Hi,
I believe that this is the right list for my question. I'm trying to get
SHA-2 HMAC support working ipsec in linux kernel (I'm configuring via
pfkey).
First, sha-384 and sha-512 as authentication algorithm always return
function not support. But I noted that my linux kernel has a sha512
kernel module (with alias for sha384). Second, sha-256 uses a 12-byte
hmac (96 bits).
Looking at the source http://lxr.linux.no/source/net/xfrm/xfrm_algo.c,
it seems to confirm that this is true. In fact, sha-384 and sha-512 are
not supported at this time and sha-256 is truncated to 96-bit.
However, the following ietf draft, which I believe is very closed to
ratification (it has already been assigned iana numbers), specifies
sha-256 to use 128-bits as hmac (page 18):
http://www.ietf.org/internet-drafts/draft-kelly-ipsec-ciph-sha2-01.txt
sha-384 is 192 bits, and sha-512 is 256 bits.
1. Is adding sha-384 and sha-512 as simple as adding to the aalg_list
structure? Can this be done for some subsequent kernel release in the
future?
2. Can the sha-256 be changed to use 128 bits? Or in order to not break
backward compatibility, another sha-256 hmac algorithm id be used for
128 bits?
Thanks,
Chinh
Chinh Nguyen wrote:
> Looking at the source http://lxr.linux.no/source/net/xfrm/xfrm_algo.c,
> it seems to confirm that this is true. In fact, sha-384 and sha-512 are
> not supported at this time and sha-256 is truncated to 96-bit.
That's normal.
HMAC usage in IPsec specifies that we only use 96-bits of the result.
This is a tradeoff in space in the packet vs absolute "security"
In addition should you be able to cause a collision in 96-bits by some
method other than brute force, you can not be sure if you guess the key
properly.
> However, the following ietf draft, which I believe is very closed to
> ratification (it has already been assigned iana numbers), specifies
> sha-256 to use 128-bits as hmac (page 18):
> http://www.ietf.org/internet-drafts/draft-kelly-ipsec-ciph-sha2-01.txt
Yes, but that's the key, not the result.
It is keyed with various sizes of bits, but the results are truncated.