TUN-9800: Migrate apt internal builds to Gitlab

* TUN-9863: Introduce Code Signing for Windows Builds

This commit adds a signing step to the build script for windows binaries.
Since we package the MSI on Linux, this commit adds another CI step that depends on package-windows and signs all of the windows packages.

To do so, we use azuresigntool which relies on a certificate stored in Azure Vault.

Closes TUN-9863

.ci/apt-internal.gitlab-ci.yml

@@ -0,0 +1,151 @@
+.register_inputs: &register_inputs
+  stage: release-internal
+  runOnBranches: "^master$"
+  COMPONENT: "common"
+
+.register_inputs_stable_bookworm: &register_inputs_stable_bookworm
+  <<: *register_inputs
+  runOnChangesTo: ['RELEASE_NOTES']
+  FLAVOR: "bookworm"
+  SERIES: "stable"
+
+.register_inputs_stable_trixie: &register_inputs_stable_trixie
+  <<: *register_inputs
+  runOnChangesTo: ['RELEASE_NOTES']
+  FLAVOR: "trixie"
+  SERIES: "stable"
+
+.register_inputs_next_bookworm: &register_inputs_next_bookworm
+  <<: *register_inputs
+  FLAVOR: "bookworm"
+  SERIES: next
+
+.register_inputs_next_trixie: &register_inputs_next_trixie
+  <<: *register_inputs
+  FLAVOR: "trixie"
+  SERIES: next
+
+################################################
+### Generate Debian Package for Internal APT ###
+################################################
+.cloudflared-apt-build: &cloudflared_apt_build
+  stage: package
+  needs:
+    - ci-image-get-image-ref
+    - linux-packaging # For consistency, we only run this job after we knew we could build the packages for external delivery
+  image: $BUILD_IMAGE
+  cache: {}
+  script:
+    - make cloudflared-deb
+  artifacts:
+    paths:
+      - cloudflared*.deb
+
+##############
+### Stable ###
+##############
+cloudflared-amd64-stable:
+  <<: *cloudflared_apt_build
+  rules:
+    - !reference [.default-rules, run-on-release]
+  variables: &amd64-stable-vars
+    GOOS: linux
+    GOARCH: amd64
+    FIPS: true
+    ORIGINAL_NAME: true
+    CGO_ENABLED: 1
+
+cloudflared-arm64-stable:
+  <<: *cloudflared_apt_build
+  rules:
+    - !reference [.default-rules, run-on-release]
+  variables: &arm64-stable-vars
+    GOOS: linux
+    GOARCH: arm64
+    FIPS: false # TUN-7595
+    ORIGINAL_NAME: true
+    CGO_ENABLED: 1
+
+############
+### Next ###
+############
+cloudflared-amd64-next:
+  <<: *cloudflared_apt_build
+  rules:
+    - !reference [.default-rules, run-on-master]
+  variables:
+    <<: *amd64-stable-vars
+    NIGHTLY: true
+
+cloudflared-arm64-next:
+  <<: *cloudflared_apt_build
+  rules:
+    - !reference [.default-rules, run-on-master]
+  variables:
+    <<: *arm64-stable-vars
+    NIGHTLY: true
+
+include:
+  - local: .ci/commons.gitlab-ci.yml
+
+  ##########################################
+  ### Publish Packages to Internal Repos ###
+  ##########################################
+  # Bookworm AMD64
+  - component: $CI_SERVER_FQDN/cloudflare/ci/apt-register/register@~latest
+    inputs:
+      <<: *register_inputs_stable_bookworm
+      jobPrefix: cloudflared-bookworm-amd64
+      needs: &amd64-stable ["cloudflared-amd64-stable"]
+
+  # Bookworm ARM64
+  - component: $CI_SERVER_FQDN/cloudflare/ci/apt-register/register@~latest
+    inputs:
+      <<: *register_inputs_stable_bookworm
+      jobPrefix: cloudflared-bookworm-arm64
+      needs: &arm64-stable ["cloudflared-arm64-stable"]
+
+  # Trixie AMD64
+  - component: $CI_SERVER_FQDN/cloudflare/ci/apt-register/register@~latest
+    inputs:
+      <<: *register_inputs_stable_trixie
+      jobPrefix: cloudflared-trixie-amd64
+      needs: *amd64-stable
+
+  # Trixie ARM64
+  - component: $CI_SERVER_FQDN/cloudflare/ci/apt-register/register@~latest
+    inputs:
+      <<: *register_inputs_stable_trixie
+      jobPrefix: cloudflared-trixie-arm64
+      needs: *arm64-stable
+
+  ##################################################
+  ### Publish Nightly Packages to Internal Repos ###
+  ##################################################
+  # Bookworm AMD64
+  - component: $CI_SERVER_FQDN/cloudflare/ci/apt-register/register@~latest
+    inputs:
+      <<: *register_inputs_next_bookworm
+      jobPrefix: cloudflared-nightly-bookworm-amd64
+      needs: &amd64-next ['cloudflared-amd64-next']
+
+  # Bookworm ARM64
+  - component: $CI_SERVER_FQDN/cloudflare/ci/apt-register/register@~latest
+    inputs:
+      <<: *register_inputs_next_bookworm
+      jobPrefix: cloudflared-nightly-bookworm-arm64
+      needs: &arm64-next ['cloudflared-arm64-next']
+
+  # Trixie AMD64
+  - component: $CI_SERVER_FQDN/cloudflare/ci/apt-register/register@~latest
+    inputs:
+      <<: *register_inputs_next_trixie
+      jobPrefix: cloudflared-nightly-trixie-amd64
+      needs: *amd64-next
+
+  # Trixie ARM64
+  - component: $CI_SERVER_FQDN/cloudflare/ci/apt-register/register@~latest
+    inputs:
+      <<: *register_inputs_next_trixie
+      jobPrefix: cloudflared-nightly-trixie-arm64
+      needs: *arm64-next

.ci/ci-image.gitlab-ci.yml

@@ -0,0 +1,31 @@
+# Builds a custom CI Image when necessary
+
+include:
+  #####################################################
+  ############## Build and Push CI Image ##############
+  #####################################################
+  - component: $CI_SERVER_FQDN/cloudflare/ci/docker-image/build-push-image@~latest
+    inputs:
+      stage: pre-build
+      jobPrefix: ci-image
+      runOnChangesTo: [".ci/image/**"]
+      runOnMR: true
+      runOnBranches: '^master$'
+      commentImageRefs: false
+      runner: vm-linux-x86-4cpu-8gb
+      EXTRA_DIB_ARGS: "--manifest=.ci/image/.docker-images"
+
+  #####################################################
+  ## Resolve the image reference for downstream jobs ##
+  #####################################################
+  - component: $CI_SERVER_FQDN/cloudflare/ci/docker-image/get-image-ref@~latest
+    inputs:
+      stage: pre-build
+      jobPrefix: ci-image
+      runOnMR: true
+      runOnBranches: '^master$'
+      IMAGE_PATH: "$REGISTRY_HOST/stash/tun/cloudflared/ci-image/master"
+      VARIABLE_NAME: BUILD_IMAGE
+      needs:
+        - job: ci-image-build-push-image
+          optional: true

.ci/commons.gitlab-ci.yml

@@ -0,0 +1,45 @@
+## A set of predefined rules to use on the different jobs
+.default-rules:
+  # Rules to run the job only on the master branch
+  run-on-master:
+    - if: $CI_COMMIT_BRANCH == $CI_DEFAULT_BRANCH
+      when: on_success
+    - when: never
+  # Rules to run the job only on merge requests
+  run-on-mr:
+    - if: $CI_COMMIT_TAG
+      when: never
+    - if: $CI_PIPELINE_SOURCE == "merge_request_event"
+      when: on_success
+    - when: never
+  # Rules to run the job on merge_requests and master branch
+  run-always:
+    - if: $CI_COMMIT_TAG
+      when: never
+    - if: $CI_PIPELINE_SOURCE == "merge_request_event"
+    - if: $CI_COMMIT_BRANCH != null && $CI_COMMIT_BRANCH == $CI_DEFAULT_BRANCH
+      when: on_success
+    - when: never
+  # Rules to run the job only when a release happens
+  run-on-release:
+    - if: $CI_COMMIT_BRANCH == $CI_DEFAULT_BRANCH
+      changes:
+          - 'RELEASE_NOTES'
+      when: on_success
+    - when: never
+
+.component-tests:
+  image: $BUILD_IMAGE
+  rules:
+    - !reference [.default-rules, run-always]
+  variables:
+    COMPONENT_TESTS_CONFIG: component-test-config.yaml
+    COMPONENT_TESTS_CONFIG_CONTENT: Y2xvdWRmbGFyZWRfYmluYXJ5OiBjbG91ZGZsYXJlZC5leGUKY3JlZGVudGlhbHNfZmlsZTogY3JlZC5qc29uCm9yaWdpbmNlcnQ6IGNlcnQucGVtCnpvbmVfZG9tYWluOiBhcmdvdHVubmVsdGVzdC5jb20Kem9uZV90YWc6IDQ4Nzk2ZjFlNzBiYjc2NjljMjliYjUxYmEyODJiZjY1
+  secrets:
+    DNS_API_TOKEN:
+      vault: gitlab/cloudflare/tun/cloudflared/_dev/_terraform_atlantis/component_tests_token/data@kv
+      file: false
+    COMPONENT_TESTS_ORIGINCERT:
+      vault: gitlab/cloudflare/tun/cloudflared/_dev/component_tests_cert_pem/data@kv
+      file: false
+  cache: {}

.ci/github.gitlab-ci.yml

@@ -0,0 +1,17 @@
+include:
+  - local: .ci/commons.gitlab-ci.yml
+
+######################################
+### Sync master branch with Github ###
+######################################
+push-github:
+  stage: sync
+  rules:
+    - !reference [.default-rules, run-on-master]
+  script:
+    - ./.ci/scripts/github-push.sh
+  secrets:
+    CLOUDFLARED_DEPLOY_SSH_KEY:
+      vault: gitlab/cloudflare/tun/cloudflared/_branch/master/cloudflared_github_ssh/data@kv
+      file: false
+  cache: {}

.ci/image/.docker-images

@@ -0,0 +1,2 @@
+images:
+  - name: ci-image

.ci/image/Dockerfile

@@ -0,0 +1,35 @@
+ARG CLOUDFLARE_DOCKER_REGISTRY_HOST
+
+FROM ${CLOUDFLARE_DOCKER_REGISTRY_HOST:-registry.cfdata.org}/stash/cf/debian-images/bookworm/main:2025.7.0@sha256:6350da2f7e728dae2c1420f6dafc38e23cacc0b399d3d5b2f40fe48d9c8ff1ca
+
+RUN apt-get update && \
+    apt-get upgrade -y && \
+    apt-get install --no-install-recommends --allow-downgrades -y \
+        build-essential \
+        git \
+        go-boring=1.24.9-1 \
+        libffi-dev \
+        procps \
+        python3-dev \
+        python3-pip \
+        python3-setuptools \
+        python3-venv \
+        # libmsi and libgcab are libraries the wixl binary depends on.
+        libmsi-dev \
+        libgcab-dev \
+        # deb and rpm build tools
+        rubygem-fpm \
+        rpm \
+        # create deb and rpm repository files
+        reprepro \
+        createrepo-c \
+        # gcc for cross architecture compilation in arm
+        gcc-aarch64-linux-gnu \
+        libc6-dev-arm64-cross && \
+    rm -rf /var/lib/apt/lists/* && \
+    # Install wixl
+    curl -o /usr/local/bin/wixl -L https://pkg.cloudflare.com/binaries/wixl && \
+    chmod a+x /usr/local/bin/wixl && \
+    mkdir -p opt
+
+WORKDIR /opt

.ci/linux.gitlab-ci.yml

@@ -0,0 +1,122 @@
+.golang-inputs: &golang_inputs
+  runOnMR: true
+  runOnBranches: '^master$'
+  outputDir: artifacts
+  runner: linux-x86-8cpu-16gb
+  stage: build
+  golangVersion: "boring-1.24"
+  imageVersion: "3371-f5539bd6f83d@sha256:a2a68f580070f9411d0d3155959ed63b700ef319b5fcc62db340e92227bbc628"
+  CGO_ENABLED: 1
+
+.default-packaging-job: &packaging-job-defaults
+  stage: package
+  needs:
+    - ci-image-get-image-ref
+  rules:
+    - !reference [.default-rules, run-on-master]
+  image: $BUILD_IMAGE
+  cache: {}
+  artifacts:
+    paths:
+      - artifacts/*
+
+include:
+  ###################
+  ### Linux Build ###
+  ###################
+  - component: $CI_SERVER_FQDN/cloudflare/ci/golang/boring-make@~latest
+    inputs:
+      <<: *golang_inputs
+      jobPrefix: linux-build
+      GOLANG_MAKE_TARGET: ci-build
+
+  ########################
+  ### Linux FIPS Build ###
+  ########################
+  - component: $CI_SERVER_FQDN/cloudflare/ci/golang/boring-make@~latest
+    inputs:
+      <<: *golang_inputs
+      jobPrefix: linux-fips-build
+      GOLANG_MAKE_TARGET: ci-fips-build
+
+  #################
+  ### Unit Tests ##
+  #################
+  - component: $CI_SERVER_FQDN/cloudflare/ci/golang/boring-make@~latest
+    inputs:
+      <<: *golang_inputs
+      stage: test
+      jobPrefix: test
+      GOLANG_MAKE_TARGET: ci-test
+
+  ######################
+  ### Unit Tests FIPS ##
+  ######################
+  - component: $CI_SERVER_FQDN/cloudflare/ci/golang/boring-make@~latest
+    inputs:
+      <<: *golang_inputs
+      stage: test
+      jobPrefix: test-fips
+      GOLANG_MAKE_TARGET: ci-fips-test
+
+  #################
+  ### Vuln Check ##
+  #################
+  - component: $CI_SERVER_FQDN/cloudflare/ci/golang/boring-make@~latest
+    inputs:
+      <<: *golang_inputs
+      runOnBranches: '^$'
+      stage: validate
+      jobPrefix: vulncheck
+      GOLANG_MAKE_TARGET: vulncheck
+
+#################################
+### Run Linux Component Tests ###
+#################################
+linux-component-tests: &linux-component-tests
+  stage: test
+  extends: .component-tests
+  needs:
+    - ci-image-get-image-ref
+    - linux-build-boring-make
+  script:
+    - ./.ci/scripts/component-tests.sh
+  variables: &component-tests-variables
+    CI: 1
+    COMPONENT_TESTS_CONFIG_CONTENT: Y2xvdWRmbGFyZWRfYmluYXJ5OiAuL2Nsb3VkZmxhcmVkCmNyZWRlbnRpYWxzX2ZpbGU6IGNyZWQuanNvbgpvcmlnaW5jZXJ0OiBjZXJ0LnBlbQp6b25lX2RvbWFpbjogYXJnb3R1bm5lbHRlc3QuY29tCnpvbmVfdGFnOiA0ODc5NmYxZTcwYmI3NjY5YzI5YmI1MWJhMjgyYmY2NQ==
+  tags:
+    - linux-x86-8cpu-16gb
+  artifacts:
+    reports:
+      junit: report.xml
+
+######################################
+### Run Linux FIPS Component Tests ###
+######################################
+linux-component-tests-fips:
+  <<: *linux-component-tests
+  needs:
+    - ci-image-get-image-ref
+    - linux-fips-build-boring-make
+  variables:
+    <<: *component-tests-variables
+    COMPONENT_TESTS_FIPS: 1
+
+################################
+####### Linux Packaging ########
+################################
+linux-packaging:
+  <<: *packaging-job-defaults
+  parallel:
+    matrix:
+      - ARCH: ["386", "amd64", "arm", "armhf", "arm64"]
+  script:
+    - ./.ci/scripts/linux/build-packages.sh ${ARCH}
+
+################################
+##### Linux FIPS Packaging #####
+################################
+linux-packaging-fips:
+  <<: *packaging-job-defaults
+  script:
+    - ./.ci/scripts/linux/build-packages-fips.sh

.ci/mac.gitlab-ci.yml

@@ -0,0 +1,66 @@
+include:
+  - local: .ci/commons.gitlab-ci.yml
+
+###############################
+### Defaults for Mac Builds ###
+###############################
+.mac-build-defaults: &mac-build-defaults
+  rules:
+    - !reference [.default-rules, run-on-mr]
+  tags:
+    - "macstadium-${RUNNER_ARCH}"
+  parallel:
+    matrix:
+      - RUNNER_ARCH: [arm, intel]
+  cache: {}
+
+######################################
+### Build Cloudflared Mac Binaries ###
+######################################
+macos-build-cloudflared: &mac-build
+  <<: *mac-build-defaults
+  stage: build
+  artifacts:
+    paths:
+      - artifacts/*
+  script:
+    - '[ "${RUNNER_ARCH}" = "arm" ] && export TARGET_ARCH=arm64'
+    - '[ "${RUNNER_ARCH}" = "intel" ] && export TARGET_ARCH=amd64'
+    - ARCH=$(uname -m)
+    - echo ARCH=$ARCH - TARGET_ARCH=$TARGET_ARCH
+    - ./.ci/scripts/mac/install-go.sh
+    - BUILD_SCRIPT=.ci/scripts/mac/build.sh
+    - if [[ ! -x ${BUILD_SCRIPT} ]] ; then exit ; fi
+    - set -euo pipefail
+    - echo "Executing ${BUILD_SCRIPT}"
+    - exec ${BUILD_SCRIPT}
+
+###############################################
+### Build and Sign Cloudflared Mac Binaries ###
+###############################################
+macos-build-and-sign-cloudflared:
+  <<: *mac-build
+  rules:
+    - !reference [.default-rules, run-on-master]
+  secrets:
+    APPLE_DEV_CA_CERT:
+      vault: gitlab/cloudflare/tun/cloudflared/_branch/master/apple_dev_ca_cert_v2/data@kv
+      file: false
+    CFD_CODE_SIGN_CERT:
+      vault: gitlab/cloudflare/tun/cloudflared/_branch/master/cfd_code_sign_cert_v2/data@kv
+      file: false
+    CFD_CODE_SIGN_KEY:
+      vault: gitlab/cloudflare/tun/cloudflared/_branch/master/cfd_code_sign_key_v2/data@kv
+      file: false
+    CFD_CODE_SIGN_PASS:
+      vault: gitlab/cloudflare/tun/cloudflared/_branch/master/cfd_code_sign_pass_v2/data@kv
+      file: false
+    CFD_INSTALLER_CERT:
+      vault: gitlab/cloudflare/tun/cloudflared/_branch/master/cfd_installer_cert_v2/data@kv
+      file: false
+    CFD_INSTALLER_KEY:
+      vault: gitlab/cloudflare/tun/cloudflared/_branch/master/cfd_installer_key_v2/data@kv
+      file: false
+    CFD_INSTALLER_PASS:
+      vault: gitlab/cloudflare/tun/cloudflared/_branch/master/cfd_installer_pass_v2/data@kv
+      file: false

.ci/release.gitlab-ci.yml

@@ -0,0 +1,133 @@
+include:
+  - local: .ci/commons.gitlab-ci.yml
+
+  ######################################
+  ### Build and Push DockerHub Image ###
+  ######################################
+  - component: $CI_SERVER_FQDN/cloudflare/ci/docker-image/build-push-image@~latest
+    inputs:
+      stage: release
+      jobPrefix: docker-hub
+      runOnMR: false
+      runOnBranches: '^master$'
+      runOnChangesTo: ['RELEASE_NOTES']
+      needs:
+        - generate-version-file
+        - release-cloudflared-to-r2
+      commentImageRefs: false
+      runner: vm-linux-x86-4cpu-8gb
+      # Based on if the CI reference is protected or not the CI component will
+      # either use _BRANCH or _PROD, therefore, to prevent the pipelines from failing
+      # we simply set both to the same value.
+      DOCKER_USER_BRANCH: &docker-hub-user svcgithubdockerhubcloudflar045
+      DOCKER_PASSWORD_BRANCH: &docker-hub-password gitlab/cloudflare/tun/cloudflared/_dev/dockerhub/svc_password/data
+      DOCKER_USER_PROD: *docker-hub-user
+      DOCKER_PASSWORD_PROD: *docker-hub-password
+      EXTRA_DIB_ARGS: --overwrite
+
+.default-release-job: &release-job-defaults
+  stage: release
+  image: $BUILD_IMAGE
+  cache:
+    paths:
+      - .cache/pip
+  variables: &release-job-variables
+    PIP_CACHE_DIR: "$CI_PROJECT_DIR/.cache/pip"
+    # KV Vars
+    KV_NAMESPACE: 380e19aa04314648949b6ad841417ebe
+    KV_ACCOUNT: &cf-account 5ab4e9dfbd435d24068829fda0077963
+    # R2 Vars
+    R2_BUCKET: cloudflared-pkgs
+    R2_ACCOUNT_ID: *cf-account
+    # APT and RPM Repository Vars
+    GPG_PUBLIC_KEY_URL: "https://pkg.cloudflare.com/cloudflare-ascii-pubkey.gpg"
+    PKG_URL: "https://pkg.cloudflare.com/cloudflared"
+    BINARY_NAME: cloudflared
+  secrets:
+    KV_API_TOKEN:
+      vault: gitlab/cloudflare/tun/cloudflared/_dev/cfd_kv_api_token/data@kv
+      file: false
+    API_KEY:
+      vault: gitlab/cloudflare/tun/cloudflared/_dev/cfd_github_api_key/data@kv
+      file: false
+    R2_CLIENT_ID:
+      vault: gitlab/cloudflare/tun/cloudflared/_dev/_terraform_atlantis/r2_api_token/client_id@kv
+      file: false
+    R2_CLIENT_SECRET:
+      vault: gitlab/cloudflare/tun/cloudflared/_dev/_terraform_atlantis/r2_api_token/client_secret@kv
+      file: false
+    LINUX_SIGNING_PUBLIC_KEY:
+      vault: gitlab/cloudflare/tun/cloudflared/_dev/gpg_v1/public_key@kv
+      file: false
+    LINUX_SIGNING_PRIVATE_KEY:
+      vault: gitlab/cloudflare/tun/cloudflared/_dev/gpg_v1/private_key@kv
+      file: false
+    LINUX_SIGNING_PUBLIC_KEY_2:
+      vault: gitlab/cloudflare/tun/cloudflared/_dev/gpg_v2/public_key@kv
+      file: false
+    LINUX_SIGNING_PRIVATE_KEY_2:
+      vault: gitlab/cloudflare/tun/cloudflared/_dev/gpg_v2/private_key@kv
+      file: false
+
+###########################################
+### Push Cloudflared Binaries to Github ###
+###########################################
+release-cloudflared-to-github:
+  <<: *release-job-defaults
+  rules:
+    - !reference [.default-rules, run-on-release]
+  needs:
+    - ci-image-get-image-ref
+    - linux-packaging
+    - linux-packaging-fips
+    - macos-build-and-sign-cloudflared
+    - windows-package-sign
+  script:
+    - ./.ci/scripts/release-target.sh github-release
+
+#########################################
+### Upload Cloudflared Binaries to R2 ###
+#########################################
+release-cloudflared-to-r2:
+  <<: *release-job-defaults
+  rules:
+    - !reference [.default-rules, run-on-release]
+  needs:
+    - ci-image-get-image-ref
+    - linux-packaging # We only release non-FIPS binaries to R2
+    - release-cloudflared-to-github
+  script:
+    - ./.ci/scripts/release-target.sh r2-linux-release
+
+#################################################
+### Upload Cloudflared Nightly Binaries to R2 ###
+#################################################
+release-cloudflared-nightly-to-r2:
+  <<: *release-job-defaults
+  rules:
+    - !reference [.default-rules, run-on-master]
+  variables:
+    <<: *release-job-variables
+    R2_BUCKET: cloudflared-pkgs-next
+    GPG_PUBLIC_KEY_URL: "https://next.pkg.cloudflare.com/cloudflare-ascii-pubkey.gpg"
+    PKG_URL: "https://next.pkg.cloudflare.com/cloudflared"
+  needs:
+    - ci-image-get-image-ref
+    - linux-packaging # We only release non-FIPS binaries to R2
+  script:
+    - ./.ci/scripts/release-target.sh r2-linux-release
+
+#############################
+### Generate Version File ###
+#############################
+generate-version-file:
+  <<: *release-job-defaults
+  rules:
+    - !reference [.default-rules, run-on-release]
+  needs:
+    - ci-image-get-image-ref
+  script:
+    - make generate-docker-version
+  artifacts:
+    paths:
+      - versions

.ci/scripts/component-tests.sh

@@ -0,0 +1,25 @@
+#!/bin/bash
+set -e -o pipefail
+
+# Fetch cloudflared from the artifacts folder
+mv ./artifacts/cloudflared ./cloudflared
+
+python3 -m venv env
+. env/bin/activate
+
+pip install --upgrade -r component-tests/requirements.txt
+
+# Creates and routes a Named Tunnel for this build. Also constructs
+# config file from env vars.
+python3 component-tests/setup.py --type create
+
+# Define the cleanup function
+cleanup() {
+    # The Named Tunnel is deleted and its route unprovisioned here.
+    python3 component-tests/setup.py --type cleanup
+}
+
+# The trap will call the cleanup function on script exit
+trap cleanup EXIT
+
+pytest component-tests -o log_cli=true --log-cli-level=INFO --junit-xml=report.xml

.ci/scripts/fmt-check.sh

@@ -1,8 +1,7 @@
 #!/bin/bash
-
 set -e -o pipefail
 
-OUTPUT=$(goimports -l -d -local github.com/cloudflare/cloudflared $(go list -mod=vendor -f '{{.Dir}}' -a ./... | fgrep -v tunnelrpc))
+OUTPUT=$(go run -mod=readonly golang.org/x/tools/cmd/goimports@v0.30.0 -l -d -local github.com/cloudflare/cloudflared $(go list -mod=vendor -f '{{.Dir}}' -a ./... | fgrep -v tunnelrpc))
 
 if [ -n "$OUTPUT" ] ; then
   PAGER=$(which colordiff || echo cat)

.ci/scripts/github-push.sh

@@ -0,0 +1,31 @@
+#!/bin/bash
+set -e -o pipefail
+
+BRANCH="master"
+TMP_PATH="$PWD/tmp"
+PRIVATE_KEY_PATH="$TMP_PATH/github-deploy-key"
+PUBLIC_KEY_GITHUB_PATH="$TMP_PATH/github.pub"
+
+mkdir -p $TMP_PATH
+
+# Setup Private Key
+echo "$CLOUDFLARED_DEPLOY_SSH_KEY" > $PRIVATE_KEY_PATH
+chmod 400 $PRIVATE_KEY_PATH
+
+# Download GitHub Public Key for KnownHostsFile
+ssh-keyscan -t ed25519 github.com > $PUBLIC_KEY_GITHUB_PATH
+
+# Setup git ssh command with the right configurations
+export GIT_SSH_COMMAND="ssh -o UserKnownHostsFile=$PUBLIC_KEY_GITHUB_PATH -o IdentitiesOnly=yes -i $PRIVATE_KEY_PATH"
+
+# Add GitHub as a new remote
+git remote add github git@github.com:cloudflare/cloudflared.git || true
+
+# GitLab doesn't pull branch references, instead it creates a new one on each pipeline.
+# Therefore, we need to manually fetch the reference to then push it to GitHub.
+git fetch origin $BRANCH:$BRANCH
+git push -u github $BRANCH
+
+if TAG="$(git describe --tags --exact-match 2>/dev/null)"; then
+  git push -u github "$TAG"
+fi

.ci/scripts/linux/build-packages.sh

@@ -0,0 +1,59 @@
+#!/bin/bash
+
+# Check if architecture argument is provided
+if [ $# -eq 0 ]; then
+    echo "Error: Architecture argument is required"
+    echo "Usage: $0 <architecture>"
+    exit 1
+fi
+
+# Parameters
+arch=$1
+
+# Get Version
+VERSION=$(git describe --tags --always --match "[0-9][0-9][0-9][0-9].*.*")
+echo $VERSION
+
+# Disable FIPS module in go-boring
+export GOEXPERIMENT=noboringcrypto
+export CGO_ENABLED=0
+
+# This controls the directory the built artifacts go into
+export ARTIFACT_DIR=artifacts/
+mkdir -p $ARTIFACT_DIR
+
+export TARGET_OS=linux
+
+unset TARGET_ARM
+export TARGET_ARCH=$arch
+
+## Support for arm platforms without hardware FPU enabled
+if [[ $arch == arm ]] ; then
+    export TARGET_ARCH=arm
+    export TARGET_ARM=5
+fi
+
+## Support for armhf builds
+if [[ $arch == armhf ]] ; then
+    export TARGET_ARCH=arm
+    export TARGET_ARM=7
+fi
+
+make cloudflared-deb
+mv cloudflared\_$VERSION\_$arch.deb $ARTIFACT_DIR/cloudflared-linux-$arch.deb
+
+# rpm packages invert the - and _ and use x86_64 instead of amd64.
+RPMVERSION=$(echo $VERSION|sed -r 's/-/_/g')
+RPMARCH=$arch
+if [ $arch == "amd64" ];then
+    RPMARCH="x86_64"
+fi
+if [ $arch == "arm64" ]; then
+    RPMARCH="aarch64"
+fi
+make cloudflared-rpm
+mv cloudflared-$RPMVERSION-1.$RPMARCH.rpm $ARTIFACT_DIR/cloudflared-linux-$RPMARCH.rpm
+
+# finally move the linux binary as well.
+mv ./cloudflared $ARTIFACT_DIR/cloudflared-linux-$arch
+

.ci/scripts/package-windows.sh

@@ -1,19 +1,23 @@
+#!/bin/bash
+python3 -m venv env
+. env/bin/activate
+pip install pynacl==1.4.0 pygithub==1.55
+
 VERSION=$(git describe --tags --always --match "[0-9][0-9][0-9][0-9].*.*")
 echo $VERSION
 
 export TARGET_OS=windows
 # This controls the directory the built artifacts go into
-export BUILT_ARTIFACT_DIR=built_artifacts/
+export BUILT_ARTIFACT_DIR=artifacts/
 export FINAL_ARTIFACT_DIR=artifacts/
 mkdir -p $BUILT_ARTIFACT_DIR
 mkdir -p $FINAL_ARTIFACT_DIR
 windowsArchs=("amd64" "386")
 for arch in ${windowsArchs[@]}; do
     export TARGET_ARCH=$arch
-    # Copy exe into final directory
+    # Copy .exe from artifacts directory
     cp $BUILT_ARTIFACT_DIR/cloudflared-windows-$arch.exe ./cloudflared.exe
     make cloudflared-msi
     # Copy msi into final directory
     mv cloudflared-$VERSION-$arch.msi $FINAL_ARTIFACT_DIR/cloudflared-windows-$arch.msi
-    cp $BUILT_ARTIFACT_DIR/cloudflared-windows-$arch.exe $FINAL_ARTIFACT_DIR/cloudflared-windows-$arch.exe 
 done

.ci/scripts/release-target.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+set -e -o pipefail
+
+# Check if a make target is provided as an argument
+if [ $# -eq 0 ]; then
+    echo "Error: Make target argument is required"
+    echo "Usage: $0 <make-target>"
+    exit 1
+fi
+
+MAKE_TARGET=$1
+
+python3 -m venv venv
+source venv/bin/activate
+
+# Our release scripts are written in python, so we should install their dependecies here.
+pip install pynacl==1.4.0 pygithub==1.55 boto3==1.22.9 python-gnupg==0.4.9
+make $MAKE_TARGET

.ci/scripts/vuln-check.sh

@@ -0,0 +1,52 @@
+#!/bin/bash
+set -e
+
+# Define the file to store the list of vulnerabilities to ignore.
+IGNORE_FILE=".vulnignore"
+
+# Check if the ignored vulnerabilities file exists. If not, create an empty one.
+if [ ! -f "$IGNORE_FILE" ]; then
+    touch "$IGNORE_FILE"
+    echo "Created an empty file to store ignored vulnerabilities: $IGNORE_FILE"
+    echo "# Add vulnerability IDs (e.g., GO-2022-0450) to ignore, one per line." >> "$IGNORE_FILE"
+    echo "# You can also add comments on the same line after the ID." >> "$IGNORE_FILE"
+    echo "" >> "$IGNORE_FILE"
+fi
+
+# Run govulncheck and capture its output.
+VULN_OUTPUT=$(go run -mod=readonly golang.org/x/vuln/cmd/govulncheck@latest ./... || true)
+
+# Print the govuln output
+echo "====================================="
+echo "Full Output of govulncheck:"
+echo "====================================="
+echo "$VULN_OUTPUT"
+echo "====================================="
+echo "End of govulncheck Output"
+echo "====================================="
+
+# Process the ignore file to remove comments and empty lines.
+# The 'cut' command gets the vulnerability ID and removes anything after the '#'.
+# The 'grep' command filters out empty lines and lines starting with '#'.
+CLEAN_IGNORES=$(grep -v '^\s*#' "$IGNORE_FILE" | cut -d'#' -f1 | sed 's/ //g' | sort -u || true)
+
+# Filter out the ignored vulnerabilities.
+UNIGNORED_VULNS=$(echo "$VULN_OUTPUT" | grep 'Vulnerability')
+
+# If the list of ignored vulnerabilities is not empty, filter them out.
+if [ -n "$CLEAN_IGNORES" ]; then
+    UNIGNORED_VULNS=$(echo "$UNIGNORED_VULNS" | grep -vFf <(echo "$CLEAN_IGNORES") || true)
+fi
+
+# If there are any vulnerabilities that were not in our ignore list, print them and exit with an error.
+if [ -n "$UNIGNORED_VULNS" ]; then
+    echo "🚨 Found new, unignored vulnerabilities:"
+    echo "-------------------------------------"
+    echo "$UNIGNORED_VULNS"
+    echo "-------------------------------------"
+    echo "Exiting with an error. ❌"
+    exit 1
+else
+    echo "🎉 No new vulnerabilities found. All clear! ✨"
+    exit 0
+fi

