projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md/raid5: unify stripe_head_state and r6_state
[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>
07a3b417 50#include <linux/async.h>
bff61975 51#include <linux/seq_file.h>
36d1c647 52#include <linux/cpu.h>
5a0e3ad6 53#include <linux/slab.h>
43b2e5d8 54#include "md.h"
bff61975 55#include "raid5.h"
54071b38 56#include "raid0.h"
ef740c37 57#include "bitmap.h"
72626685 58
1da177e4
LT		 59/*
60 * Stripe cache
61 */
62
63#define NR_STRIPES 256
64#define STRIPE_SIZE PAGE_SIZE
65#define STRIPE_SHIFT (PAGE_SHIFT - 9)
66#define STRIPE_SECTORS (STRIPE_SIZE>>9)
67#define IO_THRESHOLD 1
8b3e6cdc 68#define BYPASS_THRESHOLD 1
fccddba0 69#define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
1da177e4
LT		 70#define HASH_MASK (NR_HASH - 1)
71
fccddba0 72#define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK]))
1da177e4
LT		 73
74/* bio's attached to a stripe+device for I/O are linked together in bi_sector
75 * order without overlap. There may be several bio's per stripe+device, and
76 * a bio could span several devices.
77 * When walking this list for a particular stripe+device, we must never proceed
78 * beyond a bio that extends past this device, as the next bio might no longer
79 * be valid.
80 * This macro is used to determine the 'next' bio in the list, given the sector
81 * of the current stripe+device
82 */
83#define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL)
84/*
85 * The following can be used to debug the driver
86 */
1da177e4
LT		 87#define RAID5_PARANOIA 1
88#if RAID5_PARANOIA && defined(CONFIG_SMP)
89# define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock)
90#else
91# define CHECK_DEVLOCK()
92#endif
93
45b4233c 94#ifdef DEBUG
1da177e4
LT		 95#define inline
96#define __inline__
97#endif
98
6be9d494
BS		 99#define printk_rl(args...) ((void) (printk_ratelimit() && printk(args)))
100
960e739d 101/*
5b99c2ff
JA		 102 * We maintain a biased count of active stripes in the bottom 16 bits of
103 * bi_phys_segments, and a count of processed stripes in the upper 16 bits
960e739d
JA		 104 */
105static inline int raid5_bi_phys_segments(struct bio *bio)
106{
5b99c2ff 107 return bio->bi_phys_segments & 0xffff;
960e739d
JA		 108}
109
110static inline int raid5_bi_hw_segments(struct bio *bio)
111{
5b99c2ff 112 return (bio->bi_phys_segments >> 16) & 0xffff;
960e739d
JA		 113}
114
115static inline int raid5_dec_bi_phys_segments(struct bio *bio)
116{
117 --bio->bi_phys_segments;
118 return raid5_bi_phys_segments(bio);
119}
120
121static inline int raid5_dec_bi_hw_segments(struct bio *bio)
122{
123 unsigned short val = raid5_bi_hw_segments(bio);
124
125 --val;
5b99c2ff 126 bio->bi_phys_segments = (val << 16) | raid5_bi_phys_segments(bio);
960e739d
JA		 127 return val;
128}
129
130static inline void raid5_set_bi_hw_segments(struct bio *bio, unsigned int cnt)
131{
9b2dc8b6 132 bio->bi_phys_segments = raid5_bi_phys_segments(bio) | (cnt << 16);
960e739d
JA		 133}
134
d0dabf7e
N		 135/* Find first data disk in a raid6 stripe */
136static inline int raid6_d0(struct stripe_head *sh)
137{
67cc2b81
N		 138 if (sh->ddf_layout)
139 /* ddf always start from first device */
140 return 0;
141 /* md starts just after Q block */
d0dabf7e
N		 142 if (sh->qd_idx == sh->disks - 1)
143 return 0;
144 else
145 return sh->qd_idx + 1;
146}
16a53ecc
N		 147static inline int raid6_next_disk(int disk, int raid_disks)
148{
149 disk++;
150 return (disk < raid_disks) ? disk : 0;
151}
a4456856 152
d0dabf7e
N		 153/* When walking through the disks in a raid5, starting at raid6_d0,
154 * We need to map each disk to a 'slot', where the data disks are slot
155 * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk
156 * is raid_disks-1. This help does that mapping.
157 */
67cc2b81
N		 158static int raid6_idx_to_slot(int idx, struct stripe_head *sh,
159 int *count, int syndrome_disks)
d0dabf7e 160{
6629542e 161 int slot = *count;
67cc2b81 162
e4424fee 163 if (sh->ddf_layout)
6629542e 164 (*count)++;
d0dabf7e 165 if (idx == sh->pd_idx)
67cc2b81 166 return syndrome_disks;
d0dabf7e 167 if (idx == sh->qd_idx)
67cc2b81 168 return syndrome_disks + 1;
e4424fee 169 if (!sh->ddf_layout)
6629542e 170 (*count)++;
d0dabf7e
N		 171 return slot;
172}
173
a4456856
DW		 174static void return_io(struct bio *return_bi)
175{
176 struct bio *bi = return_bi;
177 while (bi) {
a4456856
DW		 178
179 return_bi = bi->bi_next;
180 bi->bi_next = NULL;
181 bi->bi_size = 0;
0e13fe23 182 bio_endio(bi, 0);
a4456856
DW		 183 bi = return_bi;
184 }
185}
186
1da177e4
LT		 187static void print_raid5_conf (raid5_conf_t *conf);
188
600aa109
DW		 189static int stripe_operations_active(struct stripe_head *sh)
190{
191 return sh->check_state || sh->reconstruct_state ||
192 test_bit(STRIPE_BIOFILL_RUN, &sh->state) ||
193 test_bit(STRIPE_COMPUTE_RUN, &sh->state);
194}
195
858119e1 196static void __release_stripe(raid5_conf_t *conf, struct stripe_head *sh)
1da177e4
LT		 197{
198 if (atomic_dec_and_test(&sh->count)) {
78bafebd
ES		 199 BUG_ON(!list_empty(&sh->lru));
200 BUG_ON(atomic_read(&conf->active_stripes)==0);
1da177e4 201 if (test_bit(STRIPE_HANDLE, &sh->state)) {
482c0834 202 if (test_bit(STRIPE_DELAYED, &sh->state))
1da177e4 203 list_add_tail(&sh->lru, &conf->delayed_list);
482c0834
N		 204 else if (test_bit(STRIPE_BIT_DELAY, &sh->state) &&
205 sh->bm_seq - conf->seq_write > 0)
72626685 206 list_add_tail(&sh->lru, &conf->bitmap_list);
482c0834 207 else {
72626685 208 clear_bit(STRIPE_BIT_DELAY, &sh->state);
1da177e4 209 list_add_tail(&sh->lru, &conf->handle_list);
72626685 210 }
1da177e4
LT		 211 md_wakeup_thread(conf->mddev->thread);
212 } else {
600aa109 213 BUG_ON(stripe_operations_active(sh));
1da177e4
LT		 214 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
215 atomic_dec(&conf->preread_active_stripes);
216 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD)
217 md_wakeup_thread(conf->mddev->thread);
218 }
1da177e4 219 atomic_dec(&conf->active_stripes);
ccfcc3c1
N		 220 if (!test_bit(STRIPE_EXPANDING, &sh->state)) {
221 list_add_tail(&sh->lru, &conf->inactive_list);
1da177e4 222 wake_up(&conf->wait_for_stripe);
46031f9a
RBJ		 223 if (conf->retry_read_aligned)
224 md_wakeup_thread(conf->mddev->thread);
ccfcc3c1 225 }
1da177e4
LT		 226 }
227 }
228}
d0dabf7e 229
1da177e4
LT		 230static void release_stripe(struct stripe_head *sh)
231{
232 raid5_conf_t *conf = sh->raid_conf;
233 unsigned long flags;
16a53ecc 234
1da177e4
LT		 235 spin_lock_irqsave(&conf->device_lock, flags);
236 __release_stripe(conf, sh);
237 spin_unlock_irqrestore(&conf->device_lock, flags);
238}
239
fccddba0 240static inline void remove_hash(struct stripe_head *sh)
1da177e4 241{
45b4233c
DW		 242 pr_debug("remove_hash(), stripe %llu\n",
243 (unsigned long long)sh->sector);
1da177e4 244
fccddba0 245 hlist_del_init(&sh->hash);
1da177e4
LT		 246}
247
16a53ecc 248static inline void insert_hash(raid5_conf_t *conf, struct stripe_head *sh)
1da177e4 249{
fccddba0 250 struct hlist_head *hp = stripe_hash(conf, sh->sector);
1da177e4 251
45b4233c
DW		 252 pr_debug("insert_hash(), stripe %llu\n",
253 (unsigned long long)sh->sector);
1da177e4
LT		 254
255 CHECK_DEVLOCK();
fccddba0 256 hlist_add_head(&sh->hash, hp);
1da177e4
LT		 257}
258
259
260/* find an idle stripe, make sure it is unhashed, and return it. */
261static struct stripe_head *get_free_stripe(raid5_conf_t *conf)
262{
263 struct stripe_head *sh = NULL;
264 struct list_head *first;
265
266 CHECK_DEVLOCK();
267 if (list_empty(&conf->inactive_list))
268 goto out;
269 first = conf->inactive_list.next;
270 sh = list_entry(first, struct stripe_head, lru);
271 list_del_init(first);
272 remove_hash(sh);
273 atomic_inc(&conf->active_stripes);
274out:
275 return sh;
276}
277
e4e11e38 278static void shrink_buffers(struct stripe_head *sh)
1da177e4
LT		 279{
280 struct page *p;
281 int i;
e4e11e38 282 int num = sh->raid_conf->pool_size;
1da177e4 283
e4e11e38 284 for (i = 0; i < num ; i++) {
1da177e4
LT		 285 p = sh->dev[i].page;
286 if (!p)
287 continue;
288 sh->dev[i].page = NULL;
2d1f3b5d 289 put_page(p);
1da177e4
LT		 290 }
291}
292
e4e11e38 293static int grow_buffers(struct stripe_head *sh)
1da177e4
LT		 294{
295 int i;
e4e11e38 296 int num = sh->raid_conf->pool_size;
1da177e4 297
e4e11e38 298 for (i = 0; i < num; i++) {
1da177e4
LT		 299 struct page *page;
300
301 if (!(page = alloc_page(GFP_KERNEL))) {
302 return 1;
303 }
304 sh->dev[i].page = page;
305 }
306 return 0;
307}
308
784052ec 309static void raid5_build_block(struct stripe_head *sh, int i, int previous);
911d4ee8
N		 310static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous,
311 struct stripe_head *sh);
1da177e4 312
b5663ba4 313static void init_stripe(struct stripe_head *sh, sector_t sector, int previous)
1da177e4
LT		 314{
315 raid5_conf_t *conf = sh->raid_conf;
7ecaa1e6 316 int i;
1da177e4 317
78bafebd
ES		 318 BUG_ON(atomic_read(&sh->count) != 0);
319 BUG_ON(test_bit(STRIPE_HANDLE, &sh->state));
600aa109 320 BUG_ON(stripe_operations_active(sh));
d84e0f10 321
1da177e4 322 CHECK_DEVLOCK();
45b4233c 323 pr_debug("init_stripe called, stripe %llu\n",
1da177e4
LT		 324 (unsigned long long)sh->sector);
325
326 remove_hash(sh);
16a53ecc 327
86b42c71 328 sh->generation = conf->generation - previous;
b5663ba4 329 sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks;
1da177e4 330 sh->sector = sector;
911d4ee8 331 stripe_set_idx(sector, conf, previous, sh);
1da177e4
LT		 332 sh->state = 0;
333
7ecaa1e6
N		 334
335 for (i = sh->disks; i--; ) {
1da177e4
LT		 336 struct r5dev *dev = &sh->dev[i];
337
d84e0f10 338 if (dev->toread || dev->read || dev->towrite || dev->written ||
1da177e4 339 test_bit(R5_LOCKED, &dev->flags)) {
d84e0f10 340 printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n",
1da177e4 341 (unsigned long long)sh->sector, i, dev->toread,
d84e0f10 342 dev->read, dev->towrite, dev->written,
1da177e4
LT		 343 test_bit(R5_LOCKED, &dev->flags));
344 BUG();
345 }
346 dev->flags = 0;
784052ec 347 raid5_build_block(sh, i, previous);
1da177e4
LT		 348 }
349 insert_hash(conf, sh);
350}
351
86b42c71
N		 352static struct stripe_head *__find_stripe(raid5_conf_t *conf, sector_t sector,
353 short generation)
1da177e4
LT		 354{
355 struct stripe_head *sh;
fccddba0 356 struct hlist_node *hn;
1da177e4
LT		 357
358 CHECK_DEVLOCK();
45b4233c 359 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
fccddba0 360 hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash)
86b42c71 361 if (sh->sector == sector && sh->generation == generation)
1da177e4 362 return sh;
45b4233c 363 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
1da177e4
LT		 364 return NULL;
365}
366
674806d6
N		 367/*
368 * Need to check if array has failed when deciding whether to:
369 * - start an array
370 * - remove non-faulty devices
371 * - add a spare
372 * - allow a reshape
373 * This determination is simple when no reshape is happening.
374 * However if there is a reshape, we need to carefully check
375 * both the before and after sections.
376 * This is because some failed devices may only affect one
377 * of the two sections, and some non-in_sync devices may
378 * be insync in the section most affected by failed devices.
379 */
380static int has_failed(raid5_conf_t *conf)
381{
382 int degraded;
383 int i;
384 if (conf->mddev->reshape_position == MaxSector)
385 return conf->mddev->degraded > conf->max_degraded;
386
387 rcu_read_lock();
388 degraded = 0;
389 for (i = 0; i < conf->previous_raid_disks; i++) {
390 mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev);
391 if (!rdev || test_bit(Faulty, &rdev->flags))
392 degraded++;
393 else if (test_bit(In_sync, &rdev->flags))
394 ;
395 else
396 /* not in-sync or faulty.
397 * If the reshape increases the number of devices,
398 * this is being recovered by the reshape, so
399 * this 'previous' section is not in_sync.
400 * If the number of devices is being reduced however,
401 * the device can only be part of the array if
402 * we are reverting a reshape, so this section will
403 * be in-sync.
404 */
405 if (conf->raid_disks >= conf->previous_raid_disks)
406 degraded++;
407 }
408 rcu_read_unlock();
409 if (degraded > conf->max_degraded)
410 return 1;
411 rcu_read_lock();
412 degraded = 0;
413 for (i = 0; i < conf->raid_disks; i++) {
414 mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev);
415 if (!rdev || test_bit(Faulty, &rdev->flags))
416 degraded++;
417 else if (test_bit(In_sync, &rdev->flags))
418 ;
419 else
420 /* not in-sync or faulty.
421 * If reshape increases the number of devices, this
422 * section has already been recovered, else it
423 * almost certainly hasn't.
424 */
425 if (conf->raid_disks <= conf->previous_raid_disks)
426 degraded++;
427 }
428 rcu_read_unlock();
429 if (degraded > conf->max_degraded)
430 return 1;
431 return 0;
432}
433
b5663ba4
N		 434static struct stripe_head *
435get_active_stripe(raid5_conf_t *conf, sector_t sector,
a8c906ca 436 int previous, int noblock, int noquiesce)
1da177e4
LT		 437{
438 struct stripe_head *sh;
439
45b4233c 440 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector);
1da177e4
LT		 441
442 spin_lock_irq(&conf->device_lock);
443
444 do {
72626685 445 wait_event_lock_irq(conf->wait_for_stripe,
a8c906ca 446 conf->quiesce == 0 || noquiesce,
72626685 447 conf->device_lock, /* nothing */);
86b42c71 448 sh = __find_stripe(conf, sector, conf->generation - previous);
1da177e4
LT		 449 if (!sh) {
450 if (!conf->inactive_blocked)
451 sh = get_free_stripe(conf);
452 if (noblock && sh == NULL)
453 break;
454 if (!sh) {
455 conf->inactive_blocked = 1;
456 wait_event_lock_irq(conf->wait_for_stripe,
457 !list_empty(&conf->inactive_list) &&
5036805b
N		 458 (atomic_read(&conf->active_stripes)
459 < (conf->max_nr_stripes *3/4)
1da177e4
LT		 460 || !conf->inactive_blocked),
461 conf->device_lock,
7c13edc8 462 );
1da177e4
LT		 463 conf->inactive_blocked = 0;
464 } else
b5663ba4 465 init_stripe(sh, sector, previous);
1da177e4
LT		 466 } else {
467 if (atomic_read(&sh->count)) {
ab69ae12
N		 468 BUG_ON(!list_empty(&sh->lru)
469 && !test_bit(STRIPE_EXPANDING, &sh->state));
1da177e4
LT		 470 } else {
471 if (!test_bit(STRIPE_HANDLE, &sh->state))
472 atomic_inc(&conf->active_stripes);
ff4e8d9a
N		 473 if (list_empty(&sh->lru) &&
474 !test_bit(STRIPE_EXPANDING, &sh->state))
16a53ecc
N		 475 BUG();
476 list_del_init(&sh->lru);
1da177e4
LT		 477 }
478 }
479 } while (sh == NULL);
480
481 if (sh)
482 atomic_inc(&sh->count);
483
484 spin_unlock_irq(&conf->device_lock);
485 return sh;
486}
487
6712ecf8
N		 488static void
489raid5_end_read_request(struct bio *bi, int error);
490static void
491raid5_end_write_request(struct bio *bi, int error);
91c00924 492
c4e5ac0a 493static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
91c00924
DW		 494{
495 raid5_conf_t *conf = sh->raid_conf;
496 int i, disks = sh->disks;
497
498 might_sleep();
499
500 for (i = disks; i--; ) {
501 int rw;
502 struct bio *bi;
503 mdk_rdev_t *rdev;
e9c7469b
TH		 504 if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) {
505 if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags))
506 rw = WRITE_FUA;
507 else
508 rw = WRITE;
509 } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags))
91c00924
DW		 510 rw = READ;
511 else
512 continue;
513
514 bi = &sh->dev[i].req;
515
516 bi->bi_rw = rw;
b062962e 517 if (rw & WRITE)
91c00924
DW		 518 bi->bi_end_io = raid5_end_write_request;
519 else
520 bi->bi_end_io = raid5_end_read_request;
521
522 rcu_read_lock();
523 rdev = rcu_dereference(conf->disks[i].rdev);
524 if (rdev && test_bit(Faulty, &rdev->flags))
525 rdev = NULL;
526 if (rdev)
527 atomic_inc(&rdev->nr_pending);
528 rcu_read_unlock();
529
530 if (rdev) {
c4e5ac0a 531 if (s->syncing || s->expanding || s->expanded)
91c00924
DW		 532 md_sync_acct(rdev->bdev, STRIPE_SECTORS);
533
2b7497f0
DW		 534 set_bit(STRIPE_IO_STARTED, &sh->state);
535
91c00924
DW		 536 bi->bi_bdev = rdev->bdev;
537 pr_debug("%s: for %llu schedule op %ld on disc %d\n",
e46b272b 538 __func__, (unsigned long long)sh->sector,
91c00924
DW		 539 bi->bi_rw, i);
540 atomic_inc(&sh->count);
541 bi->bi_sector = sh->sector + rdev->data_offset;
542 bi->bi_flags = 1 << BIO_UPTODATE;
543 bi->bi_vcnt = 1;
544 bi->bi_max_vecs = 1;
545 bi->bi_idx = 0;
546 bi->bi_io_vec = &sh->dev[i].vec;
547 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
548 bi->bi_io_vec[0].bv_offset = 0;
549 bi->bi_size = STRIPE_SIZE;
550 bi->bi_next = NULL;
b062962e 551 if ((rw & WRITE) &&
91c00924
DW		 552 test_bit(R5_ReWrite, &sh->dev[i].flags))
553 atomic_add(STRIPE_SECTORS,
554 &rdev->corrected_errors);
555 generic_make_request(bi);
556 } else {
b062962e 557 if (rw & WRITE)
91c00924
DW		 558 set_bit(STRIPE_DEGRADED, &sh->state);
559 pr_debug("skip op %ld on disc %d for sector %llu\n",
560 bi->bi_rw, i, (unsigned long long)sh->sector);
561 clear_bit(R5_LOCKED, &sh->dev[i].flags);
562 set_bit(STRIPE_HANDLE, &sh->state);
563 }
564 }
565}
566
567static struct dma_async_tx_descriptor *
568async_copy_data(int frombio, struct bio *bio, struct page *page,
569 sector_t sector, struct dma_async_tx_descriptor *tx)
570{
571 struct bio_vec *bvl;
572 struct page *bio_page;
573 int i;
574 int page_offset;
a08abd8c 575 struct async_submit_ctl submit;
0403e382 576 enum async_tx_flags flags = 0;
91c00924
DW		 577
578 if (bio->bi_sector >= sector)
579 page_offset = (signed)(bio->bi_sector - sector) * 512;
580 else
581 page_offset = (signed)(sector - bio->bi_sector) * -512;
a08abd8c 582
0403e382
DW		 583 if (frombio)
584 flags |= ASYNC_TX_FENCE;
585 init_async_submit(&submit, flags, tx, NULL, NULL, NULL);
586
91c00924 587 bio_for_each_segment(bvl, bio, i) {
fcde9075 588 int len = bvl->bv_len;
91c00924
DW		 589 int clen;
590 int b_offset = 0;
591
592 if (page_offset < 0) {
593 b_offset = -page_offset;
594 page_offset += b_offset;
595 len -= b_offset;
596 }
597
598 if (len > 0 && page_offset + len > STRIPE_SIZE)
599 clen = STRIPE_SIZE - page_offset;
600 else
601 clen = len;
602
603 if (clen > 0) {
fcde9075
NK		 604 b_offset += bvl->bv_offset;
605 bio_page = bvl->bv_page;
91c00924
DW		 606 if (frombio)
607 tx = async_memcpy(page, bio_page, page_offset,
a08abd8c 608 b_offset, clen, &submit);
91c00924
DW		 609 else
610 tx = async_memcpy(bio_page, page, b_offset,
a08abd8c 611 page_offset, clen, &submit);
91c00924 612 }
a08abd8c
DW		 613 /* chain the operations */
614 submit.depend_tx = tx;
615
91c00924
DW		 616 if (clen < len) /* hit end of page */
617 break;
618 page_offset += len;
619 }
620
621 return tx;
622}
623
624static void ops_complete_biofill(void *stripe_head_ref)
625{
626 struct stripe_head *sh = stripe_head_ref;
627 struct bio *return_bi = NULL;
628 raid5_conf_t *conf = sh->raid_conf;
e4d84909 629 int i;
91c00924 630
e46b272b 631 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 632 (unsigned long long)sh->sector);
633
634 /* clear completed biofills */
83de75cc 635 spin_lock_irq(&conf->device_lock);
91c00924
DW		 636 for (i = sh->disks; i--; ) {
637 struct r5dev *dev = &sh->dev[i];
91c00924
DW		 638
639 /* acknowledge completion of a biofill operation */
e4d84909
DW		 640 /* and check if we need to reply to a read request,
641 * new R5_Wantfill requests are held off until
83de75cc 642 * !STRIPE_BIOFILL_RUN
e4d84909
DW		 643 */
644 if (test_and_clear_bit(R5_Wantfill, &dev->flags)) {
91c00924 645 struct bio *rbi, *rbi2;
91c00924 646
91c00924
DW		 647 BUG_ON(!dev->read);
648 rbi = dev->read;
649 dev->read = NULL;
650 while (rbi && rbi->bi_sector <
651 dev->sector + STRIPE_SECTORS) {
652 rbi2 = r5_next_bio(rbi, dev->sector);
960e739d 653 if (!raid5_dec_bi_phys_segments(rbi)) {
91c00924
DW		 654 rbi->bi_next = return_bi;
655 return_bi = rbi;
656 }
91c00924
DW		 657 rbi = rbi2;
658 }
659 }
660 }
83de75cc
DW		 661 spin_unlock_irq(&conf->device_lock);
662 clear_bit(STRIPE_BIOFILL_RUN, &sh->state);
91c00924
DW		 663
664 return_io(return_bi);
665
e4d84909 666 set_bit(STRIPE_HANDLE, &sh->state);
91c00924
DW		 667 release_stripe(sh);
668}
669
670static void ops_run_biofill(struct stripe_head *sh)
671{
672 struct dma_async_tx_descriptor *tx = NULL;
673 raid5_conf_t *conf = sh->raid_conf;
a08abd8c 674 struct async_submit_ctl submit;
91c00924
DW		 675 int i;
676
e46b272b 677 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 678 (unsigned long long)sh->sector);
679
680 for (i = sh->disks; i--; ) {
681 struct r5dev *dev = &sh->dev[i];
682 if (test_bit(R5_Wantfill, &dev->flags)) {
683 struct bio *rbi;
684 spin_lock_irq(&conf->device_lock);
685 dev->read = rbi = dev->toread;
686 dev->toread = NULL;
687 spin_unlock_irq(&conf->device_lock);
688 while (rbi && rbi->bi_sector <
689 dev->sector + STRIPE_SECTORS) {
690 tx = async_copy_data(0, rbi, dev->page,
691 dev->sector, tx);
692 rbi = r5_next_bio(rbi, dev->sector);
693 }
694 }
695 }
696
697 atomic_inc(&sh->count);
a08abd8c
DW		 698 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL);
699 async_trigger_callback(&submit);
91c00924
DW		 700}
701
4e7d2c0a 702static void mark_target_uptodate(struct stripe_head *sh, int target)
91c00924 703{
4e7d2c0a 704 struct r5dev *tgt;
91c00924 705
4e7d2c0a
DW		 706 if (target < 0)
707 return;
91c00924 708
4e7d2c0a 709 tgt = &sh->dev[target];
91c00924
DW		 710 set_bit(R5_UPTODATE, &tgt->flags);
711 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
712 clear_bit(R5_Wantcompute, &tgt->flags);
4e7d2c0a
DW		 713}
714
ac6b53b6 715static void ops_complete_compute(void *stripe_head_ref)
91c00924
DW		 716{
717 struct stripe_head *sh = stripe_head_ref;
91c00924 718
e46b272b 719 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 720 (unsigned long long)sh->sector);
721
ac6b53b6 722 /* mark the computed target(s) as uptodate */
4e7d2c0a 723 mark_target_uptodate(sh, sh->ops.target);
ac6b53b6 724 mark_target_uptodate(sh, sh->ops.target2);
4e7d2c0a 725
ecc65c9b
DW		 726 clear_bit(STRIPE_COMPUTE_RUN, &sh->state);
727 if (sh->check_state == check_state_compute_run)
728 sh->check_state = check_state_compute_result;
91c00924
DW		 729 set_bit(STRIPE_HANDLE, &sh->state);
730 release_stripe(sh);
731}
732
d6f38f31
DW		 733/* return a pointer to the address conversion region of the scribble buffer */
734static addr_conv_t *to_addr_conv(struct stripe_head *sh,
735 struct raid5_percpu *percpu)
736{
737 return percpu->scribble + sizeof(struct page *) * (sh->disks + 2);
738}
739
740static struct dma_async_tx_descriptor *
741ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 742{
91c00924 743 int disks = sh->disks;
d6f38f31 744 struct page **xor_srcs = percpu->scribble;
91c00924
DW		 745 int target = sh->ops.target;
746 struct r5dev *tgt = &sh->dev[target];
747 struct page *xor_dest = tgt->page;
748 int count = 0;
749 struct dma_async_tx_descriptor *tx;
a08abd8c 750 struct async_submit_ctl submit;
91c00924
DW		 751 int i;
752
753 pr_debug("%s: stripe %llu block: %d\n",
e46b272b 754 __func__, (unsigned long long)sh->sector, target);
91c00924
DW		 755 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
756
757 for (i = disks; i--; )
758 if (i != target)
759 xor_srcs[count++] = sh->dev[i].page;
760
761 atomic_inc(&sh->count);
762
0403e382 763 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL,
ac6b53b6 764 ops_complete_compute, sh, to_addr_conv(sh, percpu));
91c00924 765 if (unlikely(count == 1))
a08abd8c 766 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
91c00924 767 else
a08abd8c 768 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924 769
91c00924
DW		 770 return tx;
771}
772
ac6b53b6
DW		 773/* set_syndrome_sources - populate source buffers for gen_syndrome
774 * @srcs - (struct page *) array of size sh->disks
775 * @sh - stripe_head to parse
776 *
777 * Populates srcs in proper layout order for the stripe and returns the
778 * 'count' of sources to be used in a call to async_gen_syndrome. The P
779 * destination buffer is recorded in srcs[count] and the Q destination
780 * is recorded in srcs[count+1]].
