projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
treewide: fix a few typos in comments
[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{
5b99c2ff 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;
517 if (rw == WRITE)
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;
551 if (rw == WRITE &&
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 {
557 if (rw == WRITE)
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
DW		 587 bio_for_each_segment(bvl, bio, i) {
588 int len = bio_iovec_idx(bio, i)->bv_len;
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) {
604 b_offset += bio_iovec_idx(bio, i)->bv_offset;
605 bio_page = bio_iovec_idx(bio, i)->bv_page;
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
1023 spin_lock(&sh->lock);
1024 chosen = dev->towrite;
1025 dev->towrite = NULL;
1026 BUG_ON(dev->written);
1027 wbi = dev->written = chosen;
1028 spin_unlock(&sh->lock);
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;
3f294f4f
N		 1318 sh = kmem_cache_alloc(conf->slab_cache, GFP_KERNEL);
1319 if (!sh)
1320 return 0;
e4e11e38 1321 memset(sh, 0, sizeof(*sh) + (conf->pool_size-1)*sizeof(struct r5dev));
3f294f4f
N		 1322 sh->raid_conf = conf;
1323 spin_lock_init(&sh->lock);
417b8d4a
DW		 1324 #ifdef CONFIG_MULTICORE_RAID456
1325 init_waitqueue_head(&sh->ops.wait_for_ops);
1326 #endif
3f294f4f 1327
e4e11e38
N		 1328 if (grow_buffers(sh)) {
1329 shrink_buffers(sh);
3f294f4f
N		 1330 kmem_cache_free(conf->slab_cache, sh);
1331 return 0;
1332 }
1333 /* we just created an active stripe so... */
1334 atomic_set(&sh->count, 1);
1335 atomic_inc(&conf->active_stripes);
1336 INIT_LIST_HEAD(&sh->lru);
1337 release_stripe(sh);
1338 return 1;
1339}
1340
1341static int grow_stripes(raid5_conf_t *conf, int num)
1342{
e18b890b 1343 struct kmem_cache *sc;
5e5e3e78 1344 int devs = max(conf->raid_disks, conf->previous_raid_disks);
1da177e4 1345
f4be6b43
N		 1346 if (conf->mddev->gendisk)
1347 sprintf(conf->cache_name[0],
1348 "raid%d-%s", conf->level, mdname(conf->mddev));
1349 else
1350 sprintf(conf->cache_name[0],
1351 "raid%d-%p", conf->level, conf->mddev);
1352 sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]);
1353
ad01c9e3
N		 1354 conf->active_name = 0;
1355 sc = kmem_cache_create(conf->cache_name[conf->active_name],
1da177e4 1356 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
20c2df83 1357 0, 0, NULL);
1da177e4
LT		 1358 if (!sc)
1359 return 1;
1360 conf->slab_cache = sc;
ad01c9e3 1361 conf->pool_size = devs;
16a53ecc 1362 while (num--)
3f294f4f 1363 if (!grow_one_stripe(conf))
1da177e4 1364 return 1;
1da177e4
LT		 1365 return 0;
1366}
29269553 1367
d6f38f31
DW		 1368/**
1369 * scribble_len - return the required size of the scribble region
1370 * @num - total number of disks in the array
1371 *
1372 * The size must be enough to contain:
1373 * 1/ a struct page pointer for each device in the array +2
1374 * 2/ room to convert each entry in (1) to its corresponding dma
1375 * (dma_map_page()) or page (page_address()) address.
1376 *
1377 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
1378 * calculate over all devices (not just the data blocks), using zeros in place
1379 * of the P and Q blocks.
1380 */
1381static size_t scribble_len(int num)
1382{
1383 size_t len;
1384
1385 len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2);
1386
1387 return len;
1388}
1389
ad01c9e3
N		 1390static int resize_stripes(raid5_conf_t *conf, int newsize)
1391{
1392 /* Make all the stripes able to hold 'newsize' devices.
1393 * New slots in each stripe get 'page' set to a new page.
1394 *
1395 * This happens in stages:
1396 * 1/ create a new kmem_cache and allocate the required number of
1397 * stripe_heads.
1398 * 2/ gather all the old stripe_heads and tranfer the pages across
1399 * to the new stripe_heads. This will have the side effect of
1400 * freezing the array as once all stripe_heads have been collected,
1401 * no IO will be possible. Old stripe heads are freed once their
1402 * pages have been transferred over, and the old kmem_cache is
1403 * freed when all stripes are done.
1404 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
1405 * we simple return a failre status - no need to clean anything up.
1406 * 4/ allocate new pages for the new slots in the new stripe_heads.
1407 * If this fails, we don't bother trying the shrink the
1408 * stripe_heads down again, we just leave them as they are.
1409 * As each stripe_head is processed the new one is released into
1410 * active service.
1411 *
1412 * Once step2 is started, we cannot afford to wait for a write,
1413 * so we use GFP_NOIO allocations.
1414 */
1415 struct stripe_head *osh, *nsh;
1416 LIST_HEAD(newstripes);
1417 struct disk_info *ndisks;
d6f38f31 1418 unsigned long cpu;
b5470dc5 1419 int err;
e18b890b 1420 struct kmem_cache *sc;
ad01c9e3
N		 1421 int i;
1422
1423 if (newsize <= conf->pool_size)
1424 return 0; /* never bother to shrink */
1425
b5470dc5
DW		 1426 err = md_allow_write(conf->mddev);
1427 if (err)
1428 return err;
2a2275d6 1429
ad01c9e3
N		 1430 /* Step 1 */
1431 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
1432 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
20c2df83 1433 0, 0, NULL);
ad01c9e3
N		 1434 if (!sc)
1435 return -ENOMEM;
1436
1437 for (i = conf->max_nr_stripes; i; i--) {
1438 nsh = kmem_cache_alloc(sc, GFP_KERNEL);
1439 if (!nsh)
1440 break;
1441
1442 memset(nsh, 0, sizeof(*nsh) + (newsize-1)*sizeof(struct r5dev));
1443
1444 nsh->raid_conf = conf;
1445 spin_lock_init(&nsh->lock);
417b8d4a
DW		 1446 #ifdef CONFIG_MULTICORE_RAID456
1447 init_waitqueue_head(&nsh->ops.wait_for_ops);
1448 #endif
ad01c9e3
N		 1449
1450 list_add(&nsh->lru, &newstripes);
1451 }
1452 if (i) {
1453 /* didn't get enough, give up */
1454 while (!list_empty(&newstripes)) {
1455 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1456 list_del(&nsh->lru);
1457 kmem_cache_free(sc, nsh);
1458 }
1459 kmem_cache_destroy(sc);
1460 return -ENOMEM;
1461 }
1462 /* Step 2 - Must use GFP_NOIO now.
1463 * OK, we have enough stripes, start collecting inactive
1464 * stripes and copying them over
1465 */
1466 list_for_each_entry(nsh, &newstripes, lru) {
1467 spin_lock_irq(&conf->device_lock);
1468 wait_event_lock_irq(conf->wait_for_stripe,
1469 !list_empty(&conf->inactive_list),
1470 conf->device_lock,
482c0834 1471 );
ad01c9e3
N		 1472 osh = get_free_stripe(conf);
1473 spin_unlock_irq(&conf->device_lock);
1474 atomic_set(&nsh->count, 1);
1475 for(i=0; i<conf->pool_size; i++)
1476 nsh->dev[i].page = osh->dev[i].page;
1477 for( ; i<newsize; i++)
1478 nsh->dev[i].page = NULL;
1479 kmem_cache_free(conf->slab_cache, osh);
1480 }
1481 kmem_cache_destroy(conf->slab_cache);
1482
1483 /* Step 3.
1484 * At this point, we are holding all the stripes so the array
1485 * is completely stalled, so now is a good time to resize
d6f38f31 1486 * conf->disks and the scribble region
ad01c9e3
N		 1487 */
1488 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
1489 if (ndisks) {
1490 for (i=0; i<conf->raid_disks; i++)
1491 ndisks[i] = conf->disks[i];
1492 kfree(conf->disks);
1493 conf->disks = ndisks;
1494 } else
1495 err = -ENOMEM;
1496
d6f38f31
DW		 1497 get_online_cpus();
1498 conf->scribble_len = scribble_len(newsize);
1499 for_each_present_cpu(cpu) {
1500 struct raid5_percpu *percpu;
1501 void *scribble;
1502
1503 percpu = per_cpu_ptr(conf->percpu, cpu);
1504 scribble = kmalloc(conf->scribble_len, GFP_NOIO);
1505
1506 if (scribble) {
1507 kfree(percpu->scribble);
1508 percpu->scribble = scribble;
1509 } else {
1510 err = -ENOMEM;
1511 break;
1512 }
1513 }
1514 put_online_cpus();
1515
ad01c9e3
N		 1516 /* Step 4, return new stripes to service */
1517 while(!list_empty(&newstripes)) {
1518 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1519 list_del_init(&nsh->lru);
d6f38f31 1520
ad01c9e3
N		 1521 for (i=conf->raid_disks; i < newsize; i++)
1522 if (nsh->dev[i].page == NULL) {
1523 struct page *p = alloc_page(GFP_NOIO);
1524 nsh->dev[i].page = p;
1525 if (!p)
1526 err = -ENOMEM;
1527 }
1528 release_stripe(nsh);
1529 }
1530 /* critical section pass, GFP_NOIO no longer needed */
1531
1532 conf->slab_cache = sc;
1533 conf->active_name = 1-conf->active_name;
1534 conf->pool_size = newsize;
1535 return err;
1536}
1da177e4 1537
3f294f4f 1538static int drop_one_stripe(raid5_conf_t *conf)
1da177e4
LT		 1539{
1540 struct stripe_head *sh;
1541
3f294f4f
N		 1542 spin_lock_irq(&conf->device_lock);
1543 sh = get_free_stripe(conf);
1544 spin_unlock_irq(&conf->device_lock);
1545 if (!sh)
1546 return 0;
78bafebd 1547 BUG_ON(atomic_read(&sh->count));
e4e11e38 1548 shrink_buffers(sh);
3f294f4f
N		 1549 kmem_cache_free(conf->slab_cache, sh);
1550 atomic_dec(&conf->active_stripes);
1551 return 1;
1552}
1553
1554static void shrink_stripes(raid5_conf_t *conf)
1555{
1556 while (drop_one_stripe(conf))
1557 ;
1558
29fc7e3e
N		 1559 if (conf->slab_cache)
1560 kmem_cache_destroy(conf->slab_cache);
1da177e4
LT		 1561 conf->slab_cache = NULL;
1562}
1563
6712ecf8 1564static void raid5_end_read_request(struct bio * bi, int error)
1da177e4 1565{
99c0fb5f 1566 struct stripe_head *sh = bi->bi_private;
1da177e4 1567 raid5_conf_t *conf = sh->raid_conf;
7ecaa1e6 1568 int disks = sh->disks, i;
1da177e4 1569 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
d6950432
N		 1570 char b[BDEVNAME_SIZE];
1571 mdk_rdev_t *rdev;
1da177e4 1572
1da177e4
LT		 1573
1574 for (i=0 ; i<disks; i++)
1575 if (bi == &sh->dev[i].req)
1576 break;
1577
45b4233c
DW		 1578 pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n",
1579 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
1da177e4
LT		 1580 uptodate);
1581 if (i == disks) {
1582 BUG();
6712ecf8 1583 return;
1da177e4
LT		 1584 }
1585
1586 if (uptodate) {
1da177e4 1587 set_bit(R5_UPTODATE, &sh->dev[i].flags);
4e5314b5 1588 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
d6950432 1589 rdev = conf->disks[i].rdev;
0c55e022 1590 printk_rl(KERN_INFO "md/raid:%s: read error corrected"
6be9d494
BS		 1591 " (%lu sectors at %llu on %s)\n",
1592 mdname(conf->mddev), STRIPE_SECTORS,
1593 (unsigned long long)(sh->sector
1594 + rdev->data_offset),
1595 bdevname(rdev->bdev, b));
4e5314b5
N		 1596 clear_bit(R5_ReadError, &sh->dev[i].flags);
1597 clear_bit(R5_ReWrite, &sh->dev[i].flags);
1598 }
ba22dcbf
N		 1599 if (atomic_read(&conf->disks[i].rdev->read_errors))
1600 atomic_set(&conf->disks[i].rdev->read_errors, 0);
1da177e4 1601 } else {
d6950432 1602 const char *bdn = bdevname(conf->disks[i].rdev->bdev, b);
ba22dcbf 1603 int retry = 0;
d6950432
N		 1604 rdev = conf->disks[i].rdev;
1605
1da177e4 1606 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
d6950432 1607 atomic_inc(&rdev->read_errors);
7b0bb536 1608 if (conf->mddev->degraded >= conf->max_degraded)
6be9d494 1609 printk_rl(KERN_WARNING
0c55e022 1610 "md/raid:%s: read error not correctable "
6be9d494
BS		 1611 "(sector %llu on %s).\n",
1612 mdname(conf->mddev),
1613 (unsigned long long)(sh->sector
1614 + rdev->data_offset),
1615 bdn);
ba22dcbf 1616 else if (test_bit(R5_ReWrite, &sh->dev[i].flags))
4e5314b5 1617 /* Oh, no!!! */
6be9d494 1618 printk_rl(KERN_WARNING
0c55e022 1619 "md/raid:%s: read error NOT corrected!! "
6be9d494
BS		 1620 "(sector %llu on %s).\n",
1621 mdname(conf->mddev),
1622 (unsigned long long)(sh->sector
1623 + rdev->data_offset),
1624 bdn);
d6950432 1625 else if (atomic_read(&rdev->read_errors)
ba22dcbf 1626 > conf->max_nr_stripes)
14f8d26b 1627 printk(KERN_WARNING
0c55e022 1628 "md/raid:%s: Too many read errors, failing device %s.\n",
d6950432 1629 mdname(conf->mddev), bdn);
ba22dcbf
N		 1630 else
1631 retry = 1;
1632 if (retry)
1633 set_bit(R5_ReadError, &sh->dev[i].flags);
1634 else {
4e5314b5
N		 1635 clear_bit(R5_ReadError, &sh->dev[i].flags);
1636 clear_bit(R5_ReWrite, &sh->dev[i].flags);
d6950432 1637 md_error(conf->mddev, rdev);
ba22dcbf 1638 }
1da177e4
LT		 1639 }
1640 rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
1da177e4
LT		 1641 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1642 set_bit(STRIPE_HANDLE, &sh->state);
1643 release_stripe(sh);
1da177e4
LT		 1644}
1645
d710e138 1646static void raid5_end_write_request(struct bio *bi, int error)
1da177e4 1647{
99c0fb5f 1648 struct stripe_head *sh = bi->bi_private;
1da177e4 1649 raid5_conf_t *conf = sh->raid_conf;
7ecaa1e6 1650 int disks = sh->disks, i;
1da177e4
LT		 1651 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
1652
1da177e4
LT		 1653 for (i=0 ; i<disks; i++)
1654 if (bi == &sh->dev[i].req)
1655 break;
1656
45b4233c 1657 pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n",
1da177e4
LT		 1658 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
1659 uptodate);
1660 if (i == disks) {
1661 BUG();
6712ecf8 1662 return;
1da177e4
LT		 1663 }
1664
1da177e4
LT		 1665 if (!uptodate)
1666 md_error(conf->mddev, conf->disks[i].rdev);
1667
1668 rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
1669
1670 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1671 set_bit(STRIPE_HANDLE, &sh->state);
c04be0aa 1672 release_stripe(sh);
1da177e4
LT		 1673}
1674
1675
784052ec 1676static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous);
1da177e4 1677
784052ec 1678static void raid5_build_block(struct stripe_head *sh, int i, int previous)
1da177e4
LT		 1679{
1680 struct r5dev *dev = &sh->dev[i];
1681
1682 bio_init(&dev->req);
1683 dev->req.bi_io_vec = &dev->vec;
1684 dev->req.bi_vcnt++;
1685 dev->req.bi_max_vecs++;
1686 dev->vec.bv_page = dev->page;
1687 dev->vec.bv_len = STRIPE_SIZE;
1688 dev->vec.bv_offset = 0;
1689
1690 dev->req.bi_sector = sh->sector;
1691 dev->req.bi_private = sh;
1692
1693 dev->flags = 0;
784052ec 1694 dev->sector = compute_blocknr(sh, i, previous);
1da177e4
LT		 1695}
1696
1697static void error(mddev_t *mddev, mdk_rdev_t *rdev)
1698{
1699 char b[BDEVNAME_SIZE];
7b92813c 1700 raid5_conf_t *conf = mddev->private;
0c55e022 1701 pr_debug("raid456: error called\n");
1da177e4 1702
b2d444d7 1703 if (!test_bit(Faulty, &rdev->flags)) {
850b2b42 1704 set_bit(MD_CHANGE_DEVS, &mddev->flags);
c04be0aa
N		 1705 if (test_and_clear_bit(In_sync, &rdev->flags)) {
1706 unsigned long flags;
1707 spin_lock_irqsave(&conf->device_lock, flags);
1da177e4 1708 mddev->degraded++;
c04be0aa 1709 spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4
LT		 1710 /*
1711 * if recovery was running, make sure it aborts.
1712 */
dfc70645 1713 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4 1714 }
b2d444d7 1715 set_bit(Faulty, &rdev->flags);
d710e138 1716 printk(KERN_ALERT
0c55e022 1717 "md/raid:%s: Disk failure on %s, disabling device.\n"
0c55e022
N		 1718 "md/raid:%s: Operation continuing on %d devices.\n",
1719 mdname(mddev),
1720 bdevname(rdev->bdev, b),
1721 mdname(mddev),
1722 conf->raid_disks - mddev->degraded);
1da177e4 1723 }
16a53ecc 1724}
1da177e4
LT		 1725
1726/*
1727 * Input: a 'big' sector number,
1728 * Output: index of the data and parity disk, and the sector # in them.
1729 */
112bf897 1730static sector_t raid5_compute_sector(raid5_conf_t *conf, sector_t r_sector,
911d4ee8
N		 1731 int previous, int *dd_idx,
1732 struct stripe_head *sh)
1da177e4 1733{
6e3b96ed 1734 sector_t stripe, stripe2;
35f2a591 1735 sector_t chunk_number;
1da177e4 1736 unsigned int chunk_offset;
911d4ee8 1737 int pd_idx, qd_idx;
67cc2b81 1738 int ddf_layout = 0;
1da177e4 1739 sector_t new_sector;
e183eaed
N		 1740 int algorithm = previous ? conf->prev_algo
1741 : conf->algorithm;
09c9e5fa
AN		 1742 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
1743 : conf->chunk_sectors;
112bf897
N		 1744 int raid_disks = previous ? conf->previous_raid_disks
1745 : conf->raid_disks;
1746 int data_disks = raid_disks - conf->max_degraded;
1da177e4
LT		 1747
1748 /* First compute the information on this sector */
1749
1750 /*
1751 * Compute the chunk number and the sector offset inside the chunk
1752 */
1753 chunk_offset = sector_div(r_sector, sectors_per_chunk);
1754 chunk_number = r_sector;
1da177e4
LT		 1755
1756 /*
1757 * Compute the stripe number
1758 */
35f2a591
N		 1759 stripe = chunk_number;
1760 *dd_idx = sector_div(stripe, data_disks);
6e3b96ed 1761 stripe2 = stripe;
1da177e4
LT		 1762 /*
1763 * Select the parity disk based on the user selected algorithm.
1764 */
911d4ee8 1765 pd_idx = qd_idx = ~0;
16a53ecc
N		 1766 switch(conf->level) {
1767 case 4:
911d4ee8 1768 pd_idx = data_disks;
16a53ecc
N		 1769 break;
1770 case 5:
e183eaed 1771 switch (algorithm) {
1da177e4 1772 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 1773 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 1774 if (*dd_idx >= pd_idx)
1da177e4
LT		 1775 (*dd_idx)++;
1776 break;
1777 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 1778 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 1779 if (*dd_idx >= pd_idx)
1da177e4
LT		 1780 (*dd_idx)++;
1781 break;
1782 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 1783 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 1784 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4
LT		 1785 break;
1786 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 1787 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 1788 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4 1789 break;
99c0fb5f
N		 1790 case ALGORITHM_PARITY_0:
1791 pd_idx = 0;
1792 (*dd_idx)++;
1793 break;
1794 case ALGORITHM_PARITY_N:
1795 pd_idx = data_disks;
1796 break;
1da177e4 1797 default:
99c0fb5f 1798 BUG();
16a53ecc
N		 1799 }
1800 break;
1801 case 6:
1802
e183eaed 1803 switch (algorithm) {
16a53ecc 1804 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 1805 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 1806 qd_idx = pd_idx + 1;
1807 if (pd_idx == raid_disks-1) {
99c0fb5f 1808 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 1809 qd_idx = 0;
1810 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 1811 (*dd_idx) += 2; /* D D P Q D */
1812 break;
1813 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 1814 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 1815 qd_idx = pd_idx + 1;
1816 if (pd_idx == raid_disks-1) {
99c0fb5f 1817 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 1818 qd_idx = 0;
1819 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 1820 (*dd_idx) += 2; /* D D P Q D */
1821 break;
1822 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 1823 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 1824 qd_idx = (pd_idx + 1) % raid_disks;
1825 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc
N		 1826 break;
1827 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 1828 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 1829 qd_idx = (pd_idx + 1) % raid_disks;
1830 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc 1831 break;
99c0fb5f
N		 1832
1833 case ALGORITHM_PARITY_0:
1834 pd_idx = 0;
1835 qd_idx = 1;
1836 (*dd_idx) += 2;
1837 break;
1838 case ALGORITHM_PARITY_N:
1839 pd_idx = data_disks;
1840 qd_idx = data_disks + 1;
1841 break;
1842
1843 case ALGORITHM_ROTATING_ZERO_RESTART:
1844 /* Exactly the same as RIGHT_ASYMMETRIC, but or
1845 * of blocks for computing Q is different.
