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>
55 #include <linux/async_tx.h>
61 #define NR_STRIPES 256
62 #define STRIPE_SIZE PAGE_SIZE
63 #define STRIPE_SHIFT (PAGE_SHIFT - 9)
64 #define STRIPE_SECTORS (STRIPE_SIZE>>9)
65 #define IO_THRESHOLD 1
66 #define BYPASS_THRESHOLD 1
67 #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
68 #define HASH_MASK (NR_HASH - 1)
70 #define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK]))
72 /* bio's attached to a stripe+device for I/O are linked together in bi_sector
73 * order without overlap. There may be several bio's per stripe+device, and
74 * a bio could span several devices.
75 * When walking this list for a particular stripe+device, we must never proceed
76 * beyond a bio that extends past this device, as the next bio might no longer
78 * This macro is used to determine the 'next' bio in the list, given the sector
79 * of the current stripe+device
81 #define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL)
83 * The following can be used to debug the driver
85 #define RAID5_PARANOIA 1
86 #if RAID5_PARANOIA && defined(CONFIG_SMP)
87 # define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock)
89 # define CHECK_DEVLOCK()
97 #define printk_rl(args...) ((void) (printk_ratelimit() && printk(args)))
99 #if !RAID6_USE_EMPTY_ZERO_PAGE
100 /* In .bss so it's zeroed */
101 const char raid6_empty_zero_page
[PAGE_SIZE
] __attribute__((aligned(256)));
105 * We maintain a biased count of active stripes in the bottom 16 bits of
106 * bi_phys_segments, and a count of processed stripes in the upper 16 bits
108 static inline int raid5_bi_phys_segments(struct bio
*bio
)
110 return bio
->bi_phys_segments
& 0xffff;
113 static inline int raid5_bi_hw_segments(struct bio
*bio
)
115 return (bio
->bi_phys_segments
>> 16) & 0xffff;
118 static inline int raid5_dec_bi_phys_segments(struct bio
*bio
)
120 --bio
->bi_phys_segments
;
121 return raid5_bi_phys_segments(bio
);
124 static inline int raid5_dec_bi_hw_segments(struct bio
*bio
)
126 unsigned short val
= raid5_bi_hw_segments(bio
);
129 bio
->bi_phys_segments
= (val
<< 16) | raid5_bi_phys_segments(bio
);
133 static inline void raid5_set_bi_hw_segments(struct bio
*bio
, unsigned int cnt
)
135 bio
->bi_phys_segments
= raid5_bi_phys_segments(bio
) || (cnt
<< 16);
138 static inline int raid6_next_disk(int disk
, int raid_disks
)
141 return (disk
< raid_disks
) ? disk
: 0;
144 static void return_io(struct bio
*return_bi
)
146 struct bio
*bi
= return_bi
;
149 return_bi
= bi
->bi_next
;
157 static void print_raid5_conf (raid5_conf_t
*conf
);
159 static int stripe_operations_active(struct stripe_head
*sh
)
161 return sh
->check_state
|| sh
->reconstruct_state
||
162 test_bit(STRIPE_BIOFILL_RUN
, &sh
->state
) ||
163 test_bit(STRIPE_COMPUTE_RUN
, &sh
->state
);
166 static void __release_stripe(raid5_conf_t
*conf
, struct stripe_head
*sh
)
168 if (atomic_dec_and_test(&sh
->count
)) {
169 BUG_ON(!list_empty(&sh
->lru
));
170 BUG_ON(atomic_read(&conf
->active_stripes
)==0);
171 if (test_bit(STRIPE_HANDLE
, &sh
->state
)) {
172 if (test_bit(STRIPE_DELAYED
, &sh
->state
)) {
173 list_add_tail(&sh
->lru
, &conf
->delayed_list
);
174 blk_plug_device(conf
->mddev
->queue
);
175 } else if (test_bit(STRIPE_BIT_DELAY
, &sh
->state
) &&
176 sh
->bm_seq
- conf
->seq_write
> 0) {
177 list_add_tail(&sh
->lru
, &conf
->bitmap_list
);
178 blk_plug_device(conf
->mddev
->queue
);
180 clear_bit(STRIPE_BIT_DELAY
, &sh
->state
);
181 list_add_tail(&sh
->lru
, &conf
->handle_list
);
183 md_wakeup_thread(conf
->mddev
->thread
);
185 BUG_ON(stripe_operations_active(sh
));
186 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
187 atomic_dec(&conf
->preread_active_stripes
);
188 if (atomic_read(&conf
->preread_active_stripes
) < IO_THRESHOLD
)
189 md_wakeup_thread(conf
->mddev
->thread
);
191 atomic_dec(&conf
->active_stripes
);
192 if (!test_bit(STRIPE_EXPANDING
, &sh
->state
)) {
193 list_add_tail(&sh
->lru
, &conf
->inactive_list
);
194 wake_up(&conf
->wait_for_stripe
);
195 if (conf
->retry_read_aligned
)
196 md_wakeup_thread(conf
->mddev
->thread
);
201 static void release_stripe(struct stripe_head
*sh
)
203 raid5_conf_t
*conf
= sh
->raid_conf
;
206 spin_lock_irqsave(&conf
->device_lock
, flags
);
207 __release_stripe(conf
, sh
);
208 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
211 static inline void remove_hash(struct stripe_head
*sh
)
213 pr_debug("remove_hash(), stripe %llu\n",
214 (unsigned long long)sh
->sector
);
216 hlist_del_init(&sh
->hash
);
219 static inline void insert_hash(raid5_conf_t
*conf
, struct stripe_head
*sh
)
221 struct hlist_head
*hp
= stripe_hash(conf
, sh
->sector
);
223 pr_debug("insert_hash(), stripe %llu\n",
224 (unsigned long long)sh
->sector
);
227 hlist_add_head(&sh
->hash
, hp
);
231 /* find an idle stripe, make sure it is unhashed, and return it. */
232 static struct stripe_head
*get_free_stripe(raid5_conf_t
*conf
)
234 struct stripe_head
*sh
= NULL
;
235 struct list_head
*first
;
238 if (list_empty(&conf
->inactive_list
))
240 first
= conf
->inactive_list
.next
;
241 sh
= list_entry(first
, struct stripe_head
, lru
);
242 list_del_init(first
);
244 atomic_inc(&conf
->active_stripes
);
249 static void shrink_buffers(struct stripe_head
*sh
, int num
)
254 for (i
=0; i
<num
; i
++) {
258 sh
->dev
[i
].page
= NULL
;
263 static int grow_buffers(struct stripe_head
*sh
, int num
)
267 for (i
=0; i
<num
; i
++) {
270 if (!(page
= alloc_page(GFP_KERNEL
))) {
273 sh
->dev
[i
].page
= page
;
278 static void raid5_build_block (struct stripe_head
*sh
, int i
);
280 static void init_stripe(struct stripe_head
*sh
, sector_t sector
, int pd_idx
, int disks
)
282 raid5_conf_t
*conf
= sh
->raid_conf
;
285 BUG_ON(atomic_read(&sh
->count
) != 0);
286 BUG_ON(test_bit(STRIPE_HANDLE
, &sh
->state
));
287 BUG_ON(stripe_operations_active(sh
));
290 pr_debug("init_stripe called, stripe %llu\n",
291 (unsigned long long)sh
->sector
);
301 for (i
= sh
->disks
; i
--; ) {
302 struct r5dev
*dev
= &sh
->dev
[i
];
304 if (dev
->toread
|| dev
->read
|| dev
->towrite
|| dev
->written
||
305 test_bit(R5_LOCKED
, &dev
->flags
)) {
306 printk(KERN_ERR
"sector=%llx i=%d %p %p %p %p %d\n",
307 (unsigned long long)sh
->sector
, i
, dev
->toread
,
308 dev
->read
, dev
->towrite
, dev
->written
,
309 test_bit(R5_LOCKED
, &dev
->flags
));
313 raid5_build_block(sh
, i
);
315 insert_hash(conf
, sh
);
318 static struct stripe_head
*__find_stripe(raid5_conf_t
*conf
, sector_t sector
, int disks
)
320 struct stripe_head
*sh
;
321 struct hlist_node
*hn
;
324 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector
);
325 hlist_for_each_entry(sh
, hn
, stripe_hash(conf
, sector
), hash
)
326 if (sh
->sector
== sector
&& sh
->disks
== disks
)
328 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector
);
332 static void unplug_slaves(mddev_t
*mddev
);
333 static void raid5_unplug_device(struct request_queue
*q
);
335 static struct stripe_head
*get_active_stripe(raid5_conf_t
*conf
, sector_t sector
, int disks
,
336 int pd_idx
, int noblock
)
338 struct stripe_head
*sh
;
340 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector
);
342 spin_lock_irq(&conf
->device_lock
);
345 wait_event_lock_irq(conf
->wait_for_stripe
,
347 conf
->device_lock
, /* nothing */);
348 sh
= __find_stripe(conf
, sector
, disks
);
350 if (!conf
->inactive_blocked
)
351 sh
= get_free_stripe(conf
);
352 if (noblock
&& sh
== NULL
)
355 conf
->inactive_blocked
= 1;
356 wait_event_lock_irq(conf
->wait_for_stripe
,
357 !list_empty(&conf
->inactive_list
) &&
358 (atomic_read(&conf
->active_stripes
)
359 < (conf
->max_nr_stripes
*3/4)
360 || !conf
->inactive_blocked
),
362 raid5_unplug_device(conf
->mddev
->queue
)
364 conf
->inactive_blocked
= 0;
366 init_stripe(sh
, sector
, pd_idx
, disks
);
368 if (atomic_read(&sh
->count
)) {
369 BUG_ON(!list_empty(&sh
->lru
));
371 if (!test_bit(STRIPE_HANDLE
, &sh
->state
))
372 atomic_inc(&conf
->active_stripes
);
373 if (list_empty(&sh
->lru
) &&
374 !test_bit(STRIPE_EXPANDING
, &sh
->state
))
376 list_del_init(&sh
->lru
);
379 } while (sh
== NULL
);
382 atomic_inc(&sh
->count
);
384 spin_unlock_irq(&conf
->device_lock
);
389 raid5_end_read_request(struct bio
*bi
, int error
);
391 raid5_end_write_request(struct bio
*bi
, int error
);
393 static void ops_run_io(struct stripe_head
*sh
, struct stripe_head_state
*s
)
395 raid5_conf_t
*conf
= sh
->raid_conf
;
396 int i
, disks
= sh
->disks
;
400 for (i
= disks
; i
--; ) {
404 if (test_and_clear_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
))
406 else if (test_and_clear_bit(R5_Wantread
, &sh
->dev
[i
].flags
))
411 bi
= &sh
->dev
[i
].req
;
415 bi
->bi_end_io
= raid5_end_write_request
;
417 bi
->bi_end_io
= raid5_end_read_request
;
420 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
421 if (rdev
&& test_bit(Faulty
, &rdev
->flags
))
424 atomic_inc(&rdev
->nr_pending
);
428 if (s
->syncing
|| s
->expanding
|| s
->expanded
)
429 md_sync_acct(rdev
->bdev
, STRIPE_SECTORS
);
431 set_bit(STRIPE_IO_STARTED
, &sh
->state
);
433 bi
->bi_bdev
= rdev
->bdev
;
434 pr_debug("%s: for %llu schedule op %ld on disc %d\n",
435 __func__
, (unsigned long long)sh
->sector
,
437 atomic_inc(&sh
->count
);
438 bi
->bi_sector
= sh
->sector
+ rdev
->data_offset
;
439 bi
->bi_flags
= 1 << BIO_UPTODATE
;
443 bi
->bi_io_vec
= &sh
->dev
[i
].vec
;
444 bi
->bi_io_vec
[0].bv_len
= STRIPE_SIZE
;
445 bi
->bi_io_vec
[0].bv_offset
= 0;
446 bi
->bi_size
= STRIPE_SIZE
;
449 test_bit(R5_ReWrite
, &sh
->dev
[i
].flags
))
450 atomic_add(STRIPE_SECTORS
,
451 &rdev
->corrected_errors
);
452 generic_make_request(bi
);
455 set_bit(STRIPE_DEGRADED
, &sh
->state
);
456 pr_debug("skip op %ld on disc %d for sector %llu\n",
457 bi
->bi_rw
, i
, (unsigned long long)sh
->sector
);
458 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
459 set_bit(STRIPE_HANDLE
, &sh
->state
);
464 static struct dma_async_tx_descriptor
*
465 async_copy_data(int frombio
, struct bio
*bio
, struct page
*page
,
466 sector_t sector
, struct dma_async_tx_descriptor
*tx
)
469 struct page
*bio_page
;
473 if (bio
->bi_sector
>= sector
)
474 page_offset
= (signed)(bio
->bi_sector
- sector
) * 512;
476 page_offset
= (signed)(sector
- bio
->bi_sector
) * -512;
477 bio_for_each_segment(bvl
, bio
, i
) {
478 int len
= bio_iovec_idx(bio
, i
)->bv_len
;
482 if (page_offset
< 0) {
483 b_offset
= -page_offset
;
484 page_offset
+= b_offset
;
488 if (len
> 0 && page_offset
+ len
> STRIPE_SIZE
)
489 clen
= STRIPE_SIZE
- page_offset
;
494 b_offset
+= bio_iovec_idx(bio
, i
)->bv_offset
;
495 bio_page
= bio_iovec_idx(bio
, i
)->bv_page
;
497 tx
= async_memcpy(page
, bio_page
, page_offset
,
502 tx
= async_memcpy(bio_page
, page
, b_offset
,
507 if (clen
< len
) /* hit end of page */
515 static void ops_complete_biofill(void *stripe_head_ref
)
517 struct stripe_head
*sh
= stripe_head_ref
;
518 struct bio
*return_bi
= NULL
;
519 raid5_conf_t
*conf
= sh
->raid_conf
;
522 pr_debug("%s: stripe %llu\n", __func__
,
523 (unsigned long long)sh
->sector
);
525 /* clear completed biofills */
526 spin_lock_irq(&conf
->device_lock
);
527 for (i
= sh
->disks
; i
--; ) {
528 struct r5dev
*dev
= &sh
->dev
[i
];
530 /* acknowledge completion of a biofill operation */
531 /* and check if we need to reply to a read request,
532 * new R5_Wantfill requests are held off until
533 * !STRIPE_BIOFILL_RUN
535 if (test_and_clear_bit(R5_Wantfill
, &dev
->flags
)) {
536 struct bio
*rbi
, *rbi2
;
541 while (rbi
&& rbi
->bi_sector
<
542 dev
->sector
+ STRIPE_SECTORS
) {
543 rbi2
= r5_next_bio(rbi
, dev
->sector
);
544 if (!raid5_dec_bi_phys_segments(rbi
)) {
545 rbi
->bi_next
= return_bi
;
552 spin_unlock_irq(&conf
->device_lock
);
553 clear_bit(STRIPE_BIOFILL_RUN
, &sh
->state
);
555 return_io(return_bi
);
557 set_bit(STRIPE_HANDLE
, &sh
->state
);
561 static void ops_run_biofill(struct stripe_head
*sh
)
563 struct dma_async_tx_descriptor
*tx
= NULL
;
564 raid5_conf_t
*conf
= sh
->raid_conf
;
567 pr_debug("%s: stripe %llu\n", __func__
,
568 (unsigned long long)sh
->sector
);
570 for (i
= sh
->disks
; i
--; ) {
571 struct r5dev
*dev
= &sh
->dev
[i
];
572 if (test_bit(R5_Wantfill
, &dev
->flags
)) {
574 spin_lock_irq(&conf
->device_lock
);
575 dev
->read
= rbi
= dev
->toread
;
577 spin_unlock_irq(&conf
->device_lock
);
578 while (rbi
&& rbi
->bi_sector
<
579 dev
->sector
+ STRIPE_SECTORS
) {
580 tx
= async_copy_data(0, rbi
, dev
->page
,
582 rbi
= r5_next_bio(rbi
, dev
->sector
);
587 atomic_inc(&sh
->count
);
588 async_trigger_callback(ASYNC_TX_DEP_ACK
| ASYNC_TX_ACK
, tx
,
589 ops_complete_biofill
, sh
);
592 static void ops_complete_compute5(void *stripe_head_ref
)
594 struct stripe_head
*sh
= stripe_head_ref
;
595 int target
= sh
->ops
.target
;
596 struct r5dev
*tgt
= &sh
->dev
[target
];
598 pr_debug("%s: stripe %llu\n", __func__
,
599 (unsigned long long)sh
->sector
);
601 set_bit(R5_UPTODATE
, &tgt
->flags
);
602 BUG_ON(!test_bit(R5_Wantcompute
, &tgt
->flags
));
603 clear_bit(R5_Wantcompute
, &tgt
->flags
);
604 clear_bit(STRIPE_COMPUTE_RUN
, &sh
->state
);
605 if (sh
->check_state
== check_state_compute_run
)
606 sh
->check_state
= check_state_compute_result
;
607 set_bit(STRIPE_HANDLE
, &sh
->state
);
611 static struct dma_async_tx_descriptor
*ops_run_compute5(struct stripe_head
*sh
)
613 /* kernel stack size limits the total number of disks */
614 int disks
= sh
->disks
;
615 struct page
*xor_srcs
[disks
];
616 int target
= sh
->ops
.target
;
617 struct r5dev
*tgt
= &sh
->dev
[target
];
618 struct page
*xor_dest
= tgt
->page
;
620 struct dma_async_tx_descriptor
*tx
;
623 pr_debug("%s: stripe %llu block: %d\n",
624 __func__
, (unsigned long long)sh
->sector
, target
);
625 BUG_ON(!test_bit(R5_Wantcompute
, &tgt
->flags
));
627 for (i
= disks
; i
--; )
629 xor_srcs
[count
++] = sh
->dev
[i
].page
;
631 atomic_inc(&sh
->count
);
633 if (unlikely(count
== 1))
634 tx
= async_memcpy(xor_dest
, xor_srcs
[0], 0, 0, STRIPE_SIZE
,
635 0, NULL
, ops_complete_compute5
, sh
);
637 tx
= async_xor(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
,
638 ASYNC_TX_XOR_ZERO_DST
, NULL
,
639 ops_complete_compute5
, sh
);
644 static void ops_complete_prexor(void *stripe_head_ref
)
646 struct stripe_head
*sh
= stripe_head_ref
;
648 pr_debug("%s: stripe %llu\n", __func__
,
649 (unsigned long long)sh
->sector
);
652 static struct dma_async_tx_descriptor
*
653 ops_run_prexor(struct stripe_head
*sh
, struct dma_async_tx_descriptor
*tx
)
655 /* kernel stack size limits the total number of disks */
656 int disks
= sh
->disks
;
657 struct page
*xor_srcs
[disks
];
658 int count
= 0, pd_idx
= sh
->pd_idx
, i
;
660 /* existing parity data subtracted */
661 struct page
*xor_dest
= xor_srcs
[count
++] = sh
->dev
[pd_idx
].page
;
663 pr_debug("%s: stripe %llu\n", __func__
,
664 (unsigned long long)sh
->sector
);
666 for (i
= disks
; i
--; ) {
667 struct r5dev
*dev
= &sh
->dev
[i
];
668 /* Only process blocks that are known to be uptodate */
669 if (test_bit(R5_Wantdrain
, &dev
->flags
))
670 xor_srcs
[count
++] = dev
->page
;
673 tx
= async_xor(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
,
674 ASYNC_TX_DEP_ACK
| ASYNC_TX_XOR_DROP_DST
, tx
,
675 ops_complete_prexor
, sh
);
680 static struct dma_async_tx_descriptor
*
681 ops_run_biodrain(struct stripe_head
*sh
, struct dma_async_tx_descriptor
*tx
)
683 int disks
= sh
->disks
;
686 pr_debug("%s: stripe %llu\n", __func__
,
687 (unsigned long long)sh
->sector
);
689 for (i
= disks
; i
--; ) {
690 struct r5dev
*dev
= &sh
->dev
[i
];
693 if (test_and_clear_bit(R5_Wantdrain