781 */
782static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh)
783{
784 int disks = sh->disks;
785 int syndrome_disks = sh->ddf_layout ? disks : (disks - 2);
786 int d0_idx = raid6_d0(sh);
787 int count;
788 int i;
789
790 for (i = 0; i < disks; i++)
5dd33c9a 791 srcs[i] = NULL;
ac6b53b6
DW		 792
793 count = 0;
794 i = d0_idx;
795 do {
796 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
797
798 srcs[slot] = sh->dev[i].page;
799 i = raid6_next_disk(i, disks);
800 } while (i != d0_idx);
ac6b53b6 801
e4424fee 802 return syndrome_disks;
ac6b53b6
DW		 803}
804
805static struct dma_async_tx_descriptor *
806ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu)
807{
808 int disks = sh->disks;
809 struct page **blocks = percpu->scribble;
810 int target;
811 int qd_idx = sh->qd_idx;
812 struct dma_async_tx_descriptor *tx;
813 struct async_submit_ctl submit;
814 struct r5dev *tgt;
815 struct page *dest;
816 int i;
817 int count;
818
819 if (sh->ops.target < 0)
820 target = sh->ops.target2;
821 else if (sh->ops.target2 < 0)
822 target = sh->ops.target;
91c00924 823 else
ac6b53b6
DW		 824 /* we should only have one valid target */
825 BUG();
826 BUG_ON(target < 0);
827 pr_debug("%s: stripe %llu block: %d\n",
828 __func__, (unsigned long long)sh->sector, target);
829
830 tgt = &sh->dev[target];
831 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
832 dest = tgt->page;
833
834 atomic_inc(&sh->count);
835
836 if (target == qd_idx) {
837 count = set_syndrome_sources(blocks, sh);
838 blocks[count] = NULL; /* regenerating p is not necessary */
839 BUG_ON(blocks[count+1] != dest); /* q should already be set */
0403e382
DW		 840 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
841 ops_complete_compute, sh,
ac6b53b6
DW		 842 to_addr_conv(sh, percpu));
843 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
844 } else {
845 /* Compute any data- or p-drive using XOR */
846 count = 0;
847 for (i = disks; i-- ; ) {
848 if (i == target || i == qd_idx)
849 continue;
850 blocks[count++] = sh->dev[i].page;
851 }
852
0403e382
DW		 853 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
854 NULL, ops_complete_compute, sh,
ac6b53b6
DW		 855 to_addr_conv(sh, percpu));
856 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit);
857 }
91c00924 858
91c00924
DW		 859 return tx;
860}
861
ac6b53b6
DW		 862static struct dma_async_tx_descriptor *
863ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu)
864{
865 int i, count, disks = sh->disks;
866 int syndrome_disks = sh->ddf_layout ? disks : disks-2;
867 int d0_idx = raid6_d0(sh);
868 int faila = -1, failb = -1;
869 int target = sh->ops.target;
870 int target2 = sh->ops.target2;
871 struct r5dev *tgt = &sh->dev[target];
872 struct r5dev *tgt2 = &sh->dev[target2];
873 struct dma_async_tx_descriptor *tx;
874 struct page **blocks = percpu->scribble;
875 struct async_submit_ctl submit;
876
877 pr_debug("%s: stripe %llu block1: %d block2: %d\n",
878 __func__, (unsigned long long)sh->sector, target, target2);
879 BUG_ON(target < 0 || target2 < 0);
880 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
881 BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags));
882
6c910a78 883 /* we need to open-code set_syndrome_sources to handle the
ac6b53b6
DW		 884 * slot number conversion for 'faila' and 'failb'
885 */
886 for (i = 0; i < disks ; i++)
5dd33c9a 887 blocks[i] = NULL;
ac6b53b6
DW		 888 count = 0;
889 i = d0_idx;
890 do {
891 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
892
893 blocks[slot] = sh->dev[i].page;
894
895 if (i == target)
896 faila = slot;
897 if (i == target2)
898 failb = slot;
899 i = raid6_next_disk(i, disks);
900 } while (i != d0_idx);
ac6b53b6
DW		 901
902 BUG_ON(faila == failb);
903 if (failb < faila)
904 swap(faila, failb);
905 pr_debug("%s: stripe: %llu faila: %d failb: %d\n",
906 __func__, (unsigned long long)sh->sector, faila, failb);
907
908 atomic_inc(&sh->count);
909
910 if (failb == syndrome_disks+1) {
911 /* Q disk is one of the missing disks */
912 if (faila == syndrome_disks) {
913 /* Missing P+Q, just recompute */
0403e382
DW		 914 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
915 ops_complete_compute, sh,
916 to_addr_conv(sh, percpu));
e4424fee 917 return async_gen_syndrome(blocks, 0, syndrome_disks+2,
ac6b53b6
DW		 918 STRIPE_SIZE, &submit);
919 } else {
920 struct page *dest;
921 int data_target;
922 int qd_idx = sh->qd_idx;
923
924 /* Missing D+Q: recompute D from P, then recompute Q */
925 if (target == qd_idx)
926 data_target = target2;
927 else
928 data_target = target;
929
930 count = 0;
931 for (i = disks; i-- ; ) {
932 if (i == data_target || i == qd_idx)
933 continue;
934 blocks[count++] = sh->dev[i].page;
935 }
936 dest = sh->dev[data_target].page;
0403e382
DW		 937 init_async_submit(&submit,
938 ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
939 NULL, NULL, NULL,
940 to_addr_conv(sh, percpu));
ac6b53b6
DW		 941 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE,
942 &submit);
943
944 count = set_syndrome_sources(blocks, sh);
0403e382
DW		 945 init_async_submit(&submit, ASYNC_TX_FENCE, tx,
946 ops_complete_compute, sh,
947 to_addr_conv(sh, percpu));
ac6b53b6
DW		 948 return async_gen_syndrome(blocks, 0, count+2,
949 STRIPE_SIZE, &submit);
950 }
ac6b53b6 951 } else {
6c910a78
DW		 952 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
953 ops_complete_compute, sh,
954 to_addr_conv(sh, percpu));
955 if (failb == syndrome_disks) {
956 /* We're missing D+P. */
957 return async_raid6_datap_recov(syndrome_disks+2,
958 STRIPE_SIZE, faila,
959 blocks, &submit);
960 } else {
961 /* We're missing D+D. */
962 return async_raid6_2data_recov(syndrome_disks+2,
963 STRIPE_SIZE, faila, failb,
964 blocks, &submit);
965 }
ac6b53b6
DW		 966 }
967}
968
969
91c00924
DW		 970static void ops_complete_prexor(void *stripe_head_ref)
971{
972 struct stripe_head *sh = stripe_head_ref;
973
e46b272b 974 pr_debug("%s: stripe %llu\n", __func__,
91c00924 975 (unsigned long long)sh->sector);
91c00924
DW		 976}
977
978static struct dma_async_tx_descriptor *
d6f38f31
DW		 979ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu,
980 struct dma_async_tx_descriptor *tx)
91c00924 981{
91c00924 982 int disks = sh->disks;
d6f38f31 983 struct page **xor_srcs = percpu->scribble;
91c00924 984 int count = 0, pd_idx = sh->pd_idx, i;
a08abd8c 985 struct async_submit_ctl submit;
91c00924
DW		 986
987 /* existing parity data subtracted */
988 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
989
e46b272b 990 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 991 (unsigned long long)sh->sector);
992
993 for (i = disks; i--; ) {
994 struct r5dev *dev = &sh->dev[i];
995 /* Only process blocks that are known to be uptodate */
d8ee0728 996 if (test_bit(R5_Wantdrain, &dev->flags))
91c00924
DW		 997 xor_srcs[count++] = dev->page;
998 }
999
0403e382 1000 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
d6f38f31 1001 ops_complete_prexor, sh, to_addr_conv(sh, percpu));
a08abd8c 1002 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924
DW		 1003
1004 return tx;
1005}
1006
1007static struct dma_async_tx_descriptor *
d8ee0728 1008ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
91c00924
DW		 1009{
1010 int disks = sh->disks;
d8ee0728 1011 int i;
91c00924 1012
e46b272b 1013 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1014 (unsigned long long)sh->sector);
1015
1016 for (i = disks; i--; ) {
1017 struct r5dev *dev = &sh->dev[i];
1018 struct bio *chosen;
91c00924 1019
d8ee0728 1020 if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) {
91c00924
DW		 1021 struct bio *wbi;
1022
cbe47ec5 1023 spin_lock_irq(&sh->raid_conf->device_lock);
91c00924
DW		 1024 chosen = dev->towrite;
1025 dev->towrite = NULL;
1026 BUG_ON(dev->written);
1027 wbi = dev->written = chosen;
cbe47ec5 1028 spin_unlock_irq(&sh->raid_conf->device_lock);
91c00924
DW		 1029
1030 while (wbi && wbi->bi_sector <
1031 dev->sector + STRIPE_SECTORS) {
e9c7469b
TH		 1032 if (wbi->bi_rw & REQ_FUA)
1033 set_bit(R5_WantFUA, &dev->flags);
91c00924
DW		 1034 tx = async_copy_data(1, wbi, dev->page,
1035 dev->sector, tx);
1036 wbi = r5_next_bio(wbi, dev->sector);
1037 }
1038 }
1039 }
1040
1041 return tx;
1042}
1043
ac6b53b6 1044static void ops_complete_reconstruct(void *stripe_head_ref)
91c00924
DW		 1045{
1046 struct stripe_head *sh = stripe_head_ref;
ac6b53b6
DW		 1047 int disks = sh->disks;
1048 int pd_idx = sh->pd_idx;
1049 int qd_idx = sh->qd_idx;
1050 int i;
e9c7469b 1051 bool fua = false;
91c00924 1052
e46b272b 1053 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1054 (unsigned long long)sh->sector);
1055
e9c7469b
TH		 1056 for (i = disks; i--; )
1057 fua |= test_bit(R5_WantFUA, &sh->dev[i].flags);
1058
91c00924
DW		 1059 for (i = disks; i--; ) {
1060 struct r5dev *dev = &sh->dev[i];
ac6b53b6 1061
e9c7469b 1062 if (dev->written || i == pd_idx || i == qd_idx) {
91c00924 1063 set_bit(R5_UPTODATE, &dev->flags);
e9c7469b
TH		 1064 if (fua)
1065 set_bit(R5_WantFUA, &dev->flags);
1066 }
91c00924
DW		 1067 }
1068
d8ee0728
DW		 1069 if (sh->reconstruct_state == reconstruct_state_drain_run)
1070 sh->reconstruct_state = reconstruct_state_drain_result;
1071 else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run)
1072 sh->reconstruct_state = reconstruct_state_prexor_drain_result;
1073 else {
1074 BUG_ON(sh->reconstruct_state != reconstruct_state_run);
1075 sh->reconstruct_state = reconstruct_state_result;
1076 }
91c00924
DW		 1077
1078 set_bit(STRIPE_HANDLE, &sh->state);
1079 release_stripe(sh);
1080}
1081
1082static void
ac6b53b6
DW		 1083ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu,
1084 struct dma_async_tx_descriptor *tx)
91c00924 1085{
91c00924 1086 int disks = sh->disks;
d6f38f31 1087 struct page **xor_srcs = percpu->scribble;
a08abd8c 1088 struct async_submit_ctl submit;
91c00924
DW		 1089 int count = 0, pd_idx = sh->pd_idx, i;
1090 struct page *xor_dest;
d8ee0728 1091 int prexor = 0;
91c00924 1092 unsigned long flags;
91c00924 1093
e46b272b 1094 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1095 (unsigned long long)sh->sector);
1096
1097 /* check if prexor is active which means only process blocks
1098 * that are part of a read-modify-write (written)
1099 */
d8ee0728
DW		 1100 if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
1101 prexor = 1;
91c00924
DW		 1102 xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1103 for (i = disks; i--; ) {
1104 struct r5dev *dev = &sh->dev[i];
1105 if (dev->written)
1106 xor_srcs[count++] = dev->page;
1107 }
1108 } else {
1109 xor_dest = sh->dev[pd_idx].page;
1110 for (i = disks; i--; ) {
1111 struct r5dev *dev = &sh->dev[i];
1112 if (i != pd_idx)
1113 xor_srcs[count++] = dev->page;
1114 }
1115 }
1116
91c00924
DW		 1117 /* 1/ if we prexor'd then the dest is reused as a source
1118 * 2/ if we did not prexor then we are redoing the parity
1119 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
1120 * for the synchronous xor case
1121 */
88ba2aa5 1122 flags = ASYNC_TX_ACK |
91c00924
DW		 1123 (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
1124
1125 atomic_inc(&sh->count);
1126
ac6b53b6 1127 init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh,
d6f38f31 1128 to_addr_conv(sh, percpu));
a08abd8c
DW		 1129 if (unlikely(count == 1))
1130 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
1131 else
1132 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924
DW		 1133}
1134
ac6b53b6
DW		 1135static void
1136ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu,
1137 struct dma_async_tx_descriptor *tx)
1138{
1139 struct async_submit_ctl submit;
1140 struct page **blocks = percpu->scribble;
1141 int count;
1142
1143 pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector);
1144
1145 count = set_syndrome_sources(blocks, sh);
1146
1147 atomic_inc(&sh->count);
1148
1149 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct,
1150 sh, to_addr_conv(sh, percpu));
1151 async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
91c00924
DW		 1152}
1153
1154static void ops_complete_check(void *stripe_head_ref)
1155{
1156 struct stripe_head *sh = stripe_head_ref;
91c00924 1157
e46b272b 1158 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1159 (unsigned long long)sh->sector);
1160
ecc65c9b 1161 sh->check_state = check_state_check_result;
91c00924
DW		 1162 set_bit(STRIPE_HANDLE, &sh->state);
1163 release_stripe(sh);
1164}
1165
ac6b53b6 1166static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 1167{
91c00924 1168 int disks = sh->disks;
ac6b53b6
DW		 1169 int pd_idx = sh->pd_idx;
1170 int qd_idx = sh->qd_idx;
1171 struct page *xor_dest;
d6f38f31 1172 struct page **xor_srcs = percpu->scribble;
91c00924 1173 struct dma_async_tx_descriptor *tx;
a08abd8c 1174 struct async_submit_ctl submit;
ac6b53b6
DW		 1175 int count;
1176 int i;
91c00924 1177
e46b272b 1178 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1179 (unsigned long long)sh->sector);
1180
ac6b53b6
DW		 1181 count = 0;
1182 xor_dest = sh->dev[pd_idx].page;
1183 xor_srcs[count++] = xor_dest;
91c00924 1184 for (i = disks; i--; ) {
ac6b53b6
DW		 1185 if (i == pd_idx || i == qd_idx)
1186 continue;
1187 xor_srcs[count++] = sh->dev[i].page;
91c00924
DW		 1188 }
1189
d6f38f31
DW		 1190 init_async_submit(&submit, 0, NULL, NULL, NULL,
1191 to_addr_conv(sh, percpu));
099f53cb 1192 tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
a08abd8c 1193 &sh->ops.zero_sum_result, &submit);
91c00924 1194
91c00924 1195 atomic_inc(&sh->count);
a08abd8c
DW		 1196 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL);
1197 tx = async_trigger_callback(&submit);
91c00924
DW		 1198}
1199
ac6b53b6
DW		 1200static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp)
1201{
1202 struct page **srcs = percpu->scribble;
1203 struct async_submit_ctl submit;
1204 int count;
1205
1206 pr_debug("%s: stripe %llu checkp: %d\n", __func__,
1207 (unsigned long long)sh->sector, checkp);
1208
1209 count = set_syndrome_sources(srcs, sh);
1210 if (!checkp)
1211 srcs[count] = NULL;
91c00924 1212
91c00924 1213 atomic_inc(&sh->count);
ac6b53b6
DW		 1214 init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check,
1215 sh, to_addr_conv(sh, percpu));
1216 async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE,
1217 &sh->ops.zero_sum_result, percpu->spare_page, &submit);
91c00924
DW		 1218}
1219
417b8d4a 1220static void __raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
91c00924
DW		 1221{
1222 int overlap_clear = 0, i, disks = sh->disks;
1223 struct dma_async_tx_descriptor *tx = NULL;
d6f38f31 1224 raid5_conf_t *conf = sh->raid_conf;
ac6b53b6 1225 int level = conf->level;
d6f38f31
DW		 1226 struct raid5_percpu *percpu;
1227 unsigned long cpu;
91c00924 1228
d6f38f31
DW		 1229 cpu = get_cpu();
1230 percpu = per_cpu_ptr(conf->percpu, cpu);
83de75cc 1231 if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) {
91c00924
DW		 1232 ops_run_biofill(sh);
1233 overlap_clear++;
1234 }
1235
7b3a871e 1236 if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) {
ac6b53b6
DW		 1237 if (level < 6)
1238 tx = ops_run_compute5(sh, percpu);
1239 else {
1240 if (sh->ops.target2 < 0 || sh->ops.target < 0)
1241 tx = ops_run_compute6_1(sh, percpu);
1242 else
1243 tx = ops_run_compute6_2(sh, percpu);
1244 }
1245 /* terminate the chain if reconstruct is not set to be run */
1246 if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request))
7b3a871e
DW		 1247 async_tx_ack(tx);
1248 }
91c00924 1249
600aa109 1250 if (test_bit(STRIPE_OP_PREXOR, &ops_request))
d6f38f31 1251 tx = ops_run_prexor(sh, percpu, tx);
91c00924 1252
600aa109 1253 if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) {
d8ee0728 1254 tx = ops_run_biodrain(sh, tx);
91c00924
DW		 1255 overlap_clear++;
1256 }
1257
ac6b53b6
DW		 1258 if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) {
1259 if (level < 6)
1260 ops_run_reconstruct5(sh, percpu, tx);
1261 else
1262 ops_run_reconstruct6(sh, percpu, tx);
1263 }
91c00924 1264
ac6b53b6
DW		 1265 if (test_bit(STRIPE_OP_CHECK, &ops_request)) {
1266 if (sh->check_state == check_state_run)
1267 ops_run_check_p(sh, percpu);
1268 else if (sh->check_state == check_state_run_q)
1269 ops_run_check_pq(sh, percpu, 0);
1270 else if (sh->check_state == check_state_run_pq)
1271 ops_run_check_pq(sh, percpu, 1);
1272 else
1273 BUG();
1274 }
91c00924 1275
91c00924
DW		 1276 if (overlap_clear)
1277 for (i = disks; i--; ) {
1278 struct r5dev *dev = &sh->dev[i];
1279 if (test_and_clear_bit(R5_Overlap, &dev->flags))
1280 wake_up(&sh->raid_conf->wait_for_overlap);
1281 }
d6f38f31 1282 put_cpu();
91c00924
DW		 1283}
1284
417b8d4a
DW		 1285#ifdef CONFIG_MULTICORE_RAID456
1286static void async_run_ops(void *param, async_cookie_t cookie)
1287{
1288 struct stripe_head *sh = param;
1289 unsigned long ops_request = sh->ops.request;
1290
1291 clear_bit_unlock(STRIPE_OPS_REQ_PENDING, &sh->state);
1292 wake_up(&sh->ops.wait_for_ops);
1293
1294 __raid_run_ops(sh, ops_request);
1295 release_stripe(sh);
1296}
1297
1298static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
1299{
1300 /* since handle_stripe can be called outside of raid5d context
1301 * we need to ensure sh->ops.request is de-staged before another
1302 * request arrives
1303 */
1304 wait_event(sh->ops.wait_for_ops,
1305 !test_and_set_bit_lock(STRIPE_OPS_REQ_PENDING, &sh->state));
1306 sh->ops.request = ops_request;
1307
1308 atomic_inc(&sh->count);
1309 async_schedule(async_run_ops, sh);
1310}
1311#else
1312#define raid_run_ops __raid_run_ops
1313#endif
1314
3f294f4f 1315static int grow_one_stripe(raid5_conf_t *conf)
1da177e4
LT		 1316{
1317 struct stripe_head *sh;
6ce32846 1318 sh = kmem_cache_zalloc(conf->slab_cache, GFP_KERNEL);
3f294f4f
N		 1319 if (!sh)
1320 return 0;
6ce32846 1321
3f294f4f 1322 sh->raid_conf = conf;
417b8d4a
DW		 1323 #ifdef CONFIG_MULTICORE_RAID456
1324 init_waitqueue_head(&sh->ops.wait_for_ops);
1325 #endif
3f294f4f 1326
e4e11e38
N		 1327 if (grow_buffers(sh)) {
1328 shrink_buffers(sh);
3f294f4f
N		 1329 kmem_cache_free(conf->slab_cache, sh);
1330 return 0;
1331 }
1332 /* we just created an active stripe so... */
1333 atomic_set(&sh->count, 1);
1334 atomic_inc(&conf->active_stripes);
1335 INIT_LIST_HEAD(&sh->lru);
1336 release_stripe(sh);
1337 return 1;
1338}
1339
1340static int grow_stripes(raid5_conf_t *conf, int num)
1341{
e18b890b 1342 struct kmem_cache *sc;
5e5e3e78 1343 int devs = max(conf->raid_disks, conf->previous_raid_disks);
1da177e4 1344
f4be6b43
N		 1345 if (conf->mddev->gendisk)
1346 sprintf(conf->cache_name[0],
1347 "raid%d-%s", conf->level, mdname(conf->mddev));
1348 else
1349 sprintf(conf->cache_name[0],
1350 "raid%d-%p", conf->level, conf->mddev);
1351 sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]);
1352
ad01c9e3
N		 1353 conf->active_name = 0;
1354 sc = kmem_cache_create(conf->cache_name[conf->active_name],
1da177e4 1355 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
20c2df83 1356 0, 0, NULL);
1da177e4
LT		 1357 if (!sc)
1358 return 1;
1359 conf->slab_cache = sc;
ad01c9e3 1360 conf->pool_size = devs;
16a53ecc 1361 while (num--)
3f294f4f 1362 if (!grow_one_stripe(conf))
1da177e4 1363 return 1;
1da177e4
LT		 1364 return 0;
1365}
29269553 1366
d6f38f31
DW		 1367/**
1368 * scribble_len - return the required size of the scribble region
1369 * @num - total number of disks in the array
1370 *
1371 * The size must be enough to contain:
1372 * 1/ a struct page pointer for each device in the array +2
1373 * 2/ room to convert each entry in (1) to its corresponding dma
1374 * (dma_map_page()) or page (page_address()) address.
1375 *
1376 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
1377 * calculate over all devices (not just the data blocks), using zeros in place
1378 * of the P and Q blocks.
1379 */
1380static size_t scribble_len(int num)
1381{
1382 size_t len;
1383
1384 len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2);
1385
1386 return len;
1387}
1388
ad01c9e3
N		 1389static int resize_stripes(raid5_conf_t *conf, int newsize)
1390{
1391 /* Make all the stripes able to hold 'newsize' devices.
1392 * New slots in each stripe get 'page' set to a new page.
1393 *
1394 * This happens in stages:
1395 * 1/ create a new kmem_cache and allocate the required number of
1396 * stripe_heads.
1397 * 2/ gather all the old stripe_heads and tranfer the pages across
1398 * to the new stripe_heads. This will have the side effect of
1399 * freezing the array as once all stripe_heads have been collected,
1400 * no IO will be possible. Old stripe heads are freed once their
1401 * pages have been transferred over, and the old kmem_cache is
1402 * freed when all stripes are done.
1403 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
1404 * we simple return a failre status - no need to clean anything up.
1405 * 4/ allocate new pages for the new slots in the new stripe_heads.
1406 * If this fails, we don't bother trying the shrink the
1407 * stripe_heads down again, we just leave them as they are.
1408 * As each stripe_head is processed the new one is released into
1409 * active service.
1410 *
1411 * Once step2 is started, we cannot afford to wait for a write,
1412 * so we use GFP_NOIO allocations.
1413 */
1414 struct stripe_head *osh, *nsh;
1415 LIST_HEAD(newstripes);
1416 struct disk_info *ndisks;
d6f38f31 1417 unsigned long cpu;
b5470dc5 1418 int err;
e18b890b 1419 struct kmem_cache *sc;
ad01c9e3
N		 1420 int i;
1421
1422 if (newsize <= conf->pool_size)
1423 return 0; /* never bother to shrink */
1424
b5470dc5
DW		 1425 err = md_allow_write(conf->mddev);
1426 if (err)
1427 return err;
2a2275d6 1428
ad01c9e3
N		 1429 /* Step 1 */
1430 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
1431 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
20c2df83 1432 0, 0, NULL);
ad01c9e3
N		 1433 if (!sc)
1434 return -ENOMEM;
1435
1436 for (i = conf->max_nr_stripes; i; i--) {
6ce32846 1437 nsh = kmem_cache_zalloc(sc, GFP_KERNEL);
ad01c9e3
N		 1438 if (!nsh)
1439 break;
1440
ad01c9e3 1441 nsh->raid_conf = conf;
417b8d4a
DW		 1442 #ifdef CONFIG_MULTICORE_RAID456
1443 init_waitqueue_head(&nsh->ops.wait_for_ops);
1444 #endif
ad01c9e3
N		 1445
1446 list_add(&nsh->lru, &newstripes);
1447 }
1448 if (i) {
1449 /* didn't get enough, give up */
1450 while (!list_empty(&newstripes)) {
1451 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1452 list_del(&nsh->lru);
1453 kmem_cache_free(sc, nsh);
1454 }
1455 kmem_cache_destroy(sc);
1456 return -ENOMEM;
1457 }
1458 /* Step 2 - Must use GFP_NOIO now.
1459 * OK, we have enough stripes, start collecting inactive
1460 * stripes and copying them over
1461 */
1462 list_for_each_entry(nsh, &newstripes, lru) {
1463 spin_lock_irq(&conf->device_lock);
1464 wait_event_lock_irq(conf->wait_for_stripe,
1465 !list_empty(&conf->inactive_list),
1466 conf->device_lock,
482c0834 1467 );
ad01c9e3
N		 1468 osh = get_free_stripe(conf);
1469 spin_unlock_irq(&conf->device_lock);
1470 atomic_set(&nsh->count, 1);
1471 for(i=0; i<conf->pool_size; i++)
1472 nsh->dev[i].page = osh->dev[i].page;
1473 for( ; i<newsize; i++)
1474 nsh->dev[i].page = NULL;
1475 kmem_cache_free(conf->slab_cache, osh);
1476 }
1477 kmem_cache_destroy(conf->slab_cache);
1478
1479 /* Step 3.
1480 * At this point, we are holding all the stripes so the array
1481 * is completely stalled, so now is a good time to resize
d6f38f31 1482 * conf->disks and the scribble region
ad01c9e3
N		 1483 */
1484 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
1485 if (ndisks) {
1486 for (i=0; i<conf->raid_disks; i++)
1487 ndisks[i] = conf->disks[i];
1488 kfree(conf->disks);
1489 conf->disks = ndisks;
1490 } else
1491 err = -ENOMEM;
1492
d6f38f31
DW		 1493 get_online_cpus();
1494 conf->scribble_len = scribble_len(newsize);
1495 for_each_present_cpu(cpu) {
1496 struct raid5_percpu *percpu;
1497 void *scribble;
1498
1499 percpu = per_cpu_ptr(conf->percpu, cpu);
1500 scribble = kmalloc(conf->scribble_len, GFP_NOIO);
1501
1502 if (scribble) {
1503 kfree(percpu->scribble);
1504 percpu->scribble = scribble;
1505 } else {
1506 err = -ENOMEM;
1507 break;
1508 }
1509 }
1510 put_online_cpus();
1511
ad01c9e3
N		 1512 /* Step 4, return new stripes to service */
1513 while(!list_empty(&newstripes)) {
1514 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1515 list_del_init(&nsh->lru);
d6f38f31 1516
ad01c9e3
N		 1517 for (i=conf->raid_disks; i < newsize; i++)
1518 if (nsh->dev[i].page == NULL) {
1519 struct page *p = alloc_page(GFP_NOIO);
1520 nsh->dev[i].page = p;
1521 if (!p)
1522 err = -ENOMEM;
1523 }
1524 release_stripe(nsh);
1525 }
1526 /* critical section pass, GFP_NOIO no longer needed */
1527
1528 conf->slab_cache = sc;
1529 conf->active_name = 1-conf->active_name;
1530 conf->pool_size = newsize;
1531 return err;
1532}
1da177e4 1533
3f294f4f 1534static int drop_one_stripe(raid5_conf_t *conf)
1da177e4
LT		 1535{
1536 struct stripe_head *sh;
1537
3f294f4f
N		 1538 spin_lock_irq(&conf->device_lock);
1539 sh = get_free_stripe(conf);
1540 spin_unlock_irq(&conf->device_lock);
1541 if (!sh)
1542 return 0;
78bafebd 1543 BUG_ON(atomic_read(&sh->count));
e4e11e38 1544 shrink_buffers(sh);
3f294f4f
N		 1545 kmem_cache_free(conf->slab_cache, sh);
1546 atomic_dec(&conf->active_stripes);
1547 return 1;
1548}
1549
1550static void shrink_stripes(raid5_conf_t *conf)
1551{
1552 while (drop_one_stripe(conf))
1553 ;
1554
29fc7e3e
N		 1555 if (conf->slab_cache)
1556 kmem_cache_destroy(conf->slab_cache);
1da177e4
LT		 1557 conf->slab_cache = NULL;
1558}
1559
6712ecf8 1560static void raid5_end_read_request(struct bio * bi, int error)
1da177e4 1561{
99c0fb5f 1562 struct stripe_head *sh = bi->bi_private;
1da177e4 1563 raid5_conf_t *conf = sh->raid_conf;
7ecaa1e6 1564 int disks = sh->disks, i;
1da177e4 1565 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
d6950432
N		 1566 char b[BDEVNAME_SIZE];
1567 mdk_rdev_t *rdev;
1da177e4 1568
1da177e4
LT		 1569
1570 for (i=0 ; i<disks; i++)
1571 if (bi == &sh->dev[i].req)
1572 break;
1573
45b4233c
DW		 1574 pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n",
1575 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
1da177e4
LT		 1576 uptodate);
1577 if (i == disks) {
1578 BUG();
6712ecf8 1579 return;
1da177e4
LT		 1580 }
1581
1582 if (uptodate) {
1da177e4 1583 set_bit(R5_UPTODATE, &sh->dev[i].flags);
4e5314b5 1584 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
d6950432 1585 rdev = conf->disks[i].rdev;
0c55e022 1586 printk_rl(KERN_INFO "md/raid:%s: read error corrected"
6be9d494
BS		 1587 " (%lu sectors at %llu on %s)\n",
1588 mdname(conf->mddev), STRIPE_SECTORS,
1589 (unsigned long long)(sh->sector
1590 + rdev->data_offset),
1591 bdevname(rdev->bdev, b));
4e5314b5
N		 1592 clear_bit(R5_ReadError, &sh->dev[i].flags);
1593 clear_bit(R5_ReWrite, &sh->dev[i].flags);
1594 }
ba22dcbf
N		 1595 if (atomic_read(&conf->disks[i].rdev->read_errors))
1596 atomic_set(&conf->disks[i].rdev->read_errors, 0);
1da177e4 1597 } else {
d6950432 1598 const char *bdn = bdevname(conf->disks[i].rdev->bdev, b);
ba22dcbf 1599 int retry = 0;
d6950432
N		 1600 rdev = conf->disks[i].rdev;
1601
1da177e4 1602 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
d6950432 1603 atomic_inc(&rdev->read_errors);
7b0bb536 1604 if (conf->mddev->degraded >= conf->max_degraded)
6be9d494 1605 printk_rl(KERN_WARNING
0c55e022 1606 "md/raid:%s: read error not correctable "
6be9d494
BS		 1607 "(sector %llu on %s).\n",
1608 mdname(conf->mddev),
1609 (unsigned long long)(sh->sector
1610 + rdev->data_offset),
1611 bdn);
ba22dcbf 1612 else if (test_bit(R5_ReWrite, &sh->dev[i].flags))
4e5314b5 1613 /* Oh, no!!! */
6be9d494 1614 printk_rl(KERN_WARNING
0c55e022 1615 "md/raid:%s: read error NOT corrected!! "
6be9d494
BS		 1616 "(sector %llu on %s).\n",
1617 mdname(conf->mddev),
1618 (unsigned long long)(sh->sector
1619 + rdev->data_offset),
1620 bdn);
d6950432 1621 else if (atomic_read(&rdev->read_errors)
ba22dcbf 1622 > conf->max_nr_stripes)
14f8d26b 1623 printk(KERN_WARNING
0c55e022 1624 "md/raid:%s: Too many read errors, failing device %s.\n",
d6950432 1625 mdname(conf->mddev), bdn);
ba22dcbf
N		 1626 else
1627 retry = 1;
1628 if (retry)
1629 set_bit(R5_ReadError, &sh->dev[i].flags);
1630 else {
4e5314b5
N		 1631 clear_bit(R5_ReadError, &sh->dev[i].flags);
1632 clear_bit(R5_ReWrite, &sh->dev[i].flags);
d6950432 1633 md_error(conf->mddev, rdev);
ba22dcbf 1634 }
1da177e4
LT		 1635 }
1636 rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
1da177e4
LT		 1637 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1638 set_bit(STRIPE_HANDLE, &sh->state);
1639 release_stripe(sh);
1da177e4
LT		 1640}
1641
d710e138 1642static void raid5_end_write_request(struct bio *bi, int error)
1da177e4 1643{
99c0fb5f 1644 struct stripe_head *sh = bi->bi_private;
1da177e4 1645 raid5_conf_t *conf = sh->raid_conf;
7ecaa1e6 1646 int disks = sh->disks, i;
1da177e4
LT		 1647 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
1648
1da177e4
LT		 1649 for (i=0 ; i<disks; i++)
1650 if (bi == &sh->dev[i].req)
1651 break;
1652
45b4233c 1653 pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n",
1da177e4
LT		 1654 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
1655 uptodate);
1656 if (i == disks) {
1657 BUG();
6712ecf8 1658 return;
1da177e4
LT		 1659 }
1660
1da177e4
LT		 1661 if (!uptodate)
1662 md_error(conf->mddev, conf->disks[i].rdev);
1663
1664 rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
1665
1666 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1667 set_bit(STRIPE_HANDLE, &sh->state);
c04be0aa 1668 release_stripe(sh);
1da177e4
LT		 1669}
1670
1671
784052ec 1672static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous);
1da177e4 1673
784052ec 1674static void raid5_build_block(struct stripe_head *sh, int i, int previous)
1da177e4
LT		 1675{
1676 struct r5dev *dev = &sh->dev[i];
1677
1678 bio_init(&dev->req);
1679 dev->req.bi_io_vec = &dev->vec;
1680 dev->req.bi_vcnt++;
1681 dev->req.bi_max_vecs++;
1682 dev->vec.bv_page = dev->page;
1683 dev->vec.bv_len = STRIPE_SIZE;
1684 dev->vec.bv_offset = 0;
1685
1686 dev->req.bi_sector = sh->sector;
1687 dev->req.bi_private = sh;
1688
1689 dev->flags = 0;
784052ec 1690 dev->sector = compute_blocknr(sh, i, previous);
1da177e4
LT		 1691}
1692
1693static void error(mddev_t *mddev, mdk_rdev_t *rdev)
1694{
1695 char b[BDEVNAME_SIZE];
7b92813c 1696 raid5_conf_t *conf = mddev->private;
0c55e022 1697 pr_debug("raid456: error called\n");
1da177e4 1698
6f8d0c77
N		 1699 if (test_and_clear_bit(In_sync, &rdev->flags)) {
1700 unsigned long flags;
1701 spin_lock_irqsave(&conf->device_lock, flags);
1702 mddev->degraded++;
1703 spin_unlock_irqrestore(&conf->device_lock, flags);
1704 /*
1705 * if recovery was running, make sure it aborts.
