Pass the Android state path to the bundle generator

.github/workflows/release.yml

@@ -475,132 +475,6 @@ jobs:
           path: dist/
           retention-days: 3
 
-  release_ui_gtk3:
-    # Legacy GTK3/WebKit2GTK 4.1 UI build for distros without WebKitGTK 6.0
-    # (Ubuntu 22.04, Debian 12, RHEL 9, Fedora <=39). Runs on ubuntu-22.04 so
-    # the binary links against the oldest supported glibc.
-    runs-on: ubuntu-22.04
-    outputs:
-      release_ui_gtk3_artifact_url: ${{ steps.upload_release_ui_gtk3.outputs.artifact-url }}
-    steps:
-      - name: Checkout
-        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
-        with:
-          fetch-depth: 0 # It is required for GoReleaser to work properly
-          persist-credentials: false
-
-      - name: Parse semver string
-        id: semver_parser
-        uses: netbirdio/shared-actions/actions/parse-semver@be5df6047383da2236e02243cceb857d8567c27e # v0.0.2
-
-      - name: Set snapshot flag
-        if: ${{ !startsWith(github.ref, 'refs/tags/v') }}
-        run: |
-          echo "flags=--snapshot" >> $GITHUB_ENV
-
-      - name: Set build vars
-        if: ${{ startsWith(github.ref, 'refs/tags/v') }}
-        run: |
-          if [[ "x-${{ steps.semver_parser.outputs.prerelease }}" == "x-" && "x-${{ github.repository }}" == "x-netbirdio/netbird" ]]; then
-            echo "x-${{ github.repository }}"
-            echo "x-${{ steps.semver_parser.outputs.prerelease }}"
-            echo "SKIP_PUBLISH=false" >> $GITHUB_ENV
-          else
-            echo "x-${{ github.repository }}"
-            echo "x-${{ steps.semver_parser.outputs.prerelease }}"
-          fi
-
-      - name: Set up Go
-        uses: actions/setup-go@924ae3a1cded613372ab5595356fb5720e22ba16 # v6.5.0
-        with:
-          go-version-file: "go.mod"
-          cache: false
-      - name: Cache Go modules
-        # Restore-only from the release_ui cache written by trusted runs; the
-        # module cache is identical (same go.sum) and stale build-cache
-        # entries just miss.
-        uses: actions/cache/restore@2c8a9bd7457de244a408f35966fab2fb45fda9c8 # v6.0.0
-        with:
-          path: |
-            ~/go/pkg/mod
-            ~/.cache/go-build
-          key: ${{ runner.os }}-ui-go-releaser-${{ hashFiles('**/go.sum') }}
-          restore-keys: |
-            ${{ runner.os }}-ui-go-releaser-
-
-      - name: Install modules
-        run: go mod tidy
-
-      - name: check git status
-        run: git --no-pager diff --exit-code
-
-      - name: Set up Node.js
-        uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4
-        with:
-          node-version: '22'
-
-      - name: Set up pnpm
-        uses: pnpm/action-setup@a3252b78c470c02df07e9d59298aecedc3ccdd6d # v3.0.0
-        with:
-          version: 11
-
-      - name: Install dependencies
-        run: sudo apt update && sudo apt install -y -q libgtk-3-dev libwebkit2gtk-4.1-dev
-
-      - name: Decode GPG signing key
-        if: github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name == github.repository
-        env:
-          GPG_RPM_PRIVATE_KEY: ${{ secrets.GPG_RPM_PRIVATE_KEY }}
-        run: |
-          echo "$GPG_RPM_PRIVATE_KEY" | base64 -d > /tmp/gpg-rpm-signing-key.asc
-          echo "GPG_RPM_KEY_FILE=/tmp/gpg-rpm-signing-key.asc" >> $GITHUB_ENV
-
-      - name: Install wails3 CLI
-        # Version derived from go.mod so the binding generator always matches
-        # the wails runtime the binary links against.
-        # -tags gtk3: the CLI links the wails runtime's cgo packages, and the
-        # default tags request gtk4/webkitgtk-6.0 pkg-config entries that do
-        # not exist on ubuntu-22.04.
-        run: |
-          WAILS_VERSION=$(go list -m -f '{{.Version}}' github.com/wailsapp/wails/v3)
-          go install -tags gtk3 github.com/wailsapp/wails/v3/cmd/wails3@$WAILS_VERSION
-
-      - name: Run GoReleaser
-        uses: goreleaser/goreleaser-action@5daf1e915a5f0af01ddbcd89a43b8061ff4f1a89 # v7.2.2
-        with:
-          version: ${{ env.GORELEASER_VER }}
-          args: release --config .goreleaser_ui_gtk3.yaml --clean ${{ env.flags }}
-        env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-          UPLOAD_DEBIAN_SECRET: ${{ secrets.PKG_UPLOAD_SECRET }}
-          UPLOAD_YUM_SECRET: ${{ secrets.PKG_UPLOAD_SECRET }}
-          GPG_RPM_KEY_FILE: ${{ env.GPG_RPM_KEY_FILE }}
-          NFPM_NETBIRD_UI_RPM_GTK3_PASSPHRASE: ${{ secrets.GPG_RPM_PASSPHRASE }}
-          SKIP_PUBLISH: ${{ env.SKIP_PUBLISH }}
-      - name: Verify RPM signatures
-        run: |
-          docker run --rm -v $(pwd)/dist:/dist fedora:41 bash -c '
-            dnf install -y -q rpm-sign curl >/dev/null 2>&1
-            curl -sSL https://pkgs.netbird.io/yum/repodata/repomd.xml.key -o /tmp/rpm-pub.key
-            rpm --import /tmp/rpm-pub.key
-            echo "=== Verifying RPM signatures ==="
-            for rpm_file in /dist/*.rpm; do
-              [ -f "$rpm_file" ] || continue
-              echo "--- $(basename $rpm_file) ---"
-              rpm -K "$rpm_file"
-            done
-          '
-      - name: Clean up GPG key
-        if: always()
-        run: rm -f /tmp/gpg-rpm-signing-key.asc
-      - name: upload non tags for debug purposes
-        id: upload_release_ui_gtk3
-        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a #v7.0.1
-        with:
-          name: release-ui-gtk3
-          path: dist/
-          retention-days: 3
-
   release_ui_darwin:
     runs-on: macos-latest
     outputs:
@@ -814,7 +688,7 @@ jobs:
   comment_release_artifacts:
     name: Comment release artifacts
     runs-on: ubuntu-latest
-    needs: [release, release_ui, release_ui_gtk3, release_ui_darwin]
+    needs: [release, release_ui, release_ui_darwin]
     if: ${{ always() && github.event_name == 'pull_request' && github.event.pull_request.head.repo.full_name == github.repository }}
     permissions:
       contents: read
@@ -826,14 +700,12 @@ jobs:
         env:
           RELEASE_RESULT: ${{ needs.release.result }}
           RELEASE_UI_RESULT: ${{ needs.release_ui.result }}
-          RELEASE_UI_GTK3_RESULT: ${{ needs.release_ui_gtk3.result }}
           RELEASE_UI_DARWIN_RESULT: ${{ needs.release_ui_darwin.result }}
           RELEASE_ARTIFACT_URL: ${{ needs.release.outputs.release_artifact_url }}
           LINUX_PACKAGES_ARTIFACT_URL: ${{ needs.release.outputs.linux_packages_artifact_url }}
           WINDOWS_PACKAGES_ARTIFACT_URL: ${{ needs.release.outputs.windows_packages_artifact_url }}
           MACOS_PACKAGES_ARTIFACT_URL: ${{ needs.release.outputs.macos_packages_artifact_url }}
           RELEASE_UI_ARTIFACT_URL: ${{ needs.release_ui.outputs.release_ui_artifact_url }}
-          RELEASE_UI_GTK3_ARTIFACT_URL: ${{ needs.release_ui_gtk3.outputs.release_ui_gtk3_artifact_url }}
           RELEASE_UI_DARWIN_ARTIFACT_URL: ${{ needs.release_ui_darwin.outputs.release_ui_darwin_artifact_url }}
           GHCR_IMAGES_MARKDOWN: ${{ needs.release.outputs.ghcr_images }}
         with:
@@ -856,7 +728,6 @@ jobs:
               ['Windows packages', process.env.WINDOWS_PACKAGES_ARTIFACT_URL, process.env.RELEASE_RESULT],
               ['macOS packages', process.env.MACOS_PACKAGES_ARTIFACT_URL, process.env.RELEASE_RESULT],
               ['UI artifacts', process.env.RELEASE_UI_ARTIFACT_URL, process.env.RELEASE_UI_RESULT],
-              ['UI GTK3 artifacts', process.env.RELEASE_UI_GTK3_ARTIFACT_URL, process.env.RELEASE_UI_GTK3_RESULT],
               ['UI macOS artifacts', process.env.RELEASE_UI_DARWIN_ARTIFACT_URL, process.env.RELEASE_UI_DARWIN_RESULT],
             ];
 
@@ -913,7 +784,7 @@ jobs:
 
   trigger_signer:
     runs-on: ubuntu-latest
-    needs: [release, release_ui, release_ui_gtk3, release_ui_darwin, test_windows_installer]
+    needs: [release, release_ui, release_ui_darwin, test_windows_installer]
     if: startsWith(github.ref, 'refs/tags/')
     steps:
       - name: Trigger binaries sign pipelines

.github/workflows/test-infrastructure-files.yml

@@ -257,15 +257,6 @@ jobs:
         with:
           persist-credentials: false
 
-      - name: Verify fresh-install session cookie key hardening
-        run: |
-          grep -Fxq '  SESSION_COOKIE_ENCRYPTION_KEY=$(openssl rand -base64 32)' infrastructure_files/getting-started.sh
-          grep -Fxq '    sessionCookieEncryptionKey: "$SESSION_COOKIE_ENCRYPTION_KEY"' infrastructure_files/getting-started.sh
-          grep -Fxq '  install -m 600 /dev/null config.yaml' infrastructure_files/getting-started.sh
-          grep -Fxq '  openssl rand -base64 32' infrastructure_files/getting-started-enterprise.sh
-          grep -Fxq '  NETBIRD_SESSION_COOKIE_ENCRYPTION_KEY=$(rand_b64_key)' infrastructure_files/getting-started-enterprise.sh
-          grep -Fxq '    sessionCookieEncryptionKey: "${NETBIRD_SESSION_COOKIE_ENCRYPTION_KEY}"' infrastructure_files/getting-started-enterprise.sh
-
       - name: Verify Dex retirement notice
         run: |
           if infrastructure_files/getting-started-with-dex.sh >stdout.txt 2>stderr.txt; then

.goreleaser_ui_gtk3.yaml

@@ -1,131 +0,0 @@
-version: 2
-env:
-  - SKIP_PUBLISH={{ if index .Env "SKIP_PUBLISH" }}{{ .Env.SKIP_PUBLISH }}{{ else }}true{{ end }}
-project_name: netbird-ui
-
-before:
-  hooks:
-    # Bindings are gitignored; regenerate before the frontend build so
-    # the @wailsio/runtime Vite plugin can resolve them (vite refuses to
-    # build without them).
-    # -f '-tags gtk3': the generator type-checks client/ui, whose cgo imports
-    # would otherwise resolve gtk4/webkitgtk-6.0 pkg-config entries that do
-    # not exist on ubuntu-22.04.
-    - sh -c 'cd client/ui && wails3 generate bindings -clean=true -ts -f "-tags gtk3"'
-    - sh -c 'cd client/ui/frontend && pnpm install --frozen-lockfile && pnpm build'
-
-builds:
-  # Legacy GTK3 / WebKit2GTK 4.1 build for distros without WebKitGTK 6.0
-  # (Ubuntu 22.04, Debian 12, RHEL 9, Fedora <=39). The gtk3 tag flips the
-  # Wails Linux backend to the GTK3 stack and swaps our GTK4-only XEmbed
-  # tray host for the pure-Go stub (client/ui/xembed_host_gtk3_linux.go).
-  # Must be built on the oldest supported glibc (ubuntu-22.04 runner).
-  - id: netbird-ui-gtk3
-    dir: client/ui
-    binary: netbird-ui
-    env:
-      - CGO_ENABLED=1
-    goos:
-      - linux
-    goarch:
-      - amd64
-    ldflags:
-      - -s -w -X github.com/netbirdio/netbird/version.version={{.Version}} -X main.commit={{.Commit}} -X main.date={{.CommitDate}} -X main.builtBy=goreleaser
-    mod_timestamp: "{{ .CommitTimestamp }}"
-    tags:
-      - production
-      - gtk3
-
-archives:
-  - id: linux-gtk3-arch
-    name_template: "{{ .ProjectName }}-linux-gtk3_{{ .Version }}_{{ .Os }}_{{ .Arch }}"
-    builds:
-      - netbird-ui-gtk3
-
-nfpms:
-  # Same package_name as the GTK4 packages -- the two are mutually-exclusive
-  # alternatives served from separate repo paths (see uploads below); a given
-  # distro points at exactly one of them. The file names must still differ:
-  # the Debian pool is shared storage keyed by file name, so a default-named
-  # gtk3 .deb would overwrite the stable one.
-  - maintainer: Netbird <dev@netbird.io>
-    description: Netbird client UI.
-    homepage: https://netbird.io/
-    license: BSD-3-Clause
-    vendor: NetBird
-    id: netbird_ui_deb_gtk3
-    package_name: netbird-ui
-    file_name_template: "{{ .PackageName }}-gtk3_{{ .Version }}_{{ .Os }}_{{ .Arch }}"
-    builds:
-      - netbird-ui-gtk3
-    formats:
-      - deb
-    scripts:
-      postinstall: "release_files/ui-post-install.sh"
-    contents:
-      - src: client/ui/build/linux/netbird.desktop
-        dst: /usr/share/applications/org.wails.netbird.desktop
-      - src: client/ui/build/appicon.png
-        dst: /usr/share/pixmaps/netbird.png
-    dependencies:
-      - netbird (>= 0.75.0)
-      - libgtk-3-0
-      - libwebkit2gtk-4.1-0
-
-  - maintainer: Netbird <dev@netbird.io>
-    description: Netbird client UI.
-    homepage: https://netbird.io/
-    license: BSD-3-Clause
-    vendor: NetBird
-    id: netbird_ui_rpm_gtk3
-    package_name: netbird-ui
-    file_name_template: "{{ .PackageName }}-gtk3_{{ .Version }}_{{ .Os }}_{{ .Arch }}"
-    builds:
-      - netbird-ui-gtk3
-    formats:
-      - rpm
-    scripts:
-      postinstall: "release_files/ui-post-install.sh"
-    contents:
-      - src: client/ui/build/linux/netbird.desktop
-        dst: /usr/share/applications/org.wails.netbird.desktop
-      - src: client/ui/build/appicon.png
-        dst: /usr/share/pixmaps/netbird.png
-    dependencies:
-      - netbird >= 0.75.0
-      - (gtk3 or libgtk-3-0)
-      - (webkit2gtk4.1 or libwebkit2gtk-4_1-0)
-
-    rpm:
-      signature:
-        key_file: '{{ if index .Env "GPG_RPM_KEY_FILE" }}{{ .Env.GPG_RPM_KEY_FILE }}{{ end }}'
-
-# The GTK4 UI job shares project_name, so the default checksum file name would
-# collide with it on the shared GitHub release.
-checksum:
-  name_template: "{{ .ProjectName }}_gtk3_checksums.txt"
-
-changelog:
-  disable: true
-
-uploads:
-  # The gtk3 packages reuse the netbird-ui package name, so they live in
-  # dedicated repo paths (deb distribution `gtk3`, yum path `yum-gtk3`) that
-  # legacy distros point their repo config at.
-  - name: debian-gtk3
-    skip: "{{ .Env.SKIP_PUBLISH }}"
-    ids:
-      - netbird_ui_deb_gtk3
-    mode: archive
-    target: https://pkgs.wiretrustee.com/debian/pool/{{ .ArtifactName }};deb.distribution=gtk3;deb.component=main;deb.architecture={{ if .Arm }}armhf{{ else }}{{ .Arch }}{{ end }};deb.package=
-    username: dev@wiretrustee.com
-    method: PUT
-
-  - name: yum-gtk3
-    skip: "{{ .Env.SKIP_PUBLISH }}"
-    ids:
-      - netbird_ui_rpm_gtk3
-    mode: archive
-    target: https://pkgs.wiretrustee.com/yum-gtk3/{{ .Arch }}{{ if .Arm }}{{ .Arm }}{{ end }}
-    username: dev@wiretrustee.com
-    method: PUT

AGENTS.md

@@ -14,22 +14,20 @@ in this file, not duplicated there.
 
 ## Contents
 
-- [STOP and ask the user before](#stop-and-ask-the-user-before)
-- [Quick reference](#quick-reference)
-- [Structure](#structure)
-- [Where to look](#where-to-look)
-- [Security](#security)
-- [Agent conventions](#agent-conventions)
-- [Repo-wide principles](#repo-wide-principles)
-- [Type safety](#type-safety)
-- [Concurrency and lifecycle](#concurrency-and-lifecycle)
-- [Error handling](#error-handling)
-- [Comments](#comments)
-- [Testing](#testing)
-- [Pitfalls](#pitfalls)
-- [Commits, PRs, releases](#commits-prs-releases)
-- [After you push: CI and review bots](#after-you-push-ci-and-review-bots)
-- [Discussion and support](#discussion-and-support)
+- [NetBird Agent Guidelines](#netbird-agent-guidelines)
+  - [Contents](#contents)
+  - [STOP and ask the user before](#stop-and-ask-the-user-before)
+  - [Quick reference](#quick-reference)
+  - [Structure](#structure)
+  - [Where to look](#where-to-look)
+  - [Repo-wide principles](#repo-wide-principles)
+  - [Error handling](#error-handling)
+  - [Comments](#comments)
+  - [Testing](#testing)
+  - [Pitfalls](#pitfalls)
+  - [Commits, PRs, releases](#commits-prs-releases)
+  - [After you push: CI and review bots](#after-you-push-ci-and-review-bots)
+  - [Discussion and support](#discussion-and-support)
 
 ## STOP and ask the user before
 
@@ -159,125 +157,11 @@ netbird/
 | LLM routing / Agent Network | `proxy/internal/llm/`, `agent-network/`                      |
 | End-to-end tests            | `e2e/`                                                       |
 
-## Security
-
-### Never fail open
-
-When a security check — access control, an IP restriction, an auth decision —
-hits an error such as an unparseable value, an unavailable lookup, or a state it
-does not recognize, it must **deny**. Never skip the check or allow the request
-through because the check itself failed, and make the `default` and unknown cases
-of a security-related `switch` deny rather than fall through.
-
-### Daemon RPC input is untrusted
-
-The agent runs as root (LocalSystem on Windows), so a daemon RPC crosses a
-privilege boundary: treat every field as untrusted input rather than as something
-the UI or CLI validated on the way in.
-
-When you add or change an RPC, ask what the handler does with caller input while
-running as root. If the answer touches a filesystem path, a URL or host, or a
-privileged state change, it needs a gate **in the handler** — a check in the client
-that normally calls it is not a check at all.
-
-- **A caller-supplied path the daemon opens.** Never `os.Open` it as root.
-  Constrain it, then open it *as the caller* with `ipcauth.OpenOwnedFile`, which
-  opens `O_NOFOLLOW`, requires a regular file, and refuses a file the caller does
-  not own — so a symlink or hardlink aimed at a root-only file is rejected.
-- **A caller-supplied URL or host the daemon fetches.** Restrict the scheme and
-  allow only known hosts for unprivileged callers. Prefer a lexical host
-  allowlist plus TLS verification over "resolve the host, then reject private
-  IPs": the resolve-then-trust pattern has a DNS-rebinding race (public IP at
-  check time, attacker IP at connect time), while a name allowlist has no IP
-  check to race. Never accept `http://` where `https://` is expected.
-- **A privileged state change** (SSH root login, management URL, deregistration)
-  gates on the caller identity from `ipcauth.CallerIdentity(ctx)`.
-
-Caller identity comes from the kernel — `SO_PEERCRED`, `LOCAL_PEERCRED`, or the
-named-pipe client token — and never from an RPC field. When
-`ipcauth.CallerIdentity` reports that it could not determine an identity, **deny**;
-do not fall back to treating the caller as the transport peer.
-
-## Agent conventions
-
-### Three networking modes
-
-Where packets actually flow depends on the mode the agent is running in. The
-three are not interchangeable, so establish which one a change applies to — and
-what it should do in the other two — before you write it.
-
-- **kernel mode** (Linux only): in-kernel WireGuard®. The kernel handles both
-  peer-to-peer and routed traffic, and ACLs are iptables or nftables rules. The
-  client programs kernel facilities but never sees the traffic itself.
-- **userspace mode** (wireguard-go with a TUN): wireguard-go runs in-process. The
-  kernel handles peer-to-peer traffic once it leaves the TUN, while routed traffic
-  — exit nodes and network routes — goes through the userspace forwarder, which
-  terminates the connection and re-establishes it over OS sockets. Used on
-  platforms without kernel WireGuard® or when the user opts out.
-- **netstack mode**: wireguard-go in-process with no TUN and no kernel
-  networking. The forwarder does all routing by stitching userspace sockets, and
-  listeners such as the embedded SSH and DNS servers bind on a gVisor netstack.
-  Used where the process cannot create a TUN device, such as the embedded client
-  (`client/embed/`) and the WASM build.
-
-### The overlay interface is not "WireGuard"
-
-Do not put "WireGuard" in identifiers or comments unless the code is genuinely
-coupled to WireGuard® specifically — a wireguard-go call, a handshake field, a
-kernel WireGuard® netlink attribute. For the interface, the host, peers, or
-traffic in general, say "the NetBird interface", "the interface", or "the overlay".
-Most firewall, routing, and DNS code is transport-agnostic, so a WireGuard®
-reference there is simply inaccurate and rots as the transports change.
-
-### IPv6 is a soft feature
-
-The IPv6 overlay is opt-in dual-stack, and capability can change at runtime. Treat
-it as soft rather than a requirement:
-
-- Gate local v6 paths on the interface accessor (`wgIface.Address().HasIPv6()`),
-  not on raw state fields, and skip the v6 path when the host has no v6 rather
-  than returning an error.
-- Treat an empty or unparseable peer v6 address as "no v6 for that peer" and skip
-  it, keeping the v4 path working.
-- Never let a missing v6 break v4. Fail-closed is for security checks; a
-  capability mismatch skips the v6 work and carries on.
-
-### Environment variables
-
-Name the variable in a constant and parse booleans with `strconv.ParseBool` rather
-than comparing strings inline, so an unexpected value is logged instead of
-silently meaning false:
-
-```go
-const EnvDisableFeature = "NB_DISABLE_FEATURE"
-
-func isDisabledByEnv() bool {
-    val := os.Getenv(EnvDisableFeature)
-    if val == "" {
-        return false
-    }
-    disabled, err := strconv.ParseBool(val)
-    if err != nil {
-        log.Warnf("failed to parse %s: %v", EnvDisableFeature, err)
-        return false
-    }
-    return disabled
-}
-```
-
-### Validating against protocol specs
-
-When a change depends on what a protocol actually mandates, read the specification
-text from the [IETF datatracker](https://datatracker.ietf.org/) rather than a
-summary, and check that you have the current RFC — the widely cited one for a
-protocol is often superseded. Cite the section, not just the document, so a
-reviewer can jump straight to the rule.
-
 ## Repo-wide principles
 
 1. **Run `go fmt` on every modified Go file.** Formatting is not optional.
-2. **Zero unaddressed linter warnings.** Fix what `golangci-lint` reports on code
-   you touch, and delete imports, helpers, and parameters your refactor orphaned.
+2. **Zero unaddressed diagnostics.** Fix IDE and linter warnings on code you
+   touch, and delete imports, helpers, and parameters your refactor orphaned.
    Exception: unused parameters in shared code may be consumed by builds outside
    this repository — do not remove them, ask instead.
 3. **Function comments are mandatory for exported functions**, written as full
@@ -291,12 +175,9 @@ reviewer can jump straight to the rule.
 7. **Avoid LLM-slop tells:** em dashes, hedging narration, restating the diff in
    prose, trailing summaries. Defaults, not absolute bans. Applies to code,
    comments, commit messages, and PR descriptions alike.
-8. **Concurrency: do a two-pass race analysis after every change** that touches
-   shared state, including reads of existing maps and slices. Guard them with a
-   mutex (or an atomic or channel where that fits better), keep critical
-   sections short, and run `go test -race` on the touched packages. See
-   [Concurrency and lifecycle](#concurrency-and-lifecycle) for the failure modes
-   to check for.
+8. **Concurrency: do a two-pass race analysis after every change** that adds
+   shared state. Guard maps and slices with a mutex, keep critical sections
+   short, and run `go test -race` on the touched packages.
 9. **Cross-platform builds must keep working.** The agent targets Linux, macOS,
    Windows, FreeBSD, Android, and iOS. When you add a platform-specific file,
    add the counterpart or a build-tagged fallback for the others.
@@ -304,93 +185,6 @@ reviewer can jump straight to the rule.
 11. **Never log secrets** — private keys, setup keys, tokens, PAT values — and
     keep peer IPs and hostnames out of logs above debug level.
 
-## Type safety
-
-**No bare primitives for domain concepts.** A `string` parameter for an account
-ID next to a `string` parameter for a peer ID is two bugs waiting to happen,
-because the compiler cannot catch the swap. Declare the type once and use it
-throughout, converting only at the boundaries where data enters or leaves —
-protobuf, gRPC, HTTP, an external library.
-
-```go
-type ServiceID string
-type AccountID string
-
-// Internal: typed all the way through
-func (r *Router) RemoveRoute(host SNIHost, svcID ServiceID) { ... }
-
-// Proto boundary: convert once, on the way in and on the way out
-svcID := ServiceID(mapping.GetId())
-req.ServiceId = string(svcID)
-```
-
-- **IP addresses are `netip.Addr`**, not `string` and not `net.IP`. Parse at the
-  boundary and pass the typed value inward.
-- **Always `Unmap()`** after parsing an address, after converting from `net.IP`,
-  and after extracting one from `RemoteAddr()`. This normalizes a v4-mapped v6
-  address (`::ffff:10.1.2.3`) to plain v4 so IPv4 rules match it. A stored or
-  compared mapped address silently fails to match those rules.
-- **Ports are `uint16`** internally; use `int` only where a library forces it and
-  convert immediately.
-- **Enums are a typed string with constants**, so the valid set is discoverable
-  and a typo fails to compile.
-- **Map keys follow the same rule**, and must be a real type (`type ServiceID
-  string`) rather than an alias (`type serviceID = string`) — an alias silently
-  accepts bare strings.
-
-## Concurrency and lifecycle
-
-Beyond the mutex hygiene in the principles above, check for these failure
-modes.
-
-- **Never read a struct field inside a goroutine** when another goroutine may nil
-  or reassign it. Pass the value as a parameter, or capture it into a local before
-  launching. This matters most when `Stop()` nils a field without waiting for the
-  goroutine to finish.
-
-  ```go
-  go func(ifaceName string) {   // good: passed in, cannot be nilled underneath
-      m.Start(ctx, ifaceName)
-  }(iface.Name())
-  ```
-
-- **Never wait on a channel while holding a lock the sender needs.** Copy what you
-  need out from under the lock, release it, then wait.
-
-  ```go
-  func (m *Manager) Stop() {
-      m.mu.Lock()
-      cancel, done := m.cancel, m.done
-      m.mu.Unlock()
-      if cancel != nil {
-          cancel()
-          <-done
-      }
-  }
-  ```
-
-- **`Stop`/`Close` must be idempotent** — guard on an already-stopped flag or a
-  nil cancel — and must release the state they guarded. Clear maps and caches;
-  a cancelled goroutine holding a live map still pins that memory. Note that a
-  nil map only panics on writes; reads and iteration behave like an empty map,
-  so where post-close use must be rejected, check the stopped flag explicitly.
-- **Publish coupled state only after every fallible step succeeds.** When several
-  fields form an invariant, build them into locals and assign them to the receiver
-  at the end. Assigning as you go leaves the object half-initialized when a later
-  step fails, so a readiness predicate reports ready while a coupled field is nil.
-  If an earlier step already had an external side effect — a created chain, an
-  opened handle, an inserted rule — roll it back before returning the error.
-- **Clean up what you own on constructor error paths.** Once a constructor has
-  started something, every later error path must undo it: cancel a goroutine and
-  wait for it to exit, stop a ticker, close a watcher. The object is never
-  returned, so its `Close` will never run.
-- **A failed `Start` must undo everything it started.** When a component brings up
-  several subsystems in sequence — connection manager, watchers, routing, DNS,
-  flow, persisted state — a failure partway through has to tear down the ones
-  already running, not just close the handle the error came from. Put the
-  already-started guard *before* that teardown path, so a rejected second `Start`
-  cannot dismantle the one that is running.
-
 ## Error handling
 
 Use single-assignment form when the error is only needed inside the `if`:
@@ -454,45 +248,6 @@ Log the errors you choose not to act on:
 - Close errors may be ignored for read-only operations; log them at debug for
   writes.
 
-**Do not log and return the same error.** It gets reported twice, from two places,
-and the second reader cannot tell whether it happened once or twice. Return it and
-let the caller decide. The exception is an API handler that has already written a
-response. Internal helpers return errors rather than logging and swallowing them.
-
-**Never return a typed nil as an error.** A nil `*MyError` stored in an `error`
-interface is not nil, so `err != nil` is true and callers take the failure path on
-success. Return the error only where it is actually set:
-
-```go
-if _, err := conn.Write(buf); err != nil {   // good
-    return err
-}
-return nil
-```
-
-**Accumulate with `multierror` when an operation should continue past individual
-failures** — teardown, cleanup, or setup where partial success is acceptable.
-`client/errors.FormatErrorOrNil` returns nil for an empty accumulator, so callers
-still see a plain nil on full success:
-
-```go
-func (m *Manager) Cleanup() error {
-    var merr *multierror.Error
-    for _, r := range m.resources {
-        if err := r.Close(); err != nil {
-            merr = multierror.Append(merr, fmt.Errorf("close %s: %w", r.Name, err))
-        }
-    }
-    return nberrors.FormatErrorOrNil(merr)
-}
-```
-
-| Scenario              | Approach              | Why                                       |
-| --------------------- | --------------------- | ----------------------------------------- |
-| Cleanup / teardown    | Accumulate            | Clean up as much as possible              |
-| Setup with rollback   | Abort on first error  | Partial state is invalid; undo what stuck |
-| Setup with partial OK | Accumulate            | Degraded operation is still useful        |
-
 ## Comments
 
 Comment the **why**, never the **what**. Default to no comment, and add one only
@@ -514,14 +269,10 @@ checksum = updateChecksum(checksum, oldPort, newPort)
 
 ### Length budget
 
-Neither of these is linter-enforced, so they are conventions the surrounding code
-mostly follows rather than hard limits:
-
-- **Around 90 characters per line.** Wrap the comment rather than running well past
-  it.
-- **Roughly 250 characters per comment**, about three wrapped lines. Doc comments
-  on exported identifiers may exceed it when the API genuinely needs the
-  explanation; inline comments inside a function body rarely should.
+- **90 characters per line.** Wrap the comment, do not run past it.
+- **250 characters per comment**, roughly three wrapped lines. Doc comments on
+  exported identifiers may exceed it when the API genuinely needs the
+  explanation; inline comments inside a function body may not.
 
