Bump go.yaml.in/yaml/v3 from 3.0.4 to 3.0.5

Bumps [go.yaml.in/yaml/v3](https://github.com/yaml/go-yaml) from 3.0.4 to 3.0.5. - [Commits](https://github.com/yaml/go-yaml/compare/v3.0.4...v3.0.5) --- updated-dependencies: - dependency-name: go.yaml.in/yaml/v3 dependency-version: 3.0.5 dependency-type: direct:production update-type: version-update:semver-patch ... Signed-off-by: dependabot[bot] <support@github.com>
Add an e2e Drop exit type and a roaming recovery measurement (#1819 )
2026-08-15 09:57:00 +02:00 · 2026-07-27 19:53:03 +00:00 · 2026-07-23 17:02:02 -05:00 · 2026-07-23 13:13:45 -05:00 · 2026-07-23 10:36:20 -05:00 · 2026-07-23 09:26:45 -05:00
204 changed files with 21286 additions and 5480 deletions
@@ -0,0 +1,116 @@
+name: Code-sign Windows binaries
+description: >
+  Sign every .exe under a given path in place via the DefinedNet code-signer
+  Lambda. If `role` or `bucket` is empty, logs a notice and skips signing so
+  forks and dev branches without AWS access still produce usable builds.
+
+inputs:
+  path:
+    description: "Directory whose .exe files should be signed in place"
+    required: true
+  role:
+    description: "IAM role ARN to assume via OIDC; empty disables signing"
+    required: false
+    default: ""
+  bucket:
+    description: "S3 staging bucket the code-signer Lambda reads from; empty disables signing"
+    required: false
+    default: ""
+  region:
+    description: "AWS region for the role and Lambda"
+    required: false
+    default: "us-east-2"
+  function-name:
+    description: "Code-signer Lambda function name"
+    required: false
+    default: "code-signer"
+  key-prefix:
+    description: "S3 key prefix to write under; defaults to code-signing/<owner>/<repo> of the calling repo"
+    required: false
+    default: ""
+
+runs:
+  using: composite
+  steps:
+    - name: Skip notice
+      if: inputs.role == '' || inputs.bucket == ''
+      shell: sh
+      run: echo "::notice::code-signer role or bucket not set; skipping code signing."
+
+    - name: Configure AWS credentials
+      if: inputs.role != '' && inputs.bucket != ''
+      uses: aws-actions/configure-aws-credentials@v6
+      with:
+        role-to-assume: ${{ inputs.role }}
+        aws-region: ${{ inputs.region }}
+        # Default is 12 retries to ride out IAM trust-policy propagation; once
+        # the role is stable we want a real misconfiguration to fail fast.
+        retry-max-attempts: 5
+
+    - name: Sign .exe files
+      if: inputs.role != '' && inputs.bucket != ''
+      shell: sh
+      env:
+        SIGN_PATH: ${{ inputs.path }}
+        BUCKET: ${{ inputs.bucket }}
+        FUNCTION_NAME: ${{ inputs.function-name }}
+        KEY_PREFIX: ${{ inputs.key-prefix }}
+      run: |
+        set -eu
+        # Default the prefix to this repo so the S3 key attributes the sign correctly.
+        # nebula-nightly runs this same action but writes under its own repo's prefix.
+        KEY_PREFIX="${KEY_PREFIX:-code-signing/$GITHUB_REPOSITORY}"
+        RUN="${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}"
+
+        find "$SIGN_PATH" -name '*.exe' -print | while read -r path
+        do
+            rel=${path#"$SIGN_PATH"/}
+            file=$(basename "$path")
+            name=${file%.exe}
+            prefix="${KEY_PREFIX}/${RUN}"
+            src="${prefix}/unsigned/${rel}"
+            dst="${prefix}/signed/${rel}"
+
+            echo "::group::Sign ${rel}"
+            echo "Uploading unsigned to s3://${BUCKET}/${src}"
+            aws s3 cp --no-progress "$path" "s3://${BUCKET}/${src}" >/dev/null
+
+            echo "Invoking ${FUNCTION_NAME} Lambda"
+            payload=$(jq -nc \
+                --arg s "$src" \
+                --arg d "$dst" \
+                --arg p "$name" \
+                '{source_key: $s, dest_key: $d, program_name: $p}')
+            meta=$(aws lambda invoke \
+                --function-name "$FUNCTION_NAME" \
+                --cli-binary-format raw-in-base64-out \
+                --payload "$payload" \
+                --output json \
+                /tmp/sign-resp.json)
+            if echo "$meta" | jq -e '.FunctionError != null' >/dev/null
+            then
+                echo "::endgroup::"
+                echo "::error::code-signer Lambda failed for ${rel}"
+                cat /tmp/sign-resp.json >&2
+                exit 1
+            fi
+
+            echo "Downloading signed back to ${path}"
+            aws s3 cp --no-progress "s3://${BUCKET}/${dst}" "$path" >/dev/null
+
+            aws s3 rm "s3://${BUCKET}/${src}" >/dev/null 2>&1 || true
+            aws s3 rm "s3://${BUCKET}/${dst}" >/dev/null 2>&1 || true
+
+            # Sanity-check the bytes we got back actually carry an Authenticode
+            # signature that this machine can validate end to end.
+            status=$(powershell -NoProfile -Command "(Get-AuthenticodeSignature -FilePath '$path').Status" | tr -d '\r')
+            if [ "$status" != "Valid" ]
+            then
+                echo "::endgroup::"
+                echo "::error::${rel} signature status: ${status} (expected Valid)"
+                exit 1
+            fi
+
+            echo "Signed ${rel} (sha256=$(jq -r '.sha256' /tmp/sign-resp.json), status=${status})"
+            echo "::endgroup::"
+        done
@@ -1,34 +0,0 @@
-name: gofmt
-on:
-  push:
-    branches:
-      - master
-  pull_request:
-    paths:
-      - '.github/workflows/gofmt.yml'
-      - '**.go'
-jobs:
-
-  gofmt:
-    name: Run gofmt
-    runs-on: ubuntu-latest
-    steps:
-
-    - uses: actions/checkout@v4
-
-    - uses: actions/setup-go@v6
-      with:
-        go-version: '1.25'
-        check-latest: true
-
-    - name: Install goimports
-      run: |
-        go install golang.org/x/tools/cmd/goimports@latest
-
-    - name: gofmt
-      run: |
-        if [ "$(find . -iname '*.go' | grep -v '\.pb\.go$' | xargs goimports -l)" ]
-        then
-          find . -iname '*.go' | grep -v '\.pb\.go$' | xargs goimports -d
-          exit 1
-        fi
@@ -10,11 +10,11 @@ jobs:
    name: Build Linux/BSD All
    runs-on: ubuntu-latest
    steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v7

-      - uses: actions/setup-go@v6
+      - uses: actions/setup-go@v7
        with:
-          go-version: '1.25'
+          go-version: '1.26'
          check-latest: true

      - name: Build
@@ -24,7 +24,7 @@ jobs:
          mv build/*.tar.gz release

      - name: Upload artifacts
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v7
        with:
          name: linux-latest
          path: release
@@ -32,12 +32,15 @@ jobs:
  build-windows:
    name: Build Windows
    runs-on: windows-latest
+    permissions:
+      id-token: write
+      contents: read
    steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v7

-      - uses: actions/setup-go@v6
+      - uses: actions/setup-go@v7
        with:
-          go-version: '1.25'
+          go-version: '1.26'
          check-latest: true

      - name: Build
@@ -54,8 +57,15 @@ jobs:
          mkdir build\dist\windows
          mv dist\windows\wintun build\dist\windows\

+      - name: Code-sign
+        uses: ./.github/actions/code-sign
+        with:
+          path: build
+          role: ${{ secrets.DEFINED_CODE_SIGNER_ROLE }}
+          bucket: ${{ secrets.DEFINED_CODE_SIGNER_BUCKET }}
+
      - name: Upload artifacts
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v7
        with:
          name: windows-latest
          path: build
@@ -66,16 +76,16 @@ jobs:
      HAS_SIGNING_CREDS: ${{ secrets.AC_USERNAME != '' }}
    runs-on: macos-latest
    steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v7

-      - uses: actions/setup-go@v6
+      - uses: actions/setup-go@v7
        with:
-          go-version: '1.25'
+          go-version: '1.26'
          check-latest: true

      - name: Import certificates
        if: env.HAS_SIGNING_CREDS == 'true'
-        uses: Apple-Actions/import-codesign-certs@v5
+        uses: Apple-Actions/import-codesign-certs@v7
        with:
          p12-file-base64: ${{ secrets.APPLE_DEVELOPER_CERTIFICATE_P12_BASE64 }}
          p12-password: ${{ secrets.APPLE_DEVELOPER_CERTIFICATE_PASSWORD }}
@@ -104,7 +114,7 @@ jobs:
          fi

      - name: Upload artifacts
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v7
        with:
          name: darwin-latest
          path: ./release/*
@@ -124,25 +134,25 @@ jobs:
      # be overwritten
      - name: Checkout code
        if: ${{ env.HAS_DOCKER_CREDS == 'true' }}
-        uses: actions/checkout@v4
+        uses: actions/checkout@v7

      - name: Download artifacts
        if: ${{ env.HAS_DOCKER_CREDS == 'true' }}
-        uses: actions/download-artifact@v4
+        uses: actions/download-artifact@v8
        with:
          name: linux-latest
          path: artifacts

      - name: Login to Docker Hub
        if: ${{ env.HAS_DOCKER_CREDS == 'true' }}
-        uses: docker/login-action@v3
+        uses: docker/login-action@v4
        with:
          username: ${{ vars.DOCKERHUB_USERNAME }}
          password: ${{ secrets.DOCKERHUB_TOKEN }}

      - name: Set up Docker Buildx
        if: ${{ env.HAS_DOCKER_CREDS == 'true' }}
-        uses: docker/setup-buildx-action@v3
+        uses: docker/setup-buildx-action@v4

      - name: Build and push images
        if: ${{ env.HAS_DOCKER_CREDS == 'true' }}
@@ -153,17 +163,20 @@ jobs:
          mkdir -p build/linux-{amd64,arm64}
          tar -zxvf artifacts/nebula-linux-amd64.tar.gz -C build/linux-amd64/
          tar -zxvf artifacts/nebula-linux-arm64.tar.gz -C build/linux-arm64/
-          docker buildx build . --push -f docker/Dockerfile --platform linux/amd64,linux/arm64 --tag "${DOCKER_IMAGE_REPO}:${DOCKER_IMAGE_TAG}" --tag "${DOCKER_IMAGE_REPO}:${GITHUB_REF#refs/tags/v}"
+          docker buildx build . --push -f docker/Dockerfile --platform linux/amd64,linux/arm64 \
+            --build-arg VERSION="${GITHUB_REF#refs/tags/v}" \
+            --build-arg REVISION="${GITHUB_SHA}" \
+            --tag "${DOCKER_IMAGE_REPO}:${DOCKER_IMAGE_TAG}" --tag "${DOCKER_IMAGE_REPO}:${GITHUB_REF#refs/tags/v}"

  release:
    name: Create and Upload Release
    needs: [build-linux, build-darwin, build-windows]
    runs-on: ubuntu-latest
    steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v7

      - name: Download artifacts
-        uses: actions/download-artifact@v4
+        uses: actions/download-artifact@v8
        with:
          path: artifacts

@@ -209,10 +222,11 @@ 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 ${{ github.ref_name }}" \
-            "${{ github.ref_name }}" \
+            --title "Release ${GITHUB_REF_NAME}" \
+            "${GITHUB_REF_NAME}" \
            SHASUM256.txt *-latest/*.zip *-latest/*.tar.gz
@@ -14,38 +14,119 @@ on:
      - 'go.sum'
 jobs:

-  smoke-extra:
+  smoke-extra-libvirt:
    if: github.ref == 'refs/heads/master' || contains(github.event.pull_request.labels.*.name, 'smoke-test-extra')
-    name: Run extra smoke tests
+    name: ${{ matrix.target }}
    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        target:
+          - freebsd-amd64
+          - openbsd-amd64
+          - netbsd-amd64
+          - linux-amd64-ipv6disable
+    env:
+      VAGRANT_DEFAULT_PROVIDER: libvirt
    steps:

-    - uses: actions/checkout@v4
+    - uses: actions/checkout@v7

-    - uses: actions/setup-go@v6
+    - uses: actions/setup-go@v7
      with:
-        go-version: '1.25'
+        go-version: '1.26'
        check-latest: true

    - 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
-      run: sudo apt-get update && sudo apt-get install -y vagrant virtualbox
+    - name: install vagrant and libvirt
+      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
+    - name: ${{ matrix.target }}
+      run: make smoke-vagrant/${{ matrix.target }}

-    - name: openbsd-amd64
-      run: make smoke-vagrant/openbsd-amd64
+    timeout-minutes: 30

-    - name: netbsd-amd64
-      run: make smoke-vagrant/netbsd-amd64
+  # linux-386 needs VirtualBox, which conflicts with KVM/libvirt -- isolated job.
+  smoke-extra-virtualbox:
+    if: github.ref == 'refs/heads/master' || contains(github.event.pull_request.labels.*.name, 'smoke-test-extra')
+    name: linux-386
+    runs-on: ubuntu-latest
+    env:
+      VAGRANT_DEFAULT_PROVIDER: virtualbox
+    steps:
+
+    - uses: actions/checkout@v7
+
+    - uses: actions/setup-go@v7
+      with:
+        go-version: '1.26'
+        check-latest: true
+
+    - 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 and virtualbox
+      run: |
+        sudo apt-get update && sudo apt-get install -y vagrant virtualbox
+        sudo rmmod kvm_amd kvm_intel kvm 2>/dev/null || true

    - name: linux-386
      run: make smoke-vagrant/linux-386

-    - name: linux-amd64-ipv6disable
-      run: make smoke-vagrant/linux-amd64-ipv6disable
-
    timeout-minutes: 30
+
+  smoke-windows:
+    if: github.ref == 'refs/heads/master' || contains(github.event.pull_request.labels.*.name, 'smoke-test-extra')
+    name: Run windows smoke test
+    runs-on: windows-latest
+    steps:
+
+    - uses: actions/checkout@v7
+
+    - uses: actions/setup-go@v7
+      with:
+        go-version: '1.26'
+        check-latest: true
+
+    # WSL2 + Ubuntu so the smoke can run a real linux peer with its own
+    # netns. iputils-ping is needed for the in-WSL ping check. WSL1 has no
+    # real kernel and would lack /dev/net/tun, so we have to force WSL2.
+    - uses: Vampire/setup-wsl@v3
+      with:
+        distribution: Ubuntu-24.04
+        additional-packages: iputils-ping iproute2
+
+    # Vampire/setup-wsl provisions WSL1 even when the WSL2 platform is present.
+    # Convert the distro to WSL2 explicitly before we try to use /dev/net/tun.
+    - name: convert distro to WSL2
+      shell: pwsh
+      run: |
+        wsl --set-version Ubuntu-24.04 2
+        wsl --shutdown
+        wsl --list --verbose
+
+    - name: build windows nebula
+      run: make bin-windows
+
+    - name: build linux nebula for WSL
+      shell: bash
+      env:
+        GOOS: linux
+        GOARCH: amd64
+      run: |
+        mkdir -p build/linux-amd64
+        go build -o build/linux-amd64/nebula ./cmd/nebula
+
+    - name: run smoke-windows
+      shell: pwsh
+      working-directory: ./.github/workflows/smoke
+      run: ./smoke-windows.ps1
+
+    timeout-minutes: 15
@@ -18,11 +18,11 @@ jobs:
    runs-on: ubuntu-latest
    steps:

-    - uses: actions/checkout@v4
+    - uses: actions/checkout@v7

-    - uses: actions/setup-go@v6
+    - uses: actions/setup-go@v7
      with:
-        go-version: '1.25'
+        go-version: '1.26'
        check-latest: true

    - name: build
@@ -36,6 +36,14 @@ jobs:
      working-directory: ./.github/workflows/smoke
      run: ./smoke.sh

+    - name: setup docker image ipv6
+      working-directory: ./.github/workflows/smoke
+      run: SMOKE_OVERLAY_IPV6=1 ./build.sh
+
+    - name: run smoke ipv6
+      working-directory: ./.github/workflows/smoke
+      run: SMOKE_OVERLAY_IPV6=1 ./smoke.sh
+
    - name: setup relay docker image
      working-directory: ./.github/workflows/smoke
      run: ./build-relay.sh
@@ -16,8 +16,10 @@ relay:
  am_relay: true
 EOF

-    export LIGHTHOUSES="192.168.100.1 172.17.0.2:4242"
-    export REMOTE_ALLOW_LIST='{"172.17.0.4/32": false, "172.17.0.5/32": false}'
+    # TEST-NET-3 placeholder IPs; smoke-relay.sh seds them to real container IPs.
+    # 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

@@ -5,9 +5,28 @@ 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
-# - We could make this better by launching the lighthouse first and then fetching what IP it is.
-NET="$(docker network inspect bridge -f '{{ range .IPAM.Config }}{{ .Subnet }}{{ end }}' | cut -d. -f1-3)"
+if [ "$SMOKE_OVERLAY_IPV6" ]
+then
+    LIGHTHOUSE_NIP="fd00:4242:0:0:0:ffff:c0a8:6401"
+    HOST2_NIP="fd00:4242:0:0:0:ffff:c0a8:6402"
+    HOST3_NIP="fd00:4242:0:0:0:ffff:c0a8:6403"
+    HOST4_NIP="fd00:4242:0:0:0:ffff:c0a8:6404"
+else
+    LIGHTHOUSE_NIP="192.168.100.1"
+    HOST2_NIP="192.168.100.2"
+    HOST3_NIP="192.168.100.3"
+    HOST4_NIP="192.168.100.4"
+fi
+
+# Smoke containers run on a dedicated docker network whose subnet is allocated
+# at smoke time, not known at build time. Configs are written with TEST-NET-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
@@ -25,24 +44,24 @@ NET="$(docker network inspect bridge -f '{{ range .IPAM.Config }}{{ .Subnet }}{{
        ../genconfig.sh >lighthouse1.yml

    HOST="host2" \
-        LIGHTHOUSES="192.168.100.1 $NET.2:4242" \
+        LIGHTHOUSES="$LIGHTHOUSE_NIP $LIGHTHOUSE_IP:4242" \
        ../genconfig.sh >host2.yml

    HOST="host3" \
-        LIGHTHOUSES="192.168.100.1 $NET.2:4242" \
+        LIGHTHOUSES="$LIGHTHOUSE_NIP $LIGHTHOUSE_IP:4242" \
        INBOUND='[{"port": "any", "proto": "icmp", "group": "lighthouse"}]' \
        ../genconfig.sh >host3.yml

    HOST="host4" \
-        LIGHTHOUSES="192.168.100.1 $NET.2:4242" \
+        LIGHTHOUSES="$LIGHTHOUSE_NIP $LIGHTHOUSE_IP:4242" \
        OUTBOUND='[{"port": "any", "proto": "icmp", "group": "lighthouse"}]' \
        ../genconfig.sh >host4.yml

    ../../../../nebula-cert ca -curve "${CURVE:-25519}" -name "Smoke Test"
-    ../../../../nebula-cert sign -name "lighthouse1" -groups "lighthouse,lighthouse1" -ip "192.168.100.1/24"
-    ../../../../nebula-cert sign -name "host2" -groups "host,host2" -ip "192.168.100.2/24"
-    ../../../../nebula-cert sign -name "host3" -groups "host,host3" -ip "192.168.100.3/24"
-    ../../../../nebula-cert sign -name "host4" -groups "host,host4" -ip "192.168.100.4/24"
+    ../../../../nebula-cert sign -name "lighthouse1" -groups "lighthouse,lighthouse1" -ip "$LIGHTHOUSE_NIP/24"
+    ../../../../nebula-cert sign -name "host2" -groups "host,host2" -ip "$HOST2_NIP/24"
+    ../../../../nebula-cert sign -name "host3" -groups "host,host3" -ip "$HOST3_NIP/24"
+    ../../../../nebula-cert sign -name "host4" -groups "host,host4" -ip "$HOST4_NIP/24"
 )

 docker build -t "nebula:${NAME:-smoke}" .
@@ -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
-docker run --name host2 --rm nebula:smoke-relay -config host2.yml -test
-docker run --name host3 --rm nebula:smoke-relay -config host3.yml -test
-docker run --name host4 --rm nebula:smoke-relay -config host4.yml -test
+# 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

-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 [ "$(jobs -r)" ]
 then
@@ -8,6 +8,8 @@ export VAGRANT_CWD="$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'
-sleep 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 [ "$(jobs -r)" ]
 then
@@ -0,0 +1,272 @@
+#!/usr/bin/env pwsh
+# Windows smoke test for the nebula tun + UDP + NLM code paths.
+#
+# Topology:
+#   - lighthouse runs natively on the Windows host (wintun + windows UDP)
+#   - peer runs inside WSL2 (Linux build of nebula, /dev/net/tun)
+#
+# WSL2 gives us a real netns boundary so the loopback fast-path on Windows
+# does not short-circuit the overlay -- when WSL pings the lighthouse VPN IP,
+# Linux has no idea that IP is local to the Windows host, so the packet is
+# forced through nebula. Same in reverse.
+
+$ErrorActionPreference = 'Stop'
+
+# wsl.exe emits UTF-16 LE by default which PowerShell reads as bytes, mangling
+# every captured string. WSL_UTF8 makes wsl.exe emit UTF-8 instead.
+$env:WSL_UTF8 = '1'
+
+$RepoRoot = Resolve-Path "$PSScriptRoot\..\..\.."
+$Nebula = Join-Path $RepoRoot 'nebula.exe'
+$NebulaCert = Join-Path $RepoRoot 'nebula-cert.exe'
+$NebulaLinux = Join-Path $RepoRoot 'build\linux-amd64\nebula'
+
+if (-not (Test-Path $Nebula)) { throw "missing $Nebula; run 'make bin-windows' first" }
+if (-not (Test-Path $NebulaCert)) { throw "missing $NebulaCert; run 'make bin-windows' first" }
+if (-not (Test-Path $NebulaLinux)) { throw "missing $NebulaLinux; build the linux nebula first" }
+
+# Matches the distro installed by Vampire/setup-wsl in smoke-extra.yml.
+$Distro = 'Ubuntu-24.04'
+$listed = (wsl --list --quiet 2>$null) -join "`n"
+if ($listed -notmatch [regex]::Escape($Distro)) {
+    throw "WSL distro $Distro not registered. Got: $listed"
+}
+Write-Host "Using WSL distro: $Distro"
+
+# Windows host as seen from inside WSL: WSL's default-route gateway. We extract
+# it with a regex rather than awk fields so PowerShell does not eat any '$N'
+# tokens, and tabs/double-spaces in `ip route` output do not confuse a cut.
+$ipCmd = 'ip route show default | grep -oE "([0-9]+\.){3}[0-9]+" | head -1'
+$WindowsIp = (wsl -d $Distro -- bash -c $ipCmd).Trim()
+if (-not $WindowsIp) { throw "could not determine Windows host IP from WSL" }
+Write-Host "Windows host IP from WSL: $WindowsIp"
+
+$WorkDir = Join-Path $env:TEMP 'nebula-smoke-windows'
+if (Test-Path $WorkDir) { Remove-Item -Recurse -Force $WorkDir }
+New-Item -ItemType Directory -Path $WorkDir | Out-Null
+
+$WslDir = '/tmp/nebula-smoke'
+wsl -d $Distro -- bash -c "rm -rf $WslDir && mkdir -p $WslDir" | Out-Null
+
+$DevName = 'nebula-smoke'
+$Ip1 = '192.168.241.1'
+$Ip2 = '192.168.241.2'
+$Port = 4242
+
+& $NebulaCert ca -name 'smoke-ca' -out-crt "$WorkDir\ca.crt" -out-key "$WorkDir\ca.key"
+if ($LASTEXITCODE -ne 0) { throw "nebula-cert ca failed (exit $LASTEXITCODE)" }
+
+& $NebulaCert sign -name 'lighthouse' -networks "$Ip1/24" -ca-crt "$WorkDir\ca.crt" -ca-key "$WorkDir\ca.key" -out-crt "$WorkDir\lighthouse.crt" -out-key "$WorkDir\lighthouse.key"
+if ($LASTEXITCODE -ne 0) { throw "nebula-cert sign lighthouse failed (exit $LASTEXITCODE)" }
+
+& $NebulaCert sign -name 'peer' -networks "$Ip2/24" -ca-crt "$WorkDir\ca.crt" -ca-key "$WorkDir\ca.key" -out-crt "$WorkDir\peer.crt" -out-key "$WorkDir\peer.key"
+if ($LASTEXITCODE -ne 0) { throw "nebula-cert sign peer failed (exit $LASTEXITCODE)" }
+
+# Windows lighthouse config.
+@"
+pki:
+  ca: $WorkDir\ca.crt
+  cert: $WorkDir\lighthouse.crt
+  key: $WorkDir\lighthouse.key
+static_host_map: {}
+lighthouse:
+  am_lighthouse: true
+  interval: 60
+  hosts: []
+listen:
+  host: 0.0.0.0
+  port: $Port
+tun:
+  disabled: false
+  dev: $DevName
+  drop_local_broadcast: false
+  drop_multicast: false
+  tx_queue: 500
+  mtu: 1300
+  network_category: private
+logging:
+  level: info
+  format: text
+firewall:
+  outbound_action: drop
+  inbound_action: drop
+  conntrack:
+    tcp_timeout: 12m
+    udp_timeout: 3m
+    default_timeout: 10m
+  outbound:
+    - port: any
+      proto: any
+      host: any
+  inbound:
+    - port: any
+      proto: any
+      host: any
+"@ | Out-File -FilePath "$WorkDir\lighthouse.yml" -Encoding utf8
+
+# WSL peer config (paths are POSIX, deliberately).
+@"
+pki:
+  ca: $WslDir/ca.crt
+  cert: $WslDir/peer.crt
+  key: $WslDir/peer.key
+static_host_map:
+  "${Ip1}": ["${WindowsIp}:$Port"]
+lighthouse:
+  am_lighthouse: false
+  interval: 60
+  hosts:
+    - "${Ip1}"
+listen:
+  host: 0.0.0.0
+  port: 0
+tun:
+  disabled: false
+  dev: nebula1
+  drop_local_broadcast: false
+  drop_multicast: false
+  tx_queue: 500
+  mtu: 1300
+logging:
+  level: info
+  format: text
+firewall:
+  outbound_action: drop
+  inbound_action: drop
+  conntrack:
+    tcp_timeout: 12m
+    udp_timeout: 3m
+    default_timeout: 10m
+  outbound:
+    - port: any
+      proto: any
+      host: any
+  inbound:
+    - port: any
+      proto: any
+      host: any
+"@ | Out-File -FilePath "$WorkDir\peer.yml" -Encoding utf8
+
+# Stage WSL artifacts. Convert Windows paths to WSL paths ourselves rather than
+# calling `wslpath`, because PowerShell's argument-passing to external EXEs
+# strips backslashes from path arguments in ways that are hard to escape around.
+function ConvertTo-WslPath {
+    param([string]$WindowsPath)
+    if ($WindowsPath -notmatch '^([A-Za-z]):\\(.*)$') {
+        throw "cannot convert path to WSL: $WindowsPath"
+    }
+    return "/mnt/$($matches[1].ToLower())/$($matches[2].Replace('\','/'))"
+}
+
+$WslWorkDir = ConvertTo-WslPath $WorkDir
+$WslNebulaPath = ConvertTo-WslPath $NebulaLinux
+wsl -d $Distro -- bash -c "cp '$WslWorkDir/ca.crt' '$WslWorkDir/peer.crt' '$WslWorkDir/peer.key' '$WslWorkDir/peer.yml' $WslDir/ && cp '$WslNebulaPath' $WslDir/nebula && chmod +x $WslDir/nebula"
+
+# Make sure WSL has tun support and /dev/net/tun is usable before starting
+# nebula. Diagnostics first so a fail here points at the real problem (e.g.
+# WSL1 distros do not have a real kernel and will not have tun).
+Write-Host '=== WSL diagnostic ==='
+wsl --version 2>&1 | Out-Host
+wsl --list --verbose 2>&1 | Out-Host
+wsl -d $Distro -u root -- uname -a | Out-Host
+wsl -d $Distro -u root -- bash -c "modprobe tun 2>&1 || true; mkdir -p /dev/net; [ -c /dev/net/tun ] || mknod /dev/net/tun c 10 200; chmod 600 /dev/net/tun; ls -l /dev/net/tun"
+if ($LASTEXITCODE -ne 0) { throw "failed to prepare /dev/net/tun in WSL (TUN support missing?)" }
+
+# Deliberately no New-NetFirewallRule calls here -- nebula's windows_bypass_wdf
+# feature is supposed to install WFP permit filters that let inbound traffic
+# through Windows Defender Firewall on its own. If this smoke regresses, that
+# feature regressed.
+
+$lhOut = Join-Path $WorkDir 'lighthouse.out.log'
+$lhErr = Join-Path $WorkDir 'lighthouse.err.log'
+$lhProc = Start-Process -FilePath $Nebula -ArgumentList @('-config', "$WorkDir\lighthouse.yml") `
+    -PassThru -NoNewWindow `
+    -RedirectStandardOutput $lhOut `
+    -RedirectStandardError $lhErr
+
+# Run nebula in WSL as root with no sudo + no shell wrapper. PowerShell's
+# Start-Process arg quoting mangles `bash -c "..."` strings that contain
+# spaces/redirections, so we skip bash entirely and let Start-Process do the
+# stdout/stderr capture itself.
+$peerOut = Join-Path $WorkDir 'peer.out.log'
+$peerErr = Join-Path $WorkDir 'peer.err.log'
+$peerProc = Start-Process -FilePath 'wsl' `
+    -ArgumentList @('-d', $Distro, '-u', 'root', '--', "$WslDir/nebula", '-config', "$WslDir/peer.yml") `
+    -PassThru -NoNewWindow `
+    -RedirectStandardOutput $peerOut `
+    -RedirectStandardError $peerErr
+
+function Wait-Until {
+    param([scriptblock]$Predicate, [int]$TimeoutSec, [string]$What)
+    $deadline = (Get-Date).AddSeconds($TimeoutSec)
+    while ((Get-Date) -lt $deadline) {
+        if (& $Predicate) { return }
+        Start-Sleep -Milliseconds 500
+    }
+    throw "timed out waiting for: $What"
+}
+
+try {
+    Wait-Until -TimeoutSec 30 -What "windows wintun adapter $DevName with NetworkCategory=Private" -Predicate {
+        if ($lhProc.HasExited) { throw "lighthouse exited (code $($lhProc.ExitCode)) before tun was ready" }
+        $p = Get-NetConnectionProfile -InterfaceAlias $DevName -ErrorAction SilentlyContinue
+        $p -and ("$($p.NetworkCategory)" -ieq 'Private')
+    }
+    Write-Host "OK: $DevName NetworkCategory=Private"
+
+    Wait-Until -TimeoutSec 30 -What "WSL nebula1 with $Ip2" -Predicate {
+        if ($peerProc.HasExited) { throw "peer exited (code $($peerProc.ExitCode)) before tun was ready" }
+        $r = wsl -d $Distro -u root -- bash -c "ip -o addr show nebula1 2>/dev/null | grep -q 'inet $Ip2' && echo yes"
+        ("$r").Trim() -eq 'yes'
+    }
+    Write-Host "OK: WSL nebula1 has $Ip2"
+
+    Wait-Until -TimeoutSec 30 -What "ping from WSL peer to windows lighthouse ($Ip1)" -Predicate {
+        if ($peerProc.HasExited) { throw "peer exited (code $($peerProc.ExitCode)) before ping succeeded" }
+        $r = wsl -d $Distro -u root -- bash -c "ping -c1 -W1 $Ip1 >/dev/null 2>&1 && echo OK"
+        ("$r").Trim() -eq 'OK'
+    }
+    Write-Host "OK: WSL peer -> windows lighthouse"
+
+    Wait-Until -TimeoutSec 30 -What "ping from windows lighthouse to WSL peer ($Ip2)" -Predicate {
+        $null = & ping.exe -n 1 -w 1000 $Ip2
+        $LASTEXITCODE -eq 0
+    }
+    Write-Host "OK: windows lighthouse -> WSL peer"
+
+    Write-Host ''
+    Write-Host 'All smoke checks passed.'
+}
+catch {
+    Write-Host ''
+    Write-Host '=== lighthouse stdout ==='
+    Get-Content $lhOut -ErrorAction SilentlyContinue | Out-Host
+    Write-Host '=== lighthouse stderr ==='
+    Get-Content $lhErr -ErrorAction SilentlyContinue | Out-Host
+    Write-Host '=== peer stdout ==='
+    Get-Content $peerOut -ErrorAction SilentlyContinue | Out-Host
+    Write-Host '=== peer stderr ==='
+    Get-Content $peerErr -ErrorAction SilentlyContinue | Out-Host
+    Write-Host '=== nebula WFP filters ==='
+    # Dump nebula-installed filters so we can verify they got registered with
+    # the conditions we expect.
+    $wfpDump = Join-Path $WorkDir 'wfp.xml'
+    netsh wfp show filters file=$wfpDump 2>&1 | Out-Null
+    if (Test-Path $wfpDump) {
+        Select-String -Path $wfpDump -Pattern 'Nebula' -Context 0,80 -ErrorAction SilentlyContinue | Out-Host
+    }
+    throw
+}
+finally {
+    if (-not $lhProc.HasExited) {
+        Stop-Process -Id $lhProc.Id -Force -ErrorAction SilentlyContinue
+        $lhProc.WaitForExit(5000) | Out-Null
+    }
+    wsl -d $Distro -u root -- bash -c "pkill -f $WslDir/nebula 2>/dev/null; true" | Out-Null
+    # pkill returns 1 when no match and wsl propagates that; the smoke is done
+    # so we don't want it to leak into the script's exit code.
+    $global:LASTEXITCODE = 0
+    if ($peerProc -and -not $peerProc.HasExited) {
+        Stop-Process -Id $peerProc.Id -Force -ErrorAction SilentlyContinue
+    }
+}
@@ -6,6 +6,8 @@ set -o pipefail

 mkdir -p logs

+NETWORK="nebula-smoke"
+
 cleanup() {
    echo
    echo " *** cleanup"
@@ -16,57 +18,103 @@ 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="$PREFIX.5"
+
+if [ "$SMOKE_OVERLAY_IPV6" ]
+then
+    LIGHTHOUSE_NIP="fd00:4242:0:0:0:ffff:c0a8:6401"
+    HOST2_NIP="fd00:4242:0:0:0:ffff:c0a8:6402"
+    HOST3_NIP="fd00:4242:0:0:0:ffff:c0a8:6403"
+    HOST4_NIP="fd00:4242:0:0:0:ffff:c0a8:6404"
+else
+    LIGHTHOUSE_NIP="192.168.100.1"
+    HOST2_NIP="192.168.100.2"
+    HOST3_NIP="192.168.100.3"
+    HOST4_NIP="192.168.100.4"
+fi
+
+# 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 host3 --rm "$CONTAINER" -config host3.yml -test
-docker run --name host4 --rm "$CONTAINER" -config host4.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 -v "$PWD/build/host3.yml:/nebula/host3.yml:ro" "$CONTAINER" -config host3.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 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 &
+docker exec host2 ncat -nklv 2000 &
+docker exec host3 ncat -nklv 2000 &
+docker exec host4 ncat -e '/usr/bin/echo helloagainfromhost4' -nkluv 4000 &
+docker exec host2 ncat -e '/usr/bin/echo host2' -nkluv 3000 &
+docker exec host3 ncat -e '/usr/bin/echo host3' -nkluv 3000 &

 set +x
 echo
 echo " *** Testing ping from lighthouse1"
 echo
 set -x
-docker exec lighthouse1 ping -c1 192.168.100.2
-docker exec lighthouse1 ping -c1 192.168.100.3
+docker exec lighthouse1 ping -c1 $HOST2_NIP
+docker exec lighthouse1 ping -c1 $HOST3_NIP

 set +x
 echo
 echo " *** Testing ping from host2"
 echo
 set -x
-docker exec host2 ping -c1 192.168.100.1
+docker exec host2 ping -c1 $LIGHTHOUSE_NIP
 # Should fail because not allowed by host3 inbound firewall
-! docker exec host2 ping -c1 192.168.100.3 -w5 || exit 1
+! docker exec host2 ping -c1 $HOST3_NIP -w5 || exit 1

 set +x
 echo
@@ -74,34 +122,34 @@ echo " *** Testing ncat from host2"
 echo
 set -x
 # Should fail because not allowed by host3 inbound firewall
-! docker exec host2 ncat -nzv -w5 192.168.100.3 2000 || exit 1
-! docker exec host2 ncat -nzuv -w5 192.168.100.3 3000 | grep -q host3 || exit 1
+! docker exec host2 ncat -nzv -w5 $HOST3_NIP 2000 || exit 1
+! docker exec host2 ncat -nzuv -w5 $HOST3_NIP 3000 | grep -q host3 || exit 1

 set +x
 echo
 echo " *** Testing ping from host3"
 echo
 set -x
-docker exec host3 ping -c1 192.168.100.1
-docker exec host3 ping -c1 192.168.100.2
+docker exec host3 ping -c1 $LIGHTHOUSE_NIP
+docker exec host3 ping -c1 $HOST2_NIP

 set +x
 echo
 echo " *** Testing ncat from host3"
 echo
 set -x
-docker exec host3 ncat -nzv -w5 192.168.100.2 2000
-docker exec host3 ncat -nzuv -w5 192.168.100.2 3000 | grep -q host2
+docker exec host3 ncat -nzv -w5 $HOST2_NIP 2000
+docker exec host3 ncat -nzuv -w5 $HOST2_NIP 3000 | grep -q host2

 set +x
 echo
 echo " *** Testing ping from host4"
 echo
 set -x
-docker exec host4 ping -c1 192.168.100.1
+docker exec host4 ping -c1 $LIGHTHOUSE_NIP
 # Should fail because not allowed by host4 outbound firewall
-! docker exec host4 ping -c1 192.168.100.2 -w5 || exit 1
-! docker exec host4 ping -c1 192.168.100.3 -w5 || exit 1
+! docker exec host4 ping -c1 $HOST2_NIP -w5 || exit 1
+! docker exec host4 ping -c1 $HOST3_NIP -w5 || exit 1

 set +x
 echo
@@ -109,27 +157,34 @@ echo " *** Testing ncat from host4"
 echo
 set -x
 # Should fail because not allowed by host4 outbound firewall
-! docker exec host4 ncat -nzv -w5 192.168.100.2 2000 || exit 1
-! docker exec host4 ncat -nzv -w5 192.168.100.3 2000 || exit 1
-! docker exec host4 ncat -nzuv -w5 192.168.100.2 3000 | grep -q host2 || exit 1
-! docker exec host4 ncat -nzuv -w5 192.168.100.3 3000 | grep -q host3 || exit 1
+! docker exec host4 ncat -nzv -w5 $HOST2_NIP 2000 || exit 1
+! docker exec host4 ncat -nzv -w5 $HOST3_NIP 2000 || exit 1
+! docker exec host4 ncat -nzuv -w5 $HOST2_NIP 3000 | grep -q host2 || exit 1
+! docker exec host4 ncat -nzuv -w5 $HOST3_NIP 3000 | grep -q host3 || exit 1

 set +x
 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 can ping host2 once conntrack established
-docker exec host2 ping -c1 192.168.100.4
-docker exec host4 ping -c1 192.168.100.2
+
+# host4's outbound firewall only allows ICMP to the lighthouse, so host4
+# cannot initiate UDP to host2. Once host2 initiates a flow to host4:4000,
+# conntrack must let host4's listener reply on that flow. If it doesn't,
+# the echo back from host4 never reaches host2.
+docker exec host2 sh -c "(/usr/bin/echo host2; sleep 2) | ncat -nuv $HOST4_NIP 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
@@ -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"

@@ -1,7 +1,7 @@
 # -*- mode: ruby -*-
 # vi: set ft=ruby :
 Vagrant.configure("2") do |config|
-  config.vm.box = "generic/netbsd9"
+  config.vm.box = "DefinedNet/netbsd10"

  config.vm.synced_folder "../build", "/nebula", type: "rsync"
 end
@@ -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
@@ -13,120 +13,138 @@ on:
      - 'go.sum'
 jobs:

-  test-linux:
-    name: Build all and test on ubuntu-linux
+  static:
+    name: Static checks
    runs-on: ubuntu-latest
    steps:

-    - uses: actions/checkout@v4
+    - uses: actions/checkout@v7

-    - uses: actions/setup-go@v6
+    - uses: actions/setup-go@v7
      with:
-        go-version: '1.25'
+        go-version: '1.26'
        check-latest: true

-    - name: Build
-      run: make all
+    - name: Install goimports
+      run: go install golang.org/x/tools/cmd/goimports@latest
+
+    - name: gofmt
+      run: |
+        if [ "$(find . -iname '*.go' | grep -v '\.pb\.go$' | xargs goimports -l)" ]
+        then
+          find . -iname '*.go' | grep -v '\.pb\.go$' | xargs goimports -d
+          exit 1
+        fi

    - name: Vet
      run: make vet

    - name: golangci-lint
-      uses: golangci/golangci-lint-action@v8
+      uses: golangci/golangci-lint-action@v9
      with:
-        version: v2.5
-
-    - name: Test
-      run: make test
-
-    - name: End 2 end
-      run: make e2evv
-
-    - name: Build test mobile
-      run: make build-test-mobile
-
-    - uses: actions/upload-artifact@v4
-      with:
-        name: e2e packet flow linux-latest
-        path: e2e/mermaid/linux-latest
-        if-no-files-found: warn
-
-  test-linux-boringcrypto:
-    name: Build and test on linux with boringcrypto
-    runs-on: ubuntu-latest
-    steps:
-
-    - uses: actions/checkout@v4
-
-    - uses: actions/setup-go@v6
-      with:
-        go-version: '1.25'
-        check-latest: true
-
-    - name: Build
-      run: make bin-boringcrypto
-
-    - name: Test
-      run: make test-boringcrypto
-
-    - name: End 2 end
-      run: make e2e GOEXPERIMENT=boringcrypto CGO_ENABLED=1 TEST_ENV="TEST_LOGS=1" TEST_FLAGS="-v -ldflags -checklinkname=0"
-
-  test-linux-pkcs11:
-    name: Build and test on linux with pkcs11
-    runs-on: ubuntu-latest
-    steps:
-
-    - uses: actions/checkout@v4
-
-    - uses: actions/setup-go@v6
-      with:
-        go-version: '1.25'
-        check-latest: true
-
-    - name: Build
-      run: make bin-pkcs11
-
-    - name: Test
-      run: make test-pkcs11
+        version: v2.12

  test:
-    name: Build and test on ${{ matrix.os }}
+    name: Test ${{ matrix.name }}
    runs-on: ${{ matrix.os }}
    strategy:
+      fail-fast: false
      matrix:
-        os: [windows-latest, macos-latest]
+        include:
+        - name: linux
+          os: ubuntu-latest
+          build-cmd: go build ./cmd/nebula ./cmd/nebula-cert
+          test-cmd: make test
+          e2e-cmd: make e2evv
+        - name: linux-boringcrypto
+          os: ubuntu-latest
+          build-cmd: make bin-boringcrypto
+          test-cmd: make test-boringcrypto
+          e2e-cmd: make e2e GOEXPERIMENT=boringcrypto CGO_ENABLED=1 TEST_ENV="TEST_LOGS=1" TEST_FLAGS="-v -ldflags -checklinkname=0"
+        - name: linux-pkcs11
+          os: ubuntu-latest
+          build-cmd: make bin-pkcs11
+          test-cmd: make test-pkcs11
+          e2e-cmd: ''
+        - name: macos
+          os: macos-latest
+          build-cmd: go build ./cmd/nebula ./cmd/nebula-cert
+          test-cmd: make test
+          e2e-cmd: make e2evv
+        - name: windows
+          os: windows-latest
+          build-cmd: go build ./cmd/nebula ./cmd/nebula-cert
+          test-cmd: make test
+          e2e-cmd: make e2evv
    steps:

-    - uses: actions/checkout@v4
+    - uses: actions/checkout@v7

-    - uses: actions/setup-go@v6
+    - uses: actions/setup-go@v7
      with:
-        go-version: '1.25'
+        go-version: '1.26'
        check-latest: true

-    - name: Build nebula
-      run: go build ./cmd/nebula
+    - name: Build
+      run: ${{ matrix.build-cmd }}

-    - name: Build nebula-cert
-      run: go build ./cmd/nebula-cert
-
-    - name: Vet
-      run: make vet
-
-    - name: golangci-lint
-      uses: golangci/golangci-lint-action@v8
-      with:
-        version: v2.5
+    - name: Cross-build darwin-amd64
+      if: matrix.name == 'macos'
+      run: GOARCH=amd64 go build -o /tmp/nebula-amd64 ./cmd/nebula && GOARCH=amd64 go build -o /tmp/nebula-cert-amd64 ./cmd/nebula-cert

    - name: Test
-      run: make test
+      run: ${{ matrix.test-cmd }}

    - name: End 2 end
-      run: make e2evv
+      if: matrix.e2e-cmd != ''
+      run: ${{ matrix.e2e-cmd }}

-    - uses: actions/upload-artifact@v4
+    - uses: actions/upload-artifact@v7
+      if: matrix.e2e-cmd != '' && always()
      with:
-        name: e2e packet flow ${{ matrix.os }}
-        path: e2e/mermaid/${{ matrix.os }}
+        name: e2e packet flow ${{ matrix.name }}
+        path: e2e/mermaid/
        if-no-files-found: warn
+
+  cross-build:
+    name: Cross-build ${{ matrix.name }}
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+        - {name: linux-arm,   make-target: all-cross-linux-arm}
+        - {name: linux-mips,  make-target: all-cross-linux-mips}
+        - {name: linux-other, make-target: all-cross-linux-other}
+        - {name: freebsd,     make-target: all-freebsd}
+        - {name: openbsd,     make-target: all-openbsd}
+        - {name: netbsd,      make-target: all-netbsd}
+        - {name: windows,     make-target: all-cross-windows}
+        - {name: mobile,      make-target: build-test-mobile}
+    steps:
+
+    - uses: actions/checkout@v7
+
+    - uses: actions/setup-go@v7
+      with:
+        go-version: '1.26'
+        check-latest: true
+
+    - name: Build ${{ matrix.name }}
+      run: make -j"$(nproc)" ${{ matrix.make-target }}
+
+  finish:
+    name: CI status
+    if: always()
+    needs: [static, test, cross-build]
+    runs-on: ubuntu-latest
+    steps:
+
+    - name: Fail if any upstream job failed
+      if: contains(needs.*.result, 'failure') || contains(needs.*.result, 'cancelled')
+      run: |
+        echo "upstream results: ${{ toJSON(needs) }}"
+        exit 1
+
+    - name: All upstream jobs passed
+      run: echo "ok"
@@ -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:
@@ -7,12 +7,211 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0