1706 */
1707 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4 1708 }
6f8d0c77
N		 1709 set_bit(Faulty, &rdev->flags);
1710 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1711 printk(KERN_ALERT
1712 "md/raid:%s: Disk failure on %s, disabling device.\n"
1713 "md/raid:%s: Operation continuing on %d devices.\n",
1714 mdname(mddev),
1715 bdevname(rdev->bdev, b),
1716 mdname(mddev),
1717 conf->raid_disks - mddev->degraded);
16a53ecc 1718}
1da177e4
LT		 1719
1720/*
1721 * Input: a 'big' sector number,
1722 * Output: index of the data and parity disk, and the sector # in them.
1723 */
112bf897 1724static sector_t raid5_compute_sector(raid5_conf_t *conf, sector_t r_sector,
911d4ee8
N		 1725 int previous, int *dd_idx,
1726 struct stripe_head *sh)
1da177e4 1727{
6e3b96ed 1728 sector_t stripe, stripe2;
35f2a591 1729 sector_t chunk_number;
1da177e4 1730 unsigned int chunk_offset;
911d4ee8 1731 int pd_idx, qd_idx;
67cc2b81 1732 int ddf_layout = 0;
1da177e4 1733 sector_t new_sector;
e183eaed
N		 1734 int algorithm = previous ? conf->prev_algo
1735 : conf->algorithm;
09c9e5fa
AN		 1736 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
1737 : conf->chunk_sectors;
112bf897
N		 1738 int raid_disks = previous ? conf->previous_raid_disks
1739 : conf->raid_disks;
1740 int data_disks = raid_disks - conf->max_degraded;
1da177e4
LT		 1741
1742 /* First compute the information on this sector */
1743
1744 /*
1745 * Compute the chunk number and the sector offset inside the chunk
1746 */
1747 chunk_offset = sector_div(r_sector, sectors_per_chunk);
1748 chunk_number = r_sector;
1da177e4
LT		 1749
1750 /*
1751 * Compute the stripe number
1752 */
35f2a591
N		 1753 stripe = chunk_number;
1754 *dd_idx = sector_div(stripe, data_disks);
6e3b96ed 1755 stripe2 = stripe;
1da177e4
LT		 1756 /*
1757 * Select the parity disk based on the user selected algorithm.
1758 */
911d4ee8 1759 pd_idx = qd_idx = ~0;
16a53ecc
N		 1760 switch(conf->level) {
1761 case 4:
911d4ee8 1762 pd_idx = data_disks;
16a53ecc
N		 1763 break;
1764 case 5:
e183eaed 1765 switch (algorithm) {
1da177e4 1766 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 1767 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 1768 if (*dd_idx >= pd_idx)
1da177e4
LT		 1769 (*dd_idx)++;
1770 break;
1771 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 1772 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 1773 if (*dd_idx >= pd_idx)
1da177e4
LT		 1774 (*dd_idx)++;
1775 break;
1776 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 1777 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 1778 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4
LT		 1779 break;
1780 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 1781 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 1782 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4 1783 break;
99c0fb5f
N		 1784 case ALGORITHM_PARITY_0:
1785 pd_idx = 0;
1786 (*dd_idx)++;
1787 break;
1788 case ALGORITHM_PARITY_N:
1789 pd_idx = data_disks;
1790 break;
1da177e4 1791 default:
99c0fb5f 1792 BUG();
16a53ecc
N		 1793 }
1794 break;
1795 case 6:
1796
e183eaed 1797 switch (algorithm) {
16a53ecc 1798 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 1799 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 1800 qd_idx = pd_idx + 1;
1801 if (pd_idx == raid_disks-1) {
99c0fb5f 1802 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 1803 qd_idx = 0;
1804 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 1805 (*dd_idx) += 2; /* D D P Q D */
1806 break;
1807 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 1808 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 1809 qd_idx = pd_idx + 1;
1810 if (pd_idx == raid_disks-1) {
99c0fb5f 1811 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 1812 qd_idx = 0;
1813 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 1814 (*dd_idx) += 2; /* D D P Q D */
1815 break;
1816 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 1817 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 1818 qd_idx = (pd_idx + 1) % raid_disks;
1819 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc
N		 1820 break;
1821 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 1822 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 1823 qd_idx = (pd_idx + 1) % raid_disks;
1824 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc 1825 break;
99c0fb5f
N		 1826
1827 case ALGORITHM_PARITY_0:
1828 pd_idx = 0;
1829 qd_idx = 1;
1830 (*dd_idx) += 2;
1831 break;
1832 case ALGORITHM_PARITY_N:
1833 pd_idx = data_disks;
1834 qd_idx = data_disks + 1;
1835 break;
1836
1837 case ALGORITHM_ROTATING_ZERO_RESTART:
1838 /* Exactly the same as RIGHT_ASYMMETRIC, but or
1839 * of blocks for computing Q is different.
1840 */
6e3b96ed 1841 pd_idx = sector_div(stripe2, raid_disks);
99c0fb5f
N		 1842 qd_idx = pd_idx + 1;
1843 if (pd_idx == raid_disks-1) {
1844 (*dd_idx)++; /* Q D D D P */
1845 qd_idx = 0;
1846 } else if (*dd_idx >= pd_idx)
1847 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 1848 ddf_layout = 1;
99c0fb5f
N		 1849 break;
1850
1851 case ALGORITHM_ROTATING_N_RESTART:
1852 /* Same a left_asymmetric, by first stripe is
1853 * D D D P Q rather than
1854 * Q D D D P
1855 */
6e3b96ed
N		 1856 stripe2 += 1;
1857 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 1858 qd_idx = pd_idx + 1;
1859 if (pd_idx == raid_disks-1) {
1860 (*dd_idx)++; /* Q D D D P */
1861 qd_idx = 0;
1862 } else if (*dd_idx >= pd_idx)
1863 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 1864 ddf_layout = 1;
99c0fb5f
N		 1865 break;
1866
1867 case ALGORITHM_ROTATING_N_CONTINUE:
1868 /* Same as left_symmetric but Q is before P */
6e3b96ed 1869 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 1870 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
1871 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
67cc2b81 1872 ddf_layout = 1;
99c0fb5f
N		 1873 break;
1874
1875 case ALGORITHM_LEFT_ASYMMETRIC_6:
1876 /* RAID5 left_asymmetric, with Q on last device */
6e3b96ed 1877 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 1878 if (*dd_idx >= pd_idx)
1879 (*dd_idx)++;
1880 qd_idx = raid_disks - 1;
1881 break;
1882
1883 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6e3b96ed 1884 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 1885 if (*dd_idx >= pd_idx)
1886 (*dd_idx)++;
1887 qd_idx = raid_disks - 1;
1888 break;
1889
1890 case ALGORITHM_LEFT_SYMMETRIC_6:
6e3b96ed 1891 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 1892 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
1893 qd_idx = raid_disks - 1;
1894 break;
1895
1896 case ALGORITHM_RIGHT_SYMMETRIC_6:
6e3b96ed 1897 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 1898 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
1899 qd_idx = raid_disks - 1;
1900 break;
1901
1902 case ALGORITHM_PARITY_0_6:
1903 pd_idx = 0;
1904 (*dd_idx)++;
1905 qd_idx = raid_disks - 1;
1906 break;
1907
16a53ecc 1908 default:
99c0fb5f 1909 BUG();
16a53ecc
N		 1910 }
1911 break;
1da177e4
LT		 1912 }
1913
911d4ee8
N		 1914 if (sh) {
1915 sh->pd_idx = pd_idx;
1916 sh->qd_idx = qd_idx;
67cc2b81 1917 sh->ddf_layout = ddf_layout;
911d4ee8 1918 }
1da177e4
LT		 1919 /*
1920 * Finally, compute the new sector number
1921 */
1922 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
1923 return new_sector;
1924}
1925
1926
784052ec 1927static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous)
1da177e4
LT		 1928{
1929 raid5_conf_t *conf = sh->raid_conf;
b875e531
N		 1930 int raid_disks = sh->disks;
1931 int data_disks = raid_disks - conf->max_degraded;
1da177e4 1932 sector_t new_sector = sh->sector, check;
09c9e5fa
AN		 1933 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
1934 : conf->chunk_sectors;
e183eaed
N		 1935 int algorithm = previous ? conf->prev_algo
1936 : conf->algorithm;
1da177e4
LT		 1937 sector_t stripe;
1938 int chunk_offset;
35f2a591
N		 1939 sector_t chunk_number;
1940 int dummy1, dd_idx = i;
1da177e4 1941 sector_t r_sector;
911d4ee8 1942 struct stripe_head sh2;
1da177e4 1943
16a53ecc 1944
1da177e4
LT		 1945 chunk_offset = sector_div(new_sector, sectors_per_chunk);
1946 stripe = new_sector;
1da177e4 1947
16a53ecc
N		 1948 if (i == sh->pd_idx)
1949 return 0;
1950 switch(conf->level) {
1951 case 4: break;
1952 case 5:
e183eaed 1953 switch (algorithm) {
1da177e4
LT		 1954 case ALGORITHM_LEFT_ASYMMETRIC:
1955 case ALGORITHM_RIGHT_ASYMMETRIC:
1956 if (i > sh->pd_idx)
1957 i--;
1958 break;
1959 case ALGORITHM_LEFT_SYMMETRIC:
1960 case ALGORITHM_RIGHT_SYMMETRIC:
1961 if (i < sh->pd_idx)
1962 i += raid_disks;
1963 i -= (sh->pd_idx + 1);
1964 break;
99c0fb5f
N		 1965 case ALGORITHM_PARITY_0:
1966 i -= 1;
1967 break;
1968 case ALGORITHM_PARITY_N:
1969 break;
1da177e4 1970 default:
99c0fb5f 1971 BUG();
16a53ecc
N		 1972 }
1973 break;
1974 case 6:
d0dabf7e 1975 if (i == sh->qd_idx)
16a53ecc 1976 return 0; /* It is the Q disk */
e183eaed 1977 switch (algorithm) {
16a53ecc
N		 1978 case ALGORITHM_LEFT_ASYMMETRIC:
1979 case ALGORITHM_RIGHT_ASYMMETRIC:
99c0fb5f
N		 1980 case ALGORITHM_ROTATING_ZERO_RESTART:
1981 case ALGORITHM_ROTATING_N_RESTART:
1982 if (sh->pd_idx == raid_disks-1)
1983 i--; /* Q D D D P */
16a53ecc
N		 1984 else if (i > sh->pd_idx)
1985 i -= 2; /* D D P Q D */
1986 break;
1987 case ALGORITHM_LEFT_SYMMETRIC:
1988 case ALGORITHM_RIGHT_SYMMETRIC:
1989 if (sh->pd_idx == raid_disks-1)
1990 i--; /* Q D D D P */
1991 else {
1992 /* D D P Q D */
1993 if (i < sh->pd_idx)
1994 i += raid_disks;
1995 i -= (sh->pd_idx + 2);
1996 }
1997 break;
99c0fb5f
N		 1998 case ALGORITHM_PARITY_0:
1999 i -= 2;
2000 break;
2001 case ALGORITHM_PARITY_N:
2002 break;
2003 case ALGORITHM_ROTATING_N_CONTINUE:
e4424fee 2004 /* Like left_symmetric, but P is before Q */
99c0fb5f
N		 2005 if (sh->pd_idx == 0)
2006 i--; /* P D D D Q */
e4424fee
N		 2007 else {
2008 /* D D Q P D */
2009 if (i < sh->pd_idx)
2010 i += raid_disks;
2011 i -= (sh->pd_idx + 1);
2012 }
99c0fb5f
N		 2013 break;
2014 case ALGORITHM_LEFT_ASYMMETRIC_6:
2015 case ALGORITHM_RIGHT_ASYMMETRIC_6:
2016 if (i > sh->pd_idx)
2017 i--;
2018 break;
2019 case ALGORITHM_LEFT_SYMMETRIC_6:
2020 case ALGORITHM_RIGHT_SYMMETRIC_6:
2021 if (i < sh->pd_idx)
2022 i += data_disks + 1;
2023 i -= (sh->pd_idx + 1);
2024 break;
2025 case ALGORITHM_PARITY_0_6:
2026 i -= 1;
2027 break;
16a53ecc 2028 default:
99c0fb5f 2029 BUG();
16a53ecc
N		 2030 }
2031 break;
1da177e4
LT		 2032 }
2033
2034 chunk_number = stripe * data_disks + i;
35f2a591 2035 r_sector = chunk_number * sectors_per_chunk + chunk_offset;
1da177e4 2036
112bf897 2037 check = raid5_compute_sector(conf, r_sector,
784052ec 2038 previous, &dummy1, &sh2);
911d4ee8
N		 2039 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
2040 || sh2.qd_idx != sh->qd_idx) {
0c55e022
N		 2041 printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n",
2042 mdname(conf->mddev));
1da177e4
LT		 2043 return 0;
2044 }
2045 return r_sector;
2046}
2047
2048
600aa109 2049static void
c0f7bddb 2050schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
600aa109 2051 int rcw, int expand)
e33129d8
DW		 2052{
2053 int i, pd_idx = sh->pd_idx, disks = sh->disks;
c0f7bddb
YT		 2054 raid5_conf_t *conf = sh->raid_conf;
2055 int level = conf->level;
e33129d8
DW		 2056
2057 if (rcw) {
2058 /* if we are not expanding this is a proper write request, and
2059 * there will be bios with new data to be drained into the
2060 * stripe cache
2061 */
2062 if (!expand) {
600aa109
DW		 2063 sh->reconstruct_state = reconstruct_state_drain_run;
2064 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2065 } else
2066 sh->reconstruct_state = reconstruct_state_run;
16a53ecc 2067
ac6b53b6 2068 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
e33129d8
DW		 2069
2070 for (i = disks; i--; ) {
2071 struct r5dev *dev = &sh->dev[i];
2072
2073 if (dev->towrite) {
2074 set_bit(R5_LOCKED, &dev->flags);
d8ee0728 2075 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2076 if (!expand)
2077 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2078 s->locked++;
e33129d8
DW		 2079 }
2080 }
c0f7bddb 2081 if (s->locked + conf->max_degraded == disks)
8b3e6cdc 2082 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
c0f7bddb 2083 atomic_inc(&conf->pending_full_writes);
e33129d8 2084 } else {
c0f7bddb 2085 BUG_ON(level == 6);
e33129d8
DW		 2086 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
2087 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
2088
d8ee0728 2089 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
600aa109
DW		 2090 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
2091 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
ac6b53b6 2092 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
e33129d8
DW		 2093
2094 for (i = disks; i--; ) {
2095 struct r5dev *dev = &sh->dev[i];
2096 if (i == pd_idx)
2097 continue;
2098
e33129d8
DW		 2099 if (dev->towrite &&
2100 (test_bit(R5_UPTODATE, &dev->flags) ||
d8ee0728
DW		 2101 test_bit(R5_Wantcompute, &dev->flags))) {
2102 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2103 set_bit(R5_LOCKED, &dev->flags);
2104 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2105 s->locked++;
e33129d8
DW		 2106 }
2107 }
2108 }
2109
c0f7bddb 2110 /* keep the parity disk(s) locked while asynchronous operations
e33129d8
DW		 2111 * are in flight
2112 */
2113 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
2114 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
600aa109 2115 s->locked++;
e33129d8 2116
c0f7bddb
YT		 2117 if (level == 6) {
2118 int qd_idx = sh->qd_idx;
2119 struct r5dev *dev = &sh->dev[qd_idx];
2120
2121 set_bit(R5_LOCKED, &dev->flags);
2122 clear_bit(R5_UPTODATE, &dev->flags);
2123 s->locked++;
2124 }
2125
600aa109 2126 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
e46b272b 2127 __func__, (unsigned long long)sh->sector,
600aa109 2128 s->locked, s->ops_request);
e33129d8 2129}
16a53ecc 2130
1da177e4
LT		 2131/*
2132 * Each stripe/dev can have one or more bion attached.
16a53ecc 2133 * toread/towrite point to the first in a chain.
1da177e4
LT		 2134 * The bi_next chain must be in order.
2135 */
2136static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite)
2137{
2138 struct bio **bip;
2139 raid5_conf_t *conf = sh->raid_conf;
72626685 2140 int firstwrite=0;
1da177e4 2141
cbe47ec5 2142 pr_debug("adding bi b#%llu to stripe s#%llu\n",
1da177e4
LT		 2143 (unsigned long long)bi->bi_sector,
2144 (unsigned long long)sh->sector);
2145
2146
1da177e4 2147 spin_lock_irq(&conf->device_lock);
72626685 2148 if (forwrite) {
1da177e4 2149 bip = &sh->dev[dd_idx].towrite;
72626685
N		 2150 if (*bip == NULL && sh->dev[dd_idx].written == NULL)
2151 firstwrite = 1;
2152 } else
1da177e4
LT		 2153 bip = &sh->dev[dd_idx].toread;
2154 while (*bip && (*bip)->bi_sector < bi->bi_sector) {
2155 if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector)
2156 goto overlap;
2157 bip = & (*bip)->bi_next;
2158 }
2159 if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9))
2160 goto overlap;
2161
78bafebd 2162 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
1da177e4
LT		 2163 if (*bip)
2164 bi->bi_next = *bip;
2165 *bip = bi;
960e739d 2166 bi->bi_phys_segments++;
72626685 2167
1da177e4
LT		 2168 if (forwrite) {
2169 /* check if page is covered */
2170 sector_t sector = sh->dev[dd_idx].sector;
2171 for (bi=sh->dev[dd_idx].towrite;
2172 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
2173 bi && bi->bi_sector <= sector;
2174 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
2175 if (bi->bi_sector + (bi->bi_size>>9) >= sector)
2176 sector = bi->bi_sector + (bi->bi_size>>9);
2177 }
2178 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
2179 set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags);
2180 }
cbe47ec5 2181 spin_unlock_irq(&conf->device_lock);
cbe47ec5
N		 2182
2183 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
2184 (unsigned long long)(*bip)->bi_sector,
2185 (unsigned long long)sh->sector, dd_idx);
2186
2187 if (conf->mddev->bitmap && firstwrite) {
2188 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
2189 STRIPE_SECTORS, 0);
2190 sh->bm_seq = conf->seq_flush+1;
2191 set_bit(STRIPE_BIT_DELAY, &sh->state);
2192 }
1da177e4
LT		 2193 return 1;
2194
2195 overlap:
2196 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
2197 spin_unlock_irq(&conf->device_lock);
1da177e4
LT		 2198 return 0;
2199}
2200
29269553
N		 2201static void end_reshape(raid5_conf_t *conf);
2202
911d4ee8
N		 2203static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous,
2204 struct stripe_head *sh)
ccfcc3c1 2205{
784052ec 2206 int sectors_per_chunk =
09c9e5fa 2207 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
911d4ee8 2208 int dd_idx;
2d2063ce 2209 int chunk_offset = sector_div(stripe, sectors_per_chunk);
112bf897 2210 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
2d2063ce 2211
112bf897
N		 2212 raid5_compute_sector(conf,
2213 stripe * (disks - conf->max_degraded)
b875e531 2214 *sectors_per_chunk + chunk_offset,
112bf897 2215 previous,
911d4ee8 2216 &dd_idx, sh);
ccfcc3c1
N		 2217}
2218
a4456856 2219static void
1fe797e6 2220handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh,
a4456856
DW		 2221 struct stripe_head_state *s, int disks,
2222 struct bio **return_bi)
2223{
2224 int i;
2225 for (i = disks; i--; ) {
2226 struct bio *bi;
2227 int bitmap_end = 0;
2228
2229 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
2230 mdk_rdev_t *rdev;
2231 rcu_read_lock();
2232 rdev = rcu_dereference(conf->disks[i].rdev);
2233 if (rdev && test_bit(In_sync, &rdev->flags))
2234 /* multiple read failures in one stripe */
2235 md_error(conf->mddev, rdev);
2236 rcu_read_unlock();
2237 }
2238 spin_lock_irq(&conf->device_lock);
2239 /* fail all writes first */
2240 bi = sh->dev[i].towrite;
2241 sh->dev[i].towrite = NULL;
2242 if (bi) {
2243 s->to_write--;
2244 bitmap_end = 1;
2245 }
2246
2247 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2248 wake_up(&conf->wait_for_overlap);
2249
2250 while (bi && bi->bi_sector <
2251 sh->dev[i].sector + STRIPE_SECTORS) {
2252 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
2253 clear_bit(BIO_UPTODATE, &bi->bi_flags);
960e739d 2254 if (!raid5_dec_bi_phys_segments(bi)) {
a4456856
DW		 2255 md_write_end(conf->mddev);
2256 bi->bi_next = *return_bi;
2257 *return_bi = bi;
2258 }
2259 bi = nextbi;
2260 }
2261 /* and fail all 'written' */
2262 bi = sh->dev[i].written;
2263 sh->dev[i].written = NULL;
2264 if (bi) bitmap_end = 1;
2265 while (bi && bi->bi_sector <
2266 sh->dev[i].sector + STRIPE_SECTORS) {
2267 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
2268 clear_bit(BIO_UPTODATE, &bi->bi_flags);
960e739d 2269 if (!raid5_dec_bi_phys_segments(bi)) {
a4456856
DW		 2270 md_write_end(conf->mddev);
2271 bi->bi_next = *return_bi;
2272 *return_bi = bi;
2273 }
2274 bi = bi2;
2275 }
2276
b5e98d65
DW		 2277 /* fail any reads if this device is non-operational and
2278 * the data has not reached the cache yet.
2279 */
2280 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
2281 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
2282 test_bit(R5_ReadError, &sh->dev[i].flags))) {
a4456856
DW		 2283 bi = sh->dev[i].toread;
2284 sh->dev[i].toread = NULL;
2285 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2286 wake_up(&conf->wait_for_overlap);
2287 if (bi) s->to_read--;
2288 while (bi && bi->bi_sector <
2289 sh->dev[i].sector + STRIPE_SECTORS) {
2290 struct bio *nextbi =
2291 r5_next_bio(bi, sh->dev[i].sector);
2292 clear_bit(BIO_UPTODATE, &bi->bi_flags);
960e739d 2293 if (!raid5_dec_bi_phys_segments(bi)) {
a4456856
DW		 2294 bi->bi_next = *return_bi;
2295 *return_bi = bi;
2296 }
2297 bi = nextbi;
2298 }
2299 }
2300 spin_unlock_irq(&conf->device_lock);
2301 if (bitmap_end)
2302 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
2303 STRIPE_SECTORS, 0, 0);
2304 }
2305
8b3e6cdc
DW		 2306 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2307 if (atomic_dec_and_test(&conf->pending_full_writes))
2308 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 2309}
2310
1fe797e6
DW		 2311/* fetch_block5 - checks the given member device to see if its data needs
2312 * to be read or computed to satisfy a request.
2313 *
2314 * Returns 1 when no more member devices need to be checked, otherwise returns
2315 * 0 to tell the loop in handle_stripe_fill5 to continue
f38e1219 2316 */
1fe797e6
DW		 2317static int fetch_block5(struct stripe_head *sh, struct stripe_head_state *s,
2318 int disk_idx, int disks)
f38e1219
DW		 2319{
2320 struct r5dev *dev = &sh->dev[disk_idx];
f2b3b44d 2321 struct r5dev *failed_dev = &sh->dev[s->failed_num[0]];
f38e1219 2322
f38e1219
DW		 2323 /* is the data in this block needed, and can we get it? */
2324 if (!test_bit(R5_LOCKED, &dev->flags) &&
1fe797e6
DW		 2325 !test_bit(R5_UPTODATE, &dev->flags) &&
2326 (dev->toread ||
2327 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
2328 s->syncing || s->expanding ||
2329 (s->failed &&
2330 (failed_dev->toread ||
2331 (failed_dev->towrite &&
2332 !test_bit(R5_OVERWRITE, &failed_dev->flags)))))) {
976ea8d4
DW		 2333 /* We would like to get this block, possibly by computing it,
2334 * otherwise read it if the backing disk is insync
f38e1219
DW		 2335 */
2336 if ((s->uptodate == disks - 1) &&
f2b3b44d 2337 (s->failed && disk_idx == s->failed_num[0])) {
976ea8d4
DW		 2338 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2339 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
f38e1219
DW		 2340 set_bit(R5_Wantcompute, &dev->flags);
2341 sh->ops.target = disk_idx;
ac6b53b6 2342 sh->ops.target2 = -1;
f38e1219 2343 s->req_compute = 1;
f38e1219 2344 /* Careful: from this point on 'uptodate' is in the eye
ac6b53b6 2345 * of raid_run_ops which services 'compute' operations
f38e1219
DW		 2346 * before writes. R5_Wantcompute flags a block that will
2347 * be R5_UPTODATE by the time it is needed for a
2348 * subsequent operation.
2349 */
2350 s->uptodate++;
1fe797e6 2351 return 1; /* uptodate + compute == disks */
7a1fc53c 2352 } else if (test_bit(R5_Insync, &dev->flags)) {
f38e1219
DW		 2353 set_bit(R5_LOCKED, &dev->flags);
2354 set_bit(R5_Wantread, &dev->flags);
f38e1219
DW		 2355 s->locked++;
2356 pr_debug("Reading block %d (sync=%d)\n", disk_idx,
2357 s->syncing);
2358 }
2359 }
2360
1fe797e6 2361 return 0;
f38e1219
DW		 2362}
2363
1fe797e6
DW		 2364/**
2365 * handle_stripe_fill5 - read or compute data to satisfy pending requests.
2366 */
2367static void handle_stripe_fill5(struct stripe_head *sh,
a4456856
DW		 2368 struct stripe_head_state *s, int disks)
2369{
2370 int i;
f38e1219 2371
f38e1219
DW		 2372 /* look for blocks to read/compute, skip this if a compute
2373 * is already in flight, or if the stripe contents are in the
2374 * midst of changing due to a write
2375 */
976ea8d4 2376 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
1fe797e6 2377 !sh->reconstruct_state)
f38e1219 2378 for (i = disks; i--; )
1fe797e6 2379 if (fetch_block5(sh, s, i, disks))
f38e1219 2380 break;
a4456856
DW		 2381 set_bit(STRIPE_HANDLE, &sh->state);
2382}
2383
5599becc
YT		 2384/* fetch_block6 - checks the given member device to see if its data needs
2385 * to be read or computed to satisfy a request.
