOVERY_CHECK
3273 */
ad283ea4 3274 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
ecc65c9b
DW		 3275 /* parity is correct (on disc,
3276 * not in buffer any more)
3277 */
3278 set_bit(STRIPE_INSYNC, &sh->state);
3279 else {
7f7583d4 3280 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
ecc65c9b
DW		 3281 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
3282 /* don't try to repair!! */
3283 set_bit(STRIPE_INSYNC, &sh->state);
3284 else {
3285 sh->check_state = check_state_compute_run;
976ea8d4 3286 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
ecc65c9b
DW		 3287 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3288 set_bit(R5_Wantcompute,
3289 &sh->dev[sh->pd_idx].flags);
3290 sh->ops.target = sh->pd_idx;
ac6b53b6 3291 sh->ops.target2 = -1;
ecc65c9b
DW		 3292 s->uptodate++;
3293 }
3294 }
3295 break;
3296 case check_state_compute_run:
3297 break;
3298 default:
3299 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
3300 __func__, sh->check_state,
3301 (unsigned long long) sh->sector);
3302 BUG();
a4456856
DW		 3303 }
3304}
3305
d1688a6d 3306static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh,
36d1c647 3307 struct stripe_head_state *s,
f2b3b44d 3308 int disks)
a4456856 3309{
a4456856 3310 int pd_idx = sh->pd_idx;
34e04e87 3311 int qd_idx = sh->qd_idx;
d82dfee0 3312 struct r5dev *dev;
a4456856
DW		 3313
3314 set_bit(STRIPE_HANDLE, &sh->state);
3315
3316 BUG_ON(s->failed > 2);
d82dfee0 3317
a4456856
DW		 3318 /* Want to check and possibly repair P and Q.
3319 * However there could be one 'failed' device, in which
3320 * case we can only check one of them, possibly using the
3321 * other to generate missing data
3322 */
3323
d82dfee0
DW		 3324 switch (sh->check_state) {
3325 case check_state_idle:
3326 /* start a new check operation if there are < 2 failures */
f2b3b44d 3327 if (s->failed == s->q_failed) {
d82dfee0 3328 /* The only possible failed device holds Q, so it
a4456856
DW		 3329 * makes sense to check P (If anything else were failed,
3330 * we would have used P to recreate it).
3331 */
d82dfee0 3332 sh->check_state = check_state_run;
a4456856 3333 }
f2b3b44d 3334 if (!s->q_failed && s->failed < 2) {
d82dfee0 3335 /* Q is not failed, and we didn't use it to generate
a4456856
DW		 3336 * anything, so it makes sense to check it
3337 */
d82dfee0
DW		 3338 if (sh->check_state == check_state_run)
3339 sh->check_state = check_state_run_pq;
3340 else
3341 sh->check_state = check_state_run_q;
a4456856 3342 }
a4456856 3343
d82dfee0
DW		 3344 /* discard potentially stale zero_sum_result */
3345 sh->ops.zero_sum_result = 0;
a4456856 3346
d82dfee0
DW		 3347 if (sh->check_state == check_state_run) {
3348 /* async_xor_zero_sum destroys the contents of P */
3349 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
3350 s->uptodate--;
a4456856 3351 }
d82dfee0
DW		 3352 if (sh->check_state >= check_state_run &&
3353 sh->check_state <= check_state_run_pq) {
3354 /* async_syndrome_zero_sum preserves P and Q, so
3355 * no need to mark them !uptodate here
3356 */
3357 set_bit(STRIPE_OP_CHECK, &s->ops_request);
3358 break;
a4456856
DW		 3359 }
3360
d82dfee0
DW		 3361 /* we have 2-disk failure */
3362 BUG_ON(s->failed != 2);
3363 /* fall through */
3364 case check_state_compute_result:
3365 sh->check_state = check_state_idle;
a4456856 3366
d82dfee0
DW		 3367 /* check that a write has not made the stripe insync */
3368 if (test_bit(STRIPE_INSYNC, &sh->state))
3369 break;
a4456856
DW		 3370
3371 /* now write out any block on a failed drive,
d82dfee0 3372 * or P or Q if they were recomputed
a4456856 3373 */
d82dfee0 3374 BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */
a4456856 3375 if (s->failed == 2) {
f2b3b44d 3376 dev = &sh->dev[s->failed_num[1]];
a4456856
DW		 3377 s->locked++;
3378 set_bit(R5_LOCKED, &dev->flags);
3379 set_bit(R5_Wantwrite, &dev->flags);
3380 }
3381 if (s->failed >= 1) {
f2b3b44d 3382 dev = &sh->dev[s->failed_num[0]];
a4456856
DW		 3383 s->locked++;
3384 set_bit(R5_LOCKED, &dev->flags);
3385 set_bit(R5_Wantwrite, &dev->flags);
3386 }
d82dfee0 3387 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
a4456856
DW		 3388 dev = &sh->dev[pd_idx];
3389 s->locked++;
3390 set_bit(R5_LOCKED, &dev->flags);
3391 set_bit(R5_Wantwrite, &dev->flags);
3392 }
d82dfee0 3393 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
a4456856
DW		 3394 dev = &sh->dev[qd_idx];
3395 s->locked++;
3396 set_bit(R5_LOCKED, &dev->flags);
3397 set_bit(R5_Wantwrite, &dev->flags);
3398 }
3399 clear_bit(STRIPE_DEGRADED, &sh->state);
3400
3401 set_bit(STRIPE_INSYNC, &sh->state);
d82dfee0
DW		 3402 break;
3403 case check_state_run:
3404 case check_state_run_q:
3405 case check_state_run_pq:
3406 break; /* we will be called again upon completion */
3407 case check_state_check_result:
3408 sh->check_state = check_state_idle;
3409
3410 /* handle a successful check operation, if parity is correct
3411 * we are done. Otherwise update the mismatch count and repair
3412 * parity if !MD_RECOVERY_CHECK
3413 */
3414 if (sh->ops.zero_sum_result == 0) {
3415 /* both parities are correct */
3416 if (!s->failed)
3417 set_bit(STRIPE_INSYNC, &sh->state);
3418 else {
3419 /* in contrast to the raid5 case we can validate
3420 * parity, but still have a failure to write
3421 * back
3422 */
3423 sh->check_state = check_state_compute_result;
3424 /* Returning at this point means that we may go
3425 * off and bring p and/or q uptodate again so
3426 * we make sure to check zero_sum_result again
3427 * to verify if p or q need writeback
3428 */
3429 }
3430 } else {
7f7583d4 3431 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
d82dfee0
DW		 3432 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
3433 /* don't try to repair!! */
3434 set_bit(STRIPE_INSYNC, &sh->state);
3435 else {
3436 int *target = &sh->ops.target;
3437
3438 sh->ops.target = -1;
3439 sh->ops.target2 = -1;
3440 sh->check_state = check_state_compute_run;
3441 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3442 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3443 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
3444 set_bit(R5_Wantcompute,
3445 &sh->dev[pd_idx].flags);
3446 *target = pd_idx;
3447 target = &sh->ops.target2;
3448 s->uptodate++;
3449 }
3450 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
3451 set_bit(R5_Wantcompute,
3452 &sh->dev[qd_idx].flags);
3453 *target = qd_idx;
3454 s->uptodate++;
3455 }
3456 }
3457 }
3458 break;
3459 case check_state_compute_run:
3460 break;
3461 default:
3462 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
3463 __func__, sh->check_state,
3464 (unsigned long long) sh->sector);
3465 BUG();
a4456856
DW		 3466 }
3467}
3468
d1688a6d 3469static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh)
a4456856
DW		 3470{
3471 int i;
3472
3473 /* We have read all the blocks in this stripe and now we need to
3474 * copy some of them into a target stripe for expand.
3475 */
f0a50d37 3476 struct dma_async_tx_descriptor *tx = NULL;
a4456856
DW		 3477 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3478 for (i = 0; i < sh->disks; i++)
34e04e87 3479 if (i != sh->pd_idx && i != sh->qd_idx) {
911d4ee8 3480 int dd_idx, j;
a4456856 3481 struct stripe_head *sh2;
a08abd8c 3482 struct async_submit_ctl submit;
a4456856 3483
784052ec 3484 sector_t bn = compute_blocknr(sh, i, 1);
911d4ee8
N		 3485 sector_t s = raid5_compute_sector(conf, bn, 0,
3486 &dd_idx, NULL);
a8c906ca 3487 sh2 = get_active_stripe(conf, s, 0, 1, 1);
a4456856
DW		 3488 if (sh2 == NULL)
3489 /* so far only the early blocks of this stripe
3490 * have been requested. When later blocks
3491 * get requested, we will try again
3492 */
3493 continue;
3494 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
3495 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
3496 /* must have already done this block */
3497 release_stripe(sh2);
3498 continue;
3499 }
f0a50d37
DW		 3500
3501 /* place all the copies on one channel */
a08abd8c 3502 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
f0a50d37 3503 tx = async_memcpy(sh2->dev[dd_idx].page,
88ba2aa5 3504 sh->dev[i].page, 0, 0, STRIPE_SIZE,
a08abd8c 3505 &submit);
f0a50d37 3506
a4456856
DW		 3507 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
3508 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
3509 for (j = 0; j < conf->raid_disks; j++)
3510 if (j != sh2->pd_idx &&
86c374ba 3511 j != sh2->qd_idx &&
a4456856
DW		 3512 !test_bit(R5_Expanded, &sh2->dev[j].flags))
3513 break;
3514 if (j == conf->raid_disks) {
3515 set_bit(STRIPE_EXPAND_READY, &sh2->state);
3516 set_bit(STRIPE_HANDLE, &sh2->state);
3517 }
3518 release_stripe(sh2);
f0a50d37 3519
a4456856 3520 }
a2e08551 3521 /* done submitting copies, wait for them to complete */
749586b7 3522 async_tx_quiesce(&tx);
a4456856 3523}
1da177e4
LT		 3524
3525/*
3526 * handle_stripe - do things to a stripe.
3527 *
9a3e1101
N		 3528 * We lock the stripe by setting STRIPE_ACTIVE and then examine the
3529 * state of various bits to see what needs to be done.
1da177e4 3530 * Possible results:
9a3e1101
N		 3531 * return some read requests which now have data
3532 * return some write requests which are safely on storage
1da177e4
LT		 3533 * schedule a read on some buffers
3534 * schedule a write of some buffers
3535 * return confirmation of parity correctness
3536 *
1da177e4 3537 */
a4456856 3538
acfe726b 3539static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s)
1da177e4 3540{
d1688a6d 3541 struct r5conf *conf = sh->raid_conf;
f416885e 3542 int disks = sh->disks;
474af965
N		 3543 struct r5dev *dev;
3544 int i;
9a3e1101 3545 int do_recovery = 0;
1da177e4 3546
acfe726b
N		 3547 memset(s, 0, sizeof(*s));
3548
acfe726b
N		 3549 s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3550 s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
3551 s->failed_num[0] = -1;
3552 s->failed_num[1] = -1;
1da177e4 3553
acfe726b 3554 /* Now to look around and see what can be done */
1da177e4 3555 rcu_read_lock();
16a53ecc 3556 for (i=disks; i--; ) {
3cb03002 3557 struct md_rdev *rdev;
31c176ec
N		 3558 sector_t first_bad;
3559 int bad_sectors;
3560 int is_bad = 0;
acfe726b 3561
16a53ecc 3562 dev = &sh->dev[i];
1da177e4 3563
45b4233c 3564 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
9a3e1101
N		 3565 i, dev->flags,
3566 dev->toread, dev->towrite, dev->written);
6c0069c0
YT		 3567 /* maybe we can reply to a read
3568 *
3569 * new wantfill requests are only permitted while
3570 * ops_complete_biofill is guaranteed to be inactive
3571 */
3572 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
3573 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
3574 set_bit(R5_Wantfill, &dev->flags);
1da177e4 3575
16a53ecc 3576 /* now count some things */
cc94015a
N		 3577 if (test_bit(R5_LOCKED, &dev->flags))
3578 s->locked++;
3579 if (test_bit(R5_UPTODATE, &dev->flags))
3580 s->uptodate++;
2d6e4ecc 3581 if (test_bit(R5_Wantcompute, &dev->flags)) {
cc94015a
N		 3582 s->compute++;
3583 BUG_ON(s->compute > 2);
2d6e4ecc 3584 }
1da177e4 3585
acfe726b 3586 if (test_bit(R5_Wantfill, &dev->flags))
cc94015a 3587 s->to_fill++;
acfe726b 3588 else if (dev->toread)
cc94015a 3589 s->to_read++;
16a53ecc 3590 if (dev->towrite) {
cc94015a 3591 s->to_write++;
16a53ecc 3592 if (!test_bit(R5_OVERWRITE, &dev->flags))
cc94015a 3593 s->non_overwrite++;
16a53ecc 3594 }
a4456856 3595 if (dev->written)
cc94015a 3596 s->written++;
14a75d3e
N		 3597 /* Prefer to use the replacement for reads, but only
3598 * if it is recovered enough and has no bad blocks.
3599 */
3600 rdev = rcu_dereference(conf->disks[i].replacement);
3601 if (rdev && !test_bit(Faulty, &rdev->flags) &&
3602 rdev->recovery_offset >= sh->sector + STRIPE_SECTORS &&
3603 !is_badblock(rdev, sh->sector, STRIPE_SECTORS,
3604 &first_bad, &bad_sectors))
3605 set_bit(R5_ReadRepl, &dev->flags);
3606 else {
9a3e1101
N		 3607 if (rdev)
3608 set_bit(R5_NeedReplace, &dev->flags);
14a75d3e
N		 3609 rdev = rcu_dereference(conf->disks[i].rdev);
3610 clear_bit(R5_ReadRepl, &dev->flags);
3611 }
9283d8c5
N		 3612 if (rdev && test_bit(Faulty, &rdev->flags))
3613 rdev = NULL;
31c176ec
N		 3614 if (rdev) {
3615 is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
3616 &first_bad, &bad_sectors);
3617 if (s->blocked_rdev == NULL
3618 && (test_bit(Blocked, &rdev->flags)
3619 || is_bad < 0)) {
3620 if (is_bad < 0)
3621 set_bit(BlockedBadBlocks,
3622 &rdev->flags);
3623 s->blocked_rdev = rdev;
3624 atomic_inc(&rdev->nr_pending);
3625 }
6bfe0b49 3626 }
415e72d0
N		 3627 clear_bit(R5_Insync, &dev->flags);
3628 if (!rdev)
3629 /* Not in-sync */;
31c176ec
N		 3630 else if (is_bad) {
3631 /* also not in-sync */
18b9837e
N		 3632 if (!test_bit(WriteErrorSeen, &rdev->flags) &&
3633 test_bit(R5_UPTODATE, &dev->flags)) {
31c176ec
N		 3634 /* treat as in-sync, but with a read error
3635 * which we can now try to correct
3636 */
3637 set_bit(R5_Insync, &dev->flags);
3638 set_bit(R5_ReadError, &dev->flags);
3639 }
3640 } else if (test_bit(In_sync, &rdev->flags))
415e72d0 3641 set_bit(R5_Insync, &dev->flags);
30d7a483 3642 else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
415e72d0 3643 /* in sync if before recovery_offset */
30d7a483
N		 3644 set_bit(R5_Insync, &dev->flags);
3645 else if (test_bit(R5_UPTODATE, &dev->flags) &&
3646 test_bit(R5_Expanded, &dev->flags))
3647 /* If we've reshaped into here, we assume it is Insync.
3648 * We will shortly update recovery_offset to make
3649 * it official.
3650 */
3651 set_bit(R5_Insync, &dev->flags);
3652
1cc03eb9 3653 if (test_bit(R5_WriteError, &dev->flags)) {
14a75d3e
N		 3654 /* This flag does not apply to '.replacement'
3655 * only to .rdev, so make sure to check that*/
3656 struct md_rdev *rdev2 = rcu_dereference(
3657 conf->disks[i].rdev);
3658 if (rdev2 == rdev)
3659 clear_bit(R5_Insync, &dev->flags);
3660 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
bc2607f3 3661 s->handle_bad_blocks = 1;
14a75d3e 3662 atomic_inc(&rdev2->nr_pending);
bc2607f3
N		 3663 } else
3664 clear_bit(R5_WriteError, &dev->flags);
3665 }
1cc03eb9 3666 if (test_bit(R5_MadeGood, &dev->flags)) {
14a75d3e
N		 3667 /* This flag does not apply to '.replacement'
3668 * only to .rdev, so make sure to check that*/
3669 struct md_rdev *rdev2 = rcu_dereference(
3670 conf->disks[i].rdev);
3671 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
b84db560 3672 s->handle_bad_blocks = 1;
14a75d3e 3673 atomic_inc(&rdev2->nr_pending);
b84db560
N		 3674 } else
3675 clear_bit(R5_MadeGood, &dev->flags);
3676 }
977df362
N		 3677 if (test_bit(R5_MadeGoodRepl, &dev->flags)) {
3678 struct md_rdev *rdev2 = rcu_dereference(
3679 conf->disks[i].replacement);
3680 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
3681 s->handle_bad_blocks = 1;
3682 atomic_inc(&rdev2->nr_pending);
3683 } else
3684 clear_bit(R5_MadeGoodRepl, &dev->flags);
3685 }
415e72d0 3686 if (!test_bit(R5_Insync, &dev->flags)) {
16a53ecc
N		 3687 /* The ReadError flag will just be confusing now */
3688 clear_bit(R5_ReadError, &dev->flags);
3689 clear_bit(R5_ReWrite, &dev->flags);
1da177e4 3690 }
415e72d0
N		 3691 if (test_bit(R5_ReadError, &dev->flags))
3692 clear_bit(R5_Insync, &dev->flags);
3693 if (!test_bit(R5_Insync, &dev->flags)) {
cc94015a
N		 3694 if (s->failed < 2)
3695 s->failed_num[s->failed] = i;
3696 s->failed++;
9a3e1101
N		 3697 if (rdev && !test_bit(Faulty, &rdev->flags))
3698 do_recovery = 1;
415e72d0 3699 }
1da177e4 3700 }
9a3e1101
N		 3701 if (test_bit(STRIPE_SYNCING, &sh->state)) {
3702 /* If there is a failed device being replaced,
3703 * we must be recovering.
3704 * else if we are after recovery_cp, we must be syncing
c6d2e084 3705 * else if MD_RECOVERY_REQUESTED is set, we also are syncing.
9a3e1101
N		 3706 * else we can only be replacing
3707 * sync and recovery both need to read all devices, and so
3708 * use the same flag.
3709 */
3710 if (do_recovery ||
c6d2e084 3711 sh->sector >= conf->mddev->recovery_cp ||
3712 test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery)))
9a3e1101
N		 3713 s->syncing = 1;
3714 else
3715 s->replacing = 1;
3716 }
1da177e4 3717 rcu_read_unlock();
cc94015a
N		 3718}
3719
3720static void handle_stripe(struct stripe_head *sh)
3721{
3722 struct stripe_head_state s;
d1688a6d 3723 struct r5conf *conf = sh->raid_conf;
3687c061 3724 int i;
84789554
N		 3725 int prexor;
3726 int disks = sh->disks;
474af965 3727 struct r5dev *pdev, *qdev;
cc94015a
N		 3728
3729 clear_bit(STRIPE_HANDLE, &sh->state);
257a4b42 3730 if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) {
cc94015a
N		 3731 /* already being handled, ensure it gets handled
3732 * again when current action finishes */
3733 set_bit(STRIPE_HANDLE, &sh->state);
3734 return;
3735 }
3736
f8dfcffd
N		 3737 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
3738 spin_lock(&sh->stripe_lock);
3739 /* Cannot process 'sync' concurrently with 'discard' */
3740 if (!test_bit(STRIPE_DISCARD, &sh->state) &&
3741 test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
3742 set_bit(STRIPE_SYNCING, &sh->state);
3743 clear_bit(STRIPE_INSYNC, &sh->state);
f94c0b66 3744 clear_bit(STRIPE_REPLACED, &sh->state);
f8dfcffd
N		 3745 }
3746 spin_unlock(&sh->stripe_lock);
cc94015a
N		 3747 }
3748 clear_bit(STRIPE_DELAYED, &sh->state);
3749
3750 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
3751 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
3752 (unsigned long long)sh->sector, sh->state,
3753 atomic_read(&sh->count), sh->pd_idx, sh->qd_idx,
3754 sh->check_state, sh->reconstruct_state);
3687c061 3755
acfe726b 3756 analyse_stripe(sh, &s);
c5a31000 3757
bc2607f3
N		 3758 if (s.handle_bad_blocks) {
3759 set_bit(STRIPE_HANDLE, &sh->state);
3760 goto finish;
3761 }
3762
474af965
N		 3763 if (unlikely(s.blocked_rdev)) {
3764 if (s.syncing || s.expanding || s.expanded ||
9a3e1101 3765 s.replacing || s.to_write || s.written) {
474af965
N		 3766 set_bit(STRIPE_HANDLE, &sh->state);
3767 goto finish;
3768 }
3769 /* There is nothing for the blocked_rdev to block */
3770 rdev_dec_pending(s.blocked_rdev, conf->mddev);
3771 s.blocked_rdev = NULL;
3772 }
3773
3774 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
3775 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
3776 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
3777 }
3778
3779 pr_debug("locked=%d uptodate=%d to_read=%d"
3780 " to_write=%d failed=%d failed_num=%d,%d\n",
3781 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
3782 s.failed_num[0], s.failed_num[1]);
3783 /* check if the array has lost more than max_degraded devices and,
3784 * if so, some requests might need to be failed.