, &dev
->flags
)) {
696 spin_lock(&sh
->lock
);
697 chosen
= dev
->towrite
;
699 BUG_ON(dev
->written
);
700 wbi
= dev
->written
= chosen
;
701 spin_unlock(&sh
->lock
);
703 while (wbi
&& wbi
->bi_sector
<
704 dev
->sector
+ STRIPE_SECTORS
) {
705 tx
= async_copy_data(1, wbi
, dev
->page
,
707 wbi
= r5_next_bio(wbi
, dev
->sector
);
715 static void ops_complete_postxor(void *stripe_head_ref
)
717 struct stripe_head
*sh
= stripe_head_ref
;
718 int disks
= sh
->disks
, i
, pd_idx
= sh
->pd_idx
;
720 pr_debug("%s: stripe %llu\n", __func__
,
721 (unsigned long long)sh
->sector
);
723 for (i
= disks
; i
--; ) {
724 struct r5dev
*dev
= &sh
->dev
[i
];
725 if (dev
->written
|| i
== pd_idx
)
726 set_bit(R5_UPTODATE
, &dev
->flags
);
729 if (sh
->reconstruct_state
== reconstruct_state_drain_run
)
730 sh
->reconstruct_state
= reconstruct_state_drain_result
;
731 else if (sh
->reconstruct_state
== reconstruct_state_prexor_drain_run
)
732 sh
->reconstruct_state
= reconstruct_state_prexor_drain_result
;
734 BUG_ON(sh
->reconstruct_state
!= reconstruct_state_run
);
735 sh
->reconstruct_state
= reconstruct_state_result
;
738 set_bit(STRIPE_HANDLE
, &sh
->state
);
743 ops_run_postxor(struct stripe_head
*sh
, struct dma_async_tx_descriptor
*tx
)
745 /* kernel stack size limits the total number of disks */
746 int disks
= sh
->disks
;
747 struct page
*xor_srcs
[disks
];
749 int count
= 0, pd_idx
= sh
->pd_idx
, i
;
750 struct page
*xor_dest
;
754 pr_debug("%s: stripe %llu\n", __func__
,
755 (unsigned long long)sh
->sector
);
757 /* check if prexor is active which means only process blocks
758 * that are part of a read-modify-write (written)
760 if (sh
->reconstruct_state
== reconstruct_state_prexor_drain_run
) {
762 xor_dest
= xor_srcs
[count
++] = sh
->dev
[pd_idx
].page
;
763 for (i
= disks
; i
--; ) {
764 struct r5dev
*dev
= &sh
->dev
[i
];
766 xor_srcs
[count
++] = dev
->page
;
769 xor_dest
= sh
->dev
[pd_idx
].page
;
770 for (i
= disks
; i
--; ) {
771 struct r5dev
*dev
= &sh
->dev
[i
];
773 xor_srcs
[count
++] = dev
->page
;
777 /* 1/ if we prexor'd then the dest is reused as a source
778 * 2/ if we did not prexor then we are redoing the parity
779 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
780 * for the synchronous xor case
782 flags
= ASYNC_TX_DEP_ACK
| ASYNC_TX_ACK
|
783 (prexor
? ASYNC_TX_XOR_DROP_DST
: ASYNC_TX_XOR_ZERO_DST
);
785 atomic_inc(&sh
->count
);
787 if (unlikely(count
== 1)) {
788 flags
&= ~(ASYNC_TX_XOR_DROP_DST
| ASYNC_TX_XOR_ZERO_DST
);
789 tx
= async_memcpy(xor_dest
, xor_srcs
[0], 0, 0, STRIPE_SIZE
,
790 flags
, tx
, ops_complete_postxor
, sh
);
792 tx
= async_xor(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
,
793 flags
, tx
, ops_complete_postxor
, sh
);
796 static void ops_complete_check(void *stripe_head_ref
)
798 struct stripe_head
*sh
= stripe_head_ref
;
800 pr_debug("%s: stripe %llu\n", __func__
,
801 (unsigned long long)sh
->sector
);
803 sh
->check_state
= check_state_check_result
;
804 set_bit(STRIPE_HANDLE
, &sh
->state
);
808 static void ops_run_check(struct stripe_head
*sh
)
810 /* kernel stack size limits the total number of disks */
811 int disks
= sh
->disks
;
812 struct page
*xor_srcs
[disks
];
813 struct dma_async_tx_descriptor
*tx
;
815 int count
= 0, pd_idx
= sh
->pd_idx
, i
;
816 struct page
*xor_dest
= xor_srcs
[count
++] = sh
->dev
[pd_idx
].page
;
818 pr_debug("%s: stripe %llu\n", __func__
,
819 (unsigned long long)sh
->sector
);
821 for (i
= disks
; i
--; ) {
822 struct r5dev
*dev
= &sh
->dev
[i
];
824 xor_srcs
[count
++] = dev
->page
;
827 tx
= async_xor_zero_sum(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
,
828 &sh
->ops
.zero_sum_result
, 0, NULL
, NULL
, NULL
);
830 atomic_inc(&sh
->count
);
831 tx
= async_trigger_callback(ASYNC_TX_DEP_ACK
| ASYNC_TX_ACK
, tx
,
832 ops_complete_check
, sh
);
835 static void raid5_run_ops(struct stripe_head
*sh
, unsigned long ops_request
)
837 int overlap_clear
= 0, i
, disks
= sh
->disks
;
838 struct dma_async_tx_descriptor
*tx
= NULL
;
840 if (test_bit(STRIPE_OP_BIOFILL
, &ops_request
)) {
845 if (test_bit(STRIPE_OP_COMPUTE_BLK
, &ops_request
)) {
846 tx
= ops_run_compute5(sh
);
847 /* terminate the chain if postxor is not set to be run */
848 if (tx
&& !test_bit(STRIPE_OP_POSTXOR
, &ops_request
))
852 if (test_bit(STRIPE_OP_PREXOR
, &ops_request
))
853 tx
= ops_run_prexor(sh
, tx
);
855 if (test_bit(STRIPE_OP_BIODRAIN
, &ops_request
)) {
856 tx
= ops_run_biodrain(sh
, tx
);
860 if (test_bit(STRIPE_OP_POSTXOR
, &ops_request
))
861 ops_run_postxor(sh
, tx
);
863 if (test_bit(STRIPE_OP_CHECK
, &ops_request
))
867 for (i
= disks
; i
--; ) {
868 struct r5dev
*dev
= &sh
->dev
[i
];
869 if (test_and_clear_bit(R5_Overlap
, &dev
->flags
))
870 wake_up(&sh
->raid_conf
->wait_for_overlap
);
874 static int grow_one_stripe(raid5_conf_t
*conf
)
876 struct stripe_head
*sh
;
877 sh
= kmem_cache_alloc(conf
->slab_cache
, GFP_KERNEL
);
880 memset(sh
, 0, sizeof(*sh
) + (conf
->raid_disks
-1)*sizeof(struct r5dev
));
881 sh
->raid_conf
= conf
;
882 spin_lock_init(&sh
->lock
);
884 if (grow_buffers(sh
, conf
->raid_disks
)) {
885 shrink_buffers(sh
, conf
->raid_disks
);
886 kmem_cache_free(conf
->slab_cache
, sh
);
889 sh
->disks
= conf
->raid_disks
;
890 /* we just created an active stripe so... */
891 atomic_set(&sh
->count
, 1);
892 atomic_inc(&conf
->active_stripes
);
893 INIT_LIST_HEAD(&sh
->lru
);
898 static int grow_stripes(raid5_conf_t
*conf
, int num
)
900 struct kmem_cache
*sc
;
901 int devs
= conf
->raid_disks
;
903 sprintf(conf
->cache_name
[0], "raid5-%s", mdname(conf
->mddev
));
904 sprintf(conf
->cache_name
[1], "raid5-%s-alt", mdname(conf
->mddev
));
905 conf
->active_name
= 0;
906 sc
= kmem_cache_create(conf
->cache_name
[conf
->active_name
],
907 sizeof(struct stripe_head
)+(devs
-1)*sizeof(struct r5dev
),
911 conf
->slab_cache
= sc
;
912 conf
->pool_size
= devs
;
914 if (!grow_one_stripe(conf
))
919 #ifdef CONFIG_MD_RAID5_RESHAPE
920 static int resize_stripes(raid5_conf_t
*conf
, int newsize
)
922 /* Make all the stripes able to hold 'newsize' devices.
923 * New slots in each stripe get 'page' set to a new page.
925 * This happens in stages:
926 * 1/ create a new kmem_cache and allocate the required number of
928 * 2/ gather all the old stripe_heads and tranfer the pages across
929 * to the new stripe_heads. This will have the side effect of
930 * freezing the array as once all stripe_heads have been collected,
931 * no IO will be possible. Old stripe heads are freed once their
932 * pages have been transferred over, and the old kmem_cache is
933 * freed when all stripes are done.
934 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
935 * we simple return a failre status - no need to clean anything up.
936 * 4/ allocate new pages for the new slots in the new stripe_heads.
937 * If this fails, we don't bother trying the shrink the
938 * stripe_heads down again, we just leave them as they are.
939 * As each stripe_head is processed the new one is released into
942 * Once step2 is started, we cannot afford to wait for a write,
943 * so we use GFP_NOIO allocations.
945 struct stripe_head
*osh
, *nsh
;
946 LIST_HEAD(newstripes
);
947 struct disk_info
*ndisks
;
949 struct kmem_cache
*sc
;
952 if (newsize
<= conf
->pool_size
)
953 return 0; /* never bother to shrink */
955 err
= md_allow_write(conf
->mddev
);
960 sc
= kmem_cache_create(conf
->cache_name
[1-conf
->active_name
],
961 sizeof(struct stripe_head
)+(newsize
-1)*sizeof(struct r5dev
),
966 for (i
= conf
->max_nr_stripes
; i
; i
--) {
967 nsh
= kmem_cache_alloc(sc
, GFP_KERNEL
);
971 memset(nsh
, 0, sizeof(*nsh
) + (newsize
-1)*sizeof(struct r5dev
));
973 nsh
->raid_conf
= conf
;
974 spin_lock_init(&nsh
->lock
);
976 list_add(&nsh
->lru
, &newstripes
);
979 /* didn't get enough, give up */
980 while (!list_empty(&newstripes
)) {
981 nsh
= list_entry(newstripes
.next
, struct stripe_head
, lru
);
983 kmem_cache_free(sc
, nsh
);
985 kmem_cache_destroy(sc
);
988 /* Step 2 - Must use GFP_NOIO now.
989 * OK, we have enough stripes, start collecting inactive
990 * stripes and copying them over
992 list_for_each_entry(nsh
, &newstripes
, lru
) {
993 spin_lock_irq(&conf
->device_lock
);
994 wait_event_lock_irq(conf
->wait_for_stripe
,
995 !list_empty(&conf
->inactive_list
),
997 unplug_slaves(conf
->mddev
)
999 osh
= get_free_stripe(conf
);
1000 spin_unlock_irq(&conf
->device_lock
);
1001 atomic_set(&nsh
->count
, 1);
1002 for(i
=0; i
<conf
->pool_size
; i
++)
1003 nsh
->dev
[i
].page
= osh
->dev
[i
].page
;
1004 for( ; i
<newsize
; i
++)
1005 nsh
->dev
[i
].page
= NULL
;
1006 kmem_cache_free(conf
->slab_cache
, osh
);
1008 kmem_cache_destroy(conf
->slab_cache
);
1011 * At this point, we are holding all the stripes so the array
1012 * is completely stalled, so now is a good time to resize
1015 ndisks
= kzalloc(newsize
* sizeof(struct disk_info
), GFP_NOIO
);
1017 for (i
=0; i
<conf
->raid_disks
; i
++)
1018 ndisks
[i
] = conf
->disks
[i
];
1020 conf
->disks
= ndisks
;
1024 /* Step 4, return new stripes to service */
1025 while(!list_empty(&newstripes
)) {
1026 nsh
= list_entry(newstripes
.next
, struct stripe_head
, lru
);
1027 list_del_init(&nsh
->lru
);
1028 for (i
=conf
->raid_disks
; i
< newsize
; i
++)
1029 if (nsh
->dev
[i
].page
== NULL
) {
1030 struct page
*p
= alloc_page(GFP_NOIO
);
1031 nsh
->dev
[i
].page
= p
;
1035 release_stripe(nsh
);
1037 /* critical section pass, GFP_NOIO no longer needed */
1039 conf
->slab_cache
= sc
;
1040 conf
->active_name
= 1-conf
->active_name
;
1041 conf
->pool_size
= newsize
;
1046 static int drop_one_stripe(raid5_conf_t
*conf
)
1048 struct stripe_head
*sh
;
1050 spin_lock_irq(&conf
->device_lock
);
1051 sh
= get_free_stripe(conf
);
1052 spin_unlock_irq(&conf
->device_lock
);
1055 BUG_ON(atomic_read(&sh
->count
));
1056 shrink_buffers(sh
, conf
->pool_size
);
1057 kmem_cache_free(conf
->slab_cache
, sh
);
1058 atomic_dec(&conf
->active_stripes
);
1062 static void shrink_stripes(raid5_conf_t
*conf
)
1064 while (drop_one_stripe(conf
))
1067 if (conf
->slab_cache
)
1068 kmem_cache_destroy(conf
->slab_cache
);
1069 conf
->slab_cache
= NULL
;
1072 static void raid5_end_read_request(struct bio
* bi
, int error
)
1074 struct stripe_head
*sh
= bi
->bi_private
;
1075 raid5_conf_t
*conf
= sh
->raid_conf
;
1076 int disks
= sh
->disks
, i
;
1077 int uptodate
= test_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1078 char b
[BDEVNAME_SIZE
];
1082 for (i
=0 ; i
<disks
; i
++)
1083 if (bi
== &sh
->dev
[i
].req
)
1086 pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n",
1087 (unsigned long long)sh
->sector
, i
, atomic_read(&sh
->count
),
1095 set_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
);
1096 if (test_bit(R5_ReadError
, &sh
->dev
[i
].flags
)) {
1097 rdev
= conf
->disks
[i
].rdev
;
1098 printk_rl(KERN_INFO
"raid5:%s: read error corrected"
1099 " (%lu sectors at %llu on %s)\n",
1100 mdname(conf
->mddev
), STRIPE_SECTORS
,
1101 (unsigned long long)(sh
->sector
1102 + rdev
->data_offset
),
1103 bdevname(rdev
->bdev
, b
));
1104 clear_bit(R5_ReadError
, &sh
->dev
[i
].flags
);
1105 clear_bit(R5_ReWrite
, &sh
->dev
[i
].flags
);
1107 if (atomic_read(&conf
->disks
[i
].rdev
->read_errors
))
1108 atomic_set(&conf
->disks
[i
].rdev
->read_errors
, 0);
1110 const char *bdn
= bdevname(conf
->disks
[i
].rdev
->bdev
, b
);
1112 rdev
= conf
->disks
[i
].rdev
;
1114 clear_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
);
1115 atomic_inc(&rdev
->read_errors
);
1116 if (conf
->mddev
->degraded
)
1117 printk_rl(KERN_WARNING
1118 "raid5:%s: read error not correctable "
1119 "(sector %llu on %s).\n",
1120 mdname(conf
->mddev
),
1121 (unsigned long long)(sh
->sector
1122 + rdev
->data_offset
),
1124 else if (test_bit(R5_ReWrite
, &sh
->dev
[i
].flags
))
1126 printk_rl(KERN_WARNING
1127 "raid5:%s: read error NOT corrected!! "
1128 "(sector %llu on %s).\n",
1129 mdname(conf
->mddev
),
1130 (unsigned long long)(sh
->sector
1131 + rdev
->data_offset
),
1133 else if (atomic_read(&rdev
->read_errors
)
1134 > conf
->max_nr_stripes
)
1136 "raid5:%s: Too many read errors, failing device %s.\n",
1137 mdname(conf
->mddev
), bdn
);
1141 set_bit(R5_ReadError
, &sh
->dev
[i
].flags
);
1143 clear_bit(R5_ReadError
, &sh
->dev
[i
].flags
);
1144 clear_bit(R5_ReWrite
, &sh
->dev
[i
].flags
);
1145 md_error(conf
->mddev
, rdev
);
1148 rdev_dec_pending(conf
->disks
[i
].rdev
, conf
->mddev
);
1149 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
1150 set_bit(STRIPE_HANDLE
, &sh
->state
);
1154 static void raid5_end_write_request (struct bio
*bi
, int error
)
1156 struct stripe_head
*sh
= bi
->bi_private
;
1157 raid5_conf_t
*conf
= sh
->raid_conf
;
1158 int disks
= sh
->disks
, i
;
1159 int uptodate
= test_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1161 for (i
=0 ; i
<disks
; i
++)
1162 if (bi
== &sh
->dev
[i
].req
)
1165 pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n",
1166 (unsigned long long)sh
->sector
, i
, atomic_read(&sh
->count
),
1174 md_error(conf
->mddev
, conf
->disks
[i
].rdev
);
1176 rdev_dec_pending(conf
->disks
[i
].rdev
, conf
->mddev
);
1178 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
1179 set_bit(STRIPE_HANDLE
, &sh
->state
);
1184 static sector_t
compute_blocknr(struct stripe_head
*sh
, int i
);
1186 static void raid5_build_block (struct stripe_head
*sh
, int i
)
1188 struct r5dev
*dev
= &sh
->dev
[i
];
1190 bio_init(&dev
->req
);
1191 dev
->req
.bi_io_vec
= &dev
->vec
;
1193 dev
->req
.bi_max_vecs
++;
1194 dev
->vec
.bv_page
= dev
->page
;
1195 dev
->vec
.bv_len
= STRIPE_SIZE
;
1196 dev
->vec
.bv_offset
= 0;
1198 dev
->req
.bi_sector
= sh
->sector
;
1199 dev
->req
.bi_private
= sh
;
1202 dev
->sector
= compute_blocknr(sh
, i
);
1205 static void error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1207 char b
[BDEVNAME_SIZE
];
1208 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
1209 pr_debug("raid5: error called\n");
1211 if (!test_bit(Faulty
, &rdev
->flags
)) {
1212 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
1213 if (test_and_clear_bit(In_sync
, &rdev
->flags
)) {
1214 unsigned long flags
;
1215 spin_lock_irqsave(&conf
->device_lock
, flags
);
1217 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1219 * if recovery was running, make sure it aborts.
1221 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
1223 set_bit(Faulty
, &rdev
->flags
);
1225 "raid5: Disk failure on %s, disabling device.\n"
1226 "raid5: Operation continuing on %d devices.\n",
1227 bdevname(rdev
->bdev
,b
), conf
->raid_disks
- mddev
->degraded
);
1232 * Input: a 'big' sector number,
1233 * Output: index of the data and parity disk, and the sector # in them.
1235 static sector_t
raid5_compute_sector(sector_t r_sector
, unsigned int raid_disks
,
1236 unsigned int data_disks
, unsigned int * dd_idx
,
1237 unsigned int * pd_idx
, raid5_conf_t
*conf
)
1240 unsigned long chunk_number
;
1241 unsigned int chunk_offset
;
1242 sector_t new_sector
;
1243 int sectors_per_chunk
= conf
->chunk_size
>> 9;
1245 /* First compute the information on this sector */
1248 * Compute the chunk number and the sector offset inside the chunk
1250 chunk_offset
= sector_div(r_sector
, sectors_per_chunk
);
1251 chunk_number
= r_sector
;
1252 BUG_ON(r_sector
!= chunk_number
);
1255 * Compute the stripe number
1257 stripe
= chunk_number
/ data_disks
;
1260 * Compute the data disk and parity disk indexes inside the stripe
1262 *dd_idx
= chunk_number
% data_disks
;
1265 * Select the parity disk based on the user selected algorithm.
