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->seq_write is the number of the last batch successfully written.
31 * conf->seq_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 seq_flush+1.
36 * When we are ready to do a write, if that batch hasn't been written yet,
37 * we plug the array and queue the stripe for later.
38 * When an unplug happens, we increment bm_flush, thus closing the current
40 * When we notice that bm_flush > bm_write, we write out all pending updates
41 * to the bitmap, and advance bm_write to where bm_flush was.
42 * This may occasionally write a bit out twice, but is sure never to
46 #include <linux/blkdev.h>
47 #include <linux/kthread.h>
48 #include <linux/raid/pq.h>
49 #include <linux/async_tx.h>
50 #include <linux/async.h>
51 #include <linux/seq_file.h>
52 #include <linux/cpu.h>
53 #include <linux/slab.h>
54 #include <linux/ratelimit.h>
64 #define NR_STRIPES 256
65 #define STRIPE_SIZE PAGE_SIZE
66 #define STRIPE_SHIFT (PAGE_SHIFT - 9)
67 #define STRIPE_SECTORS (STRIPE_SIZE>>9)
68 #define IO_THRESHOLD 1
69 #define BYPASS_THRESHOLD 1
70 #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
71 #define HASH_MASK (NR_HASH - 1)
73 #define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK]))
75 /* bio's attached to a stripe+device for I/O are linked together in bi_sector
76 * order without overlap. There may be several bio's per stripe+device, and
77 * a bio could span several devices.
78 * When walking this list for a particular stripe+device, we must never proceed
79 * beyond a bio that extends past this device, as the next bio might no longer
81 * This macro is used to determine the 'next' bio in the list, given the sector
82 * of the current stripe+device
84 #define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL)
86 * The following can be used to debug the driver
88 #define RAID5_PARANOIA 1
89 #if RAID5_PARANOIA && defined(CONFIG_SMP)
90 # define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock)
92 # define CHECK_DEVLOCK()
101 * We maintain a biased count of active stripes in the bottom 16 bits of
102 * bi_phys_segments, and a count of processed stripes in the upper 16 bits
104 static inline int raid5_bi_phys_segments(struct bio
*bio
)
106 return bio
->bi_phys_segments
& 0xffff;
109 static inline int raid5_bi_hw_segments(struct bio
*bio
)
111 return (bio
->bi_phys_segments
>> 16) & 0xffff;
114 static inline int raid5_dec_bi_phys_segments(struct bio
*bio
)
116 --bio
->bi_phys_segments
;
117 return raid5_bi_phys_segments(bio
);
120 static inline int raid5_dec_bi_hw_segments(struct bio
*bio
)
122 unsigned short val
= raid5_bi_hw_segments(bio
);
125 bio
->bi_phys_segments
= (val
<< 16) | raid5_bi_phys_segments(bio
);
129 static inline void raid5_set_bi_hw_segments(struct bio
*bio
, unsigned int cnt
)
131 bio
->bi_phys_segments
= raid5_bi_phys_segments(bio
) | (cnt
<< 16);
134 /* Find first data disk in a raid6 stripe */
135 static inline int raid6_d0(struct stripe_head
*sh
)
138 /* ddf always start from first device */
140 /* md starts just after Q block */
141 if (sh
->qd_idx
== sh
->disks
- 1)
144 return sh
->qd_idx
+ 1;
146 static inline int raid6_next_disk(int disk
, int raid_disks
)
149 return (disk
< raid_disks
) ? disk
: 0;
152 /* When walking through the disks in a raid5, starting at raid6_d0,
153 * We need to map each disk to a 'slot', where the data disks are slot
154 * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk
155 * is raid_disks-1. This help does that mapping.
157 static int raid6_idx_to_slot(int idx
, struct stripe_head
*sh
,
158 int *count
, int syndrome_disks
)
164 if (idx
== sh
->pd_idx
)
165 return syndrome_disks
;
166 if (idx
== sh
->qd_idx
)
167 return syndrome_disks
+ 1;
173 static void return_io(struct bio
*return_bi
)
175 struct bio
*bi
= return_bi
;
178 return_bi
= bi
->bi_next
;
186 static void print_raid5_conf (raid5_conf_t
*conf
);
188 static int stripe_operations_active(struct stripe_head
*sh
)
190 return sh
->check_state
|| sh
->reconstruct_state
||
191 test_bit(STRIPE_BIOFILL_RUN
, &sh
->state
) ||
192 test_bit(STRIPE_COMPUTE_RUN
, &sh
->state
);
195 static void __release_stripe(raid5_conf_t
*conf
, struct stripe_head
*sh
)
197 if (atomic_dec_and_test(&sh
->count
)) {
198 BUG_ON(!list_empty(&sh
->lru
));
199 BUG_ON(atomic_read(&conf
->active_stripes
)==0);
200 if (test_bit(STRIPE_HANDLE
, &sh
->state
)) {
201 if (test_bit(STRIPE_DELAYED
, &sh
->state
))
202 list_add_tail(&sh
->lru
, &conf
->delayed_list
);
203 else if (test_bit(STRIPE_BIT_DELAY
, &sh
->state
) &&
204 sh
->bm_seq
- conf
->seq_write
> 0)
205 list_add_tail(&sh
->lru
, &conf
->bitmap_list
);
207 clear_bit(STRIPE_BIT_DELAY
, &sh
->state
);
208 list_add_tail(&sh
->lru
, &conf
->handle_list
);
210 md_wakeup_thread(conf
->mddev
->thread
);
212 BUG_ON(stripe_operations_active(sh
));
213 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
214 atomic_dec(&conf
->preread_active_stripes
);
215 if (atomic_read(&conf
->preread_active_stripes
) < IO_THRESHOLD
)
216 md_wakeup_thread(conf
->mddev
->thread
);
218 atomic_dec(&conf
->active_stripes
);
219 if (!test_bit(STRIPE_EXPANDING
, &sh
->state
)) {
220 list_add_tail(&sh
->lru
, &conf
->inactive_list
);
221 wake_up(&conf
->wait_for_stripe
);
222 if (conf
->retry_read_aligned
)
223 md_wakeup_thread(conf
->mddev
->thread
);
229 static void release_stripe(struct stripe_head
*sh
)
231 raid5_conf_t
*conf
= sh
->raid_conf
;
234 spin_lock_irqsave(&conf
->device_lock
, flags
);
235 __release_stripe(conf
, sh
);
236 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
239 static inline void remove_hash(struct stripe_head
*sh
)
241 pr_debug("remove_hash(), stripe %llu\n",
242 (unsigned long long)sh
->sector
);
244 hlist_del_init(&sh
->hash
);
247 static inline void insert_hash(raid5_conf_t
*conf
, struct stripe_head
*sh
)
249 struct hlist_head
*hp
= stripe_hash(conf
, sh
->sector
);
251 pr_debug("insert_hash(), stripe %llu\n",
252 (unsigned long long)sh
->sector
);
255 hlist_add_head(&sh
->hash
, hp
);
259 /* find an idle stripe, make sure it is unhashed, and return it. */
260 static struct stripe_head
*get_free_stripe(raid5_conf_t
*conf
)
262 struct stripe_head
*sh
= NULL
;
263 struct list_head
*first
;
266 if (list_empty(&conf
->inactive_list
))
268 first
= conf
->inactive_list
.next
;
269 sh
= list_entry(first
, struct stripe_head
, lru
);
270 list_del_init(first
);
272 atomic_inc(&conf
->active_stripes
);
277 static void shrink_buffers(struct stripe_head
*sh
)
281 int num
= sh
->raid_conf
->pool_size
;
283 for (i
= 0; i
< num
; i
++) {
287 sh
->dev
[i
].page
= NULL
;
292 static int grow_buffers(struct stripe_head
*sh
)
295 int num
= sh
->raid_conf
->pool_size
;
297 for (i
= 0; i
< num
; i
++) {
300 if (!(page
= alloc_page(GFP_KERNEL
))) {
303 sh
->dev
[i
].page
= page
;
308 static void raid5_build_block(struct stripe_head
*sh
, int i
, int previous
);
309 static void stripe_set_idx(sector_t stripe
, raid5_conf_t
*conf
, int previous
,
310 struct stripe_head
*sh
);
312 static void init_stripe(struct stripe_head
*sh
, sector_t sector
, int previous
)
314 raid5_conf_t
*conf
= sh
->raid_conf
;
317 BUG_ON(atomic_read(&sh
->count
) != 0);
318 BUG_ON(test_bit(STRIPE_HANDLE
, &sh
->state
));
319 BUG_ON(stripe_operations_active(sh
));
322 pr_debug("init_stripe called, stripe %llu\n",
323 (unsigned long long)sh
->sector
);
327 sh
->generation
= conf
->generation
- previous
;
328 sh
->disks
= previous
? conf
->previous_raid_disks
: conf
->raid_disks
;
330 stripe_set_idx(sector
, conf
, previous
, sh
);
334 for (i
= sh
->disks
; i
--; ) {
335 struct r5dev
*dev
= &sh
->dev
[i
];
337 if (dev
->toread
|| dev
->read
|| dev
->towrite
|| dev
->written
||
338 test_bit(R5_LOCKED
, &dev
->flags
)) {
339 printk(KERN_ERR
"sector=%llx i=%d %p %p %p %p %d\n",
340 (unsigned long long)sh
->sector
, i
, dev
->toread
,
341 dev
->read
, dev
->towrite
, dev
->written
,
342 test_bit(R5_LOCKED
, &dev
->flags
));
346 raid5_build_block(sh
, i
, previous
);
348 insert_hash(conf
, sh
);
351 static struct stripe_head
*__find_stripe(raid5_conf_t
*conf
, sector_t sector
,
354 struct stripe_head
*sh
;
355 struct hlist_node
*hn
;
358 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector
);
359 hlist_for_each_entry(sh
, hn
, stripe_hash(conf
, sector
), hash
)
360 if (sh
->sector
== sector
&& sh
->generation
== generation
)
362 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector
);
367 * Need to check if array has failed when deciding whether to:
369 * - remove non-faulty devices
372 * This determination is simple when no reshape is happening.
373 * However if there is a reshape, we need to carefully check
374 * both the before and after sections.
375 * This is because some failed devices may only affect one
376 * of the two sections, and some non-in_sync devices may
377 * be insync in the section most affected by failed devices.
379 static int has_failed(raid5_conf_t
*conf
)
383 if (conf
->mddev
->reshape_position
== MaxSector
)
384 return conf
->mddev
->degraded
> conf
->max_degraded
;
388 for (i
= 0; i
< conf
->previous_raid_disks
; i
++) {
389 mdk_rdev_t
*rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
390 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
392 else if (test_bit(In_sync
, &rdev
->flags
))
395 /* not in-sync or faulty.
396 * If the reshape increases the number of devices,
397 * this is being recovered by the reshape, so
398 * this 'previous' section is not in_sync.
399 * If the number of devices is being reduced however,
400 * the device can only be part of the array if
401 * we are reverting a reshape, so this section will
404 if (conf
->raid_disks
>= conf
->previous_raid_disks
)
408 if (degraded
> conf
->max_degraded
)
412 for (i
= 0; i
< conf
->raid_disks
; i
++) {
413 mdk_rdev_t
*rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
414 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
416 else if (test_bit(In_sync
, &rdev
->flags
))
419 /* not in-sync or faulty.
420 * If reshape increases the number of devices, this
421 * section has already been recovered, else it
422 * almost certainly hasn't.
424 if (conf
->raid_disks
<= conf
->previous_raid_disks
)
428 if (degraded
> conf
->max_degraded
)
433 static struct stripe_head
*
434 get_active_stripe(raid5_conf_t
*conf
, sector_t sector
,
435 int previous
, int noblock
, int noquiesce
)
437 struct stripe_head
*sh
;
439 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector
);
441 spin_lock_irq(&conf
->device_lock
);
444 wait_event_lock_irq(conf
->wait_for_stripe
,
445 conf
->quiesce
== 0 || noquiesce
,
446 conf
->device_lock
, /* nothing */);
447 sh
= __find_stripe(conf
, sector
, conf
->generation
- previous
);
449 if (!conf
->inactive_blocked
)
450 sh
= get_free_stripe(conf
);
451 if (noblock
&& sh
== NULL
)
454 conf
->inactive_blocked
= 1;
455 wait_event_lock_irq(conf
->wait_for_stripe
,
456 !list_empty(&conf
->inactive_list
) &&
457 (atomic_read(&conf
->active_stripes
)
458 < (conf
->max_nr_stripes
*3/4)
459 || !conf
->inactive_blocked
),
462 conf
->inactive_blocked
= 0;
464 init_stripe(sh
, sector
, previous
);
466 if (atomic_read(&sh
->count
)) {
467 BUG_ON(!list_empty(&sh
->lru
)
468 && !test_bit(STRIPE_EXPANDING
, &sh
->state
));
470 if (!test_bit(STRIPE_HANDLE
, &sh
->state
))
471 atomic_inc(&conf
->active_stripes
);
472 if (list_empty(&sh
->lru
) &&
473 !test_bit(STRIPE_EXPANDING
, &sh
->state
))
475 list_del_init(&sh
->lru
);
478 } while (sh
== NULL
);
481 atomic_inc(&sh
->count
);
483 spin_unlock_irq(&conf
->device_lock
);
488 raid5_end_read_request(struct bio
*bi
, int error
);
490 raid5_end_write_request(struct bio
*bi
, int error
);
492 static void ops_run_io(struct stripe_head
*sh
, struct stripe_head_state
*s
)
494 raid5_conf_t
*conf
= sh
->raid_conf
;
495 int i
, disks
= sh
->disks
;
499 for (i
= disks
; i
--; ) {
503 if (test_and_clear_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
)) {
504 if (test_and_clear_bit(R5_WantFUA
, &sh
->dev
[i
].flags
))
508 } else if (test_and_clear_bit(R5_Wantread
, &sh
->dev
[i
].flags
))
513 bi
= &sh
->dev
[i
].req
;
517 bi
->bi_end_io
= raid5_end_write_request
;
519 bi
->bi_end_io
= raid5_end_read_request
;
522 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
523 if (rdev
&& test_bit(Faulty
, &rdev
->flags
))
526 atomic_inc(&rdev
->nr_pending
);
530 if (s
->syncing
|| s
->expanding
|| s
->expanded
)
531 md_sync_acct(rdev
->bdev
, STRIPE_SECTORS
);
533 set_bit(STRIPE_IO_STARTED
, &sh
->state
);
535 bi
->bi_bdev
= rdev
->bdev
;
536 pr_debug("%s: for %llu schedule op %ld on disc %d\n",
537 __func__
, (unsigned long long)sh
->sector
,
539 atomic_inc(&sh
->count
);
540 bi
->bi_sector
= sh
->sector
+ rdev
->data_offset
;
541 bi
->bi_flags
= 1 << BIO_UPTODATE
;
545 bi
->bi_io_vec
= &sh
->dev
[i
].vec
;
546 bi
->bi_io_vec
[0].bv_len
= STRIPE_SIZE
;
547 bi
->bi_io_vec
[0].bv_offset
= 0;
548 bi
->bi_size
= STRIPE_SIZE
;
550 generic_make_request(bi
);
553 set_bit(STRIPE_DEGRADED
, &sh
->state
);
554 pr_debug("skip op %ld on disc %d for sector %llu\n",
555 bi
->bi_rw
, i
, (unsigned long long)sh
->sector
);
556 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
557 set_bit(STRIPE_HANDLE
, &sh
->state
);
562 static struct dma_async_tx_descriptor
*
563 async_copy_data(int frombio
, struct bio
*bio
, struct page
*page
,
564 sector_t sector
, struct dma_async_tx_descriptor
*tx
)
567 struct page
*bio_page
;
570 struct async_submit_ctl submit
;
571 enum async_tx_flags flags
= 0;
573 if (bio
->bi_sector
>= sector
)
574 page_offset
= (signed)(bio
->bi_sector
- sector
) * 512;
576 page_offset
= (signed)(sector
- bio
->bi_sector
) * -512;
579 flags
|= ASYNC_TX_FENCE
;
580 init_async_submit(&submit
, flags
, tx
, NULL
, NULL
, NULL
);
582 bio_for_each_segment(bvl
, bio
, i
) {
583 int len
= bvl
->bv_len
;
587 if (page_offset
< 0) {
588 b_offset
= -page_offset
;
589 page_offset
+= b_offset
;
593 if (len
> 0 && page_offset
+ len
> STRIPE_SIZE
)
594 clen
= STRIPE_SIZE
- page_offset
;
599 b_offset
+= bvl
->bv_offset
;
600 bio_page
= bvl
->bv_page
;
602 tx
= async_memcpy(page
, bio_page
, page_offset
,
603 b_offset
, clen
, &submit
);
605 tx
= async_memcpy(bio_page
, page
, b_offset
,
606 page_offset
, clen
, &submit
);
608 /* chain the operations */
609 submit
.depend_tx
= tx
;
611 if (clen
< len
) /* hit end of page */
619 static void ops_complete_biofill(void *stripe_head_ref
)
621 struct stripe_head
*sh
= stripe_head_ref
;
622 struct bio
*return_bi
= NULL
;
623 raid5_conf_t
*conf
= sh
->raid_conf
;
626 pr_debug("%s: stripe %llu\n", __func__
,
627 (unsigned long long)sh
->sector
);
629 /* clear completed biofills */
630 spin_lock_irq(&conf
->device_lock
);
631 for (i
= sh
->disks
; i
--; ) {
632 struct r5dev
*dev
= &sh
->dev
[i
];
634 /* acknowledge completion of a biofill operation */
635 /* and check if we need to reply to a read request,
636 * new R5_Wantfill requests are held off until
637 * !STRIPE_BIOFILL_RUN
639 if (test_and_clear_bit(R5_Wantfill
, &dev
->flags
)) {
640 struct bio
*rbi
, *rbi2
;
645 while (rbi
&& rbi
->bi_sector
<
646 dev
->sector
+ STRIPE_SECTORS
) {
647 rbi2
= r5_next_bio(rbi
, dev
->sector
);
648 if (!raid5_dec_bi_phys_segments(rbi
)) {
649 rbi
->bi_next
= return_bi
;
656 spin_unlock_irq(&conf
->device_lock
);
657 clear_bit(STRIPE_BIOFILL_RUN
, &sh
->state
);
659 return_io(return_bi
);
661 set_bit(STRIPE_HANDLE
, &sh
->state
);
665 static void ops_run_biofill(struct stripe_head
*sh
)
667 struct dma_async_tx_descriptor
*tx
= NULL
;
668 raid5_conf_t
*conf
= sh
->raid_conf
;
669 struct async_submit_ctl submit
;
672 pr_debug("%s: stripe %llu\n", __func__
,
673 (unsigned long long)sh
->sector
);
675 for (i
= sh
->disks
; i
--; ) {
676 struct r5dev
*dev
= &sh
->dev
[i
];
677 if (test_bit(R5_Wantfill
, &dev
->flags
)) {
679 spin_lock_irq(&conf
->device_lock
);
680 dev
->read
= rbi
= dev
->toread
;
682 spin_unlock_irq(&conf
->device_lock
);
683 while (rbi
&& rbi
->bi_sector
<
684 dev
->sector
+ STRIPE_SECTORS
) {
685 tx
= async_copy_data(0, rbi
, dev
->page
,
687 rbi
= r5_next_bio(rbi
, dev
->sector
);
692 atomic_inc(&sh
->count
);
693 init_async_submit(&submit
, ASYNC_TX_ACK
, tx
, ops_complete_biofill
, sh
, NULL
);
694 async_trigger_callback(&submit
);
697 static void mark_target_uptodate(struct stripe_head
*sh
, int target
)
704 tgt
= &sh
->dev
[target
];
705 set_bit(R5_UPTODATE
, &tgt
->flags
);
706 BUG_ON(!test_bit(R5_Wantcompute
, &tgt
->flags
));
707 clear_bit(R5_Wantcompute
, &tgt
->flags
);
710 static void ops_complete_compute(void *stripe_head_ref
)
712 struct stripe_head
*sh
= stripe_head_ref
;
714 pr_debug("%s: stripe %llu\n", __func__
,
715 (unsigned long long)sh
->sector
);
717 /* mark the computed target(s) as uptodate */
718 mark_target_uptodate(sh
, sh
->ops
.target
);
719 mark_target_uptodate(sh
, sh
->ops
.target2
);
721 clear_bit(STRIPE_COMPUTE_RUN
, &sh
->state
);
722 if (sh
->check_state
== check_state_compute_run
)
723 sh
->check_state
= check_state_compute_result
;
724 set_bit(STRIPE_HANDLE
, &sh
->state
);
728 /* return a pointer to the address conversion region of the scribble buffer */
729 static addr_conv_t
*to_addr_conv(struct stripe_head
*sh
,
730 struct raid5_percpu
*percpu
)
732 return percpu
->scribble
+ sizeof(struct page
*) * (sh
->disks
+ 2);
735 static struct dma_async_tx_descriptor
*
736 ops_run_compute5(struct stripe_head
*sh
, struct raid5_percpu
*percpu
)
738 int disks
= sh
->disks
;
739 struct page
**xor_srcs
= percpu
->scribble
;
740 int target
= sh
->ops
.target
;
741 struct r5dev
*tgt
= &sh
->dev
[target
];
742 struct page
*xor_dest
= tgt
->page
;
744 struct dma_async_tx_descriptor
*tx
;
745 struct async_submit_ctl submit
;
748 pr_debug("%s: stripe %llu block: %d\n",
749 __func__
, (unsigned long long)sh
->sector
, target
);
750 BUG_ON(!test_bit(R5_Wantcompute
, &tgt
->flags
));
752 for (i
= disks
; i
--; )
754 xor_srcs
[count
++] = sh
->dev
[i
].page
;
756 atomic_inc(&sh
->count
);
758 init_async_submit(&submit
, ASYNC_TX_FENCE
|ASYNC_TX_XOR_ZERO_DST
, NULL
,
759 ops_complete_compute
, sh
, to_addr_conv(sh
, percpu
));
760 if (unlikely(count
== 1))
761 tx
= async_memcpy(xor_dest
, xor_srcs
[0], 0, 0, STRIPE_SIZE
, &submit
);
763 tx
= async_xor(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
, &submit
);
768 /* set_syndrome_sources - populate source buffers for gen_syndrome
769 * @srcs - (struct page *) array of size sh->disks
770 * @sh - stripe_head to parse
772 * Populates srcs in proper layout order for the stripe and returns the
773 * 'count' of sources to be used in a call to async_gen_syndrome. The P
774 * destination buffer is recorded in srcs[count] and the Q destination
775 * is recorded in srcs[count+1]].
