Transparent Encryption (stable/beta)

This guide explains how to configure Cilium to use IPsec based transparent encryption using Kubernetes secrets to distribute the IPsec keys. After this configuration is complete all traffic between Cilium-managed endpoints, as well as Cilium managed host traffic, will be encrypted using IPsec. This guide uses Kubernetes secrets to distribute keys. Alternatively, keys may be manually distributed, but that is not shown here.

Packets destined to the same node they were sent out of are not encrypted. This is a intended behavior as it doesn’t provide any benefits because the raw traffic on the node can be seen.

Transparent encryption is not currently supported when chaining Cilium on top of other CNI plugins. For more information, see GitHub issue #15596.

Generate & import the PSK

First, create a Kubernetes secret for the IPsec configuration to be stored. The example below demonstrates generation of the necessary IPsec configuration which will be distributed as a Kubernetes secret called cilium-ipsec-keys. A Kubernetes secret should consist of one key-value pair where the key is the name of the file to be mounted as a volume in cilium-agent pods, and the value is an IPSec configuration in the following format:

key-id encryption-algorithms PSK-in-hex-format key-size

Note

Secret resources need to be deployed in the same namespace as Cilium! In our example, we use kube-system.

In the example below we use GMC-128-AES, but any of the supported Linux algorithms may be used. To generate, use the following:

  1. $ kubectl create -n kube-system secret generic cilium-ipsec-keys \
  2.     --from-literal=keys="3 rfc4106(gcm(aes)) $(echo $(dd if=/dev/urandom count=20 bs=1 2> /dev/null| xxd -p -c 64)) 128"

The secret can be seen with kubectl -n kube-system get secret and will be listed as “cilium-ipsec-keys”.

  1. $ kubectl -n kube-system get secrets cilium-ipsec-keys
  2. NAME                TYPE     DATA   AGE
  3. cilium-ipsec-keys   Opaque   1      176m

Enable Encryption in Cilium

Note

First, make sure you have Helm 3 installed. Helm 2 is no longer supported.

Setup Helm repository:

  1. helm repo add cilium https://helm.cilium.io/

Deploy Cilium release via Helm with the following options to enable encryption:

  1. helm install cilium cilium/cilium --version 1.9.8 \
  2.   --namespace kube-system \
  3.   --set encryption.enabled=true \
  4.   --set encryption.nodeEncryption=false

These options can be provided along with other options, such as when deploying to GKE, with VXLAN tunneling:

  1. helm install cilium cilium/cilium --version 1.9.8 \
  2.   --namespace kube-system \
  3.   --set nodeinit.enabled=true \
  4.   --set nodeinit.reconfigureKubelet=true \
  5.   --set nodeinit.removeCbrBridge=true \
  6.   --set cni.binPath=/home/kubernetes/bin \
  7.   --set tunnel=vxlan \
  8.   --set encryption.enabled=true \
  9.   --set encryption.nodeEncryption=false

On GKE, Cilium can also be deployed with direct routing instead of tunneling. This requires us to enable the GKE integration and specify the native routing CIDR. As a bonus, node encryption (for transparently encrypting node-to-node traffic) can be enabled as well. See Node to node encryption below.

Note

This example builds on the steps outlined in Installation on Google GKE.

  1. export NATIVE_CIDR="$(gcloud container clusters describe $CLUSTER_NAME --zone $CLUSTER_ZONE --format 'value(clusterIpv4Cidr)')"
  2. helm install cilium cilium/cilium --version 1.9.8 \
  3.   --namespace cilium \
  4.   --set nodeinit.enabled=true \
  5.   --set nodeinit.reconfigureKubelet=true \
  6.   --set nodeinit.removeCbrBridge=true \
  7.   --set cni.binPath=/home/kubernetes/bin \
  8.   --set gke.enabled=true \
  9.   --set ipam.mode=kubernetes \
  10.   --set nativeRoutingCIDR=$NATIVE_CIDR \
  11.   --set encryption.enabled=true \
  12.   --set encryption.nodeEncryption=true

At this point the Cilium managed nodes will be using IPsec for all traffic. For further information on Cilium’s transparent encryption, see eBPF Datapath.

Encryption interface

An additional argument can be used to identify the network-facing interface. If direct routing is used and no interface is specified, the default route link is chosen by inspecting the routing tables. This will work in many cases, but depending on routing rules, users may need to specify the encryption interface as follows:

  1. --set encryption.interface=ethX

Node to node encryption

In order to enable node-to-node encryption, add:

  1. [...]
  2. --set encryption.enabled=true \
  3. --set encryption.nodeEncryption=true \
  4. --set tunnel=disabled

Note

Node to node encryption feature is tested and supported with direct routing modes. Using with encapsulation/tunneling is not currently tested or supported.

Support with tunneling mode is tracked at #13663.