.ci/scripts/windows/builds.ps1

@@ -0,0 +1,29 @@
+Set-StrictMode -Version Latest
+$ErrorActionPreference = "Stop"
+$ProgressPreference = "SilentlyContinue"
+
+$env:TARGET_OS = "windows"
+$env:LOCAL_OS = "windows"
+$TIMESTAMP_RFC3161 = "http://timestamp.digicert.com"
+
+New-Item -Path ".\artifacts" -ItemType Directory
+
+Write-Output "Building for amd64"
+$env:TARGET_ARCH = "amd64"
+$env:LOCAL_ARCH = "amd64"
+$env:CGO_ENABLED = 1
+& make cloudflared
+if ($LASTEXITCODE -ne 0) { throw "Failed to build cloudflared for amd64" }
+# Sign build
+azuresigntool.exe sign -kvu $env:KEY_VAULT_URL -kvi "$env:KEY_VAULT_CLIENT_ID" -kvs "$env:KEY_VAULT_SECRET" -kvc "$env:KEY_VAULT_CERTIFICATE" -kvt "$env:KEY_VAULT_TENANT_ID" -tr "$TIMESTAMP_RFC3161" -d "Cloudflare Tunnel Daemon" .\cloudflared.exe
+copy .\cloudflared.exe .\artifacts\cloudflared-windows-amd64.exe
+
+Write-Output "Building for 386"
+$env:TARGET_ARCH = "386"
+$env:LOCAL_ARCH = "386"
+$env:CGO_ENABLED = 0
+& make cloudflared
+if ($LASTEXITCODE -ne 0) { throw "Failed to build cloudflared for 386" }
+## Sign build
+azuresigntool.exe sign -kvu $env:KEY_VAULT_URL -kvi "$env:KEY_VAULT_CLIENT_ID" -kvs "$env:KEY_VAULT_SECRET" -kvc "$env:KEY_VAULT_CERTIFICATE" -kvt "$env:KEY_VAULT_TENANT_ID" -tr "$TIMESTAMP_RFC3161" -d "Cloudflare Tunnel Daemon" .\cloudflared.exe
+copy .\cloudflared.exe .\artifacts\cloudflared-windows-386.exe

.ci/scripts/windows/component-test.ps1

@@ -0,0 +1,40 @@
+Set-StrictMode -Version Latest
+$ErrorActionPreference = "Stop"
+$ProgressPreference = "SilentlyContinue"
+
+$env:TARGET_OS = "windows"
+$env:LOCAL_OS = "windows"
+$env:TARGET_ARCH = "amd64"
+$env:LOCAL_ARCH = "amd64"
+$env:CGO_ENABLED = 1
+
+python --version
+python -m pip --version
+
+
+Write-Host "Building cloudflared"
+& make cloudflared
+if ($LASTEXITCODE -ne 0) { throw "Failed to build cloudflared" }
+
+
+Write-Host "Running unit tests"
+# Not testing with race detector because of https://github.com/golang/go/issues/61058
+# We already test it on other platforms
+go test -failfast -v -mod=vendor ./...
+if ($LASTEXITCODE -ne 0) { throw "Failed unit tests" }
+
+
+# On Gitlab runners we need to add all of this addresses to the NO_PROXY list in order for the tests to run.
+$env:NO_PROXY = "pypi.org,files.pythonhosted.org,api.cloudflare.com,argotunneltest.com,argotunnel.com,trycloudflare.com,${env:NO_PROXY}"
+Write-Host "No Proxy: ${env:NO_PROXY}"
+Write-Host "Running component tests"
+try {
+    python -m pip --disable-pip-version-check install --upgrade -r component-tests/requirements.txt --use-pep517
+    python component-tests/setup.py --type create
+    python -m pytest component-tests -o log_cli=true --log-cli-level=INFO --junit-xml=report.xml
+    if ($LASTEXITCODE -ne 0) {
+        throw "Failed component tests"
+    }
+} finally {
+    python component-tests/setup.py --type cleanup
+}

.ci/scripts/windows/go-wrapper.ps1

@@ -0,0 +1,69 @@
+Param(
+    [string]$GoVersion,
+    [string]$ScriptToExecute
+)
+
+# The script is a wrapper that downloads a specific version
+# of go, adds it to the PATH and executes a script with that go
+# version in the path.
+
+Set-StrictMode -Version Latest
+$ErrorActionPreference = "Stop"
+$ProgressPreference = "SilentlyContinue"
+
+# Get the path to the system's temporary directory.
+$tempPath = [System.IO.Path]::GetTempPath()
+
+# Create a unique name for the new temporary folder.
+$folderName = "go_" + (Get-Random)
+
+# Join the temp path and the new folder name to create the full path.
+$fullPath = Join-Path -Path $tempPath -ChildPath $folderName
+
+# Store the current value of PATH environment variable.
+$oldPath = $env:Path
+
+# Use a try...finally block to ensure the temporrary folder and PATH are cleaned up.
+try {
+    # Create the temporary folder.
+    Write-Host "Creating temporary folder at: $fullPath"
+    $newTempFolder = New-Item -ItemType Directory -Path $fullPath -Force
+
+    # Download go
+    $url = "https://go.dev/dl/$GoVersion.windows-amd64.zip"
+    $destinationFile = Join-Path -Path $newTempFolder.FullName -ChildPath "go$GoVersion.windows-amd64.zip"
+    Write-Host "Downloading go from: $url"
+    Invoke-WebRequest -Uri $url -OutFile $destinationFile
+    Write-Host "File downloaded to: $destinationFile"
+
+    # Unzip the downloaded file.
+    Write-Host "Unzipping the file..."
+    Expand-Archive -Path $destinationFile -DestinationPath $newTempFolder.FullName -Force
+    Write-Host "File unzipped successfully."
+
+    # Define the go/bin path wich is inside the temporary folder
+    $goBinPath = Join-Path -Path $fullPath -ChildPath "go\bin"
+
+    # Add the go/bin path to the PATH environment variable.
+    $env:Path = "$goBinPath;$($env:Path)"
+    Write-Host "Added $goBinPath to the environment PATH."
+
+    go env
+    go version
+
+    & $ScriptToExecute
+} finally {
+    # Cleanup: Remove the path from the environment variable and then the temporary folder.
+    Write-Host "Starting cleanup..."
+
+    $env:Path = $oldPath
+    Write-Host "Reverted changes in the environment PATH."
+
+    # Remove the temporary folder and its contents.
+    if (Test-Path -Path $fullPath) {
+        Remove-Item -Path $fullPath -Recurse -Force
+        Write-Host "Temporary folder and its contents have been removed."
+    } else {
+        Write-Host "Temporary folder does not exist, no cleanup needed."
+    }
+}

.ci/scripts/windows/sign-msi.ps1

@@ -0,0 +1,26 @@
+# Sign Windows artifacts using azuretool
+# This script processes MSI files from the artifacts directory
+
+$ErrorActionPreference = "Stop"
+
+# Define paths
+$ARTIFACT_DIR = "artifacts"
+$TIMESTAMP_RFC3161 = "http://timestamp.digicert.com"
+
+Write-Host "Looking for Windows artifacts to sign in $ARTIFACT_DIR..."
+
+# Find all Windows MSI files
+$msiFiles = Get-ChildItem -Path $ARTIFACT_DIR -Filter "cloudflared-windows-*.msi" -ErrorAction SilentlyContinue
+
+if ($msiFiles.Count -eq 0) {
+    Write-Host "No Windows MSI files found in $ARTIFACT_DIR"
+    exit 1
+}
+
+Write-Host "Found $($msiFiles.Count) file(s) to sign:"
+foreach ($file in $msiFiles) {
+    Write-Host "Running azuretool sign for $($file.Name)"
+    azuresigntool.exe sign -kvu $env:KEY_VAULT_URL -kvi "$env:KEY_VAULT_CLIENT_ID" -kvs "$env:KEY_VAULT_SECRET" -kvc "$env:KEY_VAULT_CERTIFICATE" -kvt "$env:KEY_VAULT_TENANT_ID" -tr "$TIMESTAMP_RFC3161" -d "Cloudflare Tunnel Daemon" .\\$ARTIFACT_DIR\\$($file.Name)
+}
+
+Write-Host "Signing process completed"

.ci/windows.gitlab-ci.yml

@@ -0,0 +1,114 @@
+include:
+  - local: .ci/commons.gitlab-ci.yml
+
+###################################
+### Defaults for Windows Builds ###
+###################################
+.windows-build-defaults: &windows-build-defaults
+  rules:
+    - !reference [.default-rules, run-always]
+  tags:
+    - windows-x86
+  cache: {}
+
+##########################################
+### Build Cloudflared Windows Binaries ###
+##########################################
+windows-build-cloudflared:
+  <<: *windows-build-defaults
+  stage: build
+  script:
+    - powershell -ExecutionPolicy Bypass -File ".\.ci\scripts\windows\go-wrapper.ps1" "${GO_VERSION}" ".\.ci\scripts\windows\builds.ps1"
+  artifacts:
+    paths:
+      - artifacts/*
+
+######################################################
+### Load Environment Variables for Component Tests ###
+######################################################
+windows-load-env-variables:
+  stage: pre-build
+  extends: .component-tests
+  script:
+    - echo "COMPONENT_TESTS_CONFIG=$COMPONENT_TESTS_CONFIG" >> windows.env
+    - echo "COMPONENT_TESTS_CONFIG_CONTENT=$COMPONENT_TESTS_CONFIG_CONTENT" >> windows.env
+    - echo "DNS_API_TOKEN=$DNS_API_TOKEN" >> windows.env
+    # We have to encode the `COMPONENT_TESTS_ORIGINCERT` secret, because it content is a file, otherwise we can't export it using gitlab
+    - echo "COMPONENT_TESTS_ORIGINCERT=$(echo "$COMPONENT_TESTS_ORIGINCERT" | base64 -w0)" >> windows.env
+    - echo "KEY_VAULT_URL=$KEY_VAULT_URL" >> windows.env
+    - echo "KEY_VAULT_CLIENT_ID=$KEY_VAULT_CLIENT_ID" >> windows.env
+    - echo "KEY_VAULT_TENANT_ID=$KEY_VAULT_TENANT_ID" >> windows.env
+    - echo "KEY_VAULT_SECRET=$KEY_VAULT_SECRET" >> windows.env
+    - echo "KEY_VAULT_CERTIFICATE=$KEY_VAULT_CERTIFICATE" >> windows.env
+  variables:
+    COMPONENT_TESTS_CONFIG_CONTENT: Y2xvdWRmbGFyZWRfYmluYXJ5OiAuL2Nsb3VkZmxhcmVkLmV4ZQpjcmVkZW50aWFsc19maWxlOiBjcmVkLmpzb24Kb3JpZ2luY2VydDogY2VydC5wZW0Kem9uZV9kb21haW46IGFyZ290dW5uZWx0ZXN0LmNvbQp6b25lX3RhZzogNDg3OTZmMWU3MGJiNzY2OWMyOWJiNTFiYTI4MmJmNjU=
+  secrets:
+    KEY_VAULT_URL:
+      vault: gitlab/cloudflare/tun/cloudflared/_dev/azure_vault/app_info/key_vault_url@kv
+      file: false
+    KEY_VAULT_CLIENT_ID:
+      vault: gitlab/cloudflare/tun/cloudflared/_dev/azure_vault/app_info/key_vault_client_id@kv
+      file: false
+    KEY_VAULT_TENANT_ID:
+      vault: gitlab/cloudflare/tun/cloudflared/_dev/azure_vault/app_info/key_vault_tenant_id@kv
+      file: false
+    KEY_VAULT_SECRET:
+      vault: gitlab/cloudflare/tun/cloudflared/_dev/azure_vault/secret/key_vault_secret@kv
+      file: false
+    KEY_VAULT_CERTIFICATE:
+      vault: gitlab/cloudflare/tun/cloudflared/_dev/azure_vault/certificate/key_vault_certificate@kv
+      file: false
+  artifacts:
+    access: 'none'
+    reports:
+      dotenv: windows.env
+
+###################################
+### Run Windows Component Tests ###
+###################################
+windows-component-tests-cloudflared:
+  <<: *windows-build-defaults
+  stage: test
+  needs: ["windows-load-env-variables"]
+  script:
+    # We have to decode the secret we encoded on the `windows-load-env-variables` job
+    - $env:COMPONENT_TESTS_ORIGINCERT = [System.Text.Encoding]::UTF8.GetString([System.Convert]::FromBase64String($env:COMPONENT_TESTS_ORIGINCERT))
+    - powershell -ExecutionPolicy Bypass -File ".\.ci\scripts\windows\go-wrapper.ps1" "${GO_VERSION}" ".\.ci\scripts\windows\component-test.ps1"
+  artifacts:
+    reports:
+      junit: report.xml
+
+################################
+### Package Windows Binaries ###
+################################
+windows-package:
+  rules:
+    - !reference [.default-rules, run-on-master]
+  stage: package
+  needs:
+    - ci-image-get-image-ref
+    - windows-build-cloudflared
+  image: $BUILD_IMAGE
+  script:
+    - .ci/scripts/package-windows.sh
+  cache: {}
+  artifacts:
+    paths:
+      - artifacts/*
+
+#############################
+### Sign Windows Binaries ###
+#############################
+windows-package-sign:
+  <<: *windows-build-defaults
+  rules:
+    - !reference [.default-rules, run-on-master]
+  stage: package
+  needs:
+    - windows-package
+    - windows-load-env-variables
+  script:
+    - powershell -ExecutionPolicy Bypass -File ".\.ci\scripts\windows\sign-msi.ps1"
+  artifacts:
+    paths:
+      - artifacts/*

.gitlab-ci.yml

@@ -1,172 +1,58 @@
 variables:
-  # Define GOPATH within the project directory to allow GitLab CI to cache it.
-  # By default, Go places modules in GOMODCACHE, often outside the project.
-  # Explicitly setting GOMODCACHE ensures it's within the cached path.
-  GOPATH: "$CI_PROJECT_DIR/.go"
-  GOMODCACHE: "$GOPATH/pkg/mod"
-  GO_BIN_DIR: "$GOPATH/bin"
-
-cache:
-  # Cache Go modules and the binaries.
-  # The 'key' ensures a unique cache per branch, or you can use a fixed key
-  # for a shared cache across all branches if that fits your workflow.
-  key: "$CI_COMMIT_REF_SLUG"
-  paths:
-    - ${GOPATH}/pkg/mod/ # For Go modules
-    - ${GO_BIN_DIR}/
-
-stages: [build, release]
+  GO_VERSION: "go1.24.9"
+  GIT_DEPTH: "0"
 
 default:
   id_tokens:
     VAULT_ID_TOKEN:
       aud: https://vault.cfdata.org
 
-# This before_script is injected into every job that runs on master meaning that if there is no tag the step
-# will succeed but only write "No tag present - Skipping" to the console.
-.check_tag:
-  before_script:
-    - |
-      # Check if there is a Git tag pointing to HEAD
-      echo "Tag found: $(git tag --points-at HEAD | grep .)"
-      if git tag --points-at HEAD | grep .; then
-        echo "Tag found: $(git tag --points-at HEAD | grep .)"
-        export "VERSION=$(git tag --points-at HEAD | grep .)"
-      else
-        echo "No tag present — skipping."
-        exit 0
-      fi
+stages: [sync, pre-build, build, validate, test, package, release, release-internal, review]
 
-## A set of predefined rules to use on the different jobs
-.default_rules:
-  # Rules to run the job only on the master branch
-  run_on_master:
-    - if: $CI_COMMIT_BRANCH == $CI_DEFAULT_BRANCH
-      when: always
-    - when: never
-  # Rules to run the job only on branches that are not master. This is needed because for now
-  # we need to keep a similar behavior due to the integration with teamcity, which requires us
-  # to not trigger pipelines on tags and/or merge requests.
-  run_on_branch:
-    - if: $CI_COMMIT_TAG
-      when: never
-    - if: $CI_PIPELINE_SOURCE != "merge_request_event" && $CI_COMMIT_BRANCH != $CI_DEFAULT_BRANCH
-      when: always
-    - when: never
+include:
+  #####################################################
+  ########## Import Commons Configurations ############
+  #####################################################
+  - local: .ci/commons.gitlab-ci.yml
 
+  #####################################################
+  ########### Sync Repository with Github #############
+  #####################################################
+  - local: .ci/github.gitlab-ci.yml
 
-# Template for Go setup, including caching and installation
-.go_setup:
-  image: docker-registry.cfdata.org/stash/devtools/ci-builders/golang-1.24/master:3090-3e32590@sha256:fc81df4f8322f022d93712ee40bb1e5752fdbe9868d1e5a23fd851ad6fbecb91
-  before_script:
-    - mkdir -p ${GOPATH} ${GOMODCACHE} ${GO_BIN_DIR}
-    - export PATH=$PATH:${GO_BIN_DIR}
-    - go env -w GOMODCACHE=${GOMODCACHE} # Ensure go uses the cached module path
+  #####################################################
+  ############# Build or Fetch CI Image ###############
+  #####################################################
+  - local: .ci/ci-image.gitlab-ci.yml
 
-    # Check if govulncheck is already installed and install it if not
-    - if [ ! -f ${GO_BIN_DIR}/govulncheck ]; then
-      echo "govulncheck not found in cache, installing...";
-      go install golang.org/x/vuln/cmd/govulncheck@latest;
-      else
-      echo "govulncheck found in cache, skipping installation.";
-      fi
+  #####################################################
+  ################## Linux Builds ###################
+  #####################################################
+  - local: .ci/linux.gitlab-ci.yml
 
-# -----------------------------------------------
-# Stage 1: Build on every PR
-# -----------------------------------------------
-build_cloudflared_macos: &build
-  stage: build
-  rules:
-    - !reference [.default_rules, run_on_branch]
-  tags:
-    - "macstadium-${RUNNER_ARCH}"
-  parallel:
-    matrix:
-      - RUNNER_ARCH: [arm, intel]
-  artifacts:
-    paths:
-      - artifacts/*
-  script:
-    - '[ "${RUNNER_ARCH}" = "arm" ] && export TARGET_ARCH=arm64'
-    - '[ "${RUNNER_ARCH}" = "intel" ] && export TARGET_ARCH=amd64'
-    - ARCH=$(uname -m)
-    - echo ARCH=$ARCH - TARGET_ARCH=$TARGET_ARCH
-    - ./.teamcity/mac/install-go.sh
-    - BUILD_SCRIPT=.teamcity/mac/build.sh
-    - if [[ ! -x ${BUILD_SCRIPT} ]] ; then exit ; fi
-    - set -euo pipefail
-    - echo "Executing ${BUILD_SCRIPT}"
-    - exec ${BUILD_SCRIPT}
+  #####################################################
+  ################## Windows Builds ###################
+  #####################################################
+  - local: .ci/windows.gitlab-ci.yml
 
-vulncheck:
-  stage: build
-  extends: .go_setup
-  rules:
-    - !reference [.default_rules, run_on_branch]
-  script:
-    - make vulncheck
+  #####################################################
+  ################### macOS Builds ####################
+  #####################################################
+  - local: .ci/mac.gitlab-ci.yml
 
-# -----------------------------------------------
-# Stage 1: Build and sign only on releases
-# -----------------------------------------------
-build_and_sign_cloudflared_macos:
-  <<: *build
-  rules:
-    - !reference [.default_rules, run_on_master]
-  secrets:
-    APPLE_DEV_CA_CERT:
-      vault: gitlab/cloudflare/tun/cloudflared/_branch/master/apple_dev_ca_cert_v2/data@kv
-      file: false
-    CFD_CODE_SIGN_CERT:
-      vault: gitlab/cloudflare/tun/cloudflared/_branch/master/cfd_code_sign_cert_v2/data@kv
-      file: false
-    CFD_CODE_SIGN_KEY:
-      vault: gitlab/cloudflare/tun/cloudflared/_branch/master/cfd_code_sign_key_v2/data@kv
-      file: false
-    CFD_CODE_SIGN_PASS:
-      vault: gitlab/cloudflare/tun/cloudflared/_branch/master/cfd_code_sign_pass_v2/data@kv
-      file: false
-    CFD_INSTALLER_CERT:
-      vault: gitlab/cloudflare/tun/cloudflared/_branch/master/cfd_installer_cert_v2/data@kv
-      file: false
-    CFD_INSTALLER_KEY:
-      vault: gitlab/cloudflare/tun/cloudflared/_branch/master/cfd_installer_key_v2/data@kv
-      file: false
-    CFD_INSTALLER_PASS:
-      vault: gitlab/cloudflare/tun/cloudflared/_branch/master/cfd_installer_pass_v2/data@kv
-      file: false
+  #####################################################
+  ################# Release Packages ##################
+  #####################################################
+  - local: .ci/release.gitlab-ci.yml
 
-# -----------------------------------------------
-# Stage 2: Release to Github after building and signing
-# -----------------------------------------------
-release_cloudflared_macos_to_github:
-  stage: release
-  image: docker-registry.cfdata.org/stash/tun/docker-images/cloudflared-ci/main:6-8616fe631b76-amd64@sha256:96f4fd05e66cec03e0864c1bcf09324c130d4728eef45ee994716da499183614
-  extends: .check_tag
-  dependencies:
-    - build_and_sign_cloudflared_macos
-  rules:
-    - !reference [.default_rules, run_on_master]
-  cache:
-    paths:
-      - .cache/pip
-  variables:
-    PIP_CACHE_DIR: "$CI_PROJECT_DIR/.cache/pip"
-    KV_NAMESPACE: 380e19aa04314648949b6ad841417ebe
-    KV_ACCOUNT: 5ab4e9dfbd435d24068829fda0077963
-  secrets:
-    KV_API_TOKEN:
-      vault: gitlab/cloudflare/tun/cloudflared/_dev/cfd_kv_api_token/data@kv
-      file: false
-    API_KEY:
-      vault: gitlab/cloudflare/tun/cloudflared/_dev/cfd_github_api_key/data@kv
-      file: false
-  script:
-    - python3 --version ; pip --version  # For debugging
-    - python3 -m venv venv
-    - source venv/bin/activate
-    - pip install pynacl==1.4.0 pygithub==1.55
-    - echo $VERSION
-    - echo $TAG_EXISTS
-    - echo "Running release because tag exists."
-    - make macos-release
+  #####################################################
+  ########## Release Packages Internally ##############
+  #####################################################
+  - local: .ci/apt-internal.gitlab-ci.yml
+
+  #####################################################
+  ############## Manual Claude Review #################
+  #####################################################
+  - component: $CI_SERVER_FQDN/cloudflare/ci/ai/review@~latest
+    inputs:
+      whenToRun: "manual"

.teamcity/windows/builds.ps1

@@ -1,28 +0,0 @@
-Set-StrictMode -Version Latest
-$ErrorActionPreference = "Stop"
-$ProgressPreference = "SilentlyContinue"
-
-# Relative path to working directory
-$CloudflaredDirectory = "go\src\github.com\cloudflare\cloudflared"
-
-cd $CloudflaredDirectory
-
-Write-Output "Building for amd64"
-$env:TARGET_OS = "windows"
-$env:CGO_ENABLED = 1
-$env:TARGET_ARCH = "amd64"
-$env:Path = "$Env:Temp\go\bin;$($env:Path)"
-
-go env
-go version
-
-& make cloudflared
-if ($LASTEXITCODE -ne 0) { throw "Failed to build cloudflared for amd64" }
-copy .\cloudflared.exe .\cloudflared-windows-amd64.exe
-
-Write-Output "Building for 386"
-$env:CGO_ENABLED = 0
-$env:TARGET_ARCH = "386"
-make cloudflared
-if ($LASTEXITCODE -ne 0) { throw "Failed to build cloudflared for 386" }
-copy .\cloudflared.exe .\cloudflared-windows-386.exe

.teamcity/windows/component-test.ps1

@@ -1,47 +0,0 @@
-Set-StrictMode -Version Latest
-$ErrorActionPreference = "Stop"
-$ProgressPreference = "SilentlyContinue"
-
-$WorkingDirectory = Get-Location
-$CloudflaredDirectory = "$WorkingDirectory\go\src\github.com\cloudflare\cloudflared"
-
-go env
-go version
-
-$env:TARGET_OS = "windows"
-$env:CGO_ENABLED = 1
-$env:TARGET_ARCH = "amd64"
-$env:Path = "$Env:Temp\go\bin;$($env:Path)"
-
-python --version
-python -m pip --version
-
-cd $CloudflaredDirectory
-
-go env
-go version
-
-Write-Output "Building cloudflared"
-
-& make cloudflared
-if ($LASTEXITCODE -ne 0) { throw "Failed to build cloudflared" }
-
-echo $LASTEXITCODE
-
-Write-Output "Running unit tests"
-
-# Not testing with race detector because of https://github.com/golang/go/issues/61058
-# We already test it on other platforms
-go test -failfast -v -mod=vendor ./...
-if ($LASTEXITCODE -ne 0) { throw "Failed unit tests" }
-
-Write-Output "Running component tests"
-
-python -m pip --disable-pip-version-check install --upgrade -r component-tests/requirements.txt --use-pep517
-python component-tests/setup.py --type create
-python -m pytest component-tests -o log_cli=true --log-cli-level=INFO
-if ($LASTEXITCODE -ne 0) {
-    python component-tests/setup.py --type cleanup
-    throw "Failed component tests"
-}
-python component-tests/setup.py --type cleanup

.vulnignore

@@ -0,0 +1,3 @@
+# Add vulnerability IDs (e.g., GO-2022-0450) to ignore, one per line.
+# You can also add comments on the same line after the ID.
+GO-2025-3942 # Ignore core-dns vulnerability since we will be removing the proxy-dns feature in the near future

CHANGES.md

@@ -1,3 +1,7 @@
+## 2025.7.1
+### Notices
+- `cloudflared` will no longer officially support Debian and Ubuntu distros that reached end-of-life: `buster`, `bullseye`, `impish`, `trusty`.
+
 ## 2025.1.1
 ### New Features
 - This release introduces the use of new Post Quantum curves and the ability to use Post Quantum curves when running tunnels with the QUIC protocol this applies to non-FIPS and FIPS builds.

Dockerfile

@@ -1,7 +1,7 @@
 # use a builder image for building cloudflare
 ARG TARGET_GOOS
 ARG TARGET_GOARCH
-FROM golang:1.24.4 AS builder
+FROM golang:1.24.9 AS builder
 ENV GO111MODULE=on \
   CGO_ENABLED=0 \
   TARGET_GOOS=${TARGET_GOOS} \
@@ -27,8 +27,11 @@ LABEL org.opencontainers.image.source="https://github.com/cloudflare/cloudflared
 # copy our compiled binary
 COPY --from=builder --chown=nonroot /go/src/github.com/cloudflare/cloudflared/cloudflared /usr/local/bin/
 
-# run as non-privileged user
-USER nonroot
+# run as nonroot user
+# We need to use numeric user id's because Kubernetes doesn't support strings:
+# https://github.com/kubernetes/kubernetes/blob/v1.33.2/pkg/kubelet/kuberuntime/security_context_others.go#L49
+# The `nonroot` user maps to `65532`, from: https://github.com/GoogleContainerTools/distroless/blob/main/common/variables.bzl#L18
+USER 65532:65532
 
 # command / entrypoint of container
 ENTRYPOINT ["cloudflared", "--no-autoupdate"]

Dockerfile.amd64

@@ -1,5 +1,5 @@
 # use a builder image for building cloudflare
-FROM golang:1.24.4 AS builder
+FROM golang:1.24.9 AS builder
 ENV GO111MODULE=on \
   CGO_ENABLED=0 \
   # the CONTAINER_BUILD envvar is used set github.com/cloudflare/cloudflared/metrics.Runtime=virtual
@@ -22,8 +22,11 @@ LABEL org.opencontainers.image.source="https://github.com/cloudflare/cloudflared
 # copy our compiled binary
 COPY --from=builder --chown=nonroot /go/src/github.com/cloudflare/cloudflared/cloudflared /usr/local/bin/
 
-# run as non-privileged user
-USER nonroot
+# run as nonroot user
+# We need to use numeric user id's because Kubernetes doesn't support strings:
+# https://github.com/kubernetes/kubernetes/blob/v1.33.2/pkg/kubelet/kuberuntime/security_context_others.go#L49
+# The `nonroot` user maps to `65532`, from: https://github.com/GoogleContainerTools/distroless/blob/main/common/variables.bzl#L18
+USER 65532:65532
 
 # command / entrypoint of container
 ENTRYPOINT ["cloudflared", "--no-autoupdate"]

Dockerfile.arm64

@@ -1,5 +1,5 @@
 # use a builder image for building cloudflare
-FROM golang:1.24.4 AS builder
+FROM golang:1.24.9 AS builder
 ENV GO111MODULE=on \
   CGO_ENABLED=0 \
   # the CONTAINER_BUILD envvar is used set github.com/cloudflare/cloudflared/metrics.Runtime=virtual
@@ -22,8 +22,11 @@ LABEL org.opencontainers.image.source="https://github.com/cloudflare/cloudflared
 # copy our compiled binary
 COPY --from=builder --chown=nonroot /go/src/github.com/cloudflare/cloudflared/cloudflared /usr/local/bin/
 
-# run as non-privileged user
-USER nonroot
+# run as nonroot user
+# We need to use numeric user id's because Kubernetes doesn't support strings:
+# https://github.com/kubernetes/kubernetes/blob/v1.33.2/pkg/kubelet/kuberuntime/security_context_others.go#L49
+# The `nonroot` user maps to `65532`, from: https://github.com/GoogleContainerTools/distroless/blob/main/common/variables.bzl#L18
+USER 65532:65532
 
 # command / entrypoint of container
 ENTRYPOINT ["cloudflared", "--no-autoupdate"]

Makefile

@@ -24,7 +24,13 @@ else
 	DEB_PACKAGE_NAME := $(BINARY_NAME)
 endif
 
-DATE          := $(shell date -u -r RELEASE_NOTES '+%Y-%m-%d-%H%M UTC')
+# Use git in windows since we don't have access to the `date` tool
+ifeq ($(TARGET_OS), windows)
+	DATE := $(shell git log -1 --format="%ad" --date=format-local:'%Y-%m-%dT%H:%M UTC' -- RELEASE_NOTES)
+else
+	DATE := $(shell date -u -r RELEASE_NOTES '+%Y-%m-%d-%H:%M UTC')
+endif
+
 VERSION_FLAGS := -X "main.Version=$(VERSION)" -X "main.BuildTime=$(DATE)"
 ifdef PACKAGE_MANAGER
 	VERSION_FLAGS := $(VERSION_FLAGS) -X "github.com/cloudflare/cloudflared/cmd/cloudflared/updater.BuiltForPackageManager=$(PACKAGE_MANAGER)"
@@ -64,6 +70,8 @@ else ifeq ($(LOCAL_ARCH),x86_64)
     TARGET_ARCH ?= amd64
 else ifeq ($(LOCAL_ARCH),amd64)
     TARGET_ARCH ?= amd64
+else ifeq ($(LOCAL_ARCH),386)
+    TARGET_ARCH ?= 386
 else ifeq ($(LOCAL_ARCH),i686)
     TARGET_ARCH ?= amd64
 else ifeq ($(shell echo $(LOCAL_ARCH) | head -c 5),armv8)
@@ -120,6 +128,8 @@ endif
 #for FIPS compliance, FPM defaults to MD5.
 RPM_DIGEST := --rpm-digest sha256
 
+GO_TEST_LOG_OUTPUT = /tmp/gotest.log
+
 .PHONY: all
 all: cloudflared test
 
@@ -129,7 +139,7 @@ clean:
 
 .PHONY: vulncheck
 vulncheck:
-	@govulncheck ./...
+	@./.ci/scripts/vuln-check.sh
 
 .PHONY: cloudflared
 cloudflared:
@@ -152,11 +162,9 @@ generate-docker-version:
 
 .PHONY: test
 test: vet
-ifndef CI
-	go test -v -mod=vendor -race $(LDFLAGS) ./...
-else
-	@mkdir -p .cover
-	go test -v -mod=vendor -race $(LDFLAGS) -coverprofile=".cover/c.out" ./...
+	$Q go test -json -v -mod=vendor -race $(LDFLAGS) ./... 2>&1 | tee $(GO_TEST_LOG_OUTPUT)
+ifneq ($(FIPS), true)
+	@go run -mod=readonly github.com/gotesttools/gotestfmt/v2/cmd/gotestfmt@latest -input $(GO_TEST_LOG_OUTPUT)
 endif
 
 .PHONY: cover
@@ -174,7 +182,7 @@ fuzz:
 	@go test -fuzz=FuzzIPDecoder -fuzztime=600s ./packet
 	@go test -fuzz=FuzzICMPDecoder -fuzztime=600s ./packet
 	@go test -fuzz=FuzzSessionWrite -fuzztime=600s ./quic/v3
-	@go test -fuzz=FuzzSessionServe -fuzztime=600s ./quic/v3
+	@go test -fuzz=FuzzSessionRead -fuzztime=600s ./quic/v3
 	@go test -fuzz=FuzzRegistrationDatagram -fuzztime=600s ./quic/v3
 	@go test -fuzz=FuzzPayloadDatagram -fuzztime=600s ./quic/v3
 	@go test -fuzz=FuzzRegistrationResponseDatagram -fuzztime=600s ./quic/v3
@@ -213,10 +221,6 @@ cloudflared-deb: cloudflared cloudflared.1
 cloudflared-rpm: cloudflared cloudflared.1
 	$(call build_package,rpm)
 
-.PHONY: cloudflared-pkg
-cloudflared-pkg: cloudflared cloudflared.1
-	$(call build_package,osxpkg)
-
 .PHONY: cloudflared-msi
 cloudflared-msi:
 	wixl --define Version=$(VERSION) --define Path=$(EXECUTABLE_PATH) --output cloudflared-$(VERSION)-$(TARGET_ARCH).msi cloudflared.wxs
@@ -227,17 +231,18 @@ github-release-dryrun:
 
 .PHONY: github-release
 github-release:
-	python3 github_release.py --path $(PWD)/built_artifacts --release-version $(VERSION)
-	python3 github_message.py --release-version $(VERSION)
-
-.PHONY: macos-release
-macos-release:
 	python3 github_release.py --path $(PWD)/artifacts/ --release-version $(VERSION)
+	python3 github_message.py --release-version $(VERSION)
 
 .PHONY: r2-linux-release
 r2-linux-release:
 	python3 ./release_pkgs.py
 
+.PHONY: r2-next-linux-release
+# Publishes to a separate R2 repository during GPG key rollover, using dual-key signing.
+r2-next-linux-release:
+	python3 ./release_pkgs.py --upload-repo-file
+
 .PHONY: capnp
 capnp:
 	which capnp  # https://capnproto.org/install.html
@@ -246,7 +251,7 @@ capnp:
 
 .PHONY: vet
 vet:
-	go vet -mod=vendor github.com/cloudflare/cloudflared/...
+	$Q go vet -mod=vendor github.com/cloudflare/cloudflared/...
 
