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
)
693 int disks
= sh
->disks
;
694 int pd_idx
= sh
->pd_idx
, i
;
696 /* check if prexor is active which means only process blocks
697 * that are part of a read-modify-write (Wantprexor)
699 int prexor
= test_bit(STRIPE_OP_PREXOR
, &sh
->ops
.pending
);
701 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
702 (unsigned long long)sh
->sector
);
704 for (i
= disks
; i
--; ) {
705 struct r5dev
*dev
= &sh
->dev
[i
];
710 if (prexor
) { /* rmw */
712 test_bit(R5_Wantprexor
, &dev
->flags
))
715 if (i
!= pd_idx
&& dev
->towrite
&&
716 test_bit(R5_LOCKED
, &dev
->flags
))
723 spin_lock(&sh
->lock
);
724 chosen
= dev
->towrite
;
726 BUG_ON(dev
->written
);
727 wbi
= dev
->written
= chosen
;
728 spin_unlock(&sh
->lock
);
730 while (wbi
&& wbi
->bi_sector
<
731 dev
->sector
+ STRIPE_SECTORS
) {
732 tx
= async_copy_data(1, wbi
, dev
->page
,
734 wbi
= r5_next_bio(wbi
, dev
->sector
);
742 static void ops_complete_postxor(void *stripe_head_ref
)
744 struct stripe_head
*sh
= stripe_head_ref
;
746 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
747 (unsigned long long)sh
->sector
);
749 set_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.complete
);
750 set_bit(STRIPE_HANDLE
, &sh
->state
);
754 static void ops_complete_write(void *stripe_head_ref
)
756 struct stripe_head
*sh
= stripe_head_ref
;
757 int disks
= sh
->disks
, i
, pd_idx
= sh
->pd_idx
;
759 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
760 (unsigned long long)sh
->sector
);
762 for (i
= disks
; i
--; ) {
763 struct r5dev
*dev
= &sh
->dev
[i
];
764 if (dev
->written
|| i
== pd_idx
)
765 set_bit(R5_UPTODATE
, &dev
->flags
);
768 set_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.complete
);
769 set_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.complete
);
771 set_bit(STRIPE_HANDLE
, &sh
->state
);
776 ops_run_postxor(struct stripe_head
*sh
, struct dma_async_tx_descriptor
*tx
)
778 /* kernel stack size limits the total number of disks */
779 int disks
= sh
->disks
;
780 struct page
*xor_srcs
[disks
];
782 int count
= 0, pd_idx
= sh
->pd_idx
, i
;
783 struct page
*xor_dest
;
784 int prexor
= test_bit(STRIPE_OP_PREXOR
, &sh
->ops
.pending
);
786 dma_async_tx_callback callback
;
788 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
789 (unsigned long long)sh
->sector
);
791 /* check if prexor is active which means only process blocks
792 * that are part of a read-modify-write (written)
795 xor_dest
= xor_srcs
[count
++] = sh
->dev
[pd_idx
].page
;
796 for (i
= disks
; i
--; ) {
797 struct r5dev
*dev
= &sh
->dev
[i
];
799 xor_srcs
[count
++] = dev
->page
;
802 xor_dest
= sh
->dev
[pd_idx
].page
;
803 for (i
= disks
; i
--; ) {
804 struct r5dev
*dev
= &sh
->dev
[i
];
806 xor_srcs
[count
++] = dev
->page
;
810 /* check whether this postxor is part of a write */
811 callback
= test_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.pending
) ?
812 ops_complete_write
: ops_complete_postxor
;
814 /* 1/ if we prexor'd then the dest is reused as a source
815 * 2/ if we did not prexor then we are redoing the parity
816 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
817 * for the synchronous xor case
819 flags
= ASYNC_TX_DEP_ACK
| ASYNC_TX_ACK
|
820 (prexor
? ASYNC_TX_XOR_DROP_DST
: ASYNC_TX_XOR_ZERO_DST
);
822 atomic_inc(&sh
->count
);
824 if (unlikely(count
== 1)) {
825 flags
&= ~(ASYNC_TX_XOR_DROP_DST
| ASYNC_TX_XOR_ZERO_DST
);
826 tx
= async_memcpy(xor_dest
, xor_srcs
[0], 0, 0, STRIPE_SIZE
,
827 flags
, tx
, callback
, sh
);
829 tx
= async_xor(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
,
830 flags
, tx
, callback
, sh
);
833 static void ops_complete_check(void *stripe_head_ref
)
835 struct stripe_head
*sh
= stripe_head_ref
;
836 int pd_idx
= sh
->pd_idx
;
838 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
839 (unsigned long long)sh
->sector
);
841 if (test_and_clear_bit(STRIPE_OP_MOD_DMA_CHECK
, &sh
->ops
.pending
) &&
842 sh
->ops
.zero_sum_result
== 0)
843 set_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
845 set_bit(STRIPE_OP_CHECK
, &sh
->ops
.complete
);
846 set_bit(STRIPE_HANDLE
, &sh
->state
);
850 static void ops_run_check(struct stripe_head
*sh
)
852 /* kernel stack size limits the total number of disks */
853 int disks
= sh
->disks
;
854 struct page
*xor_srcs
[disks
];
855 struct dma_async_tx_descriptor
*tx
;
857 int count
= 0, pd_idx
= sh
->pd_idx
, i
;
858 struct page
*xor_dest
= xor_srcs
[count
++] = sh
->dev
[pd_idx
].page
;
860 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
861 (unsigned long long)sh
->sector
);
863 for (i
= disks
; i
--; ) {
864 struct r5dev
*dev
= &sh
->dev
[i
];
866 xor_srcs
[count
++] = dev
->page
;
869 tx
= async_xor_zero_sum(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
,
870 &sh
->ops
.zero_sum_result
, 0, NULL
, NULL
, NULL
);
873 set_bit(STRIPE_OP_MOD_DMA_CHECK
, &sh
->ops
.pending
);
875 clear_bit(STRIPE_OP_MOD_DMA_CHECK
, &sh
->ops
.pending
);
877 atomic_inc(&sh
->count
);
878 tx
= async_trigger_callback(ASYNC_TX_DEP_ACK
| ASYNC_TX_ACK
, tx
,
879 ops_complete_check
, sh
);
882 static void raid5_run_ops(struct stripe_head
*sh
, unsigned long pending
)
884 int overlap_clear
= 0, i
, disks
= sh
->disks
;
885 struct dma_async_tx_descriptor
*tx
= NULL
;
887 if (test_bit(STRIPE_OP_BIOFILL
, &pending
)) {
892 if (test_bit(STRIPE_OP_COMPUTE_BLK
, &pending
))
893 tx
= ops_run_compute5(sh
, pending
);
895 if (test_bit(STRIPE_OP_PREXOR
, &pending
))
896 tx
= ops_run_prexor(sh
, tx
);
898 if (test_bit(STRIPE_OP_BIODRAIN
, &pending
)) {
899 tx
= ops_run_biodrain(sh
, tx
);
903 if (test_bit(STRIPE_OP_POSTXOR
, &pending
))
904 ops_run_postxor(sh
, tx
);
906 if (test_bit(STRIPE_OP_CHECK
, &pending
))
909 if (test_bit(STRIPE_OP_IO
, &pending
))
913 for (i
= disks
; i
--; ) {
914 struct r5dev
*dev
= &sh
->dev
[i
];
915 if (test_and_clear_bit(R5_Overlap
, &dev
->flags
))
916 wake_up(&sh
->raid_conf
->wait_for_overlap
);
920 static int grow_one_stripe(raid5_conf_t
*conf
)
922 struct stripe_head
*sh
;
923 sh
= kmem_cache_alloc(conf
->slab_cache
, GFP_KERNEL
);
926 memset(sh
, 0, sizeof(*sh
) + (conf
->raid_disks
-1)*sizeof(struct r5dev
));
927 sh
->raid_conf
= conf
;
928 spin_lock_init(&sh
->lock
);
930 if (grow_buffers(sh
, conf
->raid_disks
)) {
931 shrink_buffers(sh
, conf
->raid_disks
);
932 kmem_cache_free(conf
->slab_cache
, sh
);
935 sh
->disks
= conf
->raid_disks
;
936 /* we just created an active stripe so... */
937 atomic_set(&sh
->count
, 1);
938 atomic_inc(&conf
->active_stripes
);
939 INIT_LIST_HEAD(&sh
->lru
);
944 static int grow_stripes(raid5_conf_t
*conf
, int num
)
946 struct kmem_cache
*sc
;
947 int devs
= conf
->raid_disks
;
949 sprintf(conf
->cache_name
[0], "raid5-%s", mdname(conf
->mddev
));
950 sprintf(conf
->cache_name
[1], "raid5-%s-alt", mdname(conf
->mddev
));
951 conf
->active_name
= 0;
952 sc
= kmem_cache_create(conf
->cache_name
[conf
->active_name
],
953 sizeof(struct stripe_head
)+(devs
-1)*sizeof(struct r5dev
),
957 conf
->slab_cache
= sc
;
958 conf
->pool_size
= devs
;
960 if (!grow_one_stripe(conf
))
965 #ifdef CONFIG_MD_RAID5_RESHAPE
966 static int resize_stripes(raid5_conf_t
*conf
, int newsize
)
968 /* Make all the stripes able to hold 'newsize' devices.
969 * New slots in each stripe get 'page' set to a new page.
971 * This happens in stages:
972 * 1/ create a new kmem_cache and allocate the required number of
974 * 2/ gather all the old stripe_heads and tranfer the pages across
975 * to the new stripe_heads. This will have the side effect of
976 * freezing the array as once all stripe_heads have been collected,
977 * no IO will be possible. Old stripe heads are freed once their
978 * pages have been transferred over, and the old kmem_cache is
979 * freed when all stripes are done.
980 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
981 * we simple return a failre status - no need to clean anything up.
982 * 4/ allocate new pages for the new slots in the new stripe_heads.
983 * If this fails, we don't bother trying the shrink the
984 * stripe_heads down again, we just leave them as they are.
985 * As each stripe_head is processed the new one is released into
988 * Once step2 is started, we cannot afford to wait for a write,
989 * so we use GFP_NOIO allocations.
991 struct stripe_head
*osh
, *nsh
;
992 LIST_HEAD(newstripes
);
993 struct disk_info
*ndisks
;
995 struct kmem_cache
*sc
;
998 if (newsize
<= conf
->pool_size
)
999 return 0; /* never bother to shrink */
1001 md_allow_write(conf
->mddev
);
1004 sc
= kmem_cache_create(conf
->cache_name
[1-conf
->active_name
],
1005 sizeof(struct stripe_head
)+(newsize
-1)*sizeof(struct r5dev
),
1010 for (i
= conf
->max_nr_stripes
; i
; i
--) {
1011 nsh
= kmem_cache_alloc(sc
, GFP_KERNEL
);
1015 memset(nsh
, 0, sizeof(*nsh
) + (newsize
-1)*sizeof(struct r5dev
));
1017 nsh
->raid_conf
= conf
;
1018 spin_lock_init(&nsh
->lock
);
1020 list_add(&nsh
->lru
, &newstripes
);
1023 /* didn't get enough, give up */
1024 while (!list_empty(&newstripes
)) {
1025 nsh
= list_entry(newstripes
.next
, struct stripe_head
, lru
);
1026 list_del(&nsh
->lru
);
1027 kmem_cache_free(sc
, nsh
);
1029 kmem_cache_destroy(sc
);
1032 /* Step 2 - Must use GFP_NOIO now.
1033 * OK, we have enough stripes, start collecting inactive
1034 * stripes and copying them over
1036 list_for_each_entry(nsh
, &newstripes
, lru
) {
1037 spin_lock_irq(&conf
->device_lock
);
1038 wait_event_lock_irq(conf
->wait_for_stripe
,
1039 !list_empty(&conf
->inactive_list
),
1041 unplug_slaves(conf
->mddev
)
1043 osh
= get_free_stripe(conf
);
1044 spin_unlock_irq(&conf
->device_lock
);
1045 atomic_set(&nsh
->count
, 1);
1046 for(i
=0; i
<conf
->pool_size
; i
++)
1047 nsh
->dev
[i
].page
= osh
->dev
[i
].page
;
1048 for( ; i
<newsize
; i
++)
1049 nsh
->dev
[i
].page
= NULL
;
1050 kmem_cache_free(conf
->slab_cache
, osh
);
1052 kmem_cache_destroy(conf
->slab_cache
);
1055 * At this point, we are holding all the stripes so the array
1056 * is completely stalled, so now is a good time to resize
1059 ndisks
= kzalloc(newsize
* sizeof(struct disk_info
), GFP_NOIO
);
1061 for (i
=0; i
<conf
->raid_disks
; i
++)
1062 ndisks
[i
] = conf
->disks
[i
];
1064 conf
->disks
= ndisks
;
1068 /* Step 4, return new stripes to service */
1069 while(!list_empty(&newstripes
)) {
1070 nsh
= list_entry(newstripes
.next
, struct stripe_head
, lru
);
1071 list_del_init(&nsh
->lru
);
1072 for (i
=conf
->raid_disks
; i
< newsize
; i
++)
1073 if (nsh
->dev
[i
].page
== NULL
) {
1074 struct page
*p
= alloc_page(GFP_NOIO
);
1075 nsh
->dev
[i
].page
= p
;
1079 release_stripe(nsh
);
1081 /* critical section pass, GFP_NOIO no longer needed */
1083 conf
->slab_cache
= sc
;
1084 conf
->active_name
= 1-conf
->active_name
;
1085 conf
->pool_size
= newsize
;
1090 static int drop_one_stripe(raid5_conf_t
*conf
)
1092 struct stripe_head
*sh
;
1094 spin_lock_irq(&conf
->device_lock
);
1095 sh
= get_free_stripe(conf
);
1096 spin_unlock_irq(&conf
->device_lock
);
1099 BUG_ON(atomic_read(&sh
->count
));
1100 shrink_buffers(sh
, conf
->pool_size
);
1101 kmem_cache_free(conf
->slab_cache
, sh
);
1102 atomic_dec(&conf
->active_stripes
);
1106 static void shrink_stripes(raid5_conf_t
*conf
)
1108 while (drop_one_stripe(conf
))
1111 if (conf
->slab_cache
)
1112 kmem_cache_destroy(conf
->slab_cache
);
1113 conf
->slab_cache
= NULL
;
1116 static void raid5_end_read_request(struct bio
* bi
, int error
)
1118 struct stripe_head
*sh
= bi
->bi_private
;
1119 raid5_conf_t
*conf
= sh
->raid_conf
;
1120 int disks
= sh
->disks
, i
;
1121 int uptodate
= test_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1122 char b
[BDEVNAME_SIZE
];
1126 for (i
=0 ; i
<disks
; i
++)
1127 if (bi
== &sh
->dev
[i
].req
)
1130 pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n",
1131 (unsigned long long)sh
->sector
, i
, atomic_read(&sh
->count
),
1139 set_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
);
1140 if (test_bit(R5_ReadError
, &sh
->dev
[i
].flags
)) {
1141 rdev
= conf
->disks
[i
].rdev
;
1142 printk(KERN_INFO
"raid5:%s: read error corrected (%lu sectors at %llu on %s)\n",
1143 mdname(conf
->mddev
), STRIPE_SECTORS
,
1144 (unsigned long long)sh
->sector
+ rdev
->data_offset
,
1145 bdevname(rdev
->bdev
, b
));
1146 clear_bit(R5_ReadError
, &sh
->dev
[i
].flags
);
1147 clear_bit(R5_ReWrite
, &sh
->dev
[i
].flags
);
1149 if (atomic_read(&conf
->disks
[i
].rdev
->read_errors
))
1150 atomic_set(&conf
->disks
[i
].rdev
->read_errors
, 0);
1152 const char *bdn
= bdevname(conf
->disks
[i
].rdev
->bdev
, b
);
1154 rdev
= conf
->disks
[i
].rdev
;
1156 clear_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
);
1157 atomic_inc(&rdev
->read_errors
);
1158 if (conf
->mddev
->degraded
)
1159 printk(KERN_WARNING
"raid5:%s: read error not correctable (sector %llu on %s).\n",
1160 mdname(conf
->mddev
),
1161 (unsigned long long)sh
->sector
+ rdev
->data_offset
,
1163 else if (test_bit(R5_ReWrite
, &sh
->dev
[i
].flags
))
1165 printk(KERN_WARNING
"raid5:%s: read error NOT corrected!! (sector %llu on %s).\n",
1166 mdname(conf
->mddev
),
1167 (unsigned long long)sh
->sector
+ rdev
->data_offset
,
1169 else if (atomic_read(&rdev
->read_errors
)
1170 > conf
->max_nr_stripes
)
1172 "raid5:%s: Too many read errors, failing device %s.\n",
1173 mdname(conf
->mddev
), bdn
);
1177 set_bit(R5_ReadError
, &sh
->dev
[i
].flags
);
1179 clear_bit(R5_ReadError
, &sh
->dev
[i
].flags
);
1180 clear_bit(R5_ReWrite
, &sh
->dev
[i
].flags
);
1181 md_error(conf
->mddev
, rdev
);
1184 rdev_dec_pending(conf
->disks
[i
].rdev
, conf
->mddev
);
1185 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
1186 set_bit(STRIPE_HANDLE
, &sh
->state
);
1190 static void raid5_end_write_request (struct bio
*bi
, int error
)
1192 struct stripe_head
*sh
= bi
->bi_private
;
1193 raid5_conf_t
*conf
= sh
->raid_conf
;
1194 int disks
= sh
->disks
, i
;
1195 int uptodate
= test_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1197 for (i
=0 ; i
<disks
; i
++)
1198 if (bi
== &sh
->dev
[i
].req
)
1201 pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n",
1202 (unsigned long long)sh
->sector
, i
, atomic_read(&sh
->count
),
1210 md_error(conf
->mddev
, conf
->disks
[i
].rdev
);
1212 rdev_dec_pending(conf
->disks
[i
].rdev
, conf
->mddev
);
1214 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
1215 set_bit(STRIPE_HANDLE
, &sh
->state
);
1220 static sector_t
compute_blocknr(struct stripe_head
*sh
, int i
);
1222 static void raid5_build_block (struct stripe_head
*sh
, int i
)
1224 struct r5dev
*dev
= &sh
->dev
[i
];
1226 bio_init(&dev
->req
);
1227 dev
->req
.bi_io_vec
= &dev
->vec
;
1229 dev
->req
.bi_max_vecs
++;
1230 dev
->vec
.bv_page
= dev
->page
;
1231 dev
->vec
.bv_len
= STRIPE_SIZE
;
1232 dev
->vec
.bv_offset
= 0;
1234 dev
->req
.bi_sector
= sh
->sector
;
1235 dev
->req
.bi_private
= sh
;
1238 dev
->sector
= compute_blocknr(sh
, i
);
1241 static void error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1243 char b
[BDEVNAME_SIZE
];
1244 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
1245 pr_debug("raid5: error called\n");
1247 if (!test_bit(Faulty
, &rdev
->flags
)) {
1248 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
1249 if (test_and_clear_bit(In_sync
, &rdev
->flags
)) {
1250 unsigned long flags
;
1251 spin_lock_irqsave(&conf
->device_lock
, flags
);
1253 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1255 * if recovery was running, make sure it aborts.