1846 */
6e3b96ed 1847 pd_idx = sector_div(stripe2, raid_disks);
99c0fb5f
N		 1848 qd_idx = pd_idx + 1;
1849 if (pd_idx == raid_disks-1) {
1850 (*dd_idx)++; /* Q D D D P */
1851 qd_idx = 0;
1852 } else if (*dd_idx >= pd_idx)
1853 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 1854 ddf_layout = 1;
99c0fb5f
N		 1855 break;
1856
1857 case ALGORITHM_ROTATING_N_RESTART:
1858 /* Same a left_asymmetric, by first stripe is
1859 * D D D P Q rather than
1860 * Q D D D P
1861 */
6e3b96ed
N		 1862 stripe2 += 1;
1863 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 1864 qd_idx = pd_idx + 1;
1865 if (pd_idx == raid_disks-1) {
1866 (*dd_idx)++; /* Q D D D P */
1867 qd_idx = 0;
1868 } else if (*dd_idx >= pd_idx)
1869 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 1870 ddf_layout = 1;
99c0fb5f
N		 1871 break;
1872
1873 case ALGORITHM_ROTATING_N_CONTINUE:
1874 /* Same as left_symmetric but Q is before P */
6e3b96ed 1875 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 1876 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
1877 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
67cc2b81 1878 ddf_layout = 1;
99c0fb5f
N		 1879 break;
1880
1881 case ALGORITHM_LEFT_ASYMMETRIC_6:
1882 /* RAID5 left_asymmetric, with Q on last device */
6e3b96ed 1883 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 1884 if (*dd_idx >= pd_idx)
1885 (*dd_idx)++;
1886 qd_idx = raid_disks - 1;
1887 break;
1888
1889 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6e3b96ed 1890 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 1891 if (*dd_idx >= pd_idx)
1892 (*dd_idx)++;
1893 qd_idx = raid_disks - 1;
1894 break;
1895
1896 case ALGORITHM_LEFT_SYMMETRIC_6:
6e3b96ed 1897 pd_idx = data_disks - 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_RIGHT_SYMMETRIC_6:
6e3b96ed 1903 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 1904 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
1905 qd_idx = raid_disks - 1;
1906 break;
1907
1908 case ALGORITHM_PARITY_0_6:
1909 pd_idx = 0;
1910 (*dd_idx)++;
1911 qd_idx = raid_disks - 1;
1912 break;
1913
16a53ecc 1914 default:
99c0fb5f 1915 BUG();
16a53ecc
N		 1916 }
1917 break;
1da177e4
LT		 1918 }
1919
911d4ee8
N		 1920 if (sh) {
1921 sh->pd_idx = pd_idx;
1922 sh->qd_idx = qd_idx;
67cc2b81 1923 sh->ddf_layout = ddf_layout;
911d4ee8 1924 }
1da177e4
LT		 1925 /*
1926 * Finally, compute the new sector number
1927 */
1928 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
1929 return new_sector;
1930}
1931
1932
784052ec 1933static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous)
1da177e4
LT		 1934{
1935 raid5_conf_t *conf = sh->raid_conf;
b875e531
N		 1936 int raid_disks = sh->disks;
1937 int data_disks = raid_disks - conf->max_degraded;
1da177e4 1938 sector_t new_sector = sh->sector, check;
09c9e5fa
AN		 1939 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
1940 : conf->chunk_sectors;
e183eaed
N		 1941 int algorithm = previous ? conf->prev_algo
1942 : conf->algorithm;
1da177e4
LT		 1943 sector_t stripe;
1944 int chunk_offset;
35f2a591
N		 1945 sector_t chunk_number;
1946 int dummy1, dd_idx = i;
1da177e4 1947 sector_t r_sector;
911d4ee8 1948 struct stripe_head sh2;
1da177e4 1949
16a53ecc 1950
1da177e4
LT		 1951 chunk_offset = sector_div(new_sector, sectors_per_chunk);
1952 stripe = new_sector;
1da177e4 1953
16a53ecc
N		 1954 if (i == sh->pd_idx)
1955 return 0;
1956 switch(conf->level) {
1957 case 4: break;
1958 case 5:
e183eaed 1959 switch (algorithm) {
1da177e4
LT		 1960 case ALGORITHM_LEFT_ASYMMETRIC:
1961 case ALGORITHM_RIGHT_ASYMMETRIC:
1962 if (i > sh->pd_idx)
1963 i--;
1964 break;
1965 case ALGORITHM_LEFT_SYMMETRIC:
1966 case ALGORITHM_RIGHT_SYMMETRIC:
1967 if (i < sh->pd_idx)
1968 i += raid_disks;
1969 i -= (sh->pd_idx + 1);
1970 break;
99c0fb5f
N		 1971 case ALGORITHM_PARITY_0:
1972 i -= 1;
1973 break;
1974 case ALGORITHM_PARITY_N:
1975 break;
1da177e4 1976 default:
99c0fb5f 1977 BUG();
16a53ecc
N		 1978 }
1979 break;
1980 case 6:
d0dabf7e 1981 if (i == sh->qd_idx)
16a53ecc 1982 return 0; /* It is the Q disk */
e183eaed 1983 switch (algorithm) {
16a53ecc
N		 1984 case ALGORITHM_LEFT_ASYMMETRIC:
1985 case ALGORITHM_RIGHT_ASYMMETRIC:
99c0fb5f
N		 1986 case ALGORITHM_ROTATING_ZERO_RESTART:
1987 case ALGORITHM_ROTATING_N_RESTART:
1988 if (sh->pd_idx == raid_disks-1)
1989 i--; /* Q D D D P */
16a53ecc
N		 1990 else if (i > sh->pd_idx)
1991 i -= 2; /* D D P Q D */
1992 break;
1993 case ALGORITHM_LEFT_SYMMETRIC:
1994 case ALGORITHM_RIGHT_SYMMETRIC:
1995 if (sh->pd_idx == raid_disks-1)
1996 i--; /* Q D D D P */
1997 else {
1998 /* D D P Q D */
1999 if (i < sh->pd_idx)
2000 i += raid_disks;
2001 i -= (sh->pd_idx + 2);
2002 }
2003 break;
99c0fb5f
N		 2004 case ALGORITHM_PARITY_0:
2005 i -= 2;
2006 break;
2007 case ALGORITHM_PARITY_N:
2008 break;
2009 case ALGORITHM_ROTATING_N_CONTINUE:
e4424fee 2010 /* Like left_symmetric, but P is before Q */
99c0fb5f
N		 2011 if (sh->pd_idx == 0)
2012 i--; /* P D D D Q */
e4424fee
N		 2013 else {
2014 /* D D Q P D */
2015 if (i < sh->pd_idx)
2016 i += raid_disks;
2017 i -= (sh->pd_idx + 1);
2018 }
99c0fb5f
N		 2019 break;
2020 case ALGORITHM_LEFT_ASYMMETRIC_6:
2021 case ALGORITHM_RIGHT_ASYMMETRIC_6:
2022 if (i > sh->pd_idx)
2023 i--;
2024 break;
2025 case ALGORITHM_LEFT_SYMMETRIC_6:
2026 case ALGORITHM_RIGHT_SYMMETRIC_6:
2027 if (i < sh->pd_idx)
2028 i += data_disks + 1;
2029 i -= (sh->pd_idx + 1);
2030 break;
2031 case ALGORITHM_PARITY_0_6:
2032 i -= 1;
2033 break;
16a53ecc 2034 default:
99c0fb5f 2035 BUG();
16a53ecc
N		 2036 }
2037 break;
1da177e4
LT		 2038 }
2039
2040 chunk_number = stripe * data_disks + i;
35f2a591 2041 r_sector = chunk_number * sectors_per_chunk + chunk_offset;
1da177e4 2042
112bf897 2043 check = raid5_compute_sector(conf, r_sector,
784052ec 2044 previous, &dummy1, &sh2);
911d4ee8
N		 2045 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
2046 || sh2.qd_idx != sh->qd_idx) {
0c55e022
N		 2047 printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n",
2048 mdname(conf->mddev));
1da177e4
LT		 2049 return 0;
2050 }
2051 return r_sector;
2052}
2053
2054
600aa109 2055static void
c0f7bddb 2056schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
600aa109 2057 int rcw, int expand)
e33129d8
DW		 2058{
2059 int i, pd_idx = sh->pd_idx, disks = sh->disks;
c0f7bddb
YT		 2060 raid5_conf_t *conf = sh->raid_conf;
2061 int level = conf->level;
e33129d8
DW		 2062
2063 if (rcw) {
2064 /* if we are not expanding this is a proper write request, and
2065 * there will be bios with new data to be drained into the
2066 * stripe cache
2067 */
2068 if (!expand) {
600aa109
DW		 2069 sh->reconstruct_state = reconstruct_state_drain_run;
2070 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2071 } else
2072 sh->reconstruct_state = reconstruct_state_run;
16a53ecc 2073
ac6b53b6 2074 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
e33129d8
DW		 2075
2076 for (i = disks; i--; ) {
2077 struct r5dev *dev = &sh->dev[i];
2078
2079 if (dev->towrite) {
2080 set_bit(R5_LOCKED, &dev->flags);
d8ee0728 2081 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2082 if (!expand)
2083 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2084 s->locked++;
e33129d8
DW		 2085 }
2086 }
c0f7bddb 2087 if (s->locked + conf->max_degraded == disks)
8b3e6cdc 2088 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
c0f7bddb 2089 atomic_inc(&conf->pending_full_writes);
e33129d8 2090 } else {
c0f7bddb 2091 BUG_ON(level == 6);
e33129d8
DW		 2092 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
2093 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
2094
d8ee0728 2095 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
600aa109
DW		 2096 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
2097 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
ac6b53b6 2098 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
e33129d8
DW		 2099
2100 for (i = disks; i--; ) {
2101 struct r5dev *dev = &sh->dev[i];
2102 if (i == pd_idx)
2103 continue;
2104
e33129d8
DW		 2105 if (dev->towrite &&
2106 (test_bit(R5_UPTODATE, &dev->flags) ||
d8ee0728
DW		 2107 test_bit(R5_Wantcompute, &dev->flags))) {
2108 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2109 set_bit(R5_LOCKED, &dev->flags);
2110 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2111 s->locked++;
e33129d8
DW		 2112 }
2113 }
2114 }
2115
c0f7bddb 2116 /* keep the parity disk(s) locked while asynchronous operations
e33129d8
DW		 2117 * are in flight
2118 */
2119 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
2120 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
600aa109 2121 s->locked++;
e33129d8 2122
c0f7bddb
YT		 2123 if (level == 6) {
2124 int qd_idx = sh->qd_idx;
2125 struct r5dev *dev = &sh->dev[qd_idx];
2126
2127 set_bit(R5_LOCKED, &dev->flags);
2128 clear_bit(R5_UPTODATE, &dev->flags);
2129 s->locked++;
2130 }
2131
600aa109 2132 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
e46b272b 2133 __func__, (unsigned long long)sh->sector,
600aa109 2134 s->locked, s->ops_request);
e33129d8 2135}
16a53ecc 2136
1da177e4
LT		 2137/*
2138 * Each stripe/dev can have one or more bion attached.
16a53ecc 2139 * toread/towrite point to the first in a chain.
1da177e4
LT		 2140 * The bi_next chain must be in order.
2141 */
2142static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite)
2143{
2144 struct bio **bip;
2145 raid5_conf_t *conf = sh->raid_conf;
72626685 2146 int firstwrite=0;
1da177e4 2147
45b4233c 2148 pr_debug("adding bh b#%llu to stripe s#%llu\n",
1da177e4
LT		 2149 (unsigned long long)bi->bi_sector,
2150 (unsigned long long)sh->sector);
2151
2152
2153 spin_lock(&sh->lock);
2154 spin_lock_irq(&conf->device_lock);
72626685 2155 if (forwrite) {
1da177e4 2156 bip = &sh->dev[dd_idx].towrite;
72626685
N		 2157 if (*bip == NULL && sh->dev[dd_idx].written == NULL)
2158 firstwrite = 1;
2159 } else
1da177e4
LT		 2160 bip = &sh->dev[dd_idx].toread;
2161 while (*bip && (*bip)->bi_sector < bi->bi_sector) {
2162 if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector)
2163 goto overlap;
2164 bip = & (*bip)->bi_next;
2165 }
2166 if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9))
2167 goto overlap;
2168
78bafebd 2169 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
1da177e4
LT		 2170 if (*bip)
2171 bi->bi_next = *bip;
2172 *bip = bi;
960e739d 2173 bi->bi_phys_segments++;
1da177e4
LT		 2174 spin_unlock_irq(&conf->device_lock);
2175 spin_unlock(&sh->lock);
2176
45b4233c 2177 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
1da177e4
LT		 2178 (unsigned long long)bi->bi_sector,
2179 (unsigned long long)sh->sector, dd_idx);
2180
72626685 2181 if (conf->mddev->bitmap && firstwrite) {
72626685
N		 2182 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
2183 STRIPE_SECTORS, 0);
ae3c20cc 2184 sh->bm_seq = conf->seq_flush+1;
72626685
N		 2185 set_bit(STRIPE_BIT_DELAY, &sh->state);
2186 }
2187
1da177e4
LT		 2188 if (forwrite) {
2189 /* check if page is covered */
2190 sector_t sector = sh->dev[dd_idx].sector;
2191 for (bi=sh->dev[dd_idx].towrite;
2192 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
2193 bi && bi->bi_sector <= sector;
2194 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
2195 if (bi->bi_sector + (bi->bi_size>>9) >= sector)
2196 sector = bi->bi_sector + (bi->bi_size>>9);
2197 }
2198 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
2199 set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags);
2200 }
2201 return 1;
2202
2203 overlap:
2204 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
2205 spin_unlock_irq(&conf->device_lock);
2206 spin_unlock(&sh->lock);
2207 return 0;
2208}
2209
29269553
N		 2210static void end_reshape(raid5_conf_t *conf);
2211
911d4ee8
N		 2212static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous,
2213 struct stripe_head *sh)
ccfcc3c1 2214{
784052ec 2215 int sectors_per_chunk =
09c9e5fa 2216 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
911d4ee8 2217 int dd_idx;
2d2063ce 2218 int chunk_offset = sector_div(stripe, sectors_per_chunk);
112bf897 2219 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
2d2063ce 2220
112bf897
N		 2221 raid5_compute_sector(conf,
2222 stripe * (disks - conf->max_degraded)
b875e531 2223 *sectors_per_chunk + chunk_offset,
112bf897 2224 previous,
911d4ee8 2225 &dd_idx, sh);
ccfcc3c1
N		 2226}
2227
a4456856 2228static void
1fe797e6 2229handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh,
a4456856
DW		 2230 struct stripe_head_state *s, int disks,
2231 struct bio **return_bi)
2232{
2233 int i;
2234 for (i = disks; i--; ) {
2235 struct bio *bi;
2236 int bitmap_end = 0;
2237
2238 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
2239 mdk_rdev_t *rdev;
2240 rcu_read_lock();
2241 rdev = rcu_dereference(conf->disks[i].rdev);
2242 if (rdev && test_bit(In_sync, &rdev->flags))
2243 /* multiple read failures in one stripe */
2244 md_error(conf->mddev, rdev);
2245 rcu_read_unlock();
2246 }
2247 spin_lock_irq(&conf->device_lock);
2248 /* fail all writes first */
2249 bi = sh->dev[i].towrite;
2250 sh->dev[i].towrite = NULL;
2251 if (bi) {
2252 s->to_write--;
2253 bitmap_end = 1;
2254 }
2255
2256 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2257 wake_up(&conf->wait_for_overlap);
2258
2259 while (bi && bi->bi_sector <
2260 sh->dev[i].sector + STRIPE_SECTORS) {
2261 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
2262 clear_bit(BIO_UPTODATE, &bi->bi_flags);
960e739d 2263 if (!raid5_dec_bi_phys_segments(bi)) {
a4456856
DW		 2264 md_write_end(conf->mddev);
2265 bi->bi_next = *return_bi;
2266 *return_bi = bi;
2267 }
2268 bi = nextbi;
2269 }
2270 /* and fail all 'written' */
2271 bi = sh->dev[i].written;
2272 sh->dev[i].written = NULL;
2273 if (bi) bitmap_end = 1;
2274 while (bi && bi->bi_sector <
2275 sh->dev[i].sector + STRIPE_SECTORS) {
2276 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
2277 clear_bit(BIO_UPTODATE, &bi->bi_flags);
960e739d 2278 if (!raid5_dec_bi_phys_segments(bi)) {
a4456856
DW		 2279 md_write_end(conf->mddev);
2280 bi->bi_next = *return_bi;
2281 *return_bi = bi;
2282 }
2283 bi = bi2;
2284 }
2285
b5e98d65
DW		 2286 /* fail any reads if this device is non-operational and
2287 * the data has not reached the cache yet.
2288 */
2289 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
2290 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
2291 test_bit(R5_ReadError, &sh->dev[i].flags))) {
a4456856
DW		 2292 bi = sh->dev[i].toread;
2293 sh->dev[i].toread = NULL;
2294 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2295 wake_up(&conf->wait_for_overlap);
2296 if (bi) s->to_read--;
2297 while (bi && bi->bi_sector <
2298 sh->dev[i].sector + STRIPE_SECTORS) {
2299 struct bio *nextbi =
2300 r5_next_bio(bi, sh->dev[i].sector);
2301 clear_bit(BIO_UPTODATE, &bi->bi_flags);
960e739d 2302 if (!raid5_dec_bi_phys_segments(bi)) {
a4456856
DW		 2303 bi->bi_next = *return_bi;
2304 *return_bi = bi;
2305 }
2306 bi = nextbi;
2307 }
2308 }
2309 spin_unlock_irq(&conf->device_lock);
2310 if (bitmap_end)
2311 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
2312 STRIPE_SECTORS, 0, 0);
2313 }
2314
8b3e6cdc
DW		 2315 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2316 if (atomic_dec_and_test(&conf->pending_full_writes))
2317 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 2318}
2319
1fe797e6
DW		 2320/* fetch_block5 - checks the given member device to see if its data needs
2321 * to be read or computed to satisfy a request.
2322 *
2323 * Returns 1 when no more member devices need to be checked, otherwise returns
2324 * 0 to tell the loop in handle_stripe_fill5 to continue
f38e1219 2325 */
1fe797e6
DW		 2326static int fetch_block5(struct stripe_head *sh, struct stripe_head_state *s,
2327 int disk_idx, int disks)
f38e1219
DW		 2328{
2329 struct r5dev *dev = &sh->dev[disk_idx];
2330 struct r5dev *failed_dev = &sh->dev[s->failed_num];
2331
f38e1219
DW		 2332 /* is the data in this block needed, and can we get it? */
2333 if (!test_bit(R5_LOCKED, &dev->flags) &&
1fe797e6
DW		 2334 !test_bit(R5_UPTODATE, &dev->flags) &&
2335 (dev->toread ||
2336 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
2337 s->syncing || s->expanding ||
2338 (s->failed &&
2339 (failed_dev->toread ||
2340 (failed_dev->towrite &&
2341 !test_bit(R5_OVERWRITE, &failed_dev->flags)))))) {
976ea8d4
DW		 2342 /* We would like to get this block, possibly by computing it,
2343 * otherwise read it if the backing disk is insync
f38e1219
DW		 2344 */
2345 if ((s->uptodate == disks - 1) &&
ecc65c9b 2346 (s->failed && disk_idx == s->failed_num)) {
976ea8d4
DW		 2347 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2348 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
f38e1219
DW		 2349 set_bit(R5_Wantcompute, &dev->flags);
2350 sh->ops.target = disk_idx;
ac6b53b6 2351 sh->ops.target2 = -1;
f38e1219 2352 s->req_compute = 1;
f38e1219 2353 /* Careful: from this point on 'uptodate' is in the eye
ac6b53b6 2354 * of raid_run_ops which services 'compute' operations
f38e1219
DW		 2355 * before writes. R5_Wantcompute flags a block that will
2356 * be R5_UPTODATE by the time it is needed for a
2357 * subsequent operation.
2358 */
2359 s->uptodate++;
1fe797e6 2360 return 1; /* uptodate + compute == disks */
7a1fc53c 2361 } else if (test_bit(R5_Insync, &dev->flags)) {
f38e1219
DW		 2362 set_bit(R5_LOCKED, &dev->flags);
2363 set_bit(R5_Wantread, &dev->flags);
f38e1219
DW		 2364 s->locked++;
2365 pr_debug("Reading block %d (sync=%d)\n", disk_idx,
2366 s->syncing);
2367 }
2368 }
2369
1fe797e6 2370 return 0;
f38e1219
DW		 2371}
2372
1fe797e6
DW		 2373/**
2374 * handle_stripe_fill5 - read or compute data to satisfy pending requests.
2375 */
2376static void handle_stripe_fill5(struct stripe_head *sh,
a4456856
DW		 2377 struct stripe_head_state *s, int disks)
2378{
2379 int i;
f38e1219 2380
f38e1219
DW		 2381 /* look for blocks to read/compute, skip this if a compute
2382 * is already in flight, or if the stripe contents are in the
2383 * midst of changing due to a write
2384 */
976ea8d4 2385 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
1fe797e6 2386 !sh->reconstruct_state)
f38e1219 2387 for (i = disks; i--; )
1fe797e6 2388 if (fetch_block5(sh, s, i, disks))
f38e1219 2389 break;
a4456856
DW		 2390 set_bit(STRIPE_HANDLE, &sh->state);
2391}
2392
5599becc
YT		 2393/* fetch_block6 - checks the given member device to see if its data needs
2394 * to be read or computed to satisfy a request.
2395 *
2396 * Returns 1 when no more member devices need to be checked, otherwise returns
2397 * 0 to tell the loop in handle_stripe_fill6 to continue
2398 */
2399static int fetch_block6(struct stripe_head *sh, struct stripe_head_state *s,
2400 struct r6_state *r6s, int disk_idx, int disks)
a4456856 2401{
5599becc
YT		 2402 struct r5dev *dev = &sh->dev[disk_idx];
2403 struct r5dev *fdev[2] = { &sh->dev[r6s->failed_num[0]],
2404 &sh->dev[r6s->failed_num[1]] };
2405
2406 if (!test_bit(R5_LOCKED, &dev->flags) &&
2407 !test_bit(R5_UPTODATE, &dev->flags) &&
2408 (dev->toread ||
2409 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
2410 s->syncing || s->expanding ||
2411 (s->failed >= 1 &&
2412 (fdev[0]->toread || s->to_write)) ||
2413 (s->failed >= 2 &&
2414 (fdev[1]->toread || s->to_write)))) {
2415 /* we would like to get this block, possibly by computing it,
2416 * otherwise read it if the backing disk is insync
2417 */
2418 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
2419 BUG_ON(test_bit(R5_Wantread, &dev->flags));
2420 if ((s->uptodate == disks - 1) &&
2421 (s->failed && (disk_idx == r6s->failed_num[0] ||
2422 disk_idx == r6s->failed_num[1]))) {
2423 /* have disk failed, and we're requested to fetch it;
2424 * do compute it
a4456856 2425 */
5599becc
YT		 2426 pr_debug("Computing stripe %llu block %d\n",
2427 (unsigned long long)sh->sector, disk_idx);
2428 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2429 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2430 set_bit(R5_Wantcompute, &dev->flags);
2431 sh->ops.target = disk_idx;
2432 sh->ops.target2 = -1; /* no 2nd target */
2433 s->req_compute = 1;
2434 s->uptodate++;
2435 return 1;
2436 } else if (s->uptodate == disks-2 && s->failed >= 2) {
2437 /* Computing 2-failure is *very* expensive; only
2438 * do it if failed >= 2
2439 */
2440 int other;
2441 for (other = disks; other--; ) {
2442 if (other == disk_idx)
2443 continue;
2444 if (!test_bit(R5_UPTODATE,
2445 &sh->dev[other].flags))
2446 break;
a4456856 2447 }
5599becc
YT		 2448 BUG_ON(other < 0);
2449 pr_debug("Computing stripe %llu blocks %d,%d\n",
2450 (unsigned long long)sh->sector,
2451 disk_idx, other);
2452 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2453 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2454 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
2455 set_bit(R5_Wantcompute, &sh->dev[other].flags);
2456 sh->ops.target = disk_idx;
2457 sh->ops.target2 = other;
2458 s->uptodate += 2;
2459 s->req_compute = 1;
2460 return 1;
2461 } else if (test_bit(R5_Insync, &dev->flags)) {
2462 set_bit(R5_LOCKED, &dev->flags);
2463 set_bit(R5_Wantread, &dev->flags);
2464 s->locked++;
2465 pr_debug("Reading block %d (sync=%d)\n",
2466 disk_idx, s->syncing);
a4456856
DW		 2467 }
2468 }
5599becc
YT		 2469
2470 return 0;
2471}
2472
2473/**
2474 * handle_stripe_fill6 - read or compute data to satisfy pending requests.
2475 */
2476static void handle_stripe_fill6(struct stripe_head *sh,
2477 struct stripe_head_state *s, struct r6_state *r6s,
2478 int disks)
2479{
2480 int i;
2481
2482 /* look for blocks to read/compute, skip this if a compute
2483 * is already in flight, or if the stripe contents are in the
2484 * midst of changing due to a write
2485 */
2486 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
2487 !sh->reconstruct_state)
2488 for (i = disks; i--; )
2489 if (fetch_block6(sh, s, r6s, i, disks))
2490 break;
a4456856
DW		 2491 set_bit(STRIPE_HANDLE, &sh->state);
2492}
2493
2494
1fe797e6 2495/* handle_stripe_clean_event
a4456856
DW		 2496 * any written block on an uptodate or failed drive can be returned.
2497 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
2498 * never LOCKED, so we don't need to test 'failed' directly.
