2 * raid5.c : Multiple Devices driver for Linux
3 * Copyright (C) 1996, 1997 Ingo Molnar, Miguel de Icaza, Gadi Oxman
4 * Copyright (C) 1999, 2000 Ingo Molnar
5 * Copyright (C) 2002, 2003 H. Peter Anvin
7 * RAID-4/5/6 management functions.
8 * Thanks to Penguin Computing for making the RAID-6 development possible
9 * by donating a test server!
11 * This program is free software; you can redistribute it and/or modify
12 * it under the terms of the GNU General Public License as published by
13 * the Free Software Foundation; either version 2, or (at your option)
16 * You should have received a copy of the GNU General Public License
17 * (for example /usr/src/linux/COPYING); if not, write to the Free
18 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 * The sequencing for updating the bitmap reliably is a little
25 * subtle (and I got it wrong the first time) so it deserves some
28 * We group bitmap updates into batches. Each batch has a number.
29 * We may write out several batches at once, but that isn't very important.
30 * conf->bm_write is the number of the last batch successfully written.
31 * conf->bm_flush is the number of the last batch that was closed to
33 * When we discover that we will need to write to any block in a stripe
34 * (in add_stripe_bio) we update the in-memory bitmap and record in sh->bm_seq
35 * the number of the batch it will be in. This is bm_flush+1.
36 * When we are ready to do a write, if that batch hasn't been written yet,
37 * we plug the array and queue the stripe for later.
38 * When an unplug happens, we increment bm_flush, thus closing the current
40 * When we notice that bm_flush > bm_write, we write out all pending updates
41 * to the bitmap, and advance bm_write to where bm_flush was.
42 * This may occasionally write a bit out twice, but is sure never to
46 #include <linux/module.h>
47 #include <linux/slab.h>
48 #include <linux/highmem.h>
49 #include <linux/bitops.h>
50 #include <linux/kthread.h>
51 #include <asm/atomic.h>
54 #include <linux/raid/bitmap.h>
55 #include <linux/async_tx.h>
61 #define NR_STRIPES 256
62 #define STRIPE_SIZE PAGE_SIZE
63 #define STRIPE_SHIFT (PAGE_SHIFT - 9)
64 #define STRIPE_SECTORS (STRIPE_SIZE>>9)
65 #define IO_THRESHOLD 1
66 #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
67 #define HASH_MASK (NR_HASH - 1)
69 #define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK]))
71 /* bio's attached to a stripe+device for I/O are linked together in bi_sector
72 * order without overlap. There may be several bio's per stripe+device, and
73 * a bio could span several devices.
74 * When walking this list for a particular stripe+device, we must never proceed
75 * beyond a bio that extends past this device, as the next bio might no longer
77 * This macro is used to determine the 'next' bio in the list, given the sector
78 * of the current stripe+device
80 #define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL)
82 * The following can be used to debug the driver
84 #define RAID5_PARANOIA 1
85 #if RAID5_PARANOIA && defined(CONFIG_SMP)
86 # define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock)
88 # define CHECK_DEVLOCK()
96 #if !RAID6_USE_EMPTY_ZERO_PAGE
97 /* In .bss so it's zeroed */
98 const char raid6_empty_zero_page
[PAGE_SIZE
] __attribute__((aligned(256)));
101 static inline int raid6_next_disk(int disk
, int raid_disks
)
104 return (disk
< raid_disks
) ? disk
: 0;
107 static void return_io(struct bio
*return_bi
)
109 struct bio
*bi
= return_bi
;
112 return_bi
= bi
->bi_next
;
116 test_bit(BIO_UPTODATE
, &bi
->bi_flags
)
122 static void print_raid5_conf (raid5_conf_t
*conf
);
124 static void __release_stripe(raid5_conf_t
*conf
, struct stripe_head
*sh
)
126 if (atomic_dec_and_test(&sh
->count
)) {
127 BUG_ON(!list_empty(&sh
->lru
));
128 BUG_ON(atomic_read(&conf
->active_stripes
)==0);
129 if (test_bit(STRIPE_HANDLE
, &sh
->state
)) {
130 if (test_bit(STRIPE_DELAYED
, &sh
->state
)) {
131 list_add_tail(&sh
->lru
, &conf
->delayed_list
);
132 blk_plug_device(conf
->mddev
->queue
);
133 } else if (test_bit(STRIPE_BIT_DELAY
, &sh
->state
) &&
134 sh
->bm_seq
- conf
->seq_write
> 0) {
135 list_add_tail(&sh
->lru
, &conf
->bitmap_list
);
136 blk_plug_device(conf
->mddev
->queue
);
138 clear_bit(STRIPE_BIT_DELAY
, &sh
->state
);
139 list_add_tail(&sh
->lru
, &conf
->handle_list
);
141 md_wakeup_thread(conf
->mddev
->thread
);
143 BUG_ON(sh
->ops
.pending
);
144 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
145 atomic_dec(&conf
->preread_active_stripes
);
146 if (atomic_read(&conf
->preread_active_stripes
) < IO_THRESHOLD
)
147 md_wakeup_thread(conf
->mddev
->thread
);
149 atomic_dec(&conf
->active_stripes
);
150 if (!test_bit(STRIPE_EXPANDING
, &sh
->state
)) {
151 list_add_tail(&sh
->lru
, &conf
->inactive_list
);
152 wake_up(&conf
->wait_for_stripe
);
153 if (conf
->retry_read_aligned
)
154 md_wakeup_thread(conf
->mddev
->thread
);
159 static void release_stripe(struct stripe_head
*sh
)
161 raid5_conf_t
*conf
= sh
->raid_conf
;
164 spin_lock_irqsave(&conf
->device_lock
, flags
);
165 __release_stripe(conf
, sh
);
166 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
169 static inline void remove_hash(struct stripe_head
*sh
)
171 pr_debug("remove_hash(), stripe %llu\n",
172 (unsigned long long)sh
->sector
);
174 hlist_del_init(&sh
->hash
);
177 static inline void insert_hash(raid5_conf_t
*conf
, struct stripe_head
*sh
)
179 struct hlist_head
*hp
= stripe_hash(conf
, sh
->sector
);
181 pr_debug("insert_hash(), stripe %llu\n",
182 (unsigned long long)sh
->sector
);
185 hlist_add_head(&sh
->hash
, hp
);
189 /* find an idle stripe, make sure it is unhashed, and return it. */
190 static struct stripe_head
*get_free_stripe(raid5_conf_t
*conf
)
192 struct stripe_head
*sh
= NULL
;
193 struct list_head
*first
;
196 if (list_empty(&conf
->inactive_list
))
198 first
= conf
->inactive_list
.next
;
199 sh
= list_entry(first
, struct stripe_head
, lru
);
200 list_del_init(first
);
202 atomic_inc(&conf
->active_stripes
);
207 static void shrink_buffers(struct stripe_head
*sh
, int num
)
212 for (i
=0; i
<num
; i
++) {
216 sh
->dev
[i
].page
= NULL
;
221 static int grow_buffers(struct stripe_head
*sh
, int num
)
225 for (i
=0; i
<num
; i
++) {
228 if (!(page
= alloc_page(GFP_KERNEL
))) {
231 sh
->dev
[i
].page
= page
;
236 static void raid5_build_block (struct stripe_head
*sh
, int i
);
238 static void init_stripe(struct stripe_head
*sh
, sector_t sector
, int pd_idx
, int disks
)
240 raid5_conf_t
*conf
= sh
->raid_conf
;
243 BUG_ON(atomic_read(&sh
->count
) != 0);
244 BUG_ON(test_bit(STRIPE_HANDLE
, &sh
->state
));
245 BUG_ON(sh
->ops
.pending
|| sh
->ops
.ack
|| sh
->ops
.complete
);
248 pr_debug("init_stripe called, stripe %llu\n",
249 (unsigned long long)sh
->sector
);
259 for (i
= sh
->disks
; i
--; ) {
260 struct r5dev
*dev
= &sh
->dev
[i
];
262 if (dev
->toread
|| dev
->read
|| dev
->towrite
|| dev
->written
||
263 test_bit(R5_LOCKED
, &dev
->flags
)) {
264 printk(KERN_ERR
"sector=%llx i=%d %p %p %p %p %d\n",
265 (unsigned long long)sh
->sector
, i
, dev
->toread
,
266 dev
->read
, dev
->towrite
, dev
->written
,
267 test_bit(R5_LOCKED
, &dev
->flags
));
271 raid5_build_block(sh
, i
);
273 insert_hash(conf
, sh
);
276 static struct stripe_head
*__find_stripe(raid5_conf_t
*conf
, sector_t sector
, int disks
)
278 struct stripe_head
*sh
;
279 struct hlist_node
*hn
;
282 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector
);
283 hlist_for_each_entry(sh
, hn
, stripe_hash(conf
, sector
), hash
)
284 if (sh
->sector
== sector
&& sh
->disks
== disks
)
286 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector
);
290 static void unplug_slaves(mddev_t
*mddev
);
291 static void raid5_unplug_device(struct request_queue
*q
);
293 static struct stripe_head
*get_active_stripe(raid5_conf_t
*conf
, sector_t sector
, int disks
,
294 int pd_idx
, int noblock
)
296 struct stripe_head
*sh
;
298 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector
);
300 spin_lock_irq(&conf
->device_lock
);
303 wait_event_lock_irq(conf
->wait_for_stripe
,
305 conf
->device_lock
, /* nothing */);
306 sh
= __find_stripe(conf
, sector
, disks
);
308 if (!conf
->inactive_blocked
)
309 sh
= get_free_stripe(conf
);
310 if (noblock
&& sh
== NULL
)
313 conf
->inactive_blocked
= 1;
314 wait_event_lock_irq(conf
->wait_for_stripe
,
315 !list_empty(&conf
->inactive_list
) &&
316 (atomic_read(&conf
->active_stripes
)
317 < (conf
->max_nr_stripes
*3/4)
318 || !conf
->inactive_blocked
),
320 raid5_unplug_device(conf
->mddev
->queue
)
322 conf
->inactive_blocked
= 0;
324 init_stripe(sh
, sector
, pd_idx
, disks
);
326 if (atomic_read(&sh
->count
)) {
327 BUG_ON(!list_empty(&sh
->lru
));
329 if (!test_bit(STRIPE_HANDLE
, &sh
->state
))
330 atomic_inc(&conf
->active_stripes
);
331 if (list_empty(&sh
->lru
) &&
332 !test_bit(STRIPE_EXPANDING
, &sh
->state
))
334 list_del_init(&sh
->lru
);
337 } while (sh
== NULL
);
340 atomic_inc(&sh
->count
);
342 spin_unlock_irq(&conf
->device_lock
);
346 /* test_and_ack_op() ensures that we only dequeue an operation once */
347 #define test_and_ack_op(op, pend) \
349 if (test_bit(op, &sh->ops.pending) && \
350 !test_bit(op, &sh->ops.complete)) { \
351 if (test_and_set_bit(op, &sh->ops.ack)) \
352 clear_bit(op, &pend); \
356 clear_bit(op, &pend); \
359 /* find new work to run, do not resubmit work that is already
362 static unsigned long get_stripe_work(struct stripe_head
*sh
)
364 unsigned long pending
;
367 pending
= sh
->ops
.pending
;
369 test_and_ack_op(STRIPE_OP_BIOFILL
, pending
);
370 test_and_ack_op(STRIPE_OP_COMPUTE_BLK
, pending
);
371 test_and_ack_op(STRIPE_OP_PREXOR
, pending
);
372 test_and_ack_op(STRIPE_OP_BIODRAIN
, pending
);
373 test_and_ack_op(STRIPE_OP_POSTXOR
, pending
);
374 test_and_ack_op(STRIPE_OP_CHECK
, pending
);
375 if (test_and_clear_bit(STRIPE_OP_IO
, &sh
->ops
.pending
))
378 sh
->ops
.count
-= ack
;
379 if (unlikely(sh
->ops
.count
< 0)) {
380 printk(KERN_ERR
"pending: %#lx ops.pending: %#lx ops.ack: %#lx "
381 "ops.complete: %#lx\n", pending
, sh
->ops
.pending
,
382 sh
->ops
.ack
, sh
->ops
.complete
);
390 raid5_end_read_request(struct bio
*bi
, int error
);
392 raid5_end_write_request(struct bio
*bi
, int error
);
394 static void ops_run_io(struct stripe_head
*sh
)
396 raid5_conf_t
*conf
= sh
->raid_conf
;
397 int i
, disks
= sh
->disks
;
401 for (i
= disks
; i
--; ) {
405 if (test_and_clear_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
))
407 else if (test_and_clear_bit(R5_Wantread
, &sh
->dev
[i
].flags
))
412 bi
= &sh
->dev
[i
].req
;
416 bi
->bi_end_io
= raid5_end_write_request
;
418 bi
->bi_end_io
= raid5_end_read_request
;
421 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
422 if (rdev
&& test_bit(Faulty
, &rdev
->flags
))
425 atomic_inc(&rdev
->nr_pending
);
429 if (test_bit(STRIPE_SYNCING
, &sh
->state
) ||
430 test_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
) ||
431 test_bit(STRIPE_EXPAND_READY
, &sh
->state
))
432 md_sync_acct(rdev
->bdev
, STRIPE_SECTORS
);
434 bi
->bi_bdev
= rdev
->bdev
;
435 pr_debug("%s: for %llu schedule op %ld on disc %d\n",
436 __FUNCTION__
, (unsigned long long)sh
->sector
,
438 atomic_inc(&sh
->count
);
439 bi
->bi_sector
= sh
->sector
+ rdev
->data_offset
;
440 bi
->bi_flags
= 1 << BIO_UPTODATE
;
444 bi
->bi_io_vec
= &sh
->dev
[i
].vec
;
445 bi
->bi_io_vec
[0].bv_len
= STRIPE_SIZE
;
446 bi
->bi_io_vec
[0].bv_offset
= 0;
447 bi
->bi_size
= STRIPE_SIZE
;
450 test_bit(R5_ReWrite
, &sh
->dev
[i
].flags
))
451 atomic_add(STRIPE_SECTORS
,
452 &rdev
->corrected_errors
);
453 generic_make_request(bi
);
456 set_bit(STRIPE_DEGRADED
, &sh
->state
);
457 pr_debug("skip op %ld on disc %d for sector %llu\n",
458 bi
->bi_rw
, i
, (unsigned long long)sh
->sector
);
459 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
460 set_bit(STRIPE_HANDLE
, &sh
->state
);
465 static struct dma_async_tx_descriptor
*
466 async_copy_data(int frombio
, struct bio
*bio
, struct page
*page
,
467 sector_t sector
, struct dma_async_tx_descriptor
*tx
)
470 struct page
*bio_page
;
474 if (bio
->bi_sector
>= sector
)
475 page_offset
= (signed)(bio
->bi_sector
- sector
) * 512;
477 page_offset
= (signed)(sector
- bio
->bi_sector
) * -512;
478 bio_for_each_segment(bvl
, bio
, i
) {
479 int len
= bio_iovec_idx(bio
, i
)->bv_len
;
483 if (page_offset
< 0) {
484 b_offset
= -page_offset
;
485 page_offset
+= b_offset
;
489 if (len
> 0 && page_offset
+ len
> STRIPE_SIZE
)
490 clen
= STRIPE_SIZE
- page_offset
;
495 b_offset
+= bio_iovec_idx(bio
, i
)->bv_offset
;
496 bio_page
= bio_iovec_idx(bio
, i
)->bv_page
;
498 tx
= async_memcpy(page
, bio_page
, page_offset
,
503 tx
= async_memcpy(bio_page
, page
, b_offset
,
508 if (clen
< len
) /* hit end of page */
516 static void ops_complete_biofill(void *stripe_head_ref
)
518 struct stripe_head
*sh
= stripe_head_ref
;
519 struct bio
*return_bi
= NULL
;
520 raid5_conf_t
*conf
= sh
->raid_conf
;
523 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
524 (unsigned long long)sh
->sector
);
526 /* clear completed biofills */
527 for (i
= sh
->disks
; i
--; ) {
528 struct r5dev
*dev
= &sh
->dev
[i
];
530 /* acknowledge completion of a biofill operation */
531 /* and check if we need to reply to a read request,
532 * new R5_Wantfill requests are held off until
533 * !test_bit(STRIPE_OP_BIOFILL, &sh->ops.pending)
535 if (test_and_clear_bit(R5_Wantfill
, &dev
->flags
)) {
536 struct bio
*rbi
, *rbi2
;
538 /* The access to dev->read is outside of the
539 * spin_lock_irq(&conf->device_lock), but is protected
540 * by the STRIPE_OP_BIOFILL pending bit
545 while (rbi
&& rbi
->bi_sector
<
546 dev
->sector
+ STRIPE_SECTORS
) {
547 rbi2
= r5_next_bio(rbi
, dev
->sector
);
548 spin_lock_irq(&conf
->device_lock
);
549 if (--rbi
->bi_phys_segments
== 0) {
550 rbi
->bi_next
= return_bi
;
553 spin_unlock_irq(&conf
->device_lock
);
558 set_bit(STRIPE_OP_BIOFILL
, &sh
->ops
.complete
);
560 return_io(return_bi
);
562 set_bit(STRIPE_HANDLE
, &sh
->state
);
566 static void ops_run_biofill(struct stripe_head
*sh
)
568 struct dma_async_tx_descriptor
*tx
= NULL
;
569 raid5_conf_t
*conf
= sh
->raid_conf
;
572 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
573 (unsigned long long)sh
->sector
);
575 for (i
= sh
->disks
; i
--; ) {
576 struct r5dev
*dev
= &sh
->dev
[i
];
577 if (test_bit(R5_Wantfill
, &dev
->flags
)) {
579 spin_lock_irq(&conf
->device_lock
);
580 dev
->read
= rbi
= dev
->toread
;
582 spin_unlock_irq(&conf
->device_lock
);
583 while (rbi
&& rbi
->bi_sector
<
584 dev
->sector
+ STRIPE_SECTORS
) {
585 tx
= async_copy_data(0, rbi
, dev
->page
,
587 rbi
= r5_next_bio(rbi
, dev
->sector
);
592 atomic_inc(&sh
->count
);
593 async_trigger_callback(ASYNC_TX_DEP_ACK
| ASYNC_TX_ACK
, tx
,
594 ops_complete_biofill
, sh
);
597 static void ops_complete_compute5(void *stripe_head_ref
)
599 struct stripe_head
*sh
= stripe_head_ref
;
600 int target
= sh
->ops
.target
;
601 struct r5dev
*tgt
= &sh
->dev
[target
];
603 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
604 (unsigned long long)sh
->sector
);
606 set_bit(R5_UPTODATE
, &tgt
->flags
);
607 BUG_ON(!test_bit(R5_Wantcompute
, &tgt
->flags
));
608 clear_bit(R5_Wantcompute
, &tgt
->flags
);
609 set_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.complete
);
610 set_bit(STRIPE_HANDLE
, &sh
->state
);
614 static struct dma_async_tx_descriptor
*
615 ops_run_compute5(struct stripe_head
*sh
, unsigned long pending
)
617 /* kernel stack size limits the total number of disks */
618 int disks
= sh
->disks
;
619 struct page
*xor_srcs
[disks
];
620 int target
= sh
->ops
.target
;
621 struct r5dev
*tgt
= &sh
->dev
[target
];
622 struct page
*xor_dest
= tgt
->page
;
624 struct dma_async_tx_descriptor
*tx
;
627 pr_debug("%s: stripe %llu block: %d\n",
628 __FUNCTION__
, (unsigned long long)sh
->sector
, target
);
629 BUG_ON(!test_bit(R5_Wantcompute
, &tgt
->flags
));
631 for (i
= disks
; i
--; )
633 xor_srcs
[count
++] = sh
->dev
[i
].page
;
635 atomic_inc(&sh
->count
);
637 if (unlikely(count
== 1))
638 tx
= async_memcpy(xor_dest
, xor_srcs
[0], 0, 0, STRIPE_SIZE
,
639 0, NULL
, ops_complete_compute5
, sh
);
641 tx
= async_xor(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
,
642 ASYNC_TX_XOR_ZERO_DST
, NULL
,
643 ops_complete_compute5
, sh
);
645 /* ack now if postxor is not set to be run */
646 if (tx
&& !test_bit(STRIPE_OP_POSTXOR
, &pending
))
652 static void ops_complete_prexor(void *stripe_head_ref
)
654 struct stripe_head
*sh
= stripe_head_ref
;
656 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
657 (unsigned long long)sh
->sector
);
659 set_bit(STRIPE_OP_PREXOR
, &sh
->ops
.complete
);
662 static struct dma_async_tx_descriptor
*
663 ops_run_prexor(struct stripe_head
*sh
, struct dma_async_tx_descriptor
*tx
)
665 /* kernel stack size limits the total number of disks */
666 int disks
= sh
->disks
;
667 struct page
*xor_srcs
[disks
];
668 int count
= 0, pd_idx
= sh
->pd_idx
, i
;
670 /* existing parity data subtracted */
671 struct page
*xor_dest
= xor_srcs
[count
++] = sh
->dev
[pd_idx
].page
;
673 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
674 (unsigned long long)sh
->sector
);
676 for (i
= disks
; i
--; ) {
677 struct r5dev
*dev
= &sh
->dev
[i
];
678 /* Only process blocks that are known to be uptodate */
679 if (dev
->towrite
&& test_bit(R5_Wantprexor
, &dev
->flags
))
680 xor_srcs
[count
++] = dev
->page
;
683 tx
= async_xor(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
,
684 ASYNC_TX_DEP_ACK
| ASYNC_TX_XOR_DROP_DST
, tx
,
685 ops_complete_prexor
, sh
);
690 static struct dma_async_tx_descriptor
*
691 ops_run_biodrain(struct stripe_head
*sh
, struct dma_async_tx_descriptor
*tx
,
692 unsigned long pending
)
694 int disks
= sh
->disks
;
695 int pd_idx
= sh
->pd_idx
, i
;
697 /* check if prexor is active which means only process blocks
698 * that are part of a read-modify-write (Wantprexor)
700 int prexor
= test_bit(STRIPE_OP_PREXOR
, &pending
);
702 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
703 (unsigned long long)sh
->sector
);
705 for (i
= disks
; i
--; ) {
706 struct r5dev
*dev
= &sh
->dev
[i
];
711 if (prexor
) { /* rmw */
713 test_bit(R5_Wantprexor
, &dev
->flags
))
716 if (i
!= pd_idx
&& dev
->towrite
&&
717 test_bit(R5_LOCKED
, &dev
->flags
))
724 spin_lock(&sh
->lock
);
725 chosen
= dev
->towrite
;
727 BUG_ON(dev
->written
);
728 wbi
= dev
->written
= chosen
;
729 spin_unlock(&sh
->lock
);
731 while (wbi
&& wbi
->bi_sector
<
732 dev
->sector
+ STRIPE_SECTORS
) {
733 tx
= async_copy_data(1, wbi
, dev
->page
,
735 wbi
= r5_next_bio(wbi
, dev
->sector
);
743 static void ops_complete_postxor(void *stripe_head_ref
)
745 struct stripe_head
*sh
= stripe_head_ref
;
747 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
748 (unsigned long long)sh
->sector
);
750 set_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.complete
);
751 set_bit(STRIPE_HANDLE
, &sh
->state
);
755 static void ops_complete_write(void *stripe_head_ref
)
757 struct stripe_head
*sh
= stripe_head_ref
;
758 int disks
= sh
->disks
, i
, pd_idx
= sh
->pd_idx
;
760 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
761 (unsigned long long)sh
->sector
);
763 for (i
= disks
; i
--; ) {
764 struct r5dev
*dev
= &sh
->dev
[i
];
765 if (dev
->written
|| i
== pd_idx
)
766 set_bit(R5_UPTODATE
, &dev
->flags
);
769 set_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.complete
);
770 set_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.complete
);
772 set_bit(STRIPE_HANDLE
, &sh
->state
);
777 ops_run_postxor(struct stripe_head
*sh
, struct dma_async_tx_descriptor
*tx
,
778 unsigned long pending
)
780 /* kernel stack size limits the total number of disks */
781 int disks
= sh
->disks
;
782 struct page
*xor_srcs
[disks
];
784 int count
= 0, pd_idx
= sh
->pd_idx
, i
;
785 struct page
*xor_dest
;
786 int prexor
= test_bit(STRIPE_OP_PREXOR
, &pending
);
788 dma_async_tx_callback callback
;
790 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
791 (unsigned long long)sh
->sector
);
793 /* check if prexor is active which means only process blocks
794 * that are part of a read-modify-write (written)
797 xor_dest
= xor_srcs
[count
++] = sh
->dev
[pd_idx
].page
;
798 for (i
= disks
; i
--; ) {
799 struct r5dev
*dev
= &sh
->dev
[i
];
801 xor_srcs
[count
++] = dev
->page
;
804 xor_dest
= sh
->dev
[pd_idx
].page
;
805 for (i
= disks
; i
--; ) {
806 struct r5dev
*dev
= &sh
->dev
[i
];
808 xor_srcs
[count
++] = dev
->page
;
812 /* check whether this postxor is part of a write */
813 callback
= test_bit(STRIPE_OP_BIODRAIN
, &pending
) ?