1257 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
1259 set_bit(Faulty
, &rdev
->flags
);
1261 "raid5: Disk failure on %s, disabling device."
1262 " Operation continuing on %d devices\n",
1263 bdevname(rdev
->bdev
,b
), conf
->raid_disks
- mddev
->degraded
);
1268 * Input: a 'big' sector number,
1269 * Output: index of the data and parity disk, and the sector # in them.
1271 static sector_t
raid5_compute_sector(sector_t r_sector
, unsigned int raid_disks
,
1272 unsigned int data_disks
, unsigned int * dd_idx
,
1273 unsigned int * pd_idx
, raid5_conf_t
*conf
)
1276 unsigned long chunk_number
;
1277 unsigned int chunk_offset
;
1278 sector_t new_sector
;
1279 int sectors_per_chunk
= conf
->chunk_size
>> 9;
1281 /* First compute the information on this sector */
1284 * Compute the chunk number and the sector offset inside the chunk
1286 chunk_offset
= sector_div(r_sector
, sectors_per_chunk
);
1287 chunk_number
= r_sector
;
1288 BUG_ON(r_sector
!= chunk_number
);
1291 * Compute the stripe number
1293 stripe
= chunk_number
/ data_disks
;
1296 * Compute the data disk and parity disk indexes inside the stripe
1298 *dd_idx
= chunk_number
% data_disks
;
1301 * Select the parity disk based on the user selected algorithm.
1303 switch(conf
->level
) {
1305 *pd_idx
= data_disks
;
1308 switch (conf
->algorithm
) {
1309 case ALGORITHM_LEFT_ASYMMETRIC
:
1310 *pd_idx
= data_disks
- stripe
% raid_disks
;
1311 if (*dd_idx
>= *pd_idx
)
1314 case ALGORITHM_RIGHT_ASYMMETRIC
:
1315 *pd_idx
= stripe
% raid_disks
;
1316 if (*dd_idx
>= *pd_idx
)
1319 case ALGORITHM_LEFT_SYMMETRIC
:
1320 *pd_idx
= data_disks
- stripe
% raid_disks
;
1321 *dd_idx
= (*pd_idx
+ 1 + *dd_idx
) % raid_disks
;
1323 case ALGORITHM_RIGHT_SYMMETRIC
:
1324 *pd_idx
= stripe
% raid_disks
;
1325 *dd_idx
= (*pd_idx
+ 1 + *dd_idx
) % raid_disks
;
1328 printk(KERN_ERR
"raid5: unsupported algorithm %d\n",
1334 /**** FIX THIS ****/
1335 switch (conf
->algorithm
) {
1336 case ALGORITHM_LEFT_ASYMMETRIC
:
1337 *pd_idx
= raid_disks
- 1 - (stripe
% raid_disks
);
1338 if (*pd_idx
== raid_disks
-1)
1339 (*dd_idx
)++; /* Q D D D P */
1340 else if (*dd_idx
>= *pd_idx
)
1341 (*dd_idx
) += 2; /* D D P Q D */
1343 case ALGORITHM_RIGHT_ASYMMETRIC
:
1344 *pd_idx
= stripe
% raid_disks
;
1345 if (*pd_idx
== raid_disks
-1)
1346 (*dd_idx
)++; /* Q D D D P */
1347 else if (*dd_idx
>= *pd_idx
)
1348 (*dd_idx
) += 2; /* D D P Q D */
1350 case ALGORITHM_LEFT_SYMMETRIC
:
1351 *pd_idx
= raid_disks
- 1 - (stripe
% raid_disks
);
1352 *dd_idx
= (*pd_idx
+ 2 + *dd_idx
) % raid_disks
;
1354 case ALGORITHM_RIGHT_SYMMETRIC
:
1355 *pd_idx
= stripe
% raid_disks
;
1356 *dd_idx
= (*pd_idx
+ 2 + *dd_idx
) % raid_disks
;
1359 printk (KERN_CRIT
"raid6: unsupported algorithm %d\n",
1366 * Finally, compute the new sector number
1368 new_sector
= (sector_t
)stripe
* sectors_per_chunk
+ chunk_offset
;
1373 static sector_t
compute_blocknr(struct stripe_head
*sh
, int i
)
1375 raid5_conf_t
*conf
= sh
->raid_conf
;
1376 int raid_disks
= sh
->disks
;
1377 int data_disks
= raid_disks
- conf
->max_degraded
;
1378 sector_t new_sector
= sh
->sector
, check
;
1379 int sectors_per_chunk
= conf
->chunk_size
>> 9;
1382 int chunk_number
, dummy1
, dummy2
, dd_idx
= i
;
1386 chunk_offset
= sector_div(new_sector
, sectors_per_chunk
);
1387 stripe
= new_sector
;
1388 BUG_ON(new_sector
!= stripe
);
1390 if (i
== sh
->pd_idx
)
1392 switch(conf
->level
) {
1395 switch (conf
->algorithm
) {
1396 case ALGORITHM_LEFT_ASYMMETRIC
:
1397 case ALGORITHM_RIGHT_ASYMMETRIC
:
1401 case ALGORITHM_LEFT_SYMMETRIC
:
1402 case ALGORITHM_RIGHT_SYMMETRIC
:
1405 i
-= (sh
->pd_idx
+ 1);
1408 printk(KERN_ERR
"raid5: unsupported algorithm %d\n",
1413 if (i
== raid6_next_disk(sh
->pd_idx
, raid_disks
))
1414 return 0; /* It is the Q disk */
1415 switch (conf
->algorithm
) {
1416 case ALGORITHM_LEFT_ASYMMETRIC
:
1417 case ALGORITHM_RIGHT_ASYMMETRIC
:
1418 if (sh
->pd_idx
== raid_disks
-1)
1419 i
--; /* Q D D D P */
1420 else if (i
> sh
->pd_idx
)
1421 i
-= 2; /* D D P Q D */
1423 case ALGORITHM_LEFT_SYMMETRIC
:
1424 case ALGORITHM_RIGHT_SYMMETRIC
:
1425 if (sh
->pd_idx
== raid_disks
-1)
1426 i
--; /* Q D D D P */
1431 i
-= (sh
->pd_idx
+ 2);
1435 printk (KERN_CRIT
"raid6: unsupported algorithm %d\n",
1441 chunk_number
= stripe
* data_disks
+ i
;
1442 r_sector
= (sector_t
)chunk_number
* sectors_per_chunk
+ chunk_offset
;
1444 check
= raid5_compute_sector (r_sector
, raid_disks
, data_disks
, &dummy1
, &dummy2
, conf
);
1445 if (check
!= sh
->sector
|| dummy1
!= dd_idx
|| dummy2
!= sh
->pd_idx
) {
1446 printk(KERN_ERR
"compute_blocknr: map not correct\n");
1455 * Copy data between a page in the stripe cache, and one or more bion
1456 * The page could align with the middle of the bio, or there could be
1457 * several bion, each with several bio_vecs, which cover part of the page
1458 * Multiple bion are linked together on bi_next. There may be extras
1459 * at the end of this list. We ignore them.
1461 static void copy_data(int frombio
, struct bio
*bio
,
1465 char *pa
= page_address(page
);
1466 struct bio_vec
*bvl
;
1470 if (bio
->bi_sector
>= sector
)
1471 page_offset
= (signed)(bio
->bi_sector
- sector
) * 512;
1473 page_offset
= (signed)(sector
- bio
->bi_sector
) * -512;
1474 bio_for_each_segment(bvl
, bio
, i
) {
1475 int len
= bio_iovec_idx(bio
,i
)->bv_len
;
1479 if (page_offset
< 0) {
1480 b_offset
= -page_offset
;
1481 page_offset
+= b_offset
;
1485 if (len
> 0 && page_offset
+ len
> STRIPE_SIZE
)
1486 clen
= STRIPE_SIZE
- page_offset
;
1490 char *ba
= __bio_kmap_atomic(bio
, i
, KM_USER0
);
1492 memcpy(pa
+page_offset
, ba
+b_offset
, clen
);
1494 memcpy(ba
+b_offset
, pa
+page_offset
, clen
);
1495 __bio_kunmap_atomic(ba
, KM_USER0
);
1497 if (clen
< len
) /* hit end of page */
1503 #define check_xor() do { \
1504 if (count == MAX_XOR_BLOCKS) { \
1505 xor_blocks(count, STRIPE_SIZE, dest, ptr);\
1510 static void compute_parity6(struct stripe_head
*sh
, int method
)
1512 raid6_conf_t
*conf
= sh
->raid_conf
;
1513 int i
, pd_idx
= sh
->pd_idx
, qd_idx
, d0_idx
, disks
= sh
->disks
, count
;
1515 /**** FIX THIS: This could be very bad if disks is close to 256 ****/
1518 qd_idx
= raid6_next_disk(pd_idx
, disks
);
1519 d0_idx
= raid6_next_disk(qd_idx
, disks
);
1521 pr_debug("compute_parity, stripe %llu, method %d\n",
1522 (unsigned long long)sh
->sector
, method
);
1525 case READ_MODIFY_WRITE
:
1526 BUG(); /* READ_MODIFY_WRITE N/A for RAID-6 */
1527 case RECONSTRUCT_WRITE
:
1528 for (i
= disks
; i
-- ;)
1529 if ( i
!= pd_idx
&& i
!= qd_idx
&& sh
->dev
[i
].towrite
) {
1530 chosen
= sh
->dev
[i
].towrite
;
1531 sh
->dev
[i
].towrite
= NULL
;
1533 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
1534 wake_up(&conf
->wait_for_overlap
);
1536 BUG_ON(sh
->dev
[i
].written
);
1537 sh
->dev
[i
].written
= chosen
;
1541 BUG(); /* Not implemented yet */
1544 for (i
= disks
; i
--;)
1545 if (sh
->dev
[i
].written
) {
1546 sector_t sector
= sh
->dev
[i
].sector
;
1547 struct bio
*wbi
= sh
->dev
[i
].written
;
1548 while (wbi
&& wbi
->bi_sector
< sector
+ STRIPE_SECTORS
) {
1549 copy_data(1, wbi
, sh
->dev
[i
].page
, sector
);
1550 wbi
= r5_next_bio(wbi
, sector
);
1553 set_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
1554 set_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
);
1558 // case RECONSTRUCT_WRITE:
1559 // case CHECK_PARITY:
1560 // case UPDATE_PARITY:
1561 /* Note that unlike RAID-5, the ordering of the disks matters greatly. */
1562 /* FIX: Is this ordering of drives even remotely optimal? */
1566 ptrs
[count
++] = page_address(sh
->dev
[i
].page
);
1567 if (count
<= disks
-2 && !test_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
))
1568 printk("block %d/%d not uptodate on parity calc\n", i
,count
);
1569 i
= raid6_next_disk(i
, disks
);
1570 } while ( i
!= d0_idx
);
1574 raid6_call
.gen_syndrome(disks
, STRIPE_SIZE
, ptrs
);
1577 case RECONSTRUCT_WRITE
:
1578 set_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
1579 set_bit(R5_UPTODATE
, &sh
->dev
[qd_idx
].flags
);
1580 set_bit(R5_LOCKED
, &sh
->dev
[pd_idx
].flags
);
1581 set_bit(R5_LOCKED
, &sh
->dev
[qd_idx
].flags
);
1584 set_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
1585 set_bit(R5_UPTODATE
, &sh
->dev
[qd_idx
].flags
);
1591 /* Compute one missing block */
1592 static void compute_block_1(struct stripe_head
*sh
, int dd_idx
, int nozero
)
1594 int i
, count
, disks
= sh
->disks
;
1595 void *ptr
[MAX_XOR_BLOCKS
], *dest
, *p
;
1596 int pd_idx
= sh
->pd_idx
;
1597 int qd_idx
= raid6_next_disk(pd_idx
, disks
);
1599 pr_debug("compute_block_1, stripe %llu, idx %d\n",
1600 (unsigned long long)sh
->sector
, dd_idx
);
1602 if ( dd_idx
== qd_idx
) {
1603 /* We're actually computing the Q drive */
1604 compute_parity6(sh
, UPDATE_PARITY
);
1606 dest
= page_address(sh
->dev
[dd_idx
].page
);
1607 if (!nozero
) memset(dest
, 0, STRIPE_SIZE
);
1609 for (i
= disks
; i
--; ) {
1610 if (i
== dd_idx
|| i
== qd_idx
)
1612 p
= page_address(sh
->dev
[i
].page
);
1613 if (test_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
))
1616 printk("compute_block() %d, stripe %llu, %d"
1617 " not present\n", dd_idx
,
1618 (unsigned long long)sh
->sector
, i
);
1623 xor_blocks(count
, STRIPE_SIZE
, dest
, ptr
);
1624 if (!nozero
) set_bit(R5_UPTODATE
, &sh
->dev
[dd_idx
].flags
);
1625 else clear_bit(R5_UPTODATE
, &sh
->dev
[dd_idx
].flags
);
1629 /* Compute two missing blocks */
1630 static void compute_block_2(struct stripe_head
*sh
, int dd_idx1
, int dd_idx2
)
1632 int i
, count
, disks
= sh
->disks
;
1633 int pd_idx
= sh
->pd_idx
;
1634 int qd_idx
= raid6_next_disk(pd_idx
, disks
);
1635 int d0_idx
= raid6_next_disk(qd_idx
, disks
);
1638 /* faila and failb are disk numbers relative to d0_idx */
1639 /* pd_idx become disks-2 and qd_idx become disks-1 */
1640 faila
= (dd_idx1
< d0_idx
) ? dd_idx1
+(disks
-d0_idx
) : dd_idx1
-d0_idx
;
1641 failb
= (dd_idx2
< d0_idx
) ? dd_idx2
+(disks
-d0_idx
) : dd_idx2
-d0_idx
;
1643 BUG_ON(faila
== failb
);
1644 if ( failb
< faila
) { int tmp
= faila
; faila
= failb
; failb
= tmp
; }
1646 pr_debug("compute_block_2, stripe %llu, idx %d,%d (%d,%d)\n",
1647 (unsigned long long)sh
->sector
, dd_idx1
, dd_idx2
, faila
, failb
);
1649 if ( failb
== disks
-1 ) {
1650 /* Q disk is one of the missing disks */
1651 if ( faila
== disks
-2 ) {
1652 /* Missing P+Q, just recompute */
1653 compute_parity6(sh
, UPDATE_PARITY
);
1656 /* We're missing D+Q; recompute D from P */
1657 compute_block_1(sh
, (dd_idx1
== qd_idx
) ? dd_idx2
: dd_idx1
, 0);
1658 compute_parity6(sh
, UPDATE_PARITY
); /* Is this necessary? */
1663 /* We're missing D+P or D+D; build pointer table */
1665 /**** FIX THIS: This could be very bad if disks is close to 256 ****/
1671 ptrs
[count
++] = page_address(sh
->dev
[i
].page
);
1672 i
= raid6_next_disk(i
, disks
);
1673 if (i
!= dd_idx1
&& i
!= dd_idx2
&&
1674 !test_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
))
1675 printk("compute_2 with missing block %d/%d\n", count
, i
);
1676 } while ( i
!= d0_idx
);
1678 if ( failb
== disks
-2 ) {
1679 /* We're missing D+P. */
1680 raid6_datap_recov(disks
, STRIPE_SIZE
, faila
, ptrs
);
1682 /* We're missing D+D. */
1683 raid6_2data_recov(disks
, STRIPE_SIZE
, faila
, failb
, ptrs
);
1686 /* Both the above update both missing blocks */
1687 set_bit(R5_UPTODATE
, &sh
->dev
[dd_idx1
].flags
);
1688 set_bit(R5_UPTODATE
, &sh
->dev
[dd_idx2
].flags
);
1693 handle_write_operations5(struct stripe_head
*sh
, int rcw
, int expand
)
1695 int i
, pd_idx
= sh
->pd_idx
, disks
= sh
->disks
;
1699 /* if we are not expanding this is a proper write request, and
1700 * there will be bios with new data to be drained into the
1704 set_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.pending
);
1708 set_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
);
1711 for (i
= disks
; i
--; ) {
1712 struct r5dev
*dev
= &sh
->dev
[i
];
1715 set_bit(R5_LOCKED
, &dev
->flags
);
1717 clear_bit(R5_UPTODATE
, &dev
->flags
);
1722 BUG_ON(!(test_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
) ||
1723 test_bit(R5_Wantcompute
, &sh
->dev
[pd_idx
].flags
)));
1725 set_bit(STRIPE_OP_PREXOR
, &sh
->ops
.pending
);
1726 set_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.pending
);
1727 set_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
);
1731 for (i
= disks
; i
--; ) {
1732 struct r5dev
*dev
= &sh
->dev
[i
];
1736 /* For a read-modify write there may be blocks that are
1737 * locked for reading while others are ready to be
1738 * written so we distinguish these blocks by the
1742 (test_bit(R5_UPTODATE
, &dev
->flags
) ||
1743 test_bit(R5_Wantcompute
, &dev
->flags
))) {
1744 set_bit(R5_Wantprexor
, &dev
->flags
);
1745 set_bit(R5_LOCKED
, &dev
->flags
);
1746 clear_bit(R5_UPTODATE
, &dev
->flags
);
1752 /* keep the parity disk locked while asynchronous operations
1755 set_bit(R5_LOCKED
, &sh
->dev
[pd_idx
].flags
);
1756 clear_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
1759 pr_debug("%s: stripe %llu locked: %d pending: %lx\n",
1760 __FUNCTION__
, (unsigned long long)sh
->sector
,
1761 locked
, sh
->ops
.pending
);
1767 * Each stripe/dev can have one or more bion attached.
