[Top] [All Lists]

Re: [smime] [Technical Errata Reported] RFC5084 (4774)

2016-08-18 18:51:38
FYI, BouncyCastle just fixed the default to 16 bytes a few hours ago:

On Thu, Aug 18, 2016 at 1:33 PM, Quan Nguyen <quannguyen(_at_)google(_dot_)com> 

On Thu, Aug 18, 2016 at 1:13 PM, Jim Schaad 

I did a brief look at the code you pointed to and I would run screaming
from it for several reasons.

I do not have a problem if the tag is defaulted to a length of 12 bytes,

Note that I were talking about Java Cipher which is used in different
scenarios where some requires high-level of security. It's the library
responsibility to choose a default safe option. Note that in general, all
BouncyCastle, Conscrypt, OpenJDK use 16-bytes as default. The only place
where they use/used 12-byte is where they cite/cited the RFC.

but I do not see any way to set it to a different length if desired.
There should be a version that allows for getting the tag length.  At a
minimum the length of the tag should be parameterized.

There are different ways to initialize Java Cipher and some of them allow
user to choose the authentication tag size.

This is making me want to fix the code, but not necessarily wanting to
fix the spec.  In the context of the document the default is not

If it was designed specifically for your use-case and it uses key only
once for decryption then I agree it's fine. I believe the core problem is
people only looked at section 3.2 and then applied it for *general*

Honestly, I don't know what to do in this situation.


*From:* smime [mailto:smime-bounces(_at_)ietf(_dot_)org] *On Behalf Of *Quan 
*Sent:* Thursday, August 18, 2016 12:07 PM
*To:* Russ Housley <housley(_at_)vigilsec(_dot_)com>
*Cc:* Kathleen Moriarty 
<Kathleen(_dot_)Moriarty(_dot_)ietf(_at_)gmail(_dot_)com>; IETF SMIME <
smime(_at_)ietf(_dot_)org>; David McGrew <mcgrew(_at_)cisco(_dot_)com>; 
Stephen Farrell <
*Subject:* Re: [smime] [Technical Errata Reported] RFC5084 (4774)

On Thu, Aug 18, 2016 at 11:39 AM, Russ Housley 


I just read the cited paper from Niels Ferguson <
ts/CWC-GCM/Ferguson2.pdf>.  Niels says:

   After 2^16 forgery attempts we can expect a successful forgery.

And, then Niels talks about an encrypted voice environment where this
attack might lead to disastrous consequences.

Also, Niels points out that the AES-GCM security proof is unbroken by
this attack.  He is staying within the bounds of the proven security.

Yeah, but the proven security wasn't clear about the weakness/fragility
of GCM. The attack explicitly showed how to exploit the weakness.

It is hard to imagine a protocol environment that uses CMS where 2^16
extra messages would be undetected.

In addition RFC 5084 requires automated key management.  This means that
a fresh AES-GCM key ought to be used for each of the messages.

Oh, this is interesting. The problem I saw with OpenJDK, BouncyCastle,
Conscrypt is when Java Cipher (
e/7/docs/api/javax/crypto/Cipher.html) is initialized to use GCM mode:

   1. In one case, if the parameter is ASN.1 encoded and aes-ICVlen is
missing then it's interpreted as 12-byte.

   2. In other case, they use 12-byte tag, citing RFC 5084
recommendation. For instance, see :

Note that Cipher is a *general *crypto primitive which may be used to
encrypt *multiple* messages. So there may be misunderstanding every now
and then. It's worth to note that I had a hard time convincing developers
to change because they cited RFC :(

 I recognize that attackers do not follow the specification, but it means
that they cannot just use an existing implementation.

These two observations make me wonder whether the is enough of a problem
to bother with an update to RFC 5084.  What do you think?


On Mon, Aug 15, 2016 at 1:55 PM, Russ Housley 


I do not think that we can change the DEFAULT value associated with these
OIDs.  Changing the meaning of an absent aes-ICVlen will result in too many
interoperability problems.

Yeah, I'm aware of it and I understand your concern.

 However, we could put out a very short RFC that updates RFC 5084 to
recommend the use of 16 octet authentication tags in all situations.

Thanks for doing this :) It's SGTM.


On Aug 11, 2016, at 2:49 PM, Quan Nguyen <quannguyen(_at_)google(_dot_)com> 

On Thu, Aug 11, 2016 at 11:47 AM, RFC Errata System <
rfc-editor(_at_)rfc-editor(_dot_)org> wrote:

The following errata report has been submitted for RFC5084,
"Using AES-CCM and AES-GCM Authenticated Encryption in the Cryptographic
Message Syntax (CMS)".

You may review the report below and at:

Type: Technical
Reported by: QUAN NGUYEN <quannguyen(_at_)google(_dot_)com>

Section: 3.2

Original Text
aes-ICVlen       AES-GCM-ICVlen DEFAULT 12

A length of 12 octets is RECOMMENDED.

Corrected Text
aes-ICVlen       AES-GCM-ICVlen DEFAULT 16

A length of 16 octets is RECOMMENDED.

Many JCE providers including OpenJDK, BouncyCastle, Conscrypt have a bug
to use 12 bytes authentication tag (aes-ICVlen) as default if the code path
[1] uses CMS. According to Ferguson's attack (
ts/CWC-GCM/Ferguson2.pdf), if a user encrypts 2^32 block length message,
then 12 bytes authentication tag length has only 96 - 32 = 64 bits security
which is not good enough nowadays. Furthermore, once a forgery happens then
authentication is leaked.

Sorry, I meant "authentication *key*" is leaked.

[1] In other code paths, all providers use 16 bytes authentication tag as

This erratum is currently posted as "Reported". If necessary, please
use "Reply All" to discuss whether it should be verified or
rejected. When a decision is reached, the verifying party (IESG)
can log in to change the status and edit the report, if necessary.

RFC5084 (draft-ietf-smime-cms-aes-ccm-and-gcm-03)
Title               : Using AES-CCM and AES-GCM Authenticated Encryption
in the Cryptographic Message Syntax (CMS)
Publication Date    : November 2007
Author(s)           : R. Housley
Category            : PROPOSED STANDARD
Source              : S/MIME Mail Security
Area                : Security
Stream              : IETF
Verifying Party     : IESG

smime mailing list