1267 switch(conf
->level
) {
1269 *pd_idx
= data_disks
;
1272 switch (conf
->algorithm
) {
1273 case ALGORITHM_LEFT_ASYMMETRIC
:
1274 *pd_idx
= data_disks
- stripe
% raid_disks
;
1275 if (*dd_idx
>= *pd_idx
)
1278 case ALGORITHM_RIGHT_ASYMMETRIC
:
1279 *pd_idx
= stripe
% raid_disks
;
1280 if (*dd_idx
>= *pd_idx
)
1283 case ALGORITHM_LEFT_SYMMETRIC
:
1284 *pd_idx
= data_disks
- stripe
% raid_disks
;
1285 *dd_idx
= (*pd_idx
+ 1 + *dd_idx
) % raid_disks
;
1287 case ALGORITHM_RIGHT_SYMMETRIC
:
1288 *pd_idx
= stripe
% raid_disks
;
1289 *dd_idx
= (*pd_idx
+ 1 + *dd_idx
) % raid_disks
;
1292 printk(KERN_ERR
"raid5: unsupported algorithm %d\n",
1298 /**** FIX THIS ****/
1299 switch (conf
->algorithm
) {
1300 case ALGORITHM_LEFT_ASYMMETRIC
:
1301 *pd_idx
= raid_disks
- 1 - (stripe
% raid_disks
);
1302 if (*pd_idx
== raid_disks
-1)
1303 (*dd_idx
)++; /* Q D D D P */
1304 else if (*dd_idx
>= *pd_idx
)
1305 (*dd_idx
) += 2; /* D D P Q D */
1307 case ALGORITHM_RIGHT_ASYMMETRIC
:
1308 *pd_idx
= stripe
% raid_disks
;
1309 if (*pd_idx
== raid_disks
-1)
1310 (*dd_idx
)++; /* Q D D D P */
1311 else if (*dd_idx
>= *pd_idx
)
1312 (*dd_idx
) += 2; /* D D P Q D */
1314 case ALGORITHM_LEFT_SYMMETRIC
:
1315 *pd_idx
= raid_disks
- 1 - (stripe
% raid_disks
);
1316 *dd_idx
= (*pd_idx
+ 2 + *dd_idx
) % raid_disks
;
1318 case ALGORITHM_RIGHT_SYMMETRIC
:
1319 *pd_idx
= stripe
% raid_disks
;
1320 *dd_idx
= (*pd_idx
+ 2 + *dd_idx
) % raid_disks
;
1323 printk (KERN_CRIT
"raid6: unsupported algorithm %d\n",
1330 * Finally, compute the new sector number
1332 new_sector
= (sector_t
)stripe
* sectors_per_chunk
+ chunk_offset
;
1337 static sector_t
compute_blocknr(struct stripe_head
*sh
, int i
)
1339 raid5_conf_t
*conf
= sh
->raid_conf
;
1340 int raid_disks
= sh
->disks
;
1341 int data_disks
= raid_disks
- conf
->max_degraded
;
1342 sector_t new_sector
= sh
->sector
, check
;
1343 int sectors_per_chunk
= conf
->chunk_size
>> 9;
1346 int chunk_number
, dummy1
, dummy2
, dd_idx
= i
;
1350 chunk_offset
= sector_div(new_sector
, sectors_per_chunk
);
1351 stripe
= new_sector
;
1352 BUG_ON(new_sector
!= stripe
);
1354 if (i
== sh
->pd_idx
)
1356 switch(conf
->level
) {
1359 switch (conf
->algorithm
) {
1360 case ALGORITHM_LEFT_ASYMMETRIC
:
1361 case ALGORITHM_RIGHT_ASYMMETRIC
:
1365 case ALGORITHM_LEFT_SYMMETRIC
:
1366 case ALGORITHM_RIGHT_SYMMETRIC
:
1369 i
-= (sh
->pd_idx
+ 1);
1372 printk(KERN_ERR
"raid5: unsupported algorithm %d\n",
1377 if (i
== raid6_next_disk(sh
->pd_idx
, raid_disks
))
1378 return 0; /* It is the Q disk */
1379 switch (conf
->algorithm
) {
1380 case ALGORITHM_LEFT_ASYMMETRIC
:
1381 case ALGORITHM_RIGHT_ASYMMETRIC
:
1382 if (sh
->pd_idx
== raid_disks
-1)
1383 i
--; /* Q D D D P */
1384 else if (i
> sh
->pd_idx
)
1385 i
-= 2; /* D D P Q D */
1387 case ALGORITHM_LEFT_SYMMETRIC
:
1388 case ALGORITHM_RIGHT_SYMMETRIC
:
1389 if (sh
->pd_idx
== raid_disks
-1)
1390 i
--; /* Q D D D P */
1395 i
-= (sh
->pd_idx
+ 2);
1399 printk (KERN_CRIT
"raid6: unsupported algorithm %d\n",
1405 chunk_number
= stripe
* data_disks
+ i
;
1406 r_sector
= (sector_t
)chunk_number
* sectors_per_chunk
+ chunk_offset
;
1408 check
= raid5_compute_sector (r_sector
, raid_disks
, data_disks
, &dummy1
, &dummy2
, conf
);
1409 if (check
!= sh
->sector
|| dummy1
!= dd_idx
|| dummy2
!= sh
->pd_idx
) {
1410 printk(KERN_ERR
"compute_blocknr: map not correct\n");
1419 * Copy data between a page in the stripe cache, and one or more bion
1420 * The page could align with the middle of the bio, or there could be
1421 * several bion, each with several bio_vecs, which cover part of the page
1422 * Multiple bion are linked together on bi_next. There may be extras
1423 * at the end of this list. We ignore them.
1425 static void copy_data(int frombio
, struct bio
*bio
,
1429 char *pa
= page_address(page
);
1430 struct bio_vec
*bvl
;
1434 if (bio
->bi_sector
>= sector
)
1435 page_offset
= (signed)(bio
->bi_sector
- sector
) * 512;
1437 page_offset
= (signed)(sector
- bio
->bi_sector
) * -512;
1438 bio_for_each_segment(bvl
, bio
, i
) {
1439 int len
= bio_iovec_idx(bio
,i
)->bv_len
;
1443 if (page_offset
< 0) {
1444 b_offset
= -page_offset
;
1445 page_offset
+= b_offset
;
1449 if (len
> 0 && page_offset
+ len
> STRIPE_SIZE
)
1450 clen
= STRIPE_SIZE
- page_offset
;
1454 char *ba
= __bio_kmap_atomic(bio
, i
, KM_USER0
);
1456 memcpy(pa
+page_offset
, ba
+b_offset
, clen
);
1458 memcpy(ba
+b_offset
, pa
+page_offset
, clen
);
1459 __bio_kunmap_atomic(ba
, KM_USER0
);
1461 if (clen
< len
) /* hit end of page */
1467 #define check_xor() do { \
1468 if (count == MAX_XOR_BLOCKS) { \
1469 xor_blocks(count, STRIPE_SIZE, dest, ptr);\
1474 static void compute_parity6(struct stripe_head
*sh
, int method
)
1476 raid6_conf_t
*conf
= sh
->raid_conf
;
1477 int i
, pd_idx
= sh
->pd_idx
, qd_idx
, d0_idx
, disks
= sh
->disks
, count
;
1479 /**** FIX THIS: This could be very bad if disks is close to 256 ****/
1482 qd_idx
= raid6_next_disk(pd_idx
, disks
);
1483 d0_idx
= raid6_next_disk(qd_idx
, disks
);
1485 pr_debug("compute_parity, stripe %llu, method %d\n",
1486 (unsigned long long)sh
->sector
, method
);
1489 case READ_MODIFY_WRITE
:
1490 BUG(); /* READ_MODIFY_WRITE N/A for RAID-6 */
1491 case RECONSTRUCT_WRITE
:
1492 for (i
= disks
; i
-- ;)
1493 if ( i
!= pd_idx
&& i
!= qd_idx
&& sh
->dev
[i
].towrite
) {
1494 chosen
= sh
->dev
[i
].towrite
;
1495 sh
->dev
[i
].towrite
= NULL
;
1497 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
1498 wake_up(&conf
->wait_for_overlap
);
1500 BUG_ON(sh
->dev
[i
].written
);
1501 sh
->dev
[i
].written
= chosen
;
1505 BUG(); /* Not implemented yet */
1508 for (i
= disks
; i
--;)
1509 if (sh
->dev
[i
].written
) {
1510 sector_t sector
= sh
->dev
[i
].sector
;
1511 struct bio
*wbi
= sh
->dev
[i
].written
;
1512 while (wbi
&& wbi
->bi_sector
< sector
+ STRIPE_SECTORS
) {
1513 copy_data(1, wbi
, sh
->dev
[i
].page
, sector
);
1514 wbi
= r5_next_bio(wbi
, sector
);
1517 set_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
1518 set_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
);
1522 // case RECONSTRUCT_WRITE:
1523 // case CHECK_PARITY:
1524 // case UPDATE_PARITY:
1525 /* Note that unlike RAID-5, the ordering of the disks matters greatly. */
1526 /* FIX: Is this ordering of drives even remotely optimal? */
1530 ptrs
[count
++] = page_address(sh
->dev
[i
].page
);
1531 if (count
<= disks
-2 && !test_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
))
1532 printk("block %d/%d not uptodate on parity calc\n", i
,count
);
1533 i
= raid6_next_disk(i
, disks
);
1534 } while ( i
!= d0_idx
);
1538 raid6_call
.gen_syndrome(disks
, STRIPE_SIZE
, ptrs
);
1541 case RECONSTRUCT_WRITE
:
1542 set_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
1543 set_bit(R5_UPTODATE
, &sh
->dev
[qd_idx
].flags
);
1544 set_bit(R5_LOCKED
, &sh
->dev
[pd_idx
].flags
);
1545 set_bit(R5_LOCKED
, &sh
->dev
[qd_idx
].flags
);
1548 set_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
1549 set_bit(R5_UPTODATE
, &sh
->dev
[qd_idx
].flags
);
1555 /* Compute one missing block */
1556 static void compute_block_1(struct stripe_head
*sh
, int dd_idx
, int nozero
)
1558 int i
, count
, disks
= sh
->disks
;
1559 void *ptr
[MAX_XOR_BLOCKS
], *dest
, *p
;
1560 int pd_idx
= sh
->pd_idx
;
1561 int qd_idx
= raid6_next_disk(pd_idx
, disks
);
1563 pr_debug("compute_block_1, stripe %llu, idx %d\n",
1564 (unsigned long long)sh
->sector
, dd_idx
);
1566 if ( dd_idx
== qd_idx
) {
1567 /* We're actually computing the Q drive */
1568 compute_parity6(sh
, UPDATE_PARITY
);
1570 dest
= page_address(sh
->dev
[dd_idx
].page
);
1571 if (!nozero
) memset(dest
, 0, STRIPE_SIZE
);
1573 for (i
= disks
; i
--; ) {
1574 if (i
== dd_idx
|| i
== qd_idx
)
1576 p
= page_address(sh
->dev
[i
].page
);
1577 if (test_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
))
1580 printk("compute_block() %d, stripe %llu, %d"
1581 " not present\n", dd_idx
,
1582 (unsigned long long)sh
->sector
, i
);
1587 xor_blocks(count
, STRIPE_SIZE
, dest
, ptr
);
1588 if (!nozero
) set_bit(R5_UPTODATE
, &sh
->dev
[dd_idx
].flags
);
1589 else clear_bit(R5_UPTODATE
, &sh
->dev
[dd_idx
].flags
);
1593 /* Compute two missing blocks */
1594 static void compute_block_2(struct stripe_head
*sh
, int dd_idx1
, int dd_idx2
)
1596 int i
, count
, disks
= sh
->disks
;
1597 int pd_idx
= sh
->pd_idx
;
1598 int qd_idx
= raid6_next_disk(pd_idx
, disks
);
1599 int d0_idx
= raid6_next_disk(qd_idx
, disks
);
1602 /* faila and failb are disk numbers relative to d0_idx */
1603 /* pd_idx become disks-2 and qd_idx become disks-1 */
1604 faila
= (dd_idx1
< d0_idx
) ? dd_idx1
+(disks
-d0_idx
) : dd_idx1
-d0_idx
;
1605 failb
= (dd_idx2
< d0_idx
) ? dd_idx2
+(disks
-d0_idx
) : dd_idx2
-d0_idx
;
1607 BUG_ON(faila
== failb
);
1608 if ( failb
< faila
) { int tmp
= faila
; faila
= failb
; failb
= tmp
; }
1610 pr_debug("compute_block_2, stripe %llu, idx %d,%d (%d,%d)\n",
1611 (unsigned long long)sh
->sector
, dd_idx1
, dd_idx2
, faila
, failb
);
1613 if ( failb
== disks
-1 ) {
1614 /* Q disk is one of the missing disks */
1615 if ( faila
== disks
-2 ) {
1616 /* Missing P+Q, just recompute */
1617 compute_parity6(sh
, UPDATE_PARITY
);
1620 /* We're missing D+Q; recompute D from P */
1621 compute_block_1(sh
, (dd_idx1
== qd_idx
) ? dd_idx2
: dd_idx1
, 0);
1622 compute_parity6(sh
, UPDATE_PARITY
); /* Is this necessary? */
1627 /* We're missing D+P or D+D; build pointer table */
1629 /**** FIX THIS: This could be very bad if disks is close to 256 ****/
1635 ptrs
[count
++] = page_address(sh
->dev
[i
].page
);
1636 i
= raid6_next_disk(i
, disks
);
1637 if (i
!= dd_idx1
&& i
!= dd_idx2
&&
1638 !test_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
))
1639 printk("compute_2 with missing block %d/%d\n", count
, i
);
1640 } while ( i
!= d0_idx
);
1642 if ( failb
== disks
-2 ) {
1643 /* We're missing D+P. */
1644 raid6_datap_recov(disks
, STRIPE_SIZE
, faila
, ptrs
);
1646 /* We're missing D+D. */
1647 raid6_2data_recov(disks
, STRIPE_SIZE
, faila
, failb
, ptrs
);
1650 /* Both the above update both missing blocks */
1651 set_bit(R5_UPTODATE
, &sh
->dev
[dd_idx1
].flags
);
1652 set_bit(R5_UPTODATE
, &sh
->dev
[dd_idx2
].flags
);
1657 schedule_reconstruction5(struct stripe_head
*sh
, struct stripe_head_state
*s
,
1658 int rcw
, int expand
)
1660 int i
, pd_idx
= sh
->pd_idx
, disks
= sh
->disks
;
1663 /* if we are not expanding this is a proper write request, and
1664 * there will be bios with new data to be drained into the
1668 sh
->reconstruct_state
= reconstruct_state_drain_run
;
1669 set_bit(STRIPE_OP_BIODRAIN
, &s
->ops_request
);
1671 sh
->reconstruct_state
= reconstruct_state_run
;
1673 set_bit(STRIPE_OP_POSTXOR
, &s
->ops_request
);
1675 for (i
= disks
; i
--; ) {
1676 struct r5dev
*dev
= &sh
->dev
[i
];
1679 set_bit(R5_LOCKED
, &dev
->flags
);
1680 set_bit(R5_Wantdrain
, &dev
->flags
);
1682 clear_bit(R5_UPTODATE
, &dev
->flags
);
1686 if (s
->locked
+ 1 == disks
)
1687 if (!test_and_set_bit(STRIPE_FULL_WRITE
, &sh
->state
))
1688 atomic_inc(&sh
->raid_conf
->pending_full_writes
);
1690 BUG_ON(!(test_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
) ||
1691 test_bit(R5_Wantcompute
, &sh
->dev
[pd_idx
].flags
)));
1693 sh
->reconstruct_state
= reconstruct_state_prexor_drain_run
;
1694 set_bit(STRIPE_OP_PREXOR
, &s
->ops_request
);
1695 set_bit(STRIPE_OP_BIODRAIN
, &s
->ops_request
);
1696 set_bit(STRIPE_OP_POSTXOR
, &s
->ops_request
);
1698 for (i
= disks
; i
--; ) {
1699 struct r5dev
*dev
= &sh
->dev
[i
];
1704 (test_bit(R5_UPTODATE
, &dev
->flags
) ||
1705 test_bit(R5_Wantcompute
, &dev
->flags
))) {
1706 set_bit(R5_Wantdrain
, &dev
->flags
);
1707 set_bit(R5_LOCKED
, &dev
->flags
);
1708 clear_bit(R5_UPTODATE
, &dev
->flags
);
1714 /* keep the parity disk locked while asynchronous operations
1717 set_bit(R5_LOCKED
, &sh
->dev
[pd_idx
].flags
);
1718 clear_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
1721 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
1722 __func__
, (unsigned long long)sh
->sector
,
1723 s
->locked
, s
->ops_request
);
1727 * Each stripe/dev can have one or more bion attached.
1728 * toread/towrite point to the first in a chain.
1729 * The bi_next chain must be in order.
1731 static int add_stripe_bio(struct stripe_head
*sh
, struct bio
*bi
, int dd_idx
, int forwrite
)
1734 raid5_conf_t
*conf
= sh
->raid_conf
;
1737 pr_debug("adding bh b#%llu to stripe s#%llu\n",
1738 (unsigned long long)bi
->bi_sector
,
1739 (unsigned long long)sh
->sector
);
1742 spin_lock(&sh
->lock
);
1743 spin_lock_irq(&conf
->device_lock
);
1745 bip
= &sh
->dev
[dd_idx
].towrite
;
1746 if (*bip
== NULL
&& sh
->dev
[dd_idx
].written
== NULL
)
1749 bip
= &sh
->dev
[dd_idx
].toread
;
1750 while (*bip
&& (*bip
)->bi_sector
< bi
->bi_sector
) {
1751 if ((*bip
)->bi_sector
+ ((*bip
)->bi_size
>> 9) > bi
->bi_sector
)
1753 bip
= & (*bip
)->bi_next
;
1755 if (*bip
&& (*bip
)->bi_sector
< bi
->bi_sector
+ ((bi
->bi_size
)>>9))
1758 BUG_ON(*bip
&& bi
->bi_next
&& (*bip
) != bi
->bi_next
);
1762 bi
->bi_phys_segments
++;
1763 spin_unlock_irq(&conf
->device_lock
);
1764 spin_unlock(&sh
->lock
);
1766 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
1767 (unsigned long long)bi
->bi_sector
,
1768 (unsigned long long)sh
->sector
, dd_idx
);
1770 if (conf
->mddev
->bitmap
&& firstwrite
) {
1771 bitmap_startwrite(conf
->mddev
->bitmap
, sh
->sector
,
1773 sh
->bm_seq
= conf
->seq_flush
+1;
1774 set_bit(STRIPE_BIT_DELAY
, &sh
->state
);
1778 /* check if page is covered */
1779 sector_t sector
= sh
->dev
[dd_idx
].sector
;
1780 for (bi
=sh
->dev
[dd_idx
].towrite
;
1781 sector
< sh
->dev
[dd_idx
].sector
+ STRIPE_SECTORS
&&
1782 bi
&& bi
->bi_sector
<= sector
;
1783 bi
= r5_next_bio(bi
, sh
->dev
[dd_idx
].sector
)) {
1784 if (bi
->bi_sector
+ (bi
->bi_size
>>9) >= sector
)
1785 sector
= bi
->bi_sector
+ (bi
->bi_size
>>9);
1787 if (sector
>= sh
->dev
[dd_idx
].sector
+ STRIPE_SECTORS
)
1788 set_bit(R5_OVERWRITE
, &sh
->dev
[dd_idx
].flags
);
1793 set_bit(R5_Overlap
, &sh
->dev
[dd_idx
].flags
);
1794 spin_unlock_irq(&conf
->device_lock
);
1795 spin_unlock(&sh
->lock
);
1799 static void end_reshape(raid5_conf_t
*conf
);
1801 static int page_is_zero(struct page
*p
)
1803 char *a
= page_address(p
);
1804 return ((*(u32
*)a
) == 0 &&
1805 memcmp(a
, a
+4, STRIPE_SIZE
-4)==0);
1808 static int stripe_to_pdidx(sector_t stripe
, raid5_conf_t
*conf
, int disks
)
1810 int sectors_per_chunk
= conf
->chunk_size
>> 9;
1812 int chunk_offset
= sector_div(stripe
, sectors_per_chunk
);
1814 raid5_compute_sector(stripe
* (disks
- conf
->max_degraded
)
1815 *sectors_per_chunk
+ chunk_offset
,
1816 disks
, disks
- conf
->max_degraded
,
1817 &dd_idx
, &pd_idx
, conf
);
1822 handle_failed_stripe(raid5_conf_t
*conf
, struct stripe_head
*sh
,
1823 struct stripe_head_state
*s
, int disks
,
1824 struct bio
**return_bi
)
1827 for (i
= disks
; i
--; ) {
1831 if (test_bit(R5_ReadError
, &sh
->dev
[i
].flags
)) {
1834 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
1835 if (rdev
&& test_bit(In_sync
, &rdev
->flags
))
1836 /* multiple read failures in one stripe */
1837 md_error(conf
->mddev
, rdev
);
1840 spin_lock_irq(&conf
->device_lock
);
1841 /* fail all writes first */
1842 bi
= sh
->dev
[i
].towrite
;
1843 sh
->dev
[i
].towrite
= NULL
;
1849 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
1850 wake_up(&conf
->wait_for_overlap
);
1852 while (bi
&& bi
->bi_sector
<
1853 sh
->dev
[i
].sector
+ STRIPE_SECTORS
) {
1854 struct bio
*nextbi
= r5_next_bio(bi
, sh
->dev
[i
].sector
);
1855 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1856 if (!raid5_dec_bi_phys_segments(bi
)) {
1857 md_write_end(conf
->mddev
);
1858 bi
->bi_next
= *return_bi
;
1863 /* and fail all 'written' */
1864 bi
= sh
->dev
[i
].written
;
1865 sh
->dev
[i
].written
= NULL
;
1866 if (bi
) bitmap_end
= 1;
1867 while (bi
&& bi
->bi_sector
<
1868 sh
->dev
[i
].sector
+ STRIPE_SECTORS
) {
1869 struct bio
*bi2
= r5_next_bio(bi
, sh
->dev
[i
].sector
);
1870 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1871 if (!raid5_dec_bi_phys_segments(bi
)) {
1872 md_write_end(conf
->mddev
);
1873 bi
->bi_next
= *return_bi
;
1879 /* fail any reads if this device is non-operational and
1880 * the data has not reached the cache yet.
