projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
block: fix disk->part[] dereferencing race
[deliverable/linux.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.
30 * conf->bm_write is the number of the last batch successfully written.
31 * conf->bm_flush is the number of the last batch that was closed to
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
35 * the number of the batch it will be in. This is bm_flush+1.
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
1da177e4
LT		 46#include <linux/module.h>
47#include <linux/slab.h>
1da177e4
LT		 48#include <linux/highmem.h>
49#include <linux/bitops.h>
f6705578 50#include <linux/kthread.h>
1da177e4 51#include <asm/atomic.h>
16a53ecc 52#include "raid6.h"
1da177e4 53
72626685 54#include <linux/raid/bitmap.h>
91c00924 55#include <linux/async_tx.h>
72626685 56
1da177e4
LT		 57/*
58 * Stripe cache
59 */
60
61#define NR_STRIPES 256
62#define STRIPE_SIZE PAGE_SIZE
63#define STRIPE_SHIFT (PAGE_SHIFT - 9)
64#define STRIPE_SECTORS (STRIPE_SIZE>>9)
65#define IO_THRESHOLD 1
8b3e6cdc 66#define BYPASS_THRESHOLD 1
fccddba0 67#define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
1da177e4
LT		 68#define HASH_MASK (NR_HASH - 1)
69
fccddba0 70#define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK]))
1da177e4
LT		 71
72/* bio's attached to a stripe+device for I/O are linked together in bi_sector
73 * order without overlap. There may be several bio's per stripe+device, and
74 * a bio could span several devices.
75 * When walking this list for a particular stripe+device, we must never proceed
76 * beyond a bio that extends past this device, as the next bio might no longer
77 * be valid.
78 * This macro is used to determine the 'next' bio in the list, given the sector
79 * of the current stripe+device
80 */
81#define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL)
82/*
83 * The following can be used to debug the driver
84 */
1da177e4
LT		 85#define RAID5_PARANOIA 1
86#if RAID5_PARANOIA && defined(CONFIG_SMP)
87# define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock)
88#else
89# define CHECK_DEVLOCK()
90#endif
91
45b4233c 92#ifdef DEBUG
1da177e4
LT		 93#define inline
94#define __inline__
95#endif
96
6be9d494
BS		 97#define printk_rl(args...) ((void) (printk_ratelimit() && printk(args)))
98
16a53ecc
N		 99#if !RAID6_USE_EMPTY_ZERO_PAGE
100/* In .bss so it's zeroed */
101const char raid6_empty_zero_page[PAGE_SIZE] __attribute__((aligned(256)));
102#endif
103
960e739d 104/*
5b99c2ff
JA		 105 * We maintain a biased count of active stripes in the bottom 16 bits of
106 * bi_phys_segments, and a count of processed stripes in the upper 16 bits
960e739d
JA		 107 */
108static inline int raid5_bi_phys_segments(struct bio *bio)
109{
5b99c2ff 110 return bio->bi_phys_segments & 0xffff;
960e739d
JA		 111}
112
113static inline int raid5_bi_hw_segments(struct bio *bio)
114{
5b99c2ff 115 return (bio->bi_phys_segments >> 16) & 0xffff;
960e739d
JA		 116}
117
118static inline int raid5_dec_bi_phys_segments(struct bio *bio)
119{
120 --bio->bi_phys_segments;
121 return raid5_bi_phys_segments(bio);
122}
123
124static inline int raid5_dec_bi_hw_segments(struct bio *bio)
125{
126 unsigned short val = raid5_bi_hw_segments(bio);
127
128 --val;
5b99c2ff 129 bio->bi_phys_segments = (val << 16) | raid5_bi_phys_segments(bio);
960e739d
JA		 130 return val;
131}
132
133static inline void raid5_set_bi_hw_segments(struct bio *bio, unsigned int cnt)
134{
5b99c2ff 135 bio->bi_phys_segments = raid5_bi_phys_segments(bio) || (cnt << 16);
960e739d
JA		 136}
137
16a53ecc
N		 138static inline int raid6_next_disk(int disk, int raid_disks)
139{
140 disk++;
141 return (disk < raid_disks) ? disk : 0;
142}
a4456856
DW		 143
144static void return_io(struct bio *return_bi)
145{
146 struct bio *bi = return_bi;
147 while (bi) {
a4456856
DW		 148
149 return_bi = bi->bi_next;
150 bi->bi_next = NULL;
151 bi->bi_size = 0;
0e13fe23 152 bio_endio(bi, 0);
a4456856
DW		 153 bi = return_bi;
154 }
155}
156
1da177e4
LT		 157static void print_raid5_conf (raid5_conf_t *conf);
158
600aa109
DW		 159static int stripe_operations_active(struct stripe_head *sh)
160{
161 return sh->check_state || sh->reconstruct_state ||
162 test_bit(STRIPE_BIOFILL_RUN, &sh->state) ||
163 test_bit(STRIPE_COMPUTE_RUN, &sh->state);
164}
165
858119e1 166static void __release_stripe(raid5_conf_t *conf, struct stripe_head *sh)
1da177e4
LT		 167{
168 if (atomic_dec_and_test(&sh->count)) {
78bafebd
ES		 169 BUG_ON(!list_empty(&sh->lru));
170 BUG_ON(atomic_read(&conf->active_stripes)==0);
1da177e4 171 if (test_bit(STRIPE_HANDLE, &sh->state)) {
7c785b7a 172 if (test_bit(STRIPE_DELAYED, &sh->state)) {
1da177e4 173 list_add_tail(&sh->lru, &conf->delayed_list);
7c785b7a
N		 174 blk_plug_device(conf->mddev->queue);
175 } else if (test_bit(STRIPE_BIT_DELAY, &sh->state) &&
ae3c20cc 176 sh->bm_seq - conf->seq_write > 0) {
72626685 177 list_add_tail(&sh->lru, &conf->bitmap_list);
7c785b7a
N		 178 blk_plug_device(conf->mddev->queue);
179 } else {
72626685 180 clear_bit(STRIPE_BIT_DELAY, &sh->state);
1da177e4 181 list_add_tail(&sh->lru, &conf->handle_list);
72626685 182 }
1da177e4
LT		 183 md_wakeup_thread(conf->mddev->thread);
184 } else {
600aa109 185 BUG_ON(stripe_operations_active(sh));
1da177e4
LT		 186 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
187 atomic_dec(&conf->preread_active_stripes);
188 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD)
189 md_wakeup_thread(conf->mddev->thread);
190 }
1da177e4 191 atomic_dec(&conf->active_stripes);
ccfcc3c1
N		 192 if (!test_bit(STRIPE_EXPANDING, &sh->state)) {
193 list_add_tail(&sh->lru, &conf->inactive_list);
1da177e4 194 wake_up(&conf->wait_for_stripe);
46031f9a
RBJ		 195 if (conf->retry_read_aligned)
196 md_wakeup_thread(conf->mddev->thread);
ccfcc3c1 197 }
1da177e4
LT		 198 }
199 }
200}
201static void release_stripe(struct stripe_head *sh)
202{
203 raid5_conf_t *conf = sh->raid_conf;
204 unsigned long flags;
16a53ecc 205
1da177e4
LT		 206 spin_lock_irqsave(&conf->device_lock, flags);
207 __release_stripe(conf, sh);
208 spin_unlock_irqrestore(&conf->device_lock, flags);
209}
210
fccddba0 211static inline void remove_hash(struct stripe_head *sh)
1da177e4 212{
45b4233c
DW		 213 pr_debug("remove_hash(), stripe %llu\n",
214 (unsigned long long)sh->sector);
1da177e4 215
fccddba0 216 hlist_del_init(&sh->hash);
1da177e4
LT		 217}
218
16a53ecc 219static inline void insert_hash(raid5_conf_t *conf, struct stripe_head *sh)
1da177e4 220{
fccddba0 221 struct hlist_head *hp = stripe_hash(conf, sh->sector);
1da177e4 222
45b4233c
DW		 223 pr_debug("insert_hash(), stripe %llu\n",
224 (unsigned long long)sh->sector);
1da177e4
LT		 225
226 CHECK_DEVLOCK();
fccddba0 227 hlist_add_head(&sh->hash, hp);
1da177e4
LT		 228}
229
230
231/* find an idle stripe, make sure it is unhashed, and return it. */
232static struct stripe_head *get_free_stripe(raid5_conf_t *conf)
233{
234 struct stripe_head *sh = NULL;
235 struct list_head *first;
236
237 CHECK_DEVLOCK();
238 if (list_empty(&conf->inactive_list))
239 goto out;
240 first = conf->inactive_list.next;
241 sh = list_entry(first, struct stripe_head, lru);
242 list_del_init(first);
243 remove_hash(sh);
244 atomic_inc(&conf->active_stripes);
245out:
246 return sh;
247}
248
249static void shrink_buffers(struct stripe_head *sh, int num)
250{
251 struct page *p;
252 int i;
253
254 for (i=0; i<num ; i++) {
255 p = sh->dev[i].page;
256 if (!p)
257 continue;
258 sh->dev[i].page = NULL;
2d1f3b5d 259 put_page(p);
1da177e4
LT		 260 }
261}
262
263static int grow_buffers(struct stripe_head *sh, int num)
264{
265 int i;
266
267 for (i=0; i<num; i++) {
268 struct page *page;
269
270 if (!(page = alloc_page(GFP_KERNEL))) {
271 return 1;
272 }
273 sh->dev[i].page = page;
274 }
275 return 0;
276}
277
278static void raid5_build_block (struct stripe_head *sh, int i);
279
7ecaa1e6 280static void init_stripe(struct stripe_head *sh, sector_t sector, int pd_idx, int disks)
1da177e4
LT		 281{
282 raid5_conf_t *conf = sh->raid_conf;
7ecaa1e6 283 int i;
1da177e4 284
78bafebd
ES		 285 BUG_ON(atomic_read(&sh->count) != 0);
286 BUG_ON(test_bit(STRIPE_HANDLE, &sh->state));
600aa109 287 BUG_ON(stripe_operations_active(sh));
d84e0f10 288
1da177e4 289 CHECK_DEVLOCK();
45b4233c 290 pr_debug("init_stripe called, stripe %llu\n",
1da177e4
LT		 291 (unsigned long long)sh->sector);
292
293 remove_hash(sh);
16a53ecc 294
1da177e4
LT		 295 sh->sector = sector;
296 sh->pd_idx = pd_idx;
297 sh->state = 0;
298
7ecaa1e6
N		 299 sh->disks = disks;
300
301 for (i = sh->disks; i--; ) {
1da177e4
LT		 302 struct r5dev *dev = &sh->dev[i];
303
d84e0f10 304 if (dev->toread || dev->read || dev->towrite || dev->written ||
1da177e4 305 test_bit(R5_LOCKED, &dev->flags)) {
d84e0f10 306 printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n",
1da177e4 307 (unsigned long long)sh->sector, i, dev->toread,
d84e0f10 308 dev->read, dev->towrite, dev->written,
1da177e4
LT		 309 test_bit(R5_LOCKED, &dev->flags));
310 BUG();
311 }
312 dev->flags = 0;
313 raid5_build_block(sh, i);
314 }
315 insert_hash(conf, sh);
316}
317
7ecaa1e6 318static struct stripe_head *__find_stripe(raid5_conf_t *conf, sector_t sector, int disks)
1da177e4
LT		 319{
320 struct stripe_head *sh;
fccddba0 321 struct hlist_node *hn;
1da177e4
LT		 322
323 CHECK_DEVLOCK();
45b4233c 324 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
fccddba0 325 hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash)
7ecaa1e6 326 if (sh->sector == sector && sh->disks == disks)
1da177e4 327 return sh;
45b4233c 328 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
1da177e4
LT		 329 return NULL;
330}
331
332static void unplug_slaves(mddev_t *mddev);
165125e1 333static void raid5_unplug_device(struct request_queue *q);
1da177e4 334
7ecaa1e6
N		 335static struct stripe_head *get_active_stripe(raid5_conf_t *conf, sector_t sector, int disks,
336 int pd_idx, int noblock)
1da177e4
LT		 337{
338 struct stripe_head *sh;
339
45b4233c 340 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector);
1da177e4
LT		 341
342 spin_lock_irq(&conf->device_lock);
343
344 do {
72626685
N		 345 wait_event_lock_irq(conf->wait_for_stripe,
346 conf->quiesce == 0,
347 conf->device_lock, /* nothing */);
7ecaa1e6 348 sh = __find_stripe(conf, sector, disks);
1da177e4
LT		 349 if (!sh) {
350 if (!conf->inactive_blocked)
351 sh = get_free_stripe(conf);
352 if (noblock && sh == NULL)
353 break;
354 if (!sh) {
355 conf->inactive_blocked = 1;
356 wait_event_lock_irq(conf->wait_for_stripe,
357 !list_empty(&conf->inactive_list) &&
5036805b
N		 358 (atomic_read(&conf->active_stripes)
359 < (conf->max_nr_stripes *3/4)
1da177e4
LT		 360 || !conf->inactive_blocked),
361 conf->device_lock,
f4370781 362 raid5_unplug_device(conf->mddev->queue)
1da177e4
LT		 363 );
364 conf->inactive_blocked = 0;
365 } else
7ecaa1e6 366 init_stripe(sh, sector, pd_idx, disks);
1da177e4
LT		 367 } else {
368 if (atomic_read(&sh->count)) {
78bafebd 369 BUG_ON(!list_empty(&sh->lru));
1da177e4
LT		 370 } else {
371 if (!test_bit(STRIPE_HANDLE, &sh->state))
372 atomic_inc(&conf->active_stripes);
ff4e8d9a
N		 373 if (list_empty(&sh->lru) &&
374 !test_bit(STRIPE_EXPANDING, &sh->state))
16a53ecc
N		 375 BUG();
376 list_del_init(&sh->lru);
1da177e4
LT		 377 }
378 }
379 } while (sh == NULL);
380
381 if (sh)
382 atomic_inc(&sh->count);
383
384 spin_unlock_irq(&conf->device_lock);
385 return sh;
386}
387
6712ecf8
N		 388static void
389raid5_end_read_request(struct bio *bi, int error);
390static void
391raid5_end_write_request(struct bio *bi, int error);
91c00924 392
c4e5ac0a 393static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
91c00924
DW		 394{
395 raid5_conf_t *conf = sh->raid_conf;
396 int i, disks = sh->disks;
397
398 might_sleep();
399
400 for (i = disks; i--; ) {
401 int rw;
402 struct bio *bi;
403 mdk_rdev_t *rdev;
404 if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags))
405 rw = WRITE;
406 else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags))
407 rw = READ;
408 else
409 continue;
410
411 bi = &sh->dev[i].req;
412
413 bi->bi_rw = rw;
414 if (rw == WRITE)
415 bi->bi_end_io = raid5_end_write_request;
416 else
417 bi->bi_end_io = raid5_end_read_request;
418
419 rcu_read_lock();
420 rdev = rcu_dereference(conf->disks[i].rdev);
421 if (rdev && test_bit(Faulty, &rdev->flags))
422 rdev = NULL;
423 if (rdev)
424 atomic_inc(&rdev->nr_pending);
425 rcu_read_unlock();
426
427 if (rdev) {
c4e5ac0a 428 if (s->syncing || s->expanding || s->expanded)
91c00924
DW		 429 md_sync_acct(rdev->bdev, STRIPE_SECTORS);
430
2b7497f0
DW		 431 set_bit(STRIPE_IO_STARTED, &sh->state);
432
91c00924
DW		 433 bi->bi_bdev = rdev->bdev;
434 pr_debug("%s: for %llu schedule op %ld on disc %d\n",
e46b272b 435 __func__, (unsigned long long)sh->sector,
91c00924
DW		 436 bi->bi_rw, i);
437 atomic_inc(&sh->count);
438 bi->bi_sector = sh->sector + rdev->data_offset;
439 bi->bi_flags = 1 << BIO_UPTODATE;
440 bi->bi_vcnt = 1;
441 bi->bi_max_vecs = 1;
442 bi->bi_idx = 0;
443 bi->bi_io_vec = &sh->dev[i].vec;
444 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
445 bi->bi_io_vec[0].bv_offset = 0;
446 bi->bi_size = STRIPE_SIZE;
447 bi->bi_next = NULL;
448 if (rw == WRITE &&
449 test_bit(R5_ReWrite, &sh->dev[i].flags))
450 atomic_add(STRIPE_SECTORS,
451 &rdev->corrected_errors);
452 generic_make_request(bi);
453 } else {
454 if (rw == WRITE)
455 set_bit(STRIPE_DEGRADED, &sh->state);
456 pr_debug("skip op %ld on disc %d for sector %llu\n",
457 bi->bi_rw, i, (unsigned long long)sh->sector);
458 clear_bit(R5_LOCKED, &sh->dev[i].flags);
459 set_bit(STRIPE_HANDLE, &sh->state);
460 }
461 }
462}
463
464static struct dma_async_tx_descriptor *
465async_copy_data(int frombio, struct bio *bio, struct page *page,
466 sector_t sector, struct dma_async_tx_descriptor *tx)
467{
468 struct bio_vec *bvl;
469 struct page *bio_page;
470 int i;
471 int page_offset;
472
473 if (bio->bi_sector >= sector)
474 page_offset = (signed)(bio->bi_sector - sector) * 512;
475 else
476 page_offset = (signed)(sector - bio->bi_sector) * -512;
477 bio_for_each_segment(bvl, bio, i) {
478 int len = bio_iovec_idx(bio, i)->bv_len;
479 int clen;
480 int b_offset = 0;
481
482 if (page_offset < 0) {
483 b_offset = -page_offset;
484 page_offset += b_offset;
485 len -= b_offset;
486 }
487
488 if (len > 0 && page_offset + len > STRIPE_SIZE)
489 clen = STRIPE_SIZE - page_offset;
490 else
491 clen = len;
492
493 if (clen > 0) {
494 b_offset += bio_iovec_idx(bio, i)->bv_offset;
495 bio_page = bio_iovec_idx(bio, i)->bv_page;
496 if (frombio)
497 tx = async_memcpy(page, bio_page, page_offset,
498 b_offset, clen,
eb0645a8 499 ASYNC_TX_DEP_ACK,
91c00924
DW		 500 tx, NULL, NULL);
501 else
502 tx = async_memcpy(bio_page, page, b_offset,
503 page_offset, clen,
eb0645a8 504 ASYNC_TX_DEP_ACK,
91c00924
DW		 505 tx, NULL, NULL);
506 }
507 if (clen < len) /* hit end of page */
508 break;
509 page_offset += len;
510 }
511
512 return tx;
513}
514
515static void ops_complete_biofill(void *stripe_head_ref)
516{
517 struct stripe_head *sh = stripe_head_ref;
518 struct bio *return_bi = NULL;
519 raid5_conf_t *conf = sh->raid_conf;
e4d84909 520 int i;
91c00924 521
e46b272b 522 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 523 (unsigned long long)sh->sector);
524
525 /* clear completed biofills */
83de75cc 526 spin_lock_irq(&conf->device_lock);
91c00924
DW		 527 for (i = sh->disks; i--; ) {
528 struct r5dev *dev = &sh->dev[i];
91c00924
DW		 529
530 /* acknowledge completion of a biofill operation */
e4d84909
DW		 531 /* and check if we need to reply to a read request,
532 * new R5_Wantfill requests are held off until
83de75cc 533 * !STRIPE_BIOFILL_RUN
e4d84909
DW		 534 */
535 if (test_and_clear_bit(R5_Wantfill, &dev->flags)) {
91c00924 536 struct bio *rbi, *rbi2;
91c00924 537
91c00924
DW		 538 BUG_ON(!dev->read);
539 rbi = dev->read;
540 dev->read = NULL;
541 while (rbi && rbi->bi_sector <
542 dev->sector + STRIPE_SECTORS) {
543 rbi2 = r5_next_bio(rbi, dev->sector);
960e739d 544 if (!raid5_dec_bi_phys_segments(rbi)) {
91c00924
DW		 545 rbi->bi_next = return_bi;
546 return_bi = rbi;
547 }
91c00924
DW		 548 rbi = rbi2;
549 }
550 }
551 }
83de75cc
DW		 552 spin_unlock_irq(&conf->device_lock);
553 clear_bit(STRIPE_BIOFILL_RUN, &sh->state);
91c00924
DW		 554
555 return_io(return_bi);
556
e4d84909 557 set_bit(STRIPE_HANDLE, &sh->state);
91c00924
DW		 558 release_stripe(sh);
559}
560
561static void ops_run_biofill(struct stripe_head *sh)
562{
563 struct dma_async_tx_descriptor *tx = NULL;
564 raid5_conf_t *conf = sh->raid_conf;
565 int i;
566
e46b272b 567 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 568 (unsigned long long)sh->sector);
569
570 for (i = sh->disks; i--; ) {
571 struct r5dev *dev = &sh->dev[i];
572 if (test_bit(R5_Wantfill, &dev->flags)) {
573 struct bio *rbi;
574 spin_lock_irq(&conf->device_lock);
575 dev->read = rbi = dev->toread;
576 dev->toread = NULL;
577 spin_unlock_irq(&conf->device_lock);
578 while (rbi && rbi->bi_sector <
579 dev->sector + STRIPE_SECTORS) {
580 tx = async_copy_data(0, rbi, dev->page,
581 dev->sector, tx);
582 rbi = r5_next_bio(rbi, dev->sector);
583 }
584 }
585 }
586
587 atomic_inc(&sh->count);
588 async_trigger_callback(ASYNC_TX_DEP_ACK | ASYNC_TX_ACK, tx,
589 ops_complete_biofill, sh);
590}
591
592static void ops_complete_compute5(void *stripe_head_ref)
593{
594 struct stripe_head *sh = stripe_head_ref;
595 int target = sh->ops.target;
596 struct r5dev *tgt = &sh->dev[target];
597
e46b272b 598 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 599 (unsigned long long)sh->sector);
600
601 set_bit(R5_UPTODATE, &tgt->flags);
602 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
603 clear_bit(R5_Wantcompute, &tgt->flags);
ecc65c9b
DW		 604 clear_bit(STRIPE_COMPUTE_RUN, &sh->state);
605 if (sh->check_state == check_state_compute_run)
606 sh->check_state = check_state_compute_result;
91c00924
DW		 607 set_bit(STRIPE_HANDLE, &sh->state);
608 release_stripe(sh);
609}
610
7b3a871e 611static struct dma_async_tx_descriptor *ops_run_compute5(struct stripe_head *sh)
91c00924
DW		 612{
613 /* kernel stack size limits the total number of disks */
614 int disks = sh->disks;
615 struct page *xor_srcs[disks];
616 int target = sh->ops.target;
617 struct r5dev *tgt = &sh->dev[target];
618 struct page *xor_dest = tgt->page;
619 int count = 0;
620 struct dma_async_tx_descriptor *tx;
621 int i;
622
623 pr_debug("%s: stripe %llu block: %d\n",
e46b272b 624 __func__, (unsigned long long)sh->sector, target);
91c00924
DW		 625 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
626
627 for (i = disks; i--; )
628 if (i != target)
629 xor_srcs[count++] = sh->dev[i].page;
630
631 atomic_inc(&sh->count);
632
633 if (unlikely(count == 1))
634 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE,
635 0, NULL, ops_complete_compute5, sh);
636 else
637 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
638 ASYNC_TX_XOR_ZERO_DST, NULL,
639 ops_complete_compute5, sh);
640
91c00924
DW		 641 return tx;
642}
643
644static void ops_complete_prexor(void *stripe_head_ref)
645{
646 struct stripe_head *sh = stripe_head_ref;
647
e46b272b 648 pr_debug("%s: stripe %llu\n", __func__,
91c00924 649 (unsigned long long)sh->sector);
91c00924
DW		 650}
651
652static struct dma_async_tx_descriptor *
653ops_run_prexor(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
654{
655 /* kernel stack size limits the total number of disks */
656 int disks = sh->disks;
657 struct page *xor_srcs[disks];
658 int count = 0, pd_idx = sh->pd_idx, i;
659
660 /* existing parity data subtracted */
661 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
662
e46b272b 663 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 664 (unsigned long long)sh->sector);
