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
;
111 int bytes
= bi
->bi_size
;
113 return_bi
= bi
->bi_next
;
116 bi
->bi_end_io(bi
, bytes
,
117 test_bit(BIO_UPTODATE
, &bi
->bi_flags
)
123 static void print_raid5_conf (raid5_conf_t
*conf
);
125 static void __release_stripe(raid5_conf_t
*conf
, struct stripe_head
*sh
)
127 if (atomic_dec_and_test(&sh
->count
)) {
128 BUG_ON(!list_empty(&sh
->lru
));
129 BUG_ON(atomic_read(&conf
->active_stripes
)==0);
130 if (test_bit(STRIPE_HANDLE
, &sh
->state
)) {
131 if (test_bit(STRIPE_DELAYED
, &sh
->state
)) {
132 list_add_tail(&sh
->lru
, &conf
->delayed_list
);
133 blk_plug_device(conf
->mddev
->queue
);
134 } else if (test_bit(STRIPE_BIT_DELAY
, &sh
->state
) &&
135 sh
->bm_seq
- conf
->seq_write
> 0) {
136 list_add_tail(&sh
->lru
, &conf
->bitmap_list
);
137 blk_plug_device(conf
->mddev
->queue
);
139 clear_bit(STRIPE_BIT_DELAY
, &sh
->state
);
140 list_add_tail(&sh
->lru
, &conf
->handle_list
);
142 md_wakeup_thread(conf
->mddev
->thread
);
144 BUG_ON(sh
->ops
.pending
);
145 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
146 atomic_dec(&conf
->preread_active_stripes
);
147 if (atomic_read(&conf
->preread_active_stripes
) < IO_THRESHOLD
)
148 md_wakeup_thread(conf
->mddev
->thread
);
150 atomic_dec(&conf
->active_stripes
);
151 if (!test_bit(STRIPE_EXPANDING
, &sh
->state
)) {
152 list_add_tail(&sh
->lru
, &conf
->inactive_list
);
153 wake_up(&conf
->wait_for_stripe
);
154 if (conf
->retry_read_aligned
)
155 md_wakeup_thread(conf
->mddev
->thread
);
160 static void release_stripe(struct stripe_head
*sh
)
162 raid5_conf_t
*conf
= sh
->raid_conf
;
165 spin_lock_irqsave(&conf
->device_lock
, flags
);
166 __release_stripe(conf
, sh
);
167 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
170 static inline void remove_hash(struct stripe_head
*sh
)
172 pr_debug("remove_hash(), stripe %llu\n",
173 (unsigned long long)sh
->sector
);
175 hlist_del_init(&sh
->hash
);
178 static inline void insert_hash(raid5_conf_t
*conf
, struct stripe_head
*sh
)
180 struct hlist_head
*hp
= stripe_hash(conf
, sh
->sector
);
182 pr_debug("insert_hash(), stripe %llu\n",
183 (unsigned long long)sh
->sector
);
186 hlist_add_head(&sh
->hash
, hp
);
190 /* find an idle stripe, make sure it is unhashed, and return it. */
191 static struct stripe_head
*get_free_stripe(raid5_conf_t
*conf
)
193 struct stripe_head
*sh
= NULL
;
194 struct list_head
*first
;
197 if (list_empty(&conf
->inactive_list
))
199 first
= conf
->inactive_list
.next
;
200 sh
= list_entry(first
, struct stripe_head
, lru
);
201 list_del_init(first
);
203 atomic_inc(&conf
->active_stripes
);
208 static void shrink_buffers(struct stripe_head
*sh
, int num
)
213 for (i
=0; i
<num
; i
++) {
217 sh
->dev
[i
].page
= NULL
;
222 static int grow_buffers(struct stripe_head
*sh
, int num
)
226 for (i
=0; i
<num
; i
++) {
229 if (!(page
= alloc_page(GFP_KERNEL
))) {
232 sh
->dev
[i
].page
= page
;
237 static void raid5_build_block (struct stripe_head
*sh
, int i
);
239 static void init_stripe(struct stripe_head
*sh
, sector_t sector
, int pd_idx
, int disks
)
241 raid5_conf_t
*conf
= sh
->raid_conf
;
244 BUG_ON(atomic_read(&sh
->count
) != 0);
245 BUG_ON(test_bit(STRIPE_HANDLE
, &sh
->state
));
246 BUG_ON(sh
->ops
.pending
|| sh
->ops
.ack
|| sh
->ops
.complete
);
249 pr_debug("init_stripe called, stripe %llu\n",
250 (unsigned long long)sh
->sector
);
260 for (i
= sh
->disks
; i
--; ) {
261 struct r5dev
*dev
= &sh
->dev
[i
];
263 if (dev
->toread
|| dev
->read
|| dev
->towrite
|| dev
->written
||
264 test_bit(R5_LOCKED
, &dev
->flags
)) {
265 printk(KERN_ERR
"sector=%llx i=%d %p %p %p %p %d\n",
266 (unsigned long long)sh
->sector
, i
, dev
->toread
,
267 dev
->read
, dev
->towrite
, dev
->written
,
268 test_bit(R5_LOCKED
, &dev
->flags
));
272 raid5_build_block(sh
, i
);
274 insert_hash(conf
, sh
);
277 static struct stripe_head
*__find_stripe(raid5_conf_t
*conf
, sector_t sector
, int disks
)
279 struct stripe_head
*sh
;
280 struct hlist_node
*hn
;
283 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector
);
284 hlist_for_each_entry(sh
, hn
, stripe_hash(conf
, sector
), hash
)
285 if (sh
->sector
== sector
&& sh
->disks
== disks
)
287 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector
);
291 static void unplug_slaves(mddev_t
*mddev
);
292 static void raid5_unplug_device(struct request_queue
*q
);
294 static struct stripe_head
*get_active_stripe(raid5_conf_t
*conf
, sector_t sector
, int disks
,
295 int pd_idx
, int noblock
)
297 struct stripe_head
*sh
;
299 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector
);
301 spin_lock_irq(&conf
->device_lock
);
304 wait_event_lock_irq(conf
->wait_for_stripe
,
306 conf
->device_lock
, /* nothing */);
307 sh
= __find_stripe(conf
, sector
, disks
);
309 if (!conf
->inactive_blocked
)
310 sh
= get_free_stripe(conf
);
311 if (noblock
&& sh
== NULL
)
314 conf
->inactive_blocked
= 1;
315 wait_event_lock_irq(conf
->wait_for_stripe
,
316 !list_empty(&conf
->inactive_list
) &&
317 (atomic_read(&conf
->active_stripes
)
318 < (conf
->max_nr_stripes
*3/4)
319 || !conf
->inactive_blocked
),
321 raid5_unplug_device(conf
->mddev
->queue
)
323 conf
->inactive_blocked
= 0;
325 init_stripe(sh
, sector
, pd_idx
, disks
);
327 if (atomic_read(&sh
->count
)) {
328 BUG_ON(!list_empty(&sh
->lru
));
330 if (!test_bit(STRIPE_HANDLE
, &sh
->state
))
331 atomic_inc(&conf
->active_stripes
);
332 if (list_empty(&sh
->lru
) &&
333 !test_bit(STRIPE_EXPANDING
, &sh
->state
))
335 list_del_init(&sh
->lru
);
338 } while (sh
== NULL
);
341 atomic_inc(&sh
->count
);
343 spin_unlock_irq(&conf
->device_lock
);
347 /* test_and_ack_op() ensures that we only dequeue an operation once */
348 #define test_and_ack_op(op, pend) \
350 if (test_bit(op, &sh->ops.pending) && \
351 !test_bit(op, &sh->ops.complete)) { \
352 if (test_and_set_bit(op, &sh->ops.ack)) \
353 clear_bit(op, &pend); \
357 clear_bit(op, &pend); \
360 /* find new work to run, do not resubmit work that is already
363 static unsigned long get_stripe_work(struct stripe_head
*sh
)
365 unsigned long pending
;
368 pending
= sh
->ops
.pending
;
370 test_and_ack_op(STRIPE_OP_BIOFILL
, pending
);
371 test_and_ack_op(STRIPE_OP_COMPUTE_BLK
, pending
);
372 test_and_ack_op(STRIPE_OP_PREXOR
, pending
);
373 test_and_ack_op(STRIPE_OP_BIODRAIN
, pending
);
374 test_and_ack_op(STRIPE_OP_POSTXOR
, pending
);
375 test_and_ack_op(STRIPE_OP_CHECK
, pending
);
376 if (test_and_clear_bit(STRIPE_OP_IO
, &sh
->ops
.pending
))
379 sh
->ops
.count
-= ack
;
380 if (unlikely(sh
->ops
.count
< 0)) {
381 printk(KERN_ERR
"pending: %#lx ops.pending: %#lx ops.ack: %#lx "
382 "ops.complete: %#lx\n", pending
, sh
->ops
.pending
,
383 sh
->ops
.ack
, sh
->ops
.complete
);
391 raid5_end_read_request(struct bio
*bi
, unsigned int bytes_done
, int error
);
393 raid5_end_write_request (struct bio
*bi
, unsigned int bytes_done
, int error
);
395 static void ops_run_io(struct stripe_head
*sh
)
397 raid5_conf_t
*conf
= sh
->raid_conf
;
398 int i
, disks
= sh
->disks
;
402 for (i
= disks
; i
--; ) {
406 if (test_and_clear_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
))
408 else if (test_and_clear_bit(R5_Wantread
, &sh
->dev
[i
].flags
))
413 bi
= &sh
->dev
[i
].req
;
417 bi
->bi_end_io
= raid5_end_write_request
;
419 bi
->bi_end_io
= raid5_end_read_request
;
422 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
423 if (rdev
&& test_bit(Faulty
, &rdev
->flags
))
426 atomic_inc(&rdev
->nr_pending
);
430 if (test_bit(STRIPE_SYNCING
, &sh
->state
) ||
431 test_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
) ||
432 test_bit(STRIPE_EXPAND_READY
, &sh
->state
))
433 md_sync_acct(rdev
->bdev
, STRIPE_SECTORS
);
435 bi
->bi_bdev
= rdev
->bdev
;
436 pr_debug("%s: for %llu schedule op %ld on disc %d\n",
437 __FUNCTION__
, (unsigned long long)sh
->sector
,
439 atomic_inc(&sh
->count
);
440 bi
->bi_sector
= sh
->sector
+ rdev
->data_offset
;
441 bi
->bi_flags
= 1 << BIO_UPTODATE
;
445 bi
->bi_io_vec
= &sh
->dev
[i
].vec
;
446 bi
->bi_io_vec
[0].bv_len
= STRIPE_SIZE
;
447 bi
->bi_io_vec
[0].bv_offset
= 0;
448 bi
->bi_size
= STRIPE_SIZE
;
451 test_bit(R5_ReWrite
, &sh
->dev
[i
].flags
))
452 atomic_add(STRIPE_SECTORS
,
453 &rdev
->corrected_errors
);
454 generic_make_request(bi
);
457 set_bit(STRIPE_DEGRADED
, &sh
->state
);
458 pr_debug("skip op %ld on disc %d for sector %llu\n",
459 bi
->bi_rw
, i
, (unsigned long long)sh
->sector
);
460 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
461 set_bit(STRIPE_HANDLE
, &sh
->state
);
466 static struct dma_async_tx_descriptor
*
467 async_copy_data(int frombio
, struct bio
*bio
, struct page
*page
,
468 sector_t sector
, struct dma_async_tx_descriptor
*tx
)
471 struct page
*bio_page
;
475 if (bio
->bi_sector
>= sector
)
476 page_offset
= (signed)(bio
->bi_sector
- sector
) * 512;
478 page_offset
= (signed)(sector
- bio
->bi_sector
) * -512;
479 bio_for_each_segment(bvl
, bio
, i
) {
480 int len
= bio_iovec_idx(bio
, i
)->bv_len
;
484 if (page_offset
< 0) {
485 b_offset
= -page_offset
;
486 page_offset
+= b_offset
;
490 if (len
> 0 && page_offset
+ len
> STRIPE_SIZE
)
491 clen
= STRIPE_SIZE
- page_offset
;
496 b_offset
+= bio_iovec_idx(bio
, i
)->bv_offset
;
497 bio_page
= bio_iovec_idx(bio
, i
)->bv_page
;
499 tx
= async_memcpy(page
, bio_page
, page_offset
,
504 tx
= async_memcpy(bio_page
, page
, b_offset
,
509 if (clen
< len
) /* hit end of page */
517 static void ops_complete_biofill(void *stripe_head_ref
)
519 struct stripe_head
*sh
= stripe_head_ref
;
520 struct bio
*return_bi
= NULL
;
521 raid5_conf_t
*conf
= sh
->raid_conf
;
524 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
525 (unsigned long long)sh
->sector
);
527 /* clear completed biofills */
528 for (i
= sh
->disks
; i
--; ) {
529 struct r5dev
*dev
= &sh
->dev
[i
];
531 /* acknowledge completion of a biofill operation */
532 /* and check if we need to reply to a read request,
533 * new R5_Wantfill requests are held off until
534 * !test_bit(STRIPE_OP_BIOFILL, &sh->ops.pending)
536 if (test_and_clear_bit(R5_Wantfill
, &dev
->flags
)) {
537 struct bio
*rbi
, *rbi2
;
539 /* The access to dev->read is outside of the
540 * spin_lock_irq(&conf->device_lock), but is protected
541 * by the STRIPE_OP_BIOFILL pending bit
546 while (rbi
&& rbi
->bi_sector
<
547 dev
->sector
+ STRIPE_SECTORS
) {
548 rbi2
= r5_next_bio(rbi
, dev
->sector
);
549 spin_lock_irq(&conf
->device_lock
);
550 if (--rbi
->bi_phys_segments
== 0) {
551 rbi
->bi_next
= return_bi
;
554 spin_unlock_irq(&conf
->device_lock
);
559 set_bit(STRIPE_OP_BIOFILL
, &sh
->ops
.complete
);
561 return_io(return_bi
);
563 set_bit(STRIPE_HANDLE
, &sh
->state
);
567 static void ops_run_biofill(struct stripe_head
*sh
)
569 struct dma_async_tx_descriptor
*tx
= NULL
;
570 raid5_conf_t
*conf
= sh
->raid_conf
;
573 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
574 (unsigned long long)sh
->sector
);
576 for (i
= sh
->disks
; i
--; ) {
577 struct r5dev
*dev
= &sh
->dev
[i
];
578 if (test_bit(R5_Wantfill
, &dev
->flags
)) {
580 spin_lock_irq(&conf
->device_lock
);
581 dev
->read
= rbi
= dev
->toread
;
583 spin_unlock_irq(&conf
->device_lock
);
584 while (rbi
&& rbi
->bi_sector
<
585 dev
->sector
+ STRIPE_SECTORS
) {
586 tx
= async_copy_data(0, rbi
, dev
->page
,
588 rbi
= r5_next_bio(rbi
, dev
->sector
);
593 atomic_inc(&sh
->count
);
594 async_trigger_callback(ASYNC_TX_DEP_ACK
| ASYNC_TX_ACK
, tx
,
595 ops_complete_biofill
, sh
);
598 static void ops_complete_compute5(void *stripe_head_ref
)
600 struct stripe_head
*sh
= stripe_head_ref
;
601 int target
= sh
->ops
.target
;
602 struct r5dev
*tgt
= &sh
->dev
[target
];
604 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
605 (unsigned long long)sh
->sector
);
607 set_bit(R5_UPTODATE
, &tgt
->flags
);
608 BUG_ON(!test_bit(R5_Wantcompute
, &tgt
->flags
));
609 clear_bit(R5_Wantcompute
, &tgt
->flags
);
610 set_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.complete
);
611 set_bit(STRIPE_HANDLE
, &sh
->state
);
615 static struct dma_async_tx_descriptor
*
616 ops_run_compute5(struct stripe_head
*sh
, unsigned long pending
)
618 /* kernel stack size limits the total number of disks */
619 int disks
= sh
->disks
;
620 struct page
*xor_srcs
[disks
];
621 int target
= sh
->ops
.target
;
622 struct r5dev
*tgt
= &sh
->dev
[target
];
623 struct page
*xor_dest
= tgt
->page
;
625 struct dma_async_tx_descriptor
*tx
;
628 pr_debug("%s: stripe %llu block: %d\n",
629 __FUNCTION__
, (unsigned long long)sh
->sector
, target
);
630 BUG_ON(!test_bit(R5_Wantcompute
, &tgt
->flags
));
632 for (i
= disks
; i
--; )
634 xor_srcs
[count
++] = sh
->dev
[i
].page
;
636 atomic_inc(&sh
->count
);
638 if (unlikely(count
== 1))
639 tx
= async_memcpy(xor_dest
, xor_srcs
[0], 0, 0, STRIPE_SIZE
,
640 0, NULL
, ops_complete_compute5
, sh
);
642 tx
= async_xor(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
,
643 ASYNC_TX_XOR_ZERO_DST
, NULL
,
644 ops_complete_compute5
, sh
);
646 /* ack now if postxor is not set to be run */
647 if (tx
&& !test_bit(STRIPE_OP_POSTXOR
, &pending
))
653 static void ops_complete_prexor(void *stripe_head_ref
)
655 struct stripe_head
*sh
= stripe_head_ref
;
657 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
658 (unsigned long long)sh
->sector
);
660 set_bit(STRIPE_OP_PREXOR
, &sh
->ops
.complete
);
663 static struct dma_async_tx_descriptor
*
664 ops_run_prexor(struct stripe_head
*sh
, struct dma_async_tx_descriptor
*tx
)
666 /* kernel stack size limits the total number of disks */
667 int disks
= sh
->disks
;
668 struct page
*xor_srcs
[disks
];
669 int count
= 0, pd_idx
= sh
->pd_idx
, i
;
671 /* existing parity data subtracted */
672 struct page
*xor_dest
= xor_srcs
[count
++] = sh
->dev
[pd_idx
].page
;
674 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
675 (unsigned long long)sh
->sector
);
677 for (i
= disks
; i
--; ) {
678 struct r5dev
*dev
= &sh
->dev
[i
];
679 /* Only process blocks that are known to be uptodate */
680 if (dev
->towrite
&& test_bit(R5_Wantprexor
, &dev
->flags
))
681 xor_srcs
[count
++] = dev
->page
;
684 tx
= async_xor(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
,
685 ASYNC_TX_DEP_ACK
| ASYNC_TX_XOR_DROP_DST
, tx
,
686 ops_complete_prexor
, sh
);
691 static struct dma_async_tx_descriptor
*
692 ops_run_biodrain(struct stripe_head
*sh
, struct dma_async_tx_descriptor
*tx
,
693 unsigned long pending
)
695 int disks
= sh
->disks
;
696 int pd_idx
= sh
->pd_idx
, i
;
698 /* check if prexor is active which means only process blocks
699 * that are part of a read-modify-write (Wantprexor)
701 int prexor
= test_bit(STRIPE_OP_PREXOR
, &pending
);
703 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
704 (unsigned long long)sh
->sector
);
706 for (i
= disks
; i
--; ) {
707 struct r5dev
*dev
= &sh
->dev
[i
];
712 if (prexor
) { /* rmw */
714 test_bit(R5_Wantprexor
, &dev
->flags
))
717 if (i
!= pd_idx
&& dev
->towrite
&&
718 test_bit(R5_LOCKED
, &dev
->flags
))
725 spin_lock(&sh
->lock
);
726 chosen
= dev
->towrite
;
728 BUG_ON(dev
->written
);
729 wbi
= dev
->written
= chosen
;
730 spin_unlock(&sh
->lock
);
732 while (wbi
&& wbi
->bi_sector
<
733 dev
->sector
+ STRIPE_SECTORS
) {
734 tx
= async_copy_data(1, wbi
, dev
->page
,
736 wbi
= r5_next_bio(wbi
, dev
->sector
);
744 static void ops_complete_postxor(void *stripe_head_ref
)
746 struct stripe_head
*sh
= stripe_head_ref
;
748 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
749 (unsigned long long)sh
->sector
);
751 set_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.complete
);
752 set_bit(STRIPE_HANDLE
, &sh
->state
);
756 static void ops_complete_write(void *stripe_head_ref
)
758 struct stripe_head
*sh
= stripe_head_ref
;
759 int disks
= sh
->disks
, i
, pd_idx
= sh
->pd_idx
;
761 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
762 (unsigned long long)sh
->sector
);
764 for (i
= disks
; i
--; ) {
765 struct r5dev
*dev
= &sh
->dev
[i
];
766 if (dev
->written
|| i
== pd_idx
)
767 set_bit(R5_UPTODATE
, &dev
->flags
);
770 set_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.complete
);
771 set_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.complete
);
773 set_bit(STRIPE_HANDLE
, &sh
->state
);
778 ops_run_postxor(struct stripe_head
*sh
, struct dma_async_tx_descriptor
*tx
,
779 unsigned long pending
)
781 /* kernel stack size limits the total number of disks */
782 int disks
= sh
->disks
;
783 struct page
*xor_srcs
[disks
];
785 int count
= 0, pd_idx
= sh
->pd_idx
, i
;
786 struct page
*xor_dest
;
787 int prexor
= test_bit(STRIPE_OP_PREXOR
, &pending
);
789 dma_async_tx_callback callback
;
791 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
792 (unsigned long long)sh
->sector
);
794 /* check if prexor is active which means only process blocks
795 * that are part of a read-modify-write (written)
798 xor_dest
= xor_srcs
[count
++] = sh
->dev
[pd_idx
].page
;
799 for (i
= disks
; i
--; ) {
800 struct r5dev
*dev
= &sh
->dev
[i
];
802 xor_srcs
[count
++] = dev
->page
;
805 xor_dest
= sh
->dev
[pd_idx
].page
;
806 for (i
= disks
; i
--; ) {
807 struct r5dev
*dev
= &sh
->dev
[i
];
809 xor_srcs
[count
++] = dev
->page
;
813 /* check whether this postxor is part of a write */
814 callback
= test_bit(STRIPE_OP_BIODRAIN
, &pending
) ?
