projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'v3.10.55' into update
[GitHub/mt8127/android_kernel_alcatel_ttab.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>
a9add5d9
N		 56#include <trace/events/block.h>
57
43b2e5d8 58#include "md.h"
bff61975 59#include "raid5.h"
54071b38 60#include "raid0.h"
ef740c37 61#include "bitmap.h"
72626685 62
1da177e4
LT		 63/*
64 * Stripe cache
65 */
66
67#define NR_STRIPES 256
68#define STRIPE_SIZE PAGE_SIZE
69#define STRIPE_SHIFT (PAGE_SHIFT - 9)
70#define STRIPE_SECTORS (STRIPE_SIZE>>9)
71#define IO_THRESHOLD 1
8b3e6cdc 72#define BYPASS_THRESHOLD 1
fccddba0 73#define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
1da177e4
LT		 74#define HASH_MASK (NR_HASH - 1)
75
d1688a6d 76static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect)
db298e19
N		 77{
78 int hash = (sect >> STRIPE_SHIFT) & HASH_MASK;
79 return &conf->stripe_hashtbl[hash];
80}
1da177e4
LT		 81
82/* bio's attached to a stripe+device for I/O are linked together in bi_sector
83 * order without overlap. There may be several bio's per stripe+device, and
84 * a bio could span several devices.
85 * When walking this list for a particular stripe+device, we must never proceed
86 * beyond a bio that extends past this device, as the next bio might no longer
87 * be valid.
db298e19 88 * This function is used to determine the 'next' bio in the list, given the sector
1da177e4
LT		 89 * of the current stripe+device
90 */
db298e19
N		 91static inline struct bio *r5_next_bio(struct bio *bio, sector_t sector)
92{
aa8b57aa 93 int sectors = bio_sectors(bio);
db298e19
N		 94 if (bio->bi_sector + sectors < sector + STRIPE_SECTORS)
95 return bio->bi_next;
96 else
97 return NULL;
98}
1da177e4 99
960e739d 100/*
5b99c2ff
JA		 101 * We maintain a biased count of active stripes in the bottom 16 bits of
102 * bi_phys_segments, and a count of processed stripes in the upper 16 bits
960e739d 103 */
e7836bd6 104static inline int raid5_bi_processed_stripes(struct bio *bio)
960e739d 105{
e7836bd6
SL		 106 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
107 return (atomic_read(segments) >> 16) & 0xffff;
960e739d
JA		 108}
109
e7836bd6 110static inline int raid5_dec_bi_active_stripes(struct bio *bio)
960e739d 111{
e7836bd6
SL		 112 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
113 return atomic_sub_return(1, segments) & 0xffff;
960e739d
JA		 114}
115
e7836bd6 116static inline void raid5_inc_bi_active_stripes(struct bio *bio)
960e739d 117{
e7836bd6
SL		 118 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
119 atomic_inc(segments);
960e739d
JA		 120}
121
e7836bd6
SL		 122static inline void raid5_set_bi_processed_stripes(struct bio *bio,
123 unsigned int cnt)
960e739d 124{
e7836bd6
SL		 125 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
126 int old, new;
960e739d 127
e7836bd6
SL		 128 do {
129 old = atomic_read(segments);
130 new = (old & 0xffff) | (cnt << 16);
131 } while (atomic_cmpxchg(segments, old, new) != old);
960e739d
JA		 132}
133
e7836bd6 134static inline void raid5_set_bi_stripes(struct bio *bio, unsigned int cnt)
960e739d 135{
e7836bd6
SL		 136 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
137 atomic_set(segments, cnt);
960e739d
JA		 138}
139
d0dabf7e
N		 140/* Find first data disk in a raid6 stripe */
141static inline int raid6_d0(struct stripe_head *sh)
142{
67cc2b81
N		 143 if (sh->ddf_layout)
144 /* ddf always start from first device */
145 return 0;
146 /* md starts just after Q block */
d0dabf7e
N		 147 if (sh->qd_idx == sh->disks - 1)
148 return 0;
149 else
150 return sh->qd_idx + 1;
151}
16a53ecc
N		 152static inline int raid6_next_disk(int disk, int raid_disks)
153{
154 disk++;
155 return (disk < raid_disks) ? disk : 0;
156}
a4456856 157
d0dabf7e
N		 158/* When walking through the disks in a raid5, starting at raid6_d0,
159 * We need to map each disk to a 'slot', where the data disks are slot
160 * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk
161 * is raid_disks-1. This help does that mapping.
162 */
67cc2b81
N		 163static int raid6_idx_to_slot(int idx, struct stripe_head *sh,
164 int *count, int syndrome_disks)
d0dabf7e 165{
6629542e 166 int slot = *count;
67cc2b81 167
e4424fee 168 if (sh->ddf_layout)
6629542e 169 (*count)++;
d0dabf7e 170 if (idx == sh->pd_idx)
67cc2b81 171 return syndrome_disks;
d0dabf7e 172 if (idx == sh->qd_idx)
67cc2b81 173 return syndrome_disks + 1;
e4424fee 174 if (!sh->ddf_layout)
6629542e 175 (*count)++;
d0dabf7e
N		 176 return slot;
177}
178
a4456856
DW		 179static void return_io(struct bio *return_bi)
180{
181 struct bio *bi = return_bi;
182 while (bi) {
a4456856
DW		 183
184 return_bi = bi->bi_next;
185 bi->bi_next = NULL;
186 bi->bi_size = 0;
0a82a8d1
LT		 187 trace_block_bio_complete(bdev_get_queue(bi->bi_bdev),
188 bi, 0);
0e13fe23 189 bio_endio(bi, 0);
a4456856
DW		 190 bi = return_bi;
191 }
192}
193
d1688a6d 194static void print_raid5_conf (struct r5conf *conf);
1da177e4 195
600aa109
DW		 196static int stripe_operations_active(struct stripe_head *sh)
197{
198 return sh->check_state || sh->reconstruct_state ||
199 test_bit(STRIPE_BIOFILL_RUN, &sh->state) ||
200 test_bit(STRIPE_COMPUTE_RUN, &sh->state);
201}
202
4eb788df 203static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh)
1da177e4 204{
4eb788df
SL		 205 BUG_ON(!list_empty(&sh->lru));
206 BUG_ON(atomic_read(&conf->active_stripes)==0);
207 if (test_bit(STRIPE_HANDLE, &sh->state)) {
208 if (test_bit(STRIPE_DELAYED, &sh->state) &&
209 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
210 list_add_tail(&sh->lru, &conf->delayed_list);
211 else if (test_bit(STRIPE_BIT_DELAY, &sh->state) &&
212 sh->bm_seq - conf->seq_write > 0)
213 list_add_tail(&sh->lru, &conf->bitmap_list);
214 else {
215 clear_bit(STRIPE_DELAYED, &sh->state);
216 clear_bit(STRIPE_BIT_DELAY, &sh->state);
217 list_add_tail(&sh->lru, &conf->handle_list);
218 }
219 md_wakeup_thread(conf->mddev->thread);
220 } else {
221 BUG_ON(stripe_operations_active(sh));
222 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
223 if (atomic_dec_return(&conf->preread_active_stripes)
224 < IO_THRESHOLD)
225 md_wakeup_thread(conf->mddev->thread);
226 atomic_dec(&conf->active_stripes);
227 if (!test_bit(STRIPE_EXPANDING, &sh->state)) {
228 list_add_tail(&sh->lru, &conf->inactive_list);
229 wake_up(&conf->wait_for_stripe);
230 if (conf->retry_read_aligned)
231 md_wakeup_thread(conf->mddev->thread);
1da177e4
LT		 232 }
233 }
234}
d0dabf7e 235
4eb788df
SL		 236static void __release_stripe(struct r5conf *conf, struct stripe_head *sh)
237{
238 if (atomic_dec_and_test(&sh->count))
239 do_release_stripe(conf, sh);
240}
241
1da177e4
LT		 242static void release_stripe(struct stripe_head *sh)
243{
d1688a6d 244 struct r5conf *conf = sh->raid_conf;
1da177e4 245 unsigned long flags;
16a53ecc 246
4eb788df
SL		 247 local_irq_save(flags);
248 if (atomic_dec_and_lock(&sh->count, &conf->device_lock)) {
249 do_release_stripe(conf, sh);
250 spin_unlock(&conf->device_lock);
251 }
252 local_irq_restore(flags);
1da177e4
LT		 253}
254
fccddba0 255static inline void remove_hash(struct stripe_head *sh)
1da177e4 256{
45b4233c
DW		 257 pr_debug("remove_hash(), stripe %llu\n",
258 (unsigned long long)sh->sector);
1da177e4 259
fccddba0 260 hlist_del_init(&sh->hash);
1da177e4
LT		 261}
262
d1688a6d 263static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh)
1da177e4 264{
fccddba0 265 struct hlist_head *hp = stripe_hash(conf, sh->sector);
1da177e4 266
45b4233c
DW		 267 pr_debug("insert_hash(), stripe %llu\n",
268 (unsigned long long)sh->sector);
1da177e4 269
fccddba0 270 hlist_add_head(&sh->hash, hp);
1da177e4
LT		 271}
272
273
274/* find an idle stripe, make sure it is unhashed, and return it. */
d1688a6d 275static struct stripe_head *get_free_stripe(struct r5conf *conf)
1da177e4
LT		 276{
277 struct stripe_head *sh = NULL;
278 struct list_head *first;
279
1da177e4
LT		 280 if (list_empty(&conf->inactive_list))
281 goto out;
282 first = conf->inactive_list.next;
283 sh = list_entry(first, struct stripe_head, lru);
284 list_del_init(first);
285 remove_hash(sh);
286 atomic_inc(&conf->active_stripes);
287out:
288 return sh;
289}
290
e4e11e38 291static void shrink_buffers(struct stripe_head *sh)
1da177e4
LT		 292{
293 struct page *p;
294 int i;
e4e11e38 295 int num = sh->raid_conf->pool_size;
1da177e4 296
e4e11e38 297 for (i = 0; i < num ; i++) {
1da177e4
LT		 298 p = sh->dev[i].page;
299 if (!p)
300 continue;
301 sh->dev[i].page = NULL;
2d1f3b5d 302 put_page(p);
1da177e4
LT		 303 }
304}
305
e4e11e38 306static int grow_buffers(struct stripe_head *sh)
1da177e4
LT		 307{
308 int i;
e4e11e38 309 int num = sh->raid_conf->pool_size;
1da177e4 310
e4e11e38 311 for (i = 0; i < num; i++) {
1da177e4
LT		 312 struct page *page;
313
314 if (!(page = alloc_page(GFP_KERNEL))) {
315 return 1;
316 }
317 sh->dev[i].page = page;
318 }
319 return 0;
320}
321
784052ec 322static void raid5_build_block(struct stripe_head *sh, int i, int previous);
d1688a6d 323static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
911d4ee8 324 struct stripe_head *sh);
1da177e4 325
b5663ba4 326static void init_stripe(struct stripe_head *sh, sector_t sector, int previous)
1da177e4 327{
d1688a6d 328 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 329 int i;
1da177e4 330
78bafebd
ES		 331 BUG_ON(atomic_read(&sh->count) != 0);
332 BUG_ON(test_bit(STRIPE_HANDLE, &sh->state));
600aa109 333 BUG_ON(stripe_operations_active(sh));
d84e0f10 334
45b4233c 335 pr_debug("init_stripe called, stripe %llu\n",
1da177e4
LT		 336 (unsigned long long)sh->sector);
337
338 remove_hash(sh);
16a53ecc 339
86b42c71 340 sh->generation = conf->generation - previous;
b5663ba4 341 sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks;
1da177e4 342 sh->sector = sector;
911d4ee8 343 stripe_set_idx(sector, conf, previous, sh);
1da177e4
LT		 344 sh->state = 0;
345
7ecaa1e6
N		 346
347 for (i = sh->disks; i--; ) {
1da177e4
LT		 348 struct r5dev *dev = &sh->dev[i];
349
d84e0f10 350 if (dev->toread || dev->read || dev->towrite || dev->written ||
1da177e4 351 test_bit(R5_LOCKED, &dev->flags)) {
d84e0f10 352 printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n",
1da177e4 353 (unsigned long long)sh->sector, i, dev->toread,
d84e0f10 354 dev->read, dev->towrite, dev->written,
1da177e4 355 test_bit(R5_LOCKED, &dev->flags));
8cfa7b0f 356 WARN_ON(1);
1da177e4
LT		 357 }
358 dev->flags = 0;
784052ec 359 raid5_build_block(sh, i, previous);
1da177e4
LT		 360 }
361 insert_hash(conf, sh);
362}
363
d1688a6d 364static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector,
86b42c71 365 short generation)
1da177e4
LT		 366{
367 struct stripe_head *sh;
368
45b4233c 369 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
b67bfe0d 370 hlist_for_each_entry(sh, stripe_hash(conf, sector), hash)
86b42c71 371 if (sh->sector == sector && sh->generation == generation)
1da177e4 372 return sh;
45b4233c 373 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
1da177e4
LT		 374 return NULL;
375}
376
674806d6
N		 377/*
378 * Need to check if array has failed when deciding whether to:
379 * - start an array
380 * - remove non-faulty devices
381 * - add a spare
382 * - allow a reshape
383 * This determination is simple when no reshape is happening.
384 * However if there is a reshape, we need to carefully check
385 * both the before and after sections.
386 * This is because some failed devices may only affect one
387 * of the two sections, and some non-in_sync devices may
388 * be insync in the section most affected by failed devices.
389 */
908f4fbd 390static int calc_degraded(struct r5conf *conf)
674806d6 391{
908f4fbd 392 int degraded, degraded2;
674806d6 393 int i;
674806d6
N		 394
395 rcu_read_lock();
396 degraded = 0;
397 for (i = 0; i < conf->previous_raid_disks; i++) {
3cb03002 398 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
e5c86471
N		 399 if (rdev && test_bit(Faulty, &rdev->flags))
400 rdev = rcu_dereference(conf->disks[i].replacement);
674806d6
N		 401 if (!rdev || test_bit(Faulty, &rdev->flags))
402 degraded++;
403 else if (test_bit(In_sync, &rdev->flags))
404 ;
405 else
406 /* not in-sync or faulty.
407 * If the reshape increases the number of devices,
408 * this is being recovered by the reshape, so
409 * this 'previous' section is not in_sync.
410 * If the number of devices is being reduced however,
411 * the device can only be part of the array if
412 * we are reverting a reshape, so this section will
413 * be in-sync.
414 */
415 if (conf->raid_disks >= conf->previous_raid_disks)
416 degraded++;
417 }
418 rcu_read_unlock();
908f4fbd
N		 419 if (conf->raid_disks == conf->previous_raid_disks)
420 return degraded;
674806d6 421 rcu_read_lock();
908f4fbd 422 degraded2 = 0;
674806d6 423 for (i = 0; i < conf->raid_disks; i++) {
3cb03002 424 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
e5c86471
N		 425 if (rdev && test_bit(Faulty, &rdev->flags))
426 rdev = rcu_dereference(conf->disks[i].replacement);
674806d6 427 if (!rdev || test_bit(Faulty, &rdev->flags))
908f4fbd 428 degraded2++;
674806d6
N		 429 else if (test_bit(In_sync, &rdev->flags))
430 ;
431 else
432 /* not in-sync or faulty.
433 * If reshape increases the number of devices, this
434 * section has already been recovered, else it
435 * almost certainly hasn't.
436 */
437 if (conf->raid_disks <= conf->previous_raid_disks)
908f4fbd 438 degraded2++;
674806d6
N		 439 }
440 rcu_read_unlock();
908f4fbd
N		 441 if (degraded2 > degraded)
442 return degraded2;
443 return degraded;
444}
445
446static int has_failed(struct r5conf *conf)
447{
448 int degraded;
449
450 if (conf->mddev->reshape_position == MaxSector)
451 return conf->mddev->degraded > conf->max_degraded;
452
453 degraded = calc_degraded(conf);
674806d6
N		 454 if (degraded > conf->max_degraded)
455 return 1;
456 return 0;
457}
458
b5663ba4 459static struct stripe_head *
d1688a6d 460get_active_stripe(struct r5conf *conf, sector_t sector,
a8c906ca 461 int previous, int noblock, int noquiesce)
1da177e4
LT		 462{
463 struct stripe_head *sh;
464
45b4233c 465 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector);
1da177e4
LT		 466
467 spin_lock_irq(&conf->device_lock);
468
469 do {
72626685 470 wait_event_lock_irq(conf->wait_for_stripe,
a8c906ca 471 conf->quiesce == 0 || noquiesce,
eed8c02e 472 conf->device_lock);
86b42c71 473 sh = __find_stripe(conf, sector, conf->generation - previous);
1da177e4
LT		 474 if (!sh) {
475 if (!conf->inactive_blocked)
476 sh = get_free_stripe(conf);
477 if (noblock && sh == NULL)
478 break;
479 if (!sh) {
480 conf->inactive_blocked = 1;
481 wait_event_lock_irq(conf->wait_for_stripe,
482 !list_empty(&conf->inactive_list) &&
5036805b
N		 483 (atomic_read(&conf->active_stripes)
484 < (conf->max_nr_stripes *3/4)
1da177e4 485 || !conf->inactive_blocked),
eed8c02e 486 conf->device_lock);
1da177e4
LT		 487 conf->inactive_blocked = 0;
488 } else
b5663ba4 489 init_stripe(sh, sector, previous);
1da177e4
LT		 490 } else {
491 if (atomic_read(&sh->count)) {
ab69ae12 492 BUG_ON(!list_empty(&sh->lru)
8811b596
SL		 493 && !test_bit(STRIPE_EXPANDING, &sh->state)
494 && !test_bit(STRIPE_ON_UNPLUG_LIST, &sh->state));
1da177e4
LT		 495 } else {
496 if (!test_bit(STRIPE_HANDLE, &sh->state))
497 atomic_inc(&conf->active_stripes);
ff4e8d9a
N		 498 if (list_empty(&sh->lru) &&
499 !test_bit(STRIPE_EXPANDING, &sh->state))
16a53ecc
N		 500 BUG();
501 list_del_init(&sh->lru);
1da177e4
LT		 502 }
503 }
504 } while (sh == NULL);
505
506 if (sh)
507 atomic_inc(&sh->count);
508
509 spin_unlock_irq(&conf->device_lock);
510 return sh;
511}
512
05616be5
N		 513/* Determine if 'data_offset' or 'new_data_offset' should be used
514 * in this stripe_head.
515 */
516static int use_new_offset(struct r5conf *conf, struct stripe_head *sh)
517{
518 sector_t progress = conf->reshape_progress;
519 /* Need a memory barrier to make sure we see the value
520 * of conf->generation, or ->data_offset that was set before
521 * reshape_progress was updated.
522 */
523 smp_rmb();
524 if (progress == MaxSector)
525 return 0;
526 if (sh->generation == conf->generation - 1)
527 return 0;
528 /* We are in a reshape, and this is a new-generation stripe,
529 * so use new_data_offset.
530 */
531 return 1;
532}
533
6712ecf8
N		 534static void
535raid5_end_read_request(struct bio *bi, int error);
536static void
537raid5_end_write_request(struct bio *bi, int error);
91c00924 538
c4e5ac0a 539static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
91c00924 540{
d1688a6d 541 struct r5conf *conf = sh->raid_conf;
91c00924
DW		 542 int i, disks = sh->disks;
543
544 might_sleep();
545
546 for (i = disks; i--; ) {
547 int rw;
9a3e1101 548 int replace_only = 0;
977df362
N		 549 struct bio *bi, *rbi;
550 struct md_rdev *rdev, *rrdev = NULL;
e9c7469b
TH		 551 if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) {
552 if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags))
553 rw = WRITE_FUA;
554 else
555 rw = WRITE;
9e444768 556 if (test_bit(R5_Discard, &sh->dev[i].flags))
620125f2 557 rw |= REQ_DISCARD;
e9c7469b 558 } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags))
91c00924 559 rw = READ;
9a3e1101
N		 560 else if (test_and_clear_bit(R5_WantReplace,
561 &sh->dev[i].flags)) {
562 rw = WRITE;
563 replace_only = 1;
564 } else
91c00924 565 continue;
bc0934f0
SL		 566 if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags))
567 rw |= REQ_SYNC;
91c00924
DW		 568
569 bi = &sh->dev[i].req;
977df362 570 rbi = &sh->dev[i].rreq; /* For writing to replacement */
91c00924 571
91c00924 572 rcu_read_lock();
9a3e1101 573 rrdev = rcu_dereference(conf->disks[i].replacement);
dd054fce
N		 574 smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */
575 rdev = rcu_dereference(conf->disks[i].rdev);
576 if (!rdev) {
577 rdev = rrdev;
578 rrdev = NULL;
579 }
9a3e1101
N		 580 if (rw & WRITE) {
581 if (replace_only)
582 rdev = NULL;
dd054fce
N		 583 if (rdev == rrdev)
584 /* We raced and saw duplicates */
585 rrdev = NULL;
9a3e1101 586 } else {
dd054fce 587 if (test_bit(R5_ReadRepl, &sh->dev[i].flags) && rrdev)
9a3e1101
N		 588 rdev = rrdev;
589 rrdev = NULL;
590 }
977df362 591
91c00924
DW		 592 if (rdev && test_bit(Faulty, &rdev->flags))
593 rdev = NULL;
594 if (rdev)
595 atomic_inc(&rdev->nr_pending);
977df362
N		 596 if (rrdev && test_bit(Faulty, &rrdev->flags))
597 rrdev = NULL;
598 if (rrdev)
599 atomic_inc(&rrdev->nr_pending);
91c00924
DW		 600 rcu_read_unlock();
601
73e92e51 602 /* We have already checked bad blocks for reads. Now
977df362
N		 603 * need to check for writes. We never accept write errors
604 * on the replacement, so we don't to check rrdev.
73e92e51
N		 605 */
606 while ((rw & WRITE) && rdev &&
607 test_bit(WriteErrorSeen, &rdev->flags)) {
608 sector_t first_bad;
609 int bad_sectors;
610 int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
611 &first_bad, &bad_sectors);
612 if (!bad)
613 break;
614
615 if (bad < 0) {
616 set_bit(BlockedBadBlocks, &rdev->flags);
617 if (!conf->mddev->external &&
618 conf->mddev->flags) {
619 /* It is very unlikely, but we might
620 * still need to write out the
621 * bad block log - better give it
622 * a chance*/
623 md_check_recovery(conf->mddev);
624 }
1850753d 625 /*
626 * Because md_wait_for_blocked_rdev
627 * will dec nr_pending, we must
628 * increment it first.
629 */
630 atomic_inc(&rdev->nr_pending);
73e92e51
N		 631 md_wait_for_blocked_rdev(rdev, conf->mddev);
632 } else {
633 /* Acknowledged bad block - skip the write */
634 rdev_dec_pending(rdev, conf->mddev);
635 rdev = NULL;
636 }
637 }
638
91c00924 639 if (rdev) {
9a3e1101
N		 640 if (s->syncing || s->expanding || s->expanded
641 || s->replacing)
91c00924
DW		 642 md_sync_acct(rdev->bdev, STRIPE_SECTORS);
643
2b7497f0
DW		 644 set_bit(STRIPE_IO_STARTED, &sh->state);
645
2f6db2a7 646 bio_reset(bi);
91c00924 647 bi->bi_bdev = rdev->bdev;
2f6db2a7
KO		 648 bi->bi_rw = rw;
649 bi->bi_end_io = (rw & WRITE)
650 ? raid5_end_write_request
651 : raid5_end_read_request;
652 bi->bi_private = sh;
653
91c00924 654 pr_debug("%s: for %llu schedule op %ld on disc %d\n",
e46b272b 655 __func__, (unsigned long long)sh->sector,
91c00924
DW		 656 bi->bi_rw, i);
657 atomic_inc(&sh->count);
05616be5
N		 658 if (use_new_offset(conf, sh))
659 bi->bi_sector = (sh->sector
660 + rdev->new_data_offset);
661 else
662 bi->bi_sector = (sh->sector
663 + rdev->data_offset);
3f9e7c14 664 if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
665 bi->bi_rw |= REQ_FLUSH;
666
4997b72e 667 bi->bi_vcnt = 1;
91c00924
DW		 668 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
669 bi->bi_io_vec[0].bv_offset = 0;
670 bi->bi_size = STRIPE_SIZE;
7e44a926
SL		 671 /*
672 * If this is discard request, set bi_vcnt 0. We don't
673 * want to confuse SCSI because SCSI will replace payload
674 */
675 if (rw & REQ_DISCARD)
676 bi->bi_vcnt = 0;
977df362
N		 677 if (rrdev)
678 set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags);
e3620a3a
JB		 679
680 if (conf->mddev->gendisk)
681 trace_block_bio_remap(bdev_get_queue(bi->bi_bdev),
682 bi, disk_devt(conf->mddev->gendisk),
683 sh->dev[i].sector);
91c00924 684 generic_make_request(bi);
977df362
N		 685 }
686 if (rrdev) {
9a3e1101
N		 687 if (s->syncing || s->expanding || s->expanded
688 || s->replacing)
977df362
N		 689 md_sync_acct(rrdev->bdev, STRIPE_SECTORS);
690
691 set_bit(STRIPE_IO_STARTED, &sh->state);
692
2f6db2a7 693 bio_reset(rbi);
977df362 694 rbi->bi_bdev = rrdev->bdev;
2f6db2a7
KO		 695 rbi->bi_rw = rw;
696 BUG_ON(!(rw & WRITE));
697 rbi->bi_end_io = raid5_end_write_request;
698 rbi->bi_private = sh;
699
977df362
N		 700 pr_debug("%s: for %llu schedule op %ld on "
701 "replacement disc %d\n",
702 __func__, (unsigned long long)sh->sector,
703 rbi->bi_rw, i);
704 atomic_inc(&sh->count);
05616be5
N		 705 if (use_new_offset(conf, sh))
706 rbi->bi_sector = (sh->sector
707 + rrdev->new_data_offset);
708 else
709 rbi->bi_sector = (sh->sector
710 + rrdev->data_offset);
4997b72e 711 rbi->bi_vcnt = 1;
977df362
N		 712 rbi->bi_io_vec[0].bv_len = STRIPE_SIZE;
713 rbi->bi_io_vec[0].bv_offset = 0;
714 rbi->bi_size = STRIPE_SIZE;
7e44a926
SL		 715 /*
716 * If this is discard request, set bi_vcnt 0. We don't
717 * want to confuse SCSI because SCSI will replace payload
718 */
719 if (rw & REQ_DISCARD)
720 rbi->bi_vcnt = 0;
e3620a3a
JB		 721 if (conf->mddev->gendisk)
722 trace_block_bio_remap(bdev_get_queue(rbi->bi_bdev),
723 rbi, disk_devt(conf->mddev->gendisk),
724 sh->dev[i].sector);
977df362
N		 725 generic_make_request(rbi);
726 }
727 if (!rdev && !rrdev) {
b062962e 728 if (rw & WRITE)
91c00924
DW		 729 set_bit(STRIPE_DEGRADED, &sh->state);
730 pr_debug("skip op %ld on disc %d for sector %llu\n",
731 bi->bi_rw, i, (unsigned long long)sh->sector);
732 clear_bit(R5_LOCKED, &sh->dev[i].flags);
733 set_bit(STRIPE_HANDLE, &sh->state);
734 }
735 }
736}
737
738static struct dma_async_tx_descriptor *
739async_copy_data(int frombio, struct bio *bio, struct page *page,
740 sector_t sector, struct dma_async_tx_descriptor *tx)
741{
742 struct bio_vec *bvl;
743 struct page *bio_page;
744 int i;
745 int page_offset;
a08abd8c 746 struct async_submit_ctl submit;
0403e382 747 enum async_tx_flags flags = 0;
91c00924
DW		 748
749 if (bio->bi_sector >= sector)
750 page_offset = (signed)(bio->bi_sector - sector) * 512;
751 else
752 page_offset = (signed)(sector - bio->bi_sector) * -512;
a08abd8c 753
0403e382
DW		 754 if (frombio)
755 flags |= ASYNC_TX_FENCE;
756 init_async_submit(&submit, flags, tx, NULL, NULL, NULL);
757
91c00924 758 bio_for_each_segment(bvl, bio, i) {
fcde9075 759 int len = bvl->bv_len;
91c00924
DW		 760 int clen;
761 int b_offset = 0;
762
763 if (page_offset < 0) {
764 b_offset = -page_offset;
765 page_offset += b_offset;
766 len -= b_offset;
767 }
768
769 if (len > 0 && page_offset + len > STRIPE_SIZE)
770 clen = STRIPE_SIZE - page_offset;
771 else
772 clen = len;
773
774 if (clen > 0) {
fcde9075
NK		 775 b_offset += bvl->bv_offset;
776 bio_page = bvl->bv_page;
91c00924
DW		 777 if (frombio)
778 tx = async_memcpy(page, bio_page, page_offset,
a08abd8c 779 b_offset, clen, &submit);
91c00924
DW		 780 else
781 tx = async_memcpy(bio_page, page, b_offset,
a08abd8c 782 page_offset, clen, &submit);
91c00924 783 }
a08abd8c
DW		 784 /* chain the operations */
785 submit.depend_tx = tx;
786
91c00924
DW		 787 if (clen < len) /* hit end of page */
788 break;
789 page_offset += len;
790 }
791
792 return tx;
793}
794
795static void ops_complete_biofill(void *stripe_head_ref)
796{
797 struct stripe_head *sh = stripe_head_ref;
798 struct bio *return_bi = NULL;
e4d84909 799 int i;
91c00924 800
e46b272b 801 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 802 (unsigned long long)sh->sector);
803
804 /* clear completed biofills */
805 for (i = sh->disks; i--; ) {
806 struct r5dev *dev = &sh->dev[i];
91c00924
DW		 807
808 /* acknowledge completion of a biofill operation */
e4d84909
DW		 809 /* and check if we need to reply to a read request,
810 * new R5_Wantfill requests are held off until
83de75cc 811 * !STRIPE_BIOFILL_RUN
e4d84909
DW		 812 */
813 if (test_and_clear_bit(R5_Wantfill, &dev->flags)) {
91c00924 814 struct bio *rbi, *rbi2;
91c00924 815
91c00924
DW		 816 BUG_ON(!dev->read);
817 rbi = dev->read;
818 dev->read = NULL;
819 while (rbi && rbi->bi_sector <
820 dev->sector + STRIPE_SECTORS) {
821 rbi2 = r5_next_bio(rbi, dev->sector);
e7836bd6 822 if (!raid5_dec_bi_active_stripes(rbi)) {
91c00924
DW		 823 rbi->bi_next = return_bi;
824 return_bi = rbi;
825 }
91c00924
DW		 826 rbi = rbi2;
827 }
828 }
829 }
83de75cc 830 clear_bit(STRIPE_BIOFILL_RUN, &sh->state);
91c00924
DW		 831
832 return_io(return_bi);
833
e4d84909 834 set_bit(STRIPE_HANDLE, &sh->state);
91c00924
DW		 835 release_stripe(sh);
836}
837
838static void ops_run_biofill(struct stripe_head *sh)
839{
840 struct dma_async_tx_descriptor *tx = NULL;
a08abd8c 841 struct async_submit_ctl submit;
91c00924
DW		 842 int i;
843
e46b272b 844 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 845 (unsigned long long)sh->sector);
846
847 for (i = sh->disks; i--; ) {
848 struct r5dev *dev = &sh->dev[i];
849 if (test_bit(R5_Wantfill, &dev->flags)) {
850 struct bio *rbi;
b17459c0 851 spin_lock_irq(&sh->stripe_lock);
91c00924
DW		 852 dev->read = rbi = dev->toread;
853 dev->toread = NULL;
b17459c0 854 spin_unlock_irq(&sh->stripe_lock);
91c00924
DW		 855 while (rbi && rbi->bi_sector <
856 dev->sector + STRIPE_SECTORS) {
857 tx = async_copy_data(0, rbi, dev->page,
858 dev->sector, tx);
859 rbi = r5_next_bio(rbi, dev->sector);
860 }
861 }
862 }
863
864 atomic_inc(&sh->count);
a08abd8c
DW		 865 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL);
866 async_trigger_callback(&submit);
91c00924
DW		 867}
868
4e7d2c0a 869static void mark_target_uptodate(struct stripe_head *sh, int target)
91c00924 870{
4e7d2c0a 871 struct r5dev *tgt;
91c00924 872
4e7d2c0a
DW		 873 if (target < 0)
874 return;
91c00924 875
4e7d2c0a 876 tgt = &sh->dev[target];
91c00924
DW		 877 set_bit(R5_UPTODATE, &tgt->flags);
878 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
879 clear_bit(R5_Wantcompute, &tgt->flags);
4e7d2c0a
DW		 880}
881
ac6b53b6 882static void ops_complete_compute(void *stripe_head_ref)
91c00924
DW		 883{
884 struct stripe_head *sh = stripe_head_ref;
91c00924 885
e46b272b 886 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 887 (unsigned long long)sh->sector);
888
ac6b53b6 889 /* mark the computed target(s) as uptodate */
4e7d2c0a 890 mark_target_uptodate(sh, sh->ops.target);
ac6b53b6 891 mark_target_uptodate(sh, sh->ops.target2);
4e7d2c0a 892
ecc65c9b
DW		 893 clear_bit(STRIPE_COMPUTE_RUN, &sh->state);
894 if (sh->check_state == check_state_compute_run)
895 sh->check_state = check_state_compute_result;
91c00924
DW		 896 set_bit(STRIPE_HANDLE, &sh->state);
897 release_stripe(sh);
898}
899
d6f38f31
DW		 900/* return a pointer to the address conversion region of the scribble buffer */
901static addr_conv_t *to_addr_conv(struct stripe_head *sh,
902 struct raid5_percpu *percpu)
903{
904 return percpu->scribble + sizeof(struct page *) * (sh->disks + 2);
905}
906
907static struct dma_async_tx_descriptor *
908ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 909{
91c00924 910 int disks = sh->disks;
d6f38f31 911 struct page **xor_srcs = percpu->scribble;
91c00924
DW		 912 int target = sh->ops.target;
913 struct r5dev *tgt = &sh->dev[target];
914 struct page *xor_dest = tgt->page;
915 int count = 0;
916 struct dma_async_tx_descriptor *tx;
a08abd8c 917 struct async_submit_ctl submit;
91c00924
DW		 918 int i;
919
920 pr_debug("%s: stripe %llu block: %d\n",
e46b272b 921 __func__, (unsigned long long)sh->sector, target);
91c00924
DW		 922 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
923
924 for (i = disks; i--; )
925 if (i != target)
926 xor_srcs[count++] = sh->dev[i].page;
927
928 atomic_inc(&sh->count);
929
0403e382 930 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL,
ac6b53b6 931 ops_complete_compute, sh, to_addr_conv(sh, percpu));
91c00924 932 if (unlikely(count == 1))
a08abd8c 933 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
91c00924 934 else
a08abd8c 935 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924 936
91c00924
DW		 937 return tx;
938}
939
ac6b53b6
DW		 940/* set_syndrome_sources - populate source buffers for gen_syndrome
941 * @srcs - (struct page *) array of size sh->disks
942 * @sh - stripe_head to parse
943 *
944 * Populates srcs in proper layout order for the stripe and returns the
945 * 'count' of sources to be used in a call to async_gen_syndrome. The P
946 * destination buffer is recorded in srcs[count] and the Q destination
947 * is recorded in srcs[count+1]].