814 ops_complete_write
: ops_complete_postxor
;
816 /* 1/ if we prexor'd then the dest is reused as a source
817 * 2/ if we did not prexor then we are redoing the parity
818 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
819 * for the synchronous xor case
821 flags
= ASYNC_TX_DEP_ACK
| ASYNC_TX_ACK
|
822 (prexor
? ASYNC_TX_XOR_DROP_DST
: ASYNC_TX_XOR_ZERO_DST
);
824 atomic_inc(&sh
->count
);
826 if (unlikely(count
== 1)) {
827 flags
&= ~(ASYNC_TX_XOR_DROP_DST
| ASYNC_TX_XOR_ZERO_DST
);
828 tx
= async_memcpy(xor_dest
, xor_srcs
[0], 0, 0, STRIPE_SIZE
,
829 flags
, tx
, callback
, sh
);
831 tx
= async_xor(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
,
832 flags
, tx
, callback
, sh
);
835 static void ops_complete_check(void *stripe_head_ref
)
837 struct stripe_head
*sh
= stripe_head_ref
;
838 int pd_idx
= sh
->pd_idx
;
840 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
841 (unsigned long long)sh
->sector
);
843 if (test_and_clear_bit(STRIPE_OP_MOD_DMA_CHECK
, &sh
->ops
.pending
) &&
844 sh
->ops
.zero_sum_result
== 0)
845 set_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
847 set_bit(STRIPE_OP_CHECK
, &sh
->ops
.complete
);
848 set_bit(STRIPE_HANDLE
, &sh
->state
);
852 static void ops_run_check(struct stripe_head
*sh
)
854 /* kernel stack size limits the total number of disks */
855 int disks
= sh
->disks
;
856 struct page
*xor_srcs
[disks
];
857 struct dma_async_tx_descriptor
*tx
;
859 int count
= 0, pd_idx
= sh
->pd_idx
, i
;
860 struct page
*xor_dest
= xor_srcs
[count
++] = sh
->dev
[pd_idx
].page
;
862 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
863 (unsigned long long)sh
->sector
);
865 for (i
= disks
; i
--; ) {
866 struct r5dev
*dev
= &sh
->dev
[i
];
868 xor_srcs
[count
++] = dev
->page
;
871 tx
= async_xor_zero_sum(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
,
872 &sh
->ops
.zero_sum_result
, 0, NULL
, NULL
, NULL
);
875 set_bit(STRIPE_OP_MOD_DMA_CHECK
, &sh
->ops
.pending
);
877 clear_bit(STRIPE_OP_MOD_DMA_CHECK
, &sh
->ops
.pending
);
879 atomic_inc(&sh
->count
);
880 tx
= async_trigger_callback(ASYNC_TX_DEP_ACK
| ASYNC_TX_ACK
, tx
,
881 ops_complete_check
, sh
);
884 static void raid5_run_ops(struct stripe_head
*sh
, unsigned long pending
)
886 int overlap_clear
= 0, i
, disks
= sh
->disks
;
887 struct dma_async_tx_descriptor
*tx
= NULL
;
889 if (test_bit(STRIPE_OP_BIOFILL
, &pending
)) {
894 if (test_bit(STRIPE_OP_COMPUTE_BLK
, &pending
))
895 tx
= ops_run_compute5(sh
, pending
);
897 if (test_bit(STRIPE_OP_PREXOR
, &pending
))
898 tx
= ops_run_prexor(sh
, tx
);
900 if (test_bit(STRIPE_OP_BIODRAIN
, &pending
)) {
901 tx
= ops_run_biodrain(sh
, tx
, pending
);
905 if (test_bit(STRIPE_OP_POSTXOR
, &pending
))
906 ops_run_postxor(sh
, tx
, pending
);
908 if (test_bit(STRIPE_OP_CHECK
, &pending
))
911 if (test_bit(STRIPE_OP_IO
, &pending
))
915 for (i
= disks
; i
--; ) {
916 struct r5dev
*dev
= &sh
->dev
[i
];
917 if (test_and_clear_bit(R5_Overlap
, &dev
->flags
))
918 wake_up(&sh
->raid_conf
->wait_for_overlap
);
922 static int grow_one_stripe(raid5_conf_t
*conf
)
924 struct stripe_head
*sh
;
925 sh
= kmem_cache_alloc(conf
->slab_cache
, GFP_KERNEL
);
928 memset(sh
, 0, sizeof(*sh
) + (conf
->raid_disks
-1)*sizeof(struct r5dev
));
929 sh
->raid_conf
= conf
;
930 spin_lock_init(&sh
->lock
);
932 if (grow_buffers(sh
, conf
->raid_disks
)) {
933 shrink_buffers(sh
, conf
->raid_disks
);
934 kmem_cache_free(conf
->slab_cache
, sh
);
937 sh
->disks
= conf
->raid_disks
;
938 /* we just created an active stripe so... */
939 atomic_set(&sh
->count
, 1);
940 atomic_inc(&conf
->active_stripes
);
941 INIT_LIST_HEAD(&sh
->lru
);
946 static int grow_stripes(raid5_conf_t
*conf
, int num
)
948 struct kmem_cache
*sc
;
949 int devs
= conf
->raid_disks
;
951 sprintf(conf
->cache_name
[0], "raid5-%s", mdname(conf
->mddev
));
952 sprintf(conf
->cache_name
[1], "raid5-%s-alt", mdname(conf
->mddev
));
953 conf
->active_name
= 0;
954 sc
= kmem_cache_create(conf
->cache_name
[conf
->active_name
],
955 sizeof(struct stripe_head
)+(devs
-1)*sizeof(struct r5dev
),
959 conf
->slab_cache
= sc
;
960 conf
->pool_size
= devs
;
962 if (!grow_one_stripe(conf
))
967 #ifdef CONFIG_MD_RAID5_RESHAPE
968 static int resize_stripes(raid5_conf_t
*conf
, int newsize
)
970 /* Make all the stripes able to hold 'newsize' devices.
971 * New slots in each stripe get 'page' set to a new page.
973 * This happens in stages:
974 * 1/ create a new kmem_cache and allocate the required number of
976 * 2/ gather all the old stripe_heads and tranfer the pages across
977 * to the new stripe_heads. This will have the side effect of
978 * freezing the array as once all stripe_heads have been collected,
979 * no IO will be possible. Old stripe heads are freed once their
980 * pages have been transferred over, and the old kmem_cache is
981 * freed when all stripes are done.
982 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
983 * we simple return a failre status - no need to clean anything up.
984 * 4/ allocate new pages for the new slots in the new stripe_heads.
985 * If this fails, we don't bother trying the shrink the
986 * stripe_heads down again, we just leave them as they are.
987 * As each stripe_head is processed the new one is released into
990 * Once step2 is started, we cannot afford to wait for a write,
991 * so we use GFP_NOIO allocations.
993 struct stripe_head
*osh
, *nsh
;
994 LIST_HEAD(newstripes
);
995 struct disk_info
*ndisks
;
997 struct kmem_cache
*sc
;
1000 if (newsize
<= conf
->pool_size
)
1001 return 0; /* never bother to shrink */
1003 md_allow_write(conf
->mddev
);
1006 sc
= kmem_cache_create(conf
->cache_name
[1-conf
->active_name
],
1007 sizeof(struct stripe_head
)+(newsize
-1)*sizeof(struct r5dev
),
1012 for (i
= conf
->max_nr_stripes
; i
; i
--) {
1013 nsh
= kmem_cache_alloc(sc
, GFP_KERNEL
);
1017 memset(nsh
, 0, sizeof(*nsh
) + (newsize
-1)*sizeof(struct r5dev
));
1019 nsh
->raid_conf
= conf
;
1020 spin_lock_init(&nsh
->lock
);
1022 list_add(&nsh
->lru
, &newstripes
);
1025 /* didn't get enough, give up */
1026 while (!list_empty(&newstripes
)) {
1027 nsh
= list_entry(newstripes
.next
, struct stripe_head
, lru
);
1028 list_del(&nsh
->lru
);
1029 kmem_cache_free(sc
, nsh
);
1031 kmem_cache_destroy(sc
);
1034 /* Step 2 - Must use GFP_NOIO now.
1035 * OK, we have enough stripes, start collecting inactive
1036 * stripes and copying them over
1038 list_for_each_entry(nsh
, &newstripes
, lru
) {
1039 spin_lock_irq(&conf
->device_lock
);
1040 wait_event_lock_irq(conf
->wait_for_stripe
,
1041 !list_empty(&conf
->inactive_list
),
1043 unplug_slaves(conf
->mddev
)
1045 osh
= get_free_stripe(conf
);
1046 spin_unlock_irq(&conf
->device_lock
);
1047 atomic_set(&nsh
->count
, 1);
1048 for(i
=0; i
<conf
->pool_size
; i
++)
1049 nsh
->dev
[i
].page
= osh
->dev
[i
].page
;
1050 for( ; i
<newsize
; i
++)
1051 nsh
->dev
[i
].page
= NULL
;
1052 kmem_cache_free(conf
->slab_cache
, osh
);
1054 kmem_cache_destroy(conf
->slab_cache
);
1057 * At this point, we are holding all the stripes so the array
1058 * is completely stalled, so now is a good time to resize
1061 ndisks
= kzalloc(newsize
* sizeof(struct disk_info
), GFP_NOIO
);
1063 for (i
=0; i
<conf
->raid_disks
; i
++)
1064 ndisks
[i
] = conf
->disks
[i
];
1066 conf
->disks
= ndisks
;
1070 /* Step 4, return new stripes to service */
1071 while(!list_empty(&newstripes
)) {
1072 nsh
= list_entry(newstripes
.next
, struct stripe_head
, lru
);
1073 list_del_init(&nsh
->lru
);
1074 for (i
=conf
->raid_disks
; i
< newsize
; i
++)
1075 if (nsh
->dev
[i
].page
== NULL
) {
1076 struct page
*p
= alloc_page(GFP_NOIO
);
1077 nsh
->dev
[i
].page
= p
;
1081 release_stripe(nsh
);
1083 /* critical section pass, GFP_NOIO no longer needed */
1085 conf
->slab_cache
= sc
;
1086 conf
->active_name
= 1-conf
->active_name
;
1087 conf
->pool_size
= newsize
;
1092 static int drop_one_stripe(raid5_conf_t
*conf
)
1094 struct stripe_head
*sh
;
1096 spin_lock_irq(&conf
->device_lock
);
1097 sh
= get_free_stripe(conf
);
1098 spin_unlock_irq(&conf
->device_lock
);
1101 BUG_ON(atomic_read(&sh
->count
));
1102 shrink_buffers(sh
, conf
->pool_size
);
1103 kmem_cache_free(conf
->slab_cache
, sh
);
1104 atomic_dec(&conf
->active_stripes
);
1108 static void shrink_stripes(raid5_conf_t
*conf
)
1110 while (drop_one_stripe(conf
))
1113 if (conf
->slab_cache
)
1114 kmem_cache_destroy(conf
->slab_cache
);
1115 conf
->slab_cache
= NULL
;
1118 static void raid5_end_read_request(struct bio
* bi
, int error
)
1120 struct stripe_head
*sh
= bi
->bi_private
;
1121 raid5_conf_t
*conf
= sh
->raid_conf
;
1122 int disks
= sh
->disks
, i
;
1123 int uptodate
= test_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1124 char b
[BDEVNAME_SIZE
];
1128 for (i
=0 ; i
<disks
; i
++)
1129 if (bi
== &sh
->dev
[i
].req
)
1132 pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n",
1133 (unsigned long long)sh
->sector
, i
, atomic_read(&sh
->count
),
1141 set_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
);
1142 if (test_bit(R5_ReadError
, &sh
->dev
[i
].flags
)) {
1143 rdev
= conf
->disks
[i
].rdev
;
1144 printk(KERN_INFO
"raid5:%s: read error corrected (%lu sectors at %llu on %s)\n",
1145 mdname(conf
->mddev
), STRIPE_SECTORS
,
1146 (unsigned long long)sh
->sector
+ rdev
->data_offset
,
1147 bdevname(rdev
->bdev
, b
));
1148 clear_bit(R5_ReadError
, &sh
->dev
[i
].flags
);
1149 clear_bit(R5_ReWrite
, &sh
->dev
[i
].flags
);
1151 if (atomic_read(&conf
->disks
[i
].rdev
->read_errors
))
1152 atomic_set(&conf
->disks
[i
].rdev
->read_errors
, 0);
1154 const char *bdn
= bdevname(conf
->disks
[i
].rdev
->bdev
, b
);
1156 rdev
= conf
->disks
[i
].rdev
;
1158 clear_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
);
1159 atomic_inc(&rdev
->read_errors
);
1160 if (conf
->mddev
->degraded
)
1161 printk(KERN_WARNING
"raid5:%s: read error not correctable (sector %llu on %s).\n",
1162 mdname(conf
->mddev
),
1163 (unsigned long long)sh
->sector
+ rdev
->data_offset
,
1165 else if (test_bit(R5_ReWrite
, &sh
->dev
[i
].flags
))
1167 printk(KERN_WARNING
"raid5:%s: read error NOT corrected!! (sector %llu on %s).\n",
1168 mdname(conf
->mddev
),
1169 (unsigned long long)sh
->sector
+ rdev
->data_offset
,
1171 else if (atomic_read(&rdev
->read_errors
)
1172 > conf
->max_nr_stripes
)
1174 "raid5:%s: Too many read errors, failing device %s.\n",
1175 mdname(conf
->mddev
), bdn
);
1179 set_bit(R5_ReadError
, &sh
->dev
[i
].flags
);
1181 clear_bit(R5_ReadError
, &sh
->dev
[i
].flags
);
1182 clear_bit(R5_ReWrite
, &sh
->dev
[i
].flags
);
1183 md_error(conf
->mddev
, rdev
);
1186 rdev_dec_pending(conf
->disks
[i
].rdev
, conf
->mddev
);
1187 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
1188 set_bit(STRIPE_HANDLE
, &sh
->state
);
1192 static void raid5_end_write_request (struct bio
*bi
, int error
)
1194 struct stripe_head
*sh
= bi
->bi_private
;
1195 raid5_conf_t
*conf
= sh
->raid_conf
;
1196 int disks
= sh
->disks
, i
;
1197 int uptodate
= test_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1199 for (i
=0 ; i
<disks
; i
++)
1200 if (bi
== &sh
->dev
[i
].req
)
1203 pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n",
1204 (unsigned long long)sh
->sector
, i
, atomic_read(&sh
->count
),
1212 md_error(conf
->mddev
, conf
->disks
[i
].rdev
);
1214 rdev_dec_pending(conf
->disks
[i
].rdev
, conf
->mddev
);
1216 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
1217 set_bit(STRIPE_HANDLE
, &sh
->state
);
1222 static sector_t
compute_blocknr(struct stripe_head
*sh
, int i
);
1224 static void raid5_build_block (struct stripe_head
*sh
, int i
)
1226 struct r5dev
*dev
= &sh
->dev
[i
];
1228 bio_init(&dev
->req
);
1229 dev
->req
.bi_io_vec
= &dev
->vec
;
1231 dev
->req
.bi_max_vecs
++;
1232 dev
->vec
.bv_page
= dev
->page
;
1233 dev
->vec
.bv_len
= STRIPE_SIZE
;
1234 dev
->vec
.bv_offset
= 0;
1236 dev
->req
.bi_sector
= sh
->sector
;
1237 dev
->req
.bi_private
= sh
;
1240 dev
->sector
= compute_blocknr(sh
, i
);
1243 static void error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1245 char b
[BDEVNAME_SIZE
];
1246 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
1247 pr_debug("raid5: error called\n");
1249 if (!test_bit(Faulty
, &rdev
->flags
)) {
1250 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
1251 if (test_and_clear_bit(In_sync
, &rdev
->flags
)) {
1252 unsigned long flags
;
1253 spin_lock_irqsave(&conf
->device_lock
, flags
);
1255 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1257 * if recovery was running, make sure it aborts.