1882 if (!test_bit(R5_Wantfill
, &sh
->dev
[i
].flags
) &&
1883 (!test_bit(R5_Insync
, &sh
->dev
[i
].flags
) ||
1884 test_bit(R5_ReadError
, &sh
->dev
[i
].flags
))) {
1885 bi
= sh
->dev
[i
].toread
;
1886 sh
->dev
[i
].toread
= NULL
;
1887 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
1888 wake_up(&conf
->wait_for_overlap
);
1889 if (bi
) s
->to_read
--;
1890 while (bi
&& bi
->bi_sector
<
1891 sh
->dev
[i
].sector
+ STRIPE_SECTORS
) {
1892 struct bio
*nextbi
=
1893 r5_next_bio(bi
, sh
->dev
[i
].sector
);
1894 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1895 if (!raid5_dec_bi_phys_segments(bi
)) {
1896 bi
->bi_next
= *return_bi
;
1902 spin_unlock_irq(&conf
->device_lock
);
1904 bitmap_endwrite(conf
->mddev
->bitmap
, sh
->sector
,
1905 STRIPE_SECTORS
, 0, 0);
1908 if (test_and_clear_bit(STRIPE_FULL_WRITE
, &sh
->state
))
1909 if (atomic_dec_and_test(&conf
->pending_full_writes
))
1910 md_wakeup_thread(conf
->mddev
->thread
);
1913 /* fetch_block5 - checks the given member device to see if its data needs
1914 * to be read or computed to satisfy a request.
1916 * Returns 1 when no more member devices need to be checked, otherwise returns
1917 * 0 to tell the loop in handle_stripe_fill5 to continue
1919 static int fetch_block5(struct stripe_head
*sh
, struct stripe_head_state
*s
,
1920 int disk_idx
, int disks
)
1922 struct r5dev
*dev
= &sh
->dev
[disk_idx
];
1923 struct r5dev
*failed_dev
= &sh
->dev
[s
->failed_num
];
1925 /* is the data in this block needed, and can we get it? */
1926 if (!test_bit(R5_LOCKED
, &dev
->flags
) &&
1927 !test_bit(R5_UPTODATE
, &dev
->flags
) &&
1929 (dev
->towrite
&& !test_bit(R5_OVERWRITE
, &dev
->flags
)) ||
1930 s
->syncing
|| s
->expanding
||
1932 (failed_dev
->toread
||
1933 (failed_dev
->towrite
&&
1934 !test_bit(R5_OVERWRITE
, &failed_dev
->flags
)))))) {
1935 /* We would like to get this block, possibly by computing it,
1936 * otherwise read it if the backing disk is insync
1938 if ((s
->uptodate
== disks
- 1) &&
1939 (s
->failed
&& disk_idx
== s
->failed_num
)) {
1940 set_bit(STRIPE_COMPUTE_RUN
, &sh
->state
);
1941 set_bit(STRIPE_OP_COMPUTE_BLK
, &s
->ops_request
);
1942 set_bit(R5_Wantcompute
, &dev
->flags
);
1943 sh
->ops
.target
= disk_idx
;
1945 /* Careful: from this point on 'uptodate' is in the eye
1946 * of raid5_run_ops which services 'compute' operations
1947 * before writes. R5_Wantcompute flags a block that will
1948 * be R5_UPTODATE by the time it is needed for a
1949 * subsequent operation.
1952 return 1; /* uptodate + compute == disks */
1953 } else if (test_bit(R5_Insync
, &dev
->flags
)) {
1954 set_bit(R5_LOCKED
, &dev
->flags
);
1955 set_bit(R5_Wantread
, &dev
->flags
);
1957 pr_debug("Reading block %d (sync=%d)\n", disk_idx
,
1966 * handle_stripe_fill5 - read or compute data to satisfy pending requests.
1968 static void handle_stripe_fill5(struct stripe_head
*sh
,
1969 struct stripe_head_state
*s
, int disks
)
1973 /* look for blocks to read/compute, skip this if a compute
1974 * is already in flight, or if the stripe contents are in the
1975 * midst of changing due to a write
1977 if (!test_bit(STRIPE_COMPUTE_RUN
, &sh
->state
) && !sh
->check_state
&&
1978 !sh
->reconstruct_state
)
1979 for (i
= disks
; i
--; )
1980 if (fetch_block5(sh
, s
, i
, disks
))
1982 set_bit(STRIPE_HANDLE
, &sh
->state
);
1985 static void handle_stripe_fill6(struct stripe_head
*sh
,
1986 struct stripe_head_state
*s
, struct r6_state
*r6s
,
1990 for (i
= disks
; i
--; ) {
1991 struct r5dev
*dev
= &sh
->dev
[i
];
1992 if (!test_bit(R5_LOCKED
, &dev
->flags
) &&
1993 !test_bit(R5_UPTODATE
, &dev
->flags
) &&
1994 (dev
->toread
|| (dev
->towrite
&&
1995 !test_bit(R5_OVERWRITE
, &dev
->flags
)) ||
1996 s
->syncing
|| s
->expanding
||
1998 (sh
->dev
[r6s
->failed_num
[0]].toread
||
2001 (sh
->dev
[r6s
->failed_num
[1]].toread
||
2003 /* we would like to get this block, possibly
2004 * by computing it, but we might not be able to
2006 if ((s
->uptodate
== disks
- 1) &&
2007 (s
->failed
&& (i
== r6s
->failed_num
[0] ||
2008 i
== r6s
->failed_num
[1]))) {
2009 pr_debug("Computing stripe %llu block %d\n",
2010 (unsigned long long)sh
->sector
, i
);
2011 compute_block_1(sh
, i
, 0);
2013 } else if ( s
->uptodate
== disks
-2 && s
->failed
>= 2 ) {
2014 /* Computing 2-failure is *very* expensive; only
2015 * do it if failed >= 2
2018 for (other
= disks
; other
--; ) {
2021 if (!test_bit(R5_UPTODATE
,
2022 &sh
->dev
[other
].flags
))
2026 pr_debug("Computing stripe %llu blocks %d,%d\n",
2027 (unsigned long long)sh
->sector
,
2029 compute_block_2(sh
, i
, other
);
2031 } else if (test_bit(R5_Insync
, &dev
->flags
)) {
2032 set_bit(R5_LOCKED
, &dev
->flags
);
2033 set_bit(R5_Wantread
, &dev
->flags
);
2035 pr_debug("Reading block %d (sync=%d)\n",
2040 set_bit(STRIPE_HANDLE
, &sh
->state
);
2044 /* handle_stripe_clean_event
2045 * any written block on an uptodate or failed drive can be returned.
2046 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
2047 * never LOCKED, so we don't need to test 'failed' directly.
2049 static void handle_stripe_clean_event(raid5_conf_t
*conf
,
2050 struct stripe_head
*sh
, int disks
, struct bio
**return_bi
)
2055 for (i
= disks
; i
--; )
2056 if (sh
->dev
[i
].written
) {
2058 if (!test_bit(R5_LOCKED
, &dev
->flags
) &&
2059 test_bit(R5_UPTODATE
, &dev
->flags
)) {
2060 /* We can return any write requests */
2061 struct bio
*wbi
, *wbi2
;
2063 pr_debug("Return write for disc %d\n", i
);
2064 spin_lock_irq(&conf
->device_lock
);
2066 dev
->written
= NULL
;
2067 while (wbi
&& wbi
->bi_sector
<
2068 dev
->sector
+ STRIPE_SECTORS
) {
2069 wbi2
= r5_next_bio(wbi
, dev
->sector
);
2070 if (!raid5_dec_bi_phys_segments(wbi
)) {
2071 md_write_end(conf
->mddev
);
2072 wbi
->bi_next
= *return_bi
;
2077 if (dev
->towrite
== NULL
)
2079 spin_unlock_irq(&conf
->device_lock
);
2081 bitmap_endwrite(conf
->mddev
->bitmap
,
2084 !test_bit(STRIPE_DEGRADED
, &sh
->state
),
2089 if (test_and_clear_bit(STRIPE_FULL_WRITE
, &sh
->state
))
2090 if (atomic_dec_and_test(&conf
->pending_full_writes
))
2091 md_wakeup_thread(conf
->mddev
->thread
);
2094 static void handle_stripe_dirtying5(raid5_conf_t
*conf
,
2095 struct stripe_head
*sh
, struct stripe_head_state
*s
, int disks
)
2097 int rmw
= 0, rcw
= 0, i
;
2098 for (i
= disks
; i
--; ) {
2099 /* would I have to read this buffer for read_modify_write */
2100 struct r5dev
*dev
= &sh
->dev
[i
];
2101 if ((dev
->towrite
|| i
== sh
->pd_idx
) &&
2102 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2103 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2104 test_bit(R5_Wantcompute
, &dev
->flags
))) {
2105 if (test_bit(R5_Insync
, &dev
->flags
))
2108 rmw
+= 2*disks
; /* cannot read it */
2110 /* Would I have to read this buffer for reconstruct_write */
2111 if (!test_bit(R5_OVERWRITE
, &dev
->flags
) && i
!= sh
->pd_idx
&&
2112 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2113 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2114 test_bit(R5_Wantcompute
, &dev
->flags
))) {
2115 if (test_bit(R5_Insync
, &dev
->flags
)) rcw
++;
2120 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
2121 (unsigned long long)sh
->sector
, rmw
, rcw
);
2122 set_bit(STRIPE_HANDLE
, &sh
->state
);
2123 if (rmw
< rcw
&& rmw
> 0)
2124 /* prefer read-modify-write, but need to get some data */
2125 for (i
= disks
; i
--; ) {
2126 struct r5dev
*dev
= &sh
->dev
[i
];
2127 if ((dev
->towrite
|| i
== sh
->pd_idx
) &&
2128 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2129 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2130 test_bit(R5_Wantcompute
, &dev
->flags
)) &&
2131 test_bit(R5_Insync
, &dev
->flags
)) {
2133 test_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2134 pr_debug("Read_old block "
2135 "%d for r-m-w\n", i
);
2136 set_bit(R5_LOCKED
, &dev
->flags
);
2137 set_bit(R5_Wantread
, &dev
->flags
);
2140 set_bit(STRIPE_DELAYED
, &sh
->state
);
2141 set_bit(STRIPE_HANDLE
, &sh
->state
);
2145 if (rcw
<= rmw
&& rcw
> 0)
2146 /* want reconstruct write, but need to get some data */
2147 for (i
= disks
; i
--; ) {
2148 struct r5dev
*dev
= &sh
->dev
[i
];
2149 if (!test_bit(R5_OVERWRITE
, &dev
->flags
) &&
2151 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2152 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2153 test_bit(R5_Wantcompute
, &dev
->flags
)) &&
2154 test_bit(R5_Insync
, &dev
->flags
)) {
2156 test_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2157 pr_debug("Read_old block "
2158 "%d for Reconstruct\n", i
);
2159 set_bit(R5_LOCKED
, &dev
->flags
);
2160 set_bit(R5_Wantread
, &dev
->flags
);
2163 set_bit(STRIPE_DELAYED
, &sh
->state
);
2164 set_bit(STRIPE_HANDLE
, &sh
->state
);
2168 /* now if nothing is locked, and if we have enough data,
2169 * we can start a write request
2171 /* since handle_stripe can be called at any time we need to handle the
2172 * case where a compute block operation has been submitted and then a
2173 * subsequent call wants to start a write request. raid5_run_ops only
2174 * handles the case where compute block and postxor are requested
2175 * simultaneously. If this is not the case then new writes need to be
2176 * held off until the compute completes.
2178 if ((s
->req_compute
|| !test_bit(STRIPE_COMPUTE_RUN
, &sh
->state
)) &&
2179 (s
->locked
== 0 && (rcw
== 0 || rmw
== 0) &&
2180 !test_bit(STRIPE_BIT_DELAY
, &sh
->state
)))
2181 schedule_reconstruction5(sh
, s
, rcw
== 0, 0);
2184 static void handle_stripe_dirtying6(raid5_conf_t
*conf
,
2185 struct stripe_head
*sh
, struct stripe_head_state
*s
,
2186 struct r6_state
*r6s
, int disks
)
2188 int rcw
= 0, must_compute
= 0, pd_idx
= sh
->pd_idx
, i
;
2189 int qd_idx
= r6s
->qd_idx
;
2190 for (i
= disks
; i
--; ) {
2191 struct r5dev
*dev
= &sh
->dev
[i
];
2192 /* Would I have to read this buffer for reconstruct_write */
2193 if (!test_bit(R5_OVERWRITE
, &dev
->flags
)
2194 && i
!= pd_idx
&& i
!= qd_idx
2195 && (!test_bit(R5_LOCKED
, &dev
->flags
)
2197 !test_bit(R5_UPTODATE
, &dev
->flags
)) {
2198 if (test_bit(R5_Insync
, &dev
->flags
)) rcw
++;
2200 pr_debug("raid6: must_compute: "
2201 "disk %d flags=%#lx\n", i
, dev
->flags
);
2206 pr_debug("for sector %llu, rcw=%d, must_compute=%d\n",
2207 (unsigned long long)sh
->sector
, rcw
, must_compute
);
2208 set_bit(STRIPE_HANDLE
, &sh
->state
);
2211 /* want reconstruct write, but need to get some data */
2212 for (i
= disks
; i
--; ) {
2213 struct r5dev
*dev
= &sh
->dev
[i
];
2214 if (!test_bit(R5_OVERWRITE
, &dev
->flags
)
2215 && !(s
->failed
== 0 && (i
== pd_idx
|| i
== qd_idx
))
2216 && !test_bit(R5_LOCKED
, &dev
->flags
) &&
2217 !test_bit(R5_UPTODATE
, &dev
->flags
) &&
2218 test_bit(R5_Insync
, &dev
->flags
)) {
2220 test_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2221 pr_debug("Read_old stripe %llu "
2222 "block %d for Reconstruct\n",
2223 (unsigned long long)sh
->sector
, i
);
2224 set_bit(R5_LOCKED
, &dev
->flags
);
2225 set_bit(R5_Wantread
, &dev
->flags
);
2228 pr_debug("Request delayed stripe %llu "
2229 "block %d for Reconstruct\n",
2230 (unsigned long long)sh
->sector
, i
);
2231 set_bit(STRIPE_DELAYED
, &sh
->state
);
2232 set_bit(STRIPE_HANDLE
, &sh
->state
);
2236 /* now if nothing is locked, and if we have enough data, we can start a
2239 if (s
->locked
== 0 && rcw
== 0 &&
2240 !test_bit(STRIPE_BIT_DELAY
, &sh
->state
)) {
2241 if (must_compute
> 0) {
2242 /* We have failed blocks and need to compute them */
2243 switch (s
->failed
) {
2247 compute_block_1(sh
, r6s
->failed_num
[0], 0);
2250 compute_block_2(sh
, r6s
->failed_num
[0],
2251 r6s
->failed_num
[1]);
2253 default: /* This request should have been failed? */
2258 pr_debug("Computing parity for stripe %llu\n",
2259 (unsigned long long)sh
->sector
);
2260 compute_parity6(sh
, RECONSTRUCT_WRITE
);
2261 /* now every locked buffer is ready to be written */
2262 for (i
= disks
; i
--; )
2263 if (test_bit(R5_LOCKED
, &sh
->dev
[i
].flags
)) {
2264 pr_debug("Writing stripe %llu block %d\n",
2265 (unsigned long long)sh
->sector
, i
);
2267 set_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
);
2269 if (s
->locked
== disks
)
2270 if (!test_and_set_bit(STRIPE_FULL_WRITE
, &sh
->state
))
2271 atomic_inc(&conf
->pending_full_writes
);
2272 /* after a RECONSTRUCT_WRITE, the stripe MUST be in-sync */
2273 set_bit(STRIPE_INSYNC
, &sh
->state
);
2275 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2276 atomic_dec(&conf
->preread_active_stripes
);
2277 if (atomic_read(&conf
->preread_active_stripes
) <
2279 md_wakeup_thread(conf
->mddev
->thread
);
2284 static void handle_parity_checks5(raid5_conf_t
*conf
, struct stripe_head
*sh
,
2285 struct stripe_head_state
*s
, int disks
)
2287 struct r5dev
*dev
= NULL
;
2289 set_bit(STRIPE_HANDLE
, &sh
->state
);
2291 switch (sh
->check_state
) {
2292 case check_state_idle
:
2293 /* start a new check operation if there are no failures */
2294 if (s
->failed
== 0) {
2295 BUG_ON(s
->uptodate
!= disks
);
2296 sh
->check_state
= check_state_run
;
2297 set_bit(STRIPE_OP_CHECK
, &s
->ops_request
);
2298 clear_bit(R5_UPTODATE
, &sh
->dev
[sh
->pd_idx
].flags
);
2302 dev
= &sh
->dev
[s
->failed_num
];
2304 case check_state_compute_result
:
2305 sh
->check_state
= check_state_idle
;
2307 dev
= &sh
->dev
[sh
->pd_idx
];
2309 /* check that a write has not made the stripe insync */
2310 if (test_bit(STRIPE_INSYNC
, &sh
->state
))
2313 /* either failed parity check, or recovery is happening */
2314 BUG_ON(!test_bit(R5_UPTODATE
, &dev
->flags
));
2315 BUG_ON(s
->uptodate
!= disks
);
2317 set_bit(R5_LOCKED
, &dev
->flags
);
2319 set_bit(R5_Wantwrite
, &dev
->flags
);
2321 clear_bit(STRIPE_DEGRADED
, &sh
->state
);
2322 set_bit(STRIPE_INSYNC
, &sh
->state
);
2324 case check_state_run
:
2325 break; /* we will be called again upon completion */
2326 case check_state_check_result
:
2327 sh
->check_state
= check_state_idle
;
2329 /* if a failure occurred during the check operation, leave
2330 * STRIPE_INSYNC not set and let the stripe be handled again
2335 /* handle a successful check operation, if parity is correct
2336 * we are done. Otherwise update the mismatch count and repair
2337 * parity if !MD_RECOVERY_CHECK
2339 if (sh
->ops
.zero_sum_result
== 0)
2340 /* parity is correct (on disc,
2341 * not in buffer any more)
2343 set_bit(STRIPE_INSYNC
, &sh
->state
);
2345 conf
->mddev
->resync_mismatches
+= STRIPE_SECTORS
;
2346 if (test_bit(MD_RECOVERY_CHECK
, &conf
->mddev
->recovery
))
2347 /* don't try to repair!! */
2348 set_bit(STRIPE_INSYNC
, &sh
->state
);
2350 sh
->check_state
= check_state_compute_run
;
2351 set_bit(STRIPE_COMPUTE_RUN
, &sh
->state
);
2352 set_bit(STRIPE_OP_COMPUTE_BLK
, &s
->ops_request
);
2353 set_bit(R5_Wantcompute
,
2354 &sh
->dev
[sh
->pd_idx
].flags
);
2355 sh
->ops
.target
= sh
->pd_idx
;
2360 case check_state_compute_run
:
2363 printk(KERN_ERR
"%s: unknown check_state: %d sector: %llu\n",
2364 __func__
, sh
->check_state
,
2365 (unsigned long long) sh
->sector
);
2371 static void handle_parity_checks6(raid5_conf_t
*conf
, struct stripe_head
*sh
,
2372 struct stripe_head_state
*s
,
2373 struct r6_state
*r6s
, struct page
*tmp_page
,
2376 int update_p
= 0, update_q
= 0;
2378 int pd_idx
= sh
->pd_idx
;
2379 int qd_idx
= r6s
->qd_idx
;
2381 set_bit(STRIPE_HANDLE
, &sh
->state
);
2383 BUG_ON(s
->failed
> 2);
2384 BUG_ON(s
->uptodate
< disks
);
2385 /* Want to check and possibly repair P and Q.
2386 * However there could be one 'failed' device, in which
2387 * case we can only check one of them, possibly using the
2388 * other to generate missing data
2391 /* If !tmp_page, we cannot do the calculations,
2392 * but as we have set STRIPE_HANDLE, we will soon be called
2393 * by stripe_handle with a tmp_page - just wait until then.
2396 if (s
->failed
== r6s
->q_failed
) {
2397 /* The only possible failed device holds 'Q', so it
2398 * makes sense to check P (If anything else were failed,
2399 * we would have used P to recreate it).
