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
5 * Copyright (C) 2002, 2003 H. Peter Anvin
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!
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)
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.
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
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
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
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
46 #include <linux/module.h>
47 #include <linux/slab.h>
48 #include <linux/highmem.h>
49 #include <linux/bitops.h>
50 #include <linux/kthread.h>
51 #include <asm/atomic.h>
54 #include <linux/raid/bitmap.h>
60 #define NR_STRIPES 256
61 #define STRIPE_SIZE PAGE_SIZE
62 #define STRIPE_SHIFT (PAGE_SHIFT - 9)
63 #define STRIPE_SECTORS (STRIPE_SIZE>>9)
64 #define IO_THRESHOLD 1
65 #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
66 #define HASH_MASK (NR_HASH - 1)
68 #define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK]))
70 /* bio's attached to a stripe+device for I/O are linked together in bi_sector
71 * order without overlap. There may be several bio's per stripe+device, and
72 * a bio could span several devices.
73 * When walking this list for a particular stripe+device, we must never proceed
74 * beyond a bio that extends past this device, as the next bio might no longer
76 * This macro is used to determine the 'next' bio in the list, given the sector
77 * of the current stripe+device
79 #define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL)
81 * The following can be used to debug the driver
84 #define RAID5_PARANOIA 1
85 #if RAID5_PARANOIA && defined(CONFIG_SMP)
86 # define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock)
88 # define CHECK_DEVLOCK()
91 #define PRINTK(x...) ((void)(RAID5_DEBUG && printk(x)))
97 #if !RAID6_USE_EMPTY_ZERO_PAGE
98 /* In .bss so it's zeroed */
99 const char raid6_empty_zero_page
[PAGE_SIZE
] __attribute__((aligned(256)));
102 static inline int raid6_next_disk(int disk
, int raid_disks
)
105 return (disk
< raid_disks
) ? disk
: 0;
107 static void print_raid5_conf (raid5_conf_t
*conf
);
109 static void __release_stripe(raid5_conf_t
*conf
, struct stripe_head
*sh
)
111 if (atomic_dec_and_test(&sh
->count
)) {
112 BUG_ON(!list_empty(&sh
->lru
));
113 BUG_ON(atomic_read(&conf
->active_stripes
)==0);
114 if (test_bit(STRIPE_HANDLE
, &sh
->state
)) {
115 if (test_bit(STRIPE_DELAYED
, &sh
->state
)) {
116 list_add_tail(&sh
->lru
, &conf
->delayed_list
);
117 blk_plug_device(conf
->mddev
->queue
);
118 } else if (test_bit(STRIPE_BIT_DELAY
, &sh
->state
) &&
119 sh
->bm_seq
- conf
->seq_write
> 0) {
120 list_add_tail(&sh
->lru
, &conf
->bitmap_list
);
121 blk_plug_device(conf
->mddev
->queue
);
123 clear_bit(STRIPE_BIT_DELAY
, &sh
->state
);
124 list_add_tail(&sh
->lru
, &conf
->handle_list
);
126 md_wakeup_thread(conf
->mddev
->thread
);
128 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
129 atomic_dec(&conf
->preread_active_stripes
);
130 if (atomic_read(&conf
->preread_active_stripes
) < IO_THRESHOLD
)
131 md_wakeup_thread(conf
->mddev
->thread
);
133 atomic_dec(&conf
->active_stripes
);
134 if (!test_bit(STRIPE_EXPANDING
, &sh
->state
)) {
135 list_add_tail(&sh
->lru
, &conf
->inactive_list
);
136 wake_up(&conf
->wait_for_stripe
);
137 if (conf
->retry_read_aligned
)
138 md_wakeup_thread(conf
->mddev
->thread
);
143 static void release_stripe(struct stripe_head
*sh
)
145 raid5_conf_t
*conf
= sh
->raid_conf
;
148 spin_lock_irqsave(&conf
->device_lock
, flags
);
149 __release_stripe(conf
, sh
);
150 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
153 static inline void remove_hash(struct stripe_head
*sh
)
155 PRINTK("remove_hash(), stripe %llu\n", (unsigned long long)sh
->sector
);
157 hlist_del_init(&sh
->hash
);
160 static inline void insert_hash(raid5_conf_t
*conf
, struct stripe_head
*sh
)
162 struct hlist_head
*hp
= stripe_hash(conf
, sh
->sector
);
164 PRINTK("insert_hash(), stripe %llu\n", (unsigned long long)sh
->sector
);
167 hlist_add_head(&sh
->hash
, hp
);
171 /* find an idle stripe, make sure it is unhashed, and return it. */
172 static struct stripe_head
*get_free_stripe(raid5_conf_t
*conf
)
174 struct stripe_head
*sh
= NULL
;
175 struct list_head
*first
;
178 if (list_empty(&conf
->inactive_list
))
180 first
= conf
->inactive_list
.next
;
181 sh
= list_entry(first
, struct stripe_head
, lru
);
182 list_del_init(first
);
184 atomic_inc(&conf
->active_stripes
);
189 static void shrink_buffers(struct stripe_head
*sh
, int num
)
194 for (i
=0; i
<num
; i
++) {
198 sh
->dev
[i
].page
= NULL
;
203 static int grow_buffers(struct stripe_head
*sh
, int num
)
207 for (i
=0; i
<num
; i
++) {
210 if (!(page
= alloc_page(GFP_KERNEL
))) {
213 sh
->dev
[i
].page
= page
;
218 static void raid5_build_block (struct stripe_head
*sh
, int i
);
220 static void init_stripe(struct stripe_head
*sh
, sector_t sector
, int pd_idx
, int disks
)
222 raid5_conf_t
*conf
= sh
->raid_conf
;
225 BUG_ON(atomic_read(&sh
->count
) != 0);
226 BUG_ON(test_bit(STRIPE_HANDLE
, &sh
->state
));
229 PRINTK("init_stripe called, stripe %llu\n",
230 (unsigned long long)sh
->sector
);
240 for (i
= sh
->disks
; i
--; ) {
241 struct r5dev
*dev
= &sh
->dev
[i
];
243 if (dev
->toread
|| dev
->towrite
|| dev
->written
||
244 test_bit(R5_LOCKED
, &dev
->flags
)) {
245 printk("sector=%llx i=%d %p %p %p %d\n",
246 (unsigned long long)sh
->sector
, i
, dev
->toread
,
247 dev
->towrite
, dev
->written
,
248 test_bit(R5_LOCKED
, &dev
->flags
));
252 raid5_build_block(sh
, i
);
254 insert_hash(conf
, sh
);
257 static struct stripe_head
*__find_stripe(raid5_conf_t
*conf
, sector_t sector
, int disks
)
259 struct stripe_head
*sh
;
260 struct hlist_node
*hn
;
263 PRINTK("__find_stripe, sector %llu\n", (unsigned long long)sector
);
264 hlist_for_each_entry(sh
, hn
, stripe_hash(conf
, sector
), hash
)
265 if (sh
->sector
== sector
&& sh
->disks
== disks
)
267 PRINTK("__stripe %llu not in cache\n", (unsigned long long)sector
);
271 static void unplug_slaves(mddev_t
*mddev
);
272 static void raid5_unplug_device(request_queue_t
*q
);
274 static struct stripe_head
*get_active_stripe(raid5_conf_t
*conf
, sector_t sector
, int disks
,
275 int pd_idx
, int noblock
)
277 struct stripe_head
*sh
;
279 PRINTK("get_stripe, sector %llu\n", (unsigned long long)sector
);
281 spin_lock_irq(&conf
->device_lock
);
284 wait_event_lock_irq(conf
->wait_for_stripe
,
286 conf
->device_lock
, /* nothing */);
287 sh
= __find_stripe(conf
, sector
, disks
);
289 if (!conf
->inactive_blocked
)
290 sh
= get_free_stripe(conf
);
291 if (noblock
&& sh
== NULL
)
294 conf
->inactive_blocked
= 1;
295 wait_event_lock_irq(conf
->wait_for_stripe
,
296 !list_empty(&conf
->inactive_list
) &&
297 (atomic_read(&conf
->active_stripes
)
298 < (conf
->max_nr_stripes
*3/4)
299 || !conf
->inactive_blocked
),
301 raid5_unplug_device(conf
->mddev
->queue
)
303 conf
->inactive_blocked
= 0;
305 init_stripe(sh
, sector
, pd_idx
, disks
);
307 if (atomic_read(&sh
->count
)) {
308 BUG_ON(!list_empty(&sh
->lru
));
310 if (!test_bit(STRIPE_HANDLE
, &sh
->state
))
311 atomic_inc(&conf
->active_stripes
);
312 if (list_empty(&sh
->lru
) &&
313 !test_bit(STRIPE_EXPANDING
, &sh
->state
))
315 list_del_init(&sh
->lru
);
318 } while (sh
== NULL
);
321 atomic_inc(&sh
->count
);
323 spin_unlock_irq(&conf
->device_lock
);
327 static int grow_one_stripe(raid5_conf_t
*conf
)
329 struct stripe_head
*sh
;
330 sh
= kmem_cache_alloc(conf
->slab_cache
, GFP_KERNEL
);
333 memset(sh
, 0, sizeof(*sh
) + (conf
->raid_disks
-1)*sizeof(struct r5dev
));
334 sh
->raid_conf
= conf
;
335 spin_lock_init(&sh
->lock
);
337 if (grow_buffers(sh
, conf
->raid_disks
)) {
338 shrink_buffers(sh
, conf
->raid_disks
);
339 kmem_cache_free(conf
->slab_cache
, sh
);
342 sh
->disks
= conf
->raid_disks
;
343 /* we just created an active stripe so... */
344 atomic_set(&sh
->count
, 1);
345 atomic_inc(&conf
->active_stripes
);
346 INIT_LIST_HEAD(&sh
->lru
);
351 static int grow_stripes(raid5_conf_t
*conf
, int num
)
353 struct kmem_cache
*sc
;
354 int devs
= conf
->raid_disks
;
356 sprintf(conf
->cache_name
[0], "raid5-%s", mdname(conf
->mddev
));
357 sprintf(conf
->cache_name
[1], "raid5-%s-alt", mdname(conf
->mddev
));
358 conf
->active_name
= 0;
359 sc
= kmem_cache_create(conf
->cache_name
[conf
->active_name
],
360 sizeof(struct stripe_head
)+(devs
-1)*sizeof(struct r5dev
),
364 conf
->slab_cache
= sc
;
365 conf
->pool_size
= devs
;
367 if (!grow_one_stripe(conf
))
372 #ifdef CONFIG_MD_RAID5_RESHAPE
373 static int resize_stripes(raid5_conf_t
*conf
, int newsize
)
375 /* Make all the stripes able to hold 'newsize' devices.
376 * New slots in each stripe get 'page' set to a new page.
378 * This happens in stages:
379 * 1/ create a new kmem_cache and allocate the required number of
381 * 2/ gather all the old stripe_heads and tranfer the pages across
382 * to the new stripe_heads. This will have the side effect of
383 * freezing the array as once all stripe_heads have been collected,
384 * no IO will be possible. Old stripe heads are freed once their
385 * pages have been transferred over, and the old kmem_cache is
386 * freed when all stripes are done.
387 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
388 * we simple return a failre status - no need to clean anything up.
389 * 4/ allocate new pages for the new slots in the new stripe_heads.
390 * If this fails, we don't bother trying the shrink the
391 * stripe_heads down again, we just leave them as they are.
392 * As each stripe_head is processed the new one is released into
395 * Once step2 is started, we cannot afford to wait for a write,
396 * so we use GFP_NOIO allocations.
398 struct stripe_head
*osh
, *nsh
;
399 LIST_HEAD(newstripes
);
400 struct disk_info
*ndisks
;
402 struct kmem_cache
*sc
;
405 if (newsize
<= conf
->pool_size
)
406 return 0; /* never bother to shrink */
408 md_allow_write(conf
->mddev
);
411 sc
= kmem_cache_create(conf
->cache_name
[1-conf
->active_name
],
412 sizeof(struct stripe_head
)+(newsize
-1)*sizeof(struct r5dev
),
417 for (i
= conf
->max_nr_stripes
; i
; i
--) {
418 nsh
= kmem_cache_alloc(sc
, GFP_KERNEL
);
422 memset(nsh
, 0, sizeof(*nsh
) + (newsize
-1)*sizeof(struct r5dev
));
424 nsh
->raid_conf
= conf
;
425 spin_lock_init(&nsh
->lock
);
427 list_add(&nsh
->lru
, &newstripes
);
430 /* didn't get enough, give up */
431 while (!list_empty(&newstripes
)) {
432 nsh
= list_entry(newstripes
.next
, struct stripe_head
, lru
);
434 kmem_cache_free(sc
, nsh
);
436 kmem_cache_destroy(sc
);
439 /* Step 2 - Must use GFP_NOIO now.
440 * OK, we have enough stripes, start collecting inactive
441 * stripes and copying them over
443 list_for_each_entry(nsh
, &newstripes
, lru
) {
444 spin_lock_irq(&conf
->device_lock
);
445 wait_event_lock_irq(conf
->wait_for_stripe
,
446 !list_empty(&conf
->inactive_list
),
448 unplug_slaves(conf
->mddev
)
450 osh
= get_free_stripe(conf
);
451 spin_unlock_irq(&conf
->device_lock
);
452 atomic_set(&nsh
->count
, 1);
453 for(i
=0; i
<conf
->pool_size
; i
++)
454 nsh
->dev
[i
].page
= osh
->dev
[i
].page
;
455 for( ; i
<newsize
; i
++)
456 nsh
->dev
[i
].page
= NULL
;
457 kmem_cache_free(conf
->slab_cache
, osh
);
459 kmem_cache_destroy(conf
->slab_cache
);
462 * At this point, we are holding all the stripes so the array
463 * is completely stalled, so now is a good time to resize
466 ndisks
= kzalloc(newsize
* sizeof(struct disk_info
), GFP_NOIO
);
468 for (i
=0; i
<conf
->raid_disks
; i
++)
469 ndisks
[i
] = conf
->disks
[i
];
471 conf
->disks
= ndisks
;
475 /* Step 4, return new stripes to service */
476 while(!list_empty(&newstripes
)) {
477 nsh
= list_entry(newstripes
.next
, struct stripe_head
, lru
);
478 list_del_init(&nsh
->lru
);
479 for (i
=conf
->raid_disks
; i
< newsize
; i
++)
480 if (nsh
->dev
[i
].page
== NULL
) {
481 struct page
*p
= alloc_page(GFP_NOIO
);
482 nsh
->dev
[i
].page
= p
;
488 /* critical section pass, GFP_NOIO no longer needed */
490 conf
->slab_cache
= sc
;
491 conf
->active_name
= 1-conf
->active_name
;
492 conf
->pool_size
= newsize
;
497 static int drop_one_stripe(raid5_conf_t
*conf
)
499 struct stripe_head
*sh
;
501 spin_lock_irq(&conf
->device_lock
);
502 sh
= get_free_stripe(conf
);
503 spin_unlock_irq(&conf
->device_lock
);
506 BUG_ON(atomic_read(&sh
->count
));
507 shrink_buffers(sh
, conf
->pool_size
);
508 kmem_cache_free(conf
->slab_cache
, sh
);
509 atomic_dec(&conf
->active_stripes
);
513 static void shrink_stripes(raid5_conf_t
*conf
)
515 while (drop_one_stripe(conf
))
518 if (conf
->slab_cache
)
519 kmem_cache_destroy(conf
->slab_cache
);
520 conf
->slab_cache
= NULL
;
523 static int raid5_end_read_request(struct bio
* bi
, unsigned int bytes_done
,
526 struct stripe_head
*sh
= bi
->bi_private
;
527 raid5_conf_t
*conf
= sh
->raid_conf
;
528 int disks
= sh
->disks
, i
;
529 int uptodate
= test_bit(BIO_UPTODATE
, &bi
->bi_flags
);
530 char b
[BDEVNAME_SIZE
];
536 for (i
=0 ; i
<disks
; i
++)
537 if (bi
== &sh
->dev
[i
].req
)
540 PRINTK("end_read_request %llu/%d, count: %d, uptodate %d.\n",
541 (unsigned long long)sh
->sector
, i
, atomic_read(&sh
->count
),
549 set_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
);
550 if (test_bit(R5_ReadError
, &sh
->dev
[i
].flags
)) {
551 rdev
= conf
->disks
[i
].rdev
;
552 printk(KERN_INFO
"raid5:%s: read error corrected (%lu sectors at %llu on %s)\n",
553 mdname(conf
->mddev
), STRIPE_SECTORS
,
554 (unsigned long long)sh
->sector
+ rdev
->data_offset
,
555 bdevname(rdev
->bdev
, b
));
556 clear_bit(R5_ReadError
, &sh
->dev
[i
].flags
);
557 clear_bit(R5_ReWrite
, &sh
->dev
[i
].flags
);
559 if (atomic_read(&conf
->disks
[i
].rdev
->read_errors
))
560 atomic_set(&conf
->disks
[i
].rdev
->read_errors
, 0);
562 const char *bdn
= bdevname(conf
->disks
[i
].rdev
->bdev
, b
);
564 rdev
= conf
->disks
[i
].rdev
;
566 clear_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
);
567 atomic_inc(&rdev
->read_errors
);
568 if (conf
->mddev
->degraded
)
569 printk(KERN_WARNING
"raid5:%s: read error not correctable (sector %llu on %s).\n",
571 (unsigned long long)sh
->sector
+ rdev
->data_offset
,
573 else if (test_bit(R5_ReWrite
, &sh
->dev
[i
].flags
))
575 printk(KERN_WARNING
"raid5:%s: read error NOT corrected!! (sector %llu on %s).\n",
577 (unsigned long long)sh
->sector
+ rdev
->data_offset
,
579 else if (atomic_read(&rdev
->read_errors
)
580 > conf
->max_nr_stripes
)
582 "raid5:%s: Too many read errors, failing device %s.\n",
583 mdname(conf
->mddev
), bdn
);
587 set_bit(R5_ReadError
, &sh
->dev
[i
].flags
);
589 clear_bit(R5_ReadError
, &sh
->dev
[i
].flags
);
590 clear_bit(R5_ReWrite
, &sh
->dev
[i
].flags
);
591 md_error(conf
->mddev
, rdev
);
594 rdev_dec_pending(conf
->disks
[i
].rdev
, conf
->mddev
);
595 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
596 set_bit(STRIPE_HANDLE
, &sh
->state
);
601 static int raid5_end_write_request (struct bio
*bi
, unsigned int bytes_done
,
604 struct stripe_head
*sh
= bi
->bi_private
;
605 raid5_conf_t
*conf
= sh
->raid_conf
;
606 int disks
= sh
->disks
, i
;
607 int uptodate
= test_bit(BIO_UPTODATE
, &bi
->bi_flags
);
612 for (i
=0 ; i
<disks
; i
++)
613 if (bi
== &sh
->dev
[i
].req
)
616 PRINTK("end_write_request %llu/%d, count %d, uptodate: %d.\n",
617 (unsigned long long)sh
->sector
, i
, atomic_read(&sh
->count
),
625 md_error(conf
->mddev
, conf
->disks
[i
].rdev
);
627 rdev_dec_pending(conf
->disks
[i
].rdev
, conf
->mddev
);
629 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
630 set_bit(STRIPE_HANDLE
, &sh
->state
);
636 static sector_t
compute_blocknr(struct stripe_head
*sh
, int i
);
638 static void raid5_build_block (struct stripe_head
*sh
, int i
)
640 struct r5dev
*dev
= &sh
->dev
[i
];
643 dev
->req
.bi_io_vec
= &dev
->vec
;
645 dev
->req
.bi_max_vecs
++;
646 dev
->vec
.bv_page
= dev
->page
;
647 dev
->vec
.bv_len
= STRIPE_SIZE
;
648 dev
->vec
.bv_offset
= 0;
650 dev
->req
.bi_sector
= sh
->sector
;
651 dev
->req
.bi_private
= sh
;
654 dev
->sector
= compute_blocknr(sh
, i
);
657 static void error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
659 char b
[BDEVNAME_SIZE
];
660 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
661 PRINTK("raid5: error called\n");
663 if (!test_bit(Faulty
, &rdev
->flags
)) {
664 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
665 if (test_and_clear_bit(In_sync
, &rdev
->flags
)) {
667 spin_lock_irqsave(&conf
->device_lock
, flags
);
669 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
671 * if recovery was running, make sure it aborts.