2499 */
1fe797e6 2500static void handle_stripe_clean_event(raid5_conf_t *conf,
a4456856
DW		 2501 struct stripe_head *sh, int disks, struct bio **return_bi)
2502{
2503 int i;
2504 struct r5dev *dev;
2505
2506 for (i = disks; i--; )
2507 if (sh->dev[i].written) {
2508 dev = &sh->dev[i];
2509 if (!test_bit(R5_LOCKED, &dev->flags) &&
2510 test_bit(R5_UPTODATE, &dev->flags)) {
2511 /* We can return any write requests */
2512 struct bio *wbi, *wbi2;
2513 int bitmap_end = 0;
45b4233c 2514 pr_debug("Return write for disc %d\n", i);
a4456856
DW		 2515 spin_lock_irq(&conf->device_lock);
2516 wbi = dev->written;
2517 dev->written = NULL;
2518 while (wbi && wbi->bi_sector <
2519 dev->sector + STRIPE_SECTORS) {
2520 wbi2 = r5_next_bio(wbi, dev->sector);
960e739d 2521 if (!raid5_dec_bi_phys_segments(wbi)) {
a4456856
DW		 2522 md_write_end(conf->mddev);
2523 wbi->bi_next = *return_bi;
2524 *return_bi = wbi;
2525 }
2526 wbi = wbi2;
2527 }
2528 if (dev->towrite == NULL)
2529 bitmap_end = 1;
2530 spin_unlock_irq(&conf->device_lock);
2531 if (bitmap_end)
2532 bitmap_endwrite(conf->mddev->bitmap,
2533 sh->sector,
2534 STRIPE_SECTORS,
2535 !test_bit(STRIPE_DEGRADED, &sh->state),
2536 0);
2537 }
2538 }
8b3e6cdc
DW		 2539
2540 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2541 if (atomic_dec_and_test(&conf->pending_full_writes))
2542 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 2543}
2544
1fe797e6 2545static void handle_stripe_dirtying5(raid5_conf_t *conf,
a4456856
DW		 2546 struct stripe_head *sh, struct stripe_head_state *s, int disks)
2547{
2548 int rmw = 0, rcw = 0, i;
2549 for (i = disks; i--; ) {
2550 /* would I have to read this buffer for read_modify_write */
2551 struct r5dev *dev = &sh->dev[i];
2552 if ((dev->towrite || i == sh->pd_idx) &&
2553 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 2554 !(test_bit(R5_UPTODATE, &dev->flags) ||
2555 test_bit(R5_Wantcompute, &dev->flags))) {
a4456856
DW		 2556 if (test_bit(R5_Insync, &dev->flags))
2557 rmw++;
2558 else
2559 rmw += 2*disks; /* cannot read it */
2560 }
2561 /* Would I have to read this buffer for reconstruct_write */
2562 if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx &&
2563 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 2564 !(test_bit(R5_UPTODATE, &dev->flags) ||
2565 test_bit(R5_Wantcompute, &dev->flags))) {
2566 if (test_bit(R5_Insync, &dev->flags)) rcw++;
a4456856
DW		 2567 else
2568 rcw += 2*disks;
2569 }
2570 }
45b4233c 2571 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
a4456856
DW		 2572 (unsigned long long)sh->sector, rmw, rcw);
2573 set_bit(STRIPE_HANDLE, &sh->state);
2574 if (rmw < rcw && rmw > 0)
2575 /* prefer read-modify-write, but need to get some data */
2576 for (i = disks; i--; ) {
2577 struct r5dev *dev = &sh->dev[i];
2578 if ((dev->towrite || i == sh->pd_idx) &&
2579 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 2580 !(test_bit(R5_UPTODATE, &dev->flags) ||
2581 test_bit(R5_Wantcompute, &dev->flags)) &&
a4456856
DW		 2582 test_bit(R5_Insync, &dev->flags)) {
2583 if (
2584 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
45b4233c 2585 pr_debug("Read_old block "
a4456856
DW		 2586 "%d for r-m-w\n", i);
2587 set_bit(R5_LOCKED, &dev->flags);
2588 set_bit(R5_Wantread, &dev->flags);
2589 s->locked++;
2590 } else {
2591 set_bit(STRIPE_DELAYED, &sh->state);
2592 set_bit(STRIPE_HANDLE, &sh->state);
2593 }
2594 }
2595 }
2596 if (rcw <= rmw && rcw > 0)
2597 /* want reconstruct write, but need to get some data */
2598 for (i = disks; i--; ) {
2599 struct r5dev *dev = &sh->dev[i];
2600 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
2601 i != sh->pd_idx &&
2602 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 2603 !(test_bit(R5_UPTODATE, &dev->flags) ||
2604 test_bit(R5_Wantcompute, &dev->flags)) &&
a4456856
DW		 2605 test_bit(R5_Insync, &dev->flags)) {
2606 if (
2607 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
45b4233c 2608 pr_debug("Read_old block "
a4456856
DW		 2609 "%d for Reconstruct\n", i);
2610 set_bit(R5_LOCKED, &dev->flags);
2611 set_bit(R5_Wantread, &dev->flags);
2612 s->locked++;
2613 } else {
2614 set_bit(STRIPE_DELAYED, &sh->state);
2615 set_bit(STRIPE_HANDLE, &sh->state);
2616 }
2617 }
2618 }
2619 /* now if nothing is locked, and if we have enough data,
2620 * we can start a write request
2621 */
f38e1219
DW		 2622 /* since handle_stripe can be called at any time we need to handle the
2623 * case where a compute block operation has been submitted and then a
ac6b53b6
DW		 2624 * subsequent call wants to start a write request. raid_run_ops only
2625 * handles the case where compute block and reconstruct are requested
f38e1219
DW		 2626 * simultaneously. If this is not the case then new writes need to be
2627 * held off until the compute completes.
2628 */
976ea8d4
DW		 2629 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
2630 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
2631 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
c0f7bddb 2632 schedule_reconstruction(sh, s, rcw == 0, 0);
a4456856
DW		 2633}
2634
1fe797e6 2635static void handle_stripe_dirtying6(raid5_conf_t *conf,
a4456856
DW		 2636 struct stripe_head *sh, struct stripe_head_state *s,
2637 struct r6_state *r6s, int disks)
2638{
a9b39a74 2639 int rcw = 0, pd_idx = sh->pd_idx, i;
34e04e87 2640 int qd_idx = sh->qd_idx;
a9b39a74
YT		 2641
2642 set_bit(STRIPE_HANDLE, &sh->state);
a4456856
DW		 2643 for (i = disks; i--; ) {
2644 struct r5dev *dev = &sh->dev[i];
a9b39a74
YT		 2645 /* check if we haven't enough data */
2646 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
2647 i != pd_idx && i != qd_idx &&
2648 !test_bit(R5_LOCKED, &dev->flags) &&
2649 !(test_bit(R5_UPTODATE, &dev->flags) ||
2650 test_bit(R5_Wantcompute, &dev->flags))) {
2651 rcw++;
2652 if (!test_bit(R5_Insync, &dev->flags))
2653 continue; /* it's a failed drive */
2654
2655 if (
2656 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
2657 pr_debug("Read_old stripe %llu "
2658 "block %d for Reconstruct\n",
2659 (unsigned long long)sh->sector, i);
2660 set_bit(R5_LOCKED, &dev->flags);
2661 set_bit(R5_Wantread, &dev->flags);
2662 s->locked++;
2663 } else {
2664 pr_debug("Request delayed stripe %llu "
2665 "block %d for Reconstruct\n",
2666 (unsigned long long)sh->sector, i);
2667 set_bit(STRIPE_DELAYED, &sh->state);
2668 set_bit(STRIPE_HANDLE, &sh->state);
a4456856
DW		 2669 }
2670 }
2671 }
a4456856
DW		 2672 /* now if nothing is locked, and if we have enough data, we can start a
2673 * write request
2674 */
a9b39a74
YT		 2675 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
2676 s->locked == 0 && rcw == 0 &&
a4456856 2677 !test_bit(STRIPE_BIT_DELAY, &sh->state)) {
a9b39a74 2678 schedule_reconstruction(sh, s, 1, 0);
a4456856
DW		 2679 }
2680}
2681
2682static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh,
2683 struct stripe_head_state *s, int disks)
2684{
ecc65c9b 2685 struct r5dev *dev = NULL;
bd2ab670 2686
a4456856 2687 set_bit(STRIPE_HANDLE, &sh->state);
e89f8962 2688
ecc65c9b
DW		 2689 switch (sh->check_state) {
2690 case check_state_idle:
2691 /* start a new check operation if there are no failures */
bd2ab670 2692 if (s->failed == 0) {
bd2ab670 2693 BUG_ON(s->uptodate != disks);
ecc65c9b
DW		 2694 sh->check_state = check_state_run;
2695 set_bit(STRIPE_OP_CHECK, &s->ops_request);
bd2ab670 2696 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
bd2ab670 2697 s->uptodate--;
ecc65c9b 2698 break;
bd2ab670 2699 }
ecc65c9b
DW		 2700 dev = &sh->dev[s->failed_num];
2701 /* fall through */
2702 case check_state_compute_result:
2703 sh->check_state = check_state_idle;
2704 if (!dev)
2705 dev = &sh->dev[sh->pd_idx];
2706
2707 /* check that a write has not made the stripe insync */
2708 if (test_bit(STRIPE_INSYNC, &sh->state))
2709 break;
c8894419 2710
a4456856 2711 /* either failed parity check, or recovery is happening */
a4456856
DW		 2712 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
2713 BUG_ON(s->uptodate != disks);
2714
2715 set_bit(R5_LOCKED, &dev->flags);
ecc65c9b 2716 s->locked++;
a4456856 2717 set_bit(R5_Wantwrite, &dev->flags);
830ea016 2718
a4456856 2719 clear_bit(STRIPE_DEGRADED, &sh->state);
a4456856 2720 set_bit(STRIPE_INSYNC, &sh->state);
ecc65c9b
DW		 2721 break;
2722 case check_state_run:
2723 break; /* we will be called again upon completion */
2724 case check_state_check_result:
2725 sh->check_state = check_state_idle;
2726
2727 /* if a failure occurred during the check operation, leave
2728 * STRIPE_INSYNC not set and let the stripe be handled again
2729 */
2730 if (s->failed)
2731 break;
2732
2733 /* handle a successful check operation, if parity is correct
2734 * we are done. Otherwise update the mismatch count and repair
2735 * parity if !MD_RECOVERY_CHECK
2736 */
ad283ea4 2737 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
ecc65c9b
DW		 2738 /* parity is correct (on disc,
2739 * not in buffer any more)
2740 */
2741 set_bit(STRIPE_INSYNC, &sh->state);
2742 else {
2743 conf->mddev->resync_mismatches += STRIPE_SECTORS;
2744 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
2745 /* don't try to repair!! */
2746 set_bit(STRIPE_INSYNC, &sh->state);
2747 else {
2748 sh->check_state = check_state_compute_run;
976ea8d4 2749 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
ecc65c9b
DW		 2750 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2751 set_bit(R5_Wantcompute,
2752 &sh->dev[sh->pd_idx].flags);
2753 sh->ops.target = sh->pd_idx;
ac6b53b6 2754 sh->ops.target2 = -1;
ecc65c9b
DW		 2755 s->uptodate++;
2756 }
2757 }
2758 break;
2759 case check_state_compute_run:
2760 break;
2761 default:
2762 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
2763 __func__, sh->check_state,
2764 (unsigned long long) sh->sector);
2765 BUG();
a4456856
DW		 2766 }
2767}
2768
2769
2770static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh,
36d1c647
DW		 2771 struct stripe_head_state *s,
2772 struct r6_state *r6s, int disks)
a4456856 2773{
a4456856 2774 int pd_idx = sh->pd_idx;
34e04e87 2775 int qd_idx = sh->qd_idx;
d82dfee0 2776 struct r5dev *dev;
a4456856
DW		 2777
2778 set_bit(STRIPE_HANDLE, &sh->state);
2779
2780 BUG_ON(s->failed > 2);
d82dfee0 2781
a4456856
DW		 2782 /* Want to check and possibly repair P and Q.
2783 * However there could be one 'failed' device, in which
2784 * case we can only check one of them, possibly using the
2785 * other to generate missing data
2786 */
2787
d82dfee0
DW		 2788 switch (sh->check_state) {
2789 case check_state_idle:
2790 /* start a new check operation if there are < 2 failures */
a4456856 2791 if (s->failed == r6s->q_failed) {
d82dfee0 2792 /* The only possible failed device holds Q, so it
a4456856
DW		 2793 * makes sense to check P (If anything else were failed,
2794 * we would have used P to recreate it).
2795 */
d82dfee0 2796 sh->check_state = check_state_run;
a4456856
DW		 2797 }
2798 if (!r6s->q_failed && s->failed < 2) {
d82dfee0 2799 /* Q is not failed, and we didn't use it to generate
a4456856
DW		 2800 * anything, so it makes sense to check it
2801 */
d82dfee0
DW		 2802 if (sh->check_state == check_state_run)
2803 sh->check_state = check_state_run_pq;
2804 else
2805 sh->check_state = check_state_run_q;
a4456856 2806 }
a4456856 2807
d82dfee0
DW		 2808 /* discard potentially stale zero_sum_result */
2809 sh->ops.zero_sum_result = 0;
a4456856 2810
d82dfee0
DW		 2811 if (sh->check_state == check_state_run) {
2812 /* async_xor_zero_sum destroys the contents of P */
2813 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
2814 s->uptodate--;
a4456856 2815 }
d82dfee0
DW		 2816 if (sh->check_state >= check_state_run &&
2817 sh->check_state <= check_state_run_pq) {
2818 /* async_syndrome_zero_sum preserves P and Q, so
2819 * no need to mark them !uptodate here
2820 */
2821 set_bit(STRIPE_OP_CHECK, &s->ops_request);
2822 break;
a4456856
DW		 2823 }
2824
d82dfee0
DW		 2825 /* we have 2-disk failure */
2826 BUG_ON(s->failed != 2);
2827 /* fall through */
2828 case check_state_compute_result:
2829 sh->check_state = check_state_idle;
a4456856 2830
d82dfee0
DW		 2831 /* check that a write has not made the stripe insync */
2832 if (test_bit(STRIPE_INSYNC, &sh->state))
2833 break;
a4456856
DW		 2834
2835 /* now write out any block on a failed drive,
d82dfee0 2836 * or P or Q if they were recomputed
a4456856 2837 */
d82dfee0 2838 BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */
a4456856
DW		 2839 if (s->failed == 2) {
2840 dev = &sh->dev[r6s->failed_num[1]];
2841 s->locked++;
2842 set_bit(R5_LOCKED, &dev->flags);
2843 set_bit(R5_Wantwrite, &dev->flags);
2844 }
2845 if (s->failed >= 1) {
2846 dev = &sh->dev[r6s->failed_num[0]];
2847 s->locked++;
2848 set_bit(R5_LOCKED, &dev->flags);
2849 set_bit(R5_Wantwrite, &dev->flags);
2850 }
d82dfee0 2851 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
a4456856
DW		 2852 dev = &sh->dev[pd_idx];
2853 s->locked++;
2854 set_bit(R5_LOCKED, &dev->flags);
2855 set_bit(R5_Wantwrite, &dev->flags);
2856 }
d82dfee0 2857 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
a4456856
DW		 2858 dev = &sh->dev[qd_idx];
2859 s->locked++;
2860 set_bit(R5_LOCKED, &dev->flags);
2861 set_bit(R5_Wantwrite, &dev->flags);
2862 }
2863 clear_bit(STRIPE_DEGRADED, &sh->state);
2864
2865 set_bit(STRIPE_INSYNC, &sh->state);
d82dfee0
DW		 2866 break;
2867 case check_state_run:
2868 case check_state_run_q:
2869 case check_state_run_pq:
2870 break; /* we will be called again upon completion */
2871 case check_state_check_result:
2872 sh->check_state = check_state_idle;
2873
2874 /* handle a successful check operation, if parity is correct
2875 * we are done. Otherwise update the mismatch count and repair
2876 * parity if !MD_RECOVERY_CHECK
2877 */
2878 if (sh->ops.zero_sum_result == 0) {
2879 /* both parities are correct */
2880 if (!s->failed)
2881 set_bit(STRIPE_INSYNC, &sh->state);
2882 else {
2883 /* in contrast to the raid5 case we can validate
2884 * parity, but still have a failure to write
2885 * back
2886 */
2887 sh->check_state = check_state_compute_result;
2888 /* Returning at this point means that we may go
2889 * off and bring p and/or q uptodate again so
2890 * we make sure to check zero_sum_result again
2891 * to verify if p or q need writeback
2892 */
2893 }
2894 } else {
2895 conf->mddev->resync_mismatches += STRIPE_SECTORS;
2896 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
2897 /* don't try to repair!! */
2898 set_bit(STRIPE_INSYNC, &sh->state);
2899 else {
2900 int *target = &sh->ops.target;
2901
2902 sh->ops.target = -1;
2903 sh->ops.target2 = -1;
2904 sh->check_state = check_state_compute_run;
2905 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2906 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2907 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
2908 set_bit(R5_Wantcompute,
2909 &sh->dev[pd_idx].flags);
2910 *target = pd_idx;
2911 target = &sh->ops.target2;
2912 s->uptodate++;
2913 }
2914 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
2915 set_bit(R5_Wantcompute,
2916 &sh->dev[qd_idx].flags);
2917 *target = qd_idx;
2918 s->uptodate++;
2919 }
2920 }
2921 }
2922 break;
2923 case check_state_compute_run:
2924 break;
2925 default:
2926 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
2927 __func__, sh->check_state,
2928 (unsigned long long) sh->sector);
2929 BUG();
a4456856
DW		 2930 }
2931}
2932
2933static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh,
2934 struct r6_state *r6s)
2935{
2936 int i;
2937
2938 /* We have read all the blocks in this stripe and now we need to
2939 * copy some of them into a target stripe for expand.
2940 */
f0a50d37 2941 struct dma_async_tx_descriptor *tx = NULL;
a4456856
DW		 2942 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
2943 for (i = 0; i < sh->disks; i++)
34e04e87 2944 if (i != sh->pd_idx && i != sh->qd_idx) {
911d4ee8 2945 int dd_idx, j;
a4456856 2946 struct stripe_head *sh2;
a08abd8c 2947 struct async_submit_ctl submit;
a4456856 2948
784052ec 2949 sector_t bn = compute_blocknr(sh, i, 1);
911d4ee8
N		 2950 sector_t s = raid5_compute_sector(conf, bn, 0,
2951 &dd_idx, NULL);
a8c906ca 2952 sh2 = get_active_stripe(conf, s, 0, 1, 1);
a4456856
DW		 2953 if (sh2 == NULL)
2954 /* so far only the early blocks of this stripe
2955 * have been requested. When later blocks
2956 * get requested, we will try again
2957 */
2958 continue;
2959 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
2960 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
2961 /* must have already done this block */
2962 release_stripe(sh2);
2963 continue;
2964 }
f0a50d37
DW		 2965
2966 /* place all the copies on one channel */
a08abd8c 2967 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
f0a50d37 2968 tx = async_memcpy(sh2->dev[dd_idx].page,
88ba2aa5 2969 sh->dev[i].page, 0, 0, STRIPE_SIZE,
a08abd8c 2970 &submit);
f0a50d37 2971
a4456856
DW		 2972 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
2973 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
2974 for (j = 0; j < conf->raid_disks; j++)
2975 if (j != sh2->pd_idx &&
d0dabf7e 2976 (!r6s || j != sh2->qd_idx) &&
a4456856
DW		 2977 !test_bit(R5_Expanded, &sh2->dev[j].flags))
2978 break;
2979 if (j == conf->raid_disks) {
2980 set_bit(STRIPE_EXPAND_READY, &sh2->state);
2981 set_bit(STRIPE_HANDLE, &sh2->state);
2982 }
2983 release_stripe(sh2);
f0a50d37 2984
a4456856 2985 }
a2e08551
N		 2986 /* done submitting copies, wait for them to complete */
2987 if (tx) {
2988 async_tx_ack(tx);
2989 dma_wait_for_async_tx(tx);
2990 }
a4456856 2991}
1da177e4 2992
6bfe0b49 2993
1da177e4
LT		 2994/*
2995 * handle_stripe - do things to a stripe.
2996 *
2997 * We lock the stripe and then examine the state of various bits
2998 * to see what needs to be done.
2999 * Possible results:
3000 * return some read request which now have data
3001 * return some write requests which are safely on disc
3002 * schedule a read on some buffers
3003 * schedule a write of some buffers
3004 * return confirmation of parity correctness
3005 *
1da177e4
LT		 3006 * buffers are taken off read_list or write_list, and bh_cache buffers
3007 * get BH_Lock set before the stripe lock is released.
3008 *
3009 */
a4456856 3010
1442577b 3011static void handle_stripe5(struct stripe_head *sh)
1da177e4
LT		 3012{
3013 raid5_conf_t *conf = sh->raid_conf;
a4456856
DW		 3014 int disks = sh->disks, i;
3015 struct bio *return_bi = NULL;
3016 struct stripe_head_state s;
1da177e4 3017 struct r5dev *dev;
6bfe0b49 3018 mdk_rdev_t *blocked_rdev = NULL;
e0a115e5 3019 int prexor;
729a1866 3020 int dec_preread_active = 0;
1da177e4 3021
a4456856 3022 memset(&s, 0, sizeof(s));
600aa109
DW		 3023 pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d check:%d "
3024 "reconstruct:%d\n", (unsigned long long)sh->sector, sh->state,
3025 atomic_read(&sh->count), sh->pd_idx, sh->check_state,
3026 sh->reconstruct_state);
1da177e4
LT		 3027
3028 spin_lock(&sh->lock);
3029 clear_bit(STRIPE_HANDLE, &sh->state);
3030 clear_bit(STRIPE_DELAYED, &sh->state);
3031
a4456856
DW		 3032 s.syncing = test_bit(STRIPE_SYNCING, &sh->state);
3033 s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3034 s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
def6ae26 3035
83de75cc 3036 /* Now to look around and see what can be done */
9910f16a 3037 rcu_read_lock();
1da177e4
LT		 3038 for (i=disks; i--; ) {
3039 mdk_rdev_t *rdev;
a9f326eb
N		 3040
3041 dev = &sh->dev[i];
1da177e4 3042
b5e98d65
DW		 3043 pr_debug("check %d: state 0x%lx toread %p read %p write %p "
3044 "written %p\n", i, dev->flags, dev->toread, dev->read,
3045 dev->towrite, dev->written);
3046
3047 /* maybe we can request a biofill operation
3048 *
3049 * new wantfill requests are only permitted while
83de75cc 3050 * ops_complete_biofill is guaranteed to be inactive
b5e98d65
DW		 3051 */
3052 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
83de75cc 3053 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
b5e98d65 3054 set_bit(R5_Wantfill, &dev->flags);
1da177e4
LT		 3055
3056 /* now count some things */
a4456856
DW		 3057 if (test_bit(R5_LOCKED, &dev->flags)) s.locked++;
3058 if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++;
f38e1219 3059 if (test_bit(R5_Wantcompute, &dev->flags)) s.compute++;
1da177e4 3060
b5e98d65
DW		 3061 if (test_bit(R5_Wantfill, &dev->flags))
3062 s.to_fill++;
3063 else if (dev->toread)
a4456856 3064 s.to_read++;
1da177e4 3065 if (dev->towrite) {
a4456856 3066 s.to_write++;
1da177e4 3067 if (!test_bit(R5_OVERWRITE, &dev->flags))
a4456856 3068 s.non_overwrite++;
1da177e4 3069 }
a4456856
DW		 3070 if (dev->written)
3071 s.written++;
9910f16a 3072 rdev = rcu_dereference(conf->disks[i].rdev);
ac4090d2
N		 3073 if (blocked_rdev == NULL &&
3074 rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
6bfe0b49
DW		 3075 blocked_rdev = rdev;
3076 atomic_inc(&rdev->nr_pending);
6bfe0b49 3077 }
415e72d0
N		 3078 clear_bit(R5_Insync, &dev->flags);
3079 if (!rdev)
3080 /* Not in-sync */;
3081 else if (test_bit(In_sync, &rdev->flags))
3082 set_bit(R5_Insync, &dev->flags);
3083 else {
3084 /* could be in-sync depending on recovery/reshape status */
3085 if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
3086 set_bit(R5_Insync, &dev->flags);
3087 }
3088 if (!test_bit(R5_Insync, &dev->flags)) {
14f8d26b 3089 /* The ReadError flag will just be confusing now */
4e5314b5
N		 3090 clear_bit(R5_ReadError, &dev->flags);
3091 clear_bit(R5_ReWrite, &dev->flags);
3092 }
415e72d0
N		 3093 if (test_bit(R5_ReadError, &dev->flags))
3094 clear_bit(R5_Insync, &dev->flags);
3095 if (!test_bit(R5_Insync, &dev->flags)) {
a4456856
DW		 3096 s.failed++;
3097 s.failed_num = i;
415e72d0 3098 }
1da177e4 3099 }
9910f16a 3100 rcu_read_unlock();
b5e98d65 3101
6bfe0b49 3102 if (unlikely(blocked_rdev)) {
ac4090d2
N		 3103 if (s.syncing || s.expanding || s.expanded ||
3104 s.to_write || s.written) {
3105 set_bit(STRIPE_HANDLE, &sh->state);
3106 goto unlock;
3107 }
3108 /* There is nothing for the blocked_rdev to block */
3109 rdev_dec_pending(blocked_rdev, conf->mddev);
3110 blocked_rdev = NULL;
6bfe0b49
DW		 3111 }
3112
83de75cc
DW		 3113 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
3114 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
3115 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
3116 }
b5e98d65 3117
45b4233c 3118 pr_debug("locked=%d uptodate=%d to_read=%d"
1da177e4 3119 " to_write=%d failed=%d failed_num=%d\n",
a4456856
DW		 3120 s.locked, s.uptodate, s.to_read, s.to_write,
3121 s.failed, s.failed_num);
1da177e4
LT		 3122 /* check if the array has lost two devices and, if so, some requests might
3123 * need to be failed
3124 */
a4456856 3125 if (s.failed > 1 && s.to_read+s.to_write+s.written)
1fe797e6 3126 handle_failed_stripe(conf, sh, &s, disks, &return_bi);
a4456856 3127 if (s.failed > 1 && s.syncing) {
1da177e4
LT		 3128 md_done_sync(conf->mddev, STRIPE_SECTORS,0);
3129 clear_bit(STRIPE_SYNCING, &sh->state);
a4456856 3130 s.syncing = 0;
1da177e4
LT		 3131 }
3132
3133 /* might be able to return some write requests if the parity block
3134 * is safe, or on a failed drive
3135 */
3136 dev = &sh->dev[sh->pd_idx];
a4456856
DW		 3137 if ( s.written &&
3138 ((test_bit(R5_Insync, &dev->flags) &&
3139 !test_bit(R5_LOCKED, &dev->flags) &&
3140 test_bit(R5_UPTODATE, &dev->flags)) ||
3141 (s.failed == 1 && s.failed_num == sh->pd_idx)))
1fe797e6 3142 handle_stripe_clean_event(conf, sh, disks, &return_bi);
1da177e4
LT		 3143
3144 /* Now we might consider reading some blocks, either to check/generate
3145 * parity, or to satisfy requests
3146 * or to load a block that is being partially written.