1768 * toread/towrite point to the first in a chain.
1769 * The bi_next chain must be in order.
1771 static int add_stripe_bio(struct stripe_head
*sh
, struct bio
*bi
, int dd_idx
, int forwrite
)
1774 raid5_conf_t
*conf
= sh
->raid_conf
;
1777 pr_debug("adding bh b#%llu to stripe s#%llu\n",
1778 (unsigned long long)bi
->bi_sector
,
1779 (unsigned long long)sh
->sector
);
1782 spin_lock(&sh
->lock
);
1783 spin_lock_irq(&conf
->device_lock
);
1785 bip
= &sh
->dev
[dd_idx
].towrite
;
1786 if (*bip
== NULL
&& sh
->dev
[dd_idx
].written
== NULL
)
1789 bip
= &sh
->dev
[dd_idx
].toread
;
1790 while (*bip
&& (*bip
)->bi_sector
< bi
->bi_sector
) {
1791 if ((*bip
)->bi_sector
+ ((*bip
)->bi_size
>> 9) > bi
->bi_sector
)
1793 bip
= & (*bip
)->bi_next
;
1795 if (*bip
&& (*bip
)->bi_sector
< bi
->bi_sector
+ ((bi
->bi_size
)>>9))
1798 BUG_ON(*bip
&& bi
->bi_next
&& (*bip
) != bi
->bi_next
);
1802 bi
->bi_phys_segments
++;
1803 spin_unlock_irq(&conf
->device_lock
);
1804 spin_unlock(&sh
->lock
);
1806 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
1807 (unsigned long long)bi
->bi_sector
,
1808 (unsigned long long)sh
->sector
, dd_idx
);
1810 if (conf
->mddev
->bitmap
&& firstwrite
) {
1811 bitmap_startwrite(conf
->mddev
->bitmap
, sh
->sector
,
1813 sh
->bm_seq
= conf
->seq_flush
+1;
1814 set_bit(STRIPE_BIT_DELAY
, &sh
->state
);
1818 /* check if page is covered */
1819 sector_t sector
= sh
->dev
[dd_idx
].sector
;
1820 for (bi
=sh
->dev
[dd_idx
].towrite
;
1821 sector
< sh
->dev
[dd_idx
].sector
+ STRIPE_SECTORS
&&
1822 bi
&& bi
->bi_sector
<= sector
;
1823 bi
= r5_next_bio(bi
, sh
->dev
[dd_idx
].sector
)) {
1824 if (bi
->bi_sector
+ (bi
->bi_size
>>9) >= sector
)
1825 sector
= bi
->bi_sector
+ (bi
->bi_size
>>9);
1827 if (sector
>= sh
->dev
[dd_idx
].sector
+ STRIPE_SECTORS
)
1828 set_bit(R5_OVERWRITE
, &sh
->dev
[dd_idx
].flags
);
1833 set_bit(R5_Overlap
, &sh
->dev
[dd_idx
].flags
);
1834 spin_unlock_irq(&conf
->device_lock
);
1835 spin_unlock(&sh
->lock
);
1839 static void end_reshape(raid5_conf_t
*conf
);
1841 static int page_is_zero(struct page
*p
)
1843 char *a
= page_address(p
);
1844 return ((*(u32
*)a
) == 0 &&
1845 memcmp(a
, a
+4, STRIPE_SIZE
-4)==0);
1848 static int stripe_to_pdidx(sector_t stripe
, raid5_conf_t
*conf
, int disks
)
1850 int sectors_per_chunk
= conf
->chunk_size
>> 9;
1852 int chunk_offset
= sector_div(stripe
, sectors_per_chunk
);
1854 raid5_compute_sector(stripe
* (disks
- conf
->max_degraded
)
1855 *sectors_per_chunk
+ chunk_offset
,
1856 disks
, disks
- conf
->max_degraded
,
1857 &dd_idx
, &pd_idx
, conf
);
1862 handle_requests_to_failed_array(raid5_conf_t
*conf
, struct stripe_head
*sh
,
1863 struct stripe_head_state
*s
, int disks
,
1864 struct bio
**return_bi
)
1867 for (i
= disks
; i
--; ) {
1871 if (test_bit(R5_ReadError
, &sh
->dev
[i
].flags
)) {
1874 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
1875 if (rdev
&& test_bit(In_sync
, &rdev
->flags
))
1876 /* multiple read failures in one stripe */
1877 md_error(conf
->mddev
, rdev
);
1880 spin_lock_irq(&conf
->device_lock
);
1881 /* fail all writes first */
1882 bi
= sh
->dev
[i
].towrite
;
1883 sh
->dev
[i
].towrite
= NULL
;
1889 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
1890 wake_up(&conf
->wait_for_overlap
);
1892 while (bi
&& bi
->bi_sector
<
1893 sh
->dev
[i
].sector
+ STRIPE_SECTORS
) {
1894 struct bio
*nextbi
= r5_next_bio(bi
, sh
->dev
[i
].sector
);
1895 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1896 if (--bi
->bi_phys_segments
== 0) {
1897 md_write_end(conf
->mddev
);
1898 bi
->bi_next
= *return_bi
;
1903 /* and fail all 'written' */
1904 bi
= sh
->dev
[i
].written
;
1905 sh
->dev
[i
].written
= NULL
;
1906 if (bi
) bitmap_end
= 1;
1907 while (bi
&& bi
->bi_sector
<
1908 sh
->dev
[i
].sector
+ STRIPE_SECTORS
) {
1909 struct bio
*bi2
= r5_next_bio(bi
, sh
->dev
[i
].sector
);
1910 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1911 if (--bi
->bi_phys_segments
== 0) {
1912 md_write_end(conf
->mddev
);
1913 bi
->bi_next
= *return_bi
;
1919 /* fail any reads if this device is non-operational and
1920 * the data has not reached the cache yet.
1922 if (!test_bit(R5_Wantfill
, &sh
->dev
[i
].flags
) &&
1923 (!test_bit(R5_Insync
, &sh
->dev
[i
].flags
) ||
1924 test_bit(R5_ReadError
, &sh
->dev
[i
].flags
))) {
1925 bi
= sh
->dev
[i
].toread
;
1926 sh
->dev
[i
].toread
= NULL
;
1927 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
1928 wake_up(&conf
->wait_for_overlap
);
1929 if (bi
) s
->to_read
--;
1930 while (bi
&& bi
->bi_sector
<
1931 sh
->dev
[i
].sector
+ STRIPE_SECTORS
) {
1932 struct bio
*nextbi
=
1933 r5_next_bio(bi
, sh
->dev
[i
].sector
);
1934 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1935 if (--bi
->bi_phys_segments
== 0) {
1936 bi
->bi_next
= *return_bi
;
1942 spin_unlock_irq(&conf
->device_lock
);
1944 bitmap_endwrite(conf
->mddev
->bitmap
, sh
->sector
,
1945 STRIPE_SECTORS
, 0, 0);
1950 /* __handle_issuing_new_read_requests5 - returns 0 if there are no more disks
1953 static int __handle_issuing_new_read_requests5(struct stripe_head
*sh
,
1954 struct stripe_head_state
*s
, int disk_idx
, int disks
)
1956 struct r5dev
*dev
= &sh
->dev
[disk_idx
];
1957 struct r5dev
*failed_dev
= &sh
->dev
[s
->failed_num
];
1959 /* don't schedule compute operations or reads on the parity block while
1960 * a check is in flight
1962 if ((disk_idx
== sh
->pd_idx
) &&
1963 test_bit(STRIPE_OP_CHECK
, &sh
->ops
.pending
))
1966 /* is the data in this block needed, and can we get it? */
1967 if (!test_bit(R5_LOCKED
, &dev
->flags
) &&
1968 !test_bit(R5_UPTODATE
, &dev
->flags
) && (dev
->toread
||
1969 (dev
->towrite
&& !test_bit(R5_OVERWRITE
, &dev
->flags
)) ||
1970 s
->syncing
|| s
->expanding
|| (s
->failed
&&
1971 (failed_dev
->toread
|| (failed_dev
->towrite
&&
1972 !test_bit(R5_OVERWRITE
, &failed_dev
->flags
)
1974 /* 1/ We would like to get this block, possibly by computing it,
1975 * but we might not be able to.
1977 * 2/ Since parity check operations potentially make the parity
1978 * block !uptodate it will need to be refreshed before any
1979 * compute operations on data disks are scheduled.
1981 * 3/ We hold off parity block re-reads until check operations
1984 if ((s
->uptodate
== disks
- 1) &&
1985 !test_bit(STRIPE_OP_CHECK
, &sh
->ops
.pending
)) {
1986 set_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
);
1987 set_bit(R5_Wantcompute
, &dev
->flags
);
1988 sh
->ops
.target
= disk_idx
;
1991 /* Careful: from this point on 'uptodate' is in the eye
1992 * of raid5_run_ops which services 'compute' operations
1993 * before writes. R5_Wantcompute flags a block that will
1994 * be R5_UPTODATE by the time it is needed for a
1995 * subsequent operation.
1998 return 0; /* uptodate + compute == disks */
1999 } else if ((s
->uptodate
< disks
- 1) &&
2000 test_bit(R5_Insync
, &dev
->flags
)) {
2001 /* Note: we hold off compute operations while checks are
2002 * in flight, but we still prefer 'compute' over 'read'
2003 * hence we only read if (uptodate < * disks-1)
2005 set_bit(R5_LOCKED
, &dev
->flags
);
2006 set_bit(R5_Wantread
, &dev
->flags
);
2007 if (!test_and_set_bit(STRIPE_OP_IO
, &sh
->ops
.pending
))
2010 pr_debug("Reading block %d (sync=%d)\n", disk_idx
,
2018 static void handle_issuing_new_read_requests5(struct stripe_head
*sh
,
2019 struct stripe_head_state
*s
, int disks
)
2023 /* Clear completed compute operations. Parity recovery
2024 * (STRIPE_OP_MOD_REPAIR_PD) implies a write-back which is handled
2025 * later on in this routine
2027 if (test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.complete
) &&
2028 !test_bit(STRIPE_OP_MOD_REPAIR_PD
, &sh
->ops
.pending
)) {
2029 clear_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.complete
);
2030 clear_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.ack
);
2031 clear_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
);
2034 /* look for blocks to read/compute, skip this if a compute
2035 * is already in flight, or if the stripe contents are in the
2036 * midst of changing due to a write
2038 if (!test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
) &&
2039 !test_bit(STRIPE_OP_PREXOR
, &sh
->ops
.pending
) &&
2040 !test_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
)) {
2041 for (i
= disks
; i
--; )
2042 if (__handle_issuing_new_read_requests5(
2043 sh
, s
, i
, disks
) == 0)
2046 set_bit(STRIPE_HANDLE
, &sh
->state
);
2049 static void handle_issuing_new_read_requests6(struct stripe_head
*sh
,
2050 struct stripe_head_state
*s
, struct r6_state
*r6s
,
2054 for (i
= disks
; i
--; ) {
2055 struct r5dev
*dev
= &sh
->dev
[i
];
2056 if (!test_bit(R5_LOCKED
, &dev
->flags
) &&
2057 !test_bit(R5_UPTODATE
, &dev
->flags
) &&
2058 (dev
->toread
|| (dev
->towrite
&&
2059 !test_bit(R5_OVERWRITE
, &dev
->flags
)) ||
2060 s
->syncing
|| s
->expanding
||
2062 (sh
->dev
[r6s
->failed_num
[0]].toread
||
2065 (sh
->dev
[r6s
->failed_num
[1]].toread
||
2067 /* we would like to get this block, possibly
2068 * by computing it, but we might not be able to
2070 if (s
->uptodate
== disks
-1) {
2071 pr_debug("Computing stripe %llu block %d\n",
2072 (unsigned long long)sh
->sector
, i
);
2073 compute_block_1(sh
, i
, 0);
2075 } else if ( s
->uptodate
== disks
-2 && s
->failed
>= 2 ) {
2076 /* Computing 2-failure is *very* expensive; only
2077 * do it if failed >= 2
2080 for (other
= disks
; other
--; ) {
2083 if (!test_bit(R5_UPTODATE
,
2084 &sh
->dev
[other
].flags
))
2088 pr_debug("Computing stripe %llu blocks %d,%d\n",
2089 (unsigned long long)sh
->sector
,
2091 compute_block_2(sh
, i
, other
);
2093 } else if (test_bit(R5_Insync
, &dev
->flags
)) {
2094 set_bit(R5_LOCKED
, &dev
->flags
);
2095 set_bit(R5_Wantread
, &dev
->flags
);
2097 pr_debug("Reading block %d (sync=%d)\n",
2102 set_bit(STRIPE_HANDLE
, &sh
->state
);
2106 /* handle_completed_write_requests
2107 * any written block on an uptodate or failed drive can be returned.
2108 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
2109 * never LOCKED, so we don't need to test 'failed' directly.
2111 static void handle_completed_write_requests(raid5_conf_t
*conf
,
2112 struct stripe_head
*sh
, int disks
, struct bio
**return_bi
)
2117 for (i
= disks
; i
--; )
2118 if (sh
->dev
[i
].written
) {
2120 if (!test_bit(R5_LOCKED
, &dev
->flags
) &&
2121 test_bit(R5_UPTODATE
, &dev
->flags
)) {
2122 /* We can return any write requests */
2123 struct bio
*wbi
, *wbi2
;
2125 pr_debug("Return write for disc %d\n", i
);
2126 spin_lock_irq(&conf
->device_lock
);
2128 dev
->written
= NULL
;
2129 while (wbi
&& wbi
->bi_sector
<
2130 dev
->sector
+ STRIPE_SECTORS
) {
2131 wbi2
= r5_next_bio(wbi
, dev
->sector
);
2132 if (--wbi
->bi_phys_segments
== 0) {
2133 md_write_end(conf
->mddev
);
2134 wbi
->bi_next
= *return_bi
;
2139 if (dev
->towrite
== NULL
)
2141 spin_unlock_irq(&conf
->device_lock
);
2143 bitmap_endwrite(conf
->mddev
->bitmap
,
2146 !test_bit(STRIPE_DEGRADED
, &sh
->state
),
2152 static void handle_issuing_new_write_requests5(raid5_conf_t
*conf
,
2153 struct stripe_head
*sh
, struct stripe_head_state
*s
, int disks
)
2155 int rmw
= 0, rcw
= 0, i
;
2156 for (i
= disks
; i
--; ) {
2157 /* would I have to read this buffer for read_modify_write */
2158 struct r5dev
*dev
= &sh
->dev
[i
];
2159 if ((dev
->towrite
|| i
== sh
->pd_idx
) &&
2160 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2161 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2162 test_bit(R5_Wantcompute
, &dev
->flags
))) {
2163 if (test_bit(R5_Insync
, &dev
->flags
))
2166 rmw
+= 2*disks
; /* cannot read it */
2168 /* Would I have to read this buffer for reconstruct_write */
2169 if (!test_bit(R5_OVERWRITE
, &dev
->flags
) && i
!= sh
->pd_idx
&&
2170 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2171 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2172 test_bit(R5_Wantcompute
, &dev
->flags
))) {
2173 if (test_bit(R5_Insync
, &dev
->flags
)) rcw
++;
2178 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
2179 (unsigned long long)sh
->sector
, rmw
, rcw
);
2180 set_bit(STRIPE_HANDLE
, &sh
->state
);
2181 if (rmw
< rcw
&& rmw
> 0)
2182 /* prefer read-modify-write, but need to get some data */
2183 for (i
= disks
; i
--; ) {
2184 struct r5dev
*dev
= &sh
->dev
[i
];
2185 if ((dev
->towrite
|| i
== sh
->pd_idx
) &&
2186 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2187 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2188 test_bit(R5_Wantcompute
, &dev
->flags
)) &&
2189 test_bit(R5_Insync
, &dev
->flags
)) {
2191 test_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2192 pr_debug("Read_old block "
2193 "%d for r-m-w\n", i
);
2194 set_bit(R5_LOCKED
, &dev
->flags
);
2195 set_bit(R5_Wantread
, &dev
->flags
);
2196 if (!test_and_set_bit(
2197 STRIPE_OP_IO
, &sh
->ops
.pending
))
2201 set_bit(STRIPE_DELAYED
, &sh
->state
);
2202 set_bit(STRIPE_HANDLE
, &sh
->state
);
2206 if (rcw
<= rmw
&& rcw
> 0)
2207 /* want reconstruct write, but need to get some data */
2208 for (i
= disks
; i
--; ) {
2209 struct r5dev
*dev
= &sh
->dev
[i
];
2210 if (!test_bit(R5_OVERWRITE
, &dev
->flags
) &&
2212 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2213 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2214 test_bit(R5_Wantcompute
, &dev
->flags
)) &&
2215 test_bit(R5_Insync
, &dev
->flags
)) {
2217 test_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2218 pr_debug("Read_old block "
2219 "%d for Reconstruct\n", i
);
2220 set_bit(R5_LOCKED
, &dev
->flags
);
2221 set_bit(R5_Wantread
, &dev
->flags
);
2222 if (!test_and_set_bit(
2223 STRIPE_OP_IO
, &sh
->ops
.pending
))
2227 set_bit(STRIPE_DELAYED
, &sh
->state
);
2228 set_bit(STRIPE_HANDLE
, &sh
->state
);
2232 /* now if nothing is locked, and if we have enough data,
2233 * we can start a write request
2235 /* since handle_stripe can be called at any time we need to handle the
2236 * case where a compute block operation has been submitted and then a
2237 * subsequent call wants to start a write request. raid5_run_ops only
2238 * handles the case where compute block and postxor are requested
2239 * simultaneously. If this is not the case then new writes need to be
2240 * held off until the compute completes.
