projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md/raid1: move rdev->corrected_errors counting
[linux-2.6-block.git] / drivers / md / raid5.c
CommitLineData
1da177e4
LT		 1/*
2 * raid5.c : Multiple Devices driver for Linux
3 * Copyright (C) 1996, 1997 Ingo Molnar, Miguel de Icaza, Gadi Oxman
4 * Copyright (C) 1999, 2000 Ingo Molnar
16a53ecc 5 * Copyright (C) 2002, 2003 H. Peter Anvin
1da177e4 6 *
16a53ecc
N		 7 * RAID-4/5/6 management functions.
8 * Thanks to Penguin Computing for making the RAID-6 development possible
9 * by donating a test server!
1da177e4
LT		 10 *
11 * This program is free software; you can redistribute it and/or modify
12 * it under the terms of the GNU General Public License as published by
13 * the Free Software Foundation; either version 2, or (at your option)
14 * any later version.
15 *
16 * You should have received a copy of the GNU General Public License
17 * (for example /usr/src/linux/COPYING); if not, write to the Free
18 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
19 */
20
ae3c20cc
N		 21/*
22 * BITMAP UNPLUGGING:
23 *
24 * The sequencing for updating the bitmap reliably is a little
25 * subtle (and I got it wrong the first time) so it deserves some
26 * explanation.
27 *
28 * We group bitmap updates into batches. Each batch has a number.
29 * We may write out several batches at once, but that isn't very important.
7c13edc8
N		 30 * conf->seq_write is the number of the last batch successfully written.
31 * conf->seq_flush is the number of the last batch that was closed to
ae3c20cc
N		 32 * new additions.
33 * When we discover that we will need to write to any block in a stripe
34 * (in add_stripe_bio) we update the in-memory bitmap and record in sh->bm_seq
7c13edc8 35 * the number of the batch it will be in. This is seq_flush+1.
ae3c20cc
N		 36 * When we are ready to do a write, if that batch hasn't been written yet,
37 * we plug the array and queue the stripe for later.
38 * When an unplug happens, we increment bm_flush, thus closing the current
39 * batch.
40 * When we notice that bm_flush > bm_write, we write out all pending updates
41 * to the bitmap, and advance bm_write to where bm_flush was.
42 * This may occasionally write a bit out twice, but is sure never to
43 * miss any bits.
44 */
1da177e4 45
bff61975 46#include <linux/blkdev.h>
f6705578 47#include <linux/kthread.h>
f701d589 48#include <linux/raid/pq.h>
91c00924 49#include <linux/async_tx.h>
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>
8bda470e 54#include <linux/ratelimit.h>
43b2e5d8 55#include "md.h"
bff61975 56#include "raid5.h"
54071b38 57#include "raid0.h"
ef740c37 58#include "bitmap.h"
72626685 59
1da177e4
LT		 60/*
61 * Stripe cache
62 */
63
64#define NR_STRIPES 256
65#define STRIPE_SIZE PAGE_SIZE
66#define STRIPE_SHIFT (PAGE_SHIFT - 9)
67#define STRIPE_SECTORS (STRIPE_SIZE>>9)
68#define IO_THRESHOLD 1
8b3e6cdc 69#define BYPASS_THRESHOLD 1
fccddba0 70#define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
1da177e4
LT		 71#define HASH_MASK (NR_HASH - 1)
72
fccddba0 73#define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK]))
1da177e4
LT		 74
75/* bio's attached to a stripe+device for I/O are linked together in bi_sector
76 * order without overlap. There may be several bio's per stripe+device, and
77 * a bio could span several devices.
78 * When walking this list for a particular stripe+device, we must never proceed
79 * beyond a bio that extends past this device, as the next bio might no longer
80 * be valid.
81 * This macro is used to determine the 'next' bio in the list, given the sector
82 * of the current stripe+device
83 */
84#define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL)
85/*
86 * The following can be used to debug the driver
87 */
1da177e4
LT		 88#define RAID5_PARANOIA 1
89#if RAID5_PARANOIA && defined(CONFIG_SMP)
90# define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock)
91#else
92# define CHECK_DEVLOCK()
93#endif
94
45b4233c 95#ifdef DEBUG
1da177e4
LT		 96#define inline
97#define __inline__
98#endif
99
960e739d 100/*
5b99c2ff
JA		 101 * We maintain a biased count of active stripes in the bottom 16 bits of
102 * bi_phys_segments, and a count of processed stripes in the upper 16 bits
960e739d
JA		 103 */
104static inline int raid5_bi_phys_segments(struct bio *bio)
105{
5b99c2ff 106 return bio->bi_phys_segments & 0xffff;
960e739d
JA		 107}
108
109static inline int raid5_bi_hw_segments(struct bio *bio)
110{
5b99c2ff 111 return (bio->bi_phys_segments >> 16) & 0xffff;
960e739d
JA		 112}
113
114static inline int raid5_dec_bi_phys_segments(struct bio *bio)
115{
116 --bio->bi_phys_segments;
117 return raid5_bi_phys_segments(bio);
118}
119
120static inline int raid5_dec_bi_hw_segments(struct bio *bio)
121{
122 unsigned short val = raid5_bi_hw_segments(bio);
123
124 --val;
5b99c2ff 125 bio->bi_phys_segments = (val << 16) | raid5_bi_phys_segments(bio);
960e739d
JA		 126 return val;
127}
128
129static inline void raid5_set_bi_hw_segments(struct bio *bio, unsigned int cnt)
130{
9b2dc8b6 131 bio->bi_phys_segments = raid5_bi_phys_segments(bio) | (cnt << 16);
960e739d
JA		 132}
133
d0dabf7e
N		 134/* Find first data disk in a raid6 stripe */
135static inline int raid6_d0(struct stripe_head *sh)
136{
67cc2b81
N		 137 if (sh->ddf_layout)
138 /* ddf always start from first device */
139 return 0;
140 /* md starts just after Q block */
d0dabf7e
N		 141 if (sh->qd_idx == sh->disks - 1)
142 return 0;
143 else
144 return sh->qd_idx + 1;
145}
16a53ecc
N		 146static inline int raid6_next_disk(int disk, int raid_disks)
147{
148 disk++;
149 return (disk < raid_disks) ? disk : 0;
150}
a4456856 151
d0dabf7e
N		 152/* When walking through the disks in a raid5, starting at raid6_d0,
153 * We need to map each disk to a 'slot', where the data disks are slot
154 * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk
155 * is raid_disks-1. This help does that mapping.
156 */
67cc2b81
N		 157static int raid6_idx_to_slot(int idx, struct stripe_head *sh,
158 int *count, int syndrome_disks)
d0dabf7e 159{
6629542e 160 int slot = *count;
67cc2b81 161
e4424fee 162 if (sh->ddf_layout)
6629542e 163 (*count)++;
d0dabf7e 164 if (idx == sh->pd_idx)
67cc2b81 165 return syndrome_disks;
d0dabf7e 166 if (idx == sh->qd_idx)
67cc2b81 167 return syndrome_disks + 1;
e4424fee 168 if (!sh->ddf_layout)
6629542e 169 (*count)++;
d0dabf7e
N		 170 return slot;
171}
172
a4456856
DW		 173static void return_io(struct bio *return_bi)
174{
175 struct bio *bi = return_bi;
176 while (bi) {
a4456856
DW		 177
178 return_bi = bi->bi_next;
179 bi->bi_next = NULL;
180 bi->bi_size = 0;
0e13fe23 181 bio_endio(bi, 0);
a4456856
DW		 182 bi = return_bi;
183 }
184}
185
1da177e4
LT		 186static void print_raid5_conf (raid5_conf_t *conf);
187
600aa109
DW		 188static int stripe_operations_active(struct stripe_head *sh)
189{
190 return sh->check_state || sh->reconstruct_state ||
191 test_bit(STRIPE_BIOFILL_RUN, &sh->state) ||
192 test_bit(STRIPE_COMPUTE_RUN, &sh->state);
193}
194
858119e1 195static void __release_stripe(raid5_conf_t *conf, struct stripe_head *sh)
1da177e4
LT		 196{
197 if (atomic_dec_and_test(&sh->count)) {
78bafebd
ES		 198 BUG_ON(!list_empty(&sh->lru));
199 BUG_ON(atomic_read(&conf->active_stripes)==0);
1da177e4 200 if (test_bit(STRIPE_HANDLE, &sh->state)) {
482c0834 201 if (test_bit(STRIPE_DELAYED, &sh->state))
1da177e4 202 list_add_tail(&sh->lru, &conf->delayed_list);
482c0834
N		 203 else if (test_bit(STRIPE_BIT_DELAY, &sh->state) &&
204 sh->bm_seq - conf->seq_write > 0)
72626685 205 list_add_tail(&sh->lru, &conf->bitmap_list);
482c0834 206 else {
72626685 207 clear_bit(STRIPE_BIT_DELAY, &sh->state);
1da177e4 208 list_add_tail(&sh->lru, &conf->handle_list);
72626685 209 }
1da177e4
LT		 210 md_wakeup_thread(conf->mddev->thread);
211 } else {
600aa109 212 BUG_ON(stripe_operations_active(sh));
1da177e4
LT		 213 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
214 atomic_dec(&conf->preread_active_stripes);
215 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD)
216 md_wakeup_thread(conf->mddev->thread);
217 }
1da177e4 218 atomic_dec(&conf->active_stripes);
ccfcc3c1
N		 219 if (!test_bit(STRIPE_EXPANDING, &sh->state)) {
220 list_add_tail(&sh->lru, &conf->inactive_list);
1da177e4 221 wake_up(&conf->wait_for_stripe);
46031f9a
RBJ		 222 if (conf->retry_read_aligned)
223 md_wakeup_thread(conf->mddev->thread);
ccfcc3c1 224 }
1da177e4
LT		 225 }
226 }
227}
d0dabf7e 228
1da177e4
LT		 229static void release_stripe(struct stripe_head *sh)
230{
231 raid5_conf_t *conf = sh->raid_conf;
232 unsigned long flags;
16a53ecc 233
1da177e4
LT		 234 spin_lock_irqsave(&conf->device_lock, flags);
235 __release_stripe(conf, sh);
236 spin_unlock_irqrestore(&conf->device_lock, flags);
237}
238
fccddba0 239static inline void remove_hash(struct stripe_head *sh)
1da177e4 240{
45b4233c
DW		 241 pr_debug("remove_hash(), stripe %llu\n",
242 (unsigned long long)sh->sector);
1da177e4 243
fccddba0 244 hlist_del_init(&sh->hash);
1da177e4
LT		 245}
246
16a53ecc 247static inline void insert_hash(raid5_conf_t *conf, struct stripe_head *sh)
1da177e4 248{
fccddba0 249 struct hlist_head *hp = stripe_hash(conf, sh->sector);
1da177e4 250
45b4233c
DW		 251 pr_debug("insert_hash(), stripe %llu\n",
252 (unsigned long long)sh->sector);
1da177e4
LT		 253
254 CHECK_DEVLOCK();
fccddba0 255 hlist_add_head(&sh->hash, hp);
1da177e4
LT		 256}
257
258
259/* find an idle stripe, make sure it is unhashed, and return it. */
260static struct stripe_head *get_free_stripe(raid5_conf_t *conf)
261{
262 struct stripe_head *sh = NULL;
263 struct list_head *first;
264
265 CHECK_DEVLOCK();
266 if (list_empty(&conf->inactive_list))
267 goto out;
268 first = conf->inactive_list.next;
269 sh = list_entry(first, struct stripe_head, lru);
270 list_del_init(first);
271 remove_hash(sh);
272 atomic_inc(&conf->active_stripes);
273out:
274 return sh;
275}
276
e4e11e38 277static void shrink_buffers(struct stripe_head *sh)
1da177e4
LT		 278{
279 struct page *p;
280 int i;
e4e11e38 281 int num = sh->raid_conf->pool_size;
1da177e4 282
e4e11e38 283 for (i = 0; i < num ; i++) {
1da177e4
LT		 284 p = sh->dev[i].page;
285 if (!p)
286 continue;
287 sh->dev[i].page = NULL;
2d1f3b5d 288 put_page(p);
1da177e4
LT		 289 }
290}
291
e4e11e38 292static int grow_buffers(struct stripe_head *sh)
1da177e4
LT		 293{
294 int i;
e4e11e38 295 int num = sh->raid_conf->pool_size;
1da177e4 296
e4e11e38 297 for (i = 0; i < num; i++) {
1da177e4
LT		 298 struct page *page;
299
300 if (!(page = alloc_page(GFP_KERNEL))) {
301 return 1;
302 }
303 sh->dev[i].page = page;
304 }
305 return 0;
306}
307
784052ec 308static void raid5_build_block(struct stripe_head *sh, int i, int previous);
911d4ee8
N		 309static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous,
310 struct stripe_head *sh);
1da177e4 311
b5663ba4 312static void init_stripe(struct stripe_head *sh, sector_t sector, int previous)
1da177e4
LT		 313{
314 raid5_conf_t *conf = sh->raid_conf;
7ecaa1e6 315 int i;
1da177e4 316
78bafebd
ES		 317 BUG_ON(atomic_read(&sh->count) != 0);
318 BUG_ON(test_bit(STRIPE_HANDLE, &sh->state));
600aa109 319 BUG_ON(stripe_operations_active(sh));
d84e0f10 320
1da177e4 321 CHECK_DEVLOCK();
45b4233c 322 pr_debug("init_stripe called, stripe %llu\n",
1da177e4
LT		 323 (unsigned long long)sh->sector);
324
325 remove_hash(sh);
16a53ecc 326
86b42c71 327 sh->generation = conf->generation - previous;
b5663ba4 328 sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks;
1da177e4 329 sh->sector = sector;
911d4ee8 330 stripe_set_idx(sector, conf, previous, sh);
1da177e4
LT		 331 sh->state = 0;
332
7ecaa1e6
N		 333
334 for (i = sh->disks; i--; ) {
1da177e4
LT		 335 struct r5dev *dev = &sh->dev[i];
336
d84e0f10 337 if (dev->toread || dev->read || dev->towrite || dev->written ||
1da177e4 338 test_bit(R5_LOCKED, &dev->flags)) {
d84e0f10 339 printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n",
1da177e4 340 (unsigned long long)sh->sector, i, dev->toread,
d84e0f10 341 dev->read, dev->towrite, dev->written,
1da177e4
LT		 342 test_bit(R5_LOCKED, &dev->flags));
343 BUG();
344 }
345 dev->flags = 0;
784052ec 346 raid5_build_block(sh, i, previous);
1da177e4
LT		 347 }
348 insert_hash(conf, sh);
349}
350
86b42c71
N		 351static struct stripe_head *__find_stripe(raid5_conf_t *conf, sector_t sector,
352 short generation)
1da177e4
LT		 353{
354 struct stripe_head *sh;
fccddba0 355 struct hlist_node *hn;
1da177e4
LT		 356
357 CHECK_DEVLOCK();
45b4233c 358 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
fccddba0 359 hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash)
86b42c71 360 if (sh->sector == sector && sh->generation == generation)
1da177e4 361 return sh;
45b4233c 362 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
1da177e4
LT		 363 return NULL;
364}
365
674806d6
N		 366/*
367 * Need to check if array has failed when deciding whether to:
368 * - start an array
369 * - remove non-faulty devices
370 * - add a spare
371 * - allow a reshape
372 * This determination is simple when no reshape is happening.
373 * However if there is a reshape, we need to carefully check
374 * both the before and after sections.
375 * This is because some failed devices may only affect one
376 * of the two sections, and some non-in_sync devices may
377 * be insync in the section most affected by failed devices.
378 */
379static int has_failed(raid5_conf_t *conf)
380{
381 int degraded;
382 int i;
383 if (conf->mddev->reshape_position == MaxSector)
384 return conf->mddev->degraded > conf->max_degraded;
385
386 rcu_read_lock();
387 degraded = 0;
388 for (i = 0; i < conf->previous_raid_disks; i++) {
389 mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev);
390 if (!rdev || test_bit(Faulty, &rdev->flags))
391 degraded++;
392 else if (test_bit(In_sync, &rdev->flags))
393 ;
394 else
395 /* not in-sync or faulty.
396 * If the reshape increases the number of devices,
397 * this is being recovered by the reshape, so
398 * this 'previous' section is not in_sync.
399 * If the number of devices is being reduced however,
400 * the device can only be part of the array if
401 * we are reverting a reshape, so this section will
402 * be in-sync.
403 */
404 if (conf->raid_disks >= conf->previous_raid_disks)
405 degraded++;
406 }
407 rcu_read_unlock();
408 if (degraded > conf->max_degraded)
409 return 1;
410 rcu_read_lock();
411 degraded = 0;
412 for (i = 0; i < conf->raid_disks; i++) {
413 mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev);
414 if (!rdev || test_bit(Faulty, &rdev->flags))
415 degraded++;
416 else if (test_bit(In_sync, &rdev->flags))
417 ;
418 else
419 /* not in-sync or faulty.
420 * If reshape increases the number of devices, this
421 * section has already been recovered, else it
422 * almost certainly hasn't.
423 */
424 if (conf->raid_disks <= conf->previous_raid_disks)
425 degraded++;
426 }
427 rcu_read_unlock();
428 if (degraded > conf->max_degraded)
429 return 1;
430 return 0;
431}
432
b5663ba4
N		 433static struct stripe_head *
434get_active_stripe(raid5_conf_t *conf, sector_t sector,
a8c906ca 435 int previous, int noblock, int noquiesce)
1da177e4
LT		 436{
437 struct stripe_head *sh;
438
45b4233c 439 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector);
1da177e4
LT		 440
441 spin_lock_irq(&conf->device_lock);
442
443 do {
72626685 444 wait_event_lock_irq(conf->wait_for_stripe,
a8c906ca 445 conf->quiesce == 0 || noquiesce,
72626685 446 conf->device_lock, /* nothing */);
86b42c71 447 sh = __find_stripe(conf, sector, conf->generation - previous);
1da177e4
LT		 448 if (!sh) {
449 if (!conf->inactive_blocked)
450 sh = get_free_stripe(conf);
451 if (noblock && sh == NULL)
452 break;
453 if (!sh) {
454 conf->inactive_blocked = 1;
455 wait_event_lock_irq(conf->wait_for_stripe,
456 !list_empty(&conf->inactive_list) &&
5036805b
N		 457 (atomic_read(&conf->active_stripes)
458 < (conf->max_nr_stripes *3/4)
1da177e4
LT		 459 || !conf->inactive_blocked),
460 conf->device_lock,
7c13edc8 461 );
1da177e4
LT		 462 conf->inactive_blocked = 0;
463 } else
b5663ba4 464 init_stripe(sh, sector, previous);
1da177e4
LT		 465 } else {
466 if (atomic_read(&sh->count)) {
ab69ae12
N		 467 BUG_ON(!list_empty(&sh->lru)
468 && !test_bit(STRIPE_EXPANDING, &sh->state));
1da177e4
LT		 469 } else {
470 if (!test_bit(STRIPE_HANDLE, &sh->state))
471 atomic_inc(&conf->active_stripes);
ff4e8d9a
N		 472 if (list_empty(&sh->lru) &&
473 !test_bit(STRIPE_EXPANDING, &sh->state))
16a53ecc
N		 474 BUG();
475 list_del_init(&sh->lru);
1da177e4
LT		 476 }
477 }
478 } while (sh == NULL);
479
480 if (sh)
481 atomic_inc(&sh->count);
482
483 spin_unlock_irq(&conf->device_lock);
484 return sh;
485}
486
6712ecf8
N		 487static void
488raid5_end_read_request(struct bio *bi, int error);
489static void
490raid5_end_write_request(struct bio *bi, int error);
91c00924 491
c4e5ac0a 492static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
91c00924
DW		 493{
494 raid5_conf_t *conf = sh->raid_conf;
495 int i, disks = sh->disks;
496
497 might_sleep();
498
499 for (i = disks; i--; ) {
500 int rw;
501 struct bio *bi;
502 mdk_rdev_t *rdev;
e9c7469b
TH		 503 if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) {
504 if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags))
505 rw = WRITE_FUA;
506 else
507 rw = WRITE;
508 } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags))
91c00924
DW		 509 rw = READ;
510 else
511 continue;
512
513 bi = &sh->dev[i].req;
514
515 bi->bi_rw = rw;
b062962e 516 if (rw & WRITE)
91c00924
DW		 517 bi->bi_end_io = raid5_end_write_request;
518 else
519 bi->bi_end_io = raid5_end_read_request;
520
521 rcu_read_lock();
522 rdev = rcu_dereference(conf->disks[i].rdev);
523 if (rdev && test_bit(Faulty, &rdev->flags))
524 rdev = NULL;
525 if (rdev)
526 atomic_inc(&rdev->nr_pending);
527 rcu_read_unlock();
528
529 if (rdev) {
c4e5ac0a 530 if (s->syncing || s->expanding || s->expanded)
91c00924
DW		 531 md_sync_acct(rdev->bdev, STRIPE_SECTORS);
532
2b7497f0
DW		 533 set_bit(STRIPE_IO_STARTED, &sh->state);
534
91c00924
DW		 535 bi->bi_bdev = rdev->bdev;
536 pr_debug("%s: for %llu schedule op %ld on disc %d\n",
e46b272b 537 __func__, (unsigned long long)sh->sector,
91c00924
DW		 538 bi->bi_rw, i);
539 atomic_inc(&sh->count);
540 bi->bi_sector = sh->sector + rdev->data_offset;
541 bi->bi_flags = 1 << BIO_UPTODATE;
542 bi->bi_vcnt = 1;
543 bi->bi_max_vecs = 1;
544 bi->bi_idx = 0;
545 bi->bi_io_vec = &sh->dev[i].vec;
546 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
547 bi->bi_io_vec[0].bv_offset = 0;
548 bi->bi_size = STRIPE_SIZE;
549 bi->bi_next = NULL;
b062962e 550 if ((rw & WRITE) &&
91c00924
DW		 551 test_bit(R5_ReWrite, &sh->dev[i].flags))
552 atomic_add(STRIPE_SECTORS,
553 &rdev->corrected_errors);
554 generic_make_request(bi);
555 } else {
b062962e 556 if (rw & WRITE)
91c00924
DW		 557 set_bit(STRIPE_DEGRADED, &sh->state);
558 pr_debug("skip op %ld on disc %d for sector %llu\n",
559 bi->bi_rw, i, (unsigned long long)sh->sector);
560 clear_bit(R5_LOCKED, &sh->dev[i].flags);
561 set_bit(STRIPE_HANDLE, &sh->state);
562 }
563 }
564}
565
566static struct dma_async_tx_descriptor *
567async_copy_data(int frombio, struct bio *bio, struct page *page,
568 sector_t sector, struct dma_async_tx_descriptor *tx)
569{
570 struct bio_vec *bvl;
571 struct page *bio_page;
572 int i;
573 int page_offset;
a08abd8c 574 struct async_submit_ctl submit;
0403e382 575 enum async_tx_flags flags = 0;
91c00924
DW		 576
577 if (bio->bi_sector >= sector)
578 page_offset = (signed)(bio->bi_sector - sector) * 512;
579 else
580 page_offset = (signed)(sector - bio->bi_sector) * -512;
a08abd8c 581
0403e382
DW		 582 if (frombio)
583 flags |= ASYNC_TX_FENCE;
584 init_async_submit(&submit, flags, tx, NULL, NULL, NULL);
585
91c00924 586 bio_for_each_segment(bvl, bio, i) {
fcde9075 587 int len = bvl->bv_len;
91c00924
DW		 588 int clen;
589 int b_offset = 0;
590
591 if (page_offset < 0) {
592 b_offset = -page_offset;
593 page_offset += b_offset;
594 len -= b_offset;
595 }
596
597 if (len > 0 && page_offset + len > STRIPE_SIZE)
598 clen = STRIPE_SIZE - page_offset;
599 else
600 clen = len;
601
602 if (clen > 0) {
fcde9075
NK		 603 b_offset += bvl->bv_offset;
604 bio_page = bvl->bv_page;
91c00924
DW		 605 if (frombio)
606 tx = async_memcpy(page, bio_page, page_offset,
a08abd8c 607 b_offset, clen, &submit);
91c00924
DW		 608 else
609 tx = async_memcpy(bio_page, page, b_offset,
a08abd8c 610 page_offset, clen, &submit);
91c00924 611 }
a08abd8c
DW		 612 /* chain the operations */
613 submit.depend_tx = tx;
614
91c00924
DW		 615 if (clen < len) /* hit end of page */
616 break;
617 page_offset += len;
618 }
619
620 return tx;
621}
622
623static void ops_complete_biofill(void *stripe_head_ref)
624{
625 struct stripe_head *sh = stripe_head_ref;
626 struct bio *return_bi = NULL;
627 raid5_conf_t *conf = sh->raid_conf;
e4d84909 628 int i;
91c00924 629
e46b272b 630 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 631 (unsigned long long)sh->sector);
632
633 /* clear completed biofills */
83de75cc 634 spin_lock_irq(&conf->device_lock);
91c00924
DW		 635 for (i = sh->disks; i--; ) {
636 struct r5dev *dev = &sh->dev[i];
91c00924
DW		 637
638 /* acknowledge completion of a biofill operation */
e4d84909
DW		 639 /* and check if we need to reply to a read request,
640 * new R5_Wantfill requests are held off until
83de75cc 641 * !STRIPE_BIOFILL_RUN
e4d84909
DW		 642 */
643 if (test_and_clear_bit(R5_Wantfill, &dev->flags)) {
91c00924 644 struct bio *rbi, *rbi2;
91c00924 645
91c00924
DW		 646 BUG_ON(!dev->read);
647 rbi = dev->read;
648 dev->read = NULL;
649 while (rbi && rbi->bi_sector <
650 dev->sector + STRIPE_SECTORS) {
651 rbi2 = r5_next_bio(rbi, dev->sector);
960e739d 652 if (!raid5_dec_bi_phys_segments(rbi)) {
91c00924
DW		 653 rbi->bi_next = return_bi;
654 return_bi = rbi;
655 }
91c00924
DW		 656 rbi = rbi2;
657 }
658 }
659 }
83de75cc
DW		 660 spin_unlock_irq(&conf->device_lock);
661 clear_bit(STRIPE_BIOFILL_RUN, &sh->state);
91c00924
DW		 662
663 return_io(return_bi);
664
e4d84909 665 set_bit(STRIPE_HANDLE, &sh->state);
91c00924
DW		 666 release_stripe(sh);
667}
668
669static void ops_run_biofill(struct stripe_head *sh)
670{
671 struct dma_async_tx_descriptor *tx = NULL;
672 raid5_conf_t *conf = sh->raid_conf;
a08abd8c 673 struct async_submit_ctl submit;
91c00924
DW		 674 int i;
675
e46b272b 676 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 677 (unsigned long long)sh->sector);
678
679 for (i = sh->disks; i--; ) {
680 struct r5dev *dev = &sh->dev[i];
681 if (test_bit(R5_Wantfill, &dev->flags)) {
682 struct bio *rbi;
683 spin_lock_irq(&conf->device_lock);
684 dev->read = rbi = dev->toread;
685 dev->toread = NULL;
686 spin_unlock_irq(&conf->device_lock);
687 while (rbi && rbi->bi_sector <
688 dev->sector + STRIPE_SECTORS) {
689 tx = async_copy_data(0, rbi, dev->page,
690 dev->sector, tx);
691 rbi = r5_next_bio(rbi, dev->sector);
692 }
693 }
694 }
695
696 atomic_inc(&sh->count);
a08abd8c
DW		 697 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL);
698 async_trigger_callback(&submit);
91c00924
DW		 699}
700
4e7d2c0a 701static void mark_target_uptodate(struct stripe_head *sh, int target)
91c00924 702{
4e7d2c0a 703 struct r5dev *tgt;
91c00924 704
4e7d2c0a
DW		 705 if (target < 0)
706 return;
91c00924 707
4e7d2c0a 708 tgt = &sh->dev[target];
91c00924
DW		 709 set_bit(R5_UPTODATE, &tgt->flags);
710 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
711 clear_bit(R5_Wantcompute, &tgt->flags);
4e7d2c0a
DW		 712}
713
ac6b53b6 714static void ops_complete_compute(void *stripe_head_ref)
91c00924
DW		 715{
716 struct stripe_head *sh = stripe_head_ref;
91c00924 717
e46b272b 718 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 719 (unsigned long long)sh->sector);
720
ac6b53b6 721 /* mark the computed target(s) as uptodate */
4e7d2c0a 722 mark_target_uptodate(sh, sh->ops.target);
ac6b53b6 723 mark_target_uptodate(sh, sh->ops.target2);
4e7d2c0a 724
ecc65c9b
DW		 725 clear_bit(STRIPE_COMPUTE_RUN, &sh->state);
726 if (sh->check_state == check_state_compute_run)
727 sh->check_state = check_state_compute_result;
91c00924
DW		 728 set_bit(STRIPE_HANDLE, &sh->state);
729 release_stripe(sh);
730}
731
d6f38f31
DW		 732/* return a pointer to the address conversion region of the scribble buffer */
733static addr_conv_t *to_addr_conv(struct stripe_head *sh,
734 struct raid5_percpu *percpu)
735{
736 return percpu->scribble + sizeof(struct page *) * (sh->disks + 2);
737}
738
739static struct dma_async_tx_descriptor *
740ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 741{
91c00924 742 int disks = sh->disks;
d6f38f31 743 struct page **xor_srcs = percpu->scribble;
91c00924
DW		 744 int target = sh->ops.target;
745 struct r5dev *tgt = &sh->dev[target];
746 struct page *xor_dest = tgt->page;
747 int count = 0;
748 struct dma_async_tx_descriptor *tx;
a08abd8c 749 struct async_submit_ctl submit;
91c00924
DW		 750 int i;
751
752 pr_debug("%s: stripe %llu block: %d\n",
e46b272b 753 __func__, (unsigned long long)sh->sector, target);
91c00924
DW		 754 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
755
756 for (i = disks; i--; )
757 if (i != target)
758 xor_srcs[count++] = sh->dev[i].page;
759
760 atomic_inc(&sh->count);
761
0403e382 762 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL,
ac6b53b6 763 ops_complete_compute, sh, to_addr_conv(sh, percpu));
91c00924 764 if (unlikely(count == 1))
a08abd8c 765 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
91c00924 766 else
a08abd8c 767 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924 768
91c00924
DW		 769 return tx;
770}
771
ac6b53b6
DW		 772/* set_syndrome_sources - populate source buffers for gen_syndrome
773 * @srcs - (struct page *) array of size sh->disks
774 * @sh - stripe_head to parse
775 *
776 * Populates srcs in proper layout order for the stripe and returns the
777 * 'count' of sources to be used in a call to async_gen_syndrome. The P
778 * destination buffer is recorded in srcs[count] and the Q destination
779 * is recorded in srcs[count+1]].