665
666 for (i = disks; i--; ) {
667 struct r5dev *dev = &sh->dev[i];
668 /* Only process blocks that are known to be uptodate */
d8ee0728 669 if (test_bit(R5_Wantdrain, &dev->flags))
91c00924
DW		 670 xor_srcs[count++] = dev->page;
671 }
672
673 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
674 ASYNC_TX_DEP_ACK | ASYNC_TX_XOR_DROP_DST, tx,
675 ops_complete_prexor, sh);
676
677 return tx;
678}
679
680static struct dma_async_tx_descriptor *
d8ee0728 681ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
91c00924
DW		 682{
683 int disks = sh->disks;
d8ee0728 684 int i;
91c00924 685
e46b272b 686 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 687 (unsigned long long)sh->sector);
688
689 for (i = disks; i--; ) {
690 struct r5dev *dev = &sh->dev[i];
691 struct bio *chosen;
91c00924 692
d8ee0728 693 if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) {
91c00924
DW		 694 struct bio *wbi;
695
696 spin_lock(&sh->lock);
697 chosen = dev->towrite;
698 dev->towrite = NULL;
699 BUG_ON(dev->written);
700 wbi = dev->written = chosen;
701 spin_unlock(&sh->lock);
702
703 while (wbi && wbi->bi_sector <
704 dev->sector + STRIPE_SECTORS) {
705 tx = async_copy_data(1, wbi, dev->page,
706 dev->sector, tx);
707 wbi = r5_next_bio(wbi, dev->sector);
708 }
709 }
710 }
711
712 return tx;
713}
714
715static void ops_complete_postxor(void *stripe_head_ref)
91c00924
DW		 716{
717 struct stripe_head *sh = stripe_head_ref;
718 int disks = sh->disks, i, pd_idx = sh->pd_idx;
719
e46b272b 720 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 721 (unsigned long long)sh->sector);
722
723 for (i = disks; i--; ) {
724 struct r5dev *dev = &sh->dev[i];
725 if (dev->written || i == pd_idx)
726 set_bit(R5_UPTODATE, &dev->flags);
727 }
728
d8ee0728
DW		 729 if (sh->reconstruct_state == reconstruct_state_drain_run)
730 sh->reconstruct_state = reconstruct_state_drain_result;
731 else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run)
732 sh->reconstruct_state = reconstruct_state_prexor_drain_result;
733 else {
734 BUG_ON(sh->reconstruct_state != reconstruct_state_run);
735 sh->reconstruct_state = reconstruct_state_result;
736 }
91c00924
DW		 737
738 set_bit(STRIPE_HANDLE, &sh->state);
739 release_stripe(sh);
740}
741
742static void
d8ee0728 743ops_run_postxor(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
91c00924
DW		 744{
745 /* kernel stack size limits the total number of disks */
746 int disks = sh->disks;
747 struct page *xor_srcs[disks];
748
749 int count = 0, pd_idx = sh->pd_idx, i;
750 struct page *xor_dest;
d8ee0728 751 int prexor = 0;
91c00924 752 unsigned long flags;
91c00924 753
e46b272b 754 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 755 (unsigned long long)sh->sector);
756
757 /* check if prexor is active which means only process blocks
758 * that are part of a read-modify-write (written)
759 */
d8ee0728
DW		 760 if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
761 prexor = 1;
91c00924
DW		 762 xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
763 for (i = disks; i--; ) {
764 struct r5dev *dev = &sh->dev[i];
765 if (dev->written)
766 xor_srcs[count++] = dev->page;
767 }
768 } else {
769 xor_dest = sh->dev[pd_idx].page;
770 for (i = disks; i--; ) {
771 struct r5dev *dev = &sh->dev[i];
772 if (i != pd_idx)
773 xor_srcs[count++] = dev->page;
774 }
775 }
776
91c00924
DW		 777 /* 1/ if we prexor'd then the dest is reused as a source
778 * 2/ if we did not prexor then we are redoing the parity
779 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
780 * for the synchronous xor case
781 */
782 flags = ASYNC_TX_DEP_ACK | ASYNC_TX_ACK |
783 (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
784
785 atomic_inc(&sh->count);
786
787 if (unlikely(count == 1)) {
788 flags &= ~(ASYNC_TX_XOR_DROP_DST | ASYNC_TX_XOR_ZERO_DST);
789 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE,
d8ee0728 790 flags, tx, ops_complete_postxor, sh);
91c00924
DW		 791 } else
792 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
d8ee0728 793 flags, tx, ops_complete_postxor, sh);
91c00924
DW		 794}
795
796static void ops_complete_check(void *stripe_head_ref)
797{
798 struct stripe_head *sh = stripe_head_ref;
91c00924 799
e46b272b 800 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 801 (unsigned long long)sh->sector);
802
ecc65c9b 803 sh->check_state = check_state_check_result;
91c00924
DW		 804 set_bit(STRIPE_HANDLE, &sh->state);
805 release_stripe(sh);
806}
807
808static void ops_run_check(struct stripe_head *sh)
809{
810 /* kernel stack size limits the total number of disks */
811 int disks = sh->disks;
812 struct page *xor_srcs[disks];
813 struct dma_async_tx_descriptor *tx;
814
815 int count = 0, pd_idx = sh->pd_idx, i;
816 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
817
e46b272b 818 pr_debug("%s: stripe %llu\n", __func__,
91c00924
DW		 819 (unsigned long long)sh->sector);
820
821 for (i = disks; i--; ) {
822 struct r5dev *dev = &sh->dev[i];
823 if (i != pd_idx)
824 xor_srcs[count++] = dev->page;
825 }
826
827 tx = async_xor_zero_sum(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
828 &sh->ops.zero_sum_result, 0, NULL, NULL, NULL);
829
91c00924
DW		 830 atomic_inc(&sh->count);
831 tx = async_trigger_callback(ASYNC_TX_DEP_ACK | ASYNC_TX_ACK, tx,
832 ops_complete_check, sh);
833}
834
600aa109 835static void raid5_run_ops(struct stripe_head *sh, unsigned long ops_request)
91c00924
DW		 836{
837 int overlap_clear = 0, i, disks = sh->disks;
838 struct dma_async_tx_descriptor *tx = NULL;
839
83de75cc 840 if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) {
91c00924
DW		 841 ops_run_biofill(sh);
842 overlap_clear++;
843 }
844
7b3a871e
DW		 845 if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) {
846 tx = ops_run_compute5(sh);
847 /* terminate the chain if postxor is not set to be run */
848 if (tx && !test_bit(STRIPE_OP_POSTXOR, &ops_request))
849 async_tx_ack(tx);
850 }
91c00924 851
600aa109 852 if (test_bit(STRIPE_OP_PREXOR, &ops_request))
91c00924
DW		 853 tx = ops_run_prexor(sh, tx);
854
600aa109 855 if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) {
d8ee0728 856 tx = ops_run_biodrain(sh, tx);
91c00924
DW		 857 overlap_clear++;
858 }
859
600aa109 860 if (test_bit(STRIPE_OP_POSTXOR, &ops_request))
d8ee0728 861 ops_run_postxor(sh, tx);
91c00924 862
ecc65c9b 863 if (test_bit(STRIPE_OP_CHECK, &ops_request))
91c00924
DW		 864 ops_run_check(sh);
865
91c00924
DW		 866 if (overlap_clear)
867 for (i = disks; i--; ) {
868 struct r5dev *dev = &sh->dev[i];
869 if (test_and_clear_bit(R5_Overlap, &dev->flags))
870 wake_up(&sh->raid_conf->wait_for_overlap);
871 }
872}
873
3f294f4f 874static int grow_one_stripe(raid5_conf_t *conf)
1da177e4
LT		 875{
876 struct stripe_head *sh;
3f294f4f
N		 877 sh = kmem_cache_alloc(conf->slab_cache, GFP_KERNEL);
878 if (!sh)
879 return 0;
880 memset(sh, 0, sizeof(*sh) + (conf->raid_disks-1)*sizeof(struct r5dev));
881 sh->raid_conf = conf;
882 spin_lock_init(&sh->lock);
883
884 if (grow_buffers(sh, conf->raid_disks)) {
885 shrink_buffers(sh, conf->raid_disks);
886 kmem_cache_free(conf->slab_cache, sh);
887 return 0;
888 }
7ecaa1e6 889 sh->disks = conf->raid_disks;
3f294f4f
N		 890 /* we just created an active stripe so... */
891 atomic_set(&sh->count, 1);
892 atomic_inc(&conf->active_stripes);
893 INIT_LIST_HEAD(&sh->lru);
894 release_stripe(sh);
895 return 1;
896}
897
898static int grow_stripes(raid5_conf_t *conf, int num)
899{
e18b890b 900 struct kmem_cache *sc;
1da177e4
LT		 901 int devs = conf->raid_disks;
902
42b9bebe
N		 903 sprintf(conf->cache_name[0], "raid5-%s", mdname(conf->mddev));
904 sprintf(conf->cache_name[1], "raid5-%s-alt", mdname(conf->mddev));
ad01c9e3
N		 905 conf->active_name = 0;
906 sc = kmem_cache_create(conf->cache_name[conf->active_name],
1da177e4 907 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
20c2df83 908 0, 0, NULL);
1da177e4
LT		 909 if (!sc)
910 return 1;
911 conf->slab_cache = sc;
ad01c9e3 912 conf->pool_size = devs;
16a53ecc 913 while (num--)
3f294f4f 914 if (!grow_one_stripe(conf))
1da177e4 915 return 1;
1da177e4
LT		 916 return 0;
917}
29269553
N		 918
919#ifdef CONFIG_MD_RAID5_RESHAPE
ad01c9e3
N		 920static int resize_stripes(raid5_conf_t *conf, int newsize)
921{
922 /* Make all the stripes able to hold 'newsize' devices.
923 * New slots in each stripe get 'page' set to a new page.
924 *
925 * This happens in stages:
926 * 1/ create a new kmem_cache and allocate the required number of
927 * stripe_heads.
928 * 2/ gather all the old stripe_heads and tranfer the pages across
929 * to the new stripe_heads. This will have the side effect of
930 * freezing the array as once all stripe_heads have been collected,
931 * no IO will be possible. Old stripe heads are freed once their
932 * pages have been transferred over, and the old kmem_cache is
933 * freed when all stripes are done.
934 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
935 * we simple return a failre status - no need to clean anything up.
936 * 4/ allocate new pages for the new slots in the new stripe_heads.
937 * If this fails, we don't bother trying the shrink the
938 * stripe_heads down again, we just leave them as they are.
939 * As each stripe_head is processed the new one is released into
940 * active service.
941 *
942 * Once step2 is started, we cannot afford to wait for a write,
943 * so we use GFP_NOIO allocations.
944 */
945 struct stripe_head *osh, *nsh;
946 LIST_HEAD(newstripes);
947 struct disk_info *ndisks;
b5470dc5 948 int err;
e18b890b 949 struct kmem_cache *sc;
ad01c9e3
N		 950 int i;
951
952 if (newsize <= conf->pool_size)
953 return 0; /* never bother to shrink */
954
b5470dc5
DW		 955 err = md_allow_write(conf->mddev);
956 if (err)
957 return err;
2a2275d6 958
ad01c9e3
N		 959 /* Step 1 */
960 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
961 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
20c2df83 962 0, 0, NULL);
ad01c9e3
N		 963 if (!sc)
964 return -ENOMEM;
965
966 for (i = conf->max_nr_stripes; i; i--) {
967 nsh = kmem_cache_alloc(sc, GFP_KERNEL);
968 if (!nsh)
969 break;
970
971 memset(nsh, 0, sizeof(*nsh) + (newsize-1)*sizeof(struct r5dev));
972
973 nsh->raid_conf = conf;
974 spin_lock_init(&nsh->lock);
975
976 list_add(&nsh->lru, &newstripes);
977 }
978 if (i) {
979 /* didn't get enough, give up */
980 while (!list_empty(&newstripes)) {
981 nsh = list_entry(newstripes.next, struct stripe_head, lru);
982 list_del(&nsh->lru);
983 kmem_cache_free(sc, nsh);
984 }
985 kmem_cache_destroy(sc);
986 return -ENOMEM;
987 }
988 /* Step 2 - Must use GFP_NOIO now.
989 * OK, we have enough stripes, start collecting inactive
990 * stripes and copying them over
991 */
992 list_for_each_entry(nsh, &newstripes, lru) {
993 spin_lock_irq(&conf->device_lock);
994 wait_event_lock_irq(conf->wait_for_stripe,
995 !list_empty(&conf->inactive_list),
996 conf->device_lock,
b3b46be3 997 unplug_slaves(conf->mddev)
ad01c9e3
N		 998 );
999 osh = get_free_stripe(conf);
1000 spin_unlock_irq(&conf->device_lock);
1001 atomic_set(&nsh->count, 1);
1002 for(i=0; i<conf->pool_size; i++)
1003 nsh->dev[i].page = osh->dev[i].page;
1004 for( ; i<newsize; i++)
1005 nsh->dev[i].page = NULL;
1006 kmem_cache_free(conf->slab_cache, osh);
1007 }
1008 kmem_cache_destroy(conf->slab_cache);
1009
1010 /* Step 3.
1011 * At this point, we are holding all the stripes so the array
1012 * is completely stalled, so now is a good time to resize
1013 * conf->disks.
1014 */
1015 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
1016 if (ndisks) {
1017 for (i=0; i<conf->raid_disks; i++)
1018 ndisks[i] = conf->disks[i];
1019 kfree(conf->disks);
1020 conf->disks = ndisks;
1021 } else
1022 err = -ENOMEM;
1023
1024 /* Step 4, return new stripes to service */
1025 while(!list_empty(&newstripes)) {
1026 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1027 list_del_init(&nsh->lru);
1028 for (i=conf->raid_disks; i < newsize; i++)
1029 if (nsh->dev[i].page == NULL) {
1030 struct page *p = alloc_page(GFP_NOIO);
1031 nsh->dev[i].page = p;
1032 if (!p)
1033 err = -ENOMEM;
1034 }
1035 release_stripe(nsh);
1036 }
1037 /* critical section pass, GFP_NOIO no longer needed */
1038
1039 conf->slab_cache = sc;
1040 conf->active_name = 1-conf->active_name;
1041 conf->pool_size = newsize;
1042 return err;
1043}
29269553 1044#endif
1da177e4 1045
3f294f4f 1046static int drop_one_stripe(raid5_conf_t *conf)
1da177e4
LT		 1047{
1048 struct stripe_head *sh;
1049
3f294f4f
N		 1050 spin_lock_irq(&conf->device_lock);
1051 sh = get_free_stripe(conf);
1052 spin_unlock_irq(&conf->device_lock);
1053 if (!sh)
1054 return 0;
78bafebd 1055 BUG_ON(atomic_read(&sh->count));
ad01c9e3 1056 shrink_buffers(sh, conf->pool_size);
3f294f4f
N		 1057 kmem_cache_free(conf->slab_cache, sh);
1058 atomic_dec(&conf->active_stripes);
1059 return 1;
1060}
1061
1062static void shrink_stripes(raid5_conf_t *conf)
1063{
1064 while (drop_one_stripe(conf))
1065 ;
1066
29fc7e3e
N		 1067 if (conf->slab_cache)
1068 kmem_cache_destroy(conf->slab_cache);
1da177e4
LT		 1069 conf->slab_cache = NULL;
1070}
1071
6712ecf8 1072static void raid5_end_read_request(struct bio * bi, int error)
1da177e4
LT		 1073{
1074 struct stripe_head *sh = bi->bi_private;
1075 raid5_conf_t *conf = sh->raid_conf;
7ecaa1e6 1076 int disks = sh->disks, i;
1da177e4 1077 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
d6950432
N		 1078 char b[BDEVNAME_SIZE];
1079 mdk_rdev_t *rdev;
1da177e4 1080
1da177e4
LT		 1081
1082 for (i=0 ; i<disks; i++)
1083 if (bi == &sh->dev[i].req)
1084 break;
1085
45b4233c
DW		 1086 pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n",
1087 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
1da177e4
LT		 1088 uptodate);
1089 if (i == disks) {
1090 BUG();
6712ecf8 1091 return;
1da177e4
LT		 1092 }
1093
1094 if (uptodate) {
1da177e4 1095 set_bit(R5_UPTODATE, &sh->dev[i].flags);
4e5314b5 1096 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
d6950432 1097 rdev = conf->disks[i].rdev;
6be9d494
BS		 1098 printk_rl(KERN_INFO "raid5:%s: read error corrected"
1099 " (%lu sectors at %llu on %s)\n",
1100 mdname(conf->mddev), STRIPE_SECTORS,
1101 (unsigned long long)(sh->sector
1102 + rdev->data_offset),
1103 bdevname(rdev->bdev, b));
4e5314b5
N		 1104 clear_bit(R5_ReadError, &sh->dev[i].flags);
1105 clear_bit(R5_ReWrite, &sh->dev[i].flags);
1106 }
ba22dcbf
N		 1107 if (atomic_read(&conf->disks[i].rdev->read_errors))
1108 atomic_set(&conf->disks[i].rdev->read_errors, 0);
1da177e4 1109 } else {
d6950432 1110 const char *bdn = bdevname(conf->disks[i].rdev->bdev, b);
ba22dcbf 1111 int retry = 0;
d6950432
N		 1112 rdev = conf->disks[i].rdev;
1113
1da177e4 1114 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
d6950432 1115 atomic_inc(&rdev->read_errors);
ba22dcbf 1116 if (conf->mddev->degraded)
6be9d494
BS		 1117 printk_rl(KERN_WARNING
1118 "raid5:%s: read error not correctable "
1119 "(sector %llu on %s).\n",
1120 mdname(conf->mddev),
1121 (unsigned long long)(sh->sector
1122 + rdev->data_offset),
1123 bdn);
ba22dcbf 1124 else if (test_bit(R5_ReWrite, &sh->dev[i].flags))
4e5314b5 1125 /* Oh, no!!! */
6be9d494
BS		 1126 printk_rl(KERN_WARNING
1127 "raid5:%s: read error NOT corrected!! "
1128 "(sector %llu on %s).\n",
1129 mdname(conf->mddev),
1130 (unsigned long long)(sh->sector
1131 + rdev->data_offset),
1132 bdn);
d6950432 1133 else if (atomic_read(&rdev->read_errors)
ba22dcbf 1134 > conf->max_nr_stripes)
14f8d26b 1135 printk(KERN_WARNING
d6950432
N		 1136 "raid5:%s: Too many read errors, failing device %s.\n",
1137 mdname(conf->mddev), bdn);
ba22dcbf
N		 1138 else
1139 retry = 1;
1140 if (retry)
1141 set_bit(R5_ReadError, &sh->dev[i].flags);
1142 else {
4e5314b5
N		 1143 clear_bit(R5_ReadError, &sh->dev[i].flags);
1144 clear_bit(R5_ReWrite, &sh->dev[i].flags);
d6950432 1145 md_error(conf->mddev, rdev);
ba22dcbf 1146 }
1da177e4
LT		 1147 }
1148 rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
1da177e4
LT		 1149 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1150 set_bit(STRIPE_HANDLE, &sh->state);
1151 release_stripe(sh);
1da177e4
LT		 1152}
1153
6712ecf8 1154static void raid5_end_write_request (struct bio *bi, int error)
1da177e4
LT		 1155{
1156 struct stripe_head *sh = bi->bi_private;
1157 raid5_conf_t *conf = sh->raid_conf;
7ecaa1e6 1158 int disks = sh->disks, i;
1da177e4
LT		 1159 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
1160
1da177e4
LT		 1161 for (i=0 ; i<disks; i++)
1162 if (bi == &sh->dev[i].req)
1163 break;
1164
45b4233c 1165 pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n",
1da177e4
LT		 1166 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
1167 uptodate);
1168 if (i == disks) {
1169 BUG();
6712ecf8 1170 return;
1da177e4
LT		 1171 }
1172
1da177e4
LT		 1173 if (!uptodate)
1174 md_error(conf->mddev, conf->disks[i].rdev);
1175
1176 rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
1177
1178 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1179 set_bit(STRIPE_HANDLE, &sh->state);
c04be0aa 1180 release_stripe(sh);
1da177e4
LT		 1181}
1182
1183
1184static sector_t compute_blocknr(struct stripe_head *sh, int i);
1185
1186static void raid5_build_block (struct stripe_head *sh, int i)
1187{
1188 struct r5dev *dev = &sh->dev[i];
1189
1190 bio_init(&dev->req);
1191 dev->req.bi_io_vec = &dev->vec;
1192 dev->req.bi_vcnt++;
1193 dev->req.bi_max_vecs++;
1194 dev->vec.bv_page = dev->page;
1195 dev->vec.bv_len = STRIPE_SIZE;
1196 dev->vec.bv_offset = 0;
1197
1198 dev->req.bi_sector = sh->sector;
1199 dev->req.bi_private = sh;
1200
1201 dev->flags = 0;
16a53ecc 1202 dev->sector = compute_blocknr(sh, i);
1da177e4
LT		 1203}
1204
1205static void error(mddev_t *mddev, mdk_rdev_t *rdev)
1206{
1207 char b[BDEVNAME_SIZE];
1208 raid5_conf_t *conf = (raid5_conf_t *) mddev->private;
45b4233c 1209 pr_debug("raid5: error called\n");
1da177e4 1210
b2d444d7 1211 if (!test_bit(Faulty, &rdev->flags)) {
850b2b42 1212 set_bit(MD_CHANGE_DEVS, &mddev->flags);
c04be0aa
N		 1213 if (test_and_clear_bit(In_sync, &rdev->flags)) {
1214 unsigned long flags;
1215 spin_lock_irqsave(&conf->device_lock, flags);
1da177e4 1216 mddev->degraded++;
c04be0aa 1217 spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4
LT		 1218 /*
1219 * if recovery was running, make sure it aborts.
1220 */
dfc70645 1221 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4 1222 }
b2d444d7 1223 set_bit(Faulty, &rdev->flags);
1da177e4 1224 printk (KERN_ALERT
d7a420c9
NA		 1225 "raid5: Disk failure on %s, disabling device.\n"
1226 "raid5: Operation continuing on %d devices.\n",
02c2de8c 1227 bdevname(rdev->bdev,b), conf->raid_disks - mddev->degraded);
1da177e4 1228 }
16a53ecc 1229}
1da177e4
LT		 1230
1231/*
1232 * Input: a 'big' sector number,
1233 * Output: index of the data and parity disk, and the sector # in them.
1234 */
1235static sector_t raid5_compute_sector(sector_t r_sector, unsigned int raid_disks,
1236 unsigned int data_disks, unsigned int * dd_idx,
1237 unsigned int * pd_idx, raid5_conf_t *conf)
1238{
1239 long stripe;
1240 unsigned long chunk_number;
1241 unsigned int chunk_offset;
1242 sector_t new_sector;
1243 int sectors_per_chunk = conf->chunk_size >> 9;
1244
1245 /* First compute the information on this sector */
1246
1247 /*
1248 * Compute the chunk number and the sector offset inside the chunk
1249 */
1250 chunk_offset = sector_div(r_sector, sectors_per_chunk);
1251 chunk_number = r_sector;
1252 BUG_ON(r_sector != chunk_number);
1253
1254 /*
1255 * Compute the stripe number
1256 */
1257 stripe = chunk_number / data_disks;
1258
1259 /*
1260 * Compute the data disk and parity disk indexes inside the stripe
1261 */
1262 *dd_idx = chunk_number % data_disks;
1263
1264 /*
1265 * Select the parity disk based on the user selected algorithm.
