chore: bump version to 0.13.1

fix: accept tls:// and https:// in [[forwarding]] upstreams

.SRCINFO

@@ -0,0 +1,19 @@
+pkgbase = numa-git
+	pkgdesc = Portable DNS resolver in Rust — .numa local domains, ad blocking, developer overrides, DNS-over-HTTPS
+	pkgver = 0.10.1.r0.g0000000
+	pkgrel = 1
+	url = https://github.com/razvandimescu/numa
+	arch = x86_64
+	license = MIT
+	options = !lto
+	makedepends = cargo
+	makedepends = git
+	depends = gcc-libs
+	depends = glibc
+	provides = numa
+	conflicts = numa
+	backup = etc/numa.toml
+	source = numa::git+https://github.com/razvandimescu/numa.git
+	sha256sums = SKIP
+
+pkgname = numa-git

.github/dependabot.yml

@@ -0,0 +1,34 @@
+version: 2
+updates:
+  - package-ecosystem: "cargo"
+    directory: "/"
+    schedule:
+      interval: "monthly"
+    commit-message:
+      prefix: "chore(deps)"
+    groups:
+      minor-and-patch:
+        patterns: ["*"]
+        update-types: ["minor", "patch"]
+
+  - package-ecosystem: "github-actions"
+    directory: "/"
+    schedule:
+      interval: "monthly"
+    commit-message:
+      prefix: "chore(deps)"
+    groups:
+      minor-and-patch:
+        patterns: ["*"]
+        update-types: ["minor", "patch"]
+
+  - package-ecosystem: "docker"
+    directory: "/"
+    schedule:
+      interval: "monthly"
+    commit-message:
+      prefix: "chore(deps)"
+    groups:
+      minor-and-patch:
+        patterns: ["*"]
+        update-types: ["minor", "patch"]

.github/workflows/ci.yml

@@ -3,8 +3,22 @@ name: CI
 on:
   push:
     branches: [main]
+    paths-ignore:
+      - 'site/**'
+      - 'blog/**'
+      - 'drafts/**'
+      - '*.md'
+      - 'scripts/serve-site.sh'
+      - 'scripts/generate-blog-index.sh'
   pull_request:
     branches: [main]
+    paths-ignore:
+      - 'site/**'
+      - 'blog/**'
+      - 'drafts/**'
+      - '*.md'
+      - 'scripts/serve-site.sh'
+      - 'scripts/generate-blog-index.sh'
 
 env:
   CARGO_TERM_COLOR: always
@@ -13,7 +27,7 @@ jobs:
   check:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
       - uses: dtolnay/rust-toolchain@stable
         with:
           components: rustfmt, clippy
@@ -30,7 +44,7 @@ jobs:
   check-macos:
     runs-on: macos-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
       - uses: dtolnay/rust-toolchain@stable
       - uses: Swatinem/rust-cache@v2
       - name: clippy
@@ -41,7 +55,7 @@ jobs:
   check-windows:
     runs-on: windows-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
       - uses: dtolnay/rust-toolchain@stable
       - uses: Swatinem/rust-cache@v2
       - name: build
@@ -51,7 +65,7 @@ jobs:
       - name: test
         run: cargo test
       - name: Upload binary
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v7
         with:
           name: numa-windows-x86_64
           path: target/debug/numa.exe

.github/workflows/docker.yml

@@ -0,0 +1,45 @@
+name: Docker
+
+on:
+  push:
+    tags:
+      - 'v*'
+
+permissions:
+  contents: read
+  packages: write
+
+jobs:
+  docker:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+
+      - uses: docker/setup-qemu-action@v3
+
+      - uses: docker/setup-buildx-action@v3
+
+      - uses: docker/login-action@v3
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - uses: docker/metadata-action@v5
+        id: meta
+        with:
+          images: ghcr.io/${{ github.repository }}
+          tags: |
+            type=semver,pattern={{version}}
+            type=semver,pattern={{major}}.{{minor}}
+            type=raw,value=latest
+
+      - uses: docker/build-push-action@v6
+        with:
+          context: .
+          platforms: linux/amd64,linux/arm64
+          push: true
+          tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.meta.outputs.labels }}
+          cache-from: type=gha
+          cache-to: type=gha,mode=max

.github/workflows/homebrew-bump.yml

@@ -0,0 +1,77 @@
+name: Bump Homebrew Tap
+
+on:
+  workflow_call:
+    inputs:
+      version:
+        description: 'Version to bump (e.g. 0.10.0 or v0.10.0)'
+        type: string
+        required: true
+  workflow_dispatch:
+    inputs:
+      version:
+        description: 'Version to bump (e.g. 0.10.0 or v0.10.0)'
+        required: true
+
+permissions:
+  contents: read
+
+jobs:
+  bump:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+
+      - name: Determine version
+        id: ver
+        env:
+          INPUT_VERSION: ${{ inputs.version }}
+        run: |
+          V="${INPUT_VERSION#v}"
+          echo "version=$V" >> "$GITHUB_OUTPUT"
+
+      - name: Fetch sha256 checksums from release assets
+        id: shas
+        env:
+          V: ${{ steps.ver.outputs.version }}
+        run: |
+          set -euo pipefail
+          base="https://github.com/razvandimescu/numa/releases/download/v${V}"
+          for t in macos-aarch64 macos-x86_64 linux-aarch64 linux-x86_64; do
+            sha=$(curl -fsSL "${base}/numa-${t}.tar.gz.sha256" | awk '{print $1}')
+            if [ -z "$sha" ]; then
+              echo "ERROR: failed to fetch sha256 for $t" >&2
+              exit 1
+            fi
+            key=$(echo "$t" | tr '[:lower:]-' '[:upper:]_')
+            echo "SHA_${key}=${sha}" >> "$GITHUB_ENV"
+          done
+
+      - name: Clone homebrew-tap
+        env:
+          HOMEBREW_TAP_GITHUB_TOKEN: ${{ secrets.HOMEBREW_TAP_GITHUB_TOKEN }}
+        run: |
+          git clone "https://x-access-token:${HOMEBREW_TAP_GITHUB_TOKEN}@github.com/razvandimescu/homebrew-tap.git" tap
+
+      - name: Update formula
+        env:
+          VERSION: ${{ steps.ver.outputs.version }}
+        run: |
+          python3 scripts/update-homebrew-formula.py tap/numa.rb
+          echo "--- updated numa.rb ---"
+          cat tap/numa.rb
+
+      - name: Commit and push
+        working-directory: tap
+        env:
+          V: ${{ steps.ver.outputs.version }}
+        run: |
+          if git diff --quiet; then
+            echo "numa.rb already at v${V}, nothing to commit"
+            exit 0
+          fi
+          git config user.name "github-actions[bot]"
+          git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
+          git add numa.rb
+          git commit -m "chore: bump numa to v${V}"
+          git push origin main

.github/workflows/publish-aur.yml

@@ -0,0 +1,166 @@
+# `publish-aur.yml` - Arch Linux AUR Package Workflow
+# --------------------
+# This workflow automates the validation and publishing of the 'numa-git' package to the
+# Arch User Repository (AUR). The AUR is a community-driven repository for Arch Linux users.
+#
+# Workflow Overview:
+# 1. Validate: Builds and tests the package for Arch Linux x86_64 using a clean
+#    Arch Linux container.
+# 2. Audit: Checks Rust dependencies for known security vulnerabilities using
+#    'cargo-audit'.
+# 3. Publish: If on the 'main' branch, it pushes the updated PKGBUILD and
+#    .SRCINFO to the AUR.
+#
+# Security Best Practices:
+# - SHA Pinning: All GitHub Actions are pinned to a full-length commit SHA (e.g., v6.0.2 @ SHA)
+#   to ensure the code is immutable and protects against supply-chain attacks where a tag
+#   might be maliciously moved to a compromised commit.
+# - SSH Hygiene: Uses ssh-agent to keep the private key in memory rather than on disk.
+# - Audit: Runs 'cargo audit' to prevent publishing known vulnerable dependencies.
+
+name: Publish - Arch Linux AUR Package
+
+on:
+  push:
+    branches: [main]
+    paths-ignore:
+      - 'site/**'
+      - 'blog/**'
+      - 'drafts/**'
+      - '*.md'
+      - 'scripts/serve-site.sh'
+      - 'scripts/generate-blog-index.sh'
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+jobs:
+  # The 'validate' job ensures that the PKGBUILD is correct and the software builds/tests
+  # successfully on Arch Linux before we attempt to publish it.
+  validate:
+    name: Validate PKGBUILD (${{ matrix.arch }})
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        arch: [x86_64]
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
+
+      - name: Build and Test Package
+        timeout-minutes: 60
+        env:
+          AUR_PKGNAME: ${{ secrets.AUR_PACKAGE_NAME }}
+        run: |
+          # We use a temporary directory to avoid Docker permission issues with the workspace.
+          mkdir -p build-dir
+          cp PKGBUILD build-dir/
+
+          docker run --rm -v $PWD/build-dir:/pkg -w /pkg archlinux:latest /bin/bash -c "
+            # ARCH LINUX SECURITY REQUIREMENT:
+            # 'makepkg' (the tool that builds Arch packages) refuses to run as root for safety.
+            # We must create a standard user and give them sudo access.
+
+            # Install build-time dependencies.
+            # 'base-devel' includes essential tools like gcc, make, and binutils.
+            # Install 'rust' directly to avoid the interactive virtual-package
+            # prompt for 'cargo' on current Arch images.
+            pacman -Syu --noconfirm --needed base-devel rust git sudo cargo-audit
+
+            useradd -m builduser
+            chown -R builduser:builduser /pkg
+
+            # Allow the build user to install dependencies during the build process.
+            echo 'builduser ALL=(ALL) NOPASSWD: ALL' > /etc/sudoers.d/builduser
+
+            # Fetch the source tree first so pkgver() and cargo-audit have a
+            # real Cargo.lock to inspect.
+            sudo -u builduser makepkg -o --nobuild --nocheck --nodeps --noprepare
+
+            # SECURITY AUDIT:
+            # Fail early if any dependencies have known security vulnerabilities.
+            sudo -u builduser sh -lc 'cd /pkg/src/numa && cargo audit'
+
+            # BUILD & TEST:
+            # 'makepkg -s' will:
+            # 1. Download source files (cloning this repo)
+            # 2. Run prepare(), build(), and check() (running cargo test)
+            # 3. Create the final .pkg.tar.zst package
+            sudo -u builduser makepkg -s --noconfirm
+          "
+
+  # The 'publish' job updates the AUR repository with our latest PKGBUILD and .SRCINFO.
+  publish:
+    name: Publish to AUR
+    needs: validate
+    runs-on: ubuntu-latest
+    if: github.event_name == 'push' && github.ref == 'refs/heads/main'
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
+
+      # Securely configure SSH for AUR access.
+      - name: Configure SSH
+        run: |
+          mkdir -p ~/.ssh
+          # Official AUR Ed25519 fingerprint (prevents Man-in-the-Middle attacks).
+          echo "aur.archlinux.org ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIEuBKrPzbawxA/k2g6NcyV5jmqwJ2s+zpgZGZ7tpLIcN" >> ~/.ssh/known_hosts
+
+          # Use ssh-agent to keep the private key in memory rather than writing it to disk.
+          eval $(ssh-agent -s)
+          echo "${{ secrets.AUR_SSH_PRIVATE_KEY }}" | tr -d '\r' | ssh-add -
+
+          # Export the agent socket so subsequent 'git' commands can use it.
+          echo "SSH_AUTH_SOCK=$SSH_AUTH_SOCK" >> $GITHUB_ENV
+          echo "SSH_AGENT_PID=$SSH_AGENT_PID" >> $GITHUB_ENV
+
+      - name: Push to AUR
+        env:
+          AUR_PKGNAME: ${{ secrets.AUR_PACKAGE_NAME }}
+          AUR_EMAIL: ${{ secrets.AUR_EMAIL }}
+          AUR_USER: ${{ secrets.AUR_USERNAME }}
+        run: |
+          # AUR repos are managed via Git. Each package has its own repo at:
+          # ssh://aur@aur.archlinux.org/<package-name>.git
+          git clone ssh://aur@aur.archlinux.org/$AUR_PKGNAME.git aur-repo
+
+          cp PKGBUILD aur-repo/
+          cd aur-repo
+
+          # METADATA GENERATION:
+          # '.SRCINFO' is a machine-readable version of the PKGBUILD.
+          # We must run this as a non-root user ('builduser') inside the container.
+          docker run --rm -v $(pwd):/pkg archlinux:latest /bin/bash -c "
+            pacman -Syu --noconfirm --needed binutils git sudo
+            useradd -m builduser
+            chown -R builduser:builduser /pkg
+            cd /pkg
+            sudo -u builduser git config --global --add safe.directory '*'
+            # makepkg -od fetches the source first so pkgver() can calculate the version.
+            # --noprepare skips the prepare() function, which invokes cargo and would
+            # otherwise require a full rust toolchain in this metadata-only container.
+            # pkgver() runs before prepare(), so .SRCINFO still gets the correct version.
+            sudo -u builduser makepkg -od --noprepare && sudo -u builduser makepkg --printsrcinfo > .SRCINFO
+          "
+
+          # Reclaim ownership: the in-container 'chown -R builduser:builduser /pkg'
+          # propagates through the bind mount, leaving .git/ owned by the container's
+          # builduser UID. Without this, subsequent 'git config' on the host fails with
+          # "could not lock config file .git/config: Permission denied".
+          sudo chown -R "$(id -u):$(id -g)" .
+
+          # Set the commit identity using secrets for security and auditability.
+          git config user.name "$AUR_USER"
+          git config user.email "$AUR_EMAIL"
+
+          # Stage and commit both the human-readable PKGBUILD and machine-readable .SRCINFO.
+          git add PKGBUILD .SRCINFO
+
+          if ! git diff --cached --quiet; then
+            git commit -m "chore: update PKGBUILD to ${{ github.sha }}"
+            git push origin master
+          else
+            echo "No changes to commit (metadata and PKGBUILD are already up-to-date)."
+          fi

.github/workflows/release.yml

@@ -31,7 +31,7 @@ jobs:
 
     runs-on: ${{ matrix.os }}
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
 
       - name: Install Rust
         uses: dtolnay/rust-toolchain@stable
@@ -70,7 +70,7 @@ jobs:
           (Get-FileHash "${{ matrix.name }}.zip" -Algorithm SHA256).Hash.ToLower() + "  ${{ matrix.name }}.zip" | Out-File "${{ matrix.name }}.zip.sha256" -Encoding ascii
 
       - name: Upload artifact
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v7
         with:
           name: ${{ matrix.name }}
           path: |
@@ -82,7 +82,7 @@ jobs:
   publish:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
 
       - name: Install Rust
         uses: dtolnay/rust-toolchain@stable
@@ -96,7 +96,7 @@ jobs:
     needs: [build, publish]
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/download-artifact@v4
+      - uses: actions/download-artifact@v8
         with:
           merge-multiple: true
 
@@ -108,3 +108,10 @@ jobs:
             *.tar.gz
             *.zip
             *.sha256
+
+  bump-homebrew:
+    needs: release
+    uses: ./.github/workflows/homebrew-bump.yml
+    with:
+      version: ${{ github.ref_name }}
+    secrets: inherit

.github/workflows/static.yml

@@ -30,18 +30,18 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
       - name: Install pandoc
-        run: sudo apt-get install -y pandoc
+        uses: pandoc/actions/setup@v1
       - name: Generate blog HTML
         run: make blog
       - name: Setup Pages
-        uses: actions/configure-pages@v5
+        uses: actions/configure-pages@v6
       - name: Upload artifact
-        uses: actions/upload-pages-artifact@v3
+        uses: actions/upload-pages-artifact@v4
         with:
           # Upload entire repository
           path: './site'
       - name: Deploy to GitHub Pages
         id: deployment
-        uses: actions/deploy-pages@v4
+        uses: actions/deploy-pages@v5

.gitignore

@@ -1,4 +1,8 @@
 /target
+/build-dir
 CLAUDE.md
 docs/
 site/blog/posts/
+ios/
+drafts/
+site/blog/index.html

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "numa"
-version = "0.10.0"
+version = "0.13.1"
 authors = ["razvandimescu <razvan@dimescu.com>"]
 edition = "2021"
 description = "Portable DNS resolver in Rust — .numa local domains, ad blocking, developer overrides, DNS-over-HTTPS"
@@ -10,11 +10,11 @@ keywords = ["dns", "dns-server", "ad-blocking", "reverse-proxy", "developer-tool
 categories = ["network-programming", "development-tools"]
 
 [dependencies]
-tokio = { version = "1", features = ["rt-multi-thread", "macros", "net", "time", "sync"] }
+tokio = { version = "1", features = ["rt-multi-thread", "macros", "net", "time", "sync", "signal"] }
 axum = "0.8"
 serde = { version = "1", features = ["derive"] }
 serde_json = "1"
-toml = "0.8"
+toml = "1.1"
 log = "0.4"
 env_logger = "0.11"
 reqwest = { version = "0.12", features = ["rustls-tls", "gzip", "http2"], default-features = false }
@@ -22,19 +22,24 @@ hyper = { version = "1", features = ["client", "http1", "server"] }
 hyper-util = { version = "0.1", features = ["client-legacy", "http1", "tokio"] }
 http-body-util = "0.1"
 futures = "0.3"
-socket2 = { version = "0.5", features = ["all"] }
+socket2 = { version = "0.6", features = ["all"] }
-rcgen = { version = "0.13", features = ["pem", "x509-parser"] }
+rcgen = { version = "0.14", features = ["pem", "x509-parser"] }
 time = "0.3"
 rustls = "0.23"
 tokio-rustls = "0.26"
 arc-swap = "1"
 ring = "0.17"
 rustls-pemfile = "2.2.0"
+qrcode = { version = "0.14", default-features = false, features = ["svg"] }
+webpki-roots = "1"
 
 [dev-dependencies]
-criterion = { version = "0.5", features = ["html_reports"] }
+criterion = { version = "0.8", features = ["html_reports"] }
 tower = { version = "0.5", features = ["util"] }
 http = "1"
+hickory-resolver = { version = "0.25", features = ["https-ring", "webpki-roots"] }
+hickory-proto = "0.25"
+x509-parser = "0.18"
 
 [[bench]]
 name = "hot_path"
@@ -47,3 +52,7 @@ harness = false
 [[bench]]
 name = "dnssec"
 harness = false
+
+[[bench]]
+name = "recursive_compare"
+harness = false

Dockerfile

@@ -1,4 +1,4 @@
-FROM rust:1.88-alpine AS builder
+FROM rust:1.94-alpine AS builder
 RUN apk add --no-cache musl-dev cmake make perl
 WORKDIR /app
 COPY Cargo.toml Cargo.lock ./
@@ -6,12 +6,14 @@ RUN mkdir src && echo 'fn main() {}' > src/main.rs && echo '' > src/lib.rs
 RUN cargo build --release 2>/dev/null || true
 RUN rm -rf src
 COPY src/ src/
+COPY benches/ benches/
 COPY site/ site/
 COPY numa.toml com.numa.dns.plist numa.service ./
 RUN touch src/main.rs src/lib.rs
 RUN cargo build --release
 
-FROM alpine:3.20
+FROM alpine:3.23
 COPY --from=builder /app/target/release/numa /usr/local/bin/numa
+RUN mkdir -p /root/.config/numa && printf '[server]\napi_bind_addr = "0.0.0.0"\n\n[proxy]\nenabled = true\nbind_addr = "0.0.0.0"\n' > /root/.config/numa/numa.toml
 EXPOSE 53/udp 80/tcp 443/tcp 853/tcp 5380/tcp
 ENTRYPOINT ["numa"]

Makefile

@@ -32,6 +32,19 @@ blog:
 		pandoc "$$f" --template=site/blog-template.html -o "site/blog/posts/$$name.html"; \
 		echo "  $$f → site/blog/posts/$$name.html"; \
 	done
+	@scripts/generate-blog-index.sh
+
+blog-drafts: blog
+	@if [ -d drafts ] && ls drafts/*.md >/dev/null 2>&1; then \
+		for f in drafts/*.md; do \
+			name=$$(basename "$$f" .md); \
+			pandoc "$$f" --template=site/blog-template.html -o "site/blog/posts/$$name.html"; \
+			echo "  $$f → site/blog/posts/$$name.html (draft)"; \
+		done; \
+		BLOG_INCLUDE_DRAFTS=1 scripts/generate-blog-index.sh; \
+	else \
+		echo "  No drafts found"; \
+	fi
 
 release:
 ifndef VERSION

PKGBUILD

@@ -0,0 +1,62 @@
+# Maintainer: razvandimescu <razvan@dimescu.com>
+pkgname=numa-git
+_pkgname=numa
+pkgver=0.10.1.r0.g0000000 # Placeholder — pkgver() rewrites this on each makepkg run
+pkgrel=1
+pkgdesc="Portable DNS resolver in Rust — .numa local domains, ad blocking, developer overrides, DNS-over-HTTPS"
+arch=('x86_64')
+url="https://github.com/razvandimescu/numa"
+license=('MIT')
+options=('!lto')
+depends=('gcc-libs' 'glibc')
+makedepends=('cargo' 'git' 'llvm-libs')
+provides=("$_pkgname")
+conflicts=("$_pkgname")
+backup=('etc/numa.toml')
+source=("$_pkgname::git+$url.git")
+sha256sums=('SKIP')
+
+pkgver() {
+  cd "$srcdir/$_pkgname"
+  ( set -o pipefail
+    git describe --long --tags 2>/dev/null | sed 's/\([^-]*-g\)/r\1/;s/-/./g' ||
+    printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+  ) | sed 's/^v//'
+}
+
+prepare() {
+  cd "$srcdir/$_pkgname"
+  # numa v0.10.1+ uses FHS-compliant paths on Linux by default
+  # (/var/lib/numa for data, journalctl for logs), so no source
+  # patching is needed. The earlier sed targeted /usr/local/bin/numa,
+  # which only appears in a comment in current main.
+  export RUSTUP_TOOLCHAIN=stable
+  cargo fetch --locked
+}
+
+build() {
+  cd "$srcdir/$_pkgname"
+  export RUSTUP_TOOLCHAIN=stable
+  cargo build --frozen --release
+}
+
+check() {
+  cd "$srcdir/$_pkgname"
+  export RUSTUP_TOOLCHAIN=stable
+  cargo test --frozen
+}
+
+package() {
+  cd "$srcdir/$_pkgname"
+  install -Dm755 "target/release/$_pkgname" "$pkgdir/usr/bin/$_pkgname"
+
+  # numa.service uses {{exe_path}} as a placeholder substituted by
+  # `numa install` at runtime via replace_exe_path(). For an AUR
+  # package install (no `numa install` step), we substitute it
+  # statically here so systemd gets a real ExecStart path.
+  sed 's|{{exe_path}}|/usr/bin/numa /etc/numa.toml|g' numa.service > numa.service.patched
+  install -Dm644 "numa.service.patched" "$pkgdir/usr/lib/systemd/system/numa.service"
+
+  install -Dm644 "numa.toml" "$pkgdir/etc/numa.toml"
+  install -Dm644 "LICENSE" "$pkgdir/usr/share/licenses/$pkgname/LICENSE"
+}

README.md

@@ -21,9 +21,15 @@ brew install razvandimescu/tap/numa
 # Linux
 curl -fsSL https://raw.githubusercontent.com/razvandimescu/numa/main/install.sh | sh
 
+# Arch Linux (AUR)
+yay -S numa-git
+
 # Windows — download from GitHub Releases
 # All platforms
 cargo install numa
+
+# Docker
+docker run -d --name numa --network host ghcr.io/razvandimescu/numa
 ```
 
 ```bash
@@ -69,11 +75,19 @@ DNSSEC validates the full chain of trust: RRSIG signatures, DNSKEY verification,
 
 **DNS-over-TLS listener** (RFC 7858) — accept encrypted queries on port 853 from strict clients like iOS Private DNS, systemd-resolved, or stubby. Two modes:
 
-- **Self-signed** (default) — numa generates a local CA automatically. Works on any network with zero DNS setup, but clients must manually trust the CA (on macOS/Linux add to the system trust store; on iOS install a `.mobileconfig`).
+- **Self-signed** (default) — numa generates a local CA automatically. `numa install` adds it to the system trust store on macOS, Linux (Debian/Ubuntu, Fedora/RHEL/SUSE, Arch), and Windows. On iOS, install the `.mobileconfig` from `numa setup-phone`. Firefox keeps its own NSS store and ignores the system one — trust the CA there manually if you need HTTPS for `.numa` services in Firefox.
 - **Bring-your-own cert** — point `[dot] cert_path` / `key_path` at a publicly-trusted cert (e.g., Let's Encrypt via DNS-01 challenge on a domain pointing at your numa instance). Clients connect without any trust-store setup — same UX as AdGuard Home or Cloudflare `1.1.1.1`.
 
 ALPN `"dot"` is advertised and enforced in both modes; a handshake with mismatched ALPN is rejected as a cross-protocol confusion defense.
 
+**Phone setup** — point your iPhone or Android at Numa in one step:
+
+```bash
+numa setup-phone
+```
+
+Prints a QR code. Scan it, install the profile, toggle certificate trust — your phone's DNS now routes through Numa over TLS. Requires `[mobile] enabled = true` in `numa.toml`.
+
 ## LAN Discovery
 
 Run Numa on multiple machines. They find each other automatically via mDNS:
@@ -91,6 +105,26 @@ From Machine B: `curl http://api.numa` → proxied to Machine A's port 8000. Ena
 
 **Hub mode**: run one instance with `bind_addr = "0.0.0.0:53"` and point other devices' DNS to it — they get ad blocking + `.numa` resolution without installing anything.
 
+## Docker
+
+```bash
+# Recommended — host networking (Linux)
+docker run -d --name numa --network host ghcr.io/razvandimescu/numa
+
+# Port mapping (macOS/Windows Docker Desktop)
+docker run -d --name numa -p 53:53/udp -p 53:53/tcp -p 5380:5380 ghcr.io/razvandimescu/numa
+```
+
+Dashboard at `http://localhost:5380`. The image binds the API and proxy to `0.0.0.0` by default. Override with a custom config:
+
+```bash
+docker run -d --name numa --network host \
+  -v /path/to/numa.toml:/root/.config/numa/numa.toml \
+  ghcr.io/razvandimescu/numa
+```
+
+Multi-arch: `linux/amd64` and `linux/arm64`.
+
 ## How It Compares
 
 | | Pi-hole | AdGuard Home | Unbound | Numa |
@@ -102,17 +136,22 @@ From Machine B: `curl http://api.numa` → proxied to Machine A's port 8000. Ena
 | DNSSEC validation | — | — | Yes | Yes (RSA, ECDSA, Ed25519) |
 | Ad blocking | Yes | Yes | — | 385K+ domains |
 | Web admin UI | Full | Full | — | Dashboard |
-| Encrypted upstream (DoH) | Needs cloudflared | Yes | — | Native |
+| Encrypted upstream (DoH/DoT) | Needs cloudflared | DoH only | DoT only | DoH + DoT (`tls://`) |
 | Encrypted clients (DoT listener) | Needs stunnel sidecar | Yes | Yes | Native (RFC 7858) |
+| DoH server endpoint | — | Yes | — | Yes (RFC 8484) |
+| Request hedging | — | — | — | All protocols (UDP, DoH, DoT) |
+| Serve-stale + prefetch | — | — | Prefetch at 90% TTL | RFC 8767, prefetch at 90% TTL |
+| Conditional forwarding | — | Yes | Yes | Yes (per-suffix rules) |
 | Portable (laptop) | No (appliance) | No (appliance) | Server | Single binary, macOS/Linux/Windows |
 | Community maturity | 56K stars, 10 years | 33K stars | 20 years | New |
 
 ## Performance
 
-691ns cached round-trip. ~2.0M qps throughput. Zero heap allocations in the hot path. Recursive queries average 237ms after SRTT warmup (12x improvement over round-robin). ECDSA P-256 DNSSEC verification: 174ns. [Benchmarks →](bench/)
+0.1ms cached queries — matches Unbound and AdGuard Home. Wire-level cache stores raw bytes with in-place TTL patching. Request hedging eliminates p99 spikes: cold recursive p99 538ms vs Unbound 748ms (−28%), σ 4× tighter. [Benchmarks →](benches/)
 
 ## Learn More
 
+- [Blog: DNS-over-TLS from Scratch in Rust](https://numa.rs/blog/posts/dot-from-scratch.html)
 - [Blog: Implementing DNSSEC from Scratch in Rust](https://numa.rs/blog/posts/dnssec-from-scratch.html)
 - [Blog: I Built a DNS Resolver from Scratch](https://numa.rs/blog/posts/dns-from-scratch.html)
 - [Configuration reference](numa.toml) — all options documented inline
@@ -123,10 +162,16 @@ From Machine B: `curl http://api.numa` → proxied to Machine A's port 8000. Ena
 - [x] DNS forwarding, caching, ad blocking, developer overrides
 - [x] `.numa` local domains — auto TLS, path routing, WebSocket proxy
 - [x] LAN service discovery — mDNS, cross-machine DNS + proxy
-- [x] DNS-over-HTTPS — encrypted upstream
+- [x] DNS-over-HTTPS — encrypted upstream + server endpoint (RFC 8484)
-- [x] DNS-over-TLS listener — encrypted client connections (RFC 7858, ALPN strict)
+- [x] DNS-over-TLS — encrypted client listener (RFC 7858) + upstream forwarding (`tls://`)
 - [x] Recursive resolution + DNSSEC — chain-of-trust, NSEC/NSEC3
 - [x] SRTT-based nameserver selection
+- [x] Multi-forwarder failover — multiple upstreams with SRTT ranking, fallback pool
+- [x] Request hedging — parallel requests rescue packet loss and tail latency (all protocols)
+- [x] Serve-stale + prefetch — RFC 8767, background refresh at <10% TTL and on stale serve
+- [x] Conditional forwarding — per-suffix rules for split-horizon DNS (Tailscale, VPNs)
+- [x] Cache warming — proactive resolution for configured domains
+- [x] Mobile onboarding — `setup-phone` QR flow, mobile API, mobileconfig profiles
 - [ ] pkarr integration — self-sovereign DNS via Mainline DHT
 - [ ] Global `.numa` names — DHT-backed, no registrar
 

benches/numa-bench-recursive.toml

@@ -0,0 +1,30 @@
+[server]
+bind_addr = "127.0.0.1:5454"
+api_port = 5381
+api_bind_addr = "127.0.0.1"
+data_dir = "/tmp/numa-bench"
+
+[upstream]
+mode = "recursive"
+timeout_ms = 10000
+
+[cache]
+min_ttl = 60
+max_ttl = 3600
+
+[blocking]
+enabled = false
+
+[proxy]
+port = 8080
+tls_port = 8443
+
+[dot]
+enabled = true
+port = 8530
+
+[mobile]
+enabled = false
+
+[lan]
+enabled = false

benches/numa-bench.toml

@@ -0,0 +1,31 @@
+[server]
+bind_addr = "127.0.0.1:5454"
+api_port = 5381
+api_bind_addr = "127.0.0.1"
+data_dir = "/tmp/numa-bench"
+
+[upstream]
+mode = "forward"
+address = ["https://9.9.9.9/dns-query"]
+timeout_ms = 10000
+
+[cache]
+min_ttl = 60
+max_ttl = 3600
+
+[blocking]
+enabled = false
+
+[proxy]
+port = 8080
+tls_port = 8443
+
+[dot]
+enabled = true
+port = 8530
+
+[mobile]
+enabled = false
+
+[lan]
+enabled = false

blog/dns-from-scratch.md

@@ -1,7 +1,7 @@
 ---
 title: I Built a DNS Resolver from Scratch in Rust
 description: How DNS actually works at the wire level — label compression, TTL tricks, DoH, and what surprised me building a resolver with zero DNS libraries.
-date: March 2026
+date: 2026-03-20
 ---
 
 I wanted to understand how DNS actually works. Not the "it translates domain names to IP addresses" explanation — the actual bytes on the wire. What does a DNS packet look like? How does label compression work? Why is everything crammed into 512 bytes?

blog/dnssec-from-scratch.md

@@ -1,7 +1,7 @@
 ---
 title: Implementing DNSSEC from Scratch in Rust
 description: Recursive resolution from root hints, chain-of-trust validation, NSEC/NSEC3 denial proofs, and what I learned implementing DNSSEC with zero DNS libraries.
-date: March 2026
+date: 2026-03-28
 ---
 
 In the [previous post](/blog/posts/dns-from-scratch.html) I covered how DNS works at the wire level — packet format, label compression, TTL caching, DoH. Numa was a forwarding resolver: it parsed packets, did useful things locally, and relayed the rest to Cloudflare or Quad9.
@@ -163,12 +163,12 @@ The fix has three parts:
 
 **TCP fallback.** Every outbound query tries UDP first (800ms timeout). If UDP fails or the response is truncated, retry immediately over TCP. TCP uses a 2-byte length prefix before the DNS message — trivial to implement, and it handles DNSSEC responses that exceed the UDP payload limit.
 
-**UDP auto-disable.** After 3 consecutive UDP failures, flip a global `AtomicBool` and skip UDP entirely — go TCP-first for all queries. This avoids burning 800ms per hop on a network where UDP will never work. The flag resets when the network changes (detected via LAN IP monitoring).
+**UDP auto-disable.** After 3 consecutive UDP failures, flip a global `AtomicBool` and skip UDP entirely — go TCP-first for all queries. The flag resets when the network changes (detected via LAN IP monitoring).
+
+<img src="../hostile-network.svg" alt="Latency profile on a hostile network: queries 1-3 each spend 800ms waiting for a UDP timeout before retrying over TCP, taking 1,100ms total per query. After 3 consecutive failures the UDP auto-disable flag flips, and queries 4+ go TCP-first and complete in 300ms each — 3.7× faster.">
 
 **Query minimization (RFC 7816).** When querying root servers, send only the TLD — `com` instead of `secret-project.example.com`. Root servers handle trillions of queries and are operated by 12 organizations. Minimization reduces what they learn from yours.
 
-The result: on a network that blocks UDP:53, Numa detects the block within the first 3 queries, switches to TCP, and resolves normally at 300-500ms per cold query. Cached queries remain 0ms. No manual config change needed — switch networks and it adapts.
-
 I wouldn't have found this without dogfooding. The code worked perfectly on my home network. It took a real hostile network to expose the assumption that UDP always works.
 
 ## What I learned

blog/dot-from-scratch.md

@@ -0,0 +1,176 @@
+---
+title: DNS-over-TLS from Scratch in Rust
+description: Building RFC 7858 on top of rustls — length-prefix framing, ALPN cross-protocol defense, and two bugs that only the strict clients caught.
+date: 2026-04-06
+---
+
+The [previous post](/blog/posts/dnssec-from-scratch.html) ended with "DoT — the last encrypted transport we don't support." This post is about building it.
+
+Numa now runs a DoT listener on port 853. My iPhone uses it as its system resolver, so ad blocking, DNSSEC validation, and recursive resolution follow my phone through the day. No cloud, no account, no companion app — a self-signed cert, a `.mobileconfig` profile, and a QR code in the terminal.
+
+RFC 7858 is ten pages. The hard parts weren't in the RFC. They were in cross-protocol confusion defenses, a crypto-provider init gotcha that only triggered in one specific config combination, and a certificate SAN bug iOS was happy to accept and `kdig` immediately rejected. This post is about those parts.
+
+## Why DoT when you already have DoH?
+
+Numa has shipped DoH since v0.1. Both protocols tunnel DNS over TLS; DoH wraps queries in HTTP/2, DoT is DNS-over-TCP with TLS in front. Same privacy guarantees, different wrapper.
+
+The answer to "why both" is that **phones ask for DoT by name.** iOS system DNS configures it with two fields (IP + server name) instead of a URL template. Android 9+ "Private DNS" speaks DoT natively. Linux stubs default to DoT. I wanted my phone on Numa without installing anything on the phone itself, and DoT is the protocol iOS and Android already speak for that.
+
+## The wire format is refreshingly small
+
+RFC 7858 is one sentence of wire protocol: *DNS-over-TCP (RFC 1035 §4.2.2) with TLS in front, on port 853.* DNS-over-TCP has existed since 1987 — a 2-byte length prefix followed by the DNS message. DoT is that, wrapped in a TLS session. The entire framing code is seven lines:
+
+```rust
+async fn write_framed<S>(stream: &mut S, msg: &[u8]) -> io::Result<()>
+where S: AsyncWriteExt + Unpin {
+    let mut out = Vec::with_capacity(2 + msg.len());
+    out.extend_from_slice(&(msg.len() as u16).to_be_bytes());
+    out.extend_from_slice(msg);
+    stream.write_all(&out).await?;
+    stream.flush().await
+}
+```
+
+Reads are symmetric: `read_exact` two bytes, convert to `u16`, `read_exact` that many bytes. No HTTP headers, no chunked encoding, no framing layer.
+
+## Persistent connections
+
+A fresh TCP+TLS handshake is at least 3 RTTs — about 300ms on a 100ms connection, 60× the cost of a UDP query. RFC 7858 §3.4 says clients SHOULD reuse the TCP connection for multiple queries, and every real DoT client does: iOS, Android, systemd, stubby. A single connection often carries hundreds of queries.
+
+<img src="../dot-handshake.svg" alt="Timing diagram comparing a DNS lookup over plain UDP (1 RTT), over DoT on a fresh connection (3 RTTs — TCP handshake, TLS 1.3 handshake, then the query), and over a reused DoT session (1 RTT, same as UDP).">
+
+The amortization point is the whole game. If you only ever do one query per connection, DoT is roughly 3× slower than UDP and you should not use it. If you reuse the same TLS session for a browsing session's worth of queries, the handshake is paid once and every subsequent query is effectively free.
+
+The server is a loop that reads a length-prefixed message, resolves it, writes the response framed the same way, waits for the next one. Three timeouts keep it honest:
+
+- **Handshake timeout (10s)** — a slowloris that opens TCP but never sends a ClientHello can't pin a worker.
+- **Idle timeout (30s)** — a connected client with nothing to say gets dropped.
+- **Write timeout (10s)** — a stalled reader can't hold a response buffer indefinitely.
+
+A semaphore caps concurrent connections at 512 so a burst of handshakes can't exhaust the tokio runtime.
+
+## ALPN, the cross-protocol defense that matters
+
+If DoT lives on port 853 and HTTPS on 443, what stops an HTTP/2 client from hitting 853 and getting confused replies? [Cross-protocol attacks](https://alpaca-attack.com/) exist and have had real CVEs. The defense is ALPN: during the TLS handshake the client advertises protocols, the server picks one it supports or fails. A DoT server advertises `"dot"`; a client offering only `"h2"` gets a `no_application_protocol` fatal alert before any frames are exchanged.
+
+rustls enforces this by default when you set `alpn_protocols`:
+
+```rust
+let mut config = ServerConfig::builder()
+    .with_no_client_auth()
+    .with_single_cert(certs, key)?;
+config.alpn_protocols = vec![b"dot".to_vec()];
+```
+
+"The library enforces it by default" has a latent risk: a future rustls upgrade could change the default, and the defense would quietly evaporate. I wrote a test that pins the behavior so any regression in a dependency update fails loudly:
+
+```rust
+#[tokio::test]
+async fn dot_rejects_non_dot_alpn() {
+    let (addr, cert_der) = spawn_dot_server().await;
+    let client_config = dot_client(&cert_der, vec![b"h2".to_vec()]);
+    let connector = tokio_rustls::TlsConnector::from(client_config);
+    let tcp = tokio::net::TcpStream::connect(addr).await.unwrap();
+    let result = connector
+        .connect(ServerName::try_from("numa.numa").unwrap(), tcp)
+        .await;
+    assert!(result.is_err(),
+        "DoT server must reject ALPN that doesn't include \"dot\"");
+}
+```
+
+When you're leaning on a library's default for a security-critical invariant, the test is the contract.
+
+## Two bugs that hid for days
+
+Both were fixed before v0.10 shipped. Both stayed hidden because my initial tests used *permissive* clients.
+
+### The rustls crypto provider panic
+
+rustls 0.23 requires a `CryptoProvider` installed before you can build a `ServerConfig`. Numa's HTTPS proxy calls `install_default` as a side effect when it builds its own config, so DoT "just worked" for users who enabled both — the proxy had already initialized the provider before DoT's first handshake.
+
+Then I added support for user-provided DoT certificates. Someone running DoT with their own Let's Encrypt cert, with the HTTPS proxy disabled, would hit:
+
+```
+thread 'dot' panicked at rustls-0.23.25/src/crypto/mod.rs:185:14:
+no process-level CryptoProvider available -- call
+CryptoProvider::install_default() before this point
+```
+
+The panic happened on the first client connection, not at startup. While writing the integration suite for "DoT with BYO cert, proxy disabled" — the one combination nobody had ever actually exercised — the first run panicked. Fix is two lines: call `install_default` inside `load_tls_config` so DoT can stand alone. If a side effect initializes something and you have a path that skips that side effect, you have a bug waiting for a specific deployment.
+
+### The SAN bug iOS was happy to accept
+
+Numa's self-signed DoT cert is generated on first run from a local CA alongside the data directory. It needs to match whatever `ServerName` the client sends as SNI. For the HTTPS proxy, that's the wildcard domain pattern `*.numa` (matching `frontend.numa`, `api.numa`, etc.). I initially reused the same SAN list for DoT: a wildcard `*.numa` and nothing else.
+
+On an iPhone this worked perfectly. Full browsing session, persistent connections in the log, ad blocking active. I was about to merge when I ran one last smoke test with `kdig` (GnuTLS-backed, from [Knot DNS](https://www.knot-dns.cz/)):
+
+```
+$ kdig @192.168.1.16 -p 853 +tls \
+    +tls-ca=/usr/local/var/numa/ca.pem \
+    +tls-hostname=numa.numa example.com A
+
+;; TLS, handshake failed (Error in the certificate.)
+```
+
+Huh.
+
+[RFC 6125 §6.4.3](https://datatracker.ietf.org/doc/html/rfc6125#section-6.4.3): a wildcard in a certificate's DNS-ID matches exactly one label. `*.numa` matches `frontend.numa`, but not `numa.numa`, because the wildcard wants at least one label to substitute and strict clients reject wildcards in the leftmost label under single-label TLDs as ambiguous.
+
+iOS's TLS stack is lenient and accepts it. GnuTLS, NSS (Firefox), and most non-Apple validators don't. The fix is five lines — add an explicit `numa.numa` SAN alongside the wildcard. But the lesson is the one that stuck: I wrote a commit message saying "fix an iOS bug" and had to rewrite it, because iOS was fine. The real bug was that every GnuTLS/NSS-based client on the planet would have rejected the cert, and I only found it by running one more test with a stricter tool.
+
+> Test with the strict client. The permissive client hides your bugs.
+
+## Getting your phone onto it
+
+A DoT server is useless without a way to point a phone at it. iOS won't let you type an IP and a server name into Settings directly — you install a `.mobileconfig` profile that bundles the CA as a trust anchor and the DNS settings in a single payload.
+
+Numa ships a subcommand that builds one on the fly and serves it over a QR code in the terminal:
+
+```
+$ numa setup-phone
+
+  Numa Phone Setup
+
+  Profile URL: http://192.168.1.10:8765/mobileconfig
+
+  ██████████████████████████████
+  ██                          ██
+  ██   [QR code rendered in   ██
+  ██    your terminal]        ██
+  ██                          ██
+  ██████████████████████████████
+
+  On your iPhone:
+    1. Open Camera, point at the QR code, tap the yellow banner
+    2. Allow the download when Safari asks
+    3. Open Settings — tap "Profile Downloaded" near the top
+       (or: Settings → General → VPN & Device Management → Numa DNS)
+    4. Tap Install (top right), enter passcode, Install again
+    5. Settings → General → About → Certificate Trust Settings
+       Toggle ON "Numa Local CA" — required for DoT to work
+```
+
+The same QR is available in the dashboard — click "Phone Setup" in the header and the popover renders an SVG QR code pointing at the mobileconfig URL. On mobile viewports it shows a direct download link instead.
+
+<img src="../phone-setup-dashboard.png" alt="Numa dashboard with Phone Setup popover showing QR code and install instructions">
+
+Step 4 is non-negotiable. Even though the CA is bundled in the same profile that installs the DNS settings, iOS still requires the user to explicitly toggle trust in Certificate Trust Settings. It's a deliberate iOS policy to prevent profile-based trust injection — annoying, and correct.
+
+I've been dogfooding this since v0.10 shipped in early April. The phone resolves through Numa over DoT whenever I'm home; persistent connections are visible in the log as a single source port living through dozens of queries. The one real caveat: if the laptop's LAN IP changes, the profile breaks. [RFC 9462 DDR](https://datatracker.ietf.org/doc/html/rfc9462) fixes that — Numa can respond to `_dns.resolver.arpa IN SVCB` with its current IP and iOS picks it up on each network join. Next piece of work.
+
+## What I learned
+
+**RFC-level small, API-level hard.** RFC 7858 is ten pages. The framing is trivial. But the subtle stuff — ALPN, timeouts, connection caps, handshake vs idle vs write deadlines, backoff on accept errors — isn't in the RFC. Miss any of it and you leak a DoS vector or a protocol confusion hole.
+
+**Your test matrix is your security matrix.** Both bugs in this post were hidden by lenient clients. In both cases the strict client — kdig, or a specific config combination — surfaced the bug instantly. Pick test tools for strictness, not convenience. The moment you find yourself thinking "but iOS accepts it," stop and run kdig.
+
+**Don't initialize global state via side effects.** "Module A installs a global, module B silently depends on it, disabling A breaks B" is a bug pattern that keeps coming back. Fix: have module B initialize its dependency explicitly, even if it means calling an idempotent `install_default` twice. The dependency graph should be local and obvious.
+
+## What's next
+
+- ~~**DoH server**~~ — shipped in v0.12.0. `POST /dns-query` accepts [RFC 8484](https://datatracker.ietf.org/doc/html/rfc8484) wire-format queries, so Firefox/Chrome can point their built-in DoH at Numa.
+- **DoQ server (RFC 9250)** — DNS over QUIC. Android 14+ supports it natively.
+- **DDR (RFC 9462)** — auto-discovery via `_dns.resolver.arpa IN SVCB`, so phones pick up a moved Numa instance without the installed profile going stale.
+
+The code is at [github.com/razvandimescu/numa](https://github.com/razvandimescu/numa) — the DoT listener is in [`src/dot.rs`](https://github.com/razvandimescu/numa/blob/main/src/dot.rs) and the phone onboarding flow is in [`src/setup_phone.rs`](https://github.com/razvandimescu/numa/blob/main/src/setup_phone.rs) and [`src/mobileconfig.rs`](https://github.com/razvandimescu/numa/blob/main/src/mobileconfig.rs). MIT license.

blog/fixing-doh-tail-latency.md

@@ -0,0 +1,171 @@
+---
+title: Fixing DNS tail latency with a 5-line config and a 50-line function
+description: Periodic 40-140ms DoH spikes from hyper's dispatch channel. The fix was reqwest window tuning and request hedging — Dean & Barroso's "The Tail at Scale," applied to a DNS forwarder. Same ideas took cold recursive p99 from 2.3 seconds to 538ms.
+date: 2026-04-12
+---
+
+If you're using reqwest for small HTTP/2 payloads, you probably have a tail latency problem you don't know about. Hyper's default flow control windows are 10,000× oversized for anything under 1 KB, and its dispatch channel adds periodic 40-140ms stalls that don't show up in median benchmarks.
+
+I hit this building Numa's DoH forwarding path. Median was 10ms, mean was 23ms — the tail was dragging everything.
+
+<div class="hero-metrics">
+<div class="metric-card">
+<div class="metric-vs">DoH forwarding p99</div>
+<div class="metric-value">113 → 71ms</div>
+<div class="metric-label">window tuning + request hedging</div>
+</div>
+<div class="metric-card">
+<div class="metric-vs">Cold recursive p99</div>
+<div class="metric-value">2.3s → 538ms</div>
+<div class="metric-label">NS caching, serve-stale, parallel queries</div>
+</div>
+<div class="metric-card">
+<div class="metric-vs">Forwarding σ</div>
+<div class="metric-value">31 → 13ms</div>
+<div class="metric-label">random spikes become parallel races</div>
+</div>
+</div>
+
+The fix was a 5-line reqwest config and a 50-line hedging function. This post is also an advertisement for Dean & Barroso's 2013 paper ["The Tail at Scale"](https://research.google/pubs/pub40801/) — a decade-old idea that still demolishes dispatch spikes. The same ideas later took my cold recursive p99 from 2.3 seconds to 538ms.
+
+---
+
+## The cause: hyper's dispatch channel
+
+Reqwest sits on top of hyper, which interposes an mpsc dispatch channel and a separate `ClientTask` between `.send()` and the h2 stream. I instrumented the forwarding path and confirmed: 100% of the spike time lives in the `send()` phase, and a parallel heartbeat task showed zero runtime lag during spikes. The tokio runtime was fine — the stall was internal to hyper's request scheduling.
+
+Hickory-resolver doesn't have this issue. It holds `h2::SendRequest<Bytes>` directly and calls `ready().await; send_request()` in the caller's task — no channel, no scheduling dependency. I used it as a reference point throughout.
+
+## Fix #1 — HTTP/2 window sizes
+
+Reqwest inherits hyper's HTTP/2 defaults: 2 MB stream window, 5 MB connection window. For DNS responses (~200 bytes), that's ~10,000× oversized — unnecessary WINDOW_UPDATE frames, bloated bookkeeping on every poll, and different server-side scheduling behavior.
+
+Setting both windows to the h2 spec default (64 KB) dropped my median from 13.3ms to 10.1ms:
+
+```rust
+reqwest::Client::builder()
+    .use_rustls_tls()
+    .http2_initial_stream_window_size(65_535)
+    .http2_initial_connection_window_size(65_535)
+    .http2_keep_alive_interval(Duration::from_secs(15))
+    .http2_keep_alive_while_idle(true)
+    .http2_keep_alive_timeout(Duration::from_secs(10))
+    .pool_idle_timeout(Duration::from_secs(300))
+    .pool_max_idle_per_host(1)
+    .build()
+```
+
+**Any Rust code using reqwest for tiny-payload HTTP/2 workloads — DoH, API polling, metric scraping — is probably hitting this.**
+
+## Fix #2 — Request hedging
+
+["The Tail at Scale"](https://research.google/pubs/pub40801/) (Dean & Barroso, 2013): fire a request, and if it doesn't return within your P50 latency, fire the same request in parallel. First response wins.
+
+The intuition: if 5% of requests spike due to independent random events, two parallel requests means only 0.25% of pairs spike on *both*. The tail collapses.
+
+**The surprise: hedging against the same upstream works.** HTTP/2 multiplexes streams — two `send_request()` calls on one connection become independent h2 streams. If one stalls in the dispatch channel, the other keeps making progress.
+
+```rust
+pub async fn forward_with_hedging_raw(
+    wire: &[u8],
+    primary: &Upstream,
+    secondary: &Upstream,
+    hedge_delay: Duration,
+    timeout_duration: Duration,
+) -> Result<Vec<u8>> {
+    let primary_fut = forward_query_raw(wire, primary, timeout_duration);
+    tokio::pin!(primary_fut);
+    let delay = sleep(hedge_delay);
+    tokio::pin!(delay);
+
+    // Phase 1: wait for primary to return OR the hedge delay.
+    tokio::select! {
+        result = &mut primary_fut => return result,
+        _ = &mut delay => {}
+    }
+
+    // Phase 2: hedge delay expired — fire secondary, keep primary alive.
+    let secondary_fut = forward_query_raw(wire, secondary, timeout_duration);
+    tokio::pin!(secondary_fut);
+
+    // First successful response wins.
+    tokio::select! {
+        r = primary_fut => r,
+        r = secondary_fut => r,
+    }
+}
+```
+
+The [production version](https://github.com/razvandimescu/numa/blob/main/src/forward.rs#L267) adds error handling — if one leg fails, it waits for the other. In production, Numa passes the same `&Upstream` twice when only one is configured. I extended hedging to all protocols — UDP (rescues packet loss on WiFi), DoT (rescues TLS handshake stalls). Configurable via `hedge_ms`; set to 0 to disable.
+
+**Caveat: hedging hurts on degraded networks.** When latency is consistently high (no random spikes, just slow), the hedge adds overhead with nothing to rescue. Hedging is a variance reducer, not a latency reducer — it only helps when spikes are *random*.
+
+---
+
+## Forwarding results
+
+5 iterations × 101 domains × 10 rounds, 5,050 samples per method. Hickory-resolver included as a reference (it uses h2 directly, no dispatch channel):
+
+| | Single | **Hedged** | Hickory (ref) |
+|---|---|---|---|
+| mean | 17.4ms | **14.3ms** | 16.8ms |
+| median | 10.4ms | **10.2ms** | 13.3ms |
+| p95 | 52.5ms | **28.6ms** | 37.7ms |
+| p99 | 113.4ms | **71.3ms** | 98.1ms |
+| σ | 30.6ms | **13.2ms** | 19.1ms |
+
+The internal improvement: hedging cut p95 by 45%, p99 by 37%, σ by 57%. The exact margin vs hickory varies with network conditions; the σ reduction is consistent across runs.
+
+## Recursive resolution: from 2.3 seconds to 538ms
+
+Forwarding is one job. Recursive resolution — walking from root hints through TLD nameservers to the authoritative server — is a different one. I started 15× behind Unbound on cold recursive p99 and traced it to four root causes.
+
+**1. Missing NS delegation caching.** I cached glue records (ns1's IP) but not the delegation itself. Every `.com` query walked from root. Fix: cache NS records from referral authority sections. (10 lines)
+
+**2. Expired cache entries caused full cold resolutions.** Fix: serve-stale ([RFC 8767](https://www.rfc-editor.org/rfc/rfc8767)) — return expired entries with TTL=1 while revalidating in the background. (20 lines)
+
+**3. Wasting 1,900ms per unreachable server.** 800ms UDP timeout + unconditional 1,500ms TCP fallback. Fix: 400ms UDP, TCP only for truncation. (5 lines)
+
+**4. Sequential NS queries on cold starts.** Fix: fire to the top 2 nameservers simultaneously. First response wins, SRTT recorded for both. Same hedging principle. (50 lines)
+
+<div class="before-after">
+<div class="ba-item">
+<div class="ba-label">p99 before</div>
+<div class="ba-value ba-before">2,367ms</div>
+</div>
+<div class="ba-arrow">&#8594;</div>
+<div class="ba-item">
+<div class="ba-label">p99 after</div>
+<div class="ba-value ba-after">538ms</div>
+</div>
+<div class="ba-item ba-ref">
+<div class="ba-label">Unbound (ref)</div>
+<div class="ba-value">748ms</div>
+</div>
+</div>
+
+Genuine cold benchmarks — unique subdomains, 1 query per domain, 5 iterations, 505 samples per server:
+
+| | Baseline | Final | Unbound (ref) |
+|---|---|---|---|
+| p99 | 2,367ms | **538ms** | 748ms |
+| σ | 254ms | **114ms** | 457ms |
+| median | — | 77.6ms | 74.7ms |
+
+Unbound wins median by ~4%. Where hedging shines is the tail — domains with slow or unreachable nameservers, where parallel queries turn worst-case sequential timeouts into races. Cache hits are tied at 0.1ms across Numa, Unbound, and AdGuard Home.
+
+What I'm exploring next: persistent SRTT data across restarts (currently cold-starts lose all server timing), aggressive NSEC caching to shortcut negative lookups, and adaptive hedge delays that tune themselves to observed network conditions instead of a fixed 10ms.
+
+---
+
+## Takeaways
+
+The real hero of this post is Dean & Barroso. Hedging works because **spikes are random, and two random draws rarely both lose**. It's effective for any HTTP/2 client, any language, any forwarder topology. Nobody we know of ships it by default.
+
+If you're building a Rust service that makes many small HTTP/2 requests to the same backend: check your flow control window sizes first, then implement hedging. Don't rewrite the client.
+
+Benchmarks are in [`benches/recursive_compare.rs`](https://github.com/razvandimescu/numa/blob/main/benches/recursive_compare.rs) — run them yourself. If you're using reqwest for tiny-payload workloads and try the window size fix, I'd love to hear if you see the same improvement.
+
+---
+
+Numa is a DNS resolver that runs on your laptop or phone. DoH, DoT, .numa local domains, ad blocking, developer overrides, a REST API, and all the optimization work in this post. [github.com/razvandimescu/numa](https://github.com/razvandimescu/numa).

numa.toml

@@ -2,20 +2,28 @@
 bind_addr = "0.0.0.0:53"
 api_port = 5380
 # api_bind_addr = "127.0.0.1"  # default; set to "0.0.0.0" for LAN dashboard access
-# data_dir = "/usr/local/var/numa"  # where numa stores TLS CA and cert material
+# data_dir = "/var/lib/numa"  # where numa stores TLS CA and cert material
-                                    # (default: /usr/local/var/numa on unix,
+                              # Defaults: /var/lib/numa on linux (FHS),
-                                    # %PROGRAMDATA%\numa on windows). Override for
+                              # /usr/local/var/numa on macos (homebrew prefix),
-                                    # containerized deploys or tests that can't
+                              # %PROGRAMDATA%\numa on windows. Override for
-                                    # write to the system path.
+                              # containerized deploys or tests that can't
+                              # write to the system path.
 
