PMTUD exploration, start small then grow

* SSH handshake in goroutine and defer close

.github/workflows/gofmt.yml

@@ -14,7 +14,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
 
-    - uses: actions/checkout@v4
+    - uses: actions/checkout@v6
 
     - uses: actions/setup-go@v6
       with:

.github/workflows/release.yml

@@ -10,7 +10,7 @@ jobs:
     name: Build Linux/BSD All
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
 
       - uses: actions/setup-go@v6
         with:
@@ -24,7 +24,7 @@ jobs:
           mv build/*.tar.gz release
 
       - name: Upload artifacts
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v6
         with:
           name: linux-latest
           path: release
@@ -33,7 +33,7 @@ jobs:
     name: Build Windows
     runs-on: windows-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
 
       - uses: actions/setup-go@v6
         with:
@@ -55,7 +55,7 @@ jobs:
           mv dist\windows\wintun build\dist\windows\
 
       - name: Upload artifacts
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v6
         with:
           name: windows-latest
           path: build
@@ -66,7 +66,7 @@ jobs:
       HAS_SIGNING_CREDS: ${{ secrets.AC_USERNAME != '' }}
     runs-on: macos-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
 
       - uses: actions/setup-go@v6
         with:
@@ -75,7 +75,7 @@ jobs:
 
       - name: Import certificates
         if: env.HAS_SIGNING_CREDS == 'true'
-        uses: Apple-Actions/import-codesign-certs@v5
+        uses: Apple-Actions/import-codesign-certs@v6
         with:
           p12-file-base64: ${{ secrets.APPLE_DEVELOPER_CERTIFICATE_P12_BASE64 }}
           p12-password: ${{ secrets.APPLE_DEVELOPER_CERTIFICATE_PASSWORD }}
@@ -104,7 +104,7 @@ jobs:
           fi
 
       - name: Upload artifacts
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v6
         with:
           name: darwin-latest
           path: ./release/*
@@ -124,11 +124,11 @@ jobs:
       # be overwritten
       - name: Checkout code
         if: ${{ env.HAS_DOCKER_CREDS == 'true' }}
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
 
       - name: Download artifacts
         if: ${{ env.HAS_DOCKER_CREDS == 'true' }}
-        uses: actions/download-artifact@v4
+        uses: actions/download-artifact@v7
         with:
           name: linux-latest
           path: artifacts
@@ -160,10 +160,10 @@ jobs:
     needs: [build-linux, build-darwin, build-windows]
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
 
       - name: Download artifacts
-        uses: actions/download-artifact@v4
+        uses: actions/download-artifact@v7
         with:
           path: artifacts
 
@@ -209,10 +209,11 @@ jobs:
         id: create_release
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          GITHUB_REF_NAME: ${{ github.ref_name }}
         run: |
           cd artifacts
           gh release create \
             --verify-tag \
-            --title "Release ${{ github.ref_name }}" \
+            --title "Release ${GITHUB_REF_NAME}" \
-            "${{ github.ref_name }}" \
+            "${GITHUB_REF_NAME}" \
             SHASUM256.txt *-latest/*.zip *-latest/*.tar.gz

.github/workflows/smoke-extra.yml

@@ -18,9 +18,11 @@ jobs:
     if: github.ref == 'refs/heads/master' || contains(github.event.pull_request.labels.*.name, 'smoke-test-extra')
     name: Run extra smoke tests
     runs-on: ubuntu-latest
+    env:
+      VAGRANT_DEFAULT_PROVIDER: libvirt
     steps:
 
-    - uses: actions/checkout@v4
+    - uses: actions/checkout@v6
 
     - uses: actions/setup-go@v6
       with:
@@ -30,8 +32,13 @@ jobs:
     - name: add hashicorp source
       run: wget -O- https://apt.releases.hashicorp.com/gpg | gpg --dearmor | sudo tee /usr/share/keyrings/hashicorp-archive-keyring.gpg && echo "deb [signed-by=/usr/share/keyrings/hashicorp-archive-keyring.gpg] https://apt.releases.hashicorp.com $(lsb_release -cs) main" | sudo tee /etc/apt/sources.list.d/hashicorp.list
 
-    - name: install vagrant
+    - name: install vagrant and libvirt
-      run: sudo apt-get update && sudo apt-get install -y vagrant virtualbox
+      run: |
+        sudo apt-get update && sudo apt-get install -y vagrant libvirt-daemon-system libvirt-dev
+        sudo chmod 666 /dev/kvm
+        sudo usermod -aG libvirt $(whoami)
+        sudo chmod 666 /var/run/libvirt/libvirt-sock
+        vagrant plugin install vagrant-libvirt
 
     - name: freebsd-amd64
       run: make smoke-vagrant/freebsd-amd64
@@ -42,10 +49,19 @@ jobs:
     - name: netbsd-amd64
       run: make smoke-vagrant/netbsd-amd64
 
-    - name: linux-386
-      run: make smoke-vagrant/linux-386
-
     - name: linux-amd64-ipv6disable
       run: make smoke-vagrant/linux-amd64-ipv6disable
 
+    # linux-386 runs last because it requires disabling KVM to use VirtualBox,
+    # which prevents libvirt (used by the other tests) from working after this point.
+    - name: install virtualbox for i386 test
+      run: |
+        sudo apt-get install -y virtualbox
+        sudo rmmod kvm_amd kvm_intel kvm 2>/dev/null || true
+
+    - name: linux-386
+      env:
+        VAGRANT_DEFAULT_PROVIDER: virtualbox
+      run: make smoke-vagrant/linux-386
+
     timeout-minutes: 30

.github/workflows/smoke.yml

@@ -18,7 +18,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
 
-    - uses: actions/checkout@v4
+    - uses: actions/checkout@v6
 
     - uses: actions/setup-go@v6
       with:

.github/workflows/smoke/build-relay.sh

@@ -16,8 +16,10 @@ relay:
   am_relay: true
 EOF
 
-    export LIGHTHOUSES="192.168.100.1 172.17.0.2:4242"
+    # TEST-NET-3 placeholder IPs; smoke-relay.sh seds them to real container IPs.
-    export REMOTE_ALLOW_LIST='{"172.17.0.4/32": false, "172.17.0.5/32": false}'
+    # Mapping: .2 lighthouse1, .3 host2, .4 host3, .5 host4.
+    export LIGHTHOUSES="192.168.100.1 203.0.113.2:4242"
+    export REMOTE_ALLOW_LIST='{"203.0.113.4/32": false, "203.0.113.5/32": false}'
 
     HOST="host2" ../genconfig.sh >host2.yml <<EOF
 relay:
@@ -25,7 +27,7 @@ relay:
     - 192.168.100.1
 EOF
 
-    export REMOTE_ALLOW_LIST='{"172.17.0.3/32": false}'
+    export REMOTE_ALLOW_LIST='{"203.0.113.3/32": false}'
 
     HOST="host3" ../genconfig.sh >host3.yml
 

.github/workflows/smoke/build.sh

@@ -5,9 +5,15 @@ set -e -x
 rm -rf ./build
 mkdir ./build
 
-# TODO: Assumes your docker bridge network is a /24, and the first container that launches will be .1
+# Smoke containers run on a dedicated docker network whose subnet is allocated
-# - We could make this better by launching the lighthouse first and then fetching what IP it is.
+# at smoke time, not known at build time. Configs are written with TEST-NET-3
-NET="$(docker network inspect bridge -f '{{ range .IPAM.Config }}{{ .Subnet }}{{ end }}' | cut -d. -f1-3)"
+# placeholder IPs (RFC 5737) and smoke.sh / smoke-vagrant.sh / smoke-relay.sh
+# sed the real container IPs in before starting nebula.
+#
+# Placeholder mapping (last octet == fixed container slot):
+#   203.0.113.2 -> lighthouse1, 203.0.113.3 -> host2,
+#   203.0.113.4 -> host3,       203.0.113.5 -> host4.
+LIGHTHOUSE_IP="203.0.113.2"
 
 (
     cd build
@@ -25,16 +31,16 @@ NET="$(docker network inspect bridge -f '{{ range .IPAM.Config }}{{ .Subnet }}{{
         ../genconfig.sh >lighthouse1.yml
 
     HOST="host2" \
-        LIGHTHOUSES="192.168.100.1 $NET.2:4242" \
+        LIGHTHOUSES="192.168.100.1 $LIGHTHOUSE_IP:4242" \
         ../genconfig.sh >host2.yml
 
     HOST="host3" \
-        LIGHTHOUSES="192.168.100.1 $NET.2:4242" \
+        LIGHTHOUSES="192.168.100.1 $LIGHTHOUSE_IP:4242" \
         INBOUND='[{"port": "any", "proto": "icmp", "group": "lighthouse"}]' \
         ../genconfig.sh >host3.yml
 
     HOST="host4" \
-        LIGHTHOUSES="192.168.100.1 $NET.2:4242" \
+        LIGHTHOUSES="192.168.100.1 $LIGHTHOUSE_IP:4242" \
         OUTBOUND='[{"port": "any", "proto": "icmp", "group": "lighthouse"}]' \
         ../genconfig.sh >host4.yml
 

.github/workflows/smoke/smoke-relay.sh

@@ -6,6 +6,8 @@ set -o pipefail
 
 mkdir -p logs
 
+NETWORK="nebula-smoke-relay"
+
 cleanup() {
     echo
     echo " *** cleanup"
@@ -16,22 +18,53 @@ cleanup() {
     then
         docker kill lighthouse1 host2 host3 host4
     fi
+    docker network rm "$NETWORK" >/dev/null 2>&1
 }
 
 trap cleanup EXIT
 
-docker run --name lighthouse1 --rm nebula:smoke-relay -config lighthouse1.yml -test
+# Create a dedicated smoke network with an explicit subnet (required for --ip
-docker run --name host2 --rm nebula:smoke-relay -config host2.yml -test
+# below). Probe a short list of candidates so a locally-used range doesn't
-docker run --name host3 --rm nebula:smoke-relay -config host3.yml -test
+# fail the whole test — we only need one to be free.
-docker run --name host4 --rm nebula:smoke-relay -config host4.yml -test
+docker network rm "$NETWORK" >/dev/null 2>&1 || true
+for candidate in 172.30.0.0/24 172.31.0.0/24 10.98.0.0/24 10.99.0.0/24 192.168.230.0/24; do
+    if docker network create --subnet "$candidate" "$NETWORK" >/dev/null 2>&1; then
+        break
+    fi
+done
+if ! docker network inspect "$NETWORK" >/dev/null 2>&1; then
+    echo "failed to create $NETWORK: every candidate subnet is in use" >&2
+    exit 1
+fi
 
-docker run --name lighthouse1 --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm nebula:smoke-relay -config lighthouse1.yml 2>&1 | tee logs/lighthouse1 | sed -u 's/^/  [lighthouse1]  /' &
+# Derive container IPs from the network's assigned subnet. Slots: .2 lighthouse1,
+# .3 host2, .4 host3, .5 host4 — matches the placeholders in build-relay.sh.
+SUBNET="$(docker network inspect -f '{{(index .IPAM.Config 0).Subnet}}' "$NETWORK")"
+PREFIX="${SUBNET%/*}"
+PREFIX="${PREFIX%.*}"
+LIGHTHOUSE_IP="$PREFIX.2"
+HOST2_IP="$PREFIX.3"
+HOST3_IP="$PREFIX.4"
+HOST4_IP="$PREFIX.5"
+
+# Sed the placeholder TEST-NET-3 IPs in the host configs to the real ones.
+for f in build/host2.yml build/host3.yml build/host4.yml; do
+    sed "s|203\.0\.113\.|$PREFIX.|g" "$f" >"$f.tmp"
+    mv "$f.tmp" "$f"
+done
+
+docker run --name lighthouse1 --rm nebula:smoke-relay -config lighthouse1.yml -test
+docker run --name host2 --rm -v "$PWD/build/host2.yml:/nebula/host2.yml:ro" nebula:smoke-relay -config host2.yml -test
+docker run --name host3 --rm -v "$PWD/build/host3.yml:/nebula/host3.yml:ro" nebula:smoke-relay -config host3.yml -test
+docker run --name host4 --rm -v "$PWD/build/host4.yml:/nebula/host4.yml:ro" nebula:smoke-relay -config host4.yml -test
+
+docker run --name lighthouse1 --network "$NETWORK" --ip "$LIGHTHOUSE_IP" --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm nebula:smoke-relay -config lighthouse1.yml 2>&1 | tee logs/lighthouse1 | sed -u 's/^/  [lighthouse1]  /' &
 sleep 1
-docker run --name host2 --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm nebula:smoke-relay -config host2.yml 2>&1 | tee logs/host2 | sed -u 's/^/  [host2]  /' &
+docker run --name host2 --network "$NETWORK" --ip "$HOST2_IP" -v "$PWD/build/host2.yml:/nebula/host2.yml:ro" --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm nebula:smoke-relay -config host2.yml 2>&1 | tee logs/host2 | sed -u 's/^/  [host2]  /' &
 sleep 1
-docker run --name host3 --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm nebula:smoke-relay -config host3.yml 2>&1 | tee logs/host3 | sed -u 's/^/  [host3]  /' &
+docker run --name host3 --network "$NETWORK" --ip "$HOST3_IP" -v "$PWD/build/host3.yml:/nebula/host3.yml:ro" --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm nebula:smoke-relay -config host3.yml 2>&1 | tee logs/host3 | sed -u 's/^/  [host3]  /' &
 sleep 1
-docker run --name host4 --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm nebula:smoke-relay -config host4.yml 2>&1 | tee logs/host4 | sed -u 's/^/  [host4]  /' &
+docker run --name host4 --network "$NETWORK" --ip "$HOST4_IP" -v "$PWD/build/host4.yml:/nebula/host4.yml:ro" --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm nebula:smoke-relay -config host4.yml 2>&1 | tee logs/host4 | sed -u 's/^/  [host4]  /' &
 sleep 1
 
 set +x
@@ -76,7 +109,13 @@ docker exec host4 sh -c 'kill 1'
 docker exec host3 sh -c 'kill 1'
 docker exec host2 sh -c 'kill 1'
 docker exec lighthouse1 sh -c 'kill 1'
-sleep 5
+
+# Wait up to 30s for all backgrounded jobs to exit rather than relying on a
+# fixed sleep.
+for _ in $(seq 1 30); do
+    [ -z "$(jobs -r)" ] && break
+    sleep 1
+done
 
 if [ "$(jobs -r)" ]
 then

.github/workflows/smoke/smoke-vagrant.sh

@@ -8,6 +8,8 @@ export VAGRANT_CWD="$PWD/vagrant-$1"
 
 mkdir -p logs
 
+NETWORK="nebula-smoke"
+
 cleanup() {
     echo
     echo " *** cleanup"
@@ -19,21 +21,51 @@ cleanup() {
         docker kill lighthouse1 host2
     fi
     vagrant destroy -f
+    docker network rm "$NETWORK" >/dev/null 2>&1
 }
 
 trap cleanup EXIT
 
+# Create a dedicated smoke network with an explicit subnet (required for --ip
+# below). Probe a short list of candidates so a locally-used range doesn't
+# fail the whole test — we only need one to be free.
+docker network rm "$NETWORK" >/dev/null 2>&1 || true
+for candidate in 172.30.0.0/24 172.31.0.0/24 10.98.0.0/24 10.99.0.0/24 192.168.230.0/24; do
+    if docker network create --subnet "$candidate" "$NETWORK" >/dev/null 2>&1; then
+        break
+    fi
+done
+if ! docker network inspect "$NETWORK" >/dev/null 2>&1; then
+    echo "failed to create $NETWORK: every candidate subnet is in use" >&2
+    exit 1
+fi
+
+# Derive container IPs from the network's assigned subnet. Slots: .2 lighthouse1,
+# .3 host2 — matches the placeholders in build.sh.
+SUBNET="$(docker network inspect -f '{{(index .IPAM.Config 0).Subnet}}' "$NETWORK")"
+PREFIX="${SUBNET%/*}"
+PREFIX="${PREFIX%.*}"
+LIGHTHOUSE_IP="$PREFIX.2"
+HOST2_IP="$PREFIX.3"
+
+# Sed the placeholder TEST-NET-3 IPs in the host configs to the real ones.
+# This must happen before `vagrant up` rsyncs build/ into the VM for host3.
+for f in build/host2.yml build/host3.yml; do
+    sed "s|203\.0\.113\.|$PREFIX.|g" "$f" >"$f.tmp"
+    mv "$f.tmp" "$f"
+done
+
 CONTAINER="nebula:${NAME:-smoke}"
 
 docker run --name lighthouse1 --rm "$CONTAINER" -config lighthouse1.yml -test
-docker run --name host2 --rm "$CONTAINER" -config host2.yml -test
+docker run --name host2 --rm -v "$PWD/build/host2.yml:/nebula/host2.yml:ro" "$CONTAINER" -config host2.yml -test
 
 vagrant up
 vagrant ssh -c "cd /nebula && /nebula/$1-nebula -config host3.yml -test" -- -T
 
-docker run --name lighthouse1 --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config lighthouse1.yml 2>&1 | tee logs/lighthouse1 | sed -u 's/^/  [lighthouse1]  /' &
+docker run --name lighthouse1 --network "$NETWORK" --ip "$LIGHTHOUSE_IP" --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config lighthouse1.yml 2>&1 | tee logs/lighthouse1 | sed -u 's/^/  [lighthouse1]  /' &
 sleep 1
-docker run --name host2 --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config host2.yml 2>&1 | tee logs/host2 | sed -u 's/^/  [host2]  /' &
+docker run --name host2 --network "$NETWORK" --ip "$HOST2_IP" -v "$PWD/build/host2.yml:/nebula/host2.yml:ro" --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config host2.yml 2>&1 | tee logs/host2 | sed -u 's/^/  [host2]  /' &
 sleep 1
 vagrant ssh -c "cd /nebula && sudo sh -c 'echo \$\$ >/nebula/pid && exec /nebula/$1-nebula -config host3.yml'" 2>&1 -- -T | tee logs/host3 | sed -u 's/^/  [host3]  /' &
 sleep 15
@@ -96,7 +128,14 @@ vagrant ssh -c "ping -c1 192.168.100.2" -- -T
 vagrant ssh -c "sudo xargs kill </nebula/pid" -- -T
 docker exec host2 sh -c 'kill 1'
 docker exec lighthouse1 sh -c 'kill 1'
+
+# Wait up to 30s for all backgrounded jobs to exit. vagrant ssh in particular
+# takes a beat to tear down after nebula exits on the VM, so a fixed sleep is
+# racy.
+for _ in $(seq 1 30); do
+    [ -z "$(jobs -r)" ] && break
     sleep 1
+done
 
 if [ "$(jobs -r)" ]
 then

.github/workflows/smoke/smoke.sh

@@ -6,6 +6,8 @@ set -o pipefail
 
 mkdir -p logs
 
+NETWORK="nebula-smoke"
+
 cleanup() {
     echo
     echo " *** cleanup"
@@ -16,38 +18,71 @@ cleanup() {
     then
         docker kill lighthouse1 host2 host3 host4
     fi
+    docker network rm "$NETWORK" >/dev/null 2>&1
 }
 
 trap cleanup EXIT
 
+# Create a dedicated smoke network with an explicit subnet (required for --ip
+# below). Probe a short list of candidates so a locally-used range doesn't
+# fail the whole test — we only need one to be free.
+docker network rm "$NETWORK" >/dev/null 2>&1 || true
+for candidate in 172.30.0.0/24 172.31.0.0/24 10.98.0.0/24 10.99.0.0/24 192.168.230.0/24; do
+    if docker network create --subnet "$candidate" "$NETWORK" >/dev/null 2>&1; then
+        break
+    fi
+done
+if ! docker network inspect "$NETWORK" >/dev/null 2>&1; then
+    echo "failed to create $NETWORK: every candidate subnet is in use" >&2
+    exit 1
+fi
+
+# Derive container IPs from the network's assigned subnet. Slots: .2 lighthouse1,
+# .3 host2, .4 host3, .5 host4 — matches the placeholders in build.sh.
+SUBNET="$(docker network inspect -f '{{(index .IPAM.Config 0).Subnet}}' "$NETWORK")"
+PREFIX="${SUBNET%/*}"
+PREFIX="${PREFIX%.*}"
+LIGHTHOUSE_IP="$PREFIX.2"
+HOST2_IP="$PREFIX.3"
+HOST3_IP="$PREFIX.4"
+HOST4_IP="$PREFIX.5"
+
+# Sed the placeholder TEST-NET-3 IPs in the host configs to the real ones.
+# build/lighthouse1.yml has no IPs to rewrite so it's skipped.
+for f in build/host2.yml build/host3.yml build/host4.yml; do
+    sed "s|203\.0\.113\.|$PREFIX.|g" "$f" >"$f.tmp"
+    mv "$f.tmp" "$f"
+done
+
 CONTAINER="nebula:${NAME:-smoke}"
 
 docker run --name lighthouse1 --rm "$CONTAINER" -config lighthouse1.yml -test
-docker run --name host2 --rm "$CONTAINER" -config host2.yml -test
+docker run --name host2 --rm -v "$PWD/build/host2.yml:/nebula/host2.yml:ro" "$CONTAINER" -config host2.yml -test
-docker run --name host3 --rm "$CONTAINER" -config host3.yml -test
+docker run --name host3 --rm -v "$PWD/build/host3.yml:/nebula/host3.yml:ro" "$CONTAINER" -config host3.yml -test
-docker run --name host4 --rm "$CONTAINER" -config host4.yml -test
+docker run --name host4 --rm -v "$PWD/build/host4.yml:/nebula/host4.yml:ro" "$CONTAINER" -config host4.yml -test
 
-docker run --name lighthouse1 --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config lighthouse1.yml 2>&1 | tee logs/lighthouse1 | sed -u 's/^/  [lighthouse1]  /' &
+docker run --name lighthouse1 --network "$NETWORK" --ip "$LIGHTHOUSE_IP" --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config lighthouse1.yml 2>&1 | tee logs/lighthouse1 | sed -u 's/^/  [lighthouse1]  /' &
 sleep 1
-docker run --name host2 --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config host2.yml 2>&1 | tee logs/host2 | sed -u 's/^/  [host2]  /' &
+docker run --name host2 --network "$NETWORK" --ip "$HOST2_IP" -v "$PWD/build/host2.yml:/nebula/host2.yml:ro" --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config host2.yml 2>&1 | tee logs/host2 | sed -u 's/^/  [host2]  /' &
 sleep 1
-docker run --name host3 --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config host3.yml 2>&1 | tee logs/host3 | sed -u 's/^/  [host3]  /' &
+docker run --name host3 --network "$NETWORK" --ip "$HOST3_IP" -v "$PWD/build/host3.yml:/nebula/host3.yml:ro" --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config host3.yml 2>&1 | tee logs/host3 | sed -u 's/^/  [host3]  /' &
 sleep 1
-docker run --name host4 --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config host4.yml 2>&1 | tee logs/host4 | sed -u 's/^/  [host4]  /' &
+docker run --name host4 --network "$NETWORK" --ip "$HOST4_IP" -v "$PWD/build/host4.yml:/nebula/host4.yml:ro" --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN --rm "$CONTAINER" -config host4.yml 2>&1 | tee logs/host4 | sed -u 's/^/  [host4]  /' &
 sleep 1
 
 # grab tcpdump pcaps for debugging
-docker exec lighthouse1 tcpdump -i nebula1 -q -w - -U 2>logs/lighthouse1.inside.log >logs/lighthouse1.inside.pcap &
+docker exec lighthouse1 tcpdump -i tun0 -q -w - -U 2>logs/lighthouse1.inside.log >logs/lighthouse1.inside.pcap &
 docker exec lighthouse1 tcpdump -i eth0 -q -w - -U 2>logs/lighthouse1.outside.log >logs/lighthouse1.outside.pcap &
-docker exec host2 tcpdump -i nebula1 -q -w - -U 2>logs/host2.inside.log >logs/host2.inside.pcap &
+docker exec host2 tcpdump -i tun0 -q -w - -U 2>logs/host2.inside.log >logs/host2.inside.pcap &
 docker exec host2 tcpdump -i eth0 -q -w - -U 2>logs/host2.outside.log >logs/host2.outside.pcap &
-docker exec host3 tcpdump -i nebula1 -q -w - -U 2>logs/host3.inside.log >logs/host3.inside.pcap &
+docker exec host3 tcpdump -i tun0 -q -w - -U 2>logs/host3.inside.log >logs/host3.inside.pcap &
 docker exec host3 tcpdump -i eth0 -q -w - -U 2>logs/host3.outside.log >logs/host3.outside.pcap &
-docker exec host4 tcpdump -i nebula1 -q -w - -U 2>logs/host4.inside.log >logs/host4.inside.pcap &
+docker exec host4 tcpdump -i tun0 -q -w - -U 2>logs/host4.inside.log >logs/host4.inside.pcap &
 docker exec host4 tcpdump -i eth0 -q -w - -U 2>logs/host4.outside.log >logs/host4.outside.pcap &
 
 docker exec host2 ncat -nklv 0.0.0.0 2000 &
 docker exec host3 ncat -nklv 0.0.0.0 2000 &
+docker exec host4 ncat -e '/usr/bin/echo helloagainfromhost4' -nkluv 0.0.0.0 4000 &
 docker exec host2 ncat -e '/usr/bin/echo host2' -nkluv 0.0.0.0 3000 &
 docker exec host3 ncat -e '/usr/bin/echo host3' -nkluv 0.0.0.0 3000 &
 
@@ -119,17 +154,24 @@ echo
 echo " *** Testing conntrack"
 echo
 set -x
-# host2 can ping host3 now that host3 pinged it first
+
-docker exec host2 ping -c1 192.168.100.3
+# host4's outbound firewall only allows ICMP to the lighthouse, so host4
-# host4 can ping host2 once conntrack established
+# cannot initiate UDP to host2. Once host2 initiates a flow to host4:4000,
-docker exec host2 ping -c1 192.168.100.4
+# conntrack must let host4's listener reply on that flow. If it doesn't,
-docker exec host4 ping -c1 192.168.100.2
+# the echo back from host4 never reaches host2.
+docker exec host2 sh -c "(/usr/bin/echo host2; sleep 2) | ncat -nuv 192.168.100.4 4000" | grep -q helloagainfromhost4
 
 docker exec host4 sh -c 'kill 1'
 docker exec host3 sh -c 'kill 1'
 docker exec host2 sh -c 'kill 1'
 docker exec lighthouse1 sh -c 'kill 1'
-sleep 5
+
+# Wait up to 30s for all backgrounded jobs to exit rather than relying on a
+# fixed sleep.
+for _ in $(seq 1 30); do
+    [ -z "$(jobs -r)" ] && break
+    sleep 1
+done
 
 if [ "$(jobs -r)" ]
 then

.github/workflows/smoke/vagrant-linux-amd64-ipv6disable/Vagrantfile

@@ -1,7 +1,7 @@
 # -*- mode: ruby -*-
 # vi: set ft=ruby :
 Vagrant.configure("2") do |config|
-  config.vm.box = "ubuntu/jammy64"
+  config.vm.box = "bento/ubuntu-24.04"
 
   config.vm.synced_folder "../build", "/nebula"
 

.github/workflows/smoke/vagrant-openbsd-amd64/Vagrantfile

@@ -1,7 +1,7 @@
 # -*- mode: ruby -*-
 # vi: set ft=ruby :
 Vagrant.configure("2") do |config|
-  config.vm.box = "generic/openbsd7"
+  config.vm.box = "DefinedNet/openbsd78"
 
   config.vm.synced_folder "../build", "/nebula", type: "rsync"
 end

.github/workflows/test.yml

@@ -18,7 +18,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
 
-    - uses: actions/checkout@v4
+    - uses: actions/checkout@v6
 
     - uses: actions/setup-go@v6
       with:
@@ -32,7 +32,7 @@ jobs:
       run: make vet
 
     - name: golangci-lint
-      uses: golangci/golangci-lint-action@v8
+      uses: golangci/golangci-lint-action@v9
       with:
         version: v2.5
 
@@ -45,7 +45,7 @@ jobs:
     - name: Build test mobile
       run: make build-test-mobile
 
-    - uses: actions/upload-artifact@v4
+    - uses: actions/upload-artifact@v6
       with:
         name: e2e packet flow linux-latest
         path: e2e/mermaid/linux-latest
@@ -56,7 +56,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
 
-    - uses: actions/checkout@v4
+    - uses: actions/checkout@v6
 
     - uses: actions/setup-go@v6
       with:
@@ -77,7 +77,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
 
-    - uses: actions/checkout@v4
+    - uses: actions/checkout@v6
 
     - uses: actions/setup-go@v6
       with:
@@ -98,7 +98,7 @@ jobs:
         os: [windows-latest, macos-latest]
     steps:
 
-    - uses: actions/checkout@v4
+    - uses: actions/checkout@v6
 
     - uses: actions/setup-go@v6
       with:
@@ -115,7 +115,7 @@ jobs:
       run: make vet
 
     - name: golangci-lint
-      uses: golangci/golangci-lint-action@v8
+      uses: golangci/golangci-lint-action@v9
       with:
         version: v2.5
 
@@ -125,7 +125,7 @@ jobs:
     - name: End 2 end
       run: make e2evv
 
-    - uses: actions/upload-artifact@v4
+    - uses: actions/upload-artifact@v6
       with:
         name: e2e packet flow ${{ matrix.os }}
         path: e2e/mermaid/${{ matrix.os }}

.golangci.yaml

@@ -2,7 +2,21 @@ version: "2"
 linters:
   default: none
   enable:
+    - sloglint
     - testifylint
+  settings:
+    sloglint:
+      # Enforce key-value pair form for Info/Debug/Warn/Error/Log/With and
+      # the package-level slog equivalents. Use l.Log(ctx, level, ...) for
+      # custom levels instead of LogAttrs when you can.
+      #
+      # LogAttrs is also flagged by this rule because it takes ...slog.Attr;
+      # the few legitimate sites (where attrs is built up as a []slog.Attr)
+      # carry a //nolint:sloglint with rationale.
+      kv-only: true
+      # no-mixed-args is on by default: forbids mixing kv and attrs in one call.
+      # discard-handler is on by default (since Go 1.24): suggests
+      # slog.DiscardHandler over slog.NewTextHandler(io.Discard, nil).
   exclusions:
     generated: lax
     presets:

CHANGELOG.md

@@ -7,12 +7,129 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [1.10.3] - 2026-02-06
+
+### Security
+
+- Fix an issue where blocklist bypass is possible when using curve P256 since the signature can have 2 valid representations.
+  Both fingerprint representations will be tested against the blocklist.
+  Any newly issued P256 based certificates will have their signature clamped to the low-s form.
+  Nebula will assert the low-s signature form when validating certificates in a future version. [GHSA-69x3-g4r3-p962](https://github.com/slackhq/nebula/security/advisories/GHSA-69x3-g4r3-p962)
+
 ### Changed
 
-- `default_local_cidr_any` now defaults to false, meaning that any firewall rule
+- Improve error reporting if nebula fails to start due to a tun device naming issue. (#1588)
+
+## [1.10.2] - 2026-01-21
+
+### Fixed
+
+- Fix panic when using `use_system_route_table` that was introduced in v1.10.1. (#1580)
+
+### Changed
+
+- Fix some typos in comments. (#1582)
+- Dependency updates. (#1581)
+
+## [1.10.1] - 2026-01-16
+
+See the [v1.10.1](https://github.com/slackhq/nebula/milestone/26?closed=1) milestone for a complete list of changes.
+
+### Fixed
+
+- Fix a bug where an unsafe route derived from the system route table could be lost on a config reload. (#1573)
+- Fix the PEM banner for ECDSA P256 public keys. (#1552)
+- Fix a regression on Windows from 1.9.x where nebula could fall back to a less performant UDP listener if 
+  non-critical ioctls failed. (#1568)
+- Fix a bug in handshake processing when a peer sends an unexpected public key. (#1566)
+
+### Added
+
+- Add a config option to control accepting `recv_error` packets which defaults to `always`. (#1569)
+
+### Changed
+
+- Various dependency updates. (#1541, #1549, #1550, #1557, #1558, #1560, #1561, #1570, #1571)
+
+## [1.10.0] - 2025-12-04
+
+See the [v1.10.0](https://github.com/slackhq/nebula/milestone/16?closed=1) milestone for a complete list of changes.
+
+### Added
+
+- Support for ipv6 and multiple ipv4/6 addresses in the overlay.
+  A new v2 ASN.1 based certificate format.
+  Certificates now have a unified interface for external implementations.
+  (#1212, #1216, #1345, #1359, #1381, #1419, #1464, #1466, #1451, #1476, #1467, #1481, #1399, #1488, #1492, #1495, #1468, #1521, #1535, #1538)
+- Add the ability to mark packets on linux to better target nebula packets in iptables/nftables. (#1331)
+- Add ECMP support for `unsafe_routes`. (#1332)
+- PKCS11 support for P256 keys when built with `pkcs11` tag (#1153, #1482)
+
+### Changed
+
+- **NOTE**: `default_local_cidr_any` now defaults to false, meaning that any firewall rule
   intended to target an `unsafe_routes` entry must explicitly declare it via the
   `local_cidr` field. This is almost always the intended behavior. This flag is
-  deprecated and will be removed in a future release.
+  deprecated and will be removed in a future release. (#1373)
+- Improve logging when a relay is in use on an inbound packet. (#1533)
+- Avoid fatal errors if `rountines` is > 1 on systems that don't support more than 1 routine. (#1531)
+- Log a warning if a firewall rule contains an `any` that negates a more restrictive filter. (#1513)
+- Accept encrypted CA passphrase from an environment variable. (#1421)
+- Allow handshaking with any trusted remote. (#1509)
+- Log only the count of blocklisted certificate fingerprints instead of the entire list. (#1525)
+- Don't fatal when the ssh server is unable to be configured successfully. (#1520)
+- Update to build against go v1.25. (#1483)
+- Allow projects using `nebula` as a library with userspace networking to configure the `logger` and build version. (#1239) 
+- Upgrade to `yaml.v3`. (#1148, #1371, #1438, #1478)
+
+### Fixed
+
+- Fix a potential bug with udp ipv4 only on darwin. (#1532)
+- Improve lost packet statistics. (#1441, #1537)
+- Honor `remote_allow_list` in hole punch response. (#1186)
+- Fix a panic when `tun.use_system_route_table` is `true` and a route lacks a destination. (#1437) 
+- Fix an issue when `tun.use_system_route_table: true` could result in heavy CPU utilization when many thousands of routes
+  are present. (#1326) 
+- Fix tests for 32 bit machines. (#1394)
+- Fix a possible 32bit integer underflow in config handling. (#1353)
+- Fix moving a udp address from one vpn address to another in the `static_host_map`
+  which could cause rapid re-handshaking with an incorrect remote. (#1259)
+- Improve smoke tests in environments where the docker network is not the default. (#1347)
+
+## [1.9.7] - 2025-10-10
+
+### Security
+
+- Fix an issue where Nebula could incorrectly accept and process a packet from an erroneous source IP when the sender's
+  certificate is configured with unsafe_routes (cert v1/v2) or multiple IPs (cert v2). (#1494)
+
+### Changed
+
+- Disable sending `recv_error` messages when a packet is received outside the allowable counter window. (#1459)
+- Improve error messages and remove some unnecessary fatal conditions in the Windows and generic udp listener. (#1453)
+
+## [1.9.6] - 2025-7-15
+
+### Added
+
+- Support dropping inactive tunnels. This is disabled by default in this release but can be enabled with `tunnels.drop_inactive`. See example config for more details. (#1413)
+
+### Fixed
+
+- Fix Darwin freeze due to presence of some Network Extensions (#1426)
+- Ensure the same relay tunnel is always used when multiple relay tunnels are present (#1422)
+- Fix Windows freeze due to ICMP error handling (#1412)
+- Fix relay migration panic (#1403)
+
+## [1.9.5] - 2024-12-05
+
+### Added
+
+- Gracefully ignore v2 certificates. (#1282)
+
+### Fixed
+
+- Fix relays that refuse to re-establish after one of the remote tunnel pairs breaks. (#1277)
 
 ## [1.9.4] - 2024-09-09
 
@@ -671,7 +788,14 @@ created.)
 
 - Initial public release.
 
-[Unreleased]: https://github.com/slackhq/nebula/compare/v1.9.4...HEAD
+[Unreleased]: https://github.com/slackhq/nebula/compare/v1.10.3...HEAD
+[1.10.3]: https://github.com/slackhq/nebula/releases/tag/v1.10.3
+[1.10.2]: https://github.com/slackhq/nebula/releases/tag/v1.10.2
+[1.10.1]: https://github.com/slackhq/nebula/releases/tag/v1.10.1
+[1.10.0]: https://github.com/slackhq/nebula/releases/tag/v1.10.0
+[1.9.7]: https://github.com/slackhq/nebula/releases/tag/v1.9.7
+[1.9.6]: https://github.com/slackhq/nebula/releases/tag/v1.9.6
+[1.9.5]: https://github.com/slackhq/nebula/releases/tag/v1.9.5
 [1.9.4]: https://github.com/slackhq/nebula/releases/tag/v1.9.4
 [1.9.3]: https://github.com/slackhq/nebula/releases/tag/v1.9.3
 [1.9.2]: https://github.com/slackhq/nebula/releases/tag/v1.9.2

CODEOWNERS

@@ -0,0 +1 @@
+#ECCN:Open Source

README.md

@@ -57,7 +57,7 @@ Check the [releases](https://github.com/slackhq/nebula/releases/latest) page for
     docker pull nebulaoss/nebula
     ```
 
-#### Mobile
+#### Mobile ([source code](https://github.com/DefinedNet/mobile_nebula))
 
 - [iOS](https://apps.apple.com/us/app/mobile-nebula/id1509587936?itsct=apps_box&itscg=30200)
 - [Android](https://play.google.com/store/apps/details?id=net.defined.mobile_nebula&pcampaignid=pcampaignidMKT-Other-global-all-co-prtnr-py-PartBadge-Mar2515-1)
@@ -76,6 +76,8 @@ Nebula was created to provide a mechanism for groups of hosts to communicate sec
 
 ## Getting started (quickly)
 
+**Don't want to manage your own PKI and lighthouses?** [Managed Nebula](https://www.defined.net/) from Defined Networking handles all of this for you.
+
 To set up a Nebula network, you'll need:
 
 #### 1. The [Nebula binaries](https://github.com/slackhq/nebula/releases) or [Distribution Packages](https://github.com/slackhq/nebula#distribution-packages) for your specific platform. Specifically you'll need `nebula-cert` and the specific nebula binary for each platform you use.

batch_pipeline.go

@@ -1,164 +0,0 @@
-package nebula
-
-import (
-	"net/netip"
-
-	"github.com/slackhq/nebula/overlay"
-	"github.com/slackhq/nebula/udp"
-)
-
-// batchPipelines tracks whether the inside device can operate on packet batches
-// and, if so, holds the shared packet pool sized for the virtio headroom and
-// payload limits advertised by the device. It also owns the fan-in/fan-out
-// queues between the TUN readers, encrypt/decrypt workers, and the UDP writers.
-type batchPipelines struct {
-	enabled    bool
-	inside     overlay.BatchCapableDevice
-	headroom   int
-	payloadCap int
-	pool       *overlay.PacketPool
-	batchSize  int
-	routines   int
-	rxQueues   []chan *overlay.Packet
-	txQueues   []chan queuedDatagram
-	tunQueues  []chan *overlay.Packet
-}
-
-type queuedDatagram struct {
-	packet *overlay.Packet
-	addr   netip.AddrPort
-}
-
-func (bp *batchPipelines) init(device overlay.Device, routines int, queueDepth int, maxSegments int) {
-	if device == nil || routines <= 0 {
-		return
-	}
-	bcap, ok := device.(overlay.BatchCapableDevice)
-	if !ok {
-		return
-	}
-	headroom := bcap.BatchHeadroom()
-	payload := bcap.BatchPayloadCap()
-	if maxSegments < 1 {
-		maxSegments = 1
-	}
-	requiredPayload := udp.MTU * maxSegments
-	if payload < requiredPayload {
-		payload = requiredPayload
-	}
-	batchSize := bcap.BatchSize()
-	if headroom <= 0 || payload <= 0 || batchSize <= 0 {
-		return
-	}
-	bp.enabled = true
-	bp.inside = bcap
-	bp.headroom = headroom
-	bp.payloadCap = payload
-	bp.batchSize = batchSize
-	bp.routines = routines
-	bp.pool = overlay.NewPacketPool(headroom, payload)
-	queueCap := batchSize * defaultBatchQueueDepthFactor
-	if queueDepth > 0 {
-		queueCap = queueDepth
-	}
-	if queueCap < batchSize {
-		queueCap = batchSize
-	}
-	bp.rxQueues = make([]chan *overlay.Packet, routines)
-	bp.txQueues = make([]chan queuedDatagram, routines)
-	bp.tunQueues = make([]chan *overlay.Packet, routines)
-	for i := 0; i < routines; i++ {
-		bp.rxQueues[i] = make(chan *overlay.Packet, queueCap)
-		bp.txQueues[i] = make(chan queuedDatagram, queueCap)
-		bp.tunQueues[i] = make(chan *overlay.Packet, queueCap)
-	}
-}
-
-func (bp *batchPipelines) Pool() *overlay.PacketPool {
-	if bp == nil || !bp.enabled {
-		return nil
-	}
-	return bp.pool
-}
-
-func (bp *batchPipelines) Enabled() bool {
-	return bp != nil && bp.enabled
-}
-
-func (bp *batchPipelines) batchSizeHint() int {
-	if bp == nil || bp.batchSize <= 0 {
-		return 1
-	}
-	return bp.batchSize
-}
-
-func (bp *batchPipelines) rxQueue(i int) chan *overlay.Packet {
-	if bp == nil || !bp.enabled || i < 0 || i >= len(bp.rxQueues) {
-		return nil
-	}
-	return bp.rxQueues[i]
-}
-
-func (bp *batchPipelines) txQueue(i int) chan queuedDatagram {
-	if bp == nil || !bp.enabled || i < 0 || i >= len(bp.txQueues) {
-		return nil
-	}
-	return bp.txQueues[i]
-}
-
-func (bp *batchPipelines) tunQueue(i int) chan *overlay.Packet {
-	if bp == nil || !bp.enabled || i < 0 || i >= len(bp.tunQueues) {
-		return nil
-	}
-	return bp.tunQueues[i]
-}
-
-func (bp *batchPipelines) txQueueLen(i int) int {
-	q := bp.txQueue(i)
-	if q == nil {
-		return 0
-	}
-	return len(q)
-}
-
-func (bp *batchPipelines) tunQueueLen(i int) int {
-	q := bp.tunQueue(i)
-	if q == nil {
-		return 0
-	}
-	return len(q)
-}
-
-func (bp *batchPipelines) enqueueRx(i int, pkt *overlay.Packet) bool {
-	q := bp.rxQueue(i)
-	if q == nil {
-		return false
-	}
-	q <- pkt
-	return true
-}
-
-func (bp *batchPipelines) enqueueTx(i int, pkt *overlay.Packet, addr netip.AddrPort) bool {
-	q := bp.txQueue(i)
-	if q == nil {
-		return false
-	}
-	q <- queuedDatagram{packet: pkt, addr: addr}
-	return true
-}
-
-func (bp *batchPipelines) enqueueTun(i int, pkt *overlay.Packet) bool {
-	q := bp.tunQueue(i)
-	if q == nil {
-		return false
-	}
-	q <- pkt
-	return true
-}
-
-func (bp *batchPipelines) newPacket() *overlay.Packet {
-	if bp == nil || !bp.enabled || bp.pool == nil {
-		return nil
-	}
-	return bp.pool.Get()
-}

bits.go

@@ -1,22 +1,23 @@
 package nebula
 
 import (
+	"context"
+	"log/slog"
+
 	"github.com/rcrowley/go-metrics"
-	"github.com/sirupsen/logrus"
 )
 
 type Bits struct {
 	length             uint64
 	current            uint64
 	bits               []bool
-	firstSeen          bool
 	lostCounter        metrics.Counter
 	dupeCounter        metrics.Counter
 	outOfWindowCounter metrics.Counter
 }
 
 func NewBits(bits uint64) *Bits {
-	return &Bits{
+	b := &Bits{
 		length:             bits,
 		bits:               make([]bool, bits, bits),
 		current:            0,
@@ -24,34 +25,40 @@ func NewBits(bits uint64) *Bits {
 		dupeCounter:        metrics.GetOrRegisterCounter("network.packets.duplicate", nil),
 		outOfWindowCounter: metrics.GetOrRegisterCounter("network.packets.out_of_window", nil),
 	}
+
+	// There is no counter value 0, mark it to avoid counting a lost packet later.
+	b.bits[0] = true
+	b.current = 0
+	return b
 }
 
-func (b *Bits) Check(l logrus.FieldLogger, i uint64) bool {
+func (b *Bits) Check(l *slog.Logger, i uint64) bool {
 	// If i is the next number, return true.
-	if i > b.current || (i == 0 && b.firstSeen == false && b.current < b.length) {
+	if i > b.current {
 		return true
 	}
 
-	// If i is within the window, check if it's been set already. The first window will fail this check
+	// If i is within the window, check if it's been set already.
-	if i > b.current-b.length {
+	if i > b.current-b.length || i < b.length && b.current < b.length {
-		return !b.bits[i%b.length]
-	}
-
-	// If i is within the first window
-	if i < b.length {
 		return !b.bits[i%b.length]
 	}
 
 	// Not within the window
-	l.Debugf("rejected a packet (top) %d %d\n", b.current, i)
+	if l.Enabled(context.Background(), slog.LevelDebug) {
+		l.Debug("rejected a packet (top)",
+			"current", b.current,
+			"incoming", i,
+		)
+	}
 	return false
 }
 
-func (b *Bits) Update(l *logrus.Logger, i uint64) bool {
+func (b *Bits) Update(l *slog.Logger, i uint64) bool {
 	// If i is the next number, return true and update current.
 	if i == b.current+1 {
-		// Report missed packets, we can only understand what was missed after the first window has been gone through
+		// Check if the oldest bit was lost since we are shifting the window by 1 and occupying it with this counter
-		if i > b.length && b.bits[i%b.length] == false {
+		// The very first window can only be tracked as lost once we are on the 2nd window or greater
+		if b.bits[i%b.length] == false && i > b.length {
 			b.lostCounter.Inc(1)
 		}
 		b.bits[i%b.length] = true
@@ -59,73 +66,39 @@ func (b *Bits) Update(l *logrus.Logger, i uint64) bool {
 		return true
 	}
 
-	// If i packet is greater than current but less than the maximum length of our bitmap,
+	// If i is a jump, adjust the window, record lost, update current, and return true
-	// flip everything in between to false and move ahead.
+	if i > b.current {
-	if i > b.current && i < b.current+b.length {
+		lost := int64(0)
-		// In between current and i need to be zero'd to allow those packets to come in later
+		// Zero out the bits between the current and the new counter value, limited by the window size,
-		for n := b.current + 1; n < i; n++ {
+		// since the window is shifting
+		for n := b.current + 1; n <= min(i, b.current+b.length); n++ {
+			if b.bits[n%b.length] == false && n > b.length {
+				lost++
+			}
 			b.bits[n%b.length] = false
 		}
 
-		b.bits[i%b.length] = true
+		// Only record any skipped packets as a result of the window moving further than the window length
-		b.current = i
+		// Any loss within the new window will be accounted for in future calls
-		//l.Debugf("missed %d packets between %d and %d\n", i-b.current, i, b.current)
+		lost += max(0, int64(i-b.current-b.length))
-		return true
-	}
-
-	// If i is greater than the delta between current and the total length of our bitmap,
-	// just flip everything in the map and move ahead.
-	if i >= b.current+b.length {
-		// The current window loss will be accounted for later, only record the jump as loss up until then
-		lost := maxInt64(0, int64(i-b.current-b.length))
-		//TODO: explain this
-		if b.current == 0 {
-			lost++
-		}
-
-		for n := range b.bits {
-			// Don't want to count the first window as a loss
-			//TODO: this is likely wrong, we are wanting to track only the bit slots that we aren't going to track anymore and this is marking everything as missed
-			//if b.bits[n] == false {
-			//	lost++
-			//}
-			b.bits[n] = false
-		}
-
 		b.lostCounter.Inc(lost)
 
-		if l.Level >= logrus.DebugLevel {
-			l.WithField("receiveWindow", m{"accepted": true, "currentCounter": b.current, "incomingCounter": i, "reason": "window shifting"}).
-				Debug("Receive window")
-		}
 		b.bits[i%b.length] = true
 		b.current = i
 		return true
 	}
 
-	// Allow for the 0 packet to come in within the first window
+	// If i is within the current window but below the current counter,
-	if i == 0 && b.firstSeen == false && b.current < b.length {
+	// Check to see if it's a duplicate
-		b.firstSeen = true
+	if i > b.current-b.length || i < b.length && b.current < b.length {
-		b.bits[i%b.length] = true
+		if b.current == i || b.bits[i%b.length] == true {
-		return true
+			if l.Enabled(context.Background(), slog.LevelDebug) {
-	}
+				l.Debug("Receive window",
-
+					"accepted", false,
-	// If i is within the window of current minus length (the total pat window size),
+					"currentCounter", b.current,
-	// allow it and flip to true but to NOT change current. We also have to account for the first window
+					"incomingCounter", i,
-	if ((b.current >= b.length && i > b.current-b.length) || (b.current < b.length && i < b.length)) && i <= b.current {
+					"reason", "duplicate",
-		if b.current == i {
+				)
-			if l.Level >= logrus.DebugLevel {
-				l.WithField("receiveWindow", m{"accepted": false, "currentCounter": b.current, "incomingCounter": i, "reason": "duplicate"}).
-					Debug("Receive window")
-			}
-			b.dupeCounter.Inc(1)
-			return false
-		}
-
-		if b.bits[i%b.length] == true {
-			if l.Level >= logrus.DebugLevel {
-				l.WithField("receiveWindow", m{"accepted": false, "currentCounter": b.current, "incomingCounter": i, "reason": "old duplicate"}).
-					Debug("Receive window")
 			}
 			b.dupeCounter.Inc(1)
 			return false
@@ -133,25 +106,17 @@ func (b *Bits) Update(l *logrus.Logger, i uint64) bool {
 
 		b.bits[i%b.length] = true
 		return true
-
 	}
 
 	// In all other cases, fail and don't change current.
 	b.outOfWindowCounter.Inc(1)
-	if l.Level >= logrus.DebugLevel {
+	if l.Enabled(context.Background(), slog.LevelDebug) {
-		l.WithField("accepted", false).
+		l.Debug("Receive window",
-			WithField("currentCounter", b.current).
+			"accepted", false,
-			WithField("incomingCounter", i).
+			"currentCounter", b.current,
-			WithField("reason", "nonsense").
+			"incomingCounter", i,
-			Debug("Receive window")
+			"reason", "nonsense",
+		)
 	}
 	return false
 }
-
-func maxInt64(a, b int64) int64 {
-	if a > b {
-		return a
-	}
-
-	return b
-}

bits_test.go

@@ -15,48 +15,41 @@ func TestBits(t *testing.T) {
 	assert.Len(t, b.bits, 10)
 
 	// This is initialized to zero - receive one. This should work.
-
 	assert.True(t, b.Check(l, 1))
-	u := b.Update(l, 1)
+	assert.True(t, b.Update(l, 1))
-	assert.True(t, u)
 	assert.EqualValues(t, 1, b.current)
-	g := []bool{false, true, false, false, false, false, false, false, false, false}
+	g := []bool{true, true, false, false, false, false, false, false, false, false}
 	assert.Equal(t, g, b.bits)
 
 	// Receive two
 	assert.True(t, b.Check(l, 2))
-	u = b.Update(l, 2)
+	assert.True(t, b.Update(l, 2))
-	assert.True(t, u)
 	assert.EqualValues(t, 2, b.current)
-	g = []bool{false, true, true, false, false, false, false, false, false, false}
+	g = []bool{true, true, true, false, false, false, false, false, false, false}
 	assert.Equal(t, g, b.bits)
 
 	// Receive two again - it will fail
 	assert.False(t, b.Check(l, 2))
-	u = b.Update(l, 2)
+	assert.False(t, b.Update(l, 2))
-	assert.False(t, u)
 	assert.EqualValues(t, 2, b.current)
 
 	// Jump ahead to 15, which should clear everything and set the 6th element
 	assert.True(t, b.Check(l, 15))
-	u = b.Update(l, 15)
+	assert.True(t, b.Update(l, 15))
-	assert.True(t, u)
 	assert.EqualValues(t, 15, b.current)
 	g = []bool{false, false, false, false, false, true, false, false, false, false}
 	assert.Equal(t, g, b.bits)
 
 	// Mark 14, which is allowed because it is in the window
 	assert.True(t, b.Check(l, 14))
-	u = b.Update(l, 14)
+	assert.True(t, b.Update(l, 14))
-	assert.True(t, u)
 	assert.EqualValues(t, 15, b.current)
 	g = []bool{false, false, false, false, true, true, false, false, false, false}
 	assert.Equal(t, g, b.bits)
 
 	// Mark 5, which is not allowed because it is not in the window
 	assert.False(t, b.Check(l, 5))
-	u = b.Update(l, 5)
+	assert.False(t, b.Update(l, 5))
-	assert.False(t, u)
 	assert.EqualValues(t, 15, b.current)
 	g = []bool{false, false, false, false, true, true, false, false, false, false}
 	assert.Equal(t, g, b.bits)
@@ -69,10 +62,29 @@ func TestBits(t *testing.T) {
 
 	// Walk through a few windows in order
 	b = NewBits(10)
-	for i := uint64(0); i <= 100; i++ {
+	for i := uint64(1); i <= 100; i++ {
 		assert.True(t, b.Check(l, i), "Error while checking %v", i)
 		assert.True(t, b.Update(l, i), "Error while updating %v", i)
 	}
+
+	assert.False(t, b.Check(l, 1), "Out of window check")
+}
+
+func TestBitsLargeJumps(t *testing.T) {
+	l := test.NewLogger()
+	b := NewBits(10)
+	b.lostCounter.Clear()
+
+	b = NewBits(10)
+	b.lostCounter.Clear()
+	assert.True(t, b.Update(l, 55)) // We saw packet 55 and can still track 45,46,47,48,49,50,51,52,53,54
+	assert.Equal(t, int64(45), b.lostCounter.Count())
+
+	assert.True(t, b.Update(l, 100)) // We saw packet 55 and 100 and can still track 90,91,92,93,94,95,96,97,98,99
+	assert.Equal(t, int64(89), b.lostCounter.Count())
+
+	assert.True(t, b.Update(l, 200)) // We saw packet 55, 100, and 200 and can still track 190,191,192,193,194,195,196,197,198,199
+	assert.Equal(t, int64(188), b.lostCounter.Count())
 }
 
 func TestBitsDupeCounter(t *testing.T) {
@@ -124,8 +136,7 @@ func TestBitsOutOfWindowCounter(t *testing.T) {
 	assert.False(t, b.Update(l, 0))
 	assert.Equal(t, int64(1), b.outOfWindowCounter.Count())
 
-	//tODO: make sure lostcounter doesn't increase in orderly increment
+	assert.Equal(t, int64(19), b.lostCounter.Count()) // packet 0 wasn't lost
-	assert.Equal(t, int64(20), b.lostCounter.Count())
 	assert.Equal(t, int64(0), b.dupeCounter.Count())
 	assert.Equal(t, int64(1), b.outOfWindowCounter.Count())
 }
@@ -137,8 +148,6 @@ func TestBitsLostCounter(t *testing.T) {
 	b.dupeCounter.Clear()
 	b.outOfWindowCounter.Clear()
 
-	//assert.True(t, b.Update(0))
-	assert.True(t, b.Update(l, 0))
 	assert.True(t, b.Update(l, 20))
 	assert.True(t, b.Update(l, 21))
 	assert.True(t, b.Update(l, 22))
@@ -149,7 +158,7 @@ func TestBitsLostCounter(t *testing.T) {
 	assert.True(t, b.Update(l, 27))
 	assert.True(t, b.Update(l, 28))
 	assert.True(t, b.Update(l, 29))
-	assert.Equal(t, int64(20), b.lostCounter.Count())
+	assert.Equal(t, int64(19), b.lostCounter.Count()) // packet 0 wasn't lost
 	assert.Equal(t, int64(0), b.dupeCounter.Count())
 	assert.Equal(t, int64(0), b.outOfWindowCounter.Count())
 
@@ -158,8 +167,6 @@ func TestBitsLostCounter(t *testing.T) {
 	b.dupeCounter.Clear()
 	b.outOfWindowCounter.Clear()
 
-	assert.True(t, b.Update(l, 0))
-	assert.Equal(t, int64(0), b.lostCounter.Count())
 	assert.True(t, b.Update(l, 9))
 	assert.Equal(t, int64(0), b.lostCounter.Count())
 	// 10 will set 0 index, 0 was already set, no lost packets
@@ -214,6 +221,62 @@ func TestBitsLostCounter(t *testing.T) {
 	assert.Equal(t, int64(0), b.outOfWindowCounter.Count())
 }
 
+func TestBitsLostCounterIssue1(t *testing.T) {
+	l := test.NewLogger()
+	b := NewBits(10)
+	b.lostCounter.Clear()
+	b.dupeCounter.Clear()
+	b.outOfWindowCounter.Clear()
+
+	assert.True(t, b.Update(l, 4))
+	assert.Equal(t, int64(0), b.lostCounter.Count())
+	assert.True(t, b.Update(l, 1))
+	assert.Equal(t, int64(0), b.lostCounter.Count())
+	assert.True(t, b.Update(l, 9))
+	assert.Equal(t, int64(0), b.lostCounter.Count())
+	assert.True(t, b.Update(l, 2))
+	assert.Equal(t, int64(0), b.lostCounter.Count())
+	assert.True(t, b.Update(l, 3))
+	assert.Equal(t, int64(0), b.lostCounter.Count())
+	assert.True(t, b.Update(l, 5))
+	assert.Equal(t, int64(0), b.lostCounter.Count())
+	assert.True(t, b.Update(l, 6))
+	assert.Equal(t, int64(0), b.lostCounter.Count())
+	assert.True(t, b.Update(l, 7))
+	assert.Equal(t, int64(0), b.lostCounter.Count())
+	// assert.True(t, b.Update(l, 8))
+	assert.True(t, b.Update(l, 10))
+	assert.Equal(t, int64(0), b.lostCounter.Count())
+	assert.True(t, b.Update(l, 11))
+	assert.Equal(t, int64(0), b.lostCounter.Count())
+
+	assert.True(t, b.Update(l, 14))
+	assert.Equal(t, int64(0), b.lostCounter.Count())
+	// Issue seems to be here, we reset missing packet 8 to false here and don't increment the lost counter
+	assert.True(t, b.Update(l, 19))
+	assert.Equal(t, int64(1), b.lostCounter.Count())
+	assert.True(t, b.Update(l, 12))
+	assert.Equal(t, int64(1), b.lostCounter.Count())
+	assert.True(t, b.Update(l, 13))
+	assert.Equal(t, int64(1), b.lostCounter.Count())
+	assert.True(t, b.Update(l, 15))
+	assert.Equal(t, int64(1), b.lostCounter.Count())
+	assert.True(t, b.Update(l, 16))
+	assert.Equal(t, int64(1), b.lostCounter.Count())
+	assert.True(t, b.Update(l, 17))
+	assert.Equal(t, int64(1), b.lostCounter.Count())
+	assert.True(t, b.Update(l, 18))
+	assert.Equal(t, int64(1), b.lostCounter.Count())
+	assert.True(t, b.Update(l, 20))
+	assert.Equal(t, int64(1), b.lostCounter.Count())
+	assert.True(t, b.Update(l, 21))
+
+	// We missed packet 8 above
+	assert.Equal(t, int64(1), b.lostCounter.Count())
+	assert.Equal(t, int64(0), b.dupeCounter.Count())
+	assert.Equal(t, int64(0), b.outOfWindowCounter.Count())
+}
+
 func BenchmarkBits(b *testing.B) {
 	z := NewBits(10)
 	for n := 0; n < b.N; n++ {

boring.go

@@ -1,5 +1,4 @@
 //go:build boringcrypto
-// +build boringcrypto
 
 package nebula
 

cert/ca_pool.go

@@ -1,11 +1,14 @@
 package cert
 
 import (
+	"bufio"
+	"bytes"
+	"encoding/pem"
 	"errors"
 	"fmt"
+	"io"
 	"net/netip"
 	"slices"
-	"strings"
 	"time"
 )
 
@@ -29,22 +32,46 @@ func NewCAPool() *CAPool {
 // If the pool contains any expired certificates, an ErrExpired will be
 // returned along with the pool. The caller must handle any such errors.
 func NewCAPoolFromPEM(caPEMs []byte) (*CAPool, error) {
-	pool := NewCAPool()
+	return NewCAPoolFromPEMReader(bytes.NewReader(caPEMs))
-	var err error
-	var expired bool
-	for {
-		caPEMs, err = pool.AddCAFromPEM(caPEMs)
-		if errors.Is(err, ErrExpired) {
-			expired = true
-			err = nil
 }
+
+// NewCAPoolFromPEMReader will create a new CA pool from the provided reader.
+// The reader must contain a PEM-encoded set of nebula certificates.
+func NewCAPoolFromPEMReader(r io.Reader) (*CAPool, error) {
+	pool := NewCAPool()
+
+	var expired bool
+
+	scanner := bufio.NewScanner(r)
+	scanner.Split(SplitPEM)
+
+	for scanner.Scan() {
+		pemBytes := scanner.Bytes()
+
+		block, rest := pem.Decode(pemBytes)
+		if len(bytes.TrimSpace(rest)) > 0 {
+			return nil, ErrInvalidPEMBlock
+		}
+		if block == nil {
+			return nil, ErrInvalidPEMBlock
+		}
+
+		c, err := unmarshalCertificateBlock(block)
 		if err != nil {
 			return nil, err
 		}
-		if len(caPEMs) == 0 || strings.TrimSpace(string(caPEMs)) == "" {
+
-			break
+		err = pool.AddCA(c)
+		if errors.Is(err, ErrExpired) {
+			expired = true
+			continue
+		} else if err != nil {
+			return nil, err
 		}
 	}
+	if err := scanner.Err(); err != nil {
+		return nil, ErrInvalidPEMBlock
+	}
 
 	if expired {
 		return pool, ErrExpired
@@ -141,10 +168,23 @@ func (ncp *CAPool) VerifyCertificate(now time.Time, c Certificate) (*CachedCerti
 		return nil, err
 	}
 
+	// Pre nebula v1.10.3 could generate signatures in either high or low s form and validation
+	// of signatures allowed for either. Nebula v1.10.3 and beyond clamps signature generation to low-s form
+	// but validation still allows for either. Since a change in the signature bytes affects the fingerprint, we
+	// need to test both forms until such a time comes that we enforce low-s form on signature validation.
+	fp2, err := CalculateAlternateFingerprint(c)
+	if err != nil {
+		return nil, fmt.Errorf("could not calculate alternate fingerprint to verify: %w", err)
+	}
+	if fp2 != "" && ncp.IsBlocklisted(fp2) {
+		return nil, ErrBlockListed
+	}
+
 	cc := CachedCertificate{
 		Certificate:       c,
 		InvertedGroups:    make(map[string]struct{}),
 		Fingerprint:       fp,
+		fingerprint2:      fp2,
 		signerFingerprint: signer.Fingerprint,
 	}
 
@@ -158,6 +198,11 @@ func (ncp *CAPool) VerifyCertificate(now time.Time, c Certificate) (*CachedCerti
 // VerifyCachedCertificate is the same as VerifyCertificate other than it operates on a pre-verified structure and
 // is a cheaper operation to perform as a result.
 func (ncp *CAPool) VerifyCachedCertificate(now time.Time, c *CachedCertificate) error {
+	// Check any available alternate fingerprint forms for this certificate, re P256 high-s/low-s
+	if c.fingerprint2 != "" && ncp.IsBlocklisted(c.fingerprint2) {
+		return ErrBlockListed
+	}
+
 	_, err := ncp.verify(c.Certificate, now, c.Fingerprint, c.signerFingerprint)
 	return err
 }

cert/ca_pool_test.go

@@ -1,10 +1,14 @@
 package cert
 
 import (
+	"bytes"
+	"io"
 	"net/netip"
+	"strings"
 	"testing"
 	"time"
 
+	"github.com/slackhq/nebula/cert/p256"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )
@@ -111,6 +115,60 @@ k+coOv04r+zh33ISyhbsafnYduN17p2eD7CmHvHuerguXD9f32gcxo/KsFCKEjMe
 	assert.Len(t, ppppp.CAs, 1)
 }
 