815 ops_complete_write
: ops_complete_postxor
;
817 /* 1/ if we prexor'd then the dest is reused as a source
818 * 2/ if we did not prexor then we are redoing the parity
819 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
820 * for the synchronous xor case
822 flags
= ASYNC_TX_DEP_ACK
| ASYNC_TX_ACK
|
823 (prexor
? ASYNC_TX_XOR_DROP_DST
: ASYNC_TX_XOR_ZERO_DST
);
825 atomic_inc(&sh
->count
);
827 if (unlikely(count
== 1)) {
828 flags
&= ~(ASYNC_TX_XOR_DROP_DST
| ASYNC_TX_XOR_ZERO_DST
);
829 tx
= async_memcpy(xor_dest
, xor_srcs
[0], 0, 0, STRIPE_SIZE
,
830 flags
, tx
, callback
, sh
);
832 tx
= async_xor(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
,
833 flags
, tx
, callback
, sh
);
836 static void ops_complete_check(void *stripe_head_ref
)
838 struct stripe_head
*sh
= stripe_head_ref
;
839 int pd_idx
= sh
->pd_idx
;
841 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
842 (unsigned long long)sh
->sector
);
844 if (test_and_clear_bit(STRIPE_OP_MOD_DMA_CHECK
, &sh
->ops
.pending
) &&
845 sh
->ops
.zero_sum_result
== 0)
846 set_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
848 set_bit(STRIPE_OP_CHECK
, &sh
->ops
.complete
);
849 set_bit(STRIPE_HANDLE
, &sh
->state
);
853 static void ops_run_check(struct stripe_head
*sh
)
855 /* kernel stack size limits the total number of disks */
856 int disks
= sh
->disks
;
857 struct page
*xor_srcs
[disks
];
858 struct dma_async_tx_descriptor
*tx
;
860 int count
= 0, pd_idx
= sh
->pd_idx
, i
;
861 struct page
*xor_dest
= xor_srcs
[count
++] = sh
->dev
[pd_idx
].page
;
863 pr_debug("%s: stripe %llu\n", __FUNCTION__
,
864 (unsigned long long)sh
->sector
);
866 for (i
= disks
; i
--; ) {
867 struct r5dev
*dev
= &sh
->dev
[i
];
869 xor_srcs
[count
++] = dev
->page
;
872 tx
= async_xor_zero_sum(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
,
873 &sh
->ops
.zero_sum_result
, 0, NULL
, NULL
, NULL
);
876 set_bit(STRIPE_OP_MOD_DMA_CHECK
, &sh
->ops
.pending
);
878 clear_bit(STRIPE_OP_MOD_DMA_CHECK
, &sh
->ops
.pending
);
880 atomic_inc(&sh
->count
);
881 tx
= async_trigger_callback(ASYNC_TX_DEP_ACK
| ASYNC_TX_ACK
, tx
,
882 ops_complete_check
, sh
);
885 static void raid5_run_ops(struct stripe_head
*sh
, unsigned long pending
)
887 int overlap_clear
= 0, i
, disks
= sh
->disks
;
888 struct dma_async_tx_descriptor
*tx
= NULL
;
890 if (test_bit(STRIPE_OP_BIOFILL
, &pending
)) {
895 if (test_bit(STRIPE_OP_COMPUTE_BLK
, &pending
))
896 tx
= ops_run_compute5(sh
, pending
);
898 if (test_bit(STRIPE_OP_PREXOR
, &pending
))
899 tx
= ops_run_prexor(sh
, tx
);
901 if (test_bit(STRIPE_OP_BIODRAIN
, &pending
)) {
902 tx
= ops_run_biodrain(sh
, tx
, pending
);
906 if (test_bit(STRIPE_OP_POSTXOR
, &pending
))
907 ops_run_postxor(sh
, tx
, pending
);
909 if (test_bit(STRIPE_OP_CHECK
, &pending
))
912 if (test_bit(STRIPE_OP_IO
, &pending
))
916 for (i
= disks
; i
--; ) {
917 struct r5dev
*dev
= &sh
->dev
[i
];
918 if (test_and_clear_bit(R5_Overlap
, &dev
->flags
))
919 wake_up(&sh
->raid_conf
->wait_for_overlap
);
923 static int grow_one_stripe(raid5_conf_t
*conf
)
925 struct stripe_head
*sh
;
926 sh
= kmem_cache_alloc(conf
->slab_cache
, GFP_KERNEL
);
929 memset(sh
, 0, sizeof(*sh
) + (conf
->raid_disks
-1)*sizeof(struct r5dev
));
930 sh
->raid_conf
= conf
;
931 spin_lock_init(&sh
->lock
);
933 if (grow_buffers(sh
, conf
->raid_disks
)) {
934 shrink_buffers(sh
, conf
->raid_disks
);
935 kmem_cache_free(conf
->slab_cache
, sh
);
938 sh
->disks
= conf
->raid_disks
;
939 /* we just created an active stripe so... */
940 atomic_set(&sh
->count
, 1);
941 atomic_inc(&conf
->active_stripes
);
942 INIT_LIST_HEAD(&sh
->lru
);
947 static int grow_stripes(raid5_conf_t
*conf
, int num
)
949 struct kmem_cache
*sc
;
950 int devs
= conf
->raid_disks
;
952 sprintf(conf
->cache_name
[0], "raid5-%s", mdname(conf
->mddev
));
953 sprintf(conf
->cache_name
[1], "raid5-%s-alt", mdname(conf
->mddev
));
954 conf
->active_name
= 0;
955 sc
= kmem_cache_create(conf
->cache_name
[conf
->active_name
],
956 sizeof(struct stripe_head
)+(devs
-1)*sizeof(struct r5dev
),
960 conf
->slab_cache
= sc
;
961 conf
->pool_size
= devs
;
963 if (!grow_one_stripe(conf
))
968 #ifdef CONFIG_MD_RAID5_RESHAPE
969 static int resize_stripes(raid5_conf_t
*conf
, int newsize
)
971 /* Make all the stripes able to hold 'newsize' devices.
972 * New slots in each stripe get 'page' set to a new page.
974 * This happens in stages:
975 * 1/ create a new kmem_cache and allocate the required number of
977 * 2/ gather all the old stripe_heads and tranfer the pages across
978 * to the new stripe_heads. This will have the side effect of
979 * freezing the array as once all stripe_heads have been collected,
980 * no IO will be possible. Old stripe heads are freed once their
981 * pages have been transferred over, and the old kmem_cache is
982 * freed when all stripes are done.
983 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
984 * we simple return a failre status - no need to clean anything up.
985 * 4/ allocate new pages for the new slots in the new stripe_heads.
986 * If this fails, we don't bother trying the shrink the
987 * stripe_heads down again, we just leave them as they are.
988 * As each stripe_head is processed the new one is released into
991 * Once step2 is started, we cannot afford to wait for a write,
992 * so we use GFP_NOIO allocations.
994 struct stripe_head
*osh
, *nsh
;
995 LIST_HEAD(newstripes
);
996 struct disk_info
*ndisks
;
998 struct kmem_cache
*sc
;
1001 if (newsize
<= conf
->pool_size
)
1002 return 0; /* never bother to shrink */
1004 md_allow_write(conf
->mddev
);
1007 sc
= kmem_cache_create(conf
->cache_name
[1-conf
->active_name
],
1008 sizeof(struct stripe_head
)+(newsize
-1)*sizeof(struct r5dev
),
1013 for (i
= conf
->max_nr_stripes
; i
; i
--) {
1014 nsh
= kmem_cache_alloc(sc
, GFP_KERNEL
);
1018 memset(nsh
, 0, sizeof(*nsh
) + (newsize
-1)*sizeof(struct r5dev
));
1020 nsh
->raid_conf
= conf
;
1021 spin_lock_init(&nsh
->lock
);
1023 list_add(&nsh
->lru
, &newstripes
);
1026 /* didn't get enough, give up */
1027 while (!list_empty(&newstripes
)) {
1028 nsh
= list_entry(newstripes
.next
, struct stripe_head
, lru
);
1029 list_del(&nsh
->lru
);
1030 kmem_cache_free(sc
, nsh
);
1032 kmem_cache_destroy(sc
);
1035 /* Step 2 - Must use GFP_NOIO now.
1036 * OK, we have enough stripes, start collecting inactive
1037 * stripes and copying them over
1039 list_for_each_entry(nsh
, &newstripes
, lru
) {
1040 spin_lock_irq(&conf
->device_lock
);
1041 wait_event_lock_irq(conf
->wait_for_stripe
,
1042 !list_empty(&conf
->inactive_list
),
1044 unplug_slaves(conf
->mddev
)
1046 osh
= get_free_stripe(conf
);
1047 spin_unlock_irq(&conf
->device_lock
);
1048 atomic_set(&nsh
->count
, 1);
1049 for(i
=0; i
<conf
->pool_size
; i
++)
1050 nsh
->dev
[i
].page
= osh
->dev
[i
].page
;
1051 for( ; i
<newsize
; i
++)
1052 nsh
->dev
[i
].page
= NULL
;
1053 kmem_cache_free(conf
->slab_cache
, osh
);
1055 kmem_cache_destroy(conf
->slab_cache
);
1058 * At this point, we are holding all the stripes so the array
1059 * is completely stalled, so now is a good time to resize
1062 ndisks
= kzalloc(newsize
* sizeof(struct disk_info
), GFP_NOIO
);
1064 for (i
=0; i
<conf
->raid_disks
; i
++)
1065 ndisks
[i
] = conf
->disks
[i
];
1067 conf
->disks
= ndisks
;
1071 /* Step 4, return new stripes to service */
1072 while(!list_empty(&newstripes
)) {
1073 nsh
= list_entry(newstripes
.next
, struct stripe_head
, lru
);
1074 list_del_init(&nsh
->lru
);
1075 for (i
=conf
->raid_disks
; i
< newsize
; i
++)
1076 if (nsh
->dev
[i
].page
== NULL
) {
1077 struct page
*p
= alloc_page(GFP_NOIO
);
1078 nsh
->dev
[i
].page
= p
;
1082 release_stripe(nsh
);
1084 /* critical section pass, GFP_NOIO no longer needed */
1086 conf
->slab_cache
= sc
;
1087 conf
->active_name
= 1-conf
->active_name
;
1088 conf
->pool_size
= newsize
;
1093 static int drop_one_stripe(raid5_conf_t
*conf
)
1095 struct stripe_head
*sh
;
1097 spin_lock_irq(&conf
->device_lock
);
1098 sh
= get_free_stripe(conf
);
1099 spin_unlock_irq(&conf
->device_lock
);
1102 BUG_ON(atomic_read(&sh
->count
));
1103 shrink_buffers(sh
, conf
->pool_size
);
1104 kmem_cache_free(conf
->slab_cache
, sh
);
1105 atomic_dec(&conf
->active_stripes
);
1109 static void shrink_stripes(raid5_conf_t
*conf
)
1111 while (drop_one_stripe(conf
))
1114 if (conf
->slab_cache
)
1115 kmem_cache_destroy(conf
->slab_cache
);
1116 conf
->slab_cache
= NULL
;
1119 static int raid5_end_read_request(struct bio
* bi
, unsigned int bytes_done
,
1122 struct stripe_head
*sh
= bi
->bi_private
;
1123 raid5_conf_t
*conf
= sh
->raid_conf
;
1124 int disks
= sh
->disks
, i
;
1125 int uptodate
= test_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1126 char b
[BDEVNAME_SIZE
];
1132 for (i
=0 ; i
<disks
; i
++)
1133 if (bi
== &sh
->dev
[i
].req
)
1136 pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n",
1137 (unsigned long long)sh
->sector
, i
, atomic_read(&sh
->count
),
1145 set_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
);
1146 if (test_bit(R5_ReadError
, &sh
->dev
[i
].flags
)) {
1147 rdev
= conf
->disks
[i
].rdev
;
1148 printk(KERN_INFO
"raid5:%s: read error corrected (%lu sectors at %llu on %s)\n",
1149 mdname(conf
->mddev
), STRIPE_SECTORS
,
1150 (unsigned long long)sh
->sector
+ rdev
->data_offset
,
1151 bdevname(rdev
->bdev
, b
));
1152 clear_bit(R5_ReadError
, &sh
->dev
[i
].flags
);
1153 clear_bit(R5_ReWrite
, &sh
->dev
[i
].flags
);
1155 if (atomic_read(&conf
->disks
[i
].rdev
->read_errors
))
1156 atomic_set(&conf
->disks
[i
].rdev
->read_errors
, 0);
1158 const char *bdn
= bdevname(conf
->disks
[i
].rdev
->bdev
, b
);
1160 rdev
= conf
->disks
[i
].rdev
;
1162 clear_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
);
1163 atomic_inc(&rdev
->read_errors
);
1164 if (conf
->mddev
->degraded
)
1165 printk(KERN_WARNING
"raid5:%s: read error not correctable (sector %llu on %s).\n",
1166 mdname(conf
->mddev
),
1167 (unsigned long long)sh
->sector
+ rdev
->data_offset
,
1169 else if (test_bit(R5_ReWrite
, &sh
->dev
[i
].flags
))
1171 printk(KERN_WARNING
"raid5:%s: read error NOT corrected!! (sector %llu on %s).\n",
1172 mdname(conf
->mddev
),
1173 (unsigned long long)sh
->sector
+ rdev
->data_offset
,
1175 else if (atomic_read(&rdev
->read_errors
)
1176 > conf
->max_nr_stripes
)
1178 "raid5:%s: Too many read errors, failing device %s.\n",
1179 mdname(conf
->mddev
), bdn
);
1183 set_bit(R5_ReadError
, &sh
->dev
[i
].flags
);
1185 clear_bit(R5_ReadError
, &sh
->dev
[i
].flags
);
1186 clear_bit(R5_ReWrite
, &sh
->dev
[i
].flags
);
1187 md_error(conf
->mddev
, rdev
);
1190 rdev_dec_pending(conf
->disks
[i
].rdev
, conf
->mddev
);
1191 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
1192 set_bit(STRIPE_HANDLE
, &sh
->state
);
1197 static int raid5_end_write_request (struct bio
*bi
, unsigned int bytes_done
,
1200 struct stripe_head
*sh
= bi
->bi_private
;
1201 raid5_conf_t
*conf
= sh
->raid_conf
;
1202 int disks
= sh
->disks
, i
;
1203 int uptodate
= test_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1208 for (i
=0 ; i
<disks
; i
++)
1209 if (bi
== &sh
->dev
[i
].req
)
1212 pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n",
1213 (unsigned long long)sh
->sector
, i
, atomic_read(&sh
->count
),
1221 md_error(conf
->mddev
, conf
->disks
[i
].rdev
);
1223 rdev_dec_pending(conf
->disks
[i
].rdev
, conf
->mddev
);
1225 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
1226 set_bit(STRIPE_HANDLE
, &sh
->state
);
1232 static sector_t
compute_blocknr(struct stripe_head
*sh
, int i
);
1234 static void raid5_build_block (struct stripe_head
*sh
, int i
)
1236 struct r5dev
*dev
= &sh
->dev
[i
];
1238 bio_init(&dev
->req
);
1239 dev
->req
.bi_io_vec
= &dev
->vec
;
1241 dev
->req
.bi_max_vecs
++;
1242 dev
->vec
.bv_page
= dev
->page
;
1243 dev
->vec
.bv_len
= STRIPE_SIZE
;
1244 dev
->vec
.bv_offset
= 0;
1246 dev
->req
.bi_sector
= sh
->sector
;
1247 dev
->req
.bi_private
= sh
;
1250 dev
->sector
= compute_blocknr(sh
, i
);
1253 static void error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1255 char b
[BDEVNAME_SIZE
];
1256 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
1257 pr_debug("raid5: error called\n");
1259 if (!test_bit(Faulty
, &rdev
->flags
)) {
1260 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
1261 if (test_and_clear_bit(In_sync
, &rdev
->flags
)) {
1262 unsigned long flags
;
1263 spin_lock_irqsave(&conf
->device_lock
, flags
);
1265 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1267 * if recovery was running, make sure it aborts.
1269 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
1271 set_bit(Faulty
, &rdev
->flags
);
1273 "raid5: Disk failure on %s, disabling device."
1274 " Operation continuing on %d devices\n",
1275 bdevname(rdev
->bdev
,b
), conf
->raid_disks
- mddev
->degraded
);
1280 * Input: a 'big' sector number,
1281 * Output: index of the data and parity disk, and the sector # in them.
1283 static sector_t
raid5_compute_sector(sector_t r_sector
, unsigned int raid_disks
,
1284 unsigned int data_disks
, unsigned int * dd_idx
,
1285 unsigned int * pd_idx
, raid5_conf_t
*conf
)
1288 unsigned long chunk_number
;
1289 unsigned int chunk_offset
;
1290 sector_t new_sector
;
1291 int sectors_per_chunk
= conf
->chunk_size
>> 9;
1293 /* First compute the information on this sector */
1296 * Compute the chunk number and the sector offset inside the chunk
1298 chunk_offset
= sector_div(r_sector
, sectors_per_chunk
);
1299 chunk_number
= r_sector
;
1300 BUG_ON(r_sector
!= chunk_number
);
1303 * Compute the stripe number
1305 stripe
= chunk_number
/ data_disks
;
1308 * Compute the data disk and parity disk indexes inside the stripe
1310 *dd_idx
= chunk_number
% data_disks
;
1313 * Select the parity disk based on the user selected algorithm.