2401 compute_block_1(sh
, pd_idx
, 1);
2402 if (!page_is_zero(sh
->dev
[pd_idx
].page
)) {
2403 compute_block_1(sh
, pd_idx
, 0);
2407 if (!r6s
->q_failed
&& s
->failed
< 2) {
2408 /* q is not failed, and we didn't use it to generate
2409 * anything, so it makes sense to check it
2411 memcpy(page_address(tmp_page
),
2412 page_address(sh
->dev
[qd_idx
].page
),
2414 compute_parity6(sh
, UPDATE_PARITY
);
2415 if (memcmp(page_address(tmp_page
),
2416 page_address(sh
->dev
[qd_idx
].page
),
2417 STRIPE_SIZE
) != 0) {
2418 clear_bit(STRIPE_INSYNC
, &sh
->state
);
2422 if (update_p
|| update_q
) {
2423 conf
->mddev
->resync_mismatches
+= STRIPE_SECTORS
;
2424 if (test_bit(MD_RECOVERY_CHECK
, &conf
->mddev
->recovery
))
2425 /* don't try to repair!! */
2426 update_p
= update_q
= 0;
2429 /* now write out any block on a failed drive,
2430 * or P or Q if they need it
2433 if (s
->failed
== 2) {
2434 dev
= &sh
->dev
[r6s
->failed_num
[1]];
2436 set_bit(R5_LOCKED
, &dev
->flags
);
2437 set_bit(R5_Wantwrite
, &dev
->flags
);
2439 if (s
->failed
>= 1) {
2440 dev
= &sh
->dev
[r6s
->failed_num
[0]];
2442 set_bit(R5_LOCKED
, &dev
->flags
);
2443 set_bit(R5_Wantwrite
, &dev
->flags
);
2447 dev
= &sh
->dev
[pd_idx
];
2449 set_bit(R5_LOCKED
, &dev
->flags
);
2450 set_bit(R5_Wantwrite
, &dev
->flags
);
2453 dev
= &sh
->dev
[qd_idx
];
2455 set_bit(R5_LOCKED
, &dev
->flags
);
2456 set_bit(R5_Wantwrite
, &dev
->flags
);
2458 clear_bit(STRIPE_DEGRADED
, &sh
->state
);
2460 set_bit(STRIPE_INSYNC
, &sh
->state
);
2464 static void handle_stripe_expansion(raid5_conf_t
*conf
, struct stripe_head
*sh
,
2465 struct r6_state
*r6s
)
2469 /* We have read all the blocks in this stripe and now we need to
2470 * copy some of them into a target stripe for expand.
2472 struct dma_async_tx_descriptor
*tx
= NULL
;
2473 clear_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
2474 for (i
= 0; i
< sh
->disks
; i
++)
2475 if (i
!= sh
->pd_idx
&& (!r6s
|| i
!= r6s
->qd_idx
)) {
2476 int dd_idx
, pd_idx
, j
;
2477 struct stripe_head
*sh2
;
2479 sector_t bn
= compute_blocknr(sh
, i
);
2480 sector_t s
= raid5_compute_sector(bn
, conf
->raid_disks
,
2482 conf
->max_degraded
, &dd_idx
,
2484 sh2
= get_active_stripe(conf
, s
, conf
->raid_disks
,
2487 /* so far only the early blocks of this stripe
2488 * have been requested. When later blocks
2489 * get requested, we will try again
2492 if (!test_bit(STRIPE_EXPANDING
, &sh2
->state
) ||
2493 test_bit(R5_Expanded
, &sh2
->dev
[dd_idx
].flags
)) {
2494 /* must have already done this block */
2495 release_stripe(sh2
);
2499 /* place all the copies on one channel */
2500 tx
= async_memcpy(sh2
->dev
[dd_idx
].page
,
2501 sh
->dev
[i
].page
, 0, 0, STRIPE_SIZE
,
2502 ASYNC_TX_DEP_ACK
, tx
, NULL
, NULL
);
2504 set_bit(R5_Expanded
, &sh2
->dev
[dd_idx
].flags
);
2505 set_bit(R5_UPTODATE
, &sh2
->dev
[dd_idx
].flags
);
2506 for (j
= 0; j
< conf
->raid_disks
; j
++)
2507 if (j
!= sh2
->pd_idx
&&
2508 (!r6s
|| j
!= raid6_next_disk(sh2
->pd_idx
,
2510 !test_bit(R5_Expanded
, &sh2
->dev
[j
].flags
))
2512 if (j
== conf
->raid_disks
) {
2513 set_bit(STRIPE_EXPAND_READY
, &sh2
->state
);
2514 set_bit(STRIPE_HANDLE
, &sh2
->state
);
2516 release_stripe(sh2
);
2519 /* done submitting copies, wait for them to complete */
2522 dma_wait_for_async_tx(tx
);
2528 * handle_stripe - do things to a stripe.
2530 * We lock the stripe and then examine the state of various bits
2531 * to see what needs to be done.
2533 * return some read request which now have data
2534 * return some write requests which are safely on disc
2535 * schedule a read on some buffers
2536 * schedule a write of some buffers
2537 * return confirmation of parity correctness
2539 * buffers are taken off read_list or write_list, and bh_cache buffers
2540 * get BH_Lock set before the stripe lock is released.
2544 static bool handle_stripe5(struct stripe_head
*sh
)
2546 raid5_conf_t
*conf
= sh
->raid_conf
;
2547 int disks
= sh
->disks
, i
;
2548 struct bio
*return_bi
= NULL
;
2549 struct stripe_head_state s
;
2551 mdk_rdev_t
*blocked_rdev
= NULL
;
2554 memset(&s
, 0, sizeof(s
));
2555 pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d check:%d "
2556 "reconstruct:%d\n", (unsigned long long)sh
->sector
, sh
->state
,
2557 atomic_read(&sh
->count
), sh
->pd_idx
, sh
->check_state
,
2558 sh
->reconstruct_state
);
2560 spin_lock(&sh
->lock
);
2561 clear_bit(STRIPE_HANDLE
, &sh
->state
);
2562 clear_bit(STRIPE_DELAYED
, &sh
->state
);
2564 s
.syncing
= test_bit(STRIPE_SYNCING
, &sh
->state
);
2565 s
.expanding
= test_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
2566 s
.expanded
= test_bit(STRIPE_EXPAND_READY
, &sh
->state
);
2568 /* Now to look around and see what can be done */
2570 for (i
=disks
; i
--; ) {
2572 struct r5dev
*dev
= &sh
->dev
[i
];
2573 clear_bit(R5_Insync
, &dev
->flags
);
2575 pr_debug("check %d: state 0x%lx toread %p read %p write %p "
2576 "written %p\n", i
, dev
->flags
, dev
->toread
, dev
->read
,
2577 dev
->towrite
, dev
->written
);
2579 /* maybe we can request a biofill operation
2581 * new wantfill requests are only permitted while
2582 * ops_complete_biofill is guaranteed to be inactive
2584 if (test_bit(R5_UPTODATE
, &dev
->flags
) && dev
->toread
&&
2585 !test_bit(STRIPE_BIOFILL_RUN
, &sh
->state
))
2586 set_bit(R5_Wantfill
, &dev
->flags
);
2588 /* now count some things */
2589 if (test_bit(R5_LOCKED
, &dev
->flags
)) s
.locked
++;
2590 if (test_bit(R5_UPTODATE
, &dev
->flags
)) s
.uptodate
++;
2591 if (test_bit(R5_Wantcompute
, &dev
->flags
)) s
.compute
++;
2593 if (test_bit(R5_Wantfill
, &dev
->flags
))
2595 else if (dev
->toread
)
2599 if (!test_bit(R5_OVERWRITE
, &dev
->flags
))
2604 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
2605 if (blocked_rdev
== NULL
&&
2606 rdev
&& unlikely(test_bit(Blocked
, &rdev
->flags
))) {
2607 blocked_rdev
= rdev
;
2608 atomic_inc(&rdev
->nr_pending
);
2610 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)) {
2611 /* The ReadError flag will just be confusing now */
2612 clear_bit(R5_ReadError
, &dev
->flags
);
2613 clear_bit(R5_ReWrite
, &dev
->flags
);
2615 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)
2616 || test_bit(R5_ReadError
, &dev
->flags
)) {
2620 set_bit(R5_Insync
, &dev
->flags
);
2624 if (unlikely(blocked_rdev
)) {
2625 if (s
.syncing
|| s
.expanding
|| s
.expanded
||
2626 s
.to_write
|| s
.written
) {
2627 set_bit(STRIPE_HANDLE
, &sh
->state
);
2630 /* There is nothing for the blocked_rdev to block */
2631 rdev_dec_pending(blocked_rdev
, conf
->mddev
);
2632 blocked_rdev
= NULL
;
2635 if (s
.to_fill
&& !test_bit(STRIPE_BIOFILL_RUN
, &sh
->state
)) {
2636 set_bit(STRIPE_OP_BIOFILL
, &s
.ops_request
);
2637 set_bit(STRIPE_BIOFILL_RUN
, &sh
->state
);
2640 pr_debug("locked=%d uptodate=%d to_read=%d"
2641 " to_write=%d failed=%d failed_num=%d\n",
2642 s
.locked
, s
.uptodate
, s
.to_read
, s
.to_write
,
2643 s
.failed
, s
.failed_num
);
2644 /* check if the array has lost two devices and, if so, some requests might
2647 if (s
.failed
> 1 && s
.to_read
+s
.to_write
+s
.written
)
2648 handle_failed_stripe(conf
, sh
, &s
, disks
, &return_bi
);
2649 if (s
.failed
> 1 && s
.syncing
) {
2650 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,0);
2651 clear_bit(STRIPE_SYNCING
, &sh
->state
);
2655 /* might be able to return some write requests if the parity block
2656 * is safe, or on a failed drive
2658 dev
= &sh
->dev
[sh
->pd_idx
];
2660 ((test_bit(R5_Insync
, &dev
->flags
) &&
2661 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2662 test_bit(R5_UPTODATE
, &dev
->flags
)) ||
2663 (s
.failed
== 1 && s
.failed_num
== sh
->pd_idx
)))
2664 handle_stripe_clean_event(conf
, sh
, disks
, &return_bi
);
2666 /* Now we might consider reading some blocks, either to check/generate
2667 * parity, or to satisfy requests
2668 * or to load a block that is being partially written.
2670 if (s
.to_read
|| s
.non_overwrite
||
2671 (s
.syncing
&& (s
.uptodate
+ s
.compute
< disks
)) || s
.expanding
)
2672 handle_stripe_fill5(sh
, &s
, disks
);
2674 /* Now we check to see if any write operations have recently
2678 if (sh
->reconstruct_state
== reconstruct_state_prexor_drain_result
)
2680 if (sh
->reconstruct_state
== reconstruct_state_drain_result
||
2681 sh
->reconstruct_state
== reconstruct_state_prexor_drain_result
) {
2682 sh
->reconstruct_state
= reconstruct_state_idle
;
2684 /* All the 'written' buffers and the parity block are ready to
2685 * be written back to disk
2687 BUG_ON(!test_bit(R5_UPTODATE
, &sh
->dev
[sh
->pd_idx
].flags
));
2688 for (i
= disks
; i
--; ) {
2690 if (test_bit(R5_LOCKED
, &dev
->flags
) &&
2691 (i
== sh
->pd_idx
|| dev
->written
)) {
2692 pr_debug("Writing block %d\n", i
);
2693 set_bit(R5_Wantwrite
, &dev
->flags
);
2696 if (!test_bit(R5_Insync
, &dev
->flags
) ||
2697 (i
== sh
->pd_idx
&& s
.failed
== 0))
2698 set_bit(STRIPE_INSYNC
, &sh
->state
);
2701 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2702 atomic_dec(&conf
->preread_active_stripes
);
2703 if (atomic_read(&conf
->preread_active_stripes
) <
2705 md_wakeup_thread(conf
->mddev
->thread
);
2709 /* Now to consider new write requests and what else, if anything
2710 * should be read. We do not handle new writes when:
2711 * 1/ A 'write' operation (copy+xor) is already in flight.
2712 * 2/ A 'check' operation is in flight, as it may clobber the parity
2715 if (s
.to_write
&& !sh
->reconstruct_state
&& !sh
->check_state
)
2716 handle_stripe_dirtying5(conf
, sh
, &s
, disks
);
2718 /* maybe we need to check and possibly fix the parity for this stripe
2719 * Any reads will already have been scheduled, so we just see if enough
2720 * data is available. The parity check is held off while parity
2721 * dependent operations are in flight.
2723 if (sh
->check_state
||
2724 (s
.syncing
&& s
.locked
== 0 &&
2725 !test_bit(STRIPE_COMPUTE_RUN
, &sh
->state
) &&
2726 !test_bit(STRIPE_INSYNC
, &sh
->state
)))
2727 handle_parity_checks5(conf
, sh
, &s
, disks
);
2729 if (s
.syncing
&& s
.locked
== 0 && test_bit(STRIPE_INSYNC
, &sh
->state
)) {
2730 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,1);
2731 clear_bit(STRIPE_SYNCING
, &sh
->state
);
2734 /* If the failed drive is just a ReadError, then we might need to progress
2735 * the repair/check process
2737 if (s
.failed
== 1 && !conf
->mddev
->ro
&&
2738 test_bit(R5_ReadError
, &sh
->dev
[s
.failed_num
].flags
)
2739 && !test_bit(R5_LOCKED
, &sh
->dev
[s
.failed_num
].flags
)
2740 && test_bit(R5_UPTODATE
, &sh
->dev
[s
.failed_num
].flags
)
2742 dev
= &sh
->dev
[s
.failed_num
];
2743 if (!test_bit(R5_ReWrite
, &dev
->flags
)) {
2744 set_bit(R5_Wantwrite
, &dev
->flags
);
2745 set_bit(R5_ReWrite
, &dev
->flags
);
2746 set_bit(R5_LOCKED
, &dev
->flags
);
2749 /* let's read it back */
2750 set_bit(R5_Wantread
, &dev
->flags
);
2751 set_bit(R5_LOCKED
, &dev
->flags
);
2756 /* Finish reconstruct operations initiated by the expansion process */
2757 if (sh
->reconstruct_state
== reconstruct_state_result
) {
2758 sh
->reconstruct_state
= reconstruct_state_idle
;
2759 clear_bit(STRIPE_EXPANDING
, &sh
->state
);
2760 for (i
= conf
->raid_disks
; i
--; ) {
2761 set_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
);
2762 set_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
2767 if (s
.expanded
&& test_bit(STRIPE_EXPANDING
, &sh
->state
) &&
2768 !sh
->reconstruct_state
) {
2769 /* Need to write out all blocks after computing parity */
2770 sh
->disks
= conf
->raid_disks
;
2771 sh
->pd_idx
= stripe_to_pdidx(sh
->sector
, conf
,
2773 schedule_reconstruction5(sh
, &s
, 1, 1);
2774 } else if (s
.expanded
&& !sh
->reconstruct_state
&& s
.locked
== 0) {
2775 clear_bit(STRIPE_EXPAND_READY
, &sh
->state
);
2776 atomic_dec(&conf
->reshape_stripes
);
2777 wake_up(&conf
->wait_for_overlap
);
2778 md_done_sync(conf
->mddev
, STRIPE_SECTORS
, 1);
2781 if (s
.expanding
&& s
.locked
== 0 &&
2782 !test_bit(STRIPE_COMPUTE_RUN
, &sh
->state
))
2783 handle_stripe_expansion(conf
, sh
, NULL
);
2786 spin_unlock(&sh
->lock
);
2788 /* wait for this device to become unblocked */
2789 if (unlikely(blocked_rdev
))
2790 md_wait_for_blocked_rdev(blocked_rdev
, conf
->mddev
);
2793 raid5_run_ops(sh
, s
.ops_request
);
2797 return_io(return_bi
);
2799 return blocked_rdev
== NULL
;
2802 static bool handle_stripe6(struct stripe_head
*sh
, struct page
*tmp_page
)
2804 raid6_conf_t
*conf
= sh
->raid_conf
;
2805 int disks
= sh
->disks
;
2806 struct bio
*return_bi
= NULL
;
2807 int i
, pd_idx
= sh
->pd_idx
;
2808 struct stripe_head_state s
;
2809 struct r6_state r6s
;
2810 struct r5dev
*dev
, *pdev
, *qdev
;
2811 mdk_rdev_t
*blocked_rdev
= NULL
;
2813 r6s
.qd_idx
= raid6_next_disk(pd_idx
, disks
);
2814 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
2815 "pd_idx=%d, qd_idx=%d\n",
2816 (unsigned long long)sh
->sector
, sh
->state
,
2817 atomic_read(&sh
->count
), pd_idx
, r6s
.qd_idx
);
2818 memset(&s
, 0, sizeof(s
));
2820 spin_lock(&sh
->lock
);
2821 clear_bit(STRIPE_HANDLE
, &sh
->state
);
2822 clear_bit(STRIPE_DELAYED
, &sh
->state
);
2824 s
.syncing
= test_bit(STRIPE_SYNCING
, &sh
->state
);
2825 s
.expanding
= test_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
2826 s
.expanded
= test_bit(STRIPE_EXPAND_READY
, &sh
->state
);
2827 /* Now to look around and see what can be done */
2830 for (i
=disks
; i
--; ) {
2833 clear_bit(R5_Insync
, &dev
->flags
);
2835 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
2836 i
, dev
->flags
, dev
->toread
, dev
->towrite
, dev
->written
);
2837 /* maybe we can reply to a read */
2838 if (test_bit(R5_UPTODATE
, &dev
->flags
) && dev
->toread
) {
2839 struct bio
*rbi
, *rbi2
;
2840 pr_debug("Return read for disc %d\n", i
);
2841 spin_lock_irq(&conf
->device_lock
);
2844 if (test_and_clear_bit(R5_Overlap
, &dev
->flags
))
2845 wake_up(&conf
->wait_for_overlap
);
2846 spin_unlock_irq(&conf
->device_lock
);
2847 while (rbi
&& rbi
->bi_sector
< dev
->sector
+ STRIPE_SECTORS
) {
2848 copy_data(0, rbi
, dev
->page
, dev
->sector
);
2849 rbi2
= r5_next_bio(rbi
, dev
->sector
);
2850 spin_lock_irq(&conf
->device_lock
);
2851 if (!raid5_dec_bi_phys_segments(rbi
)) {
2852 rbi
->bi_next
= return_bi
;
2855 spin_unlock_irq(&conf
->device_lock
);
2860 /* now count some things */
2861 if (test_bit(R5_LOCKED
, &dev
->flags
)) s
.locked
++;
2862 if (test_bit(R5_UPTODATE
, &dev
->flags
)) s
.uptodate
++;
2869 if (!test_bit(R5_OVERWRITE
, &dev
->flags
))
2874 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
2875 if (blocked_rdev
== NULL
&&
2876 rdev
&& unlikely(test_bit(Blocked
, &rdev
->flags
))) {
2877 blocked_rdev
= rdev
;
2878 atomic_inc(&rdev
->nr_pending
);
2880 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)) {
2881 /* The ReadError flag will just be confusing now */
2882 clear_bit(R5_ReadError
, &dev
->flags
);
2883 clear_bit(R5_ReWrite
, &dev
->flags
);
2885 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)
2886 || test_bit(R5_ReadError
, &dev
->flags
)) {
2888 r6s
.failed_num
[s
.failed
] = i
;
2891 set_bit(R5_Insync
, &dev
->flags
);
2895 if (unlikely(blocked_rdev
)) {
2896 if (s
.syncing
|| s
.expanding
|| s
.expanded
||
2897 s
.to_write
|| s
.written
) {
2898 set_bit(STRIPE_HANDLE
, &sh
->state
);
2901 /* There is nothing for the blocked_rdev to block */
2902 rdev_dec_pending(blocked_rdev
, conf
->mddev
);
2903 blocked_rdev
= NULL
;
2906 pr_debug("locked=%d uptodate=%d to_read=%d"
2907 " to_write=%d failed=%d failed_num=%d,%d\n",
2908 s
.locked
, s
.uptodate
, s
.to_read
, s
.to_write
, s
.failed
,
2909 r6s
.failed_num
[0], r6s
.failed_num
[1]);
2910 /* check if the array has lost >2 devices and, if so, some requests
2911 * might need to be failed
2913 if (s
.failed
> 2 && s
.to_read
+s
.to_write
+s
.written
)
2914 handle_failed_stripe(conf
, sh
, &s
, disks
, &return_bi
);
2915 if (s
.failed
> 2 && s
.syncing
) {
2916 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,0);
2917 clear_bit(STRIPE_SYNCING
, &sh
->state
);
2922 * might be able to return some write requests if the parity blocks
2923 * are safe, or on a failed drive
2925 pdev
= &sh
->dev
[pd_idx
];
2926 r6s
.p_failed
= (s
.failed
>= 1 && r6s
.failed_num
[0] == pd_idx
)
2927 || (s
.failed
>= 2 && r6s
.failed_num
[1] == pd_idx
);
2928 qdev
= &sh
->dev
[r6s
.qd_idx
];
2929 r6s
.q_failed
= (s
.failed
>= 1 && r6s
.failed_num
[0] == r6s
.qd_idx
)
2930 || (s
.failed
>= 2 && r6s
.failed_num
[1] == r6s
.qd_idx
);
2933 ( r6s
.p_failed
|| ((test_bit(R5_Insync
, &pdev
->flags
)
2934 && !test_bit(R5_LOCKED
, &pdev
->flags
)
2935 && test_bit(R5_UPTODATE
, &pdev
->flags
)))) &&
2936 ( r6s
.q_failed
|| ((test_bit(R5_Insync
, &qdev
->flags
)
2937 && !test_bit(R5_LOCKED
, &qdev
->flags
)
2938 && test_bit(R5_UPTODATE
, &qdev
->flags
)))))
2939 handle_stripe_clean_event(conf
, sh
, disks
, &return_bi
);
2941 /* Now we might consider reading some blocks, either to check/generate
2942 * parity, or to satisfy requests
2943 * or to load a block that is being partially written.