3147 */
a4456856 3148 if (s.to_read || s.non_overwrite ||
976ea8d4 3149 (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding)
1fe797e6 3150 handle_stripe_fill5(sh, &s, disks);
1da177e4 3151
e33129d8
DW		 3152 /* Now we check to see if any write operations have recently
3153 * completed
3154 */
e0a115e5 3155 prexor = 0;
d8ee0728 3156 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
e0a115e5 3157 prexor = 1;
d8ee0728
DW		 3158 if (sh->reconstruct_state == reconstruct_state_drain_result ||
3159 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
600aa109 3160 sh->reconstruct_state = reconstruct_state_idle;
e33129d8
DW		 3161
3162 /* All the 'written' buffers and the parity block are ready to
3163 * be written back to disk
3164 */
3165 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags));
3166 for (i = disks; i--; ) {
3167 dev = &sh->dev[i];
3168 if (test_bit(R5_LOCKED, &dev->flags) &&
3169 (i == sh->pd_idx || dev->written)) {
3170 pr_debug("Writing block %d\n", i);
3171 set_bit(R5_Wantwrite, &dev->flags);
e0a115e5
DW		 3172 if (prexor)
3173 continue;
e33129d8
DW		 3174 if (!test_bit(R5_Insync, &dev->flags) ||
3175 (i == sh->pd_idx && s.failed == 0))
3176 set_bit(STRIPE_INSYNC, &sh->state);
3177 }
3178 }
729a1866
N		 3179 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3180 dec_preread_active = 1;
e33129d8
DW		 3181 }
3182
3183 /* Now to consider new write requests and what else, if anything
3184 * should be read. We do not handle new writes when:
3185 * 1/ A 'write' operation (copy+xor) is already in flight.
3186 * 2/ A 'check' operation is in flight, as it may clobber the parity
3187 * block.
3188 */
600aa109 3189 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
1fe797e6 3190 handle_stripe_dirtying5(conf, sh, &s, disks);
1da177e4
LT		 3191
3192 /* maybe we need to check and possibly fix the parity for this stripe
e89f8962
DW		 3193 * Any reads will already have been scheduled, so we just see if enough
3194 * data is available. The parity check is held off while parity
3195 * dependent operations are in flight.
1da177e4 3196 */
ecc65c9b
DW		 3197 if (sh->check_state ||
3198 (s.syncing && s.locked == 0 &&
976ea8d4 3199 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
ecc65c9b 3200 !test_bit(STRIPE_INSYNC, &sh->state)))
a4456856 3201 handle_parity_checks5(conf, sh, &s, disks);
e89f8962 3202
a4456856 3203 if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) {
1da177e4
LT		 3204 md_done_sync(conf->mddev, STRIPE_SECTORS,1);
3205 clear_bit(STRIPE_SYNCING, &sh->state);
3206 }
4e5314b5
N		 3207
3208 /* If the failed drive is just a ReadError, then we might need to progress
3209 * the repair/check process
3210 */
a4456856
DW		 3211 if (s.failed == 1 && !conf->mddev->ro &&
3212 test_bit(R5_ReadError, &sh->dev[s.failed_num].flags)
3213 && !test_bit(R5_LOCKED, &sh->dev[s.failed_num].flags)
3214 && test_bit(R5_UPTODATE, &sh->dev[s.failed_num].flags)
4e5314b5 3215 ) {
a4456856 3216 dev = &sh->dev[s.failed_num];
4e5314b5
N		 3217 if (!test_bit(R5_ReWrite, &dev->flags)) {
3218 set_bit(R5_Wantwrite, &dev->flags);
3219 set_bit(R5_ReWrite, &dev->flags);
3220 set_bit(R5_LOCKED, &dev->flags);
a4456856 3221 s.locked++;
4e5314b5
N		 3222 } else {
3223 /* let's read it back */
3224 set_bit(R5_Wantread, &dev->flags);
3225 set_bit(R5_LOCKED, &dev->flags);
a4456856 3226 s.locked++;
4e5314b5
N		 3227 }
3228 }
3229
600aa109
DW		 3230 /* Finish reconstruct operations initiated by the expansion process */
3231 if (sh->reconstruct_state == reconstruct_state_result) {
ab69ae12 3232 struct stripe_head *sh2
a8c906ca 3233 = get_active_stripe(conf, sh->sector, 1, 1, 1);
ab69ae12
N		 3234 if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) {
3235 /* sh cannot be written until sh2 has been read.
3236 * so arrange for sh to be delayed a little
3237 */
3238 set_bit(STRIPE_DELAYED, &sh->state);
3239 set_bit(STRIPE_HANDLE, &sh->state);
3240 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
3241 &sh2->state))
3242 atomic_inc(&conf->preread_active_stripes);
3243 release_stripe(sh2);
3244 goto unlock;
3245 }
3246 if (sh2)
3247 release_stripe(sh2);
3248
600aa109 3249 sh->reconstruct_state = reconstruct_state_idle;
f0a50d37 3250 clear_bit(STRIPE_EXPANDING, &sh->state);
23397883 3251 for (i = conf->raid_disks; i--; ) {
ccfcc3c1 3252 set_bit(R5_Wantwrite, &sh->dev[i].flags);
23397883 3253 set_bit(R5_LOCKED, &sh->dev[i].flags);
efe31143 3254 s.locked++;
23397883 3255 }
f0a50d37
DW		 3256 }
3257
3258 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
600aa109 3259 !sh->reconstruct_state) {
f0a50d37
DW		 3260 /* Need to write out all blocks after computing parity */
3261 sh->disks = conf->raid_disks;
911d4ee8 3262 stripe_set_idx(sh->sector, conf, 0, sh);
c0f7bddb 3263 schedule_reconstruction(sh, &s, 1, 1);
600aa109 3264 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
ccfcc3c1 3265 clear_bit(STRIPE_EXPAND_READY, &sh->state);
f6705578 3266 atomic_dec(&conf->reshape_stripes);
ccfcc3c1
N		 3267 wake_up(&conf->wait_for_overlap);
3268 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3269 }
3270
0f94e87c 3271 if (s.expanding && s.locked == 0 &&
976ea8d4 3272 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
a4456856 3273 handle_stripe_expansion(conf, sh, NULL);
ccfcc3c1 3274
6bfe0b49 3275 unlock:
1da177e4
LT		 3276 spin_unlock(&sh->lock);
3277
6bfe0b49
DW		 3278 /* wait for this device to become unblocked */
3279 if (unlikely(blocked_rdev))
3280 md_wait_for_blocked_rdev(blocked_rdev, conf->mddev);
3281
600aa109 3282 if (s.ops_request)
ac6b53b6 3283 raid_run_ops(sh, s.ops_request);
d84e0f10 3284
c4e5ac0a 3285 ops_run_io(sh, &s);
1da177e4 3286
729a1866
N		 3287 if (dec_preread_active) {
3288 /* We delay this until after ops_run_io so that if make_request
e9c7469b 3289 * is waiting on a flush, it won't continue until the writes
729a1866
N		 3290 * have actually been submitted.
3291 */
3292 atomic_dec(&conf->preread_active_stripes);
3293 if (atomic_read(&conf->preread_active_stripes) <
3294 IO_THRESHOLD)
3295 md_wakeup_thread(conf->mddev->thread);
3296 }
a4456856 3297 return_io(return_bi);
1da177e4
LT		 3298}
3299
1442577b 3300static void handle_stripe6(struct stripe_head *sh)
1da177e4 3301{
bff61975 3302 raid5_conf_t *conf = sh->raid_conf;
f416885e 3303 int disks = sh->disks;
a4456856 3304 struct bio *return_bi = NULL;
34e04e87 3305 int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx;
a4456856
DW		 3306 struct stripe_head_state s;
3307 struct r6_state r6s;
16a53ecc 3308 struct r5dev *dev, *pdev, *qdev;
6bfe0b49 3309 mdk_rdev_t *blocked_rdev = NULL;
729a1866 3310 int dec_preread_active = 0;
1da177e4 3311
45b4233c 3312 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
6c0069c0 3313 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
a4456856 3314 (unsigned long long)sh->sector, sh->state,
6c0069c0
YT		 3315 atomic_read(&sh->count), pd_idx, qd_idx,
3316 sh->check_state, sh->reconstruct_state);
a4456856 3317 memset(&s, 0, sizeof(s));
72626685 3318
16a53ecc
N		 3319 spin_lock(&sh->lock);
3320 clear_bit(STRIPE_HANDLE, &sh->state);
3321 clear_bit(STRIPE_DELAYED, &sh->state);
3322
a4456856
DW		 3323 s.syncing = test_bit(STRIPE_SYNCING, &sh->state);
3324 s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3325 s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
16a53ecc 3326 /* Now to look around and see what can be done */
1da177e4
LT		 3327
3328 rcu_read_lock();
16a53ecc
N		 3329 for (i=disks; i--; ) {
3330 mdk_rdev_t *rdev;
3331 dev = &sh->dev[i];
1da177e4 3332
45b4233c 3333 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
16a53ecc 3334 i, dev->flags, dev->toread, dev->towrite, dev->written);
6c0069c0
YT		 3335 /* maybe we can reply to a read
3336 *
3337 * new wantfill requests are only permitted while
3338 * ops_complete_biofill is guaranteed to be inactive
3339 */
3340 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
3341 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
3342 set_bit(R5_Wantfill, &dev->flags);
1da177e4 3343
16a53ecc 3344 /* now count some things */
a4456856
DW		 3345 if (test_bit(R5_LOCKED, &dev->flags)) s.locked++;
3346 if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++;
2d6e4ecc
DW		 3347 if (test_bit(R5_Wantcompute, &dev->flags)) {
3348 s.compute++;
3349 BUG_ON(s.compute > 2);
3350 }
1da177e4 3351
6c0069c0
YT		 3352 if (test_bit(R5_Wantfill, &dev->flags)) {
3353 s.to_fill++;
3354 } else if (dev->toread)
a4456856 3355 s.to_read++;
16a53ecc 3356 if (dev->towrite) {
a4456856 3357 s.to_write++;
16a53ecc 3358 if (!test_bit(R5_OVERWRITE, &dev->flags))
a4456856 3359 s.non_overwrite++;
16a53ecc 3360 }
a4456856
DW		 3361 if (dev->written)
3362 s.written++;
16a53ecc 3363 rdev = rcu_dereference(conf->disks[i].rdev);
ac4090d2
N		 3364 if (blocked_rdev == NULL &&
3365 rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
6bfe0b49
DW		 3366 blocked_rdev = rdev;
3367 atomic_inc(&rdev->nr_pending);
6bfe0b49 3368 }
415e72d0
N		 3369 clear_bit(R5_Insync, &dev->flags);
3370 if (!rdev)
3371 /* Not in-sync */;
3372 else if (test_bit(In_sync, &rdev->flags))
3373 set_bit(R5_Insync, &dev->flags);
3374 else {
3375 /* in sync if before recovery_offset */
3376 if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
3377 set_bit(R5_Insync, &dev->flags);
3378 }
3379 if (!test_bit(R5_Insync, &dev->flags)) {
16a53ecc
N		 3380 /* The ReadError flag will just be confusing now */
3381 clear_bit(R5_ReadError, &dev->flags);
3382 clear_bit(R5_ReWrite, &dev->flags);
1da177e4 3383 }
415e72d0
N		 3384 if (test_bit(R5_ReadError, &dev->flags))
3385 clear_bit(R5_Insync, &dev->flags);
3386 if (!test_bit(R5_Insync, &dev->flags)) {
a4456856
DW		 3387 if (s.failed < 2)
3388 r6s.failed_num[s.failed] = i;
3389 s.failed++;
415e72d0 3390 }
1da177e4
LT		 3391 }
3392 rcu_read_unlock();
6bfe0b49
DW		 3393
3394 if (unlikely(blocked_rdev)) {
ac4090d2
N		 3395 if (s.syncing || s.expanding || s.expanded ||
3396 s.to_write || s.written) {
3397 set_bit(STRIPE_HANDLE, &sh->state);
3398 goto unlock;
3399 }
3400 /* There is nothing for the blocked_rdev to block */
3401 rdev_dec_pending(blocked_rdev, conf->mddev);
3402 blocked_rdev = NULL;
6bfe0b49 3403 }
ac4090d2 3404
6c0069c0
YT		 3405 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
3406 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
3407 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
3408 }
3409
45b4233c 3410 pr_debug("locked=%d uptodate=%d to_read=%d"
16a53ecc 3411 " to_write=%d failed=%d failed_num=%d,%d\n",
a4456856
DW		 3412 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
3413 r6s.failed_num[0], r6s.failed_num[1]);
3414 /* check if the array has lost >2 devices and, if so, some requests
3415 * might need to be failed
16a53ecc 3416 */
a4456856 3417 if (s.failed > 2 && s.to_read+s.to_write+s.written)
1fe797e6 3418 handle_failed_stripe(conf, sh, &s, disks, &return_bi);
a4456856 3419 if (s.failed > 2 && s.syncing) {
16a53ecc
N		 3420 md_done_sync(conf->mddev, STRIPE_SECTORS,0);
3421 clear_bit(STRIPE_SYNCING, &sh->state);
a4456856 3422 s.syncing = 0;
16a53ecc
N		 3423 }
3424
3425 /*
3426 * might be able to return some write requests if the parity blocks
3427 * are safe, or on a failed drive
3428 */
3429 pdev = &sh->dev[pd_idx];
a4456856
DW		 3430 r6s.p_failed = (s.failed >= 1 && r6s.failed_num[0] == pd_idx)
3431 || (s.failed >= 2 && r6s.failed_num[1] == pd_idx);
34e04e87
N		 3432 qdev = &sh->dev[qd_idx];
3433 r6s.q_failed = (s.failed >= 1 && r6s.failed_num[0] == qd_idx)
3434 || (s.failed >= 2 && r6s.failed_num[1] == qd_idx);
a4456856
DW		 3435
3436 if ( s.written &&
3437 ( r6s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
16a53ecc 3438 && !test_bit(R5_LOCKED, &pdev->flags)
a4456856
DW		 3439 && test_bit(R5_UPTODATE, &pdev->flags)))) &&
3440 ( r6s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
16a53ecc 3441 && !test_bit(R5_LOCKED, &qdev->flags)
a4456856 3442 && test_bit(R5_UPTODATE, &qdev->flags)))))
1fe797e6 3443 handle_stripe_clean_event(conf, sh, disks, &return_bi);
16a53ecc
N		 3444
3445 /* Now we might consider reading some blocks, either to check/generate
3446 * parity, or to satisfy requests
3447 * or to load a block that is being partially written.
3448 */
a4456856 3449 if (s.to_read || s.non_overwrite || (s.to_write && s.failed) ||
6c0069c0 3450 (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding)
1fe797e6 3451 handle_stripe_fill6(sh, &s, &r6s, disks);
16a53ecc 3452
6c0069c0
YT		 3453 /* Now we check to see if any write operations have recently
3454 * completed
3455 */
3456 if (sh->reconstruct_state == reconstruct_state_drain_result) {
6c0069c0
YT		 3457
3458 sh->reconstruct_state = reconstruct_state_idle;
3459 /* All the 'written' buffers and the parity blocks are ready to
3460 * be written back to disk
3461 */
3462 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags));
3463 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags));
3464 for (i = disks; i--; ) {
3465 dev = &sh->dev[i];
3466 if (test_bit(R5_LOCKED, &dev->flags) &&
3467 (i == sh->pd_idx || i == qd_idx ||
3468 dev->written)) {
3469 pr_debug("Writing block %d\n", i);
3470 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
3471 set_bit(R5_Wantwrite, &dev->flags);
3472 if (!test_bit(R5_Insync, &dev->flags) ||
3473 ((i == sh->pd_idx || i == qd_idx) &&
3474 s.failed == 0))
3475 set_bit(STRIPE_INSYNC, &sh->state);
3476 }
3477 }
729a1866
N		 3478 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3479 dec_preread_active = 1;
6c0069c0
YT		 3480 }
3481
a9b39a74
YT		 3482 /* Now to consider new write requests and what else, if anything
3483 * should be read. We do not handle new writes when:
3484 * 1/ A 'write' operation (copy+gen_syndrome) is already in flight.
3485 * 2/ A 'check' operation is in flight, as it may clobber the parity
3486 * block.
3487 */
3488 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
1fe797e6 3489 handle_stripe_dirtying6(conf, sh, &s, &r6s, disks);
16a53ecc
N		 3490
3491 /* maybe we need to check and possibly fix the parity for this stripe
a4456856 3492 * Any reads will already have been scheduled, so we just see if enough
6c0069c0
YT		 3493 * data is available. The parity check is held off while parity
3494 * dependent operations are in flight.
16a53ecc 3495 */
6c0069c0
YT		 3496 if (sh->check_state ||
3497 (s.syncing && s.locked == 0 &&
3498 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
3499 !test_bit(STRIPE_INSYNC, &sh->state)))
36d1c647 3500 handle_parity_checks6(conf, sh, &s, &r6s, disks);
16a53ecc 3501
a4456856 3502 if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) {
16a53ecc
N		 3503 md_done_sync(conf->mddev, STRIPE_SECTORS,1);
3504 clear_bit(STRIPE_SYNCING, &sh->state);
3505 }
3506
3507 /* If the failed drives are just a ReadError, then we might need
3508 * to progress the repair/check process
3509 */
a4456856
DW		 3510 if (s.failed <= 2 && !conf->mddev->ro)
3511 for (i = 0; i < s.failed; i++) {
3512 dev = &sh->dev[r6s.failed_num[i]];
16a53ecc
N		 3513 if (test_bit(R5_ReadError, &dev->flags)
3514 && !test_bit(R5_LOCKED, &dev->flags)
3515 && test_bit(R5_UPTODATE, &dev->flags)
3516 ) {
3517 if (!test_bit(R5_ReWrite, &dev->flags)) {
3518 set_bit(R5_Wantwrite, &dev->flags);
3519 set_bit(R5_ReWrite, &dev->flags);
3520 set_bit(R5_LOCKED, &dev->flags);
6c0069c0 3521 s.locked++;
16a53ecc
N		 3522 } else {
3523 /* let's read it back */
3524 set_bit(R5_Wantread, &dev->flags);
3525 set_bit(R5_LOCKED, &dev->flags);
6c0069c0 3526 s.locked++;
16a53ecc
N		 3527 }
3528 }
3529 }
f416885e 3530
6c0069c0
YT		 3531 /* Finish reconstruct operations initiated by the expansion process */
3532 if (sh->reconstruct_state == reconstruct_state_result) {
3533 sh->reconstruct_state = reconstruct_state_idle;
3534 clear_bit(STRIPE_EXPANDING, &sh->state);
3535 for (i = conf->raid_disks; i--; ) {
3536 set_bit(R5_Wantwrite, &sh->dev[i].flags);
3537 set_bit(R5_LOCKED, &sh->dev[i].flags);
3538 s.locked++;
3539 }
3540 }
3541
3542 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
3543 !sh->reconstruct_state) {
ab69ae12 3544 struct stripe_head *sh2
a8c906ca 3545 = get_active_stripe(conf, sh->sector, 1, 1, 1);
ab69ae12
N		 3546 if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) {
3547 /* sh cannot be written until sh2 has been read.
3548 * so arrange for sh to be delayed a little
3549 */
3550 set_bit(STRIPE_DELAYED, &sh->state);
3551 set_bit(STRIPE_HANDLE, &sh->state);
3552 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
3553 &sh2->state))
3554 atomic_inc(&conf->preread_active_stripes);
3555 release_stripe(sh2);
3556 goto unlock;
3557 }
3558 if (sh2)
3559 release_stripe(sh2);
3560
f416885e
N		 3561 /* Need to write out all blocks after computing P&Q */
3562 sh->disks = conf->raid_disks;
911d4ee8 3563 stripe_set_idx(sh->sector, conf, 0, sh);
6c0069c0
YT		 3564 schedule_reconstruction(sh, &s, 1, 1);
3565 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
f416885e
N		 3566 clear_bit(STRIPE_EXPAND_READY, &sh->state);
3567 atomic_dec(&conf->reshape_stripes);
3568 wake_up(&conf->wait_for_overlap);
3569 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3570 }
3571
0f94e87c 3572 if (s.expanding && s.locked == 0 &&
976ea8d4 3573 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
a4456856 3574 handle_stripe_expansion(conf, sh, &r6s);
f416885e 3575
6bfe0b49 3576 unlock:
16a53ecc
N		 3577 spin_unlock(&sh->lock);
3578
6bfe0b49
DW		 3579 /* wait for this device to become unblocked */
3580 if (unlikely(blocked_rdev))
3581 md_wait_for_blocked_rdev(blocked_rdev, conf->mddev);
3582
6c0069c0
YT		 3583 if (s.ops_request)
3584 raid_run_ops(sh, s.ops_request);
3585
f0e43bcd 3586 ops_run_io(sh, &s);
16a53ecc 3587
729a1866
N		 3588
3589 if (dec_preread_active) {
3590 /* We delay this until after ops_run_io so that if make_request
e9c7469b 3591 * is waiting on a flush, it won't continue until the writes
729a1866
N		 3592 * have actually been submitted.
3593 */
3594 atomic_dec(&conf->preread_active_stripes);
3595 if (atomic_read(&conf->preread_active_stripes) <
3596 IO_THRESHOLD)
3597 md_wakeup_thread(conf->mddev->thread);
3598 }
3599
f0e43bcd 3600 return_io(return_bi);
16a53ecc
N		 3601}
3602
1442577b 3603static void handle_stripe(struct stripe_head *sh)
16a53ecc
N		 3604{
3605 if (sh->raid_conf->level == 6)
1442577b 3606 handle_stripe6(sh);
16a53ecc 3607 else
1442577b 3608 handle_stripe5(sh);
16a53ecc
N		 3609}
3610
16a53ecc
N		 3611static void raid5_activate_delayed(raid5_conf_t *conf)
3612{
3613 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
3614 while (!list_empty(&conf->delayed_list)) {
3615 struct list_head *l = conf->delayed_list.next;
3616 struct stripe_head *sh;
3617 sh = list_entry(l, struct stripe_head, lru);
3618 list_del_init(l);
3619 clear_bit(STRIPE_DELAYED, &sh->state);
3620 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3621 atomic_inc(&conf->preread_active_stripes);
8b3e6cdc 3622 list_add_tail(&sh->lru, &conf->hold_list);
16a53ecc 3623 }
482c0834 3624 }
16a53ecc
N		 3625}
3626
3627static void activate_bit_delay(raid5_conf_t *conf)
3628{
3629 /* device_lock is held */
3630 struct list_head head;
3631 list_add(&head, &conf->bitmap_list);
3632 list_del_init(&conf->bitmap_list);
3633 while (!list_empty(&head)) {
3634 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
3635 list_del_init(&sh->lru);
3636 atomic_inc(&sh->count);
3637 __release_stripe(conf, sh);
3638 }
3639}
3640
11d8a6e3 3641int md_raid5_congested(mddev_t *mddev, int bits)
f022b2fd 3642{
070ec55d 3643 raid5_conf_t *conf = mddev->private;
f022b2fd
N		 3644
3645 /* No difference between reads and writes. Just check
3646 * how busy the stripe_cache is
3647 */
3fa841d7 3648
f022b2fd
N		 3649 if (conf->inactive_blocked)
3650 return 1;
3651 if (conf->quiesce)
3652 return 1;
3653 if (list_empty_careful(&conf->inactive_list))
3654 return 1;
3655
3656 return 0;
3657}
11d8a6e3
N		 3658EXPORT_SYMBOL_GPL(md_raid5_congested);
3659
3660static int raid5_congested(void *data, int bits)
3661{
3662 mddev_t *mddev = data;
3663
3664 return mddev_congested(mddev, bits) ||
3665 md_raid5_congested(mddev, bits);
3666}
f022b2fd 3667
23032a0e
RBJ		 3668/* We want read requests to align with chunks where possible,
3669 * but write requests don't need to.
