Mimi,
On Mon, Feb 11, 2019 at 03:21:03PM -0500, Mimi Zohar wrote:
> On Mon, 2019-02-11 at 22:26 +0300, Vitaly Chikunov wrote:
> > On Mon, Feb 11, 2019 at 09:21:15PM +0300, Vitaly Chikunov wrote:
> > > On Mon, Feb 11, 2019 at 09:13:00PM +0300, Vitaly Chikunov wrote:
> > > > On Mon, Feb 11, 2019 at 12:59:12PM -0500, Mimi Zohar wrote:
> > > > > On Mon, 2019-02-11 at 20:52 +0300, Vitaly Chikunov wrote:
> > > > > > On Mon, Feb 11, 2019 at 12:38:58PM -0500, Mimi Zohar wrote:
> > > > > > > On Mon, 2018-12-03 at 06:35 +0300, Vitaly Chikunov wrote:
> > > > > > > > For compatibility with older OpenSSL try to load digest by its alias if
> > > > > > > > load by its proper name is failed.
> > > > > > > >
> > > > > > > > This is configured in pkey_hash_algo by mentioning loadable alias first in the
> > > > > > > > comma separated list of algo names.
> > > > > > >
> > > > > > > After this patch, I can not verify the signature. It's failing to
> > > > > > > find the hash algorithm.
> > > > > >
> > > > > > Yes, it's fixed in "ima-evm-utils: convert sign v2 from RSA to EVP_PKEY
> > > > > > API". I didn't consider it a problem since streebog256 should not work
> > > > > > for sign/verify anyway (since RSA should not support it). That EVP_PKEY
> > > > > > patch which adds ability to sign/verify for Streebog also fixes above
> > > > > > problem.
> > > > >
> > > > > I've been having second thoughts about this patch in general, as it
> > > > > made the hash algorithm comparison unnecessarily more complex for the
> > > > > simple case. As there hasn't been a new ima-evm-utils release with
> > > > > patch, perhaps we should simple remove/revert it?
> > > >
> > > > It was only for compatibility with older openssl/gost-engine, where
> > > > "streebog256" name is not defined yet. If you don't want to allow that
> > > > users to use Streebog feel free to revert it.
> >
> > Or you may suggest simpler approach.
> >
> > Basically, we need to resole both text strings into the same PKEY_HASH_
> > id, allow any of the string pass into EVP_get_digestbyname, and resolve
> > PKEY_HASH_ id back into the correct hash name. Optionally, allow user to
> > specify new hash name on older openssl/gost-engine. This is all
> > implemented by that patch, it was not just overly complicated "hash
> > algorithm comparison".
> >
> > My wish is we retain support of older openssl/gost-engine.
>
> The following seems to fix the problem, but instead of adding it in
> strmatch(), as below, I would add it before the strmatch() call. Then
> strmatch is only called as needed.
IC. I will try to conceptually simplify the code and prepare another
patch. Probably, I will split pkey_hash_algop[] into two arrays, simple
one where strcmp() in the loop is used, and complex one for hash names
with aliases. I will try to replace strmatch() with just simple string
array loop.
>
> diff --git a/src/libimaevm.c b/src/libimaevm.c
> index d9ffa13befb0..7901215da655 100644
> --- a/src/libimaevm.c
> +++ b/src/libimaevm.c
> @@ -598,6 +598,9 @@ static int strmatch(const char *needle, const char *haystack)
> const char *p = haystack;
> const char *t;
>
> + if (strcmp(needle, haystack) == 0)
> + return 0;
> +
> for (t = strchrnul(p, ','); *p; p = t, t = strchrnul(p, ',')) {
> if (t - p == len &&
> !strncmp(needle, p, len))
>
> Mimi
