projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
block/nbd: micro-optimization in nbd request completion
[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>
43b2e5d8 56#include "md.h"
bff61975 57#include "raid5.h"
54071b38 58#include "raid0.h"
ef740c37 59#include "bitmap.h"
72626685 60
1da177e4
LT		 61/*
62 * Stripe cache
63 */
64
65#define NR_STRIPES 256
66#define STRIPE_SIZE PAGE_SIZE
67#define STRIPE_SHIFT (PAGE_SHIFT - 9)
68#define STRIPE_SECTORS (STRIPE_SIZE>>9)
69#define IO_THRESHOLD 1
8b3e6cdc 70#define BYPASS_THRESHOLD 1
fccddba0 71#define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
1da177e4
LT		 72#define HASH_MASK (NR_HASH - 1)
73
d1688a6d 74static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect)
db298e19
N		 75{
76 int hash = (sect >> STRIPE_SHIFT) & HASH_MASK;
77 return &conf->stripe_hashtbl[hash];
78}
1da177e4
LT		 79
80/* bio's attached to a stripe+device for I/O are linked together in bi_sector
81 * order without overlap. There may be several bio's per stripe+device, and
82 * a bio could span several devices.
83 * When walking this list for a particular stripe+device, we must never proceed
84 * beyond a bio that extends past this device, as the next bio might no longer
85 * be valid.
db298e19 86 * This function is used to determine the 'next' bio in the list, given the sector
1da177e4
LT		 87 * of the current stripe+device
88 */
db298e19
N		 89static inline struct bio *r5_next_bio(struct bio *bio, sector_t sector)
90{
91 int sectors = bio->bi_size >> 9;
92 if (bio->bi_sector + sectors < sector + STRIPE_SECTORS)
93 return bio->bi_next;
94 else
95 return NULL;
96}
1da177e4 97
960e739d 98/*
5b99c2ff
JA		 99 * We maintain a biased count of active stripes in the bottom 16 bits of
100 * bi_phys_segments, and a count of processed stripes in the upper 16 bits
960e739d
JA		 101 */
102static inline int raid5_bi_phys_segments(struct bio *bio)
103{
5b99c2ff 104 return bio->bi_phys_segments & 0xffff;
960e739d
JA		 105}
106
107static inline int raid5_bi_hw_segments(struct bio *bio)
108{
5b99c2ff 109 return (bio->bi_phys_segments >> 16) & 0xffff;
960e739d
JA		 110}
111
112static inline int raid5_dec_bi_phys_segments(struct bio *bio)
113{
114 --bio->bi_phys_segments;
115 return raid5_bi_phys_segments(bio);
116}
117
118static inline int raid5_dec_bi_hw_segments(struct bio *bio)
119{
120 unsigned short val = raid5_bi_hw_segments(bio);
121
122 --val;
5b99c2ff 123 bio->bi_phys_segments = (val << 16) | raid5_bi_phys_segments(bio);
960e739d
JA		 124 return val;
125}
126
127static inline void raid5_set_bi_hw_segments(struct bio *bio, unsigned int cnt)
128{
9b2dc8b6 129 bio->bi_phys_segments = raid5_bi_phys_segments(bio) | (cnt << 16);
960e739d
JA		 130}
131
d0dabf7e
N		 132/* Find first data disk in a raid6 stripe */
133static inline int raid6_d0(struct stripe_head *sh)
134{
67cc2b81
N		 135 if (sh->ddf_layout)
136 /* ddf always start from first device */
137 return 0;
138 /* md starts just after Q block */
d0dabf7e
N		 139 if (sh->qd_idx == sh->disks - 1)
140 return 0;
141 else
142 return sh->qd_idx + 1;
143}
16a53ecc
N		 144static inline int raid6_next_disk(int disk, int raid_disks)
145{
146 disk++;
147 return (disk < raid_disks) ? disk : 0;
148}
a4456856 149
d0dabf7e
N		 150/* When walking through the disks in a raid5, starting at raid6_d0,
151 * We need to map each disk to a 'slot', where the data disks are slot
152 * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk
153 * is raid_disks-1. This help does that mapping.
154 */
67cc2b81
N		 155static int raid6_idx_to_slot(int idx, struct stripe_head *sh,
156 int *count, int syndrome_disks)
d0dabf7e 157{
6629542e 158 int slot = *count;
67cc2b81 159
e4424fee 160 if (sh->ddf_layout)
6629542e 161 (*count)++;
d0dabf7e 162 if (idx == sh->pd_idx)
67cc2b81 163 return syndrome_disks;
d0dabf7e 164 if (idx == sh->qd_idx)
67cc2b81 165 return syndrome_disks + 1;
e4424fee 166 if (!sh->ddf_layout)
6629542e 167 (*count)++;
d0dabf7e
N		 168 return slot;
169}
170
a4456856
DW		 171static void return_io(struct bio *return_bi)
172{
173 struct bio *bi = return_bi;
174 while (bi) {
a4456856
DW		 175
176 return_bi = bi->bi_next;
177 bi->bi_next = NULL;
178 bi->bi_size = 0;
0e13fe23 179 bio_endio(bi, 0);
a4456856
DW		 180 bi = return_bi;
181 }
182}
183
d1688a6d 184static void print_raid5_conf (struct r5conf *conf);
1da177e4 185
600aa109
DW		 186static int stripe_operations_active(struct stripe_head *sh)
187{
188 return sh->check_state || sh->reconstruct_state ||
189 test_bit(STRIPE_BIOFILL_RUN, &sh->state) ||
190 test_bit(STRIPE_COMPUTE_RUN, &sh->state);
191}
192
d1688a6d 193static void __release_stripe(struct r5conf *conf, struct stripe_head *sh)
1da177e4
LT		 194{
195 if (atomic_dec_and_test(&sh->count)) {
78bafebd
ES		 196 BUG_ON(!list_empty(&sh->lru));
197 BUG_ON(atomic_read(&conf->active_stripes)==0);
1da177e4 198 if (test_bit(STRIPE_HANDLE, &sh->state)) {
fab363b5
SL		 199 if (test_bit(STRIPE_DELAYED, &sh->state) &&
200 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
1da177e4 201 list_add_tail(&sh->lru, &conf->delayed_list);
482c0834
N		 202 else if (test_bit(STRIPE_BIT_DELAY, &sh->state) &&
203 sh->bm_seq - conf->seq_write > 0)
72626685 204 list_add_tail(&sh->lru, &conf->bitmap_list);
482c0834 205 else {
fab363b5 206 clear_bit(STRIPE_DELAYED, &sh->state);
72626685 207 clear_bit(STRIPE_BIT_DELAY, &sh->state);
1da177e4 208 list_add_tail(&sh->lru, &conf->handle_list);
72626685 209 }
1da177e4
LT		 210 md_wakeup_thread(conf->mddev->thread);
211 } else {
600aa109 212 BUG_ON(stripe_operations_active(sh));
41fe75f6 213 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
214 if (atomic_dec_return(&conf->preread_active_stripes)
215 < IO_THRESHOLD)
1da177e4 216 md_wakeup_thread(conf->mddev->thread);
1da177e4 217 atomic_dec(&conf->active_stripes);
ccfcc3c1
N		 218 if (!test_bit(STRIPE_EXPANDING, &sh->state)) {
219 list_add_tail(&sh->lru, &conf->inactive_list);
1da177e4 220 wake_up(&conf->wait_for_stripe);
46031f9a
RBJ		 221 if (conf->retry_read_aligned)
222 md_wakeup_thread(conf->mddev->thread);
ccfcc3c1 223 }
1da177e4
LT		 224 }
225 }
226}
d0dabf7e 227
1da177e4
LT		 228static void release_stripe(struct stripe_head *sh)
229{
d1688a6d 230 struct r5conf *conf = sh->raid_conf;
1da177e4 231 unsigned long flags;
16a53ecc 232
1da177e4
LT		 233 spin_lock_irqsave(&conf->device_lock, flags);
234 __release_stripe(conf, sh);
235 spin_unlock_irqrestore(&conf->device_lock, flags);
236}
237
fccddba0 238static inline void remove_hash(struct stripe_head *sh)
1da177e4 239{
45b4233c
DW		 240 pr_debug("remove_hash(), stripe %llu\n",
241 (unsigned long long)sh->sector);
1da177e4 242
fccddba0 243 hlist_del_init(&sh->hash);
1da177e4
LT		 244}
245
d1688a6d 246static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh)
1da177e4 247{
fccddba0 248 struct hlist_head *hp = stripe_hash(conf, sh->sector);
1da177e4 249
45b4233c
DW		 250 pr_debug("insert_hash(), stripe %llu\n",
251 (unsigned long long)sh->sector);
1da177e4 252
fccddba0 253 hlist_add_head(&sh->hash, hp);
1da177e4
LT		 254}
255
256
257/* find an idle stripe, make sure it is unhashed, and return it. */
d1688a6d 258static struct stripe_head *get_free_stripe(struct r5conf *conf)
1da177e4
LT		 259{
260 struct stripe_head *sh = NULL;
261 struct list_head *first;
262
1da177e4
LT		 263 if (list_empty(&conf->inactive_list))
264 goto out;
265 first = conf->inactive_list.next;
266 sh = list_entry(first, struct stripe_head, lru);
267 list_del_init(first);
268 remove_hash(sh);
269 atomic_inc(&conf->active_stripes);
270out:
271 return sh;
272}
273
e4e11e38 274static void shrink_buffers(struct stripe_head *sh)
1da177e4
LT		 275{
276 struct page *p;
277 int i;
e4e11e38 278 int num = sh->raid_conf->pool_size;
1da177e4 279
e4e11e38 280 for (i = 0; i < num ; i++) {
1da177e4
LT		 281 p = sh->dev[i].page;
282 if (!p)
283 continue;
284 sh->dev[i].page = NULL;
2d1f3b5d 285 put_page(p);
1da177e4
LT		 286 }
287}
288
e4e11e38 289static int grow_buffers(struct stripe_head *sh)
1da177e4
LT		 290{
291 int i;
e4e11e38 292 int num = sh->raid_conf->pool_size;
1da177e4 293
e4e11e38 294 for (i = 0; i < num; i++) {
1da177e4
LT		 295 struct page *page;
296
297 if (!(page = alloc_page(GFP_KERNEL))) {
298 return 1;
299 }
300 sh->dev[i].page = page;
301 }
302 return 0;
303}
304
784052ec 305static void raid5_build_block(struct stripe_head *sh, int i, int previous);
d1688a6d 306static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
911d4ee8 307 struct stripe_head *sh);
1da177e4 308
b5663ba4 309static void init_stripe(struct stripe_head *sh, sector_t sector, int previous)
1da177e4 310{
d1688a6d 311 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 312 int i;
1da177e4 313
78bafebd
ES		 314 BUG_ON(atomic_read(&sh->count) != 0);
315 BUG_ON(test_bit(STRIPE_HANDLE, &sh->state));
600aa109 316 BUG_ON(stripe_operations_active(sh));
d84e0f10 317
45b4233c 318 pr_debug("init_stripe called, stripe %llu\n",
1da177e4
LT		 319 (unsigned long long)sh->sector);
320
321 remove_hash(sh);
16a53ecc 322
86b42c71 323 sh->generation = conf->generation - previous;
b5663ba4 324 sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks;
1da177e4 325 sh->sector = sector;
911d4ee8 326 stripe_set_idx(sector, conf, previous, sh);
1da177e4
LT		 327 sh->state = 0;
328
7ecaa1e6
N		 329
330 for (i = sh->disks; i--; ) {
1da177e4
LT		 331 struct r5dev *dev = &sh->dev[i];
332
d84e0f10 333 if (dev->toread || dev->read || dev->towrite || dev->written ||
1da177e4 334 test_bit(R5_LOCKED, &dev->flags)) {
d84e0f10 335 printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n",
1da177e4 336 (unsigned long long)sh->sector, i, dev->toread,
d84e0f10 337 dev->read, dev->towrite, dev->written,
1da177e4 338 test_bit(R5_LOCKED, &dev->flags));
8cfa7b0f 339 WARN_ON(1);
1da177e4
LT		 340 }
341 dev->flags = 0;
784052ec 342 raid5_build_block(sh, i, previous);
1da177e4
LT		 343 }
344 insert_hash(conf, sh);
345}
346
d1688a6d 347static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector,
86b42c71 348 short generation)
1da177e4
LT		 349{
350 struct stripe_head *sh;
fccddba0 351 struct hlist_node *hn;
1da177e4 352
45b4233c 353 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
fccddba0 354 hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash)
86b42c71 355 if (sh->sector == sector && sh->generation == generation)
1da177e4 356 return sh;
45b4233c 357 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
1da177e4
LT		 358 return NULL;
359}
360
674806d6
N		 361/*
362 * Need to check if array has failed when deciding whether to:
363 * - start an array
364 * - remove non-faulty devices
365 * - add a spare
366 * - allow a reshape
367 * This determination is simple when no reshape is happening.
368 * However if there is a reshape, we need to carefully check
369 * both the before and after sections.
370 * This is because some failed devices may only affect one
371 * of the two sections, and some non-in_sync devices may
372 * be insync in the section most affected by failed devices.
373 */
908f4fbd 374static int calc_degraded(struct r5conf *conf)
674806d6 375{
908f4fbd 376 int degraded, degraded2;
674806d6 377 int i;
674806d6
N		 378
379 rcu_read_lock();
380 degraded = 0;
381 for (i = 0; i < conf->previous_raid_disks; i++) {
3cb03002 382 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
674806d6
N		 383 if (!rdev || test_bit(Faulty, &rdev->flags))
384 degraded++;
385 else if (test_bit(In_sync, &rdev->flags))
386 ;
387 else
388 /* not in-sync or faulty.
389 * If the reshape increases the number of devices,
390 * this is being recovered by the reshape, so
391 * this 'previous' section is not in_sync.
392 * If the number of devices is being reduced however,
393 * the device can only be part of the array if
394 * we are reverting a reshape, so this section will
395 * be in-sync.
396 */
397 if (conf->raid_disks >= conf->previous_raid_disks)
398 degraded++;
399 }
400 rcu_read_unlock();
908f4fbd
N		 401 if (conf->raid_disks == conf->previous_raid_disks)
402 return degraded;
674806d6 403 rcu_read_lock();
908f4fbd 404 degraded2 = 0;
674806d6 405 for (i = 0; i < conf->raid_disks; i++) {
3cb03002 406 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
674806d6 407 if (!rdev || test_bit(Faulty, &rdev->flags))
908f4fbd 408 degraded2++;
674806d6
N		 409 else if (test_bit(In_sync, &rdev->flags))
410 ;
411 else
412 /* not in-sync or faulty.
413 * If reshape increases the number of devices, this
414 * section has already been recovered, else it
415 * almost certainly hasn't.
416 */
417 if (conf->raid_disks <= conf->previous_raid_disks)
908f4fbd 418 degraded2++;
674806d6
N		 419 }
420 rcu_read_unlock();
908f4fbd
N		 421 if (degraded2 > degraded)
422 return degraded2;
423 return degraded;
424}
425
426static int has_failed(struct r5conf *conf)
427{
428 int degraded;
429
430 if (conf->mddev->reshape_position == MaxSector)
431 return conf->mddev->degraded > conf->max_degraded;
432
433 degraded = calc_degraded(conf);
674806d6
N		 434 if (degraded > conf->max_degraded)
435 return 1;
436 return 0;
437}
438
b5663ba4 439static struct stripe_head *
d1688a6d 440get_active_stripe(struct r5conf *conf, sector_t sector,
a8c906ca 441 int previous, int noblock, int noquiesce)
1da177e4
LT		 442{
443 struct stripe_head *sh;
444
45b4233c 445 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector);
1da177e4
LT		 446
447 spin_lock_irq(&conf->device_lock);
448
449 do {
72626685 450 wait_event_lock_irq(conf->wait_for_stripe,
a8c906ca 451 conf->quiesce == 0 || noquiesce,
72626685 452 conf->device_lock, /* nothing */);
86b42c71 453 sh = __find_stripe(conf, sector, conf->generation - previous);
1da177e4
LT		 454 if (!sh) {
455 if (!conf->inactive_blocked)
456 sh = get_free_stripe(conf);
457 if (noblock && sh == NULL)
458 break;
459 if (!sh) {
460 conf->inactive_blocked = 1;
461 wait_event_lock_irq(conf->wait_for_stripe,
462 !list_empty(&conf->inactive_list) &&
5036805b
N		 463 (atomic_read(&conf->active_stripes)
464 < (conf->max_nr_stripes *3/4)
1da177e4
LT		 465 || !conf->inactive_blocked),
466 conf->device_lock,
7c13edc8 467 );
1da177e4
LT		 468 conf->inactive_blocked = 0;
469 } else
b5663ba4 470 init_stripe(sh, sector, previous);
1da177e4
LT		 471 } else {
472 if (atomic_read(&sh->count)) {
ab69ae12
N		 473 BUG_ON(!list_empty(&sh->lru)
474 && !test_bit(STRIPE_EXPANDING, &sh->state));
1da177e4
LT		 475 } else {
476 if (!test_bit(STRIPE_HANDLE, &sh->state))
477 atomic_inc(&conf->active_stripes);
ff4e8d9a
N		 478 if (list_empty(&sh->lru) &&
479 !test_bit(STRIPE_EXPANDING, &sh->state))
16a53ecc
N		 480 BUG();
481 list_del_init(&sh->lru);
1da177e4
LT		 482 }
483 }
484 } while (sh == NULL);
485
486 if (sh)
487 atomic_inc(&sh->count);
488
489 spin_unlock_irq(&conf->device_lock);
490 return sh;
491}
492
05616be5
N		 493/* Determine if 'data_offset' or 'new_data_offset' should be used
494 * in this stripe_head.
495 */
496static int use_new_offset(struct r5conf *conf, struct stripe_head *sh)
497{
498 sector_t progress = conf->reshape_progress;
499 /* Need a memory barrier to make sure we see the value
500 * of conf->generation, or ->data_offset that was set before
501 * reshape_progress was updated.
502 */
503 smp_rmb();
504 if (progress == MaxSector)
505 return 0;
506 if (sh->generation == conf->generation - 1)
507 return 0;
508 /* We are in a reshape, and this is a new-generation stripe,
509 * so use new_data_offset.
510 */
511 return 1;
512}
513
6712ecf8
N		 514static void
515raid5_end_read_request(struct bio *bi, int error);
516static void
517raid5_end_write_request(struct bio *bi, int error);
91c00924 518
c4e5ac0a 519static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
91c00924 520{
d1688a6d 521 struct r5conf *conf = sh->raid_conf;
91c00924
DW		 522 int i, disks = sh->disks;
523
524 might_sleep();
525
526 for (i = disks; i--; ) {
527 int rw;
9a3e1101 528 int replace_only = 0;
977df362
N		 529 struct bio *bi, *rbi;
530 struct md_rdev *rdev, *rrdev = NULL;
e9c7469b
TH		 531 if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) {
532 if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags))
533 rw = WRITE_FUA;
534 else
535 rw = WRITE;
536 } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags))
91c00924 537 rw = READ;
9a3e1101
N		 538 else if (test_and_clear_bit(R5_WantReplace,
539 &sh->dev[i].flags)) {
540 rw = WRITE;
541 replace_only = 1;
542 } else
91c00924 543 continue;
bc0934f0
SL		 544 if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags))
545 rw |= REQ_SYNC;
91c00924
DW		 546
547 bi = &sh->dev[i].req;
977df362 548 rbi = &sh->dev[i].rreq; /* For writing to replacement */
91c00924
DW		 549
550 bi->bi_rw = rw;
977df362
N		 551 rbi->bi_rw = rw;
552 if (rw & WRITE) {
91c00924 553 bi->bi_end_io = raid5_end_write_request;
977df362
N		 554 rbi->bi_end_io = raid5_end_write_request;
555 } else
91c00924
DW		 556 bi->bi_end_io = raid5_end_read_request;
557
558 rcu_read_lock();
9a3e1101 559 rrdev = rcu_dereference(conf->disks[i].replacement);
dd054fce
N		 560 smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */
561 rdev = rcu_dereference(conf->disks[i].rdev);
562 if (!rdev) {
563 rdev = rrdev;
564 rrdev = NULL;
565 }
9a3e1101
N		 566 if (rw & WRITE) {
567 if (replace_only)
568 rdev = NULL;
dd054fce
N		 569 if (rdev == rrdev)
570 /* We raced and saw duplicates */
571 rrdev = NULL;
9a3e1101 572 } else {
dd054fce 573 if (test_bit(R5_ReadRepl, &sh->dev[i].flags) && rrdev)
9a3e1101
N		 574 rdev = rrdev;
575 rrdev = NULL;
576 }
977df362 577
91c00924
DW		 578 if (rdev && test_bit(Faulty, &rdev->flags))
579 rdev = NULL;
580 if (rdev)
581 atomic_inc(&rdev->nr_pending);
977df362
N		 582 if (rrdev && test_bit(Faulty, &rrdev->flags))
583 rrdev = NULL;
584 if (rrdev)
585 atomic_inc(&rrdev->nr_pending);
91c00924
DW		 586 rcu_read_unlock();
587
73e92e51 588 /* We have already checked bad blocks for reads. Now
977df362
N		 589 * need to check for writes. We never accept write errors
590 * on the replacement, so we don't to check rrdev.
73e92e51
N		 591 */
592 while ((rw & WRITE) && rdev &&
593 test_bit(WriteErrorSeen, &rdev->flags)) {
594 sector_t first_bad;
595 int bad_sectors;
596 int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
597 &first_bad, &bad_sectors);
598 if (!bad)
599 break;
600
601 if (bad < 0) {
602 set_bit(BlockedBadBlocks, &rdev->flags);
603 if (!conf->mddev->external &&
604 conf->mddev->flags) {
605 /* It is very unlikely, but we might
606 * still need to write out the
607 * bad block log - better give it
608 * a chance*/
609 md_check_recovery(conf->mddev);
610 }
1850753d 611 /*
612 * Because md_wait_for_blocked_rdev
613 * will dec nr_pending, we must
614 * increment it first.
615 */
616 atomic_inc(&rdev->nr_pending);
73e92e51
N		 617 md_wait_for_blocked_rdev(rdev, conf->mddev);
618 } else {
619 /* Acknowledged bad block - skip the write */
620 rdev_dec_pending(rdev, conf->mddev);
621 rdev = NULL;
622 }
623 }
624
91c00924 625 if (rdev) {
9a3e1101
N		 626 if (s->syncing || s->expanding || s->expanded
627 || s->replacing)
91c00924
DW		 628 md_sync_acct(rdev->bdev, STRIPE_SECTORS);
629
2b7497f0
DW		 630 set_bit(STRIPE_IO_STARTED, &sh->state);
631
91c00924
DW		 632 bi->bi_bdev = rdev->bdev;
633 pr_debug("%s: for %llu schedule op %ld on disc %d\n",
e46b272b 634 __func__, (unsigned long long)sh->sector,
91c00924
DW		 635 bi->bi_rw, i);
636 atomic_inc(&sh->count);
05616be5
N		 637 if (use_new_offset(conf, sh))
638 bi->bi_sector = (sh->sector
639 + rdev->new_data_offset);
640 else
641 bi->bi_sector = (sh->sector
642 + rdev->data_offset);
91c00924 643 bi->bi_flags = 1 << BIO_UPTODATE;
91c00924 644 bi->bi_idx = 0;
91c00924
DW		 645 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
646 bi->bi_io_vec[0].bv_offset = 0;
647 bi->bi_size = STRIPE_SIZE;
648 bi->bi_next = NULL;
977df362
N		 649 if (rrdev)
650 set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags);
91c00924 651 generic_make_request(bi);
977df362
N		 652 }
653 if (rrdev) {
9a3e1101
N		 654 if (s->syncing || s->expanding || s->expanded
655 || s->replacing)
977df362
N		 656 md_sync_acct(rrdev->bdev, STRIPE_SECTORS);
657
658 set_bit(STRIPE_IO_STARTED, &sh->state);
659
660 rbi->bi_bdev = rrdev->bdev;
661 pr_debug("%s: for %llu schedule op %ld on "
662 "replacement disc %d\n",
663 __func__, (unsigned long long)sh->sector,
664 rbi->bi_rw, i);
665 atomic_inc(&sh->count);
05616be5
N		 666 if (use_new_offset(conf, sh))
667 rbi->bi_sector = (sh->sector
668 + rrdev->new_data_offset);
669 else
670 rbi->bi_sector = (sh->sector
671 + rrdev->data_offset);
977df362
N		 672 rbi->bi_flags = 1 << BIO_UPTODATE;
673 rbi->bi_idx = 0;
674 rbi->bi_io_vec[0].bv_len = STRIPE_SIZE;
675 rbi->bi_io_vec[0].bv_offset = 0;
676 rbi->bi_size = STRIPE_SIZE;
677 rbi->bi_next = NULL;
678 generic_make_request(rbi);
679 }
680 if (!rdev && !rrdev) {
b062962e 681 if (rw & WRITE)
91c00924
DW		 682 set_bit(STRIPE_DEGRADED, &sh->state);
683 pr_debug("skip op %ld on disc %d for sector %llu\n",
684 bi->bi_rw, i, (unsigned long long)sh->sector);
685 clear_bit(R5_LOCKED, &sh->dev[i].flags);
686 set_bit(STRIPE_HANDLE, &sh->state);
687 }
688 }
689}
690
691static struct dma_async_tx_descriptor *
692async_copy_data(int frombio, struct bio *bio, struct page *page,
693 sector_t sector, struct dma_async_tx_descriptor *tx)
694{
695 struct bio_vec *bvl;
696 struct page *bio_page;
697 int i;
698 int page_offset;
a08abd8c 699 struct async_submit_ctl submit;
0403e382 700 enum async_tx_flags flags = 0;
91c00924
DW		 701
702 if (bio->bi_sector >= sector)
703 page_offset = (signed)(bio->bi_sector - sector) * 512;
704 else
705 page_offset = (signed)(sector - bio->bi_sector) * -512;
a08abd8c 706
0403e382
DW		 707 if (frombio)
708 flags |= ASYNC_TX_FENCE;
709 init_async_submit(&submit, flags, tx, NULL, NULL, NULL);
710
91c00924 711 bio_for_each_segment(bvl, bio, i) {
fcde9075 712 int len = bvl->bv_len;
91c00924
DW		 713 int clen;
714 int b_offset = 0;
715
716 if (page_offset < 0) {
717 b_offset = -page_offset;
718 page_offset += b_offset;
719 len -= b_offset;
720 }
721
722 if (len > 0 && page_offset + len > STRIPE_SIZE)
723 clen = STRIPE_SIZE - page_offset;
724 else
725 clen = len;
726
727 if (clen > 0) {
fcde9075
NK		 728 b_offset += bvl->bv_offset;
729 bio_page = bvl->bv_page;
91c00924
DW		 730 if (frombio)
731 tx = async_memcpy(page, bio_page, page_offset,
a08abd8c 732 b_offset, clen, &submit);
91c00924
DW		 733 else
734 tx = async_memcpy(bio_page, page, b_offset,
a08abd8c 735 page_offset, clen, &submit);
91c00924 736 }
a08abd8c
DW		 737 /* chain the operations */
738 submit.depend_tx = tx;
739
91c00924
DW		 740 if (clen < len) /* hit end of page */
741 break;
742 page_offset += len;
743 }
744
745 return tx;
746}
747
748static void ops_complete_biofill(void *stripe_head_ref)
749{
750 struct stripe_head *sh = stripe_head_ref;
751 struct bio *return_bi = NULL;
d1688a6d 752 struct r5conf *conf = sh->raid_conf;
e4d84909 753 int i;
91c00924 754
e46b272b 755 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 756 (unsigned long long)sh->sector);
757
758 /* clear completed biofills */
83de75cc 759 spin_lock_irq(&conf->device_lock);
91c00924
DW		 760 for (i = sh->disks; i--; ) {
761 struct r5dev *dev = &sh->dev[i];
91c00924
DW		 762
763 /* acknowledge completion of a biofill operation */
e4d84909
DW		 764 /* and check if we need to reply to a read request,
765 * new R5_Wantfill requests are held off until
83de75cc 766 * !STRIPE_BIOFILL_RUN
e4d84909
DW		 767 */
768 if (test_and_clear_bit(R5_Wantfill, &dev->flags)) {
91c00924 769 struct bio *rbi, *rbi2;
91c00924 770
91c00924
DW		 771 BUG_ON(!dev->read);
772 rbi = dev->read;
773 dev->read = NULL;
774 while (rbi && rbi->bi_sector <
775 dev->sector + STRIPE_SECTORS) {
776 rbi2 = r5_next_bio(rbi, dev->sector);
960e739d 777 if (!raid5_dec_bi_phys_segments(rbi)) {
91c00924
DW		 778 rbi->bi_next = return_bi;
779 return_bi = rbi;
780 }
91c00924
DW		 781 rbi = rbi2;
782 }
783 }
784 }
83de75cc
DW		 785 spin_unlock_irq(&conf->device_lock);
786 clear_bit(STRIPE_BIOFILL_RUN, &sh->state);
91c00924
DW		 787
788 return_io(return_bi);
789
e4d84909 790 set_bit(STRIPE_HANDLE, &sh->state);
91c00924
DW		 791 release_stripe(sh);
792}
793
794static void ops_run_biofill(struct stripe_head *sh)
795{
796 struct dma_async_tx_descriptor *tx = NULL;
d1688a6d 797 struct r5conf *conf = sh->raid_conf;
a08abd8c 798 struct async_submit_ctl submit;
91c00924
DW		 799 int i;
800
e46b272b 801 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 802 (unsigned long long)sh->sector);
803
804 for (i = sh->disks; i--; ) {
805 struct r5dev *dev = &sh->dev[i];
806 if (test_bit(R5_Wantfill, &dev->flags)) {
807 struct bio *rbi;
808 spin_lock_irq(&conf->device_lock);
809 dev->read = rbi = dev->toread;
810 dev->toread = NULL;
811 spin_unlock_irq(&conf->device_lock);
812 while (rbi && rbi->bi_sector <
813 dev->sector + STRIPE_SECTORS) {
814 tx = async_copy_data(0, rbi, dev->page,
815 dev->sector, tx);
816 rbi = r5_next_bio(rbi, dev->sector);
817 }
818 }
819 }
820
821 atomic_inc(&sh->count);
a08abd8c
DW		 822 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL);
823 async_trigger_callback(&submit);
91c00924
DW		 824}
825
4e7d2c0a 826static void mark_target_uptodate(struct stripe_head *sh, int target)
91c00924 827{
4e7d2c0a 828 struct r5dev *tgt;
91c00924 829
4e7d2c0a
DW		 830 if (target < 0)
831 return;
91c00924 832
4e7d2c0a 833 tgt = &sh->dev[target];
91c00924
DW		 834 set_bit(R5_UPTODATE, &tgt->flags);
835 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
836 clear_bit(R5_Wantcompute, &tgt->flags);
4e7d2c0a
DW		 837}
838
ac6b53b6 839static void ops_complete_compute(void *stripe_head_ref)
91c00924
DW		 840{
841 struct stripe_head *sh = stripe_head_ref;
91c00924 842
e46b272b 843 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 844 (unsigned long long)sh->sector);
845
ac6b53b6 846 /* mark the computed target(s) as uptodate */
4e7d2c0a 847 mark_target_uptodate(sh, sh->ops.target);
ac6b53b6 848 mark_target_uptodate(sh, sh->ops.target2);
4e7d2c0a 849
ecc65c9b
DW		 850 clear_bit(STRIPE_COMPUTE_RUN, &sh->state);
851 if (sh->check_state == check_state_compute_run)
852 sh->check_state = check_state_compute_result;
91c00924
DW		 853 set_bit(STRIPE_HANDLE, &sh->state);
854 release_stripe(sh);
855}
856
d6f38f31
DW		 857/* return a pointer to the address conversion region of the scribble buffer */
858static addr_conv_t *to_addr_conv(struct stripe_head *sh,
859 struct raid5_percpu *percpu)
860{
861 return percpu->scribble + sizeof(struct page *) * (sh->disks + 2);
862}
863
864static struct dma_async_tx_descriptor *
865ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 866{
91c00924 867 int disks = sh->disks;
d6f38f31 868 struct page **xor_srcs = percpu->scribble;
91c00924
DW		 869 int target = sh->ops.target;
870 struct r5dev *tgt = &sh->dev[target];
871 struct page *xor_dest = tgt->page;
872 int count = 0;
873 struct dma_async_tx_descriptor *tx;
a08abd8c 874 struct async_submit_ctl submit;
91c00924
DW		 875 int i;
876
877 pr_debug("%s: stripe %llu block: %d\n",
e46b272b 878 __func__, (unsigned long long)sh->sector, target);
91c00924
DW		 879 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
880
881 for (i = disks; i--; )
882 if (i != target)
883 xor_srcs[count++] = sh->dev[i].page;
884
885 atomic_inc(&sh->count);
886
0403e382 887 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL,
ac6b53b6 888 ops_complete_compute, sh, to_addr_conv(sh, percpu));
91c00924 889 if (unlikely(count == 1))
a08abd8c 890 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
91c00924 891 else
a08abd8c 892 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924 893
91c00924
DW		 894 return tx;
895}
896
ac6b53b6
DW		 897/* set_syndrome_sources - populate source buffers for gen_syndrome
898 * @srcs - (struct page *) array of size sh->disks
899 * @sh - stripe_head to parse
900 *
901 * Populates srcs in proper layout order for the stripe and returns the
902 * 'count' of sources to be used in a call to async_gen_syndrome. The P
903 * destination buffer is recorded in srcs[count] and the Q destination
904 * is recorded in srcs[count+1]].