+// oneByteReader wraps a reader to return at most 1 byte per Read call,
+// exercising the streaming accumulation logic in NewCAPoolFromPEMReader.
+type oneByteReader struct {
+	r io.Reader
+}
+
+func (o *oneByteReader) Read(p []byte) (int, error) {
+	if len(p) == 0 {
+		return 0, nil
+	}
+	return o.r.Read(p[:1])
+}
+
+func TestNewCAPoolFromPEMReader_EmptyReader(t *testing.T) {
+	pool, err := NewCAPoolFromPEMReader(bytes.NewReader(nil))
+	require.NoError(t, err)
+	assert.Empty(t, pool.CAs)
+
+	pool, err = NewCAPoolFromPEMReader(strings.NewReader("   \n\t\n  "))
+	require.NoError(t, err)
+	assert.Empty(t, pool.CAs)
+}
+
+func TestNewCAPoolFromPEMReader_OneByteReads(t *testing.T) {
+	ca1, _, _, pem1 := NewTestCaCert(Version2, Curve_CURVE25519, time.Now(), time.Now().Add(time.Hour), nil, nil, nil)
+	ca2, _, _, pem2 := NewTestCaCert(Version2, Curve_CURVE25519, time.Now(), time.Now().Add(time.Hour), nil, nil, nil)
+
+	bundle := append(pem1, pem2...)
+	pool, err := NewCAPoolFromPEMReader(&oneByteReader{r: bytes.NewReader(bundle)})
+	require.NoError(t, err)
+	assert.Len(t, pool.CAs, 2)
+
+	fp1, err := ca1.Fingerprint()
+	require.NoError(t, err)
+	fp2, err := ca2.Fingerprint()
+	require.NoError(t, err)
+
+	assert.Contains(t, pool.CAs, fp1)
+	assert.Contains(t, pool.CAs, fp2)
+}
+
+func TestNewCAPoolFromPEMReader_TruncatedPEM(t *testing.T) {
+	_, err := NewCAPoolFromPEMReader(strings.NewReader("-----BEGIN NEBULA CERTIFICATE-----\npartialdata"))
+	assert.ErrorIs(t, err, ErrInvalidPEMBlock)
+}
+
+func TestNewCAPoolFromPEMReader_TrailingGarbage(t *testing.T) {
+	_, _, _, pem1 := NewTestCaCert(Version2, Curve_CURVE25519, time.Now(), time.Now().Add(time.Hour), nil, nil, nil)
+
+	bundle := append(pem1, []byte("some trailing garbage")...)
+	_, err := NewCAPoolFromPEMReader(bytes.NewReader(bundle))
+	assert.ErrorIs(t, err, ErrInvalidPEMBlock)
+}
+
 func TestCertificateV1_Verify(t *testing.T) {
 	ca, _, caKey, _ := NewTestCaCert(Version1, Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, nil)
 	c, _, _, _ := NewTestCert(Version1, Curve_CURVE25519, ca, caKey, "test cert", time.Now(), time.Now().Add(5*time.Minute), nil, nil, nil)
@@ -170,6 +228,15 @@ func TestCertificateV1_VerifyP256(t *testing.T) {
 	_, err = caPool.VerifyCertificate(time.Now(), c)
 	require.EqualError(t, err, "certificate is in the block list")
 
+	// Create a copy of the cert and swap to the alternate form for the signature
+	nc := c.Copy()
+	b, err := p256.Swap(c.Signature())
+	require.NoError(t, err)
+	require.NoError(t, nc.(*certificateV1).setSignature(b))
+
+	_, err = caPool.VerifyCertificate(time.Now(), nc)
+	require.EqualError(t, err, "certificate is in the block list")
+
 	caPool.ResetCertBlocklist()
 	_, err = caPool.VerifyCertificate(time.Now(), c)
 	require.NoError(t, err)
@@ -187,7 +254,7 @@ func TestCertificateV1_VerifyP256(t *testing.T) {
 	require.NoError(t, err)
 
 	caPool = NewCAPool()
-	b, err := caPool.AddCAFromPEM(caPem)
+	b, err = caPool.AddCAFromPEM(caPem)
 	require.NoError(t, err)
 	assert.Empty(t, b)
 
@@ -196,7 +263,17 @@ func TestCertificateV1_VerifyP256(t *testing.T) {
 	})
 
 	c, _, _, _ = NewTestCert(Version1, Curve_P256, ca, caKey, "test", time.Now(), time.Now().Add(5*time.Minute), nil, nil, []string{"test1"})
-	_, err = caPool.VerifyCertificate(time.Now(), c)
+	cc, err := caPool.VerifyCertificate(time.Now(), c)
+	require.NoError(t, err)
+
+	// Reset the blocklist and block the alternate form fingerprint
+	caPool.ResetCertBlocklist()
+	caPool.BlocklistFingerprint(cc.fingerprint2)
+	err = caPool.VerifyCachedCertificate(time.Now(), cc)
+	require.EqualError(t, err, "certificate is in the block list")
+
+	caPool.ResetCertBlocklist()
+	err = caPool.VerifyCachedCertificate(time.Now(), cc)
 	require.NoError(t, err)
 }
 
@@ -394,6 +471,15 @@ func TestCertificateV2_VerifyP256(t *testing.T) {
 	_, err = caPool.VerifyCertificate(time.Now(), c)
 	require.EqualError(t, err, "certificate is in the block list")
 
+	// Create a copy of the cert and swap to the alternate form for the signature
+	nc := c.Copy()
+	b, err := p256.Swap(c.Signature())
+	require.NoError(t, err)
+	require.NoError(t, nc.(*certificateV2).setSignature(b))
+
+	_, err = caPool.VerifyCertificate(time.Now(), nc)
+	require.EqualError(t, err, "certificate is in the block list")
+
 	caPool.ResetCertBlocklist()
 	_, err = caPool.VerifyCertificate(time.Now(), c)
 	require.NoError(t, err)
@@ -411,7 +497,7 @@ func TestCertificateV2_VerifyP256(t *testing.T) {
 	require.NoError(t, err)
 
 	caPool = NewCAPool()
-	b, err := caPool.AddCAFromPEM(caPem)
+	b, err = caPool.AddCAFromPEM(caPem)
 	require.NoError(t, err)
 	assert.Empty(t, b)
 
@@ -420,7 +506,17 @@ func TestCertificateV2_VerifyP256(t *testing.T) {
 	})
 
 	c, _, _, _ = NewTestCert(Version2, Curve_P256, ca, caKey, "test", time.Now(), time.Now().Add(5*time.Minute), nil, nil, []string{"test1"})
-	_, err = caPool.VerifyCertificate(time.Now(), c)
+	cc, err := caPool.VerifyCertificate(time.Now(), c)
+	require.NoError(t, err)
+
+	// Reset the blocklist and block the alternate form fingerprint
+	caPool.ResetCertBlocklist()
+	caPool.BlocklistFingerprint(cc.fingerprint2)
+	err = caPool.VerifyCachedCertificate(time.Now(), cc)
+	require.EqualError(t, err, "certificate is in the block list")
+
+	caPool.ResetCertBlocklist()
+	err = caPool.VerifyCachedCertificate(time.Now(), cc)
 	require.NoError(t, err)
 }
 

cert/cert.go

@@ -4,6 +4,8 @@ import (
 	"fmt"
 	"net/netip"
 	"time"
+
+	"github.com/slackhq/nebula/cert/p256"
 )
 
 type Version uint8
@@ -110,6 +112,9 @@ type CachedCertificate struct {
 	InvertedGroups    map[string]struct{}
 	Fingerprint       string
 	signerFingerprint string
+
+	// A place to store a 2nd fingerprint if the certificate could have one, such as with P256
+	fingerprint2 string
 }
 
 func (cc *CachedCertificate) String() string {
@@ -119,6 +124,7 @@ func (cc *CachedCertificate) String() string {
 // Recombine will attempt to unmarshal a certificate received in a handshake.
 // Handshakes save space by placing the peers public key in a different part of the packet, we have to
 // reassemble the actual certificate structure with that in mind.
+// Implementations MUST assert the public key is not in the raw certificate bytes if the passed in public key is not empty.
 func Recombine(v Version, rawCertBytes, publicKey []byte, curve Curve) (Certificate, error) {
 	if publicKey == nil {
 		return nil, ErrNoPeerStaticKey
@@ -151,3 +157,31 @@ func Recombine(v Version, rawCertBytes, publicKey []byte, curve Curve) (Certific
 
 	return c, nil
 }
+
+// CalculateAlternateFingerprint calculates a 2nd fingerprint representation for P256 certificates
+// CAPool blocklist testing through `VerifyCertificate` and `VerifyCachedCertificate` automatically performs this step.
+func CalculateAlternateFingerprint(c Certificate) (string, error) {
+	if c.Curve() != Curve_P256 {
+		return "", nil
+	}
+
+	nc := c.Copy()
+	b, err := p256.Swap(nc.Signature())
+	if err != nil {
+		return "", err
+	}
+
+	switch v := nc.(type) {
+	case *certificateV1:
+		err = v.setSignature(b)
+	case *certificateV2:
+		err = v.setSignature(b)
+	default:
+		return "", ErrUnknownVersion
+	}
+
+	if err != nil {
+		return "", err
+	}
+	return nc.Fingerprint()
+}

cert/cert_v1.go

@@ -426,7 +426,7 @@ func unmarshalCertificateV1(b []byte, publicKey []byte) (*certificateV1, error)
 			unsafeNetworks: make([]netip.Prefix, len(rc.Details.Subnets)/2),
 			notBefore:      time.Unix(rc.Details.NotBefore, 0),
 			notAfter:       time.Unix(rc.Details.NotAfter, 0),
-			publicKey:      make([]byte, len(rc.Details.PublicKey)),
+			publicKey:      nil,
 			isCA:           rc.Details.IsCA,
 			curve:          rc.Details.Curve,
 		},
@@ -437,11 +437,18 @@ func unmarshalCertificateV1(b []byte, publicKey []byte) (*certificateV1, error)
 	copy(nc.details.groups, rc.Details.Groups)
 	nc.details.issuer = hex.EncodeToString(rc.Details.Issuer)
 
+	// If a public key is passed in as an argument, the certificate pubkey must be empty
+	// and the passed-in pubkey copied into the cert.
 	if len(publicKey) > 0 {
-		nc.details.publicKey = publicKey
+		if len(rc.Details.PublicKey) != 0 {
+			return nil, ErrCertPubkeyPresent
 		}
-
+		nc.details.publicKey = make([]byte, len(publicKey))
+		copy(nc.details.publicKey, publicKey)
+	} else {
+		nc.details.publicKey = make([]byte, len(rc.Details.PublicKey))
 		copy(nc.details.publicKey, rc.Details.PublicKey)
+	}
 
 	var ip netip.Addr
 	for i, rawIp := range rc.Details.Ips {

cert/cert_v1_test.go

@@ -62,6 +62,62 @@ func TestCertificateV1_Marshal(t *testing.T) {
 	assert.Equal(t, nc.Groups(), nc2.Groups())
 }
 
+func TestCertificateV1_Unmarshal(t *testing.T) {
+	t.Parallel()
+	before := time.Now().Add(time.Second * -60).Round(time.Second)
+	after := time.Now().Add(time.Second * 60).Round(time.Second)
+	pubKey := []byte("1234567890abcedfghij1234567890ab")
+	invalidPubkey := []byte("00000000000000000000000000000000")
+
+	nc := certificateV1{
+		details: detailsV1{
+			name: "testing",
+			networks: []netip.Prefix{
+				mustParsePrefixUnmapped("10.1.1.1/24"),
+				mustParsePrefixUnmapped("10.1.1.2/16"),
+			},
+			unsafeNetworks: []netip.Prefix{
+				mustParsePrefixUnmapped("9.1.1.2/24"),
+				mustParsePrefixUnmapped("9.1.1.3/16"),
+			},
+			groups:    []string{"test-group1", "test-group2", "test-group3"},
+			notBefore: before,
+			notAfter:  after,
+			publicKey: pubKey,
+			isCA:      false,
+			issuer:    "1234567890abcedfghij1234567890ab",
+		},
+		signature: []byte("1234567890abcedfghij1234567890ab"),
+	}
+
+	// This certificate has a pubkey included
+	certWithPubkey, err := nc.Marshal()
+	require.NoError(t, err)
+
+	// This certificate is missing the pubkey section
+	certWithoutPubkey, err := nc.MarshalForHandshakes()
+	require.NoError(t, err)
+
+	// Cert has no pubkey and no pubkey passed in must fail to validate
+	isNil, err := unmarshalCertificateV1(certWithoutPubkey, nil)
+	require.Error(t, err)
+
+	// Cert has different pubkey than one passed in must fail
+	isNil, err = unmarshalCertificateV1(certWithPubkey, invalidPubkey)
+	require.Nil(t, isNil)
+	require.Error(t, err)
+
+	// Cert has pubkey and no pubkey argument works ok
+	_, err = unmarshalCertificateV1(certWithPubkey, nil)
+	require.NoError(t, err)
+
+	// Cert has no pubkey and valid, correctly signed pubkey passed in
+	nc2, err := unmarshalCertificateV1(certWithoutPubkey, pubKey)
+	require.NoError(t, err)
+
+	assert.Equal(t, pubKey, nc2.PublicKey())
+}
+
 func TestCertificateV1_PublicKeyPem(t *testing.T) {
 	t.Parallel()
 	before := time.Now().Add(time.Second * -60).Round(time.Second)
@@ -99,13 +155,19 @@ func TestCertificateV1_PublicKeyPem(t *testing.T) {
 AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
 AAAAAAAAAAAAAAAAAAAAAAA=
 -----END NEBULA P256 PUBLIC KEY-----
+`)
+
+	pubP256KeyPemCA := []byte(`-----BEGIN NEBULA ECDSA P256 PUBLIC KEY-----
+AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
+AAAAAAAAAAAAAAAAAAAAAAA=
+-----END NEBULA ECDSA P256 PUBLIC KEY-----
 `)
 	pubP256Key, _, _, err := UnmarshalPublicKeyFromPEM(pubP256KeyPem)
 	require.NoError(t, err)
 	nc.details.curve = Curve_P256
 	nc.details.publicKey = pubP256Key
 	assert.Equal(t, Curve_P256, nc.Curve())
-	assert.Equal(t, string(nc.MarshalPublicKeyPEM()), string(pubP256KeyPem))
+	assert.Equal(t, string(nc.MarshalPublicKeyPEM()), string(pubP256KeyPemCA))
 	assert.True(t, nc.IsCA())
 
 	nc.details.isCA = false

cert/cert_v2.go

@@ -592,7 +592,13 @@ func unmarshalCertificateV2(b []byte, publicKey []byte, curve Curve) (*certifica
 	// Maybe grab the public key
 	var rawPublicKey cryptobyte.String
 	if len(publicKey) > 0 {
-		rawPublicKey = publicKey
+		// If a public key is passed in, then the handshake certificate must
+		// not have a public key present
+		if input.PeekASN1Tag(TagCertPublicKey) {
+			return nil, ErrCertPubkeyPresent
+		}
+		rawPublicKey = make(cryptobyte.String, len(publicKey))
+		copy(rawPublicKey, publicKey)
 	} else if !input.ReadOptionalASN1(&rawPublicKey, nil, TagCertPublicKey) {
 		return nil, ErrBadFormat
 	}

cert/cert_v2_test.go

@@ -76,6 +76,58 @@ func TestCertificateV2_Marshal(t *testing.T) {
 	assert.Equal(t, nc.Groups(), nc2.Groups())
 }
 
+func TestCertificateV2_Unmarshal(t *testing.T) {
+	t.Parallel()
+	before := time.Now().Add(time.Second * -60).Round(time.Second)
+	after := time.Now().Add(time.Second * 60).Round(time.Second)
+	pubKey := []byte("1234567890abcedfghij1234567890ab")
+
+	nc := certificateV2{
+		details: detailsV2{
+			name: "testing",
+			networks: []netip.Prefix{
+				mustParsePrefixUnmapped("10.1.1.2/16"),
+				mustParsePrefixUnmapped("10.1.1.1/24"),
+			},
+			unsafeNetworks: []netip.Prefix{
+				mustParsePrefixUnmapped("9.1.1.3/16"),
+				mustParsePrefixUnmapped("9.1.1.2/24"),
+			},
+			groups:    []string{"test-group1", "test-group2", "test-group3"},
+			notBefore: before,
+			notAfter:  after,
+			isCA:      false,
+			issuer:    "1234567890abcdef1234567890abcdef",
+		},
+		signature: []byte("1234567890abcdef1234567890abcdef"),
+		publicKey: pubKey,
+	}
+
+	db, err := nc.details.Marshal()
+	require.NoError(t, err)
+	nc.rawDetails = db
+
+	certWithPubkey, err := nc.Marshal()
+	require.NoError(t, err)
+	//t.Log("Cert size:", len(b))
+	certWithoutPubkey, err := nc.MarshalForHandshakes()
+	require.NoError(t, err)
+
+	// Cert must not have a pubkey if one is passed in as an argument
+	_, err = unmarshalCertificateV2(certWithPubkey, pubKey, Curve_CURVE25519)
+	require.ErrorIs(t, err, ErrCertPubkeyPresent)
+
+	// Certs must have pubkeys
+	_, err = unmarshalCertificateV2(certWithoutPubkey, nil, Curve_CURVE25519)
+	require.ErrorIs(t, err, ErrBadFormat)
+
+	// Ensure proper unmarshal if a pubkey is passed in
+	nc2, err := unmarshalCertificateV2(certWithoutPubkey, pubKey, Curve_CURVE25519)
+	require.NoError(t, err)
+
+	assert.Equal(t, nc.PublicKey(), nc2.PublicKey())
+}
+
 func TestCertificateV2_PublicKeyPem(t *testing.T) {
 	t.Parallel()
 	before := time.Now().Add(time.Second * -60).Round(time.Second)
@@ -114,12 +166,19 @@ AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
 AAAAAAAAAAAAAAAAAAAAAAA=
 -----END NEBULA P256 PUBLIC KEY-----
 `)
+
+	pubP256KeyPemCA := []byte(`-----BEGIN NEBULA ECDSA P256 PUBLIC KEY-----
+AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
+AAAAAAAAAAAAAAAAAAAAAAA=
+-----END NEBULA ECDSA P256 PUBLIC KEY-----
+`)
+
 	pubP256Key, _, _, err := UnmarshalPublicKeyFromPEM(pubP256KeyPem)
 	require.NoError(t, err)
 	nc.curve = Curve_P256
 	nc.publicKey = pubP256Key
 	assert.Equal(t, Curve_P256, nc.Curve())
-	assert.Equal(t, string(nc.MarshalPublicKeyPEM()), string(pubP256KeyPem))
+	assert.Equal(t, string(nc.MarshalPublicKeyPEM()), string(pubP256KeyPemCA))
 	assert.True(t, nc.IsCA())
 
 	nc.details.isCA = false

cert/crypto_test.go

@@ -79,7 +79,7 @@ qrlJ69wer3ZUHFXA
 	assert.Nil(t, k)
 	assert.Equal(t, rest, invalidPem)
 
-	// Fail due to ivalid PEM format, because
+	// Fail due to invalid PEM format, because
 	// it's missing the requisite pre-encapsulation boundary.
 	curve, k, rest, err = DecryptAndUnmarshalSigningPrivateKey(passphrase, rest)
 	require.EqualError(t, err, "input did not contain a valid PEM encoded block")

cert/errors.go

@@ -21,6 +21,7 @@ var (
 	ErrPrivateKeyEncrypted        = errors.New("private key must be decrypted")
 	ErrCaNotFound                 = errors.New("could not find ca for the certificate")
 	ErrUnknownVersion             = errors.New("certificate version unrecognized")
+	ErrCertPubkeyPresent          = errors.New("certificate has unexpected pubkey present")
 
 	ErrInvalidPEMBlock                   = errors.New("input did not contain a valid PEM encoded block")
 	ErrInvalidPEMCertificateBanner       = errors.New("bytes did not contain a proper certificate banner")

cert/p256/p256.go

@@ -0,0 +1,127 @@
+package p256
+
+import (
+	"crypto/elliptic"
+	"errors"
+	"math/big"
+
+	"filippo.io/bigmod"
+
+	"golang.org/x/crypto/cryptobyte"
+	"golang.org/x/crypto/cryptobyte/asn1"
+)
+
+var halfN = new(big.Int).Rsh(elliptic.P256().Params().N, 1)
+var nMod *bigmod.Modulus
+
+func init() {
+	n, err := bigmod.NewModulus(elliptic.P256().Params().N.Bytes())
+	if err != nil {
+		panic(err)
+	}
+	nMod = n
+}
+
+func IsNormalized(sig []byte) (bool, error) {
+	r, s, err := parseSignature(sig)
+	if err != nil {
+		return false, err
+	}
+	return checkLowS(r, s), nil
+}
+
+func checkLowS(_, s []byte) bool {
+	bigS := new(big.Int).SetBytes(s)
+	// Check if S <= (N/2), because we want to include the midpoint in the set of low-s
+	return bigS.Cmp(halfN) <= 0
+}
+
+func swap(r, s []byte) ([]byte, []byte, error) {
+	var err error
+	bigS, err := bigmod.NewNat().SetBytes(s, nMod)
+	if err != nil {
+		return nil, nil, err
+	}
+	sNormalized := nMod.Nat().Sub(bigS, nMod)
+
+	result := sNormalized.Bytes(nMod)
+	for len(result) > 1 && result[0] == 0 {
+		result = result[1:]
+	}
+
+	return r, result, nil
+}
+
+func Normalize(sig []byte) ([]byte, error) {
+	r, s, err := parseSignature(sig)
+	if err != nil {
+		return nil, err
+	}
+
+	if checkLowS(r, s) {
+		return sig, nil
+	}
+
+	newR, newS, err := swap(r, s)
+	if err != nil {
+		return nil, err
+	}
+
+	return encodeSignature(newR, newS)
+}
+
+// Swap will change sig between its current form to the opposite high or low form.
+func Swap(sig []byte) ([]byte, error) {
+	r, s, err := parseSignature(sig)
+	if err != nil {
+		return nil, err
+	}
+
+	newR, newS, err := swap(r, s)
+	if err != nil {
+		return nil, err
+	}
+
+	return encodeSignature(newR, newS)
+}
+
+// parseSignature taken exactly from crypto/ecdsa/ecdsa.go
+func parseSignature(sig []byte) (r, s []byte, err error) {
+	var inner cryptobyte.String
+	input := cryptobyte.String(sig)
+	if !input.ReadASN1(&inner, asn1.SEQUENCE) ||
+		!input.Empty() ||
+		!inner.ReadASN1Integer(&r) ||
+		!inner.ReadASN1Integer(&s) ||
+		!inner.Empty() {
+		return nil, nil, errors.New("invalid ASN.1")
+	}
+	return r, s, nil
+}
+
+func encodeSignature(r, s []byte) ([]byte, error) {
+	var b cryptobyte.Builder
+	b.AddASN1(asn1.SEQUENCE, func(b *cryptobyte.Builder) {
+		addASN1IntBytes(b, r)
+		addASN1IntBytes(b, s)
+	})
+	return b.Bytes()
+}
+
+// addASN1IntBytes encodes in ASN.1 a positive integer represented as
+// a big-endian byte slice with zero or more leading zeroes.
+func addASN1IntBytes(b *cryptobyte.Builder, bytes []byte) {
+	for len(bytes) > 0 && bytes[0] == 0 {
+		bytes = bytes[1:]
+	}
+	if len(bytes) == 0 {
+		b.SetError(errors.New("invalid integer"))
+		return
+	}
+	b.AddASN1(asn1.INTEGER, func(c *cryptobyte.Builder) {
+		if bytes[0]&0x80 != 0 {
+			c.AddUint8(0)
+		}
+		c.AddBytes(bytes)
+	})
+}

cert/p256/p256_test.go

@@ -0,0 +1,28 @@
+package p256
+
+import (
+	"crypto/ecdsa"
+	"crypto/elliptic"
+	"crypto/rand"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+func TestFlipping(t *testing.T) {
+	priv, err1 := ecdsa.GenerateKey(elliptic.P256(), rand.Reader)
+	require.NoError(t, err1)
+
+	out, err := ecdsa.SignASN1(rand.Reader, priv, []byte("big chungus"))
+	require.NoError(t, err)
+
+	r, s, err := parseSignature(out)
+	require.NoError(t, err)
+
+	r, s1, err := swap(r, s)
+	require.NoError(t, err)
+	r, s2, err := swap(r, s1)
+	require.NoError(t, err)
+	require.Equal(t, s, s2)
+	require.NotEqual(t, s, s1)
+}

cert/pem.go

@@ -1,14 +1,66 @@
 package cert
 
 import (
-	"encoding/hex"
+	"bytes"
 	"encoding/pem"
+	"errors"
 	"fmt"
-	"time"
 
 	"golang.org/x/crypto/ed25519"
 )
 
+var ErrTruncatedPEMBlock = errors.New("truncated PEM block")
+
+// SplitPEM is a split function for bufio.Scanner that returns each PEM block.
+func SplitPEM(data []byte, atEOF bool) (advance int, token []byte, err error) {
+	// Look for the start of a PEM block
+	start := bytes.Index(data, []byte("-----BEGIN "))
+	if start == -1 {
+		if atEOF && len(bytes.TrimSpace(data)) > 0 {
+			// Non-whitespace content with no PEM block
+			return 0, nil, ErrTruncatedPEMBlock
+		}
+		if atEOF {
+			return len(data), nil, nil
+		}
+		// Request more data
+		return 0, nil, nil
+	}
+
+	// Look for the end marker
+	endMarkerStart := bytes.Index(data[start:], []byte("-----END "))
+	if endMarkerStart == -1 {
+		if atEOF {
+			// Incomplete PEM block at EOF
+			return 0, nil, ErrTruncatedPEMBlock
+		}
+		// Need more data to find the end
+		return 0, nil, nil
+	}
+
+	// Find the actual end of the END line (after the newline)
+	endMarkerStart += start
+	endLineEnd := bytes.IndexByte(data[endMarkerStart:], '\n')
+	var end int
+	if endLineEnd == -1 {
+		if atEOF {
+			// END marker without newline at EOF - take it anyway
+			end = len(data)
+		} else {
+			// Need more data
+			return 0, nil, nil
+		}
+	} else {
+		end = endMarkerStart + endLineEnd + 1
+	}
+
+	// Extract the PEM block
+	pemBlock := data[start:end]
+
+	// Return the valid PEM block
+	return end, pemBlock, nil
+}
+
 const ( //cert banners
 	CertificateBanner   = "NEBULA CERTIFICATE"
 	CertificateV2Banner = "NEBULA CERTIFICATE V2"
@@ -39,19 +91,7 @@ func UnmarshalCertificateFromPEM(b []byte) (Certificate, []byte, error) {
 		return nil, r, ErrInvalidPEMBlock
 	}
 
-	var c Certificate
+	c, err := unmarshalCertificateBlock(p)
-	var err error
-
-	switch p.Type {
-	// Implementations must validate the resulting certificate contains valid information
-	case CertificateBanner:
-		c, err = unmarshalCertificateV1(p.Bytes, nil)
-	case CertificateV2Banner:
-		c, err = unmarshalCertificateV2(p.Bytes, nil, Curve_CURVE25519)
-	default:
-		return nil, r, ErrInvalidPEMCertificateBanner
-	}
-
 	if err != nil {
 		return nil, r, err
 	}
@@ -60,6 +100,20 @@ func UnmarshalCertificateFromPEM(b []byte) (Certificate, []byte, error) {
 
 }
 
+// unmarshalCertificateBlock decodes a single PEM block into a certificate.
+// It expects a Nebula certificate banner and returns ErrInvalidPEMCertificateBanner otherwise.
+func unmarshalCertificateBlock(block *pem.Block) (Certificate, error) {
+	switch block.Type {
+	// Implementations must validate the resulting certificate contains valid information
+	case CertificateBanner:
+		return unmarshalCertificateV1(block.Bytes, nil)
+	case CertificateV2Banner:
+		return unmarshalCertificateV2(block.Bytes, nil, Curve_CURVE25519)
+	default:
+		return nil, ErrInvalidPEMCertificateBanner
+	}
+}
+
 func marshalCertPublicKeyToPEM(c Certificate) []byte {
 	if c.IsCA() {
 		return MarshalSigningPublicKeyToPEM(c.Curve(), c.PublicKey())
@@ -88,7 +142,7 @@ func MarshalSigningPublicKeyToPEM(curve Curve, b []byte) []byte {
 	case Curve_CURVE25519:
 		return pem.EncodeToMemory(&pem.Block{Type: Ed25519PublicKeyBanner, Bytes: b})
 	case Curve_P256:
-		return pem.EncodeToMemory(&pem.Block{Type: P256PublicKeyBanner, Bytes: b})
+		return pem.EncodeToMemory(&pem.Block{Type: ECDSAP256PublicKeyBanner, Bytes: b})
 	default:
 		return nil
 	}
@@ -140,101 +194,6 @@ func MarshalSigningPrivateKeyToPEM(curve Curve, b []byte) []byte {
 	}
 }
 
-// Backward compatibility functions for older API
-func MarshalX25519PublicKey(b []byte) []byte {
-	return MarshalPublicKeyToPEM(Curve_CURVE25519, b)
-}
-
-func MarshalX25519PrivateKey(b []byte) []byte {
-	return MarshalPrivateKeyToPEM(Curve_CURVE25519, b)
-}
-
-func MarshalPublicKey(curve Curve, b []byte) []byte {
-	return MarshalPublicKeyToPEM(curve, b)
-}
-
-func MarshalPrivateKey(curve Curve, b []byte) []byte {
-	return MarshalPrivateKeyToPEM(curve, b)
-}
-
-// NebulaCertificate is a compatibility wrapper for the old API
-type NebulaCertificate struct {
-	Details   NebulaCertificateDetails
-	Signature []byte
-	cert      Certificate
-}
-
-// NebulaCertificateDetails is a compatibility wrapper for certificate details
-type NebulaCertificateDetails struct {
-	Name      string
-	NotBefore time.Time
-	NotAfter  time.Time
-	PublicKey []byte
-	IsCA      bool
-	Issuer    []byte
-	Curve     Curve
-}
-
-// UnmarshalNebulaCertificateFromPEM provides backward compatibility with the old API
-func UnmarshalNebulaCertificateFromPEM(b []byte) (*NebulaCertificate, []byte, error) {
-	c, rest, err := UnmarshalCertificateFromPEM(b)
-	if err != nil {
-		return nil, rest, err
-	}
-
-	issuerBytes, err := func() ([]byte, error) {
-		issuer := c.Issuer()
-		if issuer == "" {
-			return nil, nil
-		}
-		decoded, err := hex.DecodeString(issuer)
-		if err != nil {
-			return nil, fmt.Errorf("failed to decode issuer fingerprint: %w", err)
-		}
-		return decoded, nil
-	}()
-	if err != nil {
-		return nil, rest, err
-	}
-
-	pubKey := c.PublicKey()
-	if pubKey != nil {
-		pubKey = append([]byte(nil), pubKey...)
-	}
-
-	sig := c.Signature()
-	if sig != nil {
-		sig = append([]byte(nil), sig...)
-	}
-
-	return &NebulaCertificate{
-		Details: NebulaCertificateDetails{
-			Name:      c.Name(),
-			NotBefore: c.NotBefore(),
-			NotAfter:  c.NotAfter(),
-			PublicKey: pubKey,
-			IsCA:      c.IsCA(),
-			Issuer:    issuerBytes,
-			Curve:     c.Curve(),
-		},
-		Signature: sig,
-		cert:      c,
-	}, rest, nil
-}
-
-// IssuerString returns the issuer in hex format for compatibility
-func (n *NebulaCertificate) IssuerString() string {
-	if n.Details.Issuer == nil {
-		return ""
-	}
-	return hex.EncodeToString(n.Details.Issuer)
-}
-
-// Certificate returns the underlying certificate (read-only)
-func (n *NebulaCertificate) Certificate() Certificate {
-	return n.cert
-}
-
 // UnmarshalPrivateKeyFromPEM will try to unmarshal the first pem block in a byte array, returning any non
 // consumed data or an error on failure
 func UnmarshalPrivateKeyFromPEM(b []byte) ([]byte, []byte, Curve, error) {

cert/pem_test.go

@@ -1,12 +1,88 @@
 package cert
 
 import (
+	"bufio"
+	"strings"
 	"testing"
 
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )
 
+func scanAll(t *testing.T, input string) ([]string, error) {
+	t.Helper()
+	scanner := bufio.NewScanner(strings.NewReader(input))
+	scanner.Split(SplitPEM)
+	var blocks []string
+	for scanner.Scan() {
+		blocks = append(blocks, scanner.Text())
+	}
+	return blocks, scanner.Err()
+}
+
+func TestSplitPEM_Single(t *testing.T) {
+	input := "-----BEGIN TEST-----\ndata\n-----END TEST-----\n"
+	blocks, err := scanAll(t, input)
+	require.NoError(t, err)
+	require.Len(t, blocks, 1)
+	require.Equal(t, input, blocks[0])
+}
+
+func TestSplitPEM_Multiple(t *testing.T) {
+	block1 := "-----BEGIN TEST-----\naaa\n-----END TEST-----\n"
+	block2 := "-----BEGIN TEST-----\nbbb\n-----END TEST-----\n"
+	blocks, err := scanAll(t, block1+block2)
+	require.NoError(t, err)
+	require.Len(t, blocks, 2)
+	require.Equal(t, block1, blocks[0])
+	require.Equal(t, block2, blocks[1])
+}
+
+func TestSplitPEM_CommentsAndWhitespaceBetweenBlocks(t *testing.T) {
+	input := "# comment\n\n-----BEGIN TEST-----\naaa\n-----END TEST-----\n\n# another comment\n\n-----BEGIN TEST-----\nbbb\n-----END TEST-----\n"
+	blocks, err := scanAll(t, input)
+	require.NoError(t, err)
+	require.Len(t, blocks, 2)
+}
+
+func TestSplitPEM_Empty(t *testing.T) {
+	blocks, err := scanAll(t, "")
+	require.NoError(t, err)
+	require.Empty(t, blocks)
+}
+
+func TestSplitPEM_WhitespaceOnly(t *testing.T) {
+	blocks, err := scanAll(t, "  \n\t\n  ")
+	require.NoError(t, err)
+	require.Empty(t, blocks)
+}
+
+func TestSplitPEM_TrailingGarbage(t *testing.T) {
+	input := "-----BEGIN TEST-----\ndata\n-----END TEST-----\ngarbage"
+	blocks, err := scanAll(t, input)
+	require.ErrorIs(t, err, ErrTruncatedPEMBlock)
+	require.Len(t, blocks, 1)
+}
+
+func TestSplitPEM_TruncatedBlock(t *testing.T) {
+	input := "-----BEGIN TEST-----\npartial data with no end"
+	_, err := scanAll(t, input)
+	require.ErrorIs(t, err, ErrTruncatedPEMBlock)
+}
+
+func TestSplitPEM_NoEndNewline(t *testing.T) {
+	input := "-----BEGIN TEST-----\ndata\n-----END TEST-----"
+	blocks, err := scanAll(t, input)
+	require.NoError(t, err)
+	require.Len(t, blocks, 1)
+	require.Equal(t, input, blocks[0])
+}
+
+func TestSplitPEM_GarbageOnly(t *testing.T) {
+	_, err := scanAll(t, "this is not PEM data")
+	require.ErrorIs(t, err, ErrTruncatedPEMBlock)
+}
+
 func TestUnmarshalCertificateFromPEM(t *testing.T) {
 	goodCert := []byte(`
 # A good cert
@@ -44,7 +120,7 @@ bzBEr00kERQxxTzTsH8cpYEgRoipvmExvg8WP8NdAJEYJosB
 	assert.Equal(t, rest, invalidPem)
 	require.EqualError(t, err, "bytes did not contain a proper certificate banner")
 
-	// Fail due to ivalid PEM format, because
+	// Fail due to invalid PEM format, because
 	// it's missing the requisite pre-encapsulation boundary.
 	cert, rest, err = UnmarshalCertificateFromPEM(rest)
 	assert.Nil(t, cert)
@@ -106,7 +182,7 @@ AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
 	assert.Equal(t, rest, invalidPem)
 	require.EqualError(t, err, "bytes did not contain a proper Ed25519/ECDSA private key banner")
 
-	// Fail due to ivalid PEM format, because
+	// Fail due to invalid PEM format, because
 	// it's missing the requisite pre-encapsulation boundary.
 	k, rest, curve, err = UnmarshalSigningPrivateKeyFromPEM(rest)
 	assert.Nil(t, k)
@@ -168,7 +244,7 @@ AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=
 	assert.Equal(t, rest, invalidPem)
 	require.EqualError(t, err, "bytes did not contain a proper private key banner")
 
-	// Fail due to ivalid PEM format, because
+	// Fail due to invalid PEM format, because
 	// it's missing the requisite pre-encapsulation boundary.
 	k, rest, curve, err = UnmarshalPrivateKeyFromPEM(rest)
 	assert.Nil(t, k)
@@ -221,7 +297,7 @@ AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=
 	require.EqualError(t, err, "bytes did not contain a proper public key banner")
 	assert.Equal(t, rest, invalidPem)
 
-	// Fail due to ivalid PEM format, because
+	// Fail due to invalid PEM format, because
 	// it's missing the requisite pre-encapsulation boundary.
 	k, rest, curve, err = UnmarshalPublicKeyFromPEM(rest)
 	assert.Nil(t, k)
@@ -299,7 +375,7 @@ AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=
 	require.EqualError(t, err, "bytes did not contain a proper public key banner")
 	assert.Equal(t, rest, invalidPem)
 
-	// Fail due to ivalid PEM format, because
+	// Fail due to invalid PEM format, because
 	// it's missing the requisite pre-encapsulation boundary.
 	k, rest, curve, err = UnmarshalPublicKeyFromPEM(rest)
 	assert.Nil(t, k)

cert/sign.go

@@ -9,6 +9,8 @@ import (
 	"fmt"
 	"net/netip"
 	"time"
+
+	"github.com/slackhq/nebula/cert/p256"
 )
 
 // TBSCertificate represents a certificate intended to be signed.
@@ -126,6 +128,13 @@ func (t *TBSCertificate) SignWith(signer Certificate, curve Curve, sp SignerLamb
 		return nil, err
 	}
 
+	if curve == Curve_P256 {
+		sig, err = p256.Normalize(sig)
+		if err != nil {
+			return nil, err
+		}
+	}
+
 	err = c.setSignature(sig)
 	if err != nil {
 		return nil, err

cert/sign_test.go

@@ -9,6 +9,7 @@ import (
 	"testing"
 	"time"
 
+	"github.com/slackhq/nebula/cert/p256"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )
@@ -89,3 +90,48 @@ func TestCertificateV1_SignP256(t *testing.T) {
 	require.NoError(t, err)
 	assert.NotNil(t, uc)
 }
+
+func TestCertificate_SignP256_AlwaysNormalized(t *testing.T) {
+	before := time.Now().Add(time.Second * -60).Round(time.Second)
+	after := time.Now().Add(time.Second * 60).Round(time.Second)
+	pubKey := []byte("01234567890abcedfghij1234567890ab1234567890abcedfghij1234567890ab")
+
+	tbs := TBSCertificate{
+		Version: Version1,
+		Name:    "testing",
+		Networks: []netip.Prefix{
+			mustParsePrefixUnmapped("10.1.1.1/24"),
+			mustParsePrefixUnmapped("10.1.1.2/16"),
+		},
+		UnsafeNetworks: []netip.Prefix{
+			mustParsePrefixUnmapped("9.1.1.2/24"),
+			mustParsePrefixUnmapped("9.1.1.3/16"),
+		},
+		Groups:    []string{"test-group1", "test-group2", "test-group3"},
+		NotBefore: before,
+		NotAfter:  after,
+		PublicKey: pubKey,
+		IsCA:      true,
+		Curve:     Curve_P256,
+	}
+
+	priv, err := ecdsa.GenerateKey(elliptic.P256(), rand.Reader)
+	require.NoError(t, err)
+	pub := elliptic.Marshal(elliptic.P256(), priv.PublicKey.X, priv.PublicKey.Y)
+	rawPriv := priv.D.FillBytes(make([]byte, 32))
+
+	for i := 0; i < 1000; i++ {
+		if i&1 == 1 {
+			tbs.Version = Version1
+		} else {
+			tbs.Version = Version2
+		}
+		c, err := tbs.Sign(nil, Curve_P256, rawPriv)
+		require.NoError(t, err)
+		assert.NotNil(t, c)
+		assert.True(t, c.CheckSignature(pub))
+		normie, err := p256.IsNormalized(c.Signature())
+		require.NoError(t, err)
+		assert.True(t, normie)
+	}
+}

cert_test/cert.go

@@ -163,3 +163,55 @@ func P256Keypair() ([]byte, []byte) {
 	pubkey := privkey.PublicKey()
 	return pubkey.Bytes(), privkey.Bytes()
 }
+
+// DummyCert is a minimal cert.Certificate implementation for testing error paths.
+type DummyCert struct {
+	Version_        cert.Version
+	Curve_          cert.Curve
+	Groups_         []string
+	IsCA_           bool
+	Issuer_         string
+	Name_           string
+	Networks_       []netip.Prefix
+	NotAfter_       time.Time
+	NotBefore_      time.Time
+	PublicKey_      []byte
+	Signature_      []byte
+	UnsafeNetworks_ []netip.Prefix
+}
+
+func (d *DummyCert) Version() cert.Version                         { return d.Version_ }
+func (d *DummyCert) Curve() cert.Curve                             { return d.Curve_ }
+func (d *DummyCert) Groups() []string                              { return d.Groups_ }
+func (d *DummyCert) IsCA() bool                                    { return d.IsCA_ }
+func (d *DummyCert) Issuer() string                                { return d.Issuer_ }
+func (d *DummyCert) Name() string                                  { return d.Name_ }
+func (d *DummyCert) Networks() []netip.Prefix                      { return d.Networks_ }
+func (d *DummyCert) NotAfter() time.Time                           { return d.NotAfter_ }
+func (d *DummyCert) NotBefore() time.Time                          { return d.NotBefore_ }
+func (d *DummyCert) PublicKey() []byte                             { return d.PublicKey_ }
+func (d *DummyCert) Signature() []byte                             { return d.Signature_ }
+func (d *DummyCert) UnsafeNetworks() []netip.Prefix                { return d.UnsafeNetworks_ }
+func (d *DummyCert) Fingerprint() (string, error)                  { return "", nil }
+func (d *DummyCert) CheckSignature(key []byte) bool                { return false }
+func (d *DummyCert) MarshalForHandshakes() ([]byte, error)         { return nil, nil }
+func (d *DummyCert) MarshalPEM() ([]byte, error)                   { return nil, nil }
+func (d *DummyCert) MarshalJSON() ([]byte, error)                  { return nil, nil }
+func (d *DummyCert) Marshal() ([]byte, error)                      { return nil, nil }
+func (d *DummyCert) String() string                                { return "dummy" }
+func (d *DummyCert) Copy() cert.Certificate                        { return d }
+func (d *DummyCert) VerifyPrivateKey(c cert.Curve, k []byte) error { return nil }
+func (d *DummyCert) Expired(time.Time) bool                        { return false }
+func (d *DummyCert) MarshalPublicKeyPEM() []byte                   { return nil }
+func (d *DummyCert) PublicKeyPEM() []byte                          { return nil }
+
+// NewTestCAPool creates a CAPool from the given CA certificates, panicking on error.
+func NewTestCAPool(cas ...cert.Certificate) *cert.CAPool {
+	pool := cert.NewCAPool()
+	for _, ca := range cas {
+		if err := pool.AddCA(ca); err != nil {
+			panic(err)
+		}
+	}
+	return pool
+}

cmd/nebula-cert/ca.go

@@ -173,6 +173,8 @@ func ca(args []string, out io.Writer, errOut io.Writer, pr PasswordReader) error
 
 	var passphrase []byte
 	if !isP11 && *cf.encryption {
+		passphrase = []byte(os.Getenv("NEBULA_CA_PASSPHRASE"))
+		if len(passphrase) == 0 {
 			for i := 0; i < 5; i++ {
 				out.Write([]byte("Enter passphrase: "))
 				passphrase, err = pr.ReadPassword()
@@ -192,6 +194,7 @@ func ca(args []string, out io.Writer, errOut io.Writer, pr PasswordReader) error
 				return fmt.Errorf("no passphrase specified, remove -encrypt flag to write out-key in plaintext")
 			}
 		}
+	}
 
 	var curve cert.Curve
 	var pub, rawPriv []byte

cmd/nebula-cert/ca_test.go

@@ -171,6 +171,17 @@ func Test_ca(t *testing.T) {
 	assert.Equal(t, pwPromptOb, ob.String())
 	assert.Empty(t, eb.String())
 
+	// test encrypted key with passphrase environment variable
+	os.Remove(keyF.Name())
+	os.Remove(crtF.Name())
+	ob.Reset()
+	eb.Reset()
+	args = []string{"-version", "1", "-encrypt", "-name", "test", "-duration", "100m", "-groups", "1,2,3,4,5", "-out-crt", crtF.Name(), "-out-key", keyF.Name()}
+	os.Setenv("NEBULA_CA_PASSPHRASE", string(passphrase))
+	require.NoError(t, ca(args, ob, eb, testpw))
+	assert.Empty(t, eb.String())
+	os.Setenv("NEBULA_CA_PASSPHRASE", "")
+
 	// read encrypted key file and verify default params
 	rb, _ = os.ReadFile(keyF.Name())
 	k, _ := pem.Decode(rb)
@@ -189,7 +200,7 @@ func Test_ca(t *testing.T) {
 	assert.Empty(t, b)
 	assert.Len(t, lKey, 64)
 
-	// test when reading passsword results in an error
+	// test when reading password results in an error
 	os.Remove(keyF.Name())
 	os.Remove(crtF.Name())
 	ob.Reset()

cmd/nebula-cert/main.go

@@ -5,10 +5,28 @@ import (
 	"fmt"
 	"io"
 	"os"
+	"runtime/debug"
+	"strings"
 )
 
+// A version string that can be set with
+//
+//	-ldflags "-X main.Build=SOMEVERSION"
+//
+// at compile-time.
 var Build string
 
+func init() {
+	if Build == "" {
+		info, ok := debug.ReadBuildInfo()
+		if !ok {
+			return
+		}
+
+		Build = strings.TrimPrefix(info.Main.Version, "v")
+	}
+}
+
 type helpError struct {
 	s string
 }

cmd/nebula-cert/sign.go

@@ -43,7 +43,7 @@ type signFlags struct {
 func newSignFlags() *signFlags {
 	sf := signFlags{set: flag.NewFlagSet("sign", flag.ContinueOnError)}
 	sf.set.Usage = func() {}
-	sf.version = sf.set.Uint("version", 0, "Optional: version of the certificate format to use, the default is to create both v1 and v2 certificates.")
+	sf.version = sf.set.Uint("version", 0, "Optional: version of the certificate format to use. The default is to match the version of the signing CA")
 	sf.caKeyPath = sf.set.String("ca-key", "ca.key", "Optional: path to the signing CA key")
 	sf.caCertPath = sf.set.String("ca-crt", "ca.crt", "Optional: path to the signing CA cert")
 	sf.name = sf.set.String("name", "", "Required: name of the cert, usually a hostname")
@@ -116,8 +116,10 @@ func signCert(args []string, out io.Writer, errOut io.Writer, pr PasswordReader)
 		// naively attempt to decode the private key as though it is not encrypted
 		caKey, _, curve, err = cert.UnmarshalSigningPrivateKeyFromPEM(rawCAKey)
 		if errors.Is(err, cert.ErrPrivateKeyEncrypted) {
-			// ask for a passphrase until we get one
 			var passphrase []byte
+			passphrase = []byte(os.Getenv("NEBULA_CA_PASSPHRASE"))
+			if len(passphrase) == 0 {
+				// ask for a passphrase until we get one
 				for i := 0; i < 5; i++ {
 					out.Write([]byte("Enter passphrase: "))
 					passphrase, err = pr.ReadPassword()
@@ -135,7 +137,7 @@ func signCert(args []string, out io.Writer, errOut io.Writer, pr PasswordReader)
 				if len(passphrase) == 0 {
 					return fmt.Errorf("cannot open encrypted ca-key without passphrase")
 				}
-
+			}
 			curve, caKey, _, err = cert.DecryptAndUnmarshalSigningPrivateKey(passphrase, rawCAKey)
 			if err != nil {
 				return fmt.Errorf("error while parsing encrypted ca-key: %s", err)
@@ -165,6 +167,10 @@ func signCert(args []string, out io.Writer, errOut io.Writer, pr PasswordReader)
 		return fmt.Errorf("ca certificate is expired")
 	}
 
+	if version == 0 {
+		version = caCert.Version()
+	}
+
 	// if no duration is given, expire one second before the root expires
 	if *sf.duration <= 0 {
 		*sf.duration = time.Until(caCert.NotAfter()) - time.Second*1
@@ -277,21 +283,19 @@ func signCert(args []string, out io.Writer, errOut io.Writer, pr PasswordReader)
 	notBefore := time.Now()
 	notAfter := notBefore.Add(*sf.duration)
 
-	if version == 0 || version == cert.Version1 {
+	switch version {
-		// Make sure we at least have an ip
+	case cert.Version1:
+		// Make sure we have only one ipv4 address
 		if len(v4Networks) != 1 {
 			return newHelpErrorf("invalid -networks definition: v1 certificates can only have a single ipv4 address")
 		}
 
-		if version == cert.Version1 {
-			// If we are asked to mint a v1 certificate only then we cant just ignore any v6 addresses
 		if len(v6Networks) > 0 {
-				return newHelpErrorf("invalid -networks definition: v1 certificates can only be ipv4")
+			return newHelpErrorf("invalid -networks definition: v1 certificates can only contain ipv4 addresses")
 		}
 
 		if len(v6UnsafeNetworks) > 0 {
-				return newHelpErrorf("invalid -unsafe-networks definition: v1 certificates can only be ipv4")
+			return newHelpErrorf("invalid -unsafe-networks definition: v1 certificates can only contain ipv4 addresses")
-			}
 		}
 
 		t := &cert.TBSCertificate{
@@ -321,9 +325,8 @@ func signCert(args []string, out io.Writer, errOut io.Writer, pr PasswordReader)
 		}
 
 		crts = append(crts, nc)
-	}
 
-	if version == 0 || version == cert.Version2 {
+	case cert.Version2:
 		t := &cert.TBSCertificate{
 			Version:        cert.Version2,
 			Name:           *sf.name,
@@ -351,6 +354,9 @@ func signCert(args []string, out io.Writer, errOut io.Writer, pr PasswordReader)
 		}
 
 		crts = append(crts, nc)
+	default:
+		// this should be unreachable
+		return fmt.Errorf("invalid version: %d", version)
 	}
 
 	if !isP11 && *sf.inPubPath == "" {

cmd/nebula-cert/sign_test.go

@@ -55,7 +55,7 @@ func Test_signHelp(t *testing.T) {
 			"  -unsafe-networks string\n"+
 			"    \tOptional: comma separated list of ip address and network in CIDR notation. Unsafe networks this cert can route for\n"+
 			"  -version uint\n"+
-			"    \tOptional: version of the certificate format to use, the default is to create both v1 and v2 certificates.\n",
+			"    \tOptional: version of the certificate format to use. The default is to match the version of the signing CA\n",
 		ob.String(),
 	)
 }
@@ -204,7 +204,7 @@ func Test_signCert(t *testing.T) {
 	ob.Reset()
 	eb.Reset()
 	args = []string{"-version", "1", "-ca-crt", caCrtF.Name(), "-ca-key", caKeyF.Name(), "-name", "test", "-ip", "1.1.1.1/24", "-out-crt", "nope", "-out-key", "nope", "-duration", "100m", "-subnets", "100::100/100"}
-	assertHelpError(t, signCert(args, ob, eb, nopw), "invalid -unsafe-networks definition: v1 certificates can only be ipv4")
+	assertHelpError(t, signCert(args, ob, eb, nopw), "invalid -unsafe-networks definition: v1 certificates can only contain ipv4 addresses")
 	assert.Empty(t, ob.String())
 	assert.Empty(t, eb.String())
 
@@ -379,6 +379,15 @@ func Test_signCert(t *testing.T) {
 	assert.Equal(t, "Enter passphrase: ", ob.String())
 	assert.Empty(t, eb.String())
 
+	// test with the proper password in the environment
+	os.Remove(crtF.Name())
+	os.Remove(keyF.Name())
+	args = []string{"-version", "1", "-ca-crt", caCrtF.Name(), "-ca-key", caKeyF.Name(), "-name", "test", "-ip", "1.1.1.1/24", "-out-crt", crtF.Name(), "-out-key", keyF.Name(), "-duration", "100m", "-subnets", "10.1.1.1/32, ,   10.2.2.2/32   ,   ,  ,, 10.5.5.5/32", "-groups", "1,,   2    ,        ,,,3,4,5"}
+	os.Setenv("NEBULA_CA_PASSPHRASE", string(passphrase))
+	require.NoError(t, signCert(args, ob, eb, testpw))
+	assert.Empty(t, eb.String())
+	os.Setenv("NEBULA_CA_PASSPHRASE", "")
+
 	// test with the wrong password
 	ob.Reset()
 	eb.Reset()
@@ -389,6 +398,17 @@ func Test_signCert(t *testing.T) {
 	assert.Equal(t, "Enter passphrase: ", ob.String())
 	assert.Empty(t, eb.String())
 
+	// test with the wrong password in environment
+	ob.Reset()
+	eb.Reset()
+
+	os.Setenv("NEBULA_CA_PASSPHRASE", "invalid password")
+	args = []string{"-version", "1", "-ca-crt", caCrtF.Name(), "-ca-key", caKeyF.Name(), "-name", "test", "-ip", "1.1.1.1/24", "-out-crt", crtF.Name(), "-out-key", keyF.Name(), "-duration", "100m", "-subnets", "10.1.1.1/32, ,   10.2.2.2/32   ,   ,  ,, 10.5.5.5/32", "-groups", "1,,   2    ,        ,,,3,4,5"}
+	require.EqualError(t, signCert(args, ob, eb, nopw), "error while parsing encrypted ca-key: invalid passphrase or corrupt private key")
+	assert.Empty(t, ob.String())
+	assert.Empty(t, eb.String())
+	os.Setenv("NEBULA_CA_PASSPHRASE", "")
+
 	// test with the user not entering a password
 	ob.Reset()
 	eb.Reset()

cmd/nebula-cert/verify.go

@@ -6,7 +6,6 @@ import (
 	"fmt"
 	"io"
 	"os"
-	"strings"
 	"time"
 
 	"github.com/slackhq/nebula/cert"
@@ -40,23 +39,17 @@ func verify(args []string, out io.Writer, errOut io.Writer) error {
 		return err
 	}
 
-	rawCACert, err := os.ReadFile(*vf.caPath)
+	caFile, err := os.Open(*vf.caPath)
 	if err != nil {
 		return fmt.Errorf("error while reading ca: %w", err)
 	}
+	defer caFile.Close()
 
-	caPool := cert.NewCAPool()
+	caPool, err := cert.NewCAPoolFromPEMReader(caFile)
-	for {
+	if err != nil && !errors.Is(err, cert.ErrExpired) {
-		rawCACert, err = caPool.AddCAFromPEM(rawCACert)
-		if err != nil {
 		return fmt.Errorf("error while adding ca cert to pool: %w", err)
 	}
 
-		if rawCACert == nil || len(rawCACert) == 0 || strings.TrimSpace(string(rawCACert)) == "" {
-			break
-		}
-	}
-
 	rawCert, err := os.ReadFile(*vf.certPath)
 	if err != nil {
 		return fmt.Errorf("unable to read crt: %w", err)

cmd/nebula-cert/verify_test.go

@@ -64,7 +64,7 @@ func Test_verify(t *testing.T) {
 	err = verify([]string{"-ca", caFile.Name(), "-crt", "does_not_exist"}, ob, eb)
 	assert.Empty(t, ob.String())
 	assert.Empty(t, eb.String())
-	require.EqualError(t, err, "error while adding ca cert to pool: input did not contain a valid PEM encoded block")
+	require.ErrorIs(t, err, cert.ErrInvalidPEMBlock)
 
 	// make a ca for later
 	caPub, caPriv, _ := ed25519.GenerateKey(rand.Reader)

cmd/nebula-service/logs_generic.go

@@ -3,8 +3,15 @@
 
 package main
 
-import "github.com/sirupsen/logrus"
+import (
+	"log/slog"
+	"os"
 
-func HookLogger(l *logrus.Logger) {
+	"github.com/slackhq/nebula/logging"
-	// Do nothing, let the logs flow to stdout/stderr
+)
+
+// newPlatformLogger returns a *slog.Logger that writes to stdout. Non-Windows
+// platforms have no special sink to integrate with.
+func newPlatformLogger() *slog.Logger {
+	return logging.NewLogger(os.Stdout)
 }

cmd/nebula-service/logs_windows.go

@@ -1,54 +1,86 @@
 package main
 
 import (
-	"fmt"
+	"context"
-	"io/ioutil"
+	"log/slog"
-	"os"
+	"strings"
+	"sync"
 
-	"github.com/kardianos/service"
+	"github.com/slackhq/nebula/logging"
-	"github.com/sirupsen/logrus"
 )
 
-// HookLogger routes the logrus logs through the service logger so that they end up in the Windows Event Viewer
+// newPlatformLogger returns a *slog.Logger that routes every log record
-// logrus output will be discarded
+// through the Windows service logger so records end up in the Windows
-func HookLogger(l *logrus.Logger) {
+// Event Log. All the heavy lifting (level management, format swap,
-	l.AddHook(newLogHook(logger))
+// timestamp toggle, WithAttrs/WithGroup) comes from logging.NewHandler;
-	l.SetOutput(ioutil.Discard)
+// this file only contributes:
+//
+//   - an io.Writer that forwards each formatted line to the service
+//     logger at the current record's Event Log severity, and
+//   - a thin severityTag that embeds *logging.Handler and overrides
+//     only Handle / WithAttrs / WithGroup, so Event Viewer's severity
+//     column and severity-based filters keep working the way they did
+//     before the slog migration.
+//
+// Format (text vs json) is carried by the embedded *logging.Handler, so
+// logging.format: json in config still produces JSON lines in Event
+// Viewer, same as the pre-slog logrus setup.
+func newPlatformLogger() *slog.Logger {
+	w := &eventLogWriter{}
+	return slog.New(&severityTag{Handler: logging.NewHandler(w), w: w})
 }
 
-type logHook struct {
+// eventLogWriter forwards slog-formatted lines to the Windows service
-	sl service.Logger
+// logger at the severity most recently stashed by severityTag.Handle.
+// The mutex serializes the stash + inner.Handle + Write cycle per record
+// across all concurrent goroutines; slog's builtin text/json handlers
+// each hold their own mutex around Write, but that only protects the
+// Write call itself, not our stash-then-handle sequence.
+type eventLogWriter struct {
+	mu    sync.Mutex
+	level slog.Level
 }
 
-func newLogHook(sl service.Logger) *logHook {
+func (w *eventLogWriter) Write(p []byte) (int, error) {
-	return &logHook{sl: sl}
+	line := strings.TrimRight(string(p), "\n")
-}
+	switch {
-
+	case w.level >= slog.LevelError:
-func (h *logHook) Fire(entry *logrus.Entry) error {
+		return len(p), logger.Error(line)
-	line, err := entry.String()
+	case w.level >= slog.LevelWarn:
-	if err != nil {
+		return len(p), logger.Warning(line)
-		fmt.Fprintf(os.Stderr, "Unable to read entry, %v", err)
-		return err
-	}
-
-	switch entry.Level {
-	case logrus.PanicLevel:
-		return h.sl.Error(line)
-	case logrus.FatalLevel:
-		return h.sl.Error(line)
-	case logrus.ErrorLevel:
-		return h.sl.Error(line)
-	case logrus.WarnLevel:
-		return h.sl.Warning(line)
-	case logrus.InfoLevel:
-		return h.sl.Info(line)
-	case logrus.DebugLevel:
-		return h.sl.Info(line)
 	default:
-		return nil
+		return len(p), logger.Info(line)
 	}
 }
 
-func (h *logHook) Levels() []logrus.Level {
+// severityTag embeds *logging.Handler to pick up everything it does for
-	return logrus.AllLevels
+// free (Enabled, SetLevel, GetLevel, SetFormat, GetFormat,
+// SetDisableTimestamp) and overrides only Handle / WithAttrs / WithGroup
+// so each record's slog.Level is stashed on the writer before formatting
+// and so derived handlers stay wrapped as severityTag rather than
+// downgrading to bare *logging.Handler.
+type severityTag struct {
+	*logging.Handler
+	w *eventLogWriter
+}
+
+func (s *severityTag) Handle(ctx context.Context, r slog.Record) error {
+	s.w.mu.Lock()
+	defer s.w.mu.Unlock()
+	s.w.level = r.Level
+	return s.Handler.Handle(ctx, r)
+}
+
+func (s *severityTag) WithAttrs(attrs []slog.Attr) slog.Handler {
+	if len(attrs) == 0 {
+		return s
+	}
+	return &severityTag{Handler: s.Handler.WithAttrs(attrs).(*logging.Handler), w: s.w}
+}
+
+func (s *severityTag) WithGroup(name string) slog.Handler {
+	if name == "" {
+		return s
+	}
+	return &severityTag{Handler: s.Handler.WithGroup(name).(*logging.Handler), w: s.w}
 }

cmd/nebula-service/main.go

@@ -4,10 +4,12 @@ import (
 	"flag"
 	"fmt"
 	"os"
+	"runtime/debug"
+	"strings"
 
-	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula"
 	"github.com/slackhq/nebula/config"
+	"github.com/slackhq/nebula/logging"
 	"github.com/slackhq/nebula/util"
 )
 
@@ -18,6 +20,17 @@ import (
 // at compile-time.
 var Build string
 
+func init() {
+	if Build == "" {
+		info, ok := debug.ReadBuildInfo()
+		if !ok {
+			return
+		}
+
+		Build = strings.TrimPrefix(info.Main.Version, "v")
+	}
+}
+
 func main() {
 	serviceFlag := flag.String("service", "", "Control the system service.")
 	configPath := flag.String("config", "", "Path to either a file or directory to load configuration from")
@@ -37,10 +50,15 @@ func main() {
 		os.Exit(0)
 	}
 
+	l := logging.NewLogger(os.Stdout)
+
 	if *serviceFlag != "" {
-		doService(configPath, configTest, Build, serviceFlag)
+		if err := doService(configPath, configTest, Build, serviceFlag); err != nil {
+			l.Error("Service command failed", "error", err)
 			os.Exit(1)
 		}
+		return
+	}
 
 	if *configPath == "" {
 		fmt.Println("-config flag must be set")
@@ -48,9 +66,6 @@ func main() {
 		os.Exit(1)
 	}
 
-	l := logrus.New()
-	l.Out = os.Stdout
-
 	c := config.NewC(l)
 	err := c.Load(*configPath)
 	if err != nil {
@@ -58,6 +73,16 @@ func main() {
 		os.Exit(1)
 	}
 
+	if err := logging.ApplyConfig(l, c); err != nil {
+		fmt.Printf("failed to apply logging config: %s", err)
+		os.Exit(1)
+	}
+	c.RegisterReloadCallback(func(c *config.C) {
+		if err := logging.ApplyConfig(l, c); err != nil {
+			l.Error("Failed to reconfigure logger on reload", "error", err)
+		}
+	})
+
 	ctrl, err := nebula.Main(c, *configTest, Build, l, nil)
 	if err != nil {
 		util.LogWithContextIfNeeded("Failed to start", err, l)
@@ -65,8 +90,20 @@ func main() {
 	}
 
 	if !*configTest {
-		ctrl.Start()
+		wait, err := ctrl.Start()
-		ctrl.ShutdownBlock()
+		if err != nil {
+			util.LogWithContextIfNeeded("Error while running", err, l)
+			os.Exit(1)
+		}
+
+		go ctrl.ShutdownBlock()
+
+		if err := wait(); err != nil {
+			l.Error("Nebula stopped due to fatal error", "error", err)
+			os.Exit(2)
+		}
+
+		l.Info("Goodbye")
 	}
 
 	os.Exit(0)

cmd/nebula-service/service.go

@@ -7,9 +7,9 @@ import (
 	"path/filepath"
 
 	"github.com/kardianos/service"
-	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula"
 	"github.com/slackhq/nebula/config"
+	"github.com/slackhq/nebula/logging"
 )
 
 var logger service.Logger
@@ -25,8 +25,7 @@ func (p *program) Start(s service.Service) error {
 	// Start should not block.
 	logger.Info("Nebula service starting.")
 