780 */
781static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh)
782{
783 int disks = sh->disks;
784 int syndrome_disks = sh->ddf_layout ? disks : (disks - 2);
785 int d0_idx = raid6_d0(sh);
786 int count;
787 int i;
788
789 for (i = 0; i < disks; i++)
5dd33c9a 790 srcs[i] = NULL;
ac6b53b6
DW		 791
792 count = 0;
793 i = d0_idx;
794 do {
795 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
796
797 srcs[slot] = sh->dev[i].page;
798 i = raid6_next_disk(i, disks);
799 } while (i != d0_idx);
ac6b53b6 800
e4424fee 801 return syndrome_disks;
ac6b53b6
DW		 802}
803
804static struct dma_async_tx_descriptor *
805ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu)
806{
807 int disks = sh->disks;
808 struct page **blocks = percpu->scribble;
809 int target;
810 int qd_idx = sh->qd_idx;
811 struct dma_async_tx_descriptor *tx;
812 struct async_submit_ctl submit;
813 struct r5dev *tgt;
814 struct page *dest;
815 int i;
816 int count;
817
818 if (sh->ops.target < 0)
819 target = sh->ops.target2;
820 else if (sh->ops.target2 < 0)
821 target = sh->ops.target;
91c00924 822 else
ac6b53b6
DW		 823 /* we should only have one valid target */
824 BUG();
825 BUG_ON(target < 0);
826 pr_debug("%s: stripe %llu block: %d\n",
827 __func__, (unsigned long long)sh->sector, target);
828
829 tgt = &sh->dev[target];
830 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
831 dest = tgt->page;
832
833 atomic_inc(&sh->count);
834
835 if (target == qd_idx) {
836 count = set_syndrome_sources(blocks, sh);
837 blocks[count] = NULL; /* regenerating p is not necessary */
838 BUG_ON(blocks[count+1] != dest); /* q should already be set */
0403e382
DW		 839 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
840 ops_complete_compute, sh,
ac6b53b6
DW		 841 to_addr_conv(sh, percpu));
842 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
843 } else {
844 /* Compute any data- or p-drive using XOR */
845 count = 0;
846 for (i = disks; i-- ; ) {
847 if (i == target || i == qd_idx)
848 continue;
849 blocks[count++] = sh->dev[i].page;
850 }
851
0403e382
DW		 852 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
853 NULL, ops_complete_compute, sh,
ac6b53b6
DW		 854 to_addr_conv(sh, percpu));
855 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit);
856 }
91c00924 857
91c00924
DW		 858 return tx;
859}
860
ac6b53b6
DW		 861static struct dma_async_tx_descriptor *
862ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu)
863{
864 int i, count, disks = sh->disks;
865 int syndrome_disks = sh->ddf_layout ? disks : disks-2;
866 int d0_idx = raid6_d0(sh);
867 int faila = -1, failb = -1;
868 int target = sh->ops.target;
869 int target2 = sh->ops.target2;
870 struct r5dev *tgt = &sh->dev[target];
871 struct r5dev *tgt2 = &sh->dev[target2];
872 struct dma_async_tx_descriptor *tx;
873 struct page **blocks = percpu->scribble;
874 struct async_submit_ctl submit;
875
876 pr_debug("%s: stripe %llu block1: %d block2: %d\n",
877 __func__, (unsigned long long)sh->sector, target, target2);
878 BUG_ON(target < 0 || target2 < 0);
879 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
880 BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags));
881
6c910a78 882 /* we need to open-code set_syndrome_sources to handle the
ac6b53b6
DW		 883 * slot number conversion for 'faila' and 'failb'
884 */
885 for (i = 0; i < disks ; i++)
5dd33c9a 886 blocks[i] = NULL;
ac6b53b6
DW		 887 count = 0;
888 i = d0_idx;
889 do {
890 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
891
892 blocks[slot] = sh->dev[i].page;
893
894 if (i == target)
895 faila = slot;
896 if (i == target2)
897 failb = slot;
898 i = raid6_next_disk(i, disks);
899 } while (i != d0_idx);
ac6b53b6
DW		 900
901 BUG_ON(faila == failb);
902 if (failb < faila)
903 swap(faila, failb);
904 pr_debug("%s: stripe: %llu faila: %d failb: %d\n",
905 __func__, (unsigned long long)sh->sector, faila, failb);
906
907 atomic_inc(&sh->count);
908
909 if (failb == syndrome_disks+1) {
910 /* Q disk is one of the missing disks */
911 if (faila == syndrome_disks) {
912 /* Missing P+Q, just recompute */
0403e382
DW		 913 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
914 ops_complete_compute, sh,
915 to_addr_conv(sh, percpu));
e4424fee 916 return async_gen_syndrome(blocks, 0, syndrome_disks+2,
ac6b53b6
DW		 917 STRIPE_SIZE, &submit);
918 } else {
919 struct page *dest;
920 int data_target;
921 int qd_idx = sh->qd_idx;
922
923 /* Missing D+Q: recompute D from P, then recompute Q */
924 if (target == qd_idx)
925 data_target = target2;
926 else
927 data_target = target;
928
929 count = 0;
930 for (i = disks; i-- ; ) {
931 if (i == data_target || i == qd_idx)
932 continue;
933 blocks[count++] = sh->dev[i].page;
934 }
935 dest = sh->dev[data_target].page;
0403e382
DW		 936 init_async_submit(&submit,
937 ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
938 NULL, NULL, NULL,
939 to_addr_conv(sh, percpu));
ac6b53b6
DW		 940 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE,
941 &submit);
942
943 count = set_syndrome_sources(blocks, sh);
0403e382
DW		 944 init_async_submit(&submit, ASYNC_TX_FENCE, tx,
945 ops_complete_compute, sh,
946 to_addr_conv(sh, percpu));
ac6b53b6
DW		 947 return async_gen_syndrome(blocks, 0, count+2,
948 STRIPE_SIZE, &submit);
949 }
ac6b53b6 950 } else {
6c910a78
DW		 951 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
952 ops_complete_compute, sh,
953 to_addr_conv(sh, percpu));
954 if (failb == syndrome_disks) {
955 /* We're missing D+P. */
956 return async_raid6_datap_recov(syndrome_disks+2,
957 STRIPE_SIZE, faila,
958 blocks, &submit);
959 } else {
960 /* We're missing D+D. */
961 return async_raid6_2data_recov(syndrome_disks+2,
962 STRIPE_SIZE, faila, failb,
963 blocks, &submit);
964 }
ac6b53b6
DW		 965 }
966}
967
968
91c00924
DW		 969static void ops_complete_prexor(void *stripe_head_ref)
970{
971 struct stripe_head *sh = stripe_head_ref;
972
e46b272b 973 pr_debug("%s: stripe %llu\n", __func__,
91c00924 974 (unsigned long long)sh->sector);
91c00924
DW		 975}
976
977static struct dma_async_tx_descriptor *
d6f38f31
DW		 978ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu,
979 struct dma_async_tx_descriptor *tx)
91c00924 980{
91c00924 981 int disks = sh->disks;
d6f38f31 982 struct page **xor_srcs = percpu->scribble;
91c00924 983 int count = 0, pd_idx = sh->pd_idx, i;
a08abd8c 984 struct async_submit_ctl submit;
91c00924
DW		 985
986 /* existing parity data subtracted */
987 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
988
e46b272b 989 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 990 (unsigned long long)sh->sector);
991
992 for (i = disks; i--; ) {
993 struct r5dev *dev = &sh->dev[i];
994 /* Only process blocks that are known to be uptodate */
d8ee0728 995 if (test_bit(R5_Wantdrain, &dev->flags))
91c00924
DW		 996 xor_srcs[count++] = dev->page;
997 }
998
0403e382 999 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
d6f38f31 1000 ops_complete_prexor, sh, to_addr_conv(sh, percpu));
a08abd8c 1001 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924
DW		 1002
1003 return tx;
1004}
1005
1006static struct dma_async_tx_descriptor *
d8ee0728 1007ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
91c00924
DW		 1008{
1009 int disks = sh->disks;
d8ee0728 1010 int i;
91c00924 1011
e46b272b 1012 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1013 (unsigned long long)sh->sector);
1014
1015 for (i = disks; i--; ) {
1016 struct r5dev *dev = &sh->dev[i];
1017 struct bio *chosen;
91c00924 1018
d8ee0728 1019 if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) {
91c00924
DW		 1020 struct bio *wbi;
1021
cbe47ec5 1022 spin_lock_irq(&sh->raid_conf->device_lock);
91c00924
DW		 1023 chosen = dev->towrite;
1024 dev->towrite = NULL;
1025 BUG_ON(dev->written);
1026 wbi = dev->written = chosen;
cbe47ec5 1027 spin_unlock_irq(&sh->raid_conf->device_lock);
91c00924
DW		 1028
1029 while (wbi && wbi->bi_sector <
1030 dev->sector + STRIPE_SECTORS) {
e9c7469b
TH		 1031 if (wbi->bi_rw & REQ_FUA)
1032 set_bit(R5_WantFUA, &dev->flags);
91c00924
DW		 1033 tx = async_copy_data(1, wbi, dev->page,
1034 dev->sector, tx);
1035 wbi = r5_next_bio(wbi, dev->sector);
1036 }
1037 }
1038 }
1039
1040 return tx;
1041}
1042
ac6b53b6 1043static void ops_complete_reconstruct(void *stripe_head_ref)
91c00924
DW		 1044{
1045 struct stripe_head *sh = stripe_head_ref;
ac6b53b6
DW		 1046 int disks = sh->disks;
1047 int pd_idx = sh->pd_idx;
1048 int qd_idx = sh->qd_idx;
1049 int i;
e9c7469b 1050 bool fua = false;
91c00924 1051
e46b272b 1052 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1053 (unsigned long long)sh->sector);
1054
e9c7469b
TH		 1055 for (i = disks; i--; )
1056 fua |= test_bit(R5_WantFUA, &sh->dev[i].flags);
1057
91c00924
DW		 1058 for (i = disks; i--; ) {
1059 struct r5dev *dev = &sh->dev[i];
ac6b53b6 1060
e9c7469b 1061 if (dev->written || i == pd_idx || i == qd_idx) {
91c00924 1062 set_bit(R5_UPTODATE, &dev->flags);
e9c7469b
TH		 1063 if (fua)
1064 set_bit(R5_WantFUA, &dev->flags);
1065 }
91c00924
DW		 1066 }
1067
d8ee0728
DW		 1068 if (sh->reconstruct_state == reconstruct_state_drain_run)
1069 sh->reconstruct_state = reconstruct_state_drain_result;
1070 else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run)
1071 sh->reconstruct_state = reconstruct_state_prexor_drain_result;
1072 else {
1073 BUG_ON(sh->reconstruct_state != reconstruct_state_run);
1074 sh->reconstruct_state = reconstruct_state_result;
1075 }
91c00924
DW		 1076
1077 set_bit(STRIPE_HANDLE, &sh->state);
1078 release_stripe(sh);
1079}
1080
1081static void
ac6b53b6
DW		 1082ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu,
1083 struct dma_async_tx_descriptor *tx)
91c00924 1084{
91c00924 1085 int disks = sh->disks;
d6f38f31 1086 struct page **xor_srcs = percpu->scribble;
a08abd8c 1087 struct async_submit_ctl submit;
91c00924
DW		 1088 int count = 0, pd_idx = sh->pd_idx, i;
1089 struct page *xor_dest;
d8ee0728 1090 int prexor = 0;
91c00924 1091 unsigned long flags;
91c00924 1092
e46b272b 1093 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1094 (unsigned long long)sh->sector);
1095
1096 /* check if prexor is active which means only process blocks
1097 * that are part of a read-modify-write (written)
1098 */
d8ee0728
DW		 1099 if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
1100 prexor = 1;
91c00924
DW		 1101 xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1102 for (i = disks; i--; ) {
1103 struct r5dev *dev = &sh->dev[i];
1104 if (dev->written)
1105 xor_srcs[count++] = dev->page;
1106 }
1107 } else {
1108 xor_dest = sh->dev[pd_idx].page;
1109 for (i = disks; i--; ) {
1110 struct r5dev *dev = &sh->dev[i];
1111 if (i != pd_idx)
1112 xor_srcs[count++] = dev->page;
1113 }
1114 }
1115
91c00924
DW		 1116 /* 1/ if we prexor'd then the dest is reused as a source
1117 * 2/ if we did not prexor then we are redoing the parity
1118 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
1119 * for the synchronous xor case
1120 */
88ba2aa5 1121 flags = ASYNC_TX_ACK |
91c00924
DW		 1122 (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
1123
1124 atomic_inc(&sh->count);
1125
ac6b53b6 1126 init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh,
d6f38f31 1127 to_addr_conv(sh, percpu));
a08abd8c
DW		 1128 if (unlikely(count == 1))
1129 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
1130 else
1131 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
91c00924
DW		 1132}
1133
ac6b53b6
DW		 1134static void
1135ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu,
1136 struct dma_async_tx_descriptor *tx)
1137{
1138 struct async_submit_ctl submit;
1139 struct page **blocks = percpu->scribble;
1140 int count;
1141
1142 pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector);
1143
1144 count = set_syndrome_sources(blocks, sh);
1145
1146 atomic_inc(&sh->count);
1147
1148 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct,
1149 sh, to_addr_conv(sh, percpu));
1150 async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
91c00924
DW		 1151}
1152
1153static void ops_complete_check(void *stripe_head_ref)
1154{
1155 struct stripe_head *sh = stripe_head_ref;
91c00924 1156
e46b272b 1157 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1158 (unsigned long long)sh->sector);
1159
ecc65c9b 1160 sh->check_state = check_state_check_result;
91c00924
DW		 1161 set_bit(STRIPE_HANDLE, &sh->state);
1162 release_stripe(sh);
1163}
1164
ac6b53b6 1165static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu)
91c00924 1166{
91c00924 1167 int disks = sh->disks;
ac6b53b6
DW		 1168 int pd_idx = sh->pd_idx;
1169 int qd_idx = sh->qd_idx;
1170 struct page *xor_dest;
d6f38f31 1171 struct page **xor_srcs = percpu->scribble;
91c00924 1172 struct dma_async_tx_descriptor *tx;
a08abd8c 1173 struct async_submit_ctl submit;
ac6b53b6
DW		 1174 int count;
1175 int i;
91c00924 1176
e46b272b 1177 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 1178 (unsigned long long)sh->sector);
1179
ac6b53b6
DW		 1180 count = 0;
1181 xor_dest = sh->dev[pd_idx].page;
1182 xor_srcs[count++] = xor_dest;
91c00924 1183 for (i = disks; i--; ) {
ac6b53b6
DW		 1184 if (i == pd_idx || i == qd_idx)
1185 continue;
1186 xor_srcs[count++] = sh->dev[i].page;
91c00924
DW		 1187 }
1188
d6f38f31
DW		 1189 init_async_submit(&submit, 0, NULL, NULL, NULL,
1190 to_addr_conv(sh, percpu));
099f53cb 1191 tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
a08abd8c 1192 &sh->ops.zero_sum_result, &submit);
91c00924 1193
91c00924 1194 atomic_inc(&sh->count);
a08abd8c
DW		 1195 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL);
1196 tx = async_trigger_callback(&submit);
91c00924
DW		 1197}
1198
ac6b53b6
DW		 1199static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp)
1200{
1201 struct page **srcs = percpu->scribble;
1202 struct async_submit_ctl submit;
1203 int count;
1204
1205 pr_debug("%s: stripe %llu checkp: %d\n", __func__,
1206 (unsigned long long)sh->sector, checkp);
1207
1208 count = set_syndrome_sources(srcs, sh);
1209 if (!checkp)
1210 srcs[count] = NULL;
91c00924 1211
91c00924 1212 atomic_inc(&sh->count);
ac6b53b6
DW		 1213 init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check,
1214 sh, to_addr_conv(sh, percpu));
1215 async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE,
1216 &sh->ops.zero_sum_result, percpu->spare_page, &submit);
91c00924
DW		 1217}
1218
417b8d4a 1219static void __raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
91c00924
DW		 1220{
1221 int overlap_clear = 0, i, disks = sh->disks;
1222 struct dma_async_tx_descriptor *tx = NULL;
d6f38f31 1223 raid5_conf_t *conf = sh->raid_conf;
ac6b53b6 1224 int level = conf->level;
d6f38f31
DW		 1225 struct raid5_percpu *percpu;
1226 unsigned long cpu;
91c00924 1227
d6f38f31
DW		 1228 cpu = get_cpu();
1229 percpu = per_cpu_ptr(conf->percpu, cpu);
83de75cc 1230 if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) {
91c00924
DW		 1231 ops_run_biofill(sh);
1232 overlap_clear++;
1233 }
1234
7b3a871e 1235 if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) {
ac6b53b6
DW		 1236 if (level < 6)
1237 tx = ops_run_compute5(sh, percpu);
1238 else {
1239 if (sh->ops.target2 < 0 || sh->ops.target < 0)
1240 tx = ops_run_compute6_1(sh, percpu);
1241 else
1242 tx = ops_run_compute6_2(sh, percpu);
1243 }
1244 /* terminate the chain if reconstruct is not set to be run */
1245 if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request))
7b3a871e
DW		 1246 async_tx_ack(tx);
1247 }
91c00924 1248
600aa109 1249 if (test_bit(STRIPE_OP_PREXOR, &ops_request))
d6f38f31 1250 tx = ops_run_prexor(sh, percpu, tx);
91c00924 1251
600aa109 1252 if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) {
d8ee0728 1253 tx = ops_run_biodrain(sh, tx);
91c00924
DW		 1254 overlap_clear++;
1255 }
1256
ac6b53b6
DW		 1257 if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) {
1258 if (level < 6)
1259 ops_run_reconstruct5(sh, percpu, tx);
1260 else
1261 ops_run_reconstruct6(sh, percpu, tx);
1262 }
91c00924 1263
ac6b53b6
DW		 1264 if (test_bit(STRIPE_OP_CHECK, &ops_request)) {
1265 if (sh->check_state == check_state_run)
1266 ops_run_check_p(sh, percpu);
1267 else if (sh->check_state == check_state_run_q)
1268 ops_run_check_pq(sh, percpu, 0);
1269 else if (sh->check_state == check_state_run_pq)
1270 ops_run_check_pq(sh, percpu, 1);
1271 else
1272 BUG();
1273 }
91c00924 1274
91c00924
DW		 1275 if (overlap_clear)
1276 for (i = disks; i--; ) {
1277 struct r5dev *dev = &sh->dev[i];
1278 if (test_and_clear_bit(R5_Overlap, &dev->flags))
1279 wake_up(&sh->raid_conf->wait_for_overlap);
1280 }
d6f38f31 1281 put_cpu();
91c00924
DW		 1282}
1283
417b8d4a
DW		 1284#ifdef CONFIG_MULTICORE_RAID456
1285static void async_run_ops(void *param, async_cookie_t cookie)
1286{
1287 struct stripe_head *sh = param;
1288 unsigned long ops_request = sh->ops.request;
1289
1290 clear_bit_unlock(STRIPE_OPS_REQ_PENDING, &sh->state);
1291 wake_up(&sh->ops.wait_for_ops);
1292
1293 __raid_run_ops(sh, ops_request);
1294 release_stripe(sh);
1295}
1296
1297static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
1298{
1299 /* since handle_stripe can be called outside of raid5d context
1300 * we need to ensure sh->ops.request is de-staged before another
1301 * request arrives
1302 */
1303 wait_event(sh->ops.wait_for_ops,
1304 !test_and_set_bit_lock(STRIPE_OPS_REQ_PENDING, &sh->state));
1305 sh->ops.request = ops_request;
1306
1307 atomic_inc(&sh->count);
1308 async_schedule(async_run_ops, sh);
1309}
1310#else
1311#define raid_run_ops __raid_run_ops
1312#endif
1313
3f294f4f 1314static int grow_one_stripe(raid5_conf_t *conf)
1da177e4
LT		 1315{
1316 struct stripe_head *sh;
6ce32846 1317 sh = kmem_cache_zalloc(conf->slab_cache, GFP_KERNEL);
3f294f4f
N		 1318 if (!sh)
1319 return 0;
6ce32846 1320
3f294f4f 1321 sh->raid_conf = conf;
417b8d4a
DW		 1322 #ifdef CONFIG_MULTICORE_RAID456
1323 init_waitqueue_head(&sh->ops.wait_for_ops);
1324 #endif
3f294f4f 1325
e4e11e38
N		 1326 if (grow_buffers(sh)) {
1327 shrink_buffers(sh);
3f294f4f
N		 1328 kmem_cache_free(conf->slab_cache, sh);
1329 return 0;
1330 }
1331 /* we just created an active stripe so... */
1332 atomic_set(&sh->count, 1);
1333 atomic_inc(&conf->active_stripes);
1334 INIT_LIST_HEAD(&sh->lru);
1335 release_stripe(sh);
1336 return 1;
1337}
1338
1339static int grow_stripes(raid5_conf_t *conf, int num)
1340{
e18b890b 1341 struct kmem_cache *sc;
5e5e3e78 1342 int devs = max(conf->raid_disks, conf->previous_raid_disks);
1da177e4 1343
f4be6b43
N		 1344 if (conf->mddev->gendisk)
1345 sprintf(conf->cache_name[0],
1346 "raid%d-%s", conf->level, mdname(conf->mddev));
1347 else
1348 sprintf(conf->cache_name[0],
1349 "raid%d-%p", conf->level, conf->mddev);
1350 sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]);
1351
ad01c9e3
N		 1352 conf->active_name = 0;
1353 sc = kmem_cache_create(conf->cache_name[conf->active_name],
1da177e4 1354 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
20c2df83 1355 0, 0, NULL);
1da177e4
LT		 1356 if (!sc)
1357 return 1;
1358 conf->slab_cache = sc;
ad01c9e3 1359 conf->pool_size = devs;
16a53ecc 1360 while (num--)
3f294f4f 1361 if (!grow_one_stripe(conf))
1da177e4 1362 return 1;
1da177e4
LT		 1363 return 0;
1364}
29269553 1365
d6f38f31
DW		 1366/**
1367 * scribble_len - return the required size of the scribble region
1368 * @num - total number of disks in the array
1369 *
1370 * The size must be enough to contain:
1371 * 1/ a struct page pointer for each device in the array +2
1372 * 2/ room to convert each entry in (1) to its corresponding dma
1373 * (dma_map_page()) or page (page_address()) address.
1374 *
1375 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
1376 * calculate over all devices (not just the data blocks), using zeros in place
1377 * of the P and Q blocks.
1378 */
1379static size_t scribble_len(int num)
1380{
1381 size_t len;
1382
1383 len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2);
1384
1385 return len;
1386}
1387
ad01c9e3
N		 1388static int resize_stripes(raid5_conf_t *conf, int newsize)
1389{
1390 /* Make all the stripes able to hold 'newsize' devices.
1391 * New slots in each stripe get 'page' set to a new page.
1392 *
1393 * This happens in stages:
1394 * 1/ create a new kmem_cache and allocate the required number of
1395 * stripe_heads.
1396 * 2/ gather all the old stripe_heads and tranfer the pages across
1397 * to the new stripe_heads. This will have the side effect of
1398 * freezing the array as once all stripe_heads have been collected,
1399 * no IO will be possible. Old stripe heads are freed once their
1400 * pages have been transferred over, and the old kmem_cache is
1401 * freed when all stripes are done.
1402 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
1403 * we simple return a failre status - no need to clean anything up.
1404 * 4/ allocate new pages for the new slots in the new stripe_heads.
1405 * If this fails, we don't bother trying the shrink the
1406 * stripe_heads down again, we just leave them as they are.
1407 * As each stripe_head is processed the new one is released into
1408 * active service.
1409 *
1410 * Once step2 is started, we cannot afford to wait for a write,
1411 * so we use GFP_NOIO allocations.
1412 */
1413 struct stripe_head *osh, *nsh;
1414 LIST_HEAD(newstripes);
1415 struct disk_info *ndisks;
d6f38f31 1416 unsigned long cpu;
b5470dc5 1417 int err;
e18b890b 1418 struct kmem_cache *sc;
ad01c9e3
N		 1419 int i;
1420
1421 if (newsize <= conf->pool_size)
1422 return 0; /* never bother to shrink */
1423
b5470dc5
DW		 1424 err = md_allow_write(conf->mddev);
1425 if (err)
1426 return err;
2a2275d6 1427
ad01c9e3
N		 1428 /* Step 1 */
1429 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
1430 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
20c2df83 1431 0, 0, NULL);
ad01c9e3
N		 1432 if (!sc)
1433 return -ENOMEM;
1434
1435 for (i = conf->max_nr_stripes; i; i--) {
6ce32846 1436 nsh = kmem_cache_zalloc(sc, GFP_KERNEL);
ad01c9e3
N		 1437 if (!nsh)
1438 break;
1439
ad01c9e3 1440 nsh->raid_conf = conf;
417b8d4a
DW		 1441 #ifdef CONFIG_MULTICORE_RAID456
1442 init_waitqueue_head(&nsh->ops.wait_for_ops);
1443 #endif
ad01c9e3
N		 1444
1445 list_add(&nsh->lru, &newstripes);
1446 }
1447 if (i) {
1448 /* didn't get enough, give up */
1449 while (!list_empty(&newstripes)) {
1450 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1451 list_del(&nsh->lru);
1452 kmem_cache_free(sc, nsh);
1453 }
1454 kmem_cache_destroy(sc);
1455 return -ENOMEM;
1456 }
1457 /* Step 2 - Must use GFP_NOIO now.
1458 * OK, we have enough stripes, start collecting inactive
1459 * stripes and copying them over
1460 */
1461 list_for_each_entry(nsh, &newstripes, lru) {
1462 spin_lock_irq(&conf->device_lock);
1463 wait_event_lock_irq(conf->wait_for_stripe,
1464 !list_empty(&conf->inactive_list),
1465 conf->device_lock,
482c0834 1466 );
ad01c9e3
N		 1467 osh = get_free_stripe(conf);
1468 spin_unlock_irq(&conf->device_lock);
1469 atomic_set(&nsh->count, 1);
1470 for(i=0; i<conf->pool_size; i++)
1471 nsh->dev[i].page = osh->dev[i].page;
1472 for( ; i<newsize; i++)
1473 nsh->dev[i].page = NULL;
1474 kmem_cache_free(conf->slab_cache, osh);
1475 }
1476 kmem_cache_destroy(conf->slab_cache);
1477
1478 /* Step 3.