905 */
906static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh)
907{
908 int disks = sh->disks;
909 int syndrome_disks = sh->ddf_layout ? disks : (disks - 2);
910 int d0_idx = raid6_d0(sh);
911 int count;
912 int i;
913
914 for (i = 0; i < disks; i++)
5dd33c9a 915 srcs[i] = NULL;
ac6b53b6
DW		 916
917 count = 0;
918 i = d0_idx;
919 do {
920 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
921
922 srcs[slot] = sh->dev[i].page;
923 i = raid6_next_disk(i, disks);
924 } while (i != d0_idx);
ac6b53b6 925
e4424fee 926 return syndrome_disks;
ac6b53b6
DW		 927}
928
929static struct dma_async_tx_descriptor *
930ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu)
931{
932 int disks = sh->disks;
933 struct page **blocks = percpu->scribble;
934 int target;
935 int qd_idx = sh->qd_idx;
936 struct dma_async_tx_descriptor *tx;
937 struct async_submit_ctl submit;
938 struct r5dev *tgt;
939 struct page *dest;
940 int i;
941 int count;
942
943 if (sh->ops.target < 0)
944 target = sh->ops.target2;
945 else if (sh->ops.target2 < 0)
946 target = sh->ops.target;
91c00924 947 else
ac6b53b6
DW		 948 /* we should only have one valid target */
949 BUG();
950 BUG_ON(target < 0);
951 pr_debug("%s: stripe %llu block: %d\n",
952 __func__, (unsigned long long)sh->sector, target);
953
954 tgt = &sh->dev[target];
955 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
956 dest = tgt->page;
957
958 atomic_inc(&sh->count);
959
960 if (target == qd_idx) {
961 count = set_syndrome_sources(blocks, sh);
962 blocks[count] = NULL; /* regenerating p is not necessary */
963 BUG_ON(blocks[count+1] != dest); /* q should already be set */
0403e382
DW		 964 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
965 ops_complete_compute, sh,
ac6b53b6
DW		 966 to_addr_conv(sh, percpu));
967 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
968 } else {
969 /* Compute any data- or p-drive using XOR */
970 count = 0;
971 for (i = disks; i-- ; ) {
972 if (i == target || i == qd_idx)
973 continue;
974 blocks[count++] = sh->dev[i].page;
975 }
976
0403e382
DW		 977 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
978 NULL, ops_complete_compute, sh,
ac6b53b6
DW		 979 to_addr_conv(sh, percpu));
980 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit);
981 }
91c00924 982
91c00924
DW		 983 return tx;
984}
985
ac6b53b6
DW		 986static struct dma_async_tx_descriptor *
987ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu)
988{
989 int i, count, disks = sh->disks;
990 int syndrome_disks = sh->ddf_layout ? disks : disks-2;
991 int d0_idx = raid6_d0(sh);
992 int faila = -1, failb = -1;
993 int target = sh->ops.target;
994 int target2 = sh->ops.target2;
995 struct r5dev *tgt = &sh->dev[target];
996 struct r5dev *tgt2 = &sh->dev[target2];
997 struct dma_async_tx_descriptor *tx;
998 struct page **blocks = percpu->scribble;
999 struct async_submit_ctl submit;
1000
1001 pr_debug("%s: stripe %llu block1: %d block2: %d\n",
1002 __func__, (unsigned long long)sh->sector, target, target2);
1003 BUG_ON(target < 0 || target2 < 0);
1004 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1005 BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags));
1006
6c910a78 1007 /* we need to open-code set_syndrome_sources to handle the
ac6b53b6
DW		 1008 * slot number conversion for 'faila' and 'failb'
1009 */
1010 for (i = 0; i < disks ; i++)
5dd33c9a 1011 blocks[i] = NULL;
ac6b53b6
DW		 1012 count = 0;
1013 i = d0_idx;
1014 do {
1015 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
1016
1017 blocks[slot] = sh->dev[i].page;
1018
1019 if (i == target)
1020 faila = slot;
1021 if (i == target2)
1022 failb = slot;
1023 i = raid6_next_disk(i, disks);
1024 } while (i != d0_idx);
ac6b53b6
DW		 1025
1026 BUG_ON(faila == failb);
1027 if (failb < faila)
1028 swap(faila, failb);
1029 pr_debug("%s: stripe: %llu faila: %d failb: %d\n",
1030 __func__, (unsigned long long)sh->sector, faila, failb);
1031
1032 atomic_inc(&sh->count);
1033
1034 if (failb == syndrome_disks+1) {
1035 /* Q disk is one of the missing disks */
1036 if (faila == syndrome_disks) {
1037 /* Missing P+Q, just recompute */
0403e382
DW		 1038 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1039 ops_complete_compute, sh,
1040 to_addr_conv(sh, percpu));
e4424fee 1041 return async_gen_syndrome(blocks, 0, syndrome_disks+2,
ac6b53b6
DW		 1042 STRIPE_SIZE, &submit);
1043 } else {
1044 struct page *dest;
1045 int data_target;
1046 int qd_idx = sh->qd_idx;
1047
1048 /* Missing D+Q: recompute D from P, then recompute Q */
1049 if (target == qd_idx)
1050 data_target = target2;
1051 else
1052 data_target = target;
1053
1054 count = 0;
1055 for (i = disks; i-- ; ) {
1056 if (i == data_target || i == qd_idx)
1057 continue;
1058 blocks[count++] = sh->dev[i].page;
1059 }
1060 dest = sh->dev[data_target].page;
0403e382
DW		 1061 init_async_submit(&submit,
1062 ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1063 NULL, NULL, NULL,
1064 to_addr_conv(sh, percpu));
ac6b53b6
DW		 1065 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE,
1066 &submit);
1067
1068 count = set_syndrome_sources(blocks, sh);
0403e382
DW		 1069 init_async_submit(&submit, ASYNC_TX_FENCE, tx,
1070 ops_complete_compute, sh,
1071 to_addr_conv(sh, percpu));
ac6b53b6
DW		 1072 return async_gen_syndrome(blocks, 0, count+2,
1073 STRIPE_SIZE, &submit);
1074 }
ac6b53b6 1075 } else {
6c910a78
DW		 1076 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1077 ops_complete_compute, sh,
1078 to_addr_conv(sh, percpu));
1079 if (failb == syndrome_disks) {
1080 /* We're missing D+P. */
1081 return async_raid6_datap_recov(syndrome_disks+2,
1082 STRIPE_SIZE, faila,
1083 blocks, &submit);
1084 } else {
1085 /* We're missing D+D. */
1086 return async_raid6_2data_recov(syndrome_disks+2,
1087 STRIPE_SIZE, faila, failb,
1088 blocks, &submit);
1089 }
ac6b53b6
DW		 1090 }
1091}
1092
1093
91c00924
DW		 1094static void ops_complete_prexor(void *stripe_head_ref)
1095{
1096 struct stripe_head *sh = stripe_head_ref;
1097
e46b272b 1098 pr_debug("%s: stripe %llu\n", __func__,
91c00924 1099 (unsigned long long)sh->sector);
91c00924
DW		 1100}
1101
1102static struct dma_async_tx_descriptor *
d6f38f31
DW		 1103ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu,
1104 struct dma_async_tx_descriptor *tx)
91c00924 1105{
91c00924 1106 int disks = sh->disks;
d6f38f31 1107 struct page **xor_srcs = percpu->scribble;
91c00924 1108 int count = 0, pd_idx = sh->pd_idx, i;
a08abd8c 1109 struct async_submit_ctl submit;
91c00924
DW		 1110
1111 /* existing parity data subtracted */
1112 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1113
e46b272b 1114 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1115 (unsigned long long)sh->sector);
1116
1117 for (i = disks; i--; ) {
1118 struct r5dev *dev = &sh->dev[i];
1119 /* Only process blocks that are known to be uptodate */
d8ee0728 1120 if (test_bit(R5_Wantdrain, &dev->flags))
91c00924
DW		 1121 xor_srcs[count++] = dev->page;
1122 }
1123
0403e382 1124 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
d6f38f31 1125 ops_complete_prexor, sh, to_addr_conv(sh, percpu));
a08abd8c 1126 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924
DW		 1127
1128 return tx;
1129}
1130
1131static struct dma_async_tx_descriptor *
d8ee0728 1132ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
91c00924
DW		 1133{
1134 int disks = sh->disks;
d8ee0728 1135 int i;
91c00924 1136
e46b272b 1137 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1138 (unsigned long long)sh->sector);
1139
1140 for (i = disks; i--; ) {
1141 struct r5dev *dev = &sh->dev[i];
1142 struct bio *chosen;
91c00924 1143
d8ee0728 1144 if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) {
91c00924
DW		 1145 struct bio *wbi;
1146
cbe47ec5 1147 spin_lock_irq(&sh->raid_conf->device_lock);
91c00924
DW		 1148 chosen = dev->towrite;
1149 dev->towrite = NULL;
1150 BUG_ON(dev->written);
1151 wbi = dev->written = chosen;
cbe47ec5 1152 spin_unlock_irq(&sh->raid_conf->device_lock);
91c00924
DW		 1153
1154 while (wbi && wbi->bi_sector <
1155 dev->sector + STRIPE_SECTORS) {
e9c7469b
TH		 1156 if (wbi->bi_rw & REQ_FUA)
1157 set_bit(R5_WantFUA, &dev->flags);
bc0934f0
SL		 1158 if (wbi->bi_rw & REQ_SYNC)
1159 set_bit(R5_SyncIO, &dev->flags);
91c00924
DW		 1160 tx = async_copy_data(1, wbi, dev->page,
1161 dev->sector, tx);
1162 wbi = r5_next_bio(wbi, dev->sector);
1163 }
1164 }
1165 }
1166
1167 return tx;
1168}
1169
ac6b53b6 1170static void ops_complete_reconstruct(void *stripe_head_ref)
91c00924
DW		 1171{
1172 struct stripe_head *sh = stripe_head_ref;
ac6b53b6
DW		 1173 int disks = sh->disks;
1174 int pd_idx = sh->pd_idx;
1175 int qd_idx = sh->qd_idx;
1176 int i;
bc0934f0 1177 bool fua = false, sync = false;
91c00924 1178
e46b272b 1179 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1180 (unsigned long long)sh->sector);
1181
bc0934f0 1182 for (i = disks; i--; ) {
e9c7469b 1183 fua |= test_bit(R5_WantFUA, &sh->dev[i].flags);
bc0934f0
SL		 1184 sync |= test_bit(R5_SyncIO, &sh->dev[i].flags);
1185 }
e9c7469b 1186
91c00924
DW		 1187 for (i = disks; i--; ) {
1188 struct r5dev *dev = &sh->dev[i];
ac6b53b6 1189
e9c7469b 1190 if (dev->written || i == pd_idx || i == qd_idx) {
91c00924 1191 set_bit(R5_UPTODATE, &dev->flags);
e9c7469b
TH		 1192 if (fua)
1193 set_bit(R5_WantFUA, &dev->flags);
bc0934f0
SL		 1194 if (sync)
1195 set_bit(R5_SyncIO, &dev->flags);
e9c7469b 1196 }
91c00924
DW		 1197 }
1198
d8ee0728
DW		 1199 if (sh->reconstruct_state == reconstruct_state_drain_run)
1200 sh->reconstruct_state = reconstruct_state_drain_result;
1201 else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run)
1202 sh->reconstruct_state = reconstruct_state_prexor_drain_result;
1203 else {
1204 BUG_ON(sh->reconstruct_state != reconstruct_state_run);
1205 sh->reconstruct_state = reconstruct_state_result;
1206 }
91c00924
DW		 1207
1208 set_bit(STRIPE_HANDLE, &sh->state);
1209 release_stripe(sh);
1210}
1211
1212static void
ac6b53b6
DW		 1213ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu,
1214 struct dma_async_tx_descriptor *tx)
91c00924 1215{
91c00924 1216 int disks = sh->disks;
d6f38f31 1217 struct page **xor_srcs = percpu->scribble;
a08abd8c 1218 struct async_submit_ctl submit;
91c00924
DW		 1219 int count = 0, pd_idx = sh->pd_idx, i;
1220 struct page *xor_dest;
d8ee0728 1221 int prexor = 0;
91c00924 1222 unsigned long flags;
91c00924 1223
e46b272b 1224 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1225 (unsigned long long)sh->sector);
1226
1227 /* check if prexor is active which means only process blocks
1228 * that are part of a read-modify-write (written)
1229 */
d8ee0728
DW		 1230 if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
1231 prexor = 1;
91c00924
DW		 1232 xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1233 for (i = disks; i--; ) {
1234 struct r5dev *dev = &sh->dev[i];
1235 if (dev->written)
1236 xor_srcs[count++] = dev->page;
1237 }
1238 } else {
1239 xor_dest = sh->dev[pd_idx].page;
1240 for (i = disks; i--; ) {
1241 struct r5dev *dev = &sh->dev[i];
1242 if (i != pd_idx)
1243 xor_srcs[count++] = dev->page;
1244 }
1245 }
1246
91c00924
DW		 1247 /* 1/ if we prexor'd then the dest is reused as a source
1248 * 2/ if we did not prexor then we are redoing the parity
1249 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
1250 * for the synchronous xor case
1251 */
88ba2aa5 1252 flags = ASYNC_TX_ACK |
91c00924
DW		 1253 (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
1254
1255 atomic_inc(&sh->count);
1256
ac6b53b6 1257 init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh,
d6f38f31 1258 to_addr_conv(sh, percpu));
a08abd8c
DW		 1259 if (unlikely(count == 1))
1260 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
1261 else
1262 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924
DW		 1263}
1264
ac6b53b6
DW		 1265static void
1266ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu,
1267 struct dma_async_tx_descriptor *tx)
1268{
1269 struct async_submit_ctl submit;
1270 struct page **blocks = percpu->scribble;
1271 int count;
1272
1273 pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector);
1274
1275 count = set_syndrome_sources(blocks, sh);
1276
1277 atomic_inc(&sh->count);
1278
1279 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct,
1280 sh, to_addr_conv(sh, percpu));
1281 async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
91c00924
DW		 1282}
1283
1284static void ops_complete_check(void *stripe_head_ref)
1285{
1286 struct stripe_head *sh = stripe_head_ref;
91c00924 1287
e46b272b 1288 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1289 (unsigned long long)sh->sector);
1290
ecc65c9b 1291 sh->check_state = check_state_check_result;
91c00924
DW		 1292 set_bit(STRIPE_HANDLE, &sh->state);
1293 release_stripe(sh);
1294}
1295
ac6b53b6 1296static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 1297{
91c00924 1298 int disks = sh->disks;
ac6b53b6
DW		 1299 int pd_idx = sh->pd_idx;
1300 int qd_idx = sh->qd_idx;
1301 struct page *xor_dest;
d6f38f31 1302 struct page **xor_srcs = percpu->scribble;
91c00924 1303 struct dma_async_tx_descriptor *tx;
a08abd8c 1304 struct async_submit_ctl submit;
ac6b53b6
DW		 1305 int count;
1306 int i;
91c00924 1307
e46b272b 1308 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1309 (unsigned long long)sh->sector);
1310
ac6b53b6
DW		 1311 count = 0;
1312 xor_dest = sh->dev[pd_idx].page;
1313 xor_srcs[count++] = xor_dest;
91c00924 1314 for (i = disks; i--; ) {
ac6b53b6
DW		 1315 if (i == pd_idx || i == qd_idx)
1316 continue;
1317 xor_srcs[count++] = sh->dev[i].page;
91c00924
DW		 1318 }
1319
d6f38f31
DW		 1320 init_async_submit(&submit, 0, NULL, NULL, NULL,
1321 to_addr_conv(sh, percpu));
099f53cb 1322 tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
a08abd8c 1323 &sh->ops.zero_sum_result, &submit);
91c00924 1324
91c00924 1325 atomic_inc(&sh->count);
a08abd8c
DW		 1326 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL);
1327 tx = async_trigger_callback(&submit);
91c00924
DW		 1328}
1329
ac6b53b6
DW		 1330static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp)
1331{
1332 struct page **srcs = percpu->scribble;
1333 struct async_submit_ctl submit;
1334 int count;
1335
1336 pr_debug("%s: stripe %llu checkp: %d\n", __func__,
1337 (unsigned long long)sh->sector, checkp);
1338
1339 count = set_syndrome_sources(srcs, sh);
1340 if (!checkp)
1341 srcs[count] = NULL;
91c00924 1342
91c00924 1343 atomic_inc(&sh->count);
ac6b53b6
DW		 1344 init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check,
1345 sh, to_addr_conv(sh, percpu));
1346 async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE,
1347 &sh->ops.zero_sum_result, percpu->spare_page, &submit);
91c00924
DW		 1348}
1349
417b8d4a 1350static void __raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
91c00924
DW		 1351{
1352 int overlap_clear = 0, i, disks = sh->disks;
1353 struct dma_async_tx_descriptor *tx = NULL;
d1688a6d 1354 struct r5conf *conf = sh->raid_conf;
ac6b53b6 1355 int level = conf->level;
d6f38f31
DW		 1356 struct raid5_percpu *percpu;
1357 unsigned long cpu;
91c00924 1358
d6f38f31
DW		 1359 cpu = get_cpu();
1360 percpu = per_cpu_ptr(conf->percpu, cpu);
83de75cc 1361 if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) {
91c00924
DW		 1362 ops_run_biofill(sh);
1363 overlap_clear++;
1364 }
1365
7b3a871e 1366 if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) {
ac6b53b6
DW		 1367 if (level < 6)
1368 tx = ops_run_compute5(sh, percpu);
1369 else {
1370 if (sh->ops.target2 < 0 || sh->ops.target < 0)
1371 tx = ops_run_compute6_1(sh, percpu);
1372 else
1373 tx = ops_run_compute6_2(sh, percpu);
1374 }
1375 /* terminate the chain if reconstruct is not set to be run */
1376 if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request))
7b3a871e
DW		 1377 async_tx_ack(tx);
1378 }
91c00924 1379
600aa109 1380 if (test_bit(STRIPE_OP_PREXOR, &ops_request))
d6f38f31 1381 tx = ops_run_prexor(sh, percpu, tx);
91c00924 1382
600aa109 1383 if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) {
d8ee0728 1384 tx = ops_run_biodrain(sh, tx);
91c00924
DW		 1385 overlap_clear++;
1386 }
1387
ac6b53b6
DW		 1388 if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) {
1389 if (level < 6)
1390 ops_run_reconstruct5(sh, percpu, tx);
1391 else
1392 ops_run_reconstruct6(sh, percpu, tx);
1393 }
91c00924 1394
ac6b53b6
DW		 1395 if (test_bit(STRIPE_OP_CHECK, &ops_request)) {
1396 if (sh->check_state == check_state_run)
1397 ops_run_check_p(sh, percpu);
1398 else if (sh->check_state == check_state_run_q)
1399 ops_run_check_pq(sh, percpu, 0);
1400 else if (sh->check_state == check_state_run_pq)
1401 ops_run_check_pq(sh, percpu, 1);
1402 else
1403 BUG();
1404 }
91c00924 1405
91c00924
DW		 1406 if (overlap_clear)
1407 for (i = disks; i--; ) {
1408 struct r5dev *dev = &sh->dev[i];
1409 if (test_and_clear_bit(R5_Overlap, &dev->flags))
1410 wake_up(&sh->raid_conf->wait_for_overlap);
1411 }
d6f38f31 1412 put_cpu();
91c00924
DW		 1413}
1414
417b8d4a
DW		 1415#ifdef CONFIG_MULTICORE_RAID456
1416static void async_run_ops(void *param, async_cookie_t cookie)
1417{
1418 struct stripe_head *sh = param;
1419 unsigned long ops_request = sh->ops.request;
1420
1421 clear_bit_unlock(STRIPE_OPS_REQ_PENDING, &sh->state);
1422 wake_up(&sh->ops.wait_for_ops);
1423
1424 __raid_run_ops(sh, ops_request);
1425 release_stripe(sh);
1426}
1427
1428static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
1429{
1430 /* since handle_stripe can be called outside of raid5d context
1431 * we need to ensure sh->ops.request is de-staged before another
1432 * request arrives
1433 */
1434 wait_event(sh->ops.wait_for_ops,
1435 !test_and_set_bit_lock(STRIPE_OPS_REQ_PENDING, &sh->state));
1436 sh->ops.request = ops_request;
1437
1438 atomic_inc(&sh->count);
1439 async_schedule(async_run_ops, sh);
1440}
1441#else
1442#define raid_run_ops __raid_run_ops
1443#endif
1444
d1688a6d 1445static int grow_one_stripe(struct r5conf *conf)
1da177e4
LT		 1446{
1447 struct stripe_head *sh;
6ce32846 1448 sh = kmem_cache_zalloc(conf->slab_cache, GFP_KERNEL);
3f294f4f
N		 1449 if (!sh)
1450 return 0;
6ce32846 1451
3f294f4f 1452 sh->raid_conf = conf;
417b8d4a
DW		 1453 #ifdef CONFIG_MULTICORE_RAID456
1454 init_waitqueue_head(&sh->ops.wait_for_ops);
1455 #endif
3f294f4f 1456
e4e11e38
N		 1457 if (grow_buffers(sh)) {
1458 shrink_buffers(sh);
3f294f4f
N		 1459 kmem_cache_free(conf->slab_cache, sh);
1460 return 0;
1461 }
1462 /* we just created an active stripe so... */
1463 atomic_set(&sh->count, 1);
1464 atomic_inc(&conf->active_stripes);
1465 INIT_LIST_HEAD(&sh->lru);
1466 release_stripe(sh);
1467 return 1;
1468}
1469
d1688a6d 1470static int grow_stripes(struct r5conf *conf, int num)
3f294f4f 1471{
e18b890b 1472 struct kmem_cache *sc;
5e5e3e78 1473 int devs = max(conf->raid_disks, conf->previous_raid_disks);
1da177e4 1474
f4be6b43
N		 1475 if (conf->mddev->gendisk)
1476 sprintf(conf->cache_name[0],
1477 "raid%d-%s", conf->level, mdname(conf->mddev));
1478 else
1479 sprintf(conf->cache_name[0],
1480 "raid%d-%p", conf->level, conf->mddev);
1481 sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]);
1482
ad01c9e3
N		 1483 conf->active_name = 0;
1484 sc = kmem_cache_create(conf->cache_name[conf->active_name],
1da177e4 1485 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
20c2df83 1486 0, 0, NULL);
1da177e4
LT		 1487 if (!sc)
1488 return 1;
1489 conf->slab_cache = sc;
ad01c9e3 1490 conf->pool_size = devs;
16a53ecc 1491 while (num--)
3f294f4f 1492 if (!grow_one_stripe(conf))
1da177e4 1493 return 1;
1da177e4
LT		 1494 return 0;
1495}
29269553 1496
d6f38f31
DW		 1497/**
1498 * scribble_len - return the required size of the scribble region
1499 * @num - total number of disks in the array
1500 *
1501 * The size must be enough to contain:
1502 * 1/ a struct page pointer for each device in the array +2
1503 * 2/ room to convert each entry in (1) to its corresponding dma
1504 * (dma_map_page()) or page (page_address()) address.
1505 *
1506 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
1507 * calculate over all devices (not just the data blocks), using zeros in place
1508 * of the P and Q blocks.
1509 */
1510static size_t scribble_len(int num)
1511{
1512 size_t len;
1513
1514 len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2);
1515
1516 return len;
1517}
1518
d1688a6d 1519static int resize_stripes(struct r5conf *conf, int newsize)
ad01c9e3
N		 1520{
1521 /* Make all the stripes able to hold 'newsize' devices.
1522 * New slots in each stripe get 'page' set to a new page.
1523 *
1524 * This happens in stages:
1525 * 1/ create a new kmem_cache and allocate the required number of
1526 * stripe_heads.
1527 * 2/ gather all the old stripe_heads and tranfer the pages across
1528 * to the new stripe_heads. This will have the side effect of
1529 * freezing the array as once all stripe_heads have been collected,
1530 * no IO will be possible. Old stripe heads are freed once their
1531 * pages have been transferred over, and the old kmem_cache is
1532 * freed when all stripes are done.
1533 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
1534 * we simple return a failre status - no need to clean anything up.
1535 * 4/ allocate new pages for the new slots in the new stripe_heads.
1536 * If this fails, we don't bother trying the shrink the
1537 * stripe_heads down again, we just leave them as they are.
1538 * As each stripe_head is processed the new one is released into
1539 * active service.
1540 *
1541 * Once step2 is started, we cannot afford to wait for a write,
1542 * so we use GFP_NOIO allocations.
1543 */
1544 struct stripe_head *osh, *nsh;
1545 LIST_HEAD(newstripes);
1546 struct disk_info *ndisks;
d6f38f31 1547 unsigned long cpu;
b5470dc5 1548 int err;
e18b890b 1549 struct kmem_cache *sc;
ad01c9e3
N		 1550 int i;
1551
1552 if (newsize <= conf->pool_size)
1553 return 0; /* never bother to shrink */
1554
b5470dc5
DW		 1555 err = md_allow_write(conf->mddev);
1556 if (err)
1557 return err;
2a2275d6 1558
ad01c9e3
N		 1559 /* Step 1 */
1560 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
1561 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
20c2df83 1562 0, 0, NULL);
ad01c9e3
N		 1563 if (!sc)
1564 return -ENOMEM;
1565
1566 for (i = conf->max_nr_stripes; i; i--) {
6ce32846 1567 nsh = kmem_cache_zalloc(sc, GFP_KERNEL);
ad01c9e3
N		 1568 if (!nsh)
1569 break;
1570
ad01c9e3 1571 nsh->raid_conf = conf;
417b8d4a
DW		 1572 #ifdef CONFIG_MULTICORE_RAID456
1573 init_waitqueue_head(&nsh->ops.wait_for_ops);
1574 #endif
ad01c9e3
N		 1575
1576 list_add(&nsh->lru, &newstripes);
1577 }
1578 if (i) {
1579 /* didn't get enough, give up */
1580 while (!list_empty(&newstripes)) {
1581 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1582 list_del(&nsh->lru);
1583 kmem_cache_free(sc, nsh);
1584 }
1585 kmem_cache_destroy(sc);
1586 return -ENOMEM;
1587 }
1588 /* Step 2 - Must use GFP_NOIO now.
1589 * OK, we have enough stripes, start collecting inactive
1590 * stripes and copying them over
1591 */
1592 list_for_each_entry(nsh, &newstripes, lru) {
1593 spin_lock_irq(&conf->device_lock);
1594 wait_event_lock_irq(conf->wait_for_stripe,
1595 !list_empty(&conf->inactive_list),
1596 conf->device_lock,
482c0834 1597 );
ad01c9e3
N		 1598 osh = get_free_stripe(conf);
1599 spin_unlock_irq(&conf->device_lock);
1600 atomic_set(&nsh->count, 1);
1601 for(i=0; i<conf->pool_size; i++)
1602 nsh->dev[i].page = osh->dev[i].page;
1603 for( ; i<newsize; i++)
1604 nsh->dev[i].page = NULL;
1605 kmem_cache_free(conf->slab_cache, osh);
1606 }
1607 kmem_cache_destroy(conf->slab_cache);
1608
1609 /* Step 3.
1610 * At this point, we are holding all the stripes so the array
1611 * is completely stalled, so now is a good time to resize
d6f38f31 1612 * conf->disks and the scribble region
ad01c9e3
N		 1613 */
1614 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
1615 if (ndisks) {
1616 for (i=0; i<conf->raid_disks; i++)
1617 ndisks[i] = conf->disks[i];
1618 kfree(conf->disks);
1619 conf->disks = ndisks;
1620 } else
1621 err = -ENOMEM;
1622
d6f38f31
DW		 1623 get_online_cpus();
1624 conf->scribble_len = scribble_len(newsize);
1625 for_each_present_cpu(cpu) {
1626 struct raid5_percpu *percpu;
1627 void *scribble;
1628
1629 percpu = per_cpu_ptr(conf->percpu, cpu);
1630 scribble = kmalloc(conf->scribble_len, GFP_NOIO);
1631
1632 if (scribble) {
1633 kfree(percpu->scribble);
1634 percpu->scribble = scribble;
1635 } else {
1636 err = -ENOMEM;
1637 break;
1638 }
1639 }
1640 put_online_cpus();
1641
ad01c9e3
N		 1642 /* Step 4, return new stripes to service */
1643 while(!list_empty(&newstripes)) {
1644 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1645 list_del_init(&nsh->lru);
d6f38f31 1646
ad01c9e3
N		 1647 for (i=conf->raid_disks; i < newsize; i++)
1648 if (nsh->dev[i].page == NULL) {
1649 struct page *p = alloc_page(GFP_NOIO);
1650 nsh->dev[i].page = p;
1651 if (!p)
1652 err = -ENOMEM;
1653 }
1654 release_stripe(nsh);
1655 }
1656 /* critical section pass, GFP_NOIO no longer needed */
1657
1658 conf->slab_cache = sc;
1659 conf->active_name = 1-conf->active_name;
1660 conf->pool_size = newsize;
1661 return err;
1662}
1da177e4 1663
d1688a6d 1664static int drop_one_stripe(struct r5conf *conf)
1da177e4
LT		 1665{
1666 struct stripe_head *sh;
1667
3f294f4f
N		 1668 spin_lock_irq(&conf->device_lock);
1669 sh = get_free_stripe(conf);
1670 spin_unlock_irq(&conf->device_lock);
1671 if (!sh)
1672 return 0;
78bafebd 1673 BUG_ON(atomic_read(&sh->count));
e4e11e38 1674 shrink_buffers(sh);
3f294f4f
N		 1675 kmem_cache_free(conf->slab_cache, sh);
1676 atomic_dec(&conf->active_stripes);
1677 return 1;
1678}
1679
d1688a6d 1680static void shrink_stripes(struct r5conf *conf)
3f294f4f
N		 1681{
1682 while (drop_one_stripe(conf))
1683 ;
1684
29fc7e3e
N		 1685 if (conf->slab_cache)
1686 kmem_cache_destroy(conf->slab_cache);
1da177e4
LT		 1687 conf->slab_cache = NULL;
1688}
1689
6712ecf8 1690static void raid5_end_read_request(struct bio * bi, int error)
1da177e4 1691{
99c0fb5f 1692 struct stripe_head *sh = bi->bi_private;
d1688a6d 1693 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 1694 int disks = sh->disks, i;
1da177e4 1695 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
d6950432 1696 char b[BDEVNAME_SIZE];
dd054fce 1697 struct md_rdev *rdev = NULL;
05616be5 1698 sector_t s;
1da177e4
LT		 1699
1700 for (i=0 ; i<disks; i++)
1701 if (bi == &sh->dev[i].req)
1702 break;
1703
45b4233c
DW		 1704 pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n",
1705 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
1da177e4
LT		 1706 uptodate);
1707 if (i == disks) {
1708 BUG();
6712ecf8 1709 return;
1da177e4 1710 }
14a75d3e 1711 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
dd054fce
N		 1712 /* If replacement finished while this request was outstanding,
1713 * 'replacement' might be NULL already.