1315 switch(conf
->level
) {
1317 *pd_idx
= data_disks
;
1320 switch (conf
->algorithm
) {
1321 case ALGORITHM_LEFT_ASYMMETRIC
:
1322 *pd_idx
= data_disks
- stripe
% raid_disks
;
1323 if (*dd_idx
>= *pd_idx
)
1326 case ALGORITHM_RIGHT_ASYMMETRIC
:
1327 *pd_idx
= stripe
% raid_disks
;
1328 if (*dd_idx
>= *pd_idx
)
1331 case ALGORITHM_LEFT_SYMMETRIC
:
1332 *pd_idx
= data_disks
- stripe
% raid_disks
;
1333 *dd_idx
= (*pd_idx
+ 1 + *dd_idx
) % raid_disks
;
1335 case ALGORITHM_RIGHT_SYMMETRIC
:
1336 *pd_idx
= stripe
% raid_disks
;
1337 *dd_idx
= (*pd_idx
+ 1 + *dd_idx
) % raid_disks
;
1340 printk(KERN_ERR
"raid5: unsupported algorithm %d\n",
1346 /**** FIX THIS ****/
1347 switch (conf
->algorithm
) {
1348 case ALGORITHM_LEFT_ASYMMETRIC
:
1349 *pd_idx
= raid_disks
- 1 - (stripe
% raid_disks
);
1350 if (*pd_idx
== raid_disks
-1)
1351 (*dd_idx
)++; /* Q D D D P */
1352 else if (*dd_idx
>= *pd_idx
)
1353 (*dd_idx
) += 2; /* D D P Q D */
1355 case ALGORITHM_RIGHT_ASYMMETRIC
:
1356 *pd_idx
= stripe
% raid_disks
;
1357 if (*pd_idx
== raid_disks
-1)
1358 (*dd_idx
)++; /* Q D D D P */
1359 else if (*dd_idx
>= *pd_idx
)
1360 (*dd_idx
) += 2; /* D D P Q D */
1362 case ALGORITHM_LEFT_SYMMETRIC
:
1363 *pd_idx
= raid_disks
- 1 - (stripe
% raid_disks
);
1364 *dd_idx
= (*pd_idx
+ 2 + *dd_idx
) % raid_disks
;
1366 case ALGORITHM_RIGHT_SYMMETRIC
:
1367 *pd_idx
= stripe
% raid_disks
;
1368 *dd_idx
= (*pd_idx
+ 2 + *dd_idx
) % raid_disks
;
1371 printk (KERN_CRIT
"raid6: unsupported algorithm %d\n",
1378 * Finally, compute the new sector number
1380 new_sector
= (sector_t
)stripe
* sectors_per_chunk
+ chunk_offset
;
1385 static sector_t
compute_blocknr(struct stripe_head
*sh
, int i
)
1387 raid5_conf_t
*conf
= sh
->raid_conf
;
1388 int raid_disks
= sh
->disks
;
1389 int data_disks
= raid_disks
- conf
->max_degraded
;
1390 sector_t new_sector
= sh
->sector
, check
;
1391 int sectors_per_chunk
= conf
->chunk_size
>> 9;
1394 int chunk_number
, dummy1
, dummy2
, dd_idx
= i
;
1398 chunk_offset
= sector_div(new_sector
, sectors_per_chunk
);
1399 stripe
= new_sector
;
1400 BUG_ON(new_sector
!= stripe
);
1402 if (i
== sh
->pd_idx
)
1404 switch(conf
->level
) {
1407 switch (conf
->algorithm
) {
1408 case ALGORITHM_LEFT_ASYMMETRIC
:
1409 case ALGORITHM_RIGHT_ASYMMETRIC
:
1413 case ALGORITHM_LEFT_SYMMETRIC
:
1414 case ALGORITHM_RIGHT_SYMMETRIC
:
1417 i
-= (sh
->pd_idx
+ 1);
1420 printk(KERN_ERR
"raid5: unsupported algorithm %d\n",
1425 if (i
== raid6_next_disk(sh
->pd_idx
, raid_disks
))
1426 return 0; /* It is the Q disk */
1427 switch (conf
->algorithm
) {
1428 case ALGORITHM_LEFT_ASYMMETRIC
:
1429 case ALGORITHM_RIGHT_ASYMMETRIC
:
1430 if (sh
->pd_idx
== raid_disks
-1)
1431 i
--; /* Q D D D P */
1432 else if (i
> sh
->pd_idx
)
1433 i
-= 2; /* D D P Q D */
1435 case ALGORITHM_LEFT_SYMMETRIC
:
1436 case ALGORITHM_RIGHT_SYMMETRIC
:
1437 if (sh
->pd_idx
== raid_disks
-1)
1438 i
--; /* Q D D D P */
1443 i
-= (sh
->pd_idx
+ 2);
1447 printk (KERN_CRIT
"raid6: unsupported algorithm %d\n",
1453 chunk_number
= stripe
* data_disks
+ i
;
1454 r_sector
= (sector_t
)chunk_number
* sectors_per_chunk
+ chunk_offset
;
1456 check
= raid5_compute_sector (r_sector
, raid_disks
, data_disks
, &dummy1
, &dummy2
, conf
);
1457 if (check
!= sh
->sector
|| dummy1
!= dd_idx
|| dummy2
!= sh
->pd_idx
) {
1458 printk(KERN_ERR
"compute_blocknr: map not correct\n");
1467 * Copy data between a page in the stripe cache, and one or more bion
1468 * The page could align with the middle of the bio, or there could be
1469 * several bion, each with several bio_vecs, which cover part of the page
1470 * Multiple bion are linked together on bi_next. There may be extras
1471 * at the end of this list. We ignore them.
1473 static void copy_data(int frombio
, struct bio
*bio
,
1477 char *pa
= page_address(page
);
1478 struct bio_vec
*bvl
;
1482 if (bio
->bi_sector
>= sector
)
1483 page_offset
= (signed)(bio
->bi_sector
- sector
) * 512;
1485 page_offset
= (signed)(sector
- bio
->bi_sector
) * -512;
1486 bio_for_each_segment(bvl
, bio
, i
) {
1487 int len
= bio_iovec_idx(bio
,i
)->bv_len
;
1491 if (page_offset
< 0) {
1492 b_offset
= -page_offset
;
1493 page_offset
+= b_offset
;
1497 if (len
> 0 && page_offset
+ len
> STRIPE_SIZE
)
1498 clen
= STRIPE_SIZE
- page_offset
;
1502 char *ba
= __bio_kmap_atomic(bio
, i
, KM_USER0
);
1504 memcpy(pa
+page_offset
, ba
+b_offset
, clen
);
1506 memcpy(ba
+b_offset
, pa
+page_offset
, clen
);
1507 __bio_kunmap_atomic(ba
, KM_USER0
);
1509 if (clen
< len
) /* hit end of page */
1515 #define check_xor() do { \
1516 if (count == MAX_XOR_BLOCKS) { \
1517 xor_blocks(count, STRIPE_SIZE, dest, ptr);\
1522 static void compute_parity6(struct stripe_head
*sh
, int method
)
1524 raid6_conf_t
*conf
= sh
->raid_conf
;
1525 int i
, pd_idx
= sh
->pd_idx
, qd_idx
, d0_idx
, disks
= sh
->disks
, count
;
1527 /**** FIX THIS: This could be very bad if disks is close to 256 ****/
1530 qd_idx
= raid6_next_disk(pd_idx
, disks
);
1531 d0_idx
= raid6_next_disk(qd_idx
, disks
);
1533 pr_debug("compute_parity, stripe %llu, method %d\n",
1534 (unsigned long long)sh
->sector
, method
);
1537 case READ_MODIFY_WRITE
:
1538 BUG(); /* READ_MODIFY_WRITE N/A for RAID-6 */
1539 case RECONSTRUCT_WRITE
:
1540 for (i
= disks
; i
-- ;)
1541 if ( i
!= pd_idx
&& i
!= qd_idx
&& sh
->dev
[i
].towrite
) {
1542 chosen
= sh
->dev
[i
].towrite
;
1543 sh
->dev
[i
].towrite
= NULL
;
1545 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
1546 wake_up(&conf
->wait_for_overlap
);
1548 BUG_ON(sh
->dev
[i
].written
);
1549 sh
->dev
[i
].written
= chosen
;
1553 BUG(); /* Not implemented yet */
1556 for (i
= disks
; i
--;)
1557 if (sh
->dev
[i
].written
) {
1558 sector_t sector
= sh
->dev
[i
].sector
;
1559 struct bio
*wbi
= sh
->dev
[i
].written
;
1560 while (wbi
&& wbi
->bi_sector
< sector
+ STRIPE_SECTORS
) {
1561 copy_data(1, wbi
, sh
->dev
[i
].page
, sector
);
1562 wbi
= r5_next_bio(wbi
, sector
);
1565 set_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
1566 set_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
);
1570 // case RECONSTRUCT_WRITE:
1571 // case CHECK_PARITY:
1572 // case UPDATE_PARITY:
1573 /* Note that unlike RAID-5, the ordering of the disks matters greatly. */
1574 /* FIX: Is this ordering of drives even remotely optimal? */
1578 ptrs
[count
++] = page_address(sh
->dev
[i
].page
);
1579 if (count
<= disks
-2 && !test_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
))
1580 printk("block %d/%d not uptodate on parity calc\n", i
,count
);
1581 i
= raid6_next_disk(i
, disks
);
1582 } while ( i
!= d0_idx
);
1586 raid6_call
.gen_syndrome(disks
, STRIPE_SIZE
, ptrs
);
1589 case RECONSTRUCT_WRITE
:
1590 set_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
1591 set_bit(R5_UPTODATE
, &sh
->dev
[qd_idx
].flags
);
1592 set_bit(R5_LOCKED
, &sh
->dev
[pd_idx
].flags
);
1593 set_bit(R5_LOCKED
, &sh
->dev
[qd_idx
].flags
);
1596 set_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
1597 set_bit(R5_UPTODATE
, &sh
->dev
[qd_idx
].flags
);
1603 /* Compute one missing block */
1604 static void compute_block_1(struct stripe_head
*sh
, int dd_idx
, int nozero
)
1606 int i
, count
, disks
= sh
->disks
;
1607 void *ptr
[MAX_XOR_BLOCKS
], *dest
, *p
;
1608 int pd_idx
= sh
->pd_idx
;
1609 int qd_idx
= raid6_next_disk(pd_idx
, disks
);
1611 pr_debug("compute_block_1, stripe %llu, idx %d\n",
1612 (unsigned long long)sh
->sector
, dd_idx
);
1614 if ( dd_idx
== qd_idx
) {
1615 /* We're actually computing the Q drive */
1616 compute_parity6(sh
, UPDATE_PARITY
);
1618 dest
= page_address(sh
->dev
[dd_idx
].page
);
1619 if (!nozero
) memset(dest
, 0, STRIPE_SIZE
);
1621 for (i
= disks
; i
--; ) {
1622 if (i
== dd_idx
|| i
== qd_idx
)
1624 p
= page_address(sh
->dev
[i
].page
);
1625 if (test_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
))
1628 printk("compute_block() %d, stripe %llu, %d"
1629 " not present\n", dd_idx
,
1630 (unsigned long long)sh
->sector
, i
);
1635 xor_blocks(count
, STRIPE_SIZE
, dest
, ptr
);
1636 if (!nozero
) set_bit(R5_UPTODATE
, &sh
->dev
[dd_idx
].flags
);
1637 else clear_bit(R5_UPTODATE
, &sh
->dev
[dd_idx
].flags
);
1641 /* Compute two missing blocks */
1642 static void compute_block_2(struct stripe_head
*sh
, int dd_idx1
, int dd_idx2
)
1644 int i
, count
, disks
= sh
->disks
;
1645 int pd_idx
= sh
->pd_idx
;
1646 int qd_idx
= raid6_next_disk(pd_idx
, disks
);
1647 int d0_idx
= raid6_next_disk(qd_idx
, disks
);
1650 /* faila and failb are disk numbers relative to d0_idx */
1651 /* pd_idx become disks-2 and qd_idx become disks-1 */
1652 faila
= (dd_idx1
< d0_idx
) ? dd_idx1
+(disks
-d0_idx
) : dd_idx1
-d0_idx
;
1653 failb
= (dd_idx2
< d0_idx
) ? dd_idx2
+(disks
-d0_idx
) : dd_idx2
-d0_idx
;
1655 BUG_ON(faila
== failb
);
1656 if ( failb
< faila
) { int tmp
= faila
; faila
= failb
; failb
= tmp
; }
1658 pr_debug("compute_block_2, stripe %llu, idx %d,%d (%d,%d)\n",
1659 (unsigned long long)sh
->sector
, dd_idx1
, dd_idx2
, faila
, failb
);
1661 if ( failb
== disks
-1 ) {
1662 /* Q disk is one of the missing disks */
1663 if ( faila
== disks
-2 ) {
1664 /* Missing P+Q, just recompute */
1665 compute_parity6(sh
, UPDATE_PARITY
);
1668 /* We're missing D+Q; recompute D from P */
1669 compute_block_1(sh
, (dd_idx1
== qd_idx
) ? dd_idx2
: dd_idx1
, 0);
1670 compute_parity6(sh
, UPDATE_PARITY
); /* Is this necessary? */
1675 /* We're missing D+P or D+D; build pointer table */
1677 /**** FIX THIS: This could be very bad if disks is close to 256 ****/
1683 ptrs
[count
++] = page_address(sh
->dev
[i
].page
);
1684 i
= raid6_next_disk(i
, disks
);
1685 if (i
!= dd_idx1
&& i
!= dd_idx2
&&
1686 !test_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
))
1687 printk("compute_2 with missing block %d/%d\n", count
, i
);
1688 } while ( i
!= d0_idx
);
1690 if ( failb
== disks
-2 ) {
1691 /* We're missing D+P. */
1692 raid6_datap_recov(disks
, STRIPE_SIZE
, faila
, ptrs
);
1694 /* We're missing D+D. */
1695 raid6_2data_recov(disks
, STRIPE_SIZE
, faila
, failb
, ptrs
);
1698 /* Both the above update both missing blocks */
1699 set_bit(R5_UPTODATE
, &sh
->dev
[dd_idx1
].flags
);
1700 set_bit(R5_UPTODATE
, &sh
->dev
[dd_idx2
].flags
);
1705 handle_write_operations5(struct stripe_head
*sh
, int rcw
, int expand
)
1707 int i
, pd_idx
= sh
->pd_idx
, disks
= sh
->disks
;
1711 /* if we are not expanding this is a proper write request, and
1712 * there will be bios with new data to be drained into the
1716 set_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.pending
);
1720 set_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
);
1723 for (i
= disks
; i
--; ) {
1724 struct r5dev
*dev
= &sh
->dev
[i
];
1727 set_bit(R5_LOCKED
, &dev
->flags
);
1729 clear_bit(R5_UPTODATE
, &dev
->flags
);
1734 BUG_ON(!(test_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
) ||
1735 test_bit(R5_Wantcompute
, &sh
->dev
[pd_idx
].flags
)));
1737 set_bit(STRIPE_OP_PREXOR
, &sh
->ops
.pending
);
1738 set_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.pending
);
1739 set_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
);
1743 for (i
= disks
; i
--; ) {
1744 struct r5dev
*dev
= &sh
->dev
[i
];
1748 /* For a read-modify write there may be blocks that are
1749 * locked for reading while others are ready to be
1750 * written so we distinguish these blocks by the
1754 (test_bit(R5_UPTODATE
, &dev
->flags
) ||
1755 test_bit(R5_Wantcompute
, &dev
->flags
))) {
1756 set_bit(R5_Wantprexor
, &dev
->flags
);
1757 set_bit(R5_LOCKED
, &dev
->flags
);
1758 clear_bit(R5_UPTODATE
, &dev
->flags
);
1764 /* keep the parity disk locked while asynchronous operations
1767 set_bit(R5_LOCKED
, &sh
->dev
[pd_idx
].flags
);
1768 clear_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
1771 pr_debug("%s: stripe %llu locked: %d pending: %lx\n",
1772 __FUNCTION__
, (unsigned long long)sh
->sector
,
1773 locked
, sh
->ops
.pending
);
1779 * Each stripe/dev can have one or more bion attached.
1780 * toread/towrite point to the first in a chain.
1781 * The bi_next chain must be in order.
1783 static int add_stripe_bio(struct stripe_head
*sh
, struct bio
*bi
, int dd_idx
, int forwrite
)
1786 raid5_conf_t
*conf
= sh
->raid_conf
;
1789 pr_debug("adding bh b#%llu to stripe s#%llu\n",
1790 (unsigned long long)bi
->bi_sector
,
1791 (unsigned long long)sh
->sector
);
1794 spin_lock(&sh
->lock
);
1795 spin_lock_irq(&conf
->device_lock
);
1797 bip
= &sh
->dev
[dd_idx
].towrite
;
1798 if (*bip
== NULL
&& sh
->dev
[dd_idx
].written
== NULL
)
1801 bip
= &sh
->dev
[dd_idx
].toread
;
1802 while (*bip
&& (*bip
)->bi_sector
< bi
->bi_sector
) {
1803 if ((*bip
)->bi_sector
+ ((*bip
)->bi_size
>> 9) > bi
->bi_sector
)
1805 bip
= & (*bip
)->bi_next
;
1807 if (*bip
&& (*bip
)->bi_sector
< bi
->bi_sector
+ ((bi
->bi_size
)>>9))
1810 BUG_ON(*bip
&& bi
->bi_next
&& (*bip
) != bi
->bi_next
);
1814 bi
->bi_phys_segments
++;
1815 spin_unlock_irq(&conf
->device_lock
);
1816 spin_unlock(&sh
->lock
);
1818 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
1819 (unsigned long long)bi
->bi_sector
,
1820 (unsigned long long)sh
->sector
, dd_idx
);
1822 if (conf
->mddev
->bitmap
&& firstwrite
) {
1823 bitmap_startwrite(conf
->mddev
->bitmap
, sh
->sector
,
1825 sh
->bm_seq
= conf
->seq_flush
+1;
1826 set_bit(STRIPE_BIT_DELAY
, &sh
->state
);
1830 /* check if page is covered */
1831 sector_t sector
= sh
->dev
[dd_idx
].sector
;
1832 for (bi
=sh
->dev
[dd_idx
].towrite
;
1833 sector
< sh
->dev
[dd_idx
].sector
+ STRIPE_SECTORS
&&
1834 bi
&& bi
->bi_sector
<= sector
;
1835 bi
= r5_next_bio(bi
, sh
->dev
[dd_idx
].sector
)) {
1836 if (bi
->bi_sector
+ (bi
->bi_size
>>9) >= sector
)
1837 sector
= bi
->bi_sector
+ (bi
->bi_size
>>9);
1839 if (sector
>= sh
->dev
[dd_idx
].sector
+ STRIPE_SECTORS
)
1840 set_bit(R5_OVERWRITE
, &sh
->dev
[dd_idx
].flags
);
1845 set_bit(R5_Overlap
, &sh
->dev
[dd_idx
].flags
);
1846 spin_unlock_irq(&conf
->device_lock
);
1847 spin_unlock(&sh
->lock
);
1851 static void end_reshape(raid5_conf_t
*conf
);
1853 static int page_is_zero(struct page
*p
)
1855 char *a
= page_address(p
);
1856 return ((*(u32
*)a
) == 0 &&
1857 memcmp(a
, a
+4, STRIPE_SIZE
-4)==0);
1860 static int stripe_to_pdidx(sector_t stripe
, raid5_conf_t
*conf
, int disks
)
1862 int sectors_per_chunk
= conf
->chunk_size
>> 9;
1864 int chunk_offset
= sector_div(stripe
, sectors_per_chunk
);
1866 raid5_compute_sector(stripe
* (disks
- conf
->max_degraded
)
1867 *sectors_per_chunk
+ chunk_offset
,
1868 disks
, disks
- conf
->max_degraded
,
1869 &dd_idx
, &pd_idx
, conf
);
1874 handle_requests_to_failed_array(raid5_conf_t
*conf
, struct stripe_head
*sh
,
1875 struct stripe_head_state
*s
, int disks
,
1876 struct bio
**return_bi
)
1879 for (i
= disks
; i
--; ) {
1883 if (test_bit(R5_ReadError
, &sh
->dev
[i
].flags
)) {
1886 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
1887 if (rdev
&& test_bit(In_sync
, &rdev
->flags
))
1888 /* multiple read failures in one stripe */
1889 md_error(conf
->mddev
, rdev
);
1892 spin_lock_irq(&conf
->device_lock
);
1893 /* fail all writes first */
1894 bi
= sh
->dev
[i
].towrite
;
1895 sh
->dev
[i
].towrite
= NULL
;
1901 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
1902 wake_up(&conf
->wait_for_overlap
);
1904 while (bi
&& bi
->bi_sector
<
1905 sh
->dev
[i
].sector
+ STRIPE_SECTORS
) {
1906 struct bio
*nextbi
= r5_next_bio(bi
, sh
->dev
[i
].sector
);
1907 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1908 if (--bi
->bi_phys_segments
== 0) {
1909 md_write_end(conf
->mddev
);
1910 bi
->bi_next
= *return_bi
;
1915 /* and fail all 'written' */
1916 bi
= sh
->dev
[i
].written
;
1917 sh
->dev
[i
].written
= NULL
;
1918 if (bi
) bitmap_end
= 1;
1919 while (bi
&& bi
->bi_sector
<
1920 sh
->dev
[i
].sector
+ STRIPE_SECTORS
) {
1921 struct bio
*bi2
= r5_next_bio(bi
, sh
->dev
[i
].sector
);
1922 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1923 if (--bi
->bi_phys_segments
== 0) {
1924 md_write_end(conf
->mddev
);
1925 bi
->bi_next
= *return_bi
;
1931 /* fail any reads if this device is non-operational and
1932 * the data has not reached the cache yet.
