[strongSwan] Calculating the generated MAC address when identity_lease is enabled
Micah R Ledbetter
me at micahrl.com
Mon Apr 9 01:15:52 CEST 2018
> A MAC address is 48 bits, so the 32 bits after the first two bytes
> are taken from the 64-bit SipHash-2-4 value (just by casting to an
> uint32_t) which is then run through htonl() and copied to the address
> buffer. To calculate the value the 128-bit key
> 0x000102030405060708090a0b0c0d0e0f is used.
Ahhh. I missed that `id` was uint32_t. That clears up my confusion.
> If you use FQDNs as client identities they will be sent in a hostname
> option (12) in the DHCP request, which you might be able to use
Ok, that sounds like a better solution. However, I am having trouble
making it work.
I put a lot of detail into my message below to provide all the context
I could, but the crux of my question now is this: can I configure the
strongSwan server to force all clients to use an FQDN as their
identifier, as you suggest? Or alternatively, can I force this by
changing the way I generate certificates?
Different clients send different identifiers
It seems that different clients send different things as identifiers:
- macOS and iOS appear to send the bare Common Name from their
certificate. For example, if the cert CN is macosclient.example.com
a conn section with
will match when the client attempts to connect.
- Windows appears to send a DER-encoded X.509 Common Name object
For example, if the cert CN is windowsclient.example.com, it sends:
... and that is matched by a conn section like
- strongSwan (clients) are configurable. They appear to send the same
thing that Windows sends by default, but can be configured to work
the same as macOS, and even more ways besides, depending on what
they have set as their 'leftid'
First attempt: rightid=%client.example.com
I found some documentation here:
> Since 5.0.1 rightid for IKEv2 connections optionally takes a % as
> prefix in front of the identity. If given it prevents the daemon from
> sending IDr in its IKE_AUTH request and will allow it to verify the
> configured identity against the subject and subjectAltNames contained
> in the responder's certificate (otherwise, it is only compared with
> the IDr returned by the responder). The IDr sent by the initiator
> might otherwise prevent the responder from finding a config if it has
> configured a different value for leftid.
However, I wasn’t able to make that work the way I expected. I
initially read this to mean that it would pluck the FQDN out of the
cert that the client sends and use that as an identity, but that
doesn't appear to be the correct reading.
I put two conn sections in my ipsec.conf:
(This isn’t ideal, because it requires a separate conn section for each
user. Is there a way to avoid that?)
With that configuration, the macOS client can connect just fine, but
the Windows client cannot. (Note that the certificates were generated
the same way, but of course with different hostnames.)
Am I misreading the documentation? Is there some way to work around
this for built-in IPSEC clients, like on Windows, which don’t appear to
let me configure what is sent as the identifier?
Further exploration: two conn sections per user
As I mentioned above, I found that Windows clients will connect if
the rightid is something like
However, the biggest problem with this is that I don’t know the
operating system that my users will be on in advance. I provide VPN
profiles for macOS and iOS (as .mobileconfig files), Windows (as a
Powershell script), and Linux (as strongSwan configuration files); my
users install whatever is appropriate for their operating system.
That means that I’d need to have two conn sections for each user.
That’s not ideal, but I could make it work if I have to.
Is there a better way?
A recap of my questions, since this got pretty long:
1. Can I configure the strongSwan server to force the clients to send
the FQDNs as identities?
2. Alternatively, can I generate certificates differently to force the
clients to send the FQDNs as identities?
3. Am I misreading the documentation about rightid=%fqdn? If so, what
is it intended to do?
4. Can I avoid using two conn sections for each user somehow?
5. Even better, can I use a single conn section to match all users,
no matter their operating system, and enforce that they send their
client identifier to the DHCP server the same way?
I don't think I said this before, but I am on Ubuntu, and my ipsec
version is: Linux strongSwan U5.3.5/K4.4.0-1054-aws
Thank you again very much for any help.
More information about the Users