777 static int set_syndrome_sources(struct page
**srcs
, struct stripe_head
*sh
)
779 int disks
= sh
->disks
;
780 int syndrome_disks
= sh
->ddf_layout
? disks
: (disks
- 2);
781 int d0_idx
= raid6_d0(sh
);
785 for (i
= 0; i
< disks
; i
++)
791 int slot
= raid6_idx_to_slot(i
, sh
, &count
, syndrome_disks
);
793 srcs
[slot
] = sh
->dev
[i
].page
;
794 i
= raid6_next_disk(i
, disks
);
795 } while (i
!= d0_idx
);
797 return syndrome_disks
;
800 static struct dma_async_tx_descriptor
*
801 ops_run_compute6_1(struct stripe_head
*sh
, struct raid5_percpu
*percpu
)
803 int disks
= sh
->disks
;
804 struct page
**blocks
= percpu
->scribble
;
806 int qd_idx
= sh
->qd_idx
;
807 struct dma_async_tx_descriptor
*tx
;
808 struct async_submit_ctl submit
;
814 if (sh
->ops
.target
< 0)
815 target
= sh
->ops
.target2
;
816 else if (sh
->ops
.target2
< 0)
817 target
= sh
->ops
.target
;
819 /* we should only have one valid target */
822 pr_debug("%s: stripe %llu block: %d\n",
823 __func__
, (unsigned long long)sh
->sector
, target
);
825 tgt
= &sh
->dev
[target
];
826 BUG_ON(!test_bit(R5_Wantcompute
, &tgt
->flags
));
829 atomic_inc(&sh
->count
);
831 if (target
== qd_idx
) {
832 count
= set_syndrome_sources(blocks
, sh
);
833 blocks
[count
] = NULL
; /* regenerating p is not necessary */
834 BUG_ON(blocks
[count
+1] != dest
); /* q should already be set */
835 init_async_submit(&submit
, ASYNC_TX_FENCE
, NULL
,
836 ops_complete_compute
, sh
,
837 to_addr_conv(sh
, percpu
));
838 tx
= async_gen_syndrome(blocks
, 0, count
+2, STRIPE_SIZE
, &submit
);
840 /* Compute any data- or p-drive using XOR */
842 for (i
= disks
; i
-- ; ) {
843 if (i
== target
|| i
== qd_idx
)
845 blocks
[count
++] = sh
->dev
[i
].page
;
848 init_async_submit(&submit
, ASYNC_TX_FENCE
|ASYNC_TX_XOR_ZERO_DST
,
849 NULL
, ops_complete_compute
, sh
,
850 to_addr_conv(sh
, percpu
));
851 tx
= async_xor(dest
, blocks
, 0, count
, STRIPE_SIZE
, &submit
);
857 static struct dma_async_tx_descriptor
*
858 ops_run_compute6_2(struct stripe_head
*sh
, struct raid5_percpu
*percpu
)
860 int i
, count
, disks
= sh
->disks
;
861 int syndrome_disks
= sh
->ddf_layout
? disks
: disks
-2;
862 int d0_idx
= raid6_d0(sh
);
863 int faila
= -1, failb
= -1;
864 int target
= sh
->ops
.target
;
865 int target2
= sh
->ops
.target2
;
866 struct r5dev
*tgt
= &sh
->dev
[target
];
867 struct r5dev
*tgt2
= &sh
->dev
[target2
];
868 struct dma_async_tx_descriptor
*tx
;
869 struct page
**blocks
= percpu
->scribble
;
870 struct async_submit_ctl submit
;
872 pr_debug("%s: stripe %llu block1: %d block2: %d\n",
873 __func__
, (unsigned long long)sh
->sector
, target
, target2
);
874 BUG_ON(target
< 0 || target2
< 0);
875 BUG_ON(!test_bit(R5_Wantcompute
, &tgt
->flags
));
876 BUG_ON(!test_bit(R5_Wantcompute
, &tgt2
->flags
));
878 /* we need to open-code set_syndrome_sources to handle the
879 * slot number conversion for 'faila' and 'failb'
881 for (i
= 0; i
< disks
; i
++)
886 int slot
= raid6_idx_to_slot(i
, sh
, &count
, syndrome_disks
);
888 blocks
[slot
] = sh
->dev
[i
].page
;
894 i
= raid6_next_disk(i
, disks
);
895 } while (i
!= d0_idx
);
897 BUG_ON(faila
== failb
);
900 pr_debug("%s: stripe: %llu faila: %d failb: %d\n",
901 __func__
, (unsigned long long)sh
->sector
, faila
, failb
);
903 atomic_inc(&sh
->count
);
905 if (failb
== syndrome_disks
+1) {
906 /* Q disk is one of the missing disks */
907 if (faila
== syndrome_disks
) {
908 /* Missing P+Q, just recompute */
909 init_async_submit(&submit
, ASYNC_TX_FENCE
, NULL
,
910 ops_complete_compute
, sh
,
911 to_addr_conv(sh
, percpu
));
912 return async_gen_syndrome(blocks
, 0, syndrome_disks
+2,
913 STRIPE_SIZE
, &submit
);
917 int qd_idx
= sh
->qd_idx
;
919 /* Missing D+Q: recompute D from P, then recompute Q */
920 if (target
== qd_idx
)
921 data_target
= target2
;
923 data_target
= target
;
926 for (i
= disks
; i
-- ; ) {
927 if (i
== data_target
|| i
== qd_idx
)
929 blocks
[count
++] = sh
->dev
[i
].page
;
931 dest
= sh
->dev
[data_target
].page
;
932 init_async_submit(&submit
,
933 ASYNC_TX_FENCE
|ASYNC_TX_XOR_ZERO_DST
,
935 to_addr_conv(sh
, percpu
));
936 tx
= async_xor(dest
, blocks
, 0, count
, STRIPE_SIZE
,
939 count
= set_syndrome_sources(blocks
, sh
);
940 init_async_submit(&submit
, ASYNC_TX_FENCE
, tx
,
941 ops_complete_compute
, sh
,
942 to_addr_conv(sh
, percpu
));
943 return async_gen_syndrome(blocks
, 0, count
+2,
944 STRIPE_SIZE
, &submit
);
947 init_async_submit(&submit
, ASYNC_TX_FENCE
, NULL
,
948 ops_complete_compute
, sh
,
949 to_addr_conv(sh
, percpu
));
950 if (failb
== syndrome_disks
) {
951 /* We're missing D+P. */
952 return async_raid6_datap_recov(syndrome_disks
+2,
956 /* We're missing D+D. */
957 return async_raid6_2data_recov(syndrome_disks
+2,
958 STRIPE_SIZE
, faila
, failb
,
965 static void ops_complete_prexor(void *stripe_head_ref
)
967 struct stripe_head
*sh
= stripe_head_ref
;
969 pr_debug("%s: stripe %llu\n", __func__
,
970 (unsigned long long)sh
->sector
);
973 static struct dma_async_tx_descriptor
*
974 ops_run_prexor(struct stripe_head
*sh
, struct raid5_percpu
*percpu
,
975 struct dma_async_tx_descriptor
*tx
)
977 int disks
= sh
->disks
;
978 struct page
**xor_srcs
= percpu
->scribble
;
979 int count
= 0, pd_idx
= sh
->pd_idx
, i
;
980 struct async_submit_ctl submit
;
982 /* existing parity data subtracted */
983 struct page
*xor_dest
= xor_srcs
[count
++] = sh
->dev
[pd_idx
].page
;
985 pr_debug("%s: stripe %llu\n", __func__
,
986 (unsigned long long)sh
->sector
);
988 for (i
= disks
; i
--; ) {
989 struct r5dev
*dev
= &sh
->dev
[i
];
990 /* Only process blocks that are known to be uptodate */
991 if (test_bit(R5_Wantdrain
, &dev
->flags
))
992 xor_srcs
[count
++] = dev
->page
;
995 init_async_submit(&submit
, ASYNC_TX_FENCE
|ASYNC_TX_XOR_DROP_DST
, tx
,
996 ops_complete_prexor
, sh
, to_addr_conv(sh
, percpu
));
997 tx
= async_xor(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
, &submit
);
1002 static struct dma_async_tx_descriptor
*
1003 ops_run_biodrain(struct stripe_head
*sh
, struct dma_async_tx_descriptor
*tx
)
1005 int disks
= sh
->disks
;
1008 pr_debug("%s: stripe %llu\n", __func__
,
1009 (unsigned long long)sh
->sector
);
1011 for (i
= disks
; i
--; ) {
1012 struct r5dev
*dev
= &sh
->dev
[i
];
1015 if (test_and_clear_bit(R5_Wantdrain
, &dev
->flags
)) {
1018 spin_lock_irq(&sh
->raid_conf
->device_lock
);
1019 chosen
= dev
->towrite
;
1020 dev
->towrite
= NULL
;
1021 BUG_ON(dev
->written
);
1022 wbi
= dev
->written
= chosen
;
1023 spin_unlock_irq(&sh
->raid_conf
->device_lock
);
1025 while (wbi
&& wbi
->bi_sector
<
1026 dev
->sector
+ STRIPE_SECTORS
) {
1027 if (wbi
->bi_rw
& REQ_FUA
)
1028 set_bit(R5_WantFUA
, &dev
->flags
);
1029 tx
= async_copy_data(1, wbi
, dev
->page
,
1031 wbi
= r5_next_bio(wbi
, dev
->sector
);
1039 static void ops_complete_reconstruct(void *stripe_head_ref
)
1041 struct stripe_head
*sh
= stripe_head_ref
;
1042 int disks
= sh
->disks
;
1043 int pd_idx
= sh
->pd_idx
;
1044 int qd_idx
= sh
->qd_idx
;
1048 pr_debug("%s: stripe %llu\n", __func__
,
1049 (unsigned long long)sh
->sector
);
1051 for (i
= disks
; i
--; )
1052 fua
|= test_bit(R5_WantFUA
, &sh
->dev
[i
].flags
);
1054 for (i
= disks
; i
--; ) {
1055 struct r5dev
*dev
= &sh
->dev
[i
];
1057 if (dev
->written
|| i
== pd_idx
|| i
== qd_idx
) {
1058 set_bit(R5_UPTODATE
, &dev
->flags
);
1060 set_bit(R5_WantFUA
, &dev
->flags
);
1064 if (sh
->reconstruct_state
== reconstruct_state_drain_run
)
1065 sh
->reconstruct_state
= reconstruct_state_drain_result
;
1066 else if (sh
->reconstruct_state
== reconstruct_state_prexor_drain_run
)
1067 sh
->reconstruct_state
= reconstruct_state_prexor_drain_result
;
1069 BUG_ON(sh
->reconstruct_state
!= reconstruct_state_run
);
1070 sh
->reconstruct_state
= reconstruct_state_result
;
1073 set_bit(STRIPE_HANDLE
, &sh
->state
);
1078 ops_run_reconstruct5(struct stripe_head
*sh
, struct raid5_percpu
*percpu
,
1079 struct dma_async_tx_descriptor
*tx
)
1081 int disks
= sh
->disks
;
1082 struct page
**xor_srcs
= percpu
->scribble
;
1083 struct async_submit_ctl submit
;
1084 int count
= 0, pd_idx
= sh
->pd_idx
, i
;
1085 struct page
*xor_dest
;
1087 unsigned long flags
;
1089 pr_debug("%s: stripe %llu\n", __func__
,
1090 (unsigned long long)sh
->sector
);
1092 /* check if prexor is active which means only process blocks
1093 * that are part of a read-modify-write (written)
1095 if (sh
->reconstruct_state
== reconstruct_state_prexor_drain_run
) {
1097 xor_dest
= xor_srcs
[count
++] = sh
->dev
[pd_idx
].page
;
1098 for (i
= disks
; i
--; ) {
1099 struct r5dev
*dev
= &sh
->dev
[i
];
1101 xor_srcs
[count
++] = dev
->page
;
1104 xor_dest
= sh
->dev
[pd_idx
].page
;
1105 for (i
= disks
; i
--; ) {
1106 struct r5dev
*dev
= &sh
->dev
[i
];
1108 xor_srcs
[count
++] = dev
->page
;
1112 /* 1/ if we prexor'd then the dest is reused as a source
1113 * 2/ if we did not prexor then we are redoing the parity
1114 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
1115 * for the synchronous xor case
1117 flags
= ASYNC_TX_ACK
|
1118 (prexor
? ASYNC_TX_XOR_DROP_DST
: ASYNC_TX_XOR_ZERO_DST
);
1120 atomic_inc(&sh
->count
);
1122 init_async_submit(&submit
, flags
, tx
, ops_complete_reconstruct
, sh
,
1123 to_addr_conv(sh
, percpu
));
1124 if (unlikely(count
== 1))
1125 tx
= async_memcpy(xor_dest
, xor_srcs
[0], 0, 0, STRIPE_SIZE
, &submit
);
1127 tx
= async_xor(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
, &submit
);
1131 ops_run_reconstruct6(struct stripe_head
*sh
, struct raid5_percpu
*percpu
,
1132 struct dma_async_tx_descriptor
*tx
)
1134 struct async_submit_ctl submit
;
1135 struct page
**blocks
= percpu
->scribble
;
1138 pr_debug("%s: stripe %llu\n", __func__
, (unsigned long long)sh
->sector
);
1140 count
= set_syndrome_sources(blocks
, sh
);
1142 atomic_inc(&sh
->count
);
1144 init_async_submit(&submit
, ASYNC_TX_ACK
, tx
, ops_complete_reconstruct
,
1145 sh
, to_addr_conv(sh
, percpu
));
1146 async_gen_syndrome(blocks
, 0, count
+2, STRIPE_SIZE
, &submit
);
1149 static void ops_complete_check(void *stripe_head_ref
)
1151 struct stripe_head
*sh
= stripe_head_ref
;
1153 pr_debug("%s: stripe %llu\n", __func__
,
1154 (unsigned long long)sh
->sector
);
1156 sh
->check_state
= check_state_check_result
;
1157 set_bit(STRIPE_HANDLE
, &sh
->state
);
1161 static void ops_run_check_p(struct stripe_head
*sh
, struct raid5_percpu
*percpu
)
1163 int disks
= sh
->disks
;
1164 int pd_idx
= sh
->pd_idx
;
1165 int qd_idx
= sh
->qd_idx
;
1166 struct page
*xor_dest
;
1167 struct page
**xor_srcs
= percpu
->scribble
;
1168 struct dma_async_tx_descriptor
*tx
;
1169 struct async_submit_ctl submit
;
1173 pr_debug("%s: stripe %llu\n", __func__
,
1174 (unsigned long long)sh
->sector
);
1177 xor_dest
= sh
->dev
[pd_idx
].page
;
1178 xor_srcs
[count
++] = xor_dest
;
1179 for (i
= disks
; i
--; ) {
1180 if (i
== pd_idx
|| i
== qd_idx
)
1182 xor_srcs
[count
++] = sh
->dev
[i
].page
;
1185 init_async_submit(&submit
, 0, NULL
, NULL
, NULL
,
1186 to_addr_conv(sh
, percpu
));
1187 tx
= async_xor_val(xor_dest
, xor_srcs
, 0, count
, STRIPE_SIZE
,
1188 &sh
->ops
.zero_sum_result
, &submit
);
1190 atomic_inc(&sh
->count
);
1191 init_async_submit(&submit
, ASYNC_TX_ACK
, tx
, ops_complete_check
, sh
, NULL
);
1192 tx
= async_trigger_callback(&submit
);
1195 static void ops_run_check_pq(struct stripe_head
*sh
, struct raid5_percpu
*percpu
, int checkp
)
1197 struct page
**srcs
= percpu
->scribble
;
1198 struct async_submit_ctl submit
;
1201 pr_debug("%s: stripe %llu checkp: %d\n", __func__
,
1202 (unsigned long long)sh
->sector
, checkp
);
1204 count
= set_syndrome_sources(srcs
, sh
);
1208 atomic_inc(&sh
->count
);
1209 init_async_submit(&submit
, ASYNC_TX_ACK
, NULL
, ops_complete_check
,
1210 sh
, to_addr_conv(sh
, percpu
));
1211 async_syndrome_val(srcs
, 0, count
+2, STRIPE_SIZE
,
1212 &sh
->ops
.zero_sum_result
, percpu
->spare_page
, &submit
);
1215 static void __raid_run_ops(struct stripe_head
*sh
, unsigned long ops_request
)
1217 int overlap_clear
= 0, i
, disks
= sh
->disks
;
1218 struct dma_async_tx_descriptor
*tx
= NULL
;
1219 raid5_conf_t
*conf
= sh
->raid_conf
;
1220 int level
= conf
->level
;
1221 struct raid5_percpu
*percpu
;
1225 percpu
= per_cpu_ptr(conf
->percpu
, cpu
);
1226 if (test_bit(STRIPE_OP_BIOFILL
, &ops_request
)) {
1227 ops_run_biofill(sh
);
1231 if (test_bit(STRIPE_OP_COMPUTE_BLK
, &ops_request
)) {
1233 tx
= ops_run_compute5(sh
, percpu
);
1235 if (sh
->ops
.target2
< 0 || sh
->ops
.target
< 0)
1236 tx
= ops_run_compute6_1(sh
, percpu
);
1238 tx
= ops_run_compute6_2(sh
, percpu
);
1240 /* terminate the chain if reconstruct is not set to be run */
1241 if (tx
&& !test_bit(STRIPE_OP_RECONSTRUCT
, &ops_request
))
1245 if (test_bit(STRIPE_OP_PREXOR
, &ops_request
))
1246 tx
= ops_run_prexor(sh
, percpu
, tx
);
1248 if (test_bit(STRIPE_OP_BIODRAIN
, &ops_request
)) {
1249 tx
= ops_run_biodrain(sh
, tx
);
1253 if (test_bit(STRIPE_OP_RECONSTRUCT
, &ops_request
)) {
1255 ops_run_reconstruct5(sh
, percpu
, tx
);
1257 ops_run_reconstruct6(sh
, percpu
, tx
);
1260 if (test_bit(STRIPE_OP_CHECK
, &ops_request
)) {
1261 if (sh
->check_state
== check_state_run
)
1262 ops_run_check_p(sh
, percpu
);
1263 else if (sh
->check_state
== check_state_run_q
)
1264 ops_run_check_pq(sh
, percpu
, 0);
1265 else if (sh
->check_state
== check_state_run_pq
)
1266 ops_run_check_pq(sh
, percpu
, 1);
1272 for (i
= disks
; i
--; ) {
1273 struct r5dev
*dev
= &sh
->dev
[i
];
1274 if (test_and_clear_bit(R5_Overlap
, &dev
->flags
))
1275 wake_up(&sh
->raid_conf
->wait_for_overlap
);
1280 #ifdef CONFIG_MULTICORE_RAID456
1281 static void async_run_ops(void *param
, async_cookie_t cookie
)
1283 struct stripe_head
*sh
= param
;
1284 unsigned long ops_request
= sh
->ops
.request
;
1286 clear_bit_unlock(STRIPE_OPS_REQ_PENDING
, &sh
->state
);
1287 wake_up(&sh
->ops
.wait_for_ops
);
1289 __raid_run_ops(sh
, ops_request
);
1293 static void raid_run_ops(struct stripe_head
*sh
, unsigned long ops_request
)
1295 /* since handle_stripe can be called outside of raid5d context
1296 * we need to ensure sh->ops.request is de-staged before another
1299 wait_event(sh
->ops
.wait_for_ops
,
1300 !test_and_set_bit_lock(STRIPE_OPS_REQ_PENDING
, &sh
->state
));
1301 sh
->ops
.request
= ops_request
;
1303 atomic_inc(&sh
->count
);
1304 async_schedule(async_run_ops
, sh
);
1307 #define raid_run_ops __raid_run_ops
1310 static int grow_one_stripe(raid5_conf_t
*conf
)
1312 struct stripe_head
*sh
;
1313 sh
= kmem_cache_zalloc(conf
->slab_cache
, GFP_KERNEL
);
1317 sh
->raid_conf
= conf
;
1318 #ifdef CONFIG_MULTICORE_RAID456
1319 init_waitqueue_head(&sh
->ops
.wait_for_ops
);
1322 if (grow_buffers(sh
)) {
1324 kmem_cache_free(conf
->slab_cache
, sh
);
1327 /* we just created an active stripe so... */
1328 atomic_set(&sh
->count
, 1);
1329 atomic_inc(&conf
->active_stripes
);
1330 INIT_LIST_HEAD(&sh
->lru
);
1335 static int grow_stripes(raid5_conf_t
*conf
, int num
)
1337 struct kmem_cache
*sc
;
1338 int devs
= max(conf
->raid_disks
, conf
->previous_raid_disks
);
1340 if (conf
->mddev
->gendisk
)
1341 sprintf(conf
->cache_name
[0],
1342 "raid%d-%s", conf
->level
, mdname(conf
->mddev
));
1344 sprintf(conf
->cache_name
[0],
1345 "raid%d-%p", conf
->level
, conf
->mddev
);
1346 sprintf(conf
->cache_name
[1], "%s-alt", conf
->cache_name
[0]);
1348 conf
->active_name
= 0;
1349 sc
= kmem_cache_create(conf
->cache_name
[conf
->active_name
],
1350 sizeof(struct stripe_head
)+(devs
-1)*sizeof(struct r5dev
),
1354 conf
->slab_cache
= sc
;
1355 conf
->pool_size
= devs
;
1357 if (!grow_one_stripe(conf
))
1363 * scribble_len - return the required size of the scribble region
1364 * @num - total number of disks in the array
1366 * The size must be enough to contain:
1367 * 1/ a struct page pointer for each device in the array +2
1368 * 2/ room to convert each entry in (1) to its corresponding dma
1369 * (dma_map_page()) or page (page_address()) address.
1371 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
1372 * calculate over all devices (not just the data blocks), using zeros in place
1373 * of the P and Q blocks.
1375 static size_t scribble_len(int num
)
1379 len
= sizeof(struct page
*) * (num
+2) + sizeof(addr_conv_t
) * (num
+2);
1384 static int resize_stripes(raid5_conf_t
*conf
, int newsize
)
1386 /* Make all the stripes able to hold 'newsize' devices.
1387 * New slots in each stripe get 'page' set to a new page.
1389 * This happens in stages:
1390 * 1/ create a new kmem_cache and allocate the required number of
1392 * 2/ gather all the old stripe_heads and tranfer the pages across
1393 * to the new stripe_heads. This will have the side effect of
1394 * freezing the array as once all stripe_heads have been collected,
1395 * no IO will be possible. Old stripe heads are freed once their
1396 * pages have been transferred over, and the old kmem_cache is
1397 * freed when all stripes are done.
1398 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
1399 * we simple return a failre status - no need to clean anything up.
1400 * 4/ allocate new pages for the new slots in the new stripe_heads.
1401 * If this fails, we don't bother trying the shrink the
1402 * stripe_heads down again, we just leave them as they are.
1403 * As each stripe_head is processed the new one is released into
1406 * Once step2 is started, we cannot afford to wait for a write,
1407 * so we use GFP_NOIO allocations.
1409 struct stripe_head
*osh
, *nsh
;
1410 LIST_HEAD(newstripes
);
1411 struct disk_info
*ndisks
;
1414 struct kmem_cache
*sc
;
1417 if (newsize
<= conf
->pool_size
)
1418 return 0; /* never bother to shrink */
1420 err
= md_allow_write(conf
->mddev
);
1425 sc
= kmem_cache_create(conf
->cache_name
[1-conf
->active_name
],
1426 sizeof(struct stripe_head
)+(newsize
-1)*sizeof(struct r5dev
),
1431 for (i
= conf
->max_nr_stripes
; i
; i
--) {
1432 nsh
= kmem_cache_zalloc(sc
, GFP_KERNEL
);
1436 nsh
->raid_conf
= conf
;
1437 #ifdef CONFIG_MULTICORE_RAID456
1438 init_waitqueue_head(&nsh
->ops
.wait_for_ops
);
1441 list_add(&nsh
->lru
, &newstripes
);
1444 /* didn't get enough, give up */
1445 while (!list_empty(&newstripes
)) {
1446 nsh
= list_entry(newstripes
.next
, struct stripe_head
, lru
);
1447 list_del(&nsh
->lru
);
1448 kmem_cache_free(sc
, nsh
);
1450 kmem_cache_destroy(sc
);
1453 /* Step 2 - Must use GFP_NOIO now.