 The budget is a smell detector, not a rule to game. Do not compress a needed
 explanation into cryptic shorthand to fit — if a block of code needs more than
@@ -578,19 +329,6 @@ up, and the 250-character budget does not apply to them.
   otherwise.
 - **Message guidance:** optional for `NoError`/`Error`; always give context for
   comparison, boolean, and collection assertions.
-- **Reproduce a bug before fixing it.** Write the test, watch it fail *for the
-  reason you expect* — a test that fails for an unrelated reason proves nothing —
-  then apply the fix and confirm it passes. Add the thin surrounding cases while
-  you are there.
-- **Use `t.Setenv`** rather than `os.Setenv` so the previous value is restored on
-  cleanup. To test the unset case, call `t.Setenv` first to register the restore,
-  then `os.Unsetenv`.
-- **Prefer `t.Cleanup` over `defer`** in any test with parallel subtests: the
-  parent function returns, running its `defer`s, while parallel subtests are
-  still suspended. Sequential subtests finish inside `t.Run`, so `defer` is safe
-  there, but `t.Cleanup` works in both cases.
-- **Explanatory comments in tests are welcome.** Describe the scenario being set
-  up; the comment budget below does not apply to them.
 
 ```go
 server, err := StartTestServer()
@@ -642,8 +380,7 @@ assert.Equal(t, expectedResult, result, "Result should match expected")
   than replacing it with your own summary: describe the change, link the issue,
   tick the checklist honestly (including "ran locally" and "single purpose"),
   and complete the documentation section. Do not tick a box you have not
-  verified, and do not delete rows that do not apply — the docs gate in CI reads
-  that section and fails when it is missing.
+  verified, and do not delete rows that do not apply.
 
 - **Keep the PR description short.** Under 1000 words on top of the template's
   own text, and usually far less — a few paragraphs. Reviewers read the diff;
@@ -702,12 +439,6 @@ assert.Equal(t, expectedResult, result, "Result should match expected")
   on their own. Propose that split to the user rather than opening one large PR
   and hoping.
 
-  Prefer GitHub's stacked pull requests for such a sequence, rather than
-  hand-managing base branches: open each PR against the branch below it instead of
-  `main`, so every PR's diff shows only its own change. Merging a layer retargets
-  the PRs above it, and branch protections and required checks on the base branch
-  still apply to each one.
-
 - **User-facing changes need a docs PR** in
   [netbirdio/docs](https://github.com/netbirdio/docs), linked from the PR
   description.

client/android/client.go

@@ -15,6 +15,7 @@ import (
 
 	log "github.com/sirupsen/logrus"
 
+	nbAnonymize "github.com/netbirdio/netbird/client/anonymize"
 	"github.com/netbirdio/netbird/client/iface/device"
 	"github.com/netbirdio/netbird/client/internal"
 	"github.com/netbirdio/netbird/client/internal/debug"
@@ -278,8 +279,10 @@ func (c *Client) GetTunSettings() (*TunSettings, error) {
 }
 
 // DebugBundle generates a debug bundle, uploads it, and returns the upload key.
-// It works both with and without a running engine.
-func (c *Client) DebugBundle(platformFiles PlatformFiles, anonymize bool) (string, error) {
+// It works both with and without a running engine. anonymizeLevel is "default"
+// or "strict"; strict also anonymizes internal IP ranges, peer names, and
+// WireGuard public keys, and implies anonymize.
+func (c *Client) DebugBundle(platformFiles PlatformFiles, anonymize bool, anonymizeLevel string) (string, error) {
 	cfg, cacheDir, cc := c.stateSnapshot()
 
 	// If the engine hasn't been started, load config from disk
@@ -298,6 +301,7 @@ func (c *Client) DebugBundle(platformFiles PlatformFiles, anonymize bool) (strin
 		InternalConfig: cfg,
 		StatusRecorder: c.recorder,
 		TempDir:        cacheDir,
+		StatePath:      platformFiles.StateFilePath(),
 	}
 
 	if cc != nil {
@@ -321,6 +325,7 @@ func (c *Client) DebugBundle(platformFiles PlatformFiles, anonymize bool) (strin
 		deps,
 		debug.BundleConfig{
 			Anonymize:         anonymize,
+			AnonymizeLevel:    nbAnonymize.ParseLevel(anonymizeLevel),
 			IncludeSystemInfo: true,
 		},
 	)

client/anonymize/anonymize.go

@@ -2,6 +2,7 @@ package anonymize
 
 import (
 	"crypto/rand"
+	"encoding/base64"
 	"fmt"
 	"math/big"
 	"net"
@@ -15,13 +16,81 @@ import (
 
 const anonTLD = ".domain"
 
+// Level selects how much the anonymizer redacts. Levels are ordered: a higher
+// level redacts strictly more. On the wire (protos, flags) levels travel as
+// their string form.
+type Level int
+
+const (
+	// LevelDefault anonymizes public IP addresses, IPv6 ULA, domains, and MAC
+	// addresses. Internal IPv4 ranges (RFC 1918, CGNAT, link-local) are
+	// preserved so support can reason about the real topology.
+	LevelDefault Level = iota
+	// LevelStrict additionally anonymizes internal IP ranges, peer names, and
+	// WireGuard public keys.
+	LevelStrict
+)
+
+// ParseLevel maps s to a Level. Empty means LevelDefault; anything
+// unrecognized maps to LevelStrict so an unknown request never yields less
+// anonymization than intended.
+func ParseLevel(s string) Level {
+	switch strings.ToLower(s) {
+	case "", LevelDefault.String():
+		return LevelDefault
+	default:
+		return LevelStrict
+	}
+}
+
+// String returns the wire form of the level: "default" or "strict".
+func (l Level) String() string {
+	if l >= LevelStrict {
+		return "strict"
+	}
+	return "default"
+}
+
+// protectedDomains are NetBird-operated suffixes that stay recognizable in an
+// anonymized bundle. At LevelStrict the labels in front of them (the peer
+// name) are still replaced, except under netbird.io, which only hosts
+// NetBird infrastructure (api, signal, flow), never peer names.
+var protectedDomains = []string{"netbird.io", "netbird.selfhosted", "netbird.cloud", "netbird.stage"}
+
+const infraDomain = "netbird.io"
+
+var (
+	macColonRegex = regexp.MustCompile(`\b[0-9a-fA-F]{2}(?::[0-9a-fA-F]{2}){5}\b`)
+	macDashRegex  = regexp.MustCompile(`\b[0-9a-fA-F]{2}(?:-[0-9a-fA-F]{2}){5}\b`)
+	wgKeyRegex    = regexp.MustCompile(`\b[A-Za-z0-9+/]{43}=`)
+)
+
 type Anonymizer struct {
 	ipAnonymizer     map[netip.Addr]netip.Addr
 	domainAnonymizer map[string]string
-	currentAnonIPv4  netip.Addr
-	currentAnonIPv6  netip.Addr
-	startAnonIPv4    netip.Addr
-	startAnonIPv6    netip.Addr
+	// domainOrder caches the keys of domainAnonymizer sorted longest-first
+	// for AnonymizeString; it is rebuilt when the map gains entries.
+	domainOrder     []string
+	labelAnonymizer map[string]string
+	labelAnonymized map[string]struct{}
+	labelCounter    uint32
+	macAnonymizer   map[string]string
+	macCounter      uint32
+	wgKeyAnonymizer map[string]string
+	wgKeyAnonymized map[string]struct{}
+	currentAnonIPv4 netip.Addr
+	currentAnonIPv6 netip.Addr
+	startAnonIPv4   netip.Addr
+	startAnonIPv6   netip.Addr
+
+	// LevelStrict also anonymizes internal ranges (RFC 1918, CGNAT,
+	// link-local), replacing them from the dedicated internal pools below so
+	// a reader can still tell an internal address from a public one.
+	level                   Level
+	currentAnonInternalIPv4 netip.Addr
+	currentAnonInternalIPv6 netip.Addr
+	startAnonInternalIPv4   netip.Addr
+	startAnonInternalIPv6   netip.Addr
 
 	domainKeyRegex *regexp.Regexp
 }
@@ -32,25 +101,50 @@ func DefaultAddresses() (netip.Addr, netip.Addr) {
 	return netip.AddrFrom4([4]byte{198, 51, 100, 0}), netip.MustParseAddr("2001:db8:ffff::")
 }
 
+// InternalAddresses returns the pool starts used in strict mode for internal
+// ranges. Both are reserved ranges that cannot collide with real addressing:
+// 198.18.0.0 (RFC 2544 benchmarking), 2001:db8:1:: (RFC 3849 documentation).
+func InternalAddresses() (netip.Addr, netip.Addr) {
+	return netip.AddrFrom4([4]byte{198, 18, 0, 0}), netip.MustParseAddr("2001:db8:1::")
+}
+
 func NewAnonymizer(startIPv4, startIPv6 netip.Addr) *Anonymizer {
+	internalIPv4, internalIPv6 := InternalAddresses()
 	return &Anonymizer{
 		ipAnonymizer:     map[netip.Addr]netip.Addr{},
 		domainAnonymizer: map[string]string{},
+		labelAnonymizer:  map[string]string{},
+		labelAnonymized:  map[string]struct{}{},
+		macAnonymizer:    map[string]string{},
+		wgKeyAnonymizer:  map[string]string{},
+		wgKeyAnonymized:  map[string]struct{}{},
 		currentAnonIPv4:  startIPv4,
 		currentAnonIPv6:  startIPv6,
 		startAnonIPv4:    startIPv4,
 		startAnonIPv6:    startIPv6,
 
+		level:                   LevelDefault,
+		currentAnonInternalIPv4: internalIPv4,
+		currentAnonInternalIPv6: internalIPv6,
+		startAnonInternalIPv4:   internalIPv4,
+		startAnonInternalIPv6:   internalIPv6,
+
 		domainKeyRegex: regexp.MustCompile(`\bdomain=([^\s,:"]+)`),
 	}
 }
 
+// SetLevel selects the anonymization level. The zero value of a new
+// Anonymizer is LevelDefault.
+func (a *Anonymizer) SetLevel(level Level) {
+	a.level = level
+}
+
 func (a *Anonymizer) AnonymizeIP(ip netip.Addr) netip.Addr {
+	// Normalize 4-in-6 addresses so ::ffff:192.168.1.1 classifies and maps
+	// like 192.168.1.1.
+	ip = ip.Unmap()
+
 	if ip.IsLoopback() ||
-		ip.IsLinkLocalUnicast() ||
-		ip.IsLinkLocalMulticast() ||
-		ip.IsInterfaceLocalMulticast() ||
-		(ip.Is4() && ip.IsPrivate()) ||
 		ip.IsUnspecified() ||
 		ip.IsMulticast() ||
 		isWellKnown(ip) ||
@@ -59,18 +153,100 @@ func (a *Anonymizer) AnonymizeIP(ip netip.Addr) netip.Addr {
 		return ip
 	}
 
+	if isInternal(ip) && a.level < LevelStrict {
+		return ip
+	}
+
 	if _, ok := a.ipAnonymizer[ip]; !ok {
-		if ip.Is4() {
-			a.ipAnonymizer[ip] = a.currentAnonIPv4
-			a.currentAnonIPv4 = a.currentAnonIPv4.Next()
-		} else {
-			a.ipAnonymizer[ip] = a.currentAnonIPv6
-			a.currentAnonIPv6 = a.currentAnonIPv6.Next()
-		}
+		a.ipAnonymizer[ip] = a.nextAnonIP(ip)
 	}
 	return a.ipAnonymizer[ip]
 }
 
+func (a *Anonymizer) nextAnonIP(ip netip.Addr) netip.Addr {
+	// At the strict level, internal addresses (including IPv6 ULA, matched
+	// by IsPrivate) come from the internal pools so they remain recognizable
+	// as internal without disclosing the real values.
+	if a.level >= LevelStrict && (isInternal(ip) || ip.IsPrivate()) {
+		if ip.Is4() {
+			anon := a.currentAnonInternalIPv4
+			a.currentAnonInternalIPv4 = a.currentAnonInternalIPv4.Next()
+			return anon
+		}
+		anon := a.currentAnonInternalIPv6
+		a.currentAnonInternalIPv6 = a.currentAnonInternalIPv6.Next()
+		return anon
+	}
+
+	if ip.Is4() {
+		anon := a.currentAnonIPv4
+		a.currentAnonIPv4 = a.currentAnonIPv4.Next()
+		return anon
+	}
+	anon := a.currentAnonIPv6
+	a.currentAnonIPv6 = a.currentAnonIPv6.Next()
+	return anon
+}
+
+// AnonymizeMAC replaces a MAC address with a consistent placeholder from the
+// locally administered range starting at 02:00:00:00:00:01, at every
+// anonymization level. Broadcast, multicast, all-zero, and already assigned
+// placeholder addresses are preserved. The colon and dash spellings of the
+// same address share one placeholder; the output keeps the input's separator.
+func (a *Anonymizer) AnonymizeMAC(mac string) string {
+	hw, err := net.ParseMAC(mac)
+	if err != nil || len(hw) != 6 {
+		return mac
+	}
+
+	if isWellKnownMAC(hw) || a.isAnonymizedMAC(hw) {
+		return mac
+	}
+
+	key := hw.String()
+	anon, ok := a.macAnonymizer[key]
+	if !ok {
+		a.macCounter++
+		anon = fmt.Sprintf("02:00:00:%02x:%02x:%02x", byte(a.macCounter>>16), byte(a.macCounter>>8), byte(a.macCounter))
+		a.macAnonymizer[key] = anon
+	}
+
+	if strings.Contains(mac, "-") {
+		anon = strings.ReplaceAll(anon, ":", "-")
+	}
+	return anon
+}
+
+// isAnonymizedMAC reports whether hw is a placeholder this anonymizer already
+// handed out, so a second pass over anonymized output leaves it unchanged.
+func (a *Anonymizer) isAnonymizedMAC(hw net.HardwareAddr) bool {
+	if hw[0] != 0x02 || hw[1] != 0 || hw[2] != 0 {
+		return false
+	}
+	value := uint32(hw[3])<<16 | uint32(hw[4])<<8 | uint32(hw[5])
+	return value <= a.macCounter
+}
+
+// AnonymizeWGKey replaces a WireGuard public key with a consistent random
+// placeholder of the same shape. Keys are only anonymized at LevelStrict;
+// placeholders already handed out pass through unchanged.
+func (a *Anonymizer) AnonymizeWGKey(key string) string {
+	if a.level < LevelStrict || !looksLikeWGKey(key) {
+		return key
+	}
+	if _, ok := a.wgKeyAnonymized[key]; ok {
+		return key
+	}
+
+	anon, ok := a.wgKeyAnonymizer[key]
+	if !ok {
+		anon = generateAnonymousKey()
+		a.wgKeyAnonymizer[key] = anon
+		a.wgKeyAnonymized[anon] = struct{}{}
+	}
+	return anon
+}
+
 func (a *Anonymizer) AnonymizeUDPAddr(addr net.UDPAddr) net.UDPAddr {
 	// Convert IP to netip.Addr
 	ip, ok := netip.AddrFromSlice(addr.IP)
@@ -89,12 +265,12 @@ func (a *Anonymizer) AnonymizeUDPAddr(addr net.UDPAddr) net.UDPAddr {
 
 // isInAnonymizedRange checks if an IP is within the range of already assigned anonymized IPs
 func (a *Anonymizer) isInAnonymizedRange(ip netip.Addr) bool {
-	if ip.Is4() && ip.Compare(a.startAnonIPv4) >= 0 && ip.Compare(a.currentAnonIPv4) <= 0 {
-		return true
-	} else if !ip.Is4() && ip.Compare(a.startAnonIPv6) >= 0 && ip.Compare(a.currentAnonIPv6) <= 0 {
-		return true
+	if ip.Is4() {
+		return inPoolRange(ip, a.startAnonIPv4, a.currentAnonIPv4) ||
+			inPoolRange(ip, a.startAnonInternalIPv4, a.currentAnonInternalIPv4)
 	}
-	return false
+	return inPoolRange(ip, a.startAnonIPv6, a.currentAnonIPv6) ||
+		inPoolRange(ip, a.startAnonInternalIPv6, a.currentAnonInternalIPv6)
 }
 
 func (a *Anonymizer) AnonymizeIPString(ip string) string {
@@ -118,14 +294,17 @@ func (a *Anonymizer) AnonymizeDomain(domain string) string {
 		baseDomain = domain[:len(domain)-1]
 	}
 
-	if strings.HasSuffix(baseDomain, "netbird.io") ||
-		strings.HasSuffix(baseDomain, "netbird.selfhosted") ||
-		strings.HasSuffix(baseDomain, "netbird.cloud") ||
-		strings.HasSuffix(baseDomain, "netbird.stage") ||
-		strings.HasSuffix(baseDomain, anonTLD) {
+	if strings.HasSuffix(baseDomain, anonTLD) {
 		return domain
 	}
 
+	if suffix := protectedSuffix(baseDomain); suffix != "" {
+		if a.level < LevelStrict || baseDomain == suffix || suffix == infraDomain {
+			return domain
+		}
+		return withTrailingDot(a.anonymizePeerName(baseDomain, suffix), hasDot)
+	}
+
 	parts := strings.Split(baseDomain, ".")
 	if len(parts) < 2 {
 		return domain
@@ -141,12 +320,53 @@ func (a *Anonymizer) AnonymizeDomain(domain string) string {
 	}
 
 	result := strings.Replace(baseDomain, baseForLookup, anonymized, 1)
-	if hasDot {
-		result += "."
+	if a.level >= LevelStrict && len(parts) > 2 {
+		prefix := strings.TrimSuffix(baseDomain, "."+baseForLookup)
+		result = a.anonymizeLabels(prefix, "host") + "." + anonymized
+		// The full mapping feeds AnonymizeString so seeded FQDNs are caught
+		// in log lines as a whole, labels included.
+		a.domainAnonymizer[baseDomain] = result
+	}
+	return withTrailingDot(result, hasDot)
+}
+
+// anonymizePeerName replaces the labels in front of a protected suffix with
+// numbered peer placeholders, keeping the suffix, and records the full
+// mapping for string replacement in logs. The numbering keeps a peer
+// recognizable across the whole bundle without disclosing its name.
+func (a *Anonymizer) anonymizePeerName(baseDomain, suffix string) string {
+	prefix := strings.TrimSuffix(baseDomain, "."+suffix)
+	result := a.anonymizeLabels(prefix, "peer") + "." + suffix
+	if result != baseDomain {
+		a.domainAnonymizer[baseDomain] = result
 	}
 	return result
 }
 
+// anonymizeLabels replaces each dot-separated label with a consistent
+// numbered placeholder ("<placeholder>-<n>"). Wildcard labels and
+// placeholders already handed out pass through unchanged.
+func (a *Anonymizer) anonymizeLabels(prefix, placeholder string) string {
+	labels := strings.Split(prefix, ".")
+	for i, label := range labels {
+		if label == "*" {
+			continue
+		}
+		if _, ok := a.labelAnonymized[label]; ok {
+			continue
+		}
+		anon, ok := a.labelAnonymizer[label]
+		if !ok {
+			a.labelCounter++
+			anon = fmt.Sprintf("%s-%d", placeholder, a.labelCounter)
+			a.labelAnonymizer[label] = anon
+			a.labelAnonymized[anon] = struct{}{}
+		}
+		labels[i] = anon
+	}
+	return strings.Join(labels, ".")
+}
+
 func (a *Anonymizer) AnonymizeURI(uri string) string {
 	u, err := url.Parse(uri)
 	if err != nil {
@@ -181,16 +401,70 @@ func (a *Anonymizer) AnonymizeString(str string) string {
 	str = ipv4Regex.ReplaceAllStringFunc(str, a.AnonymizeIPString)
 	str = ipv6Regex.ReplaceAllStringFunc(str, a.AnonymizeIPString)
 
-	for domain, anonDomain := range a.domainAnonymizer {
-		str = strings.ReplaceAll(str, domain, anonDomain)
+	for _, domain := range a.sortedDomains() {
+		str = strings.ReplaceAll(str, domain, a.domainAnonymizer[domain])
 	}
 
 	str = a.AnonymizeSchemeURI(str)
 	str = a.AnonymizeDNSLogLine(str)
 
+	// MAC handling runs after the IP passes so preserved IPv6 addresses are
+	// already out of the way; the separator guard skips matches embedded in a
+	// longer colon- or dash-separated sequence (such as an IPv6 tail).
+	str = a.anonymizeMACsInString(str, macColonRegex, ':')
+	str = a.anonymizeMACsInString(str, macDashRegex, '-')
+
+	if a.level >= LevelStrict {
+		str = wgKeyRegex.ReplaceAllStringFunc(str, a.AnonymizeWGKey)
+	}
+
 	return str
 }
 
+// sortedDomains returns the domain mappings longest-first, so a full-FQDN
+// mapping (strict level) is applied before the base-domain mapping it
+// contains. The order is rebuilt only when domainAnonymizer has grown.
+func (a *Anonymizer) sortedDomains() []string {
+	if len(a.domainOrder) == len(a.domainAnonymizer) {
+		return a.domainOrder
+	}
+
+	a.domainOrder = a.domainOrder[:0]
+	for domain := range a.domainAnonymizer {
+		a.domainOrder = append(a.domainOrder, domain)
+	}
+	slices.SortFunc(a.domainOrder, func(x, y string) int {
+		if d := len(y) - len(x); d != 0 {
+			return d
+		}
+		return strings.Compare(x, y)
+	})
+	return a.domainOrder
+}
+
+// anonymizeMACsInString replaces MAC addresses matched by re, skipping
+// matches that directly adjoin another sep so a six-group run inside a longer
+// separated sequence is left alone.
+func (a *Anonymizer) anonymizeMACsInString(str string, re *regexp.Regexp, sep byte) string {
+	matches := re.FindAllStringIndex(str, -1)
+	if len(matches) == 0 {
+		return str
+	}
+
+	var b strings.Builder
+	last := 0
+	for _, m := range matches {
+		if (m[0] > 0 && str[m[0]-1] == sep) || (m[1] < len(str) && str[m[1]] == sep) {
+			continue
+		}
+		b.WriteString(str[last:m[0]])
+		b.WriteString(a.AnonymizeMAC(str[m[0]:m[1]]))
+		last = m[1]
+	}
+	b.WriteString(str[last:])
+	return b.String()
+}
+
 // AnonymizeSchemeURI finds and anonymizes URIs with ws, wss, rel, rels, stun, stuns, turn, and turns schemes.
 func (a *Anonymizer) AnonymizeSchemeURI(text string) string {
 	re := regexp.MustCompile(`(?i)\b(wss?://|rels?://|stuns?:|turns?:|https?://)\S+\b`)
@@ -239,10 +513,79 @@ func isWellKnown(addr netip.Addr) bool {
 		"128.0.0.0", "8000::", // 2nd split subnet for default routes
 	}
 
-	if slices.Contains(wellKnown, addr.String()) {
+	return slices.Contains(wellKnown, addr.String())
+}
+
+// isInternal reports whether ip identifies a host only within the local
+// network: IPv4 private (RFC 1918), CGNAT (RFC 6598), and link-local (v4 and
+// v6). These are preserved at the default level so support can reason about
+// the real topology, and replaced from the internal pools at the strict
+// level. IPv6 ULA is deliberately not internal: its random global ID uniquely
+// fingerprints the network, so it is anonymized at every level.
+func isInternal(ip netip.Addr) bool {
+	return (ip.Is4() && ip.IsPrivate()) ||
+		ip.IsLinkLocalUnicast() ||
+		isCGNAT(ip)
+}
+
+func inPoolRange(ip, start, current netip.Addr) bool {
+	return ip.Compare(start) >= 0 && ip.Compare(current) <= 0
+}
+
+// isWellKnownMAC reports whether hw carries no stable host identity: all-zero
+// or a group address (broadcast and multicast).
+func isWellKnownMAC(hw net.HardwareAddr) bool {
+	if hw[0]&1 == 1 {
 		return true
 	}
+	for _, b := range hw {
+		if b != 0 {
+			return false
+		}
+	}
+	return true
+}
 
+// looksLikeWGKey reports whether s has the shape of a WireGuard key:
+// 44 base64 characters decoding to 32 bytes.
+func looksLikeWGKey(s string) bool {
+	if len(s) != 44 || s[43] != '=' {
+		return false
+	}
+	decoded, err := base64.StdEncoding.DecodeString(s)
+	return err == nil && len(decoded) == 32
+}
+
+func generateAnonymousKey() string {
+	buf := make([]byte, 32)
+	if _, err := rand.Read(buf); err != nil {
+		return strings.Repeat("A", 43) + "="
+	}
+	return base64.StdEncoding.EncodeToString(buf)
+}
+
+// protectedSuffix returns the protected NetBird suffix baseDomain ends with,
+// or empty. The match is label-anchored so an unrelated domain that merely
+// ends in the same characters is not preserved.
+func protectedSuffix(baseDomain string) string {
+	for _, d := range protectedDomains {
+		if baseDomain == d || strings.HasSuffix(baseDomain, "."+d) {
+			return d
+		}
+	}
+	return ""
+}
+
+func withTrailingDot(domain string, hasDot bool) string {
+	if hasDot {
+		return domain + "."
+	}
+	return domain
+}
+
+// isCGNAT reports whether addr is in 100.64.0.0/10 (RFC 6598), the range
+// NetBird assigns overlay peer addresses from.
+func isCGNAT(addr netip.Addr) bool {
 	cgnatRangeStart := netip.AddrFrom4([4]byte{100, 64, 0, 0})
 	cgnatRange := netip.PrefixFrom(cgnatRangeStart, 10)
 

client/anonymize/anonymize_test.go

@@ -1,8 +1,11 @@
 package anonymize_test
 
 import (
+	"bytes"
+	"encoding/base64"
 	"net/netip"
 	"regexp"
+	"strings"
 	"testing"
 
 	"github.com/stretchr/testify/assert"
@@ -44,6 +47,301 @@ func TestAnonymizeIP(t *testing.T) {
 	}
 }
 
+func TestParseLevel(t *testing.T) {
+	tests := []struct {
+		input  string
+		expect anonymize.Level
+	}{
+		{"", anonymize.LevelDefault},
+		{"default", anonymize.LevelDefault},
+		{"DEFAULT", anonymize.LevelDefault},
+		{"strict", anonymize.LevelStrict},
+		{"STRICT", anonymize.LevelStrict},
+		// Unknown values must never yield less anonymization than requested.
+		{"garbage", anonymize.LevelStrict},
+	}
+
+	for _, tc := range tests {
+		t.Run("input="+tc.input, func(t *testing.T) {
+			assert.Equal(t, tc.expect, anonymize.ParseLevel(tc.input), "parsed level should match")
+		})
+	}
+}
+
+func TestAnonymizeIP_DefaultLevelInternalRanges(t *testing.T) {
+	anonymizer := anonymize.NewAnonymizer(anonymize.DefaultAddresses())
+
+	tests := []struct {
+		name   string
+		ip     string
+		expect string
+	}{
+		{"RFC1918 10/8", "10.1.2.3", "10.1.2.3"},
+		{"RFC1918 172.16/12", "172.16.5.5", "172.16.5.5"},
+		{"RFC1918 192.168/16", "192.168.1.1", "192.168.1.1"},
+		{"CGNAT", "100.64.0.5", "100.64.0.5"},
+		{"IPv4 link-local", "169.254.1.1", "169.254.1.1"},
+		{"IPv6 link-local", "fe80::1", "fe80::1"},
+		// ULA is anonymized even at the default level: its random global ID
+		// uniquely fingerprints the network, unlike shared RFC 1918 space.
+		{"IPv6 ULA", "fd12:3456:789a::1", "2001:db8:ffff::"},
+		// 4-in-6 addresses classify like their unmapped IPv4 form.
+		{"4-in-6 RFC1918", "::ffff:192.168.1.1", "192.168.1.1"},
+		{"4-in-6 CGNAT", "::ffff:100.64.0.5", "100.64.0.5"},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			result := anonymizer.AnonymizeIP(netip.MustParseAddr(tc.ip))
+			assert.Equal(t, tc.expect, result.String(), "default level should preserve internal ranges except ULA")
+		})
+	}
+}
+
+func TestAnonymizeIP_StrictLevel(t *testing.T) {
+	anonymizer := anonymize.NewAnonymizer(anonymize.DefaultAddresses())
+	anonymizer.SetLevel(anonymize.LevelStrict)
+
+	// Order matters: internal pool addresses are assigned sequentially.
+	tests := []struct {
+		name   string
+		ip     string
+		expect string
+	}{
+		{"RFC1918 192.168/16", "192.168.1.1", "198.18.0.0"},
+		{"Second RFC1918", "192.168.1.2", "198.18.0.1"},
+		{"Repeated RFC1918", "192.168.1.1", "198.18.0.0"},
+		{"RFC1918 10/8", "10.1.2.3", "198.18.0.2"},
+		{"RFC1918 172.16/12", "172.16.5.5", "198.18.0.3"},
+		{"CGNAT", "100.64.0.5", "198.18.0.4"},
+		{"IPv4 link-local", "169.254.1.1", "198.18.0.5"},
+		{"Public IPv4 uses public pool", "1.2.3.4", "198.51.100.0"},
+		{"IPv6 link-local", "fe80::1", "2001:db8:1::"},
+		{"IPv6 ULA", "fd12:3456:789a::1", "2001:db8:1::1"},
+		{"Public IPv6 uses public pool", "2607:f8b0:4005:805::200e", "2001:db8:ffff::"},
+		{"Loopback IPv4", "127.0.0.1", "127.0.0.1"},
+		{"Loopback IPv6", "::1", "::1"},
+		{"Unspecified", "0.0.0.0", "0.0.0.0"},
+		{"Multicast", "224.0.0.251", "224.0.0.251"},
+		{"Well known resolver", "8.8.8.8", "8.8.8.8"},
+		{"Well known split marker", "128.0.0.0", "128.0.0.0"},
+		{"In internal pool range", "198.18.0.3", "198.18.0.3"},
+		{"In public pool range", "198.51.100.0", "198.51.100.0"},
+		{"4-in-6 repeated RFC1918", "::ffff:192.168.1.1", "198.18.0.0"},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			result := anonymizer.AnonymizeIP(netip.MustParseAddr(tc.ip))
+			assert.Equal(t, tc.expect, result.String(), "strict level should replace internal ranges from the internal pools")
+		})
+	}
+}
+
+func TestAnonymizeString_StrictInternalIPs(t *testing.T) {
+	anonymizer := anonymize.NewAnonymizer(anonymize.DefaultAddresses())
+	anonymizer.SetLevel(anonymize.LevelStrict)
+
+	input := "route 10.20.30.0/24 via 192.168.1.1 dev eth0 src 100.64.0.7"
+	firstPass := anonymizer.AnonymizeString(input)
+	secondPass := anonymizer.AnonymizeString(firstPass)
+
+	assert.NotContains(t, firstPass, "10.20.30.0", "private network address should be anonymized")
+	assert.NotContains(t, firstPass, "192.168.1.1", "private gateway should be anonymized")
+	assert.NotContains(t, firstPass, "100.64.0.7", "CGNAT address should be anonymized")
+	assert.Contains(t, firstPass, "/24", "prefix length should be preserved")
+	assert.Equal(t, firstPass, secondPass, "second pass should not further anonymize the string")
+}
+
+func TestAnonymizeMAC(t *testing.T) {
+	anonymizer := anonymize.NewAnonymizer(anonymize.DefaultAddresses())
+
+	first := anonymizer.AnonymizeMAC("aa:bb:cc:dd:ee:0f")
+	assert.Equal(t, "02:00:00:00:00:01", first, "first MAC should get the first placeholder")
+	assert.Equal(t, first, anonymizer.AnonymizeMAC("aa:bb:cc:dd:ee:0f"), "repeated MAC should map to the same placeholder")
+	assert.Equal(t, first, anonymizer.AnonymizeMAC("AA:BB:CC:DD:EE:0F"), "case should not affect the mapping")
+	assert.Equal(t, "02-00-00-00-00-01", anonymizer.AnonymizeMAC("AA-BB-CC-DD-EE-0F"), "dash form should keep its separator but share the mapping")
+
+	second := anonymizer.AnonymizeMAC("10:22:33:44:55:66")
+	assert.Equal(t, "02:00:00:00:00:02", second, "second distinct MAC should get the next placeholder")
+
+	tests := []struct {
+		name string
+		mac  string
+	}{
+		{"Broadcast", "ff:ff:ff:ff:ff:ff"},
+		{"IPv4 multicast", "01:00:5e:00:00:fb"},
+		{"IPv6 multicast", "33:33:00:00:00:01"},
+		{"All zero", "00:00:00:00:00:00"},
+		{"Assigned placeholder", "02:00:00:00:00:01"},
+		{"Invalid", "not-a-mac"},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			assert.Equal(t, tc.mac, anonymizer.AnonymizeMAC(tc.mac), "should be preserved")
+		})
+	}
+}
+
+func TestAnonymizeString_MACAddresses(t *testing.T) {
+	anonymizer := anonymize.NewAnonymizer(anonymize.DefaultAddresses())
+
+	tests := []struct {
+		name   string
+		input  string
+		expect string
+	}{
+		{
+			name:   "nftables ether rule",
+			input:  "ether saddr aa:bb:cc:dd:ee:ff drop",
+			expect: "ether saddr 02:00:00:00:00:01 drop",
+		},
+		{
+			name:   "Windows dash form",
+			input:  "Physical Address : AA-BB-CC-DD-EE-FF",
+			expect: "Physical Address : 02-00-00-00-00-01",
+		},
+		{
+			name:   "IPv6 address tail is not treated as MAC",
+			input:  "addr fe80:0:11:22:33:44:55:66 scope link",
+			expect: "addr fe80:0:11:22:33:44:55:66 scope link",
+		},
+		{
+			name:   "broadcast MAC preserved",
+			input:  "dst ff:ff:ff:ff:ff:ff type ARP",
+			expect: "dst ff:ff:ff:ff:ff:ff type ARP",
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			result := anonymizer.AnonymizeString(tc.input)
+			assert.Equal(t, tc.expect, result, "MAC addresses should be anonymized at every level")
+			assert.Equal(t, result, anonymizer.AnonymizeString(result), "second pass should not change the result")
+		})
+	}
+}
+
+func TestAnonymizeWGKey(t *testing.T) {
+	key := base64.StdEncoding.EncodeToString(bytes.Repeat([]byte{0x42}, 32))
+
+	t.Run("default level preserves keys", func(t *testing.T) {
+		anonymizer := anonymize.NewAnonymizer(anonymize.DefaultAddresses())
+		assert.Equal(t, key, anonymizer.AnonymizeWGKey(key), "default level should not touch WireGuard keys")
+	})
+
+	t.Run("strict level replaces keys", func(t *testing.T) {
+		anonymizer := anonymize.NewAnonymizer(anonymize.DefaultAddresses())
+		anonymizer.SetLevel(anonymize.LevelStrict)
+
+		anon := anonymizer.AnonymizeWGKey(key)
+		assert.NotEqual(t, key, anon, "strict level should replace the key")
+		assert.Regexp(t, `^[A-Za-z0-9+/]{43}=$`, anon, "placeholder should keep the WireGuard key shape")
+		assert.Equal(t, anon, anonymizer.AnonymizeWGKey(key), "repeated key should map to the same placeholder")
+		assert.Equal(t, anon, anonymizer.AnonymizeWGKey(anon), "an assigned placeholder should pass through unchanged")
+
+		assert.Equal(t, "not-a-key", anonymizer.AnonymizeWGKey("not-a-key"), "non-key values should be preserved")
+	})
+}
+
+func TestAnonymizeString_WGKeys(t *testing.T) {
+	key := base64.StdEncoding.EncodeToString(bytes.Repeat([]byte{0x42}, 32))
+	input := "peer " + key + " handshake completed"
+
+	t.Run("default level preserves keys", func(t *testing.T) {
+		anonymizer := anonymize.NewAnonymizer(anonymize.DefaultAddresses())
+		assert.Equal(t, input, anonymizer.AnonymizeString(input), "default level should not touch WireGuard keys in strings")
+	})
+
+	t.Run("strict level replaces keys", func(t *testing.T) {
+		anonymizer := anonymize.NewAnonymizer(anonymize.DefaultAddresses())
+		anonymizer.SetLevel(anonymize.LevelStrict)
+
+		firstPass := anonymizer.AnonymizeString(input)
+		assert.NotContains(t, firstPass, key, "the key should not survive strict anonymization")
+		assert.Equal(t, anonymizer.AnonymizeWGKey(key), extractKey(t, firstPass), "string replacement should be consistent with AnonymizeWGKey")
+		assert.Equal(t, firstPass, anonymizer.AnonymizeString(firstPass), "second pass should not change the result")
+	})
+}
+
+func extractKey(t *testing.T, logLine string) string {
+	t.Helper()
+	fields := strings.Fields(logLine)
+	require.Len(t, fields, 4, "log line should keep its structure")
+	return fields[1]
+}
+
+func TestAnonymizeDomain_StrictLevel(t *testing.T) {
+	anonymizer := anonymize.NewAnonymizer(anonymize.DefaultAddresses())
+	anonymizer.SetLevel(anonymize.LevelStrict)
+
+	t.Run("netbird peer name", func(t *testing.T) {
+		result := anonymizer.AnonymizeDomain("my-laptop.netbird.cloud")
+		assert.Regexp(t, `^peer-\d+\.netbird\.cloud$`, result, "peer name should be anonymized, suffix kept")
+		assert.NotContains(t, result, "my-laptop", "the peer name should not survive")
+		assert.Equal(t, result, anonymizer.AnonymizeDomain("my-laptop.netbird.cloud"), "repeated domain should map consistently")
+		assert.Equal(t, result, anonymizer.AnonymizeDomain(result), "an anonymized domain should pass through unchanged")
+	})
+
+	t.Run("bare netbird domain", func(t *testing.T) {
+		assert.Equal(t, "netbird.cloud", anonymizer.AnonymizeDomain("netbird.cloud"), "the bare protected suffix should be preserved")
+	})
+
+	t.Run("netbird infrastructure preserved", func(t *testing.T) {
+		assert.Equal(t, "api.netbird.io", anonymizer.AnonymizeDomain("api.netbird.io"),
+			"netbird.io hosts infrastructure, not peer names, and should stay readable")
+	})
+
+	t.Run("leading labels of other domains", func(t *testing.T) {
+		result := anonymizer.AnonymizeDomain("host1.corp.example.com")
+		assert.Regexp(t, `^host-\d+\.host-\d+\.anon-[a-zA-Z0-9]+\.domain$`, result, "every label should be anonymized")
+		for _, label := range []string{"host1", "corp", "example"} {
+			assert.NotContains(t, result, label, "no original label should survive")
+		}
+		assert.Equal(t, result, anonymizer.AnonymizeDomain("host1.corp.example.com"), "repeated domain should map consistently")
+	})
+
+	t.Run("same label maps consistently across domains", func(t *testing.T) {
+		first := anonymizer.AnonymizeDomain("shared.one.com")