1266 */
16a53ecc
N		 1267 switch(conf->level) {
1268 case 4:
1da177e4 1269 *pd_idx = data_disks;
16a53ecc
N		 1270 break;
1271 case 5:
1272 switch (conf->algorithm) {
1da177e4
LT		 1273 case ALGORITHM_LEFT_ASYMMETRIC:
1274 *pd_idx = data_disks - stripe % raid_disks;
1275 if (*dd_idx >= *pd_idx)
1276 (*dd_idx)++;
1277 break;
1278 case ALGORITHM_RIGHT_ASYMMETRIC:
1279 *pd_idx = stripe % raid_disks;
1280 if (*dd_idx >= *pd_idx)
1281 (*dd_idx)++;
1282 break;
1283 case ALGORITHM_LEFT_SYMMETRIC:
1284 *pd_idx = data_disks - stripe % raid_disks;
1285 *dd_idx = (*pd_idx + 1 + *dd_idx) % raid_disks;
1286 break;
1287 case ALGORITHM_RIGHT_SYMMETRIC:
1288 *pd_idx = stripe % raid_disks;
1289 *dd_idx = (*pd_idx + 1 + *dd_idx) % raid_disks;
1290 break;
1291 default:
14f8d26b 1292 printk(KERN_ERR "raid5: unsupported algorithm %d\n",
1da177e4 1293 conf->algorithm);
16a53ecc
N		 1294 }
1295 break;
1296 case 6:
1297
1298 /**** FIX THIS ****/
1299 switch (conf->algorithm) {
1300 case ALGORITHM_LEFT_ASYMMETRIC:
1301 *pd_idx = raid_disks - 1 - (stripe % raid_disks);
1302 if (*pd_idx == raid_disks-1)
1303 (*dd_idx)++; /* Q D D D P */
1304 else if (*dd_idx >= *pd_idx)
1305 (*dd_idx) += 2; /* D D P Q D */
1306 break;
1307 case ALGORITHM_RIGHT_ASYMMETRIC:
1308 *pd_idx = stripe % raid_disks;
1309 if (*pd_idx == raid_disks-1)
1310 (*dd_idx)++; /* Q D D D P */
1311 else if (*dd_idx >= *pd_idx)
1312 (*dd_idx) += 2; /* D D P Q D */
1313 break;
1314 case ALGORITHM_LEFT_SYMMETRIC:
1315 *pd_idx = raid_disks - 1 - (stripe % raid_disks);
1316 *dd_idx = (*pd_idx + 2 + *dd_idx) % raid_disks;
1317 break;
1318 case ALGORITHM_RIGHT_SYMMETRIC:
1319 *pd_idx = stripe % raid_disks;
1320 *dd_idx = (*pd_idx + 2 + *dd_idx) % raid_disks;
1321 break;
1322 default:
1323 printk (KERN_CRIT "raid6: unsupported algorithm %d\n",
1324 conf->algorithm);
1325 }
1326 break;
1da177e4
LT		 1327 }
1328
1329 /*
1330 * Finally, compute the new sector number
1331 */
1332 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
1333 return new_sector;
1334}
1335
1336
1337static sector_t compute_blocknr(struct stripe_head *sh, int i)
1338{
1339 raid5_conf_t *conf = sh->raid_conf;
b875e531
N		 1340 int raid_disks = sh->disks;
1341 int data_disks = raid_disks - conf->max_degraded;
1da177e4
LT		 1342 sector_t new_sector = sh->sector, check;
1343 int sectors_per_chunk = conf->chunk_size >> 9;
1344 sector_t stripe;
1345 int chunk_offset;
1346 int chunk_number, dummy1, dummy2, dd_idx = i;
1347 sector_t r_sector;
1348
16a53ecc 1349
1da177e4
LT		 1350 chunk_offset = sector_div(new_sector, sectors_per_chunk);
1351 stripe = new_sector;
1352 BUG_ON(new_sector != stripe);
1353
16a53ecc
N		 1354 if (i == sh->pd_idx)
1355 return 0;
1356 switch(conf->level) {
1357 case 4: break;
1358 case 5:
1359 switch (conf->algorithm) {
1da177e4
LT		 1360 case ALGORITHM_LEFT_ASYMMETRIC:
1361 case ALGORITHM_RIGHT_ASYMMETRIC:
1362 if (i > sh->pd_idx)
1363 i--;
1364 break;
1365 case ALGORITHM_LEFT_SYMMETRIC:
1366 case ALGORITHM_RIGHT_SYMMETRIC:
1367 if (i < sh->pd_idx)
1368 i += raid_disks;
1369 i -= (sh->pd_idx + 1);
1370 break;
1371 default:
14f8d26b 1372 printk(KERN_ERR "raid5: unsupported algorithm %d\n",
16a53ecc
N		 1373 conf->algorithm);
1374 }
1375 break;
1376 case 6:
16a53ecc
N		 1377 if (i == raid6_next_disk(sh->pd_idx, raid_disks))
1378 return 0; /* It is the Q disk */
1379 switch (conf->algorithm) {
1380 case ALGORITHM_LEFT_ASYMMETRIC:
1381 case ALGORITHM_RIGHT_ASYMMETRIC:
1382 if (sh->pd_idx == raid_disks-1)
1383 i--; /* Q D D D P */
1384 else if (i > sh->pd_idx)
1385 i -= 2; /* D D P Q D */
1386 break;
1387 case ALGORITHM_LEFT_SYMMETRIC:
1388 case ALGORITHM_RIGHT_SYMMETRIC:
1389 if (sh->pd_idx == raid_disks-1)
1390 i--; /* Q D D D P */
1391 else {
1392 /* D D P Q D */
1393 if (i < sh->pd_idx)
1394 i += raid_disks;
1395 i -= (sh->pd_idx + 2);
1396 }
1397 break;
1398 default:
1399 printk (KERN_CRIT "raid6: unsupported algorithm %d\n",
1da177e4 1400 conf->algorithm);
16a53ecc
N		 1401 }
1402 break;
1da177e4
LT		 1403 }
1404
1405 chunk_number = stripe * data_disks + i;
1406 r_sector = (sector_t)chunk_number * sectors_per_chunk + chunk_offset;
1407
1408 check = raid5_compute_sector (r_sector, raid_disks, data_disks, &dummy1, &dummy2, conf);
1409 if (check != sh->sector || dummy1 != dd_idx || dummy2 != sh->pd_idx) {
14f8d26b 1410 printk(KERN_ERR "compute_blocknr: map not correct\n");
1da177e4
LT		 1411 return 0;
1412 }
1413 return r_sector;
1414}
1415
1416
1417
1418/*
16a53ecc
N		 1419 * Copy data between a page in the stripe cache, and one or more bion
1420 * The page could align with the middle of the bio, or there could be
1421 * several bion, each with several bio_vecs, which cover part of the page
1422 * Multiple bion are linked together on bi_next. There may be extras
1423 * at the end of this list. We ignore them.
1da177e4
LT		 1424 */
1425static void copy_data(int frombio, struct bio *bio,
1426 struct page *page,
1427 sector_t sector)
1428{
1429 char *pa = page_address(page);
1430 struct bio_vec *bvl;
1431 int i;
1432 int page_offset;
1433
1434 if (bio->bi_sector >= sector)
1435 page_offset = (signed)(bio->bi_sector - sector) * 512;
1436 else
1437 page_offset = (signed)(sector - bio->bi_sector) * -512;
1438 bio_for_each_segment(bvl, bio, i) {
1439 int len = bio_iovec_idx(bio,i)->bv_len;
1440 int clen;
1441 int b_offset = 0;
1442
1443 if (page_offset < 0) {
1444 b_offset = -page_offset;
1445 page_offset += b_offset;
1446 len -= b_offset;
1447 }
1448
1449 if (len > 0 && page_offset + len > STRIPE_SIZE)
1450 clen = STRIPE_SIZE - page_offset;
1451 else clen = len;
16a53ecc 1452
1da177e4
LT		 1453 if (clen > 0) {
1454 char *ba = __bio_kmap_atomic(bio, i, KM_USER0);
1455 if (frombio)
1456 memcpy(pa+page_offset, ba+b_offset, clen);
1457 else
1458 memcpy(ba+b_offset, pa+page_offset, clen);
1459 __bio_kunmap_atomic(ba, KM_USER0);
1460 }
1461 if (clen < len) /* hit end of page */
1462 break;
1463 page_offset += len;
1464 }
1465}
1466
9bc89cd8
DW		 1467#define check_xor() do { \
1468 if (count == MAX_XOR_BLOCKS) { \
1469 xor_blocks(count, STRIPE_SIZE, dest, ptr);\
1470 count = 0; \
1471 } \
1da177e4
LT		 1472 } while(0)
1473
16a53ecc
N		 1474static void compute_parity6(struct stripe_head *sh, int method)
1475{
1476 raid6_conf_t *conf = sh->raid_conf;
f416885e 1477 int i, pd_idx = sh->pd_idx, qd_idx, d0_idx, disks = sh->disks, count;
16a53ecc
N		 1478 struct bio *chosen;
1479 /**** FIX THIS: This could be very bad if disks is close to 256 ****/
1480 void *ptrs[disks];
1481
1482 qd_idx = raid6_next_disk(pd_idx, disks);
1483 d0_idx = raid6_next_disk(qd_idx, disks);
1484
45b4233c 1485 pr_debug("compute_parity, stripe %llu, method %d\n",
16a53ecc
N		 1486 (unsigned long long)sh->sector, method);
1487
1488 switch(method) {
1489 case READ_MODIFY_WRITE:
1490 BUG(); /* READ_MODIFY_WRITE N/A for RAID-6 */
1491 case RECONSTRUCT_WRITE:
1492 for (i= disks; i-- ;)
1493 if ( i != pd_idx && i != qd_idx && sh->dev[i].towrite ) {
1494 chosen = sh->dev[i].towrite;
1495 sh->dev[i].towrite = NULL;
1496
1497 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
1498 wake_up(&conf->wait_for_overlap);
1499
52e5f9d1 1500 BUG_ON(sh->dev[i].written);
16a53ecc
N		 1501 sh->dev[i].written = chosen;
1502 }
1503 break;
1504 case CHECK_PARITY:
1505 BUG(); /* Not implemented yet */
1506 }
1507
1508 for (i = disks; i--;)
1509 if (sh->dev[i].written) {
1510 sector_t sector = sh->dev[i].sector;
1511 struct bio *wbi = sh->dev[i].written;
1512 while (wbi && wbi->bi_sector < sector + STRIPE_SECTORS) {
1513 copy_data(1, wbi, sh->dev[i].page, sector);
1514 wbi = r5_next_bio(wbi, sector);
1515 }
1516
1517 set_bit(R5_LOCKED, &sh->dev[i].flags);
1518 set_bit(R5_UPTODATE, &sh->dev[i].flags);
1519 }
1520
1521// switch(method) {
1522// case RECONSTRUCT_WRITE:
1523// case CHECK_PARITY:
1524// case UPDATE_PARITY:
1525 /* Note that unlike RAID-5, the ordering of the disks matters greatly. */
1526 /* FIX: Is this ordering of drives even remotely optimal? */
1527 count = 0;
1528 i = d0_idx;
1529 do {
1530 ptrs[count++] = page_address(sh->dev[i].page);
1531 if (count <= disks-2 && !test_bit(R5_UPTODATE, &sh->dev[i].flags))
1532 printk("block %d/%d not uptodate on parity calc\n", i,count);
1533 i = raid6_next_disk(i, disks);
1534 } while ( i != d0_idx );
1535// break;
1536// }
1537
1538 raid6_call.gen_syndrome(disks, STRIPE_SIZE, ptrs);
1539
1540 switch(method) {
1541 case RECONSTRUCT_WRITE:
1542 set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
1543 set_bit(R5_UPTODATE, &sh->dev[qd_idx].flags);
1544 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
1545 set_bit(R5_LOCKED, &sh->dev[qd_idx].flags);
1546 break;
1547 case UPDATE_PARITY:
1548 set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
1549 set_bit(R5_UPTODATE, &sh->dev[qd_idx].flags);
1550 break;
1551 }
1552}
1553
1554
1555/* Compute one missing block */
1556static void compute_block_1(struct stripe_head *sh, int dd_idx, int nozero)
1557{
f416885e 1558 int i, count, disks = sh->disks;
9bc89cd8 1559 void *ptr[MAX_XOR_BLOCKS], *dest, *p;
16a53ecc
N		 1560 int pd_idx = sh->pd_idx;
1561 int qd_idx = raid6_next_disk(pd_idx, disks);
1562
45b4233c 1563 pr_debug("compute_block_1, stripe %llu, idx %d\n",
16a53ecc
N		 1564 (unsigned long long)sh->sector, dd_idx);
1565
1566 if ( dd_idx == qd_idx ) {
1567 /* We're actually computing the Q drive */
1568 compute_parity6(sh, UPDATE_PARITY);
1569 } else {
9bc89cd8
DW		 1570 dest = page_address(sh->dev[dd_idx].page);
1571 if (!nozero) memset(dest, 0, STRIPE_SIZE);
1572 count = 0;
16a53ecc
N		 1573 for (i = disks ; i--; ) {
1574 if (i == dd_idx || i == qd_idx)
1575 continue;
1576 p = page_address(sh->dev[i].page);
1577 if (test_bit(R5_UPTODATE, &sh->dev[i].flags))
1578 ptr[count++] = p;
1579 else
1580 printk("compute_block() %d, stripe %llu, %d"
1581 " not present\n", dd_idx,
1582 (unsigned long long)sh->sector, i);
1583
1584 check_xor();
1585 }
9bc89cd8
DW		 1586 if (count)
1587 xor_blocks(count, STRIPE_SIZE, dest, ptr);
16a53ecc
N		 1588 if (!nozero) set_bit(R5_UPTODATE, &sh->dev[dd_idx].flags);
1589 else clear_bit(R5_UPTODATE, &sh->dev[dd_idx].flags);
1590 }
1591}
1592
1593/* Compute two missing blocks */
1594static void compute_block_2(struct stripe_head *sh, int dd_idx1, int dd_idx2)
1595{
f416885e 1596 int i, count, disks = sh->disks;
16a53ecc
N		 1597 int pd_idx = sh->pd_idx;
1598 int qd_idx = raid6_next_disk(pd_idx, disks);
1599 int d0_idx = raid6_next_disk(qd_idx, disks);
1600 int faila, failb;
1601
1602 /* faila and failb are disk numbers relative to d0_idx */
1603 /* pd_idx become disks-2 and qd_idx become disks-1 */
1604 faila = (dd_idx1 < d0_idx) ? dd_idx1+(disks-d0_idx) : dd_idx1-d0_idx;
1605 failb = (dd_idx2 < d0_idx) ? dd_idx2+(disks-d0_idx) : dd_idx2-d0_idx;
1606
1607 BUG_ON(faila == failb);
1608 if ( failb < faila ) { int tmp = faila; faila = failb; failb = tmp; }
1609
45b4233c 1610 pr_debug("compute_block_2, stripe %llu, idx %d,%d (%d,%d)\n",
16a53ecc
N		 1611 (unsigned long long)sh->sector, dd_idx1, dd_idx2, faila, failb);
1612
1613 if ( failb == disks-1 ) {
1614 /* Q disk is one of the missing disks */
1615 if ( faila == disks-2 ) {
1616 /* Missing P+Q, just recompute */
1617 compute_parity6(sh, UPDATE_PARITY);
1618 return;
1619 } else {
1620 /* We're missing D+Q; recompute D from P */
1621 compute_block_1(sh, (dd_idx1 == qd_idx) ? dd_idx2 : dd_idx1, 0);
1622 compute_parity6(sh, UPDATE_PARITY); /* Is this necessary? */
1623 return;
1624 }
1625 }
1626
1627 /* We're missing D+P or D+D; build pointer table */
1628 {
1629 /**** FIX THIS: This could be very bad if disks is close to 256 ****/
1630 void *ptrs[disks];
1631
1632 count = 0;
1633 i = d0_idx;
1634 do {
1635 ptrs[count++] = page_address(sh->dev[i].page);
1636 i = raid6_next_disk(i, disks);
1637 if (i != dd_idx1 && i != dd_idx2 &&
1638 !test_bit(R5_UPTODATE, &sh->dev[i].flags))
1639 printk("compute_2 with missing block %d/%d\n", count, i);
1640 } while ( i != d0_idx );
1641
1642 if ( failb == disks-2 ) {
1643 /* We're missing D+P. */
1644 raid6_datap_recov(disks, STRIPE_SIZE, faila, ptrs);
1645 } else {
1646 /* We're missing D+D. */
1647 raid6_2data_recov(disks, STRIPE_SIZE, faila, failb, ptrs);
1648 }
1649
1650 /* Both the above update both missing blocks */
1651 set_bit(R5_UPTODATE, &sh->dev[dd_idx1].flags);
1652 set_bit(R5_UPTODATE, &sh->dev[dd_idx2].flags);
1653 }
1654}
1655
600aa109 1656static void
1fe797e6 1657schedule_reconstruction5(struct stripe_head *sh, struct stripe_head_state *s,
600aa109 1658 int rcw, int expand)
e33129d8
DW		 1659{
1660 int i, pd_idx = sh->pd_idx, disks = sh->disks;
e33129d8
DW		 1661
1662 if (rcw) {
1663 /* if we are not expanding this is a proper write request, and
1664 * there will be bios with new data to be drained into the
1665 * stripe cache
1666 */
1667 if (!expand) {
600aa109
DW		 1668 sh->reconstruct_state = reconstruct_state_drain_run;
1669 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
1670 } else
1671 sh->reconstruct_state = reconstruct_state_run;
16a53ecc 1672
600aa109 1673 set_bit(STRIPE_OP_POSTXOR, &s->ops_request);
e33129d8
DW		 1674
1675 for (i = disks; i--; ) {
1676 struct r5dev *dev = &sh->dev[i];
1677
1678 if (dev->towrite) {
1679 set_bit(R5_LOCKED, &dev->flags);
d8ee0728 1680 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 1681 if (!expand)
1682 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 1683 s->locked++;
e33129d8
DW		 1684 }
1685 }
600aa109 1686 if (s->locked + 1 == disks)
8b3e6cdc
DW		 1687 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
1688 atomic_inc(&sh->raid_conf->pending_full_writes);
e33129d8
DW		 1689 } else {
1690 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
1691 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
1692
d8ee0728 1693 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
600aa109
DW		 1694 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
1695 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
1696 set_bit(STRIPE_OP_POSTXOR, &s->ops_request);
e33129d8
DW		 1697
1698 for (i = disks; i--; ) {
1699 struct r5dev *dev = &sh->dev[i];
1700 if (i == pd_idx)
1701 continue;
1702
e33129d8
DW		 1703 if (dev->towrite &&
1704 (test_bit(R5_UPTODATE, &dev->flags) ||
d8ee0728
DW		 1705 test_bit(R5_Wantcompute, &dev->flags))) {
1706 set_bit(R5_Wantdrain, &dev->flags);
e33129d8
DW		 1707 set_bit(R5_LOCKED, &dev->flags);
1708 clear_bit(R5_UPTODATE, &dev->flags);
600aa109 1709 s->locked++;
e33129d8
DW		 1710 }
1711 }
1712 }
1713
1714 /* keep the parity disk locked while asynchronous operations
1715 * are in flight
1716 */
1717 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
1718 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
600aa109 1719 s->locked++;
e33129d8 1720
600aa109 1721 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
e46b272b 1722 __func__, (unsigned long long)sh->sector,
600aa109 1723 s->locked, s->ops_request);
e33129d8 1724}
16a53ecc 1725
1da177e4
LT		 1726/*
1727 * Each stripe/dev can have one or more bion attached.
16a53ecc 1728 * toread/towrite point to the first in a chain.
1da177e4
LT		 1729 * The bi_next chain must be in order.
1730 */
1731static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite)
1732{
1733 struct bio **bip;
1734 raid5_conf_t *conf = sh->raid_conf;
72626685 1735 int firstwrite=0;
1da177e4 1736
45b4233c 1737 pr_debug("adding bh b#%llu to stripe s#%llu\n",
1da177e4
LT		 1738 (unsigned long long)bi->bi_sector,
1739 (unsigned long long)sh->sector);
1740
1741
1742 spin_lock(&sh->lock);
1743 spin_lock_irq(&conf->device_lock);
72626685 1744 if (forwrite) {
1da177e4 1745 bip = &sh->dev[dd_idx].towrite;
72626685
N		 1746 if (*bip == NULL && sh->dev[dd_idx].written == NULL)
1747 firstwrite = 1;
1748 } else
1da177e4
LT		 1749 bip = &sh->dev[dd_idx].toread;
1750 while (*bip && (*bip)->bi_sector < bi->bi_sector) {
1751 if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector)
1752 goto overlap;
1753 bip = & (*bip)->bi_next;
1754 }
1755 if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9))
1756 goto overlap;
1757
78bafebd 1758 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
1da177e4
LT		 1759 if (*bip)
1760 bi->bi_next = *bip;
1761 *bip = bi;
960e739d 1762 bi->bi_phys_segments++;
1da177e4
LT		 1763 spin_unlock_irq(&conf->device_lock);
1764 spin_unlock(&sh->lock);
1765
45b4233c 1766 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
1da177e4
LT		 1767 (unsigned long long)bi->bi_sector,
1768 (unsigned long long)sh->sector, dd_idx);
1769
72626685 1770 if (conf->mddev->bitmap && firstwrite) {
72626685
N		 1771 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
1772 STRIPE_SECTORS, 0);
ae3c20cc 1773 sh->bm_seq = conf->seq_flush+1;
72626685
N		 1774 set_bit(STRIPE_BIT_DELAY, &sh->state);
1775 }
1776
1da177e4
LT		 1777 if (forwrite) {
1778 /* check if page is covered */
1779 sector_t sector = sh->dev[dd_idx].sector;
1780 for (bi=sh->dev[dd_idx].towrite;
1781 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
1782 bi && bi->bi_sector <= sector;
1783 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
1784 if (bi->bi_sector + (bi->bi_size>>9) >= sector)
1785 sector = bi->bi_sector + (bi->bi_size>>9);
1786 }
1787 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
1788 set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags);
1789 }
1790 return 1;
1791
1792 overlap:
1793 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
1794 spin_unlock_irq(&conf->device_lock);
1795 spin_unlock(&sh->lock);
1796 return 0;
1797}
1798
29269553
N		 1799static void end_reshape(raid5_conf_t *conf);
1800
16a53ecc
N		 1801static int page_is_zero(struct page *p)
1802{
1803 char *a = page_address(p);
1804 return ((*(u32*)a) == 0 &&
1805 memcmp(a, a+4, STRIPE_SIZE-4)==0);
1806}
1807
ccfcc3c1
N		 1808static int stripe_to_pdidx(sector_t stripe, raid5_conf_t *conf, int disks)
1809{
1810 int sectors_per_chunk = conf->chunk_size >> 9;
ccfcc3c1 1811 int pd_idx, dd_idx;
2d2063ce
CQH		 1812 int chunk_offset = sector_div(stripe, sectors_per_chunk);
1813
b875e531
N		 1814 raid5_compute_sector(stripe * (disks - conf->max_degraded)
1815 *sectors_per_chunk + chunk_offset,
1816 disks, disks - conf->max_degraded,
1817 &dd_idx, &pd_idx, conf);
ccfcc3c1
N		 1818 return pd_idx;
1819}
1820
a4456856 1821static void
1fe797e6 1822handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh,
a4456856
DW		 1823 struct stripe_head_state *s, int disks,
1824 struct bio **return_bi)
1825{
1826 int i;
1827 for (i = disks; i--; ) {
1828 struct bio *bi;
1829 int bitmap_end = 0;
1830
1831 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
1832 mdk_rdev_t *rdev;
1833 rcu_read_lock();
1834 rdev = rcu_dereference(conf->disks[i].rdev);
1835 if (rdev && test_bit(In_sync, &rdev->flags))
1836 /* multiple read failures in one stripe */
1837 md_error(conf->mddev, rdev);
1838 rcu_read_unlock();
1839 }
1840 spin_lock_irq(&conf->device_lock);
1841 /* fail all writes first */
1842 bi = sh->dev[i].towrite;
1843 sh->dev[i].towrite = NULL;
1844 if (bi) {
1845 s->to_write--;
1846 bitmap_end = 1;
1847 }
1848
1849 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
1850 wake_up(&conf->wait_for_overlap);
1851
1852 while (bi && bi->bi_sector <
1853 sh->dev[i].sector + STRIPE_SECTORS) {
1854 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
1855 clear_bit(BIO_UPTODATE, &bi->bi_flags);
960e739d 1856 if (!raid5_dec_bi_phys_segments(bi)) {
a4456856
DW		 1857 md_write_end(conf->mddev);
1858 bi->bi_next = *return_bi;
1859 *return_bi = bi;
1860 }
1861 bi = nextbi;
1862 }
1863 /* and fail all 'written' */
1864 bi = sh->dev[i].written;
1865 sh->dev[i].written = NULL;
1866 if (bi) bitmap_end = 1;
1867 while (bi && bi->bi_sector <
1868 sh->dev[i].sector + STRIPE_SECTORS) {
1869 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
1870 clear_bit(BIO_UPTODATE, &bi->bi_flags);
960e739d 1871 if (!raid5_dec_bi_phys_segments(bi)) {
a4456856
DW		 1872 md_write_end(conf->mddev);
1873 bi->bi_next = *return_bi;
1874 *return_bi = bi;
1875 }
1876 bi = bi2;
1877 }
1878
b5e98d65
DW		 1879 /* fail any reads if this device is non-operational and
1880 * the data has not reached the cache yet.
1881 */
1882 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
1883 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
1884 test_bit(R5_ReadError, &sh->dev[i].flags))) {
a4456856
DW		 1885 bi = sh->dev[i].toread;
1886 sh->dev[i].toread = NULL;
1887 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
1888 wake_up(&conf->wait_for_overlap);
1889 if (bi) s->to_read--;
1890 while (bi && bi->bi_sector <
1891 sh->dev[i].sector + STRIPE_SECTORS) {
1892 struct bio *nextbi =
1893 r5_next_bio(bi, sh->dev[i].sector);
1894 clear_bit(BIO_UPTODATE, &bi->bi_flags);
960e739d 1895 if (!raid5_dec_bi_phys_segments(bi)) {
a4456856
DW		 1896 bi->bi_next = *return_bi;
1897 *return_bi = bi;
1898 }
1899 bi = nextbi;
1900 }
1901 }
1902 spin_unlock_irq(&conf->device_lock);
1903 if (bitmap_end)
1904 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
1905 STRIPE_SECTORS, 0, 0);
1906 }
1907
8b3e6cdc
DW		 1908 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
1909 if (atomic_dec_and_test(&conf->pending_full_writes))
1910 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 1911}
1912
1fe797e6
DW		 1913/* fetch_block5 - checks the given member device to see if its data needs
1914 * to be read or computed to satisfy a request.