1454 * OK, we have enough stripes, start collecting inactive
1455 * stripes and copying them over
1457 list_for_each_entry(nsh
, &newstripes
, lru
) {
1458 spin_lock_irq(&conf
->device_lock
);
1459 wait_event_lock_irq(conf
->wait_for_stripe
,
1460 !list_empty(&conf
->inactive_list
),
1463 osh
= get_free_stripe(conf
);
1464 spin_unlock_irq(&conf
->device_lock
);
1465 atomic_set(&nsh
->count
, 1);
1466 for(i
=0; i
<conf
->pool_size
; i
++)
1467 nsh
->dev
[i
].page
= osh
->dev
[i
].page
;
1468 for( ; i
<newsize
; i
++)
1469 nsh
->dev
[i
].page
= NULL
;
1470 kmem_cache_free(conf
->slab_cache
, osh
);
1472 kmem_cache_destroy(conf
->slab_cache
);
1475 * At this point, we are holding all the stripes so the array
1476 * is completely stalled, so now is a good time to resize
1477 * conf->disks and the scribble region
1479 ndisks
= kzalloc(newsize
* sizeof(struct disk_info
), GFP_NOIO
);
1481 for (i
=0; i
<conf
->raid_disks
; i
++)
1482 ndisks
[i
] = conf
->disks
[i
];
1484 conf
->disks
= ndisks
;
1489 conf
->scribble_len
= scribble_len(newsize
);
1490 for_each_present_cpu(cpu
) {
1491 struct raid5_percpu
*percpu
;
1494 percpu
= per_cpu_ptr(conf
->percpu
, cpu
);
1495 scribble
= kmalloc(conf
->scribble_len
, GFP_NOIO
);
1498 kfree(percpu
->scribble
);
1499 percpu
->scribble
= scribble
;
1507 /* Step 4, return new stripes to service */
1508 while(!list_empty(&newstripes
)) {
1509 nsh
= list_entry(newstripes
.next
, struct stripe_head
, lru
);
1510 list_del_init(&nsh
->lru
);
1512 for (i
=conf
->raid_disks
; i
< newsize
; i
++)
1513 if (nsh
->dev
[i
].page
== NULL
) {
1514 struct page
*p
= alloc_page(GFP_NOIO
);
1515 nsh
->dev
[i
].page
= p
;
1519 release_stripe(nsh
);
1521 /* critical section pass, GFP_NOIO no longer needed */
1523 conf
->slab_cache
= sc
;
1524 conf
->active_name
= 1-conf
->active_name
;
1525 conf
->pool_size
= newsize
;
1529 static int drop_one_stripe(raid5_conf_t
*conf
)
1531 struct stripe_head
*sh
;
1533 spin_lock_irq(&conf
->device_lock
);
1534 sh
= get_free_stripe(conf
);
1535 spin_unlock_irq(&conf
->device_lock
);
1538 BUG_ON(atomic_read(&sh
->count
));
1540 kmem_cache_free(conf
->slab_cache
, sh
);
1541 atomic_dec(&conf
->active_stripes
);
1545 static void shrink_stripes(raid5_conf_t
*conf
)
1547 while (drop_one_stripe(conf
))
1550 if (conf
->slab_cache
)
1551 kmem_cache_destroy(conf
->slab_cache
);
1552 conf
->slab_cache
= NULL
;
1555 static void raid5_end_read_request(struct bio
* bi
, int error
)
1557 struct stripe_head
*sh
= bi
->bi_private
;
1558 raid5_conf_t
*conf
= sh
->raid_conf
;
1559 int disks
= sh
->disks
, i
;
1560 int uptodate
= test_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1561 char b
[BDEVNAME_SIZE
];
1565 for (i
=0 ; i
<disks
; i
++)
1566 if (bi
== &sh
->dev
[i
].req
)
1569 pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n",
1570 (unsigned long long)sh
->sector
, i
, atomic_read(&sh
->count
),
1578 set_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
);
1579 if (test_bit(R5_ReadError
, &sh
->dev
[i
].flags
)) {
1580 rdev
= conf
->disks
[i
].rdev
;
1583 "md/raid:%s: read error corrected"
1584 " (%lu sectors at %llu on %s)\n",
1585 mdname(conf
->mddev
), STRIPE_SECTORS
,
1586 (unsigned long long)(sh
->sector
1587 + rdev
->data_offset
),
1588 bdevname(rdev
->bdev
, b
));
1589 atomic_add(STRIPE_SECTORS
, &rdev
->corrected_errors
);
1590 clear_bit(R5_ReadError
, &sh
->dev
[i
].flags
);
1591 clear_bit(R5_ReWrite
, &sh
->dev
[i
].flags
);
1593 if (atomic_read(&conf
->disks
[i
].rdev
->read_errors
))
1594 atomic_set(&conf
->disks
[i
].rdev
->read_errors
, 0);
1596 const char *bdn
= bdevname(conf
->disks
[i
].rdev
->bdev
, b
);
1598 rdev
= conf
->disks
[i
].rdev
;
1600 clear_bit(R5_UPTODATE
, &sh
->dev
[i
].flags
);
1601 atomic_inc(&rdev
->read_errors
);
1602 if (conf
->mddev
->degraded
>= conf
->max_degraded
)
1605 "md/raid:%s: read error not correctable "
1606 "(sector %llu on %s).\n",
1607 mdname(conf
->mddev
),
1608 (unsigned long long)(sh
->sector
1609 + rdev
->data_offset
),
1611 else if (test_bit(R5_ReWrite
, &sh
->dev
[i
].flags
))
1615 "md/raid:%s: read error NOT corrected!! "
1616 "(sector %llu on %s).\n",
1617 mdname(conf
->mddev
),
1618 (unsigned long long)(sh
->sector
1619 + rdev
->data_offset
),
1621 else if (atomic_read(&rdev
->read_errors
)
1622 > conf
->max_nr_stripes
)
1624 "md/raid:%s: Too many read errors, failing device %s.\n",
1625 mdname(conf
->mddev
), bdn
);
1629 set_bit(R5_ReadError
, &sh
->dev
[i
].flags
);
1631 clear_bit(R5_ReadError
, &sh
->dev
[i
].flags
);
1632 clear_bit(R5_ReWrite
, &sh
->dev
[i
].flags
);
1633 md_error(conf
->mddev
, rdev
);
1636 rdev_dec_pending(conf
->disks
[i
].rdev
, conf
->mddev
);
1637 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
1638 set_bit(STRIPE_HANDLE
, &sh
->state
);
1642 static void raid5_end_write_request(struct bio
*bi
, int error
)
1644 struct stripe_head
*sh
= bi
->bi_private
;
1645 raid5_conf_t
*conf
= sh
->raid_conf
;
1646 int disks
= sh
->disks
, i
;
1647 int uptodate
= test_bit(BIO_UPTODATE
, &bi
->bi_flags
);
1649 for (i
=0 ; i
<disks
; i
++)
1650 if (bi
== &sh
->dev
[i
].req
)
1653 pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n",
1654 (unsigned long long)sh
->sector
, i
, atomic_read(&sh
->count
),
1662 md_error(conf
->mddev
, conf
->disks
[i
].rdev
);
1664 rdev_dec_pending(conf
->disks
[i
].rdev
, conf
->mddev
);
1666 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
1667 set_bit(STRIPE_HANDLE
, &sh
->state
);
1672 static sector_t
compute_blocknr(struct stripe_head
*sh
, int i
, int previous
);
1674 static void raid5_build_block(struct stripe_head
*sh
, int i
, int previous
)
1676 struct r5dev
*dev
= &sh
->dev
[i
];
1678 bio_init(&dev
->req
);
1679 dev
->req
.bi_io_vec
= &dev
->vec
;
1681 dev
->req
.bi_max_vecs
++;
1682 dev
->vec
.bv_page
= dev
->page
;
1683 dev
->vec
.bv_len
= STRIPE_SIZE
;
1684 dev
->vec
.bv_offset
= 0;
1686 dev
->req
.bi_sector
= sh
->sector
;
1687 dev
->req
.bi_private
= sh
;
1690 dev
->sector
= compute_blocknr(sh
, i
, previous
);
1693 static void error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1695 char b
[BDEVNAME_SIZE
];
1696 raid5_conf_t
*conf
= mddev
->private;
1697 pr_debug("raid456: error called\n");
1699 if (test_and_clear_bit(In_sync
, &rdev
->flags
)) {
1700 unsigned long flags
;
1701 spin_lock_irqsave(&conf
->device_lock
, flags
);
1703 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1705 * if recovery was running, make sure it aborts.
1707 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
1709 set_bit(Faulty
, &rdev
->flags
);
1710 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
1712 "md/raid:%s: Disk failure on %s, disabling device.\n"
1713 "md/raid:%s: Operation continuing on %d devices.\n",
1715 bdevname(rdev
->bdev
, b
),
1717 conf
->raid_disks
- mddev
->degraded
);
1721 * Input: a 'big' sector number,
1722 * Output: index of the data and parity disk, and the sector # in them.
1724 static sector_t
raid5_compute_sector(raid5_conf_t
*conf
, sector_t r_sector
,
1725 int previous
, int *dd_idx
,
1726 struct stripe_head
*sh
)
1728 sector_t stripe
, stripe2
;
1729 sector_t chunk_number
;
1730 unsigned int chunk_offset
;
1733 sector_t new_sector
;
1734 int algorithm
= previous
? conf
->prev_algo
1736 int sectors_per_chunk
= previous
? conf
->prev_chunk_sectors
1737 : conf
->chunk_sectors
;
1738 int raid_disks
= previous
? conf
->previous_raid_disks
1740 int data_disks
= raid_disks
- conf
->max_degraded
;
1742 /* First compute the information on this sector */
1745 * Compute the chunk number and the sector offset inside the chunk
1747 chunk_offset
= sector_div(r_sector
, sectors_per_chunk
);
1748 chunk_number
= r_sector
;
1751 * Compute the stripe number
1753 stripe
= chunk_number
;
1754 *dd_idx
= sector_div(stripe
, data_disks
);
1757 * Select the parity disk based on the user selected algorithm.
1759 pd_idx
= qd_idx
= -1;
1760 switch(conf
->level
) {
1762 pd_idx
= data_disks
;
1765 switch (algorithm
) {
1766 case ALGORITHM_LEFT_ASYMMETRIC
:
1767 pd_idx
= data_disks
- sector_div(stripe2
, raid_disks
);
1768 if (*dd_idx
>= pd_idx
)
1771 case ALGORITHM_RIGHT_ASYMMETRIC
:
1772 pd_idx
= sector_div(stripe2
, raid_disks
);
1773 if (*dd_idx
>= pd_idx
)
1776 case ALGORITHM_LEFT_SYMMETRIC
:
1777 pd_idx
= data_disks
- sector_div(stripe2
, raid_disks
);
1778 *dd_idx
= (pd_idx
+ 1 + *dd_idx
) % raid_disks
;
1780 case ALGORITHM_RIGHT_SYMMETRIC
:
1781 pd_idx
= sector_div(stripe2
, raid_disks
);
1782 *dd_idx
= (pd_idx
+ 1 + *dd_idx
) % raid_disks
;
1784 case ALGORITHM_PARITY_0
:
1788 case ALGORITHM_PARITY_N
:
1789 pd_idx
= data_disks
;
1797 switch (algorithm
) {
1798 case ALGORITHM_LEFT_ASYMMETRIC
:
1799 pd_idx
= raid_disks
- 1 - sector_div(stripe2
, raid_disks
);
1800 qd_idx
= pd_idx
+ 1;
1801 if (pd_idx
== raid_disks
-1) {
1802 (*dd_idx
)++; /* Q D D D P */
1804 } else if (*dd_idx
>= pd_idx
)
1805 (*dd_idx
) += 2; /* D D P Q D */
1807 case ALGORITHM_RIGHT_ASYMMETRIC
:
1808 pd_idx
= sector_div(stripe2
, raid_disks
);
1809 qd_idx
= pd_idx
+ 1;
1810 if (pd_idx
== raid_disks
-1) {
1811 (*dd_idx
)++; /* Q D D D P */
1813 } else if (*dd_idx
>= pd_idx
)
1814 (*dd_idx
) += 2; /* D D P Q D */
1816 case ALGORITHM_LEFT_SYMMETRIC
:
1817 pd_idx
= raid_disks
- 1 - sector_div(stripe2
, raid_disks
);
1818 qd_idx
= (pd_idx
+ 1) % raid_disks
;
1819 *dd_idx
= (pd_idx
+ 2 + *dd_idx
) % raid_disks
;
1821 case ALGORITHM_RIGHT_SYMMETRIC
:
1822 pd_idx
= sector_div(stripe2
, raid_disks
);
1823 qd_idx
= (pd_idx
+ 1) % raid_disks
;
1824 *dd_idx
= (pd_idx
+ 2 + *dd_idx
) % raid_disks
;
1827 case ALGORITHM_PARITY_0
:
1832 case ALGORITHM_PARITY_N
:
1833 pd_idx
= data_disks
;
1834 qd_idx
= data_disks
+ 1;
1837 case ALGORITHM_ROTATING_ZERO_RESTART
:
1838 /* Exactly the same as RIGHT_ASYMMETRIC, but or
1839 * of blocks for computing Q is different.
1841 pd_idx
= sector_div(stripe2
, raid_disks
);
1842 qd_idx
= pd_idx
+ 1;
1843 if (pd_idx
== raid_disks
-1) {
1844 (*dd_idx
)++; /* Q D D D P */
1846 } else if (*dd_idx
>= pd_idx
)
1847 (*dd_idx
) += 2; /* D D P Q D */
1851 case ALGORITHM_ROTATING_N_RESTART
:
1852 /* Same a left_asymmetric, by first stripe is
1853 * D D D P Q rather than
1857 pd_idx
= raid_disks
- 1 - sector_div(stripe2
, raid_disks
);
1858 qd_idx
= pd_idx
+ 1;
1859 if (pd_idx
== raid_disks
-1) {
1860 (*dd_idx
)++; /* Q D D D P */
1862 } else if (*dd_idx
>= pd_idx
)
1863 (*dd_idx
) += 2; /* D D P Q D */
1867 case ALGORITHM_ROTATING_N_CONTINUE
:
1868 /* Same as left_symmetric but Q is before P */
1869 pd_idx
= raid_disks
- 1 - sector_div(stripe2
, raid_disks
);
1870 qd_idx
= (pd_idx
+ raid_disks
- 1) % raid_disks
;
1871 *dd_idx
= (pd_idx
+ 1 + *dd_idx
) % raid_disks
;
1875 case ALGORITHM_LEFT_ASYMMETRIC_6
:
1876 /* RAID5 left_asymmetric, with Q on last device */
1877 pd_idx
= data_disks
- sector_div(stripe2
, raid_disks
-1);
1878 if (*dd_idx
>= pd_idx
)
1880 qd_idx
= raid_disks
- 1;
1883 case ALGORITHM_RIGHT_ASYMMETRIC_6
:
1884 pd_idx
= sector_div(stripe2
, raid_disks
-1);
1885 if (*dd_idx
>= pd_idx
)
1887 qd_idx
= raid_disks
- 1;
1890 case ALGORITHM_LEFT_SYMMETRIC_6
:
1891 pd_idx
= data_disks
- sector_div(stripe2
, raid_disks
-1);
1892 *dd_idx
= (pd_idx
+ 1 + *dd_idx
) % (raid_disks
-1);
1893 qd_idx
= raid_disks
- 1;
1896 case ALGORITHM_RIGHT_SYMMETRIC_6
:
1897 pd_idx
= sector_div(stripe2
, raid_disks
-1);
1898 *dd_idx
= (pd_idx
+ 1 + *dd_idx
) % (raid_disks
-1);
1899 qd_idx
= raid_disks
- 1;
1902 case ALGORITHM_PARITY_0_6
:
1905 qd_idx
= raid_disks
- 1;
1915 sh
->pd_idx
= pd_idx
;
1916 sh
->qd_idx
= qd_idx
;
1917 sh
->ddf_layout
= ddf_layout
;
1920 * Finally, compute the new sector number
1922 new_sector
= (sector_t
)stripe
* sectors_per_chunk
+ chunk_offset
;
1927 static sector_t
compute_blocknr(struct stripe_head
*sh
, int i
, int previous
)
1929 raid5_conf_t
*conf
= sh
->raid_conf
;
1930 int raid_disks
= sh
->disks
;
1931 int data_disks
= raid_disks
- conf
->max_degraded
;
1932 sector_t new_sector
= sh
->sector
, check
;
1933 int sectors_per_chunk
= previous
? conf
->prev_chunk_sectors
1934 : conf
->chunk_sectors
;
1935 int algorithm
= previous
? conf
->prev_algo
1939 sector_t chunk_number
;
1940 int dummy1
, dd_idx
= i
;
1942 struct stripe_head sh2
;
1945 chunk_offset
= sector_div(new_sector
, sectors_per_chunk
);
1946 stripe
= new_sector
;
1948 if (i
== sh
->pd_idx
)
1950 switch(conf
->level
) {
1953 switch (algorithm
) {
1954 case ALGORITHM_LEFT_ASYMMETRIC
:
1955 case ALGORITHM_RIGHT_ASYMMETRIC
:
1959 case ALGORITHM_LEFT_SYMMETRIC
:
1960 case ALGORITHM_RIGHT_SYMMETRIC
:
1963 i
-= (sh
->pd_idx
+ 1);
1965 case ALGORITHM_PARITY_0
:
1968 case ALGORITHM_PARITY_N
:
1975 if (i
== sh
->qd_idx
)
1976 return 0; /* It is the Q disk */
1977 switch (algorithm
) {
1978 case ALGORITHM_LEFT_ASYMMETRIC
:
1979 case ALGORITHM_RIGHT_ASYMMETRIC
:
1980 case ALGORITHM_ROTATING_ZERO_RESTART
:
1981 case ALGORITHM_ROTATING_N_RESTART
:
1982 if (sh
->pd_idx
== raid_disks
-1)
1983 i
--; /* Q D D D P */
1984 else if (i
> sh
->pd_idx
)
1985 i
-= 2; /* D D P Q D */
1987 case ALGORITHM_LEFT_SYMMETRIC
:
1988 case ALGORITHM_RIGHT_SYMMETRIC
:
1989 if (sh
->pd_idx
== raid_disks
-1)
1990 i
--; /* Q D D D P */
1995 i
-= (sh
->pd_idx
+ 2);
1998 case ALGORITHM_PARITY_0
:
2001 case ALGORITHM_PARITY_N
:
2003 case ALGORITHM_ROTATING_N_CONTINUE
:
2004 /* Like left_symmetric, but P is before Q */
2005 if (sh
->pd_idx
== 0)
2006 i
--; /* P D D D Q */
2011 i
-= (sh
->pd_idx
+ 1);
2014 case ALGORITHM_LEFT_ASYMMETRIC_6
:
2015 case ALGORITHM_RIGHT_ASYMMETRIC_6
:
2019 case ALGORITHM_LEFT_SYMMETRIC_6
:
2020 case ALGORITHM_RIGHT_SYMMETRIC_6
:
2022 i
+= data_disks
+ 1;
2023 i
-= (sh
->pd_idx
+ 1);
2025 case ALGORITHM_PARITY_0_6
:
2034 chunk_number
= stripe
* data_disks
+ i
;
2035 r_sector
= chunk_number
* sectors_per_chunk
+ chunk_offset
;
2037 check
= raid5_compute_sector(conf
, r_sector
,
2038 previous
, &dummy1
, &sh2
);
2039 if (check
!= sh
->sector
|| dummy1
!= dd_idx
|| sh2
.pd_idx
!= sh
->pd_idx
2040 || sh2
.qd_idx
!= sh
->qd_idx
) {
2041 printk(KERN_ERR
"md/raid:%s: compute_blocknr: map not correct\n",
2042 mdname(conf
->mddev
));
2050 schedule_reconstruction(struct stripe_head
*sh
, struct stripe_head_state
*s
,
2051 int rcw
, int expand
)
2053 int i
, pd_idx
= sh
->pd_idx
, disks
= sh
->disks
;
2054 raid5_conf_t
*conf
= sh
->raid_conf
;
2055 int level
= conf
->level
;
2058 /* if we are not expanding this is a proper write request, and
2059 * there will be bios with new data to be drained into the
2063 sh
->reconstruct_state
= reconstruct_state_drain_run
;
2064 set_bit(STRIPE_OP_BIODRAIN
, &s
->ops_request
);
2066 sh
->reconstruct_state
= reconstruct_state_run
;
2068 set_bit(STRIPE_OP_RECONSTRUCT
, &s
->ops_request
);
2070 for (i
= disks
; i
--; ) {
2071 struct r5dev
*dev
= &sh
->dev
[i
];
2074 set_bit(R5_LOCKED
, &dev
->flags
);
2075 set_bit(R5_Wantdrain
, &dev
->flags
);
2077 clear_bit(R5_UPTODATE
, &dev
->flags
);
2081 if (s
->locked
+ conf
->max_degraded
== disks
)
2082 if (!test_and_set_bit(STRIPE_FULL_WRITE
, &sh
->state
))
2083 atomic_inc(&conf
->pending_full_writes
);
2086 BUG_ON(!(test_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
) ||
2087 test_bit(R5_Wantcompute
, &sh
->dev
[pd_idx
].flags
)));
2089 sh
->reconstruct_state
= reconstruct_state_prexor_drain_run
;
2090 set_bit(STRIPE_OP_PREXOR
, &s
->ops_request
);
2091 set_bit(STRIPE_OP_BIODRAIN
, &s
->ops_request
);
2092 set_bit(STRIPE_OP_RECONSTRUCT
, &s
->ops_request
);
2094 for (i
= disks
; i
--; ) {
2095 struct r5dev
*dev
= &sh
->dev
[i
];
2100 (test_bit(R5_UPTODATE
, &dev
->flags
) ||
2101 test_bit(R5_Wantcompute
, &dev
->flags
))) {
2102 set_bit(R5_Wantdrain
, &dev
->flags
);
2103 set_bit(R5_LOCKED
, &dev
->flags
);
2104 clear_bit(R5_UPTODATE
, &dev
->flags
);
2110 /* keep the parity disk(s) locked while asynchronous operations
2113 set_bit(R5_LOCKED
, &sh
->dev
[pd_idx
].flags
);
2114 clear_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
2118 int qd_idx
= sh
->qd_idx
;
2119 struct r5dev
*dev
= &sh
->dev
[qd_idx
];
2121 set_bit(R5_LOCKED
, &dev
->flags
);
2122 clear_bit(R5_UPTODATE
, &dev
->flags
);
2126 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
2127 __func__
, (unsigned long long)sh
->sector
,
2128 s
->locked
, s
->ops_request
);
2132 * Each stripe/dev can have one or more bion attached.
2133 * toread/towrite point to the first in a chain.
2134 * The bi_next chain must be in order.
2136 static int add_stripe_bio(struct stripe_head
*sh
, struct bio
*bi
, int dd_idx
, int forwrite
)
2139 raid5_conf_t
*conf
= sh
->raid_conf
;
2142 pr_debug("adding bi b#%llu to stripe s#%llu\n",
2143 (unsigned long long)bi
->bi_sector
,
2144 (unsigned long long)sh
->sector
);
2147 spin_lock_irq(&conf
->device_lock
);
2149 bip
= &sh
->dev
[dd_idx
].towrite
;
2150 if (*bip
== NULL
&& sh
->dev
[dd_idx
].written
== NULL
)
2153 bip
= &sh
->dev
[dd_idx
].toread
;
2154 while (*bip
&& (*bip
)->bi_sector
< bi
->bi_sector
) {
2155 if ((*bip
)->bi_sector
+ ((*bip
)->bi_size
>> 9) > bi
->bi_sector
)
2157 bip
= & (*bip
)->bi_next
;
2159 if (*bip
&& (*bip
)->bi_sector
< bi
->bi_sector
+ ((bi
->bi_size
)>>9))
2162 BUG_ON(*bip
&& bi
->bi_next
&& (*bip
) != bi
->bi_next
);
2166 bi
->bi_phys_segments
++;
2169 /* check if page is covered */
2170 sector_t sector
= sh
->dev
[dd_idx
].sector
;
2171 for (bi
=sh
->dev
[dd_idx
].towrite
;
2172 sector
< sh
->dev
[dd_idx
].sector
+ STRIPE_SECTORS
&&
2173 bi
&& bi
->bi_sector
<= sector
;
2174 bi
= r5_next_bio(bi
, sh
->dev
[dd_idx
].sector
)) {
2175 if (bi
->bi_sector
+ (bi
->bi_size
>>9) >= sector
)
2176 sector
= bi
->bi_sector
+ (bi
->bi_size
>>9);
2178 if (sector
>= sh
->dev
[dd_idx
].sector
+ STRIPE_SECTORS
)
2179 set_bit(R5_OVERWRITE
, &sh
->dev
[dd_idx
].flags
);
2181 spin_unlock_irq(&conf
->device_lock
);
2183 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
2184 (unsigned long long)(*bip
)->bi_sector
,
2185 (unsigned long long)sh
->sector
, dd_idx
);
2187 if (conf
->mddev
->bitmap
&& firstwrite
) {
2188 bitmap_startwrite(conf
->mddev
->bitmap
, sh
->sector
,
2190 sh
->bm_seq
= conf
->seq_flush
+1;
2191 set_bit(STRIPE_BIT_DELAY
, &sh
->state
);
2196 set_bit(R5_Overlap
, &sh
->dev
[dd_idx
].flags
);
2197 spin_unlock_irq(&conf
->device_lock
);
2201 static void end_reshape(raid5_conf_t
*conf
);
2203 static void stripe_set_idx(sector_t stripe
, raid5_conf_t
*conf
, int previous
,
2204 struct stripe_head
*sh
)
2206 int sectors_per_chunk
=
2207 previous
? conf
->prev_chunk_sectors
: conf
->chunk_sectors
;
2209 int chunk_offset
= sector_div(stripe
, sectors_per_chunk
);
2210 int disks
= previous
? conf
->previous_raid_disks
: conf
->raid_disks
;
2212 raid5_compute_sector(conf
,
2213 stripe
* (disks
- conf
->max_degraded
)
2214 *sectors_per_chunk
+ chunk_offset
,
2220 handle_failed_stripe(raid5_conf_t
*conf
, struct stripe_head
*sh
,
2221 struct stripe_head_state
*s
, int disks
,
2222 struct bio
**return_bi
)
2225 for (i
= disks
; i
--; ) {
2229 if (test_bit(R5_ReadError
, &sh
->dev
[i
].flags
)) {
2232 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
2233 if (rdev
&& test_bit(In_sync
, &rdev
->flags
))
2234 /* multiple read failures in one stripe */
2235 md_error(conf
->mddev
, rdev
);
2238 spin_lock_irq(&conf
->device_lock
);
2239 /* fail all writes first */
2240 bi
= sh
->dev
[i
].towrite
;
2241 sh
->dev
[i
].towrite
= NULL
;
2247 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
2248 wake_up(&conf
->wait_for_overlap
);
2250 while (bi
&& bi
->bi_sector
<
2251 sh
->dev
[i
].sector
+ STRIPE_SECTORS
) {
2252 struct bio
*nextbi
= r5_next_bio(bi
, sh
->dev
[i
].sector
);
2253 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
2254 if (!raid5_dec_bi_phys_segments(bi
)) {
2255 md_write_end(conf
->mddev
);
2256 bi
->bi_next
= *return_bi
;
2261 /* and fail all 'written' */
2262 bi
= sh
->dev
[i
].written
;
2263 sh
->dev
[i
].written
= NULL
;
2264 if (bi
) bitmap_end
= 1;
2265 while (bi
&& bi
->bi_sector
<
2266 sh
->dev
[i
].sector
+ STRIPE_SECTORS
) {
2267 struct bio
*bi2
= r5_next_bio(bi
, sh
->dev
[i
].sector
);
2268 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
2269 if (!raid5_dec_bi_phys_segments(bi
)) {
2270 md_write_end(conf
->mddev
);
2271 bi
->bi_next
= *return_bi
;
2277 /* fail any reads if this device is non-operational and
2278 * the data has not reached the cache yet.