1714 * In that case it moved down to 'rdev'.
1715 * rdev is not removed until all requests are finished.
1716 */
14a75d3e 1717 rdev = conf->disks[i].replacement;
dd054fce 1718 if (!rdev)
14a75d3e 1719 rdev = conf->disks[i].rdev;
1da177e4 1720
05616be5
N		 1721 if (use_new_offset(conf, sh))
1722 s = sh->sector + rdev->new_data_offset;
1723 else
1724 s = sh->sector + rdev->data_offset;
1da177e4 1725 if (uptodate) {
1da177e4 1726 set_bit(R5_UPTODATE, &sh->dev[i].flags);
4e5314b5 1727 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
14a75d3e
N		 1728 /* Note that this cannot happen on a
1729 * replacement device. We just fail those on
1730 * any error
1731 */
8bda470e
CD		 1732 printk_ratelimited(
1733 KERN_INFO
1734 "md/raid:%s: read error corrected"
1735 " (%lu sectors at %llu on %s)\n",
1736 mdname(conf->mddev), STRIPE_SECTORS,
05616be5 1737 (unsigned long long)s,
8bda470e 1738 bdevname(rdev->bdev, b));
ddd5115f 1739 atomic_add(STRIPE_SECTORS, &rdev->corrected_errors);
4e5314b5
N		 1740 clear_bit(R5_ReadError, &sh->dev[i].flags);
1741 clear_bit(R5_ReWrite, &sh->dev[i].flags);
1742 }
14a75d3e
N		 1743 if (atomic_read(&rdev->read_errors))
1744 atomic_set(&rdev->read_errors, 0);
1da177e4 1745 } else {
14a75d3e 1746 const char *bdn = bdevname(rdev->bdev, b);
ba22dcbf 1747 int retry = 0;
2e8ac303 1748 int set_bad = 0;
d6950432 1749
1da177e4 1750 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
d6950432 1751 atomic_inc(&rdev->read_errors);
14a75d3e
N		 1752 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
1753 printk_ratelimited(
1754 KERN_WARNING
1755 "md/raid:%s: read error on replacement device "
1756 "(sector %llu on %s).\n",
1757 mdname(conf->mddev),
05616be5 1758 (unsigned long long)s,
14a75d3e 1759 bdn);
2e8ac303 1760 else if (conf->mddev->degraded >= conf->max_degraded) {
1761 set_bad = 1;
8bda470e
CD		 1762 printk_ratelimited(
1763 KERN_WARNING
1764 "md/raid:%s: read error not correctable "
1765 "(sector %llu on %s).\n",
1766 mdname(conf->mddev),
05616be5 1767 (unsigned long long)s,
8bda470e 1768 bdn);
2e8ac303 1769 } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) {
4e5314b5 1770 /* Oh, no!!! */
2e8ac303 1771 set_bad = 1;
8bda470e
CD		 1772 printk_ratelimited(
1773 KERN_WARNING
1774 "md/raid:%s: read error NOT corrected!! "
1775 "(sector %llu on %s).\n",
1776 mdname(conf->mddev),
05616be5 1777 (unsigned long long)s,
8bda470e 1778 bdn);
2e8ac303 1779 } else if (atomic_read(&rdev->read_errors)
ba22dcbf 1780 > conf->max_nr_stripes)
14f8d26b 1781 printk(KERN_WARNING
0c55e022 1782 "md/raid:%s: Too many read errors, failing device %s.\n",
d6950432 1783 mdname(conf->mddev), bdn);
ba22dcbf
N		 1784 else
1785 retry = 1;
1786 if (retry)
1787 set_bit(R5_ReadError, &sh->dev[i].flags);
1788 else {
4e5314b5
N		 1789 clear_bit(R5_ReadError, &sh->dev[i].flags);
1790 clear_bit(R5_ReWrite, &sh->dev[i].flags);
2e8ac303 1791 if (!(set_bad
1792 && test_bit(In_sync, &rdev->flags)
1793 && rdev_set_badblocks(
1794 rdev, sh->sector, STRIPE_SECTORS, 0)))
1795 md_error(conf->mddev, rdev);
ba22dcbf 1796 }
1da177e4 1797 }
14a75d3e 1798 rdev_dec_pending(rdev, conf->mddev);
1da177e4
LT		 1799 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1800 set_bit(STRIPE_HANDLE, &sh->state);
1801 release_stripe(sh);
1da177e4
LT		 1802}
1803
d710e138 1804static void raid5_end_write_request(struct bio *bi, int error)
1da177e4 1805{
99c0fb5f 1806 struct stripe_head *sh = bi->bi_private;
d1688a6d 1807 struct r5conf *conf = sh->raid_conf;
7ecaa1e6 1808 int disks = sh->disks, i;
977df362 1809 struct md_rdev *uninitialized_var(rdev);
1da177e4 1810 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
b84db560
N		 1811 sector_t first_bad;
1812 int bad_sectors;
977df362 1813 int replacement = 0;
1da177e4 1814
977df362
N		 1815 for (i = 0 ; i < disks; i++) {
1816 if (bi == &sh->dev[i].req) {
1817 rdev = conf->disks[i].rdev;
1da177e4 1818 break;
977df362
N		 1819 }
1820 if (bi == &sh->dev[i].rreq) {
1821 rdev = conf->disks[i].replacement;
dd054fce
N		 1822 if (rdev)
1823 replacement = 1;
1824 else
1825 /* rdev was removed and 'replacement'
1826 * replaced it. rdev is not removed
1827 * until all requests are finished.
1828 */
1829 rdev = conf->disks[i].rdev;
977df362
N		 1830 break;
1831 }
1832 }
45b4233c 1833 pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n",
1da177e4
LT		 1834 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
1835 uptodate);
1836 if (i == disks) {
1837 BUG();
6712ecf8 1838 return;
1da177e4
LT		 1839 }
1840
977df362
N		 1841 if (replacement) {
1842 if (!uptodate)
1843 md_error(conf->mddev, rdev);
1844 else if (is_badblock(rdev, sh->sector,
1845 STRIPE_SECTORS,
1846 &first_bad, &bad_sectors))
1847 set_bit(R5_MadeGoodRepl, &sh->dev[i].flags);
1848 } else {
1849 if (!uptodate) {
1850 set_bit(WriteErrorSeen, &rdev->flags);
1851 set_bit(R5_WriteError, &sh->dev[i].flags);
3a6de292
N		 1852 if (!test_and_set_bit(WantReplacement, &rdev->flags))
1853 set_bit(MD_RECOVERY_NEEDED,
1854 &rdev->mddev->recovery);
977df362
N		 1855 } else if (is_badblock(rdev, sh->sector,
1856 STRIPE_SECTORS,
1857 &first_bad, &bad_sectors))
1858 set_bit(R5_MadeGood, &sh->dev[i].flags);
1859 }
1860 rdev_dec_pending(rdev, conf->mddev);
1da177e4 1861
977df362
N		 1862 if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags))
1863 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1da177e4 1864 set_bit(STRIPE_HANDLE, &sh->state);
c04be0aa 1865 release_stripe(sh);
1da177e4
LT		 1866}
1867
784052ec 1868static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous);
1da177e4 1869
784052ec 1870static void raid5_build_block(struct stripe_head *sh, int i, int previous)
1da177e4
LT		 1871{
1872 struct r5dev *dev = &sh->dev[i];
1873
1874 bio_init(&dev->req);
1875 dev->req.bi_io_vec = &dev->vec;
1876 dev->req.bi_vcnt++;
1877 dev->req.bi_max_vecs++;
1da177e4 1878 dev->req.bi_private = sh;
995c4275 1879 dev->vec.bv_page = dev->page;
1da177e4 1880
977df362
N		 1881 bio_init(&dev->rreq);
1882 dev->rreq.bi_io_vec = &dev->rvec;
1883 dev->rreq.bi_vcnt++;
1884 dev->rreq.bi_max_vecs++;
1885 dev->rreq.bi_private = sh;
1886 dev->rvec.bv_page = dev->page;
1887
1da177e4 1888 dev->flags = 0;
784052ec 1889 dev->sector = compute_blocknr(sh, i, previous);
1da177e4
LT		 1890}
1891
fd01b88c 1892static void error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1893{
1894 char b[BDEVNAME_SIZE];
d1688a6d 1895 struct r5conf *conf = mddev->private;
908f4fbd 1896 unsigned long flags;
0c55e022 1897 pr_debug("raid456: error called\n");
1da177e4 1898
908f4fbd
N		 1899 spin_lock_irqsave(&conf->device_lock, flags);
1900 clear_bit(In_sync, &rdev->flags);
1901 mddev->degraded = calc_degraded(conf);
1902 spin_unlock_irqrestore(&conf->device_lock, flags);
1903 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1904
de393cde 1905 set_bit(Blocked, &rdev->flags);
6f8d0c77
N		 1906 set_bit(Faulty, &rdev->flags);
1907 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1908 printk(KERN_ALERT
1909 "md/raid:%s: Disk failure on %s, disabling device.\n"
1910 "md/raid:%s: Operation continuing on %d devices.\n",
1911 mdname(mddev),
1912 bdevname(rdev->bdev, b),
1913 mdname(mddev),
1914 conf->raid_disks - mddev->degraded);
16a53ecc 1915}
1da177e4
LT		 1916
1917/*
1918 * Input: a 'big' sector number,
1919 * Output: index of the data and parity disk, and the sector # in them.
1920 */
d1688a6d 1921static sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector,
911d4ee8
N		 1922 int previous, int *dd_idx,
1923 struct stripe_head *sh)
1da177e4 1924{
6e3b96ed 1925 sector_t stripe, stripe2;
35f2a591 1926 sector_t chunk_number;
1da177e4 1927 unsigned int chunk_offset;
911d4ee8 1928 int pd_idx, qd_idx;
67cc2b81 1929 int ddf_layout = 0;
1da177e4 1930 sector_t new_sector;
e183eaed
N		 1931 int algorithm = previous ? conf->prev_algo
1932 : conf->algorithm;
09c9e5fa
AN		 1933 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
1934 : conf->chunk_sectors;
112bf897
N		 1935 int raid_disks = previous ? conf->previous_raid_disks
1936 : conf->raid_disks;
1937 int data_disks = raid_disks - conf->max_degraded;
1da177e4
LT		 1938
1939 /* First compute the information on this sector */
1940
1941 /*
1942 * Compute the chunk number and the sector offset inside the chunk
1943 */
1944 chunk_offset = sector_div(r_sector, sectors_per_chunk);
1945 chunk_number = r_sector;
1da177e4
LT		 1946
1947 /*
1948 * Compute the stripe number
1949 */
35f2a591
N		 1950 stripe = chunk_number;
1951 *dd_idx = sector_div(stripe, data_disks);
6e3b96ed 1952 stripe2 = stripe;
1da177e4
LT		 1953 /*
1954 * Select the parity disk based on the user selected algorithm.
1955 */
84789554 1956 pd_idx = qd_idx = -1;
16a53ecc
N		 1957 switch(conf->level) {
1958 case 4:
911d4ee8 1959 pd_idx = data_disks;
16a53ecc
N		 1960 break;
1961 case 5:
e183eaed 1962 switch (algorithm) {
1da177e4 1963 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 1964 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 1965 if (*dd_idx >= pd_idx)
1da177e4
LT		 1966 (*dd_idx)++;
1967 break;
1968 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 1969 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 1970 if (*dd_idx >= pd_idx)
1da177e4
LT		 1971 (*dd_idx)++;
1972 break;
1973 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 1974 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 1975 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4
LT		 1976 break;
1977 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 1978 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 1979 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4 1980 break;
99c0fb5f
N		 1981 case ALGORITHM_PARITY_0:
1982 pd_idx = 0;
1983 (*dd_idx)++;
1984 break;
1985 case ALGORITHM_PARITY_N:
1986 pd_idx = data_disks;
1987 break;
1da177e4 1988 default:
99c0fb5f 1989 BUG();
16a53ecc
N		 1990 }
1991 break;
1992 case 6:
1993
e183eaed 1994 switch (algorithm) {
16a53ecc 1995 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 1996 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 1997 qd_idx = pd_idx + 1;
1998 if (pd_idx == raid_disks-1) {
99c0fb5f 1999 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2000 qd_idx = 0;
2001 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2002 (*dd_idx) += 2; /* D D P Q D */
2003 break;
2004 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 2005 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2006 qd_idx = pd_idx + 1;
2007 if (pd_idx == raid_disks-1) {
99c0fb5f 2008 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 2009 qd_idx = 0;
2010 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 2011 (*dd_idx) += 2; /* D D P Q D */
2012 break;
2013 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 2014 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 2015 qd_idx = (pd_idx + 1) % raid_disks;
2016 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc
N		 2017 break;
2018 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 2019 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 2020 qd_idx = (pd_idx + 1) % raid_disks;
2021 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc 2022 break;
99c0fb5f
N		 2023
2024 case ALGORITHM_PARITY_0:
2025 pd_idx = 0;
2026 qd_idx = 1;
2027 (*dd_idx) += 2;
2028 break;
2029 case ALGORITHM_PARITY_N:
2030 pd_idx = data_disks;
2031 qd_idx = data_disks + 1;
2032 break;
2033
2034 case ALGORITHM_ROTATING_ZERO_RESTART:
2035 /* Exactly the same as RIGHT_ASYMMETRIC, but or
2036 * of blocks for computing Q is different.
2037 */
6e3b96ed 2038 pd_idx = sector_div(stripe2, raid_disks);
99c0fb5f
N		 2039 qd_idx = pd_idx + 1;
2040 if (pd_idx == raid_disks-1) {
2041 (*dd_idx)++; /* Q D D D P */
2042 qd_idx = 0;
2043 } else if (*dd_idx >= pd_idx)
2044 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2045 ddf_layout = 1;
99c0fb5f
N		 2046 break;
2047
2048 case ALGORITHM_ROTATING_N_RESTART:
2049 /* Same a left_asymmetric, by first stripe is
2050 * D D D P Q rather than
2051 * Q D D D P
2052 */
6e3b96ed
N		 2053 stripe2 += 1;
2054 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2055 qd_idx = pd_idx + 1;
2056 if (pd_idx == raid_disks-1) {
2057 (*dd_idx)++; /* Q D D D P */
2058 qd_idx = 0;
2059 } else if (*dd_idx >= pd_idx)
2060 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 2061 ddf_layout = 1;
99c0fb5f
N		 2062 break;
2063
2064 case ALGORITHM_ROTATING_N_CONTINUE:
2065 /* Same as left_symmetric but Q is before P */
6e3b96ed 2066 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 2067 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
2068 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
67cc2b81 2069 ddf_layout = 1;
99c0fb5f
N		 2070 break;
2071
2072 case ALGORITHM_LEFT_ASYMMETRIC_6:
2073 /* RAID5 left_asymmetric, with Q on last device */
6e3b96ed 2074 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2075 if (*dd_idx >= pd_idx)
2076 (*dd_idx)++;
2077 qd_idx = raid_disks - 1;
2078 break;
2079
2080 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6e3b96ed 2081 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2082 if (*dd_idx >= pd_idx)
2083 (*dd_idx)++;
2084 qd_idx = raid_disks - 1;
2085 break;
2086
2087 case ALGORITHM_LEFT_SYMMETRIC_6:
6e3b96ed 2088 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2089 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2090 qd_idx = raid_disks - 1;
2091 break;
2092
2093 case ALGORITHM_RIGHT_SYMMETRIC_6:
6e3b96ed 2094 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 2095 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2096 qd_idx = raid_disks - 1;
2097 break;
2098
2099 case ALGORITHM_PARITY_0_6:
2100 pd_idx = 0;
2101 (*dd_idx)++;
2102 qd_idx = raid_disks - 1;
2103 break;
2104
16a53ecc 2105 default:
99c0fb5f 2106 BUG();
16a53ecc
N		 2107 }
2108 break;
1da177e4
LT		 2109 }
2110
911d4ee8
N		 2111 if (sh) {
2112 sh->pd_idx = pd_idx;
2113 sh->qd_idx = qd_idx;
67cc2b81 2114 sh->ddf_layout = ddf_layout;
911d4ee8 2115 }
1da177e4
LT		 2116 /*
2117 * Finally, compute the new sector number
2118 */
2119 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
2120 return new_sector;
2121}
2122
2123
784052ec 2124static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous)
1da177e4 2125{
d1688a6d 2126 struct r5conf *conf = sh->raid_conf;
b875e531
N		 2127 int raid_disks = sh->disks;
2128 int data_disks = raid_disks - conf->max_degraded;
1da177e4 2129 sector_t new_sector = sh->sector, check;
09c9e5fa
AN		 2130 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2131 : conf->chunk_sectors;
e183eaed
N		 2132 int algorithm = previous ? conf->prev_algo
2133 : conf->algorithm;
1da177e4
LT		 2134 sector_t stripe;
2135 int chunk_offset;
35f2a591
N		 2136 sector_t chunk_number;
2137 int dummy1, dd_idx = i;
1da177e4 2138 sector_t r_sector;
911d4ee8 2139 struct stripe_head sh2;
1da177e4 2140
16a53ecc 2141
1da177e4
LT		 2142 chunk_offset = sector_div(new_sector, sectors_per_chunk);
2143 stripe = new_sector;
1da177e4 2144
16a53ecc
N		 2145 if (i == sh->pd_idx)
2146 return 0;
2147 switch(conf->level) {
2148 case 4: break;
2149 case 5:
e183eaed 2150 switch (algorithm) {
1da177e4
LT		 2151 case ALGORITHM_LEFT_ASYMMETRIC:
2152 case ALGORITHM_RIGHT_ASYMMETRIC:
2153 if (i > sh->pd_idx)
2154 i--;
2155 break;
2156 case ALGORITHM_LEFT_SYMMETRIC:
2157 case ALGORITHM_RIGHT_SYMMETRIC:
2158 if (i < sh->pd_idx)
2159 i += raid_disks;
2160 i -= (sh->pd_idx + 1);
2161 break;
99c0fb5f
N		 2162 case ALGORITHM_PARITY_0:
2163 i -= 1;
2164 break;
2165 case ALGORITHM_PARITY_N:
2166 break;
1da177e4 2167 default:
99c0fb5f 2168 BUG();
16a53ecc
N		 2169 }
2170 break;
2171 case 6:
d0dabf7e 2172 if (i == sh->qd_idx)
16a53ecc 2173 return 0; /* It is the Q disk */
e183eaed 2174 switch (algorithm) {
16a53ecc
N		 2175 case ALGORITHM_LEFT_ASYMMETRIC:
2176 case ALGORITHM_RIGHT_ASYMMETRIC:
99c0fb5f
N		 2177 case ALGORITHM_ROTATING_ZERO_RESTART:
2178 case ALGORITHM_ROTATING_N_RESTART:
2179 if (sh->pd_idx == raid_disks-1)
2180 i--; /* Q D D D P */
16a53ecc
N		 2181 else if (i > sh->pd_idx)
2182 i -= 2; /* D D P Q D */
2183 break;
2184 case ALGORITHM_LEFT_SYMMETRIC:
2185 case ALGORITHM_RIGHT_SYMMETRIC:
2186 if (sh->pd_idx == raid_disks-1)
2187 i--; /* Q D D D P */
2188 else {
2189 /* D D P Q D */
2190 if (i < sh->pd_idx)
2191 i += raid_disks;
2192 i -= (sh->pd_idx + 2);
2193 }
2194 break;
99c0fb5f
N		 2195 case ALGORITHM_PARITY_0:
2196 i -= 2;
2197 break;
2198 case ALGORITHM_PARITY_N:
2199 break;
2200 case ALGORITHM_ROTATING_N_CONTINUE:
e4424fee 2201 /* Like left_symmetric, but P is before Q */
99c0fb5f
N		 2202 if (sh->pd_idx == 0)
2203 i--; /* P D D D Q */
e4424fee
N		 2204 else {
2205 /* D D Q P D */
2206 if (i < sh->pd_idx)
2207 i += raid_disks;
2208 i -= (sh->pd_idx + 1);
2209 }
99c0fb5f
N		 2210 break;
2211 case ALGORITHM_LEFT_ASYMMETRIC_6:
2212 case ALGORITHM_RIGHT_ASYMMETRIC_6:
2213 if (i > sh->pd_idx)
2214 i--;
2215 break;
2216 case ALGORITHM_LEFT_SYMMETRIC_6:
2217 case ALGORITHM_RIGHT_SYMMETRIC_6:
2218 if (i < sh->pd_idx)
2219 i += data_disks + 1;
2220 i -= (sh->pd_idx + 1);
2221 break;
2222 case ALGORITHM_PARITY_0_6:
2223 i -= 1;
2224 break;
16a53ecc 2225 default:
99c0fb5f 2226 BUG();
16a53ecc
N		 2227 }
2228 break;
1da177e4
LT		 2229 }
2230
2231 chunk_number = stripe * data_disks + i;
35f2a591 2232 r_sector = chunk_number * sectors_per_chunk + chunk_offset;
1da177e4 2233
112bf897 2234 check = raid5_compute_sector(conf, r_sector,
784052ec 2235 previous, &dummy1, &sh2);
911d4ee8
N		 2236 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
2237 || sh2.qd_idx != sh->qd_idx) {
0c55e022
N		 2238 printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n",
2239 mdname(conf->mddev));
1da177e4
LT		 2240 return 0;
2241 }
2242 return r_sector;
2243}
2244
2245
600aa109 2246static void
c0f7bddb 2247schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
600aa109 2248 int rcw, int expand)
e33129d8
DW		 2249{
2250 int i, pd_idx = sh->pd_idx, disks = sh->disks;
d1688a6d 2251 struct r5conf *conf = sh->raid_conf;
c0f7bddb 2252 int level = conf->level;
e33129d8
DW		 2253
2254 if (rcw) {
2255 /* if we are not expanding this is a proper write request, and
2256 * there will be bios with new data to be drained into the
2257 * stripe cache
2258 */
2259 if (!expand) {
600aa109
DW		 2260 sh->reconstruct_state = reconstruct_state_drain_run;
2261 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2262 } else
2263 sh->reconstruct_state = reconstruct_state_run;
16a53ecc 2264
ac6b53b6 2265 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
e33129d8
DW		 2266
2267 for (i = disks; i--; ) {
2268 struct r5dev *dev = &sh->dev[i];
2269
2270 if (dev->towrite) {
2271 set_bit(R5_LOCKED, &dev->flags);
d8ee0728 2272 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2273 if (!expand)
2274 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2275 s->locked++;
e33129d8
DW		 2276 }
2277 }
c0f7bddb 2278 if (s->locked + conf->max_degraded == disks)
8b3e6cdc 2279 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
c0f7bddb 2280 atomic_inc(&conf->pending_full_writes);
e33129d8 2281 } else {
c0f7bddb 2282 BUG_ON(level == 6);
e33129d8
DW		 2283 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
2284 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
2285
d8ee0728 2286 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
600aa109
DW		 2287 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
2288 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
ac6b53b6 2289 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
e33129d8
DW		 2290
2291 for (i = disks; i--; ) {
2292 struct r5dev *dev = &sh->dev[i];
2293 if (i == pd_idx)
2294 continue;
2295
e33129d8
DW		 2296 if (dev->towrite &&
2297 (test_bit(R5_UPTODATE, &dev->flags) ||
d8ee0728
DW		 2298 test_bit(R5_Wantcompute, &dev->flags))) {
2299 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2300 set_bit(R5_LOCKED, &dev->flags);
2301 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2302 s->locked++;
e33129d8
DW		 2303 }
2304 }
2305 }
2306
c0f7bddb 2307 /* keep the parity disk(s) locked while asynchronous operations
e33129d8
DW		 2308 * are in flight
2309 */
2310 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
2311 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
600aa109 2312 s->locked++;
e33129d8 2313
c0f7bddb
YT		 2314 if (level == 6) {
2315 int qd_idx = sh->qd_idx;
2316 struct r5dev *dev = &sh->dev[qd_idx];
2317
2318 set_bit(R5_LOCKED, &dev->flags);
2319 clear_bit(R5_UPTODATE, &dev->flags);
2320 s->locked++;
2321 }
2322
600aa109 2323 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
e46b272b 2324 __func__, (unsigned long long)sh->sector,
600aa109 2325 s->locked, s->ops_request);
e33129d8 2326}
16a53ecc 2327
1da177e4
LT		 2328/*
2329 * Each stripe/dev can have one or more bion attached.
16a53ecc 2330 * toread/towrite point to the first in a chain.
1da177e4
LT		 2331 * The bi_next chain must be in order.
2332 */
2333static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite)
2334{
2335 struct bio **bip;
d1688a6d 2336 struct r5conf *conf = sh->raid_conf;
72626685 2337 int firstwrite=0;
1da177e4 2338
cbe47ec5 2339 pr_debug("adding bi b#%llu to stripe s#%llu\n",
1da177e4
LT		 2340 (unsigned long long)bi->bi_sector,
2341 (unsigned long long)sh->sector);
2342
2343
1da177e4 2344 spin_lock_irq(&conf->device_lock);
72626685 2345 if (forwrite) {
1da177e4 2346 bip = &sh->dev[dd_idx].towrite;
72626685
N		 2347 if (*bip == NULL && sh->dev[dd_idx].written == NULL)
2348 firstwrite = 1;
2349 } else
1da177e4
LT		 2350 bip = &sh->dev[dd_idx].toread;
2351 while (*bip && (*bip)->bi_sector < bi->bi_sector) {
2352 if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector)
2353 goto overlap;
2354 bip = & (*bip)->bi_next;
2355 }
2356 if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9))
2357 goto overlap;
2358
78bafebd 2359 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
1da177e4
LT		 2360 if (*bip)
2361 bi->bi_next = *bip;
2362 *bip = bi;
960e739d 2363 bi->bi_phys_segments++;
72626685 2364
1da177e4
LT		 2365 if (forwrite) {
2366 /* check if page is covered */
2367 sector_t sector = sh->dev[dd_idx].sector;
2368 for (bi=sh->dev[dd_idx].towrite;
2369 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
2370 bi && bi->bi_sector <= sector;
2371 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
2372 if (bi->bi_sector + (bi->bi_size>>9) >= sector)
2373 sector = bi->bi_sector + (bi->bi_size>>9);
2374 }
2375 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
2376 set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags);
2377 }
cbe47ec5 2378 spin_unlock_irq(&conf->device_lock);
cbe47ec5
N		 2379
2380 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
2381 (unsigned long long)(*bip)->bi_sector,
2382 (unsigned long long)sh->sector, dd_idx);
2383
2384 if (conf->mddev->bitmap && firstwrite) {
2385 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
2386 STRIPE_SECTORS, 0);
2387 sh->bm_seq = conf->seq_flush+1;
2388 set_bit(STRIPE_BIT_DELAY, &sh->state);
2389 }
1da177e4
LT		 2390 return 1;
2391
2392 overlap:
2393 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
2394 spin_unlock_irq(&conf->device_lock);
1da177e4
LT		 2395 return 0;
2396}
2397
d1688a6d 2398static void end_reshape(struct r5conf *conf);
29269553 2399
d1688a6d 2400static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
911d4ee8 2401 struct stripe_head *sh)
ccfcc3c1 2402{
784052ec 2403 int sectors_per_chunk =
09c9e5fa 2404 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
911d4ee8 2405 int dd_idx;
2d2063ce 2406 int chunk_offset = sector_div(stripe, sectors_per_chunk);
112bf897 2407 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
2d2063ce 2408
112bf897
N		 2409 raid5_compute_sector(conf,
2410 stripe * (disks - conf->max_degraded)
b875e531 2411 *sectors_per_chunk + chunk_offset,
112bf897 2412 previous,
911d4ee8 2413 &dd_idx, sh);
ccfcc3c1
N		 2414}
2415
a4456856 2416static void
d1688a6d 2417handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh,
a4456856
DW		 2418 struct stripe_head_state *s, int disks,
2419 struct bio **return_bi)
2420{
2421 int i;
2422 for (i = disks; i--; ) {
2423 struct bio *bi;
2424 int bitmap_end = 0;
2425
2426 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
3cb03002 2427 struct md_rdev *rdev;
a4456856
DW		 2428 rcu_read_lock();
2429 rdev = rcu_dereference(conf->disks[i].rdev);
2430 if (rdev && test_bit(In_sync, &rdev->flags))
7f0da59b
N		 2431 atomic_inc(&rdev->nr_pending);
2432 else
2433 rdev = NULL;
a4456856 2434 rcu_read_unlock();
7f0da59b
N		 2435 if (rdev) {
2436 if (!rdev_set_badblocks(
2437 rdev,
2438 sh->sector,
2439 STRIPE_SECTORS, 0))
2440 md_error(conf->mddev, rdev);
2441 rdev_dec_pending(rdev, conf->mddev);
2442 }
a4456856
DW		 2443 }
2444 spin_lock_irq(&conf->device_lock);
2445 /* fail all writes first */
2446 bi = sh->dev[i].towrite;
2447 sh->dev[i].towrite = NULL;
2448 if (bi) {
2449 s->to_write--;
2450 bitmap_end = 1;
2451 }
2452
2453 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2454 wake_up(&conf->wait_for_overlap);
2455
2456 while (bi && bi->bi_sector <
2457 sh->dev[i].sector + STRIPE_SECTORS) {
2458 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
2459 clear_bit(BIO_UPTODATE, &bi->bi_flags);
960e739d 2460 if (!raid5_dec_bi_phys_segments(bi)) {
a4456856
DW		 2461 md_write_end(conf->mddev);
2462 bi->bi_next = *return_bi;
2463 *return_bi = bi;
2464 }
2465 bi = nextbi;
2466 }
2467 /* and fail all 'written' */
2468 bi = sh->dev[i].written;
2469 sh->dev[i].written = NULL;
2470 if (bi) bitmap_end = 1;
2471 while (bi && bi->bi_sector <
2472 sh->dev[i].sector + STRIPE_SECTORS) {
2473 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
2474 clear_bit(BIO_UPTODATE, &bi->bi_flags);
960e739d 2475 if (!raid5_dec_bi_phys_segments(bi)) {
a4456856
DW		 2476 md_write_end(conf->mddev);
2477 bi->bi_next = *return_bi;
2478 *return_bi = bi;
2479 }
2480 bi = bi2;
2481 }
2482
b5e98d65
DW		 2483 /* fail any reads if this device is non-operational and
2484 * the data has not reached the cache yet.