948 */
949static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh)
950{
951 int disks = sh->disks;
952 int syndrome_disks = sh->ddf_layout ? disks : (disks - 2);
953 int d0_idx = raid6_d0(sh);
954 int count;
955 int i;
956
957 for (i = 0; i < disks; i++)
5dd33c9a 958 srcs[i] = NULL;
ac6b53b6
DW		 959
960 count = 0;
961 i = d0_idx;
962 do {
963 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
964
965 srcs[slot] = sh->dev[i].page;
966 i = raid6_next_disk(i, disks);
967 } while (i != d0_idx);
ac6b53b6 968
e4424fee 969 return syndrome_disks;
ac6b53b6
DW		 970}
971
972static struct dma_async_tx_descriptor *
973ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu)
974{
975 int disks = sh->disks;
976 struct page **blocks = percpu->scribble;
977 int target;
978 int qd_idx = sh->qd_idx;
979 struct dma_async_tx_descriptor *tx;
980 struct async_submit_ctl submit;
981 struct r5dev *tgt;
982 struct page *dest;
983 int i;
984 int count;
985
986 if (sh->ops.target < 0)
987 target = sh->ops.target2;
988 else if (sh->ops.target2 < 0)
989 target = sh->ops.target;
91c00924 990 else
ac6b53b6
DW		 991 /* we should only have one valid target */
992 BUG();
993 BUG_ON(target < 0);
994 pr_debug("%s: stripe %llu block: %d\n",
995 __func__, (unsigned long long)sh->sector, target);
996
997 tgt = &sh->dev[target];
998 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
999 dest = tgt->page;
1000
1001 atomic_inc(&sh->count);
1002
1003 if (target == qd_idx) {
1004 count = set_syndrome_sources(blocks, sh);
1005 blocks[count] = NULL; /* regenerating p is not necessary */
1006 BUG_ON(blocks[count+1] != dest); /* q should already be set */
0403e382
DW		 1007 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1008 ops_complete_compute, sh,
ac6b53b6
DW		 1009 to_addr_conv(sh, percpu));
1010 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
1011 } else {
1012 /* Compute any data- or p-drive using XOR */
1013 count = 0;
1014 for (i = disks; i-- ; ) {
1015 if (i == target || i == qd_idx)
1016 continue;
1017 blocks[count++] = sh->dev[i].page;
1018 }
1019
0403e382
DW		 1020 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1021 NULL, ops_complete_compute, sh,
ac6b53b6
DW		 1022 to_addr_conv(sh, percpu));
1023 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit);
1024 }
91c00924 1025
91c00924
DW		 1026 return tx;
1027}
1028
ac6b53b6
DW		 1029static struct dma_async_tx_descriptor *
1030ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu)
1031{
1032 int i, count, disks = sh->disks;
1033 int syndrome_disks = sh->ddf_layout ? disks : disks-2;
1034 int d0_idx = raid6_d0(sh);
1035 int faila = -1, failb = -1;
1036 int target = sh->ops.target;
1037 int target2 = sh->ops.target2;
1038 struct r5dev *tgt = &sh->dev[target];
1039 struct r5dev *tgt2 = &sh->dev[target2];
1040 struct dma_async_tx_descriptor *tx;
1041 struct page **blocks = percpu->scribble;
1042 struct async_submit_ctl submit;
1043
1044 pr_debug("%s: stripe %llu block1: %d block2: %d\n",
1045 __func__, (unsigned long long)sh->sector, target, target2);
1046 BUG_ON(target < 0 || target2 < 0);
1047 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1048 BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags));
1049
6c910a78 1050 /* we need to open-code set_syndrome_sources to handle the
ac6b53b6
DW		 1051 * slot number conversion for 'faila' and 'failb'
1052 */
1053 for (i = 0; i < disks ; i++)
5dd33c9a 1054 blocks[i] = NULL;
ac6b53b6
DW		 1055 count = 0;
1056 i = d0_idx;
1057 do {
1058 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
1059
1060 blocks[slot] = sh->dev[i].page;
1061
1062 if (i == target)
1063 faila = slot;
1064 if (i == target2)
1065 failb = slot;
1066 i = raid6_next_disk(i, disks);
1067 } while (i != d0_idx);
ac6b53b6
DW		 1068
1069 BUG_ON(faila == failb);
1070 if (failb < faila)
1071 swap(faila, failb);
1072 pr_debug("%s: stripe: %llu faila: %d failb: %d\n",
1073 __func__, (unsigned long long)sh->sector, faila, failb);
1074
1075 atomic_inc(&sh->count);
1076
1077 if (failb == syndrome_disks+1) {
1078 /* Q disk is one of the missing disks */
1079 if (faila == syndrome_disks) {
1080 /* Missing P+Q, just recompute */
0403e382
DW		 1081 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1082 ops_complete_compute, sh,
1083 to_addr_conv(sh, percpu));
e4424fee 1084 return async_gen_syndrome(blocks, 0, syndrome_disks+2,
ac6b53b6
DW		 1085 STRIPE_SIZE, &submit);
1086 } else {
1087 struct page *dest;
1088 int data_target;
1089 int qd_idx = sh->qd_idx;
1090
1091 /* Missing D+Q: recompute D from P, then recompute Q */
1092 if (target == qd_idx)
1093 data_target = target2;
1094 else
1095 data_target = target;
1096
1097 count = 0;
1098 for (i = disks; i-- ; ) {
1099 if (i == data_target || i == qd_idx)
1100 continue;
1101 blocks[count++] = sh->dev[i].page;
1102 }
1103 dest = sh->dev[data_target].page;
0403e382
DW		 1104 init_async_submit(&submit,
1105 ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1106 NULL, NULL, NULL,
1107 to_addr_conv(sh, percpu));
ac6b53b6
DW		 1108 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE,
1109 &submit);
1110
1111 count = set_syndrome_sources(blocks, sh);
0403e382
DW		 1112 init_async_submit(&submit, ASYNC_TX_FENCE, tx,
1113 ops_complete_compute, sh,
1114 to_addr_conv(sh, percpu));
ac6b53b6
DW		 1115 return async_gen_syndrome(blocks, 0, count+2,
1116 STRIPE_SIZE, &submit);
1117 }
ac6b53b6 1118 } else {
6c910a78
DW		 1119 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1120 ops_complete_compute, sh,
1121 to_addr_conv(sh, percpu));
1122 if (failb == syndrome_disks) {
1123 /* We're missing D+P. */
1124 return async_raid6_datap_recov(syndrome_disks+2,
1125 STRIPE_SIZE, faila,
1126 blocks, &submit);
1127 } else {
1128 /* We're missing D+D. */
1129 return async_raid6_2data_recov(syndrome_disks+2,
1130 STRIPE_SIZE, faila, failb,
1131 blocks, &submit);
1132 }
ac6b53b6
DW		 1133 }
1134}
1135
1136
91c00924
DW		 1137static void ops_complete_prexor(void *stripe_head_ref)
1138{
1139 struct stripe_head *sh = stripe_head_ref;
1140
e46b272b 1141 pr_debug("%s: stripe %llu\n", __func__,
91c00924 1142 (unsigned long long)sh->sector);
91c00924
DW		 1143}
1144
1145static struct dma_async_tx_descriptor *
d6f38f31
DW		 1146ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu,
1147 struct dma_async_tx_descriptor *tx)
91c00924 1148{
91c00924 1149 int disks = sh->disks;
d6f38f31 1150 struct page **xor_srcs = percpu->scribble;
91c00924 1151 int count = 0, pd_idx = sh->pd_idx, i;
a08abd8c 1152 struct async_submit_ctl submit;
91c00924
DW		 1153
1154 /* existing parity data subtracted */
1155 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1156
e46b272b 1157 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1158 (unsigned long long)sh->sector);
1159
1160 for (i = disks; i--; ) {
1161 struct r5dev *dev = &sh->dev[i];
1162 /* Only process blocks that are known to be uptodate */
d8ee0728 1163 if (test_bit(R5_Wantdrain, &dev->flags))
91c00924
DW		 1164 xor_srcs[count++] = dev->page;
1165 }
1166
0403e382 1167 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
d6f38f31 1168 ops_complete_prexor, sh, to_addr_conv(sh, percpu));
a08abd8c 1169 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924
DW		 1170
1171 return tx;
1172}
1173
1174static struct dma_async_tx_descriptor *
d8ee0728 1175ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
91c00924
DW		 1176{
1177 int disks = sh->disks;
d8ee0728 1178 int i;
91c00924 1179
e46b272b 1180 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1181 (unsigned long long)sh->sector);
1182
1183 for (i = disks; i--; ) {
1184 struct r5dev *dev = &sh->dev[i];
1185 struct bio *chosen;
91c00924 1186
d8ee0728 1187 if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) {
91c00924
DW		 1188 struct bio *wbi;
1189
b17459c0 1190 spin_lock_irq(&sh->stripe_lock);
91c00924
DW		 1191 chosen = dev->towrite;
1192 dev->towrite = NULL;
1193 BUG_ON(dev->written);
1194 wbi = dev->written = chosen;
b17459c0 1195 spin_unlock_irq(&sh->stripe_lock);
91c00924
DW		 1196
1197 while (wbi && wbi->bi_sector <
1198 dev->sector + STRIPE_SECTORS) {
e9c7469b
TH		 1199 if (wbi->bi_rw & REQ_FUA)
1200 set_bit(R5_WantFUA, &dev->flags);
bc0934f0
SL		 1201 if (wbi->bi_rw & REQ_SYNC)
1202 set_bit(R5_SyncIO, &dev->flags);
9e444768 1203 if (wbi->bi_rw & REQ_DISCARD)
620125f2 1204 set_bit(R5_Discard, &dev->flags);
9e444768 1205 else
620125f2
SL		 1206 tx = async_copy_data(1, wbi, dev->page,
1207 dev->sector, tx);
91c00924
DW		 1208 wbi = r5_next_bio(wbi, dev->sector);
1209 }
1210 }
1211 }
1212
1213 return tx;
1214}
1215
ac6b53b6 1216static void ops_complete_reconstruct(void *stripe_head_ref)
91c00924
DW		 1217{
1218 struct stripe_head *sh = stripe_head_ref;
ac6b53b6
DW		 1219 int disks = sh->disks;
1220 int pd_idx = sh->pd_idx;
1221 int qd_idx = sh->qd_idx;
1222 int i;
9e444768 1223 bool fua = false, sync = false, discard = false;
91c00924 1224
e46b272b 1225 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1226 (unsigned long long)sh->sector);
1227
bc0934f0 1228 for (i = disks; i--; ) {
e9c7469b 1229 fua |= test_bit(R5_WantFUA, &sh->dev[i].flags);
bc0934f0 1230 sync |= test_bit(R5_SyncIO, &sh->dev[i].flags);
9e444768 1231 discard |= test_bit(R5_Discard, &sh->dev[i].flags);
bc0934f0 1232 }
e9c7469b 1233
91c00924
DW		 1234 for (i = disks; i--; ) {
1235 struct r5dev *dev = &sh->dev[i];
ac6b53b6 1236
e9c7469b 1237 if (dev->written || i == pd_idx || i == qd_idx) {
9e444768
SL		 1238 if (!discard)
1239 set_bit(R5_UPTODATE, &dev->flags);
e9c7469b
TH		 1240 if (fua)
1241 set_bit(R5_WantFUA, &dev->flags);
bc0934f0
SL		 1242 if (sync)
1243 set_bit(R5_SyncIO, &dev->flags);
e9c7469b 1244 }
91c00924
DW		 1245 }
1246
d8ee0728
DW		 1247 if (sh->reconstruct_state == reconstruct_state_drain_run)
1248 sh->reconstruct_state = reconstruct_state_drain_result;
1249 else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run)
1250 sh->reconstruct_state = reconstruct_state_prexor_drain_result;
1251 else {
1252 BUG_ON(sh->reconstruct_state != reconstruct_state_run);
1253 sh->reconstruct_state = reconstruct_state_result;
1254 }
91c00924
DW		 1255
1256 set_bit(STRIPE_HANDLE, &sh->state);
1257 release_stripe(sh);
1258}
1259
1260static void
ac6b53b6
DW		 1261ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu,
1262 struct dma_async_tx_descriptor *tx)
91c00924 1263{
91c00924 1264 int disks = sh->disks;
d6f38f31 1265 struct page **xor_srcs = percpu->scribble;
a08abd8c 1266 struct async_submit_ctl submit;
91c00924
DW		 1267 int count = 0, pd_idx = sh->pd_idx, i;
1268 struct page *xor_dest;
d8ee0728 1269 int prexor = 0;
91c00924 1270 unsigned long flags;
91c00924 1271
e46b272b 1272 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1273 (unsigned long long)sh->sector);
1274
620125f2
SL		 1275 for (i = 0; i < sh->disks; i++) {
1276 if (pd_idx == i)
1277 continue;
1278 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1279 break;
1280 }
1281 if (i >= sh->disks) {
1282 atomic_inc(&sh->count);
620125f2
SL		 1283 set_bit(R5_Discard, &sh->dev[pd_idx].flags);
1284 ops_complete_reconstruct(sh);
1285 return;
1286 }
91c00924
DW		 1287 /* check if prexor is active which means only process blocks
1288 * that are part of a read-modify-write (written)
1289 */
d8ee0728
DW		 1290 if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
1291 prexor = 1;
91c00924
DW		 1292 xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1293 for (i = disks; i--; ) {
1294 struct r5dev *dev = &sh->dev[i];
1295 if (dev->written)
1296 xor_srcs[count++] = dev->page;
1297 }
1298 } else {
1299 xor_dest = sh->dev[pd_idx].page;
1300 for (i = disks; i--; ) {
1301 struct r5dev *dev = &sh->dev[i];
1302 if (i != pd_idx)
1303 xor_srcs[count++] = dev->page;
1304 }
1305 }
1306
91c00924
DW		 1307 /* 1/ if we prexor'd then the dest is reused as a source
1308 * 2/ if we did not prexor then we are redoing the parity
1309 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
1310 * for the synchronous xor case
1311 */
88ba2aa5 1312 flags = ASYNC_TX_ACK |
91c00924
DW		 1313 (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
1314
1315 atomic_inc(&sh->count);
1316
ac6b53b6 1317 init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh,
d6f38f31 1318 to_addr_conv(sh, percpu));
a08abd8c
DW		 1319 if (unlikely(count == 1))
1320 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
1321 else
1322 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924
DW		 1323}
1324
ac6b53b6
DW		 1325static void
1326ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu,
1327 struct dma_async_tx_descriptor *tx)
1328{
1329 struct async_submit_ctl submit;
1330 struct page **blocks = percpu->scribble;
620125f2 1331 int count, i;
ac6b53b6
DW		 1332
1333 pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector);
1334
620125f2
SL		 1335 for (i = 0; i < sh->disks; i++) {
1336 if (sh->pd_idx == i || sh->qd_idx == i)
1337 continue;
1338 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1339 break;
1340 }
1341 if (i >= sh->disks) {
1342 atomic_inc(&sh->count);
620125f2
SL		 1343 set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
1344 set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
1345 ops_complete_reconstruct(sh);
1346 return;
1347 }
1348
ac6b53b6
DW		 1349 count = set_syndrome_sources(blocks, sh);
1350
1351 atomic_inc(&sh->count);
1352
1353 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct,
1354 sh, to_addr_conv(sh, percpu));
1355 async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
91c00924
DW		 1356}
1357
1358static void ops_complete_check(void *stripe_head_ref)
1359{
1360 struct stripe_head *sh = stripe_head_ref;
91c00924 1361
e46b272b 1362 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1363 (unsigned long long)sh->sector);
1364
ecc65c9b 1365 sh->check_state = check_state_check_result;
91c00924
DW		 1366 set_bit(STRIPE_HANDLE, &sh->state);
1367 release_stripe(sh);
1368}
1369
ac6b53b6 1370static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 1371{
91c00924 1372 int disks = sh->disks;
ac6b53b6
DW		 1373 int pd_idx = sh->pd_idx;
1374 int qd_idx = sh->qd_idx;
1375 struct page *xor_dest;
d6f38f31 1376 struct page **xor_srcs = percpu->scribble;
91c00924 1377 struct dma_async_tx_descriptor *tx;
a08abd8c 1378 struct async_submit_ctl submit;
ac6b53b6
DW		 1379 int count;
1380 int i;
91c00924 1381
e46b272b 1382 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1383 (unsigned long long)sh->sector);
1384
ac6b53b6
DW		 1385 count = 0;
1386 xor_dest = sh->dev[pd_idx].page;
1387 xor_srcs[count++] = xor_dest;
91c00924 1388 for (i = disks; i--; ) {
ac6b53b6
DW		 1389 if (i == pd_idx || i == qd_idx)
1390 continue;
1391 xor_srcs[count++] = sh->dev[i].page;
91c00924
DW		 1392 }
1393
d6f38f31
DW		 1394 init_async_submit(&submit, 0, NULL, NULL, NULL,
1395 to_addr_conv(sh, percpu));
099f53cb 1396 tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
a08abd8c 1397 &sh->ops.zero_sum_result, &submit);
91c00924 1398
91c00924 1399 atomic_inc(&sh->count);
a08abd8c
DW		 1400 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL);
1401 tx = async_trigger_callback(&submit);
91c00924
DW		 1402}
1403
ac6b53b6
DW		 1404static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp)
1405{
1406 struct page **srcs = percpu->scribble;
1407 struct async_submit_ctl submit;
1408 int count;
1409
1410 pr_debug("%s: stripe %llu checkp: %d\n", __func__,
1411 (unsigned long long)sh->sector, checkp);
1412
1413 count = set_syndrome_sources(srcs, sh);
1414 if (!checkp)
1415 srcs[count] = NULL;
91c00924 1416
91c00924 1417 atomic_inc(&sh->count);
ac6b53b6
DW		 1418 init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check,
1419 sh, to_addr_conv(sh, percpu));
1420 async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE,
1421 &sh->ops.zero_sum_result, percpu->spare_page, &submit);
91c00924
DW		 1422}
1423
51acbcec 1424static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
91c00924
DW		 1425{
1426 int overlap_clear = 0, i, disks = sh->disks;
1427 struct dma_async_tx_descriptor *tx = NULL;
d1688a6d 1428 struct r5conf *conf = sh->raid_conf;
ac6b53b6 1429 int level = conf->level;
d6f38f31
DW		 1430 struct raid5_percpu *percpu;
1431 unsigned long cpu;
91c00924 1432
d6f38f31
DW		 1433 cpu = get_cpu();
1434 percpu = per_cpu_ptr(conf->percpu, cpu);
83de75cc 1435 if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) {
91c00924
DW		 1436 ops_run_biofill(sh);
1437 overlap_clear++;
1438 }
1439
7b3a871e 1440 if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) {
ac6b53b6
DW		 1441 if (level < 6)
1442 tx = ops_run_compute5(sh, percpu);
1443 else {
1444 if (sh->ops.target2 < 0 || sh->ops.target < 0)
1445 tx = ops_run_compute6_1(sh, percpu);
1446 else
1447 tx = ops_run_compute6_2(sh, percpu);
1448 }
1449 /* terminate the chain if reconstruct is not set to be run */
1450 if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request))
7b3a871e
DW		 1451 async_tx_ack(tx);
1452 }
91c00924 1453
600aa109 1454 if (test_bit(STRIPE_OP_PREXOR, &ops_request))
d6f38f31 1455 tx = ops_run_prexor(sh, percpu, tx);
91c00924 1456
600aa109 1457 if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) {
d8ee0728 1458 tx = ops_run_biodrain(sh, tx);
91c00924
DW		 1459 overlap_clear++;
1460 }
1461
ac6b53b6
DW		 1462 if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) {
1463 if (level < 6)
1464 ops_run_reconstruct5(sh, percpu, tx);
1465 else
1466 ops_run_reconstruct6(sh, percpu, tx);
1467 }
91c00924 1468
ac6b53b6
DW		 1469 if (test_bit(STRIPE_OP_CHECK, &ops_request)) {
1470 if (sh->check_state == check_state_run)
1471 ops_run_check_p(sh, percpu);
1472 else if (sh->check_state == check_state_run_q)
1473 ops_run_check_pq(sh, percpu, 0);
1474 else if (sh->check_state == check_state_run_pq)
1475 ops_run_check_pq(sh, percpu, 1);
1476 else
1477 BUG();
1478 }
91c00924 1479
91c00924
DW		 1480 if (overlap_clear)
1481 for (i = disks; i--; ) {
1482 struct r5dev *dev = &sh->dev[i];
1483 if (test_and_clear_bit(R5_Overlap, &dev->flags))
1484 wake_up(&sh->raid_conf->wait_for_overlap);
1485 }
d6f38f31 1486 put_cpu();
91c00924
DW		 1487}
1488
d1688a6d 1489static int grow_one_stripe(struct r5conf *conf)
1da177e4
LT		 1490{
1491 struct stripe_head *sh;
6ce32846 1492 sh = kmem_cache_zalloc(conf->slab_cache, GFP_KERNEL);
3f294f4f
N		 1493 if (!sh)
1494 return 0;
6ce32846 1495
3f294f4f 1496 sh->raid_conf = conf;
3f294f4f 1497
b17459c0
SL		 1498 spin_lock_init(&sh->stripe_lock);
1499
e4e11e38
N		 1500 if (grow_buffers(sh)) {
1501 shrink_buffers(sh);
3f294f4f
N		 1502 kmem_cache_free(conf->slab_cache, sh);
1503 return 0;
1504 }
1505 /* we just created an active stripe so... */
1506 atomic_set(&sh->count, 1);
1507 atomic_inc(&conf->active_stripes);
1508 INIT_LIST_HEAD(&sh->lru);
1509 release_stripe(sh);
1510 return 1;
1511}
1512
d1688a6d 1513static int grow_stripes(struct r5conf *conf, int num)
3f294f4f 1514{
e18b890b 1515 struct kmem_cache *sc;
5e5e3e78 1516 int devs = max(conf->raid_disks, conf->previous_raid_disks);
1da177e4 1517
f4be6b43
N		 1518 if (conf->mddev->gendisk)
1519 sprintf(conf->cache_name[0],
1520 "raid%d-%s", conf->level, mdname(conf->mddev));
1521 else
1522 sprintf(conf->cache_name[0],
1523 "raid%d-%p", conf->level, conf->mddev);
1524 sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]);
1525
ad01c9e3
N		 1526 conf->active_name = 0;
1527 sc = kmem_cache_create(conf->cache_name[conf->active_name],
1da177e4 1528 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
20c2df83 1529 0, 0, NULL);
1da177e4
LT		 1530 if (!sc)
1531 return 1;
1532 conf->slab_cache = sc;
ad01c9e3 1533 conf->pool_size = devs;
16a53ecc 1534 while (num--)
3f294f4f 1535 if (!grow_one_stripe(conf))
1da177e4 1536 return 1;
1da177e4
LT		 1537 return 0;
1538}
29269553 1539
d6f38f31
DW		 1540/**
1541 * scribble_len - return the required size of the scribble region
1542 * @num - total number of disks in the array
1543 *
1544 * The size must be enough to contain:
1545 * 1/ a struct page pointer for each device in the array +2
1546 * 2/ room to convert each entry in (1) to its corresponding dma
1547 * (dma_map_page()) or page (page_address()) address.
1548 *
1549 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
1550 * calculate over all devices (not just the data blocks), using zeros in place
1551 * of the P and Q blocks.
1552 */
1553static size_t scribble_len(int num)
1554{
1555 size_t len;
1556
1557 len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2);
1558
1559 return len;
1560}
1561
d1688a6d 1562static int resize_stripes(struct r5conf *conf, int newsize)
ad01c9e3
N		 1563{
1564 /* Make all the stripes able to hold 'newsize' devices.
1565 * New slots in each stripe get 'page' set to a new page.
1566 *
1567 * This happens in stages:
1568 * 1/ create a new kmem_cache and allocate the required number of
1569 * stripe_heads.
83f0d77a 1570 * 2/ gather all the old stripe_heads and transfer the pages across
ad01c9e3
N		 1571 * to the new stripe_heads. This will have the side effect of
1572 * freezing the array as once all stripe_heads have been collected,
1573 * no IO will be possible. Old stripe heads are freed once their
1574 * pages have been transferred over, and the old kmem_cache is
1575 * freed when all stripes are done.
1576 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
1577 * we simple return a failre status - no need to clean anything up.
1578 * 4/ allocate new pages for the new slots in the new stripe_heads.
1579 * If this fails, we don't bother trying the shrink the
1580 * stripe_heads down again, we just leave them as they are.
1581 * As each stripe_head is processed the new one is released into
1582 * active service.
1583 *
1584 * Once step2 is started, we cannot afford to wait for a write,
1585 * so we use GFP_NOIO allocations.
1586 */
1587 struct stripe_head *osh, *nsh;
1588 LIST_HEAD(newstripes);
1589 struct disk_info *ndisks;
d6f38f31 1590 unsigned long cpu;
b5470dc5 1591 int err;
e18b890b 1592 struct kmem_cache *sc;
ad01c9e3
N		 1593 int i;
1594
1595 if (newsize <= conf->pool_size)
1596 return 0; /* never bother to shrink */
1597
b5470dc5
DW		 1598 err = md_allow_write(conf->mddev);
1599 if (err)
1600 return err;
2a2275d6 1601
ad01c9e3
N		 1602 /* Step 1 */
1603 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
1604 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
20c2df83 1605 0, 0, NULL);
ad01c9e3
N		 1606 if (!sc)
1607 return -ENOMEM;
1608
1609 for (i = conf->max_nr_stripes; i; i--) {
6ce32846 1610 nsh = kmem_cache_zalloc(sc, GFP_KERNEL);
ad01c9e3
N		 1611 if (!nsh)
1612 break;
1613
ad01c9e3 1614 nsh->raid_conf = conf;
cb13ff69 1615 spin_lock_init(&nsh->stripe_lock);
ad01c9e3
N		 1616
1617 list_add(&nsh->lru, &newstripes);
1618 }
1619 if (i) {
1620 /* didn't get enough, give up */
1621 while (!list_empty(&newstripes)) {
1622 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1623 list_del(&nsh->lru);
1624 kmem_cache_free(sc, nsh);
1625 }
1626 kmem_cache_destroy(sc);
1627 return -ENOMEM;
1628 }
1629 /* Step 2 - Must use GFP_NOIO now.
1630 * OK, we have enough stripes, start collecting inactive
1631 * stripes and copying them over
1632 */
1633 list_for_each_entry(nsh, &newstripes, lru) {
1634 spin_lock_irq(&conf->device_lock);
1635 wait_event_lock_irq(conf->wait_for_stripe,
1636 !list_empty(&conf->inactive_list),
eed8c02e 1637 conf->device_lock);
ad01c9e3
N		 1638 osh = get_free_stripe(conf);
1639 spin_unlock_irq(&conf->device_lock);
1640 atomic_set(&nsh->count, 1);
1641 for(i=0; i<conf->pool_size; i++)
1642 nsh->dev[i].page = osh->dev[i].page;
1643 for( ; i<newsize; i++)
1644 nsh->dev[i].page = NULL;
1645 kmem_cache_free(conf->slab_cache, osh);
1646 }
1647 kmem_cache_destroy(conf->slab_cache);
1648
1649 /* Step 3.
1650 * At this point, we are holding all the stripes so the array
1651 * is completely stalled, so now is a good time to resize
d6f38f31 1652 * conf->disks and the scribble region
ad01c9e3
N		 1653 */
1654 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
1655 if (ndisks) {
1656 for (i=0; i<conf->raid_disks; i++)
1657 ndisks[i] = conf->disks[i];
1658 kfree(conf->disks);
1659 conf->disks = ndisks;
1660 } else
1661 err = -ENOMEM;
1662
d6f38f31
DW		 1663 get_online_cpus();
1664 conf->scribble_len = scribble_len(newsize);
1665 for_each_present_cpu(cpu) {
1666 struct raid5_percpu *percpu;
1667 void *scribble;
1668
1669 percpu = per_cpu_ptr(conf->percpu, cpu);
1670 scribble = kmalloc(conf->scribble_len, GFP_NOIO);
1671
1672 if (scribble) {
1673 kfree(percpu->scribble);
1674 percpu->scribble = scribble;
1675 } else {
1676 err = -ENOMEM;
1677 break;
1678 }
1679 }
1680 put_online_cpus();
1681
ad01c9e3
N		 1682 /* Step 4, return new stripes to service */
1683 while(!list_empty(&newstripes)) {
1684 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1685 list_del_init(&nsh->lru);
d6f38f31 1686
ad01c9e3
N		 1687 for (i=conf->raid_disks; i < newsize; i++)
1688 if (nsh->dev[i].page == NULL) {
1689 struct page *p = alloc_page(GFP_NOIO);
1690 nsh->dev[i].page = p;
1691 if (!p)
1692 err = -ENOMEM;
1693 }
1694 release_stripe(nsh);
1695 }
1696 /* critical section pass, GFP_NOIO no longer needed */
1697
1698 conf->slab_cache = sc;
1699 conf->active_name = 1-conf->active_name;
1700 conf->pool_size = newsize;
1701 return err;
1702}
1da177e4 1703
d1688a6d 1704static int drop_one_stripe(struct r5conf *conf)
1da177e4
LT		 1705{
1706 struct stripe_head *sh;
1707
3f294f4f
N		 1708 spin_lock_irq(&conf->device_lock);
1709 sh = get_free_stripe(conf);
1710 spin_unlock_irq(&conf->device_lock);
1711 if (!sh)
1712 return 0;
78bafebd 1713 BUG_ON(atomic_read(&sh->count));
e4e11e38 1714 shrink_buffers(sh);
3f294f4f
N		 1715 kmem_cache_free(conf->slab_cache, sh);
1716 atomic_dec(&conf->active_stripes);
1717 return 1;
1718}
1719
d1688a6d 1720static void shrink_stripes(struct r5conf *conf)
3f294f4f
N		 1721{
1722 while (drop_one_stripe(conf))
1723 ;
1724
29fc7e3e
N		 1725 if (conf->slab_cache)
1726 kmem_cache_destroy(conf->slab_cache);
1da177e4
LT		 1727 conf->slab_cache = NULL;
1728}
1729
6712ecf8 1730static void raid5_end_read_request(struct bio * bi, int error)
1da177e4 1731{
99c0fb5f 1732 struct stripe_head *sh = bi->bi_private;
d1688a6d 1733 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 1734 int disks = sh->disks, i;
1da177e4 1735 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
d6950432 1736 char b[BDEVNAME_SIZE];
dd054fce 1737 struct md_rdev *rdev = NULL;
05616be5 1738 sector_t s;
1da177e4
LT		 1739
1740 for (i=0 ; i<disks; i++)
1741 if (bi == &sh->dev[i].req)
1742 break;
1743
45b4233c
DW		 1744 pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n",
1745 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
1da177e4
LT		 1746 uptodate);
1747 if (i == disks) {
1748 BUG();
6712ecf8 1749 return;
1da177e4 1750 }
14a75d3e 1751 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
dd054fce
N		 1752 /* If replacement finished while this request was outstanding,
1753 * 'replacement' might be NULL already.
1754 * In that case it moved down to 'rdev'.
1755 * rdev is not removed until all requests are finished.
