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/blkdev.h>
47 #include <linux/raid/md_k.h>
48 #include <linux/kthread.h>
49 #include <linux/async_tx.h>
50 #include <linux/seq_file.h>
59 #define NR_STRIPES 256
60 #define STRIPE_SIZE PAGE_SIZE
61 #define STRIPE_SHIFT (PAGE_SHIFT - 9)
62 #define STRIPE_SECTORS (STRIPE_SIZE>>9)
63 #define IO_THRESHOLD 1
64 #define BYPASS_THRESHOLD 1
65 #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
66 #define HASH_MASK (NR_HASH - 1)
68 #define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK]))
70 /* bio's attached to a stripe+device for I/O are linked together in bi_sector
71 * order without overlap. There may be several bio's per stripe+device, and
72 * a bio could span several devices.
73 * When walking this list for a particular stripe+device, we must never proceed
74 * beyond a bio that extends past this device, as the next bio might no longer
76 * This macro is used to determine the 'next' bio in the list, given the sector
77 * of the current stripe+device
79 #define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL)
81 * The following can be used to debug the driver
83 #define RAID5_PARANOIA 1
84 #if RAID5_PARANOIA && defined(CONFIG_SMP)
85 # define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock)
87 # define CHECK_DEVLOCK()
95 #define printk_rl(args...) ((void) (printk_ratelimit() && printk(args)))
97 #if !RAID6_USE_EMPTY_ZERO_PAGE
98 /* In .bss so it's zeroed */
99 const char raid6_empty_zero_page
[PAGE_SIZE
] __attribute__((aligned(256)));
103 * We maintain a biased count of active stripes in the bottom 16 bits of
104 * bi_phys_segments, and a count of processed stripes in the upper 16 bits
106 static inline int raid5_bi_phys_segments(struct bio
*bio
)
108 return bio
->bi_phys_segments
& 0xffff;
111 static inline int raid5_bi_hw_segments(struct bio
*bio
)
113 return (bio
->bi_phys_segments
>> 16) & 0xffff;
116 static inline int raid5_dec_bi_phys_segments(struct bio
*bio
)
118 --bio
->bi_phys_segments
;
119 return raid5_bi_phys_segments(bio
);
122 static inline int raid5_dec_bi_hw_segments(struct bio
*bio
)
124 unsigned short val
= raid5_bi_hw_segments(bio
);
127 bio
->bi_phys_segments
= (val
<< 16) | raid5_bi_phys_segments(bio
);
131 static inline void raid5_set_bi_hw_segments(struct bio
*bio
, unsigned int cnt
)
133 bio
->bi_phys_segments
= raid5_bi_phys_segments(bio
) || (cnt
<< 16);
136 static inline int raid6_next_disk(int disk
, int raid_disks
)
139 return (disk
< raid_disks
) ? disk
: 0;
142 static void return_io(struct bio
*return_bi
)
144 struct bio
*bi
= return_bi
;
147 return_bi
= bi
->bi_next
;
155 static void print_raid5_conf (raid5_conf_t
*conf
);
157 static int stripe_operations_active(struct stripe_head
*sh
)
159 return sh
->check_state
|| sh
->reconstruct_state
||
160 test_bit(STRIPE_BIOFILL_RUN
, &sh
->state
) ||
161 test_bit(STRIPE_COMPUTE_RUN
, &sh
->state
);
164 static void __release_stripe(raid5_conf_t
*conf
, struct stripe_head
*sh
)
166 if (atomic_dec_and_test(&sh
->count
)) {
167 BUG_ON(!list_empty(&sh
->lru
));
168 BUG_ON(atomic_read(&conf
->active_stripes
)==0);
169 if (test_bit(STRIPE_HANDLE
, &sh
->state
)) {
170 if (test_bit(STRIPE_DELAYED
, &sh
->state
)) {
171 list_add_tail(&sh
->lru
, &conf
->delayed_list
);
172 blk_plug_device(conf
->mddev
->queue
);
173 } else if (test_bit(STRIPE_BIT_DELAY
, &sh
->state
) &&
174 sh
->bm_seq
- conf
->seq_write
> 0) {
175 list_add_tail(&sh
->lru
, &conf
->bitmap_list
);
176 blk_plug_device(conf
->mddev
->queue
);
178 clear_bit(STRIPE_BIT_DELAY
, &sh
->state
);
179 list_add_tail(&sh
->lru
, &conf
->handle_list
);
181 md_wakeup_thread(conf
->mddev
->thread
);
183 BUG_ON(stripe_operations_active(sh
));
184 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
185 atomic_dec(&conf
->preread_active_stripes
);
186 if (atomic_read(&conf
->preread_active_stripes
) < IO_THRESHOLD
)
187 md_wakeup_thread(conf
->mddev
->thread
);
189 atomic_dec(&conf
->active_stripes
);
190 if (!test_bit(STRIPE_EXPANDING
, &sh
->state
)) {
191 list_add_tail(&sh
->lru
, &conf
->inactive_list
);
192 wake_up(&conf
->wait_for_stripe
);
193 if (conf
->retry_read_aligned
)
194 md_wakeup_thread(conf
->mddev
->thread
);
199 static void release_stripe(struct stripe_head
*sh
)
201 raid5_conf_t
*conf
= sh
->raid_conf
;
204 spin_lock_irqsave(&conf
->device_lock
, flags
);
205 __release_stripe(conf
, sh
);
206 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
209 static inline void remove_hash(struct stripe_head
*sh
)
211 pr_debug("remove_hash(), stripe %llu\n",
212 (unsigned long long)sh
->sector
);
214 hlist_del_init(&sh
->hash
);
217 static inline void insert_hash(raid5_conf_t
*conf
, struct stripe_head
*sh
)
219 struct hlist_head
*hp
= stripe_hash(conf
, sh
->sector
);
221 pr_debug("insert_hash(), stripe %llu\n",
222 (unsigned long long)sh
->sector
);
225 hlist_add_head(&sh
->hash
, hp
);
229 /* find an idle stripe, make sure it is unhashed, and return it. */
230 static struct stripe_head
*get_free_stripe(raid5_conf_t
*conf
)
232 struct stripe_head
*sh
= NULL
;
233 struct list_head
*first
;
236 if (list_empty(&conf
->inactive_list
))
238 first
= conf
->inactive_list
.next
;
239 sh
= list_entry(first
, struct stripe_head
, lru
);
240 list_del_init(first
);
242 atomic_inc(&conf
->active_stripes
);
247 static void shrink_buffers(struct stripe_head
*sh
, int num
)
252 for (i
=0; i
<num
; i
++) {
256 sh
->dev
[i
].page
= NULL
;
261 static int grow_buffers(struct stripe_head
*sh
, int num
)
265 for (i
=0; i
<num
; i
++) {
268 if (!(page
= alloc_page(GFP_KERNEL
))) {
271 sh
->dev
[i
].page
= page
;
276 static void raid5_build_block(struct stripe_head
*sh
, int i
);
278 static void init_stripe(struct stripe_head
*sh
, sector_t sector
, int pd_idx
, int disks
)
280 raid5_conf_t
*conf
= sh
->raid_conf
;
283 BUG_ON(atomic_read(&sh
->count
) != 0);
284 BUG_ON(test_bit(STRIPE_HANDLE
, &sh
->state
));
285 BUG_ON(stripe_operations_active(sh
));
288 pr_debug("init_stripe called, stripe %llu\n",
289 (unsigned long long)sh
->sector
);
299 for (i
= sh
->disks
; i
--; ) {
300 struct r5dev
*dev
= &sh
->dev
[i
];
302 if (dev
->toread
|| dev
->read
|| dev
->towrite
|| dev
->written
||
303 test_bit(R5_LOCKED
, &dev
->flags
)) {
304 printk(KERN_ERR
"sector=%llx i=%d %p %p %p %p %d\n",
305 (unsigned long long)sh
->sector
, i
, dev
->toread
,
306 dev
->read
, dev
->towrite
, dev
->written
,
307 test_bit(R5_LOCKED
, &dev
->flags
));
311 raid5_build_block(sh
, i
);
313 insert_hash(conf
, sh
);
316 static struct stripe_head
*__find_stripe(raid5_conf_t
*conf
, sector_t sector
, int disks
)
318 struct stripe_head
*sh
;
319 struct hlist_node
*hn
;
322 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector
);
323 hlist_for_each_entry(sh
, hn
, stripe_hash(conf
, sector
), hash
)
324 if (sh
->sector
== sector
&& sh
->disks
== disks
)
326 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector
);
330 static void unplug_slaves(mddev_t
*mddev
);
331 static void raid5_unplug_device(struct request_queue
*q
);
333 static struct stripe_head
*get_active_stripe(raid5_conf_t
*conf
, sector_t sector
, int disks
,
334 int pd_idx
, int noblock
)
336 struct stripe_head
*sh
;
338 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector
);
340 spin_lock_irq(&conf
->device_lock
);
343 wait_event_lock_irq(conf
->wait_for_stripe
,
345 conf
->device_lock
, /* nothing */);
346 sh
= __find_stripe(conf
, sector
, disks
);
348 if (!conf
->inactive_blocked
)
349 sh
= get_free_stripe(conf
);
350 if (noblock
&& sh
== NULL
)
353 conf
->inactive_blocked
= 1;
354 wait_event_lock_irq(conf
->wait_for_stripe
,
355 !list_empty(&conf
->inactive_list
) &&
356 (atomic_read(&conf
->active_stripes
)
357 < (conf
->max_nr_stripes
*3/4)
358 || !conf
->inactive_blocked
),
360 raid5_unplug_device(conf
->mddev
->queue
)
362 conf
->inactive_blocked
= 0;
364 init_stripe(sh
, sector
, pd_idx
, disks
);
366 if (atomic_read(&sh
->count
)) {
367 BUG_ON(!list_empty(&sh
->lru
));
369 if (!test_bit(STRIPE_HANDLE
, &sh
->state
))
370 atomic_inc(&conf
->active_stripes
);
371 if (list_empty(&sh
->lru
) &&
372 !test_bit(STRIPE_EXPANDING
, &sh
->state
))
374 list_del_init(&sh
->lru
);
377 } while (sh
== NULL
);
380 atomic_inc(&sh
->count
);
382 spin_unlock_irq(&conf
->device_lock
);
387 raid5_end_read_request(struct bio
*bi
, int error
);
389 raid5_end_write_request(struct bio
*bi
, int error
);
391 static void ops_run_io(struct stripe_head
*sh
, struct stripe_head_state
*s
)
393 raid5_conf_t
*conf
= sh
->raid_conf
;
394 int i
, disks
= sh
->disks
;
398 for (i
= disks
; i
--; ) {
402 if (test_and_clear_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
))
404 else if (test_and_clear_bit(R5_Wantread
, &sh
->dev
[i
].flags
))
409 bi
= &sh
->dev
[i
].req
;
413 bi
->bi_end_io
= raid5_end_write_request
;
415 bi
->bi_end_io
= raid5_end_read_request
;
418 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
419 if (rdev
&& test_bit(Faulty
, &rdev
->flags
))
422 atomic_inc(&rdev
->nr_pending
);
426 if (s
->syncing
|| s
->expanding
|| s
->expanded
)
427 md_sync_acct(rdev
->bdev
, STRIPE_SECTORS
);
429 set_bit(STRIPE_IO_STARTED
, &sh
->state
);
431 bi
->bi_bdev
= rdev
->bdev
;
432 pr_debug("%s: for %llu schedule op %ld on disc %d\n",
433 __func__
, (unsigned long long)sh
->sector
,
435 atomic_inc(&sh
->count
);
436 bi
->bi_sector
= sh
->sector
+ rdev
->data_offset
;
437 bi
->bi_flags
= 1 << BIO_UPTODATE
;
441 bi
->bi_io_vec
= &sh
->dev
[i
].vec
;
442 bi
->bi_io_vec
[0].bv_len
= STRIPE_SIZE
;
443 bi
->bi_io_vec
[0].bv_offset
= 0;
444 bi
->bi_size
= STRIPE_SIZE
;
447 test_bit(R5_ReWrite
, &sh
->dev
[i
].flags
))
448 atomic_add(STRIPE_SECTORS
,
449 &rdev
->corrected_errors
);
450 generic_make_request(bi
);
453 set_bit(STRIPE_DEGRADED
, &sh
->state
);
454 pr_debug("skip op %ld on disc %d for sector %llu\n",
455 bi
->bi_rw
, i
, (unsigned long long)sh
->sector
);
456 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
457 set_bit(STRIPE_HANDLE
, &sh
->state
);
462 static struct dma_async_tx_descriptor
*
463 async_copy_data(int frombio
, struct bio
*bio
, struct page
*page
,
464 sector_t sector
, struct dma_async_tx_descriptor
*tx
)
467 struct page
*bio_page
;
471 if (bio
->bi_sector
>= sector
)
472 page_offset
= (signed)(bio
->bi_sector
- sector
) * 512;
474 page_offset
= (signed)(sector
- bio
->bi_sector
) * -512;
475 bio_for_each_segment(bvl
, bio
, i
) {
476 int len
= bio_iovec_idx(bio
, i
)->bv_len
;
480 if (page_offset
< 0) {
481 b_offset
= -page_offset
;
482 page_offset
+= b_offset
;
486 if (len
> 0 && page_offset
+ len
> STRIPE_SIZE
)
487 clen
= STRIPE_SIZE
- page_offset
;
492 b_offset
+= bio_iovec_idx(bio
, i
)->bv_offset
;
493 bio_page
= bio_iovec_idx(bio
, i
)->bv_page
;
495 tx
= async_memcpy(page
, bio_page
, page_offset
,
500 tx
= async_memcpy(bio_page
, page
, b_offset
,
505 if (clen
< len
) /* hit end of page */
513 static void ops_complete_biofill(void *stripe_head_ref
)
515 struct stripe_head
*sh
= stripe_head_ref
;
516 struct bio
*return_bi
= NULL
;
517 raid5_conf_t
*conf
= sh
->raid_conf
;
520 pr_debug("%s: stripe %llu\n", __func__
,
521 (unsigned long long)sh
->sector
);
523 /* clear completed biofills */
524 spin_lock_irq(&conf
->device_lock
);
525 for (i
= sh
->disks
; i
--; ) {
526 struct r5dev
*dev
= &sh
->dev
[i
];
528 /* acknowledge completion of a biofill operation */
529 /* and check if we need to reply to a read request,
530 * new R5_Wantfill requests are held off until
531 * !STRIPE_BIOFILL_RUN
533 if (test_and_clear_bit(R5_Wantfill
, &dev
->flags
)) {
534 struct bio
*rbi
, *rbi2
;
539 while (rbi
&& rbi
->bi_sector
<
540 dev
->sector
+ STRIPE_SECTORS
) {
541 rbi2
= r5_next_bio(rbi
, dev
->sector
);
542 if (!raid5_dec_bi_phys_segments(rbi
)) {
543 rbi
->bi_next
= return_bi
;
550 spin_unlock_irq(&conf
->device_lock
);
551 clear_bit(STRIPE_BIOFILL_RUN
, &sh
->state
);
553 return_io(return_bi
);
555 set_bit(STRIPE_HANDLE
, &sh
->state
);
559 static void ops_run_biofill(struct stripe_head
*sh
)
561 struct dma_async_tx_descriptor
*tx
= NULL
;
562 raid5_conf_t
*conf
= sh
->raid_conf
;
565 pr_debug("%s: stripe %llu\n", __func__
,
566 (unsigned long long)sh
->sector
);
568 for (i
= sh
->disks
; i
--; ) {
569 struct r5dev
*dev
= &sh
->dev
[i
];
570 if (test_bit(R5_Wantfill
, &dev
->flags
)) {
572 spin_lock_irq(&conf
->device_lock
);
573 dev
->read
= rbi
= dev
->toread
;
575 spin_unlock_irq(&conf
->device_lock
);
576 while (rbi
&& rbi
->bi_sector
<
577 dev
->sector
+ STRIPE_SECTORS
) {
578 tx
= async_copy_data(0, rbi
, dev
->page
,
580 rbi
= r5_next_bio(rbi
, dev
->sector
);
585 atomic_inc(&sh
->count
);
586 async_trigger_callback(ASYNC_TX_DEP_ACK
| ASYNC_TX_ACK
, tx
,
587 ops_complete_biofill
, sh
);
590 static void ops_complete_compute5(void *stripe_head_ref
)
592 struct stripe_head
*sh
= stripe_head_ref
;
593 int target
= sh
->ops
.target
;
594 struct r5dev
*tgt
= &sh
->dev
[target
];
596 pr_debug("%s: stripe %llu\n", __func__
,
597 (unsigned long long)sh
->sector
);
599 set_bit(R5_UPTODATE
, &tgt
->flags
);
600 BUG_ON(!test_bit(R5_Wantcompute
, &tgt
->flags
));
601 clear_bit(R5_Wantcompute
, &tgt
->flags
);
602 clear_bit(STRIPE_COMPUTE_RUN
, &sh
->state
);
603 if (sh
->check_state
== check_state_compute_run
)
604 sh
->check_state
= check_state_compute_result
;
605 set_bit(STRIPE_HANDLE
, &sh
->state
);
609 static struct dma_async_tx_descriptor
*ops_run_compute5(struct stripe_head
*sh
)
611 /* kernel stack size limits the total number of disks */
612 int disks
= sh
->disks
;
613 struct page
*xor_srcs
[disks
];
614 int target
= sh
->ops
.target
;
615 struct r5dev
*tgt
= &sh
->dev
[target
];
616 struct page
*xor_dest
= tgt
->page
;
618 struct dma_async_tx_descriptor
*tx
;
621 pr_debug("%s: stripe %llu block: %d\n",
622 __func__
, (unsigned long long)sh
->sector
, target
);
623 BUG_ON(!test_bit(R5_Wantcompute
, &tgt
->flags
));
625 for (i
= disks
; i
--; )
627 xor_srcs
[count
++] = sh
->dev
[i
].page
;
629 atomic_inc(&sh
->count
);
631 if (unlikely(count
== 1))
632 tx
= async_memcpy(xor_dest
, xor_srcs
[0], 0, 0, STRIPE_SIZE
,
633 0, NULL
, ops_complete_compute5
, sh
);
635 tx
= async_xor(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
,
636 ASYNC_TX_XOR_ZERO_DST
, NULL
,
637 ops_complete_compute5
, sh
);
642 static void ops_complete_prexor(void *stripe_head_ref
)
644 struct stripe_head
*sh
= stripe_head_ref
;
646 pr_debug("%s: stripe %llu\n", __func__
,
647 (unsigned long long)sh
->sector
);
650 static struct dma_async_tx_descriptor
*
651 ops_run_prexor(struct stripe_head
*sh
, struct dma_async_tx_descriptor
*tx
)
653 /* kernel stack size limits the total number of disks */
654 int disks
= sh
->disks
;
655 struct page
*xor_srcs
[disks
];
656 int count
= 0, pd_idx
= sh
->pd_idx
, i
;
658 /* existing parity data subtracted */
659 struct page
*xor_dest
= xor_srcs
[count
++] = sh
->dev
[pd_idx
].page
;
661 pr_debug("%s: stripe %llu\n", __func__
,
662 (unsigned long long)sh
->sector
);
664 for (i
= disks
; i
--; ) {
665 struct r5dev
*dev
= &sh
->dev
[i
];
666 /* Only process blocks that are known to be uptodate */
667 if (test_bit(R5_Wantdrain
, &dev
->flags
))
668 xor_srcs
[count
++] = dev
->page
;
671 tx
= async_xor(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
,
672 ASYNC_TX_DEP_ACK
| ASYNC_TX_XOR_DROP_DST
, tx
,
673 ops_complete_prexor
, sh
);
678 static struct dma_async_tx_descriptor
*
679 ops_run_biodrain(struct stripe_head
*sh
, struct dma_async_tx_descriptor
*tx
)
681 int disks
= sh
->disks
;
684 pr_debug("%s: stripe %llu\n", __func__
,
685 (unsigned long long)sh
->sector
);
687 for (i
= disks
; i
--; ) {
688 struct r5dev
*dev