Validate the Setup

Run a bash shell in one of the Cilium pods with kubectl -n <k8s namespace> exec -ti <cilium pod> -- bash and execute the following commands:

  1. Install tcpdump
  1. apt-get update
  2. apt-get -y install tcpdump
  1. Check that traffic is encrypted:
  1. tcpdump -n -i cilium_vxlan
  2. tcpdump: verbose output suppressed, use -v or -vv for full protocol decode
  3. listening on cilium_vxlan, link-type EN10MB (Ethernet), capture size 262144 bytes
  4. 15:16:21.626416 IP 10.60.1.1 > 10.60.0.1: ESP(spi=0x00000001,seq=0x57e2), length 180
  5. 15:16:21.626473 IP 10.60.1.1 > 10.60.0.1: ESP(spi=0x00000001,seq=0x57e3), length 180
  6. 15:16:21.627167 IP 10.60.0.1 > 10.60.1.1: ESP(spi=0x00000001,seq=0x579d), length 100
  7. 15:16:21.627296 IP 10.60.0.1 > 10.60.1.1: ESP(spi=0x00000001,seq=0x579e), length 100
  8. 15:16:21.627523 IP 10.60.0.1 > 10.60.1.1: ESP(spi=0x00000001,seq=0x579f), length 180
  9. 15:16:21.627699 IP 10.60.1.1 > 10.60.0.1: ESP(spi=0x00000001,seq=0x57e4), length 100
  10. 15:16:21.628408 IP 10.60.1.1 > 10.60.0.1: ESP(spi=0x00000001,seq=0x57e5), length 100

Key Rotation

To replace cilium-ipsec-keys secret with a new keys,

  1. KEYID=$(kubectl get secret -n kube-system cilium-ipsec-keys -o yaml | awk '/^\s*keys:/ {print $2}' | base64 -d | awk '{print $1}')
  2. if [[ $KEYID -gt 15 ]]; then KEYID=0; fi
  3. data=$(echo "{\"stringData\":{\"keys\":\"$((($KEYID+1))) "rfc4106\(gcm\(aes\)\)" $(echo $(dd if=/dev/urandom count=20 bs=1 2> /dev/null| xxd -p -c 64)) 128\"}}")
  4. kubectl patch secret -n kube-system cilium-ipsec-keys -p="${data}" -v=1

Then restart Cilium agents to transition to the new key. During transition the new and old keys will be in use. The Cilium agent keeps per endpoint data on which key is used by each endpoint and will use the correct key if either side has not yet been updated. In this way encryption will work as new keys are rolled out.

The KEYID environment variable in the above example stores the current key ID used by Cilium. The key variable is a uint8 with value between 0-16 and should be monotonically increasing every re-key with a rollover from 16 to 0. The Cilium agent will default to KEYID of zero if its not specified in the secret.

Troubleshooting

  • If the cilium Pods fail to start after enabling encryption, double-check if the IPSec Secret and Cilium are deployed in the same namespace together.

  • Make sure that the Cilium pods have kvstore connectivity:

    1. cilium status
    2. KVStore:                Ok   etcd: 1/1 connected: http://127.0.0.1:31079 - 3.3.2 (Leader)
    3. [...]

  • Check for level=warning and level=error messages in the Cilium log files

    • If there is a warning message similar to Device eth0 does not exist, use --set encryption.interface=ethX to set the encryption interface.

  • Run a bash in a Cilium and validate the following:

    • Routing rules matching on fwmark:

      1. ip rule list
      2. 1:      from all fwmark 0xd00/0xf00 lookup 200
      3. 1:      from all fwmark 0xe00/0xf00 lookup 200
      4. [...]

    • Content of routing table 200

      1. ip route list table 200
      2. local 10.60.0.0/24 dev cilium_vxlan proto 50 scope host
      3. 10.60.1.0/24 via 10.60.0.1 dev cilium_host

    • XFRM policy:

      1. ip xfrm p
      2. src 10.60.1.1/24 dst 10.60.0.1/24
      3.         dir fwd priority 0
      4.         mark 0xd00/0xf00
      5.         tmpl src 10.60.1.1 dst 10.60.0.1
      6.                 proto esp spi 0x00000001 reqid 1 mode tunnel
      7. src 10.60.1.1/24 dst 10.60.0.1/24
      8.         dir in priority 0
      9.         mark 0xd00/0xf00
      10.         tmpl src 10.60.1.1 dst 10.60.0.1
      11.                 proto esp spi 0x00000001 reqid 1 mode tunnel
      12. src 10.60.0.1/24 dst 10.60.1.1/24
      13.         dir out priority 0
      14.         mark 0xe00/0xf00
      15.         tmpl src 10.60.0.1 dst 10.60.1.1
      16.                 proto esp spi 0x00000001 reqid 1 mode tunnel

    • XFRM state:

      1. ip xfrm s
      2. src 10.60.0.1 dst 10.60.1.1
      3.         proto esp spi 0x00000001 reqid 1 mode tunnel
      4.         replay-window 0
      5.         auth-trunc hmac(sha256) 0x6162636465666768696a6b6c6d6e6f70717273747576777a797a414243444546 96
      6.         enc cbc(aes) 0x6162636465666768696a6b6c6d6e6f70717273747576777a797a414243444546
      7.         anti-replay context: seq 0x0, oseq 0xe0c0, bitmap 0x00000000
      8.         sel src 0.0.0.0/0 dst 0.0.0.0/0
      9. src 10.60.1.1 dst 10.60.0.1
      10.         proto esp spi 0x00000001 reqid 1 mode tunnel
      11.         replay-window 0
      12.         auth-trunc hmac(sha256) 0x6162636465666768696a6b6c6d6e6f70717273747576777a797a414243444546 96
      13.         enc cbc(aes) 0x6162636465666768696a6b6c6d6e6f70717273747576777a797a414243444546
      14.         anti-replay context: seq 0x0, oseq 0x0, bitmap 0x00000000
      15.         sel src 0.0.0.0/0 dst 0.0.0.0/0

Disabling Encryption

To disable the encryption, regenerate the YAML with the option encryption.enabled=false