1934 if (!test_bit(R5_Wantfill
, &sh
->dev
[i
].flags
) &&
1935 (!test_bit(R5_Insync
, &sh
->dev
[i
].flags
) ||
1936 test_bit(R5_ReadError
, &sh
->dev
[i
].flags
))) {
1937 bi
= sh
->dev
[i
].toread
;
1938 sh
->dev
[i
].toread
= NULL
;
1939 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
1940 wake_up(&conf
->wait_for_overlap
);
1941 if (bi
) s
->to_read
--;
1942 while (bi
&& bi
->bi_sector
<
1943 sh
->dev
[i
].sector
+ STRIPE_SECTORS
) {
1944 struct bio
*nextbi
=
1945 r5_next_bio(bi
, sh
->dev
[i
].sector
);
1946 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1947 if (--bi
->bi_phys_segments
== 0) {
1948 bi
->bi_next
= *return_bi
;
1954 spin_unlock_irq(&conf
->device_lock
);
1956 bitmap_endwrite(conf
->mddev
->bitmap
, sh
->sector
,
1957 STRIPE_SECTORS
, 0, 0);
1962 /* __handle_issuing_new_read_requests5 - returns 0 if there are no more disks
1965 static int __handle_issuing_new_read_requests5(struct stripe_head
*sh
,
1966 struct stripe_head_state
*s
, int disk_idx
, int disks
)
1968 struct r5dev
*dev
= &sh
->dev
[disk_idx
];
1969 struct r5dev
*failed_dev
= &sh
->dev
[s
->failed_num
];
1971 /* don't schedule compute operations or reads on the parity block while
1972 * a check is in flight
1974 if ((disk_idx
== sh
->pd_idx
) &&
1975 test_bit(STRIPE_OP_CHECK
, &sh
->ops
.pending
))
1978 /* is the data in this block needed, and can we get it? */
1979 if (!test_bit(R5_LOCKED
, &dev
->flags
) &&
1980 !test_bit(R5_UPTODATE
, &dev
->flags
) && (dev
->toread
||
1981 (dev
->towrite
&& !test_bit(R5_OVERWRITE
, &dev
->flags
)) ||
1982 s
->syncing
|| s
->expanding
|| (s
->failed
&&
1983 (failed_dev
->toread
|| (failed_dev
->towrite
&&
1984 !test_bit(R5_OVERWRITE
, &failed_dev
->flags
)
1986 /* 1/ We would like to get this block, possibly by computing it,
1987 * but we might not be able to.
1989 * 2/ Since parity check operations potentially make the parity
1990 * block !uptodate it will need to be refreshed before any
1991 * compute operations on data disks are scheduled.
1993 * 3/ We hold off parity block re-reads until check operations
1996 if ((s
->uptodate
== disks
- 1) &&
1997 !test_bit(STRIPE_OP_CHECK
, &sh
->ops
.pending
)) {
1998 set_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
);
1999 set_bit(R5_Wantcompute
, &dev
->flags
);
2000 sh
->ops
.target
= disk_idx
;
2003 /* Careful: from this point on 'uptodate' is in the eye
2004 * of raid5_run_ops which services 'compute' operations
2005 * before writes. R5_Wantcompute flags a block that will
2006 * be R5_UPTODATE by the time it is needed for a
2007 * subsequent operation.
2010 return 0; /* uptodate + compute == disks */
2011 } else if ((s
->uptodate
< disks
- 1) &&
2012 test_bit(R5_Insync
, &dev
->flags
)) {
2013 /* Note: we hold off compute operations while checks are
2014 * in flight, but we still prefer 'compute' over 'read'
2015 * hence we only read if (uptodate < * disks-1)
2017 set_bit(R5_LOCKED
, &dev
->flags
);
2018 set_bit(R5_Wantread
, &dev
->flags
);
2019 if (!test_and_set_bit(STRIPE_OP_IO
, &sh
->ops
.pending
))
2022 pr_debug("Reading block %d (sync=%d)\n", disk_idx
,
2030 static void handle_issuing_new_read_requests5(struct stripe_head
*sh
,
2031 struct stripe_head_state
*s
, int disks
)
2035 /* Clear completed compute operations. Parity recovery
2036 * (STRIPE_OP_MOD_REPAIR_PD) implies a write-back which is handled
2037 * later on in this routine
2039 if (test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.complete
) &&
2040 !test_bit(STRIPE_OP_MOD_REPAIR_PD
, &sh
->ops
.pending
)) {
2041 clear_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.complete
);
2042 clear_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.ack
);
2043 clear_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
);
2046 /* look for blocks to read/compute, skip this if a compute
2047 * is already in flight, or if the stripe contents are in the
2048 * midst of changing due to a write
2050 if (!test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
) &&
2051 !test_bit(STRIPE_OP_PREXOR
, &sh
->ops
.pending
) &&
2052 !test_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
)) {
2053 for (i
= disks
; i
--; )
2054 if (__handle_issuing_new_read_requests5(
2055 sh
, s
, i
, disks
) == 0)
2058 set_bit(STRIPE_HANDLE
, &sh
->state
);
2061 static void handle_issuing_new_read_requests6(struct stripe_head
*sh
,
2062 struct stripe_head_state
*s
, struct r6_state
*r6s
,
2066 for (i
= disks
; i
--; ) {
2067 struct r5dev
*dev
= &sh
->dev
[i
];
2068 if (!test_bit(R5_LOCKED
, &dev
->flags
) &&
2069 !test_bit(R5_UPTODATE
, &dev
->flags
) &&
2070 (dev
->toread
|| (dev
->towrite
&&
2071 !test_bit(R5_OVERWRITE
, &dev
->flags
)) ||
2072 s
->syncing
|| s
->expanding
||
2074 (sh
->dev
[r6s
->failed_num
[0]].toread
||
2077 (sh
->dev
[r6s
->failed_num
[1]].toread
||
2079 /* we would like to get this block, possibly
2080 * by computing it, but we might not be able to
2082 if (s
->uptodate
== disks
-1) {
2083 pr_debug("Computing stripe %llu block %d\n",
2084 (unsigned long long)sh
->sector
, i
);
2085 compute_block_1(sh
, i
, 0);
2087 } else if ( s
->uptodate
== disks
-2 && s
->failed
>= 2 ) {
2088 /* Computing 2-failure is *very* expensive; only
2089 * do it if failed >= 2
2092 for (other
= disks
; other
--; ) {
2095 if (!test_bit(R5_UPTODATE
,
2096 &sh
->dev
[other
].flags
))
2100 pr_debug("Computing stripe %llu blocks %d,%d\n",
2101 (unsigned long long)sh
->sector
,
2103 compute_block_2(sh
, i
, other
);
2105 } else if (test_bit(R5_Insync
, &dev
->flags
)) {
2106 set_bit(R5_LOCKED
, &dev
->flags
);
2107 set_bit(R5_Wantread
, &dev
->flags
);
2109 pr_debug("Reading block %d (sync=%d)\n",
2114 set_bit(STRIPE_HANDLE
, &sh
->state
);
2118 /* handle_completed_write_requests
2119 * any written block on an uptodate or failed drive can be returned.
2120 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
2121 * never LOCKED, so we don't need to test 'failed' directly.
2123 static void handle_completed_write_requests(raid5_conf_t
*conf
,
2124 struct stripe_head
*sh
, int disks
, struct bio
**return_bi
)
2129 for (i
= disks
; i
--; )
2130 if (sh
->dev
[i
].written
) {
2132 if (!test_bit(R5_LOCKED
, &dev
->flags
) &&
2133 test_bit(R5_UPTODATE
, &dev
->flags
)) {
2134 /* We can return any write requests */
2135 struct bio
*wbi
, *wbi2
;
2137 pr_debug("Return write for disc %d\n", i
);
2138 spin_lock_irq(&conf
->device_lock
);
2140 dev
->written
= NULL
;
2141 while (wbi
&& wbi
->bi_sector
<
2142 dev
->sector
+ STRIPE_SECTORS
) {
2143 wbi2
= r5_next_bio(wbi
, dev
->sector
);
2144 if (--wbi
->bi_phys_segments
== 0) {
2145 md_write_end(conf
->mddev
);
2146 wbi
->bi_next
= *return_bi
;
2151 if (dev
->towrite
== NULL
)
2153 spin_unlock_irq(&conf
->device_lock
);
2155 bitmap_endwrite(conf
->mddev
->bitmap
,
2158 !test_bit(STRIPE_DEGRADED
, &sh
->state
),
2164 static void handle_issuing_new_write_requests5(raid5_conf_t
*conf
,
2165 struct stripe_head
*sh
, struct stripe_head_state
*s
, int disks
)
2167 int rmw
= 0, rcw
= 0, i
;
2168 for (i
= disks
; i
--; ) {
2169 /* would I have to read this buffer for read_modify_write */
2170 struct r5dev
*dev
= &sh
->dev
[i
];
2171 if ((dev
->towrite
|| i
== sh
->pd_idx
) &&
2172 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2173 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2174 test_bit(R5_Wantcompute
, &dev
->flags
))) {
2175 if (test_bit(R5_Insync
, &dev
->flags
))
2178 rmw
+= 2*disks
; /* cannot read it */
2180 /* Would I have to read this buffer for reconstruct_write */
2181 if (!test_bit(R5_OVERWRITE
, &dev
->flags
) && i
!= sh
->pd_idx
&&
2182 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2183 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2184 test_bit(R5_Wantcompute
, &dev
->flags
))) {
2185 if (test_bit(R5_Insync
, &dev
->flags
)) rcw
++;
2190 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
2191 (unsigned long long)sh
->sector
, rmw
, rcw
);
2192 set_bit(STRIPE_HANDLE
, &sh
->state
);
2193 if (rmw
< rcw
&& rmw
> 0)
2194 /* prefer read-modify-write, but need to get some data */
2195 for (i
= disks
; i
--; ) {
2196 struct r5dev
*dev
= &sh
->dev
[i
];
2197 if ((dev
->towrite
|| i
== sh
->pd_idx
) &&
2198 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2199 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2200 test_bit(R5_Wantcompute
, &dev
->flags
)) &&
2201 test_bit(R5_Insync
, &dev
->flags
)) {
2203 test_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2204 pr_debug("Read_old block "
2205 "%d for r-m-w\n", i
);
2206 set_bit(R5_LOCKED
, &dev
->flags
);
2207 set_bit(R5_Wantread
, &dev
->flags
);
2208 if (!test_and_set_bit(
2209 STRIPE_OP_IO
, &sh
->ops
.pending
))
2213 set_bit(STRIPE_DELAYED
, &sh
->state
);
2214 set_bit(STRIPE_HANDLE
, &sh
->state
);
2218 if (rcw
<= rmw
&& rcw
> 0)
2219 /* want reconstruct write, but need to get some data */
2220 for (i
= disks
; i
--; ) {
2221 struct r5dev
*dev
= &sh
->dev
[i
];
2222 if (!test_bit(R5_OVERWRITE
, &dev
->flags
) &&
2224 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2225 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2226 test_bit(R5_Wantcompute
, &dev
->flags
)) &&
2227 test_bit(R5_Insync
, &dev
->flags
)) {
2229 test_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2230 pr_debug("Read_old block "
2231 "%d for Reconstruct\n", i
);
2232 set_bit(R5_LOCKED
, &dev
->flags
);
2233 set_bit(R5_Wantread
, &dev
->flags
);
2234 if (!test_and_set_bit(
2235 STRIPE_OP_IO
, &sh
->ops
.pending
))
2239 set_bit(STRIPE_DELAYED
, &sh
->state
);
2240 set_bit(STRIPE_HANDLE
, &sh
->state
);
2244 /* now if nothing is locked, and if we have enough data,
2245 * we can start a write request
2247 /* since handle_stripe can be called at any time we need to handle the
2248 * case where a compute block operation has been submitted and then a
2249 * subsequent call wants to start a write request. raid5_run_ops only
2250 * handles the case where compute block and postxor are requested
2251 * simultaneously. If this is not the case then new writes need to be
2252 * held off until the compute completes.
2254 if ((s
->req_compute
||
2255 !test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
)) &&
2256 (s
->locked
== 0 && (rcw
== 0 || rmw
== 0) &&
2257 !test_bit(STRIPE_BIT_DELAY
, &sh
->state
)))
2258 s
->locked
+= handle_write_operations5(sh
, rcw
== 0, 0);
2261 static void handle_issuing_new_write_requests6(raid5_conf_t
*conf
,
2262 struct stripe_head
*sh
, struct stripe_head_state
*s
,
2263 struct r6_state
*r6s
, int disks
)
2265 int rcw
= 0, must_compute
= 0, pd_idx
= sh
->pd_idx
, i
;
2266 int qd_idx
= r6s
->qd_idx
;
2267 for (i
= disks
; i
--; ) {
2268 struct r5dev
*dev
= &sh
->dev
[i
];
2269 /* Would I have to read this buffer for reconstruct_write */
2270 if (!test_bit(R5_OVERWRITE
, &dev
->flags
)
2271 && i
!= pd_idx
&& i
!= qd_idx
2272 && (!test_bit(R5_LOCKED
, &dev
->flags
)
2274 !test_bit(R5_UPTODATE
, &dev
->flags
)) {
2275 if (test_bit(R5_Insync
, &dev
->flags
)) rcw
++;
2277 pr_debug("raid6: must_compute: "
2278 "disk %d flags=%#lx\n", i
, dev
->flags
);
2283 pr_debug("for sector %llu, rcw=%d, must_compute=%d\n",
2284 (unsigned long long)sh
->sector
, rcw
, must_compute
);
2285 set_bit(STRIPE_HANDLE
, &sh
->state
);
2288 /* want reconstruct write, but need to get some data */
2289 for (i
= disks
; i
--; ) {
2290 struct r5dev
*dev
= &sh
->dev
[i
];
2291 if (!test_bit(R5_OVERWRITE
, &dev
->flags
)
2292 && !(s
->failed
== 0 && (i
== pd_idx
|| i
== qd_idx
))
2293 && !test_bit(R5_LOCKED
, &dev
->flags
) &&
2294 !test_bit(R5_UPTODATE
, &dev
->flags
) &&
2295 test_bit(R5_Insync
, &dev
->flags
)) {
2297 test_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2298 pr_debug("Read_old stripe %llu "
2299 "block %d for Reconstruct\n",
2300 (unsigned long long)sh
->sector
, i
);
2301 set_bit(R5_LOCKED
, &dev
->flags
);
2302 set_bit(R5_Wantread
, &dev
->flags
);
2305 pr_debug("Request delayed stripe %llu "
2306 "block %d for Reconstruct\n",
2307 (unsigned long long)sh
->sector
, i
);
2308 set_bit(STRIPE_DELAYED
, &sh
->state
);
2309 set_bit(STRIPE_HANDLE
, &sh
->state
);
2313 /* now if nothing is locked, and if we have enough data, we can start a
2316 if (s
->locked
== 0 && rcw
== 0 &&
2317 !test_bit(STRIPE_BIT_DELAY
, &sh
->state
)) {
2318 if (must_compute
> 0) {
2319 /* We have failed blocks and need to compute them */
2320 switch (s
->failed
) {
2324 compute_block_1(sh
, r6s
->failed_num
[0], 0);
2327 compute_block_2(sh
, r6s
->failed_num
[0],
2328 r6s
->failed_num
[1]);
2330 default: /* This request should have been failed? */
2335 pr_debug("Computing parity for stripe %llu\n",
2336 (unsigned long long)sh
->sector
);
2337 compute_parity6(sh
, RECONSTRUCT_WRITE
);
2338 /* now every locked buffer is ready to be written */
2339 for (i
= disks
; i
--; )
2340 if (test_bit(R5_LOCKED
, &sh
->dev
[i
].flags
)) {
2341 pr_debug("Writing stripe %llu block %d\n",
2342 (unsigned long long)sh
->sector
, i
);
2344 set_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
);
2346 /* after a RECONSTRUCT_WRITE, the stripe MUST be in-sync */
2347 set_bit(STRIPE_INSYNC
, &sh
->state
);
2349 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2350 atomic_dec(&conf
->preread_active_stripes
);
2351 if (atomic_read(&conf
->preread_active_stripes
) <
2353 md_wakeup_thread(conf
->mddev
->thread
);
2358 static void handle_parity_checks5(raid5_conf_t
*conf
, struct stripe_head
*sh
,
2359 struct stripe_head_state
*s
, int disks
)
2361 set_bit(STRIPE_HANDLE
, &sh
->state
);
2362 /* Take one of the following actions:
2363 * 1/ start a check parity operation if (uptodate == disks)
2364 * 2/ finish a check parity operation and act on the result
2365 * 3/ skip to the writeback section if we previously
2366 * initiated a recovery operation
2368 if (s
->failed
== 0 &&
2369 !test_bit(STRIPE_OP_MOD_REPAIR_PD
, &sh
->ops
.pending
)) {
2370 if (!test_and_set_bit(STRIPE_OP_CHECK
, &sh
->ops
.pending
)) {
2371 BUG_ON(s
->uptodate
!= disks
);
2372 clear_bit(R5_UPTODATE
, &sh
->dev
[sh
->pd_idx
].flags
);
2376 test_and_clear_bit(STRIPE_OP_CHECK
, &sh
->ops
.complete
)) {
2377 clear_bit(STRIPE_OP_CHECK
, &sh
->ops
.ack
);
2378 clear_bit(STRIPE_OP_CHECK
, &sh
->ops
.pending
);
2380 if (sh
->ops
.zero_sum_result
== 0)
2381 /* parity is correct (on disc,
2382 * not in buffer any more)
2384 set_bit(STRIPE_INSYNC
, &sh
->state
);
2386 conf
->mddev
->resync_mismatches
+=
2389 MD_RECOVERY_CHECK
, &conf
->mddev
->recovery
))
2390 /* don't try to repair!! */
2391 set_bit(STRIPE_INSYNC
, &sh
->state
);
2393 set_bit(STRIPE_OP_COMPUTE_BLK
,
2395 set_bit(STRIPE_OP_MOD_REPAIR_PD
,
2397 set_bit(R5_Wantcompute
,
2398 &sh
->dev
[sh
->pd_idx
].flags
);
2399 sh
->ops
.target
= sh
->pd_idx
;
2407 /* check if we can clear a parity disk reconstruct */
2408 if (test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.complete
) &&
2409 test_bit(STRIPE_OP_MOD_REPAIR_PD
, &sh
->ops
.pending
)) {
2411 clear_bit(STRIPE_OP_MOD_REPAIR_PD
, &sh
->ops
.pending
);
2412 clear_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.complete
);
2413 clear_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.ack
);
2414 clear_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
);
2417 /* Wait for check parity and compute block operations to complete
2420 if (!test_bit(STRIPE_INSYNC
, &sh
->state
) &&
2421 !test_bit(STRIPE_OP_CHECK
, &sh
->ops
.pending
) &&
2422 !test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
)) {
2424 /* either failed parity check, or recovery is happening */
2426 s
->failed_num
= sh
->pd_idx
;
2427 dev
= &sh
->dev
[s
->failed_num
];
2428 BUG_ON(!test_bit(R5_UPTODATE
, &dev
->flags
));
2429 BUG_ON(s
->uptodate
!= disks
);
2431 set_bit(R5_LOCKED
, &dev
->flags
);
2432 set_bit(R5_Wantwrite
, &dev
->flags
);
2433 if (!test_and_set_bit(STRIPE_OP_IO
, &sh
->ops
.pending
))
2436 clear_bit(STRIPE_DEGRADED
, &sh
->state
);
2438 set_bit(STRIPE_INSYNC
, &sh
->state
);
2443 static void handle_parity_checks6(raid5_conf_t
*conf
, struct stripe_head
*sh
,
2444 struct stripe_head_state
*s
,
2445 struct r6_state
*r6s
, struct page
*tmp_page
,
2448 int update_p
= 0, update_q
= 0;
2450 int pd_idx
= sh
->pd_idx
;
2451 int qd_idx
= r6s
->qd_idx
;
2453 set_bit(STRIPE_HANDLE
, &sh
->state
);
2455 BUG_ON(s
->failed
> 2);
2456 BUG_ON(s
->uptodate
< disks
);
2457 /* Want to check and possibly repair P and Q.
2458 * However there could be one 'failed' device, in which
2459 * case we can only check one of them, possibly using the
2460 * other to generate missing data
2463 /* If !tmp_page, we cannot do the calculations,
2464 * but as we have set STRIPE_HANDLE, we will soon be called
2465 * by stripe_handle with a tmp_page - just wait until then.
2468 if (s
->failed
== r6s
->q_failed
) {
2469 /* The only possible failed device holds 'Q', so it
2470 * makes sense to check P (If anything else were failed,
2471 * we would have used P to recreate it).