673 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
675 set_bit(Faulty
, &rdev
->flags
);
677 "raid5: Disk failure on %s, disabling device."
678 " Operation continuing on %d devices\n",
679 bdevname(rdev
->bdev
,b
), conf
->raid_disks
- mddev
->degraded
);
684 * Input: a 'big' sector number,
685 * Output: index of the data and parity disk, and the sector # in them.
687 static sector_t
raid5_compute_sector(sector_t r_sector
, unsigned int raid_disks
,
688 unsigned int data_disks
, unsigned int * dd_idx
,
689 unsigned int * pd_idx
, raid5_conf_t
*conf
)
692 unsigned long chunk_number
;
693 unsigned int chunk_offset
;
695 int sectors_per_chunk
= conf
->chunk_size
>> 9;
697 /* First compute the information on this sector */
700 * Compute the chunk number and the sector offset inside the chunk
702 chunk_offset
= sector_div(r_sector
, sectors_per_chunk
);
703 chunk_number
= r_sector
;
704 BUG_ON(r_sector
!= chunk_number
);
707 * Compute the stripe number
709 stripe
= chunk_number
/ data_disks
;
712 * Compute the data disk and parity disk indexes inside the stripe
714 *dd_idx
= chunk_number
% data_disks
;
717 * Select the parity disk based on the user selected algorithm.
719 switch(conf
->level
) {
721 *pd_idx
= data_disks
;
724 switch (conf
->algorithm
) {
725 case ALGORITHM_LEFT_ASYMMETRIC
:
726 *pd_idx
= data_disks
- stripe
% raid_disks
;
727 if (*dd_idx
>= *pd_idx
)
730 case ALGORITHM_RIGHT_ASYMMETRIC
:
731 *pd_idx
= stripe
% raid_disks
;
732 if (*dd_idx
>= *pd_idx
)
735 case ALGORITHM_LEFT_SYMMETRIC
:
736 *pd_idx
= data_disks
- stripe
% raid_disks
;
737 *dd_idx
= (*pd_idx
+ 1 + *dd_idx
) % raid_disks
;
739 case ALGORITHM_RIGHT_SYMMETRIC
:
740 *pd_idx
= stripe
% raid_disks
;
741 *dd_idx
= (*pd_idx
+ 1 + *dd_idx
) % raid_disks
;
744 printk(KERN_ERR
"raid5: unsupported algorithm %d\n",
751 switch (conf
->algorithm
) {
752 case ALGORITHM_LEFT_ASYMMETRIC
:
753 *pd_idx
= raid_disks
- 1 - (stripe
% raid_disks
);
754 if (*pd_idx
== raid_disks
-1)
755 (*dd_idx
)++; /* Q D D D P */
756 else if (*dd_idx
>= *pd_idx
)
757 (*dd_idx
) += 2; /* D D P Q D */
759 case ALGORITHM_RIGHT_ASYMMETRIC
:
760 *pd_idx
= stripe
% raid_disks
;
761 if (*pd_idx
== raid_disks
-1)
762 (*dd_idx
)++; /* Q D D D P */
763 else if (*dd_idx
>= *pd_idx
)
764 (*dd_idx
) += 2; /* D D P Q D */
766 case ALGORITHM_LEFT_SYMMETRIC
:
767 *pd_idx
= raid_disks
- 1 - (stripe
% raid_disks
);
768 *dd_idx
= (*pd_idx
+ 2 + *dd_idx
) % raid_disks
;
770 case ALGORITHM_RIGHT_SYMMETRIC
:
771 *pd_idx
= stripe
% raid_disks
;
772 *dd_idx
= (*pd_idx
+ 2 + *dd_idx
) % raid_disks
;
775 printk (KERN_CRIT
"raid6: unsupported algorithm %d\n",
782 * Finally, compute the new sector number
784 new_sector
= (sector_t
)stripe
* sectors_per_chunk
+ chunk_offset
;
789 static sector_t
compute_blocknr(struct stripe_head
*sh
, int i
)
791 raid5_conf_t
*conf
= sh
->raid_conf
;
792 int raid_disks
= sh
->disks
;
793 int data_disks
= raid_disks
- conf
->max_degraded
;
794 sector_t new_sector
= sh
->sector
, check
;
795 int sectors_per_chunk
= conf
->chunk_size
>> 9;
798 int chunk_number
, dummy1
, dummy2
, dd_idx
= i
;
802 chunk_offset
= sector_div(new_sector
, sectors_per_chunk
);
804 BUG_ON(new_sector
!= stripe
);
808 switch(conf
->level
) {
811 switch (conf
->algorithm
) {
812 case ALGORITHM_LEFT_ASYMMETRIC
:
813 case ALGORITHM_RIGHT_ASYMMETRIC
:
817 case ALGORITHM_LEFT_SYMMETRIC
:
818 case ALGORITHM_RIGHT_SYMMETRIC
:
821 i
-= (sh
->pd_idx
+ 1);
824 printk(KERN_ERR
"raid5: unsupported algorithm %d\n",
829 if (i
== raid6_next_disk(sh
->pd_idx
, raid_disks
))
830 return 0; /* It is the Q disk */
831 switch (conf
->algorithm
) {
832 case ALGORITHM_LEFT_ASYMMETRIC
:
833 case ALGORITHM_RIGHT_ASYMMETRIC
:
834 if (sh
->pd_idx
== raid_disks
-1)
836 else if (i
> sh
->pd_idx
)
837 i
-= 2; /* D D P Q D */
839 case ALGORITHM_LEFT_SYMMETRIC
:
840 case ALGORITHM_RIGHT_SYMMETRIC
:
841 if (sh
->pd_idx
== raid_disks
-1)
847 i
-= (sh
->pd_idx
+ 2);
851 printk (KERN_CRIT
"raid6: unsupported algorithm %d\n",
857 chunk_number
= stripe
* data_disks
+ i
;
858 r_sector
= (sector_t
)chunk_number
* sectors_per_chunk
+ chunk_offset
;
860 check
= raid5_compute_sector (r_sector
, raid_disks
, data_disks
, &dummy1
, &dummy2
, conf
);
861 if (check
!= sh
->sector
|| dummy1
!= dd_idx
|| dummy2
!= sh
->pd_idx
) {
862 printk(KERN_ERR
"compute_blocknr: map not correct\n");
871 * Copy data between a page in the stripe cache, and one or more bion
872 * The page could align with the middle of the bio, or there could be
873 * several bion, each with several bio_vecs, which cover part of the page
874 * Multiple bion are linked together on bi_next. There may be extras
875 * at the end of this list. We ignore them.
877 static void copy_data(int frombio
, struct bio
*bio
,
881 char *pa
= page_address(page
);
886 if (bio
->bi_sector
>= sector
)
887 page_offset
= (signed)(bio
->bi_sector
- sector
) * 512;
889 page_offset
= (signed)(sector
- bio
->bi_sector
) * -512;
890 bio_for_each_segment(bvl
, bio
, i
) {
891 int len
= bio_iovec_idx(bio
,i
)->bv_len
;
895 if (page_offset
< 0) {
896 b_offset
= -page_offset
;
897 page_offset
+= b_offset
;
901 if (len
> 0 && page_offset
+ len
> STRIPE_SIZE
)
902 clen
= STRIPE_SIZE
- page_offset
;
906 char *ba
= __bio_kmap_atomic(bio
, i
, KM_USER0
);
908 memcpy(pa
+page_offset
, ba
+b_offset
, clen
);
910 memcpy(ba
+b_offset
, pa
+page_offset
, clen
);
911 __bio_kunmap_atomic(ba
, KM_USER0
);
913 if (clen
< len
) /* hit end of page */
919 #define check_xor() do { \
920 if (count == MAX_XOR_BLOCKS) { \
921 xor_block(count, STRIPE_SIZE, ptr); \
927 static void compute_block(struct stripe_head
*sh
, int dd_idx
)
929 int i
, count
, disks
= sh
->disks
;
930 void *ptr
[MAX_XOR_BLOCKS
], *p
;
932 PRINTK("compute_block, stripe %llu, idx %d\n",
933 (unsigned long long)sh
->sector
, dd_idx
);
935 ptr
[0] = page_address(sh
->dev
[dd_idx
].page
);
936 memset(ptr
[0], 0, STRIPE_SIZE
);
938 for (i
= disks
; i
--; ) {
941 p
= page_address(sh
->dev
[i
].page
);
942 if (test_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
))
945 printk(KERN_ERR
"compute_block() %d, stripe %llu, %d"
946 " not present\n", dd_idx
,
947 (unsigned long long)sh
->sector
, i
);
952 xor_block(count
, STRIPE_SIZE
, ptr
);
953 set_bit(R5_UPTODATE
, &sh
->dev
[dd_idx
].flags
);
956 static void compute_parity5(struct stripe_head
*sh
, int method
)
958 raid5_conf_t
*conf
= sh
->raid_conf
;
959 int i
, pd_idx
= sh
->pd_idx
, disks
= sh
->disks
, count
;
960 void *ptr
[MAX_XOR_BLOCKS
];
963 PRINTK("compute_parity5, stripe %llu, method %d\n",
964 (unsigned long long)sh
->sector
, method
);
967 ptr
[0] = page_address(sh
->dev
[pd_idx
].page
);
969 case READ_MODIFY_WRITE
:
970 BUG_ON(!test_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
));
971 for (i
=disks
; i
-- ;) {
974 if (sh
->dev
[i
].towrite
&&
975 test_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
)) {
976 ptr
[count
++] = page_address(sh
->dev
[i
].page
);
977 chosen
= sh
->dev
[i
].towrite
;
978 sh
->dev
[i
].towrite
= NULL
;
980 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
981 wake_up(&conf
->wait_for_overlap
);
983 BUG_ON(sh
->dev
[i
].written
);
984 sh
->dev
[i
].written
= chosen
;
989 case RECONSTRUCT_WRITE
:
990 memset(ptr
[0], 0, STRIPE_SIZE
);
991 for (i
= disks
; i
-- ;)
992 if (i
!=pd_idx
&& sh
->dev
[i
].towrite
) {
993 chosen
= sh
->dev
[i
].towrite
;
994 sh
->dev
[i
].towrite
= NULL
;
996 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
997 wake_up(&conf
->wait_for_overlap
);
999 BUG_ON(sh
->dev
[i
].written
);
1000 sh
->dev
[i
].written
= chosen
;
1007 xor_block(count
, STRIPE_SIZE
, ptr
);
1011 for (i
= disks
; i
--;)
1012 if (sh
->dev
[i
].written
) {
1013 sector_t sector
= sh
->dev
[i
].sector
;
1014 struct bio
*wbi
= sh
->dev
[i
].written
;
1015 while (wbi
&& wbi
->bi_sector
< sector
+ STRIPE_SECTORS
) {
1016 copy_data(1, wbi
, sh
->dev
[i
].page
, sector
);
1017 wbi
= r5_next_bio(wbi
, sector
);
1020 set_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
1021 set_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
);
1025 case RECONSTRUCT_WRITE
:
1029 ptr
[count
++] = page_address(sh
->dev
[i
].page
);
1033 case READ_MODIFY_WRITE
:
1034 for (i
= disks
; i
--;)
1035 if (sh
->dev
[i
].written
) {
1036 ptr
[count
++] = page_address(sh
->dev
[i
].page
);
1041 xor_block(count
, STRIPE_SIZE
, ptr
);
1043 if (method
!= CHECK_PARITY
) {
1044 set_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
1045 set_bit(R5_LOCKED
, &sh
->dev
[pd_idx
].flags
);
1047 clear_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
1050 static void compute_parity6(struct stripe_head
*sh
, int method
)
1052 raid6_conf_t
*conf
= sh
->raid_conf
;
1053 int i
, pd_idx
= sh
->pd_idx
, qd_idx
, d0_idx
, disks
= sh
->disks
, count
;
1055 /**** FIX THIS: This could be very bad if disks is close to 256 ****/
1058 qd_idx
= raid6_next_disk(pd_idx
, disks
);
1059 d0_idx
= raid6_next_disk(qd_idx
, disks
);
1061 PRINTK("compute_parity, stripe %llu, method %d\n",
1062 (unsigned long long)sh
->sector
, method
);
1065 case READ_MODIFY_WRITE
:
1066 BUG(); /* READ_MODIFY_WRITE N/A for RAID-6 */
1067 case RECONSTRUCT_WRITE
:
1068 for (i
= disks
; i
-- ;)
1069 if ( i
!= pd_idx
&& i
!= qd_idx
&& sh
->dev
[i
].towrite
) {
1070 chosen
= sh
->dev
[i
].towrite
;
1071 sh
->dev
[i
].towrite
= NULL
;
1073 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
1074 wake_up(&conf
->wait_for_overlap
);
1076 BUG_ON(sh
->dev
[i
].written
);
1077 sh
->dev
[i
].written
= chosen
;
1081 BUG(); /* Not implemented yet */
1084 for (i
= disks
; i
--;)
1085 if (sh
->dev
[i
].written
) {
1086 sector_t sector
= sh
->dev
[i
].sector
;
1087 struct bio
*wbi
= sh
->dev
[i
].written
;
1088 while (wbi
&& wbi
->bi_sector
< sector
+ STRIPE_SECTORS
) {
1089 copy_data(1, wbi
, sh
->dev
[i
].page
, sector
);
1090 wbi
= r5_next_bio(wbi
, sector
);
1093 set_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
1094 set_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
);
1098 // case RECONSTRUCT_WRITE:
1099 // case CHECK_PARITY:
1100 // case UPDATE_PARITY:
1101 /* Note that unlike RAID-5, the ordering of the disks matters greatly. */
1102 /* FIX: Is this ordering of drives even remotely optimal? */
1106 ptrs
[count
++] = page_address(sh
->dev
[i
].page
);
1107 if (count
<= disks
-2 && !test_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
))
1108 printk("block %d/%d not uptodate on parity calc\n", i
,count
);
1109 i
= raid6_next_disk(i
, disks
);
1110 } while ( i
!= d0_idx
);
1114 raid6_call
.gen_syndrome(disks
, STRIPE_SIZE
, ptrs
);
1117 case RECONSTRUCT_WRITE
:
1118 set_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
1119 set_bit(R5_UPTODATE
, &sh
->dev
[qd_idx
].flags
);
1120 set_bit(R5_LOCKED
, &sh
->dev
[pd_idx
].flags
);
1121 set_bit(R5_LOCKED
, &sh
->dev
[qd_idx
].flags
);
1124 set_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
1125 set_bit(R5_UPTODATE
, &sh
->dev
[qd_idx
].flags
);
1131 /* Compute one missing block */
1132 static void compute_block_1(struct stripe_head
*sh
, int dd_idx
, int nozero
)
1134 int i
, count
, disks
= sh
->disks
;
1135 void *ptr
[MAX_XOR_BLOCKS
], *p
;
1136 int pd_idx
= sh
->pd_idx
;
1137 int qd_idx
= raid6_next_disk(pd_idx
, disks
);
1139 PRINTK("compute_block_1, stripe %llu, idx %d\n",
1140 (unsigned long long)sh
->sector
, dd_idx
);
1142 if ( dd_idx
== qd_idx
) {
1143 /* We're actually computing the Q drive */
1144 compute_parity6(sh
, UPDATE_PARITY
);
1146 ptr
[0] = page_address(sh
->dev
[dd_idx
].page
);
1147 if (!nozero
) memset(ptr
[0], 0, STRIPE_SIZE
);
1149 for (i
= disks
; i
--; ) {
1150 if (i
== dd_idx
|| i
== qd_idx
)
1152 p
= page_address(sh
->dev
[i
].page
);
1153 if (test_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
))
1156 printk("compute_block() %d, stripe %llu, %d"
1157 " not present\n", dd_idx
,
1158 (unsigned long long)sh
->sector
, i
);
1163 xor_block(count
, STRIPE_SIZE
, ptr
);
1164 if (!nozero
) set_bit(R5_UPTODATE
, &sh
->dev
[dd_idx
].flags
);
1165 else clear_bit(R5_UPTODATE
, &sh
->dev
[dd_idx
].flags
);
1169 /* Compute two missing blocks */
1170 static void compute_block_2(struct stripe_head
*sh
, int dd_idx1
, int dd_idx2
)
1172 int i
, count
, disks
= sh
->disks
;
1173 int pd_idx
= sh
->pd_idx
;
1174 int qd_idx
= raid6_next_disk(pd_idx
, disks
);
1175 int d0_idx
= raid6_next_disk(qd_idx
, disks
);
1178 /* faila and failb are disk numbers relative to d0_idx */
1179 /* pd_idx become disks-2 and qd_idx become disks-1 */
1180 faila
= (dd_idx1
< d0_idx
) ? dd_idx1
+(disks
-d0_idx
) : dd_idx1
-d0_idx
;
1181 failb
= (dd_idx2
< d0_idx
) ? dd_idx2
+(disks
-d0_idx
) : dd_idx2
-d0_idx
;
1183 BUG_ON(faila
== failb
);
1184 if ( failb
< faila
) { int tmp
= faila
; faila
= failb
; failb
= tmp
; }
1186 PRINTK("compute_block_2, stripe %llu, idx %d,%d (%d,%d)\n",
1187 (unsigned long long)sh
->sector
, dd_idx1
, dd_idx2
, faila
, failb
);
1189 if ( failb
== disks
-1 ) {
1190 /* Q disk is one of the missing disks */
1191 if ( faila
== disks
-2 ) {
1192 /* Missing P+Q, just recompute */
1193 compute_parity6(sh
, UPDATE_PARITY
);
1196 /* We're missing D+Q; recompute D from P */
1197 compute_block_1(sh
, (dd_idx1
== qd_idx
) ? dd_idx2
: dd_idx1
, 0);
1198 compute_parity6(sh
, UPDATE_PARITY
); /* Is this necessary? */
1203 /* We're missing D+P or D+D; build pointer table */
1205 /**** FIX THIS: This could be very bad if disks is close to 256 ****/
1211 ptrs
[count
++] = page_address(sh
->dev
[i
].page
);
1212 i
= raid6_next_disk(i
, disks
);
1213 if (i
!= dd_idx1
&& i
!= dd_idx2
&&
1214 !test_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
))
1215 printk("compute_2 with missing block %d/%d\n", count
, i
);
1216 } while ( i
!= d0_idx
);
1218 if ( failb
== disks
-2 ) {
1219 /* We're missing D+P. */
1220 raid6_datap_recov(disks
, STRIPE_SIZE
, faila
, ptrs
);
1222 /* We're missing D+D. */
1223 raid6_2data_recov(disks
, STRIPE_SIZE
, faila
, failb
, ptrs
);
1226 /* Both the above update both missing blocks */
1227 set_bit(R5_UPTODATE
, &sh
->dev
[dd_idx1
].flags
);
1228 set_bit(R5_UPTODATE
, &sh
->dev
[dd_idx2
].flags
);
1235 * Each stripe/dev can have one or more bion attached.
