diff options
author | NeilBrown <neilb@suse.de> | 2006-03-27 04:18:12 -0500 |
---|---|---|
committer | Linus Torvalds <torvalds@g5.osdl.org> | 2006-03-27 11:45:02 -0500 |
commit | b578d55fdd80140f657130abd85aebeb345755fb (patch) | |
tree | 3737b88a25b3fb1e3347cd1c4c274336aee6a462 /drivers/md/raid5.c | |
parent | f67055780caac6a99f43834795c43acf99eba6a6 (diff) |
[PATCH] md: Only checkpoint expansion progress occasionally
Instead of checkpointing at each stripe, only checkpoint when a new write
would overwrite uncheckpointed data. Block any write to the uncheckpointed
area. Arbitrarily checkpoint at least every 3Meg.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Diffstat (limited to 'drivers/md/raid5.c')
-rw-r--r-- | drivers/md/raid5.c | 53 |
1 files changed, 42 insertions, 11 deletions
diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c index 20ae32d67e21..089a32604305 100644 --- a/drivers/md/raid5.c +++ b/drivers/md/raid5.c | |||
@@ -1762,8 +1762,9 @@ static int make_request(request_queue_t *q, struct bio * bi) | |||
1762 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { | 1762 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
1763 | DEFINE_WAIT(w); | 1763 | DEFINE_WAIT(w); |
1764 | int disks; | 1764 | int disks; |
1765 | 1765 | ||
1766 | retry: | 1766 | retry: |
1767 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); | ||
1767 | if (likely(conf->expand_progress == MaxSector)) | 1768 | if (likely(conf->expand_progress == MaxSector)) |
1768 | disks = conf->raid_disks; | 1769 | disks = conf->raid_disks; |
1769 | else { | 1770 | else { |
@@ -1771,6 +1772,13 @@ static int make_request(request_queue_t *q, struct bio * bi) | |||
1771 | disks = conf->raid_disks; | 1772 | disks = conf->raid_disks; |
1772 | if (logical_sector >= conf->expand_progress) | 1773 | if (logical_sector >= conf->expand_progress) |
1773 | disks = conf->previous_raid_disks; | 1774 | disks = conf->previous_raid_disks; |
1775 | else { | ||
1776 | if (logical_sector >= conf->expand_lo) { | ||
1777 | spin_unlock_irq(&conf->device_lock); | ||
1778 | schedule(); | ||
1779 | goto retry; | ||
1780 | } | ||
1781 | } | ||
1774 | spin_unlock_irq(&conf->device_lock); | 1782 | spin_unlock_irq(&conf->device_lock); |
1775 | } | 1783 | } |
1776 | new_sector = raid5_compute_sector(logical_sector, disks, disks - 1, | 1784 | new_sector = raid5_compute_sector(logical_sector, disks, disks - 1, |
@@ -1779,7 +1787,6 @@ static int make_request(request_queue_t *q, struct bio * bi) | |||
1779 | (unsigned long long)new_sector, | 1787 | (unsigned long long)new_sector, |
1780 | (unsigned long long)logical_sector); | 1788 | (unsigned long long)logical_sector); |
1781 | 1789 | ||
1782 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); | ||
1783 | sh = get_active_stripe(conf, new_sector, disks, pd_idx, (bi->bi_rw&RWA_MASK)); | 1790 | sh = get_active_stripe(conf, new_sector, disks, pd_idx, (bi->bi_rw&RWA_MASK)); |
1784 | if (sh) { | 1791 | if (sh) { |
1785 | if (unlikely(conf->expand_progress != MaxSector)) { | 1792 | if (unlikely(conf->expand_progress != MaxSector)) { |
@@ -1877,6 +1884,7 @@ static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, i | |||
1877 | */ | 1884 | */ |
1878 | int i; | 1885 | int i; |
1879 | int dd_idx; | 1886 | int dd_idx; |
1887 | sector_t writepos, safepos, gap; | ||
1880 | 1888 | ||
1881 | if (sector_nr == 0 && | 1889 | if (sector_nr == 0 && |
1882 | conf->expand_progress != 0) { | 1890 | conf->expand_progress != 0) { |