2485 */
2486 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
2487 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
2488 test_bit(R5_ReadError, &sh->dev[i].flags))) {
a4456856
DW		 2489 bi = sh->dev[i].toread;
2490 sh->dev[i].toread = NULL;
2491 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2492 wake_up(&conf->wait_for_overlap);
2493 if (bi) s->to_read--;
2494 while (bi && bi->bi_sector <
2495 sh->dev[i].sector + STRIPE_SECTORS) {
2496 struct bio *nextbi =
2497 r5_next_bio(bi, sh->dev[i].sector);
2498 clear_bit(BIO_UPTODATE, &bi->bi_flags);
960e739d 2499 if (!raid5_dec_bi_phys_segments(bi)) {
a4456856
DW		 2500 bi->bi_next = *return_bi;
2501 *return_bi = bi;
2502 }
2503 bi = nextbi;
2504 }
2505 }
2506 spin_unlock_irq(&conf->device_lock);
2507 if (bitmap_end)
2508 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
2509 STRIPE_SECTORS, 0, 0);
8cfa7b0f
N		 2510 /* If we were in the middle of a write the parity block might
2511 * still be locked - so just clear all R5_LOCKED flags
2512 */
2513 clear_bit(R5_LOCKED, &sh->dev[i].flags);
a4456856
DW		 2514 }
2515
8b3e6cdc
DW		 2516 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2517 if (atomic_dec_and_test(&conf->pending_full_writes))
2518 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 2519}
2520
7f0da59b 2521static void
d1688a6d 2522handle_failed_sync(struct r5conf *conf, struct stripe_head *sh,
7f0da59b
N		 2523 struct stripe_head_state *s)
2524{
2525 int abort = 0;
2526 int i;
2527
7f0da59b
N		 2528 clear_bit(STRIPE_SYNCING, &sh->state);
2529 s->syncing = 0;
9a3e1101 2530 s->replacing = 0;
7f0da59b 2531 /* There is nothing more to do for sync/check/repair.
18b9837e
N		 2532 * Don't even need to abort as that is handled elsewhere
2533 * if needed, and not always wanted e.g. if there is a known
2534 * bad block here.
9a3e1101 2535 * For recover/replace we need to record a bad block on all
7f0da59b
N		 2536 * non-sync devices, or abort the recovery
2537 */
18b9837e
N		 2538 if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) {
2539 /* During recovery devices cannot be removed, so
2540 * locking and refcounting of rdevs is not needed
2541 */
2542 for (i = 0; i < conf->raid_disks; i++) {
2543 struct md_rdev *rdev = conf->disks[i].rdev;
2544 if (rdev
2545 && !test_bit(Faulty, &rdev->flags)
2546 && !test_bit(In_sync, &rdev->flags)
2547 && !rdev_set_badblocks(rdev, sh->sector,
2548 STRIPE_SECTORS, 0))
2549 abort = 1;
2550 rdev = conf->disks[i].replacement;
2551 if (rdev
2552 && !test_bit(Faulty, &rdev->flags)
2553 && !test_bit(In_sync, &rdev->flags)
2554 && !rdev_set_badblocks(rdev, sh->sector,
2555 STRIPE_SECTORS, 0))
2556 abort = 1;
2557 }
2558 if (abort)
2559 conf->recovery_disabled =
2560 conf->mddev->recovery_disabled;
7f0da59b 2561 }
18b9837e 2562 md_done_sync(conf->mddev, STRIPE_SECTORS, !abort);
7f0da59b
N		 2563}
2564
9a3e1101
N		 2565static int want_replace(struct stripe_head *sh, int disk_idx)
2566{
2567 struct md_rdev *rdev;
2568 int rv = 0;
2569 /* Doing recovery so rcu locking not required */
2570 rdev = sh->raid_conf->disks[disk_idx].replacement;
2571 if (rdev
2572 && !test_bit(Faulty, &rdev->flags)
2573 && !test_bit(In_sync, &rdev->flags)
2574 && (rdev->recovery_offset <= sh->sector
2575 || rdev->mddev->recovery_cp <= sh->sector))
2576 rv = 1;
2577
2578 return rv;
2579}
2580
93b3dbce 2581/* fetch_block - checks the given member device to see if its data needs
1fe797e6
DW		 2582 * to be read or computed to satisfy a request.
2583 *
2584 * Returns 1 when no more member devices need to be checked, otherwise returns
93b3dbce 2585 * 0 to tell the loop in handle_stripe_fill to continue
f38e1219 2586 */
93b3dbce
N		 2587static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s,
2588 int disk_idx, int disks)
a4456856 2589{
5599becc 2590 struct r5dev *dev = &sh->dev[disk_idx];
f2b3b44d
N		 2591 struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]],
2592 &sh->dev[s->failed_num[1]] };
5599becc 2593
93b3dbce 2594 /* is the data in this block needed, and can we get it? */
5599becc
YT		 2595 if (!test_bit(R5_LOCKED, &dev->flags) &&
2596 !test_bit(R5_UPTODATE, &dev->flags) &&
2597 (dev->toread ||
2598 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
2599 s->syncing || s->expanding ||
9a3e1101 2600 (s->replacing && want_replace(sh, disk_idx)) ||
5d35e09c
N		 2601 (s->failed >= 1 && fdev[0]->toread) ||
2602 (s->failed >= 2 && fdev[1]->toread) ||
93b3dbce
N		 2603 (sh->raid_conf->level <= 5 && s->failed && fdev[0]->towrite &&
2604 !test_bit(R5_OVERWRITE, &fdev[0]->flags)) ||
2605 (sh->raid_conf->level == 6 && s->failed && s->to_write))) {
5599becc
YT		 2606 /* we would like to get this block, possibly by computing it,
2607 * otherwise read it if the backing disk is insync
2608 */
2609 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
2610 BUG_ON(test_bit(R5_Wantread, &dev->flags));
2611 if ((s->uptodate == disks - 1) &&
f2b3b44d
N		 2612 (s->failed && (disk_idx == s->failed_num[0] ||
2613 disk_idx == s->failed_num[1]))) {
5599becc
YT		 2614 /* have disk failed, and we're requested to fetch it;
2615 * do compute it
a4456856 2616 */
5599becc
YT		 2617 pr_debug("Computing stripe %llu block %d\n",
2618 (unsigned long long)sh->sector, disk_idx);
2619 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2620 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2621 set_bit(R5_Wantcompute, &dev->flags);
2622 sh->ops.target = disk_idx;
2623 sh->ops.target2 = -1; /* no 2nd target */
2624 s->req_compute = 1;
93b3dbce
N		 2625 /* Careful: from this point on 'uptodate' is in the eye
2626 * of raid_run_ops which services 'compute' operations
2627 * before writes. R5_Wantcompute flags a block that will
2628 * be R5_UPTODATE by the time it is needed for a
2629 * subsequent operation.
2630 */
5599becc
YT		 2631 s->uptodate++;
2632 return 1;
2633 } else if (s->uptodate == disks-2 && s->failed >= 2) {
2634 /* Computing 2-failure is *very* expensive; only
2635 * do it if failed >= 2
2636 */
2637 int other;
2638 for (other = disks; other--; ) {
2639 if (other == disk_idx)
2640 continue;
2641 if (!test_bit(R5_UPTODATE,
2642 &sh->dev[other].flags))
2643 break;
a4456856 2644 }
5599becc
YT		 2645 BUG_ON(other < 0);
2646 pr_debug("Computing stripe %llu blocks %d,%d\n",
2647 (unsigned long long)sh->sector,
2648 disk_idx, other);
2649 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2650 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2651 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
2652 set_bit(R5_Wantcompute, &sh->dev[other].flags);
2653 sh->ops.target = disk_idx;
2654 sh->ops.target2 = other;
2655 s->uptodate += 2;
2656 s->req_compute = 1;
2657 return 1;
2658 } else if (test_bit(R5_Insync, &dev->flags)) {
2659 set_bit(R5_LOCKED, &dev->flags);
2660 set_bit(R5_Wantread, &dev->flags);
2661 s->locked++;
2662 pr_debug("Reading block %d (sync=%d)\n",
2663 disk_idx, s->syncing);
a4456856
DW		 2664 }
2665 }
5599becc
YT		 2666
2667 return 0;
2668}
2669
2670/**
93b3dbce 2671 * handle_stripe_fill - read or compute data to satisfy pending requests.
5599becc 2672 */
93b3dbce
N		 2673static void handle_stripe_fill(struct stripe_head *sh,
2674 struct stripe_head_state *s,
2675 int disks)
5599becc
YT		 2676{
2677 int i;
2678
2679 /* look for blocks to read/compute, skip this if a compute
2680 * is already in flight, or if the stripe contents are in the
2681 * midst of changing due to a write
2682 */
2683 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
2684 !sh->reconstruct_state)
2685 for (i = disks; i--; )
93b3dbce 2686 if (fetch_block(sh, s, i, disks))
5599becc 2687 break;
a4456856
DW		 2688 set_bit(STRIPE_HANDLE, &sh->state);
2689}
2690
2691
1fe797e6 2692/* handle_stripe_clean_event
a4456856
DW		 2693 * any written block on an uptodate or failed drive can be returned.
2694 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
2695 * never LOCKED, so we don't need to test 'failed' directly.
2696 */
d1688a6d 2697static void handle_stripe_clean_event(struct r5conf *conf,
a4456856
DW		 2698 struct stripe_head *sh, int disks, struct bio **return_bi)
2699{
2700 int i;
2701 struct r5dev *dev;
2702
2703 for (i = disks; i--; )
2704 if (sh->dev[i].written) {
2705 dev = &sh->dev[i];
2706 if (!test_bit(R5_LOCKED, &dev->flags) &&
2707 test_bit(R5_UPTODATE, &dev->flags)) {
2708 /* We can return any write requests */
2709 struct bio *wbi, *wbi2;
2710 int bitmap_end = 0;
45b4233c 2711 pr_debug("Return write for disc %d\n", i);
a4456856
DW		 2712 spin_lock_irq(&conf->device_lock);
2713 wbi = dev->written;
2714 dev->written = NULL;
2715 while (wbi && wbi->bi_sector <
2716 dev->sector + STRIPE_SECTORS) {
2717 wbi2 = r5_next_bio(wbi, dev->sector);
960e739d 2718 if (!raid5_dec_bi_phys_segments(wbi)) {
a4456856
DW		 2719 md_write_end(conf->mddev);
2720 wbi->bi_next = *return_bi;
2721 *return_bi = wbi;
2722 }
2723 wbi = wbi2;
2724 }
2725 if (dev->towrite == NULL)
2726 bitmap_end = 1;
2727 spin_unlock_irq(&conf->device_lock);
2728 if (bitmap_end)
2729 bitmap_endwrite(conf->mddev->bitmap,
2730 sh->sector,
2731 STRIPE_SECTORS,
2732 !test_bit(STRIPE_DEGRADED, &sh->state),
2733 0);
2734 }
2735 }
8b3e6cdc
DW		 2736
2737 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2738 if (atomic_dec_and_test(&conf->pending_full_writes))
2739 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 2740}
2741
d1688a6d 2742static void handle_stripe_dirtying(struct r5conf *conf,
c8ac1803
N		 2743 struct stripe_head *sh,
2744 struct stripe_head_state *s,
2745 int disks)
a4456856
DW		 2746{
2747 int rmw = 0, rcw = 0, i;
c8ac1803
N		 2748 if (conf->max_degraded == 2) {
2749 /* RAID6 requires 'rcw' in current implementation
2750 * Calculate the real rcw later - for now fake it
2751 * look like rcw is cheaper
2752 */
2753 rcw = 1; rmw = 2;
2754 } else for (i = disks; i--; ) {
a4456856
DW		 2755 /* would I have to read this buffer for read_modify_write */
2756 struct r5dev *dev = &sh->dev[i];
2757 if ((dev->towrite || i == sh->pd_idx) &&
2758 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 2759 !(test_bit(R5_UPTODATE, &dev->flags) ||
2760 test_bit(R5_Wantcompute, &dev->flags))) {
a4456856
DW		 2761 if (test_bit(R5_Insync, &dev->flags))
2762 rmw++;
2763 else
2764 rmw += 2*disks; /* cannot read it */
2765 }
2766 /* Would I have to read this buffer for reconstruct_write */
2767 if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx &&
2768 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 2769 !(test_bit(R5_UPTODATE, &dev->flags) ||
2770 test_bit(R5_Wantcompute, &dev->flags))) {
2771 if (test_bit(R5_Insync, &dev->flags)) rcw++;
a4456856
DW		 2772 else
2773 rcw += 2*disks;
2774 }
2775 }
45b4233c 2776 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
a4456856
DW		 2777 (unsigned long long)sh->sector, rmw, rcw);
2778 set_bit(STRIPE_HANDLE, &sh->state);
2779 if (rmw < rcw && rmw > 0)
2780 /* prefer read-modify-write, but need to get some data */
2781 for (i = disks; i--; ) {
2782 struct r5dev *dev = &sh->dev[i];
2783 if ((dev->towrite || i == sh->pd_idx) &&
2784 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 2785 !(test_bit(R5_UPTODATE, &dev->flags) ||
2786 test_bit(R5_Wantcompute, &dev->flags)) &&
a4456856
DW		 2787 test_bit(R5_Insync, &dev->flags)) {
2788 if (
2789 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
45b4233c 2790 pr_debug("Read_old block "
a4456856
DW		 2791 "%d for r-m-w\n", i);
2792 set_bit(R5_LOCKED, &dev->flags);
2793 set_bit(R5_Wantread, &dev->flags);
2794 s->locked++;
2795 } else {
2796 set_bit(STRIPE_DELAYED, &sh->state);
2797 set_bit(STRIPE_HANDLE, &sh->state);
2798 }
2799 }
2800 }
c8ac1803 2801 if (rcw <= rmw && rcw > 0) {
a4456856 2802 /* want reconstruct write, but need to get some data */
c8ac1803 2803 rcw = 0;
a4456856
DW		 2804 for (i = disks; i--; ) {
2805 struct r5dev *dev = &sh->dev[i];
2806 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
c8ac1803 2807 i != sh->pd_idx && i != sh->qd_idx &&
a4456856 2808 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219 2809 !(test_bit(R5_UPTODATE, &dev->flags) ||
c8ac1803
N		 2810 test_bit(R5_Wantcompute, &dev->flags))) {
2811 rcw++;
2812 if (!test_bit(R5_Insync, &dev->flags))
2813 continue; /* it's a failed drive */
a4456856
DW		 2814 if (
2815 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
45b4233c 2816 pr_debug("Read_old block "
a4456856
DW		 2817 "%d for Reconstruct\n", i);
2818 set_bit(R5_LOCKED, &dev->flags);
2819 set_bit(R5_Wantread, &dev->flags);
2820 s->locked++;
2821 } else {
2822 set_bit(STRIPE_DELAYED, &sh->state);
2823 set_bit(STRIPE_HANDLE, &sh->state);
2824 }
2825 }
2826 }
c8ac1803 2827 }
a4456856
DW		 2828 /* now if nothing is locked, and if we have enough data,
2829 * we can start a write request
2830 */
f38e1219
DW		 2831 /* since handle_stripe can be called at any time we need to handle the
2832 * case where a compute block operation has been submitted and then a
ac6b53b6
DW		 2833 * subsequent call wants to start a write request. raid_run_ops only
2834 * handles the case where compute block and reconstruct are requested
f38e1219
DW		 2835 * simultaneously. If this is not the case then new writes need to be
2836 * held off until the compute completes.
2837 */
976ea8d4
DW		 2838 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
2839 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
2840 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
c0f7bddb 2841 schedule_reconstruction(sh, s, rcw == 0, 0);
a4456856
DW		 2842}
2843
d1688a6d 2844static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh,
a4456856
DW		 2845 struct stripe_head_state *s, int disks)
2846{
ecc65c9b 2847 struct r5dev *dev = NULL;
bd2ab670 2848
a4456856 2849 set_bit(STRIPE_HANDLE, &sh->state);
e89f8962 2850
ecc65c9b
DW		 2851 switch (sh->check_state) {
2852 case check_state_idle:
2853 /* start a new check operation if there are no failures */
bd2ab670 2854 if (s->failed == 0) {
bd2ab670 2855 BUG_ON(s->uptodate != disks);
ecc65c9b
DW		 2856 sh->check_state = check_state_run;
2857 set_bit(STRIPE_OP_CHECK, &s->ops_request);
bd2ab670 2858 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
bd2ab670 2859 s->uptodate--;
ecc65c9b 2860 break;
bd2ab670 2861 }
f2b3b44d 2862 dev = &sh->dev[s->failed_num[0]];
ecc65c9b
DW		 2863 /* fall through */
2864 case check_state_compute_result:
2865 sh->check_state = check_state_idle;
2866 if (!dev)
2867 dev = &sh->dev[sh->pd_idx];
2868
2869 /* check that a write has not made the stripe insync */
2870 if (test_bit(STRIPE_INSYNC, &sh->state))
2871 break;
c8894419 2872
a4456856 2873 /* either failed parity check, or recovery is happening */
a4456856
DW		 2874 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
2875 BUG_ON(s->uptodate != disks);
2876
2877 set_bit(R5_LOCKED, &dev->flags);
ecc65c9b 2878 s->locked++;
a4456856 2879 set_bit(R5_Wantwrite, &dev->flags);
830ea016 2880
a4456856 2881 clear_bit(STRIPE_DEGRADED, &sh->state);
a4456856 2882 set_bit(STRIPE_INSYNC, &sh->state);
ecc65c9b
DW		 2883 break;
2884 case check_state_run:
2885 break; /* we will be called again upon completion */
2886 case check_state_check_result:
2887 sh->check_state = check_state_idle;
2888
2889 /* if a failure occurred during the check operation, leave
2890 * STRIPE_INSYNC not set and let the stripe be handled again
2891 */
2892 if (s->failed)
2893 break;
2894
2895 /* handle a successful check operation, if parity is correct
2896 * we are done. Otherwise update the mismatch count and repair
2897 * parity if !MD_RECOVERY_CHECK
2898 */
ad283ea4 2899 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
ecc65c9b
DW		 2900 /* parity is correct (on disc,
2901 * not in buffer any more)
2902 */
2903 set_bit(STRIPE_INSYNC, &sh->state);
2904 else {
2905 conf->mddev->resync_mismatches += STRIPE_SECTORS;
2906 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
2907 /* don't try to repair!! */
2908 set_bit(STRIPE_INSYNC, &sh->state);
2909 else {
2910 sh->check_state = check_state_compute_run;
976ea8d4 2911 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
ecc65c9b
DW		 2912 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2913 set_bit(R5_Wantcompute,
2914 &sh->dev[sh->pd_idx].flags);
2915 sh->ops.target = sh->pd_idx;
ac6b53b6 2916 sh->ops.target2 = -1;
ecc65c9b
DW		 2917 s->uptodate++;
2918 }
2919 }
2920 break;
2921 case check_state_compute_run:
2922 break;
2923 default:
2924 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
2925 __func__, sh->check_state,
2926 (unsigned long long) sh->sector);
2927 BUG();
a4456856
DW		 2928 }
2929}
2930
2931
d1688a6d 2932static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh,
36d1c647 2933 struct stripe_head_state *s,
f2b3b44d 2934 int disks)
a4456856 2935{
a4456856 2936 int pd_idx = sh->pd_idx;
34e04e87 2937 int qd_idx = sh->qd_idx;
d82dfee0 2938 struct r5dev *dev;
a4456856
DW		 2939
2940 set_bit(STRIPE_HANDLE, &sh->state);
2941
2942 BUG_ON(s->failed > 2);
d82dfee0 2943
a4456856
DW		 2944 /* Want to check and possibly repair P and Q.
2945 * However there could be one 'failed' device, in which
2946 * case we can only check one of them, possibly using the
2947 * other to generate missing data
2948 */
2949
d82dfee0
DW		 2950 switch (sh->check_state) {
2951 case check_state_idle:
2952 /* start a new check operation if there are < 2 failures */
f2b3b44d 2953 if (s->failed == s->q_failed) {
d82dfee0 2954 /* The only possible failed device holds Q, so it
a4456856
DW		 2955 * makes sense to check P (If anything else were failed,
2956 * we would have used P to recreate it).
2957 */
d82dfee0 2958 sh->check_state = check_state_run;
a4456856 2959 }
f2b3b44d 2960 if (!s->q_failed && s->failed < 2) {
d82dfee0 2961 /* Q is not failed, and we didn't use it to generate
a4456856
DW		 2962 * anything, so it makes sense to check it
2963 */
d82dfee0
DW		 2964 if (sh->check_state == check_state_run)
2965 sh->check_state = check_state_run_pq;
2966 else
2967 sh->check_state = check_state_run_q;
a4456856 2968 }
a4456856 2969
d82dfee0
DW		 2970 /* discard potentially stale zero_sum_result */
2971 sh->ops.zero_sum_result = 0;
a4456856 2972
d82dfee0
DW		 2973 if (sh->check_state == check_state_run) {
2974 /* async_xor_zero_sum destroys the contents of P */
2975 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
2976 s->uptodate--;
a4456856 2977 }
d82dfee0
DW		 2978 if (sh->check_state >= check_state_run &&
2979 sh->check_state <= check_state_run_pq) {
2980 /* async_syndrome_zero_sum preserves P and Q, so
2981 * no need to mark them !uptodate here
2982 */
2983 set_bit(STRIPE_OP_CHECK, &s->ops_request);
2984 break;
a4456856
DW		 2985 }
2986
d82dfee0
DW		 2987 /* we have 2-disk failure */
2988 BUG_ON(s->failed != 2);
2989 /* fall through */
2990 case check_state_compute_result:
2991 sh->check_state = check_state_idle;
a4456856 2992
d82dfee0
DW		 2993 /* check that a write has not made the stripe insync */
2994 if (test_bit(STRIPE_INSYNC, &sh->state))
2995 break;
a4456856
DW		 2996
2997 /* now write out any block on a failed drive,
d82dfee0 2998 * or P or Q if they were recomputed
a4456856 2999 */
d82dfee0 3000 BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */
a4456856 3001 if (s->failed == 2) {
f2b3b44d 3002 dev = &sh->dev[s->failed_num[1]];
a4456856
DW		 3003 s->locked++;
3004 set_bit(R5_LOCKED, &dev->flags);
3005 set_bit(R5_Wantwrite, &dev->flags);
3006 }
3007 if (s->failed >= 1) {
f2b3b44d 3008 dev = &sh->dev[s->failed_num[0]];
a4456856
DW		 3009 s->locked++;
3010 set_bit(R5_LOCKED, &dev->flags);
3011 set_bit(R5_Wantwrite, &dev->flags);
3012 }
d82dfee0 3013 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
a4456856
DW		 3014 dev = &sh->dev[pd_idx];
3015 s->locked++;
3016 set_bit(R5_LOCKED, &dev->flags);
3017 set_bit(R5_Wantwrite, &dev->flags);
3018 }
d82dfee0 3019 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
a4456856
DW		 3020 dev = &sh->dev[qd_idx];
3021 s->locked++;
3022 set_bit(R5_LOCKED, &dev->flags);
3023 set_bit(R5_Wantwrite, &dev->flags);
3024 }
3025 clear_bit(STRIPE_DEGRADED, &sh->state);
3026
3027 set_bit(STRIPE_INSYNC, &sh->state);
d82dfee0
DW		 3028 break;
3029 case check_state_run:
3030 case check_state_run_q:
3031 case check_state_run_pq:
3032 break; /* we will be called again upon completion */
3033 case check_state_check_result:
3034 sh->check_state = check_state_idle;
3035
3036 /* handle a successful check operation, if parity is correct
3037 * we are done. Otherwise update the mismatch count and repair
3038 * parity if !MD_RECOVERY_CHECK
3039 */
3040 if (sh->ops.zero_sum_result == 0) {
3041 /* both parities are correct */
3042 if (!s->failed)
3043 set_bit(STRIPE_INSYNC, &sh->state);
3044 else {
3045 /* in contrast to the raid5 case we can validate
3046 * parity, but still have a failure to write
3047 * back
3048 */
3049 sh->check_state = check_state_compute_result;
3050 /* Returning at this point means that we may go
3051 * off and bring p and/or q uptodate again so
3052 * we make sure to check zero_sum_result again
3053 * to verify if p or q need writeback
3054 */
3055 }
3056 } else {
3057 conf->mddev->resync_mismatches += STRIPE_SECTORS;
3058 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
3059 /* don't try to repair!! */
3060 set_bit(STRIPE_INSYNC, &sh->state);
3061 else {
3062 int *target = &sh->ops.target;
3063
3064 sh->ops.target = -1;
3065 sh->ops.target2 = -1;
3066 sh->check_state = check_state_compute_run;
3067 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3068 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3069 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
3070 set_bit(R5_Wantcompute,
3071 &sh->dev[pd_idx].flags);
3072 *target = pd_idx;
3073 target = &sh->ops.target2;
3074 s->uptodate++;
3075 }
3076 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
3077 set_bit(R5_Wantcompute,
3078 &sh->dev[qd_idx].flags);
3079 *target = qd_idx;
3080 s->uptodate++;
3081 }
3082 }
3083 }
3084 break;
3085 case check_state_compute_run:
3086 break;
3087 default:
3088 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
3089 __func__, sh->check_state,
3090 (unsigned long long) sh->sector);
3091 BUG();
a4456856
DW		 3092 }
3093}
3094
d1688a6d 3095static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh)
a4456856
DW		 3096{
3097 int i;
3098
3099 /* We have read all the blocks in this stripe and now we need to
3100 * copy some of them into a target stripe for expand.
3101 */
f0a50d37 3102 struct dma_async_tx_descriptor *tx = NULL;
a4456856
DW		 3103 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3104 for (i = 0; i < sh->disks; i++)
34e04e87 3105 if (i != sh->pd_idx && i != sh->qd_idx) {
911d4ee8 3106 int dd_idx, j;
a4456856 3107 struct stripe_head *sh2;
a08abd8c 3108 struct async_submit_ctl submit;
a4456856 3109
784052ec 3110 sector_t bn = compute_blocknr(sh, i, 1);
911d4ee8
N		 3111 sector_t s = raid5_compute_sector(conf, bn, 0,
3112 &dd_idx, NULL);
a8c906ca 3113 sh2 = get_active_stripe(conf, s, 0, 1, 1);
a4456856
DW		 3114 if (sh2 == NULL)
3115 /* so far only the early blocks of this stripe
3116 * have been requested. When later blocks
3117 * get requested, we will try again
3118 */
3119 continue;
3120 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
3121 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
3122 /* must have already done this block */
3123 release_stripe(sh2);
3124 continue;
3125 }
f0a50d37
DW		 3126
3127 /* place all the copies on one channel */
a08abd8c 3128 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
f0a50d37 3129 tx = async_memcpy(sh2->dev[dd_idx].page,
88ba2aa5 3130 sh->dev[i].page, 0, 0, STRIPE_SIZE,
a08abd8c 3131 &submit);
f0a50d37 3132
a4456856
DW		 3133 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
3134 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
3135 for (j = 0; j < conf->raid_disks; j++)
3136 if (j != sh2->pd_idx &&
86c374ba 3137 j != sh2->qd_idx &&
a4456856
DW		 3138 !test_bit(R5_Expanded, &sh2->dev[j].flags))
3139 break;
3140 if (j == conf->raid_disks) {
3141 set_bit(STRIPE_EXPAND_READY, &sh2->state);
3142 set_bit(STRIPE_HANDLE, &sh2->state);
3143 }
3144 release_stripe(sh2);
f0a50d37 3145
a4456856 3146 }
a2e08551
N		 3147 /* done submitting copies, wait for them to complete */
3148 if (tx) {
3149 async_tx_ack(tx);
3150 dma_wait_for_async_tx(tx);
3151 }
a4456856 3152}
1da177e4
LT		 3153
3154/*
3155 * handle_stripe - do things to a stripe.
3156 *
9a3e1101
N		 3157 * We lock the stripe by setting STRIPE_ACTIVE and then examine the
3158 * state of various bits to see what needs to be done.
1da177e4 3159 * Possible results:
9a3e1101
N		 3160 * return some read requests which now have data
3161 * return some write requests which are safely on storage
1da177e4
LT		 3162 * schedule a read on some buffers
3163 * schedule a write of some buffers
3164 * return confirmation of parity correctness
3165 *
1da177e4 3166 */
a4456856 3167
acfe726b 3168static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s)
1da177e4 3169{
d1688a6d 3170 struct r5conf *conf = sh->raid_conf;
f416885e 3171 int disks = sh->disks;
474af965
N		 3172 struct r5dev *dev;
3173 int i;
9a3e1101 3174 int do_recovery = 0;
1da177e4 3175
acfe726b
N		 3176 memset(s, 0, sizeof(*s));
3177
acfe726b
N		 3178 s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3179 s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
3180 s->failed_num[0] = -1;
3181 s->failed_num[1] = -1;
1da177e4 3182
acfe726b 3183 /* Now to look around and see what can be done */
1da177e4 3184 rcu_read_lock();
c4c1663b 3185 spin_lock_irq(&conf->device_lock);
16a53ecc 3186 for (i=disks; i--; ) {
3cb03002 3187 struct md_rdev *rdev;
31c176ec
N		 3188 sector_t first_bad;
3189 int bad_sectors;
3190 int is_bad = 0;
acfe726b 3191
16a53ecc 3192 dev = &sh->dev[i];
1da177e4 3193
45b4233c 3194 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
9a3e1101
N		 3195 i, dev->flags,
3196 dev->toread, dev->towrite, dev->written);
6c0069c0
YT		 3197 /* maybe we can reply to a read
3198 *
3199 * new wantfill requests are only permitted while
3200 * ops_complete_biofill is guaranteed to be inactive
3201 */
3202 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
3203 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
3204 set_bit(R5_Wantfill, &dev->flags);
1da177e4 3205
16a53ecc 3206 /* now count some things */
cc94015a
N		 3207 if (test_bit(R5_LOCKED, &dev->flags))
3208 s->locked++;
3209 if (test_bit(R5_UPTODATE, &dev->flags))
3210 s->uptodate++;
2d6e4ecc 3211 if (test_bit(R5_Wantcompute, &dev->flags)) {
cc94015a
N		 3212 s->compute++;
3213 BUG_ON(s->compute > 2);
2d6e4ecc 3214 }
1da177e4 3215
acfe726b 3216 if (test_bit(R5_Wantfill, &dev->flags))
cc94015a 3217 s->to_fill++;
acfe726b 3218 else if (dev->toread)
cc94015a 3219 s->to_read++;
16a53ecc 3220 if (dev->towrite) {
cc94015a 3221 s->to_write++;
16a53ecc 3222 if (!test_bit(R5_OVERWRITE, &dev->flags))
cc94015a 3223 s->non_overwrite++;
16a53ecc 3224 }
a4456856 3225 if (dev->written)
cc94015a 3226 s->written++;
14a75d3e
N		 3227 /* Prefer to use the replacement for reads, but only
3228 * if it is recovered enough and has no bad blocks.