1259 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
1261 set_bit(Faulty
, &rdev
->flags
);
1263 "raid5: Disk failure on %s, disabling device."
1264 " Operation continuing on %d devices\n",
1265 bdevname(rdev
->bdev
,b
), conf
->raid_disks
- mddev
->degraded
);
1270 * Input: a 'big' sector number,
1271 * Output: index of the data and parity disk, and the sector # in them.
1273 static sector_t
raid5_compute_sector(sector_t r_sector
, unsigned int raid_disks
,
1274 unsigned int data_disks
, unsigned int * dd_idx
,
1275 unsigned int * pd_idx
, raid5_conf_t
*conf
)
1278 unsigned long chunk_number
;
1279 unsigned int chunk_offset
;
1280 sector_t new_sector
;
1281 int sectors_per_chunk
= conf
->chunk_size
>> 9;
1283 /* First compute the information on this sector */
1286 * Compute the chunk number and the sector offset inside the chunk
1288 chunk_offset
= sector_div(r_sector
, sectors_per_chunk
);
1289 chunk_number
= r_sector
;
1290 BUG_ON(r_sector
!= chunk_number
);
1293 * Compute the stripe number
1295 stripe
= chunk_number
/ data_disks
;
1298 * Compute the data disk and parity disk indexes inside the stripe
1300 *dd_idx
= chunk_number
% data_disks
;
1303 * Select the parity disk based on the user selected algorithm.
1305 switch(conf
->level
) {
1307 *pd_idx
= data_disks
;
1310 switch (conf
->algorithm
) {
1311 case ALGORITHM_LEFT_ASYMMETRIC
:
1312 *pd_idx
= data_disks
- stripe
% raid_disks
;
1313 if (*dd_idx
>= *pd_idx
)
1316 case ALGORITHM_RIGHT_ASYMMETRIC
:
1317 *pd_idx
= stripe
% raid_disks
;
1318 if (*dd_idx
>= *pd_idx
)
1321 case ALGORITHM_LEFT_SYMMETRIC
:
1322 *pd_idx
= data_disks
- stripe
% raid_disks
;
1323 *dd_idx
= (*pd_idx
+ 1 + *dd_idx
) % raid_disks
;
1325 case ALGORITHM_RIGHT_SYMMETRIC
:
1326 *pd_idx
= stripe
% raid_disks
;
1327 *dd_idx
= (*pd_idx
+ 1 + *dd_idx
) % raid_disks
;
1330 printk(KERN_ERR
"raid5: unsupported algorithm %d\n",
1336 /**** FIX THIS ****/
1337 switch (conf
->algorithm
) {
1338 case ALGORITHM_LEFT_ASYMMETRIC
:
1339 *pd_idx
= raid_disks
- 1 - (stripe
% raid_disks
);
1340 if (*pd_idx
== raid_disks
-1)
1341 (*dd_idx
)++; /* Q D D D P */
1342 else if (*dd_idx
>= *pd_idx
)
1343 (*dd_idx
) += 2; /* D D P Q D */
1345 case ALGORITHM_RIGHT_ASYMMETRIC
:
1346 *pd_idx
= stripe
% raid_disks
;
1347 if (*pd_idx
== raid_disks
-1)
1348 (*dd_idx
)++; /* Q D D D P */
1349 else if (*dd_idx
>= *pd_idx
)
1350 (*dd_idx
) += 2; /* D D P Q D */
1352 case ALGORITHM_LEFT_SYMMETRIC
:
1353 *pd_idx
= raid_disks
- 1 - (stripe
% raid_disks
);
1354 *dd_idx
= (*pd_idx
+ 2 + *dd_idx
) % raid_disks
;
1356 case ALGORITHM_RIGHT_SYMMETRIC
:
1357 *pd_idx
= stripe
% raid_disks
;
1358 *dd_idx
= (*pd_idx
+ 2 + *dd_idx
) % raid_disks
;
1361 printk (KERN_CRIT
"raid6: unsupported algorithm %d\n",
1368 * Finally, compute the new sector number
1370 new_sector
= (sector_t
)stripe
* sectors_per_chunk
+ chunk_offset
;
1375 static sector_t
compute_blocknr(struct stripe_head
*sh
, int i
)
1377 raid5_conf_t
*conf
= sh
->raid_conf
;
1378 int raid_disks
= sh
->disks
;
1379 int data_disks
= raid_disks
- conf
->max_degraded
;
1380 sector_t new_sector
= sh
->sector
, check
;
1381 int sectors_per_chunk
= conf
->chunk_size
>> 9;
1384 int chunk_number
, dummy1
, dummy2
, dd_idx
= i
;
1388 chunk_offset
= sector_div(new_sector
, sectors_per_chunk
);
1389 stripe
= new_sector
;
1390 BUG_ON(new_sector
!= stripe
);
1392 if (i
== sh
->pd_idx
)
1394 switch(conf
->level
) {
1397 switch (conf
->algorithm
) {
1398 case ALGORITHM_LEFT_ASYMMETRIC
:
1399 case ALGORITHM_RIGHT_ASYMMETRIC
:
1403 case ALGORITHM_LEFT_SYMMETRIC
:
1404 case ALGORITHM_RIGHT_SYMMETRIC
:
1407 i
-= (sh
->pd_idx
+ 1);
1410 printk(KERN_ERR
"raid5: unsupported algorithm %d\n",
1415 if (i
== raid6_next_disk(sh
->pd_idx
, raid_disks
))
1416 return 0; /* It is the Q disk */
1417 switch (conf
->algorithm
) {
1418 case ALGORITHM_LEFT_ASYMMETRIC
:
1419 case ALGORITHM_RIGHT_ASYMMETRIC
:
1420 if (sh
->pd_idx
== raid_disks
-1)
1421 i
--; /* Q D D D P */
1422 else if (i
> sh
->pd_idx
)
1423 i
-= 2; /* D D P Q D */
1425 case ALGORITHM_LEFT_SYMMETRIC
:
1426 case ALGORITHM_RIGHT_SYMMETRIC
:
1427 if (sh
->pd_idx
== raid_disks
-1)
1428 i
--; /* Q D D D P */
1433 i
-= (sh
->pd_idx
+ 2);
1437 printk (KERN_CRIT
"raid6: unsupported algorithm %d\n",
1443 chunk_number
= stripe
* data_disks
+ i
;
1444 r_sector
= (sector_t
)chunk_number
* sectors_per_chunk
+ chunk_offset
;
1446 check
= raid5_compute_sector (r_sector
, raid_disks
, data_disks
, &dummy1
, &dummy2
, conf
);
1447 if (check
!= sh
->sector
|| dummy1
!= dd_idx
|| dummy2
!= sh
->pd_idx
) {
1448 printk(KERN_ERR
"compute_blocknr: map not correct\n");
1457 * Copy data between a page in the stripe cache, and one or more bion
1458 * The page could align with the middle of the bio, or there could be
1459 * several bion, each with several bio_vecs, which cover part of the page
1460 * Multiple bion are linked together on bi_next. There may be extras
1461 * at the end of this list. We ignore them.
1463 static void copy_data(int frombio
, struct bio
*bio
,
1467 char *pa
= page_address(page
);
1468 struct bio_vec
*bvl
;
1472 if (bio
->bi_sector
>= sector
)
1473 page_offset
= (signed)(bio
->bi_sector
- sector
) * 512;
1475 page_offset
= (signed)(sector
- bio
->bi_sector
) * -512;
1476 bio_for_each_segment(bvl
, bio
, i
) {
1477 int len
= bio_iovec_idx(bio
,i
)->bv_len
;
1481 if (page_offset
< 0) {
1482 b_offset
= -page_offset
;
1483 page_offset
+= b_offset
;
1487 if (len
> 0 && page_offset
+ len
> STRIPE_SIZE
)
1488 clen
= STRIPE_SIZE
- page_offset
;
1492 char *ba
= __bio_kmap_atomic(bio
, i
, KM_USER0
);
1494 memcpy(pa
+page_offset
, ba
+b_offset
, clen
);
1496 memcpy(ba
+b_offset
, pa
+page_offset
, clen
);
1497 __bio_kunmap_atomic(ba
, KM_USER0
);
1499 if (clen
< len
) /* hit end of page */
1505 #define check_xor() do { \
1506 if (count == MAX_XOR_BLOCKS) { \
1507 xor_blocks(count, STRIPE_SIZE, dest, ptr);\
1512 static void compute_parity6(struct stripe_head
*sh
, int method
)
1514 raid6_conf_t
*conf
= sh
->raid_conf
;
1515 int i
, pd_idx
= sh
->pd_idx
, qd_idx
, d0_idx
, disks
= sh
->disks
, count
;
1517 /**** FIX THIS: This could be very bad if disks is close to 256 ****/
1520 qd_idx
= raid6_next_disk(pd_idx
, disks
);
1521 d0_idx
= raid6_next_disk(qd_idx
, disks
);
1523 pr_debug("compute_parity, stripe %llu, method %d\n",
1524 (unsigned long long)sh
->sector
, method
);
1527 case READ_MODIFY_WRITE
:
1528 BUG(); /* READ_MODIFY_WRITE N/A for RAID-6 */
1529 case RECONSTRUCT_WRITE
:
1530 for (i
= disks
; i
-- ;)
1531 if ( i
!= pd_idx
&& i
!= qd_idx
&& sh
->dev
[i
].towrite
) {
1532 chosen
= sh
->dev
[i
].towrite
;
1533 sh
->dev
[i
].towrite
= NULL
;
1535 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
1536 wake_up(&conf
->wait_for_overlap
);
1538 BUG_ON(sh
->dev
[i
].written
);
1539 sh
->dev
[i
].written
= chosen
;
1543 BUG(); /* Not implemented yet */
1546 for (i
= disks
; i
--;)
1547 if (sh
->dev
[i
].written
) {
1548 sector_t sector
= sh
->dev
[i
].sector
;
1549 struct bio
*wbi
= sh
->dev
[i
].written
;
1550 while (wbi
&& wbi
->bi_sector
< sector
+ STRIPE_SECTORS
) {
1551 copy_data(1, wbi
, sh
->dev
[i
].page
, sector
);
1552 wbi
= r5_next_bio(wbi
, sector
);
1555 set_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
1556 set_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
);
1560 // case RECONSTRUCT_WRITE:
1561 // case CHECK_PARITY:
1562 // case UPDATE_PARITY:
1563 /* Note that unlike RAID-5, the ordering of the disks matters greatly. */
1564 /* FIX: Is this ordering of drives even remotely optimal? */
1568 ptrs
[count
++] = page_address(sh
->dev
[i
].page
);
1569 if (count
<= disks
-2 && !test_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
))
1570 printk("block %d/%d not uptodate on parity calc\n", i
,count
);
1571 i
= raid6_next_disk(i
, disks
);
1572 } while ( i
!= d0_idx
);
1576 raid6_call
.gen_syndrome(disks
, STRIPE_SIZE
, ptrs
);
1579 case RECONSTRUCT_WRITE
:
1580 set_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
1581 set_bit(R5_UPTODATE
, &sh
->dev
[qd_idx
].flags
);
1582 set_bit(R5_LOCKED
, &sh
->dev
[pd_idx
].flags
);
1583 set_bit(R5_LOCKED
, &sh
->dev
[qd_idx
].flags
);
1586 set_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
1587 set_bit(R5_UPTODATE
, &sh
->dev
[qd_idx
].flags
);
1593 /* Compute one missing block */
1594 static void compute_block_1(struct stripe_head
*sh
, int dd_idx
, int nozero
)
1596 int i
, count
, disks
= sh
->disks
;
1597 void *ptr
[MAX_XOR_BLOCKS
], *dest
, *p
;
1598 int pd_idx
= sh
->pd_idx
;
1599 int qd_idx
= raid6_next_disk(pd_idx
, disks
);
1601 pr_debug("compute_block_1, stripe %llu, idx %d\n",
1602 (unsigned long long)sh
->sector
, dd_idx
);
1604 if ( dd_idx
== qd_idx
) {
1605 /* We're actually computing the Q drive */
1606 compute_parity6(sh
, UPDATE_PARITY
);
1608 dest
= page_address(sh
->dev
[dd_idx
].page
);
1609 if (!nozero
) memset(dest
, 0, STRIPE_SIZE
);
1611 for (i
= disks
; i
--; ) {
1612 if (i
== dd_idx
|| i
== qd_idx
)
1614 p
= page_address(sh
->dev
[i
].page
);
1615 if (test_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
))
1618 printk("compute_block() %d, stripe %llu, %d"
1619 " not present\n", dd_idx
,
1620 (unsigned long long)sh
->sector
, i
);
1625 xor_blocks(count
, STRIPE_SIZE
, dest
, ptr
);
1626 if (!nozero
) set_bit(R5_UPTODATE
, &sh
->dev
[dd_idx
].flags
);
1627 else clear_bit(R5_UPTODATE
, &sh
->dev
[dd_idx
].flags
);
1631 /* Compute two missing blocks */
1632 static void compute_block_2(struct stripe_head
*sh
, int dd_idx1
, int dd_idx2
)
1634 int i
, count
, disks
= sh
->disks
;
1635 int pd_idx
= sh
->pd_idx
;
1636 int qd_idx
= raid6_next_disk(pd_idx
, disks
);
1637 int d0_idx
= raid6_next_disk(qd_idx
, disks
);
1640 /* faila and failb are disk numbers relative to d0_idx */
1641 /* pd_idx become disks-2 and qd_idx become disks-1 */
1642 faila
= (dd_idx1
< d0_idx
) ? dd_idx1
+(disks
-d0_idx
) : dd_idx1
-d0_idx
;
1643 failb
= (dd_idx2
< d0_idx
) ? dd_idx2
+(disks
-d0_idx
) : dd_idx2
-d0_idx
;
1645 BUG_ON(faila
== failb
);
1646 if ( failb
< faila
) { int tmp
= faila
; faila
= failb
; failb
= tmp
; }
1648 pr_debug("compute_block_2, stripe %llu, idx %d,%d (%d,%d)\n",
1649 (unsigned long long)sh
->sector
, dd_idx1
, dd_idx2
, faila
, failb
);
1651 if ( failb
== disks
-1 ) {
1652 /* Q disk is one of the missing disks */
1653 if ( faila
== disks
-2 ) {
1654 /* Missing P+Q, just recompute */
1655 compute_parity6(sh
, UPDATE_PARITY
);
1658 /* We're missing D+Q; recompute D from P */
1659 compute_block_1(sh
, (dd_idx1
== qd_idx
) ? dd_idx2
: dd_idx1
, 0);
1660 compute_parity6(sh
, UPDATE_PARITY
); /* Is this necessary? */
1665 /* We're missing D+P or D+D; build pointer table */
1667 /**** FIX THIS: This could be very bad if disks is close to 256 ****/
1673 ptrs
[count
++] = page_address(sh
->dev
[i
].page
);
1674 i
= raid6_next_disk(i
, disks
);
1675 if (i
!= dd_idx1
&& i
!= dd_idx2
&&
1676 !test_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
))
1677 printk("compute_2 with missing block %d/%d\n", count
, i
);
1678 } while ( i
!= d0_idx
);
1680 if ( failb
== disks
-2 ) {
1681 /* We're missing D+P. */
1682 raid6_datap_recov(disks
, STRIPE_SIZE
, faila
, ptrs
);
1684 /* We're missing D+D. */
1685 raid6_2data_recov(disks
, STRIPE_SIZE
, faila
, failb
, ptrs
);
1688 /* Both the above update both missing blocks */
1689 set_bit(R5_UPTODATE
, &sh
->dev
[dd_idx1
].flags
);
1690 set_bit(R5_UPTODATE
, &sh
->dev
[dd_idx2
].flags
);
1695 handle_write_operations5(struct stripe_head
*sh
, int rcw
, int expand
)
1697 int i
, pd_idx
= sh
->pd_idx
, disks
= sh
->disks
;
1701 /* if we are not expanding this is a proper write request, and
1702 * there will be bios with new data to be drained into the
1706 set_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.pending
);
1710 set_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
);
1713 for (i
= disks
; i
--; ) {
1714 struct r5dev
*dev
= &sh
->dev
[i
];
1717 set_bit(R5_LOCKED
, &dev
->flags
);
1719 clear_bit(R5_UPTODATE
, &dev
->flags
);
1724 BUG_ON(!(test_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
) ||
1725 test_bit(R5_Wantcompute
, &sh
->dev
[pd_idx
].flags
)));
1727 set_bit(STRIPE_OP_PREXOR
, &sh
->ops
.pending
);
1728 set_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.pending
);
1729 set_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
);
1733 for (i
= disks
; i
--; ) {
1734 struct r5dev
*dev
= &sh
->dev
[i
];
1738 /* For a read-modify write there may be blocks that are
1739 * locked for reading while others are ready to be
1740 * written so we distinguish these blocks by the
1744 (test_bit(R5_UPTODATE
, &dev
->flags
) ||
1745 test_bit(R5_Wantcompute
, &dev
->flags
))) {
1746 set_bit(R5_Wantprexor
, &dev
->flags
);
1747 set_bit(R5_LOCKED
, &dev
->flags
);
1748 clear_bit(R5_UPTODATE
, &dev
->flags
);
1754 /* keep the parity disk locked while asynchronous operations
1757 set_bit(R5_LOCKED
, &sh
->dev
[pd_idx
].flags
);
1758 clear_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
1761 pr_debug("%s: stripe %llu locked: %d pending: %lx\n",
1762 __FUNCTION__
, (unsigned long long)sh
->sector
,
1763 locked
, sh
->ops
.pending
);
1769 * Each stripe/dev can have one or more bion attached.
1770 * toread/towrite point to the first in a chain.
1771 * The bi_next chain must be in order.
1773 static int add_stripe_bio(struct stripe_head
*sh
, struct bio
*bi
, int dd_idx
, int forwrite
)
1776 raid5_conf_t
*conf
= sh
->raid_conf
;
1779 pr_debug("adding bh b#%llu to stripe s#%llu\n",
1780 (unsigned long long)bi
->bi_sector
,
1781 (unsigned long long)sh
->sector
);
1784 spin_lock(&sh
->lock
);
1785 spin_lock_irq(&conf
->device_lock
);
1787 bip
= &sh
->dev
[dd_idx
].towrite
;
1788 if (*bip
== NULL
&& sh
->dev
[dd_idx
].written
== NULL
)
1791 bip
= &sh
->dev
[dd_idx
].toread
;
1792 while (*bip
&& (*bip
)->bi_sector
< bi
->bi_sector
) {
1793 if ((*bip
)->bi_sector
+ ((*bip
)->bi_size
>> 9) > bi
->bi_sector
)
1795 bip
= & (*bip
)->bi_next
;
1797 if (*bip
&& (*bip
)->bi_sector
< bi
->bi_sector
+ ((bi
->bi_size
)>>9))
1800 BUG_ON(*bip
&& bi
->bi_next
&& (*bip
) != bi
->bi_next
);
1804 bi
->bi_phys_segments
++;
1805 spin_unlock_irq(&conf
->device_lock
);
1806 spin_unlock(&sh
->lock
);
1808 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
1809 (unsigned long long)bi
->bi_sector
,
1810 (unsigned long long)sh
->sector
, dd_idx
);
1812 if (conf
->mddev
->bitmap
&& firstwrite
) {
1813 bitmap_startwrite(conf
->mddev
->bitmap
, sh
->sector
,
1815 sh
->bm_seq
= conf
->seq_flush
+1;
1816 set_bit(STRIPE_BIT_DELAY
, &sh
->state
);
1820 /* check if page is covered */
1821 sector_t sector
= sh
->dev
[dd_idx
].sector
;
1822 for (bi
=sh
->dev
[dd_idx
].towrite
;
1823 sector
< sh
->dev
[dd_idx
].sector
+ STRIPE_SECTORS
&&
1824 bi
&& bi
->bi_sector
<= sector
;
1825 bi
= r5_next_bio(bi
, sh
->dev
[dd_idx
].sector
)) {
1826 if (bi
->bi_sector
+ (bi
->bi_size
>>9) >= sector
)
1827 sector
= bi
->bi_sector
+ (bi
->bi_size
>>9);
1829 if (sector
>= sh
->dev
[dd_idx
].sector
+ STRIPE_SECTORS
)
1830 set_bit(R5_OVERWRITE
, &sh
->dev
[dd_idx
].flags
);
1835 set_bit(R5_Overlap
, &sh
->dev
[dd_idx
].flags
);
1836 spin_unlock_irq(&conf
->device_lock
);
1837 spin_unlock(&sh
->lock
);
1841 static void end_reshape(raid5_conf_t
*conf
);
1843 static int page_is_zero(struct page
*p
)
1845 char *a
= page_address(p
);
1846 return ((*(u32
*)a
) == 0 &&
1847 memcmp(a
, a
+4, STRIPE_SIZE
-4)==0);
1850 static int stripe_to_pdidx(sector_t stripe
, raid5_conf_t
*conf
, int disks
)
1852 int sectors_per_chunk
= conf
->chunk_size
>> 9;
1854 int chunk_offset
= sector_div(stripe
, sectors_per_chunk
);
1856 raid5_compute_sector(stripe
* (disks
- conf
->max_degraded
)
1857 *sectors_per_chunk
+ chunk_offset
,
1858 disks
, disks
- conf
->max_degraded
,
1859 &dd_idx
, &pd_idx
, conf
);
1864 handle_requests_to_failed_array(raid5_conf_t
*conf
, struct stripe_head
*sh
,
1865 struct stripe_head_state
*s
, int disks
,
1866 struct bio
**return_bi
)
1869 for (i
= disks
; i
--; ) {
1873 if (test_bit(R5_ReadError
, &sh
->dev
[i
].flags
)) {
1876 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
1877 if (rdev
&& test_bit(In_sync
, &rdev
->flags
))
1878 /* multiple read failures in one stripe */
1879 md_error(conf
->mddev
, rdev
);
1882 spin_lock_irq(&conf
->device_lock
);
1883 /* fail all writes first */
1884 bi
= sh
->dev
[i
].towrite
;
1885 sh
->dev
[i
].towrite
= NULL
;
1891 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
1892 wake_up(&conf
->wait_for_overlap
);
1894 while (bi
&& bi
->bi_sector
<
1895 sh
->dev
[i
].sector
+ STRIPE_SECTORS
) {
1896 struct bio
*nextbi
= r5_next_bio(bi
, sh
->dev
[i
].sector
);
1897 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1898 if (--bi
->bi_phys_segments
== 0) {
1899 md_write_end(conf
->mddev
);
1900 bi
->bi_next
= *return_bi
;
1905 /* and fail all 'written' */
1906 bi
= sh
->dev
[i
].written
;
1907 sh
->dev
[i
].written
= NULL
;
1908 if (bi
) bitmap_end
= 1;
1909 while (bi
&& bi
->bi_sector
<
1910 sh
->dev
[i
].sector
+ STRIPE_SECTORS
) {
1911 struct bio
*bi2
= r5_next_bio(bi
, sh
->dev
[i
].sector
);
1912 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1913 if (--bi
->bi_phys_segments
== 0) {
1914 md_write_end(conf
->mddev
);
1915 bi
->bi_next
= *return_bi
;
1921 /* fail any reads if this device is non-operational and
1922 * the data has not reached the cache yet.