 .PHONY: fmt
 fmt:
@@ -255,7 +260,7 @@ fmt:
 
 .PHONY: fmt-check
 fmt-check:
-	@./fmt-check.sh
+	@./.ci/scripts/fmt-check.sh
 
 .PHONY: lint
 lint:
@@ -264,3 +269,23 @@ lint:
 .PHONY: mocks
 mocks:
 	go generate mocks/mockgen.go
+
+.PHONY: ci-build
+ci-build:
+	@GOOS=linux GOARCH=amd64 $(MAKE) cloudflared
+	@mkdir -p artifacts
+	@mv cloudflared artifacts/cloudflared
+
+.PHONY: ci-fips-build
+ci-fips-build:
+	@FIPS=true GOOS=linux GOARCH=amd64 $(MAKE) cloudflared
+	@mkdir -p artifacts
+	@mv cloudflared artifacts/cloudflared
+
+.PHONY: ci-test
+ci-test: fmt-check lint test
+	@go run -mod=readonly github.com/jstemmer/go-junit-report/v2@latest -in $(GO_TEST_LOG_OUTPUT) -parser gojson -out report.xml -set-exit-code
+
+.PHONY: ci-fips-test
+ci-fips-test:
+	@FIPS=true $(MAKE) ci-test

README.md

@@ -3,14 +3,14 @@
 Contains the command-line client for Cloudflare Tunnel, a tunneling daemon that proxies traffic from the Cloudflare network to your origins.
 This daemon sits between Cloudflare network and your origin (e.g. a webserver). Cloudflare attracts client requests and sends them to you
 via this daemon, without requiring you to poke holes on your firewall --- your origin can remain as closed as possible.
-Extensive documentation can be found in the [Cloudflare Tunnel section](https://developers.cloudflare.com/cloudflare-one/connections/connect-apps) of the Cloudflare Docs.
+Extensive documentation can be found in the [Cloudflare Tunnel section](https://developers.cloudflare.com/cloudflare-one/networks/connectors/cloudflare-tunnel) of the Cloudflare Docs.
 All usages related with proxying to your origins are available under `cloudflared tunnel help`.
 
 You can also use `cloudflared` to access Tunnel origins (that are protected with `cloudflared tunnel`) for TCP traffic
 at Layer 4 (i.e., not HTTP/websocket), which is relevant for use cases such as SSH, RDP, etc.
 Such usages are available under `cloudflared access help`.
 
-You can instead use [WARP client](https://developers.cloudflare.com/cloudflare-one/connections/connect-apps/configuration/private-networks)
+You can instead use [WARP client](https://developers.cloudflare.com/cloudflare-one/team-and-resources/devices/warp/)
 to access private origins behind Tunnels for Layer 4 traffic without requiring `cloudflared access` commands on the client side.
 
 
@@ -19,41 +19,41 @@ to access private origins behind Tunnels for Layer 4 traffic without requiring `
 Before you use Cloudflare Tunnel, you'll need to complete a few steps in the Cloudflare dashboard: you need to add a
 website to your Cloudflare account. Note that today it is possible to use Tunnel without a website (e.g. for private
 routing), but for legacy reasons this requirement is still necessary:
-1. [Add a website to Cloudflare](https://support.cloudflare.com/hc/en-us/articles/201720164-Creating-a-Cloudflare-account-and-adding-a-website)
-2. [Change your domain nameservers to Cloudflare](https://support.cloudflare.com/hc/en-us/articles/205195708)
+1. [Add a website to Cloudflare](https://developers.cloudflare.com/fundamentals/manage-domains/add-site/)
+2. [Change your domain nameservers to Cloudflare](https://developers.cloudflare.com/dns/zone-setups/full-setup/setup/)
 
 
 ## Installing `cloudflared`
 
 Downloads are available as standalone binaries, a Docker image, and Debian, RPM, and Homebrew packages. You can also find releases [here](https://github.com/cloudflare/cloudflared/releases) on the `cloudflared` GitHub repository.
 
-* You can [install on macOS](https://developers.cloudflare.com/cloudflare-one/connections/connect-apps/install-and-setup/installation#macos) via Homebrew or by downloading the [latest Darwin amd64 release](https://github.com/cloudflare/cloudflared/releases)
-* Binaries, Debian, and RPM packages for Linux [can be found here](https://developers.cloudflare.com/cloudflare-one/connections/connect-apps/install-and-setup/installation#linux)
+* You can [install on macOS](https://developers.cloudflare.com/cloudflare-one/networks/connectors/cloudflare-tunnel/downloads/#macos) via Homebrew or by downloading the [latest Darwin amd64 release](https://github.com/cloudflare/cloudflared/releases)
+* Binaries, Debian, and RPM packages for Linux [can be found here](https://developers.cloudflare.com/cloudflare-one/networks/connectors/cloudflare-tunnel/downloads/#linux)
 * A Docker image of `cloudflared` is [available on DockerHub](https://hub.docker.com/r/cloudflare/cloudflared)
-* You can install on Windows machines with the [steps here](https://developers.cloudflare.com/cloudflare-one/connections/connect-apps/install-and-setup/installation#windows)
+* You can install on Windows machines with the [steps here](https://developers.cloudflare.com/cloudflare-one/networks/connectors/cloudflare-tunnel/downloads/#windows)
 * To build from source, install the required version of go, mentioned in the [Development](#development) section below. Then you can run `make cloudflared`.
 
-User documentation for Cloudflare Tunnel can be found at https://developers.cloudflare.com/cloudflare-one/connections/connect-apps
+User documentation for Cloudflare Tunnel can be found at https://developers.cloudflare.com/cloudflare-one/networks/connectors/cloudflare-tunnel/
 
 
 ## Creating Tunnels and routing traffic
 
 Once installed, you can authenticate `cloudflared` into your Cloudflare account and begin creating Tunnels to serve traffic to your origins.
 
-* Create a Tunnel with [these instructions](https://developers.cloudflare.com/cloudflare-one/connections/connect-networks/get-started/)
+* Create a Tunnel with [these instructions](https://developers.cloudflare.com/cloudflare-one/networks/connectors/cloudflare-tunnel/get-started/)
 * Route traffic to that Tunnel:
-  * Via public [DNS records in Cloudflare](https://developers.cloudflare.com/cloudflare-one/connections/connect-apps/routing-to-tunnel/dns)
-  * Or via a public hostname guided by a [Cloudflare Load Balancer](https://developers.cloudflare.com/cloudflare-one/connections/connect-apps/routing-to-tunnel/lb)
-  * Or from [WARP client private traffic](https://developers.cloudflare.com/cloudflare-one/connections/connect-networks/private-net/)
+  * Via public [DNS records in Cloudflare](https://developers.cloudflare.com/cloudflare-one/networks/connectors/cloudflare-tunnel/routing-to-tunnel/dns/)
+  * Or via a public hostname guided by a [Cloudflare Load Balancer](https://developers.cloudflare.com/cloudflare-one/networks/connectors/cloudflare-tunnel/routing-to-tunnel/public-load-balancers/)
+  * Or from [WARP client private traffic](https://developers.cloudflare.com/cloudflare-one/networks/connectors/cloudflare-tunnel/private-net/)
 
 
 ## TryCloudflare
 
-Want to test Cloudflare Tunnel before adding a website to Cloudflare? You can do so with TryCloudflare using the documentation [available here](https://developers.cloudflare.com/cloudflare-one/connections/connect-networks/do-more-with-tunnels/trycloudflare/).
+Want to test Cloudflare Tunnel before adding a website to Cloudflare? You can do so with TryCloudflare using the documentation [available here](https://developers.cloudflare.com/cloudflare-one/networks/connectors/cloudflare-tunnel/do-more-with-tunnels/trycloudflare/).
 
 ## Deprecated versions
 
-Cloudflare currently supports versions of cloudflared that are **within one year** of the most recent release. Breaking changes unrelated to feature availability may be introduced that will impact versions released more than one year ago. You can read more about upgrading cloudflared in our [developer documentation](https://developers.cloudflare.com/cloudflare-one/connections/connect-networks/downloads/#updating-cloudflared).
+Cloudflare currently supports versions of cloudflared that are **within one year** of the most recent release. Breaking changes unrelated to feature availability may be introduced that will impact versions released more than one year ago. You can read more about upgrading cloudflared in our [developer documentation](https://developers.cloudflare.com/cloudflare-one/networks/connectors/cloudflare-tunnel/downloads/update-cloudflared/).
 
 For example, as of January 2023 Cloudflare will support cloudflared version 2023.1.1 to cloudflared 2022.1.1.
 

RELEASE_NOTES

@@ -1,3 +1,70 @@
+2025.11.1
+- 2025-11-07 TUN-9800: Fix docker hub push step
+
+2025.11.0
+- 2025-11-06 TUN-9863: Introduce Code Signing for Windows Builds
+- 2025-11-06 TUN-9800: Prefix gitlab steps with operating system
+- 2025-11-04 chore: Update cloudflared signing key name in index.html
+- 2025-10-31 chore: add claude review
+- 2025-10-31 Chore: Update documentation links in README
+- 2025-10-31 TUN-9800: Add pipelines for linux packaging
+
+2025.10.1
+- 2025-10-30 chore: Update ci image to use goboring 1.24.9
+- 2025-10-28 TUN-9849: Add cf-proxy-* to control response headers
+- 2025-10-24 TUN-9961: Add pkg.cloudflared.com index.html to git repo
+- 2025-10-23 TUN-9954: Update from go1.24.6 to go1.24.9
+- 2025-10-23 Fix systemd service installation hanging
+- 2025-10-21 TUN-9941: Use new GPG key for RPM builds
+- 2025-10-21 TUN-9941: Fix typo causing r2-release-next deployment to fail
+- 2025-10-21 TUN-9941: Lookup correct key for RPM signature
+- 2025-10-15 TUN-9919: Make RPM postinstall scriplet idempotent
+- 2025-10-14 TUN-9916: Fix the cloudflared binary path used in the component test
+
+2025.10.0
+- 2025-10-14 chore: Fix upload of RPM repo file during double signing
+- 2025-10-13 TUN-9882: Bump datagram v3 write channel capacity
+- 2025-10-10 chore: Fix import of GPG keys when two keys are provided
+- 2025-10-10 chore: Fix parameter order when uploading RPM .repo file to R2
+- 2025-10-10 TUN-9883: Add new datagram v3 feature flag
+- 2025-10-09 chore: Force usage of go-boring 1.24
+- 2025-10-08 TUN-9882: Improve metrics for datagram v3
+- 2025-10-07 GRC-16749: Add fedramp tags to catalog
+- 2025-10-07 TUN-9882: Add buffers for UDP and ICMP datagrams in datagram v3
+- 2025-10-07 TUN-9882: Add write deadline for UDP origin writes
+- 2025-09-29 TUN-9776: Support signing Debian packages with two keys for rollover
+- 2025-09-22 TUN-9800: Add pipeline to sync between gitlab and github repos
+
+2025.9.1
+- 2025-09-22 TUN-9855: Create script to ignore vulnerabilities from govuln check
+- 2025-09-19 TUN-9852: Remove fmt.Println from cloudflared access command
+
+2025.9.0
+- 2025-09-15 TUN-9820: Add support for FedRAMP in originRequest Access config
+- 2025-09-11 TUN-9800: Migrate cloudflared-ci pipelines to Gitlab CI
+- 2025-09-04 TUN-9803: Add windows builds to gitlab-ci
+- 2025-08-27 TUN-9755: Set endpoint in tunnel credentials when generating locally managed tunnel with a Fed token
+
+2025.8.1
+- 2025-08-19 AUTH-7480 update fed callback url for login helper
+- 2025-08-19 CUSTESC-53681: Correct QUIC connection management for datagram handlers
+- 2025-08-12 AUTH-7260: Add support for login interstitial auto closure
+
+2025.8.0
+- 2025-08-07 vuln: Fix GO-2025-3770 vulnerability
+- 2025-07-23 TUN-9583: set proper url and hostname for cloudflared tail command
+- 2025-07-07 TUN-9542: Remove unsupported Debian-based releases
+
+2025.7.0
+- 2025-07-03 TUN-9540: Use numeric user id for Dockerfiles
+- 2025-07-01 TUN-9161: Remove P256Kyber768Draft00PQKex curve from nonFips curve preferences
+- 2025-07-01 TUN-9531: Bump go-boring from 1.24.2 to 1.24.4
+- 2025-07-01 TUN-9511: Add metrics for virtual DNS origin
+- 2025-06-30 TUN-9470: Add OriginDialerService to include TCP
+- 2025-06-30 TUN-9473: Add --dns-resolver-addrs flag
+- 2025-06-27 TUN-9472: Add virtual DNS service
+- 2025-06-23 TUN-9469: Centralize UDP origin proxy dialing as ingress service
+
 2025.6.1
 - 2025-06-16 TUN-9467: add vulncheck to cloudflared
 - 2025-06-16 TUN-9495: Remove references to cloudflare-go

build-packages.sh

@@ -1,48 +0,0 @@
-#!/bin/bash
-VERSION=$(git describe --tags --always --match "[0-9][0-9][0-9][0-9].*.*")
-echo $VERSION
-
-# Disable FIPS module in go-boring
-export GOEXPERIMENT=noboringcrypto
-export CGO_ENABLED=0
-
-# This controls the directory the built artifacts go into
-export ARTIFACT_DIR=artifacts/
-mkdir -p $ARTIFACT_DIR
-
-linuxArchs=("386" "amd64" "arm" "armhf" "arm64")
-export TARGET_OS=linux
-for arch in ${linuxArchs[@]}; do
-    unset TARGET_ARM
-    export TARGET_ARCH=$arch
-
-    ## Support for arm platforms without hardware FPU enabled
-    if [[ $arch == arm ]] ; then
-        export TARGET_ARCH=arm
-        export TARGET_ARM=5
-    fi
-    
-    ## Support for armhf builds 
-    if [[ $arch == armhf ]] ; then
-        export TARGET_ARCH=arm
-        export TARGET_ARM=7 
-    fi
-    
-    make cloudflared-deb
-    mv cloudflared\_$VERSION\_$arch.deb $ARTIFACT_DIR/cloudflared-linux-$arch.deb
-
-    # rpm packages invert the - and _ and use x86_64 instead of amd64.
-    RPMVERSION=$(echo $VERSION|sed -r 's/-/_/g')
-    RPMARCH=$arch
-    if [ $arch == "amd64" ];then
-        RPMARCH="x86_64"
-    fi
-    if [ $arch == "arm64" ]; then
-        RPMARCH="aarch64"
-    fi
-    make cloudflared-rpm
-    mv cloudflared-$RPMVERSION-1.$RPMARCH.rpm $ARTIFACT_DIR/cloudflared-linux-$RPMARCH.rpm
-
-    # finally move the linux binary as well.
-    mv ./cloudflared $ARTIFACT_DIR/cloudflared-linux-$arch
-done

carrier/carrier.go

@@ -26,11 +26,13 @@ const (
 )
 
 type StartOptions struct {
-	AppInfo         *token.AppInfo
-	OriginURL       string
-	Headers         http.Header
-	Host            string
-	TLSClientConfig *tls.Config
+	AppInfo               *token.AppInfo
+	OriginURL             string
+	Headers               http.Header
+	Host                  string
+	TLSClientConfig       *tls.Config
+	AutoCloseInterstitial bool
+	IsFedramp             bool
 }
 
 // Connection wraps up all the needed functions to forward over the tunnel
@@ -46,7 +48,6 @@ type StdinoutStream struct{}
 // Read will read from Stdin
 func (c *StdinoutStream) Read(p []byte) (int, error) {
 	return os.Stdin.Read(p)
-
 }
 
 // Write will write to Stdout
@@ -139,7 +140,7 @@ func BuildAccessRequest(options *StartOptions, log *zerolog.Logger) (*http.Reque
 		return nil, err
 	}
 
-	token, err := token.FetchTokenWithRedirect(req.URL, options.AppInfo, log)
+	token, err := token.FetchTokenWithRedirect(req.URL, options.AppInfo, options.AutoCloseInterstitial, options.IsFedramp, log)
 	if err != nil {
 		return nil, err
 	}

catalog-info.yaml

@@ -14,4 +14,7 @@ spec:
   lifecycle: "Active"
   owner: "teams/tunnel-teams-routing"
   cf:
+    compliance:
+      fedramp-high: "pending"
+      fedramp-moderate: "yes"
     FIPS: "required"

cfsetup.yaml

@@ -1,257 +1,2 @@
-pinned_go: &pinned_go go-boring=1.24.2-1
-
-build_dir: &build_dir /cfsetup_build
-default-flavor: bookworm
-
-bullseye: &bullseye
-  build-linux:
-    build_dir: *build_dir
-    builddeps: &build_deps
-      - *pinned_go
-      - build-essential
-      - fakeroot
-      - rubygem-fpm
-      - rpm
-      - libffi-dev
-      - golangci-lint=1.64.8-2
-    pre-cache: &build_pre_cache
-      - export GOCACHE=/cfsetup_build/.cache/go-build
-      - go install golang.org/x/tools/cmd/goimports@v0.30.0
-    post-cache:
-      # Linting
-      - make lint
-      - make fmt-check
-      # Build binary for component test
-      - GOOS=linux GOARCH=amd64 make cloudflared
-  build-linux-fips:
-    build_dir: *build_dir
-    builddeps: *build_deps
-    pre-cache: *build_pre_cache
-    post-cache:
-      - export FIPS=true
-      # Build binary for component test
-      - GOOS=linux GOARCH=amd64 make cloudflared
-  cover:
-    build_dir: *build_dir
-    builddeps: *build_deps
-    pre-cache: *build_pre_cache
-    post-cache:
-      - make cover
-  # except FIPS and macos
-  build-linux-release:
-    build_dir: *build_dir
-    builddeps: &build_deps_release
-      - *pinned_go
-      - build-essential
-      - fakeroot
-      - rubygem-fpm
-      - rpm
-      - libffi-dev
-      - python3-dev
-      - python3-pip
-      - python3-setuptools
-      - wget
-      - python3-venv
-    post-cache:
-      - python3 -m venv env
-      - . /cfsetup_build/env/bin/activate
-      - pip install pynacl==1.4.0 pygithub==1.55 boto3==1.22.9 python-gnupg==0.4.9
-      # build all packages (except macos and FIPS) and move them to /cfsetup/built_artifacts
-      - ./build-packages.sh
-  # handle FIPS separately so that we built with gofips compiler
-  build-linux-fips-release:
-    build_dir: *build_dir
-    builddeps: *build_deps_release
-    post-cache:
-      # same logic as above, but for FIPS packages only
-      - ./build-packages-fips.sh
-  generate-versions-file:
-    build_dir: *build_dir
-    builddeps:
-      - *pinned_go
-      - build-essential
-    post-cache:
-      - make generate-docker-version
-  build-deb:
-    build_dir: *build_dir
-    builddeps: &build_deb_deps
-      - *pinned_go
-      - build-essential
-      - fakeroot
-      - rubygem-fpm
-    post-cache:
-      - export GOOS=linux
-      - export GOARCH=amd64
-      - make cloudflared-deb
-  build-fips-internal-deb:
-    build_dir: *build_dir
-    builddeps: &build_fips_deb_deps
-      - *pinned_go
-      - build-essential
-      - fakeroot
-      - rubygem-fpm
-    post-cache:
-      - export GOOS=linux
-      - export GOARCH=amd64
-      - export FIPS=true
-      - export ORIGINAL_NAME=true
-      - make cloudflared-deb
-  build-internal-deb-nightly-amd64:
-    build_dir: *build_dir
-    builddeps: *build_fips_deb_deps
-    post-cache:
-      - export GOOS=linux
-      - export GOARCH=amd64
-      - export NIGHTLY=true
-      - export FIPS=true
-      - export ORIGINAL_NAME=true
-      - make cloudflared-deb
-  build-internal-deb-nightly-arm64:
-    build_dir: *build_dir
-    builddeps: *build_fips_deb_deps
-    post-cache:
-      - export GOOS=linux
-      - export GOARCH=arm64
-      - export NIGHTLY=true
-      # - export FIPS=true # TUN-7595
-      - export ORIGINAL_NAME=true
-      - make cloudflared-deb
-  build-deb-arm64:
-    build_dir: *build_dir
-    builddeps: *build_deb_deps
-    post-cache:
-      - export GOOS=linux
-      - export GOARCH=arm64
-      - make cloudflared-deb
-  package-windows:
-    build_dir: *build_dir
-    builddeps:
-      - *pinned_go
-      - build-essential
-      - python3-dev
-      - libffi-dev
-      - python3-setuptools
-      - python3-pip
-      - wget
-      # libmsi and libgcab are libraries the wixl binary depends on.
-      - libmsi-dev
-      - libgcab-dev
-      - python3-venv
-    pre-cache:
-      - wget https://github.com/sudarshan-reddy/msitools/releases/download/v0.101b/wixl -P /usr/local/bin
-      - chmod a+x /usr/local/bin/wixl
-    post-cache:
-      - python3 -m venv env
-      - . env/bin/activate
-      - pip install pynacl==1.4.0 pygithub==1.55
-      - .teamcity/package-windows.sh
-  test:
-    build_dir: *build_dir
-    builddeps: &build_deps_tests
-      - *pinned_go
-      - build-essential
-      - fakeroot
-      - rubygem-fpm
-      - rpm
-      - libffi-dev
-      - gotest-to-teamcity
-    pre-cache: *build_pre_cache
-    post-cache:
-      - export GOOS=linux
-      - export GOARCH=amd64
-      - export PATH="$HOME/go/bin:$PATH"
-      - make test | gotest-to-teamcity
-  test-fips:
-    build_dir: *build_dir
-    builddeps: *build_deps_tests
-    pre-cache: *build_pre_cache
-    post-cache:
-      - export GOOS=linux
-      - export GOARCH=amd64
-      - export FIPS=true
-      - export PATH="$HOME/go/bin:$PATH"
-      - make test | gotest-to-teamcity
-  component-test:
-    build_dir: *build_dir
-    builddeps: &build_deps_component_test
-      - *pinned_go
-      - python3
-      - python3-pip
-      - python3-setuptools
-      # procps installs the ps command which is needed in test_sysv_service
-      # because the init script uses ps pid to determine if the agent is
-      # running
-      - procps
-      - python3-venv
-    pre-cache-copy-paths:
-      - component-tests/requirements.txt
-    post-cache: &component_test_post_cache
-      - python3 -m venv env
-      - . env/bin/activate
-      - pip install --upgrade -r component-tests/requirements.txt
-      # Creates and routes a Named Tunnel for this build. Also constructs
-      # config file from env vars.
-      - python3 component-tests/setup.py --type create
-      - pytest component-tests -o log_cli=true --log-cli-level=INFO
-      # The Named Tunnel is deleted and its route unprovisioned here.
-      - python3 component-tests/setup.py --type cleanup
-  component-test-fips:
-    build_dir: *build_dir
-    builddeps: *build_deps_component_test
-    pre-cache-copy-paths:
-      - component-tests/requirements.txt
-    post-cache: *component_test_post_cache
-  github-release-dryrun:
-    build_dir: *build_dir
-    builddeps:
-      - *pinned_go
-      - build-essential
-      - python3-dev
-      - libffi-dev
-      - python3-setuptools
-      - python3-pip
-      - python3-venv
-    post-cache:
-      - python3 -m venv env
-      - . env/bin/activate
-      - pip install pynacl==1.4.0 pygithub==1.55
-      - make github-release-dryrun
-  github-release:
-    build_dir: *build_dir
-    builddeps:
-      - *pinned_go
-      - build-essential
-      - python3-dev
-      - libffi-dev
-      - python3-setuptools
-      - python3-pip
-      - python3-venv
-    post-cache:
-      - python3 -m venv env
-      - . env/bin/activate
-      - pip install pynacl==1.4.0 pygithub==1.55
-      - make github-release
-  r2-linux-release:
-    build_dir: *build_dir
-    builddeps:
-      - *pinned_go
-      - build-essential
-      - fakeroot
-      - rubygem-fpm
-      - rpm
-      - wget
-      - python3-dev
-      - libffi-dev
-      - python3-setuptools
-      - python3-pip
-      - reprepro
-      - createrepo-c
-      - python3-venv
-    post-cache:
-      - python3 -m venv env
-      - . env/bin/activate
-      - pip install pynacl==1.4.0 pygithub==1.55 boto3==1.22.9 python-gnupg==0.4.9
-      - make r2-linux-release
-
-bookworm: *bullseye
-trixie: *bullseye
+# A valid cfsetup.yaml is required but we dont have any real config to specify
+dummy_key: true

client/config_test.go

@@ -45,6 +45,6 @@ func (m *mockFeatureSelector) Snapshot() features.FeatureSnapshot {
 	return features.FeatureSnapshot{
 		PostQuantum:     features.PostQuantumPrefer,
 		DatagramVersion: features.DatagramV3,
-		FeaturesList:    []string{features.FeaturePostQuantum, features.FeatureDatagramV3_1},
+		FeaturesList:    []string{features.FeaturePostQuantum, features.FeatureDatagramV3_2},
 	}
 }

cmd/cloudflared/access/carrier.go

@@ -47,6 +47,7 @@ func StartForwarder(forwarder config.Forwarder, shutdown <-chan struct{}, log *z
 	options := &carrier.StartOptions{
 		OriginURL: forwarder.URL,
 		Headers:   headers, //TODO: TUN-2688 support custom headers from config file
+		IsFedramp: forwarder.IsFedramp,
 	}
 
 	// we could add a cmd line variable for this bool if we want the SOCK5 server to be on the client side
@@ -92,6 +93,7 @@ func ssh(c *cli.Context) error {
 		OriginURL: url.String(),
 		Headers:   headers,
 		Host:      url.Host,
+		IsFedramp: c.Bool(fedrampFlag),
 	}
 
 	if connectTo := c.String(sshConnectTo); connectTo != "" {

cmd/cloudflared/access/cmd.go

@@ -51,6 +51,7 @@ Host {{.Hostname}}
   ProxyCommand {{.Cloudflared}} access ssh --hostname %h
 {{end}}
 `
+	fedrampFlag = "fedramp"
 )
 
 const sentryDSN = "https://56a9c9fa5c364ab28f34b14f35ea0f1b@sentry.io/189878"
@@ -79,6 +80,10 @@ func Commands() []*cli.Command {
 			Aliases:  []string{"forward"},
 			Category: "Access",
 			Usage:    "access <subcommand>",
+			Flags: []cli.Flag{&cli.BoolFlag{
+				Name:  fedrampFlag,
+				Usage: "use when performing operations in fedramp account",
+			}},
 			Description: `Cloudflare Access protects internal resources by securing, authenticating and monitoring access
 			per-user and by application. With Cloudflare Access, only authenticated users with the required permissions are
 			able to reach sensitive resources. The commands provided here allow you to interact with Access protected
@@ -104,6 +109,10 @@ func Commands() []*cli.Command {
 							Name:  "no-verbose",
 							Usage: "print only the jwt to stdout",
 						},
+						&cli.BoolFlag{
+							Name:  "auto-close",
+							Usage: "automatically close the auth interstitial after action",
+						},
 						&cli.StringFlag{
 							Name: appURLFlag,
 						},
@@ -322,7 +331,7 @@ func curl(c *cli.Context) error {
 			log.Info().Msg("You don't have an Access token set. Please run access token <access application> to fetch one.")
 			return run("curl", cmdArgs...)
 		}
-		tok, err = token.FetchToken(appURL, appInfo, log)
+		tok, err = token.FetchToken(appURL, appInfo, c.Bool(cfdflags.AutoCloseInterstitial), c.Bool(fedrampFlag), log)
 		if err != nil {
 			log.Err(err).Msg("Failed to refresh token")
 			return err
@@ -442,7 +451,7 @@ func sshGen(c *cli.Context) error {
 	if err != nil {
 		return err
 	}
-	cfdToken, err := token.FetchTokenWithRedirect(fetchTokenURL, appInfo, log)
+	cfdToken, err := token.FetchTokenWithRedirect(fetchTokenURL, appInfo, c.Bool(cfdflags.AutoCloseInterstitial), c.Bool(fedrampFlag), log)
 	if err != nil {
 		return err
 	}
@@ -542,7 +551,7 @@ func verifyTokenAtEdge(appUrl *url.URL, appInfo *token.AppInfo, c *cli.Context,
 	if c.IsSet(sshTokenSecretFlag) {
 		headers.Add(cfAccessClientSecretHeader, c.String(sshTokenSecretFlag))
 	}
-	options := &carrier.StartOptions{AppInfo: appInfo, OriginURL: appUrl.String(), Headers: headers}
+	options := &carrier.StartOptions{AppInfo: appInfo, OriginURL: appUrl.String(), Headers: headers, AutoCloseInterstitial: c.Bool(cfdflags.AutoCloseInterstitial), IsFedramp: c.Bool(fedrampFlag)}
 
 	if valid, err := isTokenValid(options, log); err != nil {
 		return err

cmd/cloudflared/flags/flags.go

@@ -157,4 +157,13 @@ const (
 
 	// ApiURL is the command line flag used to define the base URL of the API
 	ApiURL = "api-url"
+
+	// Virtual DNS resolver service resolver addresses to use instead of dynamically fetching them from the OS.
+	VirtualDNSServiceResolverAddresses = "dns-resolver-addrs"
+
+	// Management hostname to signify incoming management requests
+	ManagementHostname = "management-hostname"
+
+	// Automatically close the login interstitial browser window after the user makes a decision.
+	AutoCloseInterstitial = "auto-close"
 )

cmd/cloudflared/linux_service.go

@@ -60,7 +60,7 @@ After=network-online.target
 Wants=network-online.target
 
 [Service]
-TimeoutStartSec=0
+TimeoutStartSec=15
 Type=notify
 ExecStart={{ .Path }} --no-autoupdate{{ range .ExtraArgs }} {{ . }}{{ end }}
 Restart=on-failure

cmd/cloudflared/tail/cmd.go

@@ -51,6 +51,7 @@ func buildTailManagementTokenSubcommand() *cli.Command {
 
 func managementTokenCommand(c *cli.Context) error {
 	log := createLogger(c)
+
 	token, err := getManagementToken(c, log)
 	if err != nil {
 		return err
@@ -99,7 +100,7 @@ func buildTailCommand(subcommands []*cli.Command) *cli.Command {
 				EnvVars: []string{"TUNNEL_MANAGEMENT_TOKEN"},
 			},
 			&cli.StringFlag{
-				Name:    "management-hostname",
+				Name:    cfdflags.ManagementHostname,
 				Usage:   "Management hostname to signify incoming management requests",
 				EnvVars: []string{"TUNNEL_MANAGEMENT_HOSTNAME"},
 				Hidden:  true,
@@ -236,7 +237,14 @@ func getManagementToken(c *cli.Context, log *zerolog.Logger) (string, error) {
 		return "", err
 	}
 
-	client, err := userCreds.Client(c.String(cfdflags.ApiURL), buildInfo.UserAgent(), log)
+	var apiURL string
+	if userCreds.IsFEDEndpoint() {
+		apiURL = credentials.FedRampBaseApiURL
+	} else {
+		apiURL = c.String(cfdflags.ApiURL)
+	}
+
+	client, err := userCreds.Client(apiURL, buildInfo.UserAgent(), log)
 	if err != nil {
 		return "", err
 	}
@@ -261,7 +269,7 @@ func getManagementToken(c *cli.Context, log *zerolog.Logger) (string, error) {
 // buildURL will build the management url to contain the required query parameters to authenticate the request.
 func buildURL(c *cli.Context, log *zerolog.Logger) (url.URL, error) {
 	var err error
-	managementHostname := c.String("management-hostname")
+
 	token := c.String("token")
 	if token == "" {
 		token, err = getManagementToken(c, log)
@@ -269,6 +277,19 @@ func buildURL(c *cli.Context, log *zerolog.Logger) (url.URL, error) {
 			return url.URL{}, fmt.Errorf("unable to acquire management token for requested tunnel id: %w", err)
 		}
 	}
+
+	claims, err := management.ParseToken(token)
+	if err != nil {
+		return url.URL{}, fmt.Errorf("failed to determine if token is FED: %w", err)
+	}
+
+	var managementHostname string
+	if claims.IsFed() {
+		managementHostname = credentials.FedRampHostname
+	} else {
+		managementHostname = c.String(cfdflags.ManagementHostname)
+	}
+
 	query := url.Values{}
 	query.Add("access_token", token)
 	connector := c.String("connector-id")

cmd/cloudflared/tunnel/cmd.go

@@ -97,7 +97,7 @@ var (
 		"no-tls-verify",
 		"no-chunked-encoding",
 		"http2-origin",
-		"management-hostname",
+		cfdflags.ManagementHostname,
 		"service-op-ip",
 		"local-ssh-port",
 		"ssh-idle-timeout",
@@ -459,8 +459,23 @@ func StartServer(
 		}
 	}
 
+	userCreds, err := credentials.Read(c.String(cfdflags.OriginCert), log)
+	var isFEDEndpoint bool
+	if err != nil {
+		isFEDEndpoint = false
+	} else {
+		isFEDEndpoint = userCreds.IsFEDEndpoint()
+	}
+
+	var managementHostname string
+	if isFEDEndpoint {
+		managementHostname = credentials.FedRampHostname
+	} else {
+		managementHostname = c.String(cfdflags.ManagementHostname)
+	}
+
 	mgmt := management.New(
-		c.String("management-hostname"),
+		managementHostname,
 		c.Bool("management-diagnostics"),
 		serviceIP,
 		connectorID,
@@ -1042,7 +1057,7 @@ func configureProxyFlags(shouldHide bool) []cli.Flag {
 			Value:   false,
 		}),
 		altsrc.NewStringFlag(&cli.StringFlag{
-			Name:    "management-hostname",
+			Name:    cfdflags.ManagementHostname,
 			Usage:   "Management hostname to signify incoming management requests",
 			EnvVars: []string{"TUNNEL_MANAGEMENT_HOSTNAME"},
 			Hidden:  true,

cmd/cloudflared/tunnel/configuration.go

@@ -11,6 +11,7 @@ import (
 	"time"
 
 	"github.com/pkg/errors"
+	"github.com/prometheus/client_golang/prometheus"
 	"github.com/rs/zerolog"
 	"github.com/urfave/cli/v2"
 	"github.com/urfave/cli/v2/altsrc"
@@ -25,6 +26,7 @@ import (
 	"github.com/cloudflare/cloudflared/edgediscovery/allregions"
 	"github.com/cloudflare/cloudflared/features"
 	"github.com/cloudflare/cloudflared/ingress"
+	"github.com/cloudflare/cloudflared/ingress/origins"
 	"github.com/cloudflare/cloudflared/orchestration"
 	"github.com/cloudflare/cloudflared/supervisor"
 	"github.com/cloudflare/cloudflared/tlsconfig"
@@ -34,7 +36,6 @@ import (
 const (
 	secretValue       = "*****"
 	icmpFunnelTimeout = time.Second * 10
-	fedRampRegion     = "fed" // const string denoting the region used to connect to FEDRamp servers
 )
 
 var (
@@ -219,6 +220,27 @@ func prepareTunnelConfig(
 		resolvedRegion = endpoint
 	}
 
