Entropy Source for Openssl 3.8

Wed Aug 30 23:02:14 UTC 2023

Pauli,

Isn’t there a restriction on the entropy source used for FIPS? On x86, the hardware randomizer is acceptable to FIPS, for example, but on ia64, using an arbitrary entropy source that does not quite pass FIPS randomness requirements is not. I went through this trying to build a hardware source outside of the ia64 chip during my port, but it was not random enough. I worry that Manish’s requirement might end up in the same class as my ia64 randomness provider and not pass muster.

On the other hand, there are hardware HKSM devices that are far better than the Intel x86 hardware randomizer that should be supported via providers (and FIPS but I am not aware of how to do that).

Sincerely,

Randall

From: openssl-users <openssl-users-bounces at openssl.org> On Behalf Of Dr Paul Dale
Sent: Wednesday, August 30, 2023 6:51 PM
To: Manish Patidar <mann.patidar at gmail.com>
Cc: openssl-users at openssl.org
Subject: Re: Entropy Source for Openssl 3.8

Manish,

The code there is somewhat confused by the way the FIPS provider gathers it's entropy.
It doesn't access the seed source directly, instead it has call-backs into libcrypto to request entropy.
The critical function is ossl_rand_get_entropy in crypto/rand/prov_seed.c.  This function satisfies the FIPS provider's request for entropy and it doesn't access the seed source specified, instead it goes directly to OpenSSL's internal entropy gathering.

So, no there isn't a way to do what you want.

It wasn't intended to operate this way and I'll look at producing a fix.

Pauli

On 31/8/23 03:59, Manish Patidar wrote:

Hi Paul, 

I have created provider for entropy and  set the seed source using RAND_set_seed_source_type(). This is working fine non-fips mode but its not working in fips mode. 

Look like, entropy provider can't be used in fips mode. 

RAND_bytes_ex - > RAND_get0_public - > RAND_get0_primary :

Code which get entropy from provider is under #ifndef FIPS_MODULE

#ifndef FIPS_MODULE

 If (dgbl->seed == NULL) {

       dgbl->seed = rand_new_seed(ctx) 

} 

#endif

dbgl->primary = rand_new_drbg(ctx, dbgl->seed) 

In non-fips mode, dbgl->seed is not null (act as parent) but in fips mode it is NULL and parent is null. 

Is this correct?

Is there any way to set custom entropy source in fips mode? 

Regards 

Manish 

On Wed, 12 Jul 2023, 5:10 am Dr Paul Dale, <pauli at openssl.org <mailto:pauli at openssl.org> > wrote:

Manish et al,

If you create a provider that supplies an equivalent to EVP_RAND_seed that accesses your hardware entropy source, you can use this for both the FIPS and non-FIPS case.  Call RAND_set_seed_source_type() early on to use your seed source instead of the built in ones.  Alternatively, this can be done via configuration (search for "seed" in the config(5) man page for details).

Paul Dale

On 11/7/23 01:34, Manish Patidar wrote:

Thanks Pauli for your response.
Still I am not clear how to feed the h/w entropy  to openssl. 

For Non-fips mode ,  as per your suggestion, we have two option:

1. RAND_set_rand_method/RAND_set_rand_engine 

    Need implement method using EVP_RAND (EVP_RAND_fetch , EVP_RAND_CTX_new, EVP_RAND_instantiate,  EVP_RAND_seed, EVP_RAND_generate) 

    and set the method(RAND_set_rand_method)  to openssl 

    How to feed h/w entropy, using  EVP_RAND_seed ?

2. Write a new provider, if we can't use the above method
    We can refer to test/testutil/fake_random.c ?
    Internally, can we use EVP_RAND (AES_CTR DRBG)  in provider implementation ?

For Fips mode:

    We need to use AES_CTR drbg,  how to provide hardware entropy to FIPS providers.
    How to replace "seed" source, can you please provide more details. ?

Is it possible to have a common solution for both FIPS and non-FIPS mode ? 

Regards

Manish

On Mon, Jun 26, 2023 at 3:18 AM Dr Paul Dale <pauli at openssl.org <mailto:pauli at openssl.org> > wrote:

Both RAND_set_rand_method and RAND_set_rand_engine exist in 3.0.8. They 
are deprecated but I doubt they'll be removed for a long time -- per our 
policies, they won't be before OpenSSL 4.0 is released.

If you really want to avoid these two, you will have to write a provider 
that implements access to the entropy source.  You can then use this 
provider instead of OpenSSL's default sources.  I suggest looking at the 
"test" and "seed" randoms.

For FIPS usage, it would be easiest to replace the "seed" source and 
this is outside the FIPS boundary.  If you RNG is FIPS validated, it 
should be possible to use it directly, although the path is more complex.

Pauli

On 25/6/23 07:34, Manish Patidar wrote:
> Hi
> I am using Openssl 3.8 on rtos,  we have harware random entropy source 
> for RNG.   In our env, Openssl used entropy source is not available.
>
> Look like  entropy callback which used to available in earlier 
> versions, is no more supported.  I am wondering how to plungin 
> hardware entropy to Openssl.
>
> We are going to use h/w entropy in fips mode also, so we need solution 
> which works for both mode.
>
> It will be really helpful if someone guide how to use h/w entropy 
> source in openssl 3.8
>
> Regards
> Manish

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mta.openssl.org/pipermail/openssl-users/attachments/20230830/51b53df8/attachment-0001.htm>