1915 *
1916 * Returns 1 when no more member devices need to be checked, otherwise returns
1917 * 0 to tell the loop in handle_stripe_fill5 to continue
f38e1219 1918 */
1fe797e6
DW		 1919static int fetch_block5(struct stripe_head *sh, struct stripe_head_state *s,
1920 int disk_idx, int disks)
f38e1219
DW		 1921{
1922 struct r5dev *dev = &sh->dev[disk_idx];
1923 struct r5dev *failed_dev = &sh->dev[s->failed_num];
1924
f38e1219
DW		 1925 /* is the data in this block needed, and can we get it? */
1926 if (!test_bit(R5_LOCKED, &dev->flags) &&
1fe797e6
DW		 1927 !test_bit(R5_UPTODATE, &dev->flags) &&
1928 (dev->toread ||
1929 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
1930 s->syncing || s->expanding ||
1931 (s->failed &&
1932 (failed_dev->toread ||
1933 (failed_dev->towrite &&
1934 !test_bit(R5_OVERWRITE, &failed_dev->flags)))))) {
976ea8d4
DW		 1935 /* We would like to get this block, possibly by computing it,
1936 * otherwise read it if the backing disk is insync
f38e1219
DW		 1937 */
1938 if ((s->uptodate == disks - 1) &&
ecc65c9b 1939 (s->failed && disk_idx == s->failed_num)) {
976ea8d4
DW		 1940 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
1941 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
f38e1219
DW		 1942 set_bit(R5_Wantcompute, &dev->flags);
1943 sh->ops.target = disk_idx;
1944 s->req_compute = 1;
f38e1219
DW		 1945 /* Careful: from this point on 'uptodate' is in the eye
1946 * of raid5_run_ops which services 'compute' operations
1947 * before writes. R5_Wantcompute flags a block that will
1948 * be R5_UPTODATE by the time it is needed for a
1949 * subsequent operation.
1950 */
1951 s->uptodate++;
1fe797e6 1952 return 1; /* uptodate + compute == disks */
7a1fc53c 1953 } else if (test_bit(R5_Insync, &dev->flags)) {
f38e1219
DW		 1954 set_bit(R5_LOCKED, &dev->flags);
1955 set_bit(R5_Wantread, &dev->flags);
f38e1219
DW		 1956 s->locked++;
1957 pr_debug("Reading block %d (sync=%d)\n", disk_idx,
1958 s->syncing);
1959 }
1960 }
1961
1fe797e6 1962 return 0;
f38e1219
DW		 1963}
1964
1fe797e6
DW		 1965/**
1966 * handle_stripe_fill5 - read or compute data to satisfy pending requests.
1967 */
1968static void handle_stripe_fill5(struct stripe_head *sh,
a4456856
DW		 1969 struct stripe_head_state *s, int disks)
1970{
1971 int i;
f38e1219 1972
f38e1219
DW		 1973 /* look for blocks to read/compute, skip this if a compute
1974 * is already in flight, or if the stripe contents are in the
1975 * midst of changing due to a write
1976 */
976ea8d4 1977 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
1fe797e6 1978 !sh->reconstruct_state)
f38e1219 1979 for (i = disks; i--; )
1fe797e6 1980 if (fetch_block5(sh, s, i, disks))
f38e1219 1981 break;
a4456856
DW		 1982 set_bit(STRIPE_HANDLE, &sh->state);
1983}
1984
1fe797e6 1985static void handle_stripe_fill6(struct stripe_head *sh,
a4456856
DW		 1986 struct stripe_head_state *s, struct r6_state *r6s,
1987 int disks)
1988{
1989 int i;
1990 for (i = disks; i--; ) {
1991 struct r5dev *dev = &sh->dev[i];
1992 if (!test_bit(R5_LOCKED, &dev->flags) &&
1993 !test_bit(R5_UPTODATE, &dev->flags) &&
1994 (dev->toread || (dev->towrite &&
1995 !test_bit(R5_OVERWRITE, &dev->flags)) ||
1996 s->syncing || s->expanding ||
1997 (s->failed >= 1 &&
1998 (sh->dev[r6s->failed_num[0]].toread ||
1999 s->to_write)) ||
2000 (s->failed >= 2 &&
2001 (sh->dev[r6s->failed_num[1]].toread ||
2002 s->to_write)))) {
2003 /* we would like to get this block, possibly
2004 * by computing it, but we might not be able to
2005 */
c337869d
DW		 2006 if ((s->uptodate == disks - 1) &&
2007 (s->failed && (i == r6s->failed_num[0] ||
2008 i == r6s->failed_num[1]))) {
45b4233c 2009 pr_debug("Computing stripe %llu block %d\n",
a4456856
DW		 2010 (unsigned long long)sh->sector, i);
2011 compute_block_1(sh, i, 0);
2012 s->uptodate++;
2013 } else if ( s->uptodate == disks-2 && s->failed >= 2 ) {
2014 /* Computing 2-failure is *very* expensive; only
2015 * do it if failed >= 2
2016 */
2017 int other;
2018 for (other = disks; other--; ) {
2019 if (other == i)
2020 continue;
2021 if (!test_bit(R5_UPTODATE,
2022 &sh->dev[other].flags))
2023 break;
2024 }
2025 BUG_ON(other < 0);
45b4233c 2026 pr_debug("Computing stripe %llu blocks %d,%d\n",
a4456856
DW		 2027 (unsigned long long)sh->sector,
2028 i, other);
2029 compute_block_2(sh, i, other);
2030 s->uptodate += 2;
2031 } else if (test_bit(R5_Insync, &dev->flags)) {
2032 set_bit(R5_LOCKED, &dev->flags);
2033 set_bit(R5_Wantread, &dev->flags);
2034 s->locked++;
45b4233c 2035 pr_debug("Reading block %d (sync=%d)\n",
a4456856
DW		 2036 i, s->syncing);
2037 }
2038 }
2039 }
2040 set_bit(STRIPE_HANDLE, &sh->state);
2041}
2042
2043
1fe797e6 2044/* handle_stripe_clean_event
a4456856
DW		 2045 * any written block on an uptodate or failed drive can be returned.
2046 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
2047 * never LOCKED, so we don't need to test 'failed' directly.
2048 */
1fe797e6 2049static void handle_stripe_clean_event(raid5_conf_t *conf,
a4456856
DW		 2050 struct stripe_head *sh, int disks, struct bio **return_bi)
2051{
2052 int i;
2053 struct r5dev *dev;
2054
2055 for (i = disks; i--; )
2056 if (sh->dev[i].written) {
2057 dev = &sh->dev[i];
2058 if (!test_bit(R5_LOCKED, &dev->flags) &&
2059 test_bit(R5_UPTODATE, &dev->flags)) {
2060 /* We can return any write requests */
2061 struct bio *wbi, *wbi2;
2062 int bitmap_end = 0;
45b4233c 2063 pr_debug("Return write for disc %d\n", i);
a4456856
DW		 2064 spin_lock_irq(&conf->device_lock);
2065 wbi = dev->written;
2066 dev->written = NULL;
2067 while (wbi && wbi->bi_sector <
2068 dev->sector + STRIPE_SECTORS) {
2069 wbi2 = r5_next_bio(wbi, dev->sector);
960e739d 2070 if (!raid5_dec_bi_phys_segments(wbi)) {
a4456856
DW		 2071 md_write_end(conf->mddev);
2072 wbi->bi_next = *return_bi;
2073 *return_bi = wbi;
2074 }
2075 wbi = wbi2;
2076 }
2077 if (dev->towrite == NULL)
2078 bitmap_end = 1;
2079 spin_unlock_irq(&conf->device_lock);
2080 if (bitmap_end)
2081 bitmap_endwrite(conf->mddev->bitmap,
2082 sh->sector,
2083 STRIPE_SECTORS,
2084 !test_bit(STRIPE_DEGRADED, &sh->state),
2085 0);
2086 }
2087 }
8b3e6cdc
DW		 2088
2089 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2090 if (atomic_dec_and_test(&conf->pending_full_writes))
2091 md_wakeup_thread(conf->mddev->thread);
a4456856
DW		 2092}
2093
1fe797e6 2094static void handle_stripe_dirtying5(raid5_conf_t *conf,
a4456856
DW		 2095 struct stripe_head *sh, struct stripe_head_state *s, int disks)
2096{
2097 int rmw = 0, rcw = 0, i;
2098 for (i = disks; i--; ) {
2099 /* would I have to read this buffer for read_modify_write */
2100 struct r5dev *dev = &sh->dev[i];
2101 if ((dev->towrite || i == sh->pd_idx) &&
2102 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 2103 !(test_bit(R5_UPTODATE, &dev->flags) ||
2104 test_bit(R5_Wantcompute, &dev->flags))) {
a4456856
DW		 2105 if (test_bit(R5_Insync, &dev->flags))
2106 rmw++;
2107 else
2108 rmw += 2*disks; /* cannot read it */
2109 }
2110 /* Would I have to read this buffer for reconstruct_write */
2111 if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx &&
2112 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 2113 !(test_bit(R5_UPTODATE, &dev->flags) ||
2114 test_bit(R5_Wantcompute, &dev->flags))) {
2115 if (test_bit(R5_Insync, &dev->flags)) rcw++;
a4456856
DW		 2116 else
2117 rcw += 2*disks;
2118 }
2119 }
45b4233c 2120 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
a4456856
DW		 2121 (unsigned long long)sh->sector, rmw, rcw);
2122 set_bit(STRIPE_HANDLE, &sh->state);
2123 if (rmw < rcw && rmw > 0)
2124 /* prefer read-modify-write, but need to get some data */
2125 for (i = disks; i--; ) {
2126 struct r5dev *dev = &sh->dev[i];
2127 if ((dev->towrite || i == sh->pd_idx) &&
2128 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 2129 !(test_bit(R5_UPTODATE, &dev->flags) ||
2130 test_bit(R5_Wantcompute, &dev->flags)) &&
a4456856
DW		 2131 test_bit(R5_Insync, &dev->flags)) {
2132 if (
2133 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
45b4233c 2134 pr_debug("Read_old block "
a4456856
DW		 2135 "%d for r-m-w\n", i);
2136 set_bit(R5_LOCKED, &dev->flags);
2137 set_bit(R5_Wantread, &dev->flags);
2138 s->locked++;
2139 } else {
2140 set_bit(STRIPE_DELAYED, &sh->state);
2141 set_bit(STRIPE_HANDLE, &sh->state);
2142 }
2143 }
2144 }
2145 if (rcw <= rmw && rcw > 0)
2146 /* want reconstruct write, but need to get some data */
2147 for (i = disks; i--; ) {
2148 struct r5dev *dev = &sh->dev[i];
2149 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
2150 i != sh->pd_idx &&
2151 !test_bit(R5_LOCKED, &dev->flags) &&
f38e1219
DW		 2152 !(test_bit(R5_UPTODATE, &dev->flags) ||
2153 test_bit(R5_Wantcompute, &dev->flags)) &&
a4456856
DW		 2154 test_bit(R5_Insync, &dev->flags)) {
2155 if (
2156 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
45b4233c 2157 pr_debug("Read_old block "
a4456856
DW		 2158 "%d for Reconstruct\n", i);
2159 set_bit(R5_LOCKED, &dev->flags);
2160 set_bit(R5_Wantread, &dev->flags);
2161 s->locked++;
2162 } else {
2163 set_bit(STRIPE_DELAYED, &sh->state);
2164 set_bit(STRIPE_HANDLE, &sh->state);
2165 }
2166 }
2167 }
2168 /* now if nothing is locked, and if we have enough data,
2169 * we can start a write request
2170 */
f38e1219
DW		 2171 /* since handle_stripe can be called at any time we need to handle the
2172 * case where a compute block operation has been submitted and then a
2173 * subsequent call wants to start a write request. raid5_run_ops only
2174 * handles the case where compute block and postxor are requested
2175 * simultaneously. If this is not the case then new writes need to be
2176 * held off until the compute completes.
2177 */
976ea8d4
DW		 2178 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
2179 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
2180 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
1fe797e6 2181 schedule_reconstruction5(sh, s, rcw == 0, 0);
a4456856
DW		 2182}
2183
1fe797e6 2184static void handle_stripe_dirtying6(raid5_conf_t *conf,
a4456856
DW		 2185 struct stripe_head *sh, struct stripe_head_state *s,
2186 struct r6_state *r6s, int disks)
2187{
2188 int rcw = 0, must_compute = 0, pd_idx = sh->pd_idx, i;
2189 int qd_idx = r6s->qd_idx;
2190 for (i = disks; i--; ) {
2191 struct r5dev *dev = &sh->dev[i];
2192 /* Would I have to read this buffer for reconstruct_write */
2193 if (!test_bit(R5_OVERWRITE, &dev->flags)
2194 && i != pd_idx && i != qd_idx
2195 && (!test_bit(R5_LOCKED, &dev->flags)
2196 ) &&
2197 !test_bit(R5_UPTODATE, &dev->flags)) {
2198 if (test_bit(R5_Insync, &dev->flags)) rcw++;
2199 else {
45b4233c 2200 pr_debug("raid6: must_compute: "
a4456856
DW		 2201 "disk %d flags=%#lx\n", i, dev->flags);
2202 must_compute++;
2203 }
2204 }
2205 }
45b4233c 2206 pr_debug("for sector %llu, rcw=%d, must_compute=%d\n",
a4456856
DW		 2207 (unsigned long long)sh->sector, rcw, must_compute);
2208 set_bit(STRIPE_HANDLE, &sh->state);
2209
2210 if (rcw > 0)
2211 /* want reconstruct write, but need to get some data */
2212 for (i = disks; i--; ) {
2213 struct r5dev *dev = &sh->dev[i];
2214 if (!test_bit(R5_OVERWRITE, &dev->flags)
2215 && !(s->failed == 0 && (i == pd_idx || i == qd_idx))
2216 && !test_bit(R5_LOCKED, &dev->flags) &&
2217 !test_bit(R5_UPTODATE, &dev->flags) &&
2218 test_bit(R5_Insync, &dev->flags)) {
2219 if (
2220 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
45b4233c 2221 pr_debug("Read_old stripe %llu "
a4456856
DW		 2222 "block %d for Reconstruct\n",
2223 (unsigned long long)sh->sector, i);
2224 set_bit(R5_LOCKED, &dev->flags);
2225 set_bit(R5_Wantread, &dev->flags);
2226 s->locked++;
2227 } else {
45b4233c 2228 pr_debug("Request delayed stripe %llu "
a4456856
DW		 2229 "block %d for Reconstruct\n",
2230 (unsigned long long)sh->sector, i);
2231 set_bit(STRIPE_DELAYED, &sh->state);
2232 set_bit(STRIPE_HANDLE, &sh->state);
2233 }
2234 }
2235 }
2236 /* now if nothing is locked, and if we have enough data, we can start a
2237 * write request
2238 */
2239 if (s->locked == 0 && rcw == 0 &&
2240 !test_bit(STRIPE_BIT_DELAY, &sh->state)) {
2241 if (must_compute > 0) {
2242 /* We have failed blocks and need to compute them */
2243 switch (s->failed) {
2244 case 0:
2245 BUG();
2246 case 1:
2247 compute_block_1(sh, r6s->failed_num[0], 0);
2248 break;
2249 case 2:
2250 compute_block_2(sh, r6s->failed_num[0],
2251 r6s->failed_num[1]);
2252 break;
2253 default: /* This request should have been failed? */
2254 BUG();
2255 }
2256 }
2257
45b4233c 2258 pr_debug("Computing parity for stripe %llu\n",
a4456856
DW		 2259 (unsigned long long)sh->sector);
2260 compute_parity6(sh, RECONSTRUCT_WRITE);
2261 /* now every locked buffer is ready to be written */
2262 for (i = disks; i--; )
2263 if (test_bit(R5_LOCKED, &sh->dev[i].flags)) {
45b4233c 2264 pr_debug("Writing stripe %llu block %d\n",
a4456856
DW		 2265 (unsigned long long)sh->sector, i);
2266 s->locked++;
2267 set_bit(R5_Wantwrite, &sh->dev[i].flags);
2268 }
8b3e6cdc
DW		 2269 if (s->locked == disks)
2270 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
2271 atomic_inc(&conf->pending_full_writes);
a4456856
DW		 2272 /* after a RECONSTRUCT_WRITE, the stripe MUST be in-sync */
2273 set_bit(STRIPE_INSYNC, &sh->state);
2274
2275 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
2276 atomic_dec(&conf->preread_active_stripes);
2277 if (atomic_read(&conf->preread_active_stripes) <
2278 IO_THRESHOLD)
2279 md_wakeup_thread(conf->mddev->thread);
2280 }
2281 }
2282}
2283
2284static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh,
2285 struct stripe_head_state *s, int disks)
2286{
ecc65c9b 2287 struct r5dev *dev = NULL;
bd2ab670 2288
a4456856 2289 set_bit(STRIPE_HANDLE, &sh->state);
e89f8962 2290
ecc65c9b
DW		 2291 switch (sh->check_state) {
2292 case check_state_idle:
2293 /* start a new check operation if there are no failures */
bd2ab670 2294 if (s->failed == 0) {
bd2ab670 2295 BUG_ON(s->uptodate != disks);
ecc65c9b
DW		 2296 sh->check_state = check_state_run;
2297 set_bit(STRIPE_OP_CHECK, &s->ops_request);
bd2ab670 2298 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
bd2ab670 2299 s->uptodate--;
ecc65c9b 2300 break;
bd2ab670 2301 }
ecc65c9b
DW		 2302 dev = &sh->dev[s->failed_num];
2303 /* fall through */
2304 case check_state_compute_result:
2305 sh->check_state = check_state_idle;
2306 if (!dev)
2307 dev = &sh->dev[sh->pd_idx];
2308
2309 /* check that a write has not made the stripe insync */
2310 if (test_bit(STRIPE_INSYNC, &sh->state))
2311 break;
c8894419 2312
a4456856 2313 /* either failed parity check, or recovery is happening */
a4456856
DW		 2314 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
2315 BUG_ON(s->uptodate != disks);
2316
2317 set_bit(R5_LOCKED, &dev->flags);
ecc65c9b 2318 s->locked++;
a4456856 2319 set_bit(R5_Wantwrite, &dev->flags);
830ea016 2320
a4456856 2321 clear_bit(STRIPE_DEGRADED, &sh->state);
a4456856 2322 set_bit(STRIPE_INSYNC, &sh->state);
ecc65c9b
DW		 2323 break;
2324 case check_state_run:
2325 break; /* we will be called again upon completion */
2326 case check_state_check_result:
2327 sh->check_state = check_state_idle;
2328
2329 /* if a failure occurred during the check operation, leave
2330 * STRIPE_INSYNC not set and let the stripe be handled again
2331 */
2332 if (s->failed)
2333 break;
2334
2335 /* handle a successful check operation, if parity is correct
2336 * we are done. Otherwise update the mismatch count and repair
2337 * parity if !MD_RECOVERY_CHECK
2338 */
2339 if (sh->ops.zero_sum_result == 0)
2340 /* parity is correct (on disc,
2341 * not in buffer any more)
2342 */
2343 set_bit(STRIPE_INSYNC, &sh->state);
2344 else {
2345 conf->mddev->resync_mismatches += STRIPE_SECTORS;
2346 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
2347 /* don't try to repair!! */
2348 set_bit(STRIPE_INSYNC, &sh->state);
2349 else {
2350 sh->check_state = check_state_compute_run;
976ea8d4 2351 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
ecc65c9b
DW		 2352 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2353 set_bit(R5_Wantcompute,
2354 &sh->dev[sh->pd_idx].flags);
2355 sh->ops.target = sh->pd_idx;
2356 s->uptodate++;
2357 }
2358 }
2359 break;
2360 case check_state_compute_run:
2361 break;
2362 default:
2363 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
2364 __func__, sh->check_state,
2365 (unsigned long long) sh->sector);
2366 BUG();
a4456856
DW		 2367 }
2368}
2369
2370
2371static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh,
2372 struct stripe_head_state *s,
2373 struct r6_state *r6s, struct page *tmp_page,
2374 int disks)
2375{
2376 int update_p = 0, update_q = 0;
2377 struct r5dev *dev;
2378 int pd_idx = sh->pd_idx;
2379 int qd_idx = r6s->qd_idx;
2380
2381 set_bit(STRIPE_HANDLE, &sh->state);
2382
2383 BUG_ON(s->failed > 2);
2384 BUG_ON(s->uptodate < disks);
2385 /* Want to check and possibly repair P and Q.
2386 * However there could be one 'failed' device, in which
2387 * case we can only check one of them, possibly using the
2388 * other to generate missing data
2389 */
2390
2391 /* If !tmp_page, we cannot do the calculations,
2392 * but as we have set STRIPE_HANDLE, we will soon be called
2393 * by stripe_handle with a tmp_page - just wait until then.
2394 */
2395 if (tmp_page) {
2396 if (s->failed == r6s->q_failed) {
2397 /* The only possible failed device holds 'Q', so it
2398 * makes sense to check P (If anything else were failed,
2399 * we would have used P to recreate it).
2400 */
2401 compute_block_1(sh, pd_idx, 1);
2402 if (!page_is_zero(sh->dev[pd_idx].page)) {
2403 compute_block_1(sh, pd_idx, 0);
2404 update_p = 1;
2405 }
2406 }
2407 if (!r6s->q_failed && s->failed < 2) {
2408 /* q is not failed, and we didn't use it to generate
2409 * anything, so it makes sense to check it
2410 */
2411 memcpy(page_address(tmp_page),
2412 page_address(sh->dev[qd_idx].page),
2413 STRIPE_SIZE);
2414 compute_parity6(sh, UPDATE_PARITY);
2415 if (memcmp(page_address(tmp_page),
2416 page_address(sh->dev[qd_idx].page),
2417 STRIPE_SIZE) != 0) {
2418 clear_bit(STRIPE_INSYNC, &sh->state);
2419 update_q = 1;
2420 }
2421 }
2422 if (update_p || update_q) {
2423 conf->mddev->resync_mismatches += STRIPE_SECTORS;
2424 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
2425 /* don't try to repair!! */
2426 update_p = update_q = 0;
2427 }
2428
2429 /* now write out any block on a failed drive,
2430 * or P or Q if they need it
2431 */
2432
2433 if (s->failed == 2) {
2434 dev = &sh->dev[r6s->failed_num[1]];
2435 s->locked++;
2436 set_bit(R5_LOCKED, &dev->flags);
2437 set_bit(R5_Wantwrite, &dev->flags);
2438 }
2439 if (s->failed >= 1) {
2440 dev = &sh->dev[r6s->failed_num[0]];
2441 s->locked++;
2442 set_bit(R5_LOCKED, &dev->flags);
2443 set_bit(R5_Wantwrite, &dev->flags);
2444 }
2445
2446 if (update_p) {
2447 dev = &sh->dev[pd_idx];
2448 s->locked++;
2449 set_bit(R5_LOCKED, &dev->flags);
2450 set_bit(R5_Wantwrite, &dev->flags);
2451 }
2452 if (update_q) {
2453 dev = &sh->dev[qd_idx];
2454 s->locked++;
2455 set_bit(R5_LOCKED, &dev->flags);
2456 set_bit(R5_Wantwrite, &dev->flags);
2457 }
2458 clear_bit(STRIPE_DEGRADED, &sh->state);
2459
2460 set_bit(STRIPE_INSYNC, &sh->state);
2461 }
2462}
2463
2464static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh,
2465 struct r6_state *r6s)
2466{
2467 int i;
2468
2469 /* We have read all the blocks in this stripe and now we need to
2470 * copy some of them into a target stripe for expand.