3229 */
3230 rdev = rcu_dereference(conf->disks[i].replacement);
3231 if (rdev && !test_bit(Faulty, &rdev->flags) &&
3232 rdev->recovery_offset >= sh->sector + STRIPE_SECTORS &&
3233 !is_badblock(rdev, sh->sector, STRIPE_SECTORS,
3234 &first_bad, &bad_sectors))
3235 set_bit(R5_ReadRepl, &dev->flags);
3236 else {
9a3e1101
N		 3237 if (rdev)
3238 set_bit(R5_NeedReplace, &dev->flags);
14a75d3e
N		 3239 rdev = rcu_dereference(conf->disks[i].rdev);
3240 clear_bit(R5_ReadRepl, &dev->flags);
3241 }
9283d8c5
N		 3242 if (rdev && test_bit(Faulty, &rdev->flags))
3243 rdev = NULL;
31c176ec
N		 3244 if (rdev) {
3245 is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
3246 &first_bad, &bad_sectors);
3247 if (s->blocked_rdev == NULL
3248 && (test_bit(Blocked, &rdev->flags)
3249 || is_bad < 0)) {
3250 if (is_bad < 0)
3251 set_bit(BlockedBadBlocks,
3252 &rdev->flags);
3253 s->blocked_rdev = rdev;
3254 atomic_inc(&rdev->nr_pending);
3255 }
6bfe0b49 3256 }
415e72d0
N		 3257 clear_bit(R5_Insync, &dev->flags);
3258 if (!rdev)
3259 /* Not in-sync */;
31c176ec
N		 3260 else if (is_bad) {
3261 /* also not in-sync */
18b9837e
N		 3262 if (!test_bit(WriteErrorSeen, &rdev->flags) &&
3263 test_bit(R5_UPTODATE, &dev->flags)) {
31c176ec
N		 3264 /* treat as in-sync, but with a read error
3265 * which we can now try to correct
3266 */
3267 set_bit(R5_Insync, &dev->flags);
3268 set_bit(R5_ReadError, &dev->flags);
3269 }
3270 } else if (test_bit(In_sync, &rdev->flags))
415e72d0 3271 set_bit(R5_Insync, &dev->flags);
30d7a483 3272 else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
415e72d0 3273 /* in sync if before recovery_offset */
30d7a483
N		 3274 set_bit(R5_Insync, &dev->flags);
3275 else if (test_bit(R5_UPTODATE, &dev->flags) &&
3276 test_bit(R5_Expanded, &dev->flags))
3277 /* If we've reshaped into here, we assume it is Insync.
3278 * We will shortly update recovery_offset to make
3279 * it official.
3280 */
3281 set_bit(R5_Insync, &dev->flags);
3282
5d8c71f9 3283 if (rdev && test_bit(R5_WriteError, &dev->flags)) {
14a75d3e
N		 3284 /* This flag does not apply to '.replacement'
3285 * only to .rdev, so make sure to check that*/
3286 struct md_rdev *rdev2 = rcu_dereference(
3287 conf->disks[i].rdev);
3288 if (rdev2 == rdev)
3289 clear_bit(R5_Insync, &dev->flags);
3290 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
bc2607f3 3291 s->handle_bad_blocks = 1;
14a75d3e 3292 atomic_inc(&rdev2->nr_pending);
bc2607f3
N		 3293 } else
3294 clear_bit(R5_WriteError, &dev->flags);
3295 }
5d8c71f9 3296 if (rdev && test_bit(R5_MadeGood, &dev->flags)) {
14a75d3e
N		 3297 /* This flag does not apply to '.replacement'
3298 * only to .rdev, so make sure to check that*/
3299 struct md_rdev *rdev2 = rcu_dereference(
3300 conf->disks[i].rdev);
3301 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
b84db560 3302 s->handle_bad_blocks = 1;
14a75d3e 3303 atomic_inc(&rdev2->nr_pending);
b84db560
N		 3304 } else
3305 clear_bit(R5_MadeGood, &dev->flags);
3306 }
977df362
N		 3307 if (test_bit(R5_MadeGoodRepl, &dev->flags)) {
3308 struct md_rdev *rdev2 = rcu_dereference(
3309 conf->disks[i].replacement);
3310 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
3311 s->handle_bad_blocks = 1;
3312 atomic_inc(&rdev2->nr_pending);
3313 } else
3314 clear_bit(R5_MadeGoodRepl, &dev->flags);
3315 }
415e72d0 3316 if (!test_bit(R5_Insync, &dev->flags)) {
16a53ecc
N		 3317 /* The ReadError flag will just be confusing now */
3318 clear_bit(R5_ReadError, &dev->flags);
3319 clear_bit(R5_ReWrite, &dev->flags);
1da177e4 3320 }
415e72d0
N		 3321 if (test_bit(R5_ReadError, &dev->flags))
3322 clear_bit(R5_Insync, &dev->flags);
3323 if (!test_bit(R5_Insync, &dev->flags)) {
cc94015a
N		 3324 if (s->failed < 2)
3325 s->failed_num[s->failed] = i;
3326 s->failed++;
9a3e1101
N		 3327 if (rdev && !test_bit(Faulty, &rdev->flags))
3328 do_recovery = 1;
415e72d0 3329 }
1da177e4 3330 }
c4c1663b 3331 spin_unlock_irq(&conf->device_lock);
9a3e1101
N		 3332 if (test_bit(STRIPE_SYNCING, &sh->state)) {
3333 /* If there is a failed device being replaced,
3334 * we must be recovering.
3335 * else if we are after recovery_cp, we must be syncing
c6d2e084 3336 * else if MD_RECOVERY_REQUESTED is set, we also are syncing.
9a3e1101
N		 3337 * else we can only be replacing
3338 * sync and recovery both need to read all devices, and so
3339 * use the same flag.
3340 */
3341 if (do_recovery ||
c6d2e084 3342 sh->sector >= conf->mddev->recovery_cp ||
3343 test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery)))
9a3e1101
N		 3344 s->syncing = 1;
3345 else
3346 s->replacing = 1;
3347 }
1da177e4 3348 rcu_read_unlock();
cc94015a
N		 3349}
3350
3351static void handle_stripe(struct stripe_head *sh)
3352{
3353 struct stripe_head_state s;
d1688a6d 3354 struct r5conf *conf = sh->raid_conf;
3687c061 3355 int i;
84789554
N		 3356 int prexor;
3357 int disks = sh->disks;
474af965 3358 struct r5dev *pdev, *qdev;
cc94015a
N		 3359
3360 clear_bit(STRIPE_HANDLE, &sh->state);
257a4b42 3361 if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) {
cc94015a
N		 3362 /* already being handled, ensure it gets handled
3363 * again when current action finishes */
3364 set_bit(STRIPE_HANDLE, &sh->state);
3365 return;
3366 }
3367
3368 if (test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
3369 set_bit(STRIPE_SYNCING, &sh->state);
3370 clear_bit(STRIPE_INSYNC, &sh->state);
3371 }
3372 clear_bit(STRIPE_DELAYED, &sh->state);
3373
3374 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
3375 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
3376 (unsigned long long)sh->sector, sh->state,
3377 atomic_read(&sh->count), sh->pd_idx, sh->qd_idx,
3378 sh->check_state, sh->reconstruct_state);
3687c061 3379
acfe726b 3380 analyse_stripe(sh, &s);
c5a31000 3381
bc2607f3
N		 3382 if (s.handle_bad_blocks) {
3383 set_bit(STRIPE_HANDLE, &sh->state);
3384 goto finish;
3385 }
3386
474af965
N		 3387 if (unlikely(s.blocked_rdev)) {
3388 if (s.syncing || s.expanding || s.expanded ||
9a3e1101 3389 s.replacing || s.to_write || s.written) {
474af965
N		 3390 set_bit(STRIPE_HANDLE, &sh->state);
3391 goto finish;
3392 }
3393 /* There is nothing for the blocked_rdev to block */
3394 rdev_dec_pending(s.blocked_rdev, conf->mddev);
3395 s.blocked_rdev = NULL;
3396 }
3397
3398 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
3399 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
3400 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
3401 }
3402
3403 pr_debug("locked=%d uptodate=%d to_read=%d"
3404 " to_write=%d failed=%d failed_num=%d,%d\n",
3405 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
3406 s.failed_num[0], s.failed_num[1]);
3407 /* check if the array has lost more than max_degraded devices and,
3408 * if so, some requests might need to be failed.
3409 */
9a3f530f
N		 3410 if (s.failed > conf->max_degraded) {
3411 sh->check_state = 0;
3412 sh->reconstruct_state = 0;
3413 if (s.to_read+s.to_write+s.written)
3414 handle_failed_stripe(conf, sh, &s, disks, &s.return_bi);
9a3e1101 3415 if (s.syncing + s.replacing)
9a3f530f
N		 3416 handle_failed_sync(conf, sh, &s);
3417 }
474af965
N		 3418
3419 /*
3420 * might be able to return some write requests if the parity blocks
3421 * are safe, or on a failed drive
3422 */
3423 pdev = &sh->dev[sh->pd_idx];
3424 s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx)
3425 || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx);
3426 qdev = &sh->dev[sh->qd_idx];
3427 s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx)
3428 || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx)
3429 || conf->level < 6;
3430
3431 if (s.written &&
3432 (s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
3433 && !test_bit(R5_LOCKED, &pdev->flags)
3434 && test_bit(R5_UPTODATE, &pdev->flags)))) &&
3435 (s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
3436 && !test_bit(R5_LOCKED, &qdev->flags)
3437 && test_bit(R5_UPTODATE, &qdev->flags)))))
3438 handle_stripe_clean_event(conf, sh, disks, &s.return_bi);
3439
3440 /* Now we might consider reading some blocks, either to check/generate
3441 * parity, or to satisfy requests
3442 * or to load a block that is being partially written.
3443 */
3444 if (s.to_read || s.non_overwrite
3445 || (conf->level == 6 && s.to_write && s.failed)
9a3e1101
N		 3446 || (s.syncing && (s.uptodate + s.compute < disks))
3447 || s.replacing
3448 || s.expanding)
474af965
N		 3449 handle_stripe_fill(sh, &s, disks);
3450
84789554
N		 3451 /* Now we check to see if any write operations have recently
3452 * completed
3453 */
3454 prexor = 0;
3455 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
3456 prexor = 1;
3457 if (sh->reconstruct_state == reconstruct_state_drain_result ||
3458 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
3459 sh->reconstruct_state = reconstruct_state_idle;
3460
3461 /* All the 'written' buffers and the parity block are ready to
3462 * be written back to disk
3463 */
3464 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags));
3465 BUG_ON(sh->qd_idx >= 0 &&
3466 !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags));
3467 for (i = disks; i--; ) {
3468 struct r5dev *dev = &sh->dev[i];
3469 if (test_bit(R5_LOCKED, &dev->flags) &&
3470 (i == sh->pd_idx || i == sh->qd_idx ||
3471 dev->written)) {
3472 pr_debug("Writing block %d\n", i);
3473 set_bit(R5_Wantwrite, &dev->flags);
3474 if (prexor)
3475 continue;
3476 if (!test_bit(R5_Insync, &dev->flags) ||
3477 ((i == sh->pd_idx || i == sh->qd_idx) &&
3478 s.failed == 0))
3479 set_bit(STRIPE_INSYNC, &sh->state);
3480 }
3481 }
3482 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3483 s.dec_preread_active = 1;
3484 }
3485
3486 /* Now to consider new write requests and what else, if anything
3487 * should be read. We do not handle new writes when:
3488 * 1/ A 'write' operation (copy+xor) is already in flight.
3489 * 2/ A 'check' operation is in flight, as it may clobber the parity
3490 * block.
3491 */
3492 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
3493 handle_stripe_dirtying(conf, sh, &s, disks);
3494
3495 /* maybe we need to check and possibly fix the parity for this stripe
3496 * Any reads will already have been scheduled, so we just see if enough
3497 * data is available. The parity check is held off while parity
3498 * dependent operations are in flight.
3499 */
3500 if (sh->check_state ||
3501 (s.syncing && s.locked == 0 &&
3502 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
3503 !test_bit(STRIPE_INSYNC, &sh->state))) {
3504 if (conf->level == 6)
3505 handle_parity_checks6(conf, sh, &s, disks);
3506 else
3507 handle_parity_checks5(conf, sh, &s, disks);
3508 }
c5a31000 3509
9a3e1101
N		 3510 if (s.replacing && s.locked == 0
3511 && !test_bit(STRIPE_INSYNC, &sh->state)) {
3512 /* Write out to replacement devices where possible */
3513 for (i = 0; i < conf->raid_disks; i++)
3514 if (test_bit(R5_UPTODATE, &sh->dev[i].flags) &&
3515 test_bit(R5_NeedReplace, &sh->dev[i].flags)) {
3516 set_bit(R5_WantReplace, &sh->dev[i].flags);
3517 set_bit(R5_LOCKED, &sh->dev[i].flags);
3518 s.locked++;
3519 }
3520 set_bit(STRIPE_INSYNC, &sh->state);
3521 }
3522 if ((s.syncing || s.replacing) && s.locked == 0 &&
3523 test_bit(STRIPE_INSYNC, &sh->state)) {
c5a31000
N		 3524 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3525 clear_bit(STRIPE_SYNCING, &sh->state);
3526 }
3527
3528 /* If the failed drives are just a ReadError, then we might need
3529 * to progress the repair/check process
3530 */
3531 if (s.failed <= conf->max_degraded && !conf->mddev->ro)
3532 for (i = 0; i < s.failed; i++) {
3533 struct r5dev *dev = &sh->dev[s.failed_num[i]];
3534 if (test_bit(R5_ReadError, &dev->flags)
3535 && !test_bit(R5_LOCKED, &dev->flags)
3536 && test_bit(R5_UPTODATE, &dev->flags)
3537 ) {
3538 if (!test_bit(R5_ReWrite, &dev->flags)) {
3539 set_bit(R5_Wantwrite, &dev->flags);
3540 set_bit(R5_ReWrite, &dev->flags);
3541 set_bit(R5_LOCKED, &dev->flags);
3542 s.locked++;
3543 } else {
3544 /* let's read it back */
3545 set_bit(R5_Wantread, &dev->flags);
3546 set_bit(R5_LOCKED, &dev->flags);
3547 s.locked++;
3548 }
3549 }
3550 }
3551
3552
3687c061
N		 3553 /* Finish reconstruct operations initiated by the expansion process */
3554 if (sh->reconstruct_state == reconstruct_state_result) {
3555 struct stripe_head *sh_src
3556 = get_active_stripe(conf, sh->sector, 1, 1, 1);
3557 if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) {
3558 /* sh cannot be written until sh_src has been read.
3559 * so arrange for sh to be delayed a little
3560 */
3561 set_bit(STRIPE_DELAYED, &sh->state);
3562 set_bit(STRIPE_HANDLE, &sh->state);
3563 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
3564 &sh_src->state))
3565 atomic_inc(&conf->preread_active_stripes);
3566 release_stripe(sh_src);
3567 goto finish;
3568 }
3569 if (sh_src)
3570 release_stripe(sh_src);
3571
3572 sh->reconstruct_state = reconstruct_state_idle;
3573 clear_bit(STRIPE_EXPANDING, &sh->state);
3574 for (i = conf->raid_disks; i--; ) {
3575 set_bit(R5_Wantwrite, &sh->dev[i].flags);
3576 set_bit(R5_LOCKED, &sh->dev[i].flags);
3577 s.locked++;
3578 }
3579 }
f416885e 3580
3687c061
N		 3581 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
3582 !sh->reconstruct_state) {
3583 /* Need to write out all blocks after computing parity */
3584 sh->disks = conf->raid_disks;
3585 stripe_set_idx(sh->sector, conf, 0, sh);
3586 schedule_reconstruction(sh, &s, 1, 1);
3587 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
3588 clear_bit(STRIPE_EXPAND_READY, &sh->state);
3589 atomic_dec(&conf->reshape_stripes);
3590 wake_up(&conf->wait_for_overlap);
3591 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3592 }
3593
3594 if (s.expanding && s.locked == 0 &&
3595 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
3596 handle_stripe_expansion(conf, sh);
16a53ecc 3597
3687c061 3598finish:
6bfe0b49 3599 /* wait for this device to become unblocked */
5f066c63
N		 3600 if (unlikely(s.blocked_rdev)) {
3601 if (conf->mddev->external)
3602 md_wait_for_blocked_rdev(s.blocked_rdev,
3603 conf->mddev);
3604 else
3605 /* Internal metadata will immediately
3606 * be written by raid5d, so we don't
3607 * need to wait here.
3608 */
3609 rdev_dec_pending(s.blocked_rdev,
3610 conf->mddev);
3611 }
6bfe0b49 3612
bc2607f3
N		 3613 if (s.handle_bad_blocks)
3614 for (i = disks; i--; ) {
3cb03002 3615 struct md_rdev *rdev;
bc2607f3
N		 3616 struct r5dev *dev = &sh->dev[i];
3617 if (test_and_clear_bit(R5_WriteError, &dev->flags)) {
3618 /* We own a safe reference to the rdev */
3619 rdev = conf->disks[i].rdev;
3620 if (!rdev_set_badblocks(rdev, sh->sector,
3621 STRIPE_SECTORS, 0))
3622 md_error(conf->mddev, rdev);
3623 rdev_dec_pending(rdev, conf->mddev);
3624 }
b84db560
N		 3625 if (test_and_clear_bit(R5_MadeGood, &dev->flags)) {
3626 rdev = conf->disks[i].rdev;
3627 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 3628 STRIPE_SECTORS, 0);
b84db560
N		 3629 rdev_dec_pending(rdev, conf->mddev);
3630 }
977df362
N		 3631 if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) {
3632 rdev = conf->disks[i].replacement;
dd054fce
N		 3633 if (!rdev)
3634 /* rdev have been moved down */
3635 rdev = conf->disks[i].rdev;
977df362 3636 rdev_clear_badblocks(rdev, sh->sector,
c6563a8c 3637 STRIPE_SECTORS, 0);
977df362
N		 3638 rdev_dec_pending(rdev, conf->mddev);
3639 }
bc2607f3
N		 3640 }
3641
6c0069c0
YT		 3642 if (s.ops_request)
3643 raid_run_ops(sh, s.ops_request);
3644
f0e43bcd 3645 ops_run_io(sh, &s);
16a53ecc 3646
c5709ef6 3647 if (s.dec_preread_active) {
729a1866 3648 /* We delay this until after ops_run_io so that if make_request
e9c7469b 3649 * is waiting on a flush, it won't continue until the writes
729a1866
N		 3650 * have actually been submitted.
3651 */
3652 atomic_dec(&conf->preread_active_stripes);
3653 if (atomic_read(&conf->preread_active_stripes) <
3654 IO_THRESHOLD)
3655 md_wakeup_thread(conf->mddev->thread);
3656 }
3657
c5709ef6 3658 return_io(s.return_bi);
16a53ecc 3659
257a4b42 3660 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
16a53ecc
N		 3661}
3662
d1688a6d 3663static void raid5_activate_delayed(struct r5conf *conf)
16a53ecc
N		 3664{
3665 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
3666 while (!list_empty(&conf->delayed_list)) {
3667 struct list_head *l = conf->delayed_list.next;
3668 struct stripe_head *sh;
3669 sh = list_entry(l, struct stripe_head, lru);
3670 list_del_init(l);
3671 clear_bit(STRIPE_DELAYED, &sh->state);
3672 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3673 atomic_inc(&conf->preread_active_stripes);
8b3e6cdc 3674 list_add_tail(&sh->lru, &conf->hold_list);
16a53ecc 3675 }
482c0834 3676 }
16a53ecc
N		 3677}
3678
d1688a6d 3679static void activate_bit_delay(struct r5conf *conf)
16a53ecc
N		 3680{
3681 /* device_lock is held */
3682 struct list_head head;
3683 list_add(&head, &conf->bitmap_list);
3684 list_del_init(&conf->bitmap_list);
3685 while (!list_empty(&head)) {
3686 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
3687 list_del_init(&sh->lru);
3688 atomic_inc(&sh->count);
3689 __release_stripe(conf, sh);
3690 }
3691}
3692
fd01b88c 3693int md_raid5_congested(struct mddev *mddev, int bits)
f022b2fd 3694{
d1688a6d 3695 struct r5conf *conf = mddev->private;
f022b2fd
N		 3696
3697 /* No difference between reads and writes. Just check
3698 * how busy the stripe_cache is
3699 */
3fa841d7 3700
f022b2fd
N		 3701 if (conf->inactive_blocked)
3702 return 1;
3703 if (conf->quiesce)
3704 return 1;
3705 if (list_empty_careful(&conf->inactive_list))
3706 return 1;
3707
3708 return 0;
3709}
11d8a6e3
N		 3710EXPORT_SYMBOL_GPL(md_raid5_congested);
3711
3712static int raid5_congested(void *data, int bits)
3713{
fd01b88c 3714 struct mddev *mddev = data;
11d8a6e3
N		 3715
3716 return mddev_congested(mddev, bits) ||
3717 md_raid5_congested(mddev, bits);
3718}
f022b2fd 3719
23032a0e
RBJ		 3720/* We want read requests to align with chunks where possible,
3721 * but write requests don't need to.
3722 */
cc371e66
AK		 3723static int raid5_mergeable_bvec(struct request_queue *q,
3724 struct bvec_merge_data *bvm,
3725 struct bio_vec *biovec)
23032a0e 3726{
fd01b88c 3727 struct mddev *mddev = q->queuedata;
cc371e66 3728 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
23032a0e 3729 int max;
9d8f0363 3730 unsigned int chunk_sectors = mddev->chunk_sectors;
cc371e66 3731 unsigned int bio_sectors = bvm->bi_size >> 9;
23032a0e 3732
cc371e66 3733 if ((bvm->bi_rw & 1) == WRITE)
23032a0e
RBJ		 3734 return biovec->bv_len; /* always allow writes to be mergeable */
3735
664e7c41
AN		 3736 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
3737 chunk_sectors = mddev->new_chunk_sectors;
23032a0e
RBJ		 3738 max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
3739 if (max < 0) max = 0;
3740 if (max <= biovec->bv_len && bio_sectors == 0)
3741 return biovec->bv_len;
3742 else
3743 return max;
3744}
3745
f679623f 3746
fd01b88c 3747static int in_chunk_boundary(struct mddev *mddev, struct bio *bio)
f679623f
RBJ		 3748{
3749 sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
9d8f0363 3750 unsigned int chunk_sectors = mddev->chunk_sectors;
f679623f
RBJ		 3751 unsigned int bio_sectors = bio->bi_size >> 9;
3752
664e7c41
AN		 3753 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
3754 chunk_sectors = mddev->new_chunk_sectors;
f679623f
RBJ		 3755 return chunk_sectors >=
3756 ((sector & (chunk_sectors - 1)) + bio_sectors);
3757}
3758
46031f9a
RBJ		 3759/*
3760 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
3761 * later sampled by raid5d.
3762 */
d1688a6d 3763static void add_bio_to_retry(struct bio *bi,struct r5conf *conf)
46031f9a
RBJ		 3764{
3765 unsigned long flags;
3766
3767 spin_lock_irqsave(&conf->device_lock, flags);
3768
3769 bi->bi_next = conf->retry_read_aligned_list;
3770 conf->retry_read_aligned_list = bi;
3771
3772 spin_unlock_irqrestore(&conf->device_lock, flags);
3773 md_wakeup_thread(conf->mddev->thread);
3774}
3775
3776
d1688a6d 3777static struct bio *remove_bio_from_retry(struct r5conf *conf)
46031f9a
RBJ		 3778{
3779 struct bio *bi;
3780
3781 bi = conf->retry_read_aligned;
3782 if (bi) {
3783 conf->retry_read_aligned = NULL;
3784 return bi;
3785 }
3786 bi = conf->retry_read_aligned_list;
3787 if(bi) {
387bb173 3788 conf->retry_read_aligned_list = bi->bi_next;
46031f9a 3789 bi->bi_next = NULL;
960e739d
JA		 3790 /*
3791 * this sets the active strip count to 1 and the processed
3792 * strip count to zero (upper 8 bits)
3793 */
46031f9a 3794 bi->bi_phys_segments = 1; /* biased count of active stripes */
46031f9a
RBJ		 3795 }
3796
3797 return bi;
3798}
3799
3800
f679623f
RBJ		 3801/*
3802 * The "raid5_align_endio" should check if the read succeeded and if it
3803 * did, call bio_endio on the original bio (having bio_put the new bio
3804 * first).
3805 * If the read failed..
3806 */
6712ecf8 3807static void raid5_align_endio(struct bio *bi, int error)
f679623f
RBJ		 3808{
3809 struct bio* raid_bi = bi->bi_private;
fd01b88c 3810 struct mddev *mddev;
d1688a6d 3811 struct r5conf *conf;
46031f9a 3812 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
3cb03002 3813 struct md_rdev *rdev;
46031f9a 3814
f679623f 3815 bio_put(bi);
46031f9a 3816
46031f9a
RBJ		 3817 rdev = (void*)raid_bi->bi_next;
3818 raid_bi->bi_next = NULL;
2b7f2228
N		 3819 mddev = rdev->mddev;
3820 conf = mddev->private;
46031f9a
RBJ		 3821
3822 rdev_dec_pending(rdev, conf->mddev);
3823
3824 if (!error && uptodate) {
6712ecf8 3825 bio_endio(raid_bi, 0);
46031f9a
RBJ		 3826 if (atomic_dec_and_test(&conf->active_aligned_reads))
3827 wake_up(&conf->wait_for_stripe);
6712ecf8 3828 return;
46031f9a
RBJ		 3829 }
3830
3831
45b4233c 3832 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
46031f9a
RBJ		 3833
3834 add_bio_to_retry(raid_bi, conf);
f679623f
RBJ		 3835}
3836
387bb173
NB		 3837static int bio_fits_rdev(struct bio *bi)
3838{
165125e1 3839 struct request_queue *q = bdev_get_queue(bi->bi_bdev);
387bb173 3840
ae03bf63 3841 if ((bi->bi_size>>9) > queue_max_sectors(q))
387bb173
NB		 3842 return 0;
3843 blk_recount_segments(q, bi);
8a78362c 3844 if (bi->bi_phys_segments > queue_max_segments(q))
387bb173
NB		 3845 return 0;
3846
3847 if (q->merge_bvec_fn)
3848 /* it's too hard to apply the merge_bvec_fn at this stage,
3849 * just just give up
3850 */
3851 return 0;
3852
3853 return 1;
3854}
3855
3856
fd01b88c 3857static int chunk_aligned_read(struct mddev *mddev, struct bio * raid_bio)
f679623f 3858{
d1688a6d 3859 struct r5conf *conf = mddev->private;
8553fe7e 3860 int dd_idx;
f679623f 3861 struct bio* align_bi;
3cb03002 3862 struct md_rdev *rdev;
671488cc 3863 sector_t end_sector;
f679623f
RBJ		 3864
3865 if (!in_chunk_boundary(mddev, raid_bio)) {
45b4233c 3866 pr_debug("chunk_aligned_read : non aligned\n");
f679623f
RBJ		 3867 return 0;
3868 }
3869 /*
a167f663 3870 * use bio_clone_mddev to make a copy of the bio
f679623f 3871 */
a167f663 3872 align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev);
f679623f
RBJ		 3873 if (!align_bi)
3874 return 0;
3875 /*
3876 * set bi_end_io to a new function, and set bi_private to the
3877 * original bio.
3878 */
3879 align_bi->bi_end_io = raid5_align_endio;
3880 align_bi->bi_private = raid_bio;
3881 /*
3882 * compute position
3883 */
112bf897
N		 3884 align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector,
3885 0,
911d4ee8 3886 &dd_idx, NULL);
f679623f 3887
671488cc 3888 end_sector = align_bi->bi_sector + (align_bi->bi_size >> 9);
f679623f 3889 rcu_read_lock();
671488cc
N		 3890 rdev = rcu_dereference(conf->disks[dd_idx].replacement);
3891 if (!rdev || test_bit(Faulty, &rdev->flags) ||
3892 rdev->recovery_offset < end_sector) {
3893 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
3894 if (rdev &&
3895 (test_bit(Faulty, &rdev->flags) ||
3896 !(test_bit(In_sync, &rdev->flags) ||
3897 rdev->recovery_offset >= end_sector)))
3898 rdev = NULL;
3899 }
3900 if (rdev) {
31c176ec
N		 3901 sector_t first_bad;
3902 int bad_sectors;
3903
f679623f
RBJ		 3904 atomic_inc(&rdev->nr_pending);
3905 rcu_read_unlock();
46031f9a
RBJ		 3906 raid_bio->bi_next = (void*)rdev;
3907 align_bi->bi_bdev = rdev->bdev;
3908 align_bi->bi_flags &= ~(1 << BIO_SEG_VALID);
46031f9a 3909
31c176ec
N		 3910 if (!bio_fits_rdev(align_bi) ||
3911 is_badblock(rdev, align_bi->bi_sector, align_bi->bi_size>>9,
3912 &first_bad, &bad_sectors)) {
3913 /* too big in some way, or has a known bad block */
387bb173
NB		 3914 bio_put(align_bi);
3915 rdev_dec_pending(rdev, mddev);
3916 return 0;
3917 }
3918
6c0544e2 3919 /* No reshape active, so we can trust rdev->data_offset */
3920 align_bi->bi_sector += rdev->data_offset;
3921
46031f9a
RBJ		 3922 spin_lock_irq(&conf->device_lock);
3923 wait_event_lock_irq(conf->wait_for_stripe,
3924 conf->quiesce == 0,
3925 conf->device_lock, /* nothing */);
3926 atomic_inc(&conf->active_aligned_reads);
3927 spin_unlock_irq(&conf->device_lock);
3928
f679623f
RBJ		 3929 generic_make_request(align_bi);
3930 return 1;
3931 } else {
3932 rcu_read_unlock();
46031f9a 3933 bio_put(align_bi);
f679623f
RBJ		 3934 return 0;
3935 }
3936}
3937
8b3e6cdc
DW		 3938/* __get_priority_stripe - get the next stripe to process
3939 *
3940 * Full stripe writes are allowed to pass preread active stripes up until
3941 * the bypass_threshold is exceeded. In general the bypass_count
3942 * increments when the handle_list is handled before the hold_list; however, it
3943 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
3944 * stripe with in flight i/o. The bypass_count will be reset when the
3945 * head of the hold_list has changed, i.e. the head was promoted to the
3946 * handle_list.