3785 */
9a3f530f
N		 3786 if (s.failed > conf->max_degraded) {
3787 sh->check_state = 0;
3788 sh->reconstruct_state = 0;
3789 if (s.to_read+s.to_write+s.written)
3790 handle_failed_stripe(conf, sh, &s, disks, &s.return_bi);
9a3e1101 3791 if (s.syncing + s.replacing)
9a3f530f
N		 3792 handle_failed_sync(conf, sh, &s);
3793 }
474af965 3794
84789554
N		 3795 /* Now we check to see if any write operations have recently
3796 * completed
3797 */
3798 prexor = 0;
3799 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
3800 prexor = 1;
3801 if (sh->reconstruct_state == reconstruct_state_drain_result ||
3802 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
3803 sh->reconstruct_state = reconstruct_state_idle;
3804
3805 /* All the 'written' buffers and the parity block are ready to
3806 * be written back to disk
3807 */
9e444768
SL		 3808 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) &&
3809 !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags));
84789554 3810 BUG_ON(sh->qd_idx >= 0 &&
9e444768
SL		 3811 !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) &&
3812 !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags));
84789554
N		 3813 for (i = disks; i--; ) {
3814 struct r5dev *dev = &sh->dev[i];
3815 if (test_bit(R5_LOCKED, &dev->flags) &&
3816 (i == sh->pd_idx || i == sh->qd_idx ||
3817 dev->written)) {
3818 pr_debug("Writing block %d\n", i);
3819 set_bit(R5_Wantwrite, &dev->flags);
3820 if (prexor)
3821 continue;
9c4bdf69
N		 3822 if (s.failed > 1)
3823 continue;
84789554
N		 3824 if (!test_bit(R5_Insync, &dev->flags) ||
3825 ((i == sh->pd_idx || i == sh->qd_idx) &&
3826 s.failed == 0))
3827 set_bit(STRIPE_INSYNC, &sh->state);
3828 }
3829 }
3830 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3831 s.dec_preread_active = 1;
3832 }
3833
ef5b7c69
N		 3834 /*
3835 * might be able to return some write requests if the parity blocks
3836 * are safe, or on a failed drive
3837 */
3838 pdev = &sh->dev[sh->pd_idx];
3839 s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx)
3840 || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx);
3841 qdev = &sh->dev[sh->qd_idx];
3842 s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx)
3843 || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx)
3844 || conf->level < 6;
3845
3846 if (s.written &&
3847 (s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
3848 && !test_bit(R5_LOCKED, &pdev->flags)
3849 && (test_bit(R5_UPTODATE, &pdev->flags) ||
3850 test_bit(R5_Discard, &pdev->flags))))) &&
3851 (s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
3852 && !test_bit(R5_LOCKED, &qdev->flags)
3853 && (test_bit(R5_UPTODATE, &qdev->flags) ||
3854 test_bit(R5_Discard, &qdev->flags))))))
3855 handle_stripe_clean_event(conf, sh, disks, &s.return_bi);
3856
3857 /* Now we might consider reading some blocks, either to check/generate
3858 * parity, or to satisfy requests
3859 * or to load a block that is being partially written.
3860 */
3861 if (s.to_read || s.non_overwrite
3862 || (conf->level == 6 && s.to_write && s.failed)
3863 || (s.syncing && (s.uptodate + s.compute < disks))
3864 || s.replacing
3865 || s.expanding)
3866 handle_stripe_fill(sh, &s, disks);
3867
84789554
N		 3868 /* Now to consider new write requests and what else, if anything
3869 * should be read. We do not handle new writes when:
3870 * 1/ A 'write' operation (copy+xor) is already in flight.
3871 * 2/ A 'check' operation is in flight, as it may clobber the parity
3872 * block.
3873 */
3874 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
3875 handle_stripe_dirtying(conf, sh, &s, disks);
3876
3877 /* maybe we need to check and possibly fix the parity for this stripe
3878 * Any reads will already have been scheduled, so we just see if enough
3879 * data is available. The parity check is held off while parity
3880 * dependent operations are in flight.
3881 */
3882 if (sh->check_state ||
3883 (s.syncing && s.locked == 0 &&
3884 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
3885 !test_bit(STRIPE_INSYNC, &sh->state))) {
3886 if (conf->level == 6)
3887 handle_parity_checks6(conf, sh, &s, disks);
3888 else
3889 handle_parity_checks5(conf, sh, &s, disks);
3890 }
c5a31000 3891
f94c0b66
N		 3892 if ((s.replacing || s.syncing) && s.locked == 0
3893 && !test_bit(STRIPE_COMPUTE_RUN, &sh->state)
3894 && !test_bit(STRIPE_REPLACED, &sh->state)) {
9a3e1101
N		 3895 /* Write out to replacement devices where possible */
3896 for (i = 0; i < conf->raid_disks; i++)
f94c0b66
N		 3897 if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) {
3898 WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags));
9a3e1101
N		 3899 set_bit(R5_WantReplace, &sh->dev[i].flags);
3900 set_bit(R5_LOCKED, &sh->dev[i].flags);
3901 s.locked++;
3902 }
f94c0b66
N		 3903 if (s.replacing)
3904 set_bit(STRIPE_INSYNC, &sh->state);
3905 set_bit(STRIPE_REPLACED, &sh->state);
9a3e1101
N		 3906 }
3907 if ((s.syncing || s.replacing) && s.locked == 0 &&
f94c0b66 3908 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
9a3e1101 3909 test_bit(STRIPE_INSYNC, &sh->state)) {
c5a31000
N		 3910 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3911 clear_bit(STRIPE_SYNCING, &sh->state);
f8dfcffd
N		 3912 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
3913 wake_up(&conf->wait_for_overlap);
c5a31000
N		 3914 }
3915
3916 /* If the failed drives are just a ReadError, then we might need
3917 * to progress the repair/check process
3918 */
3919 if (s.failed <= conf->max_degraded && !conf->mddev->ro)
3920 for (i = 0; i < s.failed; i++) {
3921 struct r5dev *dev = &sh->dev[s.failed_num[i]];
3922 if (test_bit(R5_ReadError, &dev->flags)
3923 && !test_bit(R5_LOCKED, &dev->flags)
3924 && test_bit(R5_UPTODATE, &dev->flags)
3925 ) {
3926 if (!test_bit(R5_ReWrite, &dev->flags)) {
3927 set_bit(R5_Wantwrite, &dev->flags);
3928 set_bit(R5_ReWrite, &dev->flags);
3929 set_bit(R5_LOCKED, &dev->flags);
3930 s.locked++;
3931 } else {
3932 /* let's read it back */
3933 set_bit(R5_Wantread, &dev->flags);
3934 set_bit(R5_LOCKED, &dev->flags);
3935 s.locked++;
3936 }
3937 }
3938 }
3939
3687c061
N		 3940 /* Finish reconstruct operations initiated by the expansion process */
3941 if (sh->reconstruct_state == reconstruct_state_result) {
3942 struct stripe_head *sh_src
3943 = get_active_stripe(conf, sh->sector, 1, 1, 1);
3944 if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) {
3945 /* sh cannot be written until sh_src has been read.
3946 * so arrange for sh to be delayed a little
3947 */
3948 set_bit(STRIPE_DELAYED, &sh->state);
3949 set_bit(STRIPE_HANDLE, &sh->state);
3950 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
3951 &sh_src->state))
3952 atomic_inc(&conf->preread_active_stripes);
3953 release_stripe(sh_src);
3954 goto finish;
3955 }
3956 if (sh_src)
3957 release_stripe(sh_src);
3958
3959 sh->reconstruct_state = reconstruct_state_idle;
3960 clear_bit(STRIPE_EXPANDING, &sh->state);
3961 for (i = conf->raid_disks; i--; ) {
3962 set_bit(R5_Wantwrite, &sh->dev[i].flags);
3963 set_bit(R5_LOCKED, &sh->dev[i].flags);
3964 s.locked++;
3965 }
3966 }
f416885e 3967
3687c061
N		 3968 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
3969 !sh->reconstruct_state) {
3970 /* Need to write out all blocks after computing parity */
3971 sh->disks = conf->raid_disks;
3972 stripe_set_idx(sh->sector, conf, 0, sh);
3973 schedule_reconstruction(sh, &s, 1, 1);
3974 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
3975 clear_bit(STRIPE_EXPAND_READY, &sh->state);
3976 atomic_dec(&conf->reshape_stripes);
3977 wake_up(&conf->wait_for_overlap);
3978 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3979 }
3980
3981 if (s.expanding && s.locked == 0 &&
3982 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
3983 handle_stripe_expansion(conf, sh);
16a53ecc 3984
3687c061 3985finish:
6bfe0b49 3986 /* wait for this device to become unblocked */
5f066c63
N		 3987 if (unlikely(s.blocked_rdev)) {
3988 if (conf->mddev->external)
3989 md_wait_for_blocked_rdev(s.blocked_rdev,
3990 conf->mddev);
3991 else
3992 /* Internal metadata will immediately
3993 * be written by raid5d, so we don't
3994 * need to wait here.
3995 */
3996 rdev_dec_pending(s.blocked_rdev,
3997 conf->mddev);
3998 }
6bfe0b49 3999
bc2607f3
N		 4000 if (s.handle_bad_blocks)
4001 for (i = disks; i--; ) {
3cb03002 4002 struct md_rdev *rdev;
bc2607f3
N		 4003 struct r5dev *dev = &sh->dev[i];
4004 if (test_and_clear_bit(R5_WriteError, &dev->flags)) {
4005 /* We own a safe reference to the rdev */
4006 rdev = conf->disks[i].rdev;
4007 if (!rdev_set_badblocks(rdev, sh->sector,
4008 STRIPE_SECTORS, 0))
4009 md_error(conf->mddev, rdev);
4010 rdev_dec_pending(rdev, conf->mddev);
4011 }
b84db560
N		 4012 if (test_and_clear_bit(R5_MadeGood, &dev->flags)) {
4013 rdev = conf->disks[i].rdev;
4014 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 4015 STRIPE_SECTORS, 0);
b84db560
N		 4016 rdev_dec_pending(rdev, conf->mddev);
4017 }
977df362
N		 4018 if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) {
4019 rdev = conf->disks[i].replacement;
dd054fce
N		 4020 if (!rdev)
4021 /* rdev have been moved down */
4022 rdev = conf->disks[i].rdev;
977df362 4023 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 4024 STRIPE_SECTORS, 0);
977df362
N		 4025 rdev_dec_pending(rdev, conf->mddev);
4026 }
bc2607f3
N		 4027 }
4028
6c0069c0
YT		 4029 if (s.ops_request)
4030 raid_run_ops(sh, s.ops_request);
4031
f0e43bcd 4032 ops_run_io(sh, &s);
16a53ecc 4033
c5709ef6 4034 if (s.dec_preread_active) {
729a1866 4035 /* We delay this until after ops_run_io so that if make_request
e9c7469b 4036 * is waiting on a flush, it won't continue until the writes
729a1866
N		 4037 * have actually been submitted.
4038 */
4039 atomic_dec(&conf->preread_active_stripes);
4040 if (atomic_read(&conf->preread_active_stripes) <
4041 IO_THRESHOLD)
4042 md_wakeup_thread(conf->mddev->thread);
4043 }
4044
c5709ef6 4045 return_io(s.return_bi);
16a53ecc 4046
257a4b42 4047 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
16a53ecc
N		 4048}
4049
d1688a6d 4050static void raid5_activate_delayed(struct r5conf *conf)
16a53ecc
N		 4051{
4052 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
4053 while (!list_empty(&conf->delayed_list)) {
4054 struct list_head *l = conf->delayed_list.next;
4055 struct stripe_head *sh;
4056 sh = list_entry(l, struct stripe_head, lru);
4057 list_del_init(l);
4058 clear_bit(STRIPE_DELAYED, &sh->state);
4059 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4060 atomic_inc(&conf->preread_active_stripes);
8b3e6cdc 4061 list_add_tail(&sh->lru, &conf->hold_list);
851c30c9 4062 raid5_wakeup_stripe_thread(sh);
16a53ecc 4063 }
482c0834 4064 }
16a53ecc
N		 4065}
4066
566c09c5
SL		 4067static void activate_bit_delay(struct r5conf *conf,
4068 struct list_head *temp_inactive_list)
16a53ecc
N		 4069{
4070 /* device_lock is held */
4071 struct list_head head;
4072 list_add(&head, &conf->bitmap_list);
4073 list_del_init(&conf->bitmap_list);
4074 while (!list_empty(&head)) {
4075 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
566c09c5 4076 int hash;
16a53ecc
N		 4077 list_del_init(&sh->lru);
4078 atomic_inc(&sh->count);
566c09c5
SL		 4079 hash = sh->hash_lock_index;
4080 __release_stripe(conf, sh, &temp_inactive_list[hash]);
16a53ecc
N		 4081 }
4082}
4083
fd01b88c 4084int md_raid5_congested(struct mddev *mddev, int bits)
f022b2fd 4085{
d1688a6d 4086 struct r5conf *conf = mddev->private;
f022b2fd
N		 4087
4088 /* No difference between reads and writes. Just check
4089 * how busy the stripe_cache is
4090 */
3fa841d7 4091
f022b2fd
N		 4092 if (conf->inactive_blocked)
4093 return 1;
4094 if (conf->quiesce)
4095 return 1;
4bda556a 4096 if (atomic_read(&conf->empty_inactive_list_nr))
f022b2fd
N		 4097 return 1;
4098
4099 return 0;
4100}
11d8a6e3
N		 4101EXPORT_SYMBOL_GPL(md_raid5_congested);
4102
4103static int raid5_congested(void *data, int bits)
4104{
fd01b88c 4105 struct mddev *mddev = data;
11d8a6e3
N		 4106
4107 return mddev_congested(mddev, bits) ||
4108 md_raid5_congested(mddev, bits);
4109}
f022b2fd 4110
23032a0e
RBJ		 4111/* We want read requests to align with chunks where possible,
4112 * but write requests don't need to.
4113 */
cc371e66
AK		 4114static int raid5_mergeable_bvec(struct request_queue *q,
4115 struct bvec_merge_data *bvm,
4116 struct bio_vec *biovec)
23032a0e 4117{
fd01b88c 4118 struct mddev *mddev = q->queuedata;
cc371e66 4119 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
23032a0e 4120 int max;
9d8f0363 4121 unsigned int chunk_sectors = mddev->chunk_sectors;
cc371e66 4122 unsigned int bio_sectors = bvm->bi_size >> 9;
23032a0e 4123
cc371e66 4124 if ((bvm->bi_rw & 1) == WRITE)
23032a0e
RBJ		 4125 return biovec->bv_len; /* always allow writes to be mergeable */
4126
664e7c41
AN		 4127 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
4128 chunk_sectors = mddev->new_chunk_sectors;
23032a0e
RBJ		 4129 max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
4130 if (max < 0) max = 0;
4131 if (max <= biovec->bv_len && bio_sectors == 0)
4132 return biovec->bv_len;
4133 else
4134 return max;
4135}
4136
fd01b88c 4137static int in_chunk_boundary(struct mddev *mddev, struct bio *bio)
f679623f 4138{
4f024f37 4139 sector_t sector = bio->bi_iter.bi_sector + get_start_sect(bio->bi_bdev);
9d8f0363 4140 unsigned int chunk_sectors = mddev->chunk_sectors;
aa8b57aa 4141 unsigned int bio_sectors = bio_sectors(bio);
f679623f 4142
664e7c41
AN		 4143 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
4144 chunk_sectors = mddev->new_chunk_sectors;
f679623f
RBJ		 4145 return chunk_sectors >=
4146 ((sector & (chunk_sectors - 1)) + bio_sectors);
4147}
4148
46031f9a
RBJ		 4149/*
4150 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
4151 * later sampled by raid5d.
4152 */
d1688a6d 4153static void add_bio_to_retry(struct bio *bi,struct r5conf *conf)
46031f9a
RBJ		 4154{
4155 unsigned long flags;
4156
4157 spin_lock_irqsave(&conf->device_lock, flags);
4158
4159 bi->bi_next = conf->retry_read_aligned_list;
4160 conf->retry_read_aligned_list = bi;
4161
4162 spin_unlock_irqrestore(&conf->device_lock, flags);
4163 md_wakeup_thread(conf->mddev->thread);
4164}
4165
d1688a6d 4166static struct bio *remove_bio_from_retry(struct r5conf *conf)
46031f9a
RBJ		 4167{
4168 struct bio *bi;
4169
4170 bi = conf->retry_read_aligned;
4171 if (bi) {
4172 conf->retry_read_aligned = NULL;
4173 return bi;
4174 }
4175 bi = conf->retry_read_aligned_list;
4176 if(bi) {
387bb173 4177 conf->retry_read_aligned_list = bi->bi_next;
46031f9a 4178 bi->bi_next = NULL;
960e739d
JA		 4179 /*
4180 * this sets the active strip count to 1 and the processed
4181 * strip count to zero (upper 8 bits)
4182 */
e7836bd6 4183 raid5_set_bi_stripes(bi, 1); /* biased count of active stripes */
46031f9a
RBJ		 4184 }
4185
4186 return bi;
4187}
4188
f679623f
RBJ		 4189/*
4190 * The "raid5_align_endio" should check if the read succeeded and if it
4191 * did, call bio_endio on the original bio (having bio_put the new bio
4192 * first).
4193 * If the read failed..
4194 */
6712ecf8 4195static void raid5_align_endio(struct bio *bi, int error)
f679623f
RBJ		 4196{
4197 struct bio* raid_bi = bi->bi_private;
fd01b88c 4198 struct mddev *mddev;
d1688a6d 4199 struct r5conf *conf;
46031f9a 4200 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
3cb03002 4201 struct md_rdev *rdev;
46031f9a 4202
f679623f 4203 bio_put(bi);
46031f9a 4204
46031f9a
RBJ		 4205 rdev = (void*)raid_bi->bi_next;
4206 raid_bi->bi_next = NULL;
2b7f2228
N		 4207 mddev = rdev->mddev;
4208 conf = mddev->private;
46031f9a
RBJ		 4209
4210 rdev_dec_pending(rdev, conf->mddev);
4211
4212 if (!error && uptodate) {
0a82a8d1
LT		 4213 trace_block_bio_complete(bdev_get_queue(raid_bi->bi_bdev),
4214 raid_bi, 0);
6712ecf8 4215 bio_endio(raid_bi, 0);
46031f9a
RBJ		 4216 if (atomic_dec_and_test(&conf->active_aligned_reads))
4217 wake_up(&conf->wait_for_stripe);
6712ecf8 4218 return;
46031f9a
RBJ		 4219 }
4220
45b4233c 4221 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
46031f9a
RBJ		 4222
4223 add_bio_to_retry(raid_bi, conf);
f679623f
RBJ		 4224}
4225
387bb173
NB		 4226static int bio_fits_rdev(struct bio *bi)
4227{
165125e1 4228 struct request_queue *q = bdev_get_queue(bi->bi_bdev);
387bb173 4229
aa8b57aa 4230 if (bio_sectors(bi) > queue_max_sectors(q))
387bb173
NB		 4231 return 0;
4232 blk_recount_segments(q, bi);
8a78362c 4233 if (bi->bi_phys_segments > queue_max_segments(q))
387bb173
NB		 4234 return 0;
4235
4236 if (q->merge_bvec_fn)
4237 /* it's too hard to apply the merge_bvec_fn at this stage,
4238 * just just give up
4239 */
4240 return 0;
4241
4242 return 1;
4243}
4244
fd01b88c 4245static int chunk_aligned_read(struct mddev *mddev, struct bio * raid_bio)
f679623f 4246{
d1688a6d 4247 struct r5conf *conf = mddev->private;
8553fe7e 4248 int dd_idx;
f679623f 4249 struct bio* align_bi;
3cb03002 4250 struct md_rdev *rdev;
671488cc 4251 sector_t end_sector;
f679623f
RBJ		 4252
4253 if (!in_chunk_boundary(mddev, raid_bio)) {
45b4233c 4254 pr_debug("chunk_aligned_read : non aligned\n");
f679623f
RBJ		 4255 return 0;
4256 }
4257 /*
a167f663 4258 * use bio_clone_mddev to make a copy of the bio
f679623f 4259 */
a167f663 4260 align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev);
f679623f
RBJ		 4261 if (!align_bi)
4262 return 0;
4263 /*
4264 * set bi_end_io to a new function, and set bi_private to the
4265 * original bio.