@@ -1887,15 +1895,36 @@ static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, i | |||
1887 | return sector_nr; | 1895 | return sector_nr; |
1888 | } | 1896 | } |
1889 | 1897 | ||
1890 | /* Cannot proceed until we've updated the superblock... */ | 1898 | /* we update the metadata when there is more than 3Meg |
1891 | wait_event(conf->wait_for_overlap, | 1899 | * in the block range (that is rather arbitrary, should |
1892 | atomic_read(&conf->reshape_stripes)==0); | 1900 | * probably be time based) or when the data about to be |
1893 | mddev->reshape_position = conf->expand_progress; | 1901 | * copied would over-write the source of the data at |
1894 | 1902 | * the front of the range. | |
1895 | mddev->sb_dirty = 1; | 1903 | * i.e. one new_stripe forward from expand_progress new_maps |
1896 | md_wakeup_thread(mddev->thread); | 1904 | * to after where expand_lo old_maps to |
1897 | wait_event(mddev->sb_wait, mddev->sb_dirty == 0 || | 1905 | */ |
1898 | kthread_should_stop()); | 1906 | writepos = conf->expand_progress + |
1907 | conf->chunk_size/512*(conf->raid_disks-1); | ||
1908 | sector_div(writepos, conf->raid_disks-1); | ||
1909 | safepos = conf->expand_lo; | ||
1910 | sector_div(safepos, conf->previous_raid_disks-1); | ||
1911 | gap = conf->expand_progress - conf->expand_lo; | ||
1912 | |||
1913 | if (writepos >= safepos || | ||
1914 | gap > (conf->raid_disks-1)*3000*2 /*3Meg*/) { | ||
1915 | /* Cannot proceed until we've updated the superblock... */ | ||
1916 | wait_event(conf->wait_for_overlap, | ||
1917 | atomic_read(&conf->reshape_stripes)==0); | ||
1918 | mddev->reshape_position = conf->expand_progress; | ||
1919 | mddev->sb_dirty = 1; | ||
1920 | md_wakeup_thread(mddev->thread); | ||
1921 | wait_event(mddev->sb_wait, mddev->sb_dirty == 0 || | ||
1922 | kthread_should_stop()); | ||
1923 | spin_lock_irq(&conf->device_lock); | ||
1924 | conf->expand_lo = mddev->reshape_position; | ||
1925 | spin_unlock_irq(&conf->device_lock); | ||
1926 | wake_up(&conf->wait_for_overlap); | ||
1927 | } | ||
1899 | 1928 | ||
1900 | for (i=0; i < conf->chunk_size/512; i+= STRIPE_SECTORS) { | 1929 | for (i=0; i < conf->chunk_size/512; i+= STRIPE_SECTORS) { |
1901 | int j; | 1930 | int j; |
@@ -2322,6 +2351,7 @@ static int run(mddev_t *mddev) | |||
2322 | 2351 | ||
2323 | if (conf->expand_progress != MaxSector) { | 2352 | if (conf->expand_progress != MaxSector) { |
2324 | printk("...ok start reshape thread\n"); | 2353 | printk("...ok start reshape thread\n"); |
2354 | conf->expand_lo = conf->expand_progress; | ||
2325 | atomic_set(&conf->reshape_stripes, 0); | 2355 | atomic_set(&conf->reshape_stripes, 0); |
2326 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | 2356 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
2327 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | 2357 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
@@ -2610,6 +2640,7 @@ static int raid5_reshape(mddev_t *mddev, int raid_disks) | |||
2610 | conf->previous_raid_disks = conf->raid_disks; | 2640 | conf->previous_raid_disks = conf->raid_disks; |
2611 | conf->raid_disks = raid_disks; | 2641 | conf->raid_disks = raid_disks; |
2612 | conf->expand_progress = 0; | 2642 | conf->expand_progress = 0; |
2643 | conf->expand_lo = 0; | ||
2613 | spin_unlock_irq(&conf->device_lock); | 2644 | spin_unlock_irq(&conf->device_lock); |
2614 | 2645 | ||
2615 | /* Add some new drives, as many as will fit. | 2646 | /* Add some new drives, as many as will fit. |