2473 compute_block_1(sh
, pd_idx
, 1);
2474 if (!page_is_zero(sh
->dev
[pd_idx
].page
)) {
2475 compute_block_1(sh
, pd_idx
, 0);
2479 if (!r6s
->q_failed
&& s
->failed
< 2) {
2480 /* q is not failed, and we didn't use it to generate
2481 * anything, so it makes sense to check it
2483 memcpy(page_address(tmp_page
),
2484 page_address(sh
->dev
[qd_idx
].page
),
2486 compute_parity6(sh
, UPDATE_PARITY
);
2487 if (memcmp(page_address(tmp_page
),
2488 page_address(sh
->dev
[qd_idx
].page
),
2489 STRIPE_SIZE
) != 0) {
2490 clear_bit(STRIPE_INSYNC
, &sh
->state
);
2494 if (update_p
|| update_q
) {
2495 conf
->mddev
->resync_mismatches
+= STRIPE_SECTORS
;
2496 if (test_bit(MD_RECOVERY_CHECK
, &conf
->mddev
->recovery
))
2497 /* don't try to repair!! */
2498 update_p
= update_q
= 0;
2501 /* now write out any block on a failed drive,
2502 * or P or Q if they need it
2505 if (s
->failed
== 2) {
2506 dev
= &sh
->dev
[r6s
->failed_num
[1]];
2508 set_bit(R5_LOCKED
, &dev
->flags
);
2509 set_bit(R5_Wantwrite
, &dev
->flags
);
2511 if (s
->failed
>= 1) {
2512 dev
= &sh
->dev
[r6s
->failed_num
[0]];
2514 set_bit(R5_LOCKED
, &dev
->flags
);
2515 set_bit(R5_Wantwrite
, &dev
->flags
);
2519 dev
= &sh
->dev
[pd_idx
];
2521 set_bit(R5_LOCKED
, &dev
->flags
);
2522 set_bit(R5_Wantwrite
, &dev
->flags
);
2525 dev
= &sh
->dev
[qd_idx
];
2527 set_bit(R5_LOCKED
, &dev
->flags
);
2528 set_bit(R5_Wantwrite
, &dev
->flags
);
2530 clear_bit(STRIPE_DEGRADED
, &sh
->state
);
2532 set_bit(STRIPE_INSYNC
, &sh
->state
);
2536 static void handle_stripe_expansion(raid5_conf_t
*conf
, struct stripe_head
*sh
,
2537 struct r6_state
*r6s
)
2541 /* We have read all the blocks in this stripe and now we need to
2542 * copy some of them into a target stripe for expand.
2544 struct dma_async_tx_descriptor
*tx
= NULL
;
2545 clear_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
2546 for (i
= 0; i
< sh
->disks
; i
++)
2547 if (i
!= sh
->pd_idx
&& (!r6s
|| i
!= r6s
->qd_idx
)) {
2548 int dd_idx
, pd_idx
, j
;
2549 struct stripe_head
*sh2
;
2551 sector_t bn
= compute_blocknr(sh
, i
);
2552 sector_t s
= raid5_compute_sector(bn
, conf
->raid_disks
,
2554 conf
->max_degraded
, &dd_idx
,
2556 sh2
= get_active_stripe(conf
, s
, conf
->raid_disks
,
2559 /* so far only the early blocks of this stripe
2560 * have been requested. When later blocks
2561 * get requested, we will try again
2564 if (!test_bit(STRIPE_EXPANDING
, &sh2
->state
) ||
2565 test_bit(R5_Expanded
, &sh2
->dev
[dd_idx
].flags
)) {
2566 /* must have already done this block */
2567 release_stripe(sh2
);
2571 /* place all the copies on one channel */
2572 tx
= async_memcpy(sh2
->dev
[dd_idx
].page
,
2573 sh
->dev
[i
].page
, 0, 0, STRIPE_SIZE
,
2574 ASYNC_TX_DEP_ACK
, tx
, NULL
, NULL
);
2576 set_bit(R5_Expanded
, &sh2
->dev
[dd_idx
].flags
);
2577 set_bit(R5_UPTODATE
, &sh2
->dev
[dd_idx
].flags
);
2578 for (j
= 0; j
< conf
->raid_disks
; j
++)
2579 if (j
!= sh2
->pd_idx
&&
2580 (!r6s
|| j
!= raid6_next_disk(sh2
->pd_idx
,
2582 !test_bit(R5_Expanded
, &sh2
->dev
[j
].flags
))
2584 if (j
== conf
->raid_disks
) {
2585 set_bit(STRIPE_EXPAND_READY
, &sh2
->state
);
2586 set_bit(STRIPE_HANDLE
, &sh2
->state
);
2588 release_stripe(sh2
);
2591 /* done submitting copies, wait for them to complete */
2594 dma_wait_for_async_tx(tx
);
2599 * handle_stripe - do things to a stripe.
2601 * We lock the stripe and then examine the state of various bits
2602 * to see what needs to be done.
2604 * return some read request which now have data
2605 * return some write requests which are safely on disc
2606 * schedule a read on some buffers
2607 * schedule a write of some buffers
2608 * return confirmation of parity correctness
2610 * buffers are taken off read_list or write_list, and bh_cache buffers
2611 * get BH_Lock set before the stripe lock is released.
2615 static void handle_stripe5(struct stripe_head
*sh
)
2617 raid5_conf_t
*conf
= sh
->raid_conf
;
2618 int disks
= sh
->disks
, i
;
2619 struct bio
*return_bi
= NULL
;
2620 struct stripe_head_state s
;
2622 unsigned long pending
= 0;
2624 memset(&s
, 0, sizeof(s
));
2625 pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d "
2626 "ops=%lx:%lx:%lx\n", (unsigned long long)sh
->sector
, sh
->state
,
2627 atomic_read(&sh
->count
), sh
->pd_idx
,
2628 sh
->ops
.pending
, sh
->ops
.ack
, sh
->ops
.complete
);
2630 spin_lock(&sh
->lock
);
2631 clear_bit(STRIPE_HANDLE
, &sh
->state
);
2632 clear_bit(STRIPE_DELAYED
, &sh
->state
);
2634 s
.syncing
= test_bit(STRIPE_SYNCING
, &sh
->state
);
2635 s
.expanding
= test_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
2636 s
.expanded
= test_bit(STRIPE_EXPAND_READY
, &sh
->state
);
2637 /* Now to look around and see what can be done */
2639 /* clean-up completed biofill operations */
2640 if (test_bit(STRIPE_OP_BIOFILL
, &sh
->ops
.complete
)) {
2641 clear_bit(STRIPE_OP_BIOFILL
, &sh
->ops
.pending
);
2642 clear_bit(STRIPE_OP_BIOFILL
, &sh
->ops
.ack
);
2643 clear_bit(STRIPE_OP_BIOFILL
, &sh
->ops
.complete
);
2647 for (i
=disks
; i
--; ) {
2649 struct r5dev
*dev
= &sh
->dev
[i
];
2650 clear_bit(R5_Insync
, &dev
->flags
);
2652 pr_debug("check %d: state 0x%lx toread %p read %p write %p "
2653 "written %p\n", i
, dev
->flags
, dev
->toread
, dev
->read
,
2654 dev
->towrite
, dev
->written
);
2656 /* maybe we can request a biofill operation
2658 * new wantfill requests are only permitted while
2659 * STRIPE_OP_BIOFILL is clear
2661 if (test_bit(R5_UPTODATE
, &dev
->flags
) && dev
->toread
&&
2662 !test_bit(STRIPE_OP_BIOFILL
, &sh
->ops
.pending
))
2663 set_bit(R5_Wantfill
, &dev
->flags
);
2665 /* now count some things */
2666 if (test_bit(R5_LOCKED
, &dev
->flags
)) s
.locked
++;
2667 if (test_bit(R5_UPTODATE
, &dev
->flags
)) s
.uptodate
++;
2668 if (test_bit(R5_Wantcompute
, &dev
->flags
)) s
.compute
++;
2670 if (test_bit(R5_Wantfill
, &dev
->flags
))
2672 else if (dev
->toread
)
2676 if (!test_bit(R5_OVERWRITE
, &dev
->flags
))
2681 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
2682 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)) {
2683 /* The ReadError flag will just be confusing now */
2684 clear_bit(R5_ReadError
, &dev
->flags
);
2685 clear_bit(R5_ReWrite
, &dev
->flags
);
2687 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)
2688 || test_bit(R5_ReadError
, &dev
->flags
)) {
2692 set_bit(R5_Insync
, &dev
->flags
);
2696 if (s
.to_fill
&& !test_and_set_bit(STRIPE_OP_BIOFILL
, &sh
->ops
.pending
))
2699 pr_debug("locked=%d uptodate=%d to_read=%d"
2700 " to_write=%d failed=%d failed_num=%d\n",
2701 s
.locked
, s
.uptodate
, s
.to_read
, s
.to_write
,
2702 s
.failed
, s
.failed_num
);
2703 /* check if the array has lost two devices and, if so, some requests might
2706 if (s
.failed
> 1 && s
.to_read
+s
.to_write
+s
.written
)
2707 handle_requests_to_failed_array(conf
, sh
, &s
, disks
,
2709 if (s
.failed
> 1 && s
.syncing
) {
2710 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,0);
2711 clear_bit(STRIPE_SYNCING
, &sh
->state
);
2715 /* might be able to return some write requests if the parity block
2716 * is safe, or on a failed drive
2718 dev
= &sh
->dev
[sh
->pd_idx
];
2720 ((test_bit(R5_Insync
, &dev
->flags
) &&
2721 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2722 test_bit(R5_UPTODATE
, &dev
->flags
)) ||
2723 (s
.failed
== 1 && s
.failed_num
== sh
->pd_idx
)))
2724 handle_completed_write_requests(conf
, sh
, disks
, &return_bi
);
2726 /* Now we might consider reading some blocks, either to check/generate
2727 * parity, or to satisfy requests
2728 * or to load a block that is being partially written.
2730 if (s
.to_read
|| s
.non_overwrite
||
2731 (s
.syncing
&& (s
.uptodate
+ s
.compute
< disks
)) || s
.expanding
||
2732 test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
))
2733 handle_issuing_new_read_requests5(sh
, &s
, disks
);
2735 /* Now we check to see if any write operations have recently
2739 /* leave prexor set until postxor is done, allows us to distinguish
2740 * a rmw from a rcw during biodrain
2742 if (test_bit(STRIPE_OP_PREXOR
, &sh
->ops
.complete
) &&
2743 test_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.complete
)) {
2745 clear_bit(STRIPE_OP_PREXOR
, &sh
->ops
.complete
);
2746 clear_bit(STRIPE_OP_PREXOR
, &sh
->ops
.ack
);
2747 clear_bit(STRIPE_OP_PREXOR
, &sh
->ops
.pending
);
2749 for (i
= disks
; i
--; )
2750 clear_bit(R5_Wantprexor
, &sh
->dev
[i
].flags
);
2753 /* if only POSTXOR is set then this is an 'expand' postxor */
2754 if (test_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.complete
) &&
2755 test_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.complete
)) {
2757 clear_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.complete
);
2758 clear_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.ack
);
2759 clear_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.pending
);
2761 clear_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.complete
);
2762 clear_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.ack
);
2763 clear_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
);
2765 /* All the 'written' buffers and the parity block are ready to
2766 * be written back to disk
2768 BUG_ON(!test_bit(R5_UPTODATE
, &sh
->dev
[sh
->pd_idx
].flags
));
2769 for (i
= disks
; i
--; ) {
2771 if (test_bit(R5_LOCKED
, &dev
->flags
) &&
2772 (i
== sh
->pd_idx
|| dev
->written
)) {
2773 pr_debug("Writing block %d\n", i
);
2774 set_bit(R5_Wantwrite
, &dev
->flags
);
2775 if (!test_and_set_bit(
2776 STRIPE_OP_IO
, &sh
->ops
.pending
))
2778 if (!test_bit(R5_Insync
, &dev
->flags
) ||
2779 (i
== sh
->pd_idx
&& s
.failed
== 0))
2780 set_bit(STRIPE_INSYNC
, &sh
->state
);
2783 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2784 atomic_dec(&conf
->preread_active_stripes
);
2785 if (atomic_read(&conf
->preread_active_stripes
) <
2787 md_wakeup_thread(conf
->mddev
->thread
);
2791 /* Now to consider new write requests and what else, if anything
2792 * should be read. We do not handle new writes when:
2793 * 1/ A 'write' operation (copy+xor) is already in flight.
2794 * 2/ A 'check' operation is in flight, as it may clobber the parity
2797 if (s
.to_write
&& !test_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
) &&
2798 !test_bit(STRIPE_OP_CHECK
, &sh
->ops
.pending
))
2799 handle_issuing_new_write_requests5(conf
, sh
, &s
, disks
);
2801 /* maybe we need to check and possibly fix the parity for this stripe
2802 * Any reads will already have been scheduled, so we just see if enough
2803 * data is available. The parity check is held off while parity
2804 * dependent operations are in flight.
2806 if ((s
.syncing
&& s
.locked
== 0 &&
2807 !test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
) &&
2808 !test_bit(STRIPE_INSYNC
, &sh
->state
)) ||
2809 test_bit(STRIPE_OP_CHECK
, &sh
->ops
.pending
) ||
2810 test_bit(STRIPE_OP_MOD_REPAIR_PD
, &sh
->ops
.pending
))
2811 handle_parity_checks5(conf
, sh
, &s
, disks
);
2813 if (s
.syncing
&& s
.locked
== 0 && test_bit(STRIPE_INSYNC
, &sh
->state
)) {
2814 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,1);
2815 clear_bit(STRIPE_SYNCING
, &sh
->state
);
2818 /* If the failed drive is just a ReadError, then we might need to progress
2819 * the repair/check process
2821 if (s
.failed
== 1 && !conf
->mddev
->ro
&&
2822 test_bit(R5_ReadError
, &sh
->dev
[s
.failed_num
].flags
)
2823 && !test_bit(R5_LOCKED
, &sh
->dev
[s
.failed_num
].flags
)
2824 && test_bit(R5_UPTODATE
, &sh
->dev
[s
.failed_num
].flags
)
2826 dev
= &sh
->dev
[s
.failed_num
];
2827 if (!test_bit(R5_ReWrite
, &dev
->flags
)) {
2828 set_bit(R5_Wantwrite
, &dev
->flags
);
2829 if (!test_and_set_bit(STRIPE_OP_IO
, &sh
->ops
.pending
))
2831 set_bit(R5_ReWrite
, &dev
->flags
);
2832 set_bit(R5_LOCKED
, &dev
->flags
);
2835 /* let's read it back */
2836 set_bit(R5_Wantread
, &dev
->flags
);
2837 if (!test_and_set_bit(STRIPE_OP_IO
, &sh
->ops
.pending
))
2839 set_bit(R5_LOCKED
, &dev
->flags
);
2844 /* Finish postxor operations initiated by the expansion
2847 if (test_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.complete
) &&
2848 !test_bit(STRIPE_OP_BIODRAIN
, &sh
->ops
.pending
)) {
2850 clear_bit(STRIPE_EXPANDING
, &sh
->state
);
2852 clear_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
);
2853 clear_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.ack
);
2854 clear_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.complete
);
2856 for (i
= conf
->raid_disks
; i
--; ) {
2857 set_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
);
2858 if (!test_and_set_bit(STRIPE_OP_IO
, &sh
->ops
.pending
))
2863 if (s
.expanded
&& test_bit(STRIPE_EXPANDING
, &sh
->state
) &&
2864 !test_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
)) {
2865 /* Need to write out all blocks after computing parity */
2866 sh
->disks
= conf
->raid_disks
;
2867 sh
->pd_idx
= stripe_to_pdidx(sh
->sector
, conf
,
2869 s
.locked
+= handle_write_operations5(sh
, 1, 1);
2870 } else if (s
.expanded
&&
2871 !test_bit(STRIPE_OP_POSTXOR
, &sh
->ops
.pending
)) {
2872 clear_bit(STRIPE_EXPAND_READY
, &sh
->state
);
2873 atomic_dec(&conf
->reshape_stripes
);
2874 wake_up(&conf
->wait_for_overlap
);
2875 md_done_sync(conf
->mddev
, STRIPE_SECTORS
, 1);
2878 if (s
.expanding
&& s
.locked
== 0 &&
2879 !test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
))
2880 handle_stripe_expansion(conf
, sh
, NULL
);
2883 pending
= get_stripe_work(sh
);
2885 spin_unlock(&sh
->lock
);
2888 raid5_run_ops(sh
, pending
);
2890 return_io(return_bi
);
2894 static void handle_stripe6(struct stripe_head
*sh
, struct page
*tmp_page
)
2896 raid6_conf_t
*conf
= sh
->raid_conf
;
2897 int disks
= sh
->disks
;
2898 struct bio
*return_bi
= NULL
;
2899 int i
, pd_idx
= sh
->pd_idx
;
2900 struct stripe_head_state s
;
2901 struct r6_state r6s
;
2902 struct r5dev
*dev
, *pdev
, *qdev
;
2904 r6s
.qd_idx
= raid6_next_disk(pd_idx
, disks
);
2905 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
2906 "pd_idx=%d, qd_idx=%d\n",
2907 (unsigned long long)sh
->sector
, sh
->state
,
2908 atomic_read(&sh
->count
), pd_idx
, r6s
.qd_idx
);
2909 memset(&s
, 0, sizeof(s
));
2911 spin_lock(&sh
->lock
);
2912 clear_bit(STRIPE_HANDLE
, &sh
->state
);
2913 clear_bit(STRIPE_DELAYED
, &sh
->state
);
2915 s
.syncing
= test_bit(STRIPE_SYNCING
, &sh
->state
);
2916 s
.expanding
= test_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
2917 s
.expanded
= test_bit(STRIPE_EXPAND_READY
, &sh
->state
);
2918 /* Now to look around and see what can be done */
2921 for (i
=disks
; i
--; ) {
2924 clear_bit(R5_Insync
, &dev
->flags
);
2926 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
2927 i
, dev
->flags
, dev
->toread
, dev
->towrite
, dev
->written
);
2928 /* maybe we can reply to a read */
2929 if (test_bit(R5_UPTODATE
, &dev
->flags
) && dev
->toread
) {
2930 struct bio
*rbi
, *rbi2
;
2931 pr_debug("Return read for disc %d\n", i
);
2932 spin_lock_irq(&conf
->device_lock
);
2935 if (test_and_clear_bit(R5_Overlap
, &dev
->flags
))
2936 wake_up(&conf
->wait_for_overlap
);
2937 spin_unlock_irq(&conf
->device_lock
);
2938 while (rbi
&& rbi
->bi_sector
< dev
->sector
+ STRIPE_SECTORS
) {
2939 copy_data(0, rbi
, dev
->page
, dev
->sector
);
2940 rbi2
= r5_next_bio(rbi
, dev
->sector
);
2941 spin_lock_irq(&conf
->device_lock
);
2942 if (--rbi
->bi_phys_segments
== 0) {
2943 rbi
->bi_next
= return_bi
;
2946 spin_unlock_irq(&conf
->device_lock
);
2951 /* now count some things */
2952 if (test_bit(R5_LOCKED
, &dev
->flags
)) s
.locked
++;
2953 if (test_bit(R5_UPTODATE
, &dev
->flags
)) s
.uptodate
++;
2960 if (!test_bit(R5_OVERWRITE
, &dev
->flags
))
2965 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
2966 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)) {
2967 /* The ReadError flag will just be confusing now */
2968 clear_bit(R5_ReadError
, &dev
->flags
);
2969 clear_bit(R5_ReWrite
, &dev
->flags
);
2971 if (!rdev
|| !test_bit(In_sync
, &rdev
->flags
)
2972 || test_bit(R5_ReadError
, &dev
->flags
)) {
2974 r6s
.failed_num
[s
.failed
] = i
;
2977 set_bit(R5_Insync
, &dev
->flags
);
2980 pr_debug("locked=%d uptodate=%d to_read=%d"
2981 " to_write=%d failed=%d failed_num=%d,%d\n",
2982 s
.locked
, s
.uptodate
, s
.to_read
, s
.to_write
, s
.failed
,
2983 r6s
.failed_num
[0], r6s
.failed_num
[1]);
2984 /* check if the array has lost >2 devices and, if so, some requests
2985 * might need to be failed
2987 if (s
.failed
> 2 && s
.to_read
+s
.to_write
+s
.written
)
2988 handle_requests_to_failed_array(conf
, sh
, &s
, disks
,
2990 if (s
.failed
> 2 && s
.syncing
) {
2991 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,0);
2992 clear_bit(STRIPE_SYNCING
, &sh
->state
);
2997 * might be able to return some write requests if the parity blocks
2998 * are safe, or on a failed drive
3000 pdev
= &sh
->dev
[pd_idx
];
3001 r6s
.p_failed
= (s
.failed
>= 1 && r6s
.failed_num
[0] == pd_idx
)
3002 || (s
.failed
>= 2 && r6s
.failed_num
[1] == pd_idx
);
3003 qdev
= &sh
->dev
[r6s
.qd_idx
];
3004 r6s
.q_failed
= (s
.failed
>= 1 && r6s
.failed_num
[0] == r6s
.qd_idx
)
3005 || (s
.failed
>= 2 && r6s
.failed_num
[1] == r6s
.qd_idx
);
3008 ( r6s
.p_failed
|| ((test_bit(R5_Insync
, &pdev
->flags
)
3009 && !test_bit(R5_LOCKED
, &pdev
->flags
)
3010 && test_bit(R5_UPTODATE
, &pdev
->flags
)))) &&
3011 ( r6s
.q_failed
|| ((test_bit(R5_Insync
, &qdev
->flags
)
3012 && !test_bit(R5_LOCKED
, &qdev
->flags
)
3013 && test_bit(R5_UPTODATE
, &qdev
->flags
)))))
3014 handle_completed_write_requests(conf
, sh
, disks
, &return_bi
);
3016 /* Now we might consider reading some blocks, either to check/generate
3017 * parity, or to satisfy requests
3018 * or to load a block that is being partially written.