2386 *
2387 * Returns 1 when no more member devices need to be checked, otherwise returns
2388 * 0 to tell the loop in handle_stripe_fill6 to continue
2389 */
2390static int fetch_block6(struct stripe_head *sh, struct stripe_head_state *s,
f2b3b44d 2391 int disk_idx, int disks)
a4456856 2392{
5599becc 2393 struct r5dev *dev = &sh->dev[disk_idx];
f2b3b44d
N		 2394 struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]],
2395 &sh->dev[s->failed_num[1]] };
5599becc
YT		 2396
2397 if (!test_bit(R5_LOCKED, &dev->flags) &&
2398 !test_bit(R5_UPTODATE, &dev->flags) &&
2399 (dev->toread ||
2400 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
2401 s->syncing || s->expanding ||
2402 (s->failed >= 1 &&
2403 (fdev[0]->toread || s->to_write)) ||
2404 (s->failed >= 2 &&
2405 (fdev[1]->toread || s->to_write)))) {
2406 /* we would like to get this block, possibly by computing it,
2407 * otherwise read it if the backing disk is insync
2408 */
2409 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
2410 BUG_ON(test_bit(R5_Wantread, &dev->flags));
2411 if ((s->uptodate == disks - 1) &&
f2b3b44d
N		 2412 (s->failed && (disk_idx == s->failed_num[0] ||
2413 disk_idx == s->failed_num[1]))) {
5599becc
YT		 2414 /* have disk failed, and we're requested to fetch it;
2415 * do compute it
a4456856 2416 */
5599becc
YT		 2417 pr_debug("Computing stripe %llu block %d\n",
2418 (unsigned long long)sh->sector, disk_idx);
2419 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2420 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2421 set_bit(R5_Wantcompute, &dev->flags);
2422 sh->ops.target = disk_idx;
2423 sh->ops.target2 = -1; /* no 2nd target */
2424 s->req_compute = 1;
2425 s->uptodate++;
2426 return 1;
2427 } else if (s->uptodate == disks-2 && s->failed >= 2) {
2428 /* Computing 2-failure is *very* expensive; only
2429 * do it if failed >= 2
2430 */
2431 int other;
2432 for (other = disks; other--; ) {
2433 if (other == disk_idx)
2434 continue;
2435 if (!test_bit(R5_UPTODATE,
2436 &sh->dev[other].flags))
2437 break;
a4456856 2438 }
5599becc
YT		 2439 BUG_ON(other < 0);
2440 pr_debug("Computing stripe %llu blocks %d,%d\n",
2441 (unsigned long long)sh->sector,
2442 disk_idx, other);
2443 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2444 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2445 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
2446 set_bit(R5_Wantcompute, &sh->dev[other].flags);
2447 sh->ops.target = disk_idx;
2448 sh->ops.target2 = other;
2449 s->uptodate += 2;
2450 s->req_compute = 1;
2451 return 1;
2452 } else if (test_bit(R5_Insync, &dev->flags)) {
2453 set_bit(R5_LOCKED, &dev->flags);
2454 set_bit(R5_Wantread, &dev->flags);
2455 s->locked++;
2456 pr_debug("Reading block %d (sync=%d)\n",
2457 disk_idx, s->syncing);
a4456856
DW		 2458 }
2459 }
5599becc
YT		 2460
2461 return 0;
2462}
2463
2464/**
2465 * handle_stripe_fill6 - read or compute data to satisfy pending requests.
2466 */
2467static void handle_stripe_fill6(struct stripe_head *sh,
f2b3b44d 2468 struct stripe_head_state *s,
5599becc
YT		 2469 int disks)
2470{
2471 int i;
2472
2473 /* look for blocks to read/compute, skip this if a compute
2474 * is already in flight, or if the stripe contents are in the
2475 * midst of changing due to a write
2476 */
2477 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
2478 !sh->reconstruct_state)
2479 for (i = disks; i--; )
f2b3b44d 2480 if (fetch_block6(sh, s, i, disks))
5599becc 2481 break;
a4456856
DW		 2482 set_bit(STRIPE_HANDLE, &sh->state);
2483}
2484
2485
1fe797e6 2486/* handle_stripe_clean_event
a4456856
DW		 2487 * any written block on an uptodate or failed drive can be returned.
2488 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
2489 * never LOCKED, so we don't need to test 'failed' directly.
2490 */
1fe797e6 2491static void handle_stripe_clean_event(raid5_conf_t *conf,
a4456856
DW		 2492 struct stripe_head *sh, int disks, struct bio **return_bi)
2493{
2494 int i;
2495 struct r5dev *dev;
2496
2497 for (i = disks; i--; )
2498 if (sh->dev[i].written) {
2499 dev = &sh->dev[i];
2500 if (!test_bit(R5_LOCKED, &dev->flags) &&
2501 test_bit(R5_UPTODATE, &dev->flags)) {
2502 /* We can return any write requests */
2503 struct bio *wbi, *wbi2;
2504 int bitmap_end = 0;
45b4233c 2505 pr_debug("Return write for disc %d\n", i);
a4456856
DW		 2506 spin_lock_irq(&conf->device_lock);
2507 wbi = dev->written;
2508 dev->written = NULL;
2509 while (wbi && wbi->bi_sector <
2510 dev->sector + STRIPE_SECTORS) {
2511 wbi2 = r5_next_bio(wbi, dev->sector);
960e739d 2512 if (!raid5_dec_bi_phys_segments(wbi)) {
a4456856
DW		 2513 md_write_end(conf->mddev);
2514 wbi->bi_next = *return_bi;
2515 *return_bi = wbi;
2516 }
2517 wbi = wbi2;
2518 }
2519 if (dev->towrite == NULL)
2520 bitmap_end = 1;
2521 spin_unlock_irq(&conf->device_lock);
2522 if (bitmap_end)
2523 bitmap_endwrite(conf->mddev->bitmap,
2524 sh->sector,
2525 STRIPE_SECTORS,
2526 !test_bit(STRIPE_DEGRADED, &sh->state),
2527 0);
2528 }
2529 }
8b3e6cdc
DW		 2530
2531 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2532 if (atomic_dec_and_test(&conf->pending_full_writes))
2533 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 2534}
2535
1fe797e6 2536static void handle_stripe_dirtying5(raid5_conf_t *conf,
a4456856
DW		 2537 struct stripe_head *sh, struct stripe_head_state *s, int disks)
2538{
2539 int rmw = 0, rcw = 0, i;
2540 for (i = disks; i--; ) {
2541 /* would I have to read this buffer for read_modify_write */
2542 struct r5dev *dev = &sh->dev[i];
2543 if ((dev->towrite || i == sh->pd_idx) &&
2544 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 2545 !(test_bit(R5_UPTODATE, &dev->flags) ||
2546 test_bit(R5_Wantcompute, &dev->flags))) {
a4456856
DW		 2547 if (test_bit(R5_Insync, &dev->flags))
2548 rmw++;
2549 else
2550 rmw += 2*disks; /* cannot read it */
2551 }
2552 /* Would I have to read this buffer for reconstruct_write */
2553 if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx &&
2554 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 2555 !(test_bit(R5_UPTODATE, &dev->flags) ||
2556 test_bit(R5_Wantcompute, &dev->flags))) {
2557 if (test_bit(R5_Insync, &dev->flags)) rcw++;
a4456856
DW		 2558 else
2559 rcw += 2*disks;
2560 }
2561 }
45b4233c 2562 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
a4456856
DW		 2563 (unsigned long long)sh->sector, rmw, rcw);
2564 set_bit(STRIPE_HANDLE, &sh->state);
2565 if (rmw < rcw && rmw > 0)
2566 /* prefer read-modify-write, but need to get some data */
2567 for (i = disks; i--; ) {
2568 struct r5dev *dev = &sh->dev[i];
2569 if ((dev->towrite || i == sh->pd_idx) &&
2570 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 2571 !(test_bit(R5_UPTODATE, &dev->flags) ||
2572 test_bit(R5_Wantcompute, &dev->flags)) &&
a4456856
DW		 2573 test_bit(R5_Insync, &dev->flags)) {
2574 if (
2575 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
45b4233c 2576 pr_debug("Read_old block "
a4456856
DW		 2577 "%d for r-m-w\n", i);
2578 set_bit(R5_LOCKED, &dev->flags);
2579 set_bit(R5_Wantread, &dev->flags);
2580 s->locked++;
2581 } else {
2582 set_bit(STRIPE_DELAYED, &sh->state);
2583 set_bit(STRIPE_HANDLE, &sh->state);
2584 }
2585 }
2586 }
2587 if (rcw <= rmw && rcw > 0)
2588 /* want reconstruct write, but need to get some data */
2589 for (i = disks; i--; ) {
2590 struct r5dev *dev = &sh->dev[i];
2591 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
2592 i != sh->pd_idx &&
2593 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 2594 !(test_bit(R5_UPTODATE, &dev->flags) ||
2595 test_bit(R5_Wantcompute, &dev->flags)) &&
a4456856
DW		 2596 test_bit(R5_Insync, &dev->flags)) {
2597 if (
2598 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
45b4233c 2599 pr_debug("Read_old block "
a4456856
DW		 2600 "%d for Reconstruct\n", i);
2601 set_bit(R5_LOCKED, &dev->flags);
2602 set_bit(R5_Wantread, &dev->flags);
2603 s->locked++;
2604 } else {
2605 set_bit(STRIPE_DELAYED, &sh->state);
2606 set_bit(STRIPE_HANDLE, &sh->state);
2607 }
2608 }
2609 }
2610 /* now if nothing is locked, and if we have enough data,
2611 * we can start a write request
2612 */
f38e1219
DW		 2613 /* since handle_stripe can be called at any time we need to handle the
2614 * case where a compute block operation has been submitted and then a
ac6b53b6
DW		 2615 * subsequent call wants to start a write request. raid_run_ops only
2616 * handles the case where compute block and reconstruct are requested
f38e1219
DW		 2617 * simultaneously. If this is not the case then new writes need to be
2618 * held off until the compute completes.
2619 */
976ea8d4
DW		 2620 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
2621 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
2622 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
c0f7bddb 2623 schedule_reconstruction(sh, s, rcw == 0, 0);
a4456856
DW		 2624}
2625
1fe797e6 2626static void handle_stripe_dirtying6(raid5_conf_t *conf,
a4456856 2627 struct stripe_head *sh, struct stripe_head_state *s,
f2b3b44d 2628 int disks)
a4456856 2629{
a9b39a74 2630 int rcw = 0, pd_idx = sh->pd_idx, i;
34e04e87 2631 int qd_idx = sh->qd_idx;
a9b39a74
YT		 2632
2633 set_bit(STRIPE_HANDLE, &sh->state);
a4456856
DW		 2634 for (i = disks; i--; ) {
2635 struct r5dev *dev = &sh->dev[i];
a9b39a74
YT		 2636 /* check if we haven't enough data */
2637 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
2638 i != pd_idx && i != qd_idx &&
2639 !test_bit(R5_LOCKED, &dev->flags) &&
2640 !(test_bit(R5_UPTODATE, &dev->flags) ||
2641 test_bit(R5_Wantcompute, &dev->flags))) {
2642 rcw++;
2643 if (!test_bit(R5_Insync, &dev->flags))
2644 continue; /* it's a failed drive */
2645
2646 if (
2647 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
2648 pr_debug("Read_old stripe %llu "
2649 "block %d for Reconstruct\n",
2650 (unsigned long long)sh->sector, i);
2651 set_bit(R5_LOCKED, &dev->flags);
2652 set_bit(R5_Wantread, &dev->flags);
2653 s->locked++;
2654 } else {
2655 pr_debug("Request delayed stripe %llu "
2656 "block %d for Reconstruct\n",
2657 (unsigned long long)sh->sector, i);
2658 set_bit(STRIPE_DELAYED, &sh->state);
2659 set_bit(STRIPE_HANDLE, &sh->state);
a4456856
DW		 2660 }
2661 }
2662 }
a4456856
DW		 2663 /* now if nothing is locked, and if we have enough data, we can start a
2664 * write request
2665 */
a9b39a74
YT		 2666 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
2667 s->locked == 0 && rcw == 0 &&
a4456856 2668 !test_bit(STRIPE_BIT_DELAY, &sh->state)) {
a9b39a74 2669 schedule_reconstruction(sh, s, 1, 0);
a4456856
DW		 2670 }
2671}
2672
2673static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh,
2674 struct stripe_head_state *s, int disks)
2675{
ecc65c9b 2676 struct r5dev *dev = NULL;
bd2ab670 2677
a4456856 2678 set_bit(STRIPE_HANDLE, &sh->state);
e89f8962 2679
ecc65c9b
DW		 2680 switch (sh->check_state) {
2681 case check_state_idle:
2682 /* start a new check operation if there are no failures */
bd2ab670 2683 if (s->failed == 0) {
bd2ab670 2684 BUG_ON(s->uptodate != disks);
ecc65c9b
DW		 2685 sh->check_state = check_state_run;
2686 set_bit(STRIPE_OP_CHECK, &s->ops_request);
bd2ab670 2687 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
bd2ab670 2688 s->uptodate--;
ecc65c9b 2689 break;
bd2ab670 2690 }
f2b3b44d 2691 dev = &sh->dev[s->failed_num[0]];
ecc65c9b
DW		 2692 /* fall through */
2693 case check_state_compute_result:
2694 sh->check_state = check_state_idle;
2695 if (!dev)
2696 dev = &sh->dev[sh->pd_idx];
2697
2698 /* check that a write has not made the stripe insync */
2699 if (test_bit(STRIPE_INSYNC, &sh->state))
2700 break;
c8894419 2701
a4456856 2702 /* either failed parity check, or recovery is happening */
a4456856
DW		 2703 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
2704 BUG_ON(s->uptodate != disks);
2705
2706 set_bit(R5_LOCKED, &dev->flags);
ecc65c9b 2707 s->locked++;
a4456856 2708 set_bit(R5_Wantwrite, &dev->flags);
830ea016 2709
a4456856 2710 clear_bit(STRIPE_DEGRADED, &sh->state);
a4456856 2711 set_bit(STRIPE_INSYNC, &sh->state);
ecc65c9b
DW		 2712 break;
2713 case check_state_run:
2714 break; /* we will be called again upon completion */
2715 case check_state_check_result:
2716 sh->check_state = check_state_idle;
2717
2718 /* if a failure occurred during the check operation, leave
2719 * STRIPE_INSYNC not set and let the stripe be handled again
2720 */
2721 if (s->failed)
2722 break;
2723
2724 /* handle a successful check operation, if parity is correct
2725 * we are done. Otherwise update the mismatch count and repair
2726 * parity if !MD_RECOVERY_CHECK
2727 */
ad283ea4 2728 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
ecc65c9b
DW		 2729 /* parity is correct (on disc,
2730 * not in buffer any more)
2731 */
2732 set_bit(STRIPE_INSYNC, &sh->state);
2733 else {
2734 conf->mddev->resync_mismatches += STRIPE_SECTORS;
2735 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
2736 /* don't try to repair!! */
2737 set_bit(STRIPE_INSYNC, &sh->state);
2738 else {
2739 sh->check_state = check_state_compute_run;
976ea8d4 2740 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
ecc65c9b
DW		 2741 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2742 set_bit(R5_Wantcompute,
2743 &sh->dev[sh->pd_idx].flags);
2744 sh->ops.target = sh->pd_idx;
ac6b53b6 2745 sh->ops.target2 = -1;
ecc65c9b
DW		 2746 s->uptodate++;
2747 }
2748 }
2749 break;
2750 case check_state_compute_run:
2751 break;
2752 default:
2753 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
2754 __func__, sh->check_state,
2755 (unsigned long long) sh->sector);
2756 BUG();
a4456856
DW		 2757 }
2758}
2759
2760
2761static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh,
36d1c647 2762 struct stripe_head_state *s,
f2b3b44d 2763 int disks)
a4456856 2764{
a4456856 2765 int pd_idx = sh->pd_idx;
34e04e87 2766 int qd_idx = sh->qd_idx;
d82dfee0 2767 struct r5dev *dev;
a4456856
DW		 2768
2769 set_bit(STRIPE_HANDLE, &sh->state);
2770
2771 BUG_ON(s->failed > 2);
d82dfee0 2772
a4456856
DW		 2773 /* Want to check and possibly repair P and Q.
2774 * However there could be one 'failed' device, in which
2775 * case we can only check one of them, possibly using the
2776 * other to generate missing data
2777 */
2778
d82dfee0
DW		 2779 switch (sh->check_state) {
2780 case check_state_idle:
2781 /* start a new check operation if there are < 2 failures */
f2b3b44d 2782 if (s->failed == s->q_failed) {
d82dfee0 2783 /* The only possible failed device holds Q, so it
a4456856
DW		 2784 * makes sense to check P (If anything else were failed,
2785 * we would have used P to recreate it).
2786 */
d82dfee0 2787 sh->check_state = check_state_run;
a4456856 2788 }
f2b3b44d 2789 if (!s->q_failed && s->failed < 2) {
d82dfee0 2790 /* Q is not failed, and we didn't use it to generate
a4456856
DW		 2791 * anything, so it makes sense to check it
2792 */
d82dfee0
DW		 2793 if (sh->check_state == check_state_run)
2794 sh->check_state = check_state_run_pq;
2795 else
2796 sh->check_state = check_state_run_q;
a4456856 2797 }
a4456856 2798
d82dfee0
DW		 2799 /* discard potentially stale zero_sum_result */
2800 sh->ops.zero_sum_result = 0;
a4456856 2801
d82dfee0
DW		 2802 if (sh->check_state == check_state_run) {
2803 /* async_xor_zero_sum destroys the contents of P */
2804 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
2805 s->uptodate--;
a4456856 2806 }
d82dfee0
DW		 2807 if (sh->check_state >= check_state_run &&
2808 sh->check_state <= check_state_run_pq) {
2809 /* async_syndrome_zero_sum preserves P and Q, so
2810 * no need to mark them !uptodate here
2811 */
2812 set_bit(STRIPE_OP_CHECK, &s->ops_request);
2813 break;
a4456856
DW		 2814 }
2815
d82dfee0
DW		 2816 /* we have 2-disk failure */
2817 BUG_ON(s->failed != 2);
2818 /* fall through */
2819 case check_state_compute_result:
2820 sh->check_state = check_state_idle;
a4456856 2821
d82dfee0
DW		 2822 /* check that a write has not made the stripe insync */
2823 if (test_bit(STRIPE_INSYNC, &sh->state))
2824 break;
a4456856
DW		 2825
2826 /* now write out any block on a failed drive,
d82dfee0 2827 * or P or Q if they were recomputed
a4456856 2828 */
d82dfee0 2829 BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */
a4456856 2830 if (s->failed == 2) {
f2b3b44d 2831 dev = &sh->dev[s->failed_num[1]];
a4456856
DW		 2832 s->locked++;
2833 set_bit(R5_LOCKED, &dev->flags);
2834 set_bit(R5_Wantwrite, &dev->flags);
2835 }
2836 if (s->failed >= 1) {
f2b3b44d 2837 dev = &sh->dev[s->failed_num[0]];
a4456856
DW		 2838 s->locked++;
2839 set_bit(R5_LOCKED, &dev->flags);
2840 set_bit(R5_Wantwrite, &dev->flags);
2841 }
d82dfee0 2842 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
a4456856
DW		 2843 dev = &sh->dev[pd_idx];
2844 s->locked++;
2845 set_bit(R5_LOCKED, &dev->flags);
2846 set_bit(R5_Wantwrite, &dev->flags);
2847 }
d82dfee0 2848 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
a4456856
DW		 2849 dev = &sh->dev[qd_idx];
2850 s->locked++;
2851 set_bit(R5_LOCKED, &dev->flags);
2852 set_bit(R5_Wantwrite, &dev->flags);
2853 }
2854 clear_bit(STRIPE_DEGRADED, &sh->state);
2855
2856 set_bit(STRIPE_INSYNC, &sh->state);
d82dfee0
DW		 2857 break;
2858 case check_state_run:
2859 case check_state_run_q:
2860 case check_state_run_pq:
2861 break; /* we will be called again upon completion */
2862 case check_state_check_result:
2863 sh->check_state = check_state_idle;
2864
2865 /* handle a successful check operation, if parity is correct
2866 * we are done. Otherwise update the mismatch count and repair
2867 * parity if !MD_RECOVERY_CHECK
2868 */
2869 if (sh->ops.zero_sum_result == 0) {
2870 /* both parities are correct */
2871 if (!s->failed)
2872 set_bit(STRIPE_INSYNC, &sh->state);
2873 else {
2874 /* in contrast to the raid5 case we can validate
2875 * parity, but still have a failure to write
2876 * back
2877 */
2878 sh->check_state = check_state_compute_result;
2879 /* Returning at this point means that we may go
2880 * off and bring p and/or q uptodate again so
2881 * we make sure to check zero_sum_result again
2882 * to verify if p or q need writeback
2883 */
2884 }
2885 } else {
2886 conf->mddev->resync_mismatches += STRIPE_SECTORS;
2887 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
2888 /* don't try to repair!! */
2889 set_bit(STRIPE_INSYNC, &sh->state);
2890 else {
2891 int *target = &sh->ops.target;
2892
2893 sh->ops.target = -1;
2894 sh->ops.target2 = -1;
2895 sh->check_state = check_state_compute_run;
2896 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2897 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2898 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
2899 set_bit(R5_Wantcompute,
2900 &sh->dev[pd_idx].flags);
2901 *target = pd_idx;
2902 target = &sh->ops.target2;
2903 s->uptodate++;
2904 }
2905 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
2906 set_bit(R5_Wantcompute,
2907 &sh->dev[qd_idx].flags);
2908 *target = qd_idx;
2909 s->uptodate++;
2910 }
2911 }
2912 }
2913 break;
2914 case check_state_compute_run:
2915 break;
2916 default:
2917 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
2918 __func__, sh->check_state,
2919 (unsigned long long) sh->sector);
2920 BUG();
a4456856
DW		 2921 }
2922}
2923
2924static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh,
f2b3b44d 2925 struct stripe_head_state *r6s)
a4456856
DW		 2926{
2927 int i;
2928
2929 /* We have read all the blocks in this stripe and now we need to
2930 * copy some of them into a target stripe for expand.
2931 */
f0a50d37 2932 struct dma_async_tx_descriptor *tx = NULL;
a4456856
DW		 2933 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
2934 for (i = 0; i < sh->disks; i++)
34e04e87 2935 if (i != sh->pd_idx && i != sh->qd_idx) {
911d4ee8 2936 int dd_idx, j;
a4456856 2937 struct stripe_head *sh2;
a08abd8c 2938 struct async_submit_ctl submit;
a4456856 2939
784052ec 2940 sector_t bn = compute_blocknr(sh, i, 1);
911d4ee8
N		 2941 sector_t s = raid5_compute_sector(conf, bn, 0,
2942 &dd_idx, NULL);
a8c906ca 2943 sh2 = get_active_stripe(conf, s, 0, 1, 1);
a4456856
DW		 2944 if (sh2 == NULL)
2945 /* so far only the early blocks of this stripe
2946 * have been requested. When later blocks
2947 * get requested, we will try again
2948 */
2949 continue;
2950 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
2951 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
2952 /* must have already done this block */
2953 release_stripe(sh2);
2954 continue;
2955 }
f0a50d37
DW		 2956
2957 /* place all the copies on one channel */
a08abd8c 2958 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
f0a50d37 2959 tx = async_memcpy(sh2->dev[dd_idx].page,
88ba2aa5 2960 sh->dev[i].page, 0, 0, STRIPE_SIZE,
a08abd8c 2961 &submit);
f0a50d37 2962
a4456856
DW		 2963 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
2964 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
2965 for (j = 0; j < conf->raid_disks; j++)
2966 if (j != sh2->pd_idx &&
d0dabf7e 2967 (!r6s || j != sh2->qd_idx) &&
a4456856
DW		 2968 !test_bit(R5_Expanded, &sh2->dev[j].flags))
2969 break;
2970 if (j == conf->raid_disks) {
2971 set_bit(STRIPE_EXPAND_READY, &sh2->state);
2972 set_bit(STRIPE_HANDLE, &sh2->state);
2973 }
2974 release_stripe(sh2);
f0a50d37 2975
a4456856 2976 }
a2e08551
N		 2977 /* done submitting copies, wait for them to complete */
2978 if (tx) {
2979 async_tx_ack(tx);
2980 dma_wait_for_async_tx(tx);
2981 }
a4456856 2982}
1da177e4 2983
6bfe0b49 2984
1da177e4
LT		 2985/*
2986 * handle_stripe - do things to a stripe.
2987 *
2988 * We lock the stripe and then examine the state of various bits
2989 * to see what needs to be done.
2990 * Possible results:
2991 * return some read request which now have data
2992 * return some write requests which are safely on disc
2993 * schedule a read on some buffers
2994 * schedule a write of some buffers
2995 * return confirmation of parity correctness
2996 *
1da177e4
LT		 2997 * buffers are taken off read_list or write_list, and bh_cache buffers
2998 * get BH_Lock set before the stripe lock is released.
2999 *
3000 */
a4456856 3001
1442577b 3002static void handle_stripe5(struct stripe_head *sh)
1da177e4
LT		 3003{
3004 raid5_conf_t *conf = sh->raid_conf;
a4456856
DW		 3005 int disks = sh->disks, i;
3006 struct bio *return_bi = NULL;
3007 struct stripe_head_state s;
1da177e4 3008 struct r5dev *dev;
6bfe0b49 3009 mdk_rdev_t *blocked_rdev = NULL;
e0a115e5 3010 int prexor;
729a1866 3011 int dec_preread_active = 0;
1da177e4 3012
a4456856 3013 memset(&s, 0, sizeof(s));
600aa109
DW		 3014 pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d check:%d "
3015 "reconstruct:%d\n", (unsigned long long)sh->sector, sh->state,
3016 atomic_read(&sh->count), sh->pd_idx, sh->check_state,
3017 sh->reconstruct_state);
1da177e4 3018
a4456856
DW		 3019 s.syncing = test_bit(STRIPE_SYNCING, &sh->state);
3020 s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3021 s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
def6ae26 3022
83de75cc 3023 /* Now to look around and see what can be done */
9910f16a 3024 rcu_read_lock();
c4c1663b 3025 spin_lock_irq(&conf->device_lock);
1da177e4
LT		 3026 for (i=disks; i--; ) {
3027 mdk_rdev_t *rdev;
a9f326eb
N		 3028
3029 dev = &sh->dev[i];
1da177e4 3030
b5e98d65
DW		 3031 pr_debug("check %d: state 0x%lx toread %p read %p write %p "
3032 "written %p\n", i, dev->flags, dev->toread, dev->read,
3033 dev->towrite, dev->written);
3034
3035 /* maybe we can request a biofill operation
3036 *
3037 * new wantfill requests are only permitted while
83de75cc 3038 * ops_complete_biofill is guaranteed to be inactive
b5e98d65
DW		 3039 */
3040 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
83de75cc 3041 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
b5e98d65 3042 set_bit(R5_Wantfill, &dev->flags);
1da177e4
LT		 3043
3044 /* now count some things */
a4456856
DW		 3045 if (test_bit(R5_LOCKED, &dev->flags)) s.locked++;
3046 if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++;
f38e1219 3047 if (test_bit(R5_Wantcompute, &dev->flags)) s.compute++;
1da177e4 3048
b5e98d65
DW		 3049 if (test_bit(R5_Wantfill, &dev->flags))
3050 s.to_fill++;
3051 else if (dev->toread)
a4456856 3052 s.to_read++;
1da177e4 3053 if (dev->towrite) {
a4456856 3054 s.to_write++;
1da177e4 3055 if (!test_bit(R5_OVERWRITE, &dev->flags))
a4456856 3056 s.non_overwrite++;
1da177e4 3057 }
a4456856
DW		 3058 if (dev->written)
3059 s.written++;
9910f16a 3060 rdev = rcu_dereference(conf->disks[i].rdev);
ac4090d2
N		 3061 if (blocked_rdev == NULL &&
3062 rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
6bfe0b49
DW		 3063 blocked_rdev = rdev;
3064 atomic_inc(&rdev->nr_pending);
6bfe0b49 3065 }
415e72d0
N		 3066 clear_bit(R5_Insync, &dev->flags);
3067 if (!rdev)
3068 /* Not in-sync */;
3069 else if (test_bit(In_sync, &rdev->flags))
3070 set_bit(R5_Insync, &dev->flags);
3071 else {
3072 /* could be in-sync depending on recovery/reshape status */
3073 if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
3074 set_bit(R5_Insync, &dev->flags);
3075 }
3076 if (!test_bit(R5_Insync, &dev->flags)) {
14f8d26b 3077 /* The ReadError flag will just be confusing now */
4e5314b5
N		 3078 clear_bit(R5_ReadError, &dev->flags);
3079 clear_bit(R5_ReWrite, &dev->flags);
3080 }
415e72d0
N		 3081 if (test_bit(R5_ReadError, &dev->flags))
3082 clear_bit(R5_Insync, &dev->flags);
3083 if (!test_bit(R5_Insync, &dev->flags)) {
a4456856 3084 s.failed++;
f2b3b44d 3085 s.failed_num[0] = i;
415e72d0 3086 }
1da177e4 3087 }
c4c1663b 3088 spin_unlock_irq(&conf->device_lock);
9910f16a 3089 rcu_read_unlock();
b5e98d65 3090
6bfe0b49 3091 if (unlikely(blocked_rdev)) {
ac4090d2
N		 3092 if (s.syncing || s.expanding || s.expanded ||
3093 s.to_write || s.written) {
3094 set_bit(STRIPE_HANDLE, &sh->state);
3095 goto unlock;
3096 }
3097 /* There is nothing for the blocked_rdev to block */
3098 rdev_dec_pending(blocked_rdev, conf->mddev);
3099 blocked_rdev = NULL;
6bfe0b49
DW		 3100 }
3101
83de75cc
DW		 3102 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
3103 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
3104 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
3105 }
b5e98d65 3106
45b4233c 3107 pr_debug("locked=%d uptodate=%d to_read=%d"
1da177e4 3108 " to_write=%d failed=%d failed_num=%d\n",
a4456856 3109 s.locked, s.uptodate, s.to_read, s.to_write,
f2b3b44d 3110 s.failed, s.failed_num[0]);
1da177e4
LT		 3111 /* check if the array has lost two devices and, if so, some requests might
3112 * need to be failed
3113 */
a4456856 3114 if (s.failed > 1 && s.to_read+s.to_write+s.written)
1fe797e6 3115 handle_failed_stripe(conf, sh, &s, disks, &return_bi);
a4456856 3116 if (s.failed > 1 && s.syncing) {
1da177e4
LT		 3117 md_done_sync(conf->mddev, STRIPE_SECTORS,0);
3118 clear_bit(STRIPE_SYNCING, &sh->state);
a4456856 3119 s.syncing = 0;
1da177e4
LT		 3120 }
3121
3122 /* might be able to return some write requests if the parity block
3123 * is safe, or on a failed drive
3124 */
3125 dev = &sh->dev[sh->pd_idx];
a4456856
DW		 3126 if ( s.written &&
3127 ((test_bit(R5_Insync, &dev->flags) &&
3128 !test_bit(R5_LOCKED, &dev->flags) &&
3129 test_bit(R5_UPTODATE, &dev->flags)) ||
f2b3b44d 3130 (s.failed == 1 && s.failed_num[0] == sh->pd_idx)))
1fe797e6 3131 handle_stripe_clean_event(conf, sh, disks, &return_bi);
1da177e4
LT		 3132
3133 /* Now we might consider reading some blocks, either to check/generate
3134 * parity, or to satisfy requests
3135 * or to load a block that is being partially written.