2280 if (!test_bit(R5_Wantfill
, &sh
->dev
[i
].flags
) &&
2281 (!test_bit(R5_Insync
, &sh
->dev
[i
].flags
) ||
2282 test_bit(R5_ReadError
, &sh
->dev
[i
].flags
))) {
2283 bi
= sh
->dev
[i
].toread
;
2284 sh
->dev
[i
].toread
= NULL
;
2285 if (test_and_clear_bit(R5_Overlap
, &sh
->dev
[i
].flags
))
2286 wake_up(&conf
->wait_for_overlap
);
2287 if (bi
) s
->to_read
--;
2288 while (bi
&& bi
->bi_sector
<
2289 sh
->dev
[i
].sector
+ STRIPE_SECTORS
) {
2290 struct bio
*nextbi
=
2291 r5_next_bio(bi
, sh
->dev
[i
].sector
);
2292 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
2293 if (!raid5_dec_bi_phys_segments(bi
)) {
2294 bi
->bi_next
= *return_bi
;
2300 spin_unlock_irq(&conf
->device_lock
);
2302 bitmap_endwrite(conf
->mddev
->bitmap
, sh
->sector
,
2303 STRIPE_SECTORS
, 0, 0);
2304 /* If we were in the middle of a write the parity block might
2305 * still be locked - so just clear all R5_LOCKED flags
2307 clear_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
2310 if (test_and_clear_bit(STRIPE_FULL_WRITE
, &sh
->state
))
2311 if (atomic_dec_and_test(&conf
->pending_full_writes
))
2312 md_wakeup_thread(conf
->mddev
->thread
);
2315 /* fetch_block - checks the given member device to see if its data needs
2316 * to be read or computed to satisfy a request.
2318 * Returns 1 when no more member devices need to be checked, otherwise returns
2319 * 0 to tell the loop in handle_stripe_fill to continue
2321 static int fetch_block(struct stripe_head
*sh
, struct stripe_head_state
*s
,
2322 int disk_idx
, int disks
)
2324 struct r5dev
*dev
= &sh
->dev
[disk_idx
];
2325 struct r5dev
*fdev
[2] = { &sh
->dev
[s
->failed_num
[0]],
2326 &sh
->dev
[s
->failed_num
[1]] };
2328 /* is the data in this block needed, and can we get it? */
2329 if (!test_bit(R5_LOCKED
, &dev
->flags
) &&
2330 !test_bit(R5_UPTODATE
, &dev
->flags
) &&
2332 (dev
->towrite
&& !test_bit(R5_OVERWRITE
, &dev
->flags
)) ||
2333 s
->syncing
|| s
->expanding
||
2334 (s
->failed
>= 1 && fdev
[0]->toread
) ||
2335 (s
->failed
>= 2 && fdev
[1]->toread
) ||
2336 (sh
->raid_conf
->level
<= 5 && s
->failed
&& fdev
[0]->towrite
&&
2337 !test_bit(R5_OVERWRITE
, &fdev
[0]->flags
)) ||
2338 (sh
->raid_conf
->level
== 6 && s
->failed
&& s
->to_write
))) {
2339 /* we would like to get this block, possibly by computing it,
2340 * otherwise read it if the backing disk is insync
2342 BUG_ON(test_bit(R5_Wantcompute
, &dev
->flags
));
2343 BUG_ON(test_bit(R5_Wantread
, &dev
->flags
));
2344 if ((s
->uptodate
== disks
- 1) &&
2345 (s
->failed
&& (disk_idx
== s
->failed_num
[0] ||
2346 disk_idx
== s
->failed_num
[1]))) {
2347 /* have disk failed, and we're requested to fetch it;
2350 pr_debug("Computing stripe %llu block %d\n",
2351 (unsigned long long)sh
->sector
, disk_idx
);
2352 set_bit(STRIPE_COMPUTE_RUN
, &sh
->state
);
2353 set_bit(STRIPE_OP_COMPUTE_BLK
, &s
->ops_request
);
2354 set_bit(R5_Wantcompute
, &dev
->flags
);
2355 sh
->ops
.target
= disk_idx
;
2356 sh
->ops
.target2
= -1; /* no 2nd target */
2358 /* Careful: from this point on 'uptodate' is in the eye
2359 * of raid_run_ops which services 'compute' operations
2360 * before writes. R5_Wantcompute flags a block that will
2361 * be R5_UPTODATE by the time it is needed for a
2362 * subsequent operation.
2366 } else if (s
->uptodate
== disks
-2 && s
->failed
>= 2) {
2367 /* Computing 2-failure is *very* expensive; only
2368 * do it if failed >= 2
2371 for (other
= disks
; other
--; ) {
2372 if (other
== disk_idx
)
2374 if (!test_bit(R5_UPTODATE
,
2375 &sh
->dev
[other
].flags
))
2379 pr_debug("Computing stripe %llu blocks %d,%d\n",
2380 (unsigned long long)sh
->sector
,
2382 set_bit(STRIPE_COMPUTE_RUN
, &sh
->state
);
2383 set_bit(STRIPE_OP_COMPUTE_BLK
, &s
->ops_request
);
2384 set_bit(R5_Wantcompute
, &sh
->dev
[disk_idx
].flags
);
2385 set_bit(R5_Wantcompute
, &sh
->dev
[other
].flags
);
2386 sh
->ops
.target
= disk_idx
;
2387 sh
->ops
.target2
= other
;
2391 } else if (test_bit(R5_Insync
, &dev
->flags
)) {
2392 set_bit(R5_LOCKED
, &dev
->flags
);
2393 set_bit(R5_Wantread
, &dev
->flags
);
2395 pr_debug("Reading block %d (sync=%d)\n",
2396 disk_idx
, s
->syncing
);
2404 * handle_stripe_fill - read or compute data to satisfy pending requests.
2406 static void handle_stripe_fill(struct stripe_head
*sh
,
2407 struct stripe_head_state
*s
,
2412 /* look for blocks to read/compute, skip this if a compute
2413 * is already in flight, or if the stripe contents are in the
2414 * midst of changing due to a write
2416 if (!test_bit(STRIPE_COMPUTE_RUN
, &sh
->state
) && !sh
->check_state
&&
2417 !sh
->reconstruct_state
)
2418 for (i
= disks
; i
--; )
2419 if (fetch_block(sh
, s
, i
, disks
))
2421 set_bit(STRIPE_HANDLE
, &sh
->state
);
2425 /* handle_stripe_clean_event
2426 * any written block on an uptodate or failed drive can be returned.
2427 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
2428 * never LOCKED, so we don't need to test 'failed' directly.
2430 static void handle_stripe_clean_event(raid5_conf_t
*conf
,
2431 struct stripe_head
*sh
, int disks
, struct bio
**return_bi
)
2436 for (i
= disks
; i
--; )
2437 if (sh
->dev
[i
].written
) {
2439 if (!test_bit(R5_LOCKED
, &dev
->flags
) &&
2440 test_bit(R5_UPTODATE
, &dev
->flags
)) {
2441 /* We can return any write requests */
2442 struct bio
*wbi
, *wbi2
;
2444 pr_debug("Return write for disc %d\n", i
);
2445 spin_lock_irq(&conf
->device_lock
);
2447 dev
->written
= NULL
;
2448 while (wbi
&& wbi
->bi_sector
<
2449 dev
->sector
+ STRIPE_SECTORS
) {
2450 wbi2
= r5_next_bio(wbi
, dev
->sector
);
2451 if (!raid5_dec_bi_phys_segments(wbi
)) {
2452 md_write_end(conf
->mddev
);
2453 wbi
->bi_next
= *return_bi
;
2458 if (dev
->towrite
== NULL
)
2460 spin_unlock_irq(&conf
->device_lock
);
2462 bitmap_endwrite(conf
->mddev
->bitmap
,
2465 !test_bit(STRIPE_DEGRADED
, &sh
->state
),
2470 if (test_and_clear_bit(STRIPE_FULL_WRITE
, &sh
->state
))
2471 if (atomic_dec_and_test(&conf
->pending_full_writes
))
2472 md_wakeup_thread(conf
->mddev
->thread
);
2475 static void handle_stripe_dirtying(raid5_conf_t
*conf
,
2476 struct stripe_head
*sh
,
2477 struct stripe_head_state
*s
,
2480 int rmw
= 0, rcw
= 0, i
;
2481 if (conf
->max_degraded
== 2) {
2482 /* RAID6 requires 'rcw' in current implementation
2483 * Calculate the real rcw later - for now fake it
2484 * look like rcw is cheaper
2487 } else for (i
= disks
; i
--; ) {
2488 /* would I have to read this buffer for read_modify_write */
2489 struct r5dev
*dev
= &sh
->dev
[i
];
2490 if ((dev
->towrite
|| i
== sh
->pd_idx
) &&
2491 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2492 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2493 test_bit(R5_Wantcompute
, &dev
->flags
))) {
2494 if (test_bit(R5_Insync
, &dev
->flags
))
2497 rmw
+= 2*disks
; /* cannot read it */
2499 /* Would I have to read this buffer for reconstruct_write */
2500 if (!test_bit(R5_OVERWRITE
, &dev
->flags
) && i
!= sh
->pd_idx
&&
2501 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2502 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2503 test_bit(R5_Wantcompute
, &dev
->flags
))) {
2504 if (test_bit(R5_Insync
, &dev
->flags
)) rcw
++;
2509 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
2510 (unsigned long long)sh
->sector
, rmw
, rcw
);
2511 set_bit(STRIPE_HANDLE
, &sh
->state
);
2512 if (rmw
< rcw
&& rmw
> 0)
2513 /* prefer read-modify-write, but need to get some data */
2514 for (i
= disks
; i
--; ) {
2515 struct r5dev
*dev
= &sh
->dev
[i
];
2516 if ((dev
->towrite
|| i
== sh
->pd_idx
) &&
2517 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2518 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2519 test_bit(R5_Wantcompute
, &dev
->flags
)) &&
2520 test_bit(R5_Insync
, &dev
->flags
)) {
2522 test_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2523 pr_debug("Read_old block "
2524 "%d for r-m-w\n", i
);
2525 set_bit(R5_LOCKED
, &dev
->flags
);
2526 set_bit(R5_Wantread
, &dev
->flags
);
2529 set_bit(STRIPE_DELAYED
, &sh
->state
);
2530 set_bit(STRIPE_HANDLE
, &sh
->state
);
2534 if (rcw
<= rmw
&& rcw
> 0) {
2535 /* want reconstruct write, but need to get some data */
2537 for (i
= disks
; i
--; ) {
2538 struct r5dev
*dev
= &sh
->dev
[i
];
2539 if (!test_bit(R5_OVERWRITE
, &dev
->flags
) &&
2540 i
!= sh
->pd_idx
&& i
!= sh
->qd_idx
&&
2541 !test_bit(R5_LOCKED
, &dev
->flags
) &&
2542 !(test_bit(R5_UPTODATE
, &dev
->flags
) ||
2543 test_bit(R5_Wantcompute
, &dev
->flags
))) {
2545 if (!test_bit(R5_Insync
, &dev
->flags
))
2546 continue; /* it's a failed drive */
2548 test_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
)) {
2549 pr_debug("Read_old block "
2550 "%d for Reconstruct\n", i
);
2551 set_bit(R5_LOCKED
, &dev
->flags
);
2552 set_bit(R5_Wantread
, &dev
->flags
);
2555 set_bit(STRIPE_DELAYED
, &sh
->state
);
2556 set_bit(STRIPE_HANDLE
, &sh
->state
);
2561 /* now if nothing is locked, and if we have enough data,
2562 * we can start a write request
2564 /* since handle_stripe can be called at any time we need to handle the
2565 * case where a compute block operation has been submitted and then a
2566 * subsequent call wants to start a write request. raid_run_ops only
2567 * handles the case where compute block and reconstruct are requested
2568 * simultaneously. If this is not the case then new writes need to be
2569 * held off until the compute completes.
2571 if ((s
->req_compute
|| !test_bit(STRIPE_COMPUTE_RUN
, &sh
->state
)) &&
2572 (s
->locked
== 0 && (rcw
== 0 || rmw
== 0) &&
2573 !test_bit(STRIPE_BIT_DELAY
, &sh
->state
)))
2574 schedule_reconstruction(sh
, s
, rcw
== 0, 0);
2577 static void handle_parity_checks5(raid5_conf_t
*conf
, struct stripe_head
*sh
,
2578 struct stripe_head_state
*s
, int disks
)
2580 struct r5dev
*dev
= NULL
;
2582 set_bit(STRIPE_HANDLE
, &sh
->state
);
2584 switch (sh
->check_state
) {
2585 case check_state_idle
:
2586 /* start a new check operation if there are no failures */
2587 if (s
->failed
== 0) {
2588 BUG_ON(s
->uptodate
!= disks
);
2589 sh
->check_state
= check_state_run
;
2590 set_bit(STRIPE_OP_CHECK
, &s
->ops_request
);
2591 clear_bit(R5_UPTODATE
, &sh
->dev
[sh
->pd_idx
].flags
);
2595 dev
= &sh
->dev
[s
->failed_num
[0]];
2597 case check_state_compute_result
:
2598 sh
->check_state
= check_state_idle
;
2600 dev
= &sh
->dev
[sh
->pd_idx
];
2602 /* check that a write has not made the stripe insync */
2603 if (test_bit(STRIPE_INSYNC
, &sh
->state
))
2606 /* either failed parity check, or recovery is happening */
2607 BUG_ON(!test_bit(R5_UPTODATE
, &dev
->flags
));
2608 BUG_ON(s
->uptodate
!= disks
);
2610 set_bit(R5_LOCKED
, &dev
->flags
);
2612 set_bit(R5_Wantwrite
, &dev
->flags
);
2614 clear_bit(STRIPE_DEGRADED
, &sh
->state
);
2615 set_bit(STRIPE_INSYNC
, &sh
->state
);
2617 case check_state_run
:
2618 break; /* we will be called again upon completion */
2619 case check_state_check_result
:
2620 sh
->check_state
= check_state_idle
;
2622 /* if a failure occurred during the check operation, leave
2623 * STRIPE_INSYNC not set and let the stripe be handled again
2628 /* handle a successful check operation, if parity is correct
2629 * we are done. Otherwise update the mismatch count and repair
2630 * parity if !MD_RECOVERY_CHECK
2632 if ((sh
->ops
.zero_sum_result
& SUM_CHECK_P_RESULT
) == 0)
2633 /* parity is correct (on disc,
2634 * not in buffer any more)
2636 set_bit(STRIPE_INSYNC
, &sh
->state
);
2638 conf
->mddev
->resync_mismatches
+= STRIPE_SECTORS
;
2639 if (test_bit(MD_RECOVERY_CHECK
, &conf
->mddev
->recovery
))
2640 /* don't try to repair!! */
2641 set_bit(STRIPE_INSYNC
, &sh
->state
);
2643 sh
->check_state
= check_state_compute_run
;
2644 set_bit(STRIPE_COMPUTE_RUN
, &sh
->state
);
2645 set_bit(STRIPE_OP_COMPUTE_BLK
, &s
->ops_request
);
2646 set_bit(R5_Wantcompute
,
2647 &sh
->dev
[sh
->pd_idx
].flags
);
2648 sh
->ops
.target
= sh
->pd_idx
;
2649 sh
->ops
.target2
= -1;
2654 case check_state_compute_run
:
2657 printk(KERN_ERR
"%s: unknown check_state: %d sector: %llu\n",
2658 __func__
, sh
->check_state
,
2659 (unsigned long long) sh
->sector
);
2665 static void handle_parity_checks6(raid5_conf_t
*conf
, struct stripe_head
*sh
,
2666 struct stripe_head_state
*s
,
2669 int pd_idx
= sh
->pd_idx
;
2670 int qd_idx
= sh
->qd_idx
;
2673 set_bit(STRIPE_HANDLE
, &sh
->state
);
2675 BUG_ON(s
->failed
> 2);
2677 /* Want to check and possibly repair P and Q.
2678 * However there could be one 'failed' device, in which
2679 * case we can only check one of them, possibly using the
2680 * other to generate missing data
2683 switch (sh
->check_state
) {
2684 case check_state_idle
:
2685 /* start a new check operation if there are < 2 failures */
2686 if (s
->failed
== s
->q_failed
) {
2687 /* The only possible failed device holds Q, so it
2688 * makes sense to check P (If anything else were failed,
2689 * we would have used P to recreate it).
2691 sh
->check_state
= check_state_run
;
2693 if (!s
->q_failed
&& s
->failed
< 2) {
2694 /* Q is not failed, and we didn't use it to generate
2695 * anything, so it makes sense to check it
2697 if (sh
->check_state
== check_state_run
)
2698 sh
->check_state
= check_state_run_pq
;
2700 sh
->check_state
= check_state_run_q
;
2703 /* discard potentially stale zero_sum_result */
2704 sh
->ops
.zero_sum_result
= 0;
2706 if (sh
->check_state
== check_state_run
) {
2707 /* async_xor_zero_sum destroys the contents of P */
2708 clear_bit(R5_UPTODATE
, &sh
->dev
[pd_idx
].flags
);
2711 if (sh
->check_state
>= check_state_run
&&
2712 sh
->check_state
<= check_state_run_pq
) {
2713 /* async_syndrome_zero_sum preserves P and Q, so
2714 * no need to mark them !uptodate here
2716 set_bit(STRIPE_OP_CHECK
, &s
->ops_request
);
2720 /* we have 2-disk failure */
2721 BUG_ON(s
->failed
!= 2);
2723 case check_state_compute_result
:
2724 sh
->check_state
= check_state_idle
;
2726 /* check that a write has not made the stripe insync */
2727 if (test_bit(STRIPE_INSYNC
, &sh
->state
))
2730 /* now write out any block on a failed drive,
2731 * or P or Q if they were recomputed
2733 BUG_ON(s
->uptodate
< disks
- 1); /* We don't need Q to recover */
2734 if (s
->failed
== 2) {
2735 dev
= &sh
->dev
[s
->failed_num
[1]];
2737 set_bit(R5_LOCKED
, &dev
->flags
);
2738 set_bit(R5_Wantwrite
, &dev
->flags
);
2740 if (s
->failed
>= 1) {
2741 dev
= &sh
->dev
[s
->failed_num
[0]];
2743 set_bit(R5_LOCKED
, &dev
->flags
);
2744 set_bit(R5_Wantwrite
, &dev
->flags
);
2746 if (sh
->ops
.zero_sum_result
& SUM_CHECK_P_RESULT
) {
2747 dev
= &sh
->dev
[pd_idx
];
2749 set_bit(R5_LOCKED
, &dev
->flags
);
2750 set_bit(R5_Wantwrite
, &dev
->flags
);
2752 if (sh
->ops
.zero_sum_result
& SUM_CHECK_Q_RESULT
) {
2753 dev
= &sh
->dev
[qd_idx
];
2755 set_bit(R5_LOCKED
, &dev
->flags
);
2756 set_bit(R5_Wantwrite
, &dev
->flags
);
2758 clear_bit(STRIPE_DEGRADED
, &sh
->state
);
2760 set_bit(STRIPE_INSYNC
, &sh
->state
);
2762 case check_state_run
:
2763 case check_state_run_q
:
2764 case check_state_run_pq
:
2765 break; /* we will be called again upon completion */
2766 case check_state_check_result
:
2767 sh
->check_state
= check_state_idle
;
2769 /* handle a successful check operation, if parity is correct
2770 * we are done. Otherwise update the mismatch count and repair
2771 * parity if !MD_RECOVERY_CHECK
2773 if (sh
->ops
.zero_sum_result
== 0) {
2774 /* both parities are correct */
2776 set_bit(STRIPE_INSYNC
, &sh
->state
);
2778 /* in contrast to the raid5 case we can validate
2779 * parity, but still have a failure to write
2782 sh
->check_state
= check_state_compute_result
;
2783 /* Returning at this point means that we may go
2784 * off and bring p and/or q uptodate again so
2785 * we make sure to check zero_sum_result again
2786 * to verify if p or q need writeback
2790 conf
->mddev
->resync_mismatches
+= STRIPE_SECTORS
;
2791 if (test_bit(MD_RECOVERY_CHECK
, &conf
->mddev
->recovery
))
2792 /* don't try to repair!! */
2793 set_bit(STRIPE_INSYNC
, &sh
->state
);
2795 int *target
= &sh
->ops
.target
;
2797 sh
->ops
.target
= -1;
2798 sh
->ops
.target2
= -1;
2799 sh
->check_state
= check_state_compute_run
;
2800 set_bit(STRIPE_COMPUTE_RUN
, &sh
->state
);
2801 set_bit(STRIPE_OP_COMPUTE_BLK
, &s
->ops_request
);
2802 if (sh
->ops
.zero_sum_result
& SUM_CHECK_P_RESULT
) {
2803 set_bit(R5_Wantcompute
,
2804 &sh
->dev
[pd_idx
].flags
);
2806 target
= &sh
->ops
.target2
;
2809 if (sh
->ops
.zero_sum_result
& SUM_CHECK_Q_RESULT
) {
2810 set_bit(R5_Wantcompute
,
2811 &sh
->dev
[qd_idx
].flags
);
2818 case check_state_compute_run
:
2821 printk(KERN_ERR
"%s: unknown check_state: %d sector: %llu\n",
2822 __func__
, sh
->check_state
,
2823 (unsigned long long) sh
->sector
);
2828 static void handle_stripe_expansion(raid5_conf_t
*conf
, struct stripe_head
*sh
)
2832 /* We have read all the blocks in this stripe and now we need to
2833 * copy some of them into a target stripe for expand.