1236 * toread/towrite point to the first in a chain.
1237 * The bi_next chain must be in order.
1239 static int add_stripe_bio(struct stripe_head
*sh
, struct bio
*bi
, int dd_idx
, int forwrite
)
1242 raid5_conf_t
*conf
= sh
->raid_conf
;
1245 PRINTK("adding bh b#%llu to stripe s#%llu\n",
1246 (unsigned long long)bi
->bi_sector
,
1247 (unsigned long long)sh
->sector
);
1250 spin_lock(&sh
->lock
);
1251 spin_lock_irq(&conf
->device_lock
);
1253 bip
= &sh
->dev
[dd_idx
].towrite
;
1254 if (*bip
== NULL
&& sh
->dev
[dd_idx
].written
== NULL
)
1257 bip
= &sh
->dev
[dd_idx
].toread
;
1258 while (*bip
&& (*bip
)->bi_sector
< bi
->bi_sector
) {
1259 if ((*bip
)->bi_sector
+ ((*bip
)->bi_size
>> 9) > bi
->bi_sector
)
1261 bip
= & (*bip
)->bi_next
;
1263 if (*bip
&& (*bip
)->bi_sector
< bi
->bi_sector
+ ((bi
->bi_size
)>>9))
1266 BUG_ON(*bip
&& bi
->bi_next
&& (*bip
) != bi
->bi_next
);
1270 bi
->bi_phys_segments
++;
1271 spin_unlock_irq(&conf
->device_lock
);
1272 spin_unlock(&sh
->lock
);
1274 PRINTK("added bi b#%llu to stripe s#%llu, disk %d.\n",
1275 (unsigned long long)bi
->bi_sector
,
1276 (unsigned long long)sh
->sector
, dd_idx
);
1278 if (conf
->mddev
->bitmap
&& firstwrite
) {
1279 bitmap_startwrite(conf
->mddev
->bitmap
, sh
->sector
,
1281 sh
->bm_seq
= conf
->seq_flush
+1;
1282 set_bit(STRIPE_BIT_DELAY
, &sh
->state
);
1286 /* check if page is covered */
1287 sector_t sector
= sh
->dev
[dd_idx
].sector
;
1288 for (bi
=sh
->dev
[dd_idx
].towrite
;
1289 sector
< sh
->dev
[dd_idx
].sector
+ STRIPE_SECTORS
&&
1290 bi
&& bi
->bi_sector
<= sector
;
1291 bi
= r5_next_bio(bi
, sh
->dev
[dd_idx
].sector
)) {
1292 if (bi
->bi_sector
+ (bi
->bi_size
>>9) >= sector
)
1293 sector
= bi
->bi_sector
+ (bi
->bi_size
>>9);
1295 if (sector
>= sh
->dev
[dd_idx
].sector
+ STRIPE_SECTORS
)
1296 set_bit(R5_OVERWRITE
, &sh
->dev
[dd_idx
].flags
);
1301 set_bit(R5_Overlap
, &sh
->dev
[dd_idx
].flags
);
1302 spin_unlock_irq(&conf
->device_lock
);
1303 spin_unlock(&sh
->lock
);
1307 static void end_reshape(raid5_conf_t
*conf
);
1309 static int page_is_zero(struct page
*p
)
1311 char *a
= page_address(p
);
1312 return ((*(u32
*)a
) == 0 &&
1313 memcmp(a
, a
+4, STRIPE_SIZE
-4)==0);
1316 static int stripe_to_pdidx(sector_t stripe
, raid5_conf_t
*conf
, int disks
)
1318 int sectors_per_chunk
= conf
->chunk_size
>> 9;
1320 int chunk_offset
= sector_div(stripe
, sectors_per_chunk
);
1322 raid5_compute_sector(stripe
* (disks
- conf
->max_degraded
)
1323 *sectors_per_chunk
+ chunk_offset
,
1324 disks
, disks
- conf
->max_degraded
,
1325 &dd_idx
, &pd_idx
, conf
);
1331 * handle_stripe - do things to a stripe.
1333 * We lock the stripe and then examine the state of various bits
1334 * to see what needs to be done.
1336 * return some read request which now have data
1337 * return some write requests which are safely on disc
1338 * schedule a read on some buffers
1339 * schedule a write of some buffers
1340 * return confirmation of parity correctness
1342 * Parity calculations are done inside the stripe lock
1343 * buffers are taken off read_list or write_list, and bh_cache buffers
1344 * get BH_Lock set before the stripe lock is released.
1348 static void handle_stripe5(struct stripe_head
*sh
)
1350 raid5_conf_t
*conf
= sh
->raid_conf
;
1351 int disks
= sh
->disks
;
1352 struct bio
*return_bi
= NULL
;
1355 int syncing
, expanding
, expanded
;
1356 int locked
=0, uptodate
=0, to_read
=0, to_write
=0, failed
=0, written
=0;
1357 int non_overwrite
= 0;
1361 PRINTK("handling stripe %llu, cnt=%d, pd_idx=%d\n",
1362 (unsigned long long)sh
->sector
, atomic_read(&sh
->count
),
1365 spin_lock(&sh
->lock
);
1366 clear_bit(STRIPE_HANDLE
, &sh
->state
);
1367 clear_bit(STRIPE_DELAYED
, &sh
->state
);
1369 syncing
= test_bit(STRIPE_SYNCING
, &sh
->state
);
1370 expanding
= test_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
1371 expanded
= test_bit(STRIPE_EXPAND_READY
, &sh
->state
);
1372 /* Now to look around and see what can be done */
1375 for (i
=disks
; i
--; ) {
1378 clear_bit(R5_Insync
, &dev
->flags
);
1380 PRINTK("check %d: state 0x%lx read %p write %p written %p\n",
1381 i
, dev
->flags
, dev
->toread
, dev
->towrite
, dev
->written
);
1382 /* maybe we can reply to a read */
1383 if (test_bit(R5_UPTODATE
, &dev
->flags
) && dev
->toread
) {
1384 struct bio
*rbi
, *rbi2
;
1385 PRINTK("Return read for disc %d\n", i
);
1386 spin_lock_irq(&conf
->device_lock
);
1389 if (test_and_clear_bit(R5_Overlap
, &dev
->flags
))
1390 wake_up(&conf
->wait_for_overlap
);
1391 spin_unlock_irq(&conf
->device_lock
);
1392 while (rbi
&& rbi
->bi_sector
< dev
->sector
+ STRIPE_SECTORS
) {
1393 copy_data(0, rbi
, dev
->page
, dev
->sector
);
1394 rbi2
= r5_next_bio(rbi
, dev
->sector
);
1395 spin_lock_irq(&conf
->device_lock
);
1396 if (--rbi
->bi_phys_segments
== 0) {
1397 rbi
->bi_next
= return_bi
;
1400 spin_unlock_irq(&conf
->device_lock
);
1405 /* now count some things */
1406 if (test_bit(R5_LOCKED
, &dev
->flags
)) locked
++;
1407 if (test_bit(R5_UPTODATE
, &dev
->flags
)) uptodate
++;
1410 if (dev
->toread
) to_read
++;
1413 if (!test_bit(R5_OVERWRITE
, &dev
->flags
))
1416 if (dev
->written
) written
++;
1417 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
1418 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)) {
1419 /* The ReadError flag will just be confusing now */
1420 clear_bit(R5_ReadError
, &dev
->flags
);
1421 clear_bit(R5_ReWrite
, &dev
->flags
);
1423 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)
1424 || test_bit(R5_ReadError
, &dev
->flags
)) {
1428 set_bit(R5_Insync
, &dev
->flags
);
1431 PRINTK("locked=%d uptodate=%d to_read=%d"
1432 " to_write=%d failed=%d failed_num=%d\n",
1433 locked
, uptodate
, to_read
, to_write
, failed
, failed_num
);
1434 /* check if the array has lost two devices and, if so, some requests might
1437 if (failed
> 1 && to_read
+to_write
+written
) {
1438 for (i
=disks
; i
--; ) {
1441 if (test_bit(R5_ReadError
, &sh
->dev
[i
].flags
)) {
1444 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
1445 if (rdev
&& test_bit(In_sync
, &rdev
->flags
))
1446 /* multiple read failures in one stripe */
1447 md_error(conf
->mddev
, rdev
);
1451 spin_lock_irq(&conf
->device_lock
);
1452 /* fail all writes first */
1453 bi
= sh
->dev
[i
].towrite
;
1454 sh
->dev
[i
].towrite
= NULL
;
1455 if (bi
) { to_write
--; bitmap_end
= 1; }
1457 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
1458 wake_up(&conf
->wait_for_overlap
);
1460 while (bi
&& bi
->bi_sector
< sh
->dev
[i
].sector
+ STRIPE_SECTORS
){
1461 struct bio
*nextbi
= r5_next_bio(bi
, sh
->dev
[i
].sector
);
1462 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1463 if (--bi
->bi_phys_segments
== 0) {
1464 md_write_end(conf
->mddev
);
1465 bi
->bi_next
= return_bi
;
1470 /* and fail all 'written' */
1471 bi
= sh
->dev
[i
].written
;
1472 sh
->dev
[i
].written
= NULL
;
1473 if (bi
) bitmap_end
= 1;
1474 while (bi
&& bi
->bi_sector
< sh
->dev
[i
].sector
+ STRIPE_SECTORS
) {
1475 struct bio
*bi2
= r5_next_bio(bi
, sh
->dev
[i
].sector
);
1476 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1477 if (--bi
->bi_phys_segments
== 0) {
1478 md_write_end(conf
->mddev
);
1479 bi
->bi_next
= return_bi
;
1485 /* fail any reads if this device is non-operational */
1486 if (!test_bit(R5_Insync
, &sh
->dev
[i
].flags
) ||
1487 test_bit(R5_ReadError
, &sh
->dev
[i
].flags
)) {
1488 bi
= sh
->dev
[i
].toread
;
1489 sh
->dev
[i
].toread
= NULL
;
1490 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
1491 wake_up(&conf
->wait_for_overlap
);
1493 while (bi
&& bi
->bi_sector
< sh
->dev
[i
].sector
+ STRIPE_SECTORS
){
1494 struct bio
*nextbi
= r5_next_bio(bi
, sh
->dev
[i
].sector
);
1495 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1496 if (--bi
->bi_phys_segments
== 0) {
1497 bi
->bi_next
= return_bi
;
1503 spin_unlock_irq(&conf
->device_lock
);
1505 bitmap_endwrite(conf
->mddev
->bitmap
, sh
->sector
,
1506 STRIPE_SECTORS
, 0, 0);
1509 if (failed
> 1 && syncing
) {
1510 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,0);
1511 clear_bit(STRIPE_SYNCING
, &sh
->state
);
1515 /* might be able to return some write requests if the parity block
1516 * is safe, or on a failed drive
1518 dev
= &sh
->dev
[sh
->pd_idx
];
1520 ( (test_bit(R5_Insync
, &dev
->flags
) && !test_bit(R5_LOCKED
, &dev
->flags
) &&
1521 test_bit(R5_UPTODATE
, &dev
->flags
))
1522 || (failed
== 1 && failed_num
== sh
->pd_idx
))
1524 /* any written block on an uptodate or failed drive can be returned.
1525 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
1526 * never LOCKED, so we don't need to test 'failed' directly.
1528 for (i
=disks
; i
--; )
1529 if (sh
->dev
[i
].written
) {
1531 if (!test_bit(R5_LOCKED
, &dev
->flags
) &&
1532 test_bit(R5_UPTODATE
, &dev
->flags
) ) {
1533 /* We can return any write requests */
1534 struct bio
*wbi
, *wbi2
;
1536 PRINTK("Return write for disc %d\n", i
);
1537 spin_lock_irq(&conf
->device_lock
);
1539 dev
->written
= NULL
;
1540 while (wbi
&& wbi
->bi_sector
< dev
->sector
+ STRIPE_SECTORS
) {
1541 wbi2
= r5_next_bio(wbi
, dev
->sector
);
1542 if (--wbi
->bi_phys_segments
== 0) {
1543 md_write_end(conf
->mddev
);
1544 wbi
->bi_next
= return_bi
;
1549 if (dev
->towrite
== NULL
)
1551 spin_unlock_irq(&conf
->device_lock
);
1553 bitmap_endwrite(conf
->mddev
->bitmap
, sh
->sector
,
1555 !test_bit(STRIPE_DEGRADED
, &sh
->state
), 0);
1560 /* Now we might consider reading some blocks, either to check/generate
1561 * parity, or to satisfy requests
1562 * or to load a block that is being partially written.
