Re: [PATCH 3/5] Btrfs: pause/recover the space balance when doing remount
From: Ilya Dryomov <idryomov@gmail.com>
Date: 2012-05-17 15:08:31
On Thu, May 17, 2012 at 07:57:40PM +0800, Miao Xie wrote:
quoted hunk ↗ jump to hunk
pause the space balance threads when remounting the fs to be readonly, and recover it when remounting it from r/o to r/w Signed-off-by: Miao Xie <redacted> --- fs/btrfs/super.c | 9 ++++++++- fs/btrfs/volumes.c | 8 +++++++- 2 files changed, 15 insertions(+), 2 deletions(-)diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c index 7deb00e..ea17f0a 100644 --- a/fs/btrfs/super.c +++ b/fs/btrfs/super.c@@ -1148,6 +1148,9 @@ static int btrfs_remount(struct super_block *sb, int *flags, char *data) if (*flags & MS_RDONLY) { sb->s_flags |= MS_RDONLY; + /* pause restriper - we want to resume on remount to r/w */ + btrfs_pause_balance(root->fs_info); + ret = btrfs_commit_super(root); if (ret) goto restore;@@ -1174,7 +1177,10 @@ static int btrfs_remount(struct super_block *sb, int *flags, char *data) if (ret) goto restore; - sb->s_flags &= ~MS_RDONLY; + if (sb->s_flags & MS_RDONLY) { + sb->s_flags &= ~MS_RDONLY; + btrfs_recover_balance(fs_info->tree_root); + } } return 0;@@ -1190,6 +1196,7 @@ restore: fs_info->alloc_start = old_alloc_start; fs_info->thread_pool_size = old_thread_pool_size; fs_info->metadata_ratio = old_metadata_ratio; + return ret; }diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c index 91da8a2..c536d52 100644 --- a/fs/btrfs/volumes.c +++ b/fs/btrfs/volumes.c@@ -2833,7 +2833,13 @@ static int balance_kthread(void *data) mutex_lock(&fs_info->volume_mutex); mutex_lock(&fs_info->balance_mutex); - set_balance_control(bctl); + if (fs_info->balance_ctl) { + kfree(bctl); + bctl = fs_info->balance_ctl; + bctl->flags = bctl->flags | BTRFS_BALANCE_RESUME; + } else { + set_balance_control(bctl); + } if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) { printk(KERN_INFO "btrfs: force skipping balance\n");
This is a known bug. There is a deeper problem here, related to the fact that we restore balancing state not early enough and that we don't restore it on ro mounts at all. I have a patch in the works to fix that problem, and it also fixes this one the right way. I'm backed up with other things right now, but I'll post it as soon as I get a chance. Thanks, Ilya