+		second := anonymizer.AnonymizeDomain("shared.two.com")
+		assert.Equal(t, strings.Split(first, ".")[0], strings.Split(second, ".")[0], "the shared host label should get one placeholder")
+	})
+
+	t.Run("wildcard label preserved", func(t *testing.T) {
+		result := anonymizer.AnonymizeDomain("*.example.com")
+		assert.Regexp(t, `^\*\.anon-[a-zA-Z0-9]+\.domain$`, result, "the wildcard label should stay a wildcard")
+	})
+}
+
+func TestAnonymizeDomain_DefaultLevelKeepsPeerNames(t *testing.T) {
+	anonymizer := anonymize.NewAnonymizer(anonymize.DefaultAddresses())
+
+	assert.Equal(t, "my-laptop.netbird.cloud", anonymizer.AnonymizeDomain("my-laptop.netbird.cloud"),
+		"default level should preserve netbird FQDNs including the peer name")
+	assert.Regexp(t, `^sub\.anon-[a-zA-Z0-9]+\.domain$`, anonymizer.AnonymizeDomain("sub.example.com"),
+		"default level should keep subdomain labels")
+}
+
+func TestAnonymizeString_StrictPeerNames(t *testing.T) {
+	anonymizer := anonymize.NewAnonymizer(anonymize.DefaultAddresses())
+	anonymizer.SetLevel(anonymize.LevelStrict)
+
+	// Seed like the bundle generator does from the status: base first, then
+	// the full FQDN, so replacement must prefer the longer mapping.
+	anonBase := anonymizer.AnonymizeDomain("example.com")
+	anonPeer := anonymizer.AnonymizeDomain("peer1.netbird.cloud")
+	anonHost := anonymizer.AnonymizeDomain("host1.example.com")
+
+	logLine := "connected to peer1.netbird.cloud via host1.example.com endpoint"
+	firstPass := anonymizer.AnonymizeString(logLine)
+	assert.NotContains(t, firstPass, "peer1", "the peer name should not survive in logs")
+	assert.NotContains(t, firstPass, "host1", "the host label should not survive in logs")
+	assert.Contains(t, firstPass, anonPeer, "the seeded peer mapping should be applied")
+	assert.Contains(t, firstPass, anonHost, "the seeded host mapping should be applied, not just the base mapping")
+	assert.NotContains(t, firstPass, "host1."+anonBase, "the base mapping must not preempt the longer FQDN mapping")
+	assert.Equal(t, firstPass, anonymizer.AnonymizeString(firstPass), "second pass should not change the result")
+}
+
 func TestAnonymizeDNSLogLine(t *testing.T) {
 	anonymizer := anonymize.NewAnonymizer(netip.Addr{}, netip.Addr{})
 	tests := []struct {

client/cmd/debug.go

@@ -27,8 +27,8 @@ import (
 const errCloseConnection = "Failed to close connection: %v"
 
 var (
-	logFileCount        uint32
-	systemInfoFlag      bool
+	logFileCount             uint32
+	systemInfoFlag           bool
 	uploadBundleFlag         bool
 	uploadBundleURLFlag      string
 	uploadBundleInsecureFlag bool
@@ -156,6 +156,11 @@ func debugConfigDump(cmd *cobra.Command, _ []string) error {
 // request. Returns an error if the RPC fails or if the daemon reports
 // an upload failure reason.
 func debugBundle(cmd *cobra.Command, _ []string) error {
+	anonymizeEnabled, anonymizeLevel, err := effectiveAnonymize()
+	if err != nil {
+		return err
+	}
+
 	conn, err := getClient(cmd)
 	if err != nil {
 		return err
@@ -168,10 +173,11 @@ func debugBundle(cmd *cobra.Command, _ []string) error {
 
 	client := proto.NewDaemonServiceClient(conn)
 	request := &proto.DebugBundleRequest{
-		Anonymize:    anonymizeFlag,
-		SystemInfo:   systemInfoFlag,
-		LogFileCount: logFileCount,
-		CliVersion:   version.NetbirdVersion(),
+		Anonymize:      anonymizeEnabled,
+		AnonymizeLevel: anonymizeLevel.String(),
+		SystemInfo:     systemInfoFlag,
+		LogFileCount:   logFileCount,
+		CliVersion:     version.NetbirdVersion(),
 	}
 	if uploadBundleFlag {
 		request.UploadURL = uploadBundleURLFlag
@@ -229,6 +235,11 @@ func runForDuration(cmd *cobra.Command, args []string) error {
 		return fmt.Errorf("invalid duration format: %v", err)
 	}
 
+	anonymizeEnabled, anonymizeLevel, err := effectiveAnonymize()
+	if err != nil {
+		return err
+	}
+
 	conn, err := getClient(cmd)
 	if err != nil {
 		return err
@@ -368,10 +379,11 @@ func runForDuration(cmd *cobra.Command, args []string) error {
 	cmd.Println("Creating debug bundle...")
 
 	request := &proto.DebugBundleRequest{
-		Anonymize:    anonymizeFlag,
-		SystemInfo:   systemInfoFlag,
-		LogFileCount: logFileCount,
-		CliVersion:   version.NetbirdVersion(),
+		Anonymize:      anonymizeEnabled,
+		AnonymizeLevel: anonymizeLevel.String(),
+		SystemInfo:     systemInfoFlag,
+		LogFileCount:   logFileCount,
+		CliVersion:     version.NetbirdVersion(),
 	}
 	if uploadBundleFlag {
 		request.UploadURL = uploadBundleURLFlag

client/cmd/root.go

@@ -21,6 +21,7 @@ import (
 	"github.com/spf13/pflag"
 	"google.golang.org/grpc"
 
+	"github.com/netbirdio/netbird/client/anonymize"
 	daddr "github.com/netbirdio/netbird/client/internal/daemonaddr"
 	"github.com/netbirdio/netbird/client/internal/profilemanager"
 )
@@ -69,6 +70,7 @@ var (
 	autoConnectDisabled     bool
 	extraIFaceBlackList     []string
 	anonymizeFlag           bool
+	anonymizeLevelFlag      string
 	dnsRouteInterval        time.Duration
 	// lazyConnEnabled is the parse target for the deprecated --enable-lazy-connection
 	// flag. The flag is inert; the value is no longer read (use NB_LAZY_CONN instead).
@@ -156,7 +158,8 @@ func init() {
 	rootCmd.MarkFlagsMutuallyExclusive("setup-key", "setup-key-file")
 	rootCmd.PersistentFlags().StringVar(&preSharedKey, preSharedKeyFlag, "", "Sets WireGuard PreSharedKey property. If set, then only peers that have the same key can communicate.")
 	rootCmd.PersistentFlags().StringVarP(&hostName, "hostname", "n", "", "Sets a custom hostname for the device")
-	rootCmd.PersistentFlags().BoolVarP(&anonymizeFlag, "anonymize", "A", false, "anonymize IP addresses and non-netbird.io domains in logs and status output")
+	rootCmd.PersistentFlags().BoolVarP(&anonymizeFlag, "anonymize", "A", false, "anonymize public IP addresses, MAC addresses, and non-netbird.io domains in logs and status output; private, CGNAT, and link-local IP ranges are kept (see --anonymize-level strict)")
+	rootCmd.PersistentFlags().StringVar(&anonymizeLevelFlag, "anonymize-level", "", "anonymization level: \"default\" or \"strict\"; strict also anonymizes private, CGNAT, and link-local IP ranges, peer names, and WireGuard public keys. Setting this flag implies --anonymize")
 	rootCmd.PersistentFlags().StringVarP(&configPath, "config", "c", profilemanager.DefaultConfigPath, "Overrides the default profile file location")
 
 	rootCmd.AddCommand(upCmd)
@@ -293,6 +296,19 @@ var CLIBackOffSettings = &backoff.ExponentialBackOff{
 	Clock:               backoff.SystemClock,
 }
 
+// effectiveAnonymize resolves the --anonymize and --anonymize-level flags:
+// setting a level implies anonymization, and an invalid level is rejected.
+func effectiveAnonymize() (bool, anonymize.Level, error) {
+	if anonymizeLevelFlag == "" {
+		return anonymizeFlag, anonymize.LevelDefault, nil
+	}
+	level := anonymize.ParseLevel(anonymizeLevelFlag)
+	if !strings.EqualFold(anonymizeLevelFlag, level.String()) {
+		return false, anonymize.LevelDefault, fmt.Errorf("invalid anonymize level %q: use %q or %q", anonymizeLevelFlag, anonymize.LevelDefault.String(), anonymize.LevelStrict.String())
+	}
+	return true, level, nil
+}
+
 func getSetupKey() (string, error) {
 	if setupKeyPath != "" && setupKey == "" {
 		return getSetupKeyFromFile(setupKeyPath)

client/cmd/status.go

@@ -121,8 +121,14 @@ func statusFunc(cmd *cobra.Command, args []string) error {
 		sessionExpiresAt = ts.AsTime().UTC()
 	}
 
+	anonymizeEnabled, anonymizeLevel, err := effectiveAnonymize()
+	if err != nil {
+		return err
+	}
+
 	var outputInformationHolder = nbstatus.ConvertToStatusOutputOverview(resp.GetFullStatus(), nbstatus.ConvertOptions{
-		Anonymize:            anonymizeFlag,
+		Anonymize:            anonymizeEnabled,
+		AnonymizeLevel:       anonymizeLevel,
 		DaemonVersion:        resp.GetDaemonVersion(),
 		DaemonStatus:         nbstatus.ParseDaemonStatus(status),
 		StatusFilter:         statusFilter,

client/firewall/firewalld/firewalld.go

@@ -2,8 +2,8 @@
 // its wg interface into firewalld's "trusted" zone. This is required because
 // firewalld's nftables chains are created with NFT_CHAIN_OWNER on recent
 // versions, which returns EPERM to any other process that tries to insert
-// rules into them. Trusting the interface makes firewalld itself add the
-// accept rules to its own chains instead.
+// rules into them. The workaround mirrors what Tailscale does: let firewalld
+// itself add the accept rules to its own chains by trusting the interface.
 package firewalld
 
 // TrustedZone is the firewalld zone name used for interfaces whose traffic

client/internal/debug/debug.go

@@ -34,9 +34,8 @@ import (
 	"github.com/netbirdio/netbird/shared/netiputil"
 )
 
-const readmeContent = `Netbird debug bundle
-This debug bundle contains the following files.
-If the --anonymize flag is set, the files are anonymized to protect sensitive information.
+const readmeContent = `This debug bundle contains the following files.
+If anonymization is enabled (--anonymize / --anonymize-level), the files are anonymized to protect sensitive information.
 
 status.txt: Anonymized status information of the NetBird client.
 client.log: Most recent, anonymized client log file of the NetBird client.
@@ -70,21 +69,34 @@ capture.pcap: Packet capture in pcap format. Only present when capture was runni
 
 
 Anonymization Process
-The files in this bundle have been anonymized to protect sensitive information. Here's how the anonymization was applied:
+The files in this bundle have been anonymized to protect sensitive information. The level applied to this bundle is recorded at the top of this file. Here's how the anonymization was applied:
 
 IP Addresses
 
-IPv4 addresses are replaced with addresses starting from 198.51.100.0
-IPv6 addresses are replaced with addresses starting from 100::
+Default level:
+- Public IPv4 addresses are replaced with addresses starting from 198.51.100.0
+- Public IPv6 addresses are replaced with addresses starting from 2001:db8:ffff::
+- IPv6 unique local addresses (fc00::/7) are anonymized as well: their random global ID uniquely identifies the network.
+- IP addresses from internal IPv4 ranges and well-known addresses are not anonymized (e.g. 8.8.8.8, 100.64.0.0/10, addresses starting with 192.168., 172.16., 10., 169.254., fe80::).
+
+Strict level (--anonymize-level strict), in addition to the default level:
+- Private (RFC 1918), CGNAT (100.64.0.0/10), and link-local (169.254.0.0/16, fe80::/10) addresses are anonymized too.
+- Internal IPv4 addresses are replaced with addresses starting from 198.18.0.0 and internal IPv6 addresses with addresses starting from 2001:db8:1::, so internal addresses remain distinguishable from public ones.
+- Addresses are mapped in order of first appearance: subnet structure, allocation scheme, and gateway conventions are not preserved. Prefix lengths of networks are preserved.
+- Peer names in front of NetBird domains are replaced with numbered placeholders (e.g. peer-1.netbird.cloud), and subdomain labels of other domains with host-N placeholders.
+- WireGuard public keys are replaced with consistent placeholder keys.
 
-IP addresses from non public ranges and well known addresses are not anonymized (e.g. 8.8.8.8, 100.64.0.0/10, addresses starting with 192.168., 172.16., 10., etc.).
 Reoccuring IP addresses are replaced with the same anonymized address.
 
 Note: The anonymized IP addresses in the status file do not match those in the log and routes files. However, the anonymized IP addresses are consistent within the status file and across the routes and log files.
 
+MAC Addresses
+MAC addresses are replaced at every anonymization level with consistent placeholders counting up from 02:00:00:00:00:01. Broadcast, multicast, and all-zero addresses are kept. At the default level a preserved IPv6 link-local address may still embed a MAC address (EUI-64); the strict level anonymizes those addresses.
+
 Domains
 All domain names (except for the netbird domains) are replaced with randomly generated strings ending in ".domain". Anonymized domains are consistent across all files in the bundle.
 Reoccuring domain names are replaced with the same anonymized domain.
+At the strict level, the peer name labels in front of netbird domains are anonymized as well.
 
 Sync Response
 The network_map.json file contains the following anonymized information:
@@ -281,6 +293,7 @@ type BundleGenerator struct {
 	cliVersion     string
 
 	anonymize         bool
+	anonymizeLevel    anonymize.Level
 	includeSystemInfo bool
 	logFileCount      uint32
 
@@ -288,7 +301,10 @@ type BundleGenerator struct {
 }
 
 type BundleConfig struct {
-	Anonymize         bool
+	Anonymize bool
+	// AnonymizeLevel selects how much the anonymizer redacts.
+	// anonymize.LevelStrict implies Anonymize.
+	AnonymizeLevel    anonymize.Level
 	IncludeSystemInfo bool
 	LogFileCount      uint32
 }
@@ -327,8 +343,11 @@ func NewBundleGenerator(deps GeneratorDependencies, cfg BundleConfig) *BundleGen
 		uiLogOpener = openLogFile
 	}
 
+	anonymizer := anonymize.NewAnonymizer(anonymize.DefaultAddresses())
+	anonymizer.SetLevel(cfg.AnonymizeLevel)
+
 	return &BundleGenerator{
-		anonymizer: anonymize.NewAnonymizer(anonymize.DefaultAddresses()),
+		anonymizer: anonymizer,
 
 		internalConfig: deps.InternalConfig,
 		statusRecorder: deps.StatusRecorder,
@@ -345,7 +364,8 @@ func NewBundleGenerator(deps GeneratorDependencies, cfg BundleConfig) *BundleGen
 		daemonVersion:  deps.DaemonVersion,
 		cliVersion:     deps.CliVersion,
 
-		anonymize:         cfg.Anonymize,
+		anonymize:         cfg.Anonymize || cfg.AnonymizeLevel >= anonymize.LevelStrict,
+		anonymizeLevel:    cfg.AnonymizeLevel,
 		includeSystemInfo: cfg.IncludeSystemInfo,
 		logFileCount:      logFileCount,
 	}
@@ -485,7 +505,13 @@ func (g *BundleGenerator) addSystemInfo() {
 }
 
 func (g *BundleGenerator) addReadme() error {
-	readmeReader := strings.NewReader(readmeContent)
+	level := "none (anonymization disabled)"
+	if g.anonymize {
+		level = g.anonymizeLevel.String()
+	}
+	header := fmt.Sprintf("Netbird debug bundle\nAnonymization level applied to this bundle: %s\n", level)
+
+	readmeReader := strings.NewReader(header + readmeContent)
 	if err := g.addFileToZip(readmeReader, "README.txt"); err != nil {
 		return fmt.Errorf("add README file to zip: %w", err)
 	}
@@ -507,9 +533,10 @@ func (g *BundleGenerator) addStatus() error {
 		fullStatus := g.statusRecorder.GetFullStatus()
 		protoFullStatus := nbstatus.ToProtoFullStatus(fullStatus)
 		overview := nbstatus.ConvertToStatusOutputOverview(protoFullStatus, nbstatus.ConvertOptions{
-			Anonymize:     g.anonymize,
-			ProfileName:   profName,
-			DaemonVersion: g.daemonVersion,
+			Anonymize:      g.anonymize,
+			AnonymizeLevel: g.anonymizeLevel,
+			ProfileName:    profName,
+			DaemonVersion:  g.daemonVersion,
 		})
 		overview.CliVersion = g.cliVersion
 		statusOutput := overview.FullDetailSummary()
@@ -662,7 +689,7 @@ func (g *BundleGenerator) addCommonConfigFields(configContent *strings.Builder)
 	configContent.WriteString("NetBird Client Configuration:\n\n")
 
 	if key, err := wgtypes.ParseKey(g.internalConfig.PrivateKey); err == nil {
-		configContent.WriteString(fmt.Sprintf("PublicKey: %s\n", key.PublicKey().String()))
+		configContent.WriteString(fmt.Sprintf("PublicKey: %s\n", g.anonymizer.AnonymizeWGKey(key.PublicKey().String())))
 	}
 	configContent.WriteString(fmt.Sprintf("WgIface: %s\n", g.internalConfig.WgIface))
 	configContent.WriteString(fmt.Sprintf("WgPort: %d\n", g.internalConfig.WgPort))
@@ -952,6 +979,11 @@ func (g *BundleGenerator) addUpdateLogs() error {
 		}
 
 		baseName := filepath.Base(logFile)
+		data, err = g.anonymizeBytes(data)
+		if err != nil {
+			log.Warnf("skipping update log file %s: %v", baseName, err)
+			continue
+		}
 		if err := g.addFileToZip(bytes.NewReader(data), filepath.Join("update-logs", baseName)); err != nil {
 			return fmt.Errorf("add update log file %s to zip: %w", baseName, err)
 		}
@@ -979,6 +1011,13 @@ func (g *BundleGenerator) addCorruptedStateFiles() error {
 		}
 
 		fileName := filepath.Base(match)
+		// Corrupted state files usually fail structured JSON anonymization,
+		// so run them through the string anonymizer instead.
+		data, err = g.anonymizeBytes(data)
+		if err != nil {
+			log.Warnf("skipping corrupted state file %s: %v", fileName, err)
+			continue
+		}
 		if err := g.addFileToZip(bytes.NewReader(data), "corrupted_states/"+fileName); err != nil {
 			log.Warnf("Failed to add corrupted state file %s to zip: %v", fileName, err)
 			continue
@@ -990,6 +1029,27 @@ func (g *BundleGenerator) addCorruptedStateFiles() error {
 	return nil
 }
 
+// anonymizeBytes runs raw file content through the string anonymizer line by
+// line when anonymization is enabled. It errors instead of returning partial
+// content, so a caller never adds an unanonymized fallback to the bundle.
+func (g *BundleGenerator) anonymizeBytes(data []byte) ([]byte, error) {
+	if !g.anonymize {
+		return data, nil
+	}
+
+	var buf bytes.Buffer
+	scanner := bufio.NewScanner(bytes.NewReader(data))
+	scanner.Buffer(make([]byte, 1024*1024), 1024*1024)
+	for scanner.Scan() {
+		buf.WriteString(g.anonymizer.AnonymizeString(scanner.Text()))
+		buf.WriteByte('\n')
+	}
+	if err := scanner.Err(); err != nil {
+		return nil, fmt.Errorf("anonymize content: %w", err)
+	}
+	return buf.Bytes(), nil
+}
+
 func (g *BundleGenerator) addMetrics() error {
 	if g.clientMetrics == nil {
 		log.Debugf("skipping metrics in debug bundle: no metrics collector")
@@ -1462,6 +1522,7 @@ func anonymizeRemotePeer(peer *mgmProto.RemotePeerConfig, anonymizer *anonymize.
 	}
 
 	peer.Fqdn = anonymizer.AnonymizeDomain(peer.Fqdn)
+	peer.WgPubKey = anonymizer.AnonymizeWGKey(peer.WgPubKey)
 
 	anonymizeSSHConfig(peer.SshConfig)
 }

client/internal/debug/wgshow.go

@@ -35,14 +35,14 @@ func (g *BundleGenerator) toWGShowFormat(s *configurer.Stats) string {
 	var sb strings.Builder
 
 	sb.WriteString(fmt.Sprintf("interface: %s\n", s.DeviceName))
-	sb.WriteString(fmt.Sprintf("  public key: %s\n", s.PublicKey))
+	sb.WriteString(fmt.Sprintf("  public key: %s\n", g.anonymizer.AnonymizeWGKey(s.PublicKey)))
 	sb.WriteString(fmt.Sprintf("  listen port: %d\n", s.ListenPort))
 	if s.FWMark != 0 {
 		sb.WriteString(fmt.Sprintf("  fwmark: %#x\n", s.FWMark))
 	}
 
 	for _, peer := range s.Peers {
-		sb.WriteString(fmt.Sprintf("\npeer: %s\n", peer.PublicKey))
+		sb.WriteString(fmt.Sprintf("\npeer: %s\n", g.anonymizer.AnonymizeWGKey(peer.PublicKey)))
 		if peer.Endpoint.IP != nil {
 			if g.anonymize {
 				anonEndpoint := g.anonymizer.AnonymizeUDPAddr(peer.Endpoint)
@@ -54,7 +54,11 @@ func (g *BundleGenerator) toWGShowFormat(s *configurer.Stats) string {
 		if len(peer.AllowedIPs) > 0 {
 			var ipStrings []string
 			for _, ipnet := range peer.AllowedIPs {
-				ipStrings = append(ipStrings, ipnet.String())
+				ipStr := ipnet.String()
+				if g.anonymize {
+					ipStr = g.anonymizer.AnonymizeIPString(ipStr)
+				}
+				ipStrings = append(ipStrings, ipStr)
 			}
 			sb.WriteString(fmt.Sprintf("  allowed ips: %s\n", strings.Join(ipStrings, ", ")))
 		}

client/internal/engine.go

@@ -4,7 +4,6 @@ import (
 	"context"
 	"errors"
 	"fmt"
-	"math"
 	"math/rand"
 	"net"
 	"net/netip"
@@ -24,6 +23,7 @@ import (
 	"golang.zx2c4.com/wireguard/tun/netstack"
 	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
 
+	"github.com/netbirdio/netbird/client/anonymize"
 	nberrors "github.com/netbirdio/netbird/client/errors"
 	"github.com/netbirdio/netbird/client/firewall"
 	"github.com/netbirdio/netbird/client/firewall/firewalld"
@@ -51,7 +51,6 @@ import (
 	icemaker "github.com/netbirdio/netbird/client/internal/peer/ice"
 	"github.com/netbirdio/netbird/client/internal/peerstore"
 	"github.com/netbirdio/netbird/client/internal/portforward"
-	"github.com/netbirdio/netbird/client/internal/pqkem"
 	"github.com/netbirdio/netbird/client/internal/profilemanager"
 	"github.com/netbirdio/netbird/client/internal/relay"
 	"github.com/netbirdio/netbird/client/internal/rosenpass"
@@ -199,10 +198,6 @@ type Engine struct {
 	// rpManager is a Rosenpass manager
 	rpManager *rosenpass.Manager
 
-	// pqkemManager runs the ML-KEM post-quantum PSK exchange (gated by NB_ENABLE_PQ_MLKEM).
-	// It owns the data-path transport and peer endpoint routing.
-	pqkemManager *pqkem.Manager
-
 	// syncMsgMux is used to guarantee sequential Management Service message processing
 	syncMsgMux *sync.Mutex
 
@@ -561,11 +556,7 @@ func (e *Engine) Start(netbirdConfig *mgmProto.NetbirdConfig, mgmtURL *url.URL)
 	publicKey := e.config.WgPrivateKey.PublicKey()
 	e.flowManager = netflow.NewManager(e.wgInterface, publicKey[:], e.statusRecorder)
 
-	// Rosenpass and ML-KEM are mutually exclusive. ML-KEM (NB_ENABLE_PQ_MLKEM) takes precedence
-	if e.config.RosenpassEnabled && pqkem.Enabled() {
-		log.Warnf("rosenpass and ML-KEM post-quantum are mutually exclusive; ML-KEM is enabled, so rosenpass is disabled")
-	}
-	if e.config.RosenpassEnabled && !pqkem.Enabled() {
+	if e.config.RosenpassEnabled {
 		log.Infof("rosenpass is enabled")
 		if e.config.RosenpassPermissive {
 			log.Infof("running rosenpass in permissive mode")
@@ -654,35 +645,6 @@ func (e *Engine) Start(netbirdConfig *mgmProto.NetbirdConfig, mgmtURL *url.URL)
 		e.rpManager.SetInterface(e.wgInterface)
 	}
 
-	// Start the ML-KEM PQ manager after the interface is up so its dedicated UDP
-	// transport can bind on the WG overlay IP.
-	if pqkem.Enabled() {
-		tr, pqErr := newPQTransport(e.config.WgAddr.IP)
-		if pqErr != nil {
-			// In strict mode the peer must fail closed; silently continuing without the PQ
-			// exchange would hand out classic tunnels, so treat the bind failure as fatal.
-			if pqkem.Strict() {
-				return fmt.Errorf("pqkem: strict mode enabled but transport bind failed: %w", pqErr)
-			}
-			log.Errorf("pqkem: transport bind failed, exchange disabled: %v", pqErr)
-		} else {
-			cbHandler := pqCallbackHandler{
-				wg: e.wgInterface,
-				// On a persistent rekey failure, re-bootstrap the KEM over Signal: a
-				// fresh signalling offer starts a new exchange that overwrites the
-				// stalled PSK on both sides, recovering from a data-path desync.
-				reoffer: func(remoteKey string) {
-					if conn, ok := e.peerStore.PeerConn(remoteKey); ok {
-						conn.RequestReoffer()
-					}
-				},
-			}
-			e.pqkemManager = pqkem.NewManager(pqkem.LocalID(publicKey.String()), cbHandler, pqkem.NewLogger())
-			e.pqkemManager.Start(tr)
-			log.Infof("pqkem: enabled (udp port %d on overlay %s)", e.pqkemManager.LocalPort(), e.config.WgAddr.IP)
-		}
-	}
-
 	// if inbound conns are blocked there is no need to create the ACL manager
 	if e.firewall != nil && !e.config.BlockInbound {
 		e.acl = acl.NewDefaultManager(e.firewall)
@@ -946,10 +908,6 @@ func (e *Engine) removePeer(peerKey string) error {
 
 	e.connMgr.RemovePeerConn(peerKey)
 
-	if e.pqkemManager != nil {
-		e.pqkemManager.RemovePeer(pqkem.RemoteID(peerKey))
-	}
-
 	err := e.statusRecorder.RemovePeer(peerKey)
 	if err != nil {
 		log.Warnf("received error when removing peer %s from status recorder: %v", peerKey, err)
@@ -1428,6 +1386,7 @@ func (e *Engine) handleBundle(params *mgmProto.BundleParameters) (*mgmProto.JobR
 
 	bundleJobParams := debug.BundleConfig{
 		Anonymize:         params.Anonymize,
+		AnonymizeLevel:    anonymize.ParseLevel(params.AnonymizeLevel),
 		IncludeSystemInfo: true,
 		LogFileCount:      uint32(params.LogFileCount),
 	}
@@ -1936,10 +1895,6 @@ func (e *Engine) createPeerConn(pubKey string, allowedIPs []netip.Prefix, agentV
 		},
 		ICEConfig: e.createICEConfig(),
 	}
-	if e.pqkemManager != nil {
-		config.PQ = pqHandshaker{mgr: e.pqkemManager}
-		config.PQStrict = pqkem.Strict()
-	}
 
 	serviceDependencies := peer.ServiceDependencies{
 		StatusRecorder:     e.statusRecorder,
@@ -2123,10 +2078,6 @@ func (e *Engine) close() {
 		_ = e.rpManager.Close()
 	}
 
-	if e.pqkemManager != nil {
-		e.pqkemManager.Stop()
-	}
-
 	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
 	defer cancel()
 	if err := e.portForwardManager.GracefullyStop(ctx); err != nil {
@@ -2894,13 +2845,6 @@ func convertToOfferAnswer(msg *sProto.Message) (*peer.OfferAnswer, error) {
 
 	relayIP := decodeRelayIP(msg.GetBody().GetRelayServerIP())
 
-	// Ports are uint16 internally; the proto widens them to uint32, so validate the
-	// range before narrowing (a value that does not fit is a malformed message).
-	mlkemPort := msg.GetBody().GetMlkemPort()
-	if mlkemPort > math.MaxUint16 {
-		return nil, fmt.Errorf("invalid ML-KEM port %d in signalling message", mlkemPort)
-	}
-
 	offerAnswer := peer.OfferAnswer{
 		IceCredentials: peer.IceCredentials{
 			UFrag: remoteCred.UFrag,
@@ -2910,8 +2854,6 @@ func convertToOfferAnswer(msg *sProto.Message) (*peer.OfferAnswer, error) {
 		Version:         msg.GetBody().GetNetBirdVersion(),
 		RosenpassPubKey: rosenpassPubKey,
 		RosenpassAddr:   rosenpassAddr,
-		MlkemPayload:    msg.GetBody().GetMlkemPayload(),
-		MlkemPort:       uint16(mlkemPort),
 		RelaySrvAddress: msg.GetBody().GetRelayServerAddress(),
 		RelaySrvIP:      relayIP,
 		SessionID:       sessionID,

client/internal/peer/conn.go

@@ -3,7 +3,6 @@ package peer
 import (
 	"context"
 	"fmt"
-	"math"
 	"net"
 	"net/netip"
 	"runtime"
@@ -27,7 +26,6 @@ import (
 	"github.com/netbirdio/netbird/client/internal/portforward"
 	"github.com/netbirdio/netbird/client/internal/rosenpass"
 	"github.com/netbirdio/netbird/client/internal/stdnet"
-	"github.com/netbirdio/netbird/monotime"
 	"github.com/netbirdio/netbird/route"
 	relayClient "github.com/netbirdio/netbird/shared/relay/client"
 )
@@ -76,39 +74,6 @@ type RosenpassConfig struct {
 	PermissiveMode bool
 }
 
-// PQHandshaker attaches post-quantum ML-KEM material to signalling offers/answers and
-// feeds received material back. It is implemented by the engine over the pqkem
-// manager and is nil when the PQ exchange is disabled. remoteKey is the peer's
-// WireGuard public key.
-type PQHandshaker interface {
-	// OfferPayload returns the KEM offer to embed in an outgoing offer (nil if this
-	// peer is not the KEM initiator) and the local PQ data-path port to announce.
-	OfferPayload(remoteKey string) (payload []byte, port uint16)
-	// ShouldSendBootstrapOffer reports whether, as the controller, we should reply to a
-	// received responder offer with our own KEM offer (true only when no exchange is
-	// already in flight — so we kick the KEM once and ignore further offers).
-	ShouldSendBootstrapOffer(remoteKey string) bool
-	// AnswerPayload processes a received KEM offer (nil if absent) and returns the KEM
-	// answer to embed in the outgoing answer (nil if none) and the local PQ port.
-	AnswerPayload(remoteKey string, recvOffer []byte) (payload []byte, port uint16)
-	// OnAnswer feeds a received KEM answer (nil if absent).
-	OnAnswer(remoteKey string, recvAnswer []byte)
-	// PSK returns the peer's latest derived post-quantum PSK to program at WG
-	// peer-config time (the pull path). ok is false until one has been derived.
-	PSK(remoteKey string) (wgtypes.Key, bool)
-	// SetRemoteAddr registers the peer's data-path endpoint learned from signalling:
-	// its WG overlay IP with the announced pq UDP port (port 0 means the peer omitted
-	// it and is on the default port).
-	SetRemoteAddr(remoteKey string, addr netip.AddrPort)
-	// OnDataPathRekeyed signals a fresh WireGuard handshake for the peer; it clocks the
-	// next chained PSK rotation pushed over the data path. sinceActivity is how long
-	// ago the peer last exchanged real user data, so the rotation can be skipped for
-	// idle tunnels.
-	OnDataPathRekeyed(remoteKey string, sinceActivity time.Duration)
-	// OnDataPathDown signals the peer's tunnel went down.
-	OnDataPathDown(remoteKey string)
-}
-
 // ConnConfig is a peer Connection configuration
 type ConnConfig struct {
 	// Key is a public key of a remote peer
@@ -126,12 +91,6 @@ type ConnConfig struct {
 
 	RosenpassConfig RosenpassConfig
 
-	// PQ carries post-quantum ML-KEM material on offers/answers; nil when disabled.
-	PQ PQHandshaker
-	// PQStrict fails closed: block peer traffic until the ML-KEM PSK is established,
-	// instead of letting the tunnel come up classically and upgrading to PQ later.
-	PQStrict bool
-
 	// ICEConfig ICE protocol configuration
 	ICEConfig icemaker.Config
 }
@@ -190,11 +149,6 @@ type Conn struct {
 	// pendingFirstPacket is the lazyconn-captured handshake init, replayed once the real
 	// transport is up.
 	pendingFirstPacket []byte
-
-	// pqStrictSentinelKey is a per-conn random sentinel PSK used in PQ strict mode to
-	// fail closed: it is programmed until the real ML-KEM PSK is derived, so no session
-	// can form on a non-PQ key. Per-conn random so two strict peers never match by chance.
-	pqStrictSentinelKey *wgtypes.Key
 }
 
 // injectPendingFirstPacket replays the captured handshake through the proxy if present, else
@@ -252,16 +206,6 @@ func NewConn(config ConnConfig, services ServiceDependencies) (*Conn, error) {
 		metricsRecorder:    services.MetricsRecorder,
 	}
 
-	if config.PQ != nil && config.PQStrict {
-		// The sentinel is what makes strict mode fail closed; if we cannot generate it we
-		// must not fall back to a usable key, so fail creating the conn instead.
-		k, err := wgtypes.GenerateKey()
-		if err != nil {
-			return nil, fmt.Errorf("generate pqkem strict-mode sentinel key: %w", err)
-		}
-		conn.pqStrictSentinelKey = &k
-	}
-
 	return conn, nil
 }
 
@@ -363,8 +307,6 @@ func (conn *Conn) Close(signalToRemote bool) {
 
 	if conn.wgWatcherCancel != nil {
 		conn.wgWatcherCancel()
-		conn.wgWatcher = nil
-		conn.wgWatcherCancel = nil
 	}
 	conn.workerRelay.CloseConn()
 	if conn.workerICE != nil {
@@ -463,9 +405,6 @@ func (conn *Conn) ConnID() id.ConnID {
 
 // configureConnection starts proxying traffic from/to local Wireguard and sets connection status to StatusConnected
 func (conn *Conn) onICEConnectionIsReady(priority conntype.ConnPriority, iceConnInfo ICEConnInfo) {
-	// Read the PQ PSK before conn.mu to keep the lock order conn.mu -> manager.
-	pqPSK, pqOK := conn.pqPSK()
-
 	conn.mu.Lock()
 	defer conn.mu.Unlock()
 
@@ -483,7 +422,7 @@ func (conn *Conn) onICEConnectionIsReady(priority conntype.ConnPriority, iceConn
 	if conn.currentConnPriority > priority {
 		conn.Log.Infof("current connection priority (%s) is higher than the new one (%s), do not upgrade connection", conn.currentConnPriority, priority)
 		conn.statusICE.SetConnected()
-		conn.updateIceState(iceConnInfo, pqOK, time.Now())
+		conn.updateIceState(iceConnInfo, time.Now())
 		return
 	}
 
@@ -526,7 +465,7 @@ func (conn *Conn) onICEConnectionIsReady(priority conntype.ConnPriority, iceConn
 	updateTime := time.Now()
 	conn.enableWgWatcherIfNeeded(updateTime)
 
-	presharedKey := conn.presharedKey(iceConnInfo.RosenpassPubKey, pqPSK)
+	presharedKey := conn.presharedKey(iceConnInfo.RosenpassPubKey)
 	if err = conn.endpointUpdater.ConfigureWGEndpoint(ep, presharedKey); err != nil {
 		conn.handleConfigurationFailure(err, wgProxy)
 		return
@@ -542,14 +481,11 @@ func (conn *Conn) onICEConnectionIsReady(priority conntype.ConnPriority, iceConn
 
 	conn.currentConnPriority = priority
 	conn.statusICE.SetConnected()
-	conn.updateIceState(iceConnInfo, pqOK, updateTime)
+	conn.updateIceState(iceConnInfo, updateTime)
 	conn.doOnConnected(iceConnInfo.RosenpassPubKey, iceConnInfo.RosenpassAddr, updateTime)
 }
 
 func (conn *Conn) onICEStateDisconnected(sessionChanged bool) {
-	// Read the PQ PSK before conn.mu to keep the lock order conn.mu -> manager.
-	pqPSK, _ := conn.pqPSK()
-
 	conn.mu.Lock()
 	defer conn.mu.Unlock()
 
@@ -576,7 +512,7 @@ func (conn *Conn) onICEStateDisconnected(sessionChanged bool) {
 		// todo consider to move after the ConfigureWGEndpoint
 		conn.wgProxyRelay.Work()
 
-		presharedKey := conn.presharedKey(conn.rosenpassRemoteKey, pqPSK)
+		presharedKey := conn.presharedKey(conn.rosenpassRemoteKey)
 		if err := conn.endpointUpdater.SwitchWGEndpoint(conn.wgProxyRelay.EndpointAddr(), presharedKey); err != nil {
 			conn.Log.Errorf("failed to switch to relay conn: %v", err)
 		}
@@ -614,9 +550,6 @@ func (conn *Conn) onICEStateDisconnected(sessionChanged bool) {
 }
 
 func (conn *Conn) onRelayConnectionIsReady(rci RelayConnInfo) {
-	// Read the PQ PSK before conn.mu to keep the lock order conn.mu -> manager.
-	pqPSK, pqOK := conn.pqPSK()