3020 if (s
.to_read
|| s
.non_overwrite
|| (s
.to_write
&& s
.failed
) ||
3021 (s
.syncing
&& (s
.uptodate
< disks
)) || s
.expanding
)
3022 handle_issuing_new_read_requests6(sh
, &s
, &r6s
, disks
);
3024 /* now to consider writing and what else, if anything should be read */
3026 handle_issuing_new_write_requests6(conf
, sh
, &s
, &r6s
, disks
);
3028 /* maybe we need to check and possibly fix the parity for this stripe
3029 * Any reads will already have been scheduled, so we just see if enough
3032 if (s
.syncing
&& s
.locked
== 0 && !test_bit(STRIPE_INSYNC
, &sh
->state
))
3033 handle_parity_checks6(conf
, sh
, &s
, &r6s
, tmp_page
, disks
);
3035 if (s
.syncing
&& s
.locked
== 0 && test_bit(STRIPE_INSYNC
, &sh
->state
)) {
3036 md_done_sync(conf
->mddev
, STRIPE_SECTORS
,1);
3037 clear_bit(STRIPE_SYNCING
, &sh
->state
);
3040 /* If the failed drives are just a ReadError, then we might need
3041 * to progress the repair/check process
3043 if (s
.failed
<= 2 && !conf
->mddev
->ro
)
3044 for (i
= 0; i
< s
.failed
; i
++) {
3045 dev
= &sh
->dev
[r6s
.failed_num
[i
]];
3046 if (test_bit(R5_ReadError
, &dev
->flags
)
3047 && !test_bit(R5_LOCKED
, &dev
->flags
)
3048 && test_bit(R5_UPTODATE
, &dev
->flags
)
3050 if (!test_bit(R5_ReWrite
, &dev
->flags
)) {
3051 set_bit(R5_Wantwrite
, &dev
->flags
);
3052 set_bit(R5_ReWrite
, &dev
->flags
);
3053 set_bit(R5_LOCKED
, &dev
->flags
);
3055 /* let's read it back */
3056 set_bit(R5_Wantread
, &dev
->flags
);
3057 set_bit(R5_LOCKED
, &dev
->flags
);
3062 if (s
.expanded
&& test_bit(STRIPE_EXPANDING
, &sh
->state
)) {
3063 /* Need to write out all blocks after computing P&Q */
3064 sh
->disks
= conf
->raid_disks
;
3065 sh
->pd_idx
= stripe_to_pdidx(sh
->sector
, conf
,
3067 compute_parity6(sh
, RECONSTRUCT_WRITE
);
3068 for (i
= conf
->raid_disks
; i
-- ; ) {
3069 set_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
3071 set_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
);
3073 clear_bit(STRIPE_EXPANDING
, &sh
->state
);
3074 } else if (s
.expanded
) {
3075 clear_bit(STRIPE_EXPAND_READY
, &sh
->state
);
3076 atomic_dec(&conf
->reshape_stripes
);
3077 wake_up(&conf
->wait_for_overlap
);
3078 md_done_sync(conf
->mddev
, STRIPE_SECTORS
, 1);
3081 if (s
.expanding
&& s
.locked
== 0 &&
3082 !test_bit(STRIPE_OP_COMPUTE_BLK
, &sh
->ops
.pending
))
3083 handle_stripe_expansion(conf
, sh
, &r6s
);
3085 spin_unlock(&sh
->lock
);
3087 return_io(return_bi
);
3089 for (i
=disks
; i
-- ;) {
3093 if (test_and_clear_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
))
3095 else if (test_and_clear_bit(R5_Wantread
, &sh
->dev
[i
].flags
))
3100 bi
= &sh
->dev
[i
].req
;
3104 bi
->bi_end_io
= raid5_end_write_request
;
3106 bi
->bi_end_io
= raid5_end_read_request
;
3109 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
3110 if (rdev
&& test_bit(Faulty
, &rdev
->flags
))
3113 atomic_inc(&rdev
->nr_pending
);
3117 if (s
.syncing
|| s
.expanding
|| s
.expanded
)
3118 md_sync_acct(rdev
->bdev
, STRIPE_SECTORS
);
3120 bi
->bi_bdev
= rdev
->bdev
;
3121 pr_debug("for %llu schedule op %ld on disc %d\n",
3122 (unsigned long long)sh
->sector
, bi
->bi_rw
, i
);
3123 atomic_inc(&sh
->count
);
3124 bi
->bi_sector
= sh
->sector
+ rdev
->data_offset
;
3125 bi
->bi_flags
= 1 << BIO_UPTODATE
;
3127 bi
->bi_max_vecs
= 1;
3129 bi
->bi_io_vec
= &sh
->dev
[i
].vec
;
3130 bi
->bi_io_vec
[0].bv_len
= STRIPE_SIZE
;
3131 bi
->bi_io_vec
[0].bv_offset
= 0;
3132 bi
->bi_size
= STRIPE_SIZE
;
3135 test_bit(R5_ReWrite
, &sh
->dev
[i
].flags
))
3136 atomic_add(STRIPE_SECTORS
, &rdev
->corrected_errors
);
3137 generic_make_request(bi
);
3140 set_bit(STRIPE_DEGRADED
, &sh
->state
);
3141 pr_debug("skip op %ld on disc %d for sector %llu\n",
3142 bi
->bi_rw
, i
, (unsigned long long)sh
->sector
);
3143 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
3144 set_bit(STRIPE_HANDLE
, &sh
->state
);
3149 static void handle_stripe(struct stripe_head
*sh
, struct page
*tmp_page
)
3151 if (sh
->raid_conf
->level
== 6)
3152 handle_stripe6(sh
, tmp_page
);
3159 static void raid5_activate_delayed(raid5_conf_t
*conf
)
3161 if (atomic_read(&conf
->preread_active_stripes
) < IO_THRESHOLD
) {
3162 while (!list_empty(&conf
->delayed_list
)) {
3163 struct list_head
*l
= conf
->delayed_list
.next
;
3164 struct stripe_head
*sh
;
3165 sh
= list_entry(l
, struct stripe_head
, lru
);
3167 clear_bit(STRIPE_DELAYED
, &sh
->state
);
3168 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
))
3169 atomic_inc(&conf
->preread_active_stripes
);
3170 list_add_tail(&sh
->lru
, &conf
->handle_list
);
3175 static void activate_bit_delay(raid5_conf_t
*conf
)
3177 /* device_lock is held */
3178 struct list_head head
;
3179 list_add(&head
, &conf
->bitmap_list
);
3180 list_del_init(&conf
->bitmap_list
);
3181 while (!list_empty(&head
)) {
3182 struct stripe_head
*sh
= list_entry(head
.next
, struct stripe_head
, lru
);
3183 list_del_init(&sh
->lru
);
3184 atomic_inc(&sh
->count
);
3185 __release_stripe(conf
, sh
);
3189 static void unplug_slaves(mddev_t
*mddev
)
3191 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3195 for (i
=0; i
<mddev
->raid_disks
; i
++) {
3196 mdk_rdev_t
*rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
3197 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
) && atomic_read(&rdev
->nr_pending
)) {
3198 struct request_queue
*r_queue
= bdev_get_queue(rdev
->bdev
);
3200 atomic_inc(&rdev
->nr_pending
);
3203 if (r_queue
->unplug_fn
)
3204 r_queue
->unplug_fn(r_queue
);
3206 rdev_dec_pending(rdev
, mddev
);
3213 static void raid5_unplug_device(struct request_queue
*q
)
3215 mddev_t
*mddev
= q
->queuedata
;
3216 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3217 unsigned long flags
;
3219 spin_lock_irqsave(&conf
->device_lock
, flags
);
3221 if (blk_remove_plug(q
)) {
3223 raid5_activate_delayed(conf
);
3225 md_wakeup_thread(mddev
->thread
);
3227 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
3229 unplug_slaves(mddev
);
3232 static int raid5_issue_flush(struct request_queue
*q
, struct gendisk
*disk
,
3233 sector_t
*error_sector
)
3235 mddev_t
*mddev
= q
->queuedata
;
3236 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3240 for (i
=0; i
<mddev
->raid_disks
&& ret
== 0; i
++) {
3241 mdk_rdev_t
*rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
3242 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
)) {
3243 struct block_device
*bdev
= rdev
->bdev
;
3244 struct request_queue
*r_queue
= bdev_get_queue(bdev
);
3246 if (!r_queue
->issue_flush_fn
)
3249 atomic_inc(&rdev
->nr_pending
);
3251 ret
= r_queue
->issue_flush_fn(r_queue
, bdev
->bd_disk
,
3253 rdev_dec_pending(rdev
, mddev
);
3262 static int raid5_congested(void *data
, int bits
)
3264 mddev_t
*mddev
= data
;
3265 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3267 /* No difference between reads and writes. Just check
3268 * how busy the stripe_cache is
3270 if (conf
->inactive_blocked
)
3274 if (list_empty_careful(&conf
->inactive_list
))
3280 /* We want read requests to align with chunks where possible,
3281 * but write requests don't need to.
3283 static int raid5_mergeable_bvec(struct request_queue
*q
, struct bio
*bio
, struct bio_vec
*biovec
)
3285 mddev_t
*mddev
= q
->queuedata
;
3286 sector_t sector
= bio
->bi_sector
+ get_start_sect(bio
->bi_bdev
);
3288 unsigned int chunk_sectors
= mddev
->chunk_size
>> 9;
3289 unsigned int bio_sectors
= bio
->bi_size
>> 9;
3291 if (bio_data_dir(bio
) == WRITE
)
3292 return biovec
->bv_len
; /* always allow writes to be mergeable */
3294 max
= (chunk_sectors
- ((sector
& (chunk_sectors
- 1)) + bio_sectors
)) << 9;
3295 if (max
< 0) max
= 0;
3296 if (max
<= biovec
->bv_len
&& bio_sectors
== 0)
3297 return biovec
->bv_len
;
3303 static int in_chunk_boundary(mddev_t
*mddev
, struct bio
*bio
)
3305 sector_t sector
= bio
->bi_sector
+ get_start_sect(bio
->bi_bdev
);
3306 unsigned int chunk_sectors
= mddev
->chunk_size
>> 9;
3307 unsigned int bio_sectors
= bio
->bi_size
>> 9;
3309 return chunk_sectors
>=
3310 ((sector
& (chunk_sectors
- 1)) + bio_sectors
);
3314 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
3315 * later sampled by raid5d.
3317 static void add_bio_to_retry(struct bio
*bi
,raid5_conf_t
*conf
)
3319 unsigned long flags
;
3321 spin_lock_irqsave(&conf
->device_lock
, flags
);
3323 bi
->bi_next
= conf
->retry_read_aligned_list
;
3324 conf
->retry_read_aligned_list
= bi
;
3326 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
3327 md_wakeup_thread(conf
->mddev
->thread
);
3331 static struct bio
*remove_bio_from_retry(raid5_conf_t
*conf
)
3335 bi
= conf
->retry_read_aligned
;
3337 conf
->retry_read_aligned
= NULL
;
3340 bi
= conf
->retry_read_aligned_list
;
3342 conf
->retry_read_aligned_list
= bi
->bi_next
;
3344 bi
->bi_phys_segments
= 1; /* biased count of active stripes */
3345 bi
->bi_hw_segments
= 0; /* count of processed stripes */
3353 * The "raid5_align_endio" should check if the read succeeded and if it
3354 * did, call bio_endio on the original bio (having bio_put the new bio
3356 * If the read failed..
3358 static int raid5_align_endio(struct bio
*bi
, unsigned int bytes
, int error
)
3360 struct bio
* raid_bi
= bi
->bi_private
;
3363 int uptodate
= test_bit(BIO_UPTODATE
, &bi
->bi_flags
);
3370 mddev
= raid_bi
->bi_bdev
->bd_disk
->queue
->queuedata
;
3371 conf
= mddev_to_conf(mddev
);
3372 rdev
= (void*)raid_bi
->bi_next
;
3373 raid_bi
->bi_next
= NULL
;
3375 rdev_dec_pending(rdev
, conf
->mddev
);
3377 if (!error
&& uptodate
) {
3378 bio_endio(raid_bi
, bytes
, 0);
3379 if (atomic_dec_and_test(&conf
->active_aligned_reads
))
3380 wake_up(&conf
->wait_for_stripe
);
3385 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
3387 add_bio_to_retry(raid_bi
, conf
);
3391 static int bio_fits_rdev(struct bio
*bi
)
3393 struct request_queue
*q
= bdev_get_queue(bi
->bi_bdev
);
3395 if ((bi
->bi_size
>>9) > q
->max_sectors
)
3397 blk_recount_segments(q
, bi
);
3398 if (bi
->bi_phys_segments
> q
->max_phys_segments
||
3399 bi
->bi_hw_segments
> q
->max_hw_segments
)
3402 if (q
->merge_bvec_fn
)
3403 /* it's too hard to apply the merge_bvec_fn at this stage,
3412 static int chunk_aligned_read(struct request_queue
*q
, struct bio
* raid_bio
)
3414 mddev_t
*mddev
= q
->queuedata
;
3415 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3416 const unsigned int raid_disks
= conf
->raid_disks
;
3417 const unsigned int data_disks
= raid_disks
- conf
->max_degraded
;
3418 unsigned int dd_idx
, pd_idx
;
3419 struct bio
* align_bi
;
3422 if (!in_chunk_boundary(mddev
, raid_bio
)) {
3423 pr_debug("chunk_aligned_read : non aligned\n");
3427 * use bio_clone to make a copy of the bio
3429 align_bi
= bio_clone(raid_bio
, GFP_NOIO
);
3433 * set bi_end_io to a new function, and set bi_private to the
3436 align_bi
->bi_end_io
= raid5_align_endio
;
3437 align_bi
->bi_private
= raid_bio
;
3441 align_bi
->bi_sector
= raid5_compute_sector(raid_bio
->bi_sector
,
3449 rdev
= rcu_dereference(conf
->disks
[dd_idx
].rdev
);
3450 if (rdev
&& test_bit(In_sync
, &rdev
->flags
)) {
3451 atomic_inc(&rdev
->nr_pending
);
3453 raid_bio
->bi_next
= (void*)rdev
;
3454 align_bi
->bi_bdev
= rdev
->bdev
;
3455 align_bi
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
3456 align_bi
->bi_sector
+= rdev
->data_offset
;
3458 if (!bio_fits_rdev(align_bi
)) {
3459 /* too big in some way */
3461 rdev_dec_pending(rdev
, mddev
);
3465 spin_lock_irq(&conf
->device_lock
);
3466 wait_event_lock_irq(conf
->wait_for_stripe
,
3468 conf
->device_lock
, /* nothing */);
3469 atomic_inc(&conf
->active_aligned_reads
);
3470 spin_unlock_irq(&conf
->device_lock
);
3472 generic_make_request(align_bi
);
3482 static int make_request(struct request_queue
*q
, struct bio
* bi
)
3484 mddev_t
*mddev
= q
->queuedata
;
3485 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3486 unsigned int dd_idx
, pd_idx
;
3487 sector_t new_sector
;
3488 sector_t logical_sector
, last_sector
;
3489 struct stripe_head
*sh
;
3490 const int rw
= bio_data_dir(bi
);
3493 if (unlikely(bio_barrier(bi
))) {
3494 bio_endio(bi
, bi
->bi_size
, -EOPNOTSUPP
);
3498 md_write_start(mddev
, bi
);
3500 disk_stat_inc(mddev
->gendisk
, ios
[rw
]);
3501 disk_stat_add(mddev
->gendisk
, sectors
[rw
], bio_sectors(bi
));
3504 mddev
->reshape_position
== MaxSector
&&
3505 chunk_aligned_read(q
,bi
))
3508 logical_sector
= bi
->bi_sector
& ~((sector_t
)STRIPE_SECTORS
-1);
3509 last_sector
= bi
->bi_sector
+ (bi
->bi_size
>>9);
3511 bi
->bi_phys_segments
= 1; /* over-loaded to count active stripes */
3513 for (;logical_sector
< last_sector
; logical_sector
+= STRIPE_SECTORS
) {
3515 int disks
, data_disks
;
3518 prepare_to_wait(&conf
->wait_for_overlap
, &w
, TASK_UNINTERRUPTIBLE
);
3519 if (likely(conf
->expand_progress
== MaxSector
))
3520 disks
= conf
->raid_disks
;
3522 /* spinlock is needed as expand_progress may be
3523 * 64bit on a 32bit platform, and so it might be
3524 * possible to see a half-updated value
3525 * Ofcourse expand_progress could change after
3526 * the lock is dropped, so once we get a reference
3527 * to the stripe that we think it is, we will have
3530 spin_lock_irq(&conf
->device_lock
);
3531 disks
= conf
->raid_disks
;
3532 if (logical_sector
>= conf
->expand_progress
)
3533 disks
= conf
->previous_raid_disks
;
3535 if (logical_sector
>= conf
->expand_lo
) {
3536 spin_unlock_irq(&conf
->device_lock
);
3541 spin_unlock_irq(&conf
->device_lock
);
3543 data_disks
= disks
- conf
->max_degraded
;
3545 new_sector
= raid5_compute_sector(logical_sector
, disks
, data_disks
,
3546 &dd_idx
, &pd_idx
, conf
);
3547 pr_debug("raid5: make_request, sector %llu logical %llu\n",
3548 (unsigned long long)new_sector
,
3549 (unsigned long long)logical_sector
);
3551 sh
= get_active_stripe(conf
, new_sector
, disks
, pd_idx
, (bi
->bi_rw
&RWA_MASK
));
3553 if (unlikely(conf
->expand_progress
!= MaxSector
)) {
3554 /* expansion might have moved on while waiting for a
3555 * stripe, so we must do the range check again.
3556 * Expansion could still move past after this
3557 * test, but as we are holding a reference to
3558 * 'sh', we know that if that happens,
3559 * STRIPE_EXPANDING will get set and the expansion
3560 * won't proceed until we finish with the stripe.
3563 spin_lock_irq(&conf
->device_lock
);
3564 if (logical_sector
< conf
->expand_progress
&&
3565 disks
== conf
->previous_raid_disks
)
3566 /* mismatch, need to try again */
3568 spin_unlock_irq(&conf
->device_lock
);
3574 /* FIXME what if we get a false positive because these
3575 * are being updated.
