Fine to me.
But another question is, now it allows GC to migrate blocks belong
to atomic files, so, during migration, it may update extent cache,
once largest extent was updated, it will mark inode dirty, but after
this patch, it may lose the extent change? thoughts?
Oh, I missed that case. Maybe we could prevent updating the i_size of
atomic files in f2fs_update_inode() while allowing inode dirtying.