= &sh
->dev
[i
];
691 if (test_and_clear_bit(R5_Wantdrain
, &dev
->flags
)) {
694 spin_lock(&sh
->lock
);
695 chosen
= dev
->towrite
;
697 BUG_ON(dev
->written
);
698 wbi
= dev
->written
= chosen
;
699 spin_unlock(&sh
->lock
);
701 while (wbi
&& wbi
->bi_sector
<
702 dev
->sector
+ STRIPE_SECTORS
) {
703 tx
= async_copy_data(1, wbi
, dev
->page
,
705 wbi
= r5_next_bio(wbi
, dev
->sector
);
713 static void ops_complete_postxor(void *stripe_head_ref
)
715 struct stripe_head
*sh
= stripe_head_ref
;
716 int disks
= sh
->disks
, i
, pd_idx
= sh
->pd_idx
;
718 pr_debug("%s: stripe %llu\n", __func__
,
719 (unsigned long long)sh
->sector
);
721 for (i
= disks
; i
--; ) {
722 struct r5dev
*dev
= &sh
->dev
[i
];
723 if (dev
->written
|| i
== pd_idx
)
724 set_bit(R5_UPTODATE
, &dev
->flags
);
727 if (sh
->reconstruct_state
== reconstruct_state_drain_run
)
728 sh
->reconstruct_state
= reconstruct_state_drain_result
;
729 else if (sh
->reconstruct_state
== reconstruct_state_prexor_drain_run
)
730 sh
->reconstruct_state
= reconstruct_state_prexor_drain_result
;
732 BUG_ON(sh
->reconstruct_state
!= reconstruct_state_run
);
733 sh
->reconstruct_state
= reconstruct_state_result
;
736 set_bit(STRIPE_HANDLE
, &sh
->state
);
741 ops_run_postxor(struct stripe_head
*sh
, struct dma_async_tx_descriptor
*tx
)
743 /* kernel stack size limits the total number of disks */
744 int disks
= sh
->disks
;
745 struct page
*xor_srcs
[disks
];
747 int count
= 0, pd_idx
= sh
->pd_idx
, i
;
748 struct page
*xor_dest
;
752 pr_debug("%s: stripe %llu\n", __func__
,
753 (unsigned long long)sh
->sector
);
755 /* check if prexor is active which means only process blocks
756 * that are part of a read-modify-write (written)
758 if (sh
->reconstruct_state
== reconstruct_state_prexor_drain_run
) {
760 xor_dest
= xor_srcs
[count
++] = sh
->dev
[pd_idx
].page
;
761 for (i
= disks
; i
--; ) {
762 struct r5dev
*dev
= &sh
->dev
[i
];
764 xor_srcs
[count
++] = dev
->page
;
767 xor_dest
= sh
->dev
[pd_idx
].page
;
768 for (i
= disks
; i
--; ) {
769 struct r5dev
*dev
= &sh
->dev
[i
];
771 xor_srcs
[count
++] = dev
->page
;
775 /* 1/ if we prexor'd then the dest is reused as a source
776 * 2/ if we did not prexor then we are redoing the parity
777 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
778 * for the synchronous xor case
780 flags
= ASYNC_TX_DEP_ACK
| ASYNC_TX_ACK
|
781 (prexor
? ASYNC_TX_XOR_DROP_DST
: ASYNC_TX_XOR_ZERO_DST
);
783 atomic_inc(&sh
->count
);
785 if (unlikely(count
== 1)) {
786 flags
&= ~(ASYNC_TX_XOR_DROP_DST
| ASYNC_TX_XOR_ZERO_DST
);
787 tx
= async_memcpy(xor_dest
, xor_srcs
[0], 0, 0, STRIPE_SIZE
,
788 flags
, tx
, ops_complete_postxor
, sh
);
790 tx
= async_xor(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
,
791 flags
, tx
, ops_complete_postxor
, sh
);
794 static void ops_complete_check(void *stripe_head_ref
)
796 struct stripe_head
*sh
= stripe_head_ref
;
798 pr_debug("%s: stripe %llu\n", __func__
,
799 (unsigned long long)sh
->sector
);
801 sh
->check_state
= check_state_check_result
;
802 set_bit(STRIPE_HANDLE
, &sh
->state
);
806 static void ops_run_check(struct stripe_head
*sh
)
808 /* kernel stack size limits the total number of disks */
809 int disks
= sh
->disks
;
810 struct page
*xor_srcs
[disks
];
811 struct dma_async_tx_descriptor
*tx
;
813 int count
= 0, pd_idx
= sh
->pd_idx
, i
;
814 struct page
*xor_dest
= xor_srcs
[count
++] = sh
->dev
[pd_idx
].page
;
816 pr_debug("%s: stripe %llu\n", __func__
,
817 (unsigned long long)sh
->sector
);
819 for (i
= disks
; i
--; ) {
820 struct r5dev
*dev
= &sh
->dev
[i
];
822 xor_srcs
[count
++] = dev
->page
;
825 tx
= async_xor_zero_sum(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
,
826 &sh
->ops
.zero_sum_result
, 0, NULL
, NULL
, NULL
);
828 atomic_inc(&sh
->count
);
829 tx
= async_trigger_callback(ASYNC_TX_DEP_ACK
| ASYNC_TX_ACK
, tx
,
830 ops_complete_check
, sh
);
833 static void raid5_run_ops(struct stripe_head
*sh
, unsigned long ops_request
)
835 int overlap_clear
= 0, i
, disks
= sh
->disks
;
836 struct dma_async_tx_descriptor
*tx
= NULL
;
838 if (test_bit(STRIPE_OP_BIOFILL
, &ops_request
)) {
843 if (test_bit(STRIPE_OP_COMPUTE_BLK
, &ops_request
)) {
844 tx
= ops_run_compute5(sh
);
845 /* terminate the chain if postxor is not set to be run */
846 if (tx
&& !test_bit(STRIPE_OP_POSTXOR
, &ops_request
))
850 if (test_bit(STRIPE_OP_PREXOR
, &ops_request
))
851 tx
= ops_run_prexor(sh
, tx
);
853 if (test_bit(STRIPE_OP_BIODRAIN
, &ops_request
)) {
854 tx
= ops_run_biodrain(sh
, tx
);
858 if (test_bit(STRIPE_OP_POSTXOR
, &ops_request
))
859 ops_run_postxor(sh
, tx
);
861 if (test_bit(STRIPE_OP_CHECK
, &ops_request
))
865 for (i
= disks
; i
--; ) {
866 struct r5dev
*dev
= &sh
->dev
[i
];
867 if (test_and_clear_bit(R5_Overlap
, &dev
->flags
))
868 wake_up(&sh
->raid_conf
->wait_for_overlap
);
872 static int grow_one_stripe(raid5_conf_t
*conf
)
874 struct stripe_head
*sh
;
875 sh
= kmem_cache_alloc(conf
->slab_cache
, GFP_KERNEL
);
878 memset(sh
, 0, sizeof(*sh
) + (conf
->raid_disks
-1)*sizeof(struct r5dev
));
879 sh
->raid_conf
= conf
;
880 spin_lock_init(&sh
->lock
);
882 if (grow_buffers(sh
, conf
->raid_disks
)) {
883 shrink_buffers(sh
, conf
->raid_disks
);
884 kmem_cache_free(conf
->slab_cache
, sh
);
887 sh
->disks
= conf
->raid_disks
;
888 /* we just created an active stripe so... */
889 atomic_set(&sh
->count
, 1);
890 atomic_inc(&conf
->active_stripes
);
891 INIT_LIST_HEAD(&sh
->lru
);
896 static int grow_stripes(raid5_conf_t
*conf
, int num
)
898 struct kmem_cache
*sc
;
899 int devs
= conf
->raid_disks
;
901 sprintf(conf
->cache_name
[0], "raid5-%s", mdname(conf
->mddev
));
902 sprintf(conf
->cache_name
[1], "raid5-%s-alt", mdname(conf
->mddev
));
903 conf
->active_name
= 0;
904 sc
= kmem_cache_create(conf
->cache_name
[conf
->active_name
],
905 sizeof(struct stripe_head
)+(devs
-1)*sizeof(struct r5dev
),
909 conf
->slab_cache
= sc
;
910 conf
->pool_size
= devs
;
912 if (!grow_one_stripe(conf
))
917 #ifdef CONFIG_MD_RAID5_RESHAPE
918 static int resize_stripes(raid5_conf_t
*conf
, int newsize
)
920 /* Make all the stripes able to hold 'newsize' devices.
921 * New slots in each stripe get 'page' set to a new page.
923 * This happens in stages:
924 * 1/ create a new kmem_cache and allocate the required number of
926 * 2/ gather all the old stripe_heads and tranfer the pages across
927 * to the new stripe_heads. This will have the side effect of
928 * freezing the array as once all stripe_heads have been collected,
929 * no IO will be possible. Old stripe heads are freed once their
930 * pages have been transferred over, and the old kmem_cache is
931 * freed when all stripes are done.
932 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
933 * we simple return a failre status - no need to clean anything up.
934 * 4/ allocate new pages for the new slots in the new stripe_heads.
935 * If this fails, we don't bother trying the shrink the
936 * stripe_heads down again, we just leave them as they are.
937 * As each stripe_head is processed the new one is released into
940 * Once step2 is started, we cannot afford to wait for a write,
941 * so we use GFP_NOIO allocations.
943 struct stripe_head
*osh
, *nsh
;
944 LIST_HEAD(newstripes
);
945 struct disk_info
*ndisks
;
947 struct kmem_cache
*sc
;
950 if (newsize
<= conf
->pool_size
)
951 return 0; /* never bother to shrink */
953 err
= md_allow_write(conf
->mddev
);
958 sc
= kmem_cache_create(conf
->cache_name
[1-conf
->active_name
],
959 sizeof(struct stripe_head
)+(newsize
-1)*sizeof(struct r5dev
),
964 for (i
= conf
->max_nr_stripes
; i
; i
--) {
965 nsh
= kmem_cache_alloc(sc
, GFP_KERNEL
);
969 memset(nsh
, 0, sizeof(*nsh
) + (newsize
-1)*sizeof(struct r5dev
));
971 nsh
->raid_conf
= conf
;
972 spin_lock_init(&nsh
->lock
);
974 list_add(&nsh
->lru
, &newstripes
);
977 /* didn't get enough, give up */
978 while (!list_empty(&newstripes
)) {
979 nsh
= list_entry(newstripes
.next
, struct stripe_head
, lru
);
981 kmem_cache_free(sc
, nsh
);
983 kmem_cache_destroy(sc
);
986 /* Step 2 - Must use GFP_NOIO now.
987 * OK, we have enough stripes, start collecting inactive
988 * stripes and copying them over
990 list_for_each_entry(nsh
, &newstripes
, lru
) {
991 spin_lock_irq(&conf
->device_lock
);
992 wait_event_lock_irq(conf
->wait_for_stripe
,
993 !list_empty(&conf
->inactive_list
),
995 unplug_slaves(conf
->mddev
)
997 osh
= get_free_stripe(conf
);
998 spin_unlock_irq(&conf
->device_lock
);
999 atomic_set(&nsh
->count
, 1);
1000 for(i
=0; i
<conf
->pool_size
; i
++)
1001 nsh
->dev
[i
].page
= osh
->dev
[i
].page
;
1002 for( ; i
<newsize
; i
++)
1003 nsh
->dev
[i
].page
= NULL
;
1004 kmem_cache_free(conf
->slab_cache
, osh
);
1006 kmem_cache_destroy(conf
->slab_cache
);
1009 * At this point, we are holding all the stripes so the array
1010 * is completely stalled, so now is a good time to resize
1013 ndisks
= kzalloc(newsize
* sizeof(struct disk_info
), GFP_NOIO
);
1015 for (i
=0; i
<conf
->raid_disks
; i
++)
1016 ndisks
[i
] = conf
->disks
[i
];
1018 conf
->disks
= ndisks
;
1022 /* Step 4, return new stripes to service */
1023 while(!list_empty(&newstripes
)) {
1024 nsh
= list_entry(newstripes
.next
, struct stripe_head
, lru
);
1025 list_del_init(&nsh
->lru
);
1026 for (i
=conf
->raid_disks
; i
< newsize
; i
++)
1027 if (nsh
->dev
[i
].page
== NULL
) {
1028 struct page
*p
= alloc_page(GFP_NOIO
);
1029 nsh
->dev
[i
].page
= p
;
1033 release_stripe(nsh
);
1035 /* critical section pass, GFP_NOIO no longer needed */
1037 conf
->slab_cache
= sc
;
1038 conf
->active_name
= 1-conf
->active_name
;
1039 conf
->pool_size
= newsize
;
1044 static int drop_one_stripe(raid5_conf_t
*conf
)
1046 struct stripe_head
*sh
;
1048 spin_lock_irq(&conf
->device_lock
);
1049 sh
= get_free_stripe(conf
);
1050 spin_unlock_irq(&conf
->device_lock
);
1053 BUG_ON(atomic_read(&sh
->count
));
1054 shrink_buffers(sh
, conf
->pool_size
);
1055 kmem_cache_free(conf
->slab_cache
, sh
);
1056 atomic_dec(&conf
->active_stripes
);
1060 static void shrink_stripes(raid5_conf_t
*conf
)
1062 while (drop_one_stripe(conf
))
1065 if (conf
->slab_cache
)
1066 kmem_cache_destroy(conf
->slab_cache
);
1067 conf
->slab_cache
= NULL
;
1070 static void raid5_end_read_request(struct bio
* bi
, int error
)
1072 struct stripe_head
*sh
= bi
->bi_private
;
1073 raid5_conf_t
*conf
= sh
->raid_conf
;
1074 int disks
= sh
->disks
, i
;
1075 int uptodate
= test_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1076 char b
[BDEVNAME_SIZE
];
1080 for (i
=0 ; i
<disks
; i
++)
1081 if (bi
== &sh
->dev
[i
].req
)
1084 pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n",
1085 (unsigned long long)sh
->sector
, i
, atomic_read(&sh
->count
),
1093 set_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
);
1094 if (test_bit(R5_ReadError
, &sh
->dev
[i
].flags
)) {
1095 rdev
= conf
->disks
[i
].rdev
;
1096 printk_rl(KERN_INFO
"raid5:%s: read error corrected"
1097 " (%lu sectors at %llu on %s)\n",
1098 mdname(conf
->mddev
), STRIPE_SECTORS
,
1099 (unsigned long long)(sh
->sector
1100 + rdev
->data_offset
),
1101 bdevname(rdev
->bdev
, b
));
1102 clear_bit(R5_ReadError
, &sh
->dev
[i
].flags
);
1103 clear_bit(R5_ReWrite
, &sh
->dev
[i
].flags
);
1105 if (atomic_read(&conf
->disks
[i
].rdev
->read_errors
))
1106 atomic_set(&conf
->disks
[i
].rdev
->read_errors
, 0);
1108 const char *bdn
= bdevname(conf
->disks
[i
].rdev
->bdev
, b
);
1110 rdev
= conf
->disks
[i
].rdev
;
1112 clear_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
);
1113 atomic_inc(&rdev
->read_errors
);
1114 if (conf
->mddev
->degraded
)
1115 printk_rl(KERN_WARNING
1116 "raid5:%s: read error not correctable "
1117 "(sector %llu on %s).\n",
1118 mdname(conf
->mddev
),
1119 (unsigned long long)(sh
->sector
1120 + rdev
->data_offset
),
1122 else if (test_bit(R5_ReWrite
, &sh
->dev
[i
].flags
))
1124 printk_rl(KERN_WARNING
1125 "raid5:%s: read error NOT corrected!! "
1126 "(sector %llu on %s).\n",
1127 mdname(conf
->mddev
),
1128 (unsigned long long)(sh
->sector
1129 + rdev
->data_offset
),
1131 else if (atomic_read(&rdev
->read_errors
)
1132 > conf
->max_nr_stripes
)
1134 "raid5:%s: Too many read errors, failing device %s.\n",
1135 mdname(conf
->mddev
), bdn
);
1139 set_bit(R5_ReadError
, &sh
->dev
[i
].flags
);
1141 clear_bit(R5_ReadError
, &sh
->dev
[i
].flags
);
1142 clear_bit(R5_ReWrite
, &sh
->dev
[i
].flags
);
1143 md_error(conf
->mddev
, rdev
);
1146 rdev_dec_pending(conf
->disks
[i
].rdev
, conf
->mddev
);
1147 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
1148 set_bit(STRIPE_HANDLE
, &sh
->state
);
1152 static void raid5_end_write_request(struct bio
*bi
, int error
)
1154 struct stripe_head
*sh
= bi
->bi_private
;
1155 raid5_conf_t
*conf
= sh
->raid_conf
;
1156 int disks
= sh
->disks
, i
;
1157 int uptodate
= test_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1159 for (i
=0 ; i
<disks
; i
++)
1160 if (bi
== &sh
->dev
[i
].req
)
1163 pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n",
1164 (unsigned long long)sh
->sector
, i
, atomic_read(&sh
->count
),
1172 md_error(conf
->mddev
, conf
->disks
[i
].rdev
);
1174 rdev_dec_pending(conf
->disks
[i
].rdev
, conf
->mddev
);
1176 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
1177 set_bit(STRIPE_HANDLE
, &sh
->state
);
1182 static sector_t
compute_blocknr(struct stripe_head
*sh
, int i
);
1184 static void raid5_build_block(struct stripe_head
*sh
, int i
)
1186 struct r5dev
*dev
= &sh
->dev
[i
];
1188 bio_init(&dev
->req
);
1189 dev
->req
.bi_io_vec
= &dev
->vec
;
1191 dev
->req
.bi_max_vecs
++;
1192 dev
->vec
.bv_page
= dev
->page
;
1193 dev
->vec
.bv_len
= STRIPE_SIZE
;
1194 dev
->vec
.bv_offset
= 0;
1196 dev
->req
.bi_sector
= sh
->sector
;
1197 dev
->req
.bi_private
= sh
;
1200 dev
->sector
= compute_blocknr(sh
, i
);
1203 static void error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1205 char b
[BDEVNAME_SIZE
];
1206 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
1207 pr_debug("raid5: error called\n");
1209 if (!test_bit(Faulty
, &rdev
->flags
)) {
1210 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
1211 if (test_and_clear_bit(In_sync
, &rdev
->flags
)) {
1212 unsigned long flags
;
1213 spin_lock_irqsave(&conf
->device_lock
, flags
);
1215 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1217 * if recovery was running, make sure it aborts.
1219 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
1221 set_bit(Faulty
, &rdev
->flags
);
1223 "raid5: Disk failure on %s, disabling device.\n"
1224 "raid5: Operation continuing on %d devices.\n",
1225 bdevname(rdev
->bdev
,b
), conf
->raid_disks
- mddev
->degraded
);
1230 * Input: a 'big' sector number,
1231 * Output: index of the data and parity disk, and the sector # in them.
1233 static sector_t
raid5_compute_sector(sector_t r_sector
, unsigned int raid_disks
,
1234 unsigned int data_disks
, unsigned int * dd_idx
,
1235 unsigned int * pd_idx
, raid5_conf_t
*conf
)
1238 unsigned long chunk_number
;
1239 unsigned int chunk_offset
;
1240 sector_t new_sector
;
1241 int sectors_per_chunk
= conf
->chunk_size
>> 9;
1243 /* First compute the information on this sector */
1246 * Compute the chunk number and the sector offset inside the chunk
1248 chunk_offset
= sector_div(r_sector
, sectors_per_chunk
);
1249 chunk_number
= r_sector
;
1250 BUG_ON(r_sector
!= chunk_number
);
1253 * Compute the stripe number
1255 stripe
= chunk_number
/ data_disks
;
1258 * Compute the data disk and parity disk indexes inside the stripe
1260 *dd_idx
= chunk_number
% data_disks
;
1263 * Select the parity disk based on the user selected algorithm.