4266 */
4267 align_bi->bi_end_io = raid5_align_endio;
4268 align_bi->bi_private = raid_bio;
4269 /*
4270 * compute position
4271 */
4f024f37
KO		 4272 align_bi->bi_iter.bi_sector =
4273 raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector,
4274 0, &dd_idx, NULL);
f679623f 4275
f73a1c7d 4276 end_sector = bio_end_sector(align_bi);
f679623f 4277 rcu_read_lock();
671488cc
N		 4278 rdev = rcu_dereference(conf->disks[dd_idx].replacement);
4279 if (!rdev || test_bit(Faulty, &rdev->flags) ||
4280 rdev->recovery_offset < end_sector) {
4281 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
4282 if (rdev &&
4283 (test_bit(Faulty, &rdev->flags) ||
4284 !(test_bit(In_sync, &rdev->flags) ||
4285 rdev->recovery_offset >= end_sector)))
4286 rdev = NULL;
4287 }
4288 if (rdev) {
31c176ec
N		 4289 sector_t first_bad;
4290 int bad_sectors;
4291
f679623f
RBJ		 4292 atomic_inc(&rdev->nr_pending);
4293 rcu_read_unlock();
46031f9a
RBJ		 4294 raid_bio->bi_next = (void*)rdev;
4295 align_bi->bi_bdev = rdev->bdev;
3fd83717 4296 __clear_bit(BIO_SEG_VALID, &align_bi->bi_flags);
46031f9a 4297
31c176ec 4298 if (!bio_fits_rdev(align_bi) ||
4f024f37
KO		 4299 is_badblock(rdev, align_bi->bi_iter.bi_sector,
4300 bio_sectors(align_bi),
31c176ec
N		 4301 &first_bad, &bad_sectors)) {
4302 /* too big in some way, or has a known bad block */
387bb173
NB		 4303 bio_put(align_bi);
4304 rdev_dec_pending(rdev, mddev);
4305 return 0;
4306 }
4307
6c0544e2 4308 /* No reshape active, so we can trust rdev->data_offset */
4f024f37 4309 align_bi->bi_iter.bi_sector += rdev->data_offset;
6c0544e2 4310
46031f9a
RBJ		 4311 spin_lock_irq(&conf->device_lock);
4312 wait_event_lock_irq(conf->wait_for_stripe,
4313 conf->quiesce == 0,
eed8c02e 4314 conf->device_lock);
46031f9a
RBJ		 4315 atomic_inc(&conf->active_aligned_reads);
4316 spin_unlock_irq(&conf->device_lock);
4317
e3620a3a
JB		 4318 if (mddev->gendisk)
4319 trace_block_bio_remap(bdev_get_queue(align_bi->bi_bdev),
4320 align_bi, disk_devt(mddev->gendisk),
4f024f37 4321 raid_bio->bi_iter.bi_sector);
f679623f
RBJ		 4322 generic_make_request(align_bi);
4323 return 1;
4324 } else {
4325 rcu_read_unlock();
46031f9a 4326 bio_put(align_bi);
f679623f
RBJ		 4327 return 0;
4328 }
4329}
4330
8b3e6cdc
DW		 4331/* __get_priority_stripe - get the next stripe to process
4332 *
4333 * Full stripe writes are allowed to pass preread active stripes up until
4334 * the bypass_threshold is exceeded. In general the bypass_count
4335 * increments when the handle_list is handled before the hold_list; however, it
4336 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
4337 * stripe with in flight i/o. The bypass_count will be reset when the
4338 * head of the hold_list has changed, i.e. the head was promoted to the
4339 * handle_list.
4340 */
851c30c9 4341static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group)
8b3e6cdc 4342{
851c30c9
SL		 4343 struct stripe_head *sh = NULL, *tmp;
4344 struct list_head *handle_list = NULL;
bfc90cb0 4345 struct r5worker_group *wg = NULL;
851c30c9
SL		 4346
4347 if (conf->worker_cnt_per_group == 0) {
4348 handle_list = &conf->handle_list;
4349 } else if (group != ANY_GROUP) {
4350 handle_list = &conf->worker_groups[group].handle_list;
bfc90cb0 4351 wg = &conf->worker_groups[group];
851c30c9
SL		 4352 } else {
4353 int i;
4354 for (i = 0; i < conf->group_cnt; i++) {
4355 handle_list = &conf->worker_groups[i].handle_list;
bfc90cb0 4356 wg = &conf->worker_groups[i];
851c30c9
SL		 4357 if (!list_empty(handle_list))
4358 break;
4359 }
4360 }
8b3e6cdc
DW		 4361
4362 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
4363 __func__,
851c30c9 4364 list_empty(handle_list) ? "empty" : "busy",
8b3e6cdc
DW		 4365 list_empty(&conf->hold_list) ? "empty" : "busy",
4366 atomic_read(&conf->pending_full_writes), conf->bypass_count);
4367
851c30c9
SL		 4368 if (!list_empty(handle_list)) {
4369 sh = list_entry(handle_list->next, typeof(*sh), lru);
8b3e6cdc
DW		 4370
4371 if (list_empty(&conf->hold_list))
4372 conf->bypass_count = 0;
4373 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
4374 if (conf->hold_list.next == conf->last_hold)
4375 conf->bypass_count++;
4376 else {
4377 conf->last_hold = conf->hold_list.next;
4378 conf->bypass_count -= conf->bypass_threshold;
4379 if (conf->bypass_count < 0)
4380 conf->bypass_count = 0;
4381 }
4382 }
4383 } else if (!list_empty(&conf->hold_list) &&
4384 ((conf->bypass_threshold &&
4385 conf->bypass_count > conf->bypass_threshold) ||
4386 atomic_read(&conf->pending_full_writes) == 0)) {
851c30c9
SL		 4387
4388 list_for_each_entry(tmp, &conf->hold_list, lru) {
4389 if (conf->worker_cnt_per_group == 0 ||
4390 group == ANY_GROUP ||
4391 !cpu_online(tmp->cpu) ||
4392 cpu_to_group(tmp->cpu) == group) {
4393 sh = tmp;
4394 break;
4395 }
4396 }
4397
4398 if (sh) {
4399 conf->bypass_count -= conf->bypass_threshold;
4400 if (conf->bypass_count < 0)
4401 conf->bypass_count = 0;
4402 }
bfc90cb0 4403 wg = NULL;
851c30c9
SL		 4404 }
4405
4406 if (!sh)
8b3e6cdc
DW		 4407 return NULL;
4408
bfc90cb0
SL		 4409 if (wg) {
4410 wg->stripes_cnt--;
4411 sh->group = NULL;
4412 }
8b3e6cdc 4413 list_del_init(&sh->lru);
c7a6d35e 4414 BUG_ON(atomic_inc_return(&sh->count) != 1);
8b3e6cdc
DW		 4415 return sh;
4416}
f679623f 4417
8811b596
SL		 4418struct raid5_plug_cb {
4419 struct blk_plug_cb cb;
4420 struct list_head list;
566c09c5 4421 struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
8811b596
SL		 4422};
4423
4424static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule)
4425{
4426 struct raid5_plug_cb *cb = container_of(
4427 blk_cb, struct raid5_plug_cb, cb);
4428 struct stripe_head *sh;
4429 struct mddev *mddev = cb->cb.data;
4430 struct r5conf *conf = mddev->private;
a9add5d9 4431 int cnt = 0;
566c09c5 4432 int hash;
8811b596
SL		 4433
4434 if (cb->list.next && !list_empty(&cb->list)) {
4435 spin_lock_irq(&conf->device_lock);
4436 while (!list_empty(&cb->list)) {
4437 sh = list_first_entry(&cb->list, struct stripe_head, lru);
4438 list_del_init(&sh->lru);
4439 /*
4440 * avoid race release_stripe_plug() sees
4441 * STRIPE_ON_UNPLUG_LIST clear but the stripe
4442 * is still in our list
4443 */
4e857c58 4444 smp_mb__before_atomic();
8811b596 4445 clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state);
773ca82f
SL		 4446 /*
4447 * STRIPE_ON_RELEASE_LIST could be set here. In that
4448 * case, the count is always > 1 here
4449 */
566c09c5
SL		 4450 hash = sh->hash_lock_index;
4451 __release_stripe(conf, sh, &cb->temp_inactive_list[hash]);
a9add5d9 4452 cnt++;
8811b596
SL		 4453 }
4454 spin_unlock_irq(&conf->device_lock);
4455 }
566c09c5
SL		 4456 release_inactive_stripe_list(conf, cb->temp_inactive_list,
4457 NR_STRIPE_HASH_LOCKS);
e3620a3a
JB		 4458 if (mddev->queue)
4459 trace_block_unplug(mddev->queue, cnt, !from_schedule);
8811b596
SL		 4460 kfree(cb);
4461}
4462
4463static void release_stripe_plug(struct mddev *mddev,
4464 struct stripe_head *sh)
4465{
4466 struct blk_plug_cb *blk_cb = blk_check_plugged(
4467 raid5_unplug, mddev,
4468 sizeof(struct raid5_plug_cb));
4469 struct raid5_plug_cb *cb;
4470
4471 if (!blk_cb) {
4472 release_stripe(sh);
4473 return;
4474 }
4475
4476 cb = container_of(blk_cb, struct raid5_plug_cb, cb);
4477
566c09c5
SL		 4478 if (cb->list.next == NULL) {
4479 int i;
8811b596 4480 INIT_LIST_HEAD(&cb->list);
566c09c5
SL		 4481 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
4482 INIT_LIST_HEAD(cb->temp_inactive_list + i);
4483 }
8811b596
SL		 4484
4485 if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state))
4486 list_add_tail(&sh->lru, &cb->list);
4487 else
4488 release_stripe(sh);
4489}
4490
620125f2
SL		 4491static void make_discard_request(struct mddev *mddev, struct bio *bi)
4492{
4493 struct r5conf *conf = mddev->private;
4494 sector_t logical_sector, last_sector;
4495 struct stripe_head *sh;
4496 int remaining;
4497 int stripe_sectors;
4498
4499 if (mddev->reshape_position != MaxSector)
4500 /* Skip discard while reshape is happening */
4501 return;
4502
4f024f37
KO		 4503 logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1);
4504 last_sector = bi->bi_iter.bi_sector + (bi->bi_iter.bi_size>>9);
620125f2
SL		 4505
4506 bi->bi_next = NULL;
4507 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
4508
4509 stripe_sectors = conf->chunk_sectors *
4510 (conf->raid_disks - conf->max_degraded);
4511 logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector,
4512 stripe_sectors);
4513 sector_div(last_sector, stripe_sectors);
4514
4515 logical_sector *= conf->chunk_sectors;
4516 last_sector *= conf->chunk_sectors;
4517
4518 for (; logical_sector < last_sector;
4519 logical_sector += STRIPE_SECTORS) {
4520 DEFINE_WAIT(w);
4521 int d;
4522 again:
4523 sh = get_active_stripe(conf, logical_sector, 0, 0, 0);
4524 prepare_to_wait(&conf->wait_for_overlap, &w,
4525 TASK_UNINTERRUPTIBLE);
f8dfcffd
N		 4526 set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
4527 if (test_bit(STRIPE_SYNCING, &sh->state)) {
4528 release_stripe(sh);
4529 schedule();
4530 goto again;
4531 }
4532 clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
620125f2
SL		 4533 spin_lock_irq(&sh->stripe_lock);
4534 for (d = 0; d < conf->raid_disks; d++) {
4535 if (d == sh->pd_idx || d == sh->qd_idx)
4536 continue;
4537 if (sh->dev[d].towrite || sh->dev[d].toread) {
4538 set_bit(R5_Overlap, &sh->dev[d].flags);
4539 spin_unlock_irq(&sh->stripe_lock);
4540 release_stripe(sh);
4541 schedule();
4542 goto again;
4543 }
4544 }
f8dfcffd 4545 set_bit(STRIPE_DISCARD, &sh->state);
620125f2
SL		 4546 finish_wait(&conf->wait_for_overlap, &w);
4547 for (d = 0; d < conf->raid_disks; d++) {
4548 if (d == sh->pd_idx || d == sh->qd_idx)
4549 continue;
4550 sh->dev[d].towrite = bi;
4551 set_bit(R5_OVERWRITE, &sh->dev[d].flags);
4552 raid5_inc_bi_active_stripes(bi);
4553 }
4554 spin_unlock_irq(&sh->stripe_lock);
4555 if (conf->mddev->bitmap) {
4556 for (d = 0;
4557 d < conf->raid_disks - conf->max_degraded;
4558 d++)
4559 bitmap_startwrite(mddev->bitmap,
4560 sh->sector,
4561 STRIPE_SECTORS,
4562 0);
4563 sh->bm_seq = conf->seq_flush + 1;
4564 set_bit(STRIPE_BIT_DELAY, &sh->state);
4565 }
4566
4567 set_bit(STRIPE_HANDLE, &sh->state);
4568 clear_bit(STRIPE_DELAYED, &sh->state);
4569 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4570 atomic_inc(&conf->preread_active_stripes);
4571 release_stripe_plug(mddev, sh);
4572 }
4573
4574 remaining = raid5_dec_bi_active_stripes(bi);
4575 if (remaining == 0) {
4576 md_write_end(mddev);
4577 bio_endio(bi, 0);
4578 }
4579}
4580
b4fdcb02 4581static void make_request(struct mddev *mddev, struct bio * bi)
1da177e4 4582{
d1688a6d 4583 struct r5conf *conf = mddev->private;
911d4ee8 4584 int dd_idx;
1da177e4
LT		 4585 sector_t new_sector;
4586 sector_t logical_sector, last_sector;
4587 struct stripe_head *sh;
a362357b 4588 const int rw = bio_data_dir(bi);
49077326 4589 int remaining;
27c0f68f
SL		 4590 DEFINE_WAIT(w);
4591 bool do_prepare;
1da177e4 4592
e9c7469b
TH		 4593 if (unlikely(bi->bi_rw & REQ_FLUSH)) {
4594 md_flush_request(mddev, bi);
5a7bbad2 4595 return;
e5dcdd80
N		 4596 }
4597
3d310eb7 4598 md_write_start(mddev, bi);
06d91a5f 4599
802ba064 4600 if (rw == READ &&
52488615 4601 mddev->reshape_position == MaxSector &&
21a52c6d 4602 chunk_aligned_read(mddev,bi))
5a7bbad2 4603 return;
52488615 4604
620125f2
SL		 4605 if (unlikely(bi->bi_rw & REQ_DISCARD)) {
4606 make_discard_request(mddev, bi);
4607 return;
4608 }
4609
4f024f37 4610 logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1);
f73a1c7d 4611 last_sector = bio_end_sector(bi);
1da177e4
LT		 4612 bi->bi_next = NULL;
4613 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
06d91a5f 4614
27c0f68f 4615 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
1da177e4 4616 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
b5663ba4 4617 int previous;
c46501b2 4618 int seq;
b578d55f 4619
27c0f68f 4620 do_prepare = false;
7ecaa1e6 4621 retry:
c46501b2 4622 seq = read_seqcount_begin(&conf->gen_lock);
b5663ba4 4623 previous = 0;
27c0f68f
SL		 4624 if (do_prepare)
4625 prepare_to_wait(&conf->wait_for_overlap, &w,
4626 TASK_UNINTERRUPTIBLE);
b0f9ec04 4627 if (unlikely(conf->reshape_progress != MaxSector)) {
fef9c61f 4628 /* spinlock is needed as reshape_progress may be
df8e7f76
N		 4629 * 64bit on a 32bit platform, and so it might be
4630 * possible to see a half-updated value
aeb878b0 4631 * Of course reshape_progress could change after
df8e7f76
N		 4632 * the lock is dropped, so once we get a reference
4633 * to the stripe that we think it is, we will have
4634 * to check again.
4635 */
7ecaa1e6 4636 spin_lock_irq(&conf->device_lock);
2c810cdd 4637 if (mddev->reshape_backwards
fef9c61f
N		 4638 ? logical_sector < conf->reshape_progress
4639 : logical_sector >= conf->reshape_progress) {
b5663ba4
N		 4640 previous = 1;
4641 } else {
2c810cdd 4642 if (mddev->reshape_backwards
fef9c61f
N		 4643 ? logical_sector < conf->reshape_safe
4644 : logical_sector >= conf->reshape_safe) {
b578d55f
N		 4645 spin_unlock_irq(&conf->device_lock);
4646 schedule();
27c0f68f 4647 do_prepare = true;
b578d55f
N		 4648 goto retry;
4649 }
4650 }
7ecaa1e6
N		 4651 spin_unlock_irq(&conf->device_lock);
4652 }
16a53ecc 4653
112bf897
N		 4654 new_sector = raid5_compute_sector(conf, logical_sector,
4655 previous,
911d4ee8 4656 &dd_idx, NULL);
0c55e022 4657 pr_debug("raid456: make_request, sector %llu logical %llu\n",
c46501b2 4658 (unsigned long long)new_sector,
1da177e4
LT		 4659 (unsigned long long)logical_sector);
4660
b5663ba4 4661 sh = get_active_stripe(conf, new_sector, previous,
a8c906ca 4662 (bi->bi_rw&RWA_MASK), 0);
1da177e4 4663 if (sh) {
b0f9ec04 4664 if (unlikely(previous)) {
7ecaa1e6 4665 /* expansion might have moved on while waiting for a
df8e7f76
N		 4666 * stripe, so we must do the range check again.
4667 * Expansion could still move past after this
4668 * test, but as we are holding a reference to
4669 * 'sh', we know that if that happens,
4670 * STRIPE_EXPANDING will get set and the expansion
4671 * won't proceed until we finish with the stripe.
7ecaa1e6
N		 4672 */
4673 int must_retry = 0;
4674 spin_lock_irq(&conf->device_lock);
2c810cdd 4675 if (mddev->reshape_backwards
b0f9ec04
N		 4676 ? logical_sector >= conf->reshape_progress
4677 : logical_sector < conf->reshape_progress)
7ecaa1e6
N		 4678 /* mismatch, need to try again */
4679 must_retry = 1;
4680 spin_unlock_irq(&conf->device_lock);
4681 if (must_retry) {
4682 release_stripe(sh);
7a3ab908 4683 schedule();
27c0f68f 4684 do_prepare = true;
7ecaa1e6
N		 4685 goto retry;
4686 }
4687 }
c46501b2
N		 4688 if (read_seqcount_retry(&conf->gen_lock, seq)) {
4689 /* Might have got the wrong stripe_head
4690 * by accident
4691 */
4692 release_stripe(sh);
4693 goto retry;
4694 }
e62e58a5 4695
ffd96e35 4696 if (rw == WRITE &&
a5c308d4 4697 logical_sector >= mddev->suspend_lo &&
e464eafd
N		 4698 logical_sector < mddev->suspend_hi) {
4699 release_stripe(sh);
e62e58a5
N		 4700 /* As the suspend_* range is controlled by
4701 * userspace, we want an interruptible
4702 * wait.
4703 */
4704 flush_signals(current);
4705 prepare_to_wait(&conf->wait_for_overlap,
4706 &w, TASK_INTERRUPTIBLE);
4707 if (logical_sector >= mddev->suspend_lo &&
27c0f68f 4708 logical_sector < mddev->suspend_hi) {
e62e58a5 4709 schedule();
27c0f68f
SL		 4710 do_prepare = true;
4711 }
e464eafd
N		 4712 goto retry;
4713 }
7ecaa1e6
N		 4714
4715 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
ffd96e35 4716 !add_stripe_bio(sh, bi, dd_idx, rw)) {
7ecaa1e6
N		 4717 /* Stripe is busy expanding or
4718 * add failed due to overlap. Flush everything
1da177e4
LT		 4719 * and wait a while
4720 */
482c0834 4721 md_wakeup_thread(mddev->thread);
1da177e4
LT		 4722 release_stripe(sh);
4723 schedule();
27c0f68f 4724 do_prepare = true;
1da177e4
LT		 4725 goto retry;
4726 }
6ed3003c
N		 4727 set_bit(STRIPE_HANDLE, &sh->state);
4728 clear_bit(STRIPE_DELAYED, &sh->state);
a852d7b8 4729 if ((bi->bi_rw & REQ_SYNC) &&
729a1866
N		 4730 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4731 atomic_inc(&conf->preread_active_stripes);
8811b596 4732 release_stripe_plug(mddev, sh);
1da177e4
LT		 4733 } else {
4734 /* cannot get stripe for read-ahead, just give-up */
4735 clear_bit(BIO_UPTODATE, &bi->bi_flags);
1da177e4
LT		 4736 break;
4737 }
1da177e4 4738 }
27c0f68f 4739 finish_wait(&conf->wait_for_overlap, &w);
7c13edc8 4740
e7836bd6 4741 remaining = raid5_dec_bi_active_stripes(bi);
f6344757 4742 if (remaining == 0) {
1da177e4 4743
16a53ecc 4744 if ( rw == WRITE )
1da177e4 4745 md_write_end(mddev);
6712ecf8 4746
0a82a8d1
LT		 4747 trace_block_bio_complete(bdev_get_queue(bi->bi_bdev),
4748 bi, 0);
0e13fe23 4749 bio_endio(bi, 0);
1da177e4 4750 }
1da177e4
LT		 4751}
4752
fd01b88c 4753static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks);
b522adcd 4754
fd01b88c 4755static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped)
1da177e4 4756{
52c03291
N		 4757 /* reshaping is quite different to recovery/resync so it is
4758 * handled quite separately ... here.