1479 * At this point, we are holding all the stripes so the array
1480 * is completely stalled, so now is a good time to resize
d6f38f31 1481 * conf->disks and the scribble region
ad01c9e3
N		 1482 */
1483 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
1484 if (ndisks) {
1485 for (i=0; i<conf->raid_disks; i++)
1486 ndisks[i] = conf->disks[i];
1487 kfree(conf->disks);
1488 conf->disks = ndisks;
1489 } else
1490 err = -ENOMEM;
1491
d6f38f31
DW		 1492 get_online_cpus();
1493 conf->scribble_len = scribble_len(newsize);
1494 for_each_present_cpu(cpu) {
1495 struct raid5_percpu *percpu;
1496 void *scribble;
1497
1498 percpu = per_cpu_ptr(conf->percpu, cpu);
1499 scribble = kmalloc(conf->scribble_len, GFP_NOIO);
1500
1501 if (scribble) {
1502 kfree(percpu->scribble);
1503 percpu->scribble = scribble;
1504 } else {
1505 err = -ENOMEM;
1506 break;
1507 }
1508 }
1509 put_online_cpus();
1510
ad01c9e3
N		 1511 /* Step 4, return new stripes to service */
1512 while(!list_empty(&newstripes)) {
1513 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1514 list_del_init(&nsh->lru);
d6f38f31 1515
ad01c9e3
N		 1516 for (i=conf->raid_disks; i < newsize; i++)
1517 if (nsh->dev[i].page == NULL) {
1518 struct page *p = alloc_page(GFP_NOIO);
1519 nsh->dev[i].page = p;
1520 if (!p)
1521 err = -ENOMEM;
1522 }
1523 release_stripe(nsh);
1524 }
1525 /* critical section pass, GFP_NOIO no longer needed */
1526
1527 conf->slab_cache = sc;
1528 conf->active_name = 1-conf->active_name;
1529 conf->pool_size = newsize;
1530 return err;
1531}
1da177e4 1532
3f294f4f 1533static int drop_one_stripe(raid5_conf_t *conf)
1da177e4
LT		 1534{
1535 struct stripe_head *sh;
1536
3f294f4f
N		 1537 spin_lock_irq(&conf->device_lock);
1538 sh = get_free_stripe(conf);
1539 spin_unlock_irq(&conf->device_lock);
1540 if (!sh)
1541 return 0;
78bafebd 1542 BUG_ON(atomic_read(&sh->count));
e4e11e38 1543 shrink_buffers(sh);
3f294f4f
N		 1544 kmem_cache_free(conf->slab_cache, sh);
1545 atomic_dec(&conf->active_stripes);
1546 return 1;
1547}
1548
1549static void shrink_stripes(raid5_conf_t *conf)
1550{
1551 while (drop_one_stripe(conf))
1552 ;
1553
29fc7e3e
N		 1554 if (conf->slab_cache)
1555 kmem_cache_destroy(conf->slab_cache);
1da177e4
LT		 1556 conf->slab_cache = NULL;
1557}
1558
6712ecf8 1559static void raid5_end_read_request(struct bio * bi, int error)
1da177e4 1560{
99c0fb5f 1561 struct stripe_head *sh = bi->bi_private;
1da177e4 1562 raid5_conf_t *conf = sh->raid_conf;
7ecaa1e6 1563 int disks = sh->disks, i;
1da177e4 1564 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
d6950432
N		 1565 char b[BDEVNAME_SIZE];
1566 mdk_rdev_t *rdev;
1da177e4 1567
1da177e4
LT		 1568
1569 for (i=0 ; i<disks; i++)
1570 if (bi == &sh->dev[i].req)
1571 break;
1572
45b4233c
DW		 1573 pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n",
1574 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
1da177e4
LT		 1575 uptodate);
1576 if (i == disks) {
1577 BUG();
6712ecf8 1578 return;
1da177e4
LT		 1579 }
1580
1581 if (uptodate) {
1da177e4 1582 set_bit(R5_UPTODATE, &sh->dev[i].flags);
4e5314b5 1583 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
d6950432 1584 rdev = conf->disks[i].rdev;
8bda470e
CD		 1585 printk_ratelimited(
1586 KERN_INFO
1587 "md/raid:%s: read error corrected"
1588 " (%lu sectors at %llu on %s)\n",
1589 mdname(conf->mddev), STRIPE_SECTORS,
1590 (unsigned long long)(sh->sector
1591 + rdev->data_offset),
1592 bdevname(rdev->bdev, b));
4e5314b5
N		 1593 clear_bit(R5_ReadError, &sh->dev[i].flags);
1594 clear_bit(R5_ReWrite, &sh->dev[i].flags);
1595 }
ba22dcbf
N		 1596 if (atomic_read(&conf->disks[i].rdev->read_errors))
1597 atomic_set(&conf->disks[i].rdev->read_errors, 0);
1da177e4 1598 } else {
d6950432 1599 const char *bdn = bdevname(conf->disks[i].rdev->bdev, b);
ba22dcbf 1600 int retry = 0;
d6950432
N		 1601 rdev = conf->disks[i].rdev;
1602
1da177e4 1603 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
d6950432 1604 atomic_inc(&rdev->read_errors);
7b0bb536 1605 if (conf->mddev->degraded >= conf->max_degraded)
8bda470e
CD		 1606 printk_ratelimited(
1607 KERN_WARNING
1608 "md/raid:%s: read error not correctable "
1609 "(sector %llu on %s).\n",
1610 mdname(conf->mddev),
1611 (unsigned long long)(sh->sector
1612 + rdev->data_offset),
1613 bdn);
ba22dcbf 1614 else if (test_bit(R5_ReWrite, &sh->dev[i].flags))
4e5314b5 1615 /* Oh, no!!! */
8bda470e
CD		 1616 printk_ratelimited(
1617 KERN_WARNING
1618 "md/raid:%s: read error NOT corrected!! "
1619 "(sector %llu on %s).\n",
1620 mdname(conf->mddev),
1621 (unsigned long long)(sh->sector
1622 + rdev->data_offset),
1623 bdn);
d6950432 1624 else if (atomic_read(&rdev->read_errors)
ba22dcbf 1625 > conf->max_nr_stripes)
14f8d26b 1626 printk(KERN_WARNING
0c55e022 1627 "md/raid:%s: Too many read errors, failing device %s.\n",
d6950432 1628 mdname(conf->mddev), bdn);
ba22dcbf
N		 1629 else
1630 retry = 1;
1631 if (retry)
1632 set_bit(R5_ReadError, &sh->dev[i].flags);
1633 else {
4e5314b5
N		 1634 clear_bit(R5_ReadError, &sh->dev[i].flags);
1635 clear_bit(R5_ReWrite, &sh->dev[i].flags);
d6950432 1636 md_error(conf->mddev, rdev);
ba22dcbf 1637 }
1da177e4
LT		 1638 }
1639 rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
1da177e4
LT		 1640 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1641 set_bit(STRIPE_HANDLE, &sh->state);
1642 release_stripe(sh);
1da177e4
LT		 1643}
1644
d710e138 1645static void raid5_end_write_request(struct bio *bi, int error)
1da177e4 1646{
99c0fb5f 1647 struct stripe_head *sh = bi->bi_private;
1da177e4 1648 raid5_conf_t *conf = sh->raid_conf;
7ecaa1e6 1649 int disks = sh->disks, i;
1da177e4
LT		 1650 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
1651
1da177e4
LT		 1652 for (i=0 ; i<disks; i++)
1653 if (bi == &sh->dev[i].req)
1654 break;
1655
45b4233c 1656 pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n",
1da177e4
LT		 1657 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
1658 uptodate);
1659 if (i == disks) {
1660 BUG();
6712ecf8 1661 return;
1da177e4
LT		 1662 }
1663
1da177e4
LT		 1664 if (!uptodate)
1665 md_error(conf->mddev, conf->disks[i].rdev);
1666
1667 rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
1668
1669 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1670 set_bit(STRIPE_HANDLE, &sh->state);
c04be0aa 1671 release_stripe(sh);
1da177e4
LT		 1672}
1673
1674
784052ec 1675static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous);
1da177e4 1676
784052ec 1677static void raid5_build_block(struct stripe_head *sh, int i, int previous)
1da177e4
LT		 1678{
1679 struct r5dev *dev = &sh->dev[i];
1680
1681 bio_init(&dev->req);
1682 dev->req.bi_io_vec = &dev->vec;
1683 dev->req.bi_vcnt++;
1684 dev->req.bi_max_vecs++;
1685 dev->vec.bv_page = dev->page;
1686 dev->vec.bv_len = STRIPE_SIZE;
1687 dev->vec.bv_offset = 0;
1688
1689 dev->req.bi_sector = sh->sector;
1690 dev->req.bi_private = sh;
1691
1692 dev->flags = 0;
784052ec 1693 dev->sector = compute_blocknr(sh, i, previous);
1da177e4
LT		 1694}
1695
1696static void error(mddev_t *mddev, mdk_rdev_t *rdev)
1697{
1698 char b[BDEVNAME_SIZE];
7b92813c 1699 raid5_conf_t *conf = mddev->private;
0c55e022 1700 pr_debug("raid456: error called\n");
1da177e4 1701
6f8d0c77
N		 1702 if (test_and_clear_bit(In_sync, &rdev->flags)) {
1703 unsigned long flags;
1704 spin_lock_irqsave(&conf->device_lock, flags);
1705 mddev->degraded++;
1706 spin_unlock_irqrestore(&conf->device_lock, flags);
1707 /*
1708 * if recovery was running, make sure it aborts.
1709 */
1710 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4 1711 }
6f8d0c77
N		 1712 set_bit(Faulty, &rdev->flags);
1713 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1714 printk(KERN_ALERT
1715 "md/raid:%s: Disk failure on %s, disabling device.\n"
1716 "md/raid:%s: Operation continuing on %d devices.\n",
1717 mdname(mddev),
1718 bdevname(rdev->bdev, b),
1719 mdname(mddev),
1720 conf->raid_disks - mddev->degraded);
16a53ecc 1721}
1da177e4
LT		 1722
1723/*
1724 * Input: a 'big' sector number,
1725 * Output: index of the data and parity disk, and the sector # in them.
1726 */
112bf897 1727static sector_t raid5_compute_sector(raid5_conf_t *conf, sector_t r_sector,
911d4ee8
N		 1728 int previous, int *dd_idx,
1729 struct stripe_head *sh)
1da177e4 1730{
6e3b96ed 1731 sector_t stripe, stripe2;
35f2a591 1732 sector_t chunk_number;
1da177e4 1733 unsigned int chunk_offset;
911d4ee8 1734 int pd_idx, qd_idx;
67cc2b81 1735 int ddf_layout = 0;
1da177e4 1736 sector_t new_sector;
e183eaed
N		 1737 int algorithm = previous ? conf->prev_algo
1738 : conf->algorithm;
09c9e5fa
AN		 1739 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
1740 : conf->chunk_sectors;
112bf897
N		 1741 int raid_disks = previous ? conf->previous_raid_disks
1742 : conf->raid_disks;
1743 int data_disks = raid_disks - conf->max_degraded;
1da177e4
LT		 1744
1745 /* First compute the information on this sector */
1746
1747 /*
1748 * Compute the chunk number and the sector offset inside the chunk
1749 */
1750 chunk_offset = sector_div(r_sector, sectors_per_chunk);
1751 chunk_number = r_sector;
1da177e4
LT		 1752
1753 /*
1754 * Compute the stripe number
1755 */
35f2a591
N		 1756 stripe = chunk_number;
1757 *dd_idx = sector_div(stripe, data_disks);
6e3b96ed 1758 stripe2 = stripe;
1da177e4
LT		 1759 /*
1760 * Select the parity disk based on the user selected algorithm.
1761 */
84789554 1762 pd_idx = qd_idx = -1;
16a53ecc
N		 1763 switch(conf->level) {
1764 case 4:
911d4ee8 1765 pd_idx = data_disks;
16a53ecc
N		 1766 break;
1767 case 5:
e183eaed 1768 switch (algorithm) {
1da177e4 1769 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 1770 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 1771 if (*dd_idx >= pd_idx)
1da177e4
LT		 1772 (*dd_idx)++;
1773 break;
1774 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 1775 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 1776 if (*dd_idx >= pd_idx)
1da177e4
LT		 1777 (*dd_idx)++;
1778 break;
1779 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 1780 pd_idx = data_disks - sector_div(stripe2, raid_disks);
911d4ee8 1781 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4
LT		 1782 break;
1783 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 1784 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8 1785 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1da177e4 1786 break;
99c0fb5f
N		 1787 case ALGORITHM_PARITY_0:
1788 pd_idx = 0;
1789 (*dd_idx)++;
1790 break;
1791 case ALGORITHM_PARITY_N:
1792 pd_idx = data_disks;
1793 break;
1da177e4 1794 default:
99c0fb5f 1795 BUG();
16a53ecc
N		 1796 }
1797 break;
1798 case 6:
1799
e183eaed 1800 switch (algorithm) {
16a53ecc 1801 case ALGORITHM_LEFT_ASYMMETRIC:
6e3b96ed 1802 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 1803 qd_idx = pd_idx + 1;
1804 if (pd_idx == raid_disks-1) {
99c0fb5f 1805 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 1806 qd_idx = 0;
1807 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 1808 (*dd_idx) += 2; /* D D P Q D */
1809 break;
1810 case ALGORITHM_RIGHT_ASYMMETRIC:
6e3b96ed 1811 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 1812 qd_idx = pd_idx + 1;
1813 if (pd_idx == raid_disks-1) {
99c0fb5f 1814 (*dd_idx)++; /* Q D D D P */
911d4ee8
N		 1815 qd_idx = 0;
1816 } else if (*dd_idx >= pd_idx)
16a53ecc
N		 1817 (*dd_idx) += 2; /* D D P Q D */
1818 break;
1819 case ALGORITHM_LEFT_SYMMETRIC:
6e3b96ed 1820 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
911d4ee8
N		 1821 qd_idx = (pd_idx + 1) % raid_disks;
1822 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc
N		 1823 break;
1824 case ALGORITHM_RIGHT_SYMMETRIC:
6e3b96ed 1825 pd_idx = sector_div(stripe2, raid_disks);
911d4ee8
N		 1826 qd_idx = (pd_idx + 1) % raid_disks;
1827 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
16a53ecc 1828 break;
99c0fb5f
N		 1829
1830 case ALGORITHM_PARITY_0:
1831 pd_idx = 0;
1832 qd_idx = 1;
1833 (*dd_idx) += 2;
1834 break;
1835 case ALGORITHM_PARITY_N:
1836 pd_idx = data_disks;
1837 qd_idx = data_disks + 1;
1838 break;
1839
1840 case ALGORITHM_ROTATING_ZERO_RESTART:
1841 /* Exactly the same as RIGHT_ASYMMETRIC, but or
1842 * of blocks for computing Q is different.
1843 */
6e3b96ed 1844 pd_idx = sector_div(stripe2, raid_disks);
99c0fb5f
N		 1845 qd_idx = pd_idx + 1;
1846 if (pd_idx == raid_disks-1) {
1847 (*dd_idx)++; /* Q D D D P */
1848 qd_idx = 0;
1849 } else if (*dd_idx >= pd_idx)
1850 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 1851 ddf_layout = 1;
99c0fb5f
N		 1852 break;
1853
1854 case ALGORITHM_ROTATING_N_RESTART:
1855 /* Same a left_asymmetric, by first stripe is
1856 * D D D P Q rather than
1857 * Q D D D P
1858 */
6e3b96ed
N		 1859 stripe2 += 1;
1860 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 1861 qd_idx = pd_idx + 1;
1862 if (pd_idx == raid_disks-1) {
1863 (*dd_idx)++; /* Q D D D P */
1864 qd_idx = 0;
1865 } else if (*dd_idx >= pd_idx)
1866 (*dd_idx) += 2; /* D D P Q D */
67cc2b81 1867 ddf_layout = 1;
99c0fb5f
N		 1868 break;
1869
1870 case ALGORITHM_ROTATING_N_CONTINUE:
1871 /* Same as left_symmetric but Q is before P */
6e3b96ed 1872 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
99c0fb5f
N		 1873 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
1874 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
67cc2b81 1875 ddf_layout = 1;
99c0fb5f
N		 1876 break;
1877
1878 case ALGORITHM_LEFT_ASYMMETRIC_6:
1879 /* RAID5 left_asymmetric, with Q on last device */
6e3b96ed 1880 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 1881 if (*dd_idx >= pd_idx)
1882 (*dd_idx)++;
1883 qd_idx = raid_disks - 1;
1884 break;
1885
1886 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6e3b96ed 1887 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 1888 if (*dd_idx >= pd_idx)
1889 (*dd_idx)++;
1890 qd_idx = raid_disks - 1;
1891 break;
1892
1893 case ALGORITHM_LEFT_SYMMETRIC_6:
6e3b96ed 1894 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 1895 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
1896 qd_idx = raid_disks - 1;
1897 break;
1898
1899 case ALGORITHM_RIGHT_SYMMETRIC_6:
6e3b96ed 1900 pd_idx = sector_div(stripe2, raid_disks-1);
99c0fb5f
N		 1901 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
1902 qd_idx = raid_disks - 1;
1903 break;
1904
1905 case ALGORITHM_PARITY_0_6:
1906 pd_idx = 0;
1907 (*dd_idx)++;
1908 qd_idx = raid_disks - 1;
1909 break;
1910
16a53ecc 1911 default:
99c0fb5f 1912 BUG();
16a53ecc
N		 1913 }
1914 break;
1da177e4
LT		 1915 }
1916
911d4ee8
N		 1917 if (sh) {
1918 sh->pd_idx = pd_idx;
1919 sh->qd_idx = qd_idx;
67cc2b81 1920 sh->ddf_layout = ddf_layout;
911d4ee8 1921 }
1da177e4
LT		 1922 /*
1923 * Finally, compute the new sector number
1924 */
1925 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
1926 return new_sector;
1927}
1928
1929
784052ec 1930static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous)
1da177e4
LT		 1931{
1932 raid5_conf_t *conf = sh->raid_conf;
b875e531
N		 1933 int raid_disks = sh->disks;
1934 int data_disks = raid_disks - conf->max_degraded;
1da177e4 1935 sector_t new_sector = sh->sector, check;
09c9e5fa
AN		 1936 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
1937 : conf->chunk_sectors;
e183eaed
N		 1938 int algorithm = previous ? conf->prev_algo
1939 : conf->algorithm;
1da177e4
LT		 1940 sector_t stripe;
1941 int chunk_offset;
35f2a591
N		 1942 sector_t chunk_number;
1943 int dummy1, dd_idx = i;
1da177e4 1944 sector_t r_sector;
911d4ee8 1945 struct stripe_head sh2;
1da177e4 1946
16a53ecc 1947
1da177e4
LT		 1948 chunk_offset = sector_div(new_sector, sectors_per_chunk);
1949 stripe = new_sector;
1da177e4 1950
16a53ecc
N		 1951 if (i == sh->pd_idx)
1952 return 0;
1953 switch(conf->level) {
1954 case 4: break;
1955 case 5:
e183eaed 1956 switch (algorithm) {
1da177e4
LT		 1957 case ALGORITHM_LEFT_ASYMMETRIC:
1958 case ALGORITHM_RIGHT_ASYMMETRIC:
1959 if (i > sh->pd_idx)
1960 i--;
1961 break;
1962 case ALGORITHM_LEFT_SYMMETRIC:
1963 case ALGORITHM_RIGHT_SYMMETRIC:
1964 if (i < sh->pd_idx)
1965 i += raid_disks;
1966 i -= (sh->pd_idx + 1);
1967 break;
99c0fb5f
N		 1968 case ALGORITHM_PARITY_0:
1969 i -= 1;
1970 break;
1971 case ALGORITHM_PARITY_N:
1972 break;
1da177e4 1973 default:
99c0fb5f 1974 BUG();
16a53ecc
N		 1975 }
1976 break;
1977 case 6:
d0dabf7e 1978 if (i == sh->qd_idx)
16a53ecc 1979 return 0; /* It is the Q disk */
e183eaed 1980 switch (algorithm) {
16a53ecc
N		 1981 case ALGORITHM_LEFT_ASYMMETRIC:
1982 case ALGORITHM_RIGHT_ASYMMETRIC:
99c0fb5f
N		 1983 case ALGORITHM_ROTATING_ZERO_RESTART:
1984 case ALGORITHM_ROTATING_N_RESTART:
1985 if (sh->pd_idx == raid_disks-1)
1986 i--; /* Q D D D P */
16a53ecc
N		 1987 else if (i > sh->pd_idx)
1988 i -= 2; /* D D P Q D */
1989 break;
1990 case ALGORITHM_LEFT_SYMMETRIC:
1991 case ALGORITHM_RIGHT_SYMMETRIC:
1992 if (sh->pd_idx == raid_disks-1)
1993 i--; /* Q D D D P */
1994 else {
1995 /* D D P Q D */
1996 if (i < sh->pd_idx)
1997 i += raid_disks;
1998 i -= (sh->pd_idx + 2);
1999 }
2000 break;
99c0fb5f
N		 2001 case ALGORITHM_PARITY_0:
2002 i -= 2;
2003 break;
2004 case ALGORITHM_PARITY_N:
2005 break;
2006 case ALGORITHM_ROTATING_N_CONTINUE:
e4424fee 2007 /* Like left_symmetric, but P is before Q */
99c0fb5f
N		 2008 if (sh->pd_idx == 0)
2009 i--; /* P D D D Q */
e4424fee
N		 2010 else {
2011 /* D D Q P D */
2012 if (i < sh->pd_idx)
2013 i += raid_disks;
2014 i -= (sh->pd_idx + 1);
2015 }
99c0fb5f
N		 2016 break;
2017 case ALGORITHM_LEFT_ASYMMETRIC_6:
2018 case ALGORITHM_RIGHT_ASYMMETRIC_6:
2019 if (i > sh->pd_idx)
2020 i--;
2021 break;
2022 case ALGORITHM_LEFT_SYMMETRIC_6:
2023 case ALGORITHM_RIGHT_SYMMETRIC_6:
2024 if (i < sh->pd_idx)
2025 i += data_disks + 1;
2026 i -= (sh->pd_idx + 1);
2027 break;
2028 case ALGORITHM_PARITY_0_6:
2029 i -= 1;
2030 break;
16a53ecc 2031 default:
99c0fb5f 2032 BUG();
16a53ecc
N		 2033 }
2034 break;
1da177e4
LT		 2035 }
2036
2037 chunk_number = stripe * data_disks + i;
35f2a591 2038 r_sector = chunk_number * sectors_per_chunk + chunk_offset;
1da177e4 2039
112bf897 2040 check = raid5_compute_sector(conf, r_sector,
784052ec 2041 previous, &dummy1, &sh2);
911d4ee8
N		 2042 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
2043 || sh2.qd_idx != sh->qd_idx) {
0c55e022
N		 2044 printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n",
2045 mdname(conf->mddev));
1da177e4
LT		 2046 return 0;
2047 }
2048 return r_sector;
2049}
2050
2051
600aa109 2052static void
c0f7bddb 2053schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
600aa109 2054 int rcw, int expand)
e33129d8
DW		 2055{
2056 int i, pd_idx = sh->pd_idx, disks = sh->disks;
c0f7bddb
YT		 2057 raid5_conf_t *conf = sh->raid_conf;
2058 int level = conf->level;
e33129d8
DW		 2059
2060 if (rcw) {
2061 /* if we are not expanding this is a proper write request, and
2062 * there will be bios with new data to be drained into the
2063 * stripe cache
2064 */
2065 if (!expand) {
600aa109
DW		 2066 sh->reconstruct_state = reconstruct_state_drain_run;
2067 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2068 } else
2069 sh->reconstruct_state = reconstruct_state_run;
16a53ecc 2070
ac6b53b6 2071 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
e33129d8
DW		 2072
2073 for (i = disks; i--; ) {
2074 struct r5dev *dev = &sh->dev[i];
2075
2076 if (dev->towrite) {
2077 set_bit(R5_LOCKED, &dev->flags);
d8ee0728 2078 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2079 if (!expand)
2080 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2081 s->locked++;
e33129d8
DW		 2082 }
2083 }
c0f7bddb 2084 if (s->locked + conf->max_degraded == disks)
8b3e6cdc 2085 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
c0f7bddb 2086 atomic_inc(&conf->pending_full_writes);
e33129d8 2087 } else {
c0f7bddb 2088 BUG_ON(level == 6);
e33129d8
DW		 2089 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
2090 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
2091
d8ee0728 2092 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
600aa109
DW		 2093 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
2094 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
ac6b53b6 2095 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
e33129d8
DW		 2096
2097 for (i = disks; i--; ) {
2098 struct r5dev *dev = &sh->dev[i];
2099 if (i == pd_idx)
2100 continue;
2101
e33129d8
DW		 2102 if (dev->towrite &&
2103 (test_bit(R5_UPTODATE, &dev->flags) ||
d8ee0728
DW		 2104 test_bit(R5_Wantcompute, &dev->flags))) {
2105 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 2106 set_bit(R5_LOCKED, &dev->flags);
2107 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 2108 s->locked++;
e33129d8
DW		 2109 }
2110 }
2111 }
2112
c0f7bddb 2113 /* keep the parity disk(s) locked while asynchronous operations
e33129d8
DW		 2114 * are in flight
2115 */
2116 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
2117 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
600aa109 2118 s->locked++;
e33129d8 2119
c0f7bddb
YT		 2120 if (level == 6) {
2121 int qd_idx = sh->qd_idx;
2122 struct r5dev *dev = &sh->dev[qd_idx];
2123
2124 set_bit(R5_LOCKED, &dev->flags);
2125 clear_bit(R5_UPTODATE, &dev->flags);
2126 s->locked++;
2127 }
2128
600aa109 2129 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
e46b272b 2130 __func__, (unsigned long long)sh->sector,
600aa109 2131 s->locked, s->ops_request);
e33129d8 2132}
16a53ecc 2133
1da177e4
LT		 2134/*
2135 * Each stripe/dev can have one or more bion attached.
16a53ecc 2136 * toread/towrite point to the first in a chain.
1da177e4
LT		 2137 * The bi_next chain must be in order.
2138 */
2139static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite)
2140{
2141 struct bio **bip;
2142 raid5_conf_t *conf = sh->raid_conf;
72626685 2143 int firstwrite=0;
1da177e4 2144
cbe47ec5 2145 pr_debug("adding bi b#%llu to stripe s#%llu\n",
1da177e4
LT		 2146 (unsigned long long)bi->bi_sector,
2147 (unsigned long long)sh->sector);
2148
2149
1da177e4 2150 spin_lock_irq(&conf->device_lock);
72626685 2151 if (forwrite) {
1da177e4 2152 bip = &sh->dev[dd_idx].towrite;
72626685
N		 2153 if (*bip == NULL && sh->dev[dd_idx].written == NULL)
2154 firstwrite = 1;
2155 } else
1da177e4
LT		 2156 bip = &sh->dev[dd_idx].toread;
2157 while (*bip && (*bip)->bi_sector < bi->bi_sector) {
2158 if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector)
2159 goto overlap;
2160 bip = & (*bip)->bi_next;
2161 }
2162 if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9))
2163 goto overlap;
2164
78bafebd 2165 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
1da177e4
LT		 2166 if (*bip)
2167 bi->bi_next = *bip;
2168 *bip = bi;
960e739d 2169 bi->bi_phys_segments++;
72626685 2170
1da177e4
LT		 2171 if (forwrite) {
2172 /* check if page is covered */
2173 sector_t sector = sh->dev[dd_idx].sector;
2174 for (bi=sh->dev[dd_idx].towrite;
2175 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
2176 bi && bi->bi_sector <= sector;
2177 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
2178 if (bi->bi_sector + (bi->bi_size>>9) >= sector)
2179 sector = bi->bi_sector + (bi->bi_size>>9);
2180 }
2181 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
2182 set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags);
2183 }
cbe47ec5 2184 spin_unlock_irq(&conf->device_lock);
cbe47ec5
N		 2185
2186 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
2187 (unsigned long long)(*bip)->bi_sector,
2188 (unsigned long long)sh->sector, dd_idx);
2189
2190 if (conf->mddev->bitmap && firstwrite) {
2191 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
2192 STRIPE_SECTORS, 0);
2193 sh->bm_seq = conf->seq_flush+1;
2194 set_bit(STRIPE_BIT_DELAY, &sh->state);
2195 }
1da177e4
LT		 2196 return 1;
2197
2198 overlap:
2199 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
2200 spin_unlock_irq(&conf->device_lock);
1da177e4
LT		 2201 return 0;
2202}
2203
29269553
N		 2204static void end_reshape(raid5_conf_t *conf);
2205
911d4ee8
N		 2206static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous,
2207 struct stripe_head *sh)
ccfcc3c1 2208{
784052ec 2209 int sectors_per_chunk =
09c9e5fa 2210 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
911d4ee8 2211 int dd_idx;
2d2063ce 2212 int chunk_offset = sector_div(stripe, sectors_per_chunk);
112bf897 2213 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
2d2063ce 2214
112bf897
N		 2215 raid5_compute_sector(conf,
2216 stripe * (disks - conf->max_degraded)
b875e531 2217 *sectors_per_chunk + chunk_offset,
112bf897 2218 previous,
911d4ee8 2219 &dd_idx, sh);
ccfcc3c1
N		 2220}
2221
a4456856 2222static void
1fe797e6 2223handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh,
a4456856
DW		 2224 struct stripe_head_state *s, int disks,
2225 struct bio **return_bi)
2226{
2227 int i;
2228 for (i = disks; i--; ) {
2229 struct bio *bi;
2230 int bitmap_end = 0;
2231
2232 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
2233 mdk_rdev_t *rdev;
2234 rcu_read_lock();
2235 rdev = rcu_dereference(conf->disks[i].rdev);
2236 if (rdev && test_bit(In_sync, &rdev->flags))
2237 /* multiple read failures in one stripe */
2238 md_error(conf->mddev, rdev);
2239 rcu_read_unlock();
2240 }
2241 spin_lock_irq(&conf->device_lock);
2242 /* fail all writes first */
2243 bi = sh->dev[i].towrite;
2244 sh->dev[i].towrite = NULL;
2245 if (bi) {
2246 s->to_write--;
2247 bitmap_end = 1;
2248 }
2249
2250 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2251 wake_up(&conf->wait_for_overlap);
2252
2253 while (bi && bi->bi_sector <
2254 sh->dev[i].sector + STRIPE_SECTORS) {
2255 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
2256 clear_bit(BIO_UPTODATE, &bi->bi_flags);
960e739d 2257 if (!raid5_dec_bi_phys_segments(bi)) {
a4456856
DW		 2258 md_write_end(conf->mddev);
2259 bi->bi_next = *return_bi;
2260 *return_bi = bi;
2261 }
2262 bi = nextbi;
2263 }
2264 /* and fail all 'written' */
2265 bi = sh->dev[i].written;
2266 sh->dev[i].written = NULL;
2267 if (bi) bitmap_end = 1;
2268 while (bi && bi->bi_sector <
2269 sh->dev[i].sector + STRIPE_SECTORS) {
2270 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
2271 clear_bit(BIO_UPTODATE, &bi->bi_flags);
960e739d 2272 if (!raid5_dec_bi_phys_segments(bi)) {
a4456856
DW		 2273 md_write_end(conf->mddev);
2274 bi->bi_next = *return_bi;
2275 *return_bi = bi;
2276 }
2277 bi = bi2;
2278 }
2279
b5e98d65
DW		 2280 /* fail any reads if this device is non-operational and
2281 * the data has not reached the cache yet.