2835 struct dma_async_tx_descriptor
*tx
= NULL
;
2836 clear_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
2837 for (i
= 0; i
< sh
->disks
; i
++)
2838 if (i
!= sh
->pd_idx
&& i
!= sh
->qd_idx
) {
2840 struct stripe_head
*sh2
;
2841 struct async_submit_ctl submit
;
2843 sector_t bn
= compute_blocknr(sh
, i
, 1);
2844 sector_t s
= raid5_compute_sector(conf
, bn
, 0,
2846 sh2
= get_active_stripe(conf
, s
, 0, 1, 1);
2848 /* so far only the early blocks of this stripe
2849 * have been requested. When later blocks
2850 * get requested, we will try again
2853 if (!test_bit(STRIPE_EXPANDING
, &sh2
->state
) ||
2854 test_bit(R5_Expanded
, &sh2
->dev
[dd_idx
].flags
)) {
2855 /* must have already done this block */
2856 release_stripe(sh2
);
2860 /* place all the copies on one channel */
2861 init_async_submit(&submit
, 0, tx
, NULL
, NULL
, NULL
);
2862 tx
= async_memcpy(sh2
->dev
[dd_idx
].page
,
2863 sh
->dev
[i
].page
, 0, 0, STRIPE_SIZE
,
2866 set_bit(R5_Expanded
, &sh2
->dev
[dd_idx
].flags
);
2867 set_bit(R5_UPTODATE
, &sh2
->dev
[dd_idx
].flags
);
2868 for (j
= 0; j
< conf
->raid_disks
; j
++)
2869 if (j
!= sh2
->pd_idx
&&
2871 !test_bit(R5_Expanded
, &sh2
->dev
[j
].flags
))
2873 if (j
== conf
->raid_disks
) {
2874 set_bit(STRIPE_EXPAND_READY
, &sh2
->state
);
2875 set_bit(STRIPE_HANDLE
, &sh2
->state
);
2877 release_stripe(sh2
);
2880 /* done submitting copies, wait for them to complete */
2883 dma_wait_for_async_tx(tx
);
2889 * handle_stripe - do things to a stripe.
2891 * We lock the stripe and then examine the state of various bits
2892 * to see what needs to be done.
2894 * return some read request which now have data
2895 * return some write requests which are safely on disc
2896 * schedule a read on some buffers
2897 * schedule a write of some buffers
2898 * return confirmation of parity correctness
2900 * buffers are taken off read_list or write_list, and bh_cache buffers
2901 * get BH_Lock set before the stripe lock is released.
2905 static void analyse_stripe(struct stripe_head
*sh
, struct stripe_head_state
*s
)
2907 raid5_conf_t
*conf
= sh
->raid_conf
;
2908 int disks
= sh
->disks
;
2912 memset(s
, 0, sizeof(*s
));
2914 s
->syncing
= test_bit(STRIPE_SYNCING
, &sh
->state
);
2915 s
->expanding
= test_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
2916 s
->expanded
= test_bit(STRIPE_EXPAND_READY
, &sh
->state
);
2917 s
->failed_num
[0] = -1;
2918 s
->failed_num
[1] = -1;
2920 /* Now to look around and see what can be done */
2922 spin_lock_irq(&conf
->device_lock
);
2923 for (i
=disks
; i
--; ) {
2928 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
2929 i
, dev
->flags
, dev
->toread
, dev
->towrite
, dev
->written
);
2930 /* maybe we can reply to a read
2932 * new wantfill requests are only permitted while
2933 * ops_complete_biofill is guaranteed to be inactive
2935 if (test_bit(R5_UPTODATE
, &dev
->flags
) && dev
->toread
&&
2936 !test_bit(STRIPE_BIOFILL_RUN
, &sh
->state
))
2937 set_bit(R5_Wantfill
, &dev
->flags
);
2939 /* now count some things */
2940 if (test_bit(R5_LOCKED
, &dev
->flags
))
2942 if (test_bit(R5_UPTODATE
, &dev
->flags
))
2944 if (test_bit(R5_Wantcompute
, &dev
->flags
)) {
2946 BUG_ON(s
->compute
> 2);
2949 if (test_bit(R5_Wantfill
, &dev
->flags
))
2951 else if (dev
->toread
)
2955 if (!test_bit(R5_OVERWRITE
, &dev
->flags
))
2960 rdev
= rcu_dereference(conf
->disks
[i
].rdev
);
2961 if (s
->blocked_rdev
== NULL
&&
2962 rdev
&& unlikely(test_bit(Blocked
, &rdev
->flags
))) {
2963 s
->blocked_rdev
= rdev
;
2964 atomic_inc(&rdev
->nr_pending
);
2966 clear_bit(R5_Insync
, &dev
->flags
);
2969 else if (test_bit(In_sync
, &rdev
->flags
))
2970 set_bit(R5_Insync
, &dev
->flags
);
2972 /* in sync if before recovery_offset */
2973 if (sh
->sector
+ STRIPE_SECTORS
<= rdev
->recovery_offset
)
2974 set_bit(R5_Insync
, &dev
->flags
);
2976 if (!test_bit(R5_Insync
, &dev
->flags
)) {
2977 /* The ReadError flag will just be confusing now */
2978 clear_bit(R5_ReadError
, &dev
->flags
);
2979 clear_bit(R5_ReWrite
, &dev
->flags
);
2981 if (test_bit(R5_ReadError
, &dev
->flags
))
2982 clear_bit(R5_Insync
, &dev
->flags
);
2983 if (!test_bit(R5_Insync
, &dev
->flags
)) {
2985 s
->failed_num
[s
->failed
] = i
;
2989 spin_unlock_irq(&conf
->device_lock
);
2993 static void handle_stripe(struct stripe_head
*sh
)
2995 struct stripe_head_state s
;
2996 raid5_conf_t
*conf
= sh
->raid_conf
;
2999 int disks
= sh
->disks
;
3000 struct r5dev
*pdev
, *qdev
;
3002 clear_bit(STRIPE_HANDLE
, &sh
->state
);
3003 if (test_and_set_bit(STRIPE_ACTIVE
, &sh
->state
)) {
3004 /* already being handled, ensure it gets handled
3005 * again when current action finishes */
3006 set_bit(STRIPE_HANDLE
, &sh
->state
);
3010 if (test_and_clear_bit(STRIPE_SYNC_REQUESTED
, &sh
->state
)) {
3011 set_bit(STRIPE_SYNCING
, &sh
->state
);
3012 clear_bit(STRIPE_INSYNC
, &sh
->state
);
3014 clear_bit(STRIPE_DELAYED
, &sh
->state
);
3016 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
3017 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
3018 (unsigned long long)sh
->sector
, sh
->state
,
3019 atomic_read(&sh
->count
), sh
->pd_idx
, sh
->qd_idx
,
3020 sh
->check_state
, sh
->reconstruct_state
);
3022 analyse_stripe(sh
, &s
);
3024 if (unlikely(s
.blocked_rdev
)) {
3025 if (s
.syncing
|| s
.expanding
|| s
.expanded
||
3026 s
.to_write
|| s
.written
) {
3027 set_bit(STRIPE_HANDLE
, &sh
->state
);
3030 /* There is nothing for the blocked_rdev to block */
3031 rdev_dec_pending(s
.blocked_rdev
, conf
->mddev
);
3032 s
.blocked_rdev
= NULL
;
3035 if (s
.to_fill
&& !test_bit(STRIPE_BIOFILL_RUN
, &sh
->state
)) {
3036 set_bit(STRIPE_OP_BIOFILL
, &s
.ops_request
);
3037 set_bit(STRIPE_BIOFILL_RUN
, &sh
->state
);
3040 pr_debug("locked=%d uptodate=%d to_read=%d"
3041 " to_write=%d failed=%d failed_num=%d,%d\n",
3042 s
.locked
, s
.uptodate
, s
.to_read
, s
.to_write
, s
.failed
,
3043 s
.failed_num
[0], s
.failed_num
[1]);
3044 /* check if the array has lost more than max_degraded devices and,
3045 * if so, some requests might need to be failed.
3047 if (s
.failed
> conf
->max_degraded
&& s
.to_read
+s
.to_write
+s
.written
)
3048 handle_failed_stripe(conf
, sh
, &s
, disks
, &s
.return_bi
);
3049 if (s
.failed
> conf
->max_degraded
&& s
.syncing
) {
3050 md_done_sync(conf
->mddev
, STRIPE_SECTORS
, 0);
3051 clear_bit(STRIPE_SYNCING
, &sh
->state
);
3056 * might be able to return some write requests if the parity blocks
3057 * are safe, or on a failed drive
3059 pdev
= &sh
->dev
[sh
->pd_idx
];
3060 s
.p_failed
= (s
.failed
>= 1 && s
.failed_num
[0] == sh
->pd_idx
)
3061 || (s
.failed
>= 2 && s
.failed_num
[1] == sh
->pd_idx
);
3062 qdev
= &sh
->dev
[sh
->qd_idx
];
3063 s
.q_failed
= (s
.failed
>= 1 && s
.failed_num
[0] == sh
->qd_idx
)
3064 || (s
.failed
>= 2 && s
.failed_num
[1] == sh
->qd_idx
)
3068 (s
.p_failed
|| ((test_bit(R5_Insync
, &pdev
->flags
)
3069 && !test_bit(R5_LOCKED
, &pdev
->flags
)
3070 && test_bit(R5_UPTODATE
, &pdev
->flags
)))) &&
3071 (s
.q_failed
|| ((test_bit(R5_Insync
, &qdev
->flags
)
3072 && !test_bit(R5_LOCKED
, &qdev
->flags
)
3073 && test_bit(R5_UPTODATE
, &qdev
->flags
)))))
3074 handle_stripe_clean_event(conf
, sh
, disks
, &s
.return_bi
);
3076 /* Now we might consider reading some blocks, either to check/generate
3077 * parity, or to satisfy requests
3078 * or to load a block that is being partially written.
3080 if (s
.to_read
|| s
.non_overwrite
3081 || (conf
->level
== 6 && s
.to_write
&& s
.failed
)
3082 || (s
.syncing
&& (s
.uptodate
+ s
.compute
< disks
)) || s
.expanding
)
3083 handle_stripe_fill(sh
, &s
, disks
);
3085 /* Now we check to see if any write operations have recently
3089 if (sh
->reconstruct_state
== reconstruct_state_prexor_drain_result
)
3091 if (sh
->reconstruct_state
== reconstruct_state_drain_result
||
3092 sh
->reconstruct_state
== reconstruct_state_prexor_drain_result
) {
3093 sh
->reconstruct_state
= reconstruct_state_idle
;
3095 /* All the 'written' buffers and the parity block are ready to
3096 * be written back to disk
3098 BUG_ON(!test_bit(R5_UPTODATE
, &sh
->dev
[sh
->pd_idx
].flags
));
3099 BUG_ON(sh
->qd_idx
>= 0 &&
3100 !test_bit(R5_UPTODATE
, &sh
->dev
[sh
->qd_idx
].flags
));
3101 for (i
= disks
; i
--; ) {
3102 struct r5dev
*dev
= &sh
->dev
[i
];
3103 if (test_bit(R5_LOCKED
, &dev
->flags
) &&
3104 (i
== sh
->pd_idx
|| i
== sh
->qd_idx
||
3106 pr_debug("Writing block %d\n", i
);
3107 set_bit(R5_Wantwrite
, &dev
->flags
);
3110 if (!test_bit(R5_Insync
, &dev
->flags
) ||
3111 ((i
== sh
->pd_idx
|| i
== sh
->qd_idx
) &&
3113 set_bit(STRIPE_INSYNC
, &sh
->state
);
3116 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
))
3117 s
.dec_preread_active
= 1;
3120 /* Now to consider new write requests and what else, if anything
3121 * should be read. We do not handle new writes when:
3122 * 1/ A 'write' operation (copy+xor) is already in flight.
3123 * 2/ A 'check' operation is in flight, as it may clobber the parity
3126 if (s
.to_write
&& !sh
->reconstruct_state
&& !sh
->check_state
)
3127 handle_stripe_dirtying(conf
, sh
, &s
, disks
);
3129 /* maybe we need to check and possibly fix the parity for this stripe
3130 * Any reads will already have been scheduled, so we just see if enough
3131 * data is available. The parity check is held off while parity
3132 * dependent operations are in flight.
3134 if (sh
->check_state
||
3135 (s
.syncing
&& s
.locked
== 0 &&
3136 !test_bit(STRIPE_COMPUTE_RUN
, &sh
->state
) &&
3137 !test_bit(STRIPE_INSYNC
, &sh
->state
))) {
3138 if (conf
->level
== 6)
3139 handle_parity_checks6(conf
, sh
, &s
, disks
);
3141 handle_parity_checks5(conf
, sh
, &s
, disks
);
3144 if (s
.syncing
&& s
.locked
== 0 && test_bit(STRIPE_INSYNC
, &sh
->state
)) {
3145 md_done_sync(conf
->mddev
, STRIPE_SECTORS
, 1);
3146 clear_bit(STRIPE_SYNCING
, &sh
->state
);
3149 /* If the failed drives are just a ReadError, then we might need
3150 * to progress the repair/check process
3152 if (s
.failed
<= conf
->max_degraded
&& !conf
->mddev
->ro
)
3153 for (i
= 0; i
< s
.failed
; i
++) {
3154 struct r5dev
*dev
= &sh
->dev
[s
.failed_num
[i
]];
3155 if (test_bit(R5_ReadError
, &dev
->flags
)
3156 && !test_bit(R5_LOCKED
, &dev
->flags
)
3157 && test_bit(R5_UPTODATE
, &dev
->flags
)
3159 if (!test_bit(R5_ReWrite
, &dev
->flags
)) {
3160 set_bit(R5_Wantwrite
, &dev
->flags
);
3161 set_bit(R5_ReWrite
, &dev
->flags
);
3162 set_bit(R5_LOCKED
, &dev
->flags
);
3165 /* let's read it back */
3166 set_bit(R5_Wantread
, &dev
->flags
);
3167 set_bit(R5_LOCKED
, &dev
->flags
);
3174 /* Finish reconstruct operations initiated by the expansion process */
3175 if (sh
->reconstruct_state
== reconstruct_state_result
) {
3176 struct stripe_head
*sh_src
3177 = get_active_stripe(conf
, sh
->sector
, 1, 1, 1);
3178 if (sh_src
&& test_bit(STRIPE_EXPAND_SOURCE
, &sh_src
->state
)) {
3179 /* sh cannot be written until sh_src has been read.
3180 * so arrange for sh to be delayed a little
3182 set_bit(STRIPE_DELAYED
, &sh
->state
);
3183 set_bit(STRIPE_HANDLE
, &sh
->state
);
3184 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE
,
3186 atomic_inc(&conf
->preread_active_stripes
);
3187 release_stripe(sh_src
);
3191 release_stripe(sh_src
);
3193 sh
->reconstruct_state
= reconstruct_state_idle
;
3194 clear_bit(STRIPE_EXPANDING
, &sh
->state
);
3195 for (i
= conf
->raid_disks
; i
--; ) {
3196 set_bit(R5_Wantwrite
, &sh
->dev
[i
].flags
);
3197 set_bit(R5_LOCKED
, &sh
->dev
[i
].flags
);
3202 if (s
.expanded
&& test_bit(STRIPE_EXPANDING
, &sh
->state
) &&
3203 !sh
->reconstruct_state
) {
3204 /* Need to write out all blocks after computing parity */
3205 sh
->disks
= conf
->raid_disks
;
3206 stripe_set_idx(sh
->sector
, conf
, 0, sh
);
3207 schedule_reconstruction(sh
, &s
, 1, 1);
3208 } else if (s
.expanded
&& !sh
->reconstruct_state
&& s
.locked
== 0) {
3209 clear_bit(STRIPE_EXPAND_READY
, &sh
->state
);
3210 atomic_dec(&conf
->reshape_stripes
);
3211 wake_up(&conf
->wait_for_overlap
);
3212 md_done_sync(conf
->mddev
, STRIPE_SECTORS
, 1);
3215 if (s
.expanding
&& s
.locked
== 0 &&
3216 !test_bit(STRIPE_COMPUTE_RUN
, &sh
->state
))
3217 handle_stripe_expansion(conf
, sh
);
3220 /* wait for this device to become unblocked */
3221 if (unlikely(s
.blocked_rdev
))
3222 md_wait_for_blocked_rdev(s
.blocked_rdev
, conf
->mddev
);
3225 raid_run_ops(sh
, s
.ops_request
);
3230 if (s
.dec_preread_active
) {
3231 /* We delay this until after ops_run_io so that if make_request
3232 * is waiting on a flush, it won't continue until the writes
3233 * have actually been submitted.
3235 atomic_dec(&conf
->preread_active_stripes
);
3236 if (atomic_read(&conf
->preread_active_stripes
) <
3238 md_wakeup_thread(conf
->mddev
->thread
);
3241 return_io(s
.return_bi
);
3243 clear_bit(STRIPE_ACTIVE
, &sh
->state
);
3246 static void raid5_activate_delayed(raid5_conf_t
*conf
)
3248 if (atomic_read(&conf
->preread_active_stripes
) < IO_THRESHOLD
) {
3249 while (!list_empty(&conf
->delayed_list
)) {
3250 struct list_head
*l
= conf
->delayed_list
.next
;
3251 struct stripe_head
*sh
;
3252 sh
= list_entry(l
, struct stripe_head
, lru
);
3254 clear_bit(STRIPE_DELAYED
, &sh
->state
);
3255 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
))
3256 atomic_inc(&conf
->preread_active_stripes
);
3257 list_add_tail(&sh
->lru
, &conf
->hold_list
);
3262 static void activate_bit_delay(raid5_conf_t
*conf
)
3264 /* device_lock is held */
3265 struct list_head head
;
3266 list_add(&head
, &conf
->bitmap_list
);
3267 list_del_init(&conf
->bitmap_list
);
3268 while (!list_empty(&head
)) {
3269 struct stripe_head
*sh
= list_entry(head
.next
, struct stripe_head
, lru
);
3270 list_del_init(&sh
->lru
);
3271 atomic_inc(&sh
->count
);
3272 __release_stripe(conf
, sh
);
3276 int md_raid5_congested(mddev_t
*mddev
, int bits
)
3278 raid5_conf_t
*conf
= mddev
->private;
3280 /* No difference between reads and writes. Just check
3281 * how busy the stripe_cache is
3284 if (conf
->inactive_blocked
)
3288 if (list_empty_careful(&conf
->inactive_list
))
3293 EXPORT_SYMBOL_GPL(md_raid5_congested
);
3295 static int raid5_congested(void *data
, int bits
)
3297 mddev_t
*mddev
= data
;
3299 return mddev_congested(mddev
, bits
) ||
3300 md_raid5_congested(mddev
, bits
);
3303 /* We want read requests to align with chunks where possible,
3304 * but write requests don't need to.
3306 static int raid5_mergeable_bvec(struct request_queue
*q
,
3307 struct bvec_merge_data
*bvm
,
3308 struct bio_vec
*biovec
)
3310 mddev_t
*mddev
= q
->queuedata
;
3311 sector_t sector
= bvm
->bi_sector
+ get_start_sect(bvm
->bi_bdev
);
3313 unsigned int chunk_sectors
= mddev
->chunk_sectors
;
3314 unsigned int bio_sectors
= bvm
->bi_size
>> 9;
3316 if ((bvm
->bi_rw
& 1) == WRITE
)
3317 return biovec
->bv_len
; /* always allow writes to be mergeable */
3319 if (mddev
->new_chunk_sectors
< mddev
->chunk_sectors
)
3320 chunk_sectors
= mddev
->new_chunk_sectors
;
3321 max
= (chunk_sectors
- ((sector
& (chunk_sectors
- 1)) + bio_sectors
)) << 9;
3322 if (max
< 0) max
= 0;
3323 if (max
<= biovec
->bv_len
&& bio_sectors
== 0)
3324 return biovec
->bv_len
;
3330 static int in_chunk_boundary(mddev_t
*mddev
, struct bio
*bio
)
3332 sector_t sector
= bio
->bi_sector
+ get_start_sect(bio
->bi_bdev
);
3333 unsigned int chunk_sectors
= mddev
->chunk_sectors
;
3334 unsigned int bio_sectors
= bio
->bi_size
>> 9;
3336 if (mddev
->new_chunk_sectors
< mddev
->chunk_sectors
)
3337 chunk_sectors
= mddev
->new_chunk_sectors
;
3338 return chunk_sectors
>=
3339 ((sector
& (chunk_sectors
- 1)) + bio_sectors
);
3343 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
3344 * later sampled by raid5d.
3346 static void add_bio_to_retry(struct bio
*bi
,raid5_conf_t
*conf
)
3348 unsigned long flags
;
3350 spin_lock_irqsave(&conf
->device_lock
, flags
);
3352 bi
->bi_next
= conf
->retry_read_aligned_list
;
3353 conf
->retry_read_aligned_list
= bi
;
3355 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
3356 md_wakeup_thread(conf
->mddev
->thread
);
3360 static struct bio
*remove_bio_from_retry(raid5_conf_t
*conf
)
3364 bi
= conf
->retry_read_aligned
;
3366 conf
->retry_read_aligned
= NULL
;
3369 bi
= conf
->retry_read_aligned_list
;
3371 conf
->retry_read_aligned_list
= bi
->bi_next
;
3374 * this sets the active strip count to 1 and the processed
3375 * strip count to zero (upper 8 bits)
3377 bi
->bi_phys_segments
= 1; /* biased count of active stripes */
3385 * The "raid5_align_endio" should check if the read succeeded and if it
3386 * did, call bio_endio on the original bio (having bio_put the new bio
3388 * If the read failed..