4759 *
4760 * On each call to sync_request, we gather one chunk worth of
4761 * destination stripes and flag them as expanding.
4762 * Then we find all the source stripes and request reads.
4763 * As the reads complete, handle_stripe will copy the data
4764 * into the destination stripe and release that stripe.
4765 */
d1688a6d 4766 struct r5conf *conf = mddev->private;
1da177e4 4767 struct stripe_head *sh;
ccfcc3c1 4768 sector_t first_sector, last_sector;
f416885e
N		 4769 int raid_disks = conf->previous_raid_disks;
4770 int data_disks = raid_disks - conf->max_degraded;
4771 int new_data_disks = conf->raid_disks - conf->max_degraded;
52c03291
N		 4772 int i;
4773 int dd_idx;
c8f517c4 4774 sector_t writepos, readpos, safepos;
ec32a2bd 4775 sector_t stripe_addr;
7a661381 4776 int reshape_sectors;
ab69ae12 4777 struct list_head stripes;
52c03291 4778
fef9c61f
N		 4779 if (sector_nr == 0) {
4780 /* If restarting in the middle, skip the initial sectors */
2c810cdd 4781 if (mddev->reshape_backwards &&
fef9c61f
N		 4782 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
4783 sector_nr = raid5_size(mddev, 0, 0)
4784 - conf->reshape_progress;
2c810cdd 4785 } else if (!mddev->reshape_backwards &&
fef9c61f
N		 4786 conf->reshape_progress > 0)
4787 sector_nr = conf->reshape_progress;
f416885e 4788 sector_div(sector_nr, new_data_disks);
fef9c61f 4789 if (sector_nr) {
8dee7211
N		 4790 mddev->curr_resync_completed = sector_nr;
4791 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
fef9c61f
N		 4792 *skipped = 1;
4793 return sector_nr;
4794 }
52c03291
N		 4795 }
4796
7a661381
N		 4797 /* We need to process a full chunk at a time.
4798 * If old and new chunk sizes differ, we need to process the
4799 * largest of these
4800 */
664e7c41
AN		 4801 if (mddev->new_chunk_sectors > mddev->chunk_sectors)
4802 reshape_sectors = mddev->new_chunk_sectors;
7a661381 4803 else
9d8f0363 4804 reshape_sectors = mddev->chunk_sectors;
7a661381 4805
b5254dd5
N		 4806 /* We update the metadata at least every 10 seconds, or when
4807 * the data about to be copied would over-write the source of
4808 * the data at the front of the range. i.e. one new_stripe
4809 * along from reshape_progress new_maps to after where
4810 * reshape_safe old_maps to
52c03291 4811 */
fef9c61f 4812 writepos = conf->reshape_progress;
f416885e 4813 sector_div(writepos, new_data_disks);
c8f517c4
N		 4814 readpos = conf->reshape_progress;
4815 sector_div(readpos, data_disks);
fef9c61f 4816 safepos = conf->reshape_safe;
f416885e 4817 sector_div(safepos, data_disks);
2c810cdd 4818 if (mddev->reshape_backwards) {
ed37d83e 4819 writepos -= min_t(sector_t, reshape_sectors, writepos);
c8f517c4 4820 readpos += reshape_sectors;
7a661381 4821 safepos += reshape_sectors;
fef9c61f 4822 } else {
7a661381 4823 writepos += reshape_sectors;
ed37d83e
N		 4824 readpos -= min_t(sector_t, reshape_sectors, readpos);
4825 safepos -= min_t(sector_t, reshape_sectors, safepos);
fef9c61f 4826 }
52c03291 4827
b5254dd5
N		 4828 /* Having calculated the 'writepos' possibly use it
4829 * to set 'stripe_addr' which is where we will write to.
4830 */
4831 if (mddev->reshape_backwards) {
4832 BUG_ON(conf->reshape_progress == 0);
4833 stripe_addr = writepos;
4834 BUG_ON((mddev->dev_sectors &
4835 ~((sector_t)reshape_sectors - 1))
4836 - reshape_sectors - stripe_addr
4837 != sector_nr);
4838 } else {
4839 BUG_ON(writepos != sector_nr + reshape_sectors);
4840 stripe_addr = sector_nr;
4841 }
4842
c8f517c4
N		 4843 /* 'writepos' is the most advanced device address we might write.
4844 * 'readpos' is the least advanced device address we might read.
4845 * 'safepos' is the least address recorded in the metadata as having
4846 * been reshaped.
b5254dd5
N		 4847 * If there is a min_offset_diff, these are adjusted either by
4848 * increasing the safepos/readpos if diff is negative, or
4849 * increasing writepos if diff is positive.
4850 * If 'readpos' is then behind 'writepos', there is no way that we can
c8f517c4
N		 4851 * ensure safety in the face of a crash - that must be done by userspace
4852 * making a backup of the data. So in that case there is no particular
4853 * rush to update metadata.
4854 * Otherwise if 'safepos' is behind 'writepos', then we really need to
4855 * update the metadata to advance 'safepos' to match 'readpos' so that
4856 * we can be safe in the event of a crash.
4857 * So we insist on updating metadata if safepos is behind writepos and
4858 * readpos is beyond writepos.
4859 * In any case, update the metadata every 10 seconds.
4860 * Maybe that number should be configurable, but I'm not sure it is
4861 * worth it.... maybe it could be a multiple of safemode_delay???
4862 */
b5254dd5
N		 4863 if (conf->min_offset_diff < 0) {
4864 safepos += -conf->min_offset_diff;
4865 readpos += -conf->min_offset_diff;
4866 } else
4867 writepos += conf->min_offset_diff;
4868
2c810cdd 4869 if ((mddev->reshape_backwards
c8f517c4
N		 4870 ? (safepos > writepos && readpos < writepos)
4871 : (safepos < writepos && readpos > writepos)) ||
4872 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
52c03291
N		 4873 /* Cannot proceed until we've updated the superblock... */
4874 wait_event(conf->wait_for_overlap,
c91abf5a
N		 4875 atomic_read(&conf->reshape_stripes)==0
4876 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
4877 if (atomic_read(&conf->reshape_stripes) != 0)
4878 return 0;
fef9c61f 4879 mddev->reshape_position = conf->reshape_progress;
75d3da43 4880 mddev->curr_resync_completed = sector_nr;
c8f517c4 4881 conf->reshape_checkpoint = jiffies;
850b2b42 4882 set_bit(MD_CHANGE_DEVS, &mddev->flags);
52c03291 4883 md_wakeup_thread(mddev->thread);
850b2b42 4884 wait_event(mddev->sb_wait, mddev->flags == 0 ||
c91abf5a
N		 4885 test_bit(MD_RECOVERY_INTR, &mddev->recovery));
4886 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
4887 return 0;
52c03291 4888 spin_lock_irq(&conf->device_lock);
fef9c61f 4889 conf->reshape_safe = mddev->reshape_position;
52c03291
N		 4890 spin_unlock_irq(&conf->device_lock);
4891 wake_up(&conf->wait_for_overlap);
acb180b0 4892 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
52c03291
N		 4893 }
4894
ab69ae12 4895 INIT_LIST_HEAD(&stripes);
7a661381 4896 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
52c03291 4897 int j;
a9f326eb 4898 int skipped_disk = 0;
a8c906ca 4899 sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
52c03291
N		 4900 set_bit(STRIPE_EXPANDING, &sh->state);
4901 atomic_inc(&conf->reshape_stripes);
4902 /* If any of this stripe is beyond the end of the old
4903 * array, then we need to zero those blocks
4904 */
4905 for (j=sh->disks; j--;) {
4906 sector_t s;
4907 if (j == sh->pd_idx)
4908 continue;
f416885e 4909 if (conf->level == 6 &&
d0dabf7e 4910 j == sh->qd_idx)
f416885e 4911 continue;
784052ec 4912 s = compute_blocknr(sh, j, 0);
b522adcd 4913 if (s < raid5_size(mddev, 0, 0)) {
a9f326eb 4914 skipped_disk = 1;
52c03291
N		 4915 continue;
4916 }
4917 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
4918 set_bit(R5_Expanded, &sh->dev[j].flags);
4919 set_bit(R5_UPTODATE, &sh->dev[j].flags);
4920 }
a9f326eb 4921 if (!skipped_disk) {
52c03291
N		 4922 set_bit(STRIPE_EXPAND_READY, &sh->state);
4923 set_bit(STRIPE_HANDLE, &sh->state);
4924 }
ab69ae12 4925 list_add(&sh->lru, &stripes);
52c03291
N		 4926 }
4927 spin_lock_irq(&conf->device_lock);
2c810cdd 4928 if (mddev->reshape_backwards)
7a661381 4929 conf->reshape_progress -= reshape_sectors * new_data_disks;
fef9c61f 4930 else
7a661381 4931 conf->reshape_progress += reshape_sectors * new_data_disks;
52c03291
N		 4932 spin_unlock_irq(&conf->device_lock);
4933 /* Ok, those stripe are ready. We can start scheduling
4934 * reads on the source stripes.
4935 * The source stripes are determined by mapping the first and last
4936 * block on the destination stripes.
4937 */
52c03291 4938 first_sector =
ec32a2bd 4939 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
911d4ee8 4940 1, &dd_idx, NULL);
52c03291 4941 last_sector =
0e6e0271 4942 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
09c9e5fa 4943 * new_data_disks - 1),
911d4ee8 4944 1, &dd_idx, NULL);
58c0fed4
AN		 4945 if (last_sector >= mddev->dev_sectors)
4946 last_sector = mddev->dev_sectors - 1;
52c03291 4947 while (first_sector <= last_sector) {
a8c906ca 4948 sh = get_active_stripe(conf, first_sector, 1, 0, 1);
52c03291
N		 4949 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
4950 set_bit(STRIPE_HANDLE, &sh->state);
4951 release_stripe(sh);
4952 first_sector += STRIPE_SECTORS;
4953 }
ab69ae12
N		 4954 /* Now that the sources are clearly marked, we can release
4955 * the destination stripes
4956 */
4957 while (!list_empty(&stripes)) {
4958 sh = list_entry(stripes.next, struct stripe_head, lru);
4959 list_del_init(&sh->lru);
4960 release_stripe(sh);
4961 }
c6207277
N		 4962 /* If this takes us to the resync_max point where we have to pause,
4963 * then we need to write out the superblock.
4964 */
7a661381 4965 sector_nr += reshape_sectors;
c03f6a19
N		 4966 if ((sector_nr - mddev->curr_resync_completed) * 2
4967 >= mddev->resync_max - mddev->curr_resync_completed) {
c6207277
N		 4968 /* Cannot proceed until we've updated the superblock... */
4969 wait_event(conf->wait_for_overlap,
c91abf5a
N		 4970 atomic_read(&conf->reshape_stripes) == 0
4971 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
4972 if (atomic_read(&conf->reshape_stripes) != 0)
4973 goto ret;
fef9c61f 4974 mddev->reshape_position = conf->reshape_progress;
75d3da43 4975 mddev->curr_resync_completed = sector_nr;
c8f517c4 4976 conf->reshape_checkpoint = jiffies;
c6207277
N		 4977 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4978 md_wakeup_thread(mddev->thread);
4979 wait_event(mddev->sb_wait,
4980 !test_bit(MD_CHANGE_DEVS, &mddev->flags)
c91abf5a
N		 4981 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
4982 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
4983 goto ret;
c6207277 4984 spin_lock_irq(&conf->device_lock);
fef9c61f 4985 conf->reshape_safe = mddev->reshape_position;
c6207277
N		 4986 spin_unlock_irq(&conf->device_lock);
4987 wake_up(&conf->wait_for_overlap);
acb180b0 4988 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
c6207277 4989 }
c91abf5a 4990ret:
7a661381 4991 return reshape_sectors;
52c03291
N		 4992}
4993
4994/* FIXME go_faster isn't used */
fd01b88c 4995static inline sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster)
52c03291 4996{
d1688a6d 4997 struct r5conf *conf = mddev->private;
52c03291 4998 struct stripe_head *sh;
58c0fed4 4999 sector_t max_sector = mddev->dev_sectors;
57dab0bd 5000 sector_t sync_blocks;
16a53ecc
N		 5001 int still_degraded = 0;
5002 int i;
1da177e4 5003
72626685 5004 if (sector_nr >= max_sector) {
1da177e4 5005 /* just being told to finish up .. nothing much to do */
cea9c228 5006
29269553
N		 5007 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
5008 end_reshape(conf);
5009 return 0;
5010 }
72626685
N		 5011
5012 if (mddev->curr_resync < max_sector) /* aborted */
5013 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
5014 &sync_blocks, 1);
16a53ecc 5015 else /* completed sync */
72626685
N		 5016 conf->fullsync = 0;
5017 bitmap_close_sync(mddev->bitmap);
5018
1da177e4
LT		 5019 return 0;
5020 }
ccfcc3c1 5021
64bd660b
N		 5022 /* Allow raid5_quiesce to complete */
5023 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
5024
52c03291
N		 5025 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5026 return reshape_request(mddev, sector_nr, skipped);
f6705578 5027
c6207277
N		 5028 /* No need to check resync_max as we never do more than one
5029 * stripe, and as resync_max will always be on a chunk boundary,
5030 * if the check in md_do_sync didn't fire, there is no chance
5031 * of overstepping resync_max here
5032 */
5033
16a53ecc 5034 /* if there is too many failed drives and we are trying
1da177e4
LT		 5035 * to resync, then assert that we are finished, because there is
5036 * nothing we can do.
5037 */
3285edf1 5038 if (mddev->degraded >= conf->max_degraded &&
16a53ecc 5039 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
58c0fed4 5040 sector_t rv = mddev->dev_sectors - sector_nr;
57afd89f 5041 *skipped = 1;
1da177e4
LT		 5042 return rv;
5043 }
6f608040 5044 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
5045 !conf->fullsync &&
5046 !bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
5047 sync_blocks >= STRIPE_SECTORS) {
72626685
N		 5048 /* we can skip this block, and probably more */
5049 sync_blocks /= STRIPE_SECTORS;
5050 *skipped = 1;
5051 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
5052 }
1da177e4 5053
b47490c9
N		 5054 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
5055
a8c906ca 5056 sh = get_active_stripe(conf, sector_nr, 0, 1, 0);
1da177e4 5057 if (sh == NULL) {
a8c906ca 5058 sh = get_active_stripe(conf, sector_nr, 0, 0, 0);
1da177e4 5059 /* make sure we don't swamp the stripe cache if someone else
16a53ecc 5060 * is trying to get access
1da177e4 5061 */
66c006a5 5062 schedule_timeout_uninterruptible(1);
1da177e4 5063 }
16a53ecc
N		 5064 /* Need to check if array will still be degraded after recovery/resync
5065 * We don't need to check the 'failed' flag as when that gets set,
5066 * recovery aborts.
5067 */
f001a70c 5068 for (i = 0; i < conf->raid_disks; i++)
16a53ecc
N		 5069 if (conf->disks[i].rdev == NULL)
5070 still_degraded = 1;
5071
5072 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
5073
83206d66 5074 set_bit(STRIPE_SYNC_REQUESTED, &sh->state);
053f5b65 5075 set_bit(STRIPE_HANDLE, &sh->state);
1da177e4 5076
1da177e4
LT		 5077 release_stripe(sh);
5078
5079 return STRIPE_SECTORS;
5080}
5081
d1688a6d 5082static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio)
46031f9a
RBJ		 5083{
5084 /* We may not be able to submit a whole bio at once as there
5085 * may not be enough stripe_heads available.
5086 * We cannot pre-allocate enough stripe_heads as we may need
5087 * more than exist in the cache (if we allow ever large chunks).
5088 * So we do one stripe head at a time and record in
5089 * ->bi_hw_segments how many have been done.
5090 *
5091 * We *know* that this entire raid_bio is in one chunk, so
5092 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
5093 */
5094 struct stripe_head *sh;
911d4ee8 5095 int dd_idx;
46031f9a
RBJ		 5096 sector_t sector, logical_sector, last_sector;
5097 int scnt = 0;
5098 int remaining;
5099 int handled = 0;
5100
4f024f37
KO		 5101 logical_sector = raid_bio->bi_iter.bi_sector &
5102 ~((sector_t)STRIPE_SECTORS-1);
112bf897 5103 sector = raid5_compute_sector(conf, logical_sector,
911d4ee8 5104 0, &dd_idx, NULL);
f73a1c7d 5105 last_sector = bio_end_sector(raid_bio);
46031f9a
RBJ		 5106
5107 for (; logical_sector < last_sector;
387bb173
NB		 5108 logical_sector += STRIPE_SECTORS,
5109 sector += STRIPE_SECTORS,
5110 scnt++) {
46031f9a 5111
e7836bd6 5112 if (scnt < raid5_bi_processed_stripes(raid_bio))
46031f9a
RBJ		 5113 /* already done this stripe */
5114 continue;
5115
2844dc32 5116 sh = get_active_stripe(conf, sector, 0, 1, 1);
46031f9a
RBJ		 5117
5118 if (!sh) {
5119 /* failed to get a stripe - must wait */
e7836bd6 5120 raid5_set_bi_processed_stripes(raid_bio, scnt);
46031f9a
RBJ		 5121 conf->retry_read_aligned = raid_bio;
5122 return handled;
5123 }
5124
387bb173
NB		 5125 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) {
5126 release_stripe(sh);
e7836bd6 5127 raid5_set_bi_processed_stripes(raid_bio, scnt);
387bb173
NB		 5128 conf->retry_read_aligned = raid_bio;
5129 return handled;
5130 }
5131
3f9e7c14 5132 set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags);
36d1c647 5133 handle_stripe(sh);
46031f9a
RBJ		 5134 release_stripe(sh);
5135 handled++;
5136 }
e7836bd6 5137 remaining = raid5_dec_bi_active_stripes(raid_bio);
0a82a8d1
LT		 5138 if (remaining == 0) {
5139 trace_block_bio_complete(bdev_get_queue(raid_bio->bi_bdev),
5140 raid_bio, 0);
0e13fe23 5141 bio_endio(raid_bio, 0);
0a82a8d1 5142 }
46031f9a
RBJ		 5143 if (atomic_dec_and_test(&conf->active_aligned_reads))
5144 wake_up(&conf->wait_for_stripe);
5145 return handled;
5146}
5147
bfc90cb0 5148static int handle_active_stripes(struct r5conf *conf, int group,
566c09c5
SL		 5149 struct r5worker *worker,
5150 struct list_head *temp_inactive_list)
46a06401
SL		 5151{
5152 struct stripe_head *batch[MAX_STRIPE_BATCH], *sh;
566c09c5
SL		 5153 int i, batch_size = 0, hash;
5154 bool release_inactive = false;
46a06401
SL		 5155
5156 while (batch_size < MAX_STRIPE_BATCH &&
851c30c9 5157 (sh = __get_priority_stripe(conf, group)) != NULL)
46a06401
SL		 5158 batch[batch_size++] = sh;
5159
566c09c5
SL		 5160 if (batch_size == 0) {
5161 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5162 if (!list_empty(temp_inactive_list + i))
5163 break;
5164 if (i == NR_STRIPE_HASH_LOCKS)
5165 return batch_size;
5166 release_inactive = true;
5167 }
46a06401
SL		 5168 spin_unlock_irq(&conf->device_lock);
5169
566c09c5
SL		 5170 release_inactive_stripe_list(conf, temp_inactive_list,
5171 NR_STRIPE_HASH_LOCKS);
5172
5173 if (release_inactive) {
5174 spin_lock_irq(&conf->device_lock);
5175 return 0;
5176 }
5177
46a06401
SL		 5178 for (i = 0; i < batch_size; i++)
5179 handle_stripe(batch[i]);
5180
5181 cond_resched();
5182
5183 spin_lock_irq(&conf->device_lock);
566c09c5
SL		 5184 for (i = 0; i < batch_size; i++) {
5185 hash = batch[i]->hash_lock_index;
5186 __release_stripe(conf, batch[i], &temp_inactive_list[hash]);
5187 }
46a06401
SL		 5188 return batch_size;
5189}
46031f9a 5190
851c30c9
SL		 5191static void raid5_do_work(struct work_struct *work)
5192{
5193 struct r5worker *worker = container_of(work, struct r5worker, work);
5194 struct r5worker_group *group = worker->group;
5195 struct r5conf *conf = group->conf;
5196 int group_id = group - conf->worker_groups;
5197 int handled;
5198 struct blk_plug plug;
5199
5200 pr_debug("+++ raid5worker active\n");
5201
5202 blk_start_plug(&plug);
5203 handled = 0;
5204 spin_lock_irq(&conf->device_lock);
5205 while (1) {
5206 int batch_size, released;
5207
566c09c5 5208 released = release_stripe_list(conf, worker->temp_inactive_list);
851c30c9 5209
566c09c5
SL		 5210 batch_size = handle_active_stripes(conf, group_id, worker,
5211 worker->temp_inactive_list);
bfc90cb0 5212 worker->working = false;
851c30c9
SL		 5213 if (!batch_size && !released)
5214 break;
5215 handled += batch_size;
5216 }
5217 pr_debug("%d stripes handled\n", handled);
5218
5219 spin_unlock_irq(&conf->device_lock);
5220 blk_finish_plug(&plug);
5221
5222 pr_debug("--- raid5worker inactive\n");
5223}
5224
1da177e4
LT		 5225/*
5226 * This is our raid5 kernel thread.