 # [upstream]
 # mode = "forward"                              # "forward" (default) — relay to upstream
 #                                               # "recursive" — resolve from root hints (no address needed)
+# address = "9.9.9.9"                           # single upstream (plain UDP)
+# address = ["192.168.1.1", "9.9.9.9:5353"]    # multiple upstreams — SRTT picks fastest
 # address = "https://dns.quad9.net/dns-query"   # DNS-over-HTTPS (encrypted)
-# address = "https://cloudflare-dns.com/dns-query"  # Cloudflare DoH
+# address = "tls://9.9.9.9#dns.quad9.net"       # DNS-over-TLS (encrypted, port 853)
-# address = "9.9.9.9"                           # plain UDP
+# fallback = ["8.8.8.8", "1.1.1.1"]            # tried only when all primaries fail
-# port = 53                                     # only for forward mode, plain UDP
+# port = 53                                     # default port for addresses without :port
 # timeout_ms = 3000
+# hedge_ms = 10                                  # request hedging delay (ms). After this delay
+#                                                # without a response, fires a parallel request
+#                                                # to the same upstream. Rescues packet loss (UDP),
+#                                                # dispatch spikes (DoH), TLS stalls (DoT).
+#                                                # Set to 0 to disable. Default: 10
 # root_hints = [                                # only used in recursive mode
 #     "198.41.0.4",      # a.root-servers.net (Verisign)
 #     "199.9.14.201",    # b.root-servers.net (USC-ISI)
@@ -43,6 +51,22 @@ api_port = 5380
 #     "co", "br", "au", "ca", "jp",             # other major ccTLDs
 # ]
 
+# [[forwarding]]                                  # per-suffix conditional forwarding rules
+# suffix = "168.192.in-addr.arpa"                  # single suffix → one upstream
+# upstream = "100.90.1.63:5361"
+#
+# [[forwarding]]
+# suffix = ["home.local", "home.arpa"]             # multiple suffixes → same upstream
+# upstream = "10.0.0.1"                            # port 53 default
+#
+# [[forwarding]]                                    # DoT upstream: tls://IP[:port]#hostname
+# suffix = ["google.com", "goog"]                   # hostname is the TLS SNI / cert name
+# upstream = "tls://9.9.9.9#dns.quad9.net"          # port 853 default
+#
+# [[forwarding]]                                    # DoH upstream: full https:// URL
+# suffix = "example.corp"
+# upstream = "https://dns.quad9.net/dns-query"
+
 # [blocking]
 # enabled = true               # set to false to disable ad blocking
 # refresh_hours = 24
@@ -50,9 +74,10 @@ api_port = 5380
 # allowlist = ["example.com"]  # domains to never block
 
 [cache]
-max_entries = 10000
+max_entries = 100000
 min_ttl = 60
 max_ttl = 86400
+# warm = ["google.com", "github.com"]  # resolve at startup, refresh before TTL expiry
 
 [proxy]
 enabled = true
@@ -90,7 +115,7 @@ tld = "numa"
 
 # DNS-over-TLS listener (RFC 7858) — encrypted DNS on port 853
 # [dot]
-# enabled = false             # opt-in: accept DoT queries
+# enabled = true              # on by default; set false to disable
 # port = 853                  # standard DoT port
 # bind_addr = "0.0.0.0"       # IPv4 or IPv6; unspecified binds all interfaces
 # cert_path = "/etc/numa/dot.crt"  # PEM cert; omit to use self-signed (proxy CA if available)
@@ -101,3 +126,22 @@ tld = "numa"
 # enabled = true              # discover other Numa instances via mDNS (_numa._tcp.local)
 # broadcast_interval_secs = 30
 # peer_timeout_secs = 90
+
+# Mobile API — persistent HTTP listener serving read-only routes
+# (/health, /ca.pem, /mobileconfig, /ca.mobileconfig) on a LAN-reachable
+# port. Consumed by the iOS/Android companion apps for discovery and
+# profile fetching, and by `numa setup-phone` for QR-based onboarding.
+#
+# Opt-in because the listener binds to the LAN by default. None of the
+# exposed routes are cryptographically sensitive (no private keys, no
+# state mutations, all idempotent GETs), but enabling it does add a new
+# listener to any device on the LAN that scans port 8765.
+#
+# Safe for home LANs. Think twice before enabling on untrusted LANs
+# (office Wi-Fi, coffee shops, etc.) — an attacker on the same network
+# could run a competing Numa instance that shadows yours via mDNS and
+# trick companion apps into installing their profile instead of yours.
+[mobile]
+enabled = true                # opt-in to the mobile API listener
+# port = 8765                 # default; matches Discovery.swift defaultAPIPort
+# bind_addr = "0.0.0.0"       # default; set to "127.0.0.1" for localhost-only

scripts/generate-blog-index.sh

@@ -0,0 +1,239 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+# Generate site/blog/index.html from blog/*.md frontmatter.
+# Reads title, description, date from YAML frontmatter in each post.
+# Sorts newest first (by date string — "April 2026" > "March 2026").
+
+OUT="site/blog/index.html"
+
+# Extract frontmatter fields from a markdown file
+extract() {
+  local file="$1" field="$2"
+  sed -n '/^---$/,/^---$/p' "$file" | grep "^${field}:" | sed "s/^${field}: *//"
+}
+
+# Collect posts: "date|name|title|description" per line
+posts=""
+sources="blog/*.md"
+if [ "${BLOG_INCLUDE_DRAFTS:-}" = "1" ] && ls drafts/*.md >/dev/null 2>&1; then
+  sources="blog/*.md drafts/*.md"
+fi
+for f in $sources; do
+  name=$(basename "$f" .md)
+  title=$(extract "$f" title)
+  desc=$(extract "$f" description)
+  date=$(extract "$f" date)
+  posts+="${date}|${name}|${title}|${desc}"$'\n'
+done
+
+# Sort by ISO date (YYYY-MM-DD), newest first
+posts=$(echo "$posts" | grep -v '^$' | sort -t'|' -k1 -r)
+
+# Format ISO date (YYYY-MM-DD) to "Month YYYY"
+format_date() {
+  local months=(January February March April May June July August September October November December)
+  local y="${1%%-*}"
+  local m="${1#*-}"; m="${m%%-*}"; m=$((10#$m))
+  echo "${months[$((m-1))]} $y"
+}
+
+# Generate post list items
+items=""
+while IFS='|' read -r date name title desc; do
+  display_date=$(format_date "$date")
+  items+="    <li>
+      <a href=\"/blog/posts/${name}.html\">
+        <div class=\"post-title\">${title}</div>
+        <div class=\"post-desc\">${desc}</div>
+        <div class=\"post-date\">${display_date}</div>
+      </a>
+    </li>
+"
+done <<< "$posts"
+
+# Write the full index.html — style matches the existing hand-maintained version
+cat > "$OUT" << HTMLEOF
+<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="UTF-8">
+<meta name="viewport" content="width=device-width, initial-scale=1.0">
+<title>Blog — Numa</title>
+<meta name="description" content="Technical writing about DNS, Rust, and building infrastructure from scratch.">
+<link rel="stylesheet" href="/fonts/fonts.css">
+<style>
+*, *::before, *::after { margin: 0; padding: 0; box-sizing: border-box; }
+
+:root {
+  --bg-deep: #f5f0e8;
+  --bg-surface: #ece5da;
+  --bg-card: #faf7f2;
+  --amber: #c0623a;
+  --amber-dim: #9e4e2d;
+  --teal: #6b7c4e;
+  --text-primary: #2c2418;
+  --text-secondary: #6b5e4f;
+  --text-dim: #a39888;
+  --border: rgba(0, 0, 0, 0.08);
+  --font-display: 'Instrument Serif', Georgia, serif;
+  --font-body: 'DM Sans', system-ui, sans-serif;
+  --font-mono: 'JetBrains Mono', monospace;
+}
+
+body {
+  background: var(--bg-deep);
+  color: var(--text-primary);
+  font-family: var(--font-body);
+  font-weight: 400;
+  line-height: 1.7;
+  -webkit-font-smoothing: antialiased;
+}
+
+body::before {
+  content: '';
+  position: fixed;
+  inset: 0;
+  background-image: url("data:image/svg+xml,%3Csvg viewBox='0 0 256 256' xmlns='http://www.w3.org/2000/svg'%3E%3Cfilter id='n'%3E%3CfeTurbulence type='fractalNoise' baseFrequency='0.9' numOctaves='4' stitchTiles='stitch'/%3E%3C/filter%3E%3Crect width='100%25' height='100%25' filter='url(%23n)' opacity='0.025'/%3E%3C/svg%3E");
+  pointer-events: none;
+  z-index: 9999;
+}
+
+.blog-nav {
+  padding: 1.5rem 2rem;
+  display: flex;
+  align-items: center;
+  gap: 1.5rem;
+}
+
+.blog-nav a {
+  font-family: var(--font-mono);
+  font-size: 0.75rem;
+  letter-spacing: 0.08em;
+  text-transform: uppercase;
+  color: var(--text-dim);
+  text-decoration: none;
+  transition: color 0.2s;
+}
+.blog-nav a:hover { color: var(--amber); }
+
+.blog-nav .wordmark {
+  font-family: var(--font-display);
+  font-size: 1.4rem;
+  font-weight: 400;
+  color: var(--text-primary);
+  text-decoration: none;
+  text-transform: none;
+  letter-spacing: -0.02em;
+}
+.blog-nav .wordmark:hover { color: var(--amber); }
+
+.blog-nav .sep {
+  color: var(--text-dim);
+  font-family: var(--font-mono);
+  font-size: 0.75rem;
+}
+
+.blog-index {
+  max-width: 720px;
+  margin: 0 auto;
+  padding: 3rem 2rem 6rem;
+}
+
+.blog-index h1 {
+  font-family: var(--font-display);
+  font-weight: 400;
+  font-size: 2.5rem;
+  margin-bottom: 3rem;
+}
+
+.post-list {
+  list-style: none;
+}
+
+.post-list li {
+  padding: 1.5rem 0;
+  border-bottom: 1px solid var(--border);
+}
+
+.post-list li:first-child {
+  border-top: 1px solid var(--border);
+}
+
+.post-list a {
+  text-decoration: none;
+  display: block;
+}
+
+.post-list .post-title {
+  font-family: var(--font-display);
+  font-size: 1.4rem;
+  font-weight: 600;
+  color: var(--text-primary);
+  line-height: 1.3;
+  margin-bottom: 0.4rem;
+  transition: color 0.2s;
+}
+
+.post-list a:hover .post-title {
+  color: var(--amber);
+}
+
+.post-list .post-desc {
+  font-size: 0.95rem;
+  color: var(--text-secondary);
+  line-height: 1.5;
+  margin-bottom: 0.4rem;
+}
+
+.post-list .post-date {
+  font-family: var(--font-mono);
+  font-size: 0.72rem;
+  color: var(--text-dim);
+  letter-spacing: 0.04em;
+}
+
+.blog-footer {
+  text-align: center;
+  padding: 3rem 2rem;
+  border-top: 1px solid var(--border);
+  max-width: 720px;
+  margin: 0 auto;
+}
+
+.blog-footer a {
+  font-family: var(--font-mono);
+  font-size: 0.75rem;
+  letter-spacing: 0.08em;
+  text-transform: uppercase;
+  color: var(--text-dim);
+  text-decoration: none;
+  margin: 0 1rem;
+}
+.blog-footer a:hover { color: var(--amber); }
+</style>
+</head>
+<body>
+
+<nav class="blog-nav">
+  <a href="/" class="wordmark">Numa</a>
+  <span class="sep">/</span>
+  <a href="/blog/">Blog</a>
+</nav>
+
+<main class="blog-index">
+  <h1>Blog</h1>
+  <ul class="post-list">
+${items}  </ul>
+</main>
+
+<footer class="blog-footer">
+  <a href="https://github.com/razvandimescu/numa">GitHub</a>
+  <a href="/">Home</a>
+</footer>
+
+</body>
+</html>
+HTMLEOF
+
+echo "  blog/index.html generated ($(echo "$posts" | wc -l | tr -d ' ') posts)"

scripts/record-demo.sh

@@ -7,18 +7,19 @@
 # The script:
 #   1. Opens the dashboard in Chrome --app mode (clean, no address bar)
 #   2. Generates DNS traffic (forward, cache hit, blocked)
-#   3. Types "peekm" / "6419" into the Local Services form on camera
+#   3. Opens Phone Setup QR popover
-#   4. Shows LAN accessibility badge ("local only" / "LAN")
+#   4. Types "peekm" / "6419" into the Local Services form on camera
-#   5. Checks a blocked domain
+#   5. Shows LAN accessibility badge ("local only" / "LAN")
-#   6. Opens peekm.numa to show the proxy working
+#   6. Checks a blocked domain
-#   7. Records via ffmpeg and converts to optimized GIF
+#   7. Opens peekm.numa to show the proxy working
+#   8. Records via ffmpeg and converts to optimized GIF
 
 set -euo pipefail
 
 # --------------- Configuration ---------------
 OUTPUT="${1:-assets/hero-demo.gif}"
 PORT=5380
-RECORD_SECONDS=20
+RECORD_SECONDS=24
 VIEWPORT_W=1800
 VIEWPORT_H=1100
 FPS=12
@@ -230,8 +231,16 @@ dig @127.0.0.1 github.com +short > /dev/null 2>&1
 dig @127.0.0.1 ad.doubleclick.net +short > /dev/null 2>&1
 sleep 3
 
-# --------------- Scene 2: Add peekm service via UI (3-7s) ---------------
+# --------------- Scene 2: Phone Setup popover (3-7s) ---------------
-log "Scene 2: Adding peekm.numa service..."
+log "Scene 2: Phone Setup QR popover..."
+run_js "document.querySelector('#phoneSetup button').click();"
+sleep 3
+# Dismiss popover
+run_js "document.getElementById('phoneSetupPopover').style.display = 'none';"
+sleep 1
+
+# --------------- Scene 3: Add peekm service via UI (7-11s) ---------------
+log "Scene 3: Adding peekm.numa service..."
 
 # Services panel is now first — scroll to it
 run_js "
@@ -249,18 +258,18 @@ sleep 0.3
 run_js "document.querySelector('#serviceForm .btn-add').click();"
 sleep 2
 
-# --------------- Scene 3: Open peekm.numa (7-11s) ---------------
+# --------------- Scene 4: Open peekm.numa (11-15s) ---------------
-log "Scene 3: Opening peekm.numa in browser..."
+log "Scene 4: Opening peekm.numa in browser..."
 open "http://peekm.numa/view/peekm/README.md" 2>/dev/null || true
 sleep 4
 
-# --------------- Scene 4: Back to dashboard (11-14s) ---------------
+# --------------- Scene 5: Back to dashboard (15-18s) ---------------
-log "Scene 4: Back to dashboard — LAN badges + LOCAL queries visible..."
+log "Scene 5: Back to dashboard — LAN badges + LOCAL queries visible..."
 osascript -e "tell application \"System Events\" to set frontmost of (first process whose unix id is $CHROME_PID) to true" 2>/dev/null || true
 sleep 3
 
-# --------------- Scene 5: Check Domain blocker (14-17s) ---------------
+# --------------- Scene 6: Check Domain blocker (18-21s) ---------------
-log "Scene 5: Check Domain — blocked tracker..."
+log "Scene 6: Check Domain — blocked tracker..."
 # Scroll down to blocking panel
 run_js "
     var blockPanel = document.getElementById('blockingPanel');
@@ -273,8 +282,8 @@ sleep 0.3
 run_js "document.querySelector('#checkDomainInput').closest('form').querySelector('.btn').click();"
 sleep 2
 
-# --------------- Scene 6: Terminal-style dig overlay (17-20s) ---------------
+# --------------- Scene 7: Terminal-style dig overlay (21-24s) ---------------
-log "Scene 6: dig proof overlay..."
+log "Scene 7: dig proof overlay..."
 DIG_RESULT=$(dig @127.0.0.1 peekm.numa +short 2>/dev/null | head -1)
 run_js "
     var overlay = document.createElement('div');

scripts/release.sh

@@ -37,7 +37,7 @@ cargo update --workspace
 git add Cargo.toml Cargo.lock
 git commit -m "chore: bump version to $VERSION"
 git tag "$TAG"
-git push origin main --tags
+git push origin main "$TAG"
 
 echo
 echo "Released $TAG — GitHub Actions will build, publish to crates.io, and create the release."

scripts/serve-site.sh

@@ -0,0 +1,14 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+PORT="${1:-9000}"
+
+if [[ "${1:-}" == "--drafts" ]] || [[ "${2:-}" == "--drafts" ]]; then
+  PORT="${PORT//--drafts/9000}"  # default port if --drafts was first arg
+  make blog-drafts
+else
+  make blog
+fi
+
+echo "Serving site at http://localhost:$PORT"
+cd site && python3 -m http.server "$PORT"

scripts/update-homebrew-formula.py

@@ -0,0 +1,57 @@
+#!/usr/bin/env python3
+"""Rewrite a Homebrew formula in place: bump version, URL paths, and sha256 lines.
+
+Reads the formula path from argv[1], and the following env vars:
+  VERSION              e.g. "0.10.0" (no leading v)
+  SHA_MACOS_AARCH64
+  SHA_MACOS_X86_64
+  SHA_LINUX_AARCH64
+  SHA_LINUX_X86_64
+
+Assumptions about the formula:
+  - Has `version "X.Y.Z"` somewhere
+  - Has `url "...releases/download/vX.Y.Z/numa-<target>.tar.gz"` lines
+  - May or may not already have `sha256 "..."` lines immediately after each url
+"""
+import os
+import re
+import sys
+
+formula_path = sys.argv[1]
+version = os.environ["VERSION"].lstrip("v")
+shas = {
+    "macos-aarch64": os.environ["SHA_MACOS_AARCH64"],
+    "macos-x86_64": os.environ["SHA_MACOS_X86_64"],
+    "linux-aarch64": os.environ["SHA_LINUX_AARCH64"],
+    "linux-x86_64": os.environ["SHA_LINUX_X86_64"],
+}
+
+with open(formula_path) as f:
+    content = f.read()
+
+content = re.sub(r'version "[^"]*"', f'version "{version}"', content)
+content = re.sub(
+    r"releases/download/v[\d.]+/numa-",
+    f"releases/download/v{version}/numa-",
+    content,
+)
+content = re.sub(r'\n[ \t]*sha256 "[^"]*"', "", content)
+
+
+def add_sha(match: re.Match) -> str:
+    indent = match.group(1)
+    target = match.group(2)
+    if target not in shas:
+        return match.group(0)
+    return f'{match.group(0)}\n{indent}sha256 "{shas[target]}"'
+
+
+content = re.sub(
+    r'^([ \t]+)url "[^"]*numa-([\w-]+)\.tar\.gz"',
+    add_sha,
+    content,
+    flags=re.MULTILINE,
+)
+
+with open(formula_path, "w") as f:
+    f.write(content)

site/blog-template.html

@@ -74,6 +74,7 @@ body::before {
   font-weight: 400;
   color: var(--text-primary);
   text-decoration: none;
+  text-transform: none;
   letter-spacing: -0.02em;
 }
 .blog-nav .wordmark:hover { color: var(--amber); }
@@ -266,9 +267,105 @@ body::before {
 .blog-footer a:hover { color: var(--amber); }
 
 /* --- Responsive --- */
+/* Hero metrics cards */
+.hero-metrics {
+  display: grid;
+  grid-template-columns: repeat(3, 1fr);
+  gap: 1rem;
+  margin: 2rem 0;
+}
+.metric-card {
+  background: var(--bg-card);
+  border: 1px solid var(--border);
+  border-radius: 6px;
+  padding: 1.25rem;
+  text-align: center;
+}
+.metric-vs {
+  font-family: var(--font-mono);
+  font-size: 0.7rem;
+  letter-spacing: 0.08em;
+  text-transform: uppercase;
+  color: var(--text-dim);
+  margin-bottom: 0.5rem;
+}
+.metric-value {
+  font-family: var(--font-display);
+  font-size: 2.4rem;
+  font-weight: 400;
+  color: var(--amber);
+  line-height: 1.1;
+}
+.metric-label {
+  font-size: 0.82rem;
+  color: var(--text-secondary);
+  margin-top: 0.5rem;
+  line-height: 1.3;
+}
+
+/* Before/after progression */
+.before-after {
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  gap: 1.5rem;
+  margin: 2rem 0;
+  padding: 1.5rem;
+  background: var(--bg-card);
+  border: 1px solid var(--border);
+  border-radius: 6px;
+}
+.ba-item { text-align: center; }
+.ba-label {
+  font-family: var(--font-mono);
+  font-size: 0.7rem;
+  letter-spacing: 0.08em;
+  text-transform: uppercase;
+  color: var(--text-dim);
+  margin-bottom: 0.3rem;
+}
+.ba-value {
+  font-family: var(--font-display);
+  font-size: 1.8rem;
+  font-weight: 400;
+  color: var(--text-secondary);
+}
+.ba-before {
+  text-decoration: line-through;
+  text-decoration-color: rgba(192, 98, 58, 0.4);
+  color: var(--text-dim);
+}
+.ba-after { color: var(--amber); }
+.ba-arrow { font-size: 1.5rem; color: var(--text-dim); }
+.ba-ref {
+  border-left: 1px solid var(--border);
+  padding-left: 1.5rem;
+}
+
+/* Spike highlight */
+.spike {
+  background: rgba(192, 98, 58, 0.12);
+  padding: 0.15em 0.5em;
+  border-radius: 3px;
+  font-weight: 600;
+  color: var(--amber-dim);
+}
+
+/* Section dividers */
+.article hr {
+  border: none;
+  height: 1px;
+  background: var(--border);
+  margin: 3rem auto;
+  max-width: 120px;
+}
+
 @media (max-width: 640px) {
   .article { padding: 2rem 1.25rem 4rem; }
   .article pre { padding: 1rem; margin-left: -0.5rem; margin-right: -0.5rem; border-radius: 0; border-left: none; border-right: none; }
+  .hero-metrics { grid-template-columns: 1fr; }
+  .before-after { flex-direction: column; gap: 0.75rem; }
+  .ba-ref { border-left: none; border-top: 1px solid var(--border); padding-left: 0; padding-top: 0.75rem; }
 }
 </style>
 </head>

site/blog/dot-handshake.svg

@@ -0,0 +1,129 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 720 360" font-family="'DM Sans', system-ui, sans-serif" font-size="12">
+  <defs>
+    <marker id="arr-amber" viewBox="0 0 10 10" refX="9" refY="5" markerWidth="6" markerHeight="6" orient="auto">
+      <path d="M 0 0 L 10 5 L 0 10 z" fill="#c0623a"/>
+    </marker>
+    <marker id="arr-dim" viewBox="0 0 10 10" refX="9" refY="5" markerWidth="6" markerHeight="6" orient="auto">
+      <path d="M 0 0 L 10 5 L 0 10 z" fill="#a39888"/>
+    </marker>
+    <filter id="shadow" x="-3%" y="-3%" width="106%" height="106%">
+      <feDropShadow dx="0" dy="1" stdDeviation="2" flood-opacity="0.06"/>
+    </filter>
+  </defs>
+
+  <!-- Background -->
+  <rect width="720" height="360" rx="8" fill="#faf7f2"/>
+
+  <!-- Title -->
+  <text x="360" y="32" text-anchor="middle" font-size="15" font-weight="600" fill="#2c2418" font-family="'Instrument Serif', Georgia, serif" letter-spacing="-0.02em">UDP vs DoT — one lookup, three scenarios</text>
+  <text x="360" y="50" text-anchor="middle" font-size="11" fill="#a39888">Time flows downward. Amber = DNS work. Gray = TCP/TLS handshake overhead.</text>
+
+  <!-- ==================== Column 1: Plain UDP ==================== -->
+  <g transform="translate(20, 0)">
+    <!-- Column header -->
+    <text x="90" y="84" text-anchor="middle" font-size="13" font-weight="600" fill="#2c2418">Plain UDP DNS</text>
+    <text x="90" y="101" text-anchor="middle" font-size="10" fill="#a39888" letter-spacing="0.06em">PORT 53 · CLEARTEXT</text>
+
+    <!-- Lane labels -->
+    <text x="25" y="128" font-size="10" fill="#6b5e4f">client</text>
+    <text x="133" y="128" font-size="10" fill="#6b5e4f">server</text>
+
+    <!-- Lanes -->
+    <line x1="35" y1="138" x2="35" y2="198" stroke="#d4cbba" stroke-width="1" stroke-dasharray="2 3"/>
+    <line x1="145" y1="138" x2="145" y2="198" stroke="#d4cbba" stroke-width="1" stroke-dasharray="2 3"/>
+
+    <!-- query -->
+    <line x1="37" y1="148" x2="143" y2="160" stroke="#c0623a" stroke-width="2" marker-end="url(#arr-amber)"/>
+    <text x="90" y="143" text-anchor="middle" font-size="10" fill="#9e4e2d" font-weight="500">query</text>
+
+    <!-- response -->
+    <line x1="143" y1="178" x2="37" y2="190" stroke="#c0623a" stroke-width="2" marker-end="url(#arr-amber)"/>
+    <text x="90" y="205" text-anchor="middle" font-size="10" fill="#9e4e2d" font-weight="500">response</text>
+
+    <!-- Total cost badge -->
+    <rect x="20" y="225" width="140" height="32" rx="4" fill="#faf7f2" stroke="#d4cbba" stroke-width="1" filter="url(#shadow)"/>
+    <text x="90" y="241" text-anchor="middle" font-size="9" fill="#a39888" letter-spacing="0.04em">TOTAL LATENCY</text>
+    <text x="90" y="253" text-anchor="middle" font-size="11" font-weight="600" fill="#c0623a" font-family="'JetBrains Mono', monospace">1 × RTT</text>
+  </g>
+
+  <!-- ==================== Column 2: DoT cold ==================== -->
+  <g transform="translate(270, 0)">
+    <!-- Column header -->
+    <text x="90" y="84" text-anchor="middle" font-size="13" font-weight="600" fill="#2c2418">DoT — first query</text>
+    <text x="90" y="101" text-anchor="middle" font-size="10" fill="#a39888" letter-spacing="0.06em">PORT 853 · NEW CONNECTION</text>
+
+    <!-- Lane labels -->
+    <text x="25" y="128" font-size="10" fill="#6b5e4f">client</text>
+    <text x="133" y="128" font-size="10" fill="#6b5e4f">server</text>
+
+    <!-- Lanes -->
+    <line x1="35" y1="138" x2="35" y2="308" stroke="#d4cbba" stroke-width="1" stroke-dasharray="2 3"/>
+    <line x1="145" y1="138" x2="145" y2="308" stroke="#d4cbba" stroke-width="1" stroke-dasharray="2 3"/>
+
+    <!-- === RTT 1: TCP handshake === -->
+    <!-- SYN -->
+    <line x1="37" y1="145" x2="143" y2="153" stroke="#a39888" stroke-width="1.5" marker-end="url(#arr-dim)"/>
+    <!-- SYN-ACK -->
+    <line x1="143" y1="163" x2="37" y2="171" stroke="#a39888" stroke-width="1.5" marker-end="url(#arr-dim)"/>
+    <!-- ACK -->
+    <line x1="37" y1="181" x2="143" y2="189" stroke="#a39888" stroke-width="1.5" marker-end="url(#arr-dim)"/>
+    <!-- Label + RTT marker -->
+    <text x="168" y="170" font-size="9" fill="#a39888" font-family="'JetBrains Mono', monospace">1 rtt</text>
+    <text x="90" y="143" text-anchor="middle" font-size="9" fill="#6b5e4f" font-style="italic">TCP handshake</text>
+
+    <!-- === RTT 2: TLS 1.3 handshake === -->
+    <!-- ClientHello -->
+    <line x1="37" y1="208" x2="143" y2="216" stroke="#a39888" stroke-width="1.5" marker-end="url(#arr-dim)"/>
+    <!-- ServerHello + Cert + Finished -->
+    <line x1="143" y1="226" x2="37" y2="234" stroke="#a39888" stroke-width="1.5" marker-end="url(#arr-dim)"/>
+    <!-- Label + RTT marker -->
+    <text x="168" y="222" font-size="9" fill="#a39888" font-family="'JetBrains Mono', monospace">2 rtt</text>
+    <text x="90" y="205" text-anchor="middle" font-size="9" fill="#6b5e4f" font-style="italic">TLS 1.3 handshake</text>
+
+    <!-- === RTT 3: DNS exchange === -->
+    <!-- query (piggybacked on ClientFinished) -->
+    <line x1="37" y1="253" x2="143" y2="261" stroke="#c0623a" stroke-width="2" marker-end="url(#arr-amber)"/>
+    <!-- response -->
+    <line x1="143" y1="271" x2="37" y2="279" stroke="#c0623a" stroke-width="2" marker-end="url(#arr-amber)"/>
+    <!-- Label + RTT marker -->
+    <text x="168" y="267" font-size="9" fill="#a39888" font-family="'JetBrains Mono', monospace">3 rtt</text>
+    <text x="90" y="250" text-anchor="middle" font-size="10" fill="#9e4e2d" font-weight="500">query + response</text>
+
+    <!-- Total cost badge -->
+    <rect x="20" y="295" width="140" height="32" rx="4" fill="#faf7f2" stroke="#d4cbba" stroke-width="1" filter="url(#shadow)"/>
+    <text x="90" y="311" text-anchor="middle" font-size="9" fill="#a39888" letter-spacing="0.04em">TOTAL LATENCY</text>
+    <text x="90" y="323" text-anchor="middle" font-size="11" font-weight="600" fill="#c0623a" font-family="'JetBrains Mono', monospace">3 × RTT</text>
+  </g>
+
+  <!-- ==================== Column 3: DoT reused ==================== -->
+  <g transform="translate(520, 0)">
+    <!-- Column header -->
+    <text x="90" y="84" text-anchor="middle" font-size="13" font-weight="600" fill="#2c2418">DoT — reused session</text>
+    <text x="90" y="101" text-anchor="middle" font-size="10" fill="#a39888" letter-spacing="0.06em">PORT 853 · PERSISTENT TCP/TLS</text>
+
+    <!-- Lane labels -->
+    <text x="25" y="128" font-size="10" fill="#6b5e4f">client</text>
+    <text x="133" y="128" font-size="10" fill="#6b5e4f">server</text>
+
+    <!-- Lanes -->
+    <line x1="35" y1="138" x2="35" y2="198" stroke="#d4cbba" stroke-width="1" stroke-dasharray="2 3"/>
+    <line x1="145" y1="138" x2="145" y2="198" stroke="#d4cbba" stroke-width="1" stroke-dasharray="2 3"/>
+
+    <!-- query -->
+    <line x1="37" y1="148" x2="143" y2="160" stroke="#c0623a" stroke-width="2" marker-end="url(#arr-amber)"/>
+    <text x="90" y="143" text-anchor="middle" font-size="10" fill="#9e4e2d" font-weight="500">query</text>
+
+    <!-- response -->
+    <line x1="143" y1="178" x2="37" y2="190" stroke="#c0623a" stroke-width="2" marker-end="url(#arr-amber)"/>
+    <text x="90" y="205" text-anchor="middle" font-size="10" fill="#9e4e2d" font-weight="500">response</text>
+
+    <!-- Total cost badge -->
+    <rect x="20" y="225" width="140" height="32" rx="4" fill="#faf7f2" stroke="#d4cbba" stroke-width="1" filter="url(#shadow)"/>
+    <text x="90" y="241" text-anchor="middle" font-size="9" fill="#a39888" letter-spacing="0.04em">TOTAL LATENCY</text>
+    <text x="90" y="253" text-anchor="middle" font-size="11" font-weight="600" fill="#c0623a" font-family="'JetBrains Mono', monospace">1 × RTT</text>
+
+    <!-- Tiny caption -->
+    <text x="90" y="280" text-anchor="middle" font-size="9" fill="#a39888" font-style="italic">(handshake amortized</text>
+    <text x="90" y="292" text-anchor="middle" font-size="9" fill="#a39888" font-style="italic">across the session)</text>
+  </g>
+</svg>

site/blog/hostile-network.svg

@@ -0,0 +1,92 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 720 330" font-family="'DM Sans', system-ui, sans-serif" font-size="12">
+  <defs>
+    <filter id="shadow" x="-3%" y="-3%" width="106%" height="106%">
+      <feDropShadow dx="0" dy="1" stdDeviation="2" flood-opacity="0.06"/>
+    </filter>
+    <!-- Diagonal hatch for "wasted" UDP timeout regions. Darker warm gray
+         base + slightly darker diagonal stripes at 45°. The stripe pattern
+         is the Gantt convention for "dead/blocked time" — it reads as
+         "this time was thrown away" without needing the legend. -->
+    <pattern id="wasted-hatch" patternUnits="userSpaceOnUse" width="7" height="7" patternTransform="rotate(-45)">
+      <rect width="7" height="7" fill="#8b7f6f"/>
+      <line x1="0" y1="0" x2="0" y2="7" stroke="#3d3427" stroke-width="1.6" opacity="0.38"/>
+    </pattern>
+  </defs>
+
+  <!-- Background -->
+  <rect width="720" height="330" rx="8" fill="#faf7f2"/>
+
+  <!-- Title -->
+  <text x="360" y="32" text-anchor="middle" font-size="15" font-weight="600" fill="#2c2418" font-family="'Instrument Serif', Georgia, serif" letter-spacing="-0.02em">TCP fallback with UDP auto-disable</text>
+  <text x="360" y="50" text-anchor="middle" font-size="11" fill="#a39888">Latency profile on an ISP that blocks outbound UDP:53</text>
+
+  <!-- Legend -->
+  <g transform="translate(160, 70)">
+    <rect width="14" height="12" rx="2" fill="url(#wasted-hatch)"/>
+    <text x="22" y="10" font-size="11" fill="#6b5e4f">UDP timeout — 800 ms wasted</text>
+    <rect x="220" width="14" height="12" rx="2" fill="#c0623a"/>
+    <text x="242" y="10" font-size="11" fill="#6b5e4f">TCP — successful exchange</text>
+  </g>
+
+  <!-- Time axis -->
+  <!-- bar area: x=90 to x=570 (480px), representing 0-1200ms, scale 0.4 px/ms -->
+  <line x1="90" y1="108" x2="570" y2="108" stroke="#d4cbba" stroke-width="1"/>
+  <!-- tick marks -->
+  <line x1="90" y1="106" x2="90" y2="112" stroke="#a39888" stroke-width="1"/>
+  <line x1="210" y1="106" x2="210" y2="112" stroke="#a39888" stroke-width="1"/>
+  <line x1="330" y1="106" x2="330" y2="112" stroke="#a39888" stroke-width="1"/>
+  <line x1="410" y1="106" x2="410" y2="112" stroke="#a39888" stroke-width="1"/>
+  <line x1="530" y1="106" x2="530" y2="112" stroke="#a39888" stroke-width="1"/>
+  <!-- tick labels -->
+  <text x="90" y="102" text-anchor="middle" font-size="9" fill="#a39888" font-family="'JetBrains Mono', monospace">0</text>
+  <text x="210" y="102" text-anchor="middle" font-size="9" fill="#a39888" font-family="'JetBrains Mono', monospace">300</text>
+  <text x="330" y="102" text-anchor="middle" font-size="9" fill="#a39888" font-family="'JetBrains Mono', monospace">600</text>
+  <text x="410" y="102" text-anchor="middle" font-size="9" fill="#a39888" font-family="'JetBrains Mono', monospace">800</text>
+  <text x="530" y="102" text-anchor="middle" font-size="9" fill="#a39888" font-family="'JetBrains Mono', monospace">1100 ms</text>
+
+  <!-- ============ Phase 1: UDP-first (wasted 800ms per query) ============ -->
+
+  <!-- Query 1 -->
+  <text x="82" y="135" text-anchor="end" font-size="11" fill="#6b5e4f">query 1</text>
+  <rect x="90" y="125" width="320" height="16" rx="2" fill="url(#wasted-hatch)"/>
+  <rect x="410" y="125" width="120" height="16" rx="2" fill="#c0623a"/>
+  <text x="540" y="137" font-size="10" fill="#6b5e4f" font-family="'JetBrains Mono', monospace">1,100 ms</text>
+
+  <!-- Query 2 -->
+  <text x="82" y="159" text-anchor="end" font-size="11" fill="#6b5e4f">query 2</text>
+  <rect x="90" y="149" width="320" height="16" rx="2" fill="url(#wasted-hatch)"/>
+  <rect x="410" y="149" width="120" height="16" rx="2" fill="#c0623a"/>
+  <text x="540" y="161" font-size="10" fill="#6b5e4f" font-family="'JetBrains Mono', monospace">1,100 ms</text>
+
+  <!-- Query 3 -->
+  <text x="82" y="183" text-anchor="end" font-size="11" fill="#6b5e4f">query 3</text>
+  <rect x="90" y="173" width="320" height="16" rx="2" fill="url(#wasted-hatch)"/>
+  <rect x="410" y="173" width="120" height="16" rx="2" fill="#c0623a"/>
+  <text x="540" y="185" font-size="10" fill="#6b5e4f" font-family="'JetBrains Mono', monospace">1,100 ms</text>
+
+  <!-- State-change divider -->
+  <line x1="90" y1="206" x2="570" y2="206" stroke="#6b7c4e" stroke-width="1" stroke-dasharray="4 3"/>
+  <rect x="200" y="198" width="260" height="18" rx="9" fill="#faf7f2" stroke="#6b7c4e" stroke-width="1" filter="url(#shadow)"/>
+  <text x="330" y="210" text-anchor="middle" font-size="10" fill="#566540" font-weight="500">3 consecutive failures → UDP auto-disabled</text>
+
+  <!-- ============ Phase 2: TCP-first (UDP skipped) ============ -->
+
+  <!-- Query 4 -->
+  <text x="82" y="235" text-anchor="end" font-size="11" fill="#6b5e4f">query 4</text>
+  <rect x="90" y="225" width="120" height="16" rx="2" fill="#c0623a"/>
+  <text x="220" y="237" font-size="10" fill="#6b5e4f" font-family="'JetBrains Mono', monospace">300 ms</text>
+
+  <!-- Query 5 -->
+  <text x="82" y="259" text-anchor="end" font-size="11" fill="#6b5e4f">query 5</text>
+  <rect x="90" y="249" width="120" height="16" rx="2" fill="#c0623a"/>
+  <text x="220" y="261" font-size="10" fill="#6b5e4f" font-family="'JetBrains Mono', monospace">300 ms</text>
+
+  <!-- Speedup callout -->
+  <g transform="translate(300, 246)">
+    <line x1="0" y1="-10" x2="0" y2="22" stroke="#6b7c4e" stroke-width="1" stroke-dasharray="2 2"/>
+    <text x="10" y="6" font-size="10" fill="#566540" font-style="italic">3.7× faster — no more UDP wait</text>
+  </g>
+
+  <!-- Footer caption -->
+  <text x="360" y="298" text-anchor="middle" font-size="10" fill="#a39888" font-style="italic">The flag resets on network change (LAN IP delta). Switch back to a clean network and UDP is tried again.</text>
+</svg>

site/blog/index.html

@@ -67,6 +67,7 @@ body::before {
   font-weight: 400;
   color: var(--text-primary);
   text-decoration: none;
+  text-transform: none;
   letter-spacing: -0.02em;
 }
 .blog-nav .wordmark:hover { color: var(--amber); }
@@ -167,6 +168,20 @@ body::before {
 <main class="blog-index">
   <h1>Blog</h1>
   <ul class="post-list">
+    <li>
+      <a href="/blog/posts/fixing-doh-tail-latency.html">
+        <div class="post-title">Fixing DNS tail latency with a 5-line config and a 50-line function</div>
+        <div class="post-desc">Periodic 40-140ms DoH spikes from hyper's dispatch channel. The fix was reqwest window tuning and request hedging — Dean & Barroso's "The Tail at Scale," applied to a DNS forwarder. Same ideas took cold recursive p99 from 2.3 seconds to 538ms.</div>
+        <div class="post-date">April 2026</div>
+      </a>
+    </li>
+    <li>
+      <a href="/blog/posts/dot-from-scratch.html">
+        <div class="post-title">DNS-over-TLS from Scratch in Rust</div>
+        <div class="post-desc">Building RFC 7858 on top of rustls — length-prefix framing, ALPN cross-protocol defense, and two bugs that only the strict clients caught.</div>
+        <div class="post-date">April 2026</div>
+      </a>
+    </li>
     <li>
       <a href="/blog/posts/dnssec-from-scratch.html">
         <div class="post-title">Implementing DNSSEC from Scratch in Rust</div>
@@ -177,7 +192,7 @@ body::before {
     <li>
       <a href="/blog/posts/dns-from-scratch.html">
         <div class="post-title">I Built a DNS Resolver from Scratch in Rust</div>
-        <div class="post-desc">How DNS actually works at the wire level — label compression, TTL tricks, DoH implementation, and what I learned building a resolver with zero DNS libraries.</div>
+        <div class="post-desc">How DNS actually works at the wire level — label compression, TTL tricks, DoH, and what surprised me building a resolver with zero DNS libraries.</div>
         <div class="post-date">March 2026</div>
       </a>
     </li>

site/dashboard.html

@@ -223,6 +223,10 @@ body {
 .path-bar-fill.override { background: var(--emerald); }
 .path-bar-fill.error { background: var(--rose); }
 .path-bar-fill.blocked { background: var(--text-dim); }
+.path-bar-fill.udp { background: var(--text-dim); }
+.path-bar-fill.tcp { background: var(--violet); }
+.path-bar-fill.dot { background: var(--emerald); }
+.path-bar-fill.doh { background: var(--teal); }
 .path-pct {
   font-family: var(--font-mono);
   font-size: 0.75rem;
@@ -288,6 +292,11 @@ body {
 .path-tag.SERVFAIL { background: rgba(181, 68, 58, 0.12); color: var(--rose); }
 .path-tag.BLOCKED { background: rgba(163, 152, 136, 0.15); color: var(--text-dim); }
 .path-tag.COALESCED { background: rgba(138, 104, 158, 0.12); color: var(--violet-dim); }
+.path-tag.UDP { background: rgba(163, 152, 136, 0.15); color: var(--text-dim); }
+.path-tag.TCP { background: rgba(100, 116, 139, 0.12); color: var(--violet-dim); }
+.path-tag.DOT { background: rgba(82, 122, 82, 0.12); color: var(--emerald); }
+.path-tag.DOH { background: rgba(107, 124, 78, 0.12); color: var(--teal); }
+.src-tag { font-size: 0.6rem; color: var(--text-dim); letter-spacing: 0.02em; }
 
 /* Sidebar panels */
 .sidebar {
@@ -553,6 +562,20 @@ body {
     <div class="tagline">DNS that governs itself</div>
   </div>
   <div style="display:flex;align-items:center;gap:1.2rem;">
+    <div id="phoneSetup" style="position:relative;display:none;">
+      <button class="btn" onclick="togglePhoneSetup()" style="background:var(--bg-surface);color:var(--text-secondary);font-family:var(--font-mono);font-size:0.7rem;padding:0.35rem 0.6rem;border:1px solid var(--border);" title="Set up phone">Phone Setup</button>
+      <div id="phoneSetupPopover" style="display:none;position:absolute;top:calc(100% + 8px);right:0;z-index:100;background:var(--bg-card);border:1px solid var(--border);border-radius:10px;padding:1.2rem;width:260px;box-shadow:0 4px 20px rgba(0,0,0,0.08);">
+        <div style="font-size:0.7rem;font-weight:600;text-transform:uppercase;letter-spacing:0.1em;color:var(--text-secondary);margin-bottom:0.8rem;">Phone Setup</div>
+        <div id="qrContainer" style="display:flex;justify-content:center;margin-bottom:0.8rem;"></div>
+        <div id="phoneSetupLink" style="display:none;text-align:center;margin-bottom:0.8rem;"></div>
+        <div style="font-family:var(--font-mono);font-size:0.62rem;color:var(--text-dim);line-height:1.6;">
+          1. Scan QR &rarr; allow download<br>
+          2. Settings &rarr; Profile Downloaded &rarr; Install<br>
+          3. Settings &rarr; General &rarr; About &rarr;<br>
+          &nbsp;&nbsp;&nbsp;Certificate Trust Settings &rarr; toggle ON
+        </div>
+      </div>
+    </div>
     <button class="btn" id="pauseBtn" style="background:var(--amber);color:white;font-family:var(--font-mono);font-size:0.7rem;display:none;">Pause 5m</button>
     <button class="btn" id="toggleBtn" onclick="toggleBlocking()" style="background:var(--rose);color:white;font-family:var(--font-mono);font-size:0.7rem;display:none;"></button>
     <div class="status-badge">
@@ -607,6 +630,16 @@ body {
     </div>
   </div>
 
