diff options
Diffstat (limited to 'drivers/md/raid10.c')
-rw-r--r-- | drivers/md/raid10.c | 87 |
1 files changed, 19 insertions, 68 deletions
diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c index 3b607b28741b..e79f1c5bf71b 100644 --- a/drivers/md/raid10.c +++ b/drivers/md/raid10.c | |||
@@ -57,23 +57,16 @@ | |||
57 | */ | 57 | */ |
58 | #define NR_RAID10_BIOS 256 | 58 | #define NR_RAID10_BIOS 256 |
59 | 59 | ||
60 | static void unplug_slaves(mddev_t *mddev); | ||
61 | |||
62 | static void allow_barrier(conf_t *conf); | 60 | static void allow_barrier(conf_t *conf); |
63 | static void lower_barrier(conf_t *conf); | 61 | static void lower_barrier(conf_t *conf); |
64 | 62 | ||
65 | static void * r10bio_pool_alloc(gfp_t gfp_flags, void *data) | 63 | static void * r10bio_pool_alloc(gfp_t gfp_flags, void *data) |
66 | { | 64 | { |
67 | conf_t *conf = data; | 65 | conf_t *conf = data; |
68 | r10bio_t *r10_bio; | ||
69 | int size = offsetof(struct r10bio_s, devs[conf->copies]); | 66 | int size = offsetof(struct r10bio_s, devs[conf->copies]); |
70 | 67 | ||
71 | /* allocate a r10bio with room for raid_disks entries in the bios array */ | 68 | /* allocate a r10bio with room for raid_disks entries in the bios array */ |
72 | r10_bio = kzalloc(size, gfp_flags); | 69 | return kzalloc(size, gfp_flags); |
73 | if (!r10_bio && conf->mddev) | ||
74 | unplug_slaves(conf->mddev); | ||
75 | |||
76 | return r10_bio; | ||
77 | } | 70 | } |
78 | 71 | ||
79 | static void r10bio_pool_free(void *r10_bio, void *data) | 72 | static void r10bio_pool_free(void *r10_bio, void *data) |
@@ -106,10 +99,8 @@ static void * r10buf_pool_alloc(gfp_t gfp_flags, void *data) | |||
106 | int nalloc; | 99 | int nalloc; |
107 | 100 | ||
108 | r10_bio = r10bio_pool_alloc(gfp_flags, conf); | 101 | r10_bio = r10bio_pool_alloc(gfp_flags, conf); |
109 | if (!r10_bio) { | 102 | if (!r10_bio) |
110 | unplug_slaves(conf->mddev); | ||
111 | return NULL; | 103 | return NULL; |
112 | } | ||
113 | 104 | ||
114 | if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery)) | 105 | if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery)) |
115 | nalloc = conf->copies; /* resync */ | 106 | nalloc = conf->copies; /* resync */ |
@@ -597,37 +588,6 @@ rb_out: | |||
597 | return disk; | 588 | return disk; |
598 | } | 589 | } |
599 | 590 | ||
600 | static void unplug_slaves(mddev_t *mddev) | ||
601 | { | ||
602 | conf_t *conf = mddev->private; | ||
603 | int i; | ||
604 | |||
605 | rcu_read_lock(); | ||
606 | for (i=0; i < conf->raid_disks; i++) { | ||
607 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); | ||
608 | if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) { | ||
609 | struct request_queue *r_queue = bdev_get_queue(rdev->bdev); | ||
610 | |||
611 | atomic_inc(&rdev->nr_pending); | ||
612 | rcu_read_unlock(); | ||
613 | |||
614 | blk_unplug(r_queue); | ||
615 | |||
616 | rdev_dec_pending(rdev, mddev); | ||
617 | rcu_read_lock(); | ||
618 | } | ||
619 | } | ||
620 | rcu_read_unlock(); | ||
621 | } | ||
622 | |||
