Re: Conflict with Mickaël Salaün's blacklist patches [was [PATCH v5 0/4] Add EFI_CERT_X509_GUID support for dbx/mokx entries]

[Eric Snowberg]

> On Feb 5, 2021, at 3:27 AM, Mickaël Salaün <mic@xxxxxxxxxxx> wrote:
> 
> 
> On 05/02/2021 01:24, Eric Snowberg wrote:
>> 
>>> On Feb 4, 2021, at 1:26 AM, Mickaël Salaün <mic@xxxxxxxxxxx> wrote:
>>> 
>>> 
>>> On 04/02/2021 04:53, Eric Snowberg wrote:
>>>> 
>>>>> On Feb 3, 2021, at 11:49 AM, Mickaël Salaün <mic@xxxxxxxxxxx> wrote:
>>>>> 
>>>>> This looks good to me, and it still works for my use case. Eric's
>>>>> patchset only looks for asymmetric keys in the blacklist keyring, so
>>>>> even if we use the same keyring we don't look for the same key types. My
>>>>> patchset only allows blacklist keys (i.e. hashes, not asymmetric keys)
>>>>> to be added by user space (if authenticated), but because Eric's
>>>>> asymmetric keys are loaded with KEY_ALLOC_BYPASS_RESTRICTION, it should
>>>>> be OK for his use case.  There should be no interference between the two
>>>>> new features, but I find it a bit confusing to have such distinct use of
>>>>> keys from the same keyring depending on their type.
>>>> 
>>>> I agree, it is a bit confusing.  What is the thought of having a dbx 
>>>> keyring, similar to how the platform keyring works?
>>>> 
>>>> https://www.spinics.net/lists/linux-security-module/msg40262.html
>>>> 
>>>> 
>>>>> On 03/02/2021 17:26, David Howells wrote:
>>>>>> 
>>>>>> Eric Snowberg <eric.snowberg@xxxxxxxxxx> wrote:
>>>>>> 
>>>>>>> This is the fifth patch series for adding support for 
>>>>>>> EFI_CERT_X509_GUID entries [1].  It has been expanded to not only include
>>>>>>> dbx entries but also entries in the mokx.  Additionally my series to
>>>>>>> preload these certificate [2] has also been included.
>>>>>> 
>>>>>> Okay, I've tentatively applied this to my keys-next branch.  However, it
>>>>>> conflicts minorly with Mickaël Salaün's patches that I've previously merged on
>>>>>> the same branch.  Can you have a look at the merge commit
>>>>>> 
>>>>>> 	https://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs.git/commit/?h=keys-next&id=fdbbe7ceeb95090d09c33ce0497e0394c82aa33d
>>>>>> 
>>>>>> 	(the top patch of my keys-next branch)
>>>>>> 
>>>>>> to see if that is okay by both of you?  If so, can you give it a whirl?
>>>> 
>>>> 
>>>> I’m seeing a build error within blacklist_hashes_checked with
>>>> one of my configs.
>>>> 
>>>> The config is as follows:
>>>> 
>>>> $ grep CONFIG_SYSTEM_BLACKLIST_HASH_LIST .config
>>>> CONFIG_SYSTEM_BLACKLIST_HASH_LIST=“revocation_list"
>>>> 
>>>> $ cat certs/revocation_list
>>>> "tbs:1e125ea4f38acb7b29b0c495fd8e7602c2c3353b913811a9da3a2fb505c08a32”
>>>> 
>>>> make[1]: *** No rule to make target 'revocation_list', needed by 'certs/blacklist_hashes_checked'.  Stop.
>>> 
>>> It requires an absolute path.
>> 
>> Ok, if I use an absolute path now with CONFIG_SYSTEM_BLACKLIST_HASH_LIST 
>> it works.
>> 
>>> This is to align with other variables
>>> using the config_filename macro: CONFIG_SYSTEM_TRUSTED_KEYS,
>>> CONFIG_MODULE_SIG_KEY and now CONFIG_SYSTEM_REVOCATION_KEYS.
>> 
>> I just did a quick test with CONFIG_SYSTEM_TRUSTED_KEYS. It looks like we 
>> can use either a relative or absolute path with CONFIG_SYSTEM_TRUSTED_KEYS. 
>> Shouldn’t this be consistent?
> 
> CONFIG_SYSTEM_TRUSTED_KEYS (and similar config) works with relative path
> to $(srctree) not $(srctree)/certs as in your example.

Correct, I had "certs" in my relative path.

> We can make CONFIG_SYSTEM_BLACKLIST_HASH_LIST works with $(srctree) with
> this patch:
> 
> diff --git a/certs/Makefile b/certs/Makefile
> index eb45407ff282..92a233eaa926 100644
> --- a/certs/Makefile
> +++ b/certs/Makefile
> @@ -14,6 +14,8 @@ $(eval $(call config_filename,SYSTEM_BLACKLIST_HASH_LIST))
> 
> $(obj)/blacklist_hashes.o: $(obj)/blacklist_hashes_checked
> 
> +CFLAGS_blacklist_hashes.o += -I$(srctree)
> +
> targets += blacklist_hashes_checked

After applying this patch, CONFIG_SYSTEM_BLACKLIST_HASH_LIST now works
like the other filename macros.  It seems like this would be a good
addition.

I have done some additional testing, I am seeing a regression. The blacklist 
keyring is no longer picking up any of the hashes from the dbx during boot. 
I backed out the merge with my changes  (fdbbe7ceeb95090d09c33ce0497e0394c82aa33d) 
and still see the regression.  I then backed out Mickaël merge
(5bf1adccf5c41dbdd51d1f4de220d335d9548598) and it fixes the regression.

On a x86 with the updated dbx from uefi.org, I’d expect to see 234 bin hash entries
in the blacklist keyring.  With the current merged code, there is none.