3670 */
cc371e66
AK		 3671static int raid5_mergeable_bvec(struct request_queue *q,
3672 struct bvec_merge_data *bvm,
3673 struct bio_vec *biovec)
23032a0e
RBJ		 3674{
3675 mddev_t *mddev = q->queuedata;
cc371e66 3676 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
23032a0e 3677 int max;
9d8f0363 3678 unsigned int chunk_sectors = mddev->chunk_sectors;
cc371e66 3679 unsigned int bio_sectors = bvm->bi_size >> 9;
23032a0e 3680
cc371e66 3681 if ((bvm->bi_rw & 1) == WRITE)
23032a0e
RBJ		 3682 return biovec->bv_len; /* always allow writes to be mergeable */
3683
664e7c41
AN		 3684 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
3685 chunk_sectors = mddev->new_chunk_sectors;
23032a0e
RBJ		 3686 max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
3687 if (max < 0) max = 0;
3688 if (max <= biovec->bv_len && bio_sectors == 0)
3689 return biovec->bv_len;
3690 else
3691 return max;
3692}
3693
f679623f
RBJ		 3694
3695static int in_chunk_boundary(mddev_t *mddev, struct bio *bio)
3696{
3697 sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
9d8f0363 3698 unsigned int chunk_sectors = mddev->chunk_sectors;
f679623f
RBJ		 3699 unsigned int bio_sectors = bio->bi_size >> 9;
3700
664e7c41
AN		 3701 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
3702 chunk_sectors = mddev->new_chunk_sectors;
f679623f
RBJ		 3703 return chunk_sectors >=
3704 ((sector & (chunk_sectors - 1)) + bio_sectors);
3705}
3706
46031f9a
RBJ		 3707/*
3708 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
3709 * later sampled by raid5d.
3710 */
3711static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf)
3712{
3713 unsigned long flags;
3714
3715 spin_lock_irqsave(&conf->device_lock, flags);
3716
3717 bi->bi_next = conf->retry_read_aligned_list;
3718 conf->retry_read_aligned_list = bi;
3719
3720 spin_unlock_irqrestore(&conf->device_lock, flags);
3721 md_wakeup_thread(conf->mddev->thread);
3722}
3723
3724
3725static struct bio *remove_bio_from_retry(raid5_conf_t *conf)
3726{
3727 struct bio *bi;
3728
3729 bi = conf->retry_read_aligned;
3730 if (bi) {
3731 conf->retry_read_aligned = NULL;
3732 return bi;
3733 }
3734 bi = conf->retry_read_aligned_list;
3735 if(bi) {
387bb173 3736 conf->retry_read_aligned_list = bi->bi_next;
46031f9a 3737 bi->bi_next = NULL;
960e739d
JA		 3738 /*
3739 * this sets the active strip count to 1 and the processed
3740 * strip count to zero (upper 8 bits)
3741 */
46031f9a 3742 bi->bi_phys_segments = 1; /* biased count of active stripes */
46031f9a
RBJ		 3743 }
3744
3745 return bi;
3746}
3747
3748
f679623f
RBJ		 3749/*
3750 * The "raid5_align_endio" should check if the read succeeded and if it
3751 * did, call bio_endio on the original bio (having bio_put the new bio
3752 * first).
3753 * If the read failed..
3754 */
6712ecf8 3755static void raid5_align_endio(struct bio *bi, int error)
f679623f
RBJ		 3756{
3757 struct bio* raid_bi = bi->bi_private;
46031f9a
RBJ		 3758 mddev_t *mddev;
3759 raid5_conf_t *conf;
3760 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
3761 mdk_rdev_t *rdev;
3762
f679623f 3763 bio_put(bi);
46031f9a 3764
46031f9a
RBJ		 3765 rdev = (void*)raid_bi->bi_next;
3766 raid_bi->bi_next = NULL;
2b7f2228
N		 3767 mddev = rdev->mddev;
3768 conf = mddev->private;
46031f9a
RBJ		 3769
3770 rdev_dec_pending(rdev, conf->mddev);
3771
3772 if (!error && uptodate) {
6712ecf8 3773 bio_endio(raid_bi, 0);
46031f9a
RBJ		 3774 if (atomic_dec_and_test(&conf->active_aligned_reads))
3775 wake_up(&conf->wait_for_stripe);
6712ecf8 3776 return;
46031f9a
RBJ		 3777 }
3778
3779
45b4233c 3780 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
46031f9a
RBJ		 3781
3782 add_bio_to_retry(raid_bi, conf);
f679623f
RBJ		 3783}
3784
387bb173
NB		 3785static int bio_fits_rdev(struct bio *bi)
3786{
165125e1 3787 struct request_queue *q = bdev_get_queue(bi->bi_bdev);
387bb173 3788
ae03bf63 3789 if ((bi->bi_size>>9) > queue_max_sectors(q))
387bb173
NB		 3790 return 0;
3791 blk_recount_segments(q, bi);
8a78362c 3792 if (bi->bi_phys_segments > queue_max_segments(q))
387bb173
NB		 3793 return 0;
3794
3795 if (q->merge_bvec_fn)
3796 /* it's too hard to apply the merge_bvec_fn at this stage,
3797 * just just give up
3798 */
3799 return 0;
3800
3801 return 1;
3802}
3803
3804
21a52c6d 3805static int chunk_aligned_read(mddev_t *mddev, struct bio * raid_bio)
f679623f 3806{
070ec55d 3807 raid5_conf_t *conf = mddev->private;
8553fe7e 3808 int dd_idx;
f679623f
RBJ		 3809 struct bio* align_bi;
3810 mdk_rdev_t *rdev;
3811
3812 if (!in_chunk_boundary(mddev, raid_bio)) {
45b4233c 3813 pr_debug("chunk_aligned_read : non aligned\n");
f679623f
RBJ		 3814 return 0;
3815 }
3816 /*
a167f663 3817 * use bio_clone_mddev to make a copy of the bio
f679623f 3818 */
a167f663 3819 align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev);
f679623f
RBJ		 3820 if (!align_bi)
3821 return 0;
3822 /*
3823 * set bi_end_io to a new function, and set bi_private to the
3824 * original bio.
3825 */
3826 align_bi->bi_end_io = raid5_align_endio;
3827 align_bi->bi_private = raid_bio;
3828 /*
3829 * compute position
3830 */
112bf897
N		 3831 align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector,
3832 0,
911d4ee8 3833 &dd_idx, NULL);
f679623f
RBJ		 3834
3835 rcu_read_lock();
3836 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
3837 if (rdev && test_bit(In_sync, &rdev->flags)) {
f679623f
RBJ		 3838 atomic_inc(&rdev->nr_pending);
3839 rcu_read_unlock();
46031f9a
RBJ		 3840 raid_bio->bi_next = (void*)rdev;
3841 align_bi->bi_bdev = rdev->bdev;
3842 align_bi->bi_flags &= ~(1 << BIO_SEG_VALID);
3843 align_bi->bi_sector += rdev->data_offset;
3844
387bb173
NB		 3845 if (!bio_fits_rdev(align_bi)) {
3846 /* too big in some way */
3847 bio_put(align_bi);
3848 rdev_dec_pending(rdev, mddev);
3849 return 0;
3850 }
3851
46031f9a
RBJ		 3852 spin_lock_irq(&conf->device_lock);
3853 wait_event_lock_irq(conf->wait_for_stripe,
3854 conf->quiesce == 0,
3855 conf->device_lock, /* nothing */);
3856 atomic_inc(&conf->active_aligned_reads);
3857 spin_unlock_irq(&conf->device_lock);
3858
f679623f
RBJ		 3859 generic_make_request(align_bi);
3860 return 1;
3861 } else {
3862 rcu_read_unlock();
46031f9a 3863 bio_put(align_bi);
f679623f
RBJ		 3864 return 0;
3865 }
3866}
3867
8b3e6cdc
DW		 3868/* __get_priority_stripe - get the next stripe to process
3869 *
3870 * Full stripe writes are allowed to pass preread active stripes up until
3871 * the bypass_threshold is exceeded. In general the bypass_count
3872 * increments when the handle_list is handled before the hold_list; however, it
3873 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
3874 * stripe with in flight i/o. The bypass_count will be reset when the
3875 * head of the hold_list has changed, i.e. the head was promoted to the
3876 * handle_list.
3877 */
3878static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf)
3879{
3880 struct stripe_head *sh;
3881
3882 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
3883 __func__,
3884 list_empty(&conf->handle_list) ? "empty" : "busy",
3885 list_empty(&conf->hold_list) ? "empty" : "busy",
3886 atomic_read(&conf->pending_full_writes), conf->bypass_count);
3887
3888 if (!list_empty(&conf->handle_list)) {
3889 sh = list_entry(conf->handle_list.next, typeof(*sh), lru);
3890
3891 if (list_empty(&conf->hold_list))
3892 conf->bypass_count = 0;
3893 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
3894 if (conf->hold_list.next == conf->last_hold)
3895 conf->bypass_count++;
3896 else {
3897 conf->last_hold = conf->hold_list.next;
3898 conf->bypass_count -= conf->bypass_threshold;
3899 if (conf->bypass_count < 0)
3900 conf->bypass_count = 0;
3901 }
3902 }
3903 } else if (!list_empty(&conf->hold_list) &&
3904 ((conf->bypass_threshold &&
3905 conf->bypass_count > conf->bypass_threshold) ||
3906 atomic_read(&conf->pending_full_writes) == 0)) {
3907 sh = list_entry(conf->hold_list.next,
3908 typeof(*sh), lru);
3909 conf->bypass_count -= conf->bypass_threshold;
3910 if (conf->bypass_count < 0)
3911 conf->bypass_count = 0;
3912 } else
3913 return NULL;
3914
3915 list_del_init(&sh->lru);
3916 atomic_inc(&sh->count);
3917 BUG_ON(atomic_read(&sh->count) != 1);
3918 return sh;
3919}
f679623f 3920
21a52c6d 3921static int make_request(mddev_t *mddev, struct bio * bi)
1da177e4 3922{
070ec55d 3923 raid5_conf_t *conf = mddev->private;
911d4ee8 3924 int dd_idx;
1da177e4
LT		 3925 sector_t new_sector;
3926 sector_t logical_sector, last_sector;
3927 struct stripe_head *sh;
a362357b 3928 const int rw = bio_data_dir(bi);
49077326 3929 int remaining;
7c13edc8 3930 int plugged;
1da177e4 3931
e9c7469b
TH		 3932 if (unlikely(bi->bi_rw & REQ_FLUSH)) {
3933 md_flush_request(mddev, bi);
e5dcdd80
N		 3934 return 0;
3935 }
3936
3d310eb7 3937 md_write_start(mddev, bi);
06d91a5f 3938
802ba064 3939 if (rw == READ &&
52488615 3940 mddev->reshape_position == MaxSector &&
21a52c6d 3941 chunk_aligned_read(mddev,bi))
99c0fb5f 3942 return 0;
52488615 3943
1da177e4
LT		 3944 logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
3945 last_sector = bi->bi_sector + (bi->bi_size>>9);
3946 bi->bi_next = NULL;
3947 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
06d91a5f 3948
7c13edc8 3949 plugged = mddev_check_plugged(mddev);
1da177e4
LT		 3950 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
3951 DEFINE_WAIT(w);
16a53ecc 3952 int disks, data_disks;
b5663ba4 3953 int previous;
b578d55f 3954
7ecaa1e6 3955 retry:
b5663ba4 3956 previous = 0;
b0f9ec04 3957 disks = conf->raid_disks;
b578d55f 3958 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
b0f9ec04 3959 if (unlikely(conf->reshape_progress != MaxSector)) {
fef9c61f 3960 /* spinlock is needed as reshape_progress may be
df8e7f76
N		 3961 * 64bit on a 32bit platform, and so it might be
3962 * possible to see a half-updated value
fef9c61f 3963 * Ofcourse reshape_progress could change after
df8e7f76
N		 3964 * the lock is dropped, so once we get a reference
3965 * to the stripe that we think it is, we will have
3966 * to check again.
3967 */
7ecaa1e6 3968 spin_lock_irq(&conf->device_lock);
fef9c61f
N		 3969 if (mddev->delta_disks < 0
3970 ? logical_sector < conf->reshape_progress
3971 : logical_sector >= conf->reshape_progress) {
7ecaa1e6 3972 disks = conf->previous_raid_disks;
b5663ba4
N		 3973 previous = 1;
3974 } else {
fef9c61f
N		 3975 if (mddev->delta_disks < 0
3976 ? logical_sector < conf->reshape_safe
3977 : logical_sector >= conf->reshape_safe) {
b578d55f
N		 3978 spin_unlock_irq(&conf->device_lock);
3979 schedule();
3980 goto retry;
3981 }
3982 }
7ecaa1e6
N		 3983 spin_unlock_irq(&conf->device_lock);
3984 }
16a53ecc
N		 3985 data_disks = disks - conf->max_degraded;
3986
112bf897
N		 3987 new_sector = raid5_compute_sector(conf, logical_sector,
3988 previous,
911d4ee8 3989 &dd_idx, NULL);
0c55e022 3990 pr_debug("raid456: make_request, sector %llu logical %llu\n",
1da177e4
LT		 3991 (unsigned long long)new_sector,
3992 (unsigned long long)logical_sector);
3993
b5663ba4 3994 sh = get_active_stripe(conf, new_sector, previous,
a8c906ca 3995 (bi->bi_rw&RWA_MASK), 0);
1da177e4 3996 if (sh) {
b0f9ec04 3997 if (unlikely(previous)) {
7ecaa1e6 3998 /* expansion might have moved on while waiting for a
df8e7f76
N		 3999 * stripe, so we must do the range check again.
4000 * Expansion could still move past after this
4001 * test, but as we are holding a reference to
4002 * 'sh', we know that if that happens,
4003 * STRIPE_EXPANDING will get set and the expansion
4004 * won't proceed until we finish with the stripe.
7ecaa1e6
N		 4005 */
4006 int must_retry = 0;
4007 spin_lock_irq(&conf->device_lock);
b0f9ec04
N		 4008 if (mddev->delta_disks < 0
4009 ? logical_sector >= conf->reshape_progress
4010 : logical_sector < conf->reshape_progress)
7ecaa1e6
N		 4011 /* mismatch, need to try again */
4012 must_retry = 1;
4013 spin_unlock_irq(&conf->device_lock);
4014 if (must_retry) {
4015 release_stripe(sh);
7a3ab908 4016 schedule();
7ecaa1e6
N		 4017 goto retry;
4018 }
4019 }
e62e58a5 4020
a5c308d4
N		 4021 if (bio_data_dir(bi) == WRITE &&
4022 logical_sector >= mddev->suspend_lo &&
e464eafd
N		 4023 logical_sector < mddev->suspend_hi) {
4024 release_stripe(sh);
e62e58a5
N		 4025 /* As the suspend_* range is controlled by
4026 * userspace, we want an interruptible
4027 * wait.
4028 */
4029 flush_signals(current);
4030 prepare_to_wait(&conf->wait_for_overlap,
4031 &w, TASK_INTERRUPTIBLE);
4032 if (logical_sector >= mddev->suspend_lo &&
4033 logical_sector < mddev->suspend_hi)
4034 schedule();
e464eafd
N		 4035 goto retry;
4036 }
7ecaa1e6
N		 4037
4038 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
4039 !add_stripe_bio(sh, bi, dd_idx, (bi->bi_rw&RW_MASK))) {
4040 /* Stripe is busy expanding or
4041 * add failed due to overlap. Flush everything
1da177e4
LT		 4042 * and wait a while
4043 */
482c0834 4044 md_wakeup_thread(mddev->thread);
1da177e4
LT		 4045 release_stripe(sh);
4046 schedule();
4047 goto retry;
4048 }
4049 finish_wait(&conf->wait_for_overlap, &w);
6ed3003c
N		 4050 set_bit(STRIPE_HANDLE, &sh->state);
4051 clear_bit(STRIPE_DELAYED, &sh->state);
e9c7469b 4052 if ((bi->bi_rw & REQ_SYNC) &&
729a1866
N		 4053 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4054 atomic_inc(&conf->preread_active_stripes);
1da177e4 4055 release_stripe(sh);
1da177e4
LT		 4056 } else {
4057 /* cannot get stripe for read-ahead, just give-up */
4058 clear_bit(BIO_UPTODATE, &bi->bi_flags);
4059 finish_wait(&conf->wait_for_overlap, &w);
4060 break;
4061 }
4062
4063 }
7c13edc8
N		 4064 if (!plugged)
4065 md_wakeup_thread(mddev->thread);
4066
1da177e4 4067 spin_lock_irq(&conf->device_lock);
960e739d 4068 remaining = raid5_dec_bi_phys_segments(bi);
f6344757
N		 4069 spin_unlock_irq(&conf->device_lock);
4070 if (remaining == 0) {
1da177e4 4071
16a53ecc 4072 if ( rw == WRITE )
1da177e4 4073 md_write_end(mddev);
6712ecf8 4074
0e13fe23 4075 bio_endio(bi, 0);
1da177e4 4076 }
729a1866 4077
1da177e4
LT		 4078 return 0;
4079}
4080
b522adcd
DW		 4081static sector_t raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks);
4082
52c03291 4083static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped)
1da177e4 4084{
52c03291
N		 4085 /* reshaping is quite different to recovery/resync so it is
4086 * handled quite separately ... here.
4087 *
4088 * On each call to sync_request, we gather one chunk worth of
4089 * destination stripes and flag them as expanding.
4090 * Then we find all the source stripes and request reads.
4091 * As the reads complete, handle_stripe will copy the data
4092 * into the destination stripe and release that stripe.
4093 */
7b92813c 4094 raid5_conf_t *conf = mddev->private;
1da177e4 4095 struct stripe_head *sh;
ccfcc3c1 4096 sector_t first_sector, last_sector;
f416885e
N		 4097 int raid_disks = conf->previous_raid_disks;
4098 int data_disks = raid_disks - conf->max_degraded;
4099 int new_data_disks = conf->raid_disks - conf->max_degraded;
52c03291
N		 4100 int i;
4101 int dd_idx;
c8f517c4 4102 sector_t writepos, readpos, safepos;
ec32a2bd 4103 sector_t stripe_addr;
7a661381 4104 int reshape_sectors;
ab69ae12 4105 struct list_head stripes;
52c03291 4106
fef9c61f
N		 4107 if (sector_nr == 0) {
4108 /* If restarting in the middle, skip the initial sectors */
4109 if (mddev->delta_disks < 0 &&
4110 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
4111 sector_nr = raid5_size(mddev, 0, 0)
4112 - conf->reshape_progress;
a639755c 4113 } else if (mddev->delta_disks >= 0 &&
fef9c61f
N		 4114 conf->reshape_progress > 0)
4115 sector_nr = conf->reshape_progress;
f416885e 4116 sector_div(sector_nr, new_data_disks);
fef9c61f 4117 if (sector_nr) {
8dee7211
N		 4118 mddev->curr_resync_completed = sector_nr;
4119 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
fef9c61f
N		 4120 *skipped = 1;
4121 return sector_nr;
4122 }
52c03291
N		 4123 }
4124
7a661381
N		 4125 /* We need to process a full chunk at a time.
4126 * If old and new chunk sizes differ, we need to process the
4127 * largest of these
4128 */
664e7c41
AN		 4129 if (mddev->new_chunk_sectors > mddev->chunk_sectors)
4130 reshape_sectors = mddev->new_chunk_sectors;
7a661381 4131 else
9d8f0363 4132 reshape_sectors = mddev->chunk_sectors;
7a661381 4133
52c03291
N		 4134 /* we update the metadata when there is more than 3Meg
4135 * in the block range (that is rather arbitrary, should
4136 * probably be time based) or when the data about to be
4137 * copied would over-write the source of the data at
4138 * the front of the range.
fef9c61f
N		 4139 * i.e. one new_stripe along from reshape_progress new_maps
4140 * to after where reshape_safe old_maps to
52c03291 4141 */
fef9c61f 4142 writepos = conf->reshape_progress;
f416885e 4143 sector_div(writepos, new_data_disks);
c8f517c4
N		 4144 readpos = conf->reshape_progress;
4145 sector_div(readpos, data_disks);
fef9c61f 4146 safepos = conf->reshape_safe;
f416885e 4147 sector_div(safepos, data_disks);
fef9c61f 4148 if (mddev->delta_disks < 0) {
ed37d83e 4149 writepos -= min_t(sector_t, reshape_sectors, writepos);
c8f517c4 4150 readpos += reshape_sectors;
7a661381 4151 safepos += reshape_sectors;
fef9c61f 4152 } else {
7a661381 4153 writepos += reshape_sectors;
ed37d83e
N		 4154 readpos -= min_t(sector_t, reshape_sectors, readpos);
4155 safepos -= min_t(sector_t, reshape_sectors, safepos);
fef9c61f 4156 }
52c03291 4157
c8f517c4
N		 4158 /* 'writepos' is the most advanced device address we might write.
4159 * 'readpos' is the least advanced device address we might read.
4160 * 'safepos' is the least address recorded in the metadata as having
4161 * been reshaped.
4162 * If 'readpos' is behind 'writepos', then there is no way that we can
4163 * ensure safety in the face of a crash - that must be done by userspace
4164 * making a backup of the data. So in that case there is no particular
4165 * rush to update metadata.
4166 * Otherwise if 'safepos' is behind 'writepos', then we really need to
4167 * update the metadata to advance 'safepos' to match 'readpos' so that
4168 * we can be safe in the event of a crash.
4169 * So we insist on updating metadata if safepos is behind writepos and
4170 * readpos is beyond writepos.
4171 * In any case, update the metadata every 10 seconds.
4172 * Maybe that number should be configurable, but I'm not sure it is
4173 * worth it.... maybe it could be a multiple of safemode_delay???
4174 */
fef9c61f 4175 if ((mddev->delta_disks < 0
c8f517c4
N		 4176 ? (safepos > writepos && readpos < writepos)
4177 : (safepos < writepos && readpos > writepos)) ||
4178 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
52c03291
N		 4179 /* Cannot proceed until we've updated the superblock... */
4180 wait_event(conf->wait_for_overlap,
4181 atomic_read(&conf->reshape_stripes)==0);
fef9c61f 4182 mddev->reshape_position = conf->reshape_progress;
75d3da43 4183 mddev->curr_resync_completed = sector_nr;
c8f517c4 4184 conf->reshape_checkpoint = jiffies;
850b2b42 4185 set_bit(MD_CHANGE_DEVS, &mddev->flags);
52c03291 4186 md_wakeup_thread(mddev->thread);
850b2b42 4187 wait_event(mddev->sb_wait, mddev->flags == 0 ||
52c03291
N		 4188 kthread_should_stop());
4189 spin_lock_irq(&conf->device_lock);
fef9c61f 4190 conf->reshape_safe = mddev->reshape_position;
52c03291
N		 4191 spin_unlock_irq(&conf->device_lock);
4192 wake_up(&conf->wait_for_overlap);
acb180b0 4193 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
52c03291
N		 4194 }
4195
ec32a2bd
N		 4196 if (mddev->delta_disks < 0) {
4197 BUG_ON(conf->reshape_progress == 0);
4198 stripe_addr = writepos;
4199 BUG_ON((mddev->dev_sectors &
7a661381
N		 4200 ~((sector_t)reshape_sectors - 1))
4201 - reshape_sectors - stripe_addr
ec32a2bd
N		 4202 != sector_nr);
4203 } else {
7a661381 4204 BUG_ON(writepos != sector_nr + reshape_sectors);
ec32a2bd
N		 4205 stripe_addr = sector_nr;
4206 }
ab69ae12 4207 INIT_LIST_HEAD(&stripes);
7a661381 4208 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
52c03291 4209 int j;
a9f326eb 4210 int skipped_disk = 0;
a8c906ca 4211 sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
52c03291
N		 4212 set_bit(STRIPE_EXPANDING, &sh->state);
4213 atomic_inc(&conf->reshape_stripes);
4214 /* If any of this stripe is beyond the end of the old
4215 * array, then we need to zero those blocks
4216 */
4217 for (j=sh->disks; j--;) {
4218 sector_t s;
4219 if (j == sh->pd_idx)
4220 continue;
f416885e 4221 if (conf->level == 6 &&
d0dabf7e 4222 j == sh->qd_idx)
f416885e 4223 continue;
784052ec 4224 s = compute_blocknr(sh, j, 0);
b522adcd 4225 if (s < raid5_size(mddev, 0, 0)) {
a9f326eb 4226 skipped_disk = 1;
52c03291
N		 4227 continue;
4228 }
4229 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
4230 set_bit(R5_Expanded, &sh->dev[j].flags);
4231 set_bit(R5_UPTODATE, &sh->dev[j].flags);
4232 }
a9f326eb 4233 if (!skipped_disk) {
52c03291
N		 4234 set_bit(STRIPE_EXPAND_READY, &sh->state);
4235 set_bit(STRIPE_HANDLE, &sh->state);
4236 }
ab69ae12 4237 list_add(&sh->lru, &stripes);
52c03291
N		 4238 }
4239 spin_lock_irq(&conf->device_lock);
fef9c61f 4240 if (mddev->delta_disks < 0)
7a661381 4241 conf->reshape_progress -= reshape_sectors * new_data_disks;
fef9c61f 4242 else
7a661381 4243 conf->reshape_progress += reshape_sectors * new_data_disks;
52c03291
N		 4244 spin_unlock_irq(&conf->device_lock);
4245 /* Ok, those stripe are ready. We can start scheduling
4246 * reads on the source stripes.