3947 */
d1688a6d 3948static struct stripe_head *__get_priority_stripe(struct r5conf *conf)
8b3e6cdc
DW		 3949{
3950 struct stripe_head *sh;
3951
3952 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
3953 __func__,
3954 list_empty(&conf->handle_list) ? "empty" : "busy",
3955 list_empty(&conf->hold_list) ? "empty" : "busy",
3956 atomic_read(&conf->pending_full_writes), conf->bypass_count);
3957
3958 if (!list_empty(&conf->handle_list)) {
3959 sh = list_entry(conf->handle_list.next, typeof(*sh), lru);
3960
3961 if (list_empty(&conf->hold_list))
3962 conf->bypass_count = 0;
3963 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
3964 if (conf->hold_list.next == conf->last_hold)
3965 conf->bypass_count++;
3966 else {
3967 conf->last_hold = conf->hold_list.next;
3968 conf->bypass_count -= conf->bypass_threshold;
3969 if (conf->bypass_count < 0)
3970 conf->bypass_count = 0;
3971 }
3972 }
3973 } else if (!list_empty(&conf->hold_list) &&
3974 ((conf->bypass_threshold &&
3975 conf->bypass_count > conf->bypass_threshold) ||
3976 atomic_read(&conf->pending_full_writes) == 0)) {
3977 sh = list_entry(conf->hold_list.next,
3978 typeof(*sh), lru);
3979 conf->bypass_count -= conf->bypass_threshold;
3980 if (conf->bypass_count < 0)
3981 conf->bypass_count = 0;
3982 } else
3983 return NULL;
3984
3985 list_del_init(&sh->lru);
3986 atomic_inc(&sh->count);
3987 BUG_ON(atomic_read(&sh->count) != 1);
3988 return sh;
3989}
f679623f 3990
b4fdcb02 3991static void make_request(struct mddev *mddev, struct bio * bi)
1da177e4 3992{
d1688a6d 3993 struct r5conf *conf = mddev->private;
911d4ee8 3994 int dd_idx;
1da177e4
LT		 3995 sector_t new_sector;
3996 sector_t logical_sector, last_sector;
3997 struct stripe_head *sh;
a362357b 3998 const int rw = bio_data_dir(bi);
49077326 3999 int remaining;
1da177e4 4000
e9c7469b
TH		 4001 if (unlikely(bi->bi_rw & REQ_FLUSH)) {
4002 md_flush_request(mddev, bi);
5a7bbad2 4003 return;
e5dcdd80
N		 4004 }
4005
3d310eb7 4006 md_write_start(mddev, bi);
06d91a5f 4007
802ba064 4008 if (rw == READ &&
52488615 4009 mddev->reshape_position == MaxSector &&
21a52c6d 4010 chunk_aligned_read(mddev,bi))
5a7bbad2 4011 return;
52488615 4012
1da177e4
LT		 4013 logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
4014 last_sector = bi->bi_sector + (bi->bi_size>>9);
4015 bi->bi_next = NULL;
4016 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
06d91a5f 4017
1da177e4
LT		 4018 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
4019 DEFINE_WAIT(w);
b5663ba4 4020 int previous;
b578d55f 4021
7ecaa1e6 4022 retry:
b5663ba4 4023 previous = 0;
b578d55f 4024 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
b0f9ec04 4025 if (unlikely(conf->reshape_progress != MaxSector)) {
fef9c61f 4026 /* spinlock is needed as reshape_progress may be
df8e7f76
N		 4027 * 64bit on a 32bit platform, and so it might be
4028 * possible to see a half-updated value
aeb878b0 4029 * Of course reshape_progress could change after
df8e7f76
N		 4030 * the lock is dropped, so once we get a reference
4031 * to the stripe that we think it is, we will have
4032 * to check again.
4033 */
7ecaa1e6 4034 spin_lock_irq(&conf->device_lock);
2c810cdd 4035 if (mddev->reshape_backwards
fef9c61f
N		 4036 ? logical_sector < conf->reshape_progress
4037 : logical_sector >= conf->reshape_progress) {
b5663ba4
N		 4038 previous = 1;
4039 } else {
2c810cdd 4040 if (mddev->reshape_backwards
fef9c61f
N		 4041 ? logical_sector < conf->reshape_safe
4042 : logical_sector >= conf->reshape_safe) {
b578d55f
N		 4043 spin_unlock_irq(&conf->device_lock);
4044 schedule();
4045 goto retry;
4046 }
4047 }
7ecaa1e6
N		 4048 spin_unlock_irq(&conf->device_lock);
4049 }
16a53ecc 4050
112bf897
N		 4051 new_sector = raid5_compute_sector(conf, logical_sector,
4052 previous,
911d4ee8 4053 &dd_idx, NULL);
0c55e022 4054 pr_debug("raid456: make_request, sector %llu logical %llu\n",
1da177e4
LT		 4055 (unsigned long long)new_sector,
4056 (unsigned long long)logical_sector);
4057
b5663ba4 4058 sh = get_active_stripe(conf, new_sector, previous,
a8c906ca 4059 (bi->bi_rw&RWA_MASK), 0);
1da177e4 4060 if (sh) {
b0f9ec04 4061 if (unlikely(previous)) {
7ecaa1e6 4062 /* expansion might have moved on while waiting for a
df8e7f76
N		 4063 * stripe, so we must do the range check again.
4064 * Expansion could still move past after this
4065 * test, but as we are holding a reference to
4066 * 'sh', we know that if that happens,
4067 * STRIPE_EXPANDING will get set and the expansion
4068 * won't proceed until we finish with the stripe.
7ecaa1e6
N		 4069 */
4070 int must_retry = 0;
4071 spin_lock_irq(&conf->device_lock);
2c810cdd 4072 if (mddev->reshape_backwards
b0f9ec04
N		 4073 ? logical_sector >= conf->reshape_progress
4074 : logical_sector < conf->reshape_progress)
7ecaa1e6
N		 4075 /* mismatch, need to try again */
4076 must_retry = 1;
4077 spin_unlock_irq(&conf->device_lock);
4078 if (must_retry) {
4079 release_stripe(sh);
7a3ab908 4080 schedule();
7ecaa1e6
N		 4081 goto retry;
4082 }
4083 }
e62e58a5 4084
ffd96e35 4085 if (rw == WRITE &&
a5c308d4 4086 logical_sector >= mddev->suspend_lo &&
e464eafd
N		 4087 logical_sector < mddev->suspend_hi) {
4088 release_stripe(sh);
e62e58a5
N		 4089 /* As the suspend_* range is controlled by
4090 * userspace, we want an interruptible
4091 * wait.
4092 */
4093 flush_signals(current);
4094 prepare_to_wait(&conf->wait_for_overlap,
4095 &w, TASK_INTERRUPTIBLE);
4096 if (logical_sector >= mddev->suspend_lo &&
4097 logical_sector < mddev->suspend_hi)
4098 schedule();
e464eafd
N		 4099 goto retry;
4100 }
7ecaa1e6
N		 4101
4102 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
ffd96e35 4103 !add_stripe_bio(sh, bi, dd_idx, rw)) {
7ecaa1e6
N		 4104 /* Stripe is busy expanding or
4105 * add failed due to overlap. Flush everything
1da177e4
LT		 4106 * and wait a while
4107 */
482c0834 4108 md_wakeup_thread(mddev->thread);
1da177e4
LT		 4109 release_stripe(sh);
4110 schedule();
4111 goto retry;
4112 }
4113 finish_wait(&conf->wait_for_overlap, &w);
6ed3003c
N		 4114 set_bit(STRIPE_HANDLE, &sh->state);
4115 clear_bit(STRIPE_DELAYED, &sh->state);
e9c7469b 4116 if ((bi->bi_rw & REQ_SYNC) &&
729a1866
N		 4117 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4118 atomic_inc(&conf->preread_active_stripes);
b357f04a 4119 mddev_check_plugged(mddev);
1da177e4 4120 release_stripe(sh);
1da177e4
LT		 4121 } else {
4122 /* cannot get stripe for read-ahead, just give-up */
4123 clear_bit(BIO_UPTODATE, &bi->bi_flags);
4124 finish_wait(&conf->wait_for_overlap, &w);
4125 break;
4126 }
1da177e4 4127 }
7c13edc8 4128
1da177e4 4129 spin_lock_irq(&conf->device_lock);
960e739d 4130 remaining = raid5_dec_bi_phys_segments(bi);
f6344757
N		 4131 spin_unlock_irq(&conf->device_lock);
4132 if (remaining == 0) {
1da177e4 4133
16a53ecc 4134 if ( rw == WRITE )
1da177e4 4135 md_write_end(mddev);
6712ecf8 4136
0e13fe23 4137 bio_endio(bi, 0);
1da177e4 4138 }
1da177e4
LT		 4139}
4140
fd01b88c 4141static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks);
b522adcd 4142
fd01b88c 4143static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped)
1da177e4 4144{
52c03291
N		 4145 /* reshaping is quite different to recovery/resync so it is
4146 * handled quite separately ... here.
4147 *
4148 * On each call to sync_request, we gather one chunk worth of
4149 * destination stripes and flag them as expanding.
4150 * Then we find all the source stripes and request reads.
4151 * As the reads complete, handle_stripe will copy the data
4152 * into the destination stripe and release that stripe.
4153 */
d1688a6d 4154 struct r5conf *conf = mddev->private;
1da177e4 4155 struct stripe_head *sh;
ccfcc3c1 4156 sector_t first_sector, last_sector;
f416885e
N		 4157 int raid_disks = conf->previous_raid_disks;
4158 int data_disks = raid_disks - conf->max_degraded;
4159 int new_data_disks = conf->raid_disks - conf->max_degraded;
52c03291
N		 4160 int i;
4161 int dd_idx;
c8f517c4 4162 sector_t writepos, readpos, safepos;
ec32a2bd 4163 sector_t stripe_addr;
7a661381 4164 int reshape_sectors;
ab69ae12 4165 struct list_head stripes;
52c03291 4166
fef9c61f
N		 4167 if (sector_nr == 0) {
4168 /* If restarting in the middle, skip the initial sectors */
2c810cdd 4169 if (mddev->reshape_backwards &&
fef9c61f
N		 4170 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
4171 sector_nr = raid5_size(mddev, 0, 0)
4172 - conf->reshape_progress;
2c810cdd 4173 } else if (!mddev->reshape_backwards &&
fef9c61f
N		 4174 conf->reshape_progress > 0)
4175 sector_nr = conf->reshape_progress;
f416885e 4176 sector_div(sector_nr, new_data_disks);
fef9c61f 4177 if (sector_nr) {
8dee7211
N		 4178 mddev->curr_resync_completed = sector_nr;
4179 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
fef9c61f
N		 4180 *skipped = 1;
4181 return sector_nr;
4182 }
52c03291
N		 4183 }
4184
7a661381
N		 4185 /* We need to process a full chunk at a time.
4186 * If old and new chunk sizes differ, we need to process the
4187 * largest of these
4188 */
664e7c41
AN		 4189 if (mddev->new_chunk_sectors > mddev->chunk_sectors)
4190 reshape_sectors = mddev->new_chunk_sectors;
7a661381 4191 else
9d8f0363 4192 reshape_sectors = mddev->chunk_sectors;
7a661381 4193
b5254dd5
N		 4194 /* We update the metadata at least every 10 seconds, or when
4195 * the data about to be copied would over-write the source of
4196 * the data at the front of the range. i.e. one new_stripe
4197 * along from reshape_progress new_maps to after where
4198 * reshape_safe old_maps to
52c03291 4199 */
fef9c61f 4200 writepos = conf->reshape_progress;
f416885e 4201 sector_div(writepos, new_data_disks);
c8f517c4
N		 4202 readpos = conf->reshape_progress;
4203 sector_div(readpos, data_disks);
fef9c61f 4204 safepos = conf->reshape_safe;
f416885e 4205 sector_div(safepos, data_disks);
2c810cdd 4206 if (mddev->reshape_backwards) {
ed37d83e 4207 writepos -= min_t(sector_t, reshape_sectors, writepos);
c8f517c4 4208 readpos += reshape_sectors;
7a661381 4209 safepos += reshape_sectors;
fef9c61f 4210 } else {
7a661381 4211 writepos += reshape_sectors;
ed37d83e
N		 4212 readpos -= min_t(sector_t, reshape_sectors, readpos);
4213 safepos -= min_t(sector_t, reshape_sectors, safepos);
fef9c61f 4214 }
52c03291 4215
b5254dd5
N		 4216 /* Having calculated the 'writepos' possibly use it
4217 * to set 'stripe_addr' which is where we will write to.
4218 */
4219 if (mddev->reshape_backwards) {
4220 BUG_ON(conf->reshape_progress == 0);
4221 stripe_addr = writepos;
4222 BUG_ON((mddev->dev_sectors &
4223 ~((sector_t)reshape_sectors - 1))
4224 - reshape_sectors - stripe_addr
4225 != sector_nr);
4226 } else {
4227 BUG_ON(writepos != sector_nr + reshape_sectors);
4228 stripe_addr = sector_nr;
4229 }
4230
c8f517c4
N		 4231 /* 'writepos' is the most advanced device address we might write.
4232 * 'readpos' is the least advanced device address we might read.
4233 * 'safepos' is the least address recorded in the metadata as having
4234 * been reshaped.
b5254dd5
N		 4235 * If there is a min_offset_diff, these are adjusted either by
4236 * increasing the safepos/readpos if diff is negative, or
4237 * increasing writepos if diff is positive.
4238 * If 'readpos' is then behind 'writepos', there is no way that we can
c8f517c4
N		 4239 * ensure safety in the face of a crash - that must be done by userspace
4240 * making a backup of the data. So in that case there is no particular
4241 * rush to update metadata.
4242 * Otherwise if 'safepos' is behind 'writepos', then we really need to
4243 * update the metadata to advance 'safepos' to match 'readpos' so that
4244 * we can be safe in the event of a crash.
4245 * So we insist on updating metadata if safepos is behind writepos and
4246 * readpos is beyond writepos.
4247 * In any case, update the metadata every 10 seconds.
4248 * Maybe that number should be configurable, but I'm not sure it is
4249 * worth it.... maybe it could be a multiple of safemode_delay???
4250 */
b5254dd5
N		 4251 if (conf->min_offset_diff < 0) {
4252 safepos += -conf->min_offset_diff;
4253 readpos += -conf->min_offset_diff;
4254 } else
4255 writepos += conf->min_offset_diff;
4256
2c810cdd 4257 if ((mddev->reshape_backwards
c8f517c4
N		 4258 ? (safepos > writepos && readpos < writepos)
4259 : (safepos < writepos && readpos > writepos)) ||
4260 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
52c03291
N		 4261 /* Cannot proceed until we've updated the superblock... */
4262 wait_event(conf->wait_for_overlap,
4263 atomic_read(&conf->reshape_stripes)==0);
fef9c61f 4264 mddev->reshape_position = conf->reshape_progress;
75d3da43 4265 mddev->curr_resync_completed = sector_nr;
c8f517c4 4266 conf->reshape_checkpoint = jiffies;
850b2b42 4267 set_bit(MD_CHANGE_DEVS, &mddev->flags);
52c03291 4268 md_wakeup_thread(mddev->thread);
850b2b42 4269 wait_event(mddev->sb_wait, mddev->flags == 0 ||
52c03291
N		 4270 kthread_should_stop());
4271 spin_lock_irq(&conf->device_lock);
fef9c61f 4272 conf->reshape_safe = mddev->reshape_position;
52c03291
N		 4273 spin_unlock_irq(&conf->device_lock);
4274 wake_up(&conf->wait_for_overlap);
acb180b0 4275 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
52c03291
N		 4276 }
4277
ab69ae12 4278 INIT_LIST_HEAD(&stripes);
7a661381 4279 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
52c03291 4280 int j;
a9f326eb 4281 int skipped_disk = 0;
a8c906ca 4282 sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
52c03291
N		 4283 set_bit(STRIPE_EXPANDING, &sh->state);
4284 atomic_inc(&conf->reshape_stripes);
4285 /* If any of this stripe is beyond the end of the old
4286 * array, then we need to zero those blocks
4287 */
4288 for (j=sh->disks; j--;) {
4289 sector_t s;
4290 if (j == sh->pd_idx)
4291 continue;
f416885e 4292 if (conf->level == 6 &&
d0dabf7e 4293 j == sh->qd_idx)
f416885e 4294 continue;
784052ec 4295 s = compute_blocknr(sh, j, 0);
b522adcd 4296 if (s < raid5_size(mddev, 0, 0)) {
a9f326eb 4297 skipped_disk = 1;
52c03291
N		 4298 continue;
4299 }
4300 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
4301 set_bit(R5_Expanded, &sh->dev[j].flags);
4302 set_bit(R5_UPTODATE, &sh->dev[j].flags);
4303 }
a9f326eb 4304 if (!skipped_disk) {
52c03291
N		 4305 set_bit(STRIPE_EXPAND_READY, &sh->state);
4306 set_bit(STRIPE_HANDLE, &sh->state);
4307 }
ab69ae12 4308 list_add(&sh->lru, &stripes);
52c03291
N		 4309 }
4310 spin_lock_irq(&conf->device_lock);
2c810cdd 4311 if (mddev->reshape_backwards)
7a661381 4312 conf->reshape_progress -= reshape_sectors * new_data_disks;
fef9c61f 4313 else
7a661381 4314 conf->reshape_progress += reshape_sectors * new_data_disks;
52c03291
N		 4315 spin_unlock_irq(&conf->device_lock);
4316 /* Ok, those stripe are ready. We can start scheduling
4317 * reads on the source stripes.
4318 * The source stripes are determined by mapping the first and last
4319 * block on the destination stripes.
4320 */
52c03291 4321 first_sector =
ec32a2bd 4322 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
911d4ee8 4323 1, &dd_idx, NULL);
52c03291 4324 last_sector =
0e6e0271 4325 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
09c9e5fa 4326 * new_data_disks - 1),
911d4ee8 4327 1, &dd_idx, NULL);
58c0fed4
AN		 4328 if (last_sector >= mddev->dev_sectors)
4329 last_sector = mddev->dev_sectors - 1;
52c03291 4330 while (first_sector <= last_sector) {
a8c906ca 4331 sh = get_active_stripe(conf, first_sector, 1, 0, 1);
52c03291
N		 4332 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
4333 set_bit(STRIPE_HANDLE, &sh->state);
4334 release_stripe(sh);
4335 first_sector += STRIPE_SECTORS;
4336 }
ab69ae12
N		 4337 /* Now that the sources are clearly marked, we can release
4338 * the destination stripes
4339 */
4340 while (!list_empty(&stripes)) {
4341 sh = list_entry(stripes.next, struct stripe_head, lru);
4342 list_del_init(&sh->lru);
4343 release_stripe(sh);
4344 }
c6207277
N		 4345 /* If this takes us to the resync_max point where we have to pause,
4346 * then we need to write out the superblock.
4347 */
7a661381 4348 sector_nr += reshape_sectors;
c03f6a19
N		 4349 if ((sector_nr - mddev->curr_resync_completed) * 2
4350 >= mddev->resync_max - mddev->curr_resync_completed) {
c6207277
N		 4351 /* Cannot proceed until we've updated the superblock... */
4352 wait_event(conf->wait_for_overlap,
4353 atomic_read(&conf->reshape_stripes) == 0);
fef9c61f 4354 mddev->reshape_position = conf->reshape_progress;
75d3da43 4355 mddev->curr_resync_completed = sector_nr;
c8f517c4 4356 conf->reshape_checkpoint = jiffies;
c6207277
N		 4357 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4358 md_wakeup_thread(mddev->thread);
4359 wait_event(mddev->sb_wait,
4360 !test_bit(MD_CHANGE_DEVS, &mddev->flags)
4361 || kthread_should_stop());
4362 spin_lock_irq(&conf->device_lock);
fef9c61f 4363 conf->reshape_safe = mddev->reshape_position;
c6207277
N		 4364 spin_unlock_irq(&conf->device_lock);
4365 wake_up(&conf->wait_for_overlap);
acb180b0 4366 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
c6207277 4367 }
7a661381 4368 return reshape_sectors;
52c03291
N		 4369}
4370
4371/* FIXME go_faster isn't used */
fd01b88c 4372static inline sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster)
52c03291 4373{
d1688a6d 4374 struct r5conf *conf = mddev->private;
52c03291 4375 struct stripe_head *sh;
58c0fed4 4376 sector_t max_sector = mddev->dev_sectors;
57dab0bd 4377 sector_t sync_blocks;
16a53ecc
N		 4378 int still_degraded = 0;
4379 int i;
1da177e4 4380
72626685 4381 if (sector_nr >= max_sector) {
1da177e4 4382 /* just being told to finish up .. nothing much to do */
cea9c228 4383
29269553
N		 4384 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
4385 end_reshape(conf);
4386 return 0;
4387 }
72626685
N		 4388
4389 if (mddev->curr_resync < max_sector) /* aborted */
4390 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
4391 &sync_blocks, 1);
16a53ecc 4392 else /* completed sync */
72626685
N		 4393 conf->fullsync = 0;
4394 bitmap_close_sync(mddev->bitmap);
4395
1da177e4
LT		 4396 return 0;
4397 }
ccfcc3c1 4398
64bd660b
N		 4399 /* Allow raid5_quiesce to complete */
4400 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
4401
52c03291
N		 4402 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4403 return reshape_request(mddev, sector_nr, skipped);
f6705578 4404
c6207277
N		 4405 /* No need to check resync_max as we never do more than one
4406 * stripe, and as resync_max will always be on a chunk boundary,
4407 * if the check in md_do_sync didn't fire, there is no chance
4408 * of overstepping resync_max here
4409 */
4410
16a53ecc 4411 /* if there is too many failed drives and we are trying
1da177e4
LT		 4412 * to resync, then assert that we are finished, because there is
4413 * nothing we can do.
4414 */
3285edf1 4415 if (mddev->degraded >= conf->max_degraded &&
16a53ecc 4416 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
58c0fed4 4417 sector_t rv = mddev->dev_sectors - sector_nr;
57afd89f 4418 *skipped = 1;
1da177e4
LT		 4419 return rv;
4420 }
72626685 4421 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
3855ad9f 4422 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
72626685
N		 4423 !conf->fullsync && sync_blocks >= STRIPE_SECTORS) {
4424 /* we can skip this block, and probably more */
4425 sync_blocks /= STRIPE_SECTORS;
4426 *skipped = 1;
4427 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
4428 }
1da177e4 4429
b47490c9
N		 4430 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
4431
a8c906ca 4432 sh = get_active_stripe(conf, sector_nr, 0, 1, 0);
1da177e4 4433 if (sh == NULL) {
a8c906ca 4434 sh = get_active_stripe(conf, sector_nr, 0, 0, 0);
1da177e4 4435 /* make sure we don't swamp the stripe cache if someone else
16a53ecc 4436 * is trying to get access
1da177e4 4437 */
66c006a5 4438 schedule_timeout_uninterruptible(1);
1da177e4 4439 }
16a53ecc
N		 4440 /* Need to check if array will still be degraded after recovery/resync
4441 * We don't need to check the 'failed' flag as when that gets set,
4442 * recovery aborts.
4443 */
f001a70c 4444 for (i = 0; i < conf->raid_disks; i++)
16a53ecc
N		 4445 if (conf->disks[i].rdev == NULL)
4446 still_degraded = 1;
4447
4448 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
4449
83206d66 4450 set_bit(STRIPE_SYNC_REQUESTED, &sh->state);
1da177e4 4451
1442577b 4452 handle_stripe(sh);
1da177e4
LT		 4453 release_stripe(sh);
4454
4455 return STRIPE_SECTORS;
4456}
4457
d1688a6d 4458static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio)
46031f9a
RBJ		 4459{
4460 /* We may not be able to submit a whole bio at once as there
4461 * may not be enough stripe_heads available.
4462 * We cannot pre-allocate enough stripe_heads as we may need
4463 * more than exist in the cache (if we allow ever large chunks).
4464 * So we do one stripe head at a time and record in
4465 * ->bi_hw_segments how many have been done.
4466 *
4467 * We *know* that this entire raid_bio is in one chunk, so
4468 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
4469 */
4470 struct stripe_head *sh;
911d4ee8 4471 int dd_idx;
46031f9a
RBJ		 4472 sector_t sector, logical_sector, last_sector;
4473 int scnt = 0;
4474 int remaining;
4475 int handled = 0;
4476
4477 logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
112bf897 4478 sector = raid5_compute_sector(conf, logical_sector,
911d4ee8 4479 0, &dd_idx, NULL);
46031f9a
RBJ		 4480 last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9);
4481
4482 for (; logical_sector < last_sector;
387bb173
NB		 4483 logical_sector += STRIPE_SECTORS,
4484 sector += STRIPE_SECTORS,
4485 scnt++) {
46031f9a 4486
960e739d 4487 if (scnt < raid5_bi_hw_segments(raid_bio))
46031f9a
RBJ		 4488 /* already done this stripe */
4489 continue;
4490
a8c906ca 4491 sh = get_active_stripe(conf, sector, 0, 1, 0);
46031f9a
RBJ		 4492
4493 if (!sh) {
4494 /* failed to get a stripe - must wait */
960e739d 4495 raid5_set_bi_hw_segments(raid_bio, scnt);
46031f9a
RBJ		 4496 conf->retry_read_aligned = raid_bio;
4497 return handled;
4498 }
4499
387bb173
NB		 4500 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) {
4501 release_stripe(sh);
960e739d 4502 raid5_set_bi_hw_segments(raid_bio, scnt);
387bb173
NB		 4503 conf->retry_read_aligned = raid_bio;
4504 return handled;
4505 }
4506
36d1c647 4507 handle_stripe(sh);
46031f9a
RBJ		 4508 release_stripe(sh);
4509 handled++;
4510 }
4511 spin_lock_irq(&conf->device_lock);
960e739d 4512 remaining = raid5_dec_bi_phys_segments(raid_bio);
46031f9a 4513 spin_unlock_irq(&conf->device_lock);
0e13fe23
NB		 4514 if (remaining == 0)
4515 bio_endio(raid_bio, 0);
46031f9a
RBJ		 4516 if (atomic_dec_and_test(&conf->active_aligned_reads))
4517 wake_up(&conf->wait_for_stripe);
4518 return handled;
4519}
4520
46031f9a 4521
1da177e4
LT		 4522/*
4523 * This is our raid5 kernel thread.
4524 *
4525 * We scan the hash table for stripes which can be handled now.
4526 * During the scan, completed stripes are saved for us by the interrupt
4527 * handler, so that they will not have to wait for our next wakeup.