1924 if (!test_bit(R5_Wantfill
, &sh
->dev
[i
].flags
) &&
1925 (!test_bit(R5_Insync
, &sh
->dev
[i
].flags
) ||
1926 test_bit(R5_ReadError
, &sh
->dev
[i
].flags
))) {
1927 bi
= sh
->dev
[i
].toread
;
1928 sh
->dev
[i
].toread
= NULL
;
1929 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
1930 wake_up(&conf
->wait_for_overlap
);
1931 if (bi
) s
->to_read
--;
1932 while (bi
&& bi
->bi_sector
<
1933 sh
->dev
[i
].sector
+ STRIPE_SECTORS
) {
1934 struct bio
*nextbi
=
1935 r5_next_bio(bi
, sh
->dev
[i
].sector
);
1936 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1937 if (--bi
->bi_phys_segments
== 0) {
1938 bi
->bi_next
= *return_bi
;
1944 spin_unlock_irq(&conf
->device_lock
);
1946 bitmap_endwrite(conf
->mddev
->bitmap
, sh
->sector
,
1947 STRIPE_SECTORS
, 0, 0);
1952 /* __handle_issuing_new_read_requests5 - returns 0 if there are no more disks
1955 static int __handle_issuing_new_read_requests5(struct stripe_head
*sh
,
1956 struct stripe_head_state
*s
, int disk_idx
, int disks
)
1958 struct r5dev
*dev
= &sh
->dev
[disk_idx
];
1959 struct r5dev
*failed_dev
= &sh
->dev
[s
->failed_num
];
1961 /* don't schedule compute operations or reads on the parity block while
1962 * a check is in flight
1964 if ((disk_idx
== sh
->pd_idx
) &&
1965 test_bit(STRIPE_OP_CHECK
, &sh
->ops
.pending
))
1968 /* is the data in this block needed, and can we get it? */
1969 if (!test_bit(R5_LOCKED
, &dev
->flags
) &&
1970 !test_bit(R5_UPTODATE
, &dev
->flags
) && (dev
->toread
||
1971 (dev
->towrite
&& !test_bit(R5_OVERWRITE
, &dev
->flags
)) ||
1972 s
->syncing
|| s
->expanding
|| (s
->failed
&&
1973 (failed_dev
->toread
|| (failed_dev
->towrite
&&
1974 !test_bit(R5_OVERWRITE
, &failed_dev
->flags
)
1976 /* 1/ We would like to get this block, possibly by computing it,
1977 * but we might not be able to.
1979 * 2/ Since parity check operations potentially make the parity
1980 * block !uptodate it will need to be refreshed before any
1981 * compute operations on data disks are scheduled.
1983 * 3/ We hold off parity block re-reads until check operations
1986 if ((s
->uptodate
== disks
- 1) &&
1987 !test_bit(STRIPE_OP_CHECK
, &sh
->ops
.pending
)) {
1988 set_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
);
1989 set_bit(R5_Wantcompute
, &dev
->flags
);
1990 sh
->ops
.target
= disk_idx
;
1993 /* Careful: from this point on 'uptodate' is in the eye
1994 * of raid5_run_ops which services 'compute' operations
1995 * before writes. R5_Wantcompute flags a block that will
1996 * be R5_UPTODATE by the time it is needed for a
1997 * subsequent operation.
2000 return 0; /* uptodate + compute == disks */
2001 } else if ((s
->uptodate
< disks
- 1) &&
2002 test_bit(R5_Insync
, &dev
->flags
)) {
2003 /* Note: we hold off compute operations while checks are
2004 * in flight, but we still prefer 'compute' over 'read'
2005 * hence we only read if (uptodate < * disks-1)
2007 set_bit(R5_LOCKED
, &dev
->flags
);
2008 set_bit(R5_Wantread
, &dev
->flags
);
2009 if (!test_and_set_bit(STRIPE_OP_IO
, &sh
->ops
.pending
))
2012 pr_debug("Reading block %d (sync=%d)\n", disk_idx
,
2020 static void handle_issuing_new_read_requests5(struct stripe_head
*sh
,
2021 struct stripe_head_state
*s
, int disks
)
2025 /* Clear completed compute operations. Parity recovery
2026 * (STRIPE_OP_MOD_REPAIR_PD) implies a write-back which is handled
2027 * later on in this routine
2029 if (test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.complete
) &&
2030 !test_bit(STRIPE_OP_MOD_REPAIR_PD
, &sh
->ops
.pending
)) {
2031 clear_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.complete
);
2032 clear_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.ack
);
2033 clear_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
);
2036 /* look for blocks to read/compute, skip this if a compute
2037 * is already in flight, or if the stripe contents are in the
2038 * midst of changing due to a write
2040 if (!test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
) &&
2041 !test_bit(STRIPE_OP_PREXOR
, &sh
->ops
.pending
) &&
2042 !test_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
)) {
2043 for (i
= disks
; i
--; )
2044 if (__handle_issuing_new_read_requests5(
2045 sh
, s
, i
, disks
) == 0)
2048 set_bit(STRIPE_HANDLE
, &sh
->state
);
2051 static void handle_issuing_new_read_requests6(struct stripe_head
*sh
,
2052 struct stripe_head_state
*s
, struct r6_state
*r6s
,
2056 for (i
= disks
; i
--; ) {
2057 struct r5dev
*dev
= &sh
->dev
[i
];
2058 if (!test_bit(R5_LOCKED
, &dev
->flags
) &&
2059 !test_bit(R5_UPTODATE
, &dev
->flags
) &&
2060 (dev
->toread
|| (dev
->towrite
&&
2061 !test_bit(R5_OVERWRITE
, &dev
->flags
)) ||
2062 s
->syncing
|| s
->expanding
||
2064 (sh
->dev
[r6s
->failed_num
[0]].toread
||
2067 (sh
->dev
[r6s
->failed_num
[1]].toread
||
2069 /* we would like to get this block, possibly
2070 * by computing it, but we might not be able to
2072 if (s
->uptodate
== disks
-1) {
2073 pr_debug("Computing stripe %llu block %d\n",
2074 (unsigned long long)sh
->sector
, i
);
2075 compute_block_1(sh
, i
, 0);
2077 } else if ( s
->uptodate
== disks
-2 && s
->failed
>= 2 ) {
2078 /* Computing 2-failure is *very* expensive; only
2079 * do it if failed >= 2
2082 for (other
= disks
; other
--; ) {
2085 if (!test_bit(R5_UPTODATE
,
2086 &sh
->dev
[other
].flags
))
2090 pr_debug("Computing stripe %llu blocks %d,%d\n",
2091 (unsigned long long)sh
->sector
,
2093 compute_block_2(sh
, i
, other
);
2095 } else if (test_bit(R5_Insync
, &dev
->flags
)) {
2096 set_bit(R5_LOCKED
, &dev
->flags
);
2097 set_bit(R5_Wantread
, &dev
->flags
);
2099 pr_debug("Reading block %d (sync=%d)\n",
2104 set_bit(STRIPE_HANDLE
, &sh
->state
);
2108 /* handle_completed_write_requests
2109 * any written block on an uptodate or failed drive can be returned.
2110 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
2111 * never LOCKED, so we don't need to test 'failed' directly.
2113 static void handle_completed_write_requests(raid5_conf_t
*conf
,
2114 struct stripe_head
*sh
, int disks
, struct bio
**return_bi
)
2119 for (i
= disks
; i
--; )
2120 if (sh
->dev
[i
].written
) {
2122 if (!test_bit(R5_LOCKED
, &dev
->flags
) &&
2123 test_bit(R5_UPTODATE
, &dev
->flags
)) {
2124 /* We can return any write requests */
2125 struct bio
*wbi
, *wbi2
;
2127 pr_debug("Return write for disc %d\n", i
);
2128 spin_lock_irq(&conf
->device_lock
);
2130 dev
->written
= NULL
;
2131 while (wbi
&& wbi
->bi_sector
<
2132 dev
->sector
+ STRIPE_SECTORS
) {
2133 wbi2
= r5_next_bio(wbi
, dev
->sector
);
2134 if (--wbi
->bi_phys_segments
== 0) {
2135 md_write_end(conf
->mddev
);
2136 wbi
->bi_next
= *return_bi
;
2141 if (dev
->towrite
== NULL
)
2143 spin_unlock_irq(&conf
->device_lock
);
2145 bitmap_endwrite(conf
->mddev
->bitmap
,
2148 !test_bit(STRIPE_DEGRADED
, &sh
->state
),
2154 static void handle_issuing_new_write_requests5(raid5_conf_t
*conf
,
2155 struct stripe_head
*sh
, struct stripe_head_state
*s
, int disks
)
2157 int rmw
= 0, rcw
= 0, i
;
2158 for (i
= disks
; i
--; ) {
2159 /* would I have to read this buffer for read_modify_write */
2160 struct r5dev
*dev
= &sh
->dev
[i
];
2161 if ((dev
->towrite
|| i
== sh
->pd_idx
) &&
2162 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2163 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2164 test_bit(R5_Wantcompute
, &dev
->flags
))) {
2165 if (test_bit(R5_Insync
, &dev
->flags
))
2168 rmw
+= 2*disks
; /* cannot read it */
2170 /* Would I have to read this buffer for reconstruct_write */
2171 if (!test_bit(R5_OVERWRITE
, &dev
->flags
) && i
!= sh
->pd_idx
&&
2172 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2173 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2174 test_bit(R5_Wantcompute
, &dev
->flags
))) {
2175 if (test_bit(R5_Insync
, &dev
->flags
)) rcw
++;
2180 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
2181 (unsigned long long)sh
->sector
, rmw
, rcw
);
2182 set_bit(STRIPE_HANDLE
, &sh
->state
);
2183 if (rmw
< rcw
&& rmw
> 0)
2184 /* prefer read-modify-write, but need to get some data */
2185 for (i
= disks
; i
--; ) {
2186 struct r5dev
*dev
= &sh
->dev
[i
];
2187 if ((dev
->towrite
|| i
== sh
->pd_idx
) &&
2188 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2189 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2190 test_bit(R5_Wantcompute
, &dev
->flags
)) &&
2191 test_bit(R5_Insync
, &dev
->flags
)) {
2193 test_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2194 pr_debug("Read_old block "
2195 "%d for r-m-w\n", i
);
2196 set_bit(R5_LOCKED
, &dev
->flags
);
2197 set_bit(R5_Wantread
, &dev
->flags
);
2198 if (!test_and_set_bit(
2199 STRIPE_OP_IO
, &sh
->ops
.pending
))
2203 set_bit(STRIPE_DELAYED
, &sh
->state
);
2204 set_bit(STRIPE_HANDLE
, &sh
->state
);
2208 if (rcw
<= rmw
&& rcw
> 0)
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
) &&
2214 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2215 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2216 test_bit(R5_Wantcompute
, &dev
->flags
)) &&
2217 test_bit(R5_Insync
, &dev
->flags
)) {
2219 test_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2220 pr_debug("Read_old block "
2221 "%d for Reconstruct\n", i
);
2222 set_bit(R5_LOCKED
, &dev
->flags
);
2223 set_bit(R5_Wantread
, &dev
->flags
);
2224 if (!test_and_set_bit(
2225 STRIPE_OP_IO
, &sh
->ops
.pending
))
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,
2235 * we can start a write request
2237 /* since handle_stripe can be called at any time we need to handle the
2238 * case where a compute block operation has been submitted and then a
2239 * subsequent call wants to start a write request. raid5_run_ops only
2240 * handles the case where compute block and postxor are requested
2241 * simultaneously. If this is not the case then new writes need to be
2242 * held off until the compute completes.
2244 if ((s
->req_compute
||
2245 !test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
)) &&
2246 (s
->locked
== 0 && (rcw
== 0 || rmw
== 0) &&
2247 !test_bit(STRIPE_BIT_DELAY
, &sh
->state
)))
2248 s
->locked
+= handle_write_operations5(sh
, rcw
== 0, 0);
2251 static void handle_issuing_new_write_requests6(raid5_conf_t
*conf
,
2252 struct stripe_head
*sh
, struct stripe_head_state
*s
,
2253 struct r6_state
*r6s
, int disks
)
2255 int rcw
= 0, must_compute
= 0, pd_idx
= sh
->pd_idx
, i
;
2256 int qd_idx
= r6s
->qd_idx
;
2257 for (i
= disks
; i
--; ) {
2258 struct r5dev
*dev
= &sh
->dev
[i
];
2259 /* Would I have to read this buffer for reconstruct_write */
2260 if (!test_bit(R5_OVERWRITE
, &dev
->flags
)
2261 && i
!= pd_idx
&& i
!= qd_idx
2262 && (!test_bit(R5_LOCKED
, &dev
->flags
)
2264 !test_bit(R5_UPTODATE
, &dev
->flags
)) {
2265 if (test_bit(R5_Insync
, &dev
->flags
)) rcw
++;
2267 pr_debug("raid6: must_compute: "
2268 "disk %d flags=%#lx\n", i
, dev
->flags
);
2273 pr_debug("for sector %llu, rcw=%d, must_compute=%d\n",
2274 (unsigned long long)sh
->sector
, rcw
, must_compute
);
2275 set_bit(STRIPE_HANDLE
, &sh
->state
);
2278 /* want reconstruct write, but need to get some data */
2279 for (i
= disks
; i
--; ) {
2280 struct r5dev
*dev
= &sh
->dev
[i
];
2281 if (!test_bit(R5_OVERWRITE
, &dev
->flags
)
2282 && !(s
->failed
== 0 && (i
== pd_idx
|| i
== qd_idx
))
2283 && !test_bit(R5_LOCKED
, &dev
->flags
) &&
2284 !test_bit(R5_UPTODATE
, &dev
->flags
) &&
2285 test_bit(R5_Insync
, &dev
->flags
)) {
2287 test_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2288 pr_debug("Read_old stripe %llu "
2289 "block %d for Reconstruct\n",
2290 (unsigned long long)sh
->sector
, i
);
2291 set_bit(R5_LOCKED
, &dev
->flags
);
2292 set_bit(R5_Wantread
, &dev
->flags
);
2295 pr_debug("Request delayed stripe %llu "
2296 "block %d for Reconstruct\n",
2297 (unsigned long long)sh
->sector
, i
);
2298 set_bit(STRIPE_DELAYED
, &sh
->state
);
2299 set_bit(STRIPE_HANDLE
, &sh
->state
);
2303 /* now if nothing is locked, and if we have enough data, we can start a
2306 if (s
->locked
== 0 && rcw
== 0 &&
2307 !test_bit(STRIPE_BIT_DELAY
, &sh
->state
)) {
2308 if (must_compute
> 0) {
2309 /* We have failed blocks and need to compute them */
2310 switch (s
->failed
) {
2314 compute_block_1(sh
, r6s
->failed_num
[0], 0);
2317 compute_block_2(sh
, r6s
->failed_num
[0],
2318 r6s
->failed_num
[1]);
2320 default: /* This request should have been failed? */
2325 pr_debug("Computing parity for stripe %llu\n",
2326 (unsigned long long)sh
->sector
);
2327 compute_parity6(sh
, RECONSTRUCT_WRITE
);
2328 /* now every locked buffer is ready to be written */
2329 for (i
= disks
; i
--; )
2330 if (test_bit(R5_LOCKED
, &sh
->dev
[i
].flags
)) {
2331 pr_debug("Writing stripe %llu block %d\n",
2332 (unsigned long long)sh
->sector
, i
);
2334 set_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
);
2336 /* after a RECONSTRUCT_WRITE, the stripe MUST be in-sync */
2337 set_bit(STRIPE_INSYNC
, &sh
->state
);
2339 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2340 atomic_dec(&conf
->preread_active_stripes
);
2341 if (atomic_read(&conf
->preread_active_stripes
) <
2343 md_wakeup_thread(conf
->mddev
->thread
);
2348 static void handle_parity_checks5(raid5_conf_t
*conf
, struct stripe_head
*sh
,
2349 struct stripe_head_state
*s
, int disks
)
2351 set_bit(STRIPE_HANDLE
, &sh
->state
);
2352 /* Take one of the following actions:
2353 * 1/ start a check parity operation if (uptodate == disks)
2354 * 2/ finish a check parity operation and act on the result
2355 * 3/ skip to the writeback section if we previously
2356 * initiated a recovery operation
2358 if (s
->failed
== 0 &&
2359 !test_bit(STRIPE_OP_MOD_REPAIR_PD
, &sh
->ops
.pending
)) {
2360 if (!test_and_set_bit(STRIPE_OP_CHECK
, &sh
->ops
.pending
)) {
2361 BUG_ON(s
->uptodate
!= disks
);
2362 clear_bit(R5_UPTODATE
, &sh
->dev
[sh
->pd_idx
].flags
);
2366 test_and_clear_bit(STRIPE_OP_CHECK
, &sh
->ops
.complete
)) {
2367 clear_bit(STRIPE_OP_CHECK
, &sh
->ops
.ack
);
2368 clear_bit(STRIPE_OP_CHECK
, &sh
->ops
.pending
);
2370 if (sh
->ops
.zero_sum_result
== 0)
2371 /* parity is correct (on disc,
2372 * not in buffer any more)
2374 set_bit(STRIPE_INSYNC
, &sh
->state
);
2376 conf
->mddev
->resync_mismatches
+=
2379 MD_RECOVERY_CHECK
, &conf
->mddev
->recovery
))
2380 /* don't try to repair!! */
2381 set_bit(STRIPE_INSYNC
, &sh
->state
);
2383 set_bit(STRIPE_OP_COMPUTE_BLK
,
2385 set_bit(STRIPE_OP_MOD_REPAIR_PD
,
2387 set_bit(R5_Wantcompute
,
2388 &sh
->dev
[sh
->pd_idx
].flags
);
2389 sh
->ops
.target
= sh
->pd_idx
;
2397 /* check if we can clear a parity disk reconstruct */
2398 if (test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.complete
) &&
2399 test_bit(STRIPE_OP_MOD_REPAIR_PD
, &sh
->ops
.pending
)) {
2401 clear_bit(STRIPE_OP_MOD_REPAIR_PD
, &sh
->ops
.pending
);
2402 clear_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.complete
);
2403 clear_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.ack
);
2404 clear_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
);
2407 /* Wait for check parity and compute block operations to complete
2410 if (!test_bit(STRIPE_INSYNC
, &sh
->state
) &&
2411 !test_bit(STRIPE_OP_CHECK
, &sh
->ops
.pending
) &&
2412 !test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
)) {
2414 /* either failed parity check, or recovery is happening */
2416 s
->failed_num
= sh
->pd_idx
;
2417 dev
= &sh
->dev
[s
->failed_num
];
2418 BUG_ON(!test_bit(R5_UPTODATE
, &dev
->flags
));
2419 BUG_ON(s
->uptodate
!= disks
);
2421 set_bit(R5_LOCKED
, &dev
->flags
);
2422 set_bit(R5_Wantwrite
, &dev
->flags
);
2423 if (!test_and_set_bit(STRIPE_OP_IO
, &sh
->ops
.pending
))
2426 clear_bit(STRIPE_DEGRADED
, &sh
->state
);
2428 set_bit(STRIPE_INSYNC
, &sh
->state
);
2433 static void handle_parity_checks6(raid5_conf_t
*conf
, struct stripe_head
*sh
,
2434 struct stripe_head_state
*s
,
2435 struct r6_state
*r6s
, struct page
*tmp_page
,
2438 int update_p
= 0, update_q
= 0;
2440 int pd_idx
= sh
->pd_idx
;
2441 int qd_idx
= r6s
->qd_idx
;
2443 set_bit(STRIPE_HANDLE
, &sh
->state
);
2445 BUG_ON(s
->failed
> 2);
2446 BUG_ON(s
->uptodate
< disks
);
2447 /* Want to check and possibly repair P and Q.