+  <!-- Transport breakdown -->
+  <div class="panel">
+    <div class="panel-header">
+      <span class="panel-title">Transport</span>
+      <span class="panel-title" id="transportEncrypted" style="color: var(--text-dim)"></span>
+    </div>
+    <div class="panel-body" id="transportBars">
+    </div>
+  </div>
+
   <!-- Main grid: query log + sidebar -->
   <div class="main-grid">
     <!-- Query log -->
@@ -628,6 +661,14 @@ body {
             <option value="LOCAL">local</option>
             <option value="SERVFAIL">error</option>
           </select>
+          <select id="logFilterTransport" onchange="applyLogFilter()"
+            style="font-family:var(--font-mono);font-size:0.7rem;padding:0.25rem 0.4rem;border:1px solid var(--border);border-radius:4px;background:var(--bg-surface);color:var(--text-secondary);outline:none;">
+            <option value="">all transports</option>
+            <option value="UDP">UDP</option>
+            <option value="TCP">TCP</option>
+            <option value="DOT">DoT</option>
+            <option value="DOH">DoH</option>
+          </select>
           <span class="panel-title" id="queryCount" style="color: var(--text-dim)"></span>
         </div>
       </div>
@@ -639,6 +680,7 @@ body {
               <th>Type</th>
               <th>Domain</th>
               <th>Path</th>
+              <th>Transport</th>
               <th>Result</th>
               <th>Latency</th>
             </tr>
@@ -787,6 +829,41 @@ function formatTime(epoch) {
   return d.toLocaleTimeString([], { hour12: false });
 }
 
+let mobilePort = 8765;
+function togglePhoneSetup() {
+  const pop = document.getElementById('phoneSetupPopover');
+  const isOpen = pop.style.display !== 'none';
+  pop.style.display = isOpen ? 'none' : 'block';
+  if (!isOpen) {
+    if (window.innerWidth <= 700) {
+      document.getElementById('qrContainer').style.display = 'none';
+      const linkEl = document.getElementById('phoneSetupLink');
+      const host = window.location.hostname;
+      linkEl.style.display = 'block';
+      linkEl.innerHTML = `<a href="http://${host}:${mobilePort}/mobileconfig" style="display:inline-block;padding:0.5rem 1rem;background:var(--amber);color:white;border-radius:6px;text-decoration:none;font-family:var(--font-mono);font-size:0.75rem;">Install Profile</a>`;
+    } else {
+      fetch(API + '/qr').then(r => r.text()).then(svg => {
+        document.getElementById('qrContainer').innerHTML = svg;
+      }).catch(() => {
+        document.getElementById('qrContainer').innerHTML = '<div class="empty-state">Could not load QR</div>';
+      });
+    }
+  }
+}
+document.addEventListener('click', (e) => {
+  const setup = document.getElementById('phoneSetup');
+  if (setup && !setup.contains(e.target)) {
+    document.getElementById('phoneSetupPopover').style.display = 'none';
+  }
+});
+
+function shortSrc(addr) {
+  if (!addr) return '';
+  const ip = addr.replace(/:\d+$/, '');
+  if (ip === '127.0.0.1' || ip === '::1') return 'localhost';
+  return ip;
+}
+
 function formatRemaining(secs) {
   if (secs == null) return 'permanent';
   if (secs < 60) return `${secs}s left`;
@@ -857,6 +934,27 @@ function renderMemory(mem, stats) {
   `;
 }
 
+function renderBarChart(containerId, defs, data, total) {
+  total = total || 1;
+  document.getElementById(containerId).innerHTML = defs.map(d => {
+    const count = data[d.key] || 0;
+    const pct = ((count / total) * 100).toFixed(1);
+    return `
+      <div class="path-bar-row">
+        <span class="path-label">${d.label}</span>
+        <div class="path-bar-track">
+          <div class="path-bar-fill ${d.cls}" style="width: ${pct}%"></div>
+        </div>
+        <span class="path-pct">${pct}%</span>
+      </div>`;
+  }).join('');
+}
+
+function encryptionPct(transport) {
+  const total = (transport.udp + transport.tcp + transport.dot + transport.doh) || 1;
+  return (((transport.dot + transport.doh) / total) * 100).toFixed(0);
+}
+
 const PATH_DEFS = [
   { key: 'forwarded', label: 'Forward', cls: 'forward' },
   { key: 'recursive', label: 'Recursive', cls: 'recursive' },
@@ -868,20 +966,23 @@ const PATH_DEFS = [
 ];
 
 function renderPaths(queries) {
-  const total = queries.total || 1;
+  renderBarChart('pathBars', PATH_DEFS, queries, queries.total);
-  const container = document.getElementById('pathBars');
+}
-  container.innerHTML = PATH_DEFS.map(p => {
+
-    const count = queries[p.key] || 0;
+const TRANSPORT_DEFS = [
-    const pct = ((count / total) * 100).toFixed(1);
+  { key: 'udp', label: 'UDP',  cls: 'udp' },
-    return `
+  { key: 'tcp', label: 'TCP',  cls: 'tcp' },
-      <div class="path-bar-row">
+  { key: 'dot', label: 'DoT',  cls: 'dot' },
-        <span class="path-label">${p.label}</span>
+  { key: 'doh', label: 'DoH',  cls: 'doh' },
-        <div class="path-bar-track">
+];
-          <div class="path-bar-fill ${p.cls}" style="width: ${pct}%"></div>
+
-        </div>
+function renderTransport(transport) {
-        <span class="path-pct">${pct}%</span>
+  const total = (transport.udp + transport.tcp + transport.dot + transport.doh) || 1;
-      </div>`;
+  renderBarChart('transportBars', TRANSPORT_DEFS, transport, total);
-  }).join('');
+  const encPct = encryptionPct(transport);
+  const el = document.getElementById('transportEncrypted');
+  el.textContent = `${encPct}% encrypted`;
+  el.style.color = encPct >= 80 ? 'var(--emerald)' : encPct >= 50 ? 'var(--amber)' : 'var(--rose)';
 }
 
 function renderQueryLog(entries) {
@@ -892,6 +993,7 @@ function renderQueryLog(entries) {
 function applyLogFilter() {
   const domainFilter = document.getElementById('logFilterDomain').value.trim().toLowerCase();
   const pathFilter = document.getElementById('logFilterPath').value;
+  const transportFilter = document.getElementById('logFilterTransport').value;
 
   let filtered = lastLogEntries;
   if (domainFilter) {
@@ -900,6 +1002,9 @@ function applyLogFilter() {
   if (pathFilter) {
     filtered = filtered.filter(e => e.path === pathFilter);
   }
+  if (transportFilter) {
+    filtered = filtered.filter(e => e.transport === transportFilter);
+  }
 
   const tbody = document.getElementById('queryLogBody');
   document.getElementById('queryCount').textContent =
@@ -912,11 +1017,12 @@ function applyLogFilter() {
       ? ` <button class="btn-delete" onclick="allowDomain('${e.domain}')" title="Allow this domain" style="color:var(--emerald);font-size:0.65rem;">allow</button>`
       : '';
     return `
-    <tr>
+    <tr title="Source: ${e.src || 'unknown'}">
-      <td>${formatTime(e.timestamp_epoch)}</td>
+      <td>${formatTime(e.timestamp_epoch)}<br><span class="src-tag">${shortSrc(e.src)}</span></td>
       <td>${e.query_type}</td>
       <td class="domain-cell" title="${e.domain}">${e.domain}${allowBtn}</td>
       <td><span class="path-tag ${e.path}">${e.path}</span></td>
+      <td><span class="path-tag ${e.transport}">${e.transport}</span></td>
       <td style="white-space:nowrap;"><span style="display:inline-block;width:15px;text-align:center;">${e.dnssec === 'secure' ? '<svg title="DNSSEC verified" width="12" height="12" viewBox="0 0 24 24" fill="none" stroke="var(--emerald)" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round" style="vertical-align:-1px;"><path d="M12 22s8-4 8-10V5l-8-3-8 3v7c0 6 8 10 8 10z"/><path d="m9 12 2 2 4-4"/></svg>' : ''}</span>${e.rescode}</td>
       <td>${e.latency_ms.toFixed(1)}ms</td>
     </tr>`;
@@ -1050,6 +1156,14 @@ async function refresh() {
       }
     }
 
+    const phoneSetupEl = document.getElementById('phoneSetup');
+    if (stats.mobile && stats.mobile.enabled) {
+      phoneSetupEl.style.display = '';
+      mobilePort = stats.mobile.port;
+    } else {
+      phoneSetupEl.style.display = 'none';
+    }
+
     document.getElementById('overrideCount').textContent = stats.overrides.active;
     document.getElementById('blockedCount').textContent = formatNumber(q.blocked);
     const bl = stats.blocking;
@@ -1083,11 +1197,13 @@ async function refresh() {
 
     // QPS calculation
     const now = Date.now();
+    const encPct = encryptionPct(stats.transport);
     if (prevTotal !== null && prevTime !== null) {
       const dt = (now - prevTime) / 1000;
       const dq = q.total - prevTotal;
       const qps = dt > 0 ? (dq / dt).toFixed(1) : '0.0';
-      document.getElementById('qps').textContent = `~${qps}/s`;
+      const encTag = q.total > 0 ? ` · ${encPct}% enc` : '';
+      document.getElementById('qps').textContent = `~${qps}/s${encTag}`;
     }
     prevTotal = q.total;
     prevTime = now;
@@ -1099,6 +1215,7 @@ async function refresh() {
 
     // Panels
     renderPaths(q);
+    renderTransport(stats.transport);
     renderQueryLog(logs);
     renderOverrides(overrides);
     renderCache(cache);

site/index.html

@@ -188,11 +188,50 @@ p.lead {
   line-height: 1.8;
 }
 
+/* ===========================
+   TOP NAV
+   =========================== */
+.site-nav {
+  padding: 1.5rem 2rem;
+  display: flex;
+  align-items: center;
+  gap: 1.5rem;
+  position: relative;
+  z-index: 10;
+}
+
+.site-nav a {
+  font-family: var(--font-mono);
+  font-size: 0.75rem;
+  letter-spacing: 0.08em;
+  text-transform: uppercase;
+  color: var(--text-dim);
+  text-decoration: none;
+  transition: color 0.2s ease;
+}
+.site-nav a:hover { color: var(--amber); }
+
+.site-nav .wordmark {
+  font-family: var(--font-display);
+  font-size: 1.4rem;
+  font-weight: 400;
+  color: var(--text-primary);
+  text-transform: none;
+  letter-spacing: -0.02em;
+}
+.site-nav .wordmark:hover { color: var(--amber); }
+
+.site-nav .sep {
+  color: var(--text-dim);
+  font-family: var(--font-mono);
+  font-size: 0.75rem;
+}
+
 /* ===========================
    HERO
    =========================== */
 .hero {
-  min-height: 100vh;
+  min-height: calc(100vh - 5rem);
   display: flex;
   align-items: center;
   position: relative;
@@ -1158,6 +1197,9 @@ footer .closing {
 @media (max-width: 600px) {
   section { padding: 4rem 0; }
   .container { padding: 0 1.25rem; }
+  .site-nav { padding: 1rem 1.25rem; gap: 1rem; }
+  .site-nav .wordmark { font-size: 1.2rem; }
+  .hero { min-height: calc(100vh - 4rem); }
   .network-grid { grid-template-columns: 1fr; }
   .pipeline { flex-direction: column; align-items: stretch; gap: 0; }
   .pipeline-arrow { transform: rotate(90deg); padding: 0.15rem 0; align-self: center; }
@@ -1171,6 +1213,14 @@ footer .closing {
 </head>
 <body>
 
+<nav class="site-nav">
+  <a href="/" class="wordmark">Numa</a>
+  <span class="sep">/</span>
+  <a href="/blog/">Blog</a>
+  <span class="sep">/</span>
+  <a href="https://github.com/razvandimescu/numa" target="_blank" rel="noopener">GitHub</a>
+</nav>
+
 <!-- ==================== HERO ==================== -->
 <section class="hero">
   <div class="roman-bricks" aria-hidden="true"></div>
@@ -1243,6 +1293,8 @@ footer .closing {
           <li>Ad &amp; tracker blocking &mdash; 385K+ domains, zero config</li>
           <li>Recursive resolution &mdash; opt-in, resolve from root nameservers, no upstream needed</li>
           <li>DNSSEC validation &mdash; chain-of-trust + NSEC/NSEC3 denial proofs (RSA, ECDSA, Ed25519)</li>
+          <li>DNS-over-TLS listener &mdash; encrypted DNS for phones and strict clients (RFC 7858 with ALPN defense)</li>
+          <li>Hostile-network resilience &mdash; TCP fallback with UDP auto-disable when ISPs block port 53</li>
           <li>TTL-aware caching (sub-ms lookups)</li>
           <li>Single binary, portable &mdash; macOS, Linux, and Windows</li>
         </ul>
@@ -1261,7 +1313,7 @@ footer .closing {
         </ul>
       </div>
       <div class="layer-card reveal reveal-delay-3">
-        <div class="layer-badge">Coming Next</div>
+        <div class="layer-badge">The Vision</div>
         <h3>Self-Sovereign DNS</h3>
         <ul>
           <li>pkarr integration &mdash; DNS via Mainline DHT, no registrar needed</li>
@@ -1342,6 +1394,14 @@ footer .closing {
             <td class="cross">No</td>
             <td class="check">Root hints + full DNSSEC</td>
           </tr>
+          <tr>
+            <td>DNSSEC validation</td>
+            <td class="muted">Passthrough</td>
+            <td class="muted">Cloud only</td>
+            <td class="muted">Cloud only</td>
+            <td class="muted">Passthrough</td>
+            <td class="check">Full chain-of-trust</td>
+          </tr>
           <tr>
             <td>Ad &amp; tracker blocking</td>
             <td class="check">Yes</td>
@@ -1398,6 +1458,14 @@ footer .closing {
             <td class="cross">No</td>
             <td class="check">Built in (HTTP/2 + rustls)</td>
           </tr>
+          <tr>
+            <td>DNS-over-TLS listener</td>
+            <td class="cross">No</td>
+            <td class="muted">Cloud only</td>
+            <td class="muted">Cloud only</td>
+            <td class="check">Yes (cert required)</td>
+            <td class="check">Self-signed or BYO</td>
+          </tr>
           <tr>
             <td>Conditional forwarding</td>
             <td class="cross">No</td>
@@ -1567,11 +1635,14 @@ footer .closing {
         <dt>Resolution Modes</dt>
         <dd>Recursive (iterative from root hints, CNAME chasing, glue extraction) or Forward (DoH / plain UDP)</dd>
 
+        <dt>Listeners</dt>
+        <dd>UDP:53 + TCP:53 (plain DNS), DoT:853 (RFC 7858 + ALPN), HTTP proxy :80 / HTTPS proxy :443, dashboard :5380</dd>
+
         <dt>DNSSEC</dt>
         <dd>Chain-of-trust via ring &mdash; RSA/SHA-256, ECDSA P-256, Ed25519. NSEC/NSEC3 denial proofs. EDNS0 DO bit, 1232-byte payload (DNS Flag Day 2020).</dd>
 
         <dt>Dependencies</dt>
-        <dd>19 runtime crates &mdash; tokio, axum, hyper, ring (DNSSEC), reqwest (DoH), rcgen + rustls (TLS), socket2 (multicast), serde, and more</dd>
+        <dd>A focused set &mdash; tokio, axum, hyper, ring (DNSSEC), reqwest (DoH), rcgen + rustls + tokio-rustls (TLS/DoT), socket2 (multicast), serde. No transitive DNS library.</dd>
 
         <dt>Packet Format</dt>
         <dd>RFC 1035 compliant. EDNS0 OPT pseudo-record. Parses A, AAAA, NS, CNAME, MX, SOA, SRV, HTTPS, DNSKEY, DS, RRSIG, NSEC, NSEC3.</dd>
@@ -1586,7 +1657,7 @@ footer .closing {
 <span class="prompt">$</span> <span class="cmd">curl</span> <span class="flag">-fsSL</span> https://raw.githubusercontent.com/razvandimescu/numa/main/install.sh <span class="flag">|</span> <span class="cmd">sh</span>
 
 <span class="comment"># Run</span>
-<span class="prompt">$</span> <span class="cmd">sudo numa</span>                          <span class="comment"># bind to :53, :80, :5380</span>
+<span class="prompt">$</span> <span class="cmd">sudo numa</span>                          <span class="comment"># bind :53, :80, :443, :853, :5380</span>
 <span class="prompt">$</span> <span class="cmd">dig</span> <span class="flag">@127.0.0.1</span> google.com         <span class="comment"># test resolution</span>
 <span class="prompt">$</span> <span class="cmd">open</span> http://localhost:5380           <span class="comment"># dashboard</span>
 <span class="prompt">$</span> <span class="cmd">curl</span> <span class="flag">-X POST</span> localhost:5380/services \
@@ -1639,16 +1710,28 @@ footer .closing {
         <span class="phase">Phase 7</span>
         <span class="phase-desc">DNSSEC validation &mdash; chain-of-trust, NSEC/NSEC3 denial proofs, RSA + ECDSA + Ed25519</span>
       </div>
-      <div class="roadmap-item phase-teal">
+      <div class="roadmap-item done">
         <span class="phase">Phase 8</span>
+        <span class="phase-desc">Hostile-network resilience &mdash; TCP fallback with UDP auto-disable when ISPs block :53, RFC 7816 query minimization</span>
+      </div>
+      <div class="roadmap-item done">
+        <span class="phase">Phase 9</span>
+        <span class="phase-desc">Windows support &mdash; cross-platform install/uninstall, <code>netsh</code> DNS config, service integration</span>
+      </div>
+      <div class="roadmap-item done">
+        <span class="phase">Phase 10</span>
+        <span class="phase-desc">DNS-over-TLS listener (RFC 7858) &mdash; ALPN enforcement, persistent connections, self-signed or BYO cert</span>
+      </div>
+      <div class="roadmap-item phase-teal">
+        <span class="phase">Phase 11</span>
         <span class="phase-desc">pkarr integration &mdash; self-sovereign DNS via Mainline DHT, no registrar needed</span>
       </div>
       <div class="roadmap-item phase-teal">
-        <span class="phase">Phase 9</span>
+        <span class="phase">Phase 12</span>
         <span class="phase-desc">Global .numa names &mdash; self-publish, DHT-backed, first-come-first-served</span>
       </div>
       <div class="roadmap-item phase-teal">
-        <span class="phase">Phase 10</span>
+        <span class="phase">Phase 13</span>
         <span class="phase-desc">.onion bridge &mdash; human-readable Tor naming via Ed25519 same-key binding</span>
       </div>
     </div>

src/api.rs

@@ -57,6 +57,7 @@ pub fn router(ctx: Arc<ServerCtx>) -> Router {
         .route("/services/{name}/routes", post(add_route))
         .route("/services/{name}/routes", delete(remove_route))
         .route("/ca.pem", get(serve_ca))
+        .route("/qr", get(serve_qr))
         .route("/fonts/fonts.css", get(serve_fonts_css))
         .route(
             "/fonts/dm-sans-latin.woff2",
@@ -151,6 +152,7 @@ struct QueryLogResponse {
     domain: String,
     query_type: String,
     path: String,
+    transport: String,
     rescode: String,
     latency_ms: f64,
     dnssec: String,
@@ -166,13 +168,29 @@ struct StatsResponse {
     dnssec: bool,
     srtt: bool,
     queries: QueriesStats,
+    transport: TransportStats,
     cache: CacheStats,
     overrides: OverrideStats,
     blocking: BlockingStatsResponse,
     lan: LanStatsResponse,
+    mobile: MobileStatsResponse,
     memory: MemoryStats,
 }
 
+#[derive(Serialize)]
+struct TransportStats {
+    udp: u64,
+    tcp: u64,
+    dot: u64,
+    doh: u64,
+}
+
+#[derive(Serialize)]
+struct MobileStatsResponse {
+    enabled: bool,
+    port: u16,
+}
+
 #[derive(Serialize)]
 struct LanStatsResponse {
     enabled: bool,
@@ -403,9 +421,12 @@ async fn diagnose(
     }
 
     // Check upstream (async, no locks held)
-    let upstream = ctx.upstream.lock().unwrap().clone();
+    let upstream = ctx.upstream_pool.lock().unwrap().preferred().cloned();
-    let (upstream_matched, upstream_detail) =
+    let (upstream_matched, upstream_detail) = if let Some(ref u) = upstream {
-        forward_query_for_diagnose(&domain_lower, &upstream, ctx.timeout).await;
+        forward_query_for_diagnose(&domain_lower, u, ctx.timeout).await
+    } else {
+        (false, "no upstream configured".to_string())
+    };
     steps.push(DiagnoseStep {
         source: "upstream".to_string(),
         matched: upstream_matched,
@@ -472,6 +493,7 @@ async fn query_log(
                     domain: e.domain.clone(),
                     query_type: e.query_type.as_str().to_string(),
                     path: e.path.as_str().to_string(),
+                    transport: e.transport.as_str().to_string(),
                     rescode: e.rescode.as_str().to_string(),
                     latency_ms: e.latency_us as f64 / 1000.0,
                     dnssec: e.dnssec.as_str().to_string(),
@@ -512,7 +534,7 @@ async fn stats(State(ctx): State<Arc<ServerCtx>>) -> Json<StatsResponse> {
     let upstream = if ctx.upstream_mode == crate::config::UpstreamMode::Recursive {
         "recursive (root hints)".to_string()
     } else {
-        ctx.upstream.lock().unwrap().to_string()
+        ctx.upstream_pool.lock().unwrap().label()
     };
 
     Json(StatsResponse {
@@ -534,6 +556,12 @@ async fn stats(State(ctx): State<Arc<ServerCtx>>) -> Json<StatsResponse> {
             blocked: snap.blocked,
             errors: snap.errors,
         },
+        transport: TransportStats {
+            udp: snap.transport_udp,
+            tcp: snap.transport_tcp,
+            dot: snap.transport_dot,
+            doh: snap.transport_doh,
+        },
         cache: CacheStats {
             entries: cache_len,
             max_entries: cache_max,
@@ -551,6 +579,10 @@ async fn stats(State(ctx): State<Arc<ServerCtx>>) -> Json<StatsResponse> {
             enabled: ctx.lan_enabled,
             peers: ctx.lan_peers.lock().unwrap().list().len(),
         },
+        mobile: MobileStatsResponse {
+            enabled: ctx.mobile_enabled,
+            port: ctx.mobile_port,
+        },
         memory: MemoryStats {
             cache_bytes,
             blocklist_bytes,
@@ -592,8 +624,19 @@ async fn flush_cache_domain(
     StatusCode::NO_CONTENT
 }
 
-async fn health() -> Json<serde_json::Value> {
+/// Enriched `/health` handler shared between the main API and the mobile API.
-    Json(serde_json::json!({ "status": "ok" }))
+///
+/// Returns the cached `HealthMeta` assembled with live fields (LAN IP,
+/// uptime). Backward compatible with the previous minimal response in
+/// that `status` is still the first field and `"ok"` is still the value.
+/// The iOS companion app's `HealthInfo` Swift struct decodes the full
+/// response; any HTTP client asserting only on `"status"` keeps working.
+pub async fn health(State(ctx): State<Arc<ServerCtx>>) -> Json<crate::health::HealthResponse> {
+    let lan_ip = Some(*ctx.lan_ip.lock().unwrap());
+    Json(crate::health::HealthResponse::build(
+        &ctx.health_meta,
+        lan_ip,
+    ))
 }
 
 // --- Blocking handlers ---
@@ -905,12 +948,8 @@ async fn remove_route(
     }
 }
 
-async fn serve_ca(State(ctx): State<Arc<ServerCtx>>) -> Result<impl IntoResponse, StatusCode> {
+pub async fn serve_ca(State(ctx): State<Arc<ServerCtx>>) -> Result<impl IntoResponse, StatusCode> {
-    let ca_path = ctx.data_dir.join("ca.pem");
+    let pem = ctx.ca_pem.as_deref().ok_or(StatusCode::NOT_FOUND)?;
-    let bytes = tokio::task::spawn_blocking(move || std::fs::read(ca_path))
-        .await
-        .map_err(|_| StatusCode::INTERNAL_SERVER_ERROR)?
-        .map_err(|_| StatusCode::NOT_FOUND)?;
     Ok((
         [
             (header::CONTENT_TYPE, "application/x-pem-file"),
@@ -920,7 +959,29 @@ async fn serve_ca(State(ctx): State<Arc<ServerCtx>>) -> Result<impl IntoResponse
             ),
             (header::CACHE_CONTROL, "public, max-age=86400"),
         ],
-        bytes,
+        pem.to_string(),
+    ))
+}
+
+async fn serve_qr(State(ctx): State<Arc<ServerCtx>>) -> Result<impl IntoResponse, StatusCode> {
+    if !ctx.mobile_enabled {
+        return Err(StatusCode::NOT_FOUND);
+    }
+    let lan_ip = *ctx.lan_ip.lock().unwrap();
+    let url = format!("http://{}:{}/mobileconfig", lan_ip, ctx.mobile_port);
+    let code = qrcode::QrCode::new(&url).map_err(|_| StatusCode::INTERNAL_SERVER_ERROR)?;
+    let svg = code
+        .render::<qrcode::render::svg::Color>()
+        .min_dimensions(180, 180)
+        .dark_color(qrcode::render::svg::Color("#2c2418"))
+        .light_color(qrcode::render::svg::Color("#faf7f2"))
+        .build();
+    Ok((
+        [
+            (header::CONTENT_TYPE, "image/svg+xml"),
+            (header::CACHE_CONTROL, "no-store"),
+        ],
+        svg,
     ))
 }
 
@@ -959,44 +1020,10 @@ mod tests {
     use super::*;
     use axum::body::Body;
     use http::Request;
-    use std::sync::{Mutex, RwLock};
     use tower::ServiceExt;
 
     async fn test_ctx() -> Arc<ServerCtx> {
-        let socket = tokio::net::UdpSocket::bind("127.0.0.1:0").await.unwrap();
+        Arc::new(crate::testutil::test_ctx().await)
-        Arc::new(ServerCtx {
-            socket,
-            zone_map: std::collections::HashMap::new(),
-            cache: RwLock::new(crate::cache::DnsCache::new(100, 60, 86400)),
-            stats: Mutex::new(crate::stats::ServerStats::new()),
-            overrides: RwLock::new(crate::override_store::OverrideStore::new()),
-            blocklist: RwLock::new(crate::blocklist::BlocklistStore::new()),
-            query_log: Mutex::new(crate::query_log::QueryLog::new(100)),
-            services: Mutex::new(crate::service_store::ServiceStore::new()),
-            lan_peers: Mutex::new(crate::lan::PeerStore::new(90)),
-            forwarding_rules: Vec::new(),
-            upstream: Mutex::new(crate::forward::Upstream::Udp(
-                "127.0.0.1:53".parse().unwrap(),
-            )),
-            upstream_auto: false,
-            upstream_port: 53,
-            lan_ip: Mutex::new(std::net::Ipv4Addr::LOCALHOST),
-            timeout: std::time::Duration::from_secs(3),
-            proxy_tld: "numa".to_string(),
-            proxy_tld_suffix: ".numa".to_string(),
-            lan_enabled: false,
-            config_path: "/tmp/test-numa.toml".to_string(),
-            config_found: false,
-            config_dir: std::path::PathBuf::from("/tmp"),
-            data_dir: std::path::PathBuf::from("/tmp"),
-            tls_config: None,
-            upstream_mode: crate::config::UpstreamMode::Forward,
-            root_hints: Vec::new(),
-            srtt: RwLock::new(crate::srtt::SrttCache::new(true)),
-            inflight: Mutex::new(std::collections::HashMap::new()),
-            dnssec_enabled: false,
-            dnssec_strict: false,
-        })
     }
 
     #[tokio::test]

src/blocklist.rs

@@ -81,66 +81,70 @@ impl BlocklistStore {
         if !self.enabled {
             return false;
         }
-
         if let Some(until) = self.paused_until {
             if Instant::now() < until {
                 return false;
             }
         }
-
+        let domain = Self::normalize(domain);
-        if self.allowlist.contains(domain) {
+        if Self::find_in_set(&domain, &self.allowlist).is_some() {
             return false;
         }
-
+        Self::find_in_set(&domain, &self.domains).is_some()
-        if self.domains.contains(domain) {
-            return true;
-        }
-
-        // Walk up: ads.tracker.example.com → tracker.example.com → example.com
-        let mut d = domain;
-        while let Some(dot) = d.find('.') {
-            d = &d[dot + 1..];
-            if self.allowlist.contains(d) {
-                return false;
-            }
-            if self.domains.contains(d) {
-                return true;
-            }
-        }
-
-        false
     }
 
-    /// Check if a domain is blocked and return the reason.
     pub fn check(&self, domain: &str) -> BlockCheckResult {
-        let domain = domain.to_lowercase();
-
         if !self.enabled {
             return BlockCheckResult::disabled();
         }
 
-        if self.allowlist.contains(&domain) {
+        if let Some(until) = self.paused_until {
-            return BlockCheckResult::allowed(&domain, "exact match in allowlist");
+            if Instant::now() < until {
+                return BlockCheckResult::disabled();
+            }
         }
 
-        if self.domains.contains(&domain) {
+        let domain = Self::normalize(domain);
-            return BlockCheckResult::blocked(&domain, "exact match in blocklist");
+
+        if let Some(matched) = Self::find_in_set(&domain, &self.allowlist) {
+            let reason = if matched == domain {
+                "exact match in allowlist"
+            } else {
+                "parent domain in allowlist"
+            };
+            return BlockCheckResult::allowed(matched, reason);
         }
 
-        let mut d = domain.as_str();
+        if let Some(matched) = Self::find_in_set(&domain, &self.domains) {
-        while let Some(dot) = d.find('.') {
+            let reason = if matched == domain {
-            d = &d[dot + 1..];
+                "exact match in blocklist"
-            if self.allowlist.contains(d) {
+            } else {
-                return BlockCheckResult::allowed(d, "parent domain in allowlist");
+                "parent domain in blocklist"
-            }
+            };
-            if self.domains.contains(d) {
+            return BlockCheckResult::blocked(matched, reason);
-                return BlockCheckResult::blocked(d, "parent domain in blocklist");
-            }
         }
 
         BlockCheckResult::not_blocked()
     }
 
+    fn normalize(domain: &str) -> String {
+        domain.to_lowercase().trim_end_matches('.').to_string()
+    }
+
+    fn find_in_set<'a>(domain: &'a str, set: &HashSet<String>) -> Option<&'a str> {
+        if set.contains(domain) {
+            return Some(domain);
+        }
+        let mut d = domain;
+        while let Some(dot) = d.find('.') {
+            d = &d[dot + 1..];
+            if set.contains(d) {
+                return Some(d);
+            }
+        }
+        None
+    }
+
     /// Atomically swap in a new domain set. Build the set outside the lock,
     /// then call this to swap — keeps lock hold time sub-microsecond.
     pub fn swap_domains(&mut self, domains: HashSet<String>, sources: Vec<String>) {
@@ -172,11 +176,11 @@ impl BlocklistStore {
     }
 
     pub fn add_to_allowlist(&mut self, domain: &str) {
-        self.allowlist.insert(domain.to_lowercase());
+        self.allowlist.insert(Self::normalize(domain));
     }
 
     pub fn remove_from_allowlist(&mut self, domain: &str) -> bool {
-        self.allowlist.remove(&domain.to_lowercase())
+        self.allowlist.remove(&Self::normalize(domain))
     }
 
     pub fn allowlist(&self) -> Vec<String> {
@@ -247,6 +251,97 @@ pub fn parse_blocklist(text: &str) -> HashSet<String> {
 mod tests {
     use super::*;
 
+    fn store_with(domains: &[&str], allowlist: &[&str]) -> BlocklistStore {
+        let mut store = BlocklistStore::new();
+        store.swap_domains(domains.iter().map(|s| s.to_string()).collect(), vec![]);
+        for d in allowlist {
+            store.add_to_allowlist(d);
+        }
+        store
+    }
+
+    #[test]
+    fn exact_block() {
+        let store = store_with(&["ads.example.com"], &[]);
+        assert!(store.is_blocked("ads.example.com"));
+        assert!(!store.is_blocked("example.com"));
+    }
+
+    #[test]
+    fn parent_block_covers_subdomain() {
+        let store = store_with(&["tracker.com"], &[]);
+        assert!(store.is_blocked("tracker.com"));
+        assert!(store.is_blocked("www.tracker.com"));
+        assert!(store.is_blocked("deep.sub.tracker.com"));
+    }
+
+    #[test]
+    fn exact_allowlist_unblocks() {
+        let store = store_with(&["ads.example.com"], &["ads.example.com"]);
+        assert!(!store.is_blocked("ads.example.com"));
+    }
+
+    #[test]
+    fn parent_allowlist_unblocks_subdomain() {
+        let store = store_with(&["example.com", "www.example.com"], &["example.com"]);
+        assert!(!store.is_blocked("example.com"));
+        assert!(!store.is_blocked("www.example.com"));
+        assert!(!store.is_blocked("sub.deep.example.com"));
+    }
+
+    #[test]
+    fn allowlist_does_not_unblock_sibling() {
+        let store = store_with(
+            &["www.example.com", "ads.example.com"],
+            &["www.example.com"],
+        );
+        assert!(!store.is_blocked("www.example.com"));
+        assert!(store.is_blocked("ads.example.com"));
+    }
+
+    #[test]
+    fn check_reports_parent_allowlist() {
+        let store = store_with(
+            &["goatcounter.com", "www.goatcounter.com"],
+            &["goatcounter.com"],
+        );
+        let result = store.check("www.goatcounter.com");
+        assert!(!result.blocked);
+        assert_eq!(result.matched_rule.as_deref(), Some("goatcounter.com"));
+    }
+
+    #[test]
+    fn disabled_never_blocks() {
+        let mut store = store_with(&["ads.example.com"], &[]);
+        store.set_enabled(false);
+        assert!(!store.is_blocked("ads.example.com"));
+    }
+
+    #[test]
+    fn trailing_dot_normalized() {
+        let store = store_with(&["ads.example.com"], &["safe.example.com"]);
+        assert!(store.is_blocked("ads.example.com."));
+        assert!(!store.is_blocked("safe.example.com."));
+        let result = store.check("ads.example.com.");
+        assert!(result.blocked);
+    }
+
+    #[test]
+    fn case_insensitive() {
+        let store = store_with(&["ads.example.com"], &["safe.example.com"]);
+        assert!(store.is_blocked("ADS.Example.COM"));
+        assert!(!store.is_blocked("Safe.Example.COM"));
+    }
+
+    #[test]
+    fn domain_in_neither_list() {
+        let store = store_with(&["ads.example.com"], &[]);
+        let result = store.check("clean.example.org");
+        assert!(!result.blocked);
+        assert_eq!(result.reason, "not in blocklist");
+        assert!(result.matched_rule.is_none());
+    }
+
     #[test]
     fn heap_bytes_grows_with_domains() {
         let mut store = BlocklistStore::new();

src/buffer.rs

@@ -84,6 +84,11 @@ impl BytePacketBuffer {
 
     /// Read a qname, handling label compression (pointer jumps).
     /// Converts wire format like [3]www[6]google[3]com[0] into "www.google.com".
+    ///
+    /// Label bytes are escaped per RFC 1035 §5.1:
+    /// - literal `.` within a label → `\.`
+    /// - literal `\` → `\\`
+    /// - bytes outside `0x21..=0x7E` (excluding `.` and `\`) → `\DDD` (3-digit decimal)
     pub fn read_qname(&mut self, outstr: &mut String) -> Result<()> {
         let mut pos = self.pos();
         let mut jumped = false;
@@ -121,7 +126,18 @@ impl BytePacketBuffer {
 
                 let str_buffer = self.get_range(pos, len as usize)?;
                 for &b in str_buffer {
-                    outstr.push(b.to_ascii_lowercase() as char);
+                    let c = b.to_ascii_lowercase();
+                    match c {
+                        b'.' => outstr.push_str("\\."),
+                        b'\\' => outstr.push_str("\\\\"),
+                        0x21..=0x7E => outstr.push(c as char),
+                        _ => {
+                            outstr.push('\\');
+                            outstr.push((b'0' + c / 100) as char);
+                            outstr.push((b'0' + (c / 10) % 10) as char);
+                            outstr.push((b'0' + c % 10) as char);
+                        }
+                    }
                 }
 
                 delim = ".";
@@ -163,24 +179,68 @@ impl BytePacketBuffer {
         Ok(())
     }
 
+    /// Write a qname in wire format, parsing RFC 1035 §5.1 text escapes.
+    /// See `read_qname` for the escape grammar.
     pub fn write_qname(&mut self, qname: &str) -> Result<()> {
         if qname.is_empty() || qname == "." {
             self.write_u8(0)?;
             return Ok(());
         }
 
-        for label in qname.split('.') {
+        let bytes = qname.as_bytes();
-            let len = label.len();
+        let mut i = 0;
-            if len == 0 {
+        while i < bytes.len() {
-                continue; // skip empty labels from trailing dot
+            let len_pos = self.pos;
-            }
+            self.write_u8(0)?; // placeholder length byte, backpatched below
-            if len > 0x3f {
+            let body_start = self.pos;
-                return Err("Single label exceeds 63 characters of length".into());
+
+            while i < bytes.len() && bytes[i] != b'.' {
+                let b = bytes[i];
+                if b == b'\\' {
+                    i += 1;
+                    let c1 = *bytes.get(i).ok_or("trailing backslash in qname")?;
+                    if c1.is_ascii_digit() {
+                        let c2 = *bytes
+                            .get(i + 1)
+                            .ok_or("invalid \\DDD escape: expected 3 digits")?;
+                        let c3 = *bytes
+                            .get(i + 2)
+                            .ok_or("invalid \\DDD escape: expected 3 digits")?;
+                        if !c2.is_ascii_digit() || !c3.is_ascii_digit() {
+                            return Err("invalid \\DDD escape: expected 3 digits".into());
+                        }
+                        let val =
+                            (c1 - b'0') as u16 * 100 + (c2 - b'0') as u16 * 10 + (c3 - b'0') as u16;
+                        if val > 255 {
+                            return Err(format!("\\DDD escape out of range: {}", val).into());
+                        }
+                        self.write_u8(val as u8)?;
+                        i += 3;
+                    } else {
+                        // \. \\ and any other \X → literal next byte
+                        self.write_u8(c1)?;
+                        i += 1;
+                    }
+                } else {
+                    self.write_u8(b)?;
+                    i += 1;
+                }
+
+                if self.pos - body_start > 0x3f {
+                    return Err("Single label exceeds 63 characters of length".into());
+                }
             }
 
-            self.write_u8(len as u8)?;
+            let label_len = self.pos - body_start;
-            for b in label.as_bytes() {
+            if label_len == 0 && i < bytes.len() {
-                self.write_u8(*b)?;
+                // Empty label from leading/consecutive dots — roll back the placeholder.
+                self.pos = len_pos;
+            } else {
+                self.set(len_pos, label_len as u8)?;
+            }
+
+            if i < bytes.len() && bytes[i] == b'.' {
+                i += 1;
             }
         }
 
@@ -212,3 +272,160 @@ impl BytePacketBuffer {
         Ok(())
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn roundtrip(wire: &[u8]) -> String {
+        let mut buf = BytePacketBuffer::from_bytes(wire);
+        let mut out = String::new();
+        buf.read_qname(&mut out).unwrap();
+        out
+    }
+
+    fn write_then_read(text: &str) -> String {
+        let mut buf = BytePacketBuffer::new();
+        buf.write_qname(text).unwrap();
+        let wire_end = buf.pos();
+        buf.seek(0).unwrap();
+        let mut out = String::new();
+        buf.read_qname(&mut out).unwrap();
+        assert_eq!(
+            buf.pos(),
+            wire_end,
+            "reader should consume exactly what writer wrote"
+        );
+        out
+    }
+
+    #[test]
+    fn read_plain_domain() {
+        // [3]www[6]google[3]com[0]
+        let wire = b"\x03www\x06google\x03com\x00";
+        assert_eq!(roundtrip(wire), "www.google.com");
+    }
+
+    #[test]
+    fn read_label_with_literal_dot_is_escaped() {
+        // fanf2's example: [8]exa.mple[3]com[0] — two labels, first contains 0x2E
+        let wire = b"\x08exa.mple\x03com\x00";
+        assert_eq!(roundtrip(wire), "exa\\.mple.com");
+    }
+
+    #[test]
+    fn read_label_with_backslash_is_escaped() {
+        // [4]a\bc[3]com[0]
+        let wire = b"\x04a\\bc\x03com\x00";
+        assert_eq!(roundtrip(wire), "a\\\\bc.com");
+    }
+
+    #[test]
+    fn read_label_with_nonprintable_byte_uses_decimal_escape() {
+        // [4]\x00foo[3]com[0] — null byte at label start
+        let wire = b"\x04\x00foo\x03com\x00";
+        assert_eq!(roundtrip(wire), "\\000foo.com");
+    }
+
+    #[test]
+    fn read_label_with_space_uses_decimal_escape() {
+        // Space (0x20) is outside 0x21..=0x7E, so it must be decimal-escaped.
+        let wire = b"\x05a b c\x00";
+        assert_eq!(roundtrip(wire), "a\\032b\\032c");
+    }
+
+    #[test]
+    fn write_plain_domain() {
+        let mut buf = BytePacketBuffer::new();
+        buf.write_qname("www.google.com").unwrap();
+        assert_eq!(&buf.buf[..buf.pos], b"\x03www\x06google\x03com\x00");
+    }
+
+    #[test]
+    fn write_escaped_dot_does_not_split_label() {
+        let mut buf = BytePacketBuffer::new();
+        buf.write_qname("exa\\.mple.com").unwrap();
+        assert_eq!(&buf.buf[..buf.pos], b"\x08exa.mple\x03com\x00");
+    }
+
+    #[test]
+    fn write_escaped_backslash() {
+        let mut buf = BytePacketBuffer::new();
+        buf.write_qname("a\\\\bc.com").unwrap();
+        assert_eq!(&buf.buf[..buf.pos], b"\x04a\\bc\x03com\x00");
+    }
+
+    #[test]
+    fn write_decimal_escape_yields_raw_byte() {
+        let mut buf = BytePacketBuffer::new();
+        buf.write_qname("\\000foo.com").unwrap();
+        assert_eq!(&buf.buf[..buf.pos], b"\x04\x00foo\x03com\x00");
+    }
+
+    #[test]
+    fn write_skips_empty_labels() {
+        // Leading dot — first (empty) label is rolled back.
+        let mut buf = BytePacketBuffer::new();
+        buf.write_qname(".foo.com").unwrap();
+        assert_eq!(&buf.buf[..buf.pos], b"\x03foo\x03com\x00");
+
+        // Consecutive dots — middle empty label is rolled back.
+        let mut buf = BytePacketBuffer::new();
+        buf.write_qname("foo..com").unwrap();
+        assert_eq!(&buf.buf[..buf.pos], b"\x03foo\x03com\x00");
+    }
+
+    #[test]
+    fn write_rejects_out_of_range_decimal_escape() {
+        let mut buf = BytePacketBuffer::new();
+        assert!(buf.write_qname("\\999foo.com").is_err());
+    }
+
+    #[test]
+    fn write_rejects_trailing_backslash() {
+        let mut buf = BytePacketBuffer::new();
+        assert!(buf.write_qname("foo\\").is_err());
+    }
+
+    #[test]
+    fn write_rejects_short_decimal_escape() {
+        let mut buf = BytePacketBuffer::new();
+        assert!(buf.write_qname("\\1").is_err());
+    }
+
+    #[test]
+    fn write_rejects_label_over_63_bytes() {
+        // 64 bytes exceeds the wire-format label cap.
+        let mut buf = BytePacketBuffer::new();
+        assert!(buf.write_qname(&"a".repeat(64)).is_err());
+
+        // 63 bytes is the maximum permitted label length.
+        let mut buf = BytePacketBuffer::new();
+        assert!(buf.write_qname(&"a".repeat(63)).is_ok());
+    }
+
+    #[test]
+    fn roundtrip_preserves_dot_in_label() {
+        assert_eq!(write_then_read("exa\\.mple.com"), "exa\\.mple.com");
+    }
+
+    #[test]
+    fn roundtrip_preserves_backslash_in_label() {
+        assert_eq!(write_then_read("a\\\\b.com"), "a\\\\b.com");
+    }
+
+    #[test]
+    fn roundtrip_preserves_nonprintable_byte() {
+        assert_eq!(write_then_read("\\000foo.com"), "\\000foo.com");
+    }
+
+    #[test]
+    fn root_name_empty_and_dot_both_produce_single_zero() {
+        let mut a = BytePacketBuffer::new();
+        a.write_qname("").unwrap();
+        let mut b = BytePacketBuffer::new();
+        b.write_qname(".").unwrap();
+        assert_eq!(&a.buf[..a.pos], b"\x00");
+        assert_eq!(&b.buf[..b.pos], b"\x00");
+    }
+}

src/cache.rs

@@ -1,9 +1,26 @@
 use std::collections::HashMap;
 use std::time::{Duration, Instant};
 
+use crate::buffer::BytePacketBuffer;
 use crate::packet::DnsPacket;
 use crate::question::QueryType;
-use crate::record::DnsRecord;
+use crate::wire::WireMeta;
+
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub enum Freshness {
+    /// Within TTL, no action needed.
+    Fresh,
+    /// Within TTL but <10% remaining — trigger background prefetch.
+    NearExpiry,
+    /// Past TTL but within stale window — serve with TTL=1, trigger background refresh.
+    Stale,
+}
+
+impl Freshness {
+    pub fn needs_refresh(self) -> bool {
+        matches!(self, Freshness::NearExpiry | Freshness::Stale)
+    }
+}
 
 #[derive(Clone, Copy, Debug, Default, PartialEq, Eq)]
 pub enum DnssecStatus {
@@ -26,14 +43,16 @@ impl DnssecStatus {
 }
 
 struct CacheEntry {
-    packet: DnsPacket,
+    wire: Vec<u8>,
+    meta: WireMeta,
     inserted_at: Instant,
     ttl: Duration,
     dnssec_status: DnssecStatus,
 }
 
-/// DNS cache using a two-level map (domain -> query_type -> entry) so that
+const STALE_WINDOW: Duration = Duration::from_secs(3600);
-/// lookups can borrow `&str` instead of allocating a `String` key.
+
+/// DNS cache with serve-stale (RFC 8767). Stores raw wire bytes.
 pub struct DnsCache {
     entries: HashMap<String, HashMap<QueryType, CacheEntry>>,
     entry_count: usize,
@@ -53,54 +72,60 @@ impl DnsCache {
         }
     }
 
-    /// Read-only lookup — expired entries are left in place (cleaned up on insert).
+    /// Look up cached wire bytes, patching ID and TTLs in the returned copy.
-    pub fn lookup(&self, domain: &str, qtype: QueryType) -> Option<DnsPacket> {
+    /// Implements serve-stale (RFC 8767): expired entries within STALE_WINDOW
-        self.lookup_with_status(domain, qtype).map(|(pkt, _)| pkt)
+    /// are returned with TTL=1 and `stale=true` so callers can revalidate.
-    }
+    pub fn lookup_wire(
-
-    pub fn lookup_with_status(
         &self,
         domain: &str,
         qtype: QueryType,
-    ) -> Option<(DnsPacket, DnssecStatus)> {
+        new_id: u16,
+    ) -> Option<(Vec<u8>, DnssecStatus, Freshness)> {
         let type_map = self.entries.get(domain)?;
         let entry = type_map.get(&qtype)?;
 
         let elapsed = entry.inserted_at.elapsed();
-        if elapsed >= entry.ttl {
+        let (remaining, freshness) = if elapsed < entry.ttl {
+            let secs = (entry.ttl - elapsed).as_secs() as u32;
+            let f = if elapsed * 10 >= entry.ttl * 9 {
+                Freshness::NearExpiry
+            } else {
+                Freshness::Fresh
+            };
+            (secs.max(1), f)
+        } else if elapsed < entry.ttl + STALE_WINDOW {
+            (1, Freshness::Stale)
+        } else {
             return None;
-        }
+        };
 
-        let remaining_secs = (entry.ttl - elapsed).as_secs() as u32;
+        let mut wire = entry.wire.clone();
-        let remaining = remaining_secs.max(1);
+        crate::wire::patch_id(&mut wire, new_id);
+        crate::wire::patch_ttls(&mut wire, &entry.meta.ttl_offsets, remaining);
 
-        let mut packet = entry.packet.clone();
+        Some((wire, entry.dnssec_status, freshness))
-        adjust_ttls(&mut packet.answers, remaining);
-        adjust_ttls(&mut packet.authorities, remaining);
-        adjust_ttls(&mut packet.resources, remaining);
-
-        Some((packet, entry.dnssec_status))
     }
 
-    pub fn insert(&mut self, domain: &str, qtype: QueryType, packet: &DnsPacket) {
+    pub fn insert_wire(
-        self.insert_with_status(domain, qtype, packet, DnssecStatus::Indeterminate);
-    }
-
-    pub fn insert_with_status(
         &mut self,
         domain: &str,
         qtype: QueryType,
-        packet: &DnsPacket,
+        wire: &[u8],
         dnssec_status: DnssecStatus,
     ) {
+        let meta = match crate::wire::scan_ttl_offsets(wire) {
+            Ok(m) => m,
+            Err(_) => return, // malformed wire, skip
+        };
+
         if self.entry_count >= self.max_entries {
             self.evict_expired();
             if self.entry_count >= self.max_entries {
-                return;
+                self.evict_stalest();
             }
         }
 
-        let min_ttl = extract_min_ttl(&packet.answers)
+        let min_ttl = crate::wire::min_ttl_from_wire(wire, &meta)
             .unwrap_or(self.min_ttl)
             .clamp(self.min_ttl, self.max_ttl);
 