 ## [Unreleased]

+## [1.11.0] - 2026-07-23
+
+See the [v1.11.0](https://github.com/slackhq/nebula/milestone/25?closed=1) milestone for a complete list of changes.
+
+### Breaking
+
+- Logging has switched from logrus to Go's structured `slog`. Log output changes: levels are upper case
+  (`level=INFO`), trace prints as `level=DEBUG-4`, timestamps are always RFC3339Nano and `logging.timestamp_format`
+  is ignored, and some messages were reworded. Review any log parsing before upgrading. This is also an API break
+  for embedders, as constructors now take a `*slog.Logger`. (#1672, #1734, #1621)
+- `firewall.inbound_action` and `firewall.outbound_action` (used to set reject vs. drop policy) were each being
+  applied to the opposite direction, that is now corrected. This only affects how blocked packets are answered, not
+  which packets the firewall allows or denies. If you set either of these you are getting the behavior of the other
+  one today and likely want to swap them before upgrading. (#1798)
+- On Windows, Nebula now installs WFP PERMIT filters for the nebula adapter and the listener port by default. WFP
+  sits below Windows Defender Firewall, so any WDF inbound rules you rely on for either will no longer apply. Set
+  `tun.windows_bypass_wdf` and `listen.windows_bypass_wdf` to false to leave WDF in charge. (#1710)
+- On Windows, the nebula device is now set to the `private` network category instead of whatever Windows decided,
+  which is usually `Public`. This makes the host firewall less restrictive on the overlay. Set
+  `tun.network_category` to `unset` to keep the old behavior. (#1710)
+- Reject packets for non-TCP now use ICMP code 13, communication administratively prohibited, instead of code 3,
+  port unreachable. Anything keying off the old code needs updating. (#1766, #1768)
+- The SSH debug server's profiling commands are now confined to `sshd.sandbox_dir`, which defaults to
+  `$TMP/nebula-debug`. Relative paths resolve inside it and absolute paths outside it are rejected, so anything
+  scripting `start-cpu-profile`, `save-heap-profile`, or `save-mutex-profile` with a path elsewhere needs the
+  directory set. The directory is not created for you. (#1622)
+
+### Added
+
+- Sign the Windows release binaries. (#1718)
+- Generate IPv6 reject packets, matching the existing IPv4 behavior. (#1766, #1767, #1768)
+- Accept `-` in `nebula-cert` to read from stdin or write to stdout. (#1714)
+- Search for both `config.yml` and `config.yaml` in service and command line modes. (#1717)
+- Add version labels to the Docker/OCI images. (#1772)
+- Rebind the listener and re-query lighthouses on macOS when the underlay network changes, so devices moving
+  between wifi and wired or between networks recover without waiting for dead tunnel detection. Controlled by
+  `listen.rebind_on_network_change` (default `true`, not reloadable). (#1816)
+
 ### Changed

- `default_local_cidr_any` now defaults to false, meaning that any firewall rule
+- Reload the firewall when the unsafe networks in the certificate change. (#1719)
+- Reconfigure, start, and stop the stats listener on a config reload instead of requiring a restart. (#1670)
+- Update a static host's addresses when they change on reload. (#1713)
+- Don't require a port on ICMP firewall rules. (#1609)
+- Connection track ICMP traffic. (#1602)
+- Return `NODATA` instead of `NXDOMAIN` from the DNS server for a name that exists but has no record of the
+  requested type, so clients that query `AAAA` first (busybox/Alpine) fall through to `A`. (#1668)
+- Record the local host's details in the DNS server. (#1716)
+- Install Windows unsafe routes as link routes. (#1709)
+- Reduce relay handshake log spam, and only log a handshake send error at error level when the remote list
+  changes. (#1733, #1765, #1810)
+- Start, stop, and reload subsystems (DNS, stats, conntrack, ssh, punchy) cleanly without leaking goroutines. (#1640, #1654, #1661, #1667, #1669, #1708, #1806, #1815)
+- `Control` is now safe to stop and wait on from any lifecycle state, and a new `Control.Wait` blocks until nebula
+  has fully stopped and returns the first fatal reader error. Failed starts release the udp sockets and tun fd
+  instead of leaking them. (#1794)
+- Trigger an immediate lighthouse update when reconnecting to or adding a lighthouse instead of waiting for the next update tick. (#1645)
+- Bring the Darwin and OpenBSD tun implementations in line with the other BSDs. (#1703)
+- Update to build against go v1.26. (#1818)
+- Various dependency updates. (#1586, #1587, #1604, #1617, #1618, #1627, #1628, #1629, #1652, #1664, #1665, #1697, #1721, #1732, #1742, #1743, #1750, #1763, #1771, #1782, #1800, #1807)
+
+### Fixed
+
+- Fix a data race on a host's remote address that could send packets to the wrong address during a roam. (#1773)
+- Fix tunnels that could permanently escape connection manager monitoring. (#1752)
+- Fix a crash when reloading the SSH server's trusted keys. (#1787)
+- Fix hostmap corruption when a host has multiple overlay addresses. Each address now gets its own list instead of
+  a single shared chain, which also fixes two latent bugs on the add and makePrimary paths. (#1788, #1790)
+- Apply `remote_allow_list` IPv4 rules to 4-in-6 mapped addresses. (#1786)
+- Don't panic in the DNS server on a short or empty query name. (#1635)
+- Advance the replay window on relayed packets so a relay drops replayed frames instead of re-forwarding them. (#1751)
+- Fix a race in relay state handling. (#1753)
+- Lock replay window updates so concurrent readers can't corrupt it. (#1802)
+- Reject malformed handshakes more reliably, including invalid ed25519 key lengths. (#1601, #1756)
+- Properly handle `closetunnel` packets. (#1638)
+- Fix an IPv6 extension-header length overflow that could make the firewall parse the wrong protocol and ports. (#1789)
+- Fix relay re-establishment when a handshake arrives over a relay entry that a one-sided teardown left
+  `Disestablished`, which silently dropped every send until dead tunnel detection forced a re-handshake. (#1805)
+- Don't build new relay state on a tunnel that was just discarded. (#1796)
+- Don't delete the wrong pending hostinfo in the handshake manager. (#1811)
+- Don't call the packet reader after a UDP error on Darwin. (#1755)
+- Open the FreeBSD tun device non blocking. (#1666)
+
+## [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
+
+- 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 +870,14 @@ 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
@@ -0,0 +1 @@
+#ECCN:Open Source
@@ -60,6 +60,18 @@ ALL = $(ALL_LINUX) \
 	windows-amd64 \
 	windows-arm64

+# Cross-build shards used by .github/workflows/test.yml — same as ALL_*
+# but with the arch that has a native CI runner removed, so the cross-build
+# job is not duplicating coverage the native test jobs already give.
+ALL_CROSS_LINUX = $(filter-out linux-amd64,$(ALL_LINUX))
+
+# ALL_CROSS_LINUX further split into family sub-shards so each can run on
+# its own CI runner in parallel. Union of the three must equal
+# ALL_CROSS_LINUX; adding a new linux arch goes into the matching family.
+ALL_CROSS_LINUX_ARM   = linux-arm-5 linux-arm-6 linux-arm-7 linux-arm64
+ALL_CROSS_LINUX_MIPS  = linux-mips linux-mipsle linux-mips64 linux-mips64le linux-mips-softfloat
+ALL_CROSS_LINUX_OTHER = linux-386 linux-ppc64le linux-riscv64 linux-loong64
+
 e2e:
 	$(TEST_ENV) go test -tags=e2e_testing -count=1 $(TEST_FLAGS) ./e2e

@@ -82,6 +94,35 @@ DOCKER_BIN = build/linux-amd64/nebula build/linux-amd64/nebula-cert

 all: $(ALL:%=build/%/nebula) $(ALL:%=build/%/nebula-cert)

+all-linux: $(ALL_LINUX:%=build/%/nebula) $(ALL_LINUX:%=build/%/nebula-cert)
+
+all-freebsd: $(ALL_FREEBSD:%=build/%/nebula) $(ALL_FREEBSD:%=build/%/nebula-cert)
+
+all-openbsd: $(ALL_OPENBSD:%=build/%/nebula) $(ALL_OPENBSD:%=build/%/nebula-cert)
+
+all-netbsd: $(ALL_NETBSD:%=build/%/nebula) $(ALL_NETBSD:%=build/%/nebula-cert)
+
+all-darwin: build/darwin-amd64/nebula build/darwin-amd64/nebula-cert build/darwin-arm64/nebula build/darwin-arm64/nebula-cert
+
+all-windows: build/windows-amd64/nebula.exe build/windows-amd64/nebula-cert.exe build/windows-arm64/nebula.exe build/windows-arm64/nebula-cert.exe
+
+# CI cross-build shards. darwin-arm64 is covered by the native macos-latest
+# job; windows-amd64 is covered by the native windows-latest job; both are
+# omitted here to avoid building them a second time. darwin-amd64 stays in
+# all-cross-darwin because intel mac is only a labeled/master-time native
+# job, so PRs still need cross-build coverage for it.
+all-cross-linux: $(ALL_CROSS_LINUX:%=build/%/nebula) $(ALL_CROSS_LINUX:%=build/%/nebula-cert)
+
+all-cross-linux-arm:   $(ALL_CROSS_LINUX_ARM:%=build/%/nebula)   $(ALL_CROSS_LINUX_ARM:%=build/%/nebula-cert)
+
+all-cross-linux-mips:  $(ALL_CROSS_LINUX_MIPS:%=build/%/nebula)  $(ALL_CROSS_LINUX_MIPS:%=build/%/nebula-cert)
+
+all-cross-linux-other: $(ALL_CROSS_LINUX_OTHER:%=build/%/nebula) $(ALL_CROSS_LINUX_OTHER:%=build/%/nebula-cert)
+
+all-cross-darwin: build/darwin-amd64/nebula build/darwin-amd64/nebula-cert
+
+all-cross-windows: build/windows-arm64/nebula.exe build/windows-arm64/nebula-cert.exe
+
 docker: docker/linux-$(shell go env GOARCH)

 release: $(ALL:%=build/nebula-%.tar.gz)
@@ -227,6 +268,9 @@ smoke-relay-docker: bin-docker
 	cd .github/workflows/smoke/ && ./build-relay.sh
 	cd .github/workflows/smoke/ && ./smoke-relay.sh

+smoke-docker-ipv6: export SMOKE_OVERLAY_IPV6 = 1
+smoke-docker-ipv6: smoke-docker
+
 smoke-docker-race: BUILD_ARGS = -race
 smoke-docker-race: CGO_ENABLED = 1
 smoke-docker-race: smoke-docker
@@ -236,5 +280,5 @@ smoke-vagrant/%: bin-docker build/%/nebula
 	cd .github/workflows/smoke/ && ./smoke-vagrant.sh $*

 .FORCE:
-.PHONY: bench bench-cpu bench-cpu-long bin build-test-mobile e2e e2ev e2evv e2evvv e2evvvv proto release service smoke-docker smoke-docker-race test test-cov-html smoke-vagrant/%
+.PHONY: all all-linux all-freebsd all-openbsd all-netbsd all-darwin all-windows all-cross-linux all-cross-linux-arm all-cross-linux-mips all-cross-linux-other all-cross-darwin all-cross-windows bench bench-cpu bench-cpu-long bin build-test-mobile e2e e2ev e2evv e2evvv e2evvvv proto release service smoke-docker smoke-docker-race test test-cov-html smoke-vagrant/%
 .DEFAULT_GOAL := bin
@@ -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&amp;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.
@@ -1,157 +1,263 @@
 package nebula

 import (
+	"context"
+	"fmt"
+	"log/slog"
+	"math"
+	mathbits "math/bits"
+
 	"github.com/rcrowley/go-metrics"
-	"github.com/sirupsen/logrus"
 )

+const bitsPerWord = 64
+
+// Bits is a sliding-window anti-replay tracker. The window is stored as a
+// circular bitmap packed into uint64 words (8x denser than a []bool), so a
+// length-N window costs N/8 bytes. length must be a power of two.
 type Bits struct {
 	length             uint64
+	lengthMask         uint64
 	current            uint64
-	bits               []bool
-	firstSeen          bool
+	bits               []uint64
 	lostCounter        metrics.Counter
 	dupeCounter        metrics.Counter
 	outOfWindowCounter metrics.Counter
 }

-func NewBits(bits uint64) *Bits {
-	return &Bits{
-		length:             bits,
-		bits:               make([]bool, bits, bits),
+func NewBits(length uint64) *Bits {
+	if length == 0 || length&(length-1) != 0 {
+		panic(fmt.Sprintf("Bits length must be a power of two, got %d", length))
+	}
+
+	nWords := length / bitsPerWord
+	if nWords == 0 {
+		nWords = 1
+	}
+	b := &Bits{
+		length:             length,
+		lengthMask:         length - 1,
+		bits:               make([]uint64, nWords),
 		current:            0,
 		lostCounter:        metrics.GetOrRegisterCounter("network.packets.lost", nil),
 		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] = 1
+	return b
 }

-func (b *Bits) Check(l logrus.FieldLogger, i uint64) bool {
-	// If i is the next number, return true.
-	if i > b.current || (i == 0 && b.firstSeen == false && b.current < b.length) {
+func (b *Bits) get(i uint64) bool {
+	pos := i & b.lengthMask
+	//bit-shifting by 6 because i is a bit index, not a u64 index, and we need to find the u64 without bit in it
+	return b.bits[pos>>6]&(uint64(1)<<(pos&63)) != 0
+}
+
+func (b *Bits) set(i uint64) {
+	pos := i & b.lengthMask
+	b.bits[pos>>6] |= uint64(1) << (pos & 63)
+}
+
+// clearRange clears `count` bits starting at circular position `startPos`
+// (already masked to [0, length)) and returns how many of them were set
+// before the clear. count must be in [1, length].
+func (b *Bits) clearRange(startPos, count uint64) uint64 {
+	wasSet := uint64(0)
+	if count >= b.length {
+		for _, w := range b.bits {
+			wasSet += uint64(mathbits.OnesCount64(w))
+		}
+		clear(b.bits)
+		return wasSet
+	}
+
+	pos := startPos
+	remaining := count
+
+	// handle the potential partial word before pos becomes u64 aligned
+	word := pos >> 6
+	bit := pos & 63
+	take := uint64(64) - bit
+	if take > remaining {
+		take = remaining
+	}
+	if take > b.length-pos {
+		take = b.length - pos
+	}
+	var mask uint64
+	if take == 64 {
+		mask = math.MaxUint64
+	} else {
+		mask = ((uint64(1) << take) - 1) << bit
+	}
+	wasSet += uint64(mathbits.OnesCount64(b.bits[word] & mask))
+	b.bits[word] &^= mask
+	remaining -= take
+	pos = (pos + take) & b.lengthMask
+
+	// Clear whole words, keeping track of the number of set bits
+	for remaining >= 64 {
+		word = pos >> 6
+		wasSet += uint64(mathbits.OnesCount64(b.bits[word]))
+		b.bits[word] = 0
+		remaining -= 64
+		pos = (pos + 64) & b.lengthMask
+	}
+
+	// Clear the remaining partial word
+	if remaining > 0 {
+		word = pos >> 6
+		mask = (uint64(1) << remaining) - 1
+		wasSet += uint64(mathbits.OnesCount64(b.bits[word] & mask))
+		b.bits[word] &^= mask
+	}
+
+	return wasSet
+}
+
+func (b *Bits) strictlyWithinWindow(i uint64) bool {
+	// Handle the case where the window hasn't slid yet. This avoids u64 underflow.
+	inWarmup := b.current < b.length
+	if i < b.length && inWarmup {
 		return true
 	}

-	// If i is within the window, check if it's been set already. The first window will fail this check
+	// Next, if the packet is in-window, see if we've seen it before
 	if i > b.current-b.length {
-		return !b.bits[i%b.length]
+		return true
+	}
+	return false //not within window!
+}
+
+// Check returns true if i is within (or way out in front of) the window, and not a replay
+func (b *Bits) Check(l *slog.Logger, i uint64) bool {
+	// If i is the next number, return true.
+	if i > b.current {
+		return true
 	}

-	// If i is within the first window
-	if i < b.length {
-		return !b.bits[i%b.length]
+	if b.strictlyWithinWindow(i) {
+		return !b.get(i)
 	}

 	// 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 {
-	// If i is the next number, return true and update current.
+// Update has three branches:
+//   - i == b.current+1: fast path; advance the cursor by one and lose-count
+//     the slot we just stomped (only past warmup; see the i > b.length guard
+//     below).
+//   - i  >  b.current+1: jump path; clear all slots between current and i
+//     (or up to a full window's worth, whichever is smaller) via clearRange,
+//     then mark i. Two arms here: a warmup arm that handles the very first
+//     window before the cursor has slid, and a steady-state arm that treats
+//     every cleared empty slot as a lost packet.
+//   - i  <= b.current: in-window check for duplicates; out-of-window otherwise.
+//
+// NewBits seeds bits[0]=1 so counter 0 looks "received" — Update never
+// clears that marker during warmup (clearRange skips position 0 when
+// startPos=1), and once b.current >= b.length the marker is no longer
+// consulted. The marker prevents a fictitious "lost" hit on the first real
+// counter.
+func (b *Bits) Update(l *slog.Logger, i uint64) bool {
+	// Fast path: i is the next expected counter. Split out so the function
+	// stays small and avoids paying for the slow paths' slog argument-build
+	// stack frame on every call. The bit read/test/write is inlined to
+	// touch the backing word once.
 	if i == b.current+1 {
-		// Report missed packets, we can only understand what was missed after the first window has been gone through
-		if i > b.length && b.bits[i%b.length] == false {
+		pos := i & b.lengthMask
+		word := pos >> 6
+		mask := uint64(1) << (pos & 63)
+		w := b.bits[word]
+		if i > b.length && w&mask == 0 {
 			b.lostCounter.Inc(1)
 		}
-		b.bits[i%b.length] = true
+		b.bits[word] = w | mask
 		b.current = i
 		return true
 	}
+	return b.updateSlow(l, i)
+}

-	// If i packet is greater than current but less than the maximum length of our bitmap,
-	// flip everything in between to false and move ahead.
-	if i > b.current && i < b.current+b.length {
-		// In between current and i need to be zero'd to allow those packets to come in later
-		for n := b.current + 1; n < i; n++ {
-			b.bits[n%b.length] = false
+// updateSlow handles jumps, in-window backfill, dupes, and out-of-window.
+func (b *Bits) updateSlow(l *slog.Logger, i uint64) bool {
+	// If i is a jump, adjust the window, record lost, update current, and return true
+	if i > b.current {
+		end := i
+		if end > b.current+b.length {
+			end = b.current + b.length
+		}
+		count := end - b.current
+		startPos := (b.current + 1) & b.lengthMask
+
+		var lost int64
+		if b.current >= b.length {
+			// Steady state: every cleared slot is past warmup, so any unset
+			// bit we evict is a lost packet from the previous cycle.
+			wasSet := b.clearRange(startPos, count)
+			lost = int64(count) - int64(wasSet)
+		} else {
+			// Warmup (the very first window). Some cleared slots represent
+			// packets <= length where eviction is not "lost" in the usual
+			// sense. This branch is taken at most once per connection so we
+			// don't bother optimizing it.
+			for n := b.current + 1; n <= end; n++ {
+				if !b.get(n) && n > b.length {
+					lost++
+				}
+			}
+			b.clearRange(startPos, count)
 		}

-		b.bits[i%b.length] = true
-		b.current = i
-		//l.Debugf("missed %d packets between %d and %d\n", i-b.current, i, b.current)
-		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++
+		// Anything past the new window can never be backfilled, so it's lost.
+		if i > b.current+b.length {
+			lost += int64(i - b.current - b.length)
 		}
-
-		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.set(i)
 		b.current = i
 		return true
 	}

-	// Allow for the 0 packet to come in within the first window
-	if i == 0 && b.firstSeen == false && b.current < b.length {
-		b.firstSeen = true
-		b.bits[i%b.length] = true
-		return true
-	}
-
-	// If i is within the window of current minus length (the total pat window size),
-	// allow it and flip to true but to NOT change current. We also have to account for the first window
-	if ((b.current >= b.length && i > b.current-b.length) || (b.current < b.length && i < b.length)) && i <= b.current {
-		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")
+	// If i is within the current window but below the current counter, check to see if it's a duplicate
+	if b.strictlyWithinWindow(i) {
+		pos := i & b.lengthMask
+		word := pos >> 6
+		mask := uint64(1) << (pos & 63)
+		w := b.bits[word]
+		if b.current == i || w&mask != 0 {
+			if l.Enabled(context.Background(), slog.LevelDebug) {
+				l.Debug("Receive window",
+					"accepted", false,
+					"currentCounter", b.current,
+					"incomingCounter", i,
+					"reason", "duplicate",
+				)
 			}
 			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
-		}
-
-		b.bits[i%b.length] = true
+		b.bits[word] = w | mask
 		return true
-
 	}

 	// In all other cases, fail and don't change current.
 	b.outOfWindowCounter.Inc(1)
-	if l.Level >= logrus.DebugLevel {
-		l.WithField("accepted", false).
-			WithField("currentCounter", b.current).
-			WithField("incomingCounter", i).
-			WithField("reason", "nonsense").
-			Debug("Receive window")
+	if l.Enabled(context.Background(), slog.LevelDebug) {
+		l.Debug("Receive window",
+			"accepted", false,
+			"currentCounter", b.current,
+			"incomingCounter", i,
+			"reason", "nonsense",
+		)
 	}
 	return false
 }
-
-func maxInt64(a, b int64) int64 {
-	if a > b {
-		return a
-	}
-
-	return b
-}
@@ -7,77 +7,114 @@ import (
 	"github.com/stretchr/testify/assert"
 )

+// snapshot returns the bitmap as a []bool of length b.length, for readable
+// test assertions against the now-packed []uint64 storage.
+func (b *Bits) snapshot() []bool {
+	out := make([]bool, b.length)
+	for i := uint64(0); i < b.length; i++ {
+		out[i] = b.get(i)
+	}
+	return out
+}
+
+func TestBitsRequiresPowerOfTwo(t *testing.T) {
+	assert.Panics(t, func() { NewBits(10) })
+	assert.Panics(t, func() { NewBits(0) })
+	assert.NotPanics(t, func() { NewBits(1) })
+	assert.NotPanics(t, func() { NewBits(16) })
+	assert.NotPanics(t, func() { NewBits(1024) })
+	assert.NotPanics(t, func() { NewBits(16384) })
+}
+
 func TestBits(t *testing.T) {
 	l := test.NewLogger()
-	b := NewBits(10)
-
-	// make sure it is the right size
-	assert.Len(t, b.bits, 10)
+	b := NewBits(16)
+	assert.EqualValues(t, 16, b.length)

 	// 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, u)
+	assert.True(t, b.Update(l, 1))
 	assert.EqualValues(t, 1, b.current)
-	g := []bool{false, true, false, false, false, false, false, false, false, false}
-	assert.Equal(t, g, b.bits)
+	g := []bool{true, true, false, false, false, false, false, false, false, false, false, false, false, false, false, false}
+	assert.Equal(t, g, b.snapshot())

 	// Receive two
 	assert.True(t, b.Check(l, 2))
-	u = b.Update(l, 2)
-	assert.True(t, u)
+	assert.True(t, b.Update(l, 2))
 	assert.EqualValues(t, 2, b.current)
-	g = []bool{false, true, true, false, false, false, false, false, false, false}
-	assert.Equal(t, g, b.bits)
+	g = []bool{true, true, true, false, false, false, false, false, false, false, false, false, false, false, false, false}
+	assert.Equal(t, g, b.snapshot())

 	// Receive two again - it will fail
 	assert.False(t, b.Check(l, 2))
-	u = b.Update(l, 2)
-	assert.False(t, u)
+	assert.False(t, b.Update(l, 2))
 	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, 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)
+	// Jump ahead to 25, which clears the window and sets slot 25%16 = 9.
+	assert.True(t, b.Check(l, 25))
+	assert.True(t, b.Update(l, 25))
+	assert.EqualValues(t, 25, b.current)
+	g = []bool{false, false, false, false, false, false, false, false, false, true, false, false, false, false, false, false}
+	assert.Equal(t, g, b.snapshot())

-	// 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, 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 24, which is in window (current 25, length 16, window covers [10,25]).
+	assert.True(t, b.Check(l, 24))
+	assert.True(t, b.Update(l, 24))
+	assert.EqualValues(t, 25, b.current)
+	g = []bool{false, false, false, false, false, false, false, false, true, true, false, false, false, false, false, false}
+	assert.Equal(t, g, b.snapshot())

-	// Mark 5, which is not allowed because it is not in the window
+	// Mark 5, not allowed because 5 <= current-length (25-16=9).
 	assert.False(t, b.Check(l, 5))
-	u = 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)
+	assert.False(t, b.Update(l, 5))
+	assert.EqualValues(t, 25, b.current)
+	g = []bool{false, false, false, false, false, false, false, false, true, true, false, false, false, false, false, false}
+	assert.Equal(t, g, b.snapshot())

-	// make sure we handle wrapping around once to the current position
-	b = NewBits(10)
+	// Make sure we handle wrapping around once to the same slot. With
+	// length=16, packets 1 and 17 share slot 1.
+	b = NewBits(16)
 	assert.True(t, b.Update(l, 1))
-	assert.True(t, b.Update(l, 11))
-	assert.Equal(t, []bool{false, true, false, false, false, false, false, false, false, false}, b.bits)
+	assert.True(t, b.Update(l, 17))
+	assert.Equal(t, []bool{false, true, false, false, false, false, false, false, false, false, false, false, false, false, false, false}, b.snapshot())

 	// Walk through a few windows in order
-	b = NewBits(10)
-	for i := uint64(0); i <= 100; i++ {
+	b = NewBits(16)
+	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()
+
+	// length=16. Update(55) from current=0:
+	//   warmup, per-bit loop sees no n>16 with unset bits (slot 0 was set by
+	//   NewBits and gets re-evaluated when n=16; n=16 is not strictly > 16),
+	//   so the loop contributes 0. The jump exceeds the window so we record
+	//   55 - 0 - 16 = 39 packets fell out the back.
+	b := NewBits(16)
+	b.lostCounter.Clear()
+	assert.True(t, b.Update(l, 55))
+	assert.Equal(t, int64(39), b.lostCounter.Count())
+
+	// Update(100): clears 16 slots starting at slot 56%16=8. Only slot 7 (for
+	// packet 55) was set, so 16 - 1 = 15 evicted slots had unset bits.
+	// Plus 100 - 55 - 16 = 29 packets fell past the window. Total 44.
+	assert.True(t, b.Update(l, 100))
+	assert.Equal(t, int64(39+44), b.lostCounter.Count())
+
+	// Update(200): same shape: 16 - 1 = 15 evicted unset, plus 200 - 100 - 16 = 84 past window. Total 99.
+	assert.True(t, b.Update(l, 200))
+	assert.Equal(t, int64(39+44+99), b.lostCounter.Count())
 }

 func TestBitsDupeCounter(t *testing.T) {
 	l := test.NewLogger()
-	b := NewBits(10)
+	b := NewBits(16)
 	b.lostCounter.Clear()
 	b.dupeCounter.Clear()
 	b.outOfWindowCounter.Clear()
@@ -102,127 +139,300 @@ func TestBitsDupeCounter(t *testing.T) {

 func TestBitsOutOfWindowCounter(t *testing.T) {
 	l := test.NewLogger()
-	b := NewBits(10)
+	b := NewBits(16)
 	b.lostCounter.Clear()
 	b.dupeCounter.Clear()
 	b.outOfWindowCounter.Clear()

+	// Jump to 20 (warmup branch + 4 past-window packets).
 	assert.True(t, b.Update(l, 20))
 	assert.Equal(t, int64(0), b.outOfWindowCounter.Count())

-	assert.True(t, b.Update(l, 21))
-	assert.True(t, b.Update(l, 22))
-	assert.True(t, b.Update(l, 23))
-	assert.True(t, b.Update(l, 24))
-	assert.True(t, b.Update(l, 25))
-	assert.True(t, b.Update(l, 26))
-	assert.True(t, b.Update(l, 27))
-	assert.True(t, b.Update(l, 28))
-	assert.True(t, b.Update(l, 29))
+	// 9 single-step advances, each evicts a slot whose bit was cleared during
+	// the jump above and whose value was never seen, so each contributes 1
+	// to lostCounter.
+	for n := uint64(21); n <= 29; n++ {
+		assert.True(t, b.Update(l, n))
+	}
 	assert.Equal(t, int64(0), b.outOfWindowCounter.Count())

+	// 0 is below current-length (29-16=13) so it falls outside the window.
 	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(20), b.lostCounter.Count())
+	// 4 from the Update(20) jump + 9 from 21..29.
+	assert.Equal(t, int64(13), b.lostCounter.Count())
 	assert.Equal(t, int64(0), b.dupeCounter.Count())
 	assert.Equal(t, int64(1), b.outOfWindowCounter.Count())
 }

 func TestBitsLostCounter(t *testing.T) {
 	l := test.NewLogger()
-	b := NewBits(10)
+	b := NewBits(16)
 	b.lostCounter.Clear()
 	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))
-	assert.True(t, b.Update(l, 23))
-	assert.True(t, b.Update(l, 24))
-	assert.True(t, b.Update(l, 25))
-	assert.True(t, b.Update(l, 26))
-	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())
+	// Walk 20..29 like the original, just with a bigger window. Same
+	// reasoning as TestBitsOutOfWindowCounter: 4 past-window from Update(20),
+	// then 9 more from the unit advances.
+	for n := uint64(20); n <= 29; n++ {
+		assert.True(t, b.Update(l, n))
+	}
+	assert.Equal(t, int64(13), b.lostCounter.Count())
 	assert.Equal(t, int64(0), b.dupeCounter.Count())
 	assert.Equal(t, int64(0), b.outOfWindowCounter.Count())

-	b = NewBits(10)
+	b = NewBits(16)
 	b.lostCounter.Clear()
 	b.dupeCounter.Clear()
 	b.outOfWindowCounter.Clear()

-	assert.True(t, b.Update(l, 0))
+	// Update(15) clears the warmup window (no lost), sets slot 15.
+	assert.True(t, b.Update(l, 15))
+	assert.Equal(t, int64(0), b.lostCounter.Count())
+
+	// Update(16): slot 0 was already set (NewBits seeded it), and 16 is not
+	// strictly > length, so nothing is recorded as lost.
+	assert.True(t, b.Update(l, 16))
+	assert.Equal(t, int64(0), b.lostCounter.Count())
+
+	// Update(17): we jumped straight from 0 to 15, so slot 1 was cleared
+	// (and never re-set). 17 > 16 is past warmup, so packet 1 is recorded lost.
+	assert.True(t, b.Update(l, 17))
+	assert.Equal(t, int64(1), b.lostCounter.Count())
+
+	// Fill in 18..30 in single steps. Each i evicts slot i%16. Slots 2..14
+	// were all cleared during Update(15), and we never re-set any of them,
+	// so each i in 18..30 is a fresh lost packet — 13 more.
+	for n := uint64(18); n <= 30; n++ {
+		assert.True(t, b.Update(l, n))
+	}
+	assert.Equal(t, int64(14), b.lostCounter.Count())
+
+	// Jump ahead by exactly one window size.
+	assert.True(t, b.Update(l, 46))
+	// end = min(46, 30+16) = 46, count = 16, all slots cleared. Before the
+	// jump every slot 0..15 had been set (Update(15), (16), (17), 18..30),
+	// so wasSet=16 and 46 == current+length means no past-window slack:
+	// lost contribution = 0.
+	assert.Equal(t, int64(14), b.lostCounter.Count())
+
+	// Walk 47..55. The Update(46) jump cleared every slot, so only slot 14
+	// (for packet 46) is set when we start. Each subsequent unit step lands
+	// on a slot that was cleared and is past warmup, so it counts as lost.
+	// 9 more = 23.
+	for n := uint64(47); n <= 55; n++ {
+		assert.True(t, b.Update(l, n))
+	}
+	assert.Equal(t, int64(23), b.lostCounter.Count())
+
+	// Jump ahead by two windows: clears the window plus past-window loss.
+	assert.True(t, b.Update(l, 87))
+	// current=55, length=16. end = min(87, 71) = 71. count=16, all slots
+	// cleared. Slots set before the clear are slots 14,15,0..7 (10 total).
+	// Lost from clear = 16 - 10 = 6. Past window: 87 - 55 - 16 = 16. +22.
+	assert.Equal(t, int64(45), b.lostCounter.Count())
+	assert.Equal(t, int64(0), b.dupeCounter.Count())
+	assert.Equal(t, int64(0), b.outOfWindowCounter.Count())
+}
+
+func TestBitsLostCounterIssue1(t *testing.T) {
+	l := test.NewLogger()
+	b := NewBits(16)
+	b.lostCounter.Clear()
+	b.dupeCounter.Clear()
+	b.outOfWindowCounter.Clear()
+
+	// Receive 4, backfill 1, then 9, 2, 3, 5, 6, 7 (skip 8), 10, 11, 14.
+	// Then jump to 25 — slot 25%16=9 is being evicted, but it had been set
+	// (we received packet 9), so no spurious lost increment. The original
+	// regression was about double-counting a missing packet when its slot
+	// got cleared on a jump. With the jump path now using clearRange's
+	// word-level wasSet count, the same semantics hold.
+	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())
-	// 10 will set 0 index, 0 was already set, no lost packets
+	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())
+	// Skip packet 8.
 	assert.True(t, b.Update(l, 10))
 	assert.Equal(t, int64(0), b.lostCounter.Count())
-	// 11 will set 1 index, 1 was missed, we should see 1 packet lost
 	assert.True(t, b.Update(l, 11))
-	assert.Equal(t, int64(1), b.lostCounter.Count())
-	// Now let's fill in the window, should end up with 8 lost packets
-	assert.True(t, b.Update(l, 12))
-	assert.True(t, b.Update(l, 13))
+	assert.Equal(t, int64(0), b.lostCounter.Count())
+
 	assert.True(t, b.Update(l, 14))
+	assert.Equal(t, int64(0), b.lostCounter.Count())
+
+	// Jump to 25. With length=16, slot 25%16=9 corresponds to packet 9
+	// (which we DID receive), so its bit is set and no lost++ from that
+	// eviction. The trace below shows the only loss is packet 8.
+	assert.True(t, b.Update(l, 25))
+	// current was 14, i=25. end=min(25,30)=25. count=11. startPos=15.
+	// steady? current=14<16, so warmup branch: per-bit n=15..25, count those
+	// with !get(n) AND n>16. n=17..25 are >16. Among slots 17%16=1..25%16=9
+	// did we set slots 1..9 (packets 1..9)? Yes for all but slot 8 (packet 8
+	// was skipped). n=24 maps to slot 8 which is FALSE → lost++. All other
+	// n in 17..25 map to slots that are set. n=16 is not strictly > 16. So
+	// lost = 1.
+	assert.Equal(t, int64(1), b.lostCounter.Count())
+
+	// Fill in 12, 13, 15, 16. Each is below current=25 (in-window). 16 must
+	// recheck slot 0 — it was set by NewBits and then cleared by the
+	// Update(25) jump, so 16 backfills cleanly.
+	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.True(t, b.Update(l, 17))
-	assert.True(t, b.Update(l, 18))
-	assert.True(t, b.Update(l, 19))
-	assert.Equal(t, int64(8), b.lostCounter.Count())
+	assert.Equal(t, int64(1), b.lostCounter.Count())

-	// Jump ahead by a window size
-	assert.True(t, b.Update(l, 29))
-	assert.Equal(t, int64(8), b.lostCounter.Count())
-	// Now lets walk ahead normally through the window, the missed packets should fill in
-	assert.True(t, b.Update(l, 30))
-	assert.True(t, b.Update(l, 31))
-	assert.True(t, b.Update(l, 32))
-	assert.True(t, b.Update(l, 33))
-	assert.True(t, b.Update(l, 34))
-	assert.True(t, b.Update(l, 35))
-	assert.True(t, b.Update(l, 36))
-	assert.True(t, b.Update(l, 37))
-	assert.True(t, b.Update(l, 38))
-	// 39 packets tracked, 22 seen, 17 lost
-	assert.Equal(t, int64(17), b.lostCounter.Count())
-
-	// Jump ahead by 2 windows, should have recording 1 full window missing
-	assert.True(t, b.Update(l, 58))
-	assert.Equal(t, int64(27), b.lostCounter.Count())
-	// Now lets walk ahead normally through the window, the missed packets should fill in from this window
-	assert.True(t, b.Update(l, 59))
-	assert.True(t, b.Update(l, 60))
-	assert.True(t, b.Update(l, 61))
-	assert.True(t, b.Update(l, 62))
-	assert.True(t, b.Update(l, 63))
-	assert.True(t, b.Update(l, 64))
-	assert.True(t, b.Update(l, 65))
-	assert.True(t, b.Update(l, 66))
-	assert.True(t, b.Update(l, 67))
-	// 68 packets tracked, 32 seen, 36 missed
-	assert.Equal(t, int64(36), b.lostCounter.Count())
+	// We missed packet 8 above and that loss is still recorded once, never
+	// double-counted, never zeroed.
+	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++ {
-		for i := range z.bits {
-			z.bits[i] = true
-		}
-		for i := range z.bits {
-			z.bits[i] = false
-		}
+// TestBitsWarmupOvershoot exercises the jump path's warmup arm with an
+// overshoot past one full window. NewBits leaves current=0 with only slot 0
+// "set" by the marker. Jumping straight to length+k must (a) clear every
+// slot the jump straddles, (b) count only past-window slack (not the
+// in-window slots, which never had a "lost" tenant during warmup), and
+// (c) leave the cursor at the new counter so subsequent unit advances
+// count from steady state. The marker bit at slot 0 is irrelevant once
+// current >= length.
+func TestBitsWarmupOvershoot(t *testing.T) {
+	l := test.NewLogger()
+	b := NewBits(16)
+	b.lostCounter.Clear()

+	// Jump from current=0 to i=20 (length=16, overshoot=4).
+	// Warmup arm: counts slots in [1..16] where bit unset and n>length.
+	// Only n=16 was unset and >length: but slot 16%16=0 is the marker,
+	// so b.get(16) reads bits[0]=1 and skips. Result: 0 lost from the loop.
+	// Past-window: i - current - length = 20 - 0 - 16 = 4 lost.
+	assert.True(t, b.Update(l, 20))
+	assert.Equal(t, int64(4), b.lostCounter.Count())
+	assert.Equal(t, uint64(20), b.current)
+
+	// Steady state now (current=20 >= length=16). Unit advance to 21
+	// stomps slot 21%16=5, which was cleared by the jump and not reset,
+	// so this is +1 lost.
+	assert.True(t, b.Update(l, 21))
+	assert.Equal(t, int64(5), b.lostCounter.Count())
+}
+
+// TestBitsCheckAcrossWarmupBoundary pins the underflow trick in Check's
+// in-window clause. While in warmup, b.current-b.length underflows uint64
+// to a huge value so the first OR-clause is always false; the second
+// clause (i < length && current < length) carries the in-window check.
+// Once current >= length the regimes flip cleanly.
+func TestBitsCheckAcrossWarmupBoundary(t *testing.T) {
+	l := test.NewLogger()
+	b := NewBits(16)
+
+	// Warmup: current=0. Check(0) must read the marker (set) and return false.
+	assert.False(t, b.Check(l, 0), "marker slot should look already-received")
+	// Warmup: any 0 < i < length is in-window and unset → accepted.
+	for i := uint64(1); i < 16; i++ {
+		assert.True(t, b.Check(l, i), "warmup in-window i=%d should be accepted", i)
+	}
+	// Warmup: i >= length but > current is "next number" so accepted.
+	assert.True(t, b.Check(l, 16))
+	assert.True(t, b.Check(l, 1_000_000))
+
+	// Cross into steady state.
+	assert.True(t, b.Update(l, 100))
+	// Now current=100, length=16. In-window range is [85..100].
+	// 84 is just outside: the underflow clause activates; 84 > 100-16=84 is false.
+	// And the warmup clause is false (current >= length). So out of window.
+	assert.False(t, b.Check(l, 84))
+	// 85 sits at the boundary. 85 > 84 is true → in window, unset → accept.
+	assert.True(t, b.Check(l, 85))
+	// 100 is current itself; not strictly greater, in-window, but already set.
+	assert.False(t, b.Check(l, 100))
+	// Way out: clearly out of window.
+	assert.False(t, b.Check(l, 50))
+}
+
+// TestBitsMarkerInvariant verifies the seeded bits[0]=1 marker behaves
+// correctly across warmup and beyond. Update should never clear the marker
+// during warmup (clearRange skips position 0 when startPos=1), and once
+// current >= length the marker is no longer consulted by Check/Update on
+// the live path — but it must still report counter 0 as a duplicate while
+// we are in warmup.
+func TestBitsMarkerInvariant(t *testing.T) {
+	l := test.NewLogger()
+	b := NewBits(8)
+
+	// Counter 0 is the seeded marker; Check sees it as already received.
+	assert.False(t, b.Check(l, 0))
+	// Update(0) at current=0 hits the duplicate branch.
+	b.dupeCounter.Clear()
+	assert.False(t, b.Update(l, 0))
+	assert.Equal(t, int64(1), b.dupeCounter.Count())
+
+	// Walk forward through warmup; the marker must remain set.
+	for n := uint64(1); n <= 7; n++ {
+		assert.True(t, b.Update(l, n))
+	}
+	// Position 0 (the marker) should still read as set because we never
+	// cleared it; Update(0) still looks like a duplicate.
+	assert.False(t, b.Check(l, 0))
+
+	// Cross into steady state with a unit advance to 8: pos=0, evicts the
+	// marker bit. The lost-counter guard (i > b.length) is false (8 == 8),
+	// so this advance does NOT charge a lost packet — exactly what the
+	// marker is there to prevent.
+	b.lostCounter.Clear()
+	assert.True(t, b.Update(l, 8))
+	assert.Equal(t, int64(0), b.lostCounter.Count())
+	// The slot at pos 0 is now occupied by counter 8.
+	assert.False(t, b.Check(l, 8))
+}
+
+// BenchmarkBitsUpdateInOrder is the steady-state hot path: each call is
+// i == current+1.
+func BenchmarkBitsUpdateInOrder(b *testing.B) {
+	l := test.NewLogger()
+	z := NewBits(16384)
+	for n := 0; n < b.N; n++ {
+		z.Update(l, uint64(n)+1)
+	}
+}
+
+// BenchmarkBitsUpdateReorder simulates light reorder within the window:
+// every other packet arrives one slot behind its predecessor (forces the
+// in-window backfill branch).
+func BenchmarkBitsUpdateReorder(b *testing.B) {
+	l := test.NewLogger()
+	z := NewBits(16384)
+	for n := 0; n < b.N; n++ {
+		base := uint64(n) * 2
+		z.Update(l, base+2)
+		z.Update(l, base+1)
+	}
+}
+
+// BenchmarkBitsUpdateLargeJumps stresses the clearRange word-level path.
+func BenchmarkBitsUpdateLargeJumps(b *testing.B) {
+	l := test.NewLogger()
+	z := NewBits(16384)
+	for n := 0; n < b.N; n++ {
+		z.Update(l, uint64(n+1)*1000)
 	}
 }
@@ -1,5 +1,4 @@
 //go:build boringcrypto
-// +build boringcrypto

 package nebula

@@ -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) {
+	return NewCAPoolFromPEMReader(bytes.NewReader(caPEMs))
+}
+
+// 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 err error
+
 	var expired bool
-	for {
-		caPEMs, err = pool.AddCAFromPEM(caPEMs)
-		if errors.Is(err, ErrExpired) {
-			expired = true
-			err = nil
+
+	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
 }
@@ -172,6 +217,10 @@ func (ncp *CAPool) verify(c Certificate, now time.Time, certFp string, signerFp
 		return nil, err
 	}

+	if signer.Certificate.Curve() != c.Curve() {
+		return nil, ErrCurveMismatch
+	}
+
 	if signer.Certificate.Expired(now) {
 		return nil, ErrRootExpired
 	}
@@ -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)
 }

@@ -558,3 +654,31 @@ func TestCertificateV2_Verify_Subnets(t *testing.T) {
 	_, err = caPool.VerifyCertificate(time.Now(), c)
 	require.NoError(t, err)
 }
+
+func TestCertificateV2_CurveMismatch(t *testing.T) {
+	caIp1 := mustParsePrefixUnmapped("10.0.0.0/16")
+	caIp2 := mustParsePrefixUnmapped("192.168.0.0/24")
+	ca, _, caKey, _ := NewTestCaCert(Version2, Curve_P256, time.Now(), time.Now().Add(10*time.Minute), []netip.Prefix{caIp1, caIp2}, nil, []string{"test"})
+
+	caPem, err := ca.MarshalPEM()
+	require.NoError(t, err)
+
+	caPool := NewCAPool()
+	b, err := caPool.AddCAFromPEM(caPem)
+	require.NoError(t, err)
+	assert.Empty(t, b)
+
+	// ip is outside the network
+	cIp1 := mustParsePrefixUnmapped("10.0.0.1/24")
+	c, _, _, _ := NewTestCert(Version2, Curve_P256, ca, caKey, "test", time.Now(), time.Now().Add(5*time.Minute), []netip.Prefix{cIp1}, nil, []string{"test"})
+
+	fp, _ := c.Fingerprint()
+	_, err = caPool.verify(c, time.Now(), fp, c.Issuer())
+	require.NoError(t, err)
+	//
+	c2 := c.(*certificateV2)
+	c2.curve = Curve_CURVE25519
+	fp, _ = c.Fingerprint()
+	_, err = caPool.verify(c, time.Now(), fp, c.Issuer())
+	require.Error(t, err)
+}
@@ -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()
+}
@@ -112,6 +112,9 @@ func (c *certificateV1) CheckSignature(key []byte) bool {
 	}
 	switch c.details.curve {
 	case Curve_CURVE25519:
+		if len(key) != ed25519.PublicKeySize {
+			return false //avoids a panic internal to ed25519
+		}
 		return ed25519.Verify(key, b, c.signature)
 	case Curve_P256:
 		pubKey, err := ecdsa.ParseUncompressedPublicKey(elliptic.P256(), key)
@@ -426,7 +429,7 @@ 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,
 		},
@@ -437,12 +440,19 @@ 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)
 	}