3390 static void raid5_align_endio(struct bio
*bi
, int error
)
3392 struct bio
* raid_bi
= bi
->bi_private
;
3395 int uptodate
= test_bit(BIO_UPTODATE
, &bi
->bi_flags
);
3400 rdev
= (void*)raid_bi
->bi_next
;
3401 raid_bi
->bi_next
= NULL
;
3402 mddev
= rdev
->mddev
;
3403 conf
= mddev
->private;
3405 rdev_dec_pending(rdev
, conf
->mddev
);
3407 if (!error
&& uptodate
) {
3408 bio_endio(raid_bi
, 0);
3409 if (atomic_dec_and_test(&conf
->active_aligned_reads
))
3410 wake_up(&conf
->wait_for_stripe
);
3415 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
3417 add_bio_to_retry(raid_bi
, conf
);
3420 static int bio_fits_rdev(struct bio
*bi
)
3422 struct request_queue
*q
= bdev_get_queue(bi
->bi_bdev
);
3424 if ((bi
->bi_size
>>9) > queue_max_sectors(q
))
3426 blk_recount_segments(q
, bi
);
3427 if (bi
->bi_phys_segments
> queue_max_segments(q
))
3430 if (q
->merge_bvec_fn
)
3431 /* it's too hard to apply the merge_bvec_fn at this stage,
3440 static int chunk_aligned_read(mddev_t
*mddev
, struct bio
* raid_bio
)
3442 raid5_conf_t
*conf
= mddev
->private;
3444 struct bio
* align_bi
;
3447 if (!in_chunk_boundary(mddev
, raid_bio
)) {
3448 pr_debug("chunk_aligned_read : non aligned\n");
3452 * use bio_clone_mddev to make a copy of the bio
3454 align_bi
= bio_clone_mddev(raid_bio
, GFP_NOIO
, mddev
);
3458 * set bi_end_io to a new function, and set bi_private to the
3461 align_bi
->bi_end_io
= raid5_align_endio
;
3462 align_bi
->bi_private
= raid_bio
;
3466 align_bi
->bi_sector
= raid5_compute_sector(conf
, raid_bio
->bi_sector
,
3471 rdev
= rcu_dereference(conf
->disks
[dd_idx
].rdev
);
3472 if (rdev
&& test_bit(In_sync
, &rdev
->flags
)) {
3473 atomic_inc(&rdev
->nr_pending
);
3475 raid_bio
->bi_next
= (void*)rdev
;
3476 align_bi
->bi_bdev
= rdev
->bdev
;
3477 align_bi
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
3478 align_bi
->bi_sector
+= rdev
->data_offset
;
3480 if (!bio_fits_rdev(align_bi
)) {
3481 /* too big in some way */
3483 rdev_dec_pending(rdev
, mddev
);
3487 spin_lock_irq(&conf
->device_lock
);
3488 wait_event_lock_irq(conf
->wait_for_stripe
,
3490 conf
->device_lock
, /* nothing */);
3491 atomic_inc(&conf
->active_aligned_reads
);
3492 spin_unlock_irq(&conf
->device_lock
);
3494 generic_make_request(align_bi
);
3503 /* __get_priority_stripe - get the next stripe to process
3505 * Full stripe writes are allowed to pass preread active stripes up until
3506 * the bypass_threshold is exceeded. In general the bypass_count
3507 * increments when the handle_list is handled before the hold_list; however, it
3508 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
3509 * stripe with in flight i/o. The bypass_count will be reset when the
3510 * head of the hold_list has changed, i.e. the head was promoted to the
3513 static struct stripe_head
*__get_priority_stripe(raid5_conf_t
*conf
)
3515 struct stripe_head
*sh
;
3517 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
3519 list_empty(&conf
->handle_list
) ? "empty" : "busy",
3520 list_empty(&conf
->hold_list
) ? "empty" : "busy",
3521 atomic_read(&conf
->pending_full_writes
), conf
->bypass_count
);
3523 if (!list_empty(&conf
->handle_list
)) {
3524 sh
= list_entry(conf
->handle_list
.next
, typeof(*sh
), lru
);
3526 if (list_empty(&conf
->hold_list
))
3527 conf
->bypass_count
= 0;
3528 else if (!test_bit(STRIPE_IO_STARTED
, &sh
->state
)) {
3529 if (conf
->hold_list
.next
== conf
->last_hold
)
3530 conf
->bypass_count
++;
3532 conf
->last_hold
= conf
->hold_list
.next
;
3533 conf
->bypass_count
-= conf
->bypass_threshold
;
3534 if (conf
->bypass_count
< 0)
3535 conf
->bypass_count
= 0;
3538 } else if (!list_empty(&conf
->hold_list
) &&
3539 ((conf
->bypass_threshold
&&
3540 conf
->bypass_count
> conf
->bypass_threshold
) ||
3541 atomic_read(&conf
->pending_full_writes
) == 0)) {
3542 sh
= list_entry(conf
->hold_list
.next
,
3544 conf
->bypass_count
-= conf
->bypass_threshold
;
3545 if (conf
->bypass_count
< 0)
3546 conf
->bypass_count
= 0;
3550 list_del_init(&sh
->lru
);
3551 atomic_inc(&sh
->count
);
3552 BUG_ON(atomic_read(&sh
->count
) != 1);
3556 static int make_request(mddev_t
*mddev
, struct bio
* bi
)
3558 raid5_conf_t
*conf
= mddev
->private;
3560 sector_t new_sector
;
3561 sector_t logical_sector
, last_sector
;
3562 struct stripe_head
*sh
;
3563 const int rw
= bio_data_dir(bi
);
3567 if (unlikely(bi
->bi_rw
& REQ_FLUSH
)) {
3568 md_flush_request(mddev
, bi
);
3572 md_write_start(mddev
, bi
);
3575 mddev
->reshape_position
== MaxSector
&&
3576 chunk_aligned_read(mddev
,bi
))
3579 logical_sector
= bi
->bi_sector
& ~((sector_t
)STRIPE_SECTORS
-1);
3580 last_sector
= bi
->bi_sector
+ (bi
->bi_size
>>9);
3582 bi
->bi_phys_segments
= 1; /* over-loaded to count active stripes */
3584 plugged
= mddev_check_plugged(mddev
);
3585 for (;logical_sector
< last_sector
; logical_sector
+= STRIPE_SECTORS
) {
3587 int disks
, data_disks
;
3592 disks
= conf
->raid_disks
;
3593 prepare_to_wait(&conf
->wait_for_overlap
, &w
, TASK_UNINTERRUPTIBLE
);
3594 if (unlikely(conf
->reshape_progress
!= MaxSector
)) {
3595 /* spinlock is needed as reshape_progress may be
3596 * 64bit on a 32bit platform, and so it might be
3597 * possible to see a half-updated value
3598 * Of course reshape_progress could change after
3599 * the lock is dropped, so once we get a reference
3600 * to the stripe that we think it is, we will have
3603 spin_lock_irq(&conf
->device_lock
);
3604 if (mddev
->delta_disks
< 0
3605 ? logical_sector
< conf
->reshape_progress
3606 : logical_sector
>= conf
->reshape_progress
) {
3607 disks
= conf
->previous_raid_disks
;
3610 if (mddev
->delta_disks
< 0
3611 ? logical_sector
< conf
->reshape_safe
3612 : logical_sector
>= conf
->reshape_safe
) {
3613 spin_unlock_irq(&conf
->device_lock
);
3618 spin_unlock_irq(&conf
->device_lock
);
3620 data_disks
= disks
- conf
->max_degraded
;
3622 new_sector
= raid5_compute_sector(conf
, logical_sector
,
3625 pr_debug("raid456: make_request, sector %llu logical %llu\n",
3626 (unsigned long long)new_sector
,
3627 (unsigned long long)logical_sector
);
3629 sh
= get_active_stripe(conf
, new_sector
, previous
,
3630 (bi
->bi_rw
&RWA_MASK
), 0);
3632 if (unlikely(previous
)) {
3633 /* expansion might have moved on while waiting for a
3634 * stripe, so we must do the range check again.
3635 * Expansion could still move past after this
3636 * test, but as we are holding a reference to
3637 * 'sh', we know that if that happens,
3638 * STRIPE_EXPANDING will get set and the expansion
3639 * won't proceed until we finish with the stripe.
3642 spin_lock_irq(&conf
->device_lock
);
3643 if (mddev
->delta_disks
< 0
3644 ? logical_sector
>= conf
->reshape_progress
3645 : logical_sector
< conf
->reshape_progress
)
3646 /* mismatch, need to try again */
3648 spin_unlock_irq(&conf
->device_lock
);
3657 logical_sector
>= mddev
->suspend_lo
&&
3658 logical_sector
< mddev
->suspend_hi
) {
3660 /* As the suspend_* range is controlled by
3661 * userspace, we want an interruptible
3664 flush_signals(current
);
3665 prepare_to_wait(&conf
->wait_for_overlap
,
3666 &w
, TASK_INTERRUPTIBLE
);
3667 if (logical_sector
>= mddev
->suspend_lo
&&
3668 logical_sector
< mddev
->suspend_hi
)
3673 if (test_bit(STRIPE_EXPANDING
, &sh
->state
) ||
3674 !add_stripe_bio(sh
, bi
, dd_idx
, rw
)) {
3675 /* Stripe is busy expanding or
3676 * add failed due to overlap. Flush everything
3679 md_wakeup_thread(mddev
->thread
);
3684 finish_wait(&conf
->wait_for_overlap
, &w
);
3685 set_bit(STRIPE_HANDLE
, &sh
->state
);
3686 clear_bit(STRIPE_DELAYED
, &sh
->state
);
3687 if ((bi
->bi_rw
& REQ_SYNC
) &&
3688 !test_and_set_bit(STRIPE_PREREAD_ACTIVE
, &sh
->state
))
3689 atomic_inc(&conf
->preread_active_stripes
);
3692 /* cannot get stripe for read-ahead, just give-up */
3693 clear_bit(BIO_UPTODATE
, &bi
->bi_flags
);
3694 finish_wait(&conf
->wait_for_overlap
, &w
);
3700 md_wakeup_thread(mddev
->thread
);
3702 spin_lock_irq(&conf
->device_lock
);
3703 remaining
= raid5_dec_bi_phys_segments(bi
);
3704 spin_unlock_irq(&conf
->device_lock
);
3705 if (remaining
== 0) {
3708 md_write_end(mddev
);
3716 static sector_t
raid5_size(mddev_t
*mddev
, sector_t sectors
, int raid_disks
);
3718 static sector_t
reshape_request(mddev_t
*mddev
, sector_t sector_nr
, int *skipped
)
3720 /* reshaping is quite different to recovery/resync so it is
3721 * handled quite separately ... here.
3723 * On each call to sync_request, we gather one chunk worth of
3724 * destination stripes and flag them as expanding.
3725 * Then we find all the source stripes and request reads.
3726 * As the reads complete, handle_stripe will copy the data
3727 * into the destination stripe and release that stripe.
3729 raid5_conf_t
*conf
= mddev
->private;
3730 struct stripe_head
*sh
;
3731 sector_t first_sector
, last_sector
;
3732 int raid_disks
= conf
->previous_raid_disks
;
3733 int data_disks
= raid_disks
- conf
->max_degraded
;
3734 int new_data_disks
= conf
->raid_disks
- conf
->max_degraded
;
3737 sector_t writepos
, readpos
, safepos
;
3738 sector_t stripe_addr
;
3739 int reshape_sectors
;
3740 struct list_head stripes
;
3742 if (sector_nr
== 0) {
3743 /* If restarting in the middle, skip the initial sectors */
3744 if (mddev
->delta_disks
< 0 &&
3745 conf
->reshape_progress
< raid5_size(mddev
, 0, 0)) {
3746 sector_nr
= raid5_size(mddev
, 0, 0)
3747 - conf
->reshape_progress
;
3748 } else if (mddev
->delta_disks
>= 0 &&
3749 conf
->reshape_progress
> 0)
3750 sector_nr
= conf
->reshape_progress
;
3751 sector_div(sector_nr
, new_data_disks
);
3753 mddev
->curr_resync_completed
= sector_nr
;
3754 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
3760 /* We need to process a full chunk at a time.
3761 * If old and new chunk sizes differ, we need to process the
3764 if (mddev
->new_chunk_sectors
> mddev
->chunk_sectors
)
3765 reshape_sectors
= mddev
->new_chunk_sectors
;
3767 reshape_sectors
= mddev
->chunk_sectors
;
3769 /* we update the metadata when there is more than 3Meg
3770 * in the block range (that is rather arbitrary, should
3771 * probably be time based) or when the data about to be
3772 * copied would over-write the source of the data at
3773 * the front of the range.
3774 * i.e. one new_stripe along from reshape_progress new_maps
3775 * to after where reshape_safe old_maps to
3777 writepos
= conf
->reshape_progress
;
3778 sector_div(writepos
, new_data_disks
);
3779 readpos
= conf
->reshape_progress
;
3780 sector_div(readpos
, data_disks
);
3781 safepos
= conf
->reshape_safe
;
3782 sector_div(safepos
, data_disks
);
3783 if (mddev
->delta_disks
< 0) {
3784 writepos
-= min_t(sector_t
, reshape_sectors
, writepos
);
3785 readpos
+= reshape_sectors
;
3786 safepos
+= reshape_sectors
;
3788 writepos
+= reshape_sectors
;
3789 readpos
-= min_t(sector_t
, reshape_sectors
, readpos
);
3790 safepos
-= min_t(sector_t
, reshape_sectors
, safepos
);
3793 /* 'writepos' is the most advanced device address we might write.
3794 * 'readpos' is the least advanced device address we might read.
3795 * 'safepos' is the least address recorded in the metadata as having
3797 * If 'readpos' is behind 'writepos', then there is no way that we can
3798 * ensure safety in the face of a crash - that must be done by userspace
3799 * making a backup of the data. So in that case there is no particular
3800 * rush to update metadata.
3801 * Otherwise if 'safepos' is behind 'writepos', then we really need to
3802 * update the metadata to advance 'safepos' to match 'readpos' so that
3803 * we can be safe in the event of a crash.
3804 * So we insist on updating metadata if safepos is behind writepos and
3805 * readpos is beyond writepos.
3806 * In any case, update the metadata every 10 seconds.
3807 * Maybe that number should be configurable, but I'm not sure it is
3808 * worth it.... maybe it could be a multiple of safemode_delay???
3810 if ((mddev
->delta_disks
< 0
3811 ? (safepos
> writepos
&& readpos
< writepos
)
3812 : (safepos
< writepos
&& readpos
> writepos
)) ||
3813 time_after(jiffies
, conf
->reshape_checkpoint
+ 10*HZ
)) {
3814 /* Cannot proceed until we've updated the superblock... */
3815 wait_event(conf
->wait_for_overlap
,
3816 atomic_read(&conf
->reshape_stripes
)==0);
3817 mddev
->reshape_position
= conf
->reshape_progress
;
3818 mddev
->curr_resync_completed
= sector_nr
;
3819 conf
->reshape_checkpoint
= jiffies
;
3820 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3821 md_wakeup_thread(mddev
->thread
);
3822 wait_event(mddev
->sb_wait
, mddev
->flags
== 0 ||
3823 kthread_should_stop());
3824 spin_lock_irq(&conf
->device_lock
);
3825 conf
->reshape_safe
= mddev
->reshape_position
;
3826 spin_unlock_irq(&conf
->device_lock
);
3827 wake_up(&conf
->wait_for_overlap
);
3828 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
3831 if (mddev
->delta_disks
< 0) {
3832 BUG_ON(conf
->reshape_progress
== 0);
3833 stripe_addr
= writepos
;
3834 BUG_ON((mddev
->dev_sectors
&
3835 ~((sector_t
)reshape_sectors
- 1))
3836 - reshape_sectors
- stripe_addr
3839 BUG_ON(writepos
!= sector_nr
+ reshape_sectors
);
3840 stripe_addr
= sector_nr
;
3842 INIT_LIST_HEAD(&stripes
);
3843 for (i
= 0; i
< reshape_sectors
; i
+= STRIPE_SECTORS
) {
3845 int skipped_disk
= 0;
3846 sh
= get_active_stripe(conf
, stripe_addr
+i
, 0, 0, 1);
3847 set_bit(STRIPE_EXPANDING
, &sh
->state
);
3848 atomic_inc(&conf
->reshape_stripes
);
3849 /* If any of this stripe is beyond the end of the old
3850 * array, then we need to zero those blocks
3852 for (j
=sh
->disks
; j
--;) {
3854 if (j
== sh
->pd_idx
)
3856 if (conf
->level
== 6 &&
3859 s
= compute_blocknr(sh
, j
, 0);
3860 if (s
< raid5_size(mddev
, 0, 0)) {
3864 memset(page_address(sh
->dev
[j
].page
), 0, STRIPE_SIZE
);
3865 set_bit(R5_Expanded
, &sh
->dev
[j
].flags
);
3866 set_bit(R5_UPTODATE
, &sh
->dev
[j
].flags
);
3868 if (!skipped_disk
) {
3869 set_bit(STRIPE_EXPAND_READY
, &sh
->state
);
3870 set_bit(STRIPE_HANDLE
, &sh
->state
);
3872 list_add(&sh
->lru
, &stripes
);
3874 spin_lock_irq(&conf
->device_lock
);
3875 if (mddev
->delta_disks
< 0)
3876 conf
->reshape_progress
-= reshape_sectors
* new_data_disks
;
3878 conf
->reshape_progress
+= reshape_sectors
* new_data_disks
;
3879 spin_unlock_irq(&conf
->device_lock
);
3880 /* Ok, those stripe are ready. We can start scheduling
3881 * reads on the source stripes.
3882 * The source stripes are determined by mapping the first and last
3883 * block on the destination stripes.
3886 raid5_compute_sector(conf
, stripe_addr
*(new_data_disks
),
3889 raid5_compute_sector(conf
, ((stripe_addr
+reshape_sectors
)
3890 * new_data_disks
- 1),
3892 if (last_sector
>= mddev
->dev_sectors
)
3893 last_sector
= mddev
->dev_sectors
- 1;
3894 while (first_sector
<= last_sector
) {
3895 sh
= get_active_stripe(conf
, first_sector
, 1, 0, 1);
3896 set_bit(STRIPE_EXPAND_SOURCE
, &sh
->state
);
3897 set_bit(STRIPE_HANDLE
, &sh
->state
);
3899 first_sector
+= STRIPE_SECTORS
;
3901 /* Now that the sources are clearly marked, we can release
3902 * the destination stripes
3904 while (!list_empty(&stripes
)) {
3905 sh
= list_entry(stripes
.next
, struct stripe_head
, lru
);
3906 list_del_init(&sh
->lru
);
3909 /* If this takes us to the resync_max point where we have to pause,
3910 * then we need to write out the superblock.
3912 sector_nr
+= reshape_sectors
;
3913 if ((sector_nr
- mddev
->curr_resync_completed
) * 2
3914 >= mddev
->resync_max
- mddev
->curr_resync_completed
) {
3915 /* Cannot proceed until we've updated the superblock... */
3916 wait_event(conf
->wait_for_overlap
,
3917 atomic_read(&conf
->reshape_stripes
) == 0);
3918 mddev
->reshape_position
= conf
->reshape_progress
;
3919 mddev
->curr_resync_completed
= sector_nr
;
3920 conf
->reshape_checkpoint
= jiffies
;
3921 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3922 md_wakeup_thread(mddev
->thread
);
3923 wait_event(mddev
->sb_wait
,
3924 !test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)
3925 || kthread_should_stop());
3926 spin_lock_irq(&conf
->device_lock
);
3927 conf
->reshape_safe
= mddev
->reshape_position
;
3928 spin_unlock_irq(&conf
->device_lock
);
3929 wake_up(&conf
->wait_for_overlap
);
3930 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
3932 return reshape_sectors
;
3935 /* FIXME go_faster isn't used */
3936 static inline sector_t
sync_request(mddev_t
*mddev
, sector_t sector_nr
, int *skipped
, int go_faster
)
3938 raid5_conf_t
*conf
= mddev
->private;
3939 struct stripe_head
*sh
;
3940 sector_t max_sector
= mddev
->dev_sectors
;
3941 sector_t sync_blocks
;
3942 int still_degraded
= 0;
3945 if (sector_nr
>= max_sector
) {
3946 /* just being told to finish up .. nothing much to do */
3948 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)) {
3953 if (mddev
->curr_resync
< max_sector
) /* aborted */
3954 bitmap_end_sync(mddev
->bitmap
, mddev
->curr_resync
,
3956 else /* completed sync */
3958 bitmap_close_sync(mddev
->bitmap
);
3963 /* Allow raid5_quiesce to complete */
3964 wait_event(conf
->wait_for_overlap
, conf
->quiesce
!= 2);
3966 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
3967 return reshape_request(mddev
, sector_nr
, skipped
);
3969 /* No need to check resync_max as we never do more than one
3970 * stripe, and as resync_max will always be on a chunk boundary,
3971 * if the check in md_do_sync didn't fire, there is no chance
3972 * of overstepping resync_max here
3975 /* if there is too many failed drives and we are trying
3976 * to resync, then assert that we are finished, because there is
3977 * nothing we can do.
3979 if (mddev
->degraded
>= conf
->max_degraded
&&
3980 test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
3981 sector_t rv
= mddev
->dev_sectors
- sector_nr
;
3985 if (!bitmap_start_sync(mddev
->bitmap
, sector_nr
, &sync_blocks
, 1) &&
3986 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
) &&
3987 !conf
->fullsync
&& sync_blocks
>= STRIPE_SECTORS
) {
3988 /* we can skip this block, and probably more */
3989 sync_blocks
/= STRIPE_SECTORS
;
3991 return sync_blocks
* STRIPE_SECTORS
; /* keep things rounded to whole stripes */
3995 bitmap_cond_end_sync(mddev
->bitmap
, sector_nr
);
3997 sh
= get_active_stripe(conf
, sector_nr
, 0, 1, 0);
3999 sh
= get_active_stripe(conf
, sector_nr
, 0, 0, 0);
4000 /* make sure we don't swamp the stripe cache if someone else
4001 * is trying to get access
4003 schedule_timeout_uninterruptible(1);
4005 /* Need to check if array will still be degraded after recovery/resync
4006 * We don't need to check the 'failed' flag as when that gets set,
4009 for (i
= 0; i
< conf
->raid_disks
; i
++)
4010 if (conf
->disks
[i
].rdev
== NULL
)
4013 bitmap_start_sync(mddev
->bitmap
, sector_nr
, &sync_blocks
, still_degraded
);
4015 set_bit(STRIPE_SYNC_REQUESTED
, &sh
->state
);
4020 return STRIPE_SECTORS
;
4023 static int retry_aligned_read(raid5_conf_t
*conf
, struct bio
*raid_bio
)
4025 /* We may not be able to submit a whole bio at once as there
4026 * may not be enough stripe_heads available.
4027 * We cannot pre-allocate enough stripe_heads as we may need
4028 * more than exist in the cache (if we allow ever large chunks).
4029 * So we do one stripe head at a time and record in
4030 * ->bi_hw_segments how many have been done.
4032 * We *know* that this entire raid_bio is in one chunk, so
4033 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
4035 struct stripe_head
*sh
;
4037 sector_t sector
, logical_sector
, last_sector
;
4042 logical_sector
= raid_bio
->bi_sector
& ~((sector_t
)STRIPE_SECTORS
-1);
4043 sector
= raid5_compute_sector(conf
, logical_sector
,
4045 last_sector
= raid_bio
->bi_sector
+ (raid_bio
->bi_size
>>9);
4047 for (; logical_sector
< last_sector
;
4048 logical_sector
+= STRIPE_SECTORS
,
4049 sector
+= STRIPE_SECTORS
,
4052 if (scnt
< raid5_bi_hw_segments(raid_bio
))
4053 /* already done this stripe */
4056 sh
= get_active_stripe(conf
, sector
, 0, 1, 0);
4059 /* failed to get a stripe - must wait */
4060 raid5_set_bi_hw_segments(raid_bio
, scnt
);
4061 conf
->retry_read_aligned
= raid_bio
;
4065 set_bit(R5_ReadError
, &sh
->dev
[dd_idx
].flags
);
4066 if (!add_stripe_bio(sh
, raid_bio
, dd_idx
, 0)) {
4068 raid5_set_bi_hw_segments(raid_bio
, scnt
);
4069 conf
->retry_read_aligned
= raid_bio
;
4077 spin_lock_irq(&conf
->device_lock
);
4078 remaining
= raid5_dec_bi_phys_segments(raid_bio
);
4079 spin_unlock_irq(&conf
->device_lock
);
4081 bio_endio(raid_bio
, 0);
4082 if (atomic_dec_and_test(&conf
->active_aligned_reads
))
4083 wake_up(&conf
->wait_for_stripe
);
4089 * This is our raid5 kernel thread.
4091 * We scan the hash table for stripes which can be handled now.
4092 * During the scan, completed stripes are saved for us by the interrupt
4093 * handler, so that they will not have to wait for our next wakeup.