1564 if (to_read
|| non_overwrite
|| (syncing
&& (uptodate
< disks
)) || expanding
) {
1565 for (i
=disks
; i
--;) {
1567 if (!test_bit(R5_LOCKED
, &dev
->flags
) && !test_bit(R5_UPTODATE
, &dev
->flags
) &&
1569 (dev
->towrite
&& !test_bit(R5_OVERWRITE
, &dev
->flags
)) ||
1572 (failed
&& (sh
->dev
[failed_num
].toread
||
1573 (sh
->dev
[failed_num
].towrite
&& !test_bit(R5_OVERWRITE
, &sh
->dev
[failed_num
].flags
))))
1576 /* we would like to get this block, possibly
1577 * by computing it, but we might not be able to
1579 if (uptodate
== disks
-1) {
1580 PRINTK("Computing block %d\n", i
);
1581 compute_block(sh
, i
);
1583 } else if (test_bit(R5_Insync
, &dev
->flags
)) {
1584 set_bit(R5_LOCKED
, &dev
->flags
);
1585 set_bit(R5_Wantread
, &dev
->flags
);
1587 PRINTK("Reading block %d (sync=%d)\n",
1592 set_bit(STRIPE_HANDLE
, &sh
->state
);
1595 /* now to consider writing and what else, if anything should be read */
1598 for (i
=disks
; i
--;) {
1599 /* would I have to read this buffer for read_modify_write */
1601 if ((dev
->towrite
|| i
== sh
->pd_idx
) &&
1602 (!test_bit(R5_LOCKED
, &dev
->flags
)
1604 !test_bit(R5_UPTODATE
, &dev
->flags
)) {
1605 if (test_bit(R5_Insync
, &dev
->flags
)
1606 /* && !(!mddev->insync && i == sh->pd_idx) */
1609 else rmw
+= 2*disks
; /* cannot read it */
1611 /* Would I have to read this buffer for reconstruct_write */
1612 if (!test_bit(R5_OVERWRITE
, &dev
->flags
) && i
!= sh
->pd_idx
&&
1613 (!test_bit(R5_LOCKED
, &dev
->flags
)
1615 !test_bit(R5_UPTODATE
, &dev
->flags
)) {
1616 if (test_bit(R5_Insync
, &dev
->flags
)) rcw
++;
1617 else rcw
+= 2*disks
;
1620 PRINTK("for sector %llu, rmw=%d rcw=%d\n",
1621 (unsigned long long)sh
->sector
, rmw
, rcw
);
1622 set_bit(STRIPE_HANDLE
, &sh
->state
);
1623 if (rmw
< rcw
&& rmw
> 0)
1624 /* prefer read-modify-write, but need to get some data */
1625 for (i
=disks
; i
--;) {
1627 if ((dev
->towrite
|| i
== sh
->pd_idx
) &&
1628 !test_bit(R5_LOCKED
, &dev
->flags
) && !test_bit(R5_UPTODATE
, &dev
->flags
) &&
1629 test_bit(R5_Insync
, &dev
->flags
)) {
1630 if (test_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
))
1632 PRINTK("Read_old block %d for r-m-w\n", i
);
1633 set_bit(R5_LOCKED
, &dev
->flags
);
1634 set_bit(R5_Wantread
, &dev
->flags
);
1637 set_bit(STRIPE_DELAYED
, &sh
->state
);
1638 set_bit(STRIPE_HANDLE
, &sh
->state
);
1642 if (rcw
<= rmw
&& rcw
> 0)
1643 /* want reconstruct write, but need to get some data */
1644 for (i
=disks
; i
--;) {
1646 if (!test_bit(R5_OVERWRITE
, &dev
->flags
) && i
!= sh
->pd_idx
&&
1647 !test_bit(R5_LOCKED
, &dev
->flags
) && !test_bit(R5_UPTODATE
, &dev
->flags
) &&
1648 test_bit(R5_Insync
, &dev
->flags
)) {
1649 if (test_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
))
1651 PRINTK("Read_old block %d for Reconstruct\n", i
);
1652 set_bit(R5_LOCKED
, &dev
->flags
);
1653 set_bit(R5_Wantread
, &dev
->flags
);
1656 set_bit(STRIPE_DELAYED
, &sh
->state
);
1657 set_bit(STRIPE_HANDLE
, &sh
->state
);
1661 /* now if nothing is locked, and if we have enough data, we can start a write request */
1662 if (locked
== 0 && (rcw
== 0 ||rmw
== 0) &&
1663 !test_bit(STRIPE_BIT_DELAY
, &sh
->state
)) {
1664 PRINTK("Computing parity...\n");
1665 compute_parity5(sh
, rcw
==0 ? RECONSTRUCT_WRITE
: READ_MODIFY_WRITE
);
1666 /* now every locked buffer is ready to be written */
1668 if (test_bit(R5_LOCKED
, &sh
->dev
[i
].flags
)) {
1669 PRINTK("Writing block %d\n", i
);
1671 set_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
);
1672 if (!test_bit(R5_Insync
, &sh
->dev
[i
].flags
)
1673 || (i
==sh
->pd_idx
&& failed
== 0))
1674 set_bit(STRIPE_INSYNC
, &sh
->state
);
1676 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
1677 atomic_dec(&conf
->preread_active_stripes
);
1678 if (atomic_read(&conf
->preread_active_stripes
) < IO_THRESHOLD
)
1679 md_wakeup_thread(conf
->mddev
->thread
);
1684 /* maybe we need to check and possibly fix the parity for this stripe
1685 * Any reads will already have been scheduled, so we just see if enough data
1688 if (syncing
&& locked
== 0 &&
1689 !test_bit(STRIPE_INSYNC
, &sh
->state
)) {
1690 set_bit(STRIPE_HANDLE
, &sh
->state
);
1692 BUG_ON(uptodate
!= disks
);
1693 compute_parity5(sh
, CHECK_PARITY
);
1695 if (page_is_zero(sh
->dev
[sh
->pd_idx
].page
)) {
1696 /* parity is correct (on disc, not in buffer any more) */
1697 set_bit(STRIPE_INSYNC
, &sh
->state
);
1699 conf
->mddev
->resync_mismatches
+= STRIPE_SECTORS
;
1700 if (test_bit(MD_RECOVERY_CHECK
, &conf
->mddev
->recovery
))
1701 /* don't try to repair!! */
1702 set_bit(STRIPE_INSYNC
, &sh
->state
);
1704 compute_block(sh
, sh
->pd_idx
);
1709 if (!test_bit(STRIPE_INSYNC
, &sh
->state
)) {
1710 /* either failed parity check, or recovery is happening */
1712 failed_num
= sh
->pd_idx
;
1713 dev
= &sh
->dev
[failed_num
];
1714 BUG_ON(!test_bit(R5_UPTODATE
, &dev
->flags
));
1715 BUG_ON(uptodate
!= disks
);
1717 set_bit(R5_LOCKED
, &dev
->flags
);
1718 set_bit(R5_Wantwrite
, &dev
->flags
);
1719 clear_bit(STRIPE_DEGRADED
, &sh
->state
);
1721 set_bit(STRIPE_INSYNC
, &sh
->state
);
1724 if (syncing
&& locked
== 0 && test_bit(STRIPE_INSYNC
, &sh
->state
)) {
1725 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,1);
1726 clear_bit(STRIPE_SYNCING
, &sh
->state
);
1729 /* If the failed drive is just a ReadError, then we might need to progress
1730 * the repair/check process
1732 if (failed
== 1 && ! conf
->mddev
->ro
&&
1733 test_bit(R5_ReadError
, &sh
->dev
[failed_num
].flags
)
1734 && !test_bit(R5_LOCKED
, &sh
->dev
[failed_num
].flags
)
1735 && test_bit(R5_UPTODATE
, &sh
->dev
[failed_num
].flags
)
1737 dev
= &sh
->dev
[failed_num
];
1738 if (!test_bit(R5_ReWrite
, &dev
->flags
)) {
1739 set_bit(R5_Wantwrite
, &dev
->flags
);
1740 set_bit(R5_ReWrite
, &dev
->flags
);
1741 set_bit(R5_LOCKED
, &dev
->flags
);
1744 /* let's read it back */
1745 set_bit(R5_Wantread
, &dev
->flags
);
1746 set_bit(R5_LOCKED
, &dev
->flags
);
1751 if (expanded
&& test_bit(STRIPE_EXPANDING
, &sh
->state
)) {
1752 /* Need to write out all blocks after computing parity */
1753 sh
->disks
= conf
->raid_disks
;
1754 sh
->pd_idx
= stripe_to_pdidx(sh
->sector
, conf
, conf
->raid_disks
);
1755 compute_parity5(sh
, RECONSTRUCT_WRITE
);
1756 for (i
= conf
->raid_disks
; i
--;) {
1757 set_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
1759 set_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
);
1761 clear_bit(STRIPE_EXPANDING
, &sh
->state
);
1762 } else if (expanded
) {
1763 clear_bit(STRIPE_EXPAND_READY
, &sh
->state
);
1764 atomic_dec(&conf
->reshape_stripes
);
1765 wake_up(&conf
->wait_for_overlap
);
1766 md_done_sync(conf
->mddev
, STRIPE_SECTORS
, 1);
1769 if (expanding
&& locked
== 0) {
1770 /* We have read all the blocks in this stripe and now we need to
1771 * copy some of them into a target stripe for expand.
1773 clear_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
1774 for (i
=0; i
< sh
->disks
; i
++)
1775 if (i
!= sh
->pd_idx
) {
1776 int dd_idx
, pd_idx
, j
;
1777 struct stripe_head
*sh2
;
1779 sector_t bn
= compute_blocknr(sh
, i
);
1780 sector_t s
= raid5_compute_sector(bn
, conf
->raid_disks
,
1782 &dd_idx
, &pd_idx
, conf
);
1783 sh2
= get_active_stripe(conf
, s
, conf
->raid_disks
, pd_idx
, 1);
1785 /* so far only the early blocks of this stripe
1786 * have been requested. When later blocks
1787 * get requested, we will try again
1790 if(!test_bit(STRIPE_EXPANDING
, &sh2
->state
) ||
1791 test_bit(R5_Expanded
, &sh2
->dev
[dd_idx
].flags
)) {
1792 /* must have already done this block */
1793 release_stripe(sh2
);
1796 memcpy(page_address(sh2
->dev
[dd_idx
].page
),
1797 page_address(sh
->dev
[i
].page
),
1799 set_bit(R5_Expanded
, &sh2
->dev
[dd_idx
].flags
);
1800 set_bit(R5_UPTODATE
, &sh2
->dev
[dd_idx
].flags
);
1801 for (j
=0; j
<conf
->raid_disks
; j
++)
1802 if (j
!= sh2
->pd_idx
&&
1803 !test_bit(R5_Expanded
, &sh2
->dev
[j
].flags
))
1805 if (j
== conf
->raid_disks
) {
1806 set_bit(STRIPE_EXPAND_READY
, &sh2
->state
);
1807 set_bit(STRIPE_HANDLE
, &sh2
->state
);
1809 release_stripe(sh2
);
1813 spin_unlock(&sh
->lock
);
1815 while ((bi
=return_bi
)) {
1816 int bytes
= bi
->bi_size
;
1818 return_bi
= bi
->bi_next
;
1821 bi
->bi_end_io(bi
, bytes
,
1822 test_bit(BIO_UPTODATE
, &bi
->bi_flags
)
1825 for (i
=disks
; i
-- ;) {
1829 if (test_and_clear_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
))
1831 else if (test_and_clear_bit(R5_Wantread
, &sh
->dev
[i
].flags
))
1836 bi
= &sh
->dev
[i
].req
;
1840 bi
->bi_end_io
= raid5_end_write_request
;
1842 bi
->bi_end_io
= raid5_end_read_request
;
1845 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
1846 if (rdev
&& test_bit(Faulty
, &rdev
->flags
))
1849 atomic_inc(&rdev
->nr_pending
);
1853 if (syncing
|| expanding
|| expanded
)
1854 md_sync_acct(rdev
->bdev
, STRIPE_SECTORS
);
1856 bi
->bi_bdev
= rdev
->bdev
;
1857 PRINTK("for %llu schedule op %ld on disc %d\n",
1858 (unsigned long long)sh
->sector
, bi
->bi_rw
, i
);
1859 atomic_inc(&sh
->count
);
1860 bi
->bi_sector
= sh
->sector
+ rdev
->data_offset
;
1861 bi
->bi_flags
= 1 << BIO_UPTODATE
;
1863 bi
->bi_max_vecs
= 1;
1865 bi
->bi_io_vec
= &sh
->dev
[i
].vec
;
1866 bi
->bi_io_vec
[0].bv_len
= STRIPE_SIZE
;
1867 bi
->bi_io_vec
[0].bv_offset
= 0;
1868 bi
->bi_size
= STRIPE_SIZE
;
1871 test_bit(R5_ReWrite
, &sh
->dev
[i
].flags
))
1872 atomic_add(STRIPE_SECTORS
, &rdev
->corrected_errors
);
1873 generic_make_request(bi
);
1876 set_bit(STRIPE_DEGRADED
, &sh
->state
);
1877 PRINTK("skip op %ld on disc %d for sector %llu\n",
1878 bi
->bi_rw
, i
, (unsigned long long)sh
->sector
);
1879 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
1880 set_bit(STRIPE_HANDLE
, &sh
->state
);
1885 static void handle_stripe6(struct stripe_head
*sh
, struct page
*tmp_page
)
1887 raid6_conf_t
*conf
= sh
->raid_conf
;
1888 int disks
= sh
->disks
;
1889 struct bio
*return_bi
= NULL
;
1892 int syncing
, expanding
, expanded
;
1893 int locked
=0, uptodate
=0, to_read
=0, to_write
=0, failed
=0, written
=0;
1894 int non_overwrite
= 0;
1895 int failed_num
[2] = {0, 0};
1896 struct r5dev
*dev
, *pdev
, *qdev
;
1897 int pd_idx
= sh
->pd_idx
;
1898 int qd_idx
= raid6_next_disk(pd_idx
, disks
);
1899 int p_failed
, q_failed
;
1901 PRINTK("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d, qd_idx=%d\n",
1902 (unsigned long long)sh
->sector
, sh
->state
, atomic_read(&sh
->count
),
1905 spin_lock(&sh
->lock
);
1906 clear_bit(STRIPE_HANDLE
, &sh
->state
);
1907 clear_bit(STRIPE_DELAYED
, &sh
->state
);
1909 syncing
= test_bit(STRIPE_SYNCING
, &sh
->state
);
1910 expanding
= test_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
1911 expanded
= test_bit(STRIPE_EXPAND_READY
, &sh
->state
);
1912 /* Now to look around and see what can be done */
1915 for (i
=disks
; i
--; ) {
1918 clear_bit(R5_Insync
, &dev
->flags
);
1920 PRINTK("check %d: state 0x%lx read %p write %p written %p\n",
1921 i
, dev
->flags
, dev
->toread
, dev
->towrite
, dev
->written
);
1922 /* maybe we can reply to a read */
1923 if (test_bit(R5_UPTODATE
, &dev
->flags
) && dev
->toread
) {
1924 struct bio
*rbi
, *rbi2
;
1925 PRINTK("Return read for disc %d\n", i
);
1926 spin_lock_irq(&conf
->device_lock
);
1929 if (test_and_clear_bit(R5_Overlap
, &dev
->flags
))
1930 wake_up(&conf
->wait_for_overlap
);
1931 spin_unlock_irq(&conf
->device_lock
);
1932 while (rbi
&& rbi
->bi_sector
< dev
->sector
+ STRIPE_SECTORS
) {
1933 copy_data(0, rbi
, dev
->page
, dev
->sector
);
1934 rbi2
= r5_next_bio(rbi
, dev
->sector
);
1935 spin_lock_irq(&conf
->device_lock
);
1936 if (--rbi
->bi_phys_segments
== 0) {
1937 rbi
->bi_next
= return_bi
;
1940 spin_unlock_irq(&conf
->device_lock
);
1945 /* now count some things */
1946 if (test_bit(R5_LOCKED
, &dev
->flags
)) locked
++;
1947 if (test_bit(R5_UPTODATE
, &dev
->flags
)) uptodate
++;
1950 if (dev
->toread
) to_read
++;
1953 if (!test_bit(R5_OVERWRITE
, &dev
->flags
))
1956 if (dev
->written
) written
++;
1957 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
1958 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)) {
1959 /* The ReadError flag will just be confusing now */
1960 clear_bit(R5_ReadError
, &dev
->flags
);
1961 clear_bit(R5_ReWrite
, &dev
->flags
);
1963 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)
1964 || test_bit(R5_ReadError
, &dev
->flags
)) {
1966 failed_num
[failed
] = i
;
1969 set_bit(R5_Insync
, &dev
->flags
);
1972 PRINTK("locked=%d uptodate=%d to_read=%d"
1973 " to_write=%d failed=%d failed_num=%d,%d\n",
1974 locked
, uptodate
, to_read
, to_write
, failed
,
1975 failed_num
[0], failed_num
[1]);
1976 /* check if the array has lost >2 devices and, if so, some requests might
1979 if (failed
> 2 && to_read
+to_write
+written
) {
1980 for (i
=disks
; i
--; ) {
1983 if (test_bit(R5_ReadError
, &sh
->dev
[i
].flags
)) {
1986 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
1987 if (rdev
&& test_bit(In_sync
, &rdev
->flags
))
1988 /* multiple read failures in one stripe */
1989 md_error(conf
->mddev
, rdev
);
1993 spin_lock_irq(&conf
->device_lock
);
1994 /* fail all writes first */
1995 bi
= sh
->dev
[i
].towrite
;
1996 sh
->dev
[i
].towrite
= NULL
;
1997 if (bi
) { to_write
--; bitmap_end
= 1; }
1999 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
2000 wake_up(&conf
->wait_for_overlap
);
2002 while (bi
&& bi
->bi_sector
< sh
->dev
[i
].sector
+ STRIPE_SECTORS
){
2003 struct bio
*nextbi
= r5_next_bio(bi
, sh
->dev
[i
].sector
);
2004 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
2005 if (--bi
->bi_phys_segments
== 0) {
2006 md_write_end(conf
->mddev
);
2007 bi
->bi_next
= return_bi
;
2012 /* and fail all 'written' */
2013 bi
= sh
->dev
[i
].written
;
2014 sh
->dev
[i
].written
= NULL
;
2015 if (bi
) bitmap_end
= 1;
2016 while (bi
&& bi
->bi_sector
< sh
->dev
[i
].sector
+ STRIPE_SECTORS
) {
2017 struct bio
*bi2
= r5_next_bio(bi
, sh
->dev
[i
].sector
);
2018 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
2019 if (--bi
->bi_phys_segments
== 0) {
2020 md_write_end(conf
->mddev
);
2021 bi
->bi_next
= return_bi
;
2027 /* fail any reads if this device is non-operational */
2028 if (!test_bit(R5_Insync
, &sh
->dev
[i
].flags
) ||
2029 test_bit(R5_ReadError
, &sh
->dev
[i
].flags
)) {
2030 bi
= sh
->dev
[i
].toread
;
2031 sh
->dev
[i
].toread
= NULL
;
2032 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
2033 wake_up(&conf
->wait_for_overlap
);
2035 while (bi
&& bi
->bi_sector
< sh
->dev
[i
].sector
+ STRIPE_SECTORS
){
2036 struct bio
*nextbi
= r5_next_bio(bi
, sh
->dev
[i
].sector
);
2037 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
2038 if (--bi
->bi_phys_segments
== 0) {
2039 bi
->bi_next
= return_bi
;
2045 spin_unlock_irq(&conf
->device_lock
);
2047 bitmap_endwrite(conf
->mddev
->bitmap
, sh
->sector
,
2048 STRIPE_SECTORS
, 0, 0);
2051 if (failed
> 2 && syncing
) {
2052 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,0);
2053 clear_bit(STRIPE_SYNCING
, &sh
->state
);
2058 * might be able to return some write requests if the parity blocks
2059 * are safe, or on a failed drive
2061 pdev
= &sh
->dev
[pd_idx
];
2062 p_failed
= (failed
>= 1 && failed_num
[0] == pd_idx
)
2063 || (failed
>= 2 && failed_num
[1] == pd_idx
);
2064 qdev
= &sh
->dev
[qd_idx
];
2065 q_failed
= (failed
>= 1 && failed_num
[0] == qd_idx
)
2066 || (failed
>= 2 && failed_num
[1] == qd_idx
);
2069 ( p_failed
|| ((test_bit(R5_Insync
, &pdev
->flags
)
2070 && !test_bit(R5_LOCKED
, &pdev
->flags
)
2071 && test_bit(R5_UPTODATE
, &pdev
->flags
))) ) &&
2072 ( q_failed
|| ((test_bit(R5_Insync
, &qdev
->flags
)
2073 && !test_bit(R5_LOCKED
, &qdev
->flags
)
2074 && test_bit(R5_UPTODATE
, &qdev
->flags
))) ) ) {
2075 /* any written block on an uptodate or failed drive can be
2076 * returned. Note that if we 'wrote' to a failed drive,
2077 * it will be UPTODATE, but never LOCKED, so we don't need
2078 * to test 'failed' directly.