2242 if ((s
->req_compute
||
2243 !test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
)) &&
2244 (s
->locked
== 0 && (rcw
== 0 || rmw
== 0) &&
2245 !test_bit(STRIPE_BIT_DELAY
, &sh
->state
)))
2246 s
->locked
+= handle_write_operations5(sh
, rcw
== 0, 0);
2249 static void handle_issuing_new_write_requests6(raid5_conf_t
*conf
,
2250 struct stripe_head
*sh
, struct stripe_head_state
*s
,
2251 struct r6_state
*r6s
, int disks
)
2253 int rcw
= 0, must_compute
= 0, pd_idx
= sh
->pd_idx
, i
;
2254 int qd_idx
= r6s
->qd_idx
;
2255 for (i
= disks
; i
--; ) {
2256 struct r5dev
*dev
= &sh
->dev
[i
];
2257 /* Would I have to read this buffer for reconstruct_write */
2258 if (!test_bit(R5_OVERWRITE
, &dev
->flags
)
2259 && i
!= pd_idx
&& i
!= qd_idx
2260 && (!test_bit(R5_LOCKED
, &dev
->flags
)
2262 !test_bit(R5_UPTODATE
, &dev
->flags
)) {
2263 if (test_bit(R5_Insync
, &dev
->flags
)) rcw
++;
2265 pr_debug("raid6: must_compute: "
2266 "disk %d flags=%#lx\n", i
, dev
->flags
);
2271 pr_debug("for sector %llu, rcw=%d, must_compute=%d\n",
2272 (unsigned long long)sh
->sector
, rcw
, must_compute
);
2273 set_bit(STRIPE_HANDLE
, &sh
->state
);
2276 /* want reconstruct write, but need to get some data */
2277 for (i
= disks
; i
--; ) {
2278 struct r5dev
*dev
= &sh
->dev
[i
];
2279 if (!test_bit(R5_OVERWRITE
, &dev
->flags
)
2280 && !(s
->failed
== 0 && (i
== pd_idx
|| i
== qd_idx
))
2281 && !test_bit(R5_LOCKED
, &dev
->flags
) &&
2282 !test_bit(R5_UPTODATE
, &dev
->flags
) &&
2283 test_bit(R5_Insync
, &dev
->flags
)) {
2285 test_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2286 pr_debug("Read_old stripe %llu "
2287 "block %d for Reconstruct\n",
2288 (unsigned long long)sh
->sector
, i
);
2289 set_bit(R5_LOCKED
, &dev
->flags
);
2290 set_bit(R5_Wantread
, &dev
->flags
);
2293 pr_debug("Request delayed stripe %llu "
2294 "block %d for Reconstruct\n",
2295 (unsigned long long)sh
->sector
, i
);
2296 set_bit(STRIPE_DELAYED
, &sh
->state
);
2297 set_bit(STRIPE_HANDLE
, &sh
->state
);
2301 /* now if nothing is locked, and if we have enough data, we can start a
2304 if (s
->locked
== 0 && rcw
== 0 &&
2305 !test_bit(STRIPE_BIT_DELAY
, &sh
->state
)) {
2306 if (must_compute
> 0) {
2307 /* We have failed blocks and need to compute them */
2308 switch (s
->failed
) {
2312 compute_block_1(sh
, r6s
->failed_num
[0], 0);
2315 compute_block_2(sh
, r6s
->failed_num
[0],
2316 r6s
->failed_num
[1]);
2318 default: /* This request should have been failed? */
2323 pr_debug("Computing parity for stripe %llu\n",
2324 (unsigned long long)sh
->sector
);
2325 compute_parity6(sh
, RECONSTRUCT_WRITE
);
2326 /* now every locked buffer is ready to be written */
2327 for (i
= disks
; i
--; )
2328 if (test_bit(R5_LOCKED
, &sh
->dev
[i
].flags
)) {
2329 pr_debug("Writing stripe %llu block %d\n",
2330 (unsigned long long)sh
->sector
, i
);
2332 set_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
);
2334 /* after a RECONSTRUCT_WRITE, the stripe MUST be in-sync */
2335 set_bit(STRIPE_INSYNC
, &sh
->state
);
2337 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2338 atomic_dec(&conf
->preread_active_stripes
);
2339 if (atomic_read(&conf
->preread_active_stripes
) <
2341 md_wakeup_thread(conf
->mddev
->thread
);
2346 static void handle_parity_checks5(raid5_conf_t
*conf
, struct stripe_head
*sh
,
2347 struct stripe_head_state
*s
, int disks
)
2349 set_bit(STRIPE_HANDLE
, &sh
->state
);
2350 /* Take one of the following actions:
2351 * 1/ start a check parity operation if (uptodate == disks)
2352 * 2/ finish a check parity operation and act on the result
2353 * 3/ skip to the writeback section if we previously
2354 * initiated a recovery operation
2356 if (s
->failed
== 0 &&
2357 !test_bit(STRIPE_OP_MOD_REPAIR_PD
, &sh
->ops
.pending
)) {
2358 if (!test_and_set_bit(STRIPE_OP_CHECK
, &sh
->ops
.pending
)) {
2359 BUG_ON(s
->uptodate
!= disks
);
2360 clear_bit(R5_UPTODATE
, &sh
->dev
[sh
->pd_idx
].flags
);
2364 test_and_clear_bit(STRIPE_OP_CHECK
, &sh
->ops
.complete
)) {
2365 clear_bit(STRIPE_OP_CHECK
, &sh
->ops
.ack
);
2366 clear_bit(STRIPE_OP_CHECK
, &sh
->ops
.pending
);
2368 if (sh
->ops
.zero_sum_result
== 0)
2369 /* parity is correct (on disc,
2370 * not in buffer any more)
2372 set_bit(STRIPE_INSYNC
, &sh
->state
);
2374 conf
->mddev
->resync_mismatches
+=
2377 MD_RECOVERY_CHECK
, &conf
->mddev
->recovery
))
2378 /* don't try to repair!! */
2379 set_bit(STRIPE_INSYNC
, &sh
->state
);
2381 set_bit(STRIPE_OP_COMPUTE_BLK
,
2383 set_bit(STRIPE_OP_MOD_REPAIR_PD
,
2385 set_bit(R5_Wantcompute
,
2386 &sh
->dev
[sh
->pd_idx
].flags
);
2387 sh
->ops
.target
= sh
->pd_idx
;
2395 /* check if we can clear a parity disk reconstruct */
2396 if (test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.complete
) &&
2397 test_bit(STRIPE_OP_MOD_REPAIR_PD
, &sh
->ops
.pending
)) {
2399 clear_bit(STRIPE_OP_MOD_REPAIR_PD
, &sh
->ops
.pending
);
2400 clear_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.complete
);
2401 clear_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.ack
);
2402 clear_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
);
2405 /* Wait for check parity and compute block operations to complete
2408 if (!test_bit(STRIPE_INSYNC
, &sh
->state
) &&
2409 !test_bit(STRIPE_OP_CHECK
, &sh
->ops
.pending
) &&
2410 !test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
)) {
2412 /* either failed parity check, or recovery is happening */
2414 s
->failed_num
= sh
->pd_idx
;
2415 dev
= &sh
->dev
[s
->failed_num
];
2416 BUG_ON(!test_bit(R5_UPTODATE
, &dev
->flags
));
2417 BUG_ON(s
->uptodate
!= disks
);
2419 set_bit(R5_LOCKED
, &dev
->flags
);
2420 set_bit(R5_Wantwrite
, &dev
->flags
);
2421 if (!test_and_set_bit(STRIPE_OP_IO
, &sh
->ops
.pending
))
2424 clear_bit(STRIPE_DEGRADED
, &sh
->state
);
2426 set_bit(STRIPE_INSYNC
, &sh
->state
);
2431 static void handle_parity_checks6(raid5_conf_t
*conf
, struct stripe_head
*sh
,
2432 struct stripe_head_state
*s
,
2433 struct r6_state
*r6s
, struct page
*tmp_page
,
2436 int update_p
= 0, update_q
= 0;
2438 int pd_idx
= sh
->pd_idx
;
2439 int qd_idx
= r6s
->qd_idx
;
2441 set_bit(STRIPE_HANDLE
, &sh
->state
);
2443 BUG_ON(s
->failed
> 2);
2444 BUG_ON(s
->uptodate
< disks
);
2445 /* Want to check and possibly repair P and Q.
2446 * However there could be one 'failed' device, in which
2447 * case we can only check one of them, possibly using the
2448 * other to generate missing data
2451 /* If !tmp_page, we cannot do the calculations,
2452 * but as we have set STRIPE_HANDLE, we will soon be called
2453 * by stripe_handle with a tmp_page - just wait until then.
2456 if (s
->failed
== r6s
->q_failed
) {
2457 /* The only possible failed device holds 'Q', so it
2458 * makes sense to check P (If anything else were failed,
2459 * we would have used P to recreate it).
2461 compute_block_1(sh
, pd_idx
, 1);
2462 if (!page_is_zero(sh
->dev
[pd_idx
].page
)) {
2463 compute_block_1(sh
, pd_idx
, 0);
2467 if (!r6s
->q_failed
&& s
->failed
< 2) {
2468 /* q is not failed, and we didn't use it to generate
2469 * anything, so it makes sense to check it
2471 memcpy(page_address(tmp_page
),
2472 page_address(sh
->dev
[qd_idx
].page
),
2474 compute_parity6(sh
, UPDATE_PARITY
);
2475 if (memcmp(page_address(tmp_page
),
2476 page_address(sh
->dev
[qd_idx
].page
),
2477 STRIPE_SIZE
) != 0) {
2478 clear_bit(STRIPE_INSYNC
, &sh
->state
);
2482 if (update_p
|| update_q
) {
2483 conf
->mddev
->resync_mismatches
+= STRIPE_SECTORS
;
2484 if (test_bit(MD_RECOVERY_CHECK
, &conf
->mddev
->recovery
))
2485 /* don't try to repair!! */
2486 update_p
= update_q
= 0;
2489 /* now write out any block on a failed drive,
2490 * or P or Q if they need it
2493 if (s
->failed
== 2) {
2494 dev
= &sh
->dev
[r6s
->failed_num
[1]];
2496 set_bit(R5_LOCKED
, &dev
->flags
);
2497 set_bit(R5_Wantwrite
, &dev
->flags
);
2499 if (s
->failed
>= 1) {
2500 dev
= &sh
->dev
[r6s
->failed_num
[0]];
2502 set_bit(R5_LOCKED
, &dev
->flags
);
2503 set_bit(R5_Wantwrite
, &dev
->flags
);
2507 dev
= &sh
->dev
[pd_idx
];
2509 set_bit(R5_LOCKED
, &dev
->flags
);
2510 set_bit(R5_Wantwrite
, &dev
->flags
);
2513 dev
= &sh
->dev
[qd_idx
];
2515 set_bit(R5_LOCKED
, &dev
->flags
);
2516 set_bit(R5_Wantwrite
, &dev
->flags
);
2518 clear_bit(STRIPE_DEGRADED
, &sh
->state
);
2520 set_bit(STRIPE_INSYNC
, &sh
->state
);
2524 static void handle_stripe_expansion(raid5_conf_t
*conf
, struct stripe_head
*sh
,
2525 struct r6_state
*r6s
)
2529 /* We have read all the blocks in this stripe and now we need to
2530 * copy some of them into a target stripe for expand.
2532 struct dma_async_tx_descriptor
*tx
= NULL
;
2533 clear_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
2534 for (i
= 0; i
< sh
->disks
; i
++)
2535 if (i
!= sh
->pd_idx
&& (!r6s
|| i
!= r6s
->qd_idx
)) {
2536 int dd_idx
, pd_idx
, j
;
2537 struct stripe_head
*sh2
;
2539 sector_t bn
= compute_blocknr(sh
, i
);
2540 sector_t s
= raid5_compute_sector(bn
, conf
->raid_disks
,
2542 conf
->max_degraded
, &dd_idx
,
2544 sh2
= get_active_stripe(conf
, s
, conf
->raid_disks
,
2547 /* so far only the early blocks of this stripe
2548 * have been requested. When later blocks
2549 * get requested, we will try again
2552 if (!test_bit(STRIPE_EXPANDING
, &sh2
->state
) ||
2553 test_bit(R5_Expanded
, &sh2
->dev
[dd_idx
].flags
)) {
2554 /* must have already done this block */
2555 release_stripe(sh2
);
2559 /* place all the copies on one channel */
2560 tx
= async_memcpy(sh2
->dev
[dd_idx
].page
,
2561 sh
->dev
[i
].page
, 0, 0, STRIPE_SIZE
,
2562 ASYNC_TX_DEP_ACK
, tx
, NULL
, NULL
);
2564 set_bit(R5_Expanded
, &sh2
->dev
[dd_idx
].flags
);
2565 set_bit(R5_UPTODATE
, &sh2
->dev
[dd_idx
].flags
);
2566 for (j
= 0; j
< conf
->raid_disks
; j
++)
2567 if (j
!= sh2
->pd_idx
&&
2568 (!r6s
|| j
!= raid6_next_disk(sh2
->pd_idx
,
2570 !test_bit(R5_Expanded
, &sh2
->dev
[j
].flags
))
2572 if (j
== conf
->raid_disks
) {
2573 set_bit(STRIPE_EXPAND_READY
, &sh2
->state
);
2574 set_bit(STRIPE_HANDLE
, &sh2
->state
);
2576 release_stripe(sh2
);
2579 /* done submitting copies, wait for them to complete */
2582 dma_wait_for_async_tx(tx
);
2587 * handle_stripe - do things to a stripe.
2589 * We lock the stripe and then examine the state of various bits
2590 * to see what needs to be done.
2592 * return some read request which now have data
2593 * return some write requests which are safely on disc
2594 * schedule a read on some buffers
2595 * schedule a write of some buffers
2596 * return confirmation of parity correctness
2598 * buffers are taken off read_list or write_list, and bh_cache buffers
2599 * get BH_Lock set before the stripe lock is released.
2603 static void handle_stripe5(struct stripe_head
*sh
)
2605 raid5_conf_t
*conf
= sh
->raid_conf
;
2606 int disks
= sh
->disks
, i
;
2607 struct bio
*return_bi
= NULL
;
2608 struct stripe_head_state s
;
2610 unsigned long pending
= 0;
2612 memset(&s
, 0, sizeof(s
));
2613 pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d "
2614 "ops=%lx:%lx:%lx\n", (unsigned long long)sh
->sector
, sh
->state
,
2615 atomic_read(&sh
->count
), sh
->pd_idx
,
2616 sh
->ops
.pending
, sh
->ops
.ack
, sh
->ops
.complete
);
2618 spin_lock(&sh
->lock
);
2619 clear_bit(STRIPE_HANDLE
, &sh
->state
);
2620 clear_bit(STRIPE_DELAYED
, &sh
->state
);
2622 s
.syncing
= test_bit(STRIPE_SYNCING
, &sh
->state
);
2623 s
.expanding
= test_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
2624 s
.expanded
= test_bit(STRIPE_EXPAND_READY
, &sh
->state
);
2625 /* Now to look around and see what can be done */
2628 for (i
=disks
; i
--; ) {
2630 struct r5dev
*dev
= &sh
->dev
[i
];
2631 clear_bit(R5_Insync
, &dev
->flags
);
2633 pr_debug("check %d: state 0x%lx toread %p read %p write %p "
2634 "written %p\n", i
, dev
->flags
, dev
->toread
, dev
->read
,
2635 dev
->towrite
, dev
->written
);
2637 /* maybe we can request a biofill operation
2639 * new wantfill requests are only permitted while
2640 * STRIPE_OP_BIOFILL is clear
2642 if (test_bit(R5_UPTODATE
, &dev
->flags
) && dev
->toread
&&
2643 !test_bit(STRIPE_OP_BIOFILL
, &sh
->ops
.pending
))
2644 set_bit(R5_Wantfill
, &dev
->flags
);
2646 /* now count some things */
2647 if (test_bit(R5_LOCKED
, &dev
->flags
)) s
.locked
++;
2648 if (test_bit(R5_UPTODATE
, &dev
->flags
)) s
.uptodate
++;
2649 if (test_bit(R5_Wantcompute
, &dev
->flags
)) s
.compute
++;
2651 if (test_bit(R5_Wantfill
, &dev
->flags
))
2653 else if (dev
->toread
)
2657 if (!test_bit(R5_OVERWRITE
, &dev
->flags
))
2662 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
2663 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)) {
2664 /* The ReadError flag will just be confusing now */
2665 clear_bit(R5_ReadError
, &dev
->flags
);
2666 clear_bit(R5_ReWrite
, &dev
->flags
);
2668 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)
2669 || test_bit(R5_ReadError
, &dev
->flags
)) {
2673 set_bit(R5_Insync
, &dev
->flags
);
2677 if (s
.to_fill
&& !test_and_set_bit(STRIPE_OP_BIOFILL
, &sh
->ops
.pending
))
2680 pr_debug("locked=%d uptodate=%d to_read=%d"
2681 " to_write=%d failed=%d failed_num=%d\n",
2682 s
.locked
, s
.uptodate
, s
.to_read
, s
.to_write
,
2683 s
.failed
, s
.failed_num
);
2684 /* check if the array has lost two devices and, if so, some requests might
2687 if (s
.failed
> 1 && s
.to_read
+s
.to_write
+s
.written
)
2688 handle_requests_to_failed_array(conf
, sh
, &s
, disks
,
2690 if (s
.failed
> 1 && s
.syncing
) {
2691 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,0);
2692 clear_bit(STRIPE_SYNCING
, &sh
->state
);
2696 /* might be able to return some write requests if the parity block
2697 * is safe, or on a failed drive
2699 dev
= &sh
->dev
[sh
->pd_idx
];
2701 ((test_bit(R5_Insync
, &dev
->flags
) &&
2702 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2703 test_bit(R5_UPTODATE
, &dev
->flags
)) ||
2704 (s
.failed
== 1 && s
.failed_num
== sh
->pd_idx
)))
2705 handle_completed_write_requests(conf
, sh
, disks
, &return_bi
);
2707 /* Now we might consider reading some blocks, either to check/generate
2708 * parity, or to satisfy requests
2709 * or to load a block that is being partially written.