2448 * However there could be one 'failed' device, in which
2449 * case we can only check one of them, possibly using the
2450 * other to generate missing data
2453 /* If !tmp_page, we cannot do the calculations,
2454 * but as we have set STRIPE_HANDLE, we will soon be called
2455 * by stripe_handle with a tmp_page - just wait until then.
2458 if (s
->failed
== r6s
->q_failed
) {
2459 /* The only possible failed device holds 'Q', so it
2460 * makes sense to check P (If anything else were failed,
2461 * we would have used P to recreate it).
2463 compute_block_1(sh
, pd_idx
, 1);
2464 if (!page_is_zero(sh
->dev
[pd_idx
].page
)) {
2465 compute_block_1(sh
, pd_idx
, 0);
2469 if (!r6s
->q_failed
&& s
->failed
< 2) {
2470 /* q is not failed, and we didn't use it to generate
2471 * anything, so it makes sense to check it
2473 memcpy(page_address(tmp_page
),
2474 page_address(sh
->dev
[qd_idx
].page
),
2476 compute_parity6(sh
, UPDATE_PARITY
);
2477 if (memcmp(page_address(tmp_page
),
2478 page_address(sh
->dev
[qd_idx
].page
),
2479 STRIPE_SIZE
) != 0) {
2480 clear_bit(STRIPE_INSYNC
, &sh
->state
);
2484 if (update_p
|| update_q
) {
2485 conf
->mddev
->resync_mismatches
+= STRIPE_SECTORS
;
2486 if (test_bit(MD_RECOVERY_CHECK
, &conf
->mddev
->recovery
))
2487 /* don't try to repair!! */
2488 update_p
= update_q
= 0;
2491 /* now write out any block on a failed drive,
2492 * or P or Q if they need it
2495 if (s
->failed
== 2) {
2496 dev
= &sh
->dev
[r6s
->failed_num
[1]];
2498 set_bit(R5_LOCKED
, &dev
->flags
);
2499 set_bit(R5_Wantwrite
, &dev
->flags
);
2501 if (s
->failed
>= 1) {
2502 dev
= &sh
->dev
[r6s
->failed_num
[0]];
2504 set_bit(R5_LOCKED
, &dev
->flags
);
2505 set_bit(R5_Wantwrite
, &dev
->flags
);
2509 dev
= &sh
->dev
[pd_idx
];
2511 set_bit(R5_LOCKED
, &dev
->flags
);
2512 set_bit(R5_Wantwrite
, &dev
->flags
);
2515 dev
= &sh
->dev
[qd_idx
];
2517 set_bit(R5_LOCKED
, &dev
->flags
);
2518 set_bit(R5_Wantwrite
, &dev
->flags
);
2520 clear_bit(STRIPE_DEGRADED
, &sh
->state
);
2522 set_bit(STRIPE_INSYNC
, &sh
->state
);
2526 static void handle_stripe_expansion(raid5_conf_t
*conf
, struct stripe_head
*sh
,
2527 struct r6_state
*r6s
)
2531 /* We have read all the blocks in this stripe and now we need to
2532 * copy some of them into a target stripe for expand.
2534 struct dma_async_tx_descriptor
*tx
= NULL
;
2535 clear_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
2536 for (i
= 0; i
< sh
->disks
; i
++)
2537 if (i
!= sh
->pd_idx
&& (!r6s
|| i
!= r6s
->qd_idx
)) {
2538 int dd_idx
, pd_idx
, j
;
2539 struct stripe_head
*sh2
;
2541 sector_t bn
= compute_blocknr(sh
, i
);
2542 sector_t s
= raid5_compute_sector(bn
, conf
->raid_disks
,
2544 conf
->max_degraded
, &dd_idx
,
2546 sh2
= get_active_stripe(conf
, s
, conf
->raid_disks
,
2549 /* so far only the early blocks of this stripe
2550 * have been requested. When later blocks
2551 * get requested, we will try again
2554 if (!test_bit(STRIPE_EXPANDING
, &sh2
->state
) ||
2555 test_bit(R5_Expanded
, &sh2
->dev
[dd_idx
].flags
)) {
2556 /* must have already done this block */
2557 release_stripe(sh2
);
2561 /* place all the copies on one channel */
2562 tx
= async_memcpy(sh2
->dev
[dd_idx
].page
,
2563 sh
->dev
[i
].page
, 0, 0, STRIPE_SIZE
,
2564 ASYNC_TX_DEP_ACK
, tx
, NULL
, NULL
);
2566 set_bit(R5_Expanded
, &sh2
->dev
[dd_idx
].flags
);
2567 set_bit(R5_UPTODATE
, &sh2
->dev
[dd_idx
].flags
);
2568 for (j
= 0; j
< conf
->raid_disks
; j
++)
2569 if (j
!= sh2
->pd_idx
&&
2570 (!r6s
|| j
!= raid6_next_disk(sh2
->pd_idx
,
2572 !test_bit(R5_Expanded
, &sh2
->dev
[j
].flags
))
2574 if (j
== conf
->raid_disks
) {
2575 set_bit(STRIPE_EXPAND_READY
, &sh2
->state
);
2576 set_bit(STRIPE_HANDLE
, &sh2
->state
);
2578 release_stripe(sh2
);
2581 /* done submitting copies, wait for them to complete */
2584 dma_wait_for_async_tx(tx
);
2589 * handle_stripe - do things to a stripe.
2591 * We lock the stripe and then examine the state of various bits
2592 * to see what needs to be done.
2594 * return some read request which now have data
2595 * return some write requests which are safely on disc
2596 * schedule a read on some buffers
2597 * schedule a write of some buffers
2598 * return confirmation of parity correctness
2600 * buffers are taken off read_list or write_list, and bh_cache buffers
2601 * get BH_Lock set before the stripe lock is released.
2605 static void handle_stripe5(struct stripe_head
*sh
)
2607 raid5_conf_t
*conf
= sh
->raid_conf
;
2608 int disks
= sh
->disks
, i
;
2609 struct bio
*return_bi
= NULL
;
2610 struct stripe_head_state s
;
2612 unsigned long pending
= 0;
2614 memset(&s
, 0, sizeof(s
));
2615 pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d "
2616 "ops=%lx:%lx:%lx\n", (unsigned long long)sh
->sector
, sh
->state
,
2617 atomic_read(&sh
->count
), sh
->pd_idx
,
2618 sh
->ops
.pending
, sh
->ops
.ack
, sh
->ops
.complete
);
2620 spin_lock(&sh
->lock
);
2621 clear_bit(STRIPE_HANDLE
, &sh
->state
);
2622 clear_bit(STRIPE_DELAYED
, &sh
->state
);
2624 s
.syncing
= test_bit(STRIPE_SYNCING
, &sh
->state
);
2625 s
.expanding
= test_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
2626 s
.expanded
= test_bit(STRIPE_EXPAND_READY
, &sh
->state
);
2627 /* Now to look around and see what can be done */
2629 /* clean-up completed biofill operations */
2630 if (test_bit(STRIPE_OP_BIOFILL
, &sh
->ops
.complete
)) {
2631 clear_bit(STRIPE_OP_BIOFILL
, &sh
->ops
.pending
);
2632 clear_bit(STRIPE_OP_BIOFILL
, &sh
->ops
.ack
);
2633 clear_bit(STRIPE_OP_BIOFILL
, &sh
->ops
.complete
);
2637 for (i
=disks
; i
--; ) {
2639 struct r5dev
*dev
= &sh
->dev
[i
];
2640 clear_bit(R5_Insync
, &dev
->flags
);
2642 pr_debug("check %d: state 0x%lx toread %p read %p write %p "
2643 "written %p\n", i
, dev
->flags
, dev
->toread
, dev
->read
,
2644 dev
->towrite
, dev
->written
);
2646 /* maybe we can request a biofill operation
2648 * new wantfill requests are only permitted while
2649 * STRIPE_OP_BIOFILL is clear
2651 if (test_bit(R5_UPTODATE
, &dev
->flags
) && dev
->toread
&&
2652 !test_bit(STRIPE_OP_BIOFILL
, &sh
->ops
.pending
))
2653 set_bit(R5_Wantfill
, &dev
->flags
);
2655 /* now count some things */
2656 if (test_bit(R5_LOCKED
, &dev
->flags
)) s
.locked
++;
2657 if (test_bit(R5_UPTODATE
, &dev
->flags
)) s
.uptodate
++;
2658 if (test_bit(R5_Wantcompute
, &dev
->flags
)) s
.compute
++;
2660 if (test_bit(R5_Wantfill
, &dev
->flags
))
2662 else if (dev
->toread
)
2666 if (!test_bit(R5_OVERWRITE
, &dev
->flags
))
2671 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
2672 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)) {
2673 /* The ReadError flag will just be confusing now */
2674 clear_bit(R5_ReadError
, &dev
->flags
);
2675 clear_bit(R5_ReWrite
, &dev
->flags
);
2677 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)
2678 || test_bit(R5_ReadError
, &dev
->flags
)) {
2682 set_bit(R5_Insync
, &dev
->flags
);
2686 if (s
.to_fill
&& !test_and_set_bit(STRIPE_OP_BIOFILL
, &sh
->ops
.pending
))
2689 pr_debug("locked=%d uptodate=%d to_read=%d"
2690 " to_write=%d failed=%d failed_num=%d\n",
2691 s
.locked
, s
.uptodate
, s
.to_read
, s
.to_write
,
2692 s
.failed
, s
.failed_num
);
2693 /* check if the array has lost two devices and, if so, some requests might
2696 if (s
.failed
> 1 && s
.to_read
+s
.to_write
+s
.written
)
2697 handle_requests_to_failed_array(conf
, sh
, &s
, disks
,
2699 if (s
.failed
> 1 && s
.syncing
) {
2700 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,0);
2701 clear_bit(STRIPE_SYNCING
, &sh
->state
);
2705 /* might be able to return some write requests if the parity block
2706 * is safe, or on a failed drive
2708 dev
= &sh
->dev
[sh
->pd_idx
];
2710 ((test_bit(R5_Insync
, &dev
->flags
) &&
2711 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2712 test_bit(R5_UPTODATE
, &dev
->flags
)) ||
2713 (s
.failed
== 1 && s
.failed_num
== sh
->pd_idx
)))
2714 handle_completed_write_requests(conf
, sh
, disks
, &return_bi
);
2716 /* Now we might consider reading some blocks, either to check/generate
2717 * parity, or to satisfy requests
2718 * or to load a block that is being partially written.
2720 if (s
.to_read
|| s
.non_overwrite
||
2721 (s
.syncing
&& (s
.uptodate
+ s
.compute
< disks
)) || s
.expanding
||
2722 test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
))
2723 handle_issuing_new_read_requests5(sh
, &s
, disks
);
2725 /* Now we check to see if any write operations have recently
2729 /* leave prexor set until postxor is done, allows us to distinguish
2730 * a rmw from a rcw during biodrain
2732 if (test_bit(STRIPE_OP_PREXOR
, &sh
->ops
.complete
) &&
2733 test_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.complete
)) {
2735 clear_bit(STRIPE_OP_PREXOR
, &sh
->ops
.complete
);
2736 clear_bit(STRIPE_OP_PREXOR
, &sh
->ops
.ack
);
2737 clear_bit(STRIPE_OP_PREXOR
, &sh
->ops
.pending
);
2739 for (i
= disks
; i
--; )
2740 clear_bit(R5_Wantprexor
, &sh
->dev
[i
].flags
);
2743 /* if only POSTXOR is set then this is an 'expand' postxor */
2744 if (test_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.complete
) &&
2745 test_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.complete
)) {
2747 clear_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.complete
);
2748 clear_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.ack
);
2749 clear_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.pending
);
2751 clear_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.complete
);
2752 clear_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.ack
);
2753 clear_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
);
2755 /* All the 'written' buffers and the parity block are ready to
2756 * be written back to disk
2758 BUG_ON(!test_bit(R5_UPTODATE
, &sh
->dev
[sh
->pd_idx
].flags
));
2759 for (i
= disks
; i
--; ) {
2761 if (test_bit(R5_LOCKED
, &dev
->flags
) &&
2762 (i
== sh
->pd_idx
|| dev
->written
)) {
2763 pr_debug("Writing block %d\n", i
);
2764 set_bit(R5_Wantwrite
, &dev
->flags
);
2765 if (!test_and_set_bit(
2766 STRIPE_OP_IO
, &sh
->ops
.pending
))
2768 if (!test_bit(R5_Insync
, &dev
->flags
) ||
2769 (i
== sh
->pd_idx
&& s
.failed
== 0))
2770 set_bit(STRIPE_INSYNC
, &sh
->state
);
2773 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2774 atomic_dec(&conf
->preread_active_stripes
);
2775 if (atomic_read(&conf
->preread_active_stripes
) <
2777 md_wakeup_thread(conf
->mddev
->thread
);
2781 /* Now to consider new write requests and what else, if anything
2782 * should be read. We do not handle new writes when:
2783 * 1/ A 'write' operation (copy+xor) is already in flight.
2784 * 2/ A 'check' operation is in flight, as it may clobber the parity
2787 if (s
.to_write
&& !test_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
) &&
2788 !test_bit(STRIPE_OP_CHECK
, &sh
->ops
.pending
))
2789 handle_issuing_new_write_requests5(conf
, sh
, &s
, disks
);
2791 /* maybe we need to check and possibly fix the parity for this stripe
2792 * Any reads will already have been scheduled, so we just see if enough
2793 * data is available. The parity check is held off while parity
2794 * dependent operations are in flight.
2796 if ((s
.syncing
&& s
.locked
== 0 &&
2797 !test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
) &&
2798 !test_bit(STRIPE_INSYNC
, &sh
->state
)) ||
2799 test_bit(STRIPE_OP_CHECK
, &sh
->ops
.pending
) ||
2800 test_bit(STRIPE_OP_MOD_REPAIR_PD
, &sh
->ops
.pending
))
2801 handle_parity_checks5(conf
, sh
, &s
, disks
);
2803 if (s
.syncing
&& s
.locked
== 0 && test_bit(STRIPE_INSYNC
, &sh
->state
)) {
2804 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,1);
2805 clear_bit(STRIPE_SYNCING
, &sh
->state
);
2808 /* If the failed drive is just a ReadError, then we might need to progress
2809 * the repair/check process
2811 if (s
.failed
== 1 && !conf
->mddev
->ro
&&
2812 test_bit(R5_ReadError
, &sh
->dev
[s
.failed_num
].flags
)
2813 && !test_bit(R5_LOCKED
, &sh
->dev
[s
.failed_num
].flags
)
2814 && test_bit(R5_UPTODATE
, &sh
->dev
[s
.failed_num
].flags
)
2816 dev
= &sh
->dev
[s
.failed_num
];
2817 if (!test_bit(R5_ReWrite
, &dev
->flags
)) {
2818 set_bit(R5_Wantwrite
, &dev
->flags
);
2819 if (!test_and_set_bit(STRIPE_OP_IO
, &sh
->ops
.pending
))
2821 set_bit(R5_ReWrite
, &dev
->flags
);
2822 set_bit(R5_LOCKED
, &dev
->flags
);
2825 /* let's read it back */
2826 set_bit(R5_Wantread
, &dev
->flags
);
2827 if (!test_and_set_bit(STRIPE_OP_IO
, &sh
->ops
.pending
))
2829 set_bit(R5_LOCKED
, &dev
->flags
);
2834 /* Finish postxor operations initiated by the expansion
2837 if (test_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.complete
) &&
2838 !test_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.pending
)) {
2840 clear_bit(STRIPE_EXPANDING
, &sh
->state
);
2842 clear_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
);
2843 clear_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.ack
);
2844 clear_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.complete
);
2846 for (i
= conf
->raid_disks
; i
--; ) {
2847 set_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
);
2848 if (!test_and_set_bit(STRIPE_OP_IO
, &sh
->ops
.pending
))
2853 if (s
.expanded
&& test_bit(STRIPE_EXPANDING
, &sh
->state
) &&
2854 !test_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
)) {
2855 /* Need to write out all blocks after computing parity */
2856 sh
->disks
= conf
->raid_disks
;
2857 sh
->pd_idx
= stripe_to_pdidx(sh
->sector
, conf
,
2859 s
.locked
+= handle_write_operations5(sh
, 1, 1);
2860 } else if (s
.expanded
&&
2861 !test_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
)) {
2862 clear_bit(STRIPE_EXPAND_READY
, &sh
->state
);
2863 atomic_dec(&conf
->reshape_stripes
);
2864 wake_up(&conf
->wait_for_overlap
);
2865 md_done_sync(conf
->mddev
, STRIPE_SECTORS
, 1);
2868 if (s
.expanding
&& s
.locked
== 0)
2869 handle_stripe_expansion(conf
, sh
, NULL
);
2872 pending
= get_stripe_work(sh
);
2874 spin_unlock(&sh
->lock
);
2877 raid5_run_ops(sh
, pending
);
2879 return_io(return_bi
);
2883 static void handle_stripe6(struct stripe_head
*sh
, struct page
*tmp_page
)
2885 raid6_conf_t
*conf
= sh
->raid_conf
;
2886 int disks
= sh
->disks
;
2887 struct bio
*return_bi
= NULL
;
2888 int i
, pd_idx
= sh
->pd_idx
;
2889 struct stripe_head_state s
;
2890 struct r6_state r6s
;
2891 struct r5dev
*dev
, *pdev
, *qdev
;
2893 r6s
.qd_idx
= raid6_next_disk(pd_idx
, disks
);
2894 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
2895 "pd_idx=%d, qd_idx=%d\n",
2896 (unsigned long long)sh
->sector
, sh
->state
,
2897 atomic_read(&sh
->count
), pd_idx
, r6s
.qd_idx
);
2898 memset(&s
, 0, sizeof(s
));
2900 spin_lock(&sh
->lock
);
2901 clear_bit(STRIPE_HANDLE
, &sh
->state
);
2902 clear_bit(STRIPE_DELAYED
, &sh
->state
);
2904 s
.syncing
= test_bit(STRIPE_SYNCING
, &sh
->state
);
2905 s
.expanding
= test_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
2906 s
.expanded
= test_bit(STRIPE_EXPAND_READY
, &sh
->state
);
2907 /* Now to look around and see what can be done */
2910 for (i
=disks
; i
--; ) {
2913 clear_bit(R5_Insync
, &dev
->flags
);
2915 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
2916 i
, dev
->flags
, dev
->toread
, dev
->towrite
, dev
->written
);
2917 /* maybe we can reply to a read */
2918 if (test_bit(R5_UPTODATE
, &dev
->flags
) && dev
->toread
) {
2919 struct bio
*rbi
, *rbi2
;
2920 pr_debug("Return read for disc %d\n", i
);
2921 spin_lock_irq(&conf
->device_lock
);
2924 if (test_and_clear_bit(R5_Overlap
, &dev
->flags
))
2925 wake_up(&conf
->wait_for_overlap
);
2926 spin_unlock_irq(&conf
->device_lock
);
2927 while (rbi
&& rbi
->bi_sector
< dev
->sector
+ STRIPE_SECTORS
) {
2928 copy_data(0, rbi
, dev
->page
, dev
->sector
);
2929 rbi2
= r5_next_bio(rbi
, dev
->sector
);
2930 spin_lock_irq(&conf
->device_lock
);
2931 if (--rbi
->bi_phys_segments
== 0) {
2932 rbi
->bi_next
= return_bi
;
2935 spin_unlock_irq(&conf
->device_lock
);
2940 /* now count some things */
2941 if (test_bit(R5_LOCKED
, &dev
->flags
)) s
.locked
++;
2942 if (test_bit(R5_UPTODATE
, &dev
->flags
)) s
.uptodate
++;
2949 if (!test_bit(R5_OVERWRITE
, &dev
->flags
))
2954 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
2955 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)) {
2956 /* The ReadError flag will just be confusing now */
2957 clear_bit(R5_ReadError
, &dev
->flags
);
2958 clear_bit(R5_ReWrite
, &dev
->flags
);
2960 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)
2961 || test_bit(R5_ReadError
, &dev
->flags
)) {
2963 r6s
.failed_num
[s
.failed
] = i
;
2966 set_bit(R5_Insync
, &dev
->flags
);
2969 pr_debug("locked=%d uptodate=%d to_read=%d"
2970 " to_write=%d failed=%d failed_num=%d,%d\n",
2971 s
.locked
, s
.uptodate
, s
.to_read
, s
.to_write
, s
.failed
,
2972 r6s
.failed_num
[0], r6s
.failed_num
[1]);
2973 /* check if the array has lost >2 devices and, if so, some requests
2974 * might need to be failed
2976 if (s
.failed
> 2 && s
.to_read
+s
.to_write
+s
.written
)
2977 handle_requests_to_failed_array(conf
, sh
, &s
, disks
,
2979 if (s
.failed
> 2 && s
.syncing
) {
2980 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,0);
2981 clear_bit(STRIPE_SYNCING
, &sh
->state
);
2986 * might be able to return some write requests if the parity blocks
2987 * are safe, or on a failed drive
2989 pdev
= &sh
->dev
[pd_idx
];
2990 r6s
.p_failed
= (s
.failed
>= 1 && r6s
.failed_num
[0] == pd_idx
)
2991 || (s
.failed
>= 2 && r6s
.failed_num
[1] == pd_idx
);
2992 qdev
= &sh
->dev
[r6s
.qd_idx
];
2993 r6s
.q_failed
= (s
.failed
>= 1 && r6s
.failed_num
[0] == r6s
.qd_idx
)
2994 || (s
.failed
>= 2 && r6s
.failed_num
[1] == r6s
.qd_idx
);
2997 ( r6s
.p_failed
|| ((test_bit(R5_Insync
, &pdev
->flags
)
2998 && !test_bit(R5_LOCKED
, &pdev
->flags
)
2999 && test_bit(R5_UPTODATE
, &pdev
->flags
)))) &&
3000 ( r6s
.q_failed
|| ((test_bit(R5_Insync
, &qdev
->flags
)
3001 && !test_bit(R5_LOCKED
, &qdev
->flags
)
3002 && test_bit(R5_UPTODATE
, &qdev
->flags
)))))
3003 handle_completed_write_requests(conf
, sh
, disks
, &return_bi
);
3005 /* Now we might consider reading some blocks, either to check/generate
3006 * parity, or to satisfy requests
3007 * or to load a block that is being partially written.