-	copy(nc.details.publicKey, rc.Details.PublicKey)
-
 	var ip netip.Addr
 	for i, rawIp := range rc.Details.Ips {
 		if i%2 == 0 {
@@ -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
@@ -151,6 +151,9 @@ func (c *certificateV2) CheckSignature(key []byte) bool {

 	switch c.curve {
 	case Curve_CURVE25519:
+		if len(key) != ed25519.PublicKeySize {
+			return false //avoids a panic internal to ed25519
+		}
 		return ed25519.Verify(key, b, c.signature)
 	case Curve_P256:
 		pubKey, err := ecdsa.ParseUncompressedPublicKey(elliptic.P256(), key)
@@ -592,7 +595,13 @@ 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
 	}
@@ -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
@@ -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")
@@ -21,6 +21,8 @@ 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")
+	ErrCurveMismatch              = errors.New("certificate curve does not match CA")

 	ErrInvalidPEMBlock                   = errors.New("input did not contain a valid PEM encoded block")
 	ErrInvalidPEMCertificateBanner       = errors.New("bytes did not contain a proper certificate banner")
@@ -13,6 +13,12 @@ import (
 	"golang.org/x/crypto/ed25519"
 )

+// testCertNow is the reference "now" used to derive default before/after times
+// in NewTestCaCert and NewTestCert. Holding it fixed for the lifetime of the
+// test binary keeps CA and leaf defaults aligned at the same second, so a leaf
+// signed with default times can never expire after its CA on a rounding race.
+var testCertNow = time.Now().Round(time.Second)
+
 // NewTestCaCert will create a new ca certificate
 func NewTestCaCert(version Version, curve Curve, before, after time.Time, networks, unsafeNetworks []netip.Prefix, groups []string) (Certificate, []byte, []byte, []byte) {
 	var err error
@@ -34,10 +40,10 @@ func NewTestCaCert(version Version, curve Curve, before, after time.Time, networ
 	}

 	if before.IsZero() {
-		before = time.Now().Add(time.Second * -60).Round(time.Second)
+		before = testCertNow.Add(time.Second * -60)
 	}
 	if after.IsZero() {
-		after = time.Now().Add(time.Second * 60).Round(time.Second)
+		after = testCertNow.Add(time.Second * 60)
 	}

 	t := &TBSCertificate{
@@ -70,11 +76,11 @@ func NewTestCaCert(version Version, curve Curve, before, after time.Time, networ
 // Expiry times are defaulted if you do not pass them in
 func NewTestCert(v Version, curve Curve, ca Certificate, key []byte, name string, before, after time.Time, networks, unsafeNetworks []netip.Prefix, groups []string) (Certificate, []byte, []byte, []byte) {
 	if before.IsZero() {
-		before = time.Now().Add(time.Second * -60).Round(time.Second)
+		before = testCertNow.Add(time.Second * -60)
 	}

 	if after.IsZero() {
-		after = time.Now().Add(time.Second * 60).Round(time.Second)
+		after = testCertNow.Add(time.Second * 60)
 	}

 	if len(networks) == 0 {
@@ -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)
+	})
+}
@@ -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)
+}
@@ -1,12 +1,66 @@
 package cert

 import (
+	"bytes"
 	"encoding/pem"
+	"errors"
 	"fmt"

 	"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"
@@ -37,19 +91,7 @@ func UnmarshalCertificateFromPEM(b []byte) (Certificate, []byte, error) {
 		return nil, r, ErrInvalidPEMBlock
 	}

-	var c Certificate
-	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
-	}
-
+	c, err := unmarshalCertificateBlock(p)
 	if err != nil {
 		return nil, r, err
 	}
@@ -58,6 +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())
@@ -86,12 +142,15 @@ 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
 	}
 }

+// UnmarshalPublicKeyFromPEM will try to unmarshal the first pem block in a byte array, returning any non
+// consumed data or an error on failure. Only key-agreement (ECDH) public key banners are accepted.
+// Use UnmarshalSigningPublicKeyFromPEM for Ed25519/ECDSA banners.
 func UnmarshalPublicKeyFromPEM(b []byte) ([]byte, []byte, Curve, error) {
 	k, r := pem.Decode(b)
 	if k == nil {
@@ -100,10 +159,10 @@ func UnmarshalPublicKeyFromPEM(b []byte) ([]byte, []byte, Curve, error) {
 	var expectedLen int
 	var curve Curve
 	switch k.Type {
-	case X25519PublicKeyBanner, Ed25519PublicKeyBanner:
+	case X25519PublicKeyBanner:
 		expectedLen = 32
 		curve = Curve_CURVE25519
-	case P256PublicKeyBanner, ECDSAP256PublicKeyBanner:
+	case P256PublicKeyBanner:
 		// Uncompressed
 		expectedLen = 65
 		curve = Curve_P256
@@ -116,6 +175,33 @@ func UnmarshalPublicKeyFromPEM(b []byte) ([]byte, []byte, Curve, error) {
 	return k.Bytes, r, curve, nil
 }

+// UnmarshalSigningPublicKeyFromPEM will try to unmarshal the first pem block in a byte array, returning any non
+// consumed data or an error on failure. Only Ed25519/ECDSA public key banners are accepted.
+// Use UnmarshalPublicKeyFromPEM for X25519/P256 (ECDH) banners.
+func UnmarshalSigningPublicKeyFromPEM(b []byte) ([]byte, []byte, Curve, error) {
+	k, r := pem.Decode(b)
+	if k == nil {
+		return nil, r, 0, fmt.Errorf("input did not contain a valid PEM encoded block")
+	}
+	var expectedLen int
+	var curve Curve
+	switch k.Type {
+	case Ed25519PublicKeyBanner:
+		expectedLen = 32
+		curve = Curve_CURVE25519
+	case ECDSAP256PublicKeyBanner:
+		// Uncompressed
+		expectedLen = 65
+		curve = Curve_P256
+	default:
+		return nil, r, 0, fmt.Errorf("bytes did not contain a proper Ed25519/ECDSA public key banner")
+	}
+	if len(k.Bytes) != expectedLen {
+		return nil, r, 0, fmt.Errorf("key was not %d bytes, is invalid %s public key", expectedLen, curve)
+	}
+	return k.Bytes, r, curve, nil
+}
+
 func MarshalPrivateKeyToPEM(curve Curve, b []byte) []byte {
 	switch curve {
 	case Curve_CURVE25519:
@@ -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)
@@ -179,60 +255,6 @@ AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=
 func TestUnmarshalPublicKeyFromPEM(t *testing.T) {
 	t.Parallel()
 	pubKey := []byte(`# A good key
-----BEGIN NEBULA ED25519 PUBLIC KEY-----
-AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=
-----END NEBULA ED25519 PUBLIC KEY-----
-`)
-	shortKey := []byte(`# A short key
-----BEGIN NEBULA ED25519 PUBLIC KEY-----
-AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA==
-----END NEBULA ED25519 PUBLIC KEY-----
-`)
-	invalidBanner := []byte(`# Invalid banner
-----BEGIN NOT A NEBULA PUBLIC KEY-----
-AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=
-----END NOT A NEBULA PUBLIC KEY-----
-`)
-	invalidPem := []byte(`# Not a valid PEM format
-BEGIN NEBULA ED25519 PUBLIC KEY-----
-AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=
-END NEBULA ED25519 PUBLIC KEY-----`)
-
-	keyBundle := appendByteSlices(pubKey, shortKey, invalidBanner, invalidPem)
-
-	// Success test case
-	k, rest, curve, err := UnmarshalPublicKeyFromPEM(keyBundle)
-	assert.Len(t, k, 32)
-	assert.Equal(t, Curve_CURVE25519, curve)
-	require.NoError(t, err)
-	assert.Equal(t, rest, appendByteSlices(shortKey, invalidBanner, invalidPem))
-
-	// Fail due to short key
-	k, rest, curve, err = UnmarshalPublicKeyFromPEM(rest)
-	assert.Nil(t, k)
-	assert.Equal(t, Curve_CURVE25519, curve)
-	assert.Equal(t, rest, appendByteSlices(invalidBanner, invalidPem))
-	require.EqualError(t, err, "key was not 32 bytes, is invalid CURVE25519 public key")
-
-	// Fail due to invalid banner
-	k, rest, curve, err = UnmarshalPublicKeyFromPEM(rest)
-	assert.Nil(t, k)
-	assert.Equal(t, Curve_CURVE25519, curve)
-	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
-	// it's missing the requisite pre-encapsulation boundary.
-	k, rest, curve, err = UnmarshalPublicKeyFromPEM(rest)
-	assert.Nil(t, k)
-	assert.Equal(t, Curve_CURVE25519, curve)
-	assert.Equal(t, rest, invalidPem)
-	require.EqualError(t, err, "input did not contain a valid PEM encoded block")
-}
-
-func TestUnmarshalX25519PublicKey(t *testing.T) {
-	t.Parallel()
-	pubKey := []byte(`# A good key
 -----BEGIN NEBULA X25519 PUBLIC KEY-----
 AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=
 -----END NEBULA X25519 PUBLIC KEY-----
@@ -243,7 +265,7 @@ AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
 AAAAAAAAAAAAAAAAAAAAAAA=
 -----END NEBULA P256 PUBLIC KEY-----
 `)
-	oldPubP256Key := []byte(`# A good key
+	signingKey := []byte(`# A signing key has the wrong scope for this function
 -----BEGIN NEBULA ECDSA P256 PUBLIC KEY-----
 AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
 AAAAAAAAAAAAAAAAAAAAAAA=
@@ -264,44 +286,118 @@ AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=
 AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=
 -END NEBULA X25519 PUBLIC KEY-----`)

-	keyBundle := appendByteSlices(pubKey, pubP256Key, oldPubP256Key, shortKey, invalidBanner, invalidPem)
+	keyBundle := appendByteSlices(pubKey, pubP256Key, signingKey, shortKey, invalidBanner, invalidPem)

-	// Success test case
+	// X25519 key
 	k, rest, curve, err := UnmarshalPublicKeyFromPEM(keyBundle)
 	assert.Len(t, k, 32)
 	require.NoError(t, err)
-	assert.Equal(t, rest, appendByteSlices(pubP256Key, oldPubP256Key, shortKey, invalidBanner, invalidPem))
+	assert.Equal(t, rest, appendByteSlices(pubP256Key, signingKey, shortKey, invalidBanner, invalidPem))
 	assert.Equal(t, Curve_CURVE25519, curve)

-	// Success test case
+	// P256 key
 	k, rest, curve, err = UnmarshalPublicKeyFromPEM(rest)
 	assert.Len(t, k, 65)
 	require.NoError(t, err)
-	assert.Equal(t, rest, appendByteSlices(oldPubP256Key, shortKey, invalidBanner, invalidPem))
+	assert.Equal(t, rest, appendByteSlices(signingKey, shortKey, invalidBanner, invalidPem))
 	assert.Equal(t, Curve_P256, curve)

-	// Success test case
-	k, rest, curve, err = UnmarshalPublicKeyFromPEM(rest)
-	assert.Len(t, k, 65)
-	require.NoError(t, err)
+	// Reject a signing public key (Ed25519/ECDSA banner)
+	k, rest, _, err = UnmarshalPublicKeyFromPEM(rest)
+	assert.Nil(t, k)
 	assert.Equal(t, rest, appendByteSlices(shortKey, invalidBanner, invalidPem))
-	assert.Equal(t, Curve_P256, curve)
+	require.EqualError(t, err, "bytes did not contain a proper public key banner")

 	// Fail due to short key
-	k, rest, curve, err = UnmarshalPublicKeyFromPEM(rest)
+	k, rest, _, err = UnmarshalPublicKeyFromPEM(rest)
 	assert.Nil(t, k)
 	assert.Equal(t, rest, appendByteSlices(invalidBanner, invalidPem))
 	require.EqualError(t, err, "key was not 32 bytes, is invalid CURVE25519 public key")

 	// Fail due to invalid banner
-	k, rest, curve, err = UnmarshalPublicKeyFromPEM(rest)
+	k, rest, _, err = UnmarshalPublicKeyFromPEM(rest)
 	assert.Nil(t, k)
 	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)
+	k, rest, _, err = UnmarshalPublicKeyFromPEM(rest)
+	assert.Nil(t, k)
+	assert.Equal(t, rest, invalidPem)
+	require.EqualError(t, err, "input did not contain a valid PEM encoded block")
+}
+
+func TestUnmarshalSigningPublicKeyFromPEM(t *testing.T) {
+	t.Parallel()
+	pubKey := []byte(`# A good key
+-----BEGIN NEBULA ED25519 PUBLIC KEY-----
+AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=
+-----END NEBULA ED25519 PUBLIC KEY-----
+`)
+	pubP256Key := []byte(`# A good key
+-----BEGIN NEBULA ECDSA P256 PUBLIC KEY-----
+AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
+AAAAAAAAAAAAAAAAAAAAAAA=
+-----END NEBULA ECDSA P256 PUBLIC KEY-----
+`)
+	ecdhKey := []byte(`# A key-agreement key has the wrong scope for this function
+-----BEGIN NEBULA X25519 PUBLIC KEY-----
+AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=
+-----END NEBULA X25519 PUBLIC KEY-----
+`)
+	shortKey := []byte(`# A short key
+-----BEGIN NEBULA ED25519 PUBLIC KEY-----
+AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA==
+-----END NEBULA ED25519 PUBLIC KEY-----
+`)
+	invalidBanner := []byte(`# Invalid banner
+-----BEGIN NOT A NEBULA PUBLIC KEY-----
+AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=
+-----END NOT A NEBULA PUBLIC KEY-----
+`)
+	invalidPem := []byte(`# Not a valid PEM format
+-BEGIN NEBULA ED25519 PUBLIC KEY-----
+AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=
+-END NEBULA ED25519 PUBLIC KEY-----`)
+
+	keyBundle := appendByteSlices(pubKey, pubP256Key, ecdhKey, shortKey, invalidBanner, invalidPem)
+
+	// Ed25519 key
+	k, rest, curve, err := UnmarshalSigningPublicKeyFromPEM(keyBundle)
+	assert.Len(t, k, 32)
+	require.NoError(t, err)
+	assert.Equal(t, rest, appendByteSlices(pubP256Key, ecdhKey, shortKey, invalidBanner, invalidPem))
+	assert.Equal(t, Curve_CURVE25519, curve)
+
+	// ECDSA P256 key
+	k, rest, curve, err = UnmarshalSigningPublicKeyFromPEM(rest)
+	assert.Len(t, k, 65)
+	require.NoError(t, err)
+	assert.Equal(t, rest, appendByteSlices(ecdhKey, shortKey, invalidBanner, invalidPem))
+	assert.Equal(t, Curve_P256, curve)
+
+	// Reject a key-agreement public key (X25519/P256 banner)
+	k, rest, _, err = UnmarshalSigningPublicKeyFromPEM(rest)
+	assert.Nil(t, k)
+	assert.Equal(t, rest, appendByteSlices(shortKey, invalidBanner, invalidPem))
+	require.EqualError(t, err, "bytes did not contain a proper Ed25519/ECDSA public key banner")
+
+	// Fail due to short key
+	k, rest, _, err = UnmarshalSigningPublicKeyFromPEM(rest)
+	assert.Nil(t, k)
+	assert.Equal(t, rest, appendByteSlices(invalidBanner, invalidPem))
+	require.EqualError(t, err, "key was not 32 bytes, is invalid CURVE25519 public key")
+
+	// Fail due to invalid banner
+	k, rest, _, err = UnmarshalSigningPublicKeyFromPEM(rest)
+	assert.Nil(t, k)
+	require.EqualError(t, err, "bytes did not contain a proper Ed25519/ECDSA public key banner")
+	assert.Equal(t, rest, invalidPem)
+
+	// Fail due to invalid PEM format, because
+	// it's missing the requisite pre-encapsulation boundary.
+	k, rest, _, err = UnmarshalSigningPublicKeyFromPEM(rest)
 	assert.Nil(t, k)
 	assert.Equal(t, rest, invalidPem)
 	require.EqualError(t, err, "input did not contain a valid PEM encoded block")
@@ -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
@@ -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)
+	}
+}
@@ -14,6 +14,12 @@ import (
 	"golang.org/x/crypto/ed25519"
 )

+// testCertNow is the reference "now" used to derive default before/after times
+// in NewTestCaCert and NewTestCert. Holding it fixed for the lifetime of the
+// test binary keeps CA and leaf defaults aligned at the same second, so a leaf
+// signed with default times can never expire after its CA on a rounding race.
+var testCertNow = time.Now().Round(time.Second)
+
 // NewTestCaCert will create a new ca certificate
 func NewTestCaCert(version cert.Version, curve cert.Curve, before, after time.Time, networks, unsafeNetworks []netip.Prefix, groups []string) (cert.Certificate, []byte, []byte, []byte) {
 	var err error
@@ -35,10 +41,10 @@ func NewTestCaCert(version cert.Version, curve cert.Curve, before, after time.Ti
 	}

 	if before.IsZero() {
-		before = time.Now().Add(time.Second * -60).Round(time.Second)
+		before = testCertNow.Add(time.Second * -60)
 	}
 	if after.IsZero() {
-		after = time.Now().Add(time.Second * 60).Round(time.Second)
+		after = testCertNow.Add(time.Second * 60)
 	}

 	t := &cert.TBSCertificate{
@@ -71,11 +77,11 @@ func NewTestCaCert(version cert.Version, curve cert.Curve, before, after time.Ti
 // Expiry times are defaulted if you do not pass them in
 func NewTestCert(v cert.Version, curve cert.Curve, ca cert.Certificate, key []byte, name string, before, after time.Time, networks, unsafeNetworks []netip.Prefix, groups []string) (cert.Certificate, []byte, []byte, []byte) {
 	if before.IsZero() {
-		before = time.Now().Add(time.Second * -60).Round(time.Second)
+		before = testCertNow.Add(time.Second * -60)
 	}

 	if after.IsZero() {
-		after = time.Now().Add(time.Second * 60).Round(time.Second)
+		after = testCertNow.Add(time.Second * 60)
 	}

 	var pub, priv []byte
@@ -114,6 +120,33 @@ func NewTestCert(v cert.Version, curve cert.Curve, ca cert.Certificate, key []by
 	return c, pub, cert.MarshalPrivateKeyToPEM(curve, priv), pem
 }

+func NewTestCertDifferentVersion(c cert.Certificate, v cert.Version, ca cert.Certificate, key []byte) (cert.Certificate, []byte) {
+	nc := &cert.TBSCertificate{
+		Version:        v,
+		Curve:          c.Curve(),
+		Name:           c.Name(),
+		Networks:       c.Networks(),
+		UnsafeNetworks: c.UnsafeNetworks(),
+		Groups:         c.Groups(),
+		NotBefore:      time.Unix(c.NotBefore().Unix(), 0),
+		NotAfter:       time.Unix(c.NotAfter().Unix(), 0),
+		PublicKey:      c.PublicKey(),
+		IsCA:           false,
+	}
+
+	c, err := nc.Sign(ca, ca.Curve(), key)
+	if err != nil {
+		panic(err)
+	}
+
+	pem, err := c.MarshalPEM()
+	if err != nil {
+		panic(err)
+	}
+
+	return c, pem
+}
+
 func X25519Keypair() ([]byte, []byte) {
 	privkey := make([]byte, 32)
 	if _, err := io.ReadFull(rand.Reader, privkey); err != nil {
@@ -136,3 +169,55 @@ 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
+}
@@ -97,6 +97,19 @@ func ca(args []string, out io.Writer, errOut io.Writer, pr PasswordReader) error
 		if err = mustFlagString("out-key", cf.outKeyPath); err != nil {
 			return err
 		}
+	} else {
+		// out-key is meaningless under PKCS#11 because the private key never
+		// leaves the HSM; reject it so we never silently accept or claim a
+		// stdout slot for it.
+		outKeySet := false
+		cf.set.Visit(func(f *flag.Flag) {
+			if f.Name == "out-key" {
+				outKeySet = true
+			}
+		})
+		if outKeySet {
+			return newHelpErrorf("cannot set -out-key with -pkcs11")
+		}
 	}
 	if err := mustFlagString("out-crt", cf.outCertPath); err != nil {
 		return err
@@ -171,25 +184,37 @@ func ca(args []string, out io.Writer, errOut io.Writer, pr PasswordReader) error
 		}
 	}

+	var claims ioClaims
+	if err := reserveOutputs(&claims,
+		"out-key", *cf.outKeyPath,
+		"out-crt", *cf.outCertPath,
+		"out-qr", *cf.outQRPath,
+	); err != nil {
+		return err
+	}
+
 	var passphrase []byte
 	if !isP11 && *cf.encryption {
-		for i := 0; i < 5; i++ {
-			out.Write([]byte("Enter passphrase: "))
-			passphrase, err = pr.ReadPassword()
-
-			if err == ErrNoTerminal {
-				return fmt.Errorf("out-key must be encrypted interactively")
-			} else if err != nil {
-				return fmt.Errorf("error reading passphrase: %s", err)
-			}
-
-			if len(passphrase) > 0 {
-				break
-			}
-		}
-
+		passphrase = []byte(os.Getenv("NEBULA_CA_PASSPHRASE"))
 		if len(passphrase) == 0 {
-			return fmt.Errorf("no passphrase specified, remove -encrypt flag to write out-key in plaintext")
+			for i := 0; i < 5; i++ {
+				errOut.Write([]byte("Enter passphrase: "))
+				passphrase, err = pr.ReadPassword()
+
+				if err == ErrNoTerminal {
+					return fmt.Errorf("out-key must be encrypted interactively")
+				} else if err != nil {
+					return fmt.Errorf("error reading passphrase: %s", err)
+				}
+
+				if len(passphrase) > 0 {
+					break
+				}
+			}
+
+			if len(passphrase) == 0 {
+				return fmt.Errorf("no passphrase specified, remove -encrypt flag to write out-key in plaintext")
+			}
 		}
 	}

@@ -258,14 +283,16 @@ func ca(args []string, out io.Writer, errOut io.Writer, pr PasswordReader) error
 		Curve:          curve,
 	}

-	if !isP11 {
+	if !isP11 && !isStdio(*cf.outKeyPath) {
 		if _, err := os.Stat(*cf.outKeyPath); err == nil {
 			return fmt.Errorf("refusing to overwrite existing CA key: %s", *cf.outKeyPath)
 		}
 	}

-	if _, err := os.Stat(*cf.outCertPath); err == nil {
-		return fmt.Errorf("refusing to overwrite existing CA cert: %s", *cf.outCertPath)
+	if !isStdio(*cf.outCertPath) {
+		if _, err := os.Stat(*cf.outCertPath); err == nil {
+			return fmt.Errorf("refusing to overwrite existing CA cert: %s", *cf.outCertPath)
+		}
 	}

 	var c cert.Certificate
@@ -291,7 +318,7 @@ func ca(args []string, out io.Writer, errOut io.Writer, pr PasswordReader) error
 			b = cert.MarshalSigningPrivateKeyToPEM(curve, rawPriv)
 		}

-		err = os.WriteFile(*cf.outKeyPath, b, 0600)
+		err = writeOutput(*cf.outKeyPath, b, 0600, out)
 		if err != nil {
 			return fmt.Errorf("error while writing out-key: %s", err)
 		}
@@ -302,7 +329,7 @@ func ca(args []string, out io.Writer, errOut io.Writer, pr PasswordReader) error
 		return fmt.Errorf("error while marshalling certificate: %s", err)
 	}

-	err = os.WriteFile(*cf.outCertPath, b, 0600)
+	err = writeOutput(*cf.outCertPath, b, 0600, out)
 	if err != nil {
 		return fmt.Errorf("error while writing out-crt: %s", err)
 	}
@@ -313,7 +340,7 @@ func ca(args []string, out io.Writer, errOut io.Writer, pr PasswordReader) error
 			return fmt.Errorf("error while generating qr code: %s", err)
 		}

-		err = os.WriteFile(*cf.outQRPath, b, 0600)
+		err = writeOutput(*cf.outQRPath, b, 0600, out)
 		if err != nil {
 			return fmt.Errorf("error while writing out-qr: %s", err)
 		}
@@ -329,6 +356,7 @@ func caSummary() string {
 func caHelp(out io.Writer) {
 	cf := newCaFlags()
 	out.Write([]byte("Usage of " + os.Args[0] + " " + caSummary() + "\n"))
+	out.Write([]byte(stdioHelpText))
 	cf.set.SetOutput(out)
 	cf.set.PrintDefaults()
 }
@@ -27,6 +27,7 @@ func Test_caHelp(t *testing.T) {
 	assert.Equal(
 		t,
 		"Usage of "+os.Args[0]+" ca <flags>: create a self signed certificate authority\n"+
+			"  Pass \"-\" to any path flag to read from stdin or write to stdout.\n"+
 			"  -argon-iterations uint\n"+
 			"    \tOptional: Argon2 iterations parameter used for encrypted private key passphrase (default 1)\n"+
 			"  -argon-memory uint\n"+
@@ -84,7 +85,7 @@ func Test_ca(t *testing.T) {
 		err:      nil,
 	}

-	pwPromptOb := "Enter passphrase: "
+	pwPromptEB := "Enter passphrase: "

 	// required args
 	assertHelpError(t, ca(
@@ -168,8 +169,19 @@ func Test_ca(t *testing.T) {
 	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()}
 	require.NoError(t, ca(args, ob, eb, testpw))
-	assert.Equal(t, pwPromptOb, ob.String())
+	assert.Empty(t, ob.String())
+	assert.Equal(t, pwPromptEB, 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())
@@ -189,15 +201,15 @@ 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()
 	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()}
 	require.Error(t, ca(args, ob, eb, errpw))
-	assert.Equal(t, pwPromptOb, ob.String())
-	assert.Empty(t, eb.String())
+	assert.Empty(t, ob.String())
+	assert.Equal(t, pwPromptEB, eb.String())

 	// test when user fails to enter a password
 	os.Remove(keyF.Name())
@@ -206,8 +218,8 @@ func Test_ca(t *testing.T) {
 	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()}
 	require.EqualError(t, ca(args, ob, eb, nopw), "no passphrase specified, remove -encrypt flag to write out-key in plaintext")
-	assert.Equal(t, strings.Repeat(pwPromptOb, 5), ob.String()) // prompts 5 times before giving up
-	assert.Empty(t, eb.String())
+	assert.Empty(t, ob.String())
+	assert.Equal(t, strings.Repeat(pwPromptEB, 5), eb.String()) // prompts 5 times before giving up

 	// create valid cert/key for overwrite tests
 	os.Remove(keyF.Name())
@@ -236,3 +248,67 @@ func Test_ca(t *testing.T) {
 	os.Remove(keyF.Name())

 }
+
+func Test_ca_stdio(t *testing.T) {
+	nopw := &StubPasswordReader{}
+
+	keyF, err := os.CreateTemp("", "ca.key")
+	require.NoError(t, err)
+	os.Remove(keyF.Name())
+	defer os.Remove(keyF.Name())
+
+	crtF, err := os.CreateTemp("", "ca.crt")
+	require.NoError(t, err)
+	os.Remove(crtF.Name())
+	defer os.Remove(crtF.Name())
+
+	// out-crt on stdout, out-key on disk
+	ob := &bytes.Buffer{}
+	eb := &bytes.Buffer{}
+	require.NoError(t, ca([]string{"-name", "test-ca", "-duration", "1h", "-out-crt", "-", "-out-key", keyF.Name()}, ob, eb, nopw))
+	assert.Empty(t, eb.String())
+	c, _, err := cert.UnmarshalCertificateFromPEM(ob.Bytes())
+	require.NoError(t, err)
+	assert.True(t, c.IsCA())
+	assert.Equal(t, "test-ca", c.Name())
+
+	// out-key on stdout, out-crt on disk
+	os.Remove(keyF.Name())
+	ob.Reset()
+	eb.Reset()
+	require.NoError(t, ca([]string{"-name", "test-ca", "-duration", "1h", "-out-crt", crtF.Name(), "-out-key", "-"}, ob, eb, nopw))
+	assert.Empty(t, eb.String())
+	_, _, curve, err := cert.UnmarshalSigningPrivateKeyFromPEM(ob.Bytes())
+	require.NoError(t, err)
+	assert.Equal(t, cert.Curve_CURVE25519, curve)
+
+	// dual stdout is rejected up front
+	os.Remove(crtF.Name())
+	ob.Reset()
+	eb.Reset()
+	require.EqualError(t,
+		ca([]string{"-name", "test-ca", "-duration", "1h", "-out-crt", "-", "-out-key", "-"}, ob, eb, nopw),
+		`-out-key and -out-crt both set to "-", only one output may write to stdout`)
+	assert.Empty(t, ob.String())
+
+	// an output conflict combined with -encrypt must error BEFORE prompting
+	// for a passphrase; pr would record any read attempt
+	tracker := &trackingPasswordReader{}
+	ob.Reset()
+	eb.Reset()
+	require.EqualError(t,
+		ca([]string{"-name", "test-ca", "-duration", "1h", "-encrypt", "-out-crt", "-", "-out-key", "-"}, ob, eb, tracker),
+		`-out-key and -out-crt both set to "-", only one output may write to stdout`)
+	assert.Empty(t, ob.String())
+	assert.Empty(t, eb.String())
+	assert.Zero(t, tracker.calls, "passphrase prompt should not have been called")
+}
+
+type trackingPasswordReader struct {
+	calls int
+}
+
+func (pr *trackingPasswordReader) ReadPassword() ([]byte, error) {
+	pr.calls++
+	return []byte(""), nil
+}
@@ -42,6 +42,8 @@ func keygen(args []string, out io.Writer, errOut io.Writer) error {
 		if err = mustFlagString("out-key", cf.outKeyPath); err != nil {
 			return err
 		}
+	} else if *cf.outKeyPath != "" {
+		return newHelpErrorf("cannot set -out-key with -pkcs11")
 	}
 	if err = mustFlagString("out-pub", cf.outPubPath); err != nil {
 		return err
@@ -69,6 +71,14 @@ func keygen(args []string, out io.Writer, errOut io.Writer) error {
 		}
 	}

+	var claims ioClaims
+	if err := reserveOutputs(&claims,
+		"out-key", *cf.outKeyPath,
+		"out-pub", *cf.outPubPath,
+	); err != nil {
+		return err
+	}
+
 	if isP11 {
 		p11Client, err := pkclient.FromUrl(*cf.p11url)
 		if err != nil {
@@ -82,12 +92,12 @@ func keygen(args []string, out io.Writer, errOut io.Writer) error {
 			return fmt.Errorf("error while getting public key: %w", err)
 		}
 	} else {
-		err = os.WriteFile(*cf.outKeyPath, cert.MarshalPrivateKeyToPEM(curve, rawPriv), 0600)
+		err = writeOutput(*cf.outKeyPath, cert.MarshalPrivateKeyToPEM(curve, rawPriv), 0600, out)
 		if err != nil {
 			return fmt.Errorf("error while writing out-key: %s", err)
 		}
 	}
-	err = os.WriteFile(*cf.outPubPath, cert.MarshalPublicKeyToPEM(curve, pub), 0600)
+	err = writeOutput(*cf.outPubPath, cert.MarshalPublicKeyToPEM(curve, pub), 0600, out)
 	if err != nil {
 		return fmt.Errorf("error while writing out-pub: %s", err)
 	}
@@ -102,6 +112,7 @@ func keygenSummary() string {
 func keygenHelp(out io.Writer) {
 	cf := newKeygenFlags()
 	_, _ = out.Write([]byte("Usage of " + os.Args[0] + " " + keygenSummary() + "\n"))
+	_, _ = out.Write([]byte(stdioHelpText))
 	cf.set.SetOutput(out)
 	cf.set.PrintDefaults()
 }
@@ -20,6 +20,7 @@ func Test_keygenHelp(t *testing.T) {
 	assert.Equal(
 		t,
 		"Usage of "+os.Args[0]+" keygen <flags>: create a public/private key pair. the public key can be passed to `nebula-cert sign`\n"+
+			"  Pass \"-\" to any path flag to read from stdin or write to stdout.\n"+
 			"  -curve string\n"+
 			"    \tECDH Curve (25519, P256) (default \"25519\")\n"+
 			"  -out-key string\n"+
@@ -93,3 +94,43 @@ func Test_keygen(t *testing.T) {
 	require.NoError(t, err)
 	assert.Len(t, lPub, 32)
 }
+
+func Test_keygen_stdio(t *testing.T) {
+	keyF, err := os.CreateTemp("", "test.key")
+	require.NoError(t, err)
+	os.Remove(keyF.Name())
+	defer os.Remove(keyF.Name())
+
+	pubF, err := os.CreateTemp("", "test.pub")
+	require.NoError(t, err)
+	os.Remove(pubF.Name())
+	defer os.Remove(pubF.Name())
+
+	// out-pub on stdout, out-key on disk
+	ob := &bytes.Buffer{}
+	eb := &bytes.Buffer{}
+	require.NoError(t, keygen([]string{"-out-pub", "-", "-out-key", keyF.Name()}, ob, eb))
+	assert.Empty(t, eb.String())
+	lPub, _, curve, err := cert.UnmarshalPublicKeyFromPEM(ob.Bytes())
+	require.NoError(t, err)
+	assert.Equal(t, cert.Curve_CURVE25519, curve)
+	assert.Len(t, lPub, 32)
+
+	// out-key on stdout, out-pub on disk
+	os.Remove(keyF.Name())
+	ob.Reset()
+	eb.Reset()
+	require.NoError(t, keygen([]string{"-out-pub", pubF.Name(), "-out-key", "-"}, ob, eb))
+	assert.Empty(t, eb.String())
+	lKey, _, curve, err := cert.UnmarshalPrivateKeyFromPEM(ob.Bytes())
+	require.NoError(t, err)
+	assert.Equal(t, cert.Curve_CURVE25519, curve)
+	assert.Len(t, lKey, 32)
+
+	// both on stdout is a conflict caught up front
+	ob.Reset()
+	eb.Reset()
+	require.EqualError(t, keygen([]string{"-out-pub", "-", "-out-key", "-"}, ob, eb),
+		`-out-key and -out-pub both set to "-", only one output may write to stdout`)
+	assert.Empty(t, ob.String())
+}
@@ -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
 }
@@ -22,7 +22,9 @@ func (pr StdinPasswordReader) ReadPassword() ([]byte, error) {
 	}

 	password, err := term.ReadPassword(int(os.Stdin.Fd()))
-	fmt.Println()
+	// Terminal echo is off while reading, so the user's Enter key does not
+	// produce a visible newline. Emit one on stderr to match the prompt.
+	fmt.Fprintln(os.Stderr)

 	return password, err
 }
@@ -40,11 +40,23 @@ func printCert(args []string, out io.Writer, errOut io.Writer) error {
 		return err
 	}

-	rawCert, err := os.ReadFile(*pf.path)
+	var claims ioClaims
+	if err := reserveInputs(&claims, "path", *pf.path); err != nil {
+		return err
+	}
+	if err := reserveOutputs(&claims, "out-qr", *pf.outQRPath); err != nil {
+		return err
+	}
+
+	rawCert, err := readInput("path", *pf.path, &claims)
 	if err != nil {
 		return fmt.Errorf("unable to read cert; %s", err)
 	}

+	// When the QR is going to stdout, suppress the human-readable text/json
+	// output so the binary stream is not contaminated.
+	qrToStdout := isStdio(*pf.outQRPath)
+
 	var c cert.Certificate
 	var qrBytes []byte
 	part := 0
@@ -57,11 +69,13 @@ func printCert(args []string, out io.Writer, errOut io.Writer) error {
 			return fmt.Errorf("error while unmarshaling cert: %s", err)
 		}

-		if *pf.json {
-			jsonCerts = append(jsonCerts, c)
-		} else {
-			_, _ = out.Write([]byte(c.String()))
-			_, _ = out.Write([]byte("\n"))
+		if !qrToStdout {
+			if *pf.json {
+				jsonCerts = append(jsonCerts, c)
+			} else {
+				_, _ = out.Write([]byte(c.String()))
+				_, _ = out.Write([]byte("\n"))
+			}
 		}

 		if *pf.outQRPath != "" {
@@ -79,7 +93,7 @@ func printCert(args []string, out io.Writer, errOut io.Writer) error {
 		part++
 	}

-	if *pf.json {
+	if *pf.json && !qrToStdout {
 		b, _ := json.Marshal(jsonCerts)
 		_, _ = out.Write(b)
 		_, _ = out.Write([]byte("\n"))
@@ -91,7 +105,7 @@ func printCert(args []string, out io.Writer, errOut io.Writer) error {
 			return fmt.Errorf("error while generating qr code: %s", err)
 		}

-		err = os.WriteFile(*pf.outQRPath, b, 0600)
+		err = writeOutput(*pf.outQRPath, b, 0600, out)
 		if err != nil {
 			return fmt.Errorf("error while writing out-qr: %s", err)
 		}
@@ -107,6 +121,7 @@ func printSummary() string {
 func printHelp(out io.Writer) {
 	pf := newPrintFlags()
 	out.Write([]byte("Usage of " + os.Args[0] + " " + printSummary() + "\n"))
+	out.Write([]byte(stdioHelpText))
 	pf.set.SetOutput(out)
 	pf.set.PrintDefaults()
 }
@@ -25,6 +25,7 @@ func Test_printHelp(t *testing.T) {
 	assert.Equal(
 		t,
 		"Usage of "+os.Args[0]+" print <flags>: prints details about a certificate\n"+
+			"  Pass \"-\" to any path flag to read from stdin or write to stdout.\n"+
 			"  -json\n"+
 			"    \tOptional: outputs certificates in json format\n"+
 			"  -out-qr string\n"+
@@ -178,6 +179,44 @@ func Test_printCert(t *testing.T) {
 		ob.String(),
 	)
 	assert.Empty(t, eb.String())
+
+	// read cert from stdin
+	ob.Reset()
+	eb.Reset()
+	withStdin(t, bytes.NewReader(p))
+	err = printCert([]string{"-json", "-path", "-"}, ob, eb)
+	require.NoError(t, err)
+	assert.Equal(
+		t,
+		`[{"details":{"curve":"CURVE25519","groups":["hi"],"isCa":false,"issuer":"`+c.Issuer()+`","name":"test","networks":["10.0.0.123/8"],"notAfter":"0001-01-01T00:00:00Z","notBefore":"0001-01-01T00:00:00Z","publicKey":"`+pk+`","unsafeNetworks":[]},"fingerprint":"`+fp+`","signature":"`+sig+`","version":1}]
+`,
+		ob.String(),
+	)
+	assert.Empty(t, eb.String())
+
+	// -out-qr - sends only the PNG to stdout, suppressing the cert dump
+	ob.Reset()
+	eb.Reset()
+	withStdin(t, bytes.NewReader(p))
+	err = printCert([]string{"-path", "-", "-out-qr", "-"}, ob, eb)
+	require.NoError(t, err)
+	assert.Empty(t, eb.String())
+	stdout := ob.Bytes()
+	require.NotEmpty(t, stdout)
+	// PNG magic, no PEM/JSON noise prepended
+	assert.Equal(t, []byte{0x89, 'P', 'N', 'G', 0x0d, 0x0a, 0x1a, 0x0a}, stdout[:8])
+	assert.NotContains(t, string(stdout), "NebulaCertificate")
+	assert.NotContains(t, string(stdout), `"details"`)
+
+	// json + out-qr - still suppresses json
+	ob.Reset()
+	eb.Reset()
+	withStdin(t, bytes.NewReader(p))
+	err = printCert([]string{"-json", "-path", "-", "-out-qr", "-"}, ob, eb)
+	require.NoError(t, err)
+	assert.Empty(t, eb.String())
+	assert.Equal(t, []byte{0x89, 'P', 'N', 'G'}, ob.Bytes()[:4])
+	assert.NotContains(t, ob.String(), `"details"`)
 }

 // NewTestCaCert will generate a CA cert
@@ -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")
@@ -85,6 +85,9 @@ func signCert(args []string, out io.Writer, errOut io.Writer, pr PasswordReader)
 	if !isP11 && *sf.inPubPath != "" && *sf.outKeyPath != "" {
 		return newHelpErrorf("cannot set both -in-pub and -out-key")
 	}
+	if isP11 && *sf.outKeyPath != "" {
+		return newHelpErrorf("cannot set -out-key with -pkcs11")
+	}

 	var v4Networks []netip.Prefix
 	var v6Networks []netip.Prefix
@@ -102,13 +105,35 @@ func signCert(args []string, out io.Writer, errOut io.Writer, pr PasswordReader)
 		return newHelpErrorf("-version must be either %v or %v", cert.Version1, cert.Version2)
 	}

+	if *sf.outKeyPath == "" {
+		*sf.outKeyPath = *sf.name + ".key"
+	}
+	if *sf.outCertPath == "" {
+		*sf.outCertPath = *sf.name + ".crt"
+	}
+
+	var claims ioClaims
+	if err := reserveInputs(&claims,
+		"ca-key", *sf.caKeyPath,
+		"ca-crt", *sf.caCertPath,
+		"in-pub", *sf.inPubPath,
+	); err != nil {
+		return err
+	}
+	if err := reserveOutputs(&claims,
+		"out-key", *sf.outKeyPath,
+		"out-crt", *sf.outCertPath,
+		"out-qr", *sf.outQRPath,
+	); err != nil {
+		return err
+	}
+
 	var curve cert.Curve
 	var caKey []byte

 	if !isP11 {
 		var rawCAKey []byte
-		rawCAKey, err := os.ReadFile(*sf.caKeyPath)
-
+		rawCAKey, err = readInput("ca-key", *sf.caKeyPath, &claims)
 		if err != nil {
 			return fmt.Errorf("error while reading ca-key: %s", err)
 		}
@@ -116,26 +141,28 @@ 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
-			for i := 0; i < 5; i++ {
-				out.Write([]byte("Enter passphrase: "))
-				passphrase, err = pr.ReadPassword()
-
-				if errors.Is(err, ErrNoTerminal) {
-					return fmt.Errorf("ca-key is encrypted and must be decrypted interactively")
-				} else if err != nil {
-					return fmt.Errorf("error reading password: %s", err)
-				}
-
-				if len(passphrase) > 0 {
-					break
-				}
-			}
+			passphrase = []byte(os.Getenv("NEBULA_CA_PASSPHRASE"))
 			if len(passphrase) == 0 {
-				return fmt.Errorf("cannot open encrypted ca-key without passphrase")
-			}
+				// ask for a passphrase until we get one
+				for i := 0; i < 5; i++ {
+					errOut.Write([]byte("Enter passphrase: "))
+					passphrase, err = pr.ReadPassword()

+					if errors.Is(err, ErrNoTerminal) {
+						return fmt.Errorf("ca-key is encrypted and must be decrypted interactively")
+					} else if err != nil {
+						return fmt.Errorf("error reading password: %s", err)
+					}
+
+					if len(passphrase) > 0 {
+						break
+					}
+				}
+				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)
@@ -145,7 +172,7 @@ func signCert(args []string, out io.Writer, errOut io.Writer, pr PasswordReader)
 		}
 	}

-	rawCACert, err := os.ReadFile(*sf.caCertPath)
+	rawCACert, err := readInput("ca-crt", *sf.caCertPath, &claims)
 	if err != nil {
 		return fmt.Errorf("error while reading ca-crt: %s", err)
 	}
@@ -165,6 +192,10 @@ 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
@@ -239,7 +270,7 @@ func signCert(args []string, out io.Writer, errOut io.Writer, pr PasswordReader)

 	if *sf.inPubPath != "" {
 		var pubCurve cert.Curve
-		rawPub, err := os.ReadFile(*sf.inPubPath)
+		rawPub, err := readInput("in-pub", *sf.inPubPath, &claims)
 		if err != nil {
 			return fmt.Errorf("error while reading in-pub: %s", err)
 		}
@@ -260,16 +291,10 @@ func signCert(args []string, out io.Writer, errOut io.Writer, pr PasswordReader)
 		pub, rawPriv = newKeypair(curve)
 	}

-	if *sf.outKeyPath == "" {
-		*sf.outKeyPath = *sf.name + ".key"
-	}
-
-	if *sf.outCertPath == "" {
-		*sf.outCertPath = *sf.name + ".crt"
-	}
-
-	if _, err := os.Stat(*sf.outCertPath); err == nil {
-		return fmt.Errorf("refusing to overwrite existing cert: %s", *sf.outCertPath)
+	if !isStdio(*sf.outCertPath) {
+		if _, err := os.Stat(*sf.outCertPath); err == nil {
+			return fmt.Errorf("refusing to overwrite existing cert: %s", *sf.outCertPath)
+		}
 	}

 	var crts []cert.Certificate
@@ -277,21 +302,19 @@ 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 {
-		// Make sure we at least have an ip
+	switch version {
+	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")
-			}
+		if len(v6Networks) > 0 {
+			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")
-			}
+		if len(v6UnsafeNetworks) > 0 {
+			return newHelpErrorf("invalid -unsafe-networks definition: v1 certificates can only contain ipv4 addresses")
 		}

 		t := &cert.TBSCertificate{
@@ -321,9 +344,8 @@ 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,
@@ -351,14 +373,19 @@ 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 == "" {
-		if _, err := os.Stat(*sf.outKeyPath); err == nil {
-			return fmt.Errorf("refusing to overwrite existing key: %s", *sf.outKeyPath)
+		if !isStdio(*sf.outKeyPath) {
+			if _, err := os.Stat(*sf.outKeyPath); err == nil {
+				return fmt.Errorf("refusing to overwrite existing key: %s", *sf.outKeyPath)
+			}
 		}

-		err = os.WriteFile(*sf.outKeyPath, cert.MarshalPrivateKeyToPEM(curve, rawPriv), 0600)
+		err = writeOutput(*sf.outKeyPath, cert.MarshalPrivateKeyToPEM(curve, rawPriv), 0600, out)
 		if err != nil {
 			return fmt.Errorf("error while writing out-key: %s", err)
 		}
@@ -373,7 +400,7 @@ func signCert(args []string, out io.Writer, errOut io.Writer, pr PasswordReader)
 		b = append(b, sb...)
 	}