623 | static void raid10_unplug(struct request_queue *q) | ||
624 | { | ||
625 | mddev_t *mddev = q->queuedata; | ||
626 | |||
627 | unplug_slaves(q->queuedata); | ||
628 | md_wakeup_thread(mddev->thread); | ||
629 | } | ||
630 | |||
631 | static int raid10_congested(void *data, int bits) | 591 | static int raid10_congested(void *data, int bits) |
632 | { | 592 | { |
633 | mddev_t *mddev = data; | 593 | mddev_t *mddev = data; |
@@ -649,20 +609,16 @@ static int raid10_congested(void *data, int bits) | |||
649 | return ret; | 609 | return ret; |
650 | } | 610 | } |
651 | 611 | ||
652 | static int flush_pending_writes(conf_t *conf) | 612 | static void flush_pending_writes(conf_t *conf) |
653 | { | 613 | { |
654 | /* Any writes that have been queued but are awaiting | 614 | /* Any writes that have been queued but are awaiting |
655 | * bitmap updates get flushed here. | 615 | * bitmap updates get flushed here. |
656 | * We return 1 if any requests were actually submitted. | ||
657 | */ | 616 | */ |
658 | int rv = 0; | ||
659 | |||
660 | spin_lock_irq(&conf->device_lock); | 617 | spin_lock_irq(&conf->device_lock); |
661 | 618 | ||
662 | if (conf->pending_bio_list.head) { | 619 | if (conf->pending_bio_list.head) { |
663 | struct bio *bio; | 620 | struct bio *bio; |
664 | bio = bio_list_get(&conf->pending_bio_list); | 621 | bio = bio_list_get(&conf->pending_bio_list); |
665 | blk_remove_plug(conf->mddev->queue); | ||
666 | spin_unlock_irq(&conf->device_lock); | 622 | spin_unlock_irq(&conf->device_lock); |
667 | /* flush any pending bitmap writes to disk | 623 | /* flush any pending bitmap writes to disk |
668 | * before proceeding w/ I/O */ | 624 | * before proceeding w/ I/O */ |
@@ -674,11 +630,16 @@ static int flush_pending_writes(conf_t *conf) | |||
674 | generic_make_request(bio); | 630 | generic_make_request(bio); |
675 | bio = next; | 631 | bio = next; |
676 | } | 632 | } |
677 | rv = 1; | ||
678 | } else | 633 | } else |
679 | spin_unlock_irq(&conf->device_lock); | 634 | spin_unlock_irq(&conf->device_lock); |
680 | return rv; | ||
681 | } | 635 | } |
636 | |||
637 | static void md_kick_device(mddev_t *mddev) | ||
638 | { | ||
639 | blk_flush_plug(current); | ||
640 | md_wakeup_thread(mddev->thread); | ||
641 | } | ||
642 | |||
682 | /* Barriers.... | 643 | /* Barriers.... |
683 | * Sometimes we need to suspend IO while we do something else, | 644 | * Sometimes we need to suspend IO while we do something else, |
684 | * either some resync/recovery, or reconfigure the array. | 645 | * either some resync/recovery, or reconfigure the array. |
@@ -708,8 +669,7 @@ static void raise_barrier(conf_t *conf, int force) | |||
708 | 669 | ||
709 | /* Wait until no block IO is waiting (unless 'force') */ | 670 | /* Wait until no block IO is waiting (unless 'force') */ |
710 | wait_event_lock_irq(conf->wait_barrier, force || !conf->nr_waiting, | 671 | wait_event_lock_irq(conf->wait_barrier, force || !conf->nr_waiting, |
711 | conf->resync_lock, | 672 | conf->resync_lock, md_kick_device(conf->mddev)); |