4095 static void raid5d(mddev_t
*mddev
)
4097 struct stripe_head
*sh
;
4098 raid5_conf_t
*conf
= mddev
->private;
4100 struct blk_plug plug
;
4102 pr_debug("+++ raid5d active\n");
4104 md_check_recovery(mddev
);
4106 blk_start_plug(&plug
);
4108 spin_lock_irq(&conf
->device_lock
);
4112 if (atomic_read(&mddev
->plug_cnt
) == 0 &&
4113 !list_empty(&conf
->bitmap_list
)) {
4114 /* Now is a good time to flush some bitmap updates */
4116 spin_unlock_irq(&conf
->device_lock
);
4117 bitmap_unplug(mddev
->bitmap
);
4118 spin_lock_irq(&conf
->device_lock
);
4119 conf
->seq_write
= conf
->seq_flush
;
4120 activate_bit_delay(conf
);
4122 if (atomic_read(&mddev
->plug_cnt
) == 0)
4123 raid5_activate_delayed(conf
);
4125 while ((bio
= remove_bio_from_retry(conf
))) {
4127 spin_unlock_irq(&conf
->device_lock
);
4128 ok
= retry_aligned_read(conf
, bio
);
4129 spin_lock_irq(&conf
->device_lock
);
4135 sh
= __get_priority_stripe(conf
);
4139 spin_unlock_irq(&conf
->device_lock
);
4146 spin_lock_irq(&conf
->device_lock
);
4148 pr_debug("%d stripes handled\n", handled
);
4150 spin_unlock_irq(&conf
->device_lock
);
4152 async_tx_issue_pending_all();
4153 blk_finish_plug(&plug
);
4155 pr_debug("--- raid5d inactive\n");
4159 raid5_show_stripe_cache_size(mddev_t
*mddev
, char *page
)
4161 raid5_conf_t
*conf
= mddev
->private;
4163 return sprintf(page
, "%d\n", conf
->max_nr_stripes
);
4169 raid5_set_cache_size(mddev_t
*mddev
, int size
)
4171 raid5_conf_t
*conf
= mddev
->private;
4174 if (size
<= 16 || size
> 32768)
4176 while (size
< conf
->max_nr_stripes
) {
4177 if (drop_one_stripe(conf
))
4178 conf
->max_nr_stripes
--;
4182 err
= md_allow_write(mddev
);
4185 while (size
> conf
->max_nr_stripes
) {
4186 if (grow_one_stripe(conf
))
4187 conf
->max_nr_stripes
++;
4192 EXPORT_SYMBOL(raid5_set_cache_size
);
4195 raid5_store_stripe_cache_size(mddev_t
*mddev
, const char *page
, size_t len
)
4197 raid5_conf_t
*conf
= mddev
->private;
4201 if (len
>= PAGE_SIZE
)
4206 if (strict_strtoul(page
, 10, &new))
4208 err
= raid5_set_cache_size(mddev
, new);
4214 static struct md_sysfs_entry
4215 raid5_stripecache_size
= __ATTR(stripe_cache_size
, S_IRUGO
| S_IWUSR
,
4216 raid5_show_stripe_cache_size
,
4217 raid5_store_stripe_cache_size
);
4220 raid5_show_preread_threshold(mddev_t
*mddev
, char *page
)
4222 raid5_conf_t
*conf
= mddev
->private;
4224 return sprintf(page
, "%d\n", conf
->bypass_threshold
);
4230 raid5_store_preread_threshold(mddev_t
*mddev
, const char *page
, size_t len
)
4232 raid5_conf_t
*conf
= mddev
->private;
4234 if (len
>= PAGE_SIZE
)
4239 if (strict_strtoul(page
, 10, &new))
4241 if (new > conf
->max_nr_stripes
)
4243 conf
->bypass_threshold
= new;
4247 static struct md_sysfs_entry
4248 raid5_preread_bypass_threshold
= __ATTR(preread_bypass_threshold
,
4250 raid5_show_preread_threshold
,
4251 raid5_store_preread_threshold
);
4254 stripe_cache_active_show(mddev_t
*mddev
, char *page
)
4256 raid5_conf_t
*conf
= mddev
->private;
4258 return sprintf(page
, "%d\n", atomic_read(&conf
->active_stripes
));
4263 static struct md_sysfs_entry
4264 raid5_stripecache_active
= __ATTR_RO(stripe_cache_active
);
4266 static struct attribute
*raid5_attrs
[] = {
4267 &raid5_stripecache_size
.attr
,
4268 &raid5_stripecache_active
.attr
,
4269 &raid5_preread_bypass_threshold
.attr
,
4272 static struct attribute_group raid5_attrs_group
= {
4274 .attrs
= raid5_attrs
,
4278 raid5_size(mddev_t
*mddev
, sector_t sectors
, int raid_disks
)
4280 raid5_conf_t
*conf
= mddev
->private;
4283 sectors
= mddev
->dev_sectors
;
4285 /* size is defined by the smallest of previous and new size */
4286 raid_disks
= min(conf
->raid_disks
, conf
->previous_raid_disks
);
4288 sectors
&= ~((sector_t
)mddev
->chunk_sectors
- 1);
4289 sectors
&= ~((sector_t
)mddev
->new_chunk_sectors
- 1);
4290 return sectors
* (raid_disks
- conf
->max_degraded
);
4293 static void raid5_free_percpu(raid5_conf_t
*conf
)
4295 struct raid5_percpu
*percpu
;
4302 for_each_possible_cpu(cpu
) {
4303 percpu
= per_cpu_ptr(conf
->percpu
, cpu
);
4304 safe_put_page(percpu
->spare_page
);
4305 kfree(percpu
->scribble
);
4307 #ifdef CONFIG_HOTPLUG_CPU
4308 unregister_cpu_notifier(&conf
->cpu_notify
);
4312 free_percpu(conf
->percpu
);
4315 static void free_conf(raid5_conf_t
*conf
)
4317 shrink_stripes(conf
);
4318 raid5_free_percpu(conf
);
4320 kfree(conf
->stripe_hashtbl
);
4324 #ifdef CONFIG_HOTPLUG_CPU
4325 static int raid456_cpu_notify(struct notifier_block
*nfb
, unsigned long action
,
4328 raid5_conf_t
*conf
= container_of(nfb
, raid5_conf_t
, cpu_notify
);
4329 long cpu
= (long)hcpu
;
4330 struct raid5_percpu
*percpu
= per_cpu_ptr(conf
->percpu
, cpu
);
4333 case CPU_UP_PREPARE
:
4334 case CPU_UP_PREPARE_FROZEN
:
4335 if (conf
->level
== 6 && !percpu
->spare_page
)
4336 percpu
->spare_page
= alloc_page(GFP_KERNEL
);
4337 if (!percpu
->scribble
)
4338 percpu
->scribble
= kmalloc(conf
->scribble_len
, GFP_KERNEL
);
4340 if (!percpu
->scribble
||
4341 (conf
->level
== 6 && !percpu
->spare_page
)) {
4342 safe_put_page(percpu
->spare_page
);
4343 kfree(percpu
->scribble
);
4344 pr_err("%s: failed memory allocation for cpu%ld\n",
4346 return notifier_from_errno(-ENOMEM
);
4350 case CPU_DEAD_FROZEN
:
4351 safe_put_page(percpu
->spare_page
);
4352 kfree(percpu
->scribble
);
4353 percpu
->spare_page
= NULL
;
4354 percpu
->scribble
= NULL
;
4363 static int raid5_alloc_percpu(raid5_conf_t
*conf
)
4366 struct page
*spare_page
;
4367 struct raid5_percpu __percpu
*allcpus
;
4371 allcpus
= alloc_percpu(struct raid5_percpu
);
4374 conf
->percpu
= allcpus
;
4378 for_each_present_cpu(cpu
) {
4379 if (conf
->level
== 6) {
4380 spare_page
= alloc_page(GFP_KERNEL
);
4385 per_cpu_ptr(conf
->percpu
, cpu
)->spare_page
= spare_page
;
4387 scribble
= kmalloc(conf
->scribble_len
, GFP_KERNEL
);
4392 per_cpu_ptr(conf
->percpu
, cpu
)->scribble
= scribble
;
4394 #ifdef CONFIG_HOTPLUG_CPU
4395 conf
->cpu_notify
.notifier_call
= raid456_cpu_notify
;
4396 conf
->cpu_notify
.priority
= 0;
4398 err
= register_cpu_notifier(&conf
->cpu_notify
);
4405 static raid5_conf_t
*setup_conf(mddev_t
*mddev
)
4408 int raid_disk
, memory
, max_disks
;
4410 struct disk_info
*disk
;
4412 if (mddev
->new_level
!= 5
4413 && mddev
->new_level
!= 4
4414 && mddev
->new_level
!= 6) {
4415 printk(KERN_ERR
"md/raid:%s: raid level not set to 4/5/6 (%d)\n",
4416 mdname(mddev
), mddev
->new_level
);
4417 return ERR_PTR(-EIO
);
4419 if ((mddev
->new_level
== 5
4420 && !algorithm_valid_raid5(mddev
->new_layout
)) ||
4421 (mddev
->new_level
== 6
4422 && !algorithm_valid_raid6(mddev
->new_layout
))) {
4423 printk(KERN_ERR
"md/raid:%s: layout %d not supported\n",
4424 mdname(mddev
), mddev
->new_layout
);
4425 return ERR_PTR(-EIO
);
4427 if (mddev
->new_level
== 6 && mddev
->raid_disks
< 4) {
4428 printk(KERN_ERR
"md/raid:%s: not enough configured devices (%d, minimum 4)\n",
4429 mdname(mddev
), mddev
->raid_disks
);
4430 return ERR_PTR(-EINVAL
);
4433 if (!mddev
->new_chunk_sectors
||
4434 (mddev
->new_chunk_sectors
<< 9) % PAGE_SIZE
||
4435 !is_power_of_2(mddev
->new_chunk_sectors
)) {
4436 printk(KERN_ERR
"md/raid:%s: invalid chunk size %d\n",
4437 mdname(mddev
), mddev
->new_chunk_sectors
<< 9);
4438 return ERR_PTR(-EINVAL
);
4441 conf
= kzalloc(sizeof(raid5_conf_t
), GFP_KERNEL
);
4444 spin_lock_init(&conf
->device_lock
);
4445 init_waitqueue_head(&conf
->wait_for_stripe
);
4446 init_waitqueue_head(&conf
->wait_for_overlap
);
4447 INIT_LIST_HEAD(&conf
->handle_list
);
4448 INIT_LIST_HEAD(&conf
->hold_list
);
4449 INIT_LIST_HEAD(&conf
->delayed_list
);
4450 INIT_LIST_HEAD(&conf
->bitmap_list
);
4451 INIT_LIST_HEAD(&conf
->inactive_list
);
4452 atomic_set(&conf
->active_stripes
, 0);
4453 atomic_set(&conf
->preread_active_stripes
, 0);
4454 atomic_set(&conf
->active_aligned_reads
, 0);
4455 conf
->bypass_threshold
= BYPASS_THRESHOLD
;
4457 conf
->raid_disks
= mddev
->raid_disks
;
4458 if (mddev
->reshape_position
== MaxSector
)
4459 conf
->previous_raid_disks
= mddev
->raid_disks
;
4461 conf
->previous_raid_disks
= mddev
->raid_disks
- mddev
->delta_disks
;
4462 max_disks
= max(conf
->raid_disks
, conf
->previous_raid_disks
);
4463 conf
->scribble_len
= scribble_len(max_disks
);
4465 conf
->disks
= kzalloc(max_disks
* sizeof(struct disk_info
),
4470 conf
->mddev
= mddev
;
4472 if ((conf
->stripe_hashtbl
= kzalloc(PAGE_SIZE
, GFP_KERNEL
)) == NULL
)
4475 conf
->level
= mddev
->new_level
;
4476 if (raid5_alloc_percpu(conf
) != 0)
4479 pr_debug("raid456: run(%s) called.\n", mdname(mddev
));
4481 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4482 raid_disk
= rdev
->raid_disk
;
4483 if (raid_disk
>= max_disks
4486 disk
= conf
->disks
+ raid_disk
;
4490 if (test_bit(In_sync
, &rdev
->flags
)) {
4491 char b
[BDEVNAME_SIZE
];
4492 printk(KERN_INFO
"md/raid:%s: device %s operational as raid"
4494 mdname(mddev
), bdevname(rdev
->bdev
, b
), raid_disk
);
4495 } else if (rdev
->saved_raid_disk
!= raid_disk
)
4496 /* Cannot rely on bitmap to complete recovery */
4500 conf
->chunk_sectors
= mddev
->new_chunk_sectors
;
4501 conf
->level
= mddev
->new_level
;
4502 if (conf
->level
== 6)
4503 conf
->max_degraded
= 2;
4505 conf
->max_degraded
= 1;
4506 conf
->algorithm
= mddev
->new_layout
;
4507 conf
->max_nr_stripes
= NR_STRIPES
;
4508 conf
->reshape_progress
= mddev
->reshape_position
;
4509 if (conf
->reshape_progress
!= MaxSector
) {
4510 conf
->prev_chunk_sectors
= mddev
->chunk_sectors
;
4511 conf
->prev_algo
= mddev
->layout
;
4514 memory
= conf
->max_nr_stripes
* (sizeof(struct stripe_head
) +
4515 max_disks
* ((sizeof(struct bio
) + PAGE_SIZE
))) / 1024;
4516 if (grow_stripes(conf
, conf
->max_nr_stripes
)) {
4518 "md/raid:%s: couldn't allocate %dkB for buffers\n",
4519 mdname(mddev
), memory
);
4522 printk(KERN_INFO
"md/raid:%s: allocated %dkB\n",
4523 mdname(mddev
), memory
);
4525 conf
->thread
= md_register_thread(raid5d
, mddev
, NULL
);
4526 if (!conf
->thread
) {
4528 "md/raid:%s: couldn't allocate thread.\n",
4538 return ERR_PTR(-EIO
);
4540 return ERR_PTR(-ENOMEM
);
4544 static int only_parity(int raid_disk
, int algo
, int raid_disks
, int max_degraded
)
4547 case ALGORITHM_PARITY_0
:
4548 if (raid_disk
< max_degraded
)
4551 case ALGORITHM_PARITY_N
:
4552 if (raid_disk
>= raid_disks
- max_degraded
)
4555 case ALGORITHM_PARITY_0_6
:
4556 if (raid_disk
== 0 ||
4557 raid_disk
== raid_disks
- 1)
4560 case ALGORITHM_LEFT_ASYMMETRIC_6
:
4561 case ALGORITHM_RIGHT_ASYMMETRIC_6
:
4562 case ALGORITHM_LEFT_SYMMETRIC_6
:
4563 case ALGORITHM_RIGHT_SYMMETRIC_6
:
4564 if (raid_disk
== raid_disks
- 1)
4570 static int run(mddev_t
*mddev
)
4573 int working_disks
= 0;
4574 int dirty_parity_disks
= 0;
4576 sector_t reshape_offset
= 0;
4578 if (mddev
->recovery_cp
!= MaxSector
)
4579 printk(KERN_NOTICE
"md/raid:%s: not clean"
4580 " -- starting background reconstruction\n",
4582 if (mddev
->reshape_position
!= MaxSector
) {
4583 /* Check that we can continue the reshape.
4584 * Currently only disks can change, it must
4585 * increase, and we must be past the point where
4586 * a stripe over-writes itself
4588 sector_t here_new
, here_old
;
4590 int max_degraded
= (mddev
->level
== 6 ? 2 : 1);
4592 if (mddev
->new_level
!= mddev
->level
) {
4593 printk(KERN_ERR
"md/raid:%s: unsupported reshape "
4594 "required - aborting.\n",
4598 old_disks
= mddev
->raid_disks
- mddev
->delta_disks
;
4599 /* reshape_position must be on a new-stripe boundary, and one
4600 * further up in new geometry must map after here in old
4603 here_new
= mddev
->reshape_position
;
4604 if (sector_div(here_new
, mddev
->new_chunk_sectors
*
4605 (mddev
->raid_disks
- max_degraded
))) {
4606 printk(KERN_ERR
"md/raid:%s: reshape_position not "
4607 "on a stripe boundary\n", mdname(mddev
));
4610 reshape_offset
= here_new
* mddev
->new_chunk_sectors
;
4611 /* here_new is the stripe we will write to */
4612 here_old
= mddev
->reshape_position
;
4613 sector_div(here_old
, mddev
->chunk_sectors
*
4614 (old_disks
-max_degraded
));
4615 /* here_old is the first stripe that we might need to read
4617 if (mddev
->delta_disks
== 0) {
4618 /* We cannot be sure it is safe to start an in-place
4619 * reshape. It is only safe if user-space if monitoring
4620 * and taking constant backups.
4621 * mdadm always starts a situation like this in
4622 * readonly mode so it can take control before
4623 * allowing any writes. So just check for that.
4625 if ((here_new
* mddev
->new_chunk_sectors
!=
4626 here_old
* mddev
->chunk_sectors
) ||
4628 printk(KERN_ERR
"md/raid:%s: in-place reshape must be started"
4629 " in read-only mode - aborting\n",
4633 } else if (mddev
->delta_disks
< 0
4634 ? (here_new
* mddev
->new_chunk_sectors
<=
4635 here_old
* mddev
->chunk_sectors
)
4636 : (here_new
* mddev
->new_chunk_sectors
>=
4637 here_old
* mddev
->chunk_sectors
)) {
4638 /* Reading from the same stripe as writing to - bad */
4639 printk(KERN_ERR
"md/raid:%s: reshape_position too early for "
4640 "auto-recovery - aborting.\n",
4644 printk(KERN_INFO
"md/raid:%s: reshape will continue\n",
4646 /* OK, we should be able to continue; */
4648 BUG_ON(mddev
->level
!= mddev
->new_level
);
4649 BUG_ON(mddev
->layout
!= mddev
->new_layout
);
4650 BUG_ON(mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
);
4651 BUG_ON(mddev
->delta_disks
!= 0);
4654 if (mddev
->private == NULL
)
4655 conf
= setup_conf(mddev
);
4657 conf
= mddev
->private;
4660 return PTR_ERR(conf
);
4662 mddev
->thread
= conf
->thread
;
4663 conf
->thread
= NULL
;
4664 mddev
->private = conf
;
4667 * 0 for a fully functional array, 1 or 2 for a degraded array.
4669 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4670 if (rdev
->badblocks
.count
) {
4671 printk(KERN_ERR
"md/raid5: cannot handle bad blocks yet\n");
4674 if (rdev
->raid_disk
< 0)
4676 if (test_bit(In_sync
, &rdev
->flags
)) {
4680 /* This disc is not fully in-sync. However if it
4681 * just stored parity (beyond the recovery_offset),
4682 * when we don't need to be concerned about the
4683 * array being dirty.
4684 * When reshape goes 'backwards', we never have
4685 * partially completed devices, so we only need
4686 * to worry about reshape going forwards.
4688 /* Hack because v0.91 doesn't store recovery_offset properly. */
4689 if (mddev
->major_version
== 0 &&
4690 mddev
->minor_version
> 90)
4691 rdev
->recovery_offset
= reshape_offset
;
4693 if (rdev
->recovery_offset
< reshape_offset
) {
4694 /* We need to check old and new layout */
4695 if (!only_parity(rdev
->raid_disk
,
4698 conf
->max_degraded
))
4701 if (!only_parity(rdev
->raid_disk
,
4703 conf
->previous_raid_disks
,
4704 conf
->max_degraded
))
4706 dirty_parity_disks
++;
4709 mddev
->degraded
= (max(conf
->raid_disks
, conf
->previous_raid_disks
)
4712 if (has_failed(conf
)) {
4713 printk(KERN_ERR
"md/raid:%s: not enough operational devices"
4714 " (%d/%d failed)\n",
4715 mdname(mddev
), mddev
->degraded
, conf
->raid_disks
);
4719 /* device size must be a multiple of chunk size */
4720 mddev
->dev_sectors
&= ~(mddev
->chunk_sectors
- 1);
4721 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
4723 if (mddev
->degraded
> dirty_parity_disks
&&
4724 mddev
->recovery_cp
!= MaxSector
) {
4725 if (mddev
->ok_start_degraded
)
4727 "md/raid:%s: starting dirty degraded array"
4728 " - data corruption possible.\n",
4732 "md/raid:%s: cannot start dirty degraded array.\n",
4738 if (mddev
->degraded
== 0)
4739 printk(KERN_INFO
"md/raid:%s: raid level %d active with %d out of %d"
4740 " devices, algorithm %d\n", mdname(mddev
), conf
->level
,
4741 mddev
->raid_disks
-mddev
->degraded
, mddev
->raid_disks
,
4744 printk(KERN_ALERT
"md/raid:%s: raid level %d active with %d"
4745 " out of %d devices, algorithm %d\n",
4746 mdname(mddev
), conf
->level
,
4747 mddev
->raid_disks
- mddev
->degraded
,
4748 mddev
->raid_disks
, mddev
->new_layout
);
4750 print_raid5_conf(conf
);
4752 if (conf
->reshape_progress
!= MaxSector
) {
4753 conf
->reshape_safe
= conf
->reshape_progress
;
4754 atomic_set(&conf
->reshape_stripes
, 0);
4755 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4756 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4757 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
4758 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
4759 mddev
->sync_thread
= md_register_thread(md_do_sync
, mddev
,
4764 /* Ok, everything is just fine now */
4765 if (mddev
->to_remove
== &raid5_attrs_group
)
4766 mddev
->to_remove
= NULL
;
4767 else if (mddev
->kobj
.sd
&&
4768 sysfs_create_group(&mddev
->kobj
, &raid5_attrs_group
))
4770 "raid5: failed to create sysfs attributes for %s\n",
4772 md_set_array_sectors(mddev
, raid5_size(mddev
, 0, 0));
4776 /* read-ahead size must cover two whole stripes, which
4777 * is 2 * (datadisks) * chunksize where 'n' is the
4778 * number of raid devices
4780 int data_disks
= conf
->previous_raid_disks
- conf
->max_degraded
;
4781 int stripe
= data_disks
*
4782 ((mddev
->chunk_sectors
<< 9) / PAGE_SIZE
);
4783 if (mddev
->queue
->backing_dev_info
.ra_pages
< 2 * stripe
)
4784 mddev
->queue
->backing_dev_info
.ra_pages
= 2 * stripe
;
4786 blk_queue_merge_bvec(mddev
->queue
, raid5_mergeable_bvec
);
4788 mddev
->queue
->backing_dev_info
.congested_data
= mddev
;
4789 mddev
->queue
->backing_dev_info
.congested_fn
= raid5_congested
;
4791 chunk_size
= mddev
->chunk_sectors
<< 9;
4792 blk_queue_io_min(mddev
->queue
, chunk_size
);
4793 blk_queue_io_opt(mddev
->queue
, chunk_size
*
4794 (conf
->raid_disks
- conf
->max_degraded
));
4796 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4797 disk_stack_limits(mddev
->gendisk
, rdev
->bdev
,
4798 rdev
->data_offset
<< 9);
4803 md_unregister_thread(mddev
->thread
);
4804 mddev
->thread
= NULL
;
4806 print_raid5_conf(conf
);
4809 mddev
->private = NULL
;
4810 printk(KERN_ALERT
"md/raid:%s: failed to run raid set.\n", mdname(mddev
));
4814 static int stop(mddev_t
*mddev
)
4816 raid5_conf_t
*conf
= mddev
->private;
4818 md_unregister_thread(mddev
->thread
);
4819 mddev
->thread
= NULL
;
4821 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
4823 mddev
->private = NULL
;
4824 mddev
->to_remove
= &raid5_attrs_group
;
4829 static void print_sh(struct seq_file
*seq
, struct stripe_head
*sh
)
4833 seq_printf(seq
, "sh %llu, pd_idx %d, state %ld.\n",
4834 (unsigned long long)sh
->sector
, sh
->pd_idx
, sh
->state
);
4835 seq_printf(seq
, "sh %llu, count %d.\n",
4836 (unsigned long long)sh
->sector
, atomic_read(&sh
->count
));
4837 seq_printf(seq
, "sh %llu, ", (unsigned long long)sh
->sector
);
4838 for (i
= 0; i
< sh
->disks
; i
++) {
4839 seq_printf(seq
, "(cache%d: %p %ld) ",
4840 i
, sh
->dev
[i
].page
, sh
->dev
[i
].flags
);
4842 seq_printf(seq
, "\n");
4845 static void printall(struct seq_file
*seq
, raid5_conf_t
*conf
)
4847 struct stripe_head
*sh
;
4848 struct hlist_node
*hn
;
4851 spin_lock_irq(&conf
->device_lock
);
4852 for (i
= 0; i
< NR_HASH
; i
++) {
4853 hlist_for_each_entry(sh
, hn
, &conf
->stripe_hashtbl
[i
], hash
) {
4854 if (sh
->raid_conf
!= conf
)
4859 spin_unlock_irq(&conf
->device_lock
);
4863 static void status(struct seq_file
*seq
, mddev_t
*mddev
)
4865 raid5_conf_t
*conf
= mddev
->private;
4868 seq_printf(seq
, " level %d, %dk chunk, algorithm %d", mddev
->level
,
4869 mddev
->chunk_sectors
/ 2, mddev
->layout
);
4870 seq_printf (seq
, " [%d/%d] [", conf
->raid_disks
, conf
->raid_disks
- mddev
->degraded
);
4871 for (i
= 0; i
< conf
->raid_disks
; i
++)
4872 seq_printf (seq
, "%s",
4873 conf
->disks
[i
].rdev
&&
4874 test_bit(In_sync
, &conf
->disks
[i
].rdev
->flags
) ? "U" : "_");
4875 seq_printf (seq
, "]");
4877 seq_printf (seq
, "\n");
4878 printall(seq
, conf
);
4882 static void print_raid5_conf (raid5_conf_t
*conf
)
4885 struct disk_info
*tmp
;
4887 printk(KERN_DEBUG
"RAID conf printout:\n");
4889 printk("(conf==NULL)\n");
4892 printk(KERN_DEBUG
" --- level:%d rd:%d wd:%d\n", conf
->level
,
4894 conf
->raid_disks
- conf
->mddev
->degraded
);
4896 for (i
= 0; i
< conf
->raid_disks
; i
++) {
4897 char b
[BDEVNAME_SIZE
];
4898 tmp
= conf
->disks
+ i
;
4900 printk(KERN_DEBUG
" disk %d, o:%d, dev:%s\n",
4901 i
, !test_bit(Faulty
, &tmp
->rdev
->flags
),
4902 bdevname(tmp
->rdev
->bdev
, b
));
4906 static int raid5_spare_active(mddev_t
*mddev
)
4909 raid5_conf_t
*conf
= mddev
->private;
4910 struct disk_info
*tmp
;
4912 unsigned long flags
;
4914 for (i
= 0; i
< conf
->raid_disks
; i
++) {
4915 tmp
= conf
->disks
+ i
;
4917 && tmp
->rdev
->recovery_offset
== MaxSector
4918 && !test_bit(Faulty
, &tmp
->rdev
->flags
)
4919 && !test_and_set_bit(In_sync
, &tmp
->rdev
->flags
)) {
4921 sysfs_notify_dirent_safe(tmp
->rdev
->sysfs_state
);
4924 spin_lock_irqsave(&conf
->device_lock
, flags
);
4925 mddev
->degraded
-= count
;
4926 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
4927 print_raid5_conf(conf
);
4931 static int raid5_remove_disk(mddev_t
*mddev
, int number
)
4933 raid5_conf_t
*conf
= mddev
->private;
4936 struct disk_info
*p
= conf
->disks
+ number
;
4938 print_raid5_conf(conf
);
4941 if (number
>= conf
->raid_disks
&&
4942 conf
->reshape_progress
== MaxSector
)
4943 clear_bit(In_sync
, &rdev
->flags
);
4945 if (test_bit(In_sync
, &rdev
->flags
) ||
4946 atomic_read(&rdev
->nr_pending
)) {
4950 /* Only remove non-faulty devices if recovery
4953 if (!test_bit(Faulty
, &rdev
->flags
) &&
4954 !has_failed(conf
) &&
4955 number
< conf
->raid_disks
) {
4961 if (atomic_read(&rdev
->nr_pending
)) {
4962 /* lost the race, try later */
4969 print_raid5_conf(conf
);
4973 static int raid5_add_disk(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
4975 raid5_conf_t
*conf
= mddev
->private;
4978 struct disk_info
*p
;
4980 int last
= conf
->raid_disks
- 1;
4982 if (rdev
->badblocks
.count
)
4985 if (has_failed(conf
))
4986 /* no point adding a device */
4989 if (rdev
->raid_disk
>= 0)
4990 first
= last
= rdev
->raid_disk
;
4993 * find the disk ... but prefer rdev->saved_raid_disk
4996 if (rdev
->saved_raid_disk
>= 0 &&
4997 rdev
->saved_raid_disk
>= first
&&
4998 conf
->disks
[rdev
->saved_raid_disk
].rdev
== NULL
)
4999 disk
= rdev
->saved_raid_disk
;
5002 for ( ; disk
<= last
; disk
++)
5003 if ((p
=conf
->disks
+ disk
)->rdev
== NULL
) {
5004 clear_bit(In_sync
, &rdev
->flags
);
5005 rdev
->raid_disk
= disk
;
5007 if (rdev
->saved_raid_disk
!= disk
)
5009 rcu_assign_pointer(p
->rdev
, rdev
);
5012 print_raid5_conf(conf
);
5016 static int raid5_resize(mddev_t
*mddev
, sector_t sectors
)
5018 /* no resync is happening, and there is enough space
5019 * on all devices, so we can resize.