-	l := logrus.New()
+	l := newPlatformLogger()
-	HookLogger(l)
 
 	c := config.NewC(l)
 	err := c.Load(*p.configPath)
@@ -34,6 +33,15 @@ func (p *program) Start(s service.Service) error {
 		return fmt.Errorf("failed to load config: %s", err)
 	}
 
+	if err := logging.ApplyConfig(l, c); err != nil {
+		return fmt.Errorf("failed to apply logging config: %s", err)
+	}
+	c.RegisterReloadCallback(func(c *config.C) {
+		if err := logging.ApplyConfig(l, c); err != nil {
+			l.Error("Failed to reconfigure logger on reload", "error", err)
+		}
+	})
+
 	p.control, err = nebula.Main(c, *p.configTest, Build, l, nil)
 	if err != nil {
 		return err
@@ -57,11 +65,11 @@ func fileExists(filename string) bool {
 	return true
 }
 
-func doService(configPath *string, configTest *bool, build string, serviceFlag *string) {
+func doService(configPath *string, configTest *bool, build string, serviceFlag *string) error {
 	if *configPath == "" {
 		ex, err := os.Executable()
 		if err != nil {
-			panic(err)
+			return err
 		}
 		*configPath = filepath.Dir(ex) + "/config.yaml"
 		if !fileExists(*configPath) {
@@ -85,16 +93,16 @@ func doService(configPath *string, configTest *bool, build string, serviceFlag *
 	// Here are what the different loggers are doing:
 	// - `log` is the standard go log utility, meant to be used while the process is still attached to stdout/stderr
 	// - `logger` is the service log utility that may be attached to a special place depending on OS (Windows will have it attached to the event log)
-	// - above, in `Run` we create a `logrus.Logger` which is what nebula expects to use
+	// - in program.Start we build a *slog.Logger via newPlatformLogger; on non-Windows that is a stdout-backed slog logger, on Windows it routes records through the service logger
 	s, err := service.New(prg, svcConfig)
 	if err != nil {
-		log.Fatal(err)
+		return err
 	}
 
 	errs := make(chan error, 5)
 	logger, err = s.Logger(errs)
 	if err != nil {
-		log.Fatal(err)
+		return err
 	}
 
 	go func() {
@@ -109,18 +117,16 @@ func doService(configPath *string, configTest *bool, build string, serviceFlag *
 
 	switch *serviceFlag {
 	case "run":
-		err = s.Run()
+		if err := s.Run(); err != nil {
-		if err != nil {
 			// Route any errors to the system logger
 			logger.Error(err)
 		}
 	default:
-		err := service.Control(s, *serviceFlag)
+		if err := service.Control(s, *serviceFlag); err != nil {
-		if err != nil {
 			log.Printf("Valid actions: %q\n", service.ControlAction)
-			log.Fatal(err)
+			return err
 		}
-		return
 	}
 
+	return nil
 }

cmd/nebula/main.go

@@ -4,10 +4,12 @@ import (
 	"flag"
 	"fmt"
 	"os"
+	"runtime/debug"
+	"strings"
 
-	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula"
 	"github.com/slackhq/nebula/config"
+	"github.com/slackhq/nebula/logging"
 	"github.com/slackhq/nebula/util"
 )
 
@@ -18,6 +20,17 @@ import (
 // at compile-time.
 var Build string
 
+func init() {
+	if Build == "" {
+		info, ok := debug.ReadBuildInfo()
+		if !ok {
+			return
+		}
+
+		Build = strings.TrimPrefix(info.Main.Version, "v")
+	}
+}
+
 func main() {
 	configPath := flag.String("config", "", "Path to either a file or directory to load configuration from")
 	configTest := flag.Bool("test", false, "Test the config and print the end result. Non zero exit indicates a faulty config")
@@ -42,8 +55,7 @@ func main() {
 		os.Exit(1)
 	}
 
-	l := logrus.New()
+	l := logging.NewLogger(os.Stdout)
-	l.Out = os.Stdout
 
 	c := config.NewC(l)
 	err := c.Load(*configPath)
@@ -52,6 +64,16 @@ func main() {
 		os.Exit(1)
 	}
 
+	if err := logging.ApplyConfig(l, c); err != nil {
+		fmt.Printf("failed to apply logging config: %s", err)
+		os.Exit(1)
+	}
+	c.RegisterReloadCallback(func(c *config.C) {
+		if err := logging.ApplyConfig(l, c); err != nil {
+			l.Error("Failed to reconfigure logger on reload", "error", err)
+		}
+	})
+
 	ctrl, err := nebula.Main(c, *configTest, Build, l, nil)
 	if err != nil {
 		util.LogWithContextIfNeeded("Failed to start", err, l)
@@ -59,9 +81,21 @@ func main() {
 	}
 
 	if !*configTest {
-		ctrl.Start()
+		wait, err := ctrl.Start()
+		if err != nil {
+			util.LogWithContextIfNeeded("Error while running", err, l)
+			os.Exit(1)
+		}
+
+		go ctrl.ShutdownBlock()
 		notifyReady(l)
-		ctrl.ShutdownBlock()
+
+		if err := wait(); err != nil {
+			l.Error("Nebula stopped due to fatal error", "error", err)
+			os.Exit(2)
+		}
+
+		l.Info("Goodbye")
 	}
 
 	os.Exit(0)

cmd/nebula/notify_linux.go

@@ -1,11 +1,10 @@
 package main
 
 import (
+	"log/slog"
 	"net"
 	"os"
 	"time"
-
-	"github.com/sirupsen/logrus"
 )
 
 // SdNotifyReady tells systemd the service is ready and dependent services can now be started
@@ -13,30 +12,30 @@ import (
 // https://www.freedesktop.org/software/systemd/man/systemd.service.html
 const SdNotifyReady = "READY=1"
 
-func notifyReady(l *logrus.Logger) {
+func notifyReady(l *slog.Logger) {
 	sockName := os.Getenv("NOTIFY_SOCKET")
 	if sockName == "" {
-		l.Debugln("NOTIFY_SOCKET systemd env var not set, not sending ready signal")
+		l.Debug("NOTIFY_SOCKET systemd env var not set, not sending ready signal")
 		return
 	}
 
 	conn, err := net.DialTimeout("unixgram", sockName, time.Second)
 	if err != nil {
-		l.WithError(err).Error("failed to connect to systemd notification socket")
+		l.Error("failed to connect to systemd notification socket", "error", err)
 		return
 	}
 	defer conn.Close()
 
 	err = conn.SetWriteDeadline(time.Now().Add(time.Second))
 	if err != nil {
-		l.WithError(err).Error("failed to set the write deadline for the systemd notification socket")
+		l.Error("failed to set the write deadline for the systemd notification socket", "error", err)
 		return
 	}
 
 	if _, err = conn.Write([]byte(SdNotifyReady)); err != nil {
-		l.WithError(err).Error("failed to signal the systemd notification socket")
+		l.Error("failed to signal the systemd notification socket", "error", err)
 		return
 	}
 
-	l.Debugln("notified systemd the service is ready")
+	l.Debug("notified systemd the service is ready")
 }

cmd/nebula/notify_notlinux.go

@@ -3,8 +3,8 @@
 
 package main
 
-import "github.com/sirupsen/logrus"
+import "log/slog"
 
-func notifyReady(_ *logrus.Logger) {
+func notifyReady(_ *slog.Logger) {
 	// No init service to notify
 }

config/config.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	"log/slog"
 	"math"
 	"os"
 	"os/signal"
@@ -16,8 +17,7 @@ import (
 	"time"
 
 	"dario.cat/mergo"
-	"github.com/sirupsen/logrus"
+	"go.yaml.in/yaml/v3"
-	"gopkg.in/yaml.v3"
 )
 
 type C struct {
@@ -26,11 +26,11 @@ type C struct {
 	Settings    map[string]any
 	oldSettings map[string]any
 	callbacks   []func(*C)
-	l           *logrus.Logger
+	l           *slog.Logger
 	reloadLock  sync.Mutex
 }
 
-func NewC(l *logrus.Logger) *C {
+func NewC(l *slog.Logger) *C {
 	return &C{
 		Settings: make(map[string]any),
 		l:        l,
@@ -107,12 +107,18 @@ func (c *C) HasChanged(k string) bool {
 
 	newVals, err := yaml.Marshal(nv)
 	if err != nil {
-		c.l.WithField("config_path", k).WithError(err).Error("Error while marshaling new config")
+		c.l.Error("Error while marshaling new config",
+			"config_path", k,
+			"error", err,
+		)
 	}
 
 	oldVals, err := yaml.Marshal(ov)
 	if err != nil {
-		c.l.WithField("config_path", k).WithError(err).Error("Error while marshaling old config")
+		c.l.Error("Error while marshaling old config",
+			"config_path", k,
+			"error", err,
+		)
 	}
 
 	return string(newVals) != string(oldVals)
@@ -154,7 +160,10 @@ func (c *C) ReloadConfig() {
 
 	err := c.Load(c.path)
 	if err != nil {
-		c.l.WithField("config_path", c.path).WithError(err).Error("Error occurred while reloading config")
+		c.l.Error("Error occurred while reloading config",
+			"config_path", c.path,
+			"error", err,
+		)
 		return
 	}
 

config/config_test.go

@@ -10,7 +10,7 @@ import (
 	"github.com/slackhq/nebula/test"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
-	"gopkg.in/yaml.v3"
+	"go.yaml.in/yaml/v3"
 )
 
 func TestConfig_Load(t *testing.T) {

connection_manager.go

@@ -5,13 +5,13 @@ import (
 	"context"
 	"encoding/binary"
 	"fmt"
+	"log/slog"
 	"net/netip"
 	"sync"
 	"sync/atomic"
 	"time"
 
 	"github.com/rcrowley/go-metrics"
-	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula/cert"
 	"github.com/slackhq/nebula/config"
 	"github.com/slackhq/nebula/header"
@@ -47,10 +47,10 @@ type connectionManager struct {
 
 	metricsTxPunchy metrics.Counter
 
-	l *logrus.Logger
+	l *slog.Logger
 }
 
-func newConnectionManagerFromConfig(l *logrus.Logger, c *config.C, hm *HostMap, p *Punchy) *connectionManager {
+func newConnectionManagerFromConfig(l *slog.Logger, c *config.C, hm *HostMap, p *Punchy) *connectionManager {
 	cm := &connectionManager{
 		hostMap:         hm,
 		l:               l,
@@ -85,9 +85,10 @@ func (cm *connectionManager) reload(c *config.C, initial bool) {
 		old := cm.getInactivityTimeout()
 		cm.inactivityTimeout.Store((int64)(c.GetDuration("tunnels.inactivity_timeout", 10*time.Minute)))
 		if !initial {
-			cm.l.WithField("oldDuration", old).
+			cm.l.Info("Inactivity timeout has changed",
-				WithField("newDuration", cm.getInactivityTimeout()).
+				"oldDuration", old,
-				Info("Inactivity timeout has changed")
+				"newDuration", cm.getInactivityTimeout(),
+			)
 		}
 	}
 
@@ -95,9 +96,10 @@ func (cm *connectionManager) reload(c *config.C, initial bool) {
 		old := cm.dropInactive.Load()
 		cm.dropInactive.Store(c.GetBool("tunnels.drop_inactive", false))
 		if !initial {
-			cm.l.WithField("oldBool", old).
+			cm.l.Info("Drop inactive setting has changed",
-				WithField("newBool", cm.dropInactive.Load()).
+				"oldBool", old,
-				Info("Drop inactive setting has changed")
+				"newBool", cm.dropInactive.Load(),
+			)
 		}
 	}
 }
@@ -143,6 +145,7 @@ func (cm *connectionManager) getAndResetTrafficCheck(h *HostInfo, now time.Time)
 func (cm *connectionManager) AddTrafficWatch(h *HostInfo) {
 	if h.out.Swap(true) == false {
 		cm.trafficTimer.Add(h.localIndexId, cm.checkInterval)
+		cm.intf.pmtudManager.OnTunnelUp(h)
 	}
 }
 
@@ -178,6 +181,7 @@ func (cm *connectionManager) doTrafficCheck(localIndex uint32, p, nb, out []byte
 
 	switch decision {
 	case deleteTunnel:
+		cm.intf.pmtudManager.OnTunnelDown(hostinfo)
 		if cm.hostMap.DeleteHostInfo(hostinfo) {
 			// Only clearing the lighthouse cache if this is the last hostinfo for this vpn ip in the hostmap
 			cm.intf.lightHouse.DeleteVpnAddrs(hostinfo.vpnAddrs)
@@ -197,8 +201,15 @@ func (cm *connectionManager) doTrafficCheck(localIndex uint32, p, nb, out []byte
 		cm.tryRehandshake(hostinfo)
 
 	case sendTestPacket:
+		// Defer to pmtud if it has a confirmed PMTU > floor for this peer:
+		// the probe at the confirmed size verifies both liveness AND that
+		// the discovered PMTU still fits, so we don't burn a separate test
+		// packet on top of it. If pmtud declines (disabled, peer unsupported,
+		// or no confirmed size yet) we fall back to the regular test.
+		if !cm.intf.pmtudManager.MaybeProbeAsTest(hostinfo) {
 			cm.intf.SendMessageToHostInfo(header.Test, header.TestRequest, hostinfo, p, nb, out)
 		}
+	}
 
 	cm.resetRelayTrafficCheck(hostinfo)
 }
@@ -256,7 +267,7 @@ func (cm *connectionManager) migrateRelayUsed(oldhostinfo, newhostinfo *HostInfo
 			var err error
 			index, err = AddRelay(cm.l, newhostinfo, cm.hostMap, r.PeerAddr, nil, r.Type, Requested)
 			if err != nil {
-				cm.l.WithError(err).Error("failed to migrate relay to new hostinfo")
+				cm.l.Error("failed to migrate relay to new hostinfo", "error", err)
 				continue
 			}
 			switch r.Type {
@@ -304,16 +315,16 @@ func (cm *connectionManager) migrateRelayUsed(oldhostinfo, newhostinfo *HostInfo
 
 		msg, err := req.Marshal()
 		if err != nil {
-			cm.l.WithError(err).Error("failed to marshal Control message to migrate relay")
+			cm.l.Error("failed to marshal Control message to migrate relay", "error", err)
 		} else {
 			cm.intf.SendMessageToHostInfo(header.Control, 0, newhostinfo, msg, make([]byte, 12), make([]byte, mtu))
-			cm.l.WithFields(logrus.Fields{
+			cm.l.Info("send CreateRelayRequest",
-				"relayFrom":           req.RelayFromAddr,
+				"relayFrom", req.RelayFromAddr,
-				"relayTo":             req.RelayToAddr,
+				"relayTo", req.RelayToAddr,
-				"initiatorRelayIndex": req.InitiatorRelayIndex,
+				"initiatorRelayIndex", req.InitiatorRelayIndex,
-				"responderRelayIndex": req.ResponderRelayIndex,
+				"responderRelayIndex", req.ResponderRelayIndex,
-				"vpnAddrs":            newhostinfo.vpnAddrs}).
+				"vpnAddrs", newhostinfo.vpnAddrs,
-				Info("send CreateRelayRequest")
+			)
 		}
 	}
 }
@@ -325,7 +336,7 @@ func (cm *connectionManager) makeTrafficDecision(localIndex uint32, now time.Tim
 
 	hostinfo := cm.hostMap.Indexes[localIndex]
 	if hostinfo == nil {
-		cm.l.WithField("localIndex", localIndex).Debugln("Not found in hostmap")
+		cm.l.Debug("Not found in hostmap", "localIndex", localIndex)
 		return doNothing, nil, nil
 	}
 
@@ -345,10 +356,10 @@ func (cm *connectionManager) makeTrafficDecision(localIndex uint32, now time.Tim
 	// A hostinfo is determined alive if there is incoming traffic
 	if inTraffic {
 		decision := doNothing
-		if cm.l.Level >= logrus.DebugLevel {
+		if cm.l.Enabled(context.Background(), slog.LevelDebug) {
-			hostinfo.logger(cm.l).
+			hostinfo.logger(cm.l).Debug("Tunnel status",
-				WithField("tunnelCheck", m{"state": "alive", "method": "passive"}).
+				"tunnelCheck", m{"state": "alive", "method": "passive"},
-				Debug("Tunnel status")
+			)
 		}
 		hostinfo.pendingDeletion.Store(false)
 
@@ -375,9 +386,9 @@ func (cm *connectionManager) makeTrafficDecision(localIndex uint32, now time.Tim
 
 	if hostinfo.pendingDeletion.Load() {
 		// We have already sent a test packet and nothing was returned, this hostinfo is dead
-		hostinfo.logger(cm.l).
+		hostinfo.logger(cm.l).Info("Tunnel status",
-			WithField("tunnelCheck", m{"state": "dead", "method": "active"}).
+			"tunnelCheck", m{"state": "dead", "method": "active"},
-			Info("Tunnel status")
+		)
 
 		return deleteTunnel, hostinfo, nil
 	}
@@ -388,10 +399,10 @@ func (cm *connectionManager) makeTrafficDecision(localIndex uint32, now time.Tim
 			inactiveFor, isInactive := cm.isInactive(hostinfo, now)
 			if isInactive {
 				// Tunnel is inactive, tear it down
-				hostinfo.logger(cm.l).
+				hostinfo.logger(cm.l).Info("Dropping tunnel due to inactivity",
-					WithField("inactiveDuration", inactiveFor).
+					"inactiveDuration", inactiveFor,
-					WithField("primary", mainHostInfo).
+					"primary", mainHostInfo,
-					Info("Dropping tunnel due to inactivity")
+				)
 
 				return closeTunnel, hostinfo, primary
 			}
@@ -410,18 +421,18 @@ func (cm *connectionManager) makeTrafficDecision(localIndex uint32, now time.Tim
 			cm.sendPunch(hostinfo)
 		}
 
-		if cm.l.Level >= logrus.DebugLevel {
+		if cm.l.Enabled(context.Background(), slog.LevelDebug) {
-			hostinfo.logger(cm.l).
+			hostinfo.logger(cm.l).Debug("Tunnel status",
-				WithField("tunnelCheck", m{"state": "testing", "method": "active"}).
+				"tunnelCheck", m{"state": "testing", "method": "active"},
-				Debug("Tunnel status")
+			)
 		}
 
 		// Send a test packet to trigger an authenticated tunnel test, this should suss out any lingering tunnel issues
 		decision = sendTestPacket
 
 	} else {
-		if cm.l.Level >= logrus.DebugLevel {
+		if cm.l.Enabled(context.Background(), slog.LevelDebug) {
-			hostinfo.logger(cm.l).Debugf("Hostinfo sadness")
+			hostinfo.logger(cm.l).Debug("Hostinfo sadness")
 		}
 	}
 
@@ -493,14 +504,16 @@ func (cm *connectionManager) isInvalidCertificate(now time.Time, hostinfo *HostI
 		return false //cert is still valid! yay!
 	} else if err == cert.ErrBlockListed { //avoiding errors.Is for speed
 		// Block listed certificates should always be disconnected
-		hostinfo.logger(cm.l).WithError(err).
+		hostinfo.logger(cm.l).Info("Remote certificate is blocked, tearing down the tunnel",
-			WithField("fingerprint", remoteCert.Fingerprint).
+			"error", err,
-			Info("Remote certificate is blocked, tearing down the tunnel")
+			"fingerprint", remoteCert.Fingerprint,
+		)
 		return true
 	} else if cm.intf.disconnectInvalid.Load() {
-		hostinfo.logger(cm.l).WithError(err).
+		hostinfo.logger(cm.l).Info("Remote certificate is no longer valid, tearing down the tunnel",
-			WithField("fingerprint", remoteCert.Fingerprint).
+			"error", err,
-			Info("Remote certificate is no longer valid, tearing down the tunnel")
+			"fingerprint", remoteCert.Fingerprint,
+		)
 		return true
 	} else {
 		//if we reach here, the cert is no longer valid, but we're configured to keep tunnels from now-invalid certs open
@@ -539,10 +552,11 @@ func (cm *connectionManager) tryRehandshake(hostinfo *HostInfo) {
 	curCrtVersion := curCrt.Version()
 	myCrt := cs.getCertificate(curCrtVersion)
 	if myCrt == nil {
-		cm.l.WithField("vpnAddrs", hostinfo.vpnAddrs).
+		cm.l.Info("Re-handshaking with remote",
-			WithField("version", curCrtVersion).
+			"vpnAddrs", hostinfo.vpnAddrs,
-			WithField("reason", "local certificate removed").
+			"version", curCrtVersion,
-			Info("Re-handshaking with remote")
+			"reason", "local certificate removed",
+		)
 		cm.intf.handshakeManager.StartHandshake(hostinfo.vpnAddrs[0], nil)
 		return
 	}
@@ -550,11 +564,12 @@ func (cm *connectionManager) tryRehandshake(hostinfo *HostInfo) {
 	if peerCrt != nil && curCrtVersion < peerCrt.Certificate.Version() {
 		// if our certificate version is less than theirs, and we have a matching version available, rehandshake?
 		if cs.getCertificate(peerCrt.Certificate.Version()) != nil {
-			cm.l.WithField("vpnAddrs", hostinfo.vpnAddrs).
+			cm.l.Info("Re-handshaking with remote",
-				WithField("version", curCrtVersion).
+				"vpnAddrs", hostinfo.vpnAddrs,
-				WithField("peerVersion", peerCrt.Certificate.Version()).
+				"version", curCrtVersion,
-				WithField("reason", "local certificate version lower than peer, attempting to correct").
+				"peerVersion", peerCrt.Certificate.Version(),
-				Info("Re-handshaking with remote")
+				"reason", "local certificate version lower than peer, attempting to correct",
+			)
 			cm.intf.handshakeManager.StartHandshake(hostinfo.vpnAddrs[0], func(hh *HandshakeHostInfo) {
 				hh.initiatingVersionOverride = peerCrt.Certificate.Version()
 			})
@@ -562,17 +577,19 @@ func (cm *connectionManager) tryRehandshake(hostinfo *HostInfo) {
 		}
 	}
 	if !bytes.Equal(curCrt.Signature(), myCrt.Signature()) {
-		cm.l.WithField("vpnAddrs", hostinfo.vpnAddrs).
+		cm.l.Info("Re-handshaking with remote",
-			WithField("reason", "local certificate is not current").
+			"vpnAddrs", hostinfo.vpnAddrs,
-			Info("Re-handshaking with remote")
+			"reason", "local certificate is not current",
+		)
 
 		cm.intf.handshakeManager.StartHandshake(hostinfo.vpnAddrs[0], nil)
 		return
 	}
 	if curCrtVersion < cs.initiatingVersion {
-		cm.l.WithField("vpnAddrs", hostinfo.vpnAddrs).
+		cm.l.Info("Re-handshaking with remote",
-			WithField("reason", "current cert version < pki.initiatingVersion").
+			"vpnAddrs", hostinfo.vpnAddrs,
-			Info("Re-handshaking with remote")
+			"reason", "current cert version < pki.initiatingVersion",
+		)
 
 		cm.intf.handshakeManager.StartHandshake(hostinfo.vpnAddrs[0], nil)
 		return

connection_manager_test.go

@@ -7,9 +7,9 @@ import (
 	"testing"
 	"time"
 
-	"github.com/flynn/noise"
 	"github.com/slackhq/nebula/cert"
 	"github.com/slackhq/nebula/config"
+	"github.com/slackhq/nebula/overlay/overlaytest"
 	"github.com/slackhq/nebula/test"
 	"github.com/slackhq/nebula/udp"
 	"github.com/stretchr/testify/assert"
@@ -46,13 +46,13 @@ func Test_NewConnectionManagerTest(t *testing.T) {
 		initiatingVersion: cert.Version1,
 		privateKey:        []byte{},
 		v1Cert:            &dummyCert{version: cert.Version1},
-		v1HandshakeBytes:  []byte{},
+		v1Credential:      nil,
 	}
 
 	lh := newTestLighthouse()
 	ifce := &Interface{
 		hostMap:          hostMap,
-		inside:           &test.NoopTun{},
+		inside:           &overlaytest.NoopTun{},
 		outside:          &udp.NoopConn{},
 		firewall:         &Firewall{},
 		lightHouse:       lh,
@@ -63,10 +63,12 @@ func Test_NewConnectionManagerTest(t *testing.T) {
 	ifce.pki.cs.Store(cs)
 
 	// Create manager
-	conf := config.NewC(l)
+	conf := config.NewC(test.NewLogger())
-	punchy := NewPunchyFromConfig(l, conf)
+	punchy := NewPunchyFromConfig(test.NewLogger(), conf)
-	nc := newConnectionManagerFromConfig(l, conf, hostMap, punchy)
+	nc := newConnectionManagerFromConfig(test.NewLogger(), conf, hostMap, punchy)
 	nc.intf = ifce
+	ifce.pmtudManager = newPMTUDManagerFromConfig(test.NewLogger(), conf, ifce.inside)
+	ifce.pmtudManager.intf = ifce
 	p := []byte("")
 	nb := make([]byte, 12, 12)
 	out := make([]byte, mtu)
@@ -79,7 +81,6 @@ func Test_NewConnectionManagerTest(t *testing.T) {
 	}
 	hostinfo.ConnectionState = &ConnectionState{
 		myCert: &dummyCert{version: cert.Version1},
-		H:      &noise.HandshakeState{},
 	}
 	nc.hostMap.unlockedAddHostInfo(hostinfo, ifce)
 
@@ -129,13 +130,13 @@ func Test_NewConnectionManagerTest2(t *testing.T) {
 		initiatingVersion: cert.Version1,
 		privateKey:        []byte{},
 		v1Cert:            &dummyCert{version: cert.Version1},
-		v1HandshakeBytes:  []byte{},
+		v1Credential:      nil,
 	}
 
 	lh := newTestLighthouse()
 	ifce := &Interface{
 		hostMap:          hostMap,
-		inside:           &test.NoopTun{},
+		inside:           &overlaytest.NoopTun{},
 		outside:          &udp.NoopConn{},
 		firewall:         &Firewall{},
 		lightHouse:       lh,
@@ -146,10 +147,12 @@ func Test_NewConnectionManagerTest2(t *testing.T) {
 	ifce.pki.cs.Store(cs)
 
 	// Create manager
-	conf := config.NewC(l)
+	conf := config.NewC(test.NewLogger())
-	punchy := NewPunchyFromConfig(l, conf)
+	punchy := NewPunchyFromConfig(test.NewLogger(), conf)
-	nc := newConnectionManagerFromConfig(l, conf, hostMap, punchy)
+	nc := newConnectionManagerFromConfig(test.NewLogger(), conf, hostMap, punchy)
 	nc.intf = ifce
+	ifce.pmtudManager = newPMTUDManagerFromConfig(test.NewLogger(), conf, ifce.inside)
+	ifce.pmtudManager.intf = ifce
 	p := []byte("")
 	nb := make([]byte, 12, 12)
 	out := make([]byte, mtu)
@@ -162,7 +165,6 @@ func Test_NewConnectionManagerTest2(t *testing.T) {
 	}
 	hostinfo.ConnectionState = &ConnectionState{
 		myCert: &dummyCert{version: cert.Version1},
-		H:      &noise.HandshakeState{},
 	}
 	nc.hostMap.unlockedAddHostInfo(hostinfo, ifce)
 
@@ -214,13 +216,13 @@ func Test_NewConnectionManager_DisconnectInactive(t *testing.T) {
 		initiatingVersion: cert.Version1,
 		privateKey:        []byte{},
 		v1Cert:            &dummyCert{version: cert.Version1},
-		v1HandshakeBytes:  []byte{},
+		v1Credential:      nil,
 	}
 
 	lh := newTestLighthouse()
 	ifce := &Interface{
 		hostMap:          hostMap,
-		inside:           &test.NoopTun{},
+		inside:           &overlaytest.NoopTun{},
 		outside:          &udp.NoopConn{},
 		firewall:         &Firewall{},
 		lightHouse:       lh,
@@ -231,12 +233,12 @@ func Test_NewConnectionManager_DisconnectInactive(t *testing.T) {
 	ifce.pki.cs.Store(cs)
 
 	// Create manager
-	conf := config.NewC(l)
+	conf := config.NewC(test.NewLogger())
 	conf.Settings["tunnels"] = map[string]any{
 		"drop_inactive": true,
 	}
-	punchy := NewPunchyFromConfig(l, conf)
+	punchy := NewPunchyFromConfig(test.NewLogger(), conf)
-	nc := newConnectionManagerFromConfig(l, conf, hostMap, punchy)
+	nc := newConnectionManagerFromConfig(test.NewLogger(), conf, hostMap, punchy)
 	assert.True(t, nc.dropInactive.Load())
 	nc.intf = ifce
 
@@ -248,7 +250,6 @@ func Test_NewConnectionManager_DisconnectInactive(t *testing.T) {
 	}
 	hostinfo.ConnectionState = &ConnectionState{
 		myCert: &dummyCert{version: cert.Version1},
-		H:      &noise.HandshakeState{},
 	}
 	nc.hostMap.unlockedAddHostInfo(hostinfo, ifce)
 
@@ -341,13 +342,13 @@ func Test_NewConnectionManagerTest_DisconnectInvalid(t *testing.T) {
 	cs := &CertState{
 		privateKey:   []byte{},
 		v1Cert:       &dummyCert{},
-		v1HandshakeBytes: []byte{},
+		v1Credential: nil,
 	}
 
 	lh := newTestLighthouse()
 	ifce := &Interface{
 		hostMap:          hostMap,
-		inside:           &test.NoopTun{},
+		inside:           &overlaytest.NoopTun{},
 		outside:          &udp.NoopConn{},
 		firewall:         &Firewall{},
 		lightHouse:       lh,
@@ -360,10 +361,12 @@ func Test_NewConnectionManagerTest_DisconnectInvalid(t *testing.T) {
 	ifce.disconnectInvalid.Store(true)
 
 	// Create manager
-	conf := config.NewC(l)
+	conf := config.NewC(test.NewLogger())
-	punchy := NewPunchyFromConfig(l, conf)
+	punchy := NewPunchyFromConfig(test.NewLogger(), conf)
-	nc := newConnectionManagerFromConfig(l, conf, hostMap, punchy)
+	nc := newConnectionManagerFromConfig(test.NewLogger(), conf, hostMap, punchy)
 	nc.intf = ifce
+	ifce.pmtudManager = newPMTUDManagerFromConfig(test.NewLogger(), conf, ifce.inside)
+	ifce.pmtudManager.intf = ifce
 	ifce.connectionManager = nc
 
 	hostinfo := &HostInfo{
@@ -371,7 +374,6 @@ func Test_NewConnectionManagerTest_DisconnectInvalid(t *testing.T) {
 		ConnectionState: &ConnectionState{
 			myCert:   &dummyCert{},
 			peerCert: cachedPeerCert,
-			H:        &noise.HandshakeState{},
 		},
 	}
 	nc.hostMap.unlockedAddHostInfo(hostinfo, ifce)

connection_state.go

@@ -1,16 +1,12 @@
 package nebula
 
 import (
-	"crypto/rand"
 	"encoding/json"
-	"fmt"
 	"sync"
 	"sync/atomic"
 
-	"github.com/flynn/noise"
-	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula/cert"
-	"github.com/slackhq/nebula/noiseutil"
+	"github.com/slackhq/nebula/handshake"
 )
 
 const ReplayWindow = 1024
@@ -18,7 +14,6 @@ const ReplayWindow = 1024
 type ConnectionState struct {
 	eKey           *NebulaCipherState
 	dKey           *NebulaCipherState
-	H              *noise.HandshakeState
 	myCert         cert.Certificate
 	peerCert       *cert.CachedCertificate
 	initiator      bool
@@ -27,60 +22,24 @@ type ConnectionState struct {
 	writeLock      sync.Mutex
 }
 
-func NewConnectionState(l *logrus.Logger, cs *CertState, crt cert.Certificate, initiator bool, pattern noise.HandshakePattern) (*ConnectionState, error) {
+// newConnectionStateFromResult builds a fully-populated ConnectionState from a
-	var dhFunc noise.DHFunc
+// completed handshake.Result. It seeds messageCounter and the replay window so
-	switch crt.Curve() {
+// that the post-handshake message indices already used on the wire don't count
-	case cert.Curve_CURVE25519:
+// as missed traffic in the data plane.
-		dhFunc = noise.DH25519
+func newConnectionStateFromResult(r *handshake.Result) *ConnectionState {
-	case cert.Curve_P256:
-		if cs.pkcs11Backed {
-			dhFunc = noiseutil.DHP256PKCS11
-		} else {
-			dhFunc = noiseutil.DHP256
-		}
-	default:
-		return nil, fmt.Errorf("invalid curve: %s", crt.Curve())
-	}
-
-	var ncs noise.CipherSuite
-	if cs.cipher == "chachapoly" {
-		ncs = noise.NewCipherSuite(dhFunc, noise.CipherChaChaPoly, noise.HashSHA256)
-	} else {
-		ncs = noise.NewCipherSuite(dhFunc, noiseutil.CipherAESGCM, noise.HashSHA256)
-	}
-
-	static := noise.DHKey{Private: cs.privateKey, Public: crt.PublicKey()}
-
-	b := NewBits(ReplayWindow)
-	// Clear out bit 0, we never transmit it, and we don't want it showing as packet loss
-	b.Update(l, 0)
-
-	hs, err := noise.NewHandshakeState(noise.Config{
-		CipherSuite:   ncs,
-		Random:        rand.Reader,
-		Pattern:       pattern,
-		Initiator:     initiator,
-		StaticKeypair: static,
-		//NOTE: These should come from CertState (pki.go) when we finally implement it
-		PresharedKey:          []byte{},
-		PresharedKeyPlacement: 0,
-	})
-	if err != nil {
-		return nil, fmt.Errorf("NewConnectionState: %s", err)
-	}
-
-	// The queue and ready params prevent a counter race that would happen when
-	// sending stored packets and simultaneously accepting new traffic.
 	ci := &ConnectionState{
-		H:         hs,
+		myCert:    r.MyCert,
-		initiator: initiator,
+		initiator: r.Initiator,
-		window:    b,
+		peerCert:  r.RemoteCert,
-		myCert:    crt,
+		eKey:      NewNebulaCipherState(r.EKey),
+		dKey:      NewNebulaCipherState(r.DKey),
+		window:    NewBits(ReplayWindow),
 	}
-	// always start the counter from 2, as packet 1 and packet 2 are handshake packets.
+	ci.messageCounter.Add(r.MessageIndex)
-	ci.messageCounter.Add(2)
+	for i := uint64(1); i <= r.MessageIndex; i++ {
-
+		ci.window.Update(nil, i)
-	return ci, nil
+	}
+	return ci
 }
 
 func (cs *ConnectionState) MarshalJSON() ([]byte, error) {

connection_state_test.go

@@ -0,0 +1,114 @@
+package nebula
+
+import (
+	"net/netip"
+	"testing"
+	"time"
+
+	"github.com/flynn/noise"
+	"github.com/slackhq/nebula/cert"
+	ct "github.com/slackhq/nebula/cert_test"
+	"github.com/slackhq/nebula/handshake"
+	"github.com/slackhq/nebula/header"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// runTestHandshake runs a complete IX handshake between two freshly-built
+// peers and returns the initiator and responder Results. Used to produce
+// real cipher states for tests that need to exercise post-handshake glue.
+func runTestHandshake(t *testing.T) (initR, respR *handshake.Result) {
+	t.Helper()
+
+	ca, _, caKey, _ := ct.NewTestCaCert(
+		cert.Version2, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil,
+	)
+	caPool := ct.NewTestCAPool(ca)
+
+	makeCreds := func(name string, networks []netip.Prefix) handshake.GetCredentialFunc {
+		c, _, rawKey, _ := ct.NewTestCert(
+			cert.Version2, cert.Curve_CURVE25519, ca, caKey,
+			name, ca.NotBefore(), ca.NotAfter(), networks, nil, nil,
+		)
+		priv, _, _, err := cert.UnmarshalPrivateKeyFromPEM(rawKey)
+		require.NoError(t, err)
+		hsBytes, err := c.MarshalForHandshakes()
+		require.NoError(t, err)
+		ncs := noise.NewCipherSuite(noise.DH25519, noise.CipherChaChaPoly, noise.HashSHA256)
+		cred := handshake.NewCredential(c, hsBytes, priv, ncs)
+		return func(v cert.Version) *handshake.Credential {
+			if v == cert.Version2 {
+				return cred
+			}
+			return nil
+		}
+	}
+
+	verifier := func(c cert.Certificate) (*cert.CachedCertificate, error) {
+		return caPool.VerifyCertificate(time.Now(), c)
+	}
+
+	initCreds := makeCreds("initiator", []netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")})
+	respCreds := makeCreds("responder", []netip.Prefix{netip.MustParsePrefix("10.0.0.2/24")})
+
+	initM, err := handshake.NewMachine(
+		cert.Version2, initCreds, verifier,
+		func() (uint32, error) { return 1000, nil },
+		true, header.HandshakeIXPSK0,
+	)
+	require.NoError(t, err)
+
+	respM, err := handshake.NewMachine(
+		cert.Version2, respCreds, verifier,
+		func() (uint32, error) { return 2000, nil },
+		false, header.HandshakeIXPSK0,
+	)
+	require.NoError(t, err)
+
+	msg1, err := initM.Initiate(nil)
+	require.NoError(t, err)
+
+	resp, respR, err := respM.ProcessPacket(nil, msg1)
+	require.NoError(t, err)
+	require.NotNil(t, respR)
+
+	_, initR, err = initM.ProcessPacket(nil, resp)
+	require.NoError(t, err)
+	require.NotNil(t, initR)
+
+	return initR, respR
+}
+
+func TestNewConnectionStateFromResult(t *testing.T) {
+	initR, respR := runTestHandshake(t)
+
+	t.Run("initiator", func(t *testing.T) {
+		ci := newConnectionStateFromResult(initR)
+		assert.True(t, ci.initiator)
+		assert.Equal(t, initR.MyCert, ci.myCert)
+		assert.Equal(t, initR.RemoteCert, ci.peerCert)
+		assert.NotNil(t, ci.eKey)
+		assert.NotNil(t, ci.dKey)
+
+		// IX has 2 handshake messages; the next data-plane send is counter=3.
+		assert.Equal(t, uint64(2), ci.messageCounter.Load(),
+			"messageCounter must equal Result.MessageIndex so the next send is N+1")
+
+		// Both handshake counters must be marked seen so they don't appear lost.
+		// Check returns false if an index has already been recorded.
+		assert.False(t, ci.window.Check(nil, 1), "counter 1 must already be seen")
+		assert.False(t, ci.window.Check(nil, 2), "counter 2 must already be seen")
+		// Counter 3 is the next data-plane message and must NOT be pre-marked.
+		assert.True(t, ci.window.Check(nil, 3), "counter 3 must not be pre-seeded")
+	})
+
+	t.Run("responder", func(t *testing.T) {
+		ci := newConnectionStateFromResult(respR)
+		assert.False(t, ci.initiator)
+		assert.Equal(t, respR.MyCert, ci.myCert)
+		assert.Equal(t, respR.RemoteCert, ci.peerCert)
+		assert.NotNil(t, ci.eKey)
+		assert.NotNil(t, ci.dKey)
+		assert.Equal(t, uint64(2), ci.messageCounter.Load())
+	})
+}

control.go

@@ -2,17 +2,33 @@ package nebula
 
 import (
 	"context"
+	"errors"
+	"log/slog"
 	"net/netip"
 	"os"
 	"os/signal"
+	"sync"
 	"syscall"
 
-	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula/cert"
 	"github.com/slackhq/nebula/header"
 	"github.com/slackhq/nebula/overlay"
 )
 
+type RunState int
+
+const (
+	StateUnknown RunState = iota
+	StateReady
+	StateStarted
+	StateStopping
+	StateStopped
+)
+
+var ErrAlreadyStarted = errors.New("nebula is already started")
+var ErrAlreadyStopped = errors.New("nebula cannot be restarted")
+var ErrUnknownState = errors.New("nebula state is invalid")
+
 // Every interaction here needs to take extra care to copy memory and not return or use arguments "as is" when touching
 // core. This means copying IP objects, slices, de-referencing pointers and taking the actual value, etc
 
@@ -26,8 +42,11 @@ type controlHostLister interface {
 }
 
 type Control struct {
+	stateLock sync.Mutex
+	state     RunState
+
 	f                      *Interface
-	l                      *logrus.Logger
+	l                      *slog.Logger
 	ctx                    context.Context
 	cancel                 context.CancelFunc
 	sshStart               func()
@@ -35,6 +54,7 @@ type Control struct {
 	dnsStart               func()
 	lighthouseStart        func()
 	connectionManagerStart func(context.Context)
+	pmtudManagerStart      func(context.Context)
 }
 
 type ControlHostInfo struct {
@@ -49,10 +69,31 @@ type ControlHostInfo struct {
 	CurrentRelaysThroughMe []netip.Addr     `json:"currentRelaysThroughMe"`
 }
 
-// Start actually runs nebula, this is a nonblocking call. To block use Control.ShutdownBlock()
+// Start actually runs nebula, this is a nonblocking call.
-func (c *Control) Start() {
+// The returned function blocks until nebula has fully stopped and returns the
+// first fatal reader error (if any). A nil error means nebula shut down
+// gracefully; a non-nil error means a reader hit an unexpected failure that
+// triggered the shutdown.
+func (c *Control) Start() (func() error, error) {
+	c.stateLock.Lock()
+	defer c.stateLock.Unlock()
+	switch c.state {
+	case StateReady:
+		//yay!
+	case StateStopped, StateStopping:
+		return nil, ErrAlreadyStopped
+	case StateStarted:
+		return nil, ErrAlreadyStarted
+	default:
+		return nil, ErrUnknownState
+	}
+
 	// Activate the interface
-	c.f.activate()
+	err := c.f.activate()
+	if err != nil {
+		c.state = StateStopped
+		return nil, err
+	}
 
 	// Call all the delayed funcs that waited patiently for the interface to be created.
 	if c.sshStart != nil {
@@ -67,29 +108,58 @@ func (c *Control) Start() {
 	if c.connectionManagerStart != nil {
 		go c.connectionManagerStart(c.ctx)
 	}
+	if c.pmtudManagerStart != nil {
+		go c.pmtudManagerStart(c.ctx)
+	}
 	if c.lighthouseStart != nil {
 		c.lighthouseStart()
 	}
 
+	c.f.triggerShutdown = c.Stop
+
 	// Start reading packets.
-	c.f.run()
+	out, err := c.f.run()
+	if err != nil {
+		c.state = StateStopped
+		return nil, err
+	}
+	c.state = StateStarted
+	return out, nil
+}
+
+func (c *Control) State() RunState {
+	c.stateLock.Lock()
+	defer c.stateLock.Unlock()
+	return c.state
 }
 
 func (c *Control) Context() context.Context {
 	return c.ctx
 }
 
-// Stop signals nebula to shutdown and close all tunnels, returns after the shutdown is complete
+// Stop is a non-blocking call that signals nebula to close all tunnels and shut down
 func (c *Control) Stop() {
+	c.stateLock.Lock()
+	if c.state != StateStarted {
+		c.stateLock.Unlock()
+		// We are stopping or stopped already
+		return
+	}
+
+	c.state = StateStopping
+	c.stateLock.Unlock()
+
 	// Stop the handshakeManager (and other services), to prevent new tunnels from
 	// being created while we're shutting them all down.
 	c.cancel()
 
 	c.CloseAllTunnels(false)
 	if err := c.f.Close(); err != nil {
-		c.l.WithError(err).Error("Close interface failed")
+		c.l.Error("Close interface failed", "error", err)
 	}
-	c.l.Info("Goodbye")
+	c.stateLock.Lock()
+	c.state = StateStopped
+	c.stateLock.Unlock()
 }
 
 // ShutdownBlock will listen for and block on term and interrupt signals, calling Control.Stop() once signalled
@@ -100,7 +170,7 @@ func (c *Control) ShutdownBlock() {
 
 	rawSig := <-sigChan
 	sig := rawSig.String()
-	c.l.WithField("signal", sig).Info("Caught signal, shutting down")
+	c.l.Info("Caught signal, shutting down", "signal", sig)
 	c.Stop()
 }
 
@@ -237,8 +307,10 @@ func (c *Control) CloseAllTunnels(excludeLighthouses bool) (closed int) {
 		c.f.send(header.CloseTunnel, 0, h.ConnectionState, h, []byte{}, make([]byte, 12, 12), make([]byte, mtu))
 		c.f.closeTunnel(h)
 
-		c.l.WithField("vpnAddrs", h.vpnAddrs).WithField("udpAddr", h.remote).
+		c.l.Debug("Sending close tunnel message",
-			Debug("Sending close tunnel message")
+			"vpnAddrs", h.vpnAddrs,
+			"udpAddr", h.remote,
+		)
 		closed++
 	}
 

control_test.go

@@ -6,7 +6,6 @@ import (
 	"reflect"
 	"testing"
 
-	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula/cert"
 	"github.com/slackhq/nebula/test"
 	"github.com/stretchr/testify/assert"
@@ -79,10 +78,11 @@ func TestControl_GetHostInfoByVpnIp(t *testing.T) {
 	}, &Interface{})
 
 	c := Control{
+		state: StateReady,
 		f: &Interface{
 			hostMap: hm,
 		},
-		l: logrus.New(),
+		l: test.NewLogger(),
 	}
 
 	thi := c.GetHostInfoByVpnAddr(vpnIp, false)

control_tester.go

@@ -1,5 +1,4 @@
 //go:build e2e_testing
-// +build e2e_testing
 
 package nebula
 
@@ -174,6 +173,10 @@ func (c *Control) GetHostmap() *HostMap {
 	return c.f.hostMap
 }
 
+func (c *Control) GetF() *Interface {
+	return c.f
+}
+
 func (c *Control) GetCertState() *CertState {
 	return c.f.pki.getCertState()
 }

dist/wireshark/nebula.lua

@@ -84,30 +84,24 @@ end
 
 function nebula.prefs_changed()
     if default_settings.all_ports == nebula.prefs.all_ports and default_settings.port == nebula.prefs.port then
-        -- Nothing changed, bail
         return
     end
 
-    -- Remove our old dissector
+    -- Remove all existing registrations
     DissectorTable.get("udp.port"):remove_all(nebula)
 
-    if nebula.prefs.all_ports and default_settings.all_ports ~= nebula.prefs.all_ports then
+    if nebula.prefs.all_ports then
-        default_settings.all_port = nebula.prefs.all_ports
+        -- Register on every port for hole punch capture
-
         for i=0, 65535 do
             DissectorTable.get("udp.port"):add(i, nebula)
         end
-
+    else
-        -- no need to establish again on specific ports
+        -- Register on the configured port only
-        return
+        DissectorTable.get("udp.port"):add(nebula.prefs.port, nebula)
     end
 
-
+    default_settings.all_ports = nebula.prefs.all_ports
-    if default_settings.all_ports ~= nebula.prefs.all_ports then
-        -- Add our new port dissector
     default_settings.port = nebula.prefs.port
-        DissectorTable.get("udp.port"):add(default_settings.port, nebula)
-    end
 end
 
 DissectorTable.get("udp.port"):add(default_settings.port, nebula)

dns_server.go

@@ -1,63 +1,249 @@
 package nebula
 
 import (
+	"context"
 	"fmt"
+	"log/slog"
 	"net"
 	"net/netip"
 	"strconv"
 	"strings"
 	"sync"
+	"sync/atomic"
 
 	"github.com/gaissmai/bart"
 	"github.com/miekg/dns"
-	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula/config"
 )
 
-// This whole thing should be rewritten to use context
+type dnsServer struct {
-
-var dnsR *dnsRecords
-var dnsServer *dns.Server
-var dnsAddr string
-
-type dnsRecords struct {
 	sync.RWMutex
-	l               *logrus.Logger
+	l               *slog.Logger
+	ctx             context.Context
 	dnsMap4         map[string]netip.Addr
 	dnsMap6         map[string]netip.Addr
 	hostMap         *HostMap
 	myVpnAddrsTable *bart.Lite
+
+	mux *dns.ServeMux
+
+	// enabled mirrors `lighthouse.serve_dns && lighthouse.am_lighthouse`.
+	// Start, Add, and reload consult it so callers don't need to know the
+	// gating rules. When it toggles off via reload, accumulated records are
+	// cleared so a later re-enable starts with a fresh map populated from
+	// new handshakes.
+	enabled atomic.Bool
+
+	serverMu sync.Mutex
+	server   *dns.Server
+	// started is closed once `server` has finished binding (or after
+	// ListenAndServe returns on a bind failure). Stop waits on it before
+	// calling Shutdown to avoid the miekg/dns "server not started" race
+	// where a Shutdown that arrives before bind completes is silently
+	// ignored, leaving the listener running forever.
+	started chan struct{}
+	addr    string
 }
 
-func newDnsRecords(l *logrus.Logger, cs *CertState, hostMap *HostMap) *dnsRecords {
+// newDnsServerFromConfig builds a dnsServer, applies the initial config, and
-	return &dnsRecords{
+// registers a reload callback. The reload callback is registered before the
+// initial config is applied, so a SIGHUP can later enable, fix, or disable
+// DNS even if the initial application failed.
+//
+// The dnsServer internally gates on `lighthouse.serve_dns &&
+// lighthouse.am_lighthouse`. Start and Add are safe to call unconditionally,
+// they no-op when DNS isn't enabled. Each Start invocation owns a ctx-cancel
+// watcher that tears the listener down on nebula shutdown. The returned
+// pointer is always non-nil, even on error.
+func newDnsServerFromConfig(ctx context.Context, l *slog.Logger, cs *CertState, hostMap *HostMap, c *config.C) (*dnsServer, error) {
+	ds := &dnsServer{
 		l:               l,
+		ctx:             ctx,
 		dnsMap4:         make(map[string]netip.Addr),
 		dnsMap6:         make(map[string]netip.Addr),
 		hostMap:         hostMap,
 		myVpnAddrsTable: cs.myVpnAddrsTable,
 	}
+	ds.mux = dns.NewServeMux()
+	ds.mux.HandleFunc(".", ds.handleDnsRequest)
+
+	c.RegisterReloadCallback(func(c *config.C) {
+		if err := ds.reload(c, false); err != nil {
+			ds.l.Error("Failed to reload DNS responder from config", "error", err)
+		}
+	})
+
+	if err := ds.reload(c, true); err != nil {
+		return ds, err
+	}
+	return ds, nil
 }
 
-func (d *dnsRecords) Query(q uint16, data string) netip.Addr {
+// reload applies the latest config and reconciles the running state with it:
+//   - enabled toggled on  -> spawn a runner
+//   - enabled toggled off -> stop the runner
+//   - listen address changed (while running) -> restart on the new address
+//   - everything else     -> no-op
+//
+// On the initial call it only records configuration; Control.Start is what
+// launches the first runner via dnsStart.
+func (d *dnsServer) reload(c *config.C, initial bool) error {
+	wantsDns := c.GetBool("lighthouse.serve_dns", false)
+	amLighthouse := c.GetBool("lighthouse.am_lighthouse", false)
+	enabled := wantsDns && amLighthouse
+	newAddr := getDnsServerAddr(c)
+
+	d.serverMu.Lock()
+	running := d.server
+	runningStarted := d.started
+	sameAddr := d.addr == newAddr
+	d.addr = newAddr
+	d.enabled.Store(enabled)
+	d.serverMu.Unlock()
+
+	if initial {
+		if wantsDns && !amLighthouse {
+			d.l.Warn("DNS server refusing to run because this host is not a lighthouse.")
+		}
+		return nil
+	}
+
+	if !enabled {
+		if running != nil {
+			d.Stop()
+		}
+		// Drop any records that accumulated while enabled; a later re-enable
+		// will repopulate from fresh handshakes.
+		d.clearRecords()
+		return nil
+	}
+
+	if running == nil {
+		// Was disabled (or never started); bring it up now.
+		go d.Start()
+		return nil
+	}
+
+	if sameAddr {
+		return nil
+	}
+
+	d.shutdownServer(running, runningStarted, "reload")
+	// Old Start goroutine has now exited; bring up a fresh listener on the
+	// new address.
+	go d.Start()
+	return nil
+}
+
+// shutdownServer waits for the server to finish binding (so Shutdown actually
+// stops it rather than no-oping) and then shuts it down.
+func (d *dnsServer) shutdownServer(srv *dns.Server, started chan struct{}, reason string) {
+	if srv == nil {
+		return
+	}
+	if started != nil {
+		<-started
+	}
+	if err := srv.Shutdown(); err != nil {
+		d.l.Warn("Failed to shut down the DNS responder", "reason", reason, "error", err)
+	}
+}
+
+// Start binds and serves the DNS responder. Blocks until Stop is called or
+// the listener errors. Safe to call when DNS is disabled (returns
+// immediately). This is what Control.dnsStart points at.
+//
+// Must be invoked after the tun device is active so that lighthouse.dns.host
+// may bind to a nebula IP.
+func (d *dnsServer) Start() {
+	if !d.enabled.Load() {
+		return
+	}
+
+	started := make(chan struct{})
+	d.serverMu.Lock()
+	if d.ctx.Err() != nil {
+		d.serverMu.Unlock()
+		return
+	}
+	addr := d.addr
+	server := &dns.Server{
+		Addr:              addr,
+		Net:               "udp",
+		Handler:           d.mux,
+		NotifyStartedFunc: func() { close(started) },
+	}
+	d.server = server
+	d.started = started
+	d.serverMu.Unlock()
+
+	// Per-invocation ctx watcher. Exits when Start does, so we don't leak a
+	// watcher per reload-driven restart.
+	done := make(chan struct{})
+	go func() {
+		select {
+		case <-d.ctx.Done():
+			d.shutdownServer(server, started, "shutdown")
+		case <-done:
+		}
+	}()
+
+	d.l.Info("Starting DNS responder", "dnsListener", addr)
+	err := server.ListenAndServe()
+	close(done)
+
+	// If the listener never bound (bind error) NotifyStartedFunc never fires,
+	// so close started here to release any Stop caller waiting on it.
+	select {
+	case <-started:
+	default:
+		close(started)
+	}
+
+	if err != nil {
+		d.l.Warn("Failed to run the DNS responder", "error", err)
+	}
+}
+
+// Stop shuts down the active server, if any. Idempotent.
+func (d *dnsServer) Stop() {
+	d.serverMu.Lock()
+	srv := d.server
+	started := d.started
+	d.server = nil
+	d.started = nil
+	d.serverMu.Unlock()
+	d.shutdownServer(srv, started, "stop")
+}
+
+// Query returns the address for the given name and query type. The second
+// return value reports whether the name is known at all (in either A or AAAA),
+// which lets callers distinguish NODATA from NXDOMAIN.
+func (d *dnsServer) Query(q uint16, data string) (netip.Addr, bool) {
 	data = strings.ToLower(data)
 	d.RLock()
 	defer d.RUnlock()
+	addr4, haveV4 := d.dnsMap4[data]
+	addr6, haveV6 := d.dnsMap6[data]
+	nameExists := haveV4 || haveV6
 	switch q {
 	case dns.TypeA:
-		if r, ok := d.dnsMap4[data]; ok {
+		if haveV4 {
-			return r
+			return addr4, nameExists
 		}
 	case dns.TypeAAAA:
-		if r, ok := d.dnsMap6[data]; ok {
+		if haveV6 {
-			return r
+			return addr6, nameExists
 		}
 	}
 
-	return netip.Addr{}
+	return netip.Addr{}, nameExists
 }
 
-func (d *dnsRecords) QueryCert(data string) string {
+func (d *dnsServer) QueryCert(data string) string {
+	if len(data) < 2 {
+		return ""
+	}
 	ip, err := netip.ParseAddr(data[:len(data)-1])
 	if err != nil {
 		return ""
@@ -80,8 +266,19 @@ func (d *dnsRecords) QueryCert(data string) string {
 	return string(b)
 }
 
+// clearRecords drops all DNS records.
+func (d *dnsServer) clearRecords() {
+	d.Lock()
+	defer d.Unlock()
+	clear(d.dnsMap4)
+	clear(d.dnsMap6)
+}
+
 // Add adds the first IPv4 and IPv6 address that appears in `addresses` as the record for `host`
-func (d *dnsRecords) Add(host string, addresses []netip.Addr) {
+func (d *dnsServer) Add(host string, addresses []netip.Addr) {
+	if !d.enabled.Load() {
+		return
+	}
 	host = strings.ToLower(host)
 	d.Lock()
 	defer d.Unlock()
@@ -101,7 +298,7 @@ func (d *dnsRecords) Add(host string, addresses []netip.Addr) {
 	}
 }
 
-func (d *dnsRecords) isSelfNebulaOrLocalhost(addr string) bool {
+func (d *dnsServer) isSelfNebulaOrLocalhost(addr string) bool {
 	a, _, _ := net.SplitHostPort(addr)
 	b, err := netip.ParseAddr(a)
 	if err != nil {
@@ -116,13 +313,24 @@ func (d *dnsRecords) isSelfNebulaOrLocalhost(addr string) bool {
 	return d.myVpnAddrsTable.Contains(b)
 }
 
-func (d *dnsRecords) parseQuery(m *dns.Msg, w dns.ResponseWriter) {
+func (d *dnsServer) parseQuery(m *dns.Msg, w dns.ResponseWriter) {
+	debugEnabled := d.l.Enabled(context.Background(), slog.LevelDebug)
+	// Per RFC 2308 §2.2, a name that exists but has no record of the requested
+	// type must be answered with NOERROR and an empty answer section (NODATA),
+	// not NXDOMAIN (RFC 2308 §2.1), which is reserved for names that do not
+	// exist at all.
+	anyNameExists := false
 	for _, q := range m.Question {
 		switch q.Qtype {
 		case dns.TypeA, dns.TypeAAAA:
 			qType := dns.TypeToString[q.Qtype]
-			d.l.Debugf("Query for %s %s", qType, q.Name)
+			if debugEnabled {
-			ip := d.Query(q.Qtype, q.Name)
+				d.l.Debug("DNS query", "type", qType, "name", q.Name)
+			}
+			ip, nameExists := d.Query(q.Qtype, q.Name)
+			if nameExists {
+				anyNameExists = true
+			}
 			if ip.IsValid() {
 				rr, err := dns.NewRR(fmt.Sprintf("%s %s %s", q.Name, qType, ip))
 				if err == nil {
@@ -134,7 +342,9 @@ func (d *dnsRecords) parseQuery(m *dns.Msg, w dns.ResponseWriter) {
 			if !d.isSelfNebulaOrLocalhost(w.RemoteAddr().String()) {
 				return
 			}
-			d.l.Debugf("Query for TXT %s", q.Name)
+			if debugEnabled {
+				d.l.Debug("DNS query", "type", "TXT", "name", q.Name)
+			}
 			ip := d.QueryCert(q.Name)
 			if ip != "" {
 				rr, err := dns.NewRR(fmt.Sprintf("%s TXT %s", q.Name, ip))
@@ -145,12 +355,12 @@ func (d *dnsRecords) parseQuery(m *dns.Msg, w dns.ResponseWriter) {
 		}
 	}
 
-	if len(m.Answer) == 0 {
+	if len(m.Answer) == 0 && !anyNameExists {
 		m.Rcode = dns.RcodeNameError
 	}
 }
 
-func (d *dnsRecords) handleDnsRequest(w dns.ResponseWriter, r *dns.Msg) {
+func (d *dnsServer) handleDnsRequest(w dns.ResponseWriter, r *dns.Msg) {
 	m := new(dns.Msg)
 	m.SetReply(r)
 	m.Compress = false
@@ -163,21 +373,6 @@ func (d *dnsRecords) handleDnsRequest(w dns.ResponseWriter, r *dns.Msg) {
 	w.WriteMsg(m)
 }
 
-func dnsMain(l *logrus.Logger, cs *CertState, hostMap *HostMap, c *config.C) func() {
-	dnsR = newDnsRecords(l, cs, hostMap)
-
-	// attach request handler func
-	dns.HandleFunc(".", dnsR.handleDnsRequest)
-
-	c.RegisterReloadCallback(func(c *config.C) {
-		reloadDns(l, c)
-	})
-
-	return func() {
-		startDns(l, c)
-	}
-}
-
 func getDnsServerAddr(c *config.C) string {
 	dnsHost := strings.TrimSpace(c.GetString("lighthouse.dns.host", ""))
 	// Old guidance was to provide the literal `[::]` in `lighthouse.dns.host` but that won't resolve.
@@ -186,25 +381,3 @@ func getDnsServerAddr(c *config.C) string {
 	}
 	return net.JoinHostPort(dnsHost, strconv.Itoa(c.GetInt("lighthouse.dns.port", 53)))
 }
-
-func startDns(l *logrus.Logger, c *config.C) {
-	dnsAddr = getDnsServerAddr(c)
-	dnsServer = &dns.Server{Addr: dnsAddr, Net: "udp"}
-	l.WithField("dnsListener", dnsAddr).Info("Starting DNS responder")
-	err := dnsServer.ListenAndServe()
-	defer dnsServer.Shutdown()
-	if err != nil {
-		l.Errorf("Failed to start server: %s\n ", err.Error())
-	}
-}
-
-func reloadDns(l *logrus.Logger, c *config.C) {
-	if dnsAddr == getDnsServerAddr(c) {
-		l.Debug("No DNS server config change detected")
-		return
-	}
-
-	l.Debug("Restarting DNS server")
-	dnsServer.Shutdown()
-	go startDns(l, c)
-}

dns_server_test.go

@@ -1,19 +1,43 @@
 package nebula
 
 import (
+	"context"
+	"log/slog"
+	"net"
 	"net/netip"
+	"strconv"
 	"testing"
+	"time"
 
 	"github.com/miekg/dns"
-	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula/config"
 	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
 )
 
