Merge branch 'main' into dev

Drop the unimplemented fleet updater design, the stale project-layout sections and broken references, fix the API table, and rename the four Monitor*.cs files left over from the rename.

.gitattributes

@@ -1,4 +1,3 @@
-/.gitattributes text eol=lf
 ci/build-main-amd64.sh text eol=lf
 ci/index-backend-image.sh text eol=lf
 ci/publish-release-offer.sh text eol=lf
@@ -8,11 +7,3 @@ packaging/container/*.sh text eol=lf
 ci/release-node-images.sh text eol=lf
 packaging/install/*.sh text eol=lf
 packaging/install/*.ps1 text eol=lf
-*.sh text eol=lf
-*.init text eol=lf
-*.service text eol=lf
-*.container text eol=lf
-*.Containerfile text eol=lf
-*.ps1 text eol=lf
-*.yml text eol=lf
-*.md text eol=lf

.woodpecker/dev-backend-image.yml

@@ -1,46 +0,0 @@
-labels:
-  ci: domain-linux-native
-  platform: linux/amd64
-
-clone:
-  - name: git
-    image: plugin-git
-    settings:
-      lfs: false
-      # 要拿到 tag 才算得出开发版本号（ci/version.sh：自最近 tag 的提交数）
-      tags: true
-
-when:
-  - event: push
-    branch: main
-    evaluate: 'CI_COMMIT_MESSAGE contains "[CI]"'
-
-# 后端镜像：和 dev-image.yml（Agent 镜像）并行，各自的 restore 只还原自己要用的依赖。构建好之后、推送之前由 ci/wait-for-checks.sh 等 verify 和 web-build。
-# 前端用 web-build workflow 的产物（ci/shared-web.sh）。
-
-concurrency: 1
-
-steps:
-  - name: restore-linux-dependencies
-    image: bash
-    environment:
-      NUGET_PACKAGES: /home/justme/woodpecker/cache/nuget/packages
-    commands:
-      - python3 ci/guard-step.py restore-backend-dependencies 1800 bash ci/restore-linux-dependencies.sh dev-backend
-    depends_on: []
-
-  - name: build-linux-amd64-backend
-    image: bash
-    environment:
-      NUGET_PACKAGES: /home/justme/woodpecker/cache/nuget/packages
-      ZOT_REGISTRY: registry.oeo.one
-      ZOT_USERNAME: justme
-      ZOT_SECRET:
-        from_secret: zot_secret
-      RELEASE_API_URL: https://mo.oeo.one
-      RELEASE_SIGNING_KEY:
-        from_secret: release_signing_key
-    commands:
-      - python3 ci/guard-step.py build-linux-amd64-backend 1800 bash ci/build-dev-image.sh backend
-    depends_on:
-      - restore-linux-dependencies

.woodpecker/dev-image.yml

@@ -1,90 +0,0 @@
-labels:
-  ci: domain-linux-native
-  platform: linux/amd64
-
-clone:
-  - name: git
-    image: plugin-git
-    settings:
-      lfs: false
-      # 要拿到 tag 才算得出开发版本号（ci/version.sh：自最近 tag 的提交数）
-      tags: true
-
-when:
-  - event: push
-    branch: main
-    evaluate: 'CI_COMMIT_MESSAGE contains "[CI]"'
-
-# Agent 镜像：三个 RID 的 Agent 构建，构建完发布。后端镜像在 dev-backend-image.yml 里并行构建：Woodpecker 一个 workflow 里的步骤按依赖深度分层，
-# 下一层要等上一层全部结束，发布放在后端构建（最慢）后面会白等。
-# 镜像构建和 verify（后端快速测试）、web-build 并行跑，不再等它们：构建好之后、推送之前由 ci/wait-for-checks.sh 等这两个 workflow 都成功
-# （任何一个失败就不推；提交信息写 [SKIP TESTS] 只跳过后端测试，verify 照样是成功）。机器有 6 个 workflow 槽（WOODPECKER_MAX_WORKFLOWS），四个同时跑放得下。
-
-concurrency: 1
-
-steps:
-  - name: restore-linux-dependencies
-    image: bash
-    environment:
-      NUGET_PACKAGES: /home/justme/woodpecker/cache/nuget/packages
-    commands:
-      - python3 ci/guard-step.py restore-linux-dependencies 3600 bash ci/restore-linux-dependencies.sh dev-agent
-    depends_on: []
-
-  - name: build-arm64-agent-musl
-    image: bash
-    environment:
-      GITEA_BASE: http://127.0.0.1:3100
-      GITEA_TOKEN:
-        from_secret: gitea_token
-      RELEASE_SIGNING_KEY:
-        from_secret: release_signing_key
-    commands:
-      - python3 ci/guard-step.py build-arm64-agent-musl 3600 bash ci/build-linux-agent.sh linux-musl-arm64
-    depends_on:
-      - restore-linux-dependencies
-
-  - name: build-amd64-agent
-    image: bash
-    environment:
-      GITEA_BASE: http://127.0.0.1:3100
-      GITEA_TOKEN:
-        from_secret: gitea_token
-      RELEASE_SIGNING_KEY:
-        from_secret: release_signing_key
-    commands:
-      - python3 ci/guard-step.py build-amd64-agent 3600 bash ci/build-linux-agent.sh linux-x64
-    depends_on:
-      - restore-linux-dependencies
-
-  - name: build-arm64-agent
-    image: bash
-    environment:
-      GITEA_BASE: http://127.0.0.1:3100
-      GITEA_TOKEN:
-        from_secret: gitea_token
-      RELEASE_SIGNING_KEY:
-        from_secret: release_signing_key
-    commands:
-      - python3 ci/guard-step.py build-arm64-agent 3600 bash ci/build-linux-agent.sh linux-arm64
-    depends_on:
-      - restore-linux-dependencies
-
-  - name: publish-agent-images
-    image: bash
-    environment:
-      GITEA_BASE: http://127.0.0.1:3100
-      GITEA_TOKEN:
-        from_secret: gitea_token
-      ZOT_REGISTRY: registry.oeo.one
-      ZOT_USERNAME: justme
-      ZOT_SECRET:
-        from_secret: zot_secret
-      RELEASE_API_URL: https://mo.oeo.one
-      RELEASE_SIGNING_KEY:
-        from_secret: release_signing_key
-    commands:
-      - python3 ci/guard-step.py publish-agent-images 900 bash ci/release-node-images.sh
-    depends_on:
-      - build-amd64-agent
-      - build-arm64-agent-musl

.woodpecker/e2e.yml

@@ -1,32 +0,0 @@
-labels:
-  ci: domain-linux-native
-  platform: linux/amd64
-
-clone:
-  - name: git
-    image: plugin-git
-    settings:
-      lfs: false
-
-# 带 [CI] 的推送只出镜像，不跑慢测试；这个 workflow 管其余的：PR 和没带 [CI] 的推送
-when:
-  - event: pull_request
-  - event: push
-    branch: main
-    evaluate: 'not (CI_COMMIT_MESSAGE contains "[CI]")'
-
-# 等 verify 编译完、快的测试过了再跑（共用它的增量编译缓存，不重新编译）
-depends_on:
-  - verify
-
-steps:
-  - name: backend-e2e-tests
-    image: bash
-    # [SKIP TESTS] 和 verify 一样，跳过后端测试
-    when:
-      - evaluate: 'not (CI_COMMIT_MESSAGE contains "[SKIP TESTS]")'
-    environment:
-      NUGET_PACKAGES: /home/justme/woodpecker/cache/nuget/packages
-    commands:
-      # 标了 Category=Slow 的端到端测试，见 ci/verify-backend.sh
-      - python3 ci/guard-step.py backend-e2e-tests 1500 bash ci/verify-backend.sh slow

.woodpecker/gitea-release.yml

@@ -14,11 +14,10 @@ when:
 
 depends_on:
   - linux-release
-  - linux-agent
   - windows-release
 
 steps:
-  - name: publish-agent-oci
+  - name: publish-agent-probe-oci
     image: bash
     environment:
       GITEA_BASE: http://127.0.0.1:3100
@@ -32,7 +31,7 @@ steps:
       RELEASE_SIGNING_KEY:
         from_secret: release_signing_key
     commands:
-      - python3 ci/guard-step.py publish-agent-oci 900 bash ci/release-node-images.sh
+      - bash ci/release-node-images.sh
     depends_on: []
 
   - name: publish-gitea-release
@@ -44,6 +43,6 @@ steps:
       RELEASE_SIGNING_KEY:
         from_secret: release_signing_key
     commands:
-      - python3 ci/guard-step.py publish-gitea-release 1800 bash ci/publish-gitea-release.sh
+      - bash ci/publish-gitea-release.sh
     depends_on:
-      - publish-agent-oci
+      - publish-agent-probe-oci

.woodpecker/linux-agent.yml

@@ -1,64 +0,0 @@
-labels:
-  ci: domain-linux-native
-  platform: linux/amd64
-
-clone:
-  - name: git
-    image: plugin-git
-    settings:
-      lfs: false
-      tags: true
-
-when:
-  - event: tag
-    ref: refs/tags/v*
-
-concurrency: 1
-
-steps:
-  - name: restore-linux-dependencies
-    image: bash
-    environment:
-      NUGET_PACKAGES: /home/justme/woodpecker/cache/nuget/packages
-    commands:
-      - python3 ci/guard-step.py restore-linux-dependencies 3600 bash ci/restore-linux-dependencies.sh agent
-    depends_on: []
-
-  - name: build-arm64-agent-musl
-    image: bash
-    environment:
-      GITEA_BASE: http://127.0.0.1:3100
-      GITEA_TOKEN:
-        from_secret: gitea_token
-      RELEASE_SIGNING_KEY:
-        from_secret: release_signing_key
-    commands:
-      - python3 ci/guard-step.py build-arm64-agent-musl 3600 bash ci/build-linux-agent.sh linux-musl-arm64
-    depends_on:
-      - restore-linux-dependencies
-
-  - name: build-amd64-agent
-    image: bash
-    environment:
-      GITEA_BASE: http://127.0.0.1:3100
-      GITEA_TOKEN:
-        from_secret: gitea_token
-      RELEASE_SIGNING_KEY:
-        from_secret: release_signing_key
-    commands:
-      - python3 ci/guard-step.py build-amd64-agent 3600 bash ci/build-linux-agent.sh linux-x64
-    depends_on:
-      - restore-linux-dependencies
-
-  - name: build-arm64-agent
-    image: bash
-    environment:
-      GITEA_BASE: http://127.0.0.1:3100
-      GITEA_TOKEN:
-        from_secret: gitea_token
-      RELEASE_SIGNING_KEY:
-        from_secret: release_signing_key
-    commands:
-      - python3 ci/guard-step.py build-arm64-agent 3600 bash ci/build-linux-agent.sh linux-arm64
-    depends_on:
-      - restore-linux-dependencies

.woodpecker/linux-release.yml

@@ -22,44 +22,35 @@ steps:
     environment:
       NPM_CONFIG_CACHE: /home/justme/.npm
     commands:
-      # 版本取自 tag：先写进工作区（仓库里的版本号只是占位），前端才会带上对的版本
-      - python3 ci/guard-step.py web-build 900 bash ci/build-release-web.sh
+      - npm ci --prefix web --prefer-offline
+      - timeout 600 npm run build --prefix web
     depends_on: []
 
-  - name: restore-linux-dependencies
-    image: bash
-    environment:
-      NUGET_PACKAGES: /home/justme/woodpecker/cache/nuget/packages
-    commands:
-      - python3 ci/guard-step.py restore-linux-dependencies 1800 bash ci/restore-linux-dependencies.sh release
-    depends_on:
-      - web-build
-
   - name: linux-amd64-backend
     image: bash
     environment:
-      NUGET_PACKAGES: /home/justme/woodpecker/cache/nuget/packages
+      NUGET_PACKAGES: /home/justme/.nuget/packages
       ZOT_REGISTRY: registry.oeo.one
       ZOT_USERNAME: justme
       ZOT_SECRET:
         from_secret: zot_secret
     commands:
-      - python3 ci/guard-step.py linux-amd64-backend 1800 bash ci/release-linux.sh amd64 backend
+      - bash ci/release-linux.sh amd64 backend
     depends_on:
-      - restore-linux-dependencies
+      - web-build
 
   - name: linux-arm64-backend
     image: bash
     environment:
-      NUGET_PACKAGES: /home/justme/woodpecker/cache/nuget/packages
+      NUGET_PACKAGES: /home/justme/.nuget/packages
       ZOT_REGISTRY: registry.oeo.one
       ZOT_USERNAME: justme
       ZOT_SECRET:
         from_secret: zot_secret
     commands:
-      - python3 ci/guard-step.py linux-arm64-backend 1800 bash ci/release-linux.sh arm64 backend
+      - bash ci/release-linux.sh arm64 backend
     depends_on:
-      - restore-linux-dependencies
+      - web-build
 
   - name: backend-multiarch-index
     image: bash
@@ -73,7 +64,7 @@ steps:
       RELEASE_SIGNING_KEY:
         from_secret: release_signing_key
     commands:
-      - python3 ci/guard-step.py backend-multiarch-index 900 bash ci/index-backend-image.sh
+      - bash ci/index-backend-image.sh
     depends_on:
       - linux-amd64-backend
       - linux-arm64-backend

.woodpecker/main-amd64-build.yml

@@ -0,0 +1,62 @@
+labels:
+  ci: domain-linux-native
+  platform: linux/amd64
+
+clone:
+  - name: git
+    image: plugin-git
+    settings:
+      lfs: false
+
+when:
+  - event: push
+    branch: main
+    evaluate: 'CI_COMMIT_MESSAGE contains "[CI]"'
+
+concurrency: 1
+
+steps:
+  - name: build-linux-amd64-backend
+    image: bash
+    environment:
+      NUGET_PACKAGES: /home/justme/.nuget/packages
+      ZOT_REGISTRY: registry.oeo.one
+      ZOT_USERNAME: justme
+      ZOT_SECRET:
+        from_secret: zot_secret
+      RELEASE_API_URL: https://mo.oeo.one
+      RELEASE_SIGNING_KEY:
+        from_secret: release_signing_key
+    commands:
+      - bash ci/build-main-amd64.sh backend
+    depends_on: []
+
+  - name: build-linux-amd64-agent
+    image: bash
+    environment:
+      NUGET_PACKAGES: /home/justme/.nuget/packages
+      ZOT_REGISTRY: registry.oeo.one
+      ZOT_USERNAME: justme
+      ZOT_SECRET:
+        from_secret: zot_secret
+      RELEASE_API_URL: https://mo.oeo.one
+      RELEASE_SIGNING_KEY:
+        from_secret: release_signing_key
+    commands:
+      - bash ci/build-main-amd64.sh agent
+    depends_on: []
+
+  - name: build-linux-amd64-probe
+    image: bash
+    environment:
+      NUGET_PACKAGES: /home/justme/.nuget/packages
+      ZOT_REGISTRY: registry.oeo.one
+      ZOT_USERNAME: justme
+      ZOT_SECRET:
+        from_secret: zot_secret
+      RELEASE_API_URL: https://mo.oeo.one
+      RELEASE_SIGNING_KEY:
+        from_secret: release_signing_key
+    commands:
+      - bash ci/build-main-amd64.sh probe
+    depends_on: []

.woodpecker/verify.yml

@@ -11,21 +11,16 @@ clone:
 when:
   - event: pull_request
   - event: push
-    branch: main
+    branch: dev
 
 # On Gitea PR events, CI_COMMIT_MESSAGE is the PR title.
 steps:
-  - name: openwrt-updater-tests
-    image: bash
-    commands:
-      - python3 ci/guard-step.py openwrt-updater-tests 600 bash ci/test-ci-scripts.sh
   - name: backend-tests
     image: bash
     # [SKIP TESTS] omits Backend tests only; web-build remains a dependent workflow.
     when:
       - evaluate: 'not (CI_COMMIT_MESSAGE contains "[SKIP TESTS]")'
     environment:
-      NUGET_PACKAGES: /home/justme/woodpecker/cache/nuget/packages
+      NUGET_PACKAGES: /home/justme/.nuget/packages
     commands:
-      # 增量编译 + 快的那批测试（约 20 秒）；慢的端到端测试在 e2e workflow 里和出镜像并行跑。见 ci/verify-backend.sh
-      - python3 ci/guard-step.py backend-tests 900 bash ci/verify-backend.sh fast
+      - cd backend && dotnet test --solution MoShiro.slnx -c Release

.woodpecker/web-build.yml

@@ -7,26 +7,21 @@ clone:
     image: plugin-git
     settings:
       lfs: false
-      # [CI] 构建要算开发版本号（ci/version.sh：自最近 tag 的提交数），要拿到 tag
-      tags: true
 
 when:
   - event: pull_request
   - event: push
-    branch: main
+    branch: dev
+
+depends_on:
+  - verify
 
-# 和 verify 并行：前端构建不依赖后端测试，两个同时跑
 steps:
   - name: web-build
     image: bash
     environment:
       NPM_CONFIG_CACHE: /home/justme/.npm
     commands:
-      # node_modules 按锁文件哈希缓存，展示端和管理端并行构建，再检查 CSS，见 ci/web-build.sh
-      # main 上带 [CI] 的提交：先写入开发版本号，构建完把产物交给同一条流水线的开发镜像（ci/shared-web.sh），前端只构建一次
-      - |
-        if [ "$CI_PIPELINE_EVENT" = push ] && printf '%s' "$CI_COMMIT_MESSAGE" | grep -qF '[CI]'; then
-          export WEB_PUBLISH_SHARED=1
-          bash ci/version.sh apply
-        fi
-        python3 ci/guard-step.py web-build 900 bash ci/web-build.sh
+      - npm ci --prefix web --prefer-offline
+      - npm run build --prefix web
+      - npm run lint:css --prefix web

.woodpecker/windows-release.yml

@@ -15,24 +15,128 @@ when:
 concurrency: 1
 
 steps:
-  - name: build-macos-x64-agent
+  - name: build-linux-x64-agent
     image: powershell
     environment: &build-environment
       GITEA_BASE: http://localhost:3100
       GITEA_TOKEN:
         from_secret: gitea_token
       NUGET_PACKAGES: 'C:\Users\JustMe-NAS\.nuget\packages'
-      RID: osx-x64
+      RID: linux-x64
       COMPONENT: agent
     commands: &build-commands
       - Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass -Force
       - ./ci/release-cross-platform.ps1 -RuntimeIdentifier $env:RID -Component $env:COMPONENT
     depends_on: []
 
+  - name: build-linux-x64-probe
+    image: powershell
+    environment: &rid-linux-x64-probe
+      <<: *build-environment
+      COMPONENT: probe
+    commands: *build-commands
+    depends_on: []
+
+  - name: build-linux-x64-enroll
+    image: powershell
+    environment: &rid-linux-x64-enroll
+      <<: *build-environment
+      COMPONENT: fleet-enroll
+    commands: *build-commands
+    depends_on: []
+
+  - name: build-linux-arm64-agent
+    image: powershell
+    environment: &rid-linux-arm64-agent
+      <<: *build-environment
+      RID: linux-arm64
+    commands: *build-commands
+    depends_on: []
+
+  - name: build-linux-arm64-probe
+    image: powershell
+    environment: &rid-linux-arm64-probe
+      <<: *rid-linux-arm64-agent
+      COMPONENT: probe
+    commands: *build-commands
+    depends_on: []
+
+  - name: build-linux-arm64-enroll
+    image: powershell
+    environment: &rid-linux-arm64-enroll
+      <<: *rid-linux-arm64-agent
+      COMPONENT: fleet-enroll
+    commands: *build-commands
+    depends_on: []
+
+  - name: build-linux-musl-x64-agent
+    image: powershell
+    environment: &rid-linux-musl-x64-agent
+      <<: *build-environment
+      RID: linux-musl-x64
+    commands: *build-commands
+    depends_on: []
+
+  - name: build-linux-musl-x64-probe
+    image: powershell
+    environment: &rid-linux-musl-x64-probe
+      <<: *rid-linux-musl-x64-agent
+      COMPONENT: probe
+    commands: *build-commands
+    depends_on: []
+
+  - name: build-linux-musl-x64-enroll
+    image: powershell
+    environment: &rid-linux-musl-x64-enroll
+      <<: *rid-linux-musl-x64-agent
+      COMPONENT: fleet-enroll
+    commands: *build-commands
+    depends_on: []
+
+  - name: build-linux-musl-arm64-agent
+    image: powershell
+    environment: &rid-linux-musl-arm64-agent
+      <<: *build-environment
+      RID: linux-musl-arm64
+    commands: *build-commands
+    depends_on: []
+
+  - name: build-linux-musl-arm64-probe
+    image: powershell
+    environment: &rid-linux-musl-arm64-probe
+      <<: *rid-linux-musl-arm64-agent
+      COMPONENT: probe
+    commands: *build-commands
+    depends_on: []
+
+  - name: build-linux-musl-arm64-enroll
+    image: powershell
+    environment: &rid-linux-musl-arm64-enroll
+      <<: *rid-linux-musl-arm64-agent
+      COMPONENT: fleet-enroll
+    commands: *build-commands
+    depends_on: []
+
+  - name: build-macos-x64-agent
+    image: powershell
+    environment: &rid-macos-x64-agent
+      <<: *build-environment
+      RID: osx-x64
+    commands: *build-commands
+    depends_on: []
+
+  - name: build-macos-x64-probe
+    image: powershell
+    environment: &rid-macos-x64-probe
+      <<: *rid-macos-x64-agent
+      COMPONENT: probe
+    commands: *build-commands
+    depends_on: []
+
   - name: build-macos-x64-enroll
     image: powershell
     environment: &rid-macos-x64-enroll
-      <<: *build-environment
+      <<: *rid-macos-x64-agent
       COMPONENT: fleet-enroll
     commands: *build-commands
     depends_on: []
@@ -45,6 +149,14 @@ steps:
     commands: *build-commands
     depends_on: []
 
+  - name: build-macos-arm64-probe
+    image: powershell
+    environment: &rid-macos-arm64-probe
+      <<: *rid-macos-arm64-agent
+      COMPONENT: probe
+    commands: *build-commands
+    depends_on: []
+
   - name: build-macos-arm64-enroll
     image: powershell
     environment: &rid-macos-arm64-enroll
@@ -61,6 +173,14 @@ steps:
     commands: *build-commands
     depends_on: []
 
+  - name: build-windows-x64-probe
+    image: powershell
+    environment: &rid-windows-x64-probe
+      <<: *rid-windows-x64-agent
+      COMPONENT: probe
+    commands: *build-commands
+    depends_on: []
+
   - name: build-windows-x64-enroll
     image: powershell
     environment: &rid-windows-x64-enroll
@@ -77,6 +197,14 @@ steps:
     commands: *build-commands
     depends_on: []
 
+  - name: build-windows-arm64-probe
+    image: powershell
+    environment: &rid-windows-arm64-probe
+      <<: *rid-windows-arm64-agent
+      COMPONENT: probe
+    commands: *build-commands
+    depends_on: []
+
   - name: build-windows-arm64-enroll
     image: powershell
     environment: &rid-windows-arm64-enroll
@@ -92,11 +220,27 @@ steps:
       - Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass -Force
       - ./ci/package-cross-platform.ps1
     depends_on:
+      - build-linux-x64-agent
+      - build-linux-x64-probe
+      - build-linux-x64-enroll
+      - build-linux-arm64-agent
+      - build-linux-arm64-probe
+      - build-linux-arm64-enroll
+      - build-linux-musl-x64-agent
+      - build-linux-musl-x64-probe
+      - build-linux-musl-x64-enroll
+      - build-linux-musl-arm64-agent
+      - build-linux-musl-arm64-probe
+      - build-linux-musl-arm64-enroll
       - build-macos-x64-agent
+      - build-macos-x64-probe
       - build-macos-x64-enroll
       - build-macos-arm64-agent
+      - build-macos-arm64-probe
       - build-macos-arm64-enroll
       - build-windows-x64-agent
+      - build-windows-x64-probe
       - build-windows-x64-enroll
       - build-windows-arm64-agent
+      - build-windows-arm64-probe
       - build-windows-arm64-enroll

AGENTS.md

@@ -1,89 +1,43 @@
 # Agent Instructions
 
-MoShiro is a self-hosted network path quality monitor. Agents on home broadband, NAT networks, or public VPS hosts run active measurements and record raw per-packet and per-event data. The Backend stores it, derives metrics, scores lines per service model, and serves a REST/SSE API to two web apps. Documentation is in Chinese under [`docs/`](docs/); source comments are mostly Chinese too. [`README.md`](README.md) is the entry point.
+## Project
 
-## Mental model
+MoShiro is a self-hosted network path quality monitor. Backend coordinates Agents and Probes, stores measurements, and serves the REST/SSE API. `web/display` is the read-only overview; `web/admin` manages nodes, tests, and settings.
 
-| Term | Meaning |
-| --- | --- |
-| Agent | Unified collector process (NativeAOT, BCL only), listening and dialing by default. Each Agent connects to the Backend over `wss` and reports outgoing link health and incoming sessions. |
-| Line | One directed (dialer, listener) Agent pair, per IP family. Id `<agent_id>~<dialer_id>`, IPv6 adds `~ipv6`; series key `line:<dialer_id>` |
-| Kind | A declared measurement method (`udp_echo`, `tcp_echo`, `connect`, ...) composed from building blocks. Defined once in `MoShiro.Protocol/Catalog/Kinds` |
-| RawChunk | Self-describing columnar block, cut every 10 s, appended to segment files. The Backend never trusts node-reported statistics |
-| Metric, Model | `MetricDef` is a one-line declaration over accumulators; models score metrics at query time. Both are recomputable from raw data |
-| Plan | A sequence of on-demand test phases (for example `hy2_ladder`). Exists only in the Backend (`Application/Tests/Plans.cs`) |
-| Fleet | Product-agnostic node framework under `backend/fleet/` (identity, enrollment, RPC, link, outbox, state sync) |
-| Operator set | Passkey-signed trust root. Nodes pin its genesis fingerprint and owner at install time and verify the chain themselves |
+## Branch and workspace safety
 
-## Invariants (violations fail the build or review)
+- Treat `main` as the stable release branch and `dev` as the integration branch. Keep ordinary implementation work on `dev` or a feature branch; integrate into `main` through a pull request.
+- A push to `main` builds images only if its commit message contains `[CI]` (for a merged PR, put it in the PR or merge commit title). Without `[CI]` nothing is built; use `[CI SKIP]` to skip the pipeline entirely. When squashing or pushing to `main`, add `[CI]` unless a build is explicitly not wanted.
+- Before editing, inspect `git status`. Preserve pre-existing user changes: do not reset, clean, stash, or include unrelated files in a commit.
+- Do not commit credentials, API keys, registry passwords, or generated runtime data. CI credentials belong in Woodpecker secrets (`zot_secret`, `gitea_token`, `release_signing_key` for signing release manifests). CI registers the signed manifests at `/api/release/offer`, which needs no token because the panel verifies the signature. The release signing private key never goes into the repository or the panel; only the public key in `packaging/update/` does.
+- Keep development OCI images in the `moshiro-dev/` repository path and release images in `moshiro/`. Do not let a development build overwrite release tags such as `latest`.
 
-- Node processes are kept to measurement and reporting by build-time restrictions: process start, dynamic loading, `HttpClient`, and file writes outside `Fleet.Hosting.ConfigStore` are banned (`backend/BannedSymbols.Node.txt`). Capabilities such as updates or command execution are not part of the core; adding or changing one needs a design review against `docs/security/security.md`.
-- `backend/fleet/**` must not reference `MoShiro.*` (`ArchitectureTests`). `MoShiro.Measurement` contains no service concepts such as REALITY or Hysteria2.
-- Every RPC method is declared in a method table (`FleetMethodTable`, `MoShiroMethodTable`) with direction, type and roles. Add the declaration before the handler.
-- `Backend.Api` must not reference `Storage`. Error-code to HTTP mapping happens once, in the Api layer.
-- Server-sent parameters pass `ParamSpec` validation and then the node-local hard limits (`BudgetGuard`). Never bypass either.
-- Public API responses contain no IP, city or ASN.
-- Code style is enforced at build time (`.editorconfig`, `BannedSymbols.txt`, warnings as errors). Details in [`docs/dev/coding-style.md`](docs/dev/coding-style.md).
+## Project layout
 
-## Where to change what
+- `backend/src/MoShiro.Backend/`: service entry point and composition.
+- `backend/src/MoShiro.Backend.Api/`: REST endpoints, contracts, and OpenAPI.
+- `backend/agent/`, `backend/probe/`: node applications and their READMEs.
+- `backend/fleet/`, `backend/shared/`: reusable node framework, measurement engine, protocols, and data formats.
+- `web/display/`, `web/admin/`: Vite applications; `web/shared/`: API client, generated types, charts, and shared UI.
+- `packaging/`, `scripts/`, `.woodpecker/`, `docs/`: container/package builds, CI workflows, and project documentation.
 
-| Task | Location |
-| --- | --- |
-| New measurement method | Kind declaration in `backend/shared/MoShiro.Protocol/Catalog/Kinds`; new building block only if needed in `MoShiro.Measurement/Blocks` plus `BlockCatalog` |
-| New metric | `MetricDef` in `backend/src/MoShiro.Backend.Analysis/Metrics`, add to `MetricRegistry.All` and a preset if shown |
-| New service model | `Analysis/Models` |
-| New RPC method | `MoShiro.Protocol/Rpc` (params, `MoShiroMethodTable`), then `router.On(...)` in `AgentApp`; Backend-side handlers register in `Backend.Gateway` (`FleetGateway`) |
-| New REST endpoint | `Backend.Api/Public` or `Admin` (typed results, operationId, capability), contract DTO in `Api/Contracts`, then regenerate `backend/openapi/v1.json` and web types |
-| New capability | `Capability` constants in `Backend.Application/Auth/Capabilities.cs`; every `/api/admin` endpoint must declare one |
-| Schema change | Append a migration in `Storage.Sqlite/Schema.cs` and bump `CurrentVersion` |
-| Agent-to-Agent connection states, retry cadence, health reporting | `backend/agent/src/MoShiro.Agent/Dial` (`DialConnection` control layer, `TaskLink` QUIC channel, `LinkStatus`), `LinkHealth` in `MoShiro.Protocol/Rpc`, [`docs/design/architecture.md`](docs/design/architecture.md) §4.3 |
-| Install or update behaviour | `packaging/` (install scripts, Quadlet, compose, updater) and `docs/ops/deployment.md` |
-| Web UI | `web/display` (anonymous read-only), `web/admin`, shared code in `web/shared`. Colors only in `web/shared/theme/tokens.css` |
+## Local development and builds
 
-## Documentation
+Requirements: .NET SDK 10 and Node.js/npm. The SDK selection is defined by the root `global.json`; NuGet package versions are centralized in the root `Directory.Packages.props`.
 
-Index: [`README.md`](README.md#文档). Prefer the code and `backend/openapi/v1.json` when they disagree with a document, and fix the document. Update docs in the same change as behaviour: `docs/design` (architecture, protocol, measurement, analysis, storage, query), `docs/security`, `docs/api`, `docs/ops`, `docs/dev`, and `docs/roadmap.md` for unimplemented work. Do not record dates or change history in docs.
+- All .NET projects: `dotnet build MoShiro.slnx -c Release` from the repository root.
+- Backend build: `cd backend && dotnet build MoShiro.slnx -c Release`
+- Backend tests: `cd backend && dotnet test --solution MoShiro.slnx -c Release`
+- Frontend dependencies/build: `npm ci --prefix web && npm run build --prefix web`
+- Frontend style check: `npm run lint:css --prefix web`
+- Local backend: from `backend/`, run `dotnet run --project src/MoShiro.Backend -- --urls http://localhost:5000`.
+- Local web apps: from the repository root, run `npm ci --prefix web` then `npm run dev --prefix web`. The current dev config uses display port 5173 and admin port 5175; `/admin/` is proxied through the display server. Check the Vite configs when working on another branch because ports can differ.
+- After changing API contracts, regenerate/check `backend/openapi/v1.json` and run `npm run gen:api --prefix web` so frontend API types stay aligned.
 
-Writing style for docs, READMEs, comments and commit messages (the `stop-slop` skill lists the patterns in detail; it is written for English, the same rules apply to Chinese):
-
-- State the fact directly. Drop "not X but Y" contrasts, negation lists, rhetorical setups, and aphorisms that read like pull-quotes.
-- Name the actor. Prefer "the Agent rejects the task" over "the task is rejected".
-- No filler openers ("it is worth noting"), no closing summaries, no adverbs doing emphasis. Avoid bold for emphasis.
-- Do not claim absolutes ("never", "always", "cannot") unless the code or a build-time check enforces it. Describe current behaviour and mark planned work as planned.
-- Give concrete names, paths, values and commands instead of general statements.
-
-## Workspace safety
-
-- Inspect `git status` before editing. Preserve existing uncommitted changes; do not reset, clean, stash, or commit unrelated files.
-- `main` is the only working branch. Commit to it directly; use a pull request when review is required.
-- Never commit credentials, API keys, registry passwords, or generated runtime data. CI credentials are Woodpecker secrets (`zot_secret`, `gitea_token`, `release_signing_key`).
-- The release signing private key must not enter the repository or the panel. Only the public key under `packaging/update/` is versioned.
-
-## Build and test
-
-Requirements: .NET SDK 10 (pinned by `global.json`) and Node.js with npm. NuGet versions are centralized in `Directory.Packages.props`.
-
-| Task | Command |
-| --- | --- |
-| Build all .NET projects | `dotnet build MoShiro.slnx -c Release` |
-| Backend tests | `cd backend && dotnet test --solution MoShiro.slnx -c Release` |
-| Run Backend | `cd backend && dotnet run --project src/MoShiro.Backend -- --urls http://localhost:5000` |
-| Web dependencies and build | `npm ci --prefix web && npm run build --prefix web` |
-| Web dev servers | `npm run dev --prefix web` (ports are set in the Vite configs) |
-| Style check | `npm run lint:css --prefix web` |
-| Regenerate API types | `npm run gen:api --prefix web` |
-
-First run of the Backend prints a claim code in its log; the admin UI asks for it. Slow end-to-end tests carry `[Trait("Category", "Slow")]`. After changing API contracts, update `backend/openapi/v1.json` and regenerate the web types. Run only the validation the task requires; do not start a full tag or release build unless requested.
+Follow the user's requested validation scope. Do not start a full tag/release build unless requested.
 
 ## CI and releases
 
-Triggers, tags and image paths are defined in `.woodpecker/*.yml` and [`docs/ops/deployment.md`](docs/ops/deployment.md). Verify them there before changing or describing them. The rules below are the stable contract.
+CI triggers and image tags can change; inspect `.woodpecker/*.yml` and `docs/deployment.md` before changing or describing them. In particular, check the distinction between `[CI]` main builds, `[SKIP TESTS]`, and version-tag releases in the current branch.
 
-- Every push to `main` runs the fast Backend tests and the web build. Both must pass before images are published.
-- A commit message containing `[CI]` builds and pushes development images. `[CI SKIP]` skips the pipeline; `[SKIP TESTS]` skips only the Backend tests. Slow end-to-end tests run only for pushes without `[CI]` and for pull requests.
-- A tag `vMAJOR.MINOR.PATCH` runs the full cross-platform release pipeline. No version-bump commit is needed.
-- Versions are produced by `ci/version.sh`. The `<Version>` in `backend/Directory.Build.props` and the `version` in `web` `package.json` files are placeholders (`0.0.0-dev`) and must not be edited.
-- Linux Agent builds use `ci/build-linux-agent.sh` for glibc on amd64/arm64 and musl on arm64 in both dev and tag workflows. OCI and OpenWrt installers reuse these native outputs; Windows runners build only Windows/macOS.
-- Build stores keep NuGet cache mounts, retain image layers for 7 days, and clean temporary outputs under store locks (shared for native builds, exclusive for cleanup). Do not use `buildah prune` on these stores: it removes restore caches too.
-- Development images use the `moshiro-dev/` repository path and release images use `moshiro/`. A development build must never overwrite a release tag such as `latest`.
-- Release manifests are registered at `/api/release/offer`, which requires no token because the panel verifies the signature.
+A version tag `vMAJOR.MINOR.PATCH` runs the full cross-platform release pipeline. Development image builds and releases must remain isolated in their respective Zot repository paths.

MoShiro.slnx

@@ -27,8 +27,11 @@
   <Folder Name="/backend/agent/src/">
     <Project Path="backend/agent/src/MoShiro.Agent/MoShiro.Agent.csproj" />
   </Folder>
-  <Folder Name="/backend/agent/tests/">
-    <Project Path="backend/agent/tests/MoShiro.Agent.Tests/MoShiro.Agent.Tests.csproj" />
+  <Folder Name="/backend/probe/src/">
+    <Project Path="backend/probe/src/MoShiro.Probe/MoShiro.Probe.csproj" />
+  </Folder>
+  <Folder Name="/backend/probe/tests/">
+    <Project Path="backend/probe/tests/MoShiro.Probe.Tests/MoShiro.Probe.Tests.csproj" />
   </Folder>
   <Folder Name="/backend/src/">
     <Project Path="backend/src/MoShiro.Backend.Analysis/MoShiro.Backend.Analysis.csproj" />

README.md

@@ -1,70 +1,118 @@
 # MoShiro
 
-MoShiro 旨在解决跨网络线路质量难以量化的问题：
+MoShiro 是一个自托管的**网络线路质量观测工具**。它在若干 Probe（例如家宽出口）与多台 Agent（例如海外 VPS）之间做主动网络测量，只记录逐包、逐事件的原始数据；Backend 保存这些原始数据，推算延迟、丢包、抖动、吞吐等指标，并回答「如果在这条线路上部署某类服务，体验会怎样」；Web 界面用来查看、比较和管理。
 
-- 家宽到海外 VPS 的延迟、丢包、抖动随时段波动，单点 ping 无法反映真实体验；
-- 不同代理、游戏、流媒体服务对线路的敏感度各异，缺少统一的评价口径。
+## 项目边界
 
-它通过 Agent 之间的主动测量采集原始数据，由 Backend 推算指标，并按服务模型（REALITY、Hysteria2、Minecraft 等）对线路评分与排序。
+**它做什么**
 
-## 组件
+- 主动测量：组合 TCP、UDP、TLS、ICMP 等采集积木，只记录时间、序号、大小这类原始事实，**不转发也不记录业务内容**。
+- 线路分析：Backend 从原始数据生成时间序列指标，按服务模型（REALITY、Hysteria2、Minecraft 等）评分并对多条线路排序、对比；模型和指标可以事后重算。
+- 管理：节点注册与审批、Probe 和 Agent 之间连哪些线路、任务与按需测试、节点版本与更新，都在管理端完成。
 
-| 组件 | 职责 |
-| --- | --- |
-| Agent | 统一采集节点，默认同时监听和拨号；可部署在 VPS、家宽或 NAT 后 |
-| Backend | 注册、线路编排、存储、分析和 REST / SSE |
-| Web | 匿名只读展示端 `display` 与管理端 `admin` |
+**它不做什么**
+
+- **节点是纯采集器**：Agent 和 Probe 没有终端、命令执行、文件读写和自更新能力，也不加载外部代码。更新由节点宿主机上的更新器完成，节点进程只放面板转来的发布清单。
+- **不是主机监控**：Agent 只上报 CPU、内存、磁盘、网卡流量和运行时长这类只读的汇总数字，不采集型号、主机名、进程和任何 IP；Probe 不上报主机资源。
+- **不暴露地址**：公开接口里没有 IP、城市、ASN；国家和运营商只由 Backend 查询（节点自己不访问任何外部服务）。
+- 暂不包含告警（见文末待办）。
+
+## 信任关系
 
 ```text
-                 Backend ---- REST / SSE ---- Web
-                 ^     ^
-                 |     | 每个 Agent 独立建立 wss，汇报自身及线路状态
-                 |     |
-              Agent A <=====> Agent B
-                    直连测量
+ 管理员 ── 账号密码（Cookie 会话 + CSRF）或 API 令牌（按能力授权）──▶ Backend
+ 访客   ── 匿名只读，站点设为公开时才可用（没有 IP、城市、ASN）──────────────▶ Backend 公开接口
+
+ 节点   ── 加入令牌一次性注册，之后用自己的 ECDSA P-256 密钥签名认证 ───────▶ Backend（只存公钥）
+ Probe  ── 用同一把密钥签名建立测量会话 ───────────────────────────────────▶ Agent（只放行面板名单里的 Probe 公钥）
+ CI     ── 发布私钥对「镜像 digest 清单」签名 ──▶ Backend 原样转发 ──▶ 节点宿主机的更新器用本机公钥验签
 ```
 
-Agent 默认向其他可拨号 Agent 建立测量线路，A→B 和 B→A 分别记录；节点不连接自身。
-管理端统一展示 Agent，在线路画板中按方向开关测量、查看连接状态和立即重试。
-家宽或 NAT 后的 Agent 仍可拨出；入站测量需要其他节点能够访问其监听地址和端口。
-可在本地 `agent.json` 中显式设置 `listen.enabled: false` 关闭监听，节点身份和程序保持统一。
+- **Backend 不信任节点上报的内容**：数据按线路隔离存放；节点的身份靠签名，数据库泄露也无法冒充节点。
+- **节点不信任 Backend 的下发**：任务超出节点本机的硬上限（目标网段、码率、包速率、监听端口）就被拒绝；节点不执行任何「执行类」指令。
+- **更新不依赖面板的诚实**：清单必须由发布私钥签名；节点宿主机的更新器用本机公钥验签、只认本机配置的镜像仓库、按 digest 拉取、拒绝旧的 `seq`。面板被攻破也推不出恶意镜像。
 
-## 快速开始
+## 架构
 
-环境：.NET SDK 10（版本见 `global.json`）、Node.js 与 npm。
+```text
+                         Web（展示端 · 管理端）
+                                     ▲  REST + SSE
+        ┌──────────────────────────────────────────────────────────┐
+        │ Backend（面板）                                          │
+        │ Api · Registry · Planner · Links                         │
+        │ Releases · Analysis · LiveHub                            │
+        │ Store（段文件 + SQLite）                                 │
+        └──────────────────────────────────────────────────────────┘
+              ▲                                       ▲
+              │ wss 长连接（节点发起）                │ wss 长连接（节点发起）
+              │ 任务、数据、状态、更新通知            │ 身份、Agent 列表、状态、更新通知
+   ┌────────────────────┐                 ┌──────────────────────┐
+   │ Agent × 10+        │ ◀── 测量会话 ── │ Probe × N            │
+   │ 有公网 IP，监听    │                 │ 家宽等出口，NAT 后   │
+   └────────────────────┘                 └──────────────────────┘
+   测量会话：TLS 控制通道 + UDP/TCP 数据通道，由 Probe 主动拨出
+```
+
+两种连接：**Probe 和 Agent 都用长连接连 Backend**（身份、任务、状态、更新通知），**测量只走 Probe → Agent**（Probe 主动拨出，Agent 不需要连 Probe，家宽也不需要入站端口）。哪些 Probe 连哪些 Agent 由 Backend 决定：默认自动全连，也可以在管理端手动选择。
+
+```text
+ 采集（只记录）          传输（有界、至少一次）           分析（注册制、可重算）         展示
+ Probe / Agent 积木 ──▶ 原始列块（10 秒一块）──▶ Backend 段文件 ──▶ 指标与服务模型 ──▶ REST / SSE ──▶ Web
+```
+
+Backend 用段文件保存原始数据，用 SQLite 保存索引、指标与配置。**完整的架构、数据链路、安全边界和设计决策见 [`docs/architecture.md`](docs/architecture.md)。**
+
+## 协议
+
+- **控制链路（Fleet）**：节点到 Backend 的 `wss` 长连接（`/api/agent/v1/rpc`），JSON-RPC 2.0 信封加二进制通道；挑战加签名认证，`hello` 协商版本和能力，心跳；**期望状态同步**（Backend 下发完整快照，节点调和并回报已应用的版本和被拒绝的条目）；**可靠发件箱**（原始块至少一次投递，按 id 去重）。Fleet 与业务无关，放在 `backend/fleet/`。
+- **业务方法**：`moshiro.*`（按需测试、Probe 断开、事件、主机汇总、更新通知等），状态文档有 `task_set`（Agent 的任务集）和 `probe_agents`（Probe 的 Agent 列表）。
+- **测量数据面**：Probe 到 Agent 的 TLS 控制通道、UDP 和 TCP 数据通道（默认 `7443/tcp`、`7443/udp`、`7444/tcp`）；会话用 `session_id` 和 MAC 认证、不绑定五元组，NAT 映射变化后自动跟随。
+- **REST + SSE**：`/api/public`（展示，可匿名）、`/api/admin`（管理）；OpenAPI 在 `backend/openapi/v1.json`，前端据此生成类型。
+
+细节见 [`docs/protocol.md`](docs/protocol.md)、[`backend/fleet/README.md`](backend/fleet/README.md)、[`docs/api.md`](docs/api.md)、[`docs/enrollment.md`](docs/enrollment.md)。
+
+## 如何构建
+
+环境：.NET SDK 10（`global.json` 指定版本）、Node.js 与 npm。在仓库根目录：
 
 ```sh
-cd backend && dotnet run --project src/MoShiro.Backend -- --urls http://localhost:5000   # Backend
-npm ci --prefix web && npm run dev --prefix web                                          # Web（另开终端，仓库根目录）
+dotnet build MoShiro.slnx -c Release          # Backend、Agent、Probe、Fleet、测试和 CI 工具
+npm ci --prefix web && npm run build --prefix web   # 展示端和管理端，产物在 web/display/dist、web/admin/dist
 ```
 
-展示端 <http://localhost:5173/>，管理端 <http://localhost:5173/admin/>。首次访问管理端进入认领向导，认领码输出在后端日志。
+本地运行：先在 `backend/` 里 `dotnet run --project src/MoShiro.Backend -- --urls http://localhost:5000`，再 `npm run dev --prefix web`，展示端是 <http://localhost:5173/>，管理端是 <http://localhost:5173/admin/>。
 
-构建与测试：
+测试和检查：
 
 ```sh
-dotnet build MoShiro.slnx -c Release
-dotnet test --solution MoShiro.slnx
-npm run build --prefix web
+dotnet test --solution MoShiro.slnx            # 后端和节点的测试
+npm run lint:css --prefix web                  # 前端样式（颜色只能写在主题令牌里）
+npm run gen:api --prefix web                   # 后端改了接口之后重新生成前端类型
 ```
 
+Backend、Agent、Probe 以 OCI 镜像发布（Linux amd64、arm64），Agent 和 Probe 另有 Linux glibc/musl、macOS、Windows 的二进制包。NativeAOT 的构建细节、CI 触发规则（`[CI]`、`[SKIP TESTS]`、版本 tag）、镜像路径、容器节点的运行和更新见 [`docs/deployment.md`](docs/deployment.md)；前端见 [`web/README.md`](web/README.md)，后端见 [`backend/README.md`](backend/README.md)。
+
 ## 文档
 
-| 主题 | 文档 |
+| 文档 | 内容 |
 | --- | --- |
-| 架构与数据链路 | [architecture](docs/design/architecture.md) |
-| 线协议与方法 | [protocol](docs/design/protocol.md) |
-| 采集引擎与任务 | [measurement](docs/design/measurement.md) |
-| 分析、存储、查询 | [analysis](docs/design/analysis.md) · [storage](docs/design/storage.md) · [query](docs/design/query.md) |
-| 安全：威胁模型、访问控制、注册与身份 | [security](docs/security/security.md) · [access](docs/security/access.md) · [enrollment](docs/security/enrollment.md) |
-| REST / SSE 接口 | [api](docs/api/api.md) |
-| 构建、发布、CI、部署与更新 | [deployment](docs/ops/deployment.md) |
-| 仓库结构与代码风格 | [project-layout](docs/dev/project-layout.md) · [coding-style](docs/dev/coding-style.md) |
-| 路线图 | [roadmap](docs/roadmap.md) |
-| 子项目 | [backend](backend/README.md) · [fleet](backend/fleet/README.md) · [agent](backend/agent/README.md) · [web](web/README.md) |
+| [`docs/architecture.md`](docs/architecture.md) | 架构、数据链路、安全边界、设计决策 |
+| [`docs/security.md`](docs/security.md) | 零信任设计稿（部分已实现）：威胁模型、入网信任链、监听隐身、extension 授权 |
+| [`docs/access.md`](docs/access.md) | 访问控制与入网信任链设计（尚未实现）：账户、角色、能力、API key、提权窗口、passkey 到节点的签名链 |
+| [`docs/protocol.md`](docs/protocol.md)、[`backend/fleet/README.md`](backend/fleet/README.md) | 线协议、Fleet 节点框架（身份、注册、连接） |
+| [`docs/enrollment.md`](docs/enrollment.md) | 注册、身份、Probe 与 Agent 的接入 |
+| [`docs/measurement.md`](docs/measurement.md)、[`docs/analysis.md`](docs/analysis.md) | 采集引擎与任务模型、分析层与指标 |
+| [`docs/storage.md`](docs/storage.md)、[`docs/query.md`](docs/query.md) | 存储、展示数据查询 |
+| [`docs/api.md`](docs/api.md)、[`docs/api-guidelines.md`](docs/api-guidelines.md) | API 规范与约束 |
+| [`docs/deployment.md`](docs/deployment.md) | 构建、发布、CI、部署与更新 |
+| [`docs/project-layout.md`](docs/project-layout.md)、[`docs/coding-style.md`](docs/coding-style.md) | 仓库结构与依赖规则、代码风格 |
 