1756 */
14a75d3e 1757 rdev = conf->disks[i].replacement;
dd054fce 1758 if (!rdev)
14a75d3e 1759 rdev = conf->disks[i].rdev;
1da177e4 1760
05616be5
N		 1761 if (use_new_offset(conf, sh))
1762 s = sh->sector + rdev->new_data_offset;
1763 else
1764 s = sh->sector + rdev->data_offset;
1da177e4 1765 if (uptodate) {
1da177e4 1766 set_bit(R5_UPTODATE, &sh->dev[i].flags);
4e5314b5 1767 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
14a75d3e
N		 1768 /* Note that this cannot happen on a
1769 * replacement device. We just fail those on
1770 * any error
1771 */
8bda470e
CD		 1772 printk_ratelimited(
1773 KERN_INFO
1774 "md/raid:%s: read error corrected"
1775 " (%lu sectors at %llu on %s)\n",
1776 mdname(conf->mddev), STRIPE_SECTORS,
05616be5 1777 (unsigned long long)s,
8bda470e 1778 bdevname(rdev->bdev, b));
ddd5115f 1779 atomic_add(STRIPE_SECTORS, &rdev->corrected_errors);
4e5314b5
N		 1780 clear_bit(R5_ReadError, &sh->dev[i].flags);
1781 clear_bit(R5_ReWrite, &sh->dev[i].flags);
3f9e7c14 1782 } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
1783 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
1784
14a75d3e
N		 1785 if (atomic_read(&rdev->read_errors))
1786 atomic_set(&rdev->read_errors, 0);
1da177e4 1787 } else {
14a75d3e 1788 const char *bdn = bdevname(rdev->bdev, b);
ba22dcbf 1789 int retry = 0;
2e8ac303 1790 int set_bad = 0;
d6950432 1791
1da177e4 1792 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
d6950432 1793 atomic_inc(&rdev->read_errors);
14a75d3e
N		 1794 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
1795 printk_ratelimited(
1796 KERN_WARNING
1797 "md/raid:%s: read error on replacement device "
1798 "(sector %llu on %s).\n",
1799 mdname(conf->mddev),
05616be5 1800 (unsigned long long)s,
14a75d3e 1801 bdn);
2e8ac303 1802 else if (conf->mddev->degraded >= conf->max_degraded) {
1803 set_bad = 1;
8bda470e
CD		 1804 printk_ratelimited(
1805 KERN_WARNING
1806 "md/raid:%s: read error not correctable "
1807 "(sector %llu on %s).\n",
1808 mdname(conf->mddev),
05616be5 1809 (unsigned long long)s,
8bda470e 1810 bdn);
2e8ac303 1811 } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) {
4e5314b5 1812 /* Oh, no!!! */
2e8ac303 1813 set_bad = 1;
8bda470e
CD		 1814 printk_ratelimited(
1815 KERN_WARNING
1816 "md/raid:%s: read error NOT corrected!! "
1817 "(sector %llu on %s).\n",
1818 mdname(conf->mddev),
05616be5 1819 (unsigned long long)s,
8bda470e 1820 bdn);
2e8ac303 1821 } else if (atomic_read(&rdev->read_errors)
ba22dcbf 1822 > conf->max_nr_stripes)
14f8d26b 1823 printk(KERN_WARNING
0c55e022 1824 "md/raid:%s: Too many read errors, failing device %s.\n",
d6950432 1825 mdname(conf->mddev), bdn);
ba22dcbf
N		 1826 else
1827 retry = 1;
1828 if (retry)
3f9e7c14 1829 if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) {
1830 set_bit(R5_ReadError, &sh->dev[i].flags);
1831 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
1832 } else
1833 set_bit(R5_ReadNoMerge, &sh->dev[i].flags);
ba22dcbf 1834 else {
4e5314b5
N		 1835 clear_bit(R5_ReadError, &sh->dev[i].flags);
1836 clear_bit(R5_ReWrite, &sh->dev[i].flags);
2e8ac303 1837 if (!(set_bad
1838 && test_bit(In_sync, &rdev->flags)
1839 && rdev_set_badblocks(
1840 rdev, sh->sector, STRIPE_SECTORS, 0)))
1841 md_error(conf->mddev, rdev);
ba22dcbf 1842 }
1da177e4 1843 }
14a75d3e 1844 rdev_dec_pending(rdev, conf->mddev);
1da177e4
LT		 1845 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1846 set_bit(STRIPE_HANDLE, &sh->state);
1847 release_stripe(sh);
1da177e4
LT		 1848}
1849
d710e138 1850static void raid5_end_write_request(struct bio *bi, int error)
1da177e4 1851{
99c0fb5f 1852 struct stripe_head *sh = bi->bi_private;
d1688a6d 1853 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 1854 int disks = sh->disks, i;
977df362 1855 struct md_rdev *uninitialized_var(rdev);
1da177e4 1856 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
b84db560
N		 1857 sector_t first_bad;
1858 int bad_sectors;
977df362 1859 int replacement = 0;
1da177e4 1860
977df362
N		 1861 for (i = 0 ; i < disks; i++) {
1862 if (bi == &sh->dev[i].req) {
1863 rdev = conf->disks[i].rdev;
1da177e4 1864 break;
977df362
N		 1865 }
1866 if (bi == &sh->dev[i].rreq) {
1867 rdev = conf->disks[i].replacement;
dd054fce
N		 1868 if (rdev)
1869 replacement = 1;
1870 else
1871 /* rdev was removed and 'replacement'
1872 * replaced it. rdev is not removed
1873 * until all requests are finished.
1874 */
1875 rdev = conf->disks[i].rdev;
977df362
N		 1876 break;
1877 }
1878 }
45b4233c 1879 pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n",
1da177e4
LT		 1880 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
1881 uptodate);
1882 if (i == disks) {
1883 BUG();
6712ecf8 1884 return;
1da177e4
LT		 1885 }
1886
977df362
N		 1887 if (replacement) {
1888 if (!uptodate)
1889 md_error(conf->mddev, rdev);
1890 else if (is_badblock(rdev, sh->sector,
1891 STRIPE_SECTORS,
1892 &first_bad, &bad_sectors))
1893 set_bit(R5_MadeGoodRepl, &sh->dev[i].flags);
1894 } else {
1895 if (!uptodate) {
6ba854e9 1896 set_bit(STRIPE_DEGRADED, &sh->state);
977df362
N		 1897 set_bit(WriteErrorSeen, &rdev->flags);
1898 set_bit(R5_WriteError, &sh->dev[i].flags);
3a6de292
N		 1899 if (!test_and_set_bit(WantReplacement, &rdev->flags))
1900 set_bit(MD_RECOVERY_NEEDED,
1901 &rdev->mddev->recovery);
977df362
N		 1902 } else if (is_badblock(rdev, sh->sector,
1903 STRIPE_SECTORS,
c0b32972 1904 &first_bad, &bad_sectors)) {
977df362 1905 set_bit(R5_MadeGood, &sh->dev[i].flags);
c0b32972
N		 1906 if (test_bit(R5_ReadError, &sh->dev[i].flags))
1907 /* That was a successful write so make
1908 * sure it looks like we already did
1909 * a re-write.
1910 */
1911 set_bit(R5_ReWrite, &sh->dev[i].flags);
1912 }
977df362
N		 1913 }
1914 rdev_dec_pending(rdev, conf->mddev);
1da177e4 1915
977df362
N		 1916 if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags))
1917 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1da177e4 1918 set_bit(STRIPE_HANDLE, &sh->state);
c04be0aa 1919 release_stripe(sh);
1da177e4
LT		 1920}
1921
784052ec 1922static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous);
1da177e4 1923
784052ec 1924static void raid5_build_block(struct stripe_head *sh, int i, int previous)
1da177e4
LT		 1925{
1926 struct r5dev *dev = &sh->dev[i];
1927
1928 bio_init(&dev->req);
1929 dev->req.bi_io_vec = &dev->vec;
1930 dev->req.bi_vcnt++;
1931 dev->req.bi_max_vecs++;
1da177e4 1932 dev->req.bi_private = sh;
995c4275 1933 dev->vec.bv_page = dev->page;
1da177e4 1934
977df362
N		 1935 bio_init(&dev->rreq);
1936 dev->rreq.bi_io_vec = &dev->rvec;
1937 dev->rreq.bi_vcnt++;
1938 dev->rreq.bi_max_vecs++;
1939 dev->rreq.bi_private = sh;
1940 dev->rvec.bv_page = dev->page;
1941
1da177e4 1942 dev->flags = 0;
784052ec 1943 dev->sector = compute_blocknr(sh, i, previous);
1da177e4
LT		 1944}
1945
fd01b88c 1946static void error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1947{
1948 char b[BDEVNAME_SIZE];
d1688a6d 1949 struct r5conf *conf = mddev->private;
908f4fbd 1950 unsigned long flags;
0c55e022 1951 pr_debug("raid456: error called\n");
1da177e4 1952
908f4fbd
N		 1953 spin_lock_irqsave(&conf->device_lock, flags);
1954 clear_bit(In_sync, &rdev->flags);
1955 mddev->degraded = calc_degraded(conf);
1956 spin_unlock_irqrestore(&conf->device_lock, flags);
1957 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1958
de393cde 1959 set_bit(Blocked, &rdev->flags);
6f8d0c77
N		 1960 set_bit(Faulty, &rdev->flags);
1961 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1962 printk(KERN_ALERT
1963 "md/raid:%s: Disk failure on %s, disabling device.\n"
1964 "md/raid:%s: Operation continuing on %d devices.\n",
1965 mdname(mddev),
1966 bdevname(rdev->bdev, b),
1967 mdname(mddev),
1968 conf->raid_disks - mddev->degraded);
16a53ecc 1969}
1da177e4
LT		 1970
1971/*
1972 * Input: a 'big' sector number,
1973 * Output: index of the data and parity disk, and the sector # in them.
1974 */
d1688a6d 1975static sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector,
911d4ee8
N		 1976 int previous, int *dd_idx,
1977 struct stripe_head *sh)
1da177e4 1978{
6e3b96ed 1979 sector_t stripe, stripe2;
35f2a591 1980 sector_t chunk_number;
1da177e4 1981 unsigned int chunk_offset;
911d4ee8 1982 int pd_idx, qd_idx;
67cc2b81 1983 int ddf_layout = 0;
1da177e4 1984 sector_t new_sector;
e183eaed
N		 1985 int algorithm = previous ? conf->prev_algo
1986 : conf->algorithm;
09c9e5fa
AN		 1987 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
1988 : conf->chunk_sectors;
112bf897
N		 1989 int raid_disks = previous ? conf->previous_raid_disks
1990 : conf->raid_disks;
1991 int data_disks = raid_disks - conf->max_degraded;
1da177e4
LT		 1992
1993 /* First compute the information on this sector */
1994
1995 /*
1996 * Compute the chunk number and the sector offset inside the chunk
1997 */
1998 chunk_offset = sector_div(r_sector, sectors_per_chunk);
1999 chunk_number = r_sector;
1da177e4
LT		 2000
2001 /*
2002 * Compute the stripe number
2003 */
35f2a591
N		 2004 stripe = chunk_number;
2005 *dd_idx = sector_div(stripe, data_disks);
6e3b96ed 2006 stripe2 = stripe;
1da177e4
LT		 2007 /*
2008 * Select the parity disk based on the user selected algorithm.
2009 */
84789554 2010 pd_idx = qd_idx = -1;
16a53ecc
N		 2011 switch(conf->level) {
2012 case 4:
911d4ee8 2013 pd_idx = data_disks;
16a53ecc
N		 2014 break;
2015 case 5:
e183eaed 2016 switch (algorithm) {
1da177e4 2017 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2018 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2019 if (*dd_idx >= pd_idx)
1da177e4
LT		 2020 (*dd_idx)++;
2021 break;
2022 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2023 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2024 if (*dd_idx >= pd_idx)
1da177e4
LT		 2025 (*dd_idx)++;
2026 break;
2027 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2028 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 2029 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4
LT		 2030 break;
2031 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2032 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 2033 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4 2034 break;
99c0fb5f
N		 2035 case ALGORITHM_PARITY_0:
2036 pd_idx = 0;
2037 (*dd_idx)++;
2038 break;
2039 case ALGORITHM_PARITY_N:
2040 pd_idx = data_disks;
2041 break;
1da177e4 2042 default:
99c0fb5f 2043 BUG();
16a53ecc
N		 2044 }
2045 break;
2046 case 6:
2047
e183eaed 2048 switch (algorithm) {
16a53ecc 2049 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 2050 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2051 qd_idx = pd_idx + 1;
2052 if (pd_idx == raid_disks-1) {
99c0fb5f 2053 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2054 qd_idx = 0;
2055 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2056 (*dd_idx) += 2; /* D D P Q D */
2057 break;
2058 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2059 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2060 qd_idx = pd_idx + 1;
2061 if (pd_idx == raid_disks-1) {
99c0fb5f 2062 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2063 qd_idx = 0;
2064 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2065 (*dd_idx) += 2; /* D D P Q D */
2066 break;
2067 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2068 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2069 qd_idx = (pd_idx + 1) % raid_disks;
2070 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc
N		 2071 break;
2072 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2073 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2074 qd_idx = (pd_idx + 1) % raid_disks;
2075 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc 2076 break;
99c0fb5f
N		 2077
2078 case ALGORITHM_PARITY_0:
2079 pd_idx = 0;
2080 qd_idx = 1;
2081 (*dd_idx) += 2;
2082 break;
2083 case ALGORITHM_PARITY_N:
2084 pd_idx = data_disks;
2085 qd_idx = data_disks + 1;
2086 break;
2087
2088 case ALGORITHM_ROTATING_ZERO_RESTART:
2089 /* Exactly the same as RIGHT_ASYMMETRIC, but or
2090 * of blocks for computing Q is different.
2091 */
6e3b96ed 2092 pd_idx = sector_div(stripe2, raid_disks);
99c0fb5f
N		 2093 qd_idx = pd_idx + 1;
2094 if (pd_idx == raid_disks-1) {
2095 (*dd_idx)++; /* Q D D D P */
2096 qd_idx = 0;
2097 } else if (*dd_idx >= pd_idx)
2098 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2099 ddf_layout = 1;
99c0fb5f
N		 2100 break;
2101
2102 case ALGORITHM_ROTATING_N_RESTART:
2103 /* Same a left_asymmetric, by first stripe is
2104 * D D D P Q rather than
2105 * Q D D D P
2106 */
6e3b96ed
N		 2107 stripe2 += 1;
2108 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2109 qd_idx = pd_idx + 1;
2110 if (pd_idx == raid_disks-1) {
2111 (*dd_idx)++; /* Q D D D P */
2112 qd_idx = 0;
2113 } else if (*dd_idx >= pd_idx)
2114 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2115 ddf_layout = 1;
99c0fb5f
N		 2116 break;
2117
2118 case ALGORITHM_ROTATING_N_CONTINUE:
2119 /* Same as left_symmetric but Q is before P */
6e3b96ed 2120 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2121 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
2122 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
67cc2b81 2123 ddf_layout = 1;
99c0fb5f
N		 2124 break;
2125
2126 case ALGORITHM_LEFT_ASYMMETRIC_6:
2127 /* RAID5 left_asymmetric, with Q on last device */
6e3b96ed 2128 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2129 if (*dd_idx >= pd_idx)
2130 (*dd_idx)++;
2131 qd_idx = raid_disks - 1;
2132 break;
2133
2134 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6e3b96ed 2135 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2136 if (*dd_idx >= pd_idx)
2137 (*dd_idx)++;
2138 qd_idx = raid_disks - 1;
2139 break;
2140
2141 case ALGORITHM_LEFT_SYMMETRIC_6:
6e3b96ed 2142 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2143 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2144 qd_idx = raid_disks - 1;
2145 break;
2146
2147 case ALGORITHM_RIGHT_SYMMETRIC_6:
6e3b96ed 2148 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2149 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2150 qd_idx = raid_disks - 1;
2151 break;
2152
2153 case ALGORITHM_PARITY_0_6:
2154 pd_idx = 0;
2155 (*dd_idx)++;
2156 qd_idx = raid_disks - 1;
2157 break;
2158
16a53ecc 2159 default:
99c0fb5f 2160 BUG();
16a53ecc
N		 2161 }
2162 break;
1da177e4
LT		 2163 }
2164
911d4ee8
N		 2165 if (sh) {
2166 sh->pd_idx = pd_idx;
2167 sh->qd_idx = qd_idx;
67cc2b81 2168 sh->ddf_layout = ddf_layout;
911d4ee8 2169 }
1da177e4
LT		 2170 /*
2171 * Finally, compute the new sector number
2172 */
2173 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
2174 return new_sector;
2175}
2176
2177
784052ec 2178static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous)
1da177e4 2179{
d1688a6d 2180 struct r5conf *conf = sh->raid_conf;
b875e531
N		 2181 int raid_disks = sh->disks;
2182 int data_disks = raid_disks - conf->max_degraded;
1da177e4 2183 sector_t new_sector = sh->sector, check;
09c9e5fa
AN		 2184 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2185 : conf->chunk_sectors;
e183eaed
N		 2186 int algorithm = previous ? conf->prev_algo
2187 : conf->algorithm;
1da177e4
LT		 2188 sector_t stripe;
2189 int chunk_offset;
35f2a591
N		 2190 sector_t chunk_number;
2191 int dummy1, dd_idx = i;
1da177e4 2192 sector_t r_sector;
911d4ee8 2193 struct stripe_head sh2;
1da177e4 2194
16a53ecc 2195
1da177e4
LT		 2196 chunk_offset = sector_div(new_sector, sectors_per_chunk);
2197 stripe = new_sector;
1da177e4 2198
16a53ecc
N		 2199 if (i == sh->pd_idx)
2200 return 0;
2201 switch(conf->level) {
2202 case 4: break;
2203 case 5:
e183eaed 2204 switch (algorithm) {
1da177e4
LT		 2205 case ALGORITHM_LEFT_ASYMMETRIC:
2206 case ALGORITHM_RIGHT_ASYMMETRIC:
2207 if (i > sh->pd_idx)
2208 i--;
2209 break;
2210 case ALGORITHM_LEFT_SYMMETRIC:
2211 case ALGORITHM_RIGHT_SYMMETRIC:
2212 if (i < sh->pd_idx)
2213 i += raid_disks;
2214 i -= (sh->pd_idx + 1);
2215 break;
99c0fb5f
N		 2216 case ALGORITHM_PARITY_0:
2217 i -= 1;
2218 break;
2219 case ALGORITHM_PARITY_N:
2220 break;
1da177e4 2221 default:
99c0fb5f 2222 BUG();
16a53ecc
N		 2223 }
2224 break;
2225 case 6:
d0dabf7e 2226 if (i == sh->qd_idx)
16a53ecc 2227 return 0; /* It is the Q disk */
e183eaed 2228 switch (algorithm) {
16a53ecc
N		 2229 case ALGORITHM_LEFT_ASYMMETRIC:
2230 case ALGORITHM_RIGHT_ASYMMETRIC:
99c0fb5f
N		 2231 case ALGORITHM_ROTATING_ZERO_RESTART:
2232 case ALGORITHM_ROTATING_N_RESTART:
2233 if (sh->pd_idx == raid_disks-1)
2234 i--; /* Q D D D P */
16a53ecc
N		 2235 else if (i > sh->pd_idx)
2236 i -= 2; /* D D P Q D */
2237 break;
2238 case ALGORITHM_LEFT_SYMMETRIC:
2239 case ALGORITHM_RIGHT_SYMMETRIC:
2240 if (sh->pd_idx == raid_disks-1)
2241 i--; /* Q D D D P */
2242 else {
2243 /* D D P Q D */
2244 if (i < sh->pd_idx)
2245 i += raid_disks;
2246 i -= (sh->pd_idx + 2);
2247 }
2248 break;
99c0fb5f
N		 2249 case ALGORITHM_PARITY_0:
2250 i -= 2;
2251 break;
2252 case ALGORITHM_PARITY_N:
2253 break;
2254 case ALGORITHM_ROTATING_N_CONTINUE:
e4424fee 2255 /* Like left_symmetric, but P is before Q */
99c0fb5f
N		 2256 if (sh->pd_idx == 0)
2257 i--; /* P D D D Q */
e4424fee
N		 2258 else {
2259 /* D D Q P D */
2260 if (i < sh->pd_idx)
2261 i += raid_disks;
2262 i -= (sh->pd_idx + 1);
2263 }
99c0fb5f
N		 2264 break;
2265 case ALGORITHM_LEFT_ASYMMETRIC_6:
2266 case ALGORITHM_RIGHT_ASYMMETRIC_6:
2267 if (i > sh->pd_idx)
2268 i--;
2269 break;
2270 case ALGORITHM_LEFT_SYMMETRIC_6:
2271 case ALGORITHM_RIGHT_SYMMETRIC_6:
2272 if (i < sh->pd_idx)
2273 i += data_disks + 1;
2274 i -= (sh->pd_idx + 1);
2275 break;
2276 case ALGORITHM_PARITY_0_6:
2277 i -= 1;
2278 break;
16a53ecc 2279 default:
99c0fb5f 2280 BUG();
16a53ecc
N		 2281 }
2282 break;
1da177e4
LT		 2283 }
2284
2285 chunk_number = stripe * data_disks + i;
35f2a591 2286 r_sector = chunk_number * sectors_per_chunk + chunk_offset;
1da177e4 2287
112bf897 2288 check = raid5_compute_sector(conf, r_sector,
784052ec 2289 previous, &dummy1, &sh2);
911d4ee8
N		 2290 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
2291 || sh2.qd_idx != sh->qd_idx) {
0c55e022
N		 2292 printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n",
2293 mdname(conf->mddev));
1da177e4
LT		 2294 return 0;
2295 }
2296 return r_sector;
2297}
2298
2299
600aa109 2300static void
c0f7bddb 2301schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
600aa109 2302 int rcw, int expand)
e33129d8
DW		 2303{
2304 int i, pd_idx = sh->pd_idx, disks = sh->disks;
d1688a6d 2305 struct r5conf *conf = sh->raid_conf;
c0f7bddb 2306 int level = conf->level;
e33129d8
DW		 2307
2308 if (rcw) {
e33129d8
DW		 2309
2310 for (i = disks; i--; ) {
2311 struct r5dev *dev = &sh->dev[i];
2312
2313 if (dev->towrite) {
2314 set_bit(R5_LOCKED, &dev->flags);
d8ee0728 2315 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2316 if (!expand)
2317 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2318 s->locked++;
e33129d8
DW		 2319 }
2320 }
ce7d363a
N		 2321 /* if we are not expanding this is a proper write request, and
2322 * there will be bios with new data to be drained into the
2323 * stripe cache
2324 */
2325 if (!expand) {
2326 if (!s->locked)
2327 /* False alarm, nothing to do */
2328 return;
2329 sh->reconstruct_state = reconstruct_state_drain_run;
2330 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2331 } else
2332 sh->reconstruct_state = reconstruct_state_run;
2333
2334 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
2335
c0f7bddb 2336 if (s->locked + conf->max_degraded == disks)
8b3e6cdc 2337 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
c0f7bddb 2338 atomic_inc(&conf->pending_full_writes);
e33129d8 2339 } else {
c0f7bddb 2340 BUG_ON(level == 6);
e33129d8
DW		 2341 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
2342 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
2343
e33129d8
DW		 2344 for (i = disks; i--; ) {
2345 struct r5dev *dev = &sh->dev[i];
2346 if (i == pd_idx)
2347 continue;
2348
e33129d8
DW		 2349 if (dev->towrite &&
2350 (test_bit(R5_UPTODATE, &dev->flags) ||
d8ee0728
DW		 2351 test_bit(R5_Wantcompute, &dev->flags))) {
2352 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2353 set_bit(R5_LOCKED, &dev->flags);
2354 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2355 s->locked++;
e33129d8
DW		 2356 }
2357 }
ce7d363a
N		 2358 if (!s->locked)
2359 /* False alarm - nothing to do */
2360 return;
2361 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
2362 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
2363 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2364 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
e33129d8
DW		 2365 }
2366
c0f7bddb 2367 /* keep the parity disk(s) locked while asynchronous operations
e33129d8
DW		 2368 * are in flight
2369 */
2370 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
2371 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
600aa109 2372 s->locked++;
e33129d8 2373
c0f7bddb
YT		 2374 if (level == 6) {
2375 int qd_idx = sh->qd_idx;
2376 struct r5dev *dev = &sh->dev[qd_idx];
2377
2378 set_bit(R5_LOCKED, &dev->flags);
2379 clear_bit(R5_UPTODATE, &dev->flags);
2380 s->locked++;
2381 }
2382
600aa109 2383 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
e46b272b 2384 __func__, (unsigned long long)sh->sector,
600aa109 2385 s->locked, s->ops_request);
e33129d8 2386}
16a53ecc 2387
1da177e4
LT		 2388/*
2389 * Each stripe/dev can have one or more bion attached.
16a53ecc 2390 * toread/towrite point to the first in a chain.
1da177e4
LT		 2391 * The bi_next chain must be in order.
2392 */
2393static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite)
2394{
2395 struct bio **bip;
d1688a6d 2396 struct r5conf *conf = sh->raid_conf;
72626685 2397 int firstwrite=0;
1da177e4 2398
cbe47ec5 2399 pr_debug("adding bi b#%llu to stripe s#%llu\n",
1da177e4
LT		 2400 (unsigned long long)bi->bi_sector,
2401 (unsigned long long)sh->sector);
2402
b17459c0
SL		 2403 /*
2404 * If several bio share a stripe. The bio bi_phys_segments acts as a
2405 * reference count to avoid race. The reference count should already be
2406 * increased before this function is called (for example, in
2407 * make_request()), so other bio sharing this stripe will not free the
2408 * stripe. If a stripe is owned by one stripe, the stripe lock will
2409 * protect it.
2410 */
2411 spin_lock_irq(&sh->stripe_lock);
72626685 2412 if (forwrite) {
1da177e4 2413 bip = &sh->dev[dd_idx].towrite;
7eaf7e8e 2414 if (*bip == NULL)
72626685
N		 2415 firstwrite = 1;
2416 } else
1da177e4
LT		 2417 bip = &sh->dev[dd_idx].toread;
2418 while (*bip && (*bip)->bi_sector < bi->bi_sector) {
f73a1c7d 2419 if (bio_end_sector(*bip) > bi->bi_sector)
1da177e4
LT		 2420 goto overlap;
2421 bip = & (*bip)->bi_next;
2422 }
f73a1c7d 2423 if (*bip && (*bip)->bi_sector < bio_end_sector(bi))
1da177e4
LT		 2424 goto overlap;
2425
78bafebd 2426 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
1da177e4
LT		 2427 if (*bip)
2428 bi->bi_next = *bip;
2429 *bip = bi;
e7836bd6 2430 raid5_inc_bi_active_stripes(bi);
72626685 2431
1da177e4
LT		 2432 if (forwrite) {
2433 /* check if page is covered */
2434 sector_t sector = sh->dev[dd_idx].sector;
2435 for (bi=sh->dev[dd_idx].towrite;
2436 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
2437 bi && bi->bi_sector <= sector;
2438 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
f73a1c7d
KO		 2439 if (bio_end_sector(bi) >= sector)
2440 sector = bio_end_sector(bi);
1da177e4
LT		 2441 }
2442 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
2443 set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags);
2444 }
cbe47ec5
N		 2445
2446 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
2447 (unsigned long long)(*bip)->bi_sector,
2448 (unsigned long long)sh->sector, dd_idx);
b97390ae 2449 spin_unlock_irq(&sh->stripe_lock);
cbe47ec5
N		 2450
2451 if (conf->mddev->bitmap && firstwrite) {
2452 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
2453 STRIPE_SECTORS, 0);
2454 sh->bm_seq = conf->seq_flush+1;
2455 set_bit(STRIPE_BIT_DELAY, &sh->state);
2456 }
1da177e4
LT		 2457 return 1;
2458
2459 overlap:
2460 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
b17459c0 2461 spin_unlock_irq(&sh->stripe_lock);
1da177e4
LT		 2462 return 0;
2463}
2464
d1688a6d 2465static void end_reshape(struct r5conf *conf);
29269553 2466
d1688a6d 2467static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
911d4ee8 2468 struct stripe_head *sh)
ccfcc3c1 2469{
784052ec 2470 int sectors_per_chunk =
09c9e5fa 2471 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
911d4ee8 2472 int dd_idx;
2d2063ce 2473 int chunk_offset = sector_div(stripe, sectors_per_chunk);
112bf897 2474 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
2d2063ce 2475
112bf897
N		 2476 raid5_compute_sector(conf,
2477 stripe * (disks - conf->max_degraded)
b875e531 2478 *sectors_per_chunk + chunk_offset,
112bf897 2479 previous,
911d4ee8 2480 &dd_idx, sh);
ccfcc3c1
N		 2481}
2482
a4456856 2483static void
d1688a6d 2484handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh,
a4456856
DW		 2485 struct stripe_head_state *s, int disks,
2486 struct bio **return_bi)
2487{
2488 int i;
2489 for (i = disks; i--; ) {
2490 struct bio *bi;
2491 int bitmap_end = 0;
2492
2493 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
3cb03002 2494 struct md_rdev *rdev;
a4456856
DW		 2495 rcu_read_lock();
2496 rdev = rcu_dereference(conf->disks[i].rdev);
2497 if (rdev && test_bit(In_sync, &rdev->flags))
7f0da59b
N		 2498 atomic_inc(&rdev->nr_pending);
2499 else
2500 rdev = NULL;
a4456856 2501 rcu_read_unlock();
7f0da59b
N		 2502 if (rdev) {
2503 if (!rdev_set_badblocks(
2504 rdev,
2505 sh->sector,
2506 STRIPE_SECTORS, 0))
2507 md_error(conf->mddev, rdev);
2508 rdev_dec_pending(rdev, conf->mddev);
2509 }
a4456856 2510 }
b17459c0 2511 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 2512 /* fail all writes first */
2513 bi = sh->dev[i].towrite;
2514 sh->dev[i].towrite = NULL;
b17459c0 2515 spin_unlock_irq(&sh->stripe_lock);
1ed850f3 2516 if (bi)
a4456856 2517 bitmap_end = 1;
a4456856
DW		 2518
2519 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2520 wake_up(&conf->wait_for_overlap);
2521
2522 while (bi && bi->bi_sector <
2523 sh->dev[i].sector + STRIPE_SECTORS) {
2524 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
2525 clear_bit(BIO_UPTODATE, &bi->bi_flags);
e7836bd6 2526 if (!raid5_dec_bi_active_stripes(bi)) {
a4456856
DW		 2527 md_write_end(conf->mddev);
2528 bi->bi_next = *return_bi;
2529 *return_bi = bi;
2530 }
2531 bi = nextbi;
2532 }
7eaf7e8e
SL		 2533 if (bitmap_end)
2534 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
2535 STRIPE_SECTORS, 0, 0);
2536 bitmap_end = 0;
a4456856
DW		 2537 /* and fail all 'written' */
2538 bi = sh->dev[i].written;
2539 sh->dev[i].written = NULL;
2540 if (bi) bitmap_end = 1;
2541 while (bi && bi->bi_sector <
2542 sh->dev[i].sector + STRIPE_SECTORS) {
2543 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
2544 clear_bit(BIO_UPTODATE, &bi->bi_flags);
e7836bd6 2545 if (!raid5_dec_bi_active_stripes(bi)) {
a4456856
DW		 2546 md_write_end(conf->mddev);
2547 bi->bi_next = *return_bi;
2548 *return_bi = bi;
2549 }
2550 bi = bi2;
2551 }
2552
b5e98d65
DW		 2553 /* fail any reads if this device is non-operational and
2554 * the data has not reached the cache yet.
2555 */
2556 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
2557 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
2558 test_bit(R5_ReadError, &sh->dev[i].flags))) {
143c4d05 2559 spin_lock_irq(&sh->stripe_lock);
a4456856
DW		 2560 bi = sh->dev[i].toread;
2561 sh->dev[i].toread = NULL;
143c4d05 2562 spin_unlock_irq(&sh->stripe_lock);
a4456856
DW		 2563 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2564 wake_up(&conf->wait_for_overlap);
a4456856
DW		 2565 while (bi && bi->bi_sector <
2566 sh->dev[i].sector + STRIPE_SECTORS) {
2567 struct bio *nextbi =
2568 r5_next_bio(bi, sh->dev[i].sector);
2569 clear_bit(BIO_UPTODATE, &bi->bi_flags);
e7836bd6 2570 if (!raid5_dec_bi_active_stripes(bi)) {
a4456856
DW		 2571 bi->bi_next = *return_bi;
2572 *return_bi = bi;
2573 }
2574 bi = nextbi;
2575 }
2576 }
a4456856
DW		 2577 if (bitmap_end)
2578 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
2579 STRIPE_SECTORS, 0, 0);
8cfa7b0f
N		 2580 /* If we were in the middle of a write the parity block might
2581 * still be locked - so just clear all R5_LOCKED flags
2582 */
2583 clear_bit(R5_LOCKED, &sh->dev[i].flags);
a4456856
DW		 2584 }
2585
8b3e6cdc
DW		 2586 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2587 if (atomic_dec_and_test(&conf->pending_full_writes))
2588 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 2589}
2590
7f0da59b 2591static void
d1688a6d 2592handle_failed_sync(struct r5conf *conf, struct stripe_head *sh,
7f0da59b
N		 2593 struct stripe_head_state *s)
2594{
2595 int abort = 0;
2596 int i;
2597
7f0da59b 2598 clear_bit(STRIPE_SYNCING, &sh->state);
f8dfcffd
N		 2599 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
2600 wake_up(&conf->wait_for_overlap);
7f0da59b 2601 s->syncing = 0;
9a3e1101 2602 s->replacing = 0;
7f0da59b 2603 /* There is nothing more to do for sync/check/repair.
18b9837e
N		 2604 * Don't even need to abort as that is handled elsewhere
2605 * if needed, and not always wanted e.g. if there is a known
2606 * bad block here.