2711 if (s
.to_read
|| s
.non_overwrite
||
2712 (s
.syncing
&& (s
.uptodate
+ s
.compute
< disks
)) || s
.expanding
||
2713 test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
))
2714 handle_issuing_new_read_requests5(sh
, &s
, disks
);
2716 /* Now we check to see if any write operations have recently
2720 /* leave prexor set until postxor is done, allows us to distinguish
2721 * a rmw from a rcw during biodrain
2723 if (test_bit(STRIPE_OP_PREXOR
, &sh
->ops
.complete
) &&
2724 test_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.complete
)) {
2726 clear_bit(STRIPE_OP_PREXOR
, &sh
->ops
.complete
);
2727 clear_bit(STRIPE_OP_PREXOR
, &sh
->ops
.ack
);
2728 clear_bit(STRIPE_OP_PREXOR
, &sh
->ops
.pending
);
2730 for (i
= disks
; i
--; )
2731 clear_bit(R5_Wantprexor
, &sh
->dev
[i
].flags
);
2734 /* if only POSTXOR is set then this is an 'expand' postxor */
2735 if (test_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.complete
) &&
2736 test_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.complete
)) {
2738 clear_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.complete
);
2739 clear_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.ack
);
2740 clear_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.pending
);
2742 clear_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.complete
);
2743 clear_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.ack
);
2744 clear_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
);
2746 /* All the 'written' buffers and the parity block are ready to
2747 * be written back to disk
2749 BUG_ON(!test_bit(R5_UPTODATE
, &sh
->dev
[sh
->pd_idx
].flags
));
2750 for (i
= disks
; i
--; ) {
2752 if (test_bit(R5_LOCKED
, &dev
->flags
) &&
2753 (i
== sh
->pd_idx
|| dev
->written
)) {
2754 pr_debug("Writing block %d\n", i
);
2755 set_bit(R5_Wantwrite
, &dev
->flags
);
2756 if (!test_and_set_bit(
2757 STRIPE_OP_IO
, &sh
->ops
.pending
))
2759 if (!test_bit(R5_Insync
, &dev
->flags
) ||
2760 (i
== sh
->pd_idx
&& s
.failed
== 0))
2761 set_bit(STRIPE_INSYNC
, &sh
->state
);
2764 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2765 atomic_dec(&conf
->preread_active_stripes
);
2766 if (atomic_read(&conf
->preread_active_stripes
) <
2768 md_wakeup_thread(conf
->mddev
->thread
);
2772 /* Now to consider new write requests and what else, if anything
2773 * should be read. We do not handle new writes when:
2774 * 1/ A 'write' operation (copy+xor) is already in flight.
2775 * 2/ A 'check' operation is in flight, as it may clobber the parity
2778 if (s
.to_write
&& !test_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
) &&
2779 !test_bit(STRIPE_OP_CHECK
, &sh
->ops
.pending
))
2780 handle_issuing_new_write_requests5(conf
, sh
, &s
, disks
);
2782 /* maybe we need to check and possibly fix the parity for this stripe
2783 * Any reads will already have been scheduled, so we just see if enough
2784 * data is available. The parity check is held off while parity
2785 * dependent operations are in flight.
2787 if ((s
.syncing
&& s
.locked
== 0 &&
2788 !test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
) &&
2789 !test_bit(STRIPE_INSYNC
, &sh
->state
)) ||
2790 test_bit(STRIPE_OP_CHECK
, &sh
->ops
.pending
) ||
2791 test_bit(STRIPE_OP_MOD_REPAIR_PD
, &sh
->ops
.pending
))
2792 handle_parity_checks5(conf
, sh
, &s
, disks
);
2794 if (s
.syncing
&& s
.locked
== 0 && test_bit(STRIPE_INSYNC
, &sh
->state
)) {
2795 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,1);
2796 clear_bit(STRIPE_SYNCING
, &sh
->state
);
2799 /* If the failed drive is just a ReadError, then we might need to progress
2800 * the repair/check process
2802 if (s
.failed
== 1 && !conf
->mddev
->ro
&&
2803 test_bit(R5_ReadError
, &sh
->dev
[s
.failed_num
].flags
)
2804 && !test_bit(R5_LOCKED
, &sh
->dev
[s
.failed_num
].flags
)
2805 && test_bit(R5_UPTODATE
, &sh
->dev
[s
.failed_num
].flags
)
2807 dev
= &sh
->dev
[s
.failed_num
];
2808 if (!test_bit(R5_ReWrite
, &dev
->flags
)) {
2809 set_bit(R5_Wantwrite
, &dev
->flags
);
2810 if (!test_and_set_bit(STRIPE_OP_IO
, &sh
->ops
.pending
))
2812 set_bit(R5_ReWrite
, &dev
->flags
);
2813 set_bit(R5_LOCKED
, &dev
->flags
);
2816 /* let's read it back */
2817 set_bit(R5_Wantread
, &dev
->flags
);
2818 if (!test_and_set_bit(STRIPE_OP_IO
, &sh
->ops
.pending
))
2820 set_bit(R5_LOCKED
, &dev
->flags
);
2825 /* Finish postxor operations initiated by the expansion
2828 if (test_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.complete
) &&
2829 !test_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.pending
)) {
2831 clear_bit(STRIPE_EXPANDING
, &sh
->state
);
2833 clear_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
);
2834 clear_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.ack
);
2835 clear_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.complete
);
2837 for (i
= conf
->raid_disks
; i
--; ) {
2838 set_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
);
2839 if (!test_and_set_bit(STRIPE_OP_IO
, &sh
->ops
.pending
))
2844 if (s
.expanded
&& test_bit(STRIPE_EXPANDING
, &sh
->state
) &&
2845 !test_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
)) {
2846 /* Need to write out all blocks after computing parity */
2847 sh
->disks
= conf
->raid_disks
;
2848 sh
->pd_idx
= stripe_to_pdidx(sh
->sector
, conf
,
2850 s
.locked
+= handle_write_operations5(sh
, 1, 1);
2851 } else if (s
.expanded
&&
2852 !test_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
)) {
2853 clear_bit(STRIPE_EXPAND_READY
, &sh
->state
);
2854 atomic_dec(&conf
->reshape_stripes
);
2855 wake_up(&conf
->wait_for_overlap
);
2856 md_done_sync(conf
->mddev
, STRIPE_SECTORS
, 1);
2859 if (s
.expanding
&& s
.locked
== 0)
2860 handle_stripe_expansion(conf
, sh
, NULL
);
2863 pending
= get_stripe_work(sh
);
2865 spin_unlock(&sh
->lock
);
2868 raid5_run_ops(sh
, pending
);
2870 return_io(return_bi
);
2874 static void handle_stripe6(struct stripe_head
*sh
, struct page
*tmp_page
)
2876 raid6_conf_t
*conf
= sh
->raid_conf
;
2877 int disks
= sh
->disks
;
2878 struct bio
*return_bi
= NULL
;
2879 int i
, pd_idx
= sh
->pd_idx
;
2880 struct stripe_head_state s
;
2881 struct r6_state r6s
;
2882 struct r5dev
*dev
, *pdev
, *qdev
;
2884 r6s
.qd_idx
= raid6_next_disk(pd_idx
, disks
);
2885 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
2886 "pd_idx=%d, qd_idx=%d\n",
2887 (unsigned long long)sh
->sector
, sh
->state
,
2888 atomic_read(&sh
->count
), pd_idx
, r6s
.qd_idx
);
2889 memset(&s
, 0, sizeof(s
));
2891 spin_lock(&sh
->lock
);
2892 clear_bit(STRIPE_HANDLE
, &sh
->state
);
2893 clear_bit(STRIPE_DELAYED
, &sh
->state
);
2895 s
.syncing
= test_bit(STRIPE_SYNCING
, &sh
->state
);
2896 s
.expanding
= test_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
2897 s
.expanded
= test_bit(STRIPE_EXPAND_READY
, &sh
->state
);
2898 /* Now to look around and see what can be done */
2900 /* clean-up completed biofill operations */
2901 if (test_bit(STRIPE_OP_BIOFILL
, &sh
->ops
.complete
)) {
2902 clear_bit(STRIPE_OP_BIOFILL
, &sh
->ops
.pending
);
2903 clear_bit(STRIPE_OP_BIOFILL
, &sh
->ops
.ack
);
2904 clear_bit(STRIPE_OP_BIOFILL
, &sh
->ops
.complete
);
2908 for (i
=disks
; i
--; ) {
2911 clear_bit(R5_Insync
, &dev
->flags
);
2913 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
2914 i
, dev
->flags
, dev
->toread
, dev
->towrite
, dev
->written
);
2915 /* maybe we can reply to a read */
2916 if (test_bit(R5_UPTODATE
, &dev
->flags
) && dev
->toread
) {
2917 struct bio
*rbi
, *rbi2
;
2918 pr_debug("Return read for disc %d\n", i
);
2919 spin_lock_irq(&conf
->device_lock
);
2922 if (test_and_clear_bit(R5_Overlap
, &dev
->flags
))
2923 wake_up(&conf
->wait_for_overlap
);
2924 spin_unlock_irq(&conf
->device_lock
);
2925 while (rbi
&& rbi
->bi_sector
< dev
->sector
+ STRIPE_SECTORS
) {
2926 copy_data(0, rbi
, dev
->page
, dev
->sector
);
2927 rbi2
= r5_next_bio(rbi
, dev
->sector
);
2928 spin_lock_irq(&conf
->device_lock
);
2929 if (--rbi
->bi_phys_segments
== 0) {
2930 rbi
->bi_next
= return_bi
;
2933 spin_unlock_irq(&conf
->device_lock
);
2938 /* now count some things */
2939 if (test_bit(R5_LOCKED
, &dev
->flags
)) s
.locked
++;
2940 if (test_bit(R5_UPTODATE
, &dev
->flags
)) s
.uptodate
++;
2947 if (!test_bit(R5_OVERWRITE
, &dev
->flags
))
2952 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
2953 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)) {
2954 /* The ReadError flag will just be confusing now */
2955 clear_bit(R5_ReadError
, &dev
->flags
);
2956 clear_bit(R5_ReWrite
, &dev
->flags
);
2958 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)
2959 || test_bit(R5_ReadError
, &dev
->flags
)) {
2961 r6s
.failed_num
[s
.failed
] = i
;
2964 set_bit(R5_Insync
, &dev
->flags
);
2967 pr_debug("locked=%d uptodate=%d to_read=%d"
2968 " to_write=%d failed=%d failed_num=%d,%d\n",
2969 s
.locked
, s
.uptodate
, s
.to_read
, s
.to_write
, s
.failed
,
2970 r6s
.failed_num
[0], r6s
.failed_num
[1]);
2971 /* check if the array has lost >2 devices and, if so, some requests
2972 * might need to be failed
2974 if (s
.failed
> 2 && s
.to_read
+s
.to_write
+s
.written
)
2975 handle_requests_to_failed_array(conf
, sh
, &s
, disks
,
2977 if (s
.failed
> 2 && s
.syncing
) {
2978 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,0);
2979 clear_bit(STRIPE_SYNCING
, &sh
->state
);
2984 * might be able to return some write requests if the parity blocks
2985 * are safe, or on a failed drive
2987 pdev
= &sh
->dev
[pd_idx
];
2988 r6s
.p_failed
= (s
.failed
>= 1 && r6s
.failed_num
[0] == pd_idx
)
2989 || (s
.failed
>= 2 && r6s
.failed_num
[1] == pd_idx
);
2990 qdev
= &sh
->dev
[r6s
.qd_idx
];
2991 r6s
.q_failed
= (s
.failed
>= 1 && r6s
.failed_num
[0] == r6s
.qd_idx
)
2992 || (s
.failed
>= 2 && r6s
.failed_num
[1] == r6s
.qd_idx
);
2995 ( r6s
.p_failed
|| ((test_bit(R5_Insync
, &pdev
->flags
)
2996 && !test_bit(R5_LOCKED
, &pdev
->flags
)
2997 && test_bit(R5_UPTODATE
, &pdev
->flags
)))) &&
2998 ( r6s
.q_failed
|| ((test_bit(R5_Insync
, &qdev
->flags
)
2999 && !test_bit(R5_LOCKED
, &qdev
->flags
)
3000 && test_bit(R5_UPTODATE
, &qdev
->flags
)))))
3001 handle_completed_write_requests(conf
, sh
, disks
, &return_bi
);
3003 /* Now we might consider reading some blocks, either to check/generate
3004 * parity, or to satisfy requests
3005 * or to load a block that is being partially written.
3007 if (s
.to_read
|| s
.non_overwrite
|| (s
.to_write
&& s
.failed
) ||
3008 (s
.syncing
&& (s
.uptodate
< disks
)) || s
.expanding
)
3009 handle_issuing_new_read_requests6(sh
, &s
, &r6s
, disks
);
3011 /* now to consider writing and what else, if anything should be read */
3013 handle_issuing_new_write_requests6(conf
, sh
, &s
, &r6s
, disks
);
3015 /* maybe we need to check and possibly fix the parity for this stripe
3016 * Any reads will already have been scheduled, so we just see if enough
3019 if (s
.syncing
&& s
.locked
== 0 && !test_bit(STRIPE_INSYNC
, &sh
->state
))
3020 handle_parity_checks6(conf
, sh
, &s
, &r6s
, tmp_page
, disks
);
3022 if (s
.syncing
&& s
.locked
== 0 && test_bit(STRIPE_INSYNC
, &sh
->state
)) {
3023 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,1);
3024 clear_bit(STRIPE_SYNCING
, &sh
->state
);
3027 /* If the failed drives are just a ReadError, then we might need
3028 * to progress the repair/check process
3030 if (s
.failed
<= 2 && !conf
->mddev
->ro
)
3031 for (i
= 0; i
< s
.failed
; i
++) {
3032 dev
= &sh
->dev
[r6s
.failed_num
[i
]];
3033 if (test_bit(R5_ReadError
, &dev
->flags
)
3034 && !test_bit(R5_LOCKED
, &dev
->flags
)
3035 && test_bit(R5_UPTODATE
, &dev
->flags
)
3037 if (!test_bit(R5_ReWrite
, &dev
->flags
)) {
3038 set_bit(R5_Wantwrite
, &dev
->flags
);
3039 set_bit(R5_ReWrite
, &dev
->flags
);
3040 set_bit(R5_LOCKED
, &dev
->flags
);
3042 /* let's read it back */
3043 set_bit(R5_Wantread
, &dev
->flags
);
3044 set_bit(R5_LOCKED
, &dev
->flags
);
3049 if (s
.expanded
&& test_bit(STRIPE_EXPANDING
, &sh
->state
)) {
3050 /* Need to write out all blocks after computing P&Q */
3051 sh
->disks
= conf
->raid_disks
;
3052 sh
->pd_idx
= stripe_to_pdidx(sh
->sector
, conf
,
3054 compute_parity6(sh
, RECONSTRUCT_WRITE
);
3055 for (i
= conf
->raid_disks
; i
-- ; ) {
3056 set_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
3058 set_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
);
3060 clear_bit(STRIPE_EXPANDING
, &sh
->state
);
3061 } else if (s
.expanded
) {
3062 clear_bit(STRIPE_EXPAND_READY
, &sh
->state
);
3063 atomic_dec(&conf
->reshape_stripes
);
3064 wake_up(&conf
->wait_for_overlap
);
3065 md_done_sync(conf
->mddev
, STRIPE_SECTORS
, 1);
3068 if (s
.expanding
&& s
.locked
== 0)
3069 handle_stripe_expansion(conf
, sh
, &r6s
);
3071 spin_unlock(&sh
->lock
);
3073 return_io(return_bi
);
3075 for (i
=disks
; i
-- ;) {
3079 if (test_and_clear_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
))
3081 else if (test_and_clear_bit(R5_Wantread
, &sh
->dev
[i
].flags
))
3086 bi
= &sh
->dev
[i
].req
;
3090 bi
->bi_end_io
= raid5_end_write_request
;
3092 bi
->bi_end_io
= raid5_end_read_request
;
3095 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
3096 if (rdev
&& test_bit(Faulty
, &rdev
->flags
))
3099 atomic_inc(&rdev
->nr_pending
);
3103 if (s
.syncing
|| s
.expanding
|| s
.expanded
)
3104 md_sync_acct(rdev
->bdev
, STRIPE_SECTORS
);
3106 bi
->bi_bdev
= rdev
->bdev
;
3107 pr_debug("for %llu schedule op %ld on disc %d\n",
3108 (unsigned long long)sh
->sector
, bi
->bi_rw
, i
);
3109 atomic_inc(&sh
->count
);
3110 bi
->bi_sector
= sh
->sector
+ rdev
->data_offset
;
3111 bi
->bi_flags
= 1 << BIO_UPTODATE
;
3113 bi
->bi_max_vecs
= 1;
3115 bi
->bi_io_vec
= &sh
->dev
[i
].vec
;
3116 bi
->bi_io_vec
[0].bv_len
= STRIPE_SIZE
;
3117 bi
->bi_io_vec
[0].bv_offset
= 0;
3118 bi
->bi_size
= STRIPE_SIZE
;
3121 test_bit(R5_ReWrite
, &sh
->dev
[i
].flags
))
3122 atomic_add(STRIPE_SECTORS
, &rdev
->corrected_errors
);
3123 generic_make_request(bi
);
3126 set_bit(STRIPE_DEGRADED
, &sh
->state
);
3127 pr_debug("skip op %ld on disc %d for sector %llu\n",
3128 bi
->bi_rw
, i
, (unsigned long long)sh
->sector
);
3129 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
3130 set_bit(STRIPE_HANDLE
, &sh
->state
);
3135 static void handle_stripe(struct stripe_head
*sh
, struct page
*tmp_page
)
3137 if (sh
->raid_conf
->level
== 6)
3138 handle_stripe6(sh
, tmp_page
);
3145 static void raid5_activate_delayed(raid5_conf_t
*conf
)
3147 if (atomic_read(&conf
->preread_active_stripes
) < IO_THRESHOLD
) {
3148 while (!list_empty(&conf
->delayed_list
)) {
3149 struct list_head
*l
= conf
->delayed_list
.next
;
3150 struct stripe_head
*sh
;
3151 sh
= list_entry(l
, struct stripe_head
, lru
);
3153 clear_bit(STRIPE_DELAYED
, &sh
->state
);
3154 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
))
3155 atomic_inc(&conf
->preread_active_stripes
);
3156 list_add_tail(&sh
->lru
, &conf
->handle_list
);
3161 static void activate_bit_delay(raid5_conf_t
*conf
)
3163 /* device_lock is held */
3164 struct list_head head
;
3165 list_add(&head
, &conf
->bitmap_list
);
3166 list_del_init(&conf
->bitmap_list
);
3167 while (!list_empty(&head
)) {
3168 struct stripe_head
*sh
= list_entry(head
.next
, struct stripe_head
, lru
);
3169 list_del_init(&sh
->lru
);
3170 atomic_inc(&sh
->count
);
3171 __release_stripe(conf
, sh
);
3175 static void unplug_slaves(mddev_t
*mddev
)
3177 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3181 for (i
=0; i
<mddev
->raid_disks
; i
++) {
3182 mdk_rdev_t
*rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
3183 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
) && atomic_read(&rdev
->nr_pending
)) {
3184 struct request_queue
*r_queue
= bdev_get_queue(rdev
->bdev
);
3186 atomic_inc(&rdev
->nr_pending
);
3189 if (r_queue
->unplug_fn
)
3190 r_queue
->unplug_fn(r_queue
);
3192 rdev_dec_pending(rdev
, mddev
);
3199 static void raid5_unplug_device(struct request_queue
*q
)
3201 mddev_t
*mddev
= q
->queuedata
;
3202 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3203 unsigned long flags
;
3205 spin_lock_irqsave(&conf
->device_lock
, flags
);
3207 if (blk_remove_plug(q
)) {
3209 raid5_activate_delayed(conf
);
3211 md_wakeup_thread(mddev
->thread
);
3213 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
3215 unplug_slaves(mddev
);
3218 static int raid5_congested(void *data
, int bits
)
3220 mddev_t
*mddev
= data
;
3221 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3223 /* No difference between reads and writes. Just check
3224 * how busy the stripe_cache is
3226 if (conf
->inactive_blocked
)
3230 if (list_empty_careful(&conf
->inactive_list
))
3236 /* We want read requests to align with chunks where possible,
3237 * but write requests don't need to.