-贡献与自动化代理的工作约定见 [`AGENTS.md`](AGENTS.md)。
+代码目录：`backend/`（Backend、`fleet/` 节点框架、`shared/` 测量引擎和协议、`agent/`、`probe/`、`tests/`）、`web/`（`display` 展示端、`admin` 管理端、`shared`）、`packaging/` 和 `scripts/`（部署与维护）、`ci/` 和 `.woodpecker/`（流水线）。
+
+## 待办
+
+见 [`TODO.md`](TODO.md)。
 
 ## 许可证
 
-[Apache License 2.0](LICENSE)
+本项目采用 [Apache License 2.0](LICENSE)。

TODO.md

@@ -0,0 +1,48 @@
+# TODO
+
+按主题记录还没做的事，做完就删掉对应的条目。已经实现的东西以 [`docs/`](docs/) 和代码为准。
+
+## 测量
+
+- 路由跳数、路径漂移和 trace
+- QUIC 握手耗时：TCP 建连和 TLS 握手已经有了（`connect` 任务）。.NET 的 QUIC 依赖系统里的 msquic 原生库，和节点「只依赖 OpenSSL、NativeAOT 单文件」冲突，要做只能自己写最小的 QUIC Initial 握手探测
+- 网页长截图
+- pcap / eBPF 采集
+
+## 任务与面板
+
+- 单个 Agent 的任务（目前是线路模板加按 Agent 的覆盖，没有只属于某一个 Agent 的独立任务）
+- 原生 webhook 投递；告警（基于 webhook）。实现之前先定约束（只允许 HTTPS、禁止私网和回环、不跟随重定向、限制超时和响应大小、解析后再检查地址），见 [`docs/security.md`](docs/security.md) 第 9.2 节
+
+## 部署
+
+- 后端自动下载前端资源：用 Podman、Docker 部署时，默认拉取最新的前端资源（现在前端要自己构建，镜像里没有）。
+[security.md](docs/security.md)
+## 访客页面
+
+- 人机验证：访客页面要通过验证才能拉数据，前端访问时静默触发。候选 tiago2/cap 和 Cloudflare Turnstile。
+
+## 更新
+
+- 更新遥测（现在面板只有每个节点的「当前更新状态」）
+- **回滚路径的实机测试**：容器节点更新后观察 20 秒、失败回滚，只有单元和脚本测试，没有在真实节点上触发过。
+- **更新的灰度和自动策略**：按标签分批发布；`POLICY=auto` 的长期验证。
+- **纯二进制节点的更新**（`fleet-updater`、`fleet-release`、Windows 服务和 OpenWrt procd 适配器）。设计要点，实现时照这个来：
+  - 信任来自签名：只接受离线发布密钥签名的清单，发布源可以是面板、GitHub Releases 或任意 HTTP 服务；面板被攻破也推不出恶意二进制。
+  - 本地说了算：更新策略（`auto` / `notify` / `pinned:<version>`）和信任的公钥写在节点本地配置里，面板只能「建议」；支持多把公钥以便轮换。
+  - 清单带 `app`、`channel`、`version`、`min_from`、各 RID 制品的 `sha256` 和大小、`key_id`、签名（对规范化 JSON 签名）；本地记录已装过的最高版本，拒绝降级和旧清单重放。
+  - 流程：下载 → 校验 → 解包到 `versions/<v>/` → 原子切换 `current`（Linux 符号链接，Windows 目录联结）→ 重启服务 → 等健康信标确认 → 失败切回旧版本并把该版本标记为坏版本。
+  - 权限分离：更新进程只能写安装目录和重启被管理的服务，读不到节点身份私钥。
+  - OpenWrt：`keep_versions = 1`，下载前检查闪存剩余空间。
+  - 容器节点已经实现的那一套（清单签名、`seq` 防重放、按 digest 拉取、回滚）见 [`docs/deployment.md`](docs/deployment.md)。
+
+## 安全
+
+- **零信任**：设计稿在 [`docs/security.md`](docs/security.md)，按其中第 12 节的顺序实现（阶段零「协议方向声明表」已完成）：Probe 自己生成密钥、Agent 预握手敲门、操作员凭证和名册、测量数据的签名和归属校验、入库配额和管理端 CSP、Backend 应用层身份、extension 框架（敏感能力都放进可选安装的独立 extension，核心保持 NativeAOT）；可选的 UDP 单包授权、UDP 会话迁移的路径验证。
+- **extension**（Web SSH、TTY、更新遥测，敏感 operator）：设计时必须保住「节点是纯采集器」这条边界，见 [`README.md`](README.md) 的项目边界。
+- 对 UDP 包、控制帧、RawChunk、JSON-RPC 做模糊测试；伪造签名和重放的专项测试。
+
+## 打包
+
+- OpenWrt：procd 脚本、ipk/apk 包、sysupgrade 保留列表，在 aarch64 路由器上长时间运行验证。
+- Windows 服务适配。

backend/Directory.Build.props

@@ -13,17 +13,11 @@
     <UseArtifactsOutput>true</UseArtifactsOutput>
     <ArtifactsPath>$(MSBuildThisFileDirectory)artifacts</ArtifactsPath>
     <Deterministic>true</Deterministic>
-    <Version>0.0.0-dev</Version>
+    <Version>0.3.3</Version>
     <NuGetAudit>false</NuGetAudit>
   </PropertyGroup>
   <ItemGroup>
     <PackageReference Include="Microsoft.CodeAnalysis.BannedApiAnalyzers" PrivateAssets="all" />
     <AdditionalFiles Include="$(MSBuildThisFileDirectory)BannedSymbols.txt" />
   </ItemGroup>
-  <ItemGroup Condition="'$(UseAotAnywhere)' == 'true' and $([MSBuild]::IsOSPlatform('Linux'))">
-    <PackageReference Include="Vezel.Zig.Toolsets.linux-x64"
-                      VersionOverride="0.16.0.2"
-                      PrivateAssets="all"
-                      GeneratePathProperty="true" />
-  </ItemGroup>
 </Project>

backend/MoShiro.slnx

@@ -27,8 +27,11 @@
   <Folder Name="/agent/src/">
     <Project Path="agent/src/MoShiro.Agent/MoShiro.Agent.csproj" />
   </Folder>
-  <Folder Name="/agent/tests/">
-    <Project Path="agent/tests/MoShiro.Agent.Tests/MoShiro.Agent.Tests.csproj" />
+  <Folder Name="/probe/src/">
+    <Project Path="probe/src/MoShiro.Probe/MoShiro.Probe.csproj" />
+  </Folder>
+  <Folder Name="/probe/tests/">
+    <Project Path="probe/tests/MoShiro.Probe.Tests/MoShiro.Probe.Tests.csproj" />
   </Folder>
   <Folder Name="/backend/src/">
     <Project Path="src/MoShiro.Backend.Analysis/MoShiro.Backend.Analysis.csproj" />

backend/README.md

@@ -36,9 +36,9 @@ dotnet test --solution MoShiro.slnx
 dotnet publish src/MoShiro.Backend -c Release -r <RID> -p:PublishAot=true
 ```
 
-例如 `<RID>` 可为 `win-x64`、`linux-x64` 或 `linux-arm64`。NativeAOT 发布需要对应平台的 C/C++ 工具链。仓库根目录的 `scripts/publish.sh backend <RID>` 可构建单个平台的 Backend 可执行文件；容器发布由 Woodpecker tag 流水线完成，详见 [`../docs/ops/deployment.md`](../docs/ops/deployment.md)。
+例如 `<RID>` 可为 `win-x64`、`linux-x64` 或 `linux-arm64`。NativeAOT 发布需要对应平台的 C/C++ 工具链。仓库根目录的 `scripts/publish.sh backend <RID>` 可构建单个平台的 Backend 可执行文件；容器发布由 Woodpecker tag 流水线完成，详见 [`../docs/deployment.md`](../docs/deployment.md)。
 
-正式发布通过 Woodpecker tag 流水线生成 Linux/amd64 与 Linux/arm64 OCI 镜像，容器内监听 `8080`，数据写入挂载的 `/data` 卷。Agent 的多平台二进制包单独发布，不包含在 Backend 镜像中；Linux runner 构建三个 Linux RID，Windows runner 构建 macOS 和 Windows。
+正式发布通过 Woodpecker tag 流水线生成 Linux/amd64 与 Linux/arm64 OCI 镜像，容器内监听 `8080`，数据写入挂载的 `/data` 卷。Agent 与 Probe 的多平台二进制包单独发布，不包含在 Backend 镜像中。
 
 ## 本地运行
 
@@ -49,6 +49,6 @@ cd backend
 dotnet run --project src/MoShiro.Backend -- --urls http://localhost:5000
 ```
 
-没有默认管理员：第一次打开面板时按页面引导自己创建管理员、注册 passkey、建立信任根，页面要的认领码写在后端日志里（或用 `MoShiro__ClaimCode` 指定）。`MoShiro__InitialAdminUser` / `MoShiro__InitialAdminPassword` 只给自动化用。数据目录、公开地址、Cookie 和 CORS 等设置通过 `MoShiro` 配置节传入；完整配置说明见 [`../docs/ops/deployment.md`](../docs/ops/deployment.md)。站点默认需要登录，可在管理端修改公开设置。
+没有默认管理员：第一次打开面板时按页面引导自己创建管理员、注册 passkey、建立信任根，页面要的认领码写在后端日志里（或用 `MoShiro__ClaimCode` 指定）。`MoShiro__InitialAdminUser` / `MoShiro__InitialAdminPassword` 只给自动化用。数据目录、公开地址、Cookie 和 CORS 等设置通过 `MoShiro` 配置节传入；完整配置说明见 [`../docs/deployment.md`](../docs/deployment.md)。站点默认需要登录，可在管理端修改公开设置。
 
-OpenAPI 文档由构建生成到 `openapi/v1.json`。API 设计和端点说明见 [`../docs/api/api.md`](../docs/api/api.md)；项目总览见仓库根目录的 [`README.md`](../README.md)。
+OpenAPI 文档由构建生成到 `openapi/v1.json`。API 设计和端点说明见 [`../docs/api.md`](../docs/api.md)；项目总览见仓库根目录的 [`README.md`](../README.md)。

backend/agent/Directory.Build.props

@@ -8,7 +8,7 @@
     <EventSourceSupport>false</EventSourceSupport>
     <DebuggerSupport>false</DebuggerSupport>
     <HttpActivityPropagationSupport>false</HttpActivityPropagationSupport>
-    <Http3Support>true</Http3Support>
+    <Http3Support>false</Http3Support>
     <MetricsSupport>false</MetricsSupport>
     <UseSizeOptimizedLinq>true</UseSizeOptimizedLinq>
     <JsonSerializerIsReflectionEnabledByDefault>false</JsonSerializerIsReflectionEnabledByDefault>

backend/agent/README.md

@@ -1,22 +1,20 @@
 # MoShiro Agent
 
-Agent 是统一的采集节点，默认同时监听和拨号。每个 Agent 独立连接 Backend，接收任务集和拨号目标；节点之间通过 TLS 控制会话、QUIC Datagram 和 TCP 数据连接完成测量。
+Agent 是部署在 VPS 或其他可达网络节点上的采集进程。它向 Backend 注册并保持连接，接受 Probe 的测量会话，执行线路任务中的 Agent 侧工作，再将两端数据合并后上报。
+
+Agent 仅开放配置的控制和数据端口。它不提供远程终端、命令执行或文件管理；本地配置中的网络范围、速率和并发限制由节点自行执行，Backend 下发的任务不能越过这些限制。
+
+## 工作流程
 
 ```text
-Agent A <==== 直接测量 ====> Agent B
-   |                           |
-   +--------- wss -------------+
-                |
-             Backend
+Probe ──TLS 控制 / UDP、TCP 测量──> Agent ──长连接与数据上报──> Backend
+                                      ▲
+                                fleet-enroll 注册
 ```
 
-Backend 默认安排 Agent 之间的双向有向线路，自连接被排除。拨号方通过心跳上报控制层、TCP 和 QUIC 状态、失败原因及重试时间；监听方上报接入会话。Backend 汇总这些报告，供 Agent 详情和线路矩阵使用。测量数据由监听方合并后上传。
-
-本地 `listen.enabled` 默认 `true`。可显式关闭监听，适用于无需入站测量的节点；这仍是同一种 Agent。NAT 后的节点可主动拨出，入站可达性取决于地址和端口映射。
-
 ## 构建
 
-需要 .NET SDK 10（版本见根目录 `global.json`）。从 `backend/` 目录构建：
+需要 .NET SDK 10。为应用 根目录的 `global.json` 的 SDK 设置，请从 `backend/` 目录构建：
 
 ```sh
 cd backend
@@ -31,7 +29,7 @@ dotnet publish agent/src/MoShiro.Agent -c Release -r <RID> -p:PublishAot=true
 dotnet publish fleet/tools/fleet-enroll -c Release -r <RID> -p:PublishAot=true
 ```
 
-也可在仓库根目录运行 `bash scripts/publish.sh agent <RID>` 和 `bash scripts/publish.sh fleet-enroll <RID>`。部署包和平台说明见 [`../../docs/ops/deployment.md`](../../docs/ops/deployment.md)。
+也可在仓库根目录运行 `bash scripts/publish.sh agent <RID>` 和 `bash scripts/publish.sh fleet-enroll <RID>`。部署包和平台说明见 [`../../docs/deployment.md`](../../docs/deployment.md)。
 