+type stubDNSWriter struct{}
+
+func (stubDNSWriter) LocalAddr() net.Addr { return &net.UDPAddr{} }
+func (stubDNSWriter) RemoteAddr() net.Addr {
+	return &net.UDPAddr{IP: net.ParseIP("127.0.0.1"), Port: 5353}
+}
+func (stubDNSWriter) Write([]byte) (int, error) { return 0, nil }
+func (stubDNSWriter) WriteMsg(*dns.Msg) error   { return nil }
+func (stubDNSWriter) Close() error              { return nil }
+func (stubDNSWriter) TsigStatus() error         { return nil }
+func (stubDNSWriter) TsigTimersOnly(bool)       {}
+func (stubDNSWriter) Hijack()                   {}
+
 func TestParsequery(t *testing.T) {
-	l := logrus.New()
+	l := slog.New(slog.DiscardHandler)
 	hostMap := &HostMap{}
-	ds := newDnsRecords(l, &CertState{}, hostMap)
+	ds := &dnsServer{
+		l:       l,
+		dnsMap4: make(map[string]netip.Addr),
+		dnsMap6: make(map[string]netip.Addr),
+		hostMap: hostMap,
+	}
+	ds.enabled.Store(true)
 	addrs := []netip.Addr{
 		netip.MustParseAddr("1.2.3.4"),
 		netip.MustParseAddr("1.2.3.5"),
@@ -21,18 +45,56 @@ func TestParsequery(t *testing.T) {
 		netip.MustParseAddr("fd01::25"),
 	}
 	ds.Add("test.com.com", addrs)
+	ds.Add("v4only.com.com", []netip.Addr{netip.MustParseAddr("1.2.3.6")})
+	ds.Add("v6only.com.com", []netip.Addr{netip.MustParseAddr("fd01::26")})
 
 	m := &dns.Msg{}
 	m.SetQuestion("test.com.com", dns.TypeA)
 	ds.parseQuery(m, nil)
 	assert.NotNil(t, m.Answer)
 	assert.Equal(t, "1.2.3.4", m.Answer[0].(*dns.A).A.String())
+	assert.Equal(t, dns.RcodeSuccess, m.Rcode)
 
 	m = &dns.Msg{}
 	m.SetQuestion("test.com.com", dns.TypeAAAA)
 	ds.parseQuery(m, nil)
 	assert.NotNil(t, m.Answer)
 	assert.Equal(t, "fd01::24", m.Answer[0].(*dns.AAAA).AAAA.String())
+	assert.Equal(t, dns.RcodeSuccess, m.Rcode)
+
+	// A known name with no record of the requested type should return NODATA
+	// (NOERROR with empty answer), not NXDOMAIN.
+	m = &dns.Msg{}
+	m.SetQuestion("v4only.com.com", dns.TypeAAAA)
+	ds.parseQuery(m, nil)
+	assert.Empty(t, m.Answer)
+	assert.Equal(t, dns.RcodeSuccess, m.Rcode)
+
+	m = &dns.Msg{}
+	m.SetQuestion("v6only.com.com", dns.TypeA)
+	ds.parseQuery(m, nil)
+	assert.Empty(t, m.Answer)
+	assert.Equal(t, dns.RcodeSuccess, m.Rcode)
+
+	// An unknown name should still return NXDOMAIN.
+	m = &dns.Msg{}
+	m.SetQuestion("unknown.com.com", dns.TypeA)
+	ds.parseQuery(m, nil)
+	assert.Empty(t, m.Answer)
+	assert.Equal(t, dns.RcodeNameError, m.Rcode)
+
+	// short lookups should not fail
+	m = &dns.Msg{}
+	m.Question = []dns.Question{{Name: "", Qtype: dns.TypeTXT, Qclass: dns.ClassINET}}
+	ds.parseQuery(m, stubDNSWriter{})
+	assert.Empty(t, m.Answer)
+	assert.Equal(t, dns.RcodeNameError, m.Rcode)
+
+	m = &dns.Msg{}
+	m.Question = []dns.Question{{Name: ".", Qtype: dns.TypeTXT, Qclass: dns.ClassINET}}
+	ds.parseQuery(m, stubDNSWriter{})
+	assert.Empty(t, m.Answer)
+	assert.Equal(t, dns.RcodeNameError, m.Rcode)
 }
 
 func Test_getDnsServerAddr(t *testing.T) {
@@ -71,3 +133,208 @@ func Test_getDnsServerAddr(t *testing.T) {
 	}
 	assert.Equal(t, "[::]:1", getDnsServerAddr(c))
 }
+
+func newTestDnsServer(t *testing.T) (*dnsServer, *config.C) {
+	t.Helper()
+	sl := slog.New(slog.DiscardHandler)
+	ds := &dnsServer{
+		l:       sl,
+		ctx:     context.Background(),
+		dnsMap4: make(map[string]netip.Addr),
+		dnsMap6: make(map[string]netip.Addr),
+		hostMap: &HostMap{},
+	}
+	ds.mux = dns.NewServeMux()
+	ds.mux.HandleFunc(".", ds.handleDnsRequest)
+	return ds, config.NewC(nil)
+}
+
+func setDnsConfig(c *config.C, host string, port string, amLighthouse, serveDns bool) {
+	c.Settings["lighthouse"] = map[string]any{
+		"am_lighthouse": amLighthouse,
+		"serve_dns":     serveDns,
+		"dns": map[string]any{
+			"host": host,
+			"port": port,
+		},
+	}
+}
+
+func TestDnsServer_reload_initial_disabled(t *testing.T) {
+	ds, c := newTestDnsServer(t)
+	setDnsConfig(c, "127.0.0.1", "0", true, false)
+
+	require.NoError(t, ds.reload(c, true))
+	assert.False(t, ds.enabled.Load())
+	assert.Equal(t, "127.0.0.1:0", ds.addr)
+	assert.Nil(t, ds.server)
+}
+
+func TestDnsServer_reload_initial_enabled(t *testing.T) {
+	ds, c := newTestDnsServer(t)
+	setDnsConfig(c, "127.0.0.1", "0", true, true)
+
+	require.NoError(t, ds.reload(c, true))
+	assert.True(t, ds.enabled.Load())
+	assert.Equal(t, "127.0.0.1:0", ds.addr)
+	// initial never starts a runner; that's Control.Start's job
+	assert.Nil(t, ds.server)
+}
+
+func TestDnsServer_reload_initial_serveDnsWithoutLighthouse(t *testing.T) {
+	ds, c := newTestDnsServer(t)
+	setDnsConfig(c, "127.0.0.1", "0", false, true)
+
+	require.NoError(t, ds.reload(c, true))
+	// Wants DNS but isn't a lighthouse: gated off, no runner.
+	assert.False(t, ds.enabled.Load())
+}
+
+func TestDnsServer_reload_sameAddr_noOp(t *testing.T) {
+	ds, c := newTestDnsServer(t)
+	setDnsConfig(c, "127.0.0.1", "0", true, true)
+
+	require.NoError(t, ds.reload(c, true))
+	// No server running yet, no addr change. Reload should not spawn anything.
+	require.NoError(t, ds.reload(c, false))
+	assert.True(t, ds.enabled.Load())
+	assert.Nil(t, ds.server)
+}
+
+func TestDnsServer_StartStop_lifecycle(t *testing.T) {
+	// Bind to a real (random) UDP port so we exercise the actual
+	// ListenAndServe + Shutdown plumbing including the started-chan race fix.
+	port := freeUDPPort(t)
+
+	ds, c := newTestDnsServer(t)
+	setDnsConfig(c, "127.0.0.1", port, true, true)
+	require.NoError(t, ds.reload(c, true))
+
+	done := make(chan struct{})
+	go func() {
+		ds.Start()
+		close(done)
+	}()
+
+	waitFor(t, func() bool {
+		ds.serverMu.Lock()
+		started := ds.started
+		ds.serverMu.Unlock()
+		if started == nil {
+			return false
+		}
+		select {
+		case <-started:
+			return true
+		default:
+			return false
+		}
+	})
+
+	ds.Stop()
+	select {
+	case <-done:
+	case <-time.After(5 * time.Second):
+		t.Fatal("Start did not return after Stop")
+	}
+}
+
+func TestDnsServer_Stop_beforeBind_doesNotHang(t *testing.T) {
+	// Stop called immediately after Start should not deadlock even if bind
+	// hasn't completed yet. This exercises the started-chan close-on-bind-fail
+	// path: by binding to an obviously bad port (privileged) we get a fast
+	// bind error before NotifyStartedFunc fires.
+	ds, c := newTestDnsServer(t)
+	// Use a port that should fail to bind (negative would be invalid, use a
+	// host that won't resolve to ensure listenUDP fails quickly).
+	setDnsConfig(c, "256.256.256.256", "53", true, true)
+	require.NoError(t, ds.reload(c, true))
+
+	done := make(chan struct{})
+	go func() {
+		ds.Start()
+		close(done)
+	}()
+
+	// Give Start a moment to attempt the bind and fail.
+	select {
+	case <-done:
+		// Bind failed and Start returned; Stop should be a no-op.
+	case <-time.After(time.Second):
+		t.Fatal("Start did not return after a bad bind")
+	}
+
+	stopped := make(chan struct{})
+	go func() {
+		ds.Stop()
+		close(stopped)
+	}()
+	select {
+	case <-stopped:
+	case <-time.After(time.Second):
+		t.Fatal("Stop hung after a failed bind")
+	}
+}
+
+func TestDnsServer_reload_disable_stopsRunningServer(t *testing.T) {
+	port := freeUDPPort(t)
+	ds, c := newTestDnsServer(t)
+	setDnsConfig(c, "127.0.0.1", port, true, true)
+	require.NoError(t, ds.reload(c, true))
+
+	startReturned := make(chan struct{})
+	go func() {
+		ds.Start()
+		close(startReturned)
+	}()
+	waitForBind(t, ds)
+
+	// Toggle serve_dns off; reload should shut the running server down.
+	setDnsConfig(c, "127.0.0.1", port, true, false)
+	require.NoError(t, ds.reload(c, false))
+	select {
+	case <-startReturned:
+	case <-time.After(5 * time.Second):
+		t.Fatal("Start did not return after reload disabled DNS")
+	}
+	assert.False(t, ds.enabled.Load())
+}
+
+func freeUDPPort(t *testing.T) string {
+	t.Helper()
+	conn, err := net.ListenPacket("udp", "127.0.0.1:0")
+	require.NoError(t, err)
+	port := conn.LocalAddr().(*net.UDPAddr).Port
+	require.NoError(t, conn.Close())
+	return strconv.Itoa(port)
+}
+
+func waitForBind(t *testing.T, ds *dnsServer) {
+	t.Helper()
+	waitFor(t, func() bool {
+		ds.serverMu.Lock()
+		started := ds.started
+		ds.serverMu.Unlock()
+		if started == nil {
+			return false
+		}
+		select {
+		case <-started:
+			return true
+		default:
+			return false
+		}
+	})
+}
+
+func waitFor(t *testing.T, cond func() bool) {
+	t.Helper()
+	deadline := time.Now().Add(5 * time.Second)
+	for time.Now().Before(deadline) {
+		if cond() {
+			return
+		}
+		time.Sleep(5 * time.Millisecond)
+	}
+	t.Fatal("timed out waiting for condition")
+}

e2e/handshake_manager_test.go

@@ -0,0 +1,577 @@
+//go:build e2e_testing
+// +build e2e_testing
+
+package e2e
+
+import (
+	"net/netip"
+	"testing"
+	"time"
+
+	"github.com/slackhq/nebula"
+	"github.com/slackhq/nebula/cert"
+	"github.com/slackhq/nebula/cert_test"
+	"github.com/slackhq/nebula/e2e/router"
+	"github.com/slackhq/nebula/header"
+	"github.com/slackhq/nebula/udp"
+	"github.com/stretchr/testify/assert"
+)
+
+// makeHandshakePacket creates a handshake packet with the given parameters.
+func makeHandshakePacket(from, to netip.AddrPort, subtype header.MessageSubType, remoteIndex uint32, counter uint64) *udp.Packet {
+	data := make([]byte, 200)
+	header.Encode(data, header.Version, header.Handshake, subtype, remoteIndex, counter)
+	for i := header.Len; i < len(data); i++ {
+		data[i] = byte(i)
+	}
+	return &udp.Packet{To: to, From: from, Data: data}
+}
+
+func TestHandshakeRetransmitDuplicate(t *testing.T) {
+	t.Parallel()
+	// Verify the responder correctly handles receiving the same msg1 multiple times
+	// (retransmission). The duplicate goes through CheckAndComplete -> ErrAlreadySeen
+	// and the cached response is resent.
+
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	myControl, myVpnIpNet, myUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "me", "10.128.0.1/24", nil)
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "them", "10.128.0.2/24", nil)
+
+	myControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
+	theirControl.InjectLightHouseAddr(myVpnIpNet[0].Addr(), myUdpAddr)
+
+	myControl.Start()
+	theirControl.Start()
+
+	r := router.NewR(t, myControl, theirControl)
+	defer r.RenderFlow()
+
+	t.Log("Trigger handshake from me to them")
+	myControl.InjectTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi"))
+
+	t.Log("Grab my msg1")
+	msg1 := myControl.GetFromUDP(true)
+
+	t.Log("Inject msg1 into them, first time")
+	theirControl.InjectUDPPacket(msg1)
+	_ = theirControl.GetFromUDP(true)
+
+	t.Log("Inject the SAME msg1 again, tests ErrAlreadySeen path")
+	theirControl.InjectUDPPacket(msg1)
+	resp2 := theirControl.GetFromUDP(true)
+	assert.NotNil(t, resp2, "should get cached response on duplicate msg1")
+
+	t.Log("Complete handshake with cached response")
+	myControl.InjectUDPPacket(resp2)
+	myControl.WaitForType(1, 0, theirControl)
+
+	t.Log("Drain cached packet and verify tunnel works")
+	cachedPacket := theirControl.GetFromTun(true)
+	assertUdpPacket(t, []byte("Hi"), cachedPacket, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), 80, 80)
+	assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
+
+	t.Log("Verify only one tunnel exists on each side")
+	assert.Len(t, myControl.ListHostmapHosts(false), 1)
+	assert.Len(t, theirControl.ListHostmapHosts(false), 1)
+
+	myControl.Stop()
+	theirControl.Stop()
+}
+
+func TestHandshakeTruncatedPacketRecovery(t *testing.T) {
+	t.Parallel()
+	// Verify that a truncated handshake packet is ignored and the real
+	// packet can still complete the handshake.
+
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	myControl, myVpnIpNet, myUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "me", "10.128.0.1/24", nil)
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "them", "10.128.0.2/24", nil)
+
+	myControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
+	theirControl.InjectLightHouseAddr(myVpnIpNet[0].Addr(), myUdpAddr)
+
+	myControl.Start()
+	theirControl.Start()
+
+	r := router.NewR(t, myControl, theirControl)
+	defer r.RenderFlow()
+
+	t.Log("Trigger handshake")
+	myControl.InjectTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi"))
+
+	t.Log("Get msg1 and deliver to responder")
+	msg1 := myControl.GetFromUDP(true)
+	theirControl.InjectUDPPacket(msg1)
+
+	t.Log("Get the real response")
+	realResp := theirControl.GetFromUDP(true)
+
+	t.Log("Truncate the response and inject, should be ignored")
+	truncResp := realResp.Copy()
+	truncResp.Data = truncResp.Data[:header.Len]
+	myControl.InjectUDPPacket(truncResp)
+
+	t.Log("Verify pending handshake survived the truncated packet")
+	assert.NotEmpty(t, myControl.ListHostmapHosts(true), "pending handshake should still exist")
+
+	t.Log("Inject real response, should complete handshake")
+	myControl.InjectUDPPacket(realResp)
+	myControl.WaitForType(1, 0, theirControl)
+
+	t.Log("Drain and verify tunnel")
+	cachedPacket := theirControl.GetFromTun(true)
+	assertUdpPacket(t, []byte("Hi"), cachedPacket, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), 80, 80)
+	assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
+
+	myControl.Stop()
+	theirControl.Stop()
+}
+
+func TestHandshakeOrphanedMsg2Dropped(t *testing.T) {
+	t.Parallel()
+	// A msg2 arriving with no matching pending index should be silently dropped
+	// with no response sent and no state changes.
+
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	myControl, myVpnIpNet, myUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "me", "10.128.0.1/24", nil)
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "them", "10.128.0.2/24", nil)
+
+	myControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
+	theirControl.InjectLightHouseAddr(myVpnIpNet[0].Addr(), myUdpAddr)
+
+	myControl.Start()
+	theirControl.Start()
+
+	r := router.NewR(t, myControl, theirControl)
+	defer r.RenderFlow()
+
+	t.Log("Complete a normal handshake")
+	myControl.InjectTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi"))
+	r.RouteForAllUntilTxTun(theirControl)
+	assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
+
+	t.Log("Record hostmap state")
+	myIndexes := len(myControl.ListHostmapIndexes(false))
+
+	t.Log("Inject a fake msg2 with unknown RemoteIndex")
+	myControl.InjectUDPPacket(makeHandshakePacket(theirUdpAddr, myUdpAddr, header.HandshakeIXPSK0, 0xDEADBEEF, 2))
+
+	t.Log("Verify no new indexes created")
+	assert.Equal(t, myIndexes, len(myControl.ListHostmapIndexes(false)))
+
+	t.Log("Verify no UDP response was sent")
+	time.Sleep(100 * time.Millisecond)
+	assert.Nil(t, myControl.GetFromUDP(false), "should not send a response to orphaned msg2")
+
+	t.Log("Verify existing tunnel still works")
+	assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
+
+	myControl.Stop()
+	theirControl.Stop()
+}
+
+func TestHandshakeUnknownMessageCounter(t *testing.T) {
+	t.Parallel()
+	// A handshake packet with an unexpected message counter should be silently
+	// dropped with no side effects and no UDP response.
+
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	myControl, _, myUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "me", "10.128.0.1/24", nil)
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "them", "10.128.0.2/24", nil)
+
+	myControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
+
+	myControl.Start()
+	theirControl.Start()
+
+	t.Log("Inject handshake with MessageCounter=3")
+	myControl.InjectUDPPacket(makeHandshakePacket(theirUdpAddr, myUdpAddr, header.HandshakeIXPSK0, 0, 3))
+
+	t.Log("Inject handshake with MessageCounter=99")
+	myControl.InjectUDPPacket(makeHandshakePacket(theirUdpAddr, myUdpAddr, header.HandshakeIXPSK0, 0, 99))
+
+	t.Log("Verify no tunnels or pending handshakes")
+	assert.Empty(t, myControl.ListHostmapHosts(false))
+	assert.Empty(t, myControl.ListHostmapHosts(true))
+
+	t.Log("Verify no UDP response was sent")
+	time.Sleep(100 * time.Millisecond)
+	assert.Nil(t, myControl.GetFromUDP(false))
+
+	myControl.Stop()
+	theirControl.Stop()
+}
+
+func TestHandshakeUnknownSubtype(t *testing.T) {
+	t.Parallel()
+	// A handshake packet with an unknown subtype should be silently dropped.
+
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	myControl, _, myUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "me", "10.128.0.1/24", nil)
+	theirControl, _, theirUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "them", "10.128.0.2/24", nil)
+
+	myControl.Start()
+	theirControl.Start()
+
+	t.Log("Inject handshake with unknown subtype 99")
+	myControl.InjectUDPPacket(makeHandshakePacket(theirUdpAddr, myUdpAddr, header.MessageSubType(99), 0, 1))
+
+	t.Log("Verify no tunnels or pending handshakes")
+	assert.Empty(t, myControl.ListHostmapHosts(false))
+	assert.Empty(t, myControl.ListHostmapHosts(true))
+
+	t.Log("Verify no UDP response was sent")
+	time.Sleep(100 * time.Millisecond)
+	assert.Nil(t, myControl.GetFromUDP(false))
+
+	myControl.Stop()
+	theirControl.Stop()
+}
+
+func TestHandshakeLateResponse(t *testing.T) {
+	t.Parallel()
+	// After a handshake times out, a late response should be silently ignored
+	// with no new tunnels created.
+
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	myControl, myVpnIpNet, _, _ := newSimpleServer(cert.Version1, ca, caKey, "me", "10.128.0.1/24", m{
+		"handshakes": m{
+			"try_interval": "200ms",
+			"retries":      2,
+		},
+	})
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "them", "10.128.0.2/24", nil)
+
+	myControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
+
+	myControl.Start()
+	theirControl.Start()
+
+	t.Log("Trigger handshake from me")
+	myControl.InjectTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi"))
+
+	t.Log("Grab msg1 but don't deliver")
+	msg1 := myControl.GetFromUDP(true)
+
+	t.Log("Wait for handshake to time out")
+	for i := 0; i < 5; i++ {
+		time.Sleep(300 * time.Millisecond)
+		myControl.GetFromUDP(false)
+	}
+
+	t.Log("Confirm no pending handshakes remain")
+	assert.Empty(t, myControl.ListHostmapHosts(true))
+
+	t.Log("Deliver old msg1 to them, they create a tunnel")
+	theirControl.InjectUDPPacket(msg1)
+	resp := theirControl.GetFromUDP(true)
+	assert.NotNil(t, resp)
+
+	t.Log("Inject late response into me, should be ignored")
+	myControl.InjectUDPPacket(resp)
+
+	t.Log("No tunnel should exist on my side")
+	assert.Empty(t, myControl.ListHostmapHosts(false))
+	assert.Empty(t, myControl.ListHostmapHosts(true))
+
+	myControl.Stop()
+	theirControl.Stop()
+}
+
+func TestHandshakeSelfConnectionRejected(t *testing.T) {
+	t.Parallel()
+	// Verify that a node rejects a handshake containing its own VPN IP in the
+	// peer cert. We do this by sending the initiator's own msg1 back to itself.
+
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	myControl, myVpnIpNet, myUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "me", "10.128.0.1/24", nil)
+
+	// Need a lighthouse entry to trigger a handshake
+	myControl.InjectLightHouseAddr(netip.MustParseAddr("10.128.0.2"), netip.MustParseAddrPort("10.0.0.2:4242"))
+
+	myControl.Start()
+
+	t.Log("Trigger handshake from me")
+	myControl.InjectTunUDPPacket(netip.MustParseAddr("10.128.0.2"), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi"))
+	msg1 := myControl.GetFromUDP(true)
+
+	t.Log("Drain any handshake retransmits before injecting")
+	time.Sleep(100 * time.Millisecond)
+	for myControl.GetFromUDP(false) != nil {
+	}
+
+	t.Log("Feed my own msg1 back to me as if it came from someone else")
+	selfMsg := msg1.Copy()
+	selfMsg.From = netip.MustParseAddrPort("10.0.0.99:4242")
+	selfMsg.To = myUdpAddr
+	myControl.InjectUDPPacket(selfMsg)
+
+	t.Log("Verify no response was sent (self-connection rejected)")
+	time.Sleep(100 * time.Millisecond)
+	// Drain any further retransmits from the original handshake, then check
+	// that none of them are a handshake response (MessageCounter=2)
+	h := &header.H{}
+	for {
+		p := myControl.GetFromUDP(false)
+		if p == nil {
+			break
+		}
+		_ = h.Parse(p.Data)
+		assert.NotEqual(t, uint64(2), h.MessageCounter,
+			"should not send a stage 2 response to self-connection")
+	}
+
+	t.Log("Verify no tunnel to myself was created")
+	assert.Nil(t, myControl.GetHostInfoByVpnAddr(myVpnIpNet[0].Addr(), false))
+
+	myControl.Stop()
+}
+
+func TestHandshakeMessageCounter0Dropped(t *testing.T) {
+	t.Parallel()
+	// MessageCounter=0 is not a valid handshake message and should be dropped.
+
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	myControl, _, myUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "me", "10.128.0.1/24", nil)
+	_, _, theirUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "them", "10.128.0.2/24", nil)
+
+	myControl.Start()
+
+	t.Log("Inject handshake with MessageCounter=0")
+	myControl.InjectUDPPacket(makeHandshakePacket(theirUdpAddr, myUdpAddr, header.HandshakeIXPSK0, 0, 0))
+
+	time.Sleep(100 * time.Millisecond)
+	assert.Empty(t, myControl.ListHostmapHosts(false))
+	assert.Empty(t, myControl.ListHostmapHosts(true))
+	assert.Nil(t, myControl.GetFromUDP(false))
+
+	myControl.Stop()
+}
+
+func TestHandshakeRemoteAllowList(t *testing.T) {
+	t.Parallel()
+	// Verify that a handshake from a blocked underlay IP is dropped with no
+	// response and no state changes. Then verify the same packet from an
+	// allowed IP succeeds.
+
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	myControl, myVpnIpNet, myUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "me", "10.128.0.1/24", m{
+		"lighthouse": m{
+			"remote_allow_list": m{
+				"10.0.0.0/8": true,
+				"0.0.0.0/0":  false,
+			},
+		},
+	})
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "them", "10.128.0.2/24", nil)
+
+	myControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
+	theirControl.InjectLightHouseAddr(myVpnIpNet[0].Addr(), myUdpAddr)
+
+	myControl.Start()
+	theirControl.Start()
+
+	r := router.NewR(t, myControl, theirControl)
+	defer r.RenderFlow()
+
+	t.Log("Trigger handshake from them")
+	theirControl.InjectTunUDPPacket(myVpnIpNet[0].Addr(), 80, theirVpnIpNet[0].Addr(), 80, []byte("Hi"))
+	msg1 := theirControl.GetFromUDP(true)
+
+	t.Log("Rewrite the source to a blocked IP and inject")
+	blockedMsg := msg1.Copy()
+	blockedMsg.From = netip.MustParseAddrPort("192.168.1.1:4242")
+	myControl.InjectUDPPacket(blockedMsg)
+
+	t.Log("Verify no tunnel, no pending, no response from blocked source")
+	time.Sleep(100 * time.Millisecond)
+	assert.Empty(t, myControl.ListHostmapHosts(false))
+	assert.Empty(t, myControl.ListHostmapHosts(true))
+	assert.Nil(t, myControl.GetFromUDP(false), "should not respond to blocked source")
+
+	t.Log("Now inject the real packet from the allowed source")
+	myControl.InjectUDPPacket(msg1)
+
+	t.Log("Verify handshake completes from allowed source")
+	resp := myControl.GetFromUDP(true)
+	assert.NotNil(t, resp)
+	theirControl.InjectUDPPacket(resp)
+	theirControl.WaitForType(1, 0, myControl)
+
+	t.Log("Drain cached packet and verify tunnel works")
+	cachedPacket := myControl.GetFromTun(true)
+	assertUdpPacket(t, []byte("Hi"), cachedPacket, theirVpnIpNet[0].Addr(), myVpnIpNet[0].Addr(), 80, 80)
+	assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
+
+	myControl.Stop()
+	theirControl.Stop()
+}
+
+func TestHandshakeAlreadySeenPreferredRemote(t *testing.T) {
+	t.Parallel()
+	// When a duplicate msg1 arrives via ErrAlreadySeen, verify the tunnel
+	// remains functional and hostmap index count is stable.
+
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	myControl, myVpnIpNet, myUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "me", "10.128.0.1/24", nil)
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "them", "10.128.0.2/24", nil)
+
+	myControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
+	theirControl.InjectLightHouseAddr(myVpnIpNet[0].Addr(), myUdpAddr)
+
+	myControl.Start()
+	theirControl.Start()
+
+	r := router.NewR(t, myControl, theirControl)
+	defer r.RenderFlow()
+
+	t.Log("Complete a normal handshake via the router")
+	myControl.InjectTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi"))
+	r.RouteForAllUntilTxTun(theirControl)
+	assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
+
+	t.Log("Record hostmap state")
+	theirIndexes := len(theirControl.ListHostmapIndexes(false))
+	hi := theirControl.GetHostInfoByVpnAddr(myVpnIpNet[0].Addr(), false)
+	assert.NotNil(t, hi)
+	originalRemote := hi.CurrentRemote
+
+	t.Log("Re-trigger traffic to cause a new handshake attempt (ErrAlreadySeen)")
+	myControl.InjectTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("roam"))
+	r.RouteForAllUntilTxTun(theirControl)
+
+	t.Log("Verify tunnel still works")
+	assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
+
+	t.Log("Verify remote is still valid and index count is stable")
+	hi2 := theirControl.GetHostInfoByVpnAddr(myVpnIpNet[0].Addr(), false)
+	assert.NotNil(t, hi2)
+	assert.Equal(t, originalRemote, hi2.CurrentRemote)
+	assert.Equal(t, theirIndexes, len(theirControl.ListHostmapIndexes(false)),
+		"no extra indexes should be created from ErrAlreadySeen")
+
+	myControl.Stop()
+	theirControl.Stop()
+}
+
+func TestHandshakeWrongResponderPacketStore(t *testing.T) {
+	t.Parallel()
+	// Verify that when the wrong host responds, the cached packets are
+	// transferred to the new handshake, the evil tunnel is closed, evil's
+	// address is blocked, and the correct tunnel is eventually established.
+
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	myControl, myVpnIpNet, myUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "me", "10.128.0.100/24", nil)
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "them", "10.128.0.99/24", nil)
+	evilControl, evilVpnIpNet, evilUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "evil", "10.128.0.2/24", nil)
+
+	myControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), evilUdpAddr)
+
+	r := router.NewR(t, myControl, theirControl, evilControl)
+	defer r.RenderFlow()
+
+	myControl.Start()
+	theirControl.Start()
+	evilControl.Start()
+
+	t.Log("Send multiple packets to them (cached during handshake)")
+	myControl.InjectTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("packet1"))
+	myControl.InjectTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("packet2"))
+
+	t.Log("Route until evil tunnel is closed")
+	h := &header.H{}
+	r.RouteForAllExitFunc(func(p *udp.Packet, c *nebula.Control) router.ExitType {
+		if err := h.Parse(p.Data); err != nil {
+			panic(err)
+		}
+		if h.Type == header.CloseTunnel && p.To == evilUdpAddr {
+			return router.RouteAndExit
+		}
+		return router.KeepRouting
+	})
+
+	t.Log("Verify evil's address is blocked in the new pending handshake")
+	pendingHI := myControl.GetHostInfoByVpnAddr(theirVpnIpNet[0].Addr(), true)
+	if pendingHI != nil {
+		assert.NotContains(t, pendingHI.RemoteAddrs, evilUdpAddr,
+			"evil's address should be blocked")
+	}
+
+	t.Log("Inject correct lighthouse addr for them")
+	myControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
+
+	t.Log("Route until cached packets arrive at the real them")
+	p := r.RouteForAllUntilTxTun(theirControl)
+	assert.NotNil(t, p, "a cached packet should be delivered to the correct host")
+
+	t.Log("Verify the correct host has a tunnel")
+	assertHostInfoPair(t, myUdpAddr, theirUdpAddr, myVpnIpNet, theirVpnIpNet, myControl, theirControl)
+
+	t.Log("Verify no hostinfo artifacts from evil remain")
+	assert.Nil(t, myControl.GetHostInfoByVpnAddr(evilVpnIpNet[0].Addr(), true),
+		"no pending hostinfo for evil")
+	assert.Nil(t, myControl.GetHostInfoByVpnAddr(evilVpnIpNet[0].Addr(), false),
+		"no main hostinfo for evil")
+
+	myControl.Stop()
+	theirControl.Stop()
+	evilControl.Stop()
+}
+
+func TestHandshakeRelayComplete(t *testing.T) {
+	t.Parallel()
+	// Verify that a relay handshake completes correctly and relay state is
+	// properly maintained on all three nodes.
+
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	myControl, myVpnIpNet, _, _ := newSimpleServer(cert.Version1, ca, caKey, "me", "10.128.0.1/24", m{"relay": m{"use_relays": true}})
+	relayControl, relayVpnIpNet, relayUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "relay", "10.128.0.128/24", m{"relay": m{"am_relay": true}})
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "them", "10.128.0.2/24", m{"relay": m{"use_relays": true}})
+
+	myControl.InjectLightHouseAddr(relayVpnIpNet[0].Addr(), relayUdpAddr)
+	myControl.InjectRelays(theirVpnIpNet[0].Addr(), []netip.Addr{relayVpnIpNet[0].Addr()})
+	relayControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
+
+	r := router.NewR(t, myControl, relayControl, theirControl)
+	defer r.RenderFlow()
+
+	myControl.Start()
+	relayControl.Start()
+	theirControl.Start()
+
+	t.Log("Trigger handshake via relay")
+	myControl.InjectTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi via relay"))
+
+	p := r.RouteForAllUntilTxTun(theirControl)
+	assertUdpPacket(t, []byte("Hi via relay"), p, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), 80, 80)
+
+	t.Log("Verify bidirectional tunnel via relay")
+	assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
+
+	t.Log("Verify relay state on my side shows relay-to-me")
+	myHI := myControl.GetHostInfoByVpnAddr(theirVpnIpNet[0].Addr(), false)
+	assert.NotNil(t, myHI)
+	assert.NotEmpty(t, myHI.CurrentRelaysToMe, "should have relay-to-me for them")
+
+	t.Log("Verify relay state on their side shows relay-to-me")
+	theirHI := theirControl.GetHostInfoByVpnAddr(myVpnIpNet[0].Addr(), false)
+	assert.NotNil(t, theirHI)
+	assert.NotEmpty(t, theirHI.CurrentRelaysToMe, "should have relay-to-me for me")
+
+	t.Log("Verify relay node shows through-me relays")
+	relayHI := relayControl.GetHostInfoByVpnAddr(myVpnIpNet[0].Addr(), false)
+	assert.NotNil(t, relayHI)
+
+	myControl.Stop()
+	relayControl.Stop()
+	theirControl.Stop()
+}
+
+// NOTE: Relay V1 cert + IPv6 rejection is not tested here because
+// InjectTunUDPPacket from a V4 node to a V6 address panics in the test
+// framework. The check is in handshake_manager.go handleOutbound relay
+// logic (lines ~304-313): if the relay host has a V1 cert and either
+// address is IPv6, the relay is skipped.
+
+// NOTE: Relay reestablishment (Disestablished state transition) is covered
+// by the existing TestReestablishRelays in handshakes_test.go.

e2e/handshakes_test.go

@@ -11,7 +11,6 @@ import (
 
 	"github.com/google/gopacket"
 	"github.com/google/gopacket/layers"
-	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula"
 	"github.com/slackhq/nebula/cert"
 	"github.com/slackhq/nebula/cert_test"
@@ -20,16 +19,17 @@ import (
 	"github.com/slackhq/nebula/udp"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
-	"gopkg.in/yaml.v3"
+	"go.yaml.in/yaml/v3"
 )
 
 func BenchmarkHotPath(b *testing.B) {
 	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
-	myControl, myVpnIpNet, _, _ := newSimpleServer(cert.Version1, ca, caKey, "me", "10.128.0.1/24", nil)
+	myControl, myVpnIpNet, myUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "me", "10.128.0.1/24", nil)
 	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "them", "10.128.0.2/24", nil)
 
 	// Put their info in our lighthouse
 	myControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
+	theirControl.InjectLightHouseAddr(myVpnIpNet[0].Addr(), myUdpAddr)
 
 	// Start the servers
 	myControl.Start()
@@ -38,6 +38,9 @@ func BenchmarkHotPath(b *testing.B) {
 	r := router.NewR(b, myControl, theirControl)
 	r.CancelFlowLogs()
 
+	assertTunnel(b, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
+	b.ResetTimer()
+
 	for n := 0; n < b.N; n++ {
 		myControl.InjectTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me"))
 		_ = r.RouteForAllUntilTxTun(theirControl)
@@ -47,7 +50,41 @@ func BenchmarkHotPath(b *testing.B) {
 	theirControl.Stop()
 }
 
+func BenchmarkHotPathRelay(b *testing.B) {
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	myControl, myVpnIpNet, _, _ := newSimpleServer(cert.Version1, ca, caKey, "me     ", "10.128.0.1/24", m{"relay": m{"use_relays": true}})
+	relayControl, relayVpnIpNet, relayUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "relay  ", "10.128.0.128/24", m{"relay": m{"am_relay": true}})
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "them   ", "10.128.0.2/24", m{"relay": m{"use_relays": true}})
+
+	// Teach my how to get to the relay and that their can be reached via the relay
+	myControl.InjectLightHouseAddr(relayVpnIpNet[0].Addr(), relayUdpAddr)
+	myControl.InjectRelays(theirVpnIpNet[0].Addr(), []netip.Addr{relayVpnIpNet[0].Addr()})
+	relayControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
+
+	// Build a router so we don't have to reason who gets which packet
+	r := router.NewR(b, myControl, relayControl, theirControl)
+	r.CancelFlowLogs()
+
+	// Start the servers
+	myControl.Start()
+	relayControl.Start()
+	theirControl.Start()
+
+	assertTunnel(b, theirVpnIpNet[0].Addr(), myVpnIpNet[0].Addr(), theirControl, myControl, r)
+	b.ResetTimer()
+
+	for n := 0; n < b.N; n++ {
+		myControl.InjectTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me"))
+		_ = r.RouteForAllUntilTxTun(theirControl)
+	}
+
+	myControl.Stop()
+	theirControl.Stop()
+	relayControl.Stop()
+}
+
 func TestGoodHandshake(t *testing.T) {
+	t.Parallel()
 	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
 	myControl, myVpnIpNet, myUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "me", "10.128.0.1/24", nil)
 	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "them", "10.128.0.2/24", nil)
@@ -97,7 +134,44 @@ func TestGoodHandshake(t *testing.T) {
 	theirControl.Stop()
 }
 
+func TestGoodHandshakeNoOverlap(t *testing.T) {
+	t.Parallel()
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	myControl, myVpnIpNet, myUdpAddr, _ := newSimpleServer(cert.Version2, ca, caKey, "me", "10.128.0.1/24", nil)
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version2, ca, caKey, "them", "2001::69/24", nil) //look ma, cross-stack!
+
+	// Put their info in our lighthouse
+	myControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
+
+	// Start the servers
+	myControl.Start()
+	theirControl.Start()
+
+	empty := []byte{}
+	t.Log("do something to cause a handshake")
+	myControl.GetF().SendMessageToVpnAddr(header.Test, header.MessageNone, theirVpnIpNet[0].Addr(), empty, empty, empty)
+
+	t.Log("Have them consume my stage 0 packet. They have a tunnel now")
+	theirControl.InjectUDPPacket(myControl.GetFromUDP(true))
+
+	t.Log("Get their stage 1 packet")
+	stage1Packet := theirControl.GetFromUDP(true)
+
+	t.Log("Have me consume their stage 1 packet. I have a tunnel now")
+	myControl.InjectUDPPacket(stage1Packet)
+
+	t.Log("Wait until we see a test packet come through to make sure we give the tunnel time to complete")
+	myControl.WaitForType(header.Test, 0, theirControl)
+
+	t.Log("Make sure our host infos are correct")
+	assertHostInfoPair(t, myUdpAddr, theirUdpAddr, myVpnIpNet, theirVpnIpNet, myControl, theirControl)
+
+	myControl.Stop()
+	theirControl.Stop()
+}
+
 func TestWrongResponderHandshake(t *testing.T) {
+	t.Parallel()
 	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
 
 	myControl, myVpnIpNet, myUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "me", "10.128.0.100/24", nil)
@@ -174,6 +248,7 @@ func TestWrongResponderHandshake(t *testing.T) {
 }
 
 func TestWrongResponderHandshakeStaticHostMap(t *testing.T) {
+	t.Parallel()
 	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
 
 	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "them", "10.128.0.99/24", nil)
@@ -256,6 +331,7 @@ func TestWrongResponderHandshakeStaticHostMap(t *testing.T) {
 }
 
 func TestStage1Race(t *testing.T) {
+	t.Parallel()
 	// This tests ensures that two hosts handshaking with each other at the same time will allow traffic to flow
 	// But will eventually collapse down to a single tunnel
 
@@ -336,6 +412,7 @@ func TestStage1Race(t *testing.T) {
 }
 
 func TestUncleanShutdownRaceLoser(t *testing.T) {
+	t.Parallel()
 	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
 	myControl, myVpnIpNet, myUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "me  ", "10.128.0.1/24", nil)
 	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "them", "10.128.0.2/24", nil)
@@ -385,6 +462,7 @@ func TestUncleanShutdownRaceLoser(t *testing.T) {
 }
 
 func TestUncleanShutdownRaceWinner(t *testing.T) {
+	t.Parallel()
 	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
 	myControl, myVpnIpNet, myUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "me  ", "10.128.0.1/24", nil)
 	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "them", "10.128.0.2/24", nil)
@@ -436,6 +514,7 @@ func TestUncleanShutdownRaceWinner(t *testing.T) {
 }
 
 func TestRelays(t *testing.T) {
+	t.Parallel()
 	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
 	myControl, myVpnIpNet, _, _ := newSimpleServer(cert.Version1, ca, caKey, "me     ", "10.128.0.1/24", m{"relay": m{"use_relays": true}})
 	relayControl, relayVpnIpNet, relayUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "relay  ", "10.128.0.128/24", m{"relay": m{"am_relay": true}})
@@ -464,7 +543,38 @@ func TestRelays(t *testing.T) {
 	r.RenderHostmaps("Final hostmaps", myControl, relayControl, theirControl)
 }
 
+func TestRelaysDontCareAboutIps(t *testing.T) {
+	t.Parallel()
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	myControl, myVpnIpNet, _, _ := newSimpleServer(cert.Version2, ca, caKey, "me     ", "10.128.0.1/24", m{"relay": m{"use_relays": true}})
+	relayControl, relayVpnIpNet, relayUdpAddr, _ := newSimpleServer(cert.Version2, ca, caKey, "relay  ", "2001::9999/24", m{"relay": m{"am_relay": true}})
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version2, ca, caKey, "them   ", "10.128.0.2/24", m{"relay": m{"use_relays": true}})
+
+	// Teach my how to get to the relay and that their can be reached via the relay
+	myControl.InjectLightHouseAddr(relayVpnIpNet[0].Addr(), relayUdpAddr)
+	myControl.InjectRelays(theirVpnIpNet[0].Addr(), []netip.Addr{relayVpnIpNet[0].Addr()})
+	relayControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
+
+	// Build a router so we don't have to reason who gets which packet
+	r := router.NewR(t, myControl, relayControl, theirControl)
+	defer r.RenderFlow()
+
+	// Start the servers
+	myControl.Start()
+	relayControl.Start()
+	theirControl.Start()
+
+	t.Log("Trigger a handshake from me to them via the relay")
+	myControl.InjectTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me"))
+
+	p := r.RouteForAllUntilTxTun(theirControl)
+	r.Log("Assert the tunnel works")
+	assertUdpPacket(t, []byte("Hi from me"), p, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), 80, 80)
+	r.RenderHostmaps("Final hostmaps", myControl, relayControl, theirControl)
+}
+
 func TestReestablishRelays(t *testing.T) {
+	t.Parallel()
 	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
 	myControl, myVpnIpNet, _, _ := newSimpleServer(cert.Version1, ca, caKey, "me     ", "10.128.0.1/24", m{"relay": m{"use_relays": true}})
 	relayControl, relayVpnIpNet, relayUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "relay  ", "10.128.0.128/24", m{"relay": m{"am_relay": true}})
@@ -596,6 +706,7 @@ func TestReestablishRelays(t *testing.T) {
 }
 
 func TestStage1RaceRelays(t *testing.T) {
+	t.Parallel()
 	//NOTE: this is a race between me and relay resulting in a full tunnel from me to them via relay
 	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
 	myControl, myVpnIpNet, myUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "me     ", "10.128.0.1/24", m{"relay": m{"use_relays": true}})
@@ -643,12 +754,12 @@ func TestStage1RaceRelays(t *testing.T) {
 }
 
 func TestStage1RaceRelays2(t *testing.T) {
+	t.Parallel()
 	//NOTE: this is a race between me and relay resulting in a full tunnel from me to them via relay
 	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
 	myControl, myVpnIpNet, myUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "me     ", "10.128.0.1/24", m{"relay": m{"use_relays": true}})
 	relayControl, relayVpnIpNet, relayUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "relay  ", "10.128.0.128/24", m{"relay": m{"am_relay": true}})
 	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "them   ", "10.128.0.2/24", m{"relay": m{"use_relays": true}})
-	l := NewTestLogger()
 
 	// Teach my how to get to the relay and that their can be reached via the relay
 	myControl.InjectLightHouseAddr(relayVpnIpNet[0].Addr(), relayUdpAddr)
@@ -670,49 +781,41 @@ func TestStage1RaceRelays2(t *testing.T) {
 	theirControl.Start()
 
 	r.Log("Get a tunnel between me and relay")
-	l.Info("Get a tunnel between me and relay")
 	assertTunnel(t, myVpnIpNet[0].Addr(), relayVpnIpNet[0].Addr(), myControl, relayControl, r)
 
 	r.Log("Get a tunnel between them and relay")
-	l.Info("Get a tunnel between them and relay")
 	assertTunnel(t, theirVpnIpNet[0].Addr(), relayVpnIpNet[0].Addr(), theirControl, relayControl, r)
 
 	r.Log("Trigger a handshake from both them and me via relay to them and me")
-	l.Info("Trigger a handshake from both them and me via relay to them and me")
 	myControl.InjectTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me"))
 	theirControl.InjectTunUDPPacket(myVpnIpNet[0].Addr(), 80, theirVpnIpNet[0].Addr(), 80, []byte("Hi from them"))
 
 	//r.RouteUntilAfterMsgType(myControl, header.Control, header.MessageNone)
 	//r.RouteUntilAfterMsgType(theirControl, header.Control, header.MessageNone)
 
-	r.Log("Wait for a packet from them to me")
+	r.Log("Wait for a packet from them to me; myControl")
-	l.Info("Wait for a packet from them to me; myControl")
 	r.RouteForAllUntilTxTun(myControl)
-	l.Info("Wait for a packet from them to me; theirControl")
+	r.Log("Wait for a packet from them to me; theirControl")
 	r.RouteForAllUntilTxTun(theirControl)
 
 	r.Log("Assert the tunnel works")
-	l.Info("Assert the tunnel works")
 	assertTunnel(t, theirVpnIpNet[0].Addr(), myVpnIpNet[0].Addr(), theirControl, myControl, r)
 
 	t.Log("Wait until we remove extra tunnels")
-	l.Info("Wait until we remove extra tunnels")
+	t.Logf("Waiting for hostinfos to be removed... myControl=%d theirControl=%d relayControl=%d",
-	l.WithFields(
+		len(myControl.GetHostmap().Indexes),
-		logrus.Fields{
+		len(theirControl.GetHostmap().Indexes),
-			"myControl":    len(myControl.GetHostmap().Indexes),
+		len(relayControl.GetHostmap().Indexes),
-			"theirControl": len(theirControl.GetHostmap().Indexes),
+	)
-			"relayControl": len(relayControl.GetHostmap().Indexes),
-		}).Info("Waiting for hostinfos to be removed...")
 	hostInfos := len(myControl.GetHostmap().Indexes) + len(theirControl.GetHostmap().Indexes) + len(relayControl.GetHostmap().Indexes)
 	retries := 60
 	for hostInfos > 6 && retries > 0 {
 		hostInfos = len(myControl.GetHostmap().Indexes) + len(theirControl.GetHostmap().Indexes) + len(relayControl.GetHostmap().Indexes)
-		l.WithFields(
+		t.Logf("Waiting for hostinfos to be removed... myControl=%d theirControl=%d relayControl=%d",
-			logrus.Fields{
+			len(myControl.GetHostmap().Indexes),
-				"myControl":    len(myControl.GetHostmap().Indexes),
+			len(theirControl.GetHostmap().Indexes),
-				"theirControl": len(theirControl.GetHostmap().Indexes),
+			len(relayControl.GetHostmap().Indexes),
-				"relayControl": len(relayControl.GetHostmap().Indexes),
+		)
-			}).Info("Waiting for hostinfos to be removed...")
 		assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
 		t.Log("Connection manager hasn't ticked yet")
 		time.Sleep(time.Second)
@@ -720,7 +823,6 @@ func TestStage1RaceRelays2(t *testing.T) {
 	}
 
 	r.Log("Assert the tunnel works")
-	l.Info("Assert the tunnel works")
 	assertTunnel(t, theirVpnIpNet[0].Addr(), myVpnIpNet[0].Addr(), theirControl, myControl, r)
 
 	myControl.Stop()
@@ -729,6 +831,7 @@ func TestStage1RaceRelays2(t *testing.T) {
 }
 
 func TestRehandshakingRelays(t *testing.T) {
+	t.Parallel()
 	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
 	myControl, myVpnIpNet, _, _ := newSimpleServer(cert.Version1, ca, caKey, "me     ", "10.128.0.1/24", m{"relay": m{"use_relays": true}})
 	relayControl, relayVpnIpNet, relayUdpAddr, relayConfig := newSimpleServer(cert.Version1, ca, caKey, "relay  ", "10.128.0.128/24", m{"relay": m{"am_relay": true}})
@@ -832,6 +935,7 @@ func TestRehandshakingRelays(t *testing.T) {
 }
 
 func TestRehandshakingRelaysPrimary(t *testing.T) {
+	t.Parallel()
 	// This test is the same as TestRehandshakingRelays but one of the terminal types is a primary swap winner
 	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
 	myControl, myVpnIpNet, _, _ := newSimpleServer(cert.Version1, ca, caKey, "me     ", "10.128.0.128/24", m{"relay": m{"use_relays": true}})
@@ -936,6 +1040,7 @@ func TestRehandshakingRelaysPrimary(t *testing.T) {
 }
 
 func TestRehandshaking(t *testing.T) {
+	t.Parallel()
 	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
 	myControl, myVpnIpNet, myUdpAddr, myConfig := newSimpleServer(cert.Version1, ca, caKey, "me  ", "10.128.0.2/24", nil)
 	theirControl, theirVpnIpNet, theirUdpAddr, theirConfig := newSimpleServer(cert.Version1, ca, caKey, "them", "10.128.0.1/24", nil)
@@ -1031,6 +1136,7 @@ func TestRehandshaking(t *testing.T) {
 }
 
 func TestRehandshakingLoser(t *testing.T) {
+	t.Parallel()
 	// The purpose of this test is that the race loser renews their certificate and rehandshakes. The final tunnel
 	// Should be the one with the new certificate
 	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
@@ -1129,6 +1235,7 @@ func TestRehandshakingLoser(t *testing.T) {
 }
 
 func TestRaceRegression(t *testing.T) {
+	t.Parallel()
 	// This test forces stage 1, stage 2, stage 1 to be received by me from them
 	// We had a bug where we were not finding the duplicate handshake and responding to the final stage 1 which
 	// caused a cross-linked hostinfo
@@ -1189,6 +1296,7 @@ func TestRaceRegression(t *testing.T) {
 }
 
 func TestV2NonPrimaryWithLighthouse(t *testing.T) {
+	t.Parallel()
 	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
 	lhControl, lhVpnIpNet, lhUdpAddr, _ := newSimpleServer(cert.Version2, ca, caKey, "lh  ", "10.128.0.1/24, ff::1/64", m{"lighthouse": m{"am_lighthouse": true}})
 
@@ -1227,3 +1335,187 @@ func TestV2NonPrimaryWithLighthouse(t *testing.T) {
 	myControl.Stop()
 	theirControl.Stop()
 }
+
+func TestV2NonPrimaryWithOffNetLighthouse(t *testing.T) {
+	t.Parallel()
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	lhControl, lhVpnIpNet, lhUdpAddr, _ := newSimpleServer(cert.Version2, ca, caKey, "lh  ", "2001::1/64", m{"lighthouse": m{"am_lighthouse": true}})
+
+	o := m{
+		"static_host_map": m{
+			lhVpnIpNet[0].Addr().String(): []string{lhUdpAddr.String()},
+		},
+		"lighthouse": m{
+			"hosts": []string{lhVpnIpNet[0].Addr().String()},
+			"local_allow_list": m{
+				// Try and block our lighthouse updates from using the actual addresses assigned to this computer
+				// If we start discovering addresses the test router doesn't know about then test traffic cant flow
+				"10.0.0.0/24": true,
+				"::/0":        false,
+			},
+		},
+	}
+	myControl, myVpnIpNet, _, _ := newSimpleServer(cert.Version2, ca, caKey, "me  ", "10.128.0.2/24, ff::2/64", o)
+	theirControl, theirVpnIpNet, _, _ := newSimpleServer(cert.Version2, ca, caKey, "them", "10.128.0.3/24, ff::3/64", o)
+
+	// Build a router so we don't have to reason who gets which packet
+	r := router.NewR(t, lhControl, myControl, theirControl)
+	defer r.RenderFlow()
+
+	// Start the servers
+	lhControl.Start()
+	myControl.Start()
+	theirControl.Start()
+
+	t.Log("Stand up an ipv6 tunnel between me and them")
+	assert.True(t, myVpnIpNet[1].Addr().Is6())
+	assert.True(t, theirVpnIpNet[1].Addr().Is6())
+	assertTunnel(t, myVpnIpNet[1].Addr(), theirVpnIpNet[1].Addr(), myControl, theirControl, r)
+
+	lhControl.Stop()
+	myControl.Stop()
+	theirControl.Stop()
+}
+
+func TestLighthouseUpdateOnReload(t *testing.T) {
+	t.Parallel()
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+
+	// Create the lighthouse
+	lhControl, lhVpnIpNet, lhUdpAddr, _ := newSimpleServer(cert.Version2, ca, caKey, "lh", "10.128.0.1/24", m{"lighthouse": m{"am_lighthouse": true}})
+
+	// Create a client with NO lighthouse configured and a long update interval.
+	// The initial SendUpdate at startup will be a no-op since no lighthouses are known.
+	myControl, myVpnIpNet, _, myConfig := newSimpleServer(cert.Version2, ca, caKey, "me", "10.128.0.2/24", m{
+		"lighthouse": m{
+			"interval": 600,
+			"local_allow_list": m{
+				"10.0.0.0/24": true,
+				"::/0":        false,
+			},
+		},
+	})
+
+	r := router.NewR(t, lhControl, myControl)
+	defer r.RenderFlow()
+
+	lhControl.Start()
+	myControl.Start()
+
+	// Drain any startup packets (there should be none meaningful)
+	r.FlushAll()
+
+	// Verify lighthouse has no knowledge of the client
+	assert.Nil(t, lhControl.QueryLighthouse(myVpnIpNet[0].Addr()))
+
+	// Build a new config that adds the lighthouse
+	newSettings := make(m)
+	for k, v := range myConfig.Settings {
+		newSettings[k] = v
+	}
+	newSettings["static_host_map"] = m{
+		lhVpnIpNet[0].Addr().String(): []any{lhUdpAddr.String()},
+	}
+	newSettings["lighthouse"] = m{
+		"hosts":    []any{lhVpnIpNet[0].Addr().String()},
+		"interval": 600,
+		"local_allow_list": m{
+			"10.0.0.0/24": true,
+			"::/0":        false,
+		},
+	}
+	newCfg, err := yaml.Marshal(newSettings)
+	require.NoError(t, err)
+
+	// Reload the config. The lighthouse.hosts change triggers TriggerUpdate,
+	// which wakes the update worker. It calls SendUpdate, initiating a
+	// handshake to the new lighthouse and caching the HostUpdateNotification.
+	require.NoError(t, myConfig.ReloadConfigString(string(newCfg)))
+
+	// Route until the lighthouse receives the HostUpdateNotification.
+	// This covers: handshake stage 1, stage 2, then the cached update.
+	done := make(chan struct{})
+	go func() {
+		r.RouteForAllUntilAfterMsgTypeTo(lhControl, header.LightHouse, 0)
+		close(done)
+	}()
+	select {
+	case <-done:
+	case <-time.After(5 * time.Second):
+		t.Fatal("timed out waiting for lighthouse update after config reload")
+	}
+
+	// Verify lighthouse now has the client's addresses
+	assert.NotNil(t, lhControl.QueryLighthouse(myVpnIpNet[0].Addr()))
+
+	r.RenderHostmaps("Final hostmaps", lhControl, myControl)
+	lhControl.Stop()
+	myControl.Stop()
+}
+
+func TestGoodHandshakeUnsafeDest(t *testing.T) {
+	t.Parallel()
+	unsafePrefix := "192.168.6.0/24"
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServerWithUdpAndUnsafeNetworks(cert.Version2, ca, caKey, "spooky", "10.128.0.2/24", netip.MustParseAddrPort("10.64.0.2:4242"), unsafePrefix, nil)
+	route := m{"route": unsafePrefix, "via": theirVpnIpNet[0].Addr().String()}
+	myCfg := m{
+		"tun": m{
+			"unsafe_routes": []m{route},
+		},
+	}
+	myControl, myVpnIpNet, myUdpAddr, myConfig := newSimpleServer(cert.Version2, ca, caKey, "me", "10.128.0.1/24", myCfg)
+	t.Logf("my config %v", myConfig)
+	// Put their info in our lighthouse
+	myControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
+
+	spookyDest := netip.MustParseAddr("192.168.6.4")
+
+	// Start the servers
+	myControl.Start()
+	theirControl.Start()
+
+	t.Log("Send a udp packet through to begin standing up the tunnel, this should come out the other side")
+	myControl.InjectTunUDPPacket(spookyDest, 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me"))
+
+	t.Log("Have them consume my stage 0 packet. They have a tunnel now")
+	theirControl.InjectUDPPacket(myControl.GetFromUDP(true))
+
+	t.Log("Get their stage 1 packet so that we can play with it")
+	stage1Packet := theirControl.GetFromUDP(true)
+
+	t.Log("I consume a garbage packet with a proper nebula header for our tunnel")
+	// this should log a statement and get ignored, allowing the real handshake packet to complete the tunnel
+	badPacket := stage1Packet.Copy()
+	badPacket.Data = badPacket.Data[:len(badPacket.Data)-header.Len]
+	myControl.InjectUDPPacket(badPacket)
+
+	t.Log("Have me consume their real stage 1 packet. I have a tunnel now")
+	myControl.InjectUDPPacket(stage1Packet)
+
+	t.Log("Wait until we see my cached packet come through")
+	myControl.WaitForType(1, 0, theirControl)
+
+	t.Log("Make sure our host infos are correct")
+	assertHostInfoPair(t, myUdpAddr, theirUdpAddr, myVpnIpNet, theirVpnIpNet, myControl, theirControl)
+
+	t.Log("Get that cached packet and make sure it looks right")
+	myCachedPacket := theirControl.GetFromTun(true)
+	assertUdpPacket(t, []byte("Hi from me"), myCachedPacket, myVpnIpNet[0].Addr(), spookyDest, 80, 80)
+
+	//reply
+	theirControl.InjectTunUDPPacket(myVpnIpNet[0].Addr(), 80, spookyDest, 80, []byte("Hi from the spookyman"))
+	//wait for reply
+	theirControl.WaitForType(1, 0, myControl)
+	theirCachedPacket := myControl.GetFromTun(true)
+	assertUdpPacket(t, []byte("Hi from the spookyman"), theirCachedPacket, spookyDest, myVpnIpNet[0].Addr(), 80, 80)
+
+	t.Log("Do a bidirectional tunnel test")
+	r := router.NewR(t, myControl, theirControl)
+	defer r.RenderFlow()
+	assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
+
+	r.RenderHostmaps("Final hostmaps", myControl, theirControl)
+	myControl.Stop()
+	theirControl.Stop()
+}

e2e/helpers_test.go

@@ -4,7 +4,6 @@
 package e2e
 
 import (
-	"fmt"
 	"io"
 	"net/netip"
 	"os"
@@ -12,25 +11,27 @@ import (
 	"testing"
 	"time"
 
+	"log/slog"
+
 	"dario.cat/mergo"
 	"github.com/google/gopacket"
 	"github.com/google/gopacket/layers"
-	"github.com/sirupsen/logrus"
+
 	"github.com/slackhq/nebula"
 	"github.com/slackhq/nebula/cert"
 	"github.com/slackhq/nebula/cert_test"
 	"github.com/slackhq/nebula/config"
 	"github.com/slackhq/nebula/e2e/router"
+	"github.com/slackhq/nebula/logging"
 	"github.com/stretchr/testify/assert"
-	"gopkg.in/yaml.v3"
+	"github.com/stretchr/testify/require"
+	"go.yaml.in/yaml/v3"
 )
 
 type m = map[string]any
 
 // newSimpleServer creates a nebula instance with many assumptions
 func newSimpleServer(v cert.Version, caCrt cert.Certificate, caKey []byte, name string, sVpnNetworks string, overrides m) (*nebula.Control, []netip.Prefix, netip.AddrPort, *config.C) {
-	l := NewTestLogger()
-
 	var vpnNetworks []netip.Prefix
 	for _, sn := range strings.Split(sVpnNetworks, ",") {
 		vpnIpNet, err := netip.ParsePrefix(strings.TrimSpace(sn))
@@ -56,7 +57,54 @@ func newSimpleServer(v cert.Version, caCrt cert.Certificate, caKey []byte, name
 		budpIp[3] = 239
 		udpAddr = netip.AddrPortFrom(netip.AddrFrom16(budpIp), 4242)
 	}
-	_, _, myPrivKey, myPEM := cert_test.NewTestCert(v, cert.Curve_CURVE25519, caCrt, caKey, name, time.Now(), time.Now().Add(5*time.Minute), vpnNetworks, nil, []string{})
+	return newSimpleServerWithUdp(v, caCrt, caKey, name, sVpnNetworks, udpAddr, overrides)
+}
+
+func newSimpleServerWithUdp(v cert.Version, caCrt cert.Certificate, caKey []byte, name string, sVpnNetworks string, udpAddr netip.AddrPort, overrides m) (*nebula.Control, []netip.Prefix, netip.AddrPort, *config.C) {
+	return newSimpleServerWithUdpAndUnsafeNetworks(v, caCrt, caKey, name, sVpnNetworks, udpAddr, "", overrides)
+}
+
+func newSimpleServerWithUdpAndUnsafeNetworks(v cert.Version, caCrt cert.Certificate, caKey []byte, name string, sVpnNetworks string, udpAddr netip.AddrPort, sUnsafeNetworks string, overrides m) (*nebula.Control, []netip.Prefix, netip.AddrPort, *config.C) {
+	l := NewTestLogger()
+
+	var vpnNetworks []netip.Prefix
+	for _, sn := range strings.Split(sVpnNetworks, ",") {
+		vpnIpNet, err := netip.ParsePrefix(strings.TrimSpace(sn))
+		if err != nil {
+			panic(err)
+		}
+		vpnNetworks = append(vpnNetworks, vpnIpNet)
+	}
+
+	if len(vpnNetworks) == 0 {
+		panic("no vpn networks")
+	}
+
+	firewallInbound := []m{{
+		"proto": "any",
+		"port":  "any",
+		"host":  "any",
+	}}
+
+	var unsafeNetworks []netip.Prefix
+	if sUnsafeNetworks != "" {
+		firewallInbound = []m{{
+			"proto":      "any",
+			"port":       "any",
+			"host":       "any",
+			"local_cidr": "0.0.0.0/0",
+		}}
+
+		for _, sn := range strings.Split(sUnsafeNetworks, ",") {
+			x, err := netip.ParsePrefix(strings.TrimSpace(sn))
+			if err != nil {
+				panic(err)
+			}
+			unsafeNetworks = append(unsafeNetworks, x)
+		}
+	}
+
+	_, _, myPrivKey, myPEM := cert_test.NewTestCert(v, cert.Curve_CURVE25519, caCrt, caKey, name, time.Now(), time.Now().Add(5*time.Minute), vpnNetworks, unsafeNetworks, []string{})
 
 	caB, err := caCrt.MarshalPEM()
 	if err != nil {
@@ -76,11 +124,7 @@ func newSimpleServer(v cert.Version, caCrt cert.Certificate, caKey []byte, name
 				"port":  "any",
 				"host":  "any",
 			}},
-			"inbound": []m{{
+			"inbound": firewallInbound,
-				"proto": "any",
-				"port":  "any",
-				"host":  "any",
-			}},
 		},
 		//"handshakes": m{
 		//	"try_interval": "1s",
@@ -90,8 +134,7 @@ func newSimpleServer(v cert.Version, caCrt cert.Certificate, caKey []byte, name
 			"port": udpAddr.Port(),
 		},
 		"logging": m{
-			"timestamp_format": fmt.Sprintf("%v 15:04:05.000000", name),
+			"level": testLogLevelName(),
-			"level":            l.Level.String(),
 		},
 		"timers": m{
 			"pending_deletion_interval": 2,
@@ -192,8 +235,7 @@ func newServer(caCrt []cert.Certificate, certs []cert.Certificate, key []byte, o
 			"port": udpAddr.Port(),
 		},
 		"logging": m{
-			"timestamp_format": fmt.Sprintf("%v 15:04:05.000000", certs[0].Name()),
+			"level": testLogLevelName(),
-			"level":            l.Level.String(),
 		},
 		"timers": m{
 			"pending_deletion_interval": 2,
@@ -250,7 +292,7 @@ func deadline(t *testing.T, seconds time.Duration) doneCb {
 	}
 }
 
-func assertTunnel(t *testing.T, vpnIpA, vpnIpB netip.Addr, controlA, controlB *nebula.Control, r *router.R) {
+func assertTunnel(t testing.TB, vpnIpA, vpnIpB netip.Addr, controlA, controlB *nebula.Control, r *router.R) {
 	// Send a packet from them to me
 	controlB.InjectTunUDPPacket(vpnIpA, 80, vpnIpB, 90, []byte("Hi from B"))
 	bPacket := r.RouteForAllUntilTxTun(controlA)
@@ -262,14 +304,14 @@ func assertTunnel(t *testing.T, vpnIpA, vpnIpB netip.Addr, controlA, controlB *n
 	assertUdpPacket(t, []byte("Hello from A"), aPacket, vpnIpA, vpnIpB, 90, 80)
 }
 
-func assertHostInfoPair(t *testing.T, addrA, addrB netip.AddrPort, vpnNetsA, vpnNetsB []netip.Prefix, controlA, controlB *nebula.Control) {
+func assertHostInfoPair(t testing.TB, addrA, addrB netip.AddrPort, vpnNetsA, vpnNetsB []netip.Prefix, controlA, controlB *nebula.Control) {
 	// Get both host infos
 	//TODO: CERT-V2 we may want to loop over each vpnAddr and assert all the things
 	hBinA := controlA.GetHostInfoByVpnAddr(vpnNetsB[0].Addr(), false)
-	assert.NotNil(t, hBinA, "Host B was not found by vpnAddr in controlA")
+	require.NotNil(t, hBinA, "Host B was not found by vpnAddr in controlA")
 
 	hAinB := controlB.GetHostInfoByVpnAddr(vpnNetsA[0].Addr(), false)
-	assert.NotNil(t, hAinB, "Host A was not found by vpnAddr in controlB")
+	require.NotNil(t, hAinB, "Host A was not found by vpnAddr in controlB")
 
 	// Check that both vpn and real addr are correct
 	assert.EqualValues(t, getAddrs(vpnNetsB), hBinA.VpnAddrs, "Host B VpnIp is wrong in control A")
@@ -283,7 +325,7 @@ func assertHostInfoPair(t *testing.T, addrA, addrB netip.AddrPort, vpnNetsA, vpn
 	assert.Equal(t, hBinA.RemoteIndex, hAinB.LocalIndex, "Host B remote index does not match host A local index")
 }
 
-func assertUdpPacket(t *testing.T, expected, b []byte, fromIp, toIp netip.Addr, fromPort, toPort uint16) {
+func assertUdpPacket(t testing.TB, expected, b []byte, fromIp, toIp netip.Addr, fromPort, toPort uint16) {
 	if toIp.Is6() {
 		assertUdpPacket6(t, expected, b, fromIp, toIp, fromPort, toPort)
 	} else {
@@ -291,7 +333,7 @@ func assertUdpPacket(t *testing.T, expected, b []byte, fromIp, toIp netip.Addr,
 	}
 }
 
-func assertUdpPacket6(t *testing.T, expected, b []byte, fromIp, toIp netip.Addr, fromPort, toPort uint16) {
+func assertUdpPacket6(t testing.TB, expected, b []byte, fromIp, toIp netip.Addr, fromPort, toPort uint16) {
 	packet := gopacket.NewPacket(b, layers.LayerTypeIPv6, gopacket.Lazy)
 	v6 := packet.Layer(layers.LayerTypeIPv6).(*layers.IPv6)
 	assert.NotNil(t, v6, "No ipv6 data found")
@@ -310,7 +352,7 @@ func assertUdpPacket6(t *testing.T, expected, b []byte, fromIp, toIp netip.Addr,
 	assert.Equal(t, expected, data.Payload(), "Data was incorrect")
 }
 
-func assertUdpPacket4(t *testing.T, expected, b []byte, fromIp, toIp netip.Addr, fromPort, toPort uint16) {
+func assertUdpPacket4(t testing.TB, expected, b []byte, fromIp, toIp netip.Addr, fromPort, toPort uint16) {
 	packet := gopacket.NewPacket(b, layers.LayerTypeIPv4, gopacket.Lazy)
 	v4 := packet.Layer(layers.LayerTypeIPv4).(*layers.IPv4)
 	assert.NotNil(t, v4, "No ipv4 data found")
@@ -337,24 +379,32 @@ func getAddrs(ns []netip.Prefix) []netip.Addr {
 	return a
 }
 
-func NewTestLogger() *logrus.Logger {
+func NewTestLogger() *slog.Logger {
-	l := logrus.New()
-
 	v := os.Getenv("TEST_LOGS")
 	if v == "" {
-		l.SetOutput(io.Discard)
+		return slog.New(slog.NewTextHandler(io.Discard, nil))
-		l.SetLevel(logrus.PanicLevel)
-		return l
 	}
 
+	level := slog.LevelInfo
 	switch v {
 	case "2":
-		l.SetLevel(logrus.DebugLevel)
+		level = slog.LevelDebug
 	case "3":
-		l.SetLevel(logrus.TraceLevel)
+		level = logging.LevelTrace
-	default:
+	}
-		l.SetLevel(logrus.InfoLevel)
+	return slog.New(slog.NewTextHandler(os.Stderr, &slog.HandlerOptions{Level: level}))
 }
 