-
 	conn.mu.Lock()
 	defer conn.mu.Unlock()
 
@@ -645,7 +578,7 @@ func (conn *Conn) onRelayConnectionIsReady(rci RelayConnInfo) {
 		conn.Log.Debugf("do not switch to relay because current priority is: %s", conn.currentConnPriority.String())
 		conn.setRelayedProxy(wgProxy)
 		conn.statusRelay.SetConnected()
-		conn.updateRelayStatus(rci.relayedConn.RemoteAddr().String(), rci.rosenpassPubKey, pqOK, time.Now())
+		conn.updateRelayStatus(rci.relayedConn.RemoteAddr().String(), rci.rosenpassPubKey, time.Now())
 		return
 	}
 
@@ -656,7 +589,7 @@ func (conn *Conn) onRelayConnectionIsReady(rci RelayConnInfo) {
 	}
 	updateTime := time.Now()
 	conn.enableWgWatcherIfNeeded(updateTime)
-	if err := conn.endpointUpdater.ConfigureWGEndpoint(wgProxy.EndpointAddr(), conn.presharedKey(rci.rosenpassPubKey, pqPSK)); err != nil {
+	if err := conn.endpointUpdater.ConfigureWGEndpoint(wgProxy.EndpointAddr(), conn.presharedKey(rci.rosenpassPubKey)); err != nil {
 		if err := wgProxy.CloseConn(); err != nil {
 			conn.Log.Warnf("Failed to close relay connection: %v", err)
 		}
@@ -675,7 +608,7 @@ func (conn *Conn) onRelayConnectionIsReady(rci RelayConnInfo) {
 	conn.currentConnPriority = conntype.Relay
 	conn.statusRelay.SetConnected()
 	conn.setRelayedProxy(wgProxy)
-	conn.updateRelayStatus(rci.relayedConn.RemoteAddr().String(), rci.rosenpassPubKey, pqOK, updateTime)
+	conn.updateRelayStatus(rci.relayedConn.RemoteAddr().String(), rci.rosenpassPubKey, updateTime)
 	conn.Log.Infof("start to communicate with peer via relay")
 	conn.doOnConnected(rci.rosenpassPubKey, rci.rosenpassAddr, updateTime)
 }
@@ -737,28 +670,12 @@ func (conn *Conn) onGuardEvent() {
 	}
 }
 
-// RequestReoffer sends a fresh signalling offer for the peer, re-running the
-// post-quantum bootstrap over Signal. Used to recover from a persistent data-path
-// rekey failure: a new exchange overwrites the stalled PSK on both sides. No-op if the
-// connection is not open yet.
-func (conn *Conn) RequestReoffer() {
-	conn.mu.Lock()
-	h := conn.handshaker
-	conn.mu.Unlock()
-	if h == nil {
-		return
-	}
-	if err := h.SendOffer(); err != nil {
-		conn.Log.Debugf("pqkem: recovery re-offer failed: %v", err)
-	}
-}
-
 func (conn *Conn) onWGDisconnected(watcherCtx context.Context) {
 	conn.mu.Lock()
+	defer conn.mu.Unlock()
 
 	// watcherCtx guards against a stale watcher tearing down a connection that already superseded it.
 	if conn.ctx.Err() != nil || watcherCtx.Err() != nil {
-		conn.mu.Unlock()
 		return
 	}
 
@@ -776,15 +693,6 @@ func (conn *Conn) onWGDisconnected(watcherCtx context.Context) {
 	}
 
 	conn.escalateWGTimeoutLocked()
-	pq := conn.config.PQ
-	key := conn.config.Key
-	conn.mu.Unlock()
-
-	// Signal the PQ manager outside conn.mu: it may re-enter Conn (reoffer) under
-	// conn.mu, so calling it while holding the lock would invert the lock order.
-	if pq != nil {
-		pq.OnDataPathDown(key)
-	}
 }
 
 // escalateWGTimeoutLocked resets the peer's rosenpass state after repeated
@@ -808,14 +716,14 @@ func (conn *Conn) escalateWGTimeoutLocked() {
 	conn.onDisconnected(conn.config.WgConfig.RemoteKey)
 }
 
-func (conn *Conn) updateRelayStatus(relayServerAddr string, rosenpassPubKey []byte, pqEstablished bool, updateTime time.Time) {
+func (conn *Conn) updateRelayStatus(relayServerAddr string, rosenpassPubKey []byte, updateTime time.Time) {
 	peerState := State{
 		PubKey:             conn.config.Key,
 		ConnStatusUpdate:   updateTime,
 		ConnStatus:         conn.evalStatus(),
 		Relayed:            conn.isRelayed(),
 		RelayServerAddress: relayServerAddr,
-		RosenpassEnabled:   conn.quantumResistant(rosenpassPubKey, pqEstablished),
+		RosenpassEnabled:   isRosenpassEnabled(rosenpassPubKey),
 	}
 
 	err := conn.statusRecorder.UpdatePeerRelayedState(peerState)
@@ -824,7 +732,7 @@ func (conn *Conn) updateRelayStatus(relayServerAddr string, rosenpassPubKey []by
 	}
 }
 
-func (conn *Conn) updateIceState(iceConnInfo ICEConnInfo, pqEstablished bool, updateTime time.Time) {
+func (conn *Conn) updateIceState(iceConnInfo ICEConnInfo, updateTime time.Time) {
 	peerState := State{
 		PubKey:                     conn.config.Key,
 		ConnStatusUpdate:           updateTime,
@@ -834,7 +742,7 @@ func (conn *Conn) updateIceState(iceConnInfo ICEConnInfo, pqEstablished bool, up
 		RemoteIceCandidateType:     iceConnInfo.RemoteIceCandidateType,
 		LocalIceCandidateEndpoint:  iceConnInfo.LocalIceCandidateEndpoint,
 		RemoteIceCandidateEndpoint: iceConnInfo.RemoteIceCandidateEndpoint,
-		RosenpassEnabled:           conn.quantumResistant(iceConnInfo.RosenpassPubKey, pqEstablished),
+		RosenpassEnabled:           isRosenpassEnabled(iceConnInfo.RosenpassPubKey),
 	}
 
 	err := conn.statusRecorder.UpdatePeerICEState(peerState)
@@ -1038,35 +946,6 @@ func (conn *Conn) onWGCheckSuccess() {
 	conn.mu.Lock()
 	conn.wgTimeouts = 0
 	conn.mu.Unlock()
-
-	// A fresh WireGuard handshake clocks the post-quantum PSK rotation. Pass how long
-	// ago the peer last exchanged real user data (keepalives excluded) so the pqkem
-	// manager can skip rotation on idle tunnels — rotating then would push data-path
-	// traffic that keeps the lazy connection artificially active.
-	if conn.config.PQ != nil {
-		conn.config.PQ.OnDataPathRekeyed(conn.config.Key, conn.dataActivityAge())
-	}
-}
-
-// dataActivityAge returns how long ago the peer last exchanged real user data
-// (WireGuard keepalives excluded), per the same LastActivities signal the
-// lazy-connection inactivity monitor uses. It reports a very large duration when no
-// activity has ever been recorded, so the peer is treated as idle.
-//
-// In kernel mode there is no per-peer data-activity signal (LastActivities is
-// userspace-only), so we cannot tell active from idle. We report zero — always
-// "active" — so PSK rotation is not disabled in kernel mode. Lazy back-to-idle is
-// already limited there; the eBPF WG-activity detection (future) will supply a real
-// signal that excludes handshake/pqkem traffic.
-func (conn *Conn) dataActivityAge() time.Duration {
-	if !conn.config.WgConfig.WgInterface.IsUserspaceBind() {
-		return 0
-	}
-	last, ok := conn.config.WgConfig.WgInterface.LastActivities()[conn.config.WgConfig.RemoteKey]
-	if !ok {
-		return time.Duration(math.MaxInt64)
-	}
-	return monotime.Since(last)
 }
 
 // recordConnectionMetrics records connection stage timestamps as metrics
@@ -1107,42 +986,7 @@ func (conn *Conn) AgentVersionString() string {
 	return conn.config.AgentVersion
 }
 
-// pqPSK returns the post-quantum PSK derived for this peer, if the ML-KEM exchange has
-// produced one. It reads the manager WITHOUT conn.mu, so callers fetch it before taking
-// conn.mu: the lock order is always conn.mu -> manager, never the reverse (the manager's
-// reoffer callback re-enters Conn under conn.mu).
-func (conn *Conn) pqPSK() (*wgtypes.Key, bool) {
-	if conn.config.PQ == nil {
-		return nil, false
-	}
-	psk, ok := conn.config.PQ.PSK(conn.config.Key)
-	if !ok {
-		return nil, false
-	}
-	return &psk, true
-}
-
-// presharedKey resolves the WireGuard preshared key for the peer. pqPSK is the
-// post-quantum PSK looked up out of band via pqPSK (nil when none is derived yet), passed
-// in so the manager lock is never taken under conn.mu.
-func (conn *Conn) presharedKey(remoteRosenpassKey []byte, pqPSK *wgtypes.Key) *wgtypes.Key {
-	// Post-quantum: once the ML-KEM exchange has derived a PSK for this peer, program
-	// it here so the peer's next WireGuard handshake adopts it. Applied at peer-config
-	// time (bootstrap / reconnect); steady-state rotation is pushed separately.
-	if conn.config.PQ != nil {
-		if pqPSK != nil {
-			return pqPSK
-		}
-		if conn.config.PQStrict && conn.pqStrictSentinelKey != nil {
-			// Fail closed: program a non-matching sentinel so no session forms on a
-			// non-PQ key until the ML-KEM exchange derives the real PSK (pushed via
-			// SetPresharedKey once it converges). "pending" — turns into a "stuck"
-			// warning from the manager if the exchange keeps failing (see raiseFailure).
-			conn.Log.Debugf("pqkem: strict mode — no PQ PSK yet, blocking peer traffic until the ML-KEM exchange converges")
-			return conn.pqStrictSentinelKey
-		}
-	}
-
+func (conn *Conn) presharedKey(remoteRosenpassKey []byte) *wgtypes.Key {
 	if conn.config.RosenpassConfig.PubKey == nil {
 		return conn.config.WgConfig.PreSharedKey
 	}
@@ -1182,13 +1026,6 @@ func isRosenpassEnabled(remoteRosenpassPubKey []byte) bool {
 	return remoteRosenpassPubKey != nil
 }
 
-// quantumResistant reports whether the peer's tunnel is post-quantum protected, for
-// the status "Quantum resistance" field: either Rosenpass (the remote advertised a
-// Rosenpass key) or the ML-KEM exchange (a PQ PSK has been derived for this peer).
-func (conn *Conn) quantumResistant(remoteRosenpassPubKey []byte, pqEstablished bool) bool {
-	return isRosenpassEnabled(remoteRosenpassPubKey) || pqEstablished
-}
-
 func evalConnStatus(in connStatusInputs) guard.ConnStatus {
 	// "Relay up and needed" — the peer uses relay and the transport is connected.
 	relayUsedAndUp := in.peerUsesRelay && in.relayConnected

client/internal/peer/conn_pq_test.go

@@ -1,90 +0,0 @@
-package peer
-
-import (
-	"net/netip"
-	"testing"
-	"time"
-
-	log "github.com/sirupsen/logrus"
-	"github.com/stretchr/testify/require"
-	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
-)
-
-// fakePQ is a minimal PQHandshaker: only PSK is exercised by presharedKey, the rest
-// are no-op stubs to satisfy the interface.
-type fakePQ struct {
-	psk wgtypes.Key
-	ok  bool
-}
-
-func (f fakePQ) OfferPayload(string) ([]byte, uint16)          { return nil, 0 }
-func (f fakePQ) ShouldSendBootstrapOffer(string) bool          { return false }
-func (f fakePQ) AnswerPayload(string, []byte) ([]byte, uint16) { return nil, 0 }
-func (f fakePQ) OnAnswer(string, []byte)                       {}
-func (f fakePQ) PSK(string) (wgtypes.Key, bool)                { return f.psk, f.ok }
-func (f fakePQ) SetRemoteAddr(string, netip.AddrPort)          {}
-func (f fakePQ) OnDataPathRekeyed(string, time.Duration)       {}
-func (f fakePQ) OnDataPathDown(string)                         {}
-
-// TestConn_presharedKey_PQ covers the post-quantum branch of presharedKey across the
-// three states that matter: a derived PSK is programmed, and — before one exists —
-// strict mode blocks with a sentinel while non-strict falls open to the ordinary key.
-func TestConn_presharedKey_PQ(t *testing.T) {
-	derivedPSK, err := wgtypes.GenerateKey()
-	require.NoError(t, err)
-	nbPSK, err := wgtypes.GenerateKey()
-	require.NoError(t, err)
-
-	newConn := func() *Conn {
-		return &Conn{
-			Log: log.WithField("peer", "pq-test"),
-			config: ConnConfig{
-				Key:             "LLHf3Ma6z6mdLbriAJbqhX7+nM/B71lgw2+91q3LfhU=",
-				LocalKey:        "RRHf3Ma6z6mdLbriAJbqhX7+nM/B71lgw2+91q3LfhU=",
-				WgConfig:        WgConfig{PreSharedKey: &nbPSK},
-				RosenpassConfig: RosenpassConfig{},
-			},
-		}
-	}
-
-	t.Run("derived PSK is programmed", func(t *testing.T) {
-		for _, strict := range []bool{false, true} {
-			c := newConn()
-			c.config.PQ = fakePQ{psk: derivedPSK, ok: true}
-			c.config.PQStrict = strict
-			if strict {
-				sentinel, err := wgtypes.GenerateKey()
-				require.NoError(t, err)
-				c.pqStrictSentinelKey = &sentinel
-			}
-			pqPSK, _ := c.pqPSK()
-			got := c.presharedKey(nil, pqPSK)
-			require.NotNil(t, got)
-			require.Equal(t, derivedPSK, *got, "the derived PQ PSK must win (strict=%v)", strict)
-		}
-	})
-
-	t.Run("non-strict falls open to the ordinary key before a PSK exists", func(t *testing.T) {
-		c := newConn()
-		c.config.PQ = fakePQ{ok: false}
-		c.config.PQStrict = false
-		pqPSK, _ := c.pqPSK()
-		got := c.presharedKey(nil, pqPSK)
-		require.NotNil(t, got, "non-strict must not block")
-		require.Equal(t, nbPSK, *got, "non-strict falls through to the NetBird PSK, not a sentinel")
-	})
-
-	t.Run("strict blocks with the per-conn sentinel before a PSK exists", func(t *testing.T) {
-		sentinel, err := wgtypes.GenerateKey()
-		require.NoError(t, err)
-		c := newConn()
-		c.config.PQ = fakePQ{ok: false}
-		c.config.PQStrict = true
-		c.pqStrictSentinelKey = &sentinel
-		pqPSK, _ := c.pqPSK()
-		got := c.presharedKey(nil, pqPSK)
-		require.NotNil(t, got)
-		require.Equal(t, sentinel, *got, "strict must return the blocking sentinel")
-		require.NotEqual(t, nbPSK, *got, "the sentinel must not be the ordinary key")
-	})
-}

client/internal/peer/conn_test.go

@@ -255,8 +255,8 @@ func TestConn_presharedKey(t *testing.T) {
 			}
 			conn2.config.RosenpassConfig.PermissiveMode = test.conn2Permissive
 
-			conn1PresharedKey := conn1.presharedKey(conn2.config.RosenpassConfig.PubKey, nil)
-			conn2PresharedKey := conn2.presharedKey(conn1.config.RosenpassConfig.PubKey, nil)
+			conn1PresharedKey := conn1.presharedKey(conn2.config.RosenpassConfig.PubKey)
+			conn2PresharedKey := conn2.presharedKey(conn1.config.RosenpassConfig.PubKey)
 
 			if test.conn1ExpectedInitialKey {
 				if conn1PresharedKey == nil {
@@ -294,14 +294,14 @@ func TestConn_presharedKey_RosenpassManaged(t *testing.T) {
 	// When Rosenpass has already initialized the PSK for this peer,
 	// presharedKey must return nil to avoid UpdatePeer overwriting it.
 	conn.rosenpassInitializedPresharedKeyValidator = func(peerKey string) bool { return true }
-	if k := conn.presharedKey([]byte("remote"), nil); k != nil {
+	if k := conn.presharedKey([]byte("remote")); k != nil {
 		t.Fatalf("expected nil presharedKey when Rosenpass manages PSK, got %v", k)
 	}
 
 	// When Rosenpass hasn't taken over yet, presharedKey should provide
 	// a non-nil initial key (deterministic or from NetBird PSK).
 	conn.rosenpassInitializedPresharedKeyValidator = func(peerKey string) bool { return false }
-	if k := conn.presharedKey([]byte("remote"), nil); k == nil {
+	if k := conn.presharedKey([]byte("remote")); k == nil {
 		t.Fatalf("expected non-nil presharedKey before Rosenpass manages PSK")
 	}
 }

client/internal/peer/endpoint.go

@@ -88,7 +88,7 @@ func (e *EndpointUpdater) configureAsResponder(addr *net.UDPAddr, presharedKey *
 	var ctx context.Context
 	ctx, e.cancelFunc = context.WithCancel(context.Background())
 	e.updateWg.Add(1)
-	go e.scheduleDelayedUpdate(ctx, addr)
+	go e.scheduleDelayedUpdate(ctx, addr, presharedKey)
 
 	if err := e.updateWireGuardPeer(nil, presharedKey); err != nil {
 		e.waitForCloseTheDelayedUpdate()
@@ -107,14 +107,8 @@ func (e *EndpointUpdater) waitForCloseTheDelayedUpdate() {
 	e.updateWg.Wait()
 }
 
-// scheduleDelayedUpdate waits for the fallback period, then sets the responder's real
-// endpoint. It deliberately passes a nil preshared key so it only updates the endpoint
-// and leaves the current PSK untouched: the PSK captured when this was scheduled may be
-// stale by now (e.g. the post-quantum bootstrap derived a fresher PSK within the
-// fallback window, applied via SetPresharedKey), and re-applying the captured one would
-// revert WireGuard to a key the remote peer no longer uses — a mismatch that stalls the
-// handshake until the next retry.
-func (e *EndpointUpdater) scheduleDelayedUpdate(ctx context.Context, addr *net.UDPAddr) {
+// scheduleDelayedUpdate waits for the fallback period before updating the endpoint
+func (e *EndpointUpdater) scheduleDelayedUpdate(ctx context.Context, addr *net.UDPAddr, presharedKey *wgtypes.Key) {
 	defer e.updateWg.Done()
 	t := time.NewTimer(fallbackDelay)
 	defer t.Stop()
@@ -123,7 +117,7 @@ func (e *EndpointUpdater) scheduleDelayedUpdate(ctx context.Context, addr *net.U
 	case <-ctx.Done():
 		return
 	case <-t.C:
-		if err := e.updateWireGuardPeer(addr, nil); err != nil {
+		if err := e.updateWireGuardPeer(addr, presharedKey); err != nil {
 			e.log.Errorf("failed to update WireGuard peer, address: %s, error: %v", addr, err)
 		}
 	}

client/internal/peer/handshaker.go

@@ -39,16 +39,6 @@ type OfferAnswer struct {
 	// This value is the local Rosenpass server address when sending the message
 	RosenpassAddr string
 
-	// MlkemPayload carries the post-quantum X25519MLKEM768 handshake message
-	// (pqkem-framed offer on an OFFER, answer on an ANSWER) that seeds the
-	// WireGuard PSK. Opaque here — the pqkem library frames and parses it. Nil
-	// when the peer does not run the ML-KEM PQ exchange.
-	MlkemPayload []byte
-
-	// MlkemPort is the peer's ML-KEM PQ service UDP port (bound on its WG overlay
-	// IP) where data-path rekey messages are sent. Zero when not running the exchange.
-	MlkemPort uint16
-
 	// relay server address
 	RelaySrvAddress string
 	// RelaySrvIP is the IP the remote peer is connected to on its
@@ -91,20 +81,14 @@ type Handshaker struct {
 
 func NewHandshaker(log *log.Entry, config ConnConfig, signaler *Signaler, ice *WorkerICE, relay *WorkerRelay, metricsStages *MetricsStages) *Handshaker {
 	h := &Handshaker{
-		log:           log,
-		config:        config,
-		signaler:      signaler,
-		ice:           ice,
-		relay:         relay,
-		metricsStages: metricsStages,
-		// Buffered by 1: the single Listen goroutine can be busy handling an offer
-		// (sendAnswer does a blocking signal send) exactly when the matching answer
-		// arrives on the other channel. Unbuffered, that answer would hit the
-		// non-blocking send's default and be dropped — fatal for the post-quantum
-		// exchange, which needs the answer to converge. A 1-slot cushion lets it wait
-		// until Listen loops back, without ever blocking the signal receiver.
-		remoteOffersCh: make(chan OfferAnswer, 1),
-		remoteAnswerCh: make(chan OfferAnswer, 1),
+		log:            log,
+		config:         config,
+		signaler:       signaler,
+		ice:            ice,
+		relay:          relay,
+		metricsStages:  metricsStages,
+		remoteOffersCh: make(chan OfferAnswer),
+		remoteAnswerCh: make(chan OfferAnswer),
 	}
 	// assume remote supports ICE until we learn otherwise from received offers
 	h.remoteICESupported.Store(ice != nil)
@@ -127,9 +111,44 @@ func (h *Handshaker) Listen(ctx context.Context) {
 	for {
 		select {
 		case remoteOfferAnswer := <-h.remoteOffersCh:
-			h.handleRemoteOffer(remoteOfferAnswer)
+			h.log.Infof("received offer, running version %s, remote WireGuard listen port %d, session id: %s, remote ICE supported: %t", remoteOfferAnswer.Version, remoteOfferAnswer.WgListenPort, remoteOfferAnswer.SessionIDString(), remoteOfferAnswer.hasICECredentials())
+
+			// Record signaling received for reconnection attempts
+			if h.metricsStages != nil {
+				h.metricsStages.RecordSignalingReceived()
+			}
+
+			h.updateRemoteICEState(&remoteOfferAnswer)
+
+			if h.relayListener != nil {
+				h.relayListener.Notify(&remoteOfferAnswer)
+			}
+
+			if h.iceListener != nil && h.RemoteICESupported() {
+				h.iceListener(&remoteOfferAnswer)
+			}
+
+			if err := h.sendAnswer(); err != nil {
+				h.log.Errorf("failed to send remote offer confirmation: %s", err)
+				continue
+			}
 		case remoteOfferAnswer := <-h.remoteAnswerCh:
-			h.handleRemoteAnswer(remoteOfferAnswer)
+			h.log.Infof("received answer, running version %s, remote WireGuard listen port %d, session id: %s, remote ICE supported: %t", remoteOfferAnswer.Version, remoteOfferAnswer.WgListenPort, remoteOfferAnswer.SessionIDString(), remoteOfferAnswer.hasICECredentials())
+
+			// Record signaling received for reconnection attempts
+			if h.metricsStages != nil {
+				h.metricsStages.RecordSignalingReceived()
+			}
+
+			h.updateRemoteICEState(&remoteOfferAnswer)
+
+			if h.relayListener != nil {
+				h.relayListener.Notify(&remoteOfferAnswer)
+			}
+
+			if h.iceListener != nil && h.RemoteICESupported() {
+				h.iceListener(&remoteOfferAnswer)
+			}
 		case <-ctx.Done():
 			h.log.Infof("stop listening for remote offers and answers")
 			return
@@ -137,119 +156,6 @@ func (h *Handshaker) Listen(ctx context.Context) {
 	}
 }
 
-// onSignalReceived runs the common preamble for a received offer/answer: record the
-// signalling metric, refresh the remote ICE state, and register the peer's post-quantum
-// data-path endpoint learned from the message.
-func (h *Handshaker) onSignalReceived(remoteOfferAnswer *OfferAnswer) {
-	if h.metricsStages != nil {
-		h.metricsStages.RecordSignalingReceived()
-	}
-	h.updateRemoteICEState(remoteOfferAnswer)
-	h.pqRegisterEndpoint(remoteOfferAnswer.MlkemPort)
-}
-
-// notifyListeners hands the offer/answer to the relay and ICE workers so they bring the
-// connection up.
-func (h *Handshaker) notifyListeners(remoteOfferAnswer *OfferAnswer) {
-	if h.relayListener != nil {
-		h.relayListener.Notify(remoteOfferAnswer)
-	}
-	if h.iceListener != nil && h.RemoteICESupported() {
-		h.iceListener(remoteOfferAnswer)
-	}
-}
-
-func (h *Handshaker) handleRemoteOffer(remoteOfferAnswer OfferAnswer) {
-	h.log.Infof("received offer, running version %s, remote WireGuard listen port %d, session id: %s, remote ICE supported: %t", remoteOfferAnswer.Version, remoteOfferAnswer.WgListenPort, remoteOfferAnswer.SessionIDString(), remoteOfferAnswer.hasICECredentials())
-	h.onSignalReceived(&remoteOfferAnswer)
-
-	// If we are the controller running the KEM, a responder's offer is handled by
-	// replying with our own KEM offer, not by answering it (see pqControllerReoffer).
-	if h.pqControllerReoffer() {
-		return
-	}
-
-	// Derive+store the KEM PSK (inside sendAnswer's AnswerPayload) BEFORE bringing up the
-	// connection: the relay/ICE workers configure the WG endpoint, which pulls the PSK
-	// for the first handshake. Notifying them first would race the KEM exchange and hand
-	// the first handshake a not-yet-derived key.
-	if err := h.sendAnswer(&remoteOfferAnswer); err != nil {
-		h.log.Errorf("failed to send remote offer confirmation: %s", err)
-		return
-	}
-	h.notifyListeners(&remoteOfferAnswer)
-}
-
-func (h *Handshaker) handleRemoteAnswer(remoteOfferAnswer OfferAnswer) {
-	h.log.Infof("received answer, running version %s, remote WireGuard listen port %d, session id: %s, remote ICE supported: %t", remoteOfferAnswer.Version, remoteOfferAnswer.WgListenPort, remoteOfferAnswer.SessionIDString(), remoteOfferAnswer.hasICECredentials())
-	h.onSignalReceived(&remoteOfferAnswer)
-
-	// Feed the KEM answer (derive+store PSK) BEFORE bringing up the connection so the WG
-	// endpoint config pulls the real PSK for the first handshake instead of racing ahead
-	// of the KEM exchange.
-	if h.config.PQ != nil {
-		h.config.PQ.OnAnswer(h.config.Key, remoteOfferAnswer.MlkemPayload)
-	}
-	h.notifyListeners(&remoteOfferAnswer)
-}
-
-// pqControllerReoffer handles a responder's offer when we are the controller running the
-// KEM. The KEM material rides only the controller's offer, so the two peers derive a
-// single shared PSK (a bidirectional KEM would yield two different PSKs and WireGuard
-// would pick misaligned ones). Rather than answer the responder's (KEM-less) offer —
-// which would bring WireGuard up on a pre-PQ key before the KEM completes — we reply with
-// our own KEM offer, so the only transaction that establishes the tunnel is the one that
-// also derives the PSK. It also guarantees a responder-initiated wake still triggers a
-// KEM offer (no stuck responder). Sent exactly once per exchange; further offers while
-// one is in flight are ignored (re-sending on every responder offer would be a runaway).
-// The re-offer reuses our stable ICE session id, so the peer dedups repeats.
-//
-// Returns true when it took ownership of the offer (the caller must not answer it).
-func (h *Handshaker) pqControllerReoffer() bool {
-	if h.config.PQ == nil || !isController(h.config) {
-		return false
-	}
-	if h.config.PQ.ShouldSendBootstrapOffer(h.config.Key) {
-		h.log.Debugf("pqkem: controller received a responder offer, replying with our KEM offer instead of an answer")
-		if err := h.sendOffer(); err != nil {
-			h.log.Errorf("failed to send KEM offer in response to peer offer: %s", err)
-		}
-	} else {
-		h.log.Debugf("pqkem: controller received a responder offer but a KEM exchange is already in flight, ignoring")
-	}
-	return true
-}
-
-// pqRegisterEndpoint feeds the post-quantum handshaker the peer's data-path endpoint
-// (its WG overlay IP plus the advertised pq UDP port) learned from a remote offer/answer.
-func (h *Handshaker) pqRegisterEndpoint(remotePort uint16) {
-	if h.config.PQ == nil {
-		return
-	}
-	overlay, ok := h.pqPeerOverlayAddr()
-	if !ok {
-		return
-	}
-	// remotePort may be 0 (the peer omitted it, meaning the default port); the adapter
-	// resolves 0 to DefaultPort.
-	h.config.PQ.SetRemoteAddr(h.config.Key, netip.AddrPortFrom(overlay, remotePort))
-}
-
-// pqPeerOverlayAddr returns the peer's IPv4 overlay address for the pq data path. A
-// RemotePeerConfig carries only the peer overlay (v4 /32, optionally v6 /128) — served
-// routes are programmed on WireGuard separately and never land in WgConfig.AllowedIps —
-// so AllowedIps[0] is the v4 overlay, matching conn.AllowedIP(). The transport is v4
-// (the overlay always has v4; v6 is additive). Returns false when no v4 overlay exists.
-func (h *Handshaker) pqPeerOverlayAddr() (netip.Addr, bool) {
-	if len(h.config.WgConfig.AllowedIps) == 0 {
-		return netip.Addr{}, false
-	}
-	if a := h.config.WgConfig.AllowedIps[0].Addr().Unmap(); a.Is4() {
-		return a, true
-	}
-	return netip.Addr{}, false
-}
-
 func (h *Handshaker) SendOffer() error {
 	h.mu.Lock()
 	defer h.mu.Unlock()
@@ -289,23 +195,13 @@ func (h *Handshaker) sendOffer() error {
 	}
 
 	offer := h.buildOfferAnswer()
-	if h.config.PQ != nil {
-		offer.MlkemPayload, offer.MlkemPort = h.config.PQ.OfferPayload(h.config.Key)
-	}
 	h.log.Debugf("sending offer with serial: %s", offer.SessionIDString())
 
 	return h.signaler.SignalOffer(offer, h.config.Key)
 }
 
-func (h *Handshaker) sendAnswer(remoteOffer *OfferAnswer) error {
+func (h *Handshaker) sendAnswer() error {
 	answer := h.buildOfferAnswer()
-	if h.config.PQ != nil {
-		var recvOffer []byte
-		if remoteOffer != nil {
-			recvOffer = remoteOffer.MlkemPayload
-		}
-		answer.MlkemPayload, answer.MlkemPort = h.config.PQ.AnswerPayload(h.config.Key, recvOffer)
-	}
 	h.log.Debugf("sending answer with serial: %s", answer.SessionIDString())
 
 	return h.signaler.SignalAnswer(answer, h.config.Key)

client/internal/peer/iface.go

@@ -10,7 +10,6 @@ import (
 	"github.com/netbirdio/netbird/client/iface/configurer"
 	"github.com/netbirdio/netbird/client/iface/wgaddr"
 	"github.com/netbirdio/netbird/client/iface/wgproxy"
-	"github.com/netbirdio/netbird/monotime"
 )
 
 type WGIface interface {
@@ -20,11 +19,4 @@ type WGIface interface {
 	GetProxy() wgproxy.Proxy
 	Address() wgaddr.Address
 	RemoveEndpointAddress(key string) error
-	// LastActivities returns the last real-data activity time per peer (WireGuard
-	// keepalives excluded), used to gate post-quantum PSK rotation on active tunnels.
-	LastActivities() map[string]monotime.Time
-	// IsUserspaceBind reports whether WireGuard runs in userspace. Only there does
-	// LastActivities track per-peer data activity; in kernel mode it is unavailable,
-	// so PSK rotation cannot be gated on activity.
-	IsUserspaceBind() bool
 }

client/internal/peer/signaler.go

@@ -63,8 +63,6 @@ func (s *Signaler) signalOfferAnswer(offerAnswer OfferAnswer, remoteKey string,
 		},
 		RosenpassPubKey: offerAnswer.RosenpassPubKey,
 		RosenpassAddr:   offerAnswer.RosenpassAddr,
-		MlkemPayload:    offerAnswer.MlkemPayload,
-		MlkemPort:       int(offerAnswer.MlkemPort),
 		RelaySrvAddress: offerAnswer.RelaySrvAddress,
 		RelaySrvIP:      offerAnswer.RelaySrvIP,
 		SessionID:       sessionIDBytes,

client/internal/pqkem/bench_test.go

@@ -1,71 +0,0 @@
-package pqkem
-
-import (
-	"crypto/ecdh"
-	"crypto/mlkem"
-	"crypto/rand"
-	"testing"
-)
-
-func BenchmarkX25519Keygen(b *testing.B) {
-	c := ecdh.X25519()
-	b.ResetTimer()
-	for i := 0; i < b.N; i++ {
-		if _, err := c.GenerateKey(rand.Reader); err != nil {
-			b.Fatal(err)
-		}
-	}
-}
-
-func BenchmarkX25519ECDH(b *testing.B) {
-	c := ecdh.X25519()
-	a, err := c.GenerateKey(rand.Reader)
-	if err != nil {
-		b.Fatal(err)
-	}
-	p, err := c.GenerateKey(rand.Reader)
-	if err != nil {
-		b.Fatal(err)
-	}
-	pub := p.PublicKey()
-	b.ResetTimer()
-	for i := 0; i < b.N; i++ {
-		if _, err := a.ECDH(pub); err != nil {
-			b.Fatal(err)
-		}
-	}
-}
-
-func BenchmarkMLKEMKeygen(b *testing.B) {
-	for i := 0; i < b.N; i++ {
-		if _, err := mlkem.GenerateKey768(); err != nil {
-			b.Fatal(err)
-		}
-	}
-}
-
-func BenchmarkMLKEMEncaps(b *testing.B) {
-	dk, err := mlkem.GenerateKey768()
-	if err != nil {
-		b.Fatal(err)
-	}
-	ek := dk.EncapsulationKey()
-	b.ResetTimer()
-	for i := 0; i < b.N; i++ {
-		_, _ = ek.Encapsulate()
-	}
-}
-
-func BenchmarkMLKEMDecaps(b *testing.B) {
-	dk, err := mlkem.GenerateKey768()
-	if err != nil {
-		b.Fatal(err)
-	}
-	_, ct := dk.EncapsulationKey().Encapsulate()
-	b.ResetTimer()
-	for i := 0; i < b.N; i++ {
-		if _, err := dk.Decapsulate(ct); err != nil {
-			b.Fatal(err)
-		}
-	}
-}

client/internal/pqkem/callbacks.go

@@ -1,18 +0,0 @@
-package pqkem
-
-// CallbackHandler is implemented by the host and invoked by the library. The
-// library only reports events; the host owns the reaction. Keeping this an
-// interface — rather than touching the transport or keying directly — is what lets
-// the KEM code be extracted as a standalone library.
-type CallbackHandler interface {
-	// OnNewPSKReady fires when a fresh post-quantum PSK has been derived for a peer
-	// and must be programmed into the consumer's secure channel. It is invoked at
-	// the commit point of each side: the initiator on receiving the answer, the
-	// responder on receiving the confirm.
-	OnNewPSKReady(remoteID RemoteID, psk PSK) error
-
-	// OnRekeyFailed fires when an exchange fails to converge within the allotted
-	// time. The host should tear the peer connection down so it re-establishes, and
-	// log a WARN. The library reports the event; it does not dictate the reaction.
-	OnRekeyFailed(remoteID RemoteID) error
-}

client/internal/pqkem/capability_test.go

@@ -1,66 +0,0 @@
-package pqkem
-
-import (
-	"testing"
-
-	"github.com/stretchr/testify/assert"
-	"github.com/stretchr/testify/require"
-)
-
-// TestManager_NonCapablePeerNotOffered: a peer known not to run the KEM (it advertised
-// no PQ port over signalling) is never offered an exchange, and no failure is raised —
-// this is what stops the reoffer storm against non-PQ peers (e.g. Rosenpass peers).
-func TestManager_NonCapablePeerNotOffered(t *testing.T) {
-	wg := newFakeWG()
-	d := NewManager("bbbb", wg, nil) // initiator vs "aaaa"
-	d.Start(&loopback{ep: epB, sw: newSwitch()})
-	defer d.Stop()
-
-	d.MarkNonCapable("aaaa")
-
-	offer, err := d.SignalOffer("aaaa")
-	require.NoError(t, err)
-	assert.Nil(t, offer, "a non-capable peer must not be offered a KEM exchange")
-	assert.Empty(t, wg.failed, "a non-capable peer must not raise a rekey failure")
-}
-
-// TestManager_MarkNonCapableCancelsInFlight: if we start an exchange with a peer whose
-// capability is not yet known and then learn it does not run the KEM, the in-flight
-// exchange is cancelled and no further offer is produced (no timeout -> no failure).
-func TestManager_MarkNonCapableCancelsInFlight(t *testing.T) {
-	wg := newFakeWG()
-	d := NewManager("bbbb", wg, nil)
-	d.Start(&loopback{ep: epB, sw: newSwitch()})
-	defer d.Stop()
-
-	// Capability unknown -> the bootstrap offer goes out optimistically.
-	offer, err := d.SignalOffer("aaaa")
-	require.NoError(t, err)
-	require.NotNil(t, offer)
-
-	// Now we learn the peer is non-PQ: the exchange must be dropped.
-	d.MarkNonCapable("aaaa")
-
-	next, err := d.SignalOffer("aaaa")
-	require.NoError(t, err)
-	assert.Nil(t, next, "after learning non-capability the peer is no longer offered")
-	assert.Empty(t, wg.failed, "cancelling an in-flight exchange must not raise a failure")
-}
-
-// TestManager_EstablishedPeerNotDowngraded: a stray zero-port observation must not tear
-// down a peer we already have a working PQ session with.
-func TestManager_EstablishedPeerNotDowngraded(t *testing.T) {
-	dA, dB, _, wgB, _ := pair(t)
-	defer dA.Stop()
-	defer dB.Stop()
-
-	bootstrap(t, dA, dB)
-	require.NotEqual(t, PSK{}, wgB.psk("aaaa"), "established a PSK")
-
-	dB.MarkNonCapable("aaaa") // stray zero after establishment
-
-	// The peer keeps its derived PSK (MarkNonCapable is a no-op once established).
-	psk, ok := dB.PSK("aaaa")
-	assert.True(t, ok, "an established peer must keep its PSK despite a stray zero")