@@ -117,7 +142,8 @@ impl DnsCache {
         type_map.insert(
             qtype,
             CacheEntry {
-                packet: packet.clone(),
+                wire: wire.to_vec(),
+                meta,
                 inserted_at: Instant::now(),
                 ttl: Duration::from_secs(min_ttl as u64),
                 dnssec_status,
@@ -125,6 +151,64 @@ impl DnsCache {
         );
     }
 
+    /// Read-only lookup — expired entries are left in place (cleaned up on insert).
+    pub fn lookup(&self, domain: &str, qtype: QueryType) -> Option<DnsPacket> {
+        self.lookup_with_status(domain, qtype)
+            .map(|(pkt, _, _)| pkt)
+    }
+
+    pub fn lookup_with_status(
+        &self,
+        domain: &str,
+        qtype: QueryType,
+    ) -> Option<(DnsPacket, DnssecStatus, Freshness)> {
+        let (wire, status, freshness) = self.lookup_wire(domain, qtype, 0)?;
+        let mut buf = BytePacketBuffer::from_bytes(&wire);
+        let pkt = DnsPacket::from_buffer(&mut buf).ok()?;
+        Some((pkt, status, freshness))
+    }
+
+    pub fn insert(&mut self, domain: &str, qtype: QueryType, packet: &DnsPacket) {
+        self.insert_with_status(domain, qtype, packet, DnssecStatus::Indeterminate);
+    }
+
+    pub fn insert_with_status(
+        &mut self,
+        domain: &str,
+        qtype: QueryType,
+        packet: &DnsPacket,
+        dnssec_status: DnssecStatus,
+    ) {
+        let mut buf = BytePacketBuffer::new();
+        if packet.write(&mut buf).is_err() {
+            return;
+        }
+        self.insert_wire(domain, qtype, buf.filled(), dnssec_status);
+    }
+
+    pub fn ttl_remaining(&self, domain: &str, qtype: QueryType) -> Option<(u32, u32)> {
+        let type_map = self.entries.get(domain)?;
+        let entry = type_map.get(&qtype)?;
+        let elapsed = entry.inserted_at.elapsed();
+        if elapsed >= entry.ttl {
+            return None;
+        }
+        let total = entry.ttl.as_secs() as u32;
+        let remaining = (entry.ttl - elapsed).as_secs() as u32;
+        Some((remaining, total))
+    }
+
+    pub fn needs_warm(&self, domain: &str) -> bool {
+        for qtype in [QueryType::A, QueryType::AAAA] {
+            match self.ttl_remaining(domain, qtype) {
+                None => return true,
+                Some((remaining, total)) if remaining < total / 4 => return true,
+                _ => {}
+            }
+        }
+        false
+    }
+
     pub fn len(&self) -> usize {
         self.entry_count
     }
@@ -156,7 +240,8 @@ impl DnsCache {
                 + 1;
             total += type_map.capacity() * inner_slot;
             for entry in type_map.values() {
-                total += entry.packet.heap_bytes();
+                total += entry.wire.capacity()
+                    + entry.meta.ttl_offsets.capacity() * std::mem::size_of::<usize>();
             }
         }
         total
@@ -197,6 +282,34 @@ impl DnsCache {
         });
         self.entry_count -= count;
     }
+
+    /// Evict the single entry closest to (or furthest past) expiry.
+    fn evict_stalest(&mut self) {
+        let mut worst: Option<(String, QueryType, Duration)> = None;
+        for (domain, type_map) in &self.entries {
+            for (qtype, entry) in type_map {
+                let age = entry.inserted_at.elapsed();
+                let remaining = entry.ttl.saturating_sub(age);
+                match &worst {
+                    None => worst = Some((domain.clone(), *qtype, remaining)),
+                    Some((_, _, w)) if remaining < *w => {
+                        worst = Some((domain.clone(), *qtype, remaining));
+                    }
+                    _ => {}
+                }
+            }
+        }
+        if let Some((domain, qtype, _)) = worst {
+            if let Some(type_map) = self.entries.get_mut(&domain) {
+                if type_map.remove(&qtype).is_some() {
+                    self.entry_count -= 1;
+                }
+                if type_map.is_empty() {
+                    self.entries.remove(&domain);
+                }
+            }
+        }
+    }
 }
 
 pub struct CacheInfo {
@@ -205,20 +318,11 @@ pub struct CacheInfo {
     pub ttl_remaining: u32,
 }
 
-fn extract_min_ttl(records: &[DnsRecord]) -> Option<u32> {
-    records.iter().map(|r| r.ttl()).min()
-}
-
-fn adjust_ttls(records: &mut [DnsRecord], new_ttl: u32) {
-    for record in records.iter_mut() {
-        record.set_ttl(new_ttl);
-    }
-}
-
 #[cfg(test)]
 mod tests {
     use super::*;
     use crate::packet::DnsPacket;
+    use crate::record::DnsRecord;
 
     #[test]
     fn heap_bytes_grows_with_entries() {
@@ -233,4 +337,66 @@ mod tests {
         cache.insert("example.com", QueryType::A, &pkt);
         assert!(cache.heap_bytes() > empty);
     }
+
+    #[test]
+    fn ttl_remaining_returns_values_for_fresh_entry() {
+        let mut cache = DnsCache::new(100, 60, 3600);
+        let mut pkt = DnsPacket::new();
+        pkt.answers.push(DnsRecord::A {
+            domain: "example.com".into(),
+            addr: "1.2.3.4".parse().unwrap(),
+            ttl: 300,
+        });
+        cache.insert("example.com", QueryType::A, &pkt);
+        let (remaining, total) = cache.ttl_remaining("example.com", QueryType::A).unwrap();
+        assert_eq!(total, 300);
+        assert!(remaining <= 300);
+        assert!(remaining > 0);
+    }
+
+    #[test]
+    fn ttl_remaining_none_for_missing() {
+        let cache = DnsCache::new(100, 1, 3600);
+        assert!(cache.ttl_remaining("missing.com", QueryType::A).is_none());
+    }
+
+    #[test]
+    fn needs_warm_true_when_missing() {
+        let cache = DnsCache::new(100, 1, 3600);
+        assert!(cache.needs_warm("missing.com"));
+    }
+
+    #[test]
+    fn needs_warm_false_when_fresh() {
+        let mut cache = DnsCache::new(100, 1, 3600);
+        let mut pkt_a = DnsPacket::new();
+        pkt_a.answers.push(DnsRecord::A {
+            domain: "example.com".into(),
+            addr: "1.2.3.4".parse().unwrap(),
+            ttl: 300,
+        });
+        let mut pkt_aaaa = DnsPacket::new();
+        pkt_aaaa.answers.push(DnsRecord::AAAA {
+            domain: "example.com".into(),
+            addr: "::1".parse().unwrap(),
+            ttl: 300,
+        });
+        cache.insert("example.com", QueryType::A, &pkt_a);
+        cache.insert("example.com", QueryType::AAAA, &pkt_aaaa);
+        assert!(!cache.needs_warm("example.com"));
+    }
+
+    #[test]
+    fn needs_warm_true_when_only_a_cached() {
+        let mut cache = DnsCache::new(100, 1, 3600);
+        let mut pkt = DnsPacket::new();
+        pkt.answers.push(DnsRecord::A {
+            domain: "example.com".into(),
+            addr: "1.2.3.4".parse().unwrap(),
+            ttl: 300,
+        });
+        cache.insert("example.com", QueryType::A, &pkt);
+        // AAAA missing → needs warm
+        assert!(cache.needs_warm("example.com"));
+    }
 }

src/config.rs

@@ -31,6 +31,41 @@ pub struct Config {
     pub dnssec: DnssecConfig,
     #[serde(default)]
     pub dot: DotConfig,
+    #[serde(default)]
+    pub mobile: MobileConfig,
+    #[serde(default)]
+    pub forwarding: Vec<ForwardingRuleConfig>,
+}
+
+#[derive(Deserialize, Clone, Debug)]
+pub struct ForwardingRuleConfig {
+    #[serde(deserialize_with = "string_or_vec")]
+    pub suffix: Vec<String>,
+    pub upstream: String,
+}
+
+impl ForwardingRuleConfig {
+    fn to_runtime_rules(&self) -> Result<Vec<crate::system_dns::ForwardingRule>> {
+        let upstream = crate::forward::parse_upstream(&self.upstream, 53)
+            .map_err(|e| format!("forwarding rule for upstream '{}': {}", self.upstream, e))?;
+        Ok(self
+            .suffix
+            .iter()
+            .map(|s| crate::system_dns::ForwardingRule::new(s.clone(), upstream.clone()))
+            .collect())
+    }
+}
+
+pub fn merge_forwarding_rules(
+    config_rules: &[ForwardingRuleConfig],
+    discovered: Vec<crate::system_dns::ForwardingRule>,
+) -> Result<Vec<crate::system_dns::ForwardingRule>> {
+    let mut merged: Vec<crate::system_dns::ForwardingRule> = Vec::new();
+    for rule in config_rules {
+        merged.extend(rule.to_runtime_rules()?);
+    }
+    merged.extend(discovered);
+    Ok(merged)
 }
 
 #[derive(Deserialize)]
@@ -95,12 +130,16 @@ impl UpstreamMode {
 pub struct UpstreamConfig {
     #[serde(default)]
     pub mode: UpstreamMode,
-    #[serde(default = "default_upstream_addr")]
+    #[serde(default, deserialize_with = "string_or_vec")]
-    pub address: String,
+    pub address: Vec<String>,
     #[serde(default = "default_upstream_port")]
     pub port: u16,
+    #[serde(default)]
+    pub fallback: Vec<String>,
     #[serde(default = "default_timeout_ms")]
     pub timeout_ms: u64,
+    #[serde(default = "default_hedge_ms")]
+    pub hedge_ms: u64,
     #[serde(default = "default_root_hints")]
     pub root_hints: Vec<String>,
     #[serde(default = "default_prime_tlds")]
@@ -113,9 +152,11 @@ impl Default for UpstreamConfig {
     fn default() -> Self {
         UpstreamConfig {
             mode: UpstreamMode::default(),
-            address: default_upstream_addr(),
+            address: Vec::new(),
             port: default_upstream_port(),
+            fallback: Vec::new(),
             timeout_ms: default_timeout_ms(),
+            hedge_ms: default_hedge_ms(),
             root_hints: default_root_hints(),
             prime_tlds: default_prime_tlds(),
             srtt: default_srtt(),
@@ -123,6 +164,33 @@ impl Default for UpstreamConfig {
     }
 }
 
+fn string_or_vec<'de, D>(deserializer: D) -> std::result::Result<Vec<String>, D::Error>
+where
+    D: serde::Deserializer<'de>,
+{
+    struct Visitor;
+    impl<'de> serde::de::Visitor<'de> for Visitor {
+        type Value = Vec<String>;
+        fn expecting(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
+            f.write_str("string or array of strings")
+        }
+        fn visit_str<E: serde::de::Error>(self, v: &str) -> std::result::Result<Self::Value, E> {
+            Ok(vec![v.to_string()])
+        }
+        fn visit_seq<A: serde::de::SeqAccess<'de>>(
+            self,
+            mut seq: A,
+        ) -> std::result::Result<Self::Value, A::Error> {
+            let mut v = Vec::new();
+            while let Some(s) = seq.next_element::<String>()? {
+                v.push(s);
+            }
+            Ok(v)
+        }
+    }
+    deserializer.deserialize_any(Visitor)
+}
+
 fn default_true() -> bool {
     true
 }
@@ -200,15 +268,15 @@ fn default_root_hints() -> Vec<String> {
     ]
 }
 
-fn default_upstream_addr() -> String {
-    String::new() // empty = auto-detect from system resolver
-}
 fn default_upstream_port() -> u16 {
     53
 }
 fn default_timeout_ms() -> u64 {
     5000
 }
+fn default_hedge_ms() -> u64 {
+    10
+}
 
 #[derive(Deserialize)]
 pub struct CacheConfig {
@@ -218,6 +286,8 @@ pub struct CacheConfig {
     pub min_ttl: u32,
     #[serde(default = "default_max_ttl")]
     pub max_ttl: u32,
+    #[serde(default)]
+    pub warm: Vec<String>,
 }
 
 impl Default for CacheConfig {
@@ -226,12 +296,13 @@ impl Default for CacheConfig {
             max_entries: default_max_entries(),
             min_ttl: default_min_ttl(),
             max_ttl: default_max_ttl(),
+            warm: Vec::new(),
         }
     }
 }
 
 fn default_max_entries() -> usize {
-    10000
+    100_000
 }
 fn default_min_ttl() -> u32 {
     60
@@ -379,7 +450,7 @@ pub struct DnssecConfig {
 
 #[derive(Deserialize, Clone)]
 pub struct DotConfig {
-    #[serde(default)]
+    #[serde(default = "default_dot_enabled")]
     pub enabled: bool,
     #[serde(default = "default_dot_port")]
     pub port: u16,
@@ -396,7 +467,7 @@ pub struct DotConfig {
 impl Default for DotConfig {
     fn default() -> Self {
         DotConfig {
-            enabled: false,
+            enabled: default_dot_enabled(),
             port: default_dot_port(),
             bind_addr: default_dot_bind_addr(),
             cert_path: None,
@@ -405,6 +476,9 @@ impl Default for DotConfig {
     }
 }
 
+fn default_dot_enabled() -> bool {
+    true
+}
 fn default_dot_port() -> u16 {
     853
 }
@@ -412,6 +486,53 @@ fn default_dot_bind_addr() -> String {
     "0.0.0.0".to_string()
 }
 
+/// Configuration for the mobile API — a persistent HTTP listener that
+/// serves a read-only subset of routes (`/health`, `/ca.pem`,
+/// `/mobileconfig`, `/ca.mobileconfig`) on a LAN-reachable port, for
+/// consumption by the iOS/Android companion apps.
+///
+/// Unlike the main API (port 5380, localhost-only by default, supports
+/// state-mutating routes), the mobile API is safe to expose on the LAN
+/// because every route is idempotent and read-only.
+#[derive(Deserialize, Clone)]
+pub struct MobileConfig {
+    /// If true, spawn the mobile API listener at startup. **Default false.**
+    /// Opt-in because the listener binds to the LAN by default and exposes
+    /// a few read-only endpoints to any device on the same network (`/health`,
+    /// `/ca.pem`, `/mobileconfig`, `/ca.mobileconfig`). None of those are
+    /// cryptographically sensitive (the CA private key is never served),
+    /// but users should enable this explicitly rather than have a new
+    /// LAN-reachable port appear after an upgrade.
+    #[serde(default)]
+    pub enabled: bool,
+    /// Port for the mobile API. Default 8765.
+    #[serde(default = "default_mobile_port")]
+    pub port: u16,
+    /// Bind address for the mobile API. Default "0.0.0.0" (all interfaces)
+    /// so phones on the LAN can reach it. Set to "127.0.0.1" to restrict
+    /// to localhost — useful if you're running behind another front-end.
+    #[serde(default = "default_mobile_bind_addr")]
+    pub bind_addr: String,
+}
+
+impl Default for MobileConfig {
+    fn default() -> Self {
+        MobileConfig {
+            enabled: false,
+            port: default_mobile_port(),
+            bind_addr: default_mobile_bind_addr(),
+        }
+    }
+}
+
+fn default_mobile_port() -> u16 {
+    8765
+}
+
+fn default_mobile_bind_addr() -> String {
+    "0.0.0.0".to_string()
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -476,6 +597,220 @@ mod tests {
         assert!(config.services[0].routes[0].strip);
         assert!(!config.services[0].routes[1].strip); // default false
     }
+
+    #[test]
+    fn address_string_parses_to_vec() {
+        let config: Config = toml::from_str("[upstream]\naddress = \"1.2.3.4\"").unwrap();
+        assert_eq!(config.upstream.address, vec!["1.2.3.4"]);
+    }
+
+    #[test]
+    fn address_array_parses() {
+        let config: Config =
+            toml::from_str("[upstream]\naddress = [\"1.2.3.4\", \"5.6.7.8:5353\"]").unwrap();
+        assert_eq!(config.upstream.address, vec!["1.2.3.4", "5.6.7.8:5353"]);
+    }
+
+    #[test]
+    fn fallback_parses() {
+        let config: Config =
+            toml::from_str("[upstream]\nfallback = [\"8.8.8.8\", \"1.1.1.1\"]").unwrap();
+        assert_eq!(config.upstream.fallback, vec!["8.8.8.8", "1.1.1.1"]);
+    }
+
+    #[test]
+    fn empty_address_gives_empty_vec() {
+        let config: Config = toml::from_str("").unwrap();
+        assert!(config.upstream.address.is_empty());
+        assert!(config.upstream.fallback.is_empty());
+    }
+
+    // ── issue #82: [[forwarding]] config section ────────────────────────
+
+    #[test]
+    fn forwarding_empty_by_default() {
+        let config: Config = toml::from_str("").unwrap();
+        assert!(config.forwarding.is_empty());
+    }
+
+    #[test]
+    fn forwarding_parses_single_rule() {
+        let toml = r#"
+            [[forwarding]]
+            suffix = "home.local"
+            upstream = "100.90.1.63:5361"
+        "#;
+        let config: Config = toml::from_str(toml).unwrap();
+        assert_eq!(config.forwarding.len(), 1);
+        assert_eq!(config.forwarding[0].suffix, &["home.local"]);
+        assert_eq!(config.forwarding[0].upstream, "100.90.1.63:5361");
+    }
+
+    #[test]
+    fn forwarding_parses_reverse_dns_zone() {
+        let toml = r#"
+            [[forwarding]]
+            suffix = "168.192.in-addr.arpa"
+            upstream = "100.90.1.63:5361"
+        "#;
+        let config: Config = toml::from_str(toml).unwrap();
+        assert_eq!(config.forwarding.len(), 1);
+        assert_eq!(config.forwarding[0].suffix, &["168.192.in-addr.arpa"]);
+    }
+
+    #[test]
+    fn forwarding_parses_multiple_rules() {
+        let toml = r#"
+            [[forwarding]]
+            suffix = "168.192.in-addr.arpa"
+            upstream = "100.90.1.63:5361"
+
+            [[forwarding]]
+            suffix = "home.local"
+            upstream = "10.0.0.1"
+        "#;
+        let config: Config = toml::from_str(toml).unwrap();
+        assert_eq!(config.forwarding.len(), 2);
+        assert_eq!(config.forwarding[1].upstream, "10.0.0.1");
+    }
+
+    #[test]
+    fn forwarding_parses_suffix_array() {
+        let toml = r#"
+            [[forwarding]]
+            suffix = ["168.192.in-addr.arpa", "onsite"]
+            upstream = "192.168.88.1"
+        "#;
+        let config: Config = toml::from_str(toml).unwrap();
+        assert_eq!(config.forwarding.len(), 1);
+        assert_eq!(
+            config.forwarding[0].suffix,
+            &["168.192.in-addr.arpa", "onsite"]
+        );
+    }
+
+    #[test]
+    fn forwarding_suffix_array_expands_to_multiple_runtime_rules() {
+        let rule = ForwardingRuleConfig {
+            suffix: vec!["168.192.in-addr.arpa".to_string(), "onsite".to_string()],
+            upstream: "192.168.88.1".to_string(),
+        };
+        let runtime = rule.to_runtime_rules().unwrap();
+        assert_eq!(runtime.len(), 2);
+        assert_eq!(runtime[0].suffix, "168.192.in-addr.arpa");
+        assert_eq!(runtime[1].suffix, "onsite");
+        assert_eq!(runtime[0].upstream, runtime[1].upstream);
+    }
+
+    #[test]
+    fn forwarding_upstream_with_explicit_port() {
+        let rule = ForwardingRuleConfig {
+            suffix: vec!["home.local".to_string()],
+            upstream: "100.90.1.63:5361".to_string(),
+        };
+        let runtime = rule.to_runtime_rules().unwrap();
+        assert_eq!(runtime.len(), 1);
+        assert!(matches!(
+            runtime[0].upstream,
+            crate::forward::Upstream::Udp(_)
+        ));
+        assert_eq!(runtime[0].upstream.to_string(), "100.90.1.63:5361");
+        assert_eq!(runtime[0].suffix, "home.local");
+    }
+
+    #[test]
+    fn forwarding_upstream_defaults_to_port_53() {
+        let rule = ForwardingRuleConfig {
+            suffix: vec!["home.local".to_string()],
+            upstream: "100.90.1.63".to_string(),
+        };
+        let runtime = rule.to_runtime_rules().unwrap();
+        assert_eq!(runtime[0].upstream.to_string(), "100.90.1.63:53");
+    }
+
+    #[test]
+    fn forwarding_invalid_upstream_returns_error() {
+        let rule = ForwardingRuleConfig {
+            suffix: vec!["home.local".to_string()],
+            upstream: "not-a-valid-host".to_string(),
+        };
+        assert!(rule.to_runtime_rules().is_err());
+    }
+
+    #[test]
+    fn forwarding_upstream_accepts_dot_scheme() {
+        let rule = ForwardingRuleConfig {
+            suffix: vec!["google.com".to_string()],
+            upstream: "tls://9.9.9.9#dns.quad9.net".to_string(),
+        };
+        let runtime = rule
+            .to_runtime_rules()
+            .expect("tls:// upstream should parse");
+        assert_eq!(runtime.len(), 1);
+        assert_eq!(
+            runtime[0].upstream.to_string(),
+            "tls://9.9.9.9:853#dns.quad9.net"
+        );
+    }
+
+    #[test]
+    fn forwarding_upstream_accepts_doh_scheme() {
+        let rule = ForwardingRuleConfig {
+            suffix: vec!["goog".to_string()],
+            upstream: "https://dns.quad9.net/dns-query".to_string(),
+        };
+        let runtime = rule
+            .to_runtime_rules()
+            .expect("https:// upstream should parse");
+        assert_eq!(runtime.len(), 1);
+        assert_eq!(
+            runtime[0].upstream.to_string(),
+            "https://dns.quad9.net/dns-query"
+        );
+    }
+
+    #[test]
+    fn forwarding_config_rules_take_precedence_over_discovered() {
+        let config_rules = vec![ForwardingRuleConfig {
+            suffix: vec!["home.local".to_string()],
+            upstream: "10.0.0.1:53".to_string(),
+        }];
+        let discovered = vec![crate::system_dns::ForwardingRule::new(
+            "home.local".to_string(),
+            crate::forward::Upstream::Udp("192.168.1.1:53".parse().unwrap()),
+        )];
+        let merged = merge_forwarding_rules(&config_rules, discovered).unwrap();
+        let picked = crate::system_dns::match_forwarding_rule("host.home.local", &merged)
+            .expect("rule should match");
+        assert_eq!(picked.to_string(), "10.0.0.1:53");
+    }
+
+    #[test]
+    fn forwarding_merge_preserves_non_overlapping_discovered() {
+        let config_rules = vec![ForwardingRuleConfig {
+            suffix: vec!["home.local".to_string()],
+            upstream: "10.0.0.1:53".to_string(),
+        }];
+        let discovered = vec![crate::system_dns::ForwardingRule::new(
+            "corp.example".to_string(),
+            crate::forward::Upstream::Udp("192.168.1.1:53".parse().unwrap()),
+        )];
+        let merged = merge_forwarding_rules(&config_rules, discovered).unwrap();
+        assert_eq!(merged.len(), 2);
+        let picked = crate::system_dns::match_forwarding_rule("host.corp.example", &merged)
+            .expect("discovered rule should still match");
+        assert_eq!(picked.to_string(), "192.168.1.1:53");
+    }
+
+    #[test]
+    fn forwarding_merge_suffix_array_expands_to_multiple_rules() {
+        let config_rules = vec![ForwardingRuleConfig {
+            suffix: vec!["a.local".to_string(), "b.local".to_string()],
+            upstream: "10.0.0.1:53".to_string(),
+        }];
+        let merged = merge_forwarding_rules(&config_rules, vec![]).unwrap();
+        assert_eq!(merged.len(), 2);
+    }
 }
 
 pub struct ConfigLoad {
@@ -503,6 +838,13 @@ pub fn load_config(path: &str) -> Result<ConfigLoad> {
             let filename = p.file_name().unwrap_or(p.as_os_str());
             v.push(crate::config_dir().join(filename));
             v.push(crate::data_dir().join(filename));
+            // Interactive root and sudo'd users: always consult the XDG path
+            // so `touch ~/.config/numa/numa.toml` works regardless of whether
+            // config_dir() routed to FHS (issue #81).
+            let suggested = crate::suggested_config_path();
+            if !v.contains(&suggested) {
+                v.push(suggested);
+            }
         }
         v
     };
@@ -523,11 +865,7 @@ pub fn load_config(path: &str) -> Result<ConfigLoad> {
         }
     }
 
-    // Show config_dir candidate as the "expected" path — it's actionable
+    let display_path = crate::suggested_config_path().to_string_lossy().to_string();
-    let display_path = candidates
-        .get(1)
-        .map(|p| p.to_string_lossy().to_string())
-        .unwrap_or_else(|| resolve_path(path));
     log::info!("config not found, using defaults (create {})", display_path);
     Ok(ConfigLoad {
         config: Config::default(),

src/ctx.rs

@@ -1,7 +1,7 @@
-use std::collections::HashMap;
+use std::collections::{HashMap, HashSet};
 use std::net::SocketAddr;
 use std::path::PathBuf;
-use std::sync::{Mutex, RwLock};
+use std::sync::{Arc, Mutex, RwLock};
 use std::time::{Duration, Instant, SystemTime};
 
 use arc_swap::ArcSwap;
@@ -16,8 +16,9 @@ use crate::blocklist::BlocklistStore;
 use crate::buffer::BytePacketBuffer;
 use crate::cache::{DnsCache, DnssecStatus};
 use crate::config::{UpstreamMode, ZoneMap};
-use crate::forward::{forward_query, Upstream};
+use crate::forward::{forward_query_raw, forward_with_failover_raw, Upstream, UpstreamPool};
 use crate::header::ResultCode;
+use crate::health::HealthMeta;
 use crate::lan::PeerStore;
 use crate::override_store::OverrideStore;
 use crate::packet::DnsPacket;
@@ -26,7 +27,7 @@ use crate::question::QueryType;
 use crate::record::DnsRecord;
 use crate::service_store::ServiceStore;
 use crate::srtt::SrttCache;
-use crate::stats::{QueryPath, ServerStats};
+use crate::stats::{QueryPath, ServerStats, Transport};
 use crate::system_dns::ForwardingRule;
 
 pub struct ServerCtx {
@@ -34,6 +35,8 @@ pub struct ServerCtx {
     pub zone_map: ZoneMap,
     /// std::sync::RwLock (not tokio) — locks must never be held across .await points.
     pub cache: RwLock<DnsCache>,
+    /// Domains currently being refreshed in the background (dedup guard).
+    pub refreshing: Mutex<HashSet<(String, QueryType)>>,
     pub stats: Mutex<ServerStats>,
     pub overrides: RwLock<OverrideStore>,
     pub blocklist: RwLock<BlocklistStore>,
@@ -41,11 +44,12 @@ pub struct ServerCtx {
     pub services: Mutex<ServiceStore>,
     pub lan_peers: Mutex<PeerStore>,
     pub forwarding_rules: Vec<ForwardingRule>,
-    pub upstream: Mutex<Upstream>,
+    pub upstream_pool: Mutex<UpstreamPool>,
     pub upstream_auto: bool,
     pub upstream_port: u16,
     pub lan_ip: Mutex<std::net::Ipv4Addr>,
     pub timeout: Duration,
+    pub hedge_delay: Duration,
     pub proxy_tld: String,
     pub proxy_tld_suffix: String, // pre-computed ".{tld}" to avoid per-query allocation
     pub lan_enabled: bool,
@@ -60,6 +64,17 @@ pub struct ServerCtx {
     pub inflight: Mutex<InflightMap>,
     pub dnssec_enabled: bool,
     pub dnssec_strict: bool,
+    /// Cached health metadata (version, hostname, DoT config, CA
+    /// fingerprint, features). Shared between the main and mobile
+    /// API `/health` handlers. Built once at startup in `main.rs`.
+    pub health_meta: HealthMeta,
+    /// CA certificate in PEM form, cached at startup. `None` if no
+    /// TLS-using feature is enabled and the CA hasn't been generated.
+    /// Used by `/ca.pem`, `/mobileconfig`, and `/ca.mobileconfig`
+    /// handlers to avoid per-request disk I/O on the hot path.
+    pub ca_pem: Option<String>,
+    pub mobile_enabled: bool,
+    pub mobile_port: u16,
 }
 
 /// Transport-agnostic DNS resolution. Runs the full pipeline (overrides, blocklist,
@@ -69,9 +84,11 @@ pub struct ServerCtx {
 /// (and logging parse errors) before calling this function.
 pub async fn resolve_query(
     query: DnsPacket,
+    raw_wire: &[u8],
     src_addr: SocketAddr,
-    ctx: &ServerCtx,
+    ctx: &Arc<ServerCtx>,
-) -> crate::Result<BytePacketBuffer> {
+    transport: Transport,
+) -> crate::Result<(BytePacketBuffer, QueryPath)> {
     let start = Instant::now();
 
     let (qname, qtype) = match query.questions.first() {
@@ -79,7 +96,8 @@ pub async fn resolve_query(
         None => return Err("empty question section".into()),
     };
 
-    // Pipeline: overrides -> .tld interception -> blocklist -> local zones -> cache -> upstream
+    // Pipeline: overrides -> .localhost -> local zones -> special-use (unless forwarded)
+    //        -> .tld proxy -> blocklist -> cache -> forwarding -> recursive/upstream
     // Each lock is scoped to avoid holding MutexGuard across await points.
     let (response, path, dnssec) = {
         let override_record = ctx.overrides.read().unwrap().lookup(&qname);
@@ -98,8 +116,14 @@ pub async fn resolve_query(
                 300,
             ));
             (resp, QueryPath::Local, DnssecStatus::Indeterminate)
-        } else if is_special_use_domain(&qname) {
+        } else if let Some(records) = ctx.zone_map.get(qname.as_str()).and_then(|m| m.get(&qtype)) {
-            // RFC 6761/8880: private PTR, DDR, NAT64 — answer locally
+            let mut resp = DnsPacket::response_from(&query, ResultCode::NOERROR);
+            resp.answers = records.clone();
+            (resp, QueryPath::Local, DnssecStatus::Indeterminate)
+        } else if is_special_use_domain(&qname)
+            && crate::system_dns::match_forwarding_rule(&qname, &ctx.forwarding_rules).is_none()
+        {
+            // RFC 6761/8880: answer locally unless a forwarding rule covers this zone.
             let resp = special_use_response(&query, &qname, qtype);
             (resp, QueryPath::Local, DnssecStatus::Indeterminate)
         } else if !ctx.proxy_tld_suffix.is_empty()
@@ -146,30 +170,33 @@ pub async fn resolve_query(
                 60,
             ));
             (resp, QueryPath::Blocked, DnssecStatus::Indeterminate)
-        } else if let Some(records) = ctx.zone_map.get(qname.as_str()).and_then(|m| m.get(&qtype)) {
-            let mut resp = DnsPacket::response_from(&query, ResultCode::NOERROR);
-            resp.answers = records.clone();
-            (resp, QueryPath::Local, DnssecStatus::Indeterminate)
         } else {
             let cached = ctx.cache.read().unwrap().lookup_with_status(&qname, qtype);
-            if let Some((cached, cached_dnssec)) = cached {
+            if let Some((cached, cached_dnssec, freshness)) = cached {
+                if freshness.needs_refresh() {
+                    let key = (qname.clone(), qtype);
+                    let already = !ctx.refreshing.lock().unwrap().insert(key.clone());
+                    if !already {
+                        let ctx = Arc::clone(ctx);
+                        tokio::spawn(async move {
+                            refresh_entry(&ctx, &key.0, key.1).await;
+                            ctx.refreshing.lock().unwrap().remove(&key);
+                        });
+                    }
+                }
                 let mut resp = cached;
                 resp.header.id = query.header.id;
                 if cached_dnssec == DnssecStatus::Secure {
                     resp.header.authed_data = true;
                 }
                 (resp, QueryPath::Cached, cached_dnssec)
-            } else if let Some(fwd_addr) =
+            } else if let Some(upstream) =
                 crate::system_dns::match_forwarding_rule(&qname, &ctx.forwarding_rules)
             {
                 // Conditional forwarding takes priority over recursive mode
                 // (e.g. Tailscale .ts.net, VPC private zones)
-                let upstream = Upstream::Udp(fwd_addr);
+                match forward_and_cache(raw_wire, upstream, ctx, &qname, qtype).await {
-                match forward_query(&query, &upstream, ctx.timeout).await {
+                    Ok(resp) => (resp, QueryPath::Forwarded, DnssecStatus::Indeterminate),
-                    Ok(resp) => {
-                        ctx.cache.write().unwrap().insert(&qname, qtype, &resp);
-                        (resp, QueryPath::Forwarded, DnssecStatus::Indeterminate)
-                    }
                     Err(e) => {
                         error!(
                             "{} | {:?} {} | FORWARD ERROR | {}",
@@ -208,16 +235,27 @@ pub async fn resolve_query(
                 }
                 (resp, path, DnssecStatus::Indeterminate)
             } else {
-                let upstream =
+                let pool = ctx.upstream_pool.lock().unwrap().clone();
-                    match crate::system_dns::match_forwarding_rule(&qname, &ctx.forwarding_rules) {
+                match forward_with_failover_raw(
-                        Some(addr) => Upstream::Udp(addr),
+                    raw_wire,
-                        None => ctx.upstream.lock().unwrap().clone(),
+                    &pool,
-                    };
+                    &ctx.srtt,
-                match forward_query(&query, &upstream, ctx.timeout).await {
+                    ctx.timeout,
-                    Ok(resp) => {
+                    ctx.hedge_delay,
-                        ctx.cache.write().unwrap().insert(&qname, qtype, &resp);
+                )
-                        (resp, QueryPath::Forwarded, DnssecStatus::Indeterminate)
+                .await
-                    }
+                {
+                    Ok(resp_wire) => match cache_and_parse(ctx, &qname, qtype, &resp_wire) {
+                        Ok(resp) => (resp, QueryPath::Forwarded, DnssecStatus::Indeterminate),
+                        Err(e) => {
+                            error!("{} | {:?} {} | PARSE ERROR | {}", src_addr, qtype, qname, e);
+                            (
+                                DnsPacket::response_from(&query, ResultCode::SERVFAIL),
+                                QueryPath::UpstreamError,
+                                DnssecStatus::Indeterminate,
+                            )
+                        }
+                    },
                     Err(e) => {
                         error!(
                             "{} | {:?} {} | UPSTREAM ERROR | {}",
@@ -319,7 +357,7 @@ pub async fn resolve_query(
     // Record stats and query log
     {
         let mut s = ctx.stats.lock().unwrap();
-        let total = s.record(path);
+        let total = s.record(path, transport);
         if total.is_multiple_of(1000) {
             s.log_summary();
         }
@@ -331,19 +369,87 @@ pub async fn resolve_query(
         domain: qname,
         query_type: qtype,
         path,
+        transport,
         rescode: response.header.rescode,
         latency_us: elapsed.as_micros() as u64,
         dnssec,
     });
 
-    Ok(resp_buffer)
+    Ok((resp_buffer, path))
+}
+
+fn cache_and_parse(
+    ctx: &ServerCtx,
+    qname: &str,
+    qtype: QueryType,
+    resp_wire: &[u8],
+) -> crate::Result<DnsPacket> {
+    ctx.cache
+        .write()
+        .unwrap()
+        .insert_wire(qname, qtype, resp_wire, DnssecStatus::Indeterminate);
+    let mut buf = BytePacketBuffer::from_bytes(resp_wire);
+    DnsPacket::from_buffer(&mut buf)
+}
+
+/// Re-resolve a single (domain, qtype) and update the cache.
+/// Used for both stale-entry refresh and proactive cache warming.
+pub async fn refresh_entry(ctx: &ServerCtx, qname: &str, qtype: QueryType) {
+    let query = DnsPacket::query(0, qname, qtype);
+    if ctx.upstream_mode == UpstreamMode::Recursive {
+        if let Ok(resp) = crate::recursive::resolve_recursive(
+            qname,
+            qtype,
+            &ctx.cache,
+            &query,
+            &ctx.root_hints,
+            &ctx.srtt,
+        )
+        .await
+        {
+            ctx.cache.write().unwrap().insert(qname, qtype, &resp);
+        }
+    } else {
+        let mut buf = BytePacketBuffer::new();
+        if query.write(&mut buf).is_ok() {
+            let pool = ctx.upstream_pool.lock().unwrap().clone();
+            if let Ok(wire) = forward_with_failover_raw(
+                buf.filled(),
+                &pool,
+                &ctx.srtt,
+                ctx.timeout,
+                ctx.hedge_delay,
+            )
+            .await
+            {
+                ctx.cache.write().unwrap().insert_wire(
+                    qname,
+                    qtype,
+                    &wire,
+                    DnssecStatus::Indeterminate,
+                );
+            }
+        }
+    }
+}
+
+async fn forward_and_cache(
+    wire: &[u8],
+    upstream: &Upstream,
+    ctx: &ServerCtx,
+    qname: &str,
+    qtype: QueryType,
+) -> crate::Result<DnsPacket> {
+    let resp_wire = forward_query_raw(wire, upstream, ctx.timeout).await?;
+    cache_and_parse(ctx, qname, qtype, &resp_wire)
 }
 
-/// Handle a DNS query received over UDP. Thin wrapper around resolve_query.
 pub async fn handle_query(
     mut buffer: BytePacketBuffer,
+    raw_len: usize,
     src_addr: SocketAddr,
-    ctx: &ServerCtx,
+    ctx: &Arc<ServerCtx>,
+    transport: Transport,
 ) -> crate::Result<()> {
     let query = match DnsPacket::from_buffer(&mut buffer) {
         Ok(packet) => packet,
@@ -352,8 +458,8 @@ pub async fn handle_query(
             return Ok(());
         }
     };
-    match resolve_query(query, src_addr, ctx).await {
+    match resolve_query(query, &buffer.buf[..raw_len], src_addr, ctx, transport).await {
-        Ok(resp_buffer) => {
+        Ok((resp_buffer, _)) => {
             ctx.socket.send_to(resp_buffer.filled(), src_addr).await?;
         }
         Err(e) => {
@@ -932,4 +1038,219 @@ mod tests {
             "error message must be preserved for logging"
         );
     }
+
+    // ---- Full-pipeline resolve_query tests ----
+
+    /// Send a query through the full resolve_query pipeline and return
+    /// the parsed response + query path.
+    async fn resolve_in_test(
+        ctx: &Arc<ServerCtx>,
+        domain: &str,
+        qtype: QueryType,
+    ) -> (DnsPacket, QueryPath) {
+        let query = DnsPacket::query(0xBEEF, domain, qtype);
+        let mut buf = BytePacketBuffer::new();
+        query.write(&mut buf).unwrap();
+        let raw = &buf.buf[..buf.pos];
+        let src: SocketAddr = "127.0.0.1:1234".parse().unwrap();
+
+        let (resp_buf, path) = resolve_query(query, raw, src, ctx, Transport::Udp)
+            .await
+            .unwrap();
+
+        let mut resp_parse_buf = BytePacketBuffer::from_bytes(resp_buf.filled());
+        let resp = DnsPacket::from_buffer(&mut resp_parse_buf).unwrap();
+        (resp, path)
+    }
+
+    #[tokio::test]
+    async fn special_use_private_ptr_returns_nxdomain() {
+        let ctx = Arc::new(crate::testutil::test_ctx().await);
+        let (resp, path) =
+            resolve_in_test(&ctx, "153.188.168.192.in-addr.arpa", QueryType::PTR).await;
+        assert_eq!(path, QueryPath::Local);
+        assert_eq!(resp.header.rescode, ResultCode::NXDOMAIN);
+    }
+
+    #[tokio::test]
+    async fn forwarding_rule_overrides_special_use_domain() {
+        let mut resp = DnsPacket::new();
+        resp.header.response = true;
+        resp.header.rescode = ResultCode::NOERROR;
+        let upstream_addr = crate::testutil::mock_upstream(resp).await;
+
+        let mut ctx = crate::testutil::test_ctx().await;
+        ctx.forwarding_rules = vec![ForwardingRule::new(
+            "168.192.in-addr.arpa".to_string(),
+            Upstream::Udp(upstream_addr),
+        )];
+        let ctx = Arc::new(ctx);
+
+        let (resp, path) =
+            resolve_in_test(&ctx, "153.188.168.192.in-addr.arpa", QueryType::PTR).await;
+
+        assert_eq!(
+            path,
+            QueryPath::Forwarded,
+            "forwarding rule must take precedence over special-use NXDOMAIN"
+        );
+        assert_eq!(resp.header.rescode, ResultCode::NOERROR);
+    }
+
+    #[tokio::test]
+    async fn pipeline_override_takes_precedence() {
+        let ctx = crate::testutil::test_ctx().await;
+        ctx.overrides
+            .write()
+            .unwrap()
+            .insert("override.test", "1.2.3.4", 60, None)
+            .unwrap();
+        let ctx = Arc::new(ctx);
+
+        let (resp, path) = resolve_in_test(&ctx, "override.test", QueryType::A).await;
+        assert_eq!(path, QueryPath::Overridden);
+        assert_eq!(resp.header.rescode, ResultCode::NOERROR);
+        assert_eq!(resp.answers.len(), 1);
+    }
+
+    #[tokio::test]
+    async fn pipeline_localhost_resolves_to_loopback() {
+        let ctx = Arc::new(crate::testutil::test_ctx().await);
+
+        let (resp, path) = resolve_in_test(&ctx, "localhost", QueryType::A).await;
+        assert_eq!(path, QueryPath::Local);
+        assert_eq!(resp.header.rescode, ResultCode::NOERROR);
+        match &resp.answers[0] {
+            DnsRecord::A { addr, .. } => assert_eq!(*addr, Ipv4Addr::LOCALHOST),
+            other => panic!("expected A record, got {:?}", other),
+        }
+    }
+
+    #[tokio::test]
+    async fn pipeline_localhost_subdomain_resolves_to_loopback() {
+        let ctx = Arc::new(crate::testutil::test_ctx().await);
+
+        let (resp, path) = resolve_in_test(&ctx, "app.localhost", QueryType::A).await;
+        assert_eq!(path, QueryPath::Local);
+        assert_eq!(resp.header.rescode, ResultCode::NOERROR);
+        match &resp.answers[0] {
+            DnsRecord::A { addr, .. } => assert_eq!(*addr, Ipv4Addr::LOCALHOST),
+            other => panic!("expected A record, got {:?}", other),
+        }
+    }
+
+    #[tokio::test]
+    async fn pipeline_local_zone_returns_configured_record() {
+        let mut ctx = crate::testutil::test_ctx().await;
+        let mut inner = HashMap::new();
+        inner.insert(
+            QueryType::A,
+            vec![DnsRecord::A {
+                domain: "myapp.test".to_string(),
+                addr: Ipv4Addr::new(10, 0, 0, 42),
+                ttl: 300,
+            }],
+        );
+        ctx.zone_map.insert("myapp.test".to_string(), inner);
+        let ctx = Arc::new(ctx);
+
+        let (resp, path) = resolve_in_test(&ctx, "myapp.test", QueryType::A).await;
+        assert_eq!(path, QueryPath::Local);
+        assert_eq!(resp.header.rescode, ResultCode::NOERROR);
+        match &resp.answers[0] {
+            DnsRecord::A { addr, .. } => assert_eq!(*addr, Ipv4Addr::new(10, 0, 0, 42)),
+            other => panic!("expected A record, got {:?}", other),
+        }
+    }
+
+    #[tokio::test]
+    async fn pipeline_tld_proxy_resolves_service() {
+        let ctx = crate::testutil::test_ctx().await;
+        ctx.services.lock().unwrap().insert("grafana", 3000);
+        let ctx = Arc::new(ctx);
+
+        let (resp, path) = resolve_in_test(&ctx, "grafana.numa", QueryType::A).await;
+        assert_eq!(path, QueryPath::Local);
+        assert_eq!(resp.header.rescode, ResultCode::NOERROR);
+        match &resp.answers[0] {
+            DnsRecord::A { addr, .. } => assert_eq!(*addr, Ipv4Addr::LOCALHOST),
+            other => panic!("expected A record, got {:?}", other),
+        }
+    }
+
+    #[tokio::test]
+    async fn pipeline_blocklist_sinkhole() {
+        let ctx = crate::testutil::test_ctx().await;
+        let mut domains = std::collections::HashSet::new();
+        domains.insert("ads.tracker.test".to_string());
+        ctx.blocklist.write().unwrap().swap_domains(domains, vec![]);
+        let ctx = Arc::new(ctx);
+
+        let (resp, path) = resolve_in_test(&ctx, "ads.tracker.test", QueryType::A).await;
+        assert_eq!(path, QueryPath::Blocked);
+        assert_eq!(resp.header.rescode, ResultCode::NOERROR);
+        match &resp.answers[0] {
+            DnsRecord::A { addr, .. } => assert_eq!(*addr, Ipv4Addr::UNSPECIFIED),
+            other => panic!("expected sinkhole A record, got {:?}", other),
+        }
+    }
+
+    #[tokio::test]
+    async fn pipeline_cache_hit() {
+        let ctx = Arc::new(crate::testutil::test_ctx().await);
+
+        // Pre-populate cache with a response
+        let mut pkt = DnsPacket::new();
+        pkt.header.response = true;
+        pkt.header.rescode = ResultCode::NOERROR;
+        pkt.questions.push(crate::question::DnsQuestion {
+            name: "cached.test".to_string(),
+            qtype: QueryType::A,
+        });
+        pkt.answers.push(DnsRecord::A {
+            domain: "cached.test".to_string(),
+            addr: Ipv4Addr::new(5, 5, 5, 5),
+            ttl: 3600,
+        });
+        ctx.cache
+            .write()
+            .unwrap()
+            .insert("cached.test", QueryType::A, &pkt);
+
+        let (resp, path) = resolve_in_test(&ctx, "cached.test", QueryType::A).await;
+        assert_eq!(path, QueryPath::Cached);
+        assert_eq!(resp.header.rescode, ResultCode::NOERROR);
+    }
+
+    #[tokio::test]
+    async fn pipeline_forwarding_returns_upstream_answer() {
+        let mut upstream_resp = DnsPacket::new();
+        upstream_resp.header.response = true;
+        upstream_resp.header.rescode = ResultCode::NOERROR;
+        upstream_resp.answers.push(DnsRecord::A {
+            domain: "internal.corp".to_string(),
+            addr: Ipv4Addr::new(10, 1, 2, 3),
+            ttl: 600,
+        });
+        let upstream_addr = crate::testutil::mock_upstream(upstream_resp).await;
+
+        let mut ctx = crate::testutil::test_ctx().await;
+        ctx.forwarding_rules = vec![ForwardingRule::new(
+            "corp".to_string(),
+            Upstream::Udp(upstream_addr),
+        )];
+        let ctx = Arc::new(ctx);
+
+        let (resp, path) = resolve_in_test(&ctx, "internal.corp", QueryType::A).await;
+        assert_eq!(path, QueryPath::Forwarded);
+        assert_eq!(resp.header.rescode, ResultCode::NOERROR);
+        assert_eq!(resp.answers.len(), 1);
+        match &resp.answers[0] {
+            DnsRecord::A { domain, addr, .. } => {
+                assert_eq!(domain, "internal.corp");
+                assert_eq!(*addr, Ipv4Addr::new(10, 1, 2, 3));
+            }
+            other => panic!("expected A record, got {:?}", other),
+        }
+    }
 }

src/dnssec.rs

@@ -5,6 +5,7 @@ use log::{debug, trace};
 use ring::digest;
 use ring::signature;
 
+use crate::buffer::BytePacketBuffer;
 use crate::cache::{DnsCache, DnssecStatus};
 use crate::packet::DnsPacket;
 use crate::question::QueryType;
@@ -720,22 +721,29 @@ pub fn verify_ds(ds: &DnsRecord, dnskey: &DnsRecord, owner: &str) -> bool {
 
 // -- Canonical wire format --
 
+/// Encode a DNS name in canonical wire form per RFC 4034 §6.2:
+/// uncompressed, with ASCII letters lowercased.
+///
+/// Lowercasing happens *after* escape resolution because `\065` yields
+/// `'A'`, which canonical form must convert to `'a'`.
 pub fn name_to_wire(name: &str) -> Vec<u8> {
-    let mut wire = Vec::with_capacity(name.len() + 2);
+    let mut buf = BytePacketBuffer::new();
-    if name == "." || name.is_empty() {
+    buf.write_qname(name)
-        wire.push(0);
+        .expect("name_to_wire: input must parse as a valid DNS name");
-        return wire;
+    let mut wire = buf.filled().to_vec();
-    }
+
-    for label in name.split('.') {
+    let mut i = 0;
-        if label.is_empty() {
+    while i < wire.len() {
-            continue;
+        let label_len = wire[i] as usize;
-        }
+        if label_len == 0 {
-        wire.push(label.len() as u8);
+            break;
-        for &b in label.as_bytes() {
-            wire.push(b.to_ascii_lowercase());
         }
+        i += 1;
+        let end = i + label_len;
+        wire[i..end].make_ascii_lowercase();
+        i = end;
     }
-    wire.push(0);
+
     wire
 }
 