2282 */
2283 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
2284 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
2285 test_bit(R5_ReadError, &sh->dev[i].flags))) {
a4456856
DW		 2286 bi = sh->dev[i].toread;
2287 sh->dev[i].toread = NULL;
2288 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2289 wake_up(&conf->wait_for_overlap);
2290 if (bi) s->to_read--;
2291 while (bi && bi->bi_sector <
2292 sh->dev[i].sector + STRIPE_SECTORS) {
2293 struct bio *nextbi =
2294 r5_next_bio(bi, sh->dev[i].sector);
2295 clear_bit(BIO_UPTODATE, &bi->bi_flags);
960e739d 2296 if (!raid5_dec_bi_phys_segments(bi)) {
a4456856
DW		 2297 bi->bi_next = *return_bi;
2298 *return_bi = bi;
2299 }
2300 bi = nextbi;
2301 }
2302 }
2303 spin_unlock_irq(&conf->device_lock);
2304 if (bitmap_end)
2305 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
2306 STRIPE_SECTORS, 0, 0);
2307 }
2308
8b3e6cdc
DW		 2309 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2310 if (atomic_dec_and_test(&conf->pending_full_writes))
2311 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 2312}
2313
93b3dbce 2314/* fetch_block - checks the given member device to see if its data needs
1fe797e6
DW		 2315 * to be read or computed to satisfy a request.
2316 *
2317 * Returns 1 when no more member devices need to be checked, otherwise returns
93b3dbce 2318 * 0 to tell the loop in handle_stripe_fill to continue
f38e1219 2319 */
93b3dbce
N		 2320static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s,
2321 int disk_idx, int disks)
a4456856 2322{
5599becc 2323 struct r5dev *dev = &sh->dev[disk_idx];
f2b3b44d
N		 2324 struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]],
2325 &sh->dev[s->failed_num[1]] };
5599becc 2326
93b3dbce 2327 /* is the data in this block needed, and can we get it? */
5599becc
YT		 2328 if (!test_bit(R5_LOCKED, &dev->flags) &&
2329 !test_bit(R5_UPTODATE, &dev->flags) &&
2330 (dev->toread ||
2331 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
2332 s->syncing || s->expanding ||
5d35e09c
N		 2333 (s->failed >= 1 && fdev[0]->toread) ||
2334 (s->failed >= 2 && fdev[1]->toread) ||
93b3dbce
N		 2335 (sh->raid_conf->level <= 5 && s->failed && fdev[0]->towrite &&
2336 !test_bit(R5_OVERWRITE, &fdev[0]->flags)) ||
2337 (sh->raid_conf->level == 6 && s->failed && s->to_write))) {
5599becc
YT		 2338 /* we would like to get this block, possibly by computing it,
2339 * otherwise read it if the backing disk is insync
2340 */
2341 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
2342 BUG_ON(test_bit(R5_Wantread, &dev->flags));
2343 if ((s->uptodate == disks - 1) &&
f2b3b44d
N		 2344 (s->failed && (disk_idx == s->failed_num[0] ||
2345 disk_idx == s->failed_num[1]))) {
5599becc
YT		 2346 /* have disk failed, and we're requested to fetch it;
2347 * do compute it
a4456856 2348 */
5599becc
YT		 2349 pr_debug("Computing stripe %llu block %d\n",
2350 (unsigned long long)sh->sector, disk_idx);
2351 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2352 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2353 set_bit(R5_Wantcompute, &dev->flags);
2354 sh->ops.target = disk_idx;
2355 sh->ops.target2 = -1; /* no 2nd target */
2356 s->req_compute = 1;
93b3dbce
N		 2357 /* Careful: from this point on 'uptodate' is in the eye
2358 * of raid_run_ops which services 'compute' operations
2359 * before writes. R5_Wantcompute flags a block that will
2360 * be R5_UPTODATE by the time it is needed for a
2361 * subsequent operation.
2362 */
5599becc
YT		 2363 s->uptodate++;
2364 return 1;
2365 } else if (s->uptodate == disks-2 && s->failed >= 2) {
2366 /* Computing 2-failure is *very* expensive; only
2367 * do it if failed >= 2
2368 */
2369 int other;
2370 for (other = disks; other--; ) {
2371 if (other == disk_idx)
2372 continue;
2373 if (!test_bit(R5_UPTODATE,
2374 &sh->dev[other].flags))
2375 break;
a4456856 2376 }
5599becc
YT		 2377 BUG_ON(other < 0);
2378 pr_debug("Computing stripe %llu blocks %d,%d\n",
2379 (unsigned long long)sh->sector,
2380 disk_idx, other);
2381 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2382 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2383 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
2384 set_bit(R5_Wantcompute, &sh->dev[other].flags);
2385 sh->ops.target = disk_idx;
2386 sh->ops.target2 = other;
2387 s->uptodate += 2;
2388 s->req_compute = 1;
2389 return 1;
2390 } else if (test_bit(R5_Insync, &dev->flags)) {
2391 set_bit(R5_LOCKED, &dev->flags);
2392 set_bit(R5_Wantread, &dev->flags);
2393 s->locked++;
2394 pr_debug("Reading block %d (sync=%d)\n",
2395 disk_idx, s->syncing);
a4456856
DW		 2396 }
2397 }
5599becc
YT		 2398
2399 return 0;
2400}
2401
2402/**
93b3dbce 2403 * handle_stripe_fill - read or compute data to satisfy pending requests.
5599becc 2404 */
93b3dbce
N		 2405static void handle_stripe_fill(struct stripe_head *sh,
2406 struct stripe_head_state *s,
2407 int disks)
5599becc
YT		 2408{
2409 int i;
2410
2411 /* look for blocks to read/compute, skip this if a compute
2412 * is already in flight, or if the stripe contents are in the
2413 * midst of changing due to a write
2414 */
2415 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
2416 !sh->reconstruct_state)
2417 for (i = disks; i--; )
93b3dbce 2418 if (fetch_block(sh, s, i, disks))
5599becc 2419 break;
a4456856
DW		 2420 set_bit(STRIPE_HANDLE, &sh->state);
2421}
2422
2423
1fe797e6 2424/* handle_stripe_clean_event
a4456856
DW		 2425 * any written block on an uptodate or failed drive can be returned.
2426 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
2427 * never LOCKED, so we don't need to test 'failed' directly.
2428 */
1fe797e6 2429static void handle_stripe_clean_event(raid5_conf_t *conf,
a4456856
DW		 2430 struct stripe_head *sh, int disks, struct bio **return_bi)
2431{
2432 int i;
2433 struct r5dev *dev;
2434
2435 for (i = disks; i--; )
2436 if (sh->dev[i].written) {
2437 dev = &sh->dev[i];
2438 if (!test_bit(R5_LOCKED, &dev->flags) &&
2439 test_bit(R5_UPTODATE, &dev->flags)) {
2440 /* We can return any write requests */
2441 struct bio *wbi, *wbi2;
2442 int bitmap_end = 0;
45b4233c 2443 pr_debug("Return write for disc %d\n", i);
a4456856
DW		 2444 spin_lock_irq(&conf->device_lock);
2445 wbi = dev->written;
2446 dev->written = NULL;
2447 while (wbi && wbi->bi_sector <
2448 dev->sector + STRIPE_SECTORS) {
2449 wbi2 = r5_next_bio(wbi, dev->sector);
960e739d 2450 if (!raid5_dec_bi_phys_segments(wbi)) {
a4456856
DW		 2451 md_write_end(conf->mddev);
2452 wbi->bi_next = *return_bi;
2453 *return_bi = wbi;
2454 }
2455 wbi = wbi2;
2456 }
2457 if (dev->towrite == NULL)
2458 bitmap_end = 1;
2459 spin_unlock_irq(&conf->device_lock);
2460 if (bitmap_end)
2461 bitmap_endwrite(conf->mddev->bitmap,
2462 sh->sector,
2463 STRIPE_SECTORS,
2464 !test_bit(STRIPE_DEGRADED, &sh->state),
2465 0);
2466 }
2467 }
8b3e6cdc
DW		 2468
2469 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2470 if (atomic_dec_and_test(&conf->pending_full_writes))
2471 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 2472}
2473
c8ac1803
N		 2474static void handle_stripe_dirtying(raid5_conf_t *conf,
2475 struct stripe_head *sh,
2476 struct stripe_head_state *s,
2477 int disks)
a4456856
DW		 2478{
2479 int rmw = 0, rcw = 0, i;
c8ac1803
N		 2480 if (conf->max_degraded == 2) {
2481 /* RAID6 requires 'rcw' in current implementation
2482 * Calculate the real rcw later - for now fake it
2483 * look like rcw is cheaper
2484 */
2485 rcw = 1; rmw = 2;
2486 } else for (i = disks; i--; ) {
a4456856
DW		 2487 /* would I have to read this buffer for read_modify_write */
2488 struct r5dev *dev = &sh->dev[i];
2489 if ((dev->towrite || i == sh->pd_idx) &&
2490 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 2491 !(test_bit(R5_UPTODATE, &dev->flags) ||
2492 test_bit(R5_Wantcompute, &dev->flags))) {
a4456856
DW		 2493 if (test_bit(R5_Insync, &dev->flags))
2494 rmw++;
2495 else
2496 rmw += 2*disks; /* cannot read it */
2497 }
2498 /* Would I have to read this buffer for reconstruct_write */
2499 if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx &&
2500 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 2501 !(test_bit(R5_UPTODATE, &dev->flags) ||
2502 test_bit(R5_Wantcompute, &dev->flags))) {
2503 if (test_bit(R5_Insync, &dev->flags)) rcw++;
a4456856
DW		 2504 else
2505 rcw += 2*disks;
2506 }
2507 }
45b4233c 2508 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
a4456856
DW		 2509 (unsigned long long)sh->sector, rmw, rcw);
2510 set_bit(STRIPE_HANDLE, &sh->state);
2511 if (rmw < rcw && rmw > 0)
2512 /* prefer read-modify-write, but need to get some data */
2513 for (i = disks; i--; ) {
2514 struct r5dev *dev = &sh->dev[i];
2515 if ((dev->towrite || i == sh->pd_idx) &&
2516 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 2517 !(test_bit(R5_UPTODATE, &dev->flags) ||
2518 test_bit(R5_Wantcompute, &dev->flags)) &&
a4456856
DW		 2519 test_bit(R5_Insync, &dev->flags)) {
2520 if (
2521 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
45b4233c 2522 pr_debug("Read_old block "
a4456856
DW		 2523 "%d for r-m-w\n", i);
2524 set_bit(R5_LOCKED, &dev->flags);
2525 set_bit(R5_Wantread, &dev->flags);
2526 s->locked++;
2527 } else {
2528 set_bit(STRIPE_DELAYED, &sh->state);
2529 set_bit(STRIPE_HANDLE, &sh->state);
2530 }
2531 }
2532 }
c8ac1803 2533 if (rcw <= rmw && rcw > 0) {
a4456856 2534 /* want reconstruct write, but need to get some data */
c8ac1803 2535 rcw = 0;
a4456856
DW		 2536 for (i = disks; i--; ) {
2537 struct r5dev *dev = &sh->dev[i];
2538 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
c8ac1803 2539 i != sh->pd_idx && i != sh->qd_idx &&
a4456856 2540 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219 2541 !(test_bit(R5_UPTODATE, &dev->flags) ||
c8ac1803
N		 2542 test_bit(R5_Wantcompute, &dev->flags))) {
2543 rcw++;
2544 if (!test_bit(R5_Insync, &dev->flags))
2545 continue; /* it's a failed drive */
a4456856
DW		 2546 if (
2547 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
45b4233c 2548 pr_debug("Read_old block "
a4456856
DW		 2549 "%d for Reconstruct\n", i);
2550 set_bit(R5_LOCKED, &dev->flags);
2551 set_bit(R5_Wantread, &dev->flags);
2552 s->locked++;
2553 } else {
2554 set_bit(STRIPE_DELAYED, &sh->state);
2555 set_bit(STRIPE_HANDLE, &sh->state);
2556 }
2557 }
2558 }
c8ac1803 2559 }
a4456856
DW		 2560 /* now if nothing is locked, and if we have enough data,
2561 * we can start a write request
2562 */
f38e1219
DW		 2563 /* since handle_stripe can be called at any time we need to handle the
2564 * case where a compute block operation has been submitted and then a
ac6b53b6
DW		 2565 * subsequent call wants to start a write request. raid_run_ops only
2566 * handles the case where compute block and reconstruct are requested
f38e1219
DW		 2567 * simultaneously. If this is not the case then new writes need to be
2568 * held off until the compute completes.
2569 */
976ea8d4
DW		 2570 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
2571 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
2572 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
c0f7bddb 2573 schedule_reconstruction(sh, s, rcw == 0, 0);
a4456856
DW		 2574}
2575
a4456856
DW		 2576static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh,
2577 struct stripe_head_state *s, int disks)
2578{
ecc65c9b 2579 struct r5dev *dev = NULL;
bd2ab670 2580
a4456856 2581 set_bit(STRIPE_HANDLE, &sh->state);
e89f8962 2582
ecc65c9b
DW		 2583 switch (sh->check_state) {
2584 case check_state_idle:
2585 /* start a new check operation if there are no failures */
bd2ab670 2586 if (s->failed == 0) {
bd2ab670 2587 BUG_ON(s->uptodate != disks);
ecc65c9b
DW		 2588 sh->check_state = check_state_run;
2589 set_bit(STRIPE_OP_CHECK, &s->ops_request);
bd2ab670 2590 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
bd2ab670 2591 s->uptodate--;
ecc65c9b 2592 break;
bd2ab670 2593 }
f2b3b44d 2594 dev = &sh->dev[s->failed_num[0]];
ecc65c9b
DW		 2595 /* fall through */
2596 case check_state_compute_result:
2597 sh->check_state = check_state_idle;
2598 if (!dev)
2599 dev = &sh->dev[sh->pd_idx];
2600
2601 /* check that a write has not made the stripe insync */
2602 if (test_bit(STRIPE_INSYNC, &sh->state))
2603 break;
c8894419 2604
a4456856 2605 /* either failed parity check, or recovery is happening */
a4456856
DW		 2606 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
2607 BUG_ON(s->uptodate != disks);
2608
2609 set_bit(R5_LOCKED, &dev->flags);
ecc65c9b 2610 s->locked++;
a4456856 2611 set_bit(R5_Wantwrite, &dev->flags);
830ea016 2612
a4456856 2613 clear_bit(STRIPE_DEGRADED, &sh->state);
a4456856 2614 set_bit(STRIPE_INSYNC, &sh->state);
ecc65c9b
DW		 2615 break;
2616 case check_state_run:
2617 break; /* we will be called again upon completion */
2618 case check_state_check_result:
2619 sh->check_state = check_state_idle;
2620
2621 /* if a failure occurred during the check operation, leave
2622 * STRIPE_INSYNC not set and let the stripe be handled again
2623 */
2624 if (s->failed)
2625 break;
2626
2627 /* handle a successful check operation, if parity is correct
2628 * we are done. Otherwise update the mismatch count and repair
2629 * parity if !MD_RECOVERY_CHECK
2630 */
ad283ea4 2631 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
ecc65c9b
DW		 2632 /* parity is correct (on disc,
2633 * not in buffer any more)
2634 */
2635 set_bit(STRIPE_INSYNC, &sh->state);
2636 else {
2637 conf->mddev->resync_mismatches += STRIPE_SECTORS;
2638 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
2639 /* don't try to repair!! */
2640 set_bit(STRIPE_INSYNC, &sh->state);
2641 else {
2642 sh->check_state = check_state_compute_run;
976ea8d4 2643 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
ecc65c9b
DW		 2644 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2645 set_bit(R5_Wantcompute,
2646 &sh->dev[sh->pd_idx].flags);
2647 sh->ops.target = sh->pd_idx;
ac6b53b6 2648 sh->ops.target2 = -1;
ecc65c9b
DW		 2649 s->uptodate++;
2650 }
2651 }
2652 break;
2653 case check_state_compute_run:
2654 break;
2655 default:
2656 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
2657 __func__, sh->check_state,
2658 (unsigned long long) sh->sector);
2659 BUG();
a4456856
DW		 2660 }
2661}
2662
2663
2664static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh,
36d1c647 2665 struct stripe_head_state *s,
f2b3b44d 2666 int disks)
a4456856 2667{
a4456856 2668 int pd_idx = sh->pd_idx;
34e04e87 2669 int qd_idx = sh->qd_idx;
d82dfee0 2670 struct r5dev *dev;
a4456856
DW		 2671
2672 set_bit(STRIPE_HANDLE, &sh->state);
2673
2674 BUG_ON(s->failed > 2);
d82dfee0 2675
a4456856
DW		 2676 /* Want to check and possibly repair P and Q.
2677 * However there could be one 'failed' device, in which
2678 * case we can only check one of them, possibly using the
2679 * other to generate missing data
2680 */
2681
d82dfee0
DW		 2682 switch (sh->check_state) {
2683 case check_state_idle:
2684 /* start a new check operation if there are < 2 failures */
f2b3b44d 2685 if (s->failed == s->q_failed) {
d82dfee0 2686 /* The only possible failed device holds Q, so it
a4456856
DW		 2687 * makes sense to check P (If anything else were failed,
2688 * we would have used P to recreate it).
2689 */
d82dfee0 2690 sh->check_state = check_state_run;
a4456856 2691 }
f2b3b44d 2692 if (!s->q_failed && s->failed < 2) {
d82dfee0 2693 /* Q is not failed, and we didn't use it to generate
a4456856
DW		 2694 * anything, so it makes sense to check it
2695 */
d82dfee0
DW		 2696 if (sh->check_state == check_state_run)
2697 sh->check_state = check_state_run_pq;
2698 else
2699 sh->check_state = check_state_run_q;
a4456856 2700 }
a4456856 2701
d82dfee0
DW		 2702 /* discard potentially stale zero_sum_result */
2703 sh->ops.zero_sum_result = 0;
a4456856 2704
d82dfee0
DW		 2705 if (sh->check_state == check_state_run) {
2706 /* async_xor_zero_sum destroys the contents of P */
2707 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
2708 s->uptodate--;
a4456856 2709 }
d82dfee0
DW		 2710 if (sh->check_state >= check_state_run &&
2711 sh->check_state <= check_state_run_pq) {
2712 /* async_syndrome_zero_sum preserves P and Q, so
2713 * no need to mark them !uptodate here
2714 */
2715 set_bit(STRIPE_OP_CHECK, &s->ops_request);
2716 break;
a4456856
DW		 2717 }
2718
d82dfee0
DW		 2719 /* we have 2-disk failure */
2720 BUG_ON(s->failed != 2);
2721 /* fall through */
2722 case check_state_compute_result:
2723 sh->check_state = check_state_idle;
a4456856 2724
d82dfee0
DW		 2725 /* check that a write has not made the stripe insync */
2726 if (test_bit(STRIPE_INSYNC, &sh->state))
2727 break;
a4456856
DW		 2728
2729 /* now write out any block on a failed drive,
d82dfee0 2730 * or P or Q if they were recomputed
a4456856 2731 */
d82dfee0 2732 BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */
a4456856 2733 if (s->failed == 2) {
f2b3b44d 2734 dev = &sh->dev[s->failed_num[1]];
a4456856
DW		 2735 s->locked++;
2736 set_bit(R5_LOCKED, &dev->flags);
2737 set_bit(R5_Wantwrite, &dev->flags);
2738 }
2739 if (s->failed >= 1) {
f2b3b44d 2740 dev = &sh->dev[s->failed_num[0]];
a4456856
DW		 2741 s->locked++;
2742 set_bit(R5_LOCKED, &dev->flags);
2743 set_bit(R5_Wantwrite, &dev->flags);
2744 }
d82dfee0 2745 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
a4456856
DW		 2746 dev = &sh->dev[pd_idx];
2747 s->locked++;
2748 set_bit(R5_LOCKED, &dev->flags);
2749 set_bit(R5_Wantwrite, &dev->flags);
2750 }
d82dfee0 2751 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
a4456856
DW		 2752 dev = &sh->dev[qd_idx];
2753 s->locked++;
2754 set_bit(R5_LOCKED, &dev->flags);
2755 set_bit(R5_Wantwrite, &dev->flags);
2756 }
2757 clear_bit(STRIPE_DEGRADED, &sh->state);
2758
2759 set_bit(STRIPE_INSYNC, &sh->state);
d82dfee0
DW		 2760 break;
2761 case check_state_run:
2762 case check_state_run_q:
2763 case check_state_run_pq:
2764 break; /* we will be called again upon completion */
2765 case check_state_check_result:
2766 sh->check_state = check_state_idle;
2767
2768 /* handle a successful check operation, if parity is correct
2769 * we are done. Otherwise update the mismatch count and repair
2770 * parity if !MD_RECOVERY_CHECK
2771 */
2772 if (sh->ops.zero_sum_result == 0) {
2773 /* both parities are correct */
2774 if (!s->failed)
2775 set_bit(STRIPE_INSYNC, &sh->state);
2776 else {
2777 /* in contrast to the raid5 case we can validate
2778 * parity, but still have a failure to write
2779 * back
2780 */
2781 sh->check_state = check_state_compute_result;
2782 /* Returning at this point means that we may go
2783 * off and bring p and/or q uptodate again so
2784 * we make sure to check zero_sum_result again
2785 * to verify if p or q need writeback
2786 */
2787 }
2788 } else {
2789 conf->mddev->resync_mismatches += STRIPE_SECTORS;
2790 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
2791 /* don't try to repair!! */
2792 set_bit(STRIPE_INSYNC, &sh->state);
2793 else {
2794 int *target = &sh->ops.target;
2795
2796 sh->ops.target = -1;
2797 sh->ops.target2 = -1;
2798 sh->check_state = check_state_compute_run;
2799 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2800 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2801 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
2802 set_bit(R5_Wantcompute,
2803 &sh->dev[pd_idx].flags);
2804 *target = pd_idx;
2805 target = &sh->ops.target2;
2806 s->uptodate++;
2807 }
2808 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
2809 set_bit(R5_Wantcompute,
2810 &sh->dev[qd_idx].flags);
2811 *target = qd_idx;
2812 s->uptodate++;
2813 }
2814 }
2815 }
2816 break;
2817 case check_state_compute_run:
2818 break;
2819 default:
2820 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
2821 __func__, sh->check_state,
2822 (unsigned long long) sh->sector);
2823 BUG();
a4456856
DW		 2824 }
2825}
2826
86c374ba 2827static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh)
a4456856
DW		 2828{
2829 int i;
2830
2831 /* We have read all the blocks in this stripe and now we need to
2832 * copy some of them into a target stripe for expand.
2833 */
f0a50d37 2834 struct dma_async_tx_descriptor *tx = NULL;
a4456856
DW		 2835 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
2836 for (i = 0; i < sh->disks; i++)
34e04e87 2837 if (i != sh->pd_idx && i != sh->qd_idx) {
911d4ee8 2838 int dd_idx, j;
a4456856 2839 struct stripe_head *sh2;
a08abd8c 2840 struct async_submit_ctl submit;
a4456856 2841
784052ec 2842 sector_t bn = compute_blocknr(sh, i, 1);
911d4ee8
N		 2843 sector_t s = raid5_compute_sector(conf, bn, 0,
2844 &dd_idx, NULL);
a8c906ca 2845 sh2 = get_active_stripe(conf, s, 0, 1, 1);
a4456856
DW		 2846 if (sh2 == NULL)
2847 /* so far only the early blocks of this stripe
2848 * have been requested. When later blocks
2849 * get requested, we will try again
2850 */
2851 continue;
2852 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
2853 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
2854 /* must have already done this block */
2855 release_stripe(sh2);
2856 continue;
2857 }
f0a50d37
DW		 2858
2859 /* place all the copies on one channel */
a08abd8c 2860 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
f0a50d37 2861 tx = async_memcpy(sh2->dev[dd_idx].page,
88ba2aa5 2862 sh->dev[i].page, 0, 0, STRIPE_SIZE,
a08abd8c 2863 &submit);
f0a50d37 2864
a4456856
DW		 2865 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
2866 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
2867 for (j = 0; j < conf->raid_disks; j++)
2868 if (j != sh2->pd_idx &&
86c374ba 2869 j != sh2->qd_idx &&
a4456856
DW		 2870 !test_bit(R5_Expanded, &sh2->dev[j].flags))
2871 break;
2872 if (j == conf->raid_disks) {
2873 set_bit(STRIPE_EXPAND_READY, &sh2->state);
2874 set_bit(STRIPE_HANDLE, &sh2->state);
2875 }
2876 release_stripe(sh2);
f0a50d37 2877
a4456856 2878 }
a2e08551
N		 2879 /* done submitting copies, wait for them to complete */
2880 if (tx) {
2881 async_tx_ack(tx);
2882 dma_wait_for_async_tx(tx);
2883 }
a4456856 2884}
1da177e4 2885
6bfe0b49 2886
1da177e4
LT		 2887/*
2888 * handle_stripe - do things to a stripe.
2889 *
2890 * We lock the stripe and then examine the state of various bits
2891 * to see what needs to be done.
2892 * Possible results:
2893 * return some read request which now have data
2894 * return some write requests which are safely on disc
2895 * schedule a read on some buffers
2896 * schedule a write of some buffers
2897 * return confirmation of parity correctness
2898 *
1da177e4
LT		 2899 * buffers are taken off read_list or write_list, and bh_cache buffers
2900 * get BH_Lock set before the stripe lock is released.