712 | raid10_unplug(conf->mddev->queue)); | ||
713 | 673 | ||
714 | /* block any new IO from starting */ | 674 | /* block any new IO from starting */ |
715 | conf->barrier++; | 675 | conf->barrier++; |
@@ -717,8 +677,7 @@ static void raise_barrier(conf_t *conf, int force) | |||
717 | /* No wait for all pending IO to complete */ | 677 | /* No wait for all pending IO to complete */ |
718 | wait_event_lock_irq(conf->wait_barrier, | 678 | wait_event_lock_irq(conf->wait_barrier, |
719 | !conf->nr_pending && conf->barrier < RESYNC_DEPTH, | 679 | !conf->nr_pending && conf->barrier < RESYNC_DEPTH, |
720 | conf->resync_lock, | 680 | conf->resync_lock, md_kick_device(conf->mddev)); |
721 | raid10_unplug(conf->mddev->queue)); | ||
722 | 681 | ||
723 | spin_unlock_irq(&conf->resync_lock); | 682 | spin_unlock_irq(&conf->resync_lock); |
724 | } | 683 | } |
@@ -739,7 +698,7 @@ static void wait_barrier(conf_t *conf) | |||
739 | conf->nr_waiting++; | 698 | conf->nr_waiting++; |
740 | wait_event_lock_irq(conf->wait_barrier, !conf->barrier, | 699 | wait_event_lock_irq(conf->wait_barrier, !conf->barrier, |
741 | conf->resync_lock, | 700 | conf->resync_lock, |
742 | raid10_unplug(conf->mddev->queue)); | 701 | md_kick_device(conf->mddev)); |
743 | conf->nr_waiting--; | 702 | conf->nr_waiting--; |
744 | } | 703 | } |
745 | conf->nr_pending++; | 704 | conf->nr_pending++; |
@@ -776,7 +735,7 @@ static void freeze_array(conf_t *conf) | |||
776 | conf->nr_pending == conf->nr_queued+1, | 735 | conf->nr_pending == conf->nr_queued+1, |
777 | conf->resync_lock, | 736 | conf->resync_lock, |
778 | ({ flush_pending_writes(conf); | 737 | ({ flush_pending_writes(conf); |
779 | raid10_unplug(conf->mddev->queue); })); | 738 | md_kick_device(conf->mddev); })); |
780 | spin_unlock_irq(&conf->resync_lock); | 739 | spin_unlock_irq(&conf->resync_lock); |
781 | } | 740 | } |
782 | 741 | ||
@@ -971,7 +930,6 @@ static int make_request(mddev_t *mddev, struct bio * bio) | |||
971 | atomic_inc(&r10_bio->remaining); | 930 | atomic_inc(&r10_bio->remaining); |
972 | spin_lock_irqsave(&conf->device_lock, flags); | 931 | spin_lock_irqsave(&conf->device_lock, flags); |
973 | bio_list_add(&conf->pending_bio_list, mbio); | 932 | bio_list_add(&conf->pending_bio_list, mbio); |
974 | blk_plug_device(mddev->queue); | ||
975 | spin_unlock_irqrestore(&conf->device_lock, flags); | 933 | spin_unlock_irqrestore(&conf->device_lock, flags); |
976 | } | 934 | } |
977 | 935 | ||
@@ -988,7 +946,7 @@ static int make_request(mddev_t *mddev, struct bio * bio) | |||
988 | /* In case raid10d snuck in to freeze_array */ | 946 | /* In case raid10d snuck in to freeze_array */ |
989 | wake_up(&conf->wait_barrier); | 947 | wake_up(&conf->wait_barrier); |
990 | 948 | ||
991 | if (do_sync) | 949 | if (do_sync || !mddev->bitmap) |
992 | md_wakeup_thread(mddev->thread); | 950 | md_wakeup_thread(mddev->thread); |
993 | 951 | ||
994 | return 0; | 952 | return 0; |
@@ -1681,7 +1639,6 @@ static void raid10d(mddev_t *mddev) | |||
1681 | unsigned long flags; | 1639 | unsigned long flags; |
