[Softhsm-develop] Root of trust in SoftHSM

[Full Name]

Thanks for your answer. Please see my comments interspersed below.

-----Original Message-----
From: "(Berry) A.W. van Halderen" [berry at nlnetlabs.nl]
Date: 04/06/2021 10:01 AM
To: "Full Name" <nuncestbibendum at excite.com>
CC: softhsm-develop at lists.opendnssec.org
Subject: Re: [Softhsm-develop] Root of trust in SoftHSM

Dear Anonymous,

I should wait until later in the week until I take that drink, besides
the pub isn't open still....

On Tue, Apr 06, 2021 at 09:25:05AM -0400, Full Name wrote:
> My questions are to do with how key material that has been generated within (or loade into) SoftHSM is protected when in permanent storage. My guess is that it is encrypted with a symmetric key available to SoftHSM alone, that will be fed to AES - or something similar - to decrypt (or encrypt) the key material as necessary. On this basis my questions are the following:
> 
> 1. Is the mechanism described even close to the approach taken by SoftHSM? Either way, where is the mechanism used described?
Roughly yes this is the mechanism.  Plus the permanent storage is not loaded,
but key material is kept encrypted as much as possible.

Can you elaborate on this please? The workflow that I have in mind is one in which keys are loaded into SoftHSM and are to be available to SoftHSM afterwards, even when the host system has been rebooted after the keys above have been loaded. I.e. unless a key is explicitly evicted from SoftHSM, it should be available to it on demand. Or is this perhaps something to be managed outside SoftHSM?


> 2. Assuming hereinafter that a root symmetric key is indeed used to protect the key material in permanent storage, where is that symmetric key coming from? 
>From the storage itself, but itself is encrypted using SO and user pin.

Are the details described somewhere? I am interested in how this encryption is carried out - i.e. how is the key derived from the SO and user PINs, and what algorithm is used.

> 3. Is the root symmetric key generated based on the SO, or CO, credentials? 
see 2, not generated based on SO, but encrypted using SO/user pin.

> 4. Is the root symmetric key stored somewhere, obfuscated but in the clear? 
See above, plus that the symmetric key is kept in non-paged memory to avoid
it every getting stored and is being obfusticated.

This in fact makes SoftHSMv2 slower than SoftHSM version 1, because for many
actions it needs to decrypt data it already had decrypted earlier.
SoftHSM tries to be as secure possible, but still being software only.

> 5. Does SoftHSM provide the capability to retrieve this root symmetric key from some separate piece of hardware, such as a TPM or a YubiKey?

Not as the moment, but this is a possible feature.  On the other hand this
only really helps security if the actual decryption for which the AES key
is needed is done in the TPM.  This will hurt performance severely I think.

That is correct - TMPs are terribly slow. I was just asking from the point of view of using the TPM as the root of trust  for the master key.