-	assert.NotEqual(t, PSK{}, psk)
-}

client/internal/pqkem/concurrency_test.go

@@ -1,77 +0,0 @@
-package pqkem
-
-import (
-	"sync"
-	"testing"
-	"time"
-
-	"github.com/stretchr/testify/require"
-)
-
-// TestConcurrency_RecoversViaResignalAfterDataPathBreak exercises the A-light recovery:
-// a data-path rotation can no longer converge (OnRekeyFailed), and re-bootstrapping over
-// signalling resyncs both peers on a fresh PSK — even while the data path stays broken,
-// since the signal channel is independent of it.
-func TestConcurrency_RecoversViaResignalAfterDataPathBreak(t *testing.T) {
-	dA, dB, wgA, wgB, lbB := pair(t)
-	defer dA.Stop()
-	defer dB.Stop()
-
-	// Tighten B's timings and make a single rotation miss raise OnRekeyFailed. Set
-	// before any exchange loop spawns (the loop reads these fields).
-	dB.retryInterval = 5 * time.Millisecond
-	dB.maxRetries = 2
-	dB.maxRekeyFailures = 1
-
-	bootstrap(t, dA, dB)
-	dA.OnDataPathRekeyed("bbbb", 0)
-	dB.OnDataPathRekeyed("aaaa", 0)
-	psk1 := wgB.psk("aaaa")
-	require.NotEqual(t, PSK{}, psk1)
-	require.Equal(t, psk1, wgA.psk("bbbb"), "converged on the same PSK after bootstrap+rotation")
-
-	// Data path breaks: the rotation can no longer converge -> OnRekeyFailed.
-	lbB.drop.Store(true)
-	_, err := dB.startExchangeTest("aaaa", false, ExchangeID{})
-	require.NoError(t, err)
-	require.Eventually(t, func() bool { return failedCount(wgB) >= 1 }, time.Second, 5*time.Millisecond)
-
-	// Recovery: re-bootstrap over signalling with the data path STILL broken. It must
-	// still converge (signal is independent of the data path) on a fresh PSK.
-	bootstrap(t, dA, dB)
-	psk2 := wgB.psk("aaaa")
-	require.NotEqual(t, psk1, psk2, "recovery derived a fresh PSK")
-	require.Equal(t, psk2, wgA.psk("bbbb"), "both sides resync after recovery")
-}
-
-// TestConcurrency_ConcurrentRekeysNoRace hammers both managers with concurrent rotation
-// clocks from many goroutines. Its primary job (with -race) is to prove the single-lock
-// state machine has no data races or deadlocks under contention; a final deterministic
-// bootstrap then asserts there is no split-brain (both sides on the same PSK).
-func TestConcurrency_ConcurrentRekeysNoRace(t *testing.T) {
-	dA, dB, wgA, wgB, _ := pair(t)
-	defer dA.Stop()
-	defer dB.Stop()
-
-	bootstrap(t, dA, dB)
-
-	var wg sync.WaitGroup
-	for g := 0; g < 8; g++ {
-		wg.Add(1)
-		go func() {
-			defer wg.Done()
-			for i := 0; i < 50; i++ {
-				dB.OnDataPathRekeyed("aaaa", 0) // initiator chains a rotation
-				dA.OnDataPathRekeyed("bbbb", 0) // responder side is a no-op, still stresses the lock
-			}
-		}()
-	}
-	wg.Wait()
-
-	// The storm may leave an exchange mid-flight (concurrent cancellation). Force a
-	// clean convergence over signalling, then assert no split-brain.
-	bootstrap(t, dA, dB)
-	a, b := wgA.psk("bbbb"), wgB.psk("aaaa")
-	require.NotEqual(t, PSK{}, b)
-	require.Equal(t, a, b, "both sides converge on the same PSK, no split-brain")
-}

client/internal/pqkem/convergence.go

@@ -1,291 +0,0 @@
-package pqkem
-
-import (
-	"context"
-	"crypto/sha256"
-	"encoding/hex"
-	"fmt"
-	"time"
-)
-
-// idHex renders an exchange ID for logs.
-func idHex(id ExchangeID) string { return hex.EncodeToString(id[:]) }
-
-// pskFingerprint is a short, non-secret digest of a derived PSK: identical on both
-// peers iff they derived the same key. Logged instead of the raw PSK so debug logs
-// never carry the actual WireGuard preshared key.
-func pskFingerprint(psk PSK) string {
-	sum := sha256.Sum256(psk[:])
-	return hex.EncodeToString(sum[:8])
-}
-
-// startExchange creates a fresh initiator exchange (acknowledging ackID, zero for a
-// bootstrap) and returns the framed offer for the caller to send — pushed over the
-// data path for a chained rekey, or handed to the host for signalling when viaSignal
-// is set. Any previous in-flight exchange for the peer is cancelled.
-// startExchangeLocked must be called with m.mu held: the caller's idempotency check and
-// the exchange install stay under one lock acquisition so two concurrent starts for the
-// same peer cannot both create an exchange. It also refuses to start (and to Add to the
-// wait group) once the manager is stopping, so it never races Manager.Stop's Wait.
-func (m *Manager) startExchangeLocked(remoteID RemoteID, viaSignal bool, ackID ExchangeID) ([]byte, error) {
-	if m.rootCtx.Err() != nil {
-		return nil, fmt.Errorf("manager stopping")
-	}
-	init, err := NewInitiator()
-	if err != nil {
-		return nil, err
-	}
-	id, err := newExchangeID()
-	if err != nil {
-		return nil, err
-	}
-	raw, err := (&OfferMsg{ExchangeID: id, AckID: ackID, KEMOffer: init.Offer()}).Encode()
-	if err != nil {
-		return nil, err
-	}
-
-	ctx, cancel := context.WithCancel(m.rootCtx)
-	if old := m.exchanges[remoteID]; old != nil && old.cancel != nil {
-		old.cancel()
-	}
-	m.exchanges[remoteID] = &exchangeCtl{
-		id:        id,
-		state:     stateAwaitingAnswer,
-		startedAt: time.Now(),
-		cancel:    cancel,
-		lastSent:  raw,
-		initiator: init,
-		viaSignal: viaSignal,
-	}
-
-	m.wait.Add(1)
-	go m.initiatorLoop(ctx, remoteID, id)
-
-	via := "data-path"
-	if viaSignal {
-		via = "signal"
-	}
-	m.trace("pqkem: offer sent", "peer", remoteID, "exchange", idHex(id), "acks", idHex(ackID), "via", via)
-	return raw, nil
-}
-
-// processOffer (responder) first acknowledges the previous exchange the offer names
-// (that offer riding the data path under the freshly adopted key proves it worked),
-// then derives the PSK for the new offer, commits it optimistically, and returns the
-// framed answer. A duplicate offer returns the cached answer without re-deriving.
-func (m *Manager) processOffer(remoteID RemoteID, o *OfferMsg) ([]byte, error) {
-	m.trace("pqkem: offer received", "peer", remoteID, "exchange", idHex(o.ExchangeID), "acks", idHex(o.AckID))
-
-	if o.AckID != (ExchangeID{}) {
-		m.ackConverged(remoteID, o.AckID)
-	}
-
-	m.mu.Lock()
-	if ex := m.exchanges[remoteID]; ex != nil && ex.id == o.ExchangeID {
-		state, last := ex.state, ex.lastSent
-		m.mu.Unlock()
-		if state == stateReserved {
-			return nil, nil
-		}
-		m.trace("pqkem: duplicate offer, resending cached answer", "peer", remoteID, "exchange", idHex(o.ExchangeID))
-		return last, nil
-	}
-	// Reserve the slot so a concurrent duplicate offer bails.
-	m.exchanges[remoteID] = &exchangeCtl{id: o.ExchangeID, state: stateReserved, startedAt: time.Now()}
-	m.mu.Unlock()
-
-	answerBytes, psk, err := Respond(o.KEMOffer, m.binding(remoteID))
-	if err != nil {
-		return nil, err
-	}
-	raw, err := (&AnswerMsg{ExchangeID: o.ExchangeID, KEMAnswer: answerBytes}).Encode()
-	if err != nil {
-		return nil, err
-	}
-
-	m.mu.Lock()
-	ex := m.exchanges[remoteID]
-	if ex == nil || ex.id != o.ExchangeID {
-		m.mu.Unlock()
-		m.trace("pqkem: exchange superseded during respond, dropping answer", "peer", remoteID, "exchange", idHex(o.ExchangeID))
-		return nil, nil
-	}
-	ex.state = stateAwaitingAck
-	ex.lastSent = raw
-	ex.pendingPSK = psk
-	m.psks[remoteID] = psk
-	m.capable[remoteID] = true // a real KEM offer proves the peer runs the exchange
-	m.mu.Unlock()
-
-	m.trace("pqkem: new PSK derived", "peer", remoteID, "exchange", idHex(o.ExchangeID), "role", "responder", "psk_fp", pskFingerprint(psk))
-
-	// Commit optimistically so our data path can rekey to the new PSK.
-	if err := m.cbHandler.OnNewPSKReady(remoteID, psk); err != nil {
-		return nil, err
-	}
-	m.trace("pqkem: answer sent", "peer", remoteID, "exchange", idHex(o.ExchangeID))
-	return raw, nil
-}
-
-// processAnswer (initiator) derives and commits the PSK and parks in
-// stateAwaitingRekey; the next offer (chained from OnDataPathRekeyed) will acknowledge
-// this exchange. Only valid in stateAwaitingAnswer; advancing the state under the
-// lock makes a concurrent/duplicate answer bail.
-func (m *Manager) processAnswer(remoteID RemoteID, a *AnswerMsg) error {
-	m.mu.Lock()
-	ex := m.exchanges[remoteID]
-	if ex == nil || ex.id != a.ExchangeID || ex.state != stateAwaitingAnswer {
-		haveID := "none"
-		if ex != nil {
-			haveID = idHex(ex.id)
-		}
-		m.mu.Unlock()
-		m.trace("pqkem: unexpected answer dropped (inconsistency)", "peer", remoteID, "answer_for", idHex(a.ExchangeID), "have_exchange", haveID)
-		return nil
-	}
-	ex.state = stateAwaitingRekey
-	init := ex.initiator
-	ex.initiator = nil
-	m.mu.Unlock()
-
-	m.trace("pqkem: answer received", "peer", remoteID, "exchange", idHex(a.ExchangeID))
-
-	psk, err := init.Finish(a.KEMAnswer, m.binding(remoteID))
-	if err != nil {
-		// The state already advanced to stateAwaitingRekey and the initiator was cleared,
-		// so initiatorLoop would exit its default branch without registering a failure —
-		// leaving the peer desynced (the responder committed its PSK in processOffer).
-		// Drop the exchange and raise the failure so recovery re-bootstraps.
-		m.mu.Lock()
-		if cur := m.exchanges[remoteID]; cur != nil && cur.id == a.ExchangeID {
-			delete(m.exchanges, remoteID)
-		}
-		initial := !m.established[remoteID]
-		fail := m.registerFailureLocked(remoteID)
-		m.mu.Unlock()
-		m.raiseFailure(remoteID, fail, initial)
-		return err
-	}
-
-	// The initiator has converged: the responder must have derived the key to answer.
-	m.mu.Lock()
-	m.established[remoteID] = true
-	m.failures[remoteID] = 0
-	m.psks[remoteID] = psk
-	m.capable[remoteID] = true // a real KEM answer proves the peer runs the exchange
-	m.mu.Unlock()
-
-	m.trace("pqkem: new PSK derived", "peer", remoteID, "exchange", idHex(a.ExchangeID), "role", "initiator", "psk_fp", pskFingerprint(psk))
-
-	return m.cbHandler.OnNewPSKReady(remoteID, psk)
-}
-
-// ackConverged (responder) records convergence of the exchange named by ackID: a
-// later offer acknowledging it proves both sides operate on that exchange's key. Only
-// acts on a matching stateAwaitingAck exchange; anything else is ignored.
-func (m *Manager) ackConverged(remoteID RemoteID, ackID ExchangeID) {
-	m.mu.Lock()
-	ex := m.exchanges[remoteID]
-	if ex == nil || ex.id != ackID || ex.state != stateAwaitingAck {
-		m.mu.Unlock()
-		m.trace("pqkem: ack for unknown/mismatched exchange, ignored (inconsistency)", "peer", remoteID, "acks", idHex(ackID))
-		return
-	}
-	delete(m.exchanges, remoteID)
-	m.established[remoteID] = true
-	m.failures[remoteID] = 0
-	_ = time.Since(ex.startedAt) // convergence latency (metrics hook, later step)
-	m.mu.Unlock()
-
-	m.trace("pqkem: previous exchange confirmed by ack", "peer", remoteID, "exchange", idHex(ackID))
-}
-
-// initiatorLoop enforces the offer->answer convergence deadline and retransmits the
-// initiator's outstanding data-path offer while awaiting the answer (a
-// signalling-bootstrapped offer is retransmitted by the host, so it is not resent
-// here). Exhausting the deadline before the answer arrives is a failure. Once the
-// answer is in (state past awaitingAnswer) the loop exits: the next rotation is driven
-// by OnDataPathRekeyed, and the idle wait for it has no deadline.
-func (m *Manager) initiatorLoop(ctx context.Context, remoteID RemoteID, id ExchangeID) {
-	defer m.wait.Done()
-	t := time.NewTicker(m.retryInterval)
-	defer t.Stop()
-
-	attempts := 0
-	for {
-		select {
-		case <-ctx.Done():
-			return
-		case <-t.C:
-			m.mu.Lock()
-			ex := m.exchanges[remoteID]
-			if ex == nil || ex.id != id {
-				m.mu.Unlock()
-				return
-			}
-
-			switch ex.state {
-			case stateAwaitingAnswer:
-				if attempts >= m.maxRetries {
-					delete(m.exchanges, remoteID)
-					initial := !m.established[remoteID]
-					fail := m.registerFailureLocked(remoteID)
-					m.mu.Unlock()
-					m.raiseFailure(remoteID, fail, initial)
-					return
-				}
-				viaSignal := ex.viaSignal
-				msg := ex.lastSent
-				attempts++
-				m.mu.Unlock()
-				if !viaSignal {
-					if err := m.pushDataPath(remoteID, msg); err != nil {
-						m.logger.Warn("pqkem: offer retransmit failed", "peer", remoteID, "err", err)
-					}
-				}
-
-			default:
-				// Past awaiting the answer (converged) or superseded: the loop's job
-				// is done. The next rotation is driven externally by OnDataPathRekeyed,
-				// so there is no deadline while idle-waiting for it (that wait can be
-				// as long as the transport's natural rekey interval).
-				m.mu.Unlock()
-				return
-			}
-		}
-	}
-}
-
-// registerFailureLocked applies policy B and reports whether OnRekeyFailed is due:
-// an initial exchange (peer never established) fails immediately; a rekey tolerates
-// up to maxRekeyFailures consecutive misses (we stay on the still-valid previous
-// PSK) before failing. Assumes m.mu is held.
-func (m *Manager) registerFailureLocked(remoteID RemoteID) bool {
-	if !m.established[remoteID] {
-		return true
-	}
-	m.failures[remoteID]++
-	if m.failures[remoteID] >= m.maxRekeyFailures {
-		m.failures[remoteID] = 0
-		return true
-	}
-	return false
-}
-
-// raiseFailure reports a convergence failure. initial distinguishes a never-established
-// peer (bootstrap failed → no PQ PSK at all; in strict mode the peer stays blocked =
-// "stuck") from a rekey failure (a previous PSK is still in force and traffic continues).
-func (m *Manager) raiseFailure(remoteID RemoteID, fail, initial bool) {
-	if !fail {
-		m.logger.Warn("pqkem: rekey attempt timed out, will retry next cycle", "peer", remoteID)
-		return
-	}
-	if initial {
-		m.logger.Warn("pqkem: initial exchange failed — no PQ PSK established for peer (strict mode keeps the peer blocked until it converges)", "peer", remoteID)
-	} else {
-		m.logger.Warn("pqkem: rekey failed after retries — staying on the previous PSK", "peer", remoteID)
-	}
-	if err := m.cbHandler.OnRekeyFailed(remoteID); err != nil {
-		m.logger.Error("pqkem: OnRekeyFailed handler error", "peer", remoteID, "err", err)
-	}
-}

client/internal/pqkem/convergence_test.go

@@ -1,75 +0,0 @@
-package pqkem
-
-import (
-	"net/netip"
-	"testing"
-	"time"
-
-	"github.com/stretchr/testify/assert"
-	"github.com/stretchr/testify/require"
-)
-
-// dropTransport is a pqkem.Transport that silently discards everything.
-type dropTransport struct{}
-
-func (dropTransport) Send(netip.AddrPort, []byte) error { return nil }
-func (dropTransport) LocalPort() int                    { return 0 }
-func (dropTransport) Run(func(netip.AddrPort, []byte))  {}
-func (dropTransport) Close() error                      { return nil }
-
-func failedCount(f *fakeWG) int {
-	f.mu.Lock()
-	defer f.mu.Unlock()
-	return len(f.failed)
-}
-
-func TestManager_InitialTimeoutFailsImmediately(t *testing.T) {
-	wg := newFakeWG()
-	d := NewManager("bbbb", wg, nil) // bbbb > aaaa -> initiator
-	d.Start(dropTransport{})
-	d.retryInterval = 5 * time.Millisecond
-	d.maxRetries = 3
-	defer d.Stop()
-
-	// Bootstrap offer is produced for signalling; no answer ever comes back -> the
-	// initial exchange fails fast.
-	offer, err := d.SignalOffer("aaaa")
-	require.NoError(t, err)
-	require.NotNil(t, offer)
-
-	assert.Eventually(t, func() bool { return failedCount(wg) == 1 }, time.Second, 5*time.Millisecond)
-}
-
-func TestManager_RekeyToleratesKFailures(t *testing.T) {
-	dA, dB, _, wgB, lbB := pair(t)
-	defer dA.Stop()
-	defer dB.Stop()
-
-	// Tighten B's timings before any exchange loop spawns (the loop reads these
-	// fields, so writing them after a loop is running would race).
-	dB.retryInterval = 5 * time.Millisecond
-	dB.maxRetries = 2
-
-	// Establish: bootstrap + data-path-rekeyed so B becomes established and its data
-	// path is usable.
-	bootstrap(t, dA, dB)
-	dA.OnDataPathRekeyed("bbbb", 0)
-	dB.OnDataPathRekeyed("aaaa", 0)
-	require.NotEqual(t, PSK{}, wgB.psk("aaaa"))
-
-	// Drop B's outbound so rekeys can no longer converge.
-	lbB.drop.Store(true)
-
-	// K-1 data-path rekeys must NOT raise OnRekeyFailed.
-	for i := 0; i < DefaultMaxRekeyFailures-1; i++ {
-		_, err := dB.startExchangeTest("aaaa", false, ExchangeID{})
-		require.NoError(t, err)
-		time.Sleep(50 * time.Millisecond)
-	}
-	assert.Equal(t, 0, failedCount(wgB), "no failure before K attempts")
-
-	// The K-th failure raises it once.
-	_, err := dB.startExchangeTest("aaaa", false, ExchangeID{})
-	require.NoError(t, err)
-	assert.Eventually(t, func() bool { return failedCount(wgB) == 1 }, time.Second, 5*time.Millisecond)
-}

client/internal/pqkem/env.go

@@ -1,138 +0,0 @@
-package pqkem
-
-import (
-	"context"
-	"log/slog"
-	"os"
-	"strconv"
-	"strings"
-
-	log "github.com/sirupsen/logrus"
-)
-
-// EnvEnabled is the environment variable that turns the ML-KEM post-quantum
-// exchange on for this client. Accepts on/off aliases plus anything
-// strconv.ParseBool understands (true/false/1/0).
-const EnvEnabled = "NB_ENABLE_PQ_MLKEM"
-
-// Enabled reports whether the ML-KEM PQ exchange is enabled via the environment.
-// An empty or unrecognized value is treated as disabled.
-func Enabled() bool {
-	raw := strings.ToLower(strings.TrimSpace(os.Getenv(EnvEnabled)))
-	switch raw {
-	case "":
-		return false
-	case "on":
-		return true
-	case "off":
-		return false
-	}
-	enabled, err := strconv.ParseBool(raw)
-	if err != nil {
-		log.Warnf("failed to parse %s value %q: %v", EnvEnabled, raw, err)
-		return false
-	}
-	return enabled
-}
-
-// EnvStrict enables strict (fail-closed) mode: block peer traffic until the ML-KEM
-// PSK has been established, instead of the default opportunistic behaviour that lets
-// the tunnel come up classically and upgrades to PQ once the exchange converges.
-const EnvStrict = "NB_PQ_MLKEM_STRICT"
-
-// Strict reports whether strict (fail-closed) mode is enabled via the environment.
-// An empty or unrecognized value is treated as disabled (opportunistic).
-func Strict() bool {
-	switch strings.ToLower(strings.TrimSpace(os.Getenv(EnvStrict))) {
-	case "on":
-		return true
-	case "", "off":
-		return false
-	}
-	enabled, err := strconv.ParseBool(strings.TrimSpace(os.Getenv(EnvStrict)))
-	if err != nil {
-		log.Warnf("failed to parse %s value %q: %v", EnvStrict, os.Getenv(EnvStrict), err)
-		return false
-	}
-	return enabled
-}
-
-// EnvLogLevel overrides the ML-KEM manager's slog level (trace/debug/info/warn/error).
-// Defaults to info. The verbose per-exchange lifecycle logs are emitted at trace.
-const EnvLogLevel = "NB_PQ_MLKEM_LOG_LEVEL"
-
-// LevelTrace is a custom slog level below Debug for the verbose per-exchange lifecycle
-// logs, so they stay off unless NB_PQ_MLKEM_LOG_LEVEL=trace (and the daemon log level
-// is trace, since the records are forwarded to logrus).
-const LevelTrace = slog.LevelDebug - 4
-
-// NewLogger builds the slog logger for the ML-KEM manager. It forwards records to
-// logrus so PQ logs land in the same sink as the rest of the daemon (console +
-// client.log) rather than stdout. Verbosity is gated by EnvLogLevel.
-func NewLogger() *slog.Logger {
-	return slog.New(slogToLogrus{})
-}
-
-func logLevel() slog.Level {
-	switch strings.ToLower(strings.TrimSpace(os.Getenv(EnvLogLevel))) {
-	case "trace":
-		return LevelTrace
-	case "debug":
-		return slog.LevelDebug
-	case "warn":
-		return slog.LevelWarn
-	case "error":
-		return slog.LevelError
-	default:
-		return slog.LevelInfo
-	}
-}
-
-// slogToLogrus is a slog.Handler that forwards records to logrus, so the ML-KEM
-// manager's logs go wherever the daemon's logrus is configured (console + client.log)
-// instead of stdout. Verbosity is gated by EnvLogLevel via logLevel().
-type slogToLogrus struct {
-	fields log.Fields
-}
-
-func (h slogToLogrus) Enabled(_ context.Context, level slog.Level) bool {
-	return level >= logLevel()
-}
-
-func (h slogToLogrus) Handle(_ context.Context, r slog.Record) error {
-	fields := make(log.Fields, len(h.fields)+r.NumAttrs())
-	for k, v := range h.fields {
-		fields[k] = v
-	}
-	r.Attrs(func(a slog.Attr) bool {
-		fields[a.Key] = a.Value.Any()
-		return true
-	})
-	entry := log.WithFields(fields)
-	switch {
-	case r.Level >= slog.LevelError:
-		entry.Error(r.Message)
-	case r.Level >= slog.LevelWarn:
-		entry.Warn(r.Message)
-	case r.Level >= slog.LevelInfo:
-		entry.Info(r.Message)
-	case r.Level >= slog.LevelDebug:
-		entry.Debug(r.Message)
-	default:
-		entry.Trace(r.Message)
-	}
-	return nil
-}
-
-func (h slogToLogrus) WithAttrs(attrs []slog.Attr) slog.Handler {
-	fields := make(log.Fields, len(h.fields)+len(attrs))
-	for k, v := range h.fields {
-		fields[k] = v
-	}
-	for _, a := range attrs {
-		fields[a.Key] = a.Value.Any()
-	}
-	return slogToLogrus{fields: fields}
-}
-
-func (h slogToLogrus) WithGroup(_ string) slog.Handler { return h }

client/internal/pqkem/kem.go

@@ -1,183 +0,0 @@
-// Package pqkem is a spike (NET-1406) for a post-quantum pre-shared-key exchange
-// that could replace Rosenpass. It performs an X25519MLKEM768 hybrid key
-// encapsulation and derives a 32-byte pre-shared key (PSK).
-//
-// The exchange is a single round trip designed to ride the (already
-// authenticated) Signal offer/answer channel:
-//
-//	initiator --Offer(1216B)--> responder
-//	initiator <--Answer(1120B)-- responder
-//
-// Both sides then hold the same PSK, which is bound to the two peers' identities
-// (their peer identity keys) so the derived key cannot be transplanted
-// to a different peer pair even if the transport authentication were bypassed.
-//
-// Combiner note: this follows draft-ietf-tls-ecdhe-mlkem for X25519MLKEM768 — on
-// the wire ML-KEM ‖ X25519 (the draft deliberately reversed the share order for
-// this group), and ML-KEM_ss ‖ X25519_ss as the KDF input. The PSK is derived with
-// HKDF-SHA256 over that hybrid secret, salted with a domain-separation label and
-// bound (via the HKDF info) to the full transcript and the canonicalised peer
-// identities.
-package pqkem
-
-import (
-	"crypto/ecdh"
-	"crypto/hkdf"
-	"crypto/mlkem"
-	"crypto/rand"
-	"crypto/sha256"
-	"fmt"
-)
-
-const (
-	// OfferSize is the initiator message: ML-KEM-768 encapsulation key ‖ X25519 public key
-	// (share order per draft-ietf-tls-ecdhe-mlkem for X25519MLKEM768).
-	OfferSize = mlkem.EncapsulationKeySize768 + 32 // 1216
-	// AnswerSize is the responder message: ML-KEM-768 ciphertext ‖ X25519 public key.
-	AnswerSize = mlkem.CiphertextSize768 + 32 // 1120
-
-	pskLabel = "netbird-pq-psk-v1"
-)
-
-// PSK is the 32-byte derived pre-shared key handed to the consumer to key its channel.
-type PSK [32]byte
-
-// Binding identifies the peer pair the PSK is derived for. Callers set both
-// peer identity keys; the order does not matter (it is canonicalised).
-type Binding struct {
-	LocalID  []byte
-	RemoteID []byte
-}
-
-// Initiator holds the ephemeral secrets between Offer and Finish.
-type Initiator struct {
-	x25519  *ecdh.PrivateKey
-	mlkemDK *mlkem.DecapsulationKey768
-	offer   []byte
-}
-
-// NewInitiator generates the ephemeral X25519 + ML-KEM-768 keypairs.
-func NewInitiator() (*Initiator, error) {
-	x, err := ecdh.X25519().GenerateKey(rand.Reader)
-	if err != nil {
-		return nil, fmt.Errorf("x25519 keygen: %w", err)
-	}
-	dk, err := mlkem.GenerateKey768()
-	if err != nil {
-		return nil, fmt.Errorf("ml-kem keygen: %w", err)
-	}
-
-	offer := make([]byte, 0, OfferSize)
-	offer = append(offer, dk.EncapsulationKey().Bytes()...)
-	offer = append(offer, x.PublicKey().Bytes()...)
-
-	return &Initiator{x25519: x, mlkemDK: dk, offer: offer}, nil
-}
-
-// Offer returns the initiator message to send over Signal.
-func (i *Initiator) Offer() []byte {
-	return i.offer
-}
-
-// Finish consumes the responder's answer and derives the PSK.
-func (i *Initiator) Finish(answer []byte, b Binding) (PSK, error) {
-	if len(answer) != AnswerSize {
-		return PSK{}, fmt.Errorf("answer: got %d bytes, want %d", len(answer), AnswerSize)
-	}
-	ct := answer[:mlkem.CiphertextSize768]
-	peerX := answer[mlkem.CiphertextSize768:]
-
-	ssMLKEM, err := i.mlkemDK.Decapsulate(ct)
-	if err != nil {
-		return PSK{}, fmt.Errorf("ml-kem decapsulate: %w", err)
-	}
-	pub, err := ecdh.X25519().NewPublicKey(peerX)
-	if err != nil {
-		return PSK{}, fmt.Errorf("parse peer x25519: %w", err)
-	}
-	ssX, err := i.x25519.ECDH(pub)
-	if err != nil {
-		return PSK{}, fmt.Errorf("x25519 ecdh: %w", err)
-	}
-
-	return derivePSK(ssMLKEM, ssX, i.offer, answer, b)
-}
-
-// Respond consumes an initiator offer, produces the answer, and derives the PSK.
-func Respond(offer []byte, b Binding) (answer []byte, psk PSK, err error) {
-	if len(offer) != OfferSize {
-		return nil, PSK{}, fmt.Errorf("offer: got %d bytes, want %d", len(offer), OfferSize)
-	}
-	peerEK := offer[:mlkem.EncapsulationKeySize768]
-	peerX := offer[mlkem.EncapsulationKeySize768:]
-
-	ek, err := mlkem.NewEncapsulationKey768(peerEK)
-	if err != nil {
-		return nil, PSK{}, fmt.Errorf("parse peer ml-kem key: %w", err)
-	}
-	ssMLKEM, ct := ek.Encapsulate()
-
-	x, err := ecdh.X25519().GenerateKey(rand.Reader)
-	if err != nil {
-		return nil, PSK{}, fmt.Errorf("x25519 keygen: %w", err)
-	}
-	pub, err := ecdh.X25519().NewPublicKey(peerX)
-	if err != nil {
-		return nil, PSK{}, fmt.Errorf("parse peer x25519: %w", err)
-	}
-	ssX, err := x.ECDH(pub)
-	if err != nil {
-		return nil, PSK{}, fmt.Errorf("x25519 ecdh: %w", err)
-	}
-
-	answer = make([]byte, 0, AnswerSize)
-	answer = append(answer, ct...)
-	answer = append(answer, x.PublicKey().Bytes()...)
-
-	// derivePSK uses the same argument order on both sides; the responder's local
-	// binding is the mirror of the initiator's, canonicalised inside derivePSK.
-	psk, err = derivePSK(ssMLKEM, ssX, offer, answer, b)
-	if err != nil {
-		return nil, PSK{}, err
-	}
-	return answer, psk, nil
-}
-
-// derivePSK runs HKDF-SHA256 over the hybrid shared secret (ML-KEM_ss ‖ X25519_ss,
-// per draft-ietf-tls-ecdhe-mlkem), salted with the domain-separation label, and binds
-// the result — via the HKDF info — to the full transcript (offer ‖ answer) and the
-// canonicalised peer identities, so the PSK cannot be transplanted to another peer
-// pair or a different exchange.
-func derivePSK(ssMLKEM, ssX, offer, answer []byte, b Binding) (PSK, error) {
-	// A PSK not bound to both peer identities could be transplanted to a different peer
-	// pair, so refuse to derive one from an empty binding.
-	if len(b.LocalID) == 0 || len(b.RemoteID) == 0 {
-		return PSK{}, fmt.Errorf("empty peer identity binding")
-	}
-	lo, hi := canonicalPair(b.LocalID, b.RemoteID)
-
-	ikm := make([]byte, 0, len(ssMLKEM)+len(ssX))
-	ikm = append(ikm, ssMLKEM...)
-	ikm = append(ikm, ssX...)
-
-	info := make([]byte, 0, len(offer)+len(answer)+len(lo)+len(hi))
-	info = append(info, offer...)
-	info = append(info, answer...)
-	info = append(info, lo...)
-	info = append(info, hi...)
-
-	var psk PSK
-	key, err := hkdf.Key(sha256.New, ikm, []byte(pskLabel), string(info), len(psk))
-	if err != nil {
-		return PSK{}, fmt.Errorf("hkdf derive psk: %w", err)
-	}
-	copy(psk[:], key)
-	return psk, nil
-}
-
-func canonicalPair(a, b []byte) (lo, hi []byte) {
-	if string(a) <= string(b) {
-		return a, b
-	}
-	return b, a
-}

client/internal/pqkem/kem_framing_test.go

@@ -1,105 +0,0 @@
-package pqkem
-
-import (
-	"crypto/mlkem"
-	"testing"
-
-	"github.com/stretchr/testify/require"
-)
-
-// TestExchange_TamperedCiphertextFailsClosed verifies the core fail-closed
-// property: mutating the ML-KEM ciphertext in the answer does not error (ML-KEM
-// uses implicit rejection — Decapsulate always returns a value) but yields a
-// different shared secret, so the initiator derives a PSK that does NOT match the
-// responder's. A mismatched PSK means WireGuard passes no bytes: tamper => no data.
-func TestExchange_TamperedCiphertextFailsClosed(t *testing.T) {
-	init, err := NewInitiator()
-	require.NoError(t, err)
-
-	answer, pskB, err := Respond(init.Offer(), Binding{LocalID: wgB, RemoteID: wgA})
-	require.NoError(t, err)
-
-	tampered := append([]byte(nil), answer...)
-	tampered[0] ^= 0xff // flip a bit in the ML-KEM ciphertext
-
-	pskA, err := init.Finish(tampered, Binding{LocalID: wgA, RemoteID: wgB})
-	require.NoError(t, err, "implicit rejection: decapsulate still succeeds")
-	require.NotEqual(t, pskB, pskA, "tampered ciphertext must not yield the responder's PSK")
-}
-
-// TestExchange_TamperedX25519ShareDiverges flips a byte in the answer's X25519
-// share: the classical half of the hybrid secret changes, so the derived PSK
-// diverges from the responder's (fail-closed on the ECDH half too).
-func TestExchange_TamperedX25519ShareDiverges(t *testing.T) {
-	init, err := NewInitiator()
-	require.NoError(t, err)
-
-	answer, pskB, err := Respond(init.Offer(), Binding{LocalID: wgB, RemoteID: wgA})
-	require.NoError(t, err)
-
-	tampered := append([]byte(nil), answer...)
-	tampered[mlkem.CiphertextSize768] ^= 0x01 // first byte of the X25519 public key
-
-	pskA, err := init.Finish(tampered, Binding{LocalID: wgA, RemoteID: wgB})
-	// Either the point is rejected (error) or the ECDH differs (different PSK);
-	// in both cases the honest PSK is never reproduced.
-	if err == nil {
-		require.NotEqual(t, pskB, pskA, "tampered X25519 share must not yield the responder's PSK")
-	}
-}
-
-// TestExchange_AllZeroX25519Rejected feeds an all-zero X25519 share (a low-order
-// point) in the answer. The stdlib ECDH must reject it, so Finish errors rather
-// than deriving a PSK from a degenerate secret.
-func TestExchange_AllZeroX25519Rejected(t *testing.T) {
-	init, err := NewInitiator()
-	require.NoError(t, err)
-
-	answer, _, err := Respond(init.Offer(), Binding{LocalID: wgB, RemoteID: wgA})
-	require.NoError(t, err)
-
-	bad := append([]byte(nil), answer...)
-	for i := mlkem.CiphertextSize768; i < len(bad); i++ {
-		bad[i] = 0
-	}
-
-	_, err = init.Finish(bad, Binding{LocalID: wgA, RemoteID: wgB})
-	require.Error(t, err, "all-zero X25519 share (low-order point) must be rejected")
-}
-
-// TestExchange_SizeBoundaries locks the exact-length framing checks: one byte
-// short or long on either message is rejected, not silently truncated/padded.
-func TestExchange_SizeBoundaries(t *testing.T) {
-	init, err := NewInitiator()
-	require.NoError(t, err)
-	offer := init.Offer()
-
-	_, _, err = Respond(offer[:OfferSize-1], Binding{})
-	require.Error(t, err, "offer one byte short")
-	_, _, err = Respond(append(append([]byte(nil), offer...), 0), Binding{})
-	require.Error(t, err, "offer one byte long")
-
-	answer, _, err := Respond(offer, Binding{LocalID: wgB, RemoteID: wgA})
-	require.NoError(t, err)
-
-	_, err = init.Finish(answer[:AnswerSize-1], Binding{LocalID: wgA, RemoteID: wgB})
-	require.Error(t, err, "answer one byte short")
-	_, err = init.Finish(append(append([]byte(nil), answer...), 0), Binding{LocalID: wgA, RemoteID: wgB})
-	require.Error(t, err, "answer one byte long")
-}
-
-// TestExchange_BindingIsSymmetric confirms the canonicalisation: the two peers
-// pass their identities in opposite (Local, Remote) order yet derive the same PSK,
-// so identity binding does not depend on who is initiator vs responder.
-func TestExchange_BindingIsSymmetric(t *testing.T) {
-	init, err := NewInitiator()
-	require.NoError(t, err)
-
-	answer, pskB, err := Respond(init.Offer(), Binding{LocalID: wgB, RemoteID: wgA})
-	require.NoError(t, err)
-
-	pskA, err := init.Finish(answer, Binding{LocalID: wgA, RemoteID: wgB})
-	require.NoError(t, err)
-
-	require.Equal(t, pskB, pskA, "swapped Local/Remote order must canonicalise to the same PSK")
-}

client/internal/pqkem/kem_test.go

@@ -1,106 +0,0 @@
-package pqkem
-
-import (
-	"testing"
-	"time"
-
-	"github.com/stretchr/testify/require"
-)
-
-var (
-	wgA = []byte("peer-A-wireguard-pubkey-32bytes!")
-	wgB = []byte("peer-B-wireguard-pubkey-32bytes!")
-)
-
-func TestExchange_DerivesMatchingPSK(t *testing.T) {
-	init, err := NewInitiator()
-	require.NoError(t, err)
-
-	require.Len(t, init.Offer(), OfferSize)
-
-	answer, pskB, err := Respond(init.Offer(), Binding{LocalID: wgB, RemoteID: wgA})
-	require.NoError(t, err)
-	require.Len(t, answer, AnswerSize)
-
-	pskA, err := init.Finish(answer, Binding{LocalID: wgA, RemoteID: wgB})
-	require.NoError(t, err)
-
-	require.Equal(t, pskB, pskA, "both sides must derive the same PSK")
-	require.NotEqual(t, PSK{}, pskA, "PSK must not be zero")
-}
-
-func TestExchange_PSKBoundToPeerIdentities(t *testing.T) {
-	init, err := NewInitiator()
-	require.NoError(t, err)
-
-	answer, _, err := Respond(init.Offer(), Binding{LocalID: wgB, RemoteID: wgA})
-	require.NoError(t, err)
-
-	// Finish twice over the SAME KEM material (same offer/answer/secrets), changing only
-	// the peer identity binding: the differing PSK is attributable to the binding alone.