+	warpRoutingConfig := ingress.NewWarpRoutingConfig(&cfg.WarpRouting)
+
+	// Setup origin dialer service and virtual services
+	originDialerService := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   ingress.NewDialer(warpRoutingConfig),
+		TCPWriteTimeout: c.Duration(flags.WriteStreamTimeout),
+	}, log)
+
+	// Setup DNS Resolver Service
+	originMetrics := origins.NewMetrics(prometheus.DefaultRegisterer)
+	dnsResolverAddrs := c.StringSlice(flags.VirtualDNSServiceResolverAddresses)
+	dnsService := origins.NewDNSResolverService(origins.NewDNSDialer(), log, originMetrics)
+	if len(dnsResolverAddrs) > 0 {
+		addrs, err := parseResolverAddrPorts(dnsResolverAddrs)
+		if err != nil {
+			return nil, nil, fmt.Errorf("invalid %s provided: %w", flags.VirtualDNSServiceResolverAddresses, err)
+		}
+		dnsService = origins.NewStaticDNSResolverService(addrs, origins.NewDNSDialer(), log, originMetrics)
+	}
+	originDialerService.AddReservedService(dnsService, []netip.AddrPort{origins.VirtualDNSServiceAddr})
+
 	tunnelConfig := &supervisor.TunnelConfig{
 		ClientConfig:    clientConfig,
 		GracePeriod:     gracePeriod,
@@ -246,6 +268,8 @@ func prepareTunnelConfig(
 		DisableQUICPathMTUDiscovery:         c.Bool(flags.QuicDisablePathMTUDiscovery),
 		QUICConnectionLevelFlowControlLimit: c.Uint64(flags.QuicConnLevelFlowControlLimit),
 		QUICStreamLevelFlowControlLimit:     c.Uint64(flags.QuicStreamLevelFlowControlLimit),
+		OriginDNSService:                    dnsService,
+		OriginDialerService:                 originDialerService,
 	}
 	icmpRouter, err := newICMPRouter(c, log)
 	if err != nil {
@@ -254,10 +278,10 @@ func prepareTunnelConfig(
 		tunnelConfig.ICMPRouterServer = icmpRouter
 	}
 	orchestratorConfig := &orchestration.Config{
-		Ingress:            &ingressRules,
-		WarpRouting:        ingress.NewWarpRoutingConfig(&cfg.WarpRouting),
-		ConfigurationFlags: parseConfigFlags(c),
-		WriteTimeout:       tunnelConfig.WriteStreamTimeout,
+		Ingress:             &ingressRules,
+		WarpRouting:         warpRoutingConfig,
+		OriginDialerService: originDialerService,
+		ConfigurationFlags:  parseConfigFlags(c),
 	}
 	return tunnelConfig, orchestratorConfig, nil
 }
@@ -494,3 +518,19 @@ func findLocalAddr(dst net.IP, port int) (netip.Addr, error) {
 	localAddr := localAddrPort.Addr()
 	return localAddr, nil
 }
+
+func parseResolverAddrPorts(input []string) ([]netip.AddrPort, error) {
+	// We don't allow more than 10 resolvers to be provided statically for the resolver service.
+	if len(input) > 10 {
+		return nil, errors.New("too many addresses provided, max: 10")
+	}
+	addrs := make([]netip.AddrPort, 0, len(input))
+	for _, val := range input {
+		addr, err := netip.ParseAddrPort(val)
+		if err != nil {
+			return nil, err
+		}
+		addrs = append(addrs, addr)
+	}
+	return addrs, nil
+}

cmd/cloudflared/tunnel/login.go

@@ -12,6 +12,7 @@ import (
 	"github.com/urfave/cli/v2"
 
 	"github.com/cloudflare/cloudflared/cmd/cloudflared/cliutil"
+	cfdflags "github.com/cloudflare/cloudflared/cmd/cloudflared/flags"
 	"github.com/cloudflare/cloudflared/config"
 	"github.com/cloudflare/cloudflared/credentials"
 	"github.com/cloudflare/cloudflared/logger"
@@ -19,11 +20,10 @@ import (
 )
 
 const (
-	baseLoginURL = "https://dash.cloudflare.com/argotunnel"
-	callbackURL  = "https://login.cloudflareaccess.org/"
-	// For now these are the same but will change in the future once we know which URLs to use (TUN-8872)
-	fedBaseLoginURL      = "https://dash.cloudflare.com/argotunnel"
-	fedCallbackStoreURL  = "https://login.cloudflareaccess.org/"
+	baseLoginURL         = "https://dash.cloudflare.com/argotunnel"
+	callbackURL          = "https://login.cloudflareaccess.org/"
+	fedBaseLoginURL      = "https://dash.fed.cloudflare.com/argotunnel"
+	fedCallbackStoreURL  = "https://login.fed.cloudflareaccess.org/"
 	fedRAMPParamName     = "fedramp"
 	loginURLParamName    = "loginURL"
 	callbackURLParamName = "callbackURL"
@@ -97,6 +97,8 @@ func login(c *cli.Context) error {
 		callbackStoreURL,
 		false,
 		false,
+		c.Bool(cfdflags.AutoCloseInterstitial),
+		isFEDRamp,
 		log,
 	)
 	if err != nil {

cmd/cloudflared/tunnel/subcommand_context.go

@@ -155,10 +155,12 @@ func (sc *subcommandContext) create(name string, credentialsFilePath string, sec
 	if err != nil {
 		return nil, err
 	}
+
 	tunnelCredentials := connection.Credentials{
 		AccountTag:   credential.AccountID(),
 		TunnelSecret: tunnelSecret,
 		TunnelID:     tunnel.ID,
+		Endpoint:     credential.Endpoint(),
 	}
 	usedCertPath := false
 	if credentialsFilePath == "" {

cmd/cloudflared/tunnel/subcommands.go

@@ -241,6 +241,11 @@ var (
 		Usage:   "Overrides the remote configuration for max active private network flows (TCP/UDP) that this cloudflared instance supports",
 		EnvVars: []string{"TUNNEL_MAX_ACTIVE_FLOWS"},
 	}
+	dnsResolverAddrsFlag = &cli.StringSliceFlag{
+		Name:    flags.VirtualDNSServiceResolverAddresses,
+		Usage:   "Overrides the dynamic DNS resolver resolution to use these address:port's instead.",
+		EnvVars: []string{"TUNNEL_DNS_RESOLVER_ADDRS"},
+	}
 )
 
 func buildCreateCommand() *cli.Command {
@@ -718,6 +723,7 @@ func buildRunCommand() *cli.Command {
 		icmpv4SrcFlag,
 		icmpv6SrcFlag,
 		maxActiveFlowsFlag,
+		dnsResolverAddrsFlag,
 	}
 	flags = append(flags, configureProxyFlags(false)...)
 	return &cli.Command{

component-tests/test_management.py

@@ -107,7 +107,13 @@ class TestManagement:
             assert resp.status_code == 404, "Expected cloudflared to return 404 for /metrics"
 
 
+
+
 @retry(stop_max_attempt_number=MAX_RETRIES, wait_fixed=BACKOFF_SECS * 1000)
 def send_request(url, headers={}):
     with requests.Session() as s:
-        return s.get(url, timeout=BACKOFF_SECS, headers=headers)
+        resp = s.get(url, timeout=BACKOFF_SECS, headers=headers)
+        if resp.status_code == 530:
+            LOGGER.debug(f"Received 530 status, retrying request to {url}")
+            raise Exception(f"Received 530 status code from {url}")
+        return resp

component-tests/test_service.py

@@ -9,7 +9,7 @@ import pytest
 
 import test_logging
 from conftest import CfdModes
-from util import select_platform, start_cloudflared, wait_tunnel_ready, write_config
+from util import select_platform, skip_on_ci, start_cloudflared, wait_tunnel_ready, write_config
 
 
 def default_config_dir():
@@ -82,6 +82,7 @@ class TestServiceMode:
         os.remove(default_config_file())
         self.launchctl_cmd("list", success=False)
 
+    @skip_on_ci("we can't run sudo command on CI")
     @select_platform("Linux")
     @pytest.mark.skipif(os.path.exists("/etc/cloudflared/config.yml"),
                         reason=f"There is already a config file in default path")
@@ -98,6 +99,7 @@ class TestServiceMode:
 
         self.sysv_service_scenario(config, tmp_path, assert_log_file)
 
+    @skip_on_ci("we can't run sudo command on CI")
     @select_platform("Linux")
     @pytest.mark.skipif(os.path.exists("/etc/cloudflared/config.yml"),
                         reason=f"There is already a config file in default path")
@@ -116,6 +118,7 @@ class TestServiceMode:
 
         self.sysv_service_scenario(config, tmp_path, assert_rotating_log)
 
+    @skip_on_ci("we can't run sudo command on CI")
     @select_platform("Linux")
     @pytest.mark.skipif(os.path.exists("/etc/cloudflared/config.yml"),
                         reason=f"There is already a config file in default path")

component-tests/test_token.py

@@ -1,7 +1,7 @@
 import base64
 import json
 
-from setup import get_config_from_file, persist_origin_cert
+from setup import get_config_from_file
 from util import start_cloudflared
 
 

component-tests/test_tunnel.py

@@ -33,13 +33,20 @@ class TestTunnel:
         LOGGER.debug(config)
         with start_cloudflared(tmp_path, config, cfd_pre_args=["tunnel", "--ha-connections", "1"],  cfd_args=["run"], new_process=True):
             wait_tunnel_ready(require_min_connections=1)
-            resp = send_request(config.get_url()+"/")
-            assert resp.status_code == 503, "Expected cloudflared to return 503 for all requests with no ingress defined"
-            resp = send_request(config.get_url()+"/test")
-            assert resp.status_code == 503, "Expected cloudflared to return 503 for all requests with no ingress defined"
+            expected_status_code = 503
+            resp = send_request(config.get_url()+"/", expected_status_code)
+            assert resp.status_code == expected_status_code, "Expected cloudflared to return 503 for all requests with no ingress defined"
+            resp = send_request(config.get_url()+"/test", expected_status_code)
+            assert resp.status_code == expected_status_code, "Expected cloudflared to return 503 for all requests with no ingress defined"
 
+def retry_if_result_none(result):
+    '''
+    Returns True if the result is None, indicating that the function should be retried.
+    '''
+    return result is None
 
-@retry(stop_max_attempt_number=MAX_RETRIES, wait_fixed=BACKOFF_SECS * 1000)
-def send_request(url, headers={}):
+@retry(retry_on_result=retry_if_result_none, stop_max_attempt_number=MAX_RETRIES, wait_fixed=BACKOFF_SECS * 1000)
+def send_request(url, expected_status_code=200):
     with requests.Session() as s:
-        return s.get(url, timeout=BACKOFF_SECS, headers=headers)
+        resp = s.get(url, timeout=BACKOFF_SECS)
+        return resp if resp.status_code == expected_status_code else None

component-tests/util.py

@@ -10,7 +10,6 @@ import pytest
 
 import requests
 import yaml
-import json
 from retrying import retry
 
 from constants import METRICS_PORT, MAX_RETRIES, BACKOFF_SECS
@@ -35,6 +34,12 @@ def fips_enabled():
 nofips = pytest.mark.skipif(
         fips_enabled(), reason=f"Only runs without FIPS (COMPONENT_TESTS_FIPS=0)")
 
+def skip_on_ci(reason):
+    env_ci = os.getenv("CI")
+    running_in_ci = env_ci is not None and env_ci != "0"
+    return pytest.mark.skipif(
+        running_in_ci, reason=f"This test can't run on CI due to: {reason}")
+
 def write_config(directory, config):
     config_path = directory / "config.yml"
     with open(config_path, 'w') as outfile:
@@ -111,6 +116,7 @@ def inner_wait_tunnel_ready(tunnel_url=None, require_min_connections=1):
     metrics_url = f'http://localhost:{METRICS_PORT}/ready'
 
     with requests.Session() as s:
+        LOGGER.debug("Waiting for tunnel to be ready...")
         resp = send_request(s, metrics_url, True)
 
         ready_connections = resp.json()["readyConnections"]

config/configuration.go

@@ -242,6 +242,8 @@ type AccessConfig struct {
 
 	// AudTag is the AudTag to verify access JWT against.
 	AudTag []string `yaml:"audTag" json:"audTag"`
+
+	Environment string `yaml:"environment" json:"environment,omitempty"`
 }
 
 type IngressIPRule struct {

config/model.go

@@ -1,7 +1,7 @@
 package config
 
 import (
-	"crypto/md5"
+	"crypto/sha256"
 	"fmt"
 	"io"
 	"strings"
@@ -16,6 +16,7 @@ type Forwarder struct {
 	TokenClientID string `json:"service_token_id" yaml:"serviceTokenID"`
 	TokenSecret   string `json:"secret_token_id" yaml:"serviceTokenSecret"`
 	Destination   string `json:"destination"`
+	IsFedramp     bool   `json:"is_fedramp" yaml:"isFedramp"`
 }
 
 // Tunnel represents a tunnel that should be started
@@ -46,24 +47,24 @@ type Root struct {
 
 // Hash returns the computed values to see if the forwarder values change
 func (f *Forwarder) Hash() string {
-	h := md5.New()
-	io.WriteString(h, f.URL)
-	io.WriteString(h, f.Listener)
-	io.WriteString(h, f.TokenClientID)
-	io.WriteString(h, f.TokenSecret)
-	io.WriteString(h, f.Destination)
+	h := sha256.New()
+	_, _ = io.WriteString(h, f.URL)
+	_, _ = io.WriteString(h, f.Listener)
+	_, _ = io.WriteString(h, f.TokenClientID)
+	_, _ = io.WriteString(h, f.TokenSecret)
+	_, _ = io.WriteString(h, f.Destination)
 	return fmt.Sprintf("%x", h.Sum(nil))
 }
 
 // Hash returns the computed values to see if the forwarder values change
 func (r *DNSResolver) Hash() string {
-	h := md5.New()
-	io.WriteString(h, r.Address)
-	io.WriteString(h, strings.Join(r.Bootstraps, ","))
-	io.WriteString(h, strings.Join(r.Upstreams, ","))
-	io.WriteString(h, fmt.Sprintf("%d", r.Port))
-	io.WriteString(h, fmt.Sprintf("%d", r.MaxUpstreamConnections))
-	io.WriteString(h, fmt.Sprintf("%v", r.Enabled))
+	h := sha256.New()
+	_, _ = io.WriteString(h, r.Address)
+	_, _ = io.WriteString(h, strings.Join(r.Bootstraps, ","))
+	_, _ = io.WriteString(h, strings.Join(r.Upstreams, ","))
+	_, _ = io.WriteString(h, fmt.Sprintf("%d", r.Port))
+	_, _ = io.WriteString(h, fmt.Sprintf("%d", r.MaxUpstreamConnections))
+	_, _ = io.WriteString(h, fmt.Sprintf("%v", r.Enabled))
 	return fmt.Sprintf("%x", h.Sum(nil))
 }
 

connection/control.go

@@ -82,7 +82,7 @@ func (c *controlStream) ServeControlStream(
 	tunnelConfigGetter TunnelConfigJSONGetter,
 ) error {
 	registrationClient := c.registerClientFunc(ctx, rw, c.registerTimeout)
-
+	c.observer.logConnecting(c.connIndex, c.edgeAddress, c.protocol)
 	registrationDetails, err := registrationClient.RegisterConnection(
 		ctx,
 		c.tunnelProperties.Credentials.Auth(),

connection/errors.go

@@ -1,7 +1,6 @@
 package connection
 
 import (
-	"github.com/cloudflare/cloudflared/edgediscovery"
 	tunnelpogs "github.com/cloudflare/cloudflared/tunnelrpc/pogs"
 )
 
@@ -53,26 +52,26 @@ func serverRegistrationErrorFromRPC(err error) ServerRegisterTunnelError {
 	}
 }
 
-type muxerShutdownError struct{}
+type ControlStreamError struct{}
 
-func (e muxerShutdownError) Error() string {
-	return "muxer shutdown"
+var _ error = &ControlStreamError{}
+
+func (e *ControlStreamError) Error() string {
+	return "control stream encountered a failure while serving"
 }
 
-var errMuxerStopped = muxerShutdownError{}
+type StreamListenerError struct{}
 
-func isHandshakeErrRecoverable(err error, connIndex uint8, observer *Observer) bool {
-	log := observer.log.With().
-		Uint8(LogFieldConnIndex, connIndex).
-		Err(err).
-		Logger()
+var _ error = &StreamListenerError{}
 
-	switch err.(type) {
-	case edgediscovery.DialError:
-		log.Error().Msg("Connection unable to dial edge")
-	default:
-		log.Error().Msg("Connection failed")
-		return false
-	}
-	return true
+func (e *StreamListenerError) Error() string {
+	return "accept stream listener encountered a failure while serving"
+}
+
+type DatagramManagerError struct{}
+
+var _ error = &DatagramManagerError{}
+
+func (e *DatagramManagerError) Error() string {
+	return "datagram manager encountered a failure while serving"
 }

connection/header.go

@@ -53,7 +53,8 @@ var headerEncoding = base64.RawStdEncoding
 func IsControlResponseHeader(headerName string) bool {
 	return strings.HasPrefix(headerName, ":") ||
 		strings.HasPrefix(headerName, "cf-int-") ||
-		strings.HasPrefix(headerName, "cf-cloudflared-")
+		strings.HasPrefix(headerName, "cf-cloudflared-") ||
+		strings.HasPrefix(headerName, "cf-proxy-")
 }
 
 // isWebsocketClientHeader returns true if the header name is required by the client to upgrade properly

connection/header_test.go

@@ -1,18 +1,17 @@
 package connection
 
 import (
-	"fmt"
 	"net/http"
 	"reflect"
 	"sort"
 	"testing"
 
-	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
 )
 
 func TestSerializeHeaders(t *testing.T) {
 	request, err := http.NewRequest(http.MethodGet, "http://example.com", nil)
-	assert.NoError(t, err)
+	require.NoError(t, err)
 
 	mockHeaders := http.Header{
 		"Mock-Header-One":        {"Mock header one value", "three"},
@@ -39,22 +38,22 @@ func TestSerializeHeaders(t *testing.T) {
 	serializedHeaders := SerializeHeaders(request.Header)
 
 	// Sanity check: the headers serialized to something that's not an empty string
-	assert.NotEqual(t, "", serializedHeaders)
+	require.NotEqual(t, "", serializedHeaders)
 
 	// Deserialize back, and ensure we get the same set of headers
 	deserializedHeaders, err := DeserializeHeaders(serializedHeaders)
-	assert.NoError(t, err)
+	require.NoError(t, err)
 
-	assert.Equal(t, 13, len(deserializedHeaders))
+	require.Len(t, deserializedHeaders, 13)
 	expectedHeaders := headerToReqHeader(mockHeaders)
 
 	sort.Sort(ByName(deserializedHeaders))
 	sort.Sort(ByName(expectedHeaders))
 
-	assert.True(
+	require.True(
 		t,
 		reflect.DeepEqual(expectedHeaders, deserializedHeaders),
-		fmt.Sprintf("got = %#v, want = %#v\n", deserializedHeaders, expectedHeaders),
+		"got = %#v, want = %#v\n", deserializedHeaders, expectedHeaders,
 	)
 }
 
@@ -82,12 +81,12 @@ func headerToReqHeader(headers http.Header) (reqHeaders []HTTPHeader) {
 
 func TestSerializeNoHeaders(t *testing.T) {
 	request, err := http.NewRequest(http.MethodGet, "http://example.com", nil)
-	assert.NoError(t, err)
+	require.NoError(t, err)
 
 	serializedHeaders := SerializeHeaders(request.Header)
 	deserializedHeaders, err := DeserializeHeaders(serializedHeaders)
-	assert.NoError(t, err)
-	assert.Equal(t, 0, len(deserializedHeaders))
+	require.NoError(t, err)
+	require.Empty(t, deserializedHeaders)
 }
 
 func TestDeserializeMalformed(t *testing.T) {
@@ -102,21 +101,22 @@ func TestDeserializeMalformed(t *testing.T) {
 
 	for _, malformedValue := range malformedData {
 		_, err = DeserializeHeaders(malformedValue)
-		assert.Error(t, err)
+		require.Error(t, err)
 	}
 }
 
 func TestIsControlResponseHeader(t *testing.T) {
 	controlResponseHeaders := []string{
-		// Anything that begins with cf-int- or cf-cloudflared-
+		// Anything that begins with cf-int-, cf-cloudflared- or cf-proxy-
 		"cf-int-sample-header",
 		"cf-cloudflared-sample-header",
+		"cf-proxy-sample-header",
 		// Any http2 pseudoheader
 		":sample-pseudo-header",
 	}
 
 	for _, header := range controlResponseHeaders {
-		assert.True(t, IsControlResponseHeader(header))
+		require.True(t, IsControlResponseHeader(header))
 	}
 }
 
@@ -130,6 +130,6 @@ func TestIsNotControlResponseHeader(t *testing.T) {
 	}
 
 	for _, header := range notControlResponseHeaders {
-		assert.False(t, IsControlResponseHeader(header))
+		require.False(t, IsControlResponseHeader(header))
 	}
 }

connection/observer.go

@@ -46,6 +46,15 @@ func (o *Observer) RegisterSink(sink EventSink) {
 	o.addSinkChan <- sink
 }
 
+func (o *Observer) logConnecting(connIndex uint8, address net.IP, protocol Protocol) {
+	o.log.Debug().
+		Int(management.EventTypeKey, int(management.Cloudflared)).
+		Uint8(LogFieldConnIndex, connIndex).
+		IPAddr(LogFieldIPAddress, address).
+		Str(LogFieldProtocol, protocol.String()).
+		Msg("Registering tunnel connection")
+}
+
 func (o *Observer) logConnected(connectionID uuid.UUID, connIndex uint8, location string, address net.IP, protocol Protocol) {
 	o.log.Info().
 		Int(management.EventTypeKey, int(management.Cloudflared)).

connection/quic_connection.go

@@ -3,6 +3,7 @@ package connection
 import (
 	"bufio"
 	"context"
+	"errors"
 	"fmt"
 	"io"
 	"net"
@@ -12,7 +13,6 @@ import (
 	"sync/atomic"
 	"time"
 
-	"github.com/pkg/errors"
 	"github.com/quic-go/quic-go"
 	"github.com/rs/zerolog"
 	"golang.org/x/sync/errgroup"
@@ -65,7 +65,7 @@ func NewTunnelConnection(
 	streamWriteTimeout time.Duration,
 	gracePeriod time.Duration,
 	logger *zerolog.Logger,
-) (TunnelConnection, error) {
+) TunnelConnection {
 	return &quicConnection{
 		conn:                 conn,
 		logger:               logger,
@@ -77,10 +77,11 @@ func NewTunnelConnection(
 		rpcTimeout:           rpcTimeout,
 		streamWriteTimeout:   streamWriteTimeout,
 		gracePeriod:          gracePeriod,
-	}, nil
+	}
 }
 
 // Serve starts a QUIC connection that begins accepting streams.
+// Returning a nil error means cloudflared will exit for good and will not attempt to reconnect.
 func (q *quicConnection) Serve(ctx context.Context) error {
 	// The edge assumes the first stream is used for the control plane
 	controlStream, err := q.conn.OpenStream()
@@ -88,16 +89,16 @@ func (q *quicConnection) Serve(ctx context.Context) error {
 		return fmt.Errorf("failed to open a registration control stream: %w", err)
 	}
 
-	// If either goroutine returns nil error, we rely on this cancellation to make sure the other goroutine exits
-	// as fast as possible as well. Nil error means we want to exit for good (caller code won't retry serving this
-	// connection).
 	// If either goroutine returns a non nil error, then the error group cancels the context, thus also canceling the
-	// other goroutine as fast as possible.
-	ctx, cancel := context.WithCancel(ctx)
+	// other goroutines. We enforce returning a not-nil error for each function started in the errgroup by logging
+	// the error returned and returning a custom error type instead.
 	errGroup, ctx := errgroup.WithContext(ctx)
 
-	// In the future, if cloudflared can autonomously push traffic to the edge, we have to make sure the control
-	// stream is already fully registered before the other goroutines can proceed.
+	// Close the quic connection if any of the following routines return from the errgroup (regardless of their error)
+	// because they are no longer processing requests for the connection.
+	defer q.Close()
+
+	// Start the control stream routine
 	errGroup.Go(func() error {
 		// err is equal to nil if we exit due to unregistration. If that happens we want to wait the full
 		// amount of the grace period, allowing requests to finish before we cancel the context, which will
@@ -114,16 +115,26 @@ func (q *quicConnection) Serve(ctx context.Context) error {
 				}
 			}
 		}
-		cancel()
-		return err
+		if err != nil {
+			q.logger.Error().Err(err).Msg("failed to serve the control stream")
+		}
+		return &ControlStreamError{}
 	})
+	// Start the accept stream loop routine
 	errGroup.Go(func() error {
-		defer cancel()
-		return q.acceptStream(ctx)
+		err := q.acceptStream(ctx)
+		if err != nil {
+			q.logger.Error().Err(err).Msg("failed to accept incoming stream requests")
+		}
+		return &StreamListenerError{}
 	})
+	// Start the datagram handler routine
 	errGroup.Go(func() error {
-		defer cancel()
-		return q.datagramHandler.Serve(ctx)
+		err := q.datagramHandler.Serve(ctx)
+		if err != nil {
+			q.logger.Error().Err(err).Msg("failed to run the datagram handler")
+		}
+		return &DatagramManagerError{}
 	})
 
 	return errGroup.Wait()
@@ -140,7 +151,6 @@ func (q *quicConnection) Close() {
 }
 
 func (q *quicConnection) acceptStream(ctx context.Context) error {
-	defer q.Close()
 	for {
 		quicStream, err := q.conn.AcceptStream(ctx)
 		if err != nil {
@@ -230,7 +240,7 @@ func (q *quicConnection) dispatchRequest(ctx context.Context, stream *rpcquic.Re
 			ConnIndex: q.connIndex,
 		}), rwa.connectResponseSent
 	default:
-		return errors.Errorf("unsupported error type: %s", request.Type), false
+		return fmt.Errorf("unsupported error type: %s", request.Type), false
 	}
 }
 

connection/quic_connection_test.go

@@ -30,6 +30,7 @@ import (
 	"golang.org/x/net/nettest"
 
 	"github.com/cloudflare/cloudflared/client"
+	"github.com/cloudflare/cloudflared/config"
 	cfdflow "github.com/cloudflare/cloudflared/flow"
 
 	"github.com/cloudflare/cloudflared/datagramsession"
@@ -823,6 +824,15 @@ func testTunnelConnection(t *testing.T, serverAddr netip.AddrPort, index uint8)
 	sessionManager := datagramsession.NewManager(&log, datagramMuxer.SendToSession, sessionDemuxChan)
 	var connIndex uint8 = 0
 	packetRouter := ingress.NewPacketRouter(nil, datagramMuxer, connIndex, &log)
+	testDefaultDialer := ingress.NewDialer(ingress.WarpRoutingConfig{
+		ConnectTimeout: config.CustomDuration{Duration: 1 * time.Second},
+		TCPKeepAlive:   config.CustomDuration{Duration: 15 * time.Second},
+		MaxActiveFlows: 0,
+	})
+	originDialer := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   testDefaultDialer,
+		TCPWriteTimeout: 1 * time.Second,
+	}, &log)
 
 	datagramConn := &datagramV2Connection{
 		conn,
@@ -830,13 +840,14 @@ func testTunnelConnection(t *testing.T, serverAddr netip.AddrPort, index uint8)
 		sessionManager,
 		cfdflow.NewLimiter(0),
 		datagramMuxer,
+		originDialer,
 		packetRouter,
 		15 * time.Second,
 		0 * time.Second,
 		&log,
 	}
 
-	tunnelConn, err := NewTunnelConnection(
+	tunnelConn := NewTunnelConnection(
 		ctx,
 		conn,
 		index,
@@ -849,7 +860,6 @@ func testTunnelConnection(t *testing.T, serverAddr netip.AddrPort, index uint8)
 		0*time.Second,
 		&log,
 	)
-	require.NoError(t, err)
 	return tunnelConn, datagramConn
 }
 

connection/quic_datagram_v2.go

@@ -4,9 +4,11 @@ import (
 	"context"
 	"fmt"
 	"net"
+	"net/netip"
 	"time"
 
 	"github.com/google/uuid"
+	"github.com/pkg/errors"
 	pkgerrors "github.com/pkg/errors"
 	"github.com/quic-go/quic-go"
 	"github.com/rs/zerolog"
@@ -32,6 +34,10 @@ const (
 	demuxChanCapacity = 16
 )
 
+var (
+	errInvalidDestinationIP = errors.New("unable to parse destination IP")
+)
+
 // DatagramSessionHandler is a service that can serve datagrams for a connection and handle sessions from incoming
 // connection streams.
 type DatagramSessionHandler interface {
@@ -51,7 +57,10 @@ type datagramV2Connection struct {
 
 	// datagramMuxer mux/demux datagrams from quic connection
 	datagramMuxer *cfdquic.DatagramMuxerV2
-	packetRouter  *ingress.PacketRouter
+	// originDialer is the origin dialer for UDP requests
+	originDialer ingress.OriginUDPDialer
+	// packetRouter acts as the origin router for ICMP requests
+	packetRouter *ingress.PacketRouter
 
 	rpcTimeout         time.Duration
 	streamWriteTimeout time.Duration
@@ -61,6 +70,7 @@ type datagramV2Connection struct {
 
 func NewDatagramV2Connection(ctx context.Context,
 	conn quic.Connection,
+	originDialer ingress.OriginUDPDialer,
 	icmpRouter ingress.ICMPRouter,
 	index uint8,
 	rpcTimeout time.Duration,
@@ -79,6 +89,7 @@ func NewDatagramV2Connection(ctx context.Context,
 		sessionManager:     sessionManager,
 		flowLimiter:        flowLimiter,
 		datagramMuxer:      datagramMuxer,
+		originDialer:       originDialer,
 		packetRouter:       packetRouter,
 		rpcTimeout:         rpcTimeout,
 		streamWriteTimeout: streamWriteTimeout,
@@ -87,24 +98,17 @@ func NewDatagramV2Connection(ctx context.Context,
 }
 
 func (d *datagramV2Connection) Serve(ctx context.Context) error {
-	// If either goroutine returns nil error, we rely on this cancellation to make sure the other goroutine exits
-	// as fast as possible as well. Nil error means we want to exit for good (caller code won't retry serving this
-	// connection).
-	// If either goroutine returns a non nil error, then the error group cancels the context, thus also canceling the
-	// other goroutine as fast as possible.
-	ctx, cancel := context.WithCancel(ctx)
+	// If either goroutine from the errgroup returns at all (error or nil), we rely on its cancellation to make sure
+	// the other goroutines as well.
 	errGroup, ctx := errgroup.WithContext(ctx)
 
 	errGroup.Go(func() error {
-		defer cancel()
 		return d.sessionManager.Serve(ctx)
 	})
 	errGroup.Go(func() error {
-		defer cancel()
 		return d.datagramMuxer.ServeReceive(ctx)
 	})
 	errGroup.Go(func() error {
-		defer cancel()
 		return d.packetRouter.Serve(ctx)
 	})
 
@@ -128,12 +132,29 @@ func (q *datagramV2Connection) RegisterUdpSession(ctx context.Context, sessionID
 		tracing.EndWithErrorStatus(registerSpan, err)
 		return nil, err
 	}
+	// We need to force the net.IP to IPv4 (if it's an IPv4 address) otherwise the net.IP conversion from capnp
+	// will be a IPv4-mapped-IPv6 address.
+	// In the case that the address is IPv6 we leave it untouched and parse it as normal.
+	ip := dstIP.To4()
+	if ip == nil {
+		ip = dstIP
+	}
+	// Parse the dstIP and dstPort into a netip.AddrPort
+	// This should never fail because the IP was already parsed as a valid net.IP
+	destAddr, ok := netip.AddrFromSlice(ip)
+	if !ok {
+		log.Err(errInvalidDestinationIP).Msgf("Failed to parse destination proxy IP: %s", ip)
+		tracing.EndWithErrorStatus(registerSpan, errInvalidDestinationIP)
+		q.flowLimiter.Release()
+		return nil, errInvalidDestinationIP
+	}
+	dstAddrPort := netip.AddrPortFrom(destAddr, dstPort)
 
 	// Each session is a series of datagram from an eyeball to a dstIP:dstPort.
 	// (src port, dst IP, dst port) uniquely identifies a session, so it needs a dedicated connected socket.
-	originProxy, err := ingress.DialUDP(dstIP, dstPort)
+	originProxy, err := q.originDialer.DialUDP(dstAddrPort)
 	if err != nil {
-		log.Err(err).Msgf("Failed to create udp proxy to %s:%d", dstIP, dstPort)
+		log.Err(err).Msgf("Failed to create udp proxy to %s", dstAddrPort)
 		tracing.EndWithErrorStatus(registerSpan, err)
 		q.flowLimiter.Release()
 		return nil, err

connection/quic_datagram_v2_test.go

@@ -84,6 +84,7 @@ func TestRateLimitOnNewDatagramV2UDPSession(t *testing.T) {
 		t.Context(),
 		conn,
 		nil,
+		nil,
 		0,
 		0*time.Second,
 		0*time.Second,

credentials/credentials.go

@@ -10,6 +10,8 @@ import (
 const (
 	logFieldOriginCertPath = "originCertPath"
 	FedEndpoint            = "fed"
+	FedRampBaseApiURL      = "https://api.fed.cloudflare.com/client/v4"
+	FedRampHostname        = "management.fed.argotunnel.com"
 )
 
 type User struct {
@@ -21,6 +23,10 @@ func (c User) AccountID() string {
 	return c.cert.AccountID
 }
 
+func (c User) Endpoint() string {
+	return c.cert.Endpoint
+}
+
 func (c User) ZoneID() string {
 	return c.cert.ZoneID
 }

features/features.go

@@ -10,9 +10,10 @@ const (
 	FeaturePostQuantum       = "postquantum"
 	FeatureQUICSupportEOF    = "support_quic_eof"
 	FeatureManagementLogs    = "management_logs"
-	FeatureDatagramV3_1      = "support_datagram_v3_1"
+	FeatureDatagramV3_2      = "support_datagram_v3_2"
 
