MIME-Version: 1.0
References: <CACXcFmnEpzp-66JvENsAFTTyehM0bZoGFT6wOBPzkkbGgi7P5A@mail.gmail.com>
In-Reply-To: <CACXcFmnEpzp-66JvENsAFTTyehM0bZoGFT6wOBPzkkbGgi7P5A@mail.gmail.com>
From:   Sandy Harris <sandyinchina@gmail.com>
Date:   Tue, 14 Dec 2021 17:23:05 +0800
Message-ID: <CACXcFmnhFZAfjDC802cca9fTxwNOeNLU=F=Y6OiRyLMtoQci-A@mail.gmail.com>
Subject: Re: [RFC] random, initialize pool at compile time
To:     Linux Crypto Mailing List <linux-crypto@vger.kernel.org>,
        "Ted Ts'o" <tytso@mit.edu>, "Jason A. Donenfeld" <Jason@zx2c4.com>,
        John Denker <jsd@av8n.com>,
        Stephan Mueller <smueller@chronox.de>,
        Simo Sorce <simo@redhat.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Sandy Harris <sandyinchina@gmail.com> wrote:

> I will submit this as a patch ...

On second thought, no I won't.

The existing driver uses the gcc latent entropy plugin
to initialise the pool. I'm not sure how much entropy
that provides & worry that it might not be enough on
some systems. However it has an enormous
advantage over my proposal, in that it is likely to
give somewhat different results on every boot.

I have working code for a program to generate
initialised pool arrays & if anyone needs it, they
can certainly have it. However, I will not be trying
to get it into the mainstream kernel.

> /*
>  * Program to select random numbers for initialising things
>  * in the random(4) driver. ...

>  * Inserting random data at compile time can do no harm and
>  * will make some attacks considerably harder. ...