 ## 注册与运行
 
@@ -49,7 +47,7 @@ fleet-enroll --server https://<backend> --join <token> --name <agent-name> --ide
 moshiro-agent --config /etc/moshiro/agent.json
 ```
 
-容器节点的运行方式（podman + systemd）和宿主机更新器见 [`../../docs/ops/deployment.md`](../../docs/ops/deployment.md)。也可使用 OCI 镜像 `registry.oeo.one/moshiro/moshiro-agent:<version>`，自动选择 Linux amd64 或 arm64 镜像。镜像同时包含 `fleet-enroll`；可通过 `--entrypoint /usr/local/bin/fleet-enroll` 执行首次注册。容器默认读取命名 volume 中的 `/etc/moshiro/agent.json` 与身份文件；建议使用宿主网络，Linux ICMP 优先使用普通用户 ping socket，内核 ping_group_range 需允许映射后的 GID。镜像以 UID/GID 1654 运行，新命名 volume 由引擎按镜像目录初始化权限。
+容器节点的运行方式（podman + systemd）和宿主机更新器见 [`../../docs/deployment.md`](../../docs/deployment.md)。也可使用 OCI 镜像 `registry.oeo.one/moshiro/moshiro-agent:<version>`，自动选择 Linux amd64 或 arm64 镜像。镜像同时包含 `fleet-enroll`；可通过 `--entrypoint /usr/local/bin/fleet-enroll` 执行首次注册。容器默认读取命名 volume 中的 `/etc/moshiro/agent.json` 与身份文件；建议使用宿主网络，Linux ICMP 优先使用普通用户 ping socket，内核 ping_group_range 需允许映射后的 GID。镜像以 UID/GID 1654 运行，新命名 volume 由引擎按镜像目录初始化权限。
 
 不传 `--config` 时默认读取 `/etc/moshiro/agent.json`。Agent 默认监听控制端口 `7443/tcp`，数据端口 `7443/udp` 和 `7444/tcp`；防火墙和面板中的节点端口配置需与本地配置一致。`reporting.host_stats` 可关闭主机状态上报；启用时只上报资源使用等汇总数据，不包含主机名、进程或 IP。
 
@@ -57,9 +55,9 @@ IPv6 线路需要 Agent 实际绑定 IPv6。新配置的默认 `listen.bind` 为
 
 ## 设计说明
 
-- 每个 拨号方 使用独立的 TLS 会话和数据密钥；拨号方 断开时对应的会话资源会释放。
+- 每个 Probe 使用独立的 TLS 会话和数据密钥；Probe 断开时对应的会话资源会释放。
 - 数据面会校验会话认证、重放窗口和速率限制；无效数据包会静默丢弃。
 - 测量引擎和协议由 `backend/shared/` 提供；节点身份、注册和连接由 `backend/fleet/` 提供。
-- 详细安全边界与拓扑见 [`../../docs/design/architecture.md`](../../docs/design/architecture.md)，协议说明见 [`../../docs/design/protocol.md`](../../docs/design/protocol.md)，部署见 [`../../docs/ops/deployment.md`](../../docs/ops/deployment.md)。
+- 详细安全边界与拓扑见 [`../../docs/architecture.md`](../../docs/architecture.md)，协议说明见 [`../../docs/protocol.md`](../../docs/protocol.md)，部署见 [`../../docs/deployment.md`](../../docs/deployment.md)。
 
-Podman 系用 Quadlet（`packaging/quadlet/moshiro-agent.container`，默认自动更新），Docker 用 `packaging/compose/agent/compose.yaml`（手动更新）；配置、身份和缓存放在命名 volume 中，无需宿主配置目录、keep-id 或手动 chown。首次导入和自动更新见部署文档。
+Docker 与 rootless/rootful Podman 共用 `packaging/compose/agent/compose.yaml`，配置、身份和缓存放在命名 volume 中，无需宿主配置目录、keep-id 或手动 chown。首次导入和自动更新见部署文档。

backend/agent/src/MoShiro.Agent/AgentApp.cs

@@ -10,17 +10,13 @@ using MoShiro.Protocol;
 
 namespace MoShiro.Agent;
 
-/// <summary>
-/// 节点的组装点：与 Backend 的链路、监听侧（数据面和点对点监听，<c>listen.enabled</c> 为 false 时不启动）、拨号侧（Backend 下发的拨号目标）都在这里接起来。
-/// 没有全局状态，测试可以在同一进程里启动多个。
-/// </summary>
-public sealed class AgentApp : IDisposable, IAsyncDisposable
+/// <summary>Agent 的组装点：数据面、点对点监听、与 Backend 的链路都在这里接起来。没有全局状态，测试可以在同一进程里启动多个。</summary>
+public sealed class AgentApp : IDisposable
 {
     public const string Version = MoShiro.Protocol.BuildInfo.Version;
 
     private readonly AgentConfig _config;
-    private readonly bool _listening;
-    private readonly bool? _ipv6Supported;
+    private readonly bool _ipv6Supported;
     private readonly KnockGate _knock;
     private readonly TrustAnchor _trust;
     private readonly RestartController _restart;
@@ -28,7 +24,6 @@ public sealed class AgentApp : IDisposable, IAsyncDisposable
     private readonly LoadedNodeIdentity _identity;
     private readonly TimeProvider _time;
     private readonly INodeLog _log;
-    private readonly INodeLogSource? _logSource;
     private readonly MonoClock _clock;
     private readonly AgentCounters _counters = new();
     private readonly SessionRegistry _sessions = new();
@@ -36,34 +31,27 @@ public sealed class AgentApp : IDisposable, IAsyncDisposable
     private readonly TaskCompletionSource _ready = new(TaskCreationOptions.RunContinuationsAsynchronously);
     private readonly AgentAdmission _admission;
     private readonly Outbox _outbox = new(64 << 20);
-    private readonly RestartUpdater? _update;
-    private readonly OpenWrtUpdater? _nativeUpdate;
-    private readonly DialerHost _dialer;
+    private readonly UpdateBridge _update;
     private readonly BackendUplink _uplink;
-    private readonly QuicDatagramTransport _datagrams;
+    private readonly UdpMux _udp;
     private readonly TcpAcceptor _tcp;
     private readonly PeerListener _peers;
     private readonly PeerHub _hub;
-    private readonly ChunkAssembler _chunks;
-    private readonly RateChangeLog _rateChanges = new();
     private readonly TaskManager _tasks;
     private readonly TestRunner _tests;
     private readonly FleetLink _link;
-    private int _disposed;
 
     /// <param name="dataDirectory">配置目录（容器里的 <c>/etc/moshiro</c>）：宿主机更新器和节点在这里交换清单、请求和状态。</param>
     public AgentApp(AgentConfig config, LoadedNodeIdentity identity, string dataDirectory, TimeProvider time, INodeLog log, RestartController? restart = null)
     {
         _ownedRestart = restart is null ? new RestartController() : null;
         _restart = restart ?? _ownedRestart!;
-        _update = RestartUpdater.FromEnvironment("agent", _restart.RequestNow, time, log);
-        _nativeUpdate = OpenWrtUpdater.FromEnvironment(time, log);
+        _update = new UpdateBridge(dataDirectory, new ConfigStore(dataDirectory).Write, log);
+        NodeBinary.Prime(); // 正在运行的二进制的哈希：先在后台算好，握手时用
         _config = config;
-        _listening = config.Listen.Enabled;
         _identity = identity;
         _time = time;
         _log = log;
-        _logSource = log as INodeLogSource;
         _clock = new MonoClock(time);
         _uplink = new BackendUplink(time);
 
@@ -73,44 +61,37 @@ public sealed class AgentApp : IDisposable, IAsyncDisposable
             bind = IPAddress.Any;
         }
 
-        _ipv6Supported = _listening ? bind.AddressFamily == System.Net.Sockets.AddressFamily.InterNetworkV6 : null;
+        _ipv6Supported = bind.AddressFamily == System.Net.Sockets.AddressFamily.InterNetworkV6;
         _admission = new AgentAdmission(_streams);
-        QuicSettings quic = config.Quic.ToSettings();
-        if (quic.Validate() is { } invalid)
-        {
-            throw new InvalidOperationException("agent.json quic: " + invalid);
-        }
-
-        _datagrams = new QuicDatagramTransport(_sessions, _admission, _clock, _counters.Data, log, bind, config.Listen.QuicPorts, identity.Certificate, quic);
+        _udp = new UdpMux(_sessions, _admission, _clock, _counters.Data, log, bind, config.Listen.UdpPorts);
         // 预握手敲门：控制端口和数据端口共用一个关卡（限速和去重一起算）；名册就是期望状态里的 peers
         _trust = TrustAnchor.Load(config.Fleet.IdentityDir, log);
-        _dialer = new DialerHost(new DialEnv(identity.PeerIdentity, identity.NodeId.Value, config.Limits, quic, new DialDirectory(new ConfigStore(dataDirectory), dataDirectory, log, _trust), _clock, time, log, _rateChanges), time, log);
         _knock = new KnockGate(identity.PeerIdentity, RosterKeys, time);
         _tcp = new TcpAcceptor(bind, config.Listen.TcpPorts, identity.Certificate, _sessions, _counters.Data, _knock);
 
-        // 监听方积木的环境：和拨号方的拨号方共用同一套积木，区别只在这里装配的部分（准入策略、TCP 监听）
+        // 监听方积木的环境：和 Probe 的拨号方共用同一套积木，区别只在这里装配的部分（准入策略、TCP 监听）
         var env = new BlockEnv
         {
             Clock = _clock,
             Log = log,
-            Session = () => null, // 每个拨号方会话各用自己绑定的环境（BlockEnv.WithSession）
-            Datagrams = _datagrams,
+            Session = () => null, // 每个 Probe 会话各用自己绑定的环境（BlockEnv.WithSession）
+            Udp = _udp,
             TcpListen = _tcp,
             Admission = _admission,
             Counters = _counters.Data,
             MaxPacketsPerSecond = config.Limits.MaxTotalPps,
             MaxBulkMbps = config.Limits.MaxTotalMbps,
         };
-        _chunks = new ChunkAssembler(_streams, _clock, _counters, _outbox, identity.NodeId, time, log);
-        var targets = new TargetRunners(_chunks, identity.NodeId, _clock, time, log, _rateChanges);
-        _tasks = new TaskManager(CreateGuard, _streams, env, targets, () => _datagrams.BoundPorts, () => _tcp.BoundPorts, _trust, acceptLineTasks: _listening);
-        _hub = new PeerHub(_sessions, _clock, _chunks, _tasks, _uplink, time, log);
-        _tests = new TestRunner(_tasks, _sessions, _streams, env, _uplink, config.Limits.MaxConcurrentTests, transport => transport == "quic_datagram" ? _datagrams.BoundPorts : _tcp.BoundPorts, time, log);
+        var chunks = new ChunkAssembler(_streams, _clock, _counters, _outbox, identity.NodeId, time, log);
+        var targets = new TargetRunners(chunks, identity.NodeId, _clock, time, log);
+        _tasks = new TaskManager(CreateGuard, _streams, env, targets, () => _udp.BoundPorts, () => _tcp.BoundPorts, _trust);
+        _hub = new PeerHub(_sessions, _clock, chunks, _tasks, _uplink, time, log);
+        _tests = new TestRunner(_tasks, _sessions, _streams, env, _uplink, config.Limits.MaxConcurrentTests, transport => transport == "udp" ? _udp.BoundPorts : _tcp.BoundPorts, time, log);
 
-        _tasks.OnLineTasksChanged = (dialer, set) => _ = _hub.PushToAsync(dialer, set).AsTask();
+        _tasks.OnLineTasksChanged = (probe, set) => _ = _hub.PushToAsync(probe, set).AsTask();
         _tasks.OnPeersChanged = _hub.CloseNotIn;
-        _hub.OnDialerLost = dialer => _tests.DialerLost(dialer);
-        _admission.OnEndpointChanged = dialer => _ = _uplink.EmitEventAsync(AgentEventKinds.EndpointChanged, dialer.DatagramRemoteEndPoint?.ToString()).AsTask();
+        _hub.OnProbeLost = probe => _tests.ProbeLost(probe);
+        _admission.OnEndpointChanged = probe => _ = _uplink.EmitEventAsync(AgentEventKinds.EndpointChanged, probe.UdpEndpoint?.ToString()).AsTask();
 
         _peers = new PeerListener(
             new PeerListenerOptions
@@ -133,31 +114,13 @@ public sealed class AgentApp : IDisposable, IAsyncDisposable
 
     public void Dispose()
     {
-        if (Interlocked.Exchange(ref _disposed, 1) != 0)
-        {
-            return;
-        }
-
-        _dialer.Dispose();
-        _datagrams.Dispose();
-        _ownedRestart?.Dispose();
-    }
-
-    public async ValueTask DisposeAsync()
-    {
-        if (Interlocked.Exchange(ref _disposed, 1) != 0)
-        {
-            return;
-        }
-
-        _dialer.Dispose();
-        await _datagrams.DisposeAsync().ConfigureAwait(false);
+        _udp.Dispose();
         _ownedRestart?.Dispose();
     }
 
     internal StreamRegistry Streams => _streams;
 
-    public IReadOnlyList<int> QuicPorts => _datagrams.BoundPorts;
+    public IReadOnlyList<int> UdpPorts => _udp.BoundPorts;
 
     public IReadOnlyList<int> TcpPorts => _tcp.BoundPorts;
 
@@ -170,56 +133,40 @@ public sealed class AgentApp : IDisposable, IAsyncDisposable
 
     public FleetLink Link => _link;
 
-    /// <summary>拨号侧（测试和状态查看用）。</summary>
-    public DialerHost Dialer => _dialer;
-
-    /// <summary>对等面（拨号方连进来的会话）的路由器，测试用。</summary>
+    /// <summary>对等面（Probe 连进来的会话）的路由器，测试用。</summary>
     public RpcRouter<PeerContext> PeerRouter => _hub.Router;
 
-    /// <summary>期望状态里是否已经有这个拨号方的公钥（测试用：握手会按它校验）。</summary>
-    public bool IsPeerKnown(string dialerId, string publicKeyBase64) =>
-        _tasks.LookupPeerKey(dialerId) is { } key && Convert.ToBase64String(key) == publicKeyBase64;
+    /// <summary>期望状态里是否已经有这个 Probe 的公钥（测试用：握手会按它校验）。</summary>
+    public bool IsPeerKnown(string probeId, string publicKeyBase64) =>
+        _tasks.LookupPeerKey(probeId) is { } key && Convert.ToBase64String(key) == publicKeyBase64;
 
     public TestRunner Tests => _tests;
 
     public async Task RunAsync(CancellationToken ct)
     {
-        _log.Info("moshiro agent starting", MoShiro.Protocol.BuildInfo.Fields);
         using CancellationTokenSource linked = CancellationTokenSource.CreateLinkedTokenSource(ct);
-        Task quic = Task.CompletedTask;
-        Task tcp = Task.CompletedTask;
-        Task peers = Task.CompletedTask;
-        if (_listening)
-        {
         // 数据面和控制端口先开始监听，再连接 Backend：hello 里上报的是实际端口
-            _datagrams.Bind();
+        _udp.Bind();
         _tcp.Bind();
-            quic = _datagrams.RunAsync(linked.Token);
-            tcp = _tcp.RunAsync(linked.Token);
-            peers = _peers.RunAsync(linked.Token);
+        Task udp = _udp.RunAsync(linked.Token);
+        Task tcp = _tcp.RunAsync(linked.Token);
+        Task peers = _peers.RunAsync(linked.Token);
         await Task.WhenAny(_peers.Started.Task, peers);
         if (peers.IsFaulted)
         {
             await linked.CancelAsync();
-                await Task.WhenAll(quic, tcp, peers); // 抛出监听失败的原因（端口被占用等）
-            }
-
-            _log.Info("agent listening", ("control", ControlPort), ("quic", string.Join(',', QuicPorts)), ("tcp", string.Join(',', TcpPorts)));
-        }
-        else
-        {
-            _log.Info("agent is dial-only: listening is disabled");
+            await Task.WhenAll(udp, tcp, peers); // 抛出监听失败的原因（端口被占用等）
         }
 
         _ready.TrySetResult();
+        _log.Info("agent listening", ("control", ControlPort), ("udp", string.Join(',', UdpPorts)), ("tcp", string.Join(',', TcpPorts)));
         try
         {
-            await Task.WhenAll(quic, tcp, peers, _link.RunAsync(linked.Token), _dialer.RunAsync(linked.Token), ReportHostAsync(linked.Token), _chunks.RunTrafficReportingAsync(linked.Token));
+            await Task.WhenAll(udp, tcp, peers, _link.RunAsync(linked.Token), ReportHostAsync(linked.Token));
         }
         finally
         {
             await _tasks.StopAsync();
-            _chunks.FlushTraffic();
             await linked.CancelAsync();
         }
     }
@@ -256,14 +203,14 @@ public sealed class AgentApp : IDisposable, IAsyncDisposable
                 MaxTotalPps = _config.Limits.MaxTotalPps,
                 AllowedTargetCidrs = _config.Limits.AllowedTargetCidrs,
                 MaxConcurrentTests = _config.Limits.MaxConcurrentTests,
-                AllowedQuicPorts = _datagrams.BoundPorts,
+                AllowedUdpPorts = _udp.BoundPorts,
                 AllowedTcpPorts = _tcp.BoundPorts,
             },
-            _listening ? BlockCatalog.AgentKinds : BlockCatalog.TargetKinds);
+            BlockCatalog.AgentKinds);
 
     private FleetLink BuildLink()
     {
-        var router = new RpcRouter<LinkContext>(RpcSide.Node, MoShiroMethodTable.Table);
+        var router = new RpcRouter<LinkContext>(RpcSide.Node, MoShiroMethodTable.Table, NodeRoles.Agent);
         HashSet<string>? allowed = _config.Policy.AllowMethods is { } list ? [.. list] : null;
         bool Allow(string method) => allowed is null || allowed.Contains(method);
 
@@ -289,8 +236,7 @@ public sealed class AgentApp : IDisposable, IAsyncDisposable
         {
             router.OnNotify(MoShiroMethods.UpdateOffer, MoShiroJsonContext.Default.UpdateOfferParams, (_, offer, _) =>
             {
-                _update?.Offer(offer);
-                _nativeUpdate?.Offer();
+                _update.WriteOffer(offer);
                 return ValueTask.CompletedTask;
             });
         }
@@ -312,45 +258,18 @@ public sealed class AgentApp : IDisposable, IAsyncDisposable
         {
             router.OnNotify(MoShiroMethods.UpdateApply, MoShiroJsonContext.Default.UpdateOfferParams, (_, _, _) =>
             {
-                _update?.RequestApply();
-                _nativeUpdate?.RequestApply();
+                _update.RequestApply();
                 return ValueTask.CompletedTask;
             });
         }
 
-        if (Allow(MoShiroMethods.DialerDisconnect))
+        if (Allow(MoShiroMethods.ProbeDisconnect))
         {
             router.On(
-                MoShiroMethods.DialerDisconnect,
-                MoShiroJsonContext.Default.DialerDisconnectParams,
+                MoShiroMethods.ProbeDisconnect,
+                MoShiroJsonContext.Default.ProbeDisconnectParams,
                 MoShiroJsonContext.Default.TestRunResult,
-                async (_, request, _) => Result<TestRunResult>.Ok(new TestRunResult { Accepted = await _hub.DisconnectAsync(request.DialerId.Value) }));
-        }
-
-        if (Allow(MoShiroMethods.LogsFetch))
-        {
-            router.On(
-                MoShiroMethods.LogsFetch,
-                MoShiroJsonContext.Default.LogsFetchParams,
-                MoShiroJsonContext.Default.LogsFetchResult,
-                (_, request, _) =>
-                {
-                    NodeLogPage page = _logSource?.Read(request.AfterId, request.Limit, request.MinLevel) ?? new NodeLogPage([], 0);
-                    return ValueTask.FromResult(Result<LogsFetchResult>.Ok(new LogsFetchResult
-                    {
-                        LastId = page.LastId,
-                        Items = [.. page.Items.Select(e => new LogEntryDto { Id = e.Id, Type = e.Type, AtUtcMs = e.Record.AtUtcMs, Level = e.Record.Level, Message = e.Record.Message, Fields = e.Record.Fields })],
-                    }));
-                });
-        }
-
-        if (Allow(MoShiroMethods.LinkRetry))
-        {
-            router.OnNotify(MoShiroMethods.LinkRetry, MoShiroJsonContext.Default.LinkRetryParams, (_, request, _) =>
-            {
-                _dialer.RetryListener(request.ListenerId.Value);
-                return ValueTask.CompletedTask;
-            });
+                async (_, request, _) => Result<TestRunResult>.Ok(new TestRunResult { Accepted = await _hub.DisconnectAsync(request.ProbeId.Value) }));
         }
 
         var options = new FleetLinkOptions
@@ -358,35 +277,21 @@ public sealed class AgentApp : IDisposable, IAsyncDisposable
             Server = new Uri(_config.Fleet.Server),
             Identity = _identity.Identity,
             ServerVerifier = _trust,
+            Attestation = _update.Attest,
             NodeId = _identity.NodeId,
             NodeVersion = Version,
-            MinServerVersion = () => Environment.GetEnvironmentVariable("MOSHIRO_MIN_SERVER_VERSION") is { Length: > 0 } required ? required : _config.Fleet.MinServerVersion,
-            Capabilities = [MoShiroCapabilities.Base, MoShiroCapabilities.KindsPrefix + string.Join(',', _listening ? BlockCatalog.AgentKinds : BlockCatalog.TargetKinds)],
+            Capabilities = [MoShiroCapabilities.Base, MoShiroCapabilities.KindsPrefix + string.Join(',', BlockCatalog.AgentKinds)],
             HelloMeta = HelloMeta,
             HeartbeatExtra = HeartbeatExtra,
-            ShutdownReason = () => _restart.NowReason == "update" ? "updating" : "stopping",
         };
-        var link = new FleetLink(
-            options,
-            router,
-            [
-                new StateHandler<TaskSet>(TaskSet.StateType, MoShiroJsonContext.Default.TaskSet, _tasks.ApplyAsync),
-                new StateHandler<DialTargets>(DialTargets.StateType, MoShiroJsonContext.Default.DialTargets, (version, targets, _) =>
-                {
-                    _dialer.Directory.Apply(targets);
-                    return ValueTask.FromResult(new StateSyncResult { AppliedVersion = version });
-                }),
-            ],
-            _outbox,
-            _time,
-            _log);
+        var link = new FleetLink(options, router, [new StateHandler<TaskSet>(TaskSet.StateType, MoShiroJsonContext.Default.TaskSet, _tasks.ApplyAsync)], _outbox, _time, _log);
 
-        // 链路重新连上之后补报：在线的拨号方都告诉 Backend（Backend 重启或链路抖动期间错过了 dialer_connected）
+        // 链路重新连上之后补报：在线的 Probe 都告诉 Backend（Backend 重启或链路抖动期间错过了 probe_connected）
         link.OnConnected = async (_, _) =>
         {
-            foreach (ConnectedDialer dialer in _hub.ConnectedDialers())
+            foreach (ConnectedProbe probe in _hub.ConnectedProbes())
             {
-                await _uplink.EmitEventAsync(AgentEventKinds.DialerConnected, dialer.Id.Value, dialer);
+                await _uplink.EmitEventAsync(AgentEventKinds.ProbeConnected, probe.Id.Value, probe);
             }
         };
         return link;
@@ -397,11 +302,16 @@ public sealed class AgentApp : IDisposable, IAsyncDisposable
 
     private JsonElement? HelloMeta()
     {
-        var meta = new NodeHelloMeta
+        var meta = new AgentHelloMeta
         {
             Owners = [.. _trust.Controllers],
             Ipv6 = Ipv6Diagnostics.Sample(_ipv6Supported, Environment.GetEnvironmentVariable("MOSHIRO_NETWORK_SCOPE"), Environment.GetEnvironmentVariable("HOST_PROC")),
+            Ipv6Supported = _ipv6Supported,
+            Addresses = [.. LocalAddresses.Sample()],
             Platform = MoShiro.Measurement.PlatformDetector.Detect(),
+            TlsPublicKey = Convert.ToBase64String(_identity.Certificate.PublicKey.ExportSubjectPublicKeyInfo()),
+            ProbeTcpPort = ControlPort,
+            ProbeUdpPorts = [.. UdpPorts],
             LocalLimits = new LocalLimits
             {
                 MaxTotalMbps = _config.Limits.MaxTotalMbps,
@@ -409,17 +319,8 @@ public sealed class AgentApp : IDisposable, IAsyncDisposable
                 AllowedTargetCidrs = _config.Limits.AllowedTargetCidrs,
                 MaxConcurrentTests = _config.Limits.MaxConcurrentTests,
             },
-            Listen = !_listening
-                ? null
-                : new ListenInfo
-                {
-                    TcpPort = ControlPort,
-                    QuicPorts = [.. QuicPorts],
-                    Ipv6Supported = _ipv6Supported == true,
-                    Addresses = [.. LocalAddresses.Sample()],
-                },
         };
-        return JsonSerializer.SerializeToElement(meta, MoShiroJsonContext.Default.NodeHelloMeta);
+        return JsonSerializer.SerializeToElement(meta, MoShiroJsonContext.Default.AgentHelloMeta);
     }
 
     private long _lastIn;
@@ -432,33 +333,21 @@ public sealed class AgentApp : IDisposable, IAsyncDisposable
         long seconds = Math.Max(1, (nowMs - _lastAtMs) / 1000);
         long packetsIn = _counters.PacketsIn;
         long packetsOut = _counters.PacketsOut;
-        NodeLogSummary? logSummary = _logSource?.TakeSummary();
-        UpdateStatus? updateStatus = _nativeUpdate?.Status() ?? _update?.Status();
-        var extra = new NodeHeartbeatExtra
+        var extra = new AgentHeartbeatExtra
         {
-            RuntimeStatus = updateStatus?.State == "applying" ? NodeRuntimeStatus.Updating : !_listening && _dialer.Directory.Targets.Count == 0 ? NodeRuntimeStatus.Idle : NodeRuntimeStatus.Running,
             Ipv6 = Ipv6Diagnostics.Sample(_ipv6Supported, Environment.GetEnvironmentVariable("MOSHIRO_NETWORK_SCOPE"), Environment.GetEnvironmentVariable("HOST_PROC")),
-            Links = _dialer.SnapshotLinks(),
-            Dialers = [.. _hub.ConnectedDialers()],
+            Probes = [.. _hub.ConnectedProbes()],
+            Sessions = _sessions.All.Count,
             PpsIn = _lastAtMs == 0 ? 0 : (packetsIn - _lastIn) / seconds,
             PpsOut = _lastAtMs == 0 ? 0 : (packetsOut - _lastOut) / seconds,
             DroppedInvalid = _counters.DroppedInvalid,
             DroppedRateLimited = _counters.DroppedRateLimited,
             QueueBytes = _outbox.PendingBytes,
-            Update = updateStatus,
-            RateChanges = _rateChanges.Take(64),
-            LogSummary = logSummary is null ? null : new LogSummaryDto
-            {
-                LastId = logSummary.LastId,
-                Info = logSummary.Info,
-                Warn = logSummary.Warn,
-                Error = logSummary.Error,
-                Types = [.. logSummary.Types.Select(t => new LogTypeCountDto { Type = t.Type, Level = t.Level, Count = t.Count })],
-            },
+            Update = _update.ReadStatus(),
         };
         _lastIn = packetsIn;
         _lastOut = packetsOut;
         _lastAtMs = nowMs;
-        return JsonSerializer.SerializeToElement(extra, MoShiroJsonContext.Default.NodeHeartbeatExtra);
+        return JsonSerializer.SerializeToElement(extra, MoShiroJsonContext.Default.AgentHeartbeatExtra);
     }
 }

backend/agent/src/MoShiro.Agent/AgentConfig.cs

@@ -1,5 +1,4 @@
 using System.Text.Json.Serialization;
-using MoShiro.Measurement;
 
 namespace MoShiro.Agent;
 
@@ -15,9 +14,6 @@ public sealed class AgentConfig
     [JsonPropertyName("listen")]
     public ListenSection Listen { get; set; } = new();
 
-    [JsonPropertyName("quic")]
-    public QuicSection Quic { get; set; } = new();
-
     [JsonPropertyName("limits")]
     public LimitsSection Limits { get; set; } = new();
 
@@ -32,6 +28,16 @@ public sealed class AgentConfig
     public string? RestartAt { get; set; }
 }
 
+public sealed class ReportingSection
+{
+    /// <summary>上报主机资源占用（CPU、内存、磁盘、流量、运行时长，只有汇总数字，没有型号和 IP）。设为 false 就一个字也不上报。</summary>
+    [JsonPropertyName("host_stats")]
+    public bool HostStats { get; set; } = true;
+
+    [JsonPropertyName("host_interval_s")]
+    public int HostIntervalSeconds { get; set; } = 2;
+}
+
 public sealed class FleetSection
 {
     [JsonPropertyName("server")]
@@ -43,53 +49,24 @@ public sealed class FleetSection
     /// <summary>节点自己要求有属主（passkey）：没有就不运行。见 <c>OwnerGuard</c>。</summary>
     [JsonPropertyName("require_owner")]
     public bool RequireOwner { get; set; } = true;
-
-    /// <summary>面板至少要是哪个版本才继续连接（如 <c>0.4.2-dev.80</c>）；空 = 不要求。不够就不发任何东西，5 秒后重新握手（面板可能正在更新）。环境变量 <c>MOSHIRO_MIN_SERVER_VERSION</c> 优先。</summary>
-    [JsonPropertyName("min_server_version")]
-    public string? MinServerVersion { get; set; }
 }
 
 public sealed class ListenSection
 {
-    /// <summary>是否接受拨号方的连接。关闭时节点只拨出：不开任何端口，在 Backend 的目录里也不是拨号目标（家宽和 NAT 后的节点用这个配置）。</summary>
-    [JsonPropertyName("enabled")]
-    public bool Enabled { get; set; } = true;
-
     [JsonPropertyName("bind")]
     public string Bind { get; set; } = "::";
 
-    /// <summary>拨号方 控制通道（TLS，点对点会话）。</summary>
+    /// <summary>Probe 控制通道（TLS，点对点会话）。</summary>
     [JsonPropertyName("control_port")]
     public int ControlPort { get; set; } = 7443;
 
-    [JsonPropertyName("quic_ports")]
-    public List<int> QuicPorts { get; set; } = [7443];
+    [JsonPropertyName("udp_ports")]
+    public List<int> UdpPorts { get; set; } = [7443];
 
     [JsonPropertyName("tcp_ports")]
     public List<int> TcpPorts { get; set; } = [7444];
 }
 
-/// <summary>数据通道 QUIC 连接的传输参数；不写的项用 MsQuic 默认值。监听和拨出的连接都用这里的值。</summary>
-public sealed class QuicSection
-{
-    [JsonPropertyName("min_mtu")]
-    public int? MinMtu { get; set; }
-
-    [JsonPropertyName("max_mtu")]
-    public int? MaxMtu { get; set; }
-
-    [JsonPropertyName("pacing")]
-    public bool? Pacing { get; set; }
-
-    [JsonPropertyName("idle_timeout_ms")]
-    public int? IdleTimeoutMs { get; set; }
-
-    [JsonPropertyName("keep_alive_ms")]
-    public int? KeepAliveMs { get; set; }
-
-    public QuicSettings ToSettings() => new() { MinMtu = MinMtu, MaxMtu = MaxMtu, Pacing = Pacing, IdleTimeoutMs = IdleTimeoutMs, KeepAliveMs = KeepAliveMs };
-}
-
 public sealed class LimitsSection
 {
     [JsonPropertyName("max_total_mbps")]
@@ -98,7 +75,7 @@ public sealed class LimitsSection
     [JsonPropertyName("max_total_pps")]
     public long MaxTotalPps { get; set; } = 20_000;
 
-    /// <summary>单个 Agent 任务（外部目标）能探测的网段。默认为空：只允许公网地址，内网、回环、链路本地等一律拒绝；写了就只允许列出的网段（可以包含内网）。</summary>
+    /// <summary>默认为空：不允许执行任何目标任务。</summary>
     [JsonPropertyName("allowed_target_cidrs")]
     public List<string> AllowedTargetCidrs { get; set; } = [];
 
@@ -114,6 +91,4 @@ public sealed class PolicySection
 }
 
 [JsonSerializable(typeof(AgentConfig))]
-[JsonSerializable(typeof(DialCache))]
-[JsonSerializable(typeof(OpenWrtHealth))]
 public sealed partial class AgentConfigContext : System.Text.Json.Serialization.JsonSerializerContext;

backend/agent/src/MoShiro.Agent/Chunks/ChunkAssembler.cs

@@ -6,7 +6,7 @@ using MoShiro.Protocol;
 namespace MoShiro.Agent;
 
 /// <summary>
-/// 拨号方 的列块到达之后：找到对应的流 → 与 Agent 在数据面记录的列合并 → 盖章（块头、时钟锚点）→ 编码 → 放进发件箱（通道 16）。
+/// Probe 的列块到达之后：找到对应的流 → 与 Agent 在数据面记录的列合并 → 盖章（块头、时钟锚点）→ 编码 → 放进发件箱（通道 16）。
 /// 找不到对应流的列块（任务已经被撤销、测试已经结束）被丢弃并计数；这只会发生在配置变更的瞬间。
 /// </summary>
 public sealed class ChunkAssembler
@@ -18,7 +18,6 @@ public sealed class ChunkAssembler
     private readonly NodeId _agent;
     private readonly TimeProvider _time;
     private readonly INodeLog _log;
-    private readonly TaskTrafficReporter _traffic;
 
     public ChunkAssembler(StreamRegistry streams, MonoClock clock, AgentCounters counters, Outbox outbox, NodeId agent, TimeProvider time, INodeLog log)
     {
@@ -29,24 +28,23 @@ public sealed class ChunkAssembler
         _agent = agent;
         _time = time;
         _log = log;
-        _traffic = new TaskTrafficReporter(log, time);
     }
 
     /// <summary>
-    /// 处理某个拨号方交来的一个列块（已经解码）。返回是否被接受。
-    /// 块头里的 <c>dialer_id</c> 以认证过的会话为准（拨号方 自报的值不可信），别的拨号方的流也不会被命中。
+    /// 处理某个 Probe 交来的一个列块（已经解码）。返回是否被接受。
+    /// 块头里的 <c>probe_id</c> 以认证过的会话为准（Probe 自报的值不可信），别的 Probe 的流也不会被命中。
     /// </summary>
-    public bool Accept(RawChunk fragment, DialerSession dialer)
+    public bool Accept(RawChunk fragment, ProbeSession probe)
     {
         RawChunkHeader header = fragment.Header;
         StreamInfo? stream = header switch
         {
-            { TestId: { } test, Phase: { } phase } => _streams.FindByTest(test, phase) is { } found && ReferenceEquals(found.Owner, dialer) ? found : null,
-            { TaskId: { } task } => _streams.FindByTask(task, dialer),
+            { TestId: { } test, Phase: { } phase } => _streams.FindByTest(test, phase) is { } found && ReferenceEquals(found.Owner, probe) ? found : null,
+            { TaskId: { } task } => _streams.FindByTask(task, probe),
             _ => null,
         };
-        header.DialerId = new NodeId(dialer.DialerId);
-        header.IpVersion = dialer.IpVersion; // 地址族以认证过的控制会话为准，不接受列块自报值
+        header.ProbeId = new ProbeId(probe.ProbeId);
+        header.IpVersion = probe.IpVersion; // 地址族以认证过的控制会话为准，不接受列块自报值
         KindSpec? spec = stream is null ? null : Kinds.Find(stream.Kind);
         if (stream is null || spec is null || spec.Id != header.Kind)
         {
@@ -70,20 +68,9 @@ public sealed class ChunkAssembler
         return true;
     }
 
-    public Task RunTrafficReportingAsync(CancellationToken ct) => _traffic.RunAsync(ct);
-
-    public void FlushTraffic() => _traffic.Flush();
-
     /// <summary>Agent 自己采集的块（目标任务）直接发出。</summary>
     public void Emit(RawChunk chunk)
     {
-        byte[] payload = RawChunkCodec.Encode(chunk, out int uncompressedBytes);
-        _outbox.Enqueue(MoShiroChannels.RawChunk, chunk.Header.ChunkId.Value, payload);
-        _traffic.Record(
-            chunk.Header.TaskId?.Value ?? chunk.Header.TestId?.Value ?? "unknown",
-            chunk.Header.Kind,
-            chunk.Streams.Sum(stream => (long)stream.Count),
-            uncompressedBytes,
-            payload.Length);
+        _outbox.Enqueue(MoShiroChannels.RawChunk, chunk.Header.ChunkId.Value, RawChunkCodec.Encode(chunk));
     }
 }

backend/agent/src/MoShiro.Agent/Chunks/TaskTrafficReporter.cs

@@ -1,107 +0,0 @@
-using Fleet.Core;
-using Fleet.Rpc;
-
-namespace MoShiro.Agent;
-
-/// <summary>Aggregates per-task Agent → Backend chunk sizes into low-volume diagnostic logs.</summary>
-internal sealed class TaskTrafficReporter
-{
-    private static readonly TimeSpan s_window = TimeSpan.FromMinutes(1);
-
-    private readonly object _gate = new();
-    private readonly INodeLog _log;
-    private readonly TimeProvider _time;
-    private readonly Dictionary<(string Task, string Kind), Totals> _totals = [];
-    private long _windowStarted;
-
-    public TaskTrafficReporter(INodeLog log, TimeProvider time)
-    {
-        _log = log;
-        _time = time;
-        _windowStarted = time.GetTimestamp();
-    }
-
-    public void Record(string task, string kind, long records, int uncompressedBytes, int compressedBytes)
-    {
-        lock (_gate)
-        {
-            var key = (task, kind);
-            if (!_totals.TryGetValue(key, out Totals? totals))
-            {
-                totals = new Totals();
-                _totals.Add(key, totals);
-            }
-
-            totals.Chunks++;
-            totals.Records += records;
-            totals.UncompressedBytes += uncompressedBytes;
-            totals.CompressedBytes += compressedBytes;
-        }
-    }
-
-    public async Task RunAsync(CancellationToken ct)
-    {
-        using var timer = new PeriodicTimer(s_window, _time);
-        try
-        {
-            while (await timer.WaitForNextTickAsync(ct))
-            {
-                Flush();
-            }
-        }
-        catch (OperationCanceledException)
-        {
-            // The final snapshot is flushed by the Agent shutdown path.
-        }
-    }
-
-    public void Flush()
-    {
-        List<Snapshot> completed;
-        long windowSeconds;
-        lock (_gate)
-        {
-            if (_totals.Count == 0)
-            {
-                _windowStarted = _time.GetTimestamp();
-                return;
-            }
-
-            TimeSpan elapsed = _time.GetElapsedTime(_windowStarted);
-            windowSeconds = Math.Max(1, (long)elapsed.TotalSeconds);
-            completed = [.. _totals.Select(pair => new Snapshot(pair.Key.Task, pair.Key.Kind, pair.Value.Chunks, pair.Value.Records, pair.Value.UncompressedBytes, pair.Value.CompressedBytes))];
-            _totals.Clear();
-            _windowStarted = _time.GetTimestamp();
-        }
-
-        foreach (Snapshot item in completed)
-        {
-            double savedPercent = item.UncompressedBytes == 0
-                ? 0
-                : 100d * (item.UncompressedBytes - item.CompressedBytes) / item.UncompressedBytes;
-            long queuedBytes = item.CompressedBytes + (long)item.Chunks * OutboxFrame.HeaderBytes;
-            _log.Info(
-                "task upload traffic",
-                ("link", "agent-backend"),
-                ("task", item.Task),
-                ("kind", item.Kind),
-                ("chunks", item.Chunks),
-                ("records", item.Records),
-                ("raw_bytes", item.UncompressedBytes),
-                ("compressed_bytes", item.CompressedBytes),
-                ("queued_bytes", queuedBytes),
-                ("saved_pct", Math.Round(savedPercent, 1)),
-                ("window_s", windowSeconds));
-        }
-    }
-
-    private sealed class Totals
-    {
-        public long Chunks;
-        public long Records;
-        public long UncompressedBytes;
-        public long CompressedBytes;
-    }
-
-    private sealed record Snapshot(string Task, string Kind, long Chunks, long Records, long UncompressedBytes, long CompressedBytes);
-}

backend/agent/src/MoShiro.Agent/DataPlane/Registries.cs

@@ -34,19 +34,19 @@ public sealed class AgentCounters
     public void FragmentDiscarded() => Interlocked.Increment(ref _fragmentsDiscarded);
 }
 
-public sealed record PeerContext(DialerSession Session, MonoClock Clock);
+public sealed record PeerContext(ProbeSession Session, MonoClock Clock);
 
 /// <summary>
-/// 一个已经认证的拨号方会话：数据面的 <c>session_id</c> 和密钥（由点对点握手里的 ECDH 派生）、最近一次的 UDP 源地址、
+/// 一个已经认证的 Probe 会话：数据面的 <c>session_id</c> 和密钥（由点对点握手里的 ECDH 派生）、最近一次的 UDP 源地址、
 /// 每个流的重放窗口。数据面的包不绑定五元组，NAT 映射变化之后自动跟随新的源地址。
 /// </summary>
-public sealed class DialerSession : DataSession
+public sealed class ProbeSession : DataSession
 {
     private readonly ConcurrentDictionary<uint, ReplayWindow> _windows = new();
     private readonly ConcurrentDictionary<ulong, byte> _nonces = new();
-    private SocketAddress? _datagramSource;
+    private SocketAddress? _udpSource;
 
-    public DialerSession(PeerSessionInfo info)
+    public ProbeSession(PeerSessionInfo info)
         : base(info.Credentials.SessionIdValue, info.Credentials.Key)
     {
         Info = info;
@@ -54,20 +54,16 @@ public sealed class DialerSession : DataSession
 
     public PeerSessionInfo Info { get; }
 
-    public string DialerId => Info.PeerId;
-
-    public override string DiagnosticPeerId => DialerId;
-
-    public string? CloseReason { get; set; }
+    public string ProbeId => Info.PeerId;
 
     public string IpVersion => Info.Remote is IPEndPoint endpoint && endpoint.Address.AddressFamily == AddressFamily.InterNetworkV6 && !endpoint.Address.IsIPv4MappedToIPv6 ? IpVersions.V6 : IpVersions.V4;
 
-    public string PathKey => DialerId + "~" + IpVersion;
+    public string PathKey => ProbeId + "~" + IpVersion;
 
-    /// <summary>拨号方 的程序版本（握手里自报，只用于展示）。</summary>
+    /// <summary>Probe 的程序版本（握手里自报，只用于展示）。</summary>
     public string? Version => Info.Version;
 
-    /// <summary>拨号方 的远端地址（经过 NAT 之后的公网 IP）。</summary>
+    /// <summary>Probe 的远端地址（经过 NAT 之后的公网 IP）。</summary>
     public string? RemoteAddress => (Info.Remote as IPEndPoint)?.Address.ToString();
 
     public long ConnectedAtUtcMs { get; init; }
@@ -75,11 +71,11 @@ public sealed class DialerSession : DataSession
     public RpcSession<PeerContext>? Rpc { get; set; }
 
     /// <summary>最近一次认证通过的 UDP 包的源地址（Agent 把单向数据发往这里）。</summary>
-    public override IPEndPoint? DatagramRemoteEndPoint
+    public override IPEndPoint? UdpEndpoint
     {
         get
         {
-            SocketAddress? source = Volatile.Read(ref _datagramSource);
+            SocketAddress? source = Volatile.Read(ref _udpSource);
             return source is null ? null : (IPEndPoint)new IPEndPoint(source.Family == AddressFamily.InterNetworkV6 ? IPAddress.IPv6Any : IPAddress.Any, 0).Create(source);
         }
     }
@@ -87,7 +83,7 @@ public sealed class DialerSession : DataSession
     /// <summary>记录 UDP 源地址；NAT 映射变化后地址变了返回 true（上报 endpoint_changed）。第一次看到地址不算变化。</summary>
     public bool UpdateSource(SocketAddress source)
     {
-        SocketAddress? previous = Volatile.Read(ref _datagramSource);
+        SocketAddress? previous = Volatile.Read(ref _udpSource);
         if (previous is not null && previous.Equals(source))
         {
             return false;
@@ -100,7 +96,7 @@ public sealed class DialerSession : DataSession
             copy[i] = source[i];
         }
 
-        Volatile.Write(ref _datagramSource, copy);
+        Volatile.Write(ref _udpSource, copy);
         return previous is not null;
     }
 
@@ -133,59 +129,59 @@ public sealed class DialerSession : DataSession
     public void ForgetStream(uint streamId) => _windows.TryRemove(streamId, out _);
 }
 
-/// <summary>每个拨号方的 IPv4、IPv6 会话独立；重连只替换同地址族的旧会话。</summary>
+/// <summary>每个 Probe 的 IPv4、IPv6 会话独立；重连只替换同地址族的旧会话。</summary>
 public sealed class SessionRegistry : IDataSessions
 {
     private readonly object _gate = new();
-    private readonly Dictionary<string, DialerSession> _byDialer = new(StringComparer.Ordinal);
-    private readonly Dictionary<long, DialerSession> _bySession = [];
+    private readonly Dictionary<string, ProbeSession> _byProbe = new(StringComparer.Ordinal);
+    private readonly Dictionary<long, ProbeSession> _bySession = [];
 
-    public IReadOnlyList<DialerSession> All
+    public IReadOnlyList<ProbeSession> All
     {
         get
         {
             lock (_gate)
             {
-                return [.. _byDialer.Values];
+                return [.. _byProbe.Values];
             }
         }
     }
 
-    public DialerSession? Get(string dialerId, string ipVersion = IpVersions.V4)
+    public ProbeSession? Get(string probeId, string ipVersion = IpVersions.V4)
     {
         lock (_gate)
         {
-            return _byDialer.GetValueOrDefault(dialerId + "~" + ipVersion);
+            return _byProbe.GetValueOrDefault(probeId + "~" + ipVersion);
         }
     }
 
-    /// <summary>登记新会话；返回同一个拨号方的旧会话（如果有）。</summary>
-    public DialerSession? Replace(DialerSession session)
+    /// <summary>登记新会话；返回同一个 Probe 的旧会话（如果有）。</summary>
+    public ProbeSession? Replace(ProbeSession session)
     {
         lock (_gate)
         {
-            _byDialer.TryGetValue(session.PathKey, out DialerSession? old);
+            _byProbe.TryGetValue(session.PathKey, out ProbeSession? old);
             if (old is not null)
             {
                 _bySession.Remove(old.SessionId);
             }
 
-            _byDialer[session.PathKey] = session;
+            _byProbe[session.PathKey] = session;
             _bySession[session.SessionId] = session;
             return old;
         }
     }
 
-    public void Remove(DialerSession session)
+    public void Remove(ProbeSession session)
     {
         lock (_gate)
         {
-            if (_byDialer.TryGetValue(session.PathKey, out DialerSession? current) && ReferenceEquals(current, session))
+            if (_byProbe.TryGetValue(session.PathKey, out ProbeSession? current) && ReferenceEquals(current, session))
             {
-                _byDialer.Remove(session.PathKey);
+                _byProbe.Remove(session.PathKey);
             }
 
-            if (_bySession.TryGetValue(session.SessionId, out DialerSession? bySession) && ReferenceEquals(bySession, session))
+            if (_bySession.TryGetValue(session.SessionId, out ProbeSession? bySession) && ReferenceEquals(bySession, session))
             {
                 _bySession.Remove(session.SessionId);
             }
@@ -217,8 +213,8 @@ public sealed class StreamInfo
         KindSpec spec = Kinds.Find(kind)!;
         DeclaredPps = kind switch
         {
-            "quic_ping" or "tcp_ping" => Kinds.PeakRateHz(kind, parameters),
-            "quic_stream" => parameters.GetNumber("mbps") * 1_000_000 / 8 / Math.Max(1, parameters.GetInt("size")),
+            "udp_echo" or "tcp_echo" => parameters.GetInt("rate_hz"),
+            "udp_train" => parameters.GetNumber("mbps") * 1_000_000 / 8 / Math.Max(1, parameters.GetInt("size")),
             _ => 20,
         };
         Port = spec.Params.Any(p => p.Name == "port") ? (int)parameters.GetInt("port") : 0;
@@ -236,8 +232,8 @@ public sealed class StreamInfo
 
     public int? Phase { get; init; }
 
-    /// <summary>这个流属于哪个拨号方会话：别的会话的包即使带着这个流编号也不会被接受。</summary>
-    public DialerSession? Owner { get; init; }
+    /// <summary>这个流属于哪个 Probe 会话：别的会话的包即使带着这个流编号也不会被接受。</summary>
+    public ProbeSession? Owner { get; init; }
 
     public int Port { get; }
 
@@ -290,7 +286,7 @@ public sealed class StreamRegistry : IRecordSource
 
     public bool TryGet(uint id, out StreamInfo stream) => _streams.TryGetValue(id, out stream!);
 
-    public StreamInfo? FindByTask(TaskId task, DialerSession owner) => _streams.Values.FirstOrDefault(s => s.TaskId == task && ReferenceEquals(s.Owner, owner));
+    public StreamInfo? FindByTask(TaskId task, ProbeSession owner) => _streams.Values.FirstOrDefault(s => s.TaskId == task && ReferenceEquals(s.Owner, owner));
 
     public StreamInfo? FindByTest(TestId test, int phase) => _streams.Values.FirstOrDefault(s => s.TestId == test && s.Phase == phase);
 
@@ -314,11 +310,11 @@ public sealed class AgentAdmission : IAdmission
     }
 
     /// <summary>NAT 映射变化，源地址变了：上报 <c>endpoint_changed</c>。</summary>
-    public Action<DialerSession>? OnEndpointChanged { get; set; }
+    public Action<ProbeSession>? OnEndpointChanged { get; set; }
 
-    public Admission AdmitDatagram(DataSession session, in MeasurementDatagramHeader header, SocketAddress source, long nowUs)
+    public Admission AdmitPacket(DataSession session, in UdpHeader header, SocketAddress source, long nowUs)
     {
-        if (session is not DialerSession dialer || !_streams.TryGet(header.StreamId, out StreamInfo? stream) || !ReferenceEquals(stream.Owner, dialer) || !dialer.AcceptSequence(header.StreamId, header.Seq))
+        if (session is not ProbeSession probe || !_streams.TryGet(header.StreamId, out StreamInfo? stream) || !ReferenceEquals(stream.Owner, probe) || !probe.AcceptSequence(header.StreamId, header.Seq))
         {
             return Admission.Invalid;
         }
@@ -328,9 +324,9 @@ public sealed class AgentAdmission : IAdmission
             return Admission.RateLimited;
         }
 
-        if (dialer.UpdateSource(source))
+        if (probe.UpdateSource(source))
         {
-            OnEndpointChanged?.Invoke(dialer);
+            OnEndpointChanged?.Invoke(probe);
         }
 
         return Admission.Accepted;

backend/agent/src/MoShiro.Agent/Dial/ControlBackoff.cs

@@ -1,39 +0,0 @@
-namespace MoShiro.Agent;
-
-/// <summary>Reconnect cadence for the control connection of a single dialer-to-listener path: 1, 2, 4, 8, 16, then five 32-second waits, then once per minute.</summary>
-public sealed class ControlBackoff
-{
-    private static readonly TimeSpan[] s_delays =
-    [
-        TimeSpan.FromSeconds(1),
-        TimeSpan.FromSeconds(2),
-        TimeSpan.FromSeconds(4),
-        TimeSpan.FromSeconds(8),
-        TimeSpan.FromSeconds(16),
-        TimeSpan.FromSeconds(32),
-        TimeSpan.FromSeconds(32),
-        TimeSpan.FromSeconds(32),
-        TimeSpan.FromSeconds(32),
-        TimeSpan.FromSeconds(32),
-    ];
-
-    private const int MaxMinutes = 10;
-
-    private int _attempt;
-
-    public int Attempts => Volatile.Read(ref _attempt);
-
-    public TimeSpan Next()
-    {
-        int attempt = Interlocked.Increment(ref _attempt) - 1;
-        if (attempt < s_delays.Length)
-        {
-            return s_delays[attempt];
-        }
-
-        // 表走完还连不上（对端关了、这台机器没有这条路由）：一分钟起继续翻倍，最长十分钟，不再每分钟白白重试一次、写三条日志
-        return TimeSpan.FromMinutes(Math.Min(MaxMinutes, 1 << Math.Min(attempt - s_delays.Length, 4)));
-    }
-
-    public void Reset() => Interlocked.Exchange(ref _attempt, 0);
-}

backend/agent/src/MoShiro.Agent/Dial/DialConnection.cs

@@ -1,393 +0,0 @@
-using System.Net;
-using System.Net.Sockets;
-using System.Text.Json;
-using Fleet.Core;
-using Fleet.Identity;
-using Fleet.Link;
-using Fleet.Rpc;
-using MoShiro.Measurement;
-using MoShiro.Protocol;
-
-namespace MoShiro.Agent;
-
-/// <summary>所有连接共用的东西。</summary>
-public sealed record DialEnv(NodeIdentity Identity, string NodeId, LimitsSection Limits, QuicSettings Quic, DialDirectory Directory, MonoClock Clock, TimeProvider Time, INodeLog Log, RateChangeLog RateChanges);
-
-/// <summary>
-/// 与一个 Agent 的点对点连接，分两层：
-/// <list type="bullet">
-/// <item>控制层：TCP+TLS 长连接。认证握手 → 同步时钟 → 按 Agent 下发的 <c>line_task_set</c> 运行积木 → 每 10 秒把已完成的行切成列块放进发件箱，
-/// 发件箱和控制 RPC 共用这条连接。控制层断开才算与 Agent 失联：积木停止（数据面凭据属于控制会话），按 <see cref="ControlBackoff"/> 重拨。</item>
-/// <item>任务层：TCP 类积木各自拨号；QUIC 数据通道由 <see cref="TaskLink"/> 在控制层在线期间单独维护，断开只重拨 QUIC。</item>
-/// </list>
-/// 发件箱属于这条连接，不属于某次控制会话：控制层重连之后重发还没确认的数据块。状态与日志由 <see cref="LinkStatus"/> 统一维护。
-/// </summary>
-public sealed class DialConnection : IDisposable
-{
-    private static readonly TimeSpan s_flushInterval = TimeSpan.FromSeconds(ProtocolLimits.ChunkSeconds);
-    private static readonly TimeSpan s_clockInterval = TimeSpan.FromSeconds(30);
-
-    /// <summary>Agent 的任何帧（时钟回复、任务集）都算存活；比时钟同步间隔长得多，容得下差线路上的黑洞，但不会让一个 Agent 已经丢掉的会话永远挂着。</summary>
-    private static readonly TimeSpan s_idleLimit = TimeSpan.FromSeconds(120);
-
-    private static readonly RpcRouter<DialSession> s_router = BuildRouter();
-
-    /// <summary>对等面入站方法的路由器（只读，测试用）。</summary>
-    public static RpcRouter<DialSession> Router => s_router;
-
-    private readonly DialEnv _env;
-    private readonly LinkStatus _status;
-    private readonly Outbox _outbox = new(8 << 20);
-    private readonly SemaphoreSlim _retrySignal = new(0, 1);
-    private INodeLog _attemptLog;
-    private string _phase = "idle";
-    private long _attemptStarted;
-    private volatile DialSession? _session;
-    private volatile TaskLink? _task;
-    private QuicDatagramTransport? _activeDatagrams;
-
-    public DialConnection(DialEndpoint endpoint, DialEnv env)
-    {
-        Endpoint = endpoint;
-        _env = env;
-        _attemptLog = env.Log;
-        _status = new LinkStatus(endpoint.NodeId, endpoint.IpVersion, env.Time, env.Log);
-    }
-
-    public DialEndpoint Endpoint { get; }
-
-    /// <summary>控制层在线，且 QUIC 数据通道可发送。</summary>
-    public bool IsConnected => _session is not null && Volatile.Read(ref _activeDatagrams)?.IsConnected == true;
-
-    public int RoleCount => _session?.RoleCount ?? 0;
-
-    public long ClockOffsetUs => _session?.Clock.OffsetUs ?? 0;
-
-    /// <summary>当前线路状态（随心跳上报给面板）。</summary>
-    public LinkHealth Health() =>
-        _status.Snapshot(_outbox, _session?.TcpBlockCount ?? 0, Volatile.Read(ref _activeDatagrams)?.GetTransportSnapshot());
-
-    /// <summary>管理端手动配对：控制层在退避就马上再拨；控制层在线而 QUIC 在重拨，就跳过 QUIC 的等待。</summary>
-    public void RetryNow()
-    {
-        if (_status.Control == ControlState.Up)
-        {
-            if (_status.Quic == QuicState.Retrying)
-            {
-                _task?.RetryNow();
-            }
-
-            return;
-        }
-
-        try
-        {
-            _retrySignal.Release();
-        }
-        catch (SemaphoreFullException)
-        {
-            // 已经有一次手动重试排队。
-        }
-        catch (ObjectDisposedException)
-        {
-            // 对应线路正在被移除。
-        }
-    }
-
-    public void Dispose() => _retrySignal.Dispose();
-
-    /// <summary>控制层循环：拨号，会话结束后按退避重拨。</summary>
-    public async Task RunAsync(CancellationToken ct)
-    {
-        var backoff = new ControlBackoff();
-        while (!ct.IsCancellationRequested)
-        {
-            _attemptStarted = _env.Time.GetTimestamp();
-            _attemptLog = new ContextNodeLog(_env.Log, ("agent", Endpoint.NodeId), ("ip_version", Endpoint.IpVersion),
-                ("attempt_id", Guid.NewGuid().ToString("N")));
-            _status.Log = _attemptLog;
-            _phase = "dns";
-            _status.SetControl(ControlState.Dialing, attempt: backoff.Attempts);
-            LinkFault fault;
-            try
-            {
-                fault = await RunOnceAsync(backoff, ct);
-            }
-            catch (OperationCanceledException) when (ct.IsCancellationRequested)
-            {
-                return;
-            }
-            catch (Exception ex)
-            {
-                fault = LinkFaults.From(ex, _phase);
-            }
-
-            try
-            {
-                TimeSpan delay = backoff.Next();
-                _status.SetControl(ControlState.Backoff, fault, backoff.Attempts, delay);
-                using var waitCts = CancellationTokenSource.CreateLinkedTokenSource(ct);
-                Task delayTask = Task.Delay(delay, _env.Time, waitCts.Token);
-                Task retryTask = _retrySignal.WaitAsync(waitCts.Token);
-                Task completed = await Task.WhenAny(delayTask, retryTask);
-                await waitCts.CancelAsync();
-                if (completed == retryTask)
-                {
-                    await retryTask;
-                    _attemptLog.Info("manual link retry requested");
-                }
-                else
-                {
-                    await delayTask;
-                    try { await retryTask; }
-                    catch (OperationCanceledException) { }
-                }
-            }
-            catch (OperationCanceledException)
-            {
-                return;
-            }
-        }
-    }
-
-    /// <summary>一次控制会话：建立 → 运行 → 结束。建立失败抛出异常；建立之后结束返回结束原因。</summary>
-    private async Task<LinkFault> RunOnceAsync(ControlBackoff backoff, CancellationToken ct)
-    {
-        _attemptLog.Info("agent connection attempt started", ("phase", _phase), ("host", Endpoint.Host),
-            ("tcp_port", Endpoint.TcpPort), ("quic_ports", string.Join(',', Endpoint.QuicPorts)));
-        IPAddress address = await ResolveAsync(ct);
-        byte[] agentKey = Convert.FromBase64String(Endpoint.PublicKey);
-        (PeerSessionInfo info, TlsFrameTransport transport) = await PeerConnector.ConnectAsync(
-            address.ToString(), Endpoint.TcpPort, agentKey, _env.Identity, _env.NodeId, AgentApp.Version, 8 << 20, _env.Time, _attemptLog, ct,
-            onPhase: OnPeerPhase);
-        var data = new DataSession(info.Credentials.SessionIdValue, info.Credentials.Key);
-        var remote = (IPEndPoint)info.Remote!;
-        List<int> quicPorts = Endpoint.QuicPorts;
-        int quicPort = quicPorts.Count == 0 ? 7443 : quicPorts[0];
-        var datagrams = new QuicDatagramTransport(
-            data,
-            _env.Clock,
-            new DataPlaneCounters(),
-            _attemptLog,
-            new IPEndPoint(remote.Address, quicPort),
-            agentKey,
-            _env.Quic);
-        Volatile.Write(ref _activeDatagrams, datagrams);
-        using CancellationTokenSource cts = CancellationTokenSource.CreateLinkedTokenSource(ct);
-
-        _outbox.ResetSent();
-        var clock = new ClockEstimator();
-        var shipper = new FragmentShipper(_outbox, new NodeId(Endpoint.NodeId), new NodeId(_env.NodeId), clock, _env.Time);
-        var blockEnv = new BlockEnv
-        {
-            Clock = _env.Clock,
-            Log = _attemptLog,
-            Session = () => data,
-            Datagrams = datagrams,
-            PeerAddress = remote.Address,
-            PeerSpkiSha256 = NodeIdentity.Fingerprint(agentKey),
-            KnockKey = Knock.DeriveKey(_env.Identity, agentKey),
-            Fragments = shipper,
-            MaxPacketsPerSecond = _env.Limits.MaxTotalPps,
-            MaxBulkMbps = _env.Limits.MaxTotalMbps,
-            OnRateChange = change =>
-            {
-                change.Listener = Endpoint.NodeId;