-	DeprecatedFeatureDatagramV3 = "support_datagram_v3" // Deprecated: TUN-9291
+	DeprecatedFeatureDatagramV3   = "support_datagram_v3"   // Deprecated: TUN-9291
+	DeprecatedFeatureDatagramV3_1 = "support_datagram_v3_1" // Deprecated: TUN-9883
 )
 
 var defaultFeatures = []string{
@@ -26,6 +27,7 @@ var defaultFeatures = []string{
 // List of features that are no longer in-use.
 var deprecatedFeatures = []string{
 	DeprecatedFeatureDatagramV3,
+	DeprecatedFeatureDatagramV3_1,
 }
 
 // Features set by user provided flags
@@ -58,7 +60,7 @@ const (
 	// DatagramV2 is the currently supported datagram protocol for UDP and ICMP packets
 	DatagramV2 DatagramVersion = FeatureDatagramV2
 	// DatagramV3 is a new datagram protocol for UDP and ICMP packets. It is not backwards compatible with datagram v2.
-	DatagramV3 DatagramVersion = FeatureDatagramV3_1
+	DatagramV3 DatagramVersion = FeatureDatagramV3_2
 )
 
 // Remove any duplicate features from the list and remove deprecated features

features/selector.go

@@ -23,9 +23,10 @@ const (
 // If the TXT record is missing a key, the field will unmarshal to the default Go value
 
 type featuresRecord struct {
-	DatagramV3Percentage uint32 `json:"dv3_1"`
+	DatagramV3Percentage uint32 `json:"dv3_2"`
 
 	// DatagramV3Percentage int32 `json:"dv3"` // Removed in TUN-9291
+	// DatagramV3Percentage uint32 `json:"dv3_1"` // Removed in TUN-9883
 	// PostQuantumPercentage int32 `json:"pq"` // Removed in TUN-7970
 }
 
@@ -105,7 +106,7 @@ func (fs *featureSelector) postQuantumMode() PostQuantumMode {
 
 func (fs *featureSelector) datagramVersion() DatagramVersion {
 	// If user provides the feature via the cli, we take it as priority over remote feature evaluation
-	if slices.Contains(fs.cliFeatures, FeatureDatagramV3_1) {
+	if slices.Contains(fs.cliFeatures, FeatureDatagramV3_2) {
 		return DatagramV3
 	}
 	// If the user specifies DatagramV2, we also take that over remote

features/selector_test.go

@@ -22,15 +22,15 @@ func TestUnmarshalFeaturesRecord(t *testing.T) {
 		expectedPercentage uint32
 	}{
 		{
-			record:             []byte(`{"dv3_1":0}`),
+			record:             []byte(`{"dv3_2":0}`),
 			expectedPercentage: 0,
 		},
 		{
-			record:             []byte(`{"dv3_1":39}`),
+			record:             []byte(`{"dv3_2":39}`),
 			expectedPercentage: 39,
 		},
 		{
-			record:             []byte(`{"dv3_1":100}`),
+			record:             []byte(`{"dv3_2":100}`),
 			expectedPercentage: 100,
 		},
 		{
@@ -40,7 +40,7 @@ func TestUnmarshalFeaturesRecord(t *testing.T) {
 			record: []byte(`{"kyber":768}`), // Unmarshal to default struct if key is not present
 		},
 		{
-			record: []byte(`{"pq": 101,"dv3":100}`), // Expired keys don't unmarshal to anything
+			record: []byte(`{"pq": 101,"dv3":100,"dv3_1":100}`), // Expired keys don't unmarshal to anything
 		},
 	}
 
@@ -111,10 +111,10 @@ func TestFeaturePrecedenceEvaluationDatagramVersion(t *testing.T) {
 		},
 		{
 			name:             "user_specified_v3",
-			cli:              []string{FeatureDatagramV3_1},
+			cli:              []string{FeatureDatagramV3_2},
 			remote:           featuresRecord{},
-			expectedFeatures: dedupAndRemoveFeatures(append(defaultFeatures, FeatureDatagramV3_1)),
-			expectedVersion:  FeatureDatagramV3_1,
+			expectedFeatures: dedupAndRemoveFeatures(append(defaultFeatures, FeatureDatagramV3_2)),
+			expectedVersion:  FeatureDatagramV3_2,
 		},
 	}
 
@@ -150,6 +150,12 @@ func TestDeprecatedFeaturesRemoved(t *testing.T) {
 			remote:           featuresRecord{},
 			expectedFeatures: defaultFeatures,
 		},
+		{
+			name:             "support_datagram_v3_1",
+			cli:              []string{DeprecatedFeatureDatagramV3_1},
+			remote:           featuresRecord{},
+			expectedFeatures: defaultFeatures,
+		},
 	}
 
 	for _, test := range tests {

go.mod

@@ -10,7 +10,7 @@ require (
 	github.com/fortytw2/leaktest v1.3.0
 	github.com/fsnotify/fsnotify v1.4.9
 	github.com/getsentry/sentry-go v0.16.0
-	github.com/go-chi/chi/v5 v5.0.10
+	github.com/go-chi/chi/v5 v5.2.2
 	github.com/go-chi/cors v1.2.1
 	github.com/go-jose/go-jose/v4 v4.1.0
 	github.com/gobwas/ws v1.2.1

go.sum

@@ -58,8 +58,8 @@ github.com/gin-contrib/sse v0.1.0/go.mod h1:RHrZQHXnP2xjPF+u1gW/2HnVO7nvIa9PG3Gm
 github.com/gin-gonic/gin v1.6.3/go.mod h1:75u5sXoLsGZoRN5Sgbi1eraJ4GU3++wFwWzhwvtwp4M=
 github.com/gin-gonic/gin v1.9.1 h1:4idEAncQnU5cB7BeOkPtxjfCSye0AAm1R0RVIqJ+Jmg=
 github.com/gin-gonic/gin v1.9.1/go.mod h1:hPrL7YrpYKXt5YId3A/Tnip5kqbEAP+KLuI3SUcPTeU=
-github.com/go-chi/chi/v5 v5.0.10 h1:rLz5avzKpjqxrYwXNfmjkrYYXOyLJd37pz53UFHC6vk=
-github.com/go-chi/chi/v5 v5.0.10/go.mod h1:DslCQbL2OYiznFReuXYUmQ2hGd1aDpCnlMNITLSKoi8=
+github.com/go-chi/chi/v5 v5.2.2 h1:CMwsvRVTbXVytCk1Wd72Zy1LAsAh9GxMmSNWLHCG618=
+github.com/go-chi/chi/v5 v5.2.2/go.mod h1:L2yAIGWB3H+phAw1NxKwWM+7eUH/lU8pOMm5hHcoops=
 github.com/go-chi/cors v1.2.1 h1:xEC8UT3Rlp2QuWNEr4Fs/c2EAGVKBwy/1vHx3bppil4=
 github.com/go-chi/cors v1.2.1/go.mod h1:sSbTewc+6wYHBBCW7ytsFSn836hqM7JxpglAy2Vzc58=
 github.com/go-errors/errors v1.4.2 h1:J6MZopCL4uSllY1OfXM374weqZFFItUbrImctkmUxIA=

ingress/ingress.go

@@ -317,7 +317,7 @@ func validateIngress(ingress []config.UnvalidatedIngressRule, defaults OriginReq
 				return Ingress{}, err
 			}
 			if access.Required {
-				verifier := middleware.NewJWTValidator(access.TeamName, "", access.AudTag)
+				verifier := middleware.NewJWTValidator(access.TeamName, access.Environment, access.AudTag)
 				handlers = append(handlers, verifier)
 			}
 		}

ingress/middleware/jwtvalidator.go

@@ -6,6 +6,8 @@ import (
 	"net/http"
 
 	"github.com/coreos/go-oidc/v3/oidc"
+
+	"github.com/cloudflare/cloudflared/credentials"
 )
 
 const (
@@ -13,7 +15,8 @@ const (
 )
 
 var (
-	cloudflareAccessCertsURL = "https://%s.cloudflareaccess.com"
+	cloudflareAccessCertsURL    = "https://%s.cloudflareaccess.com"
+	cloudflareAccessFedCertsURL = "https://%s.fed.cloudflareaccess.com"
 )
 
 // JWTValidator is an implementation of Verifier that validates access based JWT tokens.
@@ -22,10 +25,14 @@ type JWTValidator struct {
 	audTags []string
 }
 
-func NewJWTValidator(teamName string, certsURL string, audTags []string) *JWTValidator {
-	if certsURL == "" {
+func NewJWTValidator(teamName string, environment string, audTags []string) *JWTValidator {
+	var certsURL string
+	if environment == credentials.FedEndpoint {
+		certsURL = fmt.Sprintf(cloudflareAccessFedCertsURL, teamName)
+	} else {
 		certsURL = fmt.Sprintf(cloudflareAccessCertsURL, teamName)
 	}
+
 	certsEndpoint := fmt.Sprintf("%s/cdn-cgi/access/certs", certsURL)
 
 	config := &oidc.Config{

ingress/origin_connection.go

@@ -19,7 +19,7 @@ import (
 type OriginConnection interface {
 	// Stream should generally be implemented as a bidirectional io.Copy.
 	Stream(ctx context.Context, tunnelConn io.ReadWriter, log *zerolog.Logger)
-	Close()
+	Close() error
 }
 
 type streamHandlerFunc func(originConn io.ReadWriter, remoteConn net.Conn, log *zerolog.Logger)
@@ -48,16 +48,7 @@ func (tc *tcpConnection) Write(b []byte) (int, error) {
 		}
 	}
 
-	nBytes, err := tc.Conn.Write(b)
-	if err != nil {
-		tc.logger.Err(err).Msg("Error writing to the TCP connection")
-	}
-
-	return nBytes, err
-}
-
-func (tc *tcpConnection) Close() {
-	tc.Conn.Close()
+	return tc.Conn.Write(b)
 }
 
 // tcpOverWSConnection is an OriginConnection that streams to TCP over WS.
@@ -75,8 +66,8 @@ func (wc *tcpOverWSConnection) Stream(ctx context.Context, tunnelConn io.ReadWri
 	wsConn.Close()
 }
 
-func (wc *tcpOverWSConnection) Close() {
-	wc.conn.Close()
+func (wc *tcpOverWSConnection) Close() error {
+	return wc.conn.Close()
 }
 
 // socksProxyOverWSConnection is an OriginConnection that streams SOCKS connections over WS.
@@ -95,5 +86,6 @@ func (sp *socksProxyOverWSConnection) Stream(ctx context.Context, tunnelConn io.
 	wsConn.Close()
 }
 
-func (sp *socksProxyOverWSConnection) Close() {
+func (sp *socksProxyOverWSConnection) Close() error {
+	return nil
 }

ingress/origin_dialer.go

@@ -0,0 +1,163 @@
+package ingress
+
+import (
+	"context"
+	"fmt"
+	"net"
+	"net/netip"
+	"sync"
+	"time"
+
+	"github.com/rs/zerolog"
+)
+
+const writeDeadlineUDP = 200 * time.Millisecond
+
+// OriginTCPDialer provides a TCP dial operation to a requested address.
+type OriginTCPDialer interface {
+	DialTCP(ctx context.Context, addr netip.AddrPort) (net.Conn, error)
+}
+
+// OriginUDPDialer provides a UDP dial operation to a requested address.
+type OriginUDPDialer interface {
+	DialUDP(addr netip.AddrPort) (net.Conn, error)
+}
+
+// OriginDialer provides both TCP and UDP dial operations to an address.
+type OriginDialer interface {
+	OriginTCPDialer
+	OriginUDPDialer
+}
+
+type OriginConfig struct {
+	// The default Dialer used if no reserved services are found for an origin request.
+	DefaultDialer OriginDialer
+	// Timeout on write operations for TCP connections to the origin.
+	TCPWriteTimeout time.Duration
+}
+
+// OriginDialerService provides a proxy TCP and UDP dialer to origin services while allowing reserved
+// services to be provided. These reserved services are assigned to specific [netip.AddrPort]s
+// and provide their own [OriginDialer]'s to handle origin dialing per protocol.
+type OriginDialerService struct {
+	// Reserved TCP services for reserved AddrPort values
+	reservedTCPServices map[netip.AddrPort]OriginTCPDialer
+	// Reserved UDP services for reserved AddrPort values
+	reservedUDPServices map[netip.AddrPort]OriginUDPDialer
+	// The default Dialer used if no reserved services are found for an origin request
+	defaultDialer  OriginDialer
+	defaultDialerM sync.RWMutex
+	// Write timeout for TCP connections
+	writeTimeout time.Duration
+
+	logger *zerolog.Logger
+}
+
+func NewOriginDialer(config OriginConfig, logger *zerolog.Logger) *OriginDialerService {
+	return &OriginDialerService{
+		reservedTCPServices: map[netip.AddrPort]OriginTCPDialer{},
+		reservedUDPServices: map[netip.AddrPort]OriginUDPDialer{},
+		defaultDialer:       config.DefaultDialer,
+		writeTimeout:        config.TCPWriteTimeout,
+		logger:              logger,
+	}
+}
+
+// AddReservedService adds a reserved virtual service to dial to.
+// Not locked and expected to be initialized before calling first dial and not afterwards.
+func (d *OriginDialerService) AddReservedService(service OriginDialer, addrs []netip.AddrPort) {
+	for _, addr := range addrs {
+		d.reservedTCPServices[addr] = service
+		d.reservedUDPServices[addr] = service
+	}
+}
+
+// UpdateDefaultDialer updates the default dialer.
+func (d *OriginDialerService) UpdateDefaultDialer(dialer *Dialer) {
+	d.defaultDialerM.Lock()
+	defer d.defaultDialerM.Unlock()
+	d.defaultDialer = dialer
+}
+
+// DialTCP will perform a dial TCP to the requested addr.
+func (d *OriginDialerService) DialTCP(ctx context.Context, addr netip.AddrPort) (net.Conn, error) {
+	conn, err := d.dialTCP(ctx, addr)
+	if err != nil {
+		return nil, err
+	}
+	// Assign the write timeout for the TCP operations
+	return &tcpConnection{
+		Conn:         conn,
+		writeTimeout: d.writeTimeout,
+		logger:       d.logger,
+	}, nil
+}
+
+func (d *OriginDialerService) dialTCP(ctx context.Context, addr netip.AddrPort) (net.Conn, error) {
+	// Check to see if any reserved services are available for this addr and call their dialer instead.
+	if dialer, ok := d.reservedTCPServices[addr]; ok {
+		return dialer.DialTCP(ctx, addr)
+	}
+	d.defaultDialerM.RLock()
+	dialer := d.defaultDialer
+	d.defaultDialerM.RUnlock()
+	return dialer.DialTCP(ctx, addr)
+}
+
+// DialUDP will perform a dial UDP to the requested addr.
+func (d *OriginDialerService) DialUDP(addr netip.AddrPort) (net.Conn, error) {
+	// Check to see if any reserved services are available for this addr and call their dialer instead.
+	if dialer, ok := d.reservedUDPServices[addr]; ok {
+		return dialer.DialUDP(addr)
+	}
+	d.defaultDialerM.RLock()
+	dialer := d.defaultDialer
+	d.defaultDialerM.RUnlock()
+	return dialer.DialUDP(addr)
+}
+
+type Dialer struct {
+	Dialer net.Dialer
+}
+
+func NewDialer(config WarpRoutingConfig) *Dialer {
+	return &Dialer{
+		Dialer: net.Dialer{
+			Timeout:   config.ConnectTimeout.Duration,
+			KeepAlive: config.TCPKeepAlive.Duration,
+		},
+	}
+}
+
+func (d *Dialer) DialTCP(ctx context.Context, dest netip.AddrPort) (net.Conn, error) {
+	conn, err := d.Dialer.DialContext(ctx, "tcp", dest.String())
+	if err != nil {
+		return nil, fmt.Errorf("unable to dial tcp to origin %s: %w", dest, err)
+	}
+
+	return conn, nil
+}
+
+func (d *Dialer) DialUDP(dest netip.AddrPort) (net.Conn, error) {
+	conn, err := d.Dialer.Dial("udp", dest.String())
+	if err != nil {
+		return nil, fmt.Errorf("unable to dial udp to origin %s: %w", dest, err)
+	}
+	return &writeDeadlineConn{
+		Conn: conn,
+	}, nil
+}
+
+// writeDeadlineConn is a wrapper around a net.Conn that sets a write deadline of 200ms.
+// This is to prevent the socket from blocking on the write operation if it were to occur. However,
+// we typically never expect this to occur except under high load or kernel issues.
+type writeDeadlineConn struct {
+	net.Conn
+}
+
+func (w *writeDeadlineConn) Write(b []byte) (int, error) {
+	if err := w.SetWriteDeadline(time.Now().Add(writeDeadlineUDP)); err != nil {
+		return 0, err
+	}
+	return w.Conn.Write(b)
+}

ingress/origin_udp_proxy.go

@@ -1,46 +0,0 @@
-package ingress
-
-import (
-	"fmt"
-	"io"
-	"net"
-	"net/netip"
-)
-
-type UDPProxy interface {
-	io.ReadWriteCloser
-	LocalAddr() net.Addr
-}
-
-type udpProxy struct {
-	*net.UDPConn
-}
-
-func DialUDP(dstIP net.IP, dstPort uint16) (UDPProxy, error) {
-	dstAddr := &net.UDPAddr{
-		IP:   dstIP,
-		Port: int(dstPort),
-	}
-
-	// We use nil as local addr to force runtime to find the best suitable local address IP given the destination
-	// address as context.
-	udpConn, err := net.DialUDP("udp", nil, dstAddr)
-	if err != nil {
-		return nil, fmt.Errorf("unable to create UDP proxy to origin (%v:%v): %w", dstIP, dstPort, err)
-	}
-
-	return &udpProxy{udpConn}, nil
-}
-
-func DialUDPAddrPort(dest netip.AddrPort) (*net.UDPConn, error) {
-	addr := net.UDPAddrFromAddrPort(dest)
-
-	// We use nil as local addr to force runtime to find the best suitable local address IP given the destination
-	// address as context.
-	udpConn, err := net.DialUDP("udp", nil, addr)
-	if err != nil {
-		return nil, fmt.Errorf("unable to dial udp to origin %s: %w", dest, err)
-	}
-
-	return udpConn, nil
-}

ingress/origins/dns.go

@@ -0,0 +1,219 @@
+package origins
+
+import (
+	"context"
+	"crypto/rand"
+	"math/big"
+	"net"
+	"net/netip"
+	"slices"
+	"sync"
+	"time"
+
+	"github.com/rs/zerolog"
+
+	"github.com/cloudflare/cloudflared/ingress"
+)
+
+const (
+	// We need a DNS record:
+	// 1. That will be around for as long as cloudflared is
+	// 2. That Cloudflare controls: to allow us to make changes if needed
+	// 3. That is an external record to a typical customer's network: enforcing that the DNS request go to the
+	//    local DNS resolver over any local /etc/host configurations setup.
+	// 4. That cloudflared would normally query: ensuring that users with a positive security model for DNS queries
+	//    don't need to adjust anything.
+	//
+	// This hostname is one that used during the edge discovery process and as such satisfies the above constraints.
+	defaultLookupHost          = "region1.v2.argotunnel.com"
+	defaultResolverPort uint16 = 53
+
+	// We want the refresh time to be short to accommodate DNS resolver changes locally, but not too frequent as to
+	// shuffle the resolver if multiple are configured.
+	refreshFreq    = 5 * time.Minute
+	refreshTimeout = 5 * time.Second
+)
+
+var (
+	// Virtual DNS service address
+	VirtualDNSServiceAddr = netip.AddrPortFrom(netip.MustParseAddr("2606:4700:0cf1:2000:0000:0000:0000:0001"), 53)
+
+	defaultResolverAddr = netip.AddrPortFrom(netip.MustParseAddr("127.0.0.1"), defaultResolverPort)
+)
+
+type netDial func(network string, address string) (net.Conn, error)
+
+// DNSResolverService will make DNS requests to the local DNS resolver via the Dial method.
+type DNSResolverService struct {
+	addresses  []netip.AddrPort
+	addressesM sync.RWMutex
+	static     bool
+	dialer     ingress.OriginDialer
+	resolver   peekResolver
+	logger     *zerolog.Logger
+	metrics    Metrics
+}
+
+func NewDNSResolverService(dialer ingress.OriginDialer, logger *zerolog.Logger, metrics Metrics) *DNSResolverService {
+	return &DNSResolverService{
+		addresses: []netip.AddrPort{defaultResolverAddr},
+		dialer:    dialer,
+		resolver:  &resolver{dialFunc: net.Dial},
+		logger:    logger,
+		metrics:   metrics,
+	}
+}
+
+func NewStaticDNSResolverService(resolverAddrs []netip.AddrPort, dialer ingress.OriginDialer, logger *zerolog.Logger, metrics Metrics) *DNSResolverService {
+	s := NewDNSResolverService(dialer, logger, metrics)
+	s.addresses = resolverAddrs
+	s.static = true
+	return s
+}
+
+func (s *DNSResolverService) DialTCP(ctx context.Context, _ netip.AddrPort) (net.Conn, error) {
+	s.metrics.IncrementDNSTCPRequests()
+	dest := s.getAddress()
+	// The dialer ignores the provided address because the request will instead go to the local DNS resolver.
+	return s.dialer.DialTCP(ctx, dest)
+}
+
+func (s *DNSResolverService) DialUDP(_ netip.AddrPort) (net.Conn, error) {
+	s.metrics.IncrementDNSUDPRequests()
+	dest := s.getAddress()
+	// The dialer ignores the provided address because the request will instead go to the local DNS resolver.
+	return s.dialer.DialUDP(dest)
+}
+
+// StartRefreshLoop is a routine that is expected to run in the background to update the DNS local resolver if
+// adjusted while the cloudflared process is running.
+// Does not run when the resolver was provided with external resolver addresses via CLI.
+func (s *DNSResolverService) StartRefreshLoop(ctx context.Context) {
+	if s.static {
+		s.logger.Debug().Msgf("Canceled DNS local resolver refresh loop because static resolver addresses were provided: %s", s.addresses)
+		return
+	}
+	// Call update first to load an address before handling traffic
+	err := s.update(ctx)
+	if err != nil {
+		s.logger.Err(err).Msg("Failed to initialize DNS local resolver")
+	}
+	for {
+		select {
+		case <-ctx.Done():
+			return
+		case <-time.Tick(refreshFreq):
+			err := s.update(ctx)
+			if err != nil {
+				s.logger.Err(err).Msg("Failed to refresh DNS local resolver")
+			}
+		}
+	}
+}
+
+func (s *DNSResolverService) update(ctx context.Context) error {
+	ctx, cancel := context.WithTimeout(ctx, refreshTimeout)
+	defer cancel()
+	// Make a standard DNS request to a well-known DNS record that will last a long time
+	_, err := s.resolver.lookupNetIP(ctx, defaultLookupHost)
+	if err != nil {
+		return err
+	}
+
+	// Validate the address before updating internal reference
+	_, address := s.resolver.addr()
+	peekAddrPort, err := netip.ParseAddrPort(address)
+	if err == nil {
+		s.setAddress(peekAddrPort)
+		return nil
+	}
+	// It's possible that the address didn't have an attached port, attempt to parse just the address and use
+	// the default port 53
+	peekAddr, err := netip.ParseAddr(address)
+	if err != nil {
+		return err
+	}
+	s.setAddress(netip.AddrPortFrom(peekAddr, defaultResolverPort))
+	return nil
+}
+
+// returns the address from the peekResolver or from the static addresses if provided.
+// If multiple addresses are provided in the static addresses pick one randomly.
+func (s *DNSResolverService) getAddress() netip.AddrPort {
+	s.addressesM.RLock()
+	defer s.addressesM.RUnlock()
+	l := len(s.addresses)
+	if l <= 0 {
+		return defaultResolverAddr
+	}
+	if l == 1 {
+		return s.addresses[0]
+	}
+	// Only initialize the random selection if there is more than one element in the list.
+	var i int64 = 0
+	r, err := rand.Int(rand.Reader, big.NewInt(int64(l)))
+	// We ignore errors from crypto rand and use index 0; this should be extremely unlikely and the
+	// list index doesn't need to be cryptographically secure, but linters insist.
+	if err == nil {
+		i = r.Int64()
+	}
+	return s.addresses[i]
+}
+
+// lock and update the address used for the local DNS resolver
+func (s *DNSResolverService) setAddress(addr netip.AddrPort) {
+	s.addressesM.Lock()
+	defer s.addressesM.Unlock()
+	if !slices.Contains(s.addresses, addr) {
+		s.logger.Debug().Msgf("Updating DNS local resolver: %s", addr)
+	}
+	// We only store one address when reading the peekResolver, so we just replace the whole list.
+	s.addresses = []netip.AddrPort{addr}
+}
+
+type peekResolver interface {
+	addr() (network string, address string)
+	lookupNetIP(ctx context.Context, host string) ([]netip.Addr, error)
+}
+
+// resolver is a shim that inspects the go runtime's DNS resolution process to capture the DNS resolver
+// address used to complete a DNS request.
+type resolver struct {
+	network  string
+	address  string
+	dialFunc netDial
+}
+
+func (r *resolver) addr() (network string, address string) {
+	return r.network, r.address
+}
+
+func (r *resolver) lookupNetIP(ctx context.Context, host string) ([]netip.Addr, error) {
+	resolver := &net.Resolver{
+		PreferGo: true,
+		// Use the peekDial to inspect the results of the DNS resolver used during the LookupIPAddr call.
+		Dial: r.peekDial,
+	}
+	return resolver.LookupNetIP(ctx, "ip", host)
+}
+
+func (r *resolver) peekDial(ctx context.Context, network, address string) (net.Conn, error) {
+	r.network = network
+	r.address = address
+	return r.dialFunc(network, address)
+}
+
+// NewDNSDialer creates a custom dialer for the DNS resolver service to utilize.
+func NewDNSDialer() *ingress.Dialer {
+	return &ingress.Dialer{
+		Dialer: net.Dialer{
+			// We want short timeouts for the DNS requests
+			Timeout: 5 * time.Second,
+			// We do not want keep alive since the edge will not reuse TCP connections per request
+			KeepAlive: -1,
+			KeepAliveConfig: net.KeepAliveConfig{
+				Enable: false,
+			},
+		},
+	}
+}

ingress/origins/dns_test.go

@@ -0,0 +1,195 @@
+package origins
+
+import (
+	"context"
+	"errors"
+	"net"
+	"net/netip"
+	"slices"
+	"testing"
+	"time"
+
+	"github.com/rs/zerolog"
+)
+
+func TestDNSResolver_DefaultResolver(t *testing.T) {
+	log := zerolog.Nop()
+	service := NewDNSResolverService(NewDNSDialer(), &log, &noopMetrics{})
+	mockResolver := &mockPeekResolver{
+		address: "127.0.0.2:53",
+	}
+	service.resolver = mockResolver
+	validateAddrs(t, []netip.AddrPort{defaultResolverAddr}, service.addresses)
+}
+
+func TestStaticDNSResolver_DefaultResolver(t *testing.T) {
+	log := zerolog.Nop()
+	addresses := []netip.AddrPort{netip.MustParseAddrPort("1.1.1.1:53"), netip.MustParseAddrPort("1.0.0.1:53")}
+	service := NewStaticDNSResolverService(addresses, NewDNSDialer(), &log, &noopMetrics{})
+	mockResolver := &mockPeekResolver{
+		address: "127.0.0.2:53",
+	}
+	service.resolver = mockResolver
+	validateAddrs(t, addresses, service.addresses)
+}
+
+func TestDNSResolver_UpdateResolverAddress(t *testing.T) {
+	log := zerolog.Nop()
+	service := NewDNSResolverService(NewDNSDialer(), &log, &noopMetrics{})
+
+	mockResolver := &mockPeekResolver{}
+	service.resolver = mockResolver
+
+	tests := []struct {
+		addr     string
+		expected netip.AddrPort
+	}{
+		{"127.0.0.2:53", netip.MustParseAddrPort("127.0.0.2:53")},
+		// missing port should be added (even though this is unlikely to happen)
+		{"127.0.0.3", netip.MustParseAddrPort("127.0.0.3:53")},
+	}
+
+	for _, test := range tests {
+		mockResolver.address = test.addr
+		// Update the resolver address
+		err := service.update(t.Context())
+		if err != nil {
+			t.Error(err)
+		}
+		// Validate expected
+		validateAddrs(t, []netip.AddrPort{test.expected}, service.addresses)
+	}
+}
+
+func TestStaticDNSResolver_RefreshLoopExits(t *testing.T) {
+	log := zerolog.Nop()
+	addresses := []netip.AddrPort{netip.MustParseAddrPort("1.1.1.1:53"), netip.MustParseAddrPort("1.0.0.1:53")}
+	service := NewStaticDNSResolverService(addresses, NewDNSDialer(), &log, &noopMetrics{})
+
+	mockResolver := &mockPeekResolver{
+		address: "127.0.0.2:53",
+	}
+	service.resolver = mockResolver
+
+	ctx, cancel := context.WithCancel(t.Context())
+	defer cancel()
+
+	go service.StartRefreshLoop(ctx)
+
+	// Wait for the refresh loop to end _and_ not update the addresses
+	time.Sleep(10 * time.Millisecond)
+
+	// Validate expected
+	validateAddrs(t, addresses, service.addresses)
+}
+
+func TestDNSResolver_UpdateResolverAddressInvalid(t *testing.T) {
+	log := zerolog.Nop()
+	service := NewDNSResolverService(NewDNSDialer(), &log, &noopMetrics{})
+	mockResolver := &mockPeekResolver{}
+	service.resolver = mockResolver
+
+	invalidAddresses := []string{
+		"999.999.999.999",
+		"localhost",
+		"255.255.255",
+	}
+
+	for _, addr := range invalidAddresses {
+		mockResolver.address = addr
+		// Update the resolver address should not update for these invalid addresses
+		err := service.update(t.Context())
+		if err == nil {
+			t.Error("service update should throw an error")
+		}
+		// Validate expected
+		validateAddrs(t, []netip.AddrPort{defaultResolverAddr}, service.addresses)
+	}
+}
+
+func TestDNSResolver_UpdateResolverErrorIgnored(t *testing.T) {
+	log := zerolog.Nop()
+	service := NewDNSResolverService(NewDNSDialer(), &log, &noopMetrics{})
+	resolverErr := errors.New("test resolver error")
+	mockResolver := &mockPeekResolver{err: resolverErr}
+	service.resolver = mockResolver
+
+	// Update the resolver address should not update when the resolver cannot complete the lookup
+	err := service.update(t.Context())
+	if err != resolverErr {
+		t.Error("service update should throw an error")
+	}
+	// Validate expected
+	validateAddrs(t, []netip.AddrPort{defaultResolverAddr}, service.addresses)
+}
+
+func TestDNSResolver_DialUDPUsesResolvedAddress(t *testing.T) {
+	log := zerolog.Nop()
+	mockDialer := &mockDialer{expected: defaultResolverAddr}
+	service := NewDNSResolverService(mockDialer, &log, &noopMetrics{})
+	mockResolver := &mockPeekResolver{}
+	service.resolver = mockResolver
+
+	// Attempt a dial to 127.0.0.2:53 which should be ignored and instead resolve to 127.0.0.1:53
+	_, err := service.DialUDP(netip.MustParseAddrPort("127.0.0.2:53"))
+	if err != nil {
+		t.Error(err)
+	}
+}
+
+func TestDNSResolver_DialTCPUsesResolvedAddress(t *testing.T) {
+	log := zerolog.Nop()
+	mockDialer := &mockDialer{expected: defaultResolverAddr}
+	service := NewDNSResolverService(mockDialer, &log, &noopMetrics{})
+	mockResolver := &mockPeekResolver{}
+	service.resolver = mockResolver
+
+	// Attempt a dial to 127.0.0.2:53 which should be ignored and instead resolve to 127.0.0.1:53
+	_, err := service.DialTCP(t.Context(), netip.MustParseAddrPort("127.0.0.2:53"))
+	if err != nil {
+		t.Error(err)
+	}
+}
+
+type mockPeekResolver struct {
+	err     error
+	address string
+}
+
+func (r *mockPeekResolver) addr() (network, address string) {
+	return "udp", r.address
+}
+
+func (r *mockPeekResolver) lookupNetIP(ctx context.Context, host string) ([]netip.Addr, error) {
+	// We can return an empty result as it doesn't matter as long as the lookup doesn't fail
+	return []netip.Addr{}, r.err
+}
+
+type mockDialer struct {
+	expected netip.AddrPort
+}
+
+func (d *mockDialer) DialTCP(ctx context.Context, addr netip.AddrPort) (net.Conn, error) {
+	if d.expected != addr {
+		return nil, errors.New("unexpected address dialed")
+	}
+	return nil, nil
+}
+
+func (d *mockDialer) DialUDP(addr netip.AddrPort) (net.Conn, error) {
+	if d.expected != addr {
+		return nil, errors.New("unexpected address dialed")
+	}
+	return nil, nil
+}
+
+func validateAddrs(t *testing.T, expected []netip.AddrPort, actual []netip.AddrPort) {
+	if len(actual) != len(expected) {
+		t.Errorf("addresses should only contain one element: %s", actual)
+	}
+	for _, e := range expected {
+		if !slices.Contains(actual, e) {
+			t.Errorf("missing address: %s in %s", e, actual)
+		}
+	}
+}

ingress/origins/metrics.go

@@ -0,0 +1,40 @@
+package origins
+
+import (
+	"github.com/prometheus/client_golang/prometheus"
+)
+
+const (
+	namespace = "cloudflared"
+	subsystem = "virtual_origins"
+)
+
+type Metrics interface {
+	IncrementDNSUDPRequests()
+	IncrementDNSTCPRequests()
+}
+
+type metrics struct {
+	dnsResolverRequests *prometheus.CounterVec
+}
+
+func (m *metrics) IncrementDNSUDPRequests() {
+	m.dnsResolverRequests.WithLabelValues("udp").Inc()
+}
+
+func (m *metrics) IncrementDNSTCPRequests() {
+	m.dnsResolverRequests.WithLabelValues("tcp").Inc()
+}
+
+func NewMetrics(registerer prometheus.Registerer) Metrics {
+	m := &metrics{
+		dnsResolverRequests: prometheus.NewCounterVec(prometheus.CounterOpts{
+			Namespace: namespace,
+			Subsystem: subsystem,
+			Name:      "dns_requests_total",
+			Help:      "Total count of DNS requests that have been proxied to the virtual DNS resolver origin",
+		}, []string{"protocol"}),
+	}
+	registerer.MustRegister(m.dnsResolverRequests)
+	return m
+}

ingress/origins/metrics_test.go

@@ -0,0 +1,6 @@
+package origins
+
+type noopMetrics struct{}
+
+func (noopMetrics) IncrementDNSUDPRequests() {}
+func (noopMetrics) IncrementDNSTCPRequests() {}

management/middleware.go

@@ -12,14 +12,7 @@ const (
 	accessClaimsCtxKey ctxKey = iota
 )
 