-	err = os.WriteFile(*sf.outCertPath, b, 0600)
+	err = writeOutput(*sf.outCertPath, b, 0600, out)
 	if err != nil {
 		return fmt.Errorf("error while writing out-crt: %s", err)
 	}
@@ -384,7 +411,7 @@ func signCert(args []string, out io.Writer, errOut io.Writer, pr PasswordReader)
 			return fmt.Errorf("error while generating qr code: %s", err)
 		}

-		err = os.WriteFile(*sf.outQRPath, b, 0600)
+		err = writeOutput(*sf.outQRPath, b, 0600, out)
 		if err != nil {
 			return fmt.Errorf("error while writing out-qr: %s", err)
 		}
@@ -434,6 +461,7 @@ func signSummary() string {
 func signHelp(out io.Writer) {
 	sf := newSignFlags()
 	out.Write([]byte("Usage of " + os.Args[0] + " " + signSummary() + "\n"))
+	out.Write([]byte(stdioHelpText))
 	sf.set.SetOutput(out)
 	sf.set.PrintDefaults()
 }
@@ -27,6 +27,7 @@ func Test_signHelp(t *testing.T) {
 	assert.Equal(
 		t,
 		"Usage of "+os.Args[0]+" sign <flags>: create and sign a certificate\n"+
+			"  Pass \"-\" to any path flag to read from stdin or write to stdout.\n"+
 			"  -ca-crt string\n"+
 			"    \tOptional: path to the signing CA cert (default \"ca.crt\")\n"+
 			"  -ca-key string\n"+
@@ -55,7 +56,7 @@ 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 +205,7 @@ 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())

@@ -376,8 +377,20 @@ func Test_signCert(t *testing.T) {
 	// test with the proper 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.NoError(t, signCert(args, ob, eb, testpw))
-	assert.Equal(t, "Enter passphrase: ", ob.String())
+	assert.Empty(t, ob.String())
+	assert.Equal(t, "Enter passphrase: ", 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))
+	ob.Reset()
+	eb.Reset()
+	require.NoError(t, signCert(args, ob, eb, testpw))
+	assert.Empty(t, ob.String())
 	assert.Empty(t, eb.String())
+	os.Setenv("NEBULA_CA_PASSPHRASE", "")

 	// test with the wrong password
 	ob.Reset()
@@ -386,8 +399,19 @@ func Test_signCert(t *testing.T) {
 	testpw.password = []byte("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.Error(t, signCert(args, ob, eb, testpw))
-	assert.Equal(t, "Enter passphrase: ", ob.String())
+	assert.Empty(t, ob.String())
+	assert.Equal(t, "Enter passphrase: ", 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()
@@ -396,8 +420,8 @@ func Test_signCert(t *testing.T) {
 	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.Error(t, signCert(args, ob, eb, nopw))
 	// normally the user hitting enter on the prompt would add newlines between these
-	assert.Equal(t, "Enter passphrase: Enter passphrase: Enter passphrase: Enter passphrase: Enter passphrase: ", ob.String())
-	assert.Empty(t, eb.String())
+	assert.Empty(t, ob.String())
+	assert.Equal(t, "Enter passphrase: Enter passphrase: Enter passphrase: Enter passphrase: Enter passphrase: ", eb.String())

 	// test an error condition
 	ob.Reset()
@@ -405,6 +429,106 @@ func Test_signCert(t *testing.T) {

 	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.Error(t, signCert(args, ob, eb, errpw))
-	assert.Equal(t, "Enter passphrase: ", ob.String())
-	assert.Empty(t, eb.String())
+	assert.Empty(t, ob.String())
+	assert.Equal(t, "Enter passphrase: ", eb.String())
+}
+
+func Test_signCert_stdio(t *testing.T) {
+	nopw := &StubPasswordReader{
+		password: []byte(""),
+		err:      nil,
+	}
+
+	caPub, caPriv, _ := ed25519.GenerateKey(rand.Reader)
+	rawCAKey := cert.MarshalSigningPrivateKeyToPEM(cert.Curve_CURVE25519, caPriv)
+
+	ca, _ := NewTestCaCert("ca", caPub, caPriv, time.Now(), time.Now().Add(time.Minute*200), nil, nil, nil)
+	rawCACrt, _ := ca.MarshalPEM()
+
+	caCrtF, err := os.CreateTemp("", "sign-cert.crt")
+	require.NoError(t, err)
+	defer os.Remove(caCrtF.Name())
+	caCrtF.Write(rawCACrt)
+
+	caKeyF, err := os.CreateTemp("", "sign-cert.key")
+	require.NoError(t, err)
+	defer os.Remove(caKeyF.Name())
+	caKeyF.Write(rawCAKey)
+
+	keyF, err := os.CreateTemp("", "sign.key")
+	require.NoError(t, err)
+	os.Remove(keyF.Name())
+	defer os.Remove(keyF.Name())
+
+	// ca-key on stdin, cert to stdout
+	withStdin(t, bytes.NewReader(rawCAKey))
+	ob := &bytes.Buffer{}
+	eb := &bytes.Buffer{}
+	args := []string{"-version", "1", "-ca-crt", caCrtF.Name(), "-ca-key", "-", "-name", "stdin-test", "-ip", "1.1.1.1/24", "-out-crt", "-", "-out-key", keyF.Name(), "-duration", "100m"}
+	require.NoError(t, signCert(args, ob, eb, nopw))
+	assert.Empty(t, eb.String())
+
+	lCrt, _, err := cert.UnmarshalCertificateFromPEM(ob.Bytes())
+	require.NoError(t, err)
+	assert.Equal(t, "stdin-test", lCrt.Name())
+	assert.True(t, lCrt.CheckSignature(caPub))
+
+	// two flags reading from stdin should error before any read attempt;
+	// otherwise an interactive shell would hang on io.ReadAll
+	stdinIn := bytes.NewReader(rawCAKey)
+	withStdin(t, stdinIn)
+	ob.Reset()
+	eb.Reset()
+	args = []string{"-version", "1", "-ca-crt", "-", "-ca-key", "-", "-name", "stdin-test", "-ip", "1.1.1.1/24", "-out-crt", "nope", "-out-key", "nope", "-duration", "100m"}
+	require.EqualError(t, signCert(args, ob, eb, nopw),
+		`-ca-key and -ca-crt both set to "-", only one input may read from stdin`)
+	assert.Equal(t, len(rawCAKey), stdinIn.Len(), "stdin should be untouched when conflict is caught up front")
+
+	// two flags writing to stdout should error before any output is written
+	// AND before stdin is consumed
+	stdinR := bytes.NewReader(rawCAKey)
+	withStdin(t, stdinR)
+	ob.Reset()
+	eb.Reset()
+	args = []string{"-version", "1", "-ca-crt", caCrtF.Name(), "-ca-key", "-", "-name", "stdin-test", "-ip", "1.1.1.1/24", "-out-crt", "-", "-out-key", "-", "-duration", "100m"}
+	require.EqualError(t, signCert(args, ob, eb, nopw),
+		`-out-key and -out-crt both set to "-", only one output may write to stdout`)
+	assert.Empty(t, ob.String())
+	// stdin should be untouched because the conflict was caught up front
+	assert.Equal(t, len(rawCAKey), stdinR.Len())
+
+	// out-key on stdout, cert on disk
+	keyF2, err := os.CreateTemp("", "sign.key")
+	require.NoError(t, err)
+	os.Remove(keyF2.Name())
+	defer os.Remove(keyF2.Name())
+	crtF, err := os.CreateTemp("", "sign.crt")
+	require.NoError(t, err)
+	os.Remove(crtF.Name())
+	defer os.Remove(crtF.Name())
+
+	ob.Reset()
+	eb.Reset()
+	args = []string{"-version", "1", "-ca-crt", caCrtF.Name(), "-ca-key", caKeyF.Name(), "-name", "stdin-test", "-ip", "1.1.1.1/24", "-out-crt", crtF.Name(), "-out-key", "-", "-duration", "100m"}
+	require.NoError(t, signCert(args, ob, eb, nopw))
+	assert.Empty(t, eb.String())
+	_, _, curve, err := cert.UnmarshalPrivateKeyFromPEM(ob.Bytes())
+	require.NoError(t, err)
+	assert.Equal(t, cert.Curve_CURVE25519, curve)
+
+	// in-pub on stdin (caller already has a keypair, only the cert is generated)
+	inPub, _ := x25519Keypair()
+	rawInPub := cert.MarshalPublicKeyToPEM(cert.Curve_CURVE25519, inPub)
+
+	withStdin(t, bytes.NewReader(rawInPub))
+	os.Remove(crtF.Name())
+	ob.Reset()
+	eb.Reset()
+	args = []string{"-version", "1", "-ca-crt", caCrtF.Name(), "-ca-key", caKeyF.Name(), "-name", "in-pub-test", "-ip", "1.1.1.1/24", "-in-pub", "-", "-out-crt", "-", "-duration", "100m"}
+	require.NoError(t, signCert(args, ob, eb, nopw))
+	assert.Empty(t, eb.String())
+	stdinCrt, _, err := cert.UnmarshalCertificateFromPEM(ob.Bytes())
+	require.NoError(t, err)
+	assert.Equal(t, "in-pub-test", stdinCrt.Name())
+	assert.Equal(t, inPub, stdinCrt.PublicKey())
 }
@@ -0,0 +1,117 @@
+package main
+
+import (
+	"fmt"
+	"io"
+	"os"
+)
+
+// stdioPath is the special path value that selects stdin (for inputs) or
+// stdout (for outputs) instead of a file on disk.
+const stdioPath = "-"
+
+// stdioHelpText is rendered just under the Usage line of each subcommand
+// help so the - convention is documented once instead of on every flag.
+const stdioHelpText = "  Pass \"-\" to any path flag to read from stdin or write to stdout.\n"
+
+// stdinReader is the source used when an input flag is set to "-".
+// It is a package level var so tests can swap in a deterministic reader.
+// Tests that mutate stdinReader cannot run with t.Parallel().
+var stdinReader io.Reader = os.Stdin
+
+// ioClaims tracks which flags have claimed stdin and stdout during a single
+// command invocation so we can refuse a second flag asking for the same
+// stream.
+type ioClaims struct {
+	in  string
+	out string
+}
+
+func (c *ioClaims) claimIn(flagName string) error {
+	if c.in != "" && c.in != flagName {
+		return fmt.Errorf("-%s and -%s both set to %q, only one input may read from stdin", c.in, flagName, stdioPath)
+	}
+	c.in = flagName
+	return nil
+}
+
+func (c *ioClaims) claimOut(flagName string) error {
+	if c.out != "" && c.out != flagName {
+		return fmt.Errorf("-%s and -%s both set to %q, only one output may write to stdout", c.out, flagName, stdioPath)
+	}
+	c.out = flagName
+	return nil
+}
+
+// reserveInputs walks alternating (flagName, path) pairs and claims stdin
+// for any path equal to stdioPath. It must be called before any input is
+// read so a conflict can be reported immediately instead of blocking on
+// io.ReadAll while waiting for input that will never arrive.
+func reserveInputs(claims *ioClaims, pairs ...string) error {
+	return reserveStdio(claims, "reserveInputs", (*ioClaims).claimIn, pairs)
+}
+
+// reserveOutputs walks alternating (flagName, path) pairs and claims stdout
+// for any path equal to stdioPath. It must be called before any output is
+// written so a conflict cannot leave one stream half written before the
+// second flag fails.
+func reserveOutputs(claims *ioClaims, pairs ...string) error {
+	return reserveStdio(claims, "reserveOutputs", (*ioClaims).claimOut, pairs)
+}
+
+func reserveStdio(claims *ioClaims, who string, claim func(*ioClaims, string) error, pairs []string) error {
+	if len(pairs)%2 != 0 {
+		panic(who + " requires alternating name, path pairs")
+	}
+	for i := 0; i < len(pairs); i += 2 {
+		name, path := pairs[i], pairs[i+1]
+		if path != stdioPath {
+			continue
+		}
+		if err := claim(claims, name); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+// readInput returns the bytes referenced by path, reading from stdin when
+// path is stdioPath.
+func readInput(flagName, path string, claims *ioClaims) ([]byte, error) {
+	if path == stdioPath {
+		if err := claims.claimIn(flagName); err != nil {
+			return nil, err
+		}
+		return io.ReadAll(stdinReader)
+	}
+	return os.ReadFile(path)
+}
+
+// openInput returns a reader for path. When path is stdioPath the returned
+// reader wraps stdin and Close is a no-op.
+func openInput(flagName, path string, claims *ioClaims) (io.ReadCloser, error) {
+	if path == stdioPath {
+		if err := claims.claimIn(flagName); err != nil {
+			return nil, err
+		}
+		return io.NopCloser(stdinReader), nil
+	}
+	return os.Open(path)
+}
+
+// writeOutput writes data to path, or to stdout when path is stdioPath. perm
+// is only used for file output. The caller must have already claimed stdout
+// via reserveOutputs before invoking with stdioPath.
+func writeOutput(path string, data []byte, perm os.FileMode, stdout io.Writer) error {
+	if path == stdioPath {
+		_, err := stdout.Write(data)
+		return err
+	}
+	return os.WriteFile(path, data, perm)
+}
+
+// isStdio reports whether path is the stdio sentinel and so should skip
+// existence checks like "refuse to overwrite".
+func isStdio(path string) bool {
+	return path == stdioPath
+}
@@ -0,0 +1,167 @@
+package main
+
+import (
+	"bytes"
+	"io"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// withStdin temporarily replaces stdinReader for the duration of t.
+func withStdin(t *testing.T, r io.Reader) {
+	t.Helper()
+	prev := stdinReader
+	stdinReader = r
+	t.Cleanup(func() { stdinReader = prev })
+}
+
+func Test_readInput_stdin(t *testing.T) {
+	withStdin(t, bytes.NewBufferString("hello"))
+	var claims ioClaims
+
+	got, err := readInput("path", "-", &claims)
+	require.NoError(t, err)
+	assert.Equal(t, []byte("hello"), got)
+	assert.Equal(t, "path", claims.in)
+}
+
+func Test_readInput_file(t *testing.T) {
+	dir := t.TempDir()
+	p := filepath.Join(dir, "f")
+	require.NoError(t, os.WriteFile(p, []byte("file"), 0600))
+	var claims ioClaims
+
+	got, err := readInput("path", p, &claims)
+	require.NoError(t, err)
+	assert.Equal(t, []byte("file"), got)
+	assert.Empty(t, claims.in)
+}
+
+func Test_readInput_doubleStdinErrors(t *testing.T) {
+	withStdin(t, bytes.NewBufferString("hello"))
+	var claims ioClaims
+
+	_, err := readInput("ca-key", "-", &claims)
+	require.NoError(t, err)
+
+	_, err = readInput("ca-crt", "-", &claims)
+	require.EqualError(t, err, `-ca-key and -ca-crt both set to "-", only one input may read from stdin`)
+}
+
+func Test_openInput_stdin(t *testing.T) {
+	withStdin(t, bytes.NewBufferString("hi"))
+	var claims ioClaims
+
+	r, err := openInput("ca", "-", &claims)
+	require.NoError(t, err)
+	defer r.Close()
+	b, err := io.ReadAll(r)
+	require.NoError(t, err)
+	assert.Equal(t, []byte("hi"), b)
+}
+
+func Test_openInput_doubleStdinErrors(t *testing.T) {
+	withStdin(t, bytes.NewBufferString("hi"))
+	var claims ioClaims
+
+	r, err := openInput("ca", "-", &claims)
+	require.NoError(t, err)
+	r.Close()
+
+	_, err = openInput("crt", "-", &claims)
+	require.EqualError(t, err, `-ca and -crt both set to "-", only one input may read from stdin`)
+}
+
+func Test_writeOutput_stdout(t *testing.T) {
+	out := &bytes.Buffer{}
+
+	err := writeOutput("-", []byte("payload"), 0600, out)
+	require.NoError(t, err)
+	assert.Equal(t, "payload", out.String())
+}
+
+func Test_writeOutput_file(t *testing.T) {
+	dir := t.TempDir()
+	p := filepath.Join(dir, "f")
+	out := &bytes.Buffer{}
+
+	err := writeOutput(p, []byte("payload"), 0600, out)
+	require.NoError(t, err)
+	assert.Empty(t, out.String())
+	got, err := os.ReadFile(p)
+	require.NoError(t, err)
+	assert.Equal(t, []byte("payload"), got)
+}
+
+func Test_reserveOutputs_noConflict(t *testing.T) {
+	var claims ioClaims
+	require.NoError(t, reserveOutputs(&claims,
+		"out-key", "/tmp/key",
+		"out-crt", "-",
+		"out-qr", "",
+	))
+	assert.Equal(t, "out-crt", claims.out)
+}
+
+func Test_reserveOutputs_conflict(t *testing.T) {
+	var claims ioClaims
+	err := reserveOutputs(&claims,
+		"out-key", "-",
+		"out-crt", "-",
+	)
+	require.EqualError(t, err, `-out-key and -out-crt both set to "-", only one output may write to stdout`)
+}
+
+func Test_reserveOutputs_panicsOnOddPairs(t *testing.T) {
+	defer func() {
+		r := recover()
+		require.NotNil(t, r)
+	}()
+	var claims ioClaims
+	_ = reserveOutputs(&claims, "out-key")
+}
+
+func Test_reserveInputs_noConflict(t *testing.T) {
+	var claims ioClaims
+	require.NoError(t, reserveInputs(&claims,
+		"ca-key", "/tmp/ca.key",
+		"ca-crt", "-",
+		"in-pub", "",
+	))
+	assert.Equal(t, "ca-crt", claims.in)
+}
+
+func Test_reserveInputs_conflict(t *testing.T) {
+	var claims ioClaims
+	err := reserveInputs(&claims,
+		"ca-key", "-",
+		"ca-crt", "-",
+	)
+	require.EqualError(t, err, `-ca-key and -ca-crt both set to "-", only one input may read from stdin`)
+}
+
+func Test_claimIn_idempotent(t *testing.T) {
+	// pre-claim then a lazy re-claim of the same flag should be a no-op
+	var claims ioClaims
+	require.NoError(t, claims.claimIn("ca-key"))
+	require.NoError(t, claims.claimIn("ca-key"))
+	assert.Equal(t, "ca-key", claims.in)
+}
+
+func Test_claimOut_idempotent(t *testing.T) {
+	var claims ioClaims
+	require.NoError(t, claims.claimOut("out-crt"))
+	require.NoError(t, claims.claimOut("out-crt"))
+	assert.Equal(t, "out-crt", claims.out)
+}
+
+func Test_isStdio(t *testing.T) {
+	assert.True(t, isStdio("-"))
+	assert.False(t, isStdio(""))
+	assert.False(t, isStdio("./-"))
+	assert.False(t, isStdio("foo"))
+}
@@ -6,7 +6,6 @@ import (
 	"fmt"
 	"io"
 	"os"
-	"strings"
 	"time"

 	"github.com/slackhq/nebula/cert"
@@ -40,24 +39,26 @@ func verify(args []string, out io.Writer, errOut io.Writer) error {
 		return err
 	}

-	rawCACert, err := os.ReadFile(*vf.caPath)
+	var claims ioClaims
+	if err := reserveInputs(&claims,
+		"ca", *vf.caPath,
+		"crt", *vf.certPath,
+	); err != nil {
+		return err
+	}
+
+	caReader, err := openInput("ca", *vf.caPath, &claims)
 	if err != nil {
 		return fmt.Errorf("error while reading ca: %w", err)
 	}
+	defer caReader.Close()

-	caPool := cert.NewCAPool()
-	for {
-		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
-		}
+	caPool, err := cert.NewCAPoolFromPEMReader(caReader)
+	if err != nil && !errors.Is(err, cert.ErrExpired) {
+		return fmt.Errorf("error while adding ca cert to pool: %w", err)
 	}

-	rawCert, err := os.ReadFile(*vf.certPath)
+	rawCert, err := readInput("crt", *vf.certPath, &claims)
 	if err != nil {
 		return fmt.Errorf("unable to read crt: %w", err)
 	}
@@ -92,6 +93,7 @@ func verifySummary() string {
 func verifyHelp(out io.Writer) {
 	vf := newVerifyFlags()
 	_, _ = out.Write([]byte("Usage of " + os.Args[0] + " " + verifySummary() + "\n"))
+	_, _ = out.Write([]byte(stdioHelpText))
 	vf.set.SetOutput(out)
 	vf.set.PrintDefaults()
 }
@@ -23,6 +23,7 @@ func Test_verifyHelp(t *testing.T) {
 	assert.Equal(
 		t,
 		"Usage of "+os.Args[0]+" verify <flags>: verifies a certificate isn't expired and was signed by a trusted authority.\n"+
+			"  Pass \"-\" to any path flag to read from stdin or write to stdout.\n"+
 			"  -ca string\n"+
 			"    \tRequired: path to a file containing one or more ca certificates\n"+
 			"  -crt string\n"+
@@ -64,7 +65,7 @@ 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)
@@ -122,3 +123,46 @@ func Test_verify(t *testing.T) {
 	assert.Empty(t, eb.String())
 	require.NoError(t, err)
 }
+
+func Test_verify_stdio(t *testing.T) {
+	ob := &bytes.Buffer{}
+	eb := &bytes.Buffer{}
+
+	caPub, caPriv, _ := ed25519.GenerateKey(rand.Reader)
+	ca, _ := NewTestCaCert("test-ca", caPub, caPriv, time.Now().Add(time.Hour*-1), time.Now().Add(time.Hour*2), nil, nil, nil)
+	caPEM, _ := ca.MarshalPEM()
+
+	crt, _ := NewTestCert(ca, caPriv, "test-cert", time.Now().Add(time.Hour*-1), time.Now().Add(time.Hour), nil, nil, nil)
+	crtPEM, _ := crt.MarshalPEM()
+
+	caFile, err := os.CreateTemp("", "verify-ca")
+	require.NoError(t, err)
+	defer os.Remove(caFile.Name())
+	caFile.Write(caPEM)
+
+	// crt on stdin, ca on disk
+	withStdin(t, bytes.NewReader(crtPEM))
+	require.NoError(t, verify([]string{"-ca", caFile.Name(), "-crt", "-"}, ob, eb))
+	assert.Empty(t, ob.String())
+	assert.Empty(t, eb.String())
+
+	// ca on stdin, crt on disk
+	certFile, err := os.CreateTemp("", "verify-cert")
+	require.NoError(t, err)
+	defer os.Remove(certFile.Name())
+	certFile.Write(crtPEM)
+
+	withStdin(t, bytes.NewReader(caPEM))
+	ob.Reset()
+	eb.Reset()
+	require.NoError(t, verify([]string{"-ca", "-", "-crt", certFile.Name()}, ob, eb))
+	assert.Empty(t, ob.String())
+	assert.Empty(t, eb.String())
+
+	// both flags on stdin should error
+	withStdin(t, bytes.NewReader(caPEM))
+	ob.Reset()
+	eb.Reset()
+	require.EqualError(t, verify([]string{"-ca", "-", "-crt", "-"}, ob, eb),
+		`-ca and -crt both set to "-", only one input may read from stdin`)
+}
@@ -3,8 +3,15 @@

 package main

-import "github.com/sirupsen/logrus"
+import (
+	"log/slog"
+	"os"

-func HookLogger(l *logrus.Logger) {
-	// Do nothing, let the logs flow to stdout/stderr
+	"github.com/slackhq/nebula/logging"
+)
+
+// 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)
 }
@@ -1,54 +1,86 @@
 package main

 import (
-	"fmt"
-	"io/ioutil"
-	"os"
+	"context"
+	"log/slog"
+	"strings"
+	"sync"

-	"github.com/kardianos/service"
-	"github.com/sirupsen/logrus"
+	"github.com/slackhq/nebula/logging"
 )

-// HookLogger routes the logrus logs through the service logger so that they end up in the Windows Event Viewer
-// logrus output will be discarded
-func HookLogger(l *logrus.Logger) {
-	l.AddHook(newLogHook(logger))
-	l.SetOutput(ioutil.Discard)
+// newPlatformLogger returns a *slog.Logger that routes every log record
+// through the Windows service logger so records end up in the Windows
+// Event Log. All the heavy lifting (level management, format swap,
+// timestamp toggle, WithAttrs/WithGroup) comes from logging.NewHandler;
+// 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 {
-	sl service.Logger
+// eventLogWriter forwards slog-formatted lines to the Windows service
+// 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 {
-	return &logHook{sl: sl}
-}
-
-func (h *logHook) Fire(entry *logrus.Entry) error {
-	line, err := entry.String()
-	if err != nil {
-		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)
+func (w *eventLogWriter) Write(p []byte) (int, error) {
+	line := strings.TrimRight(string(p), "\n")
+	switch {
+	case w.level >= slog.LevelError:
+		return len(p), logger.Error(line)
+	case w.level >= slog.LevelWarn:
+		return len(p), logger.Warning(line)
 	default:
-		return nil
+		return len(p), logger.Info(line)
 	}
 }

-func (h *logHook) Levels() []logrus.Level {
-	return logrus.AllLevels
+// severityTag embeds *logging.Handler to pick up everything it does for
+// 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}
 }
@@ -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")
@@ -37,20 +50,30 @@ func main() {
 		os.Exit(0)
 	}

+	l := logging.NewLogger(os.Stdout)
+
 	if *serviceFlag != "" {
-		doService(configPath, configTest, Build, serviceFlag)
-		os.Exit(1)
+		if *configTest {
+			fmt.Println("-test is not supported with -service, run the config test without -service")
+			os.Exit(1)
+		}
+
+		if err := doService(configPath, Build, serviceFlag); err != nil {
+			l.Error("Service command failed", "error", err)
+			os.Exit(1)
+		}
+		return
 	}

 	if *configPath == "" {
-		fmt.Println("-config flag must be set")
-		flag.Usage()
-		os.Exit(1)
+		p, err := config.DefaultPath()
+		if err != nil {
+			fmt.Println(err)
+			os.Exit(1)
+		}
+		*configPath = p
 	}

-	l := logrus.New()
-	l.Out = os.Stdout
-
 	c := config.NewC(l)
 	err := c.Load(*configPath)
 	if err != nil {
@@ -58,6 +81,16 @@ 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 +98,19 @@ func main() {
 	}

 	if !*configTest {
-		ctrl.Start()
-		ctrl.ShutdownBlock()
+		if err := ctrl.Start(); err != nil {
+			util.LogWithContextIfNeeded("Error while running", err, l)
+			os.Exit(1)
+		}
+
+		go ctrl.ShutdownBlock()
+
+		if err := ctrl.Wait(); err != nil {
+			l.Error("Nebula stopped due to fatal error", "error", err)
+			os.Exit(2)
+		}
+
+		l.Info("Goodbye")
 	}

 	os.Exit(0)
@@ -4,19 +4,17 @@ import (
 	"fmt"
 	"log"
 	"os"
-	"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

 type program struct {
 	configPath *string
-	configTest *bool
 	build      string
 	control    *nebula.Control
 }
@@ -25,8 +23,7 @@ func (p *program) Start(s service.Service) error {
 	// Start should not block.
 	logger.Info("Nebula service starting.")

-	l := logrus.New()
-	HookLogger(l)
+	l := newPlatformLogger()

 	c := config.NewC(l)
 	err := c.Load(*p.configPath)
@@ -34,39 +31,56 @@ func (p *program) Start(s service.Service) error {
 		return fmt.Errorf("failed to load config: %s", err)
 	}

-	p.control, err = nebula.Main(c, *p.configTest, Build, l, nil)
+	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, false, Build, l, nil)
 	if err != nil {
 		return err
 	}

-	p.control.Start()
+	if err := p.control.Start(); err != nil {
+		return err
+	}
+
+	// Nebula can stop itself on a fatal packet reader error, make sure to log it if it happens.
+	go func() {
+		if err := p.control.Wait(); err != nil {
+			logger.Error(fmt.Sprintf("Nebula stopped due to fatal error: %v", err))
+			os.Exit(2)
+		}
+	}()
+
 	return nil
 }

 func (p *program) Stop(s service.Service) error {
 	logger.Info("Nebula service stopping.")
+	if p.control == nil {
+		return nil
+	}
+
 	p.control.Stop()
+
+	// block until nebula has fully drained before reporting stopped.
+	// error logging is handled by Start.
+	_ = p.control.Wait()
 	return nil
 }

-func fileExists(filename string) bool {
-	_, err := os.Stat(filename)
-	if os.IsNotExist(err) {
-		return false
-	}
-	return true
-}
-
-func doService(configPath *string, configTest *bool, build string, serviceFlag *string) {
+func doService(configPath *string, build string, serviceFlag *string) error {
 	if *configPath == "" {
-		ex, err := os.Executable()
+		p, err := config.DefaultPath()
 		if err != nil {
-			panic(err)
-		}
-		*configPath = filepath.Dir(ex) + "/config.yaml"
-		if !fileExists(*configPath) {
-			*configPath = filepath.Dir(ex) + "/config.yml"
+			return err
 		}
+		*configPath = p
 	}

 	svcConfig := &service.Config{
@@ -78,23 +92,22 @@ func doService(configPath *string, configTest *bool, build string, serviceFlag *

 	prg := &program{
 		configPath: configPath,
-		configTest: configTest,
 		build:      build,
 	}

 	// 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 +122,17 @@ func doService(configPath *string, configTest *bool, build string, serviceFlag *

 	switch *serviceFlag {
 	case "run":
-		err = s.Run()
-		if err != nil {
-			// Route any errors to the system logger
+		if err := s.Run(); err != nil {
+			// Route any errors to the system logger and report the failure
 			logger.Error(err)
+			return err
 		}
 	default:
-		err := service.Control(s, *serviceFlag)
-		if err != nil {
+		if err := service.Control(s, *serviceFlag); err != nil {
 			log.Printf("Valid actions: %q\n", service.ControlAction)
-			log.Fatal(err)
+			return err
 		}
-		return
 	}

+	return nil
 }
@@ -0,0 +1,96 @@
+//go:build linux && !android && !e2e_testing
+
+package main
+
+import (
+	"fmt"
+	"net/netip"
+	"os"
+	"path/filepath"
+	"runtime"
+	"testing"
+	"time"
+
+	"github.com/slackhq/nebula"
+	"github.com/slackhq/nebula/cert"
+	cert_test "github.com/slackhq/nebula/cert_test"
+	"github.com/slackhq/nebula/config"
+	"github.com/slackhq/nebula/test"
+	"github.com/stretchr/testify/require"
+)
+
+// TestControlStopClosesOnTimer reproduces the dnclient lifecycle: nebula runs as
+// a library, and on a config update dnclient calls Stop() in-process to tear the
+// old instance down before starting a new one. This boots a real nebula (real
+// blocking UDP sockets, tun disabled), lets it run, then Stop()s it on a timer
+// and asserts it actually closes. If the reader goroutines parked in recvmmsg
+// don't wake on Close(), Wait() blocks forever and this fails with a goroutine
+// dump instead of relying on a process signal to unstick them.
+func TestControlStopClosesOnTimer(t *testing.T) {
+	l := test.NewLogger()
+	dir := t.TempDir()
+
+	before := time.Now().Add(-time.Hour)
+	after := time.Now().Add(time.Hour)
+	ca, _, caKey, caPEM := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, before, after, nil, nil, nil)
+	networks := []netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")}
+	_, _, keyPEM, certPEM := cert_test.NewTestCert(cert.Version2, cert.Curve_CURVE25519, ca, caKey, "close-on-timer", before, after, networks, nil, nil)
+
+	caPath := filepath.Join(dir, "ca.pem")
+	certPath := filepath.Join(dir, "cert.pem")
+	keyPath := filepath.Join(dir, "key.pem")
+	require.NoError(t, os.WriteFile(caPath, caPEM, 0o600))
+	require.NoError(t, os.WriteFile(certPath, certPEM, 0o600))
+	require.NoError(t, os.WriteFile(keyPath, keyPEM, 0o600))
+
+	// tun disabled so no device/root is needed; routines: 2 so we exercise the
+	// multi-socket (SO_REUSEPORT) teardown, which is where dnclient runs.
+	configBody := fmt.Sprintf(`
+pki:
+  ca: %s
+  cert: %s
+  key: %s
+listen:
+  host: 127.0.0.1
+  port: 0
+tun:
+  disabled: true
+firewall:
+  outbound:
+    - port: any
+      proto: any
+      host: any
+  inbound:
+    - port: any
+      proto: any
+      host: any
+routines: 2
+`, caPath, certPath, keyPath)
+	require.NoError(t, os.WriteFile(filepath.Join(dir, "config.yml"), []byte(configBody), 0o600))
+
+	c := config.NewC(l)
+	require.NoError(t, c.Load(dir))
+
+	ctrl, err := nebula.Main(c, false, "close-on-timer", l, nil)
+	require.NoError(t, err)
+	require.NoError(t, ctrl.Start())
+
+	// Run like a live nebula, then close on a timer, exactly as dnclient does.
+	<-time.NewTimer(5 * time.Second).C
+
+	stopped := make(chan struct{})
+	go func() {
+		ctrl.Stop() // closes the udp sockets (shutdown(2)) and the tun
+		ctrl.Wait() // blocks until every reader goroutine has returned
+		close(stopped)
+	}()
+
+	select {
+	case <-stopped:
+		t.Log("nebula closed cleanly on timer")
+	case <-time.After(10 * time.Second):
+		buf := make([]byte, 1<<20)
+		n := runtime.Stack(buf, true)
+		t.Fatalf("nebula did NOT close within 10s of Stop(): a blocking reader never woke\n%s", buf[:n])
+	}
+}
@@ -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")
@@ -37,13 +50,15 @@ func main() {
 	}

 	if *configPath == "" {
-		fmt.Println("-config flag must be set")
-		flag.Usage()
-		os.Exit(1)
+		p, err := config.DefaultPath()
+		if err != nil {
+			fmt.Println(err)
+			os.Exit(1)
+		}
+		*configPath = p
 	}

-	l := logrus.New()
-	l.Out = os.Stdout
+	l := logging.NewLogger(os.Stdout)

 	c := config.NewC(l)
 	err := c.Load(*configPath)
@@ -52,6 +67,16 @@ 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 +84,20 @@ func main() {
 	}

 	if !*configTest {
-		ctrl.Start()
+		if err := ctrl.Start(); err != nil {
+			util.LogWithContextIfNeeded("Error while running", err, l)
+			os.Exit(1)
+		}
+
+		go ctrl.ShutdownBlock()
 		notifyReady(l)
-		ctrl.ShutdownBlock()
+
+		if err := ctrl.Wait(); err != nil {
+			l.Error("Nebula stopped due to fatal error", "error", err)
+			os.Exit(2)
+		}
+
+		l.Info("Goodbye")
 	}

 	os.Exit(0)
@@ -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")
 }
@@ -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
 }
@@ -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"
-	"gopkg.in/yaml.v3"
+	"go.yaml.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
 	}

@@ -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) {
@@ -0,0 +1,29 @@
+package config
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+)
+
+// DefaultPath returns a path to a config file alongside the running executable, preferring config.yaml over config.yml.
+// If neither file exists an error is returned that names both paths checked.
+func DefaultPath() (string, error) {
+	ex, err := os.Executable()
+	if err != nil {
+		return "", err
+	}
+	return defaultPathInDir(filepath.Dir(ex))
+}
+
+func defaultPathInDir(dir string) (string, error) {
+	yamlPath := filepath.Join(dir, "config.yaml")
+	if _, err := os.Stat(yamlPath); err == nil {
+		return yamlPath, nil
+	}
+	ymlPath := filepath.Join(dir, "config.yml")
+	if _, err := os.Stat(ymlPath); err == nil {
+		return ymlPath, nil
+	}
+	return "", fmt.Errorf("no default config found at %s or %s", yamlPath, ymlPath)
+}
@@ -0,0 +1,67 @@
+package config
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestDefaultPathInDir(t *testing.T) {
+	t.Run("prefers config.yaml when both exist", func(t *testing.T) {
+		dir := t.TempDir()
+		want := filepath.Join(dir, "config.yaml")
+		other := filepath.Join(dir, "config.yml")
+		require.NoError(t, os.WriteFile(want, []byte("a: 1"), 0644))
+		require.NoError(t, os.WriteFile(other, []byte("a: 2"), 0644))
+
+		got, err := defaultPathInDir(dir)
+		require.NoError(t, err)
+		assert.Equal(t, want, got)
+	})
+
+	t.Run("returns config.yaml when only it exists", func(t *testing.T) {
+		dir := t.TempDir()
+		want := filepath.Join(dir, "config.yaml")
+		require.NoError(t, os.WriteFile(want, []byte("a: 1"), 0644))
+
+		got, err := defaultPathInDir(dir)
+		require.NoError(t, err)
+		assert.Equal(t, want, got)
+	})
+
+	t.Run("falls back to config.yml when only it exists", func(t *testing.T) {
+		dir := t.TempDir()
+		want := filepath.Join(dir, "config.yml")
+		require.NoError(t, os.WriteFile(want, []byte("a: 1"), 0644))
+
+		got, err := defaultPathInDir(dir)
+		require.NoError(t, err)
+		assert.Equal(t, want, got)
+	})
+
+	t.Run("errors when neither exists and names both paths", func(t *testing.T) {
+		dir := t.TempDir()
+		got, err := defaultPathInDir(dir)
+		assert.Empty(t, got)
+		require.Error(t, err)
+		assert.Contains(t, err.Error(), filepath.Join(dir, "config.yaml"))
+		assert.Contains(t, err.Error(), filepath.Join(dir, "config.yml"))
+	})
+}
+
+func TestDefaultPath(t *testing.T) {
+	got, err := DefaultPath()
+	if err != nil {
+		ex, exErr := os.Executable()
+		require.NoError(t, exErr)
+		assert.Contains(t, err.Error(), filepath.Dir(ex))
+		return
+	}
+	ex, err := os.Executable()
+	require.NoError(t, err)
+	assert.Equal(t, filepath.Dir(ex), filepath.Dir(got))
+	assert.Contains(t, []string{"config.yaml", "config.yml"}, filepath.Base(got))
+}
@@ -5,13 +5,12 @@ 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"
@@ -45,19 +44,16 @@ type connectionManager struct {
 	inactivityTimeout       atomic.Int64
 	dropInactive            atomic.Bool

-	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,
-		punchy:          p,
-		relayUsed:       make(map[uint32]struct{}),
-		relayUsedLock:   &sync.RWMutex{},
-		metricsTxPunchy: metrics.GetOrRegisterCounter("messages.tx.punchy", nil),
+		hostMap:       hm,
+		l:             l,
+		punchy:        p,
+		relayUsed:     make(map[uint32]struct{}),
+		relayUsedLock: &sync.RWMutex{},
 	}

 	cm.reload(c, true)
@@ -85,9 +81,10 @@ 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).
-				WithField("newDuration", cm.getInactivityTimeout()).
-				Info("Inactivity timeout has changed")
+			cm.l.Info("Inactivity timeout has changed",
+				"oldDuration", old,
+				"newDuration", cm.getInactivityTimeout(),
+			)
 		}
 	}

@@ -95,9 +92,10 @@ 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).
-				WithField("newBool", cm.dropInactive.Load()).
-				Info("Drop inactive setting has changed")
+			cm.l.Info("Drop inactive setting has changed",
+				"oldBool", old,
+				"newBool", cm.dropInactive.Load(),
+			)
 		}
 	}
 }
@@ -138,14 +136,6 @@ func (cm *connectionManager) getAndResetTrafficCheck(h *HostInfo, now time.Time)
 	return in, out
 }

-// AddTrafficWatch must be called for every new HostInfo.
-// We will continue to monitor the HostInfo until the tunnel is dropped.
-func (cm *connectionManager) AddTrafficWatch(h *HostInfo) {
-	if h.out.Swap(true) == false {
-		cm.trafficTimer.Add(h.localIndexId, cm.checkInterval)
-	}
-}
-
 func (cm *connectionManager) Start(ctx context.Context) {
 	clockSource := time.NewTicker(cm.trafficTimer.t.tickDuration)
 	defer clockSource.Stop()
@@ -256,7 +246,7 @@ 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 +294,16 @@ 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{
-				"relayFrom":           req.RelayFromAddr,
-				"relayTo":             req.RelayToAddr,
-				"initiatorRelayIndex": req.InitiatorRelayIndex,
-				"responderRelayIndex": req.ResponderRelayIndex,
-				"vpnAddrs":            newhostinfo.vpnAddrs}).
-				Info("send CreateRelayRequest")
+			cm.l.Info("send CreateRelayRequest",
+				"relayFrom", relayFrom,
+				"relayTo", relayTo,
+				"initiatorRelayIndex", req.InitiatorRelayIndex,
+				"responderRelayIndex", req.ResponderRelayIndex,
+				"vpnAddrs", newhostinfo.vpnAddrs,
+			)
 		}
 	}
 }
@@ -325,7 +315,7 @@ 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
 	}

@@ -345,16 +335,15 @@ 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 {
-			hostinfo.logger(cm.l).
-				WithField("tunnelCheck", m{"state": "alive", "method": "passive"}).
-				Debug("Tunnel status")
+		if cm.l.Enabled(context.Background(), slog.LevelDebug) {
+			hostinfo.logger(cm.l).Debug("Tunnel status",
+				"tunnelCheck", m{"state": "alive", "method": "passive"},
+			)
 		}
 		hostinfo.pendingDeletion.Store(false)

 		if mainHostInfo {
 			decision = tryRehandshake
-
 		} else {
 			if cm.shouldSwapPrimary(hostinfo) {
 				decision = swapPrimary
@@ -368,7 +357,7 @@ func (cm *connectionManager) makeTrafficDecision(localIndex uint32, now time.Tim

 		if !outTraffic {
 			// Send a punch packet to keep the NAT state alive
-			cm.sendPunch(hostinfo)
+			cm.punchy.SendPunch(hostinfo)
 		}

 		return decision, hostinfo, primary
@@ -376,9 +365,9 @@ 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).
-			WithField("tunnelCheck", m{"state": "dead", "method": "active"}).
-			Info("Tunnel status")
+		hostinfo.logger(cm.l).Info("Tunnel status",
+			"tunnelCheck", m{"state": "dead", "method": "active"},
+		)

 		return deleteTunnel, hostinfo, nil
 	}
@@ -389,40 +378,39 @@ 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).
-					WithField("inactiveDuration", inactiveFor).
-					WithField("primary", mainHostInfo).
-					Info("Dropping tunnel due to inactivity")
+				hostinfo.logger(cm.l).Info("Dropping tunnel due to inactivity",
+					"inactiveDuration", inactiveFor,
+					"primary", mainHostInfo,
+				)

 				return closeTunnel, hostinfo, primary
 			}

 			// If we aren't sending or receiving traffic then its an unused tunnel and we don't to test the tunnel.
 			// Just maintain NAT state if configured to do so.
-			cm.sendPunch(hostinfo)
+			cm.punchy.SendPunch(hostinfo)
 			cm.trafficTimer.Add(hostinfo.localIndexId, cm.checkInterval)
 			return doNothing, nil, nil
 		}

-		if cm.punchy.GetTargetEverything() {
-			// This is similar to the old punchy behavior with a slight optimization.
-			// We aren't receiving traffic but we are sending it, punch on all known
-			// ips in case we need to re-prime NAT state
-			cm.sendPunch(hostinfo)
-		}
+		// We aren't receiving traffic but we are sending it. The outbound
+		// traffic itself refreshes the primary remote's NAT state; this
+		// fans out to non-primary remotes, but only if target_all_remotes
+		// is configured.
+		cm.punchy.SendPunchToAll(hostinfo)

-		if cm.l.Level >= logrus.DebugLevel {
-			hostinfo.logger(cm.l).
-				WithField("tunnelCheck", m{"state": "testing", "method": "active"}).
-				Debug("Tunnel status")
+		if cm.l.Enabled(context.Background(), slog.LevelDebug) {
+			hostinfo.logger(cm.l).Debug("Tunnel status",
+				"tunnelCheck", m{"state": "testing", "method": "active"},
+			)
 		}

 		// 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 {
-			hostinfo.logger(cm.l).Debugf("Hostinfo sadness")
+		if cm.l.Enabled(context.Background(), slog.LevelDebug) {
+			hostinfo.logger(cm.l).Debug("Hostinfo sadness")
 		}
 	}

@@ -461,6 +449,10 @@ func (cm *connectionManager) shouldSwapPrimary(current *HostInfo) bool {
 	}

 	crt := cm.intf.pki.getCertState().getCertificate(current.ConnectionState.myCert.Version())
+	if crt == nil {
+		//my cert was reloaded away. We should definitely swap from this tunnel
+		return true
+	}
 	// If this tunnel is using the latest certificate then we should swap it to primary for a bit and see if things
 	// settle down.
 	return bytes.Equal(current.ConnectionState.myCert.Signature(), crt.Signature())
@@ -475,70 +467,84 @@ func (cm *connectionManager) swapPrimary(current, primary *HostInfo) {
 	cm.hostMap.Unlock()
 }