1265 switch(conf
->level
) {
1267 *pd_idx
= data_disks
;
1270 switch (conf
->algorithm
) {
1271 case ALGORITHM_LEFT_ASYMMETRIC
:
1272 *pd_idx
= data_disks
- stripe
% raid_disks
;
1273 if (*dd_idx
>= *pd_idx
)
1276 case ALGORITHM_RIGHT_ASYMMETRIC
:
1277 *pd_idx
= stripe
% raid_disks
;
1278 if (*dd_idx
>= *pd_idx
)
1281 case ALGORITHM_LEFT_SYMMETRIC
:
1282 *pd_idx
= data_disks
- stripe
% raid_disks
;
1283 *dd_idx
= (*pd_idx
+ 1 + *dd_idx
) % raid_disks
;
1285 case ALGORITHM_RIGHT_SYMMETRIC
:
1286 *pd_idx
= stripe
% raid_disks
;
1287 *dd_idx
= (*pd_idx
+ 1 + *dd_idx
) % raid_disks
;
1290 printk(KERN_ERR
"raid5: unsupported algorithm %d\n",
1296 /**** FIX THIS ****/
1297 switch (conf
->algorithm
) {
1298 case ALGORITHM_LEFT_ASYMMETRIC
:
1299 *pd_idx
= raid_disks
- 1 - (stripe
% raid_disks
);
1300 if (*pd_idx
== raid_disks
-1)
1301 (*dd_idx
)++; /* Q D D D P */
1302 else if (*dd_idx
>= *pd_idx
)
1303 (*dd_idx
) += 2; /* D D P Q D */
1305 case ALGORITHM_RIGHT_ASYMMETRIC
:
1306 *pd_idx
= stripe
% raid_disks
;
1307 if (*pd_idx
== raid_disks
-1)
1308 (*dd_idx
)++; /* Q D D D P */
1309 else if (*dd_idx
>= *pd_idx
)
1310 (*dd_idx
) += 2; /* D D P Q D */
1312 case ALGORITHM_LEFT_SYMMETRIC
:
1313 *pd_idx
= raid_disks
- 1 - (stripe
% raid_disks
);
1314 *dd_idx
= (*pd_idx
+ 2 + *dd_idx
) % raid_disks
;
1316 case ALGORITHM_RIGHT_SYMMETRIC
:
1317 *pd_idx
= stripe
% raid_disks
;
1318 *dd_idx
= (*pd_idx
+ 2 + *dd_idx
) % raid_disks
;
1321 printk(KERN_CRIT
"raid6: unsupported algorithm %d\n",
1328 * Finally, compute the new sector number
1330 new_sector
= (sector_t
)stripe
* sectors_per_chunk
+ chunk_offset
;
1335 static sector_t
compute_blocknr(struct stripe_head
*sh
, int i
)
1337 raid5_conf_t
*conf
= sh
->raid_conf
;
1338 int raid_disks
= sh
->disks
;
1339 int data_disks
= raid_disks
- conf
->max_degraded
;
1340 sector_t new_sector
= sh
->sector
, check
;
1341 int sectors_per_chunk
= conf
->chunk_size
>> 9;
1344 int chunk_number
, dummy1
, dummy2
, dd_idx
= i
;
1348 chunk_offset
= sector_div(new_sector
, sectors_per_chunk
);
1349 stripe
= new_sector
;
1350 BUG_ON(new_sector
!= stripe
);
1352 if (i
== sh
->pd_idx
)
1354 switch(conf
->level
) {
1357 switch (conf
->algorithm
) {
1358 case ALGORITHM_LEFT_ASYMMETRIC
:
1359 case ALGORITHM_RIGHT_ASYMMETRIC
:
1363 case ALGORITHM_LEFT_SYMMETRIC
:
1364 case ALGORITHM_RIGHT_SYMMETRIC
:
1367 i
-= (sh
->pd_idx
+ 1);
1370 printk(KERN_ERR
"raid5: unsupported algorithm %d\n",
1375 if (i
== raid6_next_disk(sh
->pd_idx
, raid_disks
))
1376 return 0; /* It is the Q disk */
1377 switch (conf
->algorithm
) {
1378 case ALGORITHM_LEFT_ASYMMETRIC
:
1379 case ALGORITHM_RIGHT_ASYMMETRIC
:
1380 if (sh
->pd_idx
== raid_disks
-1)
1381 i
--; /* Q D D D P */
1382 else if (i
> sh
->pd_idx
)
1383 i
-= 2; /* D D P Q D */
1385 case ALGORITHM_LEFT_SYMMETRIC
:
1386 case ALGORITHM_RIGHT_SYMMETRIC
:
1387 if (sh
->pd_idx
== raid_disks
-1)
1388 i
--; /* Q D D D P */
1393 i
-= (sh
->pd_idx
+ 2);
1397 printk(KERN_CRIT
"raid6: unsupported algorithm %d\n",
1403 chunk_number
= stripe
* data_disks
+ i
;
1404 r_sector
= (sector_t
)chunk_number
* sectors_per_chunk
+ chunk_offset
;
1406 check
= raid5_compute_sector(r_sector
, raid_disks
, data_disks
, &dummy1
, &dummy2
, conf
);
1407 if (check
!= sh
->sector
|| dummy1
!= dd_idx
|| dummy2
!= sh
->pd_idx
) {
1408 printk(KERN_ERR
"compute_blocknr: map not correct\n");
1417 * Copy data between a page in the stripe cache, and one or more bion
1418 * The page could align with the middle of the bio, or there could be
1419 * several bion, each with several bio_vecs, which cover part of the page
1420 * Multiple bion are linked together on bi_next. There may be extras
1421 * at the end of this list. We ignore them.
1423 static void copy_data(int frombio
, struct bio
*bio
,
1427 char *pa
= page_address(page
);
1428 struct bio_vec
*bvl
;
1432 if (bio
->bi_sector
>= sector
)
1433 page_offset
= (signed)(bio
->bi_sector
- sector
) * 512;
1435 page_offset
= (signed)(sector
- bio
->bi_sector
) * -512;
1436 bio_for_each_segment(bvl
, bio
, i
) {
1437 int len
= bio_iovec_idx(bio
,i
)->bv_len
;
1441 if (page_offset
< 0) {
1442 b_offset
= -page_offset
;
1443 page_offset
+= b_offset
;
1447 if (len
> 0 && page_offset
+ len
> STRIPE_SIZE
)
1448 clen
= STRIPE_SIZE
- page_offset
;
1452 char *ba
= __bio_kmap_atomic(bio
, i
, KM_USER0
);
1454 memcpy(pa
+page_offset
, ba
+b_offset
, clen
);
1456 memcpy(ba
+b_offset
, pa
+page_offset
, clen
);
1457 __bio_kunmap_atomic(ba
, KM_USER0
);
1459 if (clen
< len
) /* hit end of page */
1465 #define check_xor() do { \
1466 if (count == MAX_XOR_BLOCKS) { \
1467 xor_blocks(count, STRIPE_SIZE, dest, ptr);\
1472 static void compute_parity6(struct stripe_head
*sh
, int method
)
1474 raid5_conf_t
*conf
= sh
->raid_conf
;
1475 int i
, pd_idx
= sh
->pd_idx
, qd_idx
, d0_idx
, disks
= sh
->disks
, count
;
1477 /**** FIX THIS: This could be very bad if disks is close to 256 ****/
1480 qd_idx
= raid6_next_disk(pd_idx
, disks
);
1481 d0_idx
= raid6_next_disk(qd_idx
, disks
);
1483 pr_debug("compute_parity, stripe %llu, method %d\n",
1484 (unsigned long long)sh
->sector
, method
);
1487 case READ_MODIFY_WRITE
:
1488 BUG(); /* READ_MODIFY_WRITE N/A for RAID-6 */
1489 case RECONSTRUCT_WRITE
:
1490 for (i
= disks
; i
-- ;)
1491 if ( i
!= pd_idx
&& i
!= qd_idx
&& sh
->dev
[i
].towrite
) {
1492 chosen
= sh
->dev
[i
].towrite
;
1493 sh
->dev
[i
].towrite
= NULL
;
1495 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
1496 wake_up(&conf
->wait_for_overlap
);
1498 BUG_ON(sh
->dev
[i
].written
);
1499 sh
->dev
[i
].written
= chosen
;
1503 BUG(); /* Not implemented yet */
1506 for (i
= disks
; i
--;)
1507 if (sh
->dev
[i
].written
) {
1508 sector_t sector
= sh
->dev
[i
].sector
;
1509 struct bio
*wbi
= sh
->dev
[i
].written
;
1510 while (wbi
&& wbi
->bi_sector
< sector
+ STRIPE_SECTORS
) {
1511 copy_data(1, wbi
, sh
->dev
[i
].page
, sector
);
1512 wbi
= r5_next_bio(wbi
, sector
);
1515 set_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
1516 set_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
);
1520 // case RECONSTRUCT_WRITE:
1521 // case CHECK_PARITY:
1522 // case UPDATE_PARITY:
1523 /* Note that unlike RAID-5, the ordering of the disks matters greatly. */
1524 /* FIX: Is this ordering of drives even remotely optimal? */
1528 ptrs
[count
++] = page_address(sh
->dev
[i
].page
);
1529 if (count
<= disks
-2 && !test_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
))
1530 printk("block %d/%d not uptodate on parity calc\n", i
,count
);
1531 i
= raid6_next_disk(i
, disks
);
1532 } while ( i
!= d0_idx
);
1536 raid6_call
.gen_syndrome(disks
, STRIPE_SIZE
, ptrs
);
1539 case RECONSTRUCT_WRITE
:
1540 set_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
1541 set_bit(R5_UPTODATE
, &sh
->dev
[qd_idx
].flags
);
1542 set_bit(R5_LOCKED
, &sh
->dev
[pd_idx
].flags
);
1543 set_bit(R5_LOCKED
, &sh
->dev
[qd_idx
].flags
);
1546 set_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
1547 set_bit(R5_UPTODATE
, &sh
->dev
[qd_idx
].flags
);
1553 /* Compute one missing block */
1554 static void compute_block_1(struct stripe_head
*sh
, int dd_idx
, int nozero
)
1556 int i
, count
, disks
= sh
->disks
;
1557 void *ptr
[MAX_XOR_BLOCKS
], *dest
, *p
;
1558 int pd_idx
= sh
->pd_idx
;
1559 int qd_idx
= raid6_next_disk(pd_idx
, disks
);
1561 pr_debug("compute_block_1, stripe %llu, idx %d\n",
1562 (unsigned long long)sh
->sector
, dd_idx
);
1564 if ( dd_idx
== qd_idx
) {
1565 /* We're actually computing the Q drive */
1566 compute_parity6(sh
, UPDATE_PARITY
);
1568 dest
= page_address(sh
->dev
[dd_idx
].page
);
1569 if (!nozero
) memset(dest
, 0, STRIPE_SIZE
);
1571 for (i
= disks
; i
--; ) {
1572 if (i
== dd_idx
|| i
== qd_idx
)
1574 p
= page_address(sh
->dev
[i
].page
);
1575 if (test_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
))
1578 printk("compute_block() %d, stripe %llu, %d"
1579 " not present\n", dd_idx
,
1580 (unsigned long long)sh
->sector
, i
);
1585 xor_blocks(count
, STRIPE_SIZE
, dest
, ptr
);
1586 if (!nozero
) set_bit(R5_UPTODATE
, &sh
->dev
[dd_idx
].flags
);
1587 else clear_bit(R5_UPTODATE
, &sh
->dev
[dd_idx
].flags
);
1591 /* Compute two missing blocks */
1592 static void compute_block_2(struct stripe_head
*sh
, int dd_idx1
, int dd_idx2
)
1594 int i
, count
, disks
= sh
->disks
;
1595 int pd_idx
= sh
->pd_idx
;
1596 int qd_idx
= raid6_next_disk(pd_idx
, disks
);
1597 int d0_idx
= raid6_next_disk(qd_idx
, disks
);
1600 /* faila and failb are disk numbers relative to d0_idx */
1601 /* pd_idx become disks-2 and qd_idx become disks-1 */
1602 faila
= (dd_idx1
< d0_idx
) ? dd_idx1
+(disks
-d0_idx
) : dd_idx1
-d0_idx
;
1603 failb
= (dd_idx2
< d0_idx
) ? dd_idx2
+(disks
-d0_idx
) : dd_idx2
-d0_idx
;
1605 BUG_ON(faila
== failb
);
1606 if ( failb
< faila
) { int tmp
= faila
; faila
= failb
; failb
= tmp
; }
1608 pr_debug("compute_block_2, stripe %llu, idx %d,%d (%d,%d)\n",
1609 (unsigned long long)sh
->sector
, dd_idx1
, dd_idx2
, faila
, failb
);
1611 if ( failb
== disks
-1 ) {
1612 /* Q disk is one of the missing disks */
1613 if ( faila
== disks
-2 ) {
1614 /* Missing P+Q, just recompute */
1615 compute_parity6(sh
, UPDATE_PARITY
);
1618 /* We're missing D+Q; recompute D from P */
1619 compute_block_1(sh
, (dd_idx1
== qd_idx
) ? dd_idx2
: dd_idx1
, 0);
1620 compute_parity6(sh
, UPDATE_PARITY
); /* Is this necessary? */
1625 /* We're missing D+P or D+D; build pointer table */
1627 /**** FIX THIS: This could be very bad if disks is close to 256 ****/
1633 ptrs
[count
++] = page_address(sh
->dev
[i
].page
);
1634 i
= raid6_next_disk(i
, disks
);
1635 if (i
!= dd_idx1
&& i
!= dd_idx2
&&
1636 !test_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
))
1637 printk("compute_2 with missing block %d/%d\n", count
, i
);
1638 } while ( i
!= d0_idx
);
1640 if ( failb
== disks
-2 ) {
1641 /* We're missing D+P. */
1642 raid6_datap_recov(disks
, STRIPE_SIZE
, faila
, ptrs
);
1644 /* We're missing D+D. */
1645 raid6_2data_recov(disks
, STRIPE_SIZE
, faila
, failb
, ptrs
);
1648 /* Both the above update both missing blocks */
1649 set_bit(R5_UPTODATE
, &sh
->dev
[dd_idx1
].flags
);
1650 set_bit(R5_UPTODATE
, &sh
->dev
[dd_idx2
].flags
);
1655 schedule_reconstruction5(struct stripe_head
*sh
, struct stripe_head_state
*s
,
1656 int rcw
, int expand
)
1658 int i
, pd_idx
= sh
->pd_idx
, disks
= sh
->disks
;
1661 /* if we are not expanding this is a proper write request, and
1662 * there will be bios with new data to be drained into the
1666 sh
->reconstruct_state
= reconstruct_state_drain_run
;
1667 set_bit(STRIPE_OP_BIODRAIN
, &s
->ops_request
);
1669 sh
->reconstruct_state
= reconstruct_state_run
;
1671 set_bit(STRIPE_OP_POSTXOR
, &s
->ops_request
);
1673 for (i
= disks
; i
--; ) {
1674 struct r5dev
*dev
= &sh
->dev
[i
];
1677 set_bit(R5_LOCKED
, &dev
->flags
);
1678 set_bit(R5_Wantdrain
, &dev
->flags
);
1680 clear_bit(R5_UPTODATE
, &dev
->flags
);
1684 if (s
->locked
+ 1 == disks
)
1685 if (!test_and_set_bit(STRIPE_FULL_WRITE
, &sh
->state
))
1686 atomic_inc(&sh
->raid_conf
->pending_full_writes
);
1688 BUG_ON(!(test_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
) ||
1689 test_bit(R5_Wantcompute
, &sh
->dev
[pd_idx
].flags
)));
1691 sh
->reconstruct_state
= reconstruct_state_prexor_drain_run
;
1692 set_bit(STRIPE_OP_PREXOR
, &s
->ops_request
);
1693 set_bit(STRIPE_OP_BIODRAIN
, &s
->ops_request
);
1694 set_bit(STRIPE_OP_POSTXOR
, &s
->ops_request
);
1696 for (i
= disks
; i
--; ) {
1697 struct r5dev
*dev
= &sh
->dev
[i
];
1702 (test_bit(R5_UPTODATE
, &dev
->flags
) ||
1703 test_bit(R5_Wantcompute
, &dev
->flags
))) {
1704 set_bit(R5_Wantdrain
, &dev
->flags
);
1705 set_bit(R5_LOCKED
, &dev
->flags
);
1706 clear_bit(R5_UPTODATE
, &dev
->flags
);
1712 /* keep the parity disk locked while asynchronous operations
1715 set_bit(R5_LOCKED
, &sh
->dev
[pd_idx
].flags
);
1716 clear_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
1719 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
1720 __func__
, (unsigned long long)sh
->sector
,
1721 s
->locked
, s
->ops_request
);
1725 * Each stripe/dev can have one or more bion attached.
1726 * toread/towrite point to the first in a chain.
1727 * The bi_next chain must be in order.
1729 static int add_stripe_bio(struct stripe_head
*sh
, struct bio
*bi
, int dd_idx
, int forwrite
)
1732 raid5_conf_t
*conf
= sh
->raid_conf
;
1735 pr_debug("adding bh b#%llu to stripe s#%llu\n",
1736 (unsigned long long)bi
->bi_sector
,
1737 (unsigned long long)sh
->sector
);
1740 spin_lock(&sh
->lock
);
1741 spin_lock_irq(&conf
->device_lock
);
1743 bip
= &sh
->dev
[dd_idx
].towrite
;
1744 if (*bip
== NULL
&& sh
->dev
[dd_idx
].written
== NULL
)
1747 bip
= &sh
->dev
[dd_idx
].toread
;
1748 while (*bip
&& (*bip
)->bi_sector
< bi
->bi_sector
) {
1749 if ((*bip
)->bi_sector
+ ((*bip
)->bi_size
>> 9) > bi
->bi_sector
)
1751 bip
= & (*bip
)->bi_next
;
1753 if (*bip
&& (*bip
)->bi_sector
< bi
->bi_sector
+ ((bi
->bi_size
)>>9))
1756 BUG_ON(*bip
&& bi
->bi_next
&& (*bip
) != bi
->bi_next
);
1760 bi
->bi_phys_segments
++;
1761 spin_unlock_irq(&conf
->device_lock
);
1762 spin_unlock(&sh
->lock
);
1764 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
1765 (unsigned long long)bi
->bi_sector
,
1766 (unsigned long long)sh
->sector
, dd_idx
);
1768 if (conf
->mddev
->bitmap
&& firstwrite
) {
1769 bitmap_startwrite(conf
->mddev
->bitmap
, sh
->sector
,
1771 sh
->bm_seq
= conf
->seq_flush
+1;
1772 set_bit(STRIPE_BIT_DELAY
, &sh
->state
);
1776 /* check if page is covered */
1777 sector_t sector
= sh
->dev
[dd_idx
].sector
;
1778 for (bi
=sh
->dev
[dd_idx
].towrite
;
1779 sector
< sh
->dev
[dd_idx
].sector
+ STRIPE_SECTORS
&&
1780 bi
&& bi
->bi_sector
<= sector
;
1781 bi
= r5_next_bio(bi
, sh
->dev
[dd_idx
].sector
)) {
1782 if (bi
->bi_sector
+ (bi
->bi_size
>>9) >= sector
)
1783 sector
= bi
->bi_sector
+ (bi
->bi_size
>>9);
1785 if (sector
>= sh
->dev
[dd_idx
].sector
+ STRIPE_SECTORS
)
1786 set_bit(R5_OVERWRITE
, &sh
->dev
[dd_idx
].flags
);
1791 set_bit(R5_Overlap
, &sh
->dev
[dd_idx
].flags
);
1792 spin_unlock_irq(&conf
->device_lock
);
1793 spin_unlock(&sh
->lock
);
1797 static void end_reshape(raid5_conf_t
*conf
);
1799 static int page_is_zero(struct page
*p
)
1801 char *a
= page_address(p
);
1802 return ((*(u32
*)a
) == 0 &&
1803 memcmp(a
, a
+4, STRIPE_SIZE
-4)==0);
1806 static int stripe_to_pdidx(sector_t stripe
, raid5_conf_t
*conf
, int disks
)
1808 int sectors_per_chunk
= conf
->chunk_size
>> 9;
1810 int chunk_offset
= sector_div(stripe
, sectors_per_chunk
);
1812 raid5_compute_sector(stripe
* (disks
- conf
->max_degraded
)
1813 *sectors_per_chunk
+ chunk_offset
,
1814 disks
, disks
- conf
->max_degraded
,
1815 &dd_idx
, &pd_idx
, conf
);
1820 handle_failed_stripe(raid5_conf_t
*conf
, struct stripe_head
*sh
,
1821 struct stripe_head_state
*s
, int disks
,
1822 struct bio
**return_bi
)
1825 for (i
= disks
; i
--; ) {
1829 if (test_bit(R5_ReadError
, &sh
->dev
[i
].flags
)) {
1832 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
1833 if (rdev
&& test_bit(In_sync
, &rdev
->flags
))
1834 /* multiple read failures in one stripe */
1835 md_error(conf
->mddev
, rdev
);
1838 spin_lock_irq(&conf
->device_lock
);
1839 /* fail all writes first */
1840 bi
= sh
->dev
[i
].towrite
;
1841 sh
->dev
[i
].towrite
= NULL
;
1847 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
1848 wake_up(&conf
->wait_for_overlap
);
1850 while (bi
&& bi
->bi_sector
<
1851 sh
->dev
[i
].sector
+ STRIPE_SECTORS
) {
1852 struct bio
*nextbi
= r5_next_bio(bi
, sh
->dev
[i
].sector
);
1853 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1854 if (!raid5_dec_bi_phys_segments(bi
)) {
1855 md_write_end(conf
->mddev
);
1856 bi
->bi_next
= *return_bi
;
1861 /* and fail all 'written' */
1862 bi
= sh
->dev
[i
].written
;
1863 sh
->dev
[i
].written
= NULL
;
1864 if (bi
) bitmap_end
= 1;
1865 while (bi
&& bi
->bi_sector
<
1866 sh
->dev
[i
].sector
+ STRIPE_SECTORS
) {
1867 struct bio
*bi2
= r5_next_bio(bi
, sh
->dev
[i
].sector
);
1868 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1869 if (!raid5_dec_bi_phys_segments(bi
)) {
1870 md_write_end(conf
->mddev
);
1871 bi
->bi_next
= *return_bi
;
1877 /* fail any reads if this device is non-operational and
1878 * the data has not reached the cache yet.