-const (
-	connectorIDQuery = "connector_id"
-	accessTokenQuery = "access_token"
-)
-
-var (
-	errMissingAccessToken = managementError{Code: 1001, Message: "missing access_token query parameter"}
-)
+var errMissingAccessToken = managementError{Code: 1001, Message: "missing access_token query parameter"}
 
 // HTTP middleware setting the parsed access_token claims in the request context
 func ValidateAccessTokenQueryMiddleware(next http.Handler) http.Handler {
@@ -30,7 +23,7 @@ func ValidateAccessTokenQueryMiddleware(next http.Handler) http.Handler {
 			writeHTTPErrorResponse(w, errMissingAccessToken)
 			return
 		}
-		token, err := parseToken(accessToken)
+		token, err := ParseToken(accessToken)
 		if err != nil {
 			writeHTTPErrorResponse(w, errMissingAccessToken)
 			return

management/token.go

@@ -7,9 +7,12 @@ import (
 	"github.com/go-jose/go-jose/v4/jwt"
 )
 
+const tunnelstoreFEDIssuer = "fed-tunnelstore"
+
 type managementTokenClaims struct {
 	Tunnel tunnel `json:"tun"`
 	Actor  actor  `json:"actor"`
+	jwt.Claims
 }
 
 // VerifyTunnel compares the tun claim isn't empty
@@ -37,7 +40,7 @@ func (t *actor) verify() bool {
 	return t.ID != ""
 }
 
-func parseToken(token string) (*managementTokenClaims, error) {
+func ParseToken(token string) (*managementTokenClaims, error) {
 	jwt, err := jwt.ParseSigned(token, []jose.SignatureAlgorithm{jose.ES256})
 	if err != nil {
 		return nil, fmt.Errorf("malformed jwt: %v", err)
@@ -54,3 +57,7 @@ func parseToken(token string) (*managementTokenClaims, error) {
 	}
 	return &claims, nil
 }
+
+func (m *managementTokenClaims) IsFed() bool {
+	return m.Issuer == tunnelstoreFEDIssuer
+}

management/token_test.go

@@ -12,7 +12,7 @@ import (
 )
 
 const (
-	validToken = "eyJ0eXAiOiJKV1QiLCJhbGciOiJFUzI1NiIsImtpZCI6IjEifQ.eyJ0dW4iOnsiaWQiOiI3YjA5ODE0OS01MWZlLTRlZTUtYTY4Ny0zZTM3NDQ2NmVmYzciLCJhY2NvdW50X3RhZyI6ImNkMzkxZTljMDYyNmE4Zjc2Y2IxZjY3MGY2NTkxYjA1In0sImFjdG9yIjp7ImlkIjoiZGNhcnJAY2xvdWRmbGFyZS5jb20iLCJzdXBwb3J0IjpmYWxzZX0sInJlcyI6WyJsb2dzIl0sImV4cCI6MTY3NzExNzY5NiwiaWF0IjoxNjc3MTE0MDk2LCJpc3MiOiJ0dW5uZWxzdG9yZSJ9.mKenOdOy3Xi4O-grldFnAAemdlE9WajEpTDC_FwezXQTstWiRTLwU65P5jt4vNsIiZA4OJRq7bH-QYID9wf9NA"
+	validToken = "eyJ0eXAiOiJKV1QiLCJhbGciOiJFUzI1NiIsImtpZCI6IjEifQ.eyJ0dW4iOnsiaWQiOiI3YjA5ODE0OS01MWZlLTRlZTUtYTY4Ny0zZTM3NDQ2NmVmYzciLCJhY2NvdW50X3RhZyI6ImNkMzkxZTljMDYyNmE4Zjc2Y2IxZjY3MGY2NTkxYjA1In0sImFjdG9yIjp7ImlkIjoiZGNhcnJAY2xvdWRmbGFyZS5jb20iLCJzdXBwb3J0IjpmYWxzZX0sInJlcyI6WyJsb2dzIl0sImV4cCI6MTY3NzExNzY5NiwiaWF0IjoxNjc3MTE0MDk2LCJpc3MiOiJ0dW5uZWxzdG9yZSJ9.mKenOdOy3Xi4O-grldFnAAemdlE9WajEpTDC_FwezXQTstWiRTLwU65P5jt4vNsIiZA4OJRq7bH-QYID9wf9NA" // nolint: gosec
 
 	accountTag = "cd391e9c0626a8f76cb1f670f6591b05"
 	tunnelID   = "7b098149-51fe-4ee5-a687-3e374466efc7"
@@ -105,12 +105,12 @@ func TestParseToken(t *testing.T) {
 	} {
 		t.Run(test.name, func(t *testing.T) {
 			jwt := signToken(t, test.claims, key)
-			claims, err := parseToken(jwt)
+			claims, err := ParseToken(jwt)
 			if test.err != nil {
 				require.EqualError(t, err, test.err.Error())
 				return
 			}
-			require.Nil(t, err)
+			require.NoError(t, err)
 			require.Equal(t, test.claims, *claims)
 		})
 	}

orchestration/config.go

@@ -2,7 +2,6 @@ package orchestration
 
 import (
 	"encoding/json"
-	"time"
 
 	"github.com/cloudflare/cloudflared/config"
 	"github.com/cloudflare/cloudflared/ingress"
@@ -20,9 +19,9 @@ type newLocalConfig struct {
 
 // Config is the original config as read and parsed by cloudflared.
 type Config struct {
-	Ingress      *ingress.Ingress
-	WarpRouting  ingress.WarpRoutingConfig
-	WriteTimeout time.Duration
+	Ingress             *ingress.Ingress
+	WarpRouting         ingress.WarpRoutingConfig
+	OriginDialerService *ingress.OriginDialerService
 
 	// Extra settings used to configure this instance but that are not eligible for remotely management
 	// ie. (--protocol, --loglevel, ...)

orchestration/orchestrator.go

@@ -38,7 +38,9 @@ type Orchestrator struct {
 	tags   []pogs.Tag
 	// flowLimiter tracks active sessions across the tunnel and limits new sessions if they are above the limit.
 	flowLimiter cfdflow.Limiter
-	log         *zerolog.Logger
+	// Origin dialer service to manage egress socket dialing.
+	originDialerService *ingress.OriginDialerService
+	log                 *zerolog.Logger
 
 	// orchestrator must not handle any more updates after shutdownC is closed
 	shutdownC <-chan struct{}
@@ -50,18 +52,20 @@ func NewOrchestrator(ctx context.Context,
 	config *Config,
 	tags []pogs.Tag,
 	internalRules []ingress.Rule,
-	log *zerolog.Logger) (*Orchestrator, error) {
+	log *zerolog.Logger,
+) (*Orchestrator, error) {
 	o := &Orchestrator{
 		// Lowest possible version, any remote configuration will have version higher than this
 		// Starting at -1 allows a configuration migration (local to remote) to override the current configuration as it
 		// will start at version 0.
-		currentVersion: -1,
-		internalRules:  internalRules,
-		config:         config,
-		tags:           tags,
-		flowLimiter:    cfdflow.NewLimiter(config.WarpRouting.MaxActiveFlows),
-		log:            log,
-		shutdownC:      ctx.Done(),
+		currentVersion:      -1,
+		internalRules:       internalRules,
+		config:              config,
+		tags:                tags,
+		flowLimiter:         cfdflow.NewLimiter(config.WarpRouting.MaxActiveFlows),
+		originDialerService: config.OriginDialerService,
+		log:                 log,
+		shutdownC:           ctx.Done(),
 	}
 	if err := o.updateIngress(*config.Ingress, config.WarpRouting); err != nil {
 		return nil, err
@@ -175,7 +179,15 @@ func (o *Orchestrator) updateIngress(ingressRules ingress.Ingress, warpRouting i
 	// Update the flow limit since the configuration might have changed
 	o.flowLimiter.SetLimit(warpRouting.MaxActiveFlows)
 
-	proxy := proxy.NewOriginProxy(ingressRules, warpRouting, o.tags, o.flowLimiter, o.config.WriteTimeout, o.log)
+	// Update the origin dialer service with the new dialer settings
+	// We need to update the dialer here instead of creating a new instance of OriginDialerService because it has
+	// its own references and go routines. Specifically, the UDP dialer is a reference to this same service all the
+	// way into the datagram manager. Reconstructing the datagram manager is not something we currently provide during
+	// runtime in response to a configuration push except when starting a tunnel connection.
+	o.originDialerService.UpdateDefaultDialer(ingress.NewDialer(warpRouting))
+
+	// Create and replace the origin proxy with a new instance
+	proxy := proxy.NewOriginProxy(ingressRules, o.originDialerService, o.tags, o.flowLimiter, o.log)
 	o.proxy.Store(proxy)
 	o.config.Ingress = &ingressRules
 	o.config.WarpRouting = warpRouting

orchestration/orchestrator_test.go

@@ -41,6 +41,11 @@ var (
 			Value: "test",
 		},
 	}
+	testDefaultDialer = ingress.NewDialer(ingress.WarpRoutingConfig{
+		ConnectTimeout: config.CustomDuration{Duration: 1 * time.Second},
+		TCPKeepAlive:   config.CustomDuration{Duration: 15 * time.Second},
+		MaxActiveFlows: 0,
+	})
 )
 
 // TestUpdateConfiguration tests that
@@ -50,8 +55,13 @@ var (
 // - configurations can be deserialized
 // - receiving an old version is noop
 func TestUpdateConfiguration(t *testing.T) {
+	originDialer := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   testDefaultDialer,
+		TCPWriteTimeout: 1 * time.Second,
+	}, &testLogger)
 	initConfig := &Config{
-		Ingress: &ingress.Ingress{},
+		Ingress:             &ingress.Ingress{},
+		OriginDialerService: originDialer,
 	}
 	orchestrator, err := NewOrchestrator(t.Context(), initConfig, testTags, []ingress.Rule{ingress.NewManagementRule(management.New("management.argotunnel.com", false, "1.1.1.1:80", uuid.Nil, "", &testLogger, nil))}, &testLogger)
 	require.NoError(t, err)
@@ -179,8 +189,13 @@ func TestUpdateConfiguration(t *testing.T) {
 // Validates that a new version 0 will be applied if the configuration is loaded locally.
 // This will happen when a locally managed tunnel is migrated to remote configuration and receives its first configuration.
 func TestUpdateConfiguration_FromMigration(t *testing.T) {
+	originDialer := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   testDefaultDialer,
+		TCPWriteTimeout: 1 * time.Second,
+	}, &testLogger)
 	initConfig := &Config{
-		Ingress: &ingress.Ingress{},
+		Ingress:             &ingress.Ingress{},
+		OriginDialerService: originDialer,
 	}
 	orchestrator, err := NewOrchestrator(t.Context(), initConfig, testTags, []ingress.Rule{}, &testLogger)
 	require.NoError(t, err)
@@ -205,8 +220,13 @@ func TestUpdateConfiguration_FromMigration(t *testing.T) {
 
 // Validates that the default ingress rule will be set if there is no rule provided from the remote.
 func TestUpdateConfiguration_WithoutIngressRule(t *testing.T) {
+	originDialer := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   testDefaultDialer,
+		TCPWriteTimeout: 1 * time.Second,
+	}, &testLogger)
 	initConfig := &Config{
-		Ingress: &ingress.Ingress{},
+		Ingress:             &ingress.Ingress{},
+		OriginDialerService: originDialer,
 	}
 	orchestrator, err := NewOrchestrator(t.Context(), initConfig, testTags, []ingress.Rule{}, &testLogger)
 	require.NoError(t, err)
@@ -244,6 +264,11 @@ func TestConcurrentUpdateAndRead(t *testing.T) {
 	require.NoError(t, err)
 	defer tcpOrigin.Close()
 
+	originDialer := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   testDefaultDialer,
+		TCPWriteTimeout: 1 * time.Second,
+	}, &testLogger)
+
 	var (
 		configJSONV1 = []byte(fmt.Sprintf(`
 {
@@ -296,7 +321,8 @@ func TestConcurrentUpdateAndRead(t *testing.T) {
 		appliedV2 = make(chan struct{})
 
 		initConfig = &Config{
-			Ingress: &ingress.Ingress{},
+			Ingress:             &ingress.Ingress{},
+			OriginDialerService: originDialer,
 		}
 	)
 
@@ -313,7 +339,7 @@ func TestConcurrentUpdateAndRead(t *testing.T) {
 	go func() {
 		serveTCPOrigin(t, tcpOrigin, &wg)
 	}()
-	for i := 0; i < concurrentRequests; i++ {
+	for i := range concurrentRequests {
 		originProxy, err := orchestrator.GetOriginProxy()
 		require.NoError(t, err)
 		wg.Add(1)
@@ -323,48 +349,37 @@ func TestConcurrentUpdateAndRead(t *testing.T) {
 			assert.NoError(t, err, "proxyHTTP %d failed %v", i, err)
 			defer resp.Body.Close()
 
-			var warpRoutingDisabled bool
 			// The response can be from initOrigin, http_status:204 or http_status:418
 			switch resp.StatusCode {
-			// v1 proxy, warp enabled
+			// v1 proxy
 			case 200:
 				body, err := io.ReadAll(resp.Body)
 				assert.NoError(t, err)
 				assert.Equal(t, t.Name(), string(body))
-				warpRoutingDisabled = false
-			// v2 proxy, warp disabled
+			// v2 proxy
 			case 204:
 				assert.Greater(t, i, concurrentRequests/4)
-				warpRoutingDisabled = true
-			// v3 proxy, warp enabled
+			// v3 proxy
 			case 418:
 				assert.Greater(t, i, concurrentRequests/2)
-				warpRoutingDisabled = false
 			}
 
 			// Once we have originProxy, it won't be changed by configuration updates.
 			// We can infer the version by the ProxyHTTP response code
 			pr, pw := io.Pipe()
-
 			w := newRespReadWriteFlusher()
 
 			// Write TCP message and make sure it's echo back. This has to be done in a go routune since ProxyTCP doesn't
 			// return until the stream is closed.
-			if !warpRoutingDisabled {
-				wg.Add(1)
-				go func() {
-					defer wg.Done()
-					defer pw.Close()
-					tcpEyeball(t, pw, tcpBody, w)
-				}()
-			}
+			wg.Add(1)
+			go func() {
+				defer wg.Done()
+				defer pw.Close()
+				tcpEyeball(t, pw, tcpBody, w)
+			}()
 
 			err = proxyTCP(ctx, originProxy, tcpOrigin.Addr().String(), w, pr)
-			if warpRoutingDisabled {
-				assert.Error(t, err, "expect proxyTCP %d to return error", i)
-			} else {
-				assert.NoError(t, err, "proxyTCP %d failed %v", i, err)
-			}
+			assert.NoError(t, err, "proxyTCP %d failed %v", i, err)
 		}(i, originProxy)
 
 		if i == concurrentRequests/4 {
@@ -406,39 +421,47 @@ func TestOverrideWarpRoutingConfigWithLocalValues(t *testing.T) {
 		require.EqualValues(t, expectedValue, warpRouting["maxActiveFlows"])
 	}
 
-	remoteValue := uint64(100)
-	remoteIngress := ingress.Ingress{}
+	originDialer := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   testDefaultDialer,
+		TCPWriteTimeout: 1 * time.Second,
+	}, &testLogger)
+
+	// All the possible values set for MaxActiveFlows from the various points that can provide it:
+	// 1. Initialized value
+	// 2. Local CLI flag config
+	// 3. Remote configuration value
+	initValue := uint64(0)
+	localValue := uint64(100)
+	remoteValue := uint64(500)
+
+	initConfig := &Config{
+		Ingress: &ingress.Ingress{},
+		WarpRouting: ingress.WarpRoutingConfig{
+			MaxActiveFlows: initValue,
+		},
+		OriginDialerService: originDialer,
+		ConfigurationFlags: map[string]string{
+			flags.MaxActiveFlows: fmt.Sprintf("%d", localValue),
+		},
+	}
+
+	// We expect the local configuration flag to be the starting value
+	orchestrator, err := NewOrchestrator(ctx, initConfig, testTags, []ingress.Rule{}, &testLogger)
+	require.NoError(t, err)
+
+	assertMaxActiveFlows(orchestrator, localValue)
+
+	// Assigning the MaxActiveFlows in the remote config should be ignored over the local config
 	remoteWarpConfig := ingress.WarpRoutingConfig{
 		MaxActiveFlows: remoteValue,
 	}
-	remoteConfig := &Config{
-		Ingress:            &remoteIngress,
-		WarpRouting:        remoteWarpConfig,
-		ConfigurationFlags: map[string]string{},
-	}
-	orchestrator, err := NewOrchestrator(ctx, remoteConfig, testTags, []ingress.Rule{}, &testLogger)
-	require.NoError(t, err)
 
-	assertMaxActiveFlows(orchestrator, remoteValue)
-
-	// Add a local override for the maxActiveFlows
-	localValue := uint64(500)
-	remoteConfig.ConfigurationFlags[flags.MaxActiveFlows] = fmt.Sprintf("%d", localValue)
 	// Force a configuration refresh
-	err = orchestrator.updateIngress(remoteIngress, remoteWarpConfig)
+	err = orchestrator.updateIngress(ingress.Ingress{}, remoteWarpConfig)
 	require.NoError(t, err)
 
 	// Check the value being used is the local one
 	assertMaxActiveFlows(orchestrator, localValue)
-
-	// Remove local override for the maxActiveFlows
-	delete(remoteConfig.ConfigurationFlags, flags.MaxActiveFlows)
-	// Force a configuration refresh
-	err = orchestrator.updateIngress(remoteIngress, remoteWarpConfig)
-	require.NoError(t, err)
-
-	// Check the value being used is now the remote again
-	assertMaxActiveFlows(orchestrator, remoteValue)
 }
 
 func proxyHTTP(originProxy connection.OriginProxy, hostname string) (*http.Response, error) {
@@ -546,6 +569,10 @@ func updateWithValidation(t *testing.T, orchestrator *Orchestrator, version int3
 
 // TestClosePreviousProxies makes sure proxies started in the previous configuration version are shutdown
 func TestClosePreviousProxies(t *testing.T) {
+	originDialer := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   testDefaultDialer,
+		TCPWriteTimeout: 1 * time.Second,
+	}, &testLogger)
 	var (
 		hostname             = "hello.tunnel1.org"
 		configWithHelloWorld = []byte(fmt.Sprintf(`
@@ -576,7 +603,8 @@ func TestClosePreviousProxies(t *testing.T) {
 }
 `)
 		initConfig = &Config{
-			Ingress: &ingress.Ingress{},
+			Ingress:             &ingress.Ingress{},
+			OriginDialerService: originDialer,
 		}
 	)
 
@@ -638,8 +666,13 @@ func TestPersistentConnection(t *testing.T) {
 		hostname = "http://ws.tunnel.org"
 	)
 	msg := t.Name()
+	originDialer := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   testDefaultDialer,
+		TCPWriteTimeout: 1 * time.Second,
+	}, &testLogger)
 	initConfig := &Config{
-		Ingress: &ingress.Ingress{},
+		Ingress:             &ingress.Ingress{},
+		OriginDialerService: originDialer,
 	}
 	orchestrator, err := NewOrchestrator(t.Context(), initConfig, testTags, []ingress.Rule{}, &testLogger)
 	require.NoError(t, err)
@@ -752,8 +785,9 @@ func TestSerializeLocalConfig(t *testing.T) {
 		ConfigurationFlags: map[string]string{"a": "b"},
 	}
 
-	result, _ := json.Marshal(c)
-	fmt.Println(string(result))
+	result, err := json.Marshal(c)
+	require.NoError(t, err)
+	require.JSONEq(t, `{"__configuration_flags":{"a":"b"},"ingress":[],"warp-routing":{"connectTimeout":0,"tcpKeepAlive":0}}`, string(result))
 }
 
 func wsEcho(w http.ResponseWriter, r *http.Request) {

postinst.sh

@@ -1,5 +1,5 @@
 #!/bin/bash
 set -eu
-ln -s /usr/bin/cloudflared /usr/local/bin/cloudflared
+ln -sf /usr/bin/cloudflared /usr/local/bin/cloudflared
 mkdir -p /usr/local/etc/cloudflared/
 touch /usr/local/etc/cloudflared/.installedFromPackageManager || true

proxy/proxy.go

@@ -5,11 +5,11 @@ import (
 	"fmt"
 	"io"
 	"net/http"
+	"net/netip"
 	"strconv"
 	"time"
 
 	"github.com/pkg/errors"
-	pkgerrors "github.com/pkg/errors"
 	"github.com/rs/zerolog"
 	"go.opentelemetry.io/otel/attribute"
 	"go.opentelemetry.io/otel/trace"
@@ -35,7 +35,7 @@ const (
 // Proxy represents a means to Proxy between cloudflared and the origin services.
 type Proxy struct {
 	ingressRules ingress.Ingress
-	warpRouting  *ingress.WarpRoutingService
+	originDialer ingress.OriginTCPDialer
 	tags         []pogs.Tag
 	flowLimiter  cfdflow.Limiter
 	log          *zerolog.Logger
@@ -44,21 +44,19 @@ type Proxy struct {
 // NewOriginProxy returns a new instance of the Proxy struct.
 func NewOriginProxy(
 	ingressRules ingress.Ingress,
-	warpRouting ingress.WarpRoutingConfig,
+	originDialer ingress.OriginDialer,
 	tags []pogs.Tag,
 	flowLimiter cfdflow.Limiter,
-	writeTimeout time.Duration,
 	log *zerolog.Logger,
 ) *Proxy {
 	proxy := &Proxy{
 		ingressRules: ingressRules,
+		originDialer: originDialer,
 		tags:         tags,
 		flowLimiter:  flowLimiter,
 		log:          log,
 	}
 
-	proxy.warpRouting = ingress.NewWarpRoutingService(warpRouting, writeTimeout)
-
 	return proxy
 }
 
@@ -146,24 +144,18 @@ func (p *Proxy) ProxyHTTP(
 // ProxyTCP proxies to a TCP connection between the origin service and cloudflared.
 func (p *Proxy) ProxyTCP(
 	ctx context.Context,
-	rwa connection.ReadWriteAcker,
+	conn connection.ReadWriteAcker,
 	req *connection.TCPRequest,
 ) error {
 	incrementTCPRequests()
 	defer decrementTCPConcurrentRequests()
 
-	if p.warpRouting == nil {
-		err := errors.New(`cloudflared received a request from WARP client, but your configuration has disabled ingress from WARP clients. To enable this, set "warp-routing:\n\t enabled: true" in your config.yaml`)
-		p.log.Error().Msg(err.Error())
-		return err
-	}
-
 	logger := newTCPLogger(p.log, req)
 
 	// Try to start a new flow
 	if err := p.flowLimiter.Acquire(management.TCP.String()); err != nil {
 		logger.Warn().Msg("Too many concurrent flows being handled, rejecting tcp proxy")
-		return pkgerrors.Wrap(err, "failed to start tcp flow due to rate limiting")
+		return errors.Wrap(err, "failed to start tcp flow due to rate limiting")
 	}
 	defer p.flowLimiter.Release()
 
@@ -173,7 +165,14 @@ func (p *Proxy) ProxyTCP(
 	tracedCtx := tracing.NewTracedContext(serveCtx, req.CfTraceID, &logger)
 	logger.Debug().Msg("tcp proxy stream started")
 
-	if err := p.proxyStream(tracedCtx, rwa, req.Dest, p.warpRouting.Proxy, &logger); err != nil {
+	// Parse the destination into a netip.AddrPort
+	dest, err := netip.ParseAddrPort(req.Dest)
+	if err != nil {
+		logRequestError(&logger, err)
+		return err
+	}
+
+	if err := p.proxyTCPStream(tracedCtx, conn, dest, p.originDialer, &logger); err != nil {
 		logRequestError(&logger, err)
 		return err
 	}
@@ -279,14 +278,14 @@ func (p *Proxy) proxyStream(
 	tr *tracing.TracedContext,
 	rwa connection.ReadWriteAcker,
 	dest string,
-	connectionProxy ingress.StreamBasedOriginProxy,
+	originDialer ingress.StreamBasedOriginProxy,
 	logger *zerolog.Logger,
 ) error {
 	ctx := tr.Context
 	_, connectSpan := tr.Tracer().Start(ctx, "stream-connect")
 
 	start := time.Now()
-	originConn, err := connectionProxy.EstablishConnection(ctx, dest, logger)
+	originConn, err := originDialer.EstablishConnection(ctx, dest, logger)
 	if err != nil {
 		connectStreamErrors.Inc()
 		tracing.EndWithErrorStatus(connectSpan, err)
@@ -310,6 +309,45 @@ func (p *Proxy) proxyStream(
 	return nil
 }
 
+// proxyTCPStream proxies private network type TCP connections as a stream towards an available origin.
+//
+// This is different than proxyStream because it's not leveraged ingress rule services and uses the
+// originDialer from OriginDialerService.
+func (p *Proxy) proxyTCPStream(
+	tr *tracing.TracedContext,
+	tunnelConn connection.ReadWriteAcker,
+	dest netip.AddrPort,
+	originDialer ingress.OriginTCPDialer,
+	logger *zerolog.Logger,
+) error {
+	ctx := tr.Context
+	_, connectSpan := tr.Tracer().Start(ctx, "stream-connect")
+
+	start := time.Now()
+	originConn, err := originDialer.DialTCP(ctx, dest)
+	if err != nil {
+		connectStreamErrors.Inc()
+		tracing.EndWithErrorStatus(connectSpan, err)
+		return err
+	}
+	connectSpan.End()
+	defer originConn.Close()
+	logger.Debug().Msg("origin connection established")
+
+	encodedSpans := tr.GetSpans()
+
+	if err := tunnelConn.AckConnection(encodedSpans); err != nil {
+		connectStreamErrors.Inc()
+		return err
+	}
+
+	connectLatency.Observe(float64(time.Since(start).Milliseconds()))
+	logger.Debug().Msg("proxy stream acknowledged")
+
+	stream.Pipe(tunnelConn, originConn, logger)
+	return nil
+}
+
 func (p *Proxy) proxyLocalRequest(proxy ingress.HTTPLocalProxy, w connection.ResponseWriter, req *http.Request, isWebsocket bool) {
 	if isWebsocket {
 		// These headers are added since they are stripped off during an eyeball request to origintunneld, but they

proxy/proxy_test.go

@@ -33,17 +33,17 @@ import (
 	"github.com/cloudflare/cloudflared/connection"
 	"github.com/cloudflare/cloudflared/hello"
 	"github.com/cloudflare/cloudflared/ingress"
-	"github.com/cloudflare/cloudflared/logger"
 	"github.com/cloudflare/cloudflared/tracing"
 	"github.com/cloudflare/cloudflared/tunnelrpc/pogs"
 )
 
 var (
-	testTags        = []pogs.Tag{{Name: "Name", Value: "value"}}
-	noWarpRouting   = ingress.WarpRoutingConfig{}
-	testWarpRouting = ingress.WarpRoutingConfig{
-		ConnectTimeout: config.CustomDuration{Duration: time.Second},
-	}
+	testTags          = []pogs.Tag{{Name: "Name", Value: "value"}}
+	testDefaultDialer = ingress.NewDialer(ingress.WarpRoutingConfig{
+		ConnectTimeout: config.CustomDuration{Duration: 1 * time.Second},
+		TCPKeepAlive:   config.CustomDuration{Duration: 15 * time.Second},
+		MaxActiveFlows: 0,
+	})
 )
 
 type mockHTTPRespWriter struct {
@@ -162,7 +162,12 @@ func TestProxySingleOrigin(t *testing.T) {
 
 	require.NoError(t, ingressRule.StartOrigins(&log, ctx.Done()))
 
-	proxy := NewOriginProxy(ingressRule, noWarpRouting, testTags, cfdflow.NewLimiter(0), time.Duration(0), &log)
+	originDialer := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   testDefaultDialer,
+		TCPWriteTimeout: 1 * time.Second,
+	}, &log)
+
+	proxy := NewOriginProxy(ingressRule, originDialer, testTags, cfdflow.NewLimiter(0), &log)
 	t.Run("testProxyHTTP", testProxyHTTP(proxy))
 	t.Run("testProxyWebsocket", testProxyWebsocket(proxy))
 	t.Run("testProxySSE", testProxySSE(proxy))
@@ -357,7 +362,7 @@ type MultipleIngressTest struct {
 }
 
 func runIngressTestScenarios(t *testing.T, unvalidatedIngress []config.UnvalidatedIngressRule, tests []MultipleIngressTest) {
-	ingress, err := ingress.ParseIngress(&config.Configuration{
+	ingressRule, err := ingress.ParseIngress(&config.Configuration{
 		TunnelID: t.Name(),
 		Ingress:  unvalidatedIngress,
 	})
@@ -366,9 +371,14 @@ func runIngressTestScenarios(t *testing.T, unvalidatedIngress []config.Unvalidat
 	log := zerolog.Nop()
 
 	ctx, cancel := context.WithCancel(t.Context())
-	require.NoError(t, ingress.StartOrigins(&log, ctx.Done()))
+	require.NoError(t, ingressRule.StartOrigins(&log, ctx.Done()))
 
-	proxy := NewOriginProxy(ingress, noWarpRouting, testTags, cfdflow.NewLimiter(0), time.Duration(0), &log)
+	originDialer := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   testDefaultDialer,
+		TCPWriteTimeout: 1 * time.Second,
+	}, &log)
+
+	proxy := NewOriginProxy(ingressRule, originDialer, testTags, cfdflow.NewLimiter(0), &log)
 
 	for _, test := range tests {
 		responseWriter := newMockHTTPRespWriter()
@@ -416,7 +426,12 @@ func TestProxyError(t *testing.T) {
 
 	log := zerolog.Nop()
 
-	proxy := NewOriginProxy(ing, noWarpRouting, testTags, cfdflow.NewLimiter(0), time.Duration(0), &log)
+	originDialer := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   testDefaultDialer,
+		TCPWriteTimeout: 1 * time.Second,
+	}, &log)
+
+	proxy := NewOriginProxy(ing, originDialer, testTags, cfdflow.NewLimiter(0), &log)
 
 	responseWriter := newMockHTTPRespWriter()
 	req, err := http.NewRequest(http.MethodGet, "http://127.0.0.1", nil)
@@ -467,7 +482,7 @@ func (r *replayer) Bytes() []byte {
 // WS - TCP: When a tcp based ingress is configured on the origin and the
 // eyeball sends tcp packets wrapped in websockets. (E.g: cloudflared access).
 func TestConnections(t *testing.T) {
-	logger := logger.Create(nil)
+	log := zerolog.Nop()
 	replayer := &replayer{rw: bytes.NewBuffer([]byte{})}
 	type args struct {
 		ingressServiceScheme  string
@@ -475,9 +490,6 @@ func TestConnections(t *testing.T) {
 		eyeballResponseWriter connection.ResponseWriter
 		eyeballRequestBody    io.ReadCloser
 
-		// Can be set to nil to show warp routing is not enabled.
-		warpRoutingService *ingress.WarpRoutingService
-
 		// eyeball connection type.
 		connectionType connection.Type
 
@@ -488,6 +500,11 @@ func TestConnections(t *testing.T) {
 		flowLimiterResponse error
 	}
 
+	originDialer := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   testDefaultDialer,
+		TCPWriteTimeout: 0,
+	}, &log)
+
 	type want struct {
 		message []byte
 		headers http.Header
@@ -530,7 +547,6 @@ func TestConnections(t *testing.T) {
 				originService:         runEchoTCPService,
 				eyeballResponseWriter: newTCPRespWriter(replayer),
 				eyeballRequestBody:    newTCPRequestBody([]byte("test2")),
-				warpRoutingService:    ingress.NewWarpRoutingService(testWarpRouting, time.Duration(0)),
 				connectionType:        connection.TypeTCP,
 				requestHeaders: map[string][]string{
 					"Cf-Cloudflared-Proxy-Src": {"non-blank-value"},
@@ -548,7 +564,6 @@ func TestConnections(t *testing.T) {
 				originService:        runEchoWSService,
 				// eyeballResponseWriter gets set after roundtrip dial.
 				eyeballRequestBody: newPipedWSRequestBody([]byte("test3")),
-				warpRoutingService: ingress.NewWarpRoutingService(testWarpRouting, time.Duration(0)),
 				requestHeaders: map[string][]string{
 					"Cf-Cloudflared-Proxy-Src": {"non-blank-value"},
 				},
@@ -601,23 +616,6 @@ func TestConnections(t *testing.T) {
 				headers: map[string][]string{},
 			},
 		},
-		{
-			name: "tcp-tcp proxy without warpRoutingService enabled",
-			args: args{
-				ingressServiceScheme:  "tcp://",
-				originService:         runEchoTCPService,
-				eyeballResponseWriter: newTCPRespWriter(replayer),
-				eyeballRequestBody:    newTCPRequestBody([]byte("test2")),
-				connectionType:        connection.TypeTCP,
-				requestHeaders: map[string][]string{
-					"Cf-Cloudflared-Proxy-Src": {"non-blank-value"},
-				},
-			},
-			want: want{
-				message: []byte{},
-				err:     true,
-			},
-		},
 		{
 			name: "ws-ws proxy when origin is different",
 			args: args{
@@ -670,7 +668,6 @@ func TestConnections(t *testing.T) {
 				originService:         runEchoTCPService,
 				eyeballResponseWriter: newTCPRespWriter(replayer),
 				eyeballRequestBody:    newTCPRequestBody([]byte("rate-limited")),
-				warpRoutingService:    ingress.NewWarpRoutingService(testWarpRouting, time.Duration(0)),
 				connectionType:        connection.TypeTCP,
 				requestHeaders: map[string][]string{
 					"Cf-Cloudflared-Proxy-Src": {"non-blank-value"},
@@ -693,7 +690,7 @@ func TestConnections(t *testing.T) {
 			test.args.originService(t, ln)
 
 			ingressRule := createSingleIngressConfig(t, test.args.ingressServiceScheme+ln.Addr().String())
-			_ = ingressRule.StartOrigins(logger, ctx.Done())
+			_ = ingressRule.StartOrigins(&log, ctx.Done())
 