-                change.IpVersion = Endpoint.IpVersion;
-                _env.RateChanges.Add(change);
-            },
-        };
-        var session = new DialSession(blockEnv, clock, _outbox, _env);
-        var rpc = new RpcSession<DialSession>(transport, s_router, session, _env.Time);
-        rpc.OnBinary = (_, _) => ValueTask.CompletedTask;
-
-        Task run = rpc.RunAsync(cts.Token);
-        Task pump = PumpAsync(rpc, _outbox, cts.Token);
-        Task quic = Task.CompletedTask;
-        Task maintenance = Task.CompletedTask;
-        TaskLink? taskLink = null;
-        try
-        {
-            // 先同步时钟（块头里要带偏移）；Agent 随后推送线路任务集
-            _phase = "clock_sync";
-            _status.SetControl(ControlState.Syncing);
-            await SyncClockAsync(rpc, clock, rounds: 5, cts.Token);
-            _attemptLog.Info("agent clock synchronized", ("phase", _phase), ("offset_us", clock.OffsetUs));
-
-            backoff.Reset();
-            _retrySignal.Wait(0, CancellationToken.None); // 若“立即重试”恰好赶在本次拨号成功前到达，本次成功已满足请求。
-            _session = session;
-            _phase = "control_session";
-            _status.SetControl(ControlState.Up);
-            taskLink = new TaskLink(datagrams, _status, _env.Time);
-            _task = taskLink;
-            quic = taskLink.RunAsync(cts.Token);
-            maintenance = MaintainAsync(rpc, session, clock, cts.Token);
-            _attemptLog.Info("agent control session up", ("phase", _phase), ("remote", remote), ("quic_port", quicPort), ("offset_us", clock.OffsetUs));
-            Task finished = await Task.WhenAny(run, pump, maintenance, quic);
-            LinkFault fault = finished.IsFaulted && finished.Exception?.GetBaseException() is { } error
-                ? LinkFaults.From(error, _phase)
-                : LinkFaults.Closed(_phase, finished == run
-                    ? "closed by the Agent control connection"
-                    : finished == pump
-                        ? "outbox pump stopped"
-                        : finished == quic
-                            ? "QUIC link loop stopped"
-                            : "connection maintenance stopped");
-
-            await cts.CancelAsync();
-            await BoundedAsync("stop tasks", () => Task.WhenAll(Quiet(run), Quiet(pump), Quiet(maintenance), Quiet(quic)));
-            return fault;
-        }
-        finally
-        {
-            _task = null;
-            _session = null;
-            await cts.CancelAsync();
-            session.FlushLineBlocks(); // 停止前把已稳定的行交给发件箱，重连之后补传
-            await BoundedAsync("stop blocks", session.StopAllAsync);
-            await BoundedAsync("wait control", () => Quiet(run));
-            await BoundedAsync("wait pump", () => Quiet(pump));
-            await BoundedAsync("wait quic", () => Quiet(quic));
-            taskLink?.Dispose();
-            await BoundedAsync("dispose control", () => rpc.DisposeAsync().AsTask());
-            Volatile.Write(ref _activeDatagrams, null);
-            await BoundedAsync("dispose quic", () => datagrams.DisposeAsync().AsTask());
-            _status.SetQuic(QuicState.Idle);
-        }
-    }
-
-    private async Task<IPAddress> ResolveAsync(CancellationToken ct)
-    {
-        AddressFamily family = Endpoint.IpVersion == IpVersions.V6 ? AddressFamily.InterNetworkV6 : AddressFamily.InterNetwork;
-        using var resolveTimeout = new CancellationTokenSource(TimeSpan.FromSeconds(10), _env.Time);
-        using CancellationTokenSource resolveCt = CancellationTokenSource.CreateLinkedTokenSource(ct, resolveTimeout.Token);
-        IPAddress[] addresses = IPAddress.TryParse(Endpoint.Host, out IPAddress? literal)
-            ? [literal]
-            : await Dns.GetHostAddressesAsync(Endpoint.Host, family, resolveCt.Token);
-        IPAddress address = addresses.FirstOrDefault(a => a.AddressFamily == family && !a.IsIPv4MappedToIPv6)
-            ?? throw new SocketException((int)SocketError.HostNotFound);
-        _attemptLog.Info("agent address resolved", ("phase", _phase), ("remote", address),
-            ("elapsed_ms", (long)_env.Time.GetElapsedTime(_attemptStarted).TotalMilliseconds));
-        return address;
-    }
-
-    private void OnPeerPhase(string phase)
-    {
-        _phase = phase;
-        if (phase == "peer_auth")
-        {
-            _status.SetControl(ControlState.Authenticating);
-        }
-    }
-
-    private async Task BoundedAsync(string step, Func<Task> work)
-    {
-        if (!await Bounded.WaitAsync(work(), Bounded.Teardown, _env.Time))
-        {
-            _attemptLog.Warn("agent session teardown step timed out, moving on", ("phase", _phase), ("step", step), ("timeout_ms", Bounded.Teardown.TotalMilliseconds));
-        }
-    }
-
-    private static async Task Quiet(Task task)
-    {
-        try
-        {
-            await task;
-        }
-        catch (Exception ex) when (ex is OperationCanceledException or IOException or ObjectDisposedException or System.Net.Sockets.SocketException or InvalidOperationException or TimeoutException)
-        {
-            // 会话结束时的正常噪声
-        }
-    }
-
-    /// <summary>回传数据：发件箱里的块经控制连接发给 Agent。</summary>
-    private static async Task PumpAsync(RpcSession<DialSession> rpc, Outbox outbox, CancellationToken ct)
-    {
-        while (!ct.IsCancellationRequested)
-        {
-            byte[] frame = await outbox.TakeNextAsync(2 << 20, ct);
-            await rpc.SendBinaryAsync(frame, ct);
-        }
-    }
-
-    private async Task SyncClockAsync(RpcSession<DialSession> rpc, ClockEstimator clock, int rounds, CancellationToken ct)
-    {
-        for (int i = 0; i < rounds; i++)
-        {
-            long sent = _env.Clock.NowUs();
-            Result<ClockResult> reply = await rpc.RequestAsync(MoShiroMethods.Clock, new ClockParams { DialerMonoUs = sent }, MoShiroJsonContext.Default.ClockParams, MoShiroJsonContext.Default.ClockResult, ct);
-            if (reply.TryGet(out ClockResult? value, out _))
-            {
-                clock.Add(sent, value.ListenerMonoUs, _env.Clock.NowUs());
-            }
-        }
-    }
-
-    /// <summary>切块（每 10 秒）、时钟同步（每 30 秒）和控制连接的存活检查。</summary>
-    private async Task MaintainAsync(RpcSession<DialSession> rpc, DialSession session, ClockEstimator clock, CancellationToken ct)
-    {
-        using var flush = new PeriodicTimer(s_flushInterval, _env.Time);
-        long lastClockUs = _env.Clock.NowUs();
-        while (await flush.WaitForNextTickAsync(ct))
-        {
-            if (rpc.IdleFor > s_idleLimit)
-            {
-                // 监听方已经不认这个会话（它重启、重连、或者空闲超时关了会话，而 FIN 没有到达）：拨号方这边结束它，让外层按退避重拨
-                throw new ControlSilentException($"the Agent sent nothing on this session for {(long)rpc.IdleFor.TotalSeconds}s");
-            }
-
-            session.FlushLineBlocks();
-            if (_env.Clock.NowUs() - lastClockUs >= s_clockInterval.TotalMicroseconds)
-            {
-                lastClockUs = _env.Clock.NowUs();
-                await SyncClockAsync(rpc, clock, rounds: 1, ct);
-            }
-        }
-    }
-
-    private static RpcRouter<DialSession> BuildRouter() =>
-        new RpcRouter<DialSession>(RpcSide.Dialer, MoShiroMethodTable.Table)
-            .On(
-                FleetMethods.StateSync,
-                FleetRpcContext.Default.StateSyncParams,
-                FleetRpcContext.Default.StateSyncResult,
-                async (session, sync, _) =>
-                {
-                    var result = new StateSyncResult { AppliedVersion = sync.Version };
-                    if (sync.Type == LineTaskSet.StateType)
-                    {
-                        LineTaskSet? set = sync.Doc.Deserialize(MoShiroJsonContext.Default.LineTaskSet);
-                        if (set is null)
-                        {
-                            return Result<StateSyncResult>.Fail(FleetErrors.InvalidParams, "line_task_set could not be parsed");
-                        }
-
-                        result.Rejected.AddRange(await session.ApplyAsync(set));
-                    }
-
-                    return Result<StateSyncResult>.Ok(result);
-                })
-            .On(
-                MoShiroMethods.TestPhase,
-                MoShiroJsonContext.Default.TestPhaseStartParams,
-                MoShiroJsonContext.Default.TestPhaseStartResult,
-                (session, start, _) => ValueTask.FromResult(Result<TestPhaseStartResult>.Ok(new TestPhaseStartResult { Accepted = session.StartPhase(start) })))
-            .On(
-                MoShiroMethods.TestPhaseEnd,
-                MoShiroJsonContext.Default.TestPhaseEndParams,
-                MoShiroJsonContext.Default.TestPhaseEndResult,
-                async (session, end, ct) => Result<TestPhaseEndResult>.Ok(new TestPhaseEndResult { Fragments = await session.EndPhaseAsync(end, ct) }))
-            .OnNotify(
-                MoShiroMethods.TestAbort,
-                MoShiroJsonContext.Default.TestAbortParams,
-                async (session, abort, _) => await session.AbortAsync(abort.TestId))
-            .OnNotify(
-                FleetMethods.OutboxAck,
-                FleetRpcContext.Default.OutboxAckParams,
-                (session, ack, _) =>
-                {
-                    session.Outbox.Acknowledge(ack.Ids.Concat(ack.Duplicates).Select(Guid.Parse));
-                    return ValueTask.CompletedTask;
-                });
-}

backend/agent/src/MoShiro.Agent/Dial/DialDirectory.cs

@@ -1,116 +0,0 @@
-using System.Text.Json;
-using Fleet.Core;
-using Fleet.Hosting;
-using MoShiro.Protocol;
-
-namespace MoShiro.Agent;
-
-/// <summary>
-/// 本节点要拨的监听节点列表：由面板作为期望状态（<c>dial_targets</c>）直接下发；面板连不上时用上一次保存的 <c>dial_targets.json</c>。
-/// 列表变化时用 <see cref="ConfigStore"/> 写回 <c>agents.json</c>（这是配置，不是测量数据；只在变化时写，保护路由器的闪存）。
-/// </summary>
-public sealed class DialDirectory
-{
-    public const string FileName = "dial_targets.json";
-
-    private readonly object _gate = new();
-    private readonly ConfigStore _store;
-    private readonly INodeLog _log;
-    private readonly TrustAnchor _trust;
-    private List<DialEndpoint> _targets = [];
-    private long _version;
-
-    public DialDirectory(ConfigStore store, string configDirectory, INodeLog log, TrustAnchor? trust = null)
-    {
-        _trust = trust ?? TrustAnchor.None;
-        _store = store;
-        _log = log;
-        string cache = Path.Combine(configDirectory, FileName);
-        if (File.Exists(cache))
-        {
-            try
-            {
-                DialCache saved = NodeConfig.Load(cache, AgentConfigContext.Default.DialCache);
-                _targets = saved.Targets;
-                _version = saved.Version;
-            }
-            catch (Exception ex) when (ex is IOException or InvalidDataException or JsonException or InvalidOperationException)
-            {
-                _log.Warn("dial_targets.json is unreadable, waiting for the panel", ("err", ex.Message));
-            }
-        }
-    }
-
-    /// <summary>列表变化（调和连接）。</summary>
-    public event Action? Changed;
-
-    public long Version
-    {
-        get
-        {
-            lock (_gate)
-            {
-                return _version;
-            }
-        }
-    }
-
-    public IReadOnlyList<DialEndpoint> Targets
-    {
-        get
-        {
-            lock (_gate)
-            {
-                return [.. _targets];
-            }
-        }
-    }
-
-    /// <summary>应用面板下发的列表。版本不比当前新的忽略（乱序到达的旧推送）。</summary>
-    public void Apply(DialTargets result)
-    {
-        var next = new List<DialEndpoint>();
-        _trust.Update(result.TrustChain);
-        foreach (DialTarget target in result.Targets)
-        {
-            if (!string.IsNullOrWhiteSpace(target.Host))
-            {
-                next.Add(new DialEndpoint { NodeId = target.NodeId.Value, Host = target.Host, TcpPort = target.TcpPort, QuicPorts = [.. target.QuicPorts], PublicKey = target.PublicKey });
-            }
-
-            if (!string.IsNullOrWhiteSpace(target.Ipv6Host))
-            {
-                next.Add(new DialEndpoint { NodeId = target.NodeId.Value, Host = target.Ipv6Host, TcpPort = target.TcpPort, QuicPorts = [.. target.QuicPorts], PublicKey = target.PublicKey, IpVersion = IpVersions.V6 });
-            }
-        }
-
-        lock (_gate)
-        {
-            if (result.Version < _version || (result.Version == _version && Same(next, _targets)))
-            {
-                return;
-            }
-
-            _version = result.Version;
-            _targets = next;
-        }
-
-        try
-        {
-            _store.Write(FileName, JsonSerializer.SerializeToUtf8Bytes(new DialCache { Version = result.Version, Targets = next }, AgentConfigContext.Default.DialCache));
-        }
-        catch (Exception ex) when (ex is IOException or UnauthorizedAccessException)
-        {
-            _log.Warn("could not save dial_targets.json", ("err", ex.Message));
-        }
-
-        _log.Info("dial targets updated", ("version", result.Version), ("targets", next.Count));
-        Changed?.Invoke();
-    }
-
-    private static bool Same(List<DialEndpoint> a, List<DialEndpoint> b) =>
-        a.Count == b.Count && a.Zip(b).All(p => Equal(p.First, p.Second));
-
-    public static bool Equal(DialEndpoint a, DialEndpoint b) =>
-        a.NodeId == b.NodeId && a.IpVersion == b.IpVersion && a.Host == b.Host && a.TcpPort == b.TcpPort && a.QuicPorts.SequenceEqual(b.QuicPorts) && a.PublicKey == b.PublicKey;
-}

backend/agent/src/MoShiro.Agent/Dial/DialEndpoint.cs

@@ -1,40 +0,0 @@
-using System.Text.Json.Serialization;
-using MoShiro.Protocol;
-
-namespace MoShiro.Agent;
-
-/// <summary>要拨的一个监听节点的一条路径（一个 IP 族）：地址、端口和固定证书用的公钥。</summary>
-public sealed class DialEndpoint
-{
-    [JsonPropertyName("ip_version")]
-    public string IpVersion { get; set; } = IpVersions.V4;
-
-    [JsonIgnore]
-    public string PathKey => NodeId + "~" + IpVersion;
-
-    [JsonPropertyName("node_id")]
-    public string NodeId { get; set; } = "";
-
-    [JsonPropertyName("host")]
-    public string Host { get; set; } = "";
-
-    [JsonPropertyName("tcp_port")]
-    public int TcpPort { get; set; }
-
-    [JsonPropertyName("quic_ports")]
-    public List<int> QuicPorts { get; set; } = [7443];
-
-    /// <summary>监听方的 TLS 证书公钥（SPKI，base64）：证书按它的 SHA-256 指纹固定，敲门密钥由它派生。</summary>
-    [JsonPropertyName("public_key")]
-    public string PublicKey { get; set; } = "";
-}
-
-/// <summary><c>dial_targets.json</c>：最近一次从面板同步到的拨号目标（配置，不是测量数据）；面板连不上时节点靠它继续测量。</summary>
-public sealed class DialCache
-{
-    [JsonPropertyName("version")]
-    public long Version { get; set; }
-
-    [JsonPropertyName("targets")]
-    public List<DialEndpoint> Targets { get; set; } = [];
-}

backend/agent/src/MoShiro.Agent/Dial/DialSession.cs

@@ -1,205 +0,0 @@
-using System.Collections.Concurrent;
-using Fleet.Link;
-using Fleet.Rpc;
-using MoShiro.Measurement;
-using MoShiro.Protocol;
-
-namespace MoShiro.Agent;
-
-/// <summary>一次会话内的运行状态：线路任务的积木、按需测试的阶段。方法由路由器在单个入站队列里顺序调用。</summary>
-public sealed class DialSession
-{
-    private const int MaxBlocks = 16;
-
-    private readonly object _gate = new();
-    private readonly Dictionary<string, RunningBlock> _lineBlocks = [];
-    private readonly Dictionary<string, string> _lineKinds = [];
-    private readonly ConcurrentDictionary<uint, PhaseState> _phases = new();
-
-    public DialSession(BlockEnv blockEnv, ClockEstimator clock, Outbox outbox, DialEnv env)
-    {
-        BlockEnv = blockEnv;
-        Clock = clock;
-        Outbox = outbox;
-        Env = env;
-    }
-
-    public BlockEnv BlockEnv { get; }
-
-    public ClockEstimator Clock { get; }
-
-    public Outbox Outbox { get; }
-
-    public DialEnv Env { get; }
-
-    public int RoleCount
-    {
-        get
-        {
-            lock (_gate)
-            {
-                return _lineBlocks.Count;
-            }
-        }
-    }
-
-    /// <summary>正在运行的 TCP 类线路积木数（它们自带套接字和重拨，不依赖 QUIC 数据通道）。</summary>
-    public int TcpBlockCount
-    {
-        get
-        {
-            lock (_gate)
-            {
-                return _lineKinds.Values.Count(IsTcpKind);
-            }
-        }
-    }
-
-    private static bool IsTcpKind(string kind) => kind is "tcp_ping" or "tcp_connect" or "tls_hold" or "tcp_transfer";
-
-    /// <summary>调和线路任务：参数没有变化的积木继续运行，其余启动或停止。返回被拒绝的任务（不认识的 Kind、参数非法、超过上限）。</summary>
-    public async Task<List<RejectedItem>> ApplyAsync(LineTaskSet set)
-    {
-        var rejected = new List<RejectedItem>();
-        var wanted = new Dictionary<string, BlockSpec>();
-        foreach (LineTask task in set.Tasks)
-        {
-            BlockSpec? spec = BlockCatalog.DialerKinds.Contains(task.Kind)
-                ? BlockSpec.Create(task.Kind, task.Params, task.StreamId, task.Port, new FragmentScope(task.Id, null, null))
-                : null;
-            if (spec is null)
-            {
-                rejected.Add(new RejectedItem { Id = task.Id.Value, Reason = BlockCatalog.DialerKinds.Contains(task.Kind) ? "invalid_params" : "unsupported" });
-            }
-            else if (wanted.Count >= MaxBlocks)
-            {
-                rejected.Add(new RejectedItem { Id = task.Id.Value, Reason = "budget_exceeded" });
-            }
-            else
-            {
-                wanted[task.Id.Value] = spec;
-            }
-        }
-
-        var stopped = new List<RunningBlock>();
-        lock (_gate)
-        {
-            foreach ((string id, RunningBlock running) in _lineBlocks.ToList())
-            {
-                if (!wanted.TryGetValue(id, out BlockSpec? spec) || spec.Signature != running.Signature)
-                {
-                    stopped.Add(running);
-                    _lineBlocks.Remove(id);
-                    _lineKinds.Remove(id);
-                }
-            }
-
-            foreach ((string id, BlockSpec spec) in wanted)
-            {
-                if (!_lineBlocks.ContainsKey(id))
-                {
-                    _lineBlocks[id] = RunningBlock.Start(BlockCatalog.Find(spec.KindId)!.Dialer(BlockEnv, spec), spec.Signature);
-                    _lineKinds[id] = spec.KindId;
-                }
-            }
-        }
-
-        // 被停止的积木把已经稳定的行交出去（刚发出、还在等回复的行不交：它们不是丢失，只是没有等到）
-        foreach (RunningBlock running in stopped)
-        {
-            await running.StopAsync();
-            running.Block.Flush(FlushMode.Settled);
-        }
-
-        return rejected;
-    }
-
-    public void FlushLineBlocks()
-    {
-        List<RunningBlock> blocks;
-        lock (_gate)
-        {
-            blocks = [.. _lineBlocks.Values];
-        }
-
-        foreach (RunningBlock running in blocks)
-        {
-            running.Block.Flush(FlushMode.Settled);
-        }
-    }
-
-    /// <summary>按需测试的一个阶段：起对应的拨号方积木（单向数据流是接收端）。超过本地上限（或不认识）则拒绝。</summary>
-    public bool StartPhase(TestPhaseStartParams start)
-    {
-        string? direction = start.Params.TryGetValue("direction", out System.Text.Json.JsonElement d) ? d.GetString() : null;
-        if (_phases.ContainsKey(start.StreamId) || !BlockCatalog.DialerKinds.Contains(start.Kind) || BlockCatalog.Find(start.Kind)!.Supports(direction) is false)
-        {
-            return false;
-        }
-
-        if (start.Kind == "quic_stream")
-        {
-            double mbps = start.Params["mbps"].GetDouble();
-            long size = start.Params["size"].GetInt64();
-            long pps = (long)(mbps * 1_000_000 / 8 / Math.Max(1, size));
-            if (mbps > Env.Limits.MaxTotalMbps || pps > Env.Limits.MaxTotalPps)
-            {
-                return false;
-            }
-        }
-
-        BlockSpec? spec = BlockSpec.Create(start.Kind, start.Params, start.StreamId, start.Port, new FragmentScope(null, start.TestId, start.Phase));
-        if (spec is null)
-        {
-            return false;
-        }
-
-        _phases[start.StreamId] = new PhaseState(start.TestId, start.Kind, RunningBlock.Start(BlockCatalog.Find(start.Kind)!.Dialer(BlockEnv, spec), spec.Signature));
-        return true;
-    }
-
-    /// <summary>阶段结束：收尾、交出列块，返回块数。</summary>
-    public async Task<int> EndPhaseAsync(TestPhaseEndParams end, CancellationToken ct)
-    {
-        if (!_phases.TryRemove(end.StreamId, out PhaseState? phase))
-        {
-            return 0;
-        }
-
-        // 最后一批包（单向数据流）、最后几个回显的回复还在途中
-        await Task.Delay(TimeSpan.FromMilliseconds(phase.Kind == "quic_stream" ? 200 : 300), BlockEnv.Time, ct);
-        await phase.Running.StopAsync();
-        return phase.Running.Block.Flush(FlushMode.All, end.SentCount);
-    }
-
-    public async Task AbortAsync(TestId test)
-    {
-        foreach ((uint stream, PhaseState phase) in _phases.ToList())
-        {
-            if (phase.Test == test && _phases.TryRemove(stream, out _))
-            {
-                await phase.Running.StopAsync();
-            }
-        }
-    }
-
-    public async Task StopAllAsync()
-    {
-        List<RunningBlock> blocks;
-        lock (_gate)
-        {
-            blocks = [.. _lineBlocks.Values];
-            _lineBlocks.Clear();
-            _lineKinds.Clear();
-        }
-
-        blocks.AddRange(_phases.Values.Select(p => p.Running));
-        _phases.Clear();
-        foreach (RunningBlock running in blocks)
-        {
-            await running.StopAsync();
-        }
-    }
-
-    private sealed record PhaseState(TestId Test, string Kind, RunningBlock Running);
-}

backend/agent/src/MoShiro.Agent/Dial/DialerHost.cs

@@ -1,203 +0,0 @@
-using System.Collections.Concurrent;
-using Fleet.Core;
-using MoShiro.Protocol;
-
-namespace MoShiro.Agent;
-
-/// <summary>
-/// 拨号侧的组装点：每个拨号目标（监听节点 × IP 族）一个 <see cref="DialConnection"/>。目标列表变化时只调和连接：
-/// 新增的建立，移除的断开，地址或证书指纹变化的重建，其余不受影响。
-/// </summary>
-public sealed class DialerHost : IDisposable
-{
-    private readonly DialEnv _env;
-    private readonly Dictionary<string, Running> _connections = [];
-    private readonly ConcurrentDictionary<string, long> _pendingRetries = new(StringComparer.Ordinal);
-    private readonly SemaphoreSlim _changed = new(0);
-    private readonly TimeProvider _time;
-    private readonly INodeLog _log;
-
-    public DialerHost(DialEnv env, TimeProvider time, INodeLog log)
-    {
-        _env = env;
-        _time = time;
-        _log = log;
-        env.Directory.Changed += () => _changed.Release();
-    }
-
-    public DialDirectory Directory => _env.Directory;
-
-    /// <summary>某条路径的控制会话是否已经建立、QUIC 数据通道是否可发送（测试和状态查看用）。</summary>
-    public bool IsConnected(string listenerId, string ipVersion = IpVersions.V4)
-    {
-        lock (_connections)
-        {
-            return _connections.TryGetValue(listenerId + "~" + ipVersion, out Running? running) && running.Connection.IsConnected;
-        }
-    }
-
-    public DialConnection? Connection(string listenerId, string ipVersion = IpVersions.V4)
-    {
-        lock (_connections)
-        {
-            return _connections.TryGetValue(listenerId + "~" + ipVersion, out Running? running) ? running.Connection : null;
-        }
-    }
-
-    /// <summary>每条路径当前的线路健康（随心跳上报）。</summary>
-    public List<LinkHealth> SnapshotLinks()
-    {
-        DialConnection[] connections;
-        lock (_connections)
-        {
-            connections = [.. _connections.Values.Select(r => r.Connection)];
-        }
-
-        return [.. connections.Select(c => c.Health())];
-    }
-
-    /// <summary>管理端手动重试：跳过这个监听节点所有路径的退避。目标列表还没到时，30 秒内到达就补上。</summary>
-    public void RetryListener(string listenerId, bool queueIfMissing = true)
-    {
-        DialConnection[] matches;
-        lock (_connections)
-        {
-            matches = [.. _connections.Values.Select(r => r.Connection).Where(c => c.Endpoint.NodeId == listenerId)];
-        }
-
-        if (matches.Length == 0)
-        {
-            if (queueIfMissing)
-            {
-                _pendingRetries[listenerId] = _time.GetUtcNow().AddSeconds(30).ToUnixTimeMilliseconds();
-            }
-
-            _log.Warn("manual link retry requested but no matching dial target is configured", ("listener", listenerId));
-            return;
-        }
-
-        _pendingRetries.TryRemove(listenerId, out _);
-        foreach (DialConnection connection in matches)
-        {
-            connection.RetryNow();
-        }
-    }
-
-    public async Task RunAsync(CancellationToken ct)
-    {
-        try
-        {
-            while (!ct.IsCancellationRequested)
-            {
-                await ReconcileAsync(_env.Directory.Targets, ct);
-                RetryPending();
-                await _changed.WaitAsync(ct);
-            }
-        }
-        catch (OperationCanceledException)
-        {
-            // 停止
-        }
-        finally
-        {
-            await ReconcileAsync([], CancellationToken.None);
-        }
-    }
-
-    public void Dispose() => _changed.Dispose();
-
-    private void RetryPending()
-    {
-        long now = _time.GetUtcNow().ToUnixTimeMilliseconds();
-        foreach ((string listenerId, long expiresAt) in _pendingRetries.ToArray())
-        {
-            if (expiresAt <= now)
-            {
-                _pendingRetries.TryRemove(listenerId, out _);
-            }
-            else
-            {
-                RetryListener(listenerId, queueIfMissing: false);
-            }
-        }
-    }
-
-    private async Task ReconcileAsync(IReadOnlyList<DialEndpoint> targets, CancellationToken ct)
-    {
-        var wanted = targets.ToDictionary(t => t.PathKey);
-        var stopped = new List<Running>();
-        lock (_connections)
-        {
-            foreach ((string id, Running running) in _connections.ToList())
-            {
-                if (!wanted.TryGetValue(id, out DialEndpoint? endpoint) || !DialDirectory.Equal(endpoint, running.Connection.Endpoint))
-                {
-                    stopped.Add(running);
-                    _connections.Remove(id);
-                }
-            }
-        }
-
-        foreach (Running running in stopped)
-        {
-            await running.StopAsync();
-        }
-
-        if (ct.IsCancellationRequested)
-        {
-            return;
-        }
-
-        lock (_connections)
-        {
-            foreach ((string id, DialEndpoint endpoint) in wanted)
-            {
-                if (!_connections.ContainsKey(id))
-                {
-                    _connections[id] = Running.Start(new DialConnection(endpoint, _env));
-                }
-            }
-        }
-    }
-
-    private sealed class Running : IDisposable
-    {
-        private readonly CancellationTokenSource _cts = new();
-        private Task _run = Task.CompletedTask;
-
-        private Running(DialConnection connection)
-        {
-            Connection = connection;
-        }
-
-        public DialConnection Connection { get; }
-
-        public static Running Start(DialConnection connection)
-        {
-            var running = new Running(connection);
-            running._run = Task.Run(() => connection.RunAsync(running._cts.Token), CancellationToken.None);
-            return running;
-        }
-
-        public async Task StopAsync()
-        {
-            await _cts.CancelAsync();
-            try
-            {
-                await _run.WaitAsync(TimeSpan.FromSeconds(10));
-            }
-            catch (Exception ex) when (ex is OperationCanceledException or TimeoutException)
-            {
-                // 没有及时退出：不再等
-            }
-
-            Dispose();
-        }
-
-        public void Dispose()
-        {
-            Connection.Dispose();
-            _cts.Dispose();
-        }
-    }
-}

backend/agent/src/MoShiro.Agent/Dial/LinkFaults.cs

@@ -1,44 +0,0 @@
-using System.Net.Sockets;
-using System.Security.Authentication;
-using MoShiro.Protocol;
-
-namespace MoShiro.Agent;
-
-/// <summary>控制连接在已建立之后停止应答：对端不认这个会话了，或者线路中断超过了空闲上限。</summary>
-public sealed class ControlSilentException(string message) : IOException(message);
-
-/// <summary>把连接各阶段的异常归类成上报给面板的 <see cref="LinkFault"/>：失败发生在哪个阶段、属于哪一类。</summary>
-public static class LinkFaults
-{
-    private const int MaxMessage = 200;
-
-    public static LinkFault From(Exception ex, string phase)
-    {
-        Exception inner = ex is AggregateException { InnerExceptions.Count: 1 } aggregate ? aggregate.InnerExceptions[0] : ex;
-        string kind = inner switch
-        {
-            ControlSilentException => "silent",
-            OperationCanceledException or TimeoutException => "timeout",
-            AuthenticationException => "tls",
-            SocketException socket => Kind(socket.SocketErrorCode),
-            IOException { InnerException: SocketException socket } => Kind(socket.SocketErrorCode),
-            InvalidOperationException invalid when invalid.Message.Contains("authentication", StringComparison.OrdinalIgnoreCase)
-                || invalid.Message.Contains("challenge", StringComparison.OrdinalIgnoreCase) => "auth",
-            IOException => "closed",
-            _ => "other",
-        };
-        string message = kind == "timeout" && inner is OperationCanceledException ? $"{phase} did not finish in time" : inner.Message;
-        return new LinkFault { Phase = phase, Kind = kind, Message = message.Length > MaxMessage ? message[..MaxMessage] : message };
-    }
-
-    public static LinkFault Closed(string phase, string message) => new() { Phase = phase, Kind = "closed", Message = message };
-
-    private static string Kind(SocketError error) => error switch
-    {
-        SocketError.ConnectionRefused => "refused",
-        SocketError.ConnectionReset or SocketError.ConnectionAborted => "reset",
-        SocketError.HostNotFound or SocketError.NoData or SocketError.TryAgain => "dns",
-        SocketError.TimedOut => "timeout",
-        _ => "other",
-    };
-}

backend/agent/src/MoShiro.Agent/Dial/LinkStatus.cs

@@ -1,247 +0,0 @@
-using Fleet.Core;
-using Fleet.Link;
-using MoShiro.Measurement;
-using MoShiro.Protocol;
-
-namespace MoShiro.Agent;
-
-/// <summary>
-/// 一条线路（本节点拨号到一个监听节点的一个 IP 族）的状态：控制层和 QUIC 数据通道各一个状态机，状态迁移只在这里发生、只在这里记日志。
-/// 控制层每次迁移写一条 <c>link state</c>；数据通道只在“可用 ↔ 不可用”的边界上写，重拨过程中的失败只计数。
-/// </summary>
-public sealed class LinkStatus
-{
-    private const int MaxHandshakeSamples = 128;
-
-    private readonly object _gate = new();
-    private readonly string _listenerId;
-    private readonly string _ipVersion;
-    private readonly TimeProvider _time;
-    private readonly Queue<double> _handshakeMs = new();
-
-    private ControlState _control = ControlState.Dialing;
-    private long _controlSince;
-    private LinkFault? _controlFault;
-    private int _controlAttempt;
-    private long? _controlRetryAt;
-
-    private QuicState _quic = QuicState.Idle;
-    private long _quicSince;
-    private long? _quicDownSince;
-    private LinkFault? _quicFault;
-    private int _quicAttempt;
-    private long? _quicRetryAt;
-
-    private long _handshakeAttempts;
-    private long _handshakeSuccesses;
-    private long _handshakeFailures;
-    private long _handshakeTimeouts;
-
-    public LinkStatus(string listenerId, string ipVersion, TimeProvider time, INodeLog log)
-    {
-        _listenerId = listenerId;
-        _ipVersion = ipVersion;
-        _time = time;
-        Log = log;
-        _controlSince = _quicSince = Now();
-    }
-
-    /// <summary>当前这次尝试的日志（带 attempt_id 等上下文）；控制层每次重拨前由连接更新。</summary>
-    public INodeLog Log { get; set; }
-
-    public ControlState Control
-    {
-        get
-        {
-            lock (_gate)
-            {
-                return _control;
-            }
-        }
-    }
-
-    public QuicState Quic
-    {
-        get
-        {
-            lock (_gate)
-            {
-                return _quic;
-            }
-        }
-    }
-
-    public bool Ready => Control == ControlState.Up && Quic == QuicState.Ready;
-
-    public void SetControl(ControlState state, LinkFault? fault = null, int attempt = 0, TimeSpan? retryIn = null)
-    {
-        ControlState from;
-        long inStateMs;
-        lock (_gate)
-        {
-            from = _control;
-            long now = Now();
-            inStateMs = now - _controlSince;
-            if (state != from)
-            {
-                _controlSince = now;
-            }
-
-            _control = state;
-            _controlFault = state == ControlState.Up ? null : fault ?? _controlFault;
-            _controlAttempt = attempt;
-            _controlRetryAt = retryIn is { } delay ? now + (long)delay.TotalMilliseconds : null;
-        }
-
-        if (state == from)
-        {
-            return;
-        }
-
-        string reason = state == ControlState.Backoff && fault is not null ? fault.ToString() : "";
-        Log.Write(state == ControlState.Backoff ? NodeLogLevel.Warn : NodeLogLevel.Info, "link state",
-        [
-            ("layer", "control"), ("listener", _listenerId), ("ip_version", _ipVersion), ("from", Name(from)), ("to", Name(state)),
-            ("in_state_ms", inStateMs), ("attempt", attempt), ("retry_in_ms", retryIn is { } r ? (long)r.TotalMilliseconds : null),
-            ("reason", reason.Length == 0 ? null : reason),
-        ]);
-    }
-
-    public void SetQuic(QuicState state, LinkFault? fault = null, int attempt = 0, TimeSpan? retryIn = null)
-    {
-        QuicState from;
-        long? downMs = null;
-        bool logged;
-        lock (_gate)
-        {
-            from = _quic;
-            long now = Now();
-            if (state != from)
-            {
-                _quicSince = now;
-            }
-
-            if (state == QuicState.Ready)
-            {
-                downMs = _quicDownSince is { } down ? now - down : null;
-                _quicDownSince = null;
-                _quicFault = null;
-            }
-            else if (state is QuicState.Retrying or QuicState.Connecting)
-            {
-                _quicDownSince ??= now;
-                _quicFault = fault ?? _quicFault;
-            }
-            else
-            {
-                _quicDownSince = null;
-            }
-
-            // 数据通道抖动时不逐次打印：恢复可用、首次失败时各一条，重拨中的失败只计数。
-            logged = state == QuicState.Ready || (state == QuicState.Retrying && (from == QuicState.Ready || attempt == 1));
-            _quic = state;
-            _quicAttempt = attempt;
-            _quicRetryAt = retryIn is { } delay ? now + (long)delay.TotalMilliseconds : null;
-        }
-
-        if (logged)
-        {
-            Log.Write(state == QuicState.Retrying ? NodeLogLevel.Warn : NodeLogLevel.Info, "link state",
-            [
-                ("layer", "task_quic"), ("listener", _listenerId), ("ip_version", _ipVersion), ("from", Name(from)), ("to", Name(state)),
-                ("down_ms", downMs), ("attempt", attempt), ("retry_in_ms", retryIn is { } r ? (long)r.TotalMilliseconds : null),
-                ("reason", state == QuicState.Retrying ? fault?.ToString() : null),
-            ]);
-        }
-    }
-
-    /// <summary>记录一次 QUIC 握手的结果。</summary>
-    public void Handshake(bool succeeded, double? elapsedMs, bool timedOut)
-    {
-        lock (_gate)
-        {
-            _handshakeAttempts++;
-            if (succeeded)
-            {
-                _handshakeSuccesses++;
-                if (elapsedMs is { } ms)
-                {
-                    _handshakeMs.Enqueue(ms);
-                    while (_handshakeMs.Count > MaxHandshakeSamples)
-                    {
-                        _handshakeMs.Dequeue();
-                    }
-                }
-            }
-            else
-            {
-                _handshakeFailures++;
-                if (timedOut)
-                {
-                    _handshakeTimeouts++;
-                }
-            }
-        }
-    }
-
-    public LinkHealth Snapshot(Outbox outbox, int tcpBlocks, QuicTransportSnapshot? transport)
-    {
-        lock (_gate)
-        {
-            long now = Now();
-            double? p95 = null;
-            if (_handshakeMs.Count > 0)
-            {
-                double[] sorted = [.. _handshakeMs.Order()];
-                p95 = sorted[(int)Math.Ceiling(sorted.Length * 0.95) - 1];
-            }
-
-            return new LinkHealth
-            {
-                ListenerId = _listenerId,
-                IpVersion = _ipVersion,
-                Control = new ControlHealth
-                {
-                    State = _control,
-                    SinceMs = _controlSince,
-                    LastError = _controlFault,
-                    RetryAttempt = _controlAttempt,
-                    RetryInMs = _control == ControlState.Backoff ? RetryIn(_controlRetryAt, now) : null,
-                    OutboxBytes = outbox.PendingBytes,
-                    OutboxPending = outbox.PendingCount,
-                },
-                Task = new TaskHealth
-                {
-                    TcpBlocks = tcpBlocks,
-                    Quic = new QuicHealth
-                    {
-                        State = _quic,
-                        SinceMs = _quicSince,
-                        LastError = _quicFault,
-                        RetryAttempt = _quicAttempt,
-                        RetryInMs = _quic == QuicState.Retrying ? RetryIn(_quicRetryAt, now) : null,
-                        HandshakeAttempts = _handshakeAttempts,
-                        HandshakeSuccesses = _handshakeSuccesses,
-                        HandshakeFailures = _handshakeFailures,
-                        HandshakeTimeouts = _handshakeTimeouts,
-                        HandshakeP95Ms = p95,
-                        HandshakeSampleCount = _handshakeMs.Count,
-                        TransportPacketLossRate = transport?.PacketLossRate,
-                        TransportPacketsSent = transport?.PacketsSent,
-                        TransportCongestionEvents = transport?.CongestionEvents,
-                        TransportEcnCongestionEvents = transport?.EcnCongestionEvents,
-                        PathMtu = transport?.PathMtu,
-                    },
-                },
-            };
-        }
-    }
-
-    private static long? RetryIn(long? at, long now) => at is { } t ? Math.Max(0, t - now) : null;
-
-    private long Now() => _time.GetUtcNow().ToUnixTimeMilliseconds();
-
-    private static string Name(ControlState state) => state.ToString().ToLowerInvariant();
-
-    private static string Name(QuicState state) => state.ToString().ToLowerInvariant();
-}

backend/agent/src/MoShiro.Agent/Dial/QuicRedialBackoff.cs

@@ -1,31 +0,0 @@
-namespace MoShiro.Agent;
-
-/// <summary>
-/// QUIC 数据通道的重拨节奏：1、2、3、5 秒，之后固定 5 秒。数据通道断开期间的测量是缺口，所以不像控制连接那样拉长到一分钟。
-/// 抖动由调用方通过 <paramref name="random"/> 给出，避免多条线路同时重拨。
-/// </summary>
-public sealed class QuicRedialBackoff
-{
-    private static readonly TimeSpan[] s_delays = [TimeSpan.FromSeconds(1), TimeSpan.FromSeconds(2), TimeSpan.FromSeconds(3), TimeSpan.FromSeconds(5)];
-
-    private const int ShortAttempts = 12;
-
-    private int _attempt;
-
-    public int Attempts => _attempt;
-
-    public TimeSpan Next(Random? random = null)
-    {
-        TimeSpan delay = s_delays[Math.Min(_attempt, s_delays.Length - 1)];
-        if (_attempt >= ShortAttempts)
-        {
-            // 短间隔已经连续失败了这么多次：这条路径大概率是死的（对端不通、被丢弃），握手和日志都是白费，翻倍退避，最长五分钟
-            delay = TimeSpan.FromSeconds(Math.Min(300, 5 * (1 << Math.Min(_attempt - ShortAttempts + 1, 6))));
-        }
-
-        _attempt++;
-        return random is null ? delay : delay * (0.8 + (0.4 * random.NextDouble()));
-    }
-
-    public void Reset() => _attempt = 0;
-}

backend/agent/src/MoShiro.Agent/Dial/TaskLink.cs

@@ -1,162 +0,0 @@
-using Fleet.Core;
-using MoShiro.Measurement;
-using MoShiro.Protocol;
-
-namespace MoShiro.Agent;
-
-/// <summary>
-/// 任务层的 QUIC 数据通道：控制层在线期间在后台维持一条 QUIC 连接，断开后按 <see cref="QuicRedialBackoff"/> 重拨。
-/// 重拨期间控制连接和 TCP 类积木不受影响；UDP 类积木的发送在通道不可用时按各自的规则处理。
-/// </summary>
-public sealed class TaskLink : IDisposable
-{
-    /// <summary>QUIC 连接可用至少这么久才算稳定，重拨节奏随之复位。刚连上就断的通道不会把间隔重置回 1 秒。</summary>
-    private static readonly TimeSpan s_stableAfter = TimeSpan.FromSeconds(10);
-
-    /// <summary>一直在发测量报文、却这么久没有收到任何一个：QUIC 路径已经是黑洞，主动关闭重拨。</summary>
-    private static readonly TimeSpan s_silentLimit = TimeSpan.FromSeconds(30);
-
-    private static readonly TimeSpan s_watchInterval = TimeSpan.FromSeconds(5);
-    private static readonly TimeSpan s_summaryInterval = TimeSpan.FromSeconds(60);
-
-    private readonly QuicDatagramTransport _datagrams;
-    private readonly LinkStatus _status;
-    private readonly TimeProvider _time;
-    private readonly SemaphoreSlim _retrySignal = new(0, 1);
-
-    public TaskLink(QuicDatagramTransport datagrams, LinkStatus status, TimeProvider time)
-    {
-        _datagrams = datagrams;
-        _status = status;
-        _time = time;
-    }
-
-    public void Dispose() => _retrySignal.Dispose();
-
-    /// <summary>管理端手动重试：跳过当前的重拨等待。</summary>
-    public void RetryNow()
-    {
-        try
-        {
-            _retrySignal.Release();
-        }
-        catch (SemaphoreFullException)
-        {
-            // 已有一次排队
-        }
-    }
-
-    public async Task RunAsync(CancellationToken ct)
-    {
-        var backoff = new QuicRedialBackoff();
-        while (!ct.IsCancellationRequested)
-        {
-            _status.SetQuic(QuicState.Connecting, attempt: backoff.Attempts);
-            LinkFault fault;
-            long readyAt = 0;
-            try
-            {
-                try
-                {
-                    await _datagrams.ConnectAsync(ct);
-                }
-                catch (Exception ex) when (!ct.IsCancellationRequested)
-                {
-                    _status.Handshake(false, null, ex is TimeoutException or OperationCanceledException);
-                    throw;
-                }
-
-                _status.Handshake(true, _datagrams.HandshakeDurationMs, false);
-                readyAt = _time.GetTimestamp();
-                _status.SetQuic(QuicState.Ready);
-                fault = await WatchAsync(ct);
-            }
-            catch (OperationCanceledException) when (ct.IsCancellationRequested)
-            {
-                return;
-            }
-            catch (Exception ex)
-            {
-                fault = LinkFaults.From(ex, "quic_handshake");
-            }
-
-            await _datagrams.CloseClientAsync();
-            if (readyAt != 0 && _time.GetElapsedTime(readyAt) >= s_stableAfter)
-            {
-                backoff.Reset();
-            }
-
-            TimeSpan delay = backoff.Next(Random.Shared);
-            _status.SetQuic(QuicState.Retrying, fault, backoff.Attempts, delay);
-            try
-            {
-                using var wait = CancellationTokenSource.CreateLinkedTokenSource(ct);
-                Task sleep = Task.Delay(delay, _time, wait.Token);
-                Task manual = _retrySignal.WaitAsync(wait.Token);
-                await Task.WhenAny(sleep, manual);
-                await wait.CancelAsync();
-                await Task.WhenAll(Settle(sleep), Settle(manual));
-            }
-            catch (OperationCanceledException) when (ct.IsCancellationRequested)
-            {
-                return;
-            }
-        }
-    }
-
-    /// <summary>连接可用期间的看护：连接关闭（对端关闭、空闲超时）或数据面静默时返回原因。</summary>
-    private async Task<LinkFault> WatchAsync(CancellationToken ct)
-    {
-        using var timer = new PeriodicTimer(s_watchInterval, _time);
-        Task closed = _datagrams.WaitForDisconnectAsync(ct);
-        long seenIn = _datagrams.Counters.PacketsIn;
-        long seenOut = _datagrams.Counters.PacketsOut;
-        long inChangedAt = _time.GetTimestamp();
-        long lastSummaryAt = inChangedAt;
-        while (true)
-        {
-            Task tick = timer.WaitForNextTickAsync(ct).AsTask();
-            if (await Task.WhenAny(closed, tick) == closed)
-            {
-                await closed;
-                return LinkFaults.Closed("quic_data", "QUIC connection closed");
-            }
-
-            await tick;
-            long packetsIn = _datagrams.Counters.PacketsIn;
-            long packetsOut = _datagrams.Counters.PacketsOut;
-            long now = _time.GetTimestamp();
-            if (packetsIn != seenIn)
-            {
-                seenIn = packetsIn;
-                inChangedAt = now;
-            }
-            else if (packetsOut != seenOut && _time.GetElapsedTime(inChangedAt, now) > s_silentLimit)
-            {
-                return new LinkFault { Phase = "quic_data", Kind = "silent", Message = $"no datagram received for {(long)_time.GetElapsedTime(inChangedAt, now).TotalSeconds}s while sending" };
-            }
-
-            seenOut = packetsOut;
-            if (_time.GetElapsedTime(lastSummaryAt, now) >= s_summaryInterval)
-            {
-                lastSummaryAt = now;
-                QuicTransportSnapshot? transport = _datagrams.GetTransportSnapshot();
-                _status.Log.Info("agent measurement summary", ("layer", "task_quic"), ("packets_in", packetsIn), ("packets_out", packetsOut),
-                    ("dropped_invalid", _datagrams.Counters.DroppedInvalid), ("dropped_rate_limited", _datagrams.Counters.DroppedRateLimited),
-                    ("transport_packets_sent", transport?.PacketsSent), ("path_mtu", transport?.PathMtu));
-            }
-        }
-    }
-
-    private static async Task Settle(Task task)
-    {
-        try
-        {
-            await task;
-        }
-        catch (OperationCanceledException)
-        {
-            // 等待被另一条路径或停止打断
-        }
-    }
-}

backend/agent/src/MoShiro.Agent/Host/HostSampler.cs

@@ -1,13 +1,13 @@
 using System.Globalization;
 using MoShiro.Protocol;
 
-namespace MoShiro.Measurement;
+namespace MoShiro.Agent;
 
 /// <summary>
-/// 拨号方和监听方共用。主机资源占用的只读快照：CPU 占用、内存、根分区、网卡累计流量与瞬时网速、系统运行时长。
+/// 主机资源占用的只读快照：CPU 占用、内存、根分区、网卡累计流量与瞬时网速、系统运行时长。
 /// 只有汇总的数字：<b>不采集 CPU 型号、主机名、网卡名和任何 IP 地址</b>，所以面板和展示页想泄漏也泄漏不出去。
 /// Linux 读 <c>/proc</c> 和 <c>/sys</c>；容器通过 HOST_PROC、HOST_SYS、HOST_ROOT 指定宿主路径。
-/// Windows 用 GetSystemTimes、GlobalMemoryStatusEx 和 NetworkInterface（见 WindowsHost）；macOS 等其他系统只有磁盘和运行时长。读不到的项是 null。不执行任何命令，不写文件。
+/// 读不到的项是 null（例如 Windows 上的 CPU）。不执行任何命令，不写文件。
 /// </summary>
 public sealed class HostSampler
 {
@@ -18,7 +18,6 @@ public sealed class HostSampler
     private readonly Func<(long Total, long Used)?> _disk;
     private readonly string _procRoot;
     private readonly string _sysRoot;
-    private readonly bool _native;
     private (long Total, long Idle)? _previousCpu;
     private (long At, long Rx, long Tx)? _previousNet;
 
@@ -26,7 +25,6 @@ public sealed class HostSampler
         string? procRoot = null, string? sysRoot = null, string? diskRoot = null)
     {
         _time = time;
-        _native = read is null && OperatingSystem.IsWindows(); // 只有默认读法才走 Windows 的系统调用；测试注入 read 时仍按 Linux 的文本解析
         _read = read ?? ReadProc;
         _procRoot = Root(procRoot, "HOST_PROC", "/proc");
         _sysRoot = Root(sysRoot, "HOST_SYS", "/sys");
@@ -39,8 +37,7 @@ public sealed class HostSampler
         long now = _time.GetTimestamp();
         var snapshot = new HostSnapshot { AtUtcMs = _time.GetUtcNow().ToUnixTimeMilliseconds() };
 
-        (long Total, long Idle)? cpuNow = _native ? WindowsCpu() : _read(Path.Combine(_procRoot, "stat")) is { } stat ? ParseCpu(stat) : null;
-        if (cpuNow is { } cpu)
+        if (_read(Path.Combine(_procRoot, "stat")) is { } stat && ParseCpu(stat) is { } cpu)
         {
             if (_previousCpu is { } before && cpu.Total > before.Total)
             {
@@ -51,8 +48,7 @@ public sealed class HostSampler
             _previousCpu = cpu;
         }
 
-        (long Total, long Used)? memoryNow = _native ? WindowsMemory() : _read(Path.Combine(_procRoot, "meminfo")) is { } meminfo ? ParseMemory(meminfo) : null;
-        if (memoryNow is { } memory)
+        if (_read(Path.Combine(_procRoot, "meminfo")) is { } meminfo && ParseMemory(meminfo) is { } memory)
         {
             snapshot.MemTotal = memory.Total;
             snapshot.MemUsed = memory.Used;
@@ -64,21 +60,11 @@ public sealed class HostSampler
             snapshot.DiskUsed = disk.Used;
         }
 
-        (long Rx, long Tx, long? LinkMbps)? netNow = null;
-        if (_native)
-        {
-            netNow = WindowsNetwork();
-        }
-        else if (_read(Path.Combine(_procRoot, "net/dev")) is { } dev && ParseNetwork(dev) is { } parsed)
-        {
-            netNow = (parsed.Rx, parsed.Tx, ReadLinkSpeed(dev));
-        }
-
-        if (netNow is { } net)
+        if (_read(Path.Combine(_procRoot, "net/dev")) is { } dev && ParseNetwork(dev) is { } net)
         {
             snapshot.NetRxTotal = net.Rx;
             snapshot.NetTxTotal = net.Tx;
-            snapshot.NetLinkMbps = net.LinkMbps;
+            snapshot.NetLinkMbps = ReadLinkSpeed(dev);
             if (_previousNet is { } before && net.Rx >= before.Rx && net.Tx >= before.Tx)
             {
                 double seconds = Math.Max(0.001, _time.GetElapsedTime(before.At, now).TotalSeconds);
@@ -97,12 +83,6 @@ public sealed class HostSampler
         return snapshot;
     }
 
-    private static (long Total, long Idle)? WindowsCpu() => OperatingSystem.IsWindows() ? WindowsHost.Cpu() : null;
-
-    private static (long Total, long Used)? WindowsMemory() => OperatingSystem.IsWindows() ? WindowsHost.Memory() : null;
-
-    private static (long Rx, long Tx, long? LinkMbps)? WindowsNetwork() => OperatingSystem.IsWindows() ? WindowsHost.Network() : null;
-
     /// <summary>第一行 <c>cpu  user nice system idle iowait irq softirq steal …</c>：总时间与空闲时间（idle + iowait）。</summary>
     public static (long Total, long Idle)? ParseCpu(string stat)
     {

backend/agent/src/MoShiro.Agent/MoShiro.Agent.csproj

@@ -23,6 +23,4 @@
   <Import Project="Sdk.targets" Sdk="Microsoft.NET.Sdk" />
   <Import Project="Sdk.targets" Sdk="StuDev.AotAnywhere"
           Condition="'$(UseAotAnywhere)' == 'true'" />
-  <Import Project="..\..\..\shared\MoShiro.Measurement\StaticMsQuic.targets"
-          Condition="'$(StaticMsQuic)' == 'true'" />
 </Project>

backend/agent/src/MoShiro.Agent/OpenWrtUpdater.cs

@@ -1,76 +0,0 @@
-using System.Text.Json;
-using System.Text.Json.Serialization;
-using Fleet.Core;
-using Fleet.Hosting;
-using MoShiro.Measurement;
-using MoShiro.Protocol;
-
-namespace MoShiro.Agent;
-
-/// <summary>
-/// Native supervisor bridge: writes only fixed wake-up flags and a health beacon in tmpfs.
-/// Never downloads, installs, verifies releases, or executes commands. The root-owned host
-/// updater independently fetches and verifies its locally configured release feed.
-/// </summary>
-public sealed class OpenWrtUpdater
-{
-    private readonly ConfigStore _runtime;
-    private readonly string _statusPath;
-    private readonly TimeProvider _time;
-    private readonly INodeLog _log;
-
-    public OpenWrtUpdater(string runtimeDirectory, string statusPath, TimeProvider time, INodeLog log)
-    {
-        _runtime = new ConfigStore(runtimeDirectory);
-        _statusPath = statusPath;
-        _time = time;
-        _log = log;
-    }
-
-    public static OpenWrtUpdater? FromEnvironment(TimeProvider time, INodeLog log) =>
-        Environment.GetEnvironmentVariable(RestartUpdater.EnvironmentVariable) == "openwrt"
-            ? new OpenWrtUpdater("/var/run/moshiro-agent", "/var/run/moshiro-update/status.json", time, log)
-            : null;
-
-    public void Offer() => Signal("check.request");
-
-    public void RequestApply() => Signal("apply.request");
-
-    private void Signal(string fileName)
-    {
-        try
-        {
-            _runtime.Write(fileName, "check"u8.ToArray());
-        }
-        catch (Exception ex) when (ex is IOException or UnauthorizedAccessException)
-        {
-            _log.Warn("could not signal the native updater", ("err", ex.Message));
-        }
-    }
-
-    /// <summary>Called by the Fleet heartbeat: evidence that this version reached Backend.</summary>
-    public UpdateStatus? Status()
-    {
-        try
-        {
-            var health = new OpenWrtHealth(BuildInfo.Version, Environment.ProcessId, _time.GetUtcNow().ToUnixTimeSeconds());
-            _runtime.Write("health.json", JsonSerializer.SerializeToUtf8Bytes(health, AgentConfigContext.Default.OpenWrtHealth));
-            if (!File.Exists(_statusPath) || new FileInfo(_statusPath).Length > 8192)
-            {
-                return null;
-            }
-
-            return JsonSerializer.Deserialize(File.ReadAllBytes(_statusPath), MoShiroJsonContext.Default.UpdateStatus);
-        }
-        catch (Exception ex) when (ex is IOException or UnauthorizedAccessException or JsonException)
-        {
-            _log.Warn("could not read native update status", ("err", ex.Message));
-            return null;
-        }
-    }
-}
-
-public sealed record OpenWrtHealth(