9a3e1101 2607 * For recover/replace we need to record a bad block on all
7f0da59b
N		 2608 * non-sync devices, or abort the recovery
2609 */
18b9837e
N		 2610 if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) {
2611 /* During recovery devices cannot be removed, so
2612 * locking and refcounting of rdevs is not needed
2613 */
2614 for (i = 0; i < conf->raid_disks; i++) {
2615 struct md_rdev *rdev = conf->disks[i].rdev;
2616 if (rdev
2617 && !test_bit(Faulty, &rdev->flags)
2618 && !test_bit(In_sync, &rdev->flags)
2619 && !rdev_set_badblocks(rdev, sh->sector,
2620 STRIPE_SECTORS, 0))
2621 abort = 1;
2622 rdev = conf->disks[i].replacement;
2623 if (rdev
2624 && !test_bit(Faulty, &rdev->flags)
2625 && !test_bit(In_sync, &rdev->flags)
2626 && !rdev_set_badblocks(rdev, sh->sector,
2627 STRIPE_SECTORS, 0))
2628 abort = 1;
2629 }
2630 if (abort)
2631 conf->recovery_disabled =
2632 conf->mddev->recovery_disabled;
7f0da59b 2633 }
18b9837e 2634 md_done_sync(conf->mddev, STRIPE_SECTORS, !abort);
7f0da59b
N		 2635}
2636
9a3e1101
N		 2637static int want_replace(struct stripe_head *sh, int disk_idx)
2638{
2639 struct md_rdev *rdev;
2640 int rv = 0;
2641 /* Doing recovery so rcu locking not required */
2642 rdev = sh->raid_conf->disks[disk_idx].replacement;
2643 if (rdev
2644 && !test_bit(Faulty, &rdev->flags)
2645 && !test_bit(In_sync, &rdev->flags)
2646 && (rdev->recovery_offset <= sh->sector
2647 || rdev->mddev->recovery_cp <= sh->sector))
2648 rv = 1;
2649
2650 return rv;
2651}
2652
93b3dbce 2653/* fetch_block - checks the given member device to see if its data needs
1fe797e6
DW		 2654 * to be read or computed to satisfy a request.
2655 *
2656 * Returns 1 when no more member devices need to be checked, otherwise returns
93b3dbce 2657 * 0 to tell the loop in handle_stripe_fill to continue
f38e1219 2658 */
93b3dbce
N		 2659static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s,
2660 int disk_idx, int disks)
a4456856 2661{
5599becc 2662 struct r5dev *dev = &sh->dev[disk_idx];
f2b3b44d
N		 2663 struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]],
2664 &sh->dev[s->failed_num[1]] };
5599becc 2665
93b3dbce 2666 /* is the data in this block needed, and can we get it? */
5599becc
YT		 2667 if (!test_bit(R5_LOCKED, &dev->flags) &&
2668 !test_bit(R5_UPTODATE, &dev->flags) &&
2669 (dev->toread ||
2670 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
2671 s->syncing || s->expanding ||
9a3e1101 2672 (s->replacing && want_replace(sh, disk_idx)) ||
5d35e09c
N		 2673 (s->failed >= 1 && fdev[0]->toread) ||
2674 (s->failed >= 2 && fdev[1]->toread) ||
93b3dbce
N		 2675 (sh->raid_conf->level <= 5 && s->failed && fdev[0]->towrite &&
2676 !test_bit(R5_OVERWRITE, &fdev[0]->flags)) ||
2677 (sh->raid_conf->level == 6 && s->failed && s->to_write))) {
5599becc
YT		 2678 /* we would like to get this block, possibly by computing it,
2679 * otherwise read it if the backing disk is insync
2680 */
2681 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
2682 BUG_ON(test_bit(R5_Wantread, &dev->flags));
2683 if ((s->uptodate == disks - 1) &&
f2b3b44d
N		 2684 (s->failed && (disk_idx == s->failed_num[0] ||
2685 disk_idx == s->failed_num[1]))) {
5599becc
YT		 2686 /* have disk failed, and we're requested to fetch it;
2687 * do compute it
a4456856 2688 */
5599becc
YT		 2689 pr_debug("Computing stripe %llu block %d\n",
2690 (unsigned long long)sh->sector, disk_idx);
2691 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2692 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2693 set_bit(R5_Wantcompute, &dev->flags);
2694 sh->ops.target = disk_idx;
2695 sh->ops.target2 = -1; /* no 2nd target */
2696 s->req_compute = 1;
93b3dbce
N		 2697 /* Careful: from this point on 'uptodate' is in the eye
2698 * of raid_run_ops which services 'compute' operations
2699 * before writes. R5_Wantcompute flags a block that will
2700 * be R5_UPTODATE by the time it is needed for a
2701 * subsequent operation.
2702 */
5599becc
YT		 2703 s->uptodate++;
2704 return 1;
2705 } else if (s->uptodate == disks-2 && s->failed >= 2) {
2706 /* Computing 2-failure is *very* expensive; only
2707 * do it if failed >= 2
2708 */
2709 int other;
2710 for (other = disks; other--; ) {
2711 if (other == disk_idx)
2712 continue;
2713 if (!test_bit(R5_UPTODATE,
2714 &sh->dev[other].flags))
2715 break;
a4456856 2716 }
5599becc
YT		 2717 BUG_ON(other < 0);
2718 pr_debug("Computing stripe %llu blocks %d,%d\n",
2719 (unsigned long long)sh->sector,
2720 disk_idx, other);
2721 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2722 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2723 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
2724 set_bit(R5_Wantcompute, &sh->dev[other].flags);
2725 sh->ops.target = disk_idx;
2726 sh->ops.target2 = other;
2727 s->uptodate += 2;
2728 s->req_compute = 1;
2729 return 1;
2730 } else if (test_bit(R5_Insync, &dev->flags)) {
2731 set_bit(R5_LOCKED, &dev->flags);
2732 set_bit(R5_Wantread, &dev->flags);
2733 s->locked++;
2734 pr_debug("Reading block %d (sync=%d)\n",
2735 disk_idx, s->syncing);
a4456856
DW		 2736 }
2737 }
5599becc
YT		 2738
2739 return 0;
2740}
2741
2742/**
93b3dbce 2743 * handle_stripe_fill - read or compute data to satisfy pending requests.
5599becc 2744 */
93b3dbce
N		 2745static void handle_stripe_fill(struct stripe_head *sh,
2746 struct stripe_head_state *s,
2747 int disks)
5599becc
YT		 2748{
2749 int i;
2750
2751 /* look for blocks to read/compute, skip this if a compute
2752 * is already in flight, or if the stripe contents are in the
2753 * midst of changing due to a write
2754 */
2755 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
2756 !sh->reconstruct_state)
2757 for (i = disks; i--; )
93b3dbce 2758 if (fetch_block(sh, s, i, disks))
5599becc 2759 break;
a4456856
DW		 2760 set_bit(STRIPE_HANDLE, &sh->state);
2761}
2762
2763
1fe797e6 2764/* handle_stripe_clean_event
a4456856
DW		 2765 * any written block on an uptodate or failed drive can be returned.
2766 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
2767 * never LOCKED, so we don't need to test 'failed' directly.
2768 */
d1688a6d 2769static void handle_stripe_clean_event(struct r5conf *conf,
a4456856
DW		 2770 struct stripe_head *sh, int disks, struct bio **return_bi)
2771{
2772 int i;
2773 struct r5dev *dev;
f8dfcffd 2774 int discard_pending = 0;
a4456856
DW		 2775
2776 for (i = disks; i--; )
2777 if (sh->dev[i].written) {
2778 dev = &sh->dev[i];
2779 if (!test_bit(R5_LOCKED, &dev->flags) &&
9e444768 2780 (test_bit(R5_UPTODATE, &dev->flags) ||
ca64cae9 2781 test_bit(R5_Discard, &dev->flags))) {
a4456856
DW		 2782 /* We can return any write requests */
2783 struct bio *wbi, *wbi2;
45b4233c 2784 pr_debug("Return write for disc %d\n", i);
ca64cae9
N		 2785 if (test_and_clear_bit(R5_Discard, &dev->flags))
2786 clear_bit(R5_UPTODATE, &dev->flags);
a4456856
DW		 2787 wbi = dev->written;
2788 dev->written = NULL;
2789 while (wbi && wbi->bi_sector <
2790 dev->sector + STRIPE_SECTORS) {
2791 wbi2 = r5_next_bio(wbi, dev->sector);
e7836bd6 2792 if (!raid5_dec_bi_active_stripes(wbi)) {
a4456856
DW		 2793 md_write_end(conf->mddev);
2794 wbi->bi_next = *return_bi;
2795 *return_bi = wbi;
2796 }
2797 wbi = wbi2;
2798 }
7eaf7e8e
SL		 2799 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
2800 STRIPE_SECTORS,
a4456856 2801 !test_bit(STRIPE_DEGRADED, &sh->state),
7eaf7e8e 2802 0);
f8dfcffd
N		 2803 } else if (test_bit(R5_Discard, &dev->flags))
2804 discard_pending = 1;
2805 }
2806 if (!discard_pending &&
2807 test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) {
2808 clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
2809 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
2810 if (sh->qd_idx >= 0) {
2811 clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
2812 clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags);
2813 }
2814 /* now that discard is done we can proceed with any sync */
2815 clear_bit(STRIPE_DISCARD, &sh->state);
01e608d7
SL		 2816 /*
2817 * SCSI discard will change some bio fields and the stripe has
2818 * no updated data, so remove it from hash list and the stripe
2819 * will be reinitialized
2820 */
2821 spin_lock_irq(&conf->device_lock);
2822 remove_hash(sh);
2823 spin_unlock_irq(&conf->device_lock);
f8dfcffd
N		 2824 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state))
2825 set_bit(STRIPE_HANDLE, &sh->state);
2826
2827 }
8b3e6cdc
DW		 2828
2829 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2830 if (atomic_dec_and_test(&conf->pending_full_writes))
2831 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 2832}
2833
d1688a6d 2834static void handle_stripe_dirtying(struct r5conf *conf,
c8ac1803
N		 2835 struct stripe_head *sh,
2836 struct stripe_head_state *s,
2837 int disks)
a4456856
DW		 2838{
2839 int rmw = 0, rcw = 0, i;
a7854487
AL		 2840 sector_t recovery_cp = conf->mddev->recovery_cp;
2841
2842 /* RAID6 requires 'rcw' in current implementation.
2843 * Otherwise, check whether resync is now happening or should start.
2844 * If yes, then the array is dirty (after unclean shutdown or
2845 * initial creation), so parity in some stripes might be inconsistent.
2846 * In this case, we need to always do reconstruct-write, to ensure
2847 * that in case of drive failure or read-error correction, we
2848 * generate correct data from the parity.
2849 */
2850 if (conf->max_degraded == 2 ||
2851 (recovery_cp < MaxSector && sh->sector >= recovery_cp)) {
2852 /* Calculate the real rcw later - for now make it
c8ac1803
N		 2853 * look like rcw is cheaper
2854 */
2855 rcw = 1; rmw = 2;
a7854487
AL		 2856 pr_debug("force RCW max_degraded=%u, recovery_cp=%llu sh->sector=%llu\n",
2857 conf->max_degraded, (unsigned long long)recovery_cp,
2858 (unsigned long long)sh->sector);
c8ac1803 2859 } else for (i = disks; i--; ) {
a4456856
DW		 2860 /* would I have to read this buffer for read_modify_write */
2861 struct r5dev *dev = &sh->dev[i];
2862 if ((dev->towrite || i == sh->pd_idx) &&
2863 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 2864 !(test_bit(R5_UPTODATE, &dev->flags) ||
2865 test_bit(R5_Wantcompute, &dev->flags))) {
a4456856
DW		 2866 if (test_bit(R5_Insync, &dev->flags))
2867 rmw++;
2868 else
2869 rmw += 2*disks; /* cannot read it */
2870 }
2871 /* Would I have to read this buffer for reconstruct_write */
2872 if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx &&
2873 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 2874 !(test_bit(R5_UPTODATE, &dev->flags) ||
2875 test_bit(R5_Wantcompute, &dev->flags))) {
2876 if (test_bit(R5_Insync, &dev->flags)) rcw++;
a4456856
DW		 2877 else
2878 rcw += 2*disks;
2879 }
2880 }
45b4233c 2881 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
a4456856
DW		 2882 (unsigned long long)sh->sector, rmw, rcw);
2883 set_bit(STRIPE_HANDLE, &sh->state);
a9add5d9 2884 if (rmw < rcw && rmw > 0) {
a4456856 2885 /* prefer read-modify-write, but need to get some data */
e3620a3a
JB		 2886 if (conf->mddev->queue)
2887 blk_add_trace_msg(conf->mddev->queue,
2888 "raid5 rmw %llu %d",
2889 (unsigned long long)sh->sector, rmw);
a4456856
DW		 2890 for (i = disks; i--; ) {
2891 struct r5dev *dev = &sh->dev[i];
2892 if ((dev->towrite || i == sh->pd_idx) &&
2893 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 2894 !(test_bit(R5_UPTODATE, &dev->flags) ||
2895 test_bit(R5_Wantcompute, &dev->flags)) &&
a4456856
DW		 2896 test_bit(R5_Insync, &dev->flags)) {
2897 if (
2898 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
45b4233c 2899 pr_debug("Read_old block "
a9add5d9 2900 "%d for r-m-w\n", i);
a4456856
DW		 2901 set_bit(R5_LOCKED, &dev->flags);
2902 set_bit(R5_Wantread, &dev->flags);
2903 s->locked++;
2904 } else {
2905 set_bit(STRIPE_DELAYED, &sh->state);
2906 set_bit(STRIPE_HANDLE, &sh->state);
2907 }
2908 }
2909 }
a9add5d9 2910 }
c8ac1803 2911 if (rcw <= rmw && rcw > 0) {
a4456856 2912 /* want reconstruct write, but need to get some data */
a9add5d9 2913 int qread =0;
c8ac1803 2914 rcw = 0;
a4456856
DW		 2915 for (i = disks; i--; ) {
2916 struct r5dev *dev = &sh->dev[i];
2917 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
c8ac1803 2918 i != sh->pd_idx && i != sh->qd_idx &&
a4456856 2919 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219 2920 !(test_bit(R5_UPTODATE, &dev->flags) ||
c8ac1803
N		 2921 test_bit(R5_Wantcompute, &dev->flags))) {
2922 rcw++;
2923 if (!test_bit(R5_Insync, &dev->flags))
2924 continue; /* it's a failed drive */
a4456856
DW		 2925 if (
2926 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
45b4233c 2927 pr_debug("Read_old block "
a4456856
DW		 2928 "%d for Reconstruct\n", i);
2929 set_bit(R5_LOCKED, &dev->flags);
2930 set_bit(R5_Wantread, &dev->flags);
2931 s->locked++;
a9add5d9 2932 qread++;
a4456856
DW		 2933 } else {
2934 set_bit(STRIPE_DELAYED, &sh->state);
2935 set_bit(STRIPE_HANDLE, &sh->state);
2936 }
2937 }
2938 }
e3620a3a 2939 if (rcw && conf->mddev->queue)
a9add5d9
N		 2940 blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d",
2941 (unsigned long long)sh->sector,
2942 rcw, qread, test_bit(STRIPE_DELAYED, &sh->state));
c8ac1803 2943 }
a4456856
DW		 2944 /* now if nothing is locked, and if we have enough data,
2945 * we can start a write request
2946 */
f38e1219
DW		 2947 /* since handle_stripe can be called at any time we need to handle the
2948 * case where a compute block operation has been submitted and then a
ac6b53b6
DW		 2949 * subsequent call wants to start a write request. raid_run_ops only
2950 * handles the case where compute block and reconstruct are requested
f38e1219
DW		 2951 * simultaneously. If this is not the case then new writes need to be
2952 * held off until the compute completes.
2953 */
976ea8d4
DW		 2954 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
2955 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
2956 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
c0f7bddb 2957 schedule_reconstruction(sh, s, rcw == 0, 0);
a4456856
DW		 2958}
2959
d1688a6d 2960static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh,
a4456856
DW		 2961 struct stripe_head_state *s, int disks)
2962{
ecc65c9b 2963 struct r5dev *dev = NULL;
bd2ab670 2964
a4456856 2965 set_bit(STRIPE_HANDLE, &sh->state);
e89f8962 2966
ecc65c9b
DW		 2967 switch (sh->check_state) {
2968 case check_state_idle:
2969 /* start a new check operation if there are no failures */
bd2ab670 2970 if (s->failed == 0) {
bd2ab670 2971 BUG_ON(s->uptodate != disks);
ecc65c9b
DW		 2972 sh->check_state = check_state_run;
2973 set_bit(STRIPE_OP_CHECK, &s->ops_request);
bd2ab670 2974 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
bd2ab670 2975 s->uptodate--;
ecc65c9b 2976 break;
bd2ab670 2977 }
f2b3b44d 2978 dev = &sh->dev[s->failed_num[0]];
ecc65c9b
DW		 2979 /* fall through */
2980 case check_state_compute_result:
2981 sh->check_state = check_state_idle;
2982 if (!dev)
2983 dev = &sh->dev[sh->pd_idx];
2984
2985 /* check that a write has not made the stripe insync */
2986 if (test_bit(STRIPE_INSYNC, &sh->state))
2987 break;
c8894419 2988
a4456856 2989 /* either failed parity check, or recovery is happening */
a4456856
DW		 2990 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
2991 BUG_ON(s->uptodate != disks);
2992
2993 set_bit(R5_LOCKED, &dev->flags);
ecc65c9b 2994 s->locked++;
a4456856 2995 set_bit(R5_Wantwrite, &dev->flags);
830ea016 2996
a4456856 2997 clear_bit(STRIPE_DEGRADED, &sh->state);
a4456856 2998 set_bit(STRIPE_INSYNC, &sh->state);
ecc65c9b
DW		 2999 break;
3000 case check_state_run:
3001 break; /* we will be called again upon completion */
3002 case check_state_check_result:
3003 sh->check_state = check_state_idle;
3004
3005 /* if a failure occurred during the check operation, leave
3006 * STRIPE_INSYNC not set and let the stripe be handled again
3007 */
3008 if (s->failed)
3009 break;
3010
3011 /* handle a successful check operation, if parity is correct
3012 * we are done. Otherwise update the mismatch count and repair
3013 * parity if !MD_RECOVERY_CHECK
3014 */
ad283ea4 3015 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
ecc65c9b
DW		 3016 /* parity is correct (on disc,
3017 * not in buffer any more)
3018 */
3019 set_bit(STRIPE_INSYNC, &sh->state);
3020 else {
7f7583d4 3021 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
ecc65c9b
DW		 3022 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
3023 /* don't try to repair!! */
3024 set_bit(STRIPE_INSYNC, &sh->state);
3025 else {
3026 sh->check_state = check_state_compute_run;
976ea8d4 3027 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
ecc65c9b
DW		 3028 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3029 set_bit(R5_Wantcompute,
3030 &sh->dev[sh->pd_idx].flags);
3031 sh->ops.target = sh->pd_idx;
ac6b53b6 3032 sh->ops.target2 = -1;
ecc65c9b
DW		 3033 s->uptodate++;
3034 }
3035 }
3036 break;
3037 case check_state_compute_run:
3038 break;
3039 default:
3040 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
3041 __func__, sh->check_state,
3042 (unsigned long long) sh->sector);
3043 BUG();
a4456856
DW		 3044 }
3045}
3046
3047
d1688a6d 3048static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh,
36d1c647 3049 struct stripe_head_state *s,
f2b3b44d 3050 int disks)
a4456856 3051{
a4456856 3052 int pd_idx = sh->pd_idx;
34e04e87 3053 int qd_idx = sh->qd_idx;
d82dfee0 3054 struct r5dev *dev;
a4456856
DW		 3055
3056 set_bit(STRIPE_HANDLE, &sh->state);
3057
3058 BUG_ON(s->failed > 2);
d82dfee0 3059
a4456856
DW		 3060 /* Want to check and possibly repair P and Q.
3061 * However there could be one 'failed' device, in which
3062 * case we can only check one of them, possibly using the
3063 * other to generate missing data
3064 */
3065
d82dfee0
DW		 3066 switch (sh->check_state) {
3067 case check_state_idle:
3068 /* start a new check operation if there are < 2 failures */
f2b3b44d 3069 if (s->failed == s->q_failed) {
d82dfee0 3070 /* The only possible failed device holds Q, so it
a4456856
DW		 3071 * makes sense to check P (If anything else were failed,
3072 * we would have used P to recreate it).
3073 */
d82dfee0 3074 sh->check_state = check_state_run;
a4456856 3075 }
f2b3b44d 3076 if (!s->q_failed && s->failed < 2) {
d82dfee0 3077 /* Q is not failed, and we didn't use it to generate
a4456856
DW		 3078 * anything, so it makes sense to check it
3079 */
d82dfee0
DW		 3080 if (sh->check_state == check_state_run)
3081 sh->check_state = check_state_run_pq;
3082 else
3083 sh->check_state = check_state_run_q;
a4456856 3084 }
a4456856 3085
d82dfee0
DW		 3086 /* discard potentially stale zero_sum_result */
3087 sh->ops.zero_sum_result = 0;
a4456856 3088
d82dfee0
DW		 3089 if (sh->check_state == check_state_run) {
3090 /* async_xor_zero_sum destroys the contents of P */
3091 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
3092 s->uptodate--;
a4456856 3093 }
d82dfee0
DW		 3094 if (sh->check_state >= check_state_run &&
3095 sh->check_state <= check_state_run_pq) {
3096 /* async_syndrome_zero_sum preserves P and Q, so
3097 * no need to mark them !uptodate here
3098 */
3099 set_bit(STRIPE_OP_CHECK, &s->ops_request);
3100 break;
a4456856
DW		 3101 }
3102
d82dfee0
DW		 3103 /* we have 2-disk failure */
3104 BUG_ON(s->failed != 2);
3105 /* fall through */
3106 case check_state_compute_result:
3107 sh->check_state = check_state_idle;
a4456856 3108
d82dfee0
DW		 3109 /* check that a write has not made the stripe insync */
3110 if (test_bit(STRIPE_INSYNC, &sh->state))
3111 break;
a4456856
DW		 3112
3113 /* now write out any block on a failed drive,
d82dfee0 3114 * or P or Q if they were recomputed
a4456856 3115 */
d82dfee0 3116 BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */
a4456856 3117 if (s->failed == 2) {
f2b3b44d 3118 dev = &sh->dev[s->failed_num[1]];
a4456856
DW		 3119 s->locked++;
3120 set_bit(R5_LOCKED, &dev->flags);
3121 set_bit(R5_Wantwrite, &dev->flags);
3122 }
3123 if (s->failed >= 1) {
f2b3b44d 3124 dev = &sh->dev[s->failed_num[0]];
a4456856
DW		 3125 s->locked++;
3126 set_bit(R5_LOCKED, &dev->flags);
3127 set_bit(R5_Wantwrite, &dev->flags);
3128 }
d82dfee0 3129 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
a4456856
DW		 3130 dev = &sh->dev[pd_idx];
3131 s->locked++;
3132 set_bit(R5_LOCKED, &dev->flags);
3133 set_bit(R5_Wantwrite, &dev->flags);
3134 }
d82dfee0 3135 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
a4456856
DW		 3136 dev = &sh->dev[qd_idx];
3137 s->locked++;
3138 set_bit(R5_LOCKED, &dev->flags);
3139 set_bit(R5_Wantwrite, &dev->flags);
3140 }
3141 clear_bit(STRIPE_DEGRADED, &sh->state);
3142
3143 set_bit(STRIPE_INSYNC, &sh->state);
d82dfee0
DW		 3144 break;
3145 case check_state_run:
3146 case check_state_run_q:
3147 case check_state_run_pq:
3148 break; /* we will be called again upon completion */
3149 case check_state_check_result:
3150 sh->check_state = check_state_idle;
3151
3152 /* handle a successful check operation, if parity is correct
3153 * we are done. Otherwise update the mismatch count and repair
3154 * parity if !MD_RECOVERY_CHECK
3155 */
3156 if (sh->ops.zero_sum_result == 0) {
3157 /* both parities are correct */
3158 if (!s->failed)
3159 set_bit(STRIPE_INSYNC, &sh->state);
3160 else {
3161 /* in contrast to the raid5 case we can validate
3162 * parity, but still have a failure to write
3163 * back
3164 */
3165 sh->check_state = check_state_compute_result;
3166 /* Returning at this point means that we may go
3167 * off and bring p and/or q uptodate again so
3168 * we make sure to check zero_sum_result again
3169 * to verify if p or q need writeback
3170 */
3171 }
3172 } else {
7f7583d4 3173 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
d82dfee0
DW		 3174 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
3175 /* don't try to repair!! */
3176 set_bit(STRIPE_INSYNC, &sh->state);
3177 else {
3178 int *target = &sh->ops.target;
3179
3180 sh->ops.target = -1;
3181 sh->ops.target2 = -1;
3182 sh->check_state = check_state_compute_run;
3183 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3184 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3185 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
3186 set_bit(R5_Wantcompute,
3187 &sh->dev[pd_idx].flags);
3188 *target = pd_idx;
3189 target = &sh->ops.target2;
3190 s->uptodate++;
3191 }
3192 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
3193 set_bit(R5_Wantcompute,
3194 &sh->dev[qd_idx].flags);
3195 *target = qd_idx;
3196 s->uptodate++;
3197 }
3198 }
3199 }
3200 break;
3201 case check_state_compute_run:
3202 break;
3203 default:
3204 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
3205 __func__, sh->check_state,
3206 (unsigned long long) sh->sector);
3207 BUG();
a4456856
DW		 3208 }
3209}
3210
d1688a6d 3211static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh)
a4456856
DW		 3212{
3213 int i;
3214
3215 /* We have read all the blocks in this stripe and now we need to
3216 * copy some of them into a target stripe for expand.
3217 */
f0a50d37 3218 struct dma_async_tx_descriptor *tx = NULL;
a4456856
DW		 3219 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3220 for (i = 0; i < sh->disks; i++)
34e04e87 3221 if (i != sh->pd_idx && i != sh->qd_idx) {
911d4ee8 3222 int dd_idx, j;
a4456856 3223 struct stripe_head *sh2;
a08abd8c 3224 struct async_submit_ctl submit;
a4456856 3225
784052ec 3226 sector_t bn = compute_blocknr(sh, i, 1);
911d4ee8
N		 3227 sector_t s = raid5_compute_sector(conf, bn, 0,
3228 &dd_idx, NULL);
a8c906ca 3229 sh2 = get_active_stripe(conf, s, 0, 1, 1);
a4456856
DW		 3230 if (sh2 == NULL)
3231 /* so far only the early blocks of this stripe
3232 * have been requested. When later blocks
3233 * get requested, we will try again
3234 */
3235 continue;
3236 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
3237 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
3238 /* must have already done this block */
3239 release_stripe(sh2);
3240 continue;
3241 }
f0a50d37
DW		 3242
3243 /* place all the copies on one channel */
a08abd8c 3244 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
f0a50d37 3245 tx = async_memcpy(sh2->dev[dd_idx].page,
88ba2aa5 3246 sh->dev[i].page, 0, 0, STRIPE_SIZE,
a08abd8c 3247 &submit);
f0a50d37 3248
a4456856
DW		 3249 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
3250 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
3251 for (j = 0; j < conf->raid_disks; j++)
3252 if (j != sh2->pd_idx &&
86c374ba 3253 j != sh2->qd_idx &&
a4456856
DW		 3254 !test_bit(R5_Expanded, &sh2->dev[j].flags))
3255 break;
3256 if (j == conf->raid_disks) {
3257 set_bit(STRIPE_EXPAND_READY, &sh2->state);
3258 set_bit(STRIPE_HANDLE, &sh2->state);
3259 }
3260 release_stripe(sh2);
f0a50d37 3261
a4456856 3262 }
a2e08551 3263 /* done submitting copies, wait for them to complete */
749586b7 3264 async_tx_quiesce(&tx);
a4456856 3265}
1da177e4
LT		 3266
3267/*
3268 * handle_stripe - do things to a stripe.
3269 *
9a3e1101
N		 3270 * We lock the stripe by setting STRIPE_ACTIVE and then examine the
3271 * state of various bits to see what needs to be done.
1da177e4 3272 * Possible results:
9a3e1101
N		 3273 * return some read requests which now have data
3274 * return some write requests which are safely on storage
1da177e4
LT		 3275 * schedule a read on some buffers
3276 * schedule a write of some buffers
3277 * return confirmation of parity correctness
3278 *
1da177e4 3279 */
a4456856 3280
acfe726b 3281static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s)
1da177e4 3282{
d1688a6d 3283 struct r5conf *conf = sh->raid_conf;
f416885e 3284 int disks = sh->disks;
474af965
N		 3285 struct r5dev *dev;
3286 int i;
9a3e1101 3287 int do_recovery = 0;
1da177e4 3288
acfe726b
N		 3289 memset(s, 0, sizeof(*s));
3290
acfe726b
N		 3291 s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3292 s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
3293 s->failed_num[0] = -1;
3294 s->failed_num[1] = -1;
1da177e4 3295
acfe726b 3296 /* Now to look around and see what can be done */
1da177e4 3297 rcu_read_lock();
16a53ecc 3298 for (i=disks; i--; ) {
3cb03002 3299 struct md_rdev *rdev;
31c176ec
N		 3300 sector_t first_bad;
3301 int bad_sectors;
3302 int is_bad = 0;
acfe726b 3303
16a53ecc 3304 dev = &sh->dev[i];
1da177e4 3305
45b4233c 3306 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
9a3e1101
N		 3307 i, dev->flags,
3308 dev->toread, dev->towrite, dev->written);
6c0069c0
YT		 3309 /* maybe we can reply to a read
3310 *
3311 * new wantfill requests are only permitted while
3312 * ops_complete_biofill is guaranteed to be inactive
3313 */
3314 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
3315 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
3316 set_bit(R5_Wantfill, &dev->flags);
1da177e4 3317
16a53ecc 3318 /* now count some things */
cc94015a
N		 3319 if (test_bit(R5_LOCKED, &dev->flags))
3320 s->locked++;
3321 if (test_bit(R5_UPTODATE, &dev->flags))
3322 s->uptodate++;
2d6e4ecc 3323 if (test_bit(R5_Wantcompute, &dev->flags)) {
cc94015a
N		 3324 s->compute++;
3325 BUG_ON(s->compute > 2);
2d6e4ecc 3326 }
1da177e4 3327
acfe726b 3328 if (test_bit(R5_Wantfill, &dev->flags))
cc94015a 3329 s->to_fill++;
acfe726b 3330 else if (dev->toread)
cc94015a 3331 s->to_read++;
16a53ecc 3332 if (dev->towrite) {
cc94015a 3333 s->to_write++;
16a53ecc 3334 if (!test_bit(R5_OVERWRITE, &dev->flags))
cc94015a 3335 s->non_overwrite++;
16a53ecc 3336 }
a4456856 3337 if (dev->written)
cc94015a 3338 s->written++;
14a75d3e
N		 3339 /* Prefer to use the replacement for reads, but only
3340 * if it is recovered enough and has no bad blocks.
3341 */
3342 rdev = rcu_dereference(conf->disks[i].replacement);
3343 if (rdev && !test_bit(Faulty, &rdev->flags) &&
3344 rdev->recovery_offset >= sh->sector + STRIPE_SECTORS &&
3345 !is_badblock(rdev, sh->sector, STRIPE_SECTORS,
3346 &first_bad, &bad_sectors))
3347 set_bit(R5_ReadRepl, &dev->flags);
3348 else {
9a3e1101
N		 3349 if (rdev)
3350 set_bit(R5_NeedReplace, &dev->flags);
14a75d3e
N		 3351 rdev = rcu_dereference(conf->disks[i].rdev);
3352 clear_bit(R5_ReadRepl, &dev->flags);
3353 }
9283d8c5
N		 3354 if (rdev && test_bit(Faulty, &rdev->flags))
3355 rdev = NULL;
31c176ec
N		 3356 if (rdev) {
3357 is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
3358 &first_bad, &bad_sectors);
3359 if (s->blocked_rdev == NULL
3360 && (test_bit(Blocked, &rdev->flags)
3361 || is_bad < 0)) {
3362 if (is_bad < 0)
3363 set_bit(BlockedBadBlocks,
3364 &rdev->flags);
3365 s->blocked_rdev = rdev;
3366 atomic_inc(&rdev->nr_pending);
3367 }
6bfe0b49 3368 }
415e72d0
N		 3369 clear_bit(R5_Insync, &dev->flags);
3370 if (!rdev)
3371 /* Not in-sync */;
31c176ec
N		 3372 else if (is_bad) {
3373 /* also not in-sync */
18b9837e
N		 3374 if (!test_bit(WriteErrorSeen, &rdev->flags) &&
3375 test_bit(R5_UPTODATE, &dev->flags)) {
31c176ec
N		 3376 /* treat as in-sync, but with a read error
3377 * which we can now try to correct
3378 */
3379 set_bit(R5_Insync, &dev->flags);
3380 set_bit(R5_ReadError, &dev->flags);
3381 }
3382 } else if (test_bit(In_sync, &rdev->flags))
415e72d0 3383 set_bit(R5_Insync, &dev->flags);
30d7a483 3384 else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
415e72d0 3385 /* in sync if before recovery_offset */
30d7a483
N		 3386 set_bit(R5_Insync, &dev->flags);
3387 else if (test_bit(R5_UPTODATE, &dev->flags) &&
3388 test_bit(R5_Expanded, &dev->flags))
3389 /* If we've reshaped into here, we assume it is Insync.