3136 */
a4456856 3137 if (s.to_read || s.non_overwrite ||
976ea8d4 3138 (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding)
1fe797e6 3139 handle_stripe_fill5(sh, &s, disks);
1da177e4 3140
e33129d8
DW		 3141 /* Now we check to see if any write operations have recently
3142 * completed
3143 */
e0a115e5 3144 prexor = 0;
d8ee0728 3145 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
e0a115e5 3146 prexor = 1;
d8ee0728
DW		 3147 if (sh->reconstruct_state == reconstruct_state_drain_result ||
3148 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
600aa109 3149 sh->reconstruct_state = reconstruct_state_idle;
e33129d8
DW		 3150
3151 /* All the 'written' buffers and the parity block are ready to
3152 * be written back to disk
3153 */
3154 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags));
3155 for (i = disks; i--; ) {
3156 dev = &sh->dev[i];
3157 if (test_bit(R5_LOCKED, &dev->flags) &&
3158 (i == sh->pd_idx || dev->written)) {
3159 pr_debug("Writing block %d\n", i);
3160 set_bit(R5_Wantwrite, &dev->flags);
e0a115e5
DW		 3161 if (prexor)
3162 continue;
e33129d8
DW		 3163 if (!test_bit(R5_Insync, &dev->flags) ||
3164 (i == sh->pd_idx && s.failed == 0))
3165 set_bit(STRIPE_INSYNC, &sh->state);
3166 }
3167 }
729a1866
N		 3168 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3169 dec_preread_active = 1;
e33129d8
DW		 3170 }
3171
3172 /* Now to consider new write requests and what else, if anything
3173 * should be read. We do not handle new writes when:
3174 * 1/ A 'write' operation (copy+xor) is already in flight.
3175 * 2/ A 'check' operation is in flight, as it may clobber the parity
3176 * block.
3177 */
600aa109 3178 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
1fe797e6 3179 handle_stripe_dirtying5(conf, sh, &s, disks);
1da177e4
LT		 3180
3181 /* maybe we need to check and possibly fix the parity for this stripe
e89f8962
DW		 3182 * Any reads will already have been scheduled, so we just see if enough
3183 * data is available. The parity check is held off while parity
3184 * dependent operations are in flight.
1da177e4 3185 */
ecc65c9b
DW		 3186 if (sh->check_state ||
3187 (s.syncing && s.locked == 0 &&
976ea8d4 3188 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
ecc65c9b 3189 !test_bit(STRIPE_INSYNC, &sh->state)))
a4456856 3190 handle_parity_checks5(conf, sh, &s, disks);
e89f8962 3191
a4456856 3192 if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) {
1da177e4
LT		 3193 md_done_sync(conf->mddev, STRIPE_SECTORS,1);
3194 clear_bit(STRIPE_SYNCING, &sh->state);
3195 }
4e5314b5
N		 3196
3197 /* If the failed drive is just a ReadError, then we might need to progress
3198 * the repair/check process
3199 */
a4456856 3200 if (s.failed == 1 && !conf->mddev->ro &&
f2b3b44d
N		 3201 test_bit(R5_ReadError, &sh->dev[s.failed_num[0]].flags)
3202 && !test_bit(R5_LOCKED, &sh->dev[s.failed_num[0]].flags)
3203 && test_bit(R5_UPTODATE, &sh->dev[s.failed_num[0]].flags)
4e5314b5 3204 ) {
f2b3b44d 3205 dev = &sh->dev[s.failed_num[0]];
4e5314b5
N		 3206 if (!test_bit(R5_ReWrite, &dev->flags)) {
3207 set_bit(R5_Wantwrite, &dev->flags);
3208 set_bit(R5_ReWrite, &dev->flags);
3209 set_bit(R5_LOCKED, &dev->flags);
a4456856 3210 s.locked++;
4e5314b5
N		 3211 } else {
3212 /* let's read it back */
3213 set_bit(R5_Wantread, &dev->flags);
3214 set_bit(R5_LOCKED, &dev->flags);
a4456856 3215 s.locked++;
4e5314b5
N		 3216 }
3217 }
3218
600aa109
DW		 3219 /* Finish reconstruct operations initiated by the expansion process */
3220 if (sh->reconstruct_state == reconstruct_state_result) {
ab69ae12 3221 struct stripe_head *sh2
a8c906ca 3222 = get_active_stripe(conf, sh->sector, 1, 1, 1);
ab69ae12
N		 3223 if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) {
3224 /* sh cannot be written until sh2 has been read.
3225 * so arrange for sh to be delayed a little
3226 */
3227 set_bit(STRIPE_DELAYED, &sh->state);
3228 set_bit(STRIPE_HANDLE, &sh->state);
3229 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
3230 &sh2->state))
3231 atomic_inc(&conf->preread_active_stripes);
3232 release_stripe(sh2);
3233 goto unlock;
3234 }
3235 if (sh2)
3236 release_stripe(sh2);
3237
600aa109 3238 sh->reconstruct_state = reconstruct_state_idle;
f0a50d37 3239 clear_bit(STRIPE_EXPANDING, &sh->state);
23397883 3240 for (i = conf->raid_disks; i--; ) {
ccfcc3c1 3241 set_bit(R5_Wantwrite, &sh->dev[i].flags);
23397883 3242 set_bit(R5_LOCKED, &sh->dev[i].flags);
efe31143 3243 s.locked++;
23397883 3244 }
f0a50d37
DW		 3245 }
3246
3247 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
600aa109 3248 !sh->reconstruct_state) {
f0a50d37
DW		 3249 /* Need to write out all blocks after computing parity */
3250 sh->disks = conf->raid_disks;
911d4ee8 3251 stripe_set_idx(sh->sector, conf, 0, sh);
c0f7bddb 3252 schedule_reconstruction(sh, &s, 1, 1);
600aa109 3253 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
ccfcc3c1 3254 clear_bit(STRIPE_EXPAND_READY, &sh->state);
f6705578 3255 atomic_dec(&conf->reshape_stripes);
ccfcc3c1
N		 3256 wake_up(&conf->wait_for_overlap);
3257 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3258 }
3259
0f94e87c 3260 if (s.expanding && s.locked == 0 &&
976ea8d4 3261 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
a4456856 3262 handle_stripe_expansion(conf, sh, NULL);
ccfcc3c1 3263
6bfe0b49 3264 unlock:
1da177e4 3265
6bfe0b49
DW		 3266 /* wait for this device to become unblocked */
3267 if (unlikely(blocked_rdev))
3268 md_wait_for_blocked_rdev(blocked_rdev, conf->mddev);
3269
600aa109 3270 if (s.ops_request)
ac6b53b6 3271 raid_run_ops(sh, s.ops_request);
d84e0f10 3272
c4e5ac0a 3273 ops_run_io(sh, &s);
1da177e4 3274
729a1866
N		 3275 if (dec_preread_active) {
3276 /* We delay this until after ops_run_io so that if make_request
e9c7469b 3277 * is waiting on a flush, it won't continue until the writes
729a1866
N		 3278 * have actually been submitted.
3279 */
3280 atomic_dec(&conf->preread_active_stripes);
3281 if (atomic_read(&conf->preread_active_stripes) <
3282 IO_THRESHOLD)
3283 md_wakeup_thread(conf->mddev->thread);
3284 }
a4456856 3285 return_io(return_bi);
1da177e4
LT		 3286}
3287
1442577b 3288static void handle_stripe6(struct stripe_head *sh)
1da177e4 3289{
bff61975 3290 raid5_conf_t *conf = sh->raid_conf;
f416885e 3291 int disks = sh->disks;
a4456856 3292 struct bio *return_bi = NULL;
34e04e87 3293 int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx;
a4456856 3294 struct stripe_head_state s;
16a53ecc 3295 struct r5dev *dev, *pdev, *qdev;
6bfe0b49 3296 mdk_rdev_t *blocked_rdev = NULL;
729a1866 3297 int dec_preread_active = 0;
1da177e4 3298
45b4233c 3299 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
6c0069c0 3300 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
a4456856 3301 (unsigned long long)sh->sector, sh->state,
6c0069c0
YT		 3302 atomic_read(&sh->count), pd_idx, qd_idx,
3303 sh->check_state, sh->reconstruct_state);
a4456856 3304 memset(&s, 0, sizeof(s));
72626685 3305
a4456856
DW		 3306 s.syncing = test_bit(STRIPE_SYNCING, &sh->state);
3307 s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3308 s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
16a53ecc 3309 /* Now to look around and see what can be done */
1da177e4
LT		 3310
3311 rcu_read_lock();
c4c1663b 3312 spin_lock_irq(&conf->device_lock);
16a53ecc
N		 3313 for (i=disks; i--; ) {
3314 mdk_rdev_t *rdev;
3315 dev = &sh->dev[i];
1da177e4 3316
45b4233c 3317 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
16a53ecc 3318 i, dev->flags, dev->toread, dev->towrite, dev->written);
6c0069c0
YT		 3319 /* maybe we can reply to a read
3320 *
3321 * new wantfill requests are only permitted while
3322 * ops_complete_biofill is guaranteed to be inactive
3323 */
3324 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
3325 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
3326 set_bit(R5_Wantfill, &dev->flags);
1da177e4 3327
16a53ecc 3328 /* now count some things */
a4456856
DW		 3329 if (test_bit(R5_LOCKED, &dev->flags)) s.locked++;
3330 if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++;
2d6e4ecc
DW		 3331 if (test_bit(R5_Wantcompute, &dev->flags)) {
3332 s.compute++;
3333 BUG_ON(s.compute > 2);
3334 }
1da177e4 3335
6c0069c0
YT		 3336 if (test_bit(R5_Wantfill, &dev->flags)) {
3337 s.to_fill++;
3338 } else if (dev->toread)
a4456856 3339 s.to_read++;
16a53ecc 3340 if (dev->towrite) {
a4456856 3341 s.to_write++;
16a53ecc 3342 if (!test_bit(R5_OVERWRITE, &dev->flags))
a4456856 3343 s.non_overwrite++;
16a53ecc 3344 }
a4456856
DW		 3345 if (dev->written)
3346 s.written++;
16a53ecc 3347 rdev = rcu_dereference(conf->disks[i].rdev);
ac4090d2
N		 3348 if (blocked_rdev == NULL &&
3349 rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
6bfe0b49
DW		 3350 blocked_rdev = rdev;
3351 atomic_inc(&rdev->nr_pending);
6bfe0b49 3352 }
415e72d0
N		 3353 clear_bit(R5_Insync, &dev->flags);
3354 if (!rdev)
3355 /* Not in-sync */;
3356 else if (test_bit(In_sync, &rdev->flags))
3357 set_bit(R5_Insync, &dev->flags);
3358 else {
3359 /* in sync if before recovery_offset */
3360 if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
3361 set_bit(R5_Insync, &dev->flags);
3362 }
3363 if (!test_bit(R5_Insync, &dev->flags)) {
16a53ecc
N		 3364 /* The ReadError flag will just be confusing now */
3365 clear_bit(R5_ReadError, &dev->flags);
3366 clear_bit(R5_ReWrite, &dev->flags);
1da177e4 3367 }
415e72d0
N		 3368 if (test_bit(R5_ReadError, &dev->flags))
3369 clear_bit(R5_Insync, &dev->flags);
3370 if (!test_bit(R5_Insync, &dev->flags)) {
a4456856 3371 if (s.failed < 2)
f2b3b44d 3372 s.failed_num[s.failed] = i;
a4456856 3373 s.failed++;
415e72d0 3374 }
1da177e4 3375 }
c4c1663b 3376 spin_unlock_irq(&conf->device_lock);
1da177e4 3377 rcu_read_unlock();
6bfe0b49
DW		 3378
3379 if (unlikely(blocked_rdev)) {
ac4090d2
N		 3380 if (s.syncing || s.expanding || s.expanded ||
3381 s.to_write || s.written) {
3382 set_bit(STRIPE_HANDLE, &sh->state);
3383 goto unlock;
3384 }
3385 /* There is nothing for the blocked_rdev to block */
3386 rdev_dec_pending(blocked_rdev, conf->mddev);
3387 blocked_rdev = NULL;
6bfe0b49 3388 }
ac4090d2 3389
6c0069c0
YT		 3390 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
3391 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
3392 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
3393 }
3394
45b4233c 3395 pr_debug("locked=%d uptodate=%d to_read=%d"
16a53ecc 3396 " to_write=%d failed=%d failed_num=%d,%d\n",
a4456856 3397 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
f2b3b44d 3398 s.failed_num[0], s.failed_num[1]);
a4456856
DW		 3399 /* check if the array has lost >2 devices and, if so, some requests
3400 * might need to be failed
16a53ecc 3401 */
a4456856 3402 if (s.failed > 2 && s.to_read+s.to_write+s.written)
1fe797e6 3403 handle_failed_stripe(conf, sh, &s, disks, &return_bi);
a4456856 3404 if (s.failed > 2 && s.syncing) {
16a53ecc
N		 3405 md_done_sync(conf->mddev, STRIPE_SECTORS,0);
3406 clear_bit(STRIPE_SYNCING, &sh->state);
a4456856 3407 s.syncing = 0;
16a53ecc
N		 3408 }
3409
3410 /*
3411 * might be able to return some write requests if the parity blocks
3412 * are safe, or on a failed drive
3413 */
3414 pdev = &sh->dev[pd_idx];
f2b3b44d
N		 3415 s.p_failed = (s.failed >= 1 && s.failed_num[0] == pd_idx)
3416 || (s.failed >= 2 && s.failed_num[1] == pd_idx);
34e04e87 3417 qdev = &sh->dev[qd_idx];
f2b3b44d
N		 3418 s.q_failed = (s.failed >= 1 && s.failed_num[0] == qd_idx)
3419 || (s.failed >= 2 && s.failed_num[1] == qd_idx);
a4456856 3420
f2b3b44d
N		 3421 if (s.written &&
3422 (s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
16a53ecc 3423 && !test_bit(R5_LOCKED, &pdev->flags)
a4456856 3424 && test_bit(R5_UPTODATE, &pdev->flags)))) &&
f2b3b44d 3425 (s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
16a53ecc 3426 && !test_bit(R5_LOCKED, &qdev->flags)
a4456856 3427 && test_bit(R5_UPTODATE, &qdev->flags)))))
1fe797e6 3428 handle_stripe_clean_event(conf, sh, disks, &return_bi);
16a53ecc
N		 3429
3430 /* Now we might consider reading some blocks, either to check/generate
3431 * parity, or to satisfy requests
3432 * or to load a block that is being partially written.
3433 */
a4456856 3434 if (s.to_read || s.non_overwrite || (s.to_write && s.failed) ||
6c0069c0 3435 (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding)
f2b3b44d 3436 handle_stripe_fill6(sh, &s, disks);
16a53ecc 3437
6c0069c0
YT		 3438 /* Now we check to see if any write operations have recently
3439 * completed
3440 */
3441 if (sh->reconstruct_state == reconstruct_state_drain_result) {
6c0069c0
YT		 3442
3443 sh->reconstruct_state = reconstruct_state_idle;
3444 /* All the 'written' buffers and the parity blocks are ready to
3445 * be written back to disk
3446 */
3447 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags));
3448 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags));
3449 for (i = disks; i--; ) {
3450 dev = &sh->dev[i];
3451 if (test_bit(R5_LOCKED, &dev->flags) &&
3452 (i == sh->pd_idx || i == qd_idx ||
3453 dev->written)) {
3454 pr_debug("Writing block %d\n", i);
3455 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
3456 set_bit(R5_Wantwrite, &dev->flags);
3457 if (!test_bit(R5_Insync, &dev->flags) ||
3458 ((i == sh->pd_idx || i == qd_idx) &&
3459 s.failed == 0))
3460 set_bit(STRIPE_INSYNC, &sh->state);
3461 }
3462 }
729a1866
N		 3463 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3464 dec_preread_active = 1;
6c0069c0
YT		 3465 }
3466
a9b39a74
YT		 3467 /* Now to consider new write requests and what else, if anything
3468 * should be read. We do not handle new writes when:
3469 * 1/ A 'write' operation (copy+gen_syndrome) is already in flight.
3470 * 2/ A 'check' operation is in flight, as it may clobber the parity
3471 * block.
3472 */
3473 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
f2b3b44d 3474 handle_stripe_dirtying6(conf, sh, &s, disks);
16a53ecc
N		 3475
3476 /* maybe we need to check and possibly fix the parity for this stripe
a4456856 3477 * Any reads will already have been scheduled, so we just see if enough
6c0069c0
YT		 3478 * data is available. The parity check is held off while parity
3479 * dependent operations are in flight.
16a53ecc 3480 */
6c0069c0
YT		 3481 if (sh->check_state ||
3482 (s.syncing && s.locked == 0 &&
3483 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
3484 !test_bit(STRIPE_INSYNC, &sh->state)))
f2b3b44d 3485 handle_parity_checks6(conf, sh, &s, disks);
16a53ecc 3486
a4456856 3487 if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) {
16a53ecc
N		 3488 md_done_sync(conf->mddev, STRIPE_SECTORS,1);
3489 clear_bit(STRIPE_SYNCING, &sh->state);
3490 }
3491
3492 /* If the failed drives are just a ReadError, then we might need
3493 * to progress the repair/check process
3494 */
a4456856
DW		 3495 if (s.failed <= 2 && !conf->mddev->ro)
3496 for (i = 0; i < s.failed; i++) {
f2b3b44d 3497 dev = &sh->dev[s.failed_num[i]];
16a53ecc
N		 3498 if (test_bit(R5_ReadError, &dev->flags)
3499 && !test_bit(R5_LOCKED, &dev->flags)
3500 && test_bit(R5_UPTODATE, &dev->flags)
3501 ) {
3502 if (!test_bit(R5_ReWrite, &dev->flags)) {
3503 set_bit(R5_Wantwrite, &dev->flags);
3504 set_bit(R5_ReWrite, &dev->flags);
3505 set_bit(R5_LOCKED, &dev->flags);
6c0069c0 3506 s.locked++;
16a53ecc
N		 3507 } else {
3508 /* let's read it back */
3509 set_bit(R5_Wantread, &dev->flags);
3510 set_bit(R5_LOCKED, &dev->flags);
6c0069c0 3511 s.locked++;
16a53ecc
N		 3512 }
3513 }
3514 }
f416885e 3515
6c0069c0
YT		 3516 /* Finish reconstruct operations initiated by the expansion process */
3517 if (sh->reconstruct_state == reconstruct_state_result) {
3518 sh->reconstruct_state = reconstruct_state_idle;
3519 clear_bit(STRIPE_EXPANDING, &sh->state);
3520 for (i = conf->raid_disks; i--; ) {
3521 set_bit(R5_Wantwrite, &sh->dev[i].flags);
3522 set_bit(R5_LOCKED, &sh->dev[i].flags);
3523 s.locked++;
3524 }
3525 }
3526
3527 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
3528 !sh->reconstruct_state) {
ab69ae12 3529 struct stripe_head *sh2
a8c906ca 3530 = get_active_stripe(conf, sh->sector, 1, 1, 1);
ab69ae12
N		 3531 if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) {
3532 /* sh cannot be written until sh2 has been read.
3533 * so arrange for sh to be delayed a little
3534 */
3535 set_bit(STRIPE_DELAYED, &sh->state);
3536 set_bit(STRIPE_HANDLE, &sh->state);
3537 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
3538 &sh2->state))
3539 atomic_inc(&conf->preread_active_stripes);
3540 release_stripe(sh2);
3541 goto unlock;
3542 }
3543 if (sh2)
3544 release_stripe(sh2);
3545
f416885e
N		 3546 /* Need to write out all blocks after computing P&Q */
3547 sh->disks = conf->raid_disks;
911d4ee8 3548 stripe_set_idx(sh->sector, conf, 0, sh);
6c0069c0
YT		 3549 schedule_reconstruction(sh, &s, 1, 1);
3550 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
f416885e
N		 3551 clear_bit(STRIPE_EXPAND_READY, &sh->state);
3552 atomic_dec(&conf->reshape_stripes);
3553 wake_up(&conf->wait_for_overlap);
3554 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3555 }
3556
0f94e87c 3557 if (s.expanding && s.locked == 0 &&
976ea8d4 3558 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
f2b3b44d 3559 handle_stripe_expansion(conf, sh, &s);
f416885e 3560
6bfe0b49 3561 unlock:
16a53ecc 3562
6bfe0b49
DW		 3563 /* wait for this device to become unblocked */
3564 if (unlikely(blocked_rdev))
3565 md_wait_for_blocked_rdev(blocked_rdev, conf->mddev);
3566
6c0069c0
YT		 3567 if (s.ops_request)
3568 raid_run_ops(sh, s.ops_request);
3569
f0e43bcd 3570 ops_run_io(sh, &s);
16a53ecc 3571
729a1866
N		 3572
3573 if (dec_preread_active) {
3574 /* We delay this until after ops_run_io so that if make_request
e9c7469b 3575 * is waiting on a flush, it won't continue until the writes
729a1866
N		 3576 * have actually been submitted.
3577 */
3578 atomic_dec(&conf->preread_active_stripes);
3579 if (atomic_read(&conf->preread_active_stripes) <
3580 IO_THRESHOLD)
3581 md_wakeup_thread(conf->mddev->thread);
3582 }
3583
f0e43bcd 3584 return_io(return_bi);
16a53ecc
N		 3585}
3586
1442577b 3587static void handle_stripe(struct stripe_head *sh)
16a53ecc 3588{
c4c1663b
N		 3589 clear_bit(STRIPE_HANDLE, &sh->state);
3590 if (test_and_set_bit(STRIPE_ACTIVE, &sh->state)) {
3591 /* already being handled, ensure it gets handled
3592 * again when current action finishes */
3593 set_bit(STRIPE_HANDLE, &sh->state);
3594 return;
3595 }
3596
82e5a171
N		 3597 if (test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
3598 set_bit(STRIPE_SYNCING, &sh->state);
3599 clear_bit(STRIPE_INSYNC, &sh->state);
3600 }
3601 clear_bit(STRIPE_DELAYED, &sh->state);
3602
16a53ecc 3603 if (sh->raid_conf->level == 6)
1442577b 3604 handle_stripe6(sh);
16a53ecc 3605 else
1442577b 3606 handle_stripe5(sh);
c4c1663b 3607 clear_bit(STRIPE_ACTIVE, &sh->state);
16a53ecc
N		 3608}
3609
16a53ecc
N		 3610static void raid5_activate_delayed(raid5_conf_t *conf)
3611{
3612 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
3613 while (!list_empty(&conf->delayed_list)) {
3614 struct list_head *l = conf->delayed_list.next;
3615 struct stripe_head *sh;
3616 sh = list_entry(l, struct stripe_head, lru);
3617 list_del_init(l);
3618 clear_bit(STRIPE_DELAYED, &sh->state);
3619 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3620 atomic_inc(&conf->preread_active_stripes);
8b3e6cdc 3621 list_add_tail(&sh->lru, &conf->hold_list);
16a53ecc 3622 }
482c0834 3623 }
16a53ecc
N		 3624}
3625
3626static void activate_bit_delay(raid5_conf_t *conf)
3627{
3628 /* device_lock is held */
3629 struct list_head head;
3630 list_add(&head, &conf->bitmap_list);
3631 list_del_init(&conf->bitmap_list);
3632 while (!list_empty(&head)) {
3633 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
3634 list_del_init(&sh->lru);
3635 atomic_inc(&sh->count);
3636 __release_stripe(conf, sh);
3637 }
3638}
3639
11d8a6e3 3640int md_raid5_congested(mddev_t *mddev, int bits)
f022b2fd 3641{
070ec55d 3642 raid5_conf_t *conf = mddev->private;
f022b2fd
N		 3643
3644 /* No difference between reads and writes. Just check
3645 * how busy the stripe_cache is
3646 */
3fa841d7 3647
f022b2fd
N		 3648 if (conf->inactive_blocked)
3649 return 1;
3650 if (conf->quiesce)
3651 return 1;
3652 if (list_empty_careful(&conf->inactive_list))
3653 return 1;
3654
3655 return 0;
3656}
11d8a6e3
N		 3657EXPORT_SYMBOL_GPL(md_raid5_congested);
3658
3659static int raid5_congested(void *data, int bits)
3660{
3661 mddev_t *mddev = data;
3662
3663 return mddev_congested(mddev, bits) ||
3664 md_raid5_congested(mddev, bits);
3665}
f022b2fd 3666
23032a0e
RBJ		 3667/* We want read requests to align with chunks where possible,
3668 * but write requests don't need to.
3669 */
cc371e66
AK		 3670static int raid5_mergeable_bvec(struct request_queue *q,
3671 struct bvec_merge_data *bvm,
3672 struct bio_vec *biovec)
23032a0e
RBJ		 3673{
3674 mddev_t *mddev = q->queuedata;
cc371e66 3675 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
23032a0e 3676 int max;
9d8f0363 3677 unsigned int chunk_sectors = mddev->chunk_sectors;
cc371e66 3678 unsigned int bio_sectors = bvm->bi_size >> 9;
23032a0e 3679
cc371e66 3680 if ((bvm->bi_rw & 1) == WRITE)
23032a0e
RBJ		 3681 return biovec->bv_len; /* always allow writes to be mergeable */
3682
664e7c41
AN		 3683 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
3684 chunk_sectors = mddev->new_chunk_sectors;
23032a0e
RBJ		 3685 max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
3686 if (max < 0) max = 0;
3687 if (max <= biovec->bv_len && bio_sectors == 0)
3688 return biovec->bv_len;
3689 else
3690 return max;
3691}
3692
f679623f
RBJ		 3693
3694static int in_chunk_boundary(mddev_t *mddev, struct bio *bio)
3695{
3696 sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
9d8f0363 3697 unsigned int chunk_sectors = mddev->chunk_sectors;
f679623f
RBJ		 3698 unsigned int bio_sectors = bio->bi_size >> 9;
3699
664e7c41
AN		 3700 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
3701 chunk_sectors = mddev->new_chunk_sectors;
f679623f
RBJ		 3702 return chunk_sectors >=
3703 ((sector & (chunk_sectors - 1)) + bio_sectors);
3704}
3705
46031f9a
RBJ		 3706/*
3707 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
3708 * later sampled by raid5d.
3709 */
3710static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf)
3711{
3712 unsigned long flags;
3713
3714 spin_lock_irqsave(&conf->device_lock, flags);
3715
3716 bi->bi_next = conf->retry_read_aligned_list;
3717 conf->retry_read_aligned_list = bi;
3718
3719 spin_unlock_irqrestore(&conf->device_lock, flags);
3720 md_wakeup_thread(conf->mddev->thread);
3721}
3722
3723
3724static struct bio *remove_bio_from_retry(raid5_conf_t *conf)
3725{
3726 struct bio *bi;
3727
3728 bi = conf->retry_read_aligned;
3729 if (bi) {
3730 conf->retry_read_aligned = NULL;
3731 return bi;
3732 }
3733 bi = conf->retry_read_aligned_list;
3734 if(bi) {
387bb173 3735 conf->retry_read_aligned_list = bi->bi_next;
46031f9a 3736 bi->bi_next = NULL;
960e739d
JA		 3737 /*
3738 * this sets the active strip count to 1 and the processed
3739 * strip count to zero (upper 8 bits)
3740 */
46031f9a 3741 bi->bi_phys_segments = 1; /* biased count of active stripes */
46031f9a
RBJ		 3742 }
3743
3744 return bi;
3745}
3746
3747
f679623f
RBJ		 3748/*
3749 * The "raid5_align_endio" should check if the read succeeded and if it
3750 * did, call bio_endio on the original bio (having bio_put the new bio
3751 * first).
3752 * If the read failed..
3753 */
6712ecf8 3754static void raid5_align_endio(struct bio *bi, int error)
f679623f
RBJ		 3755{
3756 struct bio* raid_bi = bi->bi_private;
46031f9a
RBJ		 3757 mddev_t *mddev;
3758 raid5_conf_t *conf;
3759 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
3760 mdk_rdev_t *rdev;
3761
f679623f 3762 bio_put(bi);
46031f9a 3763
46031f9a
RBJ		 3764 rdev = (void*)raid_bi->bi_next;
3765 raid_bi->bi_next = NULL;
2b7f2228
N		 3766 mddev = rdev->mddev;
3767 conf = mddev->private;
46031f9a
RBJ		 3768
3769 rdev_dec_pending(rdev, conf->mddev);
3770
3771 if (!error && uptodate) {
6712ecf8 3772 bio_endio(raid_bi, 0);
46031f9a
RBJ		 3773 if (atomic_dec_and_test(&conf->active_aligned_reads))
3774 wake_up(&conf->wait_for_stripe);
6712ecf8 3775 return;
46031f9a
RBJ		 3776 }
3777
3778
45b4233c 3779 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
46031f9a
RBJ		 3780
3781 add_bio_to_retry(raid_bi, conf);
f679623f
RBJ		 3782}
3783
387bb173
NB		 3784static int bio_fits_rdev(struct bio *bi)
3785{
165125e1 3786 struct request_queue *q = bdev_get_queue(bi->bi_bdev);
387bb173 3787
ae03bf63 3788 if ((bi->bi_size>>9) > queue_max_sectors(q))
387bb173
NB		 3789 return 0;
3790 blk_recount_segments(q, bi);
8a78362c 3791 if (bi->bi_phys_segments > queue_max_segments(q))
387bb173
NB		 3792 return 0;
3793
3794 if (q->merge_bvec_fn)
3795 /* it's too hard to apply the merge_bvec_fn at this stage,
3796 * just just give up
3797 */
3798 return 0;
3799
3800 return 1;
3801}
3802
3803
21a52c6d 3804static int chunk_aligned_read(mddev_t *mddev, struct bio * raid_bio)
f679623f 3805{
070ec55d 3806 raid5_conf_t *conf = mddev->private;
8553fe7e 3807 int dd_idx;
f679623f
RBJ		 3808 struct bio* align_bi;
3809 mdk_rdev_t *rdev;
3810
3811 if (!in_chunk_boundary(mddev, raid_bio)) {
45b4233c 3812 pr_debug("chunk_aligned_read : non aligned\n");
f679623f
RBJ		 3813 return 0;
3814 }
3815 /*
a167f663 3816 * use bio_clone_mddev to make a copy of the bio
f679623f 3817 */
a167f663 3818 align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev);
f679623f
RBJ		 3819 if (!align_bi)
3820 return 0;
3821 /*
3822 * set bi_end_io to a new function, and set bi_private to the
3823 * original bio.