2901 *
2902 */
a4456856 2903
acfe726b 2904static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s)
1da177e4 2905{
bff61975 2906 raid5_conf_t *conf = sh->raid_conf;
f416885e 2907 int disks = sh->disks;
474af965
N		 2908 struct r5dev *dev;
2909 int i;
1da177e4 2910
acfe726b
N		 2911 memset(s, 0, sizeof(*s));
2912
2913 s->syncing = test_bit(STRIPE_SYNCING, &sh->state);
2914 s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
2915 s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
2916 s->failed_num[0] = -1;
2917 s->failed_num[1] = -1;
1da177e4 2918
acfe726b 2919 /* Now to look around and see what can be done */
1da177e4 2920 rcu_read_lock();
c4c1663b 2921 spin_lock_irq(&conf->device_lock);
16a53ecc
N		 2922 for (i=disks; i--; ) {
2923 mdk_rdev_t *rdev;
acfe726b 2924
16a53ecc 2925 dev = &sh->dev[i];
1da177e4 2926
45b4233c 2927 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
16a53ecc 2928 i, dev->flags, dev->toread, dev->towrite, dev->written);
6c0069c0
YT		 2929 /* maybe we can reply to a read
2930 *
2931 * new wantfill requests are only permitted while
2932 * ops_complete_biofill is guaranteed to be inactive
2933 */
2934 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
2935 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
2936 set_bit(R5_Wantfill, &dev->flags);
1da177e4 2937
16a53ecc 2938 /* now count some things */
cc94015a
N		 2939 if (test_bit(R5_LOCKED, &dev->flags))
2940 s->locked++;
2941 if (test_bit(R5_UPTODATE, &dev->flags))
2942 s->uptodate++;
2d6e4ecc 2943 if (test_bit(R5_Wantcompute, &dev->flags)) {
cc94015a
N		 2944 s->compute++;
2945 BUG_ON(s->compute > 2);
2d6e4ecc 2946 }
1da177e4 2947
acfe726b 2948 if (test_bit(R5_Wantfill, &dev->flags))
cc94015a 2949 s->to_fill++;
acfe726b 2950 else if (dev->toread)
cc94015a 2951 s->to_read++;
16a53ecc 2952 if (dev->towrite) {
cc94015a 2953 s->to_write++;
16a53ecc 2954 if (!test_bit(R5_OVERWRITE, &dev->flags))
cc94015a 2955 s->non_overwrite++;
16a53ecc 2956 }
a4456856 2957 if (dev->written)
cc94015a 2958 s->written++;
16a53ecc 2959 rdev = rcu_dereference(conf->disks[i].rdev);
cc94015a 2960 if (s->blocked_rdev == NULL &&
ac4090d2 2961 rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
cc94015a 2962 s->blocked_rdev = rdev;
6bfe0b49 2963 atomic_inc(&rdev->nr_pending);
6bfe0b49 2964 }
415e72d0
N		 2965 clear_bit(R5_Insync, &dev->flags);
2966 if (!rdev)
2967 /* Not in-sync */;
2968 else if (test_bit(In_sync, &rdev->flags))
2969 set_bit(R5_Insync, &dev->flags);
2970 else {
2971 /* in sync if before recovery_offset */
2972 if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
2973 set_bit(R5_Insync, &dev->flags);
2974 }
2975 if (!test_bit(R5_Insync, &dev->flags)) {
16a53ecc
N		 2976 /* The ReadError flag will just be confusing now */
2977 clear_bit(R5_ReadError, &dev->flags);
2978 clear_bit(R5_ReWrite, &dev->flags);
1da177e4 2979 }
415e72d0
N		 2980 if (test_bit(R5_ReadError, &dev->flags))
2981 clear_bit(R5_Insync, &dev->flags);
2982 if (!test_bit(R5_Insync, &dev->flags)) {
cc94015a
N		 2983 if (s->failed < 2)
2984 s->failed_num[s->failed] = i;
2985 s->failed++;
415e72d0 2986 }
1da177e4 2987 }
c4c1663b 2988 spin_unlock_irq(&conf->device_lock);
1da177e4 2989 rcu_read_unlock();
cc94015a
N		 2990}
2991
2992static void handle_stripe(struct stripe_head *sh)
2993{
2994 struct stripe_head_state s;
474af965 2995 raid5_conf_t *conf = sh->raid_conf;
3687c061 2996 int i;
84789554
N		 2997 int prexor;
2998 int disks = sh->disks;
474af965 2999 struct r5dev *pdev, *qdev;
cc94015a
N		 3000
3001 clear_bit(STRIPE_HANDLE, &sh->state);
3002 if (test_and_set_bit(STRIPE_ACTIVE, &sh->state)) {
3003 /* already being handled, ensure it gets handled
3004 * again when current action finishes */
3005 set_bit(STRIPE_HANDLE, &sh->state);
3006 return;
3007 }
3008
3009 if (test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
3010 set_bit(STRIPE_SYNCING, &sh->state);
3011 clear_bit(STRIPE_INSYNC, &sh->state);
3012 }
3013 clear_bit(STRIPE_DELAYED, &sh->state);
3014
3015 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
3016 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
3017 (unsigned long long)sh->sector, sh->state,
3018 atomic_read(&sh->count), sh->pd_idx, sh->qd_idx,
3019 sh->check_state, sh->reconstruct_state);
3687c061 3020
acfe726b 3021 analyse_stripe(sh, &s);
c5a31000 3022
474af965
N		 3023 if (unlikely(s.blocked_rdev)) {
3024 if (s.syncing || s.expanding || s.expanded ||
3025 s.to_write || s.written) {
3026 set_bit(STRIPE_HANDLE, &sh->state);
3027 goto finish;
3028 }
3029 /* There is nothing for the blocked_rdev to block */
3030 rdev_dec_pending(s.blocked_rdev, conf->mddev);
3031 s.blocked_rdev = NULL;
3032 }
3033
3034 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
3035 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
3036 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
3037 }
3038
3039 pr_debug("locked=%d uptodate=%d to_read=%d"
3040 " to_write=%d failed=%d failed_num=%d,%d\n",
3041 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
3042 s.failed_num[0], s.failed_num[1]);
3043 /* check if the array has lost more than max_degraded devices and,
3044 * if so, some requests might need to be failed.
3045 */
3046 if (s.failed > conf->max_degraded && s.to_read+s.to_write+s.written)
3047 handle_failed_stripe(conf, sh, &s, disks, &s.return_bi);
3048 if (s.failed > conf->max_degraded && s.syncing) {
3049 md_done_sync(conf->mddev, STRIPE_SECTORS, 0);
3050 clear_bit(STRIPE_SYNCING, &sh->state);
3051 s.syncing = 0;
3052 }
3053
3054 /*
3055 * might be able to return some write requests if the parity blocks
3056 * are safe, or on a failed drive
3057 */
3058 pdev = &sh->dev[sh->pd_idx];
3059 s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx)
3060 || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx);
3061 qdev = &sh->dev[sh->qd_idx];
3062 s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx)
3063 || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx)
3064 || conf->level < 6;
3065
3066 if (s.written &&
3067 (s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
3068 && !test_bit(R5_LOCKED, &pdev->flags)
3069 && test_bit(R5_UPTODATE, &pdev->flags)))) &&
3070 (s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
3071 && !test_bit(R5_LOCKED, &qdev->flags)
3072 && test_bit(R5_UPTODATE, &qdev->flags)))))
3073 handle_stripe_clean_event(conf, sh, disks, &s.return_bi);
3074
3075 /* Now we might consider reading some blocks, either to check/generate
3076 * parity, or to satisfy requests
3077 * or to load a block that is being partially written.
3078 */
3079 if (s.to_read || s.non_overwrite
3080 || (conf->level == 6 && s.to_write && s.failed)
3081 || (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding)
3082 handle_stripe_fill(sh, &s, disks);
3083
84789554
N		 3084 /* Now we check to see if any write operations have recently
3085 * completed
3086 */
3087 prexor = 0;
3088 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
3089 prexor = 1;
3090 if (sh->reconstruct_state == reconstruct_state_drain_result ||
3091 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
3092 sh->reconstruct_state = reconstruct_state_idle;
3093
3094 /* All the 'written' buffers and the parity block are ready to
3095 * be written back to disk
3096 */
3097 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags));
3098 BUG_ON(sh->qd_idx >= 0 &&
3099 !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags));
3100 for (i = disks; i--; ) {
3101 struct r5dev *dev = &sh->dev[i];
3102 if (test_bit(R5_LOCKED, &dev->flags) &&
3103 (i == sh->pd_idx || i == sh->qd_idx ||
3104 dev->written)) {
3105 pr_debug("Writing block %d\n", i);
3106 set_bit(R5_Wantwrite, &dev->flags);
3107 if (prexor)
3108 continue;
3109 if (!test_bit(R5_Insync, &dev->flags) ||
3110 ((i == sh->pd_idx || i == sh->qd_idx) &&
3111 s.failed == 0))
3112 set_bit(STRIPE_INSYNC, &sh->state);
3113 }
3114 }
3115 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3116 s.dec_preread_active = 1;
3117 }
3118
3119 /* Now to consider new write requests and what else, if anything
3120 * should be read. We do not handle new writes when:
3121 * 1/ A 'write' operation (copy+xor) is already in flight.
3122 * 2/ A 'check' operation is in flight, as it may clobber the parity
3123 * block.
3124 */
3125 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
3126 handle_stripe_dirtying(conf, sh, &s, disks);
3127
3128 /* maybe we need to check and possibly fix the parity for this stripe
3129 * Any reads will already have been scheduled, so we just see if enough
3130 * data is available. The parity check is held off while parity
3131 * dependent operations are in flight.
3132 */
3133 if (sh->check_state ||
3134 (s.syncing && s.locked == 0 &&
3135 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
3136 !test_bit(STRIPE_INSYNC, &sh->state))) {
3137 if (conf->level == 6)
3138 handle_parity_checks6(conf, sh, &s, disks);
3139 else
3140 handle_parity_checks5(conf, sh, &s, disks);
3141 }
c5a31000
N		 3142
3143 if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) {
3144 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3145 clear_bit(STRIPE_SYNCING, &sh->state);
3146 }
3147
3148 /* If the failed drives are just a ReadError, then we might need
3149 * to progress the repair/check process
3150 */
3151 if (s.failed <= conf->max_degraded && !conf->mddev->ro)
3152 for (i = 0; i < s.failed; i++) {
3153 struct r5dev *dev = &sh->dev[s.failed_num[i]];
3154 if (test_bit(R5_ReadError, &dev->flags)
3155 && !test_bit(R5_LOCKED, &dev->flags)
3156 && test_bit(R5_UPTODATE, &dev->flags)
3157 ) {
3158 if (!test_bit(R5_ReWrite, &dev->flags)) {
3159 set_bit(R5_Wantwrite, &dev->flags);
3160 set_bit(R5_ReWrite, &dev->flags);
3161 set_bit(R5_LOCKED, &dev->flags);
3162 s.locked++;
3163 } else {
3164 /* let's read it back */
3165 set_bit(R5_Wantread, &dev->flags);
3166 set_bit(R5_LOCKED, &dev->flags);
3167 s.locked++;
3168 }
3169 }
3170 }
3171
3172
3687c061
N		 3173 /* Finish reconstruct operations initiated by the expansion process */
3174 if (sh->reconstruct_state == reconstruct_state_result) {
3175 struct stripe_head *sh_src
3176 = get_active_stripe(conf, sh->sector, 1, 1, 1);
3177 if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) {
3178 /* sh cannot be written until sh_src has been read.
3179 * so arrange for sh to be delayed a little
3180 */
3181 set_bit(STRIPE_DELAYED, &sh->state);
3182 set_bit(STRIPE_HANDLE, &sh->state);
3183 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
3184 &sh_src->state))
3185 atomic_inc(&conf->preread_active_stripes);
3186 release_stripe(sh_src);
3187 goto finish;
3188 }
3189 if (sh_src)
3190 release_stripe(sh_src);
3191
3192 sh->reconstruct_state = reconstruct_state_idle;
3193 clear_bit(STRIPE_EXPANDING, &sh->state);
3194 for (i = conf->raid_disks; i--; ) {
3195 set_bit(R5_Wantwrite, &sh->dev[i].flags);
3196 set_bit(R5_LOCKED, &sh->dev[i].flags);
3197 s.locked++;
3198 }
3199 }
f416885e 3200
3687c061
N		 3201 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
3202 !sh->reconstruct_state) {
3203 /* Need to write out all blocks after computing parity */
3204 sh->disks = conf->raid_disks;
3205 stripe_set_idx(sh->sector, conf, 0, sh);
3206 schedule_reconstruction(sh, &s, 1, 1);
3207 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
3208 clear_bit(STRIPE_EXPAND_READY, &sh->state);
3209 atomic_dec(&conf->reshape_stripes);
3210 wake_up(&conf->wait_for_overlap);
3211 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3212 }
3213
3214 if (s.expanding && s.locked == 0 &&
3215 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
3216 handle_stripe_expansion(conf, sh);
16a53ecc 3217
3687c061 3218finish:
6bfe0b49 3219 /* wait for this device to become unblocked */
c5709ef6
N		 3220 if (unlikely(s.blocked_rdev))
3221 md_wait_for_blocked_rdev(s.blocked_rdev, conf->mddev);
6bfe0b49 3222
6c0069c0
YT		 3223 if (s.ops_request)
3224 raid_run_ops(sh, s.ops_request);
3225
f0e43bcd 3226 ops_run_io(sh, &s);
16a53ecc 3227
729a1866 3228
c5709ef6 3229 if (s.dec_preread_active) {
729a1866 3230 /* We delay this until after ops_run_io so that if make_request
e9c7469b 3231 * is waiting on a flush, it won't continue until the writes
729a1866
N		 3232 * have actually been submitted.
3233 */
3234 atomic_dec(&conf->preread_active_stripes);
3235 if (atomic_read(&conf->preread_active_stripes) <
3236 IO_THRESHOLD)
3237 md_wakeup_thread(conf->mddev->thread);
3238 }
3239
c5709ef6 3240 return_io(s.return_bi);
16a53ecc 3241
c4c1663b 3242 clear_bit(STRIPE_ACTIVE, &sh->state);
16a53ecc
N		 3243}
3244
16a53ecc
N		 3245static void raid5_activate_delayed(raid5_conf_t *conf)
3246{
3247 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
3248 while (!list_empty(&conf->delayed_list)) {
3249 struct list_head *l = conf->delayed_list.next;
3250 struct stripe_head *sh;
3251 sh = list_entry(l, struct stripe_head, lru);
3252 list_del_init(l);
3253 clear_bit(STRIPE_DELAYED, &sh->state);
3254 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3255 atomic_inc(&conf->preread_active_stripes);
8b3e6cdc 3256 list_add_tail(&sh->lru, &conf->hold_list);
16a53ecc 3257 }
482c0834 3258 }
16a53ecc
N		 3259}
3260
3261static void activate_bit_delay(raid5_conf_t *conf)
3262{
3263 /* device_lock is held */
3264 struct list_head head;
3265 list_add(&head, &conf->bitmap_list);
3266 list_del_init(&conf->bitmap_list);
3267 while (!list_empty(&head)) {
3268 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
3269 list_del_init(&sh->lru);
3270 atomic_inc(&sh->count);
3271 __release_stripe(conf, sh);
3272 }
3273}
3274
11d8a6e3 3275int md_raid5_congested(mddev_t *mddev, int bits)
f022b2fd 3276{
070ec55d 3277 raid5_conf_t *conf = mddev->private;
f022b2fd
N		 3278
3279 /* No difference between reads and writes. Just check
3280 * how busy the stripe_cache is
3281 */
3fa841d7 3282
f022b2fd
N		 3283 if (conf->inactive_blocked)
3284 return 1;
3285 if (conf->quiesce)
3286 return 1;
3287 if (list_empty_careful(&conf->inactive_list))
3288 return 1;
3289
3290 return 0;
3291}
11d8a6e3
N		 3292EXPORT_SYMBOL_GPL(md_raid5_congested);
3293
3294static int raid5_congested(void *data, int bits)
3295{
3296 mddev_t *mddev = data;
3297
3298 return mddev_congested(mddev, bits) ||
3299 md_raid5_congested(mddev, bits);
3300}
f022b2fd 3301
23032a0e
RBJ		 3302/* We want read requests to align with chunks where possible,
3303 * but write requests don't need to.
3304 */
cc371e66
AK		 3305static int raid5_mergeable_bvec(struct request_queue *q,
3306 struct bvec_merge_data *bvm,
3307 struct bio_vec *biovec)
23032a0e
RBJ		 3308{
3309 mddev_t *mddev = q->queuedata;
cc371e66 3310 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
23032a0e 3311 int max;
9d8f0363 3312 unsigned int chunk_sectors = mddev->chunk_sectors;
cc371e66 3313 unsigned int bio_sectors = bvm->bi_size >> 9;
23032a0e 3314
cc371e66 3315 if ((bvm->bi_rw & 1) == WRITE)
23032a0e
RBJ		 3316 return biovec->bv_len; /* always allow writes to be mergeable */
3317
664e7c41
AN		 3318 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
3319 chunk_sectors = mddev->new_chunk_sectors;
23032a0e
RBJ		 3320 max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
3321 if (max < 0) max = 0;
3322 if (max <= biovec->bv_len && bio_sectors == 0)
3323 return biovec->bv_len;
3324 else
3325 return max;
3326}
3327
f679623f
RBJ		 3328
3329static int in_chunk_boundary(mddev_t *mddev, struct bio *bio)
3330{
3331 sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
9d8f0363 3332 unsigned int chunk_sectors = mddev->chunk_sectors;
f679623f
RBJ		 3333 unsigned int bio_sectors = bio->bi_size >> 9;
3334
664e7c41
AN		 3335 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
3336 chunk_sectors = mddev->new_chunk_sectors;
f679623f
RBJ		 3337 return chunk_sectors >=
3338 ((sector & (chunk_sectors - 1)) + bio_sectors);
3339}
3340
46031f9a
RBJ		 3341/*
3342 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
3343 * later sampled by raid5d.
3344 */
3345static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf)
3346{
3347 unsigned long flags;
3348
3349 spin_lock_irqsave(&conf->device_lock, flags);
3350
3351 bi->bi_next = conf->retry_read_aligned_list;
3352 conf->retry_read_aligned_list = bi;
3353
3354 spin_unlock_irqrestore(&conf->device_lock, flags);
3355 md_wakeup_thread(conf->mddev->thread);
3356}
3357
3358
3359static struct bio *remove_bio_from_retry(raid5_conf_t *conf)
3360{
3361 struct bio *bi;
3362
3363 bi = conf->retry_read_aligned;
3364 if (bi) {
3365 conf->retry_read_aligned = NULL;
3366 return bi;
3367 }
3368 bi = conf->retry_read_aligned_list;
3369 if(bi) {
387bb173 3370 conf->retry_read_aligned_list = bi->bi_next;
46031f9a 3371 bi->bi_next = NULL;
960e739d
JA		 3372 /*
3373 * this sets the active strip count to 1 and the processed
3374 * strip count to zero (upper 8 bits)
3375 */
46031f9a 3376 bi->bi_phys_segments = 1; /* biased count of active stripes */
46031f9a
RBJ		 3377 }
3378
3379 return bi;
3380}
3381
3382
f679623f
RBJ		 3383/*
3384 * The "raid5_align_endio" should check if the read succeeded and if it
3385 * did, call bio_endio on the original bio (having bio_put the new bio
3386 * first).
3387 * If the read failed..
3388 */
6712ecf8 3389static void raid5_align_endio(struct bio *bi, int error)
f679623f
RBJ		 3390{
3391 struct bio* raid_bi = bi->bi_private;
46031f9a
RBJ		 3392 mddev_t *mddev;
3393 raid5_conf_t *conf;
3394 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
3395 mdk_rdev_t *rdev;
3396
f679623f 3397 bio_put(bi);
46031f9a 3398
46031f9a
RBJ		 3399 rdev = (void*)raid_bi->bi_next;
3400 raid_bi->bi_next = NULL;
2b7f2228
N		 3401 mddev = rdev->mddev;
3402 conf = mddev->private;
46031f9a
RBJ		 3403
3404 rdev_dec_pending(rdev, conf->mddev);
3405
3406 if (!error && uptodate) {
6712ecf8 3407 bio_endio(raid_bi, 0);
46031f9a
RBJ		 3408 if (atomic_dec_and_test(&conf->active_aligned_reads))
3409 wake_up(&conf->wait_for_stripe);
6712ecf8 3410 return;
46031f9a
RBJ		 3411 }
3412
3413
45b4233c 3414 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
46031f9a
RBJ		 3415
3416 add_bio_to_retry(raid_bi, conf);
f679623f
RBJ		 3417}
3418
387bb173
NB		 3419static int bio_fits_rdev(struct bio *bi)
3420{
165125e1 3421 struct request_queue *q = bdev_get_queue(bi->bi_bdev);
387bb173 3422
ae03bf63 3423 if ((bi->bi_size>>9) > queue_max_sectors(q))
387bb173
NB		 3424 return 0;
3425 blk_recount_segments(q, bi);
8a78362c 3426 if (bi->bi_phys_segments > queue_max_segments(q))
387bb173
NB		 3427 return 0;
3428
3429 if (q->merge_bvec_fn)
3430 /* it's too hard to apply the merge_bvec_fn at this stage,
3431 * just just give up
3432 */
3433 return 0;
3434
3435 return 1;
3436}
3437
3438
21a52c6d 3439static int chunk_aligned_read(mddev_t *mddev, struct bio * raid_bio)
f679623f 3440{
070ec55d 3441 raid5_conf_t *conf = mddev->private;
8553fe7e 3442 int dd_idx;
f679623f
RBJ		 3443 struct bio* align_bi;
3444 mdk_rdev_t *rdev;
3445
3446 if (!in_chunk_boundary(mddev, raid_bio)) {
45b4233c 3447 pr_debug("chunk_aligned_read : non aligned\n");
f679623f
RBJ		 3448 return 0;
3449 }
3450 /*
a167f663 3451 * use bio_clone_mddev to make a copy of the bio
f679623f 3452 */
a167f663 3453 align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev);
f679623f
RBJ		 3454 if (!align_bi)
3455 return 0;
3456 /*
3457 * set bi_end_io to a new function, and set bi_private to the
3458 * original bio.
3459 */
3460 align_bi->bi_end_io = raid5_align_endio;
3461 align_bi->bi_private = raid_bio;
3462 /*
3463 * compute position
3464 */
112bf897
N		 3465 align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector,
3466 0,
911d4ee8 3467 &dd_idx, NULL);
f679623f
RBJ		 3468
3469 rcu_read_lock();
3470 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
3471 if (rdev && test_bit(In_sync, &rdev->flags)) {
f679623f
RBJ		 3472 atomic_inc(&rdev->nr_pending);
3473 rcu_read_unlock();
46031f9a
RBJ		 3474 raid_bio->bi_next = (void*)rdev;
3475 align_bi->bi_bdev = rdev->bdev;
3476 align_bi->bi_flags &= ~(1 << BIO_SEG_VALID);
3477 align_bi->bi_sector += rdev->data_offset;
3478
387bb173
NB		 3479 if (!bio_fits_rdev(align_bi)) {
3480 /* too big in some way */
3481 bio_put(align_bi);
3482 rdev_dec_pending(rdev, mddev);
3483 return 0;
3484 }
3485
46031f9a
RBJ		 3486 spin_lock_irq(&conf->device_lock);
3487 wait_event_lock_irq(conf->wait_for_stripe,
3488 conf->quiesce == 0,
3489 conf->device_lock, /* nothing */);
3490 atomic_inc(&conf->active_aligned_reads);
3491 spin_unlock_irq(&conf->device_lock);
3492
f679623f
RBJ		 3493 generic_make_request(align_bi);
3494 return 1;
3495 } else {
3496 rcu_read_unlock();
46031f9a 3497 bio_put(align_bi);
f679623f
RBJ		 3498 return 0;
3499 }
3500}
3501
8b3e6cdc
DW		 3502/* __get_priority_stripe - get the next stripe to process
3503 *
3504 * Full stripe writes are allowed to pass preread active stripes up until
3505 * the bypass_threshold is exceeded. In general the bypass_count
3506 * increments when the handle_list is handled before the hold_list; however, it
3507 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
3508 * stripe with in flight i/o. The bypass_count will be reset when the
3509 * head of the hold_list has changed, i.e. the head was promoted to the
3510 * handle_list.
3511 */
3512static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf)
3513{
3514 struct stripe_head *sh;
3515
3516 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
3517 __func__,
3518 list_empty(&conf->handle_list) ? "empty" : "busy",
3519 list_empty(&conf->hold_list) ? "empty" : "busy",
3520 atomic_read(&conf->pending_full_writes), conf->bypass_count);
3521
3522 if (!list_empty(&conf->handle_list)) {
3523 sh = list_entry(conf->handle_list.next, typeof(*sh), lru);
3524
3525 if (list_empty(&conf->hold_list))
3526 conf->bypass_count = 0;
3527 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
3528 if (conf->hold_list.next == conf->last_hold)
3529 conf->bypass_count++;
3530 else {
3531 conf->last_hold = conf->hold_list.next;
3532 conf->bypass_count -= conf->bypass_threshold;
3533 if (conf->bypass_count < 0)
3534 conf->bypass_count = 0;
3535 }
3536 }
3537 } else if (!list_empty(&conf->hold_list) &&
3538 ((conf->bypass_threshold &&
3539 conf->bypass_count > conf->bypass_threshold) ||
3540 atomic_read(&conf->pending_full_writes) == 0)) {
3541 sh = list_entry(conf->hold_list.next,
3542 typeof(*sh), lru);
3543 conf->bypass_count -= conf->bypass_threshold;
3544 if (conf->bypass_count < 0)
3545 conf->bypass_count = 0;
3546 } else
3547 return NULL;
3548
3549 list_del_init(&sh->lru);
3550 atomic_inc(&sh->count);
3551 BUG_ON(atomic_read(&sh->count) != 1);
3552 return sh;
3553}
f679623f 3554
21a52c6d 3555static int make_request(mddev_t *mddev, struct bio * bi)
1da177e4 3556{
070ec55d 3557 raid5_conf_t *conf = mddev->private;
911d4ee8 3558 int dd_idx;
1da177e4
LT		 3559 sector_t new_sector;
3560 sector_t logical_sector, last_sector;
3561 struct stripe_head *sh;
a362357b 3562 const int rw = bio_data_dir(bi);
49077326 3563 int remaining;
7c13edc8 3564 int plugged;
1da177e4 3565
e9c7469b
TH		 3566 if (unlikely(bi->bi_rw & REQ_FLUSH)) {
3567 md_flush_request(mddev, bi);
e5dcdd80
N		 3568 return 0;
3569 }
3570
3d310eb7 3571 md_write_start(mddev, bi);
06d91a5f 3572
802ba064 3573 if (rw == READ &&
52488615 3574 mddev->reshape_position == MaxSector &&
21a52c6d 3575 chunk_aligned_read(mddev,bi))
99c0fb5f 3576 return 0;
52488615 3577
1da177e4
LT		 3578 logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
3579 last_sector = bi->bi_sector + (bi->bi_size>>9);
3580 bi->bi_next = NULL;
3581 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
06d91a5f 3582
7c13edc8 3583 plugged = mddev_check_plugged(mddev);
1da177e4
LT		 3584 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
3585 DEFINE_WAIT(w);
16a53ecc 3586 int disks, data_disks;
b5663ba4 3587 int previous;
b578d55f 3588
7ecaa1e6 3589 retry:
b5663ba4 3590 previous = 0;
b0f9ec04 3591 disks = conf->raid_disks;
b578d55f 3592 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
b0f9ec04 3593 if (unlikely(conf->reshape_progress != MaxSector)) {
fef9c61f 3594 /* spinlock is needed as reshape_progress may be
df8e7f76
N		 3595 * 64bit on a 32bit platform, and so it might be
3596 * possible to see a half-updated value
aeb878b0 3597 * Of course reshape_progress could change after
df8e7f76
N		 3598 * the lock is dropped, so once we get a reference
3599 * to the stripe that we think it is, we will have
3600 * to check again.
3601 */
7ecaa1e6 3602 spin_lock_irq(&conf->device_lock);
fef9c61f
N		 3603 if (mddev->delta_disks < 0
3604 ? logical_sector < conf->reshape_progress
3605 : logical_sector >= conf->reshape_progress) {
7ecaa1e6 3606 disks = conf->previous_raid_disks;
b5663ba4
N		 3607 previous = 1;
3608 } else {
fef9c61f
N		 3609 if (mddev->delta_disks < 0
3610 ? logical_sector < conf->reshape_safe
3611 : logical_sector >= conf->reshape_safe) {
b578d55f
N		 3612 spin_unlock_irq(&conf->device_lock);
3613 schedule();
3614 goto retry;
3615 }
3616 }
7ecaa1e6
N		 3617 spin_unlock_irq(&conf->device_lock);
3618 }
16a53ecc
N		 3619 data_disks = disks - conf->max_degraded;
3620
112bf897
N		 3621 new_sector = raid5_compute_sector(conf, logical_sector,
3622 previous,
911d4ee8 3623 &dd_idx, NULL);
0c55e022 3624 pr_debug("raid456: make_request, sector %llu logical %llu\n",
1da177e4
LT		 3625 (unsigned long long)new_sector,
3626 (unsigned long long)logical_sector);
3627
b5663ba4 3628 sh = get_active_stripe(conf, new_sector, previous,
a8c906ca 3629 (bi->bi_rw&RWA_MASK), 0);
1da177e4 3630 if (sh) {
b0f9ec04 3631 if (unlikely(previous)) {
7ecaa1e6 3632 /* expansion might have moved on while waiting for a
df8e7f76
N		 3633 * stripe, so we must do the range check again.
