[PATCH v4] certs: Add EFI_CERT_X509_GUID support for dbx entries

[Eric Snowberg]

> On Jan 20, 2021, at 4:26 AM, Jarkko Sakkinen <jarkko at kernel.org> wrote:
> 
> On Fri, Jan 15, 2021 at 09:49:02AM -0700, Eric Snowberg wrote:
>> 
>>> On Jan 15, 2021, at 2:15 AM, Jarkko Sakkinen <jarkko at kernel.org> wrote:
>>> 
>>> On Wed, Jan 13, 2021 at 05:11:10PM -0700, Eric Snowberg wrote:
>>>> 
>>>>> On Jan 13, 2021, at 1:41 PM, Jarkko Sakkinen <jarkko.sakkinen at linux.intel.com> wrote:
>>>>> 
>>>>> On Tue, Jan 12, 2021 at 02:57:39PM +0000, David Howells wrote:
>>>>>> Eric Snowberg <eric.snowberg at oracle.com> wrote:
>>>>>> 
>>>>>>>> On Dec 10, 2020, at 2:49 AM, David Howells <dhowells at redhat.com> wrote:
>>>>>>>> 
>>>>>>>> Eric Snowberg <eric.snowberg at oracle.com> wrote:
>>>>>>>> 
>>>>>>>>> Add support for EFI_CERT_X509_GUID dbx entries. When a EFI_CERT_X509_GUID
>>>>>>>>> is found, it is added as an asymmetrical key to the .blacklist keyring.
>>>>>>>>> Anytime the .platform keyring is used, the keys in the .blacklist keyring
>>>>>>>>> are referenced, if a matching key is found, the key will be rejected.
>>>>>>>> 
>>>>>>>> Ummm...  Why this way and not as a blacklist key which takes up less space?
>>>>>>>> I'm guessing that you're using the key chain matching logic.  We really only
>>>>>>>> need to blacklist the key IDs.
>>>>>>> 
>>>>>>> I implemented it this way so that certs in the dbx would only impact 
>>>>>>> the .platform keyring. I was under the impression we didn’t want to have 
>>>>>>> Secure Boot UEFI db/dbx certs dictate keyring functionality within the kernel
>>>>>>> itself. Meaning if we have a matching dbx cert in any other keyring (builtin,
>>>>>>> secondary, ima, etc.), it would be allowed. If that is not how you’d like to 
>>>>>>> see it done, let me know and I’ll make the change.
>>>>>> 
>>>>>> I wonder if that is that the right thing to do.  I guess this is a policy
>>>>>> decision and may depend on the particular user.
>>>>> 
>>>>> Why would you want to allow dbx entry in any keyring?
>>>> 
>>>> Today, DB and MOK certs go into the platform keyring.  These certs are only
>>>> referenced during kexec.  They can’t be used for other things like validating
>>>> kernel module signatures.  If we follow the same pattern, the DBX and MOKX entries
>>>> in the blacklist keyring should only impact kexec. 
>>>> 
>>>> Currently, Mickaël Salaün has another outstanding series to allow root to update 
>>>> the blacklist keyring.  I assume the use case for this is around certificates used 
>>>> within the kernel, for example revoking kernel module signatures.  The question I have
>>>> is, should another keyring be introduced?  One that carries DBX and MOKX, which just
>>>> correspond to certs/hashes in the platform keyring; this keyring would only be
>>>> referenced for kexec, just like the platform keyring is today. Then, the current
>>>> blacklist keyring would be used for everything internal to the kernel.
>>> 
>>> Right, I'm following actively that series.
>>> 
>>> Why couldn't user space drive this process and use that feature to do it?
>> 
>> I could see where the user would want to use both. With Mickaël Salaün’s
>> series, the blacklist keyring is updated immediately.  However it does
>> not survive a reboot.  With my patch, the blacklist keyring is updated
>> during boot, based on what is in the dbx. Neither approach needs a new 
>> kernel build.
> 
> I don't want to purposely challenge this, but why does it matter
> that it doesn't survive the boot? I'm referring here to the golden
> principle of kernel defining a mechanism, not policy. User space
> can do the population however it wants to for every boot.
> 
> E.g. systemd service could do this.
> 
> What am I missing here?

This change simply adds support for a missing type.  The kernel 
already supports cert and hash entries (EFI_CERT_X509_SHA256_GUID,
EFI_CERT_SHA256_GUID) that originate from the dbx and are loaded 
into the blacklist keyring during boot.  I’m not sure why a cert 
defined with EFI_CERT_X509_GUID should be handled in a different 
manner.

I suppose a user space tool could be created. But wouldn’t what is
currently done in the kernel in this area need to be removed?