-// isInvalidCertificate will check if we should destroy a tunnel if pki.disconnect_invalid is true and
-// the certificate is no longer valid. Block listed certificates will skip the pki.disconnect_invalid
-// check and return true.
+// isInvalidCertificate decides if we should destroy a tunnel.
+// returns true if pki.disconnect_invalid is true and the certificate is no longer valid.
+// Blocklisted certificates will skip the pki.disconnect_invalid check and return true.
 func (cm *connectionManager) isInvalidCertificate(now time.Time, hostinfo *HostInfo) bool {
 	remoteCert := hostinfo.GetCert()
 	if remoteCert == nil {
-		return false
+		return false //don't tear down tunnels for handshakes in progress
 	}

 	caPool := cm.intf.pki.GetCAPool()
 	err := caPool.VerifyCachedCertificate(now, remoteCert)
 	if err == nil {
-		return false
-	}
-
-	if !cm.intf.disconnectInvalid.Load() && err != cert.ErrBlockListed {
+		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).Info("Remote certificate is blocked, tearing down the tunnel",
+			"error", err,
+			"fingerprint", remoteCert.Fingerprint,
+		)
+		return true
+	} else if cm.intf.disconnectInvalid.Load() {
+		hostinfo.logger(cm.l).Info("Remote certificate is no longer valid, tearing down the tunnel",
+			"error", err,
+			"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
 		return false
 	}
-
-	hostinfo.logger(cm.l).WithError(err).
-		WithField("fingerprint", remoteCert.Fingerprint).
-		Info("Remote certificate is no longer valid, tearing down the tunnel")
-
-	return true
-}
-
-func (cm *connectionManager) sendPunch(hostinfo *HostInfo) {
-	if !cm.punchy.GetPunch() {
-		// Punching is disabled
-		return
-	}
-
-	if cm.intf.lightHouse.IsAnyLighthouseAddr(hostinfo.vpnAddrs) {
-		// Do not punch to lighthouses, we assume our lighthouse update interval is good enough.
-		// In the event the update interval is not sufficient to maintain NAT state then a publicly available lighthouse
-		// would lose the ability to notify us and punchy.respond would become unreliable.
-		return
-	}
-
-	if cm.punchy.GetTargetEverything() {
-		hostinfo.remotes.ForEach(cm.hostMap.GetPreferredRanges(), func(addr netip.AddrPort, preferred bool) {
-			cm.metricsTxPunchy.Inc(1)
-			cm.intf.outside.WriteTo([]byte{1}, addr)
-		})
-
-	} else if hostinfo.remote.IsValid() {
-		cm.metricsTxPunchy.Inc(1)
-		cm.intf.outside.WriteTo([]byte{1}, hostinfo.remote)
-	}
 }

 func (cm *connectionManager) tryRehandshake(hostinfo *HostInfo) {
 	cs := cm.intf.pki.getCertState()
 	curCrt := hostinfo.ConnectionState.myCert
-	myCrt := cs.getCertificate(curCrt.Version())
-	if curCrt.Version() >= cs.initiatingVersion && bytes.Equal(curCrt.Signature(), myCrt.Signature()) == true {
-		// The current tunnel is using the latest certificate and version, no need to rehandshake.
+	curCrtVersion := curCrt.Version()
+	myCrt := cs.getCertificate(curCrtVersion)
+	if myCrt == nil {
+		cm.l.Info("Re-handshaking with remote",
+			"vpnAddrs", hostinfo.vpnAddrs,
+			"version", curCrtVersion,
+			"reason", "local certificate removed",
+		)
+		cm.intf.handshakeManager.StartHandshake(hostinfo.vpnAddrs[0], nil)
 		return
 	}
+	peerCrt := hostinfo.ConnectionState.peerCert
+	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.Info("Re-handshaking with remote",
+				"vpnAddrs", hostinfo.vpnAddrs,
+				"version", curCrtVersion,
+				"peerVersion", peerCrt.Certificate.Version(),
+				"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()
+			})
+			return
+		}
+	}
+	if !bytes.Equal(curCrt.Signature(), myCrt.Signature()) {
+		cm.l.Info("Re-handshaking with remote",
+			"vpnAddrs", hostinfo.vpnAddrs,
+			"reason", "local certificate is not current",
+		)

-	cm.l.WithField("vpnAddrs", hostinfo.vpnAddrs).
-		WithField("reason", "local certificate is not current").
-		Info("Re-handshaking with remote")
+		cm.intf.handshakeManager.StartHandshake(hostinfo.vpnAddrs[0], nil)
+		return
+	}
+	if curCrtVersion < cs.initiatingVersion {
+		cm.l.Info("Re-handshaking with remote",
+			"vpnAddrs", hostinfo.vpnAddrs,
+			"reason", "current cert version < pki.initiatingVersion",
+		)

-	cm.intf.handshakeManager.StartHandshake(hostinfo.vpnAddrs[0], nil)
+		cm.intf.handshakeManager.StartHandshake(hostinfo.vpnAddrs[0], nil)
+		return
+	}
 }
@@ -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"
@@ -25,6 +25,7 @@ func newTestLighthouse() *LightHouse {
 	lighthouses := []netip.Addr{}
 	staticList := map[netip.Addr]struct{}{}

+	lh.localAddrsFn = func(*LocalAllowList) []netip.Addr { return nil }
 	lh.lighthouses.Store(&lighthouses)
 	lh.staticList.Store(&staticList)

@@ -46,13 +47,13 @@ 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,
@@ -63,9 +64,9 @@ func Test_NewConnectionManagerTest(t *testing.T) {
 	ifce.pki.cs.Store(cs)

 	// Create manager
-	conf := config.NewC(l)
-	punchy := NewPunchyFromConfig(l, conf)
-	nc := newConnectionManagerFromConfig(l, conf, hostMap, punchy)
+	conf := config.NewC(test.NewLogger())
+	punchy := NewPunchyFromConfig(test.NewLogger(), conf, nil)
+	nc := newConnectionManagerFromConfig(test.NewLogger(), conf, hostMap, punchy)
 	nc.intf = ifce
 	p := []byte("")
 	nb := make([]byte, 12, 12)
@@ -79,7 +80,6 @@ func Test_NewConnectionManagerTest(t *testing.T) {
 	}
 	hostinfo.ConnectionState = &ConnectionState{
 		myCert: &dummyCert{version: cert.Version1},
-		H:      &noise.HandshakeState{},
 	}
 	nc.hostMap.unlockedAddHostInfo(hostinfo, ifce)

@@ -129,13 +129,13 @@ 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,
@@ -146,9 +146,9 @@ func Test_NewConnectionManagerTest2(t *testing.T) {
 	ifce.pki.cs.Store(cs)

 	// Create manager
-	conf := config.NewC(l)
-	punchy := NewPunchyFromConfig(l, conf)
-	nc := newConnectionManagerFromConfig(l, conf, hostMap, punchy)
+	conf := config.NewC(test.NewLogger())
+	punchy := NewPunchyFromConfig(test.NewLogger(), conf, nil)
+	nc := newConnectionManagerFromConfig(test.NewLogger(), conf, hostMap, punchy)
 	nc.intf = ifce
 	p := []byte("")
 	nb := make([]byte, 12, 12)
@@ -162,7 +162,6 @@ func Test_NewConnectionManagerTest2(t *testing.T) {
 	}
 	hostinfo.ConnectionState = &ConnectionState{
 		myCert: &dummyCert{version: cert.Version1},
-		H:      &noise.HandshakeState{},
 	}
 	nc.hostMap.unlockedAddHostInfo(hostinfo, ifce)

@@ -214,13 +213,13 @@ 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 +230,12 @@ 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)
-	nc := newConnectionManagerFromConfig(l, conf, hostMap, punchy)
+	punchy := NewPunchyFromConfig(test.NewLogger(), conf, nil)
+	nc := newConnectionManagerFromConfig(test.NewLogger(), conf, hostMap, punchy)
 	assert.True(t, nc.dropInactive.Load())
 	nc.intf = ifce

@@ -248,7 +247,6 @@ func Test_NewConnectionManager_DisconnectInactive(t *testing.T) {
 	}
 	hostinfo.ConnectionState = &ConnectionState{
 		myCert: &dummyCert{version: cert.Version1},
-		H:      &noise.HandshakeState{},
 	}
 	nc.hostMap.unlockedAddHostInfo(hostinfo, ifce)

@@ -339,15 +337,15 @@ func Test_NewConnectionManagerTest_DisconnectInvalid(t *testing.T) {
 	cachedPeerCert, err := ncp.VerifyCertificate(now.Add(time.Second), peerCert)

 	cs := &CertState{
-		privateKey:       []byte{},
-		v1Cert:           &dummyCert{},
-		v1HandshakeBytes: []byte{},
+		privateKey:   []byte{},
+		v1Cert:       &dummyCert{},
+		v1Credential: nil,
 	}

 	lh := newTestLighthouse()
 	ifce := &Interface{
 		hostMap:          hostMap,
-		inside:           &test.NoopTun{},
+		inside:           &overlaytest.NoopTun{},
 		outside:          &udp.NoopConn{},
 		firewall:         &Firewall{},
 		lightHouse:       lh,
@@ -360,9 +358,9 @@ func Test_NewConnectionManagerTest_DisconnectInvalid(t *testing.T) {
 	ifce.disconnectInvalid.Store(true)

 	// Create manager
-	conf := config.NewC(l)
-	punchy := NewPunchyFromConfig(l, conf)
-	nc := newConnectionManagerFromConfig(l, conf, hostMap, punchy)
+	conf := config.NewC(test.NewLogger())
+	punchy := NewPunchyFromConfig(test.NewLogger(), conf, nil)
+	nc := newConnectionManagerFromConfig(test.NewLogger(), conf, hostMap, punchy)
 	nc.intf = ifce
 	ifce.connectionManager = nc

@@ -371,7 +369,6 @@ func Test_NewConnectionManagerTest_DisconnectInvalid(t *testing.T) {
 		ConnectionState: &ConnectionState{
 			myCert:   &dummyCert{},
 			peerCert: cachedPeerCert,
-			H:        &noise.HandshakeState{},
 		},
 	}
 	nc.hostMap.unlockedAddHostInfo(hostinfo, ifce)
@@ -1,86 +1,49 @@
 package nebula

 import (
-	"crypto/rand"
 	"encoding/json"
-	"fmt"
+	"log/slog"
 	"sync"
 	"sync/atomic"

-	"github.com/flynn/noise"
-	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula/cert"
+	"github.com/slackhq/nebula/handshake"
+	"github.com/slackhq/nebula/header"
 	"github.com/slackhq/nebula/noiseutil"
 )

 const ReplayWindow = 1024

 type ConnectionState struct {
-	eKey           *NebulaCipherState
-	dKey           *NebulaCipherState
-	H              *noise.HandshakeState
+	eKey           noiseutil.CipherState
+	dKey           noiseutil.CipherState
 	myCert         cert.Certificate
 	peerCert       *cert.CachedCertificate
 	initiator      bool
 	messageCounter atomic.Uint64
 	window         *Bits
+	decryptLock    sync.Mutex
 	writeLock      sync.Mutex
 }

-func NewConnectionState(l *logrus.Logger, cs *CertState, crt cert.Certificate, initiator bool, pattern noise.HandshakePattern) (*ConnectionState, error) {
-	var dhFunc noise.DHFunc
-	switch crt.Curve() {
-	case cert.Curve_CURVE25519:
-		dhFunc = noise.DH25519
-	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.
+// newConnectionStateFromResult builds a fully-populated ConnectionState from a
+// completed handshake.Result. It seeds messageCounter and the replay window so
+// that the post-handshake message indices already used on the wire don't count
+// as missed traffic in the data plane.
+func newConnectionStateFromResult(r *handshake.Result) *ConnectionState {
 	ci := &ConnectionState{
-		H:         hs,
-		initiator: initiator,
-		window:    b,
-		myCert:    crt,
+		myCert:    r.MyCert,
+		initiator: r.Initiator,
+		peerCert:  r.RemoteCert,
+		eKey:      noiseutil.NewCipherState(r.EKey, r.Cipher),
+		dKey:      noiseutil.NewCipherState(r.DKey, r.Cipher),
+		window:    NewBits(ReplayWindow),
 	}
-	// always start the counter from 2, as packet 1 and packet 2 are handshake packets.
-	ci.messageCounter.Add(2)
-
-	return ci, nil
+	ci.messageCounter.Add(r.MessageIndex)
+	for i := uint64(1); i <= r.MessageIndex; i++ {
+		ci.window.Update(nil, i)
+	}
+	return ci
 }

 func (cs *ConnectionState) MarshalJSON() ([]byte, error) {
@@ -94,3 +57,52 @@ func (cs *ConnectionState) MarshalJSON() ([]byte, error) {
 func (cs *ConnectionState) Curve() cert.Curve {
 	return cs.myCert.Curve()
 }
+
+func (cs *ConnectionState) Decrypt(l *slog.Logger, messageCounter uint64, out []byte, packet []byte, nb []byte) ([]byte, error) {
+	var err error
+	cs.decryptLock.Lock()
+	result := cs.window.Check(l, messageCounter)
+	cs.decryptLock.Unlock()
+	if !result {
+		return nil, ErrAlreadySeen
+	}
+
+	out, err = cs.dKey.DecryptDanger(out, packet[:header.Len], packet[header.Len:], messageCounter, nb)
+	if err != nil {
+		return nil, err
+	}
+
+	cs.decryptLock.Lock()
+	result = cs.window.Update(l, messageCounter)
+	cs.decryptLock.Unlock()
+	if !result {
+		return nil, ErrAlreadySeen
+	}
+	return out, nil
+}
+
+// VerifyRelay verifies AEAD protected (but not encrypted) relay frames. packet must be length-checked by the caller.
+func (cs *ConnectionState) VerifyRelay(l *slog.Logger, messageCounter uint64, packet []byte, nb []byte) error {
+	cs.decryptLock.Lock()
+	result := cs.window.Check(l, messageCounter)
+	cs.decryptLock.Unlock()
+	if !result {
+		return ErrAlreadySeen
+	}
+
+	signedPayload := packet[:len(packet)-cs.dKey.Overhead()]
+	signatureValue := packet[len(packet)-cs.dKey.Overhead():]
+	_, err := cs.dKey.DecryptDanger(nil, signedPayload, signatureValue, messageCounter, nb)
+	if err != nil {
+		return err
+	}
+
+	cs.decryptLock.Lock()
+	result = cs.window.Update(l, messageCounter)
+	cs.decryptLock.Unlock()
+	if !result {
+		return ErrAlreadySeen
+	}
+
+	return nil
+}
@@ -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())
+	})
+}
@@ -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

@@ -26,14 +42,18 @@ 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()
 	statsStart             func()
 	dnsStart               func()
 	lighthouseStart        func()
+	networkChangeStart     func(rebind func())
 	connectionManagerStart func(context.Context)
 }

@@ -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()
-func (c *Control) Start() {
+// Start actually runs nebula, this is a nonblocking call.
+// Use Wait to block until nebula has fully stopped and to learn whether a fatal reader error caused the shutdown.
+func (c *Control) Start() error {
+	c.stateLock.Lock()
+	defer c.stateLock.Unlock()
+	switch c.state {
+	case StateReady:
+		//yay!
+	case StateStopped, StateStopping:
+		return ErrAlreadyStopped
+	case StateStarted:
+		return ErrAlreadyStarted
+	default:
+		return ErrUnknownState
+	}
+
 	// Activate the interface
-	c.f.activate()
+	err := c.f.activate()
+	if err != nil {
+		// Cancel before Close so a caller returning from Wait always observes a dead Context
+		c.cancel()
+		_ = c.f.Close()
+		c.state = StateStopped
+		return err
+	}

 	// Call all the delayed funcs that waited patiently for the interface to be created.
 	if c.sshStart != nil {
@@ -64,6 +105,9 @@ func (c *Control) Start() {
 	if c.dnsStart != nil {
 		go c.dnsStart()
 	}
+	if c.networkChangeStart != nil {
+		go c.networkChangeStart(c.RebindUDPServer)
+	}
 	if c.connectionManagerStart != nil {
 		go c.connectionManagerStart(c.ctx)
 	}
@@ -71,25 +115,70 @@ func (c *Control) Start() {
 		c.lighthouseStart()
 	}

+	c.f.triggerShutdown = c.Stop
+
 	// Start reading packets.
 	c.f.run()
+	c.state = StateStarted
+	return 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 tears nebula down, closing all tunnels and releasing everything it holds.
+// Use Wait to block until the shutdown has completed.
+// A Control that has been stopped cannot be started again, Start will return ErrAlreadyStopped.
 func (c *Control) Stop() {
-	// Stop the handshakeManager (and other services), to prevent new tunnels from
-	// being created while we're shutting them all down.
-	c.cancel()
+	c.stateLock.Lock()
+	switch c.state {
+	case StateStarted:
+		// Fall through to the full teardown below

-	c.CloseAllTunnels(false)
-	if err := c.f.Close(); err != nil {
-		c.l.WithError(err).Error("Close interface failed")
+	case StateReady:
+		// Never started
+		c.cancel()
+		c.state = StateStopped
+		if err := c.f.Close(); err != nil {
+			c.l.Error("Close interface failed", "error", err)
+		}
+		c.stateLock.Unlock()
+		return
+
+	default:
+		c.stateLock.Unlock()
+		// We are stopping or stopped already
+		return
 	}
-	c.l.Info("Goodbye")
+
+	c.state = StateStopping
+	c.stateLock.Unlock()
+
+	// Closing tunnels can be slow with a large hostmap, don't hold the lock for it
+	c.cancel()
+	c.CloseAllTunnels(false)
+
+	c.stateLock.Lock()
+	c.state = StateStopped
+	if err := c.f.Close(); err != nil {
+		c.l.Error("Close interface failed", "error", err)
+	}
+	c.stateLock.Unlock()
+}
+
+// Wait blocks until nebula has fully stopped, either via Stop or an internal fatal error,
+// and returns the first fatal packet reader error if there was one.
+// It is safe to call from multiple goroutines and at any point in the lifecycle,
+// but a Wait on a Control that is never started and never stopped will block forever.
+func (c *Control) Wait() error {
+	return c.f.wait()
 }

 // ShutdownBlock will listen for and block on term and interrupt signals, calling Control.Stop() once signalled
@@ -100,13 +189,24 @@ 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()
 }

-// RebindUDPServer asks the UDP listener to rebind it's listener. Mainly used on mobile clients when interfaces change
+// RebindUDPServer asks the UDP listener to rebind it's listener. Mainly used on mobile clients when interfaces change.
 func (c *Control) RebindUDPServer() {
-	_ = c.f.outside.Rebind()
+	c.stateLock.Lock()
+	defer c.stateLock.Unlock()
+
+	if c.state != StateStarted {
+		return
+	}
+
+	// A failure here means we are likely still pinned to the interface we came up on, so the rest of this is
+	// unlikely to help. Say so instead of silently carrying on as if we rebound.
+	if err := c.f.outside.Rebind(); err != nil {
+		c.l.Error("Failed to rebind udp socket", "error", err)
+	}

 	// Trigger a lighthouse update, useful for mobile clients that should have an update interval of 0
 	c.f.lightHouse.SendUpdate()
@@ -237,8 +337,10 @@ 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).
-			Debug("Sending close tunnel message")
+		c.l.Debug("Sending close tunnel message",
+			"vpnAddrs", h.vpnAddrs,
+			"udpAddr", h.GetRemote(),
+		)
 		closed++
 	}

@@ -282,7 +384,7 @@ func copyHostInfo(h *HostInfo, preferredRanges []netip.Prefix) ControlHostInfo {
 		RemoteAddrs:            h.remotes.CopyAddrs(preferredRanges),
 		CurrentRelaysToMe:      h.relayState.CopyRelayIps(),
 		CurrentRelaysThroughMe: h.relayState.CopyRelayForIps(),
-		CurrentRemote:          h.remote,
+		CurrentRemote:          h.GetRemote(),
 	}

 	for i, a := range h.vpnAddrs {
@@ -0,0 +1,292 @@
+package nebula
+
+import (
+	"context"
+	"errors"
+	"io"
+	"net/netip"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/gaissmai/bart"
+	"github.com/slackhq/nebula/config"
+	"github.com/slackhq/nebula/routing"
+	"github.com/slackhq/nebula/test"
+	"github.com/slackhq/nebula/udp"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+type fakeDevice struct {
+	closeOnce sync.Once
+	closedCh  chan struct{}
+	closed    bool
+}
+
+func newFakeDevice() *fakeDevice {
+	return &fakeDevice{closedCh: make(chan struct{})}
+}
+
+// Read blocks until Close like a real tun with no traffic, then reports EOF
+// the same way a closed device does
+func (d *fakeDevice) Read(p []byte) (int, error) {
+	<-d.closedCh
+	return 0, io.EOF
+}
+
+func (d *fakeDevice) Write(p []byte) (int, error) { return len(p), nil }
+
+func (d *fakeDevice) Close() error {
+	d.closeOnce.Do(func() {
+		d.closed = true
+		close(d.closedCh)
+	})
+	return nil
+}
+
+func (d *fakeDevice) Activate() error                       { return nil }
+func (d *fakeDevice) Networks() []netip.Prefix              { return nil }
+func (d *fakeDevice) Name() string                          { return "fake" }
+func (d *fakeDevice) RoutesFor(netip.Addr) routing.Gateways { return nil }
+func (d *fakeDevice) SupportsMultiqueue() bool              { return false }
+func (d *fakeDevice) NewMultiQueueReader() (io.ReadWriteCloser, error) {
+	return nil, errors.New("unsupported")
+}
+
+// newReadyControl hand-builds the minimum Control that Main would have
+// produced right before Start, including the construction token NewInterface
+// takes so waiters block until Close releases the resources
+func newReadyControl(t *testing.T) (*Control, *fakeDevice, *fakeConn) {
+	l := test.NewLogger()
+	dev := newFakeDevice()
+	conn := &fakeConn{}
+	ctx, cancel := context.WithCancel(context.Background())
+
+	myVpnNet := netip.MustParsePrefix("10.128.0.1/16")
+	nt := new(bart.Lite)
+	nt.Insert(myVpnNet)
+	cs := &CertState{
+		myVpnNetworks:      []netip.Prefix{myVpnNet},
+		myVpnNetworksTable: nt,
+	}
+	lh, err := NewLightHouseFromConfig(ctx, l, config.NewC(l), cs, nil, nil)
+	require.NoError(t, err)
+
+	f := &Interface{
+		ctx:        ctx,
+		inside:     dev,
+		outside:    conn,
+		writers:    []udp.Conn{conn},
+		readers:    make([]io.ReadWriteCloser, 1),
+		routines:   1,
+		hostMap:    newHostMap(l),
+		lightHouse: lh,
+		l:          l,
+	}
+	f.wg.Add(1)
+
+	return &Control{
+		state:  StateReady,
+		f:      f,
+		l:      l,
+		ctx:    ctx,
+		cancel: cancel,
+	}, dev, conn
+}
+
+func TestControl_StopBeforeStart(t *testing.T) {
+	c, dev, conn := newReadyControl(t)
+
+	// A Stop on a never started control must release everything Main acquired
+	c.Stop()
+	assert.Equal(t, StateStopped, c.State())
+	assert.True(t, dev.closed, "the tun device should have been closed")
+	assert.True(t, conn.closed, "the udp socket should have been closed")
+	require.ErrorIs(t, c.ctx.Err(), context.Canceled, "the service context should have been cancelled")
+
+	// Wait must return promptly now that the resources are released
+	require.NoError(t, c.Wait())
+
+	// A stopped control can never be started
+	require.ErrorIs(t, c.Start(), ErrAlreadyStopped)
+
+	// A second Stop is a harmless no-op
+	c.Stop()
+	assert.Equal(t, StateStopped, c.State())
+	require.NoError(t, c.Wait())
+}
+
+func TestControl_WaitBlocksUntilStop(t *testing.T) {
+	c, _, _ := newReadyControl(t)
+
+	done := make(chan error, 1)
+	go func() { done <- c.Wait() }()
+
+	select {
+	case <-done:
+		t.Fatal("Wait returned before Stop")
+	case <-time.After(50 * time.Millisecond):
+	}
+
+	c.Stop()
+	select {
+	case err := <-done:
+		require.NoError(t, err)
+	case <-time.After(time.Second):
+		t.Fatal("Wait did not return after Stop")
+	}
+}
+
+type fakeConn struct {
+	closed  bool
+	rebinds int
+}
+
+func (c *fakeConn) Rebind() error                            { c.rebinds++; return nil }
+func (c *fakeConn) LocalAddr() (netip.AddrPort, error)       { return netip.AddrPort{}, nil }
+func (c *fakeConn) ListenOut(_ udp.EncReader) error          { return nil }
+func (c *fakeConn) WriteTo(_ []byte, _ netip.AddrPort) error { return nil }
+func (c *fakeConn) ReloadConfig(_ *config.C)                 {}
+func (c *fakeConn) SupportsMultipleReaders() bool            { return true }
+func (c *fakeConn) Close() error                             { c.closed = true; return nil }
+
+type multiqueueDevice struct {
+	*fakeDevice
+}
+
+func (d *multiqueueDevice) SupportsMultiqueue() bool { return true }
+
+func TestControl_StartMultiqueueFailureReleases(t *testing.T) {
+	dev := &multiqueueDevice{fakeDevice: newFakeDevice()}
+	conn := &fakeConn{}
+	ctx, cancel := context.WithCancel(context.Background())
+	f := &Interface{
+		ctx:      ctx,
+		inside:   dev,
+		outside:  conn,
+		writers:  []udp.Conn{conn},
+		readers:  make([]io.ReadWriteCloser, 2),
+		routines: 2,
+		l:        test.NewLogger(),
+	}
+	f.wg.Add(1)
+
+	c := &Control{
+		state:  StateReady,
+		f:      f,
+		l:      test.NewLogger(),
+		ctx:    ctx,
+		cancel: cancel,
+	}
+
+	// The second reader fails to open, everything must be released
+	require.Error(t, c.Start())
+	assert.Equal(t, StateStopped, c.State())
+	assert.True(t, dev.closed, "the tun device should have been closed")
+	assert.True(t, conn.closed, "the udp socket should have been closed")
+	require.ErrorIs(t, c.ctx.Err(), context.Canceled)
+
+	// And Wait must not hang on the construction token
+	require.NoError(t, c.Wait())
+}
+
+func TestInterface_CloseIsIdempotent(t *testing.T) {
+	dev := newFakeDevice()
+	f := &Interface{
+		inside: dev,
+		l:      test.NewLogger(),
+	}
+	f.wg.Add(1)
+
+	require.NoError(t, f.Close())
+	assert.True(t, dev.closed)
+
+	// A second Close must not double release the wg token or the device
+	require.NoError(t, f.Close())
+	require.NoError(t, f.wait())
+}
+
+func TestControl_FatalErrorReportsThroughWait(t *testing.T) {
+	c, dev, conn := newReadyControl(t)
+
+	// Mirror what Start wires up, without needing real packet readers
+	c.f.triggerShutdown = c.Stop
+	c.state = StateStarted
+
+	boom := errors.New("boom")
+	c.f.onFatal(boom)
+
+	require.ErrorIs(t, c.Wait(), boom)
+	assert.Equal(t, StateStopped, c.State())
+	assert.True(t, dev.closed)
+	assert.True(t, conn.closed)
+
+	// A second fatal error must not fire the shutdown again or replace the first
+	c.f.onFatal(errors.New("later"))
+	require.ErrorIs(t, c.Wait(), boom)
+
+	// Wait stays factual, a Stop after the death does not mask the error
+	c.Stop()
+	require.ErrorIs(t, c.Wait(), boom)
+}
+
+func TestControl_ConcurrentStopAndStart(t *testing.T) {
+	c, _, _ := newReadyControl(t)
+
+	var wg sync.WaitGroup
+	for i := 0; i < 2; i++ {
+		wg.Go(func() { c.Stop() })
+	}
+	wg.Go(func() { _ = c.Start() })
+	wg.Go(func() {
+		_ = c.Wait()
+		// A returned Wait must always observe the final state, no matter how
+		// the race resolved
+		assert.Equal(t, StateStopped, c.State())
+	})
+	wg.Wait()
+
+	// However the race resolves, the control must end fully stopped with no
+	// panic and Wait must observe the final state
+	require.NoError(t, c.Wait())
+	assert.Equal(t, StateStopped, c.State())
+	require.ErrorIs(t, c.Start(), ErrAlreadyStopped)
+}
+
+func TestControl_StartStopLifecycle(t *testing.T) {
+	c, dev, conn := newReadyControl(t)
+
+	require.NoError(t, c.Start())
+	assert.Equal(t, StateStarted, c.State())
+	require.ErrorIs(t, c.Start(), ErrAlreadyStarted)
+
+	// Stop must unpark the reader blocked in the device and release everything
+	c.Stop()
+	assert.Equal(t, StateStopped, c.State())
+	assert.True(t, dev.closed, "the tun device should have been closed")
+	assert.True(t, conn.closed, "the udp socket should have been closed")
+	require.ErrorIs(t, c.ctx.Err(), context.Canceled)
+
+	// The reader drained off a closed device, that is not a fatal error
+	require.NoError(t, c.Wait())
+	require.ErrorIs(t, c.Start(), ErrAlreadyStopped)
+}
+
+func TestControl_RebindIsGatedByState(t *testing.T) {
+	c, _, conn := newReadyControl(t)
+
+	// A rebind before Start reaches nothing, the interface is not up
+	c.RebindUDPServer()
+	assert.Equal(t, 0, conn.rebinds, "rebind before start must be a no-op")
+
+	require.NoError(t, c.Start())
+	c.RebindUDPServer()
+	assert.Equal(t, 1, conn.rebinds, "rebind while started must reach the conn")
+
+	// A rebind racing a completed stop must not touch the closed conn
+	c.Stop()
+	require.NoError(t, c.Wait())
+	c.RebindUDPServer()
+	assert.Equal(t, 1, conn.rebinds, "rebind after stop must be a no-op")
+}
@@ -1,15 +1,17 @@
 package nebula

 import (
+	"bytes"
+	"log/slog"
 	"net"
 	"net/netip"
 	"reflect"
 	"testing"

-	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula/cert"
 	"github.com/slackhq/nebula/test"
 	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
 )

 func TestControl_GetHostInfoByVpnIp(t *testing.T) {
@@ -43,8 +45,7 @@ func TestControl_GetHostInfoByVpnIp(t *testing.T) {
 	assert.True(t, ok)

 	crt := &dummyCert{}
-	hm.unlockedAddHostInfo(&HostInfo{
-		remote:  remote1,
+	hi := &HostInfo{
 		remotes: remotes,
 		ConnectionState: &ConnectionState{
 			peerCert: &cert.CachedCertificate{Certificate: crt},
@@ -57,13 +58,14 @@ func TestControl_GetHostInfoByVpnIp(t *testing.T) {
 			relayForByAddr: map[netip.Addr]*Relay{},
 			relayForByIdx:  map[uint32]*Relay{},
 		},
-	}, &Interface{})
+	}
+	hi.remote.Store(&remote1)
+	hm.unlockedAddHostInfo(hi, &Interface{})

 	vpnIp2, ok := netip.AddrFromSlice(ipNet2.IP)
 	assert.True(t, ok)

-	hm.unlockedAddHostInfo(&HostInfo{
-		remote:  remote1,
+	hi2 := &HostInfo{
 		remotes: remotes,
 		ConnectionState: &ConnectionState{
 			peerCert: nil,
@@ -76,13 +78,16 @@ func TestControl_GetHostInfoByVpnIp(t *testing.T) {
 			relayForByAddr: map[netip.Addr]*Relay{},
 			relayForByIdx:  map[uint32]*Relay{},
 		},
-	}, &Interface{})
+	}
+	hi2.remote.Store(&remote1)
+	hm.unlockedAddHostInfo(hi2, &Interface{})

 	c := Control{
+		state: StateReady,
 		f: &Interface{
 			hostMap: hm,
 		},
-		l: logrus.New(),
+		l: test.NewLogger(),
 	}

 	thi := c.GetHostInfoByVpnAddr(vpnIp, false)
@@ -119,3 +124,153 @@ func assertFields(t *testing.T, expected []string, actualStruct any) {

 	assert.Equal(t, expected, fields)
 }
+
+// alwaysAllowV4/V6 are check funcs that accept every entry (including nil pointers),
+// letting us inject a nil *V4AddrPort/*V6AddrPort into a RemoteList's reported cache
+// the same way a malformed proto message off the wire could.
+func alwaysAllowV4(netip.Addr, *V4AddrPort) bool { return true }
+func alwaysAllowV6(netip.Addr, *V6AddrPort) bool { return true }
+
+// TestGetRelays_SkipsNilRelayAddrs proves GetRelays tolerates nil entries in the
+// RelayVpnAddrs proto slice (which protoAddrToNetAddr would nil-deref on) and still
+// returns the valid relays, including the legacy OldRelayVpnAddrs.
+func TestGetRelays_SkipsNilRelayAddrs(t *testing.T) {
+	good := netip.MustParseAddr("10.0.0.9")
+
+	d := &NebulaMetaDetails{
+		OldRelayVpnAddrs: []uint32{0x0a000001}, // 10.0.0.1
+		RelayVpnAddrs: []*Addr{
+			nil,
+			netAddrToProtoAddr(good),
+			nil,
+		},
+	}
+
+	var relays []netip.Addr
+	require.NotPanics(t, func() { relays = d.GetRelays() })
+
+	assert.Equal(t, []netip.Addr{
+		netip.MustParseAddr("10.0.0.1"),
+		good,
+	}, relays)
+}
+
+// TestGetRelays_AllNil ensures an all-nil RelayVpnAddrs slice yields no relays and no panic.
+func TestGetRelays_AllNil(t *testing.T) {
+	d := &NebulaMetaDetails{RelayVpnAddrs: []*Addr{nil, nil}}
+	var relays []netip.Addr
+	require.NotPanics(t, func() { relays = d.GetRelays() })
+	assert.Empty(t, relays)
+}
+
+// TestRemoteList_CopyCache_SkipsNilReported proves CopyCache skips nil reported
+// pointers (v4 and v6) instead of nil-dereferencing them in protoV*AddrPortToNetAddrPort.
+func TestRemoteList_CopyCache_SkipsNilReported(t *testing.T) {
+	owner := netip.MustParseAddr("10.0.0.1")
+	rl := NewRemoteList([]netip.Addr{owner}, nil)
+
+	rl.unlockedSetV4(owner, owner, []*V4AddrPort{
+		nil,
+		newIp4AndPortFromString("1.2.3.4:5"),
+		nil,
+	}, alwaysAllowV4)
+
+	rl.unlockedSetV6(owner, owner, []*V6AddrPort{
+		nil,
+		newIp6AndPortFromString("[1::1]:6"),
+		nil,
+	}, alwaysAllowV6)
+
+	var cm *CacheMap
+	require.NotPanics(t, func() { cm = rl.CopyCache() })
+
+	c := (*cm)[owner.String()]
+	require.NotNil(t, c)
+	assert.ElementsMatch(t, []netip.AddrPort{
+		netip.MustParseAddrPort("1.2.3.4:5"),
+		netip.MustParseAddrPort("[1::1]:6"),
+	}, c.Reported)
+}
+
+// TestRemoteList_Rebuild_SkipsNilReported drives unlockedCollect (via Rebuild) with
+// nil reported entries and confirms only the valid addresses survive, with no panic.
+func TestRemoteList_Rebuild_SkipsNilReported(t *testing.T) {
+	owner := netip.MustParseAddr("10.0.0.1")
+	rl := NewRemoteList([]netip.Addr{owner}, nil)
+
+	rl.unlockedSetV4(owner, owner, []*V4AddrPort{
+		nil,
+		newIp4AndPortFromString("1.2.3.4:5"),
+	}, alwaysAllowV4)
+	rl.unlockedSetV6(owner, owner, []*V6AddrPort{
+		newIp6AndPortFromString("[1::1]:6"),
+		nil,
+	}, alwaysAllowV6)
+
+	require.NotPanics(t, func() { rl.Rebuild([]netip.Prefix{}) })
+
+	assert.ElementsMatch(t, []netip.AddrPort{
+		netip.MustParseAddrPort("1.2.3.4:5"),
+		netip.MustParseAddrPort("[1::1]:6"),
+	}, rl.addrs)
+}
+
+// newRelayControl marshals a NebulaControl the way it arrives on the wire so we can feed
+// it through HandleControlMsg's unmarshal + validate path.
+func newRelayControl(t *testing.T, typ NebulaControl_MessageType, from, to *Addr) []byte {
+	t.Helper()
+	msg := &NebulaControl{
+		Type:          typ,
+		RelayFromAddr: from,
+		RelayToAddr:   to,
+	}
+	b, err := msg.Marshal()
+	require.NoError(t, err)
+	return b
+}
+
+// TestRelayManager_HandleControlMsg_NilRelayAddrs verifies the validation block added to
+// HandleControlMsg: CreateRelay{Request,Response} carrying a nil RelayFromAddr or
+// RelayToAddr are dropped with a debug log rather than nil-dereferencing downstream.
+func TestRelayManager_HandleControlMsg_NilRelayAddrs(t *testing.T) {
+	good := netAddrToProtoAddr(netip.MustParseAddr("10.0.0.9"))
+
+	cases := []struct {
+		name    string
+		typ     NebulaControl_MessageType
+		from    *Addr
+		to      *Addr
+		wantLog string // debug substring expected, "" == expect no drop log
+	}{
+		{"request nil from", NebulaControl_CreateRelayRequest, nil, good, "nil RelayFromAddr"},
+		{"request nil to", NebulaControl_CreateRelayRequest, good, nil, "nil RelayToAddr"},
+		{"request both nil", NebulaControl_CreateRelayRequest, nil, nil, "nil RelayFromAddr"},
+		{"response nil from", NebulaControl_CreateRelayResponse, nil, good, "nil RelayFromAddr"},
+		{"response nil to", NebulaControl_CreateRelayResponse, good, nil, "nil RelayToAddr"},
+		// A non-relay control type is not subject to the relay-addr validation and must
+		// pass through it untouched (the final switch simply no-ops on it).
+		{"unrelated type nil addrs", NebulaControl_None, nil, nil, ""},
+	}
+
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			var buf bytes.Buffer
+			l := test.NewLoggerWithOutputAndLevel(&buf, slog.LevelDebug)
+			rm := &relayManager{l: l, hostmap: newHostMap(l)}
+			rm.useRelays.Store(true)
+
+			f := &Interface{l: l}
+			h := &HostInfo{vpnAddrs: []netip.Addr{netip.MustParseAddr("10.0.0.2")}, localIndexId: 1}
+
+			d := newRelayControl(t, tc.typ, tc.from, tc.to)
+
+			require.NotPanics(t, func() { rm.HandleControlMsg(h, d, f) })
+
+			if tc.wantLog == "" {
+				assert.NotContains(t, buf.String(), "nil Relay")
+			} else {
+				assert.Contains(t, buf.String(), tc.wantLog)
+			}
+		})
+	}
+}
@@ -1,13 +1,10 @@
 //go:build e2e_testing
-// +build e2e_testing

 package nebula

 import (
 	"net/netip"

-	"github.com/google/gopacket"
-	"github.com/google/gopacket/layers"
 	"github.com/slackhq/nebula/header"
 	"github.com/slackhq/nebula/overlay"
 	"github.com/slackhq/nebula/udp"
@@ -23,7 +20,9 @@ func (c *Control) WaitForType(msgType header.MessageType, subType header.Message
 			panic(err)
 		}
 		pipeTo.InjectUDPPacket(p)
-		if h.Type == msgType && h.Subtype == subType {
+		match := h.Type == msgType && h.Subtype == subType
+		p.Release()
+		if match {
 			return
 		}
 	}
@@ -39,7 +38,9 @@ func (c *Control) WaitForTypeByIndex(toIndex uint32, msgType header.MessageType,
 			panic(err)
 		}
 		pipeTo.InjectUDPPacket(p)
-		if h.RemoteIndex == toIndex && h.Type == msgType && h.Subtype == subType {
+		match := h.RemoteIndex == toIndex && h.Type == msgType && h.Subtype == subType
+		p.Release()
+		if match {
 			return
 		}
 	}
@@ -91,65 +92,15 @@ func (c *Control) GetTunTxChan() <-chan []byte {
 	return c.f.inside.(*overlay.TestTun).TxPackets
 }

-// InjectUDPPacket will inject a packet into the udp side of nebula
+// InjectUDPPacket injects a packet into the udp side. We copy internally so the caller keeps ownership of p.
+// The copy comes from the freelist so steady-state alloc is zero.
 func (c *Control) InjectUDPPacket(p *udp.Packet) {
-	c.f.outside.(*udp.TesterConn).Send(p)
+	c.f.outside.(*udp.TesterConn).Send(p.Copy())
 }

-// InjectTunUDPPacket puts a udp packet on the tun interface. Using UDP here because it's a simpler protocol
-func (c *Control) InjectTunUDPPacket(toAddr netip.Addr, toPort uint16, fromAddr netip.Addr, fromPort uint16, data []byte) {
-	serialize := make([]gopacket.SerializableLayer, 0)
-	var netLayer gopacket.NetworkLayer
-	if toAddr.Is6() {
-		if !fromAddr.Is6() {
-			panic("Cant send ipv6 to ipv4")
-		}
-		ip := &layers.IPv6{
-			Version:    6,
-			NextHeader: layers.IPProtocolUDP,
-			SrcIP:      fromAddr.Unmap().AsSlice(),
-			DstIP:      toAddr.Unmap().AsSlice(),
-		}
-		serialize = append(serialize, ip)
-		netLayer = ip
-	} else {
-		if !fromAddr.Is4() {
-			panic("Cant send ipv4 to ipv6")
-		}
-
-		ip := &layers.IPv4{
-			Version:  4,
-			TTL:      64,
-			Protocol: layers.IPProtocolUDP,
-			SrcIP:    fromAddr.Unmap().AsSlice(),
-			DstIP:    toAddr.Unmap().AsSlice(),
-		}
-		serialize = append(serialize, ip)
-		netLayer = ip
-	}
-
-	udp := layers.UDP{
-		SrcPort: layers.UDPPort(fromPort),
-		DstPort: layers.UDPPort(toPort),
-	}
-	err := udp.SetNetworkLayerForChecksum(netLayer)
-	if err != nil {
-		panic(err)
-	}
-
-	buffer := gopacket.NewSerializeBuffer()
-	opt := gopacket.SerializeOptions{
-		ComputeChecksums: true,
-		FixLengths:       true,
-	}
-
-	serialize = append(serialize, &udp, gopacket.Payload(data))
-	err = gopacket.SerializeLayers(buffer, opt, serialize...)
-	if err != nil {
-		panic(err)
-	}
-
-	c.f.inside.(*overlay.TestTun).Send(buffer.Bytes())
+// InjectTunPacket pushes an IP packet onto the tun interface.
+func (c *Control) InjectTunPacket(packet []byte) {
+	c.f.inside.(*overlay.TestTun).Send(packet)
 }

 func (c *Control) GetVpnAddrs() []netip.Addr {
@@ -157,7 +108,19 @@ func (c *Control) GetVpnAddrs() []netip.Addr {
 }

 func (c *Control) GetUDPAddr() netip.AddrPort {
-	return c.f.outside.(*udp.TesterConn).Addr
+	return c.f.outside.(*udp.TesterConn).GetAddr()
+}
+
+// SetUDPAddr moves this node to a new underlay address, standing in for a laptop waking up on a different
+// network. Register the new address with the router as well or nothing will route back.
+func (c *Control) SetUDPAddr(addr netip.AddrPort) {
+	c.f.outside.(*udp.TesterConn).SetAddr(addr)
+}
+
+// SetLocalAddrsFn replaces underlay address discovery so a test can advertise its simulated address instead of
+// whatever this machine's NICs happen to be. Call it before Start, SendUpdate reads it from the update worker.
+func (c *Control) SetLocalAddrsFn(fn func(*LocalAllowList) []netip.Addr) {
+	c.f.lightHouse.localAddrsFn = fn
 }

 func (c *Control) KillPendingTunnel(vpnIp netip.Addr) bool {
@@ -174,6 +137,18 @@ func (c *Control) GetHostmap() *HostMap {
 	return c.f.hostMap
 }

+// GetHostmapIndexCount returns the number of entries in the main hostmap Indexes table, holding
+// the hostmap read lock so tests can poll it while connection manager churns tunnels.
+func (c *Control) GetHostmapIndexCount() int {
+	c.f.hostMap.RLock()
+	defer c.f.hostMap.RUnlock()
+	return len(c.f.hostMap.Indexes)
+}
+
+func (c *Control) GetF() *Interface {
+	return c.f
+}
+
 func (c *Control) GetCertState() *CertState {
 	return c.f.pki.getCertState()
 }
@@ -62,7 +62,7 @@ function nebula.dissector(tvbuf, pktinfo, root)
    tree:add(pf_version, tvbuf:range(0,1))
    local type = tree:add(pf_type,    tvbuf:range(0,1))

-    local nebula_type = bit32.band(tvbuf:range(0,1):uint(), 0x0F)
+    local nebula_type = bit.band(tvbuf:range(0,1):uint(), 0x0F)
    if nebula_type == 0 then
        local stage = tvbuf(8,8):uint64()
        tree:add(pf_subtype_handshake, tvbuf:range(1,1))
@@ -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
-        default_settings.all_port = nebula.prefs.all_ports
-
+    if nebula.prefs.all_ports then
+        -- Register on every port for hole punch capture
        for i=0, 65535 do
            DissectorTable.get("udp.port"):add(i, nebula)
        end
-
-        -- no need to establish again on specific ports
-        return
+    else
+        -- Register on the configured port only
+        DissectorTable.get("udp.port"):add(nebula.prefs.port, nebula)
    end

-
-    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
+    default_settings.all_ports = nebula.prefs.all_ports
+    default_settings.port = nebula.prefs.port
 end

 DissectorTable.get("udp.port"):add(default_settings.port, nebula)
@@ -1,68 +1,277 @@
 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
-
-var dnsR *dnsRecords
-var dnsServer *dns.Server
-var dnsAddr string
-
-type dnsRecords struct {
+type dnsServer struct {
 	sync.RWMutex
-	l               *logrus.Logger
-	dnsMap4         map[string]netip.Addr
-	dnsMap6         map[string]netip.Addr
-	hostMap         *HostMap
-	myVpnAddrsTable *bart.Lite
+	l       *slog.Logger
+	ctx     context.Context
+	dnsMap4 map[string]netip.Addr
+	dnsMap6 map[string]netip.Addr
+	hostMap *HostMap
+	pki     *PKI
+
+	// selfHost is the cached FQDN we last seeded for ourselves
+	selfHost string
+
+	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 {
-	return &dnsRecords{
-		l:               l,
-		dnsMap4:         make(map[string]netip.Addr),
-		dnsMap6:         make(map[string]netip.Addr),
-		hostMap:         hostMap,
-		myVpnAddrsTable: cs.myVpnAddrsTable,
+// newDnsServerFromConfig builds a dnsServer, applies the initial config, and
+// 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, pki *PKI, 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,
+		pki:     pki,
+	}
+	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
+	}
+	ds.seedSelf()
+	return ds, nil
+}
+
+// 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 != nil
+	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 {
+			d.Stop()
+		}
+		// Drop any records that accumulated while enabled; a later re-enable
+		// will repopulate from fresh handshakes and a fresh seedSelf.
+		d.clearRecords()
+		return nil
+	}
+
+	if !running {
+		// Was disabled (or never started); bring it up now.
+		go d.Start()
+	} else if !sameAddr {
+		// Stop clears the slot before shutting down, otherwise the Start below can find the dying server and refuse
+		d.Stop()
+		go d.Start()
+	}
+
+	// Refresh the self entry every enabled reload so cert renewals that change our name or VPN addresses are picked up.
+	d.seedSelf()
+	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)
 	}
 }