1880 if (!test_bit(R5_Wantfill
, &sh
->dev
[i
].flags
) &&
1881 (!test_bit(R5_Insync
, &sh
->dev
[i
].flags
) ||
1882 test_bit(R5_ReadError
, &sh
->dev
[i
].flags
))) {
1883 bi
= sh
->dev
[i
].toread
;
1884 sh
->dev
[i
].toread
= NULL
;
1885 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
1886 wake_up(&conf
->wait_for_overlap
);
1887 if (bi
) s
->to_read
--;
1888 while (bi
&& bi
->bi_sector
<
1889 sh
->dev
[i
].sector
+ STRIPE_SECTORS
) {
1890 struct bio
*nextbi
=
1891 r5_next_bio(bi
, sh
->dev
[i
].sector
);
1892 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1893 if (!raid5_dec_bi_phys_segments(bi
)) {
1894 bi
->bi_next
= *return_bi
;
1900 spin_unlock_irq(&conf
->device_lock
);
1902 bitmap_endwrite(conf
->mddev
->bitmap
, sh
->sector
,
1903 STRIPE_SECTORS
, 0, 0);
1906 if (test_and_clear_bit(STRIPE_FULL_WRITE
, &sh
->state
))
1907 if (atomic_dec_and_test(&conf
->pending_full_writes
))
1908 md_wakeup_thread(conf
->mddev
->thread
);
1911 /* fetch_block5 - checks the given member device to see if its data needs
1912 * to be read or computed to satisfy a request.
1914 * Returns 1 when no more member devices need to be checked, otherwise returns
1915 * 0 to tell the loop in handle_stripe_fill5 to continue
1917 static int fetch_block5(struct stripe_head
*sh
, struct stripe_head_state
*s
,
1918 int disk_idx
, int disks
)
1920 struct r5dev
*dev
= &sh
->dev
[disk_idx
];
1921 struct r5dev
*failed_dev
= &sh
->dev
[s
->failed_num
];
1923 /* is the data in this block needed, and can we get it? */
1924 if (!test_bit(R5_LOCKED
, &dev
->flags
) &&
1925 !test_bit(R5_UPTODATE
, &dev
->flags
) &&
1927 (dev
->towrite
&& !test_bit(R5_OVERWRITE
, &dev
->flags
)) ||
1928 s
->syncing
|| s
->expanding
||
1930 (failed_dev
->toread
||
1931 (failed_dev
->towrite
&&
1932 !test_bit(R5_OVERWRITE
, &failed_dev
->flags
)))))) {
1933 /* We would like to get this block, possibly by computing it,
1934 * otherwise read it if the backing disk is insync
1936 if ((s
->uptodate
== disks
- 1) &&
1937 (s
->failed
&& disk_idx
== s
->failed_num
)) {
1938 set_bit(STRIPE_COMPUTE_RUN
, &sh
->state
);
1939 set_bit(STRIPE_OP_COMPUTE_BLK
, &s
->ops_request
);
1940 set_bit(R5_Wantcompute
, &dev
->flags
);
1941 sh
->ops
.target
= disk_idx
;
1943 /* Careful: from this point on 'uptodate' is in the eye
1944 * of raid5_run_ops which services 'compute' operations
1945 * before writes. R5_Wantcompute flags a block that will
1946 * be R5_UPTODATE by the time it is needed for a
1947 * subsequent operation.
1950 return 1; /* uptodate + compute == disks */
1951 } else if (test_bit(R5_Insync
, &dev
->flags
)) {
1952 set_bit(R5_LOCKED
, &dev
->flags
);
1953 set_bit(R5_Wantread
, &dev
->flags
);
1955 pr_debug("Reading block %d (sync=%d)\n", disk_idx
,
1964 * handle_stripe_fill5 - read or compute data to satisfy pending requests.
1966 static void handle_stripe_fill5(struct stripe_head
*sh
,
1967 struct stripe_head_state
*s
, int disks
)
1971 /* look for blocks to read/compute, skip this if a compute
1972 * is already in flight, or if the stripe contents are in the
1973 * midst of changing due to a write
1975 if (!test_bit(STRIPE_COMPUTE_RUN
, &sh
->state
) && !sh
->check_state
&&
1976 !sh
->reconstruct_state
)
1977 for (i
= disks
; i
--; )
1978 if (fetch_block5(sh
, s
, i
, disks
))
1980 set_bit(STRIPE_HANDLE
, &sh
->state
);
1983 static void handle_stripe_fill6(struct stripe_head
*sh
,
1984 struct stripe_head_state
*s
, struct r6_state
*r6s
,
1988 for (i
= disks
; i
--; ) {
1989 struct r5dev
*dev
= &sh
->dev
[i
];
1990 if (!test_bit(R5_LOCKED
, &dev
->flags
) &&
1991 !test_bit(R5_UPTODATE
, &dev
->flags
) &&
1992 (dev
->toread
|| (dev
->towrite
&&
1993 !test_bit(R5_OVERWRITE
, &dev
->flags
)) ||
1994 s
->syncing
|| s
->expanding
||
1996 (sh
->dev
[r6s
->failed_num
[0]].toread
||
1999 (sh
->dev
[r6s
->failed_num
[1]].toread
||
2001 /* we would like to get this block, possibly
2002 * by computing it, but we might not be able to
2004 if ((s
->uptodate
== disks
- 1) &&
2005 (s
->failed
&& (i
== r6s
->failed_num
[0] ||
2006 i
== r6s
->failed_num
[1]))) {
2007 pr_debug("Computing stripe %llu block %d\n",
2008 (unsigned long long)sh
->sector
, i
);
2009 compute_block_1(sh
, i
, 0);
2011 } else if ( s
->uptodate
== disks
-2 && s
->failed
>= 2 ) {
2012 /* Computing 2-failure is *very* expensive; only
2013 * do it if failed >= 2
2016 for (other
= disks
; other
--; ) {
2019 if (!test_bit(R5_UPTODATE
,
2020 &sh
->dev
[other
].flags
))
2024 pr_debug("Computing stripe %llu blocks %d,%d\n",
2025 (unsigned long long)sh
->sector
,
2027 compute_block_2(sh
, i
, other
);
2029 } else if (test_bit(R5_Insync
, &dev
->flags
)) {
2030 set_bit(R5_LOCKED
, &dev
->flags
);
2031 set_bit(R5_Wantread
, &dev
->flags
);
2033 pr_debug("Reading block %d (sync=%d)\n",
2038 set_bit(STRIPE_HANDLE
, &sh
->state
);
2042 /* handle_stripe_clean_event
2043 * any written block on an uptodate or failed drive can be returned.
2044 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
2045 * never LOCKED, so we don't need to test 'failed' directly.
2047 static void handle_stripe_clean_event(raid5_conf_t
*conf
,
2048 struct stripe_head
*sh
, int disks
, struct bio
**return_bi
)
2053 for (i
= disks
; i
--; )
2054 if (sh
->dev
[i
].written
) {
2056 if (!test_bit(R5_LOCKED
, &dev
->flags
) &&
2057 test_bit(R5_UPTODATE
, &dev
->flags
)) {
2058 /* We can return any write requests */
2059 struct bio
*wbi
, *wbi2
;
2061 pr_debug("Return write for disc %d\n", i
);
2062 spin_lock_irq(&conf
->device_lock
);
2064 dev
->written
= NULL
;
2065 while (wbi
&& wbi
->bi_sector
<
2066 dev
->sector
+ STRIPE_SECTORS
) {
2067 wbi2
= r5_next_bio(wbi
, dev
->sector
);
2068 if (!raid5_dec_bi_phys_segments(wbi
)) {
2069 md_write_end(conf
->mddev
);
2070 wbi
->bi_next
= *return_bi
;
2075 if (dev
->towrite
== NULL
)
2077 spin_unlock_irq(&conf
->device_lock
);
2079 bitmap_endwrite(conf
->mddev
->bitmap
,
2082 !test_bit(STRIPE_DEGRADED
, &sh
->state
),
2087 if (test_and_clear_bit(STRIPE_FULL_WRITE
, &sh
->state
))
2088 if (atomic_dec_and_test(&conf
->pending_full_writes
))
2089 md_wakeup_thread(conf
->mddev
->thread
);
2092 static void handle_stripe_dirtying5(raid5_conf_t
*conf
,
2093 struct stripe_head
*sh
, struct stripe_head_state
*s
, int disks
)
2095 int rmw
= 0, rcw
= 0, i
;
2096 for (i
= disks
; i
--; ) {
2097 /* would I have to read this buffer for read_modify_write */
2098 struct r5dev
*dev
= &sh
->dev
[i
];
2099 if ((dev
->towrite
|| i
== sh
->pd_idx
) &&
2100 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2101 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2102 test_bit(R5_Wantcompute
, &dev
->flags
))) {
2103 if (test_bit(R5_Insync
, &dev
->flags
))
2106 rmw
+= 2*disks
; /* cannot read it */
2108 /* Would I have to read this buffer for reconstruct_write */
2109 if (!test_bit(R5_OVERWRITE
, &dev
->flags
) && i
!= sh
->pd_idx
&&
2110 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2111 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2112 test_bit(R5_Wantcompute
, &dev
->flags
))) {
2113 if (test_bit(R5_Insync
, &dev
->flags
)) rcw
++;
2118 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
2119 (unsigned long long)sh
->sector
, rmw
, rcw
);
2120 set_bit(STRIPE_HANDLE
, &sh
->state
);
2121 if (rmw
< rcw
&& rmw
> 0)
2122 /* prefer read-modify-write, but need to get some data */
2123 for (i
= disks
; i
--; ) {
2124 struct r5dev
*dev
= &sh
->dev
[i
];
2125 if ((dev
->towrite
|| i
== sh
->pd_idx
) &&
2126 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2127 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2128 test_bit(R5_Wantcompute
, &dev
->flags
)) &&
2129 test_bit(R5_Insync
, &dev
->flags
)) {
2131 test_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2132 pr_debug("Read_old block "
2133 "%d for r-m-w\n", i
);
2134 set_bit(R5_LOCKED
, &dev
->flags
);
2135 set_bit(R5_Wantread
, &dev
->flags
);
2138 set_bit(STRIPE_DELAYED
, &sh
->state
);
2139 set_bit(STRIPE_HANDLE
, &sh
->state
);
2143 if (rcw
<= rmw
&& rcw
> 0)
2144 /* want reconstruct write, but need to get some data */
2145 for (i
= disks
; i
--; ) {
2146 struct r5dev
*dev
= &sh
->dev
[i
];
2147 if (!test_bit(R5_OVERWRITE
, &dev
->flags
) &&
2149 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2150 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2151 test_bit(R5_Wantcompute
, &dev
->flags
)) &&
2152 test_bit(R5_Insync
, &dev
->flags
)) {
2154 test_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2155 pr_debug("Read_old block "
2156 "%d for Reconstruct\n", i
);
2157 set_bit(R5_LOCKED
, &dev
->flags
);
2158 set_bit(R5_Wantread
, &dev
->flags
);
2161 set_bit(STRIPE_DELAYED
, &sh
->state
);
2162 set_bit(STRIPE_HANDLE
, &sh
->state
);
2166 /* now if nothing is locked, and if we have enough data,
2167 * we can start a write request
2169 /* since handle_stripe can be called at any time we need to handle the
2170 * case where a compute block operation has been submitted and then a
2171 * subsequent call wants to start a write request. raid5_run_ops only
2172 * handles the case where compute block and postxor are requested
2173 * simultaneously. If this is not the case then new writes need to be
2174 * held off until the compute completes.
2176 if ((s
->req_compute
|| !test_bit(STRIPE_COMPUTE_RUN
, &sh
->state
)) &&
2177 (s
->locked
== 0 && (rcw
== 0 || rmw
== 0) &&
2178 !test_bit(STRIPE_BIT_DELAY
, &sh
->state
)))
2179 schedule_reconstruction5(sh
, s
, rcw
== 0, 0);
2182 static void handle_stripe_dirtying6(raid5_conf_t
*conf
,
2183 struct stripe_head
*sh
, struct stripe_head_state
*s
,
2184 struct r6_state
*r6s
, int disks
)
2186 int rcw
= 0, must_compute
= 0, pd_idx
= sh
->pd_idx
, i
;
2187 int qd_idx
= r6s
->qd_idx
;
2188 for (i
= disks
; i
--; ) {
2189 struct r5dev
*dev
= &sh
->dev
[i
];
2190 /* Would I have to read this buffer for reconstruct_write */
2191 if (!test_bit(R5_OVERWRITE
, &dev
->flags
)
2192 && i
!= pd_idx
&& i
!= qd_idx
2193 && (!test_bit(R5_LOCKED
, &dev
->flags
)
2195 !test_bit(R5_UPTODATE
, &dev
->flags
)) {
2196 if (test_bit(R5_Insync
, &dev
->flags
)) rcw
++;
2198 pr_debug("raid6: must_compute: "
2199 "disk %d flags=%#lx\n", i
, dev
->flags
);
2204 pr_debug("for sector %llu, rcw=%d, must_compute=%d\n",
2205 (unsigned long long)sh
->sector
, rcw
, must_compute
);
2206 set_bit(STRIPE_HANDLE
, &sh
->state
);
2209 /* want reconstruct write, but need to get some data */
2210 for (i
= disks
; i
--; ) {
2211 struct r5dev
*dev
= &sh
->dev
[i
];
2212 if (!test_bit(R5_OVERWRITE
, &dev
->flags
)
2213 && !(s
->failed
== 0 && (i
== pd_idx
|| i
== qd_idx
))
2214 && !test_bit(R5_LOCKED
, &dev
->flags
) &&
2215 !test_bit(R5_UPTODATE
, &dev
->flags
) &&
2216 test_bit(R5_Insync
, &dev
->flags
)) {
2218 test_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2219 pr_debug("Read_old stripe %llu "
2220 "block %d for Reconstruct\n",
2221 (unsigned long long)sh
->sector
, i
);
2222 set_bit(R5_LOCKED
, &dev
->flags
);
2223 set_bit(R5_Wantread
, &dev
->flags
);
2226 pr_debug("Request delayed stripe %llu "
2227 "block %d for Reconstruct\n",
2228 (unsigned long long)sh
->sector
, i
);
2229 set_bit(STRIPE_DELAYED
, &sh
->state
);
2230 set_bit(STRIPE_HANDLE
, &sh
->state
);
2234 /* now if nothing is locked, and if we have enough data, we can start a
2237 if (s
->locked
== 0 && rcw
== 0 &&
2238 !test_bit(STRIPE_BIT_DELAY
, &sh
->state
)) {
2239 if (must_compute
> 0) {
2240 /* We have failed blocks and need to compute them */
2241 switch (s
->failed
) {
2245 compute_block_1(sh
, r6s
->failed_num
[0], 0);
2248 compute_block_2(sh
, r6s
->failed_num
[0],
2249 r6s
->failed_num
[1]);
2251 default: /* This request should have been failed? */
2256 pr_debug("Computing parity for stripe %llu\n",
2257 (unsigned long long)sh
->sector
);
2258 compute_parity6(sh
, RECONSTRUCT_WRITE
);
2259 /* now every locked buffer is ready to be written */
2260 for (i
= disks
; i
--; )
2261 if (test_bit(R5_LOCKED
, &sh
->dev
[i
].flags
)) {
2262 pr_debug("Writing stripe %llu block %d\n",
2263 (unsigned long long)sh
->sector
, i
);
2265 set_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
);
2267 if (s
->locked
== disks
)
2268 if (!test_and_set_bit(STRIPE_FULL_WRITE
, &sh
->state
))
2269 atomic_inc(&conf
->pending_full_writes
);
2270 /* after a RECONSTRUCT_WRITE, the stripe MUST be in-sync */
2271 set_bit(STRIPE_INSYNC
, &sh
->state
);
2273 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2274 atomic_dec(&conf
->preread_active_stripes
);
2275 if (atomic_read(&conf
->preread_active_stripes
) <
2277 md_wakeup_thread(conf
->mddev
->thread
);
2282 static void handle_parity_checks5(raid5_conf_t
*conf
, struct stripe_head
*sh
,
2283 struct stripe_head_state
*s
, int disks
)
2285 struct r5dev
*dev
= NULL
;
2287 set_bit(STRIPE_HANDLE
, &sh
->state
);
2289 switch (sh
->check_state
) {
2290 case check_state_idle
:
2291 /* start a new check operation if there are no failures */
2292 if (s
->failed
== 0) {
2293 BUG_ON(s
->uptodate
!= disks
);
2294 sh
->check_state
= check_state_run
;
2295 set_bit(STRIPE_OP_CHECK
, &s
->ops_request
);
2296 clear_bit(R5_UPTODATE
, &sh
->dev
[sh
->pd_idx
].flags
);
2300 dev
= &sh
->dev
[s
->failed_num
];
2302 case check_state_compute_result
:
2303 sh
->check_state
= check_state_idle
;
2305 dev
= &sh
->dev
[sh
->pd_idx
];
2307 /* check that a write has not made the stripe insync */
2308 if (test_bit(STRIPE_INSYNC
, &sh
->state
))
2311 /* either failed parity check, or recovery is happening */
2312 BUG_ON(!test_bit(R5_UPTODATE
, &dev
->flags
));
2313 BUG_ON(s
->uptodate
!= disks
);
2315 set_bit(R5_LOCKED
, &dev
->flags
);
2317 set_bit(R5_Wantwrite
, &dev
->flags
);
2319 clear_bit(STRIPE_DEGRADED
, &sh
->state
);
2320 set_bit(STRIPE_INSYNC
, &sh
->state
);
2322 case check_state_run
:
2323 break; /* we will be called again upon completion */
2324 case check_state_check_result
:
2325 sh
->check_state
= check_state_idle
;
2327 /* if a failure occurred during the check operation, leave
2328 * STRIPE_INSYNC not set and let the stripe be handled again
2333 /* handle a successful check operation, if parity is correct
2334 * we are done. Otherwise update the mismatch count and repair
2335 * parity if !MD_RECOVERY_CHECK
2337 if (sh
->ops
.zero_sum_result
== 0)
2338 /* parity is correct (on disc,
2339 * not in buffer any more)
2341 set_bit(STRIPE_INSYNC
, &sh
->state
);
2343 conf
->mddev
->resync_mismatches
+= STRIPE_SECTORS
;
2344 if (test_bit(MD_RECOVERY_CHECK
, &conf
->mddev
->recovery
))
2345 /* don't try to repair!! */
2346 set_bit(STRIPE_INSYNC
, &sh
->state
);
2348 sh
->check_state
= check_state_compute_run
;
2349 set_bit(STRIPE_COMPUTE_RUN
, &sh
->state
);
2350 set_bit(STRIPE_OP_COMPUTE_BLK
, &s
->ops_request
);
2351 set_bit(R5_Wantcompute
,
2352 &sh
->dev
[sh
->pd_idx
].flags
);
2353 sh
->ops
.target
= sh
->pd_idx
;
2358 case check_state_compute_run
:
2361 printk(KERN_ERR
"%s: unknown check_state: %d sector: %llu\n",
2362 __func__
, sh
->check_state
,
2363 (unsigned long long) sh
->sector
);
2369 static void handle_parity_checks6(raid5_conf_t
*conf
, struct stripe_head
*sh
,
2370 struct stripe_head_state
*s
,
2371 struct r6_state
*r6s
, struct page
*tmp_page
,
2374 int update_p
= 0, update_q
= 0;
2376 int pd_idx
= sh
->pd_idx
;
2377 int qd_idx
= r6s
->qd_idx
;
2379 set_bit(STRIPE_HANDLE
, &sh
->state
);
2381 BUG_ON(s
->failed
> 2);
2382 BUG_ON(s
->uptodate
< disks
);
2383 /* Want to check and possibly repair P and Q.
2384 * However there could be one 'failed' device, in which
2385 * case we can only check one of them, possibly using the
2386 * other to generate missing data
2389 /* If !tmp_page, we cannot do the calculations,
2390 * but as we have set STRIPE_HANDLE, we will soon be called
2391 * by stripe_handle with a tmp_page - just wait until then.
2394 if (s
->failed
== r6s
->q_failed
) {
2395 /* The only possible failed device holds 'Q', so it
2396 * makes sense to check P (If anything else were failed,
2397 * we would have used P to recreate it).
