Hi!

> From: Chao Yu <yuchao0@huawei.com>
> 
> [ Upstream commit 4a70e255449c9a13eed7a6eeecc85a1ea63cef76 ]
> 
> In fill_super -> init_percpu_info, we should destroy percpu counter
> in error path, otherwise memory allcoated for percpu counter will
> leak.

> diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
> index 58931d55dc1d2..c5d28e92d146e 100644
> --- a/fs/f2fs/super.c
> +++ b/fs/f2fs/super.c
> @@ -2516,8 +2516,12 @@ static int init_percpu_info(struct f2fs_sb_info *sbi)
>  	if (err)
>  		return err;
>  
> -	return percpu_counter_init(&sbi->total_valid_inode_count, 0,
> +	err = percpu_counter_init(&sbi->total_valid_inode_count, 0,
>  								GFP_KERNEL);
> +	if (err)
> +		percpu_counter_destroy(&sbi->alloc_valid_block_count);
> +
> +	return err;
>  }

Are you sure this is good idea? Normally when _init() fails, the thing
is not allocated, and there is nothing to destroy...

If you are right, there's same bug in fs/xfs/xfs_buf.c .

Best regards,
									Pavel
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html