2471 */
f0a50d37 2472 struct dma_async_tx_descriptor *tx = NULL;
a4456856
DW		 2473 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
2474 for (i = 0; i < sh->disks; i++)
a2e08551 2475 if (i != sh->pd_idx && (!r6s || i != r6s->qd_idx)) {
a4456856
DW		 2476 int dd_idx, pd_idx, j;
2477 struct stripe_head *sh2;
2478
2479 sector_t bn = compute_blocknr(sh, i);
2480 sector_t s = raid5_compute_sector(bn, conf->raid_disks,
2481 conf->raid_disks -
2482 conf->max_degraded, &dd_idx,
2483 &pd_idx, conf);
2484 sh2 = get_active_stripe(conf, s, conf->raid_disks,
2485 pd_idx, 1);
2486 if (sh2 == NULL)
2487 /* so far only the early blocks of this stripe
2488 * have been requested. When later blocks
2489 * get requested, we will try again
2490 */
2491 continue;
2492 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
2493 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
2494 /* must have already done this block */
2495 release_stripe(sh2);
2496 continue;
2497 }
f0a50d37
DW		 2498
2499 /* place all the copies on one channel */
2500 tx = async_memcpy(sh2->dev[dd_idx].page,
2501 sh->dev[i].page, 0, 0, STRIPE_SIZE,
2502 ASYNC_TX_DEP_ACK, tx, NULL, NULL);
2503
a4456856
DW		 2504 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
2505 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
2506 for (j = 0; j < conf->raid_disks; j++)
2507 if (j != sh2->pd_idx &&
a2e08551
N		 2508 (!r6s || j != raid6_next_disk(sh2->pd_idx,
2509 sh2->disks)) &&
a4456856
DW		 2510 !test_bit(R5_Expanded, &sh2->dev[j].flags))
2511 break;
2512 if (j == conf->raid_disks) {
2513 set_bit(STRIPE_EXPAND_READY, &sh2->state);
2514 set_bit(STRIPE_HANDLE, &sh2->state);
2515 }
2516 release_stripe(sh2);
f0a50d37 2517
a4456856 2518 }
a2e08551
N		 2519 /* done submitting copies, wait for them to complete */
2520 if (tx) {
2521 async_tx_ack(tx);
2522 dma_wait_for_async_tx(tx);
2523 }
a4456856 2524}
1da177e4 2525
6bfe0b49 2526
1da177e4
LT		 2527/*
2528 * handle_stripe - do things to a stripe.
2529 *
2530 * We lock the stripe and then examine the state of various bits
2531 * to see what needs to be done.
2532 * Possible results:
2533 * return some read request which now have data
2534 * return some write requests which are safely on disc
2535 * schedule a read on some buffers
2536 * schedule a write of some buffers
2537 * return confirmation of parity correctness
2538 *
1da177e4
LT		 2539 * buffers are taken off read_list or write_list, and bh_cache buffers
2540 * get BH_Lock set before the stripe lock is released.
2541 *
2542 */
a4456856 2543
df10cfbc 2544static bool handle_stripe5(struct stripe_head *sh)
1da177e4
LT		 2545{
2546 raid5_conf_t *conf = sh->raid_conf;
a4456856
DW		 2547 int disks = sh->disks, i;
2548 struct bio *return_bi = NULL;
2549 struct stripe_head_state s;
1da177e4 2550 struct r5dev *dev;
6bfe0b49 2551 mdk_rdev_t *blocked_rdev = NULL;
e0a115e5 2552 int prexor;
1da177e4 2553
a4456856 2554 memset(&s, 0, sizeof(s));
600aa109
DW		 2555 pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d check:%d "
2556 "reconstruct:%d\n", (unsigned long long)sh->sector, sh->state,
2557 atomic_read(&sh->count), sh->pd_idx, sh->check_state,
2558 sh->reconstruct_state);
1da177e4
LT		 2559
2560 spin_lock(&sh->lock);
2561 clear_bit(STRIPE_HANDLE, &sh->state);
2562 clear_bit(STRIPE_DELAYED, &sh->state);
2563
a4456856
DW		 2564 s.syncing = test_bit(STRIPE_SYNCING, &sh->state);
2565 s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
2566 s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
def6ae26 2567
83de75cc 2568 /* Now to look around and see what can be done */
9910f16a 2569 rcu_read_lock();
1da177e4
LT		 2570 for (i=disks; i--; ) {
2571 mdk_rdev_t *rdev;
a4456856 2572 struct r5dev *dev = &sh->dev[i];
1da177e4 2573 clear_bit(R5_Insync, &dev->flags);
1da177e4 2574
b5e98d65
DW		 2575 pr_debug("check %d: state 0x%lx toread %p read %p write %p "
2576 "written %p\n", i, dev->flags, dev->toread, dev->read,
2577 dev->towrite, dev->written);
2578
2579 /* maybe we can request a biofill operation
2580 *
2581 * new wantfill requests are only permitted while
83de75cc 2582 * ops_complete_biofill is guaranteed to be inactive
b5e98d65
DW		 2583 */
2584 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
83de75cc 2585 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
b5e98d65 2586 set_bit(R5_Wantfill, &dev->flags);
1da177e4
LT		 2587
2588 /* now count some things */
a4456856
DW		 2589 if (test_bit(R5_LOCKED, &dev->flags)) s.locked++;
2590 if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++;
f38e1219 2591 if (test_bit(R5_Wantcompute, &dev->flags)) s.compute++;
1da177e4 2592
b5e98d65
DW		 2593 if (test_bit(R5_Wantfill, &dev->flags))
2594 s.to_fill++;
2595 else if (dev->toread)
a4456856 2596 s.to_read++;
1da177e4 2597 if (dev->towrite) {
a4456856 2598 s.to_write++;
1da177e4 2599 if (!test_bit(R5_OVERWRITE, &dev->flags))
a4456856 2600 s.non_overwrite++;
1da177e4 2601 }
a4456856
DW		 2602 if (dev->written)
2603 s.written++;
9910f16a 2604 rdev = rcu_dereference(conf->disks[i].rdev);
ac4090d2
N		 2605 if (blocked_rdev == NULL &&
2606 rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
6bfe0b49
DW		 2607 blocked_rdev = rdev;
2608 atomic_inc(&rdev->nr_pending);
6bfe0b49 2609 }
b2d444d7 2610 if (!rdev || !test_bit(In_sync, &rdev->flags)) {
14f8d26b 2611 /* The ReadError flag will just be confusing now */
4e5314b5
N		 2612 clear_bit(R5_ReadError, &dev->flags);
2613 clear_bit(R5_ReWrite, &dev->flags);
2614 }
b2d444d7 2615 if (!rdev || !test_bit(In_sync, &rdev->flags)
4e5314b5 2616 || test_bit(R5_ReadError, &dev->flags)) {
a4456856
DW		 2617 s.failed++;
2618 s.failed_num = i;
1da177e4
LT		 2619 } else
2620 set_bit(R5_Insync, &dev->flags);
2621 }
9910f16a 2622 rcu_read_unlock();
b5e98d65 2623
6bfe0b49 2624 if (unlikely(blocked_rdev)) {
ac4090d2
N		 2625 if (s.syncing || s.expanding || s.expanded ||
2626 s.to_write || s.written) {
2627 set_bit(STRIPE_HANDLE, &sh->state);
2628 goto unlock;
2629 }
2630 /* There is nothing for the blocked_rdev to block */
2631 rdev_dec_pending(blocked_rdev, conf->mddev);
2632 blocked_rdev = NULL;
6bfe0b49
DW		 2633 }
2634
83de75cc
DW		 2635 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
2636 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
2637 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
2638 }
b5e98d65 2639
45b4233c 2640 pr_debug("locked=%d uptodate=%d to_read=%d"
1da177e4 2641 " to_write=%d failed=%d failed_num=%d\n",
a4456856
DW		 2642 s.locked, s.uptodate, s.to_read, s.to_write,
2643 s.failed, s.failed_num);
1da177e4
LT		 2644 /* check if the array has lost two devices and, if so, some requests might
2645 * need to be failed
2646 */
a4456856 2647 if (s.failed > 1 && s.to_read+s.to_write+s.written)
1fe797e6 2648 handle_failed_stripe(conf, sh, &s, disks, &return_bi);
a4456856 2649 if (s.failed > 1 && s.syncing) {
1da177e4
LT		 2650 md_done_sync(conf->mddev, STRIPE_SECTORS,0);
2651 clear_bit(STRIPE_SYNCING, &sh->state);
a4456856 2652 s.syncing = 0;
1da177e4
LT		 2653 }
2654
2655 /* might be able to return some write requests if the parity block
2656 * is safe, or on a failed drive
2657 */
2658 dev = &sh->dev[sh->pd_idx];
a4456856
DW		 2659 if ( s.written &&
2660 ((test_bit(R5_Insync, &dev->flags) &&
2661 !test_bit(R5_LOCKED, &dev->flags) &&
2662 test_bit(R5_UPTODATE, &dev->flags)) ||
2663 (s.failed == 1 && s.failed_num == sh->pd_idx)))
1fe797e6 2664 handle_stripe_clean_event(conf, sh, disks, &return_bi);
1da177e4
LT		 2665
2666 /* Now we might consider reading some blocks, either to check/generate
2667 * parity, or to satisfy requests
2668 * or to load a block that is being partially written.
2669 */
a4456856 2670 if (s.to_read || s.non_overwrite ||
976ea8d4 2671 (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding)
1fe797e6 2672 handle_stripe_fill5(sh, &s, disks);
1da177e4 2673
e33129d8
DW		 2674 /* Now we check to see if any write operations have recently
2675 * completed
2676 */
e0a115e5 2677 prexor = 0;
d8ee0728 2678 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
e0a115e5 2679 prexor = 1;
d8ee0728
DW		 2680 if (sh->reconstruct_state == reconstruct_state_drain_result ||
2681 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
600aa109 2682 sh->reconstruct_state = reconstruct_state_idle;
e33129d8
DW		 2683
2684 /* All the 'written' buffers and the parity block are ready to
2685 * be written back to disk
2686 */
2687 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags));
2688 for (i = disks; i--; ) {
2689 dev = &sh->dev[i];
2690 if (test_bit(R5_LOCKED, &dev->flags) &&
2691 (i == sh->pd_idx || dev->written)) {
2692 pr_debug("Writing block %d\n", i);
2693 set_bit(R5_Wantwrite, &dev->flags);
e0a115e5
DW		 2694 if (prexor)
2695 continue;
e33129d8
DW		 2696 if (!test_bit(R5_Insync, &dev->flags) ||
2697 (i == sh->pd_idx && s.failed == 0))
2698 set_bit(STRIPE_INSYNC, &sh->state);
2699 }
2700 }
2701 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
2702 atomic_dec(&conf->preread_active_stripes);
2703 if (atomic_read(&conf->preread_active_stripes) <
2704 IO_THRESHOLD)
2705 md_wakeup_thread(conf->mddev->thread);
2706 }
2707 }
2708
2709 /* Now to consider new write requests and what else, if anything
2710 * should be read. We do not handle new writes when:
2711 * 1/ A 'write' operation (copy+xor) is already in flight.
2712 * 2/ A 'check' operation is in flight, as it may clobber the parity
2713 * block.
2714 */
600aa109 2715 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
1fe797e6 2716 handle_stripe_dirtying5(conf, sh, &s, disks);
1da177e4
LT		 2717
2718 /* maybe we need to check and possibly fix the parity for this stripe
e89f8962
DW		 2719 * Any reads will already have been scheduled, so we just see if enough
2720 * data is available. The parity check is held off while parity
2721 * dependent operations are in flight.
1da177e4 2722 */
ecc65c9b
DW		 2723 if (sh->check_state ||
2724 (s.syncing && s.locked == 0 &&
976ea8d4 2725 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
ecc65c9b 2726 !test_bit(STRIPE_INSYNC, &sh->state)))
a4456856 2727 handle_parity_checks5(conf, sh, &s, disks);
e89f8962 2728
a4456856 2729 if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) {
1da177e4
LT		 2730 md_done_sync(conf->mddev, STRIPE_SECTORS,1);
2731 clear_bit(STRIPE_SYNCING, &sh->state);
2732 }
4e5314b5
N		 2733
2734 /* If the failed drive is just a ReadError, then we might need to progress
2735 * the repair/check process
2736 */
a4456856
DW		 2737 if (s.failed == 1 && !conf->mddev->ro &&
2738 test_bit(R5_ReadError, &sh->dev[s.failed_num].flags)
2739 && !test_bit(R5_LOCKED, &sh->dev[s.failed_num].flags)
2740 && test_bit(R5_UPTODATE, &sh->dev[s.failed_num].flags)
4e5314b5 2741 ) {
a4456856 2742 dev = &sh->dev[s.failed_num];
4e5314b5
N		 2743 if (!test_bit(R5_ReWrite, &dev->flags)) {
2744 set_bit(R5_Wantwrite, &dev->flags);
2745 set_bit(R5_ReWrite, &dev->flags);
2746 set_bit(R5_LOCKED, &dev->flags);
a4456856 2747 s.locked++;
4e5314b5
N		 2748 } else {
2749 /* let's read it back */
2750 set_bit(R5_Wantread, &dev->flags);
2751 set_bit(R5_LOCKED, &dev->flags);
a4456856 2752 s.locked++;
4e5314b5
N		 2753 }
2754 }
2755
600aa109
DW		 2756 /* Finish reconstruct operations initiated by the expansion process */
2757 if (sh->reconstruct_state == reconstruct_state_result) {
2758 sh->reconstruct_state = reconstruct_state_idle;
f0a50d37 2759 clear_bit(STRIPE_EXPANDING, &sh->state);
23397883 2760 for (i = conf->raid_disks; i--; ) {
ccfcc3c1 2761 set_bit(R5_Wantwrite, &sh->dev[i].flags);
23397883 2762 set_bit(R5_LOCKED, &sh->dev[i].flags);
efe31143 2763 s.locked++;
23397883 2764 }
f0a50d37
DW		 2765 }
2766
2767 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
600aa109 2768 !sh->reconstruct_state) {
f0a50d37
DW		 2769 /* Need to write out all blocks after computing parity */
2770 sh->disks = conf->raid_disks;
2771 sh->pd_idx = stripe_to_pdidx(sh->sector, conf,
2772 conf->raid_disks);
1fe797e6 2773 schedule_reconstruction5(sh, &s, 1, 1);
600aa109 2774 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
ccfcc3c1 2775 clear_bit(STRIPE_EXPAND_READY, &sh->state);
f6705578 2776 atomic_dec(&conf->reshape_stripes);
ccfcc3c1
N		 2777 wake_up(&conf->wait_for_overlap);
2778 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
2779 }
2780
0f94e87c 2781 if (s.expanding && s.locked == 0 &&
976ea8d4 2782 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
a4456856 2783 handle_stripe_expansion(conf, sh, NULL);
ccfcc3c1 2784
6bfe0b49 2785 unlock:
1da177e4
LT		 2786 spin_unlock(&sh->lock);
2787
6bfe0b49
DW		 2788 /* wait for this device to become unblocked */
2789 if (unlikely(blocked_rdev))
2790 md_wait_for_blocked_rdev(blocked_rdev, conf->mddev);
2791
600aa109
DW		 2792 if (s.ops_request)
2793 raid5_run_ops(sh, s.ops_request);
d84e0f10 2794
c4e5ac0a 2795 ops_run_io(sh, &s);
1da177e4 2796
a4456856 2797 return_io(return_bi);
df10cfbc
DW		 2798
2799 return blocked_rdev == NULL;
1da177e4
LT		 2800}
2801
df10cfbc 2802static bool handle_stripe6(struct stripe_head *sh, struct page *tmp_page)
1da177e4 2803{
16a53ecc 2804 raid6_conf_t *conf = sh->raid_conf;
f416885e 2805 int disks = sh->disks;
a4456856
DW		 2806 struct bio *return_bi = NULL;
2807 int i, pd_idx = sh->pd_idx;
2808 struct stripe_head_state s;
2809 struct r6_state r6s;
16a53ecc 2810 struct r5dev *dev, *pdev, *qdev;
6bfe0b49 2811 mdk_rdev_t *blocked_rdev = NULL;
1da177e4 2812
a4456856 2813 r6s.qd_idx = raid6_next_disk(pd_idx, disks);
45b4233c 2814 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
a4456856
DW		 2815 "pd_idx=%d, qd_idx=%d\n",
2816 (unsigned long long)sh->sector, sh->state,
2817 atomic_read(&sh->count), pd_idx, r6s.qd_idx);
2818 memset(&s, 0, sizeof(s));
72626685 2819
16a53ecc
N		 2820 spin_lock(&sh->lock);
2821 clear_bit(STRIPE_HANDLE, &sh->state);
2822 clear_bit(STRIPE_DELAYED, &sh->state);
2823
a4456856
DW		 2824 s.syncing = test_bit(STRIPE_SYNCING, &sh->state);
2825 s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
2826 s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
16a53ecc 2827 /* Now to look around and see what can be done */
1da177e4
LT		 2828
2829 rcu_read_lock();
16a53ecc
N		 2830 for (i=disks; i--; ) {
2831 mdk_rdev_t *rdev;
2832 dev = &sh->dev[i];
2833 clear_bit(R5_Insync, &dev->flags);
1da177e4 2834
45b4233c 2835 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
16a53ecc
N		 2836 i, dev->flags, dev->toread, dev->towrite, dev->written);
2837 /* maybe we can reply to a read */
2838 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread) {
2839 struct bio *rbi, *rbi2;
45b4233c 2840 pr_debug("Return read for disc %d\n", i);
16a53ecc
N		 2841 spin_lock_irq(&conf->device_lock);
2842 rbi = dev->toread;
2843 dev->toread = NULL;
2844 if (test_and_clear_bit(R5_Overlap, &dev->flags))
2845 wake_up(&conf->wait_for_overlap);
2846 spin_unlock_irq(&conf->device_lock);
2847 while (rbi && rbi->bi_sector < dev->sector + STRIPE_SECTORS) {
2848 copy_data(0, rbi, dev->page, dev->sector);
2849 rbi2 = r5_next_bio(rbi, dev->sector);
2850 spin_lock_irq(&conf->device_lock);
960e739d 2851 if (!raid5_dec_bi_phys_segments(rbi)) {
16a53ecc
N		 2852 rbi->bi_next = return_bi;
2853 return_bi = rbi;
2854 }
2855 spin_unlock_irq(&conf->device_lock);
2856 rbi = rbi2;
2857 }
2858 }
1da177e4 2859
16a53ecc 2860 /* now count some things */
a4456856
DW		 2861 if (test_bit(R5_LOCKED, &dev->flags)) s.locked++;
2862 if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++;
1da177e4 2863
16a53ecc 2864
a4456856
DW		 2865 if (dev->toread)
2866 s.to_read++;
16a53ecc 2867 if (dev->towrite) {
a4456856 2868 s.to_write++;
16a53ecc 2869 if (!test_bit(R5_OVERWRITE, &dev->flags))
a4456856 2870 s.non_overwrite++;
16a53ecc 2871 }
a4456856
DW		 2872 if (dev->written)
2873 s.written++;
16a53ecc 2874 rdev = rcu_dereference(conf->disks[i].rdev);
ac4090d2
N		 2875 if (blocked_rdev == NULL &&
2876 rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
6bfe0b49
DW		 2877 blocked_rdev = rdev;
2878 atomic_inc(&rdev->nr_pending);
6bfe0b49 2879 }
16a53ecc
N		 2880 if (!rdev || !test_bit(In_sync, &rdev->flags)) {
2881 /* The ReadError flag will just be confusing now */
2882 clear_bit(R5_ReadError, &dev->flags);
2883 clear_bit(R5_ReWrite, &dev->flags);
1da177e4 2884 }
16a53ecc
N		 2885 if (!rdev || !test_bit(In_sync, &rdev->flags)
2886 || test_bit(R5_ReadError, &dev->flags)) {
a4456856
DW		 2887 if (s.failed < 2)
2888 r6s.failed_num[s.failed] = i;
2889 s.failed++;
16a53ecc
N		 2890 } else
2891 set_bit(R5_Insync, &dev->flags);
1da177e4
LT		 2892 }
2893 rcu_read_unlock();
6bfe0b49
DW		 2894
2895 if (unlikely(blocked_rdev)) {
ac4090d2
N		 2896 if (s.syncing || s.expanding || s.expanded ||
2897 s.to_write || s.written) {
2898 set_bit(STRIPE_HANDLE, &sh->state);
2899 goto unlock;
2900 }
2901 /* There is nothing for the blocked_rdev to block */
2902 rdev_dec_pending(blocked_rdev, conf->mddev);
2903 blocked_rdev = NULL;
6bfe0b49 2904 }
ac4090d2 2905
45b4233c 2906 pr_debug("locked=%d uptodate=%d to_read=%d"
16a53ecc 2907 " to_write=%d failed=%d failed_num=%d,%d\n",
a4456856
DW		 2908 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
2909 r6s.failed_num[0], r6s.failed_num[1]);
2910 /* check if the array has lost >2 devices and, if so, some requests
2911 * might need to be failed
16a53ecc 2912 */
a4456856 2913 if (s.failed > 2 && s.to_read+s.to_write+s.written)
1fe797e6 2914 handle_failed_stripe(conf, sh, &s, disks, &return_bi);
a4456856 2915 if (s.failed > 2 && s.syncing) {
16a53ecc
N		 2916 md_done_sync(conf->mddev, STRIPE_SECTORS,0);
2917 clear_bit(STRIPE_SYNCING, &sh->state);
a4456856 2918 s.syncing = 0;
16a53ecc
N		 2919 }
2920
2921 /*
2922 * might be able to return some write requests if the parity blocks
2923 * are safe, or on a failed drive
2924 */
2925 pdev = &sh->dev[pd_idx];
a4456856
DW		 2926 r6s.p_failed = (s.failed >= 1 && r6s.failed_num[0] == pd_idx)
2927 || (s.failed >= 2 && r6s.failed_num[1] == pd_idx);
2928 qdev = &sh->dev[r6s.qd_idx];
2929 r6s.q_failed = (s.failed >= 1 && r6s.failed_num[0] == r6s.qd_idx)
2930 || (s.failed >= 2 && r6s.failed_num[1] == r6s.qd_idx);
2931
2932 if ( s.written &&
2933 ( r6s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
16a53ecc 2934 && !test_bit(R5_LOCKED, &pdev->flags)
a4456856
DW		 2935 && test_bit(R5_UPTODATE, &pdev->flags)))) &&
2936 ( r6s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
16a53ecc 2937 && !test_bit(R5_LOCKED, &qdev->flags)
a4456856 2938 && test_bit(R5_UPTODATE, &qdev->flags)))))
1fe797e6 2939 handle_stripe_clean_event(conf, sh, disks, &return_bi);
16a53ecc
N		 2940
2941 /* Now we might consider reading some blocks, either to check/generate
2942 * parity, or to satisfy requests
2943 * or to load a block that is being partially written.