4247 * The source stripes are determined by mapping the first and last
4248 * block on the destination stripes.
4249 */
52c03291 4250 first_sector =
ec32a2bd 4251 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
911d4ee8 4252 1, &dd_idx, NULL);
52c03291 4253 last_sector =
0e6e0271 4254 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
09c9e5fa 4255 * new_data_disks - 1),
911d4ee8 4256 1, &dd_idx, NULL);
58c0fed4
AN		 4257 if (last_sector >= mddev->dev_sectors)
4258 last_sector = mddev->dev_sectors - 1;
52c03291 4259 while (first_sector <= last_sector) {
a8c906ca 4260 sh = get_active_stripe(conf, first_sector, 1, 0, 1);
52c03291
N		 4261 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
4262 set_bit(STRIPE_HANDLE, &sh->state);
4263 release_stripe(sh);
4264 first_sector += STRIPE_SECTORS;
4265 }
ab69ae12
N		 4266 /* Now that the sources are clearly marked, we can release
4267 * the destination stripes
4268 */
4269 while (!list_empty(&stripes)) {
4270 sh = list_entry(stripes.next, struct stripe_head, lru);
4271 list_del_init(&sh->lru);
4272 release_stripe(sh);
4273 }
c6207277
N		 4274 /* If this takes us to the resync_max point where we have to pause,
4275 * then we need to write out the superblock.
4276 */
7a661381 4277 sector_nr += reshape_sectors;
c03f6a19
N		 4278 if ((sector_nr - mddev->curr_resync_completed) * 2
4279 >= mddev->resync_max - mddev->curr_resync_completed) {
c6207277
N		 4280 /* Cannot proceed until we've updated the superblock... */
4281 wait_event(conf->wait_for_overlap,
4282 atomic_read(&conf->reshape_stripes) == 0);
fef9c61f 4283 mddev->reshape_position = conf->reshape_progress;
75d3da43 4284 mddev->curr_resync_completed = sector_nr;
c8f517c4 4285 conf->reshape_checkpoint = jiffies;
c6207277
N		 4286 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4287 md_wakeup_thread(mddev->thread);
4288 wait_event(mddev->sb_wait,
4289 !test_bit(MD_CHANGE_DEVS, &mddev->flags)
4290 || kthread_should_stop());
4291 spin_lock_irq(&conf->device_lock);
fef9c61f 4292 conf->reshape_safe = mddev->reshape_position;
c6207277
N		 4293 spin_unlock_irq(&conf->device_lock);
4294 wake_up(&conf->wait_for_overlap);
acb180b0 4295 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
c6207277 4296 }
7a661381 4297 return reshape_sectors;
52c03291
N		 4298}
4299
4300/* FIXME go_faster isn't used */
4301static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
4302{
7b92813c 4303 raid5_conf_t *conf = mddev->private;
52c03291 4304 struct stripe_head *sh;
58c0fed4 4305 sector_t max_sector = mddev->dev_sectors;
57dab0bd 4306 sector_t sync_blocks;
16a53ecc
N		 4307 int still_degraded = 0;
4308 int i;
1da177e4 4309
72626685 4310 if (sector_nr >= max_sector) {
1da177e4 4311 /* just being told to finish up .. nothing much to do */
cea9c228 4312
29269553
N		 4313 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
4314 end_reshape(conf);
4315 return 0;
4316 }
72626685
N		 4317
4318 if (mddev->curr_resync < max_sector) /* aborted */
4319 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
4320 &sync_blocks, 1);
16a53ecc 4321 else /* completed sync */
72626685
N		 4322 conf->fullsync = 0;
4323 bitmap_close_sync(mddev->bitmap);
4324
1da177e4
LT		 4325 return 0;
4326 }
ccfcc3c1 4327
64bd660b
N		 4328 /* Allow raid5_quiesce to complete */
4329 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
4330
52c03291
N		 4331 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4332 return reshape_request(mddev, sector_nr, skipped);
f6705578 4333
c6207277
N		 4334 /* No need to check resync_max as we never do more than one
4335 * stripe, and as resync_max will always be on a chunk boundary,
4336 * if the check in md_do_sync didn't fire, there is no chance
4337 * of overstepping resync_max here
4338 */
4339
16a53ecc 4340 /* if there is too many failed drives and we are trying
1da177e4
LT		 4341 * to resync, then assert that we are finished, because there is
4342 * nothing we can do.
4343 */
3285edf1 4344 if (mddev->degraded >= conf->max_degraded &&
16a53ecc 4345 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
58c0fed4 4346 sector_t rv = mddev->dev_sectors - sector_nr;
57afd89f 4347 *skipped = 1;
1da177e4
LT		 4348 return rv;
4349 }
72626685 4350 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
3855ad9f 4351 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
72626685
N		 4352 !conf->fullsync && sync_blocks >= STRIPE_SECTORS) {
4353 /* we can skip this block, and probably more */
4354 sync_blocks /= STRIPE_SECTORS;
4355 *skipped = 1;
4356 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
4357 }
1da177e4 4358
b47490c9
N		 4359
4360 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
4361
a8c906ca 4362 sh = get_active_stripe(conf, sector_nr, 0, 1, 0);
1da177e4 4363 if (sh == NULL) {
a8c906ca 4364 sh = get_active_stripe(conf, sector_nr, 0, 0, 0);
1da177e4 4365 /* make sure we don't swamp the stripe cache if someone else
16a53ecc 4366 * is trying to get access
1da177e4 4367 */
66c006a5 4368 schedule_timeout_uninterruptible(1);
1da177e4 4369 }
16a53ecc
N		 4370 /* Need to check if array will still be degraded after recovery/resync
4371 * We don't need to check the 'failed' flag as when that gets set,
4372 * recovery aborts.
4373 */
f001a70c 4374 for (i = 0; i < conf->raid_disks; i++)
16a53ecc
N		 4375 if (conf->disks[i].rdev == NULL)
4376 still_degraded = 1;
4377
4378 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
4379
4380 spin_lock(&sh->lock);
1da177e4
LT		 4381 set_bit(STRIPE_SYNCING, &sh->state);
4382 clear_bit(STRIPE_INSYNC, &sh->state);
4383 spin_unlock(&sh->lock);
4384
1442577b 4385 handle_stripe(sh);
1da177e4
LT		 4386 release_stripe(sh);
4387
4388 return STRIPE_SECTORS;
4389}
4390
46031f9a
RBJ		 4391static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio)
4392{
4393 /* We may not be able to submit a whole bio at once as there
4394 * may not be enough stripe_heads available.
4395 * We cannot pre-allocate enough stripe_heads as we may need
4396 * more than exist in the cache (if we allow ever large chunks).
4397 * So we do one stripe head at a time and record in
4398 * ->bi_hw_segments how many have been done.
4399 *
4400 * We *know* that this entire raid_bio is in one chunk, so
4401 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
4402 */
4403 struct stripe_head *sh;
911d4ee8 4404 int dd_idx;
46031f9a
RBJ		 4405 sector_t sector, logical_sector, last_sector;
4406 int scnt = 0;
4407 int remaining;
4408 int handled = 0;
4409
4410 logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
112bf897 4411 sector = raid5_compute_sector(conf, logical_sector,
911d4ee8 4412 0, &dd_idx, NULL);
46031f9a
RBJ		 4413 last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9);
4414
4415 for (; logical_sector < last_sector;
387bb173
NB		 4416 logical_sector += STRIPE_SECTORS,
4417 sector += STRIPE_SECTORS,
4418 scnt++) {
46031f9a 4419
960e739d 4420 if (scnt < raid5_bi_hw_segments(raid_bio))
46031f9a
RBJ		 4421 /* already done this stripe */
4422 continue;
4423
a8c906ca 4424 sh = get_active_stripe(conf, sector, 0, 1, 0);
46031f9a
RBJ		 4425
4426 if (!sh) {
4427 /* failed to get a stripe - must wait */
960e739d 4428 raid5_set_bi_hw_segments(raid_bio, scnt);
46031f9a
RBJ		 4429 conf->retry_read_aligned = raid_bio;
4430 return handled;
4431 }
4432
4433 set_bit(R5_ReadError, &sh->dev[dd_idx].flags);
387bb173
NB		 4434 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) {
4435 release_stripe(sh);
960e739d 4436 raid5_set_bi_hw_segments(raid_bio, scnt);
387bb173
NB		 4437 conf->retry_read_aligned = raid_bio;
4438 return handled;
4439 }
4440
36d1c647 4441 handle_stripe(sh);
46031f9a
RBJ		 4442 release_stripe(sh);
4443 handled++;
4444 }
4445 spin_lock_irq(&conf->device_lock);
960e739d 4446 remaining = raid5_dec_bi_phys_segments(raid_bio);
46031f9a 4447 spin_unlock_irq(&conf->device_lock);
0e13fe23
NB		 4448 if (remaining == 0)
4449 bio_endio(raid_bio, 0);
46031f9a
RBJ		 4450 if (atomic_dec_and_test(&conf->active_aligned_reads))
4451 wake_up(&conf->wait_for_stripe);
4452 return handled;
4453}
4454
46031f9a 4455
1da177e4
LT		 4456/*
4457 * This is our raid5 kernel thread.
4458 *
4459 * We scan the hash table for stripes which can be handled now.
4460 * During the scan, completed stripes are saved for us by the interrupt
4461 * handler, so that they will not have to wait for our next wakeup.
4462 */
6ed3003c 4463static void raid5d(mddev_t *mddev)
1da177e4
LT		 4464{
4465 struct stripe_head *sh;
070ec55d 4466 raid5_conf_t *conf = mddev->private;
1da177e4 4467 int handled;
e1dfa0a2 4468 struct blk_plug plug;
1da177e4 4469
45b4233c 4470 pr_debug("+++ raid5d active\n");
1da177e4
LT		 4471
4472 md_check_recovery(mddev);
1da177e4 4473
e1dfa0a2 4474 blk_start_plug(&plug);
1da177e4
LT		 4475 handled = 0;
4476 spin_lock_irq(&conf->device_lock);
4477 while (1) {
46031f9a 4478 struct bio *bio;
1da177e4 4479
7c13edc8
N		 4480 if (atomic_read(&mddev->plug_cnt) == 0 &&
4481 !list_empty(&conf->bitmap_list)) {
4482 /* Now is a good time to flush some bitmap updates */
4483 conf->seq_flush++;
700e432d 4484 spin_unlock_irq(&conf->device_lock);
72626685 4485 bitmap_unplug(mddev->bitmap);
700e432d 4486 spin_lock_irq(&conf->device_lock);
7c13edc8 4487 conf->seq_write = conf->seq_flush;
72626685
N		 4488 activate_bit_delay(conf);
4489 }
7c13edc8
N		 4490 if (atomic_read(&mddev->plug_cnt) == 0)
4491 raid5_activate_delayed(conf);
72626685 4492
46031f9a
RBJ		 4493 while ((bio = remove_bio_from_retry(conf))) {
4494 int ok;
4495 spin_unlock_irq(&conf->device_lock);
4496 ok = retry_aligned_read(conf, bio);
4497 spin_lock_irq(&conf->device_lock);
4498 if (!ok)
4499 break;
4500 handled++;
4501 }
4502
8b3e6cdc
DW		 4503 sh = __get_priority_stripe(conf);
4504
c9f21aaf 4505 if (!sh)
1da177e4 4506 break;
1da177e4
LT		 4507 spin_unlock_irq(&conf->device_lock);
4508
4509 handled++;
417b8d4a
DW		 4510 handle_stripe(sh);
4511 release_stripe(sh);
4512 cond_resched();
1da177e4
LT		 4513
4514 spin_lock_irq(&conf->device_lock);
4515 }
45b4233c 4516 pr_debug("%d stripes handled\n", handled);
1da177e4
LT		 4517
4518 spin_unlock_irq(&conf->device_lock);
4519
c9f21aaf 4520 async_tx_issue_pending_all();
e1dfa0a2 4521 blk_finish_plug(&plug);
1da177e4 4522
45b4233c 4523 pr_debug("--- raid5d inactive\n");
1da177e4
LT		 4524}
4525
3f294f4f 4526static ssize_t
007583c9 4527raid5_show_stripe_cache_size(mddev_t *mddev, char *page)
3f294f4f 4528{
070ec55d 4529 raid5_conf_t *conf = mddev->private;
96de1e66
N		 4530 if (conf)
4531 return sprintf(page, "%d\n", conf->max_nr_stripes);
4532 else
4533 return 0;
3f294f4f
N		 4534}
4535
c41d4ac4
N		 4536int
4537raid5_set_cache_size(mddev_t *mddev, int size)
3f294f4f 4538{
070ec55d 4539 raid5_conf_t *conf = mddev->private;
b5470dc5
DW		 4540 int err;
4541
c41d4ac4 4542 if (size <= 16 || size > 32768)
3f294f4f 4543 return -EINVAL;
c41d4ac4 4544 while (size < conf->max_nr_stripes) {
3f294f4f
N		 4545 if (drop_one_stripe(conf))
4546 conf->max_nr_stripes--;
4547 else
4548 break;
4549 }
b5470dc5
DW		 4550 err = md_allow_write(mddev);
4551 if (err)
4552 return err;
c41d4ac4 4553 while (size > conf->max_nr_stripes) {
3f294f4f
N		 4554 if (grow_one_stripe(conf))
4555 conf->max_nr_stripes++;
4556 else break;
4557 }
c41d4ac4
N		 4558 return 0;
4559}
4560EXPORT_SYMBOL(raid5_set_cache_size);
4561
4562static ssize_t
4563raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len)
4564{
4565 raid5_conf_t *conf = mddev->private;
4566 unsigned long new;
4567 int err;
4568
4569 if (len >= PAGE_SIZE)
4570 return -EINVAL;
4571 if (!conf)
4572 return -ENODEV;
4573
4574 if (strict_strtoul(page, 10, &new))
4575 return -EINVAL;
4576 err = raid5_set_cache_size(mddev, new);
4577 if (err)
4578 return err;
3f294f4f
N		 4579 return len;
4580}
007583c9 4581
96de1e66
N		 4582static struct md_sysfs_entry
4583raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
4584 raid5_show_stripe_cache_size,
4585 raid5_store_stripe_cache_size);
3f294f4f 4586
8b3e6cdc
DW		 4587static ssize_t
4588raid5_show_preread_threshold(mddev_t *mddev, char *page)
4589{
070ec55d 4590 raid5_conf_t *conf = mddev->private;
8b3e6cdc
DW		 4591 if (conf)
4592 return sprintf(page, "%d\n", conf->bypass_threshold);
4593 else
4594 return 0;
4595}
4596
4597static ssize_t
4598raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len)
4599{
070ec55d 4600 raid5_conf_t *conf = mddev->private;
4ef197d8 4601 unsigned long new;
8b3e6cdc
DW		 4602 if (len >= PAGE_SIZE)
4603 return -EINVAL;
4604 if (!conf)
4605 return -ENODEV;
4606
4ef197d8 4607 if (strict_strtoul(page, 10, &new))
8b3e6cdc 4608 return -EINVAL;
4ef197d8 4609 if (new > conf->max_nr_stripes)
8b3e6cdc
DW		 4610 return -EINVAL;
4611 conf->bypass_threshold = new;
4612 return len;
4613}
4614
4615static struct md_sysfs_entry
4616raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
4617 S_IRUGO | S_IWUSR,
4618 raid5_show_preread_threshold,
4619 raid5_store_preread_threshold);
4620
3f294f4f 4621static ssize_t
96de1e66 4622stripe_cache_active_show(mddev_t *mddev, char *page)
3f294f4f 4623{
070ec55d 4624 raid5_conf_t *conf = mddev->private;
96de1e66
N		 4625 if (conf)
4626 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
4627 else
4628 return 0;
3f294f4f
N		 4629}
4630
96de1e66
N		 4631static struct md_sysfs_entry
4632raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
3f294f4f 4633
007583c9 4634static struct attribute *raid5_attrs[] = {
3f294f4f
N		 4635 &raid5_stripecache_size.attr,
4636 &raid5_stripecache_active.attr,
8b3e6cdc 4637 &raid5_preread_bypass_threshold.attr,
3f294f4f
N		 4638 NULL,
4639};
007583c9
N		 4640static struct attribute_group raid5_attrs_group = {
4641 .name = NULL,
4642 .attrs = raid5_attrs,
3f294f4f
N		 4643};
4644
80c3a6ce
DW		 4645static sector_t
4646raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks)
4647{
070ec55d 4648 raid5_conf_t *conf = mddev->private;
80c3a6ce
DW		 4649
4650 if (!sectors)
4651 sectors = mddev->dev_sectors;
5e5e3e78 4652 if (!raid_disks)
7ec05478 4653 /* size is defined by the smallest of previous and new size */
5e5e3e78 4654 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
80c3a6ce 4655
9d8f0363 4656 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
664e7c41 4657 sectors &= ~((sector_t)mddev->new_chunk_sectors - 1);
80c3a6ce
DW		 4658 return sectors * (raid_disks - conf->max_degraded);
4659}
4660
36d1c647
DW		 4661static void raid5_free_percpu(raid5_conf_t *conf)
4662{
4663 struct raid5_percpu *percpu;
4664 unsigned long cpu;
4665
4666 if (!conf->percpu)
4667 return;
4668
4669 get_online_cpus();
4670 for_each_possible_cpu(cpu) {
4671 percpu = per_cpu_ptr(conf->percpu, cpu);
4672 safe_put_page(percpu->spare_page);
d6f38f31 4673 kfree(percpu->scribble);
36d1c647
DW		 4674 }
4675#ifdef CONFIG_HOTPLUG_CPU
4676 unregister_cpu_notifier(&conf->cpu_notify);
4677#endif
4678 put_online_cpus();
4679
4680 free_percpu(conf->percpu);
4681}
4682
95fc17aa
DW		 4683static void free_conf(raid5_conf_t *conf)
4684{
4685 shrink_stripes(conf);
36d1c647 4686 raid5_free_percpu(conf);
95fc17aa
DW		 4687 kfree(conf->disks);
4688 kfree(conf->stripe_hashtbl);
4689 kfree(conf);
4690}
4691
36d1c647
DW		 4692#ifdef CONFIG_HOTPLUG_CPU
4693static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action,
4694 void *hcpu)
4695{
4696 raid5_conf_t *conf = container_of(nfb, raid5_conf_t, cpu_notify);
4697 long cpu = (long)hcpu;
4698 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
4699
4700 switch (action) {
4701 case CPU_UP_PREPARE:
4702 case CPU_UP_PREPARE_FROZEN:
d6f38f31 4703 if (conf->level == 6 && !percpu->spare_page)
36d1c647 4704 percpu->spare_page = alloc_page(GFP_KERNEL);
d6f38f31
DW		 4705 if (!percpu->scribble)
4706 percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
4707
4708 if (!percpu->scribble ||
4709 (conf->level == 6 && !percpu->spare_page)) {
4710 safe_put_page(percpu->spare_page);
4711 kfree(percpu->scribble);
36d1c647
DW		 4712 pr_err("%s: failed memory allocation for cpu%ld\n",
4713 __func__, cpu);
55af6bb5 4714 return notifier_from_errno(-ENOMEM);
36d1c647
DW		 4715 }
4716 break;
4717 case CPU_DEAD:
4718 case CPU_DEAD_FROZEN:
4719 safe_put_page(percpu->spare_page);
d6f38f31 4720 kfree(percpu->scribble);
36d1c647 4721 percpu->spare_page = NULL;
d6f38f31 4722 percpu->scribble = NULL;
36d1c647
DW		 4723 break;
4724 default:
4725 break;
4726 }
4727 return NOTIFY_OK;
4728}
4729#endif
4730
4731static int raid5_alloc_percpu(raid5_conf_t *conf)
4732{
4733 unsigned long cpu;
4734 struct page *spare_page;
a29d8b8e 4735 struct raid5_percpu __percpu *allcpus;
d6f38f31 4736 void *scribble;
36d1c647
DW		 4737 int err;
4738
36d1c647
DW		 4739 allcpus = alloc_percpu(struct raid5_percpu);
4740 if (!allcpus)
4741 return -ENOMEM;
4742 conf->percpu = allcpus;
4743
4744 get_online_cpus();
4745 err = 0;
4746 for_each_present_cpu(cpu) {
d6f38f31
DW		 4747 if (conf->level == 6) {
4748 spare_page = alloc_page(GFP_KERNEL);
4749 if (!spare_page) {
4750 err = -ENOMEM;
4751 break;
4752 }
4753 per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page;
4754 }
5e5e3e78 4755 scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
d6f38f31 4756 if (!scribble) {
36d1c647
DW		 4757 err = -ENOMEM;
4758 break;
4759 }
d6f38f31 4760 per_cpu_ptr(conf->percpu, cpu)->scribble = scribble;
36d1c647
DW		 4761 }
4762#ifdef CONFIG_HOTPLUG_CPU
4763 conf->cpu_notify.notifier_call = raid456_cpu_notify;
4764 conf->cpu_notify.priority = 0;
4765 if (err == 0)
4766 err = register_cpu_notifier(&conf->cpu_notify);
4767#endif
4768 put_online_cpus();
4769
4770 return err;
4771}
4772
91adb564 4773static raid5_conf_t *setup_conf(mddev_t *mddev)
1da177e4
LT		 4774{
4775 raid5_conf_t *conf;
5e5e3e78 4776 int raid_disk, memory, max_disks;
1da177e4
LT		 4777 mdk_rdev_t *rdev;
4778 struct disk_info *disk;
1da177e4 4779
91adb564
N		 4780 if (mddev->new_level != 5
4781 && mddev->new_level != 4
4782 && mddev->new_level != 6) {
0c55e022 4783 printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n",
91adb564
N		 4784 mdname(mddev), mddev->new_level);
4785 return ERR_PTR(-EIO);
1da177e4 4786 }
91adb564
N		 4787 if ((mddev->new_level == 5
4788 && !algorithm_valid_raid5(mddev->new_layout)) ||
4789 (mddev->new_level == 6
4790 && !algorithm_valid_raid6(mddev->new_layout))) {
0c55e022 4791 printk(KERN_ERR "md/raid:%s: layout %d not supported\n",
91adb564
N		 4792 mdname(mddev), mddev->new_layout);
4793 return ERR_PTR(-EIO);
99c0fb5f 4794 }
91adb564 4795 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
0c55e022 4796 printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n",
91adb564
N		 4797 mdname(mddev), mddev->raid_disks);
4798 return ERR_PTR(-EINVAL);
4bbf3771
N		 4799 }
4800
664e7c41
AN		 4801 if (!mddev->new_chunk_sectors ||
4802 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
4803 !is_power_of_2(mddev->new_chunk_sectors)) {
0c55e022
N		 4804 printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n",
4805 mdname(mddev), mddev->new_chunk_sectors << 9);
91adb564 4806 return ERR_PTR(-EINVAL);
f6705578
N		 4807 }
4808
91adb564
N		 4809 conf = kzalloc(sizeof(raid5_conf_t), GFP_KERNEL);
4810 if (conf == NULL)
1da177e4 4811 goto abort;
f5efd45a
DW		 4812 spin_lock_init(&conf->device_lock);
4813 init_waitqueue_head(&conf->wait_for_stripe);
4814 init_waitqueue_head(&conf->wait_for_overlap);
4815 INIT_LIST_HEAD(&conf->handle_list);
4816 INIT_LIST_HEAD(&conf->hold_list);
4817 INIT_LIST_HEAD(&conf->delayed_list);
4818 INIT_LIST_HEAD(&conf->bitmap_list);
4819 INIT_LIST_HEAD(&conf->inactive_list);
4820 atomic_set(&conf->active_stripes, 0);
4821 atomic_set(&conf->preread_active_stripes, 0);
4822 atomic_set(&conf->active_aligned_reads, 0);
4823 conf->bypass_threshold = BYPASS_THRESHOLD;
91adb564
N		 4824
4825 conf->raid_disks = mddev->raid_disks;
4826 if (mddev->reshape_position == MaxSector)
4827 conf->previous_raid_disks = mddev->raid_disks;
4828 else
f6705578 4829 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
5e5e3e78
N		 4830 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
4831 conf->scribble_len = scribble_len(max_disks);
f6705578 4832
5e5e3e78 4833 conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
b55e6bfc
N		 4834 GFP_KERNEL);
4835 if (!conf->disks)
4836 goto abort;
9ffae0cf 4837
1da177e4
LT		 4838 conf->mddev = mddev;
4839
fccddba0 4840 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
1da177e4 4841 goto abort;
1da177e4 4842
36d1c647
DW		 4843 conf->level = mddev->new_level;
4844 if (raid5_alloc_percpu(conf) != 0)
4845 goto abort;
4846
0c55e022 4847 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
1da177e4 4848
159ec1fc 4849 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4 4850 raid_disk = rdev->raid_disk;
5e5e3e78 4851 if (raid_disk >= max_disks
1da177e4
LT		 4852 || raid_disk < 0)
4853 continue;
4854 disk = conf->disks + raid_disk;
4855
4856 disk->rdev = rdev;
4857
b2d444d7 4858 if (test_bit(In_sync, &rdev->flags)) {
1da177e4 4859 char b[BDEVNAME_SIZE];
0c55e022
N		 4860 printk(KERN_INFO "md/raid:%s: device %s operational as raid"
4861 " disk %d\n",
4862 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
8c2e870a
NB		 4863 } else
4864 /* Cannot rely on bitmap to complete recovery */
4865 conf->fullsync = 1;
1da177e4
LT		 4866 }
4867
09c9e5fa 4868 conf->chunk_sectors = mddev->new_chunk_sectors;
91adb564 4869 conf->level = mddev->new_level;
16a53ecc
N		 4870 if (conf->level == 6)
4871 conf->max_degraded = 2;
4872 else
4873 conf->max_degraded = 1;
91adb564 4874 conf->algorithm = mddev->new_layout;
1da177e4 4875 conf->max_nr_stripes = NR_STRIPES;
fef9c61f 4876 conf->reshape_progress = mddev->reshape_position;
e183eaed 4877 if (conf->reshape_progress != MaxSector) {
09c9e5fa 4878 conf->prev_chunk_sectors = mddev->chunk_sectors;
e183eaed
N		 4879 conf->prev_algo = mddev->layout;
4880 }
1da177e4 4881
91adb564 4882 memory = conf->max_nr_stripes * (sizeof(struct stripe_head) +
5e5e3e78 4883 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
91adb564
N		 4884 if (grow_stripes(conf, conf->max_nr_stripes)) {
4885 printk(KERN_ERR
0c55e022
N		 4886 "md/raid:%s: couldn't allocate %dkB for buffers\n",
4887 mdname(mddev), memory);
91adb564
N		 4888 goto abort;
4889 } else
0c55e022
N		 4890 printk(KERN_INFO "md/raid:%s: allocated %dkB\n",
4891 mdname(mddev), memory);
1da177e4 4892
0da3c619 4893 conf->thread = md_register_thread(raid5d, mddev, NULL);
91adb564
N		 4894 if (!conf->thread) {
4895 printk(KERN_ERR
0c55e022 4896 "md/raid:%s: couldn't allocate thread.\n",
91adb564 4897 mdname(mddev));
16a53ecc
N		 4898 goto abort;
4899 }
91adb564
N		 4900
4901 return conf;
4902
4903 abort:
4904 if (conf) {
95fc17aa 4905 free_conf(conf);
91adb564
N		 4906 return ERR_PTR(-EIO);
4907 } else
4908 return ERR_PTR(-ENOMEM);
4909}
4910
c148ffdc
N		 4911
4912static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
4913{
4914 switch (algo) {
4915 case ALGORITHM_PARITY_0:
4916 if (raid_disk < max_degraded)
4917 return 1;
4918 break;
4919 case ALGORITHM_PARITY_N:
4920 if (raid_disk >= raid_disks - max_degraded)
4921 return 1;
4922 break;
4923 case ALGORITHM_PARITY_0_6:
4924 if (raid_disk == 0 ||
4925 raid_disk == raid_disks - 1)
4926 return 1;
4927 break;
4928 case ALGORITHM_LEFT_ASYMMETRIC_6:
4929 case ALGORITHM_RIGHT_ASYMMETRIC_6:
4930 case ALGORITHM_LEFT_SYMMETRIC_6:
4931 case ALGORITHM_RIGHT_SYMMETRIC_6:
4932 if (raid_disk == raid_disks - 1)
4933 return 1;
4934 }
4935 return 0;
4936}
4937
91adb564
N		 4938static int run(mddev_t *mddev)
4939{
4940 raid5_conf_t *conf;
9f7c2220 4941 int working_disks = 0;
c148ffdc 4942 int dirty_parity_disks = 0;
91adb564 4943 mdk_rdev_t *rdev;
c148ffdc 4944 sector_t reshape_offset = 0;
91adb564 4945
8c6ac868 4946 if (mddev->recovery_cp != MaxSector)
0c55e022 4947 printk(KERN_NOTICE "md/raid:%s: not clean"
8c6ac868
AN		 4948 " -- starting background reconstruction\n",
4949 mdname(mddev));
91adb564
N		 4950 if (mddev->reshape_position != MaxSector) {
4951 /* Check that we can continue the reshape.