3009 if (s
.to_read
|| s
.non_overwrite
|| (s
.to_write
&& s
.failed
) ||
3010 (s
.syncing
&& (s
.uptodate
< disks
)) || s
.expanding
)
3011 handle_issuing_new_read_requests6(sh
, &s
, &r6s
, disks
);
3013 /* now to consider writing and what else, if anything should be read */
3015 handle_issuing_new_write_requests6(conf
, sh
, &s
, &r6s
, disks
);
3017 /* maybe we need to check and possibly fix the parity for this stripe
3018 * Any reads will already have been scheduled, so we just see if enough
3021 if (s
.syncing
&& s
.locked
== 0 && !test_bit(STRIPE_INSYNC
, &sh
->state
))
3022 handle_parity_checks6(conf
, sh
, &s
, &r6s
, tmp_page
, disks
);
3024 if (s
.syncing
&& s
.locked
== 0 && test_bit(STRIPE_INSYNC
, &sh
->state
)) {
3025 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,1);
3026 clear_bit(STRIPE_SYNCING
, &sh
->state
);
3029 /* If the failed drives are just a ReadError, then we might need
3030 * to progress the repair/check process
3032 if (s
.failed
<= 2 && !conf
->mddev
->ro
)
3033 for (i
= 0; i
< s
.failed
; i
++) {
3034 dev
= &sh
->dev
[r6s
.failed_num
[i
]];
3035 if (test_bit(R5_ReadError
, &dev
->flags
)
3036 && !test_bit(R5_LOCKED
, &dev
->flags
)
3037 && test_bit(R5_UPTODATE
, &dev
->flags
)
3039 if (!test_bit(R5_ReWrite
, &dev
->flags
)) {
3040 set_bit(R5_Wantwrite
, &dev
->flags
);
3041 set_bit(R5_ReWrite
, &dev
->flags
);
3042 set_bit(R5_LOCKED
, &dev
->flags
);
3044 /* let's read it back */
3045 set_bit(R5_Wantread
, &dev
->flags
);
3046 set_bit(R5_LOCKED
, &dev
->flags
);
3051 if (s
.expanded
&& test_bit(STRIPE_EXPANDING
, &sh
->state
)) {
3052 /* Need to write out all blocks after computing P&Q */
3053 sh
->disks
= conf
->raid_disks
;
3054 sh
->pd_idx
= stripe_to_pdidx(sh
->sector
, conf
,
3056 compute_parity6(sh
, RECONSTRUCT_WRITE
);
3057 for (i
= conf
->raid_disks
; i
-- ; ) {
3058 set_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
3060 set_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
);
3062 clear_bit(STRIPE_EXPANDING
, &sh
->state
);
3063 } else if (s
.expanded
) {
3064 clear_bit(STRIPE_EXPAND_READY
, &sh
->state
);
3065 atomic_dec(&conf
->reshape_stripes
);
3066 wake_up(&conf
->wait_for_overlap
);
3067 md_done_sync(conf
->mddev
, STRIPE_SECTORS
, 1);
3070 if (s
.expanding
&& s
.locked
== 0)
3071 handle_stripe_expansion(conf
, sh
, &r6s
);
3073 spin_unlock(&sh
->lock
);
3075 return_io(return_bi
);
3077 for (i
=disks
; i
-- ;) {
3081 if (test_and_clear_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
))
3083 else if (test_and_clear_bit(R5_Wantread
, &sh
->dev
[i
].flags
))
3088 bi
= &sh
->dev
[i
].req
;
3092 bi
->bi_end_io
= raid5_end_write_request
;
3094 bi
->bi_end_io
= raid5_end_read_request
;
3097 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
3098 if (rdev
&& test_bit(Faulty
, &rdev
->flags
))
3101 atomic_inc(&rdev
->nr_pending
);
3105 if (s
.syncing
|| s
.expanding
|| s
.expanded
)
3106 md_sync_acct(rdev
->bdev
, STRIPE_SECTORS
);
3108 bi
->bi_bdev
= rdev
->bdev
;
3109 pr_debug("for %llu schedule op %ld on disc %d\n",
3110 (unsigned long long)sh
->sector
, bi
->bi_rw
, i
);
3111 atomic_inc(&sh
->count
);
3112 bi
->bi_sector
= sh
->sector
+ rdev
->data_offset
;
3113 bi
->bi_flags
= 1 << BIO_UPTODATE
;
3115 bi
->bi_max_vecs
= 1;
3117 bi
->bi_io_vec
= &sh
->dev
[i
].vec
;
3118 bi
->bi_io_vec
[0].bv_len
= STRIPE_SIZE
;
3119 bi
->bi_io_vec
[0].bv_offset
= 0;
3120 bi
->bi_size
= STRIPE_SIZE
;
3123 test_bit(R5_ReWrite
, &sh
->dev
[i
].flags
))
3124 atomic_add(STRIPE_SECTORS
, &rdev
->corrected_errors
);
3125 generic_make_request(bi
);
3128 set_bit(STRIPE_DEGRADED
, &sh
->state
);
3129 pr_debug("skip op %ld on disc %d for sector %llu\n",
3130 bi
->bi_rw
, i
, (unsigned long long)sh
->sector
);
3131 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
3132 set_bit(STRIPE_HANDLE
, &sh
->state
);
3137 static void handle_stripe(struct stripe_head
*sh
, struct page
*tmp_page
)
3139 if (sh
->raid_conf
->level
== 6)
3140 handle_stripe6(sh
, tmp_page
);
3147 static void raid5_activate_delayed(raid5_conf_t
*conf
)
3149 if (atomic_read(&conf
->preread_active_stripes
) < IO_THRESHOLD
) {
3150 while (!list_empty(&conf
->delayed_list
)) {
3151 struct list_head
*l
= conf
->delayed_list
.next
;
3152 struct stripe_head
*sh
;
3153 sh
= list_entry(l
, struct stripe_head
, lru
);
3155 clear_bit(STRIPE_DELAYED
, &sh
->state
);
3156 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
))
3157 atomic_inc(&conf
->preread_active_stripes
);
3158 list_add_tail(&sh
->lru
, &conf
->handle_list
);
3163 static void activate_bit_delay(raid5_conf_t
*conf
)
3165 /* device_lock is held */
3166 struct list_head head
;
3167 list_add(&head
, &conf
->bitmap_list
);
3168 list_del_init(&conf
->bitmap_list
);
3169 while (!list_empty(&head
)) {
3170 struct stripe_head
*sh
= list_entry(head
.next
, struct stripe_head
, lru
);
3171 list_del_init(&sh
->lru
);
3172 atomic_inc(&sh
->count
);
3173 __release_stripe(conf
, sh
);
3177 static void unplug_slaves(mddev_t
*mddev
)
3179 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3183 for (i
=0; i
<mddev
->raid_disks
; i
++) {
3184 mdk_rdev_t
*rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
3185 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
) && atomic_read(&rdev
->nr_pending
)) {
3186 struct request_queue
*r_queue
= bdev_get_queue(rdev
->bdev
);
3188 atomic_inc(&rdev
->nr_pending
);
3191 blk_unplug(r_queue
);
3193 rdev_dec_pending(rdev
, mddev
);
3200 static void raid5_unplug_device(struct request_queue
*q
)
3202 mddev_t
*mddev
= q
->queuedata
;
3203 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3204 unsigned long flags
;
3206 spin_lock_irqsave(&conf
->device_lock
, flags
);
3208 if (blk_remove_plug(q
)) {
3210 raid5_activate_delayed(conf
);
3212 md_wakeup_thread(mddev
->thread
);
3214 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
3216 unplug_slaves(mddev
);
3219 static int raid5_congested(void *data
, int bits
)
3221 mddev_t
*mddev
= data
;
3222 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3224 /* No difference between reads and writes. Just check
3225 * how busy the stripe_cache is
3227 if (conf
->inactive_blocked
)
3231 if (list_empty_careful(&conf
->inactive_list
))
3237 /* We want read requests to align with chunks where possible,
3238 * but write requests don't need to.
3240 static int raid5_mergeable_bvec(struct request_queue
*q
, struct bio
*bio
, struct bio_vec
*biovec
)
3242 mddev_t
*mddev
= q
->queuedata
;
3243 sector_t sector
= bio
->bi_sector
+ get_start_sect(bio
->bi_bdev
);
3245 unsigned int chunk_sectors
= mddev
->chunk_size
>> 9;
3246 unsigned int bio_sectors
= bio
->bi_size
>> 9;
3248 if (bio_data_dir(bio
) == WRITE
)
3249 return biovec
->bv_len
; /* always allow writes to be mergeable */
3251 max
= (chunk_sectors
- ((sector
& (chunk_sectors
- 1)) + bio_sectors
)) << 9;
3252 if (max
< 0) max
= 0;
3253 if (max
<= biovec
->bv_len
&& bio_sectors
== 0)
3254 return biovec
->bv_len
;
3260 static int in_chunk_boundary(mddev_t
*mddev
, struct bio
*bio
)
3262 sector_t sector
= bio
->bi_sector
+ get_start_sect(bio
->bi_bdev
);
3263 unsigned int chunk_sectors
= mddev
->chunk_size
>> 9;
3264 unsigned int bio_sectors
= bio
->bi_size
>> 9;
3266 return chunk_sectors
>=
3267 ((sector
& (chunk_sectors
- 1)) + bio_sectors
);
3271 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
3272 * later sampled by raid5d.
3274 static void add_bio_to_retry(struct bio
*bi
,raid5_conf_t
*conf
)
3276 unsigned long flags
;
3278 spin_lock_irqsave(&conf
->device_lock
, flags
);
3280 bi
->bi_next
= conf
->retry_read_aligned_list
;
3281 conf
->retry_read_aligned_list
= bi
;
3283 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
3284 md_wakeup_thread(conf
->mddev
->thread
);
3288 static struct bio
*remove_bio_from_retry(raid5_conf_t
*conf
)
3292 bi
= conf
->retry_read_aligned
;
3294 conf
->retry_read_aligned
= NULL
;
3297 bi
= conf
->retry_read_aligned_list
;
3299 conf
->retry_read_aligned_list
= bi
->bi_next
;
3301 bi
->bi_phys_segments
= 1; /* biased count of active stripes */
3302 bi
->bi_hw_segments
= 0; /* count of processed stripes */
3310 * The "raid5_align_endio" should check if the read succeeded and if it
3311 * did, call bio_endio on the original bio (having bio_put the new bio
3313 * If the read failed..
3315 static void raid5_align_endio(struct bio
*bi
, int error
)
3317 struct bio
* raid_bi
= bi
->bi_private
;
3320 int uptodate
= test_bit(BIO_UPTODATE
, &bi
->bi_flags
);
3325 mddev
= raid_bi
->bi_bdev
->bd_disk
->queue
->queuedata
;
3326 conf
= mddev_to_conf(mddev
);
3327 rdev
= (void*)raid_bi
->bi_next
;
3328 raid_bi
->bi_next
= NULL
;
3330 rdev_dec_pending(rdev
, conf
->mddev
);
3332 if (!error
&& uptodate
) {
3333 bio_endio(raid_bi
, 0);
3334 if (atomic_dec_and_test(&conf
->active_aligned_reads
))
3335 wake_up(&conf
->wait_for_stripe
);
3340 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
3342 add_bio_to_retry(raid_bi
, conf
);
3345 static int bio_fits_rdev(struct bio
*bi
)
3347 struct request_queue
*q
= bdev_get_queue(bi
->bi_bdev
);
3349 if ((bi
->bi_size
>>9) > q
->max_sectors
)
3351 blk_recount_segments(q
, bi
);
3352 if (bi
->bi_phys_segments
> q
->max_phys_segments
||
3353 bi
->bi_hw_segments
> q
->max_hw_segments
)
3356 if (q
->merge_bvec_fn
)
3357 /* it's too hard to apply the merge_bvec_fn at this stage,
3366 static int chunk_aligned_read(struct request_queue
*q
, struct bio
* raid_bio
)
3368 mddev_t
*mddev
= q
->queuedata
;
3369 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3370 const unsigned int raid_disks
= conf
->raid_disks
;
3371 const unsigned int data_disks
= raid_disks
- conf
->max_degraded
;
3372 unsigned int dd_idx
, pd_idx
;
3373 struct bio
* align_bi
;
3376 if (!in_chunk_boundary(mddev
, raid_bio
)) {
3377 pr_debug("chunk_aligned_read : non aligned\n");
3381 * use bio_clone to make a copy of the bio
3383 align_bi
= bio_clone(raid_bio
, GFP_NOIO
);
3387 * set bi_end_io to a new function, and set bi_private to the
3390 align_bi
->bi_end_io
= raid5_align_endio
;
3391 align_bi
->bi_private
= raid_bio
;
3395 align_bi
->bi_sector
= raid5_compute_sector(raid_bio
->bi_sector
,
3403 rdev
= rcu_dereference(conf
->disks
[dd_idx
].rdev
);
3404 if (rdev
&& test_bit(In_sync
, &rdev
->flags
)) {
3405 atomic_inc(&rdev
->nr_pending
);
3407 raid_bio
->bi_next
= (void*)rdev
;
3408 align_bi
->bi_bdev
= rdev
->bdev
;
3409 align_bi
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
3410 align_bi
->bi_sector
+= rdev
->data_offset
;
3412 if (!bio_fits_rdev(align_bi
)) {
3413 /* too big in some way */
3415 rdev_dec_pending(rdev
, mddev
);
3419 spin_lock_irq(&conf
->device_lock
);
3420 wait_event_lock_irq(conf
->wait_for_stripe
,
3422 conf
->device_lock
, /* nothing */);
3423 atomic_inc(&conf
->active_aligned_reads
);
3424 spin_unlock_irq(&conf
->device_lock
);
3426 generic_make_request(align_bi
);
3436 static int make_request(struct request_queue
*q
, struct bio
* bi
)
3438 mddev_t
*mddev
= q
->queuedata
;
3439 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3440 unsigned int dd_idx
, pd_idx
;
3441 sector_t new_sector
;
3442 sector_t logical_sector
, last_sector
;
3443 struct stripe_head
*sh
;
3444 const int rw
= bio_data_dir(bi
);
3447 if (unlikely(bio_barrier(bi
))) {
3448 bio_endio(bi
, -EOPNOTSUPP
);
3452 md_write_start(mddev
, bi
);
3454 disk_stat_inc(mddev
->gendisk
, ios
[rw
]);
3455 disk_stat_add(mddev
->gendisk
, sectors
[rw
], bio_sectors(bi
));
3458 mddev
->reshape_position
== MaxSector
&&
3459 chunk_aligned_read(q
,bi
))
3462 logical_sector
= bi
->bi_sector
& ~((sector_t
)STRIPE_SECTORS
-1);
3463 last_sector
= bi
->bi_sector
+ (bi
->bi_size
>>9);
3465 bi
->bi_phys_segments
= 1; /* over-loaded to count active stripes */
3467 for (;logical_sector
< last_sector
; logical_sector
+= STRIPE_SECTORS
) {
3469 int disks
, data_disks
;
3472 prepare_to_wait(&conf
->wait_for_overlap
, &w
, TASK_UNINTERRUPTIBLE
);
3473 if (likely(conf
->expand_progress
== MaxSector
))
3474 disks
= conf
->raid_disks
;
3476 /* spinlock is needed as expand_progress may be
3477 * 64bit on a 32bit platform, and so it might be
3478 * possible to see a half-updated value
3479 * Ofcourse expand_progress could change after
3480 * the lock is dropped, so once we get a reference
3481 * to the stripe that we think it is, we will have
3484 spin_lock_irq(&conf
->device_lock
);
3485 disks
= conf
->raid_disks
;
3486 if (logical_sector
>= conf
->expand_progress
)
3487 disks
= conf
->previous_raid_disks
;
3489 if (logical_sector
>= conf
->expand_lo
) {
3490 spin_unlock_irq(&conf
->device_lock
);
3495 spin_unlock_irq(&conf
->device_lock
);
3497 data_disks
= disks
- conf
->max_degraded
;
3499 new_sector
= raid5_compute_sector(logical_sector
, disks
, data_disks
,
3500 &dd_idx
, &pd_idx
, conf
);
3501 pr_debug("raid5: make_request, sector %llu logical %llu\n",
3502 (unsigned long long)new_sector
,
3503 (unsigned long long)logical_sector
);
3505 sh
= get_active_stripe(conf
, new_sector
, disks
, pd_idx
, (bi
->bi_rw
&RWA_MASK
));
3507 if (unlikely(conf
->expand_progress
!= MaxSector
)) {
3508 /* expansion might have moved on while waiting for a
3509 * stripe, so we must do the range check again.
3510 * Expansion could still move past after this
3511 * test, but as we are holding a reference to
3512 * 'sh', we know that if that happens,
3513 * STRIPE_EXPANDING will get set and the expansion
3514 * won't proceed until we finish with the stripe.
3517 spin_lock_irq(&conf
->device_lock
);
3518 if (logical_sector
< conf
->expand_progress
&&
3519 disks
== conf
->previous_raid_disks
)
3520 /* mismatch, need to try again */
3522 spin_unlock_irq(&conf
->device_lock
);
3528 /* FIXME what if we get a false positive because these
3529 * are being updated.