-	pskHonest, err := init.Finish(answer, Binding{LocalID: wgA, RemoteID: wgB})
-	require.NoError(t, err)
-
-	wgC := []byte("peer-C-wireguard-pubkey-32bytes!")
-	pskWrong, err := init.Finish(answer, Binding{LocalID: wgA, RemoteID: wgC})
-	require.NoError(t, err)
-
-	require.NotEqual(t, pskHonest, pskWrong, "PSK must be bound to the peer pair")
-}
-
-func TestExchange_RejectsEmptyBinding(t *testing.T) {
-	init, err := NewInitiator()
-	require.NoError(t, err)
-	answer, _, err := Respond(init.Offer(), Binding{LocalID: wgB, RemoteID: wgA})
-	require.NoError(t, err)
-
-	// A PSK not bound to both identities could be transplanted to another peer pair.
-	_, err = init.Finish(answer, Binding{})
-	require.Error(t, err, "empty binding must be rejected")
-	_, err = init.Finish(answer, Binding{LocalID: wgA})
-	require.Error(t, err, "missing RemoteID must be rejected")
-	_, _, err = Respond(init.Offer(), Binding{RemoteID: wgA})
-	require.Error(t, err, "missing LocalID must be rejected")
-}
-
-func TestExchange_RejectsMalformedMessages(t *testing.T) {
-	init, err := NewInitiator()
-	require.NoError(t, err)
-
-	_, _, err = Respond(init.Offer()[:10], Binding{})
-	require.Error(t, err)
-
-	_, err = init.Finish([]byte("too short"), Binding{})
-	require.Error(t, err)
-}
-
-// TestExchange_ReportSizesAndTiming is a spike measurement, not a pass/fail gate.
-// Run with: go test -run TestExchange_ReportSizesAndTiming -v ./client/internal/pqkem/
-func TestExchange_ReportSizesAndTiming(t *testing.T) {
-	const iters = 200
-
-	var tInit, tResp, tFinish time.Duration
-	for i := 0; i < iters; i++ {
-		s0 := time.Now()
-		init, err := NewInitiator()
-		require.NoError(t, err)
-		tInit += time.Since(s0)
-
-		s1 := time.Now()
-		answer, _, err := Respond(init.Offer(), Binding{LocalID: wgB, RemoteID: wgA})
-		require.NoError(t, err)
-		tResp += time.Since(s1)
-
-		s2 := time.Now()
-		_, err = init.Finish(answer, Binding{LocalID: wgA, RemoteID: wgB})
-		require.NoError(t, err)
-		tFinish += time.Since(s2)
-	}
-
-	t.Logf("wire sizes:   offer=%d B   answer=%d B   (Rosenpass static pubkey ~524160 B)", OfferSize, AnswerSize)
-	t.Logf("total on-wire per handshake: %d B  (~%.0fx smaller than RP static key)", OfferSize+AnswerSize, 524160.0/float64(OfferSize+AnswerSize))
-	t.Logf("avg NewInitiator (keygen): %s", tInit/iters)
-	t.Logf("avg Respond (encaps+dh):   %s", tResp/iters)
-	t.Logf("avg Finish (decaps+dh):    %s", tFinish/iters)
-	t.Logf("avg full handshake CPU:    %s", (tInit+tResp+tFinish)/iters)
-}

client/internal/pqkem/manager.go

@@ -1,467 +0,0 @@
-package pqkem
-
-import (
-	"context"
-	"crypto/rand"
-	"fmt"
-	"log/slog"
-	"net/netip"
-	"sync"
-	"time"
-)
-
-const (
-	// DefaultRetryInterval is how often the initiator retransmits its outstanding
-	// data-path offer while awaiting the answer.
-	DefaultRetryInterval = 2 * time.Second
-	// DefaultMaxRetries bounds how many ticks an exchange may run before it is
-	// declared failed. The convergence deadline is thus MaxRetries * RetryInterval.
-	DefaultMaxRetries = 10
-	// DefaultMaxRekeyFailures is how many consecutive rekey (non-initial) failures
-	// are tolerated before OnRekeyFailed. The initial exchange fails immediately.
-	DefaultMaxRekeyFailures = 3
-
-	// rotationActivityWindow gates rotation on recent real-data activity: a rekey
-	// clocks a rotation only if the peer exchanged user data within this window. It
-	// must stay shorter than the data path's rekey interval (WireGuard
-	// REKEY_AFTER_TIME ~120s) so the rotation's own traffic — which itself renews the
-	// activity signal — ages out before the next rekey, letting an idle tunnel stop
-	// rotating instead of self-sustaining.
-	rotationActivityWindow = 90 * time.Second
-)
-
-// LocalID and RemoteID are peer identity keys (e.g. WireGuard public keys). They are
-// distinct types so the local and a remote identity cannot be mixed up.
-type (
-	LocalID  string
-	RemoteID string
-)
-
-// Transport is the data-path socket the Manager drives (the analogue of
-// go-rosenpass's Conn). It is a dumb mover of bytes to/from endpoints: the Manager
-// owns the remoteID<->endpoint routing and hands the transport a resolved endpoint
-// to Send, and reverse-resolves the source of each inbound datagram. Its lifecycle
-// belongs to the Manager (Run at Start, Close at Stop).
-type Transport interface {
-	// Send delivers msg to the given data-path endpoint.
-	Send(endpoint netip.AddrPort, msg []byte) error
-	// LocalPort is the bound local UDP port, announced to peers so they know where
-	// to send data-path messages.
-	LocalPort() int
-	// Run starts delivering inbound datagrams as (source endpoint, msg) to onInbound
-	// and returns immediately; it runs until Close.
-	Run(onInbound func(src netip.AddrPort, msg []byte))
-	// Close stops delivery and releases the socket.
-	Close() error
-}
-
-// exchangeState is the single source of truth for an exchange's role and phase.
-type exchangeState uint8
-
-const (
-	stateReserved       exchangeState = iota // responder: deriving the answer
-	stateAwaitingAnswer                      // initiator: offer sent, awaiting the answer
-	stateAwaitingRekey                       // initiator: PSK derived+set, awaiting OnDataPathRekeyed to chain the next offer
-	stateAwaitingAck                         // responder: answer sent, awaiting the next offer that acks this exchange
-)
-
-// exchangeCtl holds all state for one in-flight exchange with a peer, under the
-// Manager's single lock. state drives every decision. lastSent is the current
-// data-path retransmit payload (the offer, for the initiator). initiator is the
-// ephemeral handle used at Finish; pendingPSK is the responder's derived key.
-// viaSignal records that the offer went to the host for the signalling channel, so
-// the loop does not retransmit it on the data path. Only the initiator runs a
-// retransmit loop, so only it sets cancel.
-type exchangeCtl struct {
-	id         ExchangeID
-	state      exchangeState
-	startedAt  time.Time
-	cancel     context.CancelFunc
-	lastSent   []byte
-	initiator  *Initiator
-	pendingPSK PSK
-	viaSignal  bool
-}
-
-// Manager is the stateful orchestrator — the analogue of go-rosenpass's Server. It
-// drives the X25519MLKEM768 exchange, owns the peer endpoint routing and the data-path
-// transport, and surfaces the derived PSK and convergence to the host via
-// CallbackHandler. It is event-driven: the bootstrap is triggered by the host
-// (SignalOffer) and each rotation is clocked by OnDataPathRekeyed. The cryptography is
-// the pure kem.go primitives; all state lives here under one lock.
-type Manager struct {
-	localID   LocalID
-	cbHandler CallbackHandler
-	logger    *slog.Logger
-
-	retryInterval    time.Duration
-	maxRetries       int
-	maxRekeyFailures int
-
-	rootCtx    context.Context
-	rootCancel context.CancelFunc
-
-	mu          sync.Mutex
-	transport   Transport
-	exchanges   map[RemoteID]*exchangeCtl   // in-flight exchange per peer
-	established map[RemoteID]bool           // peer has completed at least one exchange
-	failures    map[RemoteID]int            // consecutive rekey failures per peer
-	psks        map[RemoteID]PSK            // latest derived PSK per peer (pulled at WG peer-config time)
-	capable     map[RemoteID]bool           // peer runs the KEM (advertised a PQ port); false = known non-capable
-	peerAddrs   map[RemoteID]netip.AddrPort // remoteID -> data-path endpoint (send routing)
-	peersByAddr map[netip.AddrPort]RemoteID // reverse: source endpoint -> remoteID (inbound)
-	wait        sync.WaitGroup
-}
-
-// NewManager builds a manager for the local peer identified by its peer identity key
-// (used for the deterministic initiator role and the identity binding). A nil logger
-// falls back to slog.Default(). Install the data-path transport with Start.
-func NewManager(localID LocalID, h CallbackHandler, logger *slog.Logger) *Manager {
-	if logger == nil {
-		logger = slog.Default()
-	}
-	ctx, cancel := context.WithCancel(context.Background())
-	return &Manager{
-		localID:          localID,
-		cbHandler:        h,
-		logger:           logger,
-		retryInterval:    DefaultRetryInterval,
-		maxRetries:       DefaultMaxRetries,
-		maxRekeyFailures: DefaultMaxRekeyFailures,
-		rootCtx:          ctx,
-		rootCancel:       cancel,
-		exchanges:        make(map[RemoteID]*exchangeCtl),
-		established:      make(map[RemoteID]bool),
-		failures:         make(map[RemoteID]int),
-		psks:             make(map[RemoteID]PSK),
-		capable:          make(map[RemoteID]bool),
-		peerAddrs:        make(map[RemoteID]netip.AddrPort),
-		peersByAddr:      make(map[netip.AddrPort]RemoteID),
-	}
-}
-
-// Start installs the data-path transport and begins its inbound delivery. The Manager
-// owns it from here; Stop closes it.
-func (m *Manager) Start(t Transport) {
-	m.mu.Lock()
-	m.transport = t
-	m.mu.Unlock()
-	if t != nil {
-		t.Run(m.onDataPathInbound)
-	}
-}
-
-// LocalPort is the data-path transport's bound UDP port (0 if no transport), to be
-// announced to peers.
-func (m *Manager) LocalPort() int {
-	m.mu.Lock()
-	t := m.transport
-	m.mu.Unlock()
-	if t == nil {
-		return 0
-	}
-	return t.LocalPort()
-}
-
-// IsInitiator reports whether the local peer drives the exchange for this remote
-// peer. Roles are deterministic (lexicographic identity-key compare) so exactly one
-// side initiates, mirroring how Rosenpass picks its handshake initiator.
-func (m *Manager) IsInitiator(remoteID RemoteID) bool {
-	return string(m.localID) > string(remoteID)
-}
-
-// PSK returns the latest PSK derived for the peer, for the host to program at WG
-// peer-config time (the pull path). ok is false until an exchange has derived one.
-func (m *Manager) PSK(remoteID RemoteID) (PSK, bool) {
-	m.mu.Lock()
-	defer m.mu.Unlock()
-	psk, ok := m.psks[remoteID]
-	return psk, ok
-}
-
-// trace logs at LevelTrace, the verbose per-exchange lifecycle level gated by
-// NB_PQ_MLKEM_LOG_LEVEL=trace.
-func (m *Manager) trace(msg string, args ...any) {
-	m.logger.Log(context.Background(), LevelTrace, msg, args...)
-}
-
-// AddPeer registers where a peer's data-path messages are sent and received: its
-// overlay endpoint (IP:port). This is pure routing and says nothing about capability —
-// PQ capability is decided solely from the peer's KEM payload (see processOffer /
-// processAnswer / MarkNonCapable), never from an endpoint or port.
-func (m *Manager) AddPeer(remoteID RemoteID, endpoint netip.AddrPort) {
-	if !endpoint.IsValid() {
-		return
-	}
-	m.mu.Lock()
-	if old, ok := m.peerAddrs[remoteID]; ok {
-		delete(m.peersByAddr, old)
-	}
-	m.peerAddrs[remoteID] = endpoint
-	m.peersByAddr[endpoint] = remoteID
-	m.mu.Unlock()
-}
-
-// MarkNonCapable records that a peer does not run the KEM: it answered our offer with
-// no KEM material over signalling (the capability signal is the peer's payload, not its
-// optional data-path port). Any in-flight exchange is cancelled and further offers are
-// suppressed (see SignalOffer), so a non-PQ peer never drives the rekey-recovery storm.
-// An already-established peer is left untouched — a stray empty answer must not tear
-// down a working PQ session.
-func (m *Manager) MarkNonCapable(remoteID RemoteID) {
-	m.mu.Lock()
-	defer m.mu.Unlock()
-	if m.established[remoteID] {
-		return
-	}
-	if prev, ok := m.capable[remoteID]; ok && !prev {
-		return // already known non-capable, nothing to do
-	}
-	m.capable[remoteID] = false
-	if ex := m.exchanges[remoteID]; ex != nil {
-		if ex.cancel != nil {
-			ex.cancel()
-		}
-		delete(m.exchanges, remoteID)
-	}
-	m.trace("pqkem: peer advertises no PQ service — treating as non-capable, no KEM attempted", "peer", remoteID)
-}
-
-// RemovePeer stops any in-flight exchange for a peer and drops its state and routing.
-func (m *Manager) RemovePeer(remoteID RemoteID) {
-	m.mu.Lock()
-	if ex, ok := m.exchanges[remoteID]; ok {
-		if ex.cancel != nil {
-			ex.cancel()
-		}
-		delete(m.exchanges, remoteID)
-	}
-	delete(m.established, remoteID)
-	delete(m.failures, remoteID)
-	delete(m.psks, remoteID)
-	delete(m.capable, remoteID)
-	if ep, ok := m.peerAddrs[remoteID]; ok {
-		delete(m.peersByAddr, ep)
-		delete(m.peerAddrs, remoteID)
-	}
-	m.mu.Unlock()
-}
-
-// Stop cancels all in-flight exchanges, closes the transport, and waits for the
-// exchange goroutines to exit.
-func (m *Manager) Stop() {
-	// Cancel the root context under the lock, before Wait: startExchangeLocked checks
-	// rootCtx.Err() under the same lock before it Adds to the wait group, so once Stop
-	// has cancelled here no new Add can race Wait.
-	m.mu.Lock()
-	m.rootCancel()
-	m.mu.Unlock()
-	m.wait.Wait()
-	m.mu.Lock()
-	t := m.transport
-	m.transport = nil
-	m.exchanges = make(map[RemoteID]*exchangeCtl)
-	m.psks = make(map[RemoteID]PSK)
-	m.mu.Unlock()
-	if t != nil {
-		if err := t.Close(); err != nil {
-			m.logger.Warn("pqkem: closing data-path transport", "err", err)
-		}
-	}
-}
-
-// ---- Signalling channel (host-driven; rides the host's negotiation) ----
-
-// SignalOffer returns the KEM offer for the host to embed in its outgoing offer to
-// remoteID (bootstrap). It returns (nil, nil) when the local peer is not the
-// initiator. It is idempotent for an in-flight bootstrap: a repeat call returns the
-// same offer rather than starting a new exchange.
-//
-// A signal re-negotiation always re-bootstraps (fresh exchange): the remote may have
-// restarted and lost its PSK, so reusing a locally frozen one would desync. The derived
-// PSK still survives idle in the manager (dropped only on account-level peer removal),
-// so a pure lazy wake with no re-negotiation reuses it via the conn's WG-config pull.
-func (m *Manager) SignalOffer(remoteID RemoteID) ([]byte, error) {
-	if !m.IsInitiator(remoteID) {
-		return nil, nil
-	}
-	m.mu.Lock()
-	if capable, ok := m.capable[remoteID]; ok && !capable {
-		m.mu.Unlock()
-		return nil, nil // peer does not run the KEM; do not offer (avoids a failure/reoffer loop)
-	}
-	// Idempotent while a signalling bootstrap is in flight OR already derived a PSK but
-	// not yet chained a rotation (awaitingRekey): return the SAME offer instead of
-	// starting a new exchange. This matters when the controller both offers on its own
-	// guard AND re-offers in response to the responder's offer — without this, the
-	// second call would start a fresh exchange (a different PSK) and desync the peers.
-	if ex := m.exchanges[remoteID]; ex != nil && ex.viaSignal &&
-		(ex.state == stateAwaitingAnswer || ex.state == stateAwaitingRekey) {
-		last := ex.lastSent
-		m.mu.Unlock()
-		return last, nil
-	}
-	// Hold the lock across the check above and the install so a concurrent SignalOffer
-	// for the same peer can't also start an exchange. bootstrap offer acks nothing.
-	raw, err := m.startExchangeLocked(remoteID, true, ExchangeID{})
-	m.mu.Unlock()
-	return raw, err
-}
-
-// ShouldSendBootstrapOffer reports whether we should emit a fresh KEM offer to kick a
-// bootstrap for this peer. True only if we are the initiator, the peer is not known
-// non-capable, and no exchange is already in flight. The host uses this when it (as the
-// controller) receives the responder's offer: it replies with a KEM offer exactly once
-// to start the exchange, and ignores further responder offers while one is in flight,
-// avoiding an offer-per-offer runaway.
-func (m *Manager) ShouldSendBootstrapOffer(remoteID RemoteID) bool {
-	if !m.IsInitiator(remoteID) {
-		return false
-	}
-	m.mu.Lock()
-	defer m.mu.Unlock()
-	if capable, ok := m.capable[remoteID]; ok && !capable {
-		return false
-	}
-	return m.exchanges[remoteID] == nil
-}
-
-// SignalOnOffer processes a KEM offer the host extracted from an incoming offer and
-// returns the KEM answer for the host to embed in its outgoing answer.
-func (m *Manager) SignalOnOffer(remoteID RemoteID, offer []byte) ([]byte, error) {
-	typ, msg, err := Decode(offer)
-	if err != nil {
-		return nil, fmt.Errorf("decode signal offer from %s: %w", remoteID, err)
-	}
-	if typ != MsgOffer {
-		return nil, fmt.Errorf("expected offer from %s, got type %d", remoteID, typ)
-	}
-	return m.processOffer(remoteID, msg.(*OfferMsg))
-}
-
-// SignalOnAnswer processes a KEM answer the host extracted from an incoming answer.
-// There is no reply: the next offer (over the data path) acknowledges this exchange.
-func (m *Manager) SignalOnAnswer(remoteID RemoteID, answer []byte) error {
-	typ, msg, err := Decode(answer)
-	if err != nil {
-		return fmt.Errorf("decode signal answer from %s: %w", remoteID, err)
-	}
-	if typ != MsgAnswer {
-		return fmt.Errorf("expected answer from %s, got type %d", remoteID, typ)
-	}
-	return m.processAnswer(remoteID, msg.(*AnswerMsg))
-}
-
-// ---- Data path ----
-
-// onDataPathInbound is the transport's inbound handler: it reverse-resolves the
-// source endpoint to a peer and dispatches. Unknown sources are dropped.
-func (m *Manager) onDataPathInbound(src netip.AddrPort, msg []byte) {
-	m.mu.Lock()
-	remoteID, ok := m.peersByAddr[src]
-	m.mu.Unlock()
-	if !ok {
-		return
-	}
-	if err := m.OnDataPathMessage(remoteID, msg); err != nil {
-		m.trace("pqkem: inbound", "peer", remoteID, "err", err)
-	}
-}
-
-// OnDataPathMessage handles a KEM message received over the data path from remoteID
-// and pushes any reply back over the data path.
-func (m *Manager) OnDataPathMessage(remoteID RemoteID, raw []byte) error {
-	typ, msg, err := Decode(raw)
-	if err != nil {
-		return fmt.Errorf("decode data-path msg from %s: %w", remoteID, err)
-	}
-	switch typ {
-	case MsgOffer:
-		answer, err := m.processOffer(remoteID, msg.(*OfferMsg))
-		if err != nil {
-			return err
-		}
-		if answer == nil {
-			return nil
-		}
-		return m.pushDataPath(remoteID, answer)
-	case MsgAnswer:
-		return m.processAnswer(remoteID, msg.(*AnswerMsg))
-	default:
-		return fmt.Errorf("unhandled data-path message type %d from %s", typ, remoteID)
-	}
-}
-
-// OnDataPathRekeyed clocks the next chained PSK rotation on a fresh data-path rekey
-// (fired on first establishment AND every rekey). If we are the initiator that just
-// derived a PSK, it chains the next exchange: a fresh offer over the data path that
-// acknowledges the just-completed one (its arrival under the new key proves to the
-// responder the key works). sinceActivity is how long ago the peer last exchanged real
-// user data; past rotationActivityWindow the tunnel is treated as idle and rotation is
-// skipped — an idle tunnel has nothing to protect, and rotating would emit data-path
-// traffic that keeps the peer artificially active (see conn.onWGCheckSuccess).
-func (m *Manager) OnDataPathRekeyed(remoteID RemoteID, sinceActivity time.Duration) {
-	if sinceActivity >= rotationActivityWindow {
-		m.trace("pqkem: peer idle, skipping data-path rotation", "peer", remoteID, "since_activity", sinceActivity)
-		return
-	}
-
-	m.mu.Lock()
-	ex := m.exchanges[remoteID]
-	chain := ex != nil && ex.state == stateAwaitingRekey
-	if !chain {
-		m.mu.Unlock()
-		m.trace("pqkem: data-path rekey signal", "peer", remoteID, "chaining", false)
-		return
-	}
-	// Hold the lock across the awaitingRekey check and the install so two rekey clocks
-	// can't each start a chained exchange for the same peer.
-	offer, err := m.startExchangeLocked(remoteID, false, ex.id)
-	m.mu.Unlock()
-
-	m.trace("pqkem: data-path rekey signal", "peer", remoteID, "chaining", true)
-	if err != nil {
-		m.logger.Error("pqkem: chain offer failed to start", "peer", remoteID, "err", err)
-		return
-	}
-	if err := m.pushDataPath(remoteID, offer); err != nil {
-		m.logger.Warn("pqkem: send chain offer failed", "peer", remoteID, "err", err)
-		return
-	}
-	m.trace("pqkem: chain offer sent over data path", "peer", remoteID)
-}
-
-// OnDataPathDown notifies that the peer's data path went down. Rotations resume once
-// the host re-bootstraps over signalling on reconnect; in-flight data-path sends will
-// simply fail until then. Reserved as an explicit hook.
-func (m *Manager) OnDataPathDown(remoteID RemoteID) {}
-
-// ---- internals ----
-
-// pushDataPath resolves the peer's endpoint and sends over the data-path transport,
-// erroring if the peer is unknown or no transport is set.
-func (m *Manager) pushDataPath(remoteID RemoteID, msg []byte) error {
-	m.mu.Lock()
-	ep, ok := m.peerAddrs[remoteID]
-	t := m.transport
-	m.mu.Unlock()
-	if !ok {
-		return fmt.Errorf("no data-path endpoint for peer %s", remoteID)
-	}
-	if t == nil {
-		return fmt.Errorf("no data-path transport")
-	}
-	return t.Send(ep, msg)
-}
-
-func (m *Manager) binding(remoteID RemoteID) Binding {
-	return Binding{LocalID: []byte(m.localID), RemoteID: []byte(remoteID)}
-}
-
-func newExchangeID() (ExchangeID, error) {
-	var id ExchangeID
-	if _, err := rand.Read(id[:]); err != nil {
-		return ExchangeID{}, fmt.Errorf("generate exchange id: %w", err)
-	}
-	return id, nil
-}

client/internal/pqkem/manager_test.go

@@ -1,201 +0,0 @@
-package pqkem
-
-import (
-	"fmt"
-	"net/netip"
-	"sync"
-	"sync/atomic"
-	"testing"
-
-	"github.com/stretchr/testify/require"
-)
-
-// netSwitch is an in-memory UDP fabric: transports register their endpoint and get
-// datagrams delivered to their inbound handler.
-type netSwitch struct {
-	mu sync.Mutex
-	h  map[netip.AddrPort]func(netip.AddrPort, []byte)
-}
-
-func newSwitch() *netSwitch {
-	return &netSwitch{h: map[netip.AddrPort]func(netip.AddrPort, []byte){}}
-}
-
-func (s *netSwitch) register(ep netip.AddrPort, fn func(netip.AddrPort, []byte)) {
-	s.mu.Lock()
-	s.h[ep] = fn
-	s.mu.Unlock()
-}
-
-func (s *netSwitch) deliver(dst, src netip.AddrPort, msg []byte) error {
-	s.mu.Lock()
-	fn := s.h[dst]
-	s.mu.Unlock()
-	if fn == nil {
-		return fmt.Errorf("no route to %s", dst)
-	}
-	fn(src, msg)
-	return nil
-}
-
-// loopback is an endpoint-based pqkem.Transport over a netSwitch, with a switchable
-// drop flag.
-type loopback struct {
-	ep   netip.AddrPort
-	sw   *netSwitch
-	drop atomic.Bool
-}
-
-func (l *loopback) Send(dst netip.AddrPort, msg []byte) error {
-	if l.drop.Load() {
-		return nil
-	}
-	return l.sw.deliver(dst, l.ep, append([]byte(nil), msg...))
-}
-
-func (l *loopback) LocalPort() int                             { return int(l.ep.Port()) }
-func (l *loopback) Run(onInbound func(netip.AddrPort, []byte)) { l.sw.register(l.ep, onInbound) }
-func (l *loopback) Close() error                               { return nil }
-
-type fakeWG struct {
-	mu     sync.Mutex
-	psks   map[RemoteID]PSK
-	failed []RemoteID
-}
-
-func newFakeWG() *fakeWG { return &fakeWG{psks: map[RemoteID]PSK{}} }
-
-// startExchangeTest drives startExchangeLocked with the lock held, for tests that kick an
-// exchange directly (production callers hold m.mu across their idempotency check).
-func (m *Manager) startExchangeTest(remoteID RemoteID, viaSignal bool, ackID ExchangeID) ([]byte, error) {
-	m.mu.Lock()
-	defer m.mu.Unlock()
-	return m.startExchangeLocked(remoteID, viaSignal, ackID)
-}
-
-func (f *fakeWG) OnNewPSKReady(remoteID RemoteID, psk PSK) error {
-	f.mu.Lock()
-	defer f.mu.Unlock()
-	f.psks[remoteID] = psk
-	return nil
-}
-
-func (f *fakeWG) OnRekeyFailed(remoteID RemoteID) error {
-	f.mu.Lock()
-	defer f.mu.Unlock()
-	f.failed = append(f.failed, remoteID)
-	return nil
-}
-
-func (f *fakeWG) psk(peer RemoteID) PSK {
-	f.mu.Lock()
-	defer f.mu.Unlock()
-	return f.psks[peer]
-}
-
-var (
-	epA = netip.MustParseAddrPort("100.64.0.1:51833")
-	epB = netip.MustParseAddrPort("100.64.0.2:51833")
-)
-
-// pair builds two wired managers (B is the initiator, "bbbb" > "aaaa") sharing a
-// netSwitch, with each peer's data-path endpoint registered. lbB is B's loopback
-// (for toggling drop).
-func pair(t *testing.T) (dA, dB *Manager, wgA, wgB *fakeWG, lbB *loopback) {
-	t.Helper()
-	sw := newSwitch()
-	wgA = newFakeWG()
-	wgB = newFakeWG()
-	dA = NewManager("aaaa", wgA, nil)
-	dB = NewManager("bbbb", wgB, nil)
-	dA.Start(&loopback{ep: epA, sw: sw})
-	lbB = &loopback{ep: epB, sw: sw}
-	dB.Start(lbB)
-	dA.AddPeer("bbbb", epB)
-	dB.AddPeer("aaaa", epA)
-	return dA, dB, wgA, wgB, lbB
-}
-
-// bootstrap runs the signalling offer/answer (the test plays the host carrying bytes).
-func bootstrap(t *testing.T, dA, dB *Manager) {
-	t.Helper()
-	offer, err := dB.SignalOffer("aaaa")
-	require.NoError(t, err)
-	require.NotNil(t, offer)
-	answer, err := dA.SignalOnOffer("bbbb", offer)
-	require.NoError(t, err)
-	require.NotNil(t, answer)
-	require.NoError(t, dB.SignalOnAnswer("aaaa", answer))
-}
-
-func TestManager_BootstrapDerivesSamePSK(t *testing.T) {
-	dA, dB, wgA, wgB, _ := pair(t)
-	defer dA.Stop()
-	defer dB.Stop()
-
-	bootstrap(t, dA, dB)
-
-	pskA := wgA.psk("bbbb")
-	pskB := wgB.psk("aaaa")
-	require.NotEqual(t, PSK{}, pskA)
-	require.Equal(t, pskB, pskA, "both sides derive the same PSK from the bootstrap exchange")
-}
-
-func TestManager_ChainRotatesAndAcks(t *testing.T) {
-	dA, dB, wgA, wgB, _ := pair(t)
-	defer dA.Stop()
-	defer dB.Stop()
-
-	bootstrap(t, dA, dB)
-	psk1 := wgB.psk("aaaa")
-
-	// Data path up: B (initiator) chains the next offer over the data path, which
-	// rotates both to a fresh PSK and acknowledges A.
-	dA.OnDataPathRekeyed("bbbb", 0)
-	dB.OnDataPathRekeyed("aaaa", 0)
-
-	psk2A := wgA.psk("bbbb")
-	psk2B := wgB.psk("aaaa")
-	require.Equal(t, psk2B, psk2A, "both sides converge on the rotated PSK")
-	require.NotEqual(t, psk1, psk2B, "the chain rotated to a new PSK")
-}
-
-func TestManager_RotationSkippedWhenIdle(t *testing.T) {
-	dA, dB, wgA, wgB, _ := pair(t)
-	defer dA.Stop()
-	defer dB.Stop()
-
-	bootstrap(t, dA, dB)
-	psk1 := wgB.psk("aaaa")
-	require.NotEqual(t, PSK{}, psk1)
-
-	// Idle: the peer's last real-data activity is older than the window, so a rekey
-	// must NOT clock a rotation.
-	dA.OnDataPathRekeyed("bbbb", rotationActivityWindow)
-	dB.OnDataPathRekeyed("aaaa", rotationActivityWindow)
-	require.Equal(t, psk1, wgB.psk("aaaa"), "idle peer must not rotate the PSK")
-	require.Equal(t, psk1, wgA.psk("bbbb"), "idle peer must not rotate the PSK")
-
-	// Active: activity within the window clocks the rotation as usual.
-	dA.OnDataPathRekeyed("bbbb", rotationActivityWindow-1)
-	dB.OnDataPathRekeyed("aaaa", rotationActivityWindow-1)
-	psk2 := wgB.psk("aaaa")
-	require.NotEqual(t, psk1, psk2, "recent activity must clock a rotation")
-	require.Equal(t, psk2, wgA.psk("bbbb"), "both sides converge on the rotated PSK")
-}
-
-func TestManager_NonInitiatorReturnsNoOffer(t *testing.T) {
-	dA := NewManager("aaaa", newFakeWG(), nil)
-	defer dA.Stop()
-
-	offer, err := dA.SignalOffer("bbbb") // not the initiator vs "bbbb"
-	require.NoError(t, err)
-	require.Nil(t, offer)
-}
-
-func TestManager_StopIsIdempotent(t *testing.T) {
-	dA := NewManager("aaaa", newFakeWG(), nil)
-	dA.Start(&loopback{ep: epA, sw: newSwitch()})
-	dA.Stop()
-	dA.Stop() // must not panic or hang
-}

client/internal/pqkem/message.go

@@ -1,121 +0,0 @@
-package pqkem
-
-import (
-	"crypto/mlkem"
-	"fmt"
-)
-
-// Wire framing for the PQ-KEM exchange. Messages are self-contained, versioned,
-// transport-agnostic byte blobs: the same bytes ride the signalling channel
-// (initial bootstrap) or a data-tunnel packet (rekey). The library only ever sees
-// opaque []byte at the transport seam.
-//
-// Layout (all messages): [type:1][version:1][exchangeID:16][payload...]
-//
-// There is no confirm message: an exchange is acknowledged by the NEXT offer, which
-// carries the acked exchange's id (see OfferMsg.AckID) and — riding the data path
-// under the freshly adopted key — proves that key works.
-
-const (
-	// ProtocolVersion is bumped on any wire-incompatible change; a peer rejects
-	// messages it does not understand rather than misparsing them.
-	ProtocolVersion uint8 = 1
-
-	// ExchangeIDSize identifies one exchange so answers/acks correlate and stale
-	// messages are dropped.
-	ExchangeIDSize = 16
-
-	headerSize = 1 + 1 + ExchangeIDSize
-)
-
-// MsgType tags the two message kinds of the exchange.
-type MsgType uint8
-
-const (
-	MsgOffer MsgType = iota + 1
-	MsgAnswer
-)
-
-// ExchangeID is the per-exchange correlator. The zero value means "none" (an offer
-// that acknowledges nothing, i.e. the first exchange of a connection).
-type ExchangeID [ExchangeIDSize]byte
-
-// OfferMsg carries the initiator's public material (ML-KEM encap key ‖ X25519 pub)
-// and AckID, the id of the previous exchange this offer acknowledges (zero if none).
-type OfferMsg struct {
-	ExchangeID ExchangeID
-	AckID      ExchangeID
-	// KEMOffer is the raw Initiator.Offer() blob (OfferSize bytes).
-	KEMOffer []byte
-}
-
-// AnswerMsg carries the responder's reply (ML-KEM ciphertext ‖ X25519 pub) for the
-// round identified by ExchangeID.
-type AnswerMsg struct {
-	ExchangeID ExchangeID
-	// KEMAnswer is the raw Respond() answer blob (AnswerSize bytes).
-	KEMAnswer []byte
-}
-
-// Encode serialises the offer with its framed header (payload = AckID ‖ KEMOffer).
-func (m *OfferMsg) Encode() ([]byte, error) {
-	if len(m.KEMOffer) != OfferSize {
-		return nil, fmt.Errorf("offer payload: got %d, want %d", len(m.KEMOffer), OfferSize)
-	}
-	payload := make([]byte, 0, ExchangeIDSize+OfferSize)
-	payload = append(payload, m.AckID[:]...)
-	payload = append(payload, m.KEMOffer...)
-	return frame(MsgOffer, m.ExchangeID, payload), nil
-}
-
-// Encode serialises the answer with its framed header.
-func (m *AnswerMsg) Encode() ([]byte, error) {
-	if len(m.KEMAnswer) != AnswerSize {
-		return nil, fmt.Errorf("answer payload: got %d, want %d", len(m.KEMAnswer), AnswerSize)
-	}
-	return frame(MsgAnswer, m.ExchangeID, m.KEMAnswer), nil
-}
-
-// Decode parses a framed message into one of *OfferMsg / *AnswerMsg.
-func Decode(buf []byte) (MsgType, any, error) {
-	if len(buf) < headerSize {
-		return 0, nil, fmt.Errorf("message too short: %d bytes", len(buf))
-	}
-	typ := MsgType(buf[0])
-	if ver := buf[1]; ver != ProtocolVersion {
-		return typ, nil, fmt.Errorf("unsupported protocol version %d (want %d)", ver, ProtocolVersion)
-	}
-
-	var id ExchangeID
-	copy(id[:], buf[2:headerSize])
-	payload := buf[headerSize:]
-
-	switch typ {
-	case MsgOffer:
-		if len(payload) != ExchangeIDSize+OfferSize {
-			return typ, nil, fmt.Errorf("offer payload: got %d, want %d", len(payload), ExchangeIDSize+OfferSize)
-		}
-		var ack ExchangeID
-		copy(ack[:], payload[:ExchangeIDSize])
-		return typ, &OfferMsg{ExchangeID: id, AckID: ack, KEMOffer: payload[ExchangeIDSize:]}, nil
-	case MsgAnswer:
-		if len(payload) != AnswerSize {
-			return typ, nil, fmt.Errorf("answer payload: got %d, want %d", len(payload), AnswerSize)
-		}
-		return typ, &AnswerMsg{ExchangeID: id, KEMAnswer: payload}, nil
-	default:
-		return typ, nil, fmt.Errorf("unknown message type %d", typ)
-	}
-}
-
-func frame(typ MsgType, id ExchangeID, payload []byte) []byte {
-	buf := make([]byte, headerSize+len(payload))
-	buf[0] = byte(typ)
-	buf[1] = ProtocolVersion
-	copy(buf[2:], id[:])
-	copy(buf[headerSize:], payload)
-	return buf
-}
-
-// compile-time assurance the KEM blob sizes referenced here stay in sync with kem.go.
-var _ = [1]struct{}{}[OfferSize-(32+mlkem.EncapsulationKeySize768)]

client/internal/pqkem/message_test.go

@@ -1,57 +0,0 @@
-package pqkem
-
-import (
-	"testing"
-
-	"github.com/stretchr/testify/require"
-)
-
-func TestMessageRoundTrip(t *testing.T) {
-	init, err := NewInitiator()
-	require.NoError(t, err)
-	answer, _, err := Respond(init.Offer(), Binding{LocalID: wgB, RemoteID: wgA})
-	require.NoError(t, err)
-
-	id := ExchangeID{1, 2, 3, 4}
-	ack := ExchangeID{9, 9, 9}
-
-	offBytes, err := (&OfferMsg{ExchangeID: id, AckID: ack, KEMOffer: init.Offer()}).Encode()
-	require.NoError(t, err)
-	typ, decoded, err := Decode(offBytes)
-	require.NoError(t, err)
-	require.Equal(t, MsgOffer, typ)
-	require.Equal(t, id, decoded.(*OfferMsg).ExchangeID)
-	require.Equal(t, ack, decoded.(*OfferMsg).AckID)
-	require.Equal(t, init.Offer(), decoded.(*OfferMsg).KEMOffer)
-
-	ansBytes, err := (&AnswerMsg{ExchangeID: id, KEMAnswer: answer}).Encode()
-	require.NoError(t, err)
-	typ, decoded, err = Decode(ansBytes)
-	require.NoError(t, err)
-	require.Equal(t, MsgAnswer, typ)
-	require.Equal(t, answer, decoded.(*AnswerMsg).KEMAnswer)
-}
-
-func TestDecodeRejects(t *testing.T) {
-	// too short
-	_, _, err := Decode([]byte{1, 1})
-	require.Error(t, err)
-
-	// wrong version
-	bad := make([]byte, headerSize+ExchangeIDSize+OfferSize)
-	bad[0] = byte(MsgOffer)
