Re: [PATCH] f2fs: fix to call invalidate_mapping_pages in f2fs_move_file_range

[Chao Yu]

On 2023/5/5 2:52, Yangtao Li wrote:
> In the following scenario, after executing the move_range ioctl syscall,
> the block size of the source file is 0, but data can still be read.
>
>    # stat test
>    File: test
>    Size: 6               Blocks: 8          IO Block: 4096   regular file
>    # ./new_f2fs_io move_range test test_move_range 0 0 0
>    move range ret=0
>    # stat test
>    File: test
>    Size: 6               Blocks: 0          IO Block: 4096   regular file
>    # cat test
>    nihao
>
> Let's fix to call invalidate_mapping_pages() after __exchange_data_block()
> success.
>
> Fixes: 4dd6f977fc77 ("f2fs: support an ioctl to move a range of data blocks")
> Signed-off-by: Yangtao Li <frank.li@xxxxxxxx>
> ---
>   fs/f2fs/file.c | 3 +++
>   1 file changed, 3 insertions(+)
>
> diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
> index 78aa8cff4b41..ae7752c5cd0a 100644
> --- a/fs/f2fs/file.c
> +++ b/fs/f2fs/file.c
> @@ -2870,6 +2870,9 @@ static int f2fs_move_file_range(struct file *file_in, loff_t pos_in,
>   			f2fs_i_size_write(dst, dst_max_i_size);
>   		else if (dst_osize != dst->i_size)
>   			f2fs_i_size_write(dst, dst_osize);
> +
> +		invalidate_mapping_pages(src->i_mapping,
> +				pos_out, pos_in + len);

It needs to consider error cases?

Should we call this to drop page cache of src_inode after __clone_blkaddrs()
for each round exchange in __exchange_data_block()? and also drop page cache
of dst_indoe in roll_back case?

Thanks,

>   	}
>   	f2fs_unlock_op(sbi);
>