3824 */
3825 align_bi->bi_end_io = raid5_align_endio;
3826 align_bi->bi_private = raid_bio;
3827 /*
3828 * compute position
3829 */
112bf897
N		 3830 align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector,
3831 0,
911d4ee8 3832 &dd_idx, NULL);
f679623f
RBJ		 3833
3834 rcu_read_lock();
3835 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
3836 if (rdev && test_bit(In_sync, &rdev->flags)) {
f679623f
RBJ		 3837 atomic_inc(&rdev->nr_pending);
3838 rcu_read_unlock();
46031f9a
RBJ		 3839 raid_bio->bi_next = (void*)rdev;
3840 align_bi->bi_bdev = rdev->bdev;
3841 align_bi->bi_flags &= ~(1 << BIO_SEG_VALID);
3842 align_bi->bi_sector += rdev->data_offset;
3843
387bb173
NB		 3844 if (!bio_fits_rdev(align_bi)) {
3845 /* too big in some way */
3846 bio_put(align_bi);
3847 rdev_dec_pending(rdev, mddev);
3848 return 0;
3849 }
3850
46031f9a
RBJ		 3851 spin_lock_irq(&conf->device_lock);
3852 wait_event_lock_irq(conf->wait_for_stripe,
3853 conf->quiesce == 0,
3854 conf->device_lock, /* nothing */);
3855 atomic_inc(&conf->active_aligned_reads);
3856 spin_unlock_irq(&conf->device_lock);
3857
f679623f
RBJ		 3858 generic_make_request(align_bi);
3859 return 1;
3860 } else {
3861 rcu_read_unlock();
46031f9a 3862 bio_put(align_bi);
f679623f
RBJ		 3863 return 0;
3864 }
3865}
3866
8b3e6cdc
DW		 3867/* __get_priority_stripe - get the next stripe to process
3868 *
3869 * Full stripe writes are allowed to pass preread active stripes up until
3870 * the bypass_threshold is exceeded. In general the bypass_count
3871 * increments when the handle_list is handled before the hold_list; however, it
3872 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
3873 * stripe with in flight i/o. The bypass_count will be reset when the
3874 * head of the hold_list has changed, i.e. the head was promoted to the
3875 * handle_list.
3876 */
3877static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf)
3878{
3879 struct stripe_head *sh;
3880
3881 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
3882 __func__,
3883 list_empty(&conf->handle_list) ? "empty" : "busy",
3884 list_empty(&conf->hold_list) ? "empty" : "busy",
3885 atomic_read(&conf->pending_full_writes), conf->bypass_count);
3886
3887 if (!list_empty(&conf->handle_list)) {
3888 sh = list_entry(conf->handle_list.next, typeof(*sh), lru);
3889
3890 if (list_empty(&conf->hold_list))
3891 conf->bypass_count = 0;
3892 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
3893 if (conf->hold_list.next == conf->last_hold)
3894 conf->bypass_count++;
3895 else {
3896 conf->last_hold = conf->hold_list.next;
3897 conf->bypass_count -= conf->bypass_threshold;
3898 if (conf->bypass_count < 0)
3899 conf->bypass_count = 0;
3900 }
3901 }
3902 } else if (!list_empty(&conf->hold_list) &&
3903 ((conf->bypass_threshold &&
3904 conf->bypass_count > conf->bypass_threshold) ||
3905 atomic_read(&conf->pending_full_writes) == 0)) {
3906 sh = list_entry(conf->hold_list.next,
3907 typeof(*sh), lru);
3908 conf->bypass_count -= conf->bypass_threshold;
3909 if (conf->bypass_count < 0)
3910 conf->bypass_count = 0;
3911 } else
3912 return NULL;
3913
3914 list_del_init(&sh->lru);
3915 atomic_inc(&sh->count);
3916 BUG_ON(atomic_read(&sh->count) != 1);
3917 return sh;
3918}
f679623f 3919
21a52c6d 3920static int make_request(mddev_t *mddev, struct bio * bi)
1da177e4 3921{
070ec55d 3922 raid5_conf_t *conf = mddev->private;
911d4ee8 3923 int dd_idx;
1da177e4
LT		 3924 sector_t new_sector;
3925 sector_t logical_sector, last_sector;
3926 struct stripe_head *sh;
a362357b 3927 const int rw = bio_data_dir(bi);
49077326 3928 int remaining;
7c13edc8 3929 int plugged;
1da177e4 3930
e9c7469b
TH		 3931 if (unlikely(bi->bi_rw & REQ_FLUSH)) {
3932 md_flush_request(mddev, bi);
e5dcdd80
N		 3933 return 0;
3934 }
3935
3d310eb7 3936 md_write_start(mddev, bi);
06d91a5f 3937
802ba064 3938 if (rw == READ &&
52488615 3939 mddev->reshape_position == MaxSector &&
21a52c6d 3940 chunk_aligned_read(mddev,bi))
99c0fb5f 3941 return 0;
52488615 3942
1da177e4
LT		 3943 logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
3944 last_sector = bi->bi_sector + (bi->bi_size>>9);
3945 bi->bi_next = NULL;
3946 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
06d91a5f 3947
7c13edc8 3948 plugged = mddev_check_plugged(mddev);
1da177e4
LT		 3949 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
3950 DEFINE_WAIT(w);
16a53ecc 3951 int disks, data_disks;
b5663ba4 3952 int previous;
b578d55f 3953
7ecaa1e6 3954 retry:
b5663ba4 3955 previous = 0;
b0f9ec04 3956 disks = conf->raid_disks;
b578d55f 3957 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
b0f9ec04 3958 if (unlikely(conf->reshape_progress != MaxSector)) {
fef9c61f 3959 /* spinlock is needed as reshape_progress may be
df8e7f76
N		 3960 * 64bit on a 32bit platform, and so it might be
3961 * possible to see a half-updated value
aeb878b0 3962 * Of course reshape_progress could change after
df8e7f76
N		 3963 * the lock is dropped, so once we get a reference
3964 * to the stripe that we think it is, we will have
3965 * to check again.
3966 */
7ecaa1e6 3967 spin_lock_irq(&conf->device_lock);
fef9c61f
N		 3968 if (mddev->delta_disks < 0
3969 ? logical_sector < conf->reshape_progress
3970 : logical_sector >= conf->reshape_progress) {
7ecaa1e6 3971 disks = conf->previous_raid_disks;
b5663ba4
N		 3972 previous = 1;
3973 } else {
fef9c61f
N		 3974 if (mddev->delta_disks < 0
3975 ? logical_sector < conf->reshape_safe
3976 : logical_sector >= conf->reshape_safe) {
b578d55f
N		 3977 spin_unlock_irq(&conf->device_lock);
3978 schedule();
3979 goto retry;
3980 }
3981 }
7ecaa1e6
N		 3982 spin_unlock_irq(&conf->device_lock);
3983 }
16a53ecc
N		 3984 data_disks = disks - conf->max_degraded;
3985
112bf897
N		 3986 new_sector = raid5_compute_sector(conf, logical_sector,
3987 previous,
911d4ee8 3988 &dd_idx, NULL);
0c55e022 3989 pr_debug("raid456: make_request, sector %llu logical %llu\n",
1da177e4
LT		 3990 (unsigned long long)new_sector,
3991 (unsigned long long)logical_sector);
3992
b5663ba4 3993 sh = get_active_stripe(conf, new_sector, previous,
a8c906ca 3994 (bi->bi_rw&RWA_MASK), 0);
1da177e4 3995 if (sh) {
b0f9ec04 3996 if (unlikely(previous)) {
7ecaa1e6 3997 /* expansion might have moved on while waiting for a
df8e7f76
N		 3998 * stripe, so we must do the range check again.
3999 * Expansion could still move past after this
4000 * test, but as we are holding a reference to
4001 * 'sh', we know that if that happens,
4002 * STRIPE_EXPANDING will get set and the expansion
4003 * won't proceed until we finish with the stripe.
7ecaa1e6
N		 4004 */
4005 int must_retry = 0;
4006 spin_lock_irq(&conf->device_lock);
b0f9ec04
N		 4007 if (mddev->delta_disks < 0
4008 ? logical_sector >= conf->reshape_progress
4009 : logical_sector < conf->reshape_progress)
7ecaa1e6
N		 4010 /* mismatch, need to try again */
4011 must_retry = 1;
4012 spin_unlock_irq(&conf->device_lock);
4013 if (must_retry) {
4014 release_stripe(sh);
7a3ab908 4015 schedule();
7ecaa1e6
N		 4016 goto retry;
4017 }
4018 }
e62e58a5 4019
ffd96e35 4020 if (rw == WRITE &&
a5c308d4 4021 logical_sector >= mddev->suspend_lo &&
e464eafd
N		 4022 logical_sector < mddev->suspend_hi) {
4023 release_stripe(sh);
e62e58a5
N		 4024 /* As the suspend_* range is controlled by
4025 * userspace, we want an interruptible
4026 * wait.
4027 */
4028 flush_signals(current);
4029 prepare_to_wait(&conf->wait_for_overlap,
4030 &w, TASK_INTERRUPTIBLE);
4031 if (logical_sector >= mddev->suspend_lo &&
4032 logical_sector < mddev->suspend_hi)
4033 schedule();
e464eafd
N		 4034 goto retry;
4035 }
7ecaa1e6
N		 4036
4037 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
ffd96e35 4038 !add_stripe_bio(sh, bi, dd_idx, rw)) {
7ecaa1e6
N		 4039 /* Stripe is busy expanding or
4040 * add failed due to overlap. Flush everything
1da177e4
LT		 4041 * and wait a while
4042 */
482c0834 4043 md_wakeup_thread(mddev->thread);
1da177e4
LT		 4044 release_stripe(sh);
4045 schedule();
4046 goto retry;
4047 }
4048 finish_wait(&conf->wait_for_overlap, &w);
6ed3003c
N		 4049 set_bit(STRIPE_HANDLE, &sh->state);
4050 clear_bit(STRIPE_DELAYED, &sh->state);
e9c7469b 4051 if ((bi->bi_rw & REQ_SYNC) &&
729a1866
N		 4052 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4053 atomic_inc(&conf->preread_active_stripes);
1da177e4 4054 release_stripe(sh);
1da177e4
LT		 4055 } else {
4056 /* cannot get stripe for read-ahead, just give-up */
4057 clear_bit(BIO_UPTODATE, &bi->bi_flags);
4058 finish_wait(&conf->wait_for_overlap, &w);
4059 break;
4060 }
4061
4062 }
7c13edc8
N		 4063 if (!plugged)
4064 md_wakeup_thread(mddev->thread);
4065
1da177e4 4066 spin_lock_irq(&conf->device_lock);
960e739d 4067 remaining = raid5_dec_bi_phys_segments(bi);
f6344757
N		 4068 spin_unlock_irq(&conf->device_lock);
4069 if (remaining == 0) {
1da177e4 4070
16a53ecc 4071 if ( rw == WRITE )
1da177e4 4072 md_write_end(mddev);
6712ecf8 4073
0e13fe23 4074 bio_endio(bi, 0);
1da177e4 4075 }
729a1866 4076
1da177e4
LT		 4077 return 0;
4078}
4079
b522adcd
DW		 4080static sector_t raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks);
4081
52c03291 4082static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped)
1da177e4 4083{
52c03291
N		 4084 /* reshaping is quite different to recovery/resync so it is
4085 * handled quite separately ... here.
4086 *
4087 * On each call to sync_request, we gather one chunk worth of
4088 * destination stripes and flag them as expanding.
4089 * Then we find all the source stripes and request reads.
4090 * As the reads complete, handle_stripe will copy the data
4091 * into the destination stripe and release that stripe.
4092 */
7b92813c 4093 raid5_conf_t *conf = mddev->private;
1da177e4 4094 struct stripe_head *sh;
ccfcc3c1 4095 sector_t first_sector, last_sector;
f416885e
N		 4096 int raid_disks = conf->previous_raid_disks;
4097 int data_disks = raid_disks - conf->max_degraded;
4098 int new_data_disks = conf->raid_disks - conf->max_degraded;
52c03291
N		 4099 int i;
4100 int dd_idx;
c8f517c4 4101 sector_t writepos, readpos, safepos;
ec32a2bd 4102 sector_t stripe_addr;
7a661381 4103 int reshape_sectors;
ab69ae12 4104 struct list_head stripes;
52c03291 4105
fef9c61f
N		 4106 if (sector_nr == 0) {
4107 /* If restarting in the middle, skip the initial sectors */
4108 if (mddev->delta_disks < 0 &&
4109 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
4110 sector_nr = raid5_size(mddev, 0, 0)
4111 - conf->reshape_progress;
a639755c 4112 } else if (mddev->delta_disks >= 0 &&
fef9c61f
N		 4113 conf->reshape_progress > 0)
4114 sector_nr = conf->reshape_progress;
f416885e 4115 sector_div(sector_nr, new_data_disks);
fef9c61f 4116 if (sector_nr) {
8dee7211
N		 4117 mddev->curr_resync_completed = sector_nr;
4118 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
fef9c61f
N		 4119 *skipped = 1;
4120 return sector_nr;
4121 }
52c03291
N		 4122 }
4123
7a661381
N		 4124 /* We need to process a full chunk at a time.
4125 * If old and new chunk sizes differ, we need to process the
4126 * largest of these
4127 */
664e7c41
AN		 4128 if (mddev->new_chunk_sectors > mddev->chunk_sectors)
4129 reshape_sectors = mddev->new_chunk_sectors;
7a661381 4130 else
9d8f0363 4131 reshape_sectors = mddev->chunk_sectors;
7a661381 4132
52c03291
N		 4133 /* we update the metadata when there is more than 3Meg
4134 * in the block range (that is rather arbitrary, should
4135 * probably be time based) or when the data about to be
4136 * copied would over-write the source of the data at
4137 * the front of the range.
fef9c61f
N		 4138 * i.e. one new_stripe along from reshape_progress new_maps
4139 * to after where reshape_safe old_maps to
52c03291 4140 */
fef9c61f 4141 writepos = conf->reshape_progress;
f416885e 4142 sector_div(writepos, new_data_disks);
c8f517c4
N		 4143 readpos = conf->reshape_progress;
4144 sector_div(readpos, data_disks);
fef9c61f 4145 safepos = conf->reshape_safe;
f416885e 4146 sector_div(safepos, data_disks);
fef9c61f 4147 if (mddev->delta_disks < 0) {
ed37d83e 4148 writepos -= min_t(sector_t, reshape_sectors, writepos);
c8f517c4 4149 readpos += reshape_sectors;
7a661381 4150 safepos += reshape_sectors;
fef9c61f 4151 } else {
7a661381 4152 writepos += reshape_sectors;
ed37d83e
N		 4153 readpos -= min_t(sector_t, reshape_sectors, readpos);
4154 safepos -= min_t(sector_t, reshape_sectors, safepos);
fef9c61f 4155 }
52c03291 4156
c8f517c4
N		 4157 /* 'writepos' is the most advanced device address we might write.
4158 * 'readpos' is the least advanced device address we might read.
4159 * 'safepos' is the least address recorded in the metadata as having
4160 * been reshaped.
4161 * If 'readpos' is behind 'writepos', then there is no way that we can
4162 * ensure safety in the face of a crash - that must be done by userspace
4163 * making a backup of the data. So in that case there is no particular
4164 * rush to update metadata.
4165 * Otherwise if 'safepos' is behind 'writepos', then we really need to
4166 * update the metadata to advance 'safepos' to match 'readpos' so that
4167 * we can be safe in the event of a crash.
4168 * So we insist on updating metadata if safepos is behind writepos and
4169 * readpos is beyond writepos.
4170 * In any case, update the metadata every 10 seconds.
4171 * Maybe that number should be configurable, but I'm not sure it is
4172 * worth it.... maybe it could be a multiple of safemode_delay???
4173 */
fef9c61f 4174 if ((mddev->delta_disks < 0
c8f517c4
N		 4175 ? (safepos > writepos && readpos < writepos)
4176 : (safepos < writepos && readpos > writepos)) ||
4177 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
52c03291
N		 4178 /* Cannot proceed until we've updated the superblock... */
4179 wait_event(conf->wait_for_overlap,
4180 atomic_read(&conf->reshape_stripes)==0);
fef9c61f 4181 mddev->reshape_position = conf->reshape_progress;
75d3da43 4182 mddev->curr_resync_completed = sector_nr;
c8f517c4 4183 conf->reshape_checkpoint = jiffies;
850b2b42 4184 set_bit(MD_CHANGE_DEVS, &mddev->flags);
52c03291 4185 md_wakeup_thread(mddev->thread);
850b2b42 4186 wait_event(mddev->sb_wait, mddev->flags == 0 ||
52c03291
N		 4187 kthread_should_stop());
4188 spin_lock_irq(&conf->device_lock);
fef9c61f 4189 conf->reshape_safe = mddev->reshape_position;
52c03291
N		 4190 spin_unlock_irq(&conf->device_lock);
4191 wake_up(&conf->wait_for_overlap);
acb180b0 4192 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
52c03291
N		 4193 }
4194
ec32a2bd
N		 4195 if (mddev->delta_disks < 0) {
4196 BUG_ON(conf->reshape_progress == 0);
4197 stripe_addr = writepos;
4198 BUG_ON((mddev->dev_sectors &
7a661381
N		 4199 ~((sector_t)reshape_sectors - 1))
4200 - reshape_sectors - stripe_addr
ec32a2bd
N		 4201 != sector_nr);
4202 } else {
7a661381 4203 BUG_ON(writepos != sector_nr + reshape_sectors);
ec32a2bd
N		 4204 stripe_addr = sector_nr;
4205 }
ab69ae12 4206 INIT_LIST_HEAD(&stripes);
7a661381 4207 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
52c03291 4208 int j;
a9f326eb 4209 int skipped_disk = 0;
a8c906ca 4210 sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
52c03291
N		 4211 set_bit(STRIPE_EXPANDING, &sh->state);
4212 atomic_inc(&conf->reshape_stripes);
4213 /* If any of this stripe is beyond the end of the old
4214 * array, then we need to zero those blocks
4215 */
4216 for (j=sh->disks; j--;) {
4217 sector_t s;
4218 if (j == sh->pd_idx)
4219 continue;
f416885e 4220 if (conf->level == 6 &&
d0dabf7e 4221 j == sh->qd_idx)
f416885e 4222 continue;
784052ec 4223 s = compute_blocknr(sh, j, 0);
b522adcd 4224 if (s < raid5_size(mddev, 0, 0)) {
a9f326eb 4225 skipped_disk = 1;
52c03291
N		 4226 continue;
4227 }
4228 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
4229 set_bit(R5_Expanded, &sh->dev[j].flags);
4230 set_bit(R5_UPTODATE, &sh->dev[j].flags);
4231 }
a9f326eb 4232 if (!skipped_disk) {
52c03291
N		 4233 set_bit(STRIPE_EXPAND_READY, &sh->state);
4234 set_bit(STRIPE_HANDLE, &sh->state);
4235 }
ab69ae12 4236 list_add(&sh->lru, &stripes);
52c03291
N		 4237 }
4238 spin_lock_irq(&conf->device_lock);
fef9c61f 4239 if (mddev->delta_disks < 0)
7a661381 4240 conf->reshape_progress -= reshape_sectors * new_data_disks;
fef9c61f 4241 else
7a661381 4242 conf->reshape_progress += reshape_sectors * new_data_disks;
52c03291
N		 4243 spin_unlock_irq(&conf->device_lock);
4244 /* Ok, those stripe are ready. We can start scheduling
4245 * reads on the source stripes.
4246 * The source stripes are determined by mapping the first and last
4247 * block on the destination stripes.
4248 */
52c03291 4249 first_sector =
ec32a2bd 4250 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
911d4ee8 4251 1, &dd_idx, NULL);
52c03291 4252 last_sector =
0e6e0271 4253 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
09c9e5fa 4254 * new_data_disks - 1),
911d4ee8 4255 1, &dd_idx, NULL);
58c0fed4
AN		 4256 if (last_sector >= mddev->dev_sectors)
4257 last_sector = mddev->dev_sectors - 1;
52c03291 4258 while (first_sector <= last_sector) {
a8c906ca 4259 sh = get_active_stripe(conf, first_sector, 1, 0, 1);
52c03291
N		 4260 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
4261 set_bit(STRIPE_HANDLE, &sh->state);
4262 release_stripe(sh);
4263 first_sector += STRIPE_SECTORS;
4264 }
ab69ae12
N		 4265 /* Now that the sources are clearly marked, we can release
4266 * the destination stripes
4267 */
4268 while (!list_empty(&stripes)) {
4269 sh = list_entry(stripes.next, struct stripe_head, lru);
4270 list_del_init(&sh->lru);
4271 release_stripe(sh);
4272 }
c6207277
N		 4273 /* If this takes us to the resync_max point where we have to pause,
4274 * then we need to write out the superblock.
4275 */
7a661381 4276 sector_nr += reshape_sectors;
c03f6a19
N		 4277 if ((sector_nr - mddev->curr_resync_completed) * 2
4278 >= mddev->resync_max - mddev->curr_resync_completed) {
c6207277
N		 4279 /* Cannot proceed until we've updated the superblock... */
4280 wait_event(conf->wait_for_overlap,
4281 atomic_read(&conf->reshape_stripes) == 0);
fef9c61f 4282 mddev->reshape_position = conf->reshape_progress;
75d3da43 4283 mddev->curr_resync_completed = sector_nr;
c8f517c4 4284 conf->reshape_checkpoint = jiffies;
c6207277
N		 4285 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4286 md_wakeup_thread(mddev->thread);
4287 wait_event(mddev->sb_wait,
4288 !test_bit(MD_CHANGE_DEVS, &mddev->flags)
4289 || kthread_should_stop());
4290 spin_lock_irq(&conf->device_lock);
fef9c61f 4291 conf->reshape_safe = mddev->reshape_position;
c6207277
N		 4292 spin_unlock_irq(&conf->device_lock);
4293 wake_up(&conf->wait_for_overlap);
acb180b0 4294 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
c6207277 4295 }
7a661381 4296 return reshape_sectors;
52c03291
N		 4297}
4298
4299/* FIXME go_faster isn't used */
4300static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
4301{
7b92813c 4302 raid5_conf_t *conf = mddev->private;
52c03291 4303 struct stripe_head *sh;
58c0fed4 4304 sector_t max_sector = mddev->dev_sectors;
57dab0bd 4305 sector_t sync_blocks;
16a53ecc
N		 4306 int still_degraded = 0;
4307 int i;
1da177e4 4308
72626685 4309 if (sector_nr >= max_sector) {
1da177e4 4310 /* just being told to finish up .. nothing much to do */
cea9c228 4311
29269553
N		 4312 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
4313 end_reshape(conf);
4314 return 0;
4315 }
72626685
N		 4316
4317 if (mddev->curr_resync < max_sector) /* aborted */
4318 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
4319 &sync_blocks, 1);
16a53ecc 4320 else /* completed sync */
72626685
N		 4321 conf->fullsync = 0;
4322 bitmap_close_sync(mddev->bitmap);
4323
1da177e4
LT		 4324 return 0;
4325 }
ccfcc3c1 4326
64bd660b
N		 4327 /* Allow raid5_quiesce to complete */
4328 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
4329
52c03291
N		 4330 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4331 return reshape_request(mddev, sector_nr, skipped);
f6705578 4332
c6207277
N		 4333 /* No need to check resync_max as we never do more than one
4334 * stripe, and as resync_max will always be on a chunk boundary,
4335 * if the check in md_do_sync didn't fire, there is no chance
4336 * of overstepping resync_max here
4337 */
4338
16a53ecc 4339 /* if there is too many failed drives and we are trying
1da177e4
LT		 4340 * to resync, then assert that we are finished, because there is
4341 * nothing we can do.
4342 */
3285edf1 4343 if (mddev->degraded >= conf->max_degraded &&
16a53ecc 4344 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
58c0fed4 4345 sector_t rv = mddev->dev_sectors - sector_nr;
57afd89f 4346 *skipped = 1;
1da177e4
LT		 4347 return rv;
4348 }
72626685 4349 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
3855ad9f 4350 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
72626685
N		 4351 !conf->fullsync && sync_blocks >= STRIPE_SECTORS) {
4352 /* we can skip this block, and probably more */
4353 sync_blocks /= STRIPE_SECTORS;
4354 *skipped = 1;
4355 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
4356 }
1da177e4 4357
b47490c9
N		 4358
4359 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
4360
a8c906ca 4361 sh = get_active_stripe(conf, sector_nr, 0, 1, 0);
1da177e4 4362 if (sh == NULL) {
a8c906ca 4363 sh = get_active_stripe(conf, sector_nr, 0, 0, 0);
1da177e4 4364 /* make sure we don't swamp the stripe cache if someone else
16a53ecc 4365 * is trying to get access
1da177e4 4366 */
66c006a5 4367 schedule_timeout_uninterruptible(1);
1da177e4 4368 }
16a53ecc
N		 4369 /* Need to check if array will still be degraded after recovery/resync
4370 * We don't need to check the 'failed' flag as when that gets set,
4371 * recovery aborts.
4372 */
f001a70c 4373 for (i = 0; i < conf->raid_disks; i++)
16a53ecc
N		 4374 if (conf->disks[i].rdev == NULL)
4375 still_degraded = 1;
4376
4377 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
4378
83206d66 4379 set_bit(STRIPE_SYNC_REQUESTED, &sh->state);
1da177e4 4380
1442577b 4381 handle_stripe(sh);
1da177e4
LT		 4382 release_stripe(sh);
4383
4384 return STRIPE_SECTORS;
4385}
4386
46031f9a
RBJ		 4387static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio)
4388{
4389 /* We may not be able to submit a whole bio at once as there
4390 * may not be enough stripe_heads available.
4391 * We cannot pre-allocate enough stripe_heads as we may need
4392 * more than exist in the cache (if we allow ever large chunks).
4393 * So we do one stripe head at a time and record in
4394 * ->bi_hw_segments how many have been done.
4395 *
4396 * We *know* that this entire raid_bio is in one chunk, so
4397 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
4398 */
4399 struct stripe_head *sh;
911d4ee8 4400 int dd_idx;
46031f9a
RBJ		 4401 sector_t sector, logical_sector, last_sector;
4402 int scnt = 0;
4403 int remaining;
4404 int handled = 0;
4405
4406 logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
112bf897 4407 sector = raid5_compute_sector(conf, logical_sector,
911d4ee8 4408 0, &dd_idx, NULL);
46031f9a
RBJ		 4409 last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9);
4410
4411 for (; logical_sector < last_sector;
387bb173
NB		 4412 logical_sector += STRIPE_SECTORS,
4413 sector += STRIPE_SECTORS,
4414 scnt++) {
46031f9a 4415
960e739d 4416 if (scnt < raid5_bi_hw_segments(raid_bio))
46031f9a
RBJ		 4417 /* already done this stripe */
4418 continue;
4419
a8c906ca 4420 sh = get_active_stripe(conf, sector, 0, 1, 0);
46031f9a
RBJ		 4421
4422 if (!sh) {
4423 /* failed to get a stripe - must wait */
960e739d 4424 raid5_set_bi_hw_segments(raid_bio, scnt);
46031f9a
RBJ		 4425 conf->retry_read_aligned = raid_bio;
4426 return handled;
4427 }
4428
4429 set_bit(R5_ReadError, &sh->dev[dd_idx].flags);
387bb173
NB		 4430 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) {
4431 release_stripe(sh);
960e739d 4432 raid5_set_bi_hw_segments(raid_bio, scnt);
387bb173
NB		 4433 conf->retry_read_aligned = raid_bio;
4434 return handled;
4435 }
4436
36d1c647 4437 handle_stripe(sh);
46031f9a
RBJ		 4438 release_stripe(sh);
4439 handled++;
4440 }
4441 spin_lock_irq(&conf->device_lock);
960e739d 4442 remaining = raid5_dec_bi_phys_segments(raid_bio);
46031f9a 4443 spin_unlock_irq(&conf->device_lock);
0e13fe23
NB		 4444 if (remaining == 0)
4445 bio_endio(raid_bio, 0);
46031f9a
RBJ		 4446 if (atomic_dec_and_test(&conf->active_aligned_reads))
4447 wake_up(&conf->wait_for_stripe);
4448 return handled;
4449}
4450
46031f9a 4451
1da177e4
LT		 4452/*
4453 * This is our raid5 kernel thread.
4454 *
4455 * We scan the hash table for stripes which can be handled now.
4456 * During the scan, completed stripes are saved for us by the interrupt
4457 * handler, so that they will not have to wait for our next wakeup.