4528 */
fd01b88c 4529static void raid5d(struct mddev *mddev)
1da177e4
LT		 4530{
4531 struct stripe_head *sh;
d1688a6d 4532 struct r5conf *conf = mddev->private;
1da177e4 4533 int handled;
e1dfa0a2 4534 struct blk_plug plug;
1da177e4 4535
45b4233c 4536 pr_debug("+++ raid5d active\n");
1da177e4
LT		 4537
4538 md_check_recovery(mddev);
1da177e4 4539
e1dfa0a2 4540 blk_start_plug(&plug);
1da177e4
LT		 4541 handled = 0;
4542 spin_lock_irq(&conf->device_lock);
4543 while (1) {
46031f9a 4544 struct bio *bio;
1da177e4 4545
7c13edc8
N		 4546 if (atomic_read(&mddev->plug_cnt) == 0 &&
4547 !list_empty(&conf->bitmap_list)) {
4548 /* Now is a good time to flush some bitmap updates */
4549 conf->seq_flush++;
700e432d 4550 spin_unlock_irq(&conf->device_lock);
72626685 4551 bitmap_unplug(mddev->bitmap);
700e432d 4552 spin_lock_irq(&conf->device_lock);
7c13edc8 4553 conf->seq_write = conf->seq_flush;
72626685
N		 4554 activate_bit_delay(conf);
4555 }
7c13edc8
N		 4556 if (atomic_read(&mddev->plug_cnt) == 0)
4557 raid5_activate_delayed(conf);
72626685 4558
46031f9a
RBJ		 4559 while ((bio = remove_bio_from_retry(conf))) {
4560 int ok;
4561 spin_unlock_irq(&conf->device_lock);
4562 ok = retry_aligned_read(conf, bio);
4563 spin_lock_irq(&conf->device_lock);
4564 if (!ok)
4565 break;
4566 handled++;
4567 }
4568
8b3e6cdc
DW		 4569 sh = __get_priority_stripe(conf);
4570
c9f21aaf 4571 if (!sh)
1da177e4 4572 break;
1da177e4
LT		 4573 spin_unlock_irq(&conf->device_lock);
4574
4575 handled++;
417b8d4a
DW		 4576 handle_stripe(sh);
4577 release_stripe(sh);
4578 cond_resched();
1da177e4 4579
de393cde
N		 4580 if (mddev->flags & ~(1<<MD_CHANGE_PENDING))
4581 md_check_recovery(mddev);
4582
1da177e4
LT		 4583 spin_lock_irq(&conf->device_lock);
4584 }
45b4233c 4585 pr_debug("%d stripes handled\n", handled);
1da177e4
LT		 4586
4587 spin_unlock_irq(&conf->device_lock);
4588
c9f21aaf 4589 async_tx_issue_pending_all();
e1dfa0a2 4590 blk_finish_plug(&plug);
1da177e4 4591
45b4233c 4592 pr_debug("--- raid5d inactive\n");
1da177e4
LT		 4593}
4594
3f294f4f 4595static ssize_t
fd01b88c 4596raid5_show_stripe_cache_size(struct mddev *mddev, char *page)
3f294f4f 4597{
d1688a6d 4598 struct r5conf *conf = mddev->private;
96de1e66
N		 4599 if (conf)
4600 return sprintf(page, "%d\n", conf->max_nr_stripes);
4601 else
4602 return 0;
3f294f4f
N		 4603}
4604
c41d4ac4 4605int
fd01b88c 4606raid5_set_cache_size(struct mddev *mddev, int size)
3f294f4f 4607{
d1688a6d 4608 struct r5conf *conf = mddev->private;
b5470dc5
DW		 4609 int err;
4610
c41d4ac4 4611 if (size <= 16 || size > 32768)
3f294f4f 4612 return -EINVAL;
c41d4ac4 4613 while (size < conf->max_nr_stripes) {
3f294f4f
N		 4614 if (drop_one_stripe(conf))
4615 conf->max_nr_stripes--;
4616 else
4617 break;
4618 }
b5470dc5
DW		 4619 err = md_allow_write(mddev);
4620 if (err)
4621 return err;
c41d4ac4 4622 while (size > conf->max_nr_stripes) {
3f294f4f
N		 4623 if (grow_one_stripe(conf))
4624 conf->max_nr_stripes++;
4625 else break;
4626 }
c41d4ac4
N		 4627 return 0;
4628}
4629EXPORT_SYMBOL(raid5_set_cache_size);
4630
4631static ssize_t
fd01b88c 4632raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len)
c41d4ac4 4633{
d1688a6d 4634 struct r5conf *conf = mddev->private;
c41d4ac4
N		 4635 unsigned long new;
4636 int err;
4637
4638 if (len >= PAGE_SIZE)
4639 return -EINVAL;
4640 if (!conf)
4641 return -ENODEV;
4642
4643 if (strict_strtoul(page, 10, &new))
4644 return -EINVAL;
4645 err = raid5_set_cache_size(mddev, new);
4646 if (err)
4647 return err;
3f294f4f
N		 4648 return len;
4649}
007583c9 4650
96de1e66
N		 4651static struct md_sysfs_entry
4652raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
4653 raid5_show_stripe_cache_size,
4654 raid5_store_stripe_cache_size);
3f294f4f 4655
8b3e6cdc 4656static ssize_t
fd01b88c 4657raid5_show_preread_threshold(struct mddev *mddev, char *page)
8b3e6cdc 4658{
d1688a6d 4659 struct r5conf *conf = mddev->private;
8b3e6cdc
DW		 4660 if (conf)
4661 return sprintf(page, "%d\n", conf->bypass_threshold);
4662 else
4663 return 0;
4664}
4665
4666static ssize_t
fd01b88c 4667raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len)
8b3e6cdc 4668{
d1688a6d 4669 struct r5conf *conf = mddev->private;
4ef197d8 4670 unsigned long new;
8b3e6cdc
DW		 4671 if (len >= PAGE_SIZE)
4672 return -EINVAL;
4673 if (!conf)
4674 return -ENODEV;
4675
4ef197d8 4676 if (strict_strtoul(page, 10, &new))
8b3e6cdc 4677 return -EINVAL;
4ef197d8 4678 if (new > conf->max_nr_stripes)
8b3e6cdc
DW		 4679 return -EINVAL;
4680 conf->bypass_threshold = new;
4681 return len;
4682}
4683
4684static struct md_sysfs_entry
4685raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
4686 S_IRUGO | S_IWUSR,
4687 raid5_show_preread_threshold,
4688 raid5_store_preread_threshold);
4689
3f294f4f 4690static ssize_t
fd01b88c 4691stripe_cache_active_show(struct mddev *mddev, char *page)
3f294f4f 4692{
d1688a6d 4693 struct r5conf *conf = mddev->private;
96de1e66
N		 4694 if (conf)
4695 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
4696 else
4697 return 0;
3f294f4f
N		 4698}
4699
96de1e66
N		 4700static struct md_sysfs_entry
4701raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
3f294f4f 4702
007583c9 4703static struct attribute *raid5_attrs[] = {
3f294f4f
N		 4704 &raid5_stripecache_size.attr,
4705 &raid5_stripecache_active.attr,
8b3e6cdc 4706 &raid5_preread_bypass_threshold.attr,
3f294f4f
N		 4707 NULL,
4708};
007583c9
N		 4709static struct attribute_group raid5_attrs_group = {
4710 .name = NULL,
4711 .attrs = raid5_attrs,
3f294f4f
N		 4712};
4713
80c3a6ce 4714static sector_t
fd01b88c 4715raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks)
80c3a6ce 4716{
d1688a6d 4717 struct r5conf *conf = mddev->private;
80c3a6ce
DW		 4718
4719 if (!sectors)
4720 sectors = mddev->dev_sectors;
5e5e3e78 4721 if (!raid_disks)
7ec05478 4722 /* size is defined by the smallest of previous and new size */
5e5e3e78 4723 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
80c3a6ce 4724
9d8f0363 4725 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
664e7c41 4726 sectors &= ~((sector_t)mddev->new_chunk_sectors - 1);
80c3a6ce
DW		 4727 return sectors * (raid_disks - conf->max_degraded);
4728}
4729
d1688a6d 4730static void raid5_free_percpu(struct r5conf *conf)
36d1c647
DW		 4731{
4732 struct raid5_percpu *percpu;
4733 unsigned long cpu;
4734
4735 if (!conf->percpu)
4736 return;
4737
4738 get_online_cpus();
4739 for_each_possible_cpu(cpu) {
4740 percpu = per_cpu_ptr(conf->percpu, cpu);
4741 safe_put_page(percpu->spare_page);
d6f38f31 4742 kfree(percpu->scribble);
36d1c647
DW		 4743 }
4744#ifdef CONFIG_HOTPLUG_CPU
4745 unregister_cpu_notifier(&conf->cpu_notify);
4746#endif
4747 put_online_cpus();
4748
4749 free_percpu(conf->percpu);
4750}
4751
d1688a6d 4752static void free_conf(struct r5conf *conf)
95fc17aa
DW		 4753{
4754 shrink_stripes(conf);
36d1c647 4755 raid5_free_percpu(conf);
95fc17aa
DW		 4756 kfree(conf->disks);
4757 kfree(conf->stripe_hashtbl);
4758 kfree(conf);
4759}
4760
36d1c647
DW		 4761#ifdef CONFIG_HOTPLUG_CPU
4762static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action,
4763 void *hcpu)
4764{
d1688a6d 4765 struct r5conf *conf = container_of(nfb, struct r5conf, cpu_notify);
36d1c647
DW		 4766 long cpu = (long)hcpu;
4767 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
4768
4769 switch (action) {
4770 case CPU_UP_PREPARE:
4771 case CPU_UP_PREPARE_FROZEN:
d6f38f31 4772 if (conf->level == 6 && !percpu->spare_page)
36d1c647 4773 percpu->spare_page = alloc_page(GFP_KERNEL);
d6f38f31
DW		 4774 if (!percpu->scribble)
4775 percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
4776
4777 if (!percpu->scribble ||
4778 (conf->level == 6 && !percpu->spare_page)) {
4779 safe_put_page(percpu->spare_page);
4780 kfree(percpu->scribble);
36d1c647
DW		 4781 pr_err("%s: failed memory allocation for cpu%ld\n",
4782 __func__, cpu);
55af6bb5 4783 return notifier_from_errno(-ENOMEM);
36d1c647
DW		 4784 }
4785 break;
4786 case CPU_DEAD:
4787 case CPU_DEAD_FROZEN:
4788 safe_put_page(percpu->spare_page);
d6f38f31 4789 kfree(percpu->scribble);
36d1c647 4790 percpu->spare_page = NULL;
d6f38f31 4791 percpu->scribble = NULL;
36d1c647
DW		 4792 break;
4793 default:
4794 break;
4795 }
4796 return NOTIFY_OK;
4797}
4798#endif
4799
d1688a6d 4800static int raid5_alloc_percpu(struct r5conf *conf)
36d1c647
DW		 4801{
4802 unsigned long cpu;
4803 struct page *spare_page;
a29d8b8e 4804 struct raid5_percpu __percpu *allcpus;
d6f38f31 4805 void *scribble;
36d1c647
DW		 4806 int err;
4807
36d1c647
DW		 4808 allcpus = alloc_percpu(struct raid5_percpu);
4809 if (!allcpus)
4810 return -ENOMEM;
4811 conf->percpu = allcpus;
4812
4813 get_online_cpus();
4814 err = 0;
4815 for_each_present_cpu(cpu) {
d6f38f31
DW		 4816 if (conf->level == 6) {
4817 spare_page = alloc_page(GFP_KERNEL);
4818 if (!spare_page) {
4819 err = -ENOMEM;
4820 break;
4821 }
4822 per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page;
4823 }
5e5e3e78 4824 scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
d6f38f31 4825 if (!scribble) {
36d1c647
DW		 4826 err = -ENOMEM;
4827 break;
4828 }
d6f38f31 4829 per_cpu_ptr(conf->percpu, cpu)->scribble = scribble;
36d1c647
DW		 4830 }
4831#ifdef CONFIG_HOTPLUG_CPU
4832 conf->cpu_notify.notifier_call = raid456_cpu_notify;
4833 conf->cpu_notify.priority = 0;
4834 if (err == 0)
4835 err = register_cpu_notifier(&conf->cpu_notify);
4836#endif
4837 put_online_cpus();
4838
4839 return err;
4840}
4841
d1688a6d 4842static struct r5conf *setup_conf(struct mddev *mddev)
1da177e4 4843{
d1688a6d 4844 struct r5conf *conf;
5e5e3e78 4845 int raid_disk, memory, max_disks;
3cb03002 4846 struct md_rdev *rdev;
1da177e4 4847 struct disk_info *disk;
0232605d 4848 char pers_name[6];
1da177e4 4849
91adb564
N		 4850 if (mddev->new_level != 5
4851 && mddev->new_level != 4
4852 && mddev->new_level != 6) {
0c55e022 4853 printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n",
91adb564
N		 4854 mdname(mddev), mddev->new_level);
4855 return ERR_PTR(-EIO);
1da177e4 4856 }
91adb564
N		 4857 if ((mddev->new_level == 5
4858 && !algorithm_valid_raid5(mddev->new_layout)) ||
4859 (mddev->new_level == 6
4860 && !algorithm_valid_raid6(mddev->new_layout))) {
0c55e022 4861 printk(KERN_ERR "md/raid:%s: layout %d not supported\n",
91adb564
N		 4862 mdname(mddev), mddev->new_layout);
4863 return ERR_PTR(-EIO);
99c0fb5f 4864 }
91adb564 4865 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
0c55e022 4866 printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n",
91adb564
N		 4867 mdname(mddev), mddev->raid_disks);
4868 return ERR_PTR(-EINVAL);
4bbf3771
N		 4869 }
4870
664e7c41
AN		 4871 if (!mddev->new_chunk_sectors ||
4872 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
4873 !is_power_of_2(mddev->new_chunk_sectors)) {
0c55e022
N		 4874 printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n",
4875 mdname(mddev), mddev->new_chunk_sectors << 9);
91adb564 4876 return ERR_PTR(-EINVAL);
f6705578
N		 4877 }
4878
d1688a6d 4879 conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL);
91adb564 4880 if (conf == NULL)
1da177e4 4881 goto abort;
f5efd45a
DW		 4882 spin_lock_init(&conf->device_lock);
4883 init_waitqueue_head(&conf->wait_for_stripe);
4884 init_waitqueue_head(&conf->wait_for_overlap);
4885 INIT_LIST_HEAD(&conf->handle_list);
4886 INIT_LIST_HEAD(&conf->hold_list);
4887 INIT_LIST_HEAD(&conf->delayed_list);
4888 INIT_LIST_HEAD(&conf->bitmap_list);
4889 INIT_LIST_HEAD(&conf->inactive_list);
4890 atomic_set(&conf->active_stripes, 0);
4891 atomic_set(&conf->preread_active_stripes, 0);
4892 atomic_set(&conf->active_aligned_reads, 0);
4893 conf->bypass_threshold = BYPASS_THRESHOLD;
d890fa2b 4894 conf->recovery_disabled = mddev->recovery_disabled - 1;
91adb564
N		 4895
4896 conf->raid_disks = mddev->raid_disks;
4897 if (mddev->reshape_position == MaxSector)
4898 conf->previous_raid_disks = mddev->raid_disks;
4899 else
f6705578 4900 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
5e5e3e78
N		 4901 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
4902 conf->scribble_len = scribble_len(max_disks);
f6705578 4903
5e5e3e78 4904 conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
b55e6bfc
N		 4905 GFP_KERNEL);
4906 if (!conf->disks)
4907 goto abort;
9ffae0cf 4908
1da177e4
LT		 4909 conf->mddev = mddev;
4910
fccddba0 4911 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
1da177e4 4912 goto abort;
1da177e4 4913
36d1c647
DW		 4914 conf->level = mddev->new_level;
4915 if (raid5_alloc_percpu(conf) != 0)
4916 goto abort;
4917
0c55e022 4918 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
1da177e4 4919
dafb20fa 4920 rdev_for_each(rdev, mddev) {
1da177e4 4921 raid_disk = rdev->raid_disk;
5e5e3e78 4922 if (raid_disk >= max_disks
1da177e4
LT		 4923 || raid_disk < 0)
4924 continue;
4925 disk = conf->disks + raid_disk;
4926
17045f52
N		 4927 if (test_bit(Replacement, &rdev->flags)) {
4928 if (disk->replacement)
4929 goto abort;
4930 disk->replacement = rdev;
4931 } else {
4932 if (disk->rdev)
4933 goto abort;
4934 disk->rdev = rdev;
4935 }
1da177e4 4936
b2d444d7 4937 if (test_bit(In_sync, &rdev->flags)) {
1da177e4 4938 char b[BDEVNAME_SIZE];
0c55e022
N		 4939 printk(KERN_INFO "md/raid:%s: device %s operational as raid"
4940 " disk %d\n",
4941 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
d6b212f4 4942 } else if (rdev->saved_raid_disk != raid_disk)
8c2e870a
NB		 4943 /* Cannot rely on bitmap to complete recovery */
4944 conf->fullsync = 1;
1da177e4
LT		 4945 }
4946
09c9e5fa 4947 conf->chunk_sectors = mddev->new_chunk_sectors;
91adb564 4948 conf->level = mddev->new_level;
16a53ecc
N		 4949 if (conf->level == 6)
4950 conf->max_degraded = 2;
4951 else
4952 conf->max_degraded = 1;
91adb564 4953 conf->algorithm = mddev->new_layout;
1da177e4 4954 conf->max_nr_stripes = NR_STRIPES;
fef9c61f 4955 conf->reshape_progress = mddev->reshape_position;
e183eaed 4956 if (conf->reshape_progress != MaxSector) {
09c9e5fa 4957 conf->prev_chunk_sectors = mddev->chunk_sectors;
e183eaed
N		 4958 conf->prev_algo = mddev->layout;
4959 }
1da177e4 4960
91adb564 4961 memory = conf->max_nr_stripes * (sizeof(struct stripe_head) +
5e5e3e78 4962 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
91adb564
N		 4963 if (grow_stripes(conf, conf->max_nr_stripes)) {
4964 printk(KERN_ERR
0c55e022
N		 4965 "md/raid:%s: couldn't allocate %dkB for buffers\n",
4966 mdname(mddev), memory);
91adb564
N		 4967 goto abort;
4968 } else
0c55e022
N		 4969 printk(KERN_INFO "md/raid:%s: allocated %dkB\n",
4970 mdname(mddev), memory);
1da177e4 4971
0232605d
N		 4972 sprintf(pers_name, "raid%d", mddev->new_level);
4973 conf->thread = md_register_thread(raid5d, mddev, pers_name);
91adb564
N		 4974 if (!conf->thread) {
4975 printk(KERN_ERR
0c55e022 4976 "md/raid:%s: couldn't allocate thread.\n",
91adb564 4977 mdname(mddev));
16a53ecc
N		 4978 goto abort;
4979 }
91adb564
N		 4980
4981 return conf;
4982
4983 abort:
4984 if (conf) {
95fc17aa 4985 free_conf(conf);
91adb564
N		 4986 return ERR_PTR(-EIO);
4987 } else
4988 return ERR_PTR(-ENOMEM);
4989}
4990
c148ffdc
N		 4991
4992static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
4993{
4994 switch (algo) {
4995 case ALGORITHM_PARITY_0:
4996 if (raid_disk < max_degraded)
4997 return 1;
4998 break;
4999 case ALGORITHM_PARITY_N:
5000 if (raid_disk >= raid_disks - max_degraded)
5001 return 1;
5002 break;
5003 case ALGORITHM_PARITY_0_6:
5004 if (raid_disk == 0 ||
5005 raid_disk == raid_disks - 1)
5006 return 1;
5007 break;
5008 case ALGORITHM_LEFT_ASYMMETRIC_6:
5009 case ALGORITHM_RIGHT_ASYMMETRIC_6:
5010 case ALGORITHM_LEFT_SYMMETRIC_6:
5011 case ALGORITHM_RIGHT_SYMMETRIC_6:
5012 if (raid_disk == raid_disks - 1)
5013 return 1;
5014 }
5015 return 0;
5016}
5017
fd01b88c 5018static int run(struct mddev *mddev)
91adb564 5019{
d1688a6d 5020 struct r5conf *conf;
9f7c2220 5021 int working_disks = 0;
c148ffdc 5022 int dirty_parity_disks = 0;
3cb03002 5023 struct md_rdev *rdev;
c148ffdc 5024 sector_t reshape_offset = 0;
17045f52 5025 int i;
b5254dd5
N		 5026 long long min_offset_diff = 0;
5027 int first = 1;
91adb564 5028
8c6ac868 5029 if (mddev->recovery_cp != MaxSector)
0c55e022 5030 printk(KERN_NOTICE "md/raid:%s: not clean"
8c6ac868
AN		 5031 " -- starting background reconstruction\n",
5032 mdname(mddev));
b5254dd5
N		 5033
5034 rdev_for_each(rdev, mddev) {
5035 long long diff;
5036 if (rdev->raid_disk < 0)
5037 continue;
5038 diff = (rdev->new_data_offset - rdev->data_offset);
5039 if (first) {
5040 min_offset_diff = diff;
5041 first = 0;
5042 } else if (mddev->reshape_backwards &&
5043 diff < min_offset_diff)
5044 min_offset_diff = diff;
5045 else if (!mddev->reshape_backwards &&
5046 diff > min_offset_diff)
5047 min_offset_diff = diff;
5048 }
5049
91adb564
N		 5050 if (mddev->reshape_position != MaxSector) {
5051 /* Check that we can continue the reshape.
b5254dd5
N		 5052 * Difficulties arise if the stripe we would write to
5053 * next is at or after the stripe we would read from next.
5054 * For a reshape that changes the number of devices, this
5055 * is only possible for a very short time, and mdadm makes
5056 * sure that time appears to have past before assembling
5057 * the array. So we fail if that time hasn't passed.
5058 * For a reshape that keeps the number of devices the same
5059 * mdadm must be monitoring the reshape can keeping the
5060 * critical areas read-only and backed up. It will start
5061 * the array in read-only mode, so we check for that.
91adb564
N		 5062 */
5063 sector_t here_new, here_old;
5064 int old_disks;
18b00334 5065 int max_degraded = (mddev->level == 6 ? 2 : 1);
91adb564 5066
88ce4930 5067 if (mddev->new_level != mddev->level) {
0c55e022 5068 printk(KERN_ERR "md/raid:%s: unsupported reshape "
91adb564
N		 5069 "required - aborting.\n",
5070 mdname(mddev));
5071 return -EINVAL;
5072 }
91adb564
N		 5073 old_disks = mddev->raid_disks - mddev->delta_disks;
5074 /* reshape_position must be on a new-stripe boundary, and one
5075 * further up in new geometry must map after here in old
5076 * geometry.
5077 */
5078 here_new = mddev->reshape_position;
664e7c41 5079 if (sector_div(here_new, mddev->new_chunk_sectors *
91adb564 5080 (mddev->raid_disks - max_degraded))) {
0c55e022
N		 5081 printk(KERN_ERR "md/raid:%s: reshape_position not "
5082 "on a stripe boundary\n", mdname(mddev));
91adb564
N		 5083 return -EINVAL;
5084 }
c148ffdc 5085 reshape_offset = here_new * mddev->new_chunk_sectors;
91adb564
N		 5086 /* here_new is the stripe we will write to */
5087 here_old = mddev->reshape_position;
9d8f0363 5088 sector_div(here_old, mddev->chunk_sectors *
91adb564
N		 5089 (old_disks-max_degraded));
5090 /* here_old is the first stripe that we might need to read
5091 * from */
67ac6011 5092 if (mddev->delta_disks == 0) {
b5254dd5
N		 5093 if ((here_new * mddev->new_chunk_sectors !=
5094 here_old * mddev->chunk_sectors)) {
5095 printk(KERN_ERR "md/raid:%s: reshape position is"
5096 " confused - aborting\n", mdname(mddev));
5097 return -EINVAL;
5098 }
67ac6011 5099 /* We cannot be sure it is safe to start an in-place
b5254dd5 5100 * reshape. It is only safe if user-space is monitoring
67ac6011
N		 5101 * and taking constant backups.
5102 * mdadm always starts a situation like this in
5103 * readonly mode so it can take control before
5104 * allowing any writes. So just check for that.
5105 */
b5254dd5
N		 5106 if (abs(min_offset_diff) >= mddev->chunk_sectors &&
5107 abs(min_offset_diff) >= mddev->new_chunk_sectors)
5108 /* not really in-place - so OK */;
5109 else if (mddev->ro == 0) {
5110 printk(KERN_ERR "md/raid:%s: in-place reshape "
5111 "must be started in read-only mode "
5112 "- aborting\n",
0c55e022 5113 mdname(mddev));
67ac6011
N		 5114 return -EINVAL;
5115 }
2c810cdd 5116 } else if (mddev->reshape_backwards
b5254dd5 5117 ? (here_new * mddev->new_chunk_sectors + min_offset_diff <=
67ac6011
N		 5118 here_old * mddev->chunk_sectors)
5119 : (here_new * mddev->new_chunk_sectors >=
b5254dd5 5120 here_old * mddev->chunk_sectors + (-min_offset_diff))) {
91adb564 5121 /* Reading from the same stripe as writing to - bad */
0c55e022
N		 5122 printk(KERN_ERR "md/raid:%s: reshape_position too early for "
5123 "auto-recovery - aborting.\n",
5124 mdname(mddev));
91adb564
N		 5125 return -EINVAL;
5126 }
0c55e022
N		 5127 printk(KERN_INFO "md/raid:%s: reshape will continue\n",
5128 mdname(mddev));
91adb564
N		 5129 /* OK, we should be able to continue; */
5130 } else {
5131 BUG_ON(mddev->level != mddev->new_level);
5132 BUG_ON(mddev->layout != mddev->new_layout);
664e7c41 5133 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
91adb564 5134 BUG_ON(mddev->delta_disks != 0);
1da177e4 5135 }
91adb564 5136
245f46c2
N		 5137 if (mddev->private == NULL)
5138 conf = setup_conf(mddev);
5139 else
5140 conf = mddev->private;
5141
91adb564
N		 5142 if (IS_ERR(conf))
5143 return PTR_ERR(conf);
5144
b5254dd5 5145 conf->min_offset_diff = min_offset_diff;
91adb564
N		 5146 mddev->thread = conf->thread;
5147 conf->thread = NULL;
5148 mddev->private = conf;
5149
17045f52
N		 5150 for (i = 0; i < conf->raid_disks && conf->previous_raid_disks;
5151 i++) {
5152 rdev = conf->disks[i].rdev;
5153 if (!rdev && conf->disks[i].replacement) {
5154 /* The replacement is all we have yet */
5155 rdev = conf->disks[i].replacement;
5156 conf->disks[i].replacement = NULL;
5157 clear_bit(Replacement, &rdev->flags);
5158 conf->disks[i].rdev = rdev;
5159 }
5160 if (!rdev)
c148ffdc 5161 continue;
17045f52
N		 5162 if (conf->disks[i].replacement &&
5163 conf->reshape_progress != MaxSector) {
5164 /* replacements and reshape simply do not mix. */
5165 printk(KERN_ERR "md: cannot handle concurrent "
5166 "replacement and reshape.\n");
5167 goto abort;
5168 }
2f115882 5169 if (test_bit(In_sync, &rdev->flags)) {
91adb564 5170 working_disks++;
2f115882
N		 5171 continue;
5172 }
c148ffdc
N		 5173 /* This disc is not fully in-sync. However if it
5174 * just stored parity (beyond the recovery_offset),
5175 * when we don't need to be concerned about the
5176 * array being dirty.
5177 * When reshape goes 'backwards', we never have
5178 * partially completed devices, so we only need
5179 * to worry about reshape going forwards.
5180 */
5181 /* Hack because v0.91 doesn't store recovery_offset properly. */
5182 if (mddev->major_version == 0 &&
5183 mddev->minor_version > 90)
5184 rdev->recovery_offset = reshape_offset;
5185
c148ffdc
N		 5186 if (rdev->recovery_offset < reshape_offset) {
5187 /* We need to check old and new layout */
5188 if (!only_parity(rdev->raid_disk,
5189 conf->algorithm,
5190 conf->raid_disks,
5191 conf->max_degraded))
5192 continue;
5193 }
5194 if (!only_parity(rdev->raid_disk,
5195 conf->prev_algo,
5196 conf->previous_raid_disks,
5197 conf->max_degraded))
5198 continue;
5199 dirty_parity_disks++;
5200 }
91adb564 5201
17045f52
N		 5202 /*
5203 * 0 for a fully functional array, 1 or 2 for a degraded array.
5204 */
908f4fbd 5205 mddev->degraded = calc_degraded(conf);
91adb564 5206
674806d6 5207 if (has_failed(conf)) {
0c55e022 5208 printk(KERN_ERR "md/raid:%s: not enough operational devices"
1da177e4 5209 " (%d/%d failed)\n",
02c2de8c 5210 mdname(mddev), mddev->degraded, conf->raid_disks);
1da177e4
LT		 5211 goto abort;
5212 }
5213
91adb564 5214 /* device size must be a multiple of chunk size */
9d8f0363 5215 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
91adb564
N		 5216 mddev->resync_max_sectors = mddev->dev_sectors;
5217
c148ffdc 5218 if (mddev->degraded > dirty_parity_disks &&
1da177e4 5219 mddev->recovery_cp != MaxSector) {
6ff8d8ec
N		 5220 if (mddev->ok_start_degraded)
5221 printk(KERN_WARNING
0c55e022
N		 5222 "md/raid:%s: starting dirty degraded array"
5223 " - data corruption possible.\n",
6ff8d8ec
N		 5224 mdname(mddev));
5225 else {
5226 printk(KERN_ERR
0c55e022 5227 "md/raid:%s: cannot start dirty degraded array.\n",
6ff8d8ec
N		 5228 mdname(mddev));
5229 goto abort;
5230 }
1da177e4
LT		 5231 }
5232
1da177e4 5233 if (mddev->degraded == 0)
0c55e022
N		 5234 printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d"
5235 " devices, algorithm %d\n", mdname(mddev), conf->level,
e183eaed
N		 5236 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
5237 mddev->new_layout);
1da177e4 5238 else
0c55e022
N		 5239 printk(KERN_ALERT "md/raid:%s: raid level %d active with %d"
5240 " out of %d devices, algorithm %d\n",
5241 mdname(mddev), conf->level,
5242 mddev->raid_disks - mddev->degraded,
5243 mddev->raid_disks, mddev->new_layout);
1da177e4
LT		 5244
5245 print_raid5_conf(conf);
5246
fef9c61f 5247 if (conf->reshape_progress != MaxSector) {
fef9c61f 5248 conf->reshape_safe = conf->reshape_progress;
f6705578
N		 5249 atomic_set(&conf->reshape_stripes, 0);
5250 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5251 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5252 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5253 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5254 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 5255 "reshape");
f6705578
N		 5256 }
5257
1da177e4
LT		 5258
5259 /* Ok, everything is just fine now */
a64c876f
N		 5260 if (mddev->to_remove == &raid5_attrs_group)
5261 mddev->to_remove = NULL;
00bcb4ac
N		 5262 else if (mddev->kobj.sd &&
5263 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
5e55e2f5 5264 printk(KERN_WARNING
4a5add49 5265 "raid5: failed to create sysfs attributes for %s\n",
5e55e2f5 5266 mdname(mddev));
4a5add49 5267 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
7a5febe9 5268
4a5add49 5269 if (mddev->queue) {
9f7c2220 5270 int chunk_size;
4a5add49
N		 5271 /* read-ahead size must cover two whole stripes, which
5272 * is 2 * (datadisks) * chunksize where 'n' is the
5273 * number of raid devices
5274 */
5275 int data_disks = conf->previous_raid_disks - conf->max_degraded;
5276 int stripe = data_disks *
5277 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
5278 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
5279 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
91adb564 5280
4a5add49 5281 blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec);
f022b2fd 5282
11d8a6e3
N		 5283 mddev->queue->backing_dev_info.congested_data = mddev;
5284 mddev->queue->backing_dev_info.congested_fn = raid5_congested;
7a5febe9 5285
9f7c2220
N		 5286 chunk_size = mddev->chunk_sectors << 9;
5287 blk_queue_io_min(mddev->queue, chunk_size);
5288 blk_queue_io_opt(mddev->queue, chunk_size *
5289 (conf->raid_disks - conf->max_degraded));
8f6c2e4b 5290
05616be5 5291 rdev_for_each(rdev, mddev) {
9f7c2220
N		 5292 disk_stack_limits(mddev->gendisk, rdev->bdev,
5293 rdev->data_offset << 9);
05616be5
N		 5294 disk_stack_limits(mddev->gendisk, rdev->bdev,
5295 rdev->new_data_offset << 9);
5296 }
9f7c2220 5297 }
23032a0e 5298
1da177e4
LT		 5299 return 0;
5300abort:
01f96c0a 5301 md_unregister_thread(&mddev->thread);
e4f869d9
N		 5302 print_raid5_conf(conf);
5303 free_conf(conf);
1da177e4 5304 mddev->private = NULL;
0c55e022 5305 printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev));
1da177e4
LT		 5306 return -EIO;
5307}
5308
fd01b88c 5309static int stop(struct mddev *mddev)
1da177e4 5310{
d1688a6d 5311 struct r5conf *conf = mddev->private;
1da177e4 5312
01f96c0a 5313 md_unregister_thread(&mddev->thread);
11d8a6e3
N		 5314 if (mddev->queue)
5315 mddev->queue->backing_dev_info.congested_fn = NULL;
95fc17aa 5316 free_conf(conf);
a64c876f
N		 5317 mddev->private = NULL;
5318 mddev->to_remove = &raid5_attrs_group;
1da177e4
LT		 5319 return 0;
5320}
5321
fd01b88c 5322static void status(struct seq_file *seq, struct mddev *mddev)
1da177e4 5323{
d1688a6d 5324 struct r5conf *conf = mddev->private;
1da177e4
LT		 5325 int i;
5326
9d8f0363
AN		 5327 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
5328 mddev->chunk_sectors / 2, mddev->layout);
02c2de8c 5329 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
1da177e4
LT		 5330 for (i = 0; i < conf->raid_disks; i++)
5331 seq_printf (seq, "%s",
5332 conf->disks[i].rdev &&
b2d444d7 5333 test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_");
1da177e4 5334 seq_printf (seq, "]");
1da177e4
LT		 5335}
5336
d1688a6d 5337static void print_raid5_conf (struct r5conf *conf)
1da177e4
LT		 5338{
5339 int i;
5340 struct disk_info *tmp;
5341
0c55e022 5342 printk(KERN_DEBUG "RAID conf printout:\n");
1da177e4
LT		 5343 if (!conf) {
5344 printk("(conf==NULL)\n");
5345 return;
5346 }
0c55e022
N		 5347 printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level,
5348 conf->raid_disks,
5349 conf->raid_disks - conf->mddev->degraded);
1da177e4
LT		 5350
5351 for (i = 0; i < conf->raid_disks; i++) {
5352 char b[BDEVNAME_SIZE];
5353 tmp = conf->disks + i;
5354 if (tmp->rdev)
0c55e022
N		 5355 printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n",
5356 i, !test_bit(Faulty, &tmp->rdev->flags),
5357 bdevname(tmp->rdev->bdev, b));
1da177e4
LT		 5358 }
5359}
5360
fd01b88c 5361static int raid5_spare_active(struct mddev *mddev)
1da177e4
LT		 5362{
5363 int i;
d1688a6d 5364 struct r5conf *conf = mddev->private;
1da177e4 5365 struct disk_info *tmp;
6b965620
N		 5366 int count = 0;
5367 unsigned long flags;
1da177e4
LT		 5368
5369 for (i = 0; i < conf->raid_disks; i++) {
5370 tmp = conf->disks + i;
dd054fce
N		 5371 if (tmp->replacement
5372 && tmp->replacement->recovery_offset == MaxSector
5373 && !test_bit(Faulty, &tmp->replacement->flags)
5374 && !test_and_set_bit(In_sync, &tmp->replacement->flags)) {
5375 /* Replacement has just become active. */
5376 if (!tmp->rdev
5377 || !test_and_clear_bit(In_sync, &tmp->rdev->flags))
5378 count++;
5379 if (tmp->rdev) {
5380 /* Replaced device not technically faulty,
5381 * but we need to be sure it gets removed
5382 * and never re-added.
5383 */
5384 set_bit(Faulty, &tmp->rdev->flags);
5385 sysfs_notify_dirent_safe(
5386 tmp->rdev->sysfs_state);
5387 }
5388 sysfs_notify_dirent_safe(tmp->replacement->sysfs_state);
5389 } else if (tmp->rdev
70fffd0b 5390 && tmp->rdev->recovery_offset == MaxSector
b2d444d7 5391 && !test_bit(Faulty, &tmp->rdev->flags)
c04be0aa 5392 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
6b965620 5393 count++;
43c73ca4 5394 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
1da177e4
LT		 5395 }
5396 }
6b965620 5397 spin_lock_irqsave(&conf->device_lock, flags);
908f4fbd 5398 mddev->degraded = calc_degraded(conf);
6b965620 5399 spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4 5400 print_raid5_conf(conf);
6b965620 5401 return count;
1da177e4
LT		 5402}
5403
b8321b68 5404static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 5405{
d1688a6d 5406 struct r5conf *conf = mddev->private;
1da177e4 5407 int err = 0;
b8321b68 5408 int number = rdev->raid_disk;
657e3e4d 5409 struct md_rdev **rdevp;
1da177e4
LT		 5410 struct disk_info *p = conf->disks + number;
5411
5412 print_raid5_conf(conf);
657e3e4d
N		 5413 if (rdev == p->rdev)
5414 rdevp = &p->rdev;
5415 else if (rdev == p->replacement)
5416 rdevp = &p->replacement;
5417 else
5418 return 0;
5419
5420 if (number >= conf->raid_disks &&
5421 conf->reshape_progress == MaxSector)
5422 clear_bit(In_sync, &rdev->flags);
5423
5424 if (test_bit(In_sync, &rdev->flags) ||
5425 atomic_read(&rdev->nr_pending)) {
5426 err = -EBUSY;
5427 goto abort;
5428 }
5429 /* Only remove non-faulty devices if recovery
5430 * isn't possible.