3390 * We will shortly update recovery_offset to make
3391 * it official.
3392 */
3393 set_bit(R5_Insync, &dev->flags);
3394
c44bc449 3395 if (test_bit(R5_WriteError, &dev->flags)) {
14a75d3e
N		 3396 /* This flag does not apply to '.replacement'
3397 * only to .rdev, so make sure to check that*/
3398 struct md_rdev *rdev2 = rcu_dereference(
3399 conf->disks[i].rdev);
3400 if (rdev2 == rdev)
3401 clear_bit(R5_Insync, &dev->flags);
3402 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
bc2607f3 3403 s->handle_bad_blocks = 1;
14a75d3e 3404 atomic_inc(&rdev2->nr_pending);
bc2607f3
N		 3405 } else
3406 clear_bit(R5_WriteError, &dev->flags);
3407 }
c44bc449 3408 if (test_bit(R5_MadeGood, &dev->flags)) {
14a75d3e
N		 3409 /* This flag does not apply to '.replacement'
3410 * only to .rdev, so make sure to check that*/
3411 struct md_rdev *rdev2 = rcu_dereference(
3412 conf->disks[i].rdev);
3413 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
b84db560 3414 s->handle_bad_blocks = 1;
14a75d3e 3415 atomic_inc(&rdev2->nr_pending);
b84db560
N		 3416 } else
3417 clear_bit(R5_MadeGood, &dev->flags);
3418 }
977df362
N		 3419 if (test_bit(R5_MadeGoodRepl, &dev->flags)) {
3420 struct md_rdev *rdev2 = rcu_dereference(
3421 conf->disks[i].replacement);
3422 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
3423 s->handle_bad_blocks = 1;
3424 atomic_inc(&rdev2->nr_pending);
3425 } else
3426 clear_bit(R5_MadeGoodRepl, &dev->flags);
3427 }
415e72d0 3428 if (!test_bit(R5_Insync, &dev->flags)) {
16a53ecc
N		 3429 /* The ReadError flag will just be confusing now */
3430 clear_bit(R5_ReadError, &dev->flags);
3431 clear_bit(R5_ReWrite, &dev->flags);
1da177e4 3432 }
415e72d0
N		 3433 if (test_bit(R5_ReadError, &dev->flags))
3434 clear_bit(R5_Insync, &dev->flags);
3435 if (!test_bit(R5_Insync, &dev->flags)) {
cc94015a
N		 3436 if (s->failed < 2)
3437 s->failed_num[s->failed] = i;
3438 s->failed++;
9a3e1101
N		 3439 if (rdev && !test_bit(Faulty, &rdev->flags))
3440 do_recovery = 1;
415e72d0 3441 }
1da177e4 3442 }
9a3e1101
N		 3443 if (test_bit(STRIPE_SYNCING, &sh->state)) {
3444 /* If there is a failed device being replaced,
3445 * we must be recovering.
3446 * else if we are after recovery_cp, we must be syncing
c6d2e084 3447 * else if MD_RECOVERY_REQUESTED is set, we also are syncing.
9a3e1101
N		 3448 * else we can only be replacing
3449 * sync and recovery both need to read all devices, and so
3450 * use the same flag.
3451 */
3452 if (do_recovery ||
c6d2e084 3453 sh->sector >= conf->mddev->recovery_cp ||
3454 test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery)))
9a3e1101
N		 3455 s->syncing = 1;
3456 else
3457 s->replacing = 1;
3458 }
1da177e4 3459 rcu_read_unlock();
cc94015a
N		 3460}
3461
3462static void handle_stripe(struct stripe_head *sh)
3463{
3464 struct stripe_head_state s;
d1688a6d 3465 struct r5conf *conf = sh->raid_conf;
3687c061 3466 int i;
84789554
N		 3467 int prexor;
3468 int disks = sh->disks;
474af965 3469 struct r5dev *pdev, *qdev;
cc94015a
N		 3470
3471 clear_bit(STRIPE_HANDLE, &sh->state);
257a4b42 3472 if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) {
cc94015a
N		 3473 /* already being handled, ensure it gets handled
3474 * again when current action finishes */
3475 set_bit(STRIPE_HANDLE, &sh->state);
3476 return;
3477 }
3478
f8dfcffd
N		 3479 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
3480 spin_lock(&sh->stripe_lock);
3481 /* Cannot process 'sync' concurrently with 'discard' */
3482 if (!test_bit(STRIPE_DISCARD, &sh->state) &&
3483 test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
3484 set_bit(STRIPE_SYNCING, &sh->state);
3485 clear_bit(STRIPE_INSYNC, &sh->state);
c1dadcc1 3486 clear_bit(STRIPE_REPLACED, &sh->state);
f8dfcffd
N		 3487 }
3488 spin_unlock(&sh->stripe_lock);
cc94015a
N		 3489 }
3490 clear_bit(STRIPE_DELAYED, &sh->state);
3491
3492 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
3493 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
3494 (unsigned long long)sh->sector, sh->state,
3495 atomic_read(&sh->count), sh->pd_idx, sh->qd_idx,
3496 sh->check_state, sh->reconstruct_state);
3687c061 3497
acfe726b 3498 analyse_stripe(sh, &s);
c5a31000 3499
bc2607f3
N		 3500 if (s.handle_bad_blocks) {
3501 set_bit(STRIPE_HANDLE, &sh->state);
3502 goto finish;
3503 }
3504
474af965
N		 3505 if (unlikely(s.blocked_rdev)) {
3506 if (s.syncing || s.expanding || s.expanded ||
9a3e1101 3507 s.replacing || s.to_write || s.written) {
474af965
N		 3508 set_bit(STRIPE_HANDLE, &sh->state);
3509 goto finish;
3510 }
3511 /* There is nothing for the blocked_rdev to block */
3512 rdev_dec_pending(s.blocked_rdev, conf->mddev);
3513 s.blocked_rdev = NULL;
3514 }
3515
3516 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
3517 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
3518 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
3519 }
3520
3521 pr_debug("locked=%d uptodate=%d to_read=%d"
3522 " to_write=%d failed=%d failed_num=%d,%d\n",
3523 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
3524 s.failed_num[0], s.failed_num[1]);
3525 /* check if the array has lost more than max_degraded devices and,
3526 * if so, some requests might need to be failed.
3527 */
9a3f530f
N		 3528 if (s.failed > conf->max_degraded) {
3529 sh->check_state = 0;
3530 sh->reconstruct_state = 0;
3531 if (s.to_read+s.to_write+s.written)
3532 handle_failed_stripe(conf, sh, &s, disks, &s.return_bi);
9a3e1101 3533 if (s.syncing + s.replacing)
9a3f530f
N		 3534 handle_failed_sync(conf, sh, &s);
3535 }
474af965 3536
84789554
N		 3537 /* Now we check to see if any write operations have recently
3538 * completed
3539 */
3540 prexor = 0;
3541 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
3542 prexor = 1;
3543 if (sh->reconstruct_state == reconstruct_state_drain_result ||
3544 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
3545 sh->reconstruct_state = reconstruct_state_idle;
3546
3547 /* All the 'written' buffers and the parity block are ready to
3548 * be written back to disk
3549 */
9e444768
SL		 3550 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) &&
3551 !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags));
84789554 3552 BUG_ON(sh->qd_idx >= 0 &&
9e444768
SL		 3553 !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) &&
3554 !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags));
84789554
N		 3555 for (i = disks; i--; ) {
3556 struct r5dev *dev = &sh->dev[i];
3557 if (test_bit(R5_LOCKED, &dev->flags) &&
3558 (i == sh->pd_idx || i == sh->qd_idx ||
3559 dev->written)) {
3560 pr_debug("Writing block %d\n", i);
3561 set_bit(R5_Wantwrite, &dev->flags);
3562 if (prexor)
3563 continue;
318a3d59
N		 3564 if (s.failed > 1)
3565 continue;
84789554
N		 3566 if (!test_bit(R5_Insync, &dev->flags) ||
3567 ((i == sh->pd_idx || i == sh->qd_idx) &&
3568 s.failed == 0))
3569 set_bit(STRIPE_INSYNC, &sh->state);
3570 }
3571 }
3572 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3573 s.dec_preread_active = 1;
3574 }
3575
ef5b7c69
N		 3576 /*
3577 * might be able to return some write requests if the parity blocks
3578 * are safe, or on a failed drive
3579 */
3580 pdev = &sh->dev[sh->pd_idx];
3581 s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx)
3582 || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx);
3583 qdev = &sh->dev[sh->qd_idx];
3584 s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx)
3585 || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx)
3586 || conf->level < 6;
3587
3588 if (s.written &&
3589 (s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
3590 && !test_bit(R5_LOCKED, &pdev->flags)
3591 && (test_bit(R5_UPTODATE, &pdev->flags) ||
3592 test_bit(R5_Discard, &pdev->flags))))) &&
3593 (s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
3594 && !test_bit(R5_LOCKED, &qdev->flags)
3595 && (test_bit(R5_UPTODATE, &qdev->flags) ||
3596 test_bit(R5_Discard, &qdev->flags))))))
3597 handle_stripe_clean_event(conf, sh, disks, &s.return_bi);
3598
3599 /* Now we might consider reading some blocks, either to check/generate
3600 * parity, or to satisfy requests
3601 * or to load a block that is being partially written.
3602 */
3603 if (s.to_read || s.non_overwrite
3604 || (conf->level == 6 && s.to_write && s.failed)
3605 || (s.syncing && (s.uptodate + s.compute < disks))
3606 || s.replacing
3607 || s.expanding)
3608 handle_stripe_fill(sh, &s, disks);
3609
84789554
N		 3610 /* Now to consider new write requests and what else, if anything
3611 * should be read. We do not handle new writes when:
3612 * 1/ A 'write' operation (copy+xor) is already in flight.
3613 * 2/ A 'check' operation is in flight, as it may clobber the parity
3614 * block.
3615 */
3616 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
3617 handle_stripe_dirtying(conf, sh, &s, disks);
3618
3619 /* maybe we need to check and possibly fix the parity for this stripe
3620 * Any reads will already have been scheduled, so we just see if enough
3621 * data is available. The parity check is held off while parity
3622 * dependent operations are in flight.
3623 */
3624 if (sh->check_state ||
3625 (s.syncing && s.locked == 0 &&
3626 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
3627 !test_bit(STRIPE_INSYNC, &sh->state))) {
3628 if (conf->level == 6)
3629 handle_parity_checks6(conf, sh, &s, disks);
3630 else
3631 handle_parity_checks5(conf, sh, &s, disks);
3632 }
c5a31000 3633
c1dadcc1
N		 3634 if ((s.replacing || s.syncing) && s.locked == 0
3635 && !test_bit(STRIPE_COMPUTE_RUN, &sh->state)
3636 && !test_bit(STRIPE_REPLACED, &sh->state)) {
9a3e1101
N		 3637 /* Write out to replacement devices where possible */
3638 for (i = 0; i < conf->raid_disks; i++)
c1dadcc1
N		 3639 if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) {
3640 WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags));
9a3e1101
N		 3641 set_bit(R5_WantReplace, &sh->dev[i].flags);
3642 set_bit(R5_LOCKED, &sh->dev[i].flags);
3643 s.locked++;
3644 }
c1dadcc1
N		 3645 if (s.replacing)
3646 set_bit(STRIPE_INSYNC, &sh->state);
3647 set_bit(STRIPE_REPLACED, &sh->state);
9a3e1101
N		 3648 }
3649 if ((s.syncing || s.replacing) && s.locked == 0 &&
c1dadcc1 3650 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
9a3e1101 3651 test_bit(STRIPE_INSYNC, &sh->state)) {
c5a31000
N		 3652 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3653 clear_bit(STRIPE_SYNCING, &sh->state);
f8dfcffd
N		 3654 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
3655 wake_up(&conf->wait_for_overlap);
c5a31000
N		 3656 }
3657
3658 /* If the failed drives are just a ReadError, then we might need
3659 * to progress the repair/check process
3660 */
3661 if (s.failed <= conf->max_degraded && !conf->mddev->ro)
3662 for (i = 0; i < s.failed; i++) {
3663 struct r5dev *dev = &sh->dev[s.failed_num[i]];
3664 if (test_bit(R5_ReadError, &dev->flags)
3665 && !test_bit(R5_LOCKED, &dev->flags)
3666 && test_bit(R5_UPTODATE, &dev->flags)
3667 ) {
3668 if (!test_bit(R5_ReWrite, &dev->flags)) {
3669 set_bit(R5_Wantwrite, &dev->flags);
3670 set_bit(R5_ReWrite, &dev->flags);
3671 set_bit(R5_LOCKED, &dev->flags);
3672 s.locked++;
3673 } else {
3674 /* let's read it back */
3675 set_bit(R5_Wantread, &dev->flags);
3676 set_bit(R5_LOCKED, &dev->flags);
3677 s.locked++;
3678 }
3679 }
3680 }
3681
3682
3687c061
N		 3683 /* Finish reconstruct operations initiated by the expansion process */
3684 if (sh->reconstruct_state == reconstruct_state_result) {
3685 struct stripe_head *sh_src
3686 = get_active_stripe(conf, sh->sector, 1, 1, 1);
3687 if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) {
3688 /* sh cannot be written until sh_src has been read.
3689 * so arrange for sh to be delayed a little
3690 */
3691 set_bit(STRIPE_DELAYED, &sh->state);
3692 set_bit(STRIPE_HANDLE, &sh->state);
3693 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
3694 &sh_src->state))
3695 atomic_inc(&conf->preread_active_stripes);
3696 release_stripe(sh_src);
3697 goto finish;
3698 }
3699 if (sh_src)
3700 release_stripe(sh_src);
3701
3702 sh->reconstruct_state = reconstruct_state_idle;
3703 clear_bit(STRIPE_EXPANDING, &sh->state);
3704 for (i = conf->raid_disks; i--; ) {
3705 set_bit(R5_Wantwrite, &sh->dev[i].flags);
3706 set_bit(R5_LOCKED, &sh->dev[i].flags);
3707 s.locked++;
3708 }
3709 }
f416885e 3710
3687c061
N		 3711 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
3712 !sh->reconstruct_state) {
3713 /* Need to write out all blocks after computing parity */
3714 sh->disks = conf->raid_disks;
3715 stripe_set_idx(sh->sector, conf, 0, sh);
3716 schedule_reconstruction(sh, &s, 1, 1);
3717 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
3718 clear_bit(STRIPE_EXPAND_READY, &sh->state);
3719 atomic_dec(&conf->reshape_stripes);
3720 wake_up(&conf->wait_for_overlap);
3721 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3722 }
3723
3724 if (s.expanding && s.locked == 0 &&
3725 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
3726 handle_stripe_expansion(conf, sh);
16a53ecc 3727
3687c061 3728finish:
6bfe0b49 3729 /* wait for this device to become unblocked */
5f066c63
N		 3730 if (unlikely(s.blocked_rdev)) {
3731 if (conf->mddev->external)
3732 md_wait_for_blocked_rdev(s.blocked_rdev,
3733 conf->mddev);
3734 else
3735 /* Internal metadata will immediately
3736 * be written by raid5d, so we don't
3737 * need to wait here.
3738 */
3739 rdev_dec_pending(s.blocked_rdev,
3740 conf->mddev);
3741 }
6bfe0b49 3742
bc2607f3
N		 3743 if (s.handle_bad_blocks)
3744 for (i = disks; i--; ) {
3cb03002 3745 struct md_rdev *rdev;
bc2607f3
N		 3746 struct r5dev *dev = &sh->dev[i];
3747 if (test_and_clear_bit(R5_WriteError, &dev->flags)) {
3748 /* We own a safe reference to the rdev */
3749 rdev = conf->disks[i].rdev;
3750 if (!rdev_set_badblocks(rdev, sh->sector,
3751 STRIPE_SECTORS, 0))
3752 md_error(conf->mddev, rdev);
3753 rdev_dec_pending(rdev, conf->mddev);
3754 }
b84db560
N		 3755 if (test_and_clear_bit(R5_MadeGood, &dev->flags)) {
3756 rdev = conf->disks[i].rdev;
3757 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 3758 STRIPE_SECTORS, 0);
b84db560
N		 3759 rdev_dec_pending(rdev, conf->mddev);
3760 }
977df362
N		 3761 if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) {
3762 rdev = conf->disks[i].replacement;
dd054fce
N		 3763 if (!rdev)
3764 /* rdev have been moved down */
3765 rdev = conf->disks[i].rdev;
977df362 3766 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 3767 STRIPE_SECTORS, 0);
977df362
N		 3768 rdev_dec_pending(rdev, conf->mddev);
3769 }
bc2607f3
N		 3770 }
3771
6c0069c0
YT		 3772 if (s.ops_request)
3773 raid_run_ops(sh, s.ops_request);
3774
f0e43bcd 3775 ops_run_io(sh, &s);
16a53ecc 3776
c5709ef6 3777 if (s.dec_preread_active) {
729a1866 3778 /* We delay this until after ops_run_io so that if make_request
e9c7469b 3779 * is waiting on a flush, it won't continue until the writes
729a1866
N		 3780 * have actually been submitted.
3781 */
3782 atomic_dec(&conf->preread_active_stripes);
3783 if (atomic_read(&conf->preread_active_stripes) <
3784 IO_THRESHOLD)
3785 md_wakeup_thread(conf->mddev->thread);
3786 }
3787
c5709ef6 3788 return_io(s.return_bi);
16a53ecc 3789
257a4b42 3790 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
16a53ecc
N		 3791}
3792
d1688a6d 3793static void raid5_activate_delayed(struct r5conf *conf)
16a53ecc
N		 3794{
3795 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
3796 while (!list_empty(&conf->delayed_list)) {
3797 struct list_head *l = conf->delayed_list.next;
3798 struct stripe_head *sh;
3799 sh = list_entry(l, struct stripe_head, lru);
3800 list_del_init(l);
3801 clear_bit(STRIPE_DELAYED, &sh->state);
3802 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3803 atomic_inc(&conf->preread_active_stripes);
8b3e6cdc 3804 list_add_tail(&sh->lru, &conf->hold_list);
16a53ecc 3805 }
482c0834 3806 }
16a53ecc
N		 3807}
3808
d1688a6d 3809static void activate_bit_delay(struct r5conf *conf)
16a53ecc
N		 3810{
3811 /* device_lock is held */
3812 struct list_head head;
3813 list_add(&head, &conf->bitmap_list);
3814 list_del_init(&conf->bitmap_list);
3815 while (!list_empty(&head)) {
3816 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
3817 list_del_init(&sh->lru);
3818 atomic_inc(&sh->count);
3819 __release_stripe(conf, sh);
3820 }
3821}
3822
fd01b88c 3823int md_raid5_congested(struct mddev *mddev, int bits)
f022b2fd 3824{
d1688a6d 3825 struct r5conf *conf = mddev->private;
f022b2fd
N		 3826
3827 /* No difference between reads and writes. Just check
3828 * how busy the stripe_cache is
3829 */
3fa841d7 3830
f022b2fd
N		 3831 if (conf->inactive_blocked)
3832 return 1;
3833 if (conf->quiesce)
3834 return 1;
3835 if (list_empty_careful(&conf->inactive_list))
3836 return 1;
3837
3838 return 0;
3839}
11d8a6e3
N		 3840EXPORT_SYMBOL_GPL(md_raid5_congested);
3841
3842static int raid5_congested(void *data, int bits)
3843{
fd01b88c 3844 struct mddev *mddev = data;
11d8a6e3
N		 3845
3846 return mddev_congested(mddev, bits) ||
3847 md_raid5_congested(mddev, bits);
3848}
f022b2fd 3849
23032a0e
RBJ		 3850/* We want read requests to align with chunks where possible,
3851 * but write requests don't need to.
3852 */
cc371e66
AK		 3853static int raid5_mergeable_bvec(struct request_queue *q,
3854 struct bvec_merge_data *bvm,
3855 struct bio_vec *biovec)
23032a0e 3856{
fd01b88c 3857 struct mddev *mddev = q->queuedata;
cc371e66 3858 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
23032a0e 3859 int max;
9d8f0363 3860 unsigned int chunk_sectors = mddev->chunk_sectors;
cc371e66 3861 unsigned int bio_sectors = bvm->bi_size >> 9;
23032a0e 3862
cc371e66 3863 if ((bvm->bi_rw & 1) == WRITE)
23032a0e
RBJ		 3864 return biovec->bv_len; /* always allow writes to be mergeable */
3865
664e7c41
AN		 3866 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
3867 chunk_sectors = mddev->new_chunk_sectors;
23032a0e
RBJ		 3868 max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
3869 if (max < 0) max = 0;
3870 if (max <= biovec->bv_len && bio_sectors == 0)
3871 return biovec->bv_len;
3872 else
3873 return max;
3874}
3875
f679623f 3876
fd01b88c 3877static int in_chunk_boundary(struct mddev *mddev, struct bio *bio)
f679623f
RBJ		 3878{
3879 sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
9d8f0363 3880 unsigned int chunk_sectors = mddev->chunk_sectors;
aa8b57aa 3881 unsigned int bio_sectors = bio_sectors(bio);
f679623f 3882
664e7c41
AN		 3883 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
3884 chunk_sectors = mddev->new_chunk_sectors;
f679623f
RBJ		 3885 return chunk_sectors >=
3886 ((sector & (chunk_sectors - 1)) + bio_sectors);
3887}
3888
46031f9a
RBJ		 3889/*
3890 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
3891 * later sampled by raid5d.
3892 */
d1688a6d 3893static void add_bio_to_retry(struct bio *bi,struct r5conf *conf)
46031f9a
RBJ		 3894{
3895 unsigned long flags;
3896
3897 spin_lock_irqsave(&conf->device_lock, flags);
3898
3899 bi->bi_next = conf->retry_read_aligned_list;
3900 conf->retry_read_aligned_list = bi;
3901
3902 spin_unlock_irqrestore(&conf->device_lock, flags);
3903 md_wakeup_thread(conf->mddev->thread);
3904}
3905
3906
d1688a6d 3907static struct bio *remove_bio_from_retry(struct r5conf *conf)
46031f9a
RBJ		 3908{
3909 struct bio *bi;
3910
3911 bi = conf->retry_read_aligned;
3912 if (bi) {
3913 conf->retry_read_aligned = NULL;
3914 return bi;
3915 }
3916 bi = conf->retry_read_aligned_list;
3917 if(bi) {
387bb173 3918 conf->retry_read_aligned_list = bi->bi_next;
46031f9a 3919 bi->bi_next = NULL;
960e739d
JA		 3920 /*
3921 * this sets the active strip count to 1 and the processed
3922 * strip count to zero (upper 8 bits)
3923 */
e7836bd6 3924 raid5_set_bi_stripes(bi, 1); /* biased count of active stripes */
46031f9a
RBJ		 3925 }
3926
3927 return bi;
3928}
3929
3930
f679623f
RBJ		 3931/*
3932 * The "raid5_align_endio" should check if the read succeeded and if it
3933 * did, call bio_endio on the original bio (having bio_put the new bio
3934 * first).
3935 * If the read failed..
3936 */
6712ecf8 3937static void raid5_align_endio(struct bio *bi, int error)
f679623f
RBJ		 3938{
3939 struct bio* raid_bi = bi->bi_private;
fd01b88c 3940 struct mddev *mddev;
d1688a6d 3941 struct r5conf *conf;
46031f9a 3942 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
3cb03002 3943 struct md_rdev *rdev;
46031f9a 3944
f679623f 3945 bio_put(bi);
46031f9a 3946
46031f9a
RBJ		 3947 rdev = (void*)raid_bi->bi_next;
3948 raid_bi->bi_next = NULL;
2b7f2228
N		 3949 mddev = rdev->mddev;
3950 conf = mddev->private;
46031f9a
RBJ		 3951
3952 rdev_dec_pending(rdev, conf->mddev);
3953
3954 if (!error && uptodate) {
0a82a8d1
LT		 3955 trace_block_bio_complete(bdev_get_queue(raid_bi->bi_bdev),
3956 raid_bi, 0);
6712ecf8 3957 bio_endio(raid_bi, 0);
46031f9a
RBJ		 3958 if (atomic_dec_and_test(&conf->active_aligned_reads))
3959 wake_up(&conf->wait_for_stripe);
6712ecf8 3960 return;
46031f9a
RBJ		 3961 }
3962
3963
45b4233c 3964 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
46031f9a
RBJ		 3965
3966 add_bio_to_retry(raid_bi, conf);
f679623f
RBJ		 3967}
3968
387bb173
NB		 3969static int bio_fits_rdev(struct bio *bi)
3970{
165125e1 3971 struct request_queue *q = bdev_get_queue(bi->bi_bdev);
387bb173 3972
aa8b57aa 3973 if (bio_sectors(bi) > queue_max_sectors(q))
387bb173
NB		 3974 return 0;
3975 blk_recount_segments(q, bi);
8a78362c 3976 if (bi->bi_phys_segments > queue_max_segments(q))
387bb173
NB		 3977 return 0;
3978
3979 if (q->merge_bvec_fn)
3980 /* it's too hard to apply the merge_bvec_fn at this stage,
3981 * just just give up
3982 */
3983 return 0;
3984
3985 return 1;
3986}
3987
3988
fd01b88c 3989static int chunk_aligned_read(struct mddev *mddev, struct bio * raid_bio)
f679623f 3990{
d1688a6d 3991 struct r5conf *conf = mddev->private;
8553fe7e 3992 int dd_idx;
f679623f 3993 struct bio* align_bi;
3cb03002 3994 struct md_rdev *rdev;
671488cc 3995 sector_t end_sector;
f679623f
RBJ		 3996
3997 if (!in_chunk_boundary(mddev, raid_bio)) {
45b4233c 3998 pr_debug("chunk_aligned_read : non aligned\n");
f679623f
RBJ		 3999 return 0;
4000 }
4001 /*
a167f663 4002 * use bio_clone_mddev to make a copy of the bio
f679623f 4003 */
a167f663 4004 align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev);
f679623f
RBJ		 4005 if (!align_bi)
4006 return 0;
4007 /*
4008 * set bi_end_io to a new function, and set bi_private to the
4009 * original bio.
4010 */
4011 align_bi->bi_end_io = raid5_align_endio;
4012 align_bi->bi_private = raid_bio;
4013 /*
4014 * compute position
4015 */
112bf897
N		 4016 align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector,
4017 0,
911d4ee8 4018 &dd_idx, NULL);
f679623f 4019
f73a1c7d 4020 end_sector = bio_end_sector(align_bi);
f679623f 4021 rcu_read_lock();
671488cc
N		 4022 rdev = rcu_dereference(conf->disks[dd_idx].replacement);
4023 if (!rdev || test_bit(Faulty, &rdev->flags) ||
4024 rdev->recovery_offset < end_sector) {
4025 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
4026 if (rdev &&
4027 (test_bit(Faulty, &rdev->flags) ||
4028 !(test_bit(In_sync, &rdev->flags) ||
4029 rdev->recovery_offset >= end_sector)))
4030 rdev = NULL;
4031 }
4032 if (rdev) {
31c176ec
N		 4033 sector_t first_bad;
4034 int bad_sectors;
4035
f679623f
RBJ		 4036 atomic_inc(&rdev->nr_pending);
4037 rcu_read_unlock();
46031f9a
RBJ		 4038 raid_bio->bi_next = (void*)rdev;
4039 align_bi->bi_bdev = rdev->bdev;
4040 align_bi->bi_flags &= ~(1 << BIO_SEG_VALID);
46031f9a 4041
31c176ec 4042 if (!bio_fits_rdev(align_bi) ||
aa8b57aa 4043 is_badblock(rdev, align_bi->bi_sector, bio_sectors(align_bi),
31c176ec
N		 4044 &first_bad, &bad_sectors)) {
4045 /* too big in some way, or has a known bad block */
387bb173
NB		 4046 bio_put(align_bi);
4047 rdev_dec_pending(rdev, mddev);
4048 return 0;
4049 }
4050
6c0544e2 4051 /* No reshape active, so we can trust rdev->data_offset */
4052 align_bi->bi_sector += rdev->data_offset;
4053
46031f9a
RBJ		 4054 spin_lock_irq(&conf->device_lock);
4055 wait_event_lock_irq(conf->wait_for_stripe,
4056 conf->quiesce == 0,
eed8c02e 4057 conf->device_lock);
46031f9a
RBJ		 4058 atomic_inc(&conf->active_aligned_reads);
4059 spin_unlock_irq(&conf->device_lock);
4060
e3620a3a
JB		 4061 if (mddev->gendisk)
4062 trace_block_bio_remap(bdev_get_queue(align_bi->bi_bdev),
4063 align_bi, disk_devt(mddev->gendisk),
4064 raid_bio->bi_sector);
f679623f
RBJ		 4065 generic_make_request(align_bi);
4066 return 1;
4067 } else {
4068 rcu_read_unlock();
46031f9a 4069 bio_put(align_bi);
f679623f
RBJ		 4070 return 0;
4071 }
4072}
4073
8b3e6cdc
DW		 4074/* __get_priority_stripe - get the next stripe to process
4075 *
4076 * Full stripe writes are allowed to pass preread active stripes up until
4077 * the bypass_threshold is exceeded. In general the bypass_count
4078 * increments when the handle_list is handled before the hold_list; however, it
4079 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
4080 * stripe with in flight i/o. The bypass_count will be reset when the
4081 * head of the hold_list has changed, i.e. the head was promoted to the
4082 * handle_list.