 			// Mock flow limiter
 			ctrl := gomock.NewController(t)
@@ -702,8 +699,7 @@ func TestConnections(t *testing.T) {
 			flowLimiter.EXPECT().Acquire("tcp").AnyTimes().Return(test.args.flowLimiterResponse)
 			flowLimiter.EXPECT().Release().AnyTimes()
 
-			proxy := NewOriginProxy(ingressRule, testWarpRouting, testTags, flowLimiter, time.Duration(0), logger)
-			proxy.warpRouting = test.args.warpRoutingService
+			proxy := NewOriginProxy(ingressRule, originDialer, testTags, flowLimiter, &log)
 
 			dest := ln.Addr().String()
 			req, err := http.NewRequest(

quic/v3/datagram_test.go

@@ -1,6 +1,7 @@
 package v3_test
 
 import (
+	"crypto/rand"
 	"encoding/binary"
 	"errors"
 	"net/netip"
@@ -14,12 +15,18 @@ import (
 
 func makePayload(size int) []byte {
 	payload := make([]byte, size)
-	for i := range len(payload) {
-		payload[i] = 0xfc
-	}
+	_, _ = rand.Read(payload)
 	return payload
 }
 
+func makePayloads(size int, count int) [][]byte {
+	payloads := make([][]byte, count)
+	for i := range payloads {
+		payloads[i] = makePayload(size)
+	}
+	return payloads
+}
+
 func TestSessionRegistration_MarshalUnmarshal(t *testing.T) {
 	payload := makePayload(1280)
 	tests := []*v3.UDPSessionRegistrationDatagram{

quic/v3/manager.go

@@ -2,12 +2,11 @@ package v3
 
 import (
 	"errors"
-	"net"
-	"net/netip"
 	"sync"
 
 	"github.com/rs/zerolog"
 
+	"github.com/cloudflare/cloudflared/ingress"
 	"github.com/cloudflare/cloudflared/management"
 
 	cfdflow "github.com/cloudflare/cloudflared/flow"
@@ -38,18 +37,16 @@ type SessionManager interface {
 	UnregisterSession(requestID RequestID)
 }
 
-type DialUDP func(dest netip.AddrPort) (*net.UDPConn, error)
-
 type sessionManager struct {
 	sessions     map[RequestID]Session
 	mutex        sync.RWMutex
-	originDialer DialUDP
+	originDialer ingress.OriginUDPDialer
 	limiter      cfdflow.Limiter
 	metrics      Metrics
 	log          *zerolog.Logger
 }
 
-func NewSessionManager(metrics Metrics, log *zerolog.Logger, originDialer DialUDP, limiter cfdflow.Limiter) SessionManager {
+func NewSessionManager(metrics Metrics, log *zerolog.Logger, originDialer ingress.OriginUDPDialer, limiter cfdflow.Limiter) SessionManager {
 	return &sessionManager{
 		sessions:     make(map[RequestID]Session),
 		originDialer: originDialer,
@@ -76,7 +73,7 @@ func (s *sessionManager) RegisterSession(request *UDPSessionRegistrationDatagram
 	}
 
 	// Attempt to bind the UDP socket for the new session
-	origin, err := s.originDialer(request.Dest)
+	origin, err := s.originDialer.DialUDP(request.Dest)
 	if err != nil {
 		return nil, err
 	}

quic/v3/manager_test.go

@@ -11,6 +11,7 @@ import (
 	"github.com/stretchr/testify/require"
 	"go.uber.org/mock/gomock"
 
+	"github.com/cloudflare/cloudflared/config"
 	"github.com/cloudflare/cloudflared/mocks"
 
 	cfdflow "github.com/cloudflare/cloudflared/flow"
@@ -18,9 +19,21 @@ import (
 	v3 "github.com/cloudflare/cloudflared/quic/v3"
 )
 
+var (
+	testDefaultDialer = ingress.NewDialer(ingress.WarpRoutingConfig{
+		ConnectTimeout: config.CustomDuration{Duration: 1 * time.Second},
+		TCPKeepAlive:   config.CustomDuration{Duration: 15 * time.Second},
+		MaxActiveFlows: 0,
+	})
+)
+
 func TestRegisterSession(t *testing.T) {
 	log := zerolog.Nop()
-	manager := v3.NewSessionManager(&noopMetrics{}, &log, ingress.DialUDPAddrPort, cfdflow.NewLimiter(0))
+	originDialerService := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   testDefaultDialer,
+		TCPWriteTimeout: 0,
+	}, &log)
+	manager := v3.NewSessionManager(&noopMetrics{}, &log, originDialerService, cfdflow.NewLimiter(0))
 
 	request := v3.UDPSessionRegistrationDatagram{
 		RequestID:        testRequestID,
@@ -76,7 +89,11 @@ func TestRegisterSession(t *testing.T) {
 
 func TestGetSession_Empty(t *testing.T) {
 	log := zerolog.Nop()
-	manager := v3.NewSessionManager(&noopMetrics{}, &log, ingress.DialUDPAddrPort, cfdflow.NewLimiter(0))
+	originDialerService := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   testDefaultDialer,
+		TCPWriteTimeout: 0,
+	}, &log)
+	manager := v3.NewSessionManager(&noopMetrics{}, &log, originDialerService, cfdflow.NewLimiter(0))
 
 	_, err := manager.GetSession(testRequestID)
 	if !errors.Is(err, v3.ErrSessionNotFound) {
@@ -86,6 +103,10 @@ func TestGetSession_Empty(t *testing.T) {
 
 func TestRegisterSessionRateLimit(t *testing.T) {
 	log := zerolog.Nop()
+	originDialerService := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   testDefaultDialer,
+		TCPWriteTimeout: 0,
+	}, &log)
 	ctrl := gomock.NewController(t)
 
 	flowLimiterMock := mocks.NewMockLimiter(ctrl)
@@ -93,7 +114,7 @@ func TestRegisterSessionRateLimit(t *testing.T) {
 	flowLimiterMock.EXPECT().Acquire("udp").Return(cfdflow.ErrTooManyActiveFlows)
 	flowLimiterMock.EXPECT().Release().Times(0)
 
-	manager := v3.NewSessionManager(&noopMetrics{}, &log, ingress.DialUDPAddrPort, flowLimiterMock)
+	manager := v3.NewSessionManager(&noopMetrics{}, &log, originDialerService, flowLimiterMock)
 
 	request := v3.UDPSessionRegistrationDatagram{
 		RequestID:        testRequestID,

quic/v3/metrics.go

@@ -9,28 +9,59 @@ import (
 )
 
 const (
-	namespace = "cloudflared"
-	subsystem = "udp"
+	namespace      = "cloudflared"
+	subsystem_udp  = "udp"
+	subsystem_icmp = "icmp"
 
 	commandMetricLabel = "command"
+	reasonMetricLabel  = "reason"
 )
 
+type DroppedReason int
+
+const (
+	DroppedWriteFailed DroppedReason = iota
+	DroppedWriteDeadlineExceeded
+	DroppedWriteFull
+	DroppedWriteFlowUnknown
+	DroppedReadFailed
+	// Origin payloads that are too large to proxy.
+	DroppedReadTooLarge
+)
+
+var droppedReason = map[DroppedReason]string{
+	DroppedWriteFailed:           "write_failed",
+	DroppedWriteDeadlineExceeded: "write_deadline_exceeded",
+	DroppedWriteFull:             "write_full",
+	DroppedWriteFlowUnknown:      "write_flow_unknown",
+	DroppedReadFailed:            "read_failed",
+	DroppedReadTooLarge:          "read_too_large",
+}
+
+func (dr DroppedReason) String() string {
+	return droppedReason[dr]
+}
+
 type Metrics interface {
 	IncrementFlows(connIndex uint8)
 	DecrementFlows(connIndex uint8)
-	PayloadTooLarge(connIndex uint8)
+	FailedFlow(connIndex uint8)
 	RetryFlowResponse(connIndex uint8)
 	MigrateFlow(connIndex uint8)
 	UnsupportedRemoteCommand(connIndex uint8, command string)
+	DroppedUDPDatagram(connIndex uint8, reason DroppedReason)
+	DroppedICMPPackets(connIndex uint8, reason DroppedReason)
 }
 
 type metrics struct {
 	activeUDPFlows            *prometheus.GaugeVec
 	totalUDPFlows             *prometheus.CounterVec
-	payloadTooLarge           *prometheus.CounterVec
 	retryFlowResponses        *prometheus.CounterVec
 	migratedFlows             *prometheus.CounterVec
 	unsupportedRemoteCommands *prometheus.CounterVec
+	droppedUDPDatagrams       *prometheus.CounterVec
+	droppedICMPPackets        *prometheus.CounterVec
+	failedFlows               *prometheus.CounterVec
 }
 
 func (m *metrics) IncrementFlows(connIndex uint8) {
@@ -42,8 +73,8 @@ func (m *metrics) DecrementFlows(connIndex uint8) {
 	m.activeUDPFlows.WithLabelValues(fmt.Sprintf("%d", connIndex)).Dec()
 }
 
-func (m *metrics) PayloadTooLarge(connIndex uint8) {
-	m.payloadTooLarge.WithLabelValues(fmt.Sprintf("%d", connIndex)).Inc()
+func (m *metrics) FailedFlow(connIndex uint8) {
+	m.failedFlows.WithLabelValues(fmt.Sprintf("%d", connIndex)).Inc()
 }
 
 func (m *metrics) RetryFlowResponse(connIndex uint8) {
@@ -58,52 +89,74 @@ func (m *metrics) UnsupportedRemoteCommand(connIndex uint8, command string) {
 	m.unsupportedRemoteCommands.WithLabelValues(fmt.Sprintf("%d", connIndex), command).Inc()
 }
 
+func (m *metrics) DroppedUDPDatagram(connIndex uint8, reason DroppedReason) {
+	m.droppedUDPDatagrams.WithLabelValues(fmt.Sprintf("%d", connIndex), reason.String()).Inc()
+}
+
+func (m *metrics) DroppedICMPPackets(connIndex uint8, reason DroppedReason) {
+	m.droppedICMPPackets.WithLabelValues(fmt.Sprintf("%d", connIndex), reason.String()).Inc()
+}
+
 func NewMetrics(registerer prometheus.Registerer) Metrics {
 	m := &metrics{
 		activeUDPFlows: prometheus.NewGaugeVec(prometheus.GaugeOpts{
 			Namespace: namespace,
-			Subsystem: subsystem,
+			Subsystem: subsystem_udp,
 			Name:      "active_flows",
 			Help:      "Concurrent count of UDP flows that are being proxied to any origin",
 		}, []string{quic.ConnectionIndexMetricLabel}),
 		totalUDPFlows: prometheus.NewCounterVec(prometheus.CounterOpts{ //nolint:promlinter
 			Namespace: namespace,
-			Subsystem: subsystem,
+			Subsystem: subsystem_udp,
 			Name:      "total_flows",
 			Help:      "Total count of UDP flows that have been proxied to any origin",
 		}, []string{quic.ConnectionIndexMetricLabel}),
-		payloadTooLarge: prometheus.NewCounterVec(prometheus.CounterOpts{ //nolint:promlinter
+		failedFlows: prometheus.NewCounterVec(prometheus.CounterOpts{ //nolint:promlinter
 			Namespace: namespace,
-			Subsystem: subsystem,
-			Name:      "payload_too_large",
-			Help:      "Total count of UDP flows that have had origin payloads that are too large to proxy",
+			Subsystem: subsystem_udp,
+			Name:      "failed_flows",
+			Help:      "Total count of flows that errored and closed",
 		}, []string{quic.ConnectionIndexMetricLabel}),
 		retryFlowResponses: prometheus.NewCounterVec(prometheus.CounterOpts{ //nolint:promlinter
 			Namespace: namespace,
-			Subsystem: subsystem,
+			Subsystem: subsystem_udp,
 			Name:      "retry_flow_responses",
 			Help:      "Total count of UDP flows that have had to send their registration response more than once",
 		}, []string{quic.ConnectionIndexMetricLabel}),
 		migratedFlows: prometheus.NewCounterVec(prometheus.CounterOpts{ //nolint:promlinter
 			Namespace: namespace,
-			Subsystem: subsystem,
+			Subsystem: subsystem_udp,
 			Name:      "migrated_flows",
 			Help:      "Total count of UDP flows have been migrated across local connections",
 		}, []string{quic.ConnectionIndexMetricLabel}),
-		unsupportedRemoteCommands: prometheus.NewCounterVec(prometheus.CounterOpts{
+		unsupportedRemoteCommands: prometheus.NewCounterVec(prometheus.CounterOpts{ //nolint:promlinter
 			Namespace: namespace,
-			Subsystem: subsystem,
+			Subsystem: subsystem_udp,
 			Name:      "unsupported_remote_command_total",
-			Help:      "Total count of unsupported remote RPC commands for the ",
+			Help:      "Total count of unsupported remote RPC commands called",
 		}, []string{quic.ConnectionIndexMetricLabel, commandMetricLabel}),
+		droppedUDPDatagrams: prometheus.NewCounterVec(prometheus.CounterOpts{ //nolint:promlinter
+			Namespace: namespace,
+			Subsystem: subsystem_udp,
+			Name:      "dropped_datagrams",
+			Help:      "Total count of UDP dropped datagrams",
+		}, []string{quic.ConnectionIndexMetricLabel, reasonMetricLabel}),
+		droppedICMPPackets: prometheus.NewCounterVec(prometheus.CounterOpts{ //nolint:promlinter
+			Namespace: namespace,
+			Subsystem: subsystem_icmp,
+			Name:      "dropped_packets",
+			Help:      "Total count of ICMP dropped datagrams",
+		}, []string{quic.ConnectionIndexMetricLabel, reasonMetricLabel}),
 	}
 	registerer.MustRegister(
 		m.activeUDPFlows,
 		m.totalUDPFlows,
-		m.payloadTooLarge,
+		m.failedFlows,
 		m.retryFlowResponses,
 		m.migratedFlows,
 		m.unsupportedRemoteCommands,
+		m.droppedUDPDatagrams,
+		m.droppedICMPPackets,
 	)
 	return m
 }

quic/v3/metrics_test.go

@@ -1,10 +1,15 @@
 package v3_test
 
+import v3 "github.com/cloudflare/cloudflared/quic/v3"
+
 type noopMetrics struct{}
 
-func (noopMetrics) IncrementFlows(connIndex uint8)                           {}
-func (noopMetrics) DecrementFlows(connIndex uint8)                           {}
-func (noopMetrics) PayloadTooLarge(connIndex uint8)                          {}
-func (noopMetrics) RetryFlowResponse(connIndex uint8)                        {}
-func (noopMetrics) MigrateFlow(connIndex uint8)                              {}
-func (noopMetrics) UnsupportedRemoteCommand(connIndex uint8, command string) {}
+func (noopMetrics) IncrementFlows(connIndex uint8)                              {}
+func (noopMetrics) DecrementFlows(connIndex uint8)                              {}
+func (noopMetrics) FailedFlow(connIndex uint8)                                  {}
+func (noopMetrics) PayloadTooLarge(connIndex uint8)                             {}
+func (noopMetrics) RetryFlowResponse(connIndex uint8)                           {}
+func (noopMetrics) MigrateFlow(connIndex uint8)                                 {}
+func (noopMetrics) UnsupportedRemoteCommand(connIndex uint8, command string)    {}
+func (noopMetrics) DroppedUDPDatagram(connIndex uint8, reason v3.DroppedReason) {}
+func (noopMetrics) DroppedICMPPackets(connIndex uint8, reason v3.DroppedReason) {}

quic/v3/muxer.go

@@ -17,6 +17,9 @@ const (
 	// Allocating a 16 channel buffer here allows for the writer to be slightly faster than the reader.
 	// This has worked previously well for datagramv2, so we will start with this as well
 	demuxChanCapacity = 16
+	// This provides a small buffer for the PacketRouter to poll ICMP packets from the QUIC connection
+	// before writing them to the origin.
+	icmpDatagramChanCapacity = 128
 
 	logSrcKey      = "src"
 	logDstKey      = "dst"
@@ -59,14 +62,15 @@ type QuicConnection interface {
 }
 
 type datagramConn struct {
-	conn           QuicConnection
-	index          uint8
-	sessionManager SessionManager
-	icmpRouter     ingress.ICMPRouter
-	metrics        Metrics
-	logger         *zerolog.Logger
-	datagrams      chan []byte
-	readErrors     chan error
+	conn             QuicConnection
+	index            uint8
+	sessionManager   SessionManager
+	icmpRouter       ingress.ICMPRouter
+	metrics          Metrics
+	logger           *zerolog.Logger
+	datagrams        chan []byte
+	icmpDatagramChan chan *ICMPDatagram
+	readErrors       chan error
 
 	icmpEncoderPool sync.Pool // a pool of *packet.Encoder
 	icmpDecoderPool sync.Pool
@@ -75,14 +79,15 @@ type datagramConn struct {
 func NewDatagramConn(conn QuicConnection, sessionManager SessionManager, icmpRouter ingress.ICMPRouter, index uint8, metrics Metrics, logger *zerolog.Logger) DatagramConn {
 	log := logger.With().Uint8("datagramVersion", 3).Logger()
 	return &datagramConn{
-		conn:           conn,
-		index:          index,
-		sessionManager: sessionManager,
-		icmpRouter:     icmpRouter,
-		metrics:        metrics,
-		logger:         &log,
-		datagrams:      make(chan []byte, demuxChanCapacity),
-		readErrors:     make(chan error, 2),
+		conn:             conn,
+		index:            index,
+		sessionManager:   sessionManager,
+		icmpRouter:       icmpRouter,
+		metrics:          metrics,
+		logger:           &log,
+		datagrams:        make(chan []byte, demuxChanCapacity),
+		icmpDatagramChan: make(chan *ICMPDatagram, icmpDatagramChanCapacity),
+		readErrors:       make(chan error, 2),
 		icmpEncoderPool: sync.Pool{
 			New: func() any {
 				return packet.NewEncoder()
@@ -168,6 +173,9 @@ func (c *datagramConn) Serve(ctx context.Context) error {
 	readCtx, cancel := context.WithCancel(connCtx)
 	defer cancel()
 	go c.pollDatagrams(readCtx)
+	// Processing ICMP datagrams also monitors the reader context since the ICMP datagrams from the reader are the input
+	// for the routine.
+	go c.processICMPDatagrams(readCtx)
 	for {
 		// We make sure to monitor the context of cloudflared and the underlying connection to return if any errors occur.
 		var datagram []byte
@@ -175,64 +183,65 @@ func (c *datagramConn) Serve(ctx context.Context) error {
 		// Monitor the context of cloudflared
 		case <-ctx.Done():
 			return ctx.Err()
-		// Monitor the context of the underlying connection
+		// Monitor the context of the underlying quic connection
 		case <-connCtx.Done():
 			return connCtx.Err()
 		// Monitor for any hard errors from reading the connection
 		case err := <-c.readErrors:
 			return err
-		// Otherwise, wait and dequeue datagrams as they come in
+		// Wait and dequeue datagrams as they come in
 		case d := <-c.datagrams:
 			datagram = d
 		}
 
 		// Each incoming datagram will be processed in a new go routine to handle the demuxing and action associated.
-		go func() {
-			typ, err := ParseDatagramType(datagram)
+		typ, err := ParseDatagramType(datagram)
+		if err != nil {
+			c.logger.Err(err).Msgf("unable to parse datagram type: %d", typ)
+			continue
+		}
+		switch typ {
+		case UDPSessionRegistrationType:
+			reg := &UDPSessionRegistrationDatagram{}
+			err := reg.UnmarshalBinary(datagram)
 			if err != nil {
-				c.logger.Err(err).Msgf("unable to parse datagram type: %d", typ)
-				return
+				c.logger.Err(err).Msgf("unable to unmarshal session registration datagram")
+				continue
 			}
-			switch typ {
-			case UDPSessionRegistrationType:
-				reg := &UDPSessionRegistrationDatagram{}
-				err := reg.UnmarshalBinary(datagram)
-				if err != nil {
-					c.logger.Err(err).Msgf("unable to unmarshal session registration datagram")
-					return
-				}
-				logger := c.logger.With().Str(logFlowID, reg.RequestID.String()).Logger()
-				// We bind the new session to the quic connection context instead of cloudflared context to allow for the
-				// quic connection to close and close only the sessions bound to it. Closing of cloudflared will also
-				// initiate the close of the quic connection, so we don't have to worry about the application context
-				// in the scope of a session.
-				c.handleSessionRegistrationDatagram(connCtx, reg, &logger)
-			case UDPSessionPayloadType:
-				payload := &UDPSessionPayloadDatagram{}
-				err := payload.UnmarshalBinary(datagram)
-				if err != nil {
-					c.logger.Err(err).Msgf("unable to unmarshal session payload datagram")
-					return
-				}
-				logger := c.logger.With().Str(logFlowID, payload.RequestID.String()).Logger()
-				c.handleSessionPayloadDatagram(payload, &logger)
-			case ICMPType:
-				packet := &ICMPDatagram{}
-				err := packet.UnmarshalBinary(datagram)
-				if err != nil {
-					c.logger.Err(err).Msgf("unable to unmarshal icmp datagram")
-					return
-				}
-				c.handleICMPPacket(packet)
-			case UDPSessionRegistrationResponseType:
-				// cloudflared should never expect to receive UDP session responses as it will not initiate new
-				// sessions towards the edge.
-				c.logger.Error().Msgf("unexpected datagram type received: %d", UDPSessionRegistrationResponseType)
-				return
-			default:
-				c.logger.Error().Msgf("unknown datagram type received: %d", typ)
+			logger := c.logger.With().Str(logFlowID, reg.RequestID.String()).Logger()
+			// We bind the new session to the quic connection context instead of cloudflared context to allow for the
+			// quic connection to close and close only the sessions bound to it. Closing of cloudflared will also
+			// initiate the close of the quic connection, so we don't have to worry about the application context
+			// in the scope of a session.
+			//
+			// Additionally, we spin out the registration into a separate go routine to handle the Serve'ing of the
+			// session in a separate routine from the demuxer.
+			go c.handleSessionRegistrationDatagram(connCtx, reg, &logger)
+		case UDPSessionPayloadType:
+			payload := &UDPSessionPayloadDatagram{}
+			err := payload.UnmarshalBinary(datagram)
+			if err != nil {
+				c.logger.Err(err).Msgf("unable to unmarshal session payload datagram")
+				continue
 			}
-		}()
+			logger := c.logger.With().Str(logFlowID, payload.RequestID.String()).Logger()
+			c.handleSessionPayloadDatagram(payload, &logger)
+		case ICMPType:
+			packet := &ICMPDatagram{}
+			err := packet.UnmarshalBinary(datagram)
+			if err != nil {
+				c.logger.Err(err).Msgf("unable to unmarshal icmp datagram")
+				continue
+			}
+			c.handleICMPPacket(packet)
+		case UDPSessionRegistrationResponseType:
+			// cloudflared should never expect to receive UDP session responses as it will not initiate new
+			// sessions towards the edge.
+			c.logger.Error().Msgf("unexpected datagram type received: %d", UDPSessionRegistrationResponseType)
+			continue
+		default:
+			c.logger.Error().Msgf("unknown datagram type received: %d", typ)
+		}
 	}
 }
 
@@ -243,24 +252,21 @@ func (c *datagramConn) handleSessionRegistrationDatagram(ctx context.Context, da
 		Str(logDstKey, datagram.Dest.String()).
 		Logger()
 	session, err := c.sessionManager.RegisterSession(datagram, c)
-	switch err {
-	case nil:
-		// Continue as normal
-	case ErrSessionAlreadyRegistered:
-		// Session is already registered and likely the response got lost
-		c.handleSessionAlreadyRegistered(datagram.RequestID, &log)
-		return
-	case ErrSessionBoundToOtherConn:
-		// Session is already registered but to a different connection
-		c.handleSessionMigration(datagram.RequestID, &log)
-		return
-	case ErrSessionRegistrationRateLimited:
-		// There are too many concurrent sessions so we return an error to force a retry later
-		c.handleSessionRegistrationRateLimited(datagram, &log)
-		return
-	default:
-		log.Err(err).Msg("flow registration failure")
-		c.handleSessionRegistrationFailure(datagram.RequestID, &log)
+	if err != nil {
+		switch err {
+		case ErrSessionAlreadyRegistered:
+			// Session is already registered and likely the response got lost
+			c.handleSessionAlreadyRegistered(datagram.RequestID, &log)
+		case ErrSessionBoundToOtherConn:
+			// Session is already registered but to a different connection
+			c.handleSessionMigration(datagram.RequestID, &log)
+		case ErrSessionRegistrationRateLimited:
+			// There are too many concurrent sessions so we return an error to force a retry later
+			c.handleSessionRegistrationRateLimited(datagram, &log)
+		default:
+			log.Err(err).Msg("flow registration failure")
+			c.handleSessionRegistrationFailure(datagram.RequestID, &log)
+		}
 		return
 	}
 	log = log.With().Str(logSrcKey, session.LocalAddr().String()).Logger()
@@ -362,30 +368,54 @@ func (c *datagramConn) handleSessionRegistrationRateLimited(datagram *UDPSession
 func (c *datagramConn) handleSessionPayloadDatagram(datagram *UDPSessionPayloadDatagram, logger *zerolog.Logger) {
 	s, err := c.sessionManager.GetSession(datagram.RequestID)
 	if err != nil {
+		c.metrics.DroppedUDPDatagram(c.index, DroppedWriteFlowUnknown)
 		logger.Err(err).Msgf("unable to find flow")
 		return
 	}
-	// We ignore the bytes written to the socket because any partial write must return an error.
-	_, err = s.Write(datagram.Payload)
-	if err != nil {
-		logger.Err(err).Msgf("unable to write payload for the flow")
-		return
-	}
+	s.Write(datagram.Payload)
 }
 
-// Handles incoming ICMP datagrams.
+// Handles incoming ICMP datagrams into a serialized channel to be handled by a single consumer.
 func (c *datagramConn) handleICMPPacket(datagram *ICMPDatagram) {
 	if c.icmpRouter == nil {
 		// ICMPRouter is disabled so we drop the current packet and ignore all incoming ICMP packets
 		return
 	}
+	select {
+	case c.icmpDatagramChan <- datagram:
+	default:
+		// If the ICMP datagram channel is full, drop any additional incoming.
+		c.metrics.DroppedICMPPackets(c.index, DroppedWriteFull)
+		c.logger.Warn().Msg("failed to write icmp packet to origin: dropped")
+	}
+}
 
+// Consumes from the ICMP datagram channel to write out the ICMP requests to an origin.
+func (c *datagramConn) processICMPDatagrams(ctx context.Context) {
+	if c.icmpRouter == nil {
+		// ICMPRouter is disabled so we ignore all incoming ICMP packets
+		return
+	}
+
+	for {
+		select {
+		// If the provided context is closed we want to exit the write loop
+		case <-ctx.Done():
+			return
+		case datagram := <-c.icmpDatagramChan:
+			c.writeICMPPacket(datagram)
+		}
+	}
+}
+
+func (c *datagramConn) writeICMPPacket(datagram *ICMPDatagram) {
 	// Decode the provided ICMPDatagram as an ICMP packet
 	rawPacket := packet.RawPacket{Data: datagram.Payload}
 	cachedDecoder := c.icmpDecoderPool.Get()
 	defer c.icmpDecoderPool.Put(cachedDecoder)
 	decoder, ok := cachedDecoder.(*packet.ICMPDecoder)
 	if !ok {
+		c.metrics.DroppedICMPPackets(c.index, DroppedWriteFailed)
 		c.logger.Error().Msg("Could not get ICMPDecoder from the pool. Dropping packet")
 		return
 	}
@@ -393,6 +423,7 @@ func (c *datagramConn) handleICMPPacket(datagram *ICMPDatagram) {
 	icmp, err := decoder.Decode(rawPacket)
 
 	if err != nil {
+		c.metrics.DroppedICMPPackets(c.index, DroppedWriteFailed)
 		c.logger.Err(err).Msgf("unable to marshal icmp packet")
 		return
 	}
@@ -400,6 +431,7 @@ func (c *datagramConn) handleICMPPacket(datagram *ICMPDatagram) {
 	// If the ICMP packet's TTL is expired, we won't send it to the origin and immediately return a TTL Exceeded Message
 	if icmp.TTL <= 1 {
 		if err := c.SendICMPTTLExceed(icmp, rawPacket); err != nil {
+			c.metrics.DroppedICMPPackets(c.index, DroppedWriteFailed)
 			c.logger.Err(err).Msg("failed to return ICMP TTL exceed error")
 		}
 		return
@@ -411,6 +443,7 @@ func (c *datagramConn) handleICMPPacket(datagram *ICMPDatagram) {
 	// connection context which will have no tracing information available.
 	err = c.icmpRouter.Request(c.conn.Context(), icmp, newPacketResponder(c, c.index))
 	if err != nil {
+		c.metrics.DroppedICMPPackets(c.index, DroppedWriteFailed)
 		c.logger.Err(err).
 			Str(logSrcKey, icmp.Src.String()).
 			Str(logDstKey, icmp.Dst.String()).

quic/v3/muxer_test.go

@@ -13,6 +13,7 @@ import (
 	"testing"
 	"time"
 
+	"github.com/fortytw2/leaktest"
 	"github.com/google/gopacket/layers"
 	"github.com/rs/zerolog"
 	"github.com/stretchr/testify/assert"
@@ -88,7 +89,11 @@ func (m *mockEyeball) SendICMPTTLExceed(icmp *packet.ICMP, rawPacket packet.RawP
 
 func TestDatagramConn_New(t *testing.T) {
 	log := zerolog.Nop()
-	conn := v3.NewDatagramConn(newMockQuicConn(), v3.NewSessionManager(&noopMetrics{}, &log, ingress.DialUDPAddrPort, cfdflow.NewLimiter(0)), &noopICMPRouter{}, 0, &noopMetrics{}, &log)
+	originDialerService := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   testDefaultDialer,
+		TCPWriteTimeout: 0,
+	}, &log)
+	conn := v3.NewDatagramConn(newMockQuicConn(t.Context()), v3.NewSessionManager(&noopMetrics{}, &log, originDialerService, cfdflow.NewLimiter(0)), &noopICMPRouter{}, 0, &noopMetrics{}, &log)
 	if conn == nil {
 		t.Fatal("expected valid connection")
 	}
@@ -96,8 +101,14 @@ func TestDatagramConn_New(t *testing.T) {
 
 func TestDatagramConn_SendUDPSessionDatagram(t *testing.T) {
 	log := zerolog.Nop()
-	quic := newMockQuicConn()
-	conn := v3.NewDatagramConn(quic, v3.NewSessionManager(&noopMetrics{}, &log, ingress.DialUDPAddrPort, cfdflow.NewLimiter(0)), &noopICMPRouter{}, 0, &noopMetrics{}, &log)
+	originDialerService := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   testDefaultDialer,
+		TCPWriteTimeout: 0,
+	}, &log)
+	connCtx, connCancel := context.WithCancelCause(t.Context())
+	defer connCancel(context.Canceled)
+	quic := newMockQuicConn(connCtx)
+	conn := v3.NewDatagramConn(quic, v3.NewSessionManager(&noopMetrics{}, &log, originDialerService, cfdflow.NewLimiter(0)), &noopICMPRouter{}, 0, &noopMetrics{}, &log)
 
 	payload := []byte{0xef, 0xef}
 	err := conn.SendUDPSessionDatagram(payload)
@@ -111,8 +122,14 @@ func TestDatagramConn_SendUDPSessionDatagram(t *testing.T) {
 
 func TestDatagramConn_SendUDPSessionResponse(t *testing.T) {
 	log := zerolog.Nop()
-	quic := newMockQuicConn()
-	conn := v3.NewDatagramConn(quic, v3.NewSessionManager(&noopMetrics{}, &log, ingress.DialUDPAddrPort, cfdflow.NewLimiter(0)), &noopICMPRouter{}, 0, &noopMetrics{}, &log)
+	originDialerService := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   testDefaultDialer,
+		TCPWriteTimeout: 0,
+	}, &log)
+	connCtx, connCancel := context.WithCancelCause(t.Context())
+	defer connCancel(context.Canceled)
+	quic := newMockQuicConn(connCtx)
+	conn := v3.NewDatagramConn(quic, v3.NewSessionManager(&noopMetrics{}, &log, originDialerService, cfdflow.NewLimiter(0)), &noopICMPRouter{}, 0, &noopMetrics{}, &log)
 
 	err := conn.SendUDPSessionResponse(testRequestID, v3.ResponseDestinationUnreachable)
 	require.NoError(t, err)
@@ -133,8 +150,14 @@ func TestDatagramConn_SendUDPSessionResponse(t *testing.T) {
 
 func TestDatagramConnServe_ApplicationClosed(t *testing.T) {
 	log := zerolog.Nop()
-	quic := newMockQuicConn()
-	conn := v3.NewDatagramConn(quic, v3.NewSessionManager(&noopMetrics{}, &log, ingress.DialUDPAddrPort, cfdflow.NewLimiter(0)), &noopICMPRouter{}, 0, &noopMetrics{}, &log)
+	originDialerService := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   testDefaultDialer,
+		TCPWriteTimeout: 0,
+	}, &log)
+	connCtx, connCancel := context.WithCancelCause(t.Context())
+	defer connCancel(context.Canceled)
+	quic := newMockQuicConn(connCtx)
+	conn := v3.NewDatagramConn(quic, v3.NewSessionManager(&noopMetrics{}, &log, originDialerService, cfdflow.NewLimiter(0)), &noopICMPRouter{}, 0, &noopMetrics{}, &log)
 
 	ctx, cancel := context.WithTimeout(t.Context(), 1*time.Second)
 	defer cancel()
@@ -146,11 +169,17 @@ func TestDatagramConnServe_ApplicationClosed(t *testing.T) {
 
 func TestDatagramConnServe_ConnectionClosed(t *testing.T) {
 	log := zerolog.Nop()
-	quic := newMockQuicConn()
+	originDialerService := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   testDefaultDialer,
+		TCPWriteTimeout: 0,
+	}, &log)
+	connCtx, connCancel := context.WithCancelCause(t.Context())
+	defer connCancel(context.Canceled)
+	quic := newMockQuicConn(connCtx)
 	ctx, cancel := context.WithTimeout(t.Context(), 1*time.Second)
 	defer cancel()
 	quic.ctx = ctx
-	conn := v3.NewDatagramConn(quic, v3.NewSessionManager(&noopMetrics{}, &log, ingress.DialUDPAddrPort, cfdflow.NewLimiter(0)), &noopICMPRouter{}, 0, &noopMetrics{}, &log)
+	conn := v3.NewDatagramConn(quic, v3.NewSessionManager(&noopMetrics{}, &log, originDialerService, cfdflow.NewLimiter(0)), &noopICMPRouter{}, 0, &noopMetrics{}, &log)
 
 	err := conn.Serve(t.Context())
 	if !errors.Is(err, context.DeadlineExceeded) {
@@ -160,8 +189,12 @@ func TestDatagramConnServe_ConnectionClosed(t *testing.T) {
 
 func TestDatagramConnServe_ReceiveDatagramError(t *testing.T) {
 	log := zerolog.Nop()
+	originDialerService := ingress.NewOriginDialer(ingress.OriginConfig{
+		DefaultDialer:   testDefaultDialer,
+		TCPWriteTimeout: 0,
+	}, &log)
 	quic := &mockQuicConnReadError{err: net.ErrClosed}
-	conn := v3.NewDatagramConn(quic, v3.NewSessionManager(&noopMetrics{}, &log, ingress.DialUDPAddrPort, cfdflow.NewLimiter(0)), &noopICMPRouter{}, 0, &noopMetrics{}, &log)
+	conn := v3.NewDatagramConn(quic, v3.NewSessionManager(&noopMetrics{}, &log, originDialerService, cfdflow.NewLimiter(0)), &noopICMPRouter{}, 0, &noopMetrics{}, &log)
 
 	err := conn.Serve(t.Context())
 	if !errors.Is(err, net.ErrClosed) {
@@ -171,15 +204,17 @@ func TestDatagramConnServe_ReceiveDatagramError(t *testing.T) {
 
 func TestDatagramConnServe_SessionRegistrationRateLimit(t *testing.T) {
 	log := zerolog.Nop()
-	quic := newMockQuicConn()
+	connCtx, connCancel := context.WithCancelCause(t.Context())
+	defer connCancel(context.Canceled)
+	quic := newMockQuicConn(connCtx)
 	sessionManager := &mockSessionManager{
 		expectedRegErr: v3.ErrSessionRegistrationRateLimited,
 	}
 	conn := v3.NewDatagramConn(quic, sessionManager, &noopICMPRouter{}, 0, &noopMetrics{}, &log)
 