-func (d *dnsRecords) Query(q uint16, data string) netip.Addr {
+// 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()
+	// Re-check enabled under the lock, a disable that raced our check above snapshots the slot under it too.
+	// Two reloads in quick succession can both spawn a Start, the loser would orphan the live listener past Stop
+	if d.ctx.Err() != nil || d.server != nil || !d.enabled.Load() {
+		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)
+	}
+
+	// Release our slot, unless a reload already replaced us, so a dead listener can't block a future Start
+	d.serverMu.Lock()
+	if d.server == server {
+		d.server = nil
+		d.started = nil
+	}
+	d.serverMu.Unlock()
+
+	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 {
-			return r
+		if haveV4 {
+			return addr4, nameExists
 		}
 	case dns.TypeAAAA:
-		if r, ok := d.dnsMap6[data]; ok {
-			return r
+		if haveV6 {
+			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 ""
 	}

+	// The hostmap only ever contains peers we have handshaked with, so it never carries an entry for ourselves.
+	// Answer self lookups straight from the local cert state.
+	if cs := d.certState(); cs != nil && cs.myVpnAddrsTable != nil && cs.myVpnAddrsTable.Contains(ip) {
+		c := cs.GetDefaultCertificate()
+		if c == nil {
+			return ""
+		}
+		b, err := c.MarshalJSON()
+		if err != nil {
+			return ""
+		}
+		return string(b)
+	}
+
 	hostinfo := d.hostMap.QueryVpnAddr(ip)
 	if hostinfo == nil {
 		return ""
@@ -80,8 +289,67 @@ func (d *dnsRecords) QueryCert(data string) string {
 	return string(b)
 }

+// clearRecords drops all DNS records, including the self entry.
+func (d *dnsServer) clearRecords() {
+	d.Lock()
+	defer d.Unlock()
+	clear(d.dnsMap4)
+	clear(d.dnsMap6)
+	d.selfHost = ""
+}
+
+// seedSelf inserts (or refreshes) a record for our own cert name pointing at our VPN addresses,
+// so a single-lighthouse network can resolve the lighthouse's own hostname without the two-process workaround.
+func (d *dnsServer) seedSelf() {
+	if !d.enabled.Load() {
+		return
+	}
+	cs := d.certState()
+	if cs == nil {
+		return
+	}
+	c := cs.GetDefaultCertificate()
+	if c == nil {
+		return
+	}
+	newHost := strings.ToLower(c.Name()) + "."
+
+	d.Lock()
+	defer d.Unlock()
+	if d.selfHost != "" && d.selfHost != newHost {
+		delete(d.dnsMap4, d.selfHost)
+		delete(d.dnsMap6, d.selfHost)
+	}
+	d.selfHost = newHost
+	delete(d.dnsMap4, newHost)
+	delete(d.dnsMap6, newHost)
+	haveV4, haveV6 := false, false
+	for _, addr := range cs.myVpnAddrs {
+		if addr.Is4() && !haveV4 {
+			d.dnsMap4[newHost] = addr
+			haveV4 = true
+		} else if addr.Is6() && !haveV6 {
+			d.dnsMap6[newHost] = addr
+			haveV6 = true
+		}
+		if haveV4 && haveV6 {
+			break
+		}
+	}
+}
+
+func (d *dnsServer) certState() *CertState {
+	if d.pki == nil {
+		return nil
+	}
+	return d.pki.getCertState()
+}
+
 // 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 +369,7 @@ 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 {
@@ -112,17 +380,32 @@ func (d *dnsRecords) isSelfNebulaOrLocalhost(addr string) bool {
 		return true
 	}

+	cs := d.certState()
+	if cs == nil || cs.myVpnAddrsTable == nil {
+		return false
+	}
 	//if we found it in this table, it's good
-	return d.myVpnAddrsTable.Contains(b)
+	return cs.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)
-			ip := d.Query(q.Qtype, q.Name)
+			if debugEnabled {
+				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 +417,9 @@ 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 +430,12 @@ 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 +448,6 @@ 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 +456,3 @@ 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)
-}
@@ -1,19 +1,46 @@
 package nebula

 import (
+	"context"
+	"log/slog"
+	"net"
 	"net/netip"
+	"strconv"
 	"testing"
+	"time"

+	"github.com/gaissmai/bart"
 	"github.com/miekg/dns"
-	"github.com/sirupsen/logrus"
+	"github.com/slackhq/nebula/cert"
+	"github.com/slackhq/nebula/cert_test"
 	"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 +48,56 @@ 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 +136,496 @@ 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) {
+	port := freeUDPPort(t)
+	ds, c := newTestDnsServer(t)
+	setDnsConfig(c, "127.0.0.1", port, true, true)
+	require.NoError(t, ds.reload(c, true))
+
+	go ds.Start()
+	waitForBind(t, ds)
+
+	ds.serverMu.Lock()
+	before := ds.server
+	ds.serverMu.Unlock()
+	require.NotNil(t, before)
+
+	// Same address, so the running listener must be left alone rather than rebuilt under live queries
+	require.NoError(t, ds.reload(c, false))
+	assert.True(t, ds.enabled.Load())
+
+	ds.serverMu.Lock()
+	after := ds.server
+	ds.serverMu.Unlock()
+	assert.Same(t, before, after, "a same-address reload must not restart the listener")
+
+	ds.Stop()
+}
+
+// The branch the old sameAddr test was accidentally hitting: enabled with nothing running means reload starts it.
+func TestDnsServer_reload_whenNotRunning_starts(t *testing.T) {
+	port := freeUDPPort(t)
+	ds, c := newTestDnsServer(t)
+	setDnsConfig(c, "127.0.0.1", port, true, true)
+
+	// initial only records config, it never starts anything
+	require.NoError(t, ds.reload(c, true))
+	ds.serverMu.Lock()
+	assert.Nil(t, ds.server, "the initial reload must not start a listener")
+	ds.serverMu.Unlock()
+
+	require.NoError(t, ds.reload(c, false))
+	waitForBind(t, ds)
+
+	ds.serverMu.Lock()
+	assert.NotNil(t, ds.server, "a reload with nothing running should bring DNS up")
+	ds.serverMu.Unlock()
+
+	ds.Stop()
+}
+
+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")
+	}
+}
+
+// newTestPKI builds a minimal *PKI with a single v1 cert whose name and
+// VPN addresses are caller-provided, suitable for exercising seedSelf and
+// QueryCert self handling.
+func newTestPKI(t *testing.T, name string, addrs []netip.Addr) *PKI {
+	t.Helper()
+	networks := make([]netip.Prefix, 0, len(addrs))
+	for _, a := range addrs {
+		bits := 32
+		if a.Is6() {
+			bits = 128
+		}
+		networks = append(networks, netip.PrefixFrom(a, bits))
+	}
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil)
+	c, _, _, _ := cert_test.NewTestCert(cert.Version2, cert.Curve_CURVE25519, ca, caKey, name, time.Time{}, time.Time{}, networks, nil, nil)
+
+	addrsTable := new(bart.Lite)
+	for _, a := range addrs {
+		addrsTable.Insert(netip.PrefixFrom(a, a.BitLen()))
+	}
+
+	cs := &CertState{
+		v2Cert:            c,
+		initiatingVersion: cert.Version2,
+		myVpnAddrs:        addrs,
+		myVpnAddrsTable:   addrsTable,
+	}
+	pki := &PKI{}
+	pki.cs.Store(cs)
+	return pki
+}
+
+func TestDnsServer_seedSelf_addsOwnRecord(t *testing.T) {
+	ds, c := newTestDnsServer(t)
+	myV4 := netip.MustParseAddr("10.0.0.1")
+	myV6 := netip.MustParseAddr("fd00::1")
+	ds.pki = newTestPKI(t, "lighthouse", []netip.Addr{myV4, myV6})
+	setDnsConfig(c, "127.0.0.1", "0", true, true)
+	require.NoError(t, ds.reload(c, true))
+
+	ds.seedSelf()
+	got4, exists := ds.Query(dns.TypeA, "lighthouse.")
+	assert.True(t, exists)
+	assert.Equal(t, myV4, got4)
+	got6, exists := ds.Query(dns.TypeAAAA, "lighthouse.")
+	assert.True(t, exists)
+	assert.Equal(t, myV6, got6)
+}
+
+func TestDnsServer_seedSelf_disabled_noOp(t *testing.T) {
+	ds, c := newTestDnsServer(t)
+	ds.pki = newTestPKI(t, "lighthouse", []netip.Addr{netip.MustParseAddr("10.0.0.1")})
+	setDnsConfig(c, "127.0.0.1", "0", true, false)
+	require.NoError(t, ds.reload(c, true))
+
+	ds.seedSelf()
+	_, exists := ds.Query(dns.TypeA, "lighthouse.")
+	assert.False(t, exists)
+}
+
+func TestDnsServer_clearRecords_dropsSelfHost(t *testing.T) {
+	ds, c := newTestDnsServer(t)
+	ds.pki = newTestPKI(t, "lighthouse", []netip.Addr{netip.MustParseAddr("10.0.0.1")})
+	setDnsConfig(c, "127.0.0.1", "0", true, true)
+	require.NoError(t, ds.reload(c, true))
+	ds.seedSelf()
+	require.NotEmpty(t, ds.selfHost)
+
+	ds.clearRecords()
+	assert.Empty(t, ds.selfHost)
+	_, exists := ds.Query(dns.TypeA, "lighthouse.")
+	assert.False(t, exists)
+}
+
+func TestDnsServer_QueryCert_returnsOwnCert(t *testing.T) {
+	ds, _ := newTestDnsServer(t)
+	myV4 := netip.MustParseAddr("10.0.0.1")
+	ds.pki = newTestPKI(t, "lighthouse", []netip.Addr{myV4})
+
+	got := ds.QueryCert(myV4.String() + ".")
+	assert.NotEmpty(t, got, "TXT lookup of our own VPN address should return our cert")
+
+	other := netip.MustParseAddr("10.0.0.99")
+	assert.Empty(t, ds.QueryCert(other.String()+"."), "unknown peer IP should return nothing")
+}
+
+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")
+}
+
+// Two reloads in quick succession, or a HUP before Control.Start, can race two Starts at the same listener.
+func TestDnsServer_Start_isIdempotent(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))
+
+	go ds.Start()
+	waitForBind(t, ds)
+
+	ds.serverMu.Lock()
+	first := ds.server
+	ds.serverMu.Unlock()
+	require.NotNil(t, first)
+
+	// If the second Start replaces the tracked server, Stop kills the wrong one and the port leaks
+	done := make(chan struct{})
+	go func() {
+		ds.Start()
+		close(done)
+	}()
+	select {
+	case <-done:
+	case <-time.After(time.Second * 5):
+		t.Fatal("second Start never returned")
+	}
+
+	ds.serverMu.Lock()
+	second := ds.server
+	ds.serverMu.Unlock()
+	assert.Same(t, first, second, "a second Start must not replace the running server")
+
+	// The real proof, after Stop the port must actually be free
+	ds.Stop()
+	waitFor(t, func() bool {
+		pc, err := net.ListenPacket("udp", "127.0.0.1:"+port)
+		if err != nil {
+			return false
+		}
+		_ = pc.Close()
+		return true
+	})
+}
+
+// An address change must actually end up listening on the new port. Start's guard refuses when a server is already
+// installed, so reload has to clear the slot before shutting the old one down.
+func TestDnsServer_reload_addrChange_restarts(t *testing.T) {
+	first := freeUDPPort(t)
+	second := freeUDPPort(t)
+
+	ds, c := newTestDnsServer(t)
+	setDnsConfig(c, "127.0.0.1", first, true, true)
+	require.NoError(t, ds.reload(c, true))
+
+	go ds.Start()
+	waitForBind(t, ds)
+
+	// Cycle a few times, the failure this guards against depends on which goroutine wins serverMu
+	for i := range 8 {
+		want := second
+		if i%2 == 1 {
+			want = first
+		}
+		setDnsConfig(c, "127.0.0.1", want, true, true)
+		require.NoError(t, ds.reload(c, false))
+		waitForBind(t, ds)
+
+		ds.serverMu.Lock()
+		srv := ds.server
+		ds.serverMu.Unlock()
+		require.NotNil(t, srv, "reload left DNS down instead of restarting it")
+		require.Equal(t, "127.0.0.1:"+want, srv.Addr, "reload should be serving the new address")
+	}
+
+	// Land back on second so the port assertions below are meaningful
+	setDnsConfig(c, "127.0.0.1", second, true, true)
+	require.NoError(t, ds.reload(c, false))
+	waitForBind(t, ds)
+
+	// The old port must be released and the new one actually held
+	waitFor(t, func() bool {
+		pc, err := net.ListenPacket("udp", "127.0.0.1:"+first)
+		if err != nil {
+			return false
+		}
+		_ = pc.Close()
+		return true
+	})
+	_, err := net.ListenPacket("udp", "127.0.0.1:"+second)
+	require.Error(t, err, "the new address should be bound by the DNS responder")
+
+	ds.Stop()
+}
+
+// A listener that dies on its own must release the slot, or a later same-addr reload sees it as running and no-ops.
+func TestDnsServer_Start_bindFailure_releasesSlot(t *testing.T) {
+	port := freeUDPPort(t)
+	blocker, err := net.ListenPacket("udp", "127.0.0.1:"+port)
+	require.NoError(t, err)
+
+	ds, c := newTestDnsServer(t)
+	setDnsConfig(c, "127.0.0.1", port, true, true)
+	require.NoError(t, ds.reload(c, true))
+
+	ds.Start() // returns once the bind fails
+
+	ds.serverMu.Lock()
+	assert.Nil(t, ds.server, "a listener that failed to bind must not stay parked in the slot")
+	ds.serverMu.Unlock()
+
+	// With the slot released, a reload can retry once the port frees up
+	require.NoError(t, blocker.Close())
+	require.NoError(t, ds.reload(c, false))
+	waitForBind(t, ds)
+
+	ds.serverMu.Lock()
+	assert.NotNil(t, ds.server, "a same-addr reload should retry after a failed bind")
+	ds.serverMu.Unlock()
+
+	ds.Stop()
+}
+
+// A disable that lands while Start is between its unlocked check and the guard must not leave a listener behind.
+func TestDnsServer_Start_refusesWhenDisabledUnderLock(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))
+	require.True(t, ds.enabled.Load())
+
+	// Holding serverMu parks Start on the lock, the only way to land the disable in that window on purpose
+	ds.serverMu.Lock()
+
+	done := make(chan struct{})
+	go func() {
+		ds.Start()
+		close(done)
+	}()
+
+	select {
+	case <-done:
+		ds.serverMu.Unlock()
+		t.Fatal("Start returned early, the test never exercised the window")
+	case <-time.After(time.Millisecond * 100):
+	}
+
+	// The disable reload's critical section. It sees nothing running, so it never calls Stop.
+	ds.enabled.Store(false)
+	ds.serverMu.Unlock()
+
+	select {
+	case <-done:
+	case <-time.After(time.Second * 5):
+		t.Fatal("Start never returned")
+	}
+
+	ds.serverMu.Lock()
+	assert.Nil(t, ds.server, "Start must not install a listener a disable already cancelled")
+	ds.serverMu.Unlock()
+
+	pc, err := net.ListenPacket("udp", "127.0.0.1:"+port)
+	require.NoError(t, err, "an orphaned listener is still holding the port")
+	_ = pc.Close()
+}
@@ -1,6 +1,16 @@
 FROM gcr.io/distroless/static:latest

 ARG TARGETOS TARGETARCH
+
+ARG VERSION=dev
+ARG REVISION=unknown
+LABEL org.opencontainers.image.title="nebula" \
+      org.opencontainers.image.description="A scalable overlay networking tool with a focus on performance, simplicity and security" \
+      org.opencontainers.image.vendor="Nebula OSS" \
+      org.opencontainers.image.source="https://github.com/slackhq/nebula" \
+      org.opencontainers.image.version="${VERSION}" \
+      org.opencontainers.image.revision="${REVISION}"
+
 COPY build/$TARGETOS-$TARGETARCH/nebula /nebula
 COPY build/$TARGETOS-$TARGETARCH/nebula-cert /nebula-cert

@@ -0,0 +1,85 @@
+//go:build e2e_testing
+// +build e2e_testing
+
+package e2e
+
+import (
+	"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"
+	"github.com/stretchr/testify/require"
+)
+
+func assertTestRequestEchoed(t *testing.T, cipher string) {
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+	over := m{"cipher": cipher}
+	a, aNet, aUdp, _ := newSimpleServer(cert.Version1, ca, caKey, "a", "10.128.0.1/24", over)
+	b, bNet, bUdp, _ := newSimpleServer(cert.Version1, ca, caKey, "b", "10.128.0.2/24", over)
+
+	a.InjectLightHouseAddr(bNet[0].Addr(), bUdp)
+	b.InjectLightHouseAddr(aNet[0].Addr(), aUdp)
+	a.Start()
+	b.Start()
+	t.Cleanup(func() { a.Stop(); b.Stop() })
+	r := router.NewR(t, a, b)
+	defer r.RenderFlow()
+
+	assertTunnel(t, aNet[0].Addr(), bNet[0].Addr(), a, b, r)
+	drainUDPTx(a)
+	drainUDPTx(b)
+
+	payload := []byte("a test payload well over sixteen bytes long, wow it's so very long long long!")
+	require.Greater(t, len(payload), header.Len)
+	a.GetF().SendMessageToVpnAddr(header.Test, header.TestRequest, bNet[0].Addr(), payload, make([]byte, 12, 12), make([]byte, udp.MTU))
+
+	// Deliver A's request to B; B must echo a reply back
+	b.InjectUDPPacket(a.GetFromUDP(true))
+	reply := nextUDPTxOfType(t, b, header.Test, header.TestReply, 2*time.Second)
+
+	assert.Equal(t, aUdp, reply.To, "the reply must go back to the requester")
+	// header + echoed payload + 16-byte AEAD tag: proves the whole payload
+	// round-tripped rather than being dropped or truncated.
+	assert.Equal(t, header.Len+len(payload)+16, len(reply.Data), "the full payload must be echoed back")
+}
+
+func TestTestRequestEchoesLongPayloadAES(t *testing.T) {
+	assertTestRequestEchoed(t, "aes")
+}
+
+func TestTestRequestEchoesLongPayloadChaChaPoly(t *testing.T) {
+	assertTestRequestEchoed(t, "chachapoly")
+}
+
+// drainUDPTx empties a control's UDP tx queue without blocking.
+func drainUDPTx(c *nebula.Control) {
+	for c.GetFromUDP(false) != nil {
+	}
+}
+
+// nextUDPTxOfType returns the next packet a control transmits whose nebula
+// header matches (wantType, wantSub), skipping unrelated packets.
+// It fails the test if none arrives within the timeout.
+func nextUDPTxOfType(t *testing.T, c *nebula.Control, wantType header.MessageType, wantSub header.MessageSubType, within time.Duration) *udp.Packet {
+	t.Helper()
+	ch := c.GetUDPTxChan()
+	timeout := time.After(within)
+	for {
+		select {
+		case p := <-ch:
+			var h header.H
+			if err := h.Parse(p.Data); err == nil && h.Type == wantType && h.Subtype == wantSub {
+				return p
+			}
+		case <-timeout:
+			t.Fatalf("timed out waiting for a %v/%v packet on the udp tx queue", wantType, wantSub)
+			return nil
+		}
+	}
+}
@@ -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.InjectTunPacket(BuildTunUDPPacket(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.InjectTunPacket(BuildTunUDPPacket(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.InjectTunPacket(BuildTunUDPPacket(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.InjectTunPacket(BuildTunUDPPacket(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.InjectTunPacket(BuildTunUDPPacket(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.InjectTunPacket(BuildTunUDPPacket(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.InjectTunPacket(BuildTunUDPPacket(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.InjectTunPacket(BuildTunUDPPacket(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.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("packet1")))
+	myControl.InjectTunPacket(BuildTunUDPPacket(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.InjectTunPacket(BuildTunUDPPacket(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
+// BuildTunUDPPacket 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.
@@ -11,25 +11,26 @@ 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"
 	"github.com/slackhq/nebula/e2e/router"
 	"github.com/slackhq/nebula/header"
+	"github.com/slackhq/nebula/overlay"
 	"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()
@@ -38,16 +39,68 @@ func BenchmarkHotPath(b *testing.B) {
 	r := router.NewR(b, myControl, theirControl)
 	r.CancelFlowLogs()

+	assertTunnel(b, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
+
+	// Pre-build the IP packet bytes once so the bench measures the data plane,
+	// not gopacket SerializeLayers overhead.
+	prebuilt := BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me"))
+
+	// EnableFanIn switches the router to a 0-alloc routing path. Required
+	// for hot-path benchmarks; would conflict with GetFromUDP-using tests.
+	r.EnableFanIn()
+
+	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.InjectTunPacket(prebuilt)
+		// Release the TUN-side bytes back to the harness freelist; the bench
+		// just confirms a packet arrived, the contents aren't inspected.
+		overlay.ReleaseTunBuf(r.RouteForAllUntilTxTun(theirControl))
 	}

 	myControl.Stop()
 	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)
+
+	prebuilt := BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me"))
+	r.EnableFanIn()
+
+	b.ResetTimer()
+
+	for n := 0; n < b.N; n++ {
+		myControl.InjectTunPacket(prebuilt)
+		overlay.ReleaseTunBuf(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)
@@ -60,7 +113,7 @@ func TestGoodHandshake(t *testing.T) {
 	theirControl.Start()

 	t.Log("Send a udp packet through to begin standing up the tunnel, this should come out the other side")
-	myControl.InjectTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me"))
+	myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 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))
@@ -97,7 +150,44 @@ 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)
@@ -117,7 +207,7 @@ func TestWrongResponderHandshake(t *testing.T) {
 	evilControl.Start()

 	t.Log("Start the handshake process, we will route until we see the evil tunnel closed")
-	myControl.InjectTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me"))
+	myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me")))

 	h := &header.H{}
 	r.RouteForAllExitFunc(func(p *udp.Packet, c *nebula.Control) router.ExitType {
@@ -174,6 +264,7 @@ 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)
@@ -198,7 +289,7 @@ func TestWrongResponderHandshakeStaticHostMap(t *testing.T) {
 	evilControl.Start()

 	t.Log("Start the handshake process, we will route until we see the evil tunnel closed")
-	myControl.InjectTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me"))
+	myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me")))

 	h := &header.H{}
 	r.RouteForAllExitFunc(func(p *udp.Packet, c *nebula.Control) router.ExitType {
@@ -256,6 +347,7 @@ 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

@@ -276,8 +368,8 @@ func TestStage1Race(t *testing.T) {
 	theirControl.Start()

 	t.Log("Trigger a handshake to start on both me and them")
-	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"))
+	myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me")))
+	theirControl.InjectTunPacket(BuildTunUDPPacket(myVpnIpNet[0].Addr(), 80, theirVpnIpNet[0].Addr(), 80, []byte("Hi from them")))

 	t.Log("Get both stage 1 handshake packets")
 	myHsForThem := myControl.GetFromUDP(true)
@@ -313,7 +405,7 @@ func TestStage1Race(t *testing.T) {

 	r.Log("Spin until connection manager tears down a tunnel")

-	for len(myControl.GetHostmap().Indexes)+len(theirControl.GetHostmap().Indexes) > 2 {
+	for myControl.GetHostmapIndexCount()+theirControl.GetHostmapIndexCount() > 2 {
 		assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
 		t.Log("Connection manager hasn't ticked yet")
 		time.Sleep(time.Second)
@@ -336,6 +428,7 @@ 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)
@@ -353,18 +446,20 @@ func TestUncleanShutdownRaceLoser(t *testing.T) {
 	theirControl.Start()

 	r.Log("Trigger a handshake from me to them")
-	myControl.InjectTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me"))
+	myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me")))

 	p := r.RouteForAllUntilTxTun(theirControl)
 	assertUdpPacket(t, []byte("Hi from me"), p, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), 80, 80)

 	r.Log("Nuke my hostmap")
 	myHostmap := myControl.GetHostmap()
+	myHostmap.Lock()
 	myHostmap.Hosts = map[netip.Addr]*nebula.HostInfo{}
 	myHostmap.Indexes = map[uint32]*nebula.HostInfo{}
 	myHostmap.RemoteIndexes = map[uint32]*nebula.HostInfo{}
+	myHostmap.Unlock()

-	myControl.InjectTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me again"))
+	myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me again")))
 	p = r.RouteForAllUntilTxTun(theirControl)
 	assertUdpPacket(t, []byte("Hi from me again"), p, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), 80, 80)

@@ -372,10 +467,10 @@ func TestUncleanShutdownRaceLoser(t *testing.T) {
 	assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)

 	r.Log("Wait for the dead index to go away")
-	start := len(theirControl.GetHostmap().Indexes)
+	start := theirControl.GetHostmapIndexCount()
 	for {
 		assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
-		if len(theirControl.GetHostmap().Indexes) < start {
+		if theirControl.GetHostmapIndexCount() < start {
 			break
 		}
 		time.Sleep(time.Second)
@@ -385,6 +480,7 @@ 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)
@@ -402,7 +498,7 @@ func TestUncleanShutdownRaceWinner(t *testing.T) {
 	theirControl.Start()

 	r.Log("Trigger a handshake from me to them")
-	myControl.InjectTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me"))
+	myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me")))

 	p := r.RouteForAllUntilTxTun(theirControl)
 	assertUdpPacket(t, []byte("Hi from me"), p, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), 80, 80)
@@ -410,11 +506,13 @@ func TestUncleanShutdownRaceWinner(t *testing.T) {

 	r.Log("Nuke my hostmap")
 	theirHostmap := theirControl.GetHostmap()
+	theirHostmap.Lock()
 	theirHostmap.Hosts = map[netip.Addr]*nebula.HostInfo{}
 	theirHostmap.Indexes = map[uint32]*nebula.HostInfo{}
 	theirHostmap.RemoteIndexes = map[uint32]*nebula.HostInfo{}
+	theirHostmap.Unlock()

-	theirControl.InjectTunUDPPacket(myVpnIpNet[0].Addr(), 80, theirVpnIpNet[0].Addr(), 80, []byte("Hi from them again"))
+	theirControl.InjectTunPacket(BuildTunUDPPacket(myVpnIpNet[0].Addr(), 80, theirVpnIpNet[0].Addr(), 80, []byte("Hi from them again")))
 	p = r.RouteForAllUntilTxTun(myControl)
 	assertUdpPacket(t, []byte("Hi from them again"), p, theirVpnIpNet[0].Addr(), myVpnIpNet[0].Addr(), 80, 80)
 	r.RenderHostmaps("Derp hostmaps", myControl, theirControl)
@@ -423,10 +521,10 @@ func TestUncleanShutdownRaceWinner(t *testing.T) {
 	assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)

 	r.Log("Wait for the dead index to go away")
-	start := len(myControl.GetHostmap().Indexes)
+	start := myControl.GetHostmapIndexCount()
 	for {
 		assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
-		if len(myControl.GetHostmap().Indexes) < start {
+		if myControl.GetHostmapIndexCount() < start {
 			break
 		}
 		time.Sleep(time.Second)
@@ -436,6 +534,7 @@ 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}})
@@ -456,7 +555,37 @@ func TestRelays(t *testing.T) {
 	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"))
+	myControl.InjectTunPacket(BuildTunUDPPacket(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 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.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me")))

 	p := r.RouteForAllUntilTxTun(theirControl)
 	r.Log("Assert the tunnel works")
@@ -465,6 +594,7 @@ func TestRelays(t *testing.T) {
 }

 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}})
@@ -485,14 +615,14 @@ func TestReestablishRelays(t *testing.T) {
 	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"))
+	myControl.InjectTunPacket(BuildTunUDPPacket(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)

 	t.Log("Ensure packet traversal from them to me via the relay")
-	theirControl.InjectTunUDPPacket(myVpnIpNet[0].Addr(), 80, theirVpnIpNet[0].Addr(), 80, []byte("Hi from them"))
+	theirControl.InjectTunPacket(BuildTunUDPPacket(myVpnIpNet[0].Addr(), 80, theirVpnIpNet[0].Addr(), 80, []byte("Hi from them")))

 	p = r.RouteForAllUntilTxTun(myControl)
 	r.Log("Assert the tunnel works")
@@ -502,12 +632,12 @@ func TestReestablishRelays(t *testing.T) {
 	r.Log("Close the tunnel")
 	relayControl.CloseTunnel(theirVpnIpNet[0].Addr(), true)

-	start := len(myControl.GetHostmap().Indexes)
-	curIndexes := len(myControl.GetHostmap().Indexes)
+	start := myControl.GetHostmapIndexCount()
+	curIndexes := myControl.GetHostmapIndexCount()
 	for curIndexes >= start {
-		curIndexes = len(myControl.GetHostmap().Indexes)
+		curIndexes = myControl.GetHostmapIndexCount()
 		r.Logf("Wait for the dead index to go away:start=%v indexes, current=%v indexes", start, curIndexes)
-		myControl.InjectTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me should fail"))
+		myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me should fail")))

 		r.RouteForAllExitFunc(func(p *udp.Packet, c *nebula.Control) router.ExitType {
 			return router.RouteAndExit
@@ -524,7 +654,7 @@ func TestReestablishRelays(t *testing.T) {
 		myControl.InjectLightHouseAddr(relayVpnIpNet[0].Addr(), relayUdpAddr)
 		myControl.InjectRelays(theirVpnIpNet[0].Addr(), []netip.Addr{relayVpnIpNet[0].Addr()})
 		relayControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
-		myControl.InjectTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me"))
+		myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me")))

 		p = r.RouteForAllUntilTxTun(theirControl)
 		r.Log("Assert the tunnel works")
@@ -559,7 +689,7 @@ func TestReestablishRelays(t *testing.T) {
 	t.Log("Assert the tunnel works the other way, too")
 	for {
 		t.Log("RouteForAllUntilTxTun")
-		theirControl.InjectTunUDPPacket(myVpnIpNet[0].Addr(), 80, theirVpnIpNet[0].Addr(), 80, []byte("Hi from them"))
+		theirControl.InjectTunPacket(BuildTunUDPPacket(myVpnIpNet[0].Addr(), 80, theirVpnIpNet[0].Addr(), 80, []byte("Hi from them")))

 		p = r.RouteForAllUntilTxTun(myControl)
 		r.Log("Assert the tunnel works")
@@ -595,7 +725,72 @@ func TestReestablishRelays(t *testing.T) {

 }

+func TestRelayHandshakeOverDisestablishedEntry(t *testing.T) {
+	t.Parallel()
+	// If them tears down the tunnel while me keeps Established relay state, me's next
+	// handshake flows through the relay with no fresh CreateRelayRequest and lands on
+	// them's Disestablished terminal relay entry. them must re-establish that entry, or
+	// its first transmit deletes its only relay and the tunnel is born transmit-dead:
+	// them can receive but every send is silently dropped.
+	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(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.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me")))
+
+	p := r.RouteForAllUntilTxTun(theirControl)
+	assertUdpPacket(t, []byte("Hi from me"), p, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), 80, 80)
+	oldIdx := myControl.GetHostInfoByVpnAddr(theirVpnIpNet[0].Addr(), false).LocalIndex
+
+	t.Log("Close the tunnel on them only, marking their relay entry Disestablished")
+	theirControl.CloseTunnel(myVpnIpNet[0].Addr(), true)
+
+	t.Log("Re-handshake from me, riding the still-Established relay state")
+	myControl.ReHandshake(theirVpnIpNet[0].Addr())
+	for {
+		h := myControl.GetHostInfoByVpnAddr(theirVpnIpNet[0].Addr(), false)
+		if h != nil && h.LocalIndex != oldIdx && h.RemoteIndex != 0 {
+			break
+		}
+		r.RouteForAllExitFunc(func(*udp.Packet, *nebula.Control) router.ExitType {
+			return router.RouteAndExit
+		})
+	}
+
+	hAtThem := theirControl.GetHostInfoByVpnAddr(myVpnIpNet[0].Addr(), false)
+	require.NotNil(t, hAtThem, "them should have completed the relayed handshake")
+	require.Equal(t, []netip.Addr{relayVpnIpNet[0].Addr()}, hAtThem.CurrentRelaysToMe, "them should know a relay for the new tunnel")
+
+	t.Log("Send from them to me; their only relay entry must survive the transmit")
+	theirControl.InjectTunPacket(BuildTunUDPPacket(myVpnIpNet[0].Addr(), 80, theirVpnIpNet[0].Addr(), 80, []byte("Hi from them")))
+	require.Never(t, func() bool {
+		h := theirControl.GetHostInfoByVpnAddr(myVpnIpNet[0].Addr(), false)
+		return h == nil || len(h.CurrentRelaysToMe) == 0
+	}, time.Second, 10*time.Millisecond, "them deleted its only relay entry; the tunnel is permanently transmit-dead")
+
+	p = r.RouteForAllUntilTxTun(myControl)
+	assertUdpPacket(t, []byte("Hi from them"), p, theirVpnIpNet[0].Addr(), myVpnIpNet[0].Addr(), 80, 80)
+	r.RenderHostmaps("Final hostmaps", myControl, relayControl, theirControl)
+}
+
 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}})
@@ -628,8 +823,8 @@ func TestStage1RaceRelays(t *testing.T) {
 	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")
-	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"))
+	myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me")))
+	theirControl.InjectTunPacket(BuildTunUDPPacket(myVpnIpNet[0].Addr(), 80, theirVpnIpNet[0].Addr(), 80, []byte("Hi from them")))

 	r.Log("Wait for a packet from them to me")
 	p := r.RouteForAllUntilTxTun(myControl)
@@ -643,12 +838,12 @@ 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 +865,41 @@ 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"))
+	myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me")))
+	theirControl.InjectTunPacket(BuildTunUDPPacket(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")
-	l.Info("Wait for a packet from them to me; myControl")
+	r.Log("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")
-	l.WithFields(
-		logrus.Fields{
-			"myControl":    len(myControl.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)
+	t.Logf("Waiting for hostinfos to be removed... myControl=%d theirControl=%d relayControl=%d",
+		myControl.GetHostmapIndexCount(),
+		theirControl.GetHostmapIndexCount(),
+		relayControl.GetHostmapIndexCount(),
+	)
+	hostInfos := myControl.GetHostmapIndexCount() + theirControl.GetHostmapIndexCount() + relayControl.GetHostmapIndexCount()
 	retries := 60
 	for hostInfos > 6 && retries > 0 {
-		hostInfos = len(myControl.GetHostmap().Indexes) + len(theirControl.GetHostmap().Indexes) + len(relayControl.GetHostmap().Indexes)
-		l.WithFields(
-			logrus.Fields{
-				"myControl":    len(myControl.GetHostmap().Indexes),
-				"theirControl": len(theirControl.GetHostmap().Indexes),
-				"relayControl": len(relayControl.GetHostmap().Indexes),
-			}).Info("Waiting for hostinfos to be removed...")
+		hostInfos = myControl.GetHostmapIndexCount() + theirControl.GetHostmapIndexCount() + relayControl.GetHostmapIndexCount()
+		t.Logf("Waiting for hostinfos to be removed... myControl=%d theirControl=%d relayControl=%d",
+			myControl.GetHostmapIndexCount(),
+			theirControl.GetHostmapIndexCount(),
+			relayControl.GetHostmapIndexCount(),
+		)
 		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 +907,6 @@ 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 +915,7 @@ 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}})
@@ -749,7 +936,7 @@ func TestRehandshakingRelays(t *testing.T) {
 	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"))
+	myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me")))

 	p := r.RouteForAllUntilTxTun(theirControl)
 	r.Log("Assert the tunnel works")
@@ -805,24 +992,24 @@ func TestRehandshakingRelays(t *testing.T) {
 	assertTunnel(t, theirVpnIpNet[0].Addr(), myVpnIpNet[0].Addr(), theirControl, myControl, r)
 	r.RenderHostmaps("working hostmaps", myControl, relayControl, theirControl)
 	// We should have two hostinfos on all sides
-	for len(myControl.GetHostmap().Indexes) != 2 {
-		t.Logf("Waiting for myControl hostinfos (%v != 2) to get cleaned up from lack of use...", len(myControl.GetHostmap().Indexes))
+	for myControl.GetHostmapIndexCount() != 2 {
+		t.Logf("Waiting for myControl hostinfos (%v != 2) to get cleaned up from lack of use...", myControl.GetHostmapIndexCount())
 		r.Log("Assert the relay tunnel still works")
 		assertTunnel(t, theirVpnIpNet[0].Addr(), myVpnIpNet[0].Addr(), theirControl, myControl, r)
 		r.Log("yupitdoes")
 		time.Sleep(time.Second)
 	}
 	t.Logf("myControl hostinfos got cleaned up!")
-	for len(theirControl.GetHostmap().Indexes) != 2 {
-		t.Logf("Waiting for theirControl hostinfos (%v != 2) to get cleaned up from lack of use...", len(theirControl.GetHostmap().Indexes))
+	for theirControl.GetHostmapIndexCount() != 2 {
+		t.Logf("Waiting for theirControl hostinfos (%v != 2) to get cleaned up from lack of use...", theirControl.GetHostmapIndexCount())
 		r.Log("Assert the relay tunnel still works")
 		assertTunnel(t, theirVpnIpNet[0].Addr(), myVpnIpNet[0].Addr(), theirControl, myControl, r)
 		r.Log("yupitdoes")
 		time.Sleep(time.Second)
 	}
 	t.Logf("theirControl hostinfos got cleaned up!")
-	for len(relayControl.GetHostmap().Indexes) != 2 {
-		t.Logf("Waiting for relayControl hostinfos (%v != 2) to get cleaned up from lack of use...", len(relayControl.GetHostmap().Indexes))
+	for relayControl.GetHostmapIndexCount() != 2 {
+		t.Logf("Waiting for relayControl hostinfos (%v != 2) to get cleaned up from lack of use...", relayControl.GetHostmapIndexCount())
 		r.Log("Assert the relay tunnel still works")
 		assertTunnel(t, theirVpnIpNet[0].Addr(), myVpnIpNet[0].Addr(), theirControl, myControl, r)
 		r.Log("yupitdoes")
@@ -832,6 +1019,7 @@ 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}})
@@ -853,7 +1041,7 @@ func TestRehandshakingRelaysPrimary(t *testing.T) {
 	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"))
+	myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me")))

 	p := r.RouteForAllUntilTxTun(theirControl)
 	r.Log("Assert the tunnel works")
@@ -909,24 +1097,24 @@ func TestRehandshakingRelaysPrimary(t *testing.T) {
 	assertTunnel(t, theirVpnIpNet[0].Addr(), myVpnIpNet[0].Addr(), theirControl, myControl, r)
 	r.RenderHostmaps("working hostmaps", myControl, relayControl, theirControl)
 	// We should have two hostinfos on all sides
-	for len(myControl.GetHostmap().Indexes) != 2 {
-		t.Logf("Waiting for myControl hostinfos (%v != 2) to get cleaned up from lack of use...", len(myControl.GetHostmap().Indexes))
+	for myControl.GetHostmapIndexCount() != 2 {
+		t.Logf("Waiting for myControl hostinfos (%v != 2) to get cleaned up from lack of use...", myControl.GetHostmapIndexCount())
 		r.Log("Assert the relay tunnel still works")
 		assertTunnel(t, theirVpnIpNet[0].Addr(), myVpnIpNet[0].Addr(), theirControl, myControl, r)
 		r.Log("yupitdoes")
 		time.Sleep(time.Second)
 	}
 	t.Logf("myControl hostinfos got cleaned up!")
-	for len(theirControl.GetHostmap().Indexes) != 2 {
-		t.Logf("Waiting for theirControl hostinfos (%v != 2) to get cleaned up from lack of use...", len(theirControl.GetHostmap().Indexes))
+	for theirControl.GetHostmapIndexCount() != 2 {
+		t.Logf("Waiting for theirControl hostinfos (%v != 2) to get cleaned up from lack of use...", theirControl.GetHostmapIndexCount())
 		r.Log("Assert the relay tunnel still works")
 		assertTunnel(t, theirVpnIpNet[0].Addr(), myVpnIpNet[0].Addr(), theirControl, myControl, r)
 		r.Log("yupitdoes")
 		time.Sleep(time.Second)
 	}
 	t.Logf("theirControl hostinfos got cleaned up!")
-	for len(relayControl.GetHostmap().Indexes) != 2 {
-		t.Logf("Waiting for relayControl hostinfos (%v != 2) to get cleaned up from lack of use...", len(relayControl.GetHostmap().Indexes))
+	for relayControl.GetHostmapIndexCount() != 2 {
+		t.Logf("Waiting for relayControl hostinfos (%v != 2) to get cleaned up from lack of use...", relayControl.GetHostmapIndexCount())
 		r.Log("Assert the relay tunnel still works")
 		assertTunnel(t, theirVpnIpNet[0].Addr(), myVpnIpNet[0].Addr(), theirControl, myControl, r)
 		r.Log("yupitdoes")
@@ -936,6 +1124,7 @@ 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)
@@ -1002,7 +1191,7 @@ func TestRehandshaking(t *testing.T) {
 	theirConfig.ReloadConfigString(string(rc))

 	r.Log("Spin until there is only 1 tunnel")
-	for len(myControl.GetHostmap().Indexes)+len(theirControl.GetHostmap().Indexes) > 2 {
+	for myControl.GetHostmapIndexCount()+theirControl.GetHostmapIndexCount() > 2 {
 		assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
 		t.Log("Connection manager hasn't ticked yet")
 		time.Sleep(time.Second)
@@ -1031,6 +1220,7 @@ 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{})
@@ -1101,7 +1291,7 @@ func TestRehandshakingLoser(t *testing.T) {
 	myConfig.ReloadConfigString(string(rc))

 	r.Log("Spin until there is only 1 tunnel")
-	for len(myControl.GetHostmap().Indexes)+len(theirControl.GetHostmap().Indexes) > 2 {
+	for myControl.GetHostmapIndexCount()+theirControl.GetHostmapIndexCount() > 2 {
 		assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
 		t.Log("Connection manager hasn't ticked yet")
 		time.Sleep(time.Second)
@@ -1129,6 +1319,7 @@ 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
@@ -1152,8 +1343,8 @@ func TestRaceRegression(t *testing.T) {
 	//them rx stage:2 initiatorIndex=120607833 responderIndex=4209862089

 	t.Log("Start both handshakes")
-	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"))
+	myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me")))
+	theirControl.InjectTunPacket(BuildTunUDPPacket(myVpnIpNet[0].Addr(), 80, theirVpnIpNet[0].Addr(), 80, []byte("Hi from them")))

 	t.Log("Get both stage 1")
 	myStage1ForThem := myControl.GetFromUDP(true)
@@ -1189,6 +1380,7 @@ 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 +1419,262 @@ 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.InjectTunPacket(BuildTunUDPPacket(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.InjectTunPacket(BuildTunUDPPacket(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()
+}
+
+func TestMultiVpnAddrDeletePrimaryKeepsSecondAddr(t *testing.T) {
+	t.Parallel()
+	// Regression for the hostmap multi-vpnAddr delete bug. A dual-stack (v4+v6) V2-cert peer that
+	// handshakes twice at once ends up with two hostinfos linked in the shared next/prev chain, with the
+	// primary owning both addresses. Deleting that primary (e.g. connection manager dropping it, a
+	// CloseTunnel, a collision) must promote the surviving sibling for EVERY address. The pre-fix code
+	// unlinked the chain once per address, so it promoted the sibling for the first address and orphaned
+	// the second: the peer stayed reachable at its v4 addr but not its v6 addr despite a live tunnel.
+
+	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,fd00::1/64", nil)
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version2, ca, caKey, "them", "10.128.0.2/24,fd00::2/64", nil)
+
+	// This bug only exists for peers carrying more than one vpn address
+	require.Len(t, theirVpnIpNet, 2)
+	theirV4 := theirVpnIpNet[0].Addr()
+	theirV6 := theirVpnIpNet[1].Addr()
+
+	// Put their info in our lighthouse and vice versa
+	myControl.InjectLightHouseAddr(theirV4, theirUdpAddr)
+	theirControl.InjectLightHouseAddr(myVpnIpNet[0].Addr(), myUdpAddr)
+
+	// Build a router so we don't have to reason who gets which packet
+	r := router.NewR(t, myControl, theirControl)
+	defer r.RenderFlow()
+
+	myControl.Start()
+	theirControl.Start()
+
+	// Race a handshake so both of us build a hostinfo for the other, leaving my hostmap with a single
+	// host (them) backed by two linked hostinfos, just like TestStage1Race.
+	myControl.InjectTunPacket(BuildTunUDPPacket(theirV4, 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me")))
+	theirControl.InjectTunPacket(BuildTunUDPPacket(myVpnIpNet[0].Addr(), 80, theirV4, 80, []byte("Hi from them")))
+
+	myHsForThem := myControl.GetFromUDP(true)
+	theirHsForMe := theirControl.GetFromUDP(true)
+
+	r.InjectUDPPacket(theirControl, myControl, theirHsForMe)
+	r.InjectUDPPacket(myControl, theirControl, myHsForThem)
+
+	r.RouteForAllUntilTxTun(theirControl)
+	r.RouteForAllUntilTxTun(myControl)
+
+	r.RenderHostmaps("Racing hostmaps", myControl, theirControl)
+
+	// Two hostinfos for them means the shared next/prev chain has a sibling to promote. The Hosts map has
+	// one entry per vpn address (two, for dual stack), so the index count is what tells us there are two
+	// hostinfos.
+	require.Len(t, myControl.ListHostmapIndexes(false), 2)
+
+	// The primary owns both of their addresses
+	primaryV4 := myControl.GetHostInfoByVpnAddr(theirV4, false)
+	primaryV6 := myControl.GetHostInfoByVpnAddr(theirV6, false)
+	require.NotNil(t, primaryV4)
+	require.NotNil(t, primaryV6)
+	require.Equal(t, primaryV4.LocalIndex, primaryV6.LocalIndex, "both addrs should point at the same primary")
+
+	// Delete the primary tunnel. localOnly so we don't perturb their side, we only care about my hostmap.
+	require.True(t, myControl.CloseTunnel(theirV4, true))
+
+	// The surviving sibling must still serve BOTH addresses.
+	survivorV4 := myControl.GetHostInfoByVpnAddr(theirV4, false)
+	survivorV6 := myControl.GetHostInfoByVpnAddr(theirV6, false)
+	require.NotNil(t, survivorV4, "v4 addr should still resolve to the surviving tunnel")
+	// Pre-fix this is nil: the second address was orphaned when the primary was deleted.
+	require.NotNil(t, survivorV6, "v6 addr was orphaned after deleting the primary (multi-vpnAddr delete bug)")
+	assert.Equal(t, survivorV4.LocalIndex, survivorV6.LocalIndex, "both addrs should promote to the same survivor")
+	assert.NotEqual(t, primaryV4.LocalIndex, survivorV4.LocalIndex, "a different hostinfo should now be primary")
+
+	r.RenderHostmaps("Final hostmaps", myControl, theirControl)
+
+	myControl.Stop()
+	theirControl.Stop()
+}
@@ -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{{
-				"proto": "any",
-				"port":  "any",
-				"host":  "any",
-			}},
+			"inbound": firewallInbound,
 		},
 		//"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":            l.Level.String(),
+			"level": testLogLevelName(),
 		},
 		"timers": m{
 			"pending_deletion_interval": 2,
@@ -129,6 +172,108 @@ func newSimpleServer(v cert.Version, caCrt cert.Certificate, caKey []byte, name
 	return control, vpnNetworks, udpAddr, c
 }