2944 */
a4456856
DW		 2945 if (s.to_read || s.non_overwrite || (s.to_write && s.failed) ||
2946 (s.syncing && (s.uptodate < disks)) || s.expanding)
1fe797e6 2947 handle_stripe_fill6(sh, &s, &r6s, disks);
16a53ecc
N		 2948
2949 /* now to consider writing and what else, if anything should be read */
a4456856 2950 if (s.to_write)
1fe797e6 2951 handle_stripe_dirtying6(conf, sh, &s, &r6s, disks);
16a53ecc
N		 2952
2953 /* maybe we need to check and possibly fix the parity for this stripe
a4456856
DW		 2954 * Any reads will already have been scheduled, so we just see if enough
2955 * data is available
16a53ecc 2956 */
a4456856
DW		 2957 if (s.syncing && s.locked == 0 && !test_bit(STRIPE_INSYNC, &sh->state))
2958 handle_parity_checks6(conf, sh, &s, &r6s, tmp_page, disks);
16a53ecc 2959
a4456856 2960 if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) {
16a53ecc
N		 2961 md_done_sync(conf->mddev, STRIPE_SECTORS,1);
2962 clear_bit(STRIPE_SYNCING, &sh->state);
2963 }
2964
2965 /* If the failed drives are just a ReadError, then we might need
2966 * to progress the repair/check process
2967 */
a4456856
DW		 2968 if (s.failed <= 2 && !conf->mddev->ro)
2969 for (i = 0; i < s.failed; i++) {
2970 dev = &sh->dev[r6s.failed_num[i]];
16a53ecc
N		 2971 if (test_bit(R5_ReadError, &dev->flags)
2972 && !test_bit(R5_LOCKED, &dev->flags)
2973 && test_bit(R5_UPTODATE, &dev->flags)
2974 ) {
2975 if (!test_bit(R5_ReWrite, &dev->flags)) {
2976 set_bit(R5_Wantwrite, &dev->flags);
2977 set_bit(R5_ReWrite, &dev->flags);
2978 set_bit(R5_LOCKED, &dev->flags);
2979 } else {
2980 /* let's read it back */
2981 set_bit(R5_Wantread, &dev->flags);
2982 set_bit(R5_LOCKED, &dev->flags);
2983 }
2984 }
2985 }
f416885e 2986
a4456856 2987 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state)) {
f416885e
N		 2988 /* Need to write out all blocks after computing P&Q */
2989 sh->disks = conf->raid_disks;
2990 sh->pd_idx = stripe_to_pdidx(sh->sector, conf,
2991 conf->raid_disks);
2992 compute_parity6(sh, RECONSTRUCT_WRITE);
2993 for (i = conf->raid_disks ; i-- ; ) {
2994 set_bit(R5_LOCKED, &sh->dev[i].flags);
a4456856 2995 s.locked++;
f416885e
N		 2996 set_bit(R5_Wantwrite, &sh->dev[i].flags);
2997 }
2998 clear_bit(STRIPE_EXPANDING, &sh->state);
a4456856 2999 } else if (s.expanded) {
f416885e
N		 3000 clear_bit(STRIPE_EXPAND_READY, &sh->state);
3001 atomic_dec(&conf->reshape_stripes);
3002 wake_up(&conf->wait_for_overlap);
3003 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3004 }
3005
0f94e87c 3006 if (s.expanding && s.locked == 0 &&
976ea8d4 3007 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
a4456856 3008 handle_stripe_expansion(conf, sh, &r6s);
f416885e 3009
6bfe0b49 3010 unlock:
16a53ecc
N		 3011 spin_unlock(&sh->lock);
3012
6bfe0b49
DW		 3013 /* wait for this device to become unblocked */
3014 if (unlikely(blocked_rdev))
3015 md_wait_for_blocked_rdev(blocked_rdev, conf->mddev);
3016
f0e43bcd 3017 ops_run_io(sh, &s);
16a53ecc 3018
f0e43bcd 3019 return_io(return_bi);
df10cfbc
DW		 3020
3021 return blocked_rdev == NULL;
16a53ecc
N		 3022}
3023
df10cfbc
DW		 3024/* returns true if the stripe was handled */
3025static bool handle_stripe(struct stripe_head *sh, struct page *tmp_page)
16a53ecc
N		 3026{
3027 if (sh->raid_conf->level == 6)
df10cfbc 3028 return handle_stripe6(sh, tmp_page);
16a53ecc 3029 else
df10cfbc 3030 return handle_stripe5(sh);
16a53ecc
N		 3031}
3032
3033
3034
3035static void raid5_activate_delayed(raid5_conf_t *conf)
3036{
3037 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
3038 while (!list_empty(&conf->delayed_list)) {
3039 struct list_head *l = conf->delayed_list.next;
3040 struct stripe_head *sh;
3041 sh = list_entry(l, struct stripe_head, lru);
3042 list_del_init(l);
3043 clear_bit(STRIPE_DELAYED, &sh->state);
3044 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3045 atomic_inc(&conf->preread_active_stripes);
8b3e6cdc 3046 list_add_tail(&sh->lru, &conf->hold_list);
16a53ecc 3047 }
6ed3003c
N		 3048 } else
3049 blk_plug_device(conf->mddev->queue);
16a53ecc
N		 3050}
3051
3052static void activate_bit_delay(raid5_conf_t *conf)
3053{
3054 /* device_lock is held */
3055 struct list_head head;
3056 list_add(&head, &conf->bitmap_list);
3057 list_del_init(&conf->bitmap_list);
3058 while (!list_empty(&head)) {
3059 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
3060 list_del_init(&sh->lru);
3061 atomic_inc(&sh->count);
3062 __release_stripe(conf, sh);
3063 }
3064}
3065
3066static void unplug_slaves(mddev_t *mddev)
3067{
3068 raid5_conf_t *conf = mddev_to_conf(mddev);
3069 int i;
3070
3071 rcu_read_lock();
3072 for (i=0; i<mddev->raid_disks; i++) {
3073 mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev);
3074 if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) {
165125e1 3075 struct request_queue *r_queue = bdev_get_queue(rdev->bdev);
16a53ecc
N		 3076
3077 atomic_inc(&rdev->nr_pending);
3078 rcu_read_unlock();
3079
2ad8b1ef 3080 blk_unplug(r_queue);
16a53ecc
N		 3081
3082 rdev_dec_pending(rdev, mddev);
3083 rcu_read_lock();
3084 }
3085 }
3086 rcu_read_unlock();
3087}
3088
165125e1 3089static void raid5_unplug_device(struct request_queue *q)
16a53ecc
N		 3090{
3091 mddev_t *mddev = q->queuedata;
3092 raid5_conf_t *conf = mddev_to_conf(mddev);
3093 unsigned long flags;
3094
3095 spin_lock_irqsave(&conf->device_lock, flags);
3096
3097 if (blk_remove_plug(q)) {
3098 conf->seq_flush++;
3099 raid5_activate_delayed(conf);
72626685 3100 }
1da177e4
LT		 3101 md_wakeup_thread(mddev->thread);
3102
3103 spin_unlock_irqrestore(&conf->device_lock, flags);
3104
3105 unplug_slaves(mddev);
3106}
3107
f022b2fd
N		 3108static int raid5_congested(void *data, int bits)
3109{
3110 mddev_t *mddev = data;
3111 raid5_conf_t *conf = mddev_to_conf(mddev);
3112
3113 /* No difference between reads and writes. Just check
3114 * how busy the stripe_cache is
3115 */
3116 if (conf->inactive_blocked)
3117 return 1;
3118 if (conf->quiesce)
3119 return 1;
3120 if (list_empty_careful(&conf->inactive_list))
3121 return 1;
3122
3123 return 0;
3124}
3125
23032a0e
RBJ		 3126/* We want read requests to align with chunks where possible,
3127 * but write requests don't need to.
3128 */
cc371e66
AK		 3129static int raid5_mergeable_bvec(struct request_queue *q,
3130 struct bvec_merge_data *bvm,
3131 struct bio_vec *biovec)
23032a0e
RBJ		 3132{
3133 mddev_t *mddev = q->queuedata;
cc371e66 3134 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
23032a0e
RBJ		 3135 int max;
3136 unsigned int chunk_sectors = mddev->chunk_size >> 9;
cc371e66 3137 unsigned int bio_sectors = bvm->bi_size >> 9;
23032a0e 3138
cc371e66 3139 if ((bvm->bi_rw & 1) == WRITE)
23032a0e
RBJ		 3140 return biovec->bv_len; /* always allow writes to be mergeable */
3141
3142 max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
3143 if (max < 0) max = 0;
3144 if (max <= biovec->bv_len && bio_sectors == 0)
3145 return biovec->bv_len;
3146 else
3147 return max;
3148}
3149
f679623f
RBJ		 3150
3151static int in_chunk_boundary(mddev_t *mddev, struct bio *bio)
3152{
3153 sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
3154 unsigned int chunk_sectors = mddev->chunk_size >> 9;
3155 unsigned int bio_sectors = bio->bi_size >> 9;
3156
3157 return chunk_sectors >=
3158 ((sector & (chunk_sectors - 1)) + bio_sectors);
3159}
3160
46031f9a
RBJ		 3161/*
3162 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
3163 * later sampled by raid5d.
3164 */
3165static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf)
3166{
3167 unsigned long flags;
3168
3169 spin_lock_irqsave(&conf->device_lock, flags);
3170
3171 bi->bi_next = conf->retry_read_aligned_list;
3172 conf->retry_read_aligned_list = bi;
3173
3174 spin_unlock_irqrestore(&conf->device_lock, flags);
3175 md_wakeup_thread(conf->mddev->thread);
3176}
3177
3178
3179static struct bio *remove_bio_from_retry(raid5_conf_t *conf)
3180{
3181 struct bio *bi;
3182
3183 bi = conf->retry_read_aligned;
3184 if (bi) {
3185 conf->retry_read_aligned = NULL;
3186 return bi;
3187 }
3188 bi = conf->retry_read_aligned_list;
3189 if(bi) {
387bb173 3190 conf->retry_read_aligned_list = bi->bi_next;
46031f9a 3191 bi->bi_next = NULL;
960e739d
JA		 3192 /*
3193 * this sets the active strip count to 1 and the processed
3194 * strip count to zero (upper 8 bits)
3195 */
46031f9a 3196 bi->bi_phys_segments = 1; /* biased count of active stripes */
46031f9a
RBJ		 3197 }
3198
3199 return bi;
3200}
3201
3202
f679623f
RBJ		 3203/*
3204 * The "raid5_align_endio" should check if the read succeeded and if it
3205 * did, call bio_endio on the original bio (having bio_put the new bio
3206 * first).
3207 * If the read failed..
3208 */
6712ecf8 3209static void raid5_align_endio(struct bio *bi, int error)
f679623f
RBJ		 3210{
3211 struct bio* raid_bi = bi->bi_private;
46031f9a
RBJ		 3212 mddev_t *mddev;
3213 raid5_conf_t *conf;
3214 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
3215 mdk_rdev_t *rdev;
3216
f679623f 3217 bio_put(bi);
46031f9a
RBJ		 3218
3219 mddev = raid_bi->bi_bdev->bd_disk->queue->queuedata;
3220 conf = mddev_to_conf(mddev);
3221 rdev = (void*)raid_bi->bi_next;
3222 raid_bi->bi_next = NULL;
3223
3224 rdev_dec_pending(rdev, conf->mddev);
3225
3226 if (!error && uptodate) {
6712ecf8 3227 bio_endio(raid_bi, 0);
46031f9a
RBJ		 3228 if (atomic_dec_and_test(&conf->active_aligned_reads))
3229 wake_up(&conf->wait_for_stripe);
6712ecf8 3230 return;
46031f9a
RBJ		 3231 }
3232
3233
45b4233c 3234 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
46031f9a
RBJ		 3235
3236 add_bio_to_retry(raid_bi, conf);
f679623f
RBJ		 3237}
3238
387bb173
NB		 3239static int bio_fits_rdev(struct bio *bi)
3240{
165125e1 3241 struct request_queue *q = bdev_get_queue(bi->bi_bdev);
387bb173
NB		 3242
3243 if ((bi->bi_size>>9) > q->max_sectors)
3244 return 0;
3245 blk_recount_segments(q, bi);
960e739d 3246 if (bi->bi_phys_segments > q->max_phys_segments)
387bb173
NB		 3247 return 0;
3248
3249 if (q->merge_bvec_fn)
3250 /* it's too hard to apply the merge_bvec_fn at this stage,
3251 * just just give up
3252 */
3253 return 0;
3254
3255 return 1;
3256}
3257
3258
165125e1 3259static int chunk_aligned_read(struct request_queue *q, struct bio * raid_bio)
f679623f
RBJ		 3260{
3261 mddev_t *mddev = q->queuedata;
3262 raid5_conf_t *conf = mddev_to_conf(mddev);
3263 const unsigned int raid_disks = conf->raid_disks;
46031f9a 3264 const unsigned int data_disks = raid_disks - conf->max_degraded;
f679623f
RBJ		 3265 unsigned int dd_idx, pd_idx;
3266 struct bio* align_bi;
3267 mdk_rdev_t *rdev;
3268
3269 if (!in_chunk_boundary(mddev, raid_bio)) {
45b4233c 3270 pr_debug("chunk_aligned_read : non aligned\n");
f679623f
RBJ		 3271 return 0;
3272 }
3273 /*
3274 * use bio_clone to make a copy of the bio
3275 */
3276 align_bi = bio_clone(raid_bio, GFP_NOIO);
3277 if (!align_bi)
3278 return 0;
3279 /*
3280 * set bi_end_io to a new function, and set bi_private to the
3281 * original bio.
3282 */
3283 align_bi->bi_end_io = raid5_align_endio;
3284 align_bi->bi_private = raid_bio;
3285 /*
3286 * compute position
3287 */
3288 align_bi->bi_sector = raid5_compute_sector(raid_bio->bi_sector,
3289 raid_disks,
3290 data_disks,
3291 &dd_idx,
3292 &pd_idx,
3293 conf);
3294
3295 rcu_read_lock();
3296 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
3297 if (rdev && test_bit(In_sync, &rdev->flags)) {
f679623f
RBJ		 3298 atomic_inc(&rdev->nr_pending);
3299 rcu_read_unlock();
46031f9a
RBJ		 3300 raid_bio->bi_next = (void*)rdev;
3301 align_bi->bi_bdev = rdev->bdev;
3302 align_bi->bi_flags &= ~(1 << BIO_SEG_VALID);
3303 align_bi->bi_sector += rdev->data_offset;
3304
387bb173
NB		 3305 if (!bio_fits_rdev(align_bi)) {
3306 /* too big in some way */
3307 bio_put(align_bi);
3308 rdev_dec_pending(rdev, mddev);
3309 return 0;
3310 }
3311
46031f9a
RBJ		 3312 spin_lock_irq(&conf->device_lock);
3313 wait_event_lock_irq(conf->wait_for_stripe,
3314 conf->quiesce == 0,
3315 conf->device_lock, /* nothing */);
3316 atomic_inc(&conf->active_aligned_reads);
3317 spin_unlock_irq(&conf->device_lock);
3318
f679623f
RBJ		 3319 generic_make_request(align_bi);
3320 return 1;
3321 } else {
3322 rcu_read_unlock();
46031f9a 3323 bio_put(align_bi);
f679623f
RBJ		 3324 return 0;
3325 }
3326}
3327
8b3e6cdc
DW		 3328/* __get_priority_stripe - get the next stripe to process
3329 *
3330 * Full stripe writes are allowed to pass preread active stripes up until
3331 * the bypass_threshold is exceeded. In general the bypass_count
3332 * increments when the handle_list is handled before the hold_list; however, it
3333 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
3334 * stripe with in flight i/o. The bypass_count will be reset when the
3335 * head of the hold_list has changed, i.e. the head was promoted to the
3336 * handle_list.
3337 */
3338static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf)
3339{
3340 struct stripe_head *sh;
3341
3342 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
3343 __func__,
3344 list_empty(&conf->handle_list) ? "empty" : "busy",
3345 list_empty(&conf->hold_list) ? "empty" : "busy",
3346 atomic_read(&conf->pending_full_writes), conf->bypass_count);
3347
3348 if (!list_empty(&conf->handle_list)) {
3349 sh = list_entry(conf->handle_list.next, typeof(*sh), lru);
3350
3351 if (list_empty(&conf->hold_list))
3352 conf->bypass_count = 0;
3353 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
3354 if (conf->hold_list.next == conf->last_hold)
3355 conf->bypass_count++;
3356 else {
3357 conf->last_hold = conf->hold_list.next;
3358 conf->bypass_count -= conf->bypass_threshold;
3359 if (conf->bypass_count < 0)
3360 conf->bypass_count = 0;
3361 }
3362 }
3363 } else if (!list_empty(&conf->hold_list) &&
3364 ((conf->bypass_threshold &&
3365 conf->bypass_count > conf->bypass_threshold) ||
3366 atomic_read(&conf->pending_full_writes) == 0)) {
3367 sh = list_entry(conf->hold_list.next,
3368 typeof(*sh), lru);
3369 conf->bypass_count -= conf->bypass_threshold;
3370 if (conf->bypass_count < 0)
3371 conf->bypass_count = 0;
3372 } else
3373 return NULL;
3374
3375 list_del_init(&sh->lru);
3376 atomic_inc(&sh->count);
3377 BUG_ON(atomic_read(&sh->count) != 1);
3378 return sh;
3379}
f679623f 3380
165125e1 3381static int make_request(struct request_queue *q, struct bio * bi)
1da177e4
LT		 3382{
3383 mddev_t *mddev = q->queuedata;
3384 raid5_conf_t *conf = mddev_to_conf(mddev);
1da177e4
LT		 3385 unsigned int dd_idx, pd_idx;
3386 sector_t new_sector;
3387 sector_t logical_sector, last_sector;
3388 struct stripe_head *sh;
a362357b 3389 const int rw = bio_data_dir(bi);
f6344757 3390 int remaining;
1da177e4 3391
e5dcdd80 3392 if (unlikely(bio_barrier(bi))) {
6712ecf8 3393 bio_endio(bi, -EOPNOTSUPP);
e5dcdd80
N		 3394 return 0;
3395 }
3396
3d310eb7 3397 md_write_start(mddev, bi);
06d91a5f 3398
a362357b
JA		 3399 disk_stat_inc(mddev->gendisk, ios[rw]);
3400 disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bi));
1da177e4 3401
802ba064 3402 if (rw == READ &&
52488615
RBJ		 3403 mddev->reshape_position == MaxSector &&
3404 chunk_aligned_read(q,bi))
3405 return 0;
3406
1da177e4
LT		 3407 logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
3408 last_sector = bi->bi_sector + (bi->bi_size>>9);
3409 bi->bi_next = NULL;
3410 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
06d91a5f 3411
1da177e4
LT		 3412 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
3413 DEFINE_WAIT(w);
16a53ecc 3414 int disks, data_disks;
b578d55f 3415
7ecaa1e6 3416 retry:
b578d55f 3417 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
7ecaa1e6
N		 3418 if (likely(conf->expand_progress == MaxSector))
3419 disks = conf->raid_disks;
3420 else {
df8e7f76
N		 3421 /* spinlock is needed as expand_progress may be
3422 * 64bit on a 32bit platform, and so it might be
3423 * possible to see a half-updated value
3424 * Ofcourse expand_progress could change after
3425 * the lock is dropped, so once we get a reference
3426 * to the stripe that we think it is, we will have
3427 * to check again.
3428 */
7ecaa1e6
N		 3429 spin_lock_irq(&conf->device_lock);
3430 disks = conf->raid_disks;
3431 if (logical_sector >= conf->expand_progress)
3432 disks = conf->previous_raid_disks;
b578d55f
N		 3433 else {
3434 if (logical_sector >= conf->expand_lo) {
3435 spin_unlock_irq(&conf->device_lock);
3436 schedule();
3437 goto retry;
3438 }
3439 }
7ecaa1e6
N		 3440 spin_unlock_irq(&conf->device_lock);
3441 }
16a53ecc
N		 3442 data_disks = disks - conf->max_degraded;
3443
3444 new_sector = raid5_compute_sector(logical_sector, disks, data_disks,
7ecaa1e6 3445 &dd_idx, &pd_idx, conf);
45b4233c 3446 pr_debug("raid5: make_request, sector %llu logical %llu\n",
1da177e4
LT		 3447 (unsigned long long)new_sector,
3448 (unsigned long long)logical_sector);
3449
7ecaa1e6 3450 sh = get_active_stripe(conf, new_sector, disks, pd_idx, (bi->bi_rw&RWA_MASK));
1da177e4 3451 if (sh) {
7ecaa1e6
N		 3452 if (unlikely(conf->expand_progress != MaxSector)) {
3453 /* expansion might have moved on while waiting for a
df8e7f76
N		 3454 * stripe, so we must do the range check again.
3455 * Expansion could still move past after this
3456 * test, but as we are holding a reference to
3457 * 'sh', we know that if that happens,
3458 * STRIPE_EXPANDING will get set and the expansion
3459 * won't proceed until we finish with the stripe.
7ecaa1e6
N		 3460 */
3461 int must_retry = 0;
3462 spin_lock_irq(&conf->device_lock);
3463 if (logical_sector < conf->expand_progress &&
3464 disks == conf->previous_raid_disks)
3465 /* mismatch, need to try again */
3466 must_retry = 1;
3467 spin_unlock_irq(&conf->device_lock);
3468 if (must_retry) {
3469 release_stripe(sh);
3470 goto retry;
3471 }
3472 }
e464eafd
N		 3473 /* FIXME what if we get a false positive because these
3474 * are being updated.
3475 */
3476 if (logical_sector >= mddev->suspend_lo &&
3477 logical_sector < mddev->suspend_hi) {
3478 release_stripe(sh);
3479 schedule();
3480 goto retry;
3481 }
7ecaa1e6
N		 3482
3483 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
3484 !add_stripe_bio(sh, bi, dd_idx, (bi->bi_rw&RW_MASK))) {
3485 /* Stripe is busy expanding or
3486 * add failed due to overlap. Flush everything
1da177e4
LT		 3487 * and wait a while
3488 */
3489 raid5_unplug_device(mddev->queue);
3490 release_stripe(sh);
3491 schedule();
3492 goto retry;
3493 }
3494 finish_wait(&conf->wait_for_overlap, &w);
6ed3003c
N		 3495 set_bit(STRIPE_HANDLE, &sh->state);
3496 clear_bit(STRIPE_DELAYED, &sh->state);
1da177e4 3497 release_stripe(sh);
1da177e4
LT		 3498 } else {
3499 /* cannot get stripe for read-ahead, just give-up */
3500 clear_bit(BIO_UPTODATE, &bi->bi_flags);
3501 finish_wait(&conf->wait_for_overlap, &w);
3502 break;
3503 }
3504
3505 }
3506 spin_lock_irq(&conf->device_lock);
960e739d 3507 remaining = raid5_dec_bi_phys_segments(bi);
f6344757
N		 3508 spin_unlock_irq(&conf->device_lock);
3509 if (remaining == 0) {
1da177e4 3510
16a53ecc 3511 if ( rw == WRITE )
1da177e4 3512 md_write_end(mddev);
6712ecf8 3513
0e13fe23 3514 bio_endio(bi, 0);
1da177e4 3515 }
1da177e4
LT		 3516 return 0;
3517}
3518
52c03291 3519static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped)
1da177e4 3520{
52c03291
N		 3521 /* reshaping is quite different to recovery/resync so it is
3522 * handled quite separately ... here.
3523 *
3524 * On each call to sync_request, we gather one chunk worth of
3525 * destination stripes and flag them as expanding.
3526 * Then we find all the source stripes and request reads.
3527 * As the reads complete, handle_stripe will copy the data
3528 * into the destination stripe and release that stripe.
3529 */
1da177e4
LT		 3530 raid5_conf_t *conf = (raid5_conf_t *) mddev->private;
3531 struct stripe_head *sh;
ccfcc3c1
N		 3532 int pd_idx;
3533 sector_t first_sector, last_sector;
f416885e
N		 3534 int raid_disks = conf->previous_raid_disks;
3535 int data_disks = raid_disks - conf->max_degraded;
3536 int new_data_disks = conf->raid_disks - conf->max_degraded;
52c03291
N		 3537 int i;
3538 int dd_idx;
3539 sector_t writepos, safepos, gap;
3540
3541 if (sector_nr == 0 &&
3542 conf->expand_progress != 0) {
3543 /* restarting in the middle, skip the initial sectors */
3544 sector_nr = conf->expand_progress;
f416885e 3545 sector_div(sector_nr, new_data_disks);
52c03291
N		 3546 *skipped = 1;
3547 return sector_nr;
3548 }
3549
3550 /* we update the metadata when there is more than 3Meg
3551 * in the block range (that is rather arbitrary, should
3552 * probably be time based) or when the data about to be
3553 * copied would over-write the source of the data at
3554 * the front of the range.
3555 * i.e. one new_stripe forward from expand_progress new_maps
3556 * to after where expand_lo old_maps to
3557 */
3558 writepos = conf->expand_progress +
f416885e
N		 3559 conf->chunk_size/512*(new_data_disks);
3560 sector_div(writepos, new_data_disks);
52c03291 3561 safepos = conf->expand_lo;
f416885e 3562 sector_div(safepos, data_disks);
52c03291
N		 3563 gap = conf->expand_progress - conf->expand_lo;
3564
3565 if (writepos >= safepos ||
f416885e 3566 gap > (new_data_disks)*3000*2 /*3Meg*/) {
52c03291
N		 3567 /* Cannot proceed until we've updated the superblock... */
3568 wait_event(conf->wait_for_overlap,
3569 atomic_read(&conf->reshape_stripes)==0);
3570 mddev->reshape_position = conf->expand_progress;
850b2b42 3571 set_bit(MD_CHANGE_DEVS, &mddev->flags);
52c03291 3572 md_wakeup_thread(mddev->thread);
850b2b42 3573 wait_event(mddev->sb_wait, mddev->flags == 0 ||
52c03291
N		 3574 kthread_should_stop());
3575 spin_lock_irq(&conf->device_lock);
3576 conf->expand_lo = mddev->reshape_position;
3577 spin_unlock_irq(&conf->device_lock);
3578 wake_up(&conf->wait_for_overlap);
3579 }
3580
3581 for (i=0; i < conf->chunk_size/512; i+= STRIPE_SECTORS) {
3582 int j;
3583 int skipped = 0;
3584 pd_idx = stripe_to_pdidx(sector_nr+i, conf, conf->raid_disks);
3585 sh = get_active_stripe(conf, sector_nr+i,
3586 conf->raid_disks, pd_idx, 0);
3587 set_bit(STRIPE_EXPANDING, &sh->state);
3588 atomic_inc(&conf->reshape_stripes);
3589 /* If any of this stripe is beyond the end of the old
3590 * array, then we need to zero those blocks
3591 */
3592 for (j=sh->disks; j--;) {
3593 sector_t s;
3594 if (j == sh->pd_idx)
3595 continue;
f416885e
N		 3596 if (conf->level == 6 &&
3597 j == raid6_next_disk(sh->pd_idx, sh->disks))
3598 continue;
52c03291 3599 s = compute_blocknr(sh, j);
f233ea5c 3600 if (s < mddev->array_sectors) {
52c03291
N		 3601 skipped = 1;
3602 continue;
3603 }
3604 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
3605 set_bit(R5_Expanded, &sh->dev[j].flags);
3606 set_bit(R5_UPTODATE, &sh->dev[j].flags);
3607 }
3608 if (!skipped) {
3609 set_bit(STRIPE_EXPAND_READY, &sh->state);
3610 set_bit(STRIPE_HANDLE, &sh->state);
3611 }
3612 release_stripe(sh);
3613 }
3614 spin_lock_irq(&conf->device_lock);
6d3baf2e 3615 conf->expand_progress = (sector_nr + i) * new_data_disks;
52c03291
N		 3616 spin_unlock_irq(&conf->device_lock);
3617 /* Ok, those stripe are ready. We can start scheduling
3618 * reads on the source stripes.
