Re: [PATCH v2] fs/ntfs3: Update inode->i_mapping->a_ops on compression state

[Dave Chinner <david@fromorbit.com>]

On Fri, Jan 31, 2025 at 04:18:31PM +0300, Konstantin Komarov wrote:
> Update inode->i_mapping->a_ops when the compression state changes to
> ensure correct address space operations.
> Clear ATTR_FLAG_SPARSED/FILE_ATTRIBUTE_SPARSE_FILE when enabling
> compression to prevent flag conflicts.
> 
> v2:
> Additionally, ensure that all dirty pages are flushed and concurrent access
> to the page cache is blocked.
> 
> Fixes: 6b39bfaeec44 ("fs/ntfs3: Add support for the compression attribute")
> Reported-by: Kun Hu <huk23@m.fudan.edu.cn>, Jiaji Qin <jjtan24@m.fudan.edu.cn>
> Signed-off-by: Konstantin Komarov <almaz.alexandrovich@paragon-software.com>
> ---
>  fs/ntfs3/attrib.c  |  3 ++-
>  fs/ntfs3/file.c    | 22 ++++++++++++++++++++--
>  fs/ntfs3/frecord.c |  6 ++++--
>  3 files changed, 26 insertions(+), 5 deletions(-)
> 
> diff --git a/fs/ntfs3/attrib.c b/fs/ntfs3/attrib.c
> index af94e3737470..e946f75eb540 100644
> --- a/fs/ntfs3/attrib.c
> +++ b/fs/ntfs3/attrib.c
> @@ -2664,8 +2664,9 @@ int attr_set_compress(struct ntfs_inode *ni, bool compr)
>  		attr->nres.run_off = cpu_to_le16(run_off);
>  	}
>  
> -	/* Update data attribute flags. */
> +	/* Update attribute flags. */
>  	if (compr) {
> +		attr->flags &= ~ATTR_FLAG_SPARSED;
>  		attr->flags |= ATTR_FLAG_COMPRESSED;
>  		attr->nres.c_unit = NTFS_LZNT_CUNIT;
>  	} else {
> diff --git a/fs/ntfs3/file.c b/fs/ntfs3/file.c
> index 4d9d84cc3c6f..9b6a3f8d2e7c 100644
> --- a/fs/ntfs3/file.c
> +++ b/fs/ntfs3/file.c
> @@ -101,8 +101,26 @@ int ntfs_fileattr_set(struct mnt_idmap *idmap, struct dentry *dentry,
>  	/* Allowed to change compression for empty files and for directories only. */
>  	if (!is_dedup(ni) && !is_encrypted(ni) &&
>  	    (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))) {
> -		/* Change compress state. */
> -		int err = ni_set_compress(inode, flags & FS_COMPR_FL);
> +		int err = 0;
> +		struct address_space *mapping = inode->i_mapping;
> +
> +		/* write out all data and wait. */
> +		filemap_invalidate_lock(mapping);
> +		err = filemap_write_and_wait(mapping);
> +
> +		if (err >= 0) {
> +			/* Change compress state. */
> +			bool compr = flags & FS_COMPR_FL;
> +			err = ni_set_compress(inode, compr);
> +
> +			/* For files change a_ops too. */
> +			if (!err)
> +				mapping->a_ops = compr ? &ntfs_aops_cmpr :
> +							 &ntfs_aops;
> +		}
> +
> +		filemap_invalidate_unlock(mapping);

Holding the invalidate lock doesn't make it safe to change aops
methods. We've been down this road before - find some other way to
switch modes internally to the ntfs filesystem, because changing
aops pointers like this is *not safe* and not maintainable.

-Dave.
-- 
Dave Chinner
david@fromorbit.com