+// newServer creates a nebula instance with fewer assumptions
+func newServer(caCrt []cert.Certificate, certs []cert.Certificate, key []byte, overrides m) (*nebula.Control, []netip.Prefix, netip.AddrPort, *config.C) {
+	l := NewTestLogger()
+
+	vpnNetworks := certs[len(certs)-1].Networks()
+
+	var udpAddr netip.AddrPort
+	if vpnNetworks[0].Addr().Is4() {
+		budpIp := vpnNetworks[0].Addr().As4()
+		budpIp[1] -= 128
+		udpAddr = netip.AddrPortFrom(netip.AddrFrom4(budpIp), 4242)
+	} else {
+		budpIp := vpnNetworks[0].Addr().As16()
+		// beef for funsies
+		budpIp[2] = 190
+		budpIp[3] = 239
+		udpAddr = netip.AddrPortFrom(netip.AddrFrom16(budpIp), 4242)
+	}
+
+	caStr := ""
+	for _, ca := range caCrt {
+		x, err := ca.MarshalPEM()
+		if err != nil {
+			panic(err)
+		}
+		caStr += string(x)
+	}
+	certStr := ""
+	for _, c := range certs {
+		x, err := c.MarshalPEM()
+		if err != nil {
+			panic(err)
+		}
+		certStr += string(x)
+	}
+
+	mc := m{
+		"pki": m{
+			"ca":   caStr,
+			"cert": certStr,
+			"key":  string(key),
+		},
+		//"tun": m{"disabled": true},
+		"firewall": m{
+			"outbound": []m{{
+				"proto": "any",
+				"port":  "any",
+				"host":  "any",
+			}},
+			"inbound": []m{{
+				"proto": "any",
+				"port":  "any",
+				"host":  "any",
+			}},
+		},
+		//"handshakes": m{
+		//	"try_interval": "1s",
+		//},
+		"listen": m{
+			"host": udpAddr.Addr().String(),
+			"port": udpAddr.Port(),
+		},
+		"logging": m{
+			"level": testLogLevelName(),
+		},
+		"timers": m{
+			"pending_deletion_interval": 2,
+			"connection_alive_interval": 2,
+		},
+	}
+
+	if overrides != nil {
+		final := m{}
+		err := mergo.Merge(&final, overrides, mergo.WithAppendSlice)
+		if err != nil {
+			panic(err)
+		}
+		err = mergo.Merge(&final, mc, mergo.WithAppendSlice)
+		if err != nil {
+			panic(err)
+		}
+		mc = final
+	}
+
+	cb, err := yaml.Marshal(mc)
+	if err != nil {
+		panic(err)
+	}
+
+	c := config.NewC(l)
+	cStr := string(cb)
+	c.LoadString(cStr)
+
+	control, err := nebula.Main(c, false, "e2e-test", l, nil)
+
+	if err != nil {
+		panic(err)
+	}
+
+	return control, vpnNetworks, udpAddr, c
+}
+
 type doneCb func()

 func deadline(t *testing.T, seconds time.Duration) doneCb {
@@ -147,26 +292,26 @@ 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"))
+	controlB.InjectTunPacket(BuildTunUDPPacket(vpnIpA, 80, vpnIpB, 90, []byte("Hi from B")))
 	bPacket := r.RouteForAllUntilTxTun(controlA)
 	assertUdpPacket(t, []byte("Hi from B"), bPacket, vpnIpB, vpnIpA, 90, 80)

 	// And once more from me to them
-	controlA.InjectTunUDPPacket(vpnIpB, 80, vpnIpA, 90, []byte("Hello from A"))
+	controlA.InjectTunPacket(BuildTunUDPPacket(vpnIpB, 80, vpnIpA, 90, []byte("Hello from A")))
 	aPacket := r.RouteForAllUntilTxTun(controlB)
 	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")
@@ -180,7 +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 {
@@ -188,7 +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")
@@ -207,7 +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")
@@ -234,24 +379,87 @@ func getAddrs(ns []netip.Prefix) []netip.Addr {
 	return a
 }

-func NewTestLogger() *logrus.Logger {
-	l := logrus.New()
-
+func NewTestLogger() *slog.Logger {
 	v := os.Getenv("TEST_LOGS")
 	if v == "" {
-		l.SetOutput(io.Discard)
-		l.SetLevel(logrus.PanicLevel)
-		return l
+		return slog.New(slog.NewTextHandler(io.Discard, nil))
 	}

+	level := slog.LevelInfo
 	switch v {
 	case "2":
-		l.SetLevel(logrus.DebugLevel)
+		level = slog.LevelDebug
 	case "3":
-		l.SetLevel(logrus.TraceLevel)
-	default:
-		l.SetLevel(logrus.InfoLevel)
+		level = logging.LevelTrace
+	}
+	return slog.New(slog.NewTextHandler(os.Stderr, &slog.HandlerOptions{Level: level}))
+}
+
+// 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"
+}
+
+// BuildTunUDPPacket assembles an IP+UDP packet suitable for Control.InjectTunPacket.
+// Using UDP here because it's a simpler protocol.
+func BuildTunUDPPacket(toAddr netip.Addr, toPort uint16, fromAddr netip.Addr, fromPort uint16, data []byte) []byte {
+	serialize := make([]gopacket.SerializableLayer, 0)
+	var netLayer gopacket.NetworkLayer
+	if toAddr.Is6() {
+		if !fromAddr.Is6() {
+			panic("Cant send ipv6 to ipv4")
+		}
+		ip := &layers.IPv6{
+			Version:    6,
+			NextHeader: layers.IPProtocolUDP,
+			SrcIP:      fromAddr.Unmap().AsSlice(),
+			DstIP:      toAddr.Unmap().AsSlice(),
+		}
+		serialize = append(serialize, ip)
+		netLayer = ip
+	} else {
+		if !fromAddr.Is4() {
+			panic("Cant send ipv4 to ipv6")
+		}
+
+		ip := &layers.IPv4{
+			Version:  4,
+			TTL:      64,
+			Protocol: layers.IPProtocolUDP,
+			SrcIP:    fromAddr.Unmap().AsSlice(),
+			DstIP:    toAddr.Unmap().AsSlice(),
+		}
+		serialize = append(serialize, ip)
+		netLayer = ip
 	}

-	return l
+	udp := layers.UDP{
+		SrcPort: layers.UDPPort(fromPort),
+		DstPort: layers.UDPPort(toPort),
+	}
+	if err := udp.SetNetworkLayerForChecksum(netLayer); err != nil {
+		panic(err)
+	}
+
+	buffer := gopacket.NewSerializeBuffer()
+	opt := gopacket.SerializeOptions{
+		ComputeChecksums: true,
+		FixLengths:       true,
+	}
+
+	serialize = append(serialize, &udp, gopacket.Payload(data))
+	if err := gopacket.SerializeLayers(buffer, opt, serialize...); err != nil {
+		panic(err)
+	}
+
+	return buffer.Bytes()
 }
@@ -0,0 +1,47 @@
+//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.
+//
+// Intentionally NOT t.Parallel()'d: concurrent tests would have their own
+// goroutines running and trip the assertion.
+func TestNoGoroutineLeaks(t *testing.T) {
+	defer goleak.VerifyNone(t)
+
+	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)
+}
@@ -0,0 +1,225 @@
+//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"
+	"github.com/stretchr/testify/require"
+)
+
+// reportedAddrs is what the lighthouse would hand a peer asking where vpnAddr is.
+func reportedAddrs(t *testing.T, lh *nebula.Control, vpnAddr netip.Addr) []netip.AddrPort {
+	t.Helper()
+	cm := lh.QueryLighthouse(vpnAddr)
+	if cm == nil {
+		return nil
+	}
+	var out []netip.AddrPort
+	for _, c := range *cm {
+		out = append(out, c.Reported...)
+		out = append(out, c.Learned...)
+	}
+	return out
+}
+
+// waitForLighthouseMsg routes until a lighthouse message lands on lh, or gives up. Reports whether one arrived.
+func waitForLighthouseMsg(t *testing.T, r *router.R, lh *nebula.Control, wait time.Duration) bool {
+	t.Helper()
+	h := &header.H{}
+	return r.RouteForAllExitFuncOrTimeout(wait, func(p *udp.Packet, c *nebula.Control) router.ExitType {
+		if c != lh {
+			return router.KeepRouting
+		}
+		// Punches are a single byte and never parse, they are just not what we are after
+		if err := h.Parse(p.Data); err != nil {
+			return router.KeepRouting
+		}
+		if h.Type == header.LightHouse {
+			return router.RouteAndExit
+		}
+		return router.KeepRouting
+	})
+}
+
+// A laptop that changes networks has to tell the lighthouse promptly, otherwise the lighthouse keeps handing peers
+// the old address and their punches land nowhere. On a long lighthouse interval the only thing that closes that
+// window is the rebind, which on darwin the network change monitor drives. The e2e build compiles the monitor out,
+// so we call RebindUDPServer directly, which is the same thing the monitor does.
+func TestRebindSendsLighthouseUpdate(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", m{
+		"lighthouse": m{"am_lighthouse": true},
+	})
+
+	// 600s interval, so nothing scheduled can send an update during this test. A rebind is the only thing that can.
+	myControl, _, _, _ := newSimpleServer(cert.Version2, ca, caKey, "me", "10.128.0.2/24", m{
+		"lighthouse": m{
+			"hosts":    []any{lhVpnIpNet[0].Addr().String()},
+			"interval": 600,
+		},
+		"static_host_map": m{
+			lhVpnIpNet[0].Addr().String(): []any{lhUdpAddr.String()},
+		},
+	})
+
+	r := router.NewR(t, lhControl, myControl)
+	defer r.RenderFlow()
+
+	lhControl.Start()
+	myControl.Start()
+
+	// Let the startup registration finish, then clear everything it left behind
+	require.True(t, waitForLighthouseMsg(t, r, lhControl, time.Second*5), "expected an initial registration")
+	r.RouteFor(time.Millisecond * 400)
+
+	// Nothing should be talking to the lighthouse on its own now
+	require.False(t, waitForLighthouseMsg(t, r, lhControl, time.Millisecond*200),
+		"nothing should reach the lighthouse before the rebind")
+
+	myControl.RebindUDPServer()
+
+	assert.True(t, waitForLighthouseMsg(t, r, lhControl, time.Second*5),
+		"a rebind should push an update to the lighthouse rather than waiting out the interval")
+
+	lhControl.Stop()
+	myControl.Stop()
+}
+
+// The other half of a rebind: every live tunnel requeries the lighthouse on its next send. That query is what makes
+// the lighthouse tell the peer to punch toward our new address, which is the part that actually revives a tunnel
+// whose remote NAT state died while we were on a different network.
+func TestRebindRequeriesPeersOnNextSend(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", m{
+		"lighthouse": m{"am_lighthouse": true},
+	})
+
+	lhCfg := m{
+		"lighthouse": m{
+			"hosts":    []any{lhVpnIpNet[0].Addr().String()},
+			"interval": 600,
+			// Without this the peers advertise this machine's real addresses and then try to punch at them,
+			// which the router has no route for.
+			"local_allow_list": m{
+				"10.0.0.0/24": true,
+				"::/0":        false,
+			},
+		},
+		"static_host_map": m{
+			lhVpnIpNet[0].Addr().String(): []any{lhUdpAddr.String()},
+		},
+	}
+
+	myControl, myVpnIpNet, myUdpAddr, _ := newSimpleServer(cert.Version2, ca, caKey, "me", "10.128.0.2/24", lhCfg)
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version2, ca, caKey, "them", "10.128.0.3/24", lhCfg)
+
+	r := router.NewR(t, lhControl, myControl, theirControl)
+	defer r.RenderFlow()
+
+	lhControl.Start()
+	myControl.Start()
+	theirControl.Start()
+	r.RouteFor(time.Millisecond * 500)
+
+	// Point the peers at each other directly, this test is about the rebind and not about lighthouse discovery
+	myControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
+	theirControl.InjectLightHouseAddr(myVpnIpNet[0].Addr(), myUdpAddr)
+
+	myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("initial")))
+	r.RouteFor(time.Second)
+	require.NotNil(t, myControl.GetHostInfoByVpnAddr(theirVpnIpNet[0].Addr(), false), "expected a tunnel to them")
+	r.RouteFor(time.Millisecond * 300)
+
+	// Assert on what the peer sees rather than on lighthouse traffic. A query for them makes the lighthouse send
+	// them a punch notification, which is the whole point. Our own update to the lighthouse sends them nothing,
+	// so this cannot be satisfied by the update the rebind itself pushes.
+	myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("quiet")))
+	require.False(t, waitForLighthouseMsg(t, r, theirControl, time.Millisecond*300),
+		"an ordinary send should not requery the lighthouse")
+
+	myControl.RebindUDPServer()
+	r.RouteFor(time.Millisecond * 300) // let the update the rebind itself sends pass by
+
+	myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("after rebind")))
+	assert.True(t, waitForLighthouseMsg(t, r, theirControl, time.Second*5),
+		"the first send after a rebind should requery the lighthouse, which then tells the peer to punch at us")
+
+	lhControl.Stop()
+	myControl.Stop()
+	theirControl.Stop()
+}
+
+// The scenario this whole thing exists for: a laptop sleeps at the office and wakes up at home on a new address.
+// Until it tells the lighthouse, the lighthouse keeps handing peers the office address, so their punches land
+// nowhere and the tunnel stays dead. On a long interval the rebind is the only thing that closes that window.
+func TestRebindAdvertisesNewAddressAfterMove(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", m{
+		"lighthouse": m{"am_lighthouse": true},
+	})
+
+	myControl, myVpnIpNet, myUdpAddr, _ := newSimpleServer(cert.Version2, ca, caKey, "me", "10.128.0.2/24", m{
+		"lighthouse": m{
+			"hosts":    []any{lhVpnIpNet[0].Addr().String()},
+			"interval": 600,
+		},
+		"static_host_map": m{
+			lhVpnIpNet[0].Addr().String(): []any{lhUdpAddr.String()},
+		},
+	})
+
+	// Advertise wherever we currently are rather than this machine's real NICs, read fresh each time so a move
+	// is picked up.
+	myControl.SetLocalAddrsFn(func(*nebula.LocalAllowList) []netip.Addr {
+		return []netip.Addr{myControl.GetUDPAddr().Addr()}
+	})
+
+	r := router.NewR(t, lhControl, myControl)
+	defer r.RenderFlow()
+
+	lhControl.Start()
+	myControl.Start()
+
+	require.True(t, waitForLighthouseMsg(t, r, lhControl, time.Second*5), "expected an initial registration")
+	r.RouteFor(time.Millisecond * 400)
+
+	require.Contains(t, reportedAddrs(t, lhControl, myVpnIpNet[0].Addr()), myUdpAddr,
+		"the lighthouse should know the address we started on")
+
+	// Wake up somewhere else
+	newAddr := netip.MustParseAddrPort("10.0.0.99:4242")
+	myControl.SetUDPAddr(newAddr)
+	r.AddRoute(newAddr.Addr(), newAddr.Port(), myControl)
+
+	// Nothing has told the lighthouse, and with interval 600 nothing scheduled will
+	r.RouteFor(time.Millisecond * 400)
+	require.NotContains(t, reportedAddrs(t, lhControl, myVpnIpNet[0].Addr()), newAddr,
+		"the lighthouse should still be handing out the old address before the rebind")
+
+	myControl.RebindUDPServer()
+	require.True(t, waitForLighthouseMsg(t, r, lhControl, time.Second*5), "expected an update after the rebind")
+	r.RouteFor(time.Millisecond * 400)
+
+	assert.Contains(t, reportedAddrs(t, lhControl, myVpnIpNet[0].Addr()), newAddr,
+		"after the rebind the lighthouse should hand peers our new address")
+
+	lhControl.Stop()
+	myControl.Stop()
+}
@@ -0,0 +1,136 @@
+//go:build e2e_testing
+// +build e2e_testing
+
+package e2e
+
+import (
+	"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/udp"
+)
+
+// TestRecoveryTiming measures how long a tunnel takes to come back after the peer stops accepting our traffic,
+// which is what a laptop waking on a new network looks like from the peer's side: its NAT has no state for where
+// we are now, so everything we send disappears.
+//
+// It is a measurement, not a pass/fail assertion. Recovery is timed to the moment the peer punches back at us,
+// since that is when its NAT opens and the tunnel is usable again.
+//
+//	go test -tags e2e_testing -v -run TestRecoveryTiming ./e2e/
+func TestRecoveryTiming(t *testing.T) {
+	for _, tc := range []struct {
+		name   string
+		rebind bool
+	}{
+		{"no trigger", false},
+		{"rebind counter", true},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			d, lost := measureRecovery(t, tc.rebind)
+			t.Logf("RESULT %-16s recovered in %-9v (%d packets lost)", tc.name, d.Round(time.Millisecond), lost)
+		})
+	}
+}
+
+// measureRecovery returns how long until the peer punched back, and how many of our packets died meanwhile. When
+// rebind is true we call RebindUDPServer once the tunnel goes dark, which is what the darwin network change
+// monitor does and what iOS has always done. When false, nothing tells nebula anything is wrong.
+func measureRecovery(t *testing.T, rebind bool) (time.Duration, int) {
+	t.Helper()
+	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", m{
+		"lighthouse": m{"am_lighthouse": true},
+	})
+
+	peerCfg := m{
+		"lighthouse": m{
+			"hosts":    []any{lhVpnIpNet[0].Addr().String()},
+			"interval": 600,
+			"local_allow_list": m{
+				"10.0.0.0/24": true,
+				"::/0":        false,
+			},
+		},
+		"static_host_map": m{
+			lhVpnIpNet[0].Addr().String(): []any{lhUdpAddr.String()},
+		},
+	}
+
+	myControl, myVpnIpNet, myUdpAddr, _ := newSimpleServer(cert.Version2, ca, caKey, "me", "10.128.0.2/24", peerCfg)
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version2, ca, caKey, "them", "10.128.0.3/24", peerCfg)
+
+	r := router.NewR(t, lhControl, myControl, theirControl)
+	defer r.RenderFlow()
+	defer func() {
+		lhControl.Stop()
+		myControl.Stop()
+		theirControl.Stop()
+	}()
+
+	lhControl.Start()
+	myControl.Start()
+	theirControl.Start()
+	r.RouteFor(time.Millisecond * 500)
+
+	myControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
+	theirControl.InjectLightHouseAddr(myVpnIpNet[0].Addr(), myUdpAddr)
+
+	myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("establish")))
+	r.RouteFor(time.Second)
+	if myControl.GetHostInfoByVpnAddr(theirVpnIpNet[0].Addr(), false) == nil {
+		t.Fatal("failed to establish the tunnel we are measuring")
+	}
+	r.RouteFor(time.Millisecond * 500)
+
+	// From here the peer's NAT has no state for us, everything we send it disappears
+	start := time.Now()
+	blackholed := 0
+	var recovered time.Duration
+
+	if rebind {
+		myControl.RebindUDPServer()
+	}
+
+	// Keep the tun busy the way someone retrying a stalled connection would
+	stop := make(chan struct{})
+	defer close(stop)
+	go func() {
+		tick := time.NewTicker(time.Millisecond * 200)
+		defer tick.Stop()
+		for {
+			select {
+			case <-stop:
+				return
+			case <-tick.C:
+				myControl.InjectTunPacket(BuildTunUDPPacket(
+					theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("retry")))
+			}
+		}
+	}()
+
+	r.RouteForAllExitFuncOrTimeout(time.Second*30, func(p *udp.Packet, c *nebula.Control) router.ExitType {
+		if c == theirControl && p.From == myControl.GetUDPAddr() {
+			blackholed++
+			return router.Drop
+		}
+
+		// The peer reaching us directly is the moment its NAT opened, whether that is a punch or a handshake
+		if c == myControl && p.From == theirUdpAddr {
+			recovered = time.Since(start)
+			return router.RouteAndExit
+		}
+
+		return router.KeepRouting
+	})
+
+	if recovered == 0 {
+		t.Fatalf("no recovery within 30s (%d packets blackholed)", blackholed)
+	}
+	return recovered, blackholed
+}
@@ -13,6 +13,7 @@ import (
 	"regexp"
 	"sort"
 	"sync"
+	"sync/atomic"
 	"testing"
 	"time"

@@ -24,6 +25,19 @@ import (
 	"golang.org/x/exp/maps"
 )

+// outNatKey is the (from, to) pair used by outNat. Comparable struct, so it works as a map key without the
+// allocation cost of a string-concat key.
+type outNatKey struct {
+	from, to netip.AddrPort
+}
+
+// fannedPacket pairs a UDP TX packet with its source control so the router can route it after popping from
+// the fan-in channel.
+type fannedPacket struct {
+	from *nebula.Control
+	pkt  *udp.Packet
+}
+
 type R struct {
 	// Simple map of the ip:port registered on a control to the control
 	// Basically a router, right?
@@ -34,12 +48,28 @@ type R struct {

 	// A last used map, if an inbound packet hit the inNat map then
 	// all return packets should use the same last used inbound address for the outbound sender
-	// map[from address + ":" + to address] => ip:port to rewrite in the udp packet to receiver
-	outNat map[string]netip.AddrPort
+	outNat map[outNatKey]netip.AddrPort

 	// A map of vpn ip to the nebula control it belongs to
 	vpnControls map[netip.Addr]*nebula.Control

+	// Cached select infrastructure for RouteForAllUntilTxTun.
+	// The controls map is immutable after NewR so the cases are good for the test lifetime.
+	// We only rebuild if a different receiver is asked.
+	selRecvCtl *nebula.Control
+	selCases   []reflect.SelectCase
+	selCtls    []*nebula.Control
+
+	// Optional fan-in mode for hot-path benchmarks: one forwarder goroutine per control drains UDP TX into udpFanIn,
+	// so RouteForAllUntilTxTun can do a fixed 2-way native select instead of paying reflect.Select per call.
+	// Off by default (would otherwise interleave with tests that use GetFromUDP directly on the same control).
+	// Enabled by EnableFanIn.
+	udpFanIn  chan fannedPacket
+	stopFanIn chan struct{}
+	fanInWG   sync.WaitGroup
+	fanInMu   sync.Mutex
+	fanInOn   atomic.Bool
+
 	ignoreFlows []ignoreFlow
 	flow        []flowEntry

@@ -84,6 +114,28 @@ type packet struct {
 	packet *udp.Packet
 	tun    bool // a packet pulled off a tun device
 	rx     bool // the packet was received by a udp device
+
+	// h is the nebula header, parsed once when the packet is recorded. parseErr says why there isn't one, which
+	// the flow log reports rather than hiding. Punchy sends a single byte, so an unparseable packet is normal.
+	h        header.H
+	parseErr error
+}
+
+// fromAddr and toAddr are the addresses this packet actually travelled between. Reading them off the control
+// instead would misreport the whole history once a test moves a node. Tun packets are synthesized without
+// addresses, so they fall back to the control.
+func (p *packet) fromAddr() netip.AddrPort {
+	if p.tun || !p.packet.From.IsValid() {
+		return p.from.GetUDPAddr()
+	}
+	return p.packet.From
+}
+
+func (p *packet) toAddr() netip.AddrPort {
+	if p.tun || !p.packet.To.IsValid() {
+		return p.to.GetUDPAddr()
+	}
+	return p.packet.To
 }

 func (p *packet) WasReceived() {
@@ -101,6 +153,9 @@ const (
 	ExitNow ExitType = 1
 	// RouteAndExit routes this packet and exits immediately afterwards
 	RouteAndExit ExitType = 2
+	// Drop discards this packet without delivering it and keeps routing. Use it to simulate a blackhole, such as
+	// a restrictive NAT refusing traffic from an address it has not seen.
+	Drop ExitType = 3
 )

 type ExitFunc func(packet *udp.Packet, receiver *nebula.Control) ExitType
@@ -111,7 +166,9 @@ type ExitFunc func(packet *udp.Packet, receiver *nebula.Control) ExitType
 func NewR(t testing.TB, controls ...*nebula.Control) *R {
 	ctx, cancel := context.WithCancel(context.Background())

-	if err := os.MkdirAll("mermaid", 0755); err != nil {
+	// t.Name() contains a slash for subtests, so the flow log can land in a nested directory
+	fn := filepath.Join("mermaid", fmt.Sprintf("%s.md", t.Name()))
+	if err := os.MkdirAll(filepath.Dir(fn), 0755); err != nil {
 		panic(err)
 	}

@@ -119,10 +176,10 @@ func NewR(t testing.TB, controls ...*nebula.Control) *R {
 		controls:     make(map[netip.AddrPort]*nebula.Control),
 		vpnControls:  make(map[netip.Addr]*nebula.Control),
 		inNat:        make(map[netip.AddrPort]*nebula.Control),
-		outNat:       make(map[string]netip.AddrPort),
+		outNat:       make(map[outNatKey]netip.AddrPort),
 		flow:         []flowEntry{},
 		ignoreFlows:  []ignoreFlow{},
-		fn:           filepath.Join("mermaid", fmt.Sprintf("%s.md", t.Name())),
+		fn:           fn,
 		t:            t,
 		cancelRender: cancel,
 	}
@@ -153,8 +210,10 @@ func NewR(t testing.TB, controls ...*nebula.Control) *R {
 			case <-ctx.Done():
 				return
 			case <-clockSource.C:
+				r.Lock()
 				r.renderHostmaps("clock tick")
 				r.renderFlow()
+				r.Unlock()
 			}
 		}
 	}()
@@ -180,15 +239,21 @@ func (r *R) AddRoute(ip netip.Addr, port uint16, c *nebula.Control) {
 // RenderFlow renders the packet flow seen up until now and stops further automatic renders from happening.
 func (r *R) RenderFlow() {
 	r.cancelRender()
+	r.Lock()
+	defer r.Unlock()
 	r.renderFlow()
 }

 // CancelFlowLogs stops flow logs from being tracked and destroys any logs already collected
 func (r *R) CancelFlowLogs() {
 	r.cancelRender()
+	r.Lock()
 	r.flow = nil
+	r.Unlock()
 }

+// renderFlow writes the flow log to disk. Caller must hold r.Lock. renderFlow reads r.flow / r.additionalGraphs and
+// the *packet pointers stashed inside, all of which are mutated under the same lock by routing paths.
 func (r *R) renderFlow() {
 	if r.flow == nil {
 		return
@@ -211,7 +276,7 @@ func (r *R) renderFlow() {
 			continue
 		}

-		addr := e.packet.from.GetUDPAddr()
+		addr := e.packet.fromAddr()
 		if _, ok := participants[addr]; ok {
 			continue
 		}
@@ -230,7 +295,6 @@ func (r *R) renderFlow() {
 	}

 	// Print packets
-	h := &header.H{}
 	for _, e := range r.flow {
 		if e.packet == nil {
 			//fmt.Fprintf(f, "    note over %s: %s\n", strings.Join(participantsVals, ", "), e.note)
@@ -242,21 +306,22 @@ func (r *R) renderFlow() {
 			fmt.Fprintln(f, r.formatUdpPacket(p))

 		} else {
-			if err := h.Parse(p.packet.Data); err != nil {
-				panic(err)
-			}
-
 			line := "--x"
 			if p.rx {
 				line = "->>"
 			}

-			fmt.Fprintf(f,
-				"    %s%s%s: %s(%s), index %v, counter: %v\n",
-				normalizeName(p.from.GetUDPAddr().String()),
+			detail := fmt.Sprintf("%s(%s), index %v, counter: %v",
+				p.h.TypeName(), p.h.SubTypeName(), p.h.RemoteIndex, p.h.MessageCounter)
+			if p.parseErr != nil {
+				detail = fmt.Sprintf("unparsed, %v (%d bytes)", p.parseErr, len(p.packet.Data))
+			}
+
+			fmt.Fprintf(f, "    %s%s%s: %s\n",
+				normalizeName(p.fromAddr().String()),
 				line,
-				normalizeName(p.to.GetUDPAddr().String()),
-				h.TypeName(), h.SubTypeName(), h.RemoteIndex, h.MessageCounter,
+				normalizeName(p.toAddr().String()),
+				detail,
 			)
 		}
 	}
@@ -370,29 +435,34 @@ func (r *R) unlockedInjectFlow(from, to *nebula.Control, p *udp.Packet, tun bool

 	r.renderHostmaps(fmt.Sprintf("Packet %v", len(r.flow)))

-	if len(r.ignoreFlows) > 0 {
-		var h header.H
-		err := h.Parse(p.Data)
-		if err != nil {
-			panic(err)
-		}
+	var h header.H
+	var parseErr error
+	if !tun {
+		parseErr = h.Parse(p.Data)
+	}

-		for _, i := range r.ignoreFlows {
-			if !tun {
-				if i.messageType == h.Type && i.subType == h.Subtype {
-					return nil
-				}
-			} else if i.tun.HasValue && i.tun.IsTrue {
+	// Decide before copying, the copy comes from a freelist and an ignored packet would never be released
+	for _, i := range r.ignoreFlows {
+		if tun {
+			if i.tun.HasValue && i.tun.IsTrue {
 				return nil
 			}
+			continue
+		}
+
+		// A packet we could not parse has no type to match against, so no rule can ignore it
+		if parseErr == nil && i.messageType == h.Type && i.subType == h.Subtype {
+			return nil
 		}
 	}

 	fp := &packet{
-		from:   from,
-		to:     to,
-		packet: p.Copy(),
-		tun:    tun,
+		from:     from,
+		to:       to,
+		packet:   p.Copy(),
+		tun:      tun,
+		h:        h,
+		parseErr: parseErr,
 	}

 	r.flow = append(r.flow, flowEntry{packet: fp})
@@ -434,68 +504,157 @@ func (r *R) RouteUntilTxTun(sender *nebula.Control, receiver *nebula.Control) []
 				panic("No control for udp tx " + a.String())
 			}
 			fp := r.unlockedInjectFlow(sender, c, p, false)
-			c.InjectUDPPacket(p)
+			c.InjectUDPPacket(p) // copies internally; original is ours to release
 			fp.WasReceived()
 			r.Unlock()
+			p.Release()
 		}
 	}
 }

-// RouteForAllUntilTxTun will route for everyone and return when a packet is seen on receivers tun
-// If the router doesn't have the nebula controller for that address, we panic
+// RouteForAllUntilTxTun will route for everyone and return when a packet is seen on the receiver's tun.
+// If a control's UDP TX address can't be matched to a registered control, we panic.
+//
+// For allocation-sensitive callers (hot-path benchmarks, in particular relay
+// benches with 3+ controls), call EnableFanIn() first.
 func (r *R) RouteForAllUntilTxTun(receiver *nebula.Control) []byte {
+	if r.fanInOn.Load() {
+		return r.routeFanIn(receiver)
+	}
+	return r.routeReflect(receiver)
+}
+
+// routeFanIn is the alloc-free path used when EnableFanIn is in effect.
+func (r *R) routeFanIn(receiver *nebula.Control) []byte {
+	tunTx := receiver.GetTunTxChan()
+	for {
+		select {
+		case p := <-tunTx:
+			r.Lock()
+			if r.flow != nil {
+				np := udp.Packet{Data: make([]byte, len(p))}
+				copy(np.Data, p)
+				r.unlockedInjectFlow(receiver, receiver, &np, true)
+			}
+			r.Unlock()
+			return p
+		case fp := <-r.udpFanIn:
+			r.routeUDP(fp.from, fp.pkt)
+		}
+	}
+}
+
+// routeReflect is the default reflect.Select-based path. Pays the boxing allocation per call but doesn't interfere
+// with tests that pull packets directly from controls' UDP TX channels via GetFromUDP.
+func (r *R) routeReflect(receiver *nebula.Control) []byte {
+	sc, cm := r.selectCasesFor(receiver)
+	for {
+		x, rx, _ := reflect.Select(sc)
+		if x == 0 {
+			p := rx.Interface().([]byte)
+			r.Lock()
+			if r.flow != nil {
+				np := udp.Packet{Data: make([]byte, len(p))}
+				copy(np.Data, p)
+				r.unlockedInjectFlow(cm[x], cm[x], &np, true)
+			}
+			r.Unlock()
+			return p
+		}
+		r.routeUDP(cm[x], rx.Interface().(*udp.Packet))
+	}
+}
+
+// EnableFanIn switches RouteForAllUntilTxTun to the alloc-free fan-in path.
+// One forwarder goroutine per registered control drains UDP TX into a shared channel that RouteForAllUntilTxTun selects
+// on alongside the receiver's TUN TX channel.
+func (r *R) EnableFanIn() {
+	r.fanInMu.Lock()
+	defer r.fanInMu.Unlock()
+	if r.fanInOn.Load() {
+		return
+	}
+	r.udpFanIn = make(chan fannedPacket, 32)
+	r.stopFanIn = make(chan struct{})
+	for _, c := range r.controls {
+		r.startFanInWorker(c)
+	}
+	r.fanInOn.Store(true)
+	r.t.Cleanup(r.stopFanInWorkers)
+}
+
+// startFanInWorker spawns a goroutine that drains c's UDP TX into r.udpFanIn.
+func (r *R) startFanInWorker(c *nebula.Control) {
+	r.fanInWG.Add(1)
+	udpTx := c.GetUDPTxChan()
+	go func() {
+		defer r.fanInWG.Done()
+		for {
+			select {
+			case <-r.stopFanIn:
+				return
+			case p := <-udpTx:
+				select {
+				case <-r.stopFanIn:
+					p.Release()
+					return
+				case r.udpFanIn <- fannedPacket{from: c, pkt: p}:
+				}
+			}
+		}
+	}()
+}
+
+// stopFanInWorkers signals the fan-in goroutines to exit and waits for them.
+func (r *R) stopFanInWorkers() {
+	r.fanInMu.Lock()
+	wasOn := r.fanInOn.Swap(false)
+	r.fanInMu.Unlock()
+	if !wasOn {
+		return
+	}
+	close(r.stopFanIn)
+	r.fanInWG.Wait()
+}
+
+// routeUDP forwards a UDP TX packet from the named source control to the destination control derived from p.To,
+// releasing the source packet after InjectUDPPacket has copied its bytes into a fresh pool slot.
+func (r *R) routeUDP(from *nebula.Control, p *udp.Packet) {
+	r.Lock()
+	defer r.Unlock()
+	a := from.GetUDPAddr()
+	c := r.getControl(a, p.To, p)
+	if c == nil {
+		panic(fmt.Sprintf("No control for udp tx %s", p.To))
+	}
+	fp := r.unlockedInjectFlow(from, c, p, false)
+	c.InjectUDPPacket(p) // copies internally; original is ours to release
+	fp.WasReceived()
+	p.Release()
+}
+
+// selectCasesFor returns the SelectCase array used by routeReflect: one slot for the receiver's TUN TX channel followed
+// by one per control's UDP TX channel. Cached for the test lifetime, only rebuilt if the receiver changes.
+func (r *R) selectCasesFor(receiver *nebula.Control) ([]reflect.SelectCase, []*nebula.Control) {
+	r.Lock()
+	defer r.Unlock()
+	if r.selRecvCtl == receiver && r.selCases != nil {
+		return r.selCases, r.selCtls
+	}
 	sc := make([]reflect.SelectCase, len(r.controls)+1)
 	cm := make([]*nebula.Control, len(r.controls)+1)
-
-	i := 0
-	sc[i] = reflect.SelectCase{
-		Dir:  reflect.SelectRecv,
-		Chan: reflect.ValueOf(receiver.GetTunTxChan()),
-		Send: reflect.Value{},
-	}
-	cm[i] = receiver
-
-	i++
+	sc[0] = reflect.SelectCase{Dir: reflect.SelectRecv, Chan: reflect.ValueOf(receiver.GetTunTxChan())}
+	cm[0] = receiver
+	i := 1
 	for _, c := range r.controls {
-		sc[i] = reflect.SelectCase{
-			Dir:  reflect.SelectRecv,
-			Chan: reflect.ValueOf(c.GetUDPTxChan()),
-			Send: reflect.Value{},
-		}
-
+		sc[i] = reflect.SelectCase{Dir: reflect.SelectRecv, Chan: reflect.ValueOf(c.GetUDPTxChan())}
 		cm[i] = c
 		i++
 	}
-
-	for {
-		x, rx, _ := reflect.Select(sc)
-		r.Lock()
-
-		if x == 0 {
-			// we are the tun tx, we can exit
-			p := rx.Interface().([]byte)
-			np := udp.Packet{Data: make([]byte, len(p))}
-			copy(np.Data, p)
-
-			r.unlockedInjectFlow(cm[x], cm[x], &np, true)
-			r.Unlock()
-			return p
-
-		} else {
-			// we are a udp tx, route and continue
-			p := rx.Interface().(*udp.Packet)
-			a := cm[x].GetUDPAddr()
-			c := r.getControl(a, p.To, p)
-			if c == nil {
-				r.Unlock()
-				panic(fmt.Sprintf("No control for udp tx %s", p.To))
-			}
-			fp := r.unlockedInjectFlow(cm[x], c, p, false)
-			c.InjectUDPPacket(p)
-			fp.WasReceived()
-		}
-		r.Unlock()
-	}
+	r.selRecvCtl = receiver
+	r.selCases = sc
+	r.selCtls = cm
+	return sc, cm
 }

 // RouteExitFunc will call the whatDo func with each udp packet from sender.
@@ -522,6 +681,7 @@ func (r *R) RouteExitFunc(sender *nebula.Control, whatDo ExitFunc) {
 		switch e {
 		case ExitNow:
 			r.Unlock()
+			p.Release()
 			return

 		case RouteAndExit:
@@ -529,8 +689,13 @@ func (r *R) RouteExitFunc(sender *nebula.Control, whatDo ExitFunc) {
 			receiver.InjectUDPPacket(p)
 			fp.WasReceived()
 			r.Unlock()
+			p.Release()
 			return

+		case Drop:
+			// Record it so the flow log shows the attempt, but never hand it to the receiver
+			r.unlockedInjectFlow(sender, receiver, p, false)
+
 		case KeepRouting:
 			fp := r.unlockedInjectFlow(sender, receiver, p, false)
 			receiver.InjectUDPPacket(p)
@@ -541,6 +706,7 @@ func (r *R) RouteExitFunc(sender *nebula.Control, whatDo ExitFunc) {
 		}

 		r.Unlock()
+		p.Release()
 	}
 }

@@ -560,6 +726,85 @@ func (r *R) RouteUntilAfterMsgType(sender *nebula.Control, msgType header.Messag
 	})
 }

+// RouteFor routes everything that shows up for the given duration and then returns. Use it to let a test settle
+// deterministically rather than sleeping and hoping: a single FlushAll races a completing handshake, which queues
+// more packets right behind it.
+func (r *R) RouteFor(d time.Duration) {
+	r.RouteForAllExitFuncOrTimeout(d, func(*udp.Packet, *nebula.Control) ExitType {
+		return KeepRouting
+	})
+}
+
+// RouteForAllExitFuncOrTimeout is RouteForAllExitFunc with a deadline, reporting whether whatDo asked to exit
+// before time ran out. The unbounded version blocks forever on a quiet network, so this is what a test needs to
+// assert that something does NOT happen, or to route for a fixed settling period.
+func (r *R) RouteForAllExitFuncOrTimeout(timeout time.Duration, whatDo ExitFunc) bool {
+	sc := make([]reflect.SelectCase, 0, len(r.controls)+1)
+	cm := make([]*nebula.Control, 0, len(r.controls))
+
+	for _, c := range r.controls {
+		sc = append(sc, reflect.SelectCase{
+			Dir:  reflect.SelectRecv,
+			Chan: reflect.ValueOf(c.GetUDPTxChan()),
+			Send: reflect.Value{},
+		})
+		cm = append(cm, c)
+	}
+
+	timer := time.NewTimer(timeout)
+	defer timer.Stop()
+	sc = append(sc, reflect.SelectCase{
+		Dir:  reflect.SelectRecv,
+		Chan: reflect.ValueOf(timer.C),
+		Send: reflect.Value{},
+	})
+
+	for {
+		x, rx, _ := reflect.Select(sc)
+		if x == len(cm) {
+			return false
+		}
+
+		r.Lock()
+		p := rx.Interface().(*udp.Packet)
+		receiver := r.getControl(cm[x].GetUDPAddr(), p.To, p)
+		if receiver == nil {
+			r.Unlock()
+			panic("Can't RouteForAllExitFuncOrTimeout for host: " + p.To.String())
+		}
+
+		e := whatDo(p, receiver)
+		switch e {
+		case ExitNow:
+			r.Unlock()
+			p.Release()
+			return true
+
+		case RouteAndExit:
+			fp := r.unlockedInjectFlow(cm[x], receiver, p, false)
+			receiver.InjectUDPPacket(p)
+			fp.WasReceived()
+			r.Unlock()
+			p.Release()
+			return true
+
+		case Drop:
+			// Record it so the flow log shows the attempt, but never hand it to the receiver
+			r.unlockedInjectFlow(cm[x], receiver, p, false)
+
+		case KeepRouting:
+			fp := r.unlockedInjectFlow(cm[x], receiver, p, false)
+			receiver.InjectUDPPacket(p)
+			fp.WasReceived()
+
+		default:
+			panic(fmt.Sprintf("Unknown exitFunc return: %v", e))
+		}
+		r.Unlock()
+		p.Release()
+	}
+}
+
 func (r *R) RouteForAllUntilAfterMsgTypeTo(receiver *nebula.Control, msgType header.MessageType, subType header.MessageSubType) {
 	h := &header.H{}
 	r.RouteForAllExitFunc(func(p *udp.Packet, r *nebula.Control) ExitType {
@@ -641,6 +886,7 @@ func (r *R) RouteForAllExitFunc(whatDo ExitFunc) {
 		switch e {
 		case ExitNow:
 			r.Unlock()
+			p.Release()
 			return

 		case RouteAndExit:
@@ -648,8 +894,13 @@ func (r *R) RouteForAllExitFunc(whatDo ExitFunc) {
 			receiver.InjectUDPPacket(p)
 			fp.WasReceived()
 			r.Unlock()
+			p.Release()
 			return

+		case Drop:
+			// Record it so the flow log shows the attempt, but never hand it to the receiver
+			r.unlockedInjectFlow(cm[x], receiver, p, false)
+
 		case KeepRouting:
 			fp := r.unlockedInjectFlow(cm[x], receiver, p, false)
 			receiver.InjectUDPPacket(p)
@@ -659,6 +910,7 @@ func (r *R) RouteForAllExitFunc(whatDo ExitFunc) {
 			panic(fmt.Sprintf("Unknown exitFunc return: %v", e))
 		}
 		r.Unlock()
+		p.Release()
 	}
 }

@@ -702,19 +954,20 @@ func (r *R) FlushAll() {
 		}
 		receiver.InjectUDPPacket(p)
 		r.Unlock()
+		p.Release()
 	}
 }

 // getControl performs or seeds NAT translation and returns the control for toAddr, p from fields may change
 // This is an internal router function, the caller must hold the lock
 func (r *R) getControl(fromAddr, toAddr netip.AddrPort, p *udp.Packet) *nebula.Control {
-	if newAddr, ok := r.outNat[fromAddr.String()+":"+toAddr.String()]; ok {
+	if newAddr, ok := r.outNat[outNatKey{from: fromAddr, to: toAddr}]; ok {
 		p.From = newAddr
 	}

 	c, ok := r.inNat[toAddr]
 	if ok {
-		r.outNat[c.GetUDPAddr().String()+":"+fromAddr.String()] = toAddr
+		r.outNat[outNatKey{from: c.GetUDPAddr(), to: fromAddr}] = toAddr
 		return c
 	}