-    [property: JsonPropertyName("version")] string Version,
-    [property: JsonPropertyName("pid")] int Pid,
-    [property: JsonPropertyName("at")] long At);

backend/agent/src/MoShiro.Agent/Peer/PeerHub.cs

@@ -22,7 +22,7 @@ public sealed class BackendUplink
 
     public bool IsConnected => _link?.CurrentSession is not null;
 
-    public async ValueTask EmitEventAsync(string kind, string? detail = null, ConnectedDialer? dialer = null)
+    public async ValueTask EmitEventAsync(string kind, string? detail = null, ConnectedProbe? probe = null)
     {
         RpcSession<LinkContext>? session = _link?.CurrentSession;
         if (session is null)
@@ -30,7 +30,7 @@ public sealed class BackendUplink
             return;
         }
 
-        var parameters = new AgentEventParams { Kind = kind, AtUtcMs = _time.GetUtcNow().ToUnixTimeMilliseconds(), Detail = detail, Dialer = dialer };
+        var parameters = new AgentEventParams { Kind = kind, AtUtcMs = _time.GetUtcNow().ToUnixTimeMilliseconds(), Detail = detail, Probe = probe };
         await SafeAsync(() => session.NotifyAsync(MoShiroMethods.Event, parameters, MoShiroJsonContext.Default.AgentEventParams, CancellationToken.None));
     }
 
@@ -71,16 +71,15 @@ public sealed class BackendUplink
 }
 
 /// <summary>
-/// 拨号方 的点对点会话：每次握手成功之后登记会话（同一个拨号方的新会话取代它自己的旧会话，不同拨号方的会话并存）、为这个会话起一组线路任务的流并下发线路任务集、接收拨号方的列块并交给
-/// <see cref="ChunkAssembler"/>。会话结束时通知 Backend，并让正在运行的测试以 <c>dialer_offline</c> 结束。
+/// Probe 的点对点会话：每次握手成功之后登记会话（同一个 Probe 的新会话取代它自己的旧会话，不同 Probe 的会话并存）、为这个会话起一组线路任务的流并下发线路任务集、接收 Probe 的列块并交给
+/// <see cref="ChunkAssembler"/>。会话结束时通知 Backend，并让正在运行的测试以 <c>probe_offline</c> 结束。
 /// </summary>
 public sealed class PeerHub
 {
     private const int DedupeWindow = 1024;
 
-    /// <summary>超过这么久没有收到拨号方的任何帧就认为会话已死（拨号方 每 30 秒至少发一次时钟同步）。
-    /// 要比线路可能的中断长得多：差的线路上几十秒甚至一两分钟的黑洞时有发生（TCP RTT 能到 30 秒），75 秒就关会话会把一次中断变成几分钟的数据空白（要重新握手，而握手在差线路上也慢）。</summary>
-    private static readonly TimeSpan IdleLimit = TimeSpan.FromSeconds(180);
+    /// <summary>超过这么久没有收到 Probe 的任何帧就认为会话已死（Probe 每 30 秒至少发一次时钟同步）。</summary>
+    private static readonly TimeSpan IdleLimit = TimeSpan.FromSeconds(75);
 
         private readonly RpcRouter<PeerContext> _router;
     private readonly SessionRegistry _sessions;
@@ -106,15 +105,14 @@ public sealed class PeerHub
         _log = log;
     }
 
-    /// <summary>拨号方 会话断开（这个拨号方上的测试据此中止）。</summary>
-    public Action<DialerSession>? OnDialerLost { get; set; }
+    /// <summary>Probe 会话断开（这个 Probe 上的测试据此中止）。</summary>
+    public Action<ProbeSession>? OnProbeLost { get; set; }
 
     public async Task OnSessionAsync(PeerSessionInfo info, IRpcTransport transport, CancellationToken ct)
     {
-        var dialer = new DialerSession(info) { ConnectedAtUtcMs = _time.GetUtcNow().ToUnixTimeMilliseconds() };
-        var log = new ContextNodeLog(_log, ("dialer", dialer.DialerId), ("ip_version", dialer.IpVersion), ("control_id", Guid.NewGuid().ToString("N")));
-        var rpc = new RpcSession<PeerContext>(transport, _router, new PeerContext(dialer, _clock), _time);
-        dialer.Rpc = rpc;
+        var probe = new ProbeSession(info) { ConnectedAtUtcMs = _time.GetUtcNow().ToUnixTimeMilliseconds() };
+        var rpc = new RpcSession<PeerContext>(transport, _router, new PeerContext(probe, _clock), _time);
+        probe.Rpc = rpc;
 
         var seen = new HashSet<Guid>();
         var order = new Queue<Guid>();
@@ -137,7 +135,7 @@ public sealed class PeerHub
                 Result<RawChunk> decoded = RawChunkCodec.Decode(payload.Span);
                 if (decoded.TryGet(out RawChunk? fragment, out FleetError? error))
                 {
-                    _chunks.Accept(fragment, dialer);
+                    _chunks.Accept(fragment, probe);
                 }
                 else
                 {
@@ -149,23 +147,21 @@ public sealed class PeerHub
             await rpc.NotifyAsync(FleetMethods.OutboxAck, ack, FleetRpcContext.Default.OutboxAckParams, token);
         };
 
-        DialerSession? previous = _sessions.Replace(dialer);
+        ProbeSession? previous = _sessions.Replace(probe);
         if (previous?.Rpc is { } old)
         {
-            previous.CloseReason = "replaced";
-            log.Info("replacing previous dialer control session");
             await old.CloseAsync("replaced", CancellationToken.None);
         }
 
         // 这个会话自己的一组流：序号从 0 开始，上一个会话留下的记录不会算到它头上
-        LineTaskSet initial = _tasks.OpenSession(dialer);
+        LineTaskSet initial = _tasks.OpenSession(probe);
 
         using CancellationTokenSource linked = CancellationTokenSource.CreateLinkedTokenSource(ct);
         Task run = rpc.RunAsync(linked.Token);
-        Task watchdog = WatchIdleAsync(dialer, rpc, log, linked.Token);
-        log.Info("dialer connected", ("phase", "control_session"), ("version", info.Version ?? ""), ("remote", info.Remote));
-        await _uplink.EmitEventAsync(AgentEventKinds.DialerConnected, info.PeerId, Describe(dialer));
-        await PushAsync(dialer, initial, linked.Token);
+        Task watchdog = WatchIdleAsync(probe, rpc, linked.Token);
+        _log.Info("probe connected", ("probe", info.PeerId), ("version", info.Version ?? ""), ("remote", probe.RemoteAddress ?? ""));
+        await _uplink.EmitEventAsync(AgentEventKinds.ProbeConnected, info.PeerId, Describe(probe));
+        await PushAsync(probe, initial, linked.Token);
         try
         {
             await run;
@@ -174,35 +170,25 @@ public sealed class PeerHub
         {
             // 停止
         }
-        catch (Exception ex)
-        {
-            dialer.CloseReason ??= "control_error";
-            log.Warn("dialer control session failed", ("exception", ex.GetType().Name), ("err", ex.Message));
-            throw;
-        }
         finally
         {
-            _sessions.Remove(dialer);
+            _sessions.Remove(probe);
             await linked.CancelAsync();
             await watchdog;
             await rpc.DisposeAsync();
-            await _tasks.CloseSessionAsync(dialer);
-            log.Info("dialer disconnected", ("phase", "control_session"),
-                ("reason", dialer.CloseReason ?? (ct.IsCancellationRequested ? "service_stopping" : "peer_closed")),
-                ("duration_ms", _time.GetUtcNow().ToUnixTimeMilliseconds() - dialer.ConnectedAtUtcMs),
-                ("datagram_bound", dialer.DatagramConnection is not null), ("data_remote", dialer.DatagramRemoteEndPoint));
-            OnDialerLost?.Invoke(dialer);
-            await _uplink.EmitEventAsync(AgentEventKinds.DialerDisconnected, info.PeerId, Describe(dialer));
+            await _tasks.CloseSessionAsync(probe);
+            _log.Info("probe disconnected", ("probe", info.PeerId));
+            OnProbeLost?.Invoke(probe);
+            await _uplink.EmitEventAsync(AgentEventKinds.ProbeDisconnected, info.PeerId, Describe(probe));
         }
     }
 
     /// <summary>
-    /// 拨号方 每 30 秒发一次时钟同步请求，所以会话里至少每 30 秒有一帧：超过 <see cref="IdleLimit"/> 没有收到任何东西，
-    /// 说明拨号方已经静默死亡（断电、进程被冻结），关闭会话，让 Backend 看到 dialer_disconnected，而不是一直显示在线。
+    /// Probe 每 30 秒发一次时钟同步请求，所以会话里至少每 30 秒有一帧：超过 <see cref="IdleLimit"/> 没有收到任何东西，
+    /// 说明 Probe 已经静默死亡（断电、进程被冻结），关闭会话，让 Backend 看到 probe_disconnected，而不是一直显示在线。
     /// </summary>
-    private async Task WatchIdleAsync(DialerSession dialer, RpcSession<PeerContext> rpc, INodeLog log, CancellationToken ct)
+    private async Task WatchIdleAsync(ProbeSession probe, RpcSession<PeerContext> rpc, CancellationToken ct)
     {
-        long lastSummary = _time.GetTimestamp();
         using var timer = new PeriodicTimer(TimeSpan.FromSeconds(10), _time);
         try
         {
@@ -210,18 +196,10 @@ public sealed class PeerHub
             {
                 if (rpc.IdleFor > IdleLimit)
                 {
-                    dialer.CloseReason = "idle_timeout";
-                    log.Warn("dialer session is silent, closing it", ("phase", "control_session"), ("idle_s", (long)rpc.IdleFor.TotalSeconds), ("limit_s", IdleLimit.TotalSeconds));
+                    _log.Warn("probe session is silent, closing it", ("probe", probe.ProbeId), ("idle_s", (long)rpc.IdleFor.TotalSeconds));
                     await rpc.CloseAsync("idle", CancellationToken.None);
                     return;
                 }
-                if (_time.GetElapsedTime(lastSummary) >= TimeSpan.FromSeconds(60))
-                {
-                    lastSummary = _time.GetTimestamp();
-                    log.Info("dialer session summary", ("phase", "control_session"), ("remote", dialer.Info.Remote),
-                        ("control_idle_s", (long)rpc.IdleFor.TotalSeconds), ("datagram_bound", dialer.DatagramConnection is not null),
-                        ("data_remote", dialer.DatagramRemoteEndPoint));
-                }
             }
         }
         catch (OperationCanceledException)
@@ -230,23 +208,24 @@ public sealed class PeerHub
         }
     }
 
-    /// <summary>当前连着的拨号方（心跳和 Backend 重连之后的补报都用它）。</summary>
-    public IReadOnlyList<ConnectedDialer> ConnectedDialers() => [.. _sessions.All.Select(Describe)];
+    /// <summary>当前连着的 Probe（心跳和 Backend 重连之后的补报都用它）。</summary>
+    public IReadOnlyList<ConnectedProbe> ConnectedProbes() => [.. _sessions.All.Select(Describe)];
 
-    private static ConnectedDialer Describe(DialerSession p) => new() { Id = new NodeId(p.DialerId), Version = p.Version, Remote = p.RemoteAddress, SinceUtcMs = p.ConnectedAtUtcMs, IpVersion = p.IpVersion };
+    private static ConnectedProbe Describe(ProbeSession p) => new() { Id = new ProbeId(p.ProbeId), Version = p.Version, Remote = p.RemoteAddress, SinceUtcMs = p.ConnectedAtUtcMs, IpVersion = p.IpVersion };
 
-    /// <summary>向一个拨号方下发它的线路任务集（拨号方 连上时由 <see cref="OnSessionAsync"/> 先发一份）。</summary>
-    public async ValueTask PushToAsync(DialerSession dialer, LineTaskSet set) => await PushAsync(dialer, set, CancellationToken.None);
+    public IReadOnlyList<string> ConnectedProbeIds() => [.. _sessions.All.Select(p => p.ProbeId)];
 
-    /// <summary>Backend 要求断开某个拨号方（它会按退避重连，除非已经被吊销）。没有这个会话时返回 false。</summary>
-    public async ValueTask<bool> DisconnectAsync(string dialerId)
+    /// <summary>向一个 Probe 下发它的线路任务集（Probe 连上时由 <see cref="OnSessionAsync"/> 先发一份）。</summary>
+    public async ValueTask PushToAsync(ProbeSession probe, LineTaskSet set) => await PushAsync(probe, set, CancellationToken.None);
+
+    /// <summary>Backend 要求断开某个 Probe（它会按退避重连，除非已经被吊销）。没有这个会话时返回 false。</summary>
+    public async ValueTask<bool> DisconnectAsync(string probeId)
     {
         bool disconnected = false;
-        foreach (DialerSession dialer in _sessions.All)
+        foreach (ProbeSession probe in _sessions.All)
         {
-            if (dialer.DialerId == dialerId && dialer.Rpc is { } rpc)
+            if (probe.ProbeId == probeId && probe.Rpc is { } rpc)
             {
-                dialer.CloseReason = "disconnected";
                 await rpc.CloseAsync("disconnected", CancellationToken.None);
                 disconnected = true;
             }
@@ -255,23 +234,22 @@ public sealed class PeerHub
         return disconnected;
     }
 
-    /// <summary>名单（期望状态里的 peers）里已经没有的拨号方：立刻关闭它的会话（吊销生效）。</summary>
+    /// <summary>名单（期望状态里的 peers）里已经没有的 Probe：立刻关闭它的会话（吊销生效）。</summary>
     public void CloseNotIn(IReadOnlySet<string> allowed)
     {
-        foreach (DialerSession dialer in _sessions.All)
+        foreach (ProbeSession probe in _sessions.All)
         {
-            if (!allowed.Contains(dialer.DialerId) && dialer.Rpc is { } rpc)
+            if (!allowed.Contains(probe.ProbeId) && probe.Rpc is { } rpc)
             {
-                dialer.CloseReason = "revoked";
-                _log.Info("dialer is no longer allowed, closing its session", ("dialer", dialer.DialerId));
+                _log.Info("probe is no longer allowed, closing its session", ("probe", probe.ProbeId));
                 _ = rpc.CloseAsync("revoked", CancellationToken.None).AsTask();
             }
         }
     }
 
-    private async ValueTask PushAsync(DialerSession dialer, LineTaskSet set, CancellationToken ct)
+    private async ValueTask PushAsync(ProbeSession probe, LineTaskSet set, CancellationToken ct)
     {
-        if (dialer.Rpc is null)
+        if (probe.Rpc is null)
         {
             return;
         }
@@ -284,10 +262,10 @@ public sealed class PeerHub
         };
         try
         {
-            Result<StateSyncResult> result = await dialer.Rpc.RequestAsync(FleetMethods.StateSync, parameters, FleetRpcContext.Default.StateSyncParams, FleetRpcContext.Default.StateSyncResult, ct);
+            Result<StateSyncResult> result = await probe.Rpc.RequestAsync(FleetMethods.StateSync, parameters, FleetRpcContext.Default.StateSyncParams, FleetRpcContext.Default.StateSyncResult, ct);
             if (!result.IsSuccess)
             {
-                _log.Warn("dialer rejected the line task set", ("err", result.Error.Code));
+                _log.Warn("probe rejected the line task set", ("err", result.Error.Code));
             }
         }
         catch (Exception ex) when (ex is ObjectDisposedException or InvalidOperationException or IOException or OperationCanceledException)
@@ -302,5 +280,5 @@ public sealed class PeerHub
                 MoShiroMethods.Clock,
                 MoShiroJsonContext.Default.ClockParams,
                 MoShiroJsonContext.Default.ClockResult,
-                (context, _, _) => ValueTask.FromResult(Result<ClockResult>.Ok(new ClockResult { ListenerMonoUs = context.Clock.NowUs() })));
+                (context, _, _) => ValueTask.FromResult(Result<ClockResult>.Ok(new ClockResult { AgentMonoUs = context.Clock.NowUs() })));
 }

backend/agent/src/MoShiro.Agent/Program.cs

@@ -3,7 +3,7 @@ using Fleet.Hosting;
 using MoShiro.Agent;
 
 TimeProvider time = TimeProvider.System;
-INodeLog log = new BufferedNodeLog(new ConsoleNodeLog(time), time);
+INodeLog log = new ConsoleNodeLog(time);
 
 if (PlatformCheck.VerifyCrypto() is { } problem)
 {
@@ -43,7 +43,7 @@ catch (Exception ex) when (ex is IOException or InvalidDataException or InvalidO
 using var restartScope = restart;
 using (identity)
 {
-    await using var app = new AgentApp(config, identity, Path.GetDirectoryName(Path.GetFullPath(configPath))!, time, log, restart);
+    using var app = new AgentApp(config, identity, Path.GetDirectoryName(Path.GetFullPath(configPath))!, time, log, restart);
     await ServiceShell.RunAsync(app.RunAsync, log, restart, time);
 }
 

backend/agent/src/MoShiro.Agent/RateChangeLog.cs

@@ -1,40 +0,0 @@
-using System.Collections.Concurrent;
-using MoShiro.Protocol;
-
-namespace MoShiro.Agent;
-
-/// <summary>
-/// 自适应探测频率的状态变化，等着随下一次心跳上报给 Backend（Backend 存成事件，用于事后审计）。
-/// 和普通日志分开存：日志环形缓冲几分钟到几小时就会被覆盖，而这里的变化要能在几天后回看。有界，满了丢最旧的。
-/// </summary>
-public sealed class RateChangeLog
-{
-    public const int Capacity = 512;
-
-    private readonly ConcurrentQueue<RateChange> _queue = new();
-
-    public void Add(RateChange change)
-    {
-        _queue.Enqueue(change);
-        while (_queue.Count > Capacity && _queue.TryDequeue(out _))
-        {
-        }
-    }
-
-    /// <summary>取走最多 <paramref name="max"/> 条（最旧的在前）；没有返回 null，心跳里就不带这个字段。</summary>
-    public List<RateChange>? Take(int max)
-    {
-        List<RateChange>? taken = null;
-        while ((taken?.Count ?? 0) < max)
-        {
-            if (!_queue.TryDequeue(out RateChange? change))
-            {
-                break;
-            }
-
-            (taken ??= []).Add(change);
-        }
-
-        return taken;
-    }
-}

backend/agent/src/MoShiro.Agent/Tasks/TargetRunners.cs

@@ -7,16 +7,14 @@ using MoShiro.Protocol;
 namespace MoShiro.Agent;
 
 /// <summary>
-/// 目标任务（Agent 自己探测一个地址，没有对端）：用和线路任务同一套积木（<see cref="BlockCatalog"/> 里的拨号方，例如 icmp_ping、connect），
-/// 只是列块不交给 Agent 合并，而是自己盖章、10 秒一块直接上传。目标可以是 IP、域名或（http）URL；域名在 Agent 上解析，
-/// 每个解析出的地址都必须在本地 <c>allowed_target_cidrs</c> 内（IP 目标由 <see cref="BudgetGuard"/> 在接受任务之前检查，域名在这里检查）。
+/// 目标任务（Agent 自己探测一个地址，没有对端）：用和线路任务同一套积木（<see cref="BlockCatalog"/> 里的拨号方，例如 icmp_echo、connect），
+/// 只是列块不交给 Agent 合并，而是自己盖章、10 秒一块直接上传。目标地址必须在本地 <c>allowed_target_cidrs</c> 内
+/// （由 <see cref="BudgetGuard"/> 在接受任务之前检查）。
 /// </summary>
 public sealed class TargetRunners
 {
     private static readonly TimeSpan s_chunkSpan = TimeSpan.FromSeconds(ProtocolLimits.ChunkSeconds);
 
-    private static readonly TimeSpan s_resolveRetry = TimeSpan.FromMinutes(1);
-
     private readonly Dictionary<string, Running> _running = [];
     private readonly object _gate = new();
     private readonly ChunkAssembler _chunks;
@@ -24,11 +22,9 @@ public sealed class TargetRunners
     private readonly MonoClock _clock;
     private readonly TimeProvider _time;
     private readonly INodeLog _log;
-    private readonly RateChangeLog? _rateChanges;
 
-    public TargetRunners(ChunkAssembler chunks, NodeId agent, MonoClock clock, TimeProvider time, INodeLog log, RateChangeLog? rateChanges = null)
+    public TargetRunners(ChunkAssembler chunks, NodeId agent, MonoClock clock, TimeProvider time, INodeLog log)
     {
-        _rateChanges = rateChanges;
         _chunks = chunks;
         _agent = agent;
         _clock = clock;
@@ -37,7 +33,7 @@ public sealed class TargetRunners
     }
 
     /// <summary>让运行中的任务与期望集合一致：参数没有变化的保持，其余启动或停止。</summary>
-    public void Reconcile(IReadOnlyList<(TaskSpec Spec, TaskDecision Decision)> desired, Func<IPAddress, bool>? allow = null)
+    public void Reconcile(IReadOnlyList<(TaskSpec Spec, TaskDecision Decision)> desired)
     {
         lock (_gate)
         {
@@ -60,7 +56,7 @@ public sealed class TargetRunners
 
                 var cts = new CancellationTokenSource();
                 _running[id] = new Running(Signature(spec), cts);
-                _ = Task.Run(() => RunAsync(spec, decision, allow, cts.Token), CancellationToken.None);
+                _ = Task.Run(() => RunAsync(spec, decision, cts.Token), CancellationToken.None);
             }
         }
     }
@@ -68,34 +64,23 @@ public sealed class TargetRunners
     private static string Signature(TaskSpec spec) =>
         spec.Kind + "|" + spec.Target + "|" + string.Join(',', spec.Params.OrderBy(p => p.Key, StringComparer.Ordinal).Select(p => p.Key + "=" + p.Value.GetRawText()));
 
-    private async Task RunAsync(TaskSpec spec, TaskDecision decision, Func<IPAddress, bool>? allow, CancellationToken ct)
+    private async Task RunAsync(TaskSpec spec, TaskDecision decision, CancellationToken ct)
     {
         try
         {
-            if (!TargetHost.TryParse(spec.Target, out string host, out int targetPort, out _))
+            if (!TryResolve(spec.Target!, decision.Resolved!, spec.Kind, out IPAddress? address, out int port))
             {
-                _log.Warn("target is not valid", ("task", spec.Id.Value));
+                _log.Warn("target is not an address", ("task", spec.Id.Value));
                 return;
             }
 
-            // http 自己解析并检查每个地址；其余 Kind 在这里解析出一个允许的地址作为对端
-            IPAddress? address = null;
-            if (spec.Kind != TargetHost.HttpKind)
-            {
-                address = await ResolveAllowedAsync(spec, host, allow, ct);
-            }
-
-            int port = targetPort != 0 ? targetPort : spec.Kind == "tcp_connect" ? (int)decision.Resolved!.GetInt("port") : 0;
             var env = new BlockEnv
             {
                 Clock = _clock,
                 Log = _log,
                 Session = () => null,
                 PeerAddress = address,
-                Target = spec.Target,
-                AllowAddress = allow,
-                OnRateChange = _rateChanges is null ? null : _rateChanges.Add,
-                Fragments = new LocalSink(this, spec, address?.AddressFamily == System.Net.Sockets.AddressFamily.InterNetworkV6 ? IpVersions.V6 : IpVersions.V4),
+                Fragments = new LocalSink(this, spec, address.AddressFamily == System.Net.Sockets.AddressFamily.InterNetworkV6 ? IpVersions.V6 : IpVersions.V4),
             };
             IBlock block = BlockCatalog.Find(spec.Kind)!.Dialer(env, new BlockSpec(decision.Kind!, decision.Resolved!, 0, port, new FragmentScope(spec.Id, null, null)));
             using var cts = CancellationTokenSource.CreateLinkedTokenSource(ct);
@@ -125,32 +110,24 @@ public sealed class TargetRunners
         }
     }
 
-    /// <summary>解析目标，取第一个允许的地址；解析失败或没有允许的地址就每分钟重试（域名的解析结果会变）。</summary>
-    private async Task<IPAddress> ResolveAllowedAsync(TaskSpec spec, string host, Func<IPAddress, bool>? allow, CancellationToken ct)
+    private static bool TryResolve(string target, ParamValues values, string kind, out IPAddress address, out int port)
     {
-        while (true)
+        port = kind == "connect" ? (int)values.GetInt("port") : 0;
+        if (IPEndPoint.TryParse(target, out IPEndPoint? endpoint))
         {
-            try
+            address = endpoint.Address;
+            if (endpoint.Port != 0)
             {
-                IPAddress[] addresses = IPAddress.TryParse(host, out IPAddress? literal) ? [literal] : await Dns.GetHostAddressesAsync(host, ct);
-                IPAddress? chosen = addresses.FirstOrDefault(a => allow?.Invoke(a) ?? true);
-                if (chosen is not null)
-                {
-                    return chosen;
+                port = endpoint.Port;
             }
 
-                _log.Warn("no address of the target is allowed", ("task", spec.Id.Value), ("host", host));
-            }
-            catch (System.Net.Sockets.SocketException ex)
-            {
-                _log.Warn("target did not resolve", ("task", spec.Id.Value), ("host", host), ("err", ex.Message));
+            return true;
         }
 
-            await Task.Delay(s_resolveRetry, _time, ct);
-        }
+        return IPAddress.TryParse(target, out address!);
     }
 
-    /// <summary>目标任务的行没有拨号方一侧可以合并：积木交出的就是完整的行，Agent 自己盖章（目标、时钟锚点）并放进发件箱。</summary>
+    /// <summary>目标任务的行没有 Probe 一侧可以合并：积木交出的就是完整的行，Agent 自己盖章（目标、时钟锚点）并放进发件箱。</summary>
     private sealed class LocalSink : IFragmentSink
     {
         private readonly TargetRunners _owner;
@@ -179,7 +156,7 @@ public sealed class TargetRunners
                 Port = port,
                 StartUtcMs = startUtcMs,
                 DurationMs = Math.Clamp(durationMs, 1, ChunkFinalizer.MaxDurationMs),
-                Clock = new ChunkClock { ListenerMonoUs = anchor.ListenerMonoUs, UtcMs = anchor.UtcMs, DialerOffsetUs = 0 },
+                Clock = new ChunkClock { AgentMonoUs = anchor.AgentMonoUs, UtcMs = anchor.UtcMs, ProbeOffsetUs = 0 },
             };
             _owner._chunks.Emit(new RawChunk(header, streams));
         }

backend/agent/src/MoShiro.Agent/Tasks/TaskManager.cs

@@ -9,13 +9,13 @@ namespace MoShiro.Agent;
 
 /// <summary>
 /// 期望任务集（<c>task_set</c>）的调和器：每个任务先经过 <see cref="BudgetGuard"/>（参数校验 + 本地硬上限）。
-/// 通过的线路任务<b>为每个拨号方会话各登记一组流</b>（流编号和监听方积木都属于会话，互不干扰），并汇总成 <c>line_task_set</c> 下发给这个拨号方；
-/// 通过的目标任务在本地运行，与拨号方无关。被拒绝的任务带着原因（<c>rejected[]</c>）回报给 Backend，其余任务照常运行。
-/// 每次调用都应用完整的快照，所以是幂等的。预算（码率、包速率）按单个拨号方会话计算，会话总数由 <c>PeerListenerOptions.MaxSessions</c> 限制。
+/// 通过的线路任务<b>为每个 Probe 会话各登记一组流</b>（流编号和监听方积木都属于会话，互不干扰），并汇总成 <c>line_task_set</c> 下发给这个 Probe；
+/// 通过的目标任务在本地运行，与 Probe 无关。被拒绝的任务带着原因（<c>rejected[]</c>）回报给 Backend，其余任务照常运行。
+/// 每次调用都应用完整的快照，所以是幂等的。预算（码率、包速率）按单个 Probe 会话计算，会话总数由 <c>PeerListenerOptions.MaxSessions</c> 限制。
 /// </summary>
 public sealed class TaskManager
 {
-    // 线路任务 = 有监听方积木、并且不是只用于按需测试的 Kind（quic_stream、tcp_transfer 会占满链路）；目标任务 = 可以由 Agent 自己运行的 Kind
+    // 线路任务 = 有监听方积木、并且不是只用于按需测试的 Kind（udp_train、tcp_bulk 会占满链路）；目标任务 = 可以由 Agent 自己运行的 Kind
     private static readonly HashSet<string> s_lineKinds = [.. BlockCatalog.All.Where(p => p.Listener is not null && !p.OnDemandOnly).Select(p => p.Kind)];
     private static readonly HashSet<string> s_targetKinds = [.. BlockCatalog.All.Where(p => p.RunsAsTarget).Select(p => p.Kind)];
 
@@ -24,67 +24,65 @@ public sealed class TaskManager
     private readonly StreamRegistry _streams;
     private readonly BlockEnv _env;
     private readonly TargetRunners _targets;
-    private readonly Func<IReadOnlyList<int>> _quicPorts;
+    private readonly Func<IReadOnlyList<int>> _udpPorts;
     private readonly Func<IReadOnlyList<int>> _tcpPorts;
     private readonly TrustAnchor _trust;
-    private readonly bool _acceptLineTasks;
-    private readonly Dictionary<DialerSession, Dictionary<string, LineEntry>> _sessionLines = [];
+    private readonly Dictionary<ProbeSession, Dictionary<string, LineEntry>> _sessionLines = [];
     private volatile Dictionary<string, byte[]> _peers = new(StringComparer.Ordinal);
     private Dictionary<string, (TaskSpec Task, TaskDecision Decision)> _lineSpecs = [];
     private long _version;
     private BudgetGuard? _guard;
 
-    public TaskManager(Func<BudgetGuard> guardFactory, StreamRegistry streams, BlockEnv env, TargetRunners targets, Func<IReadOnlyList<int>> quicPorts, Func<IReadOnlyList<int>> tcpPorts, TrustAnchor? trust = null, bool acceptLineTasks = true)
+    public TaskManager(Func<BudgetGuard> guardFactory, StreamRegistry streams, BlockEnv env, TargetRunners targets, Func<IReadOnlyList<int>> udpPorts, Func<IReadOnlyList<int>> tcpPorts, TrustAnchor? trust = null)
     {
         _trust = trust ?? TrustAnchor.None;
-        _acceptLineTasks = acceptLineTasks;
         _guardFactory = guardFactory;
         _streams = streams;
         _env = env;
         _targets = targets;
-        _quicPorts = quicPorts;
+        _udpPorts = udpPorts;
         _tcpPorts = tcpPorts;
     }
 
     public BudgetGuard Guard => _guard ??= _guardFactory();
 
-    /// <summary>某个拨号方的线路任务集变化（版本递增）：由 <see cref="PeerHub"/> 推送给这个拨号方。</summary>
-    public Action<DialerSession, LineTaskSet>? OnLineTasksChanged { get; set; }
+    /// <summary>某个 Probe 的线路任务集变化（版本递增）：由 <see cref="PeerHub"/> 推送给这个 Probe。</summary>
+    public Action<ProbeSession, LineTaskSet>? OnLineTasksChanged { get; set; }
 
-    /// <summary>允许连接的拨号方变了（期望状态里的 peers）：已经不在名单里的会话由 <see cref="PeerHub"/> 关闭。</summary>
+    /// <summary>允许连接的 Probe 变了（期望状态里的 peers）：已经不在名单里的会话由 <see cref="PeerHub"/> 关闭。</summary>
     public Action<IReadOnlySet<string>>? OnPeersChanged { get; set; }
 
-    /// <summary>点对点握手时查找拨号方的公钥（SPKI）。</summary>
+    /// <summary>点对点握手时查找 Probe 的公钥（SPKI）。</summary>
     public byte[]? LookupPeerKey(string peerId) => _peers.TryGetValue(peerId, out byte[]? key) ? key : null;
 
-    /// <summary>名册里全部拨号方的公钥：敲门关卡用它们派生配对密钥。</summary>
+    /// <summary>名册里全部 Probe 的公钥：敲门关卡用它们派生配对密钥。</summary>
     public IReadOnlyList<byte[]> PeerKeys() => [.. _peers.Values];
 
-    /// <summary>一个新的拨号方会话开始：为它登记全部线路任务的流并启动监听方积木，返回要下发给它的线路任务集。</summary>
-    public LineTaskSet OpenSession(DialerSession dialer)
+    /// <summary>一个新的 Probe 会话开始：为它登记全部线路任务的流并启动监听方积木，返回要下发给它的线路任务集。</summary>
+    public LineTaskSet OpenSession(ProbeSession probe)
     {
         lock (_gate)
         {
             var entries = new Dictionary<string, LineEntry>();
             foreach ((string id, (TaskSpec task, TaskDecision decision)) in _lineSpecs)
             {
-                LineEntry entry = BuildLineEntry(task, decision, dialer);
+                LineEntry entry = BuildLineEntry(task, decision, probe);
                 Start(entry);
                 entries[id] = entry;
             }
 
-            _sessionLines[dialer] = entries;
+            _sessionLines[probe] = entries;
             return SetFor(entries);
         }
     }
 
-    /// <summary>拨号方 会话结束：释放它的所有流和积木（记录随之丢弃，下一个会话从零开始）。</summary>
-    public async Task CloseSessionAsync(DialerSession dialer)
+    /// <summary>Probe 会话结束：释放它的所有流和积木（记录随之丢弃，下一个会话从零开始）。</summary>
+    public async Task CloseSessionAsync(ProbeSession probe)
     {
         List<LineEntry> entries;
         lock (_gate)
         {
-            if (!_sessionLines.Remove(dialer, out Dictionary<string, LineEntry>? lines))
+            if (!_sessionLines.Remove(probe, out Dictionary<string, LineEntry>? lines))
             {
                 return;
             }
@@ -110,11 +108,11 @@ public sealed class TaskManager
         {
             try
             {
-                peers[peer.NodeId.Value] = Convert.FromBase64String(peer.PublicKey);
+                peers[peer.ProbeId.Value] = Convert.FromBase64String(peer.PublicKey);
             }
             catch (FormatException)
             {
-                result.Rejected.Add(new RejectedItem { Id = peer.NodeId.Value, Reason = RejectReasons.InvalidParams });
+                result.Rejected.Add(new RejectedItem { Id = peer.ProbeId.Value, Reason = RejectReasons.InvalidParams });
             }
         }
 
@@ -133,7 +131,7 @@ public sealed class TaskManager
                 continue;
             }
 
-            bool shapeOk = task.Scope == "target" ? s_targetKinds.Contains(task.Kind) : _acceptLineTasks && s_lineKinds.Contains(task.Kind);
+            bool shapeOk = task.Scope == "target" ? s_targetKinds.Contains(task.Kind) : s_lineKinds.Contains(task.Kind);
             TaskDecision decision = shapeOk ? guard.Check(task, ref usedMbps, ref usedPps) : TaskDecision.Reject(RejectReasons.Unsupported);
             if (!decision.Accepted)
             {
@@ -151,29 +149,27 @@ public sealed class TaskManager
             }
         }
 
-        _targets.Reconcile(targets, guard.AddressAllowed);
+        _targets.Reconcile(targets);
 
         // 参数没有变化的任务沿用原来的流和积木，数据面上的会话（重放窗口、记录）不被打断
         var retired = new List<LineEntry>();
-        var changes = new List<(DialerSession Dialer, LineTaskSet Set)>();
+        var changes = new List<(ProbeSession Probe, LineTaskSet Set)>();
         lock (_gate)
         {
             _lineSpecs = line;
             _version = version;
-            foreach ((DialerSession dialer, Dictionary<string, LineEntry> current) in _sessionLines)
+            foreach ((ProbeSession probe, Dictionary<string, LineEntry> current) in _sessionLines)
             {
                 var next = new Dictionary<string, LineEntry>();
-                bool lineTasksChanged = current.Count != line.Count;
                 foreach ((string id, (TaskSpec task, TaskDecision decision)) in line)
                 {
-                    LineEntry candidate = BuildLineEntry(task, decision, dialer);
+                    LineEntry candidate = BuildLineEntry(task, decision, probe);
                     if (current.TryGetValue(id, out LineEntry? existing) && existing.Signature == candidate.Signature)
                     {
                         next[id] = existing;
                     }
                     else
                     {
-                        lineTasksChanged = true;
                         Start(candidate);
                         next[id] = candidate;
                     }
@@ -193,13 +189,7 @@ public sealed class TaskManager
                     current[id] = entry;
                 }
 
-                // The Agent receives a full task snapshot for peer/trust changes and target-only tasks too.
-                // Those changes do not affect a 拨号方's line blocks, so avoid retransmitting the same full
-                // line_task_set to every 拨号方. New sessions still get their initial snapshot in OpenSession.
-                if (lineTasksChanged)
-                {
-                    changes.Add((dialer, SetFor(current)));
-                }
+                changes.Add((probe, SetFor(current)));
             }
         }
 
@@ -209,15 +199,15 @@ public sealed class TaskManager
         }
 
         OnPeersChanged?.Invoke(new HashSet<string>(peers.Keys, StringComparer.Ordinal));
-        foreach ((DialerSession dialer, LineTaskSet next) in changes)
+        foreach ((ProbeSession probe, LineTaskSet next) in changes)
         {
-            OnLineTasksChanged?.Invoke(dialer, next);
+            OnLineTasksChanged?.Invoke(probe, next);
         }
 
         return result;
     }
 
-    /// <summary>停止时释放所有拨号方会话的流和所有目标任务（Backend 失联的时候任务继续运行：节点自治）。</summary>
+    /// <summary>停止时释放所有 Probe 会话的流和所有目标任务（Backend 失联的时候任务继续运行：节点自治）。</summary>
     public async Task StopAsync()
     {
         _targets.Reconcile([]);
@@ -239,13 +229,13 @@ public sealed class TaskManager
         {
             Version = _version,
             Tasks = [.. entries.Values.Select(e => e.Task)],
-            QuicPorts = [.. _quicPorts()],
+            UdpPorts = [.. _udpPorts()],
             TcpPorts = [.. _tcpPorts()],
         };
 
-    private LineEntry BuildLineEntry(TaskSpec task, TaskDecision decision, DialerSession dialer)
+    private LineEntry BuildLineEntry(TaskSpec task, TaskDecision decision, ProbeSession probe)
     {
-        var info = new StreamInfo(_streams.Allocate(), task.Kind, decision.Resolved!) { TaskId = task.Id, Owner = dialer };
+        var info = new StreamInfo(_streams.Allocate(), task.Kind, decision.Resolved!) { TaskId = task.Id, Owner = probe };
         var lineTask = new LineTask
         {
             Id = task.Id,
@@ -255,7 +245,7 @@ public sealed class TaskManager
             Port = info.Port,
         };
         var spec = new BlockSpec(decision.Kind!, decision.Resolved, info.Id, info.Port, new FragmentScope(task.Id, null, null));
-        return new LineEntry(lineTask, info, spec, _env.WithSession(dialer));
+        return new LineEntry(lineTask, info, spec, _env.WithSession(probe));
     }
 
     /// <summary>在数据面登记这个流并启动它的监听方积木。</summary>

backend/agent/src/MoShiro.Agent/Tests/TestRunner.cs

@@ -8,9 +8,9 @@ namespace MoShiro.Agent;
 
 /// <summary>
 /// 按需测试的编排（<c>moshiro.test.run</c>）：Agent 是时间的主人。每个阶段：
-/// 登记数据面的流 → <c>moshiro.test.phase</c>（拨号方 准备好自己一侧的角色）→ 运行（单向数据流由 Agent 发包，回显类等待拨号方发包）
-/// → <c>moshiro.test.phase_end</c>（拨号方 收尾，回复它交出的列块个数）→ 等所有列块合并上传 → <c>phase_done</c>。
-/// 同一时间只能运行 <c>max_concurrent_tests</c> 个测试；拨号方 掉线或被取消时测试中止，已经上传的阶段数据保留。
+/// 登记数据面的流 → <c>moshiro.test.phase</c>（Probe 准备好自己一侧的角色）→ 运行（单向数据流由 Agent 发包，回显类等待 Probe 发包）
+/// → <c>moshiro.test.phase_end</c>（Probe 收尾，回复它交出的列块个数）→ 等所有列块合并上传 → <c>phase_done</c>。
+/// 同一时间只能运行 <c>max_concurrent_tests</c> 个测试；Probe 掉线或被取消时测试中止，已经上传的阶段数据保留。
 /// </summary>
 public sealed class TestRunner
 {
@@ -51,9 +51,9 @@ public sealed class TestRunner
             return Result<TestRunResult>.Fail(MoShiroErrors.ValidationFailed, "ip_version must be ipv4 or ipv6");
         }
 
-        if (_sessions.Get(run.DialerId.Value, run.IpVersion) is not { } dialer)
+        if (_sessions.Get(run.ProbeId.Value, run.IpVersion) is not { } probe)
         {
-            return Result<TestRunResult>.Fail(MoShiroErrors.DialerOffline, $"dialer {run.DialerId.Value} is not connected to this agent");
+            return Result<TestRunResult>.Fail(MoShiroErrors.ProbeOffline, $"probe {run.ProbeId.Value} is not connected to this agent");
         }
 
         if (run.Phases.Count == 0)
@@ -85,7 +85,7 @@ public sealed class TestRunner
             resolved.Add((phase, _tasks.Guard.CheckPhase(phase.Kind, phase.Params)));
         }
 
-        var running = new RunningTest(dialer);
+        var running = new RunningTest(probe);
         if (_running.Count >= _maxConcurrent || !_running.TryAdd(run.TestId.Value, running))
         {
             running.Dispose();
@@ -107,7 +107,7 @@ public sealed class TestRunner
 
         // 按需测试的数据端口由 Agent 决定：Backend 给的端口如果不是本机监听的端口（模板里是出厂默认值，每个 Agent 可以改，
         // 而 Backend 并不知道 TCP 数据端口），就用本机监听的第一个端口。块头里记录的是实际用的端口。
-        IReadOnlyList<int> ports = _ports(kind.Blocks.Transport == "quic_datagram" ? "quic_datagram" : "tcp");
+        IReadOnlyList<int> ports = _ports(kind.Blocks.Transport == "udp" ? "udp" : "tcp");
         bool listening = phase.Params.TryGetValue("port", out System.Text.Json.JsonElement requested) && requested.TryGetInt32(out int port) && ports.Contains(port);
         if (!listening)
         {
@@ -126,12 +126,12 @@ public sealed class TestRunner
         return false;
     }
 
-    /// <summary>某个拨号方掉线：它上面的测试中止，状态是 failed / dialer_offline；别的拨号方上的测试不受影响。</summary>
-    public void DialerLost(DialerSession dialer)
+    /// <summary>某个 Probe 掉线：它上面的测试中止，状态是 failed / probe_offline；别的 Probe 上的测试不受影响。</summary>
+    public void ProbeLost(ProbeSession probe)
     {
         foreach (RunningTest running in _running.Values)
         {
-            if (ReferenceEquals(running.Dialer, dialer))
+            if (ReferenceEquals(running.Probe, probe))
             {
                 running.Lost = true;
                 running.Cancel.Cancel();
@@ -147,26 +147,26 @@ public sealed class TestRunner
         int? lastPhase = null;
         try
         {
-            DialerSession dialer = running.Dialer;
+            ProbeSession probe = running.Probe;
             for (int index = 0; index < phases.Count; index++)
             {
                 lastPhase = index;
                 await _uplink.SendTestStatusAsync(test, TestStatuses.Running, index);
-                await RunPhaseAsync(dialer, test, index, phases[index].Phase, phases[index].Decision, ct);
+                await RunPhaseAsync(probe, test, index, phases[index].Phase, phases[index].Decision, ct);
                 await _uplink.SendTestStatusAsync(test, TestStatuses.PhaseDone, index);
             }
         }
         catch (OperationCanceledException)
         {
             finalStatus = running.Lost ? TestStatuses.Failed : TestStatuses.Aborted;
-            errorCode = running.Lost ? MoShiroErrors.DialerOffline.Value : null;
-            await TryAbortDialerAsync(running.Dialer, test);
+            errorCode = running.Lost ? MoShiroErrors.ProbeOffline.Value : null;
+            await TryAbortProbeAsync(running.Probe, test);
         }
         catch (TestFailure failure)
         {
             finalStatus = TestStatuses.Failed;
             errorCode = failure.Code;
-            await TryAbortDialerAsync(running.Dialer, test);
+            await TryAbortProbeAsync(running.Probe, test);
         }
         finally
         {
@@ -178,17 +178,17 @@ public sealed class TestRunner
         await _uplink.SendTestStatusAsync(test, finalStatus, lastPhase, errorCode);
     }
 
-    private async Task RunPhaseAsync(DialerSession dialer, TestId test, int index, TestPhase phase, TaskDecision decision, CancellationToken ct)
+    private async Task RunPhaseAsync(ProbeSession probe, TestId test, int index, TestPhase phase, TaskDecision decision, CancellationToken ct)
     {
         ParamValues values = decision.Resolved!;
-        var stream = new StreamInfo(_streams.Allocate(), phase.Kind, values) { TestId = test, Phase = index, Owner = dialer };
+        var stream = new StreamInfo(_streams.Allocate(), phase.Kind, values) { TestId = test, Phase = index, Owner = probe };
         int durationSeconds = phase.DurationSeconds > 0 ? phase.DurationSeconds : (int)(values.Values.ContainsKey("duration_s") ? values.GetInt("duration_s") : 10);
         var blockSpec = new BlockSpec(decision.Kind!, values, stream.Id, stream.Port, new FragmentScope(null, test, index));
-        stream.Running = RunningBlock.Start(BlockCatalog.Find(phase.Kind)!.Listener!(_env.WithSession(dialer), blockSpec), blockSpec.Signature);
+        stream.Running = RunningBlock.Start(BlockCatalog.Find(phase.Kind)!.Listener!(_env.WithSession(probe), blockSpec), blockSpec.Signature);
         _streams.Add(stream);
         try
         {
-            RpcSession<PeerContext> rpc = dialer.Rpc ?? throw new TestFailure(MoShiroErrors.DialerOffline.Value);
+            RpcSession<PeerContext> rpc = probe.Rpc ?? throw new TestFailure(MoShiroErrors.ProbeOffline.Value);
             var start = new TestPhaseStartParams
             {
                 TestId = test,
@@ -206,7 +206,7 @@ public sealed class TestRunner
             }
 
             // 单向数据流的发送端自己运行到结束（或失败）；其他积木是反应式的，等够时长即可
-            if (phase.Kind == "quic_stream")
+            if (phase.Kind == "udp_train")
             {
                 try
                 {
@@ -214,7 +214,7 @@ public sealed class TestRunner
                 }
                 catch (BlockFaultException)
                 {
-                    throw new TestFailure(MoShiroErrors.DialerOffline.Value);
+                    throw new TestFailure(MoShiroErrors.ProbeOffline.Value);
                 }
             }
             else
@@ -230,7 +230,7 @@ public sealed class TestRunner
                 throw new TestFailure(endError.Code.Value);
             }
 
-            // 列块是经拨号方的发件箱至少一次送达的：全部合并上传之后才进入下一阶段（超时就带着已有的数据继续）
+            // 列块是经 Probe 的发件箱至少一次送达的：全部合并上传之后才进入下一阶段（超时就带着已有的数据继续）
             DateTimeOffset deadline = _time.GetUtcNow() + s_fragmentWait;
             while (stream.FragmentsDone < finished.Fragments && _time.GetUtcNow() < deadline)
             {
@@ -246,13 +246,13 @@ public sealed class TestRunner
         {
             _streams.Remove(stream.Id);
             await stream.Running.StopAsync();
-            dialer.ForgetStream(stream.Id);
+            probe.ForgetStream(stream.Id);
         }
     }
 
-    private async Task TryAbortDialerAsync(DialerSession dialer, TestId test)
+    private async Task TryAbortProbeAsync(ProbeSession probe, TestId test)
     {
-        if (dialer.Rpc is not { } rpc)
+        if (probe.Rpc is not { } rpc)
         {
             return;
         }
@@ -267,15 +267,15 @@ public sealed class TestRunner
         }
     }
 
-    /// <summary>一个正在运行的测试：它在哪个拨号方上、怎样取消、是不是因为拨号方掉线而中止。</summary>
+    /// <summary>一个正在运行的测试：它在哪个 Probe 上、怎样取消、是不是因为 Probe 掉线而中止。</summary>
     private sealed class RunningTest : IDisposable
     {
-        public RunningTest(DialerSession dialer)
+        public RunningTest(ProbeSession probe)
         {
-            Dialer = dialer;
+            Probe = probe;
         }
 
-        public DialerSession Dialer { get; }
+        public ProbeSession Probe { get; }
 
         public CancellationTokenSource Cancel { get; } = new();
 

backend/agent/tests/MoShiro.Agent.Tests/ControlBackoffTests.cs

@@ -1,34 +0,0 @@
-
-namespace MoShiro.Agent.Tests;
-
-public sealed class ControlBackoffTests
-{
-    [Fact]
-    public void Next_UsesRequestedEscalationThenKeepsDoublingToTenMinutes()
-    {
-        var backoff = new ControlBackoff();
-        TimeSpan[] expected =
-        [
-            TimeSpan.FromSeconds(1), TimeSpan.FromSeconds(2), TimeSpan.FromSeconds(4),
-            TimeSpan.FromSeconds(8), TimeSpan.FromSeconds(16),
-            TimeSpan.FromSeconds(32), TimeSpan.FromSeconds(32), TimeSpan.FromSeconds(32),
-            TimeSpan.FromSeconds(32), TimeSpan.FromSeconds(32),
-            TimeSpan.FromMinutes(1), TimeSpan.FromMinutes(2), TimeSpan.FromMinutes(4),
-            TimeSpan.FromMinutes(8), TimeSpan.FromMinutes(10), TimeSpan.FromMinutes(10),
-        ];
-
-        Assert.Equal(expected, expected.Select(_ => backoff.Next()));
-    }
-
-    [Fact]
-    public void Reset_RestartsAtOneSecond()
-    {
-        var backoff = new ControlBackoff();
-        _ = backoff.Next();
-        _ = backoff.Next();
-
-        backoff.Reset();
-
-        Assert.Equal(TimeSpan.FromSeconds(1), backoff.Next());
-    }
-}

backend/agent/tests/MoShiro.Agent.Tests/LinkStateTests.cs

@@ -1,114 +0,0 @@
-using System.Collections.Concurrent;
-using System.Net.Sockets;
-using System.Security.Authentication;
-using Fleet.Core;
-using Fleet.Link;
-using MoShiro.Protocol;
-
-namespace MoShiro.Agent.Tests;
-
-public sealed class LinkStateTests
-{
-    [Fact]
-    public void QuicRedialBackoff_StaysShort_AndRestartsAfterReset()
-    {
-        var backoff = new QuicRedialBackoff();
-        int[] seconds = [1, 2, 3, 5, 5, 5];
-        Assert.Equal(seconds, seconds.Select(_ => (int)backoff.Next().TotalSeconds));
-
-        // 连续失败很多次（路径是死的）：翻倍退避，最长五分钟
-        int[] later = [.. Enumerable.Range(0, 12).Select(_ => (int)backoff.Next().TotalSeconds)];
-        Assert.Equal([5, 5, 5, 10, 20, 40, 80, 160, 300], later[3..]);
-
-        backoff.Reset();
-
-        Assert.Equal(TimeSpan.FromSeconds(1), backoff.Next());
-        TimeSpan jittered = new QuicRedialBackoff().Next(new Random(1));
-        Assert.InRange(jittered.TotalSeconds, 0.8, 1.2);
-    }
-
-    public static TheoryData<Exception, string> Faults => new()
-    {
-        { new OperationCanceledException(), "timeout" },
-        { new SocketException((int)SocketError.ConnectionRefused), "refused" },
-        { new SocketException((int)SocketError.HostNotFound), "dns" },
-        { new IOException("reset", new SocketException((int)SocketError.ConnectionReset)), "reset" },
-        { new AuthenticationException("bad certificate"), "tls" },
-        { new InvalidOperationException("peer authentication failed: unauthorized"), "auth" },
-        { new ControlSilentException("no frames"), "silent" },
-        { new IOException("closed"), "closed" },
-    };
-
-    [Theory]
-    [MemberData(nameof(Faults))]
-    public void LinkFaults_ClassifyTheFailureAndKeepThePhase(Exception exception, string kind)
-    {
-        LinkFault fault = LinkFaults.From(exception, "tls");
-
-        Assert.Equal(kind, fault.Kind);
-        Assert.Equal("tls", fault.Phase);
-    }
-
-    [Fact]
-    public void ControlStatus_LogsEveryTransition_AndReportsTheFaultWhileInBackoff()
-    {
-        var log = new CaptureLog();
-        var status = new LinkStatus("agt_1", "ipv4", TimeProvider.System, log);
-
-        status.SetControl(ControlState.Dialing);
-        status.SetControl(ControlState.Authenticating);
-        status.SetControl(ControlState.Syncing);
-        status.SetControl(ControlState.Up);
-        status.SetControl(ControlState.Backoff, new LinkFault { Phase = "control_session", Kind = "silent" }, attempt: 1, retryIn: TimeSpan.FromSeconds(1));
-
-        Assert.Equal(["authenticating", "syncing", "up", "backoff"], log.Items.Select(x => (string)x["to"]!));
-        LinkHealth health = status.Snapshot(new Outbox(1 << 10), 0, null);
-        Assert.Equal(ControlState.Backoff, health.Control.State);
-        Assert.Equal("silent", health.Control.LastError?.Kind);
-        Assert.InRange(health.Control.RetryInMs ?? -1, 1, 1000);
-
-        status.SetControl(ControlState.Up);
-        Assert.Null(status.Snapshot(new Outbox(1 << 10), 0, null).Control.LastError);
-    }
-
-    [Fact]
-    public void QuicStatus_LogsOnlyTheEdgesOfAnOutage_NotEveryRedial()
-    {
-        var log = new CaptureLog();
-        var status = new LinkStatus("agt_1", "ipv4", TimeProvider.System, log);
-        var fault = new LinkFault { Phase = "quic_handshake", Kind = "timeout" };
-
-        status.SetQuic(QuicState.Connecting);
-        status.SetQuic(QuicState.Ready);
-        status.SetQuic(QuicState.Retrying, fault, 1, TimeSpan.FromSeconds(1));
-        for (int attempt = 1; attempt <= 5; attempt++)
-        {
-            status.SetQuic(QuicState.Connecting, attempt: attempt);
-            status.SetQuic(QuicState.Retrying, fault, attempt + 1, TimeSpan.FromSeconds(5));
-        }
-
-        status.SetQuic(QuicState.Connecting, attempt: 6);
-        status.SetQuic(QuicState.Ready);
-
-        Assert.Equal(["ready", "retrying", "ready"], log.Items.Select(x => (string)x["to"]!));
-        Assert.NotNull(log.Items[^1]["down_ms"]);
-        Assert.True(status.Ready is false);
-        status.SetControl(ControlState.Up);
-        Assert.True(status.Ready);
-    }
-
-    private sealed class CaptureLog : INodeLog
-    {
-        private readonly ConcurrentQueue<Dictionary<string, object?>> _items = new();
-
-        public IReadOnlyList<Dictionary<string, object?>> Items => [.. _items];
-
-        public void Write(NodeLogLevel level, string message, params (string Key, object? Value)[] fields)
-        {
-            if (message == "link state")
-            {
-                _items.Enqueue(fields.ToDictionary(x => x.Key, x => x.Value));
-            }
-        }
-    }
-}

backend/agent/tests/MoShiro.Agent.Tests/OpenWrtUpdaterTests.cs

@@ -1,52 +0,0 @@
-using System.Text.Json;
-using Fleet.Core;
-using Fleet.Hosting;
-using MoShiro.Protocol;
-
-namespace MoShiro.Agent.Tests;
-
-public sealed class OpenWrtUpdaterTests : IDisposable
-{
-    private readonly string _directory = Path.Combine(Path.GetTempPath(), "moshiro-native-" + Guid.NewGuid().ToString("N"));
-
-    [Fact]
-    public void Offer_AndApply_WriteFixedFlags_NotRemoteManifestOrCommands()
-    {
-        var updater = new OpenWrtUpdater(_directory, Path.Combine(_directory, "status.json"), TimeProvider.System, NullNodeLog.Instance);
-        updater.Offer();
-        updater.RequestApply();
-        Assert.Equal("check", File.ReadAllText(Path.Combine(_directory, "check.request")));
-        Assert.Equal("check", File.ReadAllText(Path.Combine(_directory, "apply.request")));
-        Assert.Equal(2, Directory.GetFiles(_directory).Length);
-    }
-
-    [Fact]
-    public void Status_WritesHealthAndReadsHostStatus()
-    {
-        var store = new ConfigStore(_directory);
-        store.Write("status.json", "{\"state\":\"rolled_back\",\"at\":123}"u8.ToArray());
-        var updater = new OpenWrtUpdater(_directory, Path.Combine(_directory, "status.json"), TimeProvider.System, NullNodeLog.Instance);
-        Assert.Equal("rolled_back", updater.Status()?.State);
-        using JsonDocument health = JsonDocument.Parse(File.ReadAllBytes(Path.Combine(_directory, "health.json")));
-        Assert.Equal(BuildInfo.Version, health.RootElement.GetProperty("version").GetString());
-        Assert.Equal(Environment.ProcessId, health.RootElement.GetProperty("pid").GetInt32());
-        Assert.True(health.RootElement.GetProperty("at").GetInt64() > 0);
-    }
-
-    [Fact]
-    public void Status_MissingOrCorruptHostStatus_DoesNotStopCollector()
-    {
-        var updater = new OpenWrtUpdater(_directory, Path.Combine(_directory, "status.json"), TimeProvider.System, NullNodeLog.Instance);
-        Assert.Null(updater.Status());
-        new ConfigStore(_directory).Write("status.json", "not json"u8.ToArray());
-        Assert.Null(updater.Status());
-    }
-
-    public void Dispose()
-    {
-        if (Directory.Exists(_directory))
-        {
-            Directory.Delete(_directory, recursive: true);
-        }
-    }
-}

backend/agent/tests/MoShiro.Agent.Tests/TaskLinkTests.cs

@@ -1,58 +0,0 @@
-using System.Net;
-using System.Security.Cryptography;
-using System.Security.Cryptography.X509Certificates;
-using Fleet.Core;
-using Fleet.Link;
-using MoShiro.Measurement;
-using MoShiro.Protocol;
-
-namespace MoShiro.Agent.Tests;
-
-public sealed class TaskLinkTests
-{
-    [Fact]
-    public async Task QuicConnectionLoss_IsRedialedFromANewPort_WhileTheControlLayerStaysUp()
-    {
-        using ECDsa key = ECDsa.Create(ECCurve.NamedCurves.nistP256);
-        var request = new CertificateRequest("CN=localhost", key, HashAlgorithmName.SHA256);
-        using X509Certificate2 certificate = request.CreateSelfSigned(DateTimeOffset.UnixEpoch, DateTimeOffset.UnixEpoch.AddYears(100));
-        var session = new DataSession(1, new byte[32]);
-        var clock = new MonoClock(TimeProvider.System);
-        await using var server = new QuicDatagramTransport(new SingleSession(session), null, clock, new DataPlaneCounters(), NullNodeLog.Instance,
-            IPAddress.Loopback, [0], certificate);
-        server.Bind();
-        await using var client = new QuicDatagramTransport(session, clock, new DataPlaneCounters(), NullNodeLog.Instance,
-            new IPEndPoint(IPAddress.Loopback, server.BoundPorts[0]), key.ExportSubjectPublicKeyInfo());
-        var status = new LinkStatus("agt_1", "ipv4", TimeProvider.System, NullNodeLog.Instance);
-        status.SetControl(ControlState.Up);
-        using var link = new TaskLink(client, status, TimeProvider.System);
-        using var stop = new CancellationTokenSource(TimeSpan.FromSeconds(30));
-        Task run = link.RunAsync(stop.Token);
-
-        await WaitUntilAsync(() => status.Quic == QuicState.Ready, stop.Token);
-        int firstPort = client.BoundPorts[0];
-        await client.CloseClientAsync();
-
-        bool controlStayedUp = true;
-        await WaitUntilAsync(() =>
-        {
-            controlStayedUp &= status.Control == ControlState.Up;
-            return status.Quic == QuicState.Ready && client.BoundPorts[0] != firstPort;
-        }, stop.Token);
-
-        Assert.True(controlStayedUp);
-        LinkHealth health = status.Snapshot(new Outbox(1 << 10), 0, null);
-        Assert.Equal(2, health.Task.Quic.HandshakeSuccesses);
-        Assert.Equal(ControlState.Up, health.Control.State);
-        await stop.CancelAsync();
-        await run.WaitAsync(TimeSpan.FromSeconds(5));
-    }
-
-    private static async Task WaitUntilAsync(Func<bool> condition, CancellationToken ct)
-    {
-        while (!condition())
-        {
-            await Task.Delay(20, ct);
-        }
-    }
-}

backend/fleet/README.md

@@ -3,7 +3,7 @@
 「很多节点连到一个面板」这类产品都要解决同一批问题：节点怎么接入、怎么证明身份、怎么保持连接、怎么可靠上报、怎么接收配置。Fleet 把这些做成与业务无关的库，MoShiro 是它的第一个使用者；产品只写业务。Fleet 不引用任何 `MoShiro.*`，也不出现产品的名字、端口和协议前缀，由 `ArchitectureTests` 在构建时检查。
 
