summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
l---------docs/gettingstarted/developers/ipsec.rst1
-rw-r--r--src/vnet/ipsec/ipsec.rst199
2 files changed, 200 insertions, 0 deletions
diff --git a/docs/gettingstarted/developers/ipsec.rst b/docs/gettingstarted/developers/ipsec.rst
new file mode 120000
index 00000000000..46a12cdeacc
--- /dev/null
+++ b/docs/gettingstarted/developers/ipsec.rst
@@ -0,0 +1 @@
+../../../src/vnet/ipsec/ipsec.rst \ No newline at end of file
diff --git a/src/vnet/ipsec/ipsec.rst b/src/vnet/ipsec/ipsec.rst
new file mode 100644
index 00000000000..d7e02740fc3
--- /dev/null
+++ b/src/vnet/ipsec/ipsec.rst
@@ -0,0 +1,199 @@
+.. _ipsec:
+
+.. toctree::
+
+IP Security
+===========
+
+This is not a description on how IPSec works. Please read:
+
+ - https://tools.ietf.org/html/rfc4301
+ - https://tools.ietf.org/html/rfc4302
+ - https://tools.ietf.org/html/rfc4303
+
+
+I would also suggest this:
+
+ - https://wiki.strongswan.org/projects/strongswan/wiki/RouteBasedVPN
+
+
+If you're interested in cryptography, I would recommend this excellent
+introductory lecture series (there is also a book, but you'll have to
+buy it, IMHO it's worth it):
+
+ - https://www.youtube.com/channel/UC1usFRN4LCMcfIV7UjHNuQg/featured
+
+
+IPSec VPNs come in two flavours; policy and route based, the
+difference is how the Security Association (SA) is chosen.
+
+
+Route Base VPNs
+---------------
+
+There are two aspects of a route based VPN; all packets to a
+particular peer are encrypted by the same SA and routing
+decides the peer to which to forward traffic (as routing always
+does). Therefore, routing is choosing the SA. Of course the same must
+be true in reverse, that all packets from a given peer are decrypted
+with the same SA. Another way of expressing this is to say a peer is
+'protected' by this SA (really a pair of SAs; one for rx and tx).
+
+The 'standard' [#i1]_ way of representing this protected peer is by
+using a point-to-point virtual interface to which the peer is
+attached and the SA pair is associated. Prefixes
+that require protection are routed through this virtual interface and
+hence implicitly to the peer.
+
+There are three components to the model:
+
+- The SAs; An **ipsec_sa_t**, use the force, read the source.
+- The virtual interface
+- The protection - the association of the SAs to the interface.
+
+
+The protection is represented by a **ipsec_tun_protect_t**. The "tun"
+part comes from the fact that the protected interface is usually a
+tunnel. IMO It would have been better if the author had not assumed
+this [#i2]_.
+The protection associates a single TX SA and up to four RX SAs to an
+interface. Four is as many as can fit on one cache-line. Multiple RX
+SAs mean that a peer can be using any SA in the set, this is
+particularly useful during rekeying because it is not possible for the
+peers to swap their RX and TX SAs at exactly the same moment in the
+traffic stream. Instead they can add the new RX immediately, then swap
+the TX after a short delay, then remove the old RX after another short
+delay. This will minimize, if not eliminate, packet loss.
+
+The virtual interface can be represented in two ways:
+
+ - interface + encap + SA = (interface + encap) + SA = ipip-interface + SA transport mode
+
+or
+
+ - interface + encap + SA = interface + (encap + SA) = IPSec-interface + SA tunnel mode
+
+It's a question of where you add the parenthesis, from the perspective
+of the external user the effect is identical.
+
+The IPSec interface serves as the encap-free interface to be used in
+conjunction with an encap-describing tunnel mode SA. VPP supports both models.