3239 static int raid5_mergeable_bvec(struct request_queue
*q
, struct bio
*bio
, struct bio_vec
*biovec
)
3241 mddev_t
*mddev
= q
->queuedata
;
3242 sector_t sector
= bio
->bi_sector
+ get_start_sect(bio
->bi_bdev
);
3244 unsigned int chunk_sectors
= mddev
->chunk_size
>> 9;
3245 unsigned int bio_sectors
= bio
->bi_size
>> 9;
3247 if (bio_data_dir(bio
) == WRITE
)
3248 return biovec
->bv_len
; /* always allow writes to be mergeable */
3250 max
= (chunk_sectors
- ((sector
& (chunk_sectors
- 1)) + bio_sectors
)) << 9;
3251 if (max
< 0) max
= 0;
3252 if (max
<= biovec
->bv_len
&& bio_sectors
== 0)
3253 return biovec
->bv_len
;
3259 static int in_chunk_boundary(mddev_t
*mddev
, struct bio
*bio
)
3261 sector_t sector
= bio
->bi_sector
+ get_start_sect(bio
->bi_bdev
);
3262 unsigned int chunk_sectors
= mddev
->chunk_size
>> 9;
3263 unsigned int bio_sectors
= bio
->bi_size
>> 9;
3265 return chunk_sectors
>=
3266 ((sector
& (chunk_sectors
- 1)) + bio_sectors
);
3270 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
3271 * later sampled by raid5d.
3273 static void add_bio_to_retry(struct bio
*bi
,raid5_conf_t
*conf
)
3275 unsigned long flags
;
3277 spin_lock_irqsave(&conf
->device_lock
, flags
);
3279 bi
->bi_next
= conf
->retry_read_aligned_list
;
3280 conf
->retry_read_aligned_list
= bi
;
3282 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
3283 md_wakeup_thread(conf
->mddev
->thread
);
3287 static struct bio
*remove_bio_from_retry(raid5_conf_t
*conf
)
3291 bi
= conf
->retry_read_aligned
;
3293 conf
->retry_read_aligned
= NULL
;
3296 bi
= conf
->retry_read_aligned_list
;
3298 conf
->retry_read_aligned_list
= bi
->bi_next
;
3300 bi
->bi_phys_segments
= 1; /* biased count of active stripes */
3301 bi
->bi_hw_segments
= 0; /* count of processed stripes */
3309 * The "raid5_align_endio" should check if the read succeeded and if it
3310 * did, call bio_endio on the original bio (having bio_put the new bio
3312 * If the read failed..
3314 static void raid5_align_endio(struct bio
*bi
, int error
)
3316 struct bio
* raid_bi
= bi
->bi_private
;
3319 int uptodate
= test_bit(BIO_UPTODATE
, &bi
->bi_flags
);
3324 mddev
= raid_bi
->bi_bdev
->bd_disk
->queue
->queuedata
;
3325 conf
= mddev_to_conf(mddev
);
3326 rdev
= (void*)raid_bi
->bi_next
;
3327 raid_bi
->bi_next
= NULL
;
3329 rdev_dec_pending(rdev
, conf
->mddev
);
3331 if (!error
&& uptodate
) {
3332 bio_endio(raid_bi
, 0);
3333 if (atomic_dec_and_test(&conf
->active_aligned_reads
))
3334 wake_up(&conf
->wait_for_stripe
);
3339 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
3341 add_bio_to_retry(raid_bi
, conf
);
3344 static int bio_fits_rdev(struct bio
*bi
)
3346 struct request_queue
*q
= bdev_get_queue(bi
->bi_bdev
);
3348 if ((bi
->bi_size
>>9) > q
->max_sectors
)
3350 blk_recount_segments(q
, bi
);
3351 if (bi
->bi_phys_segments
> q
->max_phys_segments
||
3352 bi
->bi_hw_segments
> q
->max_hw_segments
)
3355 if (q
->merge_bvec_fn
)
3356 /* it's too hard to apply the merge_bvec_fn at this stage,
3365 static int chunk_aligned_read(struct request_queue
*q
, struct bio
* raid_bio
)
3367 mddev_t
*mddev
= q
->queuedata
;
3368 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3369 const unsigned int raid_disks
= conf
->raid_disks
;
3370 const unsigned int data_disks
= raid_disks
- conf
->max_degraded
;
3371 unsigned int dd_idx
, pd_idx
;
3372 struct bio
* align_bi
;
3375 if (!in_chunk_boundary(mddev
, raid_bio
)) {
3376 pr_debug("chunk_aligned_read : non aligned\n");
3380 * use bio_clone to make a copy of the bio
3382 align_bi
= bio_clone(raid_bio
, GFP_NOIO
);
3386 * set bi_end_io to a new function, and set bi_private to the
3389 align_bi
->bi_end_io
= raid5_align_endio
;
3390 align_bi
->bi_private
= raid_bio
;
3394 align_bi
->bi_sector
= raid5_compute_sector(raid_bio
->bi_sector
,
3402 rdev
= rcu_dereference(conf
->disks
[dd_idx
].rdev
);
3403 if (rdev
&& test_bit(In_sync
, &rdev
->flags
)) {
3404 atomic_inc(&rdev
->nr_pending
);
3406 raid_bio
->bi_next
= (void*)rdev
;
3407 align_bi
->bi_bdev
= rdev
->bdev
;
3408 align_bi
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
3409 align_bi
->bi_sector
+= rdev
->data_offset
;
3411 if (!bio_fits_rdev(align_bi
)) {
3412 /* too big in some way */
3414 rdev_dec_pending(rdev
, mddev
);
3418 spin_lock_irq(&conf
->device_lock
);
3419 wait_event_lock_irq(conf
->wait_for_stripe
,
3421 conf
->device_lock
, /* nothing */);
3422 atomic_inc(&conf
->active_aligned_reads
);
3423 spin_unlock_irq(&conf
->device_lock
);
3425 generic_make_request(align_bi
);
3435 static int make_request(struct request_queue
*q
, struct bio
* bi
)
3437 mddev_t
*mddev
= q
->queuedata
;
3438 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3439 unsigned int dd_idx
, pd_idx
;
3440 sector_t new_sector
;
3441 sector_t logical_sector
, last_sector
;
3442 struct stripe_head
*sh
;
3443 const int rw
= bio_data_dir(bi
);
3446 if (unlikely(bio_barrier(bi
))) {
3447 bio_endio(bi
, -EOPNOTSUPP
);
3451 md_write_start(mddev
, bi
);
3453 disk_stat_inc(mddev
->gendisk
, ios
[rw
]);
3454 disk_stat_add(mddev
->gendisk
, sectors
[rw
], bio_sectors(bi
));
3457 mddev
->reshape_position
== MaxSector
&&
3458 chunk_aligned_read(q
,bi
))
3461 logical_sector
= bi
->bi_sector
& ~((sector_t
)STRIPE_SECTORS
-1);
3462 last_sector
= bi
->bi_sector
+ (bi
->bi_size
>>9);
3464 bi
->bi_phys_segments
= 1; /* over-loaded to count active stripes */
3466 for (;logical_sector
< last_sector
; logical_sector
+= STRIPE_SECTORS
) {
3468 int disks
, data_disks
;
3471 prepare_to_wait(&conf
->wait_for_overlap
, &w
, TASK_UNINTERRUPTIBLE
);
3472 if (likely(conf
->expand_progress
== MaxSector
))
3473 disks
= conf
->raid_disks
;
3475 /* spinlock is needed as expand_progress may be
3476 * 64bit on a 32bit platform, and so it might be
3477 * possible to see a half-updated value
3478 * Ofcourse expand_progress could change after
3479 * the lock is dropped, so once we get a reference
3480 * to the stripe that we think it is, we will have
3483 spin_lock_irq(&conf
->device_lock
);
3484 disks
= conf
->raid_disks
;
3485 if (logical_sector
>= conf
->expand_progress
)
3486 disks
= conf
->previous_raid_disks
;
3488 if (logical_sector
>= conf
->expand_lo
) {
3489 spin_unlock_irq(&conf
->device_lock
);
3494 spin_unlock_irq(&conf
->device_lock
);
3496 data_disks
= disks
- conf
->max_degraded
;
3498 new_sector
= raid5_compute_sector(logical_sector
, disks
, data_disks
,
3499 &dd_idx
, &pd_idx
, conf
);
3500 pr_debug("raid5: make_request, sector %llu logical %llu\n",
3501 (unsigned long long)new_sector
,
3502 (unsigned long long)logical_sector
);
3504 sh
= get_active_stripe(conf
, new_sector
, disks
, pd_idx
, (bi
->bi_rw
&RWA_MASK
));
3506 if (unlikely(conf
->expand_progress
!= MaxSector
)) {
3507 /* expansion might have moved on while waiting for a
3508 * stripe, so we must do the range check again.
3509 * Expansion could still move past after this
3510 * test, but as we are holding a reference to
3511 * 'sh', we know that if that happens,
3512 * STRIPE_EXPANDING will get set and the expansion
3513 * won't proceed until we finish with the stripe.
3516 spin_lock_irq(&conf
->device_lock
);
3517 if (logical_sector
< conf
->expand_progress
&&
3518 disks
== conf
->previous_raid_disks
)
3519 /* mismatch, need to try again */
3521 spin_unlock_irq(&conf
->device_lock
);
3527 /* FIXME what if we get a false positive because these
3528 * are being updated.
3530 if (logical_sector
>= mddev
->suspend_lo
&&
3531 logical_sector
< mddev
->suspend_hi
) {
3537 if (test_bit(STRIPE_EXPANDING
, &sh
->state
) ||
3538 !add_stripe_bio(sh
, bi
, dd_idx
, (bi
->bi_rw
&RW_MASK
))) {
3539 /* Stripe is busy expanding or
3540 * add failed due to overlap. Flush everything
3543 raid5_unplug_device(mddev
->queue
);
3548 finish_wait(&conf
->wait_for_overlap
, &w
);
3549 handle_stripe(sh
, NULL
);
3552 /* cannot get stripe for read-ahead, just give-up */
3553 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
3554 finish_wait(&conf
->wait_for_overlap
, &w
);
3559 spin_lock_irq(&conf
->device_lock
);
3560 remaining
= --bi
->bi_phys_segments
;
3561 spin_unlock_irq(&conf
->device_lock
);
3562 if (remaining
== 0) {
3565 md_write_end(mddev
);
3568 test_bit(BIO_UPTODATE
, &bi
->bi_flags
)
3574 static sector_t
reshape_request(mddev_t
*mddev
, sector_t sector_nr
, int *skipped
)
3576 /* reshaping is quite different to recovery/resync so it is
3577 * handled quite separately ... here.
3579 * On each call to sync_request, we gather one chunk worth of
3580 * destination stripes and flag them as expanding.
3581 * Then we find all the source stripes and request reads.
3582 * As the reads complete, handle_stripe will copy the data
3583 * into the destination stripe and release that stripe.
3585 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
3586 struct stripe_head
*sh
;
3588 sector_t first_sector
, last_sector
;
3589 int raid_disks
= conf
->previous_raid_disks
;
3590 int data_disks
= raid_disks
- conf
->max_degraded
;
3591 int new_data_disks
= conf
->raid_disks
- conf
->max_degraded
;
3594 sector_t writepos
, safepos
, gap
;
3596 if (sector_nr
== 0 &&
3597 conf
->expand_progress
!= 0) {
3598 /* restarting in the middle, skip the initial sectors */
3599 sector_nr
= conf
->expand_progress
;
3600 sector_div(sector_nr
, new_data_disks
);
3605 /* we update the metadata when there is more than 3Meg
3606 * in the block range (that is rather arbitrary, should
3607 * probably be time based) or when the data about to be
3608 * copied would over-write the source of the data at
3609 * the front of the range.
3610 * i.e. one new_stripe forward from expand_progress new_maps
3611 * to after where expand_lo old_maps to
3613 writepos
= conf
->expand_progress
+
3614 conf
->chunk_size
/512*(new_data_disks
);
3615 sector_div(writepos
, new_data_disks
);
3616 safepos
= conf
->expand_lo
;
3617 sector_div(safepos
, data_disks
);
3618 gap
= conf
->expand_progress
- conf
->expand_lo
;
3620 if (writepos
>= safepos
||
3621 gap
> (new_data_disks
)*3000*2 /*3Meg*/) {
3622 /* Cannot proceed until we've updated the superblock... */
3623 wait_event(conf
->wait_for_overlap
,
3624 atomic_read(&conf
->reshape_stripes
)==0);
3625 mddev
->reshape_position
= conf
->expand_progress
;
3626 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3627 md_wakeup_thread(mddev
->thread
);
3628 wait_event(mddev
->sb_wait
, mddev
->flags
== 0 ||
3629 kthread_should_stop());
3630 spin_lock_irq(&conf
->device_lock
);
3631 conf
->expand_lo
= mddev
->reshape_position
;
3632 spin_unlock_irq(&conf
->device_lock
);
3633 wake_up(&conf
->wait_for_overlap
);
3636 for (i
=0; i
< conf
->chunk_size
/512; i
+= STRIPE_SECTORS
) {
3639 pd_idx
= stripe_to_pdidx(sector_nr
+i
, conf
, conf
->raid_disks
);
3640 sh
= get_active_stripe(conf
, sector_nr
+i
,
3641 conf
->raid_disks
, pd_idx
, 0);
3642 set_bit(STRIPE_EXPANDING
, &sh
->state
);
3643 atomic_inc(&conf
->reshape_stripes
);
3644 /* If any of this stripe is beyond the end of the old
3645 * array, then we need to zero those blocks
3647 for (j
=sh
->disks
; j
--;) {
3649 if (j
== sh
->pd_idx
)
3651 if (conf
->level
== 6 &&
3652 j
== raid6_next_disk(sh
->pd_idx
, sh
->disks
))
3654 s
= compute_blocknr(sh
, j
);
3655 if (s
< (mddev
->array_size
<<1)) {
3659 memset(page_address(sh
->dev
[j
].page
), 0, STRIPE_SIZE
);
3660 set_bit(R5_Expanded
, &sh
->dev
[j
].flags
);
3661 set_bit(R5_UPTODATE
, &sh
->dev
[j
].flags
);
3664 set_bit(STRIPE_EXPAND_READY
, &sh
->state
);
3665 set_bit(STRIPE_HANDLE
, &sh
->state
);
3669 spin_lock_irq(&conf
->device_lock
);
3670 conf
->expand_progress
= (sector_nr
+ i
) * new_data_disks
;
3671 spin_unlock_irq(&conf
->device_lock
);
3672 /* Ok, those stripe are ready. We can start scheduling
3673 * reads on the source stripes.