 ```text
-┌──────────────────── 产品层（MoShiro：Protocol · Measurement · Agent · Backend）────────────────────┐
+┌──────────────────── 产品层（MoShiro：Protocol · Measurement · Agent · Probe · Backend）────────────────────┐
 └────────────────────────────────────────────┬─────────────────────────────────────────────────────────────┘
                                              │ 只依赖 Fleet
 ┌───────────────────────────── Fleet ────────┴─────────────────────────────────────────────────────────────┐
@@ -19,11 +19,11 @@
 
 ## 设计原则
 
-1. 与业务无关：产品通过扩展点接入：元数据类型、RPC 方法、发件箱消息类型、期望状态文档类型。
-2. 安全默认值：Fleet 本身不提供 exec、shell、文件读写、远程下发代码这类方法；节点只响应本地注册过的方法（本地策略 `allow_methods` 还能收窄）；调用对方不支持的方法在本地直接返回 `fleet_unsupported`。
-3. 权限分离：注册（`fleet-enroll`，一次性）和常驻业务进程是两个进程；常驻进程不含注册代码，也不使用 `HttpClient`。
-4. NativeAOT：节点侧只依赖 BCL，泛型扩展点通过产品传入的 `JsonTypeInfo<T>` 序列化，不使用反射。
-5. 更新不在 Fleet 里：Fleet 不包含更新逻辑，更新由节点外部的服务管理器或更新器完成。Quadlet 装的节点和 Backend 由退出后 systemd 重新拉镜像、脚本装的节点由宿主机上的更新器验签更新，见 [`docs/ops/deployment.md`](../../docs/ops/deployment.md)；纯二进制节点的更新见 [路线图](../../docs/roadmap.md)。
+1. **与业务无关**：产品通过扩展点接入：元数据类型、RPC 方法、发件箱消息类型、期望状态文档类型。
+2. **安全默认值**：Fleet 永远不提供 exec、shell、文件读写、远程下发代码这类方法；节点只响应本地注册过的方法（本地策略 `allow_methods` 还能收窄）；调用对方不支持的方法在本地直接返回 `fleet_unsupported`。
+3. **权限分离**：注册（`fleet-enroll`，一次性）和常驻业务进程是两个进程；常驻进程不含注册代码，也不使用 `HttpClient`。
+4. **NativeAOT**：节点侧只依赖 BCL，泛型扩展点通过产品传入的 `JsonTypeInfo<T>` 序列化，不使用反射。
+5. **更新不在 Fleet 里**：节点进程没有自更新能力。容器节点的签名更新由宿主机上的更新器完成，见 [`docs/deployment.md`](../../docs/deployment.md)；纯二进制节点的更新是待办，见 [`TODO.md`](../../TODO.md)。
 
 ## 身份
 
@@ -56,7 +56,7 @@
 - JSON-RPC 2.0；方法名 `领域.动作`，`fleet.*` 保留给框架，产品用自己的前缀（MoShiro 是 `moshiro.*`）。
 - 二进制帧 `u8 channel | 正文`：`0`–`15` 框架保留，产品从 `16` 开始。
 - 错误：`error.data.code` 为 snake_case，框架的以 `fleet_` 开头，产品可以扩展。
-- 方法必须声明：每个方法在声明表（`MethodTable`）里写明发起方、请求还是通知、哪些角色可以参与；不存在双向方法，同一个方法名在一个平面里只有一个发起方。`RpcRouter` 在注册、分发和发送三处核对，违规回 `fleet_protocol_violation`，超过上限断开连接。详见 [`docs/security/security.md`](../../docs/security/security.md) 第 6 节。
+- **方法必须声明**：每个方法在声明表（`MethodTable`）里写明发起方、请求还是通知、哪些角色可以参与；不存在双向方法，同一个方法名在一个平面里只有一个发起方。`RpcRouter` 在注册、分发和发送三处核对，违规回 `fleet_protocol_violation`，超过上限断开连接。详见 [`docs/security.md`](../../docs/security.md) 第 8 节。
 - 与传输无关：同一套路由跑在 wss 上，也跑在 TLS 帧（`u32 长度 | u8 类型 | 正文`）上。
 
 | 保留方法 | 方向 | 说明 |
@@ -64,7 +64,6 @@
 | `fleet.challenge` / `fleet.auth` | 服务端→节点 / 节点→服务端 | 签名认证（`auth` 用途，nonce 30 秒内有效） |
 | `fleet.hello` | 节点→服务端 | 版本、能力、本地策略摘要、当前时间 → 返回时钟偏差；产品的元数据也放在这里 |
 | `fleet.heartbeat` | 节点→服务端 | 每 10 秒一次，产品可以附加计数 |
-| `fleet.stopping` | 节点→服务端 | 有序关闭前通知服务端，携带 `reason`（例如 `stopping` 或 `updating`），用于准确区分离线和计划停机 |
 | `fleet.state.sync` | 服务端→节点 | `{version, doc}` → `{applied_version, rejected[]}` |
 | `fleet.outbox.ack` | 服务端→节点 | `{ids[], duplicates[]}` |
 | `fleet.revoked` | 服务端→节点 | 节点被禁用或接管，随后断开 |
@@ -79,14 +78,14 @@ connect(wss) → fleet.challenge → fleet.auth → fleet.hello → 状态同步
 
 本地策略：`allow_methods`、`max_frame_bytes`、`max_inflight`。
 
-可靠发件箱：产品 `EnqueueAsync` → 有界队列（按字节和条数；满时 `DropOldest` 或 `Reject`）→ 发送 → 服务端收件箱（`IInbox`）持久化后 `fleet.outbox.ack` → 移除。至少一次，按 `msg_id`（UUIDv7）去重；在途数据默认限 4 MiB；丢弃会累计缺口，通过 `OnGap` 通知产品。只在内存里，不落盘。
+**可靠发件箱**：产品 `EnqueueAsync` → 有界队列（按字节和条数；满时 `DropOldest` 或 `Reject`）→ 发送 → 服务端收件箱（`IInbox`）持久化后 `fleet.outbox.ack` → 移除。至少一次，按 `msg_id`（UUIDv7）去重；在途数据默认限 4 MiB；丢弃会累计缺口，通过 `OnGap` 通知产品。只在内存里，不落盘。
 