3577 if (logical_sector
>= mddev
->suspend_lo
&&
3578 logical_sector
< mddev
->suspend_hi
) {
3584 if (test_bit(STRIPE_EXPANDING
, &sh
->state
) ||
3585 !add_stripe_bio(sh
, bi
, dd_idx
, (bi
->bi_rw
&RW_MASK
))) {
3586 /* Stripe is busy expanding or
3587 * add failed due to overlap. Flush everything
3590 raid5_unplug_device(mddev
->queue
);
3595 finish_wait(&conf
->wait_for_overlap
, &w
);
3596 handle_stripe(sh
, NULL
);
3599 /* cannot get stripe for read-ahead, just give-up */
3600 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
3601 finish_wait(&conf
->wait_for_overlap
, &w
);
3606 spin_lock_irq(&conf
->device_lock
);
3607 remaining
= --bi
->bi_phys_segments
;
3608 spin_unlock_irq(&conf
->device_lock
);
3609 if (remaining
== 0) {
3610 int bytes
= bi
->bi_size
;
3613 md_write_end(mddev
);
3615 bi
->bi_end_io(bi
, bytes
,
3616 test_bit(BIO_UPTODATE
, &bi
->bi_flags
)
3622 static sector_t
reshape_request(mddev_t
*mddev
, sector_t sector_nr
, int *skipped
)
3624 /* reshaping is quite different to recovery/resync so it is
3625 * handled quite separately ... here.
3627 * On each call to sync_request, we gather one chunk worth of
3628 * destination stripes and flag them as expanding.
3629 * Then we find all the source stripes and request reads.
3630 * As the reads complete, handle_stripe will copy the data
3631 * into the destination stripe and release that stripe.
3633 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
3634 struct stripe_head
*sh
;
3636 sector_t first_sector
, last_sector
;
3637 int raid_disks
= conf
->previous_raid_disks
;
3638 int data_disks
= raid_disks
- conf
->max_degraded
;
3639 int new_data_disks
= conf
->raid_disks
- conf
->max_degraded
;
3642 sector_t writepos
, safepos
, gap
;
3644 if (sector_nr
== 0 &&
3645 conf
->expand_progress
!= 0) {
3646 /* restarting in the middle, skip the initial sectors */
3647 sector_nr
= conf
->expand_progress
;
3648 sector_div(sector_nr
, new_data_disks
);
3653 /* we update the metadata when there is more than 3Meg
3654 * in the block range (that is rather arbitrary, should
3655 * probably be time based) or when the data about to be
3656 * copied would over-write the source of the data at
3657 * the front of the range.
3658 * i.e. one new_stripe forward from expand_progress new_maps
3659 * to after where expand_lo old_maps to
3661 writepos
= conf
->expand_progress
+
3662 conf
->chunk_size
/512*(new_data_disks
);
3663 sector_div(writepos
, new_data_disks
);
3664 safepos
= conf
->expand_lo
;
3665 sector_div(safepos
, data_disks
);
3666 gap
= conf
->expand_progress
- conf
->expand_lo
;
3668 if (writepos
>= safepos
||
3669 gap
> (new_data_disks
)*3000*2 /*3Meg*/) {
3670 /* Cannot proceed until we've updated the superblock... */
3671 wait_event(conf
->wait_for_overlap
,
3672 atomic_read(&conf
->reshape_stripes
)==0);
3673 mddev
->reshape_position
= conf
->expand_progress
;
3674 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3675 md_wakeup_thread(mddev
->thread
);
3676 wait_event(mddev
->sb_wait
, mddev
->flags
== 0 ||
3677 kthread_should_stop());
3678 spin_lock_irq(&conf
->device_lock
);
3679 conf
->expand_lo
= mddev
->reshape_position
;
3680 spin_unlock_irq(&conf
->device_lock
);
3681 wake_up(&conf
->wait_for_overlap
);
3684 for (i
=0; i
< conf
->chunk_size
/512; i
+= STRIPE_SECTORS
) {
3687 pd_idx
= stripe_to_pdidx(sector_nr
+i
, conf
, conf
->raid_disks
);
3688 sh
= get_active_stripe(conf
, sector_nr
+i
,
3689 conf
->raid_disks
, pd_idx
, 0);
3690 set_bit(STRIPE_EXPANDING
, &sh
->state
);
3691 atomic_inc(&conf
->reshape_stripes
);
3692 /* If any of this stripe is beyond the end of the old
3693 * array, then we need to zero those blocks
3695 for (j
=sh
->disks
; j
--;) {
3697 if (j
== sh
->pd_idx
)
3699 if (conf
->level
== 6 &&
3700 j
== raid6_next_disk(sh
->pd_idx
, sh
->disks
))
3702 s
= compute_blocknr(sh
, j
);
3703 if (s
< (mddev
->array_size
<<1)) {
3707 memset(page_address(sh
->dev
[j
].page
), 0, STRIPE_SIZE
);
3708 set_bit(R5_Expanded
, &sh
->dev
[j
].flags
);
3709 set_bit(R5_UPTODATE
, &sh
->dev
[j
].flags
);
3712 set_bit(STRIPE_EXPAND_READY
, &sh
->state
);
3713 set_bit(STRIPE_HANDLE
, &sh
->state
);
3717 spin_lock_irq(&conf
->device_lock
);
3718 conf
->expand_progress
= (sector_nr
+ i
) * new_data_disks
;
3719 spin_unlock_irq(&conf
->device_lock
);
3720 /* Ok, those stripe are ready. We can start scheduling
3721 * reads on the source stripes.
3722 * The source stripes are determined by mapping the first and last
3723 * block on the destination stripes.
3726 raid5_compute_sector(sector_nr
*(new_data_disks
),
3727 raid_disks
, data_disks
,
3728 &dd_idx
, &pd_idx
, conf
);
3730 raid5_compute_sector((sector_nr
+conf
->chunk_size
/512)
3731 *(new_data_disks
) -1,
3732 raid_disks
, data_disks
,
3733 &dd_idx
, &pd_idx
, conf
);
3734 if (last_sector
>= (mddev
->size
<<1))
3735 last_sector
= (mddev
->size
<<1)-1;
3736 while (first_sector
<= last_sector
) {
3737 pd_idx
= stripe_to_pdidx(first_sector
, conf
,
3738 conf
->previous_raid_disks
);
3739 sh
= get_active_stripe(conf
, first_sector
,
3740 conf
->previous_raid_disks
, pd_idx
, 0);
3741 set_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
3742 set_bit(STRIPE_HANDLE
, &sh
->state
);
3744 first_sector
+= STRIPE_SECTORS
;
3746 return conf
->chunk_size
>>9;
3749 /* FIXME go_faster isn't used */
3750 static inline sector_t
sync_request(mddev_t
*mddev
, sector_t sector_nr
, int *skipped
, int go_faster
)
3752 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
3753 struct stripe_head
*sh
;
3755 int raid_disks
= conf
->raid_disks
;
3756 sector_t max_sector
= mddev
->size
<< 1;
3758 int still_degraded
= 0;
3761 if (sector_nr
>= max_sector
) {
3762 /* just being told to finish up .. nothing much to do */
3763 unplug_slaves(mddev
);
3764 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)) {
3769 if (mddev
->curr_resync
< max_sector
) /* aborted */
3770 bitmap_end_sync(mddev
->bitmap
, mddev
->curr_resync
,
3772 else /* completed sync */
3774 bitmap_close_sync(mddev
->bitmap
);
3779 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
3780 return reshape_request(mddev
, sector_nr
, skipped
);
3782 /* if there is too many failed drives and we are trying
3783 * to resync, then assert that we are finished, because there is
3784 * nothing we can do.
3786 if (mddev
->degraded
>= conf
->max_degraded
&&
3787 test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
3788 sector_t rv
= (mddev
->size
<< 1) - sector_nr
;
3792 if (!bitmap_start_sync(mddev
->bitmap
, sector_nr
, &sync_blocks
, 1) &&
3793 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
) &&
3794 !conf
->fullsync
&& sync_blocks
>= STRIPE_SECTORS
) {
3795 /* we can skip this block, and probably more */
3796 sync_blocks
/= STRIPE_SECTORS
;
3798 return sync_blocks
* STRIPE_SECTORS
; /* keep things rounded to whole stripes */
3801 pd_idx
= stripe_to_pdidx(sector_nr
, conf
, raid_disks
);
3802 sh
= get_active_stripe(conf
, sector_nr
, raid_disks
, pd_idx
, 1);
3804 sh
= get_active_stripe(conf
, sector_nr
, raid_disks
, pd_idx
, 0);
3805 /* make sure we don't swamp the stripe cache if someone else
3806 * is trying to get access
3808 schedule_timeout_uninterruptible(1);
3810 /* Need to check if array will still be degraded after recovery/resync
3811 * We don't need to check the 'failed' flag as when that gets set,
3814 for (i
=0; i
<mddev
->raid_disks
; i
++)
3815 if (conf
->disks
[i
].rdev
== NULL
)
3818 bitmap_start_sync(mddev
->bitmap
, sector_nr
, &sync_blocks
, still_degraded
);
3820 spin_lock(&sh
->lock
);
3821 set_bit(STRIPE_SYNCING
, &sh
->state
);
3822 clear_bit(STRIPE_INSYNC
, &sh
->state
);
3823 spin_unlock(&sh
->lock
);
3825 handle_stripe(sh
, NULL
);
3828 return STRIPE_SECTORS
;
3831 static int retry_aligned_read(raid5_conf_t
*conf
, struct bio
*raid_bio
)
3833 /* We may not be able to submit a whole bio at once as there
3834 * may not be enough stripe_heads available.
3835 * We cannot pre-allocate enough stripe_heads as we may need
3836 * more than exist in the cache (if we allow ever large chunks).
3837 * So we do one stripe head at a time and record in
3838 * ->bi_hw_segments how many have been done.
3840 * We *know* that this entire raid_bio is in one chunk, so
3841 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
3843 struct stripe_head
*sh
;
3845 sector_t sector
, logical_sector
, last_sector
;
3850 logical_sector
= raid_bio
->bi_sector
& ~((sector_t
)STRIPE_SECTORS
-1);
3851 sector
= raid5_compute_sector( logical_sector
,
3853 conf
->raid_disks
- conf
->max_degraded
,
3857 last_sector
= raid_bio
->bi_sector
+ (raid_bio
->bi_size
>>9);
3859 for (; logical_sector
< last_sector
;
3860 logical_sector
+= STRIPE_SECTORS
,
3861 sector
+= STRIPE_SECTORS
,
3864 if (scnt
< raid_bio
->bi_hw_segments
)
3865 /* already done this stripe */
3868 sh
= get_active_stripe(conf
, sector
, conf
->raid_disks
, pd_idx
, 1);
3871 /* failed to get a stripe - must wait */
3872 raid_bio
->bi_hw_segments
= scnt
;
3873 conf
->retry_read_aligned
= raid_bio
;
3877 set_bit(R5_ReadError
, &sh
->dev
[dd_idx
].flags
);
3878 if (!add_stripe_bio(sh
, raid_bio
, dd_idx
, 0)) {
3880 raid_bio
->bi_hw_segments
= scnt
;
3881 conf
->retry_read_aligned
= raid_bio
;
3885 handle_stripe(sh
, NULL
);
3889 spin_lock_irq(&conf
->device_lock
);
3890 remaining
= --raid_bio
->bi_phys_segments
;
3891 spin_unlock_irq(&conf
->device_lock
);
3892 if (remaining
== 0) {
3893 int bytes
= raid_bio
->bi_size
;
3895 raid_bio
->bi_size
= 0;
3896 raid_bio
->bi_end_io(raid_bio
, bytes
,
3897 test_bit(BIO_UPTODATE
, &raid_bio
->bi_flags
)
3900 if (atomic_dec_and_test(&conf
->active_aligned_reads
))
3901 wake_up(&conf
->wait_for_stripe
);
3908 * This is our raid5 kernel thread.
3910 * We scan the hash table for stripes which can be handled now.
3911 * During the scan, completed stripes are saved for us by the interrupt
3912 * handler, so that they will not have to wait for our next wakeup.
3914 static void raid5d (mddev_t
*mddev
)
3916 struct stripe_head
*sh
;
3917 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3920 pr_debug("+++ raid5d active\n");
3922 md_check_recovery(mddev
);
3925 spin_lock_irq(&conf
->device_lock
);
3927 struct list_head
*first
;
3930 if (conf
->seq_flush
!= conf
->seq_write
) {
3931 int seq
= conf
->seq_flush
;
3932 spin_unlock_irq(&conf
->device_lock
);
3933 bitmap_unplug(mddev
->bitmap
);
3934 spin_lock_irq(&conf
->device_lock
);
3935 conf
->seq_write
= seq
;
3936 activate_bit_delay(conf
);
3939 if (list_empty(&conf
->handle_list
) &&
3940 atomic_read(&conf
->preread_active_stripes
) < IO_THRESHOLD
&&
3941 !blk_queue_plugged(mddev
->queue
) &&
3942 !list_empty(&conf
->delayed_list
))
3943 raid5_activate_delayed(conf
);
3945 while ((bio
= remove_bio_from_retry(conf
))) {
3947 spin_unlock_irq(&conf
->device_lock
);
3948 ok
= retry_aligned_read(conf
, bio
);
3949 spin_lock_irq(&conf
->device_lock
);
3955 if (list_empty(&conf
->handle_list
)) {
3956 async_tx_issue_pending_all();
3960 first
= conf
->handle_list
.next
;
3961 sh
= list_entry(first
, struct stripe_head
, lru
);
3963 list_del_init(first
);
3964 atomic_inc(&sh
->count
);
3965 BUG_ON(atomic_read(&sh
->count
)!= 1);
3966 spin_unlock_irq(&conf
->device_lock
);
3969 handle_stripe(sh
, conf
->spare_page
);
3972 spin_lock_irq(&conf
->device_lock
);
3974 pr_debug("%d stripes handled\n", handled
);
3976 spin_unlock_irq(&conf
->device_lock
);
3978 unplug_slaves(mddev
);
3980 pr_debug("--- raid5d inactive\n");
3984 raid5_show_stripe_cache_size(mddev_t
*mddev
, char *page
)
3986 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3988 return sprintf(page
, "%d\n", conf
->max_nr_stripes
);
3994 raid5_store_stripe_cache_size(mddev_t
*mddev
, const char *page
, size_t len
)
3996 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
3999 if (len
>= PAGE_SIZE
)
4004 new = simple_strtoul(page
, &end
, 10);
4005 if (!*page
|| (*end
&& *end
!= '\n') )
4007 if (new <= 16 || new > 32768)
4009 while (new < conf
->max_nr_stripes
) {
4010 if (drop_one_stripe(conf
))
4011 conf
->max_nr_stripes
--;
4015 md_allow_write(mddev
);
4016 while (new > conf
->max_nr_stripes
) {
4017 if (grow_one_stripe(conf
))
4018 conf
->max_nr_stripes
++;
4024 static struct md_sysfs_entry
4025 raid5_stripecache_size
= __ATTR(stripe_cache_size
, S_IRUGO
| S_IWUSR
,
4026 raid5_show_stripe_cache_size
,
4027 raid5_store_stripe_cache_size
);
4030 stripe_cache_active_show(mddev_t
*mddev
, char *page
)
4032 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4034 return sprintf(page
, "%d\n", atomic_read(&conf
->active_stripes
));
4039 static struct md_sysfs_entry
4040 raid5_stripecache_active
= __ATTR_RO(stripe_cache_active
);
4042 static struct attribute
*raid5_attrs
[] = {
4043 &raid5_stripecache_size
.attr
,
4044 &raid5_stripecache_active
.attr
,
4047 static struct attribute_group raid5_attrs_group
= {
4049 .attrs
= raid5_attrs
,
4052 static int run(mddev_t
*mddev
)
4055 int raid_disk
, memory
;
4057 struct disk_info
*disk
;
4058 struct list_head
*tmp
;
4059 int working_disks
= 0;
4061 if (mddev
->level
!= 5 && mddev
->level
!= 4 && mddev
->level
!= 6) {
4062 printk(KERN_ERR
"raid5: %s: raid level not set to 4/5/6 (%d)\n",
4063 mdname(mddev
), mddev
->level
);
4067 if (mddev
->reshape_position
!= MaxSector
) {
4068 /* Check that we can continue the reshape.
4069 * Currently only disks can change, it must
4070 * increase, and we must be past the point where
4071 * a stripe over-writes itself
4073 sector_t here_new
, here_old
;
4075 int max_degraded
= (mddev
->level
== 5 ? 1 : 2);
4077 if (mddev
->new_level
!= mddev
->level
||
4078 mddev
->new_layout
!= mddev
->layout
||
4079 mddev
->new_chunk
!= mddev
->chunk_size
) {
4080 printk(KERN_ERR
"raid5: %s: unsupported reshape "
4081 "required - aborting.\n",
4085 if (mddev
->delta_disks
<= 0) {
4086 printk(KERN_ERR
"raid5: %s: unsupported reshape "
4087 "(reduce disks) required - aborting.\n",
4091 old_disks
= mddev
->raid_disks
- mddev
->delta_disks
;
4092 /* reshape_position must be on a new-stripe boundary, and one
4093 * further up in new geometry must map after here in old
4096 here_new
= mddev
->reshape_position
;
4097 if (sector_div(here_new
, (mddev
->chunk_size
>>9)*
4098 (mddev
->raid_disks
- max_degraded
))) {
4099 printk(KERN_ERR
"raid5: reshape_position not "
4100 "on a stripe boundary\n");
4103 /* here_new is the stripe we will write to */
4104 here_old
= mddev
->reshape_position
;
4105 sector_div(here_old
, (mddev
->chunk_size
>>9)*
4106 (old_disks
-max_degraded
));
4107 /* here_old is the first stripe that we might need to read
4109 if (here_new
>= here_old
) {
4110 /* Reading from the same stripe as writing to - bad */
4111 printk(KERN_ERR
"raid5: reshape_position too early for "
4112 "auto-recovery - aborting.\n");
4115 printk(KERN_INFO
"raid5: reshape will continue\n");
4116 /* OK, we should be able to continue; */
4120 mddev
->private = kzalloc(sizeof (raid5_conf_t
), GFP_KERNEL
);
4121 if ((conf
= mddev
->private) == NULL
)
4123 if (mddev
->reshape_position
== MaxSector
) {
4124 conf
->previous_raid_disks
= conf
->raid_disks
= mddev
->raid_disks
;
4126 conf
->raid_disks
= mddev
->raid_disks
;
4127 conf
->previous_raid_disks
= mddev
->raid_disks
- mddev
->delta_disks
;
4130 conf
->disks
= kzalloc(conf
->raid_disks
* sizeof(struct disk_info
),
4135 conf
->mddev
= mddev
;
4137 if ((conf
->stripe_hashtbl
= kzalloc(PAGE_SIZE
, GFP_KERNEL
)) == NULL
)
4140 if (mddev
->level
== 6) {
4141 conf
->spare_page
= alloc_page(GFP_KERNEL
);
4142 if (!conf
->spare_page
)
4145 spin_lock_init(&conf
->device_lock
);
4146 init_waitqueue_head(&conf
->wait_for_stripe
);
4147 init_waitqueue_head(&conf
->wait_for_overlap
);
4148 INIT_LIST_HEAD(&conf
->handle_list
);
4149 INIT_LIST_HEAD(&conf
->delayed_list
);
4150 INIT_LIST_HEAD(&conf
->bitmap_list
);
4151 INIT_LIST_HEAD(&conf
->inactive_list
);
4152 atomic_set(&conf
->active_stripes
, 0);
4153 atomic_set(&conf
->preread_active_stripes
, 0);
4154 atomic_set(&conf
->active_aligned_reads
, 0);
4156 pr_debug("raid5: run(%s) called.\n", mdname(mddev
));
4158 ITERATE_RDEV(mddev
,rdev
,tmp
) {
4159 raid_disk
= rdev
->raid_disk
;
4160 if (raid_disk
>= conf
->raid_disks
4163 disk
= conf
->disks
+ raid_disk
;
4167 if (test_bit(In_sync
, &rdev
->flags
)) {
4168 char b
[BDEVNAME_SIZE
];
4169 printk(KERN_INFO
"raid5: device %s operational as raid"
4170 " disk %d\n", bdevname(rdev
->bdev
,b
),
4177 * 0 for a fully functional array, 1 or 2 for a degraded array.