2945 if (s
.to_read
|| s
.non_overwrite
|| (s
.to_write
&& s
.failed
) ||
2946 (s
.syncing
&& (s
.uptodate
< disks
)) || s
.expanding
)
2947 handle_stripe_fill6(sh
, &s
, &r6s
, disks
);
2949 /* now to consider writing and what else, if anything should be read */
2951 handle_stripe_dirtying6(conf
, sh
, &s
, &r6s
, disks
);
2953 /* maybe we need to check and possibly fix the parity for this stripe
2954 * Any reads will already have been scheduled, so we just see if enough
2957 if (s
.syncing
&& s
.locked
== 0 && !test_bit(STRIPE_INSYNC
, &sh
->state
))
2958 handle_parity_checks6(conf
, sh
, &s
, &r6s
, tmp_page
, disks
);
2960 if (s
.syncing
&& s
.locked
== 0 && test_bit(STRIPE_INSYNC
, &sh
->state
)) {
2961 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,1);
2962 clear_bit(STRIPE_SYNCING
, &sh
->state
);
2965 /* If the failed drives are just a ReadError, then we might need
2966 * to progress the repair/check process
2968 if (s
.failed
<= 2 && !conf
->mddev
->ro
)
2969 for (i
= 0; i
< s
.failed
; i
++) {
2970 dev
= &sh
->dev
[r6s
.failed_num
[i
]];
2971 if (test_bit(R5_ReadError
, &dev
->flags
)
2972 && !test_bit(R5_LOCKED
, &dev
->flags
)
2973 && test_bit(R5_UPTODATE
, &dev
->flags
)
2975 if (!test_bit(R5_ReWrite
, &dev
->flags
)) {
2976 set_bit(R5_Wantwrite
, &dev
->flags
);
2977 set_bit(R5_ReWrite
, &dev
->flags
);
2978 set_bit(R5_LOCKED
, &dev
->flags
);
2980 /* let's read it back */
2981 set_bit(R5_Wantread
, &dev
->flags
);
2982 set_bit(R5_LOCKED
, &dev
->flags
);
2987 if (s
.expanded
&& test_bit(STRIPE_EXPANDING
, &sh
->state
)) {
2988 /* Need to write out all blocks after computing P&Q */
2989 sh
->disks
= conf
->raid_disks
;
2990 sh
->pd_idx
= stripe_to_pdidx(sh
->sector
, conf
,
2992 compute_parity6(sh
, RECONSTRUCT_WRITE
);
2993 for (i
= conf
->raid_disks
; i
-- ; ) {
2994 set_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
2996 set_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
);
2998 clear_bit(STRIPE_EXPANDING
, &sh
->state
);
2999 } else if (s
.expanded
) {
3000 clear_bit(STRIPE_EXPAND_READY
, &sh
->state
);
3001 atomic_dec(&conf
->reshape_stripes
);
3002 wake_up(&conf
->wait_for_overlap
);
3003 md_done_sync(conf
->mddev
, STRIPE_SECTORS
, 1);
3006 if (s
.expanding
&& s
.locked
== 0 &&
3007 !test_bit(STRIPE_COMPUTE_RUN
, &sh
->state
))
3008 handle_stripe_expansion(conf
, sh
, &r6s
);
3011 spin_unlock(&sh
->lock
);
3013 /* wait for this device to become unblocked */
3014 if (unlikely(blocked_rdev
))
3015 md_wait_for_blocked_rdev(blocked_rdev
, conf
->mddev
);
3019 return_io(return_bi
);
3021 return blocked_rdev
== NULL
;
3024 /* returns true if the stripe was handled */
3025 static bool handle_stripe(struct stripe_head
*sh
, struct page
*tmp_page
)
3027 if (sh
->raid_conf
->level
== 6)
3028 return handle_stripe6(sh
, tmp_page
);
3030 return handle_stripe5(sh
);
3035 static void raid5_activate_delayed(raid5_conf_t
*conf
)
3037 if (atomic_read(&conf
->preread_active_stripes
) < IO_THRESHOLD
) {
3038 while (!list_empty(&conf
->delayed_list
)) {
3039 struct list_head
*l
= conf
->delayed_list
.next
;
3040 struct stripe_head
*sh
;
3041 sh
= list_entry(l
, struct stripe_head
, lru
);
3043 clear_bit(STRIPE_DELAYED
, &sh
->state
);
3044 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
))
3045 atomic_inc(&conf
->preread_active_stripes
);
3046 list_add_tail(&sh
->lru
, &conf
->hold_list
);
3049 blk_plug_device(conf
->mddev
->queue
);
3052 static void activate_bit_delay(raid5_conf_t
*conf
)
3054 /* device_lock is held */
3055 struct list_head head
;
3056 list_add(&head
, &conf
->bitmap_list
);
3057 list_del_init(&conf
->bitmap_list
);
3058 while (!list_empty(&head
)) {
3059 struct stripe_head
*sh
= list_entry(head
.next
, struct stripe_head
, lru
);
3060 list_del_init(&sh
->lru
);
3061 atomic_inc(&sh
->count
);
3062 __release_stripe(conf
, sh
);
3066 static void unplug_slaves(mddev_t
*mddev
)
3068 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3072 for (i
=0; i
<mddev
->raid_disks
; i
++) {
3073 mdk_rdev_t
*rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
3074 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
) && atomic_read(&rdev
->nr_pending
)) {
3075 struct request_queue
*r_queue
= bdev_get_queue(rdev
->bdev
);
3077 atomic_inc(&rdev
->nr_pending
);
3080 blk_unplug(r_queue
);
3082 rdev_dec_pending(rdev
, mddev
);
3089 static void raid5_unplug_device(struct request_queue
*q
)
3091 mddev_t
*mddev
= q
->queuedata
;
3092 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3093 unsigned long flags
;
3095 spin_lock_irqsave(&conf
->device_lock
, flags
);
3097 if (blk_remove_plug(q
)) {
3099 raid5_activate_delayed(conf
);
3101 md_wakeup_thread(mddev
->thread
);
3103 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
3105 unplug_slaves(mddev
);
3108 static int raid5_congested(void *data
, int bits
)
3110 mddev_t
*mddev
= data
;
3111 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3113 /* No difference between reads and writes. Just check
3114 * how busy the stripe_cache is
3116 if (conf
->inactive_blocked
)
3120 if (list_empty_careful(&conf
->inactive_list
))
3126 /* We want read requests to align with chunks where possible,
3127 * but write requests don't need to.
3129 static int raid5_mergeable_bvec(struct request_queue
*q
,
3130 struct bvec_merge_data
*bvm
,
3131 struct bio_vec
*biovec
)
3133 mddev_t
*mddev
= q
->queuedata
;
3134 sector_t sector
= bvm
->bi_sector
+ get_start_sect(bvm
->bi_bdev
);
3136 unsigned int chunk_sectors
= mddev
->chunk_size
>> 9;
3137 unsigned int bio_sectors
= bvm
->bi_size
>> 9;
3139 if ((bvm
->bi_rw
& 1) == WRITE
)
3140 return biovec
->bv_len
; /* always allow writes to be mergeable */
3142 max
= (chunk_sectors
- ((sector
& (chunk_sectors
- 1)) + bio_sectors
)) << 9;
3143 if (max
< 0) max
= 0;
3144 if (max
<= biovec
->bv_len
&& bio_sectors
== 0)
3145 return biovec
->bv_len
;
3151 static int in_chunk_boundary(mddev_t
*mddev
, struct bio
*bio
)
3153 sector_t sector
= bio
->bi_sector
+ get_start_sect(bio
->bi_bdev
);
3154 unsigned int chunk_sectors
= mddev
->chunk_size
>> 9;
3155 unsigned int bio_sectors
= bio
->bi_size
>> 9;
3157 return chunk_sectors
>=
3158 ((sector
& (chunk_sectors
- 1)) + bio_sectors
);
3162 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
3163 * later sampled by raid5d.
3165 static void add_bio_to_retry(struct bio
*bi
,raid5_conf_t
*conf
)
3167 unsigned long flags
;
3169 spin_lock_irqsave(&conf
->device_lock
, flags
);
3171 bi
->bi_next
= conf
->retry_read_aligned_list
;
3172 conf
->retry_read_aligned_list
= bi
;
3174 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
3175 md_wakeup_thread(conf
->mddev
->thread
);
3179 static struct bio
*remove_bio_from_retry(raid5_conf_t
*conf
)
3183 bi
= conf
->retry_read_aligned
;
3185 conf
->retry_read_aligned
= NULL
;
3188 bi
= conf
->retry_read_aligned_list
;
3190 conf
->retry_read_aligned_list
= bi
->bi_next
;
3193 * this sets the active strip count to 1 and the processed
3194 * strip count to zero (upper 8 bits)
3196 bi
->bi_phys_segments
= 1; /* biased count of active stripes */
3204 * The "raid5_align_endio" should check if the read succeeded and if it
3205 * did, call bio_endio on the original bio (having bio_put the new bio
3207 * If the read failed..
3209 static void raid5_align_endio(struct bio
*bi
, int error
)
3211 struct bio
* raid_bi
= bi
->bi_private
;
3214 int uptodate
= test_bit(BIO_UPTODATE
, &bi
->bi_flags
);
3219 mddev
= raid_bi
->bi_bdev
->bd_disk
->queue
->queuedata
;
3220 conf
= mddev_to_conf(mddev
);
3221 rdev
= (void*)raid_bi
->bi_next
;
3222 raid_bi
->bi_next
= NULL
;
3224 rdev_dec_pending(rdev
, conf
->mddev
);
3226 if (!error
&& uptodate
) {
3227 bio_endio(raid_bi
, 0);
3228 if (atomic_dec_and_test(&conf
->active_aligned_reads
))
3229 wake_up(&conf
->wait_for_stripe
);
3234 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
3236 add_bio_to_retry(raid_bi
, conf
);
3239 static int bio_fits_rdev(struct bio
*bi
)
3241 struct request_queue
*q
= bdev_get_queue(bi
->bi_bdev
);
3243 if ((bi
->bi_size
>>9) > q
->max_sectors
)
3245 blk_recount_segments(q
, bi
);
3246 if (bi
->bi_phys_segments
> q
->max_phys_segments
)
3249 if (q
->merge_bvec_fn
)
3250 /* it's too hard to apply the merge_bvec_fn at this stage,
3259 static int chunk_aligned_read(struct request_queue
*q
, struct bio
* raid_bio
)
3261 mddev_t
*mddev
= q
->queuedata
;
3262 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3263 const unsigned int raid_disks
= conf
->raid_disks
;
3264 const unsigned int data_disks
= raid_disks
- conf
->max_degraded
;
3265 unsigned int dd_idx
, pd_idx
;
3266 struct bio
* align_bi
;
3269 if (!in_chunk_boundary(mddev
, raid_bio
)) {
3270 pr_debug("chunk_aligned_read : non aligned\n");
3274 * use bio_clone to make a copy of the bio
3276 align_bi
= bio_clone(raid_bio
, GFP_NOIO
);
3280 * set bi_end_io to a new function, and set bi_private to the
3283 align_bi
->bi_end_io
= raid5_align_endio
;
3284 align_bi
->bi_private
= raid_bio
;
3288 align_bi
->bi_sector
= raid5_compute_sector(raid_bio
->bi_sector
,
3296 rdev
= rcu_dereference(conf
->disks
[dd_idx
].rdev
);
3297 if (rdev
&& test_bit(In_sync
, &rdev
->flags
)) {
3298 atomic_inc(&rdev
->nr_pending
);
3300 raid_bio
->bi_next
= (void*)rdev
;
3301 align_bi
->bi_bdev
= rdev
->bdev
;
3302 align_bi
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
3303 align_bi
->bi_sector
+= rdev
->data_offset
;
3305 if (!bio_fits_rdev(align_bi
)) {
3306 /* too big in some way */
3308 rdev_dec_pending(rdev
, mddev
);
3312 spin_lock_irq(&conf
->device_lock
);
3313 wait_event_lock_irq(conf
->wait_for_stripe
,
3315 conf
->device_lock
, /* nothing */);
3316 atomic_inc(&conf
->active_aligned_reads
);
3317 spin_unlock_irq(&conf
->device_lock
);
3319 generic_make_request(align_bi
);
3328 /* __get_priority_stripe - get the next stripe to process
3330 * Full stripe writes are allowed to pass preread active stripes up until
3331 * the bypass_threshold is exceeded. In general the bypass_count
3332 * increments when the handle_list is handled before the hold_list; however, it
3333 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
3334 * stripe with in flight i/o. The bypass_count will be reset when the
3335 * head of the hold_list has changed, i.e. the head was promoted to the
3338 static struct stripe_head
*__get_priority_stripe(raid5_conf_t
*conf
)
3340 struct stripe_head
*sh
;
3342 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
3344 list_empty(&conf
->handle_list
) ? "empty" : "busy",
3345 list_empty(&conf
->hold_list
) ? "empty" : "busy",
3346 atomic_read(&conf
->pending_full_writes
), conf
->bypass_count
);
3348 if (!list_empty(&conf
->handle_list
)) {
3349 sh
= list_entry(conf
->handle_list
.next
, typeof(*sh
), lru
);
3351 if (list_empty(&conf
->hold_list
))
3352 conf
->bypass_count
= 0;
3353 else if (!test_bit(STRIPE_IO_STARTED
, &sh
->state
)) {
3354 if (conf
->hold_list
.next
== conf
->last_hold
)
3355 conf
->bypass_count
++;
3357 conf
->last_hold
= conf
->hold_list
.next
;
3358 conf
->bypass_count
-= conf
->bypass_threshold
;
3359 if (conf
->bypass_count
< 0)
3360 conf
->bypass_count
= 0;
3363 } else if (!list_empty(&conf
->hold_list
) &&
3364 ((conf
->bypass_threshold
&&
3365 conf
->bypass_count
> conf
->bypass_threshold
) ||
3366 atomic_read(&conf
->pending_full_writes
) == 0)) {
3367 sh
= list_entry(conf
->hold_list
.next
,
3369 conf
->bypass_count
-= conf
->bypass_threshold
;
3370 if (conf
->bypass_count
< 0)
3371 conf
->bypass_count
= 0;
3375 list_del_init(&sh
->lru
);
3376 atomic_inc(&sh
->count
);
3377 BUG_ON(atomic_read(&sh
->count
) != 1);
3381 static int make_request(struct request_queue
*q
, struct bio
* bi
)
3383 mddev_t
*mddev
= q
->queuedata
;
3384 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3385 unsigned int dd_idx
, pd_idx
;
3386 sector_t new_sector
;
3387 sector_t logical_sector
, last_sector
;
3388 struct stripe_head
*sh
;
3389 const int rw
= bio_data_dir(bi
);
3392 if (unlikely(bio_barrier(bi
))) {
3393 bio_endio(bi
, -EOPNOTSUPP
);
3397 md_write_start(mddev
, bi
);
3399 cpu
= part_stat_lock();
3400 part_stat_inc(cpu
, &mddev
->gendisk
->part0
, ios
[rw
]);
3401 part_stat_add(cpu
, &mddev
->gendisk
->part0
, sectors
[rw
],
3406 mddev
->reshape_position
== MaxSector
&&
3407 chunk_aligned_read(q
,bi
))
3410 logical_sector
= bi
->bi_sector
& ~((sector_t
)STRIPE_SECTORS
-1);
3411 last_sector
= bi
->bi_sector
+ (bi
->bi_size
>>9);
3413 bi
->bi_phys_segments
= 1; /* over-loaded to count active stripes */
3415 for (;logical_sector
< last_sector
; logical_sector
+= STRIPE_SECTORS
) {
3417 int disks
, data_disks
;
3420 prepare_to_wait(&conf
->wait_for_overlap
, &w
, TASK_UNINTERRUPTIBLE
);
3421 if (likely(conf
->expand_progress
== MaxSector
))
3422 disks
= conf
->raid_disks
;
3424 /* spinlock is needed as expand_progress may be
3425 * 64bit on a 32bit platform, and so it might be
3426 * possible to see a half-updated value
3427 * Ofcourse expand_progress could change after
3428 * the lock is dropped, so once we get a reference
3429 * to the stripe that we think it is, we will have
3432 spin_lock_irq(&conf
->device_lock
);
3433 disks
= conf
->raid_disks
;
3434 if (logical_sector
>= conf
->expand_progress
)
3435 disks
= conf
->previous_raid_disks
;
3437 if (logical_sector
>= conf
->expand_lo
) {
3438 spin_unlock_irq(&conf
->device_lock
);
3443 spin_unlock_irq(&conf
->device_lock
);
3445 data_disks
= disks
- conf
->max_degraded
;
3447 new_sector
= raid5_compute_sector(logical_sector
, disks
, data_disks
,
3448 &dd_idx
, &pd_idx
, conf
);
3449 pr_debug("raid5: make_request, sector %llu logical %llu\n",
3450 (unsigned long long)new_sector
,
3451 (unsigned long long)logical_sector
);
3453 sh
= get_active_stripe(conf
, new_sector
, disks
, pd_idx
, (bi
->bi_rw
&RWA_MASK
));
3455 if (unlikely(conf
->expand_progress
!= MaxSector
)) {
3456 /* expansion might have moved on while waiting for a
3457 * stripe, so we must do the range check again.
3458 * Expansion could still move past after this
3459 * test, but as we are holding a reference to
3460 * 'sh', we know that if that happens,
3461 * STRIPE_EXPANDING will get set and the expansion
3462 * won't proceed until we finish with the stripe.
3465 spin_lock_irq(&conf
->device_lock
);
3466 if (logical_sector
< conf
->expand_progress
&&
3467 disks
== conf
->previous_raid_disks
)
3468 /* mismatch, need to try again */
3470 spin_unlock_irq(&conf
->device_lock
);
3476 /* FIXME what if we get a false positive because these
3477 * are being updated.
3479 if (logical_sector
>= mddev
->suspend_lo
&&
3480 logical_sector
< mddev
->suspend_hi
) {
3486 if (test_bit(STRIPE_EXPANDING
, &sh
->state
) ||
3487 !add_stripe_bio(sh
, bi
, dd_idx
, (bi
->bi_rw
&RW_MASK
))) {
3488 /* Stripe is busy expanding or
3489 * add failed due to overlap. Flush everything
3492 raid5_unplug_device(mddev
->queue
);
3497 finish_wait(&conf
->wait_for_overlap
, &w
);
3498 set_bit(STRIPE_HANDLE
, &sh
->state
);
3499 clear_bit(STRIPE_DELAYED
, &sh
->state
);
3502 /* cannot get stripe for read-ahead, just give-up */
3503 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
3504 finish_wait(&conf
->wait_for_overlap
, &w
);
3509 spin_lock_irq(&conf
->device_lock
);
3510 remaining
= raid5_dec_bi_phys_segments(bi
);
3511 spin_unlock_irq(&conf
->device_lock
);
3512 if (remaining
== 0) {
3515 md_write_end(mddev
);
3522 static sector_t
reshape_request(mddev_t
*mddev
, sector_t sector_nr
, int *skipped
)
3524 /* reshaping is quite different to recovery/resync so it is
3525 * handled quite separately ... here.
3527 * On each call to sync_request, we gather one chunk worth of
3528 * destination stripes and flag them as expanding.
3529 * Then we find all the source stripes and request reads.
3530 * As the reads complete, handle_stripe will copy the data
3531 * into the destination stripe and release that stripe.