4952 * Currently only disks can change, it must
4953 * increase, and we must be past the point where
4954 * a stripe over-writes itself
4955 */
4956 sector_t here_new, here_old;
4957 int old_disks;
18b00334 4958 int max_degraded = (mddev->level == 6 ? 2 : 1);
91adb564 4959
88ce4930 4960 if (mddev->new_level != mddev->level) {
0c55e022 4961 printk(KERN_ERR "md/raid:%s: unsupported reshape "
91adb564
N		 4962 "required - aborting.\n",
4963 mdname(mddev));
4964 return -EINVAL;
4965 }
91adb564
N		 4966 old_disks = mddev->raid_disks - mddev->delta_disks;
4967 /* reshape_position must be on a new-stripe boundary, and one
4968 * further up in new geometry must map after here in old
4969 * geometry.
4970 */
4971 here_new = mddev->reshape_position;
664e7c41 4972 if (sector_div(here_new, mddev->new_chunk_sectors *
91adb564 4973 (mddev->raid_disks - max_degraded))) {
0c55e022
N		 4974 printk(KERN_ERR "md/raid:%s: reshape_position not "
4975 "on a stripe boundary\n", mdname(mddev));
91adb564
N		 4976 return -EINVAL;
4977 }
c148ffdc 4978 reshape_offset = here_new * mddev->new_chunk_sectors;
91adb564
N		 4979 /* here_new is the stripe we will write to */
4980 here_old = mddev->reshape_position;
9d8f0363 4981 sector_div(here_old, mddev->chunk_sectors *
91adb564
N		 4982 (old_disks-max_degraded));
4983 /* here_old is the first stripe that we might need to read
4984 * from */
67ac6011
N		 4985 if (mddev->delta_disks == 0) {
4986 /* We cannot be sure it is safe to start an in-place
4987 * reshape. It is only safe if user-space if monitoring
4988 * and taking constant backups.
4989 * mdadm always starts a situation like this in
4990 * readonly mode so it can take control before
4991 * allowing any writes. So just check for that.
4992 */
4993 if ((here_new * mddev->new_chunk_sectors !=
4994 here_old * mddev->chunk_sectors) ||
4995 mddev->ro == 0) {
0c55e022
N		 4996 printk(KERN_ERR "md/raid:%s: in-place reshape must be started"
4997 " in read-only mode - aborting\n",
4998 mdname(mddev));
67ac6011
N		 4999 return -EINVAL;
5000 }
5001 } else if (mddev->delta_disks < 0
5002 ? (here_new * mddev->new_chunk_sectors <=
5003 here_old * mddev->chunk_sectors)
5004 : (here_new * mddev->new_chunk_sectors >=
5005 here_old * mddev->chunk_sectors)) {
91adb564 5006 /* Reading from the same stripe as writing to - bad */
0c55e022
N		 5007 printk(KERN_ERR "md/raid:%s: reshape_position too early for "
5008 "auto-recovery - aborting.\n",
5009 mdname(mddev));
91adb564
N		 5010 return -EINVAL;
5011 }
0c55e022
N		 5012 printk(KERN_INFO "md/raid:%s: reshape will continue\n",
5013 mdname(mddev));
91adb564
N		 5014 /* OK, we should be able to continue; */
5015 } else {
5016 BUG_ON(mddev->level != mddev->new_level);
5017 BUG_ON(mddev->layout != mddev->new_layout);
664e7c41 5018 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
91adb564 5019 BUG_ON(mddev->delta_disks != 0);
1da177e4 5020 }
91adb564 5021
245f46c2
N		 5022 if (mddev->private == NULL)
5023 conf = setup_conf(mddev);
5024 else
5025 conf = mddev->private;
5026
91adb564
N		 5027 if (IS_ERR(conf))
5028 return PTR_ERR(conf);
5029
5030 mddev->thread = conf->thread;
5031 conf->thread = NULL;
5032 mddev->private = conf;
5033
5034 /*
5035 * 0 for a fully functional array, 1 or 2 for a degraded array.
5036 */
c148ffdc
N		 5037 list_for_each_entry(rdev, &mddev->disks, same_set) {
5038 if (rdev->raid_disk < 0)
5039 continue;
2f115882 5040 if (test_bit(In_sync, &rdev->flags)) {
91adb564 5041 working_disks++;
2f115882
N		 5042 continue;
5043 }
c148ffdc
N		 5044 /* This disc is not fully in-sync. However if it
5045 * just stored parity (beyond the recovery_offset),
5046 * when we don't need to be concerned about the
5047 * array being dirty.
5048 * When reshape goes 'backwards', we never have
5049 * partially completed devices, so we only need
5050 * to worry about reshape going forwards.
5051 */
5052 /* Hack because v0.91 doesn't store recovery_offset properly. */
5053 if (mddev->major_version == 0 &&
5054 mddev->minor_version > 90)
5055 rdev->recovery_offset = reshape_offset;
5056
c148ffdc
N		 5057 if (rdev->recovery_offset < reshape_offset) {
5058 /* We need to check old and new layout */
5059 if (!only_parity(rdev->raid_disk,
5060 conf->algorithm,
5061 conf->raid_disks,
5062 conf->max_degraded))
5063 continue;
5064 }
5065 if (!only_parity(rdev->raid_disk,
5066 conf->prev_algo,
5067 conf->previous_raid_disks,
5068 conf->max_degraded))
5069 continue;
5070 dirty_parity_disks++;
5071 }
91adb564 5072
5e5e3e78
N		 5073 mddev->degraded = (max(conf->raid_disks, conf->previous_raid_disks)
5074 - working_disks);
91adb564 5075
674806d6 5076 if (has_failed(conf)) {
0c55e022 5077 printk(KERN_ERR "md/raid:%s: not enough operational devices"
1da177e4 5078 " (%d/%d failed)\n",
02c2de8c 5079 mdname(mddev), mddev->degraded, conf->raid_disks);
1da177e4
LT		 5080 goto abort;
5081 }
5082
91adb564 5083 /* device size must be a multiple of chunk size */
9d8f0363 5084 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
91adb564
N		 5085 mddev->resync_max_sectors = mddev->dev_sectors;
5086
c148ffdc 5087 if (mddev->degraded > dirty_parity_disks &&
1da177e4 5088 mddev->recovery_cp != MaxSector) {
6ff8d8ec
N		 5089 if (mddev->ok_start_degraded)
5090 printk(KERN_WARNING
0c55e022
N		 5091 "md/raid:%s: starting dirty degraded array"
5092 " - data corruption possible.\n",
6ff8d8ec
N		 5093 mdname(mddev));
5094 else {
5095 printk(KERN_ERR
0c55e022 5096 "md/raid:%s: cannot start dirty degraded array.\n",
6ff8d8ec
N		 5097 mdname(mddev));
5098 goto abort;
5099 }
1da177e4
LT		 5100 }
5101
1da177e4 5102 if (mddev->degraded == 0)
0c55e022
N		 5103 printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d"
5104 " devices, algorithm %d\n", mdname(mddev), conf->level,
e183eaed
N		 5105 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
5106 mddev->new_layout);
1da177e4 5107 else
0c55e022
N		 5108 printk(KERN_ALERT "md/raid:%s: raid level %d active with %d"
5109 " out of %d devices, algorithm %d\n",
5110 mdname(mddev), conf->level,
5111 mddev->raid_disks - mddev->degraded,
5112 mddev->raid_disks, mddev->new_layout);
1da177e4
LT		 5113
5114 print_raid5_conf(conf);
5115
fef9c61f 5116 if (conf->reshape_progress != MaxSector) {
fef9c61f 5117 conf->reshape_safe = conf->reshape_progress;
f6705578
N		 5118 atomic_set(&conf->reshape_stripes, 0);
5119 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5120 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5121 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5122 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5123 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 5124 "reshape");
f6705578
N		 5125 }
5126
1da177e4
LT		 5127
5128 /* Ok, everything is just fine now */
a64c876f
N		 5129 if (mddev->to_remove == &raid5_attrs_group)
5130 mddev->to_remove = NULL;
00bcb4ac
N		 5131 else if (mddev->kobj.sd &&
5132 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
5e55e2f5 5133 printk(KERN_WARNING
4a5add49 5134 "raid5: failed to create sysfs attributes for %s\n",
5e55e2f5 5135 mdname(mddev));
4a5add49 5136 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
7a5febe9 5137
4a5add49 5138 if (mddev->queue) {
9f7c2220 5139 int chunk_size;
4a5add49
N		 5140 /* read-ahead size must cover two whole stripes, which
5141 * is 2 * (datadisks) * chunksize where 'n' is the
5142 * number of raid devices
5143 */
5144 int data_disks = conf->previous_raid_disks - conf->max_degraded;
5145 int stripe = data_disks *
5146 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
5147 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
5148 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
91adb564 5149
4a5add49 5150 blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec);
f022b2fd 5151
11d8a6e3
N		 5152 mddev->queue->backing_dev_info.congested_data = mddev;
5153 mddev->queue->backing_dev_info.congested_fn = raid5_congested;
7a5febe9 5154
9f7c2220
N		 5155 chunk_size = mddev->chunk_sectors << 9;
5156 blk_queue_io_min(mddev->queue, chunk_size);
5157 blk_queue_io_opt(mddev->queue, chunk_size *
5158 (conf->raid_disks - conf->max_degraded));
8f6c2e4b 5159
9f7c2220
N		 5160 list_for_each_entry(rdev, &mddev->disks, same_set)
5161 disk_stack_limits(mddev->gendisk, rdev->bdev,
5162 rdev->data_offset << 9);
5163 }
23032a0e 5164
1da177e4
LT		 5165 return 0;
5166abort:
e0cf8f04 5167 md_unregister_thread(mddev->thread);
91adb564 5168 mddev->thread = NULL;
1da177e4
LT		 5169 if (conf) {
5170 print_raid5_conf(conf);
95fc17aa 5171 free_conf(conf);
1da177e4
LT		 5172 }
5173 mddev->private = NULL;
0c55e022 5174 printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev));
1da177e4
LT		 5175 return -EIO;
5176}
5177
3f294f4f 5178static int stop(mddev_t *mddev)
1da177e4 5179{
7b92813c 5180 raid5_conf_t *conf = mddev->private;
1da177e4
LT		 5181
5182 md_unregister_thread(mddev->thread);
5183 mddev->thread = NULL;
11d8a6e3
N		 5184 if (mddev->queue)
5185 mddev->queue->backing_dev_info.congested_fn = NULL;
95fc17aa 5186 free_conf(conf);
a64c876f
N		 5187 mddev->private = NULL;
5188 mddev->to_remove = &raid5_attrs_group;
1da177e4
LT		 5189 return 0;
5190}
5191
45b4233c 5192#ifdef DEBUG
d710e138 5193static void print_sh(struct seq_file *seq, struct stripe_head *sh)
1da177e4
LT		 5194{
5195 int i;
5196
16a53ecc
N		 5197 seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n",
5198 (unsigned long long)sh->sector, sh->pd_idx, sh->state);
5199 seq_printf(seq, "sh %llu, count %d.\n",
5200 (unsigned long long)sh->sector, atomic_read(&sh->count));
5201 seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector);
7ecaa1e6 5202 for (i = 0; i < sh->disks; i++) {
16a53ecc
N		 5203 seq_printf(seq, "(cache%d: %p %ld) ",
5204 i, sh->dev[i].page, sh->dev[i].flags);
1da177e4 5205 }
16a53ecc 5206 seq_printf(seq, "\n");
1da177e4
LT		 5207}
5208
d710e138 5209static void printall(struct seq_file *seq, raid5_conf_t *conf)
1da177e4
LT		 5210{
5211 struct stripe_head *sh;
fccddba0 5212 struct hlist_node *hn;
1da177e4
LT		 5213 int i;
5214
5215 spin_lock_irq(&conf->device_lock);
5216 for (i = 0; i < NR_HASH; i++) {
fccddba0 5217 hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) {
1da177e4
LT		 5218 if (sh->raid_conf != conf)
5219 continue;
16a53ecc 5220 print_sh(seq, sh);
1da177e4
LT		 5221 }
5222 }
5223 spin_unlock_irq(&conf->device_lock);
5224}
5225#endif
5226
d710e138 5227static void status(struct seq_file *seq, mddev_t *mddev)
1da177e4 5228{
7b92813c 5229 raid5_conf_t *conf = mddev->private;
1da177e4
LT		 5230 int i;
5231
9d8f0363
AN		 5232 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
5233 mddev->chunk_sectors / 2, mddev->layout);
02c2de8c 5234 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
1da177e4
LT		 5235 for (i = 0; i < conf->raid_disks; i++)
5236 seq_printf (seq, "%s",
5237 conf->disks[i].rdev &&
b2d444d7 5238 test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_");
1da177e4 5239 seq_printf (seq, "]");
45b4233c 5240#ifdef DEBUG
16a53ecc
N		 5241 seq_printf (seq, "\n");
5242 printall(seq, conf);
1da177e4
LT		 5243#endif
5244}
5245
5246static void print_raid5_conf (raid5_conf_t *conf)
5247{
5248 int i;
5249 struct disk_info *tmp;
5250
0c55e022 5251 printk(KERN_DEBUG "RAID conf printout:\n");
1da177e4
LT		 5252 if (!conf) {
5253 printk("(conf==NULL)\n");
5254 return;
5255 }
0c55e022
N		 5256 printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level,
5257 conf->raid_disks,
5258 conf->raid_disks - conf->mddev->degraded);
1da177e4
LT		 5259
5260 for (i = 0; i < conf->raid_disks; i++) {
5261 char b[BDEVNAME_SIZE];
5262 tmp = conf->disks + i;
5263 if (tmp->rdev)
0c55e022
N		 5264 printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n",
5265 i, !test_bit(Faulty, &tmp->rdev->flags),
5266 bdevname(tmp->rdev->bdev, b));
1da177e4
LT		 5267 }
5268}
5269
5270static int raid5_spare_active(mddev_t *mddev)
5271{
5272 int i;
5273 raid5_conf_t *conf = mddev->private;
5274 struct disk_info *tmp;
6b965620
N		 5275 int count = 0;
5276 unsigned long flags;
1da177e4
LT		 5277
5278 for (i = 0; i < conf->raid_disks; i++) {
5279 tmp = conf->disks + i;
5280 if (tmp->rdev
70fffd0b 5281 && tmp->rdev->recovery_offset == MaxSector
b2d444d7 5282 && !test_bit(Faulty, &tmp->rdev->flags)
c04be0aa 5283 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
6b965620 5284 count++;
43c73ca4 5285 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
1da177e4
LT		 5286 }
5287 }
6b965620
N		 5288 spin_lock_irqsave(&conf->device_lock, flags);
5289 mddev->degraded -= count;
5290 spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4 5291 print_raid5_conf(conf);
6b965620 5292 return count;
1da177e4
LT		 5293}
5294
5295static int raid5_remove_disk(mddev_t *mddev, int number)
5296{
5297 raid5_conf_t *conf = mddev->private;
5298 int err = 0;
5299 mdk_rdev_t *rdev;
5300 struct disk_info *p = conf->disks + number;
5301
5302 print_raid5_conf(conf);
5303 rdev = p->rdev;
5304 if (rdev) {
ec32a2bd
N		 5305 if (number >= conf->raid_disks &&
5306 conf->reshape_progress == MaxSector)
5307 clear_bit(In_sync, &rdev->flags);
5308
b2d444d7 5309 if (test_bit(In_sync, &rdev->flags) ||
1da177e4
LT		 5310 atomic_read(&rdev->nr_pending)) {
5311 err = -EBUSY;
5312 goto abort;
5313 }
dfc70645
N		 5314 /* Only remove non-faulty devices if recovery
5315 * isn't possible.
5316 */
5317 if (!test_bit(Faulty, &rdev->flags) &&
674806d6 5318 !has_failed(conf) &&
ec32a2bd 5319 number < conf->raid_disks) {
dfc70645
N		 5320 err = -EBUSY;
5321 goto abort;
5322 }
1da177e4 5323 p->rdev = NULL;
fbd568a3 5324 synchronize_rcu();
1da177e4
LT		 5325 if (atomic_read(&rdev->nr_pending)) {
5326 /* lost the race, try later */
5327 err = -EBUSY;
5328 p->rdev = rdev;
5329 }
5330 }
5331abort:
5332
5333 print_raid5_conf(conf);
5334 return err;
5335}
5336
5337static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
5338{
5339 raid5_conf_t *conf = mddev->private;
199050ea 5340 int err = -EEXIST;
1da177e4
LT		 5341 int disk;
5342 struct disk_info *p;
6c2fce2e
NB		 5343 int first = 0;
5344 int last = conf->raid_disks - 1;
1da177e4 5345
674806d6 5346 if (has_failed(conf))
1da177e4 5347 /* no point adding a device */
199050ea 5348 return -EINVAL;
1da177e4 5349
6c2fce2e
NB		 5350 if (rdev->raid_disk >= 0)
5351 first = last = rdev->raid_disk;
1da177e4
LT		 5352
5353 /*
16a53ecc
N		 5354 * find the disk ... but prefer rdev->saved_raid_disk
5355 * if possible.
1da177e4 5356 */
16a53ecc 5357 if (rdev->saved_raid_disk >= 0 &&
6c2fce2e 5358 rdev->saved_raid_disk >= first &&
16a53ecc
N		 5359 conf->disks[rdev->saved_raid_disk].rdev == NULL)
5360 disk = rdev->saved_raid_disk;
5361 else
6c2fce2e
NB		 5362 disk = first;
5363 for ( ; disk <= last ; disk++)
1da177e4 5364 if ((p=conf->disks + disk)->rdev == NULL) {
b2d444d7 5365 clear_bit(In_sync, &rdev->flags);
1da177e4 5366 rdev->raid_disk = disk;
199050ea 5367 err = 0;
72626685
N		 5368 if (rdev->saved_raid_disk != disk)
5369 conf->fullsync = 1;
d6065f7b 5370 rcu_assign_pointer(p->rdev, rdev);
1da177e4
LT		 5371 break;
5372 }
5373 print_raid5_conf(conf);
199050ea 5374 return err;
1da177e4
LT		 5375}
5376
5377static int raid5_resize(mddev_t *mddev, sector_t sectors)
5378{
5379 /* no resync is happening, and there is enough space
5380 * on all devices, so we can resize.
5381 * We need to make sure resync covers any new space.
5382 * If the array is shrinking we should possibly wait until
5383 * any io in the removed space completes, but it hardly seems
5384 * worth it.
5385 */
9d8f0363 5386 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
1f403624
DW		 5387 md_set_array_sectors(mddev, raid5_size(mddev, sectors,
5388 mddev->raid_disks));
b522adcd
DW		 5389 if (mddev->array_sectors >
5390 raid5_size(mddev, sectors, mddev->raid_disks))
5391 return -EINVAL;
f233ea5c 5392 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 5393 revalidate_disk(mddev->gendisk);
58c0fed4
AN		 5394 if (sectors > mddev->dev_sectors && mddev->recovery_cp == MaxSector) {
5395 mddev->recovery_cp = mddev->dev_sectors;
1da177e4
LT		 5396 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5397 }
58c0fed4 5398 mddev->dev_sectors = sectors;
4b5c7ae8 5399 mddev->resync_max_sectors = sectors;
1da177e4
LT		 5400 return 0;
5401}
5402
01ee22b4
N		 5403static int check_stripe_cache(mddev_t *mddev)
5404{
5405 /* Can only proceed if there are plenty of stripe_heads.
5406 * We need a minimum of one full stripe,, and for sensible progress
5407 * it is best to have about 4 times that.
5408 * If we require 4 times, then the default 256 4K stripe_heads will
5409 * allow for chunk sizes up to 256K, which is probably OK.
5410 * If the chunk size is greater, user-space should request more
5411 * stripe_heads first.