4458 */
6ed3003c 4459static void raid5d(mddev_t *mddev)
1da177e4
LT		 4460{
4461 struct stripe_head *sh;
070ec55d 4462 raid5_conf_t *conf = mddev->private;
1da177e4 4463 int handled;
e1dfa0a2 4464 struct blk_plug plug;
1da177e4 4465
45b4233c 4466 pr_debug("+++ raid5d active\n");
1da177e4
LT		 4467
4468 md_check_recovery(mddev);
1da177e4 4469
e1dfa0a2 4470 blk_start_plug(&plug);
1da177e4
LT		 4471 handled = 0;
4472 spin_lock_irq(&conf->device_lock);
4473 while (1) {
46031f9a 4474 struct bio *bio;
1da177e4 4475
7c13edc8
N		 4476 if (atomic_read(&mddev->plug_cnt) == 0 &&
4477 !list_empty(&conf->bitmap_list)) {
4478 /* Now is a good time to flush some bitmap updates */
4479 conf->seq_flush++;
700e432d 4480 spin_unlock_irq(&conf->device_lock);
72626685 4481 bitmap_unplug(mddev->bitmap);
700e432d 4482 spin_lock_irq(&conf->device_lock);
7c13edc8 4483 conf->seq_write = conf->seq_flush;
72626685
N		 4484 activate_bit_delay(conf);
4485 }
7c13edc8
N		 4486 if (atomic_read(&mddev->plug_cnt) == 0)
4487 raid5_activate_delayed(conf);
72626685 4488
46031f9a
RBJ		 4489 while ((bio = remove_bio_from_retry(conf))) {
4490 int ok;
4491 spin_unlock_irq(&conf->device_lock);
4492 ok = retry_aligned_read(conf, bio);
4493 spin_lock_irq(&conf->device_lock);
4494 if (!ok)
4495 break;
4496 handled++;
4497 }
4498
8b3e6cdc
DW		 4499 sh = __get_priority_stripe(conf);
4500
c9f21aaf 4501 if (!sh)
1da177e4 4502 break;
1da177e4
LT		 4503 spin_unlock_irq(&conf->device_lock);
4504
4505 handled++;
417b8d4a
DW		 4506 handle_stripe(sh);
4507 release_stripe(sh);
4508 cond_resched();
1da177e4
LT		 4509
4510 spin_lock_irq(&conf->device_lock);
4511 }
45b4233c 4512 pr_debug("%d stripes handled\n", handled);
1da177e4
LT		 4513
4514 spin_unlock_irq(&conf->device_lock);
4515
c9f21aaf 4516 async_tx_issue_pending_all();
e1dfa0a2 4517 blk_finish_plug(&plug);
1da177e4 4518
45b4233c 4519 pr_debug("--- raid5d inactive\n");
1da177e4
LT		 4520}
4521
3f294f4f 4522static ssize_t
007583c9 4523raid5_show_stripe_cache_size(mddev_t *mddev, char *page)
3f294f4f 4524{
070ec55d 4525 raid5_conf_t *conf = mddev->private;
96de1e66
N		 4526 if (conf)
4527 return sprintf(page, "%d\n", conf->max_nr_stripes);
4528 else
4529 return 0;
3f294f4f
N		 4530}
4531
c41d4ac4
N		 4532int
4533raid5_set_cache_size(mddev_t *mddev, int size)
3f294f4f 4534{
070ec55d 4535 raid5_conf_t *conf = mddev->private;
b5470dc5
DW		 4536 int err;
4537
c41d4ac4 4538 if (size <= 16 || size > 32768)
3f294f4f 4539 return -EINVAL;
c41d4ac4 4540 while (size < conf->max_nr_stripes) {
3f294f4f
N		 4541 if (drop_one_stripe(conf))
4542 conf->max_nr_stripes--;
4543 else
4544 break;
4545 }
b5470dc5
DW		 4546 err = md_allow_write(mddev);
4547 if (err)
4548 return err;
c41d4ac4 4549 while (size > conf->max_nr_stripes) {
3f294f4f
N		 4550 if (grow_one_stripe(conf))
4551 conf->max_nr_stripes++;
4552 else break;
4553 }
c41d4ac4
N		 4554 return 0;
4555}
4556EXPORT_SYMBOL(raid5_set_cache_size);
4557
4558static ssize_t
4559raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len)
4560{
4561 raid5_conf_t *conf = mddev->private;
4562 unsigned long new;
4563 int err;
4564
4565 if (len >= PAGE_SIZE)
4566 return -EINVAL;
4567 if (!conf)
4568 return -ENODEV;
4569
4570 if (strict_strtoul(page, 10, &new))
4571 return -EINVAL;
4572 err = raid5_set_cache_size(mddev, new);
4573 if (err)
4574 return err;
3f294f4f
N		 4575 return len;
4576}
007583c9 4577
96de1e66
N		 4578static struct md_sysfs_entry
4579raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
4580 raid5_show_stripe_cache_size,
4581 raid5_store_stripe_cache_size);
3f294f4f 4582
8b3e6cdc
DW		 4583static ssize_t
4584raid5_show_preread_threshold(mddev_t *mddev, char *page)
4585{
070ec55d 4586 raid5_conf_t *conf = mddev->private;
8b3e6cdc
DW		 4587 if (conf)
4588 return sprintf(page, "%d\n", conf->bypass_threshold);
4589 else
4590 return 0;
4591}
4592
4593static ssize_t
4594raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len)
4595{
070ec55d 4596 raid5_conf_t *conf = mddev->private;
4ef197d8 4597 unsigned long new;
8b3e6cdc
DW		 4598 if (len >= PAGE_SIZE)
4599 return -EINVAL;
4600 if (!conf)
4601 return -ENODEV;
4602
4ef197d8 4603 if (strict_strtoul(page, 10, &new))
8b3e6cdc 4604 return -EINVAL;
4ef197d8 4605 if (new > conf->max_nr_stripes)
8b3e6cdc
DW		 4606 return -EINVAL;
4607 conf->bypass_threshold = new;
4608 return len;
4609}
4610
4611static struct md_sysfs_entry
4612raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
4613 S_IRUGO | S_IWUSR,
4614 raid5_show_preread_threshold,
4615 raid5_store_preread_threshold);
4616
3f294f4f 4617static ssize_t
96de1e66 4618stripe_cache_active_show(mddev_t *mddev, char *page)
3f294f4f 4619{
070ec55d 4620 raid5_conf_t *conf = mddev->private;
96de1e66
N		 4621 if (conf)
4622 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
4623 else
4624 return 0;
3f294f4f
N		 4625}
4626
96de1e66
N		 4627static struct md_sysfs_entry
4628raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
3f294f4f 4629
007583c9 4630static struct attribute *raid5_attrs[] = {
3f294f4f
N		 4631 &raid5_stripecache_size.attr,
4632 &raid5_stripecache_active.attr,
8b3e6cdc 4633 &raid5_preread_bypass_threshold.attr,
3f294f4f
N		 4634 NULL,
4635};
007583c9
N		 4636static struct attribute_group raid5_attrs_group = {
4637 .name = NULL,
4638 .attrs = raid5_attrs,
3f294f4f
N		 4639};
4640
80c3a6ce
DW		 4641static sector_t
4642raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks)
4643{
070ec55d 4644 raid5_conf_t *conf = mddev->private;
80c3a6ce
DW		 4645
4646 if (!sectors)
4647 sectors = mddev->dev_sectors;
5e5e3e78 4648 if (!raid_disks)
7ec05478 4649 /* size is defined by the smallest of previous and new size */
5e5e3e78 4650 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
80c3a6ce 4651
9d8f0363 4652 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
664e7c41 4653 sectors &= ~((sector_t)mddev->new_chunk_sectors - 1);
80c3a6ce
DW		 4654 return sectors * (raid_disks - conf->max_degraded);
4655}
4656
36d1c647
DW		 4657static void raid5_free_percpu(raid5_conf_t *conf)
4658{
4659 struct raid5_percpu *percpu;
4660 unsigned long cpu;
4661
4662 if (!conf->percpu)
4663 return;
4664
4665 get_online_cpus();
4666 for_each_possible_cpu(cpu) {
4667 percpu = per_cpu_ptr(conf->percpu, cpu);
4668 safe_put_page(percpu->spare_page);
d6f38f31 4669 kfree(percpu->scribble);
36d1c647
DW		 4670 }
4671#ifdef CONFIG_HOTPLUG_CPU
4672 unregister_cpu_notifier(&conf->cpu_notify);
4673#endif
4674 put_online_cpus();
4675
4676 free_percpu(conf->percpu);
4677}
4678
95fc17aa
DW		 4679static void free_conf(raid5_conf_t *conf)
4680{
4681 shrink_stripes(conf);
36d1c647 4682 raid5_free_percpu(conf);
95fc17aa
DW		 4683 kfree(conf->disks);
4684 kfree(conf->stripe_hashtbl);
4685 kfree(conf);
4686}
4687
36d1c647
DW		 4688#ifdef CONFIG_HOTPLUG_CPU
4689static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action,
4690 void *hcpu)
4691{
4692 raid5_conf_t *conf = container_of(nfb, raid5_conf_t, cpu_notify);
4693 long cpu = (long)hcpu;
4694 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
4695
4696 switch (action) {
4697 case CPU_UP_PREPARE:
4698 case CPU_UP_PREPARE_FROZEN:
d6f38f31 4699 if (conf->level == 6 && !percpu->spare_page)
36d1c647 4700 percpu->spare_page = alloc_page(GFP_KERNEL);
d6f38f31
DW		 4701 if (!percpu->scribble)
4702 percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
4703
4704 if (!percpu->scribble ||
4705 (conf->level == 6 && !percpu->spare_page)) {
4706 safe_put_page(percpu->spare_page);
4707 kfree(percpu->scribble);
36d1c647
DW		 4708 pr_err("%s: failed memory allocation for cpu%ld\n",
4709 __func__, cpu);
55af6bb5 4710 return notifier_from_errno(-ENOMEM);
36d1c647
DW		 4711 }
4712 break;
4713 case CPU_DEAD:
4714 case CPU_DEAD_FROZEN:
4715 safe_put_page(percpu->spare_page);
d6f38f31 4716 kfree(percpu->scribble);
36d1c647 4717 percpu->spare_page = NULL;
d6f38f31 4718 percpu->scribble = NULL;
36d1c647
DW		 4719 break;
4720 default:
4721 break;
4722 }
4723 return NOTIFY_OK;
4724}
4725#endif
4726
4727static int raid5_alloc_percpu(raid5_conf_t *conf)
4728{
4729 unsigned long cpu;
4730 struct page *spare_page;
a29d8b8e 4731 struct raid5_percpu __percpu *allcpus;
d6f38f31 4732 void *scribble;
36d1c647
DW		 4733 int err;
4734
36d1c647
DW		 4735 allcpus = alloc_percpu(struct raid5_percpu);
4736 if (!allcpus)
4737 return -ENOMEM;
4738 conf->percpu = allcpus;
4739
4740 get_online_cpus();
4741 err = 0;
4742 for_each_present_cpu(cpu) {
d6f38f31
DW		 4743 if (conf->level == 6) {
4744 spare_page = alloc_page(GFP_KERNEL);
4745 if (!spare_page) {
4746 err = -ENOMEM;
4747 break;
4748 }
4749 per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page;
4750 }
5e5e3e78 4751 scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
d6f38f31 4752 if (!scribble) {
36d1c647
DW		 4753 err = -ENOMEM;
4754 break;
4755 }
d6f38f31 4756 per_cpu_ptr(conf->percpu, cpu)->scribble = scribble;
36d1c647
DW		 4757 }
4758#ifdef CONFIG_HOTPLUG_CPU
4759 conf->cpu_notify.notifier_call = raid456_cpu_notify;
4760 conf->cpu_notify.priority = 0;
4761 if (err == 0)
4762 err = register_cpu_notifier(&conf->cpu_notify);
4763#endif
4764 put_online_cpus();
4765
4766 return err;
4767}
4768
91adb564 4769static raid5_conf_t *setup_conf(mddev_t *mddev)
1da177e4
LT		 4770{
4771 raid5_conf_t *conf;
5e5e3e78 4772 int raid_disk, memory, max_disks;
1da177e4
LT		 4773 mdk_rdev_t *rdev;
4774 struct disk_info *disk;
1da177e4 4775
91adb564
N		 4776 if (mddev->new_level != 5
4777 && mddev->new_level != 4
4778 && mddev->new_level != 6) {
0c55e022 4779 printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n",
91adb564
N		 4780 mdname(mddev), mddev->new_level);
4781 return ERR_PTR(-EIO);
1da177e4 4782 }
91adb564
N		 4783 if ((mddev->new_level == 5
4784 && !algorithm_valid_raid5(mddev->new_layout)) ||
4785 (mddev->new_level == 6
4786 && !algorithm_valid_raid6(mddev->new_layout))) {
0c55e022 4787 printk(KERN_ERR "md/raid:%s: layout %d not supported\n",
91adb564
N		 4788 mdname(mddev), mddev->new_layout);
4789 return ERR_PTR(-EIO);
99c0fb5f 4790 }
91adb564 4791 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
0c55e022 4792 printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n",
91adb564
N		 4793 mdname(mddev), mddev->raid_disks);
4794 return ERR_PTR(-EINVAL);
4bbf3771
N		 4795 }
4796
664e7c41
AN		 4797 if (!mddev->new_chunk_sectors ||
4798 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
4799 !is_power_of_2(mddev->new_chunk_sectors)) {
0c55e022
N		 4800 printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n",
4801 mdname(mddev), mddev->new_chunk_sectors << 9);
91adb564 4802 return ERR_PTR(-EINVAL);
f6705578
N		 4803 }
4804
91adb564
N		 4805 conf = kzalloc(sizeof(raid5_conf_t), GFP_KERNEL);
4806 if (conf == NULL)
1da177e4 4807 goto abort;
f5efd45a
DW		 4808 spin_lock_init(&conf->device_lock);
4809 init_waitqueue_head(&conf->wait_for_stripe);
4810 init_waitqueue_head(&conf->wait_for_overlap);
4811 INIT_LIST_HEAD(&conf->handle_list);
4812 INIT_LIST_HEAD(&conf->hold_list);
4813 INIT_LIST_HEAD(&conf->delayed_list);
4814 INIT_LIST_HEAD(&conf->bitmap_list);
4815 INIT_LIST_HEAD(&conf->inactive_list);
4816 atomic_set(&conf->active_stripes, 0);
4817 atomic_set(&conf->preread_active_stripes, 0);
4818 atomic_set(&conf->active_aligned_reads, 0);
4819 conf->bypass_threshold = BYPASS_THRESHOLD;
91adb564
N		 4820
4821 conf->raid_disks = mddev->raid_disks;
4822 if (mddev->reshape_position == MaxSector)
4823 conf->previous_raid_disks = mddev->raid_disks;
4824 else
f6705578 4825 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
5e5e3e78
N		 4826 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
4827 conf->scribble_len = scribble_len(max_disks);
f6705578 4828
5e5e3e78 4829 conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
b55e6bfc
N		 4830 GFP_KERNEL);
4831 if (!conf->disks)
4832 goto abort;
9ffae0cf 4833
1da177e4
LT		 4834 conf->mddev = mddev;
4835
fccddba0 4836 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
1da177e4 4837 goto abort;
1da177e4 4838
36d1c647
DW		 4839 conf->level = mddev->new_level;
4840 if (raid5_alloc_percpu(conf) != 0)
4841 goto abort;
4842
0c55e022 4843 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
1da177e4 4844
159ec1fc 4845 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4 4846 raid_disk = rdev->raid_disk;
5e5e3e78 4847 if (raid_disk >= max_disks
1da177e4
LT		 4848 || raid_disk < 0)
4849 continue;
4850 disk = conf->disks + raid_disk;
4851
4852 disk->rdev = rdev;
4853
b2d444d7 4854 if (test_bit(In_sync, &rdev->flags)) {
1da177e4 4855 char b[BDEVNAME_SIZE];
0c55e022
N		 4856 printk(KERN_INFO "md/raid:%s: device %s operational as raid"
4857 " disk %d\n",
4858 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
d6b212f4 4859 } else if (rdev->saved_raid_disk != raid_disk)
8c2e870a
NB		 4860 /* Cannot rely on bitmap to complete recovery */
4861 conf->fullsync = 1;
1da177e4
LT		 4862 }
4863
09c9e5fa 4864 conf->chunk_sectors = mddev->new_chunk_sectors;
91adb564 4865 conf->level = mddev->new_level;
16a53ecc
N		 4866 if (conf->level == 6)
4867 conf->max_degraded = 2;
4868 else
4869 conf->max_degraded = 1;
91adb564 4870 conf->algorithm = mddev->new_layout;
1da177e4 4871 conf->max_nr_stripes = NR_STRIPES;
fef9c61f 4872 conf->reshape_progress = mddev->reshape_position;
e183eaed 4873 if (conf->reshape_progress != MaxSector) {
09c9e5fa 4874 conf->prev_chunk_sectors = mddev->chunk_sectors;
e183eaed
N		 4875 conf->prev_algo = mddev->layout;
4876 }
1da177e4 4877
91adb564 4878 memory = conf->max_nr_stripes * (sizeof(struct stripe_head) +
5e5e3e78 4879 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
91adb564
N		 4880 if (grow_stripes(conf, conf->max_nr_stripes)) {
4881 printk(KERN_ERR
0c55e022
N		 4882 "md/raid:%s: couldn't allocate %dkB for buffers\n",
4883 mdname(mddev), memory);
91adb564
N		 4884 goto abort;
4885 } else
0c55e022
N		 4886 printk(KERN_INFO "md/raid:%s: allocated %dkB\n",
4887 mdname(mddev), memory);
1da177e4 4888
0da3c619 4889 conf->thread = md_register_thread(raid5d, mddev, NULL);
91adb564
N		 4890 if (!conf->thread) {
4891 printk(KERN_ERR
0c55e022 4892 "md/raid:%s: couldn't allocate thread.\n",
91adb564 4893 mdname(mddev));
16a53ecc
N		 4894 goto abort;
4895 }
91adb564
N		 4896
4897 return conf;
4898
4899 abort:
4900 if (conf) {
95fc17aa 4901 free_conf(conf);
91adb564
N		 4902 return ERR_PTR(-EIO);
4903 } else
4904 return ERR_PTR(-ENOMEM);
4905}
4906
c148ffdc
N		 4907
4908static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
4909{
4910 switch (algo) {
4911 case ALGORITHM_PARITY_0:
4912 if (raid_disk < max_degraded)
4913 return 1;
4914 break;
4915 case ALGORITHM_PARITY_N:
4916 if (raid_disk >= raid_disks - max_degraded)
4917 return 1;
4918 break;
4919 case ALGORITHM_PARITY_0_6:
4920 if (raid_disk == 0 ||
4921 raid_disk == raid_disks - 1)
4922 return 1;
4923 break;
4924 case ALGORITHM_LEFT_ASYMMETRIC_6:
4925 case ALGORITHM_RIGHT_ASYMMETRIC_6:
4926 case ALGORITHM_LEFT_SYMMETRIC_6:
4927 case ALGORITHM_RIGHT_SYMMETRIC_6:
4928 if (raid_disk == raid_disks - 1)
4929 return 1;
4930 }
4931 return 0;
4932}
4933
91adb564
N		 4934static int run(mddev_t *mddev)
4935{
4936 raid5_conf_t *conf;
9f7c2220 4937 int working_disks = 0;
c148ffdc 4938 int dirty_parity_disks = 0;
91adb564 4939 mdk_rdev_t *rdev;
c148ffdc 4940 sector_t reshape_offset = 0;
91adb564 4941
8c6ac868 4942 if (mddev->recovery_cp != MaxSector)
0c55e022 4943 printk(KERN_NOTICE "md/raid:%s: not clean"
8c6ac868
AN		 4944 " -- starting background reconstruction\n",
4945 mdname(mddev));
91adb564
N		 4946 if (mddev->reshape_position != MaxSector) {
4947 /* Check that we can continue the reshape.
4948 * Currently only disks can change, it must
4949 * increase, and we must be past the point where
4950 * a stripe over-writes itself
4951 */
4952 sector_t here_new, here_old;
4953 int old_disks;
18b00334 4954 int max_degraded = (mddev->level == 6 ? 2 : 1);
91adb564 4955
88ce4930 4956 if (mddev->new_level != mddev->level) {
0c55e022 4957 printk(KERN_ERR "md/raid:%s: unsupported reshape "
91adb564
N		 4958 "required - aborting.\n",
4959 mdname(mddev));
4960 return -EINVAL;
4961 }
91adb564
N		 4962 old_disks = mddev->raid_disks - mddev->delta_disks;
4963 /* reshape_position must be on a new-stripe boundary, and one
4964 * further up in new geometry must map after here in old
4965 * geometry.
4966 */
4967 here_new = mddev->reshape_position;
664e7c41 4968 if (sector_div(here_new, mddev->new_chunk_sectors *
91adb564 4969 (mddev->raid_disks - max_degraded))) {
0c55e022
N		 4970 printk(KERN_ERR "md/raid:%s: reshape_position not "
4971 "on a stripe boundary\n", mdname(mddev));
91adb564
N		 4972 return -EINVAL;
4973 }
c148ffdc 4974 reshape_offset = here_new * mddev->new_chunk_sectors;
91adb564
N		 4975 /* here_new is the stripe we will write to */
4976 here_old = mddev->reshape_position;
9d8f0363 4977 sector_div(here_old, mddev->chunk_sectors *
91adb564
N		 4978 (old_disks-max_degraded));
4979 /* here_old is the first stripe that we might need to read
4980 * from */
67ac6011
N		 4981 if (mddev->delta_disks == 0) {
4982 /* We cannot be sure it is safe to start an in-place
4983 * reshape. It is only safe if user-space if monitoring
4984 * and taking constant backups.
4985 * mdadm always starts a situation like this in
4986 * readonly mode so it can take control before
4987 * allowing any writes. So just check for that.
4988 */
4989 if ((here_new * mddev->new_chunk_sectors !=
4990 here_old * mddev->chunk_sectors) ||
4991 mddev->ro == 0) {
0c55e022
N		 4992 printk(KERN_ERR "md/raid:%s: in-place reshape must be started"
4993 " in read-only mode - aborting\n",
4994 mdname(mddev));
67ac6011
N		 4995 return -EINVAL;
4996 }
4997 } else if (mddev->delta_disks < 0
4998 ? (here_new * mddev->new_chunk_sectors <=
4999 here_old * mddev->chunk_sectors)
5000 : (here_new * mddev->new_chunk_sectors >=
5001 here_old * mddev->chunk_sectors)) {
91adb564 5002 /* Reading from the same stripe as writing to - bad */
0c55e022
N		 5003 printk(KERN_ERR "md/raid:%s: reshape_position too early for "
5004 "auto-recovery - aborting.\n",
5005 mdname(mddev));
91adb564
N		 5006 return -EINVAL;
5007 }
0c55e022
N		 5008 printk(KERN_INFO "md/raid:%s: reshape will continue\n",
5009 mdname(mddev));
91adb564
N		 5010 /* OK, we should be able to continue; */
5011 } else {
5012 BUG_ON(mddev->level != mddev->new_level);
5013 BUG_ON(mddev->layout != mddev->new_layout);
664e7c41 5014 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
91adb564 5015 BUG_ON(mddev->delta_disks != 0);
1da177e4 5016 }
91adb564 5017
245f46c2
N		 5018 if (mddev->private == NULL)
5019 conf = setup_conf(mddev);
5020 else
5021 conf = mddev->private;
5022
91adb564
N		 5023 if (IS_ERR(conf))
5024 return PTR_ERR(conf);
5025
5026 mddev->thread = conf->thread;
5027 conf->thread = NULL;
5028 mddev->private = conf;
5029
5030 /*
5031 * 0 for a fully functional array, 1 or 2 for a degraded array.
5032 */
c148ffdc
N		 5033 list_for_each_entry(rdev, &mddev->disks, same_set) {
5034 if (rdev->raid_disk < 0)
5035 continue;
2f115882 5036 if (test_bit(In_sync, &rdev->flags)) {
91adb564 5037 working_disks++;
2f115882
N		 5038 continue;
5039 }
c148ffdc
N		 5040 /* This disc is not fully in-sync. However if it
5041 * just stored parity (beyond the recovery_offset),
5042 * when we don't need to be concerned about the
5043 * array being dirty.
5044 * When reshape goes 'backwards', we never have
5045 * partially completed devices, so we only need
5046 * to worry about reshape going forwards.
5047 */
5048 /* Hack because v0.91 doesn't store recovery_offset properly. */
5049 if (mddev->major_version == 0 &&
5050 mddev->minor_version > 90)
5051 rdev->recovery_offset = reshape_offset;
5052
c148ffdc
N		 5053 if (rdev->recovery_offset < reshape_offset) {
5054 /* We need to check old and new layout */
5055 if (!only_parity(rdev->raid_disk,
5056 conf->algorithm,
5057 conf->raid_disks,
5058 conf->max_degraded))
5059 continue;
5060 }
5061 if (!only_parity(rdev->raid_disk,
5062 conf->prev_algo,
5063 conf->previous_raid_disks,
5064 conf->max_degraded))
5065 continue;
5066 dirty_parity_disks++;
5067 }
91adb564 5068
5e5e3e78
N		 5069 mddev->degraded = (max(conf->raid_disks, conf->previous_raid_disks)
5070 - working_disks);
91adb564 5071
674806d6 5072 if (has_failed(conf)) {
0c55e022 5073 printk(KERN_ERR "md/raid:%s: not enough operational devices"
1da177e4 5074 " (%d/%d failed)\n",
02c2de8c 5075 mdname(mddev), mddev->degraded, conf->raid_disks);
1da177e4
LT		 5076 goto abort;
5077 }
5078
91adb564 5079 /* device size must be a multiple of chunk size */
9d8f0363 5080 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
91adb564
N		 5081 mddev->resync_max_sectors = mddev->dev_sectors;
5082
c148ffdc 5083 if (mddev->degraded > dirty_parity_disks &&
1da177e4 5084 mddev->recovery_cp != MaxSector) {
6ff8d8ec
N		 5085 if (mddev->ok_start_degraded)
5086 printk(KERN_WARNING
0c55e022
N		 5087 "md/raid:%s: starting dirty degraded array"
5088 " - data corruption possible.\n",
6ff8d8ec
N		 5089 mdname(mddev));
5090 else {
5091 printk(KERN_ERR
0c55e022 5092 "md/raid:%s: cannot start dirty degraded array.\n",
6ff8d8ec
N		 5093 mdname(mddev));
5094 goto abort;
5095 }
1da177e4
LT		 5096 }
5097
1da177e4 5098 if (mddev->degraded == 0)
0c55e022
N		 5099 printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d"
5100 " devices, algorithm %d\n", mdname(mddev), conf->level,
e183eaed
N		 5101 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
5102 mddev->new_layout);
1da177e4 5103 else
0c55e022
N		 5104 printk(KERN_ALERT "md/raid:%s: raid level %d active with %d"
5105 " out of %d devices, algorithm %d\n",
5106 mdname(mddev), conf->level,
5107 mddev->raid_disks - mddev->degraded,
5108 mddev->raid_disks, mddev->new_layout);
1da177e4
LT		 5109
5110 print_raid5_conf(conf);
5111
fef9c61f 5112 if (conf->reshape_progress != MaxSector) {
fef9c61f 5113 conf->reshape_safe = conf->reshape_progress;
f6705578
N		 5114 atomic_set(&conf->reshape_stripes, 0);
5115 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5116 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5117 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5118 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5119 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 5120 "reshape");
f6705578
N		 5121 }
5122
1da177e4
LT		 5123
5124 /* Ok, everything is just fine now */
a64c876f
N		 5125 if (mddev->to_remove == &raid5_attrs_group)
5126 mddev->to_remove = NULL;
00bcb4ac
N		 5127 else if (mddev->kobj.sd &&
5128 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
5e55e2f5 5129 printk(KERN_WARNING
4a5add49 5130 "raid5: failed to create sysfs attributes for %s\n",
5e55e2f5 5131 mdname(mddev));
4a5add49 5132 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
7a5febe9 5133
4a5add49 5134 if (mddev->queue) {
9f7c2220 5135 int chunk_size;
4a5add49
N		 5136 /* read-ahead size must cover two whole stripes, which
5137 * is 2 * (datadisks) * chunksize where 'n' is the
5138 * number of raid devices
5139 */
5140 int data_disks = conf->previous_raid_disks - conf->max_degraded;
5141 int stripe = data_disks *
5142 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
5143 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
5144 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
91adb564 5145
4a5add49 5146 blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec);
f022b2fd 5147
11d8a6e3
N		 5148 mddev->queue->backing_dev_info.congested_data = mddev;
5149 mddev->queue->backing_dev_info.congested_fn = raid5_congested;
7a5febe9 5150
9f7c2220
N		 5151 chunk_size = mddev->chunk_sectors << 9;
5152 blk_queue_io_min(mddev->queue, chunk_size);
5153 blk_queue_io_opt(mddev->queue, chunk_size *
5154 (conf->raid_disks - conf->max_degraded));
8f6c2e4b 5155
9f7c2220
N		 5156 list_for_each_entry(rdev, &mddev->disks, same_set)
5157 disk_stack_limits(mddev->gendisk, rdev->bdev,
5158 rdev->data_offset << 9);
5159 }
23032a0e 5160
1da177e4
LT		 5161 return 0;
5162abort:
e0cf8f04 5163 md_unregister_thread(mddev->thread);
91adb564 5164 mddev->thread = NULL;
1da177e4
LT		 5165 if (conf) {
5166 print_raid5_conf(conf);
95fc17aa 5167 free_conf(conf);
1da177e4
LT		 5168 }
5169 mddev->private = NULL;
0c55e022 5170 printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev));
1da177e4
LT		 5171 return -EIO;
5172}
5173
3f294f4f 5174static int stop(mddev_t *mddev)
1da177e4 5175{
7b92813c 5176 raid5_conf_t *conf = mddev->private;
1da177e4
LT		 5177
5178 md_unregister_thread(mddev->thread);
5179 mddev->thread = NULL;
11d8a6e3
N		 5180 if (mddev->queue)
5181 mddev->queue->backing_dev_info.congested_fn = NULL;
95fc17aa 5182 free_conf(conf);
a64c876f
N		 5183 mddev->private = NULL;
5184 mddev->to_remove = &raid5_attrs_group;
1da177e4
LT		 5185 return 0;
5186}
5187
45b4233c 5188#ifdef DEBUG
d710e138 5189static void print_sh(struct seq_file *seq, struct stripe_head *sh)
1da177e4
LT		 5190{
5191 int i;
5192
16a53ecc
N		 5193 seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n",
5194 (unsigned long long)sh->sector, sh->pd_idx, sh->state);
5195 seq_printf(seq, "sh %llu, count %d.\n",
5196 (unsigned long long)sh->sector, atomic_read(&sh->count));
5197 seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector);
7ecaa1e6 5198 for (i = 0; i < sh->disks; i++) {
16a53ecc
N		 5199 seq_printf(seq, "(cache%d: %p %ld) ",
5200 i, sh->dev[i].page, sh->dev[i].flags);
1da177e4 5201 }
16a53ecc 5202 seq_printf(seq, "\n");
1da177e4
LT		 5203}
5204
d710e138 5205static void printall(struct seq_file *seq, raid5_conf_t *conf)
1da177e4
LT		 5206{
5207 struct stripe_head *sh;
fccddba0 5208 struct hlist_node *hn;
1da177e4
LT		 5209 int i;
5210
5211 spin_lock_irq(&conf->device_lock);
5212 for (i = 0; i < NR_HASH; i++) {
fccddba0 5213 hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) {
1da177e4
LT		 5214 if (sh->raid_conf != conf)
5215 continue;
16a53ecc 5216 print_sh(seq, sh);
1da177e4
LT		 5217 }
5218 }
5219 spin_unlock_irq(&conf->device_lock);
5220}
5221#endif
5222
d710e138 5223static void status(struct seq_file *seq, mddev_t *mddev)
1da177e4 5224{
7b92813c 5225 raid5_conf_t *conf = mddev->private;
1da177e4
LT		 5226 int i;
5227
9d8f0363
AN		 5228 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
5229 mddev->chunk_sectors / 2, mddev->layout);
02c2de8c 5230 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
1da177e4
LT		 5231 for (i = 0; i < conf->raid_disks; i++)
5232 seq_printf (seq, "%s",
5233 conf->disks[i].rdev &&
b2d444d7 5234 test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_");
1da177e4 5235 seq_printf (seq, "]");
45b4233c 5236#ifdef DEBUG
16a53ecc
N		 5237 seq_printf (seq, "\n");
5238 printall(seq, conf);
1da177e4
LT		 5239#endif
5240}
5241
5242static void print_raid5_conf (raid5_conf_t *conf)
5243{
5244 int i;
5245 struct disk_info *tmp;
5246
0c55e022 5247 printk(KERN_DEBUG "RAID conf printout:\n");
1da177e4
LT		 5248 if (!conf) {
5249 printk("(conf==NULL)\n");
5250 return;
5251 }
0c55e022
N		 5252 printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level,
5253 conf->raid_disks,
5254 conf->raid_disks - conf->mddev->degraded);
1da177e4
LT		 5255
5256 for (i = 0; i < conf->raid_disks; i++) {
5257 char b[BDEVNAME_SIZE];
5258 tmp = conf->disks + i;
5259 if (tmp->rdev)
0c55e022
N		 5260 printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n",
5261 i, !test_bit(Faulty, &tmp->rdev->flags),
5262 bdevname(tmp->rdev->bdev, b));
1da177e4
LT		 5263 }
5264}
5265
5266static int raid5_spare_active(mddev_t *mddev)
5267{
5268 int i;
5269 raid5_conf_t *conf = mddev->private;
5270 struct disk_info *tmp;
6b965620
N		 5271 int count = 0;
5272 unsigned long flags;
1da177e4
LT		 5273
5274 for (i = 0; i < conf->raid_disks; i++) {
5275 tmp = conf->disks + i;
5276 if (tmp->rdev
70fffd0b 5277 && tmp->rdev->recovery_offset == MaxSector
b2d444d7 5278 && !test_bit(Faulty, &tmp->rdev->flags)
c04be0aa 5279 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
6b965620 5280 count++;
43c73ca4 5281 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
1da177e4
LT		 5282 }
5283 }
6b965620
N		 5284 spin_lock_irqsave(&conf->device_lock, flags);
5285 mddev->degraded -= count;
5286 spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4 5287 print_raid5_conf(conf);
6b965620 5288 return count;
1da177e4
LT		 5289}
5290
5291static int raid5_remove_disk(mddev_t *mddev, int number)
5292{
5293 raid5_conf_t *conf = mddev->private;
5294 int err = 0;
5295 mdk_rdev_t *rdev;
5296 struct disk_info *p = conf->disks + number;
5297
5298 print_raid5_conf(conf);
5299 rdev = p->rdev;
5300 if (rdev) {
ec32a2bd
N		 5301 if (number >= conf->raid_disks &&
5302 conf->reshape_progress == MaxSector)
5303 clear_bit(In_sync, &rdev->flags);
5304
b2d444d7 5305 if (test_bit(In_sync, &rdev->flags) ||
1da177e4
LT		 5306 atomic_read(&rdev->nr_pending)) {
5307 err = -EBUSY;
5308 goto abort;
5309 }
dfc70645
N		 5310 /* Only remove non-faulty devices if recovery
5311 * isn't possible.