-	return l
+// testLogLevelName returns the level name string accepted by logging.ApplyConfig
+// for the current TEST_LOGS setting. Kept in sync with NewTestLogger.
+func testLogLevelName() string {
+	switch os.Getenv("TEST_LOGS") {
+	case "2":
+		return "debug"
+	case "3":
+		return "trace"
+	case "":
+		return "info"
+	}
+	return "info"
 }

e2e/leak_test.go

@@ -0,0 +1,51 @@
+//go:build e2e_testing
+// +build e2e_testing
+
+package e2e
+
+import (
+	"testing"
+	"time"
+
+	"github.com/slackhq/nebula/cert"
+	"github.com/slackhq/nebula/cert_test"
+	"github.com/slackhq/nebula/e2e/router"
+	"go.uber.org/goleak"
+)
+
+// TestNoGoroutineLeaks brings up two nebula instances, completes a tunnel,
+// stops both, and asserts no goroutines leak past the shutdown. goleak's
+// retry mechanism gives the wg.Wait()-driven goroutines a moment to drain
+// before failing the assertion.
+//
+// IgnoreCurrent is necessary in the parallelized suite: other tests can
+// leave goroutines mid-shutdown when this one runs (Stop is async, the
+// wg.Wait() drain is not blocking on test return). We're checking that
+// *this* test's setup tears down cleanly, not that the whole suite is
+// idle at this moment. Intentionally NOT t.Parallel()'d for the same
+// reason — concurrent test goroutines would always show up.
+func TestNoGoroutineLeaks(t *testing.T) {
+	defer goleak.VerifyNone(t, goleak.IgnoreCurrent())
+
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	myControl, myVpnIpNet, myUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "me", "10.128.0.1/24", nil)
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "them", "10.128.0.2/24", nil)
+
+	myControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
+	theirControl.InjectLightHouseAddr(myVpnIpNet[0].Addr(), myUdpAddr)
+
+	myControl.Start()
+	theirControl.Start()
+
+	r := router.NewR(t, myControl, theirControl)
+	assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
+
+	myControl.Stop()
+	theirControl.Stop()
+	r.RenderFlow()
+
+	// Settle period: Stop() is non-blocking; the wg-driven goroutines need
+	// a moment to drain. goleak retries internally too, but a short explicit
+	// settle reduces flakes when the suite is busy.
+	time.Sleep(50 * time.Millisecond)
+}

e2e/tunnels_test.go

@@ -12,11 +12,14 @@ import (
 	"github.com/slackhq/nebula/cert"
 	"github.com/slackhq/nebula/cert_test"
 	"github.com/slackhq/nebula/e2e/router"
+	"github.com/slackhq/nebula/header"
+	"github.com/slackhq/nebula/udp"
 	"github.com/stretchr/testify/assert"
 	"gopkg.in/yaml.v3"
 )
 
 func TestDropInactiveTunnels(t *testing.T) {
+	t.Parallel()
 	// The goal of this test is to ensure the shortest inactivity timeout will close the tunnel on both sides
 	// under ideal conditions
 	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
@@ -61,6 +64,7 @@ func TestDropInactiveTunnels(t *testing.T) {
 }
 
 func TestCertUpgrade(t *testing.T) {
+	t.Parallel()
 	// The goal of this test is to ensure the shortest inactivity timeout will close the tunnel on both sides
 	// under ideal conditions
 	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
@@ -155,6 +159,7 @@ func TestCertUpgrade(t *testing.T) {
 }
 
 func TestCertDowngrade(t *testing.T) {
+	t.Parallel()
 	// The goal of this test is to ensure the shortest inactivity timeout will close the tunnel on both sides
 	// under ideal conditions
 	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
@@ -253,6 +258,7 @@ func TestCertDowngrade(t *testing.T) {
 }
 
 func TestCertMismatchCorrection(t *testing.T) {
+	t.Parallel()
 	// The goal of this test is to ensure the shortest inactivity timeout will close the tunnel on both sides
 	// under ideal conditions
 	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
@@ -318,3 +324,155 @@ func TestCertMismatchCorrection(t *testing.T) {
 	myControl.Stop()
 	theirControl.Stop()
 }
+
+func TestCrossStackRelaysWork(t *testing.T) {
+	t.Parallel()
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	myControl, myVpnIpNet, _, _ := newSimpleServer(cert.Version2, ca, caKey, "me     ", "10.128.0.1/24,fc00::1/64", m{"relay": m{"use_relays": true}})
+	relayControl, relayVpnIpNet, relayUdpAddr, _ := newSimpleServer(cert.Version2, ca, caKey, "relay  ", "10.128.0.128/24,fc00::128/64", m{"relay": m{"am_relay": true}})
+	theirUdp := netip.MustParseAddrPort("10.0.0.2:4242")
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServerWithUdp(cert.Version2, ca, caKey, "them   ", "fc00::2/64", theirUdp, m{"relay": m{"use_relays": true}})
+
+	//myVpnV4 := myVpnIpNet[0]
+	myVpnV6 := myVpnIpNet[1]
+	relayVpnV4 := relayVpnIpNet[0]
+	relayVpnV6 := relayVpnIpNet[1]
+	theirVpnV6 := theirVpnIpNet[0]
+
+	// Teach my how to get to the relay and that their can be reached via the relay
+	myControl.InjectLightHouseAddr(relayVpnV4.Addr(), relayUdpAddr)
+	myControl.InjectLightHouseAddr(relayVpnV6.Addr(), relayUdpAddr)
+	myControl.InjectRelays(theirVpnV6.Addr(), []netip.Addr{relayVpnV6.Addr()})
+	relayControl.InjectLightHouseAddr(theirVpnV6.Addr(), theirUdpAddr)
+
+	// Build a router so we don't have to reason who gets which packet
+	r := router.NewR(t, myControl, relayControl, theirControl)
+	defer r.RenderFlow()
+
+	// Start the servers
+	myControl.Start()
+	relayControl.Start()
+	theirControl.Start()
+
+	t.Log("Trigger a handshake from me to them via the relay")
+	myControl.InjectTunUDPPacket(theirVpnV6.Addr(), 80, myVpnV6.Addr(), 80, []byte("Hi from me"))
+
+	p := r.RouteForAllUntilTxTun(theirControl)
+	r.Log("Assert the tunnel works")
+	assertUdpPacket(t, []byte("Hi from me"), p, myVpnV6.Addr(), theirVpnV6.Addr(), 80, 80)
+
+	t.Log("reply?")
+	theirControl.InjectTunUDPPacket(myVpnV6.Addr(), 80, theirVpnV6.Addr(), 80, []byte("Hi from them"))
+	p = r.RouteForAllUntilTxTun(myControl)
+	assertUdpPacket(t, []byte("Hi from them"), p, theirVpnV6.Addr(), myVpnV6.Addr(), 80, 80)
+
+	r.RenderHostmaps("Final hostmaps", myControl, relayControl, theirControl)
+	//t.Log("finish up")
+	//myControl.Stop()
+	//theirControl.Stop()
+	//relayControl.Stop()
+}
+
+func TestCloseTunnelAuthenticated(t *testing.T) {
+	t.Parallel()
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	myControl, myVpnIpNet, myUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "me", "10.128.0.1/24", m{"tunnels": m{"drop_inactive": true, "inactivity_timeout": "5s"}})
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "them", "10.128.0.2/24", m{"tunnels": m{"drop_inactive": true, "inactivity_timeout": "10m"}})
+
+	// Share our underlay information
+	myControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
+	theirControl.InjectLightHouseAddr(myVpnIpNet[0].Addr(), myUdpAddr)
+
+	// Start the servers
+	myControl.Start()
+	theirControl.Start()
+
+	r := router.NewR(t, myControl, theirControl)
+
+	r.Log("Assert the tunnel between me and them works")
+	assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
+
+	r.Log("Close the tunnel")
+	myControl.CloseTunnel(theirVpnIpNet[0].Addr(), false)
+	r.FlushAll()
+
+	waitStart := time.Now()
+	for {
+		myIndexes := len(myControl.GetHostmap().Indexes)
+		theirIndexes := len(theirControl.GetHostmap().Indexes)
+		if myIndexes == 0 && theirIndexes == 0 {
+			break
+		}
+
+		since := time.Since(waitStart)
+		r.Logf("my tunnels: %v; their tunnels: %v; duration: %v", myIndexes, theirIndexes, since)
+		if since > time.Second*6 {
+			t.Fatal("Tunnel should have been declared inactive after 2 seconds and before 6 seconds")
+		}
+
+		time.Sleep(1 * time.Second)
+		//r.FlushAll()
+	}
+
+	r.Logf("Happy path success, tunnels were dropped within %v", time.Since(waitStart))
+
+	myControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
+	theirControl.InjectLightHouseAddr(myVpnIpNet[0].Addr(), myUdpAddr)
+	r.Log("Assert another tunnel between me and them works")
+	assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
+	hi := myControl.GetHostInfoByVpnAddr(theirVpnIpNet[0].Addr(), false)
+	if hi == nil {
+		t.Fatal("There is no hostinfo for this tunnel")
+	}
+	myHi := theirControl.GetHostInfoByVpnAddr(myVpnIpNet[0].Addr(), false)
+	if myHi == nil {
+		t.Fatal("There is no hostinfo for my tunnel")
+	}
+	r.Log("It does")
+
+	buf := make([]byte, 1024)
+	hdr := header.H{
+		Version:        1,
+		Type:           header.CloseTunnel,
+		Subtype:        0,
+		Reserved:       0,
+		RemoteIndex:    hi.RemoteIndex,
+		MessageCounter: 5,
+	}
+	out, err := hdr.Encode(buf)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	pkt := &udp.Packet{
+		To:   hi.CurrentRemote,
+		From: myHi.CurrentRemote,
+		Data: out,
+	}
+	r.InjectUDPPacket(myControl, theirControl, pkt)
+	r.Log("Injected bogus close tunnel. Let's see!")
+	waitStart = time.Now()
+	for {
+		myIndexes := len(myControl.GetHostmap().Indexes)
+		theirIndexes := len(theirControl.GetHostmap().Indexes)
+		if myIndexes == 0 {
+			t.Fatal("myIndexes should not be 0")
+		}
+		if theirIndexes == 0 {
+			t.Fatal("theirIndexes should not be 0, they should have rejected this bogus packet")
+		}
+
+		since := time.Since(waitStart)
+		r.Logf("my tunnels: %v; their tunnels: %v; duration: %v", myIndexes, theirIndexes, since)
+		if since > time.Second*4 {
+			t.Log("The tunnel would have been gone by now")
+			break
+		}
+
+		time.Sleep(1 * time.Second)
+		r.FlushAll()
+	}
+
+	myControl.Stop()
+	theirControl.Stop()
+}

examples/config.yml

@@ -144,6 +144,10 @@ listen:
   # valid values: always, never, private
   # This setting is reloadable.
   #send_recv_error: always
+  # Similar to send_recv_error, this option lets you configure if you want to accept "recv_error" packets from remote hosts.
+  # valid values: always, never, private
+  # This setting is reloadable.
+  #accept_recv_error: always
   # The so_sock option is a Linux-specific feature that allows all outgoing Nebula packets to be tagged with a specific identifier.
   # This tagging enables IP rule-based filtering. For example, it supports 0.0.0.0/0 unsafe_routes,
   # allowing for more precise routing decisions based on the packet tags. Default is 0 meaning no mark is set.
@@ -200,6 +204,12 @@ punchy:
   # Trusted SSH CA public keys. These are the public keys of the CAs that are allowed to sign SSH keys for access.
   #trusted_cas:
     #- "ssh public key string"
+  # sandbox_dir restricts file paths for profiling commands (start-cpu-profile, save-heap-profile,
+  # save-mutex-profile) to the specified directory. Relative paths will be resolved within this directory,
+  # and absolute paths outside of it will be rejected. Default is $TMP/nebula-debug.
+  # The directory is NOT automatically created.
+  # Overriding this to "" is the same as "/" and will allow overwriting any path on the host.
+  #sandbox_dir: /var/tmp/nebula-debug
 
 # EXPERIMENTAL: relay support for networks that can't establish direct connections.
 relay:
@@ -282,24 +292,21 @@ tun:
 
 # Configure logging level
 logging:
-  # panic, fatal, error, warning, info, or debug. Default is info and is reloadable.
+  # trace, debug, info, warn, or error. Default is info and is reloadable.
-  #NOTE: Debug mode can log remotely controlled/untrusted data which can quickly fill a disk in some
+  # fatal and panic are accepted for backwards compatibility and map to error.
-  # scenarios. Debug logging is also CPU intensive and will decrease performance overall.
+  #NOTE: Debug and trace modes can log remotely controlled/untrusted data which can quickly fill a disk in some
-  # Only enable debug logging while actively investigating an issue.
+  # scenarios. Debug and trace logging are also CPU intensive and will decrease performance overall.
+  # Only enable debug or trace logging while actively investigating an issue.
   level: info
-  # json or text formats currently available. Default is text
+  # json or text formats currently available. Default is text.
   format: text
-  # Disable timestamp logging. useful when output is redirected to logging system that already adds timestamps. Default is false
+  # Disable timestamp logging. Useful when output is redirected to a logging system that already adds timestamps. Default is false.
   #disable_timestamp: true
-  # timestamp format is specified in Go time format, see:
+  # Timestamps use RFC3339Nano ("2006-01-02T15:04:05.999999999Z07:00") and are not configurable.
-  #     https://golang.org/pkg/time/#pkg-constants
-  # default when `format: json`: "2006-01-02T15:04:05Z07:00" (RFC3339)
-  # default when `format: text`:
-  #     when TTY attached: seconds since beginning of execution
-  #     otherwise: "2006-01-02T15:04:05Z07:00" (RFC3339)
-  # As an example, to log as RFC3339 with millisecond precision, set to:
-  #timestamp_format: "2006-01-02T15:04:05.000Z07:00"
 
+# The stats section is reloadable. A HUP may change the backend, toggle stats
+# on or off, switch the listen/host address, or pick up new DNS for the
+# configured graphite host.
 #stats:
   #type: graphite
   #prefix: nebula
@@ -317,10 +324,12 @@ logging:
   # enables counter metrics for meta packets
   #   e.g.: `messages.tx.handshake`
   # NOTE: `message.{tx,rx}.recv_error` is always emitted
+  # Not reloadable.
   #message_metrics: false
 
   # enables detailed counter metrics for lighthouse packets
   #   e.g.: `lighthouse.rx.HostQuery`
+  # Not reloadable.
   #lighthouse_metrics: false
 
 # Handshake Manager Settings
@@ -378,13 +387,14 @@ firewall:
   # Rules are comprised of a protocol, port, and one or more of host, group, or CIDR
   # Logical evaluation is roughly: port AND proto AND (ca_sha OR ca_name) AND (host OR group OR groups OR cidr) AND (local cidr)
   # - port: Takes `0` or `any` as any, a single number `80`, a range `200-901`, or `fragment` to match second and further fragments of fragmented packets (since there is no port available).
-  #   code: same as port but makes more sense when talking about ICMP, TODO: this is not currently implemented in a way that works, use `any`
   #   proto: `any`, `tcp`, `udp`, or `icmp`
+  #          a port specification is ignored if proto is `icmp`
   #   host: `any` or a literal hostname, ie `test-host`
   #   group: `any` or a literal group name, ie `default-group`
   #   groups: Same as group but accepts a list of values. Multiple values are AND'd together and a certificate would have to contain all groups to pass
-  #   cidr: a remote CIDR, `0.0.0.0/0` is any ipv4 and `::/0` is any ipv6.
+  #   cidr: a remote CIDR, `0.0.0.0/0` is any ipv4 and `::/0` is any ipv6. `any` means any ip family and address.
-  #   local_cidr: a local CIDR, `0.0.0.0/0` is any ipv4 and `::/0` is any ipv6. This can be used to filter destinations when using unsafe_routes.
+  #   local_cidr: a local CIDR, `0.0.0.0/0` is any ipv4 and `::/0` is any ipv6. `any` means any ip family and address.
+  #     This can be used to filter destinations when using unsafe_routes.
   #     By default, this is set to only the VPN (overlay) networks assigned via the certificate networks field unless `default_local_cidr_any` is set to true.
   #     If there are unsafe_routes present in this config file, `local_cidr` should be set appropriately for the intended us case.
   #   ca_name: An issuing CA name

examples/go_service/main.go

@@ -7,9 +7,9 @@ import (
 	"net"
 	"os"
 
-	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula"
 	"github.com/slackhq/nebula/config"
+	"github.com/slackhq/nebula/logging"
 	"github.com/slackhq/nebula/overlay"
 	"github.com/slackhq/nebula/service"
 )
@@ -64,8 +64,7 @@ pki:
 		return err
 	}
 
-	logger := logrus.New()
+	logger := logging.NewLogger(os.Stdout)
-	logger.Out = os.Stdout
 
 	ctrl, err := nebula.Main(&cfg, false, "custom-app", logger, overlay.NewUserDeviceFromConfig)
 	if err != nil {

firewall.go

@@ -1,13 +1,16 @@
 package nebula
 
 import (
+	"context"
 	"crypto/sha256"
 	"encoding/hex"
 	"errors"
 	"fmt"
 	"hash/fnv"
+	"log/slog"
 	"net/netip"
 	"reflect"
+	"slices"
 	"strconv"
 	"strings"
 	"sync"
@@ -15,14 +18,13 @@ import (
 
 	"github.com/gaissmai/bart"
 	"github.com/rcrowley/go-metrics"
-	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula/cert"
 	"github.com/slackhq/nebula/config"
 	"github.com/slackhq/nebula/firewall"
 )
 
 type FirewallInterface interface {
-	AddRule(incoming bool, proto uint8, startPort int32, endPort int32, groups []string, host string, addr, localAddr netip.Prefix, caName string, caSha string) error
+	AddRule(incoming bool, proto uint8, startPort int32, endPort int32, groups []string, host string, cidr, localCidr string, caName string, caSha string) error
 }
 
 type conn struct {
@@ -66,7 +68,7 @@ type Firewall struct {
 	incomingMetrics     firewallMetrics
 	outgoingMetrics     firewallMetrics
 
-	l *logrus.Logger
+	l *slog.Logger
 }
 
 type firewallMetrics struct {
@@ -130,7 +132,7 @@ type firewallLocalCIDR struct {
 
 // NewFirewall creates a new Firewall object. A TimerWheel is created for you from the provided timeouts.
 // The certificate provided should be the highest version loaded in memory.
-func NewFirewall(l *logrus.Logger, tcpTimeout, UDPTimeout, defaultTimeout time.Duration, c cert.Certificate) *Firewall {
+func NewFirewall(l *slog.Logger, tcpTimeout, UDPTimeout, defaultTimeout time.Duration, c cert.Certificate) *Firewall {
 	//TODO: error on 0 duration
 	var tmin, tmax time.Duration
 
@@ -190,7 +192,7 @@ func NewFirewall(l *logrus.Logger, tcpTimeout, UDPTimeout, defaultTimeout time.D
 	}
 }
 
-func NewFirewallFromConfig(l *logrus.Logger, cs *CertState, c *config.C) (*Firewall, error) {
+func NewFirewallFromConfig(l *slog.Logger, cs *CertState, c *config.C) (*Firewall, error) {
 	certificate := cs.getCertificate(cert.Version2)
 	if certificate == nil {
 		certificate = cs.getCertificate(cert.Version1)
@@ -218,7 +220,7 @@ func NewFirewallFromConfig(l *logrus.Logger, cs *CertState, c *config.C) (*Firew
 	case "drop":
 		fw.InSendReject = false
 	default:
-		l.WithField("action", inboundAction).Warn("invalid firewall.inbound_action, defaulting to `drop`")
+		l.Warn("invalid firewall.inbound_action, defaulting to `drop`", "action", inboundAction)
 		fw.InSendReject = false
 	}
 
@@ -229,7 +231,7 @@ func NewFirewallFromConfig(l *logrus.Logger, cs *CertState, c *config.C) (*Firew
 	case "drop":
 		fw.OutSendReject = false
 	default:
-		l.WithField("action", inboundAction).Warn("invalid firewall.outbound_action, defaulting to `drop`")
+		l.Warn("invalid firewall.outbound_action, defaulting to `drop`", "action", outboundAction)
 		fw.OutSendReject = false
 	}
 
@@ -247,32 +249,7 @@ func NewFirewallFromConfig(l *logrus.Logger, cs *CertState, c *config.C) (*Firew
 }
 
 // AddRule properly creates the in memory rule structure for a firewall table.
-func (f *Firewall) AddRule(incoming bool, proto uint8, startPort int32, endPort int32, groups []string, host string, ip, localIp netip.Prefix, caName string, caSha string) error {
+func (f *Firewall) AddRule(incoming bool, proto uint8, startPort int32, endPort int32, groups []string, host string, cidr, localCidr, caName string, caSha string) error {
-	// Under gomobile, stringing a nil pointer with fmt causes an abort in debug mode for iOS
-	// https://github.com/golang/go/issues/14131
-	sIp := ""
-	if ip.IsValid() {
-		sIp = ip.String()
-	}
-	lIp := ""
-	if localIp.IsValid() {
-		lIp = localIp.String()
-	}
-
-	// We need this rule string because we generate a hash. Removing this will break firewall reload.
-	ruleString := fmt.Sprintf(
-		"incoming: %v, proto: %v, startPort: %v, endPort: %v, groups: %v, host: %v, ip: %v, localIp: %v, caName: %v, caSha: %s",
-		incoming, proto, startPort, endPort, groups, host, sIp, lIp, caName, caSha,
-	)
-	f.rules += ruleString + "\n"
-
-	direction := "incoming"
-	if !incoming {
-		direction = "outgoing"
-	}
-	f.l.WithField("firewallRule", m{"direction": direction, "proto": proto, "startPort": startPort, "endPort": endPort, "groups": groups, "host": host, "ip": sIp, "localIp": lIp, "caName": caName, "caSha": caSha}).
-		Info("Firewall rule added")
-
 	var (
 		ft *FirewallTable
 		fp firewallPort
@@ -290,6 +267,12 @@ func (f *Firewall) AddRule(incoming bool, proto uint8, startPort int32, endPort
 	case firewall.ProtoUDP:
 		fp = ft.UDP
 	case firewall.ProtoICMP, firewall.ProtoICMPv6:
+		//ICMP traffic doesn't have ports, so we always coerce to "any", even if a value is provided
+		if startPort != firewall.PortAny {
+			f.l.Warn("ignoring port specification for ICMP firewall rule", "startPort", startPort)
+		}
+		startPort = firewall.PortAny
+		endPort = firewall.PortAny
 		fp = ft.ICMP
 	case firewall.ProtoAny:
 		fp = ft.AnyProto
@@ -297,7 +280,22 @@ func (f *Firewall) AddRule(incoming bool, proto uint8, startPort int32, endPort
 		return fmt.Errorf("unknown protocol %v", proto)
 	}
 
-	return fp.addRule(f, startPort, endPort, groups, host, ip, localIp, caName, caSha)
+	// We need this rule string because we generate a hash. Removing this will break firewall reload.
+	ruleString := fmt.Sprintf(
+		"incoming: %v, proto: %v, startPort: %v, endPort: %v, groups: %v, host: %v, ip: %v, localIp: %v, caName: %v, caSha: %s",
+		incoming, proto, startPort, endPort, groups, host, cidr, localCidr, caName, caSha,
+	)
+	f.rules += ruleString + "\n"
+
+	direction := "incoming"
+	if !incoming {
+		direction = "outgoing"
+	}
+	f.l.Info("Firewall rule added",
+		"firewallRule", m{"direction": direction, "proto": proto, "startPort": startPort, "endPort": endPort, "groups": groups, "host": host, "cidr": cidr, "localCidr": localCidr, "caName": caName, "caSha": caSha},
+	)
+
+	return fp.addRule(f, startPort, endPort, groups, host, cidr, localCidr, caName, caSha)
 }
 
 // GetRuleHash returns a hash representation of all inbound and outbound rules
@@ -318,7 +316,7 @@ func (f *Firewall) GetRuleHashes() string {
 	return "SHA:" + f.GetRuleHash() + ",FNV:" + strconv.FormatUint(uint64(f.GetRuleHashFNV()), 10)
 }
 
-func AddFirewallRulesFromConfig(l *logrus.Logger, inbound bool, c *config.C, fw FirewallInterface) error {
+func AddFirewallRulesFromConfig(l *slog.Logger, inbound bool, c *config.C, fw FirewallInterface) error {
 	var table string
 	if inbound {
 		table = "firewall.inbound"
@@ -337,7 +335,6 @@ func AddFirewallRulesFromConfig(l *logrus.Logger, inbound bool, c *config.C, fw
 	}
 
 	for i, t := range rs {
-		var groups []string
 		r, err := convertRule(l, t, table, i)
 		if err != nil {
 			return fmt.Errorf("%s rule #%v; %s", table, i, err)
@@ -347,23 +344,10 @@ func AddFirewallRulesFromConfig(l *logrus.Logger, inbound bool, c *config.C, fw
 			return fmt.Errorf("%s rule #%v; only one of port or code should be provided", table, i)
 		}
 
-		if r.Host == "" && len(r.Groups) == 0 && r.Group == "" && r.Cidr == "" && r.LocalCidr == "" && r.CAName == "" && r.CASha == "" {
+		if r.Host == "" && len(r.Groups) == 0 && r.Cidr == "" && r.LocalCidr == "" && r.CAName == "" && r.CASha == "" {
 			return fmt.Errorf("%s rule #%v; at least one of host, group, cidr, local_cidr, ca_name, or ca_sha must be provided", table, i)
 		}
 
-		if len(r.Groups) > 0 {
-			groups = r.Groups
-		}
-
-		if r.Group != "" {
-			// Check if we have both groups and group provided in the rule config
-			if len(groups) > 0 {
-				return fmt.Errorf("%s rule #%v; only one of group or groups should be defined, both provided", table, i)
-			}
-
-			groups = []string{r.Group}
-		}
-
 		var sPort, errPort string
 		if r.Code != "" {
 			errPort = "code"
@@ -373,42 +357,55 @@ func AddFirewallRulesFromConfig(l *logrus.Logger, inbound bool, c *config.C, fw
 			sPort = r.Port
 		}
 
-		startPort, endPort, err := parsePort(sPort)
+		var proto uint8
+		var startPort, endPort int32
+		switch r.Proto {
+		case "any":
+			proto = firewall.ProtoAny
+			startPort, endPort, err = parsePort(sPort)
+		case "tcp":
+			proto = firewall.ProtoTCP
+			startPort, endPort, err = parsePort(sPort)
+		case "udp":
+			proto = firewall.ProtoUDP
+			startPort, endPort, err = parsePort(sPort)
+		case "icmp":
+			proto = firewall.ProtoICMP
+			startPort = firewall.PortAny
+			endPort = firewall.PortAny
+			if sPort != "" {
+				l.Warn("ignoring port specification for ICMP firewall rule", "port", sPort)
+			}
+		default:
+			return fmt.Errorf("%s rule #%v; proto was not understood; `%s`", table, i, r.Proto)
+		}
 		if err != nil {
 			return fmt.Errorf("%s rule #%v; %s %s", table, i, errPort, err)
 		}
 
-		var proto uint8
+		if r.Cidr != "" && r.Cidr != "any" {
-		switch r.Proto {
+			_, err = netip.ParsePrefix(r.Cidr)
-		case "any":
-			proto = firewall.ProtoAny
-		case "tcp":
-			proto = firewall.ProtoTCP
-		case "udp":
-			proto = firewall.ProtoUDP
-		case "icmp":
-			proto = firewall.ProtoICMP
-		default:
-			return fmt.Errorf("%s rule #%v; proto was not understood; `%s`", table, i, r.Proto)
-		}
-
-		var cidr netip.Prefix
-		if r.Cidr != "" {
-			cidr, err = netip.ParsePrefix(r.Cidr)
 			if err != nil {
 				return fmt.Errorf("%s rule #%v; cidr did not parse; %s", table, i, err)
 			}
 		}
 
-		var localCidr netip.Prefix
+		if r.LocalCidr != "" && r.LocalCidr != "any" {
-		if r.LocalCidr != "" {
+			_, err = netip.ParsePrefix(r.LocalCidr)
-			localCidr, err = netip.ParsePrefix(r.LocalCidr)
 			if err != nil {
 				return fmt.Errorf("%s rule #%v; local_cidr did not parse; %s", table, i, err)
 			}
 		}
 
-		err = fw.AddRule(inbound, proto, startPort, endPort, groups, r.Host, cidr, localCidr, r.CAName, r.CASha)
+		if warning := r.sanity(); warning != nil {
+			l.Warn("firewall rule sanity check",
+				"table", table,
+				"rule", i,
+				"warning", warning,
+			)
+		}
+
+		err = fw.AddRule(inbound, proto, startPort, endPort, r.Groups, r.Host, r.Cidr, r.LocalCidr, r.CAName, r.CASha)
 		if err != nil {
 			return fmt.Errorf("%s rule #%v; `%s`", table, i, err)
 		}
@@ -417,30 +414,45 @@ func AddFirewallRulesFromConfig(l *logrus.Logger, inbound bool, c *config.C, fw
 	return nil
 }
 
-var ErrInvalidRemoteIP = errors.New("remote IP is not in remote certificate subnets")
+var ErrUnknownNetworkType = errors.New("unknown network type")
-var ErrInvalidLocalIP = errors.New("local IP is not in list of handled local IPs")
+var ErrPeerRejected = errors.New("remote address is not within a network that we handle")
+var ErrInvalidRemoteIP = errors.New("remote address is not in remote certificate networks")
+var ErrInvalidLocalIP = errors.New("local address is not in list of handled local addresses")
 var ErrNoMatchingRule = errors.New("no matching rule in firewall table")
 
 // Drop returns an error if the packet should be dropped, explaining why. It
 // returns nil if the packet should not be dropped.
-func (f *Firewall) Drop(fp firewall.Packet, incoming bool, h *HostInfo, caPool *cert.CAPool, localCache *firewall.ConntrackCache) error {
+func (f *Firewall) Drop(fp firewall.Packet, incoming bool, h *HostInfo, caPool *cert.CAPool, localCache firewall.ConntrackCache) error {
 	// Check if we spoke to this tuple, if we did then allow this packet
 	if f.inConns(fp, h, caPool, localCache) {
 		return nil
 	}
 
-	// Make sure remote address matches nebula certificate
+	// Make sure remote address matches nebula certificate, and determine how to treat it
-	if h.networks != nil {
+	if h.networks == nil {
-		if !h.networks.Contains(fp.RemoteAddr) {
-			f.metrics(incoming).droppedRemoteAddr.Inc(1)
-			return ErrInvalidRemoteIP
-		}
-	} else {
 		// Simple case: Certificate has one address and no unsafe networks
 		if h.vpnAddrs[0] != fp.RemoteAddr {
 			f.metrics(incoming).droppedRemoteAddr.Inc(1)
 			return ErrInvalidRemoteIP
 		}
+	} else {
+		nwType, ok := h.networks.Lookup(fp.RemoteAddr)
+		if !ok {
+			f.metrics(incoming).droppedRemoteAddr.Inc(1)
+			return ErrInvalidRemoteIP
+		}
+		switch nwType {
+		case NetworkTypeVPN:
+			break // nothing special
+		case NetworkTypeVPNPeer:
+			f.metrics(incoming).droppedRemoteAddr.Inc(1)
+			return ErrPeerRejected // reject for now, one day this may have different FW rules
+		case NetworkTypeUnsafe:
+			break // nothing special, one day this may have different FW rules
+		default:
+			f.metrics(incoming).droppedRemoteAddr.Inc(1)
+			return ErrUnknownNetworkType //should never happen
+		}
 	}
 
 	// Make sure we are supposed to be handling this local ip address
@@ -474,7 +486,7 @@ func (f *Firewall) metrics(incoming bool) firewallMetrics {
 	}
 }
 
-// Destroy cleans up any known cyclical references so the object can be free'd my GC. This should be called if a new
+// Destroy cleans up any known cyclical references so the object can be freed by GC. This should be called if a new
 // firewall object is created
 func (f *Firewall) Destroy() {
 	//TODO: clean references if/when needed
@@ -490,10 +502,12 @@ func (f *Firewall) EmitStats() {
 	metrics.GetOrRegisterGauge("firewall.rules.hash", nil).Update(int64(f.GetRuleHashFNV()))
 }
 
-func (f *Firewall) inConns(fp firewall.Packet, h *HostInfo, caPool *cert.CAPool, localCache *firewall.ConntrackCache) bool {
+func (f *Firewall) inConns(fp firewall.Packet, h *HostInfo, caPool *cert.CAPool, localCache firewall.ConntrackCache) bool {
-	if localCache != nil && localCache.Has(fp) {
+	if localCache != nil {
+		if _, ok := localCache[fp]; ok {
 			return true
 		}
+	}
 	conntrack := f.Conntrack
 	conntrack.Lock()
 
@@ -520,26 +534,26 @@ func (f *Firewall) inConns(fp firewall.Packet, h *HostInfo, caPool *cert.CAPool,
 
 		// We now know which firewall table to check against
 		if !table.match(fp, c.incoming, h.ConnectionState.peerCert, caPool) {
-			if f.l.Level >= logrus.DebugLevel {
+			if f.l.Enabled(context.Background(), slog.LevelDebug) {
-				h.logger(f.l).
+				h.logger(f.l).Debug("dropping old conntrack entry, does not match new ruleset",
-					WithField("fwPacket", fp).
+					"fwPacket", fp,
-					WithField("incoming", c.incoming).
+					"incoming", c.incoming,
-					WithField("rulesVersion", f.rulesVersion).
+					"rulesVersion", f.rulesVersion,
-					WithField("oldRulesVersion", c.rulesVersion).
+					"oldRulesVersion", c.rulesVersion,
-					Debugln("dropping old conntrack entry, does not match new ruleset")
+				)
 			}
 			delete(conntrack.Conns, fp)
 			conntrack.Unlock()
 			return false
 		}
 
-		if f.l.Level >= logrus.DebugLevel {
+		if f.l.Enabled(context.Background(), slog.LevelDebug) {
-			h.logger(f.l).
+			h.logger(f.l).Debug("keeping old conntrack entry, does match new ruleset",
-				WithField("fwPacket", fp).
+				"fwPacket", fp,
-				WithField("incoming", c.incoming).
+				"incoming", c.incoming,
-				WithField("rulesVersion", f.rulesVersion).
+				"rulesVersion", f.rulesVersion,
-				WithField("oldRulesVersion", c.rulesVersion).
+				"oldRulesVersion", c.rulesVersion,
-				Debugln("keeping old conntrack entry, does match new ruleset")
+			)
 		}
 
 		c.rulesVersion = f.rulesVersion
@@ -557,7 +571,7 @@ func (f *Firewall) inConns(fp firewall.Packet, h *HostInfo, caPool *cert.CAPool,
 	conntrack.Unlock()
 
 	if localCache != nil {
-		localCache.Add(fp)
+		localCache[fp] = struct{}{}
 	}
 
 	return true
@@ -638,7 +652,7 @@ func (ft *FirewallTable) match(p firewall.Packet, incoming bool, c *cert.CachedC
 	return false
 }
 
-func (fp firewallPort) addRule(f *Firewall, startPort int32, endPort int32, groups []string, host string, ip, localIp netip.Prefix, caName string, caSha string) error {
+func (fp firewallPort) addRule(f *Firewall, startPort int32, endPort int32, groups []string, host string, cidr, localCidr, caName string, caSha string) error {
 	if startPort > endPort {
 		return fmt.Errorf("start port was lower than end port")
 	}
@@ -651,7 +665,7 @@ func (fp firewallPort) addRule(f *Firewall, startPort int32, endPort int32, grou
 			}
 		}
 
-		if err := fp[i].addRule(f, groups, host, ip, localIp, caName, caSha); err != nil {
+		if err := fp[i].addRule(f, groups, host, cidr, localCidr, caName, caSha); err != nil {
 			return err
 		}
 	}
@@ -665,6 +679,13 @@ func (fp firewallPort) match(p firewall.Packet, incoming bool, c *cert.CachedCer
 		return false
 	}
 
+	// this branch is here to catch traffic from FirewallTable.Any.match and FirewallTable.ICMP.match
+	if p.Protocol == firewall.ProtoICMP || p.Protocol == firewall.ProtoICMPv6 {
+		// port numbers are re-used for connection tracking of ICMP,
+		// but we don't want to actually filter on them.
+		return fp[firewall.PortAny].match(p, c, caPool)
+	}
+
 	var port int32
 
 	if p.Fragment {
@@ -682,7 +703,7 @@ func (fp firewallPort) match(p firewall.Packet, incoming bool, c *cert.CachedCer
 	return fp[firewall.PortAny].match(p, c, caPool)
 }
 
-func (fc *FirewallCA) addRule(f *Firewall, groups []string, host string, ip, localIp netip.Prefix, caName, caSha string) error {
+func (fc *FirewallCA) addRule(f *Firewall, groups []string, host string, cidr, localCidr, caName, caSha string) error {
 	fr := func() *FirewallRule {
 		return &FirewallRule{
 			Hosts:  make(map[string]*firewallLocalCIDR),
@@ -696,14 +717,14 @@ func (fc *FirewallCA) addRule(f *Firewall, groups []string, host string, ip, loc
 			fc.Any = fr()
 		}
 
-		return fc.Any.addRule(f, groups, host, ip, localIp)
+		return fc.Any.addRule(f, groups, host, cidr, localCidr)
 	}
 
 	if caSha != "" {
 		if _, ok := fc.CAShas[caSha]; !ok {
 			fc.CAShas[caSha] = fr()
 		}
-		err := fc.CAShas[caSha].addRule(f, groups, host, ip, localIp)
+		err := fc.CAShas[caSha].addRule(f, groups, host, cidr, localCidr)
 		if err != nil {
 			return err
 		}
@@ -713,7 +734,7 @@ func (fc *FirewallCA) addRule(f *Firewall, groups []string, host string, ip, loc
 		if _, ok := fc.CANames[caName]; !ok {
 			fc.CANames[caName] = fr()
 		}
-		err := fc.CANames[caName].addRule(f, groups, host, ip, localIp)
+		err := fc.CANames[caName].addRule(f, groups, host, cidr, localCidr)
 		if err != nil {
 			return err
 		}
@@ -745,24 +766,24 @@ func (fc *FirewallCA) match(p firewall.Packet, c *cert.CachedCertificate, caPool
 	return fc.CANames[s.Certificate.Name()].match(p, c)
 }
 
-func (fr *FirewallRule) addRule(f *Firewall, groups []string, host string, ip, localCIDR netip.Prefix) error {
+func (fr *FirewallRule) addRule(f *Firewall, groups []string, host, cidr, localCidr string) error {
 	flc := func() *firewallLocalCIDR {
 		return &firewallLocalCIDR{
 			LocalCIDR: new(bart.Lite),
 		}
 	}
 
-	if fr.isAny(groups, host, ip) {
+	if fr.isAny(groups, host, cidr) {
 		if fr.Any == nil {
 			fr.Any = flc()
 		}
 
-		return fr.Any.addRule(f, localCIDR)
+		return fr.Any.addRule(f, localCidr)
 	}
 
 	if len(groups) > 0 {
 		nlc := flc()
-		err := nlc.addRule(f, localCIDR)
+		err := nlc.addRule(f, localCidr)
 		if err != nil {
 			return err
 		}
@@ -778,44 +799,46 @@ func (fr *FirewallRule) addRule(f *Firewall, groups []string, host string, ip, l
 		if nlc == nil {
 			nlc = flc()
 		}
-		err := nlc.addRule(f, localCIDR)
+		err := nlc.addRule(f, localCidr)
 		if err != nil {
 			return err
 		}
 		fr.Hosts[host] = nlc
 	}
 
-	if ip.IsValid() {
+	if cidr != "" {
-		nlc, _ := fr.CIDR.Get(ip)
+		c, err := netip.ParsePrefix(cidr)
-		if nlc == nil {
-			nlc = flc()
-		}
-		err := nlc.addRule(f, localCIDR)
 		if err != nil {
 			return err
 		}
-		fr.CIDR.Insert(ip, nlc)
+		nlc, _ := fr.CIDR.Get(c)
+		if nlc == nil {
+			nlc = flc()
+		}
+		err = nlc.addRule(f, localCidr)
+		if err != nil {
+			return err
+		}
+		fr.CIDR.Insert(c, nlc)
 	}
 
 	return nil
 }
 
-func (fr *FirewallRule) isAny(groups []string, host string, ip netip.Prefix) bool {
+func (fr *FirewallRule) isAny(groups []string, host string, cidr string) bool {
-	if len(groups) == 0 && host == "" && !ip.IsValid() {
+	if len(groups) == 0 && host == "" && cidr == "" {
 		return true
 	}
 
-	for _, group := range groups {
+	if slices.Contains(groups, "any") {
-		if group == "any" {
 		return true
 	}
-	}
 
 	if host == "any" {
 		return true
 	}
 
-	if ip.IsValid() && ip.Bits() == 0 {
+	if cidr == "any" {
 		return true
 	}
 
@@ -867,8 +890,13 @@ func (fr *FirewallRule) match(p firewall.Packet, c *cert.CachedCertificate) bool
 	return false
 }
 
-func (flc *firewallLocalCIDR) addRule(f *Firewall, localIp netip.Prefix) error {
+func (flc *firewallLocalCIDR) addRule(f *Firewall, localCidr string) error {
-	if !localIp.IsValid() {
+	if localCidr == "any" {
+		flc.Any = true
+		return nil
+	}
+
+	if localCidr == "" {
 		if !f.hasUnsafeNetworks || f.defaultLocalCIDRAny {
 			flc.Any = true
 			return nil
@@ -879,12 +907,13 @@ func (flc *firewallLocalCIDR) addRule(f *Firewall, localIp netip.Prefix) error {
 		}
 		return nil
 
-	} else if localIp.Bits() == 0 {
-		flc.Any = true
-		return nil
 	}
 
-	flc.LocalCIDR.Insert(localIp)
+	c, err := netip.ParsePrefix(localCidr)
+	if err != nil {
+		return err
+	}
+	flc.LocalCIDR.Insert(c)
 	return nil
 }
 
@@ -905,7 +934,6 @@ type rule struct {
 	Code      string
 	Proto     string
 	Host      string
-	Group     string
 	Groups    []string
 	Cidr      string
 	LocalCidr string
@@ -913,7 +941,7 @@ type rule struct {
 	CASha     string
 }
 
-func convertRule(l *logrus.Logger, p any, table string, i int) (rule, error) {
+func convertRule(l *slog.Logger, p any, table string, i int) (rule, error) {
 	r := rule{}
 
 	m, ok := p.(map[string]any)
@@ -944,10 +972,14 @@ func convertRule(l *logrus.Logger, p any, table string, i int) (rule, error) {
 			return r, errors.New("group should contain a single value, an array with more than one entry was provided")
 		}
 
-		l.Warnf("%s rule #%v; group was an array with a single value, converting to simple value", table, i)
+		l.Warn("group was an array with a single value, converting to simple value",
+			"table", table,
+			"rule", i,
+		)
 		m["group"] = v[0]
 	}
-	r.Group = toString("group", m)
+
+	singleGroup := toString("group", m)
 
 	if rg, ok := m["groups"]; ok {
 		switch reflect.TypeOf(rg).Kind() {
@@ -964,52 +996,105 @@ func convertRule(l *logrus.Logger, p any, table string, i int) (rule, error) {
 		}
 	}
 
+	//flatten group vs groups
+	if singleGroup != "" {
+		// Check if we have both groups and group provided in the rule config
+		if len(r.Groups) > 0 {
+			return r, fmt.Errorf("only one of group or groups should be defined, both provided")
+		}
+		r.Groups = []string{singleGroup}
+	}
+
 	return r, nil
 }
 
-func parsePort(s string) (startPort, endPort int32, err error) {
+// sanity returns an error if the rule would be evaluated in a way that would short-circuit a configured check on a wildcard value
+// rules are evaluated as "port AND proto AND (ca_sha OR ca_name) AND (host OR group OR groups OR cidr) AND local_cidr"
+func (r *rule) sanity() error {
+	//port, proto, local_cidr are AND, no need to check here
+	//ca_sha and ca_name don't have a wildcard value, no need to check here
+	groupsEmpty := len(r.Groups) == 0
+	hostEmpty := r.Host == ""
+	cidrEmpty := r.Cidr == ""
+
+	if (groupsEmpty && hostEmpty && cidrEmpty) == true {
+		return nil //no content!
+	}
+
+	groupsHasAny := slices.Contains(r.Groups, "any")
+	if groupsHasAny && len(r.Groups) > 1 {
+		return fmt.Errorf("groups spec [%s] contains the group '\"any\". This rule will ignore the other groups specified", r.Groups)
+	}
+
+	if r.Host == "any" {
+		if !groupsEmpty {
+			return fmt.Errorf("groups specified as %s, but host=any will match any host, regardless of groups", r.Groups)
+		}
+
+		if !cidrEmpty {
+			return fmt.Errorf("cidr specified as %s, but host=any will match any host, regardless of cidr", r.Cidr)
+		}
+	}
+
+	if groupsHasAny {
+		if !hostEmpty && r.Host != "any" {
+			return fmt.Errorf("groups spec [%s] contains the group '\"any\". This rule will ignore the specified host %s", r.Groups, r.Host)
+		}
+		if !cidrEmpty {
+			return fmt.Errorf("groups spec [%s] contains the group '\"any\". This rule will ignore the specified cidr %s", r.Groups, r.Cidr)
+		}
+	}
+
+	if r.Code != "" {
+		return fmt.Errorf("code specified as [%s]. Support for 'code' will be dropped in a future release, as it has never been functional", r.Code)
+	}
+
+	//todo alert on cidr-any
+
+	return nil
+}
+
+func parsePort(s string) (int32, int32, error) {
+	var err error
+	const notAPort int32 = -2
 	if s == "any" {
-		startPort = firewall.PortAny
+		return firewall.PortAny, firewall.PortAny, nil
-		endPort = firewall.PortAny
+	}
+	if s == "fragment" {
+		return firewall.PortFragment, firewall.PortFragment, nil
+	}
+	if !strings.Contains(s, `-`) {
+		rPort, err := strconv.Atoi(s)
+		if err != nil {
+			return notAPort, notAPort, fmt.Errorf("was not a number; `%s`", s)
+		}
+		return int32(rPort), int32(rPort), nil
+	}
 
-	} else if s == "fragment" {
-		startPort = firewall.PortFragment
-		endPort = firewall.PortFragment
-
-	} else if strings.Contains(s, `-`) {
 	sPorts := strings.SplitN(s, `-`, 2)
-		sPorts[0] = strings.Trim(sPorts[0], " ")
+	for i := range sPorts {
-		sPorts[1] = strings.Trim(sPorts[1], " ")
+		sPorts[i] = strings.Trim(sPorts[i], " ")
-
+	}
 	if len(sPorts) != 2 || sPorts[0] == "" || sPorts[1] == "" {
-			return 0, 0, fmt.Errorf("appears to be a range but could not be parsed; `%s`", s)
+		return notAPort, notAPort, fmt.Errorf("appears to be a range but could not be parsed; `%s`", s)
 	}
 
 	rStartPort, err := strconv.Atoi(sPorts[0])
 	if err != nil {
-			return 0, 0, fmt.Errorf("beginning range was not a number; `%s`", sPorts[0])
+		return notAPort, notAPort, fmt.Errorf("beginning range was not a number; `%s`", sPorts[0])
 	}
 
 	rEndPort, err := strconv.Atoi(sPorts[1])
 	if err != nil {
-			return 0, 0, fmt.Errorf("ending range was not a number; `%s`", sPorts[1])
+		return notAPort, notAPort, fmt.Errorf("ending range was not a number; `%s`", sPorts[1])
 	}
 
-		startPort = int32(rStartPort)
+	startPort := int32(rStartPort)
-		endPort = int32(rEndPort)
+	endPort := int32(rEndPort)
 
 	if startPort == firewall.PortAny {
 		endPort = firewall.PortAny
 	}
 
-	} else {
+	return startPort, endPort, nil
-		rPort, err := strconv.Atoi(s)
-		if err != nil {
-			return 0, 0, fmt.Errorf("was not a number; `%s`", s)
-		}
-		startPort = int32(rPort)
-		endPort = startPort
-	}
-
-	return
 }

firewall/cache.go

@@ -1,101 +1,65 @@
 package firewall
 
 import (
-	"sync"
+	"context"
+	"log/slog"
 	"sync/atomic"
 	"time"
-
-	"github.com/sirupsen/logrus"
 )
 
 // ConntrackCache is used as a local routine cache to know if a given flow
 // has been seen in the conntrack table.
-type ConntrackCache struct {
+type ConntrackCache map[Packet]struct{}
-	mu      sync.Mutex
-	entries map[Packet]struct{}
-}
-
-func newConntrackCache() *ConntrackCache {
-	return &ConntrackCache{entries: make(map[Packet]struct{})}
-}
-
-func (c *ConntrackCache) Has(p Packet) bool {
-	if c == nil {
-		return false
-	}
-	c.mu.Lock()
-	_, ok := c.entries[p]
-	c.mu.Unlock()
-	return ok
-}
-
-func (c *ConntrackCache) Add(p Packet) {
-	if c == nil {
-		return
-	}
-	c.mu.Lock()
-	c.entries[p] = struct{}{}
-	c.mu.Unlock()
-}
-
-func (c *ConntrackCache) Len() int {
-	if c == nil {
-		return 0
-	}
-	c.mu.Lock()
-	l := len(c.entries)
-	c.mu.Unlock()
-	return l
-}
-
-func (c *ConntrackCache) Reset(capHint int) {
-	if c == nil {
-		return
-	}
-	c.mu.Lock()
-	c.entries = make(map[Packet]struct{}, capHint)
-	c.mu.Unlock()
-}
 
 type ConntrackCacheTicker struct {
 	cacheV    uint64
 	cacheTick atomic.Uint64
 
-	cache *ConntrackCache
+	l     *slog.Logger
+	cache ConntrackCache
 }
 
-func NewConntrackCacheTicker(d time.Duration) *ConntrackCacheTicker {
+func NewConntrackCacheTicker(ctx context.Context, l *slog.Logger, d time.Duration) *ConntrackCacheTicker {
 	if d == 0 {
 		return nil
 	}
 
-	c := &ConntrackCacheTicker{cache: newConntrackCache()}
+	c := &ConntrackCacheTicker{
+		l:     l,
+		cache: ConntrackCache{},
+	}
 
-	go c.tick(d)
+	go c.tick(ctx, d)
 
 	return c
 }
 
-func (c *ConntrackCacheTicker) tick(d time.Duration) {
+func (c *ConntrackCacheTicker) tick(ctx context.Context, d time.Duration) {
+	t := time.NewTicker(d)
+	defer t.Stop()
 	for {
-		time.Sleep(d)
+		select {
+		case <-ctx.Done():
+			return
+		case <-t.C:
 			c.cacheTick.Add(1)
 		}
 	}
+}
 
 // Get checks if the cache ticker has moved to the next version before returning
 // the map. If it has moved, we reset the map.
-func (c *ConntrackCacheTicker) Get(l *logrus.Logger) *ConntrackCache {
+func (c *ConntrackCacheTicker) Get() ConntrackCache {
 	if c == nil {
 		return nil
 	}
 	if tick := c.cacheTick.Load(); tick != c.cacheV {
 		c.cacheV = tick
-		if ll := c.cache.Len(); ll > 0 {
+		if ll := len(c.cache); ll > 0 {
-			if l.Level == logrus.DebugLevel {
+			if c.l.Enabled(context.Background(), slog.LevelDebug) {
-				l.WithField("len", ll).Debug("resetting conntrack cache")
+				c.l.Debug("resetting conntrack cache", "len", ll)
 			}
-			c.cache.Reset(ll)
+			c.cache = make(ConntrackCache, ll)
 		}
 	}
 

firewall/cache_test.go

@@ -0,0 +1,69 @@
+package firewall
+
+import (
+	"bytes"
+	"log/slog"
+	"strings"
+	"testing"
+
+	"github.com/slackhq/nebula/test"
+	"github.com/stretchr/testify/assert"
+)
+
+// The tests below pin the log format produced by ConntrackCacheTicker.Get
+// so changes cannot silently break what operators are grepping for. The
+// ticker's internal state (cache + cacheTick) is poked directly to avoid
+// racing a goroutine-driven tick in tests.
+
+func newFixedTicker(t *testing.T, l *slog.Logger, cacheLen int) *ConntrackCacheTicker {
+	t.Helper()
+	c := &ConntrackCacheTicker{
+		l:     l,
+		cache: make(ConntrackCache, cacheLen),
+	}
+	for i := 0; i < cacheLen; i++ {
+		c.cache[Packet{LocalPort: uint16(i) + 1}] = struct{}{}
+	}
+	c.cacheTick.Store(1) // cacheV starts at 0, so Get() takes the reset path
+	return c
+}
+
+func TestConntrackCacheTicker_Get_TextFormat(t *testing.T) {
+	buf := &bytes.Buffer{}
+	l := test.NewLoggerWithOutputAndLevel(buf, slog.LevelDebug)
+
+	c := newFixedTicker(t, l, 3)
+	c.Get()
+
+	assert.Equal(t, "level=DEBUG msg=\"resetting conntrack cache\" len=3\n", buf.String())
+}
+
+func TestConntrackCacheTicker_Get_JSONFormat(t *testing.T) {
+	buf := &bytes.Buffer{}
+	l := test.NewJSONLoggerWithOutput(buf, slog.LevelDebug)
+
+	c := newFixedTicker(t, l, 2)
+	c.Get()
+
+	assert.JSONEq(t, `{"level":"DEBUG","msg":"resetting conntrack cache","len":2}`, strings.TrimSpace(buf.String()))
+}
+
+func TestConntrackCacheTicker_Get_QuietBelowDebug(t *testing.T) {
+	buf := &bytes.Buffer{}
+	l := test.NewLoggerWithOutputAndLevel(buf, slog.LevelInfo)
+
+	c := newFixedTicker(t, l, 5)
+	c.Get()
+
+	assert.Empty(t, buf.String())
+}
+
+func TestConntrackCacheTicker_Get_QuietWhenCacheEmpty(t *testing.T) {
+	buf := &bytes.Buffer{}
+	l := test.NewLoggerWithOutputAndLevel(buf, slog.LevelDebug)
+
+	c := newFixedTicker(t, l, 0)
+	c.Get()
+
+	assert.Empty(t, buf.String())
+}

firewall/packet.go

@@ -22,7 +22,10 @@ const (
 type Packet struct {
 	LocalAddr  netip.Addr
 	RemoteAddr netip.Addr
+	// LocalPort is the destination port for incoming traffic, or the source port for outgoing. Zero for ICMP.
 	LocalPort uint16
+	// RemotePort is the source port for incoming traffic, or the destination port for outgoing.
+	// For ICMP, it's the "identifier". This is only used for connection tracking, actual firewall rules will not filter on ICMP identifier
 	RemotePort uint16
 	Protocol   uint8
 	Fragment   bool
@@ -46,6 +49,8 @@ func (fp Packet) MarshalJSON() ([]byte, error) {
 		proto = "tcp"
 	case ProtoICMP:
 		proto = "icmp"
+	case ProtoICMPv6:
+		proto = "icmpv6"
 	case ProtoUDP:
 		proto = "udp"
 	default:

go.mod

@@ -1,38 +1,39 @@
 module github.com/slackhq/nebula
 
-go 1.25
+go 1.25.0
 
 require (
 	dario.cat/mergo v1.0.2
+	filippo.io/bigmod v0.1.0
 	github.com/anmitsu/go-shlex v0.0.0-20200514113438-38f4b401e2be
 	github.com/armon/go-radix v1.0.0
-	github.com/cilium/ebpf v0.12.3
 	github.com/cyberdelia/go-metrics-graphite v0.0.0-20161219230853-39f87cc3b432
 	github.com/flynn/noise v1.1.0
-	github.com/gaissmai/bart v0.25.0
+	github.com/gaissmai/bart v0.26.0
 	github.com/gogo/protobuf v1.3.2
 	github.com/google/gopacket v1.1.19
 	github.com/kardianos/service v1.2.4
-	github.com/miekg/dns v1.1.68
+	github.com/miekg/dns v1.1.72
-	github.com/miekg/pkcs11 v1.1.2-0.20231115102856-9078ad6b9d4b
+	github.com/miekg/pkcs11 v1.1.2
 	github.com/nbrownus/go-metrics-prometheus v0.0.0-20210712211119-974a6260965f
 	github.com/prometheus/client_golang v1.23.2
 	github.com/rcrowley/go-metrics v0.0.0-20201227073835-cf1acfcdf475
-	github.com/sirupsen/logrus v1.9.3
 	github.com/skip2/go-qrcode v0.0.0-20200617195104-da1b6568686e
 	github.com/stefanberger/go-pkcs11uri v0.0.0-20230803200340-78284954bff6
 	github.com/stretchr/testify v1.11.1
 	github.com/vishvananda/netlink v1.3.1
-	golang.org/x/crypto v0.43.0
+	go.uber.org/goleak v1.3.0
+	go.yaml.in/yaml/v3 v3.0.4
+	golang.org/x/crypto v0.50.0
 	golang.org/x/exp v0.0.0-20230725093048-515e97ebf090
-	golang.org/x/net v0.45.0
+	golang.org/x/net v0.52.0
-	golang.org/x/sync v0.17.0
+	golang.org/x/sync v0.20.0
-	golang.org/x/sys v0.37.0
+	golang.org/x/sys v0.43.0
-	golang.org/x/term v0.36.0
+	golang.org/x/term v0.42.0
 	golang.zx2c4.com/wintun v0.0.0-20230126152724-0fa3db229ce2
 	golang.zx2c4.com/wireguard v0.0.0-20230325221338-052af4a8072b
-	golang.zx2c4.com/wireguard/windows v0.5.3
+	golang.zx2c4.com/wireguard/windows v0.6.1
-	google.golang.org/protobuf v1.36.8
+	google.golang.org/protobuf v1.36.11
 	gopkg.in/yaml.v3 v3.0.1
 	gvisor.dev/gvisor v0.0.0-20240423190808-9d7a357edefe
 )
@@ -49,7 +50,7 @@ require (
 	github.com/prometheus/procfs v0.16.1 // indirect
 	github.com/vishvananda/netns v0.0.5 // indirect
 	go.yaml.in/yaml/v2 v2.4.2 // indirect
-	golang.org/x/mod v0.24.0 // indirect
+	golang.org/x/mod v0.34.0 // indirect
 	golang.org/x/time v0.5.0 // indirect
-	golang.org/x/tools v0.33.0 // indirect
+	golang.org/x/tools v0.43.0 // indirect
 )