2399 compute_block_1(sh
, pd_idx
, 1);
2400 if (!page_is_zero(sh
->dev
[pd_idx
].page
)) {
2401 compute_block_1(sh
, pd_idx
, 0);
2405 if (!r6s
->q_failed
&& s
->failed
< 2) {
2406 /* q is not failed, and we didn't use it to generate
2407 * anything, so it makes sense to check it
2409 memcpy(page_address(tmp_page
),
2410 page_address(sh
->dev
[qd_idx
].page
),
2412 compute_parity6(sh
, UPDATE_PARITY
);
2413 if (memcmp(page_address(tmp_page
),
2414 page_address(sh
->dev
[qd_idx
].page
),
2415 STRIPE_SIZE
) != 0) {
2416 clear_bit(STRIPE_INSYNC
, &sh
->state
);
2420 if (update_p
|| update_q
) {
2421 conf
->mddev
->resync_mismatches
+= STRIPE_SECTORS
;
2422 if (test_bit(MD_RECOVERY_CHECK
, &conf
->mddev
->recovery
))
2423 /* don't try to repair!! */
2424 update_p
= update_q
= 0;
2427 /* now write out any block on a failed drive,
2428 * or P or Q if they need it
2431 if (s
->failed
== 2) {
2432 dev
= &sh
->dev
[r6s
->failed_num
[1]];
2434 set_bit(R5_LOCKED
, &dev
->flags
);
2435 set_bit(R5_Wantwrite
, &dev
->flags
);
2437 if (s
->failed
>= 1) {
2438 dev
= &sh
->dev
[r6s
->failed_num
[0]];
2440 set_bit(R5_LOCKED
, &dev
->flags
);
2441 set_bit(R5_Wantwrite
, &dev
->flags
);
2445 dev
= &sh
->dev
[pd_idx
];
2447 set_bit(R5_LOCKED
, &dev
->flags
);
2448 set_bit(R5_Wantwrite
, &dev
->flags
);
2451 dev
= &sh
->dev
[qd_idx
];
2453 set_bit(R5_LOCKED
, &dev
->flags
);
2454 set_bit(R5_Wantwrite
, &dev
->flags
);
2456 clear_bit(STRIPE_DEGRADED
, &sh
->state
);
2458 set_bit(STRIPE_INSYNC
, &sh
->state
);
2462 static void handle_stripe_expansion(raid5_conf_t
*conf
, struct stripe_head
*sh
,
2463 struct r6_state
*r6s
)
2467 /* We have read all the blocks in this stripe and now we need to
2468 * copy some of them into a target stripe for expand.
2470 struct dma_async_tx_descriptor
*tx
= NULL
;
2471 clear_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
2472 for (i
= 0; i
< sh
->disks
; i
++)
2473 if (i
!= sh
->pd_idx
&& (!r6s
|| i
!= r6s
->qd_idx
)) {
2474 int dd_idx
, pd_idx
, j
;
2475 struct stripe_head
*sh2
;
2477 sector_t bn
= compute_blocknr(sh
, i
);
2478 sector_t s
= raid5_compute_sector(bn
, conf
->raid_disks
,
2480 conf
->max_degraded
, &dd_idx
,
2482 sh2
= get_active_stripe(conf
, s
, conf
->raid_disks
,
2485 /* so far only the early blocks of this stripe
2486 * have been requested. When later blocks
2487 * get requested, we will try again
2490 if (!test_bit(STRIPE_EXPANDING
, &sh2
->state
) ||
2491 test_bit(R5_Expanded
, &sh2
->dev
[dd_idx
].flags
)) {
2492 /* must have already done this block */
2493 release_stripe(sh2
);
2497 /* place all the copies on one channel */
2498 tx
= async_memcpy(sh2
->dev
[dd_idx
].page
,
2499 sh
->dev
[i
].page
, 0, 0, STRIPE_SIZE
,
2500 ASYNC_TX_DEP_ACK
, tx
, NULL
, NULL
);
2502 set_bit(R5_Expanded
, &sh2
->dev
[dd_idx
].flags
);
2503 set_bit(R5_UPTODATE
, &sh2
->dev
[dd_idx
].flags
);
2504 for (j
= 0; j
< conf
->raid_disks
; j
++)
2505 if (j
!= sh2
->pd_idx
&&
2506 (!r6s
|| j
!= raid6_next_disk(sh2
->pd_idx
,
2508 !test_bit(R5_Expanded
, &sh2
->dev
[j
].flags
))
2510 if (j
== conf
->raid_disks
) {
2511 set_bit(STRIPE_EXPAND_READY
, &sh2
->state
);
2512 set_bit(STRIPE_HANDLE
, &sh2
->state
);
2514 release_stripe(sh2
);
2517 /* done submitting copies, wait for them to complete */
2520 dma_wait_for_async_tx(tx
);
2526 * handle_stripe - do things to a stripe.
2528 * We lock the stripe and then examine the state of various bits
2529 * to see what needs to be done.
2531 * return some read request which now have data
2532 * return some write requests which are safely on disc
2533 * schedule a read on some buffers
2534 * schedule a write of some buffers
2535 * return confirmation of parity correctness
2537 * buffers are taken off read_list or write_list, and bh_cache buffers
2538 * get BH_Lock set before the stripe lock is released.
2542 static bool handle_stripe5(struct stripe_head
*sh
)
2544 raid5_conf_t
*conf
= sh
->raid_conf
;
2545 int disks
= sh
->disks
, i
;
2546 struct bio
*return_bi
= NULL
;
2547 struct stripe_head_state s
;
2549 mdk_rdev_t
*blocked_rdev
= NULL
;
2552 memset(&s
, 0, sizeof(s
));
2553 pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d check:%d "
2554 "reconstruct:%d\n", (unsigned long long)sh
->sector
, sh
->state
,
2555 atomic_read(&sh
->count
), sh
->pd_idx
, sh
->check_state
,
2556 sh
->reconstruct_state
);
2558 spin_lock(&sh
->lock
);
2559 clear_bit(STRIPE_HANDLE
, &sh
->state
);
2560 clear_bit(STRIPE_DELAYED
, &sh
->state
);
2562 s
.syncing
= test_bit(STRIPE_SYNCING
, &sh
->state
);
2563 s
.expanding
= test_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
2564 s
.expanded
= test_bit(STRIPE_EXPAND_READY
, &sh
->state
);
2566 /* Now to look around and see what can be done */
2568 for (i
=disks
; i
--; ) {
2570 struct r5dev
*dev
= &sh
->dev
[i
];
2571 clear_bit(R5_Insync
, &dev
->flags
);
2573 pr_debug("check %d: state 0x%lx toread %p read %p write %p "
2574 "written %p\n", i
, dev
->flags
, dev
->toread
, dev
->read
,
2575 dev
->towrite
, dev
->written
);
2577 /* maybe we can request a biofill operation
2579 * new wantfill requests are only permitted while
2580 * ops_complete_biofill is guaranteed to be inactive
2582 if (test_bit(R5_UPTODATE
, &dev
->flags
) && dev
->toread
&&
2583 !test_bit(STRIPE_BIOFILL_RUN
, &sh
->state
))
2584 set_bit(R5_Wantfill
, &dev
->flags
);
2586 /* now count some things */
2587 if (test_bit(R5_LOCKED
, &dev
->flags
)) s
.locked
++;
2588 if (test_bit(R5_UPTODATE
, &dev
->flags
)) s
.uptodate
++;
2589 if (test_bit(R5_Wantcompute
, &dev
->flags
)) s
.compute
++;
2591 if (test_bit(R5_Wantfill
, &dev
->flags
))
2593 else if (dev
->toread
)
2597 if (!test_bit(R5_OVERWRITE
, &dev
->flags
))
2602 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
2603 if (blocked_rdev
== NULL
&&
2604 rdev
&& unlikely(test_bit(Blocked
, &rdev
->flags
))) {
2605 blocked_rdev
= rdev
;
2606 atomic_inc(&rdev
->nr_pending
);
2608 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)) {
2609 /* The ReadError flag will just be confusing now */
2610 clear_bit(R5_ReadError
, &dev
->flags
);
2611 clear_bit(R5_ReWrite
, &dev
->flags
);
2613 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)
2614 || test_bit(R5_ReadError
, &dev
->flags
)) {
2618 set_bit(R5_Insync
, &dev
->flags
);
2622 if (unlikely(blocked_rdev
)) {
2623 if (s
.syncing
|| s
.expanding
|| s
.expanded
||
2624 s
.to_write
|| s
.written
) {
2625 set_bit(STRIPE_HANDLE
, &sh
->state
);
2628 /* There is nothing for the blocked_rdev to block */
2629 rdev_dec_pending(blocked_rdev
, conf
->mddev
);
2630 blocked_rdev
= NULL
;
2633 if (s
.to_fill
&& !test_bit(STRIPE_BIOFILL_RUN
, &sh
->state
)) {
2634 set_bit(STRIPE_OP_BIOFILL
, &s
.ops_request
);
2635 set_bit(STRIPE_BIOFILL_RUN
, &sh
->state
);
2638 pr_debug("locked=%d uptodate=%d to_read=%d"
2639 " to_write=%d failed=%d failed_num=%d\n",
2640 s
.locked
, s
.uptodate
, s
.to_read
, s
.to_write
,
2641 s
.failed
, s
.failed_num
);
2642 /* check if the array has lost two devices and, if so, some requests might
2645 if (s
.failed
> 1 && s
.to_read
+s
.to_write
+s
.written
)
2646 handle_failed_stripe(conf
, sh
, &s
, disks
, &return_bi
);
2647 if (s
.failed
> 1 && s
.syncing
) {
2648 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,0);
2649 clear_bit(STRIPE_SYNCING
, &sh
->state
);
2653 /* might be able to return some write requests if the parity block
2654 * is safe, or on a failed drive
2656 dev
= &sh
->dev
[sh
->pd_idx
];
2658 ((test_bit(R5_Insync
, &dev
->flags
) &&
2659 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2660 test_bit(R5_UPTODATE
, &dev
->flags
)) ||
2661 (s
.failed
== 1 && s
.failed_num
== sh
->pd_idx
)))
2662 handle_stripe_clean_event(conf
, sh
, disks
, &return_bi
);
2664 /* Now we might consider reading some blocks, either to check/generate
2665 * parity, or to satisfy requests
2666 * or to load a block that is being partially written.
2668 if (s
.to_read
|| s
.non_overwrite
||
2669 (s
.syncing
&& (s
.uptodate
+ s
.compute
< disks
)) || s
.expanding
)
2670 handle_stripe_fill5(sh
, &s
, disks
);
2672 /* Now we check to see if any write operations have recently
2676 if (sh
->reconstruct_state
== reconstruct_state_prexor_drain_result
)
2678 if (sh
->reconstruct_state
== reconstruct_state_drain_result
||
2679 sh
->reconstruct_state
== reconstruct_state_prexor_drain_result
) {
2680 sh
->reconstruct_state
= reconstruct_state_idle
;
2682 /* All the 'written' buffers and the parity block are ready to
2683 * be written back to disk
2685 BUG_ON(!test_bit(R5_UPTODATE
, &sh
->dev
[sh
->pd_idx
].flags
));
2686 for (i
= disks
; i
--; ) {
2688 if (test_bit(R5_LOCKED
, &dev
->flags
) &&
2689 (i
== sh
->pd_idx
|| dev
->written
)) {
2690 pr_debug("Writing block %d\n", i
);
2691 set_bit(R5_Wantwrite
, &dev
->flags
);
2694 if (!test_bit(R5_Insync
, &dev
->flags
) ||
2695 (i
== sh
->pd_idx
&& s
.failed
== 0))
2696 set_bit(STRIPE_INSYNC
, &sh
->state
);
2699 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2700 atomic_dec(&conf
->preread_active_stripes
);
2701 if (atomic_read(&conf
->preread_active_stripes
) <
2703 md_wakeup_thread(conf
->mddev
->thread
);
2707 /* Now to consider new write requests and what else, if anything
2708 * should be read. We do not handle new writes when:
2709 * 1/ A 'write' operation (copy+xor) is already in flight.
2710 * 2/ A 'check' operation is in flight, as it may clobber the parity
2713 if (s
.to_write
&& !sh
->reconstruct_state
&& !sh
->check_state
)
2714 handle_stripe_dirtying5(conf
, sh
, &s
, disks
);
2716 /* maybe we need to check and possibly fix the parity for this stripe
2717 * Any reads will already have been scheduled, so we just see if enough
2718 * data is available. The parity check is held off while parity
2719 * dependent operations are in flight.
2721 if (sh
->check_state
||
2722 (s
.syncing
&& s
.locked
== 0 &&
2723 !test_bit(STRIPE_COMPUTE_RUN
, &sh
->state
) &&
2724 !test_bit(STRIPE_INSYNC
, &sh
->state
)))
2725 handle_parity_checks5(conf
, sh
, &s
, disks
);
2727 if (s
.syncing
&& s
.locked
== 0 && test_bit(STRIPE_INSYNC
, &sh
->state
)) {
2728 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,1);
2729 clear_bit(STRIPE_SYNCING
, &sh
->state
);
2732 /* If the failed drive is just a ReadError, then we might need to progress
2733 * the repair/check process
2735 if (s
.failed
== 1 && !conf
->mddev
->ro
&&
2736 test_bit(R5_ReadError
, &sh
->dev
[s
.failed_num
].flags
)
2737 && !test_bit(R5_LOCKED
, &sh
->dev
[s
.failed_num
].flags
)
2738 && test_bit(R5_UPTODATE
, &sh
->dev
[s
.failed_num
].flags
)
2740 dev
= &sh
->dev
[s
.failed_num
];
2741 if (!test_bit(R5_ReWrite
, &dev
->flags
)) {
2742 set_bit(R5_Wantwrite
, &dev
->flags
);
2743 set_bit(R5_ReWrite
, &dev
->flags
);
2744 set_bit(R5_LOCKED
, &dev
->flags
);
2747 /* let's read it back */
2748 set_bit(R5_Wantread
, &dev
->flags
);
2749 set_bit(R5_LOCKED
, &dev
->flags
);
2754 /* Finish reconstruct operations initiated by the expansion process */
2755 if (sh
->reconstruct_state
== reconstruct_state_result
) {
2756 sh
->reconstruct_state
= reconstruct_state_idle
;
2757 clear_bit(STRIPE_EXPANDING
, &sh
->state
);
2758 for (i
= conf
->raid_disks
; i
--; ) {
2759 set_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
);
2760 set_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
2765 if (s
.expanded
&& test_bit(STRIPE_EXPANDING
, &sh
->state
) &&
2766 !sh
->reconstruct_state
) {
2767 /* Need to write out all blocks after computing parity */
2768 sh
->disks
= conf
->raid_disks
;
2769 sh
->pd_idx
= stripe_to_pdidx(sh
->sector
, conf
,
2771 schedule_reconstruction5(sh
, &s
, 1, 1);
2772 } else if (s
.expanded
&& !sh
->reconstruct_state
&& s
.locked
== 0) {
2773 clear_bit(STRIPE_EXPAND_READY
, &sh
->state
);
2774 atomic_dec(&conf
->reshape_stripes
);
2775 wake_up(&conf
->wait_for_overlap
);
2776 md_done_sync(conf
->mddev
, STRIPE_SECTORS
, 1);
2779 if (s
.expanding
&& s
.locked
== 0 &&
2780 !test_bit(STRIPE_COMPUTE_RUN
, &sh
->state
))
2781 handle_stripe_expansion(conf
, sh
, NULL
);
2784 spin_unlock(&sh
->lock
);
2786 /* wait for this device to become unblocked */
2787 if (unlikely(blocked_rdev
))
2788 md_wait_for_blocked_rdev(blocked_rdev
, conf
->mddev
);
2791 raid5_run_ops(sh
, s
.ops_request
);
2795 return_io(return_bi
);
2797 return blocked_rdev
== NULL
;
2800 static bool handle_stripe6(struct stripe_head
*sh
, struct page
*tmp_page
)
2802 raid5_conf_t
*conf
= sh
->raid_conf
;
2803 int disks
= sh
->disks
;
2804 struct bio
*return_bi
= NULL
;
2805 int i
, pd_idx
= sh
->pd_idx
;
2806 struct stripe_head_state s
;
2807 struct r6_state r6s
;
2808 struct r5dev
*dev
, *pdev
, *qdev
;
2809 mdk_rdev_t
*blocked_rdev
= NULL
;
2811 r6s
.qd_idx
= raid6_next_disk(pd_idx
, disks
);
2812 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
2813 "pd_idx=%d, qd_idx=%d\n",
2814 (unsigned long long)sh
->sector
, sh
->state
,
2815 atomic_read(&sh
->count
), pd_idx
, r6s
.qd_idx
);
2816 memset(&s
, 0, sizeof(s
));
2818 spin_lock(&sh
->lock
);
2819 clear_bit(STRIPE_HANDLE
, &sh
->state
);
2820 clear_bit(STRIPE_DELAYED
, &sh
->state
);
2822 s
.syncing
= test_bit(STRIPE_SYNCING
, &sh
->state
);
2823 s
.expanding
= test_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
2824 s
.expanded
= test_bit(STRIPE_EXPAND_READY
, &sh
->state
);
2825 /* Now to look around and see what can be done */
2828 for (i
=disks
; i
--; ) {
2831 clear_bit(R5_Insync
, &dev
->flags
);
2833 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
2834 i
, dev
->flags
, dev
->toread
, dev
->towrite
, dev
->written
);
2835 /* maybe we can reply to a read */
2836 if (test_bit(R5_UPTODATE
, &dev
->flags
) && dev
->toread
) {
2837 struct bio
*rbi
, *rbi2
;
2838 pr_debug("Return read for disc %d\n", i
);
2839 spin_lock_irq(&conf
->device_lock
);
2842 if (test_and_clear_bit(R5_Overlap
, &dev
->flags
))
2843 wake_up(&conf
->wait_for_overlap
);
2844 spin_unlock_irq(&conf
->device_lock
);
2845 while (rbi
&& rbi
->bi_sector
< dev
->sector
+ STRIPE_SECTORS
) {
2846 copy_data(0, rbi
, dev
->page
, dev
->sector
);
2847 rbi2
= r5_next_bio(rbi
, dev
->sector
);
2848 spin_lock_irq(&conf
->device_lock
);
2849 if (!raid5_dec_bi_phys_segments(rbi
)) {
2850 rbi
->bi_next
= return_bi
;
2853 spin_unlock_irq(&conf
->device_lock
);
2858 /* now count some things */
2859 if (test_bit(R5_LOCKED
, &dev
->flags
)) s
.locked
++;
2860 if (test_bit(R5_UPTODATE
, &dev
->flags
)) s
.uptodate
++;
2867 if (!test_bit(R5_OVERWRITE
, &dev
->flags
))
2872 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
2873 if (blocked_rdev
== NULL
&&
2874 rdev
&& unlikely(test_bit(Blocked
, &rdev
->flags
))) {
2875 blocked_rdev
= rdev
;
2876 atomic_inc(&rdev
->nr_pending
);
2878 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)) {
2879 /* The ReadError flag will just be confusing now */
2880 clear_bit(R5_ReadError
, &dev
->flags
);
2881 clear_bit(R5_ReWrite
, &dev
->flags
);
2883 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)
2884 || test_bit(R5_ReadError
, &dev
->flags
)) {
2886 r6s
.failed_num
[s
.failed
] = i
;
2889 set_bit(R5_Insync
, &dev
->flags
);
2893 if (unlikely(blocked_rdev
)) {
2894 if (s
.syncing
|| s
.expanding
|| s
.expanded
||
2895 s
.to_write
|| s
.written
) {
2896 set_bit(STRIPE_HANDLE
, &sh
->state
);
2899 /* There is nothing for the blocked_rdev to block */
2900 rdev_dec_pending(blocked_rdev
, conf
->mddev
);
2901 blocked_rdev
= NULL
;
2904 pr_debug("locked=%d uptodate=%d to_read=%d"
2905 " to_write=%d failed=%d failed_num=%d,%d\n",
2906 s
.locked
, s
.uptodate
, s
.to_read
, s
.to_write
, s
.failed
,
2907 r6s
.failed_num
[0], r6s
.failed_num
[1]);
2908 /* check if the array has lost >2 devices and, if so, some requests
2909 * might need to be failed
2911 if (s
.failed
> 2 && s
.to_read
+s
.to_write
+s
.written
)
2912 handle_failed_stripe(conf
, sh
, &s
, disks
, &return_bi
);
2913 if (s
.failed
> 2 && s
.syncing
) {
2914 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,0);
2915 clear_bit(STRIPE_SYNCING
, &sh
->state
);
2920 * might be able to return some write requests if the parity blocks
2921 * are safe, or on a failed drive
2923 pdev
= &sh
->dev
[pd_idx
];
2924 r6s
.p_failed
= (s
.failed
>= 1 && r6s
.failed_num
[0] == pd_idx
)
2925 || (s
.failed
>= 2 && r6s
.failed_num
[1] == pd_idx
);
2926 qdev
= &sh
->dev
[r6s
.qd_idx
];
2927 r6s
.q_failed
= (s
.failed
>= 1 && r6s
.failed_num
[0] == r6s
.qd_idx
)
2928 || (s
.failed
>= 2 && r6s
.failed_num
[1] == r6s
.qd_idx
);
2931 ( r6s
.p_failed
|| ((test_bit(R5_Insync
, &pdev
->flags
)
2932 && !test_bit(R5_LOCKED
, &pdev
->flags
)
2933 && test_bit(R5_UPTODATE
, &pdev
->flags
)))) &&
2934 ( r6s
.q_failed
|| ((test_bit(R5_Insync
, &qdev
->flags
)
2935 && !test_bit(R5_LOCKED
, &qdev
->flags
)
2936 && test_bit(R5_UPTODATE
, &qdev
->flags
)))))
2937 handle_stripe_clean_event(conf
, sh
, disks
, &return_bi
);
2939 /* Now we might consider reading some blocks, either to check/generate
2940 * parity, or to satisfy requests
2941 * or to load a block that is being partially written.