4083 */
d1688a6d 4084static struct stripe_head *__get_priority_stripe(struct r5conf *conf)
8b3e6cdc
DW		 4085{
4086 struct stripe_head *sh;
4087
4088 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
4089 __func__,
4090 list_empty(&conf->handle_list) ? "empty" : "busy",
4091 list_empty(&conf->hold_list) ? "empty" : "busy",
4092 atomic_read(&conf->pending_full_writes), conf->bypass_count);
4093
4094 if (!list_empty(&conf->handle_list)) {
4095 sh = list_entry(conf->handle_list.next, typeof(*sh), lru);
4096
4097 if (list_empty(&conf->hold_list))
4098 conf->bypass_count = 0;
4099 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
4100 if (conf->hold_list.next == conf->last_hold)
4101 conf->bypass_count++;
4102 else {
4103 conf->last_hold = conf->hold_list.next;
4104 conf->bypass_count -= conf->bypass_threshold;
4105 if (conf->bypass_count < 0)
4106 conf->bypass_count = 0;
4107 }
4108 }
4109 } else if (!list_empty(&conf->hold_list) &&
4110 ((conf->bypass_threshold &&
4111 conf->bypass_count > conf->bypass_threshold) ||
4112 atomic_read(&conf->pending_full_writes) == 0)) {
4113 sh = list_entry(conf->hold_list.next,
4114 typeof(*sh), lru);
4115 conf->bypass_count -= conf->bypass_threshold;
4116 if (conf->bypass_count < 0)
4117 conf->bypass_count = 0;
4118 } else
4119 return NULL;
4120
4121 list_del_init(&sh->lru);
4122 atomic_inc(&sh->count);
4123 BUG_ON(atomic_read(&sh->count) != 1);
4124 return sh;
4125}
f679623f 4126
8811b596
SL		 4127struct raid5_plug_cb {
4128 struct blk_plug_cb cb;
4129 struct list_head list;
4130};
4131
4132static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule)
4133{
4134 struct raid5_plug_cb *cb = container_of(
4135 blk_cb, struct raid5_plug_cb, cb);
4136 struct stripe_head *sh;
4137 struct mddev *mddev = cb->cb.data;
4138 struct r5conf *conf = mddev->private;
a9add5d9 4139 int cnt = 0;
8811b596
SL		 4140
4141 if (cb->list.next && !list_empty(&cb->list)) {
4142 spin_lock_irq(&conf->device_lock);
4143 while (!list_empty(&cb->list)) {
4144 sh = list_first_entry(&cb->list, struct stripe_head, lru);
4145 list_del_init(&sh->lru);
4146 /*
4147 * avoid race release_stripe_plug() sees
4148 * STRIPE_ON_UNPLUG_LIST clear but the stripe
4149 * is still in our list
4150 */
4151 smp_mb__before_clear_bit();
4152 clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state);
4153 __release_stripe(conf, sh);
a9add5d9 4154 cnt++;
8811b596
SL		 4155 }
4156 spin_unlock_irq(&conf->device_lock);
4157 }
e3620a3a
JB		 4158 if (mddev->queue)
4159 trace_block_unplug(mddev->queue, cnt, !from_schedule);
8811b596
SL		 4160 kfree(cb);
4161}
4162
4163static void release_stripe_plug(struct mddev *mddev,
4164 struct stripe_head *sh)
4165{
4166 struct blk_plug_cb *blk_cb = blk_check_plugged(
4167 raid5_unplug, mddev,
4168 sizeof(struct raid5_plug_cb));
4169 struct raid5_plug_cb *cb;
4170
4171 if (!blk_cb) {
4172 release_stripe(sh);
4173 return;
4174 }
4175
4176 cb = container_of(blk_cb, struct raid5_plug_cb, cb);
4177
4178 if (cb->list.next == NULL)
4179 INIT_LIST_HEAD(&cb->list);
4180
4181 if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state))
4182 list_add_tail(&sh->lru, &cb->list);
4183 else
4184 release_stripe(sh);
4185}
4186
620125f2
SL		 4187static void make_discard_request(struct mddev *mddev, struct bio *bi)
4188{
4189 struct r5conf *conf = mddev->private;
4190 sector_t logical_sector, last_sector;
4191 struct stripe_head *sh;
4192 int remaining;
4193 int stripe_sectors;
4194
4195 if (mddev->reshape_position != MaxSector)
4196 /* Skip discard while reshape is happening */
4197 return;
4198
4199 logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
4200 last_sector = bi->bi_sector + (bi->bi_size>>9);
4201
4202 bi->bi_next = NULL;
4203 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
4204
4205 stripe_sectors = conf->chunk_sectors *
4206 (conf->raid_disks - conf->max_degraded);
4207 logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector,
4208 stripe_sectors);
4209 sector_div(last_sector, stripe_sectors);
4210
4211 logical_sector *= conf->chunk_sectors;
4212 last_sector *= conf->chunk_sectors;
4213
4214 for (; logical_sector < last_sector;
4215 logical_sector += STRIPE_SECTORS) {
4216 DEFINE_WAIT(w);
4217 int d;
4218 again:
4219 sh = get_active_stripe(conf, logical_sector, 0, 0, 0);
4220 prepare_to_wait(&conf->wait_for_overlap, &w,
4221 TASK_UNINTERRUPTIBLE);
f8dfcffd
N		 4222 set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
4223 if (test_bit(STRIPE_SYNCING, &sh->state)) {
4224 release_stripe(sh);
4225 schedule();
4226 goto again;
4227 }
4228 clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
620125f2
SL		 4229 spin_lock_irq(&sh->stripe_lock);
4230 for (d = 0; d < conf->raid_disks; d++) {
4231 if (d == sh->pd_idx || d == sh->qd_idx)
4232 continue;
4233 if (sh->dev[d].towrite || sh->dev[d].toread) {
4234 set_bit(R5_Overlap, &sh->dev[d].flags);
4235 spin_unlock_irq(&sh->stripe_lock);
4236 release_stripe(sh);
4237 schedule();
4238 goto again;
4239 }
4240 }
f8dfcffd 4241 set_bit(STRIPE_DISCARD, &sh->state);
620125f2
SL		 4242 finish_wait(&conf->wait_for_overlap, &w);
4243 for (d = 0; d < conf->raid_disks; d++) {
4244 if (d == sh->pd_idx || d == sh->qd_idx)
4245 continue;
4246 sh->dev[d].towrite = bi;
4247 set_bit(R5_OVERWRITE, &sh->dev[d].flags);
4248 raid5_inc_bi_active_stripes(bi);
4249 }
4250 spin_unlock_irq(&sh->stripe_lock);
4251 if (conf->mddev->bitmap) {
4252 for (d = 0;
4253 d < conf->raid_disks - conf->max_degraded;
4254 d++)
4255 bitmap_startwrite(mddev->bitmap,
4256 sh->sector,
4257 STRIPE_SECTORS,
4258 0);
4259 sh->bm_seq = conf->seq_flush + 1;
4260 set_bit(STRIPE_BIT_DELAY, &sh->state);
4261 }
4262
4263 set_bit(STRIPE_HANDLE, &sh->state);
4264 clear_bit(STRIPE_DELAYED, &sh->state);
4265 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4266 atomic_inc(&conf->preread_active_stripes);
4267 release_stripe_plug(mddev, sh);
4268 }
4269
4270 remaining = raid5_dec_bi_active_stripes(bi);
4271 if (remaining == 0) {
4272 md_write_end(mddev);
4273 bio_endio(bi, 0);
4274 }
4275}
4276
b4fdcb02 4277static void make_request(struct mddev *mddev, struct bio * bi)
1da177e4 4278{
d1688a6d 4279 struct r5conf *conf = mddev->private;
911d4ee8 4280 int dd_idx;
1da177e4
LT		 4281 sector_t new_sector;
4282 sector_t logical_sector, last_sector;
4283 struct stripe_head *sh;
a362357b 4284 const int rw = bio_data_dir(bi);
49077326 4285 int remaining;
1da177e4 4286
e9c7469b
TH		 4287 if (unlikely(bi->bi_rw & REQ_FLUSH)) {
4288 md_flush_request(mddev, bi);
5a7bbad2 4289 return;
e5dcdd80
N		 4290 }
4291
3d310eb7 4292 md_write_start(mddev, bi);
06d91a5f 4293
802ba064 4294 if (rw == READ &&
52488615 4295 mddev->reshape_position == MaxSector &&
21a52c6d 4296 chunk_aligned_read(mddev,bi))
5a7bbad2 4297 return;
52488615 4298
620125f2
SL		 4299 if (unlikely(bi->bi_rw & REQ_DISCARD)) {
4300 make_discard_request(mddev, bi);
4301 return;
4302 }
4303
1da177e4 4304 logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
f73a1c7d 4305 last_sector = bio_end_sector(bi);
1da177e4
LT		 4306 bi->bi_next = NULL;
4307 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
06d91a5f 4308
1da177e4
LT		 4309 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
4310 DEFINE_WAIT(w);
b5663ba4 4311 int previous;
b578d55f 4312
7ecaa1e6 4313 retry:
b5663ba4 4314 previous = 0;
b578d55f 4315 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
b0f9ec04 4316 if (unlikely(conf->reshape_progress != MaxSector)) {
fef9c61f 4317 /* spinlock is needed as reshape_progress may be
df8e7f76
N		 4318 * 64bit on a 32bit platform, and so it might be
4319 * possible to see a half-updated value
aeb878b0 4320 * Of course reshape_progress could change after
df8e7f76
N		 4321 * the lock is dropped, so once we get a reference
4322 * to the stripe that we think it is, we will have
4323 * to check again.
4324 */
7ecaa1e6 4325 spin_lock_irq(&conf->device_lock);
2c810cdd 4326 if (mddev->reshape_backwards
fef9c61f
N		 4327 ? logical_sector < conf->reshape_progress
4328 : logical_sector >= conf->reshape_progress) {
b5663ba4
N		 4329 previous = 1;
4330 } else {
2c810cdd 4331 if (mddev->reshape_backwards
fef9c61f
N		 4332 ? logical_sector < conf->reshape_safe
4333 : logical_sector >= conf->reshape_safe) {
b578d55f
N		 4334 spin_unlock_irq(&conf->device_lock);
4335 schedule();
4336 goto retry;
4337 }
4338 }
7ecaa1e6
N		 4339 spin_unlock_irq(&conf->device_lock);
4340 }
16a53ecc 4341
112bf897
N		 4342 new_sector = raid5_compute_sector(conf, logical_sector,
4343 previous,
911d4ee8 4344 &dd_idx, NULL);
0c55e022 4345 pr_debug("raid456: make_request, sector %llu logical %llu\n",
1da177e4
LT		 4346 (unsigned long long)new_sector,
4347 (unsigned long long)logical_sector);
4348
b5663ba4 4349 sh = get_active_stripe(conf, new_sector, previous,
a8c906ca 4350 (bi->bi_rw&RWA_MASK), 0);
1da177e4 4351 if (sh) {
b0f9ec04 4352 if (unlikely(previous)) {
7ecaa1e6 4353 /* expansion might have moved on while waiting for a
df8e7f76
N		 4354 * stripe, so we must do the range check again.
4355 * Expansion could still move past after this
4356 * test, but as we are holding a reference to
4357 * 'sh', we know that if that happens,
4358 * STRIPE_EXPANDING will get set and the expansion
4359 * won't proceed until we finish with the stripe.
7ecaa1e6
N		 4360 */
4361 int must_retry = 0;
4362 spin_lock_irq(&conf->device_lock);
2c810cdd 4363 if (mddev->reshape_backwards
b0f9ec04
N		 4364 ? logical_sector >= conf->reshape_progress
4365 : logical_sector < conf->reshape_progress)
7ecaa1e6
N		 4366 /* mismatch, need to try again */
4367 must_retry = 1;
4368 spin_unlock_irq(&conf->device_lock);
4369 if (must_retry) {
4370 release_stripe(sh);
7a3ab908 4371 schedule();
7ecaa1e6
N		 4372 goto retry;
4373 }
4374 }
e62e58a5 4375
ffd96e35 4376 if (rw == WRITE &&
a5c308d4 4377 logical_sector >= mddev->suspend_lo &&
e464eafd
N		 4378 logical_sector < mddev->suspend_hi) {
4379 release_stripe(sh);
e62e58a5
N		 4380 /* As the suspend_* range is controlled by
4381 * userspace, we want an interruptible
4382 * wait.
4383 */
4384 flush_signals(current);
4385 prepare_to_wait(&conf->wait_for_overlap,
4386 &w, TASK_INTERRUPTIBLE);
4387 if (logical_sector >= mddev->suspend_lo &&
4388 logical_sector < mddev->suspend_hi)
4389 schedule();
e464eafd
N		 4390 goto retry;
4391 }
7ecaa1e6
N		 4392
4393 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
ffd96e35 4394 !add_stripe_bio(sh, bi, dd_idx, rw)) {
7ecaa1e6
N		 4395 /* Stripe is busy expanding or
4396 * add failed due to overlap. Flush everything
1da177e4
LT		 4397 * and wait a while
4398 */
482c0834 4399 md_wakeup_thread(mddev->thread);
1da177e4
LT		 4400 release_stripe(sh);
4401 schedule();
4402 goto retry;
4403 }
4404 finish_wait(&conf->wait_for_overlap, &w);
6ed3003c
N		 4405 set_bit(STRIPE_HANDLE, &sh->state);
4406 clear_bit(STRIPE_DELAYED, &sh->state);
a852d7b8 4407 if ((bi->bi_rw & REQ_SYNC) &&
729a1866
N		 4408 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4409 atomic_inc(&conf->preread_active_stripes);
8811b596 4410 release_stripe_plug(mddev, sh);
1da177e4
LT		 4411 } else {
4412 /* cannot get stripe for read-ahead, just give-up */
4413 clear_bit(BIO_UPTODATE, &bi->bi_flags);
4414 finish_wait(&conf->wait_for_overlap, &w);
4415 break;
4416 }
1da177e4 4417 }
7c13edc8 4418
e7836bd6 4419 remaining = raid5_dec_bi_active_stripes(bi);
f6344757 4420 if (remaining == 0) {
1da177e4 4421
16a53ecc 4422 if ( rw == WRITE )
1da177e4 4423 md_write_end(mddev);
6712ecf8 4424
0a82a8d1
LT		 4425 trace_block_bio_complete(bdev_get_queue(bi->bi_bdev),
4426 bi, 0);
0e13fe23 4427 bio_endio(bi, 0);
1da177e4 4428 }
1da177e4
LT		 4429}
4430
fd01b88c 4431static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks);
b522adcd 4432
fd01b88c 4433static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped)
1da177e4 4434{
52c03291
N		 4435 /* reshaping is quite different to recovery/resync so it is
4436 * handled quite separately ... here.
4437 *
4438 * On each call to sync_request, we gather one chunk worth of
4439 * destination stripes and flag them as expanding.
4440 * Then we find all the source stripes and request reads.
4441 * As the reads complete, handle_stripe will copy the data
4442 * into the destination stripe and release that stripe.
4443 */
d1688a6d 4444 struct r5conf *conf = mddev->private;
1da177e4 4445 struct stripe_head *sh;
ccfcc3c1 4446 sector_t first_sector, last_sector;
f416885e
N		 4447 int raid_disks = conf->previous_raid_disks;
4448 int data_disks = raid_disks - conf->max_degraded;
4449 int new_data_disks = conf->raid_disks - conf->max_degraded;
52c03291
N		 4450 int i;
4451 int dd_idx;
c8f517c4 4452 sector_t writepos, readpos, safepos;
ec32a2bd 4453 sector_t stripe_addr;
7a661381 4454 int reshape_sectors;
ab69ae12 4455 struct list_head stripes;
52c03291 4456
fef9c61f
N		 4457 if (sector_nr == 0) {
4458 /* If restarting in the middle, skip the initial sectors */
2c810cdd 4459 if (mddev->reshape_backwards &&
fef9c61f
N		 4460 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
4461 sector_nr = raid5_size(mddev, 0, 0)
4462 - conf->reshape_progress;
2c810cdd 4463 } else if (!mddev->reshape_backwards &&
fef9c61f
N		 4464 conf->reshape_progress > 0)
4465 sector_nr = conf->reshape_progress;
f416885e 4466 sector_div(sector_nr, new_data_disks);
fef9c61f 4467 if (sector_nr) {
8dee7211
N		 4468 mddev->curr_resync_completed = sector_nr;
4469 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
fef9c61f
N		 4470 *skipped = 1;
4471 return sector_nr;
4472 }
52c03291
N		 4473 }
4474
7a661381
N		 4475 /* We need to process a full chunk at a time.
4476 * If old and new chunk sizes differ, we need to process the
4477 * largest of these
4478 */
664e7c41
AN		 4479 if (mddev->new_chunk_sectors > mddev->chunk_sectors)
4480 reshape_sectors = mddev->new_chunk_sectors;
7a661381 4481 else
9d8f0363 4482 reshape_sectors = mddev->chunk_sectors;
7a661381 4483
b5254dd5
N		 4484 /* We update the metadata at least every 10 seconds, or when
4485 * the data about to be copied would over-write the source of
4486 * the data at the front of the range. i.e. one new_stripe
4487 * along from reshape_progress new_maps to after where
4488 * reshape_safe old_maps to
52c03291 4489 */
fef9c61f 4490 writepos = conf->reshape_progress;
f416885e 4491 sector_div(writepos, new_data_disks);
c8f517c4
N		 4492 readpos = conf->reshape_progress;
4493 sector_div(readpos, data_disks);
fef9c61f 4494 safepos = conf->reshape_safe;
f416885e 4495 sector_div(safepos, data_disks);
2c810cdd 4496 if (mddev->reshape_backwards) {
ed37d83e 4497 writepos -= min_t(sector_t, reshape_sectors, writepos);
c8f517c4 4498 readpos += reshape_sectors;
7a661381 4499 safepos += reshape_sectors;
fef9c61f 4500 } else {
7a661381 4501 writepos += reshape_sectors;
ed37d83e
N		 4502 readpos -= min_t(sector_t, reshape_sectors, readpos);
4503 safepos -= min_t(sector_t, reshape_sectors, safepos);
fef9c61f 4504 }
52c03291 4505
b5254dd5
N		 4506 /* Having calculated the 'writepos' possibly use it
4507 * to set 'stripe_addr' which is where we will write to.
4508 */
4509 if (mddev->reshape_backwards) {
4510 BUG_ON(conf->reshape_progress == 0);
4511 stripe_addr = writepos;
4512 BUG_ON((mddev->dev_sectors &
4513 ~((sector_t)reshape_sectors - 1))
4514 - reshape_sectors - stripe_addr
4515 != sector_nr);
4516 } else {
4517 BUG_ON(writepos != sector_nr + reshape_sectors);
4518 stripe_addr = sector_nr;
4519 }
4520
c8f517c4
N		 4521 /* 'writepos' is the most advanced device address we might write.
4522 * 'readpos' is the least advanced device address we might read.
4523 * 'safepos' is the least address recorded in the metadata as having
4524 * been reshaped.
b5254dd5
N		 4525 * If there is a min_offset_diff, these are adjusted either by
4526 * increasing the safepos/readpos if diff is negative, or
4527 * increasing writepos if diff is positive.
4528 * If 'readpos' is then behind 'writepos', there is no way that we can
c8f517c4
N		 4529 * ensure safety in the face of a crash - that must be done by userspace
4530 * making a backup of the data. So in that case there is no particular
4531 * rush to update metadata.
4532 * Otherwise if 'safepos' is behind 'writepos', then we really need to
4533 * update the metadata to advance 'safepos' to match 'readpos' so that
4534 * we can be safe in the event of a crash.
4535 * So we insist on updating metadata if safepos is behind writepos and
4536 * readpos is beyond writepos.
4537 * In any case, update the metadata every 10 seconds.
4538 * Maybe that number should be configurable, but I'm not sure it is
4539 * worth it.... maybe it could be a multiple of safemode_delay???
4540 */
b5254dd5
N		 4541 if (conf->min_offset_diff < 0) {
4542 safepos += -conf->min_offset_diff;
4543 readpos += -conf->min_offset_diff;
4544 } else
4545 writepos += conf->min_offset_diff;
4546
2c810cdd 4547 if ((mddev->reshape_backwards
c8f517c4
N		 4548 ? (safepos > writepos && readpos < writepos)
4549 : (safepos < writepos && readpos > writepos)) ||
4550 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
52c03291
N		 4551 /* Cannot proceed until we've updated the superblock... */
4552 wait_event(conf->wait_for_overlap,
4553 atomic_read(&conf->reshape_stripes)==0);
fef9c61f 4554 mddev->reshape_position = conf->reshape_progress;
75d3da43 4555 mddev->curr_resync_completed = sector_nr;
c8f517c4 4556 conf->reshape_checkpoint = jiffies;
850b2b42 4557 set_bit(MD_CHANGE_DEVS, &mddev->flags);
52c03291 4558 md_wakeup_thread(mddev->thread);
850b2b42 4559 wait_event(mddev->sb_wait, mddev->flags == 0 ||
52c03291
N		 4560 kthread_should_stop());
4561 spin_lock_irq(&conf->device_lock);
fef9c61f 4562 conf->reshape_safe = mddev->reshape_position;
52c03291
N		 4563 spin_unlock_irq(&conf->device_lock);
4564 wake_up(&conf->wait_for_overlap);
acb180b0 4565 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
52c03291
N		 4566 }
4567
ab69ae12 4568 INIT_LIST_HEAD(&stripes);
7a661381 4569 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
52c03291 4570 int j;
a9f326eb 4571 int skipped_disk = 0;
a8c906ca 4572 sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
52c03291
N		 4573 set_bit(STRIPE_EXPANDING, &sh->state);
4574 atomic_inc(&conf->reshape_stripes);
4575 /* If any of this stripe is beyond the end of the old
4576 * array, then we need to zero those blocks
4577 */
4578 for (j=sh->disks; j--;) {
4579 sector_t s;
4580 if (j == sh->pd_idx)
4581 continue;
f416885e 4582 if (conf->level == 6 &&
d0dabf7e 4583 j == sh->qd_idx)
f416885e 4584 continue;
784052ec 4585 s = compute_blocknr(sh, j, 0);
b522adcd 4586 if (s < raid5_size(mddev, 0, 0)) {
a9f326eb 4587 skipped_disk = 1;
52c03291
N		 4588 continue;
4589 }
4590 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
4591 set_bit(R5_Expanded, &sh->dev[j].flags);
4592 set_bit(R5_UPTODATE, &sh->dev[j].flags);
4593 }
a9f326eb 4594 if (!skipped_disk) {
52c03291
N		 4595 set_bit(STRIPE_EXPAND_READY, &sh->state);
4596 set_bit(STRIPE_HANDLE, &sh->state);
4597 }
ab69ae12 4598 list_add(&sh->lru, &stripes);
52c03291
N		 4599 }
4600 spin_lock_irq(&conf->device_lock);
2c810cdd 4601 if (mddev->reshape_backwards)
7a661381 4602 conf->reshape_progress -= reshape_sectors * new_data_disks;
fef9c61f 4603 else
7a661381 4604 conf->reshape_progress += reshape_sectors * new_data_disks;
52c03291
N		 4605 spin_unlock_irq(&conf->device_lock);
4606 /* Ok, those stripe are ready. We can start scheduling
4607 * reads on the source stripes.
4608 * The source stripes are determined by mapping the first and last
4609 * block on the destination stripes.
4610 */
52c03291 4611 first_sector =
ec32a2bd 4612 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
911d4ee8 4613 1, &dd_idx, NULL);
52c03291 4614 last_sector =
0e6e0271 4615 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
09c9e5fa 4616 * new_data_disks - 1),
911d4ee8 4617 1, &dd_idx, NULL);
58c0fed4
AN		 4618 if (last_sector >= mddev->dev_sectors)
4619 last_sector = mddev->dev_sectors - 1;
52c03291 4620 while (first_sector <= last_sector) {
a8c906ca 4621 sh = get_active_stripe(conf, first_sector, 1, 0, 1);
52c03291
N		 4622 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
4623 set_bit(STRIPE_HANDLE, &sh->state);
4624 release_stripe(sh);
4625 first_sector += STRIPE_SECTORS;
4626 }
ab69ae12
N		 4627 /* Now that the sources are clearly marked, we can release
4628 * the destination stripes
4629 */
4630 while (!list_empty(&stripes)) {
4631 sh = list_entry(stripes.next, struct stripe_head, lru);
4632 list_del_init(&sh->lru);
4633 release_stripe(sh);
4634 }
c6207277
N		 4635 /* If this takes us to the resync_max point where we have to pause,
4636 * then we need to write out the superblock.
4637 */
7a661381 4638 sector_nr += reshape_sectors;
c03f6a19
N		 4639 if ((sector_nr - mddev->curr_resync_completed) * 2
4640 >= mddev->resync_max - mddev->curr_resync_completed) {
c6207277
N		 4641 /* Cannot proceed until we've updated the superblock... */
4642 wait_event(conf->wait_for_overlap,
4643 atomic_read(&conf->reshape_stripes) == 0);
fef9c61f 4644 mddev->reshape_position = conf->reshape_progress;
75d3da43 4645 mddev->curr_resync_completed = sector_nr;
c8f517c4 4646 conf->reshape_checkpoint = jiffies;
c6207277
N		 4647 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4648 md_wakeup_thread(mddev->thread);
4649 wait_event(mddev->sb_wait,
4650 !test_bit(MD_CHANGE_DEVS, &mddev->flags)
4651 || kthread_should_stop());
4652 spin_lock_irq(&conf->device_lock);
fef9c61f 4653 conf->reshape_safe = mddev->reshape_position;
c6207277
N		 4654 spin_unlock_irq(&conf->device_lock);
4655 wake_up(&conf->wait_for_overlap);
acb180b0 4656 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
c6207277 4657 }
7a661381 4658 return reshape_sectors;
52c03291
N		 4659}
4660
4661/* FIXME go_faster isn't used */
fd01b88c 4662static inline sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster)
52c03291 4663{
d1688a6d 4664 struct r5conf *conf = mddev->private;
52c03291 4665 struct stripe_head *sh;
58c0fed4 4666 sector_t max_sector = mddev->dev_sectors;
57dab0bd 4667 sector_t sync_blocks;
16a53ecc
N		 4668 int still_degraded = 0;
4669 int i;
1da177e4 4670
72626685 4671 if (sector_nr >= max_sector) {
1da177e4 4672 /* just being told to finish up .. nothing much to do */
cea9c228 4673
29269553
N		 4674 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
4675 end_reshape(conf);
4676 return 0;
4677 }
72626685
N		 4678
4679 if (mddev->curr_resync < max_sector) /* aborted */
4680 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
4681 &sync_blocks, 1);
16a53ecc 4682 else /* completed sync */
72626685
N		 4683 conf->fullsync = 0;
4684 bitmap_close_sync(mddev->bitmap);
4685
1da177e4
LT		 4686 return 0;
4687 }
ccfcc3c1 4688
64bd660b
N		 4689 /* Allow raid5_quiesce to complete */
4690 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
4691
52c03291
N		 4692 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4693 return reshape_request(mddev, sector_nr, skipped);
f6705578 4694
c6207277
N		 4695 /* No need to check resync_max as we never do more than one
4696 * stripe, and as resync_max will always be on a chunk boundary,
4697 * if the check in md_do_sync didn't fire, there is no chance
4698 * of overstepping resync_max here
4699 */
4700
16a53ecc 4701 /* if there is too many failed drives and we are trying
1da177e4
LT		 4702 * to resync, then assert that we are finished, because there is
4703 * nothing we can do.
4704 */
3285edf1 4705 if (mddev->degraded >= conf->max_degraded &&
16a53ecc 4706 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
58c0fed4 4707 sector_t rv = mddev->dev_sectors - sector_nr;
57afd89f 4708 *skipped = 1;
1da177e4
LT		 4709 return rv;
4710 }
6f608040 4711 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
4712 !conf->fullsync &&
4713 !bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
4714 sync_blocks >= STRIPE_SECTORS) {
72626685
N		 4715 /* we can skip this block, and probably more */
4716 sync_blocks /= STRIPE_SECTORS;
4717 *skipped = 1;
4718 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
4719 }
1da177e4 4720
b47490c9
N		 4721 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
4722
a8c906ca 4723 sh = get_active_stripe(conf, sector_nr, 0, 1, 0);
1da177e4 4724 if (sh == NULL) {
a8c906ca 4725 sh = get_active_stripe(conf, sector_nr, 0, 0, 0);
1da177e4 4726 /* make sure we don't swamp the stripe cache if someone else
16a53ecc 4727 * is trying to get access
1da177e4 4728 */
66c006a5 4729 schedule_timeout_uninterruptible(1);
1da177e4 4730 }
16a53ecc
N		 4731 /* Need to check if array will still be degraded after recovery/resync
4732 * We don't need to check the 'failed' flag as when that gets set,
4733 * recovery aborts.
4734 */
f001a70c 4735 for (i = 0; i < conf->raid_disks; i++)
16a53ecc
N		 4736 if (conf->disks[i].rdev == NULL)
4737 still_degraded = 1;
4738
4739 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
4740
83206d66 4741 set_bit(STRIPE_SYNC_REQUESTED, &sh->state);
1da177e4 4742
1442577b 4743 handle_stripe(sh);
1da177e4
LT		 4744 release_stripe(sh);
4745
4746 return STRIPE_SECTORS;
4747}
4748
d1688a6d 4749static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio)
46031f9a
RBJ		 4750{
4751 /* We may not be able to submit a whole bio at once as there
4752 * may not be enough stripe_heads available.
4753 * We cannot pre-allocate enough stripe_heads as we may need
4754 * more than exist in the cache (if we allow ever large chunks).
4755 * So we do one stripe head at a time and record in
4756 * ->bi_hw_segments how many have been done.
4757 *
4758 * We *know* that this entire raid_bio is in one chunk, so
4759 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
4760 */
4761 struct stripe_head *sh;
911d4ee8 4762 int dd_idx;
46031f9a
RBJ		 4763 sector_t sector, logical_sector, last_sector;
4764 int scnt = 0;
4765 int remaining;
4766 int handled = 0;
4767
4768 logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
112bf897 4769 sector = raid5_compute_sector(conf, logical_sector,
911d4ee8 4770 0, &dd_idx, NULL);
f73a1c7d 4771 last_sector = bio_end_sector(raid_bio);
46031f9a
RBJ		 4772
4773 for (; logical_sector < last_sector;
387bb173
NB		 4774 logical_sector += STRIPE_SECTORS,
4775 sector += STRIPE_SECTORS,
4776 scnt++) {
46031f9a 4777
e7836bd6 4778 if (scnt < raid5_bi_processed_stripes(raid_bio))
46031f9a
RBJ		 4779 /* already done this stripe */
4780 continue;
4781
a8c906ca 4782 sh = get_active_stripe(conf, sector, 0, 1, 0);
46031f9a
RBJ		 4783
4784 if (!sh) {
4785 /* failed to get a stripe - must wait */
e7836bd6 4786 raid5_set_bi_processed_stripes(raid_bio, scnt);
46031f9a
RBJ		 4787 conf->retry_read_aligned = raid_bio;
4788 return handled;
4789 }
4790
387bb173
NB		 4791 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) {
4792 release_stripe(sh);
e7836bd6 4793 raid5_set_bi_processed_stripes(raid_bio, scnt);
387bb173
NB		 4794 conf->retry_read_aligned = raid_bio;
4795 return handled;
4796 }
4797
3f9e7c14 4798 set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags);
36d1c647 4799 handle_stripe(sh);
46031f9a
RBJ		 4800 release_stripe(sh);
4801 handled++;
4802 }
e7836bd6 4803 remaining = raid5_dec_bi_active_stripes(raid_bio);
0a82a8d1
LT		 4804 if (remaining == 0) {
4805 trace_block_bio_complete(bdev_get_queue(raid_bio->bi_bdev),
4806 raid_bio, 0);
0e13fe23 4807 bio_endio(raid_bio, 0);
0a82a8d1 4808 }
46031f9a
RBJ		 4809 if (atomic_dec_and_test(&conf->active_aligned_reads))
4810 wake_up(&conf->wait_for_stripe);
4811 return handled;
4812}
4813
46a06401
SL		 4814#define MAX_STRIPE_BATCH 8
4815static int handle_active_stripes(struct r5conf *conf)
4816{
4817 struct stripe_head *batch[MAX_STRIPE_BATCH], *sh;
4818 int i, batch_size = 0;
4819
4820 while (batch_size < MAX_STRIPE_BATCH &&
4821 (sh = __get_priority_stripe(conf)) != NULL)
4822 batch[batch_size++] = sh;
4823
4824 if (batch_size == 0)
4825 return batch_size;
4826 spin_unlock_irq(&conf->device_lock);
4827
4828 for (i = 0; i < batch_size; i++)
4829 handle_stripe(batch[i]);
4830
4831 cond_resched();
4832
4833 spin_lock_irq(&conf->device_lock);
4834 for (i = 0; i < batch_size; i++)
4835 __release_stripe(conf, batch[i]);
4836 return batch_size;
4837}
46031f9a 4838
1da177e4
LT		 4839/*
4840 * This is our raid5 kernel thread.
4841 *
4842 * We scan the hash table for stripes which can be handled now.
4843 * During the scan, completed stripes are saved for us by the interrupt
4844 * handler, so that they will not have to wait for our next wakeup.