2080 for (i
=disks
; i
--; )
2081 if (sh
->dev
[i
].written
) {
2083 if (!test_bit(R5_LOCKED
, &dev
->flags
) &&
2084 test_bit(R5_UPTODATE
, &dev
->flags
) ) {
2085 /* We can return any write requests */
2087 struct bio
*wbi
, *wbi2
;
2088 PRINTK("Return write for stripe %llu disc %d\n",
2089 (unsigned long long)sh
->sector
, i
);
2090 spin_lock_irq(&conf
->device_lock
);
2092 dev
->written
= NULL
;
2093 while (wbi
&& wbi
->bi_sector
< dev
->sector
+ STRIPE_SECTORS
) {
2094 wbi2
= r5_next_bio(wbi
, dev
->sector
);
2095 if (--wbi
->bi_phys_segments
== 0) {
2096 md_write_end(conf
->mddev
);
2097 wbi
->bi_next
= return_bi
;
2102 if (dev
->towrite
== NULL
)
2104 spin_unlock_irq(&conf
->device_lock
);
2106 bitmap_endwrite(conf
->mddev
->bitmap
, sh
->sector
,
2108 !test_bit(STRIPE_DEGRADED
, &sh
->state
), 0);
2113 /* Now we might consider reading some blocks, either to check/generate
2114 * parity, or to satisfy requests
2115 * or to load a block that is being partially written.
2117 if (to_read
|| non_overwrite
|| (to_write
&& failed
) ||
2118 (syncing
&& (uptodate
< disks
)) || expanding
) {
2119 for (i
=disks
; i
--;) {
2121 if (!test_bit(R5_LOCKED
, &dev
->flags
) && !test_bit(R5_UPTODATE
, &dev
->flags
) &&
2123 (dev
->towrite
&& !test_bit(R5_OVERWRITE
, &dev
->flags
)) ||
2126 (failed
>= 1 && (sh
->dev
[failed_num
[0]].toread
|| to_write
)) ||
2127 (failed
>= 2 && (sh
->dev
[failed_num
[1]].toread
|| to_write
))
2130 /* we would like to get this block, possibly
2131 * by computing it, but we might not be able to
2133 if (uptodate
== disks
-1) {
2134 PRINTK("Computing stripe %llu block %d\n",
2135 (unsigned long long)sh
->sector
, i
);
2136 compute_block_1(sh
, i
, 0);
2138 } else if ( uptodate
== disks
-2 && failed
>= 2 ) {
2139 /* Computing 2-failure is *very* expensive; only do it if failed >= 2 */
2141 for (other
=disks
; other
--;) {
2144 if ( !test_bit(R5_UPTODATE
, &sh
->dev
[other
].flags
) )
2148 PRINTK("Computing stripe %llu blocks %d,%d\n",
2149 (unsigned long long)sh
->sector
, i
, other
);
2150 compute_block_2(sh
, i
, other
);
2152 } else if (test_bit(R5_Insync
, &dev
->flags
)) {
2153 set_bit(R5_LOCKED
, &dev
->flags
);
2154 set_bit(R5_Wantread
, &dev
->flags
);
2156 PRINTK("Reading block %d (sync=%d)\n",
2161 set_bit(STRIPE_HANDLE
, &sh
->state
);
2164 /* now to consider writing and what else, if anything should be read */
2166 int rcw
=0, must_compute
=0;
2167 for (i
=disks
; i
--;) {
2169 /* Would I have to read this buffer for reconstruct_write */
2170 if (!test_bit(R5_OVERWRITE
, &dev
->flags
)
2171 && i
!= pd_idx
&& i
!= qd_idx
2172 && (!test_bit(R5_LOCKED
, &dev
->flags
)
2174 !test_bit(R5_UPTODATE
, &dev
->flags
)) {
2175 if (test_bit(R5_Insync
, &dev
->flags
)) rcw
++;
2177 PRINTK("raid6: must_compute: disk %d flags=%#lx\n", i
, dev
->flags
);
2182 PRINTK("for sector %llu, rcw=%d, must_compute=%d\n",
2183 (unsigned long long)sh
->sector
, rcw
, must_compute
);
2184 set_bit(STRIPE_HANDLE
, &sh
->state
);
2187 /* want reconstruct write, but need to get some data */
2188 for (i
=disks
; i
--;) {
2190 if (!test_bit(R5_OVERWRITE
, &dev
->flags
)
2191 && !(failed
== 0 && (i
== pd_idx
|| i
== qd_idx
))
2192 && !test_bit(R5_LOCKED
, &dev
->flags
) && !test_bit(R5_UPTODATE
, &dev
->flags
) &&
2193 test_bit(R5_Insync
, &dev
->flags
)) {
2194 if (test_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
))
2196 PRINTK("Read_old stripe %llu block %d for Reconstruct\n",
2197 (unsigned long long)sh
->sector
, i
);
2198 set_bit(R5_LOCKED
, &dev
->flags
);
2199 set_bit(R5_Wantread
, &dev
->flags
);
2202 PRINTK("Request delayed stripe %llu block %d for Reconstruct\n",
2203 (unsigned long long)sh
->sector
, i
);
2204 set_bit(STRIPE_DELAYED
, &sh
->state
);
2205 set_bit(STRIPE_HANDLE
, &sh
->state
);
2209 /* now if nothing is locked, and if we have enough data, we can start a write request */
2210 if (locked
== 0 && rcw
== 0 &&
2211 !test_bit(STRIPE_BIT_DELAY
, &sh
->state
)) {
2212 if ( must_compute
> 0 ) {
2213 /* We have failed blocks and need to compute them */
2216 case 1: compute_block_1(sh
, failed_num
[0], 0); break;
2217 case 2: compute_block_2(sh
, failed_num
[0], failed_num
[1]); break;
2218 default: BUG(); /* This request should have been failed? */
2222 PRINTK("Computing parity for stripe %llu\n", (unsigned long long)sh
->sector
);
2223 compute_parity6(sh
, RECONSTRUCT_WRITE
);
2224 /* now every locked buffer is ready to be written */
2226 if (test_bit(R5_LOCKED
, &sh
->dev
[i
].flags
)) {
2227 PRINTK("Writing stripe %llu block %d\n",
2228 (unsigned long long)sh
->sector
, i
);
2230 set_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
);
2232 /* after a RECONSTRUCT_WRITE, the stripe MUST be in-sync */
2233 set_bit(STRIPE_INSYNC
, &sh
->state
);
2235 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2236 atomic_dec(&conf
->preread_active_stripes
);
2237 if (atomic_read(&conf
->preread_active_stripes
) < IO_THRESHOLD
)
2238 md_wakeup_thread(conf
->mddev
->thread
);
2243 /* maybe we need to check and possibly fix the parity for this stripe
2244 * Any reads will already have been scheduled, so we just see if enough data
2247 if (syncing
&& locked
== 0 && !test_bit(STRIPE_INSYNC
, &sh
->state
)) {
2248 int update_p
= 0, update_q
= 0;
2251 set_bit(STRIPE_HANDLE
, &sh
->state
);
2254 BUG_ON(uptodate
< disks
);
2255 /* Want to check and possibly repair P and Q.
2256 * However there could be one 'failed' device, in which
2257 * case we can only check one of them, possibly using the
2258 * other to generate missing data
2261 /* If !tmp_page, we cannot do the calculations,
2262 * but as we have set STRIPE_HANDLE, we will soon be called
2263 * by stripe_handle with a tmp_page - just wait until then.
2266 if (failed
== q_failed
) {
2267 /* The only possible failed device holds 'Q', so it makes
2268 * sense to check P (If anything else were failed, we would
2269 * have used P to recreate it).
2271 compute_block_1(sh
, pd_idx
, 1);
2272 if (!page_is_zero(sh
->dev
[pd_idx
].page
)) {
2273 compute_block_1(sh
,pd_idx
,0);
2277 if (!q_failed
&& failed
< 2) {
2278 /* q is not failed, and we didn't use it to generate
2279 * anything, so it makes sense to check it
2281 memcpy(page_address(tmp_page
),
2282 page_address(sh
->dev
[qd_idx
].page
),
2284 compute_parity6(sh
, UPDATE_PARITY
);
2285 if (memcmp(page_address(tmp_page
),
2286 page_address(sh
->dev
[qd_idx
].page
),
2288 clear_bit(STRIPE_INSYNC
, &sh
->state
);
2292 if (update_p
|| update_q
) {
2293 conf
->mddev
->resync_mismatches
+= STRIPE_SECTORS
;
2294 if (test_bit(MD_RECOVERY_CHECK
, &conf
->mddev
->recovery
))
2295 /* don't try to repair!! */
2296 update_p
= update_q
= 0;
2299 /* now write out any block on a failed drive,
2300 * or P or Q if they need it
2304 dev
= &sh
->dev
[failed_num
[1]];
2306 set_bit(R5_LOCKED
, &dev
->flags
);
2307 set_bit(R5_Wantwrite
, &dev
->flags
);
2310 dev
= &sh
->dev
[failed_num
[0]];
2312 set_bit(R5_LOCKED
, &dev
->flags
);
2313 set_bit(R5_Wantwrite
, &dev
->flags
);
2317 dev
= &sh
->dev
[pd_idx
];
2319 set_bit(R5_LOCKED
, &dev
->flags
);
2320 set_bit(R5_Wantwrite
, &dev
->flags
);
2323 dev
= &sh
->dev
[qd_idx
];
2325 set_bit(R5_LOCKED
, &dev
->flags
);
2326 set_bit(R5_Wantwrite
, &dev
->flags
);
2328 clear_bit(STRIPE_DEGRADED
, &sh
->state
);
2330 set_bit(STRIPE_INSYNC
, &sh
->state
);
2334 if (syncing
&& locked
== 0 && test_bit(STRIPE_INSYNC
, &sh
->state
)) {
2335 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,1);
2336 clear_bit(STRIPE_SYNCING
, &sh
->state
);
2339 /* If the failed drives are just a ReadError, then we might need
2340 * to progress the repair/check process
2342 if (failed
<= 2 && ! conf
->mddev
->ro
)
2343 for (i
=0; i
<failed
;i
++) {
2344 dev
= &sh
->dev
[failed_num
[i
]];
2345 if (test_bit(R5_ReadError
, &dev
->flags
)
2346 && !test_bit(R5_LOCKED
, &dev
->flags
)
2347 && test_bit(R5_UPTODATE
, &dev
->flags
)
2349 if (!test_bit(R5_ReWrite
, &dev
->flags
)) {
2350 set_bit(R5_Wantwrite
, &dev
->flags
);
2351 set_bit(R5_ReWrite
, &dev
->flags
);
2352 set_bit(R5_LOCKED
, &dev
->flags
);
2354 /* let's read it back */
2355 set_bit(R5_Wantread
, &dev
->flags
);
2356 set_bit(R5_LOCKED
, &dev
->flags
);
2361 if (expanded
&& test_bit(STRIPE_EXPANDING
, &sh
->state
)) {
2362 /* Need to write out all blocks after computing P&Q */
2363 sh
->disks
= conf
->raid_disks
;
2364 sh
->pd_idx
= stripe_to_pdidx(sh
->sector
, conf
,
2366 compute_parity6(sh
, RECONSTRUCT_WRITE
);
2367 for (i
= conf
->raid_disks
; i
-- ; ) {
2368 set_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
2370 set_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
);
2372 clear_bit(STRIPE_EXPANDING
, &sh
->state
);
2373 } else if (expanded
) {
2374 clear_bit(STRIPE_EXPAND_READY
, &sh
->state
);
2375 atomic_dec(&conf
->reshape_stripes
);
2376 wake_up(&conf
->wait_for_overlap
);
2377 md_done_sync(conf
->mddev
, STRIPE_SECTORS
, 1);
2380 if (expanding
&& locked
== 0) {
2381 /* We have read all the blocks in this stripe and now we need to
2382 * copy some of them into a target stripe for expand.
2384 clear_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
2385 for (i
= 0; i
< sh
->disks
; i
++)
2386 if (i
!= pd_idx
&& i
!= qd_idx
) {
2387 int dd_idx2
, pd_idx2
, j
;
2388 struct stripe_head
*sh2
;
2390 sector_t bn
= compute_blocknr(sh
, i
);
2391 sector_t s
= raid5_compute_sector(
2392 bn
, conf
->raid_disks
,
2393 conf
->raid_disks
- conf
->max_degraded
,
2394 &dd_idx2
, &pd_idx2
, conf
);
2395 sh2
= get_active_stripe(conf
, s
,
2399 /* so for only the early blocks of
2400 * this stripe have been requests.
2401 * When later blocks get requests, we
2405 if (!test_bit(STRIPE_EXPANDING
, &sh2
->state
) ||
2406 test_bit(R5_Expanded
,
2407 &sh2
->dev
[dd_idx2
].flags
)) {
2408 /* must have already done this block */
2409 release_stripe(sh2
);
2412 memcpy(page_address(sh2
->dev
[dd_idx2
].page
),
2413 page_address(sh
->dev
[i
].page
),
2415 set_bit(R5_Expanded
, &sh2
->dev
[dd_idx2
].flags
);
2416 set_bit(R5_UPTODATE
, &sh2
->dev
[dd_idx2
].flags
);
2417 for (j
= 0 ; j
< conf
->raid_disks
; j
++)
2418 if (j
!= sh2
->pd_idx
&&
2419 j
!= raid6_next_disk(sh2
->pd_idx
,
2421 !test_bit(R5_Expanded
,
2422 &sh2
->dev
[j
].flags
))
2424 if (j
== conf
->raid_disks
) {
2425 set_bit(STRIPE_EXPAND_READY
,
2427 set_bit(STRIPE_HANDLE
, &sh2
->state
);
2429 release_stripe(sh2
);
2433 spin_unlock(&sh
->lock
);
2435 while ((bi
=return_bi
)) {
2436 int bytes
= bi
->bi_size
;
2438 return_bi
= bi
->bi_next
;
2441 bi
->bi_end_io(bi
, bytes
,
2442 test_bit(BIO_UPTODATE
, &bi
->bi_flags
)
2445 for (i
=disks
; i
-- ;) {
2449 if (test_and_clear_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
))
2451 else if (test_and_clear_bit(R5_Wantread
, &sh
->dev
[i
].flags
))
2456 bi
= &sh
->dev
[i
].req
;
2460 bi
->bi_end_io
= raid5_end_write_request
;
2462 bi
->bi_end_io
= raid5_end_read_request
;
2465 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
2466 if (rdev
&& test_bit(Faulty
, &rdev
->flags
))
2469 atomic_inc(&rdev
->nr_pending
);
2473 if (syncing
|| expanding
|| expanded
)
2474 md_sync_acct(rdev
->bdev
, STRIPE_SECTORS
);
2476 bi
->bi_bdev
= rdev
->bdev
;
2477 PRINTK("for %llu schedule op %ld on disc %d\n",
2478 (unsigned long long)sh
->sector
, bi
->bi_rw
, i
);
2479 atomic_inc(&sh
->count
);
2480 bi
->bi_sector
= sh
->sector
+ rdev
->data_offset
;
2481 bi
->bi_flags
= 1 << BIO_UPTODATE
;
2483 bi
->bi_max_vecs
= 1;
2485 bi
->bi_io_vec
= &sh
->dev
[i
].vec
;
2486 bi
->bi_io_vec
[0].bv_len
= STRIPE_SIZE
;
2487 bi
->bi_io_vec
[0].bv_offset
= 0;
2488 bi
->bi_size
= STRIPE_SIZE
;
2491 test_bit(R5_ReWrite
, &sh
->dev
[i
].flags
))
2492 atomic_add(STRIPE_SECTORS
, &rdev
->corrected_errors
);
2493 generic_make_request(bi
);
2496 set_bit(STRIPE_DEGRADED
, &sh
->state
);
2497 PRINTK("skip op %ld on disc %d for sector %llu\n",
2498 bi
->bi_rw
, i
, (unsigned long long)sh
->sector
);
2499 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
2500 set_bit(STRIPE_HANDLE
, &sh
->state
);
2505 static void handle_stripe(struct stripe_head
*sh
, struct page
*tmp_page
)
2507 if (sh
->raid_conf
->level
== 6)
2508 handle_stripe6(sh
, tmp_page
);
2515 static void raid5_activate_delayed(raid5_conf_t
*conf
)
2517 if (atomic_read(&conf
->preread_active_stripes
) < IO_THRESHOLD
) {
2518 while (!list_empty(&conf
->delayed_list
)) {
2519 struct list_head
*l
= conf
->delayed_list
.next
;
2520 struct stripe_head
*sh
;
2521 sh
= list_entry(l
, struct stripe_head
, lru
);
2523 clear_bit(STRIPE_DELAYED
, &sh
->state
);
2524 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
))
2525 atomic_inc(&conf
->preread_active_stripes
);
2526 list_add_tail(&sh
->lru
, &conf
->handle_list
);
2529 blk_plug_device(conf
->mddev
->queue
);
2532 static void activate_bit_delay(raid5_conf_t
*conf
)
2534 /* device_lock is held */
2535 struct list_head head
;
2536 list_add(&head
, &conf
->bitmap_list
);
2537 list_del_init(&conf
->bitmap_list
);
2538 while (!list_empty(&head
)) {
2539 struct stripe_head
*sh
= list_entry(head
.next
, struct stripe_head
, lru
);
2540 list_del_init(&sh
->lru
);
2541 atomic_inc(&sh
->count
);
2542 __release_stripe(conf
, sh
);
2546 static void unplug_slaves(mddev_t
*mddev
)
2548 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
2552 for (i
=0; i
<mddev
->raid_disks
; i
++) {
2553 mdk_rdev_t
*rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
2554 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
) && atomic_read(&rdev
->nr_pending
)) {
2555 request_queue_t
*r_queue
= bdev_get_queue(rdev
->bdev
);
2557 atomic_inc(&rdev
->nr_pending
);
2560 if (r_queue
->unplug_fn
)
2561 r_queue
->unplug_fn(r_queue
);
2563 rdev_dec_pending(rdev
, mddev
);
2570 static void raid5_unplug_device(request_queue_t
*q
)
2572 mddev_t
*mddev
= q
->queuedata
;
2573 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
2574 unsigned long flags
;
2576 spin_lock_irqsave(&conf
->device_lock
, flags
);
2578 if (blk_remove_plug(q
)) {
2580 raid5_activate_delayed(conf
);
2582 md_wakeup_thread(mddev
->thread
);
2584 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
2586 unplug_slaves(mddev
);
2589 static int raid5_issue_flush(request_queue_t
*q
, struct gendisk
*disk
,
2590 sector_t
*error_sector
)
2592 mddev_t
*mddev
= q
->queuedata
;
2593 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
2597 for (i
=0; i
<mddev
->raid_disks
&& ret
== 0; i
++) {
2598 mdk_rdev_t
*rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
2599 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
)) {
2600 struct block_device
*bdev
= rdev
->bdev
;
2601 request_queue_t
*r_queue
= bdev_get_queue(bdev
);
2603 if (!r_queue
->issue_flush_fn
)
2606 atomic_inc(&rdev
->nr_pending
);
2608 ret
= r_queue
->issue_flush_fn(r_queue
, bdev
->bd_disk
,
2610 rdev_dec_pending(rdev
, mddev
);
2619 static int raid5_congested(void *data
, int bits
)
2621 mddev_t
*mddev
= data
;
2622 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
2624 /* No difference between reads and writes. Just check
2625 * how busy the stripe_cache is
2627 if (conf
->inactive_blocked
)
2631 if (list_empty_careful(&conf
->inactive_list
))
2637 /* We want read requests to align with chunks where possible,
2638 * but write requests don't need to.