2943 if (s
.to_read
|| s
.non_overwrite
|| (s
.to_write
&& s
.failed
) ||
2944 (s
.syncing
&& (s
.uptodate
< disks
)) || s
.expanding
)
2945 handle_stripe_fill6(sh
, &s
, &r6s
, disks
);
2947 /* now to consider writing and what else, if anything should be read */
2949 handle_stripe_dirtying6(conf
, sh
, &s
, &r6s
, disks
);
2951 /* maybe we need to check and possibly fix the parity for this stripe
2952 * Any reads will already have been scheduled, so we just see if enough
2955 if (s
.syncing
&& s
.locked
== 0 && !test_bit(STRIPE_INSYNC
, &sh
->state
))
2956 handle_parity_checks6(conf
, sh
, &s
, &r6s
, tmp_page
, disks
);
2958 if (s
.syncing
&& s
.locked
== 0 && test_bit(STRIPE_INSYNC
, &sh
->state
)) {
2959 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,1);
2960 clear_bit(STRIPE_SYNCING
, &sh
->state
);
2963 /* If the failed drives are just a ReadError, then we might need
2964 * to progress the repair/check process
2966 if (s
.failed
<= 2 && !conf
->mddev
->ro
)
2967 for (i
= 0; i
< s
.failed
; i
++) {
2968 dev
= &sh
->dev
[r6s
.failed_num
[i
]];
2969 if (test_bit(R5_ReadError
, &dev
->flags
)
2970 && !test_bit(R5_LOCKED
, &dev
->flags
)
2971 && test_bit(R5_UPTODATE
, &dev
->flags
)
2973 if (!test_bit(R5_ReWrite
, &dev
->flags
)) {
2974 set_bit(R5_Wantwrite
, &dev
->flags
);
2975 set_bit(R5_ReWrite
, &dev
->flags
);
2976 set_bit(R5_LOCKED
, &dev
->flags
);
2978 /* let's read it back */
2979 set_bit(R5_Wantread
, &dev
->flags
);
2980 set_bit(R5_LOCKED
, &dev
->flags
);
2985 if (s
.expanded
&& test_bit(STRIPE_EXPANDING
, &sh
->state
)) {
2986 /* Need to write out all blocks after computing P&Q */
2987 sh
->disks
= conf
->raid_disks
;
2988 sh
->pd_idx
= stripe_to_pdidx(sh
->sector
, conf
,
2990 compute_parity6(sh
, RECONSTRUCT_WRITE
);
2991 for (i
= conf
->raid_disks
; i
-- ; ) {
2992 set_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
2994 set_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
);
2996 clear_bit(STRIPE_EXPANDING
, &sh
->state
);
2997 } else if (s
.expanded
) {
2998 clear_bit(STRIPE_EXPAND_READY
, &sh
->state
);
2999 atomic_dec(&conf
->reshape_stripes
);
3000 wake_up(&conf
->wait_for_overlap
);
3001 md_done_sync(conf
->mddev
, STRIPE_SECTORS
, 1);
3004 if (s
.expanding
&& s
.locked
== 0 &&
3005 !test_bit(STRIPE_COMPUTE_RUN
, &sh
->state
))
3006 handle_stripe_expansion(conf
, sh
, &r6s
);
3009 spin_unlock(&sh
->lock
);
3011 /* wait for this device to become unblocked */
3012 if (unlikely(blocked_rdev
))
3013 md_wait_for_blocked_rdev(blocked_rdev
, conf
->mddev
);
3017 return_io(return_bi
);
3019 return blocked_rdev
== NULL
;
3022 /* returns true if the stripe was handled */
3023 static bool handle_stripe(struct stripe_head
*sh
, struct page
*tmp_page
)
3025 if (sh
->raid_conf
->level
== 6)
3026 return handle_stripe6(sh
, tmp_page
);
3028 return handle_stripe5(sh
);
3033 static void raid5_activate_delayed(raid5_conf_t
*conf
)
3035 if (atomic_read(&conf
->preread_active_stripes
) < IO_THRESHOLD
) {
3036 while (!list_empty(&conf
->delayed_list
)) {
3037 struct list_head
*l
= conf
->delayed_list
.next
;
3038 struct stripe_head
*sh
;
3039 sh
= list_entry(l
, struct stripe_head
, lru
);
3041 clear_bit(STRIPE_DELAYED
, &sh
->state
);
3042 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
))
3043 atomic_inc(&conf
->preread_active_stripes
);
3044 list_add_tail(&sh
->lru
, &conf
->hold_list
);
3047 blk_plug_device(conf
->mddev
->queue
);
3050 static void activate_bit_delay(raid5_conf_t
*conf
)
3052 /* device_lock is held */
3053 struct list_head head
;
3054 list_add(&head
, &conf
->bitmap_list
);
3055 list_del_init(&conf
->bitmap_list
);
3056 while (!list_empty(&head
)) {
3057 struct stripe_head
*sh
= list_entry(head
.next
, struct stripe_head
, lru
);
3058 list_del_init(&sh
->lru
);
3059 atomic_inc(&sh
->count
);
3060 __release_stripe(conf
, sh
);
3064 static void unplug_slaves(mddev_t
*mddev
)
3066 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3070 for (i
=0; i
<mddev
->raid_disks
; i
++) {
3071 mdk_rdev_t
*rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
3072 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
) && atomic_read(&rdev
->nr_pending
)) {
3073 struct request_queue
*r_queue
= bdev_get_queue(rdev
->bdev
);
3075 atomic_inc(&rdev
->nr_pending
);
3078 blk_unplug(r_queue
);
3080 rdev_dec_pending(rdev
, mddev
);
3087 static void raid5_unplug_device(struct request_queue
*q
)
3089 mddev_t
*mddev
= q
->queuedata
;
3090 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3091 unsigned long flags
;
3093 spin_lock_irqsave(&conf
->device_lock
, flags
);
3095 if (blk_remove_plug(q
)) {
3097 raid5_activate_delayed(conf
);
3099 md_wakeup_thread(mddev
->thread
);
3101 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
3103 unplug_slaves(mddev
);
3106 static int raid5_congested(void *data
, int bits
)
3108 mddev_t
*mddev
= data
;
3109 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3111 /* No difference between reads and writes. Just check
3112 * how busy the stripe_cache is
3114 if (conf
->inactive_blocked
)
3118 if (list_empty_careful(&conf
->inactive_list
))
3124 /* We want read requests to align with chunks where possible,
3125 * but write requests don't need to.
3127 static int raid5_mergeable_bvec(struct request_queue
*q
,
3128 struct bvec_merge_data
*bvm
,
3129 struct bio_vec
*biovec
)
3131 mddev_t
*mddev
= q
->queuedata
;
3132 sector_t sector
= bvm
->bi_sector
+ get_start_sect(bvm
->bi_bdev
);
3134 unsigned int chunk_sectors
= mddev
->chunk_size
>> 9;
3135 unsigned int bio_sectors
= bvm
->bi_size
>> 9;
3137 if ((bvm
->bi_rw
& 1) == WRITE
)
3138 return biovec
->bv_len
; /* always allow writes to be mergeable */
3140 max
= (chunk_sectors
- ((sector
& (chunk_sectors
- 1)) + bio_sectors
)) << 9;
3141 if (max
< 0) max
= 0;
3142 if (max
<= biovec
->bv_len
&& bio_sectors
== 0)
3143 return biovec
->bv_len
;
3149 static int in_chunk_boundary(mddev_t
*mddev
, struct bio
*bio
)
3151 sector_t sector
= bio
->bi_sector
+ get_start_sect(bio
->bi_bdev
);
3152 unsigned int chunk_sectors
= mddev
->chunk_size
>> 9;
3153 unsigned int bio_sectors
= bio
->bi_size
>> 9;
3155 return chunk_sectors
>=
3156 ((sector
& (chunk_sectors
- 1)) + bio_sectors
);
3160 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
3161 * later sampled by raid5d.
3163 static void add_bio_to_retry(struct bio
*bi
,raid5_conf_t
*conf
)
3165 unsigned long flags
;
3167 spin_lock_irqsave(&conf
->device_lock
, flags
);
3169 bi
->bi_next
= conf
->retry_read_aligned_list
;
3170 conf
->retry_read_aligned_list
= bi
;
3172 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
3173 md_wakeup_thread(conf
->mddev
->thread
);
3177 static struct bio
*remove_bio_from_retry(raid5_conf_t
*conf
)
3181 bi
= conf
->retry_read_aligned
;
3183 conf
->retry_read_aligned
= NULL
;
3186 bi
= conf
->retry_read_aligned_list
;
3188 conf
->retry_read_aligned_list
= bi
->bi_next
;
3191 * this sets the active strip count to 1 and the processed
3192 * strip count to zero (upper 8 bits)
3194 bi
->bi_phys_segments
= 1; /* biased count of active stripes */
3202 * The "raid5_align_endio" should check if the read succeeded and if it
3203 * did, call bio_endio on the original bio (having bio_put the new bio
3205 * If the read failed..
3207 static void raid5_align_endio(struct bio
*bi
, int error
)
3209 struct bio
* raid_bi
= bi
->bi_private
;
3212 int uptodate
= test_bit(BIO_UPTODATE
, &bi
->bi_flags
);
3217 mddev
= raid_bi
->bi_bdev
->bd_disk
->queue
->queuedata
;
3218 conf
= mddev_to_conf(mddev
);
3219 rdev
= (void*)raid_bi
->bi_next
;
3220 raid_bi
->bi_next
= NULL
;
3222 rdev_dec_pending(rdev
, conf
->mddev
);
3224 if (!error
&& uptodate
) {
3225 bio_endio(raid_bi
, 0);
3226 if (atomic_dec_and_test(&conf
->active_aligned_reads
))
3227 wake_up(&conf
->wait_for_stripe
);
3232 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
3234 add_bio_to_retry(raid_bi
, conf
);
3237 static int bio_fits_rdev(struct bio
*bi
)
3239 struct request_queue
*q
= bdev_get_queue(bi
->bi_bdev
);
3241 if ((bi
->bi_size
>>9) > q
->max_sectors
)
3243 blk_recount_segments(q
, bi
);
3244 if (bi
->bi_phys_segments
> q
->max_phys_segments
)
3247 if (q
->merge_bvec_fn
)
3248 /* it's too hard to apply the merge_bvec_fn at this stage,
3257 static int chunk_aligned_read(struct request_queue
*q
, struct bio
* raid_bio
)
3259 mddev_t
*mddev
= q
->queuedata
;
3260 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3261 const unsigned int raid_disks
= conf
->raid_disks
;
3262 const unsigned int data_disks
= raid_disks
- conf
->max_degraded
;
3263 unsigned int dd_idx
, pd_idx
;
3264 struct bio
* align_bi
;
3267 if (!in_chunk_boundary(mddev
, raid_bio
)) {
3268 pr_debug("chunk_aligned_read : non aligned\n");
3272 * use bio_clone to make a copy of the bio
3274 align_bi
= bio_clone(raid_bio
, GFP_NOIO
);
3278 * set bi_end_io to a new function, and set bi_private to the
3281 align_bi
->bi_end_io
= raid5_align_endio
;
3282 align_bi
->bi_private
= raid_bio
;
3286 align_bi
->bi_sector
= raid5_compute_sector(raid_bio
->bi_sector
,
3294 rdev
= rcu_dereference(conf
->disks
[dd_idx
].rdev
);
3295 if (rdev
&& test_bit(In_sync
, &rdev
->flags
)) {
3296 atomic_inc(&rdev
->nr_pending
);
3298 raid_bio
->bi_next
= (void*)rdev
;
3299 align_bi
->bi_bdev
= rdev
->bdev
;
3300 align_bi
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
3301 align_bi
->bi_sector
+= rdev
->data_offset
;
3303 if (!bio_fits_rdev(align_bi
)) {
3304 /* too big in some way */
3306 rdev_dec_pending(rdev
, mddev
);
3310 spin_lock_irq(&conf
->device_lock
);
3311 wait_event_lock_irq(conf
->wait_for_stripe
,
3313 conf
->device_lock
, /* nothing */);
3314 atomic_inc(&conf
->active_aligned_reads
);
3315 spin_unlock_irq(&conf
->device_lock
);
3317 generic_make_request(align_bi
);
3326 /* __get_priority_stripe - get the next stripe to process
3328 * Full stripe writes are allowed to pass preread active stripes up until
3329 * the bypass_threshold is exceeded. In general the bypass_count
3330 * increments when the handle_list is handled before the hold_list; however, it
3331 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
3332 * stripe with in flight i/o. The bypass_count will be reset when the
3333 * head of the hold_list has changed, i.e. the head was promoted to the
3336 static struct stripe_head
*__get_priority_stripe(raid5_conf_t
*conf
)
3338 struct stripe_head
*sh
;
3340 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
3342 list_empty(&conf
->handle_list
) ? "empty" : "busy",
3343 list_empty(&conf
->hold_list
) ? "empty" : "busy",
3344 atomic_read(&conf
->pending_full_writes
), conf
->bypass_count
);
3346 if (!list_empty(&conf
->handle_list
)) {
3347 sh
= list_entry(conf
->handle_list
.next
, typeof(*sh
), lru
);
3349 if (list_empty(&conf
->hold_list
))
3350 conf
->bypass_count
= 0;
3351 else if (!test_bit(STRIPE_IO_STARTED
, &sh
->state
)) {
3352 if (conf
->hold_list
.next
== conf
->last_hold
)
3353 conf
->bypass_count
++;
3355 conf
->last_hold
= conf
->hold_list
.next
;
3356 conf
->bypass_count
-= conf
->bypass_threshold
;
3357 if (conf
->bypass_count
< 0)
3358 conf
->bypass_count
= 0;
3361 } else if (!list_empty(&conf
->hold_list
) &&
3362 ((conf
->bypass_threshold
&&
3363 conf
->bypass_count
> conf
->bypass_threshold
) ||
3364 atomic_read(&conf
->pending_full_writes
) == 0)) {
3365 sh
= list_entry(conf
->hold_list
.next
,
3367 conf
->bypass_count
-= conf
->bypass_threshold
;
3368 if (conf
->bypass_count
< 0)
3369 conf
->bypass_count
= 0;
3373 list_del_init(&sh
->lru
);
3374 atomic_inc(&sh
->count
);
3375 BUG_ON(atomic_read(&sh
->count
) != 1);
3379 static int make_request(struct request_queue
*q
, struct bio
* bi
)
3381 mddev_t
*mddev
= q
->queuedata
;
3382 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3383 unsigned int dd_idx
, pd_idx
;
3384 sector_t new_sector
;
3385 sector_t logical_sector
, last_sector
;
3386 struct stripe_head
*sh
;
3387 const int rw
= bio_data_dir(bi
);
3390 if (unlikely(bio_barrier(bi
))) {
3391 bio_endio(bi
, -EOPNOTSUPP
);
3395 md_write_start(mddev
, bi
);
3397 cpu
= part_stat_lock();
3398 part_stat_inc(cpu
, &mddev
->gendisk
->part0
, ios
[rw
]);
3399 part_stat_add(cpu
, &mddev
->gendisk
->part0
, sectors
[rw
],
3404 mddev
->reshape_position
== MaxSector
&&
3405 chunk_aligned_read(q
,bi
))
3408 logical_sector
= bi
->bi_sector
& ~((sector_t
)STRIPE_SECTORS
-1);
3409 last_sector
= bi
->bi_sector
+ (bi
->bi_size
>>9);
3411 bi
->bi_phys_segments
= 1; /* over-loaded to count active stripes */
3413 for (;logical_sector
< last_sector
; logical_sector
+= STRIPE_SECTORS
) {
3415 int disks
, data_disks
;
3418 prepare_to_wait(&conf
->wait_for_overlap
, &w
, TASK_UNINTERRUPTIBLE
);
3419 if (likely(conf
->expand_progress
== MaxSector
))
3420 disks
= conf
->raid_disks
;
3422 /* spinlock is needed as expand_progress may be
3423 * 64bit on a 32bit platform, and so it might be
3424 * possible to see a half-updated value
3425 * Ofcourse expand_progress could change after
3426 * the lock is dropped, so once we get a reference
3427 * to the stripe that we think it is, we will have
3430 spin_lock_irq(&conf
->device_lock
);
3431 disks
= conf
->raid_disks
;
3432 if (logical_sector
>= conf
->expand_progress
)
3433 disks
= conf
->previous_raid_disks
;
3435 if (logical_sector
>= conf
->expand_lo
) {
3436 spin_unlock_irq(&conf
->device_lock
);
3441 spin_unlock_irq(&conf
->device_lock
);
3443 data_disks
= disks
- conf
->max_degraded
;
3445 new_sector
= raid5_compute_sector(logical_sector
, disks
, data_disks
,
3446 &dd_idx
, &pd_idx
, conf
);
3447 pr_debug("raid5: make_request, sector %llu logical %llu\n",
3448 (unsigned long long)new_sector
,
3449 (unsigned long long)logical_sector
);
3451 sh
= get_active_stripe(conf
, new_sector
, disks
, pd_idx
, (bi
->bi_rw
&RWA_MASK
));
3453 if (unlikely(conf
->expand_progress
!= MaxSector
)) {
3454 /* expansion might have moved on while waiting for a
3455 * stripe, so we must do the range check again.
3456 * Expansion could still move past after this
3457 * test, but as we are holding a reference to
3458 * 'sh', we know that if that happens,
3459 * STRIPE_EXPANDING will get set and the expansion
3460 * won't proceed until we finish with the stripe.
3463 spin_lock_irq(&conf
->device_lock
);
3464 if (logical_sector
< conf
->expand_progress
&&
3465 disks
== conf
->previous_raid_disks
)
3466 /* mismatch, need to try again */
3468 spin_unlock_irq(&conf
->device_lock
);
3474 /* FIXME what if we get a false positive because these
3475 * are being updated.
3477 if (logical_sector
>= mddev
->suspend_lo
&&
3478 logical_sector
< mddev
->suspend_hi
) {
3484 if (test_bit(STRIPE_EXPANDING
, &sh
->state
) ||
3485 !add_stripe_bio(sh
, bi
, dd_idx
, (bi
->bi_rw
&RW_MASK
))) {
3486 /* Stripe is busy expanding or
3487 * add failed due to overlap. Flush everything
3490 raid5_unplug_device(mddev
->queue
);
3495 finish_wait(&conf
->wait_for_overlap
, &w
);
3496 set_bit(STRIPE_HANDLE
, &sh
->state
);
3497 clear_bit(STRIPE_DELAYED
, &sh
->state
);
3500 /* cannot get stripe for read-ahead, just give-up */
3501 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
3502 finish_wait(&conf
->wait_for_overlap
, &w
);
3507 spin_lock_irq(&conf
->device_lock
);
3508 remaining
= raid5_dec_bi_phys_segments(bi
);
3509 spin_unlock_irq(&conf
->device_lock
);
3510 if (remaining
== 0) {
3513 md_write_end(mddev
);
3520 static sector_t
reshape_request(mddev_t
*mddev
, sector_t sector_nr
, int *skipped
)
3522 /* reshaping is quite different to recovery/resync so it is
3523 * handled quite separately ... here.
3525 * On each call to sync_request, we gather one chunk worth of
3526 * destination stripes and flag them as expanding.
3527 * Then we find all the source stripes and request reads.
3528 * As the reads complete, handle_stripe will copy the data
3529 * into the destination stripe and release that stripe.