3533 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
3534 struct stripe_head
*sh
;
3536 sector_t first_sector
, last_sector
;
3537 int raid_disks
= conf
->previous_raid_disks
;
3538 int data_disks
= raid_disks
- conf
->max_degraded
;
3539 int new_data_disks
= conf
->raid_disks
- conf
->max_degraded
;
3542 sector_t writepos
, safepos
, gap
;
3544 if (sector_nr
== 0 &&
3545 conf
->expand_progress
!= 0) {
3546 /* restarting in the middle, skip the initial sectors */
3547 sector_nr
= conf
->expand_progress
;
3548 sector_div(sector_nr
, new_data_disks
);
3553 /* we update the metadata when there is more than 3Meg
3554 * in the block range (that is rather arbitrary, should
3555 * probably be time based) or when the data about to be
3556 * copied would over-write the source of the data at
3557 * the front of the range.
3558 * i.e. one new_stripe forward from expand_progress new_maps
3559 * to after where expand_lo old_maps to
3561 writepos
= conf
->expand_progress
+
3562 conf
->chunk_size
/512*(new_data_disks
);
3563 sector_div(writepos
, new_data_disks
);
3564 safepos
= conf
->expand_lo
;
3565 sector_div(safepos
, data_disks
);
3566 gap
= conf
->expand_progress
- conf
->expand_lo
;
3568 if (writepos
>= safepos
||
3569 gap
> (new_data_disks
)*3000*2 /*3Meg*/) {
3570 /* Cannot proceed until we've updated the superblock... */
3571 wait_event(conf
->wait_for_overlap
,
3572 atomic_read(&conf
->reshape_stripes
)==0);
3573 mddev
->reshape_position
= conf
->expand_progress
;
3574 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3575 md_wakeup_thread(mddev
->thread
);
3576 wait_event(mddev
->sb_wait
, mddev
->flags
== 0 ||
3577 kthread_should_stop());
3578 spin_lock_irq(&conf
->device_lock
);
3579 conf
->expand_lo
= mddev
->reshape_position
;
3580 spin_unlock_irq(&conf
->device_lock
);
3581 wake_up(&conf
->wait_for_overlap
);
3584 for (i
=0; i
< conf
->chunk_size
/512; i
+= STRIPE_SECTORS
) {
3587 pd_idx
= stripe_to_pdidx(sector_nr
+i
, conf
, conf
->raid_disks
);
3588 sh
= get_active_stripe(conf
, sector_nr
+i
,
3589 conf
->raid_disks
, pd_idx
, 0);
3590 set_bit(STRIPE_EXPANDING
, &sh
->state
);
3591 atomic_inc(&conf
->reshape_stripes
);
3592 /* If any of this stripe is beyond the end of the old
3593 * array, then we need to zero those blocks
3595 for (j
=sh
->disks
; j
--;) {
3597 if (j
== sh
->pd_idx
)
3599 if (conf
->level
== 6 &&
3600 j
== raid6_next_disk(sh
->pd_idx
, sh
->disks
))
3602 s
= compute_blocknr(sh
, j
);
3603 if (s
< mddev
->array_sectors
) {
3607 memset(page_address(sh
->dev
[j
].page
), 0, STRIPE_SIZE
);
3608 set_bit(R5_Expanded
, &sh
->dev
[j
].flags
);
3609 set_bit(R5_UPTODATE
, &sh
->dev
[j
].flags
);
3612 set_bit(STRIPE_EXPAND_READY
, &sh
->state
);
3613 set_bit(STRIPE_HANDLE
, &sh
->state
);
3617 spin_lock_irq(&conf
->device_lock
);
3618 conf
->expand_progress
= (sector_nr
+ i
) * new_data_disks
;
3619 spin_unlock_irq(&conf
->device_lock
);
3620 /* Ok, those stripe are ready. We can start scheduling
3621 * reads on the source stripes.
3622 * The source stripes are determined by mapping the first and last
3623 * block on the destination stripes.
3626 raid5_compute_sector(sector_nr
*(new_data_disks
),
3627 raid_disks
, data_disks
,
3628 &dd_idx
, &pd_idx
, conf
);
3630 raid5_compute_sector((sector_nr
+conf
->chunk_size
/512)
3631 *(new_data_disks
) -1,
3632 raid_disks
, data_disks
,
3633 &dd_idx
, &pd_idx
, conf
);
3634 if (last_sector
>= (mddev
->size
<<1))
3635 last_sector
= (mddev
->size
<<1)-1;
3636 while (first_sector
<= last_sector
) {
3637 pd_idx
= stripe_to_pdidx(first_sector
, conf
,
3638 conf
->previous_raid_disks
);
3639 sh
= get_active_stripe(conf
, first_sector
,
3640 conf
->previous_raid_disks
, pd_idx
, 0);
3641 set_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
3642 set_bit(STRIPE_HANDLE
, &sh
->state
);
3644 first_sector
+= STRIPE_SECTORS
;
3646 /* If this takes us to the resync_max point where we have to pause,
3647 * then we need to write out the superblock.
3649 sector_nr
+= conf
->chunk_size
>>9;
3650 if (sector_nr
>= mddev
->resync_max
) {
3651 /* Cannot proceed until we've updated the superblock... */
3652 wait_event(conf
->wait_for_overlap
,
3653 atomic_read(&conf
->reshape_stripes
) == 0);
3654 mddev
->reshape_position
= conf
->expand_progress
;
3655 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3656 md_wakeup_thread(mddev
->thread
);
3657 wait_event(mddev
->sb_wait
,
3658 !test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)
3659 || kthread_should_stop());
3660 spin_lock_irq(&conf
->device_lock
);
3661 conf
->expand_lo
= mddev
->reshape_position
;
3662 spin_unlock_irq(&conf
->device_lock
);
3663 wake_up(&conf
->wait_for_overlap
);
3665 return conf
->chunk_size
>>9;
3668 /* FIXME go_faster isn't used */
3669 static inline sector_t
sync_request(mddev_t
*mddev
, sector_t sector_nr
, int *skipped
, int go_faster
)
3671 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
3672 struct stripe_head
*sh
;
3674 int raid_disks
= conf
->raid_disks
;
3675 sector_t max_sector
= mddev
->size
<< 1;
3677 int still_degraded
= 0;
3680 if (sector_nr
>= max_sector
) {
3681 /* just being told to finish up .. nothing much to do */
3682 unplug_slaves(mddev
);
3683 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)) {
3688 if (mddev
->curr_resync
< max_sector
) /* aborted */
3689 bitmap_end_sync(mddev
->bitmap
, mddev
->curr_resync
,
3691 else /* completed sync */
3693 bitmap_close_sync(mddev
->bitmap
);
3698 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
3699 return reshape_request(mddev
, sector_nr
, skipped
);
3701 /* No need to check resync_max as we never do more than one
3702 * stripe, and as resync_max will always be on a chunk boundary,
3703 * if the check in md_do_sync didn't fire, there is no chance
3704 * of overstepping resync_max here
3707 /* if there is too many failed drives and we are trying
3708 * to resync, then assert that we are finished, because there is
3709 * nothing we can do.
3711 if (mddev
->degraded
>= conf
->max_degraded
&&
3712 test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
3713 sector_t rv
= (mddev
->size
<< 1) - sector_nr
;
3717 if (!bitmap_start_sync(mddev
->bitmap
, sector_nr
, &sync_blocks
, 1) &&
3718 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
) &&
3719 !conf
->fullsync
&& sync_blocks
>= STRIPE_SECTORS
) {
3720 /* we can skip this block, and probably more */
3721 sync_blocks
/= STRIPE_SECTORS
;
3723 return sync_blocks
* STRIPE_SECTORS
; /* keep things rounded to whole stripes */
3727 bitmap_cond_end_sync(mddev
->bitmap
, sector_nr
);
3729 pd_idx
= stripe_to_pdidx(sector_nr
, conf
, raid_disks
);
3730 sh
= get_active_stripe(conf
, sector_nr
, raid_disks
, pd_idx
, 1);
3732 sh
= get_active_stripe(conf
, sector_nr
, raid_disks
, pd_idx
, 0);
3733 /* make sure we don't swamp the stripe cache if someone else
3734 * is trying to get access
3736 schedule_timeout_uninterruptible(1);
3738 /* Need to check if array will still be degraded after recovery/resync
3739 * We don't need to check the 'failed' flag as when that gets set,
3742 for (i
=0; i
<mddev
->raid_disks
; i
++)
3743 if (conf
->disks
[i
].rdev
== NULL
)
3746 bitmap_start_sync(mddev
->bitmap
, sector_nr
, &sync_blocks
, still_degraded
);
3748 spin_lock(&sh
->lock
);
3749 set_bit(STRIPE_SYNCING
, &sh
->state
);
3750 clear_bit(STRIPE_INSYNC
, &sh
->state
);
3751 spin_unlock(&sh
->lock
);
3753 /* wait for any blocked device to be handled */
3754 while(unlikely(!handle_stripe(sh
, NULL
)))
3758 return STRIPE_SECTORS
;
3761 static int retry_aligned_read(raid5_conf_t
*conf
, struct bio
*raid_bio
)
3763 /* We may not be able to submit a whole bio at once as there
3764 * may not be enough stripe_heads available.
3765 * We cannot pre-allocate enough stripe_heads as we may need
3766 * more than exist in the cache (if we allow ever large chunks).
3767 * So we do one stripe head at a time and record in
3768 * ->bi_hw_segments how many have been done.
3770 * We *know* that this entire raid_bio is in one chunk, so
3771 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
3773 struct stripe_head
*sh
;
3775 sector_t sector
, logical_sector
, last_sector
;
3780 logical_sector
= raid_bio
->bi_sector
& ~((sector_t
)STRIPE_SECTORS
-1);
3781 sector
= raid5_compute_sector( logical_sector
,
3783 conf
->raid_disks
- conf
->max_degraded
,
3787 last_sector
= raid_bio
->bi_sector
+ (raid_bio
->bi_size
>>9);
3789 for (; logical_sector
< last_sector
;
3790 logical_sector
+= STRIPE_SECTORS
,
3791 sector
+= STRIPE_SECTORS
,
3794 if (scnt
< raid5_bi_hw_segments(raid_bio
))
3795 /* already done this stripe */
3798 sh
= get_active_stripe(conf
, sector
, conf
->raid_disks
, pd_idx
, 1);
3801 /* failed to get a stripe - must wait */
3802 raid5_set_bi_hw_segments(raid_bio
, scnt
);
3803 conf
->retry_read_aligned
= raid_bio
;
3807 set_bit(R5_ReadError
, &sh
->dev
[dd_idx
].flags
);
3808 if (!add_stripe_bio(sh
, raid_bio
, dd_idx
, 0)) {
3810 raid5_set_bi_hw_segments(raid_bio
, scnt
);
3811 conf
->retry_read_aligned
= raid_bio
;
3815 handle_stripe(sh
, NULL
);
3819 spin_lock_irq(&conf
->device_lock
);
3820 remaining
= raid5_dec_bi_phys_segments(raid_bio
);
3821 spin_unlock_irq(&conf
->device_lock
);
3823 bio_endio(raid_bio
, 0);
3824 if (atomic_dec_and_test(&conf
->active_aligned_reads
))
3825 wake_up(&conf
->wait_for_stripe
);
3832 * This is our raid5 kernel thread.
3834 * We scan the hash table for stripes which can be handled now.
3835 * During the scan, completed stripes are saved for us by the interrupt
3836 * handler, so that they will not have to wait for our next wakeup.
3838 static void raid5d(mddev_t
*mddev
)
3840 struct stripe_head
*sh
;
3841 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3844 pr_debug("+++ raid5d active\n");
3846 md_check_recovery(mddev
);
3849 spin_lock_irq(&conf
->device_lock
);
3853 if (conf
->seq_flush
!= conf
->seq_write
) {
3854 int seq
= conf
->seq_flush
;
3855 spin_unlock_irq(&conf
->device_lock
);
3856 bitmap_unplug(mddev
->bitmap
);
3857 spin_lock_irq(&conf
->device_lock
);
3858 conf
->seq_write
= seq
;
3859 activate_bit_delay(conf
);
3862 while ((bio
= remove_bio_from_retry(conf
))) {
3864 spin_unlock_irq(&conf
->device_lock
);
3865 ok
= retry_aligned_read(conf
, bio
);
3866 spin_lock_irq(&conf
->device_lock
);
3872 sh
= __get_priority_stripe(conf
);
3876 spin_unlock_irq(&conf
->device_lock
);
3879 handle_stripe(sh
, conf
->spare_page
);
3882 spin_lock_irq(&conf
->device_lock
);
3884 pr_debug("%d stripes handled\n", handled
);
3886 spin_unlock_irq(&conf
->device_lock
);
3888 async_tx_issue_pending_all();
3889 unplug_slaves(mddev
);
3891 pr_debug("--- raid5d inactive\n");
3895 raid5_show_stripe_cache_size(mddev_t
*mddev
, char *page
)
3897 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3899 return sprintf(page
, "%d\n", conf
->max_nr_stripes
);
3905 raid5_store_stripe_cache_size(mddev_t
*mddev
, const char *page
, size_t len
)
3907 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3911 if (len
>= PAGE_SIZE
)
3916 if (strict_strtoul(page
, 10, &new))
3918 if (new <= 16 || new > 32768)
3920 while (new < conf
->max_nr_stripes
) {
3921 if (drop_one_stripe(conf
))
3922 conf
->max_nr_stripes
--;
3926 err
= md_allow_write(mddev
);
3929 while (new > conf
->max_nr_stripes
) {
3930 if (grow_one_stripe(conf
))
3931 conf
->max_nr_stripes
++;
3937 static struct md_sysfs_entry
3938 raid5_stripecache_size
= __ATTR(stripe_cache_size
, S_IRUGO
| S_IWUSR
,
3939 raid5_show_stripe_cache_size
,
3940 raid5_store_stripe_cache_size
);
3943 raid5_show_preread_threshold(mddev_t
*mddev
, char *page
)
3945 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3947 return sprintf(page
, "%d\n", conf
->bypass_threshold
);
3953 raid5_store_preread_threshold(mddev_t
*mddev
, const char *page
, size_t len
)
3955 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3957 if (len
>= PAGE_SIZE
)
3962 if (strict_strtoul(page
, 10, &new))
3964 if (new > conf
->max_nr_stripes
)
3966 conf
->bypass_threshold
= new;
3970 static struct md_sysfs_entry
3971 raid5_preread_bypass_threshold
= __ATTR(preread_bypass_threshold
,
3973 raid5_show_preread_threshold
,
3974 raid5_store_preread_threshold
);
3977 stripe_cache_active_show(mddev_t
*mddev
, char *page
)
3979 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3981 return sprintf(page
, "%d\n", atomic_read(&conf
->active_stripes
));
3986 static struct md_sysfs_entry
3987 raid5_stripecache_active
= __ATTR_RO(stripe_cache_active
);
3989 static struct attribute
*raid5_attrs
[] = {
3990 &raid5_stripecache_size
.attr
,
3991 &raid5_stripecache_active
.attr
,
3992 &raid5_preread_bypass_threshold
.attr
,
3995 static struct attribute_group raid5_attrs_group
= {
3997 .attrs
= raid5_attrs
,
4000 static int run(mddev_t
*mddev
)
4003 int raid_disk
, memory
;
4005 struct disk_info
*disk
;
4006 struct list_head
*tmp
;
4007 int working_disks
= 0;
4009 if (mddev
->level
!= 5 && mddev
->level
!= 4 && mddev
->level
!= 6) {
4010 printk(KERN_ERR
"raid5: %s: raid level not set to 4/5/6 (%d)\n",
4011 mdname(mddev
), mddev
->level
);
4015 if (mddev
->reshape_position
!= MaxSector
) {
4016 /* Check that we can continue the reshape.
4017 * Currently only disks can change, it must
4018 * increase, and we must be past the point where
4019 * a stripe over-writes itself
4021 sector_t here_new
, here_old
;
4023 int max_degraded
= (mddev
->level
== 5 ? 1 : 2);
4025 if (mddev
->new_level
!= mddev
->level
||
4026 mddev
->new_layout
!= mddev
->layout
||
4027 mddev
->new_chunk
!= mddev
->chunk_size
) {
4028 printk(KERN_ERR
"raid5: %s: unsupported reshape "
4029 "required - aborting.\n",
4033 if (mddev
->delta_disks
<= 0) {
4034 printk(KERN_ERR
"raid5: %s: unsupported reshape "
4035 "(reduce disks) required - aborting.\n",
4039 old_disks
= mddev
->raid_disks
- mddev
->delta_disks
;
4040 /* reshape_position must be on a new-stripe boundary, and one
4041 * further up in new geometry must map after here in old
4044 here_new
= mddev
->reshape_position
;
4045 if (sector_div(here_new
, (mddev
->chunk_size
>>9)*
4046 (mddev
->raid_disks
- max_degraded
))) {
4047 printk(KERN_ERR
"raid5: reshape_position not "
4048 "on a stripe boundary\n");
4051 /* here_new is the stripe we will write to */
4052 here_old
= mddev
->reshape_position
;
4053 sector_div(here_old
, (mddev
->chunk_size
>>9)*
4054 (old_disks
-max_degraded
));
4055 /* here_old is the first stripe that we might need to read
4057 if (here_new
>= here_old
) {
4058 /* Reading from the same stripe as writing to - bad */
4059 printk(KERN_ERR
"raid5: reshape_position too early for "
4060 "auto-recovery - aborting.\n");
4063 printk(KERN_INFO
"raid5: reshape will continue\n");
4064 /* OK, we should be able to continue; */
4068 mddev
->private = kzalloc(sizeof (raid5_conf_t
), GFP_KERNEL
);
4069 if ((conf
= mddev
->private) == NULL
)
4071 if (mddev
->reshape_position
== MaxSector
) {
4072 conf
->previous_raid_disks
= conf
->raid_disks
= mddev
->raid_disks
;
4074 conf
->raid_disks
= mddev
->raid_disks
;
4075 conf
->previous_raid_disks
= mddev
->raid_disks
- mddev
->delta_disks
;
4078 conf
->disks
= kzalloc(conf
->raid_disks
* sizeof(struct disk_info
),
4083 conf
->mddev
= mddev
;
4085 if ((conf
->stripe_hashtbl
= kzalloc(PAGE_SIZE
, GFP_KERNEL
)) == NULL
)
4088 if (mddev
->level
== 6) {
4089 conf
->spare_page
= alloc_page(GFP_KERNEL
);
4090 if (!conf
->spare_page
)
4093 spin_lock_init(&conf
->device_lock
);
4094 mddev
->queue
->queue_lock
= &conf
->device_lock
;
4095 init_waitqueue_head(&conf
->wait_for_stripe
);
4096 init_waitqueue_head(&conf
->wait_for_overlap
);
4097 INIT_LIST_HEAD(&conf
->handle_list
);
4098 INIT_LIST_HEAD(&conf
->hold_list
);
4099 INIT_LIST_HEAD(&conf
->delayed_list
);
4100 INIT_LIST_HEAD(&conf
->bitmap_list
);
4101 INIT_LIST_HEAD(&conf
->inactive_list
);
4102 atomic_set(&conf
->active_stripes
, 0);
4103 atomic_set(&conf
->preread_active_stripes
, 0);
4104 atomic_set(&conf
->active_aligned_reads
, 0);
4105 conf
->bypass_threshold
= BYPASS_THRESHOLD
;
4107 pr_debug("raid5: run(%s) called.\n", mdname(mddev
));
4109 rdev_for_each(rdev
, tmp
, mddev
) {
4110 raid_disk
= rdev
->raid_disk
;
4111 if (raid_disk
>= conf
->raid_disks
4114 disk
= conf
->disks
+ raid_disk
;
4118 if (test_bit(In_sync
, &rdev
->flags
)) {
4119 char b
[BDEVNAME_SIZE
];
4120 printk(KERN_INFO
"raid5: device %s operational as raid"
4121 " disk %d\n", bdevname(rdev
->bdev
,b
),
4125 /* Cannot rely on bitmap to complete recovery */
4130 * 0 for a fully functional array, 1 or 2 for a degraded array.