3634 * Expansion could still move past after this
3635 * test, but as we are holding a reference to
3636 * 'sh', we know that if that happens,
3637 * STRIPE_EXPANDING will get set and the expansion
3638 * won't proceed until we finish with the stripe.
7ecaa1e6
N		 3639 */
3640 int must_retry = 0;
3641 spin_lock_irq(&conf->device_lock);
b0f9ec04
N		 3642 if (mddev->delta_disks < 0
3643 ? logical_sector >= conf->reshape_progress
3644 : logical_sector < conf->reshape_progress)
7ecaa1e6
N		 3645 /* mismatch, need to try again */
3646 must_retry = 1;
3647 spin_unlock_irq(&conf->device_lock);
3648 if (must_retry) {
3649 release_stripe(sh);
7a3ab908 3650 schedule();
7ecaa1e6
N		 3651 goto retry;
3652 }
3653 }
e62e58a5 3654
ffd96e35 3655 if (rw == WRITE &&
a5c308d4 3656 logical_sector >= mddev->suspend_lo &&
e464eafd
N		 3657 logical_sector < mddev->suspend_hi) {
3658 release_stripe(sh);
e62e58a5
N		 3659 /* As the suspend_* range is controlled by
3660 * userspace, we want an interruptible
3661 * wait.
3662 */
3663 flush_signals(current);
3664 prepare_to_wait(&conf->wait_for_overlap,
3665 &w, TASK_INTERRUPTIBLE);
3666 if (logical_sector >= mddev->suspend_lo &&
3667 logical_sector < mddev->suspend_hi)
3668 schedule();
e464eafd
N		 3669 goto retry;
3670 }
7ecaa1e6
N		 3671
3672 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
ffd96e35 3673 !add_stripe_bio(sh, bi, dd_idx, rw)) {
7ecaa1e6
N		 3674 /* Stripe is busy expanding or
3675 * add failed due to overlap. Flush everything
1da177e4
LT		 3676 * and wait a while
3677 */
482c0834 3678 md_wakeup_thread(mddev->thread);
1da177e4
LT		 3679 release_stripe(sh);
3680 schedule();
3681 goto retry;
3682 }
3683 finish_wait(&conf->wait_for_overlap, &w);
6ed3003c
N		 3684 set_bit(STRIPE_HANDLE, &sh->state);
3685 clear_bit(STRIPE_DELAYED, &sh->state);
e9c7469b 3686 if ((bi->bi_rw & REQ_SYNC) &&
729a1866
N		 3687 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3688 atomic_inc(&conf->preread_active_stripes);
1da177e4 3689 release_stripe(sh);
1da177e4
LT		 3690 } else {
3691 /* cannot get stripe for read-ahead, just give-up */
3692 clear_bit(BIO_UPTODATE, &bi->bi_flags);
3693 finish_wait(&conf->wait_for_overlap, &w);
3694 break;
3695 }
3696
3697 }
7c13edc8
N		 3698 if (!plugged)
3699 md_wakeup_thread(mddev->thread);
3700
1da177e4 3701 spin_lock_irq(&conf->device_lock);
960e739d 3702 remaining = raid5_dec_bi_phys_segments(bi);
f6344757
N		 3703 spin_unlock_irq(&conf->device_lock);
3704 if (remaining == 0) {
1da177e4 3705
16a53ecc 3706 if ( rw == WRITE )
1da177e4 3707 md_write_end(mddev);
6712ecf8 3708
0e13fe23 3709 bio_endio(bi, 0);
1da177e4 3710 }
729a1866 3711
1da177e4
LT		 3712 return 0;
3713}
3714
b522adcd
DW		 3715static sector_t raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks);
3716
52c03291 3717static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped)
1da177e4 3718{
52c03291
N		 3719 /* reshaping is quite different to recovery/resync so it is
3720 * handled quite separately ... here.
3721 *
3722 * On each call to sync_request, we gather one chunk worth of
3723 * destination stripes and flag them as expanding.
3724 * Then we find all the source stripes and request reads.
3725 * As the reads complete, handle_stripe will copy the data
3726 * into the destination stripe and release that stripe.
3727 */
7b92813c 3728 raid5_conf_t *conf = mddev->private;
1da177e4 3729 struct stripe_head *sh;
ccfcc3c1 3730 sector_t first_sector, last_sector;
f416885e
N		 3731 int raid_disks = conf->previous_raid_disks;
3732 int data_disks = raid_disks - conf->max_degraded;
3733 int new_data_disks = conf->raid_disks - conf->max_degraded;
52c03291
N		 3734 int i;
3735 int dd_idx;
c8f517c4 3736 sector_t writepos, readpos, safepos;
ec32a2bd 3737 sector_t stripe_addr;
7a661381 3738 int reshape_sectors;
ab69ae12 3739 struct list_head stripes;
52c03291 3740
fef9c61f
N		 3741 if (sector_nr == 0) {
3742 /* If restarting in the middle, skip the initial sectors */
3743 if (mddev->delta_disks < 0 &&
3744 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
3745 sector_nr = raid5_size(mddev, 0, 0)
3746 - conf->reshape_progress;
a639755c 3747 } else if (mddev->delta_disks >= 0 &&
fef9c61f
N		 3748 conf->reshape_progress > 0)
3749 sector_nr = conf->reshape_progress;
f416885e 3750 sector_div(sector_nr, new_data_disks);
fef9c61f 3751 if (sector_nr) {
8dee7211
N		 3752 mddev->curr_resync_completed = sector_nr;
3753 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
fef9c61f
N		 3754 *skipped = 1;
3755 return sector_nr;
3756 }
52c03291
N		 3757 }
3758
7a661381
N		 3759 /* We need to process a full chunk at a time.
3760 * If old and new chunk sizes differ, we need to process the
3761 * largest of these
3762 */
664e7c41
AN		 3763 if (mddev->new_chunk_sectors > mddev->chunk_sectors)
3764 reshape_sectors = mddev->new_chunk_sectors;
7a661381 3765 else
9d8f0363 3766 reshape_sectors = mddev->chunk_sectors;
7a661381 3767
52c03291
N		 3768 /* we update the metadata when there is more than 3Meg
3769 * in the block range (that is rather arbitrary, should
3770 * probably be time based) or when the data about to be
3771 * copied would over-write the source of the data at
3772 * the front of the range.
fef9c61f
N		 3773 * i.e. one new_stripe along from reshape_progress new_maps
3774 * to after where reshape_safe old_maps to
52c03291 3775 */
fef9c61f 3776 writepos = conf->reshape_progress;
f416885e 3777 sector_div(writepos, new_data_disks);
c8f517c4
N		 3778 readpos = conf->reshape_progress;
3779 sector_div(readpos, data_disks);
fef9c61f 3780 safepos = conf->reshape_safe;
f416885e 3781 sector_div(safepos, data_disks);
fef9c61f 3782 if (mddev->delta_disks < 0) {
ed37d83e 3783 writepos -= min_t(sector_t, reshape_sectors, writepos);
c8f517c4 3784 readpos += reshape_sectors;
7a661381 3785 safepos += reshape_sectors;
fef9c61f 3786 } else {
7a661381 3787 writepos += reshape_sectors;
ed37d83e
N		 3788 readpos -= min_t(sector_t, reshape_sectors, readpos);
3789 safepos -= min_t(sector_t, reshape_sectors, safepos);
fef9c61f 3790 }
52c03291 3791
c8f517c4
N		 3792 /* 'writepos' is the most advanced device address we might write.
3793 * 'readpos' is the least advanced device address we might read.
3794 * 'safepos' is the least address recorded in the metadata as having
3795 * been reshaped.
3796 * If 'readpos' is behind 'writepos', then there is no way that we can
3797 * ensure safety in the face of a crash - that must be done by userspace
3798 * making a backup of the data. So in that case there is no particular
3799 * rush to update metadata.
3800 * Otherwise if 'safepos' is behind 'writepos', then we really need to
3801 * update the metadata to advance 'safepos' to match 'readpos' so that
3802 * we can be safe in the event of a crash.
3803 * So we insist on updating metadata if safepos is behind writepos and
3804 * readpos is beyond writepos.
3805 * In any case, update the metadata every 10 seconds.
3806 * Maybe that number should be configurable, but I'm not sure it is
3807 * worth it.... maybe it could be a multiple of safemode_delay???
3808 */
fef9c61f 3809 if ((mddev->delta_disks < 0
c8f517c4
N		 3810 ? (safepos > writepos && readpos < writepos)
3811 : (safepos < writepos && readpos > writepos)) ||
3812 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
52c03291
N		 3813 /* Cannot proceed until we've updated the superblock... */
3814 wait_event(conf->wait_for_overlap,
3815 atomic_read(&conf->reshape_stripes)==0);
fef9c61f 3816 mddev->reshape_position = conf->reshape_progress;
75d3da43 3817 mddev->curr_resync_completed = sector_nr;
c8f517c4 3818 conf->reshape_checkpoint = jiffies;
850b2b42 3819 set_bit(MD_CHANGE_DEVS, &mddev->flags);
52c03291 3820 md_wakeup_thread(mddev->thread);
850b2b42 3821 wait_event(mddev->sb_wait, mddev->flags == 0 ||
52c03291
N		 3822 kthread_should_stop());
3823 spin_lock_irq(&conf->device_lock);
fef9c61f 3824 conf->reshape_safe = mddev->reshape_position;
52c03291
N		 3825 spin_unlock_irq(&conf->device_lock);
3826 wake_up(&conf->wait_for_overlap);
acb180b0 3827 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
52c03291
N		 3828 }
3829
ec32a2bd
N		 3830 if (mddev->delta_disks < 0) {
3831 BUG_ON(conf->reshape_progress == 0);
3832 stripe_addr = writepos;
3833 BUG_ON((mddev->dev_sectors &
7a661381
N		 3834 ~((sector_t)reshape_sectors - 1))
3835 - reshape_sectors - stripe_addr
ec32a2bd
N		 3836 != sector_nr);
3837 } else {
7a661381 3838 BUG_ON(writepos != sector_nr + reshape_sectors);
ec32a2bd
N		 3839 stripe_addr = sector_nr;
3840 }
ab69ae12 3841 INIT_LIST_HEAD(&stripes);
7a661381 3842 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
52c03291 3843 int j;
a9f326eb 3844 int skipped_disk = 0;
a8c906ca 3845 sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
52c03291
N		 3846 set_bit(STRIPE_EXPANDING, &sh->state);
3847 atomic_inc(&conf->reshape_stripes);
3848 /* If any of this stripe is beyond the end of the old
3849 * array, then we need to zero those blocks
3850 */
3851 for (j=sh->disks; j--;) {
3852 sector_t s;
3853 if (j == sh->pd_idx)
3854 continue;
f416885e 3855 if (conf->level == 6 &&
d0dabf7e 3856 j == sh->qd_idx)
f416885e 3857 continue;
784052ec 3858 s = compute_blocknr(sh, j, 0);
b522adcd 3859 if (s < raid5_size(mddev, 0, 0)) {
a9f326eb 3860 skipped_disk = 1;
52c03291
N		 3861 continue;
3862 }
3863 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
3864 set_bit(R5_Expanded, &sh->dev[j].flags);
3865 set_bit(R5_UPTODATE, &sh->dev[j].flags);
3866 }
a9f326eb 3867 if (!skipped_disk) {
52c03291
N		 3868 set_bit(STRIPE_EXPAND_READY, &sh->state);
3869 set_bit(STRIPE_HANDLE, &sh->state);
3870 }
ab69ae12 3871 list_add(&sh->lru, &stripes);
52c03291
N		 3872 }
3873 spin_lock_irq(&conf->device_lock);
fef9c61f 3874 if (mddev->delta_disks < 0)
7a661381 3875 conf->reshape_progress -= reshape_sectors * new_data_disks;
fef9c61f 3876 else
7a661381 3877 conf->reshape_progress += reshape_sectors * new_data_disks;
52c03291
N		 3878 spin_unlock_irq(&conf->device_lock);
3879 /* Ok, those stripe are ready. We can start scheduling
3880 * reads on the source stripes.
3881 * The source stripes are determined by mapping the first and last
3882 * block on the destination stripes.
3883 */
52c03291 3884 first_sector =
ec32a2bd 3885 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
911d4ee8 3886 1, &dd_idx, NULL);
52c03291 3887 last_sector =
0e6e0271 3888 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
09c9e5fa 3889 * new_data_disks - 1),
911d4ee8 3890 1, &dd_idx, NULL);
58c0fed4
AN		 3891 if (last_sector >= mddev->dev_sectors)
3892 last_sector = mddev->dev_sectors - 1;
52c03291 3893 while (first_sector <= last_sector) {
a8c906ca 3894 sh = get_active_stripe(conf, first_sector, 1, 0, 1);
52c03291
N		 3895 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3896 set_bit(STRIPE_HANDLE, &sh->state);
3897 release_stripe(sh);
3898 first_sector += STRIPE_SECTORS;
3899 }
ab69ae12
N		 3900 /* Now that the sources are clearly marked, we can release
3901 * the destination stripes
3902 */
3903 while (!list_empty(&stripes)) {
3904 sh = list_entry(stripes.next, struct stripe_head, lru);
3905 list_del_init(&sh->lru);
3906 release_stripe(sh);
3907 }
c6207277
N		 3908 /* If this takes us to the resync_max point where we have to pause,
3909 * then we need to write out the superblock.
3910 */
7a661381 3911 sector_nr += reshape_sectors;
c03f6a19
N		 3912 if ((sector_nr - mddev->curr_resync_completed) * 2
3913 >= mddev->resync_max - mddev->curr_resync_completed) {
c6207277
N		 3914 /* Cannot proceed until we've updated the superblock... */
3915 wait_event(conf->wait_for_overlap,
3916 atomic_read(&conf->reshape_stripes) == 0);
fef9c61f 3917 mddev->reshape_position = conf->reshape_progress;
75d3da43 3918 mddev->curr_resync_completed = sector_nr;
c8f517c4 3919 conf->reshape_checkpoint = jiffies;
c6207277
N		 3920 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3921 md_wakeup_thread(mddev->thread);
3922 wait_event(mddev->sb_wait,
3923 !test_bit(MD_CHANGE_DEVS, &mddev->flags)
3924 || kthread_should_stop());
3925 spin_lock_irq(&conf->device_lock);
fef9c61f 3926 conf->reshape_safe = mddev->reshape_position;
c6207277
N		 3927 spin_unlock_irq(&conf->device_lock);
3928 wake_up(&conf->wait_for_overlap);
acb180b0 3929 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
c6207277 3930 }
7a661381 3931 return reshape_sectors;
52c03291
N		 3932}
3933
3934/* FIXME go_faster isn't used */
3935static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
3936{
7b92813c 3937 raid5_conf_t *conf = mddev->private;
52c03291 3938 struct stripe_head *sh;
58c0fed4 3939 sector_t max_sector = mddev->dev_sectors;
57dab0bd 3940 sector_t sync_blocks;
16a53ecc
N		 3941 int still_degraded = 0;
3942 int i;
1da177e4 3943
72626685 3944 if (sector_nr >= max_sector) {
1da177e4 3945 /* just being told to finish up .. nothing much to do */
cea9c228 3946
29269553
N		 3947 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
3948 end_reshape(conf);
3949 return 0;
3950 }
72626685
N		 3951
3952 if (mddev->curr_resync < max_sector) /* aborted */
3953 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
3954 &sync_blocks, 1);
16a53ecc 3955 else /* completed sync */
72626685
N		 3956 conf->fullsync = 0;
3957 bitmap_close_sync(mddev->bitmap);
3958
1da177e4
LT		 3959 return 0;
3960 }
ccfcc3c1 3961
64bd660b
N		 3962 /* Allow raid5_quiesce to complete */
3963 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
3964
52c03291
N		 3965 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3966 return reshape_request(mddev, sector_nr, skipped);
f6705578 3967
c6207277
N		 3968 /* No need to check resync_max as we never do more than one
3969 * stripe, and as resync_max will always be on a chunk boundary,
3970 * if the check in md_do_sync didn't fire, there is no chance
3971 * of overstepping resync_max here
3972 */
3973
16a53ecc 3974 /* if there is too many failed drives and we are trying
1da177e4
LT		 3975 * to resync, then assert that we are finished, because there is
3976 * nothing we can do.
3977 */
3285edf1 3978 if (mddev->degraded >= conf->max_degraded &&
16a53ecc 3979 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
58c0fed4 3980 sector_t rv = mddev->dev_sectors - sector_nr;
57afd89f 3981 *skipped = 1;
1da177e4
LT		 3982 return rv;
3983 }
72626685 3984 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
3855ad9f 3985 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
72626685
N		 3986 !conf->fullsync && sync_blocks >= STRIPE_SECTORS) {
3987 /* we can skip this block, and probably more */
3988 sync_blocks /= STRIPE_SECTORS;
3989 *skipped = 1;
3990 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
3991 }
1da177e4 3992
b47490c9
N		 3993
3994 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
3995
a8c906ca 3996 sh = get_active_stripe(conf, sector_nr, 0, 1, 0);
1da177e4 3997 if (sh == NULL) {
a8c906ca 3998 sh = get_active_stripe(conf, sector_nr, 0, 0, 0);
1da177e4 3999 /* make sure we don't swamp the stripe cache if someone else
16a53ecc 4000 * is trying to get access
1da177e4 4001 */
66c006a5 4002 schedule_timeout_uninterruptible(1);
1da177e4 4003 }
16a53ecc
N		 4004 /* Need to check if array will still be degraded after recovery/resync
4005 * We don't need to check the 'failed' flag as when that gets set,
4006 * recovery aborts.
4007 */
f001a70c 4008 for (i = 0; i < conf->raid_disks; i++)
16a53ecc
N		 4009 if (conf->disks[i].rdev == NULL)
4010 still_degraded = 1;
4011
4012 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
4013
83206d66 4014 set_bit(STRIPE_SYNC_REQUESTED, &sh->state);
1da177e4 4015
1442577b 4016 handle_stripe(sh);
1da177e4
LT		 4017 release_stripe(sh);
4018
4019 return STRIPE_SECTORS;
4020}
4021
46031f9a
RBJ		 4022static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio)
4023{
4024 /* We may not be able to submit a whole bio at once as there
4025 * may not be enough stripe_heads available.
4026 * We cannot pre-allocate enough stripe_heads as we may need
4027 * more than exist in the cache (if we allow ever large chunks).
4028 * So we do one stripe head at a time and record in
4029 * ->bi_hw_segments how many have been done.
4030 *
4031 * We *know* that this entire raid_bio is in one chunk, so
4032 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
4033 */
4034 struct stripe_head *sh;
911d4ee8 4035 int dd_idx;
46031f9a
RBJ		 4036 sector_t sector, logical_sector, last_sector;
4037 int scnt = 0;
4038 int remaining;
4039 int handled = 0;
4040
4041 logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
112bf897 4042 sector = raid5_compute_sector(conf, logical_sector,
911d4ee8 4043 0, &dd_idx, NULL);
46031f9a
RBJ		 4044 last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9);
4045
4046 for (; logical_sector < last_sector;
387bb173
NB		 4047 logical_sector += STRIPE_SECTORS,
4048 sector += STRIPE_SECTORS,
4049 scnt++) {
46031f9a 4050
960e739d 4051 if (scnt < raid5_bi_hw_segments(raid_bio))
46031f9a
RBJ		 4052 /* already done this stripe */
4053 continue;
4054
a8c906ca 4055 sh = get_active_stripe(conf, sector, 0, 1, 0);
46031f9a
RBJ		 4056
4057 if (!sh) {
4058 /* failed to get a stripe - must wait */
960e739d 4059 raid5_set_bi_hw_segments(raid_bio, scnt);
46031f9a
RBJ		 4060 conf->retry_read_aligned = raid_bio;
4061 return handled;
4062 }
4063
4064 set_bit(R5_ReadError, &sh->dev[dd_idx].flags);
387bb173
NB		 4065 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) {
4066 release_stripe(sh);
960e739d 4067 raid5_set_bi_hw_segments(raid_bio, scnt);
387bb173
NB		 4068 conf->retry_read_aligned = raid_bio;
4069 return handled;
4070 }
4071
36d1c647 4072 handle_stripe(sh);
46031f9a
RBJ		 4073 release_stripe(sh);
4074 handled++;
4075 }
4076 spin_lock_irq(&conf->device_lock);
960e739d 4077 remaining = raid5_dec_bi_phys_segments(raid_bio);
46031f9a 4078 spin_unlock_irq(&conf->device_lock);
0e13fe23
NB		 4079 if (remaining == 0)
4080 bio_endio(raid_bio, 0);
46031f9a
RBJ		 4081 if (atomic_dec_and_test(&conf->active_aligned_reads))
4082 wake_up(&conf->wait_for_stripe);
4083 return handled;
4084}
4085
46031f9a 4086
1da177e4
LT		 4087/*
4088 * This is our raid5 kernel thread.
4089 *
4090 * We scan the hash table for stripes which can be handled now.
4091 * During the scan, completed stripes are saved for us by the interrupt
4092 * handler, so that they will not have to wait for our next wakeup.