5412 */
5413 raid5_conf_t *conf = mddev->private;
5414 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
5415 > conf->max_nr_stripes ||
5416 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
5417 > conf->max_nr_stripes) {
0c55e022
N		 5418 printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n",
5419 mdname(mddev),
01ee22b4
N		 5420 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
5421 / STRIPE_SIZE)*4);
5422 return 0;
5423 }
5424 return 1;
5425}
5426
50ac168a 5427static int check_reshape(mddev_t *mddev)
29269553 5428{
070ec55d 5429 raid5_conf_t *conf = mddev->private;
29269553 5430
88ce4930
N		 5431 if (mddev->delta_disks == 0 &&
5432 mddev->new_layout == mddev->layout &&
664e7c41 5433 mddev->new_chunk_sectors == mddev->chunk_sectors)
50ac168a 5434 return 0; /* nothing to do */
dba034ee
N		 5435 if (mddev->bitmap)
5436 /* Cannot grow a bitmap yet */
5437 return -EBUSY;
674806d6 5438 if (has_failed(conf))
ec32a2bd
N		 5439 return -EINVAL;
5440 if (mddev->delta_disks < 0) {
5441 /* We might be able to shrink, but the devices must
5442 * be made bigger first.
5443 * For raid6, 4 is the minimum size.
5444 * Otherwise 2 is the minimum
5445 */
5446 int min = 2;
5447 if (mddev->level == 6)
5448 min = 4;
5449 if (mddev->raid_disks + mddev->delta_disks < min)
5450 return -EINVAL;
5451 }
29269553 5452
01ee22b4 5453 if (!check_stripe_cache(mddev))
29269553 5454 return -ENOSPC;
29269553 5455
ec32a2bd 5456 return resize_stripes(conf, conf->raid_disks + mddev->delta_disks);
63c70c4f
N		 5457}
5458
5459static int raid5_start_reshape(mddev_t *mddev)
5460{
070ec55d 5461 raid5_conf_t *conf = mddev->private;
63c70c4f 5462 mdk_rdev_t *rdev;
63c70c4f 5463 int spares = 0;
c04be0aa 5464 unsigned long flags;
63c70c4f 5465
f416885e 5466 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
63c70c4f
N		 5467 return -EBUSY;
5468
01ee22b4
N		 5469 if (!check_stripe_cache(mddev))
5470 return -ENOSPC;
5471
159ec1fc 5472 list_for_each_entry(rdev, &mddev->disks, same_set)
469518a3
N		 5473 if (!test_bit(In_sync, &rdev->flags)
5474 && !test_bit(Faulty, &rdev->flags))
29269553 5475 spares++;
63c70c4f 5476
f416885e 5477 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
29269553
N		 5478 /* Not enough devices even to make a degraded array
5479 * of that size
5480 */
5481 return -EINVAL;
5482
ec32a2bd
N		 5483 /* Refuse to reduce size of the array. Any reductions in
5484 * array size must be through explicit setting of array_size
5485 * attribute.
5486 */
5487 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
5488 < mddev->array_sectors) {
0c55e022 5489 printk(KERN_ERR "md/raid:%s: array size must be reduced "
ec32a2bd
N		 5490 "before number of disks\n", mdname(mddev));
5491 return -EINVAL;
5492 }
5493
f6705578 5494 atomic_set(&conf->reshape_stripes, 0);
29269553
N		 5495 spin_lock_irq(&conf->device_lock);
5496 conf->previous_raid_disks = conf->raid_disks;
63c70c4f 5497 conf->raid_disks += mddev->delta_disks;
09c9e5fa
AN		 5498 conf->prev_chunk_sectors = conf->chunk_sectors;
5499 conf->chunk_sectors = mddev->new_chunk_sectors;
88ce4930
N		 5500 conf->prev_algo = conf->algorithm;
5501 conf->algorithm = mddev->new_layout;
fef9c61f
N		 5502 if (mddev->delta_disks < 0)
5503 conf->reshape_progress = raid5_size(mddev, 0, 0);
5504 else
5505 conf->reshape_progress = 0;
5506 conf->reshape_safe = conf->reshape_progress;
86b42c71 5507 conf->generation++;
29269553
N		 5508 spin_unlock_irq(&conf->device_lock);
5509
5510 /* Add some new drives, as many as will fit.
5511 * We know there are enough to make the newly sized array work.
3424bf6a
N		 5512 * Don't add devices if we are reducing the number of
5513 * devices in the array. This is because it is not possible
5514 * to correctly record the "partially reconstructed" state of
5515 * such devices during the reshape and confusion could result.
29269553 5516 */
87a8dec9
N		 5517 if (mddev->delta_disks >= 0) {
5518 int added_devices = 0;
5519 list_for_each_entry(rdev, &mddev->disks, same_set)
5520 if (rdev->raid_disk < 0 &&
5521 !test_bit(Faulty, &rdev->flags)) {
5522 if (raid5_add_disk(mddev, rdev) == 0) {
5523 char nm[20];
5524 if (rdev->raid_disk
5525 >= conf->previous_raid_disks) {
5526 set_bit(In_sync, &rdev->flags);
5527 added_devices++;
5528 } else
5529 rdev->recovery_offset = 0;
5530 sprintf(nm, "rd%d", rdev->raid_disk);
5531 if (sysfs_create_link(&mddev->kobj,
5532 &rdev->kobj, nm))
5533 /* Failure here is OK */;
50da0840 5534 }
87a8dec9
N		 5535 } else if (rdev->raid_disk >= conf->previous_raid_disks
5536 && !test_bit(Faulty, &rdev->flags)) {
5537 /* This is a spare that was manually added */
5538 set_bit(In_sync, &rdev->flags);
5539 added_devices++;
5540 }
29269553 5541
87a8dec9
N		 5542 /* When a reshape changes the number of devices,
5543 * ->degraded is measured against the larger of the
5544 * pre and post number of devices.
5545 */
ec32a2bd 5546 spin_lock_irqsave(&conf->device_lock, flags);
9eb07c25 5547 mddev->degraded += (conf->raid_disks - conf->previous_raid_disks)
ec32a2bd
N		 5548 - added_devices;
5549 spin_unlock_irqrestore(&conf->device_lock, flags);
5550 }
63c70c4f 5551 mddev->raid_disks = conf->raid_disks;
e516402c 5552 mddev->reshape_position = conf->reshape_progress;
850b2b42 5553 set_bit(MD_CHANGE_DEVS, &mddev->flags);
f6705578 5554
29269553
N		 5555 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5556 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5557 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5558 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5559 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 5560 "reshape");
29269553
N		 5561 if (!mddev->sync_thread) {
5562 mddev->recovery = 0;
5563 spin_lock_irq(&conf->device_lock);
5564 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
fef9c61f 5565 conf->reshape_progress = MaxSector;
29269553
N		 5566 spin_unlock_irq(&conf->device_lock);
5567 return -EAGAIN;
5568 }
c8f517c4 5569 conf->reshape_checkpoint = jiffies;
29269553
N		 5570 md_wakeup_thread(mddev->sync_thread);
5571 md_new_event(mddev);
5572 return 0;
5573}
29269553 5574
ec32a2bd
N		 5575/* This is called from the reshape thread and should make any
5576 * changes needed in 'conf'
5577 */
29269553
N		 5578static void end_reshape(raid5_conf_t *conf)
5579{
29269553 5580
f6705578 5581 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
f6705578 5582
f6705578 5583 spin_lock_irq(&conf->device_lock);
cea9c228 5584 conf->previous_raid_disks = conf->raid_disks;
fef9c61f 5585 conf->reshape_progress = MaxSector;
f6705578 5586 spin_unlock_irq(&conf->device_lock);
b0f9ec04 5587 wake_up(&conf->wait_for_overlap);
16a53ecc
N		 5588
5589 /* read-ahead size must cover two whole stripes, which is
5590 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
5591 */
4a5add49 5592 if (conf->mddev->queue) {
cea9c228 5593 int data_disks = conf->raid_disks - conf->max_degraded;
09c9e5fa 5594 int stripe = data_disks * ((conf->chunk_sectors << 9)
cea9c228 5595 / PAGE_SIZE);
16a53ecc
N		 5596 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
5597 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
5598 }
29269553 5599 }
29269553
N		 5600}
5601
ec32a2bd
N		 5602/* This is called from the raid5d thread with mddev_lock held.
5603 * It makes config changes to the device.
5604 */
cea9c228
N		 5605static void raid5_finish_reshape(mddev_t *mddev)
5606{
070ec55d 5607 raid5_conf_t *conf = mddev->private;
cea9c228
N		 5608
5609 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5610
ec32a2bd
N		 5611 if (mddev->delta_disks > 0) {
5612 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
5613 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 5614 revalidate_disk(mddev->gendisk);
ec32a2bd
N		 5615 } else {
5616 int d;
ec32a2bd
N		 5617 mddev->degraded = conf->raid_disks;
5618 for (d = 0; d < conf->raid_disks ; d++)
5619 if (conf->disks[d].rdev &&
5620 test_bit(In_sync,
5621 &conf->disks[d].rdev->flags))
5622 mddev->degraded--;
5623 for (d = conf->raid_disks ;
5624 d < conf->raid_disks - mddev->delta_disks;
1a67dde0
N		 5625 d++) {
5626 mdk_rdev_t *rdev = conf->disks[d].rdev;
5627 if (rdev && raid5_remove_disk(mddev, d) == 0) {
5628 char nm[20];
5629 sprintf(nm, "rd%d", rdev->raid_disk);
5630 sysfs_remove_link(&mddev->kobj, nm);
5631 rdev->raid_disk = -1;
5632 }
5633 }
cea9c228 5634 }
88ce4930 5635 mddev->layout = conf->algorithm;
09c9e5fa 5636 mddev->chunk_sectors = conf->chunk_sectors;
ec32a2bd
N		 5637 mddev->reshape_position = MaxSector;
5638 mddev->delta_disks = 0;
cea9c228
N		 5639 }
5640}
5641
72626685
N		 5642static void raid5_quiesce(mddev_t *mddev, int state)
5643{
070ec55d 5644 raid5_conf_t *conf = mddev->private;
72626685
N		 5645
5646 switch(state) {
e464eafd
N		 5647 case 2: /* resume for a suspend */
5648 wake_up(&conf->wait_for_overlap);
5649 break;
5650
72626685
N		 5651 case 1: /* stop all writes */
5652 spin_lock_irq(&conf->device_lock);
64bd660b
N		 5653 /* '2' tells resync/reshape to pause so that all
5654 * active stripes can drain
5655 */
5656 conf->quiesce = 2;
72626685 5657 wait_event_lock_irq(conf->wait_for_stripe,
46031f9a
RBJ		 5658 atomic_read(&conf->active_stripes) == 0 &&
5659 atomic_read(&conf->active_aligned_reads) == 0,
72626685 5660 conf->device_lock, /* nothing */);
64bd660b 5661 conf->quiesce = 1;
72626685 5662 spin_unlock_irq(&conf->device_lock);
64bd660b
N		 5663 /* allow reshape to continue */
5664 wake_up(&conf->wait_for_overlap);
72626685
N		 5665 break;
5666
5667 case 0: /* re-enable writes */
5668 spin_lock_irq(&conf->device_lock);
5669 conf->quiesce = 0;
5670 wake_up(&conf->wait_for_stripe);
e464eafd 5671 wake_up(&conf->wait_for_overlap);
72626685
N		 5672 spin_unlock_irq(&conf->device_lock);
5673 break;
5674 }
72626685 5675}
b15c2e57 5676
d562b0c4 5677
f1b29bca 5678static void *raid45_takeover_raid0(mddev_t *mddev, int level)
54071b38 5679{
f1b29bca 5680 struct raid0_private_data *raid0_priv = mddev->private;
d76c8420 5681 sector_t sectors;
54071b38 5682
f1b29bca
DW		 5683 /* for raid0 takeover only one zone is supported */
5684 if (raid0_priv->nr_strip_zones > 1) {
0c55e022
N		 5685 printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n",
5686 mdname(mddev));
f1b29bca
DW		 5687 return ERR_PTR(-EINVAL);
5688 }
5689
3b71bd93
N		 5690 sectors = raid0_priv->strip_zone[0].zone_end;
5691 sector_div(sectors, raid0_priv->strip_zone[0].nb_dev);
5692 mddev->dev_sectors = sectors;
f1b29bca 5693 mddev->new_level = level;
54071b38
TM		 5694 mddev->new_layout = ALGORITHM_PARITY_N;
5695 mddev->new_chunk_sectors = mddev->chunk_sectors;
5696 mddev->raid_disks += 1;
5697 mddev->delta_disks = 1;
5698 /* make sure it will be not marked as dirty */
5699 mddev->recovery_cp = MaxSector;
5700
5701 return setup_conf(mddev);
5702}
5703
5704
d562b0c4
N		 5705static void *raid5_takeover_raid1(mddev_t *mddev)
5706{
5707 int chunksect;
5708
5709 if (mddev->raid_disks != 2 ||
5710 mddev->degraded > 1)
5711 return ERR_PTR(-EINVAL);
5712
5713 /* Should check if there are write-behind devices? */
5714
5715 chunksect = 64*2; /* 64K by default */
5716
5717 /* The array must be an exact multiple of chunksize */
5718 while (chunksect && (mddev->array_sectors & (chunksect-1)))
5719 chunksect >>= 1;
5720
5721 if ((chunksect<<9) < STRIPE_SIZE)
5722 /* array size does not allow a suitable chunk size */
5723 return ERR_PTR(-EINVAL);
5724
5725 mddev->new_level = 5;
5726 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
664e7c41 5727 mddev->new_chunk_sectors = chunksect;
d562b0c4
N		 5728
5729 return setup_conf(mddev);
5730}
5731
fc9739c6
N		 5732static void *raid5_takeover_raid6(mddev_t *mddev)
5733{
5734 int new_layout;
5735
5736 switch (mddev->layout) {
5737 case ALGORITHM_LEFT_ASYMMETRIC_6:
5738 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
5739 break;
5740 case ALGORITHM_RIGHT_ASYMMETRIC_6:
5741 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
5742 break;
5743 case ALGORITHM_LEFT_SYMMETRIC_6:
5744 new_layout = ALGORITHM_LEFT_SYMMETRIC;
5745 break;
5746 case ALGORITHM_RIGHT_SYMMETRIC_6:
5747 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
5748 break;
5749 case ALGORITHM_PARITY_0_6:
5750 new_layout = ALGORITHM_PARITY_0;
5751 break;
5752 case ALGORITHM_PARITY_N:
5753 new_layout = ALGORITHM_PARITY_N;
5754 break;
5755 default:
5756 return ERR_PTR(-EINVAL);
5757 }
5758 mddev->new_level = 5;
5759 mddev->new_layout = new_layout;
5760 mddev->delta_disks = -1;
5761 mddev->raid_disks -= 1;
5762 return setup_conf(mddev);
5763}
5764
d562b0c4 5765
50ac168a 5766static int raid5_check_reshape(mddev_t *mddev)
b3546035 5767{
88ce4930
N		 5768 /* For a 2-drive array, the layout and chunk size can be changed
5769 * immediately as not restriping is needed.
5770 * For larger arrays we record the new value - after validation
5771 * to be used by a reshape pass.
b3546035 5772 */
070ec55d 5773 raid5_conf_t *conf = mddev->private;
597a711b 5774 int new_chunk = mddev->new_chunk_sectors;
b3546035 5775
597a711b 5776 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
b3546035
N		 5777 return -EINVAL;
5778 if (new_chunk > 0) {
0ba459d2 5779 if (!is_power_of_2(new_chunk))
b3546035 5780 return -EINVAL;
597a711b 5781 if (new_chunk < (PAGE_SIZE>>9))
b3546035 5782 return -EINVAL;
597a711b 5783 if (mddev->array_sectors & (new_chunk-1))
b3546035
N		 5784 /* not factor of array size */
5785 return -EINVAL;
5786 }
5787
5788 /* They look valid */
5789
88ce4930 5790 if (mddev->raid_disks == 2) {
597a711b
N		 5791 /* can make the change immediately */
5792 if (mddev->new_layout >= 0) {
5793 conf->algorithm = mddev->new_layout;
5794 mddev->layout = mddev->new_layout;
88ce4930
N		 5795 }
5796 if (new_chunk > 0) {
597a711b
N		 5797 conf->chunk_sectors = new_chunk ;
5798 mddev->chunk_sectors = new_chunk;
88ce4930
N		 5799 }
5800 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5801 md_wakeup_thread(mddev->thread);
b3546035 5802 }
50ac168a 5803 return check_reshape(mddev);
88ce4930
N		 5804}
5805
50ac168a 5806static int raid6_check_reshape(mddev_t *mddev)
88ce4930 5807{
597a711b 5808 int new_chunk = mddev->new_chunk_sectors;
50ac168a 5809
597a711b 5810 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
88ce4930 5811 return -EINVAL;
b3546035 5812 if (new_chunk > 0) {
0ba459d2 5813 if (!is_power_of_2(new_chunk))
88ce4930 5814 return -EINVAL;
597a711b 5815 if (new_chunk < (PAGE_SIZE >> 9))
88ce4930 5816 return -EINVAL;
597a711b 5817 if (mddev->array_sectors & (new_chunk-1))
88ce4930
N		 5818 /* not factor of array size */
5819 return -EINVAL;
b3546035 5820 }
88ce4930
N		 5821
5822 /* They look valid */
50ac168a 5823 return check_reshape(mddev);
b3546035
N		 5824}
5825
d562b0c4
N		 5826static void *raid5_takeover(mddev_t *mddev)
5827{
5828 /* raid5 can take over:
f1b29bca 5829 * raid0 - if there is only one strip zone - make it a raid4 layout
d562b0c4
N		 5830 * raid1 - if there are two drives. We need to know the chunk size
5831 * raid4 - trivial - just use a raid4 layout.
5832 * raid6 - Providing it is a *_6 layout
d562b0c4 5833 */
f1b29bca
DW		 5834 if (mddev->level == 0)
5835 return raid45_takeover_raid0(mddev, 5);
d562b0c4
N		 5836 if (mddev->level == 1)
5837 return raid5_takeover_raid1(mddev);
e9d4758f
N		 5838 if (mddev->level == 4) {
5839 mddev->new_layout = ALGORITHM_PARITY_N;
5840 mddev->new_level = 5;
5841 return setup_conf(mddev);
5842 }
fc9739c6
N		 5843 if (mddev->level == 6)
5844 return raid5_takeover_raid6(mddev);
d562b0c4
N		 5845
5846 return ERR_PTR(-EINVAL);
5847}
5848
a78d38a1
N		 5849static void *raid4_takeover(mddev_t *mddev)
5850{
f1b29bca
DW		 5851 /* raid4 can take over:
5852 * raid0 - if there is only one strip zone
5853 * raid5 - if layout is right
a78d38a1 5854 */
f1b29bca
DW		 5855 if (mddev->level == 0)
5856 return raid45_takeover_raid0(mddev, 4);
a78d38a1
N		 5857 if (mddev->level == 5 &&
5858 mddev->layout == ALGORITHM_PARITY_N) {
5859 mddev->new_layout = 0;
5860 mddev->new_level = 4;
5861 return setup_conf(mddev);
5862 }
5863 return ERR_PTR(-EINVAL);
5864}
d562b0c4 5865
245f46c2
N		 5866static struct mdk_personality raid5_personality;
5867
5868static void *raid6_takeover(mddev_t *mddev)
5869{
5870 /* Currently can only take over a raid5. We map the
5871 * personality to an equivalent raid6 personality
5872 * with the Q block at the end.
5873 */
5874 int new_layout;
5875
5876 if (mddev->pers != &raid5_personality)
5877 return ERR_PTR(-EINVAL);
5878 if (mddev->degraded > 1)
5879 return ERR_PTR(-EINVAL);
5880 if (mddev->raid_disks > 253)
5881 return ERR_PTR(-EINVAL);
5882 if (mddev->raid_disks < 3)
5883 return ERR_PTR(-EINVAL);
5884
5885 switch (mddev->layout) {
5886 case ALGORITHM_LEFT_ASYMMETRIC:
5887 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
5888 break;
5889 case ALGORITHM_RIGHT_ASYMMETRIC:
5890 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
5891 break;
5892 case ALGORITHM_LEFT_SYMMETRIC:
5893 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
5894 break;
5895 case ALGORITHM_RIGHT_SYMMETRIC:
5896 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
5897 break;
5898 case ALGORITHM_PARITY_0:
5899 new_layout = ALGORITHM_PARITY_0_6;
5900 break;
5901 case ALGORITHM_PARITY_N:
5902 new_layout = ALGORITHM_PARITY_N;
5903 break;
5904 default:
5905 return ERR_PTR(-EINVAL);
5906 }
5907 mddev->new_level = 6;
5908 mddev->new_layout = new_layout;
5909 mddev->delta_disks = 1;
5910 mddev->raid_disks += 1;
5911 return setup_conf(mddev);
5912}
5913
5914
16a53ecc
N		 5915static struct mdk_personality raid6_personality =
5916{
5917 .name = "raid6",
5918 .level = 6,
5919 .owner = THIS_MODULE,
5920 .make_request = make_request,
5921 .run = run,
5922 .stop = stop,
5923 .status = status,
5924 .error_handler = error,
5925 .hot_add_disk = raid5_add_disk,
5926 .hot_remove_disk= raid5_remove_disk,
5927 .spare_active = raid5_spare_active,
5928 .sync_request = sync_request,
5929 .resize = raid5_resize,
80c3a6ce 5930 .size = raid5_size,
50ac168a 5931 .check_reshape = raid6_check_reshape,
f416885e 5932 .start_reshape = raid5_start_reshape,
cea9c228 5933 .finish_reshape = raid5_finish_reshape,
16a53ecc 5934 .quiesce = raid5_quiesce,
245f46c2 5935 .takeover = raid6_takeover,
16a53ecc 5936};
2604b703 5937static struct mdk_personality raid5_personality =
1da177e4
LT		 5938{
5939 .name = "raid5",
2604b703 5940 .level = 5,
1da177e4
LT		 5941 .owner = THIS_MODULE,
5942 .make_request = make_request,
5943 .run = run,
5944 .stop = stop,
5945 .status = status,
5946 .error_handler = error,
5947 .hot_add_disk = raid5_add_disk,
5948 .hot_remove_disk= raid5_remove_disk,
5949 .spare_active = raid5_spare_active,
5950 .sync_request = sync_request,
5951 .resize = raid5_resize,
80c3a6ce 5952 .size = raid5_size,
63c70c4f
N		 5953 .check_reshape = raid5_check_reshape,
5954 .start_reshape = raid5_start_reshape,
cea9c228 5955 .finish_reshape = raid5_finish_reshape,
72626685 5956 .quiesce = raid5_quiesce,
d562b0c4 5957 .takeover = raid5_takeover,
1da177e4
LT		 5958};
5959
2604b703 5960static struct mdk_personality raid4_personality =
1da177e4 5961{
2604b703
N		 5962 .name = "raid4",
5963 .level = 4,
5964 .owner = THIS_MODULE,
5965 .make_request = make_request,
5966 .run = run,
5967 .stop = stop,
5968 .status = status,
5969 .error_handler = error,
5970 .hot_add_disk = raid5_add_disk,
5971 .hot_remove_disk= raid5_remove_disk,
5972 .spare_active = raid5_spare_active,
5973 .sync_request = sync_request,
5974 .resize = raid5_resize,
80c3a6ce 5975 .size = raid5_size,
3d37890b
N		 5976 .check_reshape = raid5_check_reshape,
5977 .start_reshape = raid5_start_reshape,
cea9c228 5978 .finish_reshape = raid5_finish_reshape,
2604b703 5979 .quiesce = raid5_quiesce,
a78d38a1 5980 .takeover = raid4_takeover,
2604b703
N		 5981};
5982
5983static int __init raid5_init(void)
5984{
16a53ecc 5985 register_md_personality(&raid6_personality);
2604b703
N		 5986 register_md_personality(&raid5_personality);
5987 register_md_personality(&raid4_personality);
5988 return 0;
1da177e4
LT		 5989}
5990
2604b703 5991static void raid5_exit(void)
1da177e4 5992{
16a53ecc 5993 unregister_md_personality(&raid6_personality);
2604b703
N		 5994 unregister_md_personality(&raid5_personality);
5995 unregister_md_personality(&raid4_personality);
1da177e4
LT		 5996}
5997
5998module_init(raid5_init);
5999module_exit(raid5_exit);
6000MODULE_LICENSE("GPL");
0efb9e61 6001MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
1da177e4 6002MODULE_ALIAS("md-personality-4"); /* RAID5 */
d9d166c2
N		 6003MODULE_ALIAS("md-raid5");
6004MODULE_ALIAS("md-raid4");
2604b703
N		 6005MODULE_ALIAS("md-level-5");
6006MODULE_ALIAS("md-level-4");
16a53ecc
N		 6007MODULE_ALIAS("md-personality-8"); /* RAID6 */
6008MODULE_ALIAS("md-raid6");
6009MODULE_ALIAS("md-level-6");
6010
6011/* This used to be two separate modules, they were: */
6012MODULE_ALIAS("raid5");
6013MODULE_ALIAS("raid6");
kernel source for telekom puls (alcatel/mediatek)