3619 * The source stripes are determined by mapping the first and last
3620 * block on the destination stripes.
3621 */
52c03291 3622 first_sector =
f416885e 3623 raid5_compute_sector(sector_nr*(new_data_disks),
52c03291
N		 3624 raid_disks, data_disks,
3625 &dd_idx, &pd_idx, conf);
3626 last_sector =
3627 raid5_compute_sector((sector_nr+conf->chunk_size/512)
f416885e 3628 *(new_data_disks) -1,
52c03291
N		 3629 raid_disks, data_disks,
3630 &dd_idx, &pd_idx, conf);
3631 if (last_sector >= (mddev->size<<1))
3632 last_sector = (mddev->size<<1)-1;
3633 while (first_sector <= last_sector) {
f416885e
N		 3634 pd_idx = stripe_to_pdidx(first_sector, conf,
3635 conf->previous_raid_disks);
52c03291
N		 3636 sh = get_active_stripe(conf, first_sector,
3637 conf->previous_raid_disks, pd_idx, 0);
3638 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3639 set_bit(STRIPE_HANDLE, &sh->state);
3640 release_stripe(sh);
3641 first_sector += STRIPE_SECTORS;
3642 }
c6207277
N		 3643 /* If this takes us to the resync_max point where we have to pause,
3644 * then we need to write out the superblock.
3645 */
3646 sector_nr += conf->chunk_size>>9;
3647 if (sector_nr >= mddev->resync_max) {
3648 /* Cannot proceed until we've updated the superblock... */
3649 wait_event(conf->wait_for_overlap,
3650 atomic_read(&conf->reshape_stripes) == 0);
3651 mddev->reshape_position = conf->expand_progress;
3652 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3653 md_wakeup_thread(mddev->thread);
3654 wait_event(mddev->sb_wait,
3655 !test_bit(MD_CHANGE_DEVS, &mddev->flags)
3656 || kthread_should_stop());
3657 spin_lock_irq(&conf->device_lock);
3658 conf->expand_lo = mddev->reshape_position;
3659 spin_unlock_irq(&conf->device_lock);
3660 wake_up(&conf->wait_for_overlap);
3661 }
52c03291
N		 3662 return conf->chunk_size>>9;
3663}
3664
3665/* FIXME go_faster isn't used */
3666static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
3667{
3668 raid5_conf_t *conf = (raid5_conf_t *) mddev->private;
3669 struct stripe_head *sh;
3670 int pd_idx;
1da177e4 3671 int raid_disks = conf->raid_disks;
72626685
N		 3672 sector_t max_sector = mddev->size << 1;
3673 int sync_blocks;
16a53ecc
N		 3674 int still_degraded = 0;
3675 int i;
1da177e4 3676
72626685 3677 if (sector_nr >= max_sector) {
1da177e4
LT		 3678 /* just being told to finish up .. nothing much to do */
3679 unplug_slaves(mddev);
29269553
N		 3680 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
3681 end_reshape(conf);
3682 return 0;
3683 }
72626685
N		 3684
3685 if (mddev->curr_resync < max_sector) /* aborted */
3686 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
3687 &sync_blocks, 1);
16a53ecc 3688 else /* completed sync */
72626685
N		 3689 conf->fullsync = 0;
3690 bitmap_close_sync(mddev->bitmap);
3691
1da177e4
LT		 3692 return 0;
3693 }
ccfcc3c1 3694
52c03291
N		 3695 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3696 return reshape_request(mddev, sector_nr, skipped);
f6705578 3697
c6207277
N		 3698 /* No need to check resync_max as we never do more than one
3699 * stripe, and as resync_max will always be on a chunk boundary,
3700 * if the check in md_do_sync didn't fire, there is no chance
3701 * of overstepping resync_max here
3702 */
3703
16a53ecc 3704 /* if there is too many failed drives and we are trying
1da177e4
LT		 3705 * to resync, then assert that we are finished, because there is
3706 * nothing we can do.
3707 */
3285edf1 3708 if (mddev->degraded >= conf->max_degraded &&
16a53ecc 3709 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
57afd89f
N		 3710 sector_t rv = (mddev->size << 1) - sector_nr;
3711 *skipped = 1;
1da177e4
LT		 3712 return rv;
3713 }
72626685 3714 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
3855ad9f 3715 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
72626685
N		 3716 !conf->fullsync && sync_blocks >= STRIPE_SECTORS) {
3717 /* we can skip this block, and probably more */
3718 sync_blocks /= STRIPE_SECTORS;
3719 *skipped = 1;
3720 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
3721 }
1da177e4 3722
b47490c9
N		 3723
3724 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
3725
ccfcc3c1 3726 pd_idx = stripe_to_pdidx(sector_nr, conf, raid_disks);
7ecaa1e6 3727 sh = get_active_stripe(conf, sector_nr, raid_disks, pd_idx, 1);
1da177e4 3728 if (sh == NULL) {
7ecaa1e6 3729 sh = get_active_stripe(conf, sector_nr, raid_disks, pd_idx, 0);
1da177e4 3730 /* make sure we don't swamp the stripe cache if someone else
16a53ecc 3731 * is trying to get access
1da177e4 3732 */
66c006a5 3733 schedule_timeout_uninterruptible(1);
1da177e4 3734 }
16a53ecc
N		 3735 /* Need to check if array will still be degraded after recovery/resync
3736 * We don't need to check the 'failed' flag as when that gets set,
3737 * recovery aborts.
3738 */
3739 for (i=0; i<mddev->raid_disks; i++)
3740 if (conf->disks[i].rdev == NULL)
3741 still_degraded = 1;
3742
3743 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
3744
3745 spin_lock(&sh->lock);
1da177e4
LT		 3746 set_bit(STRIPE_SYNCING, &sh->state);
3747 clear_bit(STRIPE_INSYNC, &sh->state);
3748 spin_unlock(&sh->lock);
3749
df10cfbc
DW		 3750 /* wait for any blocked device to be handled */
3751 while(unlikely(!handle_stripe(sh, NULL)))
3752 ;
1da177e4
LT		 3753 release_stripe(sh);
3754
3755 return STRIPE_SECTORS;
3756}
3757
46031f9a
RBJ		 3758static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio)
3759{
3760 /* We may not be able to submit a whole bio at once as there
3761 * may not be enough stripe_heads available.
3762 * We cannot pre-allocate enough stripe_heads as we may need
3763 * more than exist in the cache (if we allow ever large chunks).
3764 * So we do one stripe head at a time and record in
3765 * ->bi_hw_segments how many have been done.
3766 *
3767 * We *know* that this entire raid_bio is in one chunk, so
3768 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
3769 */
3770 struct stripe_head *sh;
3771 int dd_idx, pd_idx;
3772 sector_t sector, logical_sector, last_sector;
3773 int scnt = 0;
3774 int remaining;
3775 int handled = 0;
3776
3777 logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
3778 sector = raid5_compute_sector( logical_sector,
3779 conf->raid_disks,
3780 conf->raid_disks - conf->max_degraded,
3781 &dd_idx,
3782 &pd_idx,
3783 conf);
3784 last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9);
3785
3786 for (; logical_sector < last_sector;
387bb173
NB		 3787 logical_sector += STRIPE_SECTORS,
3788 sector += STRIPE_SECTORS,
3789 scnt++) {
46031f9a 3790
960e739d 3791 if (scnt < raid5_bi_hw_segments(raid_bio))
46031f9a
RBJ		 3792 /* already done this stripe */
3793 continue;
3794
3795 sh = get_active_stripe(conf, sector, conf->raid_disks, pd_idx, 1);
3796
3797 if (!sh) {
3798 /* failed to get a stripe - must wait */
960e739d 3799 raid5_set_bi_hw_segments(raid_bio, scnt);
46031f9a
RBJ		 3800 conf->retry_read_aligned = raid_bio;
3801 return handled;
3802 }
3803
3804 set_bit(R5_ReadError, &sh->dev[dd_idx].flags);
387bb173
NB		 3805 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) {
3806 release_stripe(sh);
960e739d 3807 raid5_set_bi_hw_segments(raid_bio, scnt);
387bb173
NB		 3808 conf->retry_read_aligned = raid_bio;
3809 return handled;
3810 }
3811
46031f9a
RBJ		 3812 handle_stripe(sh, NULL);
3813 release_stripe(sh);
3814 handled++;
3815 }
3816 spin_lock_irq(&conf->device_lock);
960e739d 3817 remaining = raid5_dec_bi_phys_segments(raid_bio);
46031f9a 3818 spin_unlock_irq(&conf->device_lock);
0e13fe23
NB		 3819 if (remaining == 0)
3820 bio_endio(raid_bio, 0);
46031f9a
RBJ		 3821 if (atomic_dec_and_test(&conf->active_aligned_reads))
3822 wake_up(&conf->wait_for_stripe);
3823 return handled;
3824}
3825
3826
3827
1da177e4
LT		 3828/*
3829 * This is our raid5 kernel thread.
3830 *
3831 * We scan the hash table for stripes which can be handled now.
3832 * During the scan, completed stripes are saved for us by the interrupt
3833 * handler, so that they will not have to wait for our next wakeup.
3834 */
6ed3003c 3835static void raid5d(mddev_t *mddev)
1da177e4
LT		 3836{
3837 struct stripe_head *sh;
3838 raid5_conf_t *conf = mddev_to_conf(mddev);
3839 int handled;
3840
45b4233c 3841 pr_debug("+++ raid5d active\n");
1da177e4
LT		 3842
3843 md_check_recovery(mddev);
1da177e4
LT		 3844
3845 handled = 0;
3846 spin_lock_irq(&conf->device_lock);
3847 while (1) {
46031f9a 3848 struct bio *bio;
1da177e4 3849
ae3c20cc 3850 if (conf->seq_flush != conf->seq_write) {
72626685 3851 int seq = conf->seq_flush;
700e432d 3852 spin_unlock_irq(&conf->device_lock);
72626685 3853 bitmap_unplug(mddev->bitmap);
700e432d 3854 spin_lock_irq(&conf->device_lock);
72626685
N		 3855 conf->seq_write = seq;
3856 activate_bit_delay(conf);
3857 }
3858
46031f9a
RBJ		 3859 while ((bio = remove_bio_from_retry(conf))) {
3860 int ok;
3861 spin_unlock_irq(&conf->device_lock);
3862 ok = retry_aligned_read(conf, bio);
3863 spin_lock_irq(&conf->device_lock);
3864 if (!ok)
3865 break;
3866 handled++;
3867 }
3868
8b3e6cdc
DW		 3869 sh = __get_priority_stripe(conf);
3870
c9f21aaf 3871 if (!sh)
1da177e4 3872 break;
1da177e4
LT		 3873 spin_unlock_irq(&conf->device_lock);
3874
3875 handled++;
16a53ecc 3876 handle_stripe(sh, conf->spare_page);
1da177e4
LT		 3877 release_stripe(sh);
3878
3879 spin_lock_irq(&conf->device_lock);
3880 }
45b4233c 3881 pr_debug("%d stripes handled\n", handled);
1da177e4
LT		 3882
3883 spin_unlock_irq(&conf->device_lock);
3884
c9f21aaf 3885 async_tx_issue_pending_all();
1da177e4
LT		 3886 unplug_slaves(mddev);
3887
45b4233c 3888 pr_debug("--- raid5d inactive\n");
1da177e4
LT		 3889}
3890
3f294f4f 3891static ssize_t
007583c9 3892raid5_show_stripe_cache_size(mddev_t *mddev, char *page)
3f294f4f 3893{
007583c9 3894 raid5_conf_t *conf = mddev_to_conf(mddev);
96de1e66
N		 3895 if (conf)
3896 return sprintf(page, "%d\n", conf->max_nr_stripes);
3897 else
3898 return 0;
3f294f4f
N		 3899}
3900
3901static ssize_t
007583c9 3902raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len)
3f294f4f 3903{
007583c9 3904 raid5_conf_t *conf = mddev_to_conf(mddev);
4ef197d8 3905 unsigned long new;
b5470dc5
DW		 3906 int err;
3907
3f294f4f
N		 3908 if (len >= PAGE_SIZE)
3909 return -EINVAL;
96de1e66
N		 3910 if (!conf)
3911 return -ENODEV;
3f294f4f 3912
4ef197d8 3913 if (strict_strtoul(page, 10, &new))
3f294f4f
N		 3914 return -EINVAL;
3915 if (new <= 16 || new > 32768)
3916 return -EINVAL;
3917 while (new < conf->max_nr_stripes) {
3918 if (drop_one_stripe(conf))
3919 conf->max_nr_stripes--;
3920 else
3921 break;
3922 }
b5470dc5
DW		 3923 err = md_allow_write(mddev);
3924 if (err)
3925 return err;
3f294f4f
N		 3926 while (new > conf->max_nr_stripes) {
3927 if (grow_one_stripe(conf))
3928 conf->max_nr_stripes++;
3929 else break;
3930 }
3931 return len;
3932}
007583c9 3933
96de1e66
N		 3934static struct md_sysfs_entry
3935raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
3936 raid5_show_stripe_cache_size,
3937 raid5_store_stripe_cache_size);
3f294f4f 3938
8b3e6cdc
DW		 3939static ssize_t
3940raid5_show_preread_threshold(mddev_t *mddev, char *page)
3941{
3942 raid5_conf_t *conf = mddev_to_conf(mddev);
3943 if (conf)
3944 return sprintf(page, "%d\n", conf->bypass_threshold);
3945 else
3946 return 0;
3947}
3948
3949static ssize_t
3950raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len)
3951{
3952 raid5_conf_t *conf = mddev_to_conf(mddev);
4ef197d8 3953 unsigned long new;
8b3e6cdc
DW		 3954 if (len >= PAGE_SIZE)
3955 return -EINVAL;
3956 if (!conf)
3957 return -ENODEV;
3958
4ef197d8 3959 if (strict_strtoul(page, 10, &new))
8b3e6cdc 3960 return -EINVAL;
4ef197d8 3961 if (new > conf->max_nr_stripes)
8b3e6cdc
DW		 3962 return -EINVAL;
3963 conf->bypass_threshold = new;
3964 return len;
3965}
3966
3967static struct md_sysfs_entry
3968raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
3969 S_IRUGO | S_IWUSR,
3970 raid5_show_preread_threshold,
3971 raid5_store_preread_threshold);
3972
3f294f4f 3973static ssize_t
96de1e66 3974stripe_cache_active_show(mddev_t *mddev, char *page)
3f294f4f 3975{
007583c9 3976 raid5_conf_t *conf = mddev_to_conf(mddev);
96de1e66
N		 3977 if (conf)
3978 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
3979 else
3980 return 0;
3f294f4f
N		 3981}
3982
96de1e66
N		 3983static struct md_sysfs_entry
3984raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
3f294f4f 3985
007583c9 3986static struct attribute *raid5_attrs[] = {
3f294f4f
N		 3987 &raid5_stripecache_size.attr,
3988 &raid5_stripecache_active.attr,
8b3e6cdc 3989 &raid5_preread_bypass_threshold.attr,
3f294f4f
N		 3990 NULL,
3991};
007583c9
N		 3992static struct attribute_group raid5_attrs_group = {
3993 .name = NULL,
3994 .attrs = raid5_attrs,
3f294f4f
N		 3995};
3996
72626685 3997static int run(mddev_t *mddev)
1da177e4
LT		 3998{
3999 raid5_conf_t *conf;
4000 int raid_disk, memory;
4001 mdk_rdev_t *rdev;
4002 struct disk_info *disk;
4003 struct list_head *tmp;
02c2de8c 4004 int working_disks = 0;
1da177e4 4005
16a53ecc
N		 4006 if (mddev->level != 5 && mddev->level != 4 && mddev->level != 6) {
4007 printk(KERN_ERR "raid5: %s: raid level not set to 4/5/6 (%d)\n",
14f8d26b 4008 mdname(mddev), mddev->level);
1da177e4
LT		 4009 return -EIO;
4010 }
4011
f6705578
N		 4012 if (mddev->reshape_position != MaxSector) {
4013 /* Check that we can continue the reshape.
4014 * Currently only disks can change, it must
4015 * increase, and we must be past the point where
4016 * a stripe over-writes itself
4017 */
4018 sector_t here_new, here_old;
4019 int old_disks;
f416885e 4020 int max_degraded = (mddev->level == 5 ? 1 : 2);
f6705578
N		 4021
4022 if (mddev->new_level != mddev->level ||
4023 mddev->new_layout != mddev->layout ||
4024 mddev->new_chunk != mddev->chunk_size) {
f416885e
N		 4025 printk(KERN_ERR "raid5: %s: unsupported reshape "
4026 "required - aborting.\n",
f6705578
N		 4027 mdname(mddev));
4028 return -EINVAL;
4029 }
4030 if (mddev->delta_disks <= 0) {
f416885e
N		 4031 printk(KERN_ERR "raid5: %s: unsupported reshape "
4032 "(reduce disks) required - aborting.\n",
f6705578
N		 4033 mdname(mddev));
4034 return -EINVAL;
4035 }
4036 old_disks = mddev->raid_disks - mddev->delta_disks;
4037 /* reshape_position must be on a new-stripe boundary, and one
f416885e
N		 4038 * further up in new geometry must map after here in old
4039 * geometry.
f6705578
N		 4040 */
4041 here_new = mddev->reshape_position;
f416885e
N		 4042 if (sector_div(here_new, (mddev->chunk_size>>9)*
4043 (mddev->raid_disks - max_degraded))) {
4044 printk(KERN_ERR "raid5: reshape_position not "
4045 "on a stripe boundary\n");
f6705578
N		 4046 return -EINVAL;
4047 }
4048 /* here_new is the stripe we will write to */
4049 here_old = mddev->reshape_position;
f416885e
N		 4050 sector_div(here_old, (mddev->chunk_size>>9)*
4051 (old_disks-max_degraded));
4052 /* here_old is the first stripe that we might need to read
4053 * from */
f6705578
N		 4054 if (here_new >= here_old) {
4055 /* Reading from the same stripe as writing to - bad */
f416885e
N		 4056 printk(KERN_ERR "raid5: reshape_position too early for "
4057 "auto-recovery - aborting.\n");
f6705578
N		 4058 return -EINVAL;
4059 }
4060 printk(KERN_INFO "raid5: reshape will continue\n");
4061 /* OK, we should be able to continue; */
4062 }
4063
4064
b55e6bfc 4065 mddev->private = kzalloc(sizeof (raid5_conf_t), GFP_KERNEL);
1da177e4
LT		 4066 if ((conf = mddev->private) == NULL)
4067 goto abort;
f6705578
N		 4068 if (mddev->reshape_position == MaxSector) {
4069 conf->previous_raid_disks = conf->raid_disks = mddev->raid_disks;
4070 } else {
4071 conf->raid_disks = mddev->raid_disks;
4072 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
4073 }
4074
4075 conf->disks = kzalloc(conf->raid_disks * sizeof(struct disk_info),
b55e6bfc
N		 4076 GFP_KERNEL);
4077 if (!conf->disks)
4078 goto abort;
9ffae0cf 4079
1da177e4
LT		 4080 conf->mddev = mddev;
4081
fccddba0 4082 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
1da177e4 4083 goto abort;
1da177e4 4084
16a53ecc
N		 4085 if (mddev->level == 6) {
4086 conf->spare_page = alloc_page(GFP_KERNEL);
4087 if (!conf->spare_page)
4088 goto abort;
4089 }
1da177e4 4090 spin_lock_init(&conf->device_lock);
e7e72bf6 4091 mddev->queue->queue_lock = &conf->device_lock;
1da177e4
LT		 4092 init_waitqueue_head(&conf->wait_for_stripe);
4093 init_waitqueue_head(&conf->wait_for_overlap);
4094 INIT_LIST_HEAD(&conf->handle_list);
8b3e6cdc 4095 INIT_LIST_HEAD(&conf->hold_list);
1da177e4 4096 INIT_LIST_HEAD(&conf->delayed_list);
72626685 4097 INIT_LIST_HEAD(&conf->bitmap_list);
1da177e4
LT		 4098 INIT_LIST_HEAD(&conf->inactive_list);
4099 atomic_set(&conf->active_stripes, 0);
4100 atomic_set(&conf->preread_active_stripes, 0);
46031f9a 4101 atomic_set(&conf->active_aligned_reads, 0);
8b3e6cdc 4102 conf->bypass_threshold = BYPASS_THRESHOLD;
1da177e4 4103
45b4233c 4104 pr_debug("raid5: run(%s) called.\n", mdname(mddev));
1da177e4 4105
d089c6af 4106 rdev_for_each(rdev, tmp, mddev) {
1da177e4 4107 raid_disk = rdev->raid_disk;
f6705578 4108 if (raid_disk >= conf->raid_disks
1da177e4
LT		 4109 || raid_disk < 0)
4110 continue;
4111 disk = conf->disks + raid_disk;
4112
4113 disk->rdev = rdev;
4114
b2d444d7 4115 if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 4116 char b[BDEVNAME_SIZE];
4117 printk(KERN_INFO "raid5: device %s operational as raid"
4118 " disk %d\n", bdevname(rdev->bdev,b),
4119 raid_disk);
02c2de8c 4120 working_disks++;
8c2e870a
NB		 4121 } else
4122 /* Cannot rely on bitmap to complete recovery */
4123 conf->fullsync = 1;
1da177e4
LT		 4124 }
4125
1da177e4 4126 /*
16a53ecc 4127 * 0 for a fully functional array, 1 or 2 for a degraded array.