2640 static int raid5_mergeable_bvec(request_queue_t
*q
, struct bio
*bio
, struct bio_vec
*biovec
)
2642 mddev_t
*mddev
= q
->queuedata
;
2643 sector_t sector
= bio
->bi_sector
+ get_start_sect(bio
->bi_bdev
);
2645 unsigned int chunk_sectors
= mddev
->chunk_size
>> 9;
2646 unsigned int bio_sectors
= bio
->bi_size
>> 9;
2648 if (bio_data_dir(bio
) == WRITE
)
2649 return biovec
->bv_len
; /* always allow writes to be mergeable */
2651 max
= (chunk_sectors
- ((sector
& (chunk_sectors
- 1)) + bio_sectors
)) << 9;
2652 if (max
< 0) max
= 0;
2653 if (max
<= biovec
->bv_len
&& bio_sectors
== 0)
2654 return biovec
->bv_len
;
2660 static int in_chunk_boundary(mddev_t
*mddev
, struct bio
*bio
)
2662 sector_t sector
= bio
->bi_sector
+ get_start_sect(bio
->bi_bdev
);
2663 unsigned int chunk_sectors
= mddev
->chunk_size
>> 9;
2664 unsigned int bio_sectors
= bio
->bi_size
>> 9;
2666 return chunk_sectors
>=
2667 ((sector
& (chunk_sectors
- 1)) + bio_sectors
);
2671 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
2672 * later sampled by raid5d.
2674 static void add_bio_to_retry(struct bio
*bi
,raid5_conf_t
*conf
)
2676 unsigned long flags
;
2678 spin_lock_irqsave(&conf
->device_lock
, flags
);
2680 bi
->bi_next
= conf
->retry_read_aligned_list
;
2681 conf
->retry_read_aligned_list
= bi
;
2683 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
2684 md_wakeup_thread(conf
->mddev
->thread
);
2688 static struct bio
*remove_bio_from_retry(raid5_conf_t
*conf
)
2692 bi
= conf
->retry_read_aligned
;
2694 conf
->retry_read_aligned
= NULL
;
2697 bi
= conf
->retry_read_aligned_list
;
2699 conf
->retry_read_aligned_list
= bi
->bi_next
;
2701 bi
->bi_phys_segments
= 1; /* biased count of active stripes */
2702 bi
->bi_hw_segments
= 0; /* count of processed stripes */
2710 * The "raid5_align_endio" should check if the read succeeded and if it
2711 * did, call bio_endio on the original bio (having bio_put the new bio
2713 * If the read failed..
2715 static int raid5_align_endio(struct bio
*bi
, unsigned int bytes
, int error
)
2717 struct bio
* raid_bi
= bi
->bi_private
;
2720 int uptodate
= test_bit(BIO_UPTODATE
, &bi
->bi_flags
);
2727 mddev
= raid_bi
->bi_bdev
->bd_disk
->queue
->queuedata
;
2728 conf
= mddev_to_conf(mddev
);
2729 rdev
= (void*)raid_bi
->bi_next
;
2730 raid_bi
->bi_next
= NULL
;
2732 rdev_dec_pending(rdev
, conf
->mddev
);
2734 if (!error
&& uptodate
) {
2735 bio_endio(raid_bi
, bytes
, 0);
2736 if (atomic_dec_and_test(&conf
->active_aligned_reads
))
2737 wake_up(&conf
->wait_for_stripe
);
2742 PRINTK("raid5_align_endio : io error...handing IO for a retry\n");
2744 add_bio_to_retry(raid_bi
, conf
);
2748 static int bio_fits_rdev(struct bio
*bi
)
2750 request_queue_t
*q
= bdev_get_queue(bi
->bi_bdev
);
2752 if ((bi
->bi_size
>>9) > q
->max_sectors
)
2754 blk_recount_segments(q
, bi
);
2755 if (bi
->bi_phys_segments
> q
->max_phys_segments
||
2756 bi
->bi_hw_segments
> q
->max_hw_segments
)
2759 if (q
->merge_bvec_fn
)
2760 /* it's too hard to apply the merge_bvec_fn at this stage,
2769 static int chunk_aligned_read(request_queue_t
*q
, struct bio
* raid_bio
)
2771 mddev_t
*mddev
= q
->queuedata
;
2772 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
2773 const unsigned int raid_disks
= conf
->raid_disks
;
2774 const unsigned int data_disks
= raid_disks
- conf
->max_degraded
;
2775 unsigned int dd_idx
, pd_idx
;
2776 struct bio
* align_bi
;
2779 if (!in_chunk_boundary(mddev
, raid_bio
)) {
2780 PRINTK("chunk_aligned_read : non aligned\n");
2784 * use bio_clone to make a copy of the bio
2786 align_bi
= bio_clone(raid_bio
, GFP_NOIO
);
2790 * set bi_end_io to a new function, and set bi_private to the
2793 align_bi
->bi_end_io
= raid5_align_endio
;
2794 align_bi
->bi_private
= raid_bio
;
2798 align_bi
->bi_sector
= raid5_compute_sector(raid_bio
->bi_sector
,
2806 rdev
= rcu_dereference(conf
->disks
[dd_idx
].rdev
);
2807 if (rdev
&& test_bit(In_sync
, &rdev
->flags
)) {
2808 atomic_inc(&rdev
->nr_pending
);
2810 raid_bio
->bi_next
= (void*)rdev
;
2811 align_bi
->bi_bdev
= rdev
->bdev
;
2812 align_bi
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
2813 align_bi
->bi_sector
+= rdev
->data_offset
;
2815 if (!bio_fits_rdev(align_bi
)) {
2816 /* too big in some way */
2818 rdev_dec_pending(rdev
, mddev
);
2822 spin_lock_irq(&conf
->device_lock
);
2823 wait_event_lock_irq(conf
->wait_for_stripe
,
2825 conf
->device_lock
, /* nothing */);
2826 atomic_inc(&conf
->active_aligned_reads
);
2827 spin_unlock_irq(&conf
->device_lock
);
2829 generic_make_request(align_bi
);
2839 static int make_request(request_queue_t
*q
, struct bio
* bi
)
2841 mddev_t
*mddev
= q
->queuedata
;
2842 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
2843 unsigned int dd_idx
, pd_idx
;
2844 sector_t new_sector
;
2845 sector_t logical_sector
, last_sector
;
2846 struct stripe_head
*sh
;
2847 const int rw
= bio_data_dir(bi
);
2850 if (unlikely(bio_barrier(bi
))) {
2851 bio_endio(bi
, bi
->bi_size
, -EOPNOTSUPP
);
2855 md_write_start(mddev
, bi
);
2857 disk_stat_inc(mddev
->gendisk
, ios
[rw
]);
2858 disk_stat_add(mddev
->gendisk
, sectors
[rw
], bio_sectors(bi
));
2861 mddev
->reshape_position
== MaxSector
&&
2862 chunk_aligned_read(q
,bi
))
2865 logical_sector
= bi
->bi_sector
& ~((sector_t
)STRIPE_SECTORS
-1);
2866 last_sector
= bi
->bi_sector
+ (bi
->bi_size
>>9);
2868 bi
->bi_phys_segments
= 1; /* over-loaded to count active stripes */
2870 for (;logical_sector
< last_sector
; logical_sector
+= STRIPE_SECTORS
) {
2872 int disks
, data_disks
;
2875 prepare_to_wait(&conf
->wait_for_overlap
, &w
, TASK_UNINTERRUPTIBLE
);
2876 if (likely(conf
->expand_progress
== MaxSector
))
2877 disks
= conf
->raid_disks
;
2879 /* spinlock is needed as expand_progress may be
2880 * 64bit on a 32bit platform, and so it might be
2881 * possible to see a half-updated value
2882 * Ofcourse expand_progress could change after
2883 * the lock is dropped, so once we get a reference
2884 * to the stripe that we think it is, we will have
2887 spin_lock_irq(&conf
->device_lock
);
2888 disks
= conf
->raid_disks
;
2889 if (logical_sector
>= conf
->expand_progress
)
2890 disks
= conf
->previous_raid_disks
;
2892 if (logical_sector
>= conf
->expand_lo
) {
2893 spin_unlock_irq(&conf
->device_lock
);
2898 spin_unlock_irq(&conf
->device_lock
);
2900 data_disks
= disks
- conf
->max_degraded
;
2902 new_sector
= raid5_compute_sector(logical_sector
, disks
, data_disks
,
2903 &dd_idx
, &pd_idx
, conf
);
2904 PRINTK("raid5: make_request, sector %llu logical %llu\n",
2905 (unsigned long long)new_sector
,
2906 (unsigned long long)logical_sector
);
2908 sh
= get_active_stripe(conf
, new_sector
, disks
, pd_idx
, (bi
->bi_rw
&RWA_MASK
));
2910 if (unlikely(conf
->expand_progress
!= MaxSector
)) {
2911 /* expansion might have moved on while waiting for a
2912 * stripe, so we must do the range check again.
2913 * Expansion could still move past after this
2914 * test, but as we are holding a reference to
2915 * 'sh', we know that if that happens,
2916 * STRIPE_EXPANDING will get set and the expansion
2917 * won't proceed until we finish with the stripe.
2920 spin_lock_irq(&conf
->device_lock
);
2921 if (logical_sector
< conf
->expand_progress
&&
2922 disks
== conf
->previous_raid_disks
)
2923 /* mismatch, need to try again */
2925 spin_unlock_irq(&conf
->device_lock
);
2931 /* FIXME what if we get a false positive because these
2932 * are being updated.
2934 if (logical_sector
>= mddev
->suspend_lo
&&
2935 logical_sector
< mddev
->suspend_hi
) {
2941 if (test_bit(STRIPE_EXPANDING
, &sh
->state
) ||
2942 !add_stripe_bio(sh
, bi
, dd_idx
, (bi
->bi_rw
&RW_MASK
))) {
2943 /* Stripe is busy expanding or
2944 * add failed due to overlap. Flush everything
2947 raid5_unplug_device(mddev
->queue
);
2952 finish_wait(&conf
->wait_for_overlap
, &w
);
2953 set_bit(STRIPE_HANDLE
, &sh
->state
);
2954 clear_bit(STRIPE_DELAYED
, &sh
->state
);
2957 /* cannot get stripe for read-ahead, just give-up */
2958 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
2959 finish_wait(&conf
->wait_for_overlap
, &w
);
2964 spin_lock_irq(&conf
->device_lock
);
2965 remaining
= --bi
->bi_phys_segments
;
2966 spin_unlock_irq(&conf
->device_lock
);
2967 if (remaining
== 0) {
2968 int bytes
= bi
->bi_size
;
2971 md_write_end(mddev
);
2973 bi
->bi_end_io(bi
, bytes
,
2974 test_bit(BIO_UPTODATE
, &bi
->bi_flags
)
2980 static sector_t
reshape_request(mddev_t
*mddev
, sector_t sector_nr
, int *skipped
)
2982 /* reshaping is quite different to recovery/resync so it is
2983 * handled quite separately ... here.
2985 * On each call to sync_request, we gather one chunk worth of
2986 * destination stripes and flag them as expanding.
2987 * Then we find all the source stripes and request reads.
2988 * As the reads complete, handle_stripe will copy the data
2989 * into the destination stripe and release that stripe.
2991 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
2992 struct stripe_head
*sh
;
2994 sector_t first_sector
, last_sector
;
2995 int raid_disks
= conf
->previous_raid_disks
;
2996 int data_disks
= raid_disks
- conf
->max_degraded
;
2997 int new_data_disks
= conf
->raid_disks
- conf
->max_degraded
;
3000 sector_t writepos
, safepos
, gap
;
3002 if (sector_nr
== 0 &&
3003 conf
->expand_progress
!= 0) {
3004 /* restarting in the middle, skip the initial sectors */
3005 sector_nr
= conf
->expand_progress
;
3006 sector_div(sector_nr
, new_data_disks
);
3011 /* we update the metadata when there is more than 3Meg
3012 * in the block range (that is rather arbitrary, should
3013 * probably be time based) or when the data about to be
3014 * copied would over-write the source of the data at
3015 * the front of the range.
3016 * i.e. one new_stripe forward from expand_progress new_maps
3017 * to after where expand_lo old_maps to
3019 writepos
= conf
->expand_progress
+
3020 conf
->chunk_size
/512*(new_data_disks
);
3021 sector_div(writepos
, new_data_disks
);
3022 safepos
= conf
->expand_lo
;
3023 sector_div(safepos
, data_disks
);
3024 gap
= conf
->expand_progress
- conf
->expand_lo
;
3026 if (writepos
>= safepos
||
3027 gap
> (new_data_disks
)*3000*2 /*3Meg*/) {
3028 /* Cannot proceed until we've updated the superblock... */
3029 wait_event(conf
->wait_for_overlap
,
3030 atomic_read(&conf
->reshape_stripes
)==0);
3031 mddev
->reshape_position
= conf
->expand_progress
;
3032 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3033 md_wakeup_thread(mddev
->thread
);
3034 wait_event(mddev
->sb_wait
, mddev
->flags
== 0 ||
3035 kthread_should_stop());
3036 spin_lock_irq(&conf
->device_lock
);
3037 conf
->expand_lo
= mddev
->reshape_position
;
3038 spin_unlock_irq(&conf
->device_lock
);
3039 wake_up(&conf
->wait_for_overlap
);
3042 for (i
=0; i
< conf
->chunk_size
/512; i
+= STRIPE_SECTORS
) {
3045 pd_idx
= stripe_to_pdidx(sector_nr
+i
, conf
, conf
->raid_disks
);
3046 sh
= get_active_stripe(conf
, sector_nr
+i
,
3047 conf
->raid_disks
, pd_idx
, 0);
3048 set_bit(STRIPE_EXPANDING
, &sh
->state
);
3049 atomic_inc(&conf
->reshape_stripes
);
3050 /* If any of this stripe is beyond the end of the old
3051 * array, then we need to zero those blocks
3053 for (j
=sh
->disks
; j
--;) {
3055 if (j
== sh
->pd_idx
)
3057 if (conf
->level
== 6 &&
3058 j
== raid6_next_disk(sh
->pd_idx
, sh
->disks
))
3060 s
= compute_blocknr(sh
, j
);
3061 if (s
< (mddev
->array_size
<<1)) {
3065 memset(page_address(sh
->dev
[j
].page
), 0, STRIPE_SIZE
);
3066 set_bit(R5_Expanded
, &sh
->dev
[j
].flags
);
3067 set_bit(R5_UPTODATE
, &sh
->dev
[j
].flags
);
3070 set_bit(STRIPE_EXPAND_READY
, &sh
->state
);
3071 set_bit(STRIPE_HANDLE
, &sh
->state
);
3075 spin_lock_irq(&conf
->device_lock
);
3076 conf
->expand_progress
= (sector_nr
+ i
) * new_data_disks
;
3077 spin_unlock_irq(&conf
->device_lock
);
3078 /* Ok, those stripe are ready. We can start scheduling
3079 * reads on the source stripes.