@@ -1475,6 +1483,23 @@ mod tests {
         );
     }
 
+    #[test]
+    fn name_to_wire_escaped_dot_in_label_is_not_a_separator() {
+        // `exa\.mple.com` is two labels: `exa.mple` (8 bytes including the 0x2E) and `com`.
+        let wire = name_to_wire("exa\\.mple.com");
+        assert_eq!(
+            wire,
+            vec![8, b'e', b'x', b'a', b'.', b'm', b'p', b'l', b'e', 3, b'c', b'o', b'm', 0]
+        );
+    }
+
+    #[test]
+    fn name_to_wire_decimal_escape_is_lowercased() {
+        // \065 = 'A', must become 'a' in canonical form.
+        let wire = name_to_wire("\\065bc.com");
+        assert_eq!(wire, vec![3, b'a', b'b', b'c', 3, b'c', b'o', b'm', 0]);
+    }
+
     #[test]
     fn parent_zone_cases() {
         assert_eq!(parent_zone("example.com"), "com");

src/doh.rs

@@ -0,0 +1,224 @@
+use std::net::SocketAddr;
+
+use axum::body::Bytes;
+use axum::extract::{Request, State};
+use axum::response::{IntoResponse, Response};
+use hyper::StatusCode;
+use log::warn;
+
+use crate::buffer::BytePacketBuffer;
+use crate::ctx::{resolve_query, ServerCtx};
+use crate::header::ResultCode;
+use crate::packet::DnsPacket;
+use crate::stats::Transport;
+
+const MAX_DNS_MSG: usize = 4096;
+const DOH_CONTENT_TYPE: &str = "application/dns-message";
+
+pub async fn doh_post(State(state): State<super::proxy::DohState>, req: Request) -> Response {
+    let host = super::proxy::extract_host(&req);
+    if !is_doh_host(host.as_deref(), &state.ctx.proxy_tld) {
+        return StatusCode::NOT_FOUND.into_response();
+    }
+
+    let content_type = req
+        .headers()
+        .get(hyper::header::CONTENT_TYPE)
+        .and_then(|v| v.to_str().ok())
+        .unwrap_or("");
+    if !content_type.starts_with(DOH_CONTENT_TYPE) {
+        return StatusCode::UNSUPPORTED_MEDIA_TYPE.into_response();
+    }
+
+    let body = match axum::body::to_bytes(req.into_body(), MAX_DNS_MSG).await {
+        Ok(b) => b,
+        Err(_) => {
+            return (StatusCode::PAYLOAD_TOO_LARGE, "body exceeds 4096 bytes").into_response()
+        }
+    };
+
+    if body.is_empty() {
+        return (StatusCode::BAD_REQUEST, "empty body").into_response();
+    }
+
+    let src = state
+        .remote_addr
+        .unwrap_or_else(|| SocketAddr::from(([127, 0, 0, 1], 0)));
+
+    resolve_doh(&body, src, &state.ctx).await
+}
+
+fn is_doh_host(host: Option<&str>, tld: &str) -> bool {
+    let h = match host {
+        Some(h) => h,
+        None => return false,
+    };
+    let base = strip_port(h).unwrap_or(h);
+    is_loopback_host(base) || is_tld_match(base, tld)
+}
+
+fn strip_port(h: &str) -> Option<&str> {
+    if h.starts_with('[') {
+        // [::1]:443 → [::1]
+        let (base, port) = h.rsplit_once("]:")?;
+        port.bytes()
+            .all(|b| b.is_ascii_digit())
+            .then(|| &h[..base.len() + 1])
+    } else {
+        let (base, port) = h.rsplit_once(':')?;
+        // Bare IPv6 like "::1" has multiple colons — not a port suffix
+        if base.contains(':') {
+            return None;
+        }
+        port.bytes().all(|b| b.is_ascii_digit()).then_some(base)
+    }
+}
+
+fn is_loopback_host(h: &str) -> bool {
+    matches!(h, "127.0.0.1" | "::1" | "[::1]" | "localhost")
+}
+
+fn is_tld_match(h: &str, tld: &str) -> bool {
+    h == tld
+        || (h.len() == 2 * tld.len() + 1
+            && h.starts_with(tld)
+            && h.as_bytes().get(tld.len()) == Some(&b'.')
+            && h.ends_with(tld))
+}
+
+async fn resolve_doh(
+    dns_bytes: &[u8],
+    src: SocketAddr,
+    ctx: &std::sync::Arc<ServerCtx>,
+) -> Response {
+    let mut buffer = BytePacketBuffer::from_bytes(dns_bytes);
+    let query = match DnsPacket::from_buffer(&mut buffer) {
+        Ok(q) => q,
+        Err(e) => {
+            warn!("DoH: parse error from {}: {}", src, e);
+            let query_id = u16::from_be_bytes([
+                dns_bytes.first().copied().unwrap_or(0),
+                dns_bytes.get(1).copied().unwrap_or(0),
+            ]);
+            let mut resp = DnsPacket::new();
+            resp.header.id = query_id;
+            resp.header.response = true;
+            resp.header.rescode = ResultCode::FORMERR;
+            return serialize_response(&resp);
+        }
+    };
+
+    let query_id = query.header.id;
+    let query_rd = query.header.recursion_desired;
+    let questions = query.questions.clone();
+
+    match resolve_query(query, dns_bytes, src, ctx, Transport::Doh).await {
+        Ok((resp_buffer, _)) => {
+            let min_ttl = extract_min_ttl(resp_buffer.filled());
+            dns_response(resp_buffer.filled(), min_ttl)
+        }
+        Err(e) => {
+            warn!("DoH: resolve error for {}: {}", src, e);
+            let mut resp = DnsPacket::new();
+            resp.header.id = query_id;
+            resp.header.response = true;
+            resp.header.recursion_desired = query_rd;
+            resp.header.recursion_available = true;
+            resp.header.rescode = ResultCode::SERVFAIL;
+            resp.questions = questions;
+            serialize_response(&resp)
+        }
+    }
+}
+
+fn extract_min_ttl(wire: &[u8]) -> u32 {
+    crate::wire::scan_ttl_offsets(wire)
+        .ok()
+        .and_then(|meta| crate::wire::min_ttl_from_wire(wire, &meta))
+        .unwrap_or(0)
+}
+
+fn dns_response(wire: &[u8], min_ttl: u32) -> Response {
+    (
+        StatusCode::OK,
+        [
+            (hyper::header::CONTENT_TYPE, DOH_CONTENT_TYPE),
+            (
+                hyper::header::CACHE_CONTROL,
+                &format!("max-age={}", min_ttl),
+            ),
+        ],
+        Bytes::copy_from_slice(wire),
+    )
+        .into_response()
+}
+
+fn serialize_response(pkt: &DnsPacket) -> Response {
+    let mut buf = BytePacketBuffer::new();
+    match pkt.write(&mut buf) {
+        Ok(_) => dns_response(buf.filled(), 0),
+        Err(_) => StatusCode::INTERNAL_SERVER_ERROR.into_response(),
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::buffer::BytePacketBuffer;
+    use crate::header::ResultCode;
+    use crate::packet::DnsPacket;
+    use crate::record::DnsRecord;
+
+    #[test]
+    fn is_doh_host_matches_tld() {
+        assert!(is_doh_host(Some("numa"), "numa"));
+        assert!(is_doh_host(Some("numa.numa"), "numa"));
+        assert!(is_doh_host(Some("127.0.0.1"), "numa"));
+        assert!(is_doh_host(Some("127.0.0.1:443"), "numa"));
+        assert!(is_doh_host(Some("::1"), "numa"));
+        assert!(is_doh_host(Some("[::1]"), "numa"));
+        assert!(is_doh_host(Some("[::1]:443"), "numa"));
+        assert!(is_doh_host(Some("localhost"), "numa"));
+        assert!(is_doh_host(Some("localhost:443"), "numa"));
+        assert!(!is_doh_host(Some("foo.numa"), "numa"));
+        assert!(!is_doh_host(None, "numa"));
+    }
+
+    #[test]
+    fn extract_min_ttl_from_response() {
+        let mut pkt = DnsPacket::new();
+        pkt.header.response = true;
+        pkt.answers.push(DnsRecord::A {
+            domain: "example.com".to_string(),
+            addr: std::net::Ipv4Addr::new(1, 2, 3, 4),
+            ttl: 300,
+        });
+        pkt.answers.push(DnsRecord::A {
+            domain: "example.com".to_string(),
+            addr: std::net::Ipv4Addr::new(5, 6, 7, 8),
+            ttl: 60,
+        });
+        let mut buf = BytePacketBuffer::new();
+        pkt.write(&mut buf).unwrap();
+        assert_eq!(extract_min_ttl(buf.filled()), 60);
+    }
+
+    #[test]
+    fn extract_min_ttl_no_answers() {
+        let mut pkt = DnsPacket::new();
+        pkt.header.response = true;
+        let mut buf = BytePacketBuffer::new();
+        pkt.write(&mut buf).unwrap();
+        assert_eq!(extract_min_ttl(buf.filled()), 0);
+    }
+
+    #[test]
+    fn serialize_formerr_response() {
+        let mut pkt = DnsPacket::new();
+        pkt.header.id = 0xABCD;
+        pkt.header.response = true;
+        pkt.header.rescode = ResultCode::FORMERR;
+        let resp = serialize_response(&pkt);
+        assert_eq!(resp.status(), StatusCode::OK);
+    }
+}

src/dot.rs

@@ -15,6 +15,7 @@ use crate::config::DotConfig;
 use crate::ctx::{resolve_query, ServerCtx};
 use crate::header::ResultCode;
 use crate::packet::DnsPacket;
+use crate::stats::Transport;
 
 const MAX_CONNECTIONS: usize = 512;
 const IDLE_TIMEOUT: Duration = Duration::from_secs(30);
@@ -153,8 +154,11 @@ async fn accept_loop(listener: TcpListener, acceptor: TlsAcceptor, ctx: Arc<Serv
 
 /// Handle a single persistent DoT connection (RFC 7858).
 /// Reads length-prefixed DNS queries until EOF, idle timeout, or error.
-async fn handle_dot_connection<S>(mut stream: S, remote_addr: SocketAddr, ctx: &ServerCtx)
+async fn handle_dot_connection<S>(
-where
+    mut stream: S,
+    remote_addr: SocketAddr,
+    ctx: &std::sync::Arc<ServerCtx>,
+) where
     S: AsyncReadExt + AsyncWriteExt + Unpin,
 {
     loop {
@@ -177,8 +181,6 @@ where
             break;
         };
 
-        // Parse query up-front so we can echo its question section in SERVFAIL
-        // responses when resolve_query fails.
         let query = match DnsPacket::from_buffer(&mut buffer) {
             Ok(q) => q,
             Err(e) => {
@@ -200,8 +202,16 @@ where
             }
         };
 
-        match resolve_query(query.clone(), remote_addr, ctx).await {
+        match resolve_query(
-            Ok(resp_buffer) => {
+            query.clone(),
+            &buffer.buf[..msg_len],
+            remote_addr,
+            ctx,
+            Transport::Dot,
+        )
+        .await
+        {
+            Ok((resp_buffer, _)) => {
                 if write_framed(&mut stream, resp_buffer.filled())
                     .await
                     .is_err()
@@ -269,7 +279,7 @@ where
 mod tests {
     use super::*;
     use std::collections::HashMap;
-    use std::sync::{Mutex, RwLock};
+    use std::sync::Mutex;
 
     use rcgen::{CertificateParams, DnType, KeyPair};
     use rustls::pki_types::{CertificateDer, PrivateKeyDer, PrivatePkcs8KeyDer, ServerName};
@@ -334,54 +344,29 @@ mod tests {
     async fn spawn_dot_server() -> (SocketAddr, CertificateDer<'static>) {
         let (server_tls, cert_der) = test_tls_configs();
 
-        let socket = tokio::net::UdpSocket::bind("127.0.0.1:0").await.unwrap();
+        let upstream_addr = crate::testutil::blackhole_upstream();
-        // Bind an unresponsive upstream and leak it so it lives for the test duration.
+
-        let blackhole = Box::leak(Box::new(std::net::UdpSocket::bind("127.0.0.1:0").unwrap()));
+        let mut ctx = crate::testutil::test_ctx().await;
-        let upstream_addr = blackhole.local_addr().unwrap();
+        ctx.zone_map = {
-        let ctx = Arc::new(ServerCtx {
+            let mut m = HashMap::new();
-            socket,
+            let mut inner = HashMap::new();
-            zone_map: {
+            inner.insert(
-                let mut m = HashMap::new();
+                QueryType::A,
-                let mut inner = HashMap::new();
+                vec![DnsRecord::A {
-                inner.insert(
+                    domain: "dot-test.example".to_string(),
-                    QueryType::A,
+                    addr: std::net::Ipv4Addr::new(10, 0, 0, 1),
-                    vec![DnsRecord::A {
+                    ttl: 300,
-                        domain: "dot-test.example".to_string(),
+                }],
-                        addr: std::net::Ipv4Addr::new(10, 0, 0, 1),
+            );
-                        ttl: 300,
+            m.insert("dot-test.example".to_string(), inner);
-                    }],
+            m
-                );
+        };
-                m.insert("dot-test.example".to_string(), inner);
+        ctx.upstream_pool = Mutex::new(crate::forward::UpstreamPool::new(
-                m
+            vec![crate::forward::Upstream::Udp(upstream_addr)],
-            },
+            vec![],
-            cache: RwLock::new(crate::cache::DnsCache::new(100, 60, 86400)),
+        ));
-            stats: Mutex::new(crate::stats::ServerStats::new()),
+        ctx.tls_config = Some(arc_swap::ArcSwap::from(server_tls));
-            overrides: RwLock::new(crate::override_store::OverrideStore::new()),
+        let ctx = Arc::new(ctx);
-            blocklist: RwLock::new(crate::blocklist::BlocklistStore::new()),
-            query_log: Mutex::new(crate::query_log::QueryLog::new(100)),
-            services: Mutex::new(crate::service_store::ServiceStore::new()),
-            lan_peers: Mutex::new(crate::lan::PeerStore::new(90)),
-            forwarding_rules: Vec::new(),
-            upstream: Mutex::new(crate::forward::Upstream::Udp(upstream_addr)),
-            upstream_auto: false,
-            upstream_port: 53,
-            lan_ip: Mutex::new(std::net::Ipv4Addr::LOCALHOST),
-            timeout: Duration::from_millis(200),
-            proxy_tld: "numa".to_string(),
-            proxy_tld_suffix: ".numa".to_string(),
-            lan_enabled: false,
-            config_path: String::new(),
-            config_found: false,
-            config_dir: std::path::PathBuf::from("/tmp"),
-            data_dir: std::path::PathBuf::from("/tmp"),
-            tls_config: Some(arc_swap::ArcSwap::from(server_tls)),
-            upstream_mode: crate::config::UpstreamMode::Forward,
-            root_hints: Vec::new(),
-            srtt: RwLock::new(crate::srtt::SrttCache::new(true)),
-            inflight: Mutex::new(HashMap::new()),
-            dnssec_enabled: false,
-            dnssec_strict: false,
-        });
 
         let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
         let addr = listener.local_addr().unwrap();

src/forward.rs

@@ -1,12 +1,14 @@
 use std::fmt;
-use std::net::SocketAddr;
+use std::net::{IpAddr, SocketAddr};
-use std::time::Duration;
+use std::sync::RwLock;
+use std::time::{Duration, Instant};
 
 use tokio::net::UdpSocket;
 use tokio::time::timeout;
 
 use crate::buffer::BytePacketBuffer;
 use crate::packet::DnsPacket;
+use crate::srtt::SrttCache;
 use crate::Result;
 
 #[derive(Clone)]
@@ -16,6 +18,11 @@ pub enum Upstream {
         url: String,
         client: reqwest::Client,
     },
+    Dot {
+        addr: SocketAddr,
+        tls_name: Option<String>,
+        connector: tokio_rustls::TlsConnector,
+    },
 }
 
 impl PartialEq for Upstream {
@@ -23,16 +30,138 @@ impl PartialEq for Upstream {
         match (self, other) {
             (Self::Udp(a), Self::Udp(b)) => a == b,
             (Self::Doh { url: a, .. }, Self::Doh { url: b, .. }) => a == b,
+            (Self::Dot { addr: a, .. }, Self::Dot { addr: b, .. }) => a == b,
             _ => false,
         }
     }
 }
 
+impl fmt::Debug for Upstream {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        fmt::Display::fmt(self, f)
+    }
+}
+
 impl fmt::Display for Upstream {
     fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         match self {
             Upstream::Udp(addr) => write!(f, "{}", addr),
             Upstream::Doh { url, .. } => f.write_str(url),
+            Upstream::Dot { addr, tls_name, .. } => match tls_name {
+                Some(name) => write!(f, "tls://{}#{}", addr, name),
+                None => write!(f, "tls://{}", addr),
+            },
+        }
+    }
+}
+
+pub(crate) fn parse_upstream_addr(
+    s: &str,
+    default_port: u16,
+) -> std::result::Result<SocketAddr, String> {
+    // Try full socket addr first: "1.2.3.4:5353" or "[::1]:5353"
+    if let Ok(addr) = s.parse::<SocketAddr>() {
+        return Ok(addr);
+    }
+    // Bare IP: "1.2.3.4" or "::1"
+    if let Ok(ip) = s.parse::<IpAddr>() {
+        return Ok(SocketAddr::new(ip, default_port));
+    }
+    Err(format!("invalid upstream address: {}", s))
+}
+
+pub fn parse_upstream(s: &str, default_port: u16) -> Result<Upstream> {
+    if s.starts_with("https://") {
+        let client = reqwest::Client::builder()
+            .use_rustls_tls()
+            .http2_initial_stream_window_size(65_535)
+            .http2_initial_connection_window_size(65_535)
+            .http2_keep_alive_interval(Duration::from_secs(15))
+            .http2_keep_alive_while_idle(true)
+            .http2_keep_alive_timeout(Duration::from_secs(10))
+            .pool_idle_timeout(Duration::from_secs(300))
+            .pool_max_idle_per_host(1)
+            .build()
+            .unwrap_or_default();
+        return Ok(Upstream::Doh {
+            url: s.to_string(),
+            client,
+        });
+    }
+    // tls://IP:PORT#hostname  or  tls://IP#hostname  (default port 853)
+    if let Some(rest) = s.strip_prefix("tls://") {
+        let (addr_part, tls_name) = match rest.find('#') {
+            Some(i) => (&rest[..i], Some(rest[i + 1..].to_string())),
+            None => (rest, None),
+        };
+        let addr = parse_upstream_addr(addr_part, 853)?;
+        let connector = build_dot_connector()?;
+        return Ok(Upstream::Dot {
+            addr,
+            tls_name,
+            connector,
+        });
+    }
+    let addr = parse_upstream_addr(s, default_port)?;
+    Ok(Upstream::Udp(addr))
+}
+
+fn build_dot_connector() -> Result<tokio_rustls::TlsConnector> {
+    let _ = rustls::crypto::ring::default_provider().install_default();
+    let mut root_store = rustls::RootCertStore::empty();
+    root_store.extend(webpki_roots::TLS_SERVER_ROOTS.iter().cloned());
+    let config = rustls::ClientConfig::builder()
+        .with_root_certificates(root_store)
+        .with_no_client_auth();
+    Ok(tokio_rustls::TlsConnector::from(std::sync::Arc::new(
+        config,
+    )))
+}
+
+#[derive(Clone)]
+pub struct UpstreamPool {
+    primary: Vec<Upstream>,
+    fallback: Vec<Upstream>,
+}
+
+impl UpstreamPool {
+    pub fn new(primary: Vec<Upstream>, fallback: Vec<Upstream>) -> Self {
+        Self { primary, fallback }
+    }
+
+    pub fn preferred(&self) -> Option<&Upstream> {
+        self.primary.first().or(self.fallback.first())
+    }
+
+    pub fn set_primary(&mut self, primary: Vec<Upstream>) {
+        self.primary = primary;
+    }
+
+    /// Update the primary upstream if `new_addr` (parsed with `port`) differs
+    /// from the current preferred upstream. Returns `true` if the pool changed.
+    pub fn maybe_update_primary(&mut self, new_addr: &str, port: u16) -> bool {
+        let Ok(new_sock) = format!("{}:{}", new_addr, port).parse::<SocketAddr>() else {
+            return false;
+        };
+        let new_upstream = Upstream::Udp(new_sock);
+        if self.preferred() == Some(&new_upstream) {
+            return false;
+        }
+        self.primary = vec![new_upstream];
+        true
+    }
+
+    pub fn label(&self) -> String {
+        match self.preferred() {
+            Some(u) => {
+                let total = self.primary.len() + self.fallback.len();
+                if total > 1 {
+                    format!("{} (+{} more)", u, total - 1)
+                } else {
+                    u.to_string()
+                }
+            }
+            None => "none".to_string(),
         }
     }
 }
@@ -42,10 +171,11 @@ pub async fn forward_query(
     upstream: &Upstream,
     timeout_duration: Duration,
 ) -> Result<DnsPacket> {
-    match upstream {
+    let mut send_buffer = BytePacketBuffer::new();
-        Upstream::Udp(addr) => forward_udp(query, *addr, timeout_duration).await,
+    query.write(&mut send_buffer)?;
-        Upstream::Doh { url, client } => forward_doh(query, url, client, timeout_duration).await,
+    let data = forward_query_raw(send_buffer.filled(), upstream, timeout_duration).await?;
-    }
+    let mut recv_buffer = BytePacketBuffer::from_bytes(&data);
+    DnsPacket::from_buffer(&mut recv_buffer)
 }
 
 pub(crate) async fn forward_udp(
@@ -53,24 +183,10 @@ pub(crate) async fn forward_udp(
     upstream: SocketAddr,
     timeout_duration: Duration,
 ) -> Result<DnsPacket> {
-    let socket = UdpSocket::bind("0.0.0.0:0").await?;
-
     let mut send_buffer = BytePacketBuffer::new();
     query.write(&mut send_buffer)?;
-
+    let data = forward_udp_raw(send_buffer.filled(), upstream, timeout_duration).await?;
-    socket.send_to(send_buffer.filled(), upstream).await?;
+    let mut recv_buffer = BytePacketBuffer::from_bytes(&data);
-
-    let mut recv_buffer = BytePacketBuffer::new();
-    let (size, _) = timeout(timeout_duration, socket.recv_from(&mut recv_buffer.buf)).await??;
-
-    if size == recv_buffer.buf.len() {
-        log::debug!(
-            "upstream response truncated ({} bytes, buffer {})",
-            size,
-            recv_buffer.buf.len()
-        );
-    }
-
     DnsPacket::from_buffer(&mut recv_buffer)
 }
 
@@ -107,22 +223,209 @@ pub(crate) async fn forward_tcp(
     DnsPacket::from_buffer(&mut recv_buffer)
 }
 
-async fn forward_doh(
+async fn forward_dot_raw(
-    query: &DnsPacket,
+    wire: &[u8],
+    addr: SocketAddr,
+    tls_name: &Option<String>,
+    connector: &tokio_rustls::TlsConnector,
+    timeout_duration: Duration,
+) -> Result<Vec<u8>> {
+    use rustls::pki_types::ServerName;
+    use tokio::io::{AsyncReadExt, AsyncWriteExt};
+    use tokio::net::TcpStream;
+
+    let server_name = match tls_name {
+        Some(name) => ServerName::try_from(name.clone())?,
+        None => ServerName::try_from(addr.ip().to_string())?,
+    };
+
+    let tcp = timeout(timeout_duration, TcpStream::connect(addr)).await??;
+    let mut tls = timeout(timeout_duration, connector.connect(server_name, tcp)).await??;
+
+    let mut outbuf = Vec::with_capacity(2 + wire.len());
+    outbuf.extend_from_slice(&(wire.len() as u16).to_be_bytes());
+    outbuf.extend_from_slice(wire);
+    timeout(timeout_duration, tls.write_all(&outbuf)).await??;
+
+    let mut len_buf = [0u8; 2];
+    timeout(timeout_duration, tls.read_exact(&mut len_buf)).await??;
+    let resp_len = u16::from_be_bytes(len_buf) as usize;
+
+    let mut data = vec![0u8; resp_len];
+    timeout(timeout_duration, tls.read_exact(&mut data)).await??;
+
+    Ok(data)
+}
+
+pub async fn forward_query_raw(
+    wire: &[u8],
+    upstream: &Upstream,
+    timeout_duration: Duration,
+) -> Result<Vec<u8>> {
+    match upstream {
+        Upstream::Udp(addr) => forward_udp_raw(wire, *addr, timeout_duration).await,
+        Upstream::Doh { url, client } => forward_doh_raw(wire, url, client, timeout_duration).await,
+        Upstream::Dot {
+            addr,
+            tls_name,
+            connector,
+        } => forward_dot_raw(wire, *addr, tls_name, connector, timeout_duration).await,
+    }
+}
+
+pub async fn forward_with_hedging_raw(
+    wire: &[u8],
+    primary: &Upstream,
+    secondary: &Upstream,
+    hedge_delay: Duration,
+    timeout_duration: Duration,
+) -> Result<Vec<u8>> {
+    use tokio::time::sleep;
+
+    let primary_fut = forward_query_raw(wire, primary, timeout_duration);
+    tokio::pin!(primary_fut);
+
+    let delay = sleep(hedge_delay);
+    tokio::pin!(delay);
+
+    // Phase 1: wait for either primary to return, or the hedge delay.
+    tokio::select! {
+        result = &mut primary_fut => return result,
+        _ = &mut delay => {}
+    }
+
+    // Phase 2: hedge delay expired — fire secondary while still polling primary.
+    let secondary_fut = forward_query_raw(wire, secondary, timeout_duration);
+    tokio::pin!(secondary_fut);
+
+    // First successful response wins. If one errors, wait for the other.
+    let mut primary_err: Option<crate::Error> = None;
+    let mut secondary_err: Option<crate::Error> = None;
+
+    loop {
+        tokio::select! {
+            r = &mut primary_fut, if primary_err.is_none() => {
+                match r {
+                    Ok(resp) => return Ok(resp),
+                    Err(e) => {
+                        if let Some(se) = secondary_err.take() {
+                            return Err(se);
+                        }
+                        primary_err = Some(e);
+                    }
+                }
+            }
+            r = &mut secondary_fut, if secondary_err.is_none() => {
+                match r {
+                    Ok(resp) => return Ok(resp),
+                    Err(e) => {
+                        if let Some(pe) = primary_err.take() {
+                            return Err(pe);
+                        }
+                        secondary_err = Some(e);
+                    }
+                }
+            }
+        }
+
+        match (primary_err, secondary_err) {
+            (Some(pe), Some(_)) => return Err(pe),
+            (pe, se) => {
+                primary_err = pe;
+                secondary_err = se;
+            }
+        }
+    }
+}
+
+pub async fn forward_with_failover_raw(
+    wire: &[u8],
+    pool: &UpstreamPool,
+    srtt: &RwLock<SrttCache>,
+    timeout_duration: Duration,
+    hedge_delay: Duration,
+) -> Result<Vec<u8>> {
+    let mut candidates: Vec<(usize, u64)> = pool
+        .primary
+        .iter()
+        .enumerate()
+        .map(|(i, u)| {
+            let rtt = match u {
+                Upstream::Udp(addr) => srtt.read().unwrap().get(addr.ip()),
+                _ => 0,
+            };
+            (i, rtt)
+        })
+        .collect();
+    candidates.sort_by_key(|&(_, rtt)| rtt);
+
+    let all_upstreams: Vec<&Upstream> = candidates
+        .iter()
+        .map(|&(i, _)| &pool.primary[i])
+        .chain(pool.fallback.iter())
+        .collect();
+
+    let mut last_err: Option<Box<dyn std::error::Error + Send + Sync>> = None;
+
+    for upstream in &all_upstreams {
+        let start = Instant::now();
+        let result = if !hedge_delay.is_zero() {
+            // Hedge against the same upstream: independent h2 streams (DoH),
+            // independent UDP packets (plain DNS), or independent TLS
+            // connections (DoT). Rescues packet loss, dispatch spikes, and
+            // TLS handshake stalls.
+            forward_with_hedging_raw(wire, upstream, upstream, hedge_delay, timeout_duration).await
+        } else {
+            forward_query_raw(wire, upstream, timeout_duration).await
+        };
+        match result {
+            Ok(resp) => {
+                if let Upstream::Udp(addr) = upstream {
+                    let rtt_ms = start.elapsed().as_millis() as u64;
+                    srtt.write().unwrap().record_rtt(addr.ip(), rtt_ms, false);
+                }
+                return Ok(resp);
+            }
+            Err(e) => {
+                if let Upstream::Udp(addr) = upstream {
+                    srtt.write().unwrap().record_failure(addr.ip());
+                }
+                log::debug!("upstream {} failed: {}", upstream, e);
+                last_err = Some(e);
+            }
+        }
+    }
+
+    Err(last_err.unwrap_or_else(|| "no upstream configured".into()))
+}
+
+async fn forward_udp_raw(
+    wire: &[u8],
+    upstream: SocketAddr,
+    timeout_duration: Duration,
+) -> Result<Vec<u8>> {
+    let socket = UdpSocket::bind("0.0.0.0:0").await?;
+    socket.send_to(wire, upstream).await?;
+
+    let mut recv_buf = vec![0u8; 4096];
+    let (size, _) = timeout(timeout_duration, socket.recv_from(&mut recv_buf)).await??;
+    recv_buf.truncate(size);
+    Ok(recv_buf)
+}
+
+async fn forward_doh_raw(
+    wire: &[u8],
     url: &str,
     client: &reqwest::Client,
     timeout_duration: Duration,
-) -> Result<DnsPacket> {
+) -> Result<Vec<u8>> {
-    let mut send_buffer = BytePacketBuffer::new();
-    query.write(&mut send_buffer)?;
-
     let resp = timeout(
         timeout_duration,
         client
             .post(url)
             .header("content-type", "application/dns-message")
             .header("accept", "application/dns-message")
-            .body(send_buffer.filled().to_vec())
+            .body(wire.to_vec())
             .send(),
     )
     .await??
@@ -130,9 +433,25 @@ async fn forward_doh(
 
     let bytes = resp.bytes().await?;
     log::debug!("DoH response: {} bytes", bytes.len());
+    Ok(bytes.to_vec())
+}
 
-    let mut recv_buffer = BytePacketBuffer::from_bytes(&bytes);
+/// Send a lightweight keepalive query to a DoH upstream to prevent
-    DnsPacket::from_buffer(&mut recv_buffer)
+/// the HTTP/2 + TLS connection from going idle and being torn down.
+pub async fn keepalive_doh(upstream: &Upstream) {
+    if let Upstream::Doh { url, client } = upstream {
+        // Query for . NS — minimal, always succeeds, response is small
+        let wire: &[u8] = &[
+            0x00, 0x00, // ID
+            0x01, 0x00, // flags: RD=1
+            0x00, 0x01, // QDCOUNT=1
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, // AN=0, NS=0, AR=0
+            0x00, // root name (.)
+            0x00, 0x02, // type NS
+            0x00, 0x01, // class IN
+        ];
+        let _ = forward_doh_raw(wire, url, client, Duration::from_secs(5)).await;
+    }
 }
 
 #[cfg(test)]
@@ -271,4 +590,121 @@ mod tests {
         let result = forward_query(&make_query(), &upstream, Duration::from_millis(100)).await;
         assert!(result.is_err());
     }
+
+    #[test]
+    fn parse_addr_ip_only() {
+        let addr = parse_upstream_addr("1.2.3.4", 53).unwrap();
+        assert_eq!(addr, "1.2.3.4:53".parse::<SocketAddr>().unwrap());
+    }
+
+    #[test]
+    fn parse_addr_ip_port() {
+        let addr = parse_upstream_addr("1.2.3.4:5353", 53).unwrap();
+        assert_eq!(addr, "1.2.3.4:5353".parse::<SocketAddr>().unwrap());
+    }
+
+    #[test]
+    fn parse_addr_ipv6_bracketed() {
+        let addr = parse_upstream_addr("[::1]:5553", 53).unwrap();
+        assert_eq!(addr, "[::1]:5553".parse::<SocketAddr>().unwrap());
+    }
+
+    #[test]
+    fn parse_addr_ipv6_bare() {
+        let addr = parse_upstream_addr("::1", 53).unwrap();
+        assert_eq!(addr, "[::1]:53".parse::<SocketAddr>().unwrap());
+    }
+
+    #[test]
+    fn pool_label_single() {
+        let pool = UpstreamPool::new(vec![Upstream::Udp("1.2.3.4:53".parse().unwrap())], vec![]);
+        assert_eq!(pool.label(), "1.2.3.4:53");
+    }
+
+    #[test]
+    fn pool_label_multi() {
+        let pool = UpstreamPool::new(
+            vec![Upstream::Udp("1.2.3.4:53".parse().unwrap())],
+            vec![Upstream::Udp("8.8.8.8:53".parse().unwrap())],
+        );
+        assert_eq!(pool.label(), "1.2.3.4:53 (+1 more)");
+    }
+
+    #[tokio::test]
+    async fn failover_tries_next_on_failure() {
+        // First upstream is unreachable, second responds
+        let query = make_query();
+        let response_bytes = to_wire(&make_response(&query));
+
+        let app = axum::Router::new().route(
+            "/dns-query",
+            axum::routing::post(move || {
+                let body = response_bytes.clone();
+                async move {
+                    (
+                        [(axum::http::header::CONTENT_TYPE, "application/dns-message")],
+                        body,
+                    )
+                }
+            }),
+        );
+
+        let listener = tokio::net::TcpListener::bind("127.0.0.1:0").await.unwrap();
+        let good_addr = listener.local_addr().unwrap();
+        tokio::spawn(axum::serve(listener, app).into_future());
+
+        // Unreachable UDP upstream + working DoH upstream
+        let pool = UpstreamPool::new(
+            vec![
+                Upstream::Udp("127.0.0.1:1".parse().unwrap()), // will fail
+                Upstream::Doh {
+                    url: format!("http://{}/dns-query", good_addr),
+                    client: reqwest::Client::new(),
+                },
+            ],
+            vec![],
+        );
+
+        let srtt = RwLock::new(SrttCache::new(true));
+        let wire = to_wire(&query);
+        let resp_wire = forward_with_failover_raw(
+            &wire,
+            &pool,
+            &srtt,
+            Duration::from_millis(500),
+            Duration::ZERO,
+        )
+        .await
+        .expect("should fail over to second upstream");
+
+        let mut buf = BytePacketBuffer::from_bytes(&resp_wire);
+        let result = DnsPacket::from_buffer(&mut buf).unwrap();
+        assert_eq!(result.header.id, 0xABCD);
+        assert_eq!(result.answers.len(), 1);
+    }
+
+    #[test]
+    fn maybe_update_primary_swaps_when_different() {
+        let mut pool = UpstreamPool::new(
+            vec![Upstream::Udp("1.2.3.4:53".parse().unwrap())],
+            vec![Upstream::Udp("8.8.8.8:53".parse().unwrap())],
+        );
+        assert!(pool.maybe_update_primary("5.6.7.8", 53));
+        assert_eq!(pool.preferred().unwrap().to_string(), "5.6.7.8:53");
+    }
+
+    #[test]
+    fn maybe_update_primary_noop_when_same() {
+        let mut pool =
+            UpstreamPool::new(vec![Upstream::Udp("1.2.3.4:53".parse().unwrap())], vec![]);
+        assert!(!pool.maybe_update_primary("1.2.3.4", 53));
+    }
+
+    #[test]
+    fn maybe_update_primary_rejects_invalid_addr() {
+        let mut pool =
+            UpstreamPool::new(vec![Upstream::Udp("1.2.3.4:53".parse().unwrap())], vec![]);
+        assert!(!pool.maybe_update_primary("not-an-ip", 53));
+        assert_eq!(pool.preferred().unwrap().to_string(), "1.2.3.4:53");
+    }
 }

src/health.rs

@@ -0,0 +1,258 @@
+//! Health metadata and `/health` response shape, shared between the main
+//! HTTP API and the mobile API.
+//!
+//! The static fields (version, hostname, DoT config, CA fingerprint,
+//! feature list) are computed once at startup and stored in [`HealthMeta`]
+//! on `ServerCtx`. Per-request fields (uptime, LAN IP) are computed live.
+//! Both handlers call [`HealthResponse::build`] to assemble the JSON
+//! response from `HealthMeta` + live inputs.
+//!
+//! JSON schema is documented in `docs/implementation/ios-companion-app.md`
+//! §4.2. The iOS companion app's `HealthInfo` struct is the canonical
+//! consumer; any change to this response must keep that struct decoding
+//! cleanly (all consumed fields are optional on the Swift side, but
+//! `lan_ip` is load-bearing for the pipeline).
+
+use std::net::Ipv4Addr;
+use std::path::Path;
+use std::time::Instant;
+
+use ring::digest::{digest, SHA256};
+use serde::Serialize;
+
+/// Immutable health metadata cached on `ServerCtx`. Built once at startup
+/// from config + file-system state (CA cert).
+#[derive(Clone)]
+pub struct HealthMeta {
+    pub version: &'static str,
+    pub hostname: String,
+    pub sni: String,
+    pub dot_enabled: bool,
+    pub dot_port: u16,
+    pub api_port: u16,
+    pub ca_fingerprint_sha256: Option<String>,
+    pub features: Vec<String>,
+    pub started_at: Instant,
+}
+
+impl HealthMeta {
+    /// Minimal `HealthMeta` for unit tests that construct a `ServerCtx`
+    /// without needing the real startup flow (CA file reads, hostname
+    /// detection, etc.). Deterministic values so test JSON assertions
+    /// stay stable.
+    #[cfg(test)]
+    pub fn test_fixture() -> Self {
+        HealthMeta {
+            version: env!("CARGO_PKG_VERSION"),
+            hostname: "test-host".to_string(),
+            sni: "numa.numa".to_string(),
+            dot_enabled: false,
+            dot_port: 853,
+            api_port: 8765,
+            ca_fingerprint_sha256: None,
+            features: vec![],
+            started_at: Instant::now(),
+        }
+    }
+
+    /// Build a new HealthMeta from config + startup-time environment.
+    /// Call once at server boot; the returned value is cheap to clone
+    /// (small number of short strings) and lives on `ServerCtx`.
+    ///
+    /// The argument count is deliberate — each flag corresponds to a
+    /// specific config value and is clearly named at the call site.
+    /// Collapsing into a struct hides nothing meaningful for a one-call
+    /// initializer.
+    #[allow(clippy::too_many_arguments)]
+    pub fn build(
+        data_dir: &Path,
+        dot_enabled: bool,
+        dot_port: u16,
+        api_port: u16,
+        dnssec_enabled: bool,
+        recursive_enabled: bool,
+        mdns_enabled: bool,
+        blocking_enabled: bool,
+        doh_enabled: bool,
+    ) -> Self {
+        let ca_path = data_dir.join("ca.pem");
+        let ca_fingerprint_sha256 = compute_ca_fingerprint(&ca_path);
+
+        let mut features = Vec::new();
+        if doh_enabled {
+            features.push("doh".to_string());
+        }
+        if dot_enabled {
+            features.push("dot".to_string());
+        }
+        if recursive_enabled {
+            features.push("recursive".to_string());
+        }
+        if blocking_enabled {
+            features.push("blocking".to_string());
+        }
+        if mdns_enabled {
+            features.push("mdns".to_string());
+        }
+        if dnssec_enabled {
+            features.push("dnssec".to_string());
+        }
+
+        HealthMeta {
+            version: env!("CARGO_PKG_VERSION"),
+            hostname: crate::hostname(),
+            sni: "numa.numa".to_string(),
+            dot_enabled,
+            dot_port,
+            api_port,
+            ca_fingerprint_sha256,
+            features,
+            started_at: Instant::now(),
+        }
+    }
+}
+
+/// JSON response shape returned by `GET /health` on both main and mobile APIs.
+///
+/// Fields are organized to match the iOS companion app's
+/// `HealthInfo` Swift struct — see `ios-companion-app.md` §4.2.
+#[derive(Serialize)]
+pub struct HealthResponse {
+    pub status: &'static str,
+    pub version: &'static str,
+    pub uptime_secs: u64,
+    pub hostname: String,
+    pub lan_ip: Option<String>,
+    pub sni: String,
+    pub dot: DotBlock,
+    pub api: ApiBlock,
+    pub ca: CaBlock,
+    pub features: Vec<String>,
+}
+
+#[derive(Serialize)]
+pub struct DotBlock {
+    pub enabled: bool,
+    pub port: Option<u16>,
+}
+
+#[derive(Serialize)]
+pub struct ApiBlock {
+    pub port: u16,
+}
+
+#[derive(Serialize)]
+pub struct CaBlock {
+    pub present: bool,
+    pub fingerprint_sha256: Option<String>,
+}
+
+impl HealthResponse {
+    /// Assemble a fresh `HealthResponse` from the cached metadata and
+    /// the current LAN IP (which may change across network transitions).
+    /// Pass `None` for `lan_ip` if detection fails — the response still
+    /// returns 200 OK, just without the LAN address.
+    pub fn build(meta: &HealthMeta, lan_ip: Option<Ipv4Addr>) -> Self {
+        HealthResponse {
+            status: "ok",
+            version: meta.version,
+            uptime_secs: meta.started_at.elapsed().as_secs(),
+            hostname: meta.hostname.clone(),
+            lan_ip: lan_ip.map(|ip| ip.to_string()),
+            sni: meta.sni.clone(),
+            dot: DotBlock {
+                enabled: meta.dot_enabled,
+                port: if meta.dot_enabled {
+                    Some(meta.dot_port)
+                } else {
+                    None
+                },
+            },
+            api: ApiBlock {
+                port: meta.api_port,
+            },
+            ca: CaBlock {
+                present: meta.ca_fingerprint_sha256.is_some(),
+                fingerprint_sha256: meta.ca_fingerprint_sha256.clone(),
+            },
+            features: meta.features.clone(),
+        }
+    }
+}
+
+/// Read the CA cert at `ca_path` and return its SHA-256 fingerprint as a
+/// lowercase hex string, or None if the file doesn't exist or can't be read.
+///
+/// Hashes the raw PEM bytes for simplicity. A more canonical SPKI-based
+/// fingerprint would require parsing the PEM → DER → extracting
+/// SubjectPublicKeyInfo, which adds complexity without meaningful benefit
+/// for our use case (the iOS app uses the fingerprint only for display
+/// and to detect rotation).
+fn compute_ca_fingerprint(ca_path: &Path) -> Option<String> {
+    let pem = std::fs::read(ca_path).ok()?;
+    let hash = digest(&SHA256, &pem);
+    let hex: String = hash.as_ref().iter().map(|b| format!("{:02x}", b)).collect();
+    Some(hex)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn health_response_contains_required_fields() {
+        let meta = HealthMeta {
+            version: "0.10.0",
+            hostname: "test-host".to_string(),
+            sni: "numa.numa".to_string(),
+            dot_enabled: true,
+            dot_port: 853,
+            api_port: 8765,
+            ca_fingerprint_sha256: Some("abcd1234".to_string()),
+            features: vec!["dot".to_string(), "dnssec".to_string()],
+            started_at: Instant::now(),
+        };
+
+        let response = HealthResponse::build(&meta, Some(Ipv4Addr::new(192, 168, 1, 50)));
+        let json = serde_json::to_string(&response).unwrap();
+
+        assert!(json.contains("\"status\":\"ok\""));
+        assert!(json.contains("\"version\":\"0.10.0\""));
+        assert!(json.contains("\"hostname\":\"test-host\""));
+        assert!(json.contains("\"lan_ip\":\"192.168.1.50\""));
+        assert!(json.contains("\"sni\":\"numa.numa\""));
+        assert!(json.contains("\"port\":853"));
+        assert!(json.contains("\"port\":8765"));
+        assert!(json.contains("\"fingerprint_sha256\":\"abcd1234\""));
+        assert!(json.contains("\"features\":[\"dot\",\"dnssec\"]"));
+    }
+
+    #[test]
+    fn health_response_omits_dot_port_when_disabled() {
+        let meta = HealthMeta {
+            version: "0.10.0",
+            hostname: "t".to_string(),
+            sni: "numa.numa".to_string(),
+            dot_enabled: false,
+            dot_port: 853,
+            api_port: 8765,
+            ca_fingerprint_sha256: None,
+            features: vec![],
+            started_at: Instant::now(),
+        };
+
+        let response = HealthResponse::build(&meta, None);
+        let json = serde_json::to_string(&response).unwrap();
+
+        assert!(json.contains("\"enabled\":false"));
+        assert!(json.contains("\"dot\":{\"enabled\":false,\"port\":null}"));
+        assert!(json.contains("\"present\":false"));
+        assert!(json.contains("\"lan_ip\":null"));
+    }
+
+    #[test]
+    fn ca_fingerprint_returns_none_for_missing_file() {
+        let fp = compute_ca_fingerprint(Path::new("/nonexistent/ca.pem"));
+        assert!(fp.is_none());
+    }
+}

src/lan.rs

@@ -9,6 +9,7 @@ use crate::buffer::BytePacketBuffer;
 use crate::config::LanConfig;
 use crate::ctx::ServerCtx;
 use crate::header::DnsHeader;
+use crate::health::HealthMeta;
 use crate::question::{DnsQuestion, QueryType};
 
 // --- Constants ---
@@ -18,6 +19,18 @@ const MDNS_PORT: u16 = 5353;
 const SERVICE_TYPE: &str = "_numa._tcp.local";
 const MDNS_TTL: u32 = 120;
 
+// TXT record key prefixes (including the trailing `=`). Shared between
+// the sender (`build_announcement`) and the receiver (`parse_mdns_response`)
+// to prevent drift — both sides match on the same literal, not on two
+// independent string constants that could diverge.
+const TXT_SERVICES: &str = "services=";
+const TXT_ID: &str = "id=";
+const TXT_VERSION: &str = "version=";
+const TXT_API_PORT: &str = "api_port=";
+const TXT_PROTO: &str = "proto=";
+const TXT_DOT_PORT: &str = "dot_port=";
+const TXT_CA_FP: &str = "ca_fp=";
+
 // --- Peer Store ---
 
 pub struct PeerStore {
@@ -97,14 +110,16 @@ pub fn detect_lan_ip() -> Option<Ipv4Addr> {
     }
 }
 
+/// Short hostname for mDNS instance names (`<short>._numa._tcp.local`).
+/// Truncates at the first `.` so `macbook-pro.local` becomes `macbook-pro`.
+/// Uses the shared `crate::hostname()` helper as the source.
 fn get_hostname() -> String {
-    std::process::Command::new("hostname")
+    crate::hostname()
-        .output()
+        .split('.')
-        .ok()
+        .next()
-        .and_then(|o| String::from_utf8(o.stdout).ok())
+        .filter(|s| !s.is_empty())
-        .map(|h| h.trim().split('.').next().unwrap_or("numa").to_string())
+        .unwrap_or("numa")
-        .filter(|h| !h.is_empty())
+        .to_string()
-        .unwrap_or_else(|| "numa".to_string())
 }
 
 /// Generate a per-process instance ID for self-filtering on multi-instance hosts
@@ -168,13 +183,22 @@ pub async fn start_lan_discovery(ctx: Arc<ServerCtx>, config: &LanConfig) {
                     .map(|e| (e.name.clone(), e.target_port))
                     .collect()
             };
-            if services.is_empty() {
+            // Note: we always announce ourselves, even when the
-                continue;
+            // services list is empty. The announcement still carries
-            }
+            // the mobile API port + version + CA fingerprint in TXT,
+            // which is what the iOS companion app browses for via
+            // NWBrowser on `_numa._tcp.local`. Other Numa peers
+            // receive these empty-services announcements too and
+            // correctly ignore them in parse_mdns_response (the
+            // receiver only processes when services is non-empty).
             let current_ip = *sender_ctx.lan_ip.lock().unwrap();
-            if let Ok(pkt) =
+            if let Ok(pkt) = build_announcement(
-                build_announcement(&sender_hostname, current_ip, &services, &sender_instance_id)
+                &sender_hostname,
-            {
+                current_ip,
+                &services,
+                &sender_instance_id,
+                &sender_ctx.health_meta,
+            ) {
                 let _ = sender_socket.send_to(pkt.filled(), dest).await;
             }
         }
@@ -240,6 +264,7 @@ fn build_announcement(
     ip: Ipv4Addr,
     services: &[(String, u16)],
     inst_id: &str,
+    meta: &HealthMeta,
 ) -> crate::Result<BytePacketBuffer> {
     let mut buf = BytePacketBuffer::new();
     let instance_name = format!("{}._numa._tcp.local", hostname);
@@ -260,7 +285,11 @@ fn build_announcement(
     patch_rdlen(&mut buf, rdlen_pos, rdata_start)?;
 
     // SRV: <instance>._numa._tcp.local → <hostname>.local
-    // Port in SRV is informational; actual service ports are in TXT
+    // Port = mobile API port, which is what the iOS companion app resolves
+    // the SRV record for. Legacy Numa peers don't read the SRV port (see
+    // parse_mdns_response — it only uses TXT services= for peer discovery),
+    // so changing the SRV port from "first service's port" to the mobile
+    // API port is backwards compatible.
     write_record_header(
         &mut buf,
         &instance_name,
@@ -273,11 +302,13 @@ fn build_announcement(
     let rdata_start = buf.pos();
     buf.write_u16(0)?; // priority
     buf.write_u16(0)?; // weight
-    buf.write_u16(services.first().map(|(_, p)| *p).unwrap_or(0))?; // first service port for SRV display
+    buf.write_u16(meta.api_port)?; // mobile API port, for iOS companion app
     buf.write_qname(&host_local)?;
     patch_rdlen(&mut buf, rdlen_pos, rdata_start)?;
 
-    // TXT: services + instance ID for self-filtering
+    // TXT: legacy peer-discovery entries (services, id) + enriched entries
+    // for the iOS companion app (version, api_port, proto, dot_port, ca_fp).
+    // All in one TXT RRset per mDNS convention.
     write_record_header(
         &mut buf,
         &instance_name,
@@ -293,8 +324,21 @@ fn build_announcement(
         .map(|(name, port)| format!("{}:{}", name, port))
         .collect::<Vec<_>>()
         .join(",");
-    write_txt_string(&mut buf, &format!("services={}", svc_str))?;
+    // Legacy peer-discovery entries (consumed by parse_mdns_response)
-    write_txt_string(&mut buf, &format!("id={}", inst_id))?;
+    write_txt_string(&mut buf, &format!("{}{}", TXT_SERVICES, svc_str))?;
+    write_txt_string(&mut buf, &format!("{}{}", TXT_ID, inst_id))?;
+    // Enriched entries (consumed by the iOS/Android companion apps)
+    write_txt_string(&mut buf, &format!("{}{}", TXT_VERSION, meta.version))?;
+    write_txt_string(&mut buf, &format!("{}{}", TXT_API_PORT, meta.api_port))?;
+    if meta.dot_enabled {
+        write_txt_string(&mut buf, &format!("{}dot", TXT_PROTO))?;
+        write_txt_string(&mut buf, &format!("{}{}", TXT_DOT_PORT, meta.dot_port))?;
+    } else {
+        write_txt_string(&mut buf, &format!("{}plain", TXT_PROTO))?;
+    }
+    if let Some(fp) = &meta.ca_fingerprint_sha256 {
+        write_txt_string(&mut buf, &format!("{}{}", TXT_CA_FP, fp))?;
+    }
     patch_rdlen(&mut buf, rdlen_pos, rdata_start)?;
 
