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

[Mickaël Salaün]

On 06/02/2021 02:14, Eric Snowberg wrote:
> 
>> On Feb 5, 2021, at 3:27 AM, Mickaël Salaün <mic at digikod.net> wrote:
>>
>>
>> On 05/02/2021 01:24, Eric Snowberg wrote:
>>>
>>>> On Feb 4, 2021, at 1:26 AM, Mickaël Salaün <mic at digikod.net> wrote:
>>>>
>>>>
>>>> On 04/02/2021 04:53, Eric Snowberg wrote:
>>>>>
>>>>>> On Feb 3, 2021, at 11:49 AM, Mickaël Salaün <mic at digikod.net> 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 at oracle.com> 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'll send a patch with this.

> 
> 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.

Hum, I missed a part in refactoring (commit
f78e50c8f750c0ac6767ac1ed006360cf77c56c4). :/
Could you please test the following patch?

diff --git a/certs/blacklist.c b/certs/blacklist.c
index 07c592ae5307..f998a2e85ddc 100644
--- a/certs/blacklist.c
+++ b/certs/blacklist.c
@@ -197,13 +197,16 @@ int mark_hash_blacklisted(const u8 *hash, size_t
hash_len,
                enum blacklist_hash_type hash_type)
 {
        const char *buffer;
+       int err;

        buffer = get_raw_hash(hash, hash_len, hash_type);
        if (IS_ERR(buffer))
                return PTR_ERR(buffer);
+       err = mark_raw_hash_blacklisted(buffer);
        kfree(buffer);
-       return 0;
+       return err;
 }


Is it possible to test these kind of dbx blacklist with Qemu?