3080 * The source stripes are determined by mapping the first and last
3081 * block on the destination stripes.
3084 raid5_compute_sector(sector_nr
*(new_data_disks
),
3085 raid_disks
, data_disks
,
3086 &dd_idx
, &pd_idx
, conf
);
3088 raid5_compute_sector((sector_nr
+conf
->chunk_size
/512)
3089 *(new_data_disks
) -1,
3090 raid_disks
, data_disks
,
3091 &dd_idx
, &pd_idx
, conf
);
3092 if (last_sector
>= (mddev
->size
<<1))
3093 last_sector
= (mddev
->size
<<1)-1;
3094 while (first_sector
<= last_sector
) {
3095 pd_idx
= stripe_to_pdidx(first_sector
, conf
,
3096 conf
->previous_raid_disks
);
3097 sh
= get_active_stripe(conf
, first_sector
,
3098 conf
->previous_raid_disks
, pd_idx
, 0);
3099 set_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
3100 set_bit(STRIPE_HANDLE
, &sh
->state
);
3102 first_sector
+= STRIPE_SECTORS
;
3104 return conf
->chunk_size
>>9;
3107 /* FIXME go_faster isn't used */
3108 static inline sector_t
sync_request(mddev_t
*mddev
, sector_t sector_nr
, int *skipped
, int go_faster
)
3110 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
3111 struct stripe_head
*sh
;
3113 int raid_disks
= conf
->raid_disks
;
3114 sector_t max_sector
= mddev
->size
<< 1;
3116 int still_degraded
= 0;
3119 if (sector_nr
>= max_sector
) {
3120 /* just being told to finish up .. nothing much to do */
3121 unplug_slaves(mddev
);
3122 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)) {
3127 if (mddev
->curr_resync
< max_sector
) /* aborted */
3128 bitmap_end_sync(mddev
->bitmap
, mddev
->curr_resync
,
3130 else /* completed sync */
3132 bitmap_close_sync(mddev
->bitmap
);
3137 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
3138 return reshape_request(mddev
, sector_nr
, skipped
);
3140 /* if there is too many failed drives and we are trying
3141 * to resync, then assert that we are finished, because there is
3142 * nothing we can do.
3144 if (mddev
->degraded
>= conf
->max_degraded
&&
3145 test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
3146 sector_t rv
= (mddev
->size
<< 1) - sector_nr
;
3150 if (!bitmap_start_sync(mddev
->bitmap
, sector_nr
, &sync_blocks
, 1) &&
3151 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
) &&
3152 !conf
->fullsync
&& sync_blocks
>= STRIPE_SECTORS
) {
3153 /* we can skip this block, and probably more */
3154 sync_blocks
/= STRIPE_SECTORS
;
3156 return sync_blocks
* STRIPE_SECTORS
; /* keep things rounded to whole stripes */
3159 pd_idx
= stripe_to_pdidx(sector_nr
, conf
, raid_disks
);
3160 sh
= get_active_stripe(conf
, sector_nr
, raid_disks
, pd_idx
, 1);
3162 sh
= get_active_stripe(conf
, sector_nr
, raid_disks
, pd_idx
, 0);
3163 /* make sure we don't swamp the stripe cache if someone else
3164 * is trying to get access
3166 schedule_timeout_uninterruptible(1);
3168 /* Need to check if array will still be degraded after recovery/resync
3169 * We don't need to check the 'failed' flag as when that gets set,
3172 for (i
=0; i
<mddev
->raid_disks
; i
++)
3173 if (conf
->disks
[i
].rdev
== NULL
)
3176 bitmap_start_sync(mddev
->bitmap
, sector_nr
, &sync_blocks
, still_degraded
);
3178 spin_lock(&sh
->lock
);
3179 set_bit(STRIPE_SYNCING
, &sh
->state
);
3180 clear_bit(STRIPE_INSYNC
, &sh
->state
);
3181 spin_unlock(&sh
->lock
);
3183 handle_stripe(sh
, NULL
);
3186 return STRIPE_SECTORS
;
3189 static int retry_aligned_read(raid5_conf_t
*conf
, struct bio
*raid_bio
)
3191 /* We may not be able to submit a whole bio at once as there
3192 * may not be enough stripe_heads available.
3193 * We cannot pre-allocate enough stripe_heads as we may need
3194 * more than exist in the cache (if we allow ever large chunks).
3195 * So we do one stripe head at a time and record in
3196 * ->bi_hw_segments how many have been done.
3198 * We *know* that this entire raid_bio is in one chunk, so
3199 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
3201 struct stripe_head
*sh
;
3203 sector_t sector
, logical_sector
, last_sector
;
3208 logical_sector
= raid_bio
->bi_sector
& ~((sector_t
)STRIPE_SECTORS
-1);
3209 sector
= raid5_compute_sector( logical_sector
,
3211 conf
->raid_disks
- conf
->max_degraded
,
3215 last_sector
= raid_bio
->bi_sector
+ (raid_bio
->bi_size
>>9);
3217 for (; logical_sector
< last_sector
;
3218 logical_sector
+= STRIPE_SECTORS
,
3219 sector
+= STRIPE_SECTORS
,
3222 if (scnt
< raid_bio
->bi_hw_segments
)
3223 /* already done this stripe */
3226 sh
= get_active_stripe(conf
, sector
, conf
->raid_disks
, pd_idx
, 1);
3229 /* failed to get a stripe - must wait */
3230 raid_bio
->bi_hw_segments
= scnt
;
3231 conf
->retry_read_aligned
= raid_bio
;
3235 set_bit(R5_ReadError
, &sh
->dev
[dd_idx
].flags
);
3236 if (!add_stripe_bio(sh
, raid_bio
, dd_idx
, 0)) {
3238 raid_bio
->bi_hw_segments
= scnt
;
3239 conf
->retry_read_aligned
= raid_bio
;
3243 handle_stripe(sh
, NULL
);
3247 spin_lock_irq(&conf
->device_lock
);
3248 remaining
= --raid_bio
->bi_phys_segments
;
3249 spin_unlock_irq(&conf
->device_lock
);
3250 if (remaining
== 0) {
3251 int bytes
= raid_bio
->bi_size
;
3253 raid_bio
->bi_size
= 0;
3254 raid_bio
->bi_end_io(raid_bio
, bytes
,
3255 test_bit(BIO_UPTODATE
, &raid_bio
->bi_flags
)
3258 if (atomic_dec_and_test(&conf
->active_aligned_reads
))
3259 wake_up(&conf
->wait_for_stripe
);
3266 * This is our raid5 kernel thread.
3268 * We scan the hash table for stripes which can be handled now.
3269 * During the scan, completed stripes are saved for us by the interrupt
3270 * handler, so that they will not have to wait for our next wakeup.
3272 static void raid5d(mddev_t
*mddev
)
3274 struct stripe_head
*sh
;
3275 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3278 PRINTK("+++ raid5d active\n");
3280 md_check_recovery(mddev
);
3283 spin_lock_irq(&conf
->device_lock
);
3285 struct list_head
*first
;
3288 if (conf
->seq_flush
!= conf
->seq_write
) {
3289 int seq
= conf
->seq_flush
;
3290 spin_unlock_irq(&conf
->device_lock
);
3291 bitmap_unplug(mddev
->bitmap
);
3292 spin_lock_irq(&conf
->device_lock
);
3293 conf
->seq_write
= seq
;
3294 activate_bit_delay(conf
);
3297 while ((bio
= remove_bio_from_retry(conf
))) {
3299 spin_unlock_irq(&conf
->device_lock
);
3300 ok
= retry_aligned_read(conf
, bio
);
3301 spin_lock_irq(&conf
->device_lock
);
3307 if (list_empty(&conf
->handle_list
))
3310 first
= conf
->handle_list
.next
;
3311 sh
= list_entry(first
, struct stripe_head
, lru
);
3313 list_del_init(first
);
3314 atomic_inc(&sh
->count
);
3315 BUG_ON(atomic_read(&sh
->count
)!= 1);
3316 spin_unlock_irq(&conf
->device_lock
);
3319 handle_stripe(sh
, conf
->spare_page
);
3322 spin_lock_irq(&conf
->device_lock
);
3324 PRINTK("%d stripes handled\n", handled
);
3326 spin_unlock_irq(&conf
->device_lock
);
3328 unplug_slaves(mddev
);
3330 PRINTK("--- raid5d inactive\n");
3334 raid5_show_stripe_cache_size(mddev_t
*mddev
, char *page
)
3336 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3338 return sprintf(page
, "%d\n", conf
->max_nr_stripes
);
3344 raid5_store_stripe_cache_size(mddev_t
*mddev
, const char *page
, size_t len
)
3346 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3349 if (len
>= PAGE_SIZE
)
3354 new = simple_strtoul(page
, &end
, 10);
3355 if (!*page
|| (*end
&& *end
!= '\n') )
3357 if (new <= 16 || new > 32768)
3359 while (new < conf
->max_nr_stripes
) {
3360 if (drop_one_stripe(conf
))
3361 conf
->max_nr_stripes
--;
3365 md_allow_write(mddev
);
3366 while (new > conf
->max_nr_stripes
) {
3367 if (grow_one_stripe(conf
))
3368 conf
->max_nr_stripes
++;
3374 static struct md_sysfs_entry
3375 raid5_stripecache_size
= __ATTR(stripe_cache_size
, S_IRUGO
| S_IWUSR
,
3376 raid5_show_stripe_cache_size
,
3377 raid5_store_stripe_cache_size
);
3380 stripe_cache_active_show(mddev_t
*mddev
, char *page
)
3382 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3384 return sprintf(page
, "%d\n", atomic_read(&conf
->active_stripes
));
3389 static struct md_sysfs_entry
3390 raid5_stripecache_active
= __ATTR_RO(stripe_cache_active
);
3392 static struct attribute
*raid5_attrs
[] = {
3393 &raid5_stripecache_size
.attr
,
3394 &raid5_stripecache_active
.attr
,
3397 static struct attribute_group raid5_attrs_group
= {
3399 .attrs
= raid5_attrs
,
3402 static int run(mddev_t
*mddev
)
3405 int raid_disk
, memory
;
3407 struct disk_info
*disk
;
3408 struct list_head
*tmp
;
3409 int working_disks
= 0;
3411 if (mddev
->level
!= 5 && mddev
->level
!= 4 && mddev
->level
!= 6) {
3412 printk(KERN_ERR
"raid5: %s: raid level not set to 4/5/6 (%d)\n",
3413 mdname(mddev
), mddev
->level
);
3417 if (mddev
->reshape_position
!= MaxSector
) {
3418 /* Check that we can continue the reshape.
3419 * Currently only disks can change, it must
3420 * increase, and we must be past the point where
3421 * a stripe over-writes itself
3423 sector_t here_new
, here_old
;
3425 int max_degraded
= (mddev
->level
== 5 ? 1 : 2);
3427 if (mddev
->new_level
!= mddev
->level
||
3428 mddev
->new_layout
!= mddev
->layout
||
3429 mddev
->new_chunk
!= mddev
->chunk_size
) {
3430 printk(KERN_ERR
"raid5: %s: unsupported reshape "
3431 "required - aborting.\n",
3435 if (mddev
->delta_disks
<= 0) {
3436 printk(KERN_ERR
"raid5: %s: unsupported reshape "
3437 "(reduce disks) required - aborting.\n",
3441 old_disks
= mddev
->raid_disks
- mddev
->delta_disks
;
3442 /* reshape_position must be on a new-stripe boundary, and one
3443 * further up in new geometry must map after here in old
3446 here_new
= mddev
->reshape_position
;
3447 if (sector_div(here_new
, (mddev
->chunk_size
>>9)*
3448 (mddev
->raid_disks
- max_degraded
))) {
3449 printk(KERN_ERR
"raid5: reshape_position not "
3450 "on a stripe boundary\n");
3453 /* here_new is the stripe we will write to */
3454 here_old
= mddev
->reshape_position
;
3455 sector_div(here_old
, (mddev
->chunk_size
>>9)*
3456 (old_disks
-max_degraded
));
3457 /* here_old is the first stripe that we might need to read
3459 if (here_new
>= here_old
) {
3460 /* Reading from the same stripe as writing to - bad */
3461 printk(KERN_ERR
"raid5: reshape_position too early for "
3462 "auto-recovery - aborting.\n");
3465 printk(KERN_INFO
"raid5: reshape will continue\n");
3466 /* OK, we should be able to continue; */
3470 mddev
->private = kzalloc(sizeof (raid5_conf_t
), GFP_KERNEL
);
3471 if ((conf
= mddev
->private) == NULL
)
3473 if (mddev
->reshape_position
== MaxSector
) {
3474 conf
->previous_raid_disks
= conf
->raid_disks
= mddev
->raid_disks
;
3476 conf
->raid_disks
= mddev
->raid_disks
;
3477 conf
->previous_raid_disks
= mddev
->raid_disks
- mddev
->delta_disks
;
3480 conf
->disks
= kzalloc(conf
->raid_disks
* sizeof(struct disk_info
),
3485 conf
->mddev
= mddev
;
3487 if ((conf
->stripe_hashtbl
= kzalloc(PAGE_SIZE
, GFP_KERNEL
)) == NULL
)
3490 if (mddev
->level
== 6) {
3491 conf
->spare_page
= alloc_page(GFP_KERNEL
);
3492 if (!conf
->spare_page
)
3495 spin_lock_init(&conf
->device_lock
);
3496 init_waitqueue_head(&conf
->wait_for_stripe
);
3497 init_waitqueue_head(&conf
->wait_for_overlap
);
3498 INIT_LIST_HEAD(&conf
->handle_list
);
3499 INIT_LIST_HEAD(&conf
->delayed_list
);
3500 INIT_LIST_HEAD(&conf
->bitmap_list
);
3501 INIT_LIST_HEAD(&conf
->inactive_list
);
3502 atomic_set(&conf
->active_stripes
, 0);
3503 atomic_set(&conf
->preread_active_stripes
, 0);
3504 atomic_set(&conf
->active_aligned_reads
, 0);
3506 PRINTK("raid5: run(%s) called.\n", mdname(mddev
));
3508 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3509 raid_disk
= rdev
->raid_disk
;
3510 if (raid_disk
>= conf
->raid_disks
3513 disk
= conf
->disks
+ raid_disk
;
3517 if (test_bit(In_sync
, &rdev
->flags
)) {
3518 char b
[BDEVNAME_SIZE
];
3519 printk(KERN_INFO
"raid5: device %s operational as raid"
3520 " disk %d\n", bdevname(rdev
->bdev
,b
),
3527 * 0 for a fully functional array, 1 or 2 for a degraded array.