     // A: <hostname>.local → IP
@@ -408,7 +452,7 @@ fn parse_mdns_response(data: &[u8]) -> Option<MdnsAnnouncement> {
                         break;
                     }
                     if let Ok(txt) = std::str::from_utf8(&data[pos..pos + txt_len]) {
-                        if let Some(val) = txt.strip_prefix("services=") {
+                        if let Some(val) = txt.strip_prefix(TXT_SERVICES) {
                             let svcs: Vec<(String, u16)> = val
                                 .split(',')
                                 .filter_map(|s| {
@@ -421,7 +465,7 @@ fn parse_mdns_response(data: &[u8]) -> Option<MdnsAnnouncement> {
                             if !svcs.is_empty() {
                                 txt_services = Some(svcs);
                             }
-                        } else if let Some(id) = txt.strip_prefix("id=") {
+                        } else if let Some(id) = txt.strip_prefix(TXT_ID) {
                             peer_instance_id = Some(id.to_string());
                         }
                     }

src/lib.rs

@@ -5,10 +5,14 @@ pub mod cache;
 pub mod config;
 pub mod ctx;
 pub mod dnssec;
+pub mod doh;
 pub mod dot;
 pub mod forward;
 pub mod header;
+pub mod health;
 pub mod lan;
+pub mod mobile_api;
+pub mod mobileconfig;
 pub mod override_store;
 pub mod packet;
 pub mod proxy;
@@ -17,16 +21,74 @@ pub mod question;
 pub mod record;
 pub mod recursive;
 pub mod service_store;
+pub mod setup_phone;
 pub mod srtt;
 pub mod stats;
 pub mod system_dns;
 pub mod tls;
+pub mod wire;
+
+#[cfg(test)]
+pub(crate) mod testutil;
 
 pub type Error = Box<dyn std::error::Error + Send + Sync>;
 pub type Result<T> = std::result::Result<T, Error>;
 
+/// Detect the machine hostname via the `hostname` command. Returns the
+/// full hostname (e.g., `macbook-pro.local`), or `"numa"` if the command
+/// fails. Call sites that need the short form (e.g., mDNS instance
+/// names) should truncate at the first `.`.
+pub fn hostname() -> String {
+    std::process::Command::new("hostname")
+        .output()
+        .ok()
+        .and_then(|o| String::from_utf8(o.stdout).ok())
+        .map(|h| h.trim().to_string())
+        .filter(|h| !h.is_empty())
+        .unwrap_or_else(|| "numa".to_string())
+}
+
+/// Path to suggest to an interactive user when asking them to create
+/// `numa.toml`. Prefers `$HOME/.config/numa/numa.toml` when HOME is set
+/// (actionable without sudo); falls back to `config_dir()` otherwise.
+///
+/// Note: `config_dir()` routes interactive root to FHS (`/var/lib/numa`)
+/// so that runtime state like `services.json` stays continuous with the
+/// installed daemon. This helper exists specifically to give advisories
+/// and `load_config` an XDG-aware path for user-authored config, without
+/// moving runtime state out of FHS — see issue #81.
+pub(crate) fn suggested_config_path() -> std::path::PathBuf {
+    #[cfg(not(windows))]
+    {
+        resolve_suggested_config_path(std::env::var("HOME").ok().as_deref(), config_dir)
+    }
+    #[cfg(windows)]
+    {
+        config_dir().join("numa.toml")
+    }
+}
+
+#[cfg(not(windows))]
+fn resolve_suggested_config_path<F>(home: Option<&str>, fallback_dir: F) -> std::path::PathBuf
+where
+    F: FnOnce() -> std::path::PathBuf,
+{
+    if let Some(home) = home {
+        if !home.is_empty() && home != "/" {
+            return std::path::PathBuf::from(home)
+                .join(".config")
+                .join("numa")
+                .join("numa.toml");
+        }
+    }
+    fallback_dir().join("numa.toml")
+}
+
 /// Shared config directory for persistent data (services.json, etc).
-/// Unix: ~/.config/numa/ (or /usr/local/var/numa/ when running as root daemon)
+/// Unix users: ~/.config/numa/
+/// Linux root daemon: /var/lib/numa (FHS) — falls back to /usr/local/var/numa
+///                    if a pre-v0.10.1 install already lives there.
+/// macOS root daemon: /usr/local/var/numa (Homebrew prefix)
 /// Windows: %APPDATA%\numa
 pub fn config_dir() -> std::path::PathBuf {
     #[cfg(windows)]
@@ -63,13 +125,15 @@ fn config_dir_unix() -> std::path::PathBuf {
     }
 
     // Running as root daemon (launchd/systemd) — use system-wide path
-    std::path::PathBuf::from("/usr/local/var/numa")
+    daemon_data_dir()
 }
 
 /// Default system-wide data directory for TLS certs. Overridable via
 /// `[server] data_dir = "..."` in numa.toml — this function only provides
 /// the fallback when the config doesn't set it.
-/// Unix: /usr/local/var/numa
+/// Linux: /var/lib/numa (FHS) — falls back to /usr/local/var/numa if a
+///        pre-v0.10.1 install already has data there.
+/// macOS: /usr/local/var/numa (Homebrew prefix)
 /// Windows: %PROGRAMDATA%\numa
 pub fn data_dir() -> std::path::PathBuf {
     #[cfg(windows)]
@@ -81,6 +145,131 @@ pub fn data_dir() -> std::path::PathBuf {
     }
     #[cfg(not(windows))]
     {
+        daemon_data_dir()
+    }
+}
+
+/// Resolve the system-wide data directory for the running platform.
+/// Honors backwards compatibility with pre-v0.10.1 installs that still
+/// have their CA cert + services.json under `/usr/local/var/numa`.
+#[cfg(not(windows))]
+fn daemon_data_dir() -> std::path::PathBuf {
+    #[cfg(target_os = "linux")]
+    {
+        std::path::PathBuf::from(resolve_linux_data_dir(
+            std::path::Path::new("/usr/local/var/numa").exists(),
+            std::path::Path::new("/var/lib/numa").exists(),
+        ))
+    }
+    #[cfg(target_os = "macos")]
+    {
+        // macOS uses the Homebrew prefix convention; no FHS migration needed.
         std::path::PathBuf::from("/usr/local/var/numa")
     }
 }
+
+/// Extracted as a pure function so the migration logic is unit-testable
+/// without touching the real filesystem.
+#[cfg(any(target_os = "linux", test))]
+fn resolve_linux_data_dir(legacy_exists: bool, fhs_exists: bool) -> &'static str {
+    if legacy_exists && !fhs_exists {
+        "/usr/local/var/numa"
+    } else {
+        "/var/lib/numa"
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn linux_data_dir_fresh_install_uses_fhs() {
+        assert_eq!(resolve_linux_data_dir(false, false), "/var/lib/numa");
+    }
+
+    #[test]
+    fn linux_data_dir_upgrading_install_keeps_legacy() {
+        // Migration must keep legacy so the user doesn't lose their CA on upgrade.
+        assert_eq!(resolve_linux_data_dir(true, false), "/usr/local/var/numa");
+    }
+
+    #[test]
+    fn linux_data_dir_after_migration_uses_fhs() {
+        assert_eq!(resolve_linux_data_dir(true, true), "/var/lib/numa");
+    }
+
+    #[test]
+    fn linux_data_dir_only_fhs_uses_fhs() {
+        assert_eq!(resolve_linux_data_dir(false, true), "/var/lib/numa");
+    }
+
+    #[cfg(not(windows))]
+    fn fhs() -> std::path::PathBuf {
+        std::path::PathBuf::from("/var/lib/numa")
+    }
+
+    #[cfg(not(windows))]
+    #[test]
+    fn suggested_config_path_prefers_home() {
+        assert_eq!(
+            resolve_suggested_config_path(Some("/home/alice"), fhs),
+            std::path::PathBuf::from("/home/alice/.config/numa/numa.toml"),
+        );
+    }
+
+    #[cfg(not(windows))]
+    #[test]
+    fn suggested_config_path_prefers_root_home_over_fhs() {
+        // Interactive root: HOME=/root is a real user context, not a daemon signal.
+        // Advisory must point where load_config will actually look — issue #81.
+        assert_eq!(
+            resolve_suggested_config_path(Some("/root"), fhs),
+            std::path::PathBuf::from("/root/.config/numa/numa.toml"),
+        );
+    }
+
+    #[cfg(not(windows))]
+    #[test]
+    fn suggested_config_path_falls_back_when_home_unset() {
+        assert_eq!(
+            resolve_suggested_config_path(None, fhs),
+            std::path::PathBuf::from("/var/lib/numa/numa.toml"),
+        );
+    }
+
+    #[cfg(not(windows))]
+    #[test]
+    fn suggested_config_path_falls_back_when_home_is_root() {
+        // systemd services sometimes have HOME=/ — don't treat that as a real home.
+        assert_eq!(
+            resolve_suggested_config_path(Some("/"), fhs),
+            std::path::PathBuf::from("/var/lib/numa/numa.toml"),
+        );
+    }
+
+    #[cfg(not(windows))]
+    #[test]
+    fn suggested_config_path_falls_back_when_home_is_empty() {
+        assert_eq!(
+            resolve_suggested_config_path(Some(""), fhs),
+            std::path::PathBuf::from("/var/lib/numa/numa.toml"),
+        );
+    }
+
+    #[cfg(not(windows))]
+    #[test]
+    fn suggested_config_path_skips_fallback_when_home_valid() {
+        // Happy path shouldn't probe the filesystem via config_dir().
+        let called = std::cell::Cell::new(false);
+        let fallback = || {
+            called.set(true);
+            std::path::PathBuf::from("/should/not/be/used")
+        };
+        let _ = resolve_suggested_config_path(Some("/home/alice"), fallback);
+        assert!(
+            !called.get(),
+            "fallback must not be invoked when HOME is valid"
+        );
+    }
+}

src/main.rs

@@ -11,11 +11,11 @@ use numa::buffer::BytePacketBuffer;
 use numa::cache::DnsCache;
 use numa::config::{build_zone_map, load_config, ConfigLoad};
 use numa::ctx::{handle_query, ServerCtx};
-use numa::forward::Upstream;
+use numa::forward::{parse_upstream, Upstream, UpstreamPool};
 use numa::override_store::OverrideStore;
 use numa::query_log::QueryLog;
 use numa::service_store::ServiceStore;
-use numa::stats::ServerStats;
+use numa::stats::{ServerStats, Transport};
 use numa::system_dns::{
     discover_system_dns, install_service, restart_service, service_status, uninstall_service,
 };
@@ -54,6 +54,9 @@ async fn main() -> numa::Result<()> {
                 }
             };
         }
+        "setup-phone" => {
+            return numa::setup_phone::run().await.map_err(|e| e.into());
+        }
         "lan" => {
             let sub = std::env::args().nth(2).unwrap_or_default();
             let config_path = std::env::args()
@@ -85,12 +88,27 @@ async fn main() -> numa::Result<()> {
             eprintln!("  service status  Check if the service is running");
             eprintln!("  lan on          Enable LAN service discovery (mDNS)");
             eprintln!("  lan off         Disable LAN service discovery");
+            eprintln!("  setup-phone     Generate a QR code to install Numa DoT on a phone");
             eprintln!("  help            Show this help");
             eprintln!();
             eprintln!("Config path defaults to numa.toml");
             return Ok(());
         }
-        _ => {}
+        _ => {
+            if !arg1.is_empty()
+                && arg1 != "run"
+                && !arg1.contains('/')
+                && !arg1.contains('\\')
+                && !arg1.ends_with(".toml")
+            {
+                eprintln!(
+                    "\x1b[1;38;2;192;98;58mNuma\x1b[0m — unknown command: \x1b[1m{}\x1b[0m\n",
+                    arg1
+                );
+                eprintln!("Run \x1b[1mnuma help\x1b[0m for a list of commands.");
+                std::process::exit(1);
+            }
+        }
     }
 
     let config_path = if arg1.is_empty() || arg1 == "run" {
@@ -111,18 +129,18 @@ async fn main() -> numa::Result<()> {
 
     let root_hints = numa::recursive::parse_root_hints(&config.upstream.root_hints);
 
-    let (resolved_mode, upstream_auto, upstream, upstream_label) = match config.upstream.mode {
+    let recursive_pool = || {
+        let dummy = UpstreamPool::new(vec![Upstream::Udp("0.0.0.0:0".parse().unwrap())], vec![]);
+        (dummy, "recursive (root hints)".to_string())
+    };
+
+    let (resolved_mode, upstream_auto, pool, upstream_label) = match config.upstream.mode {
         numa::config::UpstreamMode::Auto => {
             info!("auto mode: probing recursive resolution...");
             if numa::recursive::probe_recursive(&root_hints).await {
                 info!("recursive probe succeeded — self-sovereign mode");
-                let dummy = Upstream::Udp("0.0.0.0:0".parse().unwrap());
+                let (pool, label) = recursive_pool();
-                (
+                (numa::config::UpstreamMode::Recursive, false, pool, label)
-                    numa::config::UpstreamMode::Recursive,
-                    false,
-                    dummy,
-                    "recursive (root hints)".to_string(),
-                )
             } else {
                 log::warn!("recursive probe failed — falling back to Quad9 DoH");
                 let client = reqwest::Client::builder()
@@ -131,55 +149,45 @@ async fn main() -> numa::Result<()> {
                     .unwrap_or_default();
                 let url = DOH_FALLBACK.to_string();
                 let label = url.clone();
-                (
+                let pool = UpstreamPool::new(vec![Upstream::Doh { url, client }], vec![]);
-                    numa::config::UpstreamMode::Forward,
+                (numa::config::UpstreamMode::Forward, false, pool, label)
-                    false,
-                    Upstream::Doh { url, client },
-                    label,
-                )
             }
         }
         numa::config::UpstreamMode::Recursive => {
-            let dummy = Upstream::Udp("0.0.0.0:0".parse().unwrap());
+            let (pool, label) = recursive_pool();
-            (
+            (numa::config::UpstreamMode::Recursive, false, pool, label)
-                numa::config::UpstreamMode::Recursive,
-                false,
-                dummy,
-                "recursive (root hints)".to_string(),
-            )
         }
         numa::config::UpstreamMode::Forward => {
-            let upstream_addr = if config.upstream.address.is_empty() {
+            let addrs = if config.upstream.address.is_empty() {
-                system_dns
+                let detected = system_dns
                     .default_upstream
                     .or_else(numa::system_dns::detect_dhcp_dns)
                     .unwrap_or_else(|| {
                         info!("could not detect system DNS, falling back to Quad9 DoH");
                         DOH_FALLBACK.to_string()
-                    })
+                    });
+                vec![detected]
             } else {
                 config.upstream.address.clone()
             };
 
-            let upstream: Upstream = if upstream_addr.starts_with("https://") {
+            let primary: Vec<Upstream> = addrs
-                let client = reqwest::Client::builder()
+                .iter()
-                    .use_rustls_tls()
+                .map(|s| parse_upstream(s, config.upstream.port))
-                    .build()
+                .collect::<numa::Result<Vec<_>>>()?;
-                    .unwrap_or_default();
+            let fallback: Vec<Upstream> = config
-                Upstream::Doh {
+                .upstream
-                    url: upstream_addr,
+                .fallback
-                    client,
+                .iter()
-                }
+                .map(|s| parse_upstream(s, config.upstream.port))
-            } else {
+                .collect::<numa::Result<Vec<_>>>()?;
-                let addr: SocketAddr =
+
-                    format!("{}:{}", upstream_addr, config.upstream.port).parse()?;
+            let pool = UpstreamPool::new(primary, fallback);
-                Upstream::Udp(addr)
+            let label = pool.label();
-            };
-            let label = upstream.to_string();
             (
                 numa::config::UpstreamMode::Forward,
                 config.upstream.address.is_empty(),
-                upstream,
+                pool,
                 label,
             )
         }
@@ -202,7 +210,13 @@ async fn main() -> numa::Result<()> {
     }
     service_store.load_persisted();
 
-    let forwarding_rules = system_dns.forwarding_rules;
+    for fwd in &config.forwarding {
+        for suffix in &fwd.suffix {
+            info!("forwarding .{} to {} (config rule)", suffix, fwd.upstream);
+        }
+    }
+    let forwarding_rules =
+        numa::config::merge_forwarding_rules(&config.forwarding, system_dns.forwarding_rules)?;
 
     // Resolve data_dir from config, falling back to the platform default.
     // Used for TLS CA storage below and stored on ServerCtx for runtime use.
@@ -223,7 +237,11 @@ async fn main() -> numa::Result<()> {
         ) {
             Ok(tls_config) => Some(ArcSwap::from(tls_config)),
             Err(e) => {
-                log::warn!("TLS setup failed, HTTPS proxy disabled: {}", e);
+                if let Some(advisory) = numa::tls::try_data_dir_advisory(&e, &resolved_data_dir) {
+                    eprint!("{}", advisory);
+                } else {
+                    log::warn!("TLS setup failed, HTTPS proxy disabled: {}", e);
+                }
                 None
             }
         }
@@ -231,14 +249,43 @@ async fn main() -> numa::Result<()> {
         None
     };
 
+    let doh_enabled = initial_tls.is_some();
+    let health_meta = numa::health::HealthMeta::build(
+        &resolved_data_dir,
+        config.dot.enabled,
+        config.dot.port,
+        config.mobile.port,
+        config.dnssec.enabled,
+        resolved_mode == numa::config::UpstreamMode::Recursive,
+        config.lan.enabled,
+        config.blocking.enabled,
+        doh_enabled,
+    );
+
+    let ca_pem = std::fs::read_to_string(resolved_data_dir.join("ca.pem")).ok();
+
+    let socket = match UdpSocket::bind(&config.server.bind_addr).await {
+        Ok(s) => s,
+        Err(e) => {
+            if let Some(advisory) =
+                numa::system_dns::try_port53_advisory(&config.server.bind_addr, &e)
+            {
+                eprint!("{}", advisory);
+                std::process::exit(1);
+            }
+            return Err(e.into());
+        }
+    };
+
     let ctx = Arc::new(ServerCtx {
-        socket: UdpSocket::bind(&config.server.bind_addr).await?,
+        socket,
         zone_map: build_zone_map(&config.zones)?,
         cache: RwLock::new(DnsCache::new(
             config.cache.max_entries,
             config.cache.min_ttl,
             config.cache.max_ttl,
         )),
+        refreshing: Mutex::new(std::collections::HashSet::new()),
         stats: Mutex::new(ServerStats::new()),
         overrides: RwLock::new(OverrideStore::new()),
         blocklist: RwLock::new(blocklist),
@@ -246,11 +293,12 @@ async fn main() -> numa::Result<()> {
         services: Mutex::new(service_store),
         lan_peers: Mutex::new(numa::lan::PeerStore::new(config.lan.peer_timeout_secs)),
         forwarding_rules,
-        upstream: Mutex::new(upstream),
+        upstream_pool: Mutex::new(pool),
         upstream_auto,
         upstream_port: config.upstream.port,
         lan_ip: Mutex::new(numa::lan::detect_lan_ip().unwrap_or(std::net::Ipv4Addr::LOCALHOST)),
         timeout: Duration::from_millis(config.upstream.timeout_ms),
+        hedge_delay: Duration::from_millis(config.upstream.hedge_ms),
         proxy_tld_suffix: if config.proxy.tld.is_empty() {
             String::new()
         } else {
@@ -269,6 +317,10 @@ async fn main() -> numa::Result<()> {
         inflight: std::sync::Mutex::new(std::collections::HashMap::new()),
         dnssec_enabled: config.dnssec.enabled,
         dnssec_strict: config.dnssec.strict,
+        health_meta,
+        ca_pem,
+        mobile_enabled: config.mobile.enabled,
+        mobile_port: config.mobile.port,
     });
 
     let zone_count: usize = ctx.zone_map.values().map(|m| m.len()).sum();
@@ -360,6 +412,9 @@ async fn main() -> numa::Result<()> {
         g,
         &format!("max {} entries", config.cache.max_entries),
     );
+    if !config.cache.warm.is_empty() {
+        row("Warm", g, &format!("{} domains", config.cache.warm.len()));
+    }
     row(
         "Blocking",
         g,
@@ -386,6 +441,13 @@ async fn main() -> numa::Result<()> {
     if config.dot.enabled {
         row("DoT", g, &format!("tls://:{}", config.dot.port));
     }
+    if doh_enabled {
+        row(
+            "DoH",
+            g,
+            &format!("https://:{}/dns-query", config.proxy.tls_port),
+        );
+    }
     if config.lan.enabled {
         row("LAN", g, "mDNS (_numa._tcp.local)");
     }
@@ -442,6 +504,23 @@ async fn main() -> numa::Result<()> {
         });
     }
 
+    // Spawn cache warming for user-configured domains
+    if !config.cache.warm.is_empty() {
+        let warm_ctx = Arc::clone(&ctx);
+        let warm_domains = config.cache.warm.clone();
+        tokio::spawn(async move {
+            cache_warm_loop(warm_ctx, warm_domains).await;
+        });
+    }
+
+    // Spawn DoH connection keepalive — prevents idle TLS teardown
+    {
+        let keepalive_ctx = Arc::clone(&ctx);
+        tokio::spawn(async move {
+            doh_keepalive_loop(keepalive_ctx).await;
+        });
+    }
+
     // Spawn HTTP API server
     let api_ctx = Arc::clone(&ctx);
     let api_addr: SocketAddr = format!("{}:{}", config.server.api_bind_addr, api_port).parse()?;
@@ -452,6 +531,21 @@ async fn main() -> numa::Result<()> {
         axum::serve(listener, app).await.unwrap();
     });
 
+    // Spawn Mobile API listener (read-only subset for iOS/Android companion
+    // apps, LAN-bound by default so phones can reach it). Only idempotent
+    // GETs; no state-mutating routes are exposed here regardless of
+    // the main API's bind address.
+    if config.mobile.enabled {
+        let mobile_ctx = Arc::clone(&ctx);
+        let mobile_bind = config.mobile.bind_addr.clone();
+        let mobile_port = config.mobile.port;
+        tokio::spawn(async move {
+            if let Err(e) = numa::mobile_api::start(mobile_ctx, mobile_bind, mobile_port).await {
+                log::warn!("Mobile API listener failed: {}", e);
+            }
+        });
+    }
+
     let proxy_bind: std::net::Ipv4Addr = config
         .proxy
         .bind_addr
@@ -506,7 +600,7 @@ async fn main() -> numa::Result<()> {
     #[allow(clippy::infinite_loop)]
     loop {
         let mut buffer = BytePacketBuffer::new();
-        let (_, src_addr) = match ctx.socket.recv_from(&mut buffer.buf).await {
+        let (len, src_addr) = match ctx.socket.recv_from(&mut buffer.buf).await {
             Ok(r) => r,
             Err(e) if e.kind() == std::io::ErrorKind::ConnectionReset => {
                 // Windows delivers ICMP port-unreachable as ConnectionReset on UDP sockets
@@ -514,10 +608,9 @@ async fn main() -> numa::Result<()> {
             }
             Err(e) => return Err(e.into()),
         };
-
         let ctx = Arc::clone(&ctx);
         tokio::spawn(async move {
-            if let Err(e) = handle_query(buffer, src_addr, &ctx).await {
+            if let Err(e) = handle_query(buffer, len, src_addr, &ctx, Transport::Udp).await {
                 error!("{} | HANDLER ERROR | {}", src_addr, e);
             }
         });
@@ -546,27 +639,17 @@ async fn network_watch_loop(ctx: Arc<numa::ctx::ServerCtx>) {
             }
         }
 
-        // Re-detect upstream every 30s or on LAN IP change (UDP only —
+        // Re-detect upstream every 30s or on LAN IP change (auto-detect only)
-        // DoH upstreams are explicitly configured via URL, not auto-detected)
+        if ctx.upstream_auto && (changed || tick.is_multiple_of(6)) {
-        if ctx.upstream_auto
-            && matches!(*ctx.upstream.lock().unwrap(), Upstream::Udp(_))
-            && (changed || tick.is_multiple_of(6))
-        {
             let dns_info = numa::system_dns::discover_system_dns();
             let new_addr = dns_info
                 .default_upstream
                 .or_else(numa::system_dns::detect_dhcp_dns)
                 .unwrap_or_else(|| QUAD9_IP.to_string());
-            if let Ok(new_sock) =
+            let mut pool = ctx.upstream_pool.lock().unwrap();
-                format!("{}:{}", new_addr, ctx.upstream_port).parse::<SocketAddr>()
+            if pool.maybe_update_primary(&new_addr, ctx.upstream_port) {
-            {
+                info!("upstream changed → {}", pool.label());
-                let new_upstream = Upstream::Udp(new_sock);
+                changed = true;
-                let mut upstream = ctx.upstream.lock().unwrap();
-                if *upstream != new_upstream {
-                    info!("upstream changed: {} → {}", upstream, new_upstream);
-                    *upstream = new_upstream;
-                    changed = true;
-                }
             }
         }
 
@@ -673,3 +756,45 @@ async fn load_blocklists(ctx: &ServerCtx, lists: &[String]) {
         downloaded.len()
     );
 }
+
+async fn warm_domain(ctx: &ServerCtx, domain: &str) {
+    for qtype in [
+        numa::question::QueryType::A,
+        numa::question::QueryType::AAAA,
+    ] {
+        numa::ctx::refresh_entry(ctx, domain, qtype).await;
+    }
+}
+
+async fn doh_keepalive_loop(ctx: Arc<ServerCtx>) {
+    let mut interval = tokio::time::interval(Duration::from_secs(25));
+    interval.tick().await; // skip first immediate tick
+    loop {
+        interval.tick().await;
+        let pool = ctx.upstream_pool.lock().unwrap().clone();
+        if let Some(upstream) = pool.preferred() {
+            numa::forward::keepalive_doh(upstream).await;
+        }
+    }
+}
+
+async fn cache_warm_loop(ctx: Arc<ServerCtx>, domains: Vec<String>) {
+    tokio::time::sleep(Duration::from_secs(2)).await;
+
+    for domain in &domains {
+        warm_domain(&ctx, domain).await;
+    }
+    info!("cache warm: {} domains resolved at startup", domains.len());
+
+    let mut interval = tokio::time::interval(Duration::from_secs(30));
+    interval.tick().await;
+    loop {
+        interval.tick().await;
+        for domain in &domains {
+            let refresh = ctx.cache.read().unwrap().needs_warm(domain);
+            if refresh {
+                warm_domain(&ctx, domain).await;
+            }
+        }
+    }
+}

src/mobile_api.rs

@@ -0,0 +1,107 @@
+//! Mobile API — persistent HTTP listener for iOS/Android companion apps.
+//!
+//! Read-only subset of Numa's HTTP surface served on a separate port
+//! (default 8765) bound to the LAN. Unlike the main API on port 5380
+//! (which defaults to `127.0.0.1` and serves mutating routes like
+//! `DELETE /services/{name}` or `PUT /blocking/toggle`), this listener
+//! is safe to expose on the LAN because every route is idempotent and
+//! read-only.
+//!
+//! Routes (all GET):
+//!
+//! - `/health` — enriched status + metadata, shares the handler with the
+//!   main API via `crate::api::health`
+//! - `/ca.pem` — Numa local CA in PEM form, shares the handler with the
+//!   main API via `crate::api::serve_ca`
+//! - `/mobileconfig` — combined CA + DNS settings profile (Full mode)
+//! - `/ca.mobileconfig` — CA-only trust profile (no DNS override)
+//!
+//! The mobile API does NOT include the mutating routes (overrides, cache
+//! flush, blocking toggle, service CRUD, etc.). Even if a user sets
+//! `api_bind_addr` to `0.0.0.0` for the main API, those routes stay on
+//! port 5380; the mobile API on port 8765 never serves them. This is the
+//! primary security boundary: anything exposed to the LAN is read-only.
+
+use std::net::Ipv4Addr;
+use std::sync::Arc;
+
+use axum::extract::State;
+use axum::http::{header, StatusCode};
+use axum::response::IntoResponse;
+use axum::routing::get;
+use axum::Router;
+use log::info;
+
+use crate::ctx::ServerCtx;
+use crate::mobileconfig::{build_mobileconfig, ProfileMode};
+
+/// Content-Disposition for the full CA + DNS profile download.
+const FULL_PROFILE_DISPOSITION: &str = "attachment; filename=\"numa.mobileconfig\"";
+
+/// Content-Disposition for the CA-only profile download.
+const CA_ONLY_PROFILE_DISPOSITION: &str = "attachment; filename=\"numa-ca.mobileconfig\"";
+
+/// Build the axum router for the mobile API.
+///
+/// Shares handler functions with the main API where possible (`health`,
+/// `serve_ca`) so the response shapes are identical across both ports.
+pub fn router(ctx: Arc<ServerCtx>) -> Router {
+    Router::new()
+        .route("/health", get(crate::api::health))
+        .route("/ca.pem", get(crate::api::serve_ca))
+        .route("/mobileconfig", get(serve_full_mobileconfig))
+        .route("/ca.mobileconfig", get(serve_ca_only_mobileconfig))
+        .with_state(ctx)
+}
+
+/// Start the mobile API listener on `bind_addr:port`. Runs until the
+/// caller cancels the spawned task. Logs the URL on successful bind.
+pub async fn start(ctx: Arc<ServerCtx>, bind_addr: String, port: u16) -> crate::Result<()> {
+    let addr: std::net::SocketAddr = format!("{}:{}", bind_addr, port).parse()?;
+    let listener = tokio::net::TcpListener::bind(addr).await?;
+
+    info!("Mobile API listening on http://{}", addr);
+
+    let app = router(ctx);
+    axum::serve(listener, app).await?;
+
+    Ok(())
+}
+
+/// Serve the full mobileconfig profile (CA + DNS settings), with the
+/// DNS payload pointing at the current LAN IP. Each request reads the
+/// fresh LAN IP from `ctx.lan_ip` so the profile always reflects the
+/// laptop's current network state.
+async fn serve_full_mobileconfig(
+    State(ctx): State<Arc<ServerCtx>>,
+) -> Result<impl IntoResponse, StatusCode> {
+    let ca_pem = ctx.ca_pem.as_deref().ok_or(StatusCode::NOT_FOUND)?;
+    let lan_ip: Ipv4Addr = *ctx.lan_ip.lock().unwrap();
+    let profile = build_mobileconfig(ProfileMode::Full { lan_ip }, ca_pem);
+    Ok(profile_response(profile, FULL_PROFILE_DISPOSITION))
+}
+
+/// Serve the CA-only mobileconfig profile. Trusts the Numa local CA but
+/// does NOT change the device's DNS settings. Used by the iOS companion
+/// app's DoT mode, where the app configures DNS via `NEDNSSettingsManager`
+/// and only needs the system trust store to accept Numa's self-signed cert.
+async fn serve_ca_only_mobileconfig(
+    State(ctx): State<Arc<ServerCtx>>,
+) -> Result<impl IntoResponse, StatusCode> {
+    let ca_pem = ctx.ca_pem.as_deref().ok_or(StatusCode::NOT_FOUND)?;
+    let profile = build_mobileconfig(ProfileMode::CaOnly, ca_pem);
+    Ok(profile_response(profile, CA_ONLY_PROFILE_DISPOSITION))
+}
+
+/// Shared response constructor for both mobileconfig variants.
+/// Identical headers; only the Content-Disposition filename differs.
+fn profile_response(profile: String, disposition: &'static str) -> impl IntoResponse {
+    (
+        [
+            (header::CONTENT_TYPE, "application/x-apple-aspen-config"),
+            (header::CONTENT_DISPOSITION, disposition),
+            (header::CACHE_CONTROL, "no-store"),
+        ],
+        profile,
+    )
+}

src/mobileconfig.rs

@@ -0,0 +1,305 @@
+//! Apple `.mobileconfig` profile generator.
+//!
+//! Builds iOS Configuration Profiles that Numa serves to phones for one-tap
+//! CA trust and DNS-over-TLS setup. The plist structure is hand-rendered
+//! via `format!` — no plist crate dependency, deterministic output, small
+//! binary footprint.
+//!
+//! Two modes:
+//!
+//! - [`ProfileMode::Full`]: CA trust payload + DNS settings payload pointing
+//!   at a specific LAN IP over DoT. This is what `numa setup-phone` has
+//!   always produced — the user scans a QR, installs this profile, and the
+//!   phone is configured for DoT through Numa in a single step (after the
+//!   iOS Certificate Trust Settings toggle, which is a separate system
+//!   gate we can't bypass).
+//!
+//! - [`ProfileMode::CaOnly`]: CA trust payload only, no DNS settings. Used
+//!   by the future iOS companion app flow where `NEDNSSettingsManager`
+//!   configures DNS programmatically and we only need the system trust
+//!   store to accept Numa's DoT cert. Installing this profile does NOT
+//!   change the user's DNS at all.
+//!
+//! Payload identifiers and UUIDs are fixed (not randomized) so iOS replaces
+//! the existing profile on re-install rather than accumulating duplicates.
+//! The `Full` and `CaOnly` profiles have distinct top-level UUIDs so they
+//! can coexist as separate installed profiles, but they share the same CA
+//! payload UUID since the CA itself is the same trust anchor in both.
+
+use std::net::Ipv4Addr;
+
+/// Top-level UUID and PayloadIdentifier for the full profile (CA + DNS).
+/// Changing this breaks in-place replacement on existing iOS installs.
+const FULL_PROFILE_UUID: &str = "F1E2D3C4-B5A6-7890-1234-567890ABCDEF";
+const FULL_PROFILE_ID: &str = "com.numa.dns.profile";
+
+/// Top-level UUID and PayloadIdentifier for the CA-only profile.
+/// Distinct from `FULL_PROFILE_UUID` so a user can install one, the other,
+/// or both without the latest install silently replacing a different mode.
+const CA_ONLY_PROFILE_UUID: &str = "F2E3D4C5-B6A7-8901-2345-67890ABCDEF0";
+const CA_ONLY_PROFILE_ID: &str = "com.numa.dns.ca.profile";
+
+/// CA trust payload UUID. Same in both modes — iOS will see "the same CA
+/// trust anchor" regardless of which wrapping profile contains it.
+const CA_PAYLOAD_UUID: &str = "B2C3D4E5-F6A7-8901-BCDE-F12345678901";
+const CA_PAYLOAD_ID: &str = "com.numa.dns.ca";
+
+/// DNS settings payload UUID (Full mode only).
+const DNS_PAYLOAD_UUID: &str = "A1B2C3D4-E5F6-7890-ABCD-EF1234567890";
+const DNS_PAYLOAD_ID: &str = "com.numa.dns.dot";
+
+/// Profile mode determines which payloads are included in the generated
+/// `.mobileconfig`.
+#[derive(Debug, Clone)]
+pub enum ProfileMode {
+    /// Full profile: CA trust anchor + managed DNS settings payload
+    /// pointing at the given LAN IP over DoT. This is what the classic
+    /// `numa setup-phone` QR flow serves.
+    Full { lan_ip: Ipv4Addr },
+
+    /// CA-only profile: just the trust anchor, no DNS settings. For use
+    /// with the iOS companion app which manages DNS programmatically via
+    /// `NEDNSSettingsManager` and only needs the system trust store to
+    /// accept Numa's self-signed DoT cert.
+    CaOnly,
+}
+
+/// Build a full `.mobileconfig` profile as an XML plist string.
+pub fn build_mobileconfig(mode: ProfileMode, ca_pem: &str) -> String {
+    let ca_payload = build_ca_payload(ca_pem);
+
+    match mode {
+        ProfileMode::Full { lan_ip } => {
+            let dns_payload = build_dns_payload(lan_ip);
+            let payloads = format!("{}\n{}", ca_payload, dns_payload);
+            let description = format!(
+                "Trusts the Numa local CA and routes DNS queries to Numa over DoT on your local network ({lan_ip})"
+            );
+            wrap_plist(
+                &payloads,
+                FULL_PROFILE_UUID,
+                FULL_PROFILE_ID,
+                &description,
+                "Numa DNS",
+            )
+        }
+        ProfileMode::CaOnly => wrap_plist(
+            &ca_payload,
+            CA_ONLY_PROFILE_UUID,
+            CA_ONLY_PROFILE_ID,
+            "Trusts the Numa local Certificate Authority. Does not change your DNS settings.",
+            "Numa CA",
+        ),
+    }
+}
+
+/// Strip the PEM header/footer and newlines from a CA cert, leaving raw
+/// base64 for embedding in a plist `<data>` block.
+fn pem_to_base64(pem: &str) -> String {
+    pem.lines()
+        .filter(|line| !line.starts_with("-----"))
+        .collect::<String>()
+}
+
+/// Wrap the base64 CA cert at 52 chars per line for plist readability
+/// (matches Apple convention in hand-written profiles).
+fn chunk_base64(base64: &str) -> String {
+    base64
+        .chars()
+        .collect::<Vec<_>>()
+        .chunks(52)
+        .map(|chunk| format!("\t\t\t{}", chunk.iter().collect::<String>()))
+        .collect::<Vec<_>>()
+        .join("\n")
+}
+
+/// Render the `com.apple.security.root` payload dict containing the CA cert.
+fn build_ca_payload(ca_pem: &str) -> String {
+    let ca_wrapped = chunk_base64(&pem_to_base64(ca_pem));
+    format!(
+        r#"		<dict>
+			<key>PayloadCertificateFileName</key>
+			<string>numa-ca.pem</string>
+			<key>PayloadContent</key>
+			<data>
+{ca}
+			</data>
+			<key>PayloadDescription</key>
+			<string>Numa local Certificate Authority — required for DoT trust</string>
+			<key>PayloadDisplayName</key>
+			<string>Numa Local CA</string>
+			<key>PayloadIdentifier</key>
+			<string>{ca_id}</string>
+			<key>PayloadType</key>
+			<string>com.apple.security.root</string>
+			<key>PayloadUUID</key>
+			<string>{ca_uuid}</string>
+			<key>PayloadVersion</key>
+			<integer>1</integer>
+		</dict>"#,
+        ca = ca_wrapped,
+        ca_id = CA_PAYLOAD_ID,
+        ca_uuid = CA_PAYLOAD_UUID,
+    )
+}
+
+/// Render the `com.apple.dnsSettings.managed` payload dict for Full mode.
+fn build_dns_payload(lan_ip: Ipv4Addr) -> String {
+    format!(
+        r#"		<dict>
+			<key>DNSSettings</key>
+			<dict>
+				<key>DNSProtocol</key>
+				<string>TLS</string>
+				<key>ServerAddresses</key>
+				<array>
+					<string>{ip}</string>
+				</array>
+				<key>ServerName</key>
+				<string>numa.numa</string>
+			</dict>
+			<key>OnDemandRules</key>
+			<array>
+				<dict>
+					<key>Action</key>
+					<string>Connect</string>
+					<key>InterfaceTypeMatch</key>
+					<string>WiFi</string>
+				</dict>
+				<dict>
+					<key>Action</key>
+					<string>Disconnect</string>
+				</dict>
+			</array>
+			<key>PayloadDescription</key>
+			<string>Routes DNS queries through Numa over DoT when on Wi-Fi</string>
+			<key>PayloadDisplayName</key>
+			<string>Numa DNS-over-TLS</string>
+			<key>PayloadIdentifier</key>
+			<string>{dns_id}</string>
+			<key>PayloadType</key>
+			<string>com.apple.dnsSettings.managed</string>
+			<key>PayloadUUID</key>
+			<string>{dns_uuid}</string>
+			<key>PayloadVersion</key>
+			<integer>1</integer>
+		</dict>"#,
+        ip = lan_ip,
+        dns_id = DNS_PAYLOAD_ID,
+        dns_uuid = DNS_PAYLOAD_UUID,
+    )
+}
+
+/// Wrap one or more payload dicts in the top-level plist structure
+/// with Configuration type, PayloadContent array, and profile metadata.
+fn wrap_plist(
+    payloads: &str,
+    top_uuid: &str,
+    top_id: &str,
+    description: &str,
+    display_name: &str,
+) -> String {
+    format!(
+        r#"<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
+<plist version="1.0">
+<dict>
+	<key>PayloadContent</key>
+	<array>
+{payloads}
+	</array>
+	<key>PayloadDescription</key>
+	<string>{description}</string>
+	<key>PayloadDisplayName</key>
+	<string>{display_name}</string>
+	<key>PayloadIdentifier</key>
+	<string>{top_id}</string>
+	<key>PayloadRemovalDisallowed</key>
+	<false/>
+	<key>PayloadType</key>
+	<string>Configuration</string>
+	<key>PayloadUUID</key>
+	<string>{top_uuid}</string>
+	<key>PayloadVersion</key>
+	<integer>1</integer>
+</dict>
+</plist>
+"#,
+        payloads = payloads,
+        description = description,
+        display_name = display_name,
+        top_id = top_id,
+        top_uuid = top_uuid,
+    )
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    const SAMPLE_PEM: &str =
+        "-----BEGIN CERTIFICATE-----\nMIIBkDCCATagAwIBAgIUTEST\n-----END CERTIFICATE-----\n";
+
+    #[test]
+    fn pem_to_base64_strips_headers() {
+        let pem = "-----BEGIN CERTIFICATE-----\nABCDEF\nGHIJKL\n-----END CERTIFICATE-----\n";
+        assert_eq!(pem_to_base64(pem), "ABCDEFGHIJKL");
+    }
+
+    #[test]
+    fn full_profile_contains_ip_and_ca() {
+        let config = build_mobileconfig(
+            ProfileMode::Full {
+                lan_ip: Ipv4Addr::new(192, 168, 1, 100),
+            },
+            SAMPLE_PEM,
+        );
+        assert!(config.contains("192.168.1.100"));
+        assert!(config.contains("MIIBkDCCATagAwIBAgIUTEST"));
+        assert!(config.contains("com.apple.security.root"));
+        assert!(config.contains("com.apple.dnsSettings.managed"));
+        assert!(config.contains("DNSProtocol"));
+        assert!(config.contains(FULL_PROFILE_UUID));
+        assert!(config.contains(FULL_PROFILE_ID));
+    }
+
+    #[test]
+    fn ca_only_profile_contains_ca_but_not_dns() {
+        let config = build_mobileconfig(ProfileMode::CaOnly, SAMPLE_PEM);
+        assert!(config.contains("MIIBkDCCATagAwIBAgIUTEST"));
+        assert!(config.contains("com.apple.security.root"));
+        assert!(!config.contains("com.apple.dnsSettings.managed"));
+        assert!(!config.contains("DNSProtocol"));
+        assert!(!config.contains("ServerAddresses"));
+        assert!(config.contains(CA_ONLY_PROFILE_UUID));
+        assert!(config.contains(CA_ONLY_PROFILE_ID));
+    }
+
+    #[test]
+    fn full_and_ca_only_have_distinct_top_uuids() {
+        let full = build_mobileconfig(
+            ProfileMode::Full {
+                lan_ip: Ipv4Addr::new(10, 0, 0, 1),
+            },
+            SAMPLE_PEM,
+        );
+        let ca_only = build_mobileconfig(ProfileMode::CaOnly, SAMPLE_PEM);
+        assert!(full.contains(FULL_PROFILE_UUID));
+        assert!(!full.contains(CA_ONLY_PROFILE_UUID));
+        assert!(ca_only.contains(CA_ONLY_PROFILE_UUID));
+        assert!(!ca_only.contains(FULL_PROFILE_UUID));
+    }
+
+    #[test]
+    fn both_modes_share_ca_payload_uuid() {
+        let full = build_mobileconfig(
+            ProfileMode::Full {
+                lan_ip: Ipv4Addr::new(10, 0, 0, 1),
+            },
+            SAMPLE_PEM,
+        );
+        let ca_only = build_mobileconfig(ProfileMode::CaOnly, SAMPLE_PEM);
+        assert!(full.contains(CA_PAYLOAD_UUID));
+        assert!(ca_only.contains(CA_PAYLOAD_UUID));
+    }
+}

src/proxy.rs

@@ -4,7 +4,7 @@ use std::sync::Arc;
 use axum::body::Body;
 use axum::extract::{Request, State};
 use axum::response::IntoResponse;
-use axum::routing::any;
+use axum::routing::{any, post};
 use axum::Router;
 use http_body_util::BodyExt;
 use hyper::StatusCode;
@@ -18,6 +18,14 @@ use crate::ctx::ServerCtx;
 
 type HttpClient = Client<hyper_util::client::legacy::connect::HttpConnector, Body>;
 
+/// State passed to the DoH handler. Includes the remote address so
+/// `resolve_query` can log the client IP.
+#[derive(Clone)]
+pub struct DohState {
+    pub ctx: Arc<ServerCtx>,
+    pub remote_addr: Option<std::net::SocketAddr>,
+}
+
 #[derive(Clone)]
 struct ProxyState {
     ctx: Arc<ServerCtx>,
@@ -74,9 +82,17 @@ pub async fn start_proxy_tls(ctx: Arc<ServerCtx>, port: u16, bind_addr: Ipv4Addr
 
     // Hold a separate Arc so we can access tls_config after ctx moves into ProxyState
     let tls_holder = Arc::clone(&ctx);
-    let state = ProxyState { ctx, client };
+    let proxy_state = ProxyState {
+        ctx: Arc::clone(&ctx),
+        client,
+    };
 
-    let app = Router::new().fallback(any(proxy_handler)).with_state(state);
+    // DoH route (RFC 8484) served only on the TLS listener.
+    // DohState.remote_addr is set per-connection below.
+    let doh_state = DohState {
+        ctx,
+        remote_addr: None,
+    };
 
     loop {
         let (tcp_stream, remote_addr) = match listener.accept().await {
@@ -91,7 +107,17 @@ pub async fn start_proxy_tls(ctx: Arc<ServerCtx>, port: u16, bind_addr: Ipv4Addr
         // unwrap safe: guarded by is_none() check above
         let acceptor =
             TlsAcceptor::from(Arc::clone(&*tls_holder.tls_config.as_ref().unwrap().load()));
-        let app = app.clone();
+
+        let mut conn_doh_state = doh_state.clone();
+        conn_doh_state.remote_addr = Some(remote_addr);
+
+        let app = Router::new()
+            .route(
+                "/dns-query",
+                post(crate::doh::doh_post).with_state(conn_doh_state),
+            )
+            .fallback(any(proxy_handler))
+            .with_state(proxy_state.clone());
 
         tokio::spawn(async move {
             let tls_stream = match acceptor.accept(tcp_stream).await {
@@ -232,7 +258,7 @@ pre .str {{ color: #d48a5a }}
     )
 }
 
-fn extract_host(req: &Request) -> Option<String> {
+pub fn extract_host(req: &Request) -> Option<String> {
     req.headers()
         .get(hyper::header::HOST)
         .and_then(|v| v.to_str().ok())

src/query_log.rs

@@ -5,7 +5,7 @@ use std::time::SystemTime;
 use crate::cache::DnssecStatus;
 use crate::header::ResultCode;
 use crate::question::QueryType;
-use crate::stats::QueryPath;
+use crate::stats::{QueryPath, Transport};
 
 pub struct QueryLogEntry {
     pub timestamp: SystemTime,
@@ -13,6 +13,7 @@ pub struct QueryLogEntry {
     pub domain: String,
     pub query_type: QueryType,
     pub path: QueryPath,
+    pub transport: Transport,
     pub rescode: ResultCode,
     pub latency_us: u64,
     pub dnssec: DnssecStatus,
@@ -107,6 +108,7 @@ mod tests {
             domain: "example.com".into(),
             query_type: QueryType::A,
             path: QueryPath::Forwarded,
+            transport: Transport::Udp,
             rescode: ResultCode::NOERROR,
             latency_us: 500,
             dnssec: DnssecStatus::Indeterminate,

src/recursive.rs

@@ -15,8 +15,8 @@ use crate::srtt::SrttCache;
 
 const MAX_REFERRAL_DEPTH: u8 = 10;
 const MAX_CNAME_DEPTH: u8 = 8;
-const NS_QUERY_TIMEOUT: Duration = Duration::from_millis(800);
+const NS_QUERY_TIMEOUT: Duration = Duration::from_millis(400);
-const TCP_TIMEOUT: Duration = Duration::from_millis(1500);
+const TCP_TIMEOUT: Duration = Duration::from_millis(400);
 const UDP_FAIL_THRESHOLD: u8 = 3;
 
 static QUERY_ID: AtomicU16 = AtomicU16::new(1);
@@ -202,23 +202,24 @@ pub(crate) fn resolve_iterative<'a>(
         let mut ns_idx = 0;
 
         for _ in 0..MAX_REFERRAL_DEPTH {
-            let ns_addr = match ns_addrs.get(ns_idx) {
+            if ns_idx >= ns_addrs.len() {
-                Some(addr) => *addr,
+                return Err("no nameserver available".into());
-                None => return Err("no nameserver available".into()),
+            }
-            };
 
             let (q_name, q_type) = minimize_query(qname, qtype, &current_zone);
 
             debug!(
-                "recursive: querying {} for {:?} {} (zone: {}, depth {})",
+                "recursive: querying {} (+ hedge) for {:?} {} (zone: {}, depth {})",
-                ns_addr, q_type, q_name, current_zone, referral_depth
+                ns_addrs[ns_idx], q_type, q_name, current_zone, referral_depth
             );
 
-            let response = match send_query(q_name, q_type, ns_addr, srtt).await {
+            let response = match send_query_hedged(q_name, q_type, &ns_addrs[ns_idx..], srtt).await
+            {
                 Ok(r) => r,
                 Err(e) => {
-                    debug!("recursive: NS {} failed: {}", ns_addr, e);
+                    debug!("recursive: NS query failed: {}", e);
-                    ns_idx += 1;
+                    let remaining = ns_addrs.len().saturating_sub(ns_idx);
+                    ns_idx += remaining.min(2);
                     continue;
                 }
             };
@@ -228,6 +229,9 @@ pub(crate) fn resolve_iterative<'a>(
             {
                 if let Some(zone) = referral_zone(&response) {
                     current_zone = zone;
+                    let mut cache_w = cache.write().unwrap();
+                    cache_ns_delegation(&mut cache_w, &current_zone, &response);
+                    drop(cache_w);
                 }
                 let mut all_ns = extract_ns_from_records(&response.answers);
                 if all_ns.is_empty() {
@@ -296,6 +300,7 @@ pub(crate) fn resolve_iterative<'a>(
 
             {
                 let mut cache_w = cache.write().unwrap();
+                cache_ns_delegation(&mut cache_w, &current_zone, &response);
                 cache_ds_from_authority(&mut cache_w, &response);
             }
             let mut new_ns_addrs = resolve_ns_addrs_from_glue(&response, &ns_names, cache);
@@ -560,6 +565,23 @@ fn cache_ds_from_authority(cache: &mut DnsCache, response: &DnsPacket) {
     }
 }
 
+/// Cache NS delegation records from a referral response so that
+/// `find_closest_ns` can skip re-querying TLD servers on subsequent lookups.
+fn cache_ns_delegation(cache: &mut DnsCache, zone: &str, response: &DnsPacket) {
+    let ns_records: Vec<_> = response
+        .authorities
+        .iter()
+        .filter(|r| matches!(r, DnsRecord::NS { .. }))
+        .cloned()
+        .collect();
+    if ns_records.is_empty() {
+        return;
+    }
+    let mut pkt = make_glue_packet();
+    pkt.answers = ns_records;
+    cache.insert(zone, QueryType::NS, &pkt);
+}
+
 fn make_glue_packet() -> DnsPacket {
     let mut pkt = DnsPacket::new();
     pkt.header.response = true;
@@ -587,6 +609,115 @@ async fn tcp_with_srtt(
     }
 }
 