 	// Setup the muxer
-	ctx, cancel := context.WithCancel(t.Context())
-	defer cancel()
+	ctx, cancel := context.WithCancelCause(t.Context())
+	defer cancel(context.Canceled)
 	done := make(chan error, 1)
 	go func() {
 		done <- conn.Serve(ctx)
@@ -199,9 +234,12 @@ func TestDatagramConnServe_SessionRegistrationRateLimit(t *testing.T) {
 
 	require.EqualValues(t, testRequestID, resp.RequestID)
 	require.EqualValues(t, v3.ResponseTooManyActiveFlows, resp.ResponseType)
+
+	assertContextClosed(t, ctx, done, cancel)
 }
 
 func TestDatagramConnServe_ErrorDatagramTypes(t *testing.T) {
+	defer leaktest.Check(t)()
 	for _, test := range []struct {
 		name     string
 		input    []byte
@@ -226,7 +264,9 @@ func TestDatagramConnServe_ErrorDatagramTypes(t *testing.T) {
 		t.Run(test.name, func(t *testing.T) {
 			logOutput := new(LockedBuffer)
 			log := zerolog.New(logOutput)
-			quic := newMockQuicConn()
+			connCtx, connCancel := context.WithCancelCause(t.Context())
+			defer connCancel(context.Canceled)
+			quic := newMockQuicConn(connCtx)
 			quic.send <- test.input
 			conn := v3.NewDatagramConn(quic, &mockSessionManager{}, &noopICMPRouter{}, 0, &noopMetrics{}, &log)
 
@@ -265,8 +305,11 @@ func (b *LockedBuffer) String() string {
 }
 
 func TestDatagramConnServe_RegisterSession_SessionManagerError(t *testing.T) {
+	defer leaktest.Check(t)()
 	log := zerolog.Nop()
-	quic := newMockQuicConn()
+	connCtx, connCancel := context.WithCancelCause(t.Context())
+	defer connCancel(context.Canceled)
+	quic := newMockQuicConn(connCtx)
 	expectedErr := errors.New("unable to register session")
 	sessionManager := mockSessionManager{expectedRegErr: expectedErr}
 	conn := v3.NewDatagramConn(quic, &sessionManager, &noopICMPRouter{}, 0, &noopMetrics{}, &log)
@@ -300,8 +343,11 @@ func TestDatagramConnServe_RegisterSession_SessionManagerError(t *testing.T) {
 }
 
 func TestDatagramConnServe(t *testing.T) {
+	defer leaktest.Check(t)()
 	log := zerolog.Nop()
-	quic := newMockQuicConn()
+	connCtx, connCancel := context.WithCancelCause(t.Context())
+	defer connCancel(context.Canceled)
+	quic := newMockQuicConn(connCtx)
 	session := newMockSession()
 	sessionManager := mockSessionManager{session: &session}
 	conn := v3.NewDatagramConn(quic, &sessionManager, &noopICMPRouter{}, 0, &noopMetrics{}, &log)
@@ -348,8 +394,11 @@ func TestDatagramConnServe(t *testing.T) {
 // instances causes inteference resulting in multiple different raw packets being decoded
 // as the same decoded packet.
 func TestDatagramConnServeDecodeMultipleICMPInParallel(t *testing.T) {
+	defer leaktest.Check(t)()
 	log := zerolog.Nop()
-	quic := newMockQuicConn()
+	connCtx, connCancel := context.WithCancelCause(t.Context())
+	defer connCancel(context.Canceled)
+	quic := newMockQuicConn(connCtx)
 	session := newMockSession()
 	sessionManager := mockSessionManager{session: &session}
 	router := newMockICMPRouter()
@@ -389,10 +438,14 @@ func TestDatagramConnServeDecodeMultipleICMPInParallel(t *testing.T) {
 	wg := sync.WaitGroup{}
 	var receivedPackets []*packet.ICMP
 	go func() {
-		for ctx.Err() == nil {
-			icmpPacket := <-router.recv
-			receivedPackets = append(receivedPackets, icmpPacket)
-			wg.Done()
+		for {
+			select {
+			case <-ctx.Done():
+				return
+			case icmpPacket := <-router.recv:
+				receivedPackets = append(receivedPackets, icmpPacket)
+				wg.Done()
+			}
 		}
 	}()
 
@@ -428,8 +481,11 @@ func TestDatagramConnServeDecodeMultipleICMPInParallel(t *testing.T) {
 }
 
 func TestDatagramConnServe_RegisterTwice(t *testing.T) {
+	defer leaktest.Check(t)()
 	log := zerolog.Nop()
-	quic := newMockQuicConn()
+	connCtx, connCancel := context.WithCancelCause(t.Context())
+	defer connCancel(context.Canceled)
+	quic := newMockQuicConn(connCtx)
 	session := newMockSession()
 	sessionManager := mockSessionManager{session: &session}
 	conn := v3.NewDatagramConn(quic, &sessionManager, &noopICMPRouter{}, 0, &noopMetrics{}, &log)
@@ -490,12 +546,17 @@ func TestDatagramConnServe_RegisterTwice(t *testing.T) {
 }
 
 func TestDatagramConnServe_MigrateConnection(t *testing.T) {
+	defer leaktest.Check(t)()
 	log := zerolog.Nop()
-	quic := newMockQuicConn()
+	connCtx, connCancel := context.WithCancelCause(t.Context())
+	defer connCancel(context.Canceled)
+	quic := newMockQuicConn(connCtx)
 	session := newMockSession()
 	sessionManager := mockSessionManager{session: &session}
 	conn := v3.NewDatagramConn(quic, &sessionManager, &noopICMPRouter{}, 0, &noopMetrics{}, &log)
-	quic2 := newMockQuicConn()
+	conn2Ctx, conn2Cancel := context.WithCancelCause(t.Context())
+	defer conn2Cancel(context.Canceled)
+	quic2 := newMockQuicConn(conn2Ctx)
 	conn2 := v3.NewDatagramConn(quic2, &sessionManager, &noopICMPRouter{}, 1, &noopMetrics{}, &log)
 
 	// Setup the muxer
@@ -573,8 +634,11 @@ func TestDatagramConnServe_MigrateConnection(t *testing.T) {
 }
 
 func TestDatagramConnServe_Payload_GetSessionError(t *testing.T) {
+	defer leaktest.Check(t)()
 	log := zerolog.Nop()
-	quic := newMockQuicConn()
+	connCtx, connCancel := context.WithCancelCause(t.Context())
+	defer connCancel(context.Canceled)
+	quic := newMockQuicConn(connCtx)
 	// mockSessionManager will return the ErrSessionNotFound for any session attempting to be queried by the muxer
 	sessionManager := mockSessionManager{session: nil, expectedGetErr: v3.ErrSessionNotFound}
 	conn := v3.NewDatagramConn(quic, &sessionManager, &noopICMPRouter{}, 0, &noopMetrics{}, &log)
@@ -600,9 +664,12 @@ func TestDatagramConnServe_Payload_GetSessionError(t *testing.T) {
 	assertContextClosed(t, ctx, done, cancel)
 }
 
-func TestDatagramConnServe_Payload(t *testing.T) {
+func TestDatagramConnServe_Payloads(t *testing.T) {
+	defer leaktest.Check(t)()
 	log := zerolog.Nop()
-	quic := newMockQuicConn()
+	connCtx, connCancel := context.WithCancelCause(t.Context())
+	defer connCancel(context.Canceled)
+	quic := newMockQuicConn(connCtx)
 	session := newMockSession()
 	sessionManager := mockSessionManager{session: &session}
 	conn := v3.NewDatagramConn(quic, &sessionManager, &noopICMPRouter{}, 0, &noopMetrics{}, &log)
@@ -615,15 +682,26 @@ func TestDatagramConnServe_Payload(t *testing.T) {
 		done <- conn.Serve(ctx)
 	}()
 
-	// Send new session registration
-	expectedPayload := []byte{0xef, 0xef}
-	datagram := newSessionPayloadDatagram(testRequestID, expectedPayload)
-	quic.send <- datagram
+	// Send session payloads
+	expectedPayloads := makePayloads(256, 16)
+	go func() {
+		for _, payload := range expectedPayloads {
+			datagram := newSessionPayloadDatagram(testRequestID, payload)
+			quic.send <- datagram
+		}
+	}()
 
-	// Session should receive the payload
-	payload := <-session.recv
-	if !slices.Equal(expectedPayload, payload) {
-		t.Fatalf("expected session receieve the payload sent via the muxer")
+	// Session should receive the payloads (in-order)
+	for i, payload := range expectedPayloads {
+		select {
+		case recv := <-session.recv:
+			if !slices.Equal(recv, payload) {
+				t.Fatalf("expected session receieve the payload[%d] sent via the muxer: (%x) (%x)", i, recv[:16], payload[:16])
+			}
+		case err := <-ctx.Done():
+			// we expect the payload to return before the context to cancel on the session
+			t.Fatal(err)
+		}
 	}
 
 	// Cancel the muxer Serve context and make sure it closes with the expected error
@@ -631,8 +709,11 @@ func TestDatagramConnServe_Payload(t *testing.T) {
 }
 
 func TestDatagramConnServe_ICMPDatagram_TTLDecremented(t *testing.T) {
+	defer leaktest.Check(t)()
 	log := zerolog.Nop()
-	quic := newMockQuicConn()
+	connCtx, connCancel := context.WithCancelCause(t.Context())
+	defer connCancel(context.Canceled)
+	quic := newMockQuicConn(connCtx)
 	router := newMockICMPRouter()
 	conn := v3.NewDatagramConn(quic, &mockSessionManager{}, router, 0, &noopMetrics{}, &log)
 
@@ -677,8 +758,11 @@ func TestDatagramConnServe_ICMPDatagram_TTLDecremented(t *testing.T) {
 }
 
 func TestDatagramConnServe_ICMPDatagram_TTLExceeded(t *testing.T) {
+	defer leaktest.Check(t)()
 	log := zerolog.Nop()
-	quic := newMockQuicConn()
+	connCtx, connCancel := context.WithCancelCause(t.Context())
+	defer connCancel(context.Canceled)
+	quic := newMockQuicConn(connCtx)
 	router := newMockICMPRouter()
 	conn := v3.NewDatagramConn(quic, &mockSessionManager{}, router, 0, &noopMetrics{}, &log)
 
@@ -797,9 +881,9 @@ type mockQuicConn struct {
 	recv chan []byte
 }
 
-func newMockQuicConn() *mockQuicConn {
+func newMockQuicConn(ctx context.Context) *mockQuicConn {
 	return &mockQuicConn{
-		ctx:  context.Background(),
+		ctx:  ctx,
 		send: make(chan []byte, 1),
 		recv: make(chan []byte, 1),
 	}
@@ -817,7 +901,12 @@ func (m *mockQuicConn) SendDatagram(payload []byte) error {
 }
 
 func (m *mockQuicConn) ReceiveDatagram(_ context.Context) ([]byte, error) {
-	return <-m.send, nil
+	select {
+	case <-m.ctx.Done():
+		return nil, m.ctx.Err()
+	case b := <-m.send:
+		return b, nil
+	}
 }
 
 type mockQuicConnReadError struct {
@@ -881,11 +970,10 @@ func (m *mockSession) Serve(ctx context.Context) error {
 	return v3.SessionCloseErr
 }
 
-func (m *mockSession) Write(payload []byte) (n int, err error) {
+func (m *mockSession) Write(payload []byte) {
 	b := make([]byte, len(payload))
 	copy(b, payload)
 	m.recv <- b
-	return len(b), nil
 }
 
 func (m *mockSession) Close() error {

quic/v3/session.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"io"
 	"net"
+	"os"
 	"sync"
 	"sync/atomic"
 	"time"
@@ -22,6 +23,11 @@ const (
 	// this value (maxDatagramPayloadLen).
 	maxOriginUDPPacketSize = 1500
 
+	// The maximum amount of datagrams a session will queue up before it begins dropping datagrams.
+	// This channel buffer is small because we assume that the dedicated writer to the origin is typically
+	// fast enought to keep the channel empty.
+	writeChanCapacity = 512
+
 	logFlowID        = "flowID"
 	logPacketSizeKey = "packetSize"
 )
@@ -49,7 +55,7 @@ func newSessionIdleErr(timeout time.Duration) error {
 }
 
 type Session interface {
-	io.WriteCloser
+	io.Closer
 	ID() RequestID
 	ConnectionID() uint8
 	RemoteAddr() net.Addr
@@ -58,6 +64,7 @@ type Session interface {
 	Migrate(eyeball DatagramConn, ctx context.Context, logger *zerolog.Logger)
 	// Serve starts the event loop for processing UDP packets
 	Serve(ctx context.Context) error
+	Write(payload []byte)
 }
 
 type session struct {
@@ -67,12 +74,18 @@ type session struct {
 	originAddr     net.Addr
 	localAddr      net.Addr
 	eyeball        atomic.Pointer[DatagramConn]
+	writeChan      chan []byte
 	// activeAtChan is used to communicate the last read/write time
 	activeAtChan chan time.Time
-	closeChan    chan error
-	contextChan  chan context.Context
-	metrics      Metrics
-	log          *zerolog.Logger
+	errChan      chan error
+	// The close channel signal only exists for the write loop because the read loop is always waiting on a read
+	// from the UDP socket to the origin. To close the read loop we close the socket.
+	// Additionally, we can't close the writeChan to indicate that writes are complete because the producer (edge)
+	// side may still be trying to write to this session.
+	closeWrite  chan struct{}
+	contextChan chan context.Context
+	metrics     Metrics
+	log         *zerolog.Logger
 
 	// A special close function that we wrap with sync.Once to make sure it is only called once
 	closeFn func() error
@@ -89,10 +102,12 @@ func NewSession(
 	log *zerolog.Logger,
 ) Session {
 	logger := log.With().Str(logFlowID, id.String()).Logger()
-	// closeChan has two slots to allow for both writers (the closeFn and the Serve routine) to both be able to
-	// write to the channel without blocking since there is only ever one value read from the closeChan by the
+	writeChan := make(chan []byte, writeChanCapacity)
+	// errChan has three slots to allow for all writers (the closeFn, the read loop and the write loop) to
+	// write to the channel without blocking since there is only ever one value read from the errChan by the
 	// waitForCloseCondition.
-	closeChan := make(chan error, 2)
+	errChan := make(chan error, 3)
+	closeWrite := make(chan struct{})
 	session := &session{
 		id:             id,
 		closeAfterIdle: closeAfterIdle,
@@ -100,10 +115,12 @@ func NewSession(
 		originAddr:     originAddr,
 		localAddr:      localAddr,
 		eyeball:        atomic.Pointer[DatagramConn]{},
+		writeChan:      writeChan,
 		// activeAtChan has low capacity. It can be full when there are many concurrent read/write. markActive() will
 		// drop instead of blocking because last active time only needs to be an approximation
 		activeAtChan: make(chan time.Time, 1),
-		closeChan:    closeChan,
+		errChan:      errChan,
+		closeWrite:   closeWrite,
 		// contextChan is an unbounded channel to help enforce one active migration of a session at a time.
 		contextChan: make(chan context.Context),
 		metrics:     metrics,
@@ -111,9 +128,12 @@ func NewSession(
 		closeFn: sync.OnceValue(func() error {
 			// We don't want to block on sending to the close channel if it is already full
 			select {
-			case closeChan <- SessionCloseErr:
+			case errChan <- SessionCloseErr:
 			default:
 			}
+			// Indicate to the write loop that the session is now closed
+			close(closeWrite)
+			// Close the socket directly to unblock the read loop and cause it to also end
 			return origin.Close()
 		}),
 	}
@@ -154,66 +174,115 @@ func (s *session) Migrate(eyeball DatagramConn, ctx context.Context, logger *zer
 }
 
 func (s *session) Serve(ctx context.Context) error {
-	go func() {
-		// QUIC implementation copies data to another buffer before returning https://github.com/quic-go/quic-go/blob/v0.24.0/session.go#L1967-L1975
-		// This makes it safe to share readBuffer between iterations
-		readBuffer := [maxOriginUDPPacketSize + DatagramPayloadHeaderLen]byte{}
-		// To perform a zero copy write when passing the datagram to the connection, we prepare the buffer with
-		// the required datagram header information. We can reuse this buffer for this session since the header is the
-		// same for the each read.
-		_ = MarshalPayloadHeaderTo(s.id, readBuffer[:DatagramPayloadHeaderLen])
-		for {
-			// Read from the origin UDP socket
-			n, err := s.origin.Read(readBuffer[DatagramPayloadHeaderLen:])
-			if err != nil {
-				if errors.Is(err, io.EOF) ||
-					errors.Is(err, io.ErrUnexpectedEOF) {
-					s.log.Debug().Msgf("flow (origin) connection closed: %v", err)
-				}
-				s.closeChan <- err
-				return
-			}
-			if n < 0 {
-				s.log.Warn().Int(logPacketSizeKey, n).Msg("flow (origin) packet read was negative and was dropped")
-				continue
-			}
-			if n > maxDatagramPayloadLen {
-				connectionIndex := s.ConnectionID()
-				s.metrics.PayloadTooLarge(connectionIndex)
-				s.log.Error().Int(logPacketSizeKey, n).Msg("flow (origin) packet read was too large and was dropped")
-				continue
-			}
-			// We need to synchronize on the eyeball in-case that the connection was migrated. This should be rarely a point
-			// of lock contention, as a migration can only happen during startup of a session before traffic flow.
-			eyeball := *(s.eyeball.Load())
-			// Sending a packet to the session does block on the [quic.Connection], however, this is okay because it
-			// will cause back-pressure to the kernel buffer if the writes are not fast enough to the edge.
-			err = eyeball.SendUDPSessionDatagram(readBuffer[:DatagramPayloadHeaderLen+n])
-			if err != nil {
-				s.closeChan <- err
-				return
-			}
-			// Mark the session as active since we proxied a valid packet from the origin.
-			s.markActive()
-		}
-	}()
+	go s.writeLoop()
+	go s.readLoop()
 	return s.waitForCloseCondition(ctx, s.closeAfterIdle)
 }
 
-func (s *session) Write(payload []byte) (n int, err error) {
-	n, err = s.origin.Write(payload)
-	if err != nil {
-		s.log.Err(err).Msg("failed to write payload to flow (remote)")
-		return n, err
+// Read datagrams from the origin and write them to the connection.
+func (s *session) readLoop() {
+	// QUIC implementation copies data to another buffer before returning https://github.com/quic-go/quic-go/blob/v0.24.0/session.go#L1967-L1975
+	// This makes it safe to share readBuffer between iterations
+	readBuffer := [maxOriginUDPPacketSize + DatagramPayloadHeaderLen]byte{}
+	// To perform a zero copy write when passing the datagram to the connection, we prepare the buffer with
+	// the required datagram header information. We can reuse this buffer for this session since the header is the
+	// same for the each read.
+	_ = MarshalPayloadHeaderTo(s.id, readBuffer[:DatagramPayloadHeaderLen])
+	for {
+		// Read from the origin UDP socket
+		n, err := s.origin.Read(readBuffer[DatagramPayloadHeaderLen:])
+		if err != nil {
+			if isConnectionClosed(err) {
+				s.log.Debug().Msgf("flow (read) connection closed: %v", err)
+			}
+			s.closeSession(err)
+			return
+		}
+		if n < 0 {
+			s.metrics.DroppedUDPDatagram(s.ConnectionID(), DroppedReadFailed)
+			s.log.Warn().Int(logPacketSizeKey, n).Msg("flow (origin) packet read was negative and was dropped")
+			continue
+		}
+		if n > maxDatagramPayloadLen {
+			s.metrics.DroppedUDPDatagram(s.ConnectionID(), DroppedReadTooLarge)
+			s.log.Error().Int(logPacketSizeKey, n).Msg("flow (origin) packet read was too large and was dropped")
+			continue
+		}
+		// We need to synchronize on the eyeball in-case that the connection was migrated. This should be rarely a point
+		// of lock contention, as a migration can only happen during startup of a session before traffic flow.
+		eyeball := *(s.eyeball.Load())
+		// Sending a packet to the session does block on the [quic.Connection], however, this is okay because it
+		// will cause back-pressure to the kernel buffer if the writes are not fast enough to the edge.
+		err = eyeball.SendUDPSessionDatagram(readBuffer[:DatagramPayloadHeaderLen+n])
+		if err != nil {
+			s.closeSession(err)
+			return
+		}
+		// Mark the session as active since we proxied a valid packet from the origin.
+		s.markActive()
 	}
-	// Write must return a non-nil error if it returns n < len(p). https://pkg.go.dev/io#Writer
-	if n < len(payload) {
-		s.log.Err(io.ErrShortWrite).Msg("failed to write the full payload to flow (remote)")
-		return n, io.ErrShortWrite
+}
+
+func (s *session) Write(payload []byte) {
+	select {
+	case s.writeChan <- payload:
+	default:
+		s.metrics.DroppedUDPDatagram(s.ConnectionID(), DroppedWriteFull)
+		s.log.Error().Msg("failed to write flow payload to origin: dropped")
+	}
+}
+
+// Read datagrams from the write channel to the origin.
+func (s *session) writeLoop() {
+	for {
+		select {
+		case <-s.closeWrite:
+			// When the closeWrite channel is closed, we will no longer write to the origin and end this
+			// goroutine since the session is now closed.
+			return
+		case payload := <-s.writeChan:
+			n, err := s.origin.Write(payload)
+			if err != nil {
+				// Check if this is a write deadline exceeded to the connection
+				if errors.Is(err, os.ErrDeadlineExceeded) {
+					s.metrics.DroppedUDPDatagram(s.ConnectionID(), DroppedWriteDeadlineExceeded)
+					s.log.Warn().Err(err).Msg("flow (write) deadline exceeded: dropping packet")
+					continue
+				}
+				if isConnectionClosed(err) {
+					s.log.Debug().Msgf("flow (write) connection closed: %v", err)
+				}
+				s.log.Err(err).Msg("failed to write flow payload to origin")
+				s.closeSession(err)
+				// If we fail to write to the origin socket, we need to end the writer and close the session
+				return
+			}
+			// Write must return a non-nil error if it returns n < len(p). https://pkg.go.dev/io#Writer
+			if n < len(payload) {
+				s.metrics.DroppedUDPDatagram(s.ConnectionID(), DroppedWriteFailed)
+				s.log.Err(io.ErrShortWrite).Msg("failed to write the full flow payload to origin")
+				continue
+			}
+			// Mark the session as active since we successfully proxied a packet to the origin.
+			s.markActive()
+		}
+	}
+}
+
+func isConnectionClosed(err error) bool {
+	return errors.Is(err, net.ErrClosed) || errors.Is(err, io.EOF) || errors.Is(err, io.ErrUnexpectedEOF)
+}
+
+// Send an error to the error channel to report that an error has either happened on the tunnel or origin side of the
+// proxied connection.
+func (s *session) closeSession(err error) {
+	select {
+	case s.errChan <- err:
+	default:
+		// In the case that the errChan is already full, we will skip over it and return as to not block
+		// the caller because we should start cleaning up the session.
+		s.log.Warn().Msg("error channel was full")
 	}
-	// Mark the session as active since we proxied a packet to the origin.
-	s.markActive()
-	return n, err
 }
 
 // ResetIdleTimer will restart the current idle timer.
@@ -240,7 +309,8 @@ func (s *session) Close() error {
 
 func (s *session) waitForCloseCondition(ctx context.Context, closeAfterIdle time.Duration) error {
 	connCtx := ctx
-	// Closing the session at the end cancels read so Serve() can return
+	// Closing the session at the end cancels read so Serve() can return, additionally, it closes the
+	// closeWrite channel which indicates to the write loop to return.
 	defer s.Close()
 	if closeAfterIdle == 0 {
 		// Provided that the default caller doesn't specify one
@@ -260,7 +330,10 @@ func (s *session) waitForCloseCondition(ctx context.Context, closeAfterIdle time
 			// still be active on the existing connection.
 			connCtx = newContext
 			continue
-		case reason := <-s.closeChan:
+		case reason := <-s.errChan:
+			// Any error returned here is from the read or write loops indicating that it can no longer process datagrams
+			// and as such the session needs to close.
+			s.metrics.FailedFlow(s.ConnectionID())
 			return reason
 		case <-checkIdleTimer.C:
 			// The check idle timer will only return after an idle period since the last active

quic/v3/session_fuzz_test.go

@@ -4,20 +4,24 @@ import (
 	"testing"
 )
 
-// FuzzSessionWrite verifies that we don't run into any panics when writing variable sized payloads to the origin.
+// FuzzSessionWrite verifies that we don't run into any panics when writing a single variable sized payload to the origin.
 func FuzzSessionWrite(f *testing.F) {
-	f.Fuzz(func(t *testing.T, b []byte) {
-		testSessionWrite(t, b)
-	})
-}
-
-// FuzzSessionServe verifies that we don't run into any panics when reading variable sized payloads from the origin.
-func FuzzSessionServe(f *testing.F) {
 	f.Fuzz(func(t *testing.T, b []byte) {
 		// The origin transport read is bound to 1280 bytes
 		if len(b) > 1280 {
 			b = b[:1280]
 		}
-		testSessionServe_Origin(t, b)
+		testSessionWrite(t, [][]byte{b})
+	})
+}
+
+// FuzzSessionRead verifies that we don't run into any panics when reading a single variable sized payload from the origin.
+func FuzzSessionRead(f *testing.F) {
+	f.Fuzz(func(t *testing.T, b []byte) {
+		// The origin transport read is bound to 1280 bytes
+		if len(b) > 1280 {
+			b = b[:1280]
+		}
+		testSessionRead(t, [][]byte{b})
 	})
 }

quic/v3/session_test.go

@@ -31,60 +31,61 @@ func TestSessionNew(t *testing.T) {
 	}
 }
 
-func testSessionWrite(t *testing.T, payload []byte) {
+func testSessionWrite(t *testing.T, payloads [][]byte) {
 	log := zerolog.Nop()
 	origin, server := net.Pipe()
 	defer origin.Close()
 	defer server.Close()
-	// Start origin server read
-	serverRead := make(chan []byte, 1)
+	// Start origin server reads
+	serverRead := make(chan []byte, len(payloads))
 	go func() {
-		read := make([]byte, 1500)
-		_, _ = server.Read(read[:])
-		serverRead <- read
+		for range len(payloads) {
+			buf := make([]byte, 1500)
+			_, _ = server.Read(buf[:])
+			serverRead <- buf
+		}
+		close(serverRead)
 	}()
-	// Create session and write to origin
+
+	// Create a session
 	session := v3.NewSession(testRequestID, 5*time.Second, origin, testOriginAddr, testLocalAddr, &noopEyeball{}, &noopMetrics{}, &log)
-	n, err := session.Write(payload)
 	defer session.Close()
-	if err != nil {
-		t.Fatal(err)
-	}
-	if n != len(payload) {
-		t.Fatal("unable to write the whole payload")
+	// Start the Serve to begin the writeLoop
+	ctx, cancel := context.WithCancelCause(t.Context())
+	defer cancel(context.Canceled)
+	done := make(chan error)
+	go func() {
+		done <- session.Serve(ctx)
+	}()
+	// Write the payloads to the session
+	for _, payload := range payloads {
+		session.Write(payload)
 	}
 
-	read := <-serverRead
-	if !slices.Equal(payload, read[:len(payload)]) {
-		t.Fatal("payload provided from origin and read value are not the same")
+	// Read from the origin to ensure the payloads were received (in-order)
+	for i, payload := range payloads {
+		read := <-serverRead
+		if !slices.Equal(payload, read[:len(payload)]) {
+			t.Fatalf("payload[%d] provided from origin and read value are not the same (%x) and (%x)", i, payload[:16], read[:16])
+		}
+	}
+	_, more := <-serverRead
+	if more {
+		t.Fatalf("expected the session to have all of the origin payloads received: %d", len(serverRead))
+	}
+
+	assertContextClosed(t, ctx, done, cancel)
+}
+
+func TestSessionWrite(t *testing.T) {
+	defer leaktest.Check(t)()
+	for i := range 1280 {
+		payloads := makePayloads(i, 16)
+		testSessionWrite(t, payloads)
 	}
 }
 
-func TestSessionWrite_Max(t *testing.T) {
-	defer leaktest.Check(t)()
-	payload := makePayload(1280)
-	testSessionWrite(t, payload)
-}
-
-func TestSessionWrite_Min(t *testing.T) {
-	defer leaktest.Check(t)()
-	payload := makePayload(0)
-	testSessionWrite(t, payload)
-}
-
-func TestSessionServe_OriginMax(t *testing.T) {
-	defer leaktest.Check(t)()
-	payload := makePayload(1280)
-	testSessionServe_Origin(t, payload)
-}
-
-func TestSessionServe_OriginMin(t *testing.T) {
-	defer leaktest.Check(t)()
-	payload := makePayload(0)
-	testSessionServe_Origin(t, payload)
-}
-
-func testSessionServe_Origin(t *testing.T, payload []byte) {
+func testSessionRead(t *testing.T, payloads [][]byte) {
 	log := zerolog.Nop()
 	origin, server := net.Pipe()
 	defer origin.Close()
@@ -100,37 +101,42 @@ func testSessionServe_Origin(t *testing.T, payload []byte) {
 		done <- session.Serve(ctx)
 	}()
 
-	// Write from the origin server
-	_, err := server.Write(payload)
-	if err != nil {
-		t.Fatal(err)
-	}
-
-	select {
-	case data := <-eyeball.recvData:
-		// check received data matches provided from origin
-		expectedData := makePayload(1500)
-		_ = v3.MarshalPayloadHeaderTo(testRequestID, expectedData[:])
-		copy(expectedData[17:], payload)
-		if !slices.Equal(expectedData[:v3.DatagramPayloadHeaderLen+len(payload)], data) {
-			t.Fatal("expected datagram did not equal expected")
+	// Write from the origin server to the eyeball
+	go func() {
+		for _, payload := range payloads {
+			_, _ = server.Write(payload)
+		}
+	}()
+
+	// Read from the eyeball to ensure the payloads were received (in-order)
+	for i, payload := range payloads {
+		select {
+		case data := <-eyeball.recvData:
+			// check received data matches provided from origin
+			expectedData := makePayload(1500)
+			_ = v3.MarshalPayloadHeaderTo(testRequestID, expectedData[:])
+			copy(expectedData[17:], payload)
+			if !slices.Equal(expectedData[:v3.DatagramPayloadHeaderLen+len(payload)], data) {
+				t.Fatalf("expected datagram[%d] did not equal expected", i)
+			}
+		case err := <-ctx.Done():
+			// we expect the payload to return before the context to cancel on the session
+			t.Fatal(err)
 		}
-		cancel(errExpectedContextCanceled)
-	case err := <-ctx.Done():
-		// we expect the payload to return before the context to cancel on the session
-		t.Fatal(err)
 	}
 
-	err = <-done
-	if !errors.Is(err, context.Canceled) {
-		t.Fatal(err)
-	}
-	if !errors.Is(context.Cause(ctx), errExpectedContextCanceled) {
-		t.Fatal(err)
+	assertContextClosed(t, ctx, done, cancel)
+}
+
+func TestSessionRead(t *testing.T) {
+	defer leaktest.Check(t)()
+	for i := range 1280 {
+		payloads := makePayloads(i, 16)
+		testSessionRead(t, payloads)
 	}
 }
 
-func TestSessionServe_OriginTooLarge(t *testing.T) {
+func TestSessionRead_OriginTooLarge(t *testing.T) {
 	defer leaktest.Check(t)()
 	log := zerolog.Nop()
 	eyeball := newMockEyeball()
@@ -317,6 +323,8 @@ func TestSessionServe_IdleTimeout(t *testing.T) {
 	closeAfterIdle := 2 * time.Second
 	session := v3.NewSession(testRequestID, closeAfterIdle, origin, testOriginAddr, testLocalAddr, &noopEyeball{}, &noopMetrics{}, &log)
 	err := session.Serve(t.Context())
+
+	// Session should idle timeout if no reads or writes occur
 	if !errors.Is(err, v3.SessionIdleErr{}) {
 		t.Fatal(err)
 	}