-	bad[1] = ProtocolVersion + 1
-	_, _, err = Decode(bad)
-	require.Error(t, err)
-
-	// unknown type
-	bad2 := make([]byte, headerSize)
-	bad2[0] = 99
-	bad2[1] = ProtocolVersion
-	_, _, err = Decode(bad2)
-	require.Error(t, err)
-
-	// offer with wrong payload size
-	_, err = (&OfferMsg{KEMOffer: []byte{1, 2, 3}}).Encode()
-	require.Error(t, err)
-}

client/internal/pqkem_adapter.go

@@ -1,160 +0,0 @@
-package internal
-
-import (
-	"net/netip"
-	"time"
-
-	log "github.com/sirupsen/logrus"
-	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
-
-	"github.com/netbirdio/netbird/client/internal/pqkem"
-)
-
-// pqPresharedKeySetter is the subset of the WireGuard interface the ML-KEM callback
-// needs: programming a peer's preshared key. *iface.WGIface satisfies it.
-type pqPresharedKeySetter interface {
-	SetPresharedKey(peerKey string, psk wgtypes.Key, updateOnly bool) error
-}
-
-// pqCallbackHandler programs the derived PQ PSK onto the WireGuard peer. It is the
-// engine-side implementation of pqkem.CallbackHandler.
-type pqCallbackHandler struct {
-	wg pqPresharedKeySetter
-	// reoffer re-bootstraps the KEM over Signal for a peer (a fresh signalling offer)
-	// to recover from a persistent data-path rekey failure. Nil disables recovery.
-	reoffer func(remoteKey string)
-}
-
-// OnNewPSKReady programs the freshly derived PSK for the peer (updateOnly: a no-op
-// if the peer is not present, mirroring Rosenpass).
-func (h pqCallbackHandler) OnNewPSKReady(remoteID pqkem.RemoteID, psk pqkem.PSK) error {
-	// updateOnly: applies to an already-configured peer (rotation). At bootstrap the
-	// peer is not configured yet, so this is a no-op there and the PSK is instead
-	// pulled at peer-config time (pqHandshaker.PSK / conn.presharedKey).
-	log.Tracef("pqkem: programming PSK for peer %s", remoteID)
-	return h.wg.SetPresharedKey(string(remoteID), wgtypes.Key(psk), true)
-}
-
-// OnRekeyFailed reports a failed PQ (re)key convergence and re-bootstraps the KEM over
-// Signal to recover: a fresh signalling offer starts a new exchange that overwrites the
-// stalled PSK on both sides, resyncing after a persistent data-path desync. The tunnel
-// stays up on the previous PSK meanwhile (the Signal channel is independent of the
-// broken data path).
-func (h pqCallbackHandler) OnRekeyFailed(remoteID pqkem.RemoteID) error {
-	log.Warnf("pqkem: post-quantum rekey failed for peer %s, re-bootstrapping over signal", remoteID)
-	if h.reoffer != nil {
-		h.reoffer(string(remoteID))
-	}
-	return nil
-}
-
-// pqHandshaker adapts the pqkem manager to peer.PQHandshaker (string peer keys),
-// wiring the host's signalling offers/answers to the KEM exchange.
-type pqHandshaker struct {
-	mgr *pqkem.Manager
-}
-
-// announcedPort is the PQ data-path port to advertise to peers. It is omitted (0) when
-// the manager is on DefaultPort, since peers assume the default when no port is sent;
-// only a non-default (collision-forced) port is announced explicitly.
-func (p pqHandshaker) announcedPort() uint16 {
-	if port := p.mgr.LocalPort(); port != DefaultPort {
-		return uint16(port)
-	}
-	return 0
-}
-
-// OfferPayload builds the KEM offer to attach to an outgoing signalling offer for the
-// peer, plus the data-path port to announce (0 when on DefaultPort). Payload is nil when
-// this side has no offer to send.
-func (p pqHandshaker) OfferPayload(remoteKey string) ([]byte, uint16) {
-	payload, err := p.mgr.SignalOffer(pqkem.RemoteID(remoteKey))
-	if err != nil {
-		log.Warnf("pqkem: build offer for %s: %v", remoteKey, err)
-	}
-	return payload, p.announcedPort()
-}
-
-// ShouldSendBootstrapOffer reports whether the controller should reply to the peer's
-// KEM-less offer with its own bootstrap offer instead of an answer.
-func (p pqHandshaker) ShouldSendBootstrapOffer(remoteKey string) bool {
-	return p.mgr.ShouldSendBootstrapOffer(pqkem.RemoteID(remoteKey))
-}
-
-// AnswerPayload processes a received KEM offer (nil when absent) and returns the KEM
-// answer to attach to the outgoing signalling answer, plus the data-path port to announce
-// (0 when on DefaultPort). An empty offer is treated as a capability signal.
-func (p pqHandshaker) AnswerPayload(remoteKey string, recvOffer []byte) ([]byte, uint16) {
-	if len(recvOffer) == 0 {
-		// Capability signal (responder side): the KEM offer flows initiator->responder,
-		// so if we are the responder for this peer (it is the KEM initiator by role) an
-		// empty offer means it does not run the KEM. If we are the initiator, an empty
-		// offer is normal — the peer is the responder and puts its material in the
-		// answer — so we must not flag it.
-		if !p.mgr.IsInitiator(pqkem.RemoteID(remoteKey)) {
-			p.mgr.MarkNonCapable(pqkem.RemoteID(remoteKey))
-		}
-		return nil, p.announcedPort()
-	}
-	payload, err := p.mgr.SignalOnOffer(pqkem.RemoteID(remoteKey), recvOffer)
-	if err != nil {
-		log.Warnf("pqkem: build answer for %s: %v", remoteKey, err)
-	}
-	return payload, p.announcedPort()
-}
-
-// OnAnswer feeds a received KEM answer (nil when absent) into the exchange. An empty
-// answer to our offer is treated as a capability signal on the initiator side.
-func (p pqHandshaker) OnAnswer(remoteKey string, recvAnswer []byte) {
-	if len(recvAnswer) == 0 {
-		// Capability signal (initiator side): the KEM answer flows responder->initiator,
-		// so an empty answer to our offer means the peer does not run the KEM — mark it
-		// non-capable to stop offering (no failure/reoffer storm). Only meaningful when
-		// we are the initiator: as the responder we also receive an (empty) answer to
-		// our own non-KEM offer from a perfectly capable peer, which must not be flagged.
-		if p.mgr.IsInitiator(pqkem.RemoteID(remoteKey)) {
-			p.mgr.MarkNonCapable(pqkem.RemoteID(remoteKey))
-		}
-		return
-	}
-	if err := p.mgr.SignalOnAnswer(pqkem.RemoteID(remoteKey), recvAnswer); err != nil {
-		log.Warnf("pqkem: process answer from %s: %v", remoteKey, err)
-	}
-}
-
-// PSK exposes the peer's derived PSK for the conn to program at WG peer-config time.
-func (p pqHandshaker) PSK(remoteKey string) (wgtypes.Key, bool) {
-	psk, ok := p.mgr.PSK(pqkem.RemoteID(remoteKey))
-	if !ok {
-		return wgtypes.Key{}, false
-	}
-	return wgtypes.Key(psk), true
-}
-
-// SetRemoteAddr registers the peer's data-path endpoint learned from signalling. A
-// zero port means the peer omitted it (it is on DefaultPort), so we resolve it here —
-// DefaultPort lives in this package, not in peer. Sends only ever fire once the tunnel
-// is up (clocked by OnDataPathRekeyed), so registering here is safe even before
-// connection-up.
-func (p pqHandshaker) SetRemoteAddr(remoteKey string, addr netip.AddrPort) {
-	if !addr.Addr().IsValid() {
-		return
-	}
-	port := addr.Port()
-	if port == 0 {
-		port = DefaultPort
-	}
-	p.mgr.AddPeer(pqkem.RemoteID(remoteKey), netip.AddrPortFrom(addr.Addr(), port))
-}
-
-// OnDataPathRekeyed clocks the next chained PSK rotation on a fresh WG handshake.
-// sinceActivity is how long ago the peer last exchanged real user data; the manager
-// skips rotation for idle tunnels.
-func (p pqHandshaker) OnDataPathRekeyed(remoteKey string, sinceActivity time.Duration) {
-	p.mgr.OnDataPathRekeyed(pqkem.RemoteID(remoteKey), sinceActivity)
-}
-
-// OnDataPathDown signals the peer's tunnel went down.
-func (p pqHandshaker) OnDataPathDown(remoteKey string) {
-	p.mgr.OnDataPathDown(pqkem.RemoteID(remoteKey))
-}

client/internal/pqkem_adapter_test.go

@@ -1,37 +0,0 @@
-package internal
-
-import (
-	"testing"
-
-	"github.com/stretchr/testify/require"
-
-	"github.com/netbirdio/netbird/client/internal/pqkem"
-)
-
-type pqNoopHandler struct{}
-
-func (pqNoopHandler) OnNewPSKReady(pqkem.RemoteID, pqkem.PSK) error { return nil }
-func (pqNoopHandler) OnRekeyFailed(pqkem.RemoteID) error            { return nil }
-
-// TestPQAdapter_CapabilityRoleAware locks the role-aware capability signal: the KEM
-// payload only flows initiator-offer -> responder-answer, so an empty message in the
-// other direction comes from a perfectly capable peer and must NOT flag it. Only the
-// message that should carry material (the answer we receive as initiator) marks a peer
-// non-capable when empty.
-func TestPQAdapter_CapabilityRoleAware(t *testing.T) {
-	// localID "zzzz" > "aaaa" => this manager is the KEM initiator for peer "aaaa".
-	mgr := pqkem.NewManager("zzzz", pqNoopHandler{}, nil)
-	defer mgr.Stop()
-	h := pqHandshaker{mgr: mgr}
-
-	// An empty OFFER from our peer is normal here: as the initiator's responder it puts
-	// its material in the answer, not the offer. It must not disable our offering.
-	h.AnswerPayload("aaaa", nil)
-	payload, _ := h.OfferPayload("aaaa")
-	require.NotNil(t, payload, "an empty offer from a responder-role peer must not mark it non-capable")
-
-	// An empty ANSWER to our offer means the peer does not run the KEM -> stop offering.
-	h.OnAnswer("aaaa", nil)
-	payload2, _ := h.OfferPayload("aaaa")
-	require.Nil(t, payload2, "an empty answer to our offer marks the peer non-capable, so we stop offering")
-}

client/internal/pqkem_transport.go

@@ -1,76 +0,0 @@
-package internal
-
-import (
-	"fmt"
-	"net"
-	"net/netip"
-
-	log "github.com/sirupsen/logrus"
-)
-
-// DefaultPort is the preferred UDP port for the ML-KEM data-path service, bound on
-// the WG overlay IP. Since each client owns a distinct overlay IP, this port is
-// almost always free, so it need not be announced (peers assume it). A peer only
-// announces Body.mlkemPort when a collision forced it onto a different port.
-const DefaultPort = 51833
-
-// pqTransport is the ML-KEM data-path transport: a dumb UDP socket bound on the WG
-// overlay IP. It implements pqkem.Transport — the manager owns the remoteID<->endpoint
-// routing and drives this socket's lifecycle (Run / Close).
-type pqTransport struct {
-	conn *net.UDPConn
-	port int
-}
-
-// newPQTransport binds a UDP socket on the WG overlay IP, preferring DefaultPort and
-// falling back to an OS-assigned ephemeral port if it is in use. Call it after the WG
-// interface is up so the overlay IP is assigned; when the bound port is not
-// DefaultPort it must be announced to peers via Body.mlkemPort.
-func newPQTransport(overlayIP netip.Addr) (*pqTransport, error) {
-	if !overlayIP.IsValid() {
-		return nil, fmt.Errorf("invalid overlay IP for pqkem transport")
-	}
-	// The WG overlay always carries an IPv4 address (v6 is additive, never standalone),
-	// so the transport binds over IPv4. Unmap first so AsSlice() yields 4 bytes for an
-	// IPv4-mapped IPv6 address (a hardcoded "udp4" would otherwise fail on its 16 bytes).
-	overlayIP = overlayIP.Unmap()
-	ip := net.IP(overlayIP.AsSlice())
-	conn, err := net.ListenUDP("udp4", &net.UDPAddr{IP: ip, Port: DefaultPort})
-	if err != nil {
-		log.Debugf("pqkem: default port %d unavailable on %s (%v), using an ephemeral port", DefaultPort, overlayIP, err)
-		conn, err = net.ListenUDP("udp4", &net.UDPAddr{IP: ip, Port: 0})
-		if err != nil {
-			return nil, fmt.Errorf("bind pqkem udp on overlay %s: %w", overlayIP, err)
-		}
-	}
-	return &pqTransport{conn: conn, port: conn.LocalAddr().(*net.UDPAddr).Port}, nil
-}
-
-// Send implements pqkem.Transport.
-func (t *pqTransport) Send(endpoint netip.AddrPort, msg []byte) error {
-	_, err := t.conn.WriteToUDPAddrPort(msg, endpoint)
-	return err
-}
-
-// LocalPort implements pqkem.Transport.
-func (t *pqTransport) LocalPort() int { return t.port }
-
-// Run implements pqkem.Transport: the receive loop, delivering each datagram as
-// (source endpoint, msg). Exits when the socket is closed.
-func (t *pqTransport) Run(onInbound func(src netip.AddrPort, msg []byte)) {
-	go func() {
-		buf := make([]byte, 2048)
-		for {
-			n, src, err := t.conn.ReadFromUDPAddrPort(buf)
-			if err != nil {
-				return
-			}
-			msg := make([]byte, n)
-			copy(msg, buf[:n])
-			onInbound(src, msg)
-		}
-	}()
-}
-
-// Close implements pqkem.Transport.
-func (t *pqTransport) Close() error { return t.conn.Close() }

client/ios/NetBirdSDK/client.go

@@ -14,6 +14,7 @@ import (
 
 	log "github.com/sirupsen/logrus"
 
+	nbAnonymize "github.com/netbirdio/netbird/client/anonymize"
 	"github.com/netbirdio/netbird/client/internal"
 	"github.com/netbirdio/netbird/client/internal/auth"
 	"github.com/netbirdio/netbird/client/internal/debug"
@@ -200,8 +201,10 @@ func (c *Client) Stop() {
 // DebugBundle generates a debug bundle, uploads it and returns the upload key.
 // It works with or without a running engine: when the engine is up it reuses
 // the live config, sync response and client metrics; otherwise it loads the
-// config from disk (or the preloaded tvOS config).
-func (c *Client) DebugBundle(anonymize bool) (string, error) {
+// config from disk (or the preloaded tvOS config). anonymizeLevel is "default"
+// or "strict"; strict also anonymizes internal IP ranges, peer names, and
+// WireGuard public keys, and implies anonymize.
+func (c *Client) DebugBundle(anonymize bool, anonymizeLevel string) (string, error) {
 	cfg, cc := c.stateSnapshot()
 
 	// If the engine hasn't been started, load config so we can reach management.
@@ -251,6 +254,7 @@ func (c *Client) DebugBundle(anonymize bool) (string, error) {
 		deps,
 		debug.BundleConfig{
 			Anonymize:         anonymize,
+			AnonymizeLevel:    nbAnonymize.ParseLevel(anonymizeLevel),
 			IncludeSystemInfo: true,
 		},
 	)

client/proto/daemon.pb.go

@@ -2781,6 +2781,11 @@ type DebugBundleRequest struct {
 	// untrusted TLS certificate. Restricted to privileged callers; for
 	// self-hosted upload servers.
 	UploadInsecure bool `protobuf:"varint,7,opt,name=uploadInsecure,proto3" json:"uploadInsecure,omitempty"`
+	// anonymizeLevel selects how much the anonymizer redacts: "default"
+	// (or empty) keeps internal IP ranges, "strict" also anonymizes them.
+	// Unknown values are treated as "strict". Only meaningful with anonymize;
+	// "strict" implies it.
+	AnonymizeLevel string `protobuf:"bytes,8,opt,name=anonymizeLevel,proto3" json:"anonymizeLevel,omitempty"`
 	unknownFields  protoimpl.UnknownFields
 	sizeCache      protoimpl.SizeCache
 }
@@ -2857,6 +2862,13 @@ func (x *DebugBundleRequest) GetUploadInsecure() bool {
 	return false
 }
 
+func (x *DebugBundleRequest) GetAnonymizeLevel() string {
+	if x != nil {
+		return x.AnonymizeLevel
+	}
+	return ""
+}
+
 type DebugBundleResponse struct {
 	state               protoimpl.MessageState `protogen:"open.v1"`
 	Path                string                 `protobuf:"bytes,1,opt,name=path,proto3" json:"path,omitempty"`
@@ -7253,7 +7265,7 @@ const file_daemon_proto_rawDesc = "" +
 	"\x12translatedHostname\x18\x04 \x01(\tR\x12translatedHostname\x128\n" +
 	"\x0etranslatedPort\x18\x05 \x01(\v2\x10.daemon.PortInfoR\x0etranslatedPort\"G\n" +
 	"\x17ForwardingRulesResponse\x12,\n" +
-	"\x05rules\x18\x01 \x03(\v2\x16.daemon.ForwardingRuleR\x05rules\"\xdc\x01\n" +
+	"\x05rules\x18\x01 \x03(\v2\x16.daemon.ForwardingRuleR\x05rules\"\x84\x02\n" +
 	"\x12DebugBundleRequest\x12\x1c\n" +
 	"\tanonymize\x18\x01 \x01(\bR\tanonymize\x12\x1e\n" +
 	"\n" +
@@ -7264,7 +7276,8 @@ const file_daemon_proto_rawDesc = "" +
 	"\n" +
 	"cliVersion\x18\x06 \x01(\tR\n" +
 	"cliVersion\x12&\n" +
-	"\x0euploadInsecure\x18\a \x01(\bR\x0euploadInsecure\"}\n" +
+	"\x0euploadInsecure\x18\a \x01(\bR\x0euploadInsecure\x12&\n" +
+	"\x0eanonymizeLevel\x18\b \x01(\tR\x0eanonymizeLevel\"}\n" +
 	"\x13DebugBundleResponse\x12\x12\n" +
 	"\x04path\x18\x01 \x01(\tR\x04path\x12 \n" +
 	"\vuploadedKey\x18\x02 \x01(\tR\vuploadedKey\x120\n" +

client/proto/daemon.proto

@@ -540,6 +540,11 @@ message DebugBundleRequest {
   // untrusted TLS certificate. Restricted to privileged callers; for
   // self-hosted upload servers.
   bool uploadInsecure = 7;
+  // anonymizeLevel selects how much the anonymizer redacts: "default"
+  // (or empty) keeps internal IP ranges, "strict" also anonymizes them.
+  // Unknown values are treated as "strict". Only meaningful with anonymize;
+  // "strict" implies it.
+  string anonymizeLevel = 8;
 }
 
 message DebugBundleResponse {

client/server/debug.go

@@ -16,6 +16,7 @@ import (
 	"google.golang.org/grpc/codes"
 	gstatus "google.golang.org/grpc/status"
 
+	"github.com/netbirdio/netbird/client/anonymize"
 	"github.com/netbirdio/netbird/client/internal/debug"
 	"github.com/netbirdio/netbird/client/internal/ipcauth"
 	"github.com/netbirdio/netbird/client/proto"
@@ -122,6 +123,7 @@ func (s *Server) generateDebugBundle(req *proto.DebugBundleRequest, uiOpener deb
 		},
 		debug.BundleConfig{
 			Anonymize:         req.GetAnonymize(),
+			AnonymizeLevel:    anonymize.ParseLevel(req.GetAnonymizeLevel()),
 			IncludeSystemInfo: req.GetSystemInfo(),
 			LogFileCount:      req.GetLogFileCount(),
 		},

client/status/status.go

@@ -46,7 +46,10 @@ func ParseDaemonStatus(s string) DaemonStatus {
 
 // ConvertOptions holds parameters for ConvertToStatusOutputOverview.
 type ConvertOptions struct {
-	Anonymize            bool
+	Anonymize bool
+	// AnonymizeLevel selects how much the anonymizer redacts. Only
+	// meaningful when Anonymize is set.
+	AnonymizeLevel       anonymize.Level
 	DaemonVersion        string
 	DaemonStatus         DaemonStatus
 	StatusFilter         string
@@ -217,6 +220,7 @@ func ConvertToStatusOutputOverview(pbFullStatus *proto.FullStatus, opts ConvertO
 
 	if opts.Anonymize {
 		anonymizer := anonymize.NewAnonymizer(anonymize.DefaultAddresses())
+		anonymizer.SetLevel(opts.AnonymizeLevel)
 		anonymizeOverview(anonymizer, &overview)
 	}
 
@@ -976,6 +980,7 @@ func timeAgo(t time.Time) string {
 
 func anonymizePeerDetail(a *anonymize.Anonymizer, peer *PeerStateDetailOutput) {
 	peer.FQDN = a.AnonymizeDomain(peer.FQDN)
+	peer.PubKey = a.AnonymizeWGKey(peer.PubKey)
 	if localIP, port, err := net.SplitHostPort(peer.IceCandidateEndpoint.Local); err == nil {
 		peer.IceCandidateEndpoint.Local = fmt.Sprintf("%s:%s", a.AnonymizeIPString(localIP), port)
 	}
@@ -1007,6 +1012,7 @@ func anonymizeOverview(a *anonymize.Anonymizer, overview *OutputOverview) {
 	overview.SignalState.URL = a.AnonymizeURI(overview.SignalState.URL)
 	overview.SignalState.Error = a.AnonymizeString(overview.SignalState.Error)
 
+	overview.PubKey = a.AnonymizeWGKey(overview.PubKey)
 	overview.IP = a.AnonymizeIPString(overview.IP)
 	overview.IPv6 = a.AnonymizeIPString(overview.IPv6)
 	for i, detail := range overview.Relays.Details {

client/ui/frontend/src/contexts/DebugBundleContext.tsx

@@ -71,10 +71,12 @@ type BundleOptions = {
     hasWindow: boolean;
     totalSec: number;
     uploadUrl: string;
-    anonymize: boolean;
+    anonymizeLevel: AnonymizeLevel;
     systemInfo: boolean;
 };
 
+export type AnonymizeLevel = "none" | "default" | "strict";
+
 const startCaptureBestEffort = async (totalSec: number, pcap: CaptureState) => {
     try {
         // Mirror the CLI's safety margin: window + 30s, server caps at 10m.
@@ -187,7 +189,10 @@ const runBundleFlow = async (
 
     if (opts.uploadUrl) setStage({ kind: "uploading" });
     const result = await DebugSvc.Bundle({
-        anonymize: opts.anonymize,
+        anonymize: opts.anonymizeLevel !== "none",
+        // The daemon only knows "default" and "strict"; "none" is expressed
+        // through the anonymize flag being off.
+        anonymizeLevel: opts.anonymizeLevel === "strict" ? "strict" : "default",
         systemInfo: opts.systemInfo,
         uploadUrl: opts.uploadUrl,
         logFileCount,
@@ -198,7 +203,7 @@ const runBundleFlow = async (
 };
 
 const useDebugBundle = () => {
-    const [anonymize, setAnonymize] = useState(false);
+    const [anonymizeLevel, setAnonymizeLevel] = useState<AnonymizeLevel>("none");
     const [systemInfo, setSystemInfo] = useState(true);
     const [upload, setUpload] = useState(true);
     const [trace, setTrace] = useState(true);
@@ -240,7 +245,7 @@ const useDebugBundle = () => {
             hasWindow: capture && totalSec > 0,
             totalSec,
             uploadUrl: upload ? NETBIRD_UPLOAD_URL : "",
-            anonymize,
+            anonymizeLevel,
             systemInfo,
         };
 
@@ -272,8 +277,8 @@ const useDebugBundle = () => {
     };
 
     return {
-        anonymize,
-        setAnonymize,
+        anonymizeLevel,
+        setAnonymizeLevel,
         systemInfo,
         setSystemInfo,
         upload,

client/ui/frontend/src/hooks/useKeepConnectedOnQuit.ts

@@ -1,35 +0,0 @@
-import { useCallback, useEffect, useState } from "react";
-import { Preferences } from "@bindings/services";
-
-export const useKeepConnectedOnQuit = () => {
-    const [keepConnected, setKeepConnected] = useState<boolean | null>(null);
-
-    useEffect(() => {
-        let cancelled = false;
-        Preferences.Get()
-            .then((prefs) => {
-                if (cancelled) return;
-                setKeepConnected(prefs?.keepConnectedOnQuit ?? false);
-            })
-            .catch((err: unknown) => {
-                if (cancelled) return;
-                console.warn("[useKeepConnectedOnQuit] load preferences failed", err);
-                setKeepConnected(false);
-            });
-        return () => {
-            cancelled = true;
-        };
-    }, []);
-
-    const setKeepConnectedOnQuit = useCallback(async (keep: boolean) => {
-        setKeepConnected(keep);
-        try {
-            await Preferences.SetKeepConnectedOnQuit(keep);
-        } catch (err: unknown) {
-            setKeepConnected(!keep);
-            console.error("[useKeepConnectedOnQuit] SetKeepConnectedOnQuit failed", err);
-        }
-    }, []);
-
-    return { keepConnected, setKeepConnectedOnQuit };
-};

client/ui/frontend/src/modules/settings/SettingsGeneral.tsx

@@ -11,7 +11,6 @@ import { ManagementServerSwitch } from "@/components/ManagementServerSwitch.tsx"
 import { ManagementMode, useManagementUrl } from "@/hooks/useManagementUrl.ts";
 import { LanguagePicker } from "@/components/LanguagePicker.tsx";
 import { useRestrictions } from "@/contexts/RestrictionsContext.tsx";
-import { useKeepConnectedOnQuit } from "@/hooks/useKeepConnectedOnQuit.ts";
 
 export function SettingsGeneral() {
     const { t } = useTranslation();
@@ -20,7 +19,6 @@ export function SettingsGeneral() {
     const { mode, setMode, setUrl, displayUrl, showError, canSave, save, checking, unreachable } =
         useManagementUrl();
     const { mdm, features } = useRestrictions();
-    const { keepConnected, setKeepConnectedOnQuit } = useKeepConnectedOnQuit();
 
     const inputRef = useRef<HTMLInputElement>(null);
     const managementUrlId = useId();
@@ -59,15 +57,6 @@ export function SettingsGeneral() {
                         helpText={t("settings.general.autostart.help")}
                     />
                 )}
-                <FancyToggleSwitch
-                    value={keepConnected ?? false}
-                    onChange={(v) => {
-                        void setKeepConnectedOnQuit(v);
-                    }}
-                    loading={keepConnected === null}
-                    label={t("settings.general.keepConnectedOnQuit.label")}
-                    helpText={t("settings.general.keepConnectedOnQuit.help")}
-                />
             </SectionGroup>
 
             {!mdm.managementURL && !features.disableUpdateSettings && (

client/ui/frontend/src/modules/settings/SettingsTroubleshooting.tsx

@@ -1,6 +1,6 @@
 import { useId, type ReactNode } from "react";
 import { Trans, useTranslation } from "react-i18next";
-import { CircleCheckBig, FolderOpen, Loader2 } from "lucide-react";
+import { ChevronDown, CircleCheckBig, FolderOpen, Info, Loader2 } from "lucide-react";
 import { Browser } from "@wailsio/runtime";
 import { Debug as DebugSvc } from "@bindings/services";
 import type { DebugBundleResult } from "@bindings/services/models.js";
@@ -8,13 +8,22 @@ import { Button } from "@/components/buttons/Button";
 import { DialogActions } from "@/components/dialog/DialogActions";
 import { DialogDescription } from "@/components/dialog/DialogDescription";
 import { DialogHeading } from "@/components/dialog/DialogHeading";
+import {
+    DropdownMenu,
+    DropdownMenuContent,
+    DropdownMenuRadioGroup,
+    DropdownMenuRadioItem,
+    DropdownMenuTrigger,
+} from "@/components/DropdownMenu";
 import FancyToggleSwitch from "@/components/switches/FancyToggleSwitch";
 import HelpText from "@/components/typography/HelpText.tsx";
 import { Input } from "@/components/inputs/Input";
 import { Label } from "@/components/typography/Label";
 import { SquareIcon } from "@/components/SquareIcon";
+import { Tooltip } from "@/components/Tooltip";
+import { cn } from "@/lib/cn";
 import { formatRemaining } from "@/lib/formatters";
-import type { DebugStage } from "@/contexts/DebugBundleContext";
+import type { AnonymizeLevel, DebugStage } from "@/contexts/DebugBundleContext";
 import { useDebugBundleContext } from "@/contexts/DebugBundleContext";
 import { SectionGroup, SettingsBottomBar } from "@/modules/settings/SettingsSection.tsx";
 
@@ -24,8 +33,8 @@ export function SettingsTroubleshooting() {
     const { t } = useTranslation();
     const durationId = useId();
     const {
-        anonymize,
-        setAnonymize,
+        anonymizeLevel,
+        setAnonymizeLevel,
         systemInfo,
         setSystemInfo,
         upload,
@@ -55,12 +64,71 @@ export function SettingsTroubleshooting() {
 
     return (
         <SectionGroup title={t("settings.troubleshooting.section.title")}>
-            <FancyToggleSwitch
-                value={anonymize}
-                onChange={setAnonymize}
-                label={t("settings.troubleshooting.anonymize.label")}
-                helpText={t("settings.troubleshooting.anonymize.help")}
-            />
+            <div className={"flex items-center justify-between gap-6"}>
+                <div className={"max-w-md flex-1"}>
+                    <Label as={"div"}>
+                        <span className={"inline-flex items-center gap-1.5"}>
+                            {t("settings.troubleshooting.anonymize.label")}
+                            <Tooltip
+                                content={
+                                    <div className={"max-w-xs whitespace-normal leading-relaxed"}>
+                                        {t("settings.troubleshooting.anonymize.info")}
+                                    </div>
+                                }
+                            >
+                                <Info
+                                    size={14}
+                                    aria-label={t("settings.troubleshooting.anonymize.label")}
+                                    className={"shrink-0 cursor-default text-nb-gray-400"}
+                                />
+                            </Tooltip>
+                        </span>
+                    </Label>
+                    <HelpText margin={false}>
+                        {t("settings.troubleshooting.anonymize.help")}
+                    </HelpText>
+                </div>
+                <div className={"shrink-0"}>
+                    <DropdownMenu>
+                        <DropdownMenuTrigger asChild>
+                            <button
+                                type={"button"}
+                                aria-label={t("settings.troubleshooting.anonymize.label")}
+                                className={cn(
+                                    "inline-flex h-[40px] min-w-[160px] items-center justify-between gap-2 px-3",
+                                    "rounded-md border bg-white dark:bg-nb-gray-900",
+                                    "border-neutral-200 dark:border-nb-gray-700",
+                                    "cursor-default text-xs font-semibold text-nb-gray-100 outline-none",
+                                    "hover:border-nb-gray-600 data-[state=open]:border-nb-gray-600",
+                                )}
+                            >
+                                {t(`settings.troubleshooting.anonymize.${anonymizeLevel}`)}
+                                <ChevronDown
+                                    size={16}
+                                    aria-hidden={"true"}
+                                    className={"shrink-0 text-nb-gray-200"}
+                                />
+                            </button>
+                        </DropdownMenuTrigger>
+                        <DropdownMenuContent align={"end"} className={"min-w-[160px]"}>
+                            <DropdownMenuRadioGroup
+                                value={anonymizeLevel}
+                                onValueChange={(v) => setAnonymizeLevel(v as AnonymizeLevel)}
+                            >
+                                <DropdownMenuRadioItem value={"none"}>
+                                    {t("settings.troubleshooting.anonymize.none")}
+                                </DropdownMenuRadioItem>
+                                <DropdownMenuRadioItem value={"default"}>
+                                    {t("settings.troubleshooting.anonymize.default")}
+                                </DropdownMenuRadioItem>
+                                <DropdownMenuRadioItem value={"strict"}>
+                                    {t("settings.troubleshooting.anonymize.strict")}
+                                </DropdownMenuRadioItem>
+                            </DropdownMenuRadioGroup>
+                        </DropdownMenuContent>
+                    </DropdownMenu>
+                </div>
+            </div>
             <FancyToggleSwitch
                 value={systemInfo}
                 onChange={setSystemInfo}

client/ui/i18n/locales/de/common.json

@@ -551,14 +551,6 @@
     "settings.general.autostart.errorTitle": {
         "message": "Ändern des Autostarts fehlgeschlagen"
     },
-    "settings.general.keepConnectedOnQuit.label": {
-        "message": "Nach dem Beenden verbunden bleiben",
-        "description": "Toggle label: keep the VPN connection up after quitting the UI."
-    },
-    "settings.general.keepConnectedOnQuit.help": {
-        "message": "Die Verbindung bleibt im Hintergrund bestehen, nachdem Sie NetBird schließen. Sie endet erst, wenn Sie sie selbst trennen.",
-        "description": "Helper text for the stay-connected-after-quitting toggle."
-    },
     "settings.general.language.label": {
         "message": "Anzeigesprache"
     },

client/ui/i18n/locales/en/common.json

@@ -735,14 +735,6 @@
         "message": "Autostart Change Failed",
         "description": "Error-dialog title when changing the autostart setting fails."
     },
-    "settings.general.keepConnectedOnQuit.label": {
-        "message": "Stay Connected After Quitting",
-        "description": "Toggle label: keep the VPN connection up after quitting the UI."
-    },
-    "settings.general.keepConnectedOnQuit.help": {
-        "message": "The connection stays up in the background after you close NetBird. It only stops when you disconnect it yourself.",
-        "description": "Helper text for the stay-connected-after-quitting toggle."
-    },
     "settings.general.language.label": {
         "message": "Display Language",
         "description": "Label for the display-language picker."
@@ -1013,11 +1005,27 @@
     },
     "settings.troubleshooting.anonymize.label": {
         "message": "Anonymize Sensitive Information",
-        "description": "Toggle label: anonymize sensitive information in the bundle."
+        "description": "Label for the anonymization level dropdown (None, Default, Strict)."
     },
     "settings.troubleshooting.anonymize.help": {
-        "message": "Hides public IP addresses and non-NetBird domains from logs.",
-        "description": "Helper text for anonymizing logs (hides public IPs and non-NetBird domains)."
+        "message": "Hides IP addresses, domains, and other sensitive values.",
+        "description": "Helper text under the anonymization dropdown. The level details live in the info tooltip."
+    },
+    "settings.troubleshooting.anonymize.info": {
+        "message": "Default keeps internal IPv4 addresses and peer names readable for support. Strict additionally anonymizes private (RFC 1918), CGNAT, and link-local IP addresses, peer names, and WireGuard public keys. Recurring values map to the same placeholder, so peers stay distinguishable. Use Strict when sharing the bundle outside your organization.",