5227 *
5228 * We scan the hash table for stripes which can be handled now.
5229 * During the scan, completed stripes are saved for us by the interrupt
5230 * handler, so that they will not have to wait for our next wakeup.
5231 */
4ed8731d 5232static void raid5d(struct md_thread *thread)
1da177e4 5233{
4ed8731d 5234 struct mddev *mddev = thread->mddev;
d1688a6d 5235 struct r5conf *conf = mddev->private;
1da177e4 5236 int handled;
e1dfa0a2 5237 struct blk_plug plug;
1da177e4 5238
45b4233c 5239 pr_debug("+++ raid5d active\n");
1da177e4
LT		 5240
5241 md_check_recovery(mddev);
1da177e4 5242
e1dfa0a2 5243 blk_start_plug(&plug);
1da177e4
LT		 5244 handled = 0;
5245 spin_lock_irq(&conf->device_lock);
5246 while (1) {
46031f9a 5247 struct bio *bio;
773ca82f
SL		 5248 int batch_size, released;
5249
566c09c5 5250 released = release_stripe_list(conf, conf->temp_inactive_list);
1da177e4 5251
0021b7bc 5252 if (
7c13edc8
N		 5253 !list_empty(&conf->bitmap_list)) {
5254 /* Now is a good time to flush some bitmap updates */
5255 conf->seq_flush++;
700e432d 5256 spin_unlock_irq(&conf->device_lock);
72626685 5257 bitmap_unplug(mddev->bitmap);
700e432d 5258 spin_lock_irq(&conf->device_lock);
7c13edc8 5259 conf->seq_write = conf->seq_flush;
566c09c5 5260 activate_bit_delay(conf, conf->temp_inactive_list);
72626685 5261 }
0021b7bc 5262 raid5_activate_delayed(conf);
72626685 5263
46031f9a
RBJ		 5264 while ((bio = remove_bio_from_retry(conf))) {
5265 int ok;
5266 spin_unlock_irq(&conf->device_lock);
5267 ok = retry_aligned_read(conf, bio);
5268 spin_lock_irq(&conf->device_lock);
5269 if (!ok)
5270 break;
5271 handled++;
5272 }
5273
566c09c5
SL		 5274 batch_size = handle_active_stripes(conf, ANY_GROUP, NULL,
5275 conf->temp_inactive_list);
773ca82f 5276 if (!batch_size && !released)
1da177e4 5277 break;
46a06401 5278 handled += batch_size;
1da177e4 5279
46a06401
SL		 5280 if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) {
5281 spin_unlock_irq(&conf->device_lock);
de393cde 5282 md_check_recovery(mddev);
46a06401
SL		 5283 spin_lock_irq(&conf->device_lock);
5284 }
1da177e4 5285 }
45b4233c 5286 pr_debug("%d stripes handled\n", handled);
1da177e4
LT		 5287
5288 spin_unlock_irq(&conf->device_lock);
5289
c9f21aaf 5290 async_tx_issue_pending_all();
e1dfa0a2 5291 blk_finish_plug(&plug);
1da177e4 5292
45b4233c 5293 pr_debug("--- raid5d inactive\n");
1da177e4
LT		 5294}
5295
3f294f4f 5296static ssize_t
fd01b88c 5297raid5_show_stripe_cache_size(struct mddev *mddev, char *page)
3f294f4f 5298{
d1688a6d 5299 struct r5conf *conf = mddev->private;
96de1e66
N		 5300 if (conf)
5301 return sprintf(page, "%d\n", conf->max_nr_stripes);
5302 else
5303 return 0;
3f294f4f
N		 5304}
5305
c41d4ac4 5306int
fd01b88c 5307raid5_set_cache_size(struct mddev *mddev, int size)
3f294f4f 5308{
d1688a6d 5309 struct r5conf *conf = mddev->private;
b5470dc5 5310 int err;
566c09c5 5311 int hash;
b5470dc5 5312
c41d4ac4 5313 if (size <= 16 || size > 32768)
3f294f4f 5314 return -EINVAL;
566c09c5 5315 hash = (conf->max_nr_stripes - 1) % NR_STRIPE_HASH_LOCKS;
c41d4ac4 5316 while (size < conf->max_nr_stripes) {
566c09c5 5317 if (drop_one_stripe(conf, hash))
3f294f4f
N		 5318 conf->max_nr_stripes--;
5319 else
5320 break;
566c09c5
SL		 5321 hash--;
5322 if (hash < 0)
5323 hash = NR_STRIPE_HASH_LOCKS - 1;
3f294f4f 5324 }
b5470dc5
DW		 5325 err = md_allow_write(mddev);
5326 if (err)
5327 return err;
566c09c5 5328 hash = conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS;
c41d4ac4 5329 while (size > conf->max_nr_stripes) {
566c09c5 5330 if (grow_one_stripe(conf, hash))
3f294f4f
N		 5331 conf->max_nr_stripes++;
5332 else break;
566c09c5 5333 hash = (hash + 1) % NR_STRIPE_HASH_LOCKS;
3f294f4f 5334 }
c41d4ac4
N		 5335 return 0;
5336}
5337EXPORT_SYMBOL(raid5_set_cache_size);
5338
5339static ssize_t
fd01b88c 5340raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len)
c41d4ac4 5341{
d1688a6d 5342 struct r5conf *conf = mddev->private;
c41d4ac4
N		 5343 unsigned long new;
5344 int err;
5345
5346 if (len >= PAGE_SIZE)
5347 return -EINVAL;
5348 if (!conf)
5349 return -ENODEV;
5350
b29bebd6 5351 if (kstrtoul(page, 10, &new))
c41d4ac4
N		 5352 return -EINVAL;
5353 err = raid5_set_cache_size(mddev, new);
5354 if (err)
5355 return err;
3f294f4f
N		 5356 return len;
5357}
007583c9 5358
96de1e66
N		 5359static struct md_sysfs_entry
5360raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
5361 raid5_show_stripe_cache_size,
5362 raid5_store_stripe_cache_size);
3f294f4f 5363
8b3e6cdc 5364static ssize_t
fd01b88c 5365raid5_show_preread_threshold(struct mddev *mddev, char *page)
8b3e6cdc 5366{
d1688a6d 5367 struct r5conf *conf = mddev->private;
8b3e6cdc
DW		 5368 if (conf)
5369 return sprintf(page, "%d\n", conf->bypass_threshold);
5370 else
5371 return 0;
5372}
5373
5374static ssize_t
fd01b88c 5375raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len)
8b3e6cdc 5376{
d1688a6d 5377 struct r5conf *conf = mddev->private;
4ef197d8 5378 unsigned long new;
8b3e6cdc
DW		 5379 if (len >= PAGE_SIZE)
5380 return -EINVAL;
5381 if (!conf)
5382 return -ENODEV;
5383
b29bebd6 5384 if (kstrtoul(page, 10, &new))
8b3e6cdc 5385 return -EINVAL;
4ef197d8 5386 if (new > conf->max_nr_stripes)
8b3e6cdc
DW		 5387 return -EINVAL;
5388 conf->bypass_threshold = new;
5389 return len;
5390}
5391
5392static struct md_sysfs_entry
5393raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
5394 S_IRUGO | S_IWUSR,
5395 raid5_show_preread_threshold,
5396 raid5_store_preread_threshold);
5397
d592a996
SL		 5398static ssize_t
5399raid5_show_skip_copy(struct mddev *mddev, char *page)
5400{
5401 struct r5conf *conf = mddev->private;
5402 if (conf)
5403 return sprintf(page, "%d\n", conf->skip_copy);
5404 else
5405 return 0;
5406}
5407
5408static ssize_t
5409raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len)
5410{
5411 struct r5conf *conf = mddev->private;
5412 unsigned long new;
5413 if (len >= PAGE_SIZE)
5414 return -EINVAL;
5415 if (!conf)
5416 return -ENODEV;
5417
5418 if (kstrtoul(page, 10, &new))
5419 return -EINVAL;
5420 new = !!new;
5421 if (new == conf->skip_copy)
5422 return len;
5423
5424 mddev_suspend(mddev);
5425 conf->skip_copy = new;
5426 if (new)
5427 mddev->queue->backing_dev_info.capabilities |=
5428 BDI_CAP_STABLE_WRITES;
5429 else
5430 mddev->queue->backing_dev_info.capabilities &=
5431 ~BDI_CAP_STABLE_WRITES;
5432 mddev_resume(mddev);
5433 return len;
5434}
5435
5436static struct md_sysfs_entry
5437raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR,
5438 raid5_show_skip_copy,
5439 raid5_store_skip_copy);
5440
3f294f4f 5441static ssize_t
fd01b88c 5442stripe_cache_active_show(struct mddev *mddev, char *page)
3f294f4f 5443{
d1688a6d 5444 struct r5conf *conf = mddev->private;
96de1e66
N		 5445 if (conf)
5446 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
5447 else
5448 return 0;
3f294f4f
N		 5449}
5450
96de1e66
N		 5451static struct md_sysfs_entry
5452raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
3f294f4f 5453
b721420e
SL		 5454static ssize_t
5455raid5_show_group_thread_cnt(struct mddev *mddev, char *page)
5456{
5457 struct r5conf *conf = mddev->private;
5458 if (conf)
5459 return sprintf(page, "%d\n", conf->worker_cnt_per_group);
5460 else
5461 return 0;
5462}
5463
60aaf933 5464static int alloc_thread_groups(struct r5conf *conf, int cnt,
5465 int *group_cnt,
5466 int *worker_cnt_per_group,
5467 struct r5worker_group **worker_groups);
b721420e
SL		 5468static ssize_t
5469raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len)
5470{
5471 struct r5conf *conf = mddev->private;
5472 unsigned long new;
5473 int err;
60aaf933 5474 struct r5worker_group *new_groups, *old_groups;
5475 int group_cnt, worker_cnt_per_group;
b721420e
SL		 5476
5477 if (len >= PAGE_SIZE)
5478 return -EINVAL;
5479 if (!conf)
5480 return -ENODEV;
5481
5482 if (kstrtoul(page, 10, &new))
5483 return -EINVAL;
5484
5485 if (new == conf->worker_cnt_per_group)
5486 return len;
5487
5488 mddev_suspend(mddev);
5489
5490 old_groups = conf->worker_groups;
d206dcfa 5491 if (old_groups)
5492 flush_workqueue(raid5_wq);
5493
60aaf933 5494 err = alloc_thread_groups(conf, new,
5495 &group_cnt, &worker_cnt_per_group,
5496 &new_groups);
5497 if (!err) {
5498 spin_lock_irq(&conf->device_lock);
5499 conf->group_cnt = group_cnt;
5500 conf->worker_cnt_per_group = worker_cnt_per_group;
5501 conf->worker_groups = new_groups;
5502 spin_unlock_irq(&conf->device_lock);
b721420e 5503
b721420e
SL		 5504 if (old_groups)
5505 kfree(old_groups[0].workers);
5506 kfree(old_groups);
5507 }
5508
5509 mddev_resume(mddev);
5510
5511 if (err)
5512 return err;
5513 return len;
5514}
5515
5516static struct md_sysfs_entry
5517raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR,
5518 raid5_show_group_thread_cnt,
5519 raid5_store_group_thread_cnt);
5520
007583c9 5521static struct attribute *raid5_attrs[] = {
3f294f4f
N		 5522 &raid5_stripecache_size.attr,
5523 &raid5_stripecache_active.attr,
8b3e6cdc 5524 &raid5_preread_bypass_threshold.attr,
b721420e 5525 &raid5_group_thread_cnt.attr,
d592a996 5526 &raid5_skip_copy.attr,
3f294f4f
N		 5527 NULL,
5528};
007583c9
N		 5529static struct attribute_group raid5_attrs_group = {
5530 .name = NULL,
5531 .attrs = raid5_attrs,
3f294f4f
N		 5532};
5533
60aaf933 5534static int alloc_thread_groups(struct r5conf *conf, int cnt,
5535 int *group_cnt,
5536 int *worker_cnt_per_group,
5537 struct r5worker_group **worker_groups)
851c30c9 5538{
566c09c5 5539 int i, j, k;
851c30c9
SL		 5540 ssize_t size;
5541 struct r5worker *workers;
5542
60aaf933 5543 *worker_cnt_per_group = cnt;
851c30c9 5544 if (cnt == 0) {
60aaf933 5545 *group_cnt = 0;
5546 *worker_groups = NULL;
851c30c9
SL		 5547 return 0;
5548 }
60aaf933 5549 *group_cnt = num_possible_nodes();
851c30c9 5550 size = sizeof(struct r5worker) * cnt;
60aaf933 5551 workers = kzalloc(size * *group_cnt, GFP_NOIO);
5552 *worker_groups = kzalloc(sizeof(struct r5worker_group) *
5553 *group_cnt, GFP_NOIO);
5554 if (!*worker_groups || !workers) {
851c30c9 5555 kfree(workers);
60aaf933 5556 kfree(*worker_groups);
851c30c9
SL		 5557 return -ENOMEM;
5558 }
5559
60aaf933 5560 for (i = 0; i < *group_cnt; i++) {
851c30c9
SL		 5561 struct r5worker_group *group;
5562
0c775d52 5563 group = &(*worker_groups)[i];
851c30c9
SL		 5564 INIT_LIST_HEAD(&group->handle_list);
5565 group->conf = conf;
5566 group->workers = workers + i * cnt;
5567
5568 for (j = 0; j < cnt; j++) {
566c09c5
SL		 5569 struct r5worker *worker = group->workers + j;
5570 worker->group = group;
5571 INIT_WORK(&worker->work, raid5_do_work);
5572
5573 for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++)
5574 INIT_LIST_HEAD(worker->temp_inactive_list + k);
851c30c9
SL		 5575 }
5576 }
5577
5578 return 0;
5579}
5580
5581static void free_thread_groups(struct r5conf *conf)
5582{
5583 if (conf->worker_groups)
5584 kfree(conf->worker_groups[0].workers);
5585 kfree(conf->worker_groups);
5586 conf->worker_groups = NULL;
5587}
5588
80c3a6ce 5589static sector_t
fd01b88c 5590raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks)
80c3a6ce 5591{
d1688a6d 5592 struct r5conf *conf = mddev->private;
80c3a6ce
DW		 5593
5594 if (!sectors)
5595 sectors = mddev->dev_sectors;
5e5e3e78 5596 if (!raid_disks)
7ec05478 5597 /* size is defined by the smallest of previous and new size */
5e5e3e78 5598 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
80c3a6ce 5599
9d8f0363 5600 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
664e7c41 5601 sectors &= ~((sector_t)mddev->new_chunk_sectors - 1);
80c3a6ce
DW		 5602 return sectors * (raid_disks - conf->max_degraded);
5603}
5604
789b5e03
ON		 5605static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
5606{
5607 safe_put_page(percpu->spare_page);
5608 kfree(percpu->scribble);
5609 percpu->spare_page = NULL;
5610 percpu->scribble = NULL;
5611}
5612
5613static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
5614{
5615 if (conf->level == 6 && !percpu->spare_page)
5616 percpu->spare_page = alloc_page(GFP_KERNEL);
5617 if (!percpu->scribble)
5618 percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
5619
5620 if (!percpu->scribble || (conf->level == 6 && !percpu->spare_page)) {
5621 free_scratch_buffer(conf, percpu);
5622 return -ENOMEM;
5623 }
5624
5625 return 0;
5626}
5627
d1688a6d 5628static void raid5_free_percpu(struct r5conf *conf)
36d1c647 5629{
36d1c647
DW		 5630 unsigned long cpu;
5631
5632 if (!conf->percpu)
5633 return;
5634
36d1c647
DW		 5635#ifdef CONFIG_HOTPLUG_CPU
5636 unregister_cpu_notifier(&conf->cpu_notify);
5637#endif
789b5e03
ON		 5638
5639 get_online_cpus();
5640 for_each_possible_cpu(cpu)
5641 free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
36d1c647
DW		 5642 put_online_cpus();
5643
5644 free_percpu(conf->percpu);
5645}
5646
d1688a6d 5647static void free_conf(struct r5conf *conf)
95fc17aa 5648{
851c30c9 5649 free_thread_groups(conf);
95fc17aa 5650 shrink_stripes(conf);
36d1c647 5651 raid5_free_percpu(conf);
95fc17aa
DW		 5652 kfree(conf->disks);
5653 kfree(conf->stripe_hashtbl);
5654 kfree(conf);
5655}
5656
36d1c647
DW		 5657#ifdef CONFIG_HOTPLUG_CPU
5658static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action,
5659 void *hcpu)
5660{
d1688a6d 5661 struct r5conf *conf = container_of(nfb, struct r5conf, cpu_notify);
36d1c647
DW		 5662 long cpu = (long)hcpu;
5663 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
5664
5665 switch (action) {
5666 case CPU_UP_PREPARE:
5667 case CPU_UP_PREPARE_FROZEN:
789b5e03 5668 if (alloc_scratch_buffer(conf, percpu)) {
36d1c647
DW		 5669 pr_err("%s: failed memory allocation for cpu%ld\n",
5670 __func__, cpu);
55af6bb5 5671 return notifier_from_errno(-ENOMEM);
36d1c647
DW		 5672 }
5673 break;
5674 case CPU_DEAD:
5675 case CPU_DEAD_FROZEN:
789b5e03 5676 free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
36d1c647
DW		 5677 break;
5678 default:
5679 break;
5680 }
5681 return NOTIFY_OK;
5682}
5683#endif
5684
d1688a6d 5685static int raid5_alloc_percpu(struct r5conf *conf)
36d1c647
DW		 5686{
5687 unsigned long cpu;
789b5e03 5688 int err = 0;
36d1c647 5689
789b5e03
ON		 5690 conf->percpu = alloc_percpu(struct raid5_percpu);
5691 if (!conf->percpu)
36d1c647 5692 return -ENOMEM;
789b5e03
ON		 5693
5694#ifdef CONFIG_HOTPLUG_CPU
5695 conf->cpu_notify.notifier_call = raid456_cpu_notify;
5696 conf->cpu_notify.priority = 0;
5697 err = register_cpu_notifier(&conf->cpu_notify);
5698 if (err)
5699 return err;
5700#endif
36d1c647
DW		 5701
5702 get_online_cpus();
36d1c647 5703 for_each_present_cpu(cpu) {
789b5e03
ON		 5704 err = alloc_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
5705 if (err) {
5706 pr_err("%s: failed memory allocation for cpu%ld\n",
5707 __func__, cpu);
36d1c647
DW		 5708 break;
5709 }
36d1c647 5710 }
36d1c647
DW		 5711 put_online_cpus();
5712
5713 return err;
5714}
5715
d1688a6d 5716static struct r5conf *setup_conf(struct mddev *mddev)
1da177e4 5717{
d1688a6d 5718 struct r5conf *conf;
5e5e3e78 5719 int raid_disk, memory, max_disks;
3cb03002 5720 struct md_rdev *rdev;
1da177e4 5721 struct disk_info *disk;
0232605d 5722 char pers_name[6];
566c09c5 5723 int i;
60aaf933 5724 int group_cnt, worker_cnt_per_group;
5725 struct r5worker_group *new_group;
1da177e4 5726
91adb564
N		 5727 if (mddev->new_level != 5
5728 && mddev->new_level != 4
5729 && mddev->new_level != 6) {
0c55e022 5730 printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n",
91adb564
N		 5731 mdname(mddev), mddev->new_level);
5732 return ERR_PTR(-EIO);
1da177e4 5733 }
91adb564
N		 5734 if ((mddev->new_level == 5
5735 && !algorithm_valid_raid5(mddev->new_layout)) ||
5736 (mddev->new_level == 6
5737 && !algorithm_valid_raid6(mddev->new_layout))) {
0c55e022 5738 printk(KERN_ERR "md/raid:%s: layout %d not supported\n",
91adb564
N		 5739 mdname(mddev), mddev->new_layout);
5740 return ERR_PTR(-EIO);
99c0fb5f 5741 }
91adb564 5742 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
0c55e022 5743 printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n",
91adb564
N		 5744 mdname(mddev), mddev->raid_disks);
5745 return ERR_PTR(-EINVAL);
4bbf3771
N		 5746 }
5747
664e7c41
AN		 5748 if (!mddev->new_chunk_sectors ||
5749 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
5750 !is_power_of_2(mddev->new_chunk_sectors)) {
0c55e022
N		 5751 printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n",
5752 mdname(mddev), mddev->new_chunk_sectors << 9);
91adb564 5753 return ERR_PTR(-EINVAL);
f6705578
N		 5754 }
5755
d1688a6d 5756 conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL);
91adb564 5757 if (conf == NULL)
1da177e4 5758 goto abort;
851c30c9 5759 /* Don't enable multi-threading by default*/
60aaf933 5760 if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group,
5761 &new_group)) {
5762 conf->group_cnt = group_cnt;
5763 conf->worker_cnt_per_group = worker_cnt_per_group;
5764 conf->worker_groups = new_group;
5765 } else
851c30c9 5766 goto abort;
f5efd45a 5767 spin_lock_init(&conf->device_lock);
c46501b2 5768 seqcount_init(&conf->gen_lock);
f5efd45a
DW		 5769 init_waitqueue_head(&conf->wait_for_stripe);
5770 init_waitqueue_head(&conf->wait_for_overlap);
5771 INIT_LIST_HEAD(&conf->handle_list);
5772 INIT_LIST_HEAD(&conf->hold_list);
5773 INIT_LIST_HEAD(&conf->delayed_list);
5774 INIT_LIST_HEAD(&conf->bitmap_list);
773ca82f 5775 init_llist_head(&conf->released_stripes);
f5efd45a
DW		 5776 atomic_set(&conf->active_stripes, 0);
5777 atomic_set(&conf->preread_active_stripes, 0);
5778 atomic_set(&conf->active_aligned_reads, 0);
5779 conf->bypass_threshold = BYPASS_THRESHOLD;
d890fa2b 5780 conf->recovery_disabled = mddev->recovery_disabled - 1;
91adb564
N		 5781
5782 conf->raid_disks = mddev->raid_disks;
5783 if (mddev->reshape_position == MaxSector)
5784 conf->previous_raid_disks = mddev->raid_disks;
5785 else
f6705578 5786 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
5e5e3e78
N		 5787 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
5788 conf->scribble_len = scribble_len(max_disks);
f6705578 5789
5e5e3e78 5790 conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
b55e6bfc
N		 5791 GFP_KERNEL);
5792 if (!conf->disks)
5793 goto abort;
9ffae0cf 5794
1da177e4
LT		 5795 conf->mddev = mddev;
5796
fccddba0 5797 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
1da177e4 5798 goto abort;
1da177e4 5799
566c09c5
SL		 5800 /* We init hash_locks[0] separately to that it can be used
5801 * as the reference lock in the spin_lock_nest_lock() call
5802 * in lock_all_device_hash_locks_irq in order to convince
5803 * lockdep that we know what we are doing.