3529 mddev
->degraded
= conf
->raid_disks
- working_disks
;
3530 conf
->mddev
= mddev
;
3531 conf
->chunk_size
= mddev
->chunk_size
;
3532 conf
->level
= mddev
->level
;
3533 if (conf
->level
== 6)
3534 conf
->max_degraded
= 2;
3536 conf
->max_degraded
= 1;
3537 conf
->algorithm
= mddev
->layout
;
3538 conf
->max_nr_stripes
= NR_STRIPES
;
3539 conf
->expand_progress
= mddev
->reshape_position
;
3541 /* device size must be a multiple of chunk size */
3542 mddev
->size
&= ~(mddev
->chunk_size
/1024 -1);
3543 mddev
->resync_max_sectors
= mddev
->size
<< 1;
3545 if (conf
->level
== 6 && conf
->raid_disks
< 4) {
3546 printk(KERN_ERR
"raid6: not enough configured devices for %s (%d, minimum 4)\n",
3547 mdname(mddev
), conf
->raid_disks
);
3550 if (!conf
->chunk_size
|| conf
->chunk_size
% 4) {
3551 printk(KERN_ERR
"raid5: invalid chunk size %d for %s\n",
3552 conf
->chunk_size
, mdname(mddev
));
3555 if (conf
->algorithm
> ALGORITHM_RIGHT_SYMMETRIC
) {
3557 "raid5: unsupported parity algorithm %d for %s\n",
3558 conf
->algorithm
, mdname(mddev
));
3561 if (mddev
->degraded
> conf
->max_degraded
) {
3562 printk(KERN_ERR
"raid5: not enough operational devices for %s"
3563 " (%d/%d failed)\n",
3564 mdname(mddev
), mddev
->degraded
, conf
->raid_disks
);
3568 if (mddev
->degraded
> 0 &&
3569 mddev
->recovery_cp
!= MaxSector
) {
3570 if (mddev
->ok_start_degraded
)
3572 "raid5: starting dirty degraded array: %s"
3573 "- data corruption possible.\n",
3577 "raid5: cannot start dirty degraded array for %s\n",
3584 mddev
->thread
= md_register_thread(raid5d
, mddev
, "%s_raid5");
3585 if (!mddev
->thread
) {
3587 "raid5: couldn't allocate thread for %s\n",
3592 memory
= conf
->max_nr_stripes
* (sizeof(struct stripe_head
) +
3593 conf
->raid_disks
* ((sizeof(struct bio
) + PAGE_SIZE
))) / 1024;
3594 if (grow_stripes(conf
, conf
->max_nr_stripes
)) {
3596 "raid5: couldn't allocate %dkB for buffers\n", memory
);
3597 shrink_stripes(conf
);
3598 md_unregister_thread(mddev
->thread
);
3601 printk(KERN_INFO
"raid5: allocated %dkB for %s\n",
3602 memory
, mdname(mddev
));
3604 if (mddev
->degraded
== 0)
3605 printk("raid5: raid level %d set %s active with %d out of %d"
3606 " devices, algorithm %d\n", conf
->level
, mdname(mddev
),
3607 mddev
->raid_disks
-mddev
->degraded
, mddev
->raid_disks
,
3610 printk(KERN_ALERT
"raid5: raid level %d set %s active with %d"
3611 " out of %d devices, algorithm %d\n", conf
->level
,
3612 mdname(mddev
), mddev
->raid_disks
- mddev
->degraded
,
3613 mddev
->raid_disks
, conf
->algorithm
);
3615 print_raid5_conf(conf
);
3617 if (conf
->expand_progress
!= MaxSector
) {
3618 printk("...ok start reshape thread\n");
3619 conf
->expand_lo
= conf
->expand_progress
;
3620 atomic_set(&conf
->reshape_stripes
, 0);
3621 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
3622 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
3623 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
3624 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
3625 mddev
->sync_thread
= md_register_thread(md_do_sync
, mddev
,
3629 /* read-ahead size must cover two whole stripes, which is
3630 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
3633 int data_disks
= conf
->previous_raid_disks
- conf
->max_degraded
;
3634 int stripe
= data_disks
*
3635 (mddev
->chunk_size
/ PAGE_SIZE
);
3636 if (mddev
->queue
->backing_dev_info
.ra_pages
< 2 * stripe
)
3637 mddev
->queue
->backing_dev_info
.ra_pages
= 2 * stripe
;
3640 /* Ok, everything is just fine now */
3641 if (sysfs_create_group(&mddev
->kobj
, &raid5_attrs_group
))
3643 "raid5: failed to create sysfs attributes for %s\n",
3646 mddev
->queue
->unplug_fn
= raid5_unplug_device
;
3647 mddev
->queue
->issue_flush_fn
= raid5_issue_flush
;
3648 mddev
->queue
->backing_dev_info
.congested_data
= mddev
;
3649 mddev
->queue
->backing_dev_info
.congested_fn
= raid5_congested
;
3651 mddev
->array_size
= mddev
->size
* (conf
->previous_raid_disks
-
3652 conf
->max_degraded
);
3654 blk_queue_merge_bvec(mddev
->queue
, raid5_mergeable_bvec
);
3659 print_raid5_conf(conf
);
3660 safe_put_page(conf
->spare_page
);
3662 kfree(conf
->stripe_hashtbl
);
3665 mddev
->private = NULL
;
3666 printk(KERN_ALERT
"raid5: failed to run raid set %s\n", mdname(mddev
));
3672 static int stop(mddev_t
*mddev
)
3674 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
3676 md_unregister_thread(mddev
->thread
);
3677 mddev
->thread
= NULL
;
3678 shrink_stripes(conf
);
3679 kfree(conf
->stripe_hashtbl
);
3680 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
3681 blk_sync_queue(mddev
->queue
); /* the unplug fn references 'conf'*/
3682 sysfs_remove_group(&mddev
->kobj
, &raid5_attrs_group
);
3685 mddev
->private = NULL
;
3690 static void print_sh (struct seq_file
*seq
, struct stripe_head
*sh
)
3694 seq_printf(seq
, "sh %llu, pd_idx %d, state %ld.\n",
3695 (unsigned long long)sh
->sector
, sh
->pd_idx
, sh
->state
);
3696 seq_printf(seq
, "sh %llu, count %d.\n",
3697 (unsigned long long)sh
->sector
, atomic_read(&sh
->count
));
3698 seq_printf(seq
, "sh %llu, ", (unsigned long long)sh
->sector
);
3699 for (i
= 0; i
< sh
->disks
; i
++) {
3700 seq_printf(seq
, "(cache%d: %p %ld) ",
3701 i
, sh
->dev
[i
].page
, sh
->dev
[i
].flags
);
3703 seq_printf(seq
, "\n");
3706 static void printall (struct seq_file
*seq
, raid5_conf_t
*conf
)
3708 struct stripe_head
*sh
;
3709 struct hlist_node
*hn
;
3712 spin_lock_irq(&conf
->device_lock
);
3713 for (i
= 0; i
< NR_HASH
; i
++) {
3714 hlist_for_each_entry(sh
, hn
, &conf
->stripe_hashtbl
[i
], hash
) {
3715 if (sh
->raid_conf
!= conf
)
3720 spin_unlock_irq(&conf
->device_lock
);
3724 static void status (struct seq_file
*seq
, mddev_t
*mddev
)
3726 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
3729 seq_printf (seq
, " level %d, %dk chunk, algorithm %d", mddev
->level
, mddev
->chunk_size
>> 10, mddev
->layout
);
3730 seq_printf (seq
, " [%d/%d] [", conf
->raid_disks
, conf
->raid_disks
- mddev
->degraded
);
3731 for (i
= 0; i
< conf
->raid_disks
; i
++)
3732 seq_printf (seq
, "%s",
3733 conf
->disks
[i
].rdev
&&
3734 test_bit(In_sync
, &conf
->disks
[i
].rdev
->flags
) ? "U" : "_");
3735 seq_printf (seq
, "]");
3737 seq_printf (seq
, "\n");
3738 printall(seq
, conf
);
3742 static void print_raid5_conf (raid5_conf_t
*conf
)
3745 struct disk_info
*tmp
;
3747 printk("RAID5 conf printout:\n");
3749 printk("(conf==NULL)\n");
3752 printk(" --- rd:%d wd:%d\n", conf
->raid_disks
,
3753 conf
->raid_disks
- conf
->mddev
->degraded
);
3755 for (i
= 0; i
< conf
->raid_disks
; i
++) {
3756 char b
[BDEVNAME_SIZE
];
3757 tmp
= conf
->disks
+ i
;
3759 printk(" disk %d, o:%d, dev:%s\n",
3760 i
, !test_bit(Faulty
, &tmp
->rdev
->flags
),
3761 bdevname(tmp
->rdev
->bdev
,b
));
3765 static int raid5_spare_active(mddev_t
*mddev
)
3768 raid5_conf_t
*conf
= mddev
->private;
3769 struct disk_info
*tmp
;
3771 for (i
= 0; i
< conf
->raid_disks
; i
++) {
3772 tmp
= conf
->disks
+ i
;
3774 && !test_bit(Faulty
, &tmp
->rdev
->flags
)
3775 && !test_and_set_bit(In_sync
, &tmp
->rdev
->flags
)) {
3776 unsigned long flags
;
3777 spin_lock_irqsave(&conf
->device_lock
, flags
);
3779 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
3782 print_raid5_conf(conf
);
3786 static int raid5_remove_disk(mddev_t
*mddev
, int number
)
3788 raid5_conf_t
*conf
= mddev
->private;
3791 struct disk_info
*p
= conf
->disks
+ number
;
3793 print_raid5_conf(conf
);
3796 if (test_bit(In_sync
, &rdev
->flags
) ||
3797 atomic_read(&rdev
->nr_pending
)) {
3803 if (atomic_read(&rdev
->nr_pending
)) {
3804 /* lost the race, try later */
3811 print_raid5_conf(conf
);
3815 static int raid5_add_disk(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
3817 raid5_conf_t
*conf
= mddev
->private;
3820 struct disk_info
*p
;
3822 if (mddev
->degraded
> conf
->max_degraded
)
3823 /* no point adding a device */
3827 * find the disk ... but prefer rdev->saved_raid_disk
3830 if (rdev
->saved_raid_disk
>= 0 &&
3831 conf
->disks
[rdev
->saved_raid_disk
].rdev
== NULL
)
3832 disk
= rdev
->saved_raid_disk
;
3835 for ( ; disk
< conf
->raid_disks
; disk
++)
3836 if ((p
=conf
->disks
+ disk
)->rdev
== NULL
) {
3837 clear_bit(In_sync
, &rdev
->flags
);
3838 rdev
->raid_disk
= disk
;
3840 if (rdev
->saved_raid_disk
!= disk
)
3842 rcu_assign_pointer(p
->rdev
, rdev
);
3845 print_raid5_conf(conf
);
3849 static int raid5_resize(mddev_t
*mddev
, sector_t sectors
)
3851 /* no resync is happening, and there is enough space
3852 * on all devices, so we can resize.
3853 * We need to make sure resync covers any new space.
3854 * If the array is shrinking we should possibly wait until
3855 * any io in the removed space completes, but it hardly seems
3858 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3860 sectors
&= ~((sector_t
)mddev
->chunk_size
/512 - 1);
3861 mddev
->array_size
= (sectors
* (mddev
->raid_disks
-conf
->max_degraded
))>>1;
3862 set_capacity(mddev
->gendisk
, mddev
->array_size
<< 1);
3864 if (sectors
/2 > mddev
->size
&& mddev
->recovery_cp
== MaxSector
) {
3865 mddev
->recovery_cp
= mddev
->size
<< 1;
3866 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3868 mddev
->size
= sectors
/2;
3869 mddev
->resync_max_sectors
= sectors
;
3873 #ifdef CONFIG_MD_RAID5_RESHAPE
3874 static int raid5_check_reshape(mddev_t
*mddev
)
3876 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3879 if (mddev
->delta_disks
< 0 ||
3880 mddev
->new_level
!= mddev
->level
)
3881 return -EINVAL
; /* Cannot shrink array or change level yet */
3882 if (mddev
->delta_disks
== 0)
3883 return 0; /* nothing to do */
3885 /* Can only proceed if there are plenty of stripe_heads.
3886 * We need a minimum of one full stripe,, and for sensible progress
3887 * it is best to have about 4 times that.
3888 * If we require 4 times, then the default 256 4K stripe_heads will
3889 * allow for chunk sizes up to 256K, which is probably OK.
3890 * If the chunk size is greater, user-space should request more
3891 * stripe_heads first.
3893 if ((mddev
->chunk_size
/ STRIPE_SIZE
) * 4 > conf
->max_nr_stripes
||
3894 (mddev
->new_chunk
/ STRIPE_SIZE
) * 4 > conf
->max_nr_stripes
) {
3895 printk(KERN_WARNING
"raid5: reshape: not enough stripes. Needed %lu\n",
3896 (mddev
->chunk_size
/ STRIPE_SIZE
)*4);
3900 err
= resize_stripes(conf
, conf
->raid_disks
+ mddev
->delta_disks
);
3904 if (mddev
->degraded
> conf
->max_degraded
)
3906 /* looks like we might be able to manage this */
3910 static int raid5_start_reshape(mddev_t
*mddev
)
3912 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3914 struct list_head
*rtmp
;
3916 int added_devices
= 0;
3917 unsigned long flags
;
3919 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3922 ITERATE_RDEV(mddev
, rdev
, rtmp
)
3923 if (rdev
->raid_disk
< 0 &&
3924 !test_bit(Faulty
, &rdev
->flags
))
3927 if (spares
- mddev
->degraded
< mddev
->delta_disks
- conf
->max_degraded
)
3928 /* Not enough devices even to make a degraded array
3933 atomic_set(&conf
->reshape_stripes
, 0);
3934 spin_lock_irq(&conf
->device_lock
);
3935 conf
->previous_raid_disks
= conf
->raid_disks
;
3936 conf
->raid_disks
+= mddev
->delta_disks
;
3937 conf
->expand_progress
= 0;
3938 conf
->expand_lo
= 0;
3939 spin_unlock_irq(&conf
->device_lock
);
3941 /* Add some new drives, as many as will fit.
3942 * We know there are enough to make the newly sized array work.
3944 ITERATE_RDEV(mddev
, rdev
, rtmp
)
3945 if (rdev
->raid_disk
< 0 &&
3946 !test_bit(Faulty
, &rdev
->flags
)) {
3947 if (raid5_add_disk(mddev
, rdev
)) {
3949 set_bit(In_sync
, &rdev
->flags
);
3951 rdev
->recovery_offset
= 0;
3952 sprintf(nm
, "rd%d", rdev
->raid_disk
);
3953 if (sysfs_create_link(&mddev
->kobj
,
3956 "raid5: failed to create "
3957 " link %s for %s\n",
3963 spin_lock_irqsave(&conf
->device_lock
, flags
);
3964 mddev
->degraded
= (conf
->raid_disks
- conf
->previous_raid_disks
) - added_devices
;
3965 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
3966 mddev
->raid_disks
= conf
->raid_disks
;
3967 mddev
->reshape_position
= 0;
3968 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3970 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
3971 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
3972 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
3973 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
3974 mddev
->sync_thread
= md_register_thread(md_do_sync
, mddev
,
3976 if (!mddev
->sync_thread
) {
3977 mddev
->recovery
= 0;
3978 spin_lock_irq(&conf
->device_lock
);
3979 mddev
->raid_disks
= conf
->raid_disks
= conf
->previous_raid_disks
;
3980 conf
->expand_progress
= MaxSector
;
3981 spin_unlock_irq(&conf
->device_lock
);
3984 md_wakeup_thread(mddev
->sync_thread
);
3985 md_new_event(mddev
);
3990 static void end_reshape(raid5_conf_t
*conf
)
3992 struct block_device
*bdev
;
3994 if (!test_bit(MD_RECOVERY_INTR
, &conf
->mddev
->recovery
)) {
3995 conf
->mddev
->array_size
= conf
->mddev
->size
*
3996 (conf
->raid_disks
- conf
->max_degraded
);
3997 set_capacity(conf
->mddev
->gendisk
, conf
->mddev
->array_size
<< 1);
3998 conf
->mddev
->changed
= 1;
4000 bdev
= bdget_disk(conf
->mddev
->gendisk
, 0);
4002 mutex_lock(&bdev
->bd_inode
->i_mutex
);
4003 i_size_write(bdev
->bd_inode
, (loff_t
)conf
->mddev
->array_size
<< 10);
4004 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
4007 spin_lock_irq(&conf
->device_lock
);
4008 conf
->expand_progress
= MaxSector
;
4009 spin_unlock_irq(&conf
->device_lock
);
4010 conf
->mddev
->reshape_position
= MaxSector
;
4012 /* read-ahead size must cover two whole stripes, which is
4013 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
4016 int data_disks
= conf
->previous_raid_disks
- conf
->max_degraded
;
4017 int stripe
= data_disks
*
4018 (conf
->mddev
->chunk_size
/ PAGE_SIZE
);
4019 if (conf
->mddev
->queue
->backing_dev_info
.ra_pages
< 2 * stripe
)
4020 conf
->mddev
->queue
->backing_dev_info
.ra_pages
= 2 * stripe
;
4025 static void raid5_quiesce(mddev_t
*mddev
, int state
)
4027 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4030 case 2: /* resume for a suspend */
4031 wake_up(&conf
->wait_for_overlap
);
4034 case 1: /* stop all writes */
4035 spin_lock_irq(&conf
->device_lock
);
4037 wait_event_lock_irq(conf
->wait_for_stripe
,
4038 atomic_read(&conf
->active_stripes
) == 0 &&
4039 atomic_read(&conf
->active_aligned_reads
) == 0,
4040 conf
->device_lock
, /* nothing */);
4041 spin_unlock_irq(&conf
->device_lock
);
4044 case 0: /* re-enable writes */
4045 spin_lock_irq(&conf
->device_lock
);
4047 wake_up(&conf
->wait_for_stripe
);
4048 wake_up(&conf
->wait_for_overlap
);
4049 spin_unlock_irq(&conf
->device_lock
);
4054 static struct mdk_personality raid6_personality
=
4058 .owner
= THIS_MODULE
,
4059 .make_request
= make_request
,
4063 .error_handler
= error
,
4064 .hot_add_disk
= raid5_add_disk
,
4065 .hot_remove_disk
= raid5_remove_disk
,
4066 .spare_active
= raid5_spare_active
,
4067 .sync_request
= sync_request
,
4068 .resize
= raid5_resize
,
4069 #ifdef CONFIG_MD_RAID5_RESHAPE
4070 .check_reshape
= raid5_check_reshape
,
4071 .start_reshape
= raid5_start_reshape
,
4073 .quiesce
= raid5_quiesce
,
4075 static struct mdk_personality raid5_personality
=
4079 .owner
= THIS_MODULE
,
4080 .make_request
= make_request
,
4084 .error_handler
= error
,
4085 .hot_add_disk
= raid5_add_disk
,
4086 .hot_remove_disk
= raid5_remove_disk
,
4087 .spare_active
= raid5_spare_active
,
4088 .sync_request
= sync_request
,
4089 .resize
= raid5_resize
,
4090 #ifdef CONFIG_MD_RAID5_RESHAPE
4091 .check_reshape
= raid5_check_reshape
,
4092 .start_reshape
= raid5_start_reshape
,
4094 .quiesce
= raid5_quiesce
,
4097 static struct mdk_personality raid4_personality
=
4101 .owner
= THIS_MODULE
,
4102 .make_request
= make_request
,
4106 .error_handler
= error
,
4107 .hot_add_disk
= raid5_add_disk
,
4108 .hot_remove_disk
= raid5_remove_disk
,
4109 .spare_active
= raid5_spare_active
,
4110 .sync_request
= sync_request
,
4111 .resize
= raid5_resize
,
4112 #ifdef CONFIG_MD_RAID5_RESHAPE
4113 .check_reshape
= raid5_check_reshape
,
4114 .start_reshape
= raid5_start_reshape
,
4116 .quiesce
= raid5_quiesce
,
4119 static int __init
raid5_init(void)
4123 e
= raid6_select_algo();
4126 register_md_personality(&raid6_personality
);
4127 register_md_personality(&raid5_personality
);
4128 register_md_personality(&raid4_personality
);
4132 static void raid5_exit(void)
4134 unregister_md_personality(&raid6_personality
);
4135 unregister_md_personality(&raid5_personality
);
4136 unregister_md_personality(&raid4_personality
);
4139 module_init(raid5_init
);
4140 module_exit(raid5_exit
);
4141 MODULE_LICENSE("GPL");
4142 MODULE_ALIAS("md-personality-4"); /* RAID5 */
4143 MODULE_ALIAS("md-raid5");
4144 MODULE_ALIAS("md-raid4");
4145 MODULE_ALIAS("md-level-5");
4146 MODULE_ALIAS("md-level-4");
4147 MODULE_ALIAS("md-personality-8"); /* RAID6 */
4148 MODULE_ALIAS("md-raid6");
4149 MODULE_ALIAS("md-level-6");
4151 /* This used to be two separate modules, they were: */
4152 MODULE_ALIAS("raid5");
4153 MODULE_ALIAS("raid6");