3531 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
3532 struct stripe_head
*sh
;
3534 sector_t first_sector
, last_sector
;
3535 int raid_disks
= conf
->previous_raid_disks
;
3536 int data_disks
= raid_disks
- conf
->max_degraded
;
3537 int new_data_disks
= conf
->raid_disks
- conf
->max_degraded
;
3540 sector_t writepos
, safepos
, gap
;
3542 if (sector_nr
== 0 &&
3543 conf
->expand_progress
!= 0) {
3544 /* restarting in the middle, skip the initial sectors */
3545 sector_nr
= conf
->expand_progress
;
3546 sector_div(sector_nr
, new_data_disks
);
3551 /* we update the metadata when there is more than 3Meg
3552 * in the block range (that is rather arbitrary, should
3553 * probably be time based) or when the data about to be
3554 * copied would over-write the source of the data at
3555 * the front of the range.
3556 * i.e. one new_stripe forward from expand_progress new_maps
3557 * to after where expand_lo old_maps to
3559 writepos
= conf
->expand_progress
+
3560 conf
->chunk_size
/512*(new_data_disks
);
3561 sector_div(writepos
, new_data_disks
);
3562 safepos
= conf
->expand_lo
;
3563 sector_div(safepos
, data_disks
);
3564 gap
= conf
->expand_progress
- conf
->expand_lo
;
3566 if (writepos
>= safepos
||
3567 gap
> (new_data_disks
)*3000*2 /*3Meg*/) {
3568 /* Cannot proceed until we've updated the superblock... */
3569 wait_event(conf
->wait_for_overlap
,
3570 atomic_read(&conf
->reshape_stripes
)==0);
3571 mddev
->reshape_position
= conf
->expand_progress
;
3572 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3573 md_wakeup_thread(mddev
->thread
);
3574 wait_event(mddev
->sb_wait
, mddev
->flags
== 0 ||
3575 kthread_should_stop());
3576 spin_lock_irq(&conf
->device_lock
);
3577 conf
->expand_lo
= mddev
->reshape_position
;
3578 spin_unlock_irq(&conf
->device_lock
);
3579 wake_up(&conf
->wait_for_overlap
);
3582 for (i
=0; i
< conf
->chunk_size
/512; i
+= STRIPE_SECTORS
) {
3585 pd_idx
= stripe_to_pdidx(sector_nr
+i
, conf
, conf
->raid_disks
);
3586 sh
= get_active_stripe(conf
, sector_nr
+i
,
3587 conf
->raid_disks
, pd_idx
, 0);
3588 set_bit(STRIPE_EXPANDING
, &sh
->state
);
3589 atomic_inc(&conf
->reshape_stripes
);
3590 /* If any of this stripe is beyond the end of the old
3591 * array, then we need to zero those blocks
3593 for (j
=sh
->disks
; j
--;) {
3595 if (j
== sh
->pd_idx
)
3597 if (conf
->level
== 6 &&
3598 j
== raid6_next_disk(sh
->pd_idx
, sh
->disks
))
3600 s
= compute_blocknr(sh
, j
);
3601 if (s
< mddev
->array_sectors
) {
3605 memset(page_address(sh
->dev
[j
].page
), 0, STRIPE_SIZE
);
3606 set_bit(R5_Expanded
, &sh
->dev
[j
].flags
);
3607 set_bit(R5_UPTODATE
, &sh
->dev
[j
].flags
);
3610 set_bit(STRIPE_EXPAND_READY
, &sh
->state
);
3611 set_bit(STRIPE_HANDLE
, &sh
->state
);
3615 spin_lock_irq(&conf
->device_lock
);
3616 conf
->expand_progress
= (sector_nr
+ i
) * new_data_disks
;
3617 spin_unlock_irq(&conf
->device_lock
);
3618 /* Ok, those stripe are ready. We can start scheduling
3619 * reads on the source stripes.
3620 * The source stripes are determined by mapping the first and last
3621 * block on the destination stripes.
3624 raid5_compute_sector(sector_nr
*(new_data_disks
),
3625 raid_disks
, data_disks
,
3626 &dd_idx
, &pd_idx
, conf
);
3628 raid5_compute_sector((sector_nr
+conf
->chunk_size
/512)
3629 *(new_data_disks
) -1,
3630 raid_disks
, data_disks
,
3631 &dd_idx
, &pd_idx
, conf
);
3632 if (last_sector
>= (mddev
->size
<<1))
3633 last_sector
= (mddev
->size
<<1)-1;
3634 while (first_sector
<= last_sector
) {
3635 pd_idx
= stripe_to_pdidx(first_sector
, conf
,
3636 conf
->previous_raid_disks
);
3637 sh
= get_active_stripe(conf
, first_sector
,
3638 conf
->previous_raid_disks
, pd_idx
, 0);
3639 set_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
3640 set_bit(STRIPE_HANDLE
, &sh
->state
);
3642 first_sector
+= STRIPE_SECTORS
;
3644 /* If this takes us to the resync_max point where we have to pause,
3645 * then we need to write out the superblock.
3647 sector_nr
+= conf
->chunk_size
>>9;
3648 if (sector_nr
>= mddev
->resync_max
) {
3649 /* Cannot proceed until we've updated the superblock... */
3650 wait_event(conf
->wait_for_overlap
,
3651 atomic_read(&conf
->reshape_stripes
) == 0);
3652 mddev
->reshape_position
= conf
->expand_progress
;
3653 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3654 md_wakeup_thread(mddev
->thread
);
3655 wait_event(mddev
->sb_wait
,
3656 !test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)
3657 || kthread_should_stop());
3658 spin_lock_irq(&conf
->device_lock
);
3659 conf
->expand_lo
= mddev
->reshape_position
;
3660 spin_unlock_irq(&conf
->device_lock
);
3661 wake_up(&conf
->wait_for_overlap
);
3663 return conf
->chunk_size
>>9;
3666 /* FIXME go_faster isn't used */
3667 static inline sector_t
sync_request(mddev_t
*mddev
, sector_t sector_nr
, int *skipped
, int go_faster
)
3669 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
3670 struct stripe_head
*sh
;
3672 int raid_disks
= conf
->raid_disks
;
3673 sector_t max_sector
= mddev
->size
<< 1;
3675 int still_degraded
= 0;
3678 if (sector_nr
>= max_sector
) {
3679 /* just being told to finish up .. nothing much to do */
3680 unplug_slaves(mddev
);
3681 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)) {
3686 if (mddev
->curr_resync
< max_sector
) /* aborted */
3687 bitmap_end_sync(mddev
->bitmap
, mddev
->curr_resync
,
3689 else /* completed sync */
3691 bitmap_close_sync(mddev
->bitmap
);
3696 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
3697 return reshape_request(mddev
, sector_nr
, skipped
);
3699 /* No need to check resync_max as we never do more than one
3700 * stripe, and as resync_max will always be on a chunk boundary,
3701 * if the check in md_do_sync didn't fire, there is no chance
3702 * of overstepping resync_max here
3705 /* if there is too many failed drives and we are trying
3706 * to resync, then assert that we are finished, because there is
3707 * nothing we can do.
3709 if (mddev
->degraded
>= conf
->max_degraded
&&
3710 test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
3711 sector_t rv
= (mddev
->size
<< 1) - sector_nr
;
3715 if (!bitmap_start_sync(mddev
->bitmap
, sector_nr
, &sync_blocks
, 1) &&
3716 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
) &&
3717 !conf
->fullsync
&& sync_blocks
>= STRIPE_SECTORS
) {
3718 /* we can skip this block, and probably more */
3719 sync_blocks
/= STRIPE_SECTORS
;
3721 return sync_blocks
* STRIPE_SECTORS
; /* keep things rounded to whole stripes */
3725 bitmap_cond_end_sync(mddev
->bitmap
, sector_nr
);
3727 pd_idx
= stripe_to_pdidx(sector_nr
, conf
, raid_disks
);
3728 sh
= get_active_stripe(conf
, sector_nr
, raid_disks
, pd_idx
, 1);
3730 sh
= get_active_stripe(conf
, sector_nr
, raid_disks
, pd_idx
, 0);
3731 /* make sure we don't swamp the stripe cache if someone else
3732 * is trying to get access
3734 schedule_timeout_uninterruptible(1);
3736 /* Need to check if array will still be degraded after recovery/resync
3737 * We don't need to check the 'failed' flag as when that gets set,
3740 for (i
=0; i
<mddev
->raid_disks
; i
++)
3741 if (conf
->disks
[i
].rdev
== NULL
)
3744 bitmap_start_sync(mddev
->bitmap
, sector_nr
, &sync_blocks
, still_degraded
);
3746 spin_lock(&sh
->lock
);
3747 set_bit(STRIPE_SYNCING
, &sh
->state
);
3748 clear_bit(STRIPE_INSYNC
, &sh
->state
);
3749 spin_unlock(&sh
->lock
);
3751 /* wait for any blocked device to be handled */
3752 while(unlikely(!handle_stripe(sh
, NULL
)))
3756 return STRIPE_SECTORS
;
3759 static int retry_aligned_read(raid5_conf_t
*conf
, struct bio
*raid_bio
)
3761 /* We may not be able to submit a whole bio at once as there
3762 * may not be enough stripe_heads available.
3763 * We cannot pre-allocate enough stripe_heads as we may need
3764 * more than exist in the cache (if we allow ever large chunks).
3765 * So we do one stripe head at a time and record in
3766 * ->bi_hw_segments how many have been done.
3768 * We *know* that this entire raid_bio is in one chunk, so
3769 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
3771 struct stripe_head
*sh
;
3773 sector_t sector
, logical_sector
, last_sector
;
3778 logical_sector
= raid_bio
->bi_sector
& ~((sector_t
)STRIPE_SECTORS
-1);
3779 sector
= raid5_compute_sector( logical_sector
,
3781 conf
->raid_disks
- conf
->max_degraded
,
3785 last_sector
= raid_bio
->bi_sector
+ (raid_bio
->bi_size
>>9);
3787 for (; logical_sector
< last_sector
;
3788 logical_sector
+= STRIPE_SECTORS
,
3789 sector
+= STRIPE_SECTORS
,
3792 if (scnt
< raid5_bi_hw_segments(raid_bio
))
3793 /* already done this stripe */
3796 sh
= get_active_stripe(conf
, sector
, conf
->raid_disks
, pd_idx
, 1);
3799 /* failed to get a stripe - must wait */
3800 raid5_set_bi_hw_segments(raid_bio
, scnt
);
3801 conf
->retry_read_aligned
= raid_bio
;
3805 set_bit(R5_ReadError
, &sh
->dev
[dd_idx
].flags
);
3806 if (!add_stripe_bio(sh
, raid_bio
, dd_idx
, 0)) {
3808 raid5_set_bi_hw_segments(raid_bio
, scnt
);
3809 conf
->retry_read_aligned
= raid_bio
;
3813 handle_stripe(sh
, NULL
);
3817 spin_lock_irq(&conf
->device_lock
);
3818 remaining
= raid5_dec_bi_phys_segments(raid_bio
);
3819 spin_unlock_irq(&conf
->device_lock
);
3821 bio_endio(raid_bio
, 0);
3822 if (atomic_dec_and_test(&conf
->active_aligned_reads
))
3823 wake_up(&conf
->wait_for_stripe
);
3830 * This is our raid5 kernel thread.
3832 * We scan the hash table for stripes which can be handled now.
3833 * During the scan, completed stripes are saved for us by the interrupt
3834 * handler, so that they will not have to wait for our next wakeup.
3836 static void raid5d(mddev_t
*mddev
)
3838 struct stripe_head
*sh
;
3839 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3842 pr_debug("+++ raid5d active\n");
3844 md_check_recovery(mddev
);
3847 spin_lock_irq(&conf
->device_lock
);
3851 if (conf
->seq_flush
!= conf
->seq_write
) {
3852 int seq
= conf
->seq_flush
;
3853 spin_unlock_irq(&conf
->device_lock
);
3854 bitmap_unplug(mddev
->bitmap
);
3855 spin_lock_irq(&conf
->device_lock
);
3856 conf
->seq_write
= seq
;
3857 activate_bit_delay(conf
);
3860 while ((bio
= remove_bio_from_retry(conf
))) {
3862 spin_unlock_irq(&conf
->device_lock
);
3863 ok
= retry_aligned_read(conf
, bio
);
3864 spin_lock_irq(&conf
->device_lock
);
3870 sh
= __get_priority_stripe(conf
);
3874 spin_unlock_irq(&conf
->device_lock
);
3877 handle_stripe(sh
, conf
->spare_page
);
3880 spin_lock_irq(&conf
->device_lock
);
3882 pr_debug("%d stripes handled\n", handled
);
3884 spin_unlock_irq(&conf
->device_lock
);
3886 async_tx_issue_pending_all();
3887 unplug_slaves(mddev
);
3889 pr_debug("--- raid5d inactive\n");
3893 raid5_show_stripe_cache_size(mddev_t
*mddev
, char *page
)
3895 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3897 return sprintf(page
, "%d\n", conf
->max_nr_stripes
);
3903 raid5_store_stripe_cache_size(mddev_t
*mddev
, const char *page
, size_t len
)
3905 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3909 if (len
>= PAGE_SIZE
)
3914 if (strict_strtoul(page
, 10, &new))
3916 if (new <= 16 || new > 32768)
3918 while (new < conf
->max_nr_stripes
) {
3919 if (drop_one_stripe(conf
))
3920 conf
->max_nr_stripes
--;
3924 err
= md_allow_write(mddev
);
3927 while (new > conf
->max_nr_stripes
) {
3928 if (grow_one_stripe(conf
))
3929 conf
->max_nr_stripes
++;
3935 static struct md_sysfs_entry
3936 raid5_stripecache_size
= __ATTR(stripe_cache_size
, S_IRUGO
| S_IWUSR
,
3937 raid5_show_stripe_cache_size
,
3938 raid5_store_stripe_cache_size
);
3941 raid5_show_preread_threshold(mddev_t
*mddev
, char *page
)
3943 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3945 return sprintf(page
, "%d\n", conf
->bypass_threshold
);
3951 raid5_store_preread_threshold(mddev_t
*mddev
, const char *page
, size_t len
)
3953 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3955 if (len
>= PAGE_SIZE
)
3960 if (strict_strtoul(page
, 10, &new))
3962 if (new > conf
->max_nr_stripes
)
3964 conf
->bypass_threshold
= new;
3968 static struct md_sysfs_entry
3969 raid5_preread_bypass_threshold
= __ATTR(preread_bypass_threshold
,
3971 raid5_show_preread_threshold
,
3972 raid5_store_preread_threshold
);
3975 stripe_cache_active_show(mddev_t
*mddev
, char *page
)
3977 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3979 return sprintf(page
, "%d\n", atomic_read(&conf
->active_stripes
));
3984 static struct md_sysfs_entry
3985 raid5_stripecache_active
= __ATTR_RO(stripe_cache_active
);
3987 static struct attribute
*raid5_attrs
[] = {
3988 &raid5_stripecache_size
.attr
,
3989 &raid5_stripecache_active
.attr
,
3990 &raid5_preread_bypass_threshold
.attr
,
3993 static struct attribute_group raid5_attrs_group
= {
3995 .attrs
= raid5_attrs
,
3998 static int run(mddev_t
*mddev
)
4001 int raid_disk
, memory
;
4003 struct disk_info
*disk
;
4004 int working_disks
= 0;
4006 if (mddev
->level
!= 5 && mddev
->level
!= 4 && mddev
->level
!= 6) {
4007 printk(KERN_ERR
"raid5: %s: raid level not set to 4/5/6 (%d)\n",
4008 mdname(mddev
), mddev
->level
);
4012 if (mddev
->chunk_size
< PAGE_SIZE
) {
4013 printk(KERN_ERR
"md/raid5: chunk_size must be at least "
4014 "PAGE_SIZE but %d < %ld\n",
4015 mddev
->chunk_size
, PAGE_SIZE
);
4019 if (mddev
->reshape_position
!= MaxSector
) {
4020 /* Check that we can continue the reshape.
4021 * Currently only disks can change, it must
4022 * increase, and we must be past the point where
4023 * a stripe over-writes itself
4025 sector_t here_new
, here_old
;
4027 int max_degraded
= (mddev
->level
== 5 ? 1 : 2);
4029 if (mddev
->new_level
!= mddev
->level
||
4030 mddev
->new_layout
!= mddev
->layout
||
4031 mddev
->new_chunk
!= mddev
->chunk_size
) {
4032 printk(KERN_ERR
"raid5: %s: unsupported reshape "
4033 "required - aborting.\n",
4037 if (mddev
->delta_disks
<= 0) {
4038 printk(KERN_ERR
"raid5: %s: unsupported reshape "
4039 "(reduce disks) required - aborting.\n",
4043 old_disks
= mddev
->raid_disks
- mddev
->delta_disks
;
4044 /* reshape_position must be on a new-stripe boundary, and one
4045 * further up in new geometry must map after here in old
4048 here_new
= mddev
->reshape_position
;
4049 if (sector_div(here_new
, (mddev
->chunk_size
>>9)*
4050 (mddev
->raid_disks
- max_degraded
))) {
4051 printk(KERN_ERR
"raid5: reshape_position not "
4052 "on a stripe boundary\n");
4055 /* here_new is the stripe we will write to */
4056 here_old
= mddev
->reshape_position
;
4057 sector_div(here_old
, (mddev
->chunk_size
>>9)*
4058 (old_disks
-max_degraded
));
4059 /* here_old is the first stripe that we might need to read
4061 if (here_new
>= here_old
) {
4062 /* Reading from the same stripe as writing to - bad */
4063 printk(KERN_ERR
"raid5: reshape_position too early for "
4064 "auto-recovery - aborting.\n");
4067 printk(KERN_INFO
"raid5: reshape will continue\n");
4068 /* OK, we should be able to continue; */
4072 mddev
->private = kzalloc(sizeof (raid5_conf_t
), GFP_KERNEL
);
4073 if ((conf
= mddev
->private) == NULL
)
4075 if (mddev
->reshape_position
== MaxSector
) {
4076 conf
->previous_raid_disks
= conf
->raid_disks
= mddev
->raid_disks
;
4078 conf
->raid_disks
= mddev
->raid_disks
;
4079 conf
->previous_raid_disks
= mddev
->raid_disks
- mddev
->delta_disks
;
4082 conf
->disks
= kzalloc(conf
->raid_disks
* sizeof(struct disk_info
),
4087 conf
->mddev
= mddev
;
4089 if ((conf
->stripe_hashtbl
= kzalloc(PAGE_SIZE
, GFP_KERNEL
)) == NULL
)
4092 if (mddev
->level
== 6) {
4093 conf
->spare_page
= alloc_page(GFP_KERNEL
);
4094 if (!conf
->spare_page
)
4097 spin_lock_init(&conf
->device_lock
);
4098 mddev
->queue
->queue_lock
= &conf
->device_lock
;
4099 init_waitqueue_head(&conf
->wait_for_stripe
);
4100 init_waitqueue_head(&conf
->wait_for_overlap
);
4101 INIT_LIST_HEAD(&conf
->handle_list
);
4102 INIT_LIST_HEAD(&conf
->hold_list
);
4103 INIT_LIST_HEAD(&conf
->delayed_list
);
4104 INIT_LIST_HEAD(&conf
->bitmap_list
);
4105 INIT_LIST_HEAD(&conf
->inactive_list
);
4106 atomic_set(&conf
->active_stripes
, 0);
4107 atomic_set(&conf
->preread_active_stripes
, 0);
4108 atomic_set(&conf
->active_aligned_reads
, 0);
4109 conf
->bypass_threshold
= BYPASS_THRESHOLD
;
4111 pr_debug("raid5: run(%s) called.\n", mdname(mddev
));
4113 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4114 raid_disk
= rdev
->raid_disk
;
4115 if (raid_disk
>= conf
->raid_disks
4118 disk
= conf
->disks
+ raid_disk
;
4122 if (test_bit(In_sync
, &rdev
->flags
)) {
4123 char b
[BDEVNAME_SIZE
];
4124 printk(KERN_INFO
"raid5: device %s operational as raid"
4125 " disk %d\n", bdevname(rdev
->bdev
,b
),
4129 /* Cannot rely on bitmap to complete recovery */
4134 * 0 for a fully functional array, 1 or 2 for a degraded array.