4093 */
6ed3003c 4094static void raid5d(mddev_t *mddev)
1da177e4
LT		 4095{
4096 struct stripe_head *sh;
070ec55d 4097 raid5_conf_t *conf = mddev->private;
1da177e4 4098 int handled;
e1dfa0a2 4099 struct blk_plug plug;
1da177e4 4100
45b4233c 4101 pr_debug("+++ raid5d active\n");
1da177e4
LT		 4102
4103 md_check_recovery(mddev);
1da177e4 4104
e1dfa0a2 4105 blk_start_plug(&plug);
1da177e4
LT		 4106 handled = 0;
4107 spin_lock_irq(&conf->device_lock);
4108 while (1) {
46031f9a 4109 struct bio *bio;
1da177e4 4110
7c13edc8
N		 4111 if (atomic_read(&mddev->plug_cnt) == 0 &&
4112 !list_empty(&conf->bitmap_list)) {
4113 /* Now is a good time to flush some bitmap updates */
4114 conf->seq_flush++;
700e432d 4115 spin_unlock_irq(&conf->device_lock);
72626685 4116 bitmap_unplug(mddev->bitmap);
700e432d 4117 spin_lock_irq(&conf->device_lock);
7c13edc8 4118 conf->seq_write = conf->seq_flush;
72626685
N		 4119 activate_bit_delay(conf);
4120 }
7c13edc8
N		 4121 if (atomic_read(&mddev->plug_cnt) == 0)
4122 raid5_activate_delayed(conf);
72626685 4123
46031f9a
RBJ		 4124 while ((bio = remove_bio_from_retry(conf))) {
4125 int ok;
4126 spin_unlock_irq(&conf->device_lock);
4127 ok = retry_aligned_read(conf, bio);
4128 spin_lock_irq(&conf->device_lock);
4129 if (!ok)
4130 break;
4131 handled++;
4132 }
4133
8b3e6cdc
DW		 4134 sh = __get_priority_stripe(conf);
4135
c9f21aaf 4136 if (!sh)
1da177e4 4137 break;
1da177e4
LT		 4138 spin_unlock_irq(&conf->device_lock);
4139
4140 handled++;
417b8d4a
DW		 4141 handle_stripe(sh);
4142 release_stripe(sh);
4143 cond_resched();
1da177e4
LT		 4144
4145 spin_lock_irq(&conf->device_lock);
4146 }
45b4233c 4147 pr_debug("%d stripes handled\n", handled);
1da177e4
LT		 4148
4149 spin_unlock_irq(&conf->device_lock);
4150
c9f21aaf 4151 async_tx_issue_pending_all();
e1dfa0a2 4152 blk_finish_plug(&plug);
1da177e4 4153
45b4233c 4154 pr_debug("--- raid5d inactive\n");
1da177e4
LT		 4155}
4156
3f294f4f 4157static ssize_t
007583c9 4158raid5_show_stripe_cache_size(mddev_t *mddev, char *page)
3f294f4f 4159{
070ec55d 4160 raid5_conf_t *conf = mddev->private;
96de1e66
N		 4161 if (conf)
4162 return sprintf(page, "%d\n", conf->max_nr_stripes);
4163 else
4164 return 0;
3f294f4f
N		 4165}
4166
c41d4ac4
N		 4167int
4168raid5_set_cache_size(mddev_t *mddev, int size)
3f294f4f 4169{
070ec55d 4170 raid5_conf_t *conf = mddev->private;
b5470dc5
DW		 4171 int err;
4172
c41d4ac4 4173 if (size <= 16 || size > 32768)
3f294f4f 4174 return -EINVAL;
c41d4ac4 4175 while (size < conf->max_nr_stripes) {
3f294f4f
N		 4176 if (drop_one_stripe(conf))
4177 conf->max_nr_stripes--;
4178 else
4179 break;
4180 }
b5470dc5
DW		 4181 err = md_allow_write(mddev);
4182 if (err)
4183 return err;
c41d4ac4 4184 while (size > conf->max_nr_stripes) {
3f294f4f
N		 4185 if (grow_one_stripe(conf))
4186 conf->max_nr_stripes++;
4187 else break;
4188 }
c41d4ac4
N		 4189 return 0;
4190}
4191EXPORT_SYMBOL(raid5_set_cache_size);
4192
4193static ssize_t
4194raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len)
4195{
4196 raid5_conf_t *conf = mddev->private;
4197 unsigned long new;
4198 int err;
4199
4200 if (len >= PAGE_SIZE)
4201 return -EINVAL;
4202 if (!conf)
4203 return -ENODEV;
4204
4205 if (strict_strtoul(page, 10, &new))
4206 return -EINVAL;
4207 err = raid5_set_cache_size(mddev, new);
4208 if (err)
4209 return err;
3f294f4f
N		 4210 return len;
4211}
007583c9 4212
96de1e66
N		 4213static struct md_sysfs_entry
4214raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
4215 raid5_show_stripe_cache_size,
4216 raid5_store_stripe_cache_size);
3f294f4f 4217
8b3e6cdc
DW		 4218static ssize_t
4219raid5_show_preread_threshold(mddev_t *mddev, char *page)
4220{
070ec55d 4221 raid5_conf_t *conf = mddev->private;
8b3e6cdc
DW		 4222 if (conf)
4223 return sprintf(page, "%d\n", conf->bypass_threshold);
4224 else
4225 return 0;
4226}
4227
4228static ssize_t
4229raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len)
4230{
070ec55d 4231 raid5_conf_t *conf = mddev->private;
4ef197d8 4232 unsigned long new;
8b3e6cdc
DW		 4233 if (len >= PAGE_SIZE)
4234 return -EINVAL;
4235 if (!conf)
4236 return -ENODEV;
4237
4ef197d8 4238 if (strict_strtoul(page, 10, &new))
8b3e6cdc 4239 return -EINVAL;
4ef197d8 4240 if (new > conf->max_nr_stripes)
8b3e6cdc
DW		 4241 return -EINVAL;
4242 conf->bypass_threshold = new;
4243 return len;
4244}
4245
4246static struct md_sysfs_entry
4247raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
4248 S_IRUGO | S_IWUSR,
4249 raid5_show_preread_threshold,
4250 raid5_store_preread_threshold);
4251
3f294f4f 4252static ssize_t
96de1e66 4253stripe_cache_active_show(mddev_t *mddev, char *page)
3f294f4f 4254{
070ec55d 4255 raid5_conf_t *conf = mddev->private;
96de1e66
N		 4256 if (conf)
4257 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
4258 else
4259 return 0;
3f294f4f
N		 4260}
4261
96de1e66
N		 4262static struct md_sysfs_entry
4263raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
3f294f4f 4264
007583c9 4265static struct attribute *raid5_attrs[] = {
3f294f4f
N		 4266 &raid5_stripecache_size.attr,
4267 &raid5_stripecache_active.attr,
8b3e6cdc 4268 &raid5_preread_bypass_threshold.attr,
3f294f4f
N		 4269 NULL,
4270};
007583c9
N		 4271static struct attribute_group raid5_attrs_group = {
4272 .name = NULL,
4273 .attrs = raid5_attrs,
3f294f4f
N		 4274};
4275
80c3a6ce
DW		 4276static sector_t
4277raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks)
4278{
070ec55d 4279 raid5_conf_t *conf = mddev->private;
80c3a6ce
DW		 4280
4281 if (!sectors)
4282 sectors = mddev->dev_sectors;
5e5e3e78 4283 if (!raid_disks)
7ec05478 4284 /* size is defined by the smallest of previous and new size */
5e5e3e78 4285 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
80c3a6ce 4286
9d8f0363 4287 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
664e7c41 4288 sectors &= ~((sector_t)mddev->new_chunk_sectors - 1);
80c3a6ce
DW		 4289 return sectors * (raid_disks - conf->max_degraded);
4290}
4291
36d1c647
DW		 4292static void raid5_free_percpu(raid5_conf_t *conf)
4293{
4294 struct raid5_percpu *percpu;
4295 unsigned long cpu;
4296
4297 if (!conf->percpu)
4298 return;
4299
4300 get_online_cpus();
4301 for_each_possible_cpu(cpu) {
4302 percpu = per_cpu_ptr(conf->percpu, cpu);
4303 safe_put_page(percpu->spare_page);
d6f38f31 4304 kfree(percpu->scribble);
36d1c647
DW		 4305 }
4306#ifdef CONFIG_HOTPLUG_CPU
4307 unregister_cpu_notifier(&conf->cpu_notify);
4308#endif
4309 put_online_cpus();
4310
4311 free_percpu(conf->percpu);
4312}
4313
95fc17aa
DW		 4314static void free_conf(raid5_conf_t *conf)
4315{
4316 shrink_stripes(conf);
36d1c647 4317 raid5_free_percpu(conf);
95fc17aa
DW		 4318 kfree(conf->disks);
4319 kfree(conf->stripe_hashtbl);
4320 kfree(conf);
4321}
4322
36d1c647
DW		 4323#ifdef CONFIG_HOTPLUG_CPU
4324static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action,
4325 void *hcpu)
4326{
4327 raid5_conf_t *conf = container_of(nfb, raid5_conf_t, cpu_notify);
4328 long cpu = (long)hcpu;
4329 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
4330
4331 switch (action) {
4332 case CPU_UP_PREPARE:
4333 case CPU_UP_PREPARE_FROZEN:
d6f38f31 4334 if (conf->level == 6 && !percpu->spare_page)
36d1c647 4335 percpu->spare_page = alloc_page(GFP_KERNEL);
d6f38f31
DW		 4336 if (!percpu->scribble)
4337 percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
4338
4339 if (!percpu->scribble ||
4340 (conf->level == 6 && !percpu->spare_page)) {
4341 safe_put_page(percpu->spare_page);
4342 kfree(percpu->scribble);
36d1c647
DW		 4343 pr_err("%s: failed memory allocation for cpu%ld\n",
4344 __func__, cpu);
55af6bb5 4345 return notifier_from_errno(-ENOMEM);
36d1c647
DW		 4346 }
4347 break;
4348 case CPU_DEAD:
4349 case CPU_DEAD_FROZEN:
4350 safe_put_page(percpu->spare_page);
d6f38f31 4351 kfree(percpu->scribble);
36d1c647 4352 percpu->spare_page = NULL;
d6f38f31 4353 percpu->scribble = NULL;
36d1c647
DW		 4354 break;
4355 default:
4356 break;
4357 }
4358 return NOTIFY_OK;
4359}
4360#endif
4361
4362static int raid5_alloc_percpu(raid5_conf_t *conf)
4363{
4364 unsigned long cpu;
4365 struct page *spare_page;
a29d8b8e 4366 struct raid5_percpu __percpu *allcpus;
d6f38f31 4367 void *scribble;
36d1c647
DW		 4368 int err;
4369
36d1c647
DW		 4370 allcpus = alloc_percpu(struct raid5_percpu);
4371 if (!allcpus)
4372 return -ENOMEM;
4373 conf->percpu = allcpus;
4374
4375 get_online_cpus();
4376 err = 0;
4377 for_each_present_cpu(cpu) {
d6f38f31
DW		 4378 if (conf->level == 6) {
4379 spare_page = alloc_page(GFP_KERNEL);
4380 if (!spare_page) {
4381 err = -ENOMEM;
4382 break;
4383 }
4384 per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page;
4385 }
5e5e3e78 4386 scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
d6f38f31 4387 if (!scribble) {
36d1c647
DW		 4388 err = -ENOMEM;
4389 break;
4390 }
d6f38f31 4391 per_cpu_ptr(conf->percpu, cpu)->scribble = scribble;
36d1c647
DW		 4392 }
4393#ifdef CONFIG_HOTPLUG_CPU
4394 conf->cpu_notify.notifier_call = raid456_cpu_notify;
4395 conf->cpu_notify.priority = 0;
4396 if (err == 0)
4397 err = register_cpu_notifier(&conf->cpu_notify);
4398#endif
4399 put_online_cpus();
4400
4401 return err;
4402}
4403
91adb564 4404static raid5_conf_t *setup_conf(mddev_t *mddev)
1da177e4
LT		 4405{
4406 raid5_conf_t *conf;
5e5e3e78 4407 int raid_disk, memory, max_disks;
1da177e4
LT		 4408 mdk_rdev_t *rdev;
4409 struct disk_info *disk;
1da177e4 4410
91adb564
N		 4411 if (mddev->new_level != 5
4412 && mddev->new_level != 4
4413 && mddev->new_level != 6) {
0c55e022 4414 printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n",
91adb564
N		 4415 mdname(mddev), mddev->new_level);
4416 return ERR_PTR(-EIO);
1da177e4 4417 }
91adb564
N		 4418 if ((mddev->new_level == 5
4419 && !algorithm_valid_raid5(mddev->new_layout)) ||
4420 (mddev->new_level == 6
4421 && !algorithm_valid_raid6(mddev->new_layout))) {
0c55e022 4422 printk(KERN_ERR "md/raid:%s: layout %d not supported\n",
91adb564
N		 4423 mdname(mddev), mddev->new_layout);
4424 return ERR_PTR(-EIO);
99c0fb5f 4425 }
91adb564 4426 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
0c55e022 4427 printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n",
91adb564
N		 4428 mdname(mddev), mddev->raid_disks);
4429 return ERR_PTR(-EINVAL);
4bbf3771
N		 4430 }
4431
664e7c41
AN		 4432 if (!mddev->new_chunk_sectors ||
4433 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
4434 !is_power_of_2(mddev->new_chunk_sectors)) {
0c55e022
N		 4435 printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n",
4436 mdname(mddev), mddev->new_chunk_sectors << 9);
91adb564 4437 return ERR_PTR(-EINVAL);
f6705578
N		 4438 }
4439
91adb564
N		 4440 conf = kzalloc(sizeof(raid5_conf_t), GFP_KERNEL);
4441 if (conf == NULL)
1da177e4 4442 goto abort;
f5efd45a
DW		 4443 spin_lock_init(&conf->device_lock);
4444 init_waitqueue_head(&conf->wait_for_stripe);
4445 init_waitqueue_head(&conf->wait_for_overlap);
4446 INIT_LIST_HEAD(&conf->handle_list);
4447 INIT_LIST_HEAD(&conf->hold_list);
4448 INIT_LIST_HEAD(&conf->delayed_list);
4449 INIT_LIST_HEAD(&conf->bitmap_list);
4450 INIT_LIST_HEAD(&conf->inactive_list);
4451 atomic_set(&conf->active_stripes, 0);
4452 atomic_set(&conf->preread_active_stripes, 0);
4453 atomic_set(&conf->active_aligned_reads, 0);
4454 conf->bypass_threshold = BYPASS_THRESHOLD;
91adb564
N		 4455
4456 conf->raid_disks = mddev->raid_disks;
4457 if (mddev->reshape_position == MaxSector)
4458 conf->previous_raid_disks = mddev->raid_disks;
4459 else
f6705578 4460 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
5e5e3e78
N		 4461 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
4462 conf->scribble_len = scribble_len(max_disks);
f6705578 4463
5e5e3e78 4464 conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
b55e6bfc
N		 4465 GFP_KERNEL);
4466 if (!conf->disks)
4467 goto abort;
9ffae0cf 4468
1da177e4
LT		 4469 conf->mddev = mddev;
4470
fccddba0 4471 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
1da177e4 4472 goto abort;
1da177e4 4473
36d1c647
DW		 4474 conf->level = mddev->new_level;
4475 if (raid5_alloc_percpu(conf) != 0)
4476 goto abort;
4477
0c55e022 4478 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
1da177e4 4479
159ec1fc 4480 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4 4481 raid_disk = rdev->raid_disk;
5e5e3e78 4482 if (raid_disk >= max_disks
1da177e4
LT		 4483 || raid_disk < 0)
4484 continue;
4485 disk = conf->disks + raid_disk;
4486
4487 disk->rdev = rdev;
4488
b2d444d7 4489 if (test_bit(In_sync, &rdev->flags)) {
1da177e4 4490 char b[BDEVNAME_SIZE];
0c55e022
N		 4491 printk(KERN_INFO "md/raid:%s: device %s operational as raid"
4492 " disk %d\n",
4493 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
d6b212f4 4494 } else if (rdev->saved_raid_disk != raid_disk)
8c2e870a
NB		 4495 /* Cannot rely on bitmap to complete recovery */
4496 conf->fullsync = 1;
1da177e4
LT		 4497 }
4498
09c9e5fa 4499 conf->chunk_sectors = mddev->new_chunk_sectors;
91adb564 4500 conf->level = mddev->new_level;
16a53ecc
N		 4501 if (conf->level == 6)
4502 conf->max_degraded = 2;
4503 else
4504 conf->max_degraded = 1;
91adb564 4505 conf->algorithm = mddev->new_layout;
1da177e4 4506 conf->max_nr_stripes = NR_STRIPES;
fef9c61f 4507 conf->reshape_progress = mddev->reshape_position;
e183eaed 4508 if (conf->reshape_progress != MaxSector) {
09c9e5fa 4509 conf->prev_chunk_sectors = mddev->chunk_sectors;
e183eaed
N		 4510 conf->prev_algo = mddev->layout;
4511 }
1da177e4 4512
91adb564 4513 memory = conf->max_nr_stripes * (sizeof(struct stripe_head) +
5e5e3e78 4514 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
91adb564
N		 4515 if (grow_stripes(conf, conf->max_nr_stripes)) {
4516 printk(KERN_ERR
0c55e022
N		 4517 "md/raid:%s: couldn't allocate %dkB for buffers\n",
4518 mdname(mddev), memory);
91adb564
N		 4519 goto abort;
4520 } else
0c55e022
N		 4521 printk(KERN_INFO "md/raid:%s: allocated %dkB\n",
4522 mdname(mddev), memory);
1da177e4 4523
0da3c619 4524 conf->thread = md_register_thread(raid5d, mddev, NULL);
91adb564
N		 4525 if (!conf->thread) {
4526 printk(KERN_ERR
0c55e022 4527 "md/raid:%s: couldn't allocate thread.\n",
91adb564 4528 mdname(mddev));
16a53ecc
N		 4529 goto abort;
4530 }
91adb564
N		 4531
4532 return conf;
4533
4534 abort:
4535 if (conf) {
95fc17aa 4536 free_conf(conf);
91adb564
N		 4537 return ERR_PTR(-EIO);
4538 } else
4539 return ERR_PTR(-ENOMEM);
4540}
4541
c148ffdc
N		 4542
4543static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
4544{
4545 switch (algo) {
4546 case ALGORITHM_PARITY_0:
4547 if (raid_disk < max_degraded)
4548 return 1;
4549 break;
4550 case ALGORITHM_PARITY_N:
4551 if (raid_disk >= raid_disks - max_degraded)
4552 return 1;
4553 break;
4554 case ALGORITHM_PARITY_0_6:
4555 if (raid_disk == 0 ||
4556 raid_disk == raid_disks - 1)
4557 return 1;
4558 break;
4559 case ALGORITHM_LEFT_ASYMMETRIC_6:
4560 case ALGORITHM_RIGHT_ASYMMETRIC_6:
4561 case ALGORITHM_LEFT_SYMMETRIC_6:
4562 case ALGORITHM_RIGHT_SYMMETRIC_6:
4563 if (raid_disk == raid_disks - 1)
4564 return 1;
4565 }
4566 return 0;
4567}
4568
91adb564
N		 4569static int run(mddev_t *mddev)
4570{
4571 raid5_conf_t *conf;
9f7c2220 4572 int working_disks = 0;
c148ffdc 4573 int dirty_parity_disks = 0;
91adb564 4574 mdk_rdev_t *rdev;
c148ffdc 4575 sector_t reshape_offset = 0;
91adb564 4576
8c6ac868 4577 if (mddev->recovery_cp != MaxSector)
0c55e022 4578 printk(KERN_NOTICE "md/raid:%s: not clean"
8c6ac868
AN		 4579 " -- starting background reconstruction\n",
4580 mdname(mddev));
91adb564
N		 4581 if (mddev->reshape_position != MaxSector) {
4582 /* Check that we can continue the reshape.
4583 * Currently only disks can change, it must
4584 * increase, and we must be past the point where
4585 * a stripe over-writes itself
4586 */
4587 sector_t here_new, here_old;
4588 int old_disks;
18b00334 4589 int max_degraded = (mddev->level == 6 ? 2 : 1);
91adb564 4590
88ce4930 4591 if (mddev->new_level != mddev->level) {
0c55e022 4592 printk(KERN_ERR "md/raid:%s: unsupported reshape "
91adb564
N		 4593 "required - aborting.\n",
4594 mdname(mddev));
4595 return -EINVAL;
4596 }
91adb564
N		 4597 old_disks = mddev->raid_disks - mddev->delta_disks;
4598 /* reshape_position must be on a new-stripe boundary, and one
4599 * further up in new geometry must map after here in old
4600 * geometry.
4601 */
4602 here_new = mddev->reshape_position;
664e7c41 4603 if (sector_div(here_new, mddev->new_chunk_sectors *
91adb564 4604 (mddev->raid_disks - max_degraded))) {
0c55e022
N		 4605 printk(KERN_ERR "md/raid:%s: reshape_position not "
4606 "on a stripe boundary\n", mdname(mddev));
91adb564
N		 4607 return -EINVAL;
4608 }
c148ffdc 4609 reshape_offset = here_new * mddev->new_chunk_sectors;
91adb564
N		 4610 /* here_new is the stripe we will write to */
4611 here_old = mddev->reshape_position;
9d8f0363 4612 sector_div(here_old, mddev->chunk_sectors *
91adb564
N		 4613 (old_disks-max_degraded));
4614 /* here_old is the first stripe that we might need to read
4615 * from */
67ac6011
N		 4616 if (mddev->delta_disks == 0) {
4617 /* We cannot be sure it is safe to start an in-place
4618 * reshape. It is only safe if user-space if monitoring
4619 * and taking constant backups.
4620 * mdadm always starts a situation like this in
4621 * readonly mode so it can take control before
4622 * allowing any writes. So just check for that.
4623 */
4624 if ((here_new * mddev->new_chunk_sectors !=
4625 here_old * mddev->chunk_sectors) ||
4626 mddev->ro == 0) {
0c55e022
N		 4627 printk(KERN_ERR "md/raid:%s: in-place reshape must be started"
4628 " in read-only mode - aborting\n",
4629 mdname(mddev));
67ac6011
N		 4630 return -EINVAL;
4631 }
4632 } else if (mddev->delta_disks < 0
4633 ? (here_new * mddev->new_chunk_sectors <=
4634 here_old * mddev->chunk_sectors)
4635 : (here_new * mddev->new_chunk_sectors >=
4636 here_old * mddev->chunk_sectors)) {
91adb564 4637 /* Reading from the same stripe as writing to - bad */
0c55e022
N		 4638 printk(KERN_ERR "md/raid:%s: reshape_position too early for "
4639 "auto-recovery - aborting.\n",
4640 mdname(mddev));
91adb564
N		 4641 return -EINVAL;
4642 }
0c55e022
N		 4643 printk(KERN_INFO "md/raid:%s: reshape will continue\n",
4644 mdname(mddev));
91adb564
N		 4645 /* OK, we should be able to continue; */
4646 } else {
4647 BUG_ON(mddev->level != mddev->new_level);
4648 BUG_ON(mddev->layout != mddev->new_layout);
664e7c41 4649 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
91adb564 4650 BUG_ON(mddev->delta_disks != 0);
1da177e4 4651 }
91adb564 4652
245f46c2
N		 4653 if (mddev->private == NULL)
4654 conf = setup_conf(mddev);
4655 else
4656 conf = mddev->private;
4657
91adb564
N		 4658 if (IS_ERR(conf))
4659 return PTR_ERR(conf);
4660
4661 mddev->thread = conf->thread;
4662 conf->thread = NULL;
4663 mddev->private = conf;
4664
4665 /*
4666 * 0 for a fully functional array, 1 or 2 for a degraded array.
4667 */
c148ffdc
N		 4668 list_for_each_entry(rdev, &mddev->disks, same_set) {
4669 if (rdev->raid_disk < 0)
4670 continue;
2f115882 4671 if (test_bit(In_sync, &rdev->flags)) {
91adb564 4672 working_disks++;
2f115882
N		 4673 continue;
4674 }
c148ffdc
N		 4675 /* This disc is not fully in-sync. However if it
4676 * just stored parity (beyond the recovery_offset),
4677 * when we don't need to be concerned about the
4678 * array being dirty.
4679 * When reshape goes 'backwards', we never have
4680 * partially completed devices, so we only need
4681 * to worry about reshape going forwards.
4682 */
4683 /* Hack because v0.91 doesn't store recovery_offset properly. */
4684 if (mddev->major_version == 0 &&
4685 mddev->minor_version > 90)
4686 rdev->recovery_offset = reshape_offset;
4687
c148ffdc
N		 4688 if (rdev->recovery_offset < reshape_offset) {
4689 /* We need to check old and new layout */
4690 if (!only_parity(rdev->raid_disk,
4691 conf->algorithm,
4692 conf->raid_disks,
4693 conf->max_degraded))
4694 continue;
4695 }
4696 if (!only_parity(rdev->raid_disk,
4697 conf->prev_algo,
4698 conf->previous_raid_disks,
4699 conf->max_degraded))
4700 continue;
4701 dirty_parity_disks++;
4702 }
91adb564 4703
5e5e3e78
N		 4704 mddev->degraded = (max(conf->raid_disks, conf->previous_raid_disks)
4705 - working_disks);
91adb564 4706
674806d6 4707 if (has_failed(conf)) {
0c55e022 4708 printk(KERN_ERR "md/raid:%s: not enough operational devices"
1da177e4 4709 " (%d/%d failed)\n",
02c2de8c 4710 mdname(mddev), mddev->degraded, conf->raid_disks);
1da177e4
LT		 4711 goto abort;
4712 }
4713
91adb564 4714 /* device size must be a multiple of chunk size */
9d8f0363 4715 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
91adb564
N		 4716 mddev->resync_max_sectors = mddev->dev_sectors;
4717
c148ffdc 4718 if (mddev->degraded > dirty_parity_disks &&
1da177e4 4719 mddev->recovery_cp != MaxSector) {
6ff8d8ec
N		 4720 if (mddev->ok_start_degraded)
4721 printk(KERN_WARNING
0c55e022
N		 4722 "md/raid:%s: starting dirty degraded array"
4723 " - data corruption possible.\n",
6ff8d8ec
N		 4724 mdname(mddev));
4725 else {
4726 printk(KERN_ERR
0c55e022 4727 "md/raid:%s: cannot start dirty degraded array.\n",
6ff8d8ec
N		 4728 mdname(mddev));
4729 goto abort;
4730 }
1da177e4
LT		 4731 }
4732
1da177e4 4733 if (mddev->degraded == 0)
0c55e022
N		 4734 printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d"
4735 " devices, algorithm %d\n", mdname(mddev), conf->level,
e183eaed
N		 4736 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
4737 mddev->new_layout);
1da177e4 4738 else
0c55e022
N		 4739 printk(KERN_ALERT "md/raid:%s: raid level %d active with %d"
4740 " out of %d devices, algorithm %d\n",
4741 mdname(mddev), conf->level,
4742 mddev->raid_disks - mddev->degraded,
4743 mddev->raid_disks, mddev->new_layout);
1da177e4
LT		 4744
4745 print_raid5_conf(conf);
4746
fef9c61f 4747 if (conf->reshape_progress != MaxSector) {
fef9c61f 4748 conf->reshape_safe = conf->reshape_progress;
f6705578
N		 4749 atomic_set(&conf->reshape_stripes, 0);
4750 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4751 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4752 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
4753 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4754 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 4755 "reshape");
f6705578
N		 4756 }
4757
1da177e4
LT		 4758
4759 /* Ok, everything is just fine now */
a64c876f
N		 4760 if (mddev->to_remove == &raid5_attrs_group)
4761 mddev->to_remove = NULL;
00bcb4ac
N		 4762 else if (mddev->kobj.sd &&
4763 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
5e55e2f5 4764 printk(KERN_WARNING
4a5add49 4765 "raid5: failed to create sysfs attributes for %s\n",
5e55e2f5 4766 mdname(mddev));
4a5add49 4767 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
7a5febe9 4768
4a5add49 4769 if (mddev->queue) {
9f7c2220 4770 int chunk_size;
4a5add49
N		 4771 /* read-ahead size must cover two whole stripes, which
4772 * is 2 * (datadisks) * chunksize where 'n' is the
4773 * number of raid devices
4774 */
4775 int data_disks = conf->previous_raid_disks - conf->max_degraded;
4776 int stripe = data_disks *
4777 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
4778 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
4779 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
91adb564 4780
4a5add49 4781 blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec);
f022b2fd 4782
11d8a6e3
N		 4783 mddev->queue->backing_dev_info.congested_data = mddev;
4784 mddev->queue->backing_dev_info.congested_fn = raid5_congested;
7a5febe9 4785
9f7c2220
N		 4786 chunk_size = mddev->chunk_sectors << 9;
4787 blk_queue_io_min(mddev->queue, chunk_size);
4788 blk_queue_io_opt(mddev->queue, chunk_size *
4789 (conf->raid_disks - conf->max_degraded));
8f6c2e4b 4790
9f7c2220
N		 4791 list_for_each_entry(rdev, &mddev->disks, same_set)
4792 disk_stack_limits(mddev->gendisk, rdev->bdev,
4793 rdev->data_offset << 9);
4794 }
23032a0e 4795
1da177e4
LT		 4796 return 0;
4797abort:
e0cf8f04 4798 md_unregister_thread(mddev->thread);
91adb564 4799 mddev->thread = NULL;
1da177e4
LT		 4800 if (conf) {
4801 print_raid5_conf(conf);
95fc17aa 4802 free_conf(conf);
1da177e4
LT		 4803 }
4804 mddev->private = NULL;
0c55e022 4805 printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev));
1da177e4
LT		 4806 return -EIO;
4807}
4808
3f294f4f 4809static int stop(mddev_t *mddev)
1da177e4 4810{
7b92813c 4811 raid5_conf_t *conf = mddev->private;
1da177e4
LT		 4812
4813 md_unregister_thread(mddev->thread);
4814 mddev->thread = NULL;
11d8a6e3
N		 4815 if (mddev->queue)
4816 mddev->queue->backing_dev_info.congested_fn = NULL;
95fc17aa 4817 free_conf(conf);
a64c876f
N		 4818 mddev->private = NULL;
4819 mddev->to_remove = &raid5_attrs_group;
1da177e4
LT		 4820 return 0;
4821}
4822
45b4233c 4823#ifdef DEBUG
d710e138 4824static void print_sh(struct seq_file *seq, struct stripe_head *sh)
1da177e4
LT		 4825{
4826 int i;
4827
16a53ecc
N		 4828 seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n",
4829 (unsigned long long)sh->sector, sh->pd_idx, sh->state);
4830 seq_printf(seq, "sh %llu, count %d.\n",
4831 (unsigned long long)sh->sector, atomic_read(&sh->count));
4832 seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector);
7ecaa1e6 4833 for (i = 0; i < sh->disks; i++) {
16a53ecc
N		 4834 seq_printf(seq, "(cache%d: %p %ld) ",
4835 i, sh->dev[i].page, sh->dev[i].flags);
1da177e4 4836 }
16a53ecc 4837 seq_printf(seq, "\n");
1da177e4
LT		 4838}
4839
d710e138 4840static void printall(struct seq_file *seq, raid5_conf_t *conf)
1da177e4
LT		 4841{
4842 struct stripe_head *sh;
fccddba0 4843 struct hlist_node *hn;
1da177e4
LT		 4844 int i;
4845
4846 spin_lock_irq(&conf->device_lock);
4847 for (i = 0; i < NR_HASH; i++) {
fccddba0 4848 hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) {
1da177e4
LT		 4849 if (sh->raid_conf != conf)
4850 continue;
16a53ecc 4851 print_sh(seq, sh);
1da177e4
LT		 4852 }
4853 }
4854 spin_unlock_irq(&conf->device_lock);
4855}
4856#endif
4857
d710e138 4858static void status(struct seq_file *seq, mddev_t *mddev)
1da177e4 4859{
7b92813c 4860 raid5_conf_t *conf = mddev->private;
1da177e4
LT		 4861 int i;
4862
9d8f0363
AN		 4863 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
4864 mddev->chunk_sectors / 2, mddev->layout);
02c2de8c 4865 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
1da177e4
LT		 4866 for (i = 0; i < conf->raid_disks; i++)
4867 seq_printf (seq, "%s",
4868 conf->disks[i].rdev &&
b2d444d7 4869 test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_");
1da177e4 4870 seq_printf (seq, "]");
45b4233c 4871#ifdef DEBUG
16a53ecc
N		 4872 seq_printf (seq, "\n");
4873 printall(seq, conf);
1da177e4
LT		 4874#endif
4875}
4876
4877static void print_raid5_conf (raid5_conf_t *conf)
4878{
4879 int i;
4880 struct disk_info *tmp;
4881
0c55e022 4882 printk(KERN_DEBUG "RAID conf printout:\n");
1da177e4
LT		 4883 if (!conf) {
4884 printk("(conf==NULL)\n");
4885 return;
4886 }
0c55e022
N		 4887 printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level,
4888 conf->raid_disks,
4889 conf->raid_disks - conf->mddev->degraded);
1da177e4
LT		 4890
4891 for (i = 0; i < conf->raid_disks; i++) {
4892 char b[BDEVNAME_SIZE];
4893 tmp = conf->disks + i;
4894 if (tmp->rdev)
0c55e022
N		 4895 printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n",
4896 i, !test_bit(Faulty, &tmp->rdev->flags),
4897 bdevname(tmp->rdev->bdev, b));
1da177e4
LT		 4898 }
4899}
4900
4901static int raid5_spare_active(mddev_t *mddev)
4902{
4903 int i;
4904 raid5_conf_t *conf = mddev->private;
4905 struct disk_info *tmp;
6b965620
N		 4906 int count = 0;
4907 unsigned long flags;
1da177e4
LT		 4908
4909 for (i = 0; i < conf->raid_disks; i++) {
4910 tmp = conf->disks + i;
4911 if (tmp->rdev
70fffd0b 4912 && tmp->rdev->recovery_offset == MaxSector
b2d444d7 4913 && !test_bit(Faulty, &tmp->rdev->flags)
c04be0aa 4914 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
6b965620 4915 count++;
43c73ca4 4916 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
1da177e4
LT		 4917 }
4918 }
6b965620
N		 4919 spin_lock_irqsave(&conf->device_lock, flags);
4920 mddev->degraded -= count;
4921 spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4 4922 print_raid5_conf(conf);
6b965620 4923 return count;
1da177e4
LT		 4924}
4925
4926static int raid5_remove_disk(mddev_t *mddev, int number)
4927{
4928 raid5_conf_t *conf = mddev->private;
4929 int err = 0;
4930 mdk_rdev_t *rdev;
4931 struct disk_info *p = conf->disks + number;
4932
4933 print_raid5_conf(conf);
4934 rdev = p->rdev;
4935 if (rdev) {
ec32a2bd
N		 4936 if (number >= conf->raid_disks &&
4937 conf->reshape_progress == MaxSector)
4938 clear_bit(In_sync, &rdev->flags);
4939
b2d444d7 4940 if (test_bit(In_sync, &rdev->flags) ||
1da177e4
LT		 4941 atomic_read(&rdev->nr_pending)) {
4942 err = -EBUSY;
4943 goto abort;
4944 }
dfc70645
N		 4945 /* Only remove non-faulty devices if recovery
4946 * isn't possible.