-期望状态同步：服务端 `IStateProvider<TDoc>` 给每个节点算 `{version, doc}`，变化时 `NotifyChanged` 推送；节点 `StateHandler<TDoc>` 调和后返回 `{applied_version, rejected[]}`。下发的是完整快照，天然幂等，重连后服务端总会重发当前版本。`rejected[]` 带结构化原因（`unsupported`、`budget_exceeded`、`policy_denied`），面板可以直接展示。
+**期望状态同步**：服务端 `IStateProvider<TDoc>` 给每个节点算 `{version, doc}`，变化时 `NotifyChanged` 推送；节点 `StateHandler<TDoc>` 调和后返回 `{applied_version, rejected[]}`。下发的是**完整快照**，天然幂等，重连后服务端总会重发当前版本。`rejected[]` 带结构化原因（`unsupported`、`budget_exceeded`、`policy_denied`），面板可以直接展示。
 
 ## 点对点会话
 
 ```text
-节点 A（发起方，如 Agent A）── TLS（按 SPKI 校验证书）──> 节点 B（监听方，如 Agent B）
+节点 A（发起方，如 Probe）── TLS（按 SPKI 校验证书）──> 节点 B（监听方，如 Agent）
     <── fleet.peer.challenge ──
     ── fleet.peer.auth（peer 用途，A 的私钥签名）──>
     B 用服务端下发的对端公钥列表验签（经 state.sync 下发），之后复用同一套 Rpc：路由、二进制通道、发件箱
@@ -100,6 +99,6 @@ connect(wss) → fleet.challenge → fleet.auth → fleet.hello → 状态同步
 
 - `ModuleHost`：按依赖顺序启动模块、相反顺序停止，隔离故障，重启时退避。
 - `NodeConfig`：源生成 JSON 读取，环境变量覆盖，启动时校验。
-- `ConfigStore`：常驻进程里唯一允许写文件的地方，只能写配置目录，写入原子替换、权限 0600；父目录不存在会创建。
+- `ConfigStore`：常驻进程里**唯一**允许写文件的地方，只能写配置目录，写入原子替换、权限 0600；父目录不存在会创建。
 - `ServiceShell`：SIGTERM、Ctrl+C，systemd `sd_notify`、OpenWrt procd（前台运行）和 Windows 服务。
 - 服务端接口：`IFleetStore`（节点、令牌、公钥、状态版本）、`IStateProvider<TDoc>`、`IInbox`、`INodeEventSink`、`IEnrollmentHook`。MoShiro 把 `IFleetStore` 放进自己的 `moshiro.db`，和业务数据由同一个单写入者写入。

backend/fleet/src/Fleet.Core/BinaryAttestation.cs

@@ -0,0 +1,43 @@
+using System.Security.Cryptography;
+
+namespace Fleet.Core;
+
+/// <summary>
+/// 节点握手时自述的「我跑的是哪个二进制」：正在运行的可执行文件的 SHA-256，以及（容器节点）宿主机更新器验过的 CI 签名清单和签名。
+/// 服务端在节点第一次握手成功的那一刻把哈希绑定到这个节点，之后哈希对不上就不让它继续运行。
+/// </summary>
+public sealed record BinaryAttestation(string Sha256, string? Manifest, string? Signature);
+
+/// <summary>
+/// 正在运行的二进制的哈希。Linux 上读 <c>/proc/self/exe</c>：它指向进程启动时映射的那个映像，磁盘上的文件之后被换掉也不影响，
+/// 所以「运行期间被替换、重启后换了别的」的情况下一次握手就会对不上。算一次，之后缓存。
+/// </summary>
+public static class NodeBinary
+{
+    private static readonly Lazy<string?> s_hash = new(Compute);
+
+    /// <summary>小写十六进制；读不到可执行文件时为 null（这样的节点不做绑定）。</summary>
+    public static string? Sha256Hex => s_hash.Value;
+
+    /// <summary>在后台先算好，免得第一次握手等它。</summary>
+    public static void Prime() => _ = Task.Run(() => s_hash.Value);
+
+    private static string? Compute()
+    {
+        try
+        {
+            string? path = OperatingSystem.IsLinux() && File.Exists("/proc/self/exe") ? "/proc/self/exe" : Environment.ProcessPath;
+            if (path is null)
+            {
+                return null;
+            }
+
+            using var stream = new FileStream(path, FileMode.Open, FileAccess.Read, FileShare.ReadWrite | FileShare.Delete, 1 << 16);
+            return Convert.ToHexStringLower(SHA256.HashData(stream));
+        }
+        catch (Exception ex) when (ex is IOException or UnauthorizedAccessException)
+        {
+            return null;
+        }
+    }
+}

backend/fleet/src/Fleet.Core/Bounded.cs

@@ -1,32 +0,0 @@
-using System.Net.Sockets;
-using System.Net.WebSockets;
-
-namespace Fleet.Core;
-
-/// <summary>
-/// 收尾类的等待（关闭、释放、等任务结束）一律要有上限：对端没了、原生层不回调、某个处理器卡住时，
-/// 没有上限的等待会让状态机永远停在「已连接」，既不重连也不报错。超时就放弃等待并交给调用方记日志；会话结束时的常见异常按正常噪声吞掉。
-/// </summary>
-public static class Bounded
-{
-    /// <summary>收尾的一步最多等这么久。</summary>
-    public static readonly TimeSpan Teardown = TimeSpan.FromSeconds(5);
-
-    /// <summary>等 <paramref name="work"/> 最多 <paramref name="limit"/>；超时返回 false（任务继续在后台收尾，不再等它），否则返回 true。</summary>
-    public static async Task<bool> WaitAsync(Task work, TimeSpan limit, TimeProvider time)
-    {
-        try
-        {
-            await work.WaitAsync(limit, time);
-            return true;
-        }
-        catch (TimeoutException)
-        {
-            return false;
-        }
-        catch (Exception ex) when (ex is OperationCanceledException or IOException or ObjectDisposedException or InvalidOperationException or SocketException or WebSocketException)
-        {
-            return true; // 会话结束时的正常噪声：这一步已经结束了
-        }
-    }
-}

backend/fleet/src/Fleet.Core/ContextNodeLog.cs

@@ -1,8 +0,0 @@
-namespace Fleet.Core;
-
-/// <summary>Adds stable diagnostic context to every event in one operation.</summary>
-public sealed class ContextNodeLog(INodeLog inner, params (string Key, object? Value)[] context) : INodeLog
-{
-    public void Write(NodeLogLevel level, string message, params (string Key, object? Value)[] fields) =>
-        inner.Write(level, message, [.. context, .. fields]);
-}

backend/fleet/src/Fleet.Core/ErrorCode.cs

@@ -27,7 +27,6 @@ public static class FleetErrors
     public static readonly ErrorCode JoinTokenExpired = new("fleet_join_token_expired");
     public static readonly ErrorCode ProofInvalid = new("fleet_proof_invalid");
     public static readonly ErrorCode FrameTooLarge = new("fleet_frame_too_large");
-
-    /// <summary>对方的版本低于要求的最低版本：等一会儿（对方可能正在更新）再握手。</summary>
-    public static readonly ErrorCode VersionRequired = new("fleet_version_required");
+    /// <summary>节点现在跑的二进制和它第一次握手时绑定的不一致，也不在 CI 签名的更新清单里。</summary>
+    public static readonly ErrorCode BinaryMismatch = new("fleet_binary_mismatch");
 }

backend/fleet/src/Fleet.Core/NodeSupport.cs

@@ -9,35 +9,12 @@ public enum NodeLogLevel
     Error,
 }
 
-/// <summary>节点日志：字段名统一为 logfmt（<c>ts level msg agent dialer session task result err</c>）。逐包事件不记录，只累加计数。</summary>
+/// <summary>节点日志：字段名统一为 logfmt（<c>ts level msg agent probe session task result err</c>）。逐包事件不记录，只累加计数。</summary>
 public interface INodeLog
 {
     void Write(NodeLogLevel level, string message, params (string Key, object? Value)[] fields);
 }
 
-/// <summary>One structured node log line.</summary>
-public sealed record NodeLogRecord(long AtUtcMs, string Level, string Message, Dictionary<string, string> Fields);
-
-/// <summary>A retained log line: <paramref name="Id"/> grows with every line, <paramref name="Type"/> is a stable hash of the message text (the template), so counters can be sent without the text.</summary>
-public sealed record NodeLogEntry(long Id, uint Type, NodeLogRecord Record);
-
-/// <summary>How many lines of one type and level were written since the last summary.</summary>
-public sealed record NodeLogTypeCount(uint Type, string Level, long Count);
-
-/// <summary>What a node tells the control plane about its logs without sending any text: counters since the last summary and the newest line id.</summary>
-public sealed record NodeLogSummary(long LastId, long Info, long Warn, long Error, IReadOnlyList<NodeLogTypeCount> Types);
-
-public sealed record NodeLogPage(IReadOnlyList<NodeLogEntry> Items, long LastId);
-
-/// <summary>Bounded local log retention. The control plane gets counters in heartbeats and pulls the text when it needs it.</summary>
-public interface INodeLogSource
-{
-    NodeLogSummary TakeSummary();
-
-    /// <summary>Lines with an id above <paramref name="afterId"/>, oldest first, at most <paramref name="limit"/>, at or above <paramref name="minLevel"/> (info, warn or error).</summary>
-    NodeLogPage Read(long afterId, int limit, string? minLevel);
-}
-
 public sealed class NullNodeLog : INodeLog
 {
     public static NullNodeLog Instance { get; } = new();

backend/fleet/src/Fleet.Core/NodeVersions.cs

@@ -1,87 +0,0 @@
-using System.Globalization;
-
-namespace Fleet.Core;
-
-/// <summary>
-/// 版本号比较，给「要求的最低版本」用。认得 <c>主.次.修订</c>，可以带预发布后缀；开发构建 <c>0.4.2-dev.73-5245f03</c> 按 <c>dev.73</c> 的序号比较，
-/// 提交哈希不参与比较。同一个主次修订里，正式版高于任何预发布版。认不出来的版本号不拒绝（返回 false），免得一个奇怪的版本字符串把节点挡在门外。
-/// </summary>
-public static class NodeVersions
-{
-    private readonly record struct Parsed(int Major, int Minor, int Patch, bool Prerelease, long Dev, string Suffix);
-
-    /// <summary><paramref name="actual"/> 低于 <paramref name="required"/> 返回 true；任何一个为空或无法解析返回 false。</summary>
-    public static bool IsBelow(string? actual, string? required)
-    {
-        if (string.IsNullOrWhiteSpace(required) || string.IsNullOrWhiteSpace(actual))
-        {
-            return false;
-        }
-
-        if (!TryParse(actual, out Parsed have) || !TryParse(required, out Parsed need))
-        {
-            return false;
-        }
-
-        return Compare(have, need) < 0;
-    }
-
-    private static int Compare(Parsed a, Parsed b)
-    {
-        int core = (a.Major, a.Minor, a.Patch).CompareTo((b.Major, b.Minor, b.Patch));
-        if (core != 0)
-        {
-            return core;
-        }
-
-        if (a.Prerelease != b.Prerelease)
-        {
-            return a.Prerelease ? -1 : 1; // 正式版高于预发布版
-        }
-
-        if (!a.Prerelease)
-        {
-            return 0;
-        }
-
-        int dev = a.Dev.CompareTo(b.Dev);
-        return dev != 0 ? dev : 0;
-    }
-
-    private static bool TryParse(string text, out Parsed parsed)
-    {
-        parsed = default;
-        string value = text.Trim().TrimStart('v');
-        int plus = value.IndexOf('+', StringComparison.Ordinal);
-        if (plus >= 0)
-        {
-            value = value[..plus];
-        }
-
-        int dash = value.IndexOf('-', StringComparison.Ordinal);
-        string core = dash < 0 ? value : value[..dash];
-        string suffix = dash < 0 ? "" : value[(dash + 1)..];
-        string[] parts = core.Split('.');
-        if (parts.Length != 3
-            || !int.TryParse(parts[0], NumberStyles.None, CultureInfo.InvariantCulture, out int major)
-            || !int.TryParse(parts[1], NumberStyles.None, CultureInfo.InvariantCulture, out int minor)
-            || !int.TryParse(parts[2], NumberStyles.None, CultureInfo.InvariantCulture, out int patch))
-        {
-            return false;
-        }
-
-        long dev = 0;
-        if (suffix.StartsWith("dev.", StringComparison.Ordinal))
-        {
-            string number = suffix[4..];
-            int end = number.IndexOf('-', StringComparison.Ordinal);
-            if (!long.TryParse(end < 0 ? number : number[..end], NumberStyles.None, CultureInfo.InvariantCulture, out dev))
-            {
-                return false;
-            }
-        }
-
-        parsed = new Parsed(major, minor, patch, suffix.Length > 0, dev, suffix);
-        return true;
-    }
-}

backend/fleet/src/Fleet.Core/VarInt.cs

@@ -51,4 +51,16 @@ public static class VarInt
 
         return false;
     }
+
+    public static bool TryReadUInt32LittleEndian(ReadOnlySpan<byte> source, out uint value)
+    {
+        if (source.Length < 4)
+        {
+            value = 0;
+            return false;
+        }
+
+        value = BinaryPrimitives.ReadUInt32LittleEndian(source);
+        return true;
+    }
 }

backend/fleet/src/Fleet.Hosting/BufferedNodeLog.cs

@@ -1,147 +0,0 @@
-using System.Globalization;
-using Fleet.Core;
-
-namespace Fleet.Hosting;
-
-/// <summary>
-/// Writes logs locally and keeps a bounded ring of recent lines. Heartbeats carry only counters (see <see cref="TakeSummary"/>);
-/// the text leaves the node when the control plane asks for it (see <see cref="Read"/>).
-/// </summary>
-public sealed class BufferedNodeLog : INodeLog, INodeLogSource
-{
-    private const int Capacity = 2048;
-    private const int MaxFields = 20;
-    private const int MaxMessageLength = 512;
-    private const int MaxFieldValueLength = 256;
-    private const int MaxPageBytes = 96 << 10;
-    private const int MaxSummaryTypes = 8;
-
-    private readonly object _gate = new();
-    private readonly LinkedList<NodeLogEntry> _ring = [];
-    private readonly Dictionary<(uint Type, string Level), long> _counts = [];
-    private readonly INodeLog _local;
-    private readonly TimeProvider _time;
-    private long _lastId;
-    private long _info;
-    private long _warn;
-    private long _error;
-
-    public BufferedNodeLog(INodeLog local, TimeProvider time)
-    {
-        _local = local;
-        _time = time;
-    }
-
-    public void Write(NodeLogLevel level, string message, params (string Key, object? Value)[] fields)
-    {
-        _local.Write(level, message, fields);
-
-        var safeFields = new Dictionary<string, string>(Math.Min(fields.Length, MaxFields), StringComparer.Ordinal);
-        foreach ((string key, object? value) in fields.Take(MaxFields))
-        {
-            string safeValue = IsSensitiveKey(key) ? "[redacted]" : Convert.ToString(value, CultureInfo.InvariantCulture) ?? "";
-            safeFields[key.Length <= 64 ? key : key[..64]] = Truncate(safeValue, MaxFieldValueLength);
-        }
-
-        string levelName = level.ToString().ToLowerInvariant();
-        var record = new NodeLogRecord(_time.GetUtcNow().ToUnixTimeMilliseconds(), levelName, Truncate(message, MaxMessageLength), safeFields);
-        uint type = TypeOf(record.Message);
-        lock (_gate)
-        {
-            _ring.AddLast(new NodeLogEntry(++_lastId, type, record));
-            if (_ring.Count > Capacity)
-            {
-                _ring.RemoveFirst();
-            }
-
-            _counts[(type, levelName)] = _counts.GetValueOrDefault((type, levelName)) + 1;
-            switch (level)
-            {
-                case NodeLogLevel.Error:
-                    _error++;
-                    break;
-                case NodeLogLevel.Warn:
-                    _warn++;
-                    break;
-                default:
-                    _info++;
-                    break;
-            }
-        }
-    }
-
-    public NodeLogSummary TakeSummary()
-    {
-        lock (_gate)
-        {
-            var summary = new NodeLogSummary(
-                _lastId, _info, _warn, _error,
-                [.. _counts.OrderByDescending(pair => pair.Value).Take(MaxSummaryTypes).Select(pair => new NodeLogTypeCount(pair.Key.Type, pair.Key.Level, pair.Value))]);
-            _counts.Clear();
-            _info = _warn = _error = 0;
-            return summary;
-        }
-    }
-
-    public NodeLogPage Read(long afterId, int limit, string? minLevel)
-    {
-        int rank = LevelRank(minLevel);
-        lock (_gate)
-        {
-            var items = new List<NodeLogEntry>(Math.Min(Math.Max(limit, 0), 200));
-            int bytes = 0;
-            foreach (NodeLogEntry entry in _ring)
-            {
-                if (entry.Id <= afterId || LevelRank(entry.Record.Level) < rank)
-                {
-                    continue;
-                }
-
-                int entryBytes = 64 + entry.Record.Message.Length + entry.Record.Fields.Sum(pair => pair.Key.Length + pair.Value.Length + 8);
-                if (items.Count >= Math.Clamp(limit, 0, 200) || (items.Count > 0 && bytes + entryBytes > MaxPageBytes))
-                {
-                    break;
-                }
-
-                items.Add(entry);
-                bytes += entryBytes;
-            }
-
-            return new NodeLogPage(items, _lastId);
-        }
-    }
-
-    /// <summary>The line's type: a stable 32-bit FNV-1a hash of its message text, so the same template always maps to the same number.</summary>
-    public static uint TypeOf(string message)
-    {
-        uint hash = 2166136261;
-        foreach (char c in message)
-        {
-            hash = (hash ^ c) * 16777619;
-        }
-
-        return hash;
-    }
-
-    private static int LevelRank(string? level) => level switch
-    {
-        "error" => 2,
-        "warn" => 1,
-        _ => 0,
-    };
-
-    private static bool IsSensitiveKey(string key)
-    {
-        string lower = key.ToLowerInvariant();
-        return lower.Contains("token", StringComparison.Ordinal)
-            || lower.Contains("secret", StringComparison.Ordinal)
-            || lower.Contains("password", StringComparison.Ordinal)
-            || lower.Contains("private", StringComparison.Ordinal)
-            || lower.Contains("signature", StringComparison.Ordinal)
-            || lower.Contains("authorization", StringComparison.Ordinal)
-            || lower.Contains("cookie", StringComparison.Ordinal)
-            || lower == "mac";
-    }
-
-    private static string Truncate(string value, int maxLength) => value.Length <= maxLength ? value : value[..maxLength];
-}

backend/fleet/src/Fleet.Hosting/Hosting.cs

@@ -156,9 +156,6 @@ public static class ServiceShell
             log.Info("stopping", ("result", "SIGHUP"));
             cts.Cancel();
         });
-        // 有序退出卡住（比如某个已经死掉的会话收不到对端的关闭回复）时不能一直挂着：supervisor 的 stop 超时之前自己退出，由它重新拉起
-        using var exited = new CancellationTokenSource();
-        using CancellationTokenRegistration deadline = cts.Token.Register(() => _ = ExitIfStuckAsync(time ?? TimeProvider.System, log, exited.Token));
         Task trimmer = TrimHeapAsync(cts.Token);
         Task scheduled = restart is null ? Task.CompletedTask : WaitForRestartAsync(restart, time ?? TimeProvider.System, log, cts);
         try
@@ -174,28 +171,9 @@ public static class ServiceShell
             await cts.CancelAsync();
             await trimmer;
             await scheduled;
-            await exited.CancelAsync();
         }
     }
 
-    /// <summary>退出请求发出之后给有序退出的时间（要短于容器的 StopTimeout，否则 supervisor 会先强杀）。</summary>
-    public static readonly TimeSpan ShutdownGrace = TimeSpan.FromSeconds(10);
-
-    private static async Task ExitIfStuckAsync(TimeProvider time, INodeLog log, CancellationToken exited)
-    {
-        try
-        {
-            await Task.Delay(ShutdownGrace, time, exited);
-        }
-        catch (OperationCanceledException)
-        {
-            return; // 正常退出了
-        }
-
-        log.Warn("orderly shutdown did not finish in time, exiting", ("grace_s", (long)ShutdownGrace.TotalSeconds));
-        Environment.Exit(0);
-    }
-
     private static async Task WaitForRestartAsync(RestartController restart, TimeProvider time, INodeLog log, CancellationTokenSource cts)
     {
         while (!cts.IsCancellationRequested)
@@ -203,7 +181,7 @@ public static class ServiceShell
             // 先取令牌、再取设置：中间被改动的话令牌会被取消，下一轮重新排
             CancellationToken changed = restart.ChangeToken;
             RestartSchedule? schedule = restart.Current;
-            using CancellationTokenSource wait = CancellationTokenSource.CreateLinkedTokenSource(cts.Token, changed, restart.NowToken);
+            using CancellationTokenSource wait = CancellationTokenSource.CreateLinkedTokenSource(cts.Token, changed);
             try
             {
                 if (schedule is null)
@@ -218,13 +196,6 @@ public static class ServiceShell
             }
             catch (OperationCanceledException)
             {
-                if (restart.NowReason is { } reason)
-                {
-                    log.Info("stopping", ("result", reason));
-                    await cts.CancelAsync();
-                    return;
-                }
-
                 continue; // 进程在退出，或者设置被改了：循环条件和下一轮会处理
             }
 
@@ -287,7 +258,7 @@ public static class NodeConfig
 
 /// <summary>
 /// 常驻业务进程里唯一允许写文件的地方，而且只能写配置/身份目录（原子替换，类 Unix 上权限 0600）。
-/// 例如 拨号方 把最新的 Agent 列表写回 bootstrap。测量数据本来就不落盘。
+/// 例如 Probe 把最新的 Agent 列表写回 bootstrap。测量数据本来就不落盘。
 /// </summary>
 public sealed class ConfigStore
 {

backend/fleet/src/Fleet.Hosting/OwnerGuard.cs

@@ -3,7 +3,7 @@ using Fleet.Core;
 namespace Fleet.Hosting;
 
 /// <summary>
-/// 节点自己要求有属主（passkey）：拨号方和监听方共用这一份，启动时检查。
+/// 节点自己要求有属主（passkey）：Agent 和 Probe 共用这一份，启动时检查。
 /// <para>
 /// 属主是安装命令带来、注册时固定在 <c>trust.json</c> 里的 passkey 凭证 id（必须是操作员集合里的凭证）。节点据此知道自己绑定在哪个 passkey 上，
 /// 才敢把更多的接口和操作交给它（extension 会话、加密内容都只认属主的 passkey）；没有属主的节点没有人能授权它的敏感能力，所以不运行。

backend/fleet/src/Fleet.Hosting/RestartController.cs

@@ -8,9 +8,7 @@ public sealed class RestartController : IDisposable
     private readonly RestartSchedule? _local;
     private readonly object _gate = new();
     private RestartSchedule? _remote;
-    private readonly CancellationTokenSource _now = new();
     private CancellationTokenSource _changed = new();
-    private string? _nowReason;
 
     public RestartController(RestartSchedule? local = null)
     {
@@ -43,32 +41,6 @@ public sealed class RestartController : IDisposable
         }
     }
 
-    /// <summary>应用要求现在就有序退出（比如管理员请求更新）；退出的原因，没有就是 null。</summary>
-    public string? NowReason
-    {
-        get
-        {
-            lock (_gate)
-            {
-                return _nowReason;
-            }
-        }
-    }
-
-    /// <summary>应用要求现在就退出的令牌。</summary>
-    public CancellationToken NowToken => _now.Token;
-
-    /// <summary>要求进程现在就有序退出，由服务管理器拉起；不管有没有设置定时重启。</summary>
-    public void RequestNow(string reason)
-    {
-        lock (_gate)
-        {
-            _nowReason ??= reason;
-        }
-
-        _now.Cancel();
-    }
-
     /// <summary>面板下发的设置；为空表示面板关闭了定时重启。返回 false 表示文本不合法（节点忽略它）。</summary>
     public bool SetRemote(string? text)
     {
@@ -96,7 +68,6 @@ public sealed class RestartController : IDisposable
         lock (_gate)
         {
             _changed.Dispose();
-            _now.Dispose();
         }
     }
 }

backend/fleet/src/Fleet.Hosting/TrustAnchor.cs

@@ -25,7 +25,7 @@ public sealed partial class TrustFileContext : JsonSerializerContext;
 /// <summary>
 /// 节点的信任根：安装命令里带来的创世指纹和属主，在注册时固定到本地，之后 Backend 改不了。
 /// 操作员集合的新版本随期望状态下发，节点只接受从固定的创世出发、签名完整、版本更新的链。
-/// 属主只是被固定和保存：节点现在没有 extension，等做 extension 时由节点本地强制（见 docs/security/access.md 第 9 节）。
+/// 属主只是被固定和保存：节点现在没有 extension，等做 extension 时由节点本地强制（见 docs/access.md 第 8 节）。
 /// </summary>
 public sealed class TrustAnchor : IServerVerifier
 {

backend/fleet/src/Fleet.Identity/OperatorTrust.cs

@@ -20,7 +20,7 @@ public sealed record OperatorKey(string CredentialId, byte[] PublicKeySpki);
 public sealed record OperatorSet(int Version, string Previous, string RpId, IReadOnlyList<string> Origins, IReadOnlyList<OperatorKey> Keys, IReadOnlyList<byte[]>? Servers = null);
 
 /// <summary>
-/// 信任根：操作员（passkey）→ 操作员集合 → 节点本地固定的属主。设计见 docs/security/access.md 第 8 节。
+/// 信任根：操作员（passkey）→ 操作员集合 → 节点本地固定的属主。设计见 docs/access.md 第 7 节。
 /// 安装命令里带着创世指纹和属主（passkey 的凭证 id）：节点把它们固定在本地，之后只认从创世出发、由操作员 passkey 签过名的集合版本，
 /// 属主必须是集合里的凭证。Backend 能转发集合，但签不出新的版本，也改不了节点已经固定的东西。
 /// 全是纯函数，只用 BCL，节点和 Backend 共用。语句是 <c>key=value</c> 的文本，第一行是带版本的标题；签名覆盖整段文本。

backend/fleet/src/Fleet.Identity/WebAuthnAssertion.cs

@@ -7,7 +7,7 @@ namespace Fleet.Identity;
 
 /// <summary>
 /// 验证一次 WebAuthn 断言（<c>navigator.credentials.get</c> 的结果）。只支持 ES256（P-256 的 ECDSA），纯函数，只用 BCL：
-/// 面板用它验证 passkey 登录，节点验证操作员集合的签名也用它（docs/security/access.md 第 8 节）。
+/// 面板用它验证 passkey 登录，节点以后用它验证 passkey 签出的委托（docs/access.md 第 7 节）。
 /// 不解析 CBOR——注册时的 attestation 解析只在 Backend 里。
 /// </summary>
 public static class WebAuthnAssertion

backend/fleet/src/Fleet.Link/FleetLink.cs

@@ -1,6 +1,3 @@
-using System.Net;
-using System.Net.Http;
-using System.Net.Sockets;
 using System.Net.WebSockets;
 using System.Text.Json;
 using System.Text.Json.Serialization.Metadata;
@@ -52,10 +49,7 @@ public sealed class FleetLinkOptions
 
     public string NodeVersion { get; init; } = "0.0.0";
 
-    /// <summary>服务端至少要是哪个版本（每次握手重新取）；空表示不要求。不够就关掉这条连接，等 <see cref="VersionRetry"/> 再握手，服务端可能正在更新。</summary>
-    public Func<string?>? MinServerVersion { get; init; }
-
-    /// <summary>框架之外的能力，例如 <c>moshiro.v1</c> 与 <c>moshiro.kinds:quic_ping,tcp_ping</c>。</summary>
+    /// <summary>框架之外的能力，例如 <c>moshiro.v1</c> 与 <c>moshiro.kinds:udp_echo,tcp_echo</c>。</summary>
     public IReadOnlyList<string> Capabilities { get; init; } = [];
 
     /// <summary>附加到 <c>fleet.hello</c> 的产品元数据。</summary>
@@ -64,12 +58,6 @@ public sealed class FleetLinkOptions
     /// <summary>附加到 <c>fleet.heartbeat</c> 的进程内计数（不能包含主机信息）。</summary>
     public Func<JsonElement?>? HeartbeatExtra { get; init; }
 
-    /// <summary>Optional reason for an orderly client shutdown, so the server can distinguish updates from service stops.</summary>
-    public Func<string>? ShutdownReason { get; init; }
-
-    /// <summary>连上服务器的整个过程（TCP、TLS、升级）最多等多久；卡住就放弃，由外层退避重试。</summary>
-    public TimeSpan ConnectTimeout { get; init; } = TimeSpan.FromSeconds(20);
-
     public TimeSpan HeartbeatInterval { get; init; } = TimeSpan.FromSeconds(10);
 
     public int MaxFrameBytes { get; init; } = 16 << 20;
@@ -84,6 +72,9 @@ public sealed class FleetLinkOptions
     /// 服务器必须在挑战里证明自己，之后这条链路用应用层加密；否则按旧的方式（没有固定信任根的节点）。
     /// </summary>
     public IServerVerifier? ServerVerifier { get; init; }
+
+    /// <summary>握手时自述正在运行的二进制（哈希和 CI 签名的清单）；没有就不上报（旧行为）。</summary>
+    public Func<BinaryAttestation?>? Attestation { get; init; }
 }
 
 /// <summary>
@@ -103,13 +94,6 @@ public sealed class FleetLink
     private volatile RpcSession<LinkContext>? _session;
     private volatile bool _stopped;
 
-    /// <summary>版本不够时重新握手的间隔：固定，不退避（对方更新完马上就能连上）。</summary>
-    public static readonly TimeSpan VersionRetry = TimeSpan.FromSeconds(5);
-
-    private static readonly TimeSpan s_tcpTimeout = TimeSpan.FromSeconds(4);
-    private TimeSpan? _fixedRetry;
-    private int _failedConnects;
-
     public FleetLink(FleetLinkOptions options, RpcRouter<LinkContext> router, IEnumerable<IStateHandler> handlers, Outbox outbox, TimeProvider time, INodeLog log)
     {
         _options = options;
@@ -156,7 +140,7 @@ public sealed class FleetLink
             {
                 return;
             }
-            catch (Exception ex)
+            catch (Exception ex) when (ex is WebSocketException or IOException or HttpRequestException or InvalidOperationException or TimeoutException)
             {
                 reason = ex.Message;
             }
@@ -173,68 +157,12 @@ public sealed class FleetLink
                 return;
             }
 
-            TimeSpan delay = _fixedRetry ?? _options.Backoff.Next();
-            _fixedRetry = null;
+            TimeSpan delay = _options.Backoff.Next();
             _log.Warn("link down, retrying", ("err", reason), ("retry_ms", (long)delay.TotalMilliseconds));
             await Task.Delay(delay, _time, ct);
         }
     }
 
-    /// <summary>
-    /// 连上面板：整个握手（TCP、TLS、升级）限时，卡住就放弃重试；每个地址的 TCP 连接也单独限时。
-    /// 失败之后下一次换着先试另一个地址族（IPv6 / IPv4）：有的网络上到某个地址族的 TCP 能连上、之后的 TLS 却被卡住，换一个往往就通。
-    /// </summary>
-    private async Task ConnectAsync(ClientWebSocket socket, Uri uri, CancellationToken ct)
-    {
-        int attempt = Volatile.Read(ref _failedConnects);
-        using var handler = new SocketsHttpHandler { ConnectCallback = (context, token) => ConnectTcpAsync(context, attempt, token) };
-        using var invoker = new HttpMessageInvoker(handler, disposeHandler: false);
-        using CancellationTokenSource timeout = CancellationTokenSource.CreateLinkedTokenSource(ct);
-        timeout.CancelAfter(_options.ConnectTimeout);
-        try
-        {
-            await socket.ConnectAsync(uri, invoker, timeout.Token);
-            Volatile.Write(ref _failedConnects, 0);
-        }
-        catch (OperationCanceledException) when (!ct.IsCancellationRequested)
-        {
-            Interlocked.Increment(ref _failedConnects);
-            throw new TimeoutException("connecting to the server timed out");
-        }
-        catch
-        {
-            Interlocked.Increment(ref _failedConnects);
-            throw;
-        }
-    }
-
-    private static async ValueTask<Stream> ConnectTcpAsync(SocketsHttpConnectionContext context, int attempt, CancellationToken ct)
-    {
-        IPAddress[] addresses = await Dns.GetHostAddressesAsync(context.DnsEndPoint.Host, ct);
-        // 第一次按 DNS 的顺序；之后奇数次先试 IPv6，偶数次先试 IPv4
-        IEnumerable<IPAddress> ordered = attempt == 0 ? addresses
-            : addresses.OrderBy(a => (a.AddressFamily == AddressFamily.InterNetworkV6) == (attempt % 2 == 1) ? 0 : 1);
-        Exception? last = null;
-        foreach (IPAddress address in ordered)
-        {
-            var tcp = new Socket(address.AddressFamily, SocketType.Stream, ProtocolType.Tcp) { NoDelay = true };
-            using CancellationTokenSource limit = CancellationTokenSource.CreateLinkedTokenSource(ct);
-            limit.CancelAfter(s_tcpTimeout);
-            try
-            {
-                await tcp.ConnectAsync(new IPEndPoint(address, context.DnsEndPoint.Port), limit.Token);
-                return new NetworkStream(tcp, ownsSocket: true);
-            }
-            catch (Exception ex) when (ex is SocketException or OperationCanceledException && !ct.IsCancellationRequested)
-            {
-                tcp.Dispose();
-                last = ex;
-            }
-        }
-
-        throw last ?? new SocketException((int)SocketError.HostNotFound);
-    }
-
     /// <summary>当前这条链路是不是应用层加密的（服务器证明了身份）。</summary>
     public bool IsSecure => _secure && IsConnected;
 
@@ -244,7 +172,7 @@ public sealed class FleetLink
         socket.Options.KeepAliveInterval = TimeSpan.FromSeconds(15);
         socket.Options.KeepAliveTimeout = TimeSpan.FromSeconds(20); // 对端 20 秒没有回 Pong 就判定连接已死
         var builder = new UriBuilder(_options.Server) { Scheme = _options.Server.Scheme == "https" ? "wss" : "ws", Path = RpcPath };
-        await ConnectAsync(socket, builder.Uri, ct);
+        await socket.ConnectAsync(builder.Uri, ct);
         await using var transport = new SecureTransport(new WsTransport(socket, _options.MaxFrameBytes));
 
         string? failure = await AuthenticateAsync(transport, ct);
@@ -256,8 +184,7 @@ public sealed class FleetLink
         _secure = transport.Secure;
 
         var session = new RpcSession<LinkContext>(transport, _router, new LinkContext(this), _time);
-        // Keep the active WebSocket receive alive until the planned shutdown marker and Close frame have been sent.
-        using var sessionCts = new CancellationTokenSource();
+        using CancellationTokenSource sessionCts = CancellationTokenSource.CreateLinkedTokenSource(ct);
         Task run = session.RunAsync(sessionCts.Token);
         try
         {
@@ -268,22 +195,10 @@ public sealed class FleetLink
                 NowUtcMs = _time.GetUtcNow().ToUnixTimeMilliseconds(),
                 Meta = _options.HelloMeta?.Invoke(),
             };
-            Result<HelloResult> helloResult = await session.RequestAsync(FleetMethods.Hello, hello, FleetRpcContext.Default.HelloParams, FleetRpcContext.Default.HelloResult, ct);
+            Result<HelloResult> helloResult = await session.RequestAsync(FleetMethods.Hello, hello, FleetRpcContext.Default.HelloParams, FleetRpcContext.Default.HelloResult, sessionCts.Token);
             if (!helloResult.IsSuccess)
             {
-                if (helloResult.Error.Code == FleetErrors.VersionRequired)
-                {
-                    _fixedRetry = VersionRetry; // 我们的版本低于服务端要求的：等自己更新，每 5 秒再试一次
-                }
-
-                return "hello failed: " + helloResult.Error.Code + (_fixedRetry is null ? "" : ": " + helloResult.Error.Message);
-            }
-
-            string? requiredServer = _options.MinServerVersion?.Invoke();
-            if (NodeVersions.IsBelow(helloResult.Value.ServerVersion, requiredServer))
-            {
-                _fixedRetry = VersionRetry; // 服务端版本低于我们要求的（它可能正在更新）：不发任何东西，5 秒后重新握手
-                return $"server version {helloResult.Value.ServerVersion} is below the required {requiredServer}";
+                return "hello failed: " + helloResult.Error.Code;
             }
 
             Outbox.ResetSent();
@@ -296,39 +211,31 @@ public sealed class FleetLink
                 await OnConnected(session, sessionCts.Token);
             }
 
-            Task cancellation = Task.Delay(Timeout.Infinite, ct);
-            Task finished = await Task.WhenAny(run, pump, heartbeat, cancellation);
-            string finishedReason = DescribeFinishedTask(finished, run, pump, heartbeat, session.TransportFailure);
-            _session = null; // 链路已经结束：上报方不能再往这条会话上写，收尾卡住时也一样
-            if (ct.IsCancellationRequested)
-            {
-                try
-                {
-                    using var closeTimeout = new CancellationTokenSource(TimeSpan.FromSeconds(1), _time);
-                    string reason = _options.ShutdownReason?.Invoke() ?? "stopping";
-                    await session.NotifyAsync(FleetMethods.Stopping, new StoppingParams { Reason = reason }, FleetRpcContext.Default.StoppingParams, closeTimeout.Token);
-                    await session.CloseAsync(reason, closeTimeout.Token);
-                }
-                catch (Exception ex) when (ex is OperationCanceledException or WebSocketException or IOException or ObjectDisposedException or InvalidOperationException)
-                {
-                    // The socket may already be gone; shutdown must still complete.
-                }
-            }
-
+            Task finished = await Task.WhenAny(run, pump, heartbeat);
             await sessionCts.CancelAsync();
-            await Bounded.WaitAsync(Task.WhenAll(Quiet(run), Quiet(pump), Quiet(heartbeat)), Bounded.Teardown, _time);
-            return finishedReason;
+            await Task.WhenAll(Quiet(run), Quiet(pump), Quiet(heartbeat));
+            return finished == run ? "closed by server" : "link task failed";
         }
         finally
         {
             _session = null;
             await sessionCts.CancelAsync();
-            await Bounded.WaitAsync(Quiet(run), Bounded.Teardown, _time);
-            await Bounded.WaitAsync(session.DisposeAsync().AsTask(), Bounded.Teardown, _time);
+            await Quiet(run);
+            await session.DisposeAsync();
         }
     }
 
     /// <summary>挑战 → 签名 → 认证结果。失败返回原因；服务端禁用该节点时不再重试。</summary>
+    private void Attest(AuthParams auth)
+    {
+        if (_options.Attestation?.Invoke() is { } attestation)
+        {
+            auth.Binary = attestation.Sha256;
+            auth.Release = attestation.Manifest;
+            auth.ReleaseSignature = attestation.Signature;
+        }
+    }
+
     private async Task<string?> AuthenticateAsync(SecureTransport transport, CancellationToken ct)
     {
         using var timeout = new CancellationTokenSource(TimeSpan.FromSeconds(15), _time);
@@ -363,6 +270,7 @@ public sealed class FleetLink
                     Signature = Convert.ToBase64String(_options.Identity.Sign(SignaturePurpose.AuthSecure, _options.NodeId.Value, nonce.Nonce, serverEphemeralText, nodeEphemeralText)),
                     NodeEphemeral = nodeEphemeralText,
                 };
+                Attest(authParams);
                 RpcMessage secureRequest = RpcMessage.Request(1, FleetMethods.Auth, authParams, FleetRpcContext.Default.AuthParams);
                 await transport.SendAsync(new RpcFrame(FrameKind.Text, JsonSerializer.SerializeToUtf8Bytes(secureRequest, FleetRpcContext.Default.RpcMessage)), linked.Token);
                 transport.Enable(LinkCipher.ForNode(ephemeral, serverEphemeral, nonce.Nonce));
@@ -370,6 +278,7 @@ public sealed class FleetLink
             else
             {
                 authParams = new AuthParams { NodeId = _options.NodeId.Value, Signature = Convert.ToBase64String(_options.Identity.Sign(SignaturePurpose.Auth, _options.NodeId.Value, nonce.Nonce)) };
+                Attest(authParams);
                 RpcMessage request = RpcMessage.Request(1, FleetMethods.Auth, authParams, FleetRpcContext.Default.AuthParams);
                 await transport.SendAsync(new RpcFrame(FrameKind.Text, JsonSerializer.SerializeToUtf8Bytes(request, FleetRpcContext.Default.RpcMessage)), linked.Token);
             }
@@ -484,29 +393,9 @@ public sealed class FleetLink
         {
             await task;
         }
-        catch (OperationCanceledException)
-        {
-            // Session cancellation is expected during reconnect and shutdown.
-        }
-        catch (Exception ex) when (ex is WebSocketException or IOException or ObjectDisposedException or TimeoutException or InvalidOperationException)
+        catch (Exception ex) when (ex is OperationCanceledException or WebSocketException or IOException or ObjectDisposedException)
         {
             // 会话结束时的正常异常
         }
     }
-
-    private static string DescribeFinishedTask(Task finished, Task run, Task pump, Task heartbeat, string? transportFailure)
-    {
-        if (finished.IsFaulted)
-        {
-            string taskName = finished == run ? "receive" : finished == pump ? "outbox send" : "heartbeat";
-            return $"{taskName} failed: {finished.Exception?.GetBaseException().Message ?? "unknown error"}";
-        }
-
-        if (finished == run && transportFailure is not null)
-        {
-            return "transport send failed: " + transportFailure;
-        }
-
-        return finished == run ? "closed by server" : "link task stopped";
-    }
 }

backend/fleet/src/Fleet.Link/Peer.cs

@@ -22,23 +22,14 @@ public sealed class PeerListenerOptions
     /// <summary>按对端 ID 查找允许连接的公钥（SPKI）。不在列表里的对端一律拒绝（期望状态里的 peers）。</summary>
     public required Func<string, byte[]?> LookupPeerKey { get; init; }
 
-    /// <summary>预握手敲门：每个连接先过这一关（名册里的 拨号方 才算得出敲门包），不通过就直接断开，什么都不写。</summary>
+    /// <summary>预握手敲门：每个连接先过这一关（名册里的 Probe 才算得出敲门包），不通过就直接断开，什么都不写。</summary>
     public required KnockGate Knock { get; init; }
 
     public int MaxFrameBytes { get; init; } = 8 << 20;
 
-    /// <summary>
-    /// 同时在线的会话数上限，只数已经通过敲门的对端。每个拨号方每个地址族一个会话，所以要容得下「名册里的节点数 × 2」。
-    /// 之前这个上限连还没敲门的连接一起数：节点多起来之后，已建立的会话把名额占满，后到的拨号方（哪怕在名册里）一连上就被断开，
-    /// 而且一直连不上，重试只会得到同样的结果。
-    /// </summary>
-    public int MaxSessions { get; init; } = 64;
+    public int MaxSessions { get; init; } = 8;
 
-    /// <summary>还没通过敲门的连接数上限（陌生人、扫描器、还在握手的）：敲门有读超时，占着名额的时间有限，但它们不能挤占已通过敲门的会话。</summary>
-    public int MaxPending { get; init; } = 64;
-
-    /// <summary>敲门、TLS、挑战应答整个握手的上限。线路差的时候（RTT 好几秒、还丢包）握手本来就要很久，10 秒做不完就会一直连不上；敲门已经挡住了不认识的连接，放宽不会放大被占连接的风险。</summary>
-    public TimeSpan HandshakeTimeout { get; init; } = PeerTimings.Handshake;
+    public TimeSpan HandshakeTimeout { get; init; } = TimeSpan.FromSeconds(10);
 }
 
 /// <summary>
@@ -52,7 +43,6 @@ public sealed class PeerListener
     private readonly INodeLog _log;
     private readonly TimeProvider _time;
     private int _sessions;
-    private int _pending;
 
     public PeerListener(PeerListenerOptions options, Func<PeerSessionInfo, IRpcTransport, CancellationToken, Task> onSession, INodeLog log, TimeProvider time)
     {
@@ -83,9 +73,9 @@ public sealed class PeerListener
             while (!ct.IsCancellationRequested)
             {
                 TcpClient client = await listener.AcceptTcpClientAsync(ct);
-                if (Interlocked.Increment(ref _pending) > _options.MaxPending)
+                if (Interlocked.Increment(ref _sessions) > _options.MaxSessions)
                 {
-                    Interlocked.Decrement(ref _pending);
+                    Interlocked.Decrement(ref _sessions);
                     client.Dispose();
                     continue;
                 }
@@ -105,11 +95,6 @@ public sealed class PeerListener
 
     private async Task HandleAsync(TcpClient client, CancellationToken ct)
     {
-        bool admitted = false;
-        bool counted = false;
-        string phase = "knock";
-        long started = _time.GetTimestamp();
-        EndPoint? remote = client.Client.RemoteEndPoint;
         try
         {
             client.NoDelay = true;
@@ -122,36 +107,19 @@ public sealed class PeerListener
                 return;
             }
 
-            admitted = true;
-            Interlocked.Decrement(ref _pending);
-            if (Interlocked.Increment(ref _sessions) > _options.MaxSessions)
-            {
-                Interlocked.Decrement(ref _sessions);
-                _log.Warn("peer session limit reached", ("remote", remote), ("limit", _options.MaxSessions));
-                return;
-            }
-
-            counted = true;
-            phase = "tls";
-            _log.Info("peer knock accepted", ("remote", remote), ("phase", phase));
             var tls = new SslStream(stream, leaveInnerStreamOpen: false);
             await tls.AuthenticateAsServerAsync(
                 new SslServerAuthenticationOptions { ServerCertificate = _options.Certificate, ClientCertificateRequired = false, EnabledSslProtocols = System.Security.Authentication.SslProtocols.Tls12 | System.Security.Authentication.SslProtocols.Tls13 },
                 linked.Token);
-            phase = "peer_auth";
-            _log.Info("peer TLS established", ("remote", remote), ("phase", phase), ("tls", tls.SslProtocol));
             var transport = new TlsFrameTransport(tls, _options.MaxFrameBytes);
             PeerSessionInfo? info = await HandshakeAsync(transport, client.Client.RemoteEndPoint, linked.Token);
             if (info is null)
             {
-                _log.Warn("peer authentication rejected", ("remote", remote), ("phase", phase));
                 await transport.DisposeAsync();
                 return;
             }
 
-            phase = "control_session";
-            _log.Info("peer session", ("dialer", info.PeerId), ("result", "established"), ("remote", remote),
-                ("elapsed_ms", (long)_time.GetElapsedTime(started).TotalMilliseconds));
+            _log.Info("peer session", ("probe", info.PeerId), ("result", "established"));
             await using (transport)
             {
                 await _onSession(info, transport, ct);
@@ -159,29 +127,14 @@ public sealed class PeerListener
         }
         catch (Exception ex) when (ex is IOException or System.Security.Authentication.AuthenticationException or OperationCanceledException or SocketException or ObjectDisposedException)
         {
-            // Invalid knocks stay silent. Only peers that passed admission get diagnostics.
-            if (admitted && !ct.IsCancellationRequested)
-            {
-                _log.Warn("peer control connection failed", ("remote", remote), ("phase", phase),
-                    ("elapsed_ms", (long)_time.GetElapsedTime(started).TotalMilliseconds),
-                    ("exception", ex.GetType().Name), ("err", ex.Message),
-                    ("socket_error", ex is SocketException socket ? socket.SocketErrorCode : null));
-            }
+            // 握手失败或连接断开：这个会话结束，监听继续
         }
         finally
         {
             client.Dispose();
-            if (!admitted)
-            {
-                Interlocked.Decrement(ref _pending);
-            }
-
-            if (counted)
-            {
             Interlocked.Decrement(ref _sessions);
         }
     }
-    }
 
     private async Task<PeerSessionInfo?> HandshakeAsync(TlsFrameTransport transport, EndPoint? remote, CancellationToken ct)
     {
@@ -255,9 +208,8 @@ public sealed class PeerListener
     }
 }
 
-/// <summary>发起点对点会话（拨号方 一侧）：按 SPKI 指纹固定对端证书，用身份私钥签名，派生同样的数据面凭据。</summary>
-/// <summary>点对点会话的套接字：开启 TCP keepalive（空闲 30 秒后每 10 秒探测，连续 6 次无响应才断开，约 90 秒），主机真的没了的时候连接不会一直挂着；
-/// 但线路短暂中断（几十秒的黑洞）不能把会话杀掉——线路中断本来就是要测的东西，会话得扛过去，恢复之后接着测。</summary>
+/// <summary>发起点对点会话（Probe 一侧）：按 SPKI 指纹固定对端证书，用身份私钥签名，派生同样的数据面凭据。</summary>
+/// <summary>点对点会话的套接字：开启 TCP keepalive（空闲 15 秒后每 5 秒探测，连续 3 次无响应就断开，约 30 秒），主机丢失或路径被切断时连接不会一直挂着。</summary>
 internal static class PeerSockets
 {
     public static void EnableKeepAlive(Socket socket)
@@ -265,9 +217,9 @@ internal static class PeerSockets
         try
         {
             socket.SetSocketOption(SocketOptionLevel.Socket, SocketOptionName.KeepAlive, true);
-            socket.SetSocketOption(SocketOptionLevel.Tcp, SocketOptionName.TcpKeepAliveTime, 30);
-            socket.SetSocketOption(SocketOptionLevel.Tcp, SocketOptionName.TcpKeepAliveInterval, 10);
-            socket.SetSocketOption(SocketOptionLevel.Tcp, SocketOptionName.TcpKeepAliveRetryCount, 6);
+            socket.SetSocketOption(SocketOptionLevel.Tcp, SocketOptionName.TcpKeepAliveTime, 15);
+            socket.SetSocketOption(SocketOptionLevel.Tcp, SocketOptionName.TcpKeepAliveInterval, 5);
+            socket.SetSocketOption(SocketOptionLevel.Tcp, SocketOptionName.TcpKeepAliveRetryCount, 3);
         }
         catch (Exception ex) when (ex is SocketException or ArgumentException or PlatformNotSupportedException)
         {
@@ -276,51 +228,27 @@ internal static class PeerSockets
     }
 }
 
-/// <summary>点对点会话的时间常数，拨号方和监听方两侧共用。</summary>
-public static class PeerTimings
-{
-    /// <summary>建立会话（连接、敲门、TLS、挑战应答）的总时限。</summary>
-    public static readonly TimeSpan Handshake = TimeSpan.FromSeconds(30);
-}
-
 public static class PeerConnector
 {
-    public static Task<(PeerSessionInfo Info, TlsFrameTransport Transport)> ConnectAsync(
-        string host, int port, byte[] agentPublicKey, NodeIdentity identity, string peerId, string? version, int maxFrameBytes, TimeProvider time, CancellationToken ct) =>
-        ConnectAsync(host, port, agentPublicKey, identity, peerId, version, maxFrameBytes, time, NullNodeLog.Instance, ct);
-
     public static async Task<(PeerSessionInfo Info, TlsFrameTransport Transport)> ConnectAsync(
-        string host, int port, byte[] agentPublicKey, NodeIdentity identity, string peerId, string? version, int maxFrameBytes, TimeProvider time, INodeLog log, CancellationToken ct,
-        Action<string>? onPhase = null)
+        string host, int port, byte[] agentPublicKey, NodeIdentity identity, string peerId, string? version, int maxFrameBytes, TimeProvider time, CancellationToken ct)
     {
         // 证书按公钥的指纹固定；敲门密钥由双方身份密钥派生
         byte[] expectedSpkiSha256 = NodeIdentity.Fingerprint(agentPublicKey);
         TcpClient client = IPAddress.TryParse(host, out IPAddress? address) ? new TcpClient(address.AddressFamily) : new TcpClient();
         client.NoDelay = true;
         PeerSockets.EnableKeepAlive(client.Client);
-        string phase = "tcp_connect";
-        onPhase?.Invoke(phase);
-        long started = time.GetTimestamp();
         try
         {
-            using var timeout = new CancellationTokenSource(PeerTimings.Handshake, time);
+            using var timeout = new CancellationTokenSource(TimeSpan.FromSeconds(10), time);
             using CancellationTokenSource linked = CancellationTokenSource.CreateLinkedTokenSource(ct, timeout.Token);
             await client.ConnectAsync(host, port, linked.Token);
-            log.Info("peer TCP connected", ("remote", client.Client.RemoteEndPoint), ("local", client.Client.LocalEndPoint),
-                ("phase", phase), ("elapsed_ms", (long)time.GetElapsedTime(started).TotalMilliseconds));
-            phase = "knock_send";
-            onPhase?.Invoke(phase);
             Stream stream = client.GetStream();
             await stream.WriteAsync(Knock.Build(Knock.DeriveKey(identity, agentPublicKey), time.GetUtcNow().ToUnixTimeMilliseconds()), linked.Token);
-            phase = "tls";
-            onPhase?.Invoke(phase);
             var tls = new SslStream(
                 stream, leaveInnerStreamOpen: false,
                 (_, certificate, _, _) => certificate is not null && SpkiMatches(certificate, expectedSpkiSha256));
             await tls.AuthenticateAsClientAsync(new SslClientAuthenticationOptions { TargetHost = host }, linked.Token);
-            log.Info("peer TLS established", ("phase", phase), ("tls", tls.SslProtocol), ("elapsed_ms", (long)time.GetElapsedTime(started).TotalMilliseconds));
-            phase = "peer_auth";
-            onPhase?.Invoke(phase);
             var transport = new TlsFrameTransport(tls, maxFrameBytes);
 
             RpcFrame? first = await transport.ReceiveAsync(linked.Token);
@@ -351,19 +279,10 @@ public static class PeerConnector
                 throw new InvalidOperationException("session id mismatch: the key agreement did not produce the same credentials");
             }
 
-            log.Info("peer authentication established", ("phase", phase), ("elapsed_ms", (long)time.GetElapsedTime(started).TotalMilliseconds));
             return (new PeerSessionInfo(peerId, credentials, client.Client.RemoteEndPoint), transport);
         }
-        catch (Exception ex)
+        catch
         {
-            if (!ct.IsCancellationRequested)
-            {
-                bool timedOut = ex is OperationCanceledException;
-                log.Warn("peer control handshake failed", ("phase", phase), ("remote", host), ("port", port),
-                    ("elapsed_ms", (long)time.GetElapsedTime(started).TotalMilliseconds), ("timeout_ms", PeerTimings.Handshake.TotalMilliseconds),
-                    ("exception", timedOut ? "Timeout" : ex.GetType().Name), ("err", timedOut ? $"{phase} did not finish within {PeerTimings.Handshake.TotalSeconds:0}s" : ex.Message),
-                    ("socket_error", ex is SocketException socket ? socket.SocketErrorCode : null));
-            }
             client.Dispose();
             throw;
         }

backend/fleet/src/Fleet.Link/Transports.cs

@@ -7,7 +7,6 @@ namespace Fleet.Link;
 /// <summary>WebSocket 传输（节点 → 服务端）：文本帧是 JSON-RPC，二进制帧是发件箱。</summary>
 public sealed class WsTransport : IRpcTransport
 {
-    private static readonly TimeSpan s_sendTimeout = TimeSpan.FromSeconds(20);
     private readonly WebSocket _socket;
     private readonly int _maxFrameBytes;
     private readonly SemaphoreSlim _sendLock = new(1, 1);
@@ -20,34 +19,16 @@ public sealed class WsTransport : IRpcTransport
 
     public async ValueTask SendAsync(RpcFrame frame, CancellationToken ct)
     {
-        using var timeout = new CancellationTokenSource(s_sendTimeout);
-        using CancellationTokenSource linked = CancellationTokenSource.CreateLinkedTokenSource(ct, timeout.Token);
-        bool lockTaken = false;
+        await _sendLock.WaitAsync(ct);
         try
         {
-            await _sendLock.WaitAsync(linked.Token);
-            lockTaken = true;
-            if (_socket.State != WebSocketState.Open)
-            {
-                throw new WebSocketException($"WebSocket is not open (state: {_socket.State})");
-            }
-
-            await _socket.SendAsync(frame.Data, frame.Kind == FrameKind.Text ? WebSocketMessageType.Text : WebSocketMessageType.Binary, endOfMessage: true, linked.Token);
-        }
-        catch (OperationCanceledException) when (timeout.IsCancellationRequested && !ct.IsCancellationRequested)
-        {
-            // A wedged send otherwise holds the lock and prevents heartbeat/reconnect from making progress.
-            _socket.Abort();
-            throw new TimeoutException($"WebSocket send timed out after {s_sendTimeout.TotalSeconds:0} seconds");
+            await _socket.SendAsync(frame.Data, frame.Kind == FrameKind.Text ? WebSocketMessageType.Text : WebSocketMessageType.Binary, endOfMessage: true, ct);
         }
         finally
-        {
-            if (lockTaken)
         {
             _sendLock.Release();
         }
     }
-    }
 
     public async ValueTask<RpcFrame?> ReceiveAsync(CancellationToken ct)
     {
@@ -112,13 +93,12 @@ public sealed class WsTransport : IRpcTransport
 }
 
 /// <summary>
-/// TLS 帧传输（拨号方 → 监听方）：<c>u32 LE 正文长度 | u8 类型(0 文本, 1 二进制) | 正文</c>。
+/// TLS 帧传输（Probe → Agent）：<c>u32 LE 正文长度 | u8 类型(0 文本, 1 二进制) | 正文</c>。
 /// 比 WebSocket 轻得多，Agent 不需要内置 HTTP 服务端。
 /// </summary>
 public sealed class TlsFrameTransport : IRpcTransport
 {
     private const int HeaderBytes = 5;
-    private static readonly TimeSpan s_sendTimeout = TimeSpan.FromSeconds(20);
 
     private readonly Stream _stream;
     private readonly int _maxFrameBytes;
@@ -140,31 +120,18 @@ public sealed class TlsFrameTransport : IRpcTransport
         byte[] header = new byte[HeaderBytes];
         BinaryPrimitives.WriteUInt32LittleEndian(header, (uint)frame.Data.Length);
         header[4] = (byte)frame.Kind;
-        // 对端黑洞时写会一直挂着并占着发送锁，心跳、时钟同步和重连都跟着卡住：超时就掐掉这条连接
-        using var timeout = new CancellationTokenSource(s_sendTimeout);
-        using CancellationTokenSource linked = CancellationTokenSource.CreateLinkedTokenSource(ct, timeout.Token);
-        bool lockTaken = false;
+        await _sendLock.WaitAsync(ct);
         try
         {
-            await _sendLock.WaitAsync(linked.Token);
-            lockTaken = true;
-            await _stream.WriteAsync(header, linked.Token);
-            await _stream.WriteAsync(frame.Data, linked.Token);
-            await _stream.FlushAsync(linked.Token);
-        }
-        catch (OperationCanceledException) when (timeout.IsCancellationRequested && !ct.IsCancellationRequested)
-        {
-            await _stream.DisposeAsync();
-            throw new TimeoutException($"peer send timed out after {s_sendTimeout.TotalSeconds:0} seconds");
+            await _stream.WriteAsync(header, ct);
+            await _stream.WriteAsync(frame.Data, ct);
+            await _stream.FlushAsync(ct);
         }
         finally
-        {
-            if (lockTaken)
         {
             _sendLock.Release();
         }
     }
-    }
 
     public async ValueTask<RpcFrame?> ReceiveAsync(CancellationToken ct)
     {

backend/fleet/src/Fleet.Rpc/FleetMethods.cs

@@ -10,7 +10,6 @@ public static class FleetMethods
     public const string Auth = "fleet.auth";
     public const string Hello = "fleet.hello";
     public const string Heartbeat = "fleet.heartbeat";
-    public const string Stopping = "fleet.stopping";
     public const string StateSync = "fleet.state.sync";
     public const string OutboxAck = "fleet.outbox.ack";
     public const string Revoked = "fleet.revoked";
@@ -18,7 +17,7 @@ public static class FleetMethods
 
 /// <summary>
 /// 框架自己的方法声明，两个平面各一份。认证和握手阶段由连接代码直接处理的（<c>Inline</c>）不经过路由器。
-/// 同一个方法名在两个平面里可以有不同的发起方：<c>fleet.state.sync</c> 在节点面是服务端发给节点，在对等面是 监听方发给拨号方。
+/// 同一个方法名在两个平面里可以有不同的发起方：<c>fleet.state.sync</c> 在节点面是服务端发给节点，在对等面是 Agent 发给 Probe。
 /// </summary>
 public static class FleetMethodTable
 {
@@ -29,7 +28,6 @@ public static class FleetMethodTable
         MethodSpec.Request(FleetMethods.Auth, RpcSide.Node).AsInline(),
         MethodSpec.Request(FleetMethods.Hello, RpcSide.Node).AsInline(),
         MethodSpec.Notification(FleetMethods.Heartbeat, RpcSide.Node).AsInline(),
-        MethodSpec.Notification(FleetMethods.Stopping, RpcSide.Node).AsInline(),
         MethodSpec.Request(FleetMethods.StateSync, RpcSide.Server),
         MethodSpec.Notification(FleetMethods.OutboxAck, RpcSide.Server),
         MethodSpec.Notification(FleetMethods.Revoked, RpcSide.Server),
@@ -42,6 +40,12 @@ public static class FleetMethodTable
     ]);
 }
 
+/// <summary>二进制通道：0–15 保留给框架，产品从 16 开始。</summary>
+public static class FleetChannels
+{
+    public const byte FirstProductChannel = 16;
+}
+
 public sealed class ChallengeParams
 {
     [JsonPropertyName("nonce")]
@@ -71,6 +75,17 @@ public sealed class AuthParams
     /// <summary>节点的临时 ECDH 公钥（SPKI，base64）：有它就是安全链路，签名覆盖双方的临时公钥，认证结果起所有帧都是加密的。</summary>
     [JsonPropertyName("node_eph")]
     public string? NodeEphemeral { get; set; }
+
+    /// <summary>正在运行的二进制的 SHA-256（小写十六进制）：服务端在第一次握手时绑定，之后核对。旧节点没有这个字段。</summary>
+    [JsonPropertyName("binary")]
+    public string? Binary { get; set; }
+
+    /// <summary>安装这个二进制时宿主机更新器验过的 CI 签名清单（原文）和签名（base64）：服务端用 CI 公钥验过，才相信它列出的官方哈希。</summary>
+    [JsonPropertyName("release")]
+    public string? Release { get; set; }
+
+    [JsonPropertyName("release_sig")]
+    public string? ReleaseSignature { get; set; }
 }
 
 public sealed class AuthResult
@@ -91,7 +106,7 @@ public sealed class HelloParams
     [JsonPropertyName("now_utc_ms")]
     public long NowUtcMs { get; set; }
 
-    /// <summary>产品自定义的元数据（例如 MoShiro 的 tls_spki、listen 信息、local_limits 摘要）。</summary>
+    /// <summary>产品自定义的元数据（例如 MoShiro 的 tls_spki、probe_ports、local_limits 摘要）。</summary>
     [JsonPropertyName("meta")]
     public JsonElement? Meta { get; set; }
 }
@@ -104,10 +119,6 @@ public sealed class HelloResult
     /// <summary>节点时钟 − 服务端时钟（毫秒，未扣除网络往返）。</summary>
     [JsonPropertyName("clock_offset_ms")]
     public long ClockOffsetMs { get; set; }
-
-    /// <summary>服务端的版本：节点要求了最低服务端版本时，用它判断。</summary>
-    [JsonPropertyName("server_version")]
-    public string? ServerVersion { get; set; }
 }
 
 public sealed class HeartbeatParams
@@ -162,12 +173,6 @@ public sealed class RevokedParams
     public string Reason { get; set; } = "";
 }
 
-public sealed class StoppingParams
-{
-    [JsonPropertyName("reason")]
-    public string Reason { get; set; } = "stopping";
-}
-
 [JsonSourceGenerationOptions(DefaultIgnoreCondition = JsonIgnoreCondition.WhenWritingNull)]
 [JsonSerializable(typeof(RpcMessage))]
 [JsonSerializable(typeof(ChallengeParams))]
@@ -176,7 +181,6 @@ public sealed class StoppingParams
 [JsonSerializable(typeof(HelloParams))]
 [JsonSerializable(typeof(HelloResult))]
 [JsonSerializable(typeof(HeartbeatParams))]
-[JsonSerializable(typeof(StoppingParams))]
 [JsonSerializable(typeof(StateSyncParams))]
 [JsonSerializable(typeof(StateSyncResult))]
 [JsonSerializable(typeof(OutboxAckParams))]

backend/fleet/src/Fleet.Rpc/MethodTable.cs

@@ -1,6 +1,6 @@
 namespace Fleet.Rpc;
 
-/// <summary>连接的哪一侧。节点面：服务端和节点；对等面：拨号方和监听方。</summary>
+/// <summary>连接的哪一侧。节点面：服务端和节点；对等面：拨号方（Probe）和监听方（Agent）。</summary>
 public enum RpcSide
 {
     Server,

backend/fleet/src/Fleet.Rpc/PeerMethods.cs

@@ -3,7 +3,7 @@ using System.Text.Json.Serialization;
 namespace Fleet.Rpc;
 
 /// <summary>
-/// 节点间点对点会话的握手（拨号方 → 监听方，TLS 帧）。除了签名认证，双方各生成一个临时 ECDH 密钥，
+/// 节点间点对点会话的握手（Probe → Agent，TLS 帧）。除了签名认证，双方各生成一个临时 ECDH 密钥，
 /// 经过 HKDF 派生出数据面使用的 <c>session_id</c> 与 <c>session_key</c>（所以不需要单独的“下发密钥”方法）。
 /// </summary>
 public static class PeerMethods

backend/fleet/src/Fleet.Rpc/RpcRouter.cs

@@ -11,7 +11,7 @@ public readonly record struct RpcViolation(string Method, string Reason);
 public readonly record struct RpcDispatch(RpcMessage? Response, RpcViolation? Violation);
 
 /// <summary>
-/// 方法路由，与传输方式无关，也是协议方向的<b>唯一卡口</b>：节点和服务端、拨号方和监听方 的入站分发都经过这里。
+/// 方法路由，与传输方式无关，也是协议方向的<b>唯一卡口</b>：节点和服务端、Probe 和 Agent 的入站分发都经过这里。
 /// <typeparamref name="TContext"/> 由传输层提供（例如调用方节点）。路由器知道自己在连接的哪一侧（<see cref="Side"/>）和声明表，
 /// 注册处理函数时核对方向、类型和角色，分发时拒绝没有声明或不该由对方发起的方法，发送时核对自己有没有资格发起。
 /// </summary>

backend/fleet/src/Fleet.Rpc/RpcSession.cs

@@ -1,5 +1,4 @@
 using System.Collections.Concurrent;
-using System.Net.WebSockets;
 using System.Text.Json;
 using System.Text.Json.Serialization.Metadata;
 using System.Threading.Channels;
@@ -17,7 +16,7 @@ public readonly record struct RpcFrame(FrameKind Kind, ReadOnlyMemory<byte> Data
 
 /// <summary>
 /// 一条双向的、保序的消息通道：文本帧承载 JSON-RPC，二进制帧承载发件箱。
-/// 节点到服务端用 WebSocket，拨号方到监听方 用 TLS 帧；会话引擎对两者一视同仁。
+/// 节点到服务端用 WebSocket，Probe 到 Agent 用 TLS 帧；会话引擎对两者一视同仁。
 /// </summary>
 public interface IRpcTransport : IAsyncDisposable
 {
@@ -50,7 +49,6 @@ public sealed class RpcSession<TContext> : IAsyncDisposable
     private readonly IRpcTransport _transport;
     private readonly RpcRouter<TContext> _router;
     private readonly TContext _context;
-    private static readonly TimeSpan s_dispatcherGrace = TimeSpan.FromSeconds(5);
     private readonly TimeProvider _time;
     private readonly RpcSessionOptions _options;
     private readonly ConcurrentDictionary<long, TaskCompletionSource<RpcMessage>> _pending = new();
@@ -60,7 +58,6 @@ public sealed class RpcSession<TContext> : IAsyncDisposable
     private long _lastReceive;
     private int _violations;
     private bool _violated;
-    private string? _transportFailure;
 
     public RpcSession(IRpcTransport transport, RpcRouter<TContext> router, TContext context, TimeProvider time, RpcSessionOptions? options = null)
     {
@@ -79,9 +76,6 @@ public sealed class RpcSession<TContext> : IAsyncDisposable
     /// <summary>对端上一次发来任何帧到现在的时间。静默死亡（断电、进程被冻结）的对端不会关闭连接，调用方据此设置空闲超时。</summary>
     public TimeSpan IdleFor => _time.GetElapsedTime(Volatile.Read(ref _lastReceive));
 
-    /// <summary>The first transport send failure observed by this session, if any.</summary>
-    public string? TransportFailure => Volatile.Read(ref _transportFailure);
-
     /// <summary>读帧并分发，直到连接关闭或被取消。</summary>
     public async Task RunAsync(CancellationToken ct)
     {
@@ -136,8 +130,14 @@ public sealed class RpcSession<TContext> : IAsyncDisposable
             }
 
             await linked.CancelAsync();
-            // 某个入站处理器卡住了也不能拖住整个会话的结束（链路据此重连）：它随取消令牌自己收尾
-            await Bounded.WaitAsync(dispatcher, s_dispatcherGrace, _time);
+            try
+            {
+                await dispatcher;
+            }
+            catch (OperationCanceledException)
+            {
+                // 正常结束
+            }
         }
     }
 
@@ -197,11 +197,11 @@ public sealed class RpcSession<TContext> : IAsyncDisposable
             _router.CheckSend(message.Method, message.Id is null ? MethodKind.Notification : MethodKind.Request);
         }
 
-        return SendTransportAsync(new RpcFrame(FrameKind.Text, JsonSerializer.SerializeToUtf8Bytes(message, FleetRpcContext.Default.RpcMessage)), ct);
+        return _transport.SendAsync(new RpcFrame(FrameKind.Text, JsonSerializer.SerializeToUtf8Bytes(message, FleetRpcContext.Default.RpcMessage)), ct);
     }
 
     public ValueTask SendBinaryAsync(ReadOnlyMemory<byte> frame, CancellationToken ct) =>
-        SendTransportAsync(new RpcFrame(FrameKind.Binary, frame), ct);
+        _transport.SendAsync(new RpcFrame(FrameKind.Binary, frame), ct);
 
     public ValueTask CloseAsync(string reason, CancellationToken ct) => _transport.CloseAsync(reason, ct);
 
@@ -239,30 +239,6 @@ public sealed class RpcSession<TContext> : IAsyncDisposable
         }
     }
 
-    private async ValueTask SendTransportAsync(RpcFrame frame, CancellationToken ct)
-    {
-        try
-        {
-            await _transport.SendAsync(frame, ct);
-        }
-        catch (Exception ex) when (!ct.IsCancellationRequested && ex is WebSocketException or IOException or TimeoutException or ObjectDisposedException)
-        {
-            Interlocked.CompareExchange(ref _transportFailure, ex.Message, null);
-            try
-            {
-                // A failed send must also stop the receive loop. Otherwise callers that log and swallow this
-                // exception can leave the session marked connected forever on an already-aborted socket.
-                _stop.Cancel();
-            }
-            catch (ObjectDisposedException)
-            {
-                // Disposal raced with the failed send.
-            }
-
-            throw;
-        }
-    }
-
     private static RpcMessage? Parse(ReadOnlyMemory<byte> data)
     {
         try

backend/fleet/src/Fleet.Server.Abstractions/FleetServerOptions.cs

@@ -5,7 +5,7 @@ public sealed class FleetServerOptions
     /// <summary>节点 ID 的类型前缀，MoShiro 使用 <c>agt</c>。</summary>
     public string NodeIdPrefix { get; set; } = "node";
 
-    /// <summary>加入令牌的「种类」对应的节点 ID 前缀（例如 <c>edge</c> → <c>edg</c>）；没有列出的种类用 <see cref="NodeIdPrefix"/>。</summary>
+    /// <summary>加入令牌的「种类」对应的节点 ID 前缀（例如 <c>probe</c> → <c>prb</c>）；没有列出的种类用 <see cref="NodeIdPrefix"/>。</summary>
     public Dictionary<string, string> NodeIdPrefixes { get; } = new(StringComparer.Ordinal);
 
     public int MaxFrameBytes { get; set; } = 16 << 20;
@@ -16,18 +16,6 @@ public sealed class FleetServerOptions
 
     public TimeSpan RequestTimeout { get; set; } = TimeSpan.FromSeconds(15);
 
-    /// <summary>处理节点的一条消息（心跳、hello、收件箱）最多用这么久：钩子卡在数据库或外部查询里时，不能让这条连接停在「在线」而收不到任何心跳超时。超过就断开，节点按退避重连。</summary>
-    public TimeSpan HandlerTimeout { get; set; } = TimeSpan.FromSeconds(30);
-
     /// <summary>节点违反协议（方法没有声明、方向、类型或角色不对）的次数上限；超过就断开这条连接。</summary>
     public int MaxViolations { get; set; } = 20;
 }
-
-/// <summary>握手时的版本要求：服务端的版本（告诉节点），以及节点至少要是哪个版本。每次握手都重新取，所以改配置不用重启。</summary>
-public interface IFleetVersionPolicy
-{
-    string? ServerVersion { get; }
-
-    /// <summary>空表示不要求。</summary>
-    string? MinNodeVersion { get; }
-}

backend/fleet/src/Fleet.Server.Abstractions/ServerContracts.cs

@@ -6,17 +6,6 @@ namespace Fleet.Server;
 
 public sealed record EnrollmentContext(string? RemoteIp, JsonElement? ProductMeta);
 
-/// <summary>
-/// 节点的真实来源地址：前面有反向代理时，套接字上只有代理的地址，产品在这里决定信不信转发头。
-/// 没有注册就用套接字地址。
-/// </summary>
-public interface IClientAddressSource
-{
-    /// <param name="socket">套接字上看到的对端地址。</param>
-    /// <param name="header">按名字取请求头的所有值。</param>
-    string? Resolve(System.Net.IPAddress? socket, Func<string, IEnumerable<string>> header);
-}
-
 /// <summary>
 /// 服务器的身份密钥：操作员用 passkey 把它的公钥签进操作员集合。有它的时候，认证挑战会带上服务器的签名，节点据此确认对面是真的服务器，
 /// 并在这条链路上用应用层加密。没有注册就是旧的行为（没有服务器身份）。
@@ -30,6 +19,21 @@ public interface IServerIdentity
 }
 
 /// <summary>操作员集合的链（passkey 签的，创世在前）。没有初始化信任根时是空的。节点注册时用它确认「这个面板确实是我的操作员的」。</summary>
+/// <summary>节点握手时自述的二进制（旧节点没有）。</summary>
+public sealed record BinaryEvidence(string? Sha256, string? Manifest, string? Signature);
+
+/// <summary>是否让这个节点继续：不放行时 <c>Reason</c> 会原样告诉节点（这是它自己的信息，不是秘密）。</summary>
+public sealed record BinaryDecision(bool Accept, string Reason = "");
+
+/// <summary>
+/// 服务端对节点二进制的把关：节点第一次握手成功的那一刻绑定它的二进制哈希，之后哈希对不上、又不在 CI 签名的更新清单里，就不让它继续运行。
+/// 没有注册实现时不做任何检查。
+/// </summary>
+public interface IBinaryAttestor
+{
+    ValueTask<BinaryDecision> CheckAsync(NodeRecord node, BinaryEvidence evidence, CancellationToken ct);
+}
+
 public interface IOperatorTrustSource
 {
     ValueTask<IReadOnlyList<SignedStatementDto>> ChainAsync(CancellationToken ct);
@@ -49,8 +53,7 @@ public enum NodeEventKind
     Heartbeat,
 }
 
- /// <param name="RemoteIp">这条连接的来源地址（经过 <see cref="IClientAddressSource"/>）；没有就是 null。</param>
-public sealed record NodeEvent(NodeEventKind Kind, NodeRecord Node, long AtMs, HelloParams? Hello = null, JsonElement? Extra = null, string? Reason = null, string? RemoteIp = null);
+public sealed record NodeEvent(NodeEventKind Kind, NodeRecord Node, long AtMs, HelloParams? Hello = null, JsonElement? Extra = null);
 
 public interface INodeEventSink
 {

backend/fleet/src/Fleet.Server.Enrollment/FleetEnrollmentEndpoints.cs