5312 */
5313 if (!test_bit(Faulty, &rdev->flags) &&
674806d6 5314 !has_failed(conf) &&
ec32a2bd 5315 number < conf->raid_disks) {
dfc70645
N		 5316 err = -EBUSY;
5317 goto abort;
5318 }
1da177e4 5319 p->rdev = NULL;
fbd568a3 5320 synchronize_rcu();
1da177e4
LT		 5321 if (atomic_read(&rdev->nr_pending)) {
5322 /* lost the race, try later */
5323 err = -EBUSY;
5324 p->rdev = rdev;
5325 }
5326 }
5327abort:
5328
5329 print_raid5_conf(conf);
5330 return err;
5331}
5332
5333static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
5334{
5335 raid5_conf_t *conf = mddev->private;
199050ea 5336 int err = -EEXIST;
1da177e4
LT		 5337 int disk;
5338 struct disk_info *p;
6c2fce2e
NB		 5339 int first = 0;
5340 int last = conf->raid_disks - 1;
1da177e4 5341
674806d6 5342 if (has_failed(conf))
1da177e4 5343 /* no point adding a device */
199050ea 5344 return -EINVAL;
1da177e4 5345
6c2fce2e
NB		 5346 if (rdev->raid_disk >= 0)
5347 first = last = rdev->raid_disk;
1da177e4
LT		 5348
5349 /*
16a53ecc
N		 5350 * find the disk ... but prefer rdev->saved_raid_disk
5351 * if possible.
1da177e4 5352 */
16a53ecc 5353 if (rdev->saved_raid_disk >= 0 &&
6c2fce2e 5354 rdev->saved_raid_disk >= first &&
16a53ecc
N		 5355 conf->disks[rdev->saved_raid_disk].rdev == NULL)
5356 disk = rdev->saved_raid_disk;
5357 else
6c2fce2e
NB		 5358 disk = first;
5359 for ( ; disk <= last ; disk++)
1da177e4 5360 if ((p=conf->disks + disk)->rdev == NULL) {
b2d444d7 5361 clear_bit(In_sync, &rdev->flags);
1da177e4 5362 rdev->raid_disk = disk;
199050ea 5363 err = 0;
72626685
N		 5364 if (rdev->saved_raid_disk != disk)
5365 conf->fullsync = 1;
d6065f7b 5366 rcu_assign_pointer(p->rdev, rdev);
1da177e4
LT		 5367 break;
5368 }
5369 print_raid5_conf(conf);
199050ea 5370 return err;
1da177e4
LT		 5371}
5372
5373static int raid5_resize(mddev_t *mddev, sector_t sectors)
5374{
5375 /* no resync is happening, and there is enough space
5376 * on all devices, so we can resize.
5377 * We need to make sure resync covers any new space.
5378 * If the array is shrinking we should possibly wait until
5379 * any io in the removed space completes, but it hardly seems
5380 * worth it.
5381 */
9d8f0363 5382 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
1f403624
DW		 5383 md_set_array_sectors(mddev, raid5_size(mddev, sectors,
5384 mddev->raid_disks));
b522adcd
DW		 5385 if (mddev->array_sectors >
5386 raid5_size(mddev, sectors, mddev->raid_disks))
5387 return -EINVAL;
f233ea5c 5388 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 5389 revalidate_disk(mddev->gendisk);
b098636c
N		 5390 if (sectors > mddev->dev_sectors &&
5391 mddev->recovery_cp > mddev->dev_sectors) {
58c0fed4 5392 mddev->recovery_cp = mddev->dev_sectors;
1da177e4
LT		 5393 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5394 }
58c0fed4 5395 mddev->dev_sectors = sectors;
4b5c7ae8 5396 mddev->resync_max_sectors = sectors;
1da177e4
LT		 5397 return 0;
5398}
5399
01ee22b4
N		 5400static int check_stripe_cache(mddev_t *mddev)
5401{
5402 /* Can only proceed if there are plenty of stripe_heads.
5403 * We need a minimum of one full stripe,, and for sensible progress
5404 * it is best to have about 4 times that.
5405 * If we require 4 times, then the default 256 4K stripe_heads will
5406 * allow for chunk sizes up to 256K, which is probably OK.
5407 * If the chunk size is greater, user-space should request more
5408 * stripe_heads first.
5409 */
5410 raid5_conf_t *conf = mddev->private;
5411 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
5412 > conf->max_nr_stripes ||
5413 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
5414 > conf->max_nr_stripes) {
0c55e022
N		 5415 printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n",
5416 mdname(mddev),
01ee22b4
N		 5417 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
5418 / STRIPE_SIZE)*4);
5419 return 0;
5420 }
5421 return 1;
5422}
5423
50ac168a 5424static int check_reshape(mddev_t *mddev)
29269553 5425{
070ec55d 5426 raid5_conf_t *conf = mddev->private;
29269553 5427
88ce4930
N		 5428 if (mddev->delta_disks == 0 &&
5429 mddev->new_layout == mddev->layout &&
664e7c41 5430 mddev->new_chunk_sectors == mddev->chunk_sectors)
50ac168a 5431 return 0; /* nothing to do */
dba034ee
N		 5432 if (mddev->bitmap)
5433 /* Cannot grow a bitmap yet */
5434 return -EBUSY;
674806d6 5435 if (has_failed(conf))
ec32a2bd
N		 5436 return -EINVAL;
5437 if (mddev->delta_disks < 0) {
5438 /* We might be able to shrink, but the devices must
5439 * be made bigger first.
5440 * For raid6, 4 is the minimum size.
5441 * Otherwise 2 is the minimum
5442 */
5443 int min = 2;
5444 if (mddev->level == 6)
5445 min = 4;
5446 if (mddev->raid_disks + mddev->delta_disks < min)
5447 return -EINVAL;
5448 }
29269553 5449
01ee22b4 5450 if (!check_stripe_cache(mddev))
29269553 5451 return -ENOSPC;
29269553 5452
ec32a2bd 5453 return resize_stripes(conf, conf->raid_disks + mddev->delta_disks);
63c70c4f
N		 5454}
5455
5456static int raid5_start_reshape(mddev_t *mddev)
5457{
070ec55d 5458 raid5_conf_t *conf = mddev->private;
63c70c4f 5459 mdk_rdev_t *rdev;
63c70c4f 5460 int spares = 0;
c04be0aa 5461 unsigned long flags;
63c70c4f 5462
f416885e 5463 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
63c70c4f
N		 5464 return -EBUSY;
5465
01ee22b4
N		 5466 if (!check_stripe_cache(mddev))
5467 return -ENOSPC;
5468
159ec1fc 5469 list_for_each_entry(rdev, &mddev->disks, same_set)
469518a3
N		 5470 if (!test_bit(In_sync, &rdev->flags)
5471 && !test_bit(Faulty, &rdev->flags))
29269553 5472 spares++;
63c70c4f 5473
f416885e 5474 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
29269553
N		 5475 /* Not enough devices even to make a degraded array
5476 * of that size
5477 */
5478 return -EINVAL;
5479
ec32a2bd
N		 5480 /* Refuse to reduce size of the array. Any reductions in
5481 * array size must be through explicit setting of array_size
5482 * attribute.
5483 */
5484 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
5485 < mddev->array_sectors) {
0c55e022 5486 printk(KERN_ERR "md/raid:%s: array size must be reduced "
ec32a2bd
N		 5487 "before number of disks\n", mdname(mddev));
5488 return -EINVAL;
5489 }
5490
f6705578 5491 atomic_set(&conf->reshape_stripes, 0);
29269553
N		 5492 spin_lock_irq(&conf->device_lock);
5493 conf->previous_raid_disks = conf->raid_disks;
63c70c4f 5494 conf->raid_disks += mddev->delta_disks;
09c9e5fa
AN		 5495 conf->prev_chunk_sectors = conf->chunk_sectors;
5496 conf->chunk_sectors = mddev->new_chunk_sectors;
88ce4930
N		 5497 conf->prev_algo = conf->algorithm;
5498 conf->algorithm = mddev->new_layout;
fef9c61f
N		 5499 if (mddev->delta_disks < 0)
5500 conf->reshape_progress = raid5_size(mddev, 0, 0);
5501 else
5502 conf->reshape_progress = 0;
5503 conf->reshape_safe = conf->reshape_progress;
86b42c71 5504 conf->generation++;
29269553
N		 5505 spin_unlock_irq(&conf->device_lock);
5506
5507 /* Add some new drives, as many as will fit.
5508 * We know there are enough to make the newly sized array work.
3424bf6a
N		 5509 * Don't add devices if we are reducing the number of
5510 * devices in the array. This is because it is not possible
5511 * to correctly record the "partially reconstructed" state of
5512 * such devices during the reshape and confusion could result.
29269553 5513 */
87a8dec9
N		 5514 if (mddev->delta_disks >= 0) {
5515 int added_devices = 0;
5516 list_for_each_entry(rdev, &mddev->disks, same_set)
5517 if (rdev->raid_disk < 0 &&
5518 !test_bit(Faulty, &rdev->flags)) {
5519 if (raid5_add_disk(mddev, rdev) == 0) {
5520 char nm[20];
5521 if (rdev->raid_disk
5522 >= conf->previous_raid_disks) {
5523 set_bit(In_sync, &rdev->flags);
5524 added_devices++;
5525 } else
5526 rdev->recovery_offset = 0;
5527 sprintf(nm, "rd%d", rdev->raid_disk);
5528 if (sysfs_create_link(&mddev->kobj,
5529 &rdev->kobj, nm))
5530 /* Failure here is OK */;
50da0840 5531 }
87a8dec9
N		 5532 } else if (rdev->raid_disk >= conf->previous_raid_disks
5533 && !test_bit(Faulty, &rdev->flags)) {
5534 /* This is a spare that was manually added */
5535 set_bit(In_sync, &rdev->flags);
5536 added_devices++;
5537 }
29269553 5538
87a8dec9
N		 5539 /* When a reshape changes the number of devices,
5540 * ->degraded is measured against the larger of the
5541 * pre and post number of devices.
5542 */
ec32a2bd 5543 spin_lock_irqsave(&conf->device_lock, flags);
9eb07c25 5544 mddev->degraded += (conf->raid_disks - conf->previous_raid_disks)
ec32a2bd
N		 5545 - added_devices;
5546 spin_unlock_irqrestore(&conf->device_lock, flags);
5547 }
63c70c4f 5548 mddev->raid_disks = conf->raid_disks;
e516402c 5549 mddev->reshape_position = conf->reshape_progress;
850b2b42 5550 set_bit(MD_CHANGE_DEVS, &mddev->flags);
f6705578 5551
29269553
N		 5552 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5553 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5554 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5555 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5556 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 5557 "reshape");
29269553
N		 5558 if (!mddev->sync_thread) {
5559 mddev->recovery = 0;
5560 spin_lock_irq(&conf->device_lock);
5561 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
fef9c61f 5562 conf->reshape_progress = MaxSector;
29269553
N		 5563 spin_unlock_irq(&conf->device_lock);
5564 return -EAGAIN;
5565 }
c8f517c4 5566 conf->reshape_checkpoint = jiffies;
29269553
N		 5567 md_wakeup_thread(mddev->sync_thread);
5568 md_new_event(mddev);
5569 return 0;
5570}
29269553 5571
ec32a2bd
N		 5572/* This is called from the reshape thread and should make any
5573 * changes needed in 'conf'
5574 */
29269553
N		 5575static void end_reshape(raid5_conf_t *conf)
5576{
29269553 5577
f6705578 5578 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
f6705578 5579
f6705578 5580 spin_lock_irq(&conf->device_lock);
cea9c228 5581 conf->previous_raid_disks = conf->raid_disks;
fef9c61f 5582 conf->reshape_progress = MaxSector;
f6705578 5583 spin_unlock_irq(&conf->device_lock);
b0f9ec04 5584 wake_up(&conf->wait_for_overlap);
16a53ecc
N		 5585
5586 /* read-ahead size must cover two whole stripes, which is
5587 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
5588 */
4a5add49 5589 if (conf->mddev->queue) {
cea9c228 5590 int data_disks = conf->raid_disks - conf->max_degraded;
09c9e5fa 5591 int stripe = data_disks * ((conf->chunk_sectors << 9)
cea9c228 5592 / PAGE_SIZE);
16a53ecc
N		 5593 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
5594 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
5595 }
29269553 5596 }
29269553
N		 5597}
5598
ec32a2bd
N		 5599/* This is called from the raid5d thread with mddev_lock held.
5600 * It makes config changes to the device.
5601 */
cea9c228
N		 5602static void raid5_finish_reshape(mddev_t *mddev)
5603{
070ec55d 5604 raid5_conf_t *conf = mddev->private;
cea9c228
N		 5605
5606 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5607
ec32a2bd
N		 5608 if (mddev->delta_disks > 0) {
5609 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
5610 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 5611 revalidate_disk(mddev->gendisk);
ec32a2bd
N		 5612 } else {
5613 int d;
ec32a2bd
N		 5614 mddev->degraded = conf->raid_disks;
5615 for (d = 0; d < conf->raid_disks ; d++)
5616 if (conf->disks[d].rdev &&
5617 test_bit(In_sync,
5618 &conf->disks[d].rdev->flags))
5619 mddev->degraded--;
5620 for (d = conf->raid_disks ;
5621 d < conf->raid_disks - mddev->delta_disks;
1a67dde0
N		 5622 d++) {
5623 mdk_rdev_t *rdev = conf->disks[d].rdev;
5624 if (rdev && raid5_remove_disk(mddev, d) == 0) {
5625 char nm[20];
5626 sprintf(nm, "rd%d", rdev->raid_disk);
5627 sysfs_remove_link(&mddev->kobj, nm);
5628 rdev->raid_disk = -1;
5629 }
5630 }
cea9c228 5631 }
88ce4930 5632 mddev->layout = conf->algorithm;
09c9e5fa 5633 mddev->chunk_sectors = conf->chunk_sectors;
ec32a2bd
N		 5634 mddev->reshape_position = MaxSector;
5635 mddev->delta_disks = 0;
cea9c228
N		 5636 }
5637}
5638
72626685
N		 5639static void raid5_quiesce(mddev_t *mddev, int state)
5640{
070ec55d 5641 raid5_conf_t *conf = mddev->private;
72626685
N		 5642
5643 switch(state) {
e464eafd
N		 5644 case 2: /* resume for a suspend */
5645 wake_up(&conf->wait_for_overlap);
5646 break;
5647
72626685
N		 5648 case 1: /* stop all writes */
5649 spin_lock_irq(&conf->device_lock);
64bd660b
N		 5650 /* '2' tells resync/reshape to pause so that all
5651 * active stripes can drain
5652 */
5653 conf->quiesce = 2;
72626685 5654 wait_event_lock_irq(conf->wait_for_stripe,
46031f9a
RBJ		 5655 atomic_read(&conf->active_stripes) == 0 &&
5656 atomic_read(&conf->active_aligned_reads) == 0,
72626685 5657 conf->device_lock, /* nothing */);
64bd660b 5658 conf->quiesce = 1;
72626685 5659 spin_unlock_irq(&conf->device_lock);
64bd660b
N		 5660 /* allow reshape to continue */
5661 wake_up(&conf->wait_for_overlap);
72626685
N		 5662 break;
5663
5664 case 0: /* re-enable writes */
5665 spin_lock_irq(&conf->device_lock);
5666 conf->quiesce = 0;
5667 wake_up(&conf->wait_for_stripe);
e464eafd 5668 wake_up(&conf->wait_for_overlap);
72626685
N		 5669 spin_unlock_irq(&conf->device_lock);
5670 break;
5671 }
72626685 5672}
b15c2e57 5673
d562b0c4 5674
f1b29bca 5675static void *raid45_takeover_raid0(mddev_t *mddev, int level)
54071b38 5676{
f1b29bca 5677 struct raid0_private_data *raid0_priv = mddev->private;
d76c8420 5678 sector_t sectors;
54071b38 5679
f1b29bca
DW		 5680 /* for raid0 takeover only one zone is supported */
5681 if (raid0_priv->nr_strip_zones > 1) {
0c55e022
N		 5682 printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n",
5683 mdname(mddev));
f1b29bca
DW		 5684 return ERR_PTR(-EINVAL);
5685 }
5686
3b71bd93
N		 5687 sectors = raid0_priv->strip_zone[0].zone_end;
5688 sector_div(sectors, raid0_priv->strip_zone[0].nb_dev);
5689 mddev->dev_sectors = sectors;
f1b29bca 5690 mddev->new_level = level;
54071b38
TM		 5691 mddev->new_layout = ALGORITHM_PARITY_N;
5692 mddev->new_chunk_sectors = mddev->chunk_sectors;
5693 mddev->raid_disks += 1;
5694 mddev->delta_disks = 1;
5695 /* make sure it will be not marked as dirty */
5696 mddev->recovery_cp = MaxSector;
5697
5698 return setup_conf(mddev);
5699}
5700
5701
d562b0c4
N		 5702static void *raid5_takeover_raid1(mddev_t *mddev)
5703{
5704 int chunksect;
5705
5706 if (mddev->raid_disks != 2 ||
5707 mddev->degraded > 1)
5708 return ERR_PTR(-EINVAL);
5709
5710 /* Should check if there are write-behind devices? */
5711
5712 chunksect = 64*2; /* 64K by default */
5713
5714 /* The array must be an exact multiple of chunksize */
5715 while (chunksect && (mddev->array_sectors & (chunksect-1)))
5716 chunksect >>= 1;
5717
5718 if ((chunksect<<9) < STRIPE_SIZE)
5719 /* array size does not allow a suitable chunk size */
5720 return ERR_PTR(-EINVAL);
5721
5722 mddev->new_level = 5;
5723 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
664e7c41 5724 mddev->new_chunk_sectors = chunksect;
d562b0c4
N		 5725
5726 return setup_conf(mddev);
5727}
5728
fc9739c6
N		 5729static void *raid5_takeover_raid6(mddev_t *mddev)
5730{
5731 int new_layout;
5732
5733 switch (mddev->layout) {
5734 case ALGORITHM_LEFT_ASYMMETRIC_6:
5735 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
5736 break;
5737 case ALGORITHM_RIGHT_ASYMMETRIC_6:
5738 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
5739 break;
5740 case ALGORITHM_LEFT_SYMMETRIC_6:
5741 new_layout = ALGORITHM_LEFT_SYMMETRIC;
5742 break;
5743 case ALGORITHM_RIGHT_SYMMETRIC_6:
5744 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
5745 break;
5746 case ALGORITHM_PARITY_0_6:
5747 new_layout = ALGORITHM_PARITY_0;
5748 break;
5749 case ALGORITHM_PARITY_N:
5750 new_layout = ALGORITHM_PARITY_N;
5751 break;
5752 default:
5753 return ERR_PTR(-EINVAL);
5754 }
5755 mddev->new_level = 5;
5756 mddev->new_layout = new_layout;
5757 mddev->delta_disks = -1;
5758 mddev->raid_disks -= 1;
5759 return setup_conf(mddev);
5760}
5761
d562b0c4 5762
50ac168a 5763static int raid5_check_reshape(mddev_t *mddev)
b3546035 5764{
88ce4930
N		 5765 /* For a 2-drive array, the layout and chunk size can be changed
5766 * immediately as not restriping is needed.
5767 * For larger arrays we record the new value - after validation
5768 * to be used by a reshape pass.
b3546035 5769 */
070ec55d 5770 raid5_conf_t *conf = mddev->private;
597a711b 5771 int new_chunk = mddev->new_chunk_sectors;
b3546035 5772
597a711b 5773 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
b3546035
N		 5774 return -EINVAL;
5775 if (new_chunk > 0) {
0ba459d2 5776 if (!is_power_of_2(new_chunk))
b3546035 5777 return -EINVAL;
597a711b 5778 if (new_chunk < (PAGE_SIZE>>9))
b3546035 5779 return -EINVAL;
597a711b 5780 if (mddev->array_sectors & (new_chunk-1))
b3546035
N		 5781 /* not factor of array size */
5782 return -EINVAL;
5783 }
5784
5785 /* They look valid */
5786
88ce4930 5787 if (mddev->raid_disks == 2) {
597a711b
N		 5788 /* can make the change immediately */
5789 if (mddev->new_layout >= 0) {
5790 conf->algorithm = mddev->new_layout;
5791 mddev->layout = mddev->new_layout;
88ce4930
N		 5792 }
5793 if (new_chunk > 0) {
597a711b
N		 5794 conf->chunk_sectors = new_chunk ;
5795 mddev->chunk_sectors = new_chunk;
88ce4930
N		 5796 }
5797 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5798 md_wakeup_thread(mddev->thread);
b3546035 5799 }
50ac168a 5800 return check_reshape(mddev);
88ce4930
N		 5801}
5802
50ac168a 5803static int raid6_check_reshape(mddev_t *mddev)
88ce4930 5804{
597a711b 5805 int new_chunk = mddev->new_chunk_sectors;
50ac168a 5806
597a711b 5807 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
88ce4930 5808 return -EINVAL;
b3546035 5809 if (new_chunk > 0) {
0ba459d2 5810 if (!is_power_of_2(new_chunk))
88ce4930 5811 return -EINVAL;
597a711b 5812 if (new_chunk < (PAGE_SIZE >> 9))
88ce4930 5813 return -EINVAL;
597a711b 5814 if (mddev->array_sectors & (new_chunk-1))
88ce4930
N		 5815 /* not factor of array size */
5816 return -EINVAL;
b3546035 5817 }
88ce4930
N		 5818
5819 /* They look valid */
50ac168a 5820 return check_reshape(mddev);
b3546035
N		 5821}
5822
d562b0c4
N		 5823static void *raid5_takeover(mddev_t *mddev)
5824{
5825 /* raid5 can take over:
f1b29bca 5826 * raid0 - if there is only one strip zone - make it a raid4 layout
d562b0c4
N		 5827 * raid1 - if there are two drives. We need to know the chunk size
5828 * raid4 - trivial - just use a raid4 layout.
5829 * raid6 - Providing it is a *_6 layout
d562b0c4 5830 */
f1b29bca
DW		 5831 if (mddev->level == 0)
5832 return raid45_takeover_raid0(mddev, 5);
d562b0c4
N		 5833 if (mddev->level == 1)
5834 return raid5_takeover_raid1(mddev);
e9d4758f
N		 5835 if (mddev->level == 4) {
5836 mddev->new_layout = ALGORITHM_PARITY_N;
5837 mddev->new_level = 5;
5838 return setup_conf(mddev);
5839 }
fc9739c6
N		 5840 if (mddev->level == 6)
5841 return raid5_takeover_raid6(mddev);
d562b0c4
N		 5842
5843 return ERR_PTR(-EINVAL);
5844}
5845
a78d38a1
N		 5846static void *raid4_takeover(mddev_t *mddev)
5847{
f1b29bca
DW		 5848 /* raid4 can take over:
5849 * raid0 - if there is only one strip zone
5850 * raid5 - if layout is right
a78d38a1 5851 */
f1b29bca
DW		 5852 if (mddev->level == 0)
5853 return raid45_takeover_raid0(mddev, 4);
a78d38a1
N		 5854 if (mddev->level == 5 &&
5855 mddev->layout == ALGORITHM_PARITY_N) {
5856 mddev->new_layout = 0;
5857 mddev->new_level = 4;
5858 return setup_conf(mddev);
5859 }
5860 return ERR_PTR(-EINVAL);
5861}
d562b0c4 5862
245f46c2
N		 5863static struct mdk_personality raid5_personality;
5864
5865static void *raid6_takeover(mddev_t *mddev)
5866{
5867 /* Currently can only take over a raid5. We map the
5868 * personality to an equivalent raid6 personality
5869 * with the Q block at the end.
5870 */
5871 int new_layout;
5872
5873 if (mddev->pers != &raid5_personality)
5874 return ERR_PTR(-EINVAL);
5875 if (mddev->degraded > 1)
5876 return ERR_PTR(-EINVAL);
5877 if (mddev->raid_disks > 253)
5878 return ERR_PTR(-EINVAL);
5879 if (mddev->raid_disks < 3)
5880 return ERR_PTR(-EINVAL);
5881
5882 switch (mddev->layout) {
5883 case ALGORITHM_LEFT_ASYMMETRIC:
5884 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
5885 break;
5886 case ALGORITHM_RIGHT_ASYMMETRIC:
5887 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
5888 break;
5889 case ALGORITHM_LEFT_SYMMETRIC:
5890 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
5891 break;
5892 case ALGORITHM_RIGHT_SYMMETRIC:
5893 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
5894 break;
5895 case ALGORITHM_PARITY_0:
5896 new_layout = ALGORITHM_PARITY_0_6;
5897 break;
5898 case ALGORITHM_PARITY_N:
5899 new_layout = ALGORITHM_PARITY_N;
5900 break;
5901 default:
5902 return ERR_PTR(-EINVAL);
5903 }
5904 mddev->new_level = 6;
5905 mddev->new_layout = new_layout;
5906 mddev->delta_disks = 1;
5907 mddev->raid_disks += 1;
5908 return setup_conf(mddev);
5909}
5910
5911
16a53ecc
N		 5912static struct mdk_personality raid6_personality =
5913{
5914 .name = "raid6",
5915 .level = 6,
5916 .owner = THIS_MODULE,
5917 .make_request = make_request,
5918 .run = run,
5919 .stop = stop,
5920 .status = status,
5921 .error_handler = error,
5922 .hot_add_disk = raid5_add_disk,
5923 .hot_remove_disk= raid5_remove_disk,
5924 .spare_active = raid5_spare_active,
5925 .sync_request = sync_request,
5926 .resize = raid5_resize,
80c3a6ce 5927 .size = raid5_size,
50ac168a 5928 .check_reshape = raid6_check_reshape,
f416885e 5929 .start_reshape = raid5_start_reshape,
cea9c228 5930 .finish_reshape = raid5_finish_reshape,
16a53ecc 5931 .quiesce = raid5_quiesce,
245f46c2 5932 .takeover = raid6_takeover,
16a53ecc 5933};
2604b703 5934static struct mdk_personality raid5_personality =
1da177e4
LT		 5935{
5936 .name = "raid5",
2604b703 5937 .level = 5,
1da177e4
LT		 5938 .owner = THIS_MODULE,
5939 .make_request = make_request,
5940 .run = run,
5941 .stop = stop,
5942 .status = status,
5943 .error_handler = error,
5944 .hot_add_disk = raid5_add_disk,
5945 .hot_remove_disk= raid5_remove_disk,
5946 .spare_active = raid5_spare_active,
5947 .sync_request = sync_request,
5948 .resize = raid5_resize,
80c3a6ce 5949 .size = raid5_size,
63c70c4f
N		 5950 .check_reshape = raid5_check_reshape,
5951 .start_reshape = raid5_start_reshape,
cea9c228 5952 .finish_reshape = raid5_finish_reshape,
72626685 5953 .quiesce = raid5_quiesce,
d562b0c4 5954 .takeover = raid5_takeover,
1da177e4
LT		 5955};
5956
2604b703 5957static struct mdk_personality raid4_personality =
1da177e4 5958{
2604b703
N		 5959 .name = "raid4",
5960 .level = 4,
5961 .owner = THIS_MODULE,
5962 .make_request = make_request,
5963 .run = run,
5964 .stop = stop,
5965 .status = status,
5966 .error_handler = error,
5967 .hot_add_disk = raid5_add_disk,
5968 .hot_remove_disk= raid5_remove_disk,
5969 .spare_active = raid5_spare_active,
5970 .sync_request = sync_request,
5971 .resize = raid5_resize,
80c3a6ce 5972 .size = raid5_size,
3d37890b
N		 5973 .check_reshape = raid5_check_reshape,
5974 .start_reshape = raid5_start_reshape,
cea9c228 5975 .finish_reshape = raid5_finish_reshape,
2604b703 5976 .quiesce = raid5_quiesce,
a78d38a1 5977 .takeover = raid4_takeover,
2604b703
N		 5978};
5979
5980static int __init raid5_init(void)
5981{
16a53ecc 5982 register_md_personality(&raid6_personality);
2604b703
N		 5983 register_md_personality(&raid5_personality);
5984 register_md_personality(&raid4_personality);
5985 return 0;
1da177e4
LT		 5986}
5987
2604b703 5988static void raid5_exit(void)
1da177e4 5989{
16a53ecc 5990 unregister_md_personality(&raid6_personality);
2604b703
N		 5991 unregister_md_personality(&raid5_personality);
5992 unregister_md_personality(&raid4_personality);
1da177e4
LT		 5993}
5994
5995module_init(raid5_init);
5996module_exit(raid5_exit);
5997MODULE_LICENSE("GPL");
0efb9e61 5998MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
1da177e4 5999MODULE_ALIAS("md-personality-4"); /* RAID5 */
d9d166c2
N		 6000MODULE_ALIAS("md-raid5");
6001MODULE_ALIAS("md-raid4");
2604b703
N		 6002MODULE_ALIAS("md-level-5");
6003MODULE_ALIAS("md-level-4");
16a53ecc
N		 6004MODULE_ALIAS("md-personality-8"); /* RAID6 */
6005MODULE_ALIAS("md-raid6");
6006MODULE_ALIAS("md-level-6");
6007
6008/* This used to be two separate modules, they were: */
6009MODULE_ALIAS("raid5");
6010MODULE_ALIAS("raid6");
kernel source for telekom puls (alcatel/mediatek)