4947 */
4948 if (!test_bit(Faulty, &rdev->flags) &&
674806d6 4949 !has_failed(conf) &&
ec32a2bd 4950 number < conf->raid_disks) {
dfc70645
N		 4951 err = -EBUSY;
4952 goto abort;
4953 }
1da177e4 4954 p->rdev = NULL;
fbd568a3 4955 synchronize_rcu();
1da177e4
LT		 4956 if (atomic_read(&rdev->nr_pending)) {
4957 /* lost the race, try later */
4958 err = -EBUSY;
4959 p->rdev = rdev;
4960 }
4961 }
4962abort:
4963
4964 print_raid5_conf(conf);
4965 return err;
4966}
4967
4968static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
4969{
4970 raid5_conf_t *conf = mddev->private;
199050ea 4971 int err = -EEXIST;
1da177e4
LT		 4972 int disk;
4973 struct disk_info *p;
6c2fce2e
NB		 4974 int first = 0;
4975 int last = conf->raid_disks - 1;
1da177e4 4976
674806d6 4977 if (has_failed(conf))
1da177e4 4978 /* no point adding a device */
199050ea 4979 return -EINVAL;
1da177e4 4980
6c2fce2e
NB		 4981 if (rdev->raid_disk >= 0)
4982 first = last = rdev->raid_disk;
1da177e4
LT		 4983
4984 /*
16a53ecc
N		 4985 * find the disk ... but prefer rdev->saved_raid_disk
4986 * if possible.
1da177e4 4987 */
16a53ecc 4988 if (rdev->saved_raid_disk >= 0 &&
6c2fce2e 4989 rdev->saved_raid_disk >= first &&
16a53ecc
N		 4990 conf->disks[rdev->saved_raid_disk].rdev == NULL)
4991 disk = rdev->saved_raid_disk;
4992 else
6c2fce2e
NB		 4993 disk = first;
4994 for ( ; disk <= last ; disk++)
1da177e4 4995 if ((p=conf->disks + disk)->rdev == NULL) {
b2d444d7 4996 clear_bit(In_sync, &rdev->flags);
1da177e4 4997 rdev->raid_disk = disk;
199050ea 4998 err = 0;
72626685
N		 4999 if (rdev->saved_raid_disk != disk)
5000 conf->fullsync = 1;
d6065f7b 5001 rcu_assign_pointer(p->rdev, rdev);
1da177e4
LT		 5002 break;
5003 }
5004 print_raid5_conf(conf);
199050ea 5005 return err;
1da177e4
LT		 5006}
5007
5008static int raid5_resize(mddev_t *mddev, sector_t sectors)
5009{
5010 /* no resync is happening, and there is enough space
5011 * on all devices, so we can resize.
5012 * We need to make sure resync covers any new space.
5013 * If the array is shrinking we should possibly wait until
5014 * any io in the removed space completes, but it hardly seems
5015 * worth it.
5016 */
9d8f0363 5017 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
1f403624
DW		 5018 md_set_array_sectors(mddev, raid5_size(mddev, sectors,
5019 mddev->raid_disks));
b522adcd
DW		 5020 if (mddev->array_sectors >
5021 raid5_size(mddev, sectors, mddev->raid_disks))
5022 return -EINVAL;
f233ea5c 5023 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 5024 revalidate_disk(mddev->gendisk);
b098636c
N		 5025 if (sectors > mddev->dev_sectors &&
5026 mddev->recovery_cp > mddev->dev_sectors) {
58c0fed4 5027 mddev->recovery_cp = mddev->dev_sectors;
1da177e4
LT		 5028 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5029 }
58c0fed4 5030 mddev->dev_sectors = sectors;
4b5c7ae8 5031 mddev->resync_max_sectors = sectors;
1da177e4
LT		 5032 return 0;
5033}
5034
01ee22b4
N		 5035static int check_stripe_cache(mddev_t *mddev)
5036{
5037 /* Can only proceed if there are plenty of stripe_heads.
5038 * We need a minimum of one full stripe,, and for sensible progress
5039 * it is best to have about 4 times that.
5040 * If we require 4 times, then the default 256 4K stripe_heads will
5041 * allow for chunk sizes up to 256K, which is probably OK.
5042 * If the chunk size is greater, user-space should request more
5043 * stripe_heads first.
5044 */
5045 raid5_conf_t *conf = mddev->private;
5046 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
5047 > conf->max_nr_stripes ||
5048 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
5049 > conf->max_nr_stripes) {
0c55e022
N		 5050 printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n",
5051 mdname(mddev),
01ee22b4
N		 5052 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
5053 / STRIPE_SIZE)*4);
5054 return 0;
5055 }
5056 return 1;
5057}
5058
50ac168a 5059static int check_reshape(mddev_t *mddev)
29269553 5060{
070ec55d 5061 raid5_conf_t *conf = mddev->private;
29269553 5062
88ce4930
N		 5063 if (mddev->delta_disks == 0 &&
5064 mddev->new_layout == mddev->layout &&
664e7c41 5065 mddev->new_chunk_sectors == mddev->chunk_sectors)
50ac168a 5066 return 0; /* nothing to do */
dba034ee
N		 5067 if (mddev->bitmap)
5068 /* Cannot grow a bitmap yet */
5069 return -EBUSY;
674806d6 5070 if (has_failed(conf))
ec32a2bd
N		 5071 return -EINVAL;
5072 if (mddev->delta_disks < 0) {
5073 /* We might be able to shrink, but the devices must
5074 * be made bigger first.
5075 * For raid6, 4 is the minimum size.
5076 * Otherwise 2 is the minimum
5077 */
5078 int min = 2;
5079 if (mddev->level == 6)
5080 min = 4;
5081 if (mddev->raid_disks + mddev->delta_disks < min)
5082 return -EINVAL;
5083 }
29269553 5084
01ee22b4 5085 if (!check_stripe_cache(mddev))
29269553 5086 return -ENOSPC;
29269553 5087
ec32a2bd 5088 return resize_stripes(conf, conf->raid_disks + mddev->delta_disks);
63c70c4f
N		 5089}
5090
5091static int raid5_start_reshape(mddev_t *mddev)
5092{
070ec55d 5093 raid5_conf_t *conf = mddev->private;
63c70c4f 5094 mdk_rdev_t *rdev;
63c70c4f 5095 int spares = 0;
c04be0aa 5096 unsigned long flags;
63c70c4f 5097
f416885e 5098 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
63c70c4f
N		 5099 return -EBUSY;
5100
01ee22b4
N		 5101 if (!check_stripe_cache(mddev))
5102 return -ENOSPC;
5103
159ec1fc 5104 list_for_each_entry(rdev, &mddev->disks, same_set)
469518a3
N		 5105 if (!test_bit(In_sync, &rdev->flags)
5106 && !test_bit(Faulty, &rdev->flags))
29269553 5107 spares++;
63c70c4f 5108
f416885e 5109 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
29269553
N		 5110 /* Not enough devices even to make a degraded array
5111 * of that size
5112 */
5113 return -EINVAL;
5114
ec32a2bd
N		 5115 /* Refuse to reduce size of the array. Any reductions in
5116 * array size must be through explicit setting of array_size
5117 * attribute.
5118 */
5119 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
5120 < mddev->array_sectors) {
0c55e022 5121 printk(KERN_ERR "md/raid:%s: array size must be reduced "
ec32a2bd
N		 5122 "before number of disks\n", mdname(mddev));
5123 return -EINVAL;
5124 }
5125
f6705578 5126 atomic_set(&conf->reshape_stripes, 0);
29269553
N		 5127 spin_lock_irq(&conf->device_lock);
5128 conf->previous_raid_disks = conf->raid_disks;
63c70c4f 5129 conf->raid_disks += mddev->delta_disks;
09c9e5fa
AN		 5130 conf->prev_chunk_sectors = conf->chunk_sectors;
5131 conf->chunk_sectors = mddev->new_chunk_sectors;
88ce4930
N		 5132 conf->prev_algo = conf->algorithm;
5133 conf->algorithm = mddev->new_layout;
fef9c61f
N		 5134 if (mddev->delta_disks < 0)
5135 conf->reshape_progress = raid5_size(mddev, 0, 0);
5136 else
5137 conf->reshape_progress = 0;
5138 conf->reshape_safe = conf->reshape_progress;
86b42c71 5139 conf->generation++;
29269553
N		 5140 spin_unlock_irq(&conf->device_lock);
5141
5142 /* Add some new drives, as many as will fit.
5143 * We know there are enough to make the newly sized array work.
3424bf6a
N		 5144 * Don't add devices if we are reducing the number of
5145 * devices in the array. This is because it is not possible
5146 * to correctly record the "partially reconstructed" state of
5147 * such devices during the reshape and confusion could result.
29269553 5148 */
87a8dec9
N		 5149 if (mddev->delta_disks >= 0) {
5150 int added_devices = 0;
5151 list_for_each_entry(rdev, &mddev->disks, same_set)
5152 if (rdev->raid_disk < 0 &&
5153 !test_bit(Faulty, &rdev->flags)) {
5154 if (raid5_add_disk(mddev, rdev) == 0) {
87a8dec9
N		 5155 if (rdev->raid_disk
5156 >= conf->previous_raid_disks) {
5157 set_bit(In_sync, &rdev->flags);
5158 added_devices++;
5159 } else
5160 rdev->recovery_offset = 0;
36fad858
NK		 5161
5162 if (sysfs_link_rdev(mddev, rdev))
87a8dec9 5163 /* Failure here is OK */;
50da0840 5164 }
87a8dec9
N		 5165 } else if (rdev->raid_disk >= conf->previous_raid_disks
5166 && !test_bit(Faulty, &rdev->flags)) {
5167 /* This is a spare that was manually added */
5168 set_bit(In_sync, &rdev->flags);
5169 added_devices++;
5170 }
29269553 5171
87a8dec9
N		 5172 /* When a reshape changes the number of devices,
5173 * ->degraded is measured against the larger of the
5174 * pre and post number of devices.
5175 */
ec32a2bd 5176 spin_lock_irqsave(&conf->device_lock, flags);
9eb07c25 5177 mddev->degraded += (conf->raid_disks - conf->previous_raid_disks)
ec32a2bd
N		 5178 - added_devices;
5179 spin_unlock_irqrestore(&conf->device_lock, flags);
5180 }
63c70c4f 5181 mddev->raid_disks = conf->raid_disks;
e516402c 5182 mddev->reshape_position = conf->reshape_progress;
850b2b42 5183 set_bit(MD_CHANGE_DEVS, &mddev->flags);
f6705578 5184
29269553
N		 5185 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5186 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5187 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5188 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5189 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
0da3c619 5190 "reshape");
29269553
N		 5191 if (!mddev->sync_thread) {
5192 mddev->recovery = 0;
5193 spin_lock_irq(&conf->device_lock);
5194 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
fef9c61f 5195 conf->reshape_progress = MaxSector;
29269553
N		 5196 spin_unlock_irq(&conf->device_lock);
5197 return -EAGAIN;
5198 }
c8f517c4 5199 conf->reshape_checkpoint = jiffies;
29269553
N		 5200 md_wakeup_thread(mddev->sync_thread);
5201 md_new_event(mddev);
5202 return 0;
5203}
29269553 5204
ec32a2bd
N		 5205/* This is called from the reshape thread and should make any
5206 * changes needed in 'conf'
5207 */
29269553
N		 5208static void end_reshape(raid5_conf_t *conf)
5209{
29269553 5210
f6705578 5211 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
f6705578 5212
f6705578 5213 spin_lock_irq(&conf->device_lock);
cea9c228 5214 conf->previous_raid_disks = conf->raid_disks;
fef9c61f 5215 conf->reshape_progress = MaxSector;
f6705578 5216 spin_unlock_irq(&conf->device_lock);
b0f9ec04 5217 wake_up(&conf->wait_for_overlap);
16a53ecc
N		 5218
5219 /* read-ahead size must cover two whole stripes, which is
5220 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
5221 */
4a5add49 5222 if (conf->mddev->queue) {
cea9c228 5223 int data_disks = conf->raid_disks - conf->max_degraded;
09c9e5fa 5224 int stripe = data_disks * ((conf->chunk_sectors << 9)
cea9c228 5225 / PAGE_SIZE);
16a53ecc
N		 5226 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
5227 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
5228 }
29269553 5229 }
29269553
N		 5230}
5231
ec32a2bd
N		 5232/* This is called from the raid5d thread with mddev_lock held.
5233 * It makes config changes to the device.
5234 */
cea9c228
N		 5235static void raid5_finish_reshape(mddev_t *mddev)
5236{
070ec55d 5237 raid5_conf_t *conf = mddev->private;
cea9c228
N		 5238
5239 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5240
ec32a2bd
N		 5241 if (mddev->delta_disks > 0) {
5242 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
5243 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 5244 revalidate_disk(mddev->gendisk);
ec32a2bd
N		 5245 } else {
5246 int d;
ec32a2bd
N		 5247 mddev->degraded = conf->raid_disks;
5248 for (d = 0; d < conf->raid_disks ; d++)
5249 if (conf->disks[d].rdev &&
5250 test_bit(In_sync,
5251 &conf->disks[d].rdev->flags))
5252 mddev->degraded--;
5253 for (d = conf->raid_disks ;
5254 d < conf->raid_disks - mddev->delta_disks;
1a67dde0
N		 5255 d++) {
5256 mdk_rdev_t *rdev = conf->disks[d].rdev;
5257 if (rdev && raid5_remove_disk(mddev, d) == 0) {
36fad858 5258 sysfs_unlink_rdev(mddev, rdev);
1a67dde0
N		 5259 rdev->raid_disk = -1;
5260 }
5261 }
cea9c228 5262 }
88ce4930 5263 mddev->layout = conf->algorithm;
09c9e5fa 5264 mddev->chunk_sectors = conf->chunk_sectors;
ec32a2bd
N		 5265 mddev->reshape_position = MaxSector;
5266 mddev->delta_disks = 0;
cea9c228
N		 5267 }
5268}
5269
72626685
N		 5270static void raid5_quiesce(mddev_t *mddev, int state)
5271{
070ec55d 5272 raid5_conf_t *conf = mddev->private;
72626685
N		 5273
5274 switch(state) {
e464eafd
N		 5275 case 2: /* resume for a suspend */
5276 wake_up(&conf->wait_for_overlap);
5277 break;
5278
72626685
N		 5279 case 1: /* stop all writes */
5280 spin_lock_irq(&conf->device_lock);
64bd660b
N		 5281 /* '2' tells resync/reshape to pause so that all
5282 * active stripes can drain
5283 */
5284 conf->quiesce = 2;
72626685 5285 wait_event_lock_irq(conf->wait_for_stripe,
46031f9a
RBJ		 5286 atomic_read(&conf->active_stripes) == 0 &&
5287 atomic_read(&conf->active_aligned_reads) == 0,
72626685 5288 conf->device_lock, /* nothing */);
64bd660b 5289 conf->quiesce = 1;
72626685 5290 spin_unlock_irq(&conf->device_lock);
64bd660b
N		 5291 /* allow reshape to continue */
5292 wake_up(&conf->wait_for_overlap);
72626685
N		 5293 break;
5294
5295 case 0: /* re-enable writes */
5296 spin_lock_irq(&conf->device_lock);
5297 conf->quiesce = 0;
5298 wake_up(&conf->wait_for_stripe);
e464eafd 5299 wake_up(&conf->wait_for_overlap);
72626685
N		 5300 spin_unlock_irq(&conf->device_lock);
5301 break;
5302 }
72626685 5303}
b15c2e57 5304
d562b0c4 5305
f1b29bca 5306static void *raid45_takeover_raid0(mddev_t *mddev, int level)
54071b38 5307{
f1b29bca 5308 struct raid0_private_data *raid0_priv = mddev->private;
d76c8420 5309 sector_t sectors;
54071b38 5310
f1b29bca
DW		 5311 /* for raid0 takeover only one zone is supported */
5312 if (raid0_priv->nr_strip_zones > 1) {
0c55e022
N		 5313 printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n",
5314 mdname(mddev));
f1b29bca
DW		 5315 return ERR_PTR(-EINVAL);
5316 }
5317
3b71bd93
N		 5318 sectors = raid0_priv->strip_zone[0].zone_end;
5319 sector_div(sectors, raid0_priv->strip_zone[0].nb_dev);
5320 mddev->dev_sectors = sectors;
f1b29bca 5321 mddev->new_level = level;
54071b38
TM		 5322 mddev->new_layout = ALGORITHM_PARITY_N;
5323 mddev->new_chunk_sectors = mddev->chunk_sectors;
5324 mddev->raid_disks += 1;
5325 mddev->delta_disks = 1;
5326 /* make sure it will be not marked as dirty */
5327 mddev->recovery_cp = MaxSector;
5328
5329 return setup_conf(mddev);
5330}
5331
5332
d562b0c4
N		 5333static void *raid5_takeover_raid1(mddev_t *mddev)
5334{
5335 int chunksect;
5336
5337 if (mddev->raid_disks != 2 ||
5338 mddev->degraded > 1)
5339 return ERR_PTR(-EINVAL);
5340
5341 /* Should check if there are write-behind devices? */
5342
5343 chunksect = 64*2; /* 64K by default */
5344
5345 /* The array must be an exact multiple of chunksize */
5346 while (chunksect && (mddev->array_sectors & (chunksect-1)))
5347 chunksect >>= 1;
5348
5349 if ((chunksect<<9) < STRIPE_SIZE)
5350 /* array size does not allow a suitable chunk size */
5351 return ERR_PTR(-EINVAL);
5352
5353 mddev->new_level = 5;
5354 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
664e7c41 5355 mddev->new_chunk_sectors = chunksect;
d562b0c4
N		 5356
5357 return setup_conf(mddev);
5358}
5359
fc9739c6
N		 5360static void *raid5_takeover_raid6(mddev_t *mddev)
5361{
5362 int new_layout;
5363
5364 switch (mddev->layout) {
5365 case ALGORITHM_LEFT_ASYMMETRIC_6:
5366 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
5367 break;
5368 case ALGORITHM_RIGHT_ASYMMETRIC_6:
5369 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
5370 break;
5371 case ALGORITHM_LEFT_SYMMETRIC_6:
5372 new_layout = ALGORITHM_LEFT_SYMMETRIC;
5373 break;
5374 case ALGORITHM_RIGHT_SYMMETRIC_6:
5375 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
5376 break;
5377 case ALGORITHM_PARITY_0_6:
5378 new_layout = ALGORITHM_PARITY_0;
5379 break;
5380 case ALGORITHM_PARITY_N:
5381 new_layout = ALGORITHM_PARITY_N;
5382 break;
5383 default:
5384 return ERR_PTR(-EINVAL);
5385 }
5386 mddev->new_level = 5;
5387 mddev->new_layout = new_layout;
5388 mddev->delta_disks = -1;
5389 mddev->raid_disks -= 1;
5390 return setup_conf(mddev);
5391}
5392
d562b0c4 5393
50ac168a 5394static int raid5_check_reshape(mddev_t *mddev)
b3546035 5395{
88ce4930
N		 5396 /* For a 2-drive array, the layout and chunk size can be changed
5397 * immediately as not restriping is needed.
5398 * For larger arrays we record the new value - after validation
5399 * to be used by a reshape pass.
b3546035 5400 */
070ec55d 5401 raid5_conf_t *conf = mddev->private;
597a711b 5402 int new_chunk = mddev->new_chunk_sectors;
b3546035 5403
597a711b 5404 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
b3546035
N		 5405 return -EINVAL;
5406 if (new_chunk > 0) {
0ba459d2 5407 if (!is_power_of_2(new_chunk))
b3546035 5408 return -EINVAL;
597a711b 5409 if (new_chunk < (PAGE_SIZE>>9))
b3546035 5410 return -EINVAL;
597a711b 5411 if (mddev->array_sectors & (new_chunk-1))
b3546035
N		 5412 /* not factor of array size */
5413 return -EINVAL;
5414 }
5415
5416 /* They look valid */
5417
88ce4930 5418 if (mddev->raid_disks == 2) {
597a711b
N		 5419 /* can make the change immediately */
5420 if (mddev->new_layout >= 0) {
5421 conf->algorithm = mddev->new_layout;
5422 mddev->layout = mddev->new_layout;
88ce4930
N		 5423 }
5424 if (new_chunk > 0) {
597a711b
N		 5425 conf->chunk_sectors = new_chunk ;
5426 mddev->chunk_sectors = new_chunk;
88ce4930
N		 5427 }
5428 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5429 md_wakeup_thread(mddev->thread);
b3546035 5430 }
50ac168a 5431 return check_reshape(mddev);
88ce4930
N		 5432}
5433
50ac168a 5434static int raid6_check_reshape(mddev_t *mddev)
88ce4930 5435{
597a711b 5436 int new_chunk = mddev->new_chunk_sectors;
50ac168a 5437
597a711b 5438 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
88ce4930 5439 return -EINVAL;
b3546035 5440 if (new_chunk > 0) {
0ba459d2 5441 if (!is_power_of_2(new_chunk))
88ce4930 5442 return -EINVAL;
597a711b 5443 if (new_chunk < (PAGE_SIZE >> 9))
88ce4930 5444 return -EINVAL;
597a711b 5445 if (mddev->array_sectors & (new_chunk-1))
88ce4930
N		 5446 /* not factor of array size */
5447 return -EINVAL;
b3546035 5448 }
88ce4930
N		 5449
5450 /* They look valid */
50ac168a 5451 return check_reshape(mddev);
b3546035
N		 5452}
5453
d562b0c4
N		 5454static void *raid5_takeover(mddev_t *mddev)
5455{
5456 /* raid5 can take over:
f1b29bca 5457 * raid0 - if there is only one strip zone - make it a raid4 layout
d562b0c4
N		 5458 * raid1 - if there are two drives. We need to know the chunk size
5459 * raid4 - trivial - just use a raid4 layout.
5460 * raid6 - Providing it is a *_6 layout
d562b0c4 5461 */
f1b29bca
DW		 5462 if (mddev->level == 0)
5463 return raid45_takeover_raid0(mddev, 5);
d562b0c4
N		 5464 if (mddev->level == 1)
5465 return raid5_takeover_raid1(mddev);
e9d4758f
N		 5466 if (mddev->level == 4) {
5467 mddev->new_layout = ALGORITHM_PARITY_N;
5468 mddev->new_level = 5;
5469 return setup_conf(mddev);
5470 }
fc9739c6
N		 5471 if (mddev->level == 6)
5472 return raid5_takeover_raid6(mddev);
d562b0c4
N		 5473
5474 return ERR_PTR(-EINVAL);
5475}
5476
a78d38a1
N		 5477static void *raid4_takeover(mddev_t *mddev)
5478{
f1b29bca
DW		 5479 /* raid4 can take over:
5480 * raid0 - if there is only one strip zone
5481 * raid5 - if layout is right
a78d38a1 5482 */
f1b29bca
DW		 5483 if (mddev->level == 0)
5484 return raid45_takeover_raid0(mddev, 4);
a78d38a1
N		 5485 if (mddev->level == 5 &&
5486 mddev->layout == ALGORITHM_PARITY_N) {
5487 mddev->new_layout = 0;
5488 mddev->new_level = 4;
5489 return setup_conf(mddev);
5490 }
5491 return ERR_PTR(-EINVAL);
5492}
d562b0c4 5493
245f46c2
N		 5494static struct mdk_personality raid5_personality;
5495
5496static void *raid6_takeover(mddev_t *mddev)
5497{
5498 /* Currently can only take over a raid5. We map the
5499 * personality to an equivalent raid6 personality
5500 * with the Q block at the end.
5501 */
5502 int new_layout;
5503
5504 if (mddev->pers != &raid5_personality)
5505 return ERR_PTR(-EINVAL);
5506 if (mddev->degraded > 1)
5507 return ERR_PTR(-EINVAL);
5508 if (mddev->raid_disks > 253)
5509 return ERR_PTR(-EINVAL);
5510 if (mddev->raid_disks < 3)
5511 return ERR_PTR(-EINVAL);
5512
5513 switch (mddev->layout) {
5514 case ALGORITHM_LEFT_ASYMMETRIC:
5515 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
5516 break;
5517 case ALGORITHM_RIGHT_ASYMMETRIC:
5518 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
5519 break;
5520 case ALGORITHM_LEFT_SYMMETRIC:
5521 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
5522 break;
5523 case ALGORITHM_RIGHT_SYMMETRIC:
5524 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
5525 break;
5526 case ALGORITHM_PARITY_0:
5527 new_layout = ALGORITHM_PARITY_0_6;
5528 break;
5529 case ALGORITHM_PARITY_N:
5530 new_layout = ALGORITHM_PARITY_N;
5531 break;
5532 default:
5533 return ERR_PTR(-EINVAL);
5534 }
5535 mddev->new_level = 6;
5536 mddev->new_layout = new_layout;
5537 mddev->delta_disks = 1;
5538 mddev->raid_disks += 1;
5539 return setup_conf(mddev);
5540}
5541
5542
16a53ecc
N		 5543static struct mdk_personality raid6_personality =
5544{
5545 .name = "raid6",
5546 .level = 6,
5547 .owner = THIS_MODULE,
5548 .make_request = make_request,
5549 .run = run,
5550 .stop = stop,
5551 .status = status,
5552 .error_handler = error,
5553 .hot_add_disk = raid5_add_disk,
5554 .hot_remove_disk= raid5_remove_disk,
5555 .spare_active = raid5_spare_active,
5556 .sync_request = sync_request,
5557 .resize = raid5_resize,
80c3a6ce 5558 .size = raid5_size,
50ac168a 5559 .check_reshape = raid6_check_reshape,
f416885e 5560 .start_reshape = raid5_start_reshape,
cea9c228 5561 .finish_reshape = raid5_finish_reshape,
16a53ecc 5562 .quiesce = raid5_quiesce,
245f46c2 5563 .takeover = raid6_takeover,
16a53ecc 5564};
2604b703 5565static struct mdk_personality raid5_personality =
1da177e4
LT		 5566{
5567 .name = "raid5",
2604b703 5568 .level = 5,
1da177e4
LT		 5569 .owner = THIS_MODULE,
5570 .make_request = make_request,
5571 .run = run,
5572 .stop = stop,
5573 .status = status,
5574 .error_handler = error,
5575 .hot_add_disk = raid5_add_disk,
5576 .hot_remove_disk= raid5_remove_disk,
5577 .spare_active = raid5_spare_active,
5578 .sync_request = sync_request,
5579 .resize = raid5_resize,
80c3a6ce 5580 .size = raid5_size,
63c70c4f
N		 5581 .check_reshape = raid5_check_reshape,
5582 .start_reshape = raid5_start_reshape,
cea9c228 5583 .finish_reshape = raid5_finish_reshape,
72626685 5584 .quiesce = raid5_quiesce,
d562b0c4 5585 .takeover = raid5_takeover,
1da177e4
LT		 5586};
5587
2604b703 5588static struct mdk_personality raid4_personality =
1da177e4 5589{
2604b703
N		 5590 .name = "raid4",
5591 .level = 4,
5592 .owner = THIS_MODULE,
5593 .make_request = make_request,
5594 .run = run,
5595 .stop = stop,
5596 .status = status,
5597 .error_handler = error,
5598 .hot_add_disk = raid5_add_disk,
5599 .hot_remove_disk= raid5_remove_disk,
5600 .spare_active = raid5_spare_active,
5601 .sync_request = sync_request,
5602 .resize = raid5_resize,
80c3a6ce 5603 .size = raid5_size,
3d37890b
N		 5604 .check_reshape = raid5_check_reshape,
5605 .start_reshape = raid5_start_reshape,
cea9c228 5606 .finish_reshape = raid5_finish_reshape,
2604b703 5607 .quiesce = raid5_quiesce,
a78d38a1 5608 .takeover = raid4_takeover,
2604b703
N		 5609};
5610
5611static int __init raid5_init(void)
5612{
16a53ecc 5613 register_md_personality(&raid6_personality);
2604b703
N		 5614 register_md_personality(&raid5_personality);
5615 register_md_personality(&raid4_personality);
5616 return 0;
1da177e4
LT		 5617}
5618
2604b703 5619static void raid5_exit(void)
1da177e4 5620{
16a53ecc 5621 unregister_md_personality(&raid6_personality);
2604b703
N		 5622 unregister_md_personality(&raid5_personality);
5623 unregister_md_personality(&raid4_personality);
1da177e4
LT		 5624}
5625
5626module_init(raid5_init);
5627module_exit(raid5_exit);
5628MODULE_LICENSE("GPL");
0efb9e61 5629MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
1da177e4 5630MODULE_ALIAS("md-personality-4"); /* RAID5 */
d9d166c2
N		 5631MODULE_ALIAS("md-raid5");
5632MODULE_ALIAS("md-raid4");
2604b703
N		 5633MODULE_ALIAS("md-level-5");
5634MODULE_ALIAS("md-level-4");
16a53ecc
N		 5635MODULE_ALIAS("md-personality-8"); /* RAID6 */
5636MODULE_ALIAS("md-raid6");
5637MODULE_ALIAS("md-level-6");
5638
5639/* This used to be two separate modules, they were: */
5640MODULE_ALIAS("raid5");
5641MODULE_ALIAS("raid6");
Linux 6.x block layer and io_uring tree(s)