5431 */
5432 if (!test_bit(Faulty, &rdev->flags) &&
5433 mddev->recovery_disabled != conf->recovery_disabled &&
5434 !has_failed(conf) &&
dd054fce 5435 (!p->replacement || p->replacement == rdev) &&
657e3e4d
N		 5436 number < conf->raid_disks) {
5437 err = -EBUSY;
5438 goto abort;
5439 }
5440 *rdevp = NULL;
5441 synchronize_rcu();
5442 if (atomic_read(&rdev->nr_pending)) {
5443 /* lost the race, try later */
5444 err = -EBUSY;
5445 *rdevp = rdev;
dd054fce
N		 5446 } else if (p->replacement) {
5447 /* We must have just cleared 'rdev' */
5448 p->rdev = p->replacement;
5449 clear_bit(Replacement, &p->replacement->flags);
5450 smp_mb(); /* Make sure other CPUs may see both as identical
5451 * but will never see neither - if they are careful
5452 */
5453 p->replacement = NULL;
5454 clear_bit(WantReplacement, &rdev->flags);
5455 } else
5456 /* We might have just removed the Replacement as faulty-
5457 * clear the bit just in case
5458 */
5459 clear_bit(WantReplacement, &rdev->flags);
1da177e4
LT		 5460abort:
5461
5462 print_raid5_conf(conf);
5463 return err;
5464}
5465
fd01b88c 5466static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 5467{
d1688a6d 5468 struct r5conf *conf = mddev->private;
199050ea 5469 int err = -EEXIST;
1da177e4
LT		 5470 int disk;
5471 struct disk_info *p;
6c2fce2e
NB		 5472 int first = 0;
5473 int last = conf->raid_disks - 1;
1da177e4 5474
7f0da59b
N		 5475 if (mddev->recovery_disabled == conf->recovery_disabled)
5476 return -EBUSY;
5477
dc10c643 5478 if (rdev->saved_raid_disk < 0 && has_failed(conf))
1da177e4 5479 /* no point adding a device */
199050ea 5480 return -EINVAL;
1da177e4 5481
6c2fce2e
NB		 5482 if (rdev->raid_disk >= 0)
5483 first = last = rdev->raid_disk;
1da177e4
LT		 5484
5485 /*
16a53ecc
N		 5486 * find the disk ... but prefer rdev->saved_raid_disk
5487 * if possible.
1da177e4 5488 */
16a53ecc 5489 if (rdev->saved_raid_disk >= 0 &&
6c2fce2e 5490 rdev->saved_raid_disk >= first &&
16a53ecc 5491 conf->disks[rdev->saved_raid_disk].rdev == NULL)
5cfb22a1
N		 5492 first = rdev->saved_raid_disk;
5493
5494 for (disk = first; disk <= last; disk++) {
7bfec5f3
N		 5495 p = conf->disks + disk;
5496 if (p->rdev == NULL) {
b2d444d7 5497 clear_bit(In_sync, &rdev->flags);
1da177e4 5498 rdev->raid_disk = disk;
199050ea 5499 err = 0;
72626685
N		 5500 if (rdev->saved_raid_disk != disk)
5501 conf->fullsync = 1;
d6065f7b 5502 rcu_assign_pointer(p->rdev, rdev);
5cfb22a1 5503 goto out;
1da177e4 5504 }
5cfb22a1
N		 5505 }
5506 for (disk = first; disk <= last; disk++) {
5507 p = conf->disks + disk;
7bfec5f3
N		 5508 if (test_bit(WantReplacement, &p->rdev->flags) &&
5509 p->replacement == NULL) {
5510 clear_bit(In_sync, &rdev->flags);
5511 set_bit(Replacement, &rdev->flags);
5512 rdev->raid_disk = disk;
5513 err = 0;
5514 conf->fullsync = 1;
5515 rcu_assign_pointer(p->replacement, rdev);
5516 break;
5517 }
5518 }
5cfb22a1 5519out:
1da177e4 5520 print_raid5_conf(conf);
199050ea 5521 return err;
1da177e4
LT		 5522}
5523
fd01b88c 5524static int raid5_resize(struct mddev *mddev, sector_t sectors)
1da177e4
LT		 5525{
5526 /* no resync is happening, and there is enough space
5527 * on all devices, so we can resize.
5528 * We need to make sure resync covers any new space.
5529 * If the array is shrinking we should possibly wait until
5530 * any io in the removed space completes, but it hardly seems
5531 * worth it.
5532 */
a4a6125a 5533 sector_t newsize;
9d8f0363 5534 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
a4a6125a
N		 5535 newsize = raid5_size(mddev, sectors, mddev->raid_disks);
5536 if (mddev->external_size &&
5537 mddev->array_sectors > newsize)
b522adcd 5538 return -EINVAL;
a4a6125a
N		 5539 if (mddev->bitmap) {
5540 int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0);
5541 if (ret)
5542 return ret;
5543 }
5544 md_set_array_sectors(mddev, newsize);
f233ea5c 5545 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 5546 revalidate_disk(mddev->gendisk);
b098636c
N		 5547 if (sectors > mddev->dev_sectors &&
5548 mddev->recovery_cp > mddev->dev_sectors) {
58c0fed4 5549 mddev->recovery_cp = mddev->dev_sectors;
1da177e4
LT		 5550 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5551 }
58c0fed4 5552 mddev->dev_sectors = sectors;
4b5c7ae8 5553 mddev->resync_max_sectors = sectors;
1da177e4
LT		 5554 return 0;
5555}
5556
fd01b88c 5557static int check_stripe_cache(struct mddev *mddev)
01ee22b4
N		 5558{
5559 /* Can only proceed if there are plenty of stripe_heads.
5560 * We need a minimum of one full stripe,, and for sensible progress
5561 * it is best to have about 4 times that.
5562 * If we require 4 times, then the default 256 4K stripe_heads will
5563 * allow for chunk sizes up to 256K, which is probably OK.
5564 * If the chunk size is greater, user-space should request more
5565 * stripe_heads first.
5566 */
d1688a6d 5567 struct r5conf *conf = mddev->private;
01ee22b4
N		 5568 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
5569 > conf->max_nr_stripes ||
5570 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
5571 > conf->max_nr_stripes) {
0c55e022
N		 5572 printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n",
5573 mdname(mddev),
01ee22b4
N		 5574 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
5575 / STRIPE_SIZE)*4);
5576 return 0;
5577 }
5578 return 1;
5579}
5580
fd01b88c 5581static int check_reshape(struct mddev *mddev)
29269553 5582{
d1688a6d 5583 struct r5conf *conf = mddev->private;
29269553 5584
88ce4930
N		 5585 if (mddev->delta_disks == 0 &&
5586 mddev->new_layout == mddev->layout &&
664e7c41 5587 mddev->new_chunk_sectors == mddev->chunk_sectors)
50ac168a 5588 return 0; /* nothing to do */
674806d6 5589 if (has_failed(conf))
ec32a2bd
N		 5590 return -EINVAL;
5591 if (mddev->delta_disks < 0) {
5592 /* We might be able to shrink, but the devices must
5593 * be made bigger first.
5594 * For raid6, 4 is the minimum size.
5595 * Otherwise 2 is the minimum
5596 */
5597 int min = 2;
5598 if (mddev->level == 6)
5599 min = 4;
5600 if (mddev->raid_disks + mddev->delta_disks < min)
5601 return -EINVAL;
5602 }
29269553 5603
01ee22b4 5604 if (!check_stripe_cache(mddev))
29269553 5605 return -ENOSPC;
29269553 5606
ec32a2bd 5607 return resize_stripes(conf, conf->raid_disks + mddev->delta_disks);
63c70c4f
N		 5608}
5609
fd01b88c 5610static int raid5_start_reshape(struct mddev *mddev)
63c70c4f 5611{
d1688a6d 5612 struct r5conf *conf = mddev->private;
3cb03002 5613 struct md_rdev *rdev;
63c70c4f 5614 int spares = 0;
c04be0aa 5615 unsigned long flags;
63c70c4f 5616
f416885e 5617 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
63c70c4f
N		 5618 return -EBUSY;
5619
01ee22b4
N		 5620 if (!check_stripe_cache(mddev))
5621 return -ENOSPC;
5622
30b67645
N		 5623 if (has_failed(conf))
5624 return -EINVAL;
5625
c6563a8c 5626 rdev_for_each(rdev, mddev) {
469518a3
N		 5627 if (!test_bit(In_sync, &rdev->flags)
5628 && !test_bit(Faulty, &rdev->flags))
29269553 5629 spares++;
c6563a8c 5630 }
63c70c4f 5631
f416885e 5632 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
29269553
N		 5633 /* Not enough devices even to make a degraded array
5634 * of that size
5635 */
5636 return -EINVAL;
5637
ec32a2bd
N		 5638 /* Refuse to reduce size of the array. Any reductions in
5639 * array size must be through explicit setting of array_size
5640 * attribute.
5641 */
5642 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
5643 < mddev->array_sectors) {
0c55e022 5644 printk(KERN_ERR "md/raid:%s: array size must be reduced "
ec32a2bd
N		 5645 "before number of disks\n", mdname(mddev));
5646 return -EINVAL;
5647 }
5648
f6705578 5649 atomic_set(&conf->reshape_stripes, 0);
29269553
N		 5650 spin_lock_irq(&conf->device_lock);
5651 conf->previous_raid_disks = conf->raid_disks;
63c70c4f 5652 conf->raid_disks += mddev->delta_disks;
09c9e5fa
AN		 5653 conf->prev_chunk_sectors = conf->chunk_sectors;
5654 conf->chunk_sectors = mddev->new_chunk_sectors;
88ce4930
N		 5655 conf->prev_algo = conf->algorithm;
5656 conf->algorithm = mddev->new_layout;
05616be5
N		 5657 conf->generation++;
5658 /* Code that selects data_offset needs to see the generation update
5659 * if reshape_progress has been set - so a memory barrier needed.
5660 */
5661 smp_mb();
2c810cdd 5662 if (mddev->reshape_backwards)
fef9c61f
N		 5663 conf->reshape_progress = raid5_size(mddev, 0, 0);
5664 else
5665 conf->reshape_progress = 0;
5666 conf->reshape_safe = conf->reshape_progress;
29269553
N		 5667 spin_unlock_irq(&conf->device_lock);
5668
5669 /* Add some new drives, as many as will fit.
5670 * We know there are enough to make the newly sized array work.
3424bf6a
N		 5671 * Don't add devices if we are reducing the number of
5672 * devices in the array. This is because it is not possible
5673 * to correctly record the "partially reconstructed" state of
5674 * such devices during the reshape and confusion could result.
29269553 5675 */
87a8dec9 5676 if (mddev->delta_disks >= 0) {
dafb20fa 5677 rdev_for_each(rdev, mddev)
87a8dec9
N		 5678 if (rdev->raid_disk < 0 &&
5679 !test_bit(Faulty, &rdev->flags)) {
5680 if (raid5_add_disk(mddev, rdev) == 0) {
87a8dec9 5681 if (rdev->raid_disk
9d4c7d87 5682 >= conf->previous_raid_disks)
87a8dec9 5683 set_bit(In_sync, &rdev->flags);
9d4c7d87 5684 else
87a8dec9 5685 rdev->recovery_offset = 0;
36fad858
NK		 5686
5687 if (sysfs_link_rdev(mddev, rdev))
87a8dec9 5688 /* Failure here is OK */;
50da0840 5689 }
87a8dec9
N		 5690 } else if (rdev->raid_disk >= conf->previous_raid_disks
5691 && !test_bit(Faulty, &rdev->flags)) {
5692 /* This is a spare that was manually added */
5693 set_bit(In_sync, &rdev->flags);
87a8dec9 5694 }
29269553 5695
87a8dec9
N		 5696 /* When a reshape changes the number of devices,
5697 * ->degraded is measured against the larger of the
5698 * pre and post number of devices.
5699 */
ec32a2bd 5700 spin_lock_irqsave(&conf->device_lock, flags);
908f4fbd 5701 mddev->degraded = calc_degraded(conf);
ec32a2bd
N		 5702 spin_unlock_irqrestore(&conf->device_lock, flags);
5703 }
63c70c4f 5704 mddev->raid_disks = conf->raid_disks;
e516402c 5705 mddev->reshape_position = conf->reshape_progress;
850b2b42 5706 set_bit(MD_CHANGE_DEVS, &mddev->flags);
f6705578 5707
29269553
N		 5708 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5709 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5710 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5711 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5712 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 5713 "reshape");
29269553
N		 5714 if (!mddev->sync_thread) {
5715 mddev->recovery = 0;
5716 spin_lock_irq(&conf->device_lock);
5717 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
05616be5
N		 5718 rdev_for_each(rdev, mddev)
5719 rdev->new_data_offset = rdev->data_offset;
5720 smp_wmb();
fef9c61f 5721 conf->reshape_progress = MaxSector;
1e3fa9bd 5722 mddev->reshape_position = MaxSector;
29269553
N		 5723 spin_unlock_irq(&conf->device_lock);
5724 return -EAGAIN;
5725 }
c8f517c4 5726 conf->reshape_checkpoint = jiffies;
29269553
N		 5727 md_wakeup_thread(mddev->sync_thread);
5728 md_new_event(mddev);
5729 return 0;
5730}
29269553 5731
ec32a2bd
N		 5732/* This is called from the reshape thread and should make any
5733 * changes needed in 'conf'
5734 */
d1688a6d 5735static void end_reshape(struct r5conf *conf)
29269553 5736{
29269553 5737
f6705578 5738 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
05616be5 5739 struct md_rdev *rdev;
f6705578 5740
f6705578 5741 spin_lock_irq(&conf->device_lock);
cea9c228 5742 conf->previous_raid_disks = conf->raid_disks;
05616be5
N		 5743 rdev_for_each(rdev, conf->mddev)
5744 rdev->data_offset = rdev->new_data_offset;
5745 smp_wmb();
fef9c61f 5746 conf->reshape_progress = MaxSector;
f6705578 5747 spin_unlock_irq(&conf->device_lock);
b0f9ec04 5748 wake_up(&conf->wait_for_overlap);
16a53ecc
N		 5749
5750 /* read-ahead size must cover two whole stripes, which is
5751 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
5752 */
4a5add49 5753 if (conf->mddev->queue) {
cea9c228 5754 int data_disks = conf->raid_disks - conf->max_degraded;
09c9e5fa 5755 int stripe = data_disks * ((conf->chunk_sectors << 9)
cea9c228 5756 / PAGE_SIZE);
16a53ecc
N		 5757 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
5758 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
5759 }
29269553 5760 }
29269553
N		 5761}
5762
ec32a2bd
N		 5763/* This is called from the raid5d thread with mddev_lock held.
5764 * It makes config changes to the device.
5765 */
fd01b88c 5766static void raid5_finish_reshape(struct mddev *mddev)
cea9c228 5767{
d1688a6d 5768 struct r5conf *conf = mddev->private;
cea9c228
N		 5769
5770 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5771
ec32a2bd
N		 5772 if (mddev->delta_disks > 0) {
5773 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
5774 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 5775 revalidate_disk(mddev->gendisk);
ec32a2bd
N		 5776 } else {
5777 int d;
908f4fbd
N		 5778 spin_lock_irq(&conf->device_lock);
5779 mddev->degraded = calc_degraded(conf);
5780 spin_unlock_irq(&conf->device_lock);
ec32a2bd
N		 5781 for (d = conf->raid_disks ;
5782 d < conf->raid_disks - mddev->delta_disks;
1a67dde0 5783 d++) {
3cb03002 5784 struct md_rdev *rdev = conf->disks[d].rdev;
da7613b8
N		 5785 if (rdev)
5786 clear_bit(In_sync, &rdev->flags);
5787 rdev = conf->disks[d].replacement;
5788 if (rdev)
5789 clear_bit(In_sync, &rdev->flags);
1a67dde0 5790 }
cea9c228 5791 }
88ce4930 5792 mddev->layout = conf->algorithm;
09c9e5fa 5793 mddev->chunk_sectors = conf->chunk_sectors;
ec32a2bd
N		 5794 mddev->reshape_position = MaxSector;
5795 mddev->delta_disks = 0;
2c810cdd 5796 mddev->reshape_backwards = 0;
cea9c228
N		 5797 }
5798}
5799
fd01b88c 5800static void raid5_quiesce(struct mddev *mddev, int state)
72626685 5801{
d1688a6d 5802 struct r5conf *conf = mddev->private;
72626685
N		 5803
5804 switch(state) {
e464eafd
N		 5805 case 2: /* resume for a suspend */
5806 wake_up(&conf->wait_for_overlap);
5807 break;
5808
72626685
N		 5809 case 1: /* stop all writes */
5810 spin_lock_irq(&conf->device_lock);
64bd660b
N		 5811 /* '2' tells resync/reshape to pause so that all
5812 * active stripes can drain
5813 */
5814 conf->quiesce = 2;
72626685 5815 wait_event_lock_irq(conf->wait_for_stripe,
46031f9a
RBJ		 5816 atomic_read(&conf->active_stripes) == 0 &&
5817 atomic_read(&conf->active_aligned_reads) == 0,
72626685 5818 conf->device_lock, /* nothing */);
64bd660b 5819 conf->quiesce = 1;
72626685 5820 spin_unlock_irq(&conf->device_lock);
64bd660b
N		 5821 /* allow reshape to continue */
5822 wake_up(&conf->wait_for_overlap);
72626685
N		 5823 break;
5824
5825 case 0: /* re-enable writes */
5826 spin_lock_irq(&conf->device_lock);
5827 conf->quiesce = 0;
5828 wake_up(&conf->wait_for_stripe);
e464eafd 5829 wake_up(&conf->wait_for_overlap);
72626685
N		 5830 spin_unlock_irq(&conf->device_lock);
5831 break;
5832 }
72626685 5833}
b15c2e57 5834
d562b0c4 5835
fd01b88c 5836static void *raid45_takeover_raid0(struct mddev *mddev, int level)
54071b38 5837{
e373ab10 5838 struct r0conf *raid0_conf = mddev->private;
d76c8420 5839 sector_t sectors;
54071b38 5840
f1b29bca 5841 /* for raid0 takeover only one zone is supported */
e373ab10 5842 if (raid0_conf->nr_strip_zones > 1) {
0c55e022
N		 5843 printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n",
5844 mdname(mddev));
f1b29bca
DW		 5845 return ERR_PTR(-EINVAL);
5846 }
5847
e373ab10
N		 5848 sectors = raid0_conf->strip_zone[0].zone_end;
5849 sector_div(sectors, raid0_conf->strip_zone[0].nb_dev);
3b71bd93 5850 mddev->dev_sectors = sectors;
f1b29bca 5851 mddev->new_level = level;
54071b38
TM		 5852 mddev->new_layout = ALGORITHM_PARITY_N;
5853 mddev->new_chunk_sectors = mddev->chunk_sectors;
5854 mddev->raid_disks += 1;
5855 mddev->delta_disks = 1;
5856 /* make sure it will be not marked as dirty */
5857 mddev->recovery_cp = MaxSector;
5858
5859 return setup_conf(mddev);
5860}
5861
5862
fd01b88c 5863static void *raid5_takeover_raid1(struct mddev *mddev)
d562b0c4
N		 5864{
5865 int chunksect;
5866
5867 if (mddev->raid_disks != 2 ||
5868 mddev->degraded > 1)
5869 return ERR_PTR(-EINVAL);
5870
5871 /* Should check if there are write-behind devices? */
5872
5873 chunksect = 64*2; /* 64K by default */
5874
5875 /* The array must be an exact multiple of chunksize */
5876 while (chunksect && (mddev->array_sectors & (chunksect-1)))
5877 chunksect >>= 1;
5878
5879 if ((chunksect<<9) < STRIPE_SIZE)
5880 /* array size does not allow a suitable chunk size */
5881 return ERR_PTR(-EINVAL);
5882
5883 mddev->new_level = 5;
5884 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
664e7c41 5885 mddev->new_chunk_sectors = chunksect;
d562b0c4
N		 5886
5887 return setup_conf(mddev);
5888}
5889
fd01b88c 5890static void *raid5_takeover_raid6(struct mddev *mddev)
fc9739c6
N		 5891{
5892 int new_layout;
5893
5894 switch (mddev->layout) {
5895 case ALGORITHM_LEFT_ASYMMETRIC_6:
5896 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
5897 break;
5898 case ALGORITHM_RIGHT_ASYMMETRIC_6:
5899 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
5900 break;
5901 case ALGORITHM_LEFT_SYMMETRIC_6:
5902 new_layout = ALGORITHM_LEFT_SYMMETRIC;
5903 break;
5904 case ALGORITHM_RIGHT_SYMMETRIC_6:
5905 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
5906 break;
5907 case ALGORITHM_PARITY_0_6:
5908 new_layout = ALGORITHM_PARITY_0;
5909 break;
5910 case ALGORITHM_PARITY_N:
5911 new_layout = ALGORITHM_PARITY_N;
5912 break;
5913 default:
5914 return ERR_PTR(-EINVAL);
5915 }
5916 mddev->new_level = 5;
5917 mddev->new_layout = new_layout;
5918 mddev->delta_disks = -1;
5919 mddev->raid_disks -= 1;
5920 return setup_conf(mddev);
5921}
5922
d562b0c4 5923
fd01b88c 5924static int raid5_check_reshape(struct mddev *mddev)
b3546035 5925{
88ce4930
N		 5926 /* For a 2-drive array, the layout and chunk size can be changed
5927 * immediately as not restriping is needed.
5928 * For larger arrays we record the new value - after validation
5929 * to be used by a reshape pass.
b3546035 5930 */
d1688a6d 5931 struct r5conf *conf = mddev->private;
597a711b 5932 int new_chunk = mddev->new_chunk_sectors;
b3546035 5933
597a711b 5934 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
b3546035
N		 5935 return -EINVAL;
5936 if (new_chunk > 0) {
0ba459d2 5937 if (!is_power_of_2(new_chunk))
b3546035 5938 return -EINVAL;
597a711b 5939 if (new_chunk < (PAGE_SIZE>>9))
b3546035 5940 return -EINVAL;
597a711b 5941 if (mddev->array_sectors & (new_chunk-1))
b3546035
N		 5942 /* not factor of array size */
5943 return -EINVAL;
5944 }
5945
5946 /* They look valid */
5947
88ce4930 5948 if (mddev->raid_disks == 2) {
597a711b
N		 5949 /* can make the change immediately */
5950 if (mddev->new_layout >= 0) {
5951 conf->algorithm = mddev->new_layout;
5952 mddev->layout = mddev->new_layout;
88ce4930
N		 5953 }
5954 if (new_chunk > 0) {
597a711b
N		 5955 conf->chunk_sectors = new_chunk ;
5956 mddev->chunk_sectors = new_chunk;
88ce4930
N		 5957 }
5958 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5959 md_wakeup_thread(mddev->thread);
b3546035 5960 }
50ac168a 5961 return check_reshape(mddev);
88ce4930
N		 5962}
5963
fd01b88c 5964static int raid6_check_reshape(struct mddev *mddev)
88ce4930 5965{
597a711b 5966 int new_chunk = mddev->new_chunk_sectors;
50ac168a 5967
597a711b 5968 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
88ce4930 5969 return -EINVAL;
b3546035 5970 if (new_chunk > 0) {
0ba459d2 5971 if (!is_power_of_2(new_chunk))
88ce4930 5972 return -EINVAL;
597a711b 5973 if (new_chunk < (PAGE_SIZE >> 9))
88ce4930 5974 return -EINVAL;
597a711b 5975 if (mddev->array_sectors & (new_chunk-1))
88ce4930
N		 5976 /* not factor of array size */
5977 return -EINVAL;
b3546035 5978 }
88ce4930
N		 5979
5980 /* They look valid */
50ac168a 5981 return check_reshape(mddev);
b3546035
N		 5982}
5983
fd01b88c 5984static void *raid5_takeover(struct mddev *mddev)
d562b0c4
N		 5985{
5986 /* raid5 can take over:
f1b29bca 5987 * raid0 - if there is only one strip zone - make it a raid4 layout
d562b0c4
N		 5988 * raid1 - if there are two drives. We need to know the chunk size
5989 * raid4 - trivial - just use a raid4 layout.
5990 * raid6 - Providing it is a *_6 layout
d562b0c4 5991 */
f1b29bca
DW		 5992 if (mddev->level == 0)
5993 return raid45_takeover_raid0(mddev, 5);
d562b0c4
N		 5994 if (mddev->level == 1)
5995 return raid5_takeover_raid1(mddev);
e9d4758f
N		 5996 if (mddev->level == 4) {
5997 mddev->new_layout = ALGORITHM_PARITY_N;
5998 mddev->new_level = 5;
5999 return setup_conf(mddev);
6000 }
fc9739c6
N		 6001 if (mddev->level == 6)
6002 return raid5_takeover_raid6(mddev);
d562b0c4
N		 6003
6004 return ERR_PTR(-EINVAL);
6005}
6006
fd01b88c 6007static void *raid4_takeover(struct mddev *mddev)
a78d38a1 6008{
f1b29bca
DW		 6009 /* raid4 can take over:
6010 * raid0 - if there is only one strip zone
6011 * raid5 - if layout is right
a78d38a1 6012 */
f1b29bca
DW		 6013 if (mddev->level == 0)
6014 return raid45_takeover_raid0(mddev, 4);
a78d38a1
N		 6015 if (mddev->level == 5 &&
6016 mddev->layout == ALGORITHM_PARITY_N) {
6017 mddev->new_layout = 0;
6018 mddev->new_level = 4;
6019 return setup_conf(mddev);
6020 }
6021 return ERR_PTR(-EINVAL);
6022}
d562b0c4 6023
84fc4b56 6024static struct md_personality raid5_personality;
245f46c2 6025
fd01b88c 6026static void *raid6_takeover(struct mddev *mddev)
245f46c2
N		 6027{
6028 /* Currently can only take over a raid5. We map the
6029 * personality to an equivalent raid6 personality
6030 * with the Q block at the end.
6031 */
6032 int new_layout;
6033
6034 if (mddev->pers != &raid5_personality)
6035 return ERR_PTR(-EINVAL);
6036 if (mddev->degraded > 1)
6037 return ERR_PTR(-EINVAL);
6038 if (mddev->raid_disks > 253)
6039 return ERR_PTR(-EINVAL);
6040 if (mddev->raid_disks < 3)
6041 return ERR_PTR(-EINVAL);
6042
6043 switch (mddev->layout) {
6044 case ALGORITHM_LEFT_ASYMMETRIC:
6045 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
6046 break;
6047 case ALGORITHM_RIGHT_ASYMMETRIC:
6048 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
6049 break;
6050 case ALGORITHM_LEFT_SYMMETRIC:
6051 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
6052 break;
6053 case ALGORITHM_RIGHT_SYMMETRIC:
6054 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
6055 break;
6056 case ALGORITHM_PARITY_0:
6057 new_layout = ALGORITHM_PARITY_0_6;
6058 break;
6059 case ALGORITHM_PARITY_N:
6060 new_layout = ALGORITHM_PARITY_N;
6061 break;
6062 default:
6063 return ERR_PTR(-EINVAL);
6064 }
6065 mddev->new_level = 6;
6066 mddev->new_layout = new_layout;
6067 mddev->delta_disks = 1;
6068 mddev->raid_disks += 1;
6069 return setup_conf(mddev);
6070}
6071
6072
84fc4b56 6073static struct md_personality raid6_personality =
16a53ecc
N		 6074{
6075 .name = "raid6",
6076 .level = 6,
6077 .owner = THIS_MODULE,
6078 .make_request = make_request,
6079 .run = run,
6080 .stop = stop,
6081 .status = status,
6082 .error_handler = error,
6083 .hot_add_disk = raid5_add_disk,
6084 .hot_remove_disk= raid5_remove_disk,
6085 .spare_active = raid5_spare_active,
6086 .sync_request = sync_request,
6087 .resize = raid5_resize,
80c3a6ce 6088 .size = raid5_size,
50ac168a 6089 .check_reshape = raid6_check_reshape,
f416885e 6090 .start_reshape = raid5_start_reshape,
cea9c228 6091 .finish_reshape = raid5_finish_reshape,
16a53ecc 6092 .quiesce = raid5_quiesce,
245f46c2 6093 .takeover = raid6_takeover,
16a53ecc 6094};
84fc4b56 6095static struct md_personality raid5_personality =
1da177e4
LT		 6096{
6097 .name = "raid5",
2604b703 6098 .level = 5,
1da177e4
LT		 6099 .owner = THIS_MODULE,
6100 .make_request = make_request,
6101 .run = run,
6102 .stop = stop,
6103 .status = status,
6104 .error_handler = error,
6105 .hot_add_disk = raid5_add_disk,
6106 .hot_remove_disk= raid5_remove_disk,
6107 .spare_active = raid5_spare_active,
6108 .sync_request = sync_request,
6109 .resize = raid5_resize,
80c3a6ce 6110 .size = raid5_size,
63c70c4f
N		 6111 .check_reshape = raid5_check_reshape,
6112 .start_reshape = raid5_start_reshape,
cea9c228 6113 .finish_reshape = raid5_finish_reshape,
72626685 6114 .quiesce = raid5_quiesce,
d562b0c4 6115 .takeover = raid5_takeover,
1da177e4
LT		 6116};
6117
84fc4b56 6118static struct md_personality raid4_personality =
1da177e4 6119{
2604b703
N		 6120 .name = "raid4",
6121 .level = 4,
6122 .owner = THIS_MODULE,
6123 .make_request = make_request,
6124 .run = run,
6125 .stop = stop,
6126 .status = status,
6127 .error_handler = error,
6128 .hot_add_disk = raid5_add_disk,
6129 .hot_remove_disk= raid5_remove_disk,
6130 .spare_active = raid5_spare_active,
6131 .sync_request = sync_request,
6132 .resize = raid5_resize,
80c3a6ce 6133 .size = raid5_size,
3d37890b
N		 6134 .check_reshape = raid5_check_reshape,
6135 .start_reshape = raid5_start_reshape,
cea9c228 6136 .finish_reshape = raid5_finish_reshape,
2604b703 6137 .quiesce = raid5_quiesce,
a78d38a1 6138 .takeover = raid4_takeover,
2604b703
N		 6139};
6140
6141static int __init raid5_init(void)
6142{
16a53ecc 6143 register_md_personality(&raid6_personality);
2604b703
N		 6144 register_md_personality(&raid5_personality);
6145 register_md_personality(&raid4_personality);
6146 return 0;
1da177e4
LT		 6147}
6148
2604b703 6149static void raid5_exit(void)
1da177e4 6150{
16a53ecc 6151 unregister_md_personality(&raid6_personality);
2604b703
N		 6152 unregister_md_personality(&raid5_personality);
6153 unregister_md_personality(&raid4_personality);
1da177e4
LT		 6154}
6155
6156module_init(raid5_init);
6157module_exit(raid5_exit);
6158MODULE_LICENSE("GPL");
0efb9e61 6159MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
1da177e4 6160MODULE_ALIAS("md-personality-4"); /* RAID5 */
d9d166c2
N		 6161MODULE_ALIAS("md-raid5");
6162MODULE_ALIAS("md-raid4");
2604b703
N		 6163MODULE_ALIAS("md-level-5");
6164MODULE_ALIAS("md-level-4");
16a53ecc
N		 6165MODULE_ALIAS("md-personality-8"); /* RAID6 */
6166MODULE_ALIAS("md-raid6");
6167MODULE_ALIAS("md-level-6");
6168
6169/* This used to be two separate modules, they were: */
6170MODULE_ALIAS("raid5");
6171MODULE_ALIAS("raid6");
kernel source for telekom puls (alcatel/mediatek)