4132 mddev
->degraded
= conf
->raid_disks
- working_disks
;
4133 conf
->mddev
= mddev
;
4134 conf
->chunk_size
= mddev
->chunk_size
;
4135 conf
->level
= mddev
->level
;
4136 if (conf
->level
== 6)
4137 conf
->max_degraded
= 2;
4139 conf
->max_degraded
= 1;
4140 conf
->algorithm
= mddev
->layout
;
4141 conf
->max_nr_stripes
= NR_STRIPES
;
4142 conf
->expand_progress
= mddev
->reshape_position
;
4144 /* device size must be a multiple of chunk size */
4145 mddev
->size
&= ~(mddev
->chunk_size
/1024 -1);
4146 mddev
->resync_max_sectors
= mddev
->size
<< 1;
4148 if (conf
->level
== 6 && conf
->raid_disks
< 4) {
4149 printk(KERN_ERR
"raid6: not enough configured devices for %s (%d, minimum 4)\n",
4150 mdname(mddev
), conf
->raid_disks
);
4153 if (!conf
->chunk_size
|| conf
->chunk_size
% 4) {
4154 printk(KERN_ERR
"raid5: invalid chunk size %d for %s\n",
4155 conf
->chunk_size
, mdname(mddev
));
4158 if (conf
->algorithm
> ALGORITHM_RIGHT_SYMMETRIC
) {
4160 "raid5: unsupported parity algorithm %d for %s\n",
4161 conf
->algorithm
, mdname(mddev
));
4164 if (mddev
->degraded
> conf
->max_degraded
) {
4165 printk(KERN_ERR
"raid5: not enough operational devices for %s"
4166 " (%d/%d failed)\n",
4167 mdname(mddev
), mddev
->degraded
, conf
->raid_disks
);
4171 if (mddev
->degraded
> 0 &&
4172 mddev
->recovery_cp
!= MaxSector
) {
4173 if (mddev
->ok_start_degraded
)
4175 "raid5: starting dirty degraded array: %s"
4176 "- data corruption possible.\n",
4180 "raid5: cannot start dirty degraded array for %s\n",
4187 mddev
->thread
= md_register_thread(raid5d
, mddev
, "%s_raid5");
4188 if (!mddev
->thread
) {
4190 "raid5: couldn't allocate thread for %s\n",
4195 memory
= conf
->max_nr_stripes
* (sizeof(struct stripe_head
) +
4196 conf
->raid_disks
* ((sizeof(struct bio
) + PAGE_SIZE
))) / 1024;
4197 if (grow_stripes(conf
, conf
->max_nr_stripes
)) {
4199 "raid5: couldn't allocate %dkB for buffers\n", memory
);
4200 shrink_stripes(conf
);
4201 md_unregister_thread(mddev
->thread
);
4204 printk(KERN_INFO
"raid5: allocated %dkB for %s\n",
4205 memory
, mdname(mddev
));
4207 if (mddev
->degraded
== 0)
4208 printk("raid5: raid level %d set %s active with %d out of %d"
4209 " devices, algorithm %d\n", conf
->level
, mdname(mddev
),
4210 mddev
->raid_disks
-mddev
->degraded
, mddev
->raid_disks
,
4213 printk(KERN_ALERT
"raid5: raid level %d set %s active with %d"
4214 " out of %d devices, algorithm %d\n", conf
->level
,
4215 mdname(mddev
), mddev
->raid_disks
- mddev
->degraded
,
4216 mddev
->raid_disks
, conf
->algorithm
);
4218 print_raid5_conf(conf
);
4220 if (conf
->expand_progress
!= MaxSector
) {
4221 printk("...ok start reshape thread\n");
4222 conf
->expand_lo
= conf
->expand_progress
;
4223 atomic_set(&conf
->reshape_stripes
, 0);
4224 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4225 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4226 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
4227 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
4228 mddev
->sync_thread
= md_register_thread(md_do_sync
, mddev
,
4232 /* read-ahead size must cover two whole stripes, which is
4233 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
4236 int data_disks
= conf
->previous_raid_disks
- conf
->max_degraded
;
4237 int stripe
= data_disks
*
4238 (mddev
->chunk_size
/ PAGE_SIZE
);
4239 if (mddev
->queue
->backing_dev_info
.ra_pages
< 2 * stripe
)
4240 mddev
->queue
->backing_dev_info
.ra_pages
= 2 * stripe
;
4243 /* Ok, everything is just fine now */
4244 if (sysfs_create_group(&mddev
->kobj
, &raid5_attrs_group
))
4246 "raid5: failed to create sysfs attributes for %s\n",
4249 mddev
->queue
->unplug_fn
= raid5_unplug_device
;
4250 mddev
->queue
->backing_dev_info
.congested_data
= mddev
;
4251 mddev
->queue
->backing_dev_info
.congested_fn
= raid5_congested
;
4253 mddev
->array_sectors
= 2 * mddev
->size
* (conf
->previous_raid_disks
-
4254 conf
->max_degraded
);
4256 blk_queue_merge_bvec(mddev
->queue
, raid5_mergeable_bvec
);
4261 print_raid5_conf(conf
);
4262 safe_put_page(conf
->spare_page
);
4264 kfree(conf
->stripe_hashtbl
);
4267 mddev
->private = NULL
;
4268 printk(KERN_ALERT
"raid5: failed to run raid set %s\n", mdname(mddev
));
4274 static int stop(mddev_t
*mddev
)
4276 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
4278 md_unregister_thread(mddev
->thread
);
4279 mddev
->thread
= NULL
;
4280 shrink_stripes(conf
);
4281 kfree(conf
->stripe_hashtbl
);
4282 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
4283 blk_sync_queue(mddev
->queue
); /* the unplug fn references 'conf'*/
4284 sysfs_remove_group(&mddev
->kobj
, &raid5_attrs_group
);
4287 mddev
->private = NULL
;
4292 static void print_sh (struct seq_file
*seq
, struct stripe_head
*sh
)
4296 seq_printf(seq
, "sh %llu, pd_idx %d, state %ld.\n",
4297 (unsigned long long)sh
->sector
, sh
->pd_idx
, sh
->state
);
4298 seq_printf(seq
, "sh %llu, count %d.\n",
4299 (unsigned long long)sh
->sector
, atomic_read(&sh
->count
));
4300 seq_printf(seq
, "sh %llu, ", (unsigned long long)sh
->sector
);
4301 for (i
= 0; i
< sh
->disks
; i
++) {
4302 seq_printf(seq
, "(cache%d: %p %ld) ",
4303 i
, sh
->dev
[i
].page
, sh
->dev
[i
].flags
);
4305 seq_printf(seq
, "\n");
4308 static void printall (struct seq_file
*seq
, raid5_conf_t
*conf
)
4310 struct stripe_head
*sh
;
4311 struct hlist_node
*hn
;
4314 spin_lock_irq(&conf
->device_lock
);
4315 for (i
= 0; i
< NR_HASH
; i
++) {
4316 hlist_for_each_entry(sh
, hn
, &conf
->stripe_hashtbl
[i
], hash
) {
4317 if (sh
->raid_conf
!= conf
)
4322 spin_unlock_irq(&conf
->device_lock
);
4326 static void status (struct seq_file
*seq
, mddev_t
*mddev
)
4328 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
4331 seq_printf (seq
, " level %d, %dk chunk, algorithm %d", mddev
->level
, mddev
->chunk_size
>> 10, mddev
->layout
);
4332 seq_printf (seq
, " [%d/%d] [", conf
->raid_disks
, conf
->raid_disks
- mddev
->degraded
);
4333 for (i
= 0; i
< conf
->raid_disks
; i
++)
4334 seq_printf (seq
, "%s",
4335 conf
->disks
[i
].rdev
&&
4336 test_bit(In_sync
, &conf
->disks
[i
].rdev
->flags
) ? "U" : "_");
4337 seq_printf (seq
, "]");
4339 seq_printf (seq
, "\n");
4340 printall(seq
, conf
);
4344 static void print_raid5_conf (raid5_conf_t
*conf
)
4347 struct disk_info
*tmp
;
4349 printk("RAID5 conf printout:\n");
4351 printk("(conf==NULL)\n");
4354 printk(" --- rd:%d wd:%d\n", conf
->raid_disks
,
4355 conf
->raid_disks
- conf
->mddev
->degraded
);
4357 for (i
= 0; i
< conf
->raid_disks
; i
++) {
4358 char b
[BDEVNAME_SIZE
];
4359 tmp
= conf
->disks
+ i
;
4361 printk(" disk %d, o:%d, dev:%s\n",
4362 i
, !test_bit(Faulty
, &tmp
->rdev
->flags
),
4363 bdevname(tmp
->rdev
->bdev
,b
));
4367 static int raid5_spare_active(mddev_t
*mddev
)
4370 raid5_conf_t
*conf
= mddev
->private;
4371 struct disk_info
*tmp
;
4373 for (i
= 0; i
< conf
->raid_disks
; i
++) {
4374 tmp
= conf
->disks
+ i
;
4376 && !test_bit(Faulty
, &tmp
->rdev
->flags
)
4377 && !test_and_set_bit(In_sync
, &tmp
->rdev
->flags
)) {
4378 unsigned long flags
;
4379 spin_lock_irqsave(&conf
->device_lock
, flags
);
4381 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
4384 print_raid5_conf(conf
);
4388 static int raid5_remove_disk(mddev_t
*mddev
, int number
)
4390 raid5_conf_t
*conf
= mddev
->private;
4393 struct disk_info
*p
= conf
->disks
+ number
;
4395 print_raid5_conf(conf
);
4398 if (test_bit(In_sync
, &rdev
->flags
) ||
4399 atomic_read(&rdev
->nr_pending
)) {
4403 /* Only remove non-faulty devices if recovery
4406 if (!test_bit(Faulty
, &rdev
->flags
) &&
4407 mddev
->degraded
<= conf
->max_degraded
) {
4413 if (atomic_read(&rdev
->nr_pending
)) {
4414 /* lost the race, try later */
4421 print_raid5_conf(conf
);
4425 static int raid5_add_disk(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
4427 raid5_conf_t
*conf
= mddev
->private;
4430 struct disk_info
*p
;
4432 int last
= conf
->raid_disks
- 1;
4434 if (mddev
->degraded
> conf
->max_degraded
)
4435 /* no point adding a device */
4438 if (rdev
->raid_disk
>= 0)
4439 first
= last
= rdev
->raid_disk
;
4442 * find the disk ... but prefer rdev->saved_raid_disk
4445 if (rdev
->saved_raid_disk
>= 0 &&
4446 rdev
->saved_raid_disk
>= first
&&
4447 conf
->disks
[rdev
->saved_raid_disk
].rdev
== NULL
)
4448 disk
= rdev
->saved_raid_disk
;
4451 for ( ; disk
<= last
; disk
++)
4452 if ((p
=conf
->disks
+ disk
)->rdev
== NULL
) {
4453 clear_bit(In_sync
, &rdev
->flags
);
4454 rdev
->raid_disk
= disk
;
4456 if (rdev
->saved_raid_disk
!= disk
)
4458 rcu_assign_pointer(p
->rdev
, rdev
);
4461 print_raid5_conf(conf
);
4465 static int raid5_resize(mddev_t
*mddev
, sector_t sectors
)
4467 /* no resync is happening, and there is enough space
4468 * on all devices, so we can resize.
4469 * We need to make sure resync covers any new space.
4470 * If the array is shrinking we should possibly wait until
4471 * any io in the removed space completes, but it hardly seems
4474 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4476 sectors
&= ~((sector_t
)mddev
->chunk_size
/512 - 1);
4477 mddev
->array_sectors
= sectors
* (mddev
->raid_disks
4478 - conf
->max_degraded
);
4479 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4481 if (sectors
/2 > mddev
->size
&& mddev
->recovery_cp
== MaxSector
) {
4482 mddev
->recovery_cp
= mddev
->size
<< 1;
4483 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4485 mddev
->size
= sectors
/2;
4486 mddev
->resync_max_sectors
= sectors
;
4490 #ifdef CONFIG_MD_RAID5_RESHAPE
4491 static int raid5_check_reshape(mddev_t
*mddev
)
4493 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4496 if (mddev
->delta_disks
< 0 ||
4497 mddev
->new_level
!= mddev
->level
)
4498 return -EINVAL
; /* Cannot shrink array or change level yet */
4499 if (mddev
->delta_disks
== 0)
4500 return 0; /* nothing to do */
4502 /* Cannot grow a bitmap yet */
4505 /* Can only proceed if there are plenty of stripe_heads.
4506 * We need a minimum of one full stripe,, and for sensible progress
4507 * it is best to have about 4 times that.
4508 * If we require 4 times, then the default 256 4K stripe_heads will
4509 * allow for chunk sizes up to 256K, which is probably OK.
4510 * If the chunk size is greater, user-space should request more
4511 * stripe_heads first.
4513 if ((mddev
->chunk_size
/ STRIPE_SIZE
) * 4 > conf
->max_nr_stripes
||
4514 (mddev
->new_chunk
/ STRIPE_SIZE
) * 4 > conf
->max_nr_stripes
) {
4515 printk(KERN_WARNING
"raid5: reshape: not enough stripes. Needed %lu\n",
4516 (mddev
->chunk_size
/ STRIPE_SIZE
)*4);
4520 err
= resize_stripes(conf
, conf
->raid_disks
+ mddev
->delta_disks
);
4524 if (mddev
->degraded
> conf
->max_degraded
)
4526 /* looks like we might be able to manage this */
4530 static int raid5_start_reshape(mddev_t
*mddev
)
4532 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4534 struct list_head
*rtmp
;
4536 int added_devices
= 0;
4537 unsigned long flags
;
4539 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4542 rdev_for_each(rdev
, rtmp
, mddev
)
4543 if (rdev
->raid_disk
< 0 &&
4544 !test_bit(Faulty
, &rdev
->flags
))
4547 if (spares
- mddev
->degraded
< mddev
->delta_disks
- conf
->max_degraded
)
4548 /* Not enough devices even to make a degraded array
4553 atomic_set(&conf
->reshape_stripes
, 0);
4554 spin_lock_irq(&conf
->device_lock
);
4555 conf
->previous_raid_disks
= conf
->raid_disks
;
4556 conf
->raid_disks
+= mddev
->delta_disks
;
4557 conf
->expand_progress
= 0;
4558 conf
->expand_lo
= 0;
4559 spin_unlock_irq(&conf
->device_lock
);
4561 /* Add some new drives, as many as will fit.
4562 * We know there are enough to make the newly sized array work.
4564 rdev_for_each(rdev
, rtmp
, mddev
)
4565 if (rdev
->raid_disk
< 0 &&
4566 !test_bit(Faulty
, &rdev
->flags
)) {
4567 if (raid5_add_disk(mddev
, rdev
) == 0) {
4569 set_bit(In_sync
, &rdev
->flags
);
4571 rdev
->recovery_offset
= 0;
4572 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4573 if (sysfs_create_link(&mddev
->kobj
,
4576 "raid5: failed to create "
4577 " link %s for %s\n",
4583 spin_lock_irqsave(&conf
->device_lock
, flags
);
4584 mddev
->degraded
= (conf
->raid_disks
- conf
->previous_raid_disks
) - added_devices
;
4585 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
4586 mddev
->raid_disks
= conf
->raid_disks
;
4587 mddev
->reshape_position
= 0;
4588 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
4590 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4591 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4592 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
4593 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
4594 mddev
->sync_thread
= md_register_thread(md_do_sync
, mddev
,
4596 if (!mddev
->sync_thread
) {
4597 mddev
->recovery
= 0;
4598 spin_lock_irq(&conf
->device_lock
);
4599 mddev
->raid_disks
= conf
->raid_disks
= conf
->previous_raid_disks
;
4600 conf
->expand_progress
= MaxSector
;
4601 spin_unlock_irq(&conf
->device_lock
);
4604 md_wakeup_thread(mddev
->sync_thread
);
4605 md_new_event(mddev
);
4610 static void end_reshape(raid5_conf_t
*conf
)
4612 struct block_device
*bdev
;
4614 if (!test_bit(MD_RECOVERY_INTR
, &conf
->mddev
->recovery
)) {
4615 conf
->mddev
->array_sectors
= 2 * conf
->mddev
->size
*
4616 (conf
->raid_disks
- conf
->max_degraded
);
4617 set_capacity(conf
->mddev
->gendisk
, conf
->mddev
->array_sectors
);
4618 conf
->mddev
->changed
= 1;
4620 bdev
= bdget_disk(conf
->mddev
->gendisk
, 0);
4622 mutex_lock(&bdev
->bd_inode
->i_mutex
);
4623 i_size_write(bdev
->bd_inode
,
4624 (loff_t
)conf
->mddev
->array_sectors
<< 9);
4625 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
4628 spin_lock_irq(&conf
->device_lock
);
4629 conf
->expand_progress
= MaxSector
;
4630 spin_unlock_irq(&conf
->device_lock
);
4631 conf
->mddev
->reshape_position
= MaxSector
;
4633 /* read-ahead size must cover two whole stripes, which is
4634 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
4637 int data_disks
= conf
->previous_raid_disks
- conf
->max_degraded
;
4638 int stripe
= data_disks
*
4639 (conf
->mddev
->chunk_size
/ PAGE_SIZE
);
4640 if (conf
->mddev
->queue
->backing_dev_info
.ra_pages
< 2 * stripe
)
4641 conf
->mddev
->queue
->backing_dev_info
.ra_pages
= 2 * stripe
;
4646 static void raid5_quiesce(mddev_t
*mddev
, int state
)
4648 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4651 case 2: /* resume for a suspend */
4652 wake_up(&conf
->wait_for_overlap
);
4655 case 1: /* stop all writes */
4656 spin_lock_irq(&conf
->device_lock
);
4658 wait_event_lock_irq(conf
->wait_for_stripe
,
4659 atomic_read(&conf
->active_stripes
) == 0 &&
4660 atomic_read(&conf
->active_aligned_reads
) == 0,
4661 conf
->device_lock
, /* nothing */);
4662 spin_unlock_irq(&conf
->device_lock
);
4665 case 0: /* re-enable writes */
4666 spin_lock_irq(&conf
->device_lock
);
4668 wake_up(&conf
->wait_for_stripe
);
4669 wake_up(&conf
->wait_for_overlap
);
4670 spin_unlock_irq(&conf
->device_lock
);
4675 static struct mdk_personality raid6_personality
=
4679 .owner
= THIS_MODULE
,
4680 .make_request
= make_request
,
4684 .error_handler
= error
,
4685 .hot_add_disk
= raid5_add_disk
,
4686 .hot_remove_disk
= raid5_remove_disk
,
4687 .spare_active
= raid5_spare_active
,
4688 .sync_request
= sync_request
,
4689 .resize
= raid5_resize
,
4690 #ifdef CONFIG_MD_RAID5_RESHAPE
4691 .check_reshape
= raid5_check_reshape
,
4692 .start_reshape
= raid5_start_reshape
,
4694 .quiesce
= raid5_quiesce
,
4696 static struct mdk_personality raid5_personality
=
4700 .owner
= THIS_MODULE
,
4701 .make_request
= make_request
,
4705 .error_handler
= error
,
4706 .hot_add_disk
= raid5_add_disk
,
4707 .hot_remove_disk
= raid5_remove_disk
,
4708 .spare_active
= raid5_spare_active
,
4709 .sync_request
= sync_request
,
4710 .resize
= raid5_resize
,
4711 #ifdef CONFIG_MD_RAID5_RESHAPE
4712 .check_reshape
= raid5_check_reshape
,
4713 .start_reshape
= raid5_start_reshape
,
4715 .quiesce
= raid5_quiesce
,
4718 static struct mdk_personality raid4_personality
=
4722 .owner
= THIS_MODULE
,
4723 .make_request
= make_request
,
4727 .error_handler
= error
,
4728 .hot_add_disk
= raid5_add_disk
,
4729 .hot_remove_disk
= raid5_remove_disk
,
4730 .spare_active
= raid5_spare_active
,
4731 .sync_request
= sync_request
,
4732 .resize
= raid5_resize
,
4733 #ifdef CONFIG_MD_RAID5_RESHAPE
4734 .check_reshape
= raid5_check_reshape
,
4735 .start_reshape
= raid5_start_reshape
,
4737 .quiesce
= raid5_quiesce
,
4740 static int __init
raid5_init(void)
4744 e
= raid6_select_algo();
4747 register_md_personality(&raid6_personality
);
4748 register_md_personality(&raid5_personality
);
4749 register_md_personality(&raid4_personality
);
4753 static void raid5_exit(void)
4755 unregister_md_personality(&raid6_personality
);
4756 unregister_md_personality(&raid5_personality
);
4757 unregister_md_personality(&raid4_personality
);
4760 module_init(raid5_init
);
4761 module_exit(raid5_exit
);
4762 MODULE_LICENSE("GPL");
4763 MODULE_ALIAS("md-personality-4"); /* RAID5 */
4764 MODULE_ALIAS("md-raid5");
4765 MODULE_ALIAS("md-raid4");
4766 MODULE_ALIAS("md-level-5");
4767 MODULE_ALIAS("md-level-4");
4768 MODULE_ALIAS("md-personality-8"); /* RAID6 */
4769 MODULE_ALIAS("md-raid6");
4770 MODULE_ALIAS("md-level-6");
4772 /* This used to be two separate modules, they were: */
4773 MODULE_ALIAS("raid5");
4774 MODULE_ALIAS("raid6");