@@ -0,0 +1,125 @@
+//go:build e2e_testing
+// +build e2e_testing
+
+package e2e
+
+import (
+	"crypto/ed25519"
+	"crypto/rand"
+	"encoding/pem"
+	"net"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/slackhq/nebula/cert"
+	"github.com/slackhq/nebula/cert_test"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"golang.org/x/crypto/ssh"
+)
+
+func TestSSHDLifecycle(t *testing.T) {
+	// TestSSHDLifecycle exercises the in-process sshd through several config reloads and a Control.Stop.
+	ca, _, caKey, _ := cert_test.NewTestCaCert(
+		cert.Version1, cert.Curve_CURVE25519,
+		time.Now(), time.Now().Add(10*time.Minute),
+		nil, nil, []string{},
+	)
+
+	hostKeyPEM := generateSSHHostKey(t)
+	clientSigner, clientAuthKey := generateSSHClientKey(t)
+	sshdAddr := allocLoopbackPort(t)
+
+	overrides := m{
+		"sshd": m{
+			"enabled":  true,
+			"listen":   sshdAddr,
+			"host_key": hostKeyPEM,
+			"authorized_users": []m{{
+				"user": "tester",
+				"keys": []string{clientAuthKey},
+			}},
+		},
+	}
+	control, _, _, _ := newSimpleServer(cert.Version1, ca, caKey, "sshd-test", "10.222.0.1/24", overrides)
+	control.Start()
+	t.Cleanup(func() { control.Stop() })
+
+	// sshd binds in a goroutine after Start returns; wait for it.
+	require.Eventually(t, func() bool { return canDial(sshdAddr) }, 2*time.Second, 25*time.Millisecond,
+		"sshd never started listening")
+
+	for i := 1; i <= 3; i++ {
+		out := sshExecReload(t, sshdAddr, clientSigner)
+		assert.Contains(t, out, "Reloading config", "reload cycle %d", i)
+		require.Eventually(t, func() bool { return canDial(sshdAddr) }, 2*time.Second, 25*time.Millisecond,
+			"sshd not listening after reload cycle %d", i)
+	}
+
+	control.Stop()
+	require.Eventually(t, func() bool { return !canDial(sshdAddr) }, 2*time.Second, 25*time.Millisecond,
+		"sshd still listening after Control.Stop")
+}
+
+func canDial(addr string) bool {
+	c, err := net.DialTimeout("tcp", addr, 100*time.Millisecond)
+	if err != nil {
+		return false
+	}
+	_ = c.Close()
+	return true
+}
+
+// allocLoopbackPort grabs an unused TCP port on 127.0.0.1, closes it, and returns the address. There
+// is a small race between releasing the port and the sshd reclaiming it; in practice the OS keeps the
+// port available long enough for the test to bind it.
+func allocLoopbackPort(t *testing.T) string {
+	t.Helper()
+	l, err := net.Listen("tcp", "127.0.0.1:0")
+	require.NoError(t, err)
+	addr := l.Addr().String()
+	require.NoError(t, l.Close())
+	return addr
+}
+
+func generateSSHHostKey(t *testing.T) string {
+	t.Helper()
+	_, priv, err := ed25519.GenerateKey(rand.Reader)
+	require.NoError(t, err)
+	block, err := ssh.MarshalPrivateKey(priv, "nebula-e2e-host")
+	require.NoError(t, err)
+	return string(pem.EncodeToMemory(block))
+}
+
+func generateSSHClientKey(t *testing.T) (ssh.Signer, string) {
+	t.Helper()
+	_, priv, err := ed25519.GenerateKey(rand.Reader)
+	require.NoError(t, err)
+	signer, err := ssh.NewSignerFromKey(priv)
+	require.NoError(t, err)
+	auth := strings.TrimSpace(string(ssh.MarshalAuthorizedKey(signer.PublicKey())))
+	return signer, auth
+}
+
+func sshExecReload(t *testing.T, addr string, signer ssh.Signer) string {
+	t.Helper()
+	cfg := &ssh.ClientConfig{
+		User:            "tester",
+		Auth:            []ssh.AuthMethod{ssh.PublicKeys(signer)},
+		HostKeyCallback: ssh.InsecureIgnoreHostKey(),
+		Timeout:         2 * time.Second,
+	}
+	client, err := ssh.Dial("tcp", addr, cfg)
+	require.NoError(t, err)
+	defer client.Close()
+
+	sess, err := client.NewSession()
+	require.NoError(t, err)
+	defer sess.Close()
+
+	// reload tears the channel down before sending exit-status, so Output returns an error on the
+	// channel close. The output buffer still has whatever the reload callback wrote before that.
+	out, _ := sess.Output("reload")
+	return string(out)
+}
@@ -4,15 +4,23 @@
 package e2e

 import (
+	"fmt"
+	"net/netip"
 	"testing"
 	"time"

 	"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"
+	"github.com/stretchr/testify/require"
+	"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{})
@@ -35,8 +43,8 @@ func TestDropInactiveTunnels(t *testing.T) {
 	r.Log("Go inactive and wait for the tunnels to get dropped")
 	waitStart := time.Now()
 	for {
-		myIndexes := len(myControl.GetHostmap().Indexes)
-		theirIndexes := len(theirControl.GetHostmap().Indexes)
+		myIndexes := myControl.GetHostmapIndexCount()
+		theirIndexes := theirControl.GetHostmapIndexCount()
 		if myIndexes == 0 && theirIndexes == 0 {
 			break
 		}
@@ -55,3 +63,511 @@ func TestDropInactiveTunnels(t *testing.T) {
 	myControl.Stop()
 	theirControl.Stop()
 }
+
+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{})
+	caB, err := ca.MarshalPEM()
+	if err != nil {
+		panic(err)
+	}
+	ca2, _, caKey2, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+
+	ca2B, err := ca2.MarshalPEM()
+	if err != nil {
+		panic(err)
+	}
+	caStr := fmt.Sprintf("%s\n%s", caB, ca2B)
+
+	myCert, _, myPrivKey, _ := cert_test.NewTestCert(cert.Version1, cert.Curve_CURVE25519, ca, caKey, "me", time.Now(), time.Now().Add(5*time.Minute), []netip.Prefix{netip.MustParsePrefix("10.128.0.1/24")}, nil, []string{})
+	_, myCert2Pem := cert_test.NewTestCertDifferentVersion(myCert, cert.Version2, ca2, caKey2)
+
+	theirCert, _, theirPrivKey, _ := cert_test.NewTestCert(cert.Version1, cert.Curve_CURVE25519, ca, caKey, "them", time.Now(), time.Now().Add(5*time.Minute), []netip.Prefix{netip.MustParsePrefix("10.128.0.2/24")}, nil, []string{})
+	theirCert2, _ := cert_test.NewTestCertDifferentVersion(theirCert, cert.Version2, ca2, caKey2)
+
+	myControl, myVpnIpNet, myUdpAddr, myC := newServer([]cert.Certificate{ca, ca2}, []cert.Certificate{myCert}, myPrivKey, m{})
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newServer([]cert.Certificate{ca, ca2}, []cert.Certificate{theirCert, theirCert2}, theirPrivKey, m{})
+
+	// 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)
+	defer r.RenderFlow()
+
+	r.Log("Assert the tunnel between me and them works")
+	assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
+	r.Log("yay")
+	//todo ???
+	time.Sleep(1 * time.Second)
+	r.FlushAll()
+
+	mc := m{
+		"pki": m{
+			"ca":   caStr,
+			"cert": string(myCert2Pem),
+			"key":  string(myPrivKey),
+		},
+		//"tun": m{"disabled": true},
+		"firewall": myC.Settings["firewall"],
+		//"handshakes": m{
+		//	"try_interval": "1s",
+		//},
+		"listen":  myC.Settings["listen"],
+		"logging": myC.Settings["logging"],
+		"timers":  myC.Settings["timers"],
+	}
+
+	cb, err := yaml.Marshal(mc)
+	if err != nil {
+		panic(err)
+	}
+
+	r.Logf("reload new v2-only config")
+	err = myC.ReloadConfigString(string(cb))
+	assert.NoError(t, err)
+	r.Log("yay, spin until their sees it")
+	waitStart := time.Now()
+	for {
+		assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
+		c := theirControl.GetHostInfoByVpnAddr(myVpnIpNet[0].Addr(), false)
+		if c == nil {
+			r.Log("nil")
+		} else {
+			version := c.Cert.Version()
+			r.Logf("version %d", version)
+			if version == cert.Version2 {
+				break
+			}
+		}
+		since := time.Since(waitStart)
+		if since > time.Second*10 {
+			t.Fatal("Cert should be new by now")
+		}
+		time.Sleep(time.Second)
+	}
+
+	r.RenderHostmaps("Final hostmaps", myControl, theirControl)
+
+	myControl.Stop()
+	theirControl.Stop()
+}
+
+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{})
+	caB, err := ca.MarshalPEM()
+	if err != nil {
+		panic(err)
+	}
+	ca2, _, caKey2, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+
+	ca2B, err := ca2.MarshalPEM()
+	if err != nil {
+		panic(err)
+	}
+	caStr := fmt.Sprintf("%s\n%s", caB, ca2B)
+
+	myCert, _, myPrivKey, myCertPem := cert_test.NewTestCert(cert.Version1, cert.Curve_CURVE25519, ca, caKey, "me", time.Now(), time.Now().Add(5*time.Minute), []netip.Prefix{netip.MustParsePrefix("10.128.0.1/24")}, nil, []string{})
+	myCert2, _ := cert_test.NewTestCertDifferentVersion(myCert, cert.Version2, ca2, caKey2)
+
+	theirCert, _, theirPrivKey, _ := cert_test.NewTestCert(cert.Version1, cert.Curve_CURVE25519, ca, caKey, "them", time.Now(), time.Now().Add(5*time.Minute), []netip.Prefix{netip.MustParsePrefix("10.128.0.2/24")}, nil, []string{})
+	theirCert2, _ := cert_test.NewTestCertDifferentVersion(theirCert, cert.Version2, ca2, caKey2)
+
+	myControl, myVpnIpNet, myUdpAddr, myC := newServer([]cert.Certificate{ca, ca2}, []cert.Certificate{myCert2}, myPrivKey, m{})
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newServer([]cert.Certificate{ca, ca2}, []cert.Certificate{theirCert, theirCert2}, theirPrivKey, m{})
+
+	// 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)
+	defer r.RenderFlow()
+
+	r.Log("Assert the tunnel between me and them works")
+	//assertTunnel(t, theirVpnIpNet[0].Addr(), myVpnIpNet[0].Addr(), theirControl, myControl, r)
+	//r.Log("yay")
+	assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
+	r.Log("yay")
+	//todo ???
+	time.Sleep(1 * time.Second)
+	r.FlushAll()
+
+	mc := m{
+		"pki": m{
+			"ca":   caStr,
+			"cert": string(myCertPem),
+			"key":  string(myPrivKey),
+		},
+		"firewall": myC.Settings["firewall"],
+		"listen":   myC.Settings["listen"],
+		"logging":  myC.Settings["logging"],
+		"timers":   myC.Settings["timers"],
+	}
+
+	cb, err := yaml.Marshal(mc)
+	if err != nil {
+		panic(err)
+	}
+
+	r.Logf("reload new v1-only config")
+	err = myC.ReloadConfigString(string(cb))
+	assert.NoError(t, err)
+	r.Log("yay, spin until their sees it")
+	waitStart := time.Now()
+	for {
+		assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
+		c := theirControl.GetHostInfoByVpnAddr(myVpnIpNet[0].Addr(), false)
+		c2 := myControl.GetHostInfoByVpnAddr(theirVpnIpNet[0].Addr(), false)
+		if c == nil || c2 == nil {
+			r.Log("nil")
+		} else {
+			version := c.Cert.Version()
+			theirVersion := c2.Cert.Version()
+			r.Logf("version %d,%d", version, theirVersion)
+			if version == cert.Version1 {
+				break
+			}
+		}
+		since := time.Since(waitStart)
+		if since > time.Second*5 {
+			r.Log("it is unusual that the cert is not new yet, but not a failure yet")
+		}
+		if since > time.Second*10 {
+			r.Log("wtf")
+			t.Fatal("Cert should be new by now")
+		}
+		time.Sleep(time.Second)
+	}
+
+	r.RenderHostmaps("Final hostmaps", myControl, theirControl)
+
+	myControl.Stop()
+	theirControl.Stop()
+}
+
+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{})
+	ca2, _, caKey2, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+
+	myCert, _, myPrivKey, _ := cert_test.NewTestCert(cert.Version1, cert.Curve_CURVE25519, ca, caKey, "me", time.Now(), time.Now().Add(5*time.Minute), []netip.Prefix{netip.MustParsePrefix("10.128.0.1/24")}, nil, []string{})
+	myCert2, _ := cert_test.NewTestCertDifferentVersion(myCert, cert.Version2, ca2, caKey2)
+
+	theirCert, _, theirPrivKey, _ := cert_test.NewTestCert(cert.Version1, cert.Curve_CURVE25519, ca, caKey, "them", time.Now(), time.Now().Add(5*time.Minute), []netip.Prefix{netip.MustParsePrefix("10.128.0.2/24")}, nil, []string{})
+	theirCert2, _ := cert_test.NewTestCertDifferentVersion(theirCert, cert.Version2, ca2, caKey2)
+
+	myControl, myVpnIpNet, myUdpAddr, _ := newServer([]cert.Certificate{ca, ca2}, []cert.Certificate{myCert2}, myPrivKey, m{})
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newServer([]cert.Certificate{ca, ca2}, []cert.Certificate{theirCert, theirCert2}, theirPrivKey, m{})
+
+	// 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)
+	defer r.RenderFlow()
+
+	r.Log("Assert the tunnel between me and them works")
+	//assertTunnel(t, theirVpnIpNet[0].Addr(), myVpnIpNet[0].Addr(), theirControl, myControl, r)
+	//r.Log("yay")
+	assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
+	r.Log("yay")
+	//todo ???
+	time.Sleep(1 * time.Second)
+	r.FlushAll()
+
+	waitStart := time.Now()
+	for {
+		assertTunnel(t, myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), myControl, theirControl, r)
+		c := theirControl.GetHostInfoByVpnAddr(myVpnIpNet[0].Addr(), false)
+		c2 := myControl.GetHostInfoByVpnAddr(theirVpnIpNet[0].Addr(), false)
+		if c == nil || c2 == nil {
+			r.Log("nil")
+		} else {
+			version := c.Cert.Version()
+			theirVersion := c2.Cert.Version()
+			r.Logf("version %d,%d", version, theirVersion)
+			if version == theirVersion {
+				break
+			}
+		}
+		since := time.Since(waitStart)
+		if since > time.Second*5 {
+			r.Log("wtf")
+		}
+		if since > time.Second*10 {
+			r.Log("wtf")
+			t.Fatal("Cert should be new by now")
+		}
+		time.Sleep(time.Second)
+	}
+
+	r.RenderHostmaps("Final hostmaps", myControl, theirControl)
+
+	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.InjectTunPacket(BuildTunUDPPacket(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.InjectTunPacket(BuildTunUDPPacket(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()
+}
+
+// TestRelayReplayProtection asserts that a relay (forwarding-type) node rejects
+// replayed relay frames. A captured relay frame, re-injected with the same
+// message counter, must be dropped by the replay window rather than re-forwarded
+// to the relay target. Before the fix, handleOutsideRelayPacket authenticated the
+// frame but never advanced the replay window, so every replay was re-forwarded.
+func TestRelayReplayProtection(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}})
+
+	myVpnV6 := myVpnIpNet[1]
+	relayVpnV4 := relayVpnIpNet[0]
+	relayVpnV6 := relayVpnIpNet[1]
+	theirVpnV6 := theirVpnIpNet[0]
+
+	// Teach me how to reach the relay and that them is reachable 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)
+
+	r := router.NewR(t, myControl, relayControl, theirControl)
+	defer r.RenderFlow()
+
+	myControl.Start()
+	relayControl.Start()
+	theirControl.Start()
+
+	// Establish the relayed tunnel in both directions so all handshakes complete.
+	t.Log("Establish the relayed tunnel")
+	myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnV6.Addr(), 80, myVpnV6.Addr(), 80, []byte("Hi from me")))
+	p := r.RouteForAllUntilTxTun(theirControl)
+	assertUdpPacket(t, []byte("Hi from me"), p, myVpnV6.Addr(), theirVpnV6.Addr(), 80, 80)
+	theirControl.InjectTunPacket(BuildTunUDPPacket(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)
+
+	// Drain anything still queued on me's UDP tx so the next packet we pull is the
+	// relay frame we are about to generate.
+	for myControl.GetFromUDP(false) != nil {
+	}
+
+	// Capture a single legitimate relay frame that me transmits toward the relay.
+	t.Log("Capture a relay frame from me -> relay")
+	myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnV6.Addr(), 80, myVpnV6.Addr(), 80, []byte("replay me")))
+	relayFrame := myControl.GetFromUDP(true)
+	require.Equal(t, relayUdpAddr, relayFrame.To, "captured frame should be addressed to the relay")
+	var fh header.H
+	require.NoError(t, fh.Parse(relayFrame.Data))
+	require.Equal(t, header.Message, fh.Type)
+	require.Equal(t, header.MessageRelay, fh.Subtype)
+
+	// drainForwards counts relay frames the relay forwards toward them within the
+	// settle window. We match on destination + (Message, MessageRelay) so the
+	// relay's own direct traffic to them can't be miscounted.
+	drainForwards := func(settle time.Duration) int {
+		ch := relayControl.GetUDPTxChan()
+		count := 0
+		for {
+			select {
+			case pkt := <-ch:
+				var ph header.H
+				if pkt.To == theirUdpAddr && ph.Parse(pkt.Data) == nil &&
+					ph.Type == header.Message && ph.Subtype == header.MessageRelay {
+					count++
+				}
+				pkt.Release()
+			case <-time.After(settle):
+				return count
+			}
+		}
+	}
+
+	// First delivery of the captured frame: the relay should forward it once.
+	t.Log("Deliver the captured frame once; relay forwards it to them")
+	relayControl.InjectUDPPacket(relayFrame)
+	require.Equal(t, 1, drainForwards(200*time.Millisecond), "relay should forward the first, legitimate copy")
+
+	// Replay the exact same frame several times. A correct replay window rejects
+	// these duplicates so the relay forwards none of them.
+	t.Log("Replay the captured frame; relay must drop the duplicates")
+	const replays = 3
+	for i := 0; i < replays; i++ {
+		relayControl.InjectUDPPacket(relayFrame)
+	}
+	forwarded := drainForwards(200 * time.Millisecond)
+	assert.Equal(t, 0, forwarded, "relay re-forwarded %d/%d replayed relay frames; replay protection is ineffective on relay tunnels", forwarded, replays)
+
+	r.RenderHostmaps("Final hostmaps", myControl, relayControl, theirControl)
+}
+
+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 := myControl.GetHostmapIndexCount()
+		theirIndexes := theirControl.GetHostmapIndexCount()
+		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 := myControl.GetHostmapIndexCount()
+		theirIndexes := theirControl.GetHostmapIndexCount()
+		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()
+}
@@ -138,12 +138,32 @@ listen:
  # max, net.core.rmem_max and net.core.wmem_max
  #read_buffer: 10485760
  #write_buffer: 10485760
+
+  # On Windows only
+  # When true, Nebula installs a WFP (Windows Filtering Platform) PERMIT filter scoped to UDP at the listener port.
+  # WFP sits below Windows Defender Firewall, so this lets peer handshakes reach Nebula's outside socket regardless
+  # of WDF's inbound rules.
+  # Default true; set to false to leave WDF in charge of inbound decisions on the listener port. Not reloadable.
+  #windows_bypass_wdf: true
+
+  # On macOS only
+  # macOS scopes the udp socket to the interface it was created on, so moving between networks (wifi to wired,
+  # office to home) leaves Nebula sending out an interface that no longer has a route. When true, Nebula watches
+  # the routing socket and rebinds the listener once the change settles.
+  # iOS does not use this, the host app drives the same rebind itself.
+  # Default true. Not reloadable.
+  #rebind_on_network_change: true
+
  # By default, Nebula replies to packets it has no tunnel for with a "recv_error" packet. This packet helps speed up reconnection
  # in the case that Nebula on either side did not shut down cleanly. This response can be abused as a way to discover if Nebula is running
  # on a host though. This option lets you configure if you want to send "recv_error" packets always, never, or only to private network remotes.
  # 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.
@@ -159,17 +179,21 @@ listen:

 punchy:
  # Continues to punch inbound/outbound at a regular interval to avoid expiration of firewall nat mappings
+  # This setting is reloadable.
  punch: true

  # respond means that a node you are trying to reach will connect back out to you if your hole punching fails
  # this is extremely useful if one node is behind a difficult nat, such as a symmetric NAT
  # Default is false
+  # This setting is reloadable.
  #respond: true

  # delays a punch response for misbehaving NATs, default is 1 second.
+  # This setting is reloadable.
  #delay: 1s

  # set the delay before attempting punchy.respond. Default is 5 seconds. respond must be true to take effect.
+  # This setting is reloadable.
  #respond_delay: 5s

 # Cipher allows you to choose between the available ciphers for your network. Options are chachapoly or aes
@@ -200,6 +224,12 @@ 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:
@@ -272,6 +302,24 @@ tun:
    #  metric: 100
    #  install: true

+  # On Windows only, sets the network category of the nebula interface. Without this, Windows often
+  # leaves the network as "Unidentified" and treats it as Public, which makes the host firewall more
+  # restrictive than you usually want for an overlay between trusted peers. Valid values:
+  #   private             - treat the nebula network as a private/trusted network (default)
+  #   public              - treat it as a public/untrusted network
+  #   domain              - treat it as a domain-authenticated network
+  #   unset               - leave whatever Windows decided alone
+  # Not reloadable.
+  #network_category: private
+
+  # On Windows only
+  # When true, Nebula installs a WFP (Windows Filtering Platform) PERMIT filter scoped to the nebula adapter LUID.
+  # WFP sits below Windows Defender Firewall, so this lets inbound traffic through regardless of WDF rules.
+  # Filters are auto-removed when the adapter goes away.
+  # See listen.windows_bypass_wdf for the matching control over inbound to nebula's outside UDP listener.
+  # Default true; set to false to leave WDF in charge of inbound decisions on the nebula interface. Not reloadable.
+  #windows_bypass_wdf: true
+
  # On linux only, set to true to manage unsafe routes directly on the system route table with gateway routes instead of
  # in nebula configuration files. Default false, not reloadable.
  #use_system_route_table: false
@@ -282,24 +330,21 @@ tun:

 # Configure logging level
 logging:
-  # panic, fatal, error, warning, info, or debug. Default is info and is reloadable.
-  #NOTE: Debug mode can log remotely controlled/untrusted data which can quickly fill a disk in some
-  # scenarios. Debug logging is also CPU intensive and will decrease performance overall.
-  # Only enable debug logging while actively investigating an issue.
+  # trace, debug, info, warn, or error. Default is info and is reloadable.
+  # fatal and panic are accepted for backwards compatibility and map to error.
+  #NOTE: Debug and trace modes can log remotely controlled/untrusted data which can quickly fill a disk in some
+  # 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:
-  #     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"
+  # Timestamps use RFC3339Nano ("2006-01-02T15:04:05.999999999Z07:00") and are not configurable.

+# 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 +362,12 @@ 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
@@ -358,7 +405,7 @@ firewall:
  #   `drop` (default): silently drop the packet.
  #   `reject`: send a reject reply.
  #     - For TCP, this will be a RST "Connection Reset" packet.
-  #     - For other protocols, this will be an ICMP port unreachable packet.
+  #     - For other protocols, this will be an ICMP "Destination unreachable: Communication administratively prohibited" packet.
  outbound_action: drop
  inbound_action: drop

@@ -378,13 +425,14 @@ 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.
-  #   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.
+  #   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. `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
@@ -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.Out = os.Stdout
+	logger := logging.NewLogger(os.Stdout)

 	ctrl, err := nebula.Main(&cfg, false, "custom-app", logger, overlay.NewUserDeviceFromConfig)
 	if err != nil {
@@ -8,6 +8,15 @@ Before=sshd.service
 Type=notify
 NotifyAccess=main
 SyslogIdentifier=nebula
+
+# Uncomment to run as an unprivileged user with only CAP_NET_ADMIN. Requires a
+# nebula user that owns the config directory. Add CAP_NET_BIND_SERVICE to both
+# lines if any listener (lighthouse DNS, listen.port, stats, sshd) binds <1024.
+#User=nebula
+#Group=nebula
+#CapabilityBoundingSet=CAP_NET_ADMIN
+#AmbientCapabilities=CAP_NET_ADMIN
+
 ExecReload=/bin/kill -HUP $MAINPID
 ExecStart=/usr/local/bin/nebula -config /etc/nebula/config.yml
 Restart=always
@@ -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 {
@@ -42,8 +44,8 @@ type Firewall struct {
 	InRules  *FirewallTable
 	OutRules *FirewallTable

-	InSendReject  bool
-	OutSendReject bool
+	InboundSendReject  bool
+	OutboundSendReject bool

 	//TODO: we should have many more options for TCP, an option for ICMP, and mimic the kernel a bit better
 	// https://www.kernel.org/doc/Documentation/networking/nf_conntrack-sysctl.txt
@@ -56,8 +58,9 @@ type Firewall struct {
 	routableNetworks *bart.Lite

 	// assignedNetworks is a list of vpn networks assigned to us in the certificate.
-	assignedNetworks  []netip.Prefix
-	hasUnsafeNetworks bool
+	assignedNetworks []netip.Prefix
+	// unsafeNetworks is the list of unsafe networks issued to us in the certificate
+	unsafeNetworks []netip.Prefix

 	rules        string
 	rulesVersion uint16
@@ -66,7 +69,7 @@ type Firewall struct {
 	incomingMetrics     firewallMetrics
 	outgoingMetrics     firewallMetrics

-	l *logrus.Logger
+	l *slog.Logger
 }

 type firewallMetrics struct {
@@ -130,7 +133,7 @@ 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

@@ -156,10 +159,9 @@ func NewFirewall(l *logrus.Logger, tcpTimeout, UDPTimeout, defaultTimeout time.D
 		assignedNetworks = append(assignedNetworks, network)
 	}

-	hasUnsafeNetworks := false
-	for _, n := range c.UnsafeNetworks() {
+	unsafeNetworks := c.UnsafeNetworks()
+	for _, n := range unsafeNetworks {
 		routableNetworks.Insert(n)
-		hasUnsafeNetworks = true
 	}

 	return &Firewall{
@@ -167,15 +169,15 @@ func NewFirewall(l *logrus.Logger, tcpTimeout, UDPTimeout, defaultTimeout time.D
 			Conns:      make(map[firewall.Packet]*conn),
 			TimerWheel: NewTimerWheel[firewall.Packet](tmin, tmax),
 		},
-		InRules:           newFirewallTable(),
-		OutRules:          newFirewallTable(),
-		TCPTimeout:        tcpTimeout,
-		UDPTimeout:        UDPTimeout,
-		DefaultTimeout:    defaultTimeout,
-		routableNetworks:  routableNetworks,
-		assignedNetworks:  assignedNetworks,
-		hasUnsafeNetworks: hasUnsafeNetworks,
-		l:                 l,
+		InRules:          newFirewallTable(),
+		OutRules:         newFirewallTable(),
+		TCPTimeout:       tcpTimeout,
+		UDPTimeout:       UDPTimeout,
+		DefaultTimeout:   defaultTimeout,
+		routableNetworks: routableNetworks,
+		assignedNetworks: assignedNetworks,
+		unsafeNetworks:   unsafeNetworks,
+		l:                l,

 		incomingMetrics: firewallMetrics{
 			droppedLocalAddr:  metrics.GetOrRegisterCounter("firewall.incoming.dropped.local_addr", nil),
@@ -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)
@@ -214,23 +216,23 @@ func NewFirewallFromConfig(l *logrus.Logger, cs *CertState, c *config.C) (*Firew
 	inboundAction := c.GetString("firewall.inbound_action", "drop")
 	switch inboundAction {
 	case "reject":
-		fw.InSendReject = true
+		fw.InboundSendReject = true
 	case "drop":
-		fw.InSendReject = false
+		fw.InboundSendReject = false
 	default:
-		l.WithField("action", inboundAction).Warn("invalid firewall.inbound_action, defaulting to `drop`")
-		fw.InSendReject = false
+		l.Warn("invalid firewall.inbound_action, defaulting to `drop`", "action", inboundAction)
+		fw.InboundSendReject = false
 	}

 	outboundAction := c.GetString("firewall.outbound_action", "drop")
 	switch outboundAction {
 	case "reject":
-		fw.OutSendReject = true
+		fw.OutboundSendReject = true
 	case "drop":
-		fw.OutSendReject = false
+		fw.OutboundSendReject = false
 	default:
-		l.WithField("action", inboundAction).Warn("invalid firewall.outbound_action, defaulting to `drop`")
-		fw.OutSendReject = false
+		l.Warn("invalid firewall.outbound_action, defaulting to `drop`", "action", outboundAction)
+		fw.OutboundSendReject = false
 	}

 	err := AddFirewallRulesFromConfig(l, false, c, fw)
@@ -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 {
-	// 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")
-
+func (f *Firewall) AddRule(incoming bool, proto uint8, startPort int32, endPort int32, groups []string, host string, cidr, localCidr, caName string, caSha string) error {
 	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
-		switch r.Proto {
-		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 r.Cidr != "" && r.Cidr != "any" {
+			_, 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 != "" {
-			localCidr, err = netip.ParsePrefix(r.LocalCidr)
+		if r.LocalCidr != "" && r.LocalCidr != "any" {
+			_, 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 +414,40 @@ 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 ErrInvalidLocalIP = errors.New("local IP is not in list of handled local IPs")
+var ErrUnknownNetworkType = errors.New("unknown network type")
+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 {
-	// 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
-	if h.networks != nil {
-		if !h.networks.Contains(fp.RemoteAddr) {
-			f.metrics(incoming).droppedRemoteAddr.Inc(1)
-			return ErrInvalidRemoteIP
-		}
-	} else {
+	// Make sure remote address matches nebula certificate, and determine how to treat it
+	if h.networks == nil {
 		// 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
@@ -449,6 +456,11 @@ func (f *Firewall) Drop(fp firewall.Packet, incoming bool, h *HostInfo, caPool *
 		return ErrInvalidLocalIP
 	}

+	// Check if we spoke to this tuple, if we did then allow this packet
+	if f.inConns(fp, h, caPool, localCache) {
+		return nil
+	}
+
 	table := f.OutRules
 	if incoming {
 		table = f.InRules
@@ -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
@@ -522,26 +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 {
-				h.logger(f.l).
-					WithField("fwPacket", fp).
-					WithField("incoming", c.incoming).
-					WithField("rulesVersion", f.rulesVersion).
-					WithField("oldRulesVersion", c.rulesVersion).
-					Debugln("dropping old conntrack entry, does not match new ruleset")
+			if f.l.Enabled(context.Background(), slog.LevelDebug) {
+				h.logger(f.l).Debug("dropping old conntrack entry, does not match new ruleset",
+					"fwPacket", fp,
+					"incoming", c.incoming,
+					"rulesVersion", f.rulesVersion,
+					"oldRulesVersion", c.rulesVersion,
+				)
 			}
 			delete(conntrack.Conns, fp)
 			conntrack.Unlock()
 			return false
 		}

-		if f.l.Level >= logrus.DebugLevel {
-			h.logger(f.l).
-				WithField("fwPacket", fp).
-				WithField("incoming", c.incoming).
-				WithField("rulesVersion", f.rulesVersion).
-				WithField("oldRulesVersion", c.rulesVersion).
-				Debugln("keeping old conntrack entry, does match new ruleset")
+		if f.l.Enabled(context.Background(), slog.LevelDebug) {
+			h.logger(f.l).Debug("keeping old conntrack entry, does match new ruleset",
+				"fwPacket", fp,
+				"incoming", c.incoming,
+				"rulesVersion", f.rulesVersion,
+				"oldRulesVersion", c.rulesVersion,
+			)
 		}

 		c.rulesVersion = f.rulesVersion
@@ -640,7 +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")
 	}
@@ -653,7 +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
 		}
 	}
@@ -667,6 +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 {
@@ -684,7 +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),
@@ -698,14 +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
 		}
@@ -715,7 +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
 		}
@@ -747,24 +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
 		}
@@ -780,44 +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() {
-		nlc, _ := fr.CIDR.Get(ip)
-		if nlc == nil {
-			nlc = flc()
-		}
-		err := nlc.addRule(f, localCIDR)
+	if cidr != "" {
+		c, err := netip.ParsePrefix(cidr)
 		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 {
-	if len(groups) == 0 && host == "" && !ip.IsValid() {
+func (fr *FirewallRule) isAny(groups []string, host string, cidr string) bool {
+	if len(groups) == 0 && host == "" && cidr == "" {
 		return true
 	}

-	for _, group := range groups {
-		if group == "any" {
-			return true
-		}
+	if slices.Contains(groups, "any") {
+		return true
 	}

 	if host == "any" {
 		return true
 	}

-	if ip.IsValid() && ip.Bits() == 0 {
+	if cidr == "any" {
 		return true
 	}

@@ -869,9 +890,14 @@ func (fr *FirewallRule) match(p firewall.Packet, c *cert.CachedCertificate) bool
 	return false
 }

-func (flc *firewallLocalCIDR) addRule(f *Firewall, localIp netip.Prefix) error {
-	if !localIp.IsValid() {
-		if !f.hasUnsafeNetworks || f.defaultLocalCIDRAny {
+func (flc *firewallLocalCIDR) addRule(f *Firewall, localCidr string) error {
+	if localCidr == "any" {
+		flc.Any = true
+		return nil
+	}
+
+	if localCidr == "" {
+		if len(f.unsafeNetworks) == 0 || f.defaultLocalCIDRAny {
 			flc.Any = true
 			return nil
 		}
@@ -881,12 +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
 }

@@ -907,7 +934,6 @@ type rule struct {
 	Code      string
 	Proto     string
 	Host      string
-	Group     string
 	Groups    []string
 	Cidr      string
 	LocalCidr string
@@ -915,7 +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)
@@ -946,10 +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() {
@@ -966,52 +996,122 @@ 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) {
-	if s == "any" {
-		startPort = firewall.PortAny
-		endPort = firewall.PortAny
+// 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 == ""

-	} 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], " ")
-		sPorts[1] = strings.Trim(sPorts[1], " ")
-
-		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)
-		}
-
-		rStartPort, err := strconv.Atoi(sPorts[0])
-		if err != nil {
-			return 0, 0, 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])
-		}
-
-		startPort = int32(rStartPort)
-		endPort = int32(rEndPort)
-
-		if startPort == firewall.PortAny {
-			endPort = firewall.PortAny
-		}
-
-	} else {
-		rPort, err := strconv.Atoi(s)
-		if err != nil {
-			return 0, 0, fmt.Errorf("was not a number; `%s`", s)
-		}
-		startPort = int32(rPort)
-		endPort = startPort
+	if (groupsEmpty && hostEmpty && cidrEmpty) == true {
+		return nil //no content!
 	}

-	return
+	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) {
+	const notAPort int32 = -2
+	if s == "any" {
+		return firewall.PortAny, firewall.PortAny, nil
+	}
+	if s == "fragment" {
+		return firewall.PortFragment, firewall.PortFragment, nil
+	}
+	if !strings.Contains(s, `-`) {
+		rPort, err := parsePortValue("", s)
+		if err != nil {
+			return notAPort, notAPort, err
+		}
+		return rPort, rPort, nil
+	}
+
+	sPorts := strings.SplitN(s, `-`, 2)
+	for i := range sPorts {
+		sPorts[i] = strings.Trim(sPorts[i], " ")
+	}
+	if len(sPorts) != 2 || sPorts[0] == "" || sPorts[1] == "" {
+		return notAPort, notAPort, fmt.Errorf("appears to be a range but could not be parsed; `%s`", s)
+	}
+
+	startPort, err := parsePortValue("beginning range ", sPorts[0])
+	if err != nil {
+		return notAPort, notAPort, err
+	}
+
+	endPort, err := parsePortValue("ending range ", sPorts[1])
+	if err != nil {
+		return notAPort, notAPort, err
+	}
+
+	if startPort == firewall.PortAny {
+		endPort = firewall.PortAny
+	}
+
+	return startPort, endPort, nil
+}
+
+// parsePortValue accepts a base-10 decimal in [0, 65535] and returns it
+// widened to int32. Using strconv.ParseUint with bitSize 16 rejects
+// negative input, out-of-range input (>65535), and any non-decimal byte
+// by construction, so the int32 widening that follows is provably safe
+// and cannot collide with firewall.PortAny (0) or firewall.PortFragment
+// (-1) via integer truncation.
+//
+// prefix is prepended to both error messages so callers can disambiguate
+// the single-port path (prefix="") from the range bounds (prefix="beginning
+// range " / "ending range "), preserving the historical error strings.
+func parsePortValue(prefix, s string) (int32, error) {
+	n, err := strconv.ParseUint(s, 10, 16)
+	if err == nil {
+		return int32(n), nil
+	}
+	if errors.Is(err, strconv.ErrRange) {
+		return 0, fmt.Errorf("%sout of range [0,65535]; `%s`", prefix, s)
+	}
+	return 0, fmt.Errorf("%swas not a number; `%s`", prefix, s)
 }
@@ -1,10 +1,10 @@
 package firewall

 import (
+	"context"
+	"log/slog"
 	"sync/atomic"
 	"time"
-
-	"github.com/sirupsen/logrus"
 )

 // ConntrackCache is used as a local routine cache to know if a given flow
@@ -15,41 +15,49 @@ type ConntrackCacheTicker struct {
 	cacheV    uint64
 	cacheTick atomic.Uint64

+	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{
+		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)
-		c.cacheTick.Add(1)
+		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 := len(c.cache); ll > 0 {
-			if l.Level == logrus.DebugLevel {
-				l.WithField("len", ll).Debug("resetting conntrack cache")
+			if c.l.Enabled(context.Background(), slog.LevelDebug) {
+				c.l.Debug("resetting conntrack cache", "len", ll)
 			}
 			c.cache = make(ConntrackCache, ll)
 		}
@@ -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())
+}
@@ -22,7 +22,10 @@ const (
 type Packet struct {
 	LocalAddr  netip.Addr
 	RemoteAddr netip.Addr
-	LocalPort  uint16
+	// 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:
@@ -1,37 +1,39 @@
 module github.com/slackhq/nebula

-go 1.25
+go 1.26.0

 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/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 v0.28.0
 	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/pkcs11 v1.1.2-0.20231115102856-9078ad6b9d4b
+	github.com/kardianos/service v1.3.0
+	github.com/miekg/dns v1.1.72
+	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 v3.0.5
+	golang.org/x/crypto v0.54.0
 	golang.org/x/exp v0.0.0-20230725093048-515e97ebf090
-	golang.org/x/net v0.45.0
-	golang.org/x/sync v0.17.0
-	golang.org/x/sys v0.37.0
-	golang.org/x/term v0.36.0
+	golang.org/x/net v0.57.0
+	golang.org/x/sync v0.22.0
+	golang.org/x/sys v0.47.0
+	golang.org/x/term v0.45.0
 	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
-	google.golang.org/protobuf v1.36.8
+	golang.zx2c4.com/wireguard/windows v1.0.1
+	google.golang.org/protobuf v1.36.11
 	gopkg.in/yaml.v3 v3.0.1
 	gvisor.dev/gvisor v0.0.0-20240423190808-9d7a357edefe
 )
@@ -48,7 +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 v0.36.0 // indirect
 	golang.org/x/time v0.5.0 // indirect
-	golang.org/x/tools v0.33.0 // indirect
+	golang.org/x/tools v0.45.0 // indirect
 )
@@ -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=
@@ -24,8 +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/gaissmai/bart v0.25.0 h1:eqiokVPqM3F94vJ0bTHXHtH91S8zkKL+bKh+BsGOsJM=
-github.com/gaissmai/bart v0.25.0/go.mod h1:GREWQfTLRWz/c5FTOsIw+KkscuFkIV5t8Rp7Nd1Td5c=
+github.com/gaissmai/bart v0.28.0 h1:89yZLo8NmyqD0RYgJ3QO9HhqqGGw+oWhf90cZm69Lko=
+github.com/gaissmai/bart v0.28.0/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=
@@ -64,8 +66,8 @@ github.com/json-iterator/go v1.1.10/go.mod h1:KdQUCv79m/52Kvf8AW2vK1V8akMuk1QjK/
 github.com/json-iterator/go v1.1.11/go.mod h1:KdQUCv79m/52Kvf8AW2vK1V8akMuk1QjK/uOdHXbAo4=
 github.com/julienschmidt/httprouter v1.2.0/go.mod h1:SYymIcj16QtmaHHD7aYtjjsJG7VTCxuUUipMqKk8s4w=
 github.com/julienschmidt/httprouter v1.3.0/go.mod h1:JR6WtHb+2LUe8TCKY3cZOxFyyO8IZAc4RVcycCCAKdM=
-github.com/kardianos/service v1.2.4 h1:XNlGtZOYNx2u91urOdg/Kfmc+gfmuIo1Dd3rEi2OgBk=
-github.com/kardianos/service v1.2.4/go.mod h1:E4V9ufUuY82F7Ztlu1eN9VXWIQxg8NoLQlmFe0MtrXc=
+github.com/kardianos/service v1.3.0 h1:/LGy+xPP2TM+GLTiCZ2di7cy0Jd/qrawlTUfqKYFdTI=
+github.com/kardianos/service v1.3.0/go.mod h1:E4V9ufUuY82F7Ztlu1eN9VXWIQxg8NoLQlmFe0MtrXc=
 github.com/kisielk/errcheck v1.5.0/go.mod h1:pFxgyoBC7bSaBwPgfKdkLd5X25qrDl4LWUI2bnpBCr8=
 github.com/kisielk/gotool v1.0.0/go.mod h1:XhKaO+MFFWcvkIS/tQcRk01m1F5IRFswLeQ+oQHNcck=
 github.com/klauspost/compress v1.18.0 h1:c/Cqfb0r+Yi+JtIEq73FWXVkRonBlf0CRNYc8Zttxdo=
@@ -83,10 +85,10 @@ github.com/kr/text v0.1.0/go.mod h1:4Jbv+DJW3UT/LiOwJeYQe1efqtUx/iVham/4vfdArNI=
 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 v1.1.68/go.mod h1:fujopn7TB3Pu3JM69XaawiU0wqjpL9/8xGop5UrTPps=
-github.com/miekg/pkcs11 v1.1.2-0.20231115102856-9078ad6b9d4b h1:J/AzCvg5z0Hn1rqZUJjpbzALUmkKX0Zwbc/i4fw7Sfk=
-github.com/miekg/pkcs11 v1.1.2-0.20231115102856-9078ad6b9d4b/go.mod h1:XsNlhZGX73bx86s2hdc/FuaLm2CPZJemRLMA+WTFxgs=
+github.com/miekg/dns v1.1.72 h1:vhmr+TF2A3tuoGNkLDFK9zi36F2LS+hKTRW0Uf8kbzI=
+github.com/miekg/dns v1.1.72/go.mod h1:+EuEPhdHOsfk6Wk5TT2CzssZdqkmFhf8r+aVyDEToIs=
+github.com/miekg/pkcs11 v1.1.2 h1:/VxmeAX5qU6Q3EwafypogwWbYryHFmF2RpkJmw3m4MQ=
+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=
@@ -131,8 +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=
@@ -155,21 +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 v3.0.5 h1:N6y/pJk8buWs9NY5ERU2HSMfm+IuD/OtfdAnq6kESPw=
+go.yaml.in/yaml/v3 v3.0.5/go.mod h1:HVTZu1O7/Vkt2N+BFy8Zza+lnLsABggaTM2ZpNIGuKg=
 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 v0.43.0/go.mod h1:BFbav4mRNlXJL4wNeejLpWxB7wMbc79PdRGhWKncxR0=
+golang.org/x/crypto v0.54.0 h1:YLIA59K4fiNzHzjnZt2tUJQjQtUWfWbeHBqKtk3eScw=
+golang.org/x/crypto v0.54.0/go.mod h1:KWL8ny2AZdGR2cWmzeHrp2azQPGogOv+HeQaVEXC2dk=
 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 v0.24.0/go.mod h1:IXM97Txy2VM4PJ3gI61r1YEk/gAj6zAHN3AdZt6S9Ww=
+golang.org/x/mod v0.36.0 h1:JJjpVx6myfUsUdAzZuOSTTmRE0PfZeNWzzvKrP7amb4=
+golang.org/x/mod v0.36.0/go.mod h1:moc6ELqsWcOw5Ef3xVprK5ul/MvtVvkIXLziUOICjUQ=
 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=
@@ -180,8 +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 v0.45.0/go.mod h1:ECOoLqd5U3Lhyeyo/QDCEVQ4sNgYsqvCZ722XogGieY=
+golang.org/x/net v0.57.0 h1:K5+3DljvIuDG9/Jv9rvyMywYNFCQ9RSUY6OOTTkT+tE=
+golang.org/x/net v0.57.0/go.mod h1:KpXc8iv+r3XplLAG/f7Jsf9RPszJzdR0f58q9vGOuEU=
 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=
@@ -189,8 +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 v0.17.0/go.mod h1:9KTHXmSnoGruLpwFjVSX0lNNA75CykiMECbovNTZqGI=
+golang.org/x/sync v0.22.0 h1:SZjpbeLmrCk4xhRSZFNZW5gFUeCeFgjekvI/+gfScek=
+golang.org/x/sync v0.22.0/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=
@@ -204,14 +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 v0.37.0/go.mod h1:OgkHotnGiDImocRcuBABYBEXf8A9a87e/uXjp9XT3ks=
+golang.org/x/sys v0.47.0 h1:o7XGOvZQCADBQQ4Y7VNq2dRWQR7JmOUW8Kxx4ZsNgWs=
+golang.org/x/sys v0.47.0/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 v0.36.0/go.mod h1:Qu394IJq6V6dCBRgwqshf3mPF85AqzYEzofzRdZkWss=
+golang.org/x/term v0.45.0 h1:NwWyBmoJCbfTHpxrWoZ9C6/VxOf7ic219I8xZZFdrf0=
+golang.org/x/term v0.45.0/go.mod h1:9aqxs0blBcrm/n0L9QW0aRVD+ktan8ssZromtqJC43w=
 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=
@@ -222,8 +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 v0.33.0/go.mod h1:CIJMaWEY88juyUfo7UbgPqbC8rU2OqfAV1h2Qp0oMYI=
+golang.org/x/tools v0.45.0 h1:18qN3FAooORvApf5XjCXgsuayZOEtXf6JK18I3+ONa8=
+golang.org/x/tools v0.45.0/go.mod h1:LuUGqqaXcXMEFEruIVJVm5mgDD8vww/z/SR1gQ4uE/0=
 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=
@@ -232,8 +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 v0.5.3/go.mod h1:9TEe8TJmtwyQebdFwAkEWOPr3prrtqm+REGFifP60hI=
+golang.zx2c4.com/wireguard/windows v1.0.1 h1:eOxiDVbywPC+ZQqvdCK7x+ZwWXKbYv50TtH8ysFIbw8=
+golang.zx2c4.com/wireguard/windows v1.0.1/go.mod h1:+fbT3FFdX4zzYDLwJh5+HPEcNN/3HyNdzhNSVsQM+zs=
 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=
@@ -242,8 +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 v1.36.8/go.mod h1:fuxRtAxBytpl4zzqUh6/eyUujkJdNiuEkXntxiD/uRU=
+google.golang.org/protobuf v1.36.11 h1:fV6ZwhNocDyBLK0dj+fg8ektcVegBBuEolpbTQyBNVE=
+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=
@@ -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
@@ -0,0 +1,22 @@
+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")
+	ErrInvalidRemoteIndex      = errors.New("peer sent an invalid index in handshake payload")
+	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")
+)
--- a/Show More
+++ b/Show More