+/// Smart NS query: fire to two servers simultaneously when SRTT is unknown
+/// (cold queries), or to the best server with SRTT-based hedge when known.
+async fn send_query_hedged(
+    qname: &str,
+    qtype: QueryType,
+    servers: &[SocketAddr],
+    srtt: &RwLock<SrttCache>,
+) -> crate::Result<DnsPacket> {
+    if servers.is_empty() {
+        return Err("no nameserver available".into());
+    }
+    if servers.len() == 1 {
+        return send_query(qname, qtype, servers[0], srtt).await;
+    }
+
+    let primary = servers[0];
+    let secondary = servers[1];
+    let primary_known = srtt.read().unwrap().is_known(primary.ip());
+
+    if !primary_known {
+        // Cold: fire both simultaneously, first response wins
+        debug!(
+            "recursive: parallel query to {} and {} for {:?} {}",
+            primary, secondary, qtype, qname
+        );
+        let fut_a = send_query(qname, qtype, primary, srtt);
+        let fut_b = send_query(qname, qtype, secondary, srtt);
+        tokio::pin!(fut_a);
+        tokio::pin!(fut_b);
+
+        // First Ok wins. If one errors, wait for the other.
+        let mut a_done = false;
+        let mut b_done = false;
+        let mut a_err: Option<crate::Error> = None;
+        let mut b_err: Option<crate::Error> = None;
+
+        loop {
+            tokio::select! {
+                r = &mut fut_a, if !a_done => {
+                    match r {
+                        Ok(resp) => return Ok(resp),
+                        Err(e) => { a_done = true; a_err = Some(e); }
+                    }
+                }
+                r = &mut fut_b, if !b_done => {
+                    match r {
+                        Ok(resp) => return Ok(resp),
+                        Err(e) => { b_done = true; b_err = Some(e); }
+                    }
+                }
+            }
+            match (a_err.take(), b_err.take()) {
+                (Some(e), Some(_)) => return Err(e),
+                (a, b) => {
+                    a_err = a;
+                    b_err = b;
+                }
+            }
+        }
+    } else {
+        // Warm: send to best, hedge after SRTT × 3 if slow
+        let hedge_ms = srtt.read().unwrap().get(primary.ip()) * 3;
+        let hedge_delay = Duration::from_millis(hedge_ms.max(50));
+
+        let fut_a = send_query(qname, qtype, primary, srtt);
+        tokio::pin!(fut_a);
+        let delay = tokio::time::sleep(hedge_delay);
+        tokio::pin!(delay);
+
+        tokio::select! {
+            r = &mut fut_a => return r,
+            _ = &mut delay => {}
+        }
+
+        debug!(
+            "recursive: hedging {} -> {} after {}ms for {:?} {}",
+            primary, secondary, hedge_ms, qtype, qname
+        );
+        let fut_b = send_query(qname, qtype, secondary, srtt);
+        tokio::pin!(fut_b);
+
+        // First Ok wins; if one errors, wait for the other.
+        let mut a_err: Option<crate::Error> = None;
+        let mut b_err: Option<crate::Error> = None;
+        loop {
+            tokio::select! {
+                r = &mut fut_a, if a_err.is_none() => {
+                    match r {
+                        Ok(resp) => return Ok(resp),
+                        Err(e) => {
+                            if b_err.is_some() { return Err(e); }
+                            a_err = Some(e);
+                        }
+                    }
+                }
+                r = &mut fut_b, if b_err.is_none() => {
+                    match r {
+                        Ok(resp) => return Ok(resp),
+                        Err(e) => {
+                            if let Some(ae) = a_err.take() { return Err(ae); }
+                            b_err = Some(e);
+                        }
+                    }
+                }
+            }
+        }
+    }
+}
+
 async fn send_query(
     qname: &str,
     qtype: QueryType,
@@ -634,9 +765,13 @@ async fn send_query(
                     "send_query: {} consecutive UDP failures — switching to TCP-first",
                     fails
                 );
+                // Now that UDP is disabled, retry this query via TCP
+                return tcp_with_srtt(&query, server, srtt, start).await;
             }
-            debug!("send_query: UDP failed for {}: {}, trying TCP", server, e);
+            // UDP works in general (priming succeeded) but this server timed out.
-            tcp_with_srtt(&query, server, srtt, start).await
+            // Don't waste another 400ms on TCP — the server is unreachable.
+            srtt.write().unwrap().record_failure(server.ip());
+            Err(e)
         }
     }
 }
@@ -678,6 +813,10 @@ mod tests {
     use super::*;
     use std::net::{Ipv4Addr, Ipv6Addr};
 
+    /// Tests that mutate the global UDP_DISABLED / UDP_FAILURES flags must hold
+    /// this lock to avoid racing with each other under `cargo test` parallelism.
+    static UDP_STATE_LOCK: std::sync::Mutex<()> = std::sync::Mutex::new(());
+
     #[test]
     fn extract_ns_from_authority() {
         let mut pkt = DnsPacket::new();
@@ -916,10 +1055,11 @@ mod tests {
     }
 
     /// TCP-only server returns authoritative answer directly.
-    /// Verifies: UDP fails → TCP fallback → resolves.
+    /// Verifies: when UDP is disabled, TCP-first resolves.
     #[tokio::test]
     async fn tcp_fallback_resolves_when_udp_blocked() {
-        UDP_DISABLED.store(false, Ordering::Relaxed);
+        let _guard = UDP_STATE_LOCK.lock().unwrap();
+        UDP_DISABLED.store(true, Ordering::Relaxed);
         UDP_FAILURES.store(0, Ordering::Release);
 
         let server_addr = spawn_tcp_dns_server(|query| {
@@ -950,49 +1090,32 @@ mod tests {
         }
     }
 
-    /// Full iterative resolution through TCP-only mock: root referral → authoritative answer.
+    /// TCP round-trip through mock: query → authoritative answer via forward_tcp.
-    /// The mock plays both roles (returns referral for NS queries, answer for A queries).
+    /// Uses forward_tcp directly to avoid dependence on the global UDP_DISABLED flag
+    /// which is shared across concurrent tests.
     #[tokio::test]
     async fn tcp_only_iterative_resolution() {
-        UDP_DISABLED.store(true, Ordering::Release); // Skip UDP entirely for speed
-
         let server_addr = spawn_tcp_dns_server(|query| {
             let q = match query.questions.first() {
                 Some(q) => q,
                 None => return DnsPacket::response_from(query, ResultCode::SERVFAIL),
             };
 
-            if q.qtype == QueryType::NS || q.name == "com" {
+            let mut resp = DnsPacket::response_from(query, ResultCode::NOERROR);
-                // Return referral — NS points back to ourselves (same IP, port 53 in glue
+            resp.header.authoritative_answer = true;
-                // won't work, but cache will have our address from root_hints)
+            resp.answers.push(DnsRecord::A {
-                let mut resp = DnsPacket::new();
+                domain: q.name.clone(),
-                resp.header.id = query.header.id;
+                addr: Ipv4Addr::new(10, 0, 0, 42),
-                resp.header.response = true;
+                ttl: 300,
-                resp.header.rescode = ResultCode::NOERROR;
+            });
-                resp.questions = query.questions.clone();
+            resp
-                resp.authorities.push(DnsRecord::NS {
-                    domain: "com".into(),
-                    host: "ns1.com".into(),
-                    ttl: 3600,
-                });
-                resp
-            } else {
-                // Return authoritative answer
-                let mut resp = DnsPacket::response_from(query, ResultCode::NOERROR);
-                resp.header.authoritative_answer = true;
-                resp.answers.push(DnsRecord::A {
-                    domain: q.name.clone(),
-                    addr: Ipv4Addr::new(10, 0, 0, 42),
-                    ttl: 300,
-                });
-                resp
-            }
         })
         .await;
 
-        let srtt = RwLock::new(SrttCache::new(true));
+        let query = DnsPacket::query(0x1234, "hello.example.com", QueryType::A);
-        let result = send_query("hello.example.com", QueryType::A, server_addr, &srtt).await;
+        let resp = crate::forward::forward_tcp(&query, server_addr, TCP_TIMEOUT)
-        let resp = result.expect("TCP-only send_query should work");
+            .await
+            .expect("TCP query should work");
         assert_eq!(resp.header.rescode, ResultCode::NOERROR);
         match &resp.answers[0] {
             DnsRecord::A { addr, .. } => assert_eq!(*addr, Ipv4Addr::new(10, 0, 0, 42)),
@@ -1002,7 +1125,8 @@ mod tests {
 
     #[tokio::test]
     async fn tcp_fallback_handles_nxdomain() {
-        UDP_DISABLED.store(false, Ordering::Relaxed);
+        let _guard = UDP_STATE_LOCK.lock().unwrap();
+        UDP_DISABLED.store(true, Ordering::Relaxed);
         UDP_FAILURES.store(0, Ordering::Release);
 
         let server_addr = spawn_tcp_dns_server(|query| {
@@ -1034,6 +1158,7 @@ mod tests {
 
     #[tokio::test]
     async fn udp_auto_disable_resets() {
+        let _guard = UDP_STATE_LOCK.lock().unwrap();
         UDP_DISABLED.store(true, Ordering::Release);
         UDP_FAILURES.store(5, Ordering::Relaxed);
 

src/setup_phone.rs

@@ -0,0 +1,126 @@
+//! `numa setup-phone` CLI — thin QR wrapper over the persistent mobile API.
+//!
+//! Before the mobile API existed, this command spawned its own one-shot
+//! HTTP server on port 8765 to serve a freshly-generated mobileconfig
+//! for a single download. That role now belongs to
+//! [`crate::mobile_api`], which runs persistently alongside the main
+//! API and serves `/mobileconfig` at the same port whenever Numa is
+//! running.
+//!
+//! This command is now a thin terminal-side wrapper:
+//!
+//!   1. Detect the current LAN IP
+//!   2. Render a terminal QR code pointing at
+//!      `http://<lan_ip>:8765/mobileconfig`
+//!   3. Print install instructions and exit
+//!
+//! The user scans the QR, iOS fetches the profile from the mobile API
+//! (which is always up as long as `numa` is running), installs, and the
+//! user walks through Settings → Certificate Trust Settings to enable
+//! trust.
+//!
+//! Numa must be running for the profile download to succeed; if the
+//! mobile API is not listening on port 8765, the download will fail
+//! and the user will see Safari's "Cannot Connect to Server" error.
+//! The CLI prints a reminder about this at the bottom of the output.
+
+use qrcode::render::unicode;
+use qrcode::QrCode;
+
+/// Default port where the persistent mobile API serves `/mobileconfig`.
+/// Matches `MobileConfig::default().port` in `config.rs`. If the user
+/// has overridden `[mobile] port = N` in `numa.toml`, they'll need to
+/// adjust the URL manually — this CLI uses the default without parsing
+/// `numa.toml`.
+const SETUP_PORT: u16 = 8765;
+
+fn render_qr(url: &str) -> Result<String, String> {
+    let code = QrCode::new(url).map_err(|e| format!("failed to encode QR: {}", e))?;
+    Ok(code
+        .render::<unicode::Dense1x2>()
+        .dark_color(unicode::Dense1x2::Light)
+        .light_color(unicode::Dense1x2::Dark)
+        .build())
+}
+
+/// Run the `numa setup-phone` flow.
+pub async fn run() -> Result<(), String> {
+    let lan_ip = crate::lan::detect_lan_ip()
+        .ok_or("could not detect LAN IP — are you connected to a network?")?;
+
+    let addr = std::net::SocketAddr::from(([127, 0, 0, 1], SETUP_PORT));
+    let api_reachable = tokio::time::timeout(
+        std::time::Duration::from_millis(500),
+        tokio::net::TcpStream::connect(addr),
+    )
+    .await
+    .map(|r| r.is_ok())
+    .unwrap_or(false);
+
+    if !api_reachable {
+        eprintln!();
+        eprintln!(
+            "  \x1b[1;38;2;192;98;58mNuma\x1b[0m — mobile API is not reachable on port {}.",
+            SETUP_PORT
+        );
+        eprintln!();
+        eprintln!("  The phone won't be able to download the profile until the mobile");
+        eprintln!("  API is running. Add this to your numa.toml and restart Numa:");
+        eprintln!();
+        eprintln!("    [mobile]");
+        eprintln!("    enabled = true");
+        eprintln!();
+        return Err("mobile API not running".into());
+    }
+
+    let url = format!("http://{}:{}/mobileconfig", lan_ip, SETUP_PORT);
+    let qr = render_qr(&url)?;
+
+    eprintln!();
+    eprintln!("  \x1b[1;38;2;192;98;58mNuma Phone Setup\x1b[0m");
+    eprintln!();
+    eprintln!("  Profile URL: \x1b[36m{}\x1b[0m", url);
+    eprintln!();
+    for line in qr.lines() {
+        eprintln!("  {}", line);
+    }
+    eprintln!();
+    eprintln!("  \x1b[1mOn your iPhone:\x1b[0m");
+    eprintln!("    1. Open Camera, point at the QR code, tap the yellow banner");
+    eprintln!("    2. Allow the download when Safari asks");
+    eprintln!("    3. Open Settings — tap \"Profile Downloaded\" near the top");
+    eprintln!("       (or: Settings → General → VPN & Device Management → Numa DNS)");
+    eprintln!("    4. Tap Install (top right), enter passcode, Install again");
+    eprintln!("    5. \x1b[1mSettings → General → About → Certificate Trust Settings\x1b[0m");
+    eprintln!("       Toggle ON \"Numa Local CA\" — required for DoT to work");
+    eprintln!();
+    eprintln!(
+        "  \x1b[33mNote:\x1b[0m profile uses your laptop's current IP ({}). If your",
+        lan_ip
+    );
+    eprintln!("  laptop changes networks, re-scan this QR — iOS will replace the");
+    eprintln!("  existing profile automatically (fixed UUID).");
+    eprintln!();
+    eprintln!(
+        "  \x1b[90mThe profile is served by Numa's persistent mobile API on port {}.\x1b[0m",
+        SETUP_PORT
+    );
+    eprintln!("  \x1b[90mMake sure `numa` is running before scanning. If it's not,\x1b[0m");
+    eprintln!("  \x1b[90mstart it with `sudo numa install` or run it interactively.\x1b[0m");
+    eprintln!();
+
+    Ok(())
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn render_qr_produces_unicode() {
+        let qr = render_qr("http://192.168.1.9:8765/mobileconfig").unwrap();
+        assert!(!qr.is_empty());
+        // Dense1x2 uses these block characters
+        assert!(qr.chars().any(|c| matches!(c, '█' | '▀' | '▄' | ' ')));
+    }
+}

src/srtt.rs

@@ -45,6 +45,11 @@ impl SrttCache {
         }
     }
 
+    /// Whether we have observed RTT data for this IP.
+    pub fn is_known(&self, ip: IpAddr) -> bool {
+        self.entries.contains_key(&ip)
+    }
+
     /// Apply time-based decay: each DECAY_AFTER_SECS period halves distance to INITIAL.
     fn decayed_srtt(entry: &SrttEntry) -> u64 {
         Self::decay_for_age(entry.srtt_ms, entry.updated_at.elapsed().as_secs())

src/stats.rs

@@ -97,9 +97,32 @@ pub struct ServerStats {
     queries_local: u64,
     queries_overridden: u64,
     upstream_errors: u64,
+    transport_udp: u64,
+    transport_tcp: u64,
+    transport_dot: u64,
+    transport_doh: u64,
     started_at: Instant,
 }
 
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub enum Transport {
+    Udp,
+    Tcp,
+    Dot,
+    Doh,
+}
+
+impl Transport {
+    pub fn as_str(&self) -> &'static str {
+        match self {
+            Transport::Udp => "UDP",
+            Transport::Tcp => "TCP",
+            Transport::Dot => "DOT",
+            Transport::Doh => "DOH",
+        }
+    }
+}
+
 #[derive(Clone, Copy, Debug, PartialEq, Eq)]
 pub enum QueryPath {
     Local,
@@ -167,11 +190,15 @@ impl ServerStats {
             queries_local: 0,
             queries_overridden: 0,
             upstream_errors: 0,
+            transport_udp: 0,
+            transport_tcp: 0,
+            transport_dot: 0,
+            transport_doh: 0,
             started_at: Instant::now(),
         }
     }
 
-    pub fn record(&mut self, path: QueryPath) -> u64 {
+    pub fn record(&mut self, path: QueryPath, transport: Transport) -> u64 {
         self.queries_total += 1;
         match path {
             QueryPath::Local => self.queries_local += 1,
@@ -183,6 +210,12 @@ impl ServerStats {
             QueryPath::Overridden => self.queries_overridden += 1,
             QueryPath::UpstreamError => self.upstream_errors += 1,
         }
+        match transport {
+            Transport::Udp => self.transport_udp += 1,
+            Transport::Tcp => self.transport_tcp += 1,
+            Transport::Dot => self.transport_dot += 1,
+            Transport::Doh => self.transport_doh += 1,
+        }
         self.queries_total
     }
 
@@ -206,6 +239,10 @@ impl ServerStats {
             overridden: self.queries_overridden,
             blocked: self.queries_blocked,
             errors: self.upstream_errors,
+            transport_udp: self.transport_udp,
+            transport_tcp: self.transport_tcp,
+            transport_dot: self.transport_dot,
+            transport_doh: self.transport_doh,
         }
     }
 
@@ -242,4 +279,8 @@ pub struct StatsSnapshot {
     pub overridden: u64,
     pub blocked: u64,
     pub errors: u64,
+    pub transport_udp: u64,
+    pub transport_tcp: u64,
+    pub transport_dot: u64,
+    pub transport_doh: u64,
 }

src/system_dns.rs

@@ -2,6 +2,19 @@ use std::net::SocketAddr;
 
 use log::info;
 
+use crate::forward::Upstream;
+
+fn print_recursive_hint() {
+    let is_recursive = crate::config::load_config("numa.toml")
+        .map(|c| c.config.upstream.mode == crate::config::UpstreamMode::Recursive)
+        .unwrap_or(false);
+    if !is_recursive {
+        eprintln!("  Want full DNS sovereignty? Add to numa.toml:");
+        eprintln!("    [upstream]");
+        eprintln!("    mode = \"recursive\"\n");
+    }
+}
+
 fn is_loopback_or_stub(addr: &str) -> bool {
     matches!(addr, "127.0.0.1" | "127.0.0.53" | "0.0.0.0" | "::1" | "")
 }
@@ -11,7 +24,18 @@ fn is_loopback_or_stub(addr: &str) -> bool {
 pub struct ForwardingRule {
     pub suffix: String,
     dot_suffix: String, // pre-computed ".suffix" for zero-alloc matching
-    pub upstream: SocketAddr,
+    pub upstream: Upstream,
+}
+
+impl ForwardingRule {
+    pub fn new(suffix: String, upstream: Upstream) -> Self {
+        let dot_suffix = format!(".{}", suffix);
+        Self {
+            suffix,
+            dot_suffix,
+            upstream,
+        }
+    }
 }
 
 /// Result of system DNS discovery — default upstream + conditional forwarding rules.
@@ -46,6 +70,61 @@ pub fn discover_system_dns() -> SystemDnsInfo {
     }
 }
 
+/// Advisory for port-53 bind failures (EADDRINUSE or EACCES); `None`
+/// if not applicable so the caller can fall back to the raw error.
+pub fn try_port53_advisory(bind_addr: &str, err: &std::io::Error) -> Option<String> {
+    if !is_port_53(bind_addr) {
+        return None;
+    }
+    let (title, cause) = match err.kind() {
+        std::io::ErrorKind::AddrInUse => (
+            "port 53 is already in use",
+            "Another process is already bound to port 53. On Linux this is\n  \
+             typically systemd-resolved; on Windows, the DNS Client service.",
+        ),
+        std::io::ErrorKind::PermissionDenied => (
+            "permission denied",
+            "Port 53 is privileged — binding it requires root on Linux/macOS\n  \
+             or Administrator on Windows.",
+        ),
+        _ => return None,
+    };
+    let o = "\x1b[1;38;2;192;98;58m"; // bold orange
+    let r = "\x1b[0m";
+    Some(format!(
+        "
+{o}Numa{r} — cannot bind to {bind_addr}: {title}.
+
+  {cause}
+
+  Fix — pick one:
+
+    1. Install Numa as the system resolver (frees port 53):
+
+         sudo numa install       (on Windows, run as Administrator)
+
+    2. Run on a non-privileged port for testing.
+       Create {} with:
+
+         [server]
+         bind_addr = \"127.0.0.1:5354\"
+         api_port  = 5380
+
+       Then run:  numa
+       Test with: dig @127.0.0.1 -p 5354 example.com
+
+",
+        crate::suggested_config_path().display()
+    ))
+}
+
+fn is_port_53(bind_addr: &str) -> bool {
+    bind_addr
+        .parse::<SocketAddr>()
+        .map(|s| s.port() == 53)
+        .unwrap_or(false)
+}
+
 #[cfg(target_os = "macos")]
 fn discover_macos() -> SystemDnsInfo {
     use log::{debug, warn};
@@ -155,12 +234,8 @@ fn discover_macos() -> SystemDnsInfo {
 
 #[cfg(any(target_os = "macos", target_os = "linux"))]
 fn make_rule(domain: &str, nameserver: &str) -> Option<ForwardingRule> {
-    let addr: SocketAddr = format!("{}:53", nameserver).parse().ok()?;
+    let addr = crate::forward::parse_upstream_addr(nameserver, 53).ok()?;
-    Some(ForwardingRule {
+    Some(ForwardingRule::new(domain.to_string(), Upstream::Udp(addr)))
-        dot_suffix: format!(".{}", domain),
-        suffix: domain.to_string(),
-        upstream: addr,
-    })
 }
 
 #[cfg(target_os = "linux")]
@@ -174,6 +249,9 @@ fn discover_linux() -> SystemDnsInfo {
     let default_upstream = if let Some(ns) = upstream {
         info!("detected system upstream: {}", ns);
         Some(ns)
+    } else if let Some(ns) = resolvectl_dns_server() {
+        info!("detected system upstream via resolvectl: {}", ns);
+        Some(ns)
     } else {
         // Fallback to backup from a previous `numa install`
         let backup = {
@@ -214,7 +292,18 @@ fn discover_linux() -> SystemDnsInfo {
     }
 }
 
-/// Parse resolv.conf in a single pass, extracting both the first non-loopback
+/// Yield each `nameserver` address from resolv.conf content. No filtering —
+/// callers decide what counts as a real upstream.
+#[cfg(any(target_os = "linux", test))]
+fn iter_nameservers(content: &str) -> impl Iterator<Item = &str> {
+    content.lines().filter_map(|line| {
+        let mut parts = line.split_whitespace();
+        (parts.next() == Some("nameserver")).then_some(())?;
+        parts.next()
+    })
+}
+
+/// Parse resolv.conf in a single pass, extracting the first non-loopback
 /// nameserver and all search domains.
 #[cfg(target_os = "linux")]
 fn parse_resolv_conf(path: &str) -> (Option<String>, Vec<String>) {
@@ -222,19 +311,13 @@ fn parse_resolv_conf(path: &str) -> (Option<String>, Vec<String>) {
         Ok(t) => t,
         Err(_) => return (None, Vec::new()),
     };
-    let mut upstream = None;
+    let upstream = iter_nameservers(&text)
+        .find(|ns| !is_loopback_or_stub(ns))
+        .map(str::to_string);
     let mut search_domains = Vec::new();
     for line in text.lines() {
         let line = line.trim();
-        if line.starts_with("nameserver") {
+        if line.starts_with("search") || line.starts_with("domain") {
-            if upstream.is_none() {
-                if let Some(ns) = line.split_whitespace().nth(1) {
-                    if !is_loopback_or_stub(ns) {
-                        upstream = Some(ns.to_string());
-                    }
-                }
-            }
-        } else if line.starts_with("search") || line.starts_with("domain") {
             for domain in line.split_whitespace().skip(1) {
                 search_domains.push(domain.to_string());
             }
@@ -243,6 +326,21 @@ fn parse_resolv_conf(path: &str) -> (Option<String>, Vec<String>) {
     (upstream, search_domains)
 }
 
+/// True if the resolv.conf *content* appears to be written by numa itself,
+/// or has no real upstream — either way, it's not a safe source of truth
+/// for a backup.
+#[cfg(any(target_os = "linux", test))]
+fn resolv_conf_is_numa_managed(content: &str) -> bool {
+    content.contains("Generated by Numa") || !resolv_conf_has_real_upstream(content)
+}
+
+/// True if the resolv.conf content has at least one non-loopback, non-stub
+/// nameserver. An all-loopback resolv.conf is self-referential.
+#[cfg(any(target_os = "linux", test))]
+fn resolv_conf_has_real_upstream(content: &str) -> bool {
+    iter_nameservers(content).any(|ns| !is_loopback_or_stub(ns))
+}
+
 /// Query resolvectl for the real upstream DNS server (e.g. VPC resolver on AWS).
 #[cfg(target_os = "linux")]
 fn resolvectl_dns_server() -> Option<String> {
@@ -526,9 +624,19 @@ fn enable_dnscache() {
         .status();
 }
 
+/// True if the backup map has at least one real upstream (non-loopback, non-stub).
+#[cfg(any(windows, test))]
+fn backup_has_real_upstream_windows(
+    interfaces: &std::collections::HashMap<String, WindowsInterfaceDns>,
+) -> bool {
+    interfaces
+        .values()
+        .any(|iface| iface.servers.iter().any(|s| !is_loopback_or_stub(s)))
+}
+
 #[cfg(windows)]
 fn install_windows() -> Result<(), String> {
-    let interfaces = get_windows_interfaces()?;
+    let mut interfaces = get_windows_interfaces()?;
     if interfaces.is_empty() {
         return Err("no active network interfaces found".to_string());
     }
@@ -538,9 +646,30 @@ fn install_windows() -> Result<(), String> {
         std::fs::create_dir_all(parent)
             .map_err(|e| format!("failed to create {}: {}", parent.display(), e))?;
     }
-    let json = serde_json::to_string_pretty(&interfaces)
+
-        .map_err(|e| format!("failed to serialize backup: {}", e))?;
+    // Preserve an existing useful backup rather than overwriting it with
-    std::fs::write(&path, json).map_err(|e| format!("failed to write backup: {}", e))?;
+    // numa-managed state (which would be self-referential after uninstall).
+    let existing: Option<std::collections::HashMap<String, WindowsInterfaceDns>> =
+        std::fs::read_to_string(&path)
+            .ok()
+            .and_then(|json| serde_json::from_str(&json).ok());
+    let has_useful_existing = existing
+        .as_ref()
+        .map(backup_has_real_upstream_windows)
+        .unwrap_or(false);
+
+    if has_useful_existing {
+        eprintln!("  Existing DNS backup preserved at {}", path.display());
+    } else {
+        // Filter loopback/stub addresses before saving so a fresh backup
+        // captured from already-numa-managed state isn't self-referential.
+        for iface in interfaces.values_mut() {
+            iface.servers.retain(|s| !is_loopback_or_stub(s));
+        }
+        let json = serde_json::to_string_pretty(&interfaces)
+            .map_err(|e| format!("failed to serialize backup: {}", e))?;
+        std::fs::write(&path, json).map_err(|e| format!("failed to write backup: {}", e))?;
+    }
 
     for name in interfaces.keys() {
         let status = std::process::Command::new("netsh")
@@ -570,16 +699,17 @@ fn install_windows() -> Result<(), String> {
     let needs_reboot = disable_dnscache()?;
     register_autostart();
 
-    eprintln!("\n  Original DNS saved to {}", path.display());
+    eprintln!();
+    if !has_useful_existing {
+        eprintln!("  Original DNS saved to {}", path.display());
+    }
     eprintln!("  Run 'numa uninstall' to restore.\n");
     if needs_reboot {
         eprintln!("  *** Reboot required. Numa will start automatically. ***\n");
     } else {
         eprintln!("  Numa will start automatically on next boot.\n");
     }
-    eprintln!("  Want full DNS sovereignty? Add to numa.toml:");
+    print_recursive_hint();
-    eprintln!("    [upstream]");
-    eprintln!("    mode = \"recursive\"\n");
     Ok(())
 }
 
@@ -694,10 +824,13 @@ fn uninstall_windows() -> Result<(), String> {
 /// Find the upstream for a domain by checking forwarding rules.
 /// Returns None if no rule matches (use default upstream).
 /// Zero-allocation on the hot path — dot_suffix is pre-computed.
-pub fn match_forwarding_rule(domain: &str, rules: &[ForwardingRule]) -> Option<SocketAddr> {
+pub fn match_forwarding_rule<'a>(
+    domain: &str,
+    rules: &'a [ForwardingRule],
+) -> Option<&'a Upstream> {
     for rule in rules {
         if domain == rule.suffix || domain.ends_with(&rule.dot_suffix) {
-            return Some(rule.upstream);
+            return Some(&rule.upstream);
         }
     }
     None
@@ -754,27 +887,60 @@ fn get_dns_servers(service: &str) -> Result<Vec<String>, String> {
     }
 }
 
+/// True if the backup map has at least one real upstream (non-loopback, non-stub).
+/// An all-loopback backup is self-referential — restoring it is a no-op.
+#[cfg(any(target_os = "macos", test))]
+fn backup_has_real_upstream_macos(
+    servers: &std::collections::HashMap<String, Vec<String>>,
+) -> bool {
+    servers
+        .values()
+        .any(|list| list.iter().any(|s| !is_loopback_or_stub(s)))
+}
+
 #[cfg(target_os = "macos")]
 fn install_macos() -> Result<(), String> {
     use std::collections::HashMap;
 
     let services = get_network_services()?;
-    let mut original: HashMap<String, Vec<String>> = HashMap::new();
-
-    // Save current DNS for each service
-    for service in &services {
-        let servers = get_dns_servers(service)?;
-        original.insert(service.clone(), servers);
-    }
-
-    // Save backup
     let dir = numa_data_dir();
     std::fs::create_dir_all(&dir)
         .map_err(|e| format!("failed to create {}: {}", dir.display(), e))?;
 
-    let json = serde_json::to_string_pretty(&original)
+    // If a useful backup already exists (at least one non-loopback upstream),
-        .map_err(|e| format!("failed to serialize backup: {}", e))?;
+    // preserve it — overwriting would destroy the original DNS state when
-    std::fs::write(backup_path(), json).map_err(|e| format!("failed to write backup: {}", e))?;
+    // re-installing on top of a numa-managed configuration.
+    let existing_backup: Option<HashMap<String, Vec<String>>> =
+        std::fs::read_to_string(backup_path())
+            .ok()
+            .and_then(|json| serde_json::from_str(&json).ok());
+    let has_useful_existing = existing_backup
+        .as_ref()
+        .map(backup_has_real_upstream_macos)
+        .unwrap_or(false);
+
+    if has_useful_existing {
+        eprintln!(
+            "  Existing DNS backup preserved at {}",
+            backup_path().display()
+        );
+    } else {
+        // Capture fresh, filtering out loopback and stub addresses so we
+        // never record a self-referential backup.
+        let mut original: HashMap<String, Vec<String>> = HashMap::new();
+        for service in &services {
+            let servers: Vec<String> = get_dns_servers(service)?
+                .into_iter()
+                .filter(|s| !is_loopback_or_stub(s))
+                .collect();
+            original.insert(service.clone(), servers);
+        }
+
+        let json = serde_json::to_string_pretty(&original)
+            .map_err(|e| format!("failed to serialize backup: {}", e))?;
+        std::fs::write(backup_path(), json)
+            .map_err(|e| format!("failed to write backup: {}", e))?;
+    }
 
     // Set DNS to 127.0.0.1 and add "numa" search domain for each service
     for service in &services {
@@ -795,7 +961,10 @@ fn install_macos() -> Result<(), String> {
             .status();
     }
 
-    eprintln!("\n  Original DNS saved to {}", backup_path().display());
+    eprintln!();
+    if !has_useful_existing {
+        eprintln!("  Original DNS saved to {}", backup_path().display());
+    }
     eprintln!("  Run 'sudo numa uninstall' to restore.\n");
 
     Ok(())
@@ -990,14 +1159,23 @@ fn install_service_macos() -> Result<(), String> {
     std::fs::write(PLIST_DEST, plist)
         .map_err(|e| format!("failed to write {}: {}", PLIST_DEST, e))?;
 
-    // Load the service first so numa is listening before DNS redirect
+    // Modern launchctl API: explicitly tear down any existing in-memory
+    // state, then bootstrap fresh from the on-disk plist. The deprecated
+    // `load -w` returns exit 0 even when it cannot actually reload (label
+    // already in launchd state), silently leaving the daemon running a
+    // stale binary path after `numa install` rewrites the plist on disk —
+    // which is exactly what `brew upgrade numa` does.
+    let _ = std::process::Command::new("launchctl")
+        .args(["bootout", "system", PLIST_DEST])
+        .status();
+
     let status = std::process::Command::new("launchctl")
-        .args(["load", "-w", PLIST_DEST])
+        .args(["bootstrap", "system", PLIST_DEST])
         .status()
         .map_err(|e| format!("failed to run launchctl: {}", e))?;
 
     if !status.success() {
-        return Err("launchctl load failed".to_string());
+        return Err("launchctl bootstrap failed".to_string());
     }
 
     // Wait for numa to be ready before redirecting DNS
@@ -1010,7 +1188,7 @@ fn install_service_macos() -> Result<(), String> {
     if !api_up {
         // Service failed to start — don't redirect DNS to a dead endpoint
         let _ = std::process::Command::new("launchctl")
-            .args(["unload", PLIST_DEST])
+            .args(["bootout", "system", PLIST_DEST])
             .status();
         return Err(
             "numa service did not start (port 53 may be in use). Service unloaded.".to_string(),
@@ -1025,9 +1203,7 @@ fn install_service_macos() -> Result<(), String> {
     eprintln!("  Numa will auto-start on boot and restart if killed.");
     eprintln!("  Logs: /usr/local/var/log/numa.log");
     eprintln!("  Run 'sudo numa uninstall' to restore original DNS.\n");
-    eprintln!("  Want full DNS sovereignty? Add to numa.toml:");
+    print_recursive_hint();
-    eprintln!("    [upstream]");
-    eprintln!("    mode = \"recursive\"\n");
     Ok(())
 }
 
@@ -1038,22 +1214,25 @@ fn uninstall_service_macos() -> Result<(), String> {
         eprintln!("  warning: failed to restore system DNS: {}", e);
     }
 
-    // Remove plist first so service won't restart on boot even if unload fails
+    // Bootout the service from launchd's in-memory state BEFORE removing
-    if let Err(e) = std::fs::remove_file(PLIST_DEST) {
+    // the plist. The modern API needs the file path as the specifier;
-        if e.kind() != std::io::ErrorKind::NotFound {
+    // doing this in the wrong order would leave the service loaded in
-            return Err(format!("failed to remove {}: {}", PLIST_DEST, e));
+    // memory until reboot. (Deprecated `unload -w` had the same issue.)
+    let bootout_status = std::process::Command::new("launchctl")
+        .args(["bootout", "system", PLIST_DEST])
+        .status();
+    if let Ok(s) = bootout_status {
+        if !s.success() {
+            eprintln!(
+                "  warning: launchctl bootout returned non-zero (service may not have been loaded)"
+            );
         }
     }
 
-    // Unload the service
+    // Remove plist so the service won't restart on boot
-    let status = std::process::Command::new("launchctl")
+    if let Err(e) = std::fs::remove_file(PLIST_DEST) {
-        .args(["unload", "-w", PLIST_DEST])
+        if e.kind() != std::io::ErrorKind::NotFound {
-        .status();
+            return Err(format!("failed to remove {}: {}", PLIST_DEST, e));
-    if let Ok(s) = status {
-        if !s.success() {
-            eprintln!(
-                "  warning: launchctl unload returned non-zero (service may still be running)"
-            );
         }
     }
 
@@ -1132,11 +1311,31 @@ fn install_linux() -> Result<(), String> {
             .map_err(|e| format!("failed to create {}: {}", parent.display(), e))?;
     }
 
-    // Back up current resolv.conf (ignore NotFound)
+    // Back up current resolv.conf, but never overwrite a useful existing
-    match std::fs::copy(resolv, &backup) {
+    // backup with a numa-managed file — that would leave uninstall with
-        Ok(_) => eprintln!("  Saved /etc/resolv.conf to {}", backup.display()),
+    // nothing to restore to.
-        Err(e) if e.kind() == std::io::ErrorKind::NotFound => {}
+    let current = std::fs::read_to_string(resolv).ok();
-        Err(e) => return Err(format!("failed to backup /etc/resolv.conf: {}", e)),
+    let current_is_numa_managed = current
+        .as_deref()
+        .map(resolv_conf_is_numa_managed)
+        .unwrap_or(false);
+    let existing_backup_is_useful = std::fs::read_to_string(&backup)
+        .ok()
+        .as_deref()
+        .map(resolv_conf_has_real_upstream)
+        .unwrap_or(false);
+
+    if existing_backup_is_useful {
+        eprintln!(
+            "  Existing resolv.conf backup preserved at {}",
+            backup.display()
+        );
+    } else if current_is_numa_managed {
+        eprintln!("  warning: /etc/resolv.conf is already numa-managed; no fresh backup written");
+    } else if let Some(content) = current.as_deref() {
+        std::fs::write(&backup, content)
+            .map_err(|e| format!("failed to backup /etc/resolv.conf: {}", e))?;
+        eprintln!("  Saved /etc/resolv.conf to {}", backup.display());
     }
 
     if resolv
@@ -1209,9 +1408,7 @@ fn install_service_linux() -> Result<(), String> {
     eprintln!("  Numa will auto-start on boot and restart if killed.");
     eprintln!("  Logs: journalctl -u numa -f");
     eprintln!("  Run 'sudo numa uninstall' to restore original DNS.\n");
-    eprintln!("  Want full DNS sovereignty? Add to numa.toml:");
+    print_recursive_hint();
-    eprintln!("    [upstream]");
-    eprintln!("    mode = \"recursive\"\n");
     Ok(())
 }
 
@@ -1278,102 +1475,209 @@ fn run_systemctl(args: &[&str]) -> Result<(), String> {
 
 // --- CA trust management ---
 
+/// One Linux trust-store backend (Debian, Fedora pki, Arch p11-kit).
+#[cfg(target_os = "linux")]
+struct LinuxTrustStore {
+    name: &'static str,
+    anchor_dir: &'static str,
+    anchor_file: &'static str,
+    refresh_install: &'static [&'static str],
+    refresh_uninstall: &'static [&'static str],
+}
+
+// If you change this table, update tests/docker/install-trust.sh to match —
+// it asserts the same paths/commands against real distro images.
+#[cfg(target_os = "linux")]
+const LINUX_TRUST_STORES: &[LinuxTrustStore] = &[
+    // Debian / Ubuntu / Mint
+    LinuxTrustStore {
+        name: "debian",
+        anchor_dir: "/usr/local/share/ca-certificates",
+        anchor_file: "numa-local-ca.crt",
+        refresh_install: &["update-ca-certificates"],
+        refresh_uninstall: &["update-ca-certificates", "--fresh"],
+    },
+    // Fedora / RHEL / CentOS / SUSE (p11-kit via update-ca-trust wrapper)
+    LinuxTrustStore {
+        name: "pki",
+        anchor_dir: "/etc/pki/ca-trust/source/anchors",
+        anchor_file: "numa-local-ca.pem",
+        refresh_install: &["update-ca-trust", "extract"],
+        refresh_uninstall: &["update-ca-trust", "extract"],
+    },
+    // Arch / Manjaro (raw p11-kit)
+    LinuxTrustStore {
+        name: "p11kit",
+        anchor_dir: "/etc/ca-certificates/trust-source/anchors",
+        anchor_file: "numa-local-ca.pem",
+        refresh_install: &["trust", "extract-compat"],
+        refresh_uninstall: &["trust", "extract-compat"],
+    },
+];
+
+#[cfg(target_os = "linux")]
+fn detect_linux_trust_store() -> Option<&'static LinuxTrustStore> {
+    LINUX_TRUST_STORES
+        .iter()
+        .find(|s| std::path::Path::new(s.anchor_dir).is_dir())
+}
+
 fn trust_ca() -> Result<(), String> {
-    let ca_path = crate::data_dir().join("ca.pem");
+    let ca_path = crate::data_dir().join(crate::tls::CA_FILE_NAME);
     if !ca_path.exists() {
         return Err("CA not generated yet — start numa first to create certificates".into());
     }
 
     #[cfg(target_os = "macos")]
-    {
+    let result = trust_ca_macos(&ca_path);
-        let status = std::process::Command::new("security")
-            .args([
-                "add-trusted-cert",
-                "-d",
-                "-r",
-                "trustRoot",
-                "-k",
-                "/Library/Keychains/System.keychain",
-            ])
-            .arg(&ca_path)
-            .status()
-            .map_err(|e| format!("security: {}", e))?;
-        if !status.success() {
-            return Err("security add-trusted-cert failed".into());
-        }
-        eprintln!("  Trusted Numa CA in system keychain");
-    }
-
     #[cfg(target_os = "linux")]
-    {
+    let result = trust_ca_linux(&ca_path);
-        let dest = std::path::Path::new("/usr/local/share/ca-certificates/numa-local-ca.crt");
+    #[cfg(windows)]
-        std::fs::copy(&ca_path, dest).map_err(|e| format!("copy CA: {}", e))?;
+    let result = trust_ca_windows(&ca_path);
-        let status = std::process::Command::new("update-ca-certificates")
+    #[cfg(not(any(target_os = "macos", target_os = "linux", windows)))]
-            .status()
+    let result = Err::<(), String>("CA trust not supported on this OS".to_string());
-            .map_err(|e| format!("update-ca-certificates: {}", e))?;
-        if !status.success() {
-            return Err("update-ca-certificates failed".into());
-        }
-        eprintln!("  Trusted Numa CA system-wide");
-    }
 
-    #[cfg(not(any(target_os = "macos", target_os = "linux")))]
+    result
-    {
-        Err("CA trust not supported on this OS".into())
-    }
-
-    #[cfg(any(target_os = "macos", target_os = "linux"))]
-    Ok(())
 }
 
 fn untrust_ca() -> Result<(), String> {
-    let ca_path = crate::data_dir().join("ca.pem");
-
     #[cfg(target_os = "macos")]
+    let result = untrust_ca_macos();
+    #[cfg(target_os = "linux")]
+    let result = untrust_ca_linux();
+    #[cfg(windows)]
+    let result = untrust_ca_windows();
+    #[cfg(not(any(target_os = "macos", target_os = "linux", windows)))]
+    let result = Ok::<(), String>(());
+
+    result
+}
+
+#[cfg(target_os = "macos")]
+fn trust_ca_macos(ca_path: &std::path::Path) -> Result<(), String> {
+    let status = std::process::Command::new("security")
+        .args([
+            "add-trusted-cert",
+            "-d",
+            "-r",
+            "trustRoot",
+            "-k",
+            "/Library/Keychains/System.keychain",
+        ])
+        .arg(ca_path)
+        .status()
+        .map_err(|e| format!("security: {}", e))?;
+    if !status.success() {
+        return Err("security add-trusted-cert failed".into());
+    }
+    eprintln!("  Trusted Numa CA in system keychain");
+    Ok(())
+}
+
+#[cfg(target_os = "macos")]
+fn untrust_ca_macos() -> Result<(), String> {
+    if let Ok(out) = std::process::Command::new("security")
+        .args([
+            "find-certificate",
+            "-c",
+            crate::tls::CA_COMMON_NAME,
+            "-a",
+            "-Z",
+            "/Library/Keychains/System.keychain",
+        ])
+        .output()
     {
-        // Find all Numa CA certs by hash and delete each one
+        let stdout = String::from_utf8_lossy(&out.stdout);
-        if let Ok(out) = std::process::Command::new("security")
+        for line in stdout.lines() {
-            .args([
+            if let Some(hash) = line.strip_prefix("SHA-1 hash: ") {
-                "find-certificate",
+                let hash = hash.trim();
-                "-c",
+                let _ = std::process::Command::new("security")
-                "Numa Local CA",
+                    .args([
-                "-a",
+                        "delete-certificate",
-                "-Z",
+                        "-Z",
-                "/Library/Keychains/System.keychain",
+                        hash,
-            ])
+                        "/Library/Keychains/System.keychain",
-            .output()
+                    ])
-        {
+                    .output();
-            let stdout = String::from_utf8_lossy(&out.stdout);
-            for line in stdout.lines() {
-                if let Some(hash) = line.strip_prefix("SHA-1 hash: ") {
-                    let hash = hash.trim();
-                    let _ = std::process::Command::new("security")
-                        .args([
-                            "delete-certificate",
-                            "-Z",
-                            hash,
-                            "/Library/Keychains/System.keychain",
-                        ])
-                        .output();
-                }
             }
         }
-        eprintln!("  Removed Numa CA from system keychain");
     }
+    eprintln!("  Removed Numa CA from system keychain");
+    Ok(())
+}
 
-    #[cfg(target_os = "linux")]
+#[cfg(target_os = "linux")]
-    {
+fn trust_ca_linux(ca_path: &std::path::Path) -> Result<(), String> {
-        let dest = std::path::Path::new("/usr/local/share/ca-certificates/numa-local-ca.crt");
+    let store = detect_linux_trust_store().ok_or_else(|| {
-        if dest.exists() {
+        let names: Vec<&str> = LINUX_TRUST_STORES.iter().map(|s| s.name).collect();
-            let _ = std::fs::remove_file(dest);
+        format!(
-            let _ = std::process::Command::new("update-ca-certificates")
+            "no supported CA trust store found (tried: {}). \
-                .arg("--fresh")
+             Please report at https://github.com/razvandimescu/numa/issues",
-                .status();
+            names.join(", ")
-            eprintln!("  Removed Numa CA from system trust store");
+        )
+    })?;
+
+    let dest = std::path::Path::new(store.anchor_dir).join(store.anchor_file);
+    std::fs::copy(ca_path, &dest).map_err(|e| format!("copy CA to {}: {}", dest.display(), e))?;
+
+    run_refresh(store.name, store.refresh_install)?;
+    eprintln!("  Trusted Numa CA system-wide ({})", store.name);
+    Ok(())
+}
+
+#[cfg(target_os = "linux")]
+fn untrust_ca_linux() -> Result<(), String> {
+    let Some(store) = detect_linux_trust_store() else {
+        return Ok(());
+    };
+
+    let dest = std::path::Path::new(store.anchor_dir).join(store.anchor_file);
+    match std::fs::remove_file(&dest) {
+        Ok(()) => {
+            let _ = run_refresh(store.name, store.refresh_uninstall);
+            eprintln!("  Removed Numa CA from system trust store ({})", store.name);
         }
+        Err(e) if e.kind() == std::io::ErrorKind::NotFound => {}
+        Err(_) => {} // best-effort uninstall
     }
+    Ok(())
+}
 
-    let _ = ca_path; // suppress unused warning on other platforms
+#[cfg(target_os = "linux")]
+fn run_refresh(store_name: &str, argv: &[&str]) -> Result<(), String> {
+    let (cmd, args) = argv
+        .split_first()
+        .expect("refresh command must be non-empty");
+    let status = std::process::Command::new(cmd)
+        .args(args)
+        .status()
+        .map_err(|e| format!("{} ({}): {}", cmd, store_name, e))?;
+    if !status.success() {
+        return Err(format!("{} ({}) failed", cmd, store_name));
+    }
+    Ok(())
+}
+
+#[cfg(windows)]
+fn trust_ca_windows(ca_path: &std::path::Path) -> Result<(), String> {
+    let status = std::process::Command::new("certutil")
+        .args(["-addstore", "-f", "Root"])
+        .arg(ca_path)
+        .status()
+        .map_err(|e| format!("certutil: {}", e))?;
+    if !status.success() {
+        return Err("certutil -addstore Root failed (run as Administrator?)".into());
+    }
+    eprintln!("  Trusted Numa CA in Windows Root store");
+    Ok(())
+}
+
+#[cfg(windows)]
+fn untrust_ca_windows() -> Result<(), String> {
+    let _ = std::process::Command::new("certutil")
+        .args(["-delstore", "Root", crate::tls::CA_COMMON_NAME])
+        .status();
+    eprintln!("  Removed Numa CA from Windows Root store");
     Ok(())
 }
 
@@ -1432,6 +1736,82 @@ Wireless LAN adapter Wi-Fi:
         assert!(!result.contains("{{exe_path}}"));
     }
 