+
+A route based VPN could impose 0, 1 or 2 encaps. the support matrix for these use cases is:
+
+.. code-block:: console
+
+
+ | 0 | 1 | 2 |
+ --------------------------
+ ipip | N | Y | Y |
+ ipsec | P | Y | P |
+
+Where P = potentially.
+
+Ipsec could potentially support 0 encap (i.e. transport mode) since
+neither the interface nor the SA *requires* encap. However, for a
+route based VPN to use transport mode is probably wrong since one
+shouldn't use transport mode for transit traffic, since without encap
+it is not guaranteed to return. IPSec could potentially support 2
+encaps, but that would require the SA to describe both, something it
+does not do at this time.
+
+Internally the difference is that the mid-chain adjacency for the IPSec
+interface has no associated encap (whereas for an ipip tunnel it
+describes the peer). Consequently, features on the output arc see
+packets without any encap. Since the protecting SAs are in tunnel
+mode, they apply the encap. The mid-chain adj is stacked only once the
+protecting SA is known, since only then is the peer known. Otherwise
+the VLIB graph nodes used are the same:
+
+.. code-block:: console
+
+ (routing) --> ipX-michain --> espX-encrypt --> adj-midchain-tx --> (routing)
+
+ where X = 4 or 6.
+
+
+Some benefits to the ipsec interface:
+
+- it is slightly more efficient since the encapsulating IP header has its checksum updated only once.
+- even when the interface is admin up traffic cannot be sent to a peer
+ unless the SA is available (since it's the SA that determines the
+ encap). With ipip interfaces a client must use the admin state to
+ prevent sending until the SA is available.
+
+The best recommendations I can make are:
+
+- pick a model that supports your use case
+- make sure any other features you wish to use are supported by the model
+- choose the model that best fits your control plane's model.
+
+
+Multi-point Interfaces
+^^^^^^^^^^^^^^^^^^^^^^
+
+As mentioned above route based VPNs protect all packets destined to
+a given peer with the same SA pair. This protection was modelled using
+a virtual p2p interface, so one could legitimately reason that
+all traffic through the interface is protected with the SA pair or all
+traffic to the peer is protected, since they are one in the
+same. However, when we consider multi-point interfaces, we have to
+think of protection applying to the peers on the link.
+
+When using IPSec protection on a P2MP link the **ipsec_tun_protect_t**
+will be specific to a particular peer (in the P2P case this peer is
+the usual special all zero address).
+
+All other aspects of using route based VPNs remains the same. The
+routes are resolved via specific peers on the interface, i.e.
+
+.. code-block:: console
+
+ ip route add 10.0.0.0/8 via 192.168.1.1 mipip0
+
+
+rather than
+
+.. code-block:: console
+
+ ip route add 10.0.0.0/8 via ipip0
+
+
+but one should always use a next-hop on a multi-access interface, so
+this is not a restriction.
+
+The data-path is unchanged, in both P2P and P2MP case the SA to
+use for TX comes from the adjacency, and for RX it's the SPI that
+matches to the SA and interface.
+
+
+Policy Based VPNs
+-----------------
+
+At the risk of stating the obvious, in a policy based VPN the SA is
+chosen based on a specific IPSec policy. A policy describes what
+attributes of the packets to match and what action to take if
+matched. Actions are:
+
+- bypass: Ignore it
+- discard: Drop it
+- protect: Either encrypt or decrypt with a specific SA
+
+The 'resolve' action which (as per-RFC4301) states that an IKE session
+should be initiated, is not supported.
+
+Policies are stored in a security policy database (SPD). An SPD is
+attached to an interface. Packets that ingress and egress the
+interface are matched against the policies in the attached SPD.
+This is IPSec as described in RFC4301.
+
+
+.. rubric:: Footnotes:
+
+.. [#i1] Standard in inverted commas because, at least to my
+ knowledge, there is no official standard (RFC) that states it
+ should be this way. It is probably this way because routers
+ model/implement/restrict/etc IPSec as an interface
+ input/output feature.
+.. [#i2] That's a self criticism.
+