5020 * We need to make sure resync covers any new space.
5021 * If the array is shrinking we should possibly wait until
5022 * any io in the removed space completes, but it hardly seems
5025 sectors
&= ~((sector_t
)mddev
->chunk_sectors
- 1);
5026 md_set_array_sectors(mddev
, raid5_size(mddev
, sectors
,
5027 mddev
->raid_disks
));
5028 if (mddev
->array_sectors
>
5029 raid5_size(mddev
, sectors
, mddev
->raid_disks
))
5031 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
5032 revalidate_disk(mddev
->gendisk
);
5033 if (sectors
> mddev
->dev_sectors
&&
5034 mddev
->recovery_cp
> mddev
->dev_sectors
) {
5035 mddev
->recovery_cp
= mddev
->dev_sectors
;
5036 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5038 mddev
->dev_sectors
= sectors
;
5039 mddev
->resync_max_sectors
= sectors
;
5043 static int check_stripe_cache(mddev_t
*mddev
)
5045 /* Can only proceed if there are plenty of stripe_heads.
5046 * We need a minimum of one full stripe,, and for sensible progress
5047 * it is best to have about 4 times that.
5048 * If we require 4 times, then the default 256 4K stripe_heads will
5049 * allow for chunk sizes up to 256K, which is probably OK.
5050 * If the chunk size is greater, user-space should request more
5051 * stripe_heads first.
5053 raid5_conf_t
*conf
= mddev
->private;
5054 if (((mddev
->chunk_sectors
<< 9) / STRIPE_SIZE
) * 4
5055 > conf
->max_nr_stripes
||
5056 ((mddev
->new_chunk_sectors
<< 9) / STRIPE_SIZE
) * 4
5057 > conf
->max_nr_stripes
) {
5058 printk(KERN_WARNING
"md/raid:%s: reshape: not enough stripes. Needed %lu\n",
5060 ((max(mddev
->chunk_sectors
, mddev
->new_chunk_sectors
) << 9)
5067 static int check_reshape(mddev_t
*mddev
)
5069 raid5_conf_t
*conf
= mddev
->private;
5071 if (mddev
->delta_disks
== 0 &&
5072 mddev
->new_layout
== mddev
->layout
&&
5073 mddev
->new_chunk_sectors
== mddev
->chunk_sectors
)
5074 return 0; /* nothing to do */
5076 /* Cannot grow a bitmap yet */
5078 if (has_failed(conf
))
5080 if (mddev
->delta_disks
< 0) {
5081 /* We might be able to shrink, but the devices must
5082 * be made bigger first.
5083 * For raid6, 4 is the minimum size.
5084 * Otherwise 2 is the minimum
5087 if (mddev
->level
== 6)
5089 if (mddev
->raid_disks
+ mddev
->delta_disks
< min
)
5093 if (!check_stripe_cache(mddev
))
5096 return resize_stripes(conf
, conf
->raid_disks
+ mddev
->delta_disks
);
5099 static int raid5_start_reshape(mddev_t
*mddev
)
5101 raid5_conf_t
*conf
= mddev
->private;
5104 unsigned long flags
;
5106 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
5109 if (!check_stripe_cache(mddev
))
5112 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
5113 if (!test_bit(In_sync
, &rdev
->flags
)
5114 && !test_bit(Faulty
, &rdev
->flags
))
5117 if (spares
- mddev
->degraded
< mddev
->delta_disks
- conf
->max_degraded
)
5118 /* Not enough devices even to make a degraded array
5123 /* Refuse to reduce size of the array. Any reductions in
5124 * array size must be through explicit setting of array_size
5127 if (raid5_size(mddev
, 0, conf
->raid_disks
+ mddev
->delta_disks
)
5128 < mddev
->array_sectors
) {
5129 printk(KERN_ERR
"md/raid:%s: array size must be reduced "
5130 "before number of disks\n", mdname(mddev
));
5134 atomic_set(&conf
->reshape_stripes
, 0);
5135 spin_lock_irq(&conf
->device_lock
);
5136 conf
->previous_raid_disks
= conf
->raid_disks
;
5137 conf
->raid_disks
+= mddev
->delta_disks
;
5138 conf
->prev_chunk_sectors
= conf
->chunk_sectors
;
5139 conf
->chunk_sectors
= mddev
->new_chunk_sectors
;
5140 conf
->prev_algo
= conf
->algorithm
;
5141 conf
->algorithm
= mddev
->new_layout
;
5142 if (mddev
->delta_disks
< 0)
5143 conf
->reshape_progress
= raid5_size(mddev
, 0, 0);
5145 conf
->reshape_progress
= 0;
5146 conf
->reshape_safe
= conf
->reshape_progress
;
5148 spin_unlock_irq(&conf
->device_lock
);
5150 /* Add some new drives, as many as will fit.
5151 * We know there are enough to make the newly sized array work.
5152 * Don't add devices if we are reducing the number of
5153 * devices in the array. This is because it is not possible
5154 * to correctly record the "partially reconstructed" state of
5155 * such devices during the reshape and confusion could result.
5157 if (mddev
->delta_disks
>= 0) {
5158 int added_devices
= 0;
5159 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
5160 if (rdev
->raid_disk
< 0 &&
5161 !test_bit(Faulty
, &rdev
->flags
)) {
5162 if (raid5_add_disk(mddev
, rdev
) == 0) {
5164 >= conf
->previous_raid_disks
) {
5165 set_bit(In_sync
, &rdev
->flags
);
5168 rdev
->recovery_offset
= 0;
5170 if (sysfs_link_rdev(mddev
, rdev
))
5171 /* Failure here is OK */;
5173 } else if (rdev
->raid_disk
>= conf
->previous_raid_disks
5174 && !test_bit(Faulty
, &rdev
->flags
)) {
5175 /* This is a spare that was manually added */
5176 set_bit(In_sync
, &rdev
->flags
);
5180 /* When a reshape changes the number of devices,
5181 * ->degraded is measured against the larger of the
5182 * pre and post number of devices.
5184 spin_lock_irqsave(&conf
->device_lock
, flags
);
5185 mddev
->degraded
+= (conf
->raid_disks
- conf
->previous_raid_disks
)
5187 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
5189 mddev
->raid_disks
= conf
->raid_disks
;
5190 mddev
->reshape_position
= conf
->reshape_progress
;
5191 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
5193 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
5194 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
5195 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
5196 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
5197 mddev
->sync_thread
= md_register_thread(md_do_sync
, mddev
,
5199 if (!mddev
->sync_thread
) {
5200 mddev
->recovery
= 0;
5201 spin_lock_irq(&conf
->device_lock
);
5202 mddev
->raid_disks
= conf
->raid_disks
= conf
->previous_raid_disks
;
5203 conf
->reshape_progress
= MaxSector
;
5204 spin_unlock_irq(&conf
->device_lock
);
5207 conf
->reshape_checkpoint
= jiffies
;
5208 md_wakeup_thread(mddev
->sync_thread
);
5209 md_new_event(mddev
);
5213 /* This is called from the reshape thread and should make any
5214 * changes needed in 'conf'
5216 static void end_reshape(raid5_conf_t
*conf
)
5219 if (!test_bit(MD_RECOVERY_INTR
, &conf
->mddev
->recovery
)) {
5221 spin_lock_irq(&conf
->device_lock
);
5222 conf
->previous_raid_disks
= conf
->raid_disks
;
5223 conf
->reshape_progress
= MaxSector
;
5224 spin_unlock_irq(&conf
->device_lock
);
5225 wake_up(&conf
->wait_for_overlap
);
5227 /* read-ahead size must cover two whole stripes, which is
5228 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
5230 if (conf
->mddev
->queue
) {
5231 int data_disks
= conf
->raid_disks
- conf
->max_degraded
;
5232 int stripe
= data_disks
* ((conf
->chunk_sectors
<< 9)
5234 if (conf
->mddev
->queue
->backing_dev_info
.ra_pages
< 2 * stripe
)
5235 conf
->mddev
->queue
->backing_dev_info
.ra_pages
= 2 * stripe
;
5240 /* This is called from the raid5d thread with mddev_lock held.
5241 * It makes config changes to the device.
5243 static void raid5_finish_reshape(mddev_t
*mddev
)
5245 raid5_conf_t
*conf
= mddev
->private;
5247 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
5249 if (mddev
->delta_disks
> 0) {
5250 md_set_array_sectors(mddev
, raid5_size(mddev
, 0, 0));
5251 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
5252 revalidate_disk(mddev
->gendisk
);
5255 mddev
->degraded
= conf
->raid_disks
;
5256 for (d
= 0; d
< conf
->raid_disks
; d
++)
5257 if (conf
->disks
[d
].rdev
&&
5259 &conf
->disks
[d
].rdev
->flags
))
5261 for (d
= conf
->raid_disks
;
5262 d
< conf
->raid_disks
- mddev
->delta_disks
;
5264 mdk_rdev_t
*rdev
= conf
->disks
[d
].rdev
;
5265 if (rdev
&& raid5_remove_disk(mddev
, d
) == 0) {
5266 sysfs_unlink_rdev(mddev
, rdev
);
5267 rdev
->raid_disk
= -1;
5271 mddev
->layout
= conf
->algorithm
;
5272 mddev
->chunk_sectors
= conf
->chunk_sectors
;
5273 mddev
->reshape_position
= MaxSector
;
5274 mddev
->delta_disks
= 0;
5278 static void raid5_quiesce(mddev_t
*mddev
, int state
)
5280 raid5_conf_t
*conf
= mddev
->private;
5283 case 2: /* resume for a suspend */
5284 wake_up(&conf
->wait_for_overlap
);
5287 case 1: /* stop all writes */
5288 spin_lock_irq(&conf
->device_lock
);
5289 /* '2' tells resync/reshape to pause so that all
5290 * active stripes can drain
5293 wait_event_lock_irq(conf
->wait_for_stripe
,
5294 atomic_read(&conf
->active_stripes
) == 0 &&
5295 atomic_read(&conf
->active_aligned_reads
) == 0,
5296 conf
->device_lock
, /* nothing */);
5298 spin_unlock_irq(&conf
->device_lock
);
5299 /* allow reshape to continue */
5300 wake_up(&conf
->wait_for_overlap
);
5303 case 0: /* re-enable writes */
5304 spin_lock_irq(&conf
->device_lock
);
5306 wake_up(&conf
->wait_for_stripe
);
5307 wake_up(&conf
->wait_for_overlap
);
5308 spin_unlock_irq(&conf
->device_lock
);
5314 static void *raid45_takeover_raid0(mddev_t
*mddev
, int level
)
5316 struct raid0_private_data
*raid0_priv
= mddev
->private;
5319 /* for raid0 takeover only one zone is supported */
5320 if (raid0_priv
->nr_strip_zones
> 1) {
5321 printk(KERN_ERR
"md/raid:%s: cannot takeover raid0 with more than one zone.\n",
5323 return ERR_PTR(-EINVAL
);
5326 sectors
= raid0_priv
->strip_zone
[0].zone_end
;
5327 sector_div(sectors
, raid0_priv
->strip_zone
[0].nb_dev
);
5328 mddev
->dev_sectors
= sectors
;
5329 mddev
->new_level
= level
;
5330 mddev
->new_layout
= ALGORITHM_PARITY_N
;
5331 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
5332 mddev
->raid_disks
+= 1;
5333 mddev
->delta_disks
= 1;
5334 /* make sure it will be not marked as dirty */
5335 mddev
->recovery_cp
= MaxSector
;
5337 return setup_conf(mddev
);
5341 static void *raid5_takeover_raid1(mddev_t
*mddev
)
5345 if (mddev
->raid_disks
!= 2 ||
5346 mddev
->degraded
> 1)
5347 return ERR_PTR(-EINVAL
);
5349 /* Should check if there are write-behind devices? */
5351 chunksect
= 64*2; /* 64K by default */
5353 /* The array must be an exact multiple of chunksize */
5354 while (chunksect
&& (mddev
->array_sectors
& (chunksect
-1)))
5357 if ((chunksect
<<9) < STRIPE_SIZE
)
5358 /* array size does not allow a suitable chunk size */
5359 return ERR_PTR(-EINVAL
);
5361 mddev
->new_level
= 5;
5362 mddev
->new_layout
= ALGORITHM_LEFT_SYMMETRIC
;
5363 mddev
->new_chunk_sectors
= chunksect
;
5365 return setup_conf(mddev
);
5368 static void *raid5_takeover_raid6(mddev_t
*mddev
)
5372 switch (mddev
->layout
) {
5373 case ALGORITHM_LEFT_ASYMMETRIC_6
:
5374 new_layout
= ALGORITHM_LEFT_ASYMMETRIC
;
5376 case ALGORITHM_RIGHT_ASYMMETRIC_6
:
5377 new_layout
= ALGORITHM_RIGHT_ASYMMETRIC
;
5379 case ALGORITHM_LEFT_SYMMETRIC_6
:
5380 new_layout
= ALGORITHM_LEFT_SYMMETRIC
;
5382 case ALGORITHM_RIGHT_SYMMETRIC_6
:
5383 new_layout
= ALGORITHM_RIGHT_SYMMETRIC
;
5385 case ALGORITHM_PARITY_0_6
:
5386 new_layout
= ALGORITHM_PARITY_0
;
5388 case ALGORITHM_PARITY_N
:
5389 new_layout
= ALGORITHM_PARITY_N
;
5392 return ERR_PTR(-EINVAL
);
5394 mddev
->new_level
= 5;
5395 mddev
->new_layout
= new_layout
;
5396 mddev
->delta_disks
= -1;
5397 mddev
->raid_disks
-= 1;
5398 return setup_conf(mddev
);
5402 static int raid5_check_reshape(mddev_t
*mddev
)
5404 /* For a 2-drive array, the layout and chunk size can be changed
5405 * immediately as not restriping is needed.
5406 * For larger arrays we record the new value - after validation
5407 * to be used by a reshape pass.
5409 raid5_conf_t
*conf
= mddev
->private;
5410 int new_chunk
= mddev
->new_chunk_sectors
;
5412 if (mddev
->new_layout
>= 0 && !algorithm_valid_raid5(mddev
->new_layout
))
5414 if (new_chunk
> 0) {
5415 if (!is_power_of_2(new_chunk
))
5417 if (new_chunk
< (PAGE_SIZE
>>9))
5419 if (mddev
->array_sectors
& (new_chunk
-1))
5420 /* not factor of array size */
5424 /* They look valid */
5426 if (mddev
->raid_disks
== 2) {
5427 /* can make the change immediately */
5428 if (mddev
->new_layout
>= 0) {
5429 conf
->algorithm
= mddev
->new_layout
;
5430 mddev
->layout
= mddev
->new_layout
;
5432 if (new_chunk
> 0) {
5433 conf
->chunk_sectors
= new_chunk
;
5434 mddev
->chunk_sectors
= new_chunk
;
5436 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
5437 md_wakeup_thread(mddev
->thread
);
5439 return check_reshape(mddev
);
5442 static int raid6_check_reshape(mddev_t
*mddev
)
5444 int new_chunk
= mddev
->new_chunk_sectors
;
5446 if (mddev
->new_layout
>= 0 && !algorithm_valid_raid6(mddev
->new_layout
))
5448 if (new_chunk
> 0) {
5449 if (!is_power_of_2(new_chunk
))
5451 if (new_chunk
< (PAGE_SIZE
>> 9))
5453 if (mddev
->array_sectors
& (new_chunk
-1))
5454 /* not factor of array size */
5458 /* They look valid */
5459 return check_reshape(mddev
);
5462 static void *raid5_takeover(mddev_t
*mddev
)
5464 /* raid5 can take over:
5465 * raid0 - if there is only one strip zone - make it a raid4 layout
5466 * raid1 - if there are two drives. We need to know the chunk size
5467 * raid4 - trivial - just use a raid4 layout.
5468 * raid6 - Providing it is a *_6 layout
5470 if (mddev
->level
== 0)
5471 return raid45_takeover_raid0(mddev
, 5);
5472 if (mddev
->level
== 1)
5473 return raid5_takeover_raid1(mddev
);
5474 if (mddev
->level
== 4) {
5475 mddev
->new_layout
= ALGORITHM_PARITY_N
;
5476 mddev
->new_level
= 5;
5477 return setup_conf(mddev
);
5479 if (mddev
->level
== 6)
5480 return raid5_takeover_raid6(mddev
);
5482 return ERR_PTR(-EINVAL
);
5485 static void *raid4_takeover(mddev_t
*mddev
)
5487 /* raid4 can take over:
5488 * raid0 - if there is only one strip zone
5489 * raid5 - if layout is right
5491 if (mddev
->level
== 0)
5492 return raid45_takeover_raid0(mddev
, 4);
5493 if (mddev
->level
== 5 &&
5494 mddev
->layout
== ALGORITHM_PARITY_N
) {
5495 mddev
->new_layout
= 0;
5496 mddev
->new_level
= 4;
5497 return setup_conf(mddev
);
5499 return ERR_PTR(-EINVAL
);
5502 static struct mdk_personality raid5_personality
;
5504 static void *raid6_takeover(mddev_t
*mddev
)
5506 /* Currently can only take over a raid5. We map the
5507 * personality to an equivalent raid6 personality
5508 * with the Q block at the end.
5512 if (mddev
->pers
!= &raid5_personality
)
5513 return ERR_PTR(-EINVAL
);
5514 if (mddev
->degraded
> 1)
5515 return ERR_PTR(-EINVAL
);
5516 if (mddev
->raid_disks
> 253)
5517 return ERR_PTR(-EINVAL
);
5518 if (mddev
->raid_disks
< 3)
5519 return ERR_PTR(-EINVAL
);
5521 switch (mddev
->layout
) {
5522 case ALGORITHM_LEFT_ASYMMETRIC
:
5523 new_layout
= ALGORITHM_LEFT_ASYMMETRIC_6
;
5525 case ALGORITHM_RIGHT_ASYMMETRIC
:
5526 new_layout
= ALGORITHM_RIGHT_ASYMMETRIC_6
;
5528 case ALGORITHM_LEFT_SYMMETRIC
:
5529 new_layout
= ALGORITHM_LEFT_SYMMETRIC_6
;
5531 case ALGORITHM_RIGHT_SYMMETRIC
:
5532 new_layout
= ALGORITHM_RIGHT_SYMMETRIC_6
;
5534 case ALGORITHM_PARITY_0
:
5535 new_layout
= ALGORITHM_PARITY_0_6
;
5537 case ALGORITHM_PARITY_N
:
5538 new_layout
= ALGORITHM_PARITY_N
;
5541 return ERR_PTR(-EINVAL
);
5543 mddev
->new_level
= 6;
5544 mddev
->new_layout
= new_layout
;
5545 mddev
->delta_disks
= 1;
5546 mddev
->raid_disks
+= 1;
5547 return setup_conf(mddev
);
5551 static struct mdk_personality raid6_personality
=
5555 .owner
= THIS_MODULE
,
5556 .make_request
= make_request
,
5560 .error_handler
= error
,
5561 .hot_add_disk
= raid5_add_disk
,
5562 .hot_remove_disk
= raid5_remove_disk
,
5563 .spare_active
= raid5_spare_active
,
5564 .sync_request
= sync_request
,
5565 .resize
= raid5_resize
,
5567 .check_reshape
= raid6_check_reshape
,
5568 .start_reshape
= raid5_start_reshape
,
5569 .finish_reshape
= raid5_finish_reshape
,
5570 .quiesce
= raid5_quiesce
,
5571 .takeover
= raid6_takeover
,
5573 static struct mdk_personality raid5_personality
=
5577 .owner
= THIS_MODULE
,
5578 .make_request
= make_request
,
5582 .error_handler
= error
,
5583 .hot_add_disk
= raid5_add_disk
,
5584 .hot_remove_disk
= raid5_remove_disk
,
5585 .spare_active
= raid5_spare_active
,
5586 .sync_request
= sync_request
,
5587 .resize
= raid5_resize
,
5589 .check_reshape
= raid5_check_reshape
,
5590 .start_reshape
= raid5_start_reshape
,
5591 .finish_reshape
= raid5_finish_reshape
,
5592 .quiesce
= raid5_quiesce
,
5593 .takeover
= raid5_takeover
,
5596 static struct mdk_personality raid4_personality
=
5600 .owner
= THIS_MODULE
,
5601 .make_request
= make_request
,
5605 .error_handler
= error
,
5606 .hot_add_disk
= raid5_add_disk
,
5607 .hot_remove_disk
= raid5_remove_disk
,
5608 .spare_active
= raid5_spare_active
,
5609 .sync_request
= sync_request
,
5610 .resize
= raid5_resize
,
5612 .check_reshape
= raid5_check_reshape
,
5613 .start_reshape
= raid5_start_reshape
,
5614 .finish_reshape
= raid5_finish_reshape
,
5615 .quiesce
= raid5_quiesce
,
5616 .takeover
= raid4_takeover
,
5619 static int __init
raid5_init(void)
5621 register_md_personality(&raid6_personality
);
5622 register_md_personality(&raid5_personality
);
5623 register_md_personality(&raid4_personality
);
5627 static void raid5_exit(void)
5629 unregister_md_personality(&raid6_personality
);
5630 unregister_md_personality(&raid5_personality
);
5631 unregister_md_personality(&raid4_personality
);
5634 module_init(raid5_init
);
5635 module_exit(raid5_exit
);
5636 MODULE_LICENSE("GPL");
5637 MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
5638 MODULE_ALIAS("md-personality-4"); /* RAID5 */
5639 MODULE_ALIAS("md-raid5");
5640 MODULE_ALIAS("md-raid4");
5641 MODULE_ALIAS("md-level-5");
5642 MODULE_ALIAS("md-level-4");
5643 MODULE_ALIAS("md-personality-8"); /* RAID6 */
5644 MODULE_ALIAS("md-raid6");
5645 MODULE_ALIAS("md-level-6");
5647 /* This used to be two separate modules, they were: */
5648 MODULE_ALIAS("raid5");
5649 MODULE_ALIAS("raid6");