go.sum

@@ -1,6 +1,8 @@
 cloud.google.com/go v0.34.0/go.mod h1:aQUYkXzVsufM+DwF1aE+0xfcU+56JwCaLick0ClmMTw=
 dario.cat/mergo v1.0.2 h1:85+piFYR1tMbRrLcDwR18y4UKJ3aH1Tbzi24VRW1TK8=
 dario.cat/mergo v1.0.2/go.mod h1:E/hbnu0NxMFBjpMIE34DRGLWqDy0g5FuKDhCb31ngxA=
+filippo.io/bigmod v0.1.0 h1:UNzDk7y9ADKST+axd9skUpBQeW7fG2KrTZyOE4uGQy8=
+filippo.io/bigmod v0.1.0/go.mod h1:OjOXDNlClLblvXdwgFFOQFJEocLhhtai8vGLy0JCZlI=
 github.com/alecthomas/template v0.0.0-20160405071501-a0175ee3bccc/go.mod h1:LOuyumcjzFXgccqObfd/Ljyb9UuFJ6TxHnclSeseNhc=
 github.com/alecthomas/template v0.0.0-20190718012654-fb15b899a751/go.mod h1:LOuyumcjzFXgccqObfd/Ljyb9UuFJ6TxHnclSeseNhc=
 github.com/alecthomas/units v0.0.0-20151022065526-2efee857e7cf/go.mod h1:ybxpYRFXyAe+OPACYpWeL0wqObRcbAqCMya13uyzqw0=
@@ -17,8 +19,6 @@ github.com/beorn7/perks v1.0.1/go.mod h1:G2ZrVWU2WbWT9wwq4/hrbKbnv/1ERSJQ0ibhJ6r
 github.com/cespare/xxhash/v2 v2.1.1/go.mod h1:VGX0DQ3Q6kWi7AoAeZDth3/j3BFtOZR5XLFGgcrjCOs=
 github.com/cespare/xxhash/v2 v2.3.0 h1:UL815xU9SqsFlibzuggzjXhog7bL6oX9BbNZnL2UFvs=
 github.com/cespare/xxhash/v2 v2.3.0/go.mod h1:VGX0DQ3Q6kWi7AoAeZDth3/j3BFtOZR5XLFGgcrjCOs=
-github.com/cilium/ebpf v0.12.3 h1:8ht6F9MquybnY97at+VDZb3eQQr8ev79RueWeVaEcG4=
-github.com/cilium/ebpf v0.12.3/go.mod h1:TctK1ivibvI3znr66ljgi4hqOT8EYQjz1KWBfb1UVgM=
 github.com/cyberdelia/go-metrics-graphite v0.0.0-20161219230853-39f87cc3b432 h1:M5QgkYacWj0Xs8MhpIK/5uwU02icXpEoSo9sM2aRCps=
 github.com/cyberdelia/go-metrics-graphite v0.0.0-20161219230853-39f87cc3b432/go.mod h1:xwIwAxMvYnVrGJPe2FKx5prTrnAjGOD8zvDOnxnrrkM=
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
@@ -26,10 +26,8 @@ github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c
 github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/flynn/noise v1.1.0 h1:KjPQoQCEFdZDiP03phOvGi11+SVVhBG2wOWAorLsstg=
 github.com/flynn/noise v1.1.0/go.mod h1:xbMo+0i6+IGbYdJhF31t2eR1BIU0CYc12+BNAKwUTag=
-github.com/frankban/quicktest v1.14.5 h1:dfYrrRyLtiqT9GyKXgdh+k4inNeTvmGbuSgZ3lx3GhA=
+github.com/gaissmai/bart v0.26.0 h1:xOZ57E9hJLBiQaSyeZa9wgWhGuzfGACgqp4BE77OkO0=
-github.com/frankban/quicktest v1.14.5/go.mod h1:4ptaffx2x8+WTWXmUCuVU6aPUX1/Mz7zb5vbUoiM6w0=
+github.com/gaissmai/bart v0.26.0/go.mod h1:GREWQfTLRWz/c5FTOsIw+KkscuFkIV5t8Rp7Nd1Td5c=
-github.com/gaissmai/bart v0.25.0 h1:eqiokVPqM3F94vJ0bTHXHtH91S8zkKL+bKh+BsGOsJM=
-github.com/gaissmai/bart v0.25.0/go.mod h1:GREWQfTLRWz/c5FTOsIw+KkscuFkIV5t8Rp7Nd1Td5c=
 github.com/go-kit/kit v0.8.0/go.mod h1:xBxKIO96dXMWWy0MnWVtmwkA9/13aqxPnvrjFYMA2as=
 github.com/go-kit/kit v0.9.0/go.mod h1:xBxKIO96dXMWWy0MnWVtmwkA9/13aqxPnvrjFYMA2as=
 github.com/go-kit/log v0.1.0/go.mod h1:zbhenjAZHb184qTLMA9ZjW7ThYL0H2mk7Q6pNt4vbaY=
@@ -82,16 +80,15 @@ github.com/kr/pretty v0.2.1/go.mod h1:ipq/a2n7PKx3OHsz4KJII5eveXtPO4qwEXGdVfWzfn
 github.com/kr/pretty v0.3.1 h1:flRD4NNwYAUpkphVc1HcthR4KEIFJ65n8Mw5qdRn3LE=
 github.com/kr/pretty v0.3.1/go.mod h1:hoEshYVHaxMs3cyo3Yncou5ZscifuDolrwPKZanG3xk=
 github.com/kr/pty v1.1.1/go.mod h1:pFQYn66WHrOpPYNljwOMqo10TkYh1fy3cYio2l3bCsQ=
+github.com/kr/text v0.1.0 h1:45sCR5RtlFHMR4UwH9sdQ5TC8v0qDQCHnXt+kaKSTVE=
 github.com/kr/text v0.1.0/go.mod h1:4Jbv+DJW3UT/LiOwJeYQe1efqtUx/iVham/4vfdArNI=
-github.com/kr/text v0.2.0 h1:5Nx0Ya0ZqY2ygV366QzturHI13Jq95ApcVaJBhpS+AY=
-github.com/kr/text v0.2.0/go.mod h1:eLer722TekiGuMkidMxC/pM04lWEeraHUUmBw8l2grE=
 github.com/kylelemons/godebug v1.1.0 h1:RPNrshWIDI6G2gRW9EHilWtl7Z6Sb1BR0xunSBf0SNc=
 github.com/kylelemons/godebug v1.1.0/go.mod h1:9/0rRGxNHcop5bhtWyNeEfOS8JIWk580+fNqagV/RAw=
 github.com/matttproud/golang_protobuf_extensions v1.0.1/go.mod h1:D8He9yQNgCq6Z5Ld7szi9bcBfOoFv/3dc6xSMkL2PC0=
-github.com/miekg/dns v1.1.68 h1:jsSRkNozw7G/mnmXULynzMNIsgY2dHC8LO6U6Ij2JEA=
+github.com/miekg/dns v1.1.72 h1:vhmr+TF2A3tuoGNkLDFK9zi36F2LS+hKTRW0Uf8kbzI=
-github.com/miekg/dns v1.1.68/go.mod h1:fujopn7TB3Pu3JM69XaawiU0wqjpL9/8xGop5UrTPps=
+github.com/miekg/dns v1.1.72/go.mod h1:+EuEPhdHOsfk6Wk5TT2CzssZdqkmFhf8r+aVyDEToIs=
-github.com/miekg/pkcs11 v1.1.2-0.20231115102856-9078ad6b9d4b h1:J/AzCvg5z0Hn1rqZUJjpbzALUmkKX0Zwbc/i4fw7Sfk=
+github.com/miekg/pkcs11 v1.1.2 h1:/VxmeAX5qU6Q3EwafypogwWbYryHFmF2RpkJmw3m4MQ=
-github.com/miekg/pkcs11 v1.1.2-0.20231115102856-9078ad6b9d4b/go.mod h1:XsNlhZGX73bx86s2hdc/FuaLm2CPZJemRLMA+WTFxgs=
+github.com/miekg/pkcs11 v1.1.2/go.mod h1:XsNlhZGX73bx86s2hdc/FuaLm2CPZJemRLMA+WTFxgs=
 github.com/modern-go/concurrent v0.0.0-20180228061459-e0a39a4cb421/go.mod h1:6dJC0mAP4ikYIbvyc7fijjWJddQyLn8Ig3JB5CqoB9Q=
 github.com/modern-go/concurrent v0.0.0-20180306012644-bacd9c7ef1dd/go.mod h1:6dJC0mAP4ikYIbvyc7fijjWJddQyLn8Ig3JB5CqoB9Q=
 github.com/modern-go/reflect2 v0.0.0-20180701023420-4b7aa43c6742/go.mod h1:bx2lNnkwVCuqBIxFjflWJWanXIb3RllmbCylyMrvgv0=
@@ -136,8 +133,6 @@ github.com/rogpeppe/go-internal v1.10.0/go.mod h1:UQnix2H7Ngw/k4C5ijL5+65zddjncj
 github.com/sirupsen/logrus v1.2.0/go.mod h1:LxeOpSwHxABJmUn/MG1IvRgCAasNZTLOkJPxbbu5VWo=
 github.com/sirupsen/logrus v1.4.2/go.mod h1:tLMulIdttU9McNUspp0xgXVQah82FyeX6MwdIuYE2rE=
 github.com/sirupsen/logrus v1.6.0/go.mod h1:7uNnSEd1DgxDLC74fIahvMZmmYsHGZGEOFrfsX/uA88=
-github.com/sirupsen/logrus v1.9.3 h1:dueUQJ1C2q9oE3F7wvmSGAaVtTmUizReu6fjN8uqzbQ=
-github.com/sirupsen/logrus v1.9.3/go.mod h1:naHLuLoDiP4jHNo9R0sCBMtWGeIprob74mVsIT4qYEQ=
 github.com/skip2/go-qrcode v0.0.0-20200617195104-da1b6568686e h1:MRM5ITcdelLK2j1vwZ3Je0FKVCfqOLp5zO6trqMLYs0=
 github.com/skip2/go-qrcode v0.0.0-20200617195104-da1b6568686e/go.mod h1:XV66xRDqSt+GTGFMVlhk3ULuV0y9ZmzeVGR4mloJI3M=
 github.com/stefanberger/go-pkcs11uri v0.0.0-20230803200340-78284954bff6 h1:pnnLyeX7o/5aX8qUQ69P/mLojDqwda8hFOCBTmP/6hw=
@@ -160,21 +155,23 @@ go.uber.org/goleak v1.3.0 h1:2K3zAYmnTNqV73imy9J1T3WC+gmCePx2hEGkimedGto=
 go.uber.org/goleak v1.3.0/go.mod h1:CoHD4mav9JJNrW/WLlf7HGZPjdw8EucARQHekz1X6bE=
 go.yaml.in/yaml/v2 v2.4.2 h1:DzmwEr2rDGHl7lsFgAHxmNz/1NlQ7xLIrlN2h5d1eGI=
 go.yaml.in/yaml/v2 v2.4.2/go.mod h1:081UH+NErpNdqlCXm3TtEran0rJZGxAYx9hb/ELlsPU=
+go.yaml.in/yaml/v3 v3.0.4 h1:tfq32ie2Jv2UxXFdLJdh3jXuOzWiL1fo0bu/FbuKpbc=
+go.yaml.in/yaml/v3 v3.0.4/go.mod h1:DhzuOOF2ATzADvBadXxruRBLzYTpT36CKvDb3+aBEFg=
 golang.org/x/crypto v0.0.0-20180904163835-0709b304e793/go.mod h1:6SG95UA2DQfeDnfUPMdvaQW0Q7yPrPDi9nlGo2tz2b4=
 golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACkg1iLfiJU5Ep61QUkGW8qpdssI0+w=
 golang.org/x/crypto v0.0.0-20191011191535-87dc89f01550/go.mod h1:yigFU9vqHzYiE8UmvKecakEJjdnWj3jj499lnFckfCI=
 golang.org/x/crypto v0.0.0-20200622213623-75b288015ac9/go.mod h1:LzIPMQfyMNhhGPhUkYOs5KpL4U8rLKemX1yGLhDgUto=
 golang.org/x/crypto v0.0.0-20210322153248-0c34fe9e7dc2/go.mod h1:T9bdIzuCu7OtxOm1hfPfRQxPLYneinmdGuTeoZ9dtd4=
-golang.org/x/crypto v0.43.0 h1:dduJYIi3A3KOfdGOHX8AVZ/jGiyPa3IbBozJ5kNuE04=
+golang.org/x/crypto v0.50.0 h1:zO47/JPrL6vsNkINmLoo/PH1gcxpls50DNogFvB5ZGI=
-golang.org/x/crypto v0.43.0/go.mod h1:BFbav4mRNlXJL4wNeejLpWxB7wMbc79PdRGhWKncxR0=
+golang.org/x/crypto v0.50.0/go.mod h1:3muZ7vA7PBCE6xgPX7nkzzjiUq87kRItoJQM1Yo8S+Q=
 golang.org/x/exp v0.0.0-20230725093048-515e97ebf090 h1:Di6/M8l0O2lCLc6VVRWhgCiApHV8MnQurBnFSHsQtNY=
 golang.org/x/exp v0.0.0-20230725093048-515e97ebf090/go.mod h1:FXUEEKJgO7OQYeo8N01OfiKP8RXMtf6e8aTskBGqWdc=
 golang.org/x/lint v0.0.0-20200302205851-738671d3881b/go.mod h1:3xt1FjdF8hUf6vQPIChWIBhFzV8gjjsPE/fR3IyQdNY=
 golang.org/x/mod v0.1.1-0.20191105210325-c90efee705ee/go.mod h1:QqPTAvyqsEbceGzBzNggFXnrqF1CaUcvgkdR5Ot7KZg=
 golang.org/x/mod v0.2.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
 golang.org/x/mod v0.3.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
-golang.org/x/mod v0.24.0 h1:ZfthKaKaT4NrhGVZHO1/WDTwGES4De8KtWO0SIbNJMU=
+golang.org/x/mod v0.34.0 h1:xIHgNUUnW6sYkcM5Jleh05DvLOtwc6RitGHbDk4akRI=
-golang.org/x/mod v0.24.0/go.mod h1:IXM97Txy2VM4PJ3gI61r1YEk/gAj6zAHN3AdZt6S9Ww=
+golang.org/x/mod v0.34.0/go.mod h1:ykgH52iCZe79kzLLMhyCUzhMci+nQj+0XkbXpNYtVjY=
 golang.org/x/net v0.0.0-20180724234803-3673e40ba225/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
 golang.org/x/net v0.0.0-20181114220301-adae6a3d119a/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
 golang.org/x/net v0.0.0-20190108225652-1e06a53dbb7e/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
@@ -185,8 +182,8 @@ golang.org/x/net v0.0.0-20200226121028-0de0cce0169b/go.mod h1:z5CRVTTTmAJ677TzLL
 golang.org/x/net v0.0.0-20200625001655-4c5254603344/go.mod h1:/O7V0waA8r7cgGh81Ro3o1hOxt32SMVPicZroKQ2sZA=
 golang.org/x/net v0.0.0-20201021035429-f5854403a974/go.mod h1:sp8m0HH+o8qH0wwXwYZr8TS3Oi6o0r6Gce1SSxlDquU=
 golang.org/x/net v0.0.0-20210226172049-e18ecbb05110/go.mod h1:m0MpNAwzfU5UDzcl9v0D8zg8gWTRqZa9RBIspLL5mdg=
-golang.org/x/net v0.45.0 h1:RLBg5JKixCy82FtLJpeNlVM0nrSqpCRYzVU1n8kj0tM=
+golang.org/x/net v0.52.0 h1:He/TN1l0e4mmR3QqHMT2Xab3Aj3L9qjbhRm78/6jrW0=
-golang.org/x/net v0.45.0/go.mod h1:ECOoLqd5U3Lhyeyo/QDCEVQ4sNgYsqvCZ722XogGieY=
+golang.org/x/net v0.52.0/go.mod h1:R1MAz7uMZxVMualyPXb+VaqGSa3LIaUqk0eEt3w36Sw=
 golang.org/x/oauth2 v0.0.0-20190226205417-e64efc72b421/go.mod h1:gOpvHmFTYa4IltrdGE7lF6nIHvwfUNPOp7c8zoXwtLw=
 golang.org/x/sync v0.0.0-20181108010431-42b317875d0f/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
 golang.org/x/sync v0.0.0-20181221193216-37e7f081c4d4/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
@@ -194,8 +191,8 @@ golang.org/x/sync v0.0.0-20190423024810-112230192c58/go.mod h1:RxMgew5VJxzue5/jJ
 golang.org/x/sync v0.0.0-20190911185100-cd5d95a43a6e/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
 golang.org/x/sync v0.0.0-20201020160332-67f06af15bc9/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
 golang.org/x/sync v0.0.0-20201207232520-09787c993a3a/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
-golang.org/x/sync v0.17.0 h1:l60nONMj9l5drqw6jlhIELNv9I0A4OFgRsG9k2oT9Ug=
+golang.org/x/sync v0.20.0 h1:e0PTpb7pjO8GAtTs2dQ6jYa5BWYlMuX047Dco/pItO4=
-golang.org/x/sync v0.17.0/go.mod h1:9KTHXmSnoGruLpwFjVSX0lNNA75CykiMECbovNTZqGI=
+golang.org/x/sync v0.20.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
 golang.org/x/sys v0.0.0-20180905080454-ebe1bf3edb33/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
 golang.org/x/sys v0.0.0-20181116152217-5ac8a444bdc5/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
 golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
@@ -209,14 +206,13 @@ golang.org/x/sys v0.0.0-20200930185726-fdedc70b468f/go.mod h1:h1NjWce9XRLGQEsW7w
 golang.org/x/sys v0.0.0-20201119102817-f84b799fce68/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20210124154548-22da62e12c0c/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20210603081109-ebe580a85c40/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.0.0-20220715151400-c0bba94af5f8/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.2.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.10.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.37.0 h1:fdNQudmxPjkdUTPnLn5mdQv7Zwvbvpaxqs831goi9kQ=
+golang.org/x/sys v0.43.0 h1:Rlag2XtaFTxp19wS8MXlJwTvoh8ArU6ezoyFsMyCTNI=
-golang.org/x/sys v0.37.0/go.mod h1:OgkHotnGiDImocRcuBABYBEXf8A9a87e/uXjp9XT3ks=
+golang.org/x/sys v0.43.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
 golang.org/x/term v0.0.0-20201126162022-7de9c90e9dd1/go.mod h1:bj7SfCRtBDWHUb9snDiAeCFNEtKQo2Wmx5Cou7ajbmo=
-golang.org/x/term v0.36.0 h1:zMPR+aF8gfksFprF/Nc/rd1wRS1EI6nDBGyWAvDzx2Q=
+golang.org/x/term v0.42.0 h1:UiKe+zDFmJobeJ5ggPwOshJIVt6/Ft0rcfrXZDLWAWY=
-golang.org/x/term v0.36.0/go.mod h1:Qu394IJq6V6dCBRgwqshf3mPF85AqzYEzofzRdZkWss=
+golang.org/x/term v0.42.0/go.mod h1:Dq/D+snpsbazcBG5+F9Q1n2rXV8Ma+71xEjTRufARgY=
 golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
 golang.org/x/text v0.3.2/go.mod h1:bEr9sfX3Q8Zfm5fL9x+3itogRgK3+ptLWKqgva+5dAk=
 golang.org/x/text v0.3.3/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
@@ -227,8 +223,8 @@ golang.org/x/tools v0.0.0-20191119224855-298f0cb1881e/go.mod h1:b+2E5dAYhXwXZwtn
 golang.org/x/tools v0.0.0-20200130002326-2f3ba24bd6e7/go.mod h1:TB2adYChydJhpapKDTa4BR/hXlZSLoq2Wpct/0txZ28=
 golang.org/x/tools v0.0.0-20200619180055-7c47624df98f/go.mod h1:EkVYQZoAsY45+roYkvgYkIh4xh/qjgUK9TdY2XT94GE=
 golang.org/x/tools v0.0.0-20210106214847-113979e3529a/go.mod h1:emZCQorbCU4vsT4fOWvOPXz4eW1wZW4PmDk9uLelYpA=
-golang.org/x/tools v0.33.0 h1:4qz2S3zmRxbGIhDIAgjxvFutSvH5EfnsYrRBj0UI0bc=
+golang.org/x/tools v0.43.0 h1:12BdW9CeB3Z+J/I/wj34VMl8X+fEXBxVR90JeMX5E7s=
-golang.org/x/tools v0.33.0/go.mod h1:CIJMaWEY88juyUfo7UbgPqbC8rU2OqfAV1h2Qp0oMYI=
+golang.org/x/tools v0.43.0/go.mod h1:uHkMso649BX2cZK6+RpuIPXS3ho2hZo4FVwfoy1vIk0=
 golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
@@ -237,8 +233,8 @@ golang.zx2c4.com/wintun v0.0.0-20230126152724-0fa3db229ce2 h1:B82qJJgjvYKsXS9jeu
 golang.zx2c4.com/wintun v0.0.0-20230126152724-0fa3db229ce2/go.mod h1:deeaetjYA+DHMHg+sMSMI58GrEteJUUzzw7en6TJQcI=
 golang.zx2c4.com/wireguard v0.0.0-20230325221338-052af4a8072b h1:J1CaxgLerRR5lgx3wnr6L04cJFbWoceSK9JWBdglINo=
 golang.zx2c4.com/wireguard v0.0.0-20230325221338-052af4a8072b/go.mod h1:tqur9LnfstdR9ep2LaJT4lFUl0EjlHtge+gAjmsHUG4=
-golang.zx2c4.com/wireguard/windows v0.5.3 h1:On6j2Rpn3OEMXqBq00QEDC7bWSZrPIHKIus8eIuExIE=
+golang.zx2c4.com/wireguard/windows v0.6.1 h1:XMaKojH1Hs/raMrmnir4n35nTvzvWj7NmSYzHn2F4qU=
-golang.zx2c4.com/wireguard/windows v0.5.3/go.mod h1:9TEe8TJmtwyQebdFwAkEWOPr3prrtqm+REGFifP60hI=
+golang.zx2c4.com/wireguard/windows v0.6.1/go.mod h1:04aqInu5GYuTFvMuDw/rKBAF7mHrltW/3rekpfbbZDM=
 google.golang.org/appengine v1.4.0/go.mod h1:xpcJRLb0r/rnEns0DIKYYv+WjYCduHsrkT7/EB5XEv4=
 google.golang.org/protobuf v0.0.0-20200109180630-ec00e32a8dfd/go.mod h1:DFci5gLYBciE7Vtevhsrf46CRTquxDuWsQurQQe4oz8=
 google.golang.org/protobuf v0.0.0-20200221191635-4d8936d0db64/go.mod h1:kwYJMbMJ01Woi6D6+Kah6886xMZcty6N08ah7+eCXa0=
@@ -247,8 +243,8 @@ google.golang.org/protobuf v1.20.1-0.20200309200217-e05f789c0967/go.mod h1:A+miE
 google.golang.org/protobuf v1.21.0/go.mod h1:47Nbq4nVaFHyn7ilMalzfO3qCViNmqZ2kzikPIcrTAo=
 google.golang.org/protobuf v1.23.0/go.mod h1:EGpADcykh3NcUnDUJcl1+ZksZNG86OlYog2l/sGQquU=
 google.golang.org/protobuf v1.26.0-rc.1/go.mod h1:jlhhOSvTdKEhbULTjvd4ARK9grFBp09yW+WbY/TyQbw=
-google.golang.org/protobuf v1.36.8 h1:xHScyCOEuuwZEc6UtSOvPbAT4zRh0xcNRYekJwfqyMc=
+google.golang.org/protobuf v1.36.11 h1:fV6ZwhNocDyBLK0dj+fg8ektcVegBBuEolpbTQyBNVE=
-google.golang.org/protobuf v1.36.8/go.mod h1:fuxRtAxBytpl4zzqUh6/eyUujkJdNiuEkXntxiD/uRU=
+google.golang.org/protobuf v1.36.11/go.mod h1:HTf+CrKn2C3g5S8VImy6tdcUvCska2kB7j23XfzDpco=
 gopkg.in/alecthomas/kingpin.v2 v2.2.6/go.mod h1:FMv+mEhP44yOT+4EoQTLFTRgOQ1FBLkstjWtayDeSgw=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
 gopkg.in/check.v1 v1.0.0-20190902080502-41f04d3bba15/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=

handshake/credential.go

@@ -0,0 +1,57 @@
+package handshake
+
+import (
+	"crypto/rand"
+
+	"github.com/flynn/noise"
+	"github.com/slackhq/nebula/cert"
+)
+
+// Credential holds everything needed to participate in a handshake
+// at a given cert version. Version and Curve are read from Cert; the public
+// half of the static keypair likewise comes from Cert.PublicKey().
+type Credential struct {
+	Cert        cert.Certificate  // the certificate
+	Bytes       []byte            // pre-marshaled certificate bytes
+	privateKey  []byte            // static private key (public half lives in Cert)
+	cipherSuite noise.CipherSuite // pre-built cipher suite (DH + cipher + hash)
+}
+
+// NewCredential creates a Credential with all material needed for handshake
+// participation. The cipherSuite should be pre-built by the caller with the
+// appropriate DH function, cipher, and hash.
+func NewCredential(
+	c cert.Certificate,
+	hsBytes []byte,
+	privateKey []byte,
+	cipherSuite noise.CipherSuite,
+) *Credential {
+	return &Credential{
+		Cert:        c,
+		Bytes:       hsBytes,
+		privateKey:  privateKey,
+		cipherSuite: cipherSuite,
+	}
+}
+
+// buildHandshakeState creates a noise.HandshakeState from this credential.
+func (hc *Credential) buildHandshakeState(initiator bool, pattern noise.HandshakePattern) (*noise.HandshakeState, error) {
+	return noise.NewHandshakeState(noise.Config{
+		CipherSuite:           hc.cipherSuite,
+		Random:                rand.Reader,
+		Pattern:               pattern,
+		Initiator:             initiator,
+		StaticKeypair:         noise.DHKey{Private: hc.privateKey, Public: hc.Cert.PublicKey()},
+		PresharedKey:          []byte{},
+		PresharedKeyPlacement: 0,
+	})
+}
+
+// GetCredentialFunc returns the handshake credential for the given version,
+// or nil if that version is not available.
+//
+// Implementations must return credentials drawn from a snapshot stable for
+// the lifetime of any single Machine. The Machine may call this multiple
+// times during a handshake (e.g. when negotiating to the peer's version)
+// and assumes the underlying static keypair is consistent across calls.
+type GetCredentialFunc func(v cert.Version) *Credential

handshake/errors.go

@@ -0,0 +1,21 @@
+package handshake
+
+import "errors"
+
+var (
+	ErrInitiateOnResponder     = errors.New("initiate called on responder")
+	ErrInitiateAlreadyCalled   = errors.New("initiate already called")
+	ErrInitiateNotCalled       = errors.New("initiate must be called before ProcessPacket for initiators")
+	ErrPacketTooShort          = errors.New("packet too short")
+	ErrPublicKeyMismatch       = errors.New("public key mismatch between certificate and handshake")
+	ErrIncompleteHandshake     = errors.New("handshake completed without receiving required content")
+	ErrMachineFailed           = errors.New("handshake machine has failed")
+	ErrUnknownSubtype          = errors.New("unknown handshake subtype")
+	ErrMissingContent          = errors.New("expected handshake content but message was empty")
+	ErrUnexpectedContent       = errors.New("received unexpected handshake content")
+	ErrIndexAllocation         = errors.New("failed to allocate local index")
+	ErrNoCredential            = errors.New("no handshake credential available for cert version")
+	ErrAsymmetricCipherKeys    = errors.New("noise produced only one cipher key")
+	ErrMultiMessageUnsupported = errors.New("multi-message handshake patterns are not yet supported by the manager")
+	ErrSubtypeMismatch         = errors.New("packet subtype does not match handshake machine subtype")
+)

handshake/handshake.proto

@@ -0,0 +1,29 @@
+// This file documents the wire format the nebula handshake speaks. It is
+// not run through protoc; the encoder/decoder in payload.go is hand-written
+// against this shape directly to keep the parser narrow and panic-free.
+//
+// Any change to the wire format must be reflected here, and adding a new
+// field requires updating MarshalPayload / unmarshalPayloadDetails together
+// with the field-uniqueness and wire-type checks in those functions.
+
+syntax = "proto3";
+package nebula.handshake;
+
+message NebulaHandshake {
+  NebulaHandshakeDetails Details = 1;
+  bytes Hmac = 2;
+}
+
+message NebulaHandshakeDetails {
+  bytes Cert = 1;
+  uint32 InitiatorIndex = 2;
+  uint32 ResponderIndex = 3;
+  // Cookie was reserved for an anti-DoS mechanism that was never
+  // implemented. No released version of nebula has ever populated it; the
+  // hand-written parser silently skips it on read.
+  uint64 Cookie = 4 [deprecated = true];
+  uint64 Time = 5;
+  uint32 CertVersion = 8;
+  // reserved for WIP multiport
+  reserved 6, 7;
+}

handshake/helpers_test.go

@@ -0,0 +1,116 @@
+package handshake
+
+import (
+	"net/netip"
+	"testing"
+	"time"
+
+	"github.com/flynn/noise"
+	"github.com/slackhq/nebula/cert"
+	ct "github.com/slackhq/nebula/cert_test"
+	"github.com/slackhq/nebula/header"
+	"github.com/stretchr/testify/require"
+)
+
+// testCertState holds cert material for a test peer.
+type testCertState struct {
+	version cert.Version
+	creds   map[cert.Version]*Credential
+}
+
+func (s *testCertState) getCredential(v cert.Version) *Credential {
+	return s.creds[v]
+}
+
+func newTestCertState(
+	t *testing.T, ca cert.Certificate, caKey []byte, name string, networks []netip.Prefix,
+) *testCertState {
+	return newTestCertStateWithCipher(t, ca, caKey, name, networks, noise.CipherChaChaPoly)
+}
+
+func newTestCertStateWithCipher(
+	t *testing.T, ca cert.Certificate, caKey []byte, name string, networks []netip.Prefix,
+	cipher noise.CipherFunc,
+) *testCertState {
+	t.Helper()
+	c, _, rawPrivKey, _ := ct.NewTestCert(
+		cert.Version2, cert.Curve_CURVE25519, ca, caKey,
+		name, ca.NotBefore(), ca.NotAfter(), networks, nil, nil,
+	)
+
+	priv, _, _, err := cert.UnmarshalPrivateKeyFromPEM(rawPrivKey)
+	require.NoError(t, err)
+
+	hsBytes, err := c.MarshalForHandshakes()
+	require.NoError(t, err)
+
+	ncs := noise.NewCipherSuite(noise.DH25519, cipher, noise.HashSHA256)
+	return &testCertState{
+		version: cert.Version2,
+		creds: map[cert.Version]*Credential{
+			cert.Version2: NewCredential(c, hsBytes, priv, ncs),
+		},
+	}
+}
+
+func testVerifier(pool *cert.CAPool) CertVerifier {
+	return func(c cert.Certificate) (*cert.CachedCertificate, error) {
+		return pool.VerifyCertificate(time.Now(), c)
+	}
+}
+
+func newTestMachine(
+	t *testing.T,
+	cs *testCertState,
+	verifier CertVerifier,
+	initiator bool,
+	localIndex uint32,
+) *Machine {
+	t.Helper()
+	m, err := NewMachine(
+		cs.version, cs.getCredential,
+		verifier, func() (uint32, error) { return localIndex, nil },
+		initiator, header.HandshakeIXPSK0,
+	)
+	require.NoError(t, err)
+	return m
+}
+
+func initiateHandshake(
+	t *testing.T,
+	initCS *testCertState, initVerifier CertVerifier,
+	respCS *testCertState, respVerifier CertVerifier,
+) (initM, respM *Machine, respResult *Result, resp []byte, err error) {
+	t.Helper()
+	initM = newTestMachine(t, initCS, initVerifier, true, 100)
+	msg1, merr := initM.Initiate(nil)
+	require.NoError(t, merr)
+
+	respM = newTestMachine(t, respCS, respVerifier, false, 200)
+	resp, respResult, err = respM.ProcessPacket(nil, msg1)
+	return
+}
+
+func doFullHandshake(
+	t *testing.T, initCS, respCS *testCertState, caPool *cert.CAPool,
+) (initResult, respResult *Result) {
+	t.Helper()
+	v := testVerifier(caPool)
+
+	initM := newTestMachine(t, initCS, v, true, 1000)
+	respM := newTestMachine(t, respCS, v, false, 2000)
+
+	msg1, err := initM.Initiate(nil)
+	require.NoError(t, err)
+
+	resp, respResult, err := respM.ProcessPacket(nil, msg1)
+	require.NoError(t, err)
+	require.NotNil(t, respResult)
+	require.NotEmpty(t, resp)
+
+	_, initResult, err = initM.ProcessPacket(nil, resp)
+	require.NoError(t, err)
+	require.NotNil(t, initResult)
+
+	return initResult, respResult
+}

handshake/machine.go

@@ -0,0 +1,444 @@
+package handshake
+
+import (
+	"bytes"
+	"fmt"
+	"slices"
+	"time"
+
+	"github.com/flynn/noise"
+	"github.com/slackhq/nebula/cert"
+	"github.com/slackhq/nebula/header"
+)
+
+// IndexAllocator is called by the Machine to allocate a local index for the
+// handshake. It is called at most once, when the first outgoing message that
+// carries a payload is built.
+//
+// Implementations MUST NOT return 0. Zero is reserved as a sentinel meaning
+// "no index assigned" on the wire and in the payload-presence checks. If an
+// allocator ever returned 0, a legitimate handshake's payload could be
+// indistinguishable from an empty one and would be rejected.
+type IndexAllocator func() (uint32, error)
+
+// CertVerifier is called by the Machine after reconstructing the peer's
+// certificate from the handshake. The verifier performs all validation
+// (CA trust, expiry, policy checks, allow lists).
+type CertVerifier func(cert.Certificate) (*cert.CachedCertificate, error)
+
+// Result contains the results of a successful handshake.
+// Returned by ProcessPacket when the handshake is complete.
+type Result struct {
+	EKey          *noise.CipherState
+	DKey          *noise.CipherState
+	MyCert        cert.Certificate
+	RemoteCert    *cert.CachedCertificate
+	RemoteIndex   uint32
+	LocalIndex    uint32
+	HandshakeTime uint64
+	MessageIndex  uint64 // number of messages exchanged during the handshake
+	Initiator     bool
+}
+
+// Machine drives a Noise handshake through N messages. It handles Noise
+// protocol operations, certificate reconstruction, and payload encoding.
+// Certificate validation is delegated to the caller via CertVerifier.
+//
+// A Machine is not safe for concurrent use. The caller must ensure that
+// Initiate and ProcessPacket are not called concurrently.
+//
+// Error contract: when ProcessPacket or Initiate returns an error, callers
+// must check Failed() to decide what to do next. If Failed() is false the
+// underlying noise state was not advanced (the packet was rejected before
+// ReadMessage took effect, or the rejection is non-fatal like a stale
+// retransmit) and the Machine can accept another packet. If Failed() is
+// true the Machine is unrecoverable and the caller must abandon it.
+type Machine struct {
+	hs             *noise.HandshakeState
+	getCred        GetCredentialFunc
+	allocIndex     IndexAllocator
+	verifier       CertVerifier
+	result         *Result
+	msgs           []msgFlags
+	myVersion      cert.Version
+	subtype        header.MessageSubType
+	indexAllocated bool
+	remoteCertSet  bool
+	payloadSet     bool
+	failed         bool
+}
+
+// NewMachine creates a handshake state machine. The subtype determines both
+// the noise pattern and the per-message content layout. The credential for
+// `version` is fetched via getCred and used to seed the noise.HandshakeState.
+// IndexAllocator is called lazily when the first outgoing payload is built.
+func NewMachine(
+	version cert.Version,
+	getCred GetCredentialFunc,
+	verifier CertVerifier,
+	allocIndex IndexAllocator,
+	initiator bool,
+	subtype header.MessageSubType,
+) (*Machine, error) {
+	info, err := subtypeInfoFor(subtype)
+	if err != nil {
+		return nil, err
+	}
+
+	cred := getCred(version)
+	if cred == nil {
+		return nil, fmt.Errorf("%w: %v", ErrNoCredential, version)
+	}
+
+	hs, err := cred.buildHandshakeState(initiator, info.pattern)
+	if err != nil {
+		return nil, fmt.Errorf("build noise state: %w", err)
+	}
+
+	return &Machine{
+		hs:         hs,
+		subtype:    subtype,
+		msgs:       info.msgs,
+		getCred:    getCred,
+		allocIndex: allocIndex,
+		verifier:   verifier,
+		myVersion:  version,
+		result: &Result{
+			Initiator: initiator,
+		},
+	}, nil
+}
+
+// Failed returns true if the Machine is in an unrecoverable state.
+func (m *Machine) Failed() bool {
+	return m.failed
+}
+
+// Subtype returns the handshake subtype this Machine was built for.
+func (m *Machine) Subtype() header.MessageSubType {
+	return m.subtype
+}
+
+// MessageIndex returns the noise handshake message index, which equals the
+// wire counter of the most recently sent or received message.
+func (m *Machine) MessageIndex() int {
+	return m.hs.MessageIndex()
+}
+
+// requireComplete checks that both a peer cert and payload have been received.
+// Marks the machine as failed if not.
+func (m *Machine) requireComplete() error {
+	if !m.payloadSet || !m.remoteCertSet {
+		m.failed = true
+		return ErrIncompleteHandshake
+	}
+	return nil
+}
+
+// myMsgFlags returns the flags for the current outgoing message.
+func (m *Machine) myMsgFlags() msgFlags {
+	idx := m.hs.MessageIndex()
+	if idx < len(m.msgs) {
+		return m.msgs[idx]
+	}
+	return msgFlags{}
+}
+
+// peerMsgFlags returns the flags for the message we just read.
+func (m *Machine) peerMsgFlags() msgFlags {
+	idx := m.hs.MessageIndex() - 1
+	if idx >= 0 && idx < len(m.msgs) {
+		return m.msgs[idx]
+	}
+	return msgFlags{}
+}
+
+// Initiate produces the first handshake message. Only valid for initiators,
+// and must be called exactly once before ProcessPacket.
+//
+// out is a destination buffer the message is appended to and returned. Pass
+// nil to allocate fresh, or pass a re-used buffer sliced to length 0 (e.g.
+// buf[:0]) with sufficient capacity to avoid allocation.
+//
+// An error return may not indicate a fatal condition, check Failed() to
+// determine if the Machine can still be used.
+func (m *Machine) Initiate(out []byte) ([]byte, error) {
+	if m.failed {
+		return nil, ErrMachineFailed
+	}
+	if !m.result.Initiator {
+		m.failed = true
+		return nil, ErrInitiateOnResponder
+	}
+	if m.hs.MessageIndex() != 0 {
+		m.failed = true
+		return nil, ErrInitiateAlreadyCalled
+	}
+
+	// At MessageIndex=0 with RemoteIndex still zero, buildResponse produces
+	// header counter 1 and remote index 0, which is what the initial message needs.
+	out, _, _, err := m.buildResponse(out)
+	if err != nil {
+		m.failed = true
+		return nil, err
+	}
+	return out, nil
+}
+
+// ProcessPacket handles an incoming handshake message. It advances the Noise
+// state, validates the peer certificate via the verifier, and optionally
+// produces a response.
+//
+// out is a destination buffer the response is appended to and returned. Pass
+// nil to allocate fresh, or pass a re-used buffer sliced to length 0 (e.g.
+// buf[:0]) with sufficient capacity to avoid allocation. The returned slice
+// is nil when no outgoing message is produced (handshake complete on this
+// side, or final message of a multi-message pattern).
+//
+// Returns a non-nil Result when the handshake is complete.
+// An error return may not indicate a fatal condition, check Failed() to
+// determine if the Machine can still be used.
+func (m *Machine) ProcessPacket(out, packet []byte) ([]byte, *Result, error) {
+	if m.failed {
+		return nil, nil, ErrMachineFailed
+	}
+	if len(packet) < header.Len {
+		return nil, nil, ErrPacketTooShort
+	}
+	// Reject packets whose subtype doesn't match the one this Machine was
+	// built for. A pending handshake that suddenly receives a different
+	// subtype on its index is either a stray packet that matched by chance
+	// or a peer protocol violation; drop it without failing the Machine so
+	// the legitimate retransmit can still complete.
+	if header.MessageSubType(packet[1]) != m.subtype {
+		return nil, nil, ErrSubtypeMismatch
+	}
+	if m.result.Initiator && m.hs.MessageIndex() == 0 {
+		m.failed = true
+		return nil, nil, ErrInitiateNotCalled
+	}
+
+	// The (eKey, dKey) ordering here is correct for IX, where the initiator
+	// completes the handshake by reading the responder's stage-2 message.
+	// noise returns (cs1, cs2) where cs1 is the initiator->responder cipher.
+	// For 3-message patterns where a responder finishes by reading the final
+	// message, this ordering would be wrong; revisit when XX/pqIX lands.
+	msg, eKey, dKey, err := m.hs.ReadMessage(nil, packet[header.Len:])
+	if err != nil {
+		// Noise ReadMessage failed. The noise library checkpoints and rolls back
+		// on failure, so the Machine is still alive. The caller can retry with
+		// a different packet.
+		return nil, nil, fmt.Errorf("noise ReadMessage: %w", err)
+	}
+
+	// From here on, noise state has advanced. Any error is fatal.
+	flags := m.peerMsgFlags()
+
+	if err := m.processPayload(msg, flags); err != nil {
+		return nil, nil, err
+	}
+
+	// If ReadMessage derived keys, the handshake is complete. Noise should
+	// always produce both keys together; asymmetry is a protocol invariant
+	// violation.
+	if eKey != nil || dKey != nil {
+		if eKey == nil || dKey == nil {
+			m.failed = true
+			return nil, nil, ErrAsymmetricCipherKeys
+		}
+		if err := m.requireComplete(); err != nil {
+			return nil, nil, err
+		}
+		return nil, m.completed(eKey, dKey), nil
+	}
+
+	// ReadMessage didn't complete, produce the next outgoing message
+	out, dk, ek, err := m.buildResponse(out)
+	if err != nil {
+		m.failed = true
+		return nil, nil, err
+	}
+
+	if ek != nil || dk != nil {
+		if ek == nil || dk == nil {
+			m.failed = true
+			return nil, nil, ErrAsymmetricCipherKeys
+		}
+		if err := m.requireComplete(); err != nil {
+			return nil, nil, err
+		}
+		return out, m.completed(ek, dk), nil
+	}
+
+	return out, nil, nil
+}
+
+func (m *Machine) completed(eKey, dKey *noise.CipherState) *Result {
+	m.result.EKey = eKey
+	m.result.DKey = dKey
+	m.result.MessageIndex = uint64(m.hs.MessageIndex())
+	return m.result
+}
+
+func (m *Machine) processPayload(msg []byte, flags msgFlags) error {
+	if len(msg) == 0 {
+		if flags.expectsPayload || flags.expectsCert {
+			m.failed = true
+			return ErrMissingContent
+		}
+		return nil
+	}
+
+	payload, err := UnmarshalPayload(msg)
+	if err != nil {
+		m.failed = true
+		return fmt.Errorf("unmarshal handshake: %w", err)
+	}
+
+	// Assert the payload contains exactly what we expect
+	hasPayloadData := payload.InitiatorIndex != 0 || payload.ResponderIndex != 0 || payload.Time != 0
+	if hasPayloadData != flags.expectsPayload {
+		m.failed = true
+		return ErrUnexpectedContent
+	}
+
+	hasCertData := len(payload.Cert) > 0
+	if hasCertData != flags.expectsCert {
+		m.failed = true
+		return ErrUnexpectedContent
+	}
+
+	// Process payload
+	if flags.expectsPayload {
+		if m.result.Initiator {
+			m.result.RemoteIndex = payload.ResponderIndex
+		} else {
+			m.result.RemoteIndex = payload.InitiatorIndex
+		}
+		m.result.HandshakeTime = payload.Time
+		m.payloadSet = true
+	}
+
+	// Process certificate
+	if flags.expectsCert {
+		if err := m.validateCert(payload); err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+func (m *Machine) validateCert(payload Payload) error {
+	cred := m.getCred(m.myVersion)
+	if cred == nil {
+		m.failed = true
+		return fmt.Errorf("%w: %v", ErrNoCredential, m.myVersion)
+	}
+	rc, err := cert.Recombine(
+		cert.Version(payload.CertVersion),
+		payload.Cert,
+		m.hs.PeerStatic(),
+		cred.Cert.Curve(),
+	)
+	if err != nil {
+		m.failed = true
+		return fmt.Errorf("recombine cert: %w", err)
+	}
+
+	if !bytes.Equal(rc.PublicKey(), m.hs.PeerStatic()) {
+		m.failed = true
+		return ErrPublicKeyMismatch
+	}
+
+	// Version negotiation, if the peer sent a different version and we have it, switch
+	if rc.Version() != m.myVersion {
+		if m.getCred(rc.Version()) != nil {
+			m.myVersion = rc.Version()
+		}
+	}
+
+	verified, err := m.verifier(rc)
+	if err != nil {
+		m.failed = true
+		return fmt.Errorf("verify cert: %w", err)
+	}
+
+	m.result.RemoteCert = verified
+	m.remoteCertSet = true
+	return nil
+}
+
+func (m *Machine) marshalOutgoing(flags msgFlags) ([]byte, error) {
+	if !flags.expectsPayload && !flags.expectsCert {
+		return nil, nil
+	}
+
+	var p Payload
+	if flags.expectsPayload {
+		if !m.indexAllocated {
+			index, err := m.allocIndex()
+			if err != nil {
+				return nil, fmt.Errorf("%w: %w", ErrIndexAllocation, err)
+			}
+			m.result.LocalIndex = index
+			m.indexAllocated = true
+		}
+
+		if m.result.Initiator {
+			p.InitiatorIndex = m.result.LocalIndex
+		} else {
+			p.ResponderIndex = m.result.LocalIndex
+			p.InitiatorIndex = m.result.RemoteIndex
+		}
+		p.Time = uint64(time.Now().UnixNano())
+	}
+	if flags.expectsCert {
+		cred := m.getCred(m.myVersion)
+		if cred == nil {
+			return nil, fmt.Errorf("%w: %v", ErrNoCredential, m.myVersion)
+		}
+		p.Cert = cred.Bytes
+		p.CertVersion = uint32(cred.Cert.Version())
+		m.result.MyCert = cred.Cert
+	}
+
+	return MarshalPayload(nil, p), nil
+}
+
+func (m *Machine) buildResponse(out []byte) ([]byte, *noise.CipherState, *noise.CipherState, error) {
+	flags := m.myMsgFlags()
+	hsBytes, err := m.marshalOutgoing(flags)
+	if err != nil {
+		return nil, nil, nil, err
+	}
+
+	// Extend out by header.Len to make room for the header. slices.Grow is a
+	// no-op when the cap is already sufficient (the zero-copy case where the
+	// caller passed a pre-sized buffer). header.Encode overwrites the new
+	// bytes, so they don't need to be zeroed.
+	start := len(out)
+	out = slices.Grow(out, header.Len)[:start+header.Len]
+	header.Encode(
+		out[start:],
+		header.Version, header.Handshake, m.subtype,
+		m.result.RemoteIndex,
+		uint64(m.hs.MessageIndex()+1),
+	)
+
+	// noise.WriteMessage appends the encrypted handshake message to out,
+	// reusing capacity when present.
+	//
+	// The (dKey, eKey) ordering here is correct for IX, where the responder
+	// completes the handshake by writing the stage-2 message. noise returns
+	// (cs1, cs2) where cs1 is the initiator->responder cipher (which is the
+	// responder's decrypt key). For 3-message patterns where an initiator
+	// finishes by writing the final message, this ordering would be wrong;
+	// revisit when XX/pqIX lands.
+	out, dKey, eKey, err := m.hs.WriteMessage(out, hsBytes)
+	if err != nil {
+		return nil, nil, nil, fmt.Errorf("noise WriteMessage: %w", err)
+	}
+
+	return out, dKey, eKey, nil
+}

handshake/machine_test.go

@@ -0,0 +1,662 @@
+package handshake
+
+import (
+	"net/netip"
+	"testing"
+	"time"
+
+	"github.com/flynn/noise"
+	"github.com/slackhq/nebula/cert"
+	ct "github.com/slackhq/nebula/cert_test"
+	"github.com/slackhq/nebula/header"
+	"github.com/slackhq/nebula/noiseutil"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestMachineIXHappyPath(t *testing.T) {
+	ca, _, caKey, _ := ct.NewTestCaCert(
+		cert.Version2, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil,
+	)
+	caPool := ct.NewTestCAPool(ca)
+
+	initCS := newTestCertState(t, ca, caKey, "initiator", []netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")})
+	respCS := newTestCertState(t, ca, caKey, "responder", []netip.Prefix{netip.MustParsePrefix("10.0.0.2/24")})
+
+	initR, respR := doFullHandshake(t, initCS, respCS, caPool)
+
+	assert.Equal(t, "responder", initR.RemoteCert.Certificate.Name())
+	assert.Equal(t, "initiator", respR.RemoteCert.Certificate.Name())
+
+	assert.Equal(t, uint32(1000), initR.LocalIndex)
+	assert.Equal(t, uint32(2000), initR.RemoteIndex)
+	assert.Equal(t, uint32(2000), respR.LocalIndex)
+	assert.Equal(t, uint32(1000), respR.RemoteIndex)
+
+	assert.Equal(t, uint64(2), initR.MessageIndex, "IX has 2 messages")
+	assert.Equal(t, uint64(2), respR.MessageIndex, "IX has 2 messages")
+
+	ct1, err := initR.EKey.Encrypt(nil, nil, []byte("hello"))
+	require.NoError(t, err)
+	pt1, err := respR.DKey.Decrypt(nil, nil, ct1)
+	require.NoError(t, err)
+	assert.Equal(t, []byte("hello"), pt1)
+
+	ct2, err := respR.EKey.Encrypt(nil, nil, []byte("world"))
+	require.NoError(t, err)
+	pt2, err := initR.DKey.Decrypt(nil, nil, ct2)
+	require.NoError(t, err)
+	assert.Equal(t, []byte("world"), pt2)
+}
+
+func TestMachineInitiateErrors(t *testing.T) {
+	ca, _, caKey, _ := ct.NewTestCaCert(
+		cert.Version2, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil,
+	)
+	caPool := ct.NewTestCAPool(ca)
+	cs := newTestCertState(t, ca, caKey, "test", []netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")})
+	v := testVerifier(caPool)
+
+	t.Run("initiate on responder", func(t *testing.T) {
+		m := newTestMachine(t, cs, v, false, 100)
+		_, err := m.Initiate(nil)
+		require.ErrorIs(t, err, ErrInitiateOnResponder)
+		assert.True(t, m.Failed())
+	})
+
+	t.Run("initiate called twice", func(t *testing.T) {
+		m := newTestMachine(t, cs, v, true, 100)
+		_, err := m.Initiate(nil)
+		require.NoError(t, err)
+		_, err = m.Initiate(nil)
+		require.ErrorIs(t, err, ErrInitiateAlreadyCalled)
+		assert.True(t, m.Failed())
+	})
+
+	t.Run("process packet before initiate on initiator", func(t *testing.T) {
+		m := newTestMachine(t, cs, v, true, 100)
+		_, _, err := m.ProcessPacket(nil, make([]byte, 100))
+		require.ErrorIs(t, err, ErrInitiateNotCalled)
+		assert.True(t, m.Failed())
+	})
+
+	t.Run("calling failed machine", func(t *testing.T) {
+		m := newTestMachine(t, cs, v, false, 100)
+		_, err := m.Initiate(nil) // fails: responder
+		require.Error(t, err)
+		_, err = m.Initiate(nil) // fails: already failed
+		require.ErrorIs(t, err, ErrMachineFailed)
+	})
+}
+
+func TestMachineProcessPacketErrors(t *testing.T) {
+	ca, _, caKey, _ := ct.NewTestCaCert(
+		cert.Version2, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil,
+	)
+	caPool := ct.NewTestCAPool(ca)
+	cs := newTestCertState(t, ca, caKey, "test", []netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")})
+	v := testVerifier(caPool)
+
+	t.Run("packet too short", func(t *testing.T) {
+		m := newTestMachine(t, cs, v, false, 100)
+		_, _, err := m.ProcessPacket(nil, []byte{1, 2, 3})
+		require.ErrorIs(t, err, ErrPacketTooShort)
+		assert.False(t, m.Failed(), "short packet should not kill machine")
+	})
+
+	t.Run("noise decryption failure is recoverable", func(t *testing.T) {
+		initCS := newTestCertState(t, ca, caKey, "init", []netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")})
+		initM := newTestMachine(t, initCS, v, true, 100)
+		msg1, err := initM.Initiate(nil)
+		require.NoError(t, err)
+
+		respM := newTestMachine(t, cs, v, false, 200)
+		resp, _, err := respM.ProcessPacket(nil, msg1)
+		require.NoError(t, err)
+
+		corrupted := make([]byte, len(resp))
+		copy(corrupted, resp)
+		for i := header.Len; i < len(corrupted); i++ {
+			corrupted[i] ^= 0xff
+		}
+		_, _, err = initM.ProcessPacket(nil, corrupted)
+		require.Error(t, err)
+		assert.False(t, initM.Failed(), "noise failure should be recoverable")
+
+		// And the machine should still complete a real handshake afterward.
+		_, result, err := initM.ProcessPacket(nil, resp)
+		require.NoError(t, err)
+		require.NotNil(t, result, "initiator should complete on the legitimate response")
+	})
+
+	t.Run("invalid cert is fatal", func(t *testing.T) {
+		otherCA, _, otherCAKey, _ := ct.NewTestCaCert(
+			cert.Version2, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil,
+		)
+		otherCS := newTestCertState(t, otherCA, otherCAKey, "other", []netip.Prefix{netip.MustParsePrefix("10.0.0.2/24")})
+
+		initM := newTestMachine(t, otherCS, testVerifier(ct.NewTestCAPool(otherCA)), true, 100)
+		msg1, err := initM.Initiate(nil)
+		require.NoError(t, err)
+
+		respM := newTestMachine(t, cs, v, false, 200)
+		_, _, err = respM.ProcessPacket(nil, msg1)
+		require.Error(t, err)
+		assert.True(t, respM.Failed(), "cert validation failure should kill machine")
+	})
+
+	t.Run("subtype mismatch is recoverable", func(t *testing.T) {
+		initCS := newTestCertState(t, ca, caKey, "init", []netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")})
+		initM := newTestMachine(t, initCS, v, true, 100)
+		msg1, err := initM.Initiate(nil)
+		require.NoError(t, err)
+
+		// Mutate the subtype byte (offset 1 in the header) to a value the
+		// responder Machine wasn't built for.
+		bad := make([]byte, len(msg1))
+		copy(bad, msg1)
+		bad[1] = 0xff
+
+		respM := newTestMachine(t, cs, v, false, 200)
+		_, _, err = respM.ProcessPacket(nil, bad)
+		require.ErrorIs(t, err, ErrSubtypeMismatch)
+		assert.False(t, respM.Failed(), "subtype mismatch should not kill the machine")
+
+		// And the machine should still complete a real handshake afterward.
+		resp, result, err := respM.ProcessPacket(nil, msg1)
+		require.NoError(t, err)
+		require.NotNil(t, result, "responder should complete on the legitimate stage-1 packet")
+		assert.NotEmpty(t, resp, "responder should produce a stage-2 reply")
+	})
+}
+
+// TestMachineProcessPayload exercises processPayload's internal validation
+// directly. Most of these failure modes can't be reached black-box once the
+// subtype check at the top of ProcessPacket gates external callers, so we
+// drive them by hand here for coverage.
+func TestMachineProcessPayload(t *testing.T) {
+	ca, _, caKey, _ := ct.NewTestCaCert(
+		cert.Version2, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil,
+	)
+	caPool := ct.NewTestCAPool(ca)
+	cs := newTestCertState(t, ca, caKey, "test", []netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")})
+	v := testVerifier(caPool)
+
+	t.Run("empty message with expects fails", func(t *testing.T) {
+		m := newTestMachine(t, cs, v, false, 100)
+		err := m.processPayload(nil, msgFlags{expectsPayload: true, expectsCert: true})
+		require.ErrorIs(t, err, ErrMissingContent)
+		assert.True(t, m.Failed())
+	})
+
+	t.Run("empty message with no expects passes", func(t *testing.T) {
+		m := newTestMachine(t, cs, v, false, 100)
+		err := m.processPayload(nil, msgFlags{})
+		require.NoError(t, err)
+		assert.False(t, m.Failed())
+	})
+
+	t.Run("malformed protobuf is fatal", func(t *testing.T) {
+		m := newTestMachine(t, cs, v, false, 100)
+		err := m.processPayload([]byte{0xff, 0xff, 0xff}, msgFlags{expectsPayload: true, expectsCert: true})
+		require.Error(t, err)
+		assert.True(t, m.Failed())
+	})
+
+	t.Run("unexpected payload data is fatal", func(t *testing.T) {
+		m := newTestMachine(t, cs, v, false, 100)
+		// A payload with index data when none was expected.
+		bytes := MarshalPayload(nil, Payload{InitiatorIndex: 42, Time: 1})
+		err := m.processPayload(bytes, msgFlags{expectsPayload: false, expectsCert: false})
+		require.ErrorIs(t, err, ErrUnexpectedContent)
+		assert.True(t, m.Failed())
+	})
+
+	t.Run("unexpected cert data is fatal", func(t *testing.T) {
+		m := newTestMachine(t, cs, v, false, 100)
+		// A payload with cert when none was expected.
+		bytes := MarshalPayload(nil, Payload{Cert: []byte{1, 2, 3}, CertVersion: 2})
+		err := m.processPayload(bytes, msgFlags{expectsPayload: false, expectsCert: false})
+		require.ErrorIs(t, err, ErrUnexpectedContent)
+		assert.True(t, m.Failed())
+	})
+
+	t.Run("missing payload data when expected is fatal", func(t *testing.T) {
+		m := newTestMachine(t, cs, v, false, 100)
+		// Cert present, but no index/time fields.
+		bytes := MarshalPayload(nil, Payload{Cert: []byte{1, 2, 3}, CertVersion: 2})
+		err := m.processPayload(bytes, msgFlags{expectsPayload: true, expectsCert: true})
+		require.ErrorIs(t, err, ErrUnexpectedContent)
+		assert.True(t, m.Failed())
+	})
+}
+
+// TestMachineRequireComplete checks the fail-on-incomplete-handshake path
+// directly. Like processPayload above this isn't reachable from a normal IX
+// flow, so we drive it by hand.
+func TestMachineRequireComplete(t *testing.T) {
+	ca, _, caKey, _ := ct.NewTestCaCert(
+		cert.Version2, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil,
+	)
+	caPool := ct.NewTestCAPool(ca)
+	cs := newTestCertState(t, ca, caKey, "test", []netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")})
+	v := testVerifier(caPool)
+
+	t.Run("missing both fails", func(t *testing.T) {
+		m := newTestMachine(t, cs, v, false, 100)
+		err := m.requireComplete()
+		require.ErrorIs(t, err, ErrIncompleteHandshake)
+		assert.True(t, m.Failed())
+	})
+
+	t.Run("payload only fails", func(t *testing.T) {
+		m := newTestMachine(t, cs, v, false, 100)
+		m.payloadSet = true
+		err := m.requireComplete()
+		require.ErrorIs(t, err, ErrIncompleteHandshake)
+		assert.True(t, m.Failed())
+	})
+
+	t.Run("cert only fails", func(t *testing.T) {
+		m := newTestMachine(t, cs, v, false, 100)
+		m.remoteCertSet = true
+		err := m.requireComplete()
+		require.ErrorIs(t, err, ErrIncompleteHandshake)
+		assert.True(t, m.Failed())
+	})
+
+	t.Run("both set passes", func(t *testing.T) {
+		m := newTestMachine(t, cs, v, false, 100)
+		m.payloadSet = true
+		m.remoteCertSet = true
+		err := m.requireComplete()
+		require.NoError(t, err)
+		assert.False(t, m.Failed())
+	})
+}
+
+func TestMachineAESCipher(t *testing.T) {
+	ca, _, caKey, _ := ct.NewTestCaCert(
+		cert.Version2, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil,
+	)
+	caPool := ct.NewTestCAPool(ca)
+
+	initCS := newTestCertStateWithCipher(
+		t, ca, caKey, "init",
+		[]netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")},
+		noiseutil.CipherAESGCM,
+	)
+	respCS := newTestCertStateWithCipher(
+		t, ca, caKey, "resp",
+		[]netip.Prefix{netip.MustParsePrefix("10.0.0.2/24")},
+		noiseutil.CipherAESGCM,
+	)
+
+	initR, respR := doFullHandshake(t, initCS, respCS, caPool)
+
+	ct1, err := initR.EKey.Encrypt(nil, nil, []byte("works"))
+	require.NoError(t, err)
+	pt1, err := respR.DKey.Decrypt(nil, nil, ct1)
+	require.NoError(t, err)
+	assert.Equal(t, []byte("works"), pt1)
+
+	ct2, err := respR.EKey.Encrypt(nil, nil, []byte("back"))
+	require.NoError(t, err)
+	pt2, err := initR.DKey.Decrypt(nil, nil, ct2)
+	require.NoError(t, err)
+	assert.Equal(t, []byte("back"), pt2)
+}
+
+func TestResultFields(t *testing.T) {
+	ca, _, caKey, _ := ct.NewTestCaCert(
+		cert.Version2, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil,
+	)
+	caPool := ct.NewTestCAPool(ca)
+	initCS := newTestCertState(t, ca, caKey, "init", []netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")})
+	respCS := newTestCertState(t, ca, caKey, "resp", []netip.Prefix{netip.MustParsePrefix("10.0.0.2/24")})
+
+	initR, respR := doFullHandshake(t, initCS, respCS, caPool)
+
+	assert.True(t, initR.Initiator)
+	assert.False(t, respR.Initiator)
+	assert.NotZero(t, initR.HandshakeTime)
+	assert.NotZero(t, respR.HandshakeTime)
+	assert.NotNil(t, initR.RemoteCert)
+	assert.NotNil(t, respR.RemoteCert)
+}
+
+func TestMachineBufferReuse(t *testing.T) {
+	ca, _, caKey, _ := ct.NewTestCaCert(
+		cert.Version2, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil,
+	)
+	caPool := ct.NewTestCAPool(ca)
+	initCS := newTestCertState(t, ca, caKey, "init", []netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")})
+	respCS := newTestCertState(t, ca, caKey, "resp", []netip.Prefix{netip.MustParsePrefix("10.0.0.2/24")})
+	v := testVerifier(caPool)
+
+	initM := newTestMachine(t, initCS, v, true, 1000)
+	respM := newTestMachine(t, respCS, v, false, 2000)
+
+	msg1, err := initM.Initiate(nil)
+	require.NoError(t, err)
+
+	t.Run("response writes into provided buffer", func(t *testing.T) {
+		buf := make([]byte, 0, 4096)
+		resp, result, err := respM.ProcessPacket(buf, msg1)
+		require.NoError(t, err)
+		require.NotNil(t, result)
+
+		assert.NotEmpty(t, resp, "response should have content")
+		assert.Equal(t, &buf[:1][0], &resp[:1][0],
+			"response should reuse the provided buffer's backing array")
+	})
+
+	t.Run("initiate writes into provided buffer", func(t *testing.T) {
+		initM2 := newTestMachine(t, initCS, v, true, 3000)
+		buf := make([]byte, 0, 4096)
+		msg, err := initM2.Initiate(buf)
+		require.NoError(t, err)
+
+		assert.NotEmpty(t, msg, "initiate should have content")
+		assert.Equal(t, &buf[:1][0], &msg[:1][0],
+			"initiate should reuse the provided buffer's backing array")
+	})
+
+	t.Run("nil out still works", func(t *testing.T) {
+		initM2 := newTestMachine(t, initCS, v, true, 4000)
+		respM2 := newTestMachine(t, respCS, v, false, 5000)
+
+		msg1, err := initM2.Initiate(nil)
+		require.NoError(t, err)
+
+		resp, _, err := respM2.ProcessPacket(nil, msg1)
+		require.NoError(t, err)
+
+		out, result, err := initM2.ProcessPacket(nil, resp)
+		require.NoError(t, err)
+		assert.NotNil(t, result)
+		assert.Nil(t, out, "initiator should have no response for IX msg2")
+	})
+}
+
+func TestMachineMsgIndexTracking(t *testing.T) {
+	ca, _, caKey, _ := ct.NewTestCaCert(
+		cert.Version2, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil,
+	)
+	caPool := ct.NewTestCAPool(ca)
+	initCS := newTestCertState(t, ca, caKey, "init", []netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")})
+	respCS := newTestCertState(t, ca, caKey, "resp", []netip.Prefix{netip.MustParsePrefix("10.0.0.2/24")})
+	v := testVerifier(caPool)
+
+	initM := newTestMachine(t, initCS, v, true, 100)
+	respM := newTestMachine(t, respCS, v, false, 200)
+
+	msg1, err := initM.Initiate(nil)
+	require.NoError(t, err)
+
+	resp1, result1, err := respM.ProcessPacket(nil, msg1)
+	require.NoError(t, err)
+	assert.NotNil(t, result1)
+
+	_, result2, err := initM.ProcessPacket(nil, resp1)
+	require.NoError(t, err)
+	assert.NotNil(t, result2)
+}
+
+func TestMachineThreeMessagePattern(t *testing.T) {
+	registerTestXXInfo(t)
+
+	// Use HandshakeXX (3 messages) to verify the Machine handles multi-message
+	// patterns correctly. XX flow:
+	//   msg1 (I->R): [E]           - payload only, no cert
+	//   msg2 (R->I): [E, ee, S, es] - payload + cert
+	//   msg3 (I->R): [S, se]       - cert only (no payload, not first two)
+
+	ca, _, caKey, _ := ct.NewTestCaCert(
+		cert.Version2, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil,
+	)
+	caPool := ct.NewTestCAPool(ca)
+	v := testVerifier(caPool)
+
+	initCS := newTestCertState(t, ca, caKey, "init", []netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")})
+	respCS := newTestCertState(t, ca, caKey, "resp", []netip.Prefix{netip.MustParsePrefix("10.0.0.2/24")})
+
+	initM, err := NewMachine(
+		cert.Version2,
+		initCS.getCredential, v,
+		func() (uint32, error) { return 1000, nil },
+		true, header.HandshakeXXPSK0,
+	)
+	require.NoError(t, err)
+
+	respM, err := NewMachine(
+		cert.Version2,
+		respCS.getCredential, v,
+		func() (uint32, error) { return 2000, nil },
+		false, header.HandshakeXXPSK0,
+	)
+	require.NoError(t, err)
+
+	// msg1: initiator -> responder (E only, no cert)
+	msg1, err := initM.Initiate(nil)
+	require.NoError(t, err)
+	assert.NotEmpty(t, msg1)
+
+	// Responder processes msg1, should not complete yet, should produce msg2
+	msg2, result, err := respM.ProcessPacket(nil, msg1)
+	require.NoError(t, err)
+	assert.Nil(t, result, "XX should not complete on msg1")
+	assert.NotEmpty(t, msg2, "responder should produce msg2")
+
+	// Initiator processes msg2: gets responder's cert, produces msg3, and
+	// completes (WriteMessage for msg3 derives keys)
+	msg3, initResult, err := initM.ProcessPacket(nil, msg2)
+	require.NoError(t, err)
+	require.NotNil(t, initResult, "XX initiator should complete after reading msg2 and writing msg3")
+	assert.NotEmpty(t, msg3, "initiator should produce msg3")
+	assert.Equal(t, "resp", initResult.RemoteCert.Certificate.Name())
+
+	// Responder processes msg3: gets initiator's cert and completes
+	_, respResult, err := respM.ProcessPacket(nil, msg3)
+	require.NoError(t, err)
+	require.NotNil(t, respResult, "XX responder should complete on msg3")
+	assert.Equal(t, "init", respResult.RemoteCert.Certificate.Name())
+
+	assert.Equal(t, uint64(3), initResult.MessageIndex, "XX has 3 messages")
+	assert.Equal(t, uint64(3), respResult.MessageIndex, "XX has 3 messages")
+
+	// Verify keys work
+	ct1, err := initResult.EKey.Encrypt(nil, nil, []byte("three messages"))
+	require.NoError(t, err)
+	pt1, err := respResult.DKey.Decrypt(nil, nil, ct1)
+	require.NoError(t, err)
+	assert.Equal(t, []byte("three messages"), pt1)
+}
+
+// NOTE: ErrIncompleteHandshake is tested implicitly. It can't be triggered with
+// IX since the cert is always in the payload. A 3-message pattern test (HybridIX)
+// should exercise the case where cert arrives in msg3 and verify that completing
+// without it fails.
+
+func TestMachineExpiredCert(t *testing.T) {
+	ca, _, caKey, _ := ct.NewTestCaCert(
+		cert.Version2, cert.Curve_CURVE25519,
+		time.Now().Add(-24*time.Hour), time.Now().Add(24*time.Hour),
+		nil, nil, nil,
+	)
+	caPool := ct.NewTestCAPool(ca)
+
+	expCert, _, expKeyPEM, _ := ct.NewTestCert(
+		cert.Version2, cert.Curve_CURVE25519, ca, caKey,
+		"expired", time.Now().Add(-2*time.Hour), time.Now().Add(-1*time.Hour),
+		[]netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")}, nil, nil,
+	)
+	expKey, _, _, err := cert.UnmarshalPrivateKeyFromPEM(expKeyPEM)
+	require.NoError(t, err)
+	expHsBytes, err := expCert.MarshalForHandshakes()
+	require.NoError(t, err)
+	ncs := noise.NewCipherSuite(noise.DH25519, noise.CipherChaChaPoly, noise.HashSHA256)
+
+	expiredCS := &testCertState{
+		version: cert.Version2,
+		creds: map[cert.Version]*Credential{
+			cert.Version2: NewCredential(expCert, expHsBytes, expKey, ncs),
+		},
+	}
+
+	respCS := newTestCertState(
+		t, ca, caKey, "responder",
+		[]netip.Prefix{netip.MustParsePrefix("10.0.0.2/24")},
+	)
+
+	_, respM, _, _, err := initiateHandshake(
+		t, expiredCS, testVerifier(caPool),
+		respCS, testVerifier(caPool),
+	)
+	require.ErrorContains(t, err, "verify cert")
+	assert.True(t, respM.Failed())
+}
+
+func TestMachineNoCertNetworks(t *testing.T) {
+	ca, _, caKey, _ := ct.NewTestCaCert(
+		cert.Version2, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil,
+	)
+	caPool := ct.NewTestCAPool(ca)
+
+	caHsBytes, err := ca.MarshalForHandshakes()
+	require.NoError(t, err)
+	ncs := noise.NewCipherSuite(noise.DH25519, noise.CipherChaChaPoly, noise.HashSHA256)
+
+	noNetCS := &testCertState{
+		version: cert.Version2,
+		creds: map[cert.Version]*Credential{
+			cert.Version2: NewCredential(ca, caHsBytes, caKey, ncs),
+		},
+	}
+
+	respCS := newTestCertState(
+		t, ca, caKey, "responder",
+		[]netip.Prefix{netip.MustParsePrefix("10.0.0.2/24")},
+	)
+
+	_, respM, _, _, err := initiateHandshake(
+		t, noNetCS, testVerifier(caPool),
+		respCS, testVerifier(caPool),
+	)
+	require.Error(t, err)
+	assert.True(t, respM.Failed())
+}
+
+func TestMachineDifferentCAs(t *testing.T) {
+	ca1, _, caKey1, _ := ct.NewTestCaCert(
+		cert.Version2, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil,
+	)
+	ca2, _, caKey2, _ := ct.NewTestCaCert(
+		cert.Version2, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil,
+	)
+
+	initCS := newTestCertState(
+		t, ca1, caKey1, "init",
+		[]netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")},
+	)
+	respCS := newTestCertState(
+		t, ca2, caKey2, "resp",
+		[]netip.Prefix{netip.MustParsePrefix("10.0.0.2/24")},
+	)
+
+	_, respM, _, _, err := initiateHandshake(
+		t, initCS, testVerifier(ct.NewTestCAPool(ca1)),
+		respCS, testVerifier(ct.NewTestCAPool(ca2)),
+	)
+	require.ErrorContains(t, err, "verify cert")
+	assert.True(t, respM.Failed())
+}
+
+func TestMachineVersionNegotiation(t *testing.T) {
+	ca1, _, caKey1, _ := ct.NewTestCaCert(
+		cert.Version1, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil,
+	)
+	ca2, _, caKey2, _ := ct.NewTestCaCert(
+		cert.Version2, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil,
+	)
+	caPool := ct.NewTestCAPool(ca1, ca2)
+
+	makeMultiVersionResp := func(t *testing.T) *testCertState {
+		t.Helper()
+		respCertV1, _, respKeyPEM, _ := ct.NewTestCert(
+			cert.Version1, cert.Curve_CURVE25519, ca1, caKey1, "resp",
+			ca1.NotBefore(), ca1.NotAfter(),
+			[]netip.Prefix{netip.MustParsePrefix("10.0.0.2/24")}, nil, nil,
+		)
+		respKey, _, _, _ := cert.UnmarshalPrivateKeyFromPEM(respKeyPEM)
+		respCertV2, _ := ct.NewTestCertDifferentVersion(respCertV1, cert.Version2, ca2, caKey2)
+		respHsV1, _ := respCertV1.MarshalForHandshakes()
+		respHsV2, _ := respCertV2.MarshalForHandshakes()
+		ncs := noise.NewCipherSuite(noise.DH25519, noise.CipherChaChaPoly, noise.HashSHA256)
+		return &testCertState{
+			version: cert.Version1,
+			creds: map[cert.Version]*Credential{
+				cert.Version1: NewCredential(respCertV1, respHsV1, respKey, ncs),
+				cert.Version2: NewCredential(respCertV2, respHsV2, respKey, ncs),
+			},
+		}
+	}
+
+	t.Run("responder matches initiator version", func(t *testing.T) {
+		initCS := newTestCertState(
+			t, ca2, caKey2, "init",
+			[]netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")},
+		)
+		respCS := makeMultiVersionResp(t)
+		v := testVerifier(caPool)
+
+		initM, _, respResult, resp, err := initiateHandshake(
+			t, initCS, v,
+			respCS, v,
+		)
+		require.NoError(t, err)
+		require.NotNil(t, respResult)
+
+		assert.Equal(t, cert.Version2, respResult.MyCert.Version(),
+			"responder should negotiate to initiator's version")
+
+		_, initResult, err := initM.ProcessPacket(nil, resp)
+		require.NoError(t, err)
+		require.NotNil(t, initResult)
+		assert.Equal(t, cert.Version2, initResult.RemoteCert.Certificate.Version(),
+			"initiator should see V2 cert from responder")
+	})
+
+	t.Run("responder keeps version when no match available", func(t *testing.T) {
+		initCS := newTestCertState(
+			t, ca2, caKey2, "init",
+			[]netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")},
+		)
+
+		respCert, _, respKeyPEM, _ := ct.NewTestCert(
+			cert.Version1, cert.Curve_CURVE25519, ca1, caKey1, "resp",
+			ca1.NotBefore(), ca1.NotAfter(),
+			[]netip.Prefix{netip.MustParsePrefix("10.0.0.2/24")}, nil, nil,
+		)
+		respKey, _, _, _ := cert.UnmarshalPrivateKeyFromPEM(respKeyPEM)
+		respHs, _ := respCert.MarshalForHandshakes()
+		ncs := noise.NewCipherSuite(noise.DH25519, noise.CipherChaChaPoly, noise.HashSHA256)
+		respCS := &testCertState{
+			version: cert.Version1,
+			creds: map[cert.Version]*Credential{
+				cert.Version1: NewCredential(respCert, respHs, respKey, ncs),
+			},
+		}
+
+		v := testVerifier(caPool)
+		_, _, respResult, _, err := initiateHandshake(
+			t, initCS, v,
+			respCS, v,
+		)
+		require.NoError(t, err)
+		require.NotNil(t, respResult)
+
+		assert.Equal(t, cert.Version1, respResult.MyCert.Version(),
+			"responder should keep V1 when V2 not available")
+	})
+}