5804 */
5805 spin_lock_init(conf->hash_locks);
5806 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
5807 spin_lock_init(conf->hash_locks + i);
5808
5809 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5810 INIT_LIST_HEAD(conf->inactive_list + i);
5811
5812 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5813 INIT_LIST_HEAD(conf->temp_inactive_list + i);
5814
36d1c647
DW		 5815 conf->level = mddev->new_level;
5816 if (raid5_alloc_percpu(conf) != 0)
5817 goto abort;
5818
0c55e022 5819 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
1da177e4 5820
dafb20fa 5821 rdev_for_each(rdev, mddev) {
1da177e4 5822 raid_disk = rdev->raid_disk;
5e5e3e78 5823 if (raid_disk >= max_disks
1da177e4
LT		 5824 || raid_disk < 0)
5825 continue;
5826 disk = conf->disks + raid_disk;
5827
17045f52
N		 5828 if (test_bit(Replacement, &rdev->flags)) {
5829 if (disk->replacement)
5830 goto abort;
5831 disk->replacement = rdev;
5832 } else {
5833 if (disk->rdev)
5834 goto abort;
5835 disk->rdev = rdev;
5836 }
1da177e4 5837
b2d444d7 5838 if (test_bit(In_sync, &rdev->flags)) {
1da177e4 5839 char b[BDEVNAME_SIZE];
0c55e022
N		 5840 printk(KERN_INFO "md/raid:%s: device %s operational as raid"
5841 " disk %d\n",
5842 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
d6b212f4 5843 } else if (rdev->saved_raid_disk != raid_disk)
8c2e870a
NB		 5844 /* Cannot rely on bitmap to complete recovery */
5845 conf->fullsync = 1;
1da177e4
LT		 5846 }
5847
09c9e5fa 5848 conf->chunk_sectors = mddev->new_chunk_sectors;
91adb564 5849 conf->level = mddev->new_level;
16a53ecc
N		 5850 if (conf->level == 6)
5851 conf->max_degraded = 2;
5852 else
5853 conf->max_degraded = 1;
91adb564 5854 conf->algorithm = mddev->new_layout;
fef9c61f 5855 conf->reshape_progress = mddev->reshape_position;
e183eaed 5856 if (conf->reshape_progress != MaxSector) {
09c9e5fa 5857 conf->prev_chunk_sectors = mddev->chunk_sectors;
e183eaed
N		 5858 conf->prev_algo = mddev->layout;
5859 }
1da177e4 5860
91adb564 5861 memory = conf->max_nr_stripes * (sizeof(struct stripe_head) +
5e5e3e78 5862 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
4bda556a 5863 atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS);
566c09c5 5864 if (grow_stripes(conf, NR_STRIPES)) {
91adb564 5865 printk(KERN_ERR
0c55e022
N		 5866 "md/raid:%s: couldn't allocate %dkB for buffers\n",
5867 mdname(mddev), memory);
91adb564
N		 5868 goto abort;
5869 } else
0c55e022
N		 5870 printk(KERN_INFO "md/raid:%s: allocated %dkB\n",
5871 mdname(mddev), memory);
1da177e4 5872
0232605d
N		 5873 sprintf(pers_name, "raid%d", mddev->new_level);
5874 conf->thread = md_register_thread(raid5d, mddev, pers_name);
91adb564
N		 5875 if (!conf->thread) {
5876 printk(KERN_ERR
0c55e022 5877 "md/raid:%s: couldn't allocate thread.\n",
91adb564 5878 mdname(mddev));
16a53ecc
N		 5879 goto abort;
5880 }
91adb564
N		 5881
5882 return conf;
5883
5884 abort:
5885 if (conf) {
95fc17aa 5886 free_conf(conf);
91adb564
N		 5887 return ERR_PTR(-EIO);
5888 } else
5889 return ERR_PTR(-ENOMEM);
5890}
5891
c148ffdc
N		 5892static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
5893{
5894 switch (algo) {
5895 case ALGORITHM_PARITY_0:
5896 if (raid_disk < max_degraded)
5897 return 1;
5898 break;
5899 case ALGORITHM_PARITY_N:
5900 if (raid_disk >= raid_disks - max_degraded)
5901 return 1;
5902 break;
5903 case ALGORITHM_PARITY_0_6:
f72ffdd6 5904 if (raid_disk == 0 ||
c148ffdc
N		 5905 raid_disk == raid_disks - 1)
5906 return 1;
5907 break;
5908 case ALGORITHM_LEFT_ASYMMETRIC_6:
5909 case ALGORITHM_RIGHT_ASYMMETRIC_6:
5910 case ALGORITHM_LEFT_SYMMETRIC_6:
5911 case ALGORITHM_RIGHT_SYMMETRIC_6:
5912 if (raid_disk == raid_disks - 1)
5913 return 1;
5914 }
5915 return 0;
5916}
5917
fd01b88c 5918static int run(struct mddev *mddev)
91adb564 5919{
d1688a6d 5920 struct r5conf *conf;
9f7c2220 5921 int working_disks = 0;
c148ffdc 5922 int dirty_parity_disks = 0;
3cb03002 5923 struct md_rdev *rdev;
c148ffdc 5924 sector_t reshape_offset = 0;
17045f52 5925 int i;
b5254dd5
N		 5926 long long min_offset_diff = 0;
5927 int first = 1;
91adb564 5928
8c6ac868 5929 if (mddev->recovery_cp != MaxSector)
0c55e022 5930 printk(KERN_NOTICE "md/raid:%s: not clean"
8c6ac868
AN		 5931 " -- starting background reconstruction\n",
5932 mdname(mddev));
b5254dd5
N		 5933
5934 rdev_for_each(rdev, mddev) {
5935 long long diff;
5936 if (rdev->raid_disk < 0)
5937 continue;
5938 diff = (rdev->new_data_offset - rdev->data_offset);
5939 if (first) {
5940 min_offset_diff = diff;
5941 first = 0;
5942 } else if (mddev->reshape_backwards &&
5943 diff < min_offset_diff)
5944 min_offset_diff = diff;
5945 else if (!mddev->reshape_backwards &&
5946 diff > min_offset_diff)
5947 min_offset_diff = diff;
5948 }
5949
91adb564
N		 5950 if (mddev->reshape_position != MaxSector) {
5951 /* Check that we can continue the reshape.
b5254dd5
N		 5952 * Difficulties arise if the stripe we would write to
5953 * next is at or after the stripe we would read from next.
5954 * For a reshape that changes the number of devices, this
5955 * is only possible for a very short time, and mdadm makes
5956 * sure that time appears to have past before assembling
5957 * the array. So we fail if that time hasn't passed.
5958 * For a reshape that keeps the number of devices the same
5959 * mdadm must be monitoring the reshape can keeping the
5960 * critical areas read-only and backed up. It will start
5961 * the array in read-only mode, so we check for that.
91adb564
N		 5962 */
5963 sector_t here_new, here_old;
5964 int old_disks;
18b00334 5965 int max_degraded = (mddev->level == 6 ? 2 : 1);
91adb564 5966
88ce4930 5967 if (mddev->new_level != mddev->level) {
0c55e022 5968 printk(KERN_ERR "md/raid:%s: unsupported reshape "
91adb564
N		 5969 "required - aborting.\n",
5970 mdname(mddev));
5971 return -EINVAL;
5972 }
91adb564
N		 5973 old_disks = mddev->raid_disks - mddev->delta_disks;
5974 /* reshape_position must be on a new-stripe boundary, and one
5975 * further up in new geometry must map after here in old
5976 * geometry.
5977 */
5978 here_new = mddev->reshape_position;
664e7c41 5979 if (sector_div(here_new, mddev->new_chunk_sectors *
91adb564 5980 (mddev->raid_disks - max_degraded))) {
0c55e022
N		 5981 printk(KERN_ERR "md/raid:%s: reshape_position not "
5982 "on a stripe boundary\n", mdname(mddev));
91adb564
N		 5983 return -EINVAL;
5984 }
c148ffdc 5985 reshape_offset = here_new * mddev->new_chunk_sectors;
91adb564
N		 5986 /* here_new is the stripe we will write to */
5987 here_old = mddev->reshape_position;
9d8f0363 5988 sector_div(here_old, mddev->chunk_sectors *
91adb564
N		 5989 (old_disks-max_degraded));
5990 /* here_old is the first stripe that we might need to read
5991 * from */
67ac6011 5992 if (mddev->delta_disks == 0) {
b5254dd5
N		 5993 if ((here_new * mddev->new_chunk_sectors !=
5994 here_old * mddev->chunk_sectors)) {
5995 printk(KERN_ERR "md/raid:%s: reshape position is"
5996 " confused - aborting\n", mdname(mddev));
5997 return -EINVAL;
5998 }
67ac6011 5999 /* We cannot be sure it is safe to start an in-place
b5254dd5 6000 * reshape. It is only safe if user-space is monitoring
67ac6011
N		 6001 * and taking constant backups.
6002 * mdadm always starts a situation like this in
6003 * readonly mode so it can take control before
6004 * allowing any writes. So just check for that.
6005 */
b5254dd5
N		 6006 if (abs(min_offset_diff) >= mddev->chunk_sectors &&
6007 abs(min_offset_diff) >= mddev->new_chunk_sectors)
6008 /* not really in-place - so OK */;
6009 else if (mddev->ro == 0) {
6010 printk(KERN_ERR "md/raid:%s: in-place reshape "
6011 "must be started in read-only mode "
6012 "- aborting\n",
0c55e022 6013 mdname(mddev));
67ac6011
N		 6014 return -EINVAL;
6015 }
2c810cdd 6016 } else if (mddev->reshape_backwards
b5254dd5 6017 ? (here_new * mddev->new_chunk_sectors + min_offset_diff <=
67ac6011
N		 6018 here_old * mddev->chunk_sectors)
6019 : (here_new * mddev->new_chunk_sectors >=
b5254dd5 6020 here_old * mddev->chunk_sectors + (-min_offset_diff))) {
91adb564 6021 /* Reading from the same stripe as writing to - bad */
0c55e022
N		 6022 printk(KERN_ERR "md/raid:%s: reshape_position too early for "
6023 "auto-recovery - aborting.\n",
6024 mdname(mddev));
91adb564
N		 6025 return -EINVAL;
6026 }
0c55e022
N		 6027 printk(KERN_INFO "md/raid:%s: reshape will continue\n",
6028 mdname(mddev));
91adb564
N		 6029 /* OK, we should be able to continue; */
6030 } else {
6031 BUG_ON(mddev->level != mddev->new_level);
6032 BUG_ON(mddev->layout != mddev->new_layout);
664e7c41 6033 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
91adb564 6034 BUG_ON(mddev->delta_disks != 0);
1da177e4 6035 }
91adb564 6036
245f46c2
N		 6037 if (mddev->private == NULL)
6038 conf = setup_conf(mddev);
6039 else
6040 conf = mddev->private;
6041
91adb564
N		 6042 if (IS_ERR(conf))
6043 return PTR_ERR(conf);
6044
b5254dd5 6045 conf->min_offset_diff = min_offset_diff;
91adb564
N		 6046 mddev->thread = conf->thread;
6047 conf->thread = NULL;
6048 mddev->private = conf;
6049
17045f52
N		 6050 for (i = 0; i < conf->raid_disks && conf->previous_raid_disks;
6051 i++) {
6052 rdev = conf->disks[i].rdev;
6053 if (!rdev && conf->disks[i].replacement) {
6054 /* The replacement is all we have yet */
6055 rdev = conf->disks[i].replacement;
6056 conf->disks[i].replacement = NULL;
6057 clear_bit(Replacement, &rdev->flags);
6058 conf->disks[i].rdev = rdev;
6059 }
6060 if (!rdev)
c148ffdc 6061 continue;
17045f52
N		 6062 if (conf->disks[i].replacement &&
6063 conf->reshape_progress != MaxSector) {
6064 /* replacements and reshape simply do not mix. */
6065 printk(KERN_ERR "md: cannot handle concurrent "
6066 "replacement and reshape.\n");
6067 goto abort;
6068 }
2f115882 6069 if (test_bit(In_sync, &rdev->flags)) {
91adb564 6070 working_disks++;
2f115882
N		 6071 continue;
6072 }
c148ffdc
N		 6073 /* This disc is not fully in-sync. However if it
6074 * just stored parity (beyond the recovery_offset),
6075 * when we don't need to be concerned about the
6076 * array being dirty.
6077 * When reshape goes 'backwards', we never have
6078 * partially completed devices, so we only need
6079 * to worry about reshape going forwards.
6080 */
6081 /* Hack because v0.91 doesn't store recovery_offset properly. */
6082 if (mddev->major_version == 0 &&
6083 mddev->minor_version > 90)
6084 rdev->recovery_offset = reshape_offset;
5026d7a9 6085
c148ffdc
N		 6086 if (rdev->recovery_offset < reshape_offset) {
6087 /* We need to check old and new layout */
6088 if (!only_parity(rdev->raid_disk,
6089 conf->algorithm,
6090 conf->raid_disks,
6091 conf->max_degraded))
6092 continue;
6093 }
6094 if (!only_parity(rdev->raid_disk,
6095 conf->prev_algo,
6096 conf->previous_raid_disks,
6097 conf->max_degraded))
6098 continue;
6099 dirty_parity_disks++;
6100 }
91adb564 6101
17045f52
N		 6102 /*
6103 * 0 for a fully functional array, 1 or 2 for a degraded array.
6104 */
908f4fbd 6105 mddev->degraded = calc_degraded(conf);
91adb564 6106
674806d6 6107 if (has_failed(conf)) {
0c55e022 6108 printk(KERN_ERR "md/raid:%s: not enough operational devices"
1da177e4 6109 " (%d/%d failed)\n",
02c2de8c 6110 mdname(mddev), mddev->degraded, conf->raid_disks);
1da177e4
LT		 6111 goto abort;
6112 }
6113
91adb564 6114 /* device size must be a multiple of chunk size */
9d8f0363 6115 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
91adb564
N		 6116 mddev->resync_max_sectors = mddev->dev_sectors;
6117
c148ffdc 6118 if (mddev->degraded > dirty_parity_disks &&
1da177e4 6119 mddev->recovery_cp != MaxSector) {
6ff8d8ec
N		 6120 if (mddev->ok_start_degraded)
6121 printk(KERN_WARNING
0c55e022
N		 6122 "md/raid:%s: starting dirty degraded array"
6123 " - data corruption possible.\n",
6ff8d8ec
N		 6124 mdname(mddev));
6125 else {
6126 printk(KERN_ERR
0c55e022 6127 "md/raid:%s: cannot start dirty degraded array.\n",
6ff8d8ec
N		 6128 mdname(mddev));
6129 goto abort;
6130 }
1da177e4
LT		 6131 }
6132
1da177e4 6133 if (mddev->degraded == 0)
0c55e022
N		 6134 printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d"
6135 " devices, algorithm %d\n", mdname(mddev), conf->level,
e183eaed
N		 6136 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
6137 mddev->new_layout);
1da177e4 6138 else
0c55e022
N		 6139 printk(KERN_ALERT "md/raid:%s: raid level %d active with %d"
6140 " out of %d devices, algorithm %d\n",
6141 mdname(mddev), conf->level,
6142 mddev->raid_disks - mddev->degraded,
6143 mddev->raid_disks, mddev->new_layout);
1da177e4
LT		 6144
6145 print_raid5_conf(conf);
6146
fef9c61f 6147 if (conf->reshape_progress != MaxSector) {
fef9c61f 6148 conf->reshape_safe = conf->reshape_progress;
f6705578
N		 6149 atomic_set(&conf->reshape_stripes, 0);
6150 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6151 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6152 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6153 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6154 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 6155 "reshape");
f6705578
N		 6156 }
6157
1da177e4 6158 /* Ok, everything is just fine now */
a64c876f
N		 6159 if (mddev->to_remove == &raid5_attrs_group)
6160 mddev->to_remove = NULL;
00bcb4ac
N		 6161 else if (mddev->kobj.sd &&
6162 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
5e55e2f5 6163 printk(KERN_WARNING
4a5add49 6164 "raid5: failed to create sysfs attributes for %s\n",
5e55e2f5 6165 mdname(mddev));
4a5add49 6166 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
7a5febe9 6167
4a5add49 6168 if (mddev->queue) {
9f7c2220 6169 int chunk_size;
620125f2 6170 bool discard_supported = true;
4a5add49
N		 6171 /* read-ahead size must cover two whole stripes, which
6172 * is 2 * (datadisks) * chunksize where 'n' is the
6173 * number of raid devices
6174 */
6175 int data_disks = conf->previous_raid_disks - conf->max_degraded;
6176 int stripe = data_disks *
6177 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
6178 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
6179 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
91adb564 6180
4a5add49 6181 blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec);
f022b2fd 6182
11d8a6e3
N		 6183 mddev->queue->backing_dev_info.congested_data = mddev;
6184 mddev->queue->backing_dev_info.congested_fn = raid5_congested;
7a5febe9 6185
9f7c2220
N		 6186 chunk_size = mddev->chunk_sectors << 9;
6187 blk_queue_io_min(mddev->queue, chunk_size);
6188 blk_queue_io_opt(mddev->queue, chunk_size *
6189 (conf->raid_disks - conf->max_degraded));
c78afc62 6190 mddev->queue->limits.raid_partial_stripes_expensive = 1;
620125f2
SL		 6191 /*
6192 * We can only discard a whole stripe. It doesn't make sense to
6193 * discard data disk but write parity disk
6194 */
6195 stripe = stripe * PAGE_SIZE;
4ac6875e
N		 6196 /* Round up to power of 2, as discard handling
6197 * currently assumes that */
6198 while ((stripe-1) & stripe)
6199 stripe = (stripe | (stripe-1)) + 1;
620125f2
SL		 6200 mddev->queue->limits.discard_alignment = stripe;
6201 mddev->queue->limits.discard_granularity = stripe;
6202 /*
6203 * unaligned part of discard request will be ignored, so can't
8e0e99ba 6204 * guarantee discard_zeroes_data
620125f2
SL		 6205 */
6206 mddev->queue->limits.discard_zeroes_data = 0;
8f6c2e4b 6207
5026d7a9
PA		 6208 blk_queue_max_write_same_sectors(mddev->queue, 0);
6209
05616be5 6210 rdev_for_each(rdev, mddev) {
9f7c2220
N		 6211 disk_stack_limits(mddev->gendisk, rdev->bdev,
6212 rdev->data_offset << 9);
05616be5
N		 6213 disk_stack_limits(mddev->gendisk, rdev->bdev,
6214 rdev->new_data_offset << 9);
620125f2
SL		 6215 /*
6216 * discard_zeroes_data is required, otherwise data
6217 * could be lost. Consider a scenario: discard a stripe
6218 * (the stripe could be inconsistent if
6219 * discard_zeroes_data is 0); write one disk of the
6220 * stripe (the stripe could be inconsistent again
6221 * depending on which disks are used to calculate
6222 * parity); the disk is broken; The stripe data of this
6223 * disk is lost.