3531 if (logical_sector
>= mddev
->suspend_lo
&&
3532 logical_sector
< mddev
->suspend_hi
) {
3538 if (test_bit(STRIPE_EXPANDING
, &sh
->state
) ||
3539 !add_stripe_bio(sh
, bi
, dd_idx
, (bi
->bi_rw
&RW_MASK
))) {
3540 /* Stripe is busy expanding or
3541 * add failed due to overlap. Flush everything
3544 raid5_unplug_device(mddev
->queue
);
3549 finish_wait(&conf
->wait_for_overlap
, &w
);
3550 handle_stripe(sh
, NULL
);
3553 /* cannot get stripe for read-ahead, just give-up */
3554 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
3555 finish_wait(&conf
->wait_for_overlap
, &w
);
3560 spin_lock_irq(&conf
->device_lock
);
3561 remaining
= --bi
->bi_phys_segments
;
3562 spin_unlock_irq(&conf
->device_lock
);
3563 if (remaining
== 0) {
3566 md_write_end(mddev
);
3569 test_bit(BIO_UPTODATE
, &bi
->bi_flags
)
3575 static sector_t
reshape_request(mddev_t
*mddev
, sector_t sector_nr
, int *skipped
)
3577 /* reshaping is quite different to recovery/resync so it is
3578 * handled quite separately ... here.
3580 * On each call to sync_request, we gather one chunk worth of
3581 * destination stripes and flag them as expanding.
3582 * Then we find all the source stripes and request reads.
3583 * As the reads complete, handle_stripe will copy the data
3584 * into the destination stripe and release that stripe.
3586 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
3587 struct stripe_head
*sh
;
3589 sector_t first_sector
, last_sector
;
3590 int raid_disks
= conf
->previous_raid_disks
;
3591 int data_disks
= raid_disks
- conf
->max_degraded
;
3592 int new_data_disks
= conf
->raid_disks
- conf
->max_degraded
;
3595 sector_t writepos
, safepos
, gap
;
3597 if (sector_nr
== 0 &&
3598 conf
->expand_progress
!= 0) {
3599 /* restarting in the middle, skip the initial sectors */
3600 sector_nr
= conf
->expand_progress
;
3601 sector_div(sector_nr
, new_data_disks
);
3606 /* we update the metadata when there is more than 3Meg
3607 * in the block range (that is rather arbitrary, should
3608 * probably be time based) or when the data about to be
3609 * copied would over-write the source of the data at
3610 * the front of the range.
3611 * i.e. one new_stripe forward from expand_progress new_maps
3612 * to after where expand_lo old_maps to
3614 writepos
= conf
->expand_progress
+
3615 conf
->chunk_size
/512*(new_data_disks
);
3616 sector_div(writepos
, new_data_disks
);
3617 safepos
= conf
->expand_lo
;
3618 sector_div(safepos
, data_disks
);
3619 gap
= conf
->expand_progress
- conf
->expand_lo
;
3621 if (writepos
>= safepos
||
3622 gap
> (new_data_disks
)*3000*2 /*3Meg*/) {
3623 /* Cannot proceed until we've updated the superblock... */
3624 wait_event(conf
->wait_for_overlap
,
3625 atomic_read(&conf
->reshape_stripes
)==0);
3626 mddev
->reshape_position
= conf
->expand_progress
;
3627 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3628 md_wakeup_thread(mddev
->thread
);
3629 wait_event(mddev
->sb_wait
, mddev
->flags
== 0 ||
3630 kthread_should_stop());
3631 spin_lock_irq(&conf
->device_lock
);
3632 conf
->expand_lo
= mddev
->reshape_position
;
3633 spin_unlock_irq(&conf
->device_lock
);
3634 wake_up(&conf
->wait_for_overlap
);
3637 for (i
=0; i
< conf
->chunk_size
/512; i
+= STRIPE_SECTORS
) {
3640 pd_idx
= stripe_to_pdidx(sector_nr
+i
, conf
, conf
->raid_disks
);
3641 sh
= get_active_stripe(conf
, sector_nr
+i
,
3642 conf
->raid_disks
, pd_idx
, 0);
3643 set_bit(STRIPE_EXPANDING
, &sh
->state
);
3644 atomic_inc(&conf
->reshape_stripes
);
3645 /* If any of this stripe is beyond the end of the old
3646 * array, then we need to zero those blocks
3648 for (j
=sh
->disks
; j
--;) {
3650 if (j
== sh
->pd_idx
)
3652 if (conf
->level
== 6 &&
3653 j
== raid6_next_disk(sh
->pd_idx
, sh
->disks
))
3655 s
= compute_blocknr(sh
, j
);
3656 if (s
< (mddev
->array_size
<<1)) {
3660 memset(page_address(sh
->dev
[j
].page
), 0, STRIPE_SIZE
);
3661 set_bit(R5_Expanded
, &sh
->dev
[j
].flags
);
3662 set_bit(R5_UPTODATE
, &sh
->dev
[j
].flags
);
3665 set_bit(STRIPE_EXPAND_READY
, &sh
->state
);
3666 set_bit(STRIPE_HANDLE
, &sh
->state
);
3670 spin_lock_irq(&conf
->device_lock
);
3671 conf
->expand_progress
= (sector_nr
+ i
) * new_data_disks
;
3672 spin_unlock_irq(&conf
->device_lock
);
3673 /* Ok, those stripe are ready. We can start scheduling
3674 * reads on the source stripes.
3675 * The source stripes are determined by mapping the first and last
3676 * block on the destination stripes.
3679 raid5_compute_sector(sector_nr
*(new_data_disks
),
3680 raid_disks
, data_disks
,
3681 &dd_idx
, &pd_idx
, conf
);
3683 raid5_compute_sector((sector_nr
+conf
->chunk_size
/512)
3684 *(new_data_disks
) -1,
3685 raid_disks
, data_disks
,
3686 &dd_idx
, &pd_idx
, conf
);
3687 if (last_sector
>= (mddev
->size
<<1))
3688 last_sector
= (mddev
->size
<<1)-1;
3689 while (first_sector
<= last_sector
) {
3690 pd_idx
= stripe_to_pdidx(first_sector
, conf
,
3691 conf
->previous_raid_disks
);
3692 sh
= get_active_stripe(conf
, first_sector
,
3693 conf
->previous_raid_disks
, pd_idx
, 0);
3694 set_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
3695 set_bit(STRIPE_HANDLE
, &sh
->state
);
3697 first_sector
+= STRIPE_SECTORS
;
3699 return conf
->chunk_size
>>9;
3702 /* FIXME go_faster isn't used */
3703 static inline sector_t
sync_request(mddev_t
*mddev
, sector_t sector_nr
, int *skipped
, int go_faster
)
3705 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
3706 struct stripe_head
*sh
;
3708 int raid_disks
= conf
->raid_disks
;
3709 sector_t max_sector
= mddev
->size
<< 1;
3711 int still_degraded
= 0;
3714 if (sector_nr
>= max_sector
) {
3715 /* just being told to finish up .. nothing much to do */
3716 unplug_slaves(mddev
);
3717 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)) {
3722 if (mddev
->curr_resync
< max_sector
) /* aborted */
3723 bitmap_end_sync(mddev
->bitmap
, mddev
->curr_resync
,
3725 else /* completed sync */
3727 bitmap_close_sync(mddev
->bitmap
);
3732 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
3733 return reshape_request(mddev
, sector_nr
, skipped
);
3735 /* if there is too many failed drives and we are trying
3736 * to resync, then assert that we are finished, because there is
3737 * nothing we can do.
3739 if (mddev
->degraded
>= conf
->max_degraded
&&
3740 test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
3741 sector_t rv
= (mddev
->size
<< 1) - sector_nr
;
3745 if (!bitmap_start_sync(mddev
->bitmap
, sector_nr
, &sync_blocks
, 1) &&
3746 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
) &&
3747 !conf
->fullsync
&& sync_blocks
>= STRIPE_SECTORS
) {
3748 /* we can skip this block, and probably more */
3749 sync_blocks
/= STRIPE_SECTORS
;
3751 return sync_blocks
* STRIPE_SECTORS
; /* keep things rounded to whole stripes */
3754 pd_idx
= stripe_to_pdidx(sector_nr
, conf
, raid_disks
);
3755 sh
= get_active_stripe(conf
, sector_nr
, raid_disks
, pd_idx
, 1);
3757 sh
= get_active_stripe(conf
, sector_nr
, raid_disks
, pd_idx
, 0);
3758 /* make sure we don't swamp the stripe cache if someone else
3759 * is trying to get access
3761 schedule_timeout_uninterruptible(1);
3763 /* Need to check if array will still be degraded after recovery/resync
3764 * We don't need to check the 'failed' flag as when that gets set,
3767 for (i
=0; i
<mddev
->raid_disks
; i
++)
3768 if (conf
->disks
[i
].rdev
== NULL
)
3771 bitmap_start_sync(mddev
->bitmap
, sector_nr
, &sync_blocks
, still_degraded
);
3773 spin_lock(&sh
->lock
);
3774 set_bit(STRIPE_SYNCING
, &sh
->state
);
3775 clear_bit(STRIPE_INSYNC
, &sh
->state
);
3776 spin_unlock(&sh
->lock
);
3778 handle_stripe(sh
, NULL
);
3781 return STRIPE_SECTORS
;
3784 static int retry_aligned_read(raid5_conf_t
*conf
, struct bio
*raid_bio
)
3786 /* We may not be able to submit a whole bio at once as there
3787 * may not be enough stripe_heads available.
3788 * We cannot pre-allocate enough stripe_heads as we may need
3789 * more than exist in the cache (if we allow ever large chunks).
3790 * So we do one stripe head at a time and record in
3791 * ->bi_hw_segments how many have been done.
3793 * We *know* that this entire raid_bio is in one chunk, so
3794 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
3796 struct stripe_head
*sh
;
3798 sector_t sector
, logical_sector
, last_sector
;
3803 logical_sector
= raid_bio
->bi_sector
& ~((sector_t
)STRIPE_SECTORS
-1);
3804 sector
= raid5_compute_sector( logical_sector
,
3806 conf
->raid_disks
- conf
->max_degraded
,
3810 last_sector
= raid_bio
->bi_sector
+ (raid_bio
->bi_size
>>9);
3812 for (; logical_sector
< last_sector
;
3813 logical_sector
+= STRIPE_SECTORS
,
3814 sector
+= STRIPE_SECTORS
,
3817 if (scnt
< raid_bio
->bi_hw_segments
)
3818 /* already done this stripe */
3821 sh
= get_active_stripe(conf
, sector
, conf
->raid_disks
, pd_idx
, 1);
3824 /* failed to get a stripe - must wait */
3825 raid_bio
->bi_hw_segments
= scnt
;
3826 conf
->retry_read_aligned
= raid_bio
;
3830 set_bit(R5_ReadError
, &sh
->dev
[dd_idx
].flags
);
3831 if (!add_stripe_bio(sh
, raid_bio
, dd_idx
, 0)) {
3833 raid_bio
->bi_hw_segments
= scnt
;
3834 conf
->retry_read_aligned
= raid_bio
;
3838 handle_stripe(sh
, NULL
);
3842 spin_lock_irq(&conf
->device_lock
);
3843 remaining
= --raid_bio
->bi_phys_segments
;
3844 spin_unlock_irq(&conf
->device_lock
);
3845 if (remaining
== 0) {
3847 raid_bio
->bi_end_io(raid_bio
,
3848 test_bit(BIO_UPTODATE
, &raid_bio
->bi_flags
)
3851 if (atomic_dec_and_test(&conf
->active_aligned_reads
))
3852 wake_up(&conf
->wait_for_stripe
);
3859 * This is our raid5 kernel thread.
3861 * We scan the hash table for stripes which can be handled now.
3862 * During the scan, completed stripes are saved for us by the interrupt
3863 * handler, so that they will not have to wait for our next wakeup.
3865 static void raid5d (mddev_t
*mddev
)
3867 struct stripe_head
*sh
;
3868 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3871 pr_debug("+++ raid5d active\n");
3873 md_check_recovery(mddev
);
3876 spin_lock_irq(&conf
->device_lock
);
3878 struct list_head
*first
;
3881 if (conf
->seq_flush
!= conf
->seq_write
) {
3882 int seq
= conf
->seq_flush
;
3883 spin_unlock_irq(&conf
->device_lock
);
3884 bitmap_unplug(mddev
->bitmap
);
3885 spin_lock_irq(&conf
->device_lock
);
3886 conf
->seq_write
= seq
;
3887 activate_bit_delay(conf
);
3890 if (list_empty(&conf
->handle_list
) &&
3891 atomic_read(&conf
->preread_active_stripes
) < IO_THRESHOLD
&&
3892 !blk_queue_plugged(mddev
->queue
) &&
3893 !list_empty(&conf
->delayed_list
))
3894 raid5_activate_delayed(conf
);
3896 while ((bio
= remove_bio_from_retry(conf
))) {
3898 spin_unlock_irq(&conf
->device_lock
);
3899 ok
= retry_aligned_read(conf
, bio
);
3900 spin_lock_irq(&conf
->device_lock
);
3906 if (list_empty(&conf
->handle_list
)) {
3907 async_tx_issue_pending_all();
3911 first
= conf
->handle_list
.next
;
3912 sh
= list_entry(first
, struct stripe_head
, lru
);
3914 list_del_init(first
);
3915 atomic_inc(&sh
->count
);
3916 BUG_ON(atomic_read(&sh
->count
)!= 1);
3917 spin_unlock_irq(&conf
->device_lock
);
3920 handle_stripe(sh
, conf
->spare_page
);
3923 spin_lock_irq(&conf
->device_lock
);
3925 pr_debug("%d stripes handled\n", handled
);
3927 spin_unlock_irq(&conf
->device_lock
);
3929 unplug_slaves(mddev
);
3931 pr_debug("--- raid5d inactive\n");
3935 raid5_show_stripe_cache_size(mddev_t
*mddev
, char *page
)
3937 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3939 return sprintf(page
, "%d\n", conf
->max_nr_stripes
);
3945 raid5_store_stripe_cache_size(mddev_t
*mddev
, const char *page
, size_t len
)
3947 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3950 if (len
>= PAGE_SIZE
)
3955 new = simple_strtoul(page
, &end
, 10);
3956 if (!*page
|| (*end
&& *end
!= '\n') )
3958 if (new <= 16 || new > 32768)
3960 while (new < conf
->max_nr_stripes
) {
3961 if (drop_one_stripe(conf
))
3962 conf
->max_nr_stripes
--;
3966 md_allow_write(mddev
);
3967 while (new > conf
->max_nr_stripes
) {
3968 if (grow_one_stripe(conf
))
3969 conf
->max_nr_stripes
++;
3975 static struct md_sysfs_entry
3976 raid5_stripecache_size
= __ATTR(stripe_cache_size
, S_IRUGO
| S_IWUSR
,
3977 raid5_show_stripe_cache_size
,
3978 raid5_store_stripe_cache_size
);
3981 stripe_cache_active_show(mddev_t
*mddev
, char *page
)
3983 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3985 return sprintf(page
, "%d\n", atomic_read(&conf
->active_stripes
));
3990 static struct md_sysfs_entry
3991 raid5_stripecache_active
= __ATTR_RO(stripe_cache_active
);
3993 static struct attribute
*raid5_attrs
[] = {
3994 &raid5_stripecache_size
.attr
,
3995 &raid5_stripecache_active
.attr
,
3998 static struct attribute_group raid5_attrs_group
= {
4000 .attrs
= raid5_attrs
,
4003 static int run(mddev_t
*mddev
)
4006 int raid_disk
, memory
;
4008 struct disk_info
*disk
;
4009 struct list_head
*tmp
;
4010 int working_disks
= 0;
4012 if (mddev
->level
!= 5 && mddev
->level
!= 4 && mddev
->level
!= 6) {
4013 printk(KERN_ERR
"raid5: %s: raid level not set to 4/5/6 (%d)\n",
4014 mdname(mddev
), mddev
->level
);
4018 if (mddev
->reshape_position
!= MaxSector
) {
4019 /* Check that we can continue the reshape.
4020 * Currently only disks can change, it must
4021 * increase, and we must be past the point where
4022 * a stripe over-writes itself
4024 sector_t here_new
, here_old
;
4026 int max_degraded
= (mddev
->level
== 5 ? 1 : 2);
4028 if (mddev
->new_level
!= mddev
->level
||
4029 mddev
->new_layout
!= mddev
->layout
||
4030 mddev
->new_chunk
!= mddev
->chunk_size
) {
4031 printk(KERN_ERR
"raid5: %s: unsupported reshape "
4032 "required - aborting.\n",
4036 if (mddev
->delta_disks
<= 0) {
4037 printk(KERN_ERR
"raid5: %s: unsupported reshape "
4038 "(reduce disks) required - aborting.\n",
4042 old_disks
= mddev
->raid_disks
- mddev
->delta_disks
;
4043 /* reshape_position must be on a new-stripe boundary, and one
4044 * further up in new geometry must map after here in old
4047 here_new
= mddev
->reshape_position
;
4048 if (sector_div(here_new
, (mddev
->chunk_size
>>9)*
4049 (mddev
->raid_disks
- max_degraded
))) {
4050 printk(KERN_ERR
"raid5: reshape_position not "
4051 "on a stripe boundary\n");
4054 /* here_new is the stripe we will write to */
4055 here_old
= mddev
->reshape_position
;
4056 sector_div(here_old
, (mddev
->chunk_size
>>9)*
4057 (old_disks
-max_degraded
));
4058 /* here_old is the first stripe that we might need to read
4060 if (here_new
>= here_old
) {
4061 /* Reading from the same stripe as writing to - bad */
4062 printk(KERN_ERR
"raid5: reshape_position too early for "
4063 "auto-recovery - aborting.\n");
4066 printk(KERN_INFO
"raid5: reshape will continue\n");
4067 /* OK, we should be able to continue; */
4071 mddev
->private = kzalloc(sizeof (raid5_conf_t
), GFP_KERNEL
);
4072 if ((conf
= mddev
->private) == NULL
)
4074 if (mddev
->reshape_position
== MaxSector
) {
4075 conf
->previous_raid_disks
= conf
->raid_disks
= mddev
->raid_disks
;
4077 conf
->raid_disks
= mddev
->raid_disks
;
4078 conf
->previous_raid_disks
= mddev
->raid_disks
- mddev
->delta_disks
;
4081 conf
->disks
= kzalloc(conf
->raid_disks
* sizeof(struct disk_info
),
4086 conf
->mddev
= mddev
;
4088 if ((conf
->stripe_hashtbl
= kzalloc(PAGE_SIZE
, GFP_KERNEL
)) == NULL
)
4091 if (mddev
->level
== 6) {
4092 conf
->spare_page
= alloc_page(GFP_KERNEL
);
4093 if (!conf
->spare_page
)
4096 spin_lock_init(&conf
->device_lock
);
4097 init_waitqueue_head(&conf
->wait_for_stripe
);
4098 init_waitqueue_head(&conf
->wait_for_overlap
);
4099 INIT_LIST_HEAD(&conf
->handle_list
);
4100 INIT_LIST_HEAD(&conf
->delayed_list
);
4101 INIT_LIST_HEAD(&conf
->bitmap_list
);
4102 INIT_LIST_HEAD(&conf
->inactive_list
);
4103 atomic_set(&conf
->active_stripes
, 0);
4104 atomic_set(&conf
->preread_active_stripes
, 0);
4105 atomic_set(&conf
->active_aligned_reads
, 0);
4107 pr_debug("raid5: run(%s) called.\n", mdname(mddev
));
4109 ITERATE_RDEV(mddev
,rdev
,tmp
) {
4110 raid_disk
= rdev
->raid_disk
;
4111 if (raid_disk
>= conf
->raid_disks
4114 disk
= conf
->disks
+ raid_disk
;
4118 if (test_bit(In_sync
, &rdev
->flags
)) {
4119 char b
[BDEVNAME_SIZE
];
4120 printk(KERN_INFO
"raid5: device %s operational as raid"
4121 " disk %d\n", bdevname(rdev
->bdev
,b
),
4128 * 0 for a fully functional array, 1 or 2 for a degraded array.