1682 | conf_t *conf = mddev->private; | 1640 | conf_t *conf = mddev->private; |
1683 | struct list_head *head = &conf->retry_list; | 1641 | struct list_head *head = &conf->retry_list; |
1684 | int unplug=0; | ||
1685 | mdk_rdev_t *rdev; | 1642 | mdk_rdev_t *rdev; |
1686 | 1643 | ||
1687 | md_check_recovery(mddev); | 1644 | md_check_recovery(mddev); |
@@ -1689,7 +1646,7 @@ static void raid10d(mddev_t *mddev) | |||
1689 | for (;;) { | 1646 | for (;;) { |
1690 | char b[BDEVNAME_SIZE]; | 1647 | char b[BDEVNAME_SIZE]; |
1691 | 1648 | ||
1692 | unplug += flush_pending_writes(conf); | 1649 | flush_pending_writes(conf); |
1693 | 1650 | ||
1694 | spin_lock_irqsave(&conf->device_lock, flags); | 1651 | spin_lock_irqsave(&conf->device_lock, flags); |
1695 | if (list_empty(head)) { | 1652 | if (list_empty(head)) { |
@@ -1703,13 +1660,11 @@ static void raid10d(mddev_t *mddev) | |||
1703 | 1660 | ||
1704 | mddev = r10_bio->mddev; | 1661 | mddev = r10_bio->mddev; |
1705 | conf = mddev->private; | 1662 | conf = mddev->private; |
1706 | if (test_bit(R10BIO_IsSync, &r10_bio->state)) { | 1663 | if (test_bit(R10BIO_IsSync, &r10_bio->state)) |
1707 | sync_request_write(mddev, r10_bio); | 1664 | sync_request_write(mddev, r10_bio); |
1708 | unplug = 1; | 1665 | else if (test_bit(R10BIO_IsRecover, &r10_bio->state)) |
1709 | } else if (test_bit(R10BIO_IsRecover, &r10_bio->state)) { | ||
1710 | recovery_request_write(mddev, r10_bio); | 1666 | recovery_request_write(mddev, r10_bio); |
1711 | unplug = 1; | 1667 | else { |
1712 | } else { | ||
1713 | int mirror; | 1668 | int mirror; |
1714 | /* we got a read error. Maybe the drive is bad. Maybe just | 1669 | /* we got a read error. Maybe the drive is bad. Maybe just |
1715 | * the block and we can fix it. | 1670 | * the block and we can fix it. |
@@ -1756,14 +1711,11 @@ static void raid10d(mddev_t *mddev) | |||
1756 | bio->bi_rw = READ | do_sync; | 1711 | bio->bi_rw = READ | do_sync; |
1757 | bio->bi_private = r10_bio; | 1712 | bio->bi_private = r10_bio; |
1758 | bio->bi_end_io = raid10_end_read_request; | 1713 | bio->bi_end_io = raid10_end_read_request; |
1759 | unplug = 1; | ||
1760 | generic_make_request(bio); | 1714 | generic_make_request(bio); |
1761 | } | 1715 | } |
1762 | } | 1716 | } |
1763 | cond_resched(); | 1717 | cond_resched(); |
1764 | } | 1718 | } |
1765 | if (unplug) | ||
1766 | unplug_slaves(mddev); | ||
1767 | } | 1719 | } |
1768 | 1720 | ||
1769 | 1721 | ||
@@ -2376,7 +2328,6 @@ static int run(mddev_t *mddev) | |||
2376 | md_set_array_sectors(mddev, size); | 2328 | md_set_array_sectors(mddev, size); |
2377 | mddev->resync_max_sectors = size; | 2329 | mddev->resync_max_sectors = size; |
2378 | 2330 | ||
2379 | mddev->queue->unplug_fn = raid10_unplug; | ||
2380 | mddev->queue->backing_dev_info.congested_fn = raid10_congested; | 2331 | mddev->queue->backing_dev_info.congested_fn = raid10_congested; |
2381 | mddev->queue->backing_dev_info.congested_data = mddev; | 2332 | mddev->queue->backing_dev_info.congested_data = mddev; |
2382 | 2333 | ||