3674 * The source stripes are determined by mapping the first and last
3675 * block on the destination stripes.
3678 raid5_compute_sector(sector_nr
*(new_data_disks
),
3679 raid_disks
, data_disks
,
3680 &dd_idx
, &pd_idx
, conf
);
3682 raid5_compute_sector((sector_nr
+conf
->chunk_size
/512)
3683 *(new_data_disks
) -1,
3684 raid_disks
, data_disks
,
3685 &dd_idx
, &pd_idx
, conf
);
3686 if (last_sector
>= (mddev
->size
<<1))
3687 last_sector
= (mddev
->size
<<1)-1;
3688 while (first_sector
<= last_sector
) {
3689 pd_idx
= stripe_to_pdidx(first_sector
, conf
,
3690 conf
->previous_raid_disks
);
3691 sh
= get_active_stripe(conf
, first_sector
,
3692 conf
->previous_raid_disks
, pd_idx
, 0);
3693 set_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
3694 set_bit(STRIPE_HANDLE
, &sh
->state
);
3696 first_sector
+= STRIPE_SECTORS
;
3698 return conf
->chunk_size
>>9;
3701 /* FIXME go_faster isn't used */
3702 static inline sector_t
sync_request(mddev_t
*mddev
, sector_t sector_nr
, int *skipped
, int go_faster
)
3704 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
3705 struct stripe_head
*sh
;
3707 int raid_disks
= conf
->raid_disks
;
3708 sector_t max_sector
= mddev
->size
<< 1;
3710 int still_degraded
= 0;
3713 if (sector_nr
>= max_sector
) {
3714 /* just being told to finish up .. nothing much to do */
3715 unplug_slaves(mddev
);
3716 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)) {
3721 if (mddev
->curr_resync
< max_sector
) /* aborted */
3722 bitmap_end_sync(mddev
->bitmap
, mddev
->curr_resync
,
3724 else /* completed sync */
3726 bitmap_close_sync(mddev
->bitmap
);
3731 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
3732 return reshape_request(mddev
, sector_nr
, skipped
);
3734 /* if there is too many failed drives and we are trying
3735 * to resync, then assert that we are finished, because there is
3736 * nothing we can do.
3738 if (mddev
->degraded
>= conf
->max_degraded
&&
3739 test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
3740 sector_t rv
= (mddev
->size
<< 1) - sector_nr
;
3744 if (!bitmap_start_sync(mddev
->bitmap
, sector_nr
, &sync_blocks
, 1) &&
3745 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
) &&
3746 !conf
->fullsync
&& sync_blocks
>= STRIPE_SECTORS
) {
3747 /* we can skip this block, and probably more */
3748 sync_blocks
/= STRIPE_SECTORS
;
3750 return sync_blocks
* STRIPE_SECTORS
; /* keep things rounded to whole stripes */
3753 pd_idx
= stripe_to_pdidx(sector_nr
, conf
, raid_disks
);
3754 sh
= get_active_stripe(conf
, sector_nr
, raid_disks
, pd_idx
, 1);
3756 sh
= get_active_stripe(conf
, sector_nr
, raid_disks
, pd_idx
, 0);
3757 /* make sure we don't swamp the stripe cache if someone else
3758 * is trying to get access
3760 schedule_timeout_uninterruptible(1);
3762 /* Need to check if array will still be degraded after recovery/resync
3763 * We don't need to check the 'failed' flag as when that gets set,
3766 for (i
=0; i
<mddev
->raid_disks
; i
++)
3767 if (conf
->disks
[i
].rdev
== NULL
)
3770 bitmap_start_sync(mddev
->bitmap
, sector_nr
, &sync_blocks
, still_degraded
);
3772 spin_lock(&sh
->lock
);
3773 set_bit(STRIPE_SYNCING
, &sh
->state
);
3774 clear_bit(STRIPE_INSYNC
, &sh
->state
);
3775 spin_unlock(&sh
->lock
);
3777 handle_stripe(sh
, NULL
);
3780 return STRIPE_SECTORS
;
3783 static int retry_aligned_read(raid5_conf_t
*conf
, struct bio
*raid_bio
)
3785 /* We may not be able to submit a whole bio at once as there
3786 * may not be enough stripe_heads available.
3787 * We cannot pre-allocate enough stripe_heads as we may need
3788 * more than exist in the cache (if we allow ever large chunks).
3789 * So we do one stripe head at a time and record in
3790 * ->bi_hw_segments how many have been done.
3792 * We *know* that this entire raid_bio is in one chunk, so
3793 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
3795 struct stripe_head
*sh
;
3797 sector_t sector
, logical_sector
, last_sector
;
3802 logical_sector
= raid_bio
->bi_sector
& ~((sector_t
)STRIPE_SECTORS
-1);
3803 sector
= raid5_compute_sector( logical_sector
,
3805 conf
->raid_disks
- conf
->max_degraded
,
3809 last_sector
= raid_bio
->bi_sector
+ (raid_bio
->bi_size
>>9);
3811 for (; logical_sector
< last_sector
;
3812 logical_sector
+= STRIPE_SECTORS
,
3813 sector
+= STRIPE_SECTORS
,
3816 if (scnt
< raid_bio
->bi_hw_segments
)
3817 /* already done this stripe */
3820 sh
= get_active_stripe(conf
, sector
, conf
->raid_disks
, pd_idx
, 1);
3823 /* failed to get a stripe - must wait */
3824 raid_bio
->bi_hw_segments
= scnt
;
3825 conf
->retry_read_aligned
= raid_bio
;
3829 set_bit(R5_ReadError
, &sh
->dev
[dd_idx
].flags
);
3830 if (!add_stripe_bio(sh
, raid_bio
, dd_idx
, 0)) {
3832 raid_bio
->bi_hw_segments
= scnt
;
3833 conf
->retry_read_aligned
= raid_bio
;
3837 handle_stripe(sh
, NULL
);
3841 spin_lock_irq(&conf
->device_lock
);
3842 remaining
= --raid_bio
->bi_phys_segments
;
3843 spin_unlock_irq(&conf
->device_lock
);
3844 if (remaining
== 0) {
3846 raid_bio
->bi_end_io(raid_bio
,
3847 test_bit(BIO_UPTODATE
, &raid_bio
->bi_flags
)
3850 if (atomic_dec_and_test(&conf
->active_aligned_reads
))
3851 wake_up(&conf
->wait_for_stripe
);
3858 * This is our raid5 kernel thread.
3860 * We scan the hash table for stripes which can be handled now.
3861 * During the scan, completed stripes are saved for us by the interrupt
3862 * handler, so that they will not have to wait for our next wakeup.
3864 static void raid5d (mddev_t
*mddev
)
3866 struct stripe_head
*sh
;
3867 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3870 pr_debug("+++ raid5d active\n");
3872 md_check_recovery(mddev
);
3875 spin_lock_irq(&conf
->device_lock
);
3877 struct list_head
*first
;
3880 if (conf
->seq_flush
!= conf
->seq_write
) {
3881 int seq
= conf
->seq_flush
;
3882 spin_unlock_irq(&conf
->device_lock
);
3883 bitmap_unplug(mddev
->bitmap
);
3884 spin_lock_irq(&conf
->device_lock
);
3885 conf
->seq_write
= seq
;
3886 activate_bit_delay(conf
);
3889 if (list_empty(&conf
->handle_list
) &&
3890 atomic_read(&conf
->preread_active_stripes
) < IO_THRESHOLD
&&
3891 !blk_queue_plugged(mddev
->queue
) &&
3892 !list_empty(&conf
->delayed_list
))
3893 raid5_activate_delayed(conf
);
3895 while ((bio
= remove_bio_from_retry(conf
))) {
3897 spin_unlock_irq(&conf
->device_lock
);
3898 ok
= retry_aligned_read(conf
, bio
);
3899 spin_lock_irq(&conf
->device_lock
);
3905 if (list_empty(&conf
->handle_list
)) {
3906 async_tx_issue_pending_all();
3910 first
= conf
->handle_list
.next
;
3911 sh
= list_entry(first
, struct stripe_head
, lru
);
3913 list_del_init(first
);
3914 atomic_inc(&sh
->count
);
3915 BUG_ON(atomic_read(&sh
->count
)!= 1);
3916 spin_unlock_irq(&conf
->device_lock
);
3919 handle_stripe(sh
, conf
->spare_page
);
3922 spin_lock_irq(&conf
->device_lock
);
3924 pr_debug("%d stripes handled\n", handled
);
3926 spin_unlock_irq(&conf
->device_lock
);
3928 unplug_slaves(mddev
);
3930 pr_debug("--- raid5d inactive\n");
3934 raid5_show_stripe_cache_size(mddev_t
*mddev
, char *page
)
3936 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3938 return sprintf(page
, "%d\n", conf
->max_nr_stripes
);
3944 raid5_store_stripe_cache_size(mddev_t
*mddev
, const char *page
, size_t len
)
3946 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3949 if (len
>= PAGE_SIZE
)
3954 new = simple_strtoul(page
, &end
, 10);
3955 if (!*page
|| (*end
&& *end
!= '\n') )
3957 if (new <= 16 || new > 32768)
3959 while (new < conf
->max_nr_stripes
) {
3960 if (drop_one_stripe(conf
))
3961 conf
->max_nr_stripes
--;
3965 md_allow_write(mddev
);
3966 while (new > conf
->max_nr_stripes
) {
3967 if (grow_one_stripe(conf
))
3968 conf
->max_nr_stripes
++;
3974 static struct md_sysfs_entry
3975 raid5_stripecache_size
= __ATTR(stripe_cache_size
, S_IRUGO
| S_IWUSR
,
3976 raid5_show_stripe_cache_size
,
3977 raid5_store_stripe_cache_size
);
3980 stripe_cache_active_show(mddev_t
*mddev
, char *page
)
3982 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3984 return sprintf(page
, "%d\n", atomic_read(&conf
->active_stripes
));
3989 static struct md_sysfs_entry
3990 raid5_stripecache_active
= __ATTR_RO(stripe_cache_active
);
3992 static struct attribute
*raid5_attrs
[] = {
3993 &raid5_stripecache_size
.attr
,
3994 &raid5_stripecache_active
.attr
,
3997 static struct attribute_group raid5_attrs_group
= {
3999 .attrs
= raid5_attrs
,
4002 static int run(mddev_t
*mddev
)
4005 int raid_disk
, memory
;
4007 struct disk_info
*disk
;
4008 struct list_head
*tmp
;
4009 int working_disks
= 0;
4011 if (mddev
->level
!= 5 && mddev
->level
!= 4 && mddev
->level
!= 6) {
4012 printk(KERN_ERR
"raid5: %s: raid level not set to 4/5/6 (%d)\n",
4013 mdname(mddev
), mddev
->level
);
4017 if (mddev
->reshape_position
!= MaxSector
) {
4018 /* Check that we can continue the reshape.
4019 * Currently only disks can change, it must
4020 * increase, and we must be past the point where
4021 * a stripe over-writes itself
4023 sector_t here_new
, here_old
;
4025 int max_degraded
= (mddev
->level
== 5 ? 1 : 2);
4027 if (mddev
->new_level
!= mddev
->level
||
4028 mddev
->new_layout
!= mddev
->layout
||
4029 mddev
->new_chunk
!= mddev
->chunk_size
) {
4030 printk(KERN_ERR
"raid5: %s: unsupported reshape "
4031 "required - aborting.\n",
4035 if (mddev
->delta_disks
<= 0) {
4036 printk(KERN_ERR
"raid5: %s: unsupported reshape "
4037 "(reduce disks) required - aborting.\n",
4041 old_disks
= mddev
->raid_disks
- mddev
->delta_disks
;
4042 /* reshape_position must be on a new-stripe boundary, and one
4043 * further up in new geometry must map after here in old
4046 here_new
= mddev
->reshape_position
;
4047 if (sector_div(here_new
, (mddev
->chunk_size
>>9)*
4048 (mddev
->raid_disks
- max_degraded
))) {
4049 printk(KERN_ERR
"raid5: reshape_position not "
4050 "on a stripe boundary\n");
4053 /* here_new is the stripe we will write to */
4054 here_old
= mddev
->reshape_position
;
4055 sector_div(here_old
, (mddev
->chunk_size
>>9)*
4056 (old_disks
-max_degraded
));
4057 /* here_old is the first stripe that we might need to read
4059 if (here_new
>= here_old
) {
4060 /* Reading from the same stripe as writing to - bad */
4061 printk(KERN_ERR
"raid5: reshape_position too early for "
4062 "auto-recovery - aborting.\n");
4065 printk(KERN_INFO
"raid5: reshape will continue\n");
4066 /* OK, we should be able to continue; */
4070 mddev
->private = kzalloc(sizeof (raid5_conf_t
), GFP_KERNEL
);
4071 if ((conf
= mddev
->private) == NULL
)
4073 if (mddev
->reshape_position
== MaxSector
) {
4074 conf
->previous_raid_disks
= conf
->raid_disks
= mddev
->raid_disks
;
4076 conf
->raid_disks
= mddev
->raid_disks
;
4077 conf
->previous_raid_disks
= mddev
->raid_disks
- mddev
->delta_disks
;
4080 conf
->disks
= kzalloc(conf
->raid_disks
* sizeof(struct disk_info
),
4085 conf
->mddev
= mddev
;
4087 if ((conf
->stripe_hashtbl
= kzalloc(PAGE_SIZE
, GFP_KERNEL
)) == NULL
)
4090 if (mddev
->level
== 6) {
4091 conf
->spare_page
= alloc_page(GFP_KERNEL
);
4092 if (!conf
->spare_page
)
4095 spin_lock_init(&conf
->device_lock
);
4096 init_waitqueue_head(&conf
->wait_for_stripe
);
4097 init_waitqueue_head(&conf
->wait_for_overlap
);
4098 INIT_LIST_HEAD(&conf
->handle_list
);
4099 INIT_LIST_HEAD(&conf
->delayed_list
);
4100 INIT_LIST_HEAD(&conf
->bitmap_list
);
4101 INIT_LIST_HEAD(&conf
->inactive_list
);
4102 atomic_set(&conf
->active_stripes
, 0);
4103 atomic_set(&conf
->preread_active_stripes
, 0);
4104 atomic_set(&conf
->active_aligned_reads
, 0);
4106 pr_debug("raid5: run(%s) called.\n", mdname(mddev
));
4108 ITERATE_RDEV(mddev
,rdev
,tmp
) {
4109 raid_disk
= rdev
->raid_disk
;
4110 if (raid_disk
>= conf
->raid_disks
4113 disk
= conf
->disks
+ raid_disk
;
4117 if (test_bit(In_sync
, &rdev
->flags
)) {
4118 char b
[BDEVNAME_SIZE
];
4119 printk(KERN_INFO
"raid5: device %s operational as raid"
4120 " disk %d\n", bdevname(rdev
->bdev
,b
),
4127 * 0 for a fully functional array, 1 or 2 for a degraded array.