4130 mddev
->degraded
= conf
->raid_disks
- working_disks
;
4131 conf
->mddev
= mddev
;
4132 conf
->chunk_size
= mddev
->chunk_size
;
4133 conf
->level
= mddev
->level
;
4134 if (conf
->level
== 6)
4135 conf
->max_degraded
= 2;
4137 conf
->max_degraded
= 1;
4138 conf
->algorithm
= mddev
->layout
;
4139 conf
->max_nr_stripes
= NR_STRIPES
;
4140 conf
->expand_progress
= mddev
->reshape_position
;
4142 /* device size must be a multiple of chunk size */
4143 mddev
->size
&= ~(mddev
->chunk_size
/1024 -1);
4144 mddev
->resync_max_sectors
= mddev
->size
<< 1;
4146 if (conf
->level
== 6 && conf
->raid_disks
< 4) {
4147 printk(KERN_ERR
"raid6: not enough configured devices for %s (%d, minimum 4)\n",
4148 mdname(mddev
), conf
->raid_disks
);
4151 if (!conf
->chunk_size
|| conf
->chunk_size
% 4) {
4152 printk(KERN_ERR
"raid5: invalid chunk size %d for %s\n",
4153 conf
->chunk_size
, mdname(mddev
));
4156 if (conf
->algorithm
> ALGORITHM_RIGHT_SYMMETRIC
) {
4158 "raid5: unsupported parity algorithm %d for %s\n",
4159 conf
->algorithm
, mdname(mddev
));
4162 if (mddev
->degraded
> conf
->max_degraded
) {
4163 printk(KERN_ERR
"raid5: not enough operational devices for %s"
4164 " (%d/%d failed)\n",
4165 mdname(mddev
), mddev
->degraded
, conf
->raid_disks
);
4169 if (mddev
->degraded
> 0 &&
4170 mddev
->recovery_cp
!= MaxSector
) {
4171 if (mddev
->ok_start_degraded
)
4173 "raid5: starting dirty degraded array: %s"
4174 "- data corruption possible.\n",
4178 "raid5: cannot start dirty degraded array for %s\n",
4185 mddev
->thread
= md_register_thread(raid5d
, mddev
, "%s_raid5");
4186 if (!mddev
->thread
) {
4188 "raid5: couldn't allocate thread for %s\n",
4193 memory
= conf
->max_nr_stripes
* (sizeof(struct stripe_head
) +
4194 conf
->raid_disks
* ((sizeof(struct bio
) + PAGE_SIZE
))) / 1024;
4195 if (grow_stripes(conf
, conf
->max_nr_stripes
)) {
4197 "raid5: couldn't allocate %dkB for buffers\n", memory
);
4198 shrink_stripes(conf
);
4199 md_unregister_thread(mddev
->thread
);
4202 printk(KERN_INFO
"raid5: allocated %dkB for %s\n",
4203 memory
, mdname(mddev
));
4205 if (mddev
->degraded
== 0)
4206 printk("raid5: raid level %d set %s active with %d out of %d"
4207 " devices, algorithm %d\n", conf
->level
, mdname(mddev
),
4208 mddev
->raid_disks
-mddev
->degraded
, mddev
->raid_disks
,
4211 printk(KERN_ALERT
"raid5: raid level %d set %s active with %d"
4212 " out of %d devices, algorithm %d\n", conf
->level
,
4213 mdname(mddev
), mddev
->raid_disks
- mddev
->degraded
,
4214 mddev
->raid_disks
, conf
->algorithm
);
4216 print_raid5_conf(conf
);
4218 if (conf
->expand_progress
!= MaxSector
) {
4219 printk("...ok start reshape thread\n");
4220 conf
->expand_lo
= conf
->expand_progress
;
4221 atomic_set(&conf
->reshape_stripes
, 0);
4222 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4223 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4224 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
4225 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
4226 mddev
->sync_thread
= md_register_thread(md_do_sync
, mddev
,
4230 /* read-ahead size must cover two whole stripes, which is
4231 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
4234 int data_disks
= conf
->previous_raid_disks
- conf
->max_degraded
;
4235 int stripe
= data_disks
*
4236 (mddev
->chunk_size
/ PAGE_SIZE
);
4237 if (mddev
->queue
->backing_dev_info
.ra_pages
< 2 * stripe
)
4238 mddev
->queue
->backing_dev_info
.ra_pages
= 2 * stripe
;
4241 /* Ok, everything is just fine now */
4242 if (sysfs_create_group(&mddev
->kobj
, &raid5_attrs_group
))
4244 "raid5: failed to create sysfs attributes for %s\n",
4247 mddev
->queue
->unplug_fn
= raid5_unplug_device
;
4248 mddev
->queue
->backing_dev_info
.congested_data
= mddev
;
4249 mddev
->queue
->backing_dev_info
.congested_fn
= raid5_congested
;
4251 mddev
->array_size
= mddev
->size
* (conf
->previous_raid_disks
-
4252 conf
->max_degraded
);
4254 blk_queue_merge_bvec(mddev
->queue
, raid5_mergeable_bvec
);
4259 print_raid5_conf(conf
);
4260 safe_put_page(conf
->spare_page
);
4262 kfree(conf
->stripe_hashtbl
);
4265 mddev
->private = NULL
;
4266 printk(KERN_ALERT
"raid5: failed to run raid set %s\n", mdname(mddev
));
4272 static int stop(mddev_t
*mddev
)
4274 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
4276 md_unregister_thread(mddev
->thread
);
4277 mddev
->thread
= NULL
;
4278 shrink_stripes(conf
);
4279 kfree(conf
->stripe_hashtbl
);
4280 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
4281 blk_sync_queue(mddev
->queue
); /* the unplug fn references 'conf'*/
4282 sysfs_remove_group(&mddev
->kobj
, &raid5_attrs_group
);
4285 mddev
->private = NULL
;
4290 static void print_sh (struct seq_file
*seq
, struct stripe_head
*sh
)
4294 seq_printf(seq
, "sh %llu, pd_idx %d, state %ld.\n",
4295 (unsigned long long)sh
->sector
, sh
->pd_idx
, sh
->state
);
4296 seq_printf(seq
, "sh %llu, count %d.\n",
4297 (unsigned long long)sh
->sector
, atomic_read(&sh
->count
));
4298 seq_printf(seq
, "sh %llu, ", (unsigned long long)sh
->sector
);
4299 for (i
= 0; i
< sh
->disks
; i
++) {
4300 seq_printf(seq
, "(cache%d: %p %ld) ",
4301 i
, sh
->dev
[i
].page
, sh
->dev
[i
].flags
);
4303 seq_printf(seq
, "\n");
4306 static void printall (struct seq_file
*seq
, raid5_conf_t
*conf
)
4308 struct stripe_head
*sh
;
4309 struct hlist_node
*hn
;
4312 spin_lock_irq(&conf
->device_lock
);
4313 for (i
= 0; i
< NR_HASH
; i
++) {
4314 hlist_for_each_entry(sh
, hn
, &conf
->stripe_hashtbl
[i
], hash
) {
4315 if (sh
->raid_conf
!= conf
)
4320 spin_unlock_irq(&conf
->device_lock
);
4324 static void status (struct seq_file
*seq
, mddev_t
*mddev
)
4326 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
4329 seq_printf (seq
, " level %d, %dk chunk, algorithm %d", mddev
->level
, mddev
->chunk_size
>> 10, mddev
->layout
);
4330 seq_printf (seq
, " [%d/%d] [", conf
->raid_disks
, conf
->raid_disks
- mddev
->degraded
);
4331 for (i
= 0; i
< conf
->raid_disks
; i
++)
4332 seq_printf (seq
, "%s",
4333 conf
->disks
[i
].rdev
&&
4334 test_bit(In_sync
, &conf
->disks
[i
].rdev
->flags
) ? "U" : "_");
4335 seq_printf (seq
, "]");
4337 seq_printf (seq
, "\n");
4338 printall(seq
, conf
);
4342 static void print_raid5_conf (raid5_conf_t
*conf
)
4345 struct disk_info
*tmp
;
4347 printk("RAID5 conf printout:\n");
4349 printk("(conf==NULL)\n");
4352 printk(" --- rd:%d wd:%d\n", conf
->raid_disks
,
4353 conf
->raid_disks
- conf
->mddev
->degraded
);
4355 for (i
= 0; i
< conf
->raid_disks
; i
++) {
4356 char b
[BDEVNAME_SIZE
];
4357 tmp
= conf
->disks
+ i
;
4359 printk(" disk %d, o:%d, dev:%s\n",
4360 i
, !test_bit(Faulty
, &tmp
->rdev
->flags
),
4361 bdevname(tmp
->rdev
->bdev
,b
));
4365 static int raid5_spare_active(mddev_t
*mddev
)
4368 raid5_conf_t
*conf
= mddev
->private;
4369 struct disk_info
*tmp
;
4371 for (i
= 0; i
< conf
->raid_disks
; i
++) {
4372 tmp
= conf
->disks
+ i
;
4374 && !test_bit(Faulty
, &tmp
->rdev
->flags
)
4375 && !test_and_set_bit(In_sync
, &tmp
->rdev
->flags
)) {
4376 unsigned long flags
;
4377 spin_lock_irqsave(&conf
->device_lock
, flags
);
4379 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
4382 print_raid5_conf(conf
);
4386 static int raid5_remove_disk(mddev_t
*mddev
, int number
)
4388 raid5_conf_t
*conf
= mddev
->private;
4391 struct disk_info
*p
= conf
->disks
+ number
;
4393 print_raid5_conf(conf
);
4396 if (test_bit(In_sync
, &rdev
->flags
) ||
4397 atomic_read(&rdev
->nr_pending
)) {
4403 if (atomic_read(&rdev
->nr_pending
)) {
4404 /* lost the race, try later */
4411 print_raid5_conf(conf
);
4415 static int raid5_add_disk(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
4417 raid5_conf_t
*conf
= mddev
->private;
4420 struct disk_info
*p
;
4422 if (mddev
->degraded
> conf
->max_degraded
)
4423 /* no point adding a device */
4427 * find the disk ... but prefer rdev->saved_raid_disk
4430 if (rdev
->saved_raid_disk
>= 0 &&
4431 conf
->disks
[rdev
->saved_raid_disk
].rdev
== NULL
)
4432 disk
= rdev
->saved_raid_disk
;
4435 for ( ; disk
< conf
->raid_disks
; disk
++)
4436 if ((p
=conf
->disks
+ disk
)->rdev
== NULL
) {
4437 clear_bit(In_sync
, &rdev
->flags
);
4438 rdev
->raid_disk
= disk
;
4440 if (rdev
->saved_raid_disk
!= disk
)
4442 rcu_assign_pointer(p
->rdev
, rdev
);
4445 print_raid5_conf(conf
);
4449 static int raid5_resize(mddev_t
*mddev
, sector_t sectors
)
4451 /* no resync is happening, and there is enough space
4452 * on all devices, so we can resize.
4453 * We need to make sure resync covers any new space.
4454 * If the array is shrinking we should possibly wait until
4455 * any io in the removed space completes, but it hardly seems
4458 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4460 sectors
&= ~((sector_t
)mddev
->chunk_size
/512 - 1);
4461 mddev
->array_size
= (sectors
* (mddev
->raid_disks
-conf
->max_degraded
))>>1;
4462 set_capacity(mddev
->gendisk
, mddev
->array_size
<< 1);
4464 if (sectors
/2 > mddev
->size
&& mddev
->recovery_cp
== MaxSector
) {
4465 mddev
->recovery_cp
= mddev
->size
<< 1;
4466 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4468 mddev
->size
= sectors
/2;
4469 mddev
->resync_max_sectors
= sectors
;
4473 #ifdef CONFIG_MD_RAID5_RESHAPE
4474 static int raid5_check_reshape(mddev_t
*mddev
)
4476 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4479 if (mddev
->delta_disks
< 0 ||
4480 mddev
->new_level
!= mddev
->level
)
4481 return -EINVAL
; /* Cannot shrink array or change level yet */
4482 if (mddev
->delta_disks
== 0)
4483 return 0; /* nothing to do */
4485 /* Can only proceed if there are plenty of stripe_heads.
4486 * We need a minimum of one full stripe,, and for sensible progress
4487 * it is best to have about 4 times that.
4488 * If we require 4 times, then the default 256 4K stripe_heads will
4489 * allow for chunk sizes up to 256K, which is probably OK.
4490 * If the chunk size is greater, user-space should request more
4491 * stripe_heads first.
4493 if ((mddev
->chunk_size
/ STRIPE_SIZE
) * 4 > conf
->max_nr_stripes
||
4494 (mddev
->new_chunk
/ STRIPE_SIZE
) * 4 > conf
->max_nr_stripes
) {
4495 printk(KERN_WARNING
"raid5: reshape: not enough stripes. Needed %lu\n",
4496 (mddev
->chunk_size
/ STRIPE_SIZE
)*4);
4500 err
= resize_stripes(conf
, conf
->raid_disks
+ mddev
->delta_disks
);
4504 if (mddev
->degraded
> conf
->max_degraded
)
4506 /* looks like we might be able to manage this */
4510 static int raid5_start_reshape(mddev_t
*mddev
)
4512 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4514 struct list_head
*rtmp
;
4516 int added_devices
= 0;
4517 unsigned long flags
;
4519 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4522 ITERATE_RDEV(mddev
, rdev
, rtmp
)
4523 if (rdev
->raid_disk
< 0 &&
4524 !test_bit(Faulty
, &rdev
->flags
))
4527 if (spares
- mddev
->degraded
< mddev
->delta_disks
- conf
->max_degraded
)
4528 /* Not enough devices even to make a degraded array
4533 atomic_set(&conf
->reshape_stripes
, 0);
4534 spin_lock_irq(&conf
->device_lock
);
4535 conf
->previous_raid_disks
= conf
->raid_disks
;
4536 conf
->raid_disks
+= mddev
->delta_disks
;
4537 conf
->expand_progress
= 0;
4538 conf
->expand_lo
= 0;
4539 spin_unlock_irq(&conf
->device_lock
);
4541 /* Add some new drives, as many as will fit.
4542 * We know there are enough to make the newly sized array work.
4544 ITERATE_RDEV(mddev
, rdev
, rtmp
)
4545 if (rdev
->raid_disk
< 0 &&
4546 !test_bit(Faulty
, &rdev
->flags
)) {
4547 if (raid5_add_disk(mddev
, rdev
)) {
4549 set_bit(In_sync
, &rdev
->flags
);
4551 rdev
->recovery_offset
= 0;
4552 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4553 if (sysfs_create_link(&mddev
->kobj
,
4556 "raid5: failed to create "
4557 " link %s for %s\n",
4563 spin_lock_irqsave(&conf
->device_lock
, flags
);
4564 mddev
->degraded
= (conf
->raid_disks
- conf
->previous_raid_disks
) - added_devices
;
4565 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
4566 mddev
->raid_disks
= conf
->raid_disks
;
4567 mddev
->reshape_position
= 0;
4568 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
4570 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4571 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4572 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
4573 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
4574 mddev
->sync_thread
= md_register_thread(md_do_sync
, mddev
,
4576 if (!mddev
->sync_thread
) {
4577 mddev
->recovery
= 0;
4578 spin_lock_irq(&conf
->device_lock
);
4579 mddev
->raid_disks
= conf
->raid_disks
= conf
->previous_raid_disks
;
4580 conf
->expand_progress
= MaxSector
;
4581 spin_unlock_irq(&conf
->device_lock
);
4584 md_wakeup_thread(mddev
->sync_thread
);
4585 md_new_event(mddev
);
4590 static void end_reshape(raid5_conf_t
*conf
)
4592 struct block_device
*bdev
;
4594 if (!test_bit(MD_RECOVERY_INTR
, &conf
->mddev
->recovery
)) {
4595 conf
->mddev
->array_size
= conf
->mddev
->size
*
4596 (conf
->raid_disks
- conf
->max_degraded
);
4597 set_capacity(conf
->mddev
->gendisk
, conf
->mddev
->array_size
<< 1);
4598 conf
->mddev
->changed
= 1;
4600 bdev
= bdget_disk(conf
->mddev
->gendisk
, 0);
4602 mutex_lock(&bdev
->bd_inode
->i_mutex
);
4603 i_size_write(bdev
->bd_inode
, (loff_t
)conf
->mddev
->array_size
<< 10);
4604 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
4607 spin_lock_irq(&conf
->device_lock
);
4608 conf
->expand_progress
= MaxSector
;
4609 spin_unlock_irq(&conf
->device_lock
);
4610 conf
->mddev
->reshape_position
= MaxSector
;
4612 /* read-ahead size must cover two whole stripes, which is
4613 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
4616 int data_disks
= conf
->previous_raid_disks
- conf
->max_degraded
;
4617 int stripe
= data_disks
*
4618 (conf
->mddev
->chunk_size
/ PAGE_SIZE
);
4619 if (conf
->mddev
->queue
->backing_dev_info
.ra_pages
< 2 * stripe
)
4620 conf
->mddev
->queue
->backing_dev_info
.ra_pages
= 2 * stripe
;
4625 static void raid5_quiesce(mddev_t
*mddev
, int state
)
4627 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4630 case 2: /* resume for a suspend */
4631 wake_up(&conf
->wait_for_overlap
);
4634 case 1: /* stop all writes */
4635 spin_lock_irq(&conf
->device_lock
);
4637 wait_event_lock_irq(conf
->wait_for_stripe
,
4638 atomic_read(&conf
->active_stripes
) == 0 &&
4639 atomic_read(&conf
->active_aligned_reads
) == 0,
4640 conf
->device_lock
, /* nothing */);
4641 spin_unlock_irq(&conf
->device_lock
);
4644 case 0: /* re-enable writes */
4645 spin_lock_irq(&conf
->device_lock
);
4647 wake_up(&conf
->wait_for_stripe
);
4648 wake_up(&conf
->wait_for_overlap
);
4649 spin_unlock_irq(&conf
->device_lock
);
4654 static struct mdk_personality raid6_personality
=
4658 .owner
= THIS_MODULE
,
4659 .make_request
= make_request
,
4663 .error_handler
= error
,
4664 .hot_add_disk
= raid5_add_disk
,
4665 .hot_remove_disk
= raid5_remove_disk
,
4666 .spare_active
= raid5_spare_active
,
4667 .sync_request
= sync_request
,
4668 .resize
= raid5_resize
,
4669 #ifdef CONFIG_MD_RAID5_RESHAPE
4670 .check_reshape
= raid5_check_reshape
,
4671 .start_reshape
= raid5_start_reshape
,
4673 .quiesce
= raid5_quiesce
,
4675 static struct mdk_personality raid5_personality
=
4679 .owner
= THIS_MODULE
,
4680 .make_request
= make_request
,
4684 .error_handler
= error
,
4685 .hot_add_disk
= raid5_add_disk
,
4686 .hot_remove_disk
= raid5_remove_disk
,
4687 .spare_active
= raid5_spare_active
,
4688 .sync_request
= sync_request
,
4689 .resize
= raid5_resize
,
4690 #ifdef CONFIG_MD_RAID5_RESHAPE
4691 .check_reshape
= raid5_check_reshape
,
4692 .start_reshape
= raid5_start_reshape
,
4694 .quiesce
= raid5_quiesce
,
4697 static struct mdk_personality raid4_personality
=
4701 .owner
= THIS_MODULE
,
4702 .make_request
= make_request
,
4706 .error_handler
= error
,
4707 .hot_add_disk
= raid5_add_disk
,
4708 .hot_remove_disk
= raid5_remove_disk
,
4709 .spare_active
= raid5_spare_active
,
4710 .sync_request
= sync_request
,
4711 .resize
= raid5_resize
,
4712 #ifdef CONFIG_MD_RAID5_RESHAPE
4713 .check_reshape
= raid5_check_reshape
,
4714 .start_reshape
= raid5_start_reshape
,
4716 .quiesce
= raid5_quiesce
,
4719 static int __init
raid5_init(void)
4723 e
= raid6_select_algo();
4726 register_md_personality(&raid6_personality
);
4727 register_md_personality(&raid5_personality
);
4728 register_md_personality(&raid4_personality
);
4732 static void raid5_exit(void)
4734 unregister_md_personality(&raid6_personality
);
4735 unregister_md_personality(&raid5_personality
);
4736 unregister_md_personality(&raid4_personality
);
4739 module_init(raid5_init
);
4740 module_exit(raid5_exit
);
4741 MODULE_LICENSE("GPL");
4742 MODULE_ALIAS("md-personality-4"); /* RAID5 */
4743 MODULE_ALIAS("md-raid5");
4744 MODULE_ALIAS("md-raid4");
4745 MODULE_ALIAS("md-level-5");
4746 MODULE_ALIAS("md-level-4");
4747 MODULE_ALIAS("md-personality-8"); /* RAID6 */
4748 MODULE_ALIAS("md-raid6");
4749 MODULE_ALIAS("md-level-6");
4751 /* This used to be two separate modules, they were: */
4752 MODULE_ALIAS("raid5");
4753 MODULE_ALIAS("raid6");