Re: [PATCH] f2fs: keep pinned fallocate failures section-aligned

[Chao Yu <chao@kernel.org>]

On 7/23/26 21:36, Wenjie Qi wrote:
> Pinned fallocate allocates one section at a time. If f2fs_map_blocks()
> fails after allocating part of the current section, the error path still
> adds the partial length to expanded and clears FADVISE_TRUNC_BIT. A failed
> fallocate can then leave a pinned file with a non-section-aligned size.

Do you have a reproducer? maybe inject memory allocation failure?

Thanks,

> 
> Count only complete sections on failure and keep FADVISE_TRUNC_BIT set
> until the failed section preallocation has been removed.
> 
> Fixes: f5a53edcf01e ("f2fs: support aligned pinned file")
> Cc: stable@kernel.org
> Signed-off-by: Wenjie Qi <qiwenjie@xiaomi.com>
> ---
>   fs/f2fs/file.c | 17 ++++++++++++++---
>   1 file changed, 14 insertions(+), 3 deletions(-)
> 
> diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
> index c54897a25981..83af73f3831e 100644
> --- a/fs/f2fs/file.c
> +++ b/fs/f2fs/file.c
> @@ -1966,11 +1966,12 @@ static int f2fs_expand_inode_data(struct inode *inode, loff_t offset,
>   
>   		map.m_seg_type = CURSEG_COLD_DATA_PINNED;
>   		err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_PRE_DIO);
> -		file_dont_truncate(inode);
> +		if (!err)
> +			file_dont_truncate(inode);
>   
>   		f2fs_up_write(&sbi->pin_sem);
>   
> -		expanded += map.m_len;
> +		expanded += err ? rounddown(map.m_len, sec_blks) : map.m_len;
>   		sec_len -= map.m_len;
>   		map.m_lblk += map.m_len;
>   		if (!err && sec_len)
> @@ -1986,7 +1987,7 @@ static int f2fs_expand_inode_data(struct inode *inode, loff_t offset,
>   		pgoff_t last_off;
>   
>   		if (!expanded)
> -			return err;
> +			goto out;
>   
>   		last_off = pg_start + expanded - 1;
>   
> @@ -2004,6 +2005,16 @@ static int f2fs_expand_inode_data(struct inode *inode, loff_t offset,
>   			f2fs_i_size_write(inode, new_size);
>   	}
>   
> +out:
> +	if (err && f2fs_is_pinned_file(inode) && file_should_truncate(inode)) {
> +		f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
> +		filemap_invalidate_lock(inode->i_mapping);
> +		if (!f2fs_truncate(inode))
> +			file_dont_truncate(inode);
> +		filemap_invalidate_unlock(inode->i_mapping);
> +		f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
> +	}
> +
>   	return err;
>   }
>