1da177e4 4128 */
02c2de8c 4129 mddev->degraded = conf->raid_disks - working_disks;
1da177e4
LT		 4130 conf->mddev = mddev;
4131 conf->chunk_size = mddev->chunk_size;
4132 conf->level = mddev->level;
16a53ecc
N		 4133 if (conf->level == 6)
4134 conf->max_degraded = 2;
4135 else
4136 conf->max_degraded = 1;
1da177e4
LT		 4137 conf->algorithm = mddev->layout;
4138 conf->max_nr_stripes = NR_STRIPES;
f6705578 4139 conf->expand_progress = mddev->reshape_position;
1da177e4
LT		 4140
4141 /* device size must be a multiple of chunk size */
4142 mddev->size &= ~(mddev->chunk_size/1024 -1);
b1581566 4143 mddev->resync_max_sectors = mddev->size << 1;
1da177e4 4144
16a53ecc
N		 4145 if (conf->level == 6 && conf->raid_disks < 4) {
4146 printk(KERN_ERR "raid6: not enough configured devices for %s (%d, minimum 4)\n",
4147 mdname(mddev), conf->raid_disks);
4148 goto abort;
4149 }
1da177e4
LT		 4150 if (!conf->chunk_size || conf->chunk_size % 4) {
4151 printk(KERN_ERR "raid5: invalid chunk size %d for %s\n",
4152 conf->chunk_size, mdname(mddev));
4153 goto abort;
4154 }
4155 if (conf->algorithm > ALGORITHM_RIGHT_SYMMETRIC) {
4156 printk(KERN_ERR
4157 "raid5: unsupported parity algorithm %d for %s\n",
4158 conf->algorithm, mdname(mddev));
4159 goto abort;
4160 }
16a53ecc 4161 if (mddev->degraded > conf->max_degraded) {
1da177e4
LT		 4162 printk(KERN_ERR "raid5: not enough operational devices for %s"
4163 " (%d/%d failed)\n",
02c2de8c 4164 mdname(mddev), mddev->degraded, conf->raid_disks);
1da177e4
LT		 4165 goto abort;
4166 }
4167
16a53ecc 4168 if (mddev->degraded > 0 &&
1da177e4 4169 mddev->recovery_cp != MaxSector) {
6ff8d8ec
N		 4170 if (mddev->ok_start_degraded)
4171 printk(KERN_WARNING
4172 "raid5: starting dirty degraded array: %s"
4173 "- data corruption possible.\n",
4174 mdname(mddev));
4175 else {
4176 printk(KERN_ERR
4177 "raid5: cannot start dirty degraded array for %s\n",
4178 mdname(mddev));
4179 goto abort;
4180 }
1da177e4
LT		 4181 }
4182
4183 {
4184 mddev->thread = md_register_thread(raid5d, mddev, "%s_raid5");
4185 if (!mddev->thread) {
4186 printk(KERN_ERR
4187 "raid5: couldn't allocate thread for %s\n",
4188 mdname(mddev));
4189 goto abort;
4190 }
4191 }
5036805b 4192 memory = conf->max_nr_stripes * (sizeof(struct stripe_head) +
1da177e4
LT		 4193 conf->raid_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
4194 if (grow_stripes(conf, conf->max_nr_stripes)) {
4195 printk(KERN_ERR
4196 "raid5: couldn't allocate %dkB for buffers\n", memory);
4197 shrink_stripes(conf);
4198 md_unregister_thread(mddev->thread);
4199 goto abort;
4200 } else
4201 printk(KERN_INFO "raid5: allocated %dkB for %s\n",
4202 memory, mdname(mddev));
4203
4204 if (mddev->degraded == 0)
4205 printk("raid5: raid level %d set %s active with %d out of %d"
4206 " devices, algorithm %d\n", conf->level, mdname(mddev),
4207 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
4208 conf->algorithm);
4209 else
4210 printk(KERN_ALERT "raid5: raid level %d set %s active with %d"
4211 " out of %d devices, algorithm %d\n", conf->level,
4212 mdname(mddev), mddev->raid_disks - mddev->degraded,
4213 mddev->raid_disks, conf->algorithm);
4214
4215 print_raid5_conf(conf);
4216
f6705578
N		 4217 if (conf->expand_progress != MaxSector) {
4218 printk("...ok start reshape thread\n");
b578d55f 4219 conf->expand_lo = conf->expand_progress;
f6705578
N		 4220 atomic_set(&conf->reshape_stripes, 0);
4221 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4222 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4223 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
4224 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4225 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
4226 "%s_reshape");
f6705578
N		 4227 }
4228
1da177e4 4229 /* read-ahead size must cover two whole stripes, which is
16a53ecc 4230 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
1da177e4
LT		 4231 */
4232 {
16a53ecc
N		 4233 int data_disks = conf->previous_raid_disks - conf->max_degraded;
4234 int stripe = data_disks *
8932c2e0 4235 (mddev->chunk_size / PAGE_SIZE);
1da177e4
LT		 4236 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
4237 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
4238 }
4239
4240 /* Ok, everything is just fine now */
5e55e2f5
N		 4241 if (sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
4242 printk(KERN_WARNING
4243 "raid5: failed to create sysfs attributes for %s\n",
4244 mdname(mddev));
7a5febe9
N		 4245
4246 mddev->queue->unplug_fn = raid5_unplug_device;
f022b2fd 4247 mddev->queue->backing_dev_info.congested_data = mddev;
041ae52e 4248 mddev->queue->backing_dev_info.congested_fn = raid5_congested;
f022b2fd 4249
f233ea5c 4250 mddev->array_sectors = 2 * mddev->size * (conf->previous_raid_disks -
16a53ecc 4251 conf->max_degraded);
7a5febe9 4252
23032a0e
RBJ		 4253 blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec);
4254
1da177e4
LT		 4255 return 0;
4256abort:
4257 if (conf) {
4258 print_raid5_conf(conf);
16a53ecc 4259 safe_put_page(conf->spare_page);
b55e6bfc 4260 kfree(conf->disks);
fccddba0 4261 kfree(conf->stripe_hashtbl);
1da177e4
LT		 4262 kfree(conf);
4263 }
4264 mddev->private = NULL;
4265 printk(KERN_ALERT "raid5: failed to run raid set %s\n", mdname(mddev));
4266 return -EIO;
4267}
4268
4269
4270
3f294f4f 4271static int stop(mddev_t *mddev)
1da177e4
LT		 4272{
4273 raid5_conf_t *conf = (raid5_conf_t *) mddev->private;
4274
4275 md_unregister_thread(mddev->thread);
4276 mddev->thread = NULL;
4277 shrink_stripes(conf);
fccddba0 4278 kfree(conf->stripe_hashtbl);
041ae52e 4279 mddev->queue->backing_dev_info.congested_fn = NULL;
1da177e4 4280 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
007583c9 4281 sysfs_remove_group(&mddev->kobj, &raid5_attrs_group);
b55e6bfc 4282 kfree(conf->disks);
96de1e66 4283 kfree(conf);
1da177e4
LT		 4284 mddev->private = NULL;
4285 return 0;
4286}
4287
45b4233c 4288#ifdef DEBUG
16a53ecc 4289static void print_sh (struct seq_file *seq, struct stripe_head *sh)
1da177e4
LT		 4290{
4291 int i;
4292
16a53ecc
N		 4293 seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n",
4294 (unsigned long long)sh->sector, sh->pd_idx, sh->state);
4295 seq_printf(seq, "sh %llu, count %d.\n",
4296 (unsigned long long)sh->sector, atomic_read(&sh->count));
4297 seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector);
7ecaa1e6 4298 for (i = 0; i < sh->disks; i++) {
16a53ecc
N		 4299 seq_printf(seq, "(cache%d: %p %ld) ",
4300 i, sh->dev[i].page, sh->dev[i].flags);
1da177e4 4301 }
16a53ecc 4302 seq_printf(seq, "\n");
1da177e4
LT		 4303}
4304
16a53ecc 4305static void printall (struct seq_file *seq, raid5_conf_t *conf)
1da177e4
LT		 4306{
4307 struct stripe_head *sh;
fccddba0 4308 struct hlist_node *hn;
1da177e4
LT		 4309 int i;
4310
4311 spin_lock_irq(&conf->device_lock);
4312 for (i = 0; i < NR_HASH; i++) {
fccddba0 4313 hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) {
1da177e4
LT		 4314 if (sh->raid_conf != conf)
4315 continue;
16a53ecc 4316 print_sh(seq, sh);
1da177e4
LT		 4317 }
4318 }
4319 spin_unlock_irq(&conf->device_lock);
4320}
4321#endif
4322
4323static void status (struct seq_file *seq, mddev_t *mddev)
4324{
4325 raid5_conf_t *conf = (raid5_conf_t *) mddev->private;
4326 int i;
4327
4328 seq_printf (seq, " level %d, %dk chunk, algorithm %d", mddev->level, mddev->chunk_size >> 10, mddev->layout);
02c2de8c 4329 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
1da177e4
LT		 4330 for (i = 0; i < conf->raid_disks; i++)
4331 seq_printf (seq, "%s",
4332 conf->disks[i].rdev &&
b2d444d7 4333 test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_");
1da177e4 4334 seq_printf (seq, "]");
45b4233c 4335#ifdef DEBUG
16a53ecc
N		 4336 seq_printf (seq, "\n");
4337 printall(seq, conf);
1da177e4
LT		 4338#endif
4339}
4340
4341static void print_raid5_conf (raid5_conf_t *conf)
4342{
4343 int i;
4344 struct disk_info *tmp;
4345
4346 printk("RAID5 conf printout:\n");
4347 if (!conf) {
4348 printk("(conf==NULL)\n");
4349 return;
4350 }
02c2de8c
N		 4351 printk(" --- rd:%d wd:%d\n", conf->raid_disks,
4352 conf->raid_disks - conf->mddev->degraded);
1da177e4
LT		 4353
4354 for (i = 0; i < conf->raid_disks; i++) {
4355 char b[BDEVNAME_SIZE];
4356 tmp = conf->disks + i;
4357 if (tmp->rdev)
4358 printk(" disk %d, o:%d, dev:%s\n",
b2d444d7 4359 i, !test_bit(Faulty, &tmp->rdev->flags),
1da177e4
LT		 4360 bdevname(tmp->rdev->bdev,b));
4361 }
4362}
4363
4364static int raid5_spare_active(mddev_t *mddev)
4365{
4366 int i;
4367 raid5_conf_t *conf = mddev->private;
4368 struct disk_info *tmp;
4369
4370 for (i = 0; i < conf->raid_disks; i++) {
4371 tmp = conf->disks + i;
4372 if (tmp->rdev
b2d444d7 4373 && !test_bit(Faulty, &tmp->rdev->flags)
c04be0aa
N		 4374 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
4375 unsigned long flags;
4376 spin_lock_irqsave(&conf->device_lock, flags);
1da177e4 4377 mddev->degraded--;
c04be0aa 4378 spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4
LT		 4379 }
4380 }
4381 print_raid5_conf(conf);
4382 return 0;
4383}
4384
4385static int raid5_remove_disk(mddev_t *mddev, int number)
4386{
4387 raid5_conf_t *conf = mddev->private;
4388 int err = 0;
4389 mdk_rdev_t *rdev;
4390 struct disk_info *p = conf->disks + number;
4391
4392 print_raid5_conf(conf);
4393 rdev = p->rdev;
4394 if (rdev) {
b2d444d7 4395 if (test_bit(In_sync, &rdev->flags) ||
1da177e4
LT		 4396 atomic_read(&rdev->nr_pending)) {
4397 err = -EBUSY;
4398 goto abort;
4399 }
dfc70645
N		 4400 /* Only remove non-faulty devices if recovery
4401 * isn't possible.
4402 */
4403 if (!test_bit(Faulty, &rdev->flags) &&
4404 mddev->degraded <= conf->max_degraded) {
4405 err = -EBUSY;
4406 goto abort;
4407 }
1da177e4 4408 p->rdev = NULL;
fbd568a3 4409 synchronize_rcu();
1da177e4
LT		 4410 if (atomic_read(&rdev->nr_pending)) {
4411 /* lost the race, try later */
4412 err = -EBUSY;
4413 p->rdev = rdev;
4414 }
4415 }
4416abort:
4417
4418 print_raid5_conf(conf);
4419 return err;
4420}
4421
4422static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
4423{
4424 raid5_conf_t *conf = mddev->private;
199050ea 4425 int err = -EEXIST;
1da177e4
LT		 4426 int disk;
4427 struct disk_info *p;
6c2fce2e
NB		 4428 int first = 0;
4429 int last = conf->raid_disks - 1;
1da177e4 4430
16a53ecc 4431 if (mddev->degraded > conf->max_degraded)
1da177e4 4432 /* no point adding a device */
199050ea 4433 return -EINVAL;
1da177e4 4434
6c2fce2e
NB		 4435 if (rdev->raid_disk >= 0)
4436 first = last = rdev->raid_disk;
1da177e4
LT		 4437
4438 /*
16a53ecc
N		 4439 * find the disk ... but prefer rdev->saved_raid_disk
4440 * if possible.
1da177e4 4441 */
16a53ecc 4442 if (rdev->saved_raid_disk >= 0 &&
6c2fce2e 4443 rdev->saved_raid_disk >= first &&
16a53ecc
N		 4444 conf->disks[rdev->saved_raid_disk].rdev == NULL)
4445 disk = rdev->saved_raid_disk;
4446 else
6c2fce2e
NB		 4447 disk = first;
4448 for ( ; disk <= last ; disk++)
1da177e4 4449 if ((p=conf->disks + disk)->rdev == NULL) {
b2d444d7 4450 clear_bit(In_sync, &rdev->flags);
1da177e4 4451 rdev->raid_disk = disk;
199050ea 4452 err = 0;
72626685
N		 4453 if (rdev->saved_raid_disk != disk)
4454 conf->fullsync = 1;
d6065f7b 4455 rcu_assign_pointer(p->rdev, rdev);
1da177e4
LT		 4456 break;
4457 }
4458 print_raid5_conf(conf);
199050ea 4459 return err;
1da177e4
LT		 4460}
4461
4462static int raid5_resize(mddev_t *mddev, sector_t sectors)
4463{
4464 /* no resync is happening, and there is enough space
4465 * on all devices, so we can resize.
4466 * We need to make sure resync covers any new space.
4467 * If the array is shrinking we should possibly wait until
4468 * any io in the removed space completes, but it hardly seems
4469 * worth it.
4470 */
16a53ecc
N		 4471 raid5_conf_t *conf = mddev_to_conf(mddev);
4472
1da177e4 4473 sectors &= ~((sector_t)mddev->chunk_size/512 - 1);
f233ea5c
AN		 4474 mddev->array_sectors = sectors * (mddev->raid_disks
4475 - conf->max_degraded);
4476 set_capacity(mddev->gendisk, mddev->array_sectors);
44ce6294 4477 mddev->changed = 1;
1da177e4
LT		 4478 if (sectors/2 > mddev->size && mddev->recovery_cp == MaxSector) {
4479 mddev->recovery_cp = mddev->size << 1;
4480 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4481 }
4482 mddev->size = sectors /2;
4b5c7ae8 4483 mddev->resync_max_sectors = sectors;
1da177e4
LT		 4484 return 0;
4485}
4486
29269553 4487#ifdef CONFIG_MD_RAID5_RESHAPE
63c70c4f 4488static int raid5_check_reshape(mddev_t *mddev)
29269553
N		 4489{
4490 raid5_conf_t *conf = mddev_to_conf(mddev);
4491 int err;
29269553 4492
63c70c4f
N		 4493 if (mddev->delta_disks < 0 ||
4494 mddev->new_level != mddev->level)
4495 return -EINVAL; /* Cannot shrink array or change level yet */
4496 if (mddev->delta_disks == 0)
29269553 4497 return 0; /* nothing to do */
dba034ee
N		 4498 if (mddev->bitmap)
4499 /* Cannot grow a bitmap yet */
4500 return -EBUSY;
29269553
N		 4501
4502 /* Can only proceed if there are plenty of stripe_heads.
4503 * We need a minimum of one full stripe,, and for sensible progress
4504 * it is best to have about 4 times that.
4505 * If we require 4 times, then the default 256 4K stripe_heads will
4506 * allow for chunk sizes up to 256K, which is probably OK.
4507 * If the chunk size is greater, user-space should request more
4508 * stripe_heads first.
4509 */
63c70c4f
N		 4510 if ((mddev->chunk_size / STRIPE_SIZE) * 4 > conf->max_nr_stripes ||
4511 (mddev->new_chunk / STRIPE_SIZE) * 4 > conf->max_nr_stripes) {
29269553
N		 4512 printk(KERN_WARNING "raid5: reshape: not enough stripes. Needed %lu\n",
4513 (mddev->chunk_size / STRIPE_SIZE)*4);
4514 return -ENOSPC;
4515 }
4516
63c70c4f
N		 4517 err = resize_stripes(conf, conf->raid_disks + mddev->delta_disks);
4518 if (err)
4519 return err;
4520
b4c4c7b8
N		 4521 if (mddev->degraded > conf->max_degraded)
4522 return -EINVAL;
63c70c4f
N		 4523 /* looks like we might be able to manage this */
4524 return 0;
4525}
4526
4527static int raid5_start_reshape(mddev_t *mddev)
4528{
4529 raid5_conf_t *conf = mddev_to_conf(mddev);
4530 mdk_rdev_t *rdev;
4531 struct list_head *rtmp;
4532 int spares = 0;
4533 int added_devices = 0;
c04be0aa 4534 unsigned long flags;
63c70c4f 4535
f416885e 4536 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
63c70c4f
N		 4537 return -EBUSY;
4538
d089c6af 4539 rdev_for_each(rdev, rtmp, mddev)
29269553
N		 4540 if (rdev->raid_disk < 0 &&
4541 !test_bit(Faulty, &rdev->flags))
4542 spares++;
63c70c4f 4543
f416885e 4544 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
29269553
N		 4545 /* Not enough devices even to make a degraded array
4546 * of that size
4547 */
4548 return -EINVAL;
4549
f6705578 4550 atomic_set(&conf->reshape_stripes, 0);
29269553
N		 4551 spin_lock_irq(&conf->device_lock);
4552 conf->previous_raid_disks = conf->raid_disks;
63c70c4f 4553 conf->raid_disks += mddev->delta_disks;
29269553 4554 conf->expand_progress = 0;
b578d55f 4555 conf->expand_lo = 0;
29269553
N		 4556 spin_unlock_irq(&conf->device_lock);
4557
4558 /* Add some new drives, as many as will fit.
4559 * We know there are enough to make the newly sized array work.
4560 */
d089c6af 4561 rdev_for_each(rdev, rtmp, mddev)
29269553
N		 4562 if (rdev->raid_disk < 0 &&
4563 !test_bit(Faulty, &rdev->flags)) {
199050ea 4564 if (raid5_add_disk(mddev, rdev) == 0) {
29269553
N		 4565 char nm[20];
4566 set_bit(In_sync, &rdev->flags);
29269553 4567 added_devices++;
5fd6c1dc 4568 rdev->recovery_offset = 0;
29269553 4569 sprintf(nm, "rd%d", rdev->raid_disk);
5e55e2f5
N		 4570 if (sysfs_create_link(&mddev->kobj,
4571 &rdev->kobj, nm))
4572 printk(KERN_WARNING
4573 "raid5: failed to create "
4574 " link %s for %s\n",
4575 nm, mdname(mddev));
29269553
N		 4576 } else
4577 break;
4578 }
4579
c04be0aa 4580 spin_lock_irqsave(&conf->device_lock, flags);
63c70c4f 4581 mddev->degraded = (conf->raid_disks - conf->previous_raid_disks) - added_devices;
c04be0aa 4582 spin_unlock_irqrestore(&conf->device_lock, flags);
63c70c4f 4583 mddev->raid_disks = conf->raid_disks;
f6705578 4584 mddev->reshape_position = 0;
850b2b42 4585 set_bit(MD_CHANGE_DEVS, &mddev->flags);
f6705578 4586
29269553
N		 4587 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4588 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4589 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
4590 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4591 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
4592 "%s_reshape");
4593 if (!mddev->sync_thread) {
4594 mddev->recovery = 0;
4595 spin_lock_irq(&conf->device_lock);
4596 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
4597 conf->expand_progress = MaxSector;
4598 spin_unlock_irq(&conf->device_lock);
4599 return -EAGAIN;
4600 }
4601 md_wakeup_thread(mddev->sync_thread);
4602 md_new_event(mddev);
4603 return 0;
4604}
4605#endif
4606
4607static void end_reshape(raid5_conf_t *conf)
4608{
4609 struct block_device *bdev;
4610
f6705578 4611 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
f233ea5c 4612 conf->mddev->array_sectors = 2 * conf->mddev->size *
f416885e 4613 (conf->raid_disks - conf->max_degraded);
f233ea5c 4614 set_capacity(conf->mddev->gendisk, conf->mddev->array_sectors);
44ce6294 4615 conf->mddev->changed = 1;
f6705578
N		 4616
4617 bdev = bdget_disk(conf->mddev->gendisk, 0);
4618 if (bdev) {
4619 mutex_lock(&bdev->bd_inode->i_mutex);
f233ea5c
AN		 4620 i_size_write(bdev->bd_inode,
4621 (loff_t)conf->mddev->array_sectors << 9);
f6705578
N		 4622 mutex_unlock(&bdev->bd_inode->i_mutex);
4623 bdput(bdev);
4624 }
4625 spin_lock_irq(&conf->device_lock);
4626 conf->expand_progress = MaxSector;
4627 spin_unlock_irq(&conf->device_lock);
4628 conf->mddev->reshape_position = MaxSector;
16a53ecc
N		 4629
4630 /* read-ahead size must cover two whole stripes, which is
4631 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
4632 */
4633 {
4634 int data_disks = conf->previous_raid_disks - conf->max_degraded;
4635 int stripe = data_disks *
4636 (conf->mddev->chunk_size / PAGE_SIZE);
4637 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
4638 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
4639 }
29269553 4640 }
29269553
N		 4641}
4642
72626685
N		 4643static void raid5_quiesce(mddev_t *mddev, int state)
4644{
4645 raid5_conf_t *conf = mddev_to_conf(mddev);
4646
4647 switch(state) {
e464eafd
N		 4648 case 2: /* resume for a suspend */
4649 wake_up(&conf->wait_for_overlap);
4650 break;
4651
72626685
N		 4652 case 1: /* stop all writes */
4653 spin_lock_irq(&conf->device_lock);
4654 conf->quiesce = 1;
4655 wait_event_lock_irq(conf->wait_for_stripe,
46031f9a
RBJ		 4656 atomic_read(&conf->active_stripes) == 0 &&
4657 atomic_read(&conf->active_aligned_reads) == 0,
72626685
N		 4658 conf->device_lock, /* nothing */);
4659 spin_unlock_irq(&conf->device_lock);
4660 break;
4661
4662 case 0: /* re-enable writes */
4663 spin_lock_irq(&conf->device_lock);
4664 conf->quiesce = 0;
4665 wake_up(&conf->wait_for_stripe);
e464eafd 4666 wake_up(&conf->wait_for_overlap);
72626685
N		 4667 spin_unlock_irq(&conf->device_lock);
4668 break;
4669 }
72626685 4670}
b15c2e57 4671
16a53ecc
N		 4672static struct mdk_personality raid6_personality =
4673{
4674 .name = "raid6",
4675 .level = 6,
4676 .owner = THIS_MODULE,
4677 .make_request = make_request,
4678 .run = run,
4679 .stop = stop,
4680 .status = status,
4681 .error_handler = error,
4682 .hot_add_disk = raid5_add_disk,
4683 .hot_remove_disk= raid5_remove_disk,
4684 .spare_active = raid5_spare_active,
4685 .sync_request = sync_request,
4686 .resize = raid5_resize,
f416885e
N		 4687#ifdef CONFIG_MD_RAID5_RESHAPE
4688 .check_reshape = raid5_check_reshape,
4689 .start_reshape = raid5_start_reshape,
4690#endif
16a53ecc
N		 4691 .quiesce = raid5_quiesce,
4692};
2604b703 4693static struct mdk_personality raid5_personality =
1da177e4
LT		 4694{
4695 .name = "raid5",
2604b703 4696 .level = 5,
1da177e4
LT		 4697 .owner = THIS_MODULE,
4698 .make_request = make_request,
4699 .run = run,
4700 .stop = stop,
4701 .status = status,
4702 .error_handler = error,
4703 .hot_add_disk = raid5_add_disk,
4704 .hot_remove_disk= raid5_remove_disk,
4705 .spare_active = raid5_spare_active,
4706 .sync_request = sync_request,
4707 .resize = raid5_resize,
29269553 4708#ifdef CONFIG_MD_RAID5_RESHAPE
63c70c4f
N		 4709 .check_reshape = raid5_check_reshape,
4710 .start_reshape = raid5_start_reshape,
29269553 4711#endif
72626685 4712 .quiesce = raid5_quiesce,
1da177e4
LT		 4713};
4714
2604b703 4715static struct mdk_personality raid4_personality =
1da177e4 4716{
2604b703
N		 4717 .name = "raid4",
4718 .level = 4,
4719 .owner = THIS_MODULE,
4720 .make_request = make_request,
4721 .run = run,
4722 .stop = stop,
4723 .status = status,
4724 .error_handler = error,
4725 .hot_add_disk = raid5_add_disk,
4726 .hot_remove_disk= raid5_remove_disk,
4727 .spare_active = raid5_spare_active,
4728 .sync_request = sync_request,
4729 .resize = raid5_resize,
3d37890b
N		 4730#ifdef CONFIG_MD_RAID5_RESHAPE
4731 .check_reshape = raid5_check_reshape,
4732 .start_reshape = raid5_start_reshape,
4733#endif
2604b703
N		 4734 .quiesce = raid5_quiesce,
4735};
4736
4737static int __init raid5_init(void)
4738{
16a53ecc
N		 4739 int e;
4740
4741 e = raid6_select_algo();
4742 if ( e )
4743 return e;
4744 register_md_personality(&raid6_personality);
2604b703
N		 4745 register_md_personality(&raid5_personality);
4746 register_md_personality(&raid4_personality);
4747 return 0;
1da177e4
LT		 4748}
4749
2604b703 4750static void raid5_exit(void)
1da177e4 4751{
16a53ecc 4752 unregister_md_personality(&raid6_personality);
2604b703
N		 4753 unregister_md_personality(&raid5_personality);
4754 unregister_md_personality(&raid4_personality);
1da177e4
LT		 4755}
4756
4757module_init(raid5_init);
4758module_exit(raid5_exit);
4759MODULE_LICENSE("GPL");
4760MODULE_ALIAS("md-personality-4"); /* RAID5 */
d9d166c2
N		 4761MODULE_ALIAS("md-raid5");
4762MODULE_ALIAS("md-raid4");
2604b703
N		 4763MODULE_ALIAS("md-level-5");
4764MODULE_ALIAS("md-level-4");
16a53ecc
N		 4765MODULE_ALIAS("md-personality-8"); /* RAID6 */
4766MODULE_ALIAS("md-raid6");
4767MODULE_ALIAS("md-level-6");
4768
4769/* This used to be two separate modules, they were: */
4770MODULE_ALIAS("raid5");
4771MODULE_ALIAS("raid6");
Linux kernel - Deliverables
This page took 0.689755 seconds and 5 git commands to generate.