4136 mddev
->degraded
= conf
->raid_disks
- working_disks
;
4137 conf
->mddev
= mddev
;
4138 conf
->chunk_size
= mddev
->chunk_size
;
4139 conf
->level
= mddev
->level
;
4140 if (conf
->level
== 6)
4141 conf
->max_degraded
= 2;
4143 conf
->max_degraded
= 1;
4144 conf
->algorithm
= mddev
->layout
;
4145 conf
->max_nr_stripes
= NR_STRIPES
;
4146 conf
->expand_progress
= mddev
->reshape_position
;
4148 /* device size must be a multiple of chunk size */
4149 mddev
->size
&= ~(mddev
->chunk_size
/1024 -1);
4150 mddev
->resync_max_sectors
= mddev
->size
<< 1;
4152 if (conf
->level
== 6 && conf
->raid_disks
< 4) {
4153 printk(KERN_ERR
"raid6: not enough configured devices for %s (%d, minimum 4)\n",
4154 mdname(mddev
), conf
->raid_disks
);
4157 if (!conf
->chunk_size
|| conf
->chunk_size
% 4) {
4158 printk(KERN_ERR
"raid5: invalid chunk size %d for %s\n",
4159 conf
->chunk_size
, mdname(mddev
));
4162 if (conf
->algorithm
> ALGORITHM_RIGHT_SYMMETRIC
) {
4164 "raid5: unsupported parity algorithm %d for %s\n",
4165 conf
->algorithm
, mdname(mddev
));
4168 if (mddev
->degraded
> conf
->max_degraded
) {
4169 printk(KERN_ERR
"raid5: not enough operational devices for %s"
4170 " (%d/%d failed)\n",
4171 mdname(mddev
), mddev
->degraded
, conf
->raid_disks
);
4175 if (mddev
->degraded
> 0 &&
4176 mddev
->recovery_cp
!= MaxSector
) {
4177 if (mddev
->ok_start_degraded
)
4179 "raid5: starting dirty degraded array: %s"
4180 "- data corruption possible.\n",
4184 "raid5: cannot start dirty degraded array for %s\n",
4191 mddev
->thread
= md_register_thread(raid5d
, mddev
, "%s_raid5");
4192 if (!mddev
->thread
) {
4194 "raid5: couldn't allocate thread for %s\n",
4199 memory
= conf
->max_nr_stripes
* (sizeof(struct stripe_head
) +
4200 conf
->raid_disks
* ((sizeof(struct bio
) + PAGE_SIZE
))) / 1024;
4201 if (grow_stripes(conf
, conf
->max_nr_stripes
)) {
4203 "raid5: couldn't allocate %dkB for buffers\n", memory
);
4204 shrink_stripes(conf
);
4205 md_unregister_thread(mddev
->thread
);
4208 printk(KERN_INFO
"raid5: allocated %dkB for %s\n",
4209 memory
, mdname(mddev
));
4211 if (mddev
->degraded
== 0)
4212 printk("raid5: raid level %d set %s active with %d out of %d"
4213 " devices, algorithm %d\n", conf
->level
, mdname(mddev
),
4214 mddev
->raid_disks
-mddev
->degraded
, mddev
->raid_disks
,
4217 printk(KERN_ALERT
"raid5: raid level %d set %s active with %d"
4218 " out of %d devices, algorithm %d\n", conf
->level
,
4219 mdname(mddev
), mddev
->raid_disks
- mddev
->degraded
,
4220 mddev
->raid_disks
, conf
->algorithm
);
4222 print_raid5_conf(conf
);
4224 if (conf
->expand_progress
!= MaxSector
) {
4225 printk("...ok start reshape thread\n");
4226 conf
->expand_lo
= conf
->expand_progress
;
4227 atomic_set(&conf
->reshape_stripes
, 0);
4228 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4229 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4230 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
4231 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
4232 mddev
->sync_thread
= md_register_thread(md_do_sync
, mddev
,
4236 /* read-ahead size must cover two whole stripes, which is
4237 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
4240 int data_disks
= conf
->previous_raid_disks
- conf
->max_degraded
;
4241 int stripe
= data_disks
*
4242 (mddev
->chunk_size
/ PAGE_SIZE
);
4243 if (mddev
->queue
->backing_dev_info
.ra_pages
< 2 * stripe
)
4244 mddev
->queue
->backing_dev_info
.ra_pages
= 2 * stripe
;
4247 /* Ok, everything is just fine now */
4248 if (sysfs_create_group(&mddev
->kobj
, &raid5_attrs_group
))
4250 "raid5: failed to create sysfs attributes for %s\n",
4253 mddev
->queue
->unplug_fn
= raid5_unplug_device
;
4254 mddev
->queue
->backing_dev_info
.congested_data
= mddev
;
4255 mddev
->queue
->backing_dev_info
.congested_fn
= raid5_congested
;
4257 mddev
->array_sectors
= 2 * mddev
->size
* (conf
->previous_raid_disks
-
4258 conf
->max_degraded
);
4260 blk_queue_merge_bvec(mddev
->queue
, raid5_mergeable_bvec
);
4265 print_raid5_conf(conf
);
4266 safe_put_page(conf
->spare_page
);
4268 kfree(conf
->stripe_hashtbl
);
4271 mddev
->private = NULL
;
4272 printk(KERN_ALERT
"raid5: failed to run raid set %s\n", mdname(mddev
));
4278 static int stop(mddev_t
*mddev
)
4280 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
4282 md_unregister_thread(mddev
->thread
);
4283 mddev
->thread
= NULL
;
4284 shrink_stripes(conf
);
4285 kfree(conf
->stripe_hashtbl
);
4286 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
4287 blk_sync_queue(mddev
->queue
); /* the unplug fn references 'conf'*/
4288 sysfs_remove_group(&mddev
->kobj
, &raid5_attrs_group
);
4291 mddev
->private = NULL
;
4296 static void print_sh(struct seq_file
*seq
, struct stripe_head
*sh
)
4300 seq_printf(seq
, "sh %llu, pd_idx %d, state %ld.\n",
4301 (unsigned long long)sh
->sector
, sh
->pd_idx
, sh
->state
);
4302 seq_printf(seq
, "sh %llu, count %d.\n",
4303 (unsigned long long)sh
->sector
, atomic_read(&sh
->count
));
4304 seq_printf(seq
, "sh %llu, ", (unsigned long long)sh
->sector
);
4305 for (i
= 0; i
< sh
->disks
; i
++) {
4306 seq_printf(seq
, "(cache%d: %p %ld) ",
4307 i
, sh
->dev
[i
].page
, sh
->dev
[i
].flags
);
4309 seq_printf(seq
, "\n");
4312 static void printall(struct seq_file
*seq
, raid5_conf_t
*conf
)
4314 struct stripe_head
*sh
;
4315 struct hlist_node
*hn
;
4318 spin_lock_irq(&conf
->device_lock
);
4319 for (i
= 0; i
< NR_HASH
; i
++) {
4320 hlist_for_each_entry(sh
, hn
, &conf
->stripe_hashtbl
[i
], hash
) {
4321 if (sh
->raid_conf
!= conf
)
4326 spin_unlock_irq(&conf
->device_lock
);
4330 static void status(struct seq_file
*seq
, mddev_t
*mddev
)
4332 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
4335 seq_printf (seq
, " level %d, %dk chunk, algorithm %d", mddev
->level
, mddev
->chunk_size
>> 10, mddev
->layout
);
4336 seq_printf (seq
, " [%d/%d] [", conf
->raid_disks
, conf
->raid_disks
- mddev
->degraded
);
4337 for (i
= 0; i
< conf
->raid_disks
; i
++)
4338 seq_printf (seq
, "%s",
4339 conf
->disks
[i
].rdev
&&
4340 test_bit(In_sync
, &conf
->disks
[i
].rdev
->flags
) ? "U" : "_");
4341 seq_printf (seq
, "]");
4343 seq_printf (seq
, "\n");
4344 printall(seq
, conf
);
4348 static void print_raid5_conf (raid5_conf_t
*conf
)
4351 struct disk_info
*tmp
;
4353 printk("RAID5 conf printout:\n");
4355 printk("(conf==NULL)\n");
4358 printk(" --- rd:%d wd:%d\n", conf
->raid_disks
,
4359 conf
->raid_disks
- conf
->mddev
->degraded
);
4361 for (i
= 0; i
< conf
->raid_disks
; i
++) {
4362 char b
[BDEVNAME_SIZE
];
4363 tmp
= conf
->disks
+ i
;
4365 printk(" disk %d, o:%d, dev:%s\n",
4366 i
, !test_bit(Faulty
, &tmp
->rdev
->flags
),
4367 bdevname(tmp
->rdev
->bdev
,b
));
4371 static int raid5_spare_active(mddev_t
*mddev
)
4374 raid5_conf_t
*conf
= mddev
->private;
4375 struct disk_info
*tmp
;
4377 for (i
= 0; i
< conf
->raid_disks
; i
++) {
4378 tmp
= conf
->disks
+ i
;
4380 && !test_bit(Faulty
, &tmp
->rdev
->flags
)
4381 && !test_and_set_bit(In_sync
, &tmp
->rdev
->flags
)) {
4382 unsigned long flags
;
4383 spin_lock_irqsave(&conf
->device_lock
, flags
);
4385 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
4388 print_raid5_conf(conf
);
4392 static int raid5_remove_disk(mddev_t
*mddev
, int number
)
4394 raid5_conf_t
*conf
= mddev
->private;
4397 struct disk_info
*p
= conf
->disks
+ number
;
4399 print_raid5_conf(conf
);
4402 if (test_bit(In_sync
, &rdev
->flags
) ||
4403 atomic_read(&rdev
->nr_pending
)) {
4407 /* Only remove non-faulty devices if recovery
4410 if (!test_bit(Faulty
, &rdev
->flags
) &&
4411 mddev
->degraded
<= conf
->max_degraded
) {
4417 if (atomic_read(&rdev
->nr_pending
)) {
4418 /* lost the race, try later */
4425 print_raid5_conf(conf
);
4429 static int raid5_add_disk(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
4431 raid5_conf_t
*conf
= mddev
->private;
4434 struct disk_info
*p
;
4436 int last
= conf
->raid_disks
- 1;
4438 if (mddev
->degraded
> conf
->max_degraded
)
4439 /* no point adding a device */
4442 if (rdev
->raid_disk
>= 0)
4443 first
= last
= rdev
->raid_disk
;
4446 * find the disk ... but prefer rdev->saved_raid_disk
4449 if (rdev
->saved_raid_disk
>= 0 &&
4450 rdev
->saved_raid_disk
>= first
&&
4451 conf
->disks
[rdev
->saved_raid_disk
].rdev
== NULL
)
4452 disk
= rdev
->saved_raid_disk
;
4455 for ( ; disk
<= last
; disk
++)
4456 if ((p
=conf
->disks
+ disk
)->rdev
== NULL
) {
4457 clear_bit(In_sync
, &rdev
->flags
);
4458 rdev
->raid_disk
= disk
;
4460 if (rdev
->saved_raid_disk
!= disk
)
4462 rcu_assign_pointer(p
->rdev
, rdev
);
4465 print_raid5_conf(conf
);
4469 static int raid5_resize(mddev_t
*mddev
, sector_t sectors
)
4471 /* no resync is happening, and there is enough space
4472 * on all devices, so we can resize.
4473 * We need to make sure resync covers any new space.
4474 * If the array is shrinking we should possibly wait until
4475 * any io in the removed space completes, but it hardly seems
4478 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4480 sectors
&= ~((sector_t
)mddev
->chunk_size
/512 - 1);
4481 mddev
->array_sectors
= sectors
* (mddev
->raid_disks
4482 - conf
->max_degraded
);
4483 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4485 if (sectors
/2 > mddev
->size
&& mddev
->recovery_cp
== MaxSector
) {
4486 mddev
->recovery_cp
= mddev
->size
<< 1;
4487 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4489 mddev
->size
= sectors
/2;
4490 mddev
->resync_max_sectors
= sectors
;
4494 #ifdef CONFIG_MD_RAID5_RESHAPE
4495 static int raid5_check_reshape(mddev_t
*mddev
)
4497 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4500 if (mddev
->delta_disks
< 0 ||
4501 mddev
->new_level
!= mddev
->level
)
4502 return -EINVAL
; /* Cannot shrink array or change level yet */
4503 if (mddev
->delta_disks
== 0)
4504 return 0; /* nothing to do */
4506 /* Cannot grow a bitmap yet */
4509 /* Can only proceed if there are plenty of stripe_heads.
4510 * We need a minimum of one full stripe,, and for sensible progress
4511 * it is best to have about 4 times that.
4512 * If we require 4 times, then the default 256 4K stripe_heads will
4513 * allow for chunk sizes up to 256K, which is probably OK.
4514 * If the chunk size is greater, user-space should request more
4515 * stripe_heads first.
4517 if ((mddev
->chunk_size
/ STRIPE_SIZE
) * 4 > conf
->max_nr_stripes
||
4518 (mddev
->new_chunk
/ STRIPE_SIZE
) * 4 > conf
->max_nr_stripes
) {
4519 printk(KERN_WARNING
"raid5: reshape: not enough stripes. Needed %lu\n",
4520 (mddev
->chunk_size
/ STRIPE_SIZE
)*4);
4524 err
= resize_stripes(conf
, conf
->raid_disks
+ mddev
->delta_disks
);
4528 if (mddev
->degraded
> conf
->max_degraded
)
4530 /* looks like we might be able to manage this */
4534 static int raid5_start_reshape(mddev_t
*mddev
)
4536 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4539 int added_devices
= 0;
4540 unsigned long flags
;
4542 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4545 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4546 if (rdev
->raid_disk
< 0 &&
4547 !test_bit(Faulty
, &rdev
->flags
))
4550 if (spares
- mddev
->degraded
< mddev
->delta_disks
- conf
->max_degraded
)
4551 /* Not enough devices even to make a degraded array
4556 atomic_set(&conf
->reshape_stripes
, 0);
4557 spin_lock_irq(&conf
->device_lock
);
4558 conf
->previous_raid_disks
= conf
->raid_disks
;
4559 conf
->raid_disks
+= mddev
->delta_disks
;
4560 conf
->expand_progress
= 0;
4561 conf
->expand_lo
= 0;
4562 spin_unlock_irq(&conf
->device_lock
);
4564 /* Add some new drives, as many as will fit.
4565 * We know there are enough to make the newly sized array work.
4567 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4568 if (rdev
->raid_disk
< 0 &&
4569 !test_bit(Faulty
, &rdev
->flags
)) {
4570 if (raid5_add_disk(mddev
, rdev
) == 0) {
4572 set_bit(In_sync
, &rdev
->flags
);
4574 rdev
->recovery_offset
= 0;
4575 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4576 if (sysfs_create_link(&mddev
->kobj
,
4579 "raid5: failed to create "
4580 " link %s for %s\n",
4586 spin_lock_irqsave(&conf
->device_lock
, flags
);
4587 mddev
->degraded
= (conf
->raid_disks
- conf
->previous_raid_disks
) - added_devices
;
4588 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
4589 mddev
->raid_disks
= conf
->raid_disks
;
4590 mddev
->reshape_position
= 0;
4591 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
4593 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4594 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4595 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
4596 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
4597 mddev
->sync_thread
= md_register_thread(md_do_sync
, mddev
,
4599 if (!mddev
->sync_thread
) {
4600 mddev
->recovery
= 0;
4601 spin_lock_irq(&conf
->device_lock
);
4602 mddev
->raid_disks
= conf
->raid_disks
= conf
->previous_raid_disks
;
4603 conf
->expand_progress
= MaxSector
;
4604 spin_unlock_irq(&conf
->device_lock
);
4607 md_wakeup_thread(mddev
->sync_thread
);
4608 md_new_event(mddev
);
4613 static void end_reshape(raid5_conf_t
*conf
)
4615 struct block_device
*bdev
;
4617 if (!test_bit(MD_RECOVERY_INTR
, &conf
->mddev
->recovery
)) {
4618 conf
->mddev
->array_sectors
= 2 * conf
->mddev
->size
*
4619 (conf
->raid_disks
- conf
->max_degraded
);
4620 set_capacity(conf
->mddev
->gendisk
, conf
->mddev
->array_sectors
);
4621 conf
->mddev
->changed
= 1;
4623 bdev
= bdget_disk(conf
->mddev
->gendisk
, 0);
4625 mutex_lock(&bdev
->bd_inode
->i_mutex
);
4626 i_size_write(bdev
->bd_inode
,
4627 (loff_t
)conf
->mddev
->array_sectors
<< 9);
4628 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
4631 spin_lock_irq(&conf
->device_lock
);
4632 conf
->expand_progress
= MaxSector
;
4633 spin_unlock_irq(&conf
->device_lock
);
4634 conf
->mddev
->reshape_position
= MaxSector
;
4636 /* read-ahead size must cover two whole stripes, which is
4637 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
4640 int data_disks
= conf
->previous_raid_disks
- conf
->max_degraded
;
4641 int stripe
= data_disks
*
4642 (conf
->mddev
->chunk_size
/ PAGE_SIZE
);
4643 if (conf
->mddev
->queue
->backing_dev_info
.ra_pages
< 2 * stripe
)
4644 conf
->mddev
->queue
->backing_dev_info
.ra_pages
= 2 * stripe
;
4649 static void raid5_quiesce(mddev_t
*mddev
, int state
)
4651 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4654 case 2: /* resume for a suspend */
4655 wake_up(&conf
->wait_for_overlap
);
4658 case 1: /* stop all writes */
4659 spin_lock_irq(&conf
->device_lock
);
4661 wait_event_lock_irq(conf
->wait_for_stripe
,
4662 atomic_read(&conf
->active_stripes
) == 0 &&
4663 atomic_read(&conf
->active_aligned_reads
) == 0,
4664 conf
->device_lock
, /* nothing */);
4665 spin_unlock_irq(&conf
->device_lock
);
4668 case 0: /* re-enable writes */
4669 spin_lock_irq(&conf
->device_lock
);
4671 wake_up(&conf
->wait_for_stripe
);
4672 wake_up(&conf
->wait_for_overlap
);
4673 spin_unlock_irq(&conf
->device_lock
);
4678 static struct mdk_personality raid6_personality
=
4682 .owner
= THIS_MODULE
,
4683 .make_request
= make_request
,
4687 .error_handler
= error
,
4688 .hot_add_disk
= raid5_add_disk
,
4689 .hot_remove_disk
= raid5_remove_disk
,
4690 .spare_active
= raid5_spare_active
,
4691 .sync_request
= sync_request
,
4692 .resize
= raid5_resize
,
4693 #ifdef CONFIG_MD_RAID5_RESHAPE
4694 .check_reshape
= raid5_check_reshape
,
4695 .start_reshape
= raid5_start_reshape
,
4697 .quiesce
= raid5_quiesce
,
4699 static struct mdk_personality raid5_personality
=
4703 .owner
= THIS_MODULE
,
4704 .make_request
= make_request
,
4708 .error_handler
= error
,
4709 .hot_add_disk
= raid5_add_disk
,
4710 .hot_remove_disk
= raid5_remove_disk
,
4711 .spare_active
= raid5_spare_active
,
4712 .sync_request
= sync_request
,
4713 .resize
= raid5_resize
,
4714 #ifdef CONFIG_MD_RAID5_RESHAPE
4715 .check_reshape
= raid5_check_reshape
,
4716 .start_reshape
= raid5_start_reshape
,
4718 .quiesce
= raid5_quiesce
,
4721 static struct mdk_personality raid4_personality
=
4725 .owner
= THIS_MODULE
,
4726 .make_request
= make_request
,
4730 .error_handler
= error
,
4731 .hot_add_disk
= raid5_add_disk
,
4732 .hot_remove_disk
= raid5_remove_disk
,
4733 .spare_active
= raid5_spare_active
,
4734 .sync_request
= sync_request
,
4735 .resize
= raid5_resize
,
4736 #ifdef CONFIG_MD_RAID5_RESHAPE
4737 .check_reshape
= raid5_check_reshape
,
4738 .start_reshape
= raid5_start_reshape
,
4740 .quiesce
= raid5_quiesce
,
4743 static int __init
raid5_init(void)
4747 e
= raid6_select_algo();
4750 register_md_personality(&raid6_personality
);
4751 register_md_personality(&raid5_personality
);
4752 register_md_personality(&raid4_personality
);
4756 static void raid5_exit(void)
4758 unregister_md_personality(&raid6_personality
);
4759 unregister_md_personality(&raid5_personality
);
4760 unregister_md_personality(&raid4_personality
);
4763 module_init(raid5_init
);
4764 module_exit(raid5_exit
);
4765 MODULE_LICENSE("GPL");
4766 MODULE_ALIAS("md-personality-4"); /* RAID5 */
4767 MODULE_ALIAS("md-raid5");
4768 MODULE_ALIAS("md-raid4");
4769 MODULE_ALIAS("md-level-5");
4770 MODULE_ALIAS("md-level-4");
4771 MODULE_ALIAS("md-personality-8"); /* RAID6 */
4772 MODULE_ALIAS("md-raid6");
4773 MODULE_ALIAS("md-level-6");
4775 /* This used to be two separate modules, they were: */
4776 MODULE_ALIAS("raid5");
4777 MODULE_ALIAS("raid6");