Re: [PATCH] certs: Prevent spurious errors on repeated blacklisting

["Mickaël Salaün" <mic@digikod.net>]

On 07/11/2022 17:35, Thomas Weißschuh wrote:
> On 2022-11-07 17:20+0100, Mickaël Salaün wrote:
>> On 07/11/2022 16:55, Thomas Weißschuh wrote:
>>> On 2022-11-07 14:12+0100, Mickaël Salaün wrote:
>>>> This is a follow-up of
>>>> https://lore.kernel.org/r/c8c65713-5cda-43ad-8018-20f2e32e4432@t-8ch.de
>>>>
>>>> Added Jarkko, Mark Pearson, Eric Snowberg and more ML in Cc.
>>>>
>>>>
>>>> On 04/11/2022 02:47, Thomas Weißschuh wrote:
>>>>> When the blacklist keyring was changed to allow updates from the root
>>>>> user it gained an ->update() function that disallows all updates.
>>>>> When the a hash is blacklisted multiple times from the builtin or
>>>>> firmware-provided blacklist this spams prominent logs during boot:
>>>>>
>>>>> [    0.890814] blacklist: Problem blacklisting hash (-13)
>>>>>
>>>>> As all these repeated calls to mark_raw_hash_blacklisted() would create
>>>>> the same keyring entry again anyways these errors can be safely ignored.
>>>>
>>>> These errors can indeed be safely ignored, however they highlight issues
>>>> with some firmware vendors not checking nor optimizing their blocked hashes.
>>>> This raises security concerns, and it should be fixed by firmware vendors.
>>>
>>> Thanks, I was not aware that these are worth fixing.
>>>
>>>>> Fixes: 6364d106e041 ("certs: Allow root user to append signed hashes to the blacklist keyring")
>>>>> Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
>>>>> ---
>>>>>     certs/blacklist.c | 4 +++-
>>>>>     1 file changed, 3 insertions(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/certs/blacklist.c b/certs/blacklist.c
>>>>> index 41f10601cc72..5f7f2882ced7 100644
>>>>> --- a/certs/blacklist.c
>>>>> +++ b/certs/blacklist.c
>>>>> @@ -191,7 +191,9 @@ static int mark_raw_hash_blacklisted(const char *hash)
>>>>>     				   BLACKLIST_KEY_PERM,
>>>>>     				   KEY_ALLOC_NOT_IN_QUOTA |
>>>>>     				   KEY_ALLOC_BUILT_IN);
>>>>> -	if (IS_ERR(key)) {
>>>>> +
>>>>> +	/* Blacklisting the same hash twice fails but would be idempotent */
>>>>> +	if (IS_ERR(key) && PTR_ERR(key) != -EACCES) {
>>>>
>>>> We should not hide EACCES errors. This logs issues, which is correct for
>>>> duplicate hashes, and can help firmware vendors to fix their database. I'd
>>>> really like to see a different log message instead: change the duplicate
>>>> entry error code from EACCES to EEXIST, and call pr_warn for this specific
>>>> case.
>>>
>>> Returning EACCES would require some deeper changes to how the keyring is set up
>>
>> I guess you meant EEXIST?
> 
> Indeed, sorry.
> 
>>> or even changes to the keyring core itself to introduce a key_create() (without
>>> update) function.
>>>
>>> Is this something you would take a look at, or should I try to do it?
>>> (I have no previous knowledge about the keyring subsystem)
>>
>> Please take a look. I think it should not be too complex.
> 
> Will do.
> 
> My plan is to create a new function key_create() that does takes the core logic
> of key_create_or_update() and fails with EEXIST if needed.
> 
>>> In any case it probably would also be good to log the problematic hashes
>>> themselves, so users can properly report the issue to their firmware vendors.
>>
>> Agree
> 
> I'll send a patch for that, too.

Good!

Jarkko, David, any though?

> 
> As for this patch's Fixes-tag, it could refer to either the commit that
> introduced the logging in the first place or the one that actively started to
> trigger it:
> * 734114f8782f ("KEYS: Add a system blacklist keyring")
> * 6364d106e041 ("certs: Allow root user to append signed hashes to the blacklist keyring")
> 
> Personally I'd tend to use the latter.

Even if commit 6364d106e041 is not directly the cause of the issue, it 
makes it visible, so I agree that you should keep the current Fixes tag.


> 
>>>>>     		pr_err("Problem blacklisting hash (%ld)\n", PTR_ERR(key));
>>>>>     		return PTR_ERR(key);
>>>>>     	}
>>>>>
>>>>> base-commit: ee6050c8af96bba2f81e8b0793a1fc2f998fcd20