4845 */
4ed8731d 4846static void raid5d(struct md_thread *thread)
1da177e4 4847{
4ed8731d 4848 struct mddev *mddev = thread->mddev;
d1688a6d 4849 struct r5conf *conf = mddev->private;
1da177e4 4850 int handled;
e1dfa0a2 4851 struct blk_plug plug;
1da177e4 4852
45b4233c 4853 pr_debug("+++ raid5d active\n");
1da177e4
LT		 4854
4855 md_check_recovery(mddev);
1da177e4 4856
e1dfa0a2 4857 blk_start_plug(&plug);
1da177e4
LT		 4858 handled = 0;
4859 spin_lock_irq(&conf->device_lock);
4860 while (1) {
46031f9a 4861 struct bio *bio;
46a06401 4862 int batch_size;
1da177e4 4863
0021b7bc 4864 if (
7c13edc8
N		 4865 !list_empty(&conf->bitmap_list)) {
4866 /* Now is a good time to flush some bitmap updates */
4867 conf->seq_flush++;
700e432d 4868 spin_unlock_irq(&conf->device_lock);
72626685 4869 bitmap_unplug(mddev->bitmap);
700e432d 4870 spin_lock_irq(&conf->device_lock);
7c13edc8 4871 conf->seq_write = conf->seq_flush;
72626685
N		 4872 activate_bit_delay(conf);
4873 }
0021b7bc 4874 raid5_activate_delayed(conf);
72626685 4875
46031f9a
RBJ		 4876 while ((bio = remove_bio_from_retry(conf))) {
4877 int ok;
4878 spin_unlock_irq(&conf->device_lock);
4879 ok = retry_aligned_read(conf, bio);
4880 spin_lock_irq(&conf->device_lock);
4881 if (!ok)
4882 break;
4883 handled++;
4884 }
4885
46a06401
SL		 4886 batch_size = handle_active_stripes(conf);
4887 if (!batch_size)
1da177e4 4888 break;
46a06401 4889 handled += batch_size;
1da177e4 4890
46a06401
SL		 4891 if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) {
4892 spin_unlock_irq(&conf->device_lock);
de393cde 4893 md_check_recovery(mddev);
46a06401
SL		 4894 spin_lock_irq(&conf->device_lock);
4895 }
1da177e4 4896 }
45b4233c 4897 pr_debug("%d stripes handled\n", handled);
1da177e4
LT		 4898
4899 spin_unlock_irq(&conf->device_lock);
4900
c9f21aaf 4901 async_tx_issue_pending_all();
e1dfa0a2 4902 blk_finish_plug(&plug);
1da177e4 4903
45b4233c 4904 pr_debug("--- raid5d inactive\n");
1da177e4
LT		 4905}
4906
3f294f4f 4907static ssize_t
fd01b88c 4908raid5_show_stripe_cache_size(struct mddev *mddev, char *page)
3f294f4f 4909{
d1688a6d 4910 struct r5conf *conf = mddev->private;
96de1e66
N		 4911 if (conf)
4912 return sprintf(page, "%d\n", conf->max_nr_stripes);
4913 else
4914 return 0;
3f294f4f
N		 4915}
4916
c41d4ac4 4917int
fd01b88c 4918raid5_set_cache_size(struct mddev *mddev, int size)
3f294f4f 4919{
d1688a6d 4920 struct r5conf *conf = mddev->private;
b5470dc5
DW		 4921 int err;
4922
c41d4ac4 4923 if (size <= 16 || size > 32768)
3f294f4f 4924 return -EINVAL;
c41d4ac4 4925 while (size < conf->max_nr_stripes) {
3f294f4f
N		 4926 if (drop_one_stripe(conf))
4927 conf->max_nr_stripes--;
4928 else
4929 break;
4930 }
b5470dc5
DW		 4931 err = md_allow_write(mddev);
4932 if (err)
4933 return err;
c41d4ac4 4934 while (size > conf->max_nr_stripes) {
3f294f4f
N		 4935 if (grow_one_stripe(conf))
4936 conf->max_nr_stripes++;
4937 else break;
4938 }
c41d4ac4
N		 4939 return 0;
4940}
4941EXPORT_SYMBOL(raid5_set_cache_size);
4942
4943static ssize_t
fd01b88c 4944raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len)
c41d4ac4 4945{
d1688a6d 4946 struct r5conf *conf = mddev->private;
c41d4ac4
N		 4947 unsigned long new;
4948 int err;
4949
4950 if (len >= PAGE_SIZE)
4951 return -EINVAL;
4952 if (!conf)
4953 return -ENODEV;
4954
4955 if (strict_strtoul(page, 10, &new))
4956 return -EINVAL;
4957 err = raid5_set_cache_size(mddev, new);
4958 if (err)
4959 return err;
3f294f4f
N		 4960 return len;
4961}
007583c9 4962
96de1e66
N		 4963static struct md_sysfs_entry
4964raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
4965 raid5_show_stripe_cache_size,
4966 raid5_store_stripe_cache_size);
3f294f4f 4967
8b3e6cdc 4968static ssize_t
fd01b88c 4969raid5_show_preread_threshold(struct mddev *mddev, char *page)
8b3e6cdc 4970{
d1688a6d 4971 struct r5conf *conf = mddev->private;
8b3e6cdc
DW		 4972 if (conf)
4973 return sprintf(page, "%d\n", conf->bypass_threshold);
4974 else
4975 return 0;
4976}
4977
4978static ssize_t
fd01b88c 4979raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len)
8b3e6cdc 4980{
d1688a6d 4981 struct r5conf *conf = mddev->private;
4ef197d8 4982 unsigned long new;
8b3e6cdc
DW		 4983 if (len >= PAGE_SIZE)
4984 return -EINVAL;
4985 if (!conf)
4986 return -ENODEV;
4987
4ef197d8 4988 if (strict_strtoul(page, 10, &new))
8b3e6cdc 4989 return -EINVAL;
4ef197d8 4990 if (new > conf->max_nr_stripes)
8b3e6cdc
DW		 4991 return -EINVAL;
4992 conf->bypass_threshold = new;
4993 return len;
4994}
4995
4996static struct md_sysfs_entry
4997raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
4998 S_IRUGO | S_IWUSR,
4999 raid5_show_preread_threshold,
5000 raid5_store_preread_threshold);
5001
3f294f4f 5002static ssize_t
fd01b88c 5003stripe_cache_active_show(struct mddev *mddev, char *page)
3f294f4f 5004{
d1688a6d 5005 struct r5conf *conf = mddev->private;
96de1e66
N		 5006 if (conf)
5007 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
5008 else
5009 return 0;
3f294f4f
N		 5010}
5011
96de1e66
N		 5012static struct md_sysfs_entry
5013raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
3f294f4f 5014
007583c9 5015static struct attribute *raid5_attrs[] = {
3f294f4f
N		 5016 &raid5_stripecache_size.attr,
5017 &raid5_stripecache_active.attr,
8b3e6cdc 5018 &raid5_preread_bypass_threshold.attr,
3f294f4f
N		 5019 NULL,
5020};
007583c9
N		 5021static struct attribute_group raid5_attrs_group = {
5022 .name = NULL,
5023 .attrs = raid5_attrs,
3f294f4f
N		 5024};
5025
80c3a6ce 5026static sector_t
fd01b88c 5027raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks)
80c3a6ce 5028{
d1688a6d 5029 struct r5conf *conf = mddev->private;
80c3a6ce
DW		 5030
5031 if (!sectors)
5032 sectors = mddev->dev_sectors;
5e5e3e78 5033 if (!raid_disks)
7ec05478 5034 /* size is defined by the smallest of previous and new size */
5e5e3e78 5035 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
80c3a6ce 5036
9d8f0363 5037 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
664e7c41 5038 sectors &= ~((sector_t)mddev->new_chunk_sectors - 1);
80c3a6ce
DW		 5039 return sectors * (raid_disks - conf->max_degraded);
5040}
5041
4d4ef86d
ON		 5042static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
5043{
5044 safe_put_page(percpu->spare_page);
5045 kfree(percpu->scribble);
5046 percpu->spare_page = NULL;
5047 percpu->scribble = NULL;
5048}
5049
5050static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
5051{
5052 if (conf->level == 6 && !percpu->spare_page)
5053 percpu->spare_page = alloc_page(GFP_KERNEL);
5054 if (!percpu->scribble)
5055 percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
5056
5057 if (!percpu->scribble || (conf->level == 6 && !percpu->spare_page)) {
5058 free_scratch_buffer(conf, percpu);
5059 return -ENOMEM;
5060 }
5061
5062 return 0;
5063}
5064
d1688a6d 5065static void raid5_free_percpu(struct r5conf *conf)
36d1c647 5066{
36d1c647
DW		 5067 unsigned long cpu;
5068
5069 if (!conf->percpu)
5070 return;
5071
36d1c647
DW		 5072#ifdef CONFIG_HOTPLUG_CPU
5073 unregister_cpu_notifier(&conf->cpu_notify);
5074#endif
4d4ef86d
ON		 5075
5076 get_online_cpus();
5077 for_each_possible_cpu(cpu)
5078 free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
36d1c647
DW		 5079 put_online_cpus();
5080
5081 free_percpu(conf->percpu);
5082}
5083
d1688a6d 5084static void free_conf(struct r5conf *conf)
95fc17aa
DW		 5085{
5086 shrink_stripes(conf);
36d1c647 5087 raid5_free_percpu(conf);
95fc17aa
DW		 5088 kfree(conf->disks);
5089 kfree(conf->stripe_hashtbl);
5090 kfree(conf);
5091}
5092
36d1c647
DW		 5093#ifdef CONFIG_HOTPLUG_CPU
5094static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action,
5095 void *hcpu)
5096{
d1688a6d 5097 struct r5conf *conf = container_of(nfb, struct r5conf, cpu_notify);
36d1c647
DW		 5098 long cpu = (long)hcpu;
5099 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
5100
5101 switch (action) {
5102 case CPU_UP_PREPARE:
5103 case CPU_UP_PREPARE_FROZEN:
4d4ef86d 5104 if (alloc_scratch_buffer(conf, percpu)) {
36d1c647
DW		 5105 pr_err("%s: failed memory allocation for cpu%ld\n",
5106 __func__, cpu);
55af6bb5 5107 return notifier_from_errno(-ENOMEM);
36d1c647
DW		 5108 }
5109 break;
5110 case CPU_DEAD:
5111 case CPU_DEAD_FROZEN:
4d4ef86d 5112 free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
36d1c647
DW		 5113 break;
5114 default:
5115 break;
5116 }
5117 return NOTIFY_OK;
5118}
5119#endif
5120
d1688a6d 5121static int raid5_alloc_percpu(struct r5conf *conf)
36d1c647
DW		 5122{
5123 unsigned long cpu;
4d4ef86d 5124 int err = 0;
36d1c647 5125
4d4ef86d
ON		 5126 conf->percpu = alloc_percpu(struct raid5_percpu);
5127 if (!conf->percpu)
36d1c647 5128 return -ENOMEM;
4d4ef86d
ON		 5129
5130#ifdef CONFIG_HOTPLUG_CPU
5131 conf->cpu_notify.notifier_call = raid456_cpu_notify;
5132 conf->cpu_notify.priority = 0;
5133 err = register_cpu_notifier(&conf->cpu_notify);
5134 if (err)
5135 return err;
5136#endif
36d1c647
DW		 5137
5138 get_online_cpus();
36d1c647 5139 for_each_present_cpu(cpu) {
4d4ef86d
ON		 5140 err = alloc_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
5141 if (err) {
5142 pr_err("%s: failed memory allocation for cpu%ld\n",
5143 __func__, cpu);
36d1c647
DW		 5144 break;
5145 }
36d1c647 5146 }
36d1c647
DW		 5147 put_online_cpus();
5148
5149 return err;
5150}
5151
d1688a6d 5152static struct r5conf *setup_conf(struct mddev *mddev)
1da177e4 5153{
d1688a6d 5154 struct r5conf *conf;
5e5e3e78 5155 int raid_disk, memory, max_disks;
3cb03002 5156 struct md_rdev *rdev;
1da177e4 5157 struct disk_info *disk;
0232605d 5158 char pers_name[6];
1da177e4 5159
91adb564
N		 5160 if (mddev->new_level != 5
5161 && mddev->new_level != 4
5162 && mddev->new_level != 6) {
0c55e022 5163 printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n",
91adb564
N		 5164 mdname(mddev), mddev->new_level);
5165 return ERR_PTR(-EIO);
1da177e4 5166 }
91adb564
N		 5167 if ((mddev->new_level == 5
5168 && !algorithm_valid_raid5(mddev->new_layout)) ||
5169 (mddev->new_level == 6
5170 && !algorithm_valid_raid6(mddev->new_layout))) {
0c55e022 5171 printk(KERN_ERR "md/raid:%s: layout %d not supported\n",
91adb564
N		 5172 mdname(mddev), mddev->new_layout);
5173 return ERR_PTR(-EIO);
99c0fb5f 5174 }
91adb564 5175 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
0c55e022 5176 printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n",
91adb564
N		 5177 mdname(mddev), mddev->raid_disks);
5178 return ERR_PTR(-EINVAL);
4bbf3771
N		 5179 }
5180
664e7c41
AN		 5181 if (!mddev->new_chunk_sectors ||
5182 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
5183 !is_power_of_2(mddev->new_chunk_sectors)) {
0c55e022
N		 5184 printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n",
5185 mdname(mddev), mddev->new_chunk_sectors << 9);
91adb564 5186 return ERR_PTR(-EINVAL);
f6705578
N		 5187 }
5188
d1688a6d 5189 conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL);
91adb564 5190 if (conf == NULL)
1da177e4 5191 goto abort;
f5efd45a
DW		 5192 spin_lock_init(&conf->device_lock);
5193 init_waitqueue_head(&conf->wait_for_stripe);
5194 init_waitqueue_head(&conf->wait_for_overlap);
5195 INIT_LIST_HEAD(&conf->handle_list);
5196 INIT_LIST_HEAD(&conf->hold_list);
5197 INIT_LIST_HEAD(&conf->delayed_list);
5198 INIT_LIST_HEAD(&conf->bitmap_list);
5199 INIT_LIST_HEAD(&conf->inactive_list);
5200 atomic_set(&conf->active_stripes, 0);
5201 atomic_set(&conf->preread_active_stripes, 0);
5202 atomic_set(&conf->active_aligned_reads, 0);
5203 conf->bypass_threshold = BYPASS_THRESHOLD;
d890fa2b 5204 conf->recovery_disabled = mddev->recovery_disabled - 1;
91adb564
N		 5205
5206 conf->raid_disks = mddev->raid_disks;
5207 if (mddev->reshape_position == MaxSector)
5208 conf->previous_raid_disks = mddev->raid_disks;
5209 else
f6705578 5210 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
5e5e3e78
N		 5211 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
5212 conf->scribble_len = scribble_len(max_disks);
f6705578 5213
5e5e3e78 5214 conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
b55e6bfc
N		 5215 GFP_KERNEL);
5216 if (!conf->disks)
5217 goto abort;
9ffae0cf 5218
1da177e4
LT		 5219 conf->mddev = mddev;
5220
fccddba0 5221 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
1da177e4 5222 goto abort;
1da177e4 5223
36d1c647
DW		 5224 conf->level = mddev->new_level;
5225 if (raid5_alloc_percpu(conf) != 0)
5226 goto abort;
5227
0c55e022 5228 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
1da177e4 5229
dafb20fa 5230 rdev_for_each(rdev, mddev) {
1da177e4 5231 raid_disk = rdev->raid_disk;
5e5e3e78 5232 if (raid_disk >= max_disks
1da177e4
LT		 5233 || raid_disk < 0)
5234 continue;
5235 disk = conf->disks + raid_disk;
5236
17045f52
N		 5237 if (test_bit(Replacement, &rdev->flags)) {
5238 if (disk->replacement)
5239 goto abort;
5240 disk->replacement = rdev;
5241 } else {
5242 if (disk->rdev)
5243 goto abort;
5244 disk->rdev = rdev;
5245 }
1da177e4 5246
b2d444d7 5247 if (test_bit(In_sync, &rdev->flags)) {
1da177e4 5248 char b[BDEVNAME_SIZE];
0c55e022
N		 5249 printk(KERN_INFO "md/raid:%s: device %s operational as raid"
5250 " disk %d\n",
5251 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
d6b212f4 5252 } else if (rdev->saved_raid_disk != raid_disk)
8c2e870a
NB		 5253 /* Cannot rely on bitmap to complete recovery */
5254 conf->fullsync = 1;
1da177e4
LT		 5255 }
5256
09c9e5fa 5257 conf->chunk_sectors = mddev->new_chunk_sectors;
91adb564 5258 conf->level = mddev->new_level;
16a53ecc
N		 5259 if (conf->level == 6)
5260 conf->max_degraded = 2;
5261 else
5262 conf->max_degraded = 1;
91adb564 5263 conf->algorithm = mddev->new_layout;
1da177e4 5264 conf->max_nr_stripes = NR_STRIPES;
fef9c61f 5265 conf->reshape_progress = mddev->reshape_position;
e183eaed 5266 if (conf->reshape_progress != MaxSector) {
09c9e5fa 5267 conf->prev_chunk_sectors = mddev->chunk_sectors;
e183eaed
N		 5268 conf->prev_algo = mddev->layout;
5269 }
1da177e4 5270
91adb564 5271 memory = conf->max_nr_stripes * (sizeof(struct stripe_head) +
5e5e3e78 5272 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
91adb564
N		 5273 if (grow_stripes(conf, conf->max_nr_stripes)) {
5274 printk(KERN_ERR
0c55e022
N		 5275 "md/raid:%s: couldn't allocate %dkB for buffers\n",
5276 mdname(mddev), memory);
91adb564
N		 5277 goto abort;
5278 } else
0c55e022
N		 5279 printk(KERN_INFO "md/raid:%s: allocated %dkB\n",
5280 mdname(mddev), memory);
1da177e4 5281
0232605d
N		 5282 sprintf(pers_name, "raid%d", mddev->new_level);
5283 conf->thread = md_register_thread(raid5d, mddev, pers_name);
91adb564
N		 5284 if (!conf->thread) {
5285 printk(KERN_ERR
0c55e022 5286 "md/raid:%s: couldn't allocate thread.\n",
91adb564 5287 mdname(mddev));
16a53ecc
N		 5288 goto abort;
5289 }
91adb564
N		 5290
5291 return conf;
5292
5293 abort:
5294 if (conf) {
95fc17aa 5295 free_conf(conf);
91adb564
N		 5296 return ERR_PTR(-EIO);
5297 } else
5298 return ERR_PTR(-ENOMEM);
5299}
5300
c148ffdc
N		 5301
5302static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
5303{
5304 switch (algo) {
5305 case ALGORITHM_PARITY_0:
5306 if (raid_disk < max_degraded)
5307 return 1;
5308 break;
5309 case ALGORITHM_PARITY_N:
5310 if (raid_disk >= raid_disks - max_degraded)
5311 return 1;
5312 break;
5313 case ALGORITHM_PARITY_0_6:
5314 if (raid_disk == 0 ||
5315 raid_disk == raid_disks - 1)
5316 return 1;
5317 break;
5318 case ALGORITHM_LEFT_ASYMMETRIC_6:
5319 case ALGORITHM_RIGHT_ASYMMETRIC_6:
5320 case ALGORITHM_LEFT_SYMMETRIC_6:
5321 case ALGORITHM_RIGHT_SYMMETRIC_6:
5322 if (raid_disk == raid_disks - 1)
5323 return 1;
5324 }
5325 return 0;
5326}
5327
fd01b88c 5328static int run(struct mddev *mddev)
91adb564 5329{
d1688a6d 5330 struct r5conf *conf;
9f7c2220 5331 int working_disks = 0;
c148ffdc 5332 int dirty_parity_disks = 0;
3cb03002 5333 struct md_rdev *rdev;
c148ffdc 5334 sector_t reshape_offset = 0;
17045f52 5335 int i;
b5254dd5
N		 5336 long long min_offset_diff = 0;
5337 int first = 1;
91adb564 5338
8c6ac868 5339 if (mddev->recovery_cp != MaxSector)
0c55e022 5340 printk(KERN_NOTICE "md/raid:%s: not clean"
8c6ac868
AN		 5341 " -- starting background reconstruction\n",
5342 mdname(mddev));
b5254dd5
N		 5343
5344 rdev_for_each(rdev, mddev) {
5345 long long diff;
5346 if (rdev->raid_disk < 0)
5347 continue;
5348 diff = (rdev->new_data_offset - rdev->data_offset);
5349 if (first) {
5350 min_offset_diff = diff;
5351 first = 0;
5352 } else if (mddev->reshape_backwards &&
5353 diff < min_offset_diff)
5354 min_offset_diff = diff;
5355 else if (!mddev->reshape_backwards &&
5356 diff > min_offset_diff)
5357 min_offset_diff = diff;
5358 }
5359
91adb564
N		 5360 if (mddev->reshape_position != MaxSector) {
5361 /* Check that we can continue the reshape.
b5254dd5
N		 5362 * Difficulties arise if the stripe we would write to
5363 * next is at or after the stripe we would read from next.
5364 * For a reshape that changes the number of devices, this
5365 * is only possible for a very short time, and mdadm makes
5366 * sure that time appears to have past before assembling
5367 * the array. So we fail if that time hasn't passed.
5368 * For a reshape that keeps the number of devices the same
5369 * mdadm must be monitoring the reshape can keeping the
5370 * critical areas read-only and backed up. It will start
5371 * the array in read-only mode, so we check for that.
91adb564
N		 5372 */
5373 sector_t here_new, here_old;
5374 int old_disks;
18b00334 5375 int max_degraded = (mddev->level == 6 ? 2 : 1);
91adb564 5376
88ce4930 5377 if (mddev->new_level != mddev->level) {
0c55e022 5378 printk(KERN_ERR "md/raid:%s: unsupported reshape "
91adb564
N		 5379 "required - aborting.\n",
5380 mdname(mddev));
5381 return -EINVAL;
5382 }
91adb564
N		 5383 old_disks = mddev->raid_disks - mddev->delta_disks;
5384 /* reshape_position must be on a new-stripe boundary, and one
5385 * further up in new geometry must map after here in old
5386 * geometry.
5387 */
5388 here_new = mddev->reshape_position;
664e7c41 5389 if (sector_div(here_new, mddev->new_chunk_sectors *
91adb564 5390 (mddev->raid_disks - max_degraded))) {
0c55e022
N		 5391 printk(KERN_ERR "md/raid:%s: reshape_position not "
5392 "on a stripe boundary\n", mdname(mddev));
91adb564
N		 5393 return -EINVAL;
5394 }
c148ffdc 5395 reshape_offset = here_new * mddev->new_chunk_sectors;
91adb564
N		 5396 /* here_new is the stripe we will write to */
5397 here_old = mddev->reshape_position;
9d8f0363 5398 sector_div(here_old, mddev->chunk_sectors *
91adb564
N		 5399 (old_disks-max_degraded));
5400 /* here_old is the first stripe that we might need to read
5401 * from */
67ac6011 5402 if (mddev->delta_disks == 0) {
b5254dd5
N		 5403 if ((here_new * mddev->new_chunk_sectors !=
5404 here_old * mddev->chunk_sectors)) {
5405 printk(KERN_ERR "md/raid:%s: reshape position is"
5406 " confused - aborting\n", mdname(mddev));
5407 return -EINVAL;
5408 }
67ac6011 5409 /* We cannot be sure it is safe to start an in-place
b5254dd5 5410 * reshape. It is only safe if user-space is monitoring
67ac6011
N		 5411 * and taking constant backups.
5412 * mdadm always starts a situation like this in
5413 * readonly mode so it can take control before
5414 * allowing any writes. So just check for that.
5415 */
b5254dd5
N		 5416 if (abs(min_offset_diff) >= mddev->chunk_sectors &&
5417 abs(min_offset_diff) >= mddev->new_chunk_sectors)
5418 /* not really in-place - so OK */;
5419 else if (mddev->ro == 0) {
5420 printk(KERN_ERR "md/raid:%s: in-place reshape "
5421 "must be started in read-only mode "
5422 "- aborting\n",
0c55e022 5423 mdname(mddev));
67ac6011
N		 5424 return -EINVAL;
5425 }
2c810cdd 5426 } else if (mddev->reshape_backwards
b5254dd5 5427 ? (here_new * mddev->new_chunk_sectors + min_offset_diff <=
67ac6011
N		 5428 here_old * mddev->chunk_sectors)
5429 : (here_new * mddev->new_chunk_sectors >=
b5254dd5 5430 here_old * mddev->chunk_sectors + (-min_offset_diff))) {
91adb564 5431 /* Reading from the same stripe as writing to - bad */
0c55e022
N		 5432 printk(KERN_ERR "md/raid:%s: reshape_position too early for "
5433 "auto-recovery - aborting.\n",
5434 mdname(mddev));
91adb564
N		 5435 return -EINVAL;
5436 }
0c55e022
N		 5437 printk(KERN_INFO "md/raid:%s: reshape will continue\n",
5438 mdname(mddev));
91adb564
N		 5439 /* OK, we should be able to continue; */
5440 } else {
5441 BUG_ON(mddev->level != mddev->new_level);
5442 BUG_ON(mddev->layout != mddev->new_layout);
664e7c41 5443 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
91adb564 5444 BUG_ON(mddev->delta_disks != 0);
1da177e4 5445 }
91adb564 5446
245f46c2
N		 5447 if (mddev->private == NULL)
5448 conf = setup_conf(mddev);
5449 else
5450 conf = mddev->private;
5451
91adb564
N		 5452 if (IS_ERR(conf))
5453 return PTR_ERR(conf);
5454
b5254dd5 5455 conf->min_offset_diff = min_offset_diff;
91adb564
N		 5456 mddev->thread = conf->thread;
5457 conf->thread = NULL;
5458 mddev->private = conf;
5459
17045f52
N		 5460 for (i = 0; i < conf->raid_disks && conf->previous_raid_disks;
5461 i++) {
5462 rdev = conf->disks[i].rdev;
5463 if (!rdev && conf->disks[i].replacement) {
5464 /* The replacement is all we have yet */
5465 rdev = conf->disks[i].replacement;
5466 conf->disks[i].replacement = NULL;
5467 clear_bit(Replacement, &rdev->flags);
5468 conf->disks[i].rdev = rdev;
5469 }
5470 if (!rdev)
c148ffdc 5471 continue;
17045f52
N		 5472 if (conf->disks[i].replacement &&
5473 conf->reshape_progress != MaxSector) {
5474 /* replacements and reshape simply do not mix. */
5475 printk(KERN_ERR "md: cannot handle concurrent "
5476 "replacement and reshape.\n");
5477 goto abort;
5478 }
2f115882 5479 if (test_bit(In_sync, &rdev->flags)) {
91adb564 5480 working_disks++;
2f115882
N		 5481 continue;
5482 }
c148ffdc
N		 5483 /* This disc is not fully in-sync. However if it
5484 * just stored parity (beyond the recovery_offset),
5485 * when we don't need to be concerned about the
5486 * array being dirty.
5487 * When reshape goes 'backwards', we never have
5488 * partially completed devices, so we only need
5489 * to worry about reshape going forwards.
5490 */
5491 /* Hack because v0.91 doesn't store recovery_offset properly. */
5492 if (mddev->major_version == 0 &&
5493 mddev->minor_version > 90)
5494 rdev->recovery_offset = reshape_offset;
5026d7a9 5495
c148ffdc
N		 5496 if (rdev->recovery_offset < reshape_offset) {
5497 /* We need to check old and new layout */
5498 if (!only_parity(rdev->raid_disk,
5499 conf->algorithm,
5500 conf->raid_disks,
5501 conf->max_degraded))
5502 continue;
5503 }
5504 if (!only_parity(rdev->raid_disk,
5505 conf->prev_algo,
5506 conf->previous_raid_disks,
5507 conf->max_degraded))
5508 continue;
5509 dirty_parity_disks++;
5510 }
91adb564 5511
17045f52
N		 5512 /*
5513 * 0 for a fully functional array, 1 or 2 for a degraded array.
5514 */
908f4fbd 5515 mddev->degraded = calc_degraded(conf);
91adb564 5516
674806d6 5517 if (has_failed(conf)) {
0c55e022 5518 printk(KERN_ERR "md/raid:%s: not enough operational devices"
1da177e4 5519 " (%d/%d failed)\n",
02c2de8c 5520 mdname(mddev), mddev->degraded, conf->raid_disks);
1da177e4
LT		 5521 goto abort;
5522 }
5523
91adb564 5524 /* device size must be a multiple of chunk size */
9d8f0363 5525 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
91adb564
N		 5526 mddev->resync_max_sectors = mddev->dev_sectors;
5527
c148ffdc 5528 if (mddev->degraded > dirty_parity_disks &&
1da177e4 5529 mddev->recovery_cp != MaxSector) {
6ff8d8ec
N		 5530 if (mddev->ok_start_degraded)
5531 printk(KERN_WARNING
0c55e022
N		 5532 "md/raid:%s: starting dirty degraded array"
5533 " - data corruption possible.\n",
6ff8d8ec
N		 5534 mdname(mddev));
5535 else {
5536 printk(KERN_ERR
0c55e022 5537 "md/raid:%s: cannot start dirty degraded array.\n",
6ff8d8ec
N		 5538 mdname(mddev));
5539 goto abort;
5540 }
1da177e4
LT		 5541 }
5542
1da177e4 5543 if (mddev->degraded == 0)
0c55e022
N		 5544 printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d"
5545 " devices, algorithm %d\n", mdname(mddev), conf->level,
e183eaed
N		 5546 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
5547 mddev->new_layout);
1da177e4 5548 else
0c55e022
N		 5549 printk(KERN_ALERT "md/raid:%s: raid level %d active with %d"
5550 " out of %d devices, algorithm %d\n",
5551 mdname(mddev), conf->level,
5552 mddev->raid_disks - mddev->degraded,
5553 mddev->raid_disks, mddev->new_layout);
1da177e4
LT		 5554
5555 print_raid5_conf(conf);
5556
fef9c61f 5557 if (conf->reshape_progress != MaxSector) {
fef9c61f 5558 conf->reshape_safe = conf->reshape_progress;
f6705578
N		 5559 atomic_set(&conf->reshape_stripes, 0);
5560 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5561 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5562 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5563 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5564 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 5565 "reshape");
f6705578
N		 5566 }
5567
1da177e4
LT		 5568
5569 /* Ok, everything is just fine now */
a64c876f
N		 5570 if (mddev->to_remove == &raid5_attrs_group)
5571 mddev->to_remove = NULL;
00bcb4ac
N		 5572 else if (mddev->kobj.sd &&
5573 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
5e55e2f5 5574 printk(KERN_WARNING
4a5add49 5575 "raid5: failed to create sysfs attributes for %s\n",
5e55e2f5 5576 mdname(mddev));
4a5add49 5577 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
7a5febe9 5578
4a5add49 5579 if (mddev->queue) {
9f7c2220 5580 int chunk_size;
620125f2 5581 bool discard_supported = true;
4a5add49
N		 5582 /* read-ahead size must cover two whole stripes, which
5583 * is 2 * (datadisks) * chunksize where 'n' is the
5584 * number of raid devices
5585 */
5586 int data_disks = conf->previous_raid_disks - conf->max_degraded;
5587 int stripe = data_disks *
5588 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
5589 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
5590 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
91adb564 5591
4a5add49 5592 blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec);
f022b2fd 5593
11d8a6e3
N		 5594 mddev->queue->backing_dev_info.congested_data = mddev;
5595 mddev->queue->backing_dev_info.congested_fn = raid5_congested;
7a5febe9 5596
9f7c2220
N		 5597 chunk_size = mddev->chunk_sectors << 9;
5598 blk_queue_io_min(mddev->queue, chunk_size);
5599 blk_queue_io_opt(mddev->queue, chunk_size *
5600 (conf->raid_disks - conf->max_degraded));
620125f2
SL		 5601 /*
5602 * We can only discard a whole stripe. It doesn't make sense to
5603 * discard data disk but write parity disk
5604 */
5605 stripe = stripe * PAGE_SIZE;
4ac6875e
N		 5606 /* Round up to power of 2, as discard handling
5607 * currently assumes that */
5608 while ((stripe-1) & stripe)
5609 stripe = (stripe | (stripe-1)) + 1;
620125f2
SL		 5610 mddev->queue->limits.discard_alignment = stripe;
5611 mddev->queue->limits.discard_granularity = stripe;
5612 /*
5613 * unaligned part of discard request will be ignored, so can't
5614 * guarantee discard_zerors_data
5615 */
5616 mddev->queue->limits.discard_zeroes_data = 0;
8f6c2e4b 5617
5026d7a9
PA		 5618 blk_queue_max_write_same_sectors(mddev->queue, 0);
5619
05616be5 5620 rdev_for_each(rdev, mddev) {
9f7c2220
N		 5621 disk_stack_limits(mddev->gendisk, rdev->bdev,
5622 rdev->data_offset << 9);
05616be5
N		 5623 disk_stack_limits(mddev->gendisk, rdev->bdev,
5624 rdev->new_data_offset << 9);
620125f2
SL		 5625 /*
5626 * discard_zeroes_data is required, otherwise data
5627 * could be lost. Consider a scenario: discard a stripe
5628 * (the stripe could be inconsistent if
5629 * discard_zeroes_data is 0); write one disk of the
5630 * stripe (the stripe could be inconsistent again
5631 * depending on which disks are used to calculate
5632 * parity); the disk is broken; The stripe data of this
5633 * disk is lost.
5634 */
5635 if (!blk_queue_discard(bdev_get_queue(rdev->bdev)) ||
5636 !bdev_get_queue(rdev->bdev)->
5637 limits.discard_zeroes_data)
5638 discard_supported = false;
05616be5 5639 }
620125f2
SL		 5640
5641 if (discard_supported &&
5642 mddev->queue->limits.max_discard_sectors >= stripe &&
5643 mddev->queue->limits.discard_granularity >= stripe)
5644 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
5645 mddev->queue);
5646 else
5647 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
5648 mddev->queue);
9f7c2220 5649 }
23032a0e 5650
1da177e4
LT		 5651 return 0;
5652abort:
01f96c0a 5653 md_unregister_thread(&mddev->thread);
e4f869d9
N		 5654 print_raid5_conf(conf);
5655 free_conf(conf);
1da177e4 5656 mddev->private = NULL;
0c55e022 5657 printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev));
1da177e4
LT		 5658 return -EIO;
5659}
5660
fd01b88c 5661static int stop(struct mddev *mddev)
1da177e4 5662{
d1688a6d 5663 struct r5conf *conf = mddev->private;
1da177e4 5664
01f96c0a 5665 md_unregister_thread(&mddev->thread);
11d8a6e3
N		 5666 if (mddev->queue)
5667 mddev->queue->backing_dev_info.congested_fn = NULL;
95fc17aa 5668 free_conf(conf);
a64c876f
N		 5669 mddev->private = NULL;
5670 mddev->to_remove = &raid5_attrs_group;
1da177e4
LT		 5671 return 0;
5672}
5673
fd01b88c 5674static void status(struct seq_file *seq, struct mddev *mddev)
1da177e4 5675{
d1688a6d 5676 struct r5conf *conf = mddev->private;
1da177e4
LT		 5677 int i;
5678
9d8f0363
AN		 5679 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
5680 mddev->chunk_sectors / 2, mddev->layout);
02c2de8c 5681 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
1da177e4
LT		 5682 for (i = 0; i < conf->raid_disks; i++)
5683 seq_printf (seq, "%s",
5684 conf->disks[i].rdev &&
b2d444d7 5685 test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_");
1da177e4 5686 seq_printf (seq, "]");
1da177e4
LT		 5687}
5688
d1688a6d 5689static void print_raid5_conf (struct r5conf *conf)
1da177e4
LT		 5690{
5691 int i;
5692 struct disk_info *tmp;
5693
0c55e022 5694 printk(KERN_DEBUG "RAID conf printout:\n");
1da177e4
LT		 5695 if (!conf) {
5696 printk("(conf==NULL)\n");
5697 return;
5698 }
0c55e022
N		 5699 printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level,
5700 conf->raid_disks,
5701 conf->raid_disks - conf->mddev->degraded);
1da177e4
LT		 5702
5703 for (i = 0; i < conf->raid_disks; i++) {
5704 char b[BDEVNAME_SIZE];
5705 tmp = conf->disks + i;
5706 if (tmp->rdev)
0c55e022
N		 5707 printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n",
5708 i, !test_bit(Faulty, &tmp->rdev->flags),
5709 bdevname(tmp->rdev->bdev, b));
1da177e4
LT		 5710 }
5711}
5712
fd01b88c 5713static int raid5_spare_active(struct mddev *mddev)
1da177e4
LT		 5714{
5715 int i;
d1688a6d 5716 struct r5conf *conf = mddev->private;
1da177e4 5717 struct disk_info *tmp;
6b965620
N		 5718 int count = 0;
5719 unsigned long flags;
1da177e4
LT		 5720
5721 for (i = 0; i < conf->raid_disks; i++) {
5722 tmp = conf->disks + i;
dd054fce
N		 5723 if (tmp->replacement
5724 && tmp->replacement->recovery_offset == MaxSector
5725 && !test_bit(Faulty, &tmp->replacement->flags)
5726 && !test_and_set_bit(In_sync, &tmp->replacement->flags)) {
5727 /* Replacement has just become active. */
5728 if (!tmp->rdev
5729 || !test_and_clear_bit(In_sync, &tmp->rdev->flags))
5730 count++;
5731 if (tmp->rdev) {
5732 /* Replaced device not technically faulty,
5733 * but we need to be sure it gets removed
5734 * and never re-added.