+        "description": "Info tooltip explaining the anonymization levels. 'RFC 1918', 'CGNAT', 'link-local', and 'WireGuard' are technical terms — keep them."
+    },
+    "settings.troubleshooting.anonymize.none": {
+        "message": "None",
+        "description": "Dropdown option: no anonymization."
+    },
+    "settings.troubleshooting.anonymize.default": {
+        "message": "Default",
+        "description": "Dropdown option: default anonymization level."
+    },
+    "settings.troubleshooting.anonymize.strict": {
+        "message": "Strict",
+        "description": "Dropdown option: strict anonymization level."
     },
     "settings.troubleshooting.systemInfo.label": {
         "message": "Include System Information",

client/ui/i18n/locales/es/common.json

@@ -551,14 +551,6 @@
     "settings.general.autostart.errorTitle": {
         "message": "Error al cambiar el inicio automático"
     },
-    "settings.general.keepConnectedOnQuit.label": {
-        "message": "Permanecer conectado al salir",
-        "description": "Toggle label: keep the VPN connection up after quitting the UI."
-    },
-    "settings.general.keepConnectedOnQuit.help": {
-        "message": "La conexión sigue activa en segundo plano después de cerrar NetBird. Solo se detiene cuando la desconectas tú.",
-        "description": "Helper text for the stay-connected-after-quitting toggle."
-    },
     "settings.general.language.label": {
         "message": "Idioma de la interfaz"
     },

client/ui/i18n/locales/fr/common.json

@@ -551,14 +551,6 @@
     "settings.general.autostart.errorTitle": {
         "message": "Échec de la modification du démarrage automatique"
     },
-    "settings.general.keepConnectedOnQuit.label": {
-        "message": "Rester connecté après la fermeture",
-        "description": "Toggle label: keep the VPN connection up after quitting the UI."
-    },
-    "settings.general.keepConnectedOnQuit.help": {
-        "message": "La connexion reste active en arrière-plan après la fermeture de NetBird. Elle ne s'arrête que si vous la coupez vous-même.",
-        "description": "Helper text for the stay-connected-after-quitting toggle."
-    },
     "settings.general.language.label": {
         "message": "Langue d’affichage"
     },

client/ui/i18n/locales/hu/common.json

@@ -551,14 +551,6 @@
     "settings.general.autostart.errorTitle": {
         "message": "Az automatikus indítás módosítása sikertelen"
     },
-    "settings.general.keepConnectedOnQuit.label": {
-        "message": "Kapcsolat megtartása kilépéskor",
-        "description": "Toggle label: keep the VPN connection up after quitting the UI."
-    },
-    "settings.general.keepConnectedOnQuit.help": {
-        "message": "A kapcsolat a háttérben megmarad, miután bezárod a NetBirdöt. Csak akkor szakad meg, ha te magad bontod.",
-        "description": "Helper text for the stay-connected-after-quitting toggle."
-    },
     "settings.general.language.label": {
         "message": "Megjelenítési nyelv"
     },

client/ui/i18n/locales/it/common.json

@@ -551,14 +551,6 @@
     "settings.general.autostart.errorTitle": {
         "message": "Modifica avvio automatico non riuscita"
     },
-    "settings.general.keepConnectedOnQuit.label": {
-        "message": "Resta connesso dopo la chiusura",
-        "description": "Toggle label: keep the VPN connection up after quitting the UI."
-    },
-    "settings.general.keepConnectedOnQuit.help": {
-        "message": "La connessione resta attiva in background dopo la chiusura di NetBird. Si interrompe solo quando la disconnetti tu.",
-        "description": "Helper text for the stay-connected-after-quitting toggle."
-    },
     "settings.general.language.label": {
         "message": "Lingua dell'interfaccia"
     },

client/ui/i18n/locales/ja/common.json

@@ -551,14 +551,6 @@
     "settings.general.autostart.errorTitle": {
         "message": "自動起動の変更に失敗しました"
     },
-    "settings.general.keepConnectedOnQuit.label": {
-        "message": "終了後も接続を維持",
-        "description": "Toggle label: keep the VPN connection up after quitting the UI."
-    },
-    "settings.general.keepConnectedOnQuit.help": {
-        "message": "NetBird を閉じたあとも接続はバックグラウンドで維持されます。自分で切断したときにだけ停止します。",
-        "description": "Helper text for the stay-connected-after-quitting toggle."
-    },
     "settings.general.language.label": {
         "message": "表示言語"
     },

client/ui/i18n/locales/pt/common.json

@@ -551,14 +551,6 @@
     "settings.general.autostart.errorTitle": {
         "message": "Falha ao alterar o início automático"
     },
-    "settings.general.keepConnectedOnQuit.label": {
-        "message": "Permanecer conectado ao sair",
-        "description": "Toggle label: keep the VPN connection up after quitting the UI."
-    },
-    "settings.general.keepConnectedOnQuit.help": {
-        "message": "A conexão continua ativa em segundo plano depois de fechar o NetBird. Ela só para quando você mesmo a desconecta.",
-        "description": "Helper text for the stay-connected-after-quitting toggle."
-    },
     "settings.general.language.label": {
         "message": "Idioma de exibição"
     },

client/ui/i18n/locales/ru/common.json

@@ -551,14 +551,6 @@
     "settings.general.autostart.errorTitle": {
         "message": "Не удалось изменить автозапуск"
     },
-    "settings.general.keepConnectedOnQuit.label": {
-        "message": "Оставаться подключённым после выхода",
-        "description": "Toggle label: keep the VPN connection up after quitting the UI."
-    },
-    "settings.general.keepConnectedOnQuit.help": {
-        "message": "Соединение остаётся активным в фоне после закрытия NetBird. Оно прервётся, только когда вы отключите его сами.",
-        "description": "Helper text for the stay-connected-after-quitting toggle."
-    },
     "settings.general.language.label": {
         "message": "Язык интерфейса"
     },

client/ui/i18n/locales/zh-CN/common.json

@@ -551,14 +551,6 @@
     "settings.general.autostart.errorTitle": {
         "message": "更改自启动设置失败"
     },
-    "settings.general.keepConnectedOnQuit.label": {
-        "message": "退出后保持连接",
-        "description": "Toggle label: keep the VPN connection up after quitting the UI."
-    },
-    "settings.general.keepConnectedOnQuit.help": {
-        "message": "关闭 NetBird 后，连接会在后台保持。只有你自己断开时才会停止。",
-        "description": "Helper text for the stay-connected-after-quitting toggle."
-    },
     "settings.general.language.label": {
         "message": "显示语言"
     },

client/ui/main.go

@@ -180,7 +180,6 @@ func main() {
 		WindowManager:   windowManager,
 		Session:         authSession,
 		Localizer:       localizer,
-		Preferences:     prefStore,
 	})
 	listenForShowSignal(context.Background(), tray)
 

client/ui/preferences/store.go

@@ -58,10 +58,6 @@ type UIPreferences struct {
 	// decision has run for this OS user. It only ever transitions to true
 	// and is never reset, so the default-on flow runs at most once, ever.
 	AutostartInitialized bool `json:"autostartInitialized"`
-	// KeepConnectedOnQuit leaves the daemon connected when the GUI quits.
-	// Its false zero value preserves the historical disconnect-on-quit
-	// behaviour for preference files written before the field existed.
-	KeepConnectedOnQuit bool `json:"keepConnectedOnQuit"`
 }
 
 // LanguageValidator rejects SetLanguage inputs with no shipped bundle.
@@ -187,26 +183,6 @@ func (s *Store) SetAutostartInitialized(done bool) error {
 	return nil
 }
 
-// SetKeepConnectedOnQuit persists the disconnect-on-quit opt-out. No-op if unchanged.
-func (s *Store) SetKeepConnectedOnQuit(keep bool) error {
-	s.mu.Lock()
-	if s.current.KeepConnectedOnQuit == keep {
-		s.mu.Unlock()
-		return nil
-	}
-	next := s.current
-	next.KeepConnectedOnQuit = keep
-	if err := s.persistLocked(next); err != nil {
-		s.mu.Unlock()
-		return fmt.Errorf("persist preferences: %w", err)
-	}
-	s.current = next
-	s.mu.Unlock()
-
-	s.broadcast(next)
-	return nil
-}
-
 // SetLanguage validates, persists, and broadcasts. No-op if unchanged.
 func (s *Store) SetLanguage(lang i18n.LanguageCode) error {
 	if lang == "" {

client/ui/preferences/store_test.go

@@ -238,42 +238,6 @@ func TestStore_SetAutostartInitializedPersistsAcrossReload(t *testing.T) {
 	assert.True(t, reloaded.Get().AutostartInitialized, "marker must survive a reload from disk")
 }
 
-func TestStore_SetKeepConnectedOnQuitPersistsAcrossReload(t *testing.T) {
-	withTempConfigDir(t)
-	emitter := &recordingEmitter{}
-	s, err := NewStore(nil, emitter)
-	require.NoError(t, err)
-
-	assert.False(t, s.Get().KeepConnectedOnQuit, "quitting must disconnect by default")
-
-	require.NoError(t, s.SetKeepConnectedOnQuit(true))
-	assert.True(t, s.Get().KeepConnectedOnQuit, "Get should reflect the persisted opt-out")
-	require.Len(t, emitter.calledWith(EventPreferencesChanged), 1, "first write should broadcast")
-
-	require.NoError(t, s.SetKeepConnectedOnQuit(true))
-	assert.Len(t, emitter.calledWith(EventPreferencesChanged), 1, "idempotent write should not broadcast again")
-
-	reloaded, err := NewStore(nil, nil)
-	require.NoError(t, err)
-	assert.True(t, reloaded.Get().KeepConnectedOnQuit, "opt-out must survive a reload from disk")
-}
-
-func TestStore_KeepConnectedOnQuitDefaultsFalseForPreExistingFile(t *testing.T) {
-	withTempConfigDir(t)
-
-	// A preferences file written before the field existed must keep the
-	// historical disconnect-on-quit behaviour rather than silently opting out.
-	path, err := preferencesPath()
-	require.NoError(t, err)
-	require.NoError(t, os.MkdirAll(filepath.Dir(path), 0o755))
-	require.NoError(t, os.WriteFile(path, []byte(`{"language":"en","viewMode":"default"}`), 0o600))
-
-	s, err := NewStore(nil, nil)
-	require.NoError(t, err)
-	assert.False(t, s.Get().KeepConnectedOnQuit, "a file predating the field must not opt out of disconnect-on-quit")
-	assert.True(t, s.ExistedAtLoad(), "the pre-existing file must be seen on disk")
-}
-
 func TestStore_ExistedAtLoad(t *testing.T) {
 	withTempConfigDir(t)
 

client/ui/services/cursor_linux.go

@@ -49,11 +49,5 @@ func getCursorPosition(app *application.App) (application.Point, bool) {
 	if app == nil || app.Screen == nil {
 		return p, true
 	}
-	// The wails GTK3 backend caches screens from the active window; a tray app
-	// has none at startup, so the cache is empty and PhysicalToDipPoint would
-	// dereference a nil nearest screen. Raw pixels are correct there anyway.
-	if app.Screen.ScreenNearestPhysicalPoint(p) == nil {
-		return p, true
-	}
 	return app.Screen.PhysicalToDipPoint(p), true
 }

client/ui/services/debug.go

@@ -15,10 +15,13 @@ import (
 )
 
 type DebugBundleParams struct {
-	Anonymize    bool   `json:"anonymize"`
-	SystemInfo   bool   `json:"systemInfo"`
-	UploadURL    string `json:"uploadUrl"`
-	LogFileCount uint32 `json:"logFileCount"`
+	Anonymize bool `json:"anonymize"`
+	// AnonymizeLevel is "default" or "strict"; strict also anonymizes
+	// private IP ranges, peer names, and WireGuard public keys.
+	AnonymizeLevel string `json:"anonymizeLevel"`
+	SystemInfo     bool   `json:"systemInfo"`
+	UploadURL      string `json:"uploadUrl"`
+	LogFileCount   uint32 `json:"logFileCount"`
 }
 
 // DebugBundleResult: Path is set for local-only bundles, UploadedKey on upload
@@ -48,11 +51,12 @@ func (s *Debug) Bundle(ctx context.Context, p DebugBundleParams) (DebugBundleRes
 		return DebugBundleResult{}, err
 	}
 	resp, err := cli.DebugBundle(ctx, &proto.DebugBundleRequest{
-		Anonymize:    p.Anonymize,
-		SystemInfo:   p.SystemInfo,
-		UploadURL:    p.UploadURL,
-		LogFileCount: p.LogFileCount,
-		CliVersion:   version.NetbirdVersion(),
+		Anonymize:      p.Anonymize,
+		AnonymizeLevel: p.AnonymizeLevel,
+		SystemInfo:     p.SystemInfo,
+		UploadURL:      p.UploadURL,
+		LogFileCount:   p.LogFileCount,
+		CliVersion:     version.NetbirdVersion(),
 	})
 	if err != nil {
 		return DebugBundleResult{}, err

client/ui/services/preferences.go

@@ -34,7 +34,3 @@ func (s *Preferences) SetViewMode(_ context.Context, mode preferences.ViewMode)
 func (s *Preferences) SetOnboardingCompleted(_ context.Context, done bool) error {
 	return s.store.SetOnboardingCompleted(done)
 }
-
-func (s *Preferences) SetKeepConnectedOnQuit(_ context.Context, keep bool) error {
-	return s.store.SetKeepConnectedOnQuit(keep)
-}

client/ui/tray.go

@@ -16,7 +16,6 @@ import (
 
 	"github.com/netbirdio/netbird/client/ui/authsession"
 	"github.com/netbirdio/netbird/client/ui/i18n"
-	"github.com/netbirdio/netbird/client/ui/preferences"
 	"github.com/netbirdio/netbird/client/ui/services"
 	"github.com/netbirdio/netbird/version"
 )
@@ -51,9 +50,8 @@ type TrayServices struct {
 	WindowManager   *services.WindowManager
 	// Session is bound to authsession directly because the services wrapper
 	// only re-exposes the React subset.
-	Session     *authsession.Session
-	Localizer   *Localizer
-	Preferences *preferences.Store
+	Session   *authsession.Session
+	Localizer *Localizer
 }
 
 type Tray struct {
@@ -463,12 +461,10 @@ func (t *Tray) handleQuit() {
 	t.profileMu.Unlock()
 	t.svc.DaemonFeed.CancelProfileSwitch()
 
-	if t.svc.Preferences == nil || !t.svc.Preferences.Get().KeepConnectedOnQuit {
-		ctx, cancel := context.WithTimeout(context.Background(), quitDownTimeout)
-		defer cancel()
-		if err := t.svc.Connection.Down(ctx); err != nil {
-			log.Errorf("disconnect on quit: %v", err)
-		}
+	ctx, cancel := context.WithTimeout(context.Background(), quitDownTimeout)
+	defer cancel()
+	if err := t.svc.Connection.Down(ctx); err != nil {
+		log.Errorf("disconnect on quit: %v", err)
 	}
 	t.app.Quit()
 }

client/ui/xembed_host_gtk3_linux.go

@@ -1,40 +0,0 @@
-//go:build linux && gtk3 && !(linux && 386)
-
-package main
-
-import (
-	"errors"
-
-	"github.com/godbus/dbus/v5"
-)
-
-// The legacy GTK3 / WebKit2GTK 4.1 build (-tags gtk3) drops the in-process
-// XEmbed StatusNotifierWatcher entirely. The real implementation
-// (xembed_host_linux.go + xembed_tray_linux.c) links GTK4 and uses GTK4-only
-// popup-menu APIs that have no drop-in GTK3 equivalent, so rather than port the
-// C layer we stub the host out on gtk3 builds. The tray still works on every
-// desktop that ships its own StatusNotifierWatcher (KDE, GNOME+AppIndicator,
-// Cinnamon/xapp, XFCE, …); only the minimal-WM fallback (Fluxbox/OpenBox/i3/
-// dwm/vanilla GNOME) is unavailable on gtk3 packages. See LINUX-TRAY.md.
-
-// xembedHost is a placeholder so the package compiles on gtk3 builds; the real
-// type (with X11/GTK4 state) lives in xembed_host_linux.go. It is never
-// instantiated here because xembedTrayAvailable always reports false.
-type xembedHost struct{}
-
-// run satisfies the call in tray_watcher_linux.go; unreachable on gtk3 because
-// newXembedHost never returns a non-nil host.
-func (*xembedHost) run() {}
-
-// xembedTrayAvailable always reports false on gtk3 builds, so the watcher probe
-// loop in startStatusNotifierWatcher exits immediately and newXembedHost is
-// never reached. recenter_linux.go's predicate becomes a harmless no-op too.
-func xembedTrayAvailable() bool {
-	return false
-}
-
-// newXembedHost exists only to satisfy the reference in tray_watcher_linux.go;
-// it is unreachable because xembedTrayAvailable returns false on gtk3.
-func newXembedHost(conn *dbus.Conn, busName string, objPath dbus.ObjectPath) (*xembedHost, error) {
-	return nil, errors.New("xembed host unsupported on gtk3 build")
-}

client/ui/xembed_host_linux.go

@@ -1,4 +1,4 @@
-//go:build linux && !gtk3 && !(linux && 386)
+//go:build linux && !(linux && 386)
 
 package main
 

client/ui/xembed_tray_linux.c

@@ -1,5 +1,3 @@
-//go:build linux && !gtk3 && !(linux && 386)
-
 #include "xembed_tray_linux.h"
 
 #include <X11/Xatom.h>

infrastructure_files/getting-started-enterprise.sh

@@ -262,7 +262,6 @@ init_environment() {
   POSTGRES_DB="netbird"
   POSTGRES_PASSWORD=$(rand_secret)
   NETBIRD_ENCRYPTION_KEY=$(rand_b64_key)
-  NETBIRD_SESSION_COOKIE_ENCRYPTION_KEY=$(rand_b64_key)
   NETBIRD_RELAY_AUTH_SECRET=$(rand_secret)
 
   POSTGRES_DSN="host=postgres user=${POSTGRES_USER} password=${POSTGRES_PASSWORD} dbname=${POSTGRES_DB} port=5432 sslmode=disable TimeZone=UTC"
@@ -697,7 +696,6 @@ server:
     issuer: "https://${NETBIRD_DOMAIN}/oauth2"
     localAuthDisabled: false
     signKeyRefreshEnabled: false
-    sessionCookieEncryptionKey: "${NETBIRD_SESSION_COOKIE_ENCRYPTION_KEY}"
     dashboardRedirectURIs:
       - "https://${NETBIRD_DOMAIN}/nb-auth"
       - "https://${NETBIRD_DOMAIN}/nb-silent-auth"

infrastructure_files/getting-started.sh

@@ -348,7 +348,6 @@ initialize_default_values() {
   NETBIRD_RELAY_AUTH_SECRET=$(openssl rand -base64 32 | sed "$SED_STRIP_PADDING")
   # Note: DataStoreEncryptionKey must keep base64 padding (=) for Go's base64.StdEncoding
   DATASTORE_ENCRYPTION_KEY=$(openssl rand -base64 32)
-  SESSION_COOKIE_ENCRYPTION_KEY=$(openssl rand -base64 32)
   NETBIRD_STUN_PORT=3478
 
   # Docker images
@@ -528,8 +527,7 @@ generate_configuration_files() {
 
   # Common files for all configurations
   render_dashboard_env > dashboard.env
-  install -m 600 /dev/null config.yaml
-  render_combined_yaml >> config.yaml
+  render_combined_yaml > config.yaml
   return 0
 }
 
@@ -913,7 +911,6 @@ server:
   auth:
     issuer: "$NETBIRD_HTTP_PROTOCOL://$NETBIRD_DOMAIN/oauth2"
     signKeyRefreshEnabled: true
-    sessionCookieEncryptionKey: "$SESSION_COOKIE_ENCRYPTION_KEY"
     dashboardRedirectURIs:
       - "$NETBIRD_HTTP_PROTOCOL://$NETBIRD_DOMAIN/nb-auth"
       - "$NETBIRD_HTTP_PROTOCOL://$NETBIRD_DOMAIN/nb-silent-auth"

release_files/darwin_pkg/preinstall

@@ -4,67 +4,9 @@ set -x
 
 LOG_FILE=/var/log/netbird/client_pre_install.log
 AGENT=/usr/local/bin/netbird
-UI_PROCESS=netbird-ui
 
 mkdir -p /var/log/netbird/
 
-# wait_for_ui_exit polls for up to $1 seconds, returning 0 as soon as no UI
-# process is left and 1 if one is still running when the time is up.
-wait_for_ui_exit() {
-    waited=0
-    while [ "$waited" -lt "$1" ]; do
-        pgrep -x "$UI_PROCESS" > /dev/null 2>&1 || return 0
-        sleep 1
-        waited=$((waited + 1))
-    done
-    return 1
-}
-
-# request_ui_quit asks the UI to quit from inside the console user's session and
-# reports whether the request could be sent at all. The installer runs as root
-# outside that session, so a quit Apple event sent straight from here always
-# fails with -600.
-request_ui_quit() {
-    console_user=$(stat -f%Su /dev/console 2>/dev/null)
-    case "$console_user" in
-        ""|root|loginwindow|_mbsetupuser)
-            echo "No active GUI user session (console user: '${console_user:-none}'); skipping the quit request."
-            return 1
-            ;;
-    esac
-
-    uid=$(id -u "$console_user" 2>/dev/null)
-    if [ -z "$uid" ]; then
-        echo "Could not resolve uid for console user '$console_user'; skipping the quit request."
-        return 1
-    fi
-
-    echo "Asking the NetBird UI to quit as console user $console_user (uid $uid)."
-    launchctl asuser "$uid" sudo -u "$console_user" -H osascript -e 'quit app "NetBird"' || true
-}
-
-# quit_ui stops a running UI so the app bundle can be replaced underneath it. A
-# UI process that survives the install keeps serving the old binary until it is
-# quit by hand, so anything still running once the quit request is out of the
-# way is signalled. Waiting for a graceful exit only makes sense when a quit
-# request was actually sent.
-quit_ui() {
-    if request_ui_quit && wait_for_ui_exit 10; then
-        return 0
-    fi
-
-    pgrep -x "$UI_PROCESS" > /dev/null 2>&1 || return 0
-
-    echo "NetBird UI still running; terminating it."
-    pkill -x "$UI_PROCESS" || true
-    if wait_for_ui_exit 3; then
-        return 0
-    fi
-
-    echo "NetBird UI ignored SIGTERM; killing it."
-    pkill -KILL -x "$UI_PROCESS" || true
-}
-
 {
     # check if it was installed with brew
     brew list --formula | grep netbird
@@ -73,9 +15,10 @@ quit_ui() {
         echo "NetBird has been installed with Brew. Please use Brew to update the package."
         exit 1
     fi
-    quit_ui
+    osascript -e 'quit app "Netbird"' || true
     $AGENT service stop || true
 
     echo "Preinstall complete"
     exit 0 # all good
 } &> $LOG_FILE
+

shared/management/proto/management.proto

@@ -110,6 +110,10 @@ message BundleParameters {
   int64  bundle_for_time  = 2;
   int32  log_file_count   = 3;
   bool   anonymize        = 4;
+  // anonymize_level selects how much the anonymizer redacts: "default"
+  // (or empty) keeps internal IP ranges, "strict" also anonymizes them.
+  // Unknown values are treated as "strict".
+  string anonymize_level  = 5;
 }
 
 message BundleResult {

shared/signal/client/client.go

@@ -52,11 +52,6 @@ type CredentialPayload struct {
 	Credential      *Credential
 	RosenpassPubKey []byte
 	RosenpassAddr   string
-	// MlkemPayload is the opaque post-quantum KEM handshake message riding this
-	// OFFER/ANSWER (see Body.mlkemPayload). Nil when not running the PQ exchange.
-	MlkemPayload []byte
-	// MlkemPort is the sender's ML-KEM PQ service UDP port (0 when not running).
-	MlkemPort       int
 	RelaySrvAddress string
 	RelaySrvIP      netip.Addr
 	SessionID       []byte
@@ -94,13 +89,6 @@ func MarshalCredential(myKey wgtypes.Key, remoteKey string, p CredentialPayload)
 	if p.RelaySrvIP.IsValid() {
 		body.RelayServerIP = p.RelaySrvIP.Unmap().AsSlice()
 	}
-	if len(p.MlkemPayload) > 0 {
-		body.MlkemPayload = p.MlkemPayload
-	}
-	if p.MlkemPort > 0 {
-		port := uint32(p.MlkemPort)
-		body.MlkemPort = &port
-	}
 	return &proto.Message{
 		Key:       myKey.PublicKey().String(),
 		RemoteKey: remoteKey,

shared/signal/proto/signalexchange.pb.go

@@ -239,16 +239,6 @@ type Body struct {
 	// fallback dial target when DNS resolution of relayServerAddress fails.
 	// SNI/TLS verification still uses relayServerAddress.
 	RelayServerIP []byte `protobuf:"bytes,11,opt,name=relayServerIP,proto3,oneof" json:"relayServerIP,omitempty"`
-	// mlkemPayload carries a post-quantum X25519MLKEM768 handshake message that
-	// seeds the WireGuard PSK, riding this Body's OFFER/ANSWER: on an OFFER it is
-	// the KEM offer, on an ANSWER the KEM answer. It is opaque to signal — the
-	// pqkem library frames and parses it. Absent when the sender does not run the
-	// ML-KEM PQ exchange; unknown to older clients, which ignore it.
-	MlkemPayload []byte `protobuf:"bytes,12,opt,name=mlkemPayload,proto3,oneof" json:"mlkemPayload,omitempty"`
-	// mlkemPort is the UDP port of the sender's ML-KEM PQ service, bound on its
-	// WireGuard overlay IP. Peers send subsequent rekey messages there over the
-	// data path. Zero/absent when the ML-KEM PQ exchange is not running.
-	MlkemPort *uint32 `protobuf:"varint,13,opt,name=mlkemPort,proto3,oneof" json:"mlkemPort,omitempty"`
 }
 
 func (x *Body) Reset() {
@@ -353,20 +343,6 @@ func (x *Body) GetRelayServerIP() []byte {
 	return nil
 }
 
-func (x *Body) GetMlkemPayload() []byte {
-	if x != nil {
-		return x.MlkemPayload
-	}
-	return nil
-}
-
-func (x *Body) GetMlkemPort() uint32 {
-	if x != nil && x.MlkemPort != nil {
-		return *x.MlkemPort
-	}
-	return 0
-}
-
 // Mode indicates a connection mode
 type Mode struct {
 	state         protoimpl.MessageState
@@ -490,7 +466,7 @@ var file_signalexchange_proto_rawDesc = []byte{
 	0x52, 0x09, 0x72, 0x65, 0x6d, 0x6f, 0x74, 0x65, 0x4b, 0x65, 0x79, 0x12, 0x28, 0x0a, 0x04, 0x62,
 	0x6f, 0x64, 0x79, 0x18, 0x04, 0x20, 0x01, 0x28, 0x0b, 0x32, 0x14, 0x2e, 0x73, 0x69, 0x67, 0x6e,
 	0x61, 0x6c, 0x65, 0x78, 0x63, 0x68, 0x61, 0x6e, 0x67, 0x65, 0x2e, 0x42, 0x6f, 0x64, 0x79, 0x52,
-	0x04, 0x62, 0x6f, 0x64, 0x79, 0x22, 0xbd, 0x05, 0x0a, 0x04, 0x42, 0x6f, 0x64, 0x79, 0x12, 0x2d,
+	0x04, 0x62, 0x6f, 0x64, 0x79, 0x22, 0xd2, 0x04, 0x0a, 0x04, 0x42, 0x6f, 0x64, 0x79, 0x12, 0x2d,
 	0x0a, 0x04, 0x74, 0x79, 0x70, 0x65, 0x18, 0x01, 0x20, 0x01, 0x28, 0x0e, 0x32, 0x19, 0x2e, 0x73,
 	0x69, 0x67, 0x6e, 0x61, 0x6c, 0x65, 0x78, 0x63, 0x68, 0x61, 0x6e, 0x67, 0x65, 0x2e, 0x42, 0x6f,
 	0x64, 0x79, 0x2e, 0x54, 0x79, 0x70, 0x65, 0x52, 0x04, 0x74, 0x79, 0x70, 0x65, 0x12, 0x18, 0x0a,
@@ -518,46 +494,39 @@ var file_signalexchange_proto_rawDesc = []byte{
 	0x52, 0x09, 0x73, 0x65, 0x73, 0x73, 0x69, 0x6f, 0x6e, 0x49, 0x64, 0x88, 0x01, 0x01, 0x12, 0x29,
 	0x0a, 0x0d, 0x72, 0x65, 0x6c, 0x61, 0x79, 0x53, 0x65, 0x72, 0x76, 0x65, 0x72, 0x49, 0x50, 0x18,
 	0x0b, 0x20, 0x01, 0x28, 0x0c, 0x48, 0x02, 0x52, 0x0d, 0x72, 0x65, 0x6c, 0x61, 0x79, 0x53, 0x65,
-	0x72, 0x76, 0x65, 0x72, 0x49, 0x50, 0x88, 0x01, 0x01, 0x12, 0x27, 0x0a, 0x0c, 0x6d, 0x6c, 0x6b,
-	0x65, 0x6d, 0x50, 0x61, 0x79, 0x6c, 0x6f, 0x61, 0x64, 0x18, 0x0c, 0x20, 0x01, 0x28, 0x0c, 0x48,
-	0x03, 0x52, 0x0c, 0x6d, 0x6c, 0x6b, 0x65, 0x6d, 0x50, 0x61, 0x79, 0x6c, 0x6f, 0x61, 0x64, 0x88,
-	0x01, 0x01, 0x12, 0x21, 0x0a, 0x09, 0x6d, 0x6c, 0x6b, 0x65, 0x6d, 0x50, 0x6f, 0x72, 0x74, 0x18,
-	0x0d, 0x20, 0x01, 0x28, 0x0d, 0x48, 0x04, 0x52, 0x09, 0x6d, 0x6c, 0x6b, 0x65, 0x6d, 0x50, 0x6f,
-	0x72, 0x74, 0x88, 0x01, 0x01, 0x22, 0x52, 0x0a, 0x04, 0x54, 0x79, 0x70, 0x65, 0x12, 0x09, 0x0a,
-	0x05, 0x4f, 0x46, 0x46, 0x45, 0x52, 0x10, 0x00, 0x12, 0x0a, 0x0a, 0x06, 0x41, 0x4e, 0x53, 0x57,
-	0x45, 0x52, 0x10, 0x01, 0x12, 0x0d, 0x0a, 0x09, 0x43, 0x41, 0x4e, 0x44, 0x49, 0x44, 0x41, 0x54,
-	0x45, 0x10, 0x02, 0x12, 0x08, 0x0a, 0x04, 0x4d, 0x4f, 0x44, 0x45, 0x10, 0x04, 0x12, 0x0b, 0x0a,
-	0x07, 0x47, 0x4f, 0x5f, 0x49, 0x44, 0x4c, 0x45, 0x10, 0x05, 0x12, 0x0d, 0x0a, 0x09, 0x48, 0x45,
-	0x41, 0x52, 0x54, 0x42, 0x45, 0x41, 0x54, 0x10, 0x06, 0x42, 0x15, 0x0a, 0x13, 0x5f, 0x72, 0x65,
-	0x6c, 0x61, 0x79, 0x53, 0x65, 0x72, 0x76, 0x65, 0x72, 0x41, 0x64, 0x64, 0x72, 0x65, 0x73, 0x73,
-	0x42, 0x0c, 0x0a, 0x0a, 0x5f, 0x73, 0x65, 0x73, 0x73, 0x69, 0x6f, 0x6e, 0x49, 0x64, 0x42, 0x10,
-	0x0a, 0x0e, 0x5f, 0x72, 0x65, 0x6c, 0x61, 0x79, 0x53, 0x65, 0x72, 0x76, 0x65, 0x72, 0x49, 0x50,
-	0x42, 0x0f, 0x0a, 0x0d, 0x5f, 0x6d, 0x6c, 0x6b, 0x65, 0x6d, 0x50, 0x61, 0x79, 0x6c, 0x6f, 0x61,
-	0x64, 0x42, 0x0c, 0x0a, 0x0a, 0x5f, 0x6d, 0x6c, 0x6b, 0x65, 0x6d, 0x50, 0x6f, 0x72, 0x74, 0x4a,
-	0x04, 0x08, 0x09, 0x10, 0x0a, 0x22, 0x2e, 0x0a, 0x04, 0x4d, 0x6f, 0x64, 0x65, 0x12, 0x1b, 0x0a,
-	0x06, 0x64, 0x69, 0x72, 0x65, 0x63, 0x74, 0x18, 0x01, 0x20, 0x01, 0x28, 0x08, 0x48, 0x00, 0x52,
-	0x06, 0x64, 0x69, 0x72, 0x65, 0x63, 0x74, 0x88, 0x01, 0x01, 0x42, 0x09, 0x0a, 0x07, 0x5f, 0x64,
-	0x69, 0x72, 0x65, 0x63, 0x74, 0x22, 0x6d, 0x0a, 0x0f, 0x52, 0x6f, 0x73, 0x65, 0x6e, 0x70, 0x61,
-	0x73, 0x73, 0x43, 0x6f, 0x6e, 0x66, 0x69, 0x67, 0x12, 0x28, 0x0a, 0x0f, 0x72, 0x6f, 0x73, 0x65,
-	0x6e, 0x70, 0x61, 0x73, 0x73, 0x50, 0x75, 0x62, 0x4b, 0x65, 0x79, 0x18, 0x01, 0x20, 0x01, 0x28,
-	0x0c, 0x52, 0x0f, 0x72, 0x6f, 0x73, 0x65, 0x6e, 0x70, 0x61, 0x73, 0x73, 0x50, 0x75, 0x62, 0x4b,
-	0x65, 0x79, 0x12, 0x30, 0x0a, 0x13, 0x72, 0x6f, 0x73, 0x65, 0x6e, 0x70, 0x61, 0x73, 0x73, 0x53,
-	0x65, 0x72, 0x76, 0x65, 0x72, 0x41, 0x64, 0x64, 0x72, 0x18, 0x02, 0x20, 0x01, 0x28, 0x09, 0x52,
-	0x13, 0x72, 0x6f, 0x73, 0x65, 0x6e, 0x70, 0x61, 0x73, 0x73, 0x53, 0x65, 0x72, 0x76, 0x65, 0x72,
-	0x41, 0x64, 0x64, 0x72, 0x32, 0xb9, 0x01, 0x0a, 0x0e, 0x53, 0x69, 0x67, 0x6e, 0x61, 0x6c, 0x45,
-	0x78, 0x63, 0x68, 0x61, 0x6e, 0x67, 0x65, 0x12, 0x4c, 0x0a, 0x04, 0x53, 0x65, 0x6e, 0x64, 0x12,
-	0x20, 0x2e, 0x73, 0x69, 0x67, 0x6e, 0x61, 0x6c, 0x65, 0x78, 0x63, 0x68, 0x61, 0x6e, 0x67, 0x65,
-	0x2e, 0x45, 0x6e, 0x63, 0x72, 0x79, 0x70, 0x74, 0x65, 0x64, 0x4d, 0x65, 0x73, 0x73, 0x61, 0x67,
-	0x65, 0x1a, 0x20, 0x2e, 0x73, 0x69, 0x67, 0x6e, 0x61, 0x6c, 0x65, 0x78, 0x63, 0x68, 0x61, 0x6e,
-	0x67, 0x65, 0x2e, 0x45, 0x6e, 0x63, 0x72, 0x79, 0x70, 0x74, 0x65, 0x64, 0x4d, 0x65, 0x73, 0x73,
-	0x61, 0x67, 0x65, 0x22, 0x00, 0x12, 0x59, 0x0a, 0x0d, 0x43, 0x6f, 0x6e, 0x6e, 0x65, 0x63, 0x74,
-	0x53, 0x74, 0x72, 0x65, 0x61, 0x6d, 0x12, 0x20, 0x2e, 0x73, 0x69, 0x67, 0x6e, 0x61, 0x6c, 0x65,
+	0x72, 0x76, 0x65, 0x72, 0x49, 0x50, 0x88, 0x01, 0x01, 0x22, 0x52, 0x0a, 0x04, 0x54, 0x79, 0x70,
+	0x65, 0x12, 0x09, 0x0a, 0x05, 0x4f, 0x46, 0x46, 0x45, 0x52, 0x10, 0x00, 0x12, 0x0a, 0x0a, 0x06,
+	0x41, 0x4e, 0x53, 0x57, 0x45, 0x52, 0x10, 0x01, 0x12, 0x0d, 0x0a, 0x09, 0x43, 0x41, 0x4e, 0x44,
+	0x49, 0x44, 0x41, 0x54, 0x45, 0x10, 0x02, 0x12, 0x08, 0x0a, 0x04, 0x4d, 0x4f, 0x44, 0x45, 0x10,
+	0x04, 0x12, 0x0b, 0x0a, 0x07, 0x47, 0x4f, 0x5f, 0x49, 0x44, 0x4c, 0x45, 0x10, 0x05, 0x12, 0x0d,
+	0x0a, 0x09, 0x48, 0x45, 0x41, 0x52, 0x54, 0x42, 0x45, 0x41, 0x54, 0x10, 0x06, 0x42, 0x15, 0x0a,
+	0x13, 0x5f, 0x72, 0x65, 0x6c, 0x61, 0x79, 0x53, 0x65, 0x72, 0x76, 0x65, 0x72, 0x41, 0x64, 0x64,
+	0x72, 0x65, 0x73, 0x73, 0x42, 0x0c, 0x0a, 0x0a, 0x5f, 0x73, 0x65, 0x73, 0x73, 0x69, 0x6f, 0x6e,
+	0x49, 0x64, 0x42, 0x10, 0x0a, 0x0e, 0x5f, 0x72, 0x65, 0x6c, 0x61, 0x79, 0x53, 0x65, 0x72, 0x76,
+	0x65, 0x72, 0x49, 0x50, 0x4a, 0x04, 0x08, 0x09, 0x10, 0x0a, 0x22, 0x2e, 0x0a, 0x04, 0x4d, 0x6f,
+	0x64, 0x65, 0x12, 0x1b, 0x0a, 0x06, 0x64, 0x69, 0x72, 0x65, 0x63, 0x74, 0x18, 0x01, 0x20, 0x01,
+	0x28, 0x08, 0x48, 0x00, 0x52, 0x06, 0x64, 0x69, 0x72, 0x65, 0x63, 0x74, 0x88, 0x01, 0x01, 0x42,
+	0x09, 0x0a, 0x07, 0x5f, 0x64, 0x69, 0x72, 0x65, 0x63, 0x74, 0x22, 0x6d, 0x0a, 0x0f, 0x52, 0x6f,
+	0x73, 0x65, 0x6e, 0x70, 0x61, 0x73, 0x73, 0x43, 0x6f, 0x6e, 0x66, 0x69, 0x67, 0x12, 0x28, 0x0a,
+	0x0f, 0x72, 0x6f, 0x73, 0x65, 0x6e, 0x70, 0x61, 0x73, 0x73, 0x50, 0x75, 0x62, 0x4b, 0x65, 0x79,
+	0x18, 0x01, 0x20, 0x01, 0x28, 0x0c, 0x52, 0x0f, 0x72, 0x6f, 0x73, 0x65, 0x6e, 0x70, 0x61, 0x73,
+	0x73, 0x50, 0x75, 0x62, 0x4b, 0x65, 0x79, 0x12, 0x30, 0x0a, 0x13, 0x72, 0x6f, 0x73, 0x65, 0x6e,
+	0x70, 0x61, 0x73, 0x73, 0x53, 0x65, 0x72, 0x76, 0x65, 0x72, 0x41, 0x64, 0x64, 0x72, 0x18, 0x02,
+	0x20, 0x01, 0x28, 0x09, 0x52, 0x13, 0x72, 0x6f, 0x73, 0x65, 0x6e, 0x70, 0x61, 0x73, 0x73, 0x53,
+	0x65, 0x72, 0x76, 0x65, 0x72, 0x41, 0x64, 0x64, 0x72, 0x32, 0xb9, 0x01, 0x0a, 0x0e, 0x53, 0x69,
+	0x67, 0x6e, 0x61, 0x6c, 0x45, 0x78, 0x63, 0x68, 0x61, 0x6e, 0x67, 0x65, 0x12, 0x4c, 0x0a, 0x04,
+	0x53, 0x65, 0x6e, 0x64, 0x12, 0x20, 0x2e, 0x73, 0x69, 0x67, 0x6e, 0x61, 0x6c, 0x65, 0x78, 0x63,
+	0x68, 0x61, 0x6e, 0x67, 0x65, 0x2e, 0x45, 0x6e, 0x63, 0x72, 0x79, 0x70, 0x74, 0x65, 0x64, 0x4d,
+	0x65, 0x73, 0x73, 0x61, 0x67, 0x65, 0x1a, 0x20, 0x2e, 0x73, 0x69, 0x67, 0x6e, 0x61, 0x6c, 0x65,
 	0x78, 0x63, 0x68, 0x61, 0x6e, 0x67, 0x65, 0x2e, 0x45, 0x6e, 0x63, 0x72, 0x79, 0x70, 0x74, 0x65,
-	0x64, 0x4d, 0x65, 0x73, 0x73, 0x61, 0x67, 0x65, 0x1a, 0x20, 0x2e, 0x73, 0x69, 0x67, 0x6e, 0x61,
-	0x6c, 0x65, 0x78, 0x63, 0x68, 0x61, 0x6e, 0x67, 0x65, 0x2e, 0x45, 0x6e, 0x63, 0x72, 0x79, 0x70,
-	0x74, 0x65, 0x64, 0x4d, 0x65, 0x73, 0x73, 0x61, 0x67, 0x65, 0x22, 0x00, 0x28, 0x01, 0x30, 0x01,
-	0x42, 0x08, 0x5a, 0x06, 0x2f, 0x70, 0x72, 0x6f, 0x74, 0x6f, 0x62, 0x06, 0x70, 0x72, 0x6f, 0x74,
-	0x6f, 0x33,
+	0x64, 0x4d, 0x65, 0x73, 0x73, 0x61, 0x67, 0x65, 0x22, 0x00, 0x12, 0x59, 0x0a, 0x0d, 0x43, 0x6f,
+	0x6e, 0x6e, 0x65, 0x63, 0x74, 0x53, 0x74, 0x72, 0x65, 0x61, 0x6d, 0x12, 0x20, 0x2e, 0x73, 0x69,
+	0x67, 0x6e, 0x61, 0x6c, 0x65, 0x78, 0x63, 0x68, 0x61, 0x6e, 0x67, 0x65, 0x2e, 0x45, 0x6e, 0x63,
+	0x72, 0x79, 0x70, 0x74, 0x65, 0x64, 0x4d, 0x65, 0x73, 0x73, 0x61, 0x67, 0x65, 0x1a, 0x20, 0x2e,
+	0x73, 0x69, 0x67, 0x6e, 0x61, 0x6c, 0x65, 0x78, 0x63, 0x68, 0x61, 0x6e, 0x67, 0x65, 0x2e, 0x45,
+	0x6e, 0x63, 0x72, 0x79, 0x70, 0x74, 0x65, 0x64, 0x4d, 0x65, 0x73, 0x73, 0x61, 0x67, 0x65, 0x22,
+	0x00, 0x28, 0x01, 0x30, 0x01, 0x42, 0x08, 0x5a, 0x06, 0x2f, 0x70, 0x72, 0x6f, 0x74, 0x6f, 0x62,
+	0x06, 0x70, 0x72, 0x6f, 0x74, 0x6f, 0x33,
 }
 
 var (

shared/signal/proto/signalexchange.proto

@@ -75,18 +75,6 @@ message Body {
   // fallback dial target when DNS resolution of relayServerAddress fails.
   // SNI/TLS verification still uses relayServerAddress.
   optional bytes relayServerIP = 11;
-
-  // mlkemPayload carries a post-quantum X25519MLKEM768 handshake message that
-  // seeds the WireGuard PSK, riding this Body's OFFER/ANSWER: on an OFFER it is
-  // the KEM offer, on an ANSWER the KEM answer. It is opaque to signal — the
-  // pqkem library frames and parses it. Absent when the sender does not run the
-  // ML-KEM PQ exchange; unknown to older clients, which ignore it.
-  optional bytes mlkemPayload = 12;
-
-  // mlkemPort is the UDP port of the sender's ML-KEM PQ service, bound on its
-  // WireGuard overlay IP. Peers send subsequent rekey messages there over the
-  // data path. Zero/absent when the ML-KEM PQ exchange is not running.
-  optional uint32 mlkemPort = 13;
 }
 
 // Mode indicates a connection mode