Re: [PATCH 1/1 linux-next] udf: improve udf_get_filename() error management

[Jan Kara <jack@suse.cz>]

On Mon 16-03-15 21:20:23, Fabian Frederick wrote:
> Let udf_get_filename() return errors instead of 0.
> 
> udf_pc_to_char() now returns error accordingly.
> udf_readdir() and udf_find_entry() process is done on
> positive result.
> 
> Note that error propagation has still to be done in the
> later functions.
> 
> Suggested-by: Jan Kara <jack@suse.cz>
> Signed-off-by: Fabian Frederick <fabf@skynet.be>
...
> diff --git a/fs/udf/unicode.c b/fs/udf/unicode.c
> index b84fee3..4911c1d 100644
> --- a/fs/udf/unicode.c
> +++ b/fs/udf/unicode.c
> @@ -338,15 +338,17 @@ int udf_get_filename(struct super_block *sb, uint8_t *sname, int slen,
>  		     uint8_t *dname, int dlen)
>  {
>  	struct ustr *filename, *unifilename;
> -	int len = 0;
> +	int ret = 0;
>  
>  	filename = kmalloc(sizeof(struct ustr), GFP_NOFS);
>  	if (!filename)
> -		return 0;
> +		return -ENOMEM;
>  
>  	unifilename = kmalloc(sizeof(struct ustr), GFP_NOFS);
> -	if (!unifilename)
> +	if (!unifilename) {
> +		ret = -ENOMEM;
>  		goto out1;
> +	}
>  
>  	if (udf_build_ustr_exact(unifilename, sname, slen))
>  		goto out2;
> @@ -367,14 +369,14 @@ int udf_get_filename(struct super_block *sb, uint8_t *sname, int slen,
>  	} else
>  		goto out2;
>  
> -	len = udf_translate_to_linux(dname, dlen,
> +	ret = udf_translate_to_linux(dname, dlen,
>  				     filename->u_name, filename->u_len,
>  				     unifilename->u_name, unifilename->u_len);
>  out2:
>  	kfree(unifilename);
>  out1:
>  	kfree(filename);
> -	return len;
> +	return ret;
>  }
  I think you need to do more work around udf_get_filename() if you want
error handling to function correctly.  The test in udf_build_ustr_exact()
which can make it fail is just bogus - remove it. udf_CS0toUTF8() and
udf_CS0toNLS() may fail as well - make them return error (-EINVAL?) and
propagate that up. The else branch in udf_get_filename() should be a BUG()
- mount code makes sure one of UDF_FLAG_NLS_MAP and UDF_FLAG_UTF8 is set.
That should deal with all the error paths so we really return value < 0 if
we fail and positive value if we succeed.

								Honza
-- 
Jan Kara <jack@suse.cz>
SUSE Labs, CR