5735 */
5736 set_bit(Faulty, &tmp->rdev->flags);
5737 sysfs_notify_dirent_safe(
5738 tmp->rdev->sysfs_state);
5739 }
5740 sysfs_notify_dirent_safe(tmp->replacement->sysfs_state);
5741 } else if (tmp->rdev
70fffd0b 5742 && tmp->rdev->recovery_offset == MaxSector
b2d444d7 5743 && !test_bit(Faulty, &tmp->rdev->flags)
c04be0aa 5744 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
6b965620 5745 count++;
43c73ca4 5746 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
1da177e4
LT		 5747 }
5748 }
6b965620 5749 spin_lock_irqsave(&conf->device_lock, flags);
908f4fbd 5750 mddev->degraded = calc_degraded(conf);
6b965620 5751 spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4 5752 print_raid5_conf(conf);
6b965620 5753 return count;
1da177e4
LT		 5754}
5755
b8321b68 5756static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 5757{
d1688a6d 5758 struct r5conf *conf = mddev->private;
1da177e4 5759 int err = 0;
b8321b68 5760 int number = rdev->raid_disk;
657e3e4d 5761 struct md_rdev **rdevp;
1da177e4
LT		 5762 struct disk_info *p = conf->disks + number;
5763
5764 print_raid5_conf(conf);
657e3e4d
N		 5765 if (rdev == p->rdev)
5766 rdevp = &p->rdev;
5767 else if (rdev == p->replacement)
5768 rdevp = &p->replacement;
5769 else
5770 return 0;
5771
5772 if (number >= conf->raid_disks &&
5773 conf->reshape_progress == MaxSector)
5774 clear_bit(In_sync, &rdev->flags);
5775
5776 if (test_bit(In_sync, &rdev->flags) ||
5777 atomic_read(&rdev->nr_pending)) {
5778 err = -EBUSY;
5779 goto abort;
5780 }
5781 /* Only remove non-faulty devices if recovery
5782 * isn't possible.
5783 */
5784 if (!test_bit(Faulty, &rdev->flags) &&
5785 mddev->recovery_disabled != conf->recovery_disabled &&
5786 !has_failed(conf) &&
dd054fce 5787 (!p->replacement || p->replacement == rdev) &&
657e3e4d
N		 5788 number < conf->raid_disks) {
5789 err = -EBUSY;
5790 goto abort;
5791 }
5792 *rdevp = NULL;
5793 synchronize_rcu();
5794 if (atomic_read(&rdev->nr_pending)) {
5795 /* lost the race, try later */
5796 err = -EBUSY;
5797 *rdevp = rdev;
dd054fce
N		 5798 } else if (p->replacement) {
5799 /* We must have just cleared 'rdev' */
5800 p->rdev = p->replacement;
5801 clear_bit(Replacement, &p->replacement->flags);
5802 smp_mb(); /* Make sure other CPUs may see both as identical
5803 * but will never see neither - if they are careful
5804 */
5805 p->replacement = NULL;
5806 clear_bit(WantReplacement, &rdev->flags);
5807 } else
5808 /* We might have just removed the Replacement as faulty-
5809 * clear the bit just in case
5810 */
5811 clear_bit(WantReplacement, &rdev->flags);
1da177e4
LT		 5812abort:
5813
5814 print_raid5_conf(conf);
5815 return err;
5816}
5817
fd01b88c 5818static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 5819{
d1688a6d 5820 struct r5conf *conf = mddev->private;
199050ea 5821 int err = -EEXIST;
1da177e4
LT		 5822 int disk;
5823 struct disk_info *p;
6c2fce2e
NB		 5824 int first = 0;
5825 int last = conf->raid_disks - 1;
1da177e4 5826
7f0da59b
N		 5827 if (mddev->recovery_disabled == conf->recovery_disabled)
5828 return -EBUSY;
5829
dc10c643 5830 if (rdev->saved_raid_disk < 0 && has_failed(conf))
1da177e4 5831 /* no point adding a device */
199050ea 5832 return -EINVAL;
1da177e4 5833
6c2fce2e
NB		 5834 if (rdev->raid_disk >= 0)
5835 first = last = rdev->raid_disk;
1da177e4
LT		 5836
5837 /*
16a53ecc
N		 5838 * find the disk ... but prefer rdev->saved_raid_disk
5839 * if possible.
1da177e4 5840 */
16a53ecc 5841 if (rdev->saved_raid_disk >= 0 &&
6c2fce2e 5842 rdev->saved_raid_disk >= first &&
16a53ecc 5843 conf->disks[rdev->saved_raid_disk].rdev == NULL)
5cfb22a1
N		 5844 first = rdev->saved_raid_disk;
5845
5846 for (disk = first; disk <= last; disk++) {
7bfec5f3
N		 5847 p = conf->disks + disk;
5848 if (p->rdev == NULL) {
b2d444d7 5849 clear_bit(In_sync, &rdev->flags);
1da177e4 5850 rdev->raid_disk = disk;
199050ea 5851 err = 0;
72626685
N		 5852 if (rdev->saved_raid_disk != disk)
5853 conf->fullsync = 1;
d6065f7b 5854 rcu_assign_pointer(p->rdev, rdev);
5cfb22a1 5855 goto out;
1da177e4 5856 }
5cfb22a1
N		 5857 }
5858 for (disk = first; disk <= last; disk++) {
5859 p = conf->disks + disk;
7bfec5f3
N		 5860 if (test_bit(WantReplacement, &p->rdev->flags) &&
5861 p->replacement == NULL) {
5862 clear_bit(In_sync, &rdev->flags);
5863 set_bit(Replacement, &rdev->flags);
5864 rdev->raid_disk = disk;
5865 err = 0;
5866 conf->fullsync = 1;
5867 rcu_assign_pointer(p->replacement, rdev);
5868 break;
5869 }
5870 }
5cfb22a1 5871out:
1da177e4 5872 print_raid5_conf(conf);
199050ea 5873 return err;
1da177e4
LT		 5874}
5875
fd01b88c 5876static int raid5_resize(struct mddev *mddev, sector_t sectors)
1da177e4
LT		 5877{
5878 /* no resync is happening, and there is enough space
5879 * on all devices, so we can resize.
5880 * We need to make sure resync covers any new space.
5881 * If the array is shrinking we should possibly wait until
5882 * any io in the removed space completes, but it hardly seems
5883 * worth it.
5884 */
a4a6125a 5885 sector_t newsize;
9d8f0363 5886 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
a4a6125a
N		 5887 newsize = raid5_size(mddev, sectors, mddev->raid_disks);
5888 if (mddev->external_size &&
5889 mddev->array_sectors > newsize)
b522adcd 5890 return -EINVAL;
a4a6125a
N		 5891 if (mddev->bitmap) {
5892 int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0);
5893 if (ret)
5894 return ret;
5895 }
5896 md_set_array_sectors(mddev, newsize);
f233ea5c 5897 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 5898 revalidate_disk(mddev->gendisk);
b098636c
N		 5899 if (sectors > mddev->dev_sectors &&
5900 mddev->recovery_cp > mddev->dev_sectors) {
58c0fed4 5901 mddev->recovery_cp = mddev->dev_sectors;
1da177e4
LT		 5902 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5903 }
58c0fed4 5904 mddev->dev_sectors = sectors;
4b5c7ae8 5905 mddev->resync_max_sectors = sectors;
1da177e4
LT		 5906 return 0;
5907}
5908
fd01b88c 5909static int check_stripe_cache(struct mddev *mddev)
01ee22b4
N		 5910{
5911 /* Can only proceed if there are plenty of stripe_heads.
5912 * We need a minimum of one full stripe,, and for sensible progress
5913 * it is best to have about 4 times that.
5914 * If we require 4 times, then the default 256 4K stripe_heads will
5915 * allow for chunk sizes up to 256K, which is probably OK.
5916 * If the chunk size is greater, user-space should request more
5917 * stripe_heads first.
5918 */
d1688a6d 5919 struct r5conf *conf = mddev->private;
01ee22b4
N		 5920 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
5921 > conf->max_nr_stripes ||
5922 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
5923 > conf->max_nr_stripes) {
0c55e022
N		 5924 printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n",
5925 mdname(mddev),
01ee22b4
N		 5926 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
5927 / STRIPE_SIZE)*4);
5928 return 0;
5929 }
5930 return 1;
5931}
5932
fd01b88c 5933static int check_reshape(struct mddev *mddev)
29269553 5934{
d1688a6d 5935 struct r5conf *conf = mddev->private;
29269553 5936
88ce4930
N		 5937 if (mddev->delta_disks == 0 &&
5938 mddev->new_layout == mddev->layout &&
664e7c41 5939 mddev->new_chunk_sectors == mddev->chunk_sectors)
50ac168a 5940 return 0; /* nothing to do */
674806d6 5941 if (has_failed(conf))
ec32a2bd
N		 5942 return -EINVAL;
5943 if (mddev->delta_disks < 0) {
5944 /* We might be able to shrink, but the devices must
5945 * be made bigger first.
5946 * For raid6, 4 is the minimum size.
5947 * Otherwise 2 is the minimum
5948 */
5949 int min = 2;
5950 if (mddev->level == 6)
5951 min = 4;
5952 if (mddev->raid_disks + mddev->delta_disks < min)
5953 return -EINVAL;
5954 }
29269553 5955
01ee22b4 5956 if (!check_stripe_cache(mddev))
29269553 5957 return -ENOSPC;
29269553 5958
e56108d6
N		 5959 return resize_stripes(conf, (conf->previous_raid_disks
5960 + mddev->delta_disks));
63c70c4f
N		 5961}
5962
fd01b88c 5963static int raid5_start_reshape(struct mddev *mddev)
63c70c4f 5964{
d1688a6d 5965 struct r5conf *conf = mddev->private;
3cb03002 5966 struct md_rdev *rdev;
63c70c4f 5967 int spares = 0;
c04be0aa 5968 unsigned long flags;
63c70c4f 5969
f416885e 5970 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
63c70c4f
N		 5971 return -EBUSY;
5972
01ee22b4
N		 5973 if (!check_stripe_cache(mddev))
5974 return -ENOSPC;
5975
30b67645
N		 5976 if (has_failed(conf))
5977 return -EINVAL;
5978
c6563a8c 5979 rdev_for_each(rdev, mddev) {
469518a3
N		 5980 if (!test_bit(In_sync, &rdev->flags)
5981 && !test_bit(Faulty, &rdev->flags))
29269553 5982 spares++;
c6563a8c 5983 }
63c70c4f 5984
f416885e 5985 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
29269553
N		 5986 /* Not enough devices even to make a degraded array
5987 * of that size
5988 */
5989 return -EINVAL;
5990
ec32a2bd
N		 5991 /* Refuse to reduce size of the array. Any reductions in
5992 * array size must be through explicit setting of array_size
5993 * attribute.
5994 */
5995 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
5996 < mddev->array_sectors) {
0c55e022 5997 printk(KERN_ERR "md/raid:%s: array size must be reduced "
ec32a2bd
N		 5998 "before number of disks\n", mdname(mddev));
5999 return -EINVAL;
6000 }
6001
f6705578 6002 atomic_set(&conf->reshape_stripes, 0);
29269553
N		 6003 spin_lock_irq(&conf->device_lock);
6004 conf->previous_raid_disks = conf->raid_disks;
63c70c4f 6005 conf->raid_disks += mddev->delta_disks;
09c9e5fa
AN		 6006 conf->prev_chunk_sectors = conf->chunk_sectors;
6007 conf->chunk_sectors = mddev->new_chunk_sectors;
88ce4930
N		 6008 conf->prev_algo = conf->algorithm;
6009 conf->algorithm = mddev->new_layout;
05616be5
N		 6010 conf->generation++;
6011 /* Code that selects data_offset needs to see the generation update
6012 * if reshape_progress has been set - so a memory barrier needed.
6013 */
6014 smp_mb();
2c810cdd 6015 if (mddev->reshape_backwards)
fef9c61f
N		 6016 conf->reshape_progress = raid5_size(mddev, 0, 0);
6017 else
6018 conf->reshape_progress = 0;
6019 conf->reshape_safe = conf->reshape_progress;
29269553
N		 6020 spin_unlock_irq(&conf->device_lock);
6021
6022 /* Add some new drives, as many as will fit.
6023 * We know there are enough to make the newly sized array work.
3424bf6a
N		 6024 * Don't add devices if we are reducing the number of
6025 * devices in the array. This is because it is not possible
6026 * to correctly record the "partially reconstructed" state of
6027 * such devices during the reshape and confusion could result.
29269553 6028 */
87a8dec9 6029 if (mddev->delta_disks >= 0) {
dafb20fa 6030 rdev_for_each(rdev, mddev)
87a8dec9
N		 6031 if (rdev->raid_disk < 0 &&
6032 !test_bit(Faulty, &rdev->flags)) {
6033 if (raid5_add_disk(mddev, rdev) == 0) {
87a8dec9 6034 if (rdev->raid_disk
9d4c7d87 6035 >= conf->previous_raid_disks)
87a8dec9 6036 set_bit(In_sync, &rdev->flags);
9d4c7d87 6037 else
87a8dec9 6038 rdev->recovery_offset = 0;
36fad858
NK		 6039
6040 if (sysfs_link_rdev(mddev, rdev))
87a8dec9 6041 /* Failure here is OK */;
50da0840 6042 }
87a8dec9
N		 6043 } else if (rdev->raid_disk >= conf->previous_raid_disks
6044 && !test_bit(Faulty, &rdev->flags)) {
6045 /* This is a spare that was manually added */
6046 set_bit(In_sync, &rdev->flags);
87a8dec9 6047 }
29269553 6048
87a8dec9
N		 6049 /* When a reshape changes the number of devices,
6050 * ->degraded is measured against the larger of the
6051 * pre and post number of devices.
6052 */
ec32a2bd 6053 spin_lock_irqsave(&conf->device_lock, flags);
908f4fbd 6054 mddev->degraded = calc_degraded(conf);
ec32a2bd
N		 6055 spin_unlock_irqrestore(&conf->device_lock, flags);
6056 }
63c70c4f 6057 mddev->raid_disks = conf->raid_disks;
e516402c 6058 mddev->reshape_position = conf->reshape_progress;
850b2b42 6059 set_bit(MD_CHANGE_DEVS, &mddev->flags);
f6705578 6060
29269553
N		 6061 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6062 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6063 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6064 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6065 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 6066 "reshape");
29269553
N		 6067 if (!mddev->sync_thread) {
6068 mddev->recovery = 0;
6069 spin_lock_irq(&conf->device_lock);
6070 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
05616be5
N		 6071 rdev_for_each(rdev, mddev)
6072 rdev->new_data_offset = rdev->data_offset;
6073 smp_wmb();
fef9c61f 6074 conf->reshape_progress = MaxSector;
1e3fa9bd 6075 mddev->reshape_position = MaxSector;
29269553
N		 6076 spin_unlock_irq(&conf->device_lock);
6077 return -EAGAIN;
6078 }
c8f517c4 6079 conf->reshape_checkpoint = jiffies;
29269553
N		 6080 md_wakeup_thread(mddev->sync_thread);
6081 md_new_event(mddev);
6082 return 0;
6083}
29269553 6084
ec32a2bd
N		 6085/* This is called from the reshape thread and should make any
6086 * changes needed in 'conf'
6087 */
d1688a6d 6088static void end_reshape(struct r5conf *conf)
29269553 6089{
29269553 6090
f6705578 6091 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
05616be5 6092 struct md_rdev *rdev;
f6705578 6093
f6705578 6094 spin_lock_irq(&conf->device_lock);
cea9c228 6095 conf->previous_raid_disks = conf->raid_disks;
05616be5
N		 6096 rdev_for_each(rdev, conf->mddev)
6097 rdev->data_offset = rdev->new_data_offset;
6098 smp_wmb();
fef9c61f 6099 conf->reshape_progress = MaxSector;
f6705578 6100 spin_unlock_irq(&conf->device_lock);
b0f9ec04 6101 wake_up(&conf->wait_for_overlap);
16a53ecc
N		 6102
6103 /* read-ahead size must cover two whole stripes, which is
6104 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
6105 */
4a5add49 6106 if (conf->mddev->queue) {
cea9c228 6107 int data_disks = conf->raid_disks - conf->max_degraded;
09c9e5fa 6108 int stripe = data_disks * ((conf->chunk_sectors << 9)
cea9c228 6109 / PAGE_SIZE);
16a53ecc
N		 6110 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
6111 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
6112 }
29269553 6113 }
29269553
N		 6114}
6115
ec32a2bd
N		 6116/* This is called from the raid5d thread with mddev_lock held.
6117 * It makes config changes to the device.
6118 */
fd01b88c 6119static void raid5_finish_reshape(struct mddev *mddev)
cea9c228 6120{
d1688a6d 6121 struct r5conf *conf = mddev->private;
cea9c228
N		 6122
6123 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6124
ec32a2bd
N		 6125 if (mddev->delta_disks > 0) {
6126 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
6127 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 6128 revalidate_disk(mddev->gendisk);
ec32a2bd
N		 6129 } else {
6130 int d;
908f4fbd
N		 6131 spin_lock_irq(&conf->device_lock);
6132 mddev->degraded = calc_degraded(conf);
6133 spin_unlock_irq(&conf->device_lock);
ec32a2bd
N		 6134 for (d = conf->raid_disks ;
6135 d < conf->raid_disks - mddev->delta_disks;
1a67dde0 6136 d++) {
3cb03002 6137 struct md_rdev *rdev = conf->disks[d].rdev;
da7613b8
N		 6138 if (rdev)
6139 clear_bit(In_sync, &rdev->flags);
6140 rdev = conf->disks[d].replacement;
6141 if (rdev)
6142 clear_bit(In_sync, &rdev->flags);
1a67dde0 6143 }
cea9c228 6144 }
88ce4930 6145 mddev->layout = conf->algorithm;
09c9e5fa 6146 mddev->chunk_sectors = conf->chunk_sectors;
ec32a2bd
N		 6147 mddev->reshape_position = MaxSector;
6148 mddev->delta_disks = 0;
2c810cdd 6149 mddev->reshape_backwards = 0;
cea9c228
N		 6150 }
6151}
6152
fd01b88c 6153static void raid5_quiesce(struct mddev *mddev, int state)
72626685 6154{
d1688a6d 6155 struct r5conf *conf = mddev->private;
72626685
N		 6156
6157 switch(state) {
e464eafd
N		 6158 case 2: /* resume for a suspend */
6159 wake_up(&conf->wait_for_overlap);
6160 break;
6161
72626685
N		 6162 case 1: /* stop all writes */
6163 spin_lock_irq(&conf->device_lock);
64bd660b
N		 6164 /* '2' tells resync/reshape to pause so that all
6165 * active stripes can drain
6166 */
6167 conf->quiesce = 2;
72626685 6168 wait_event_lock_irq(conf->wait_for_stripe,
46031f9a
RBJ		 6169 atomic_read(&conf->active_stripes) == 0 &&
6170 atomic_read(&conf->active_aligned_reads) == 0,
eed8c02e 6171 conf->device_lock);
64bd660b 6172 conf->quiesce = 1;
72626685 6173 spin_unlock_irq(&conf->device_lock);
64bd660b
N		 6174 /* allow reshape to continue */
6175 wake_up(&conf->wait_for_overlap);
72626685
N		 6176 break;
6177
6178 case 0: /* re-enable writes */
6179 spin_lock_irq(&conf->device_lock);
6180 conf->quiesce = 0;
6181 wake_up(&conf->wait_for_stripe);
e464eafd 6182 wake_up(&conf->wait_for_overlap);
72626685
N		 6183 spin_unlock_irq(&conf->device_lock);
6184 break;
6185 }
72626685 6186}
b15c2e57 6187
d562b0c4 6188
fd01b88c 6189static void *raid45_takeover_raid0(struct mddev *mddev, int level)
54071b38 6190{
e373ab10 6191 struct r0conf *raid0_conf = mddev->private;
d76c8420 6192 sector_t sectors;
54071b38 6193
f1b29bca 6194 /* for raid0 takeover only one zone is supported */
e373ab10 6195 if (raid0_conf->nr_strip_zones > 1) {
0c55e022
N		 6196 printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n",
6197 mdname(mddev));
f1b29bca
DW		 6198 return ERR_PTR(-EINVAL);
6199 }
6200
e373ab10
N		 6201 sectors = raid0_conf->strip_zone[0].zone_end;
6202 sector_div(sectors, raid0_conf->strip_zone[0].nb_dev);
3b71bd93 6203 mddev->dev_sectors = sectors;
f1b29bca 6204 mddev->new_level = level;
54071b38
TM		 6205 mddev->new_layout = ALGORITHM_PARITY_N;
6206 mddev->new_chunk_sectors = mddev->chunk_sectors;
6207 mddev->raid_disks += 1;
6208 mddev->delta_disks = 1;
6209 /* make sure it will be not marked as dirty */
6210 mddev->recovery_cp = MaxSector;
6211
6212 return setup_conf(mddev);
6213}
6214
6215
fd01b88c 6216static void *raid5_takeover_raid1(struct mddev *mddev)
d562b0c4
N		 6217{
6218 int chunksect;
6219
6220 if (mddev->raid_disks != 2 ||
6221 mddev->degraded > 1)
6222 return ERR_PTR(-EINVAL);
6223
6224 /* Should check if there are write-behind devices? */
6225
6226 chunksect = 64*2; /* 64K by default */
6227
6228 /* The array must be an exact multiple of chunksize */
6229 while (chunksect && (mddev->array_sectors & (chunksect-1)))
6230 chunksect >>= 1;
6231
6232 if ((chunksect<<9) < STRIPE_SIZE)
6233 /* array size does not allow a suitable chunk size */
6234 return ERR_PTR(-EINVAL);
6235
6236 mddev->new_level = 5;
6237 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
664e7c41 6238 mddev->new_chunk_sectors = chunksect;
d562b0c4
N		 6239
6240 return setup_conf(mddev);
6241}
6242
fd01b88c 6243static void *raid5_takeover_raid6(struct mddev *mddev)
fc9739c6
N		 6244{
6245 int new_layout;
6246
6247 switch (mddev->layout) {
6248 case ALGORITHM_LEFT_ASYMMETRIC_6:
6249 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
6250 break;
6251 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6252 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
6253 break;
6254 case ALGORITHM_LEFT_SYMMETRIC_6:
6255 new_layout = ALGORITHM_LEFT_SYMMETRIC;
6256 break;
6257 case ALGORITHM_RIGHT_SYMMETRIC_6:
6258 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
6259 break;
6260 case ALGORITHM_PARITY_0_6:
6261 new_layout = ALGORITHM_PARITY_0;
6262 break;
6263 case ALGORITHM_PARITY_N:
6264 new_layout = ALGORITHM_PARITY_N;
6265 break;
6266 default:
6267 return ERR_PTR(-EINVAL);
6268 }
6269 mddev->new_level = 5;
6270 mddev->new_layout = new_layout;
6271 mddev->delta_disks = -1;
6272 mddev->raid_disks -= 1;
6273 return setup_conf(mddev);
6274}
6275
d562b0c4 6276
fd01b88c 6277static int raid5_check_reshape(struct mddev *mddev)
b3546035 6278{
88ce4930
N		 6279 /* For a 2-drive array, the layout and chunk size can be changed
6280 * immediately as not restriping is needed.
6281 * For larger arrays we record the new value - after validation
6282 * to be used by a reshape pass.
b3546035 6283 */
d1688a6d 6284 struct r5conf *conf = mddev->private;
597a711b 6285 int new_chunk = mddev->new_chunk_sectors;
b3546035 6286
597a711b 6287 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
b3546035
N		 6288 return -EINVAL;
6289 if (new_chunk > 0) {
0ba459d2 6290 if (!is_power_of_2(new_chunk))
b3546035 6291 return -EINVAL;
597a711b 6292 if (new_chunk < (PAGE_SIZE>>9))
b3546035 6293 return -EINVAL;
597a711b 6294 if (mddev->array_sectors & (new_chunk-1))
b3546035
N		 6295 /* not factor of array size */
6296 return -EINVAL;
6297 }
6298
6299 /* They look valid */
6300
88ce4930 6301 if (mddev->raid_disks == 2) {
597a711b
N		 6302 /* can make the change immediately */
6303 if (mddev->new_layout >= 0) {
6304 conf->algorithm = mddev->new_layout;
6305 mddev->layout = mddev->new_layout;
88ce4930
N		 6306 }
6307 if (new_chunk > 0) {
597a711b
N		 6308 conf->chunk_sectors = new_chunk ;
6309 mddev->chunk_sectors = new_chunk;
88ce4930
N		 6310 }
6311 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6312 md_wakeup_thread(mddev->thread);
b3546035 6313 }
50ac168a 6314 return check_reshape(mddev);
88ce4930
N		 6315}
6316
fd01b88c 6317static int raid6_check_reshape(struct mddev *mddev)
88ce4930 6318{
597a711b 6319 int new_chunk = mddev->new_chunk_sectors;
50ac168a 6320
597a711b 6321 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
88ce4930 6322 return -EINVAL;
b3546035 6323 if (new_chunk > 0) {
0ba459d2 6324 if (!is_power_of_2(new_chunk))
88ce4930 6325 return -EINVAL;
597a711b 6326 if (new_chunk < (PAGE_SIZE >> 9))
88ce4930 6327 return -EINVAL;
597a711b 6328 if (mddev->array_sectors & (new_chunk-1))
88ce4930
N		 6329 /* not factor of array size */
6330 return -EINVAL;
b3546035 6331 }
88ce4930
N		 6332
6333 /* They look valid */
50ac168a 6334 return check_reshape(mddev);
b3546035
N		 6335}
6336
fd01b88c 6337static void *raid5_takeover(struct mddev *mddev)
d562b0c4
N		 6338{
6339 /* raid5 can take over:
f1b29bca 6340 * raid0 - if there is only one strip zone - make it a raid4 layout
d562b0c4
N		 6341 * raid1 - if there are two drives. We need to know the chunk size
6342 * raid4 - trivial - just use a raid4 layout.
6343 * raid6 - Providing it is a *_6 layout
d562b0c4 6344 */
f1b29bca
DW		 6345 if (mddev->level == 0)
6346 return raid45_takeover_raid0(mddev, 5);
d562b0c4
N		 6347 if (mddev->level == 1)
6348 return raid5_takeover_raid1(mddev);
e9d4758f
N		 6349 if (mddev->level == 4) {
6350 mddev->new_layout = ALGORITHM_PARITY_N;
6351 mddev->new_level = 5;
6352 return setup_conf(mddev);
6353 }
fc9739c6
N		 6354 if (mddev->level == 6)
6355 return raid5_takeover_raid6(mddev);
d562b0c4
N		 6356
6357 return ERR_PTR(-EINVAL);
6358}
6359
fd01b88c 6360static void *raid4_takeover(struct mddev *mddev)
a78d38a1 6361{
f1b29bca
DW		 6362 /* raid4 can take over:
6363 * raid0 - if there is only one strip zone
6364 * raid5 - if layout is right
a78d38a1 6365 */
f1b29bca
DW		 6366 if (mddev->level == 0)
6367 return raid45_takeover_raid0(mddev, 4);
a78d38a1
N		 6368 if (mddev->level == 5 &&
6369 mddev->layout == ALGORITHM_PARITY_N) {
6370 mddev->new_layout = 0;
6371 mddev->new_level = 4;
6372 return setup_conf(mddev);
6373 }
6374 return ERR_PTR(-EINVAL);
6375}
d562b0c4 6376
84fc4b56 6377static struct md_personality raid5_personality;
245f46c2 6378
fd01b88c 6379static void *raid6_takeover(struct mddev *mddev)
245f46c2
N		 6380{
6381 /* Currently can only take over a raid5. We map the
6382 * personality to an equivalent raid6 personality
6383 * with the Q block at the end.
6384 */
6385 int new_layout;
6386
6387 if (mddev->pers != &raid5_personality)
6388 return ERR_PTR(-EINVAL);
6389 if (mddev->degraded > 1)
6390 return ERR_PTR(-EINVAL);
6391 if (mddev->raid_disks > 253)
6392 return ERR_PTR(-EINVAL);
6393 if (mddev->raid_disks < 3)
6394 return ERR_PTR(-EINVAL);
6395
6396 switch (mddev->layout) {
6397 case ALGORITHM_LEFT_ASYMMETRIC:
6398 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
6399 break;
6400 case ALGORITHM_RIGHT_ASYMMETRIC:
6401 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
6402 break;
6403 case ALGORITHM_LEFT_SYMMETRIC:
6404 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
6405 break;
6406 case ALGORITHM_RIGHT_SYMMETRIC:
6407 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
6408 break;
6409 case ALGORITHM_PARITY_0:
6410 new_layout = ALGORITHM_PARITY_0_6;
6411 break;
6412 case ALGORITHM_PARITY_N:
6413 new_layout = ALGORITHM_PARITY_N;
6414 break;
6415 default:
6416 return ERR_PTR(-EINVAL);
6417 }
6418 mddev->new_level = 6;
6419 mddev->new_layout = new_layout;
6420 mddev->delta_disks = 1;
6421 mddev->raid_disks += 1;
6422 return setup_conf(mddev);
6423}
6424
6425
84fc4b56 6426static struct md_personality raid6_personality =
16a53ecc
N		 6427{
6428 .name = "raid6",
6429 .level = 6,
6430 .owner = THIS_MODULE,
6431 .make_request = make_request,
6432 .run = run,
6433 .stop = stop,
6434 .status = status,
6435 .error_handler = error,
6436 .hot_add_disk = raid5_add_disk,
6437 .hot_remove_disk= raid5_remove_disk,
6438 .spare_active = raid5_spare_active,
6439 .sync_request = sync_request,
6440 .resize = raid5_resize,
80c3a6ce 6441 .size = raid5_size,
50ac168a 6442 .check_reshape = raid6_check_reshape,
f416885e 6443 .start_reshape = raid5_start_reshape,
cea9c228 6444 .finish_reshape = raid5_finish_reshape,
16a53ecc 6445 .quiesce = raid5_quiesce,
245f46c2 6446 .takeover = raid6_takeover,
16a53ecc 6447};
84fc4b56 6448static struct md_personality raid5_personality =
1da177e4
LT		 6449{
6450 .name = "raid5",
2604b703 6451 .level = 5,
1da177e4
LT		 6452 .owner = THIS_MODULE,
6453 .make_request = make_request,
6454 .run = run,
6455 .stop = stop,
6456 .status = status,
6457 .error_handler = error,
6458 .hot_add_disk = raid5_add_disk,
6459 .hot_remove_disk= raid5_remove_disk,
6460 .spare_active = raid5_spare_active,
6461 .sync_request = sync_request,
6462 .resize = raid5_resize,
80c3a6ce 6463 .size = raid5_size,
63c70c4f
N		 6464 .check_reshape = raid5_check_reshape,
6465 .start_reshape = raid5_start_reshape,
cea9c228 6466 .finish_reshape = raid5_finish_reshape,
72626685 6467 .quiesce = raid5_quiesce,
d562b0c4 6468 .takeover = raid5_takeover,
1da177e4
LT		 6469};
6470
84fc4b56 6471static struct md_personality raid4_personality =
1da177e4 6472{
2604b703
N		 6473 .name = "raid4",
6474 .level = 4,
6475 .owner = THIS_MODULE,
6476 .make_request = make_request,
6477 .run = run,
6478 .stop = stop,
6479 .status = status,
6480 .error_handler = error,
6481 .hot_add_disk = raid5_add_disk,
6482 .hot_remove_disk= raid5_remove_disk,
6483 .spare_active = raid5_spare_active,
6484 .sync_request = sync_request,
6485 .resize = raid5_resize,
80c3a6ce 6486 .size = raid5_size,
3d37890b
N		 6487 .check_reshape = raid5_check_reshape,
6488 .start_reshape = raid5_start_reshape,
cea9c228 6489 .finish_reshape = raid5_finish_reshape,
2604b703 6490 .quiesce = raid5_quiesce,
a78d38a1 6491 .takeover = raid4_takeover,
2604b703
N		 6492};
6493
6494static int __init raid5_init(void)
6495{
16a53ecc 6496 register_md_personality(&raid6_personality);
2604b703
N		 6497 register_md_personality(&raid5_personality);
6498 register_md_personality(&raid4_personality);
6499 return 0;
1da177e4
LT		 6500}
6501
2604b703 6502static void raid5_exit(void)
1da177e4 6503{
16a53ecc 6504 unregister_md_personality(&raid6_personality);
2604b703
N		 6505 unregister_md_personality(&raid5_personality);
6506 unregister_md_personality(&raid4_personality);
1da177e4
LT		 6507}
6508
6509module_init(raid5_init);
6510module_exit(raid5_exit);
6511MODULE_LICENSE("GPL");
0efb9e61 6512MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
1da177e4 6513MODULE_ALIAS("md-personality-4"); /* RAID5 */
d9d166c2
N		 6514MODULE_ALIAS("md-raid5");
6515MODULE_ALIAS("md-raid4");
2604b703
N		 6516MODULE_ALIAS("md-level-5");
6517MODULE_ALIAS("md-level-4");
16a53ecc
N		 6518MODULE_ALIAS("md-personality-8"); /* RAID6 */
6519MODULE_ALIAS("md-raid6");
6520MODULE_ALIAS("md-level-6");
6521
6522/* This used to be two separate modules, they were: */
6523MODULE_ALIAS("raid5");
6524MODULE_ALIAS("raid6");
kernel source for telekom puls (alcatel/mediatek)