handshake/patterns.go

@@ -0,0 +1,54 @@
+package handshake
+
+import (
+	"fmt"
+
+	"github.com/flynn/noise"
+	"github.com/slackhq/nebula/header"
+)
+
+// msgFlags tracks what application data a handshake message carries.
+type msgFlags struct {
+	expectsPayload bool // message carries indexes and time
+	expectsCert    bool // message carries the certificate
+}
+
+// subtypeInfo bundles the noise pattern with the per-message flags for a
+// given handshake subtype.
+type subtypeInfo struct {
+	pattern noise.HandshakePattern
+	msgs    []msgFlags
+}
+
+// subtypeInfos defines the noise pattern and message content layout for each
+// handshake subtype.
+var subtypeInfos = map[header.MessageSubType]subtypeInfo{
+	// IX: 2 messages, both carry payload and cert
+	header.HandshakeIXPSK0: {
+		pattern: noise.HandshakeIX,
+		msgs: []msgFlags{
+			{expectsPayload: true, expectsCert: true},
+			{expectsPayload: true, expectsCert: true},
+		},
+	},
+
+	// XX: 3 messages
+	//   msg1 (I->R): payload only
+	//   msg2 (R->I): payload + cert
+	//   msg3 (I->R): cert only
+	//header.HandshakeXXPSK0: {
+	//	pattern: noise.HandshakeXX,
+	//	msgs: []msgFlags{
+	//		{expectsPayload: true, expectsCert: false},
+	//		{expectsPayload: true, expectsCert: true},
+	//		{expectsPayload: false, expectsCert: true},
+	//	},
+	//},
+}
+
+func subtypeInfoFor(subtype header.MessageSubType) (subtypeInfo, error) {
+	if info, ok := subtypeInfos[subtype]; ok {
+		return info, nil
+	}
+	return subtypeInfo{}, fmt.Errorf("%w: %d", ErrUnknownSubtype, subtype)
+}

handshake/patterns_test.go

@@ -0,0 +1,63 @@
+package handshake
+
+import (
+	"testing"
+
+	"github.com/flynn/noise"
+	"github.com/slackhq/nebula/header"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestSubtypeInfo(t *testing.T) {
+	t.Run("IX", func(t *testing.T) {
+		info, err := subtypeInfoFor(header.HandshakeIXPSK0)
+		require.NoError(t, err)
+		assert.Equal(t, noise.HandshakeIX.Name, info.pattern.Name)
+		require.Len(t, info.msgs, 2)
+		// msg1: payload + cert
+		assert.True(t, info.msgs[0].expectsPayload)
+		assert.True(t, info.msgs[0].expectsCert)
+		// msg2: payload + cert
+		assert.True(t, info.msgs[1].expectsPayload)
+		assert.True(t, info.msgs[1].expectsCert)
+	})
+
+	t.Run("XX", func(t *testing.T) {
+		registerTestXXInfo(t)
+		info, err := subtypeInfoFor(header.HandshakeXXPSK0)
+		require.NoError(t, err)
+		assert.Equal(t, noise.HandshakeXX.Name, info.pattern.Name)
+		require.Len(t, info.msgs, 3)
+		// msg1: payload only
+		assert.True(t, info.msgs[0].expectsPayload)
+		assert.False(t, info.msgs[0].expectsCert)
+		// msg2: payload + cert
+		assert.True(t, info.msgs[1].expectsPayload)
+		assert.True(t, info.msgs[1].expectsCert)
+		// msg3: cert only
+		assert.False(t, info.msgs[2].expectsPayload)
+		assert.True(t, info.msgs[2].expectsCert)
+	})
+
+	t.Run("unknown subtype returns error", func(t *testing.T) {
+		_, err := subtypeInfoFor(99)
+		require.ErrorIs(t, err, ErrUnknownSubtype)
+	})
+}
+
+// registerTestXXInfo temporarily registers XX subtype info for testing.
+func registerTestXXInfo(t *testing.T) {
+	t.Helper()
+	subtypeInfos[header.HandshakeXXPSK0] = subtypeInfo{
+		pattern: noise.HandshakeXX,
+		msgs: []msgFlags{
+			{expectsPayload: true, expectsCert: false},
+			{expectsPayload: true, expectsCert: true},
+			{expectsPayload: false, expectsCert: true},
+		},
+	}
+	t.Cleanup(func() {
+		delete(subtypeInfos, header.HandshakeXXPSK0)
+	})
+}

handshake/payload.go

@@ -0,0 +1,173 @@
+package handshake
+
+import (
+	"errors"
+	"math"
+
+	"google.golang.org/protobuf/encoding/protowire"
+)
+
+var (
+	errInvalidHandshakeMessage = errors.New("invalid handshake message")
+	errInvalidHandshakeDetails = errors.New("invalid handshake details")
+)
+
+// Payload represents the decoded fields of a handshake message.
+// Wire format is protobuf-compatible with NebulaHandshake{Details: NebulaHandshakeDetails{...}}.
+type Payload struct {
+	Cert           []byte
+	InitiatorIndex uint32
+	ResponderIndex uint32
+	Time           uint64
+	CertVersion    uint32
+}
+
+// Proto field numbers for NebulaHandshakeDetails
+const (
+	fieldCert           = 1 // bytes
+	fieldInitiatorIndex = 2 // uint32
+	fieldResponderIndex = 3 // uint32
+	fieldTime           = 5 // uint64
+	fieldCertVersion    = 8 // uint32
+)
+
+// MarshalPayload encodes a handshake payload in protobuf wire format compatible
+// with NebulaHandshake{Details: NebulaHandshakeDetails{...}}.
+// Returns out (which may be nil), with the marshalled Payload appended to it.
+func MarshalPayload(out []byte, p Payload) []byte {
+	var details []byte
+
+	if len(p.Cert) > 0 {
+		details = protowire.AppendTag(details, fieldCert, protowire.BytesType)
+		details = protowire.AppendBytes(details, p.Cert)
+	}
+	if p.InitiatorIndex != 0 {
+		details = protowire.AppendTag(details, fieldInitiatorIndex, protowire.VarintType)
+		details = protowire.AppendVarint(details, uint64(p.InitiatorIndex))
+	}
+	if p.ResponderIndex != 0 {
+		details = protowire.AppendTag(details, fieldResponderIndex, protowire.VarintType)
+		details = protowire.AppendVarint(details, uint64(p.ResponderIndex))
+	}
+	if p.Time != 0 {
+		details = protowire.AppendTag(details, fieldTime, protowire.VarintType)
+		details = protowire.AppendVarint(details, p.Time)
+	}
+	if p.CertVersion != 0 {
+		details = protowire.AppendTag(details, fieldCertVersion, protowire.VarintType)
+		details = protowire.AppendVarint(details, uint64(p.CertVersion))
+	}
+
+	out = protowire.AppendTag(out, 1, protowire.BytesType)
+	out = protowire.AppendBytes(out, details)
+
+	return out
+}
+
+// UnmarshalPayload decodes a protobuf-encoded NebulaHandshake message.
+func UnmarshalPayload(b []byte) (Payload, error) {
+	var p Payload
+
+	for len(b) > 0 {
+		num, typ, n := protowire.ConsumeTag(b)
+		if n < 0 {
+			return p, errInvalidHandshakeMessage
+		}
+		b = b[n:]
+
+		switch {
+		case num == 1 && typ == protowire.BytesType:
+			details, n := protowire.ConsumeBytes(b)
+			if n < 0 {
+				return p, errInvalidHandshakeMessage
+			}
+			b = b[n:]
+			if err := unmarshalPayloadDetails(&p, details); err != nil {
+				return p, err
+			}
+		default:
+			n := protowire.ConsumeFieldValue(num, typ, b)
+			if n < 0 {
+				return p, errInvalidHandshakeMessage
+			}
+			b = b[n:]
+		}
+	}
+
+	return p, nil
+}
+
+func unmarshalPayloadDetails(p *Payload, b []byte) error {
+	for len(b) > 0 {
+		num, typ, n := protowire.ConsumeTag(b)
+		if n < 0 {
+			return errInvalidHandshakeDetails
+		}
+		b = b[n:]
+
+		// For known field numbers, reject any non-matching wire type as a
+		// hard error rather than silently skipping. The caller will catch
+		// missing-field cases downstream, but a wire-type mismatch on a tag
+		// we know is a peer protocol violation worth flagging here.
+		// Repeated occurrences of a singular field follow proto3 last-wins.
+		switch num {
+		case fieldCert:
+			if typ != protowire.BytesType {
+				return errInvalidHandshakeDetails
+			}
+			v, n := protowire.ConsumeBytes(b)
+			if n < 0 {
+				return errInvalidHandshakeDetails
+			}
+			p.Cert = append([]byte(nil), v...)
+			b = b[n:]
+		case fieldInitiatorIndex:
+			if typ != protowire.VarintType {
+				return errInvalidHandshakeDetails
+			}
+			v, n := protowire.ConsumeVarint(b)
+			if n < 0 || v > math.MaxUint32 {
+				return errInvalidHandshakeDetails
+			}
+			p.InitiatorIndex = uint32(v)
+			b = b[n:]
+		case fieldResponderIndex:
+			if typ != protowire.VarintType {
+				return errInvalidHandshakeDetails
+			}
+			v, n := protowire.ConsumeVarint(b)
+			if n < 0 || v > math.MaxUint32 {
+				return errInvalidHandshakeDetails
+			}
+			p.ResponderIndex = uint32(v)
+			b = b[n:]
+		case fieldTime:
+			if typ != protowire.VarintType {
+				return errInvalidHandshakeDetails
+			}
+			v, n := protowire.ConsumeVarint(b)
+			if n < 0 {
+				return errInvalidHandshakeDetails
+			}
+			p.Time = v
+			b = b[n:]
+		case fieldCertVersion:
+			if typ != protowire.VarintType {
+				return errInvalidHandshakeDetails
+			}
+			v, n := protowire.ConsumeVarint(b)
+			if n < 0 || v > math.MaxUint32 {
+				return errInvalidHandshakeDetails
+			}
+			p.CertVersion = uint32(v)
+			b = b[n:]
+		default:
+			n := protowire.ConsumeFieldValue(num, typ, b)
+			if n < 0 {
+				return errInvalidHandshakeDetails
+			}
+			b = b[n:]
+		}
+	}
+	return nil
+}

handshake/payload_test.go

@@ -0,0 +1,361 @@
+package handshake
+
+import (
+	"bytes"
+	"math"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"google.golang.org/protobuf/encoding/protowire"
+)
+
+func TestPayloadRoundTrip(t *testing.T) {
+	t.Run("all fields set", func(t *testing.T) {
+		data := MarshalPayload(nil, Payload{
+			Cert:           []byte("test-cert-bytes"),
+			CertVersion:    2,
+			InitiatorIndex: 12345,
+			ResponderIndex: 67890,
+			Time:           1234567890,
+		})
+
+		got, err := UnmarshalPayload(data)
+		require.NoError(t, err)
+
+		assert.Equal(t, []byte("test-cert-bytes"), got.Cert)
+		assert.Equal(t, uint32(12345), got.InitiatorIndex)
+		assert.Equal(t, uint32(67890), got.ResponderIndex)
+		assert.Equal(t, uint64(1234567890), got.Time)
+		assert.Equal(t, uint32(2), got.CertVersion)
+	})
+
+	t.Run("minimal fields", func(t *testing.T) {
+		data := MarshalPayload(nil, Payload{InitiatorIndex: 1})
+
+		got, err := UnmarshalPayload(data)
+		require.NoError(t, err)
+
+		assert.Equal(t, uint32(1), got.InitiatorIndex)
+		assert.Equal(t, uint32(0), got.ResponderIndex)
+		assert.Equal(t, uint64(0), got.Time)
+		assert.Nil(t, got.Cert)
+	})
+
+	t.Run("empty payload", func(t *testing.T) {
+		data := MarshalPayload(nil, Payload{})
+
+		got, err := UnmarshalPayload(data)
+		require.NoError(t, err)
+
+		assert.Equal(t, uint32(0), got.InitiatorIndex)
+	})
+
+	t.Run("large cert bytes", func(t *testing.T) {
+		bigCert := make([]byte, 4096)
+		for i := range bigCert {
+			bigCert[i] = byte(i % 256)
+		}
+
+		data := MarshalPayload(nil, Payload{
+			Cert:           bigCert,
+			CertVersion:    2,
+			InitiatorIndex: 999,
+		})
+
+		got, err := UnmarshalPayload(data)
+		require.NoError(t, err)
+
+		assert.Equal(t, bigCert, got.Cert)
+		assert.Equal(t, uint32(999), got.InitiatorIndex)
+	})
+
+	t.Run("append to existing buffer", func(t *testing.T) {
+		prefix := []byte("prefix")
+		data := MarshalPayload(prefix, Payload{InitiatorIndex: 42})
+
+		assert.Equal(t, []byte("prefix"), data[:6])
+
+		got, err := UnmarshalPayload(data[6:])
+		require.NoError(t, err)
+		assert.Equal(t, uint32(42), got.InitiatorIndex)
+	})
+}
+
+func TestPayloadUnknownFields(t *testing.T) {
+	t.Run("unknown field in outer message is skipped", func(t *testing.T) {
+		// Marshal a normal payload then append an unknown field (field 99, varint)
+		data := MarshalPayload(nil, Payload{InitiatorIndex: 42})
+		data = protowire.AppendTag(data, 99, protowire.VarintType)
+		data = protowire.AppendVarint(data, 12345)
+
+		got, err := UnmarshalPayload(data)
+		require.NoError(t, err)
+		assert.Equal(t, uint32(42), got.InitiatorIndex)
+	})
+
+	t.Run("unknown field in details is skipped", func(t *testing.T) {
+		// Build details with a known field + unknown field
+		var details []byte
+		details = protowire.AppendTag(details, fieldInitiatorIndex, protowire.VarintType)
+		details = protowire.AppendVarint(details, 77)
+		// Unknown field 50, varint
+		details = protowire.AppendTag(details, 50, protowire.VarintType)
+		details = protowire.AppendVarint(details, 9999)
+		// Another known field after the unknown one
+		details = protowire.AppendTag(details, fieldResponderIndex, protowire.VarintType)
+		details = protowire.AppendVarint(details, 88)
+
+		// Wrap in outer message
+		var data []byte
+		data = protowire.AppendTag(data, 1, protowire.BytesType)
+		data = protowire.AppendBytes(data, details)
+
+		got, err := UnmarshalPayload(data)
+		require.NoError(t, err)
+		assert.Equal(t, uint32(77), got.InitiatorIndex)
+		assert.Equal(t, uint32(88), got.ResponderIndex)
+	})
+
+	t.Run("reserved fields 6 and 7 are skipped", func(t *testing.T) {
+		// Fields 6 and 7 are reserved in the proto definition
+		var details []byte
+		details = protowire.AppendTag(details, fieldInitiatorIndex, protowire.VarintType)
+		details = protowire.AppendVarint(details, 100)
+		details = protowire.AppendTag(details, 6, protowire.VarintType)
+		details = protowire.AppendVarint(details, 1)
+		details = protowire.AppendTag(details, 7, protowire.VarintType)
+		details = protowire.AppendVarint(details, 2)
+
+		var data []byte
+		data = protowire.AppendTag(data, 1, protowire.BytesType)
+		data = protowire.AppendBytes(data, details)
+
+		got, err := UnmarshalPayload(data)
+		require.NoError(t, err)
+		assert.Equal(t, uint32(100), got.InitiatorIndex)
+	})
+}
+
+func TestPayloadBytesConsumed(t *testing.T) {
+	t.Run("all bytes consumed on valid input", func(t *testing.T) {
+		original := Payload{
+			Cert:           []byte("cert"),
+			CertVersion:    2,
+			InitiatorIndex: 100,
+			ResponderIndex: 200,
+			Time:           999,
+		}
+		data := MarshalPayload(nil, original)
+
+		got, err := UnmarshalPayload(data)
+		require.NoError(t, err)
+
+		// Re-marshal and compare — proves we consumed and reproduced all fields
+		remarshaled := MarshalPayload(nil, got)
+		assert.Equal(t, data, remarshaled)
+	})
+}
+
+// wrapDetails wraps raw detail bytes in the outer NebulaHandshake envelope
+// so UnmarshalPayload can reach unmarshalPayloadDetails.
+func wrapDetails(details []byte) []byte {
+	var out []byte
+	out = protowire.AppendTag(out, 1, protowire.BytesType)
+	out = protowire.AppendBytes(out, details)
+	return out
+}
+
+func TestPayloadUnmarshalErrors(t *testing.T) {
+	t.Run("nil input", func(t *testing.T) {
+		got, err := UnmarshalPayload(nil)
+		require.NoError(t, err)
+		assert.Equal(t, uint32(0), got.InitiatorIndex)
+	})
+
+	t.Run("truncated outer tag", func(t *testing.T) {
+		_, err := UnmarshalPayload([]byte{0x80})
+		assert.Error(t, err)
+	})
+
+	t.Run("truncated outer details field", func(t *testing.T) {
+		_, err := UnmarshalPayload([]byte{0x0a, 0x64, 0x01, 0x02, 0x03, 0x04, 0x05})
+		assert.Error(t, err)
+	})
+
+	t.Run("truncated outer unknown field", func(t *testing.T) {
+		// Valid tag for unknown field 99 varint, but no value follows
+		var data []byte
+		data = protowire.AppendTag(data, 99, protowire.VarintType)
+		_, err := UnmarshalPayload(data)
+		assert.Error(t, err)
+	})
+
+	t.Run("truncated details tag", func(t *testing.T) {
+		_, err := UnmarshalPayload(wrapDetails([]byte{0x80}))
+		assert.Error(t, err)
+	})
+
+	t.Run("truncated cert bytes", func(t *testing.T) {
+		// Field 1 (cert), bytes type, length 10 but only 2 bytes
+		var details []byte
+		details = protowire.AppendTag(details, fieldCert, protowire.BytesType)
+		details = append(details, 0x0a, 0x01, 0x02) // length 10, only 2 bytes
+		_, err := UnmarshalPayload(wrapDetails(details))
+		assert.Error(t, err)
+	})
+
+	t.Run("truncated initiator index varint", func(t *testing.T) {
+		var details []byte
+		details = protowire.AppendTag(details, fieldInitiatorIndex, protowire.VarintType)
+		details = append(details, 0x80) // incomplete varint
+		_, err := UnmarshalPayload(wrapDetails(details))
+		assert.Error(t, err)
+	})
+
+	t.Run("truncated responder index varint", func(t *testing.T) {
+		var details []byte
+		details = protowire.AppendTag(details, fieldResponderIndex, protowire.VarintType)
+		details = append(details, 0x80)
+		_, err := UnmarshalPayload(wrapDetails(details))
+		assert.Error(t, err)
+	})
+
+	t.Run("truncated time varint", func(t *testing.T) {
+		var details []byte
+		details = protowire.AppendTag(details, fieldTime, protowire.VarintType)
+		details = append(details, 0x80)
+		_, err := UnmarshalPayload(wrapDetails(details))
+		assert.Error(t, err)
+	})
+
+	t.Run("truncated cert version varint", func(t *testing.T) {
+		var details []byte
+		details = protowire.AppendTag(details, fieldCertVersion, protowire.VarintType)
+		details = append(details, 0x80)
+		_, err := UnmarshalPayload(wrapDetails(details))
+		assert.Error(t, err)
+	})
+
+	t.Run("truncated unknown field in details", func(t *testing.T) {
+		var details []byte
+		details = protowire.AppendTag(details, 50, protowire.VarintType)
+		details = append(details, 0x80) // incomplete varint
+		_, err := UnmarshalPayload(wrapDetails(details))
+		assert.Error(t, err)
+	})
+
+	t.Run("cert with wrong wire type rejected", func(t *testing.T) {
+		// fieldCert as Varint instead of Bytes.
+		var details []byte
+		details = protowire.AppendTag(details, fieldCert, protowire.VarintType)
+		details = protowire.AppendVarint(details, 42)
+		_, err := UnmarshalPayload(wrapDetails(details))
+		assert.Error(t, err)
+	})
+
+	t.Run("initiator index with wrong wire type rejected", func(t *testing.T) {
+		// fieldInitiatorIndex as Bytes instead of Varint.
+		var details []byte
+		details = protowire.AppendTag(details, fieldInitiatorIndex, protowire.BytesType)
+		details = protowire.AppendBytes(details, []byte{1, 2, 3})
+		_, err := UnmarshalPayload(wrapDetails(details))
+		assert.Error(t, err)
+	})
+
+	t.Run("time with wrong wire type rejected", func(t *testing.T) {
+		var details []byte
+		details = protowire.AppendTag(details, fieldTime, protowire.BytesType)
+		details = protowire.AppendBytes(details, []byte{1, 2, 3})
+		_, err := UnmarshalPayload(wrapDetails(details))
+		assert.Error(t, err)
+	})
+
+	t.Run("cert version with wrong wire type rejected", func(t *testing.T) {
+		var details []byte
+		details = protowire.AppendTag(details, fieldCertVersion, protowire.BytesType)
+		details = protowire.AppendBytes(details, []byte{1, 2, 3})
+		_, err := UnmarshalPayload(wrapDetails(details))
+		assert.Error(t, err)
+	})
+
+	t.Run("repeated singular field follows proto3 last-wins", func(t *testing.T) {
+		// Per proto3, multiple instances of a singular field are accepted and
+		// the last value wins. We keep this behavior so that peers using
+		// alternative encoders aren't rejected.
+		var details []byte
+		details = protowire.AppendTag(details, fieldInitiatorIndex, protowire.VarintType)
+		details = protowire.AppendVarint(details, 1)
+		details = protowire.AppendTag(details, fieldInitiatorIndex, protowire.VarintType)
+		details = protowire.AppendVarint(details, 42)
+		got, err := UnmarshalPayload(wrapDetails(details))
+		require.NoError(t, err)
+		assert.Equal(t, uint32(42), got.InitiatorIndex)
+	})
+
+	t.Run("initiator index varint overflow rejected", func(t *testing.T) {
+		var details []byte
+		details = protowire.AppendTag(details, fieldInitiatorIndex, protowire.VarintType)
+		details = protowire.AppendVarint(details, math.MaxUint32+1)
+		_, err := UnmarshalPayload(wrapDetails(details))
+		assert.Error(t, err)
+	})
+
+	t.Run("cert version varint overflow rejected", func(t *testing.T) {
+		var details []byte
+		details = protowire.AppendTag(details, fieldCertVersion, protowire.VarintType)
+		details = protowire.AppendVarint(details, math.MaxUint32+1)
+		_, err := UnmarshalPayload(wrapDetails(details))
+		assert.Error(t, err)
+	})
+
+}
+
+// FuzzPayload feeds arbitrary bytes through UnmarshalPayload to confirm it
+// never panics, and for any input that parses cleanly, that re-marshal +
+// re-parse is a fix-point. Inputs come from an authenticated peer (post-
+// noise-decrypt), so the threat model is "valid peer behaving arbitrarily,"
+// not "unauthenticated injection."
+func FuzzPayload(f *testing.F) {
+	// Seed corpus with a handful of known-good shapes.
+	f.Add(MarshalPayload(nil, Payload{}))
+	f.Add(MarshalPayload(nil, Payload{Cert: []byte{1, 2, 3}, CertVersion: 2}))
+	f.Add(MarshalPayload(nil, Payload{InitiatorIndex: 42, Time: 1}))
+	f.Add(MarshalPayload(nil, Payload{
+		Cert:           []byte("seed-cert"),
+		InitiatorIndex: 1,
+		ResponderIndex: 2,
+		Time:           3,
+		CertVersion:    2,
+	}))
+	f.Add([]byte{})
+	f.Add([]byte{0xff})
+
+	f.Fuzz(func(t *testing.T, data []byte) {
+		p1, err := UnmarshalPayload(data)
+		if err != nil {
+			return
+		}
+
+		// For any input that parses, re-marshaling and re-parsing must
+		// yield an equivalent Payload. This catches dispatch bugs (e.g.
+		// emitting a field on marshal that we don't accept on parse) and
+		// any non-idempotent parsing behavior.
+		b2 := MarshalPayload(nil, p1)
+		p2, err := UnmarshalPayload(b2)
+		if err != nil {
+			t.Fatalf("re-parse of self-marshaled payload failed: %v\nintermediate: %x\n", err, b2)
+		}
+		if !payloadsEqual(p1, p2) {
+			t.Fatalf("re-marshal not idempotent\nfirst:  %+v\nsecond: %+v", p1, p2)
+		}
+	})
+}
+
+func payloadsEqual(a, b Payload) bool {
+	return bytes.Equal(a.Cert, b.Cert) &&
+		a.InitiatorIndex == b.InitiatorIndex &&
+		a.ResponderIndex == b.ResponderIndex &&
+		a.Time == b.Time &&
+		a.CertVersion == b.CertVersion
+}

handshake_ix.go

@@ -1,683 +0,0 @@
-package nebula
-
-import (
-	"net/netip"
-	"slices"
-	"time"
-
-	"github.com/flynn/noise"
-	"github.com/sirupsen/logrus"
-	"github.com/slackhq/nebula/cert"
-	"github.com/slackhq/nebula/header"
-)
-
-// NOISE IX Handshakes
-
-// This function constructs a handshake packet, but does not actually send it
-// Sending is done by the handshake manager
-func ixHandshakeStage0(f *Interface, hh *HandshakeHostInfo) bool {
-	err := f.handshakeManager.allocateIndex(hh)
-	if err != nil {
-		f.l.WithError(err).WithField("vpnAddrs", hh.hostinfo.vpnAddrs).
-			WithField("handshake", m{"stage": 0, "style": "ix_psk0"}).Error("Failed to generate index")
-		return false
-	}
-
-	cs := f.pki.getCertState()
-	v := cs.initiatingVersion
-	if hh.initiatingVersionOverride != cert.VersionPre1 {
-		v = hh.initiatingVersionOverride
-	} else if v < cert.Version2 {
-		// If we're connecting to a v6 address we should encourage use of a V2 cert
-		for _, a := range hh.hostinfo.vpnAddrs {
-			if a.Is6() {
-				v = cert.Version2
-				break
-			}
-		}
-	}
-
-	crt := cs.getCertificate(v)
-	if crt == nil {
-		f.l.WithField("vpnAddrs", hh.hostinfo.vpnAddrs).
-			WithField("handshake", m{"stage": 0, "style": "ix_psk0"}).
-			WithField("certVersion", v).
-			Error("Unable to handshake with host because no certificate is available")
-		return false
-	}
-
-	crtHs := cs.getHandshakeBytes(v)
-	if crtHs == nil {
-		f.l.WithField("vpnAddrs", hh.hostinfo.vpnAddrs).
-			WithField("handshake", m{"stage": 0, "style": "ix_psk0"}).
-			WithField("certVersion", v).
-			Error("Unable to handshake with host because no certificate handshake bytes is available")
-		return false
-	}
-
-	ci, err := NewConnectionState(f.l, cs, crt, true, noise.HandshakeIX)
-	if err != nil {
-		f.l.WithError(err).WithField("vpnAddrs", hh.hostinfo.vpnAddrs).
-			WithField("handshake", m{"stage": 0, "style": "ix_psk0"}).
-			WithField("certVersion", v).
-			Error("Failed to create connection state")
-		return false
-	}
-	hh.hostinfo.ConnectionState = ci
-
-	hs := &NebulaHandshake{
-		Details: &NebulaHandshakeDetails{
-			InitiatorIndex: hh.hostinfo.localIndexId,
-			Time:           uint64(time.Now().UnixNano()),
-			Cert:           crtHs,
-			CertVersion:    uint32(v),
-		},
-	}
-
-	hsBytes, err := hs.Marshal()
-	if err != nil {
-		f.l.WithError(err).WithField("vpnAddrs", hh.hostinfo.vpnAddrs).
-			WithField("certVersion", v).
-			WithField("handshake", m{"stage": 0, "style": "ix_psk0"}).Error("Failed to marshal handshake message")
-		return false
-	}
-
-	h := header.Encode(make([]byte, header.Len), header.Version, header.Handshake, header.HandshakeIXPSK0, 0, 1)
-
-	msg, _, _, err := ci.H.WriteMessage(h, hsBytes)
-	if err != nil {
-		f.l.WithError(err).WithField("vpnAddrs", hh.hostinfo.vpnAddrs).
-			WithField("handshake", m{"stage": 0, "style": "ix_psk0"}).Error("Failed to call noise.WriteMessage")
-		return false
-	}
-
-	// We are sending handshake packet 1, so we don't expect to receive
-	// handshake packet 1 from the responder
-	ci.window.Update(f.l, 1)
-
-	hh.hostinfo.HandshakePacket[0] = msg
-	hh.ready = true
-	return true
-}
-
-func ixHandshakeStage1(f *Interface, addr netip.AddrPort, via *ViaSender, packet []byte, h *header.H) {
-	cs := f.pki.getCertState()
-	crt := cs.GetDefaultCertificate()
-	if crt == nil {
-		f.l.WithField("udpAddr", addr).
-			WithField("handshake", m{"stage": 0, "style": "ix_psk0"}).
-			WithField("certVersion", cs.initiatingVersion).
-			Error("Unable to handshake with host because no certificate is available")
-		return
-	}
-
-	ci, err := NewConnectionState(f.l, cs, crt, false, noise.HandshakeIX)
-	if err != nil {
-		f.l.WithError(err).WithField("udpAddr", addr).
-			WithField("handshake", m{"stage": 1, "style": "ix_psk0"}).
-			Error("Failed to create connection state")
-		return
-	}
-
-	// Mark packet 1 as seen so it doesn't show up as missed
-	ci.window.Update(f.l, 1)
-
-	msg, _, _, err := ci.H.ReadMessage(nil, packet[header.Len:])
-	if err != nil {
-		f.l.WithError(err).WithField("udpAddr", addr).
-			WithField("handshake", m{"stage": 1, "style": "ix_psk0"}).
-			Error("Failed to call noise.ReadMessage")
-		return
-	}
-
-	hs := &NebulaHandshake{}
-	err = hs.Unmarshal(msg)
-	if err != nil || hs.Details == nil {
-		f.l.WithError(err).WithField("udpAddr", addr).
-			WithField("handshake", m{"stage": 1, "style": "ix_psk0"}).
-			Error("Failed unmarshal handshake message")
-		return
-	}
-
-	rc, err := cert.Recombine(cert.Version(hs.Details.CertVersion), hs.Details.Cert, ci.H.PeerStatic(), ci.Curve())
-	if err != nil {
-		f.l.WithError(err).WithField("udpAddr", addr).
-			WithField("handshake", m{"stage": 1, "style": "ix_psk0"}).
-			Info("Handshake did not contain a certificate")
-		return
-	}
-
-	remoteCert, err := f.pki.GetCAPool().VerifyCertificate(time.Now(), rc)
-	if err != nil {
-		fp, fperr := rc.Fingerprint()
-		if fperr != nil {
-			fp = "<error generating certificate fingerprint>"
-		}
-
-		e := f.l.WithError(err).WithField("udpAddr", addr).
-			WithField("handshake", m{"stage": 1, "style": "ix_psk0"}).
-			WithField("certVpnNetworks", rc.Networks()).
-			WithField("certFingerprint", fp)
-
-		if f.l.Level >= logrus.DebugLevel {
-			e = e.WithField("cert", rc)
-		}
-
-		e.Info("Invalid certificate from host")
-		return
-	}
-
-	if remoteCert.Certificate.Version() != ci.myCert.Version() {
-		// We started off using the wrong certificate version, lets see if we can match the version that was sent to us
-		myCertOtherVersion := cs.getCertificate(remoteCert.Certificate.Version())
-		if myCertOtherVersion == nil {
-			if f.l.Level >= logrus.DebugLevel {
-				f.l.WithError(err).WithFields(m{
-					"udpAddr":   addr,
-					"handshake": m{"stage": 1, "style": "ix_psk0"},
-					"cert":      remoteCert,
-				}).Debug("Might be unable to handshake with host due to missing certificate version")
-			}
-		} else {
-			// Record the certificate we are actually using
-			ci.myCert = myCertOtherVersion
-		}
-	}
-
-	if len(remoteCert.Certificate.Networks()) == 0 {
-		f.l.WithError(err).WithField("udpAddr", addr).
-			WithField("cert", remoteCert).
-			WithField("handshake", m{"stage": 1, "style": "ix_psk0"}).
-			Info("No networks in certificate")
-		return
-	}
-
-	var vpnAddrs []netip.Addr
-	var filteredNetworks []netip.Prefix
-	certName := remoteCert.Certificate.Name()
-	certVersion := remoteCert.Certificate.Version()
-	fingerprint := remoteCert.Fingerprint
-	issuer := remoteCert.Certificate.Issuer()
-
-	for _, network := range remoteCert.Certificate.Networks() {
-		vpnAddr := network.Addr()
-		if f.myVpnAddrsTable.Contains(vpnAddr) {
-			f.l.WithField("vpnAddr", vpnAddr).WithField("udpAddr", addr).
-				WithField("certName", certName).
-				WithField("certVersion", certVersion).
-				WithField("fingerprint", fingerprint).
-				WithField("issuer", issuer).
-				WithField("handshake", m{"stage": 1, "style": "ix_psk0"}).Error("Refusing to handshake with myself")
-			return
-		}
-
-		// vpnAddrs outside our vpn networks are of no use to us, filter them out
-		if !f.myVpnNetworksTable.Contains(vpnAddr) {
-			continue
-		}
-
-		filteredNetworks = append(filteredNetworks, network)
-		vpnAddrs = append(vpnAddrs, vpnAddr)
-	}
-
-	if len(vpnAddrs) == 0 {
-		f.l.WithError(err).WithField("udpAddr", addr).
-			WithField("certName", certName).
-			WithField("certVersion", certVersion).
-			WithField("fingerprint", fingerprint).
-			WithField("issuer", issuer).
-			WithField("handshake", m{"stage": 1, "style": "ix_psk0"}).Error("No usable vpn addresses from host, refusing handshake")
-		return
-	}
-
-	if addr.IsValid() {
-		// addr can be invalid when the tunnel is being relayed.
-		// We only want to apply the remote allow list for direct tunnels here
-		if !f.lightHouse.GetRemoteAllowList().AllowAll(vpnAddrs, addr.Addr()) {
-			f.l.WithField("vpnAddrs", vpnAddrs).WithField("udpAddr", addr).Debug("lighthouse.remote_allow_list denied incoming handshake")
-			return
-		}
-	}
-
-	myIndex, err := generateIndex(f.l)
-	if err != nil {
-		f.l.WithError(err).WithField("vpnAddrs", vpnAddrs).WithField("udpAddr", addr).
-			WithField("certName", certName).
-			WithField("certVersion", certVersion).
-			WithField("fingerprint", fingerprint).
-			WithField("issuer", issuer).
-			WithField("handshake", m{"stage": 1, "style": "ix_psk0"}).Error("Failed to generate index")
-		return
-	}
-
-	hostinfo := &HostInfo{
-		ConnectionState:   ci,
-		localIndexId:      myIndex,
-		remoteIndexId:     hs.Details.InitiatorIndex,
-		vpnAddrs:          vpnAddrs,
-		HandshakePacket:   make(map[uint8][]byte, 0),
-		lastHandshakeTime: hs.Details.Time,
-		relayState: RelayState{
-			relays:         nil,
-			relayForByAddr: map[netip.Addr]*Relay{},
-			relayForByIdx:  map[uint32]*Relay{},
-		},
-	}
-
-	f.l.WithField("vpnAddrs", vpnAddrs).WithField("udpAddr", addr).
-		WithField("certName", certName).
-		WithField("certVersion", certVersion).
-		WithField("fingerprint", fingerprint).
-		WithField("issuer", issuer).
-		WithField("initiatorIndex", hs.Details.InitiatorIndex).WithField("responderIndex", hs.Details.ResponderIndex).
-		WithField("remoteIndex", h.RemoteIndex).WithField("handshake", m{"stage": 1, "style": "ix_psk0"}).
-		Info("Handshake message received")
-
-	hs.Details.ResponderIndex = myIndex
-	hs.Details.Cert = cs.getHandshakeBytes(ci.myCert.Version())
-	if hs.Details.Cert == nil {
-		f.l.WithField("vpnAddrs", vpnAddrs).WithField("udpAddr", addr).
-			WithField("certName", certName).
-			WithField("certVersion", certVersion).
-			WithField("fingerprint", fingerprint).
-			WithField("issuer", issuer).
-			WithField("initiatorIndex", hs.Details.InitiatorIndex).WithField("responderIndex", hs.Details.ResponderIndex).
-			WithField("remoteIndex", h.RemoteIndex).WithField("handshake", m{"stage": 1, "style": "ix_psk0"}).
-			WithField("certVersion", ci.myCert.Version()).
-			Error("Unable to handshake with host because no certificate handshake bytes is available")
-		return
-	}
-
-	hs.Details.CertVersion = uint32(ci.myCert.Version())
-	// Update the time in case their clock is way off from ours
-	hs.Details.Time = uint64(time.Now().UnixNano())
-
-	hsBytes, err := hs.Marshal()
-	if err != nil {
-		f.l.WithError(err).WithField("vpnAddrs", hostinfo.vpnAddrs).WithField("udpAddr", addr).
-			WithField("certName", certName).
-			WithField("certVersion", certVersion).
-			WithField("fingerprint", fingerprint).
-			WithField("issuer", issuer).
-			WithField("handshake", m{"stage": 1, "style": "ix_psk0"}).Error("Failed to marshal handshake message")
-		return
-	}
-
-	nh := header.Encode(make([]byte, header.Len), header.Version, header.Handshake, header.HandshakeIXPSK0, hs.Details.InitiatorIndex, 2)
-	msg, dKey, eKey, err := ci.H.WriteMessage(nh, hsBytes)
-	if err != nil {
-		f.l.WithError(err).WithField("vpnAddrs", hostinfo.vpnAddrs).WithField("udpAddr", addr).
-			WithField("certName", certName).
-			WithField("certVersion", certVersion).
-			WithField("fingerprint", fingerprint).
-			WithField("issuer", issuer).
-			WithField("handshake", m{"stage": 1, "style": "ix_psk0"}).Error("Failed to call noise.WriteMessage")
-		return
-	} else if dKey == nil || eKey == nil {
-		f.l.WithField("vpnAddrs", hostinfo.vpnAddrs).WithField("udpAddr", addr).
-			WithField("certName", certName).
-			WithField("certVersion", certVersion).
-			WithField("fingerprint", fingerprint).
-			WithField("issuer", issuer).
-			WithField("handshake", m{"stage": 1, "style": "ix_psk0"}).Error("Noise did not arrive at a key")
-		return
-	}
-
-	hostinfo.HandshakePacket[0] = make([]byte, len(packet[header.Len:]))
-	copy(hostinfo.HandshakePacket[0], packet[header.Len:])
-
-	// Regardless of whether you are the sender or receiver, you should arrive here
-	// and complete standing up the connection.
-	hostinfo.HandshakePacket[2] = make([]byte, len(msg))
-	copy(hostinfo.HandshakePacket[2], msg)
-
-	// We are sending handshake packet 2, so we don't expect to receive
-	// handshake packet 2 from the initiator.
-	ci.window.Update(f.l, 2)
-
-	ci.peerCert = remoteCert
-	ci.dKey = NewNebulaCipherState(dKey)
-	ci.eKey = NewNebulaCipherState(eKey)
-
-	hostinfo.remotes = f.lightHouse.QueryCache(vpnAddrs)
-	hostinfo.SetRemote(addr)
-	hostinfo.buildNetworks(filteredNetworks, remoteCert.Certificate.UnsafeNetworks())
-
-	existing, err := f.handshakeManager.CheckAndComplete(hostinfo, 0, f)
-	if err != nil {
-		switch err {
-		case ErrAlreadySeen:
-			// Update remote if preferred
-			if existing.SetRemoteIfPreferred(f.hostMap, addr) {
-				// Send a test packet to ensure the other side has also switched to
-				// the preferred remote
-				f.SendMessageToVpnAddr(header.Test, header.TestRequest, vpnAddrs[0], []byte(""), make([]byte, 12, 12), make([]byte, mtu))
-			}
-
-			msg = existing.HandshakePacket[2]
-			f.messageMetrics.Tx(header.Handshake, header.MessageSubType(msg[1]), 1)
-			if addr.IsValid() {
-				err := f.outside.WriteTo(msg, addr)
-				if err != nil {
-					f.l.WithField("vpnAddrs", existing.vpnAddrs).WithField("udpAddr", addr).
-						WithField("handshake", m{"stage": 2, "style": "ix_psk0"}).WithField("cached", true).
-						WithError(err).Error("Failed to send handshake message")
-				} else {
-					f.l.WithField("vpnAddrs", existing.vpnAddrs).WithField("udpAddr", addr).
-						WithField("handshake", m{"stage": 2, "style": "ix_psk0"}).WithField("cached", true).
-						Info("Handshake message sent")
-				}
-				return
-			} else {
-				if via == nil {
-					f.l.Error("Handshake send failed: both addr and via are nil.")
-					return
-				}
-				hostinfo.relayState.InsertRelayTo(via.relayHI.vpnAddrs[0])
-				f.SendVia(via.relayHI, via.relay, msg, make([]byte, 12), make([]byte, mtu), false)
-				f.l.WithField("vpnAddrs", existing.vpnAddrs).WithField("relay", via.relayHI.vpnAddrs[0]).
-					WithField("handshake", m{"stage": 2, "style": "ix_psk0"}).WithField("cached", true).
-					Info("Handshake message sent")
-				return
-			}
-		case ErrExistingHostInfo:
-			// This means there was an existing tunnel and this handshake was older than the one we are currently based on
-			f.l.WithField("vpnAddrs", vpnAddrs).WithField("udpAddr", addr).
-				WithField("certName", certName).
-				WithField("certVersion", certVersion).
-				WithField("oldHandshakeTime", existing.lastHandshakeTime).
-				WithField("newHandshakeTime", hostinfo.lastHandshakeTime).
-				WithField("fingerprint", fingerprint).
-				WithField("issuer", issuer).
-				WithField("initiatorIndex", hs.Details.InitiatorIndex).WithField("responderIndex", hs.Details.ResponderIndex).
-				WithField("remoteIndex", h.RemoteIndex).WithField("handshake", m{"stage": 1, "style": "ix_psk0"}).
-				Info("Handshake too old")
-
-			// Send a test packet to trigger an authenticated tunnel test, this should suss out any lingering tunnel issues
-			f.SendMessageToVpnAddr(header.Test, header.TestRequest, vpnAddrs[0], []byte(""), make([]byte, 12, 12), make([]byte, mtu))
-			return
-		case ErrLocalIndexCollision:
-			// This means we failed to insert because of collision on localIndexId. Just let the next handshake packet retry
-			f.l.WithField("vpnAddrs", vpnAddrs).WithField("udpAddr", addr).
-				WithField("certName", certName).
-				WithField("certVersion", certVersion).
-				WithField("fingerprint", fingerprint).
-				WithField("issuer", issuer).
-				WithField("initiatorIndex", hs.Details.InitiatorIndex).WithField("responderIndex", hs.Details.ResponderIndex).
-				WithField("remoteIndex", h.RemoteIndex).WithField("handshake", m{"stage": 1, "style": "ix_psk0"}).
-				WithField("localIndex", hostinfo.localIndexId).WithField("collision", existing.vpnAddrs).
-				Error("Failed to add HostInfo due to localIndex collision")
-			return
-		default:
-			// Shouldn't happen, but just in case someone adds a new error type to CheckAndComplete
-			// And we forget to update it here
-			f.l.WithError(err).WithField("vpnAddrs", vpnAddrs).WithField("udpAddr", addr).
-				WithField("certName", certName).
-				WithField("certVersion", certVersion).
-				WithField("fingerprint", fingerprint).
-				WithField("issuer", issuer).
-				WithField("initiatorIndex", hs.Details.InitiatorIndex).WithField("responderIndex", hs.Details.ResponderIndex).
-				WithField("remoteIndex", h.RemoteIndex).WithField("handshake", m{"stage": 1, "style": "ix_psk0"}).
-				Error("Failed to add HostInfo to HostMap")
-			return
-		}
-	}
-
-	// Do the send
-	f.messageMetrics.Tx(header.Handshake, header.MessageSubType(msg[1]), 1)
-	if addr.IsValid() {
-		err = f.outside.WriteTo(msg, addr)
-		if err != nil {
-			f.l.WithField("vpnAddrs", vpnAddrs).WithField("udpAddr", addr).
-				WithField("certName", certName).
-				WithField("certVersion", certVersion).
-				WithField("fingerprint", fingerprint).
-				WithField("issuer", issuer).
-				WithField("initiatorIndex", hs.Details.InitiatorIndex).WithField("responderIndex", hs.Details.ResponderIndex).
-				WithField("remoteIndex", h.RemoteIndex).WithField("handshake", m{"stage": 2, "style": "ix_psk0"}).
-				WithError(err).Error("Failed to send handshake")
-		} else {
-			f.l.WithField("vpnAddrs", vpnAddrs).WithField("udpAddr", addr).
-				WithField("certName", certName).
-				WithField("certVersion", certVersion).
-				WithField("fingerprint", fingerprint).
-				WithField("issuer", issuer).
-				WithField("initiatorIndex", hs.Details.InitiatorIndex).WithField("responderIndex", hs.Details.ResponderIndex).
-				WithField("remoteIndex", h.RemoteIndex).WithField("handshake", m{"stage": 2, "style": "ix_psk0"}).
-				Info("Handshake message sent")
-		}
-	} else {
-		if via == nil {
-			f.l.Error("Handshake send failed: both addr and via are nil.")
-			return
-		}
-		hostinfo.relayState.InsertRelayTo(via.relayHI.vpnAddrs[0])
-		// I successfully received a handshake. Just in case I marked this tunnel as 'Disestablished', ensure
-		// it's correctly marked as working.
-		via.relayHI.relayState.UpdateRelayForByIdxState(via.remoteIdx, Established)
-		f.SendVia(via.relayHI, via.relay, msg, make([]byte, 12), make([]byte, mtu), false)
-		f.l.WithField("vpnAddrs", vpnAddrs).WithField("relay", via.relayHI.vpnAddrs[0]).
-			WithField("certName", certName).
-			WithField("certVersion", certVersion).
-			WithField("fingerprint", fingerprint).
-			WithField("issuer", issuer).
-			WithField("initiatorIndex", hs.Details.InitiatorIndex).WithField("responderIndex", hs.Details.ResponderIndex).
-			WithField("remoteIndex", h.RemoteIndex).WithField("handshake", m{"stage": 2, "style": "ix_psk0"}).
-			Info("Handshake message sent")
-	}
-
-	f.connectionManager.AddTrafficWatch(hostinfo)
-
-	hostinfo.remotes.RefreshFromHandshake(vpnAddrs)
-
-	return
-}
-
-func ixHandshakeStage2(f *Interface, addr netip.AddrPort, via *ViaSender, hh *HandshakeHostInfo, packet []byte, h *header.H) bool {
-	if hh == nil {
-		// Nothing here to tear down, got a bogus stage 2 packet
-		return true
-	}
-
-	hh.Lock()
-	defer hh.Unlock()
-
-	hostinfo := hh.hostinfo
-	if addr.IsValid() {
-		// The vpnAddr we know about is the one we tried to handshake with, use it to apply the remote allow list.
-		if !f.lightHouse.GetRemoteAllowList().AllowAll(hostinfo.vpnAddrs, addr.Addr()) {
-			f.l.WithField("vpnAddrs", hostinfo.vpnAddrs).WithField("udpAddr", addr).Debug("lighthouse.remote_allow_list denied incoming handshake")
-			return false
-		}
-	}
-
-	ci := hostinfo.ConnectionState
-	msg, eKey, dKey, err := ci.H.ReadMessage(nil, packet[header.Len:])
-	if err != nil {
-		f.l.WithError(err).WithField("vpnAddrs", hostinfo.vpnAddrs).WithField("udpAddr", addr).
-			WithField("handshake", m{"stage": 2, "style": "ix_psk0"}).WithField("header", h).
-			Error("Failed to call noise.ReadMessage")
-
-		// We don't want to tear down the connection on a bad ReadMessage because it could be an attacker trying
-		// to DOS us. Every other error condition after should to allow a possible good handshake to complete in the
-		// near future
-		return false
-	} else if dKey == nil || eKey == nil {
-		f.l.WithField("vpnAddrs", hostinfo.vpnAddrs).WithField("udpAddr", addr).
-			WithField("handshake", m{"stage": 2, "style": "ix_psk0"}).
-			Error("Noise did not arrive at a key")
-
-		// This should be impossible in IX but just in case, if we get here then there is no chance to recover
-		// the handshake state machine. Tear it down
-		return true
-	}
-
-	hs := &NebulaHandshake{}
-	err = hs.Unmarshal(msg)
-	if err != nil || hs.Details == nil {
-		f.l.WithError(err).WithField("vpnAddrs", hostinfo.vpnAddrs).WithField("udpAddr", addr).
-			WithField("handshake", m{"stage": 2, "style": "ix_psk0"}).Error("Failed unmarshal handshake message")
-
-		// The handshake state machine is complete, if things break now there is no chance to recover. Tear down and start again
-		return true
-	}
-
-	rc, err := cert.Recombine(cert.Version(hs.Details.CertVersion), hs.Details.Cert, ci.H.PeerStatic(), ci.Curve())
-	if err != nil {
-		f.l.WithError(err).WithField("udpAddr", addr).
-			WithField("vpnAddrs", hostinfo.vpnAddrs).
-			WithField("handshake", m{"stage": 2, "style": "ix_psk0"}).
-			Info("Handshake did not contain a certificate")
-		return true
-	}
-
-	remoteCert, err := f.pki.GetCAPool().VerifyCertificate(time.Now(), rc)
-	if err != nil {
-		fp, err := rc.Fingerprint()
-		if err != nil {
-			fp = "<error generating certificate fingerprint>"
-		}
-
-		e := f.l.WithError(err).WithField("udpAddr", addr).
-			WithField("vpnAddrs", hostinfo.vpnAddrs).
-			WithField("handshake", m{"stage": 2, "style": "ix_psk0"}).
-			WithField("certFingerprint", fp).
-			WithField("certVpnNetworks", rc.Networks())
-
-		if f.l.Level >= logrus.DebugLevel {
-			e = e.WithField("cert", rc)
-		}
-
-		e.Info("Invalid certificate from host")
-		return true
-	}
-
-	if len(remoteCert.Certificate.Networks()) == 0 {
-		f.l.WithError(err).WithField("udpAddr", addr).
-			WithField("vpnAddrs", hostinfo.vpnAddrs).
-			WithField("cert", remoteCert).
-			WithField("handshake", m{"stage": 2, "style": "ix_psk0"}).
-			Info("No networks in certificate")
-		return true
-	}
-
-	vpnNetworks := remoteCert.Certificate.Networks()
-	certName := remoteCert.Certificate.Name()
-	certVersion := remoteCert.Certificate.Version()
-	fingerprint := remoteCert.Fingerprint
-	issuer := remoteCert.Certificate.Issuer()
-
-	hostinfo.remoteIndexId = hs.Details.ResponderIndex
-	hostinfo.lastHandshakeTime = hs.Details.Time
-
-	// Store their cert and our symmetric keys
-	ci.peerCert = remoteCert
-	ci.dKey = NewNebulaCipherState(dKey)
-	ci.eKey = NewNebulaCipherState(eKey)
-
-	// Make sure the current udpAddr being used is set for responding
-	if addr.IsValid() {
-		hostinfo.SetRemote(addr)
-	} else {
-		hostinfo.relayState.InsertRelayTo(via.relayHI.vpnAddrs[0])
-	}
-
-	var vpnAddrs []netip.Addr
-	var filteredNetworks []netip.Prefix
-	for _, network := range vpnNetworks {
-		// vpnAddrs outside our vpn networks are of no use to us, filter them out
-		vpnAddr := network.Addr()
-		if !f.myVpnNetworksTable.Contains(vpnAddr) {
-			continue
-		}
-
-		filteredNetworks = append(filteredNetworks, network)
-		vpnAddrs = append(vpnAddrs, vpnAddr)
-	}
-
-	if len(vpnAddrs) == 0 {
-		f.l.WithError(err).WithField("udpAddr", addr).
-			WithField("certName", certName).
-			WithField("certVersion", certVersion).
-			WithField("fingerprint", fingerprint).
-			WithField("issuer", issuer).
-			WithField("handshake", m{"stage": 2, "style": "ix_psk0"}).Error("No usable vpn addresses from host, refusing handshake")
-		return true
-	}
-
-	// Ensure the right host responded
-	if !slices.Contains(vpnAddrs, hostinfo.vpnAddrs[0]) {
-		f.l.WithField("intendedVpnAddrs", hostinfo.vpnAddrs).WithField("haveVpnNetworks", vpnNetworks).
-			WithField("udpAddr", addr).
-			WithField("certName", certName).
-			WithField("certVersion", certVersion).
-			WithField("handshake", m{"stage": 2, "style": "ix_psk0"}).
-			Info("Incorrect host responded to handshake")
-
-		// Release our old handshake from pending, it should not continue
-		f.handshakeManager.DeleteHostInfo(hostinfo)
-
-		// Create a new hostinfo/handshake for the intended vpn ip
-		f.handshakeManager.StartHandshake(hostinfo.vpnAddrs[0], func(newHH *HandshakeHostInfo) {
-			// Block the current used address
-			newHH.hostinfo.remotes = hostinfo.remotes
-			newHH.hostinfo.remotes.BlockRemote(addr)
-
-			f.l.WithField("blockedUdpAddrs", newHH.hostinfo.remotes.CopyBlockedRemotes()).
-				WithField("vpnNetworks", vpnNetworks).
-				WithField("remotes", newHH.hostinfo.remotes.CopyAddrs(f.hostMap.GetPreferredRanges())).
-				Info("Blocked addresses for handshakes")
-
-			// Swap the packet store to benefit the original intended recipient
-			newHH.packetStore = hh.packetStore
-			hh.packetStore = []*cachedPacket{}
-
-			// Finally, put the correct vpn addrs in the host info, tell them to close the tunnel, and return true to tear down
-			hostinfo.vpnAddrs = vpnAddrs
-			f.sendCloseTunnel(hostinfo)
-		})
-
-		return true
-	}
-
-	// Mark packet 2 as seen so it doesn't show up as missed
-	ci.window.Update(f.l, 2)
-
-	duration := time.Since(hh.startTime).Nanoseconds()
-	f.l.WithField("vpnAddrs", vpnAddrs).WithField("udpAddr", addr).
-		WithField("certName", certName).
-		WithField("certVersion", certVersion).
-		WithField("fingerprint", fingerprint).
-		WithField("issuer", issuer).
-		WithField("initiatorIndex", hs.Details.InitiatorIndex).WithField("responderIndex", hs.Details.ResponderIndex).
-		WithField("remoteIndex", h.RemoteIndex).WithField("handshake", m{"stage": 2, "style": "ix_psk0"}).
-		WithField("durationNs", duration).
-		WithField("sentCachedPackets", len(hh.packetStore)).
-		Info("Handshake message received")
-
-	// Build up the radix for the firewall if we have subnets in the cert
-	hostinfo.vpnAddrs = vpnAddrs
-	hostinfo.buildNetworks(filteredNetworks, remoteCert.Certificate.UnsafeNetworks())
-
-	// Complete our handshake and update metrics, this will replace any existing tunnels for the vpnAddrs here
-	f.handshakeManager.Complete(hostinfo, f)
-	f.connectionManager.AddTrafficWatch(hostinfo)
-
-	if f.l.Level >= logrus.DebugLevel {
-		hostinfo.logger(f.l).Debugf("Sending %d stored packets", len(hh.packetStore))
-	}
-
-	if len(hh.packetStore) > 0 {
-		nb := make([]byte, 12, 12)
-		out := make([]byte, mtu)
-		for _, cp := range hh.packetStore {
-			cp.callback(cp.messageType, cp.messageSubType, hostinfo, cp.packet, nb, out)
-		}
-		f.cachedPacketMetrics.sent.Inc(int64(len(hh.packetStore)))
-	}
-
-	hostinfo.remotes.RefreshFromHandshake(vpnAddrs)
-	f.metricHandshakes.Update(duration)
-
-	return false
-}

handshake_manager_test.go

@@ -5,6 +5,7 @@ import (
 	"testing"
 	"time"
 
+	"github.com/gaissmai/bart"
 	"github.com/slackhq/nebula/cert"
 	"github.com/slackhq/nebula/header"
 	"github.com/slackhq/nebula/test"
@@ -27,7 +28,7 @@ func Test_NewHandshakeManagerVpnIp(t *testing.T) {
 		initiatingVersion: cert.Version1,
 		privateKey:        []byte{},
 		v1Cert:            &dummyCert{version: cert.Version1},
-		v1HandshakeBytes:  []byte{},
+		v1Credential:      nil,
 	}
 
 	blah := NewHandshakeManager(l, mainHM, lh, &udp.NoopConn{}, defaultHandshakeConfig)
@@ -100,3 +101,137 @@ func (mw *mockEncWriter) GetHostInfo(_ netip.Addr) *HostInfo {
 func (mw *mockEncWriter) GetCertState() *CertState {
 	return &CertState{initiatingVersion: cert.Version2}
 }
+
+func TestValidatePeerCert(t *testing.T) {
+	l := test.NewLogger()
+
+	myNetwork := netip.MustParsePrefix("10.0.0.1/24")
+	myAddrTable := new(bart.Lite)
+	myAddrTable.Insert(netip.PrefixFrom(myNetwork.Addr(), myNetwork.Addr().BitLen()))
+	myNetTable := new(bart.Lite)
+	myNetTable.Insert(myNetwork.Masked())
+
+	newHM := func() *HandshakeManager {
+		hm := NewHandshakeManager(l, newHostMap(l), newTestLighthouse(), &udp.NoopConn{}, defaultHandshakeConfig)
+		hm.f = &Interface{
+			handshakeManager:   hm,
+			pki:                &PKI{},
+			l:                  l,
+			myVpnAddrsTable:    myAddrTable,
+			myVpnNetworksTable: myNetTable,
+			lightHouse:         hm.lightHouse,
+		}
+		return hm
+	}
+
+	cached := func(networks ...netip.Prefix) *cert.CachedCertificate {
+		return &cert.CachedCertificate{
+			Certificate: &dummyCert{name: "peer", networks: networks},
+		}
+	}
+
+	via := ViaSender{
+		UdpAddr:   netip.MustParseAddrPort("198.51.100.7:4242"),
+		IsRelayed: true, // skip the remote allow list (covered separately)
+	}
+
+	t.Run("addr inside our networks sets anyVpnAddrsInCommon", func(t *testing.T) {
+		hm := newHM()
+		// 10.0.0.2 falls inside our 10.0.0.0/24
+		addrs, common, ok := hm.validatePeerCert(via, cached(netip.MustParsePrefix("10.0.0.2/24")))
+		assert.True(t, ok)
+		assert.True(t, common)
+		assert.Equal(t, []netip.Addr{netip.MustParseAddr("10.0.0.2")}, addrs)
+	})
+
+	t.Run("addr outside our networks leaves anyVpnAddrsInCommon false", func(t *testing.T) {
+		hm := newHM()
+		addrs, common, ok := hm.validatePeerCert(via, cached(netip.MustParsePrefix("192.168.1.5/24")))
+		assert.True(t, ok)
+		assert.False(t, common)
+		assert.Equal(t, []netip.Addr{netip.MustParseAddr("192.168.1.5")}, addrs)
+	})
+
+	t.Run("any matching network is enough", func(t *testing.T) {
+		hm := newHM()
+		addrs, common, ok := hm.validatePeerCert(via, cached(
+			netip.MustParsePrefix("192.168.1.5/24"),
+			netip.MustParsePrefix("10.0.0.42/24"),
+		))
+		assert.True(t, ok)
+		assert.True(t, common)
+		assert.Len(t, addrs, 2)
+	})
+
+	t.Run("self-handshake is rejected", func(t *testing.T) {
+		hm := newHM()
+		// 10.0.0.1 is in myVpnAddrsTable
+		addrs, common, ok := hm.validatePeerCert(via, cached(netip.MustParsePrefix("10.0.0.1/24")))
+		assert.False(t, ok)
+		assert.False(t, common)
+		assert.Nil(t, addrs)
+	})
+
+	t.Run("cert with no networks is rejected", func(t *testing.T) {
+		hm := newHM()
+		addrs, common, ok := hm.validatePeerCert(via, cached())
+		assert.False(t, ok)
+		assert.False(t, common)
+		assert.Nil(t, addrs)
+	})
+}
+
+func TestHandleIncomingDispatch(t *testing.T) {
+	l := test.NewLogger()
+
+	newHM := func() *HandshakeManager {
+		hm := NewHandshakeManager(l, newHostMap(l), newTestLighthouse(), &udp.NoopConn{}, defaultHandshakeConfig)
+		hm.f = &Interface{
+			handshakeManager: hm,
+			pki:              &PKI{},
+			l:                l,
+		}
+		return hm
+	}
+
+	via := ViaSender{
+		UdpAddr:   netip.MustParseAddrPort("198.51.100.7:4242"),
+		IsRelayed: true, // bypass remote allow list
+	}
+
+	// A packet body of zero length is fine for these tests: dispatch is
+	// gated on header fields, and we assert that we never reach noise/cert
+	// processing for any of the malformed shapes here.
+	pkt := make([]byte, header.Len)
+
+	t.Run("unsupported subtype dropped", func(t *testing.T) {
+		hm := newHM()
+		h := &header.H{Type: header.Handshake, Subtype: header.MessageSubType(99), MessageCounter: 1}
+		hm.HandleIncoming(via, pkt, h)
+		assert.Empty(t, hm.indexes, "no pending handshake should be created")
+	})
+
+	t.Run("stage-1 with non-zero RemoteIndex dropped", func(t *testing.T) {
+		hm := newHM()
+		h := &header.H{
+			Type:           header.Handshake,
+			Subtype:        header.HandshakeIXPSK0,
+			RemoteIndex:    0xdeadbeef,
+			MessageCounter: 1,
+		}
+		hm.HandleIncoming(via, pkt, h)
+		assert.Empty(t, hm.indexes, "spoofed stage-1 must not create a pending machine")
+	})
+
+	t.Run("continuation with no matching pending index dropped", func(t *testing.T) {
+		hm := newHM()
+		h := &header.H{
+			Type:           header.Handshake,
+			Subtype:        header.HandshakeIXPSK0,
+			RemoteIndex:    0xcafef00d,
+			MessageCounter: 2,
+		}
+		hm.HandleIncoming(via, pkt, h)
+		assert.Empty(t, hm.indexes, "orphan stage-2 must not create state")
+	})
+}

header/header.go

@@ -57,6 +57,8 @@ const (
 const (
 	TestRequest      MessageSubType = 0
 	TestReply        MessageSubType = 1
+	MTUDProbeRequest MessageSubType = 2
+	MTUDProbeReply   MessageSubType = 3
 )
 
 const (
@@ -69,6 +71,8 @@ var ErrHeaderTooShort = errors.New("header is too short")
 var subTypeTestMap = map[MessageSubType]string{
 	TestRequest:      "testRequest",
 	TestReply:        "testReply",
+	MTUDProbeRequest: "mtudProbeRequest",
+	MTUDProbeReply:   "mtudProbeReply",
 }
 
 var subTypeNoneMap = map[MessageSubType]string{0: "none"}

hostmap.go

@@ -1,7 +1,11 @@
 package nebula
 
 import (
+	"context"
+	"encoding/json"
 	"errors"
+	"fmt"
+	"log/slog"
 	"net"
 	"net/netip"
 	"slices"
@@ -11,10 +15,10 @@ import (
 