4179 mddev
->degraded
= conf
->raid_disks
- working_disks
;
4180 conf
->mddev
= mddev
;
4181 conf
->chunk_size
= mddev
->chunk_size
;
4182 conf
->level
= mddev
->level
;
4183 if (conf
->level
== 6)
4184 conf
->max_degraded
= 2;
4186 conf
->max_degraded
= 1;
4187 conf
->algorithm
= mddev
->layout
;
4188 conf
->max_nr_stripes
= NR_STRIPES
;
4189 conf
->expand_progress
= mddev
->reshape_position
;
4191 /* device size must be a multiple of chunk size */
4192 mddev
->size
&= ~(mddev
->chunk_size
/1024 -1);
4193 mddev
->resync_max_sectors
= mddev
->size
<< 1;
4195 if (conf
->level
== 6 && conf
->raid_disks
< 4) {
4196 printk(KERN_ERR
"raid6: not enough configured devices for %s (%d, minimum 4)\n",
4197 mdname(mddev
), conf
->raid_disks
);
4200 if (!conf
->chunk_size
|| conf
->chunk_size
% 4) {
4201 printk(KERN_ERR
"raid5: invalid chunk size %d for %s\n",
4202 conf
->chunk_size
, mdname(mddev
));
4205 if (conf
->algorithm
> ALGORITHM_RIGHT_SYMMETRIC
) {
4207 "raid5: unsupported parity algorithm %d for %s\n",
4208 conf
->algorithm
, mdname(mddev
));
4211 if (mddev
->degraded
> conf
->max_degraded
) {
4212 printk(KERN_ERR
"raid5: not enough operational devices for %s"
4213 " (%d/%d failed)\n",
4214 mdname(mddev
), mddev
->degraded
, conf
->raid_disks
);
4218 if (mddev
->degraded
> 0 &&
4219 mddev
->recovery_cp
!= MaxSector
) {
4220 if (mddev
->ok_start_degraded
)
4222 "raid5: starting dirty degraded array: %s"
4223 "- data corruption possible.\n",
4227 "raid5: cannot start dirty degraded array for %s\n",
4234 mddev
->thread
= md_register_thread(raid5d
, mddev
, "%s_raid5");
4235 if (!mddev
->thread
) {
4237 "raid5: couldn't allocate thread for %s\n",
4242 memory
= conf
->max_nr_stripes
* (sizeof(struct stripe_head
) +
4243 conf
->raid_disks
* ((sizeof(struct bio
) + PAGE_SIZE
))) / 1024;
4244 if (grow_stripes(conf
, conf
->max_nr_stripes
)) {
4246 "raid5: couldn't allocate %dkB for buffers\n", memory
);
4247 shrink_stripes(conf
);
4248 md_unregister_thread(mddev
->thread
);
4251 printk(KERN_INFO
"raid5: allocated %dkB for %s\n",
4252 memory
, mdname(mddev
));
4254 if (mddev
->degraded
== 0)
4255 printk("raid5: raid level %d set %s active with %d out of %d"
4256 " devices, algorithm %d\n", conf
->level
, mdname(mddev
),
4257 mddev
->raid_disks
-mddev
->degraded
, mddev
->raid_disks
,
4260 printk(KERN_ALERT
"raid5: raid level %d set %s active with %d"
4261 " out of %d devices, algorithm %d\n", conf
->level
,
4262 mdname(mddev
), mddev
->raid_disks
- mddev
->degraded
,
4263 mddev
->raid_disks
, conf
->algorithm
);
4265 print_raid5_conf(conf
);
4267 if (conf
->expand_progress
!= MaxSector
) {
4268 printk("...ok start reshape thread\n");
4269 conf
->expand_lo
= conf
->expand_progress
;
4270 atomic_set(&conf
->reshape_stripes
, 0);
4271 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4272 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4273 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
4274 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
4275 mddev
->sync_thread
= md_register_thread(md_do_sync
, mddev
,
4279 /* read-ahead size must cover two whole stripes, which is
4280 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
4283 int data_disks
= conf
->previous_raid_disks
- conf
->max_degraded
;
4284 int stripe
= data_disks
*
4285 (mddev
->chunk_size
/ PAGE_SIZE
);
4286 if (mddev
->queue
->backing_dev_info
.ra_pages
< 2 * stripe
)
4287 mddev
->queue
->backing_dev_info
.ra_pages
= 2 * stripe
;
4290 /* Ok, everything is just fine now */
4291 if (sysfs_create_group(&mddev
->kobj
, &raid5_attrs_group
))
4293 "raid5: failed to create sysfs attributes for %s\n",
4296 mddev
->queue
->unplug_fn
= raid5_unplug_device
;
4297 mddev
->queue
->issue_flush_fn
= raid5_issue_flush
;
4298 mddev
->queue
->backing_dev_info
.congested_data
= mddev
;
4299 mddev
->queue
->backing_dev_info
.congested_fn
= raid5_congested
;
4301 mddev
->array_size
= mddev
->size
* (conf
->previous_raid_disks
-
4302 conf
->max_degraded
);
4304 blk_queue_merge_bvec(mddev
->queue
, raid5_mergeable_bvec
);
4309 print_raid5_conf(conf
);
4310 safe_put_page(conf
->spare_page
);
4312 kfree(conf
->stripe_hashtbl
);
4315 mddev
->private = NULL
;
4316 printk(KERN_ALERT
"raid5: failed to run raid set %s\n", mdname(mddev
));
4322 static int stop(mddev_t
*mddev
)
4324 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
4326 md_unregister_thread(mddev
->thread
);
4327 mddev
->thread
= NULL
;
4328 shrink_stripes(conf
);
4329 kfree(conf
->stripe_hashtbl
);
4330 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
4331 blk_sync_queue(mddev
->queue
); /* the unplug fn references 'conf'*/
4332 sysfs_remove_group(&mddev
->kobj
, &raid5_attrs_group
);
4335 mddev
->private = NULL
;
4340 static void print_sh (struct seq_file
*seq
, struct stripe_head
*sh
)
4344 seq_printf(seq
, "sh %llu, pd_idx %d, state %ld.\n",
4345 (unsigned long long)sh
->sector
, sh
->pd_idx
, sh
->state
);
4346 seq_printf(seq
, "sh %llu, count %d.\n",
4347 (unsigned long long)sh
->sector
, atomic_read(&sh
->count
));
4348 seq_printf(seq
, "sh %llu, ", (unsigned long long)sh
->sector
);
4349 for (i
= 0; i
< sh
->disks
; i
++) {
4350 seq_printf(seq
, "(cache%d: %p %ld) ",
4351 i
, sh
->dev
[i
].page
, sh
->dev
[i
].flags
);
4353 seq_printf(seq
, "\n");
4356 static void printall (struct seq_file
*seq
, raid5_conf_t
*conf
)
4358 struct stripe_head
*sh
;
4359 struct hlist_node
*hn
;
4362 spin_lock_irq(&conf
->device_lock
);
4363 for (i
= 0; i
< NR_HASH
; i
++) {
4364 hlist_for_each_entry(sh
, hn
, &conf
->stripe_hashtbl
[i
], hash
) {
4365 if (sh
->raid_conf
!= conf
)
4370 spin_unlock_irq(&conf
->device_lock
);
4374 static void status (struct seq_file
*seq
, mddev_t
*mddev
)
4376 raid5_conf_t
*conf
= (raid5_conf_t
*) mddev
->private;
4379 seq_printf (seq
, " level %d, %dk chunk, algorithm %d", mddev
->level
, mddev
->chunk_size
>> 10, mddev
->layout
);
4380 seq_printf (seq
, " [%d/%d] [", conf
->raid_disks
, conf
->raid_disks
- mddev
->degraded
);
4381 for (i
= 0; i
< conf
->raid_disks
; i
++)
4382 seq_printf (seq
, "%s",
4383 conf
->disks
[i
].rdev
&&
4384 test_bit(In_sync
, &conf
->disks
[i
].rdev
->flags
) ? "U" : "_");
4385 seq_printf (seq
, "]");
4387 seq_printf (seq
, "\n");
4388 printall(seq
, conf
);
4392 static void print_raid5_conf (raid5_conf_t
*conf
)
4395 struct disk_info
*tmp
;
4397 printk("RAID5 conf printout:\n");
4399 printk("(conf==NULL)\n");
4402 printk(" --- rd:%d wd:%d\n", conf
->raid_disks
,
4403 conf
->raid_disks
- conf
->mddev
->degraded
);
4405 for (i
= 0; i
< conf
->raid_disks
; i
++) {
4406 char b
[BDEVNAME_SIZE
];
4407 tmp
= conf
->disks
+ i
;
4409 printk(" disk %d, o:%d, dev:%s\n",
4410 i
, !test_bit(Faulty
, &tmp
->rdev
->flags
),
4411 bdevname(tmp
->rdev
->bdev
,b
));
4415 static int raid5_spare_active(mddev_t
*mddev
)
4418 raid5_conf_t
*conf
= mddev
->private;
4419 struct disk_info
*tmp
;
4421 for (i
= 0; i
< conf
->raid_disks
; i
++) {
4422 tmp
= conf
->disks
+ i
;
4424 && !test_bit(Faulty
, &tmp
->rdev
->flags
)
4425 && !test_and_set_bit(In_sync
, &tmp
->rdev
->flags
)) {
4426 unsigned long flags
;
4427 spin_lock_irqsave(&conf
->device_lock
, flags
);
4429 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
4432 print_raid5_conf(conf
);
4436 static int raid5_remove_disk(mddev_t
*mddev
, int number
)
4438 raid5_conf_t
*conf
= mddev
->private;
4441 struct disk_info
*p
= conf
->disks
+ number
;
4443 print_raid5_conf(conf
);
4446 if (test_bit(In_sync
, &rdev
->flags
) ||
4447 atomic_read(&rdev
->nr_pending
)) {
4453 if (atomic_read(&rdev
->nr_pending
)) {
4454 /* lost the race, try later */
4461 print_raid5_conf(conf
);
4465 static int raid5_add_disk(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
4467 raid5_conf_t
*conf
= mddev
->private;
4470 struct disk_info
*p
;
4472 if (mddev
->degraded
> conf
->max_degraded
)
4473 /* no point adding a device */
4477 * find the disk ... but prefer rdev->saved_raid_disk
4480 if (rdev
->saved_raid_disk
>= 0 &&
4481 conf
->disks
[rdev
->saved_raid_disk
].rdev
== NULL
)
4482 disk
= rdev
->saved_raid_disk
;
4485 for ( ; disk
< conf
->raid_disks
; disk
++)
4486 if ((p
=conf
->disks
+ disk
)->rdev
== NULL
) {
4487 clear_bit(In_sync
, &rdev
->flags
);
4488 rdev
->raid_disk
= disk
;
4490 if (rdev
->saved_raid_disk
!= disk
)
4492 rcu_assign_pointer(p
->rdev
, rdev
);
4495 print_raid5_conf(conf
);
4499 static int raid5_resize(mddev_t
*mddev
, sector_t sectors
)
4501 /* no resync is happening, and there is enough space
4502 * on all devices, so we can resize.
4503 * We need to make sure resync covers any new space.
4504 * If the array is shrinking we should possibly wait until
4505 * any io in the removed space completes, but it hardly seems
4508 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4510 sectors
&= ~((sector_t
)mddev
->chunk_size
/512 - 1);
4511 mddev
->array_size
= (sectors
* (mddev
->raid_disks
-conf
->max_degraded
))>>1;
4512 set_capacity(mddev
->gendisk
, mddev
->array_size
<< 1);
4514 if (sectors
/2 > mddev
->size
&& mddev
->recovery_cp
== MaxSector
) {
4515 mddev
->recovery_cp
= mddev
->size
<< 1;
4516 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4518 mddev
->size
= sectors
/2;
4519 mddev
->resync_max_sectors
= sectors
;
4523 #ifdef CONFIG_MD_RAID5_RESHAPE
4524 static int raid5_check_reshape(mddev_t
*mddev
)
4526 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4529 if (mddev
->delta_disks
< 0 ||
4530 mddev
->new_level
!= mddev
->level
)
4531 return -EINVAL
; /* Cannot shrink array or change level yet */
4532 if (mddev
->delta_disks
== 0)
4533 return 0; /* nothing to do */
4535 /* Can only proceed if there are plenty of stripe_heads.
4536 * We need a minimum of one full stripe,, and for sensible progress
4537 * it is best to have about 4 times that.
4538 * If we require 4 times, then the default 256 4K stripe_heads will
4539 * allow for chunk sizes up to 256K, which is probably OK.
4540 * If the chunk size is greater, user-space should request more
4541 * stripe_heads first.
4543 if ((mddev
->chunk_size
/ STRIPE_SIZE
) * 4 > conf
->max_nr_stripes
||
4544 (mddev
->new_chunk
/ STRIPE_SIZE
) * 4 > conf
->max_nr_stripes
) {
4545 printk(KERN_WARNING
"raid5: reshape: not enough stripes. Needed %lu\n",
4546 (mddev
->chunk_size
/ STRIPE_SIZE
)*4);
4550 err
= resize_stripes(conf
, conf
->raid_disks
+ mddev
->delta_disks
);
4554 if (mddev
->degraded
> conf
->max_degraded
)
4556 /* looks like we might be able to manage this */
4560 static int raid5_start_reshape(mddev_t
*mddev
)
4562 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4564 struct list_head
*rtmp
;
4566 int added_devices
= 0;
4567 unsigned long flags
;
4569 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4572 ITERATE_RDEV(mddev
, rdev
, rtmp
)
4573 if (rdev
->raid_disk
< 0 &&
4574 !test_bit(Faulty
, &rdev
->flags
))
4577 if (spares
- mddev
->degraded
< mddev
->delta_disks
- conf
->max_degraded
)
4578 /* Not enough devices even to make a degraded array
4583 atomic_set(&conf
->reshape_stripes
, 0);
4584 spin_lock_irq(&conf
->device_lock
);
4585 conf
->previous_raid_disks
= conf
->raid_disks
;
4586 conf
->raid_disks
+= mddev
->delta_disks
;
4587 conf
->expand_progress
= 0;
4588 conf
->expand_lo
= 0;
4589 spin_unlock_irq(&conf
->device_lock
);
4591 /* Add some new drives, as many as will fit.
4592 * We know there are enough to make the newly sized array work.
4594 ITERATE_RDEV(mddev
, rdev
, rtmp
)
4595 if (rdev
->raid_disk
< 0 &&
4596 !test_bit(Faulty
, &rdev
->flags
)) {
4597 if (raid5_add_disk(mddev
, rdev
)) {
4599 set_bit(In_sync
, &rdev
->flags
);
4601 rdev
->recovery_offset
= 0;
4602 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4603 if (sysfs_create_link(&mddev
->kobj
,
4606 "raid5: failed to create "
4607 " link %s for %s\n",
4613 spin_lock_irqsave(&conf
->device_lock
, flags
);
4614 mddev
->degraded
= (conf
->raid_disks
- conf
->previous_raid_disks
) - added_devices
;
4615 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
4616 mddev
->raid_disks
= conf
->raid_disks
;
4617 mddev
->reshape_position
= 0;
4618 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
4620 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4621 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4622 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
4623 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
4624 mddev
->sync_thread
= md_register_thread(md_do_sync
, mddev
,
4626 if (!mddev
->sync_thread
) {
4627 mddev
->recovery
= 0;
4628 spin_lock_irq(&conf
->device_lock
);
4629 mddev
->raid_disks
= conf
->raid_disks
= conf
->previous_raid_disks
;
4630 conf
->expand_progress
= MaxSector
;
4631 spin_unlock_irq(&conf
->device_lock
);
4634 md_wakeup_thread(mddev
->sync_thread
);
4635 md_new_event(mddev
);
4640 static void end_reshape(raid5_conf_t
*conf
)
4642 struct block_device
*bdev
;
4644 if (!test_bit(MD_RECOVERY_INTR
, &conf
->mddev
->recovery
)) {
4645 conf
->mddev
->array_size
= conf
->mddev
->size
*
4646 (conf
->raid_disks
- conf
->max_degraded
);
4647 set_capacity(conf
->mddev
->gendisk
, conf
->mddev
->array_size
<< 1);
4648 conf
->mddev
->changed
= 1;
4650 bdev
= bdget_disk(conf
->mddev
->gendisk
, 0);
4652 mutex_lock(&bdev
->bd_inode
->i_mutex
);
4653 i_size_write(bdev
->bd_inode
, (loff_t
)conf
->mddev
->array_size
<< 10);
4654 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
4657 spin_lock_irq(&conf
->device_lock
);
4658 conf
->expand_progress
= MaxSector
;
4659 spin_unlock_irq(&conf
->device_lock
);
4660 conf
->mddev
->reshape_position
= MaxSector
;
4662 /* read-ahead size must cover two whole stripes, which is
4663 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
4666 int data_disks
= conf
->previous_raid_disks
- conf
->max_degraded
;
4667 int stripe
= data_disks
*
4668 (conf
->mddev
->chunk_size
/ PAGE_SIZE
);
4669 if (conf
->mddev
->queue
->backing_dev_info
.ra_pages
< 2 * stripe
)
4670 conf
->mddev
->queue
->backing_dev_info
.ra_pages
= 2 * stripe
;
4675 static void raid5_quiesce(mddev_t
*mddev
, int state
)
4677 raid5_conf_t
*conf
= mddev_to_conf(mddev
);
4680 case 2: /* resume for a suspend */
4681 wake_up(&conf
->wait_for_overlap
);
4684 case 1: /* stop all writes */
4685 spin_lock_irq(&conf
->device_lock
);
4687 wait_event_lock_irq(conf
->wait_for_stripe
,
4688 atomic_read(&conf
->active_stripes
) == 0 &&
4689 atomic_read(&conf
->active_aligned_reads
) == 0,
4690 conf
->device_lock
, /* nothing */);
4691 spin_unlock_irq(&conf
->device_lock
);
4694 case 0: /* re-enable writes */
4695 spin_lock_irq(&conf
->device_lock
);
4697 wake_up(&conf
->wait_for_stripe
);
4698 wake_up(&conf
->wait_for_overlap
);
4699 spin_unlock_irq(&conf
->device_lock
);
4704 static struct mdk_personality raid6_personality
=
4708 .owner
= THIS_MODULE
,
4709 .make_request
= make_request
,
4713 .error_handler
= error
,
4714 .hot_add_disk
= raid5_add_disk
,
4715 .hot_remove_disk
= raid5_remove_disk
,
4716 .spare_active
= raid5_spare_active
,
4717 .sync_request
= sync_request
,
4718 .resize
= raid5_resize
,
4719 #ifdef CONFIG_MD_RAID5_RESHAPE
4720 .check_reshape
= raid5_check_reshape
,
4721 .start_reshape
= raid5_start_reshape
,
4723 .quiesce
= raid5_quiesce
,
4725 static struct mdk_personality raid5_personality
=
4729 .owner
= THIS_MODULE
,
4730 .make_request
= make_request
,
4734 .error_handler
= error
,
4735 .hot_add_disk
= raid5_add_disk
,
4736 .hot_remove_disk
= raid5_remove_disk
,
4737 .spare_active
= raid5_spare_active
,
4738 .sync_request
= sync_request
,
4739 .resize
= raid5_resize
,
4740 #ifdef CONFIG_MD_RAID5_RESHAPE
4741 .check_reshape
= raid5_check_reshape
,
4742 .start_reshape
= raid5_start_reshape
,
4744 .quiesce
= raid5_quiesce
,
4747 static struct mdk_personality raid4_personality
=
4751 .owner
= THIS_MODULE
,
4752 .make_request
= make_request
,
4756 .error_handler
= error
,
4757 .hot_add_disk
= raid5_add_disk
,
4758 .hot_remove_disk
= raid5_remove_disk
,
4759 .spare_active
= raid5_spare_active
,
4760 .sync_request
= sync_request
,
4761 .resize
= raid5_resize
,
4762 #ifdef CONFIG_MD_RAID5_RESHAPE
4763 .check_reshape
= raid5_check_reshape
,
4764 .start_reshape
= raid5_start_reshape
,
4766 .quiesce
= raid5_quiesce
,
4769 static int __init
raid5_init(void)
4773 e
= raid6_select_algo();
4776 register_md_personality(&raid6_personality
);
4777 register_md_personality(&raid5_personality
);
4778 register_md_personality(&raid4_personality
);
4782 static void raid5_exit(void)
4784 unregister_md_personality(&raid6_personality
);
4785 unregister_md_personality(&raid5_personality
);
4786 unregister_md_personality(&raid4_personality
);
4789 module_init(raid5_init
);
4790 module_exit(raid5_exit
);
4791 MODULE_LICENSE("GPL");
4792 MODULE_ALIAS("md-personality-4"); /* RAID5 */
4793 MODULE_ALIAS("md-raid5");
4794 MODULE_ALIAS("md-raid4");
4795 MODULE_ALIAS("md-level-5");
4796 MODULE_ALIAS("md-level-4");
4797 MODULE_ALIAS("md-personality-8"); /* RAID6 */
4798 MODULE_ALIAS("md-raid6");
4799 MODULE_ALIAS("md-level-6");
4801 /* This used to be two separate modules, they were: */
4802 MODULE_ALIAS("raid5");
4803 MODULE_ALIAS("raid6");