4129 mddev
->degraded
= conf
->raid_disks
- working_disks
;
4130 conf
->mddev
= mddev
;
4131 conf
->chunk_size
= mddev
->chunk_size
;
4132 conf
->level
= mddev
->level
;
4133 if (conf
->level
== 6)
4134 conf
->max_degraded
= 2;
4136 conf
->max_degraded
= 1;
4137 conf
->algorithm
= mddev
->layout
;
4138 conf
->max_nr_stripes
= NR_STRIPES
;
4139 conf
->expand_progress
= mddev
->reshape_position
;
4141 /* device size must be a multiple of chunk size */
4142 mddev
->size
&= ~(mddev
->chunk_size
/1024 -1);
4143 mddev
->resync_max_sectors
= mddev
->size
<< 1;
4145 if (conf
->level
== 6 && conf
->raid_disks
< 4) {
4146 printk(KERN_ERR
"raid6: not enough configured devices for %s (%d, minimum 4)\n",
4147 mdname(mddev
), conf
->raid_disks
);
4150 if (!conf
->chunk_size
|| conf
->chunk_size
% 4) {
4151 printk(KERN_ERR
"raid5: invalid chunk size %d for %s\n",
4152 conf
->chunk_size
, mdname(mddev
));
4155 if (conf
->algorithm
> ALGORITHM_RIGHT_SYMMETRIC
) {
4157 "raid5: unsupported parity algorithm %d for %s\n",
4158 conf
->algorithm
, mdname(mddev
));
4161 if (mddev
->degraded
> conf
->max_degraded
) {
4162 printk(KERN_ERR
"raid5: not enough operational devices for %s"
4163 " (%d/%d failed)\n",
4164 mdname(mddev
), mddev
->degraded
, conf
->raid_disks
);
4168 if (mddev
->degraded
> 0 &&
4169 mddev
->recovery_cp
!= MaxSector
) {
4170 if (mddev
->ok_start_degraded
)
4172 "raid5: starting dirty degraded array: %s"
4173 "- data corruption possible.\n",
4177 "raid5: cannot start dirty degraded array for %s\n",
4184 mddev
->thread
= md_register_thread(raid5d
, mddev
, "%s_raid5");
4185 if (!mddev
->thread
) {
4187 "raid5: couldn't allocate thread for %s\n",
4192 memory
= conf
->max_nr_stripes
* (sizeof(struct stripe_head
) +
4193 conf
->raid_disks
* ((sizeof(struct bio
) + PAGE_SIZE
))) / 1024;
4194 if (grow_stripes(conf
, conf
->max_nr_stripes
)) {
4196 "raid5: couldn't allocate %dkB for buffers\n", memory
);
4197 shrink_stripes(conf
);
4198 md_unregister_thread(mddev
->thread
);
4201 printk(KERN_INFO
"raid5: allocated %dkB for %s\n",
4202 memory
, mdname(mddev
));
4204 if (mddev
->degraded
== 0)
4205 printk("raid5: raid level %d set %s active with %d out of %d"
4206 " devices, algorithm %d\n", conf
->level
, mdname(mddev
),
4207 mddev
->raid_disks
-mddev
->degraded
, mddev
->raid_disks
,
4210 printk(KERN_ALERT
"raid5: raid level %d set %s active with %d"
4211 " out of %d devices, algorithm %d\n", conf
->level
,
4212 mdname(mddev
), mddev
->raid_disks
- mddev
->degraded
,
4213 mddev
->raid_disks
, conf
->algorithm
);
4215 print_raid5_conf(conf
);
4217 if (conf
->expand_progress
!= MaxSector
) {
4218 printk("...ok start reshape thread\n");
4219 conf
->expand_lo
= conf
->expand_progress
;
4220 atomic_set(&conf
->reshape_stripes
, 0);
4221 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4222 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4223 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
4224 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
4225 mddev
->sync_thread
= md_register_thread(md_do_sync
, mddev
,
4229 /* read-ahead size must cover two whole stripes, which is
4230 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
4233 int data_disks
= conf
->previous_raid_disks
- conf
->max_degraded
;
4234 int stripe
= data_disks
*
4235 (mddev
->chunk_size
/ PAGE_SIZE
);
4236 if (mddev
->queue
->backing_dev_info
.ra_pages
< 2 * stripe
)
4237 mddev
->queue
->backing_dev_info
.ra_pages
= 2 * stripe
;
4240 /* Ok, everything is just fine now */
4241 if (sysfs_create_group(&mddev
->kobj
, &raid5_attrs_group
))
4243 "raid5: failed to create sysfs attributes for %s\n",
4246 mddev
->queue
->unplug_fn
= raid5_unplug_device
;
4247 mddev
->queue
->backing_dev_info
.congested_data
= mddev
;
4248 mddev
->queue
->backing_dev_info
.congested_fn
= raid5_congested
;
4250 mddev
->array_size
= mddev
->size
* (conf
->previous_raid_disks
-
4251 conf
->max_degraded
);
4253 blk_queue_merge_bvec(mddev
->queue
, raid5_mergeable_bvec
);
4258 print_raid5_conf(conf
);
4259 safe_put_page(conf
->spare_page
);
4261 kfree(conf
->stripe_hashtbl
);
4264 mddev
->private = NULL
;
4265 printk(KERN_ALERT
"raid5: failed to run raid set %s\n", mdname(mddev
));
4271 static int stop(mddev_t
*mddev
)
4273 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
4275 md_unregister_thread(mddev
->thread
);
4276 mddev
->thread
= NULL
;
4277 shrink_stripes(conf
);
4278 kfree(conf
->stripe_hashtbl
);
4279 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
4280 blk_sync_queue(mddev
->queue
); /* the unplug fn references 'conf'*/
4281 sysfs_remove_group(&mddev
->kobj
, &raid5_attrs_group
);
4284 mddev
->private = NULL
;
4289 static void print_sh (struct seq_file
*seq
, struct stripe_head
*sh
)
4293 seq_printf(seq
, "sh %llu, pd_idx %d, state %ld.\n",
4294 (unsigned long long)sh
->sector
, sh
->pd_idx
, sh
->state
);
4295 seq_printf(seq
, "sh %llu, count %d.\n",
4296 (unsigned long long)sh
->sector
, atomic_read(&sh
->count
));
4297 seq_printf(seq
, "sh %llu, ", (unsigned long long)sh
->sector
);
4298 for (i
= 0; i
< sh
->disks
; i
++) {
4299 seq_printf(seq
, "(cache%d: %p %ld) ",
4300 i
, sh
->dev
[i
].page
, sh
->dev
[i
].flags
);
4302 seq_printf(seq
, "\n");
4305 static void printall (struct seq_file
*seq
, raid5_conf_t
*conf
)
4307 struct stripe_head
*sh
;
4308 struct hlist_node
*hn
;
4311 spin_lock_irq(&conf
->device_lock
);
4312 for (i
= 0; i
< NR_HASH
; i
++) {
4313 hlist_for_each_entry(sh
, hn
, &conf
->stripe_hashtbl
[i
], hash
) {
4314 if (sh
->raid_conf
!= conf
)
4319 spin_unlock_irq(&conf
->device_lock
);
4323 static void status (struct seq_file
*seq
, mddev_t
*mddev
)
4325 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
4328 seq_printf (seq
, " level %d, %dk chunk, algorithm %d", mddev
->level
, mddev
->chunk_size
>> 10, mddev
->layout
);
4329 seq_printf (seq
, " [%d/%d] [", conf
->raid_disks
, conf
->raid_disks
- mddev
->degraded
);
4330 for (i
= 0; i
< conf
->raid_disks
; i
++)
4331 seq_printf (seq
, "%s",
4332 conf
->disks
[i
].rdev
&&
4333 test_bit(In_sync
, &conf
->disks
[i
].rdev
->flags
) ? "U" : "_");
4334 seq_printf (seq
, "]");
4336 seq_printf (seq
, "\n");
4337 printall(seq
, conf
);
4341 static void print_raid5_conf (raid5_conf_t
*conf
)
4344 struct disk_info
*tmp
;
4346 printk("RAID5 conf printout:\n");
4348 printk("(conf==NULL)\n");
4351 printk(" --- rd:%d wd:%d\n", conf
->raid_disks
,
4352 conf
->raid_disks
- conf
->mddev
->degraded
);
4354 for (i
= 0; i
< conf
->raid_disks
; i
++) {
4355 char b
[BDEVNAME_SIZE
];
4356 tmp
= conf
->disks
+ i
;
4358 printk(" disk %d, o:%d, dev:%s\n",
4359 i
, !test_bit(Faulty
, &tmp
->rdev
->flags
),
4360 bdevname(tmp
->rdev
->bdev
,b
));
4364 static int raid5_spare_active(mddev_t
*mddev
)
4367 raid5_conf_t
*conf
= mddev
->private;
4368 struct disk_info
*tmp
;
4370 for (i
= 0; i
< conf
->raid_disks
; i
++) {
4371 tmp
= conf
->disks
+ i
;
4373 && !test_bit(Faulty
, &tmp
->rdev
->flags
)
4374 && !test_and_set_bit(In_sync
, &tmp
->rdev
->flags
)) {
4375 unsigned long flags
;
4376 spin_lock_irqsave(&conf
->device_lock
, flags
);
4378 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
4381 print_raid5_conf(conf
);
4385 static int raid5_remove_disk(mddev_t
*mddev
, int number
)
4387 raid5_conf_t
*conf
= mddev
->private;
4390 struct disk_info
*p
= conf
->disks
+ number
;
4392 print_raid5_conf(conf
);
4395 if (test_bit(In_sync
, &rdev
->flags
) ||
4396 atomic_read(&rdev
->nr_pending
)) {
4402 if (atomic_read(&rdev
->nr_pending
)) {
4403 /* lost the race, try later */
4410 print_raid5_conf(conf
);
4414 static int raid5_add_disk(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
4416 raid5_conf_t
*conf
= mddev
->private;
4419 struct disk_info
*p
;
4421 if (mddev
->degraded
> conf
->max_degraded
)
4422 /* no point adding a device */
4426 * find the disk ... but prefer rdev->saved_raid_disk
4429 if (rdev
->saved_raid_disk
>= 0 &&
4430 conf
->disks
[rdev
->saved_raid_disk
].rdev
== NULL
)
4431 disk
= rdev
->saved_raid_disk
;
4434 for ( ; disk
< conf
->raid_disks
; disk
++)
4435 if ((p
=conf
->disks
+ disk
)->rdev
== NULL
) {
4436 clear_bit(In_sync
, &rdev
->flags
);
4437 rdev
->raid_disk
= disk
;
4439 if (rdev
->saved_raid_disk
!= disk
)
4441 rcu_assign_pointer(p
->rdev
, rdev
);
4444 print_raid5_conf(conf
);
4448 static int raid5_resize(mddev_t
*mddev
, sector_t sectors
)
4450 /* no resync is happening, and there is enough space
4451 * on all devices, so we can resize.
4452 * We need to make sure resync covers any new space.
4453 * If the array is shrinking we should possibly wait until
4454 * any io in the removed space completes, but it hardly seems
4457 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4459 sectors
&= ~((sector_t
)mddev
->chunk_size
/512 - 1);
4460 mddev
->array_size
= (sectors
* (mddev
->raid_disks
-conf
->max_degraded
))>>1;
4461 set_capacity(mddev
->gendisk
, mddev
->array_size
<< 1);
4463 if (sectors
/2 > mddev
->size
&& mddev
->recovery_cp
== MaxSector
) {
4464 mddev
->recovery_cp
= mddev
->size
<< 1;
4465 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4467 mddev
->size
= sectors
/2;
4468 mddev
->resync_max_sectors
= sectors
;
4472 #ifdef CONFIG_MD_RAID5_RESHAPE
4473 static int raid5_check_reshape(mddev_t
*mddev
)
4475 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4478 if (mddev
->delta_disks
< 0 ||
4479 mddev
->new_level
!= mddev
->level
)
4480 return -EINVAL
; /* Cannot shrink array or change level yet */
4481 if (mddev
->delta_disks
== 0)
4482 return 0; /* nothing to do */
4484 /* Can only proceed if there are plenty of stripe_heads.
4485 * We need a minimum of one full stripe,, and for sensible progress
4486 * it is best to have about 4 times that.
4487 * If we require 4 times, then the default 256 4K stripe_heads will
4488 * allow for chunk sizes up to 256K, which is probably OK.
4489 * If the chunk size is greater, user-space should request more
4490 * stripe_heads first.
4492 if ((mddev
->chunk_size
/ STRIPE_SIZE
) * 4 > conf
->max_nr_stripes
||
4493 (mddev
->new_chunk
/ STRIPE_SIZE
) * 4 > conf
->max_nr_stripes
) {
4494 printk(KERN_WARNING
"raid5: reshape: not enough stripes. Needed %lu\n",
4495 (mddev
->chunk_size
/ STRIPE_SIZE
)*4);
4499 err
= resize_stripes(conf
, conf
->raid_disks
+ mddev
->delta_disks
);
4503 if (mddev
->degraded
> conf
->max_degraded
)
4505 /* looks like we might be able to manage this */
4509 static int raid5_start_reshape(mddev_t
*mddev
)
4511 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4513 struct list_head
*rtmp
;
4515 int added_devices
= 0;
4516 unsigned long flags
;
4518 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4521 ITERATE_RDEV(mddev
, rdev
, rtmp
)
4522 if (rdev
->raid_disk
< 0 &&
4523 !test_bit(Faulty
, &rdev
->flags
))
4526 if (spares
- mddev
->degraded
< mddev
->delta_disks
- conf
->max_degraded
)
4527 /* Not enough devices even to make a degraded array
4532 atomic_set(&conf
->reshape_stripes
, 0);
4533 spin_lock_irq(&conf
->device_lock
);
4534 conf
->previous_raid_disks
= conf
->raid_disks
;
4535 conf
->raid_disks
+= mddev
->delta_disks
;
4536 conf
->expand_progress
= 0;
4537 conf
->expand_lo
= 0;
4538 spin_unlock_irq(&conf
->device_lock
);
4540 /* Add some new drives, as many as will fit.
4541 * We know there are enough to make the newly sized array work.
4543 ITERATE_RDEV(mddev
, rdev
, rtmp
)
4544 if (rdev
->raid_disk
< 0 &&
4545 !test_bit(Faulty
, &rdev
->flags
)) {
4546 if (raid5_add_disk(mddev
, rdev
)) {
4548 set_bit(In_sync
, &rdev
->flags
);
4550 rdev
->recovery_offset
= 0;
4551 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4552 if (sysfs_create_link(&mddev
->kobj
,
4555 "raid5: failed to create "
4556 " link %s for %s\n",
4562 spin_lock_irqsave(&conf
->device_lock
, flags
);
4563 mddev
->degraded
= (conf
->raid_disks
- conf
->previous_raid_disks
) - added_devices
;
4564 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
4565 mddev
->raid_disks
= conf
->raid_disks
;
4566 mddev
->reshape_position
= 0;
4567 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
4569 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4570 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4571 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
4572 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
4573 mddev
->sync_thread
= md_register_thread(md_do_sync
, mddev
,
4575 if (!mddev
->sync_thread
) {
4576 mddev
->recovery
= 0;
4577 spin_lock_irq(&conf
->device_lock
);
4578 mddev
->raid_disks
= conf
->raid_disks
= conf
->previous_raid_disks
;
4579 conf
->expand_progress
= MaxSector
;
4580 spin_unlock_irq(&conf
->device_lock
);
4583 md_wakeup_thread(mddev
->sync_thread
);
4584 md_new_event(mddev
);
4589 static void end_reshape(raid5_conf_t
*conf
)
4591 struct block_device
*bdev
;
4593 if (!test_bit(MD_RECOVERY_INTR
, &conf
->mddev
->recovery
)) {
4594 conf
->mddev
->array_size
= conf
->mddev
->size
*
4595 (conf
->raid_disks
- conf
->max_degraded
);
4596 set_capacity(conf
->mddev
->gendisk
, conf
->mddev
->array_size
<< 1);
4597 conf
->mddev
->changed
= 1;
4599 bdev
= bdget_disk(conf
->mddev
->gendisk
, 0);
4601 mutex_lock(&bdev
->bd_inode
->i_mutex
);
4602 i_size_write(bdev
->bd_inode
, (loff_t
)conf
->mddev
->array_size
<< 10);
4603 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
4606 spin_lock_irq(&conf
->device_lock
);
4607 conf
->expand_progress
= MaxSector
;
4608 spin_unlock_irq(&conf
->device_lock
);
4609 conf
->mddev
->reshape_position
= MaxSector
;
4611 /* read-ahead size must cover two whole stripes, which is
4612 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
4615 int data_disks
= conf
->previous_raid_disks
- conf
->max_degraded
;
4616 int stripe
= data_disks
*
4617 (conf
->mddev
->chunk_size
/ PAGE_SIZE
);
4618 if (conf
->mddev
->queue
->backing_dev_info
.ra_pages
< 2 * stripe
)
4619 conf
->mddev
->queue
->backing_dev_info
.ra_pages
= 2 * stripe
;
4624 static void raid5_quiesce(mddev_t
*mddev
, int state
)
4626 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4629 case 2: /* resume for a suspend */
4630 wake_up(&conf
->wait_for_overlap
);
4633 case 1: /* stop all writes */
4634 spin_lock_irq(&conf
->device_lock
);
4636 wait_event_lock_irq(conf
->wait_for_stripe
,
4637 atomic_read(&conf
->active_stripes
) == 0 &&
4638 atomic_read(&conf
->active_aligned_reads
) == 0,
4639 conf
->device_lock
, /* nothing */);
4640 spin_unlock_irq(&conf
->device_lock
);
4643 case 0: /* re-enable writes */
4644 spin_lock_irq(&conf
->device_lock
);
4646 wake_up(&conf
->wait_for_stripe
);
4647 wake_up(&conf
->wait_for_overlap
);
4648 spin_unlock_irq(&conf
->device_lock
);
4653 static struct mdk_personality raid6_personality
=
4657 .owner
= THIS_MODULE
,
4658 .make_request
= make_request
,
4662 .error_handler
= error
,
4663 .hot_add_disk
= raid5_add_disk
,
4664 .hot_remove_disk
= raid5_remove_disk
,
4665 .spare_active
= raid5_spare_active
,
4666 .sync_request
= sync_request
,
4667 .resize
= raid5_resize
,
4668 #ifdef CONFIG_MD_RAID5_RESHAPE
4669 .check_reshape
= raid5_check_reshape
,
4670 .start_reshape
= raid5_start_reshape
,
4672 .quiesce
= raid5_quiesce
,
4674 static struct mdk_personality raid5_personality
=
4678 .owner
= THIS_MODULE
,
4679 .make_request
= make_request
,
4683 .error_handler
= error
,
4684 .hot_add_disk
= raid5_add_disk
,
4685 .hot_remove_disk
= raid5_remove_disk
,
4686 .spare_active
= raid5_spare_active
,
4687 .sync_request
= sync_request
,
4688 .resize
= raid5_resize
,
4689 #ifdef CONFIG_MD_RAID5_RESHAPE
4690 .check_reshape
= raid5_check_reshape
,
4691 .start_reshape
= raid5_start_reshape
,
4693 .quiesce
= raid5_quiesce
,
4696 static struct mdk_personality raid4_personality
=
4700 .owner
= THIS_MODULE
,
4701 .make_request
= make_request
,
4705 .error_handler
= error
,
4706 .hot_add_disk
= raid5_add_disk
,
4707 .hot_remove_disk
= raid5_remove_disk
,
4708 .spare_active
= raid5_spare_active
,
4709 .sync_request
= sync_request
,
4710 .resize
= raid5_resize
,
4711 #ifdef CONFIG_MD_RAID5_RESHAPE
4712 .check_reshape
= raid5_check_reshape
,
4713 .start_reshape
= raid5_start_reshape
,
4715 .quiesce
= raid5_quiesce
,
4718 static int __init
raid5_init(void)
4722 e
= raid6_select_algo();
4725 register_md_personality(&raid6_personality
);
4726 register_md_personality(&raid5_personality
);
4727 register_md_personality(&raid4_personality
);
4731 static void raid5_exit(void)
4733 unregister_md_personality(&raid6_personality
);
4734 unregister_md_personality(&raid5_personality
);
4735 unregister_md_personality(&raid4_personality
);
4738 module_init(raid5_init
);
4739 module_exit(raid5_exit
);
4740 MODULE_LICENSE("GPL");
4741 MODULE_ALIAS("md-personality-4"); /* RAID5 */
4742 MODULE_ALIAS("md-raid5");
4743 MODULE_ALIAS("md-raid4");
4744 MODULE_ALIAS("md-level-5");
4745 MODULE_ALIAS("md-level-4");
4746 MODULE_ALIAS("md-personality-8"); /* RAID6 */
4747 MODULE_ALIAS("md-raid6");
4748 MODULE_ALIAS("md-level-6");
4750 /* This used to be two separate modules, they were: */
4751 MODULE_ALIAS("raid5");
4752 MODULE_ALIAS("raid6");