 	"github.com/gaissmai/bart"
 	"github.com/rcrowley/go-metrics"
-	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula/cert"
 	"github.com/slackhq/nebula/config"
 	"github.com/slackhq/nebula/header"
+	"github.com/slackhq/nebula/logging"
 )
 
 const defaultPromoteEvery = 1000       // Count of packets sent before we try moving a tunnel to a preferred underlay ip address
@@ -58,7 +62,7 @@ type HostMap struct {
 	RemoteIndexes   map[uint32]*HostInfo
 	Hosts           map[netip.Addr]*HostInfo
 	preferredRanges atomic.Pointer[[]netip.Prefix]
-	l               *logrus.Logger
+	l               *slog.Logger
 }
 
 // For synchronization, treat the pointed-to Relay struct as immutable. To edit the Relay
@@ -212,6 +216,18 @@ func (rs *RelayState) InsertRelay(ip netip.Addr, idx uint32, r *Relay) {
 	rs.relayForByIdx[idx] = r
 }
 
+type NetworkType uint8
+
+const (
+	NetworkTypeUnknown NetworkType = iota
+	// NetworkTypeVPN is a network that overlaps one or more of the vpnNetworks in our certificate
+	NetworkTypeVPN
+	// NetworkTypeVPNPeer is a network that does not overlap one of our networks
+	NetworkTypeVPNPeer
+	// NetworkTypeUnsafe is a network from Certificate.UnsafeNetworks()
+	NetworkTypeUnsafe
+)
+
 type HostInfo struct {
 	remote          netip.AddrPort
 	remotes         *RemoteList
@@ -225,8 +241,8 @@ type HostInfo struct {
 	// vpn networks but were removed because they are not usable
 	vpnAddrs []netip.Addr
 
-	// networks are both all vpn and unsafe networks assigned to this host
+	// networks is a combination of specific vpn addresses (not prefixes!) and full unsafe networks assigned to this host.
-	networks   *bart.Lite
+	networks   *bart.Table[NetworkType]
 	relayState RelayState
 
 	// HandshakePacket records the packets used to create this hostinfo
@@ -264,9 +280,25 @@ type HostInfo struct {
 }
 
 type ViaSender struct {
+	UdpAddr   netip.AddrPort
 	relayHI   *HostInfo // relayHI is the host info object of the relay
 	remoteIdx uint32    // remoteIdx is the index included in the header of the received packet
 	relay     *Relay    // relay contains the rest of the relay information, including the PeerIP of the host trying to communicate with us.
+	IsRelayed bool      // IsRelayed is true if the packet was sent through a relay
+}
+
+func (v ViaSender) String() string {
+	if v.IsRelayed {
+		return fmt.Sprintf("%s (relayed)", v.UdpAddr)
+	}
+	return v.UdpAddr.String()
+}
+
+func (v ViaSender) MarshalJSON() ([]byte, error) {
+	if v.IsRelayed {
+		return json.Marshal(m{"relay": v.UdpAddr})
+	}
+	return json.Marshal(m{"direct": v.UdpAddr})
 }
 
 type cachedPacket struct {
@@ -283,7 +315,7 @@ type cachedPacketMetrics struct {
 	dropped metrics.Counter
 }
 
-func NewHostMapFromConfig(l *logrus.Logger, c *config.C) *HostMap {
+func NewHostMapFromConfig(l *slog.Logger, c *config.C) *HostMap {
 	hm := newHostMap(l)
 
 	hm.reload(c, true)
@@ -291,13 +323,12 @@ func NewHostMapFromConfig(l *logrus.Logger, c *config.C) *HostMap {
 		hm.reload(c, false)
 	})
 
-	l.WithField("preferredRanges", hm.GetPreferredRanges()).
+	l.Info("Main HostMap created", "preferredRanges", hm.GetPreferredRanges())
-		Info("Main HostMap created")
 
 	return hm
 }
 
-func newHostMap(l *logrus.Logger) *HostMap {
+func newHostMap(l *slog.Logger) *HostMap {
 	return &HostMap{
 		Indexes:       map[uint32]*HostInfo{},
 		Relays:        map[uint32]*HostInfo{},
@@ -316,7 +347,10 @@ func (hm *HostMap) reload(c *config.C, initial bool) {
 			preferredRange, err := netip.ParsePrefix(rawPreferredRange)
 
 			if err != nil {
-				hm.l.WithError(err).WithField("range", rawPreferredRanges).Warn("Failed to parse preferred ranges, ignoring")
+				hm.l.Warn("Failed to parse preferred ranges, ignoring",
+					"error", err,
+					"range", rawPreferredRanges,
+				)
 				continue
 			}
 
@@ -325,7 +359,10 @@ func (hm *HostMap) reload(c *config.C, initial bool) {
 
 		oldRanges := hm.preferredRanges.Swap(&preferredRanges)
 		if !initial {
-			hm.l.WithField("oldPreferredRanges", *oldRanges).WithField("newPreferredRanges", preferredRanges).Info("preferred_ranges changed")
+			hm.l.Info("preferred_ranges changed",
+				"oldPreferredRanges", *oldRanges,
+				"newPreferredRanges", preferredRanges,
+			)
 		}
 	}
 }
@@ -458,10 +495,11 @@ func (hm *HostMap) unlockedInnerDeleteHostInfo(hostinfo *HostInfo, addr netip.Ad
 		hm.Indexes = map[uint32]*HostInfo{}
 	}
 
-	if hm.l.Level >= logrus.DebugLevel {
+	if hm.l.Enabled(context.Background(), slog.LevelDebug) {
-		hm.l.WithField("hostMap", m{"mapTotalSize": len(hm.Hosts),
+		hm.l.Debug("Hostmap hostInfo deleted",
-			"vpnAddrs": hostinfo.vpnAddrs, "indexNumber": hostinfo.localIndexId, "remoteIndexNumber": hostinfo.remoteIndexId}).
+			"hostMap", m{"mapTotalSize": len(hm.Hosts),
-			Debug("Hostmap hostInfo deleted")
+				"vpnAddrs": hostinfo.vpnAddrs, "indexNumber": hostinfo.localIndexId, "remoteIndexNumber": hostinfo.remoteIndexId},
+		)
 	}
 
 	if isLastHostinfo {
@@ -574,9 +612,9 @@ func (hm *HostMap) queryVpnAddr(vpnIp netip.Addr, promoteIfce *Interface) *HostI
 // unlockedAddHostInfo assumes you have a write-lock and will add a hostinfo object to the hostmap Indexes and RemoteIndexes maps.
 // If an entry exists for the Hosts table (vpnIp -> hostinfo) then the provided hostinfo will be made primary
 func (hm *HostMap) unlockedAddHostInfo(hostinfo *HostInfo, f *Interface) {
-	if f.serveDns {
+	if f.dnsServer != nil {
 		remoteCert := hostinfo.ConnectionState.peerCert
-		dnsR.Add(remoteCert.Certificate.Name()+".", hostinfo.vpnAddrs)
+		f.dnsServer.Add(remoteCert.Certificate.Name()+".", hostinfo.vpnAddrs)
 	}
 	for _, addr := range hostinfo.vpnAddrs {
 		hm.unlockedInnerAddHostInfo(addr, hostinfo, f)
@@ -585,10 +623,11 @@ func (hm *HostMap) unlockedAddHostInfo(hostinfo *HostInfo, f *Interface) {
 	hm.Indexes[hostinfo.localIndexId] = hostinfo
 	hm.RemoteIndexes[hostinfo.remoteIndexId] = hostinfo
 
-	if hm.l.Level >= logrus.DebugLevel {
+	if hm.l.Enabled(context.Background(), slog.LevelDebug) {
-		hm.l.WithField("hostMap", m{"vpnAddrs": hostinfo.vpnAddrs, "mapTotalSize": len(hm.Hosts),
+		hm.l.Debug("Hostmap vpnIp added",
-			"hostinfo": m{"existing": true, "localIndexId": hostinfo.localIndexId, "vpnAddrs": hostinfo.vpnAddrs}}).
+			"hostMap", m{"vpnAddrs": hostinfo.vpnAddrs, "mapTotalSize": len(hm.Hosts),
-			Debug("Hostmap vpnIp added")
+				"hostinfo": m{"existing": true, "localIndexId": hostinfo.localIndexId, "vpnAddrs": hostinfo.vpnAddrs}},
+		)
 	}
 }
 
@@ -682,6 +721,7 @@ func (i *HostInfo) GetCert() *cert.CachedCertificate {
 	return nil
 }
 
+// TODO: Maybe use ViaSender here?
 func (i *HostInfo) SetRemote(remote netip.AddrPort) {
 	// We copy here because we likely got this remote from a source that reuses the object
 	if i.remote != remote {
@@ -692,14 +732,14 @@ func (i *HostInfo) SetRemote(remote netip.AddrPort) {
 
 // SetRemoteIfPreferred returns true if the remote was changed. The lastRoam
 // time on the HostInfo will also be updated.
-func (i *HostInfo) SetRemoteIfPreferred(hm *HostMap, newRemote netip.AddrPort) bool {
+func (i *HostInfo) SetRemoteIfPreferred(hm *HostMap, via ViaSender) bool {
-	if !newRemote.IsValid() {
+	if via.IsRelayed {
-		// relays have nil udp Addrs
 		return false
 	}
+
 	currentRemote := i.remote
 	if !currentRemote.IsValid() {
-		i.SetRemote(newRemote)
+		i.SetRemote(via.UdpAddr)
 		return true
 	}
 
@@ -712,7 +752,7 @@ func (i *HostInfo) SetRemoteIfPreferred(hm *HostMap, newRemote netip.AddrPort) b
 			return false
 		}
 
-		if l.Contains(newRemote.Addr()) {
+		if l.Contains(via.UdpAddr.Addr()) {
 			newIsPreferred = true
 		}
 	}
@@ -722,7 +762,7 @@ func (i *HostInfo) SetRemoteIfPreferred(hm *HostMap, newRemote netip.AddrPort) b
 		i.lastRoam = time.Now()
 		i.lastRoamRemote = currentRemote
 
-		i.SetRemote(newRemote)
+		i.SetRemote(via.UdpAddr)
 
 		return true
 	}
@@ -730,35 +770,44 @@ func (i *HostInfo) SetRemoteIfPreferred(hm *HostMap, newRemote netip.AddrPort) b
 	return false
 }
 
-func (i *HostInfo) buildNetworks(networks, unsafeNetworks []netip.Prefix) {
+// buildNetworks fills in the networks field of HostInfo. It accepts a cert.Certificate so you never ever mix the network types up.
-	if len(networks) == 1 && len(unsafeNetworks) == 0 {
+func (i *HostInfo) buildNetworks(myVpnNetworksTable *bart.Lite, c cert.Certificate) {
-		// Simple case, no CIDRTree needed
+	if len(c.Networks()) == 1 && len(c.UnsafeNetworks()) == 0 {
-		return
+		if myVpnNetworksTable.Contains(c.Networks()[0].Addr()) {
+			return // Simple case, no BART needed
+		}
 	}
 
-	i.networks = new(bart.Lite)
+	i.networks = new(bart.Table[NetworkType])
-	for _, network := range networks {
+	for _, network := range c.Networks() {
 		nprefix := netip.PrefixFrom(network.Addr(), network.Addr().BitLen())
-		i.networks.Insert(nprefix)
+		if myVpnNetworksTable.Contains(network.Addr()) {
-	}
+			i.networks.Insert(nprefix, NetworkTypeVPN)
-
+		} else {
-	for _, network := range unsafeNetworks {
+			i.networks.Insert(nprefix, NetworkTypeVPNPeer)
-		i.networks.Insert(network)
 		}
 	}
 
-func (i *HostInfo) logger(l *logrus.Logger) *logrus.Entry {
+	for _, network := range c.UnsafeNetworks() {
+		i.networks.Insert(network, NetworkTypeUnsafe)
+	}
+}
+
+// logger returns a derived slog.Logger with per-hostinfo fields pre-bound.
+func (i *HostInfo) logger(l *slog.Logger) *slog.Logger {
 	if i == nil {
-		return logrus.NewEntry(l)
+		return l
 	}
 
-	li := l.WithField("vpnAddrs", i.vpnAddrs).
+	li := l.With(
-		WithField("localIndex", i.localIndexId).
+		"vpnAddrs", i.vpnAddrs,
-		WithField("remoteIndex", i.remoteIndexId)
+		"localIndex", i.localIndexId,
+		"remoteIndex", i.remoteIndexId,
+	)
 
 	if connState := i.ConnectionState; connState != nil {
 		if peerCert := connState.peerCert; peerCert != nil {
-			li = li.WithField("certName", peerCert.Certificate.Name())
+			li = li.With("certName", peerCert.Certificate.Name())
 		}
 	}
 
@@ -767,14 +816,17 @@ func (i *HostInfo) logger(l *logrus.Logger) *logrus.Entry {
 
 // Utility functions
 
-func localAddrs(l *logrus.Logger, allowList *LocalAllowList) []netip.Addr {
+func localAddrs(l *slog.Logger, allowList *LocalAllowList) []netip.Addr {
 	//FIXME: This function is pretty garbage
 	var finalAddrs []netip.Addr
 	ifaces, _ := net.Interfaces()
 	for _, i := range ifaces {
 		allow := allowList.AllowName(i.Name)
-		if l.Level >= logrus.TraceLevel {
+		if l.Enabled(context.Background(), logging.LevelTrace) {
-			l.WithField("interfaceName", i.Name).WithField("allow", allow).Trace("localAllowList.AllowName")
+			l.Log(context.Background(), logging.LevelTrace, "localAllowList.AllowName",
+				"interfaceName", i.Name,
+				"allow", allow,
+			)
 		}
 
 		if !allow {
@@ -792,8 +844,8 @@ func localAddrs(l *logrus.Logger, allowList *LocalAllowList) []netip.Addr {
 			}
 
 			if !addr.IsValid() {
-				if l.Level >= logrus.DebugLevel {
+				if l.Enabled(context.Background(), slog.LevelDebug) {
-					l.WithField("localAddr", rawAddr).Debug("addr was invalid")
+					l.Debug("addr was invalid", "localAddr", rawAddr)
 				}
 				continue
 			}
@@ -801,8 +853,11 @@ func localAddrs(l *logrus.Logger, allowList *LocalAllowList) []netip.Addr {
 
 			if addr.IsLoopback() == false && addr.IsLinkLocalUnicast() == false {
 				isAllowed := allowList.Allow(addr)
-				if l.Level >= logrus.TraceLevel {
+				if l.Enabled(context.Background(), logging.LevelTrace) {
-					l.WithField("localAddr", addr).WithField("allowed", isAllowed).Trace("localAllowList.Allow")
+					l.Log(context.Background(), logging.LevelTrace, "localAllowList.Allow",
+						"localAddr", addr,
+						"allowed", isAllowed,
+					)
 				}
 				if !isAllowed {
 					continue

hostmap_test.go

@@ -196,7 +196,7 @@ func TestHostMap_DeleteHostInfo(t *testing.T) {
 
 func TestHostMap_reload(t *testing.T) {
 	l := test.NewLogger()
-	c := config.NewC(l)
+	c := config.NewC(test.NewLogger())
 
 	hm := NewHostMapFromConfig(l, c)
 

hostmap_tester.go

@@ -1,5 +1,4 @@
 //go:build e2e_testing
-// +build e2e_testing
 
 package nebula
 

inside.go

@@ -1,23 +1,25 @@
 package nebula
 
 import (
+	"context"
+	"log/slog"
 	"net/netip"
-	"unsafe"
 
-	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula/firewall"
 	"github.com/slackhq/nebula/header"
 	"github.com/slackhq/nebula/iputil"
 	"github.com/slackhq/nebula/noiseutil"
-	"github.com/slackhq/nebula/overlay"
 	"github.com/slackhq/nebula/routing"
 )
 
-func (f *Interface) consumeInsidePacket(packet []byte, fwPacket *firewall.Packet, nb, out []byte, q int, localCache *firewall.ConntrackCache) {
+func (f *Interface) consumeInsidePacket(packet []byte, fwPacket *firewall.Packet, nb, out []byte, q int, localCache firewall.ConntrackCache) {
 	err := newPacket(packet, false, fwPacket)
 	if err != nil {
-		if f.l.Level >= logrus.DebugLevel {
+		if f.l.Enabled(context.Background(), slog.LevelDebug) {
-			f.l.WithField("packet", packet).Debugf("Error while validating outbound packet: %s", err)
+			f.l.Debug("Error while validating outbound packet",
+				"packet", packet,
+				"error", err,
+			)
 		}
 		return
 	}
@@ -37,7 +39,7 @@ func (f *Interface) consumeInsidePacket(packet []byte, fwPacket *firewall.Packet
 		if immediatelyForwardToSelf {
 			_, err := f.readers[q].Write(packet)
 			if err != nil {
-				f.l.WithError(err).Error("Failed to forward to tun")
+				f.l.Error("Failed to forward to tun", "error", err)
 			}
 		}
 		// Otherwise, drop. On linux, we should never see these packets - Linux
@@ -56,10 +58,11 @@ func (f *Interface) consumeInsidePacket(packet []byte, fwPacket *firewall.Packet
 
 	if hostinfo == nil {
 		f.rejectInside(packet, out, q)
-		if f.l.Level >= logrus.DebugLevel {
+		if f.l.Enabled(context.Background(), slog.LevelDebug) {
-			f.l.WithField("vpnAddr", fwPacket.RemoteAddr).
+			f.l.Debug("dropping outbound packet, vpnAddr not in our vpn networks or in unsafe networks",
-				WithField("fwPacket", fwPacket).
+				"vpnAddr", fwPacket.RemoteAddr,
-				Debugln("dropping outbound packet, vpnAddr not in our vpn networks or in unsafe networks")
+				"fwPacket", fwPacket,
+			)
 		}
 		return
 	}
@@ -74,11 +77,11 @@ func (f *Interface) consumeInsidePacket(packet []byte, fwPacket *firewall.Packet
 
 	} else {
 		f.rejectInside(packet, out, q)
-		if f.l.Level >= logrus.DebugLevel {
+		if f.l.Enabled(context.Background(), slog.LevelDebug) {
-			hostinfo.logger(f.l).
+			hostinfo.logger(f.l).Debug("dropping outbound packet",
-				WithField("fwPacket", fwPacket).
+				"fwPacket", fwPacket,
-				WithField("reason", dropReason).
+				"reason", dropReason,
-				Debugln("dropping outbound packet")
+			)
 		}
 	}
 }
@@ -95,7 +98,7 @@ func (f *Interface) rejectInside(packet []byte, out []byte, q int) {
 
 	_, err := f.readers[q].Write(out)
 	if err != nil {
-		f.l.WithError(err).Error("Failed to write to tun")
+		f.l.Error("Failed to write to tun", "error", err)
 	}
 }
 
@@ -110,11 +113,11 @@ func (f *Interface) rejectOutside(packet []byte, ci *ConnectionState, hostinfo *
 	}
 
 	if len(out) > iputil.MaxRejectPacketSize {
-		if f.l.GetLevel() >= logrus.InfoLevel {
+		if f.l.Enabled(context.Background(), slog.LevelInfo) {
-			f.l.
+			f.l.Info("rejectOutside: packet too big, not sending",
-				WithField("packet", packet).
+				"packet", packet,
-				WithField("outPacket", out).
+				"outPacket", out,
-				Info("rejectOutside: packet too big, not sending")
+			)
 		}
 		return
 	}
@@ -122,9 +125,10 @@ func (f *Interface) rejectOutside(packet []byte, ci *ConnectionState, hostinfo *
 	f.sendNoMetrics(header.Message, 0, ci, hostinfo, netip.AddrPort{}, out, nb, packet, q)
 }
 
-// Handshake will attempt to initiate a tunnel with the provided vpn address if it is within our vpn networks. This is a no-op if the tunnel is already established or being established
+// Handshake will attempt to initiate a tunnel with the provided vpn address. This is a no-op if the tunnel is already established or being established
+// it does not check if it is within our vpn networks!
 func (f *Interface) Handshake(vpnAddr netip.Addr) {
-	f.getOrHandshakeNoRouting(vpnAddr, nil)
+	f.handshakeManager.GetOrHandshake(vpnAddr, nil)
 }
 
 // getOrHandshakeNoRouting returns nil if the vpnAddr is not routable.
@@ -140,7 +144,6 @@ func (f *Interface) getOrHandshakeNoRouting(vpnAddr netip.Addr, cacheCallback fu
 // getOrHandshakeConsiderRouting will try to find the HostInfo to handle this packet, starting a handshake if necessary.
 // If the 2nd return var is false then the hostinfo is not ready to be used in a tunnel.
 func (f *Interface) getOrHandshakeConsiderRouting(fwPacket *firewall.Packet, cacheCallback func(*HandshakeHostInfo)) (*HostInfo, bool) {
-
 	destinationAddr := fwPacket.RemoteAddr
 
 	hostinfo, ready := f.getOrHandshakeNoRouting(destinationAddr, cacheCallback)
@@ -186,10 +189,11 @@ func (f *Interface) getOrHandshakeConsiderRouting(fwPacket *firewall.Packet, cac
 		// This would also need to interact with unsafe_route updates through reloading the config or
 		// use of the use_system_route_table option
 
-		if f.l.Level >= logrus.DebugLevel {
+		if f.l.Enabled(context.Background(), slog.LevelDebug) {
-			f.l.WithField("destination", destinationAddr).
+			f.l.Debug("Calculated gateway for ECMP not available, attempting other gateways",
-				WithField("originalGateway", gatewayAddr).
+				"destination", destinationAddr,
-				Debugln("Calculated gateway for ECMP not available, attempting other gateways")
+				"originalGateway", gatewayAddr,
+			)
 		}
 
 		for i := range gateways {
@@ -215,17 +219,18 @@ func (f *Interface) sendMessageNow(t header.MessageType, st header.MessageSubTyp
 	fp := &firewall.Packet{}
 	err := newPacket(p, false, fp)
 	if err != nil {
-		f.l.Warnf("error while parsing outgoing packet for firewall check; %v", err)
+		f.l.Warn("error while parsing outgoing packet for firewall check", "error", err)
 		return
 	}
 
 	// check if packet is in outbound fw rules
 	dropReason := f.firewall.Drop(*fp, false, hostinfo, f.pki.GetCAPool(), nil)
 	if dropReason != nil {
-		if f.l.Level >= logrus.DebugLevel {
+		if f.l.Enabled(context.Background(), slog.LevelDebug) {
-			f.l.WithField("fwPacket", fp).
+			f.l.Debug("dropping cached packet",
-				WithField("reason", dropReason).
+				"fwPacket", fp,
-				Debugln("dropping cached packet")
+				"reason", dropReason,
+			)
 		}
 		return
 	}
@@ -233,16 +238,18 @@ func (f *Interface) sendMessageNow(t header.MessageType, st header.MessageSubTyp
 	f.sendNoMetrics(header.Message, st, hostinfo.ConnectionState, hostinfo, netip.AddrPort{}, p, nb, out, 0)
 }
 
-// SendMessageToVpnAddr handles real addr:port lookup and sends to the current best known address for vpnAddr
+// SendMessageToVpnAddr handles real addr:port lookup and sends to the current best known address for vpnAddr.
+// This function ignores myVpnNetworksTable, and will always attempt to treat the address as a vpnAddr
 func (f *Interface) SendMessageToVpnAddr(t header.MessageType, st header.MessageSubType, vpnAddr netip.Addr, p, nb, out []byte) {
-	hostInfo, ready := f.getOrHandshakeNoRouting(vpnAddr, func(hh *HandshakeHostInfo) {
+	hostInfo, ready := f.handshakeManager.GetOrHandshake(vpnAddr, func(hh *HandshakeHostInfo) {
 		hh.cachePacket(f.l, t, st, p, f.SendMessageToHostInfo, f.cachedPacketMetrics)
 	})
 
 	if hostInfo == nil {
-		if f.l.Level >= logrus.DebugLevel {
+		if f.l.Enabled(context.Background(), slog.LevelDebug) {
-			f.l.WithField("vpnAddr", vpnAddr).
+			f.l.Debug("dropping SendMessageToVpnAddr, vpnAddr not in our vpn networks or in unsafe routes",
-				Debugln("dropping SendMessageToVpnAddr, vpnAddr not in our vpn networks or in unsafe routes")
+				"vpnAddr", vpnAddr,
+			)
 		}
 		return
 	}
@@ -298,12 +305,12 @@ func (f *Interface) SendVia(via *HostInfo,
 		if noiseutil.EncryptLockNeeded {
 			via.ConnectionState.writeLock.Unlock()
 		}
-		via.logger(f.l).
+		via.logger(f.l).Error("SendVia out buffer not large enough for relay",
-			WithField("outCap", cap(out)).
+			"outCap", cap(out),
-			WithField("payloadLen", len(ad)).
+			"payloadLen", len(ad),
-			WithField("headerLen", len(out)).
+			"headerLen", len(out),
-			WithField("cipherOverhead", via.ConnectionState.eKey.Overhead()).
+			"cipherOverhead", via.ConnectionState.eKey.Overhead(),
-			Error("SendVia out buffer not large enough for relay")
+		)
 		return
 	}
 
@@ -323,12 +330,12 @@ func (f *Interface) SendVia(via *HostInfo,
 		via.ConnectionState.writeLock.Unlock()
 	}
 	if err != nil {
-		via.logger(f.l).WithError(err).Info("Failed to EncryptDanger in sendVia")
+		via.logger(f.l).Info("Failed to EncryptDanger in sendVia", "error", err)
 		return
 	}
 	err = f.writers[0].WriteTo(out, via.remote)
 	if err != nil {
-		via.logger(f.l).WithError(err).Info("Failed to WriteTo in sendVia")
+		via.logger(f.l).Info("Failed to WriteTo in sendVia", "error", err)
 	}
 	f.connectionManager.RelayUsed(relay.LocalIndex)
 }
@@ -337,21 +344,9 @@ func (f *Interface) sendNoMetrics(t header.MessageType, st header.MessageSubType
 	if ci.eKey == nil {
 		return
 	}
-	target := remote
+	useRelay := !remote.IsValid() && !hostinfo.remote.IsValid()
-	if !target.IsValid() {
-		target = hostinfo.remote
-	}
-	useRelay := !target.IsValid()
 	fullOut := out
 
-	var pkt *overlay.Packet
-	if !useRelay && f.batches.Enabled() {
-		pkt = f.batches.newPacket()
-		if pkt != nil {
-			out = pkt.Payload()[:0]
-		}
-	}
-
 	if useRelay {
 		if len(out) < header.Len {
 			// out always has a capacity of mtu, but not always a length greater than the header.Len.
@@ -379,91 +374,58 @@ func (f *Interface) sendNoMetrics(t header.MessageType, st header.MessageSubType
 		// finally used again. This tunnel would eventually be torn down and recreated if this action didn't help.
 		f.lightHouse.QueryServer(hostinfo.vpnAddrs[0])
 		hostinfo.lastRebindCount = f.rebindCount
-		if f.l.Level >= logrus.DebugLevel {
+		if f.l.Enabled(context.Background(), slog.LevelDebug) {
-			f.l.WithField("vpnAddrs", hostinfo.vpnAddrs).Debug("Lighthouse update triggered for punch due to rebind counter")
+			f.l.Debug("Lighthouse update triggered for punch due to rebind counter",
+				"vpnAddrs", hostinfo.vpnAddrs,
+			)
 		}
 	}
 
 	var err error
-	if len(p) > 0 && slicesOverlap(out, p) {
-		tmp := make([]byte, len(p))
-		copy(tmp, p)
-		p = tmp
-	}
 	out, err = ci.eKey.EncryptDanger(out, out, p, c, nb)
 	if noiseutil.EncryptLockNeeded {
 		ci.writeLock.Unlock()
 	}
 	if err != nil {
-		if pkt != nil {
+		hostinfo.logger(f.l).Error("Failed to encrypt outgoing packet",
-			pkt.Release()
+			"error", err,
-		}
+			"udpAddr", remote,
-		hostinfo.logger(f.l).WithError(err).
+			"counter", c,
-			WithField("udpAddr", target).WithField("counter", c).
+			"attemptedCounter", c,
-			WithField("attemptedCounter", c).
+		)
-			Error("Failed to encrypt outgoing packet")
 		return
 	}
 
-	if target.IsValid() {
+	if remote.IsValid() {
-		if pkt != nil {
+		err = f.writers[q].WriteTo(out, remote)
-			pkt.Len = len(out)
+		if err != nil {
-			if f.l.Level >= logrus.DebugLevel {
+			hostinfo.logger(f.l).Error("Failed to write outgoing packet",
-				f.l.WithFields(logrus.Fields{
+				"error", err,
-					"queue":        q,
+				"udpAddr", remote,
-					"dest":         target,
+			)
-					"payload_len":  pkt.Len,
-					"use_batches":  true,
-					"remote_index": hostinfo.remoteIndexId,
-				}).Debug("enqueueing packet to UDP batch queue")
 		}
-			if f.tryQueuePacket(q, pkt, target) {
+	} else if hostinfo.remote.IsValid() {
-				return
+		err = f.writers[q].WriteTo(out, hostinfo.remote)
+		if err != nil {
+			hostinfo.logger(f.l).Error("Failed to write outgoing packet",
+				"error", err,
+				"udpAddr", remote,
+			)
 		}
-			if f.l.Level >= logrus.DebugLevel {
+	} else {
-				f.l.WithFields(logrus.Fields{
-					"queue": q,
-					"dest":  target,
-				}).Debug("failed to enqueue packet; falling back to immediate send")
-			}
-			f.writeImmediatePacket(q, pkt, target, hostinfo)
-			return
-		}
-		if f.tryQueueDatagram(q, out, target) {
-			return
-		}
-		f.writeImmediate(q, out, target, hostinfo)
-		return
-	}
-
-	// fall back to relay path
-	if pkt != nil {
-		pkt.Release()
-	}
-
 		// Try to send via a relay
 		for _, relayIP := range hostinfo.relayState.CopyRelayIps() {
 			relayHostInfo, relay, err := f.hostMap.QueryVpnAddrsRelayFor(hostinfo.vpnAddrs, relayIP)
 			if err != nil {
 				hostinfo.relayState.DeleteRelay(relayIP)
-			hostinfo.logger(f.l).WithField("relay", relayIP).WithError(err).Info("sendNoMetrics failed to find HostInfo")
+				hostinfo.logger(f.l).Info("sendNoMetrics failed to find HostInfo",
+					"relay", relayIP,
+					"error", err,
+				)
 				continue
 			}
 			f.SendVia(relayHostInfo, relay, out, nb, fullOut[:header.Len+len(out)], true)
 			break
 		}
 	}
-
-// slicesOverlap reports whether the two byte slices share any portion of memory.
-// cipher.AEAD.Seal requires plaintext and dst to live in disjoint regions.
-func slicesOverlap(a, b []byte) bool {
-	if len(a) == 0 || len(b) == 0 {
-		return false
-	}
-
-	aStart := uintptr(unsafe.Pointer(&a[0]))
-	aEnd := aStart + uintptr(len(a))
-	bStart := uintptr(unsafe.Pointer(&b[0]))
-	bEnd := bStart + uintptr(len(b))
-	return aStart < bEnd && bStart < aEnd
 }

inside_bsd.go

@@ -1,5 +1,4 @@
 //go:build darwin || dragonfly || freebsd || netbsd || openbsd
-// +build darwin dragonfly freebsd netbsd openbsd
 
 package nebula
 

inside_generic.go

@@ -1,5 +1,4 @@
 //go:build !darwin && !dragonfly && !freebsd && !netbsd && !openbsd
-// +build !darwin,!dragonfly,!freebsd,!netbsd,!openbsd
 
 package nebula
 

lighthouse.go

@@ -5,6 +5,7 @@ import (
 	"encoding/binary"
 	"errors"
 	"fmt"
+	"log/slog"
 	"net"
 	"net/netip"
 	"slices"
@@ -15,10 +16,10 @@ import (
 
 	"github.com/gaissmai/bart"
 	"github.com/rcrowley/go-metrics"
-	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula/cert"
 	"github.com/slackhq/nebula/config"
 	"github.com/slackhq/nebula/header"
+	"github.com/slackhq/nebula/logging"
 	"github.com/slackhq/nebula/udp"
 	"github.com/slackhq/nebula/util"
 )
@@ -69,18 +70,19 @@ type LightHouse struct {
 	// Addr's of relays that can be used by peers to access me
 	relaysForMe atomic.Pointer[[]netip.Addr]
 
+	updateTrigger chan struct{}
 	queryChan     chan netip.Addr
 
 	calculatedRemotes atomic.Pointer[bart.Table[[]*calculatedRemote]] // Maps VpnAddr to []*calculatedRemote
 
 	metrics           *MessageMetrics
 	metricHolepunchTx metrics.Counter
-	l                 *logrus.Logger
+	l                 *slog.Logger
 }
 
 // NewLightHouseFromConfig will build a Lighthouse struct from the values provided in the config object
 // addrMap should be nil unless this is during a config reload
-func NewLightHouseFromConfig(ctx context.Context, l *logrus.Logger, c *config.C, cs *CertState, pc udp.Conn, p *Punchy) (*LightHouse, error) {
+func NewLightHouseFromConfig(ctx context.Context, l *slog.Logger, c *config.C, cs *CertState, pc udp.Conn, p *Punchy) (*LightHouse, error) {
 	amLighthouse := c.GetBool("lighthouse.am_lighthouse", false)
 	nebulaPort := uint32(c.GetInt("listen.port", 0))
 	if amLighthouse && nebulaPort == 0 {
@@ -105,6 +107,7 @@ func NewLightHouseFromConfig(ctx context.Context, l *logrus.Logger, c *config.C,
 		nebulaPort:         nebulaPort,
 		punchConn:          pc,
 		punchy:             p,
+		updateTrigger:      make(chan struct{}, 1),
 		queryChan:          make(chan netip.Addr, c.GetUint32("handshakes.query_buffer", 64)),
 		l:                  l,
 	}
@@ -131,7 +134,7 @@ func NewLightHouseFromConfig(ctx context.Context, l *logrus.Logger, c *config.C,
 		case *util.ContextualError:
 			v.Log(l)
 		case error:
-			l.WithError(err).Error("failed to reload lighthouse")
+			l.Error("failed to reload lighthouse", "error", err)
 		}
 	})
 
@@ -203,8 +206,10 @@ func (lh *LightHouse) reload(c *config.C, initial bool) error {
 			//TODO: we could technically insert all returned addrs instead of just the first one if a dns lookup was used
 			addr := addrs[0].Unmap()
 			if lh.myVpnNetworksTable.Contains(addr) {
-				lh.l.WithField("addr", rawAddr).WithField("entry", i+1).
+				lh.l.Warn("Ignoring lighthouse.advertise_addrs report because it is within the nebula network range",
-					Warn("Ignoring lighthouse.advertise_addrs report because it is within the nebula network range")
+					"addr", rawAddr,
+					"entry", i+1,
+				)
 				continue
 			}
 
@@ -222,7 +227,9 @@ func (lh *LightHouse) reload(c *config.C, initial bool) error {
 		lh.interval.Store(int64(c.GetInt("lighthouse.interval", 10)))
 
 		if !initial {
-			lh.l.Infof("lighthouse.interval changed to %v", lh.interval.Load())
+			lh.l.Info("lighthouse.interval changed",
+				"interval", lh.interval.Load(),
+			)
 
 			if lh.updateCancel != nil {
 				// May not always have a running routine
@@ -316,6 +323,7 @@ func (lh *LightHouse) reload(c *config.C, initial bool) error {
 		if !initial {
 			//NOTE: we are not tearing down existing lighthouse connections because they might be used for non lighthouse traffic
 			lh.l.Info("lighthouse.hosts has changed")
+			lh.TriggerUpdate()
 		}
 	}
 
@@ -333,9 +341,12 @@ func (lh *LightHouse) reload(c *config.C, initial bool) error {
 			for _, v := range c.GetStringSlice("relay.relays", nil) {
 				configRIP, err := netip.ParseAddr(v)
 				if err != nil {
-					lh.l.WithField("relay", v).WithError(err).Warn("Parse relay from config failed")
+					lh.l.Warn("Parse relay from config failed",
+						"relay", v,
+						"error", err,
+					)
 				} else {
-					lh.l.WithField("relay", v).Info("Read relay from config")
+					lh.l.Info("Read relay from config", "relay", v)
 					relaysForMe = append(relaysForMe, configRIP)
 				}
 			}
@@ -360,7 +371,10 @@ func (lh *LightHouse) parseLighthouses(c *config.C) ([]netip.Addr, error) {
 		}
 
 		if !lh.myVpnNetworksTable.Contains(addr) {
-			return nil, util.NewContextualError("lighthouse host is not in our networks, invalid", m{"vpnAddr": addr, "networks": lh.myVpnNetworks}, nil)
+			lh.l.Warn("lighthouse host is not within our networks, lighthouse functionality will work but layer 3 network traffic to the lighthouse will not",
+				"vpnAddr", addr,
+				"networks", lh.myVpnNetworks,
+			)
 		}
 		out[i] = addr
 	}
@@ -431,7 +445,11 @@ func (lh *LightHouse) loadStaticMap(c *config.C, staticList map[netip.Addr]struc
 		}
 
 		if !lh.myVpnNetworksTable.Contains(vpnAddr) {
-			return util.NewContextualError("static_host_map key is not in our network, invalid", m{"vpnAddr": vpnAddr, "networks": lh.myVpnNetworks, "entry": i + 1}, nil)
+			lh.l.Warn("static_host_map key is not within our networks, layer 3 network traffic to this host will not work",
+				"vpnAddr", vpnAddr,
+				"networks", lh.myVpnNetworks,
+				"entry", i+1,
+			)
 		}
 
 		vals, ok := v.([]any)
@@ -532,12 +550,13 @@ func (lh *LightHouse) DeleteVpnAddrs(allVpnAddrs []netip.Addr) {
 	lh.Lock()
 	rm, ok := lh.addrMap[allVpnAddrs[0]]
 	if ok {
+		debugEnabled := lh.l.Enabled(context.Background(), slog.LevelDebug)
 		for _, addr := range allVpnAddrs {
 			srm := lh.addrMap[addr]
 			if srm == rm {
 				delete(lh.addrMap, addr)
-				if lh.l.Level >= logrus.DebugLevel {
+				if debugEnabled {
-					lh.l.Debugf("deleting %s from lighthouse.", addr)
+					lh.l.Debug("deleting from lighthouse", "vpnAddr", addr)
 				}
 			}
 		}
@@ -654,9 +673,12 @@ func (lh *LightHouse) unlockedGetRemoteList(allAddrs []netip.Addr) *RemoteList {
 
 func (lh *LightHouse) shouldAdd(vpnAddrs []netip.Addr, to netip.Addr) bool {
 	allow := lh.GetRemoteAllowList().AllowAll(vpnAddrs, to)
-	if lh.l.Level >= logrus.TraceLevel {
+	if lh.l.Enabled(context.Background(), logging.LevelTrace) {
-		lh.l.WithField("vpnAddrs", vpnAddrs).WithField("udpAddr", to).WithField("allow", allow).
+		lh.l.Log(context.Background(), logging.LevelTrace, "remoteAllowList.Allow",
-			Trace("remoteAllowList.Allow")
+			"vpnAddrs", vpnAddrs,
+			"udpAddr", to,
+			"allow", allow,
+		)
 	}
 	if !allow {
 		return false
@@ -673,9 +695,12 @@ func (lh *LightHouse) shouldAdd(vpnAddrs []netip.Addr, to netip.Addr) bool {
 func (lh *LightHouse) unlockedShouldAddV4(vpnAddr netip.Addr, to *V4AddrPort) bool {
 	udpAddr := protoV4AddrPortToNetAddrPort(to)
 	allow := lh.GetRemoteAllowList().Allow(vpnAddr, udpAddr.Addr())
-	if lh.l.Level >= logrus.TraceLevel {
+	if lh.l.Enabled(context.Background(), logging.LevelTrace) {
-		lh.l.WithField("vpnAddr", vpnAddr).WithField("udpAddr", udpAddr).WithField("allow", allow).
+		lh.l.Log(context.Background(), logging.LevelTrace, "remoteAllowList.Allow",
-			Trace("remoteAllowList.Allow")
+			"vpnAddr", vpnAddr,
+			"udpAddr", udpAddr,
+			"allow", allow,
+		)
 	}
 
 	if !allow {
@@ -693,9 +718,12 @@ func (lh *LightHouse) unlockedShouldAddV4(vpnAddr netip.Addr, to *V4AddrPort) bo
 func (lh *LightHouse) unlockedShouldAddV6(vpnAddr netip.Addr, to *V6AddrPort) bool {
 	udpAddr := protoV6AddrPortToNetAddrPort(to)
 	allow := lh.GetRemoteAllowList().Allow(vpnAddr, udpAddr.Addr())
-	if lh.l.Level >= logrus.TraceLevel {
+	if lh.l.Enabled(context.Background(), logging.LevelTrace) {
-		lh.l.WithField("vpnAddr", vpnAddr).WithField("udpAddr", udpAddr).WithField("allow", allow).
+		lh.l.Log(context.Background(), logging.LevelTrace, "remoteAllowList.Allow",
-			Trace("remoteAllowList.Allow")
+			"vpnAddr", vpnAddr,
+			"udpAddr", udpAddr,
+			"allow", allow,
+		)
 	}
 
 	if !allow {
@@ -711,23 +739,16 @@ func (lh *LightHouse) unlockedShouldAddV6(vpnAddr netip.Addr, to *V6AddrPort) bo
 
 func (lh *LightHouse) IsLighthouseAddr(vpnAddr netip.Addr) bool {
 	l := lh.GetLighthouses()
-	for i := range l {
+	return slices.Contains(l, vpnAddr)
-		if l[i] == vpnAddr {
-			return true
-		}
-	}
-	return false
 }
 
 func (lh *LightHouse) IsAnyLighthouseAddr(vpnAddrs []netip.Addr) bool {
 	l := lh.GetLighthouses()
 	for i := range vpnAddrs {
-		for j := range l {
+		if slices.Contains(l, vpnAddrs[i]) {
-			if l[j] == vpnAddrs[i] {
 			return true
 		}
 	}
-	}
 	return false
 }
 
@@ -777,8 +798,10 @@ func (lh *LightHouse) innerQueryServer(addr netip.Addr, nb, out []byte) {
 
 		if v == cert.Version1 {
 			if !addr.Is4() {
-				lh.l.WithField("queryVpnAddr", addr).WithField("lighthouseAddr", lhVpnAddr).
+				lh.l.Error("Can't query lighthouse for v6 address using a v1 protocol",
-					Error("Can't query lighthouse for v6 address using a v1 protocol")
+					"queryVpnAddr", addr,
+					"lighthouseAddr", lhVpnAddr,
+				)
 				continue
 			}
 
@@ -789,9 +812,11 @@ func (lh *LightHouse) innerQueryServer(addr netip.Addr, nb, out []byte) {
 
 				v1Query, err = msg.Marshal()
 				if err != nil {
-					lh.l.WithError(err).WithField("queryVpnAddr", addr).
+					lh.l.Error("Failed to marshal lighthouse v1 query payload",
-						WithField("lighthouseAddr", lhVpnAddr).
+						"error", err,
-						Error("Failed to marshal lighthouse v1 query payload")
+						"queryVpnAddr", addr,
+						"lighthouseAddr", lhVpnAddr,
+					)
 					continue
 				}
 			}
@@ -806,9 +831,11 @@ func (lh *LightHouse) innerQueryServer(addr netip.Addr, nb, out []byte) {
 
 				v2Query, err = msg.Marshal()
 				if err != nil {
-					lh.l.WithError(err).WithField("queryVpnAddr", addr).
+					lh.l.Error("Failed to marshal lighthouse v2 query payload",
-						WithField("lighthouseAddr", lhVpnAddr).
+						"error", err,
-						Error("Failed to marshal lighthouse v2 query payload")
+						"queryVpnAddr", addr,
+						"lighthouseAddr", lhVpnAddr,
+					)
 					continue
 				}
 			}
@@ -817,7 +844,11 @@ func (lh *LightHouse) innerQueryServer(addr netip.Addr, nb, out []byte) {
 			queried++
 
 		} else {
-			lh.l.Debugf("Can not query lighthouse for %v using unknown protocol version: %v", addr, v)
+			lh.l.Debug("unsupported protocol version",
+				"op", "query",
+				"queryVpnAddr", addr,
+				"version", v,
+			)
 			continue
 		}
 	}
@@ -846,11 +877,24 @@ func (lh *LightHouse) StartUpdateWorker() {
 				return
 			case <-clockSource.C:
 				continue
+			case <-lh.updateTrigger:
+				continue
 			}
 		}
 	}()
 }
 
+// TriggerUpdate requests an immediate lighthouse update. This is a non-blocking
+// operation intended to be called after a handshake completes with a lighthouse,
+// so the lighthouse has our current addresses without waiting for the next
+// periodic update.
+func (lh *LightHouse) TriggerUpdate() {
+	select {
+	case lh.updateTrigger <- struct{}{}:
+	default:
+	}
+}
+
 func (lh *LightHouse) SendUpdate() {
 	var v4 []*V4AddrPort
 	var v6 []*V6AddrPort
@@ -896,8 +940,9 @@ func (lh *LightHouse) SendUpdate() {
 		if v == cert.Version1 {
 			if v1Update == nil {
 				if !lh.myVpnNetworks[0].Addr().Is4() {
-					lh.l.WithField("lighthouseAddr", lhVpnAddr).
+					lh.l.Warn("cannot update lighthouse using v1 protocol without an IPv4 address",
-						Warn("cannot update lighthouse using v1 protocol without an IPv4 address")
+						"lighthouseAddr", lhVpnAddr,
+					)
 					continue
 				}
 				var relays []uint32
@@ -921,8 +966,10 @@ func (lh *LightHouse) SendUpdate() {
 
 				v1Update, err = msg.Marshal()
 				if err != nil {
-					lh.l.WithError(err).WithField("lighthouseAddr", lhVpnAddr).
+					lh.l.Error("Error while marshaling for lighthouse v1 update",
-						Error("Error while marshaling for lighthouse v1 update")
+						"error", err,
+						"lighthouseAddr", lhVpnAddr,
+					)
 					continue
 				}
 			}
@@ -948,8 +995,10 @@ func (lh *LightHouse) SendUpdate() {
 
 				v2Update, err = msg.Marshal()
 				if err != nil {
-					lh.l.WithError(err).WithField("lighthouseAddr", lhVpnAddr).
+					lh.l.Error("Error while marshaling for lighthouse v2 update",
-						Error("Error while marshaling for lighthouse v2 update")
+						"error", err,
+						"lighthouseAddr", lhVpnAddr,
+					)
 					continue
 				}
 			}
@@ -958,7 +1007,10 @@ func (lh *LightHouse) SendUpdate() {
 			updated++
 
 		} else {
-			lh.l.Debugf("Can not update lighthouse using unknown protocol version: %v", v)
+			lh.l.Debug("unsupported protocol version",
+				"op", "update",
+				"version", v,
+			)
 			continue
 		}
 	}
@@ -972,7 +1024,7 @@ type LightHouseHandler struct {
 	out  []byte
 	pb   []byte
 	meta *NebulaMeta
-	l    *logrus.Logger
+	l    *slog.Logger
 }
 
 func (lh *LightHouse) NewRequestHandler() *LightHouseHandler {
@@ -1021,14 +1073,19 @@ func (lhh *LightHouseHandler) HandleRequest(rAddr netip.AddrPort, fromVpnAddrs [
 	n := lhh.resetMeta()
 	err := n.Unmarshal(p)
 	if err != nil {
-		lhh.l.WithError(err).WithField("vpnAddrs", fromVpnAddrs).WithField("udpAddr", rAddr).
+		lhh.l.Error("Failed to unmarshal lighthouse packet",
-			Error("Failed to unmarshal lighthouse packet")
+			"error", err,
+			"vpnAddrs", fromVpnAddrs,
+			"udpAddr", rAddr,
+		)
 		return
 	}
 
 	if n.Details == nil {
-		lhh.l.WithField("vpnAddrs", fromVpnAddrs).WithField("udpAddr", rAddr).
+		lhh.l.Error("Invalid lighthouse update",
-			Error("Invalid lighthouse update")
+			"vpnAddrs", fromVpnAddrs,
+			"udpAddr", rAddr,
+		)
 		return
 	}
 
@@ -1056,25 +1113,29 @@ func (lhh *LightHouseHandler) HandleRequest(rAddr netip.AddrPort, fromVpnAddrs [
 func (lhh *LightHouseHandler) handleHostQuery(n *NebulaMeta, fromVpnAddrs []netip.Addr, addr netip.AddrPort, w EncWriter) {
 	// Exit if we don't answer queries
 	if !lhh.lh.amLighthouse {
-		if lhh.l.Level >= logrus.DebugLevel {
+		if lhh.l.Enabled(context.Background(), slog.LevelDebug) {
-			lhh.l.Debugln("I don't answer queries, but received from: ", addr)
+			lhh.l.Debug("I don't answer queries, but received one", "from", addr)
 		}
 		return
 	}
 
 	queryVpnAddr, useVersion, err := n.Details.GetVpnAddrAndVersion()
 	if err != nil {
-		if lhh.l.Level >= logrus.DebugLevel {
+		if lhh.l.Enabled(context.Background(), slog.LevelDebug) {
-			lhh.l.WithField("from", fromVpnAddrs).WithField("details", n.Details).
+			lhh.l.Debug("Dropping malformed HostQuery",
-				Debugln("Dropping malformed HostQuery")
+				"from", fromVpnAddrs,
+				"details", n.Details,
+			)
 		}
 		return
 	}
 	if useVersion == cert.Version1 && queryVpnAddr.Is6() {
 		// this case really shouldn't be possible to represent, but reject it anyway.
-		if lhh.l.Level >= logrus.DebugLevel {
+		if lhh.l.Enabled(context.Background(), slog.LevelDebug) {
-			lhh.l.WithField("vpnAddrs", fromVpnAddrs).WithField("queryVpnAddr", queryVpnAddr).
+			lhh.l.Debug("invalid vpn addr for v1 handleHostQuery",
-				Debugln("invalid vpn addr for v1 handleHostQuery")
+				"vpnAddrs", fromVpnAddrs,
+				"queryVpnAddr", queryVpnAddr,
+			)
 		}
 		return
 	}
@@ -1099,7 +1160,10 @@ func (lhh *LightHouseHandler) handleHostQuery(n *NebulaMeta, fromVpnAddrs []neti
 	}
 
 	if err != nil {
-		lhh.l.WithError(err).WithField("vpnAddrs", fromVpnAddrs).Error("Failed to marshal lighthouse host query reply")
+		lhh.l.Error("Failed to marshal lighthouse host query reply",
+			"error", err,
+			"vpnAddrs", fromVpnAddrs,
+		)
 		return
 	}
 
@@ -1127,8 +1191,10 @@ func (lhh *LightHouseHandler) sendHostPunchNotification(n *NebulaMeta, fromVpnAd
 			if ok {
 				whereToPunch = newDest
 			} else {
-				if lhh.l.Level >= logrus.DebugLevel {
+				if lhh.l.Enabled(context.Background(), slog.LevelDebug) {
-					lhh.l.WithField("to", crt.Networks()).Debugln("unable to punch to host, no addresses in common")
+					lhh.l.Debug("unable to punch to host, no addresses in common",
+						"to", crt.Networks(),
+					)
 				}
 			}
 		}
@@ -1154,7 +1220,10 @@ func (lhh *LightHouseHandler) sendHostPunchNotification(n *NebulaMeta, fromVpnAd
 	}
 
 	if err != nil {
-		lhh.l.WithError(err).WithField("vpnAddrs", fromVpnAddrs).Error("Failed to marshal lighthouse host was queried for")
+		lhh.l.Error("Failed to marshal lighthouse host was queried for",
+			"error", err,
+			"vpnAddrs", fromVpnAddrs,
+		)
 		return
 	}
 
@@ -1196,8 +1265,11 @@ func (lhh *LightHouseHandler) coalesceAnswers(v cert.Version, c *cache, n *Nebul
 				n.Details.RelayVpnAddrs = append(n.Details.RelayVpnAddrs, netAddrToProtoAddr(r))
 			}
 		} else {
-			if lhh.l.Level >= logrus.DebugLevel {
+			if lhh.l.Enabled(context.Background(), slog.LevelDebug) {
-				lhh.l.WithField("version", v).Debug("unsupported protocol version")
+				lhh.l.Debug("unsupported protocol version",
+					"op", "coalesceAnswers",
+					"version", v,
+				)
 			}
 		}
 	}
@@ -1210,8 +1282,11 @@ func (lhh *LightHouseHandler) handleHostQueryReply(n *NebulaMeta, fromVpnAddrs [
 
 	certVpnAddr, _, err := n.Details.GetVpnAddrAndVersion()
 	if err != nil {
-		if lhh.l.Level >= logrus.DebugLevel {
+		if lhh.l.Enabled(context.Background(), slog.LevelDebug) {
-			lhh.l.WithError(err).WithField("vpnAddrs", fromVpnAddrs).Error("dropping malformed HostQueryReply")
+			lhh.l.Error("dropping malformed HostQueryReply",
+				"error", err,
+				"vpnAddrs", fromVpnAddrs,
+			)
 		}
 		return
 	}
@@ -1236,8 +1311,8 @@ func (lhh *LightHouseHandler) handleHostQueryReply(n *NebulaMeta, fromVpnAddrs [
 
 func (lhh *LightHouseHandler) handleHostUpdateNotification(n *NebulaMeta, fromVpnAddrs []netip.Addr, w EncWriter) {
 	if !lhh.lh.amLighthouse {
-		if lhh.l.Level >= logrus.DebugLevel {
+		if lhh.l.Enabled(context.Background(), slog.LevelDebug) {
-			lhh.l.Debugln("I am not a lighthouse, do not take host updates: ", fromVpnAddrs)
+			lhh.l.Debug("I am not a lighthouse, do not take host updates", "from", fromVpnAddrs)
 		}
 		return
 	}
@@ -1260,8 +1335,11 @@ func (lhh *LightHouseHandler) handleHostUpdateNotification(n *NebulaMeta, fromVp
 
 	//Simple check that the host sent this not someone else, if detailsVpnAddr is filled
 	if detailsVpnAddr.IsValid() && !slices.Contains(fromVpnAddrs, detailsVpnAddr) {
-		if lhh.l.Level >= logrus.DebugLevel {
+		if lhh.l.Enabled(context.Background(), slog.LevelDebug) {
-			lhh.l.WithField("vpnAddrs", fromVpnAddrs).WithField("answer", detailsVpnAddr).Debugln("Host sent invalid update")
+			lhh.l.Debug("Host sent invalid update",
+				"vpnAddrs", fromVpnAddrs,
+				"answer", detailsVpnAddr,
+			)
 		}
 		return
 	}
@@ -1283,7 +1361,9 @@ func (lhh *LightHouseHandler) handleHostUpdateNotification(n *NebulaMeta, fromVp
 	switch useVersion {
 	case cert.Version1:
 		if !fromVpnAddrs[0].Is4() {
-			lhh.l.WithField("vpnAddrs", fromVpnAddrs).Error("Can not send HostUpdateNotificationAck for a ipv6 vpn ip in a v1 message")
+			lhh.l.Error("Can not send HostUpdateNotificationAck for a ipv6 vpn ip in a v1 message",
+				"vpnAddrs", fromVpnAddrs,
+			)
 			return
 		}
 		vpnAddrB := fromVpnAddrs[0].As4()
@@ -1291,13 +1371,16 @@ func (lhh *LightHouseHandler) handleHostUpdateNotification(n *NebulaMeta, fromVp
 	case cert.Version2:
 		// do nothing, we want to send a blank message
 	default:
-		lhh.l.WithField("useVersion", useVersion).Error("invalid protocol version")
+		lhh.l.Error("invalid protocol version", "useVersion", useVersion)
 		return
 	}
 
 	ln, err := n.MarshalTo(lhh.pb)
 	if err != nil {
-		lhh.l.WithError(err).WithField("vpnAddrs", fromVpnAddrs).Error("Failed to marshal lighthouse host update ack")
+		lhh.l.Error("Failed to marshal lighthouse host update ack",
+			"error", err,
+			"vpnAddrs", fromVpnAddrs,
+		)
 		return
 	}
 
@@ -1314,8 +1397,11 @@ func (lhh *LightHouseHandler) handleHostPunchNotification(n *NebulaMeta, fromVpn
 
 	detailsVpnAddr, _, err := n.Details.GetVpnAddrAndVersion()
 	if err != nil {
-		if lhh.l.Level >= logrus.DebugLevel {
+		if lhh.l.Enabled(context.Background(), slog.LevelDebug) {
-			lhh.l.WithField("details", n.Details).WithError(err).Debugln("dropping invalid HostPunchNotification")
+			lhh.l.Debug("dropping invalid HostPunchNotification",
+				"details", n.Details,
+				"error", err,
+			)
 		}
 		return
 	}
@@ -1332,17 +1418,27 @@ func (lhh *LightHouseHandler) handleHostPunchNotification(n *NebulaMeta, fromVpn
 			lhh.lh.punchConn.WriteTo(empty, vpnPeer)
 		}()
 
-		if lhh.l.Level >= logrus.DebugLevel {
+		if lhh.l.Enabled(context.Background(), slog.LevelDebug) {
-			lhh.l.Debugf("Punching on %v for %v", vpnPeer, logVpnAddr)
+			lhh.l.Debug("Punching",
+				"vpnPeer", vpnPeer,
+				"logVpnAddr", logVpnAddr,
+			)
 		}
 	}
 
+	remoteAllowList := lhh.lh.GetRemoteAllowList()
 	for _, a := range n.Details.V4AddrPorts {
-		punch(protoV4AddrPortToNetAddrPort(a), detailsVpnAddr)
+		b := protoV4AddrPortToNetAddrPort(a)
+		if remoteAllowList.Allow(detailsVpnAddr, b.Addr()) {
+			punch(b, detailsVpnAddr)
+		}
 	}
 
 	for _, a := range n.Details.V6AddrPorts {
-		punch(protoV6AddrPortToNetAddrPort(a), detailsVpnAddr)
+		b := protoV6AddrPortToNetAddrPort(a)
+		if remoteAllowList.Allow(detailsVpnAddr, b.Addr()) {
+			punch(b, detailsVpnAddr)
+		}
 	}
 
 	// This sends a nebula test packet to the host trying to contact us. In the case
@@ -1351,8 +1447,10 @@ func (lhh *LightHouseHandler) handleHostPunchNotification(n *NebulaMeta, fromVpn
 	if lhh.lh.punchy.GetRespond() {
 		go func() {
 			time.Sleep(lhh.lh.punchy.GetRespondDelay())
-			if lhh.l.Level >= logrus.DebugLevel {
+			if lhh.l.Enabled(context.Background(), slog.LevelDebug) {
-				lhh.l.Debugf("Sending a nebula test packet to vpn addr %s", detailsVpnAddr)
+				lhh.l.Debug("Sending a nebula test packet",
+					"vpnAddr", detailsVpnAddr,
+				)
 			}
 			//NOTE: we have to allocate a new output buffer here since we are spawning a new goroutine
 			// for each punchBack packet. We should move this into a timerwheel or a single goroutine

lighthouse_test.go

@@ -1,7 +1,6 @@
 package nebula
 
 import (
-	"context"
 	"encoding/binary"
 	"fmt"
 	"net/netip"
@@ -14,7 +13,7 @@ import (
 	"github.com/slackhq/nebula/test"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
-	"gopkg.in/yaml.v3"
+	"go.yaml.in/yaml/v3"
 )
 
 func TestOldIPv4Only(t *testing.T) {
@@ -42,14 +41,14 @@ func Test_lhStaticMapping(t *testing.T) {
 	c := config.NewC(l)
 	c.Settings["lighthouse"] = map[string]any{"hosts": []any{lh1}}
 	c.Settings["static_host_map"] = map[string]any{lh1: []any{"1.1.1.1:4242"}}
-	_, err := NewLightHouseFromConfig(context.Background(), l, c, cs, nil, nil)
+	_, err := NewLightHouseFromConfig(t.Context(), l, c, cs, nil, nil)
 	require.NoError(t, err)
 
 	lh2 := "10.128.0.3"
 	c = config.NewC(l)
 	c.Settings["lighthouse"] = map[string]any{"hosts": []any{lh1, lh2}}
 	c.Settings["static_host_map"] = map[string]any{lh1: []any{"100.1.1.1:4242"}}
-	_, err = NewLightHouseFromConfig(context.Background(), l, c, cs, nil, nil)
+	_, err = NewLightHouseFromConfig(t.Context(), l, c, cs, nil, nil)
 	require.EqualError(t, err, "lighthouse 10.128.0.3 does not have a static_host_map entry")
 }
 
@@ -71,7 +70,7 @@ func TestReloadLighthouseInterval(t *testing.T) {
 	}
 
 	c.Settings["static_host_map"] = map[string]any{lh1: []any{"1.1.1.1:4242"}}
-	lh, err := NewLightHouseFromConfig(context.Background(), l, c, cs, nil, nil)
+	lh, err := NewLightHouseFromConfig(t.Context(), l, c, cs, nil, nil)
 	require.NoError(t, err)
 	lh.ifce = &mockEncWriter{}
 
@@ -99,7 +98,7 @@ func BenchmarkLighthouseHandleRequest(b *testing.B) {
 	}
 
 	c := config.NewC(l)
-	lh, err := NewLightHouseFromConfig(context.Background(), l, c, cs, nil, nil)
+	lh, err := NewLightHouseFromConfig(b.Context(), l, c, cs, nil, nil)
 	require.NoError(b, err)
 
 	hAddr := netip.MustParseAddrPort("4.5.6.7:12345")
@@ -202,7 +201,7 @@ func TestLighthouse_Memory(t *testing.T) {
 		myVpnNetworks:      []netip.Prefix{myVpnNet},
 		myVpnNetworksTable: nt,
 	}
-	lh, err := NewLightHouseFromConfig(context.Background(), l, c, cs, nil, nil)
+	lh, err := NewLightHouseFromConfig(t.Context(), l, c, cs, nil, nil)
 	lh.ifce = &mockEncWriter{}
 	require.NoError(t, err)
 	lhh := lh.NewRequestHandler()
@@ -288,7 +287,7 @@ func TestLighthouse_reload(t *testing.T) {
 		myVpnNetworksTable: nt,
 	}
 
-	lh, err := NewLightHouseFromConfig(context.Background(), l, c, cs, nil, nil)
+	lh, err := NewLightHouseFromConfig(t.Context(), l, c, cs, nil, nil)
 	require.NoError(t, err)
 
 	nc := map[string]any{
@@ -523,7 +522,7 @@ func TestLighthouse_Dont_Delete_Static_Hosts(t *testing.T) {
 		myVpnNetworks:      []netip.Prefix{myVpnNet},
 		myVpnNetworksTable: nt,
 	}
-	lh, err := NewLightHouseFromConfig(context.Background(), l, c, cs, nil, nil)
+	lh, err := NewLightHouseFromConfig(t.Context(), l, c, cs, nil, nil)
 	require.NoError(t, err)
 	lh.ifce = &mockEncWriter{}
 
@@ -589,7 +588,7 @@ func TestLighthouse_DeletesWork(t *testing.T) {
 		myVpnNetworks:      []netip.Prefix{myVpnNet},
 		myVpnNetworksTable: nt,
 	}
-	lh, err := NewLightHouseFromConfig(context.Background(), l, c, cs, nil, nil)
+	lh, err := NewLightHouseFromConfig(t.Context(), l, c, cs, nil, nil)
 	require.NoError(t, err)
 	lh.ifce = &mockEncWriter{}
 

logger.go

@@ -1,45 +0,0 @@
-package nebula
-
-import (
-	"fmt"
-	"strings"
-	"time"
-
-	"github.com/sirupsen/logrus"
-	"github.com/slackhq/nebula/config"
-)
-
-func configLogger(l *logrus.Logger, c *config.C) error {
-	// set up our logging level
-	logLevel, err := logrus.ParseLevel(strings.ToLower(c.GetString("logging.level", "info")))
-	if err != nil {
-		return fmt.Errorf("%s; possible levels: %s", err, logrus.AllLevels)
-	}
-	l.SetLevel(logLevel)
-
-	disableTimestamp := c.GetBool("logging.disable_timestamp", false)
-	timestampFormat := c.GetString("logging.timestamp_format", "")
-	fullTimestamp := (timestampFormat != "")
-	if timestampFormat == "" {
-		timestampFormat = time.RFC3339
-	}
-
-	logFormat := strings.ToLower(c.GetString("logging.format", "text"))
-	switch logFormat {
-	case "text":
-		l.Formatter = &logrus.TextFormatter{
-			TimestampFormat:  timestampFormat,
-			FullTimestamp:    fullTimestamp,
-			DisableTimestamp: disableTimestamp,
-		}
-	case "json":
-		l.Formatter = &logrus.JSONFormatter{
-			TimestampFormat:  timestampFormat,
-			DisableTimestamp: disableTimestamp,
-		}
-	default:
-		return fmt.Errorf("unknown log format `%s`. possible formats: %s", logFormat, []string{"text", "json"})
-	}
-
-	return nil
-}