+    #[test]
+    fn macos_backup_real_upstream_detection() {
+        use std::collections::HashMap;
+        let mut map: HashMap<String, Vec<String>> = HashMap::new();
+
+        // Empty backup → no real upstream
+        assert!(!backup_has_real_upstream_macos(&map));
+
+        // All-loopback backup → still no real upstream (the bug case)
+        map.insert("Wi-Fi".into(), vec!["127.0.0.1".into()]);
+        map.insert("Ethernet".into(), vec!["::1".into()]);
+        assert!(!backup_has_real_upstream_macos(&map));
+
+        // One real entry → useful
+        map.insert("Tailscale".into(), vec!["192.168.1.1".into()]);
+        assert!(backup_has_real_upstream_macos(&map));
+    }
+
+    #[test]
+    fn windows_backup_filters_loopback() {
+        use std::collections::HashMap;
+        let mut map: HashMap<String, WindowsInterfaceDns> = HashMap::new();
+
+        // Empty backup → no real upstream
+        assert!(!backup_has_real_upstream_windows(&map));
+
+        // All-loopback backup → still no real upstream (the bug case)
+        map.insert(
+            "Wi-Fi".into(),
+            WindowsInterfaceDns {
+                dhcp: false,
+                servers: vec!["127.0.0.1".into()],
+            },
+        );
+        map.insert(
+            "Ethernet".into(),
+            WindowsInterfaceDns {
+                dhcp: false,
+                servers: vec!["::1".into(), "0.0.0.0".into()],
+            },
+        );
+        assert!(!backup_has_real_upstream_windows(&map));
+
+        // One real entry alongside loopback → useful
+        map.insert(
+            "Ethernet 2".into(),
+            WindowsInterfaceDns {
+                dhcp: false,
+                servers: vec!["192.168.1.1".into()],
+            },
+        );
+        assert!(backup_has_real_upstream_windows(&map));
+    }
+
+    #[test]
+    fn resolv_conf_real_upstream_detection() {
+        let real = "nameserver 192.168.1.1\nsearch lan\n";
+        assert!(resolv_conf_has_real_upstream(real));
+        assert!(!resolv_conf_is_numa_managed(real));
+
+        let self_ref = "nameserver 127.0.0.1\nsearch numa\n";
+        assert!(!resolv_conf_has_real_upstream(self_ref));
+        assert!(resolv_conf_is_numa_managed(self_ref));
+
+        let numa_marker =
+            "# Generated by Numa — run 'sudo numa uninstall' to restore\nnameserver 127.0.0.1\nsearch numa\n";
+        assert!(resolv_conf_is_numa_managed(numa_marker));
+
+        let systemd_stub = "nameserver 127.0.0.53\noptions edns0\n";
+        assert!(!resolv_conf_has_real_upstream(systemd_stub));
+
+        let mixed = "nameserver 127.0.0.1\nnameserver 1.1.1.1\n";
+        assert!(resolv_conf_has_real_upstream(mixed));
+        assert!(!resolv_conf_is_numa_managed(mixed));
+    }
+
     #[test]
     fn parse_ipconfig_skips_disconnected() {
         let sample = "\
@@ -1448,4 +1828,43 @@ Wireless LAN adapter Wi-Fi:
         assert_eq!(result.len(), 1);
         assert!(result.contains_key("Wi-Fi"));
     }
+
+    #[test]
+    fn try_port53_advisory_addr_in_use() {
+        let err = std::io::Error::from(std::io::ErrorKind::AddrInUse);
+        let msg = try_port53_advisory("0.0.0.0:53", &err).expect("should advise on port 53");
+        assert!(msg.contains("cannot bind to"));
+        assert!(msg.contains("already in use"));
+        assert!(msg.contains("numa install"));
+        assert!(msg.contains("bind_addr"));
+    }
+
+    #[test]
+    fn try_port53_advisory_permission_denied() {
+        let err = std::io::Error::from(std::io::ErrorKind::PermissionDenied);
+        let msg = try_port53_advisory("0.0.0.0:53", &err).expect("should advise on port 53");
+        assert!(msg.contains("cannot bind to"));
+        assert!(msg.contains("permission denied"));
+        assert!(msg.contains("numa install"));
+        assert!(msg.contains("bind_addr"));
+    }
+
+    #[test]
+    fn try_port53_advisory_skips_non_53_ports() {
+        let err = std::io::Error::from(std::io::ErrorKind::AddrInUse);
+        assert!(try_port53_advisory("127.0.0.1:5354", &err).is_none());
+        assert!(try_port53_advisory("[::]:853", &err).is_none());
+    }
+
+    #[test]
+    fn try_port53_advisory_skips_unrelated_error_kinds() {
+        let err = std::io::Error::from(std::io::ErrorKind::NotFound);
+        assert!(try_port53_advisory("0.0.0.0:53", &err).is_none());
+    }
+
+    #[test]
+    fn try_port53_advisory_skips_malformed_bind_addr() {
+        let err = std::io::Error::from(std::io::ErrorKind::AddrInUse);
+        assert!(try_port53_advisory("not-an-address", &err).is_none());
+    }
 }

src/testutil.rs

@@ -0,0 +1,95 @@
+use std::collections::{HashMap, HashSet};
+use std::net::{Ipv4Addr, SocketAddr};
+use std::path::PathBuf;
+use std::sync::{Mutex, RwLock};
+use std::time::Duration;
+
+use tokio::net::UdpSocket;
+
+use crate::blocklist::BlocklistStore;
+use crate::buffer::BytePacketBuffer;
+use crate::cache::DnsCache;
+use crate::config::UpstreamMode;
+use crate::ctx::ServerCtx;
+use crate::forward::{Upstream, UpstreamPool};
+use crate::health::HealthMeta;
+use crate::lan::PeerStore;
+use crate::override_store::OverrideStore;
+use crate::packet::DnsPacket;
+use crate::query_log::QueryLog;
+use crate::service_store::ServiceStore;
+use crate::srtt::SrttCache;
+use crate::stats::ServerStats;
+/// Minimal `ServerCtx` for tests. Override fields after construction
+/// (all fields are `pub`), then wrap in `Arc`.
+pub async fn test_ctx() -> ServerCtx {
+    let socket = UdpSocket::bind("127.0.0.1:0").await.unwrap();
+    ServerCtx {
+        socket,
+        zone_map: HashMap::new(),
+        cache: RwLock::new(DnsCache::new(100, 60, 86400)),
+        refreshing: Mutex::new(HashSet::new()),
+        stats: Mutex::new(ServerStats::new()),
+        overrides: RwLock::new(OverrideStore::new()),
+        blocklist: RwLock::new(BlocklistStore::new()),
+        query_log: Mutex::new(QueryLog::new(100)),
+        services: Mutex::new(ServiceStore::new()),
+        lan_peers: Mutex::new(PeerStore::new(90)),
+        forwarding_rules: Vec::new(),
+        upstream_pool: Mutex::new(UpstreamPool::new(
+            vec![Upstream::Udp("127.0.0.1:53".parse().unwrap())],
+            vec![],
+        )),
+        upstream_auto: false,
+        upstream_port: 53,
+        lan_ip: Mutex::new(Ipv4Addr::LOCALHOST),
+        timeout: Duration::from_millis(200),
+        hedge_delay: Duration::ZERO,
+        proxy_tld: "numa".to_string(),
+        proxy_tld_suffix: ".numa".to_string(),
+        lan_enabled: false,
+        config_path: "/tmp/test-numa.toml".to_string(),
+        config_found: false,
+        config_dir: PathBuf::from("/tmp"),
+        data_dir: PathBuf::from("/tmp"),
+        tls_config: None,
+        upstream_mode: UpstreamMode::Forward,
+        root_hints: Vec::new(),
+        srtt: RwLock::new(SrttCache::new(true)),
+        inflight: Mutex::new(HashMap::new()),
+        dnssec_enabled: false,
+        dnssec_strict: false,
+        health_meta: HealthMeta::test_fixture(),
+        ca_pem: None,
+        mobile_enabled: false,
+        mobile_port: 8765,
+    }
+}
+
+/// Spawn a UDP socket that replies to the first DNS query with the given
+/// response packet (patching the query ID to match). Returns the socket address.
+pub async fn mock_upstream(response: DnsPacket) -> SocketAddr {
+    let sock = UdpSocket::bind("127.0.0.1:0").await.unwrap();
+    let addr = sock.local_addr().unwrap();
+    tokio::spawn(async move {
+        let mut buf = [0u8; 512];
+        let (_, src) = sock.recv_from(&mut buf).await.unwrap();
+        let query_id = u16::from_be_bytes([buf[0], buf[1]]);
+        let mut resp = response;
+        resp.header.id = query_id;
+        let mut out = BytePacketBuffer::new();
+        resp.write(&mut out).unwrap();
+        sock.send_to(out.filled(), src).await.unwrap();
+    });
+    addr
+}
+
+/// UDP socket that accepts connections but never replies.
+/// Useful as an upstream that triggers timeouts.
+pub fn blackhole_upstream() -> SocketAddr {
+    let sock = std::net::UdpSocket::bind("127.0.0.1:0").unwrap();
+    let addr = sock.local_addr().unwrap();
+    // Leak so it stays bound for the duration of the test process.
+    Box::leak(Box::new(sock));
+    addr
+}

src/tls.rs

@@ -5,7 +5,9 @@ use std::sync::Arc;
 use log::{info, warn};
 
 use crate::ctx::ServerCtx;
-use rcgen::{BasicConstraints, CertificateParams, DnType, IsCa, KeyPair, KeyUsagePurpose, SanType};
+use rcgen::{
+    BasicConstraints, CertificateParams, DnType, IsCa, Issuer, KeyPair, KeyUsagePurpose, SanType,
+};
 use rustls::pki_types::{CertificateDer, PrivateKeyDer, PrivatePkcs8KeyDer};
 use rustls::ServerConfig;
 use time::{Duration, OffsetDateTime};
@@ -13,6 +15,13 @@ use time::{Duration, OffsetDateTime};
 const CA_VALIDITY_DAYS: i64 = 3650; // 10 years
 const CERT_VALIDITY_DAYS: i64 = 365; // 1 year
 
+/// Common Name on Numa's local CA. Referenced by trust-store helpers
+/// (`security`, `certutil`) when locating the cert for removal.
+pub const CA_COMMON_NAME: &str = "Numa Local CA";
+
+/// Filename of the CA certificate inside the data dir.
+pub const CA_FILE_NAME: &str = "ca.pem";
+
 /// Collect all service + LAN peer names and regenerate the TLS cert.
 pub fn regenerate_tls(ctx: &ServerCtx) {
     let tls = match &ctx.tls_config {
@@ -33,6 +42,41 @@ pub fn regenerate_tls(ctx: &ServerCtx) {
     }
 }
 
+/// Advisory for TLS-setup failures caused by a non-writable data dir;
+/// `None` if not applicable so the caller can fall back to the raw error.
+pub fn try_data_dir_advisory(err: &crate::Error, data_dir: &Path) -> Option<String> {
+    let io_err = err.downcast_ref::<std::io::Error>()?;
+    if io_err.kind() != std::io::ErrorKind::PermissionDenied {
+        return None;
+    }
+    let o = "\x1b[1;38;2;192;98;58m";
+    let r = "\x1b[0m";
+    Some(format!(
+        "
+{o}Numa{r} — HTTPS proxy disabled: cannot write TLS CA to {}.
+
+  The data directory is not writable by the current user. Numa needs
+  to persist a local Certificate Authority there to serve .numa over
+  HTTPS. DNS resolution and plain-HTTP proxy continue to work.
+
+  Fix — pick one:
+
+    1. Install Numa as the system resolver (sets up a writable data dir):
+
+         sudo numa install       (on Windows, run as Administrator)
+
+    2. Point data_dir at a path you can write.
+       Create {} with:
+
+         [server]
+         data_dir = \"/path/you/can/write\"
+
+",
+        data_dir.display(),
+        crate::suggested_config_path().display()
+    ))
+}
+
 /// Build a TLS config with a cert covering all provided service names.
 /// Wildcards under single-label TLDs (*.numa) are rejected by browsers,
 /// so we list each service explicitly as a SAN.
@@ -46,8 +90,8 @@ pub fn build_tls_config(
     alpn: Vec<Vec<u8>>,
     data_dir: &Path,
 ) -> crate::Result<Arc<ServerConfig>> {
-    let (ca_cert, ca_key) = ensure_ca(data_dir)?;
+    let (ca_der, issuer) = ensure_ca(data_dir)?;
-    let (cert_chain, key) = generate_service_cert(&ca_cert, &ca_key, tld, service_names)?;
+    let (cert_chain, key) = generate_service_cert(&ca_der, &issuer, tld, service_names)?;
 
     // Ensure a crypto provider is installed (rustls needs one)
     let _ = rustls::crypto::ring::default_provider().install_default();
@@ -65,18 +109,20 @@ pub fn build_tls_config(
     Ok(Arc::new(config))
 }
 
-fn ensure_ca(dir: &Path) -> crate::Result<(rcgen::Certificate, KeyPair)> {
+fn ensure_ca(dir: &Path) -> crate::Result<(CertificateDer<'static>, Issuer<'static, KeyPair>)> {
     let ca_key_path = dir.join("ca.key");
-    let ca_cert_path = dir.join("ca.pem");
+    let ca_cert_path = dir.join(CA_FILE_NAME);
 
     if ca_key_path.exists() && ca_cert_path.exists() {
         let key_pem = std::fs::read_to_string(&ca_key_path)?;
         let cert_pem = std::fs::read_to_string(&ca_cert_path)?;
         let key_pair = KeyPair::from_pem(&key_pem)?;
-        let params = CertificateParams::from_ca_cert_pem(&cert_pem)?;
+        let ca_der = rustls_pemfile::certs(&mut cert_pem.as_bytes())
-        let cert = params.self_signed(&key_pair)?;
+            .next()
+            .ok_or("empty CA PEM file")??;
+        let issuer = Issuer::from_ca_cert_der(&ca_der, key_pair)?;
         info!("loaded CA from {:?}", ca_cert_path);
-        return Ok((cert, key_pair));
+        return Ok((ca_der, issuer));
     }
 
     // Generate new CA
@@ -86,7 +132,7 @@ fn ensure_ca(dir: &Path) -> crate::Result<(rcgen::Certificate, KeyPair)> {
     let mut params = CertificateParams::default();
     params
         .distinguished_name
-        .push(DnType::CommonName, "Numa Local CA");
+        .push(DnType::CommonName, CA_COMMON_NAME);
     params.is_ca = IsCa::Ca(BasicConstraints::Unconstrained);
     params.key_usages = vec![KeyUsagePurpose::KeyCertSign, KeyUsagePurpose::CrlSign];
     params.not_before = OffsetDateTime::now_utc();
@@ -104,14 +150,16 @@ fn ensure_ca(dir: &Path) -> crate::Result<(rcgen::Certificate, KeyPair)> {
     }
 
     info!("generated CA at {:?}", ca_cert_path);
-    Ok((cert, key_pair))
+    let ca_der = cert.der().clone();
+    let issuer = Issuer::new(params, key_pair);
+    Ok((ca_der, issuer))
 }
 
 /// Generate a cert with explicit SANs for each service name.
 /// Always regenerated at startup (~5ms) — no disk caching needed.
 fn generate_service_cert(
-    ca_cert: &rcgen::Certificate,
+    ca_der: &CertificateDer<'static>,
-    ca_key: &KeyPair,
+    issuer: &Issuer<'_, KeyPair>,
     tld: &str,
     service_names: &[String],
 ) -> crate::Result<(Vec<CertificateDer<'static>>, PrivateKeyDer<'static>)> {
@@ -138,15 +186,26 @@ fn generate_service_cert(
         }
     }
 
-    if sans.is_empty() {
+    // Loopback IP SANs so browsers can reach DoH at https://127.0.0.1/dns-query
-        return Err("no valid service names for TLS cert".into());
+    sans.push(SanType::IpAddress(std::net::IpAddr::V4(
+        std::net::Ipv4Addr::LOCALHOST,
+    )));
+    sans.push(SanType::IpAddress(std::net::IpAddr::V6(
+        std::net::Ipv6Addr::LOCALHOST,
+    )));
+
+    for name in ["localhost", tld] {
+        match name.to_string().try_into() {
+            Ok(ia5) => sans.push(SanType::DnsName(ia5)),
+            Err(e) => warn!("invalid SAN {}: {}", name, e),
+        }
     }
 
     params.subject_alt_names = sans;
     params.not_before = OffsetDateTime::now_utc();
     params.not_after = OffsetDateTime::now_utc() + Duration::days(CERT_VALIDITY_DAYS);
 
-    let cert = params.signed_by(&key_pair, ca_cert, ca_key)?;
+    let cert = params.signed_by(&key_pair, issuer)?;
 
     info!(
         "generated TLS cert for: {}",
@@ -157,9 +216,107 @@ fn generate_service_cert(
             .join(", ")
     );
 
-    let cert_der = CertificateDer::from(cert.der().to_vec());
+    let cert_der = cert.der().clone();
-    let ca_der = CertificateDer::from(ca_cert.der().to_vec());
+    let ca_cert_der = ca_der.clone();
     let key_der = PrivateKeyDer::Pkcs8(PrivatePkcs8KeyDer::from(key_pair.serialize_der()));
 
-    Ok((vec![cert_der, ca_der], key_der))
+    Ok((vec![cert_der, ca_cert_der], key_der))
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::path::PathBuf;
+
+    #[test]
+    fn try_data_dir_advisory_permission_denied() {
+        let err: crate::Error =
+            Box::new(std::io::Error::from(std::io::ErrorKind::PermissionDenied));
+        let path = PathBuf::from("/usr/local/var/numa");
+        let msg = try_data_dir_advisory(&err, &path).expect("should advise");
+        assert!(msg.contains("HTTPS proxy disabled"));
+        assert!(msg.contains("/usr/local/var/numa"));
+        assert!(msg.contains("numa install"));
+        assert!(msg.contains("data_dir"));
+    }
+
+    #[test]
+    fn try_data_dir_advisory_skips_other_io_kinds() {
+        let err: crate::Error = Box::new(std::io::Error::from(std::io::ErrorKind::NotFound));
+        assert!(try_data_dir_advisory(&err, &PathBuf::from("/x")).is_none());
+    }
+
+    #[test]
+    fn try_data_dir_advisory_skips_non_io_errors() {
+        let err: crate::Error = "rcgen failure".into();
+        assert!(try_data_dir_advisory(&err, &PathBuf::from("/x")).is_none());
+    }
+
+    #[test]
+    fn service_cert_contains_expected_sans() {
+        use x509_parser::prelude::GeneralName;
+
+        let dir = std::env::temp_dir().join(format!("numa-test-san-{}", std::process::id()));
+        let _ = std::fs::remove_dir_all(&dir);
+        let (ca_der, issuer) = ensure_ca(&dir).unwrap();
+
+        let names = vec!["grafana".into(), "router".into()];
+        let (chain, _) = generate_service_cert(&ca_der, &issuer, "numa", &names).unwrap();
+        assert_eq!(chain.len(), 2, "chain should be [leaf, CA]");
+
+        let (_, cert) = x509_parser::parse_x509_certificate(chain[0].as_ref()).unwrap();
+        let san = cert
+            .tbs_certificate
+            .subject_alternative_name()
+            .unwrap()
+            .unwrap();
+
+        let dns: Vec<&str> = san
+            .value
+            .general_names
+            .iter()
+            .filter_map(|gn| match gn {
+                GeneralName::DNSName(s) => Some(*s),
+                _ => None,
+            })
+            .collect();
+
+        let ips: Vec<std::net::IpAddr> = san
+            .value
+            .general_names
+            .iter()
+            .filter_map(|gn| match gn {
+                GeneralName::IPAddress(b) => match b.len() {
+                    4 => Some(std::net::IpAddr::V4(std::net::Ipv4Addr::new(
+                        b[0], b[1], b[2], b[3],
+                    ))),
+                    16 => {
+                        let a: [u8; 16] = (*b).try_into().unwrap();
+                        Some(std::net::IpAddr::V6(std::net::Ipv6Addr::from(a)))
+                    }
+                    _ => None,
+                },
+                _ => None,
+            })
+            .collect();
+
+        // DNS SANs
+        assert!(dns.contains(&"*.numa"), "missing wildcard SAN");
+        assert!(dns.contains(&"grafana.numa"), "missing service SAN");
+        assert!(dns.contains(&"router.numa"), "missing service SAN");
+        assert!(dns.contains(&"localhost"), "missing localhost SAN");
+        assert!(dns.contains(&"numa"), "missing bare TLD SAN");
+
+        // IP SANs
+        assert!(
+            ips.contains(&std::net::IpAddr::V4(std::net::Ipv4Addr::LOCALHOST)),
+            "missing 127.0.0.1 SAN"
+        );
+        assert!(
+            ips.contains(&std::net::IpAddr::V6(std::net::Ipv6Addr::LOCALHOST)),
+            "missing ::1 SAN"
+        );
+
+        let _ = std::fs::remove_dir_all(&dir);
+    }
 }

tests/docker/hold53.py

@@ -0,0 +1,5 @@
+import socket, signal
+s = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
+s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 0)
+s.bind(("", 53))
+signal.pause()

tests/docker/install-trust.sh

@@ -0,0 +1,123 @@
+#!/usr/bin/env bash
+#
+# Cross-distro CA trust contract test for issue #35.
+#
+# Runs the exact shell commands `src/system_dns.rs::trust_ca_linux` would run
+# on each Linux trust-store family (Debian, Fedora pki, Arch p11-kit), and
+# asserts the certificate ends up in (and is removed from) the system bundle.
+#
+# This is a contract test, not an integration test: it doesn't drive the Rust
+# code (that would need systemd-in-container). It verifies the assumptions in
+# `LINUX_TRUST_STORES` against the real distro behavior. If you change that
+# table in src/system_dns.rs, update the per-distro cases below to match.
+#
+# Requirements: docker, openssl (host).
+# Usage:        ./tests/docker/install-trust.sh
+
+set -euo pipefail
+
+cd "$(dirname "$0")/../.."
+
+GREEN="\033[32m"; RED="\033[31m"; RESET="\033[0m"
+
+# Self-signed CA fixture, mounted into each container as ca.pem.
+# basicConstraints=CA:TRUE is required — without it, Debian's
+# update-ca-certificates silently skips the cert during bundle build.
+FIXTURE_DIR=$(mktemp -d)
+trap 'rm -rf "$FIXTURE_DIR"' EXIT
+openssl req -x509 -newkey rsa:2048 -nodes -days 1 \
+    -keyout "$FIXTURE_DIR/ca.key" \
+    -out    "$FIXTURE_DIR/ca.pem" \
+    -subj   "/CN=Numa Local CA Test $(date +%s)" \
+    -addext "basicConstraints=critical,CA:TRUE" \
+    -addext "keyUsage=critical,keyCertSign,cRLSign" >/dev/null 2>&1
+
+# Distro bundles store certs differently — Debian writes raw PEM only,
+# Fedora prepends "# CN" comment headers, Arch via extract-compat is
+# raw PEM. To detect cert presence uniformly we grep for a deterministic
+# substring of the base64 body (first base64 line is unique per cert).
+CERT_TAG=$(sed -n '2p' "$FIXTURE_DIR/ca.pem")
+
+PASSED=0; FAILED=0
+
+run_case() {
+    local distro="$1"; shift
+    local image="$1"; shift
+    local platform="$1"; shift
+    local script="$1"
+
+    printf "── %s (%s) ──\n" "$distro" "$image"
+    if docker run --rm \
+        --platform "$platform" \
+        --security-opt seccomp=unconfined \
+        -e CERT_TAG="$CERT_TAG" \
+        -e DEBIAN_FRONTEND=noninteractive \
+        -v "$FIXTURE_DIR/ca.pem:/fixture/ca.pem:ro" \
+        "$image" bash -c "$script"; then
+        printf "${GREEN}✓${RESET} %s\n\n" "$distro"
+        PASSED=$((PASSED + 1))
+    else
+        printf "${RED}✗${RESET} %s\n\n" "$distro"
+        FAILED=$((FAILED + 1))
+    fi
+}
+
+# Debian / Ubuntu / Mint — anchor: /usr/local/share/ca-certificates/*.crt
+run_case "debian" "debian:stable" "linux/amd64" '
+    set -e
+    apt-get update -qq
+    apt-get install -qq -y ca-certificates >/dev/null
+    install -m 0644 /fixture/ca.pem /usr/local/share/ca-certificates/numa-local-ca.crt
+    update-ca-certificates >/dev/null 2>&1
+    grep -q "$CERT_TAG" /etc/ssl/certs/ca-certificates.crt
+    echo "  install: cert present in bundle"
+    rm /usr/local/share/ca-certificates/numa-local-ca.crt
+    update-ca-certificates --fresh >/dev/null 2>&1
+    if grep -q "$CERT_TAG" /etc/ssl/certs/ca-certificates.crt; then
+        echo "  uninstall: cert STILL present (regression)" >&2
+        exit 1
+    fi
+    echo "  uninstall: cert removed from bundle"
+'
+
+# Fedora / RHEL / CentOS / SUSE — anchor: /etc/pki/ca-trust/source/anchors/*.pem
+run_case "fedora" "fedora:latest" "linux/amd64" '
+    set -e
+    dnf install -q -y ca-certificates >/dev/null
+    install -m 0644 /fixture/ca.pem /etc/pki/ca-trust/source/anchors/numa-local-ca.pem
+    update-ca-trust extract
+    grep -q "$CERT_TAG" /etc/pki/ca-trust/extracted/pem/tls-ca-bundle.pem
+    echo "  install: cert present in bundle"
+    rm /etc/pki/ca-trust/source/anchors/numa-local-ca.pem
+    update-ca-trust extract
+    if grep -q "$CERT_TAG" /etc/pki/ca-trust/extracted/pem/tls-ca-bundle.pem; then
+        echo "  uninstall: cert STILL present (regression)" >&2
+        exit 1
+    fi
+    echo "  uninstall: cert removed from bundle"
+'
+
+# Arch / Manjaro — anchor: /etc/ca-certificates/trust-source/anchors/*.pem
+# archlinux:latest is x86_64-only; --platform forces emulation on Apple Silicon.
+run_case "arch" "archlinux:latest" "linux/amd64" '
+    set -e
+    # pacman 7+ filters syscalls in its own sandbox; disable for Rosetta/qemu emulation.
+    sed -i "s/^#DisableSandboxSyscalls/DisableSandboxSyscalls/" /etc/pacman.conf
+    pacman -Sy --noconfirm --needed ca-certificates p11-kit >/dev/null 2>&1
+    install -m 0644 /fixture/ca.pem /etc/ca-certificates/trust-source/anchors/numa-local-ca.pem
+    trust extract-compat
+    grep -q "$CERT_TAG" /etc/ssl/certs/ca-certificates.crt
+    echo "  install: cert present in bundle"
+    rm /etc/ca-certificates/trust-source/anchors/numa-local-ca.pem
+    trust extract-compat
+    if grep -q "$CERT_TAG" /etc/ssl/certs/ca-certificates.crt; then
+        echo "  uninstall: cert STILL present (regression)" >&2
+        exit 1
+    fi
+    echo "  uninstall: cert removed from bundle"
+'
+
+printf "── summary ──\n"
+printf "  ${GREEN}passed${RESET}: %d\n" "$PASSED"
+printf "  ${RED}failed${RESET}: %d\n" "$FAILED"
+[ "$FAILED" -eq 0 ]

tests/docker/issue-81.sh

@@ -0,0 +1,164 @@
+#!/usr/bin/env bash
+#
+# End-to-end validation of the issue #81 fix (config path advisory).
+#
+# Builds numa from two source trees — the buggy baseline and the fix
+# candidate — inside one debian:bookworm container, then runs four
+# scenarios to prove:
+#
+#   1. replication/main  — reporter's sequence, bug confirmed
+#   2. replication/fix   — reporter's sequence, bug is gone
+#   3. existing/main     — pre-installed config at FHS data dir still loads
+#   4. existing/fix      — same, unchanged by the fix (no regression)
+#
+# Scenarios 3 and 4 guard against the fear that the fix might change
+# candidate order and break existing daemon installs (including the
+# macOS Homebrew-prefix layout at /usr/local/var/numa/).
+#
+# Usage:
+#   MAIN_SRC=/path/to/main-checkout FIX_SRC=/path/to/fix-worktree \
+#     ./tests/docker/issue-81.sh
+#
+# Defaults: MAIN_SRC = $(git rev-parse --show-toplevel), FIX_SRC = same.
+
+set -euo pipefail
+
+MAIN_SRC="${MAIN_SRC:-$(git rev-parse --show-toplevel)}"
+FIX_SRC="${FIX_SRC:-$MAIN_SRC}"
+
+GREEN="\033[32m"; RED="\033[31m"; RESET="\033[0m"
+
+echo "── issue #81 validation ──"
+echo "  main: $MAIN_SRC"
+echo "  fix:  $FIX_SRC"
+echo
+
+docker run --rm \
+    --platform linux/amd64 \
+    -v "$MAIN_SRC:/main:ro" \
+    -v "$FIX_SRC:/fix:ro" \
+    -v "$(dirname "$0")/hold53.py:/tmp/hold53.py:ro" \
+    -v numa-port53-cargo:/root/.cargo \
+    -v numa-port53-target:/work/target \
+    debian:bookworm bash -c '
+set -euo pipefail
+
+# Paths and ports used by all scenarios — keep in one place so the
+# heredocs and the verdict greps cannot drift.
+XDG_CONFIG="/root/.config/numa/numa.toml"
+FHS_CONFIG="/var/lib/numa/numa.toml"
+TEST_PORT="5354"
+TEST_API_PORT="5380"
+
+apt-get update -qq && apt-get install -y -qq curl build-essential python3 2>&1 | tail -1
+if ! command -v cargo &>/dev/null; then
+    curl -sSf https://sh.rustup.rs | sh -s -- -y --profile minimal --quiet
+fi
+. "$HOME/.cargo/env"
+
+build_from() {
+    local label="$1"; local src="$2"
+    mkdir -p "/work/$label"
+    tar -C "$src" --exclude=./target --exclude=./.git -cf - . | tar -C "/work/$label" -xf -
+    (cd "/work/$label" && cargo build --release --locked 2>&1 | tail -1)
+    cp "/work/$label/target/release/numa" "/work/numa-$label"
+}
+
+build_from main /main
+build_from fix /fix
+
+holder=0
+stop_holder() {
+    if [ "$holder" -ne 0 ]; then
+        kill "$holder" 2>/dev/null || true
+        wait "$holder" 2>/dev/null || true
+        holder=0
+    fi
+}
+trap stop_holder EXIT
+
+start_holder() {
+    python3 /tmp/hold53.py &
+    holder=$!
+    sleep 0.3
+}
+
+write_test_config() {
+    local path="$1"
+    mkdir -p "$(dirname "$path")"
+    cat > "$path" <<EOF
+[server]
+bind_addr = "127.0.0.1:$TEST_PORT"
+api_port = $TEST_API_PORT
+EOF
+}
+
+verdict() {
+    local label="$1"; local expected="$2"; local file="$3"
+    # "cannot bind to" is printed by the advisory when numa fails to start.
+    # Its absence is a reliable proxy for "numa bound successfully" because
+    # the banner-only log we capture contains no other failure surface.
+    if grep -q "cannot bind to" "$file"; then
+        echo "  [$label] did not bind $TEST_PORT — numa ignored the XDG config"
+        [ "$expected" = "ignored" ] && return 0 || return 1
+    else
+        echo "  [$label] bound $TEST_PORT — config loaded"
+        [ "$expected" = "bound" ] && return 0 || return 1
+    fi
+}
+
+scenario_replication() {
+    local label="$1"; local bin="/work/numa-$label"; local expected="$2"
+    echo
+    echo "════════ REPLICATION / $label ════════"
+    rm -rf /root/.config/numa /var/lib/numa
+    mkdir -p "$(dirname "$XDG_CONFIG")"
+
+    start_holder
+    set +e
+    timeout 5 "$bin" > /tmp/run1.txt 2>&1
+    set -e
+    echo "── step 1: advisory printed by $label ──"
+    grep -E "Create .* with:" /tmp/run1.txt | sed "s/^/  /" || echo "  <no advisory line>"
+
+    write_test_config "$XDG_CONFIG"
+    echo "── step 2: wrote config at $XDG_CONFIG ──"
+
+    set +e
+    timeout 3 "$bin" > /tmp/run2.txt 2>&1
+    set -e
+    stop_holder
+
+    verdict "$label" "$expected" /tmp/run2.txt
+}
+
+scenario_existing_install() {
+    local label="$1"; local bin="/work/numa-$label"
+    echo
+    echo "════════ EXISTING INSTALL / $label ════════"
+    rm -rf /root/.config/numa /var/lib/numa
+    write_test_config "$FHS_CONFIG"
+
+    start_holder
+    set +e
+    timeout 3 "$bin" > /tmp/run.txt 2>&1
+    set -e
+    stop_holder
+
+    verdict "$label" "bound" /tmp/run.txt
+}
+
+RC=0
+scenario_replication main ignored || RC=1
+scenario_replication fix bound || RC=1
+scenario_existing_install main || RC=1
+scenario_existing_install fix || RC=1
+
+echo
+if [ "$RC" -eq 0 ]; then
+    echo "── all scenarios matched expectations ──"
+else
+    echo "── FAILURE: one or more scenarios diverged ──"
+fi
+exit $RC
+'

tests/docker/smoke-arch.sh

@@ -0,0 +1,147 @@
+#!/usr/bin/env bash
+#
+# Arch Linux compatibility smoke test.
+#
+# Builds numa from source inside an archlinux:latest container, runs it
+# in forward mode on port 5354, and verifies a single DNS query returns
+# an A record. Validates the "Arch compatible" claim end-to-end before
+# release announcements.
+#
+# Dogfooding: the test numa forwards to the host's running numa via
+# host.docker.internal (Docker Desktop's host gateway). This avoids the
+# Docker NAT/UDP issues with public resolvers and exercises the realistic
+# numa-on-numa shape. Requires the host to be running numa on port 53.
+#
+# First run is slow (~8-12 min): image pull + pacman + cold cargo build.
+# No caching across runs.
+#
+# Requirements: docker, host running numa on 0.0.0.0:53
+# Usage:        ./tests/docker/smoke-arch.sh
+
+set -euo pipefail
+
+cd "$(dirname "$0")/../.."
+
+GREEN="\033[32m"; RED="\033[31m"; RESET="\033[0m"
+
+# Precondition: the test numa-on-arch forwards to the host numa as its
+# upstream (dogfood pattern). Fail fast with a clear error if there is
+# no working DNS on the host, rather than letting the dig inside the
+# container time out with "deadline has elapsed".
+if ! dig @127.0.0.1 google.com A +short +time=1 +tries=1 >/dev/null 2>&1; then
+    printf "${RED}error:${RESET} host numa is not answering on 127.0.0.1:53\n" >&2
+    echo "  This test forwards to the host numa via host.docker.internal." >&2
+    echo "  Start numa on the host first (sudo numa install), then rerun." >&2
+    exit 1
+fi
+
+echo "── building + running numa on archlinux:latest ──"
+echo "  (first run is slow: image pull + pacman + cold cargo build, ~8-12 min)"
+echo
+
+docker run --rm \
+    --platform linux/amd64 \
+    --security-opt seccomp=unconfined \
+    -v "$PWD:/src:ro" \
+    -v numa-arch-cargo:/root/.cargo \
+    -v numa-arch-target:/work/target \
+    archlinux:latest bash -c '
+    set -e
+
+    # pacman 7+ filters syscalls in its own sandbox; disable for Rosetta/qemu
+    sed -i "s/^#DisableSandboxSyscalls/DisableSandboxSyscalls/" /etc/pacman.conf
+
+    echo "── pacman: installing build + runtime deps ──"
+    pacman -Sy --noconfirm --needed rust gcc pkgconf cmake make perl bind 2>&1 | tail -3
+    echo
+
+    # Copy source to a writable workdir, skipping target/ + .git so we
+    # do not pull in the host (macOS) build artifacts.
+    mkdir -p /work
+    tar -C /src --exclude=./target --exclude=./.git -cf - . | tar -C /work -xf -
+    cd /work
+
+    echo "── cargo build --release --locked ──"
+    cargo build --release --locked 2>&1 | tail -5
+    echo
+
+    # Dogfood: forward to the host numa via host.docker.internal.
+    # numa parses upstream.address as a literal SocketAddr, so we resolve
+    # the hostname to an IPv4 address first (force v4 — getent hosts may
+    # return IPv6 first, and IPv6 addresses need bracketed addr:port form).
+    HOST_IP=$(getent ahostsv4 host.docker.internal | awk "/STREAM/ {print \$1; exit}")
+    if [ -z "$HOST_IP" ]; then
+        echo "  ✗ could not resolve host.docker.internal to IPv4 (not on Docker Desktop?)"
+        exit 1
+    fi
+    echo "── starting numa on :5354 (forward to host numa at $HOST_IP:53) ──"
+    # Intentionally NOT setting [server] data_dir — we want to exercise the
+    # default code path (data_dir() → daemon_data_dir() → /var/lib/numa) so
+    # the FHS-path assertion below verifies the live wiring, not just the
+    # unit-tested helper.
+    cat > /tmp/numa.toml <<EOF
+[server]
+bind_addr = "127.0.0.1:5354"
+api_port = 5381
+
+[upstream]
+mode = "forward"
+address = "$HOST_IP"
+port = 53
+EOF
+
+    ./target/release/numa /tmp/numa.toml > /tmp/numa.log 2>&1 &
+    NUMA_PID=$!
+
+    # Poll for readiness — numa is ready when it answers a query
+    READY=0
+    for i in 1 2 3 4 5 6 7 8; do
+        sleep 1
+        if dig @127.0.0.1 -p 5354 google.com A +short +time=1 +tries=1 2>/dev/null \
+            | grep -qE "^[0-9]+\.[0-9]+\.[0-9]+\.[0-9]+$"; then
+            READY=1
+            break
+        fi
+    done
+
+    if [ "$READY" -ne 1 ]; then
+        echo "  ✗ numa did not return an A record after 8s"
+        echo "  numa log:"
+        cat /tmp/numa.log
+        kill $NUMA_PID 2>/dev/null || true
+        exit 1
+    fi
+
+    echo "── dig @127.0.0.1 -p 5354 google.com A ──"
+    ANSWER=$(dig @127.0.0.1 -p 5354 google.com A +short +time=2 +tries=1)
+    echo "$ANSWER" | sed "s/^/  /"
+
+    kill $NUMA_PID 2>/dev/null || true
+
+    # FHS path assertion: the default data dir on Linux must be /var/lib/numa
+    # (not the legacy /usr/local/var/numa). The CA cert generated at startup
+    # is the canonical proof that numa wrote to the right place.
+    echo
+    echo "── FHS path check ──"
+    if [ -f /var/lib/numa/ca.pem ]; then
+        echo "  ✓ CA cert at /var/lib/numa/ca.pem (FHS path)"
+    else
+        echo "  ✗ CA cert NOT at /var/lib/numa/ca.pem"
+        echo "  ls /var/lib/numa/:"
+        ls -la /var/lib/numa/ 2>&1 | sed "s/^/    /"
+        echo "  ls /usr/local/var/numa/:"
+        ls -la /usr/local/var/numa/ 2>&1 | sed "s/^/    /"
+        exit 1
+    fi
+    if [ -e /usr/local/var/numa ]; then
+        echo "  ✗ legacy path /usr/local/var/numa unexpectedly exists on a fresh container"
+        exit 1
+    fi
+    echo "  ✓ legacy path /usr/local/var/numa absent (fresh install used FHS)"
+
+    echo
+    echo "  ✓ numa built, ran, answered a forward query, and used the FHS data dir on Arch"
+'
+
+echo
+printf "${GREEN}── smoke-arch passed ──${RESET}\n"

tests/docker/smoke-port53.sh

@@ -0,0 +1,138 @@
+#!/usr/bin/env bash
+#
+# Port-53 conflict advisory integration test.
+#
+# Builds numa from source inside a debian:bookworm container, pre-binds
+# port 53 with a UDP socket, then runs numa bare (default bind_addr
+# 0.0.0.0:53). Verifies:
+#   - process exits with code 1
+#   - stderr contains the advisory ("cannot bind to")
+#   - stderr contains both fix suggestions ("numa install", "bind_addr")
+#
+# This is the end-to-end test for the fix in:
+#   src/main.rs — AddrInUse match arm → eprint advisory + process::exit(1)
+#
+# No systemd-resolved needed — the conflict is simulated by a Python
+# UDP socket held open before numa starts.
+#
+# Requirements: docker
+# Usage:        ./tests/docker/smoke-port53.sh
+
+set -euo pipefail
+
+cd "$(dirname "$0")/../.."
+
+GREEN="\033[32m"; RED="\033[31m"; RESET="\033[0m"
+
+pass() { printf "  ${GREEN}✓${RESET} %s\n" "$1"; }
+fail() { printf "  ${RED}✗${RESET} %s\n" "$1"; printf "    %s\n" "$2"; FAILED=$((FAILED+1)); }
+FAILED=0
+
+echo "── smoke-port53: building + testing numa on debian:bookworm ──"
+echo "  (first run is slow: image pull + cold cargo build, ~5-8 min)"
+echo
+
+OUTPUT=$(docker run --rm \
+    --platform linux/amd64 \
+    -v "$PWD:/src:ro" \
+    -v numa-port53-cargo:/root/.cargo \
+    -v numa-port53-target:/work/target \
+    debian:bookworm bash -c '
+set -e
+
+apt-get update -qq && apt-get install -y -qq curl build-essential python3 2>&1 | tail -3
+
+# Install rustup if not already in the cargo cache volume
+if ! command -v cargo &>/dev/null; then
+    curl -sSf https://sh.rustup.rs | sh -s -- -y --profile minimal --quiet
+fi
+. "$HOME/.cargo/env"
+
+# Copy source to a writable workdir
+mkdir -p /work
+tar -C /src --exclude=./target --exclude=./.git -cf - . | tar -C /work -xf -
+cd /work
+
+echo "── cargo build --release --locked ──"
+cargo build --release --locked 2>&1 | tail -5
+echo
+
+# Write the holder script to a file to avoid quoting hell.
+# Holds port 53 until killed — no sleep race.
+cat > /tmp/hold53.py << '"'"'PYEOF'"'"'
+import socket, signal
+s = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
+s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 0)
+s.bind(("", 53))
+signal.pause()
+PYEOF
+
+python3 /tmp/hold53.py &
+HOLDER_PID=$!
+
+# Verify the holder is actually up before proceeding
+sleep 0.3
+if ! kill -0 $HOLDER_PID 2>/dev/null; then
+    echo "holder_failed=1"
+    exit 1
+fi
+
+echo "── running numa with port 53 already bound ──"
+# timeout 5: guards against numa not exiting (advisory not fired, bug present)
+# Capture stderr to a file so the exit code is not clobbered by || or $()
+set +e
+timeout 5 ./target/release/numa > /tmp/numa-stderr.txt 2>&1
+EXIT_CODE=$?
+set -e
+STDERR=$(cat /tmp/numa-stderr.txt)
+
+kill $HOLDER_PID 2>/dev/null || true
+
+echo "exit_code=$EXIT_CODE"
+printf "%s" "$STDERR" | sed "s/^/  numa: /"
+' 2>&1)
+
+echo "$OUTPUT"
+
+echo
+echo "── assertions ──"
+
+if echo "$OUTPUT" | grep -q "holder_failed=1"; then
+    echo "  SETUP FAILED: could not pre-bind port 53 inside container"
+    exit 1
+fi
+
+EXIT_CODE=$(echo "$OUTPUT" | grep '^exit_code=' | cut -d= -f2)
+
+if [ "${EXIT_CODE:-}" = "1" ]; then
+    pass "exits with code 1"
+else
+    fail "exits with code 1" "got: exit_code=${EXIT_CODE:-<missing>}"
+fi
+
+if echo "$OUTPUT" | grep -q "cannot bind to"; then
+    pass "advisory printed to stderr"
+else
+    fail "advisory printed to stderr" "stderr did not contain 'cannot bind to'"
+fi
+
+if echo "$OUTPUT" | grep -q "numa install"; then
+    pass "advisory offers 'sudo numa install'"
+else
+    fail "advisory offers 'sudo numa install'" "not found in output"
+fi
+
+if echo "$OUTPUT" | grep -q "bind_addr"; then
+    pass "advisory offers non-privileged port alternative"
+else
+    fail "advisory offers non-privileged port alternative" "'bind_addr' not found in output"
+fi
+
+echo
+if [ "$FAILED" -eq 0 ]; then
+    printf "${GREEN}── smoke-port53 passed ──${RESET}\n"
+    exit 0
+else
+    printf "${RED}── smoke-port53 failed ($FAILED assertion(s)) ──${RESET}\n"
+    exit 1
+fi

tests/integration.sh

@@ -53,7 +53,17 @@ CONF
     echo "Starting Numa on :$PORT ($SUITE_NAME)..."
     RUST_LOG=info "$BINARY" "$CONFIG" > "$LOG" 2>&1 &
     NUMA_PID=$!
-    sleep 4
+    sleep 2
+
+    # Wait for blocklist to load (if blocking is enabled in this suite)
+    if echo "$SUITE_CONFIG" | grep -q 'enabled = true'; then
+        for i in $(seq 1 20); do
+            LOADED=$(curl -sf http://127.0.0.1:$API_PORT/blocking/stats 2>/dev/null \
+                | grep -o '"domains_loaded":[0-9]*' | cut -d: -f2)
+            if [ "${LOADED:-0}" -gt 0 ]; then break; fi
+            sleep 1
+        done
+    fi
 
     if ! kill -0 "$NUMA_PID" 2>/dev/null; then
         echo "Failed to start Numa:"
@@ -622,6 +632,54 @@ CONF
             "10.0.0.1" \
             "$($KDIG +short dot-test.example A 2>/dev/null)"
 
+        echo ""
+        echo "=== DNS-over-HTTPS (RFC 8484) ==="
+
+        DOH_QUERY_FILE=/tmp/numa-doh-query.bin
+        DOH_RESP_FILE=/tmp/numa-doh-resp.bin
+
+        # Build DNS wire-format query for dot-test.example A
+        printf '\x00\x01\x01\x00\x00\x01\x00\x00\x00\x00\x00\x00\x08dot-test\x07example\x00\x00\x01\x00\x01' > "$DOH_QUERY_FILE"
+
+        # POST valid DoH query
+        DOH_CODE=$(curl -sk -X POST \
+            --resolve "numa.numa:$PROXY_HTTPS_PORT:127.0.0.1" \
+            -H "Content-Type: application/dns-message" \
+            --data-binary @"$DOH_QUERY_FILE" \
+            --cacert "$CA" \
+            -o "$DOH_RESP_FILE" \
+            -w "%{http_code}" \
+            "https://numa.numa:$PROXY_HTTPS_PORT/dns-query")
+        check "DoH POST returns HTTP 200" "200" "$DOH_CODE"
+
+        # Check response contains IP 10.0.0.1 (hex: 0a000001)
+        DOH_HEX=$(xxd -p "$DOH_RESP_FILE" | tr -d '\n')
+        if echo "$DOH_HEX" | grep -q "0a000001"; then
+            check "DoH response resolves dot-test.example → 10.0.0.1" "found" "found"
+        else
+            check "DoH response resolves dot-test.example → 10.0.0.1" "0a000001" "$DOH_HEX"
+        fi
+
+        # Wrong Content-Type → 415
+        DOH_CT_CODE=$(curl -sk -X POST \
+            -H "Host: numa.numa" \
+            -H "Content-Type: text/plain" \
+            --data-binary @"$DOH_QUERY_FILE" \
+            -o /dev/null -w "%{http_code}" \
+            "https://127.0.0.1:$PROXY_HTTPS_PORT/dns-query")
+        check "DoH wrong Content-Type → 415" "415" "$DOH_CT_CODE"
+
+        # Wrong host → 404 (DoH only serves numa.numa)
+        DOH_HOST_CODE=$(curl -sk -X POST \
+            -H "Host: foo.numa" \
+            -H "Content-Type: application/dns-message" \
+            --data-binary @"$DOH_QUERY_FILE" \
+            -o /dev/null -w "%{http_code}" \
+            "https://127.0.0.1:$PROXY_HTTPS_PORT/dns-query")
+        check "DoH wrong host → 404" "404" "$DOH_HOST_CODE"
+
+        rm -f "$DOH_QUERY_FILE" "$DOH_RESP_FILE"
+
         echo ""
         echo "=== Proxy TLS works with DoT enabled ==="
 

tests/manual/install-trust-macos.sh

@@ -0,0 +1,94 @@
+#!/usr/bin/env bash
+#
+# Manual macOS CA trust contract test.
+#
+# Mirrors src/system_dns.rs::trust_ca_macos / untrust_ca_macos by running
+# the same `security` shell commands against a fixture cert with a unique
+# CN. Safe to run alongside a production numa install:
+#
+#   - Test cert CN = "Numa Local CA Test <pid-ts>", always strictly longer
+#     than the production CN "Numa Local CA". `security find-certificate -c`
+#     does substring matching, so the test's search for $TEST_CN can never
+#     match the production cert (the search term is longer than the prod CN).
+#   - All deletes use `delete-certificate -Z <hash>`, which only touches the
+#     cert with that exact hash. Production and test certs have different
+#     hashes by construction (different key material), so the delete cannot
+#     reach the production cert even if a CN search somehow returned both.
+#
+# Mutates the System keychain (briefly). Cleans up on success or interrupt.
+# Requires sudo for `security add-trusted-cert` and `delete-certificate`.
+#
+# Usage: ./tests/manual/install-trust-macos.sh
+
+set -euo pipefail
+
+if [[ "$OSTYPE" != darwin* ]]; then
+    echo "This test is macOS-only." >&2
+    exit 1
+fi
+
+GREEN="\033[32m"; RED="\033[31m"; RESET="\033[0m"
+
+# Production constant from src/tls.rs::CA_COMMON_NAME — keep in sync.
+PROD_CN="Numa Local CA"
+KEYCHAIN="/Library/Keychains/System.keychain"
+
+# Notice if production numa is already installed. We proceed regardless —
+# see header for why coexistence is safe (unique CN + by-hash deletion).
+if security find-certificate -c "$PROD_CN" "$KEYCHAIN" >/dev/null 2>&1; then
+    echo "  note: production '$PROD_CN' detected — proceeding alongside (test cert can't touch it)"
+    echo
+fi
+
+# Unique CN ensures the test cert can never collide with production.
+TEST_CN="Numa Local CA Test $$-$(date +%s)"
+FIXTURE_DIR=$(mktemp -d)
+
+cleanup() {
+    # Best-effort: remove any test certs by hash if still present.
+    if security find-certificate -c "$TEST_CN" "$KEYCHAIN" >/dev/null 2>&1; then
+        echo "  cleanup: removing leftover test cert"
+        security find-certificate -c "$TEST_CN" -a -Z "$KEYCHAIN" 2>/dev/null \
+            | awk '/^SHA-1 hash:/ {print $NF}' \
+            | while read -r hash; do
+                sudo security delete-certificate -Z "$hash" "$KEYCHAIN" >/dev/null 2>&1 || true
+            done
+    fi
+    rm -rf "$FIXTURE_DIR"
+}
+trap cleanup EXIT
+
+echo "── generating fixture CA ──"
+openssl req -x509 -newkey rsa:2048 -nodes -days 1 \
+    -keyout "$FIXTURE_DIR/ca.key" \
+    -out    "$FIXTURE_DIR/ca.pem" \
+    -subj   "/CN=$TEST_CN" \
+    -addext "basicConstraints=critical,CA:TRUE" \
+    -addext "keyUsage=critical,keyCertSign,cRLSign" >/dev/null 2>&1
+echo "  CN: $TEST_CN"
+echo
+
+echo "── trust step (mirrors trust_ca_macos) ──"
+sudo security add-trusted-cert -d -r trustRoot -k "$KEYCHAIN" "$FIXTURE_DIR/ca.pem"
+if security find-certificate -c "$TEST_CN" "$KEYCHAIN" >/dev/null 2>&1; then
+    printf "  ${GREEN}✓${RESET} test cert found in keychain\n"
+else
+    printf "  ${RED}✗${RESET} test cert NOT found after add-trusted-cert\n"
+    exit 1
+fi
+echo
+
+echo "── untrust step (mirrors untrust_ca_macos) ──"
+security find-certificate -c "$TEST_CN" -a -Z "$KEYCHAIN" 2>/dev/null \
+    | awk '/^SHA-1 hash:/ {print $NF}' \
+    | while read -r hash; do
+        sudo security delete-certificate -Z "$hash" "$KEYCHAIN" >/dev/null
+    done
+if security find-certificate -c "$TEST_CN" "$KEYCHAIN" >/dev/null 2>&1; then
+    printf "  ${RED}✗${RESET} test cert STILL present after delete (regression)\n"
+    exit 1
+fi
+printf "  ${GREEN}✓${RESET} test cert removed from keychain\n"
+echo
+
+printf "${GREEN}all checks passed${RESET}\n"