- From: Frederick Hirsch <Frederick.Hirsch@nokia.com>
- Date: Thu, 10 Dec 2009 09:58:04 -0500
- To: ext Pratik Datta <pratik.datta@oracle.com>
- Cc: Frederick Hirsch <Frederick.Hirsch@nokia.com>, "Martin, Cynthia E." <cemartin@mitre.org>, XMLSec WG Public List <public-xmlsec@w3.org>
Clarification comments/questions inline regards, Frederick Frederick Hirsch Nokia On Dec 8, 2009, at 3:16 PM, ext Pratik Datta wrote: > I did a little more investigation on Java crypto libraries and found > out that Padding is usually not supported with AES-GCM. Unlike AES- > CBC, which needs padding, AES-GCM doesn’t need padding even though > AES-GCM is a block cipher. Also Java crypto libraries place the > authentication tag at the end. I suppose other crypto libraries > will do similar things. Based on that here is the revised proposed > text. > > Section 5.1, (add this to the list of algorithms.) > > http://www.w3.org/2009/xmlenc11#aes128-gcm > http://www.w3.org/2009/xmlenc11#aes256-gcm > Specifically, add to list of Block Encryption ciphers as Optional, as follows: 5. Optional AES128-GCM http://www.w3.org/2009/xmlenc11#aes128-gcm 6. Optional AES256-GCM http://www.w3.org/2009/xmlenc11#aes256-gcm > > Section 5.2.3 AES-GCM (add new section) > 5.2.4 (already have 5.2.3), but identifiers at beginning, flagged as optional Identifier: http://www.w3.org/2009/xmlenc11#aes128-gcm (Optional) http://www.w3.org/2009/xmlenc11#aes256-gcm (Optional) > AES-GCM is an authenticated encryption mechanism. I.e. it is > equivalent to doing these two operations in one step - HMAC signing > followed by i.e. > AES-CBC encryption. It is very attractive from performance point of > view, because the cost of AES-GCM is similar to regular AES-CBC > encryption, yet it achieves the same result as encryption + HMAC > signing.. Also AES-GCM can be pipelined so it is amenable to > hardware acceleration.. s/.././ > > Identifiers. > http://www.w3.org/2009/xmlenc11#aes128-gcm > http://www.w3.org/2009/xmlenc11#aes256-gcm > at beginning > > AES-GCM is used with a 96 bit Initialization Vector (IV), and a 128 > bit Authentication Tag (T). The cipher text contains the IV first, > followed by the encrypted octets and finally the Authentication tag. > No padding should be used during encryption. During decryption the > implementation should compare the authentication tag computed during > decryption with the specified Authentication Tag, and fail if they > don’t match. > > do we have an example? > Pratik > > > From: Pratik Datta > Sent: Monday, November 16, 2009 9:29 AM > To: Martin, Cynthia E. > Cc: Frederick Hirsch; XMLSec WG Public List > Subject: Re: Proposal for adding AES-GCM to XML Encryption 1.1 > > Cynthia/Magnus, > > Here are some decisions that I took for the AES-GCM. > • IV Size: The algorithm is simplest with a 96 bit IV, so I decided > to go with a 96 bit IV > • Authentication tag Size: The algorithm supports many different > tag sizes, but I decided to go with 128 bit, since that is what TLS > chose. > • Location of Authentication tag: I considered these three options > • putting it at the end : The end of the ciphertext has the > padding, and if I put the Authentication tag in the end, it might > make it harder to implement the padding > • putting it at the beginning : This is what I went for, but > streamability is adversely affected with this solution, because > during encryption you can compute the authentication tag only after > encrypting the whole data, but then you would have to go back to put > in the authentication tag in the beginning of the cipherText > • putting it as a separate tag : this would change the schema, and > it is not worth doing that. That would just make it harder for > implementations. > > > Pratik > > On 11/10/2009 5:51 AM, Martin, Cynthia E. wrote: > I have no objections- I can review it this morning. > > Cynthia > > ________________________________________ > From: public-xmlsec-request@w3.org [public-xmlsec-request@w3.org] On > Behalf Of Frederick Hirsch [Frederick.Hirsch@nokia.com] > Sent: Tuesday, November 10, 2009 8:38 AM > To: ext pratik datta > Cc: Frederick Hirsch; XMLSec WG Public List > Subject: Re: Proposal for adding AES-GCM to XML Encryption 1.1 > > Does anyone object to adding this as optional to the XML Encryption > 1.1 specification before Last Call? > > Who can review the text Pratik sent? > > regards, Frederick > > Frederick Hirsch > Nokia > > > > On Nov 9, 2009, at 6:13 PM, ext pratik datta wrote: > > > It will be optional. > > At this point I am not in a position to interop with this, but maybe > in > a few months. > > Pratik > > On 11/9/2009 12:25 PM, Frederick Hirsch wrote: > > Pratik > > Are you proposing we add it as an Optional or Required to implement > algorithm? > > Who is in a position to interop test this? > > regards, Frederick > > Frederick Hirsch, Nokia > Chair XML Security WG > > > > On Nov 9, 2009, at 3:18 PM, ext pratik datta wrote: > > > I am not sure how important AES-GCM is, but we can consider > adding it > to XML Encryption 1.1. > > NSA suite B requires AES-GCM as a TLS Cipher suite. (see RFC 5430 > http://www.rfc-archive.org/getrfc.php?rfc=5430) > > > > Here is a preliminary proposal for adding AES-GCM (I had a brief > discussion about GCM with Brian in the F2F) > > > Section 5.1, (add this to the list of algorithms.) > > http://www.w3.org/2009/xmlenc11#aes128-gcm > http://www.w3.org/2009/xmlenc11#aes256-gcm > > > Section 5.2.3 AES-GCM (add new section) > > AES-GCM is an authenticated encryption mechanism. I.e. it is > equivalent > to doing these two operations in one step - HMAC signing followed by > AES-CBC encryption. It is very attractive from performance point of > view, because the cost of AES-GCM is similar to regular AES-CBC > encryption, yet it achieves the same result as encryption + HMAC > signing.. Also AES-GCM can be pipelined so it is amenable to > hardware > acceleration.. > > Identifiers. > http://www.w3.org/2009/xmlenc11#aes128-gcm > http://www.w3.org/2009/xmlenc11#aes256-gcm > > > AES-GCM is used with a 96 bit Initialization Vector (IV), and a > 128 bit > Authentication Tag (T). The cipher text contains the IV first, > followed > by the T and then finally the encrypted octets. Decryption should > fail > if the authentication tag computed during decryption does not > match the > specified Authentication Tag. > > > > > Pratik > > > > > > > > > > >
Received on Thursday, 10 December 2009 14:58:55 UTC