Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1651896 > unrolled thread

Problem with new X.509 is_hash_blacklisted() interface

Back to article view | Back to linux.kernel

Contents

  Problem with new X.509 is_hash_blacklisted() interface James Bottomley <James.Bottomley@HansenPartnership.com> - 2017-05-27 17:10 +0200
    Re: Problem with new X.509 is_hash_blacklisted() interface Ard Biesheuvel <ard.biesheuvel@linaro.org> - 2017-05-30 12:50 +0200

#1651896 — Problem with new X.509 is_hash_blacklisted() interface

Added by

commit 436529562df2748fd9918f578205b22cf8ced277
Author: David Howells <dhowells@redhat.com>
Date:   Mon Apr 3 16:07:25 2017 +0100

    X.509: Allow X.509 certs to be blacklisted
 
Ironically it duplicates a UEFI bug we've been struggling with for a
while in the pkcs11 handlers:  namely if you have a blacklist based on
certificate hashes, an interface which only takes a hash cannot
definitively tell you if the certificate is on the blacklist or not
because the hash the cert is blacklisted by may be a different
algorithm from the hash you feed in to is_hash_blacklisted().  This
means that the only safe way to use the interface is to construct every
possible hash of the cert and feed them one at a time into
is_hash_blacklisted().  This makes it an almost unusable API.

I suggest you deprecate this interface immediately and introduce an
is_cert_blacklisted() one which takes a pointer to the TBS data.  Then
the implementation can loop over the blacklists, see the hash type and
construct the hash of the TBS data for comparison (caching the hashes
for efficiency).  That way you'll be assured of a definitive answer and
an easy API.

It might be reasonable to cc linux-efi on future kernel keyring stuff,
because some of the other issues may have also come up in the UEFI
keyrings.

James

[toc] | [next] | [standalone]

#1653099

On 27 May 2017 at 15:05, James Bottomley
<James.Bottomley@hansenpartnership.com> wrote:
> Added by
>
> commit 436529562df2748fd9918f578205b22cf8ced277
> Author: David Howells <dhowells@redhat.com>
> Date:   Mon Apr 3 16:07:25 2017 +0100
>
>     X.509: Allow X.509 certs to be blacklisted
>
> Ironically it duplicates a UEFI bug we've been struggling with for a
> while in the pkcs11 handlers:  namely if you have a blacklist based on
> certificate hashes, an interface which only takes a hash cannot
> definitively tell you if the certificate is on the blacklist or not
> because the hash the cert is blacklisted by may be a different
> algorithm from the hash you feed in to is_hash_blacklisted().  This
> means that the only safe way to use the interface is to construct every
> possible hash of the cert and feed them one at a time into
> is_hash_blacklisted().  This makes it an almost unusable API.
>
> I suggest you deprecate this interface immediately and introduce an
> is_cert_blacklisted() one which takes a pointer to the TBS data.  Then
> the implementation can loop over the blacklists, see the hash type and
> construct the hash of the TBS data for comparison (caching the hashes
> for efficiency).  That way you'll be assured of a definitive answer and
> an easy API.
>
> It might be reasonable to cc linux-efi on future kernel keyring stuff,
> because some of the other issues may have also come up in the UEFI
> keyrings.
>

Hi James,

Thanks for highlighting this. I agree that this should be addressed
asap, given that this code has not appeared in a release yet (it was
added this cycle)

Perhaps redundantly, I'd like to emphasize that this is really not a
UEFI specific issue, it applies to any application of X.509 that does
not restrict the set of permitted hash algorithms to a single one.

Regards,
Ard.

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web