@@ -73,7 +73,7 @@ public static class FleetEnrollmentEndpoints
             return Problem(StatusCodes.Status400BadRequest, FleetErrors.InvalidParams, "request body is not valid");
         }
 
-        Result<EnrollResponse> result = await service.EnrollAsync(request, ClientIp(http), http.RequestAborted);
+        Result<EnrollResponse> result = await service.EnrollAsync(request, http.Connection.RemoteIpAddress?.ToString(), http.RequestAborted);
         if (result.TryGet(out EnrollResponse? response, out FleetError? error))
         {
             return Results.Json(response, EnrollmentJsonContext.Default.EnrollResponse, statusCode: StatusCodes.Status201Created);
@@ -88,11 +88,6 @@ public static class FleetEnrollmentEndpoints
         return Problem(status, error.Code, error.Message);
     }
 
-    private static string? ClientIp(HttpContext http) =>
-        http.RequestServices.GetService<IClientAddressSource>() is { } source
-            ? source.Resolve(http.Connection.RemoteIpAddress, name => http.Request.Headers[name].Select(v => v ?? ""))
-            : http.Connection.RemoteIpAddress?.ToString();
-
     private static async Task<EnrollRequest?> ReadBodyAsync(HttpContext http, CancellationToken ct)
     {
         if (http.Request.ContentLength > MaxBodyBytes)

backend/fleet/src/Fleet.Server.Link/FleetLinkHost.cs

@@ -25,10 +25,10 @@ public sealed class FleetLinkHost : IFleetConnections, IStateNotifier
         IEnumerable<INodeEventSink> sinks,
         FleetRouter router,
         IServerIdentity? serverIdentity = null,
-        IFleetVersionPolicy? versionPolicy = null)
+        IBinaryAttestor? binaryAttestor = null)
     {
-        VersionPolicy = versionPolicy;
         ServerIdentity = serverIdentity;
+        BinaryAttestor = binaryAttestor;
         Store = store;
         Options = options;
         Time = time;
@@ -45,7 +45,8 @@ public sealed class FleetLinkHost : IFleetConnections, IStateNotifier
     /// <summary>服务器的身份密钥；没有时节点连接用旧的明文认证。</summary>
     internal IServerIdentity? ServerIdentity { get; }
 
-    internal IFleetVersionPolicy? VersionPolicy { get; }
+    /// <summary>节点二进制的把关；没有时不检查。</summary>
+    internal IBinaryAttestor? BinaryAttestor { get; }
 
     internal FleetServerOptions Options { get; }
 
@@ -71,9 +72,9 @@ public sealed class FleetLinkHost : IFleetConnections, IStateNotifier
         _connections.TryGetValue(node, out NodeConnection? c) ? c.Session : null;
 
     /// <summary>处理一条 WebSocket 连接，直到连接结束。</summary>
-    public async Task RunConnectionAsync(WebSocket socket, string? remoteIp, CancellationToken ct)
+    public async Task RunConnectionAsync(WebSocket socket, CancellationToken ct)
     {
-        await using var connection = new NodeConnection(this, socket, remoteIp);
+        using var connection = new NodeConnection(this, socket);
         await connection.RunAsync(ct);
     }
 

backend/fleet/src/Fleet.Server.Link/FleetLog.cs

@@ -10,9 +10,6 @@ internal static partial class FleetLog
     [LoggerMessage(Level = LogLevel.Information, Message = "node disconnected node={Node} reason={Reason}")]
     public static partial void Disconnected(ILogger logger, string node, string reason);
 
-    [LoggerMessage(Level = LogLevel.Warning, Message = "node below the required version node={Node} version={Version} required={Required}")]
-    public static partial void VersionTooOld(ILogger logger, string node, string version, string required);
-
     [LoggerMessage(Level = LogLevel.Warning, Message = "node auth rejected err={Err}")]
     public static partial void AuthRejected(ILogger logger, string err);
 

backend/fleet/src/Fleet.Server.Link/FleetServerBuilder.cs

@@ -29,10 +29,10 @@ public sealed class FleetServerBuilder
     internal Func<NodeRecord, string?>? RoleOf => _roleOf;
 
     /// <summary>
-    /// 方法声明表（框架的加产品的，见 <see cref="MethodTable.Combine"/>）和可选的节点角色解析（声明表里限定了角色的方法才需要）。方法表必须设置：
+    /// 方法声明表（框架的加产品的，见 <see cref="MethodTable.Combine"/>）和节点的角色解析（例如按节点 ID 前缀区分 Agent 和 Probe）。必须设置：
     /// 路由器据此核对每个处理函数的方向、类型和角色，并拒绝没有声明或不该由节点发起的方法。
     /// </summary>
-    public FleetServerBuilder WithMethods(MethodTable methods, Func<NodeRecord, string?>? roleOf = null)
+    public FleetServerBuilder WithMethods(MethodTable methods, Func<NodeRecord, string?> roleOf)
     {
         _methods = methods;
         _roleOf = roleOf;
@@ -111,7 +111,7 @@ public static class FleetServerExtensions
             provider.GetServices<INodeEventSink>(),
             provider.GetRequiredService<FleetRouter>(),
             provider.GetService<IServerIdentity>(),
-            provider.GetService<IFleetVersionPolicy>()));
+            provider.GetService<IBinaryAttestor>()));
         services.AddSingleton<IFleetConnections>(provider => provider.GetRequiredService<FleetLinkHost>());
         services.AddSingleton<IStateNotifier>(provider => provider.GetRequiredService<FleetLinkHost>());
         return builder;
@@ -129,10 +129,7 @@ public static class FleetServerExtensions
             }
 
             using System.Net.WebSockets.WebSocket socket = await http.WebSockets.AcceptWebSocketAsync();
-            string? remote = http.RequestServices.GetService<IClientAddressSource>() is { } source
-                ? source.Resolve(http.Connection.RemoteIpAddress, name => http.Request.Headers[name].Select(v => v ?? ""))
-                : http.Connection.RemoteIpAddress?.ToString();
-            await http.RequestServices.GetRequiredService<FleetLinkHost>().RunConnectionAsync(socket, remote, http.RequestAborted);
+            await http.RequestServices.GetRequiredService<FleetLinkHost>().RunConnectionAsync(socket, http.RequestAborted);
         });
         return app;
     }

backend/fleet/src/Fleet.Server.Link/NodeConnection.Auth.cs

@@ -88,6 +88,18 @@ internal sealed partial class NodeConnection
             return false;
         }
 
+        // 二进制：第一次握手成功的这一刻绑定，之后哈希对不上又不在 CI 签名的更新清单里，就不让它继续运行
+        if (_host.BinaryAttestor is { } attestor)
+        {
+            BinaryDecision decision = await attestor.CheckAsync(node, new BinaryEvidence(auth!.Binary, auth.Release, auth.ReleaseSignature), ct);
+            if (!decision.Accept)
+            {
+                await SendAsync(RpcMessage.Failure(message.Id, new FleetError(FleetErrors.BinaryMismatch, decision.Reason)), ct);
+                await RejectAsync(null, "binary mismatch: " + decision.Reason, ct);
+                return false;
+            }
+        }
+
         _session = new NodeSessionInfo(node.Id, _host.Time.GetUtcNow().ToUnixTimeMilliseconds(), node.NodeVersion, [], null);
         await SendAsync(
             RpcMessage.Success(message.Id.Value, new AuthResult { Status = StatusName(node.Status) }, FleetRpcContext.Default.AuthResult),

backend/fleet/src/Fleet.Server.Link/NodeConnection.Receive.cs

@@ -34,20 +34,12 @@ internal sealed partial class NodeConnection
             {
                 if (TryParse(frame.Data, out RpcMessage? message))
                 {
-                    if (!await Bounded.WaitAsync(HandleMessageAsync(message, ct), _host.Options.HandlerTimeout, _host.Time))
-                    {
-                        _closeReason ??= "handler timeout";
-                        return;
-                    }
+                    await HandleMessageAsync(message, ct);
                 }
             }
             else
             {
-                if (!await Bounded.WaitAsync(HandleBinaryAsync(frame.Data, ct), _host.Options.HandlerTimeout, _host.Time))
-                {
-                    _closeReason ??= "handler timeout";
-                    return;
-                }
+                await HandleBinaryAsync(frame.Data, ct);
             }
         }
     }
@@ -72,17 +64,6 @@ internal sealed partial class NodeConnection
             case FleetMethods.Heartbeat:
                 await HandleHeartbeatAsync(message, ct);
                 return;
-            case FleetMethods.Stopping:
-                try
-                {
-                    string? reason = message.Params?.Deserialize(FleetRpcContext.Default.StoppingParams)?.Reason;
-                    _closeReason = reason is "updating" ? "updating" : "stopping";
-                }
-                catch (JsonException)
-                {
-                    _closeReason = "stopping";
-                }
-                return;
         }
 
         if (_node!.Status != NodeStatus.Active)
@@ -151,20 +132,6 @@ internal sealed partial class NodeConnection
             return;
         }
 
-        string? minVersion = _host.VersionPolicy?.MinNodeVersion;
-        if (NodeVersions.IsBelow(hello.NodeVersion, minVersion))
-        {
-            // 版本不够：不登记在线，告诉节点要求的版本，关闭连接。节点等几秒（它可能正在更新）再握手。
-            FleetLog.VersionTooOld(_host.Logger, _node!.Id.Value, hello.NodeVersion, minVersion!);
-            if (message.Id is not null)
-            {
-                await SendAsync(RpcMessage.Failure(message.Id, new FleetError(FleetErrors.VersionRequired, $"node version {hello.NodeVersion} is below the required {minVersion}")), ct);
-            }
-
-            await CloseAsync("version required");
-            return;
-        }
-
         long nowMs = _host.Time.GetUtcNow().ToUnixTimeMilliseconds();
         _session = _session! with
         {
@@ -183,7 +150,7 @@ internal sealed partial class NodeConnection
             await SendAsync(
                 RpcMessage.Success(
                     message.Id.Value,
-                    new HelloResult { ServerTimeMs = nowMs, ClockOffsetMs = hello.NowUtcMs - nowMs, ServerVersion = _host.VersionPolicy?.ServerVersion },
+                    new HelloResult { ServerTimeMs = nowMs, ClockOffsetMs = hello.NowUtcMs - nowMs },
                     FleetRpcContext.Default.HelloResult),
                 ct);
         }

backend/fleet/src/Fleet.Server.Link/NodeConnection.Send.cs

@@ -26,12 +26,6 @@ internal sealed partial class NodeConnection
 
         try
         {
-            if (_disposed)
-            {
-                return;
-            }
-
-            _lifetime.CancelAfter(s_closeTimeout);
             if (_socket.State == WebSocketState.Open)
             {
                 await _socket.CloseOutputAsync(WebSocketCloseStatus.NormalClosure, reason, CancellationToken.None);
@@ -46,6 +40,7 @@ internal sealed partial class NodeConnection
             _sendLock.Release();
         }
 
+        _lifetime.CancelAfter(s_closeTimeout);
     }
 
     public async ValueTask RevokeAsync(string reason, CancellationToken ct)
@@ -75,7 +70,7 @@ internal sealed partial class NodeConnection
 
         try
         {
-            if (_disposed || _socket.State != WebSocketState.Open)
+            if (_socket.State != WebSocketState.Open)
             {
                 return false;
             }

backend/fleet/src/Fleet.Server.Link/NodeConnection.State.cs

@@ -102,7 +102,7 @@ internal sealed partial class NodeConnection
 
     private async Task FireAsync(NodeEventKind kind, CancellationToken ct, HelloParams? hello = null, JsonElement? extra = null)
     {
-        var nodeEvent = new NodeEvent(kind, _node!, _host.Time.GetUtcNow().ToUnixTimeMilliseconds(), hello, extra, kind == NodeEventKind.Offline ? _closeReason : null, _remoteIp);
+        var nodeEvent = new NodeEvent(kind, _node!, _host.Time.GetUtcNow().ToUnixTimeMilliseconds(), hello, extra);
         foreach (INodeEventSink sink in _host.Sinks)
         {
             try

backend/fleet/src/Fleet.Server.Link/NodeConnection.Wire.cs

@@ -17,7 +17,6 @@ internal sealed partial class NodeConnection
             ValueWebSocketReceiveResult result = await _socket.ReceiveAsync(buffer, ct);
             if (result.MessageType == WebSocketMessageType.Close)
             {
-                _closeReason ??= _socket.CloseStatusDescription;
                 return null;
             }
 

backend/fleet/src/Fleet.Server.Link/NodeConnection.cs

@@ -11,7 +11,7 @@ namespace Fleet.Server;
 /// 一个节点的 WebSocket 会话：挑战 → 签名认证 → hello → 状态推送 / 收件箱 / 产品 RPC。
 /// 收到的消息按顺序处理，所以产品的处理函数不能等待“向同一个节点发起的请求”，否则会死锁。
 /// </summary>
-internal sealed partial class NodeConnection : IAsyncDisposable
+internal sealed partial class NodeConnection : IDisposable
 {
     private static readonly TimeSpan s_authTimeout = TimeSpan.FromSeconds(15);
     private static readonly TimeSpan s_presenceRefresh = TimeSpan.FromSeconds(60);
@@ -31,17 +31,13 @@ internal sealed partial class NodeConnection : IAsyncDisposable
     private NodeRecord? _node;
     private NodeSessionInfo? _session;
     private string? _closeReason;
-    private bool _disposed;
     private bool _helloDone;
     private int _violations;
 
-    private readonly string? _remoteIp;
-
-    public NodeConnection(FleetLinkHost host, WebSocket socket, string? remoteIp = null)
+    public NodeConnection(FleetLinkHost host, WebSocket socket)
     {
         _host = host;
         _socket = socket;
-        _remoteIp = remoteIp;
     }
 
     public NodeId NodeId => _node!.Id;
@@ -138,19 +134,10 @@ internal sealed partial class NodeConnection : IAsyncDisposable
         }
     }
 
-    public async ValueTask DisposeAsync()
+    public void Dispose()
     {
-        await _sendLock.WaitAsync();
-        try
-        {
-            _disposed = true;
         _cipher?.Dispose();
         _lifetime.Dispose();
-        }
-        finally
-        {
-            // API callers may already be waiting to send; let them observe _disposed.
-            _sendLock.Release();
-        }
+        _sendLock.Dispose();
     }
 }

backend/fleet/tests/Fleet.Tests/BoundedTests.cs

@@ -1,28 +0,0 @@
-using Fleet.Core;
-
-namespace Fleet.Tests;
-
-public sealed class BoundedTests
-{
-    [Fact]
-    public async Task WaitAsync_ReturnsFalse_WhenTheWorkNeverFinishes()
-    {
-        var never = new TaskCompletionSource();
-        Assert.False(await Bounded.WaitAsync(never.Task, TimeSpan.FromMilliseconds(100), TimeProvider.System));
-    }
-
-    [Fact]
-    public async Task WaitAsync_ReturnsTrue_WhenTheWorkFinishes_OrFailsWithSessionNoise()
-    {
-        Assert.True(await Bounded.WaitAsync(Task.CompletedTask, TimeSpan.FromSeconds(5), TimeProvider.System));
-        Assert.True(await Bounded.WaitAsync(Task.FromException(new IOException("closed")), TimeSpan.FromSeconds(5), TimeProvider.System));
-        Assert.True(await Bounded.WaitAsync(Task.FromException(new ObjectDisposedException("x")), TimeSpan.FromSeconds(5), TimeProvider.System));
-        Assert.True(await Bounded.WaitAsync(Task.FromCanceled(new CancellationToken(true)), TimeSpan.FromSeconds(5), TimeProvider.System));
-    }
-
-    [Fact]
-    public async Task WaitAsync_DoesNotHideRealFailures()
-    {
-        await Assert.ThrowsAsync<FormatException>(() => Bounded.WaitAsync(Task.FromException(new FormatException()), TimeSpan.FromSeconds(5), TimeProvider.System));
-    }
-}

backend/fleet/tests/Fleet.Tests/BufferedNodeLogTests.cs

@@ -1,65 +0,0 @@
-using Fleet.Core;
-using Fleet.Hosting;
-
-namespace Fleet.Tests;
-
-public sealed class BufferedNodeLogTests
-{
-    [Fact]
-    public void Summary_CountsByLevelAndTypeWithoutText_AndResetsAfterEachTake()
-    {
-        var log = new BufferedNodeLog(NullNodeLog.Instance, TimeProvider.System);
-        for (int i = 0; i < 5; i++)
-        {
-            log.Warn("peer control handshake failed", ("attempt", i));
-        }
-
-        log.Info("link up");
-        log.Error("MsQuic QUIC transport shutdown");
-
-        NodeLogSummary first = log.TakeSummary();
-        Assert.Equal((1L, 5L, 1L), (first.Info, first.Warn, first.Error));
-        Assert.Equal(7, first.LastId);
-        NodeLogTypeCount top = first.Types[0];
-        Assert.Equal((BufferedNodeLog.TypeOf("peer control handshake failed"), "warn", 5L), (top.Type, top.Level, top.Count));
-
-        NodeLogSummary second = log.TakeSummary();
-        Assert.Equal((0L, 0L, 0L), (second.Info, second.Warn, second.Error));
-        Assert.Empty(second.Types);
-        Assert.Equal(7, second.LastId);
-    }
-
-    [Fact]
-    public void Read_ReturnsLinesAfterAnIdAtOrAboveALevel_OldestFirst_AndTheRingKeepsTheNewest()
-    {
-        var log = new BufferedNodeLog(NullNodeLog.Instance, TimeProvider.System);
-        log.Info("one");
-        log.Warn("two");
-        log.Error("three");
-        log.Info("four");
-
-        Assert.Equal(["two", "three"], log.Read(1, 2, null).Items.Select(e => e.Record.Message));
-        Assert.Equal(["three"], log.Read(0, 10, "error").Items.Select(e => e.Record.Message));
-        Assert.Equal(["two", "three"], log.Read(0, 10, "warn").Items.Select(e => e.Record.Message));
-
-        for (int i = 0; i < 3000; i++)
-        {
-            log.Info("filler", ("n", i));
-        }
-
-        NodeLogPage page = log.Read(0, 200, null);
-        Assert.Equal(200, page.Items.Count);
-        Assert.True(page.Items[0].Id > 4); // 最早的几条已经被环挤掉
-        Assert.Equal(3004, page.LastId);
-    }
-
-    [Fact]
-    public void SensitiveFieldsAreRedacted()
-    {
-        var log = new BufferedNodeLog(NullNodeLog.Instance, TimeProvider.System);
-        log.Info("auth", ("token", "abc"), ("agent", "agt_x"));
-        NodeLogEntry entry = Assert.Single(log.Read(0, 10, null).Items);
-        Assert.Equal("[redacted]", entry.Record.Fields["token"]);
-        Assert.Equal("agt_x", entry.Record.Fields["agent"]);
-    }
-}

backend/fleet/tests/Fleet.Tests/KnockTests.cs

@@ -15,14 +15,14 @@ public sealed class KnockTests
     [Fact]
     public void PairKey_IsTheSameOnBothSides_AndDiffersPerPair()
     {
-        using NodeIdentity dialer = NodeIdentity.Generate();
+        using NodeIdentity probe = NodeIdentity.Generate();
         using NodeIdentity agent = NodeIdentity.Generate();
         using NodeIdentity other = NodeIdentity.Generate();
-        byte[] fromDialer = Knock.DeriveKey(dialer, agent.PublicKeySpki);
-        byte[] fromAgent = Knock.DeriveKey(agent, dialer.PublicKeySpki);
-        Assert.Equal(fromDialer, fromAgent);
-        Assert.Equal(32, fromDialer.Length);
-        Assert.NotEqual(fromDialer, Knock.DeriveKey(other, agent.PublicKeySpki));
+        byte[] fromProbe = Knock.DeriveKey(probe, agent.PublicKeySpki);
+        byte[] fromAgent = Knock.DeriveKey(agent, probe.PublicKeySpki);
+        Assert.Equal(fromProbe, fromAgent);
+        Assert.Equal(32, fromProbe.Length);
+        Assert.NotEqual(fromProbe, Knock.DeriveKey(other, agent.PublicKeySpki));
     }
 
     [Fact]
@@ -61,36 +61,36 @@ public sealed class KnockTests
     public void Gate_AdmitsRosterMembersOnce_AndRejectsReplaysStrangersAndExpiredPackets()
     {
         using NodeIdentity agent = NodeIdentity.Generate();
-        using NodeIdentity dialer = NodeIdentity.Generate();
+        using NodeIdentity probe = NodeIdentity.Generate();
         using NodeIdentity stranger = NodeIdentity.Generate();
-        var roster = new List<byte[]> { dialer.PublicKeySpki };
+        var roster = new List<byte[]> { probe.PublicKeySpki };
         var gate = new KnockGate(agent, () => roster, new Clock());
         long now = 1_700_000_000_000;
 
-        byte[] packet = Knock.Build(Knock.DeriveKey(dialer, agent.PublicKeySpki), now);
+        byte[] packet = Knock.Build(Knock.DeriveKey(probe, agent.PublicKeySpki), now);
         Assert.True(gate.Accept(packet, now));
         Assert.False(gate.Accept(packet, now), "the same nonce cannot be used twice");
         Assert.False(gate.Accept(Knock.Build(Knock.DeriveKey(stranger, agent.PublicKeySpki), now), now), "not on the roster");
-        Assert.False(gate.Accept(Knock.Build(Knock.DeriveKey(dialer, agent.PublicKeySpki), now - 10 * 60_000), now), "expired");
+        Assert.False(gate.Accept(Knock.Build(Knock.DeriveKey(probe, agent.PublicKeySpki), now - 10 * 60_000), now), "expired");
 
         // 名册变了：新成员马上能进，被移除的马上进不来
         using NodeIdentity second = NodeIdentity.Generate();
         roster.Add(second.PublicKeySpki);
         Assert.True(gate.Accept(Knock.Build(Knock.DeriveKey(second, agent.PublicKeySpki), now), now));
         roster.RemoveAt(0);
-        Assert.False(gate.Accept(Knock.Build(Knock.DeriveKey(dialer, agent.PublicKeySpki), now), now));
+        Assert.False(gate.Accept(Knock.Build(Knock.DeriveKey(probe, agent.PublicKeySpki), now), now));
     }
 
     [Fact]
     public async Task Gate_RateLimitsEachSource_AndReadsAtMost32Bytes()
     {
         using NodeIdentity agent = NodeIdentity.Generate();
-        using NodeIdentity dialer = NodeIdentity.Generate();
+        using NodeIdentity probe = NodeIdentity.Generate();
         var clock = new Clock();
-        var gate = new KnockGate(agent, () => [dialer.PublicKeySpki], clock, attemptsPerPeriod: 3);
+        var gate = new KnockGate(agent, () => [probe.PublicKeySpki], clock, attemptsPerPeriod: 3);
         IPAddress noisy = IPAddress.Parse("203.0.113.9");
         IPAddress quiet = IPAddress.Parse("203.0.113.10");
-        byte[] key = Knock.DeriveKey(dialer, agent.PublicKeySpki);
+        byte[] key = Knock.DeriveKey(probe, agent.PublicKeySpki);
 
         for (int i = 0; i < 3; i++)
         {

backend/fleet/tests/Fleet.Tests/LinkConnectTimeoutTests.cs

@@ -1,106 +0,0 @@
-using System.Net;
-using System.Net.Sockets;
-using Fleet.Core;
-using Fleet.Identity;
-using Fleet.Link;
-using Fleet.Rpc;
-
-namespace Fleet.Tests;
-
-/// <summary>面板那头只接住连接、什么都不回（线路把握手吞了、代理卡住）：节点要在限定时间内放弃这次连接重试，而不是一直等着。</summary>
-public sealed class LinkConnectTimeoutTests
-{
-    [Fact]
-    public async Task ASilentServer_IsGivenUp_AfterTheConnectTimeout_AndRetried()
-    {
-        using var listener = new TcpListener(IPAddress.Loopback, 0);
-        listener.Start();
-        var accepted = new List<TcpClient>();
-        using var acceptLoop = new CancellationTokenSource();
-        Task accepting = Task.Run(async () =>
-        {
-            try
-            {
-                while (true)
-                {
-                    accepted.Add(await listener.AcceptTcpClientAsync(acceptLoop.Token));
-                }
-            }
-            catch (OperationCanceledException)
-            {
-                // 测试结束
-            }
-        });
-
-        var options = new FleetLinkOptions
-        {
-            Server = new Uri($"http://127.0.0.1:{((IPEndPoint)listener.LocalEndpoint).Port}"),
-            Identity = NodeIdentity.Generate(),
-            NodeId = new NodeId("dlr_test"),
-            ConnectTimeout = TimeSpan.FromMilliseconds(500),
-            Backoff = new Backoff(TimeSpan.FromMilliseconds(10), TimeSpan.FromMilliseconds(50)),
-        };
-        var router = new RpcRouter<LinkContext>(RpcSide.Node, FleetMethodTable.Table);
-        var link = new FleetLink(options, router, [], new Outbox(1 << 16), TimeProvider.System, new SilentLog());
-        var reasons = new List<string>();
-        link.OnDisconnected = reason =>
-        {
-            lock (reasons)
-            {
-                reasons.Add(reason);
-            }
-        };
-
-        using var stop = new CancellationTokenSource();
-        Task running = link.RunAsync(stop.Token);
-        try
-        {
-            using var deadline = new CancellationTokenSource(TimeSpan.FromSeconds(10));
-            while (true)
-            {
-                lock (reasons)
-                {
-                    if (reasons.Count >= 2)
-                    {
-                        break;
-                    }
-                }
-
-                await Task.Delay(50, deadline.Token);
-            }
-
-            lock (reasons)
-            {
-                Assert.All(reasons, reason => Assert.Contains("timed out", reason, StringComparison.OrdinalIgnoreCase));
-            }
-
-            Assert.True(accepted.Count >= 2, "the second attempt connects again");
-        }
-        finally
-        {
-            await stop.CancelAsync();
-            try
-            {
-                await running;
-            }
-            catch (OperationCanceledException) when (stop.IsCancellationRequested)
-            {
-                // Canceling the retry delay ends RunAsync with the requested cancellation.
-            }
-
-            await acceptLoop.CancelAsync();
-            await accepting;
-            foreach (TcpClient client in accepted)
-            {
-                client.Dispose();
-            }
-        }
-    }
-
-    private sealed class SilentLog : INodeLog
-    {
-        public void Write(NodeLogLevel level, string message, params (string Key, object? Value)[] fields)
-        {
-        }
-    }
-}

backend/fleet/tests/Fleet.Tests/LinkTests.cs

@@ -275,27 +275,6 @@ public sealed class LinkTests
         await run.WaitAsync(TimeSpan.FromSeconds(5));
     }
 
-    [Fact]
-    public async Task Session_RunEnds_EvenWhenAnInboundHandlerIsStuck()
-    {
-        (TlsFrameTransport left, TlsFrameTransport right) = await PairAsync();
-        MethodTable table = TestTables.Of(TestTables.Req("right.stuck"));
-        var stuck = new TaskCompletionSource<Result<Number>>(TaskCreationOptions.RunContinuationsAsynchronously); // 不理会取消令牌：模拟卡在原生层的处理器
-        var rightRouter = new RpcRouter<string>(RpcSide.Listener, table).On("right.stuck", LinkJson.Default.Number, LinkJson.Default.Number, (_, _, _) => new ValueTask<Result<Number>>(stuck.Task));
-        await using var leftSession = new RpcSession<string>(left, new RpcRouter<string>(RpcSide.Dialer, table), "left", TimeProvider.System);
-        await using var rightSession = new RpcSession<string>(right, rightRouter, "right", TimeProvider.System);
-        Task leftRun = leftSession.RunAsync(default);
-        Task rightRun = rightSession.RunAsync(default);
-
-        _ = leftSession.RequestAsync("right.stuck", new Number { Value = 1 }, LinkJson.Default.Number, LinkJson.Default.Number, default, TimeSpan.FromSeconds(30)).AsTask();
-        await Task.Delay(300); // 请求到达并进入处理器
-        await leftSession.CloseAsync("bye", default);
-
-        await rightRun.WaitAsync(TimeSpan.FromSeconds(10)); // 以前这里会一直等到处理器返回
-        stuck.TrySetResult(Result<Number>.Ok(new Number { Value = 0 }));
-        await leftRun.WaitAsync(TimeSpan.FromSeconds(5));
-    }
-
     // ───────────────────────── 点对点握手 ─────────────────────────
 
     private sealed class PeerFixture : IAsyncDisposable
@@ -303,7 +282,7 @@ public sealed class LinkTests
         private readonly CancellationTokenSource _stop = new();
         private readonly NodeIdentity _identity;
 
-        public PeerFixture(Func<string, byte[]?> lookup, Func<PeerSessionInfo, IRpcTransport, CancellationToken, Task>? onSession = null, int maxSessions = 64)
+        public PeerFixture(Func<string, byte[]?> lookup, Func<PeerSessionInfo, IRpcTransport, CancellationToken, Task>? onSession = null)
         {
             Certificate = NodeCertificate.LoadPfx(NodeCertificate.CreateSelfSignedPfx("agent", TimeSpan.FromDays(30), TimeProvider.System));
             _identity = NodeIdentity.FromPkcs8(Certificate.GetECDsaPrivateKey()!.ExportPkcs8PrivateKey());
@@ -314,7 +293,6 @@ public sealed class LinkTests
                     Certificate = Certificate,
                     LookupPeerKey = lookup,
                     Knock = new KnockGate(_identity, () => KnockPeers, TimeProvider.System),
-                    MaxSessions = maxSessions,
                 },
                 onSession ?? ((_, transport, ct) => KeepOpenAsync(transport, ct)),
                 NullNodeLog.Instance, TimeProvider.System);
@@ -329,10 +307,10 @@ public sealed class LinkTests
 
         public byte[] Spki => NodeCertificate.SpkiFingerprint(Certificate);
 
-        /// <summary>Agent 的 TLS 证书公钥：拨号方 用它固定证书、派生敲门密钥。</summary>
+        /// <summary>Agent 的 TLS 证书公钥：Probe 用它固定证书、派生敲门密钥。</summary>
         public byte[] PublicKey => Certificate.PublicKey.ExportSubjectPublicKeyInfo();
 
-        /// <summary>名册：能敲开门的 拨号方 公钥。</summary>
+        /// <summary>名册：能敲开门的 Probe 公钥。</summary>
         public List<byte[]> KnockPeers { get; } = [];
 
         public async ValueTask DisposeAsync()
@@ -353,90 +331,81 @@ public sealed class LinkTests
     }
 
     [Fact]
-    public async Task PeerHandshake_AuthenticatesTheDialer_AndBothSidesDeriveTheSameDataPlaneCredentials()
+    public async Task PeerHandshake_AuthenticatesTheProbe_AndBothSidesDeriveTheSameDataPlaneCredentials()
     {
-        using NodeIdentity dialer = NodeIdentity.Generate();
+        using NodeIdentity probe = NodeIdentity.Generate();
         var seen = new TaskCompletionSource<PeerSessionInfo>();
-        await using var agent = new PeerFixture(id => id == "dlr_1" ? dialer.PublicKeySpki : null, (info, transport, ct) =>
+        await using var agent = new PeerFixture(id => id == "prb_1" ? probe.PublicKeySpki : null, (info, transport, ct) =>
         {
             seen.TrySetResult(info);
             return Task.Delay(Timeout.Infinite, ct).ContinueWith(_ => { }, CancellationToken.None);
         });
-        agent.KnockPeers.Add(dialer.PublicKeySpki);
+        agent.KnockPeers.Add(probe.PublicKeySpki);
         await agent.Listener.Started.Task;
 
-        var logs = new Fleet.Hosting.BufferedNodeLog(NullNodeLog.Instance, TimeProvider.System);
-        var scoped = new ContextNodeLog(logs, ("attempt_id", "test-attempt"));
-        (PeerSessionInfo client, TlsFrameTransport transport) = await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, dialer, "dlr_1", null, 1 << 20, TimeProvider.System, scoped, default);
+        (PeerSessionInfo client, TlsFrameTransport transport) = await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, probe, "prb_1", null, 1 << 20, TimeProvider.System, default);
         await using (transport)
         {
             PeerSessionInfo server = await seen.Task.WaitAsync(TimeSpan.FromSeconds(5));
-            Assert.Equal("dlr_1", server.PeerId);
+            Assert.Equal("prb_1", server.PeerId);
             Assert.Equal(client.Credentials.Key, server.Credentials.Key);
             Assert.Equal(client.Credentials.SessionId, server.Credentials.SessionId);
-            NodeLogPage page = logs.Read(0, 32, null);
-            Assert.Equal(new[] { "tcp_connect", "tls", "peer_auth" }, page.Items.Select(x => x.Record.Fields["phase"]));
-            Assert.All(page.Items, entry => Assert.Equal("test-attempt", entry.Record.Fields["attempt_id"]));
         }
     }
 
     [Fact]
-    public async Task PeerHandshake_RejectsUnknownDialers_AndForgedSignatures_WithTheSameError()
+    public async Task PeerHandshake_RejectsUnknownProbes_AndForgedSignatures_WithTheSameError()
     {
-        using NodeIdentity dialer = NodeIdentity.Generate();
+        using NodeIdentity probe = NodeIdentity.Generate();
         using NodeIdentity impostor = NodeIdentity.Generate();
-        await using var agent = new PeerFixture(id => id == "dlr_1" ? dialer.PublicKeySpki : null);
-        agent.KnockPeers.AddRange([dialer.PublicKeySpki, impostor.PublicKeySpki]); // 敲门通过之后，名册外的 ID 和伪造的签名仍然被拒绝
+        await using var agent = new PeerFixture(id => id == "prb_1" ? probe.PublicKeySpki : null);
+        agent.KnockPeers.AddRange([probe.PublicKeySpki, impostor.PublicKeySpki]); // 敲门通过之后，名册外的 ID 和伪造的签名仍然被拒绝
         await agent.Listener.Started.Task;
 
-        var logs = new Fleet.Hosting.BufferedNodeLog(NullNodeLog.Instance, TimeProvider.System);
         InvalidOperationException unknown = await Assert.ThrowsAsync<InvalidOperationException>(async () =>
-            await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, dialer, "dlr_unknown", null, 1 << 20, TimeProvider.System, logs, default));
-        NodeLogRecord failure = Assert.Single(logs.Read(0, 32, null).Items, entry => entry.Record.Message == "peer control handshake failed").Record;
-        Assert.Equal("peer_auth", failure.Fields["phase"]);
-        Assert.Equal("InvalidOperationException", failure.Fields["exception"]);
+            await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, probe, "prb_unknown", null, 1 << 20, TimeProvider.System, default));
         InvalidOperationException forged = await Assert.ThrowsAsync<InvalidOperationException>(async () =>
-            await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, impostor, "dlr_1", null, 1 << 20, TimeProvider.System, default));
+            await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, impostor, "prb_1", null, 1 << 20, TimeProvider.System, default));
         Assert.Equal(unknown.Message, forged.Message);
         Assert.Contains("fleet_unauthorized", forged.Message);
 
         // 正主仍然可以连接
-        (_, TlsFrameTransport ok) = await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, dialer, "dlr_1", null, 1 << 20, TimeProvider.System, default);
+        (_, TlsFrameTransport ok) = await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, probe, "prb_1", null, 1 << 20, TimeProvider.System, default);
         await ok.DisposeAsync();
     }
 
     [Fact]
     public async Task PeerHandshake_PinsTheAgentCertificate_ByItsSpkiFingerprint()
     {
-        using NodeIdentity dialer = NodeIdentity.Generate();
-        await using var agent = new PeerFixture(_ => dialer.PublicKeySpki);
-        agent.KnockPeers.Add(dialer.PublicKeySpki);
+        using NodeIdentity probe = NodeIdentity.Generate();
+        await using var agent = new PeerFixture(_ => probe.PublicKeySpki);
+        agent.KnockPeers.Add(probe.PublicKeySpki);
         await agent.Listener.Started.Task;
 
         // 公钥不对：敲门密钥对不上，Agent 直接断开（拿到的不是证书，连不是它的证书都看不到）
         using NodeIdentity another = NodeIdentity.Generate();
         await Assert.ThrowsAnyAsync<Exception>(async () =>
-            await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, another.PublicKeySpki, dialer, "dlr_1", null, 1 << 20, TimeProvider.System, default));
+            await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, another.PublicKeySpki, probe, "prb_1", null, 1 << 20, TimeProvider.System, default));
     }
 
     [Fact]
     public async Task PeerSession_CarriesRpcBothWays_AfterTheHandshake()
     {
-        using NodeIdentity dialer = NodeIdentity.Generate();
+        using NodeIdentity probe = NodeIdentity.Generate();
         MethodTable table = TestTables.Of(TestTables.Req("agent.echo"));
         var router = new RpcRouter<string>(RpcSide.Listener, table).On("agent.echo", LinkJson.Default.Number, LinkJson.Default.Number, (_, p, _) => ValueTask.FromResult(Result<Number>.Ok(new Number { Value = p.Value + 1 })));
         await using var agent = new PeerFixture(
-            _ => dialer.PublicKeySpki,
+            _ => probe.PublicKeySpki,
             async (info, transport, ct) =>
             {
                 await using var session = new RpcSession<string>(transport, router, info.PeerId, TimeProvider.System);
                 await session.RunAsync(ct);
             });
-        agent.KnockPeers.Add(dialer.PublicKeySpki);
+        agent.KnockPeers.Add(probe.PublicKeySpki);
         await agent.Listener.Started.Task;
 
-        (_, TlsFrameTransport transport) = await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, dialer, "dlr_1", null, 1 << 20, TimeProvider.System, default);
-        await using var client = new RpcSession<string>(transport, new RpcRouter<string>(RpcSide.Dialer, table), "dialer", TimeProvider.System);
+        (_, TlsFrameTransport transport) = await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, probe, "prb_1", null, 1 << 20, TimeProvider.System, default);
+        await using var client = new RpcSession<string>(transport, new RpcRouter<string>(RpcSide.Dialer, table), "probe", TimeProvider.System);
         Task run = client.RunAsync(default);
         Result<Number> result = await client.RequestAsync("agent.echo", new Number { Value = 41 }, LinkJson.Default.Number, LinkJson.Default.Number, default);
         Assert.Equal(42, result.Value.Value);
@@ -447,9 +416,9 @@ public sealed class LinkTests
     [Fact]
     public async Task PeerListener_LimitsConcurrentSessions_AndSurvivesGarbageConnections()
     {
-        using NodeIdentity dialer = NodeIdentity.Generate();
-        await using var agent = new PeerFixture(_ => dialer.PublicKeySpki);
-        agent.KnockPeers.Add(dialer.PublicKeySpki);
+        using NodeIdentity probe = NodeIdentity.Generate();
+        await using var agent = new PeerFixture(_ => probe.PublicKeySpki);
+        agent.KnockPeers.Add(probe.PublicKeySpki);
         await agent.Listener.Started.Task;
 
         // 不说 TLS 的连接：握手失败，监听继续
@@ -459,75 +428,16 @@ public sealed class LinkTests
             await garbage.GetStream().WriteAsync("GET / HTTP/1.1\r\n\r\n"u8.ToArray());
         }
 
-        (_, TlsFrameTransport ok) = await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, dialer, "dlr_1", null, 1 << 20, TimeProvider.System, default);
+        (_, TlsFrameTransport ok) = await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, probe, "prb_1", null, 1 << 20, TimeProvider.System, default);
         await ok.DisposeAsync();
     }
 
-    /// <summary>名额只数通过敲门的会话：一堆还没敲门的连接占着，名册里的拨号方照样连得上（之前它们共用一个上限，节点多了之后后到的拨号方被永远拒绝）。</summary>
-    [Fact]
-    public async Task PeerListener_SessionLimitCountsOnlyAdmittedPeers_SoSilentConnectionsCannotLockOutTheRoster()
-    {
-        using NodeIdentity first = NodeIdentity.Generate();
-        using NodeIdentity second = NodeIdentity.Generate();
-        using NodeIdentity third = NodeIdentity.Generate();
-        await using var agent = new PeerFixture(id => id == "dlr_1" ? first.PublicKeySpki : id == "dlr_2" ? second.PublicKeySpki : third.PublicKeySpki, maxSessions: 2);
-        agent.KnockPeers.AddRange([first.PublicKeySpki, second.PublicKeySpki, third.PublicKeySpki]);
-        await agent.Listener.Started.Task;
-
-        // 20 个什么都不说的连接：占着待敲门的名额，但不占会话名额
-        var silent = new List<TcpClient>();
-        try
-        {
-            for (int i = 0; i < 20; i++)
-            {
-                var client = new TcpClient();
-                await client.ConnectAsync(IPAddress.Loopback, agent.Listener.Port);
-                silent.Add(client);
-            }
-
-            (_, TlsFrameTransport a) = await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, first, "dlr_1", null, 1 << 20, TimeProvider.System, default);
-            (_, TlsFrameTransport b) = await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, second, "dlr_2", null, 1 << 20, TimeProvider.System, default);
-
-            // 会话名额（2）满了：第三个在名册里的拨号方被拒绝，一个会话结束之后又能连上
-            await Assert.ThrowsAnyAsync<Exception>(async () =>
-            {
-                (_, TlsFrameTransport refused) = await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, third, "dlr_3", null, 1 << 20, TimeProvider.System, default);
-                await refused.DisposeAsync();
-            });
-            await a.DisposeAsync();
-            bool reconnected = false;
-            for (int attempt = 0; attempt < 50 && !reconnected; attempt++)
-            {
-                try
-                {
-                    (_, TlsFrameTransport again) = await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, third, "dlr_3", null, 1 << 20, TimeProvider.System, default);
-                    await again.DisposeAsync();
-                    reconnected = true;
-                }
-                catch (Exception ex) when (ex is IOException or System.Security.Authentication.AuthenticationException or InvalidOperationException)
-                {
-                    await Task.Delay(100);
-                }
-            }
-
-            Assert.True(reconnected);
-            await b.DisposeAsync();
-        }
-        finally
-        {
-            foreach (TcpClient client in silent)
-            {
-                client.Dispose();
-            }
-        }
-    }
-
     [Fact]
     public async Task PeerListener_WritesNothingToStrangers_NotEvenATlsServerHello()
     {
-        using NodeIdentity dialer = NodeIdentity.Generate();
-        await using var agent = new PeerFixture(_ => dialer.PublicKeySpki);
-        agent.KnockPeers.Add(dialer.PublicKeySpki);
+        using NodeIdentity probe = NodeIdentity.Generate();
+        await using var agent = new PeerFixture(_ => probe.PublicKeySpki);
+        agent.KnockPeers.Add(probe.PublicKeySpki);
         await agent.Listener.Started.Task;
 
         // 一个像样的 TLS ClientHello 开头、随机数据、什么都不发：都拿不到任何字节（连接被直接关闭）
@@ -555,8 +465,8 @@ public sealed class LinkTests
             Assert.Equal(0, read);
         }
 
-        // 名册里的 拨号方 仍然进得来
-        (_, TlsFrameTransport ok) = await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, dialer, "dlr_1", null, 1 << 20, TimeProvider.System, default);
+        // 名册里的 Probe 仍然进得来
+        (_, TlsFrameTransport ok) = await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, probe, "prb_1", null, 1 << 20, TimeProvider.System, default);
         await ok.DisposeAsync();
     }
 

backend/fleet/tests/Fleet.Tests/MethodDirectionTests.cs

@@ -34,10 +34,10 @@ public sealed class MethodDirectionTests
         // t.ping 声明成请求：不能按通知注册
         Assert.Throws<InvalidOperationException>(() => server.OnNotify("t.ping", EchoContext.Default.AddParams, (_, _, _) => ValueTask.CompletedTask));
 
-        // 节点这一侧：t.note 只给 agent 角色，dialer 注册不了；t.push 是它该处理的
-        var dialer = new RpcRouter<string>(RpcSide.Node, s_table, "dialer");
-        dialer.On("t.push", EchoContext.Default.AddParams, EchoContext.Default.AddResult, (_, _, _) => throw new NotSupportedException());
-        Assert.Throws<InvalidOperationException>(() => new RpcRouter<string>(RpcSide.Server, s_table, "dialer").OnNotify("t.note", EchoContext.Default.AddParams, (_, _, _) => ValueTask.CompletedTask));
+        // 节点这一侧：t.note 只给 agent 角色，probe 注册不了；t.push 是它该处理的
+        var probe = new RpcRouter<string>(RpcSide.Node, s_table, "probe");
+        probe.On("t.push", EchoContext.Default.AddParams, EchoContext.Default.AddResult, (_, _, _) => throw new NotSupportedException());
+        Assert.Throws<InvalidOperationException>(() => new RpcRouter<string>(RpcSide.Server, s_table, "probe").OnNotify("t.note", EchoContext.Default.AddParams, (_, _, _) => ValueTask.CompletedTask));
 
         server.On("t.ping", EchoContext.Default.AddParams, EchoContext.Default.AddResult, (_, _, _) => ValueTask.FromResult(Result<AddResult>.Ok(new AddResult())));
     }
@@ -99,7 +99,7 @@ public sealed class MethodDirectionTests
     {
         RpcRouter<string> server = Server().OnNotify("t.note", EchoContext.Default.AddParams, (_, _, _) => ValueTask.CompletedTask);
         Assert.Null((await server.DispatchCheckedAsync("agent", Notification("t.note"), default)).Violation);
-        Assert.NotNull((await server.DispatchCheckedAsync("dialer", Notification("t.note"), default)).Violation);
+        Assert.NotNull((await server.DispatchCheckedAsync("probe", Notification("t.note"), default)).Violation);
         Assert.NotNull((await server.DispatchCheckedAsync(null!, Notification("t.note"), default)).Violation);
     }
 
@@ -115,7 +115,7 @@ public sealed class MethodDirectionTests
         // 对端的角色也要允许
         var node = new RpcRouter<string>(RpcSide.Node, s_table, "agent");
         node.CheckSend("t.note", MethodKind.Notification, "agent");
-        Assert.Throws<InvalidOperationException>(() => node.CheckSend("t.note", MethodKind.Notification, "dialer"));
+        Assert.Throws<InvalidOperationException>(() => node.CheckSend("t.note", MethodKind.Notification, "probe"));
     }
 
     [Fact]
@@ -126,7 +126,7 @@ public sealed class MethodDirectionTests
 
         // 服务端要处理 t.ping 和 t.note；t.inline 在握手阶段直接处理，不算缺
         Assert.Equal(["t.note", "t.ping"], s_table.MissingHandlers(RpcSide.Server, null, []));
-        Assert.Equal(["t.ping"], s_table.MissingHandlers(RpcSide.Server, "dialer", []));
+        Assert.Equal(["t.ping"], s_table.MissingHandlers(RpcSide.Server, "probe", []));
         Assert.Equal(["t.note"], s_table.MissingHandlers(RpcSide.Server, "agent", ["t.ping"]));
         Assert.Empty(s_table.MissingHandlers(RpcSide.Server, "agent", ["t.ping", "t.note"]));
         // 同名方法在对等面的接收方是拨号方？不：t.ping 在对等面由监听方发起，拨号方处理

backend/fleet/tests/Fleet.Tests/NodeVersionsTests.cs

@@ -1,21 +0,0 @@
-using Fleet.Core;
-
-namespace Fleet.Tests;
-
-public sealed class NodeVersionsTests
-{
-    [Theory]
-    [InlineData("0.4.2-dev.73-5245f03", "0.4.2-dev.74", true)]
-    [InlineData("0.4.2-dev.73-5245f03", "0.4.2-dev.73-abcdef0", false)] // 哈希不参与比较
-    [InlineData("0.4.2-dev.100-aaaaaaa", "0.4.2-dev.99", false)] // 序号按数字比，不是按字符串
-    [InlineData("0.4.2-dev.999", "0.4.2", true)] // 正式版高于预发布版
-    [InlineData("0.4.2", "0.4.2-dev.5", false)]
-    [InlineData("0.4.1", "0.4.2-dev.1", true)]
-    [InlineData("v0.5.0", "0.4.9", false)]
-    [InlineData("0.4.2-dev.73-5245f03", "", false)] // 没有要求
-    [InlineData("", "0.4.2", false)] // 认不出来的版本号不拒绝
-    [InlineData("not-a-version", "0.4.2", false)]
-    [InlineData("0.4.2", "garbage", false)]
-    public void IsBelow_ComparesCoreThenReleaseThenTheDevCounter(string actual, string required, bool expected) =>
-        Assert.Equal(expected, NodeVersions.IsBelow(actual, required));
-}

backend/fleet/tests/Fleet.Tests/OwnerGuardTests.cs

@@ -4,7 +4,7 @@ using Fleet.Identity;
 
 namespace Fleet.Tests;
 
-/// <summary>节点自己要求有属主（passkey）：拨号方和监听方共用这一份，没有属主就不运行。</summary>
+/// <summary>节点自己要求有属主（passkey）：Agent 和 Probe 共用这一份，没有属主就不运行。</summary>
 public sealed class OwnerGuardTests : IDisposable
 {
     private readonly TestAuthenticator _alice = new();

backend/fleet/tools/fleet-enroll/Program.cs

@@ -2,7 +2,7 @@ using System.Text.Json.Nodes;
 using Fleet.Core;
 using Fleet.Enrollment.Client;
 
-// fleet-enroll --server https://panel --join jt_xxx --name jp-1 --identity-dir /etc/moshiro [--meta '{"tcp_port":7443,...}'] [--version 1.0]
+// fleet-enroll --server https://panel --join jt_xxx --name jp-1 --identity-dir /etc/moshiro [--meta '{"probe_tcp_port":7443,...}'] [--version 1.0]
 // 一次性运行：生成身份、注册、写入身份目录，然后退出。
 string? Arg(string name) => args.SkipWhile(a => a != name).Skip(1).FirstOrDefault();
 

backend/fleet/tools/fleet-enroll/fleet-enroll.csproj

@@ -33,6 +33,4 @@
   <Import Project="Sdk.targets" Sdk="Microsoft.NET.Sdk" />
   <Import Project="Sdk.targets" Sdk="StuDev.AotAnywhere"
           Condition="'$(UseAotAnywhere)' == 'true'" />
-  <Import Project="..\..\..\shared\MoShiro.Measurement\StaticMsQuic.targets"
-          Condition="'$(StaticMsQuic)' == 'true'" />
 </Project>

backend/probe/Directory.Build.props

@@ -0,0 +1,21 @@
+<Project>
+  <Import Project="$([MSBuild]::GetPathOfFileAbove('Directory.Build.props', '$(MSBuildThisFileDirectory)../'))" />
+  <PropertyGroup>
+    <OptimizationPreference>Size</OptimizationPreference>
+    <UseSystemResourceKeys>true</UseSystemResourceKeys>
+    <StackTraceSupport>false</StackTraceSupport>
+    <IlcFoldIdenticalMethodBodies>true</IlcFoldIdenticalMethodBodies>
+    <EventSourceSupport>false</EventSourceSupport>
+    <DebuggerSupport>false</DebuggerSupport>
+    <HttpActivityPropagationSupport>false</HttpActivityPropagationSupport>
+    <Http3Support>false</Http3Support>
+    <MetricsSupport>false</MetricsSupport>
+    <UseSizeOptimizedLinq>true</UseSizeOptimizedLinq>
+    <JsonSerializerIsReflectionEnabledByDefault>false</JsonSerializerIsReflectionEnabledByDefault>
+    <ServerGarbageCollection>false</ServerGarbageCollection>
+    <ConcurrentGarbageCollection>false</ConcurrentGarbageCollection>
+  </PropertyGroup>
+  <ItemGroup>
+    <AdditionalFiles Include="$(MSBuildThisFileDirectory)../BannedSymbols.Node.txt" />
+  </ItemGroup>
+</Project>

backend/probe/README.md

@@ -0,0 +1,59 @@
+# MoShiro Probe
+
+Probe 是网络测量的发起端，通常运行在家宽或其他需要观测的网络出口。它连接 Backend（身份、Agent 列表、状态）和 Agent（测量），执行线路任务，并把本端采集的原始数据交给 Agent 汇总上报。Probe 不提供远程命令或文件管理功能。
+
+## 工作流程
+
+```text
+Backend ──任务──> Agent <──测量会话──> Probe
+   │  ▲           │                     │
+   │  └──汇总并上报数据 <───┘                     │
+   └─────Agent 列表、状态（Fleet 长连接）───────────┘
+```
+
+Probe 用 `probe.json` 读取面板地址（`fleet.server`）、身份目录（`fleet.identity_dir`，由 `fleet-enroll` 注册时生成）和本地测试限制。连上面板之后，面板下发要连接的 Agent 列表；Probe 将列表缓存到同目录的 `agents.json`，面板暂时连不上时靠它继续测量。身份目录里有私钥，请限制访问权限，不要提交到版本控制。
+
+容器节点的运行方式（podman + systemd）和宿主机更新器见 [`../../docs/deployment.md`](../../docs/deployment.md)；Probe 也可以用 OCI 镜像 `registry.oeo.one/moshiro/moshiro-probe:<version>`，容器默认读取 `/etc/moshiro/probe.json`，命名 volume 保存身份配置和缓存，镜像以 UID/GID 1654 运行。
+
+## 构建
+
+需要 .NET SDK 10。为应用 根目录的 `global.json` 的 SDK 设置，请从 `backend/` 目录构建：
+
+```sh
+cd backend
+dotnet build probe/src/MoShiro.Probe -c Release
+```
+
+发布 NativeAOT 程序需要对应平台的 C/C++ 工具链；`<RID>` 应与构建环境和目标平台匹配：
+
+```sh
+dotnet publish probe/src/MoShiro.Probe -c Release -r <RID> -p:PublishAot=true
+```
+
+也可在仓库根目录运行 `bash scripts/publish.sh probe <RID>`，再使用 `scripts/package.sh probe <RID>` 生成部署包。支持平台和安装流程见 [`../../docs/deployment.md`](../../docs/deployment.md)。
+
+## 配置与运行
+
+在管理端添加 Probe（生成 probe 角色的加入令牌），在目标机上注册（密钥由 Probe 自己生成）：
+
+```sh
+fleet-enroll --server https://panel.example.com --join <加入令牌> --name home --identity-dir /etc/moshiro
+```
+
+写 `/etc/moshiro/probe.json`：`{"fleet":{"server":"https://panel.example.com","identity_dir":"/etc/moshiro"}}`（`install.sh probe --server … --join … --name …` 会一并完成）。随后运行：
+
+```sh
+moshiro-probe --config /etc/moshiro/probe.json
+```
+
+不传 `--config` 时默认读取 `/etc/moshiro/probe.json`。配置中的 `limits` 是本地硬上限，Backend 无法远程放宽；路由器等低性能设备可调低测试码率和包速率，避免 Probe 自身成为测量瓶颈。部署时可使用仓库的 `packaging/` 文件和安装脚本。
+
+也可使用 OCI 镜像 `registry.oeo.one/moshiro/moshiro-probe:<version>`，自动选择 Linux amd64 或 arm64 镜像。容器默认读取 `/etc/moshiro/probe.json`，并在同一目录维护 `agents.json` 缓存，配置与缓存存放在命名 volume 中。建议使用宿主网络；当前独立 ICMP 目标任务由 Agent 执行。
+
+## 设计说明
+
+- Probe 只执行编译进程序的测量模块。它不落盘测量数据，也不能执行命令或自行更新（更新由宿主机上的更新器完成，Probe 只写清单和请求文件）。
+- 测量模块由 `backend/shared/MoShiro.Measurement/` 与 Agent 共用；Probe 负责发起连接、调度任务和记录本端数据。
+- 测量模型和协议细节见 [`../../docs/measurement.md`](../../docs/measurement.md) 与 [`../../docs/protocol.md`](../../docs/protocol.md)；整体架构见 [`../../docs/architecture.md`](../../docs/architecture.md)。
+
+Docker 与 rootless/rootful Podman 共用 `packaging/compose/probe/compose.yaml`，配置、身份和缓存放在命名 volume 中，无需宿主配置目录、keep-id 或手动 chown。首次导入和自动更新见部署文档。

backend/probe/src/MoShiro.Probe/Link/AgentConnection.cs

@@ -0,0 +1,422 @@
+using System.Collections.Concurrent;
+using System.Net;
+using System.Net.Sockets;
+using System.Text.Json;
+using Fleet.Core;
+using Fleet.Identity;
+using Fleet.Link;
+using Fleet.Rpc;
+using MoShiro.Measurement;
+using MoShiro.Protocol;
+
+namespace MoShiro.Probe;
+
+/// <summary>所有连接共用的东西。</summary>
+public sealed record ConnectionEnv(NodeIdentity Identity, string ProbeId, ProbeLimits Limits, AgentDirectory Directory, MonoClock Clock, TimeProvider Time, INodeLog Log);
+
+/// <summary>
+/// 与一个 Agent 的点对点会话：认证握手 → 同步时钟 → 按 Agent 下发的 <c>line_task_set</c> 运行角色 →
+/// 每 10 秒把已完成的行切成列块放进发件箱 → 响应按需测试的阶段。会话断开时所有角色停止（凭据属于会话），按退避重连。
+/// 会话内的所有状态（套接字、发件箱、角色）随会话一起创建和释放。
+/// </summary>
+public sealed class AgentConnection
+{
+    private static readonly TimeSpan s_flushInterval = TimeSpan.FromSeconds(ProtocolLimits.ChunkSeconds);
+    private static readonly TimeSpan s_clockInterval = TimeSpan.FromSeconds(30);
+    private static readonly RpcRouter<ConnectionSession> s_router = BuildRouter();
+
+    /// <summary>对等面入站方法的路由器（只读，测试用）。</summary>
+    public static RpcRouter<ConnectionSession> Router => s_router;
+
+    private readonly ConnectionEnv _env;
+    private volatile ConnectionSession? _session;
+
+    public AgentConnection(KnownAgent endpoint, ConnectionEnv env)
+    {
+        Endpoint = endpoint;
+        _env = env;
+    }
+
+    public KnownAgent Endpoint { get; }
+
+    public bool IsConnected => _session is not null;
+
+    public int RoleCount => _session?.RoleCount ?? 0;
+
+    public long ClockOffsetUs => _session?.Clock.OffsetUs ?? 0;
+
+    public async Task RunAsync(CancellationToken ct)
+    {
+        var backoff = new Backoff(TimeSpan.FromSeconds(1), TimeSpan.FromSeconds(20));
+        while (!ct.IsCancellationRequested)
+        {
+            try
+            {
+                string reason = await RunOnceAsync(backoff, ct);
+                _env.Log.Warn("agent session ended", ("agent", Endpoint.AgentId), ("err", reason));
+            }
+            catch (OperationCanceledException) when (ct.IsCancellationRequested)
+            {
+                return;
+            }
+            catch (Exception ex) when (ex is IOException or System.Net.Sockets.SocketException or InvalidOperationException or System.Security.Authentication.AuthenticationException or OperationCanceledException or FormatException)
+            {
+                _env.Log.Warn("agent connection failed", ("agent", Endpoint.AgentId), ("err", ex.Message));
+            }
+
+            try
+            {
+                await Task.Delay(backoff.Next(), _env.Time, ct);
+            }
+            catch (OperationCanceledException)
+            {
+                return;
+            }
+        }
+    }
+
+    private async Task<string> RunOnceAsync(Backoff backoff, CancellationToken ct)
+    {
+        AddressFamily family = Endpoint.IpVersion == IpVersions.V6 ? AddressFamily.InterNetworkV6 : AddressFamily.InterNetwork;
+        using var resolveTimeout = new CancellationTokenSource(TimeSpan.FromSeconds(10), _env.Time);
+        using CancellationTokenSource resolveCt = CancellationTokenSource.CreateLinkedTokenSource(ct, resolveTimeout.Token);
+        IPAddress[] addresses = IPAddress.TryParse(Endpoint.Host, out IPAddress? literal)
+            ? [literal]
+            : await Dns.GetHostAddressesAsync(Endpoint.Host, family, resolveCt.Token);
+        IPAddress? address = addresses.FirstOrDefault(a => a.AddressFamily == family && !a.IsIPv4MappedToIPv6);
+        if (address is null)
+        {
+            throw new SocketException((int)SocketError.HostNotFound);
+        }
+
+        byte[] agentKey = Convert.FromBase64String(Endpoint.PublicKey);
+        (PeerSessionInfo info, TlsFrameTransport transport) = await PeerConnector.ConnectAsync(
+            address.ToString(), Endpoint.TcpPort, agentKey, _env.Identity, _env.ProbeId, ProbeApp.Version, 8 << 20, _env.Time, ct);
+        var remote = (IPEndPoint)info.Remote!;
+        var data = new DataSession(info.Credentials.SessionIdValue, info.Credentials.Key);
+        using var udp = new UdpMux(new SingleSession(data), null, _env.Clock, new DataPlaneCounters(), _env.Log, remote.AddressFamily == AddressFamily.InterNetworkV6 ? IPAddress.IPv6Any : IPAddress.Any, []);
+        udp.Bind();
+        var outbox = new Outbox(8 << 20);
+        var clock = new ClockEstimator();
+        var shipper = new FragmentShipper(outbox, new NodeId(Endpoint.AgentId), new ProbeId(_env.ProbeId), clock, _env.Time);
+        var blockEnv = new BlockEnv
+        {
+            Clock = _env.Clock,
+            Log = _env.Log,
+            Session = () => data,
+            Udp = udp,
+            PeerAddress = remote.Address,
+            PeerSpkiSha256 = NodeIdentity.Fingerprint(agentKey),
+            KnockKey = Knock.DeriveKey(_env.Identity, agentKey),
+            Fragments = shipper,
+            MaxPacketsPerSecond = _env.Limits.MaxTotalPps,
+            MaxBulkMbps = _env.Limits.MaxTestMbps,
+        };
+        var session = new ConnectionSession(blockEnv, clock, outbox, _env);
+        var rpc = new RpcSession<ConnectionSession>(transport, s_router, session, _env.Time);
+        rpc.OnBinary = (_, _) => ValueTask.CompletedTask;
+
+        using CancellationTokenSource cts = CancellationTokenSource.CreateLinkedTokenSource(ct);
+        Task run = rpc.RunAsync(cts.Token);
+        Task receive = udp.RunAsync(cts.Token);
+        Task pump = PumpAsync(rpc, outbox, cts.Token);
+        try
+        {
+            // 先同步时钟（块头里要带偏移）；Agent 随后推送线路任务集
+            await SyncClockAsync(rpc, clock, rounds: 5, cts.Token);
+            backoff.Reset();
+            _session = session;
+            _env.Log.Info("agent session up", ("agent", Endpoint.AgentId), ("offset_us", clock.OffsetUs));
+            Task maintenance = MaintainAsync(rpc, session, clock, cts.Token);
+            Task finished = await Task.WhenAny(run, receive, pump, maintenance);
+            await cts.CancelAsync();
+            await Task.WhenAll(Quiet(run), Quiet(receive), Quiet(pump), Quiet(maintenance));
+            return finished == run ? "closed by the agent" : "session task failed";
+        }
+        finally
+        {
+            _session = null;
+            await cts.CancelAsync();
+            await session.StopAllAsync();
+            await Quiet(run);
+            await Quiet(receive);
+            await Quiet(pump);
+            await rpc.DisposeAsync();
+        }
+    }
+
+    private static async Task Quiet(Task task)
+    {
+        try
+        {
+            await task;
+        }
+        catch (Exception ex) when (ex is OperationCanceledException or IOException or ObjectDisposedException or System.Net.Sockets.SocketException or InvalidOperationException)
+        {
+            // 会话结束时的正常噪声
+        }
+    }
+
+    private static async Task PumpAsync(RpcSession<ConnectionSession> rpc, Outbox outbox, CancellationToken ct)
+    {
+        while (!ct.IsCancellationRequested)
+        {
+            byte[] frame = await outbox.TakeNextAsync(2 << 20, ct);
+            await rpc.SendBinaryAsync(frame, ct);
+        }
+    }
+
+    private async Task SyncClockAsync(RpcSession<ConnectionSession> rpc, ClockEstimator clock, int rounds, CancellationToken ct)
+    {
+        for (int i = 0; i < rounds; i++)
+        {
+            long sent = _env.Clock.NowUs();
+            Result<ClockResult> reply = await rpc.RequestAsync(MoShiroMethods.Clock, new ClockParams { ProbeMonoUs = sent }, MoShiroJsonContext.Default.ClockParams, MoShiroJsonContext.Default.ClockResult, ct);
+            if (reply.TryGet(out ClockResult? value, out _))
+            {
+                clock.Add(sent, value.AgentMonoUs, _env.Clock.NowUs());
+            }
+        }
+    }
+
+    /// <summary>切块（每 10 秒）和时钟同步（每 30 秒）。</summary>
+    private async Task MaintainAsync(RpcSession<ConnectionSession> rpc, ConnectionSession session, ClockEstimator clock, CancellationToken ct)
+    {
+        using var flush = new PeriodicTimer(s_flushInterval, _env.Time);
+        long lastClockUs = _env.Clock.NowUs();
+        while (await flush.WaitForNextTickAsync(ct))
+        {
+            session.FlushLineBlocks();
+            if (_env.Clock.NowUs() - lastClockUs >= s_clockInterval.TotalMicroseconds)
+            {
+                lastClockUs = _env.Clock.NowUs();
+                await SyncClockAsync(rpc, clock, rounds: 1, ct);
+            }
+        }
+    }
+
+    private static RpcRouter<ConnectionSession> BuildRouter() =>
+        new RpcRouter<ConnectionSession>(RpcSide.Dialer, MoShiroMethodTable.Table)
+            .On(
+                FleetMethods.StateSync,
+                FleetRpcContext.Default.StateSyncParams,
+                FleetRpcContext.Default.StateSyncResult,
+                async (session, sync, _) =>
+                {
+                    var result = new StateSyncResult { AppliedVersion = sync.Version };
+                    if (sync.Type == LineTaskSet.StateType)
+                    {
+                        LineTaskSet? set = sync.Doc.Deserialize(MoShiroJsonContext.Default.LineTaskSet);
+                        if (set is null)
+                        {
+                            return Result<StateSyncResult>.Fail(FleetErrors.InvalidParams, "line_task_set could not be parsed");
+                        }
+
+                        result.Rejected.AddRange(await session.ApplyAsync(set));
+                    }
+
+                    return Result<StateSyncResult>.Ok(result);
+                })
+            .On(
+                MoShiroMethods.TestPhase,
+                MoShiroJsonContext.Default.TestPhaseStartParams,
+                MoShiroJsonContext.Default.TestPhaseStartResult,
+                (session, start, _) => ValueTask.FromResult(Result<TestPhaseStartResult>.Ok(new TestPhaseStartResult { Accepted = session.StartPhase(start) })))
+            .On(
+                MoShiroMethods.TestPhaseEnd,
+                MoShiroJsonContext.Default.TestPhaseEndParams,
+                MoShiroJsonContext.Default.TestPhaseEndResult,
+                async (session, end, ct) => Result<TestPhaseEndResult>.Ok(new TestPhaseEndResult { Fragments = await session.EndPhaseAsync(end, ct) }))
+            .OnNotify(
+                MoShiroMethods.TestAbort,
+                MoShiroJsonContext.Default.TestAbortParams,
+                async (session, abort, _) => await session.AbortAsync(abort.TestId))
+            .OnNotify(
+                FleetMethods.OutboxAck,
+                FleetRpcContext.Default.OutboxAckParams,
+                (session, ack, _) =>
+                {
+                    session.Outbox.Acknowledge(ack.Ids.Concat(ack.Duplicates).Select(Guid.Parse));
+                    return ValueTask.CompletedTask;
+                });
+}
+
+/// <summary>一次会话内的运行状态：线路任务的积木、按需测试的阶段。方法由路由器在单个入站队列里顺序调用。</summary>
+public sealed class ConnectionSession
+{
+    private const int MaxBlocks = 16;
+
+    private readonly object _gate = new();
+    private readonly Dictionary<string, RunningBlock> _lineBlocks = [];
+    private readonly ConcurrentDictionary<uint, PhaseState> _phases = new();
+
+    public ConnectionSession(BlockEnv blockEnv, ClockEstimator clock, Outbox outbox, ConnectionEnv env)
+    {
+        BlockEnv = blockEnv;
+        Clock = clock;
+        Outbox = outbox;
+        Env = env;
+    }
+
+    public BlockEnv BlockEnv { get; }
+
+    public ClockEstimator Clock { get; }
+
+    public Outbox Outbox { get; }
+
+    public ConnectionEnv Env { get; }
+
+    public int RoleCount
+    {
+        get
+        {
+            lock (_gate)
+            {
+                return _lineBlocks.Count;
+            }
+        }
+    }
+
+    /// <summary>调和线路任务：参数没有变化的积木继续运行，其余启动或停止。返回被拒绝的任务（不认识的 Kind、参数非法、超过上限）。</summary>
+    public async Task<List<RejectedItem>> ApplyAsync(LineTaskSet set)
+    {
+        var rejected = new List<RejectedItem>();
+        var wanted = new Dictionary<string, BlockSpec>();
+        foreach (LineTask task in set.Tasks)
+        {
+            BlockSpec? spec = BlockCatalog.ProbeKinds.Contains(task.Kind)
+                ? BlockSpec.Create(task.Kind, task.Params, task.StreamId, task.Port, new FragmentScope(task.Id, null, null))
+                : null;
+            if (spec is null)
+            {
+                rejected.Add(new RejectedItem { Id = task.Id.Value, Reason = BlockCatalog.ProbeKinds.Contains(task.Kind) ? "invalid_params" : "unsupported" });
+            }
+            else if (wanted.Count >= MaxBlocks)
+            {
+                rejected.Add(new RejectedItem { Id = task.Id.Value, Reason = "budget_exceeded" });
+            }
+            else
+            {
+                wanted[task.Id.Value] = spec;
+            }
+        }
+
+        var stopped = new List<RunningBlock>();
+        lock (_gate)
+        {
+            foreach ((string id, RunningBlock running) in _lineBlocks.ToList())
+            {
+                if (!wanted.TryGetValue(id, out BlockSpec? spec) || spec.Signature != running.Signature)
+                {
+                    stopped.Add(running);
+                    _lineBlocks.Remove(id);
+                }
+            }
+
+            foreach ((string id, BlockSpec spec) in wanted)
+            {
+                if (!_lineBlocks.ContainsKey(id))
+                {
+                    _lineBlocks[id] = RunningBlock.Start(BlockCatalog.Find(spec.KindId)!.Dialer(BlockEnv, spec), spec.Signature);
+                }
+            }
+        }
+
+        // 被停止的积木把已经稳定的行交出去（刚发出、还在等回复的行不交：它们不是丢失，只是没有等到）
+        foreach (RunningBlock running in stopped)
+        {
+            await running.StopAsync();
+            running.Block.Flush(FlushMode.Settled);
+        }
+
+        return rejected;
+    }
+
+    public void FlushLineBlocks()
+    {
+        List<RunningBlock> blocks;
+        lock (_gate)
+        {
+            blocks = [.. _lineBlocks.Values];
+        }
+
+        foreach (RunningBlock running in blocks)
+        {
+            running.Block.Flush(FlushMode.Settled);
+        }
+    }
+
+    /// <summary>按需测试的一个阶段：起对应的拨号方积木（单向数据流是接收端）。超过本地上限（或不认识）则拒绝。</summary>
+    public bool StartPhase(TestPhaseStartParams start)
+    {
+        string? direction = start.Params.TryGetValue("direction", out System.Text.Json.JsonElement d) ? d.GetString() : null;
+        if (_phases.ContainsKey(start.StreamId) || !BlockCatalog.ProbeKinds.Contains(start.Kind) || BlockCatalog.Find(start.Kind)!.Supports(direction) is false)
+        {
+            return false;
+        }
+
+        if (start.Kind == "udp_train")
+        {
+            double mbps = start.Params["mbps"].GetDouble();
+            long size = start.Params["size"].GetInt64();
+            long pps = (long)(mbps * 1_000_000 / 8 / Math.Max(1, size));
+            if (mbps > Env.Limits.MaxTestMbps || pps > Env.Limits.MaxTotalPps)
+            {
+                return false;
+            }
+        }
+
+        BlockSpec? spec = BlockSpec.Create(start.Kind, start.Params, start.StreamId, start.Port, new FragmentScope(null, start.TestId, start.Phase));
+        if (spec is null)
+        {
+            return false;
+        }
+
+        _phases[start.StreamId] = new PhaseState(start.TestId, start.Kind, RunningBlock.Start(BlockCatalog.Find(start.Kind)!.Dialer(BlockEnv, spec), spec.Signature));
+        return true;