6224 */
6225 if (!blk_queue_discard(bdev_get_queue(rdev->bdev)) ||
6226 !bdev_get_queue(rdev->bdev)->
6227 limits.discard_zeroes_data)
6228 discard_supported = false;
8e0e99ba
N		 6229 /* Unfortunately, discard_zeroes_data is not currently
6230 * a guarantee - just a hint. So we only allow DISCARD
6231 * if the sysadmin has confirmed that only safe devices
6232 * are in use by setting a module parameter.
6233 */
6234 if (!devices_handle_discard_safely) {
6235 if (discard_supported) {
6236 pr_info("md/raid456: discard support disabled due to uncertainty.\n");
6237 pr_info("Set raid456.devices_handle_discard_safely=Y to override.\n");
6238 }
6239 discard_supported = false;
6240 }
05616be5 6241 }
620125f2
SL		 6242
6243 if (discard_supported &&
6244 mddev->queue->limits.max_discard_sectors >= stripe &&
6245 mddev->queue->limits.discard_granularity >= stripe)
6246 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
6247 mddev->queue);
6248 else
6249 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
6250 mddev->queue);
9f7c2220 6251 }
23032a0e 6252
1da177e4
LT		 6253 return 0;
6254abort:
01f96c0a 6255 md_unregister_thread(&mddev->thread);
e4f869d9
N		 6256 print_raid5_conf(conf);
6257 free_conf(conf);
1da177e4 6258 mddev->private = NULL;
0c55e022 6259 printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev));
1da177e4
LT		 6260 return -EIO;
6261}
6262
fd01b88c 6263static int stop(struct mddev *mddev)
1da177e4 6264{
d1688a6d 6265 struct r5conf *conf = mddev->private;
1da177e4 6266
01f96c0a 6267 md_unregister_thread(&mddev->thread);
11d8a6e3
N		 6268 if (mddev->queue)
6269 mddev->queue->backing_dev_info.congested_fn = NULL;
95fc17aa 6270 free_conf(conf);
a64c876f
N		 6271 mddev->private = NULL;
6272 mddev->to_remove = &raid5_attrs_group;
1da177e4
LT		 6273 return 0;
6274}
6275
fd01b88c 6276static void status(struct seq_file *seq, struct mddev *mddev)
1da177e4 6277{
d1688a6d 6278 struct r5conf *conf = mddev->private;
1da177e4
LT		 6279 int i;
6280
9d8f0363
AN		 6281 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
6282 mddev->chunk_sectors / 2, mddev->layout);
02c2de8c 6283 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
1da177e4
LT		 6284 for (i = 0; i < conf->raid_disks; i++)
6285 seq_printf (seq, "%s",
6286 conf->disks[i].rdev &&
b2d444d7 6287 test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_");
1da177e4 6288 seq_printf (seq, "]");
1da177e4
LT		 6289}
6290
d1688a6d 6291static void print_raid5_conf (struct r5conf *conf)
1da177e4
LT		 6292{
6293 int i;
6294 struct disk_info *tmp;
6295
0c55e022 6296 printk(KERN_DEBUG "RAID conf printout:\n");
1da177e4
LT		 6297 if (!conf) {
6298 printk("(conf==NULL)\n");
6299 return;
6300 }
0c55e022
N		 6301 printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level,
6302 conf->raid_disks,
6303 conf->raid_disks - conf->mddev->degraded);
1da177e4
LT		 6304
6305 for (i = 0; i < conf->raid_disks; i++) {
6306 char b[BDEVNAME_SIZE];
6307 tmp = conf->disks + i;
6308 if (tmp->rdev)
0c55e022
N		 6309 printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n",
6310 i, !test_bit(Faulty, &tmp->rdev->flags),
6311 bdevname(tmp->rdev->bdev, b));
1da177e4
LT		 6312 }
6313}
6314
fd01b88c 6315static int raid5_spare_active(struct mddev *mddev)
1da177e4
LT		 6316{
6317 int i;
d1688a6d 6318 struct r5conf *conf = mddev->private;
1da177e4 6319 struct disk_info *tmp;
6b965620
N		 6320 int count = 0;
6321 unsigned long flags;
1da177e4
LT		 6322
6323 for (i = 0; i < conf->raid_disks; i++) {
6324 tmp = conf->disks + i;
dd054fce
N		 6325 if (tmp->replacement
6326 && tmp->replacement->recovery_offset == MaxSector
6327 && !test_bit(Faulty, &tmp->replacement->flags)
6328 && !test_and_set_bit(In_sync, &tmp->replacement->flags)) {
6329 /* Replacement has just become active. */
6330 if (!tmp->rdev
6331 || !test_and_clear_bit(In_sync, &tmp->rdev->flags))
6332 count++;
6333 if (tmp->rdev) {
6334 /* Replaced device not technically faulty,
6335 * but we need to be sure it gets removed
6336 * and never re-added.
6337 */
6338 set_bit(Faulty, &tmp->rdev->flags);
6339 sysfs_notify_dirent_safe(
6340 tmp->rdev->sysfs_state);
6341 }
6342 sysfs_notify_dirent_safe(tmp->replacement->sysfs_state);
6343 } else if (tmp->rdev
70fffd0b 6344 && tmp->rdev->recovery_offset == MaxSector
b2d444d7 6345 && !test_bit(Faulty, &tmp->rdev->flags)
c04be0aa 6346 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
6b965620 6347 count++;
43c73ca4 6348 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
1da177e4
LT		 6349 }
6350 }
6b965620 6351 spin_lock_irqsave(&conf->device_lock, flags);
908f4fbd 6352 mddev->degraded = calc_degraded(conf);
6b965620 6353 spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4 6354 print_raid5_conf(conf);
6b965620 6355 return count;
1da177e4
LT		 6356}
6357
b8321b68 6358static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 6359{
d1688a6d 6360 struct r5conf *conf = mddev->private;
1da177e4 6361 int err = 0;
b8321b68 6362 int number = rdev->raid_disk;
657e3e4d 6363 struct md_rdev **rdevp;
1da177e4
LT		 6364 struct disk_info *p = conf->disks + number;
6365
6366 print_raid5_conf(conf);
657e3e4d
N		 6367 if (rdev == p->rdev)
6368 rdevp = &p->rdev;
6369 else if (rdev == p->replacement)
6370 rdevp = &p->replacement;
6371 else
6372 return 0;
6373
6374 if (number >= conf->raid_disks &&
6375 conf->reshape_progress == MaxSector)
6376 clear_bit(In_sync, &rdev->flags);
6377
6378 if (test_bit(In_sync, &rdev->flags) ||
6379 atomic_read(&rdev->nr_pending)) {
6380 err = -EBUSY;
6381 goto abort;
6382 }
6383 /* Only remove non-faulty devices if recovery
6384 * isn't possible.
6385 */
6386 if (!test_bit(Faulty, &rdev->flags) &&
6387 mddev->recovery_disabled != conf->recovery_disabled &&
6388 !has_failed(conf) &&
dd054fce 6389 (!p->replacement || p->replacement == rdev) &&
657e3e4d
N		 6390 number < conf->raid_disks) {
6391 err = -EBUSY;
6392 goto abort;
6393 }
6394 *rdevp = NULL;
6395 synchronize_rcu();
6396 if (atomic_read(&rdev->nr_pending)) {
6397 /* lost the race, try later */
6398 err = -EBUSY;
6399 *rdevp = rdev;
dd054fce
N		 6400 } else if (p->replacement) {
6401 /* We must have just cleared 'rdev' */
6402 p->rdev = p->replacement;
6403 clear_bit(Replacement, &p->replacement->flags);
6404 smp_mb(); /* Make sure other CPUs may see both as identical
6405 * but will never see neither - if they are careful
6406 */
6407 p->replacement = NULL;
6408 clear_bit(WantReplacement, &rdev->flags);
6409 } else
6410 /* We might have just removed the Replacement as faulty-
6411 * clear the bit just in case
6412 */
6413 clear_bit(WantReplacement, &rdev->flags);
1da177e4
LT		 6414abort:
6415
6416 print_raid5_conf(conf);
6417 return err;
6418}
6419
fd01b88c 6420static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 6421{
d1688a6d 6422 struct r5conf *conf = mddev->private;
199050ea 6423 int err = -EEXIST;
1da177e4
LT		 6424 int disk;
6425 struct disk_info *p;
6c2fce2e
NB		 6426 int first = 0;
6427 int last = conf->raid_disks - 1;
1da177e4 6428
7f0da59b
N		 6429 if (mddev->recovery_disabled == conf->recovery_disabled)
6430 return -EBUSY;
6431
dc10c643 6432 if (rdev->saved_raid_disk < 0 && has_failed(conf))
1da177e4 6433 /* no point adding a device */
199050ea 6434 return -EINVAL;
1da177e4 6435
6c2fce2e
NB		 6436 if (rdev->raid_disk >= 0)
6437 first = last = rdev->raid_disk;
1da177e4
LT		 6438
6439 /*
16a53ecc
N		 6440 * find the disk ... but prefer rdev->saved_raid_disk
6441 * if possible.
1da177e4 6442 */
16a53ecc 6443 if (rdev->saved_raid_disk >= 0 &&
6c2fce2e 6444 rdev->saved_raid_disk >= first &&
16a53ecc 6445 conf->disks[rdev->saved_raid_disk].rdev == NULL)
5cfb22a1
N		 6446 first = rdev->saved_raid_disk;
6447
6448 for (disk = first; disk <= last; disk++) {
7bfec5f3
N		 6449 p = conf->disks + disk;
6450 if (p->rdev == NULL) {
b2d444d7 6451 clear_bit(In_sync, &rdev->flags);
1da177e4 6452 rdev->raid_disk = disk;
199050ea 6453 err = 0;
72626685
N		 6454 if (rdev->saved_raid_disk != disk)
6455 conf->fullsync = 1;
d6065f7b 6456 rcu_assign_pointer(p->rdev, rdev);
5cfb22a1 6457 goto out;
1da177e4 6458 }
5cfb22a1
N		 6459 }
6460 for (disk = first; disk <= last; disk++) {
6461 p = conf->disks + disk;
7bfec5f3
N		 6462 if (test_bit(WantReplacement, &p->rdev->flags) &&
6463 p->replacement == NULL) {
6464 clear_bit(In_sync, &rdev->flags);
6465 set_bit(Replacement, &rdev->flags);
6466 rdev->raid_disk = disk;
6467 err = 0;
6468 conf->fullsync = 1;
6469 rcu_assign_pointer(p->replacement, rdev);
6470 break;
6471 }
6472 }
5cfb22a1 6473out:
1da177e4 6474 print_raid5_conf(conf);
199050ea 6475 return err;
1da177e4
LT		 6476}
6477
fd01b88c 6478static int raid5_resize(struct mddev *mddev, sector_t sectors)
1da177e4
LT		 6479{
6480 /* no resync is happening, and there is enough space
6481 * on all devices, so we can resize.
6482 * We need to make sure resync covers any new space.
6483 * If the array is shrinking we should possibly wait until
6484 * any io in the removed space completes, but it hardly seems
6485 * worth it.
6486 */
a4a6125a 6487 sector_t newsize;
9d8f0363 6488 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
a4a6125a
N		 6489 newsize = raid5_size(mddev, sectors, mddev->raid_disks);
6490 if (mddev->external_size &&
6491 mddev->array_sectors > newsize)
b522adcd 6492 return -EINVAL;
a4a6125a
N		 6493 if (mddev->bitmap) {
6494 int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0);
6495 if (ret)
6496 return ret;
6497 }
6498 md_set_array_sectors(mddev, newsize);
f233ea5c 6499 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 6500 revalidate_disk(mddev->gendisk);
b098636c
N		 6501 if (sectors > mddev->dev_sectors &&
6502 mddev->recovery_cp > mddev->dev_sectors) {
58c0fed4 6503 mddev->recovery_cp = mddev->dev_sectors;
1da177e4
LT		 6504 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6505 }
58c0fed4 6506 mddev->dev_sectors = sectors;
4b5c7ae8 6507 mddev->resync_max_sectors = sectors;
1da177e4
LT		 6508 return 0;
6509}
6510
fd01b88c 6511static int check_stripe_cache(struct mddev *mddev)
01ee22b4
N		 6512{
6513 /* Can only proceed if there are plenty of stripe_heads.
6514 * We need a minimum of one full stripe,, and for sensible progress
6515 * it is best to have about 4 times that.
6516 * If we require 4 times, then the default 256 4K stripe_heads will
6517 * allow for chunk sizes up to 256K, which is probably OK.
6518 * If the chunk size is greater, user-space should request more
6519 * stripe_heads first.
6520 */
d1688a6d 6521 struct r5conf *conf = mddev->private;
01ee22b4
N		 6522 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
6523 > conf->max_nr_stripes ||
6524 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
6525 > conf->max_nr_stripes) {
0c55e022
N		 6526 printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n",
6527 mdname(mddev),
01ee22b4
N		 6528 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
6529 / STRIPE_SIZE)*4);
6530 return 0;
6531 }
6532 return 1;
6533}
6534
fd01b88c 6535static int check_reshape(struct mddev *mddev)
29269553 6536{
d1688a6d 6537 struct r5conf *conf = mddev->private;
29269553 6538
88ce4930
N		 6539 if (mddev->delta_disks == 0 &&
6540 mddev->new_layout == mddev->layout &&
664e7c41 6541 mddev->new_chunk_sectors == mddev->chunk_sectors)
50ac168a 6542 return 0; /* nothing to do */
674806d6 6543 if (has_failed(conf))
ec32a2bd 6544 return -EINVAL;
fdcfbbb6 6545 if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) {
ec32a2bd
N		 6546 /* We might be able to shrink, but the devices must
6547 * be made bigger first.
6548 * For raid6, 4 is the minimum size.
6549 * Otherwise 2 is the minimum
6550 */
6551 int min = 2;
6552 if (mddev->level == 6)
6553 min = 4;
6554 if (mddev->raid_disks + mddev->delta_disks < min)
6555 return -EINVAL;
6556 }
29269553 6557
01ee22b4 6558 if (!check_stripe_cache(mddev))
29269553 6559 return -ENOSPC;
29269553 6560
e56108d6
N		 6561 return resize_stripes(conf, (conf->previous_raid_disks
6562 + mddev->delta_disks));
63c70c4f
N		 6563}
6564
fd01b88c 6565static int raid5_start_reshape(struct mddev *mddev)
63c70c4f 6566{
d1688a6d 6567 struct r5conf *conf = mddev->private;
3cb03002 6568 struct md_rdev *rdev;
63c70c4f 6569 int spares = 0;
c04be0aa 6570 unsigned long flags;
63c70c4f 6571
f416885e 6572 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
63c70c4f
N		 6573 return -EBUSY;
6574
01ee22b4
N		 6575 if (!check_stripe_cache(mddev))
6576 return -ENOSPC;
6577
30b67645
N		 6578 if (has_failed(conf))
6579 return -EINVAL;
6580
c6563a8c 6581 rdev_for_each(rdev, mddev) {
469518a3
N		 6582 if (!test_bit(In_sync, &rdev->flags)
6583 && !test_bit(Faulty, &rdev->flags))
29269553 6584 spares++;
c6563a8c 6585 }
63c70c4f 6586
f416885e 6587 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
29269553
N		 6588 /* Not enough devices even to make a degraded array
6589 * of that size
6590 */
6591 return -EINVAL;
6592
ec32a2bd
N		 6593 /* Refuse to reduce size of the array. Any reductions in
6594 * array size must be through explicit setting of array_size
6595 * attribute.
6596 */
6597 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
6598 < mddev->array_sectors) {
0c55e022 6599 printk(KERN_ERR "md/raid:%s: array size must be reduced "
ec32a2bd
N		 6600 "before number of disks\n", mdname(mddev));
6601 return -EINVAL;
6602 }
6603
f6705578 6604 atomic_set(&conf->reshape_stripes, 0);
29269553 6605 spin_lock_irq(&conf->device_lock);
c46501b2 6606 write_seqcount_begin(&conf->gen_lock);
29269553 6607 conf->previous_raid_disks = conf->raid_disks;
63c70c4f 6608 conf->raid_disks += mddev->delta_disks;
09c9e5fa
AN		 6609 conf->prev_chunk_sectors = conf->chunk_sectors;
6610 conf->chunk_sectors = mddev->new_chunk_sectors;
88ce4930
N		 6611 conf->prev_algo = conf->algorithm;
6612 conf->algorithm = mddev->new_layout;
05616be5
N		 6613 conf->generation++;
6614 /* Code that selects data_offset needs to see the generation update
6615 * if reshape_progress has been set - so a memory barrier needed.
6616 */
6617 smp_mb();
2c810cdd 6618 if (mddev->reshape_backwards)
fef9c61f
N		 6619 conf->reshape_progress = raid5_size(mddev, 0, 0);
6620 else
6621 conf->reshape_progress = 0;
6622 conf->reshape_safe = conf->reshape_progress;
c46501b2 6623 write_seqcount_end(&conf->gen_lock);
29269553
N		 6624 spin_unlock_irq(&conf->device_lock);
6625
4d77e3ba
N		 6626 /* Now make sure any requests that proceeded on the assumption
6627 * the reshape wasn't running - like Discard or Read - have
6628 * completed.
6629 */
6630 mddev_suspend(mddev);
6631 mddev_resume(mddev);
6632
29269553
N		 6633 /* Add some new drives, as many as will fit.
6634 * We know there are enough to make the newly sized array work.
3424bf6a
N		 6635 * Don't add devices if we are reducing the number of
6636 * devices in the array. This is because it is not possible
6637 * to correctly record the "partially reconstructed" state of
6638 * such devices during the reshape and confusion could result.
29269553 6639 */
87a8dec9 6640 if (mddev->delta_disks >= 0) {
dafb20fa 6641 rdev_for_each(rdev, mddev)
87a8dec9
N		 6642 if (rdev->raid_disk < 0 &&
6643 !test_bit(Faulty, &rdev->flags)) {
6644 if (raid5_add_disk(mddev, rdev) == 0) {
87a8dec9 6645 if (rdev->raid_disk
9d4c7d87 6646 >= conf->previous_raid_disks)
87a8dec9 6647 set_bit(In_sync, &rdev->flags);
9d4c7d87 6648 else
87a8dec9 6649 rdev->recovery_offset = 0;
36fad858
NK		 6650
6651 if (sysfs_link_rdev(mddev, rdev))
87a8dec9 6652 /* Failure here is OK */;
50da0840 6653 }
87a8dec9
N		 6654 } else if (rdev->raid_disk >= conf->previous_raid_disks
6655 && !test_bit(Faulty, &rdev->flags)) {
6656 /* This is a spare that was manually added */
6657 set_bit(In_sync, &rdev->flags);
87a8dec9 6658 }
29269553 6659
87a8dec9
N		 6660 /* When a reshape changes the number of devices,
6661 * ->degraded is measured against the larger of the
6662 * pre and post number of devices.
6663 */
ec32a2bd 6664 spin_lock_irqsave(&conf->device_lock, flags);
908f4fbd 6665 mddev->degraded = calc_degraded(conf);
ec32a2bd
N		 6666 spin_unlock_irqrestore(&conf->device_lock, flags);
6667 }
63c70c4f 6668 mddev->raid_disks = conf->raid_disks;
e516402c 6669 mddev->reshape_position = conf->reshape_progress;
850b2b42 6670 set_bit(MD_CHANGE_DEVS, &mddev->flags);
f6705578 6671
29269553
N		 6672 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6673 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6674 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6675 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6676 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 6677 "reshape");
29269553
N		 6678 if (!mddev->sync_thread) {
6679 mddev->recovery = 0;
6680 spin_lock_irq(&conf->device_lock);
ba8805b9 6681 write_seqcount_begin(&conf->gen_lock);
29269553 6682 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
ba8805b9
N		 6683 mddev->new_chunk_sectors =
6684 conf->chunk_sectors = conf->prev_chunk_sectors;
6685 mddev->new_layout = conf->algorithm = conf->prev_algo;
05616be5
N		 6686 rdev_for_each(rdev, mddev)
6687 rdev->new_data_offset = rdev->data_offset;
6688 smp_wmb();
ba8805b9 6689 conf->generation --;
fef9c61f 6690 conf->reshape_progress = MaxSector;
1e3fa9bd 6691 mddev->reshape_position = MaxSector;
ba8805b9 6692 write_seqcount_end(&conf->gen_lock);
29269553
N		 6693 spin_unlock_irq(&conf->device_lock);
6694 return -EAGAIN;
6695 }
c8f517c4 6696 conf->reshape_checkpoint = jiffies;
29269553
N		 6697 md_wakeup_thread(mddev->sync_thread);
6698 md_new_event(mddev);
6699 return 0;
6700}
29269553 6701
ec32a2bd
N		 6702/* This is called from the reshape thread and should make any
6703 * changes needed in 'conf'
6704 */
d1688a6d 6705static void end_reshape(struct r5conf *conf)
29269553 6706{
29269553 6707
f6705578 6708 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
05616be5 6709 struct md_rdev *rdev;
f6705578 6710
f6705578 6711 spin_lock_irq(&conf->device_lock);
cea9c228 6712 conf->previous_raid_disks = conf->raid_disks;
05616be5
N		 6713 rdev_for_each(rdev, conf->mddev)
6714 rdev->data_offset = rdev->new_data_offset;
6715 smp_wmb();
fef9c61f 6716 conf->reshape_progress = MaxSector;
f6705578 6717 spin_unlock_irq(&conf->device_lock);
b0f9ec04 6718 wake_up(&conf->wait_for_overlap);
16a53ecc
N		 6719
6720 /* read-ahead size must cover two whole stripes, which is
6721 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
6722 */
4a5add49 6723 if (conf->mddev->queue) {
cea9c228 6724 int data_disks = conf->raid_disks - conf->max_degraded;
09c9e5fa 6725 int stripe = data_disks * ((conf->chunk_sectors << 9)
cea9c228 6726 / PAGE_SIZE);
16a53ecc
N		 6727 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
6728 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
6729 }
29269553 6730 }
29269553
N		 6731}
6732
ec32a2bd
N		 6733/* This is called from the raid5d thread with mddev_lock held.
6734 * It makes config changes to the device.
6735 */
fd01b88c 6736static void raid5_finish_reshape(struct mddev *mddev)
cea9c228 6737{
d1688a6d 6738 struct r5conf *conf = mddev->private;
cea9c228
N		 6739
6740 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6741
ec32a2bd
N		 6742 if (mddev->delta_disks > 0) {
6743 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
6744 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 6745 revalidate_disk(mddev->gendisk);
ec32a2bd
N		 6746 } else {
6747 int d;
908f4fbd
N		 6748 spin_lock_irq(&conf->device_lock);
6749 mddev->degraded = calc_degraded(conf);
6750 spin_unlock_irq(&conf->device_lock);
ec32a2bd
N		 6751 for (d = conf->raid_disks ;
6752 d < conf->raid_disks - mddev->delta_disks;
1a67dde0 6753 d++) {
3cb03002 6754 struct md_rdev *rdev = conf->disks[d].rdev;
da7613b8
N		 6755 if (rdev)
6756 clear_bit(In_sync, &rdev->flags);
6757 rdev = conf->disks[d].replacement;
6758 if (rdev)
6759 clear_bit(In_sync, &rdev->flags);
1a67dde0 6760 }
cea9c228 6761 }
88ce4930 6762 mddev->layout = conf->algorithm;
09c9e5fa 6763 mddev->chunk_sectors = conf->chunk_sectors;
ec32a2bd
N		 6764 mddev->reshape_position = MaxSector;
6765 mddev->delta_disks = 0;
2c810cdd 6766 mddev->reshape_backwards = 0;
cea9c228
N		 6767 }
6768}
6769
fd01b88c 6770static void raid5_quiesce(struct mddev *mddev, int state)
72626685 6771{
d1688a6d 6772 struct r5conf *conf = mddev->private;
72626685
N		 6773
6774 switch(state) {
e464eafd
N		 6775 case 2: /* resume for a suspend */
6776 wake_up(&conf->wait_for_overlap);
6777 break;
6778
72626685 6779 case 1: /* stop all writes */
566c09c5 6780 lock_all_device_hash_locks_irq(conf);
64bd660b
N		 6781 /* '2' tells resync/reshape to pause so that all
6782 * active stripes can drain
6783 */
6784 conf->quiesce = 2;
566c09c5 6785 wait_event_cmd(conf->wait_for_stripe,
46031f9a
RBJ		 6786 atomic_read(&conf->active_stripes) == 0 &&
6787 atomic_read(&conf->active_aligned_reads) == 0,
566c09c5
SL		 6788 unlock_all_device_hash_locks_irq(conf),
6789 lock_all_device_hash_locks_irq(conf));
64bd660b 6790 conf->quiesce = 1;
566c09c5 6791 unlock_all_device_hash_locks_irq(conf);
64bd660b
N		 6792 /* allow reshape to continue */
6793 wake_up(&conf->wait_for_overlap);
72626685
N		 6794 break;
6795
6796 case 0: /* re-enable writes */
566c09c5 6797 lock_all_device_hash_locks_irq(conf);
72626685
N		 6798 conf->quiesce = 0;
6799 wake_up(&conf->wait_for_stripe);
e464eafd 6800 wake_up(&conf->wait_for_overlap);
566c09c5 6801 unlock_all_device_hash_locks_irq(conf);
72626685
N		 6802 break;
6803 }
72626685 6804}
b15c2e57 6805
fd01b88c 6806static void *raid45_takeover_raid0(struct mddev *mddev, int level)
54071b38 6807{
e373ab10 6808 struct r0conf *raid0_conf = mddev->private;
d76c8420 6809 sector_t sectors;
54071b38 6810
f1b29bca 6811 /* for raid0 takeover only one zone is supported */
e373ab10 6812 if (raid0_conf->nr_strip_zones > 1) {
0c55e022
N		 6813 printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n",
6814 mdname(mddev));
f1b29bca
DW		 6815 return ERR_PTR(-EINVAL);
6816 }
6817
e373ab10
N		 6818 sectors = raid0_conf->strip_zone[0].zone_end;
6819 sector_div(sectors, raid0_conf->strip_zone[0].nb_dev);
3b71bd93 6820 mddev->dev_sectors = sectors;
f1b29bca 6821 mddev->new_level = level;
54071b38
TM		 6822 mddev->new_layout = ALGORITHM_PARITY_N;
6823 mddev->new_chunk_sectors = mddev->chunk_sectors;
6824 mddev->raid_disks += 1;
6825 mddev->delta_disks = 1;
6826 /* make sure it will be not marked as dirty */
6827 mddev->recovery_cp = MaxSector;
6828
6829 return setup_conf(mddev);
6830}
6831
fd01b88c 6832static void *raid5_takeover_raid1(struct mddev *mddev)
d562b0c4
N		 6833{
6834 int chunksect;
6835
6836 if (mddev->raid_disks != 2 ||
6837 mddev->degraded > 1)
6838 return ERR_PTR(-EINVAL);
6839
6840 /* Should check if there are write-behind devices? */
6841
6842 chunksect = 64*2; /* 64K by default */
6843
6844 /* The array must be an exact multiple of chunksize */
6845 while (chunksect && (mddev->array_sectors & (chunksect-1)))
6846 chunksect >>= 1;
6847
6848 if ((chunksect<<9) < STRIPE_SIZE)
6849 /* array size does not allow a suitable chunk size */
6850 return ERR_PTR(-EINVAL);
6851
6852 mddev->new_level = 5;
6853 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
664e7c41 6854 mddev->new_chunk_sectors = chunksect;
d562b0c4
N		 6855
6856 return setup_conf(mddev);
6857}
6858
fd01b88c 6859static void *raid5_takeover_raid6(struct mddev *mddev)
fc9739c6
N		 6860{
6861 int new_layout;
6862
6863 switch (mddev->layout) {
6864 case ALGORITHM_LEFT_ASYMMETRIC_6:
6865 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
6866 break;
6867 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6868 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
6869 break;
6870 case ALGORITHM_LEFT_SYMMETRIC_6:
6871 new_layout = ALGORITHM_LEFT_SYMMETRIC;
6872 break;
6873 case ALGORITHM_RIGHT_SYMMETRIC_6:
6874 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
6875 break;
6876 case ALGORITHM_PARITY_0_6:
6877 new_layout = ALGORITHM_PARITY_0;
6878 break;
6879 case ALGORITHM_PARITY_N:
6880 new_layout = ALGORITHM_PARITY_N;
6881 break;
6882 default:
6883 return ERR_PTR(-EINVAL);
6884 }
6885 mddev->new_level = 5;
6886 mddev->new_layout = new_layout;
6887 mddev->delta_disks = -1;
6888 mddev->raid_disks -= 1;
6889 return setup_conf(mddev);
6890}
6891
fd01b88c 6892static int raid5_check_reshape(struct mddev *mddev)
b3546035 6893{
88ce4930
N		 6894 /* For a 2-drive array, the layout and chunk size can be changed
6895 * immediately as not restriping is needed.
6896 * For larger arrays we record the new value - after validation
6897 * to be used by a reshape pass.
b3546035 6898 */
d1688a6d 6899 struct r5conf *conf = mddev->private;
597a711b 6900 int new_chunk = mddev->new_chunk_sectors;
b3546035 6901
597a711b 6902 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
b3546035
N		 6903 return -EINVAL;
6904 if (new_chunk > 0) {
0ba459d2 6905 if (!is_power_of_2(new_chunk))
b3546035 6906 return -EINVAL;
597a711b 6907 if (new_chunk < (PAGE_SIZE>>9))
b3546035 6908 return -EINVAL;
597a711b 6909 if (mddev->array_sectors & (new_chunk-1))
b3546035
N		 6910 /* not factor of array size */
6911 return -EINVAL;
6912 }
6913
6914 /* They look valid */
6915
88ce4930 6916 if (mddev->raid_disks == 2) {
597a711b
N		 6917 /* can make the change immediately */
6918 if (mddev->new_layout >= 0) {
6919 conf->algorithm = mddev->new_layout;
6920 mddev->layout = mddev->new_layout;
88ce4930
N		 6921 }
6922 if (new_chunk > 0) {
597a711b
N		 6923 conf->chunk_sectors = new_chunk ;
6924 mddev->chunk_sectors = new_chunk;
88ce4930
N		 6925 }
6926 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6927 md_wakeup_thread(mddev->thread);
b3546035 6928 }
50ac168a 6929 return check_reshape(mddev);
88ce4930
N		 6930}
6931
fd01b88c 6932static int raid6_check_reshape(struct mddev *mddev)
88ce4930 6933{
597a711b 6934 int new_chunk = mddev->new_chunk_sectors;
50ac168a 6935
597a711b 6936 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
88ce4930 6937 return -EINVAL;
b3546035 6938 if (new_chunk > 0) {
0ba459d2 6939 if (!is_power_of_2(new_chunk))
88ce4930 6940 return -EINVAL;
597a711b 6941 if (new_chunk < (PAGE_SIZE >> 9))
88ce4930 6942 return -EINVAL;
597a711b 6943 if (mddev->array_sectors & (new_chunk-1))
88ce4930
N		 6944 /* not factor of array size */
6945 return -EINVAL;
b3546035 6946 }
88ce4930
N		 6947
6948 /* They look valid */
50ac168a 6949 return check_reshape(mddev);
b3546035
N		 6950}
6951
fd01b88c 6952static void *raid5_takeover(struct mddev *mddev)
d562b0c4
N		 6953{
6954 /* raid5 can take over:
f1b29bca 6955 * raid0 - if there is only one strip zone - make it a raid4 layout
d562b0c4
N		 6956 * raid1 - if there are two drives. We need to know the chunk size
6957 * raid4 - trivial - just use a raid4 layout.
6958 * raid6 - Providing it is a *_6 layout
d562b0c4 6959 */
f1b29bca
DW		 6960 if (mddev->level == 0)
6961 return raid45_takeover_raid0(mddev, 5);
d562b0c4
N		 6962 if (mddev->level == 1)
6963 return raid5_takeover_raid1(mddev);
e9d4758f
N		 6964 if (mddev->level == 4) {
6965 mddev->new_layout = ALGORITHM_PARITY_N;
6966 mddev->new_level = 5;
6967 return setup_conf(mddev);
6968 }
fc9739c6
N		 6969 if (mddev->level == 6)
6970 return raid5_takeover_raid6(mddev);
d562b0c4
N		 6971
6972 return ERR_PTR(-EINVAL);
6973}
6974
fd01b88c 6975static void *raid4_takeover(struct mddev *mddev)
a78d38a1 6976{
f1b29bca
DW		 6977 /* raid4 can take over:
6978 * raid0 - if there is only one strip zone
6979 * raid5 - if layout is right
a78d38a1 6980 */
f1b29bca
DW		 6981 if (mddev->level == 0)
6982 return raid45_takeover_raid0(mddev, 4);
a78d38a1
N		 6983 if (mddev->level == 5 &&
6984 mddev->layout == ALGORITHM_PARITY_N) {
6985 mddev->new_layout = 0;
6986 mddev->new_level = 4;
6987 return setup_conf(mddev);
6988 }
6989 return ERR_PTR(-EINVAL);
6990}
d562b0c4 6991
84fc4b56 6992static struct md_personality raid5_personality;
245f46c2 6993
fd01b88c 6994static void *raid6_takeover(struct mddev *mddev)
245f46c2
N		 6995{
6996 /* Currently can only take over a raid5. We map the
6997 * personality to an equivalent raid6 personality
6998 * with the Q block at the end.
6999 */
7000 int new_layout;
7001
7002 if (mddev->pers != &raid5_personality)
7003 return ERR_PTR(-EINVAL);
7004 if (mddev->degraded > 1)
7005 return ERR_PTR(-EINVAL);
7006 if (mddev->raid_disks > 253)
7007 return ERR_PTR(-EINVAL);
7008 if (mddev->raid_disks < 3)
7009 return ERR_PTR(-EINVAL);
7010
7011 switch (mddev->layout) {
7012 case ALGORITHM_LEFT_ASYMMETRIC:
7013 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
7014 break;
7015 case ALGORITHM_RIGHT_ASYMMETRIC:
7016 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
7017 break;
7018 case ALGORITHM_LEFT_SYMMETRIC:
7019 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
7020 break;
7021 case ALGORITHM_RIGHT_SYMMETRIC:
7022 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
7023 break;
7024 case ALGORITHM_PARITY_0:
7025 new_layout = ALGORITHM_PARITY_0_6;
7026 break;
7027 case ALGORITHM_PARITY_N:
7028 new_layout = ALGORITHM_PARITY_N;
7029 break;
7030 default:
7031 return ERR_PTR(-EINVAL);
7032 }
7033 mddev->new_level = 6;
7034 mddev->new_layout = new_layout;
7035 mddev->delta_disks = 1;
7036 mddev->raid_disks += 1;
7037 return setup_conf(mddev);
7038}
7039
84fc4b56 7040static struct md_personality raid6_personality =
16a53ecc
N		 7041{
7042 .name = "raid6",
7043 .level = 6,
7044 .owner = THIS_MODULE,
7045 .make_request = make_request,
7046 .run = run,
7047 .stop = stop,
7048 .status = status,
7049 .error_handler = error,
7050 .hot_add_disk = raid5_add_disk,
7051 .hot_remove_disk= raid5_remove_disk,
7052 .spare_active = raid5_spare_active,
7053 .sync_request = sync_request,
7054 .resize = raid5_resize,
80c3a6ce 7055 .size = raid5_size,
50ac168a 7056 .check_reshape = raid6_check_reshape,
f416885e 7057 .start_reshape = raid5_start_reshape,
cea9c228 7058 .finish_reshape = raid5_finish_reshape,
16a53ecc 7059 .quiesce = raid5_quiesce,
245f46c2 7060 .takeover = raid6_takeover,
16a53ecc 7061};
84fc4b56 7062static struct md_personality raid5_personality =
1da177e4
LT		 7063{
7064 .name = "raid5",
2604b703 7065 .level = 5,
1da177e4
LT		 7066 .owner = THIS_MODULE,
7067 .make_request = make_request,
7068 .run = run,
7069 .stop = stop,
7070 .status = status,
7071 .error_handler = error,
7072 .hot_add_disk = raid5_add_disk,
7073 .hot_remove_disk= raid5_remove_disk,
7074 .spare_active = raid5_spare_active,
7075 .sync_request = sync_request,
7076 .resize = raid5_resize,
80c3a6ce 7077 .size = raid5_size,
63c70c4f
N		 7078 .check_reshape = raid5_check_reshape,
7079 .start_reshape = raid5_start_reshape,
cea9c228 7080 .finish_reshape = raid5_finish_reshape,
72626685 7081 .quiesce = raid5_quiesce,
d562b0c4 7082 .takeover = raid5_takeover,
1da177e4
LT		 7083};
7084
84fc4b56 7085static struct md_personality raid4_personality =
1da177e4 7086{
2604b703
N		 7087 .name = "raid4",
7088 .level = 4,
7089 .owner = THIS_MODULE,
7090 .make_request = make_request,
7091 .run = run,
7092 .stop = stop,
7093 .status = status,
7094 .error_handler = error,
7095 .hot_add_disk = raid5_add_disk,
7096 .hot_remove_disk= raid5_remove_disk,
7097 .spare_active = raid5_spare_active,
7098 .sync_request = sync_request,
7099 .resize = raid5_resize,
80c3a6ce 7100 .size = raid5_size,
3d37890b
N		 7101 .check_reshape = raid5_check_reshape,
7102 .start_reshape = raid5_start_reshape,
cea9c228 7103 .finish_reshape = raid5_finish_reshape,
2604b703 7104 .quiesce = raid5_quiesce,
a78d38a1 7105 .takeover = raid4_takeover,
2604b703
N		 7106};
7107
7108static int __init raid5_init(void)
7109{
851c30c9
SL		 7110 raid5_wq = alloc_workqueue("raid5wq",
7111 WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0);
7112 if (!raid5_wq)
7113 return -ENOMEM;
16a53ecc 7114 register_md_personality(&raid6_personality);
2604b703
N		 7115 register_md_personality(&raid5_personality);
7116 register_md_personality(&raid4_personality);
7117 return 0;
1da177e4
LT		 7118}
7119
2604b703 7120static void raid5_exit(void)
1da177e4 7121{
16a53ecc 7122 unregister_md_personality(&raid6_personality);
2604b703
N		 7123 unregister_md_personality(&raid5_personality);
7124 unregister_md_personality(&raid4_personality);
851c30c9 7125 destroy_workqueue(raid5_wq);
1da177e4
LT		 7126}
7127
7128module_init(raid5_init);
7129module_exit(raid5_exit);
7130MODULE_LICENSE("GPL");
0efb9e61 7131MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
1da177e4 7132MODULE_ALIAS("md-personality-4"); /* RAID5 */
d9d166c2
N		 7133MODULE_ALIAS("md-raid5");
7134MODULE_ALIAS("md-raid4");
2604b703
N		 7135MODULE_ALIAS("md-level-5");
7136MODULE_ALIAS("md-level-4");
16a53ecc
N		 7137MODULE_ALIAS("md-personality-8"); /* RAID6 */
7138MODULE_ALIAS("md-raid6");
7139MODULE_ALIAS("md-level-6");
7140
7141/* This used to be two separate modules, they were: */
7142MODULE_ALIAS("raid5");
7143MODULE_ALIAS("raid6");
Linux 6.x block layer and io_uring tree(s)