Re: [PATCH] fscrypto: make filename crypto functions return 0 on success

On Wed, Sep 14, 2016 at 07:54:41PM -0700, Eric Biggers wrote:

On Wed, Sep 14, 2016 at 05:10:21PM -0700, Jaegeuk Kim wrote:

quoted

Hi Eric,

On Wed, Sep 14, 2016 at 04:00:45PM -0700, Eric Biggers wrote:

quoted

Several filename crypto functions: fname_decrypt(),
fscrypt_fname_disk_to_usr(), and fscrypt_fname_usr_to_disk(), returned
the output length on success or -errno on failure.  However, the output
length was redundant with the value written to 'oname->len'.  It is also
potentially error-prone to make callers have to check for '< 0' instead
of '!= 0'.

At a glance, you missed some places like
- fscrypt_fname_usr_to_disk() in ext4_symlink() and f2fs_symlink()
- fscrypt_fname_disk_to_usr() in dx_show_leaf() and f2fs_fill_dentries()

Those callers don't have to be updated because they only check for an error
value --- they don't use the return value as a length.  I did check all the
callers of fscrypt_fname_{usr_to_disk,disk_to_usr()}, and only the three I
update in the patch used the return value as a length.

Hmm, your patch tries to change the return value and make its usage more
consistent. So, I think it should have been considered to change if (err < 0)
to if (err) entirely in order to avoid any potential mistake; this patch already
addresses it in some cases, but not all.

e.g., in f2fs_symlink,

	err = fscrypt_fname_usr_to_disk(inode, &istr, &ostr);
	if (err < 0)
		goto err_out;

BTW, in dx_show_leaf(),
	res = fscrypt_fname_disk_to_usr(dir,
		0, 0, &de_name,
		&fname_crypto_str);
	if (res < 0)
		...

IMHO, it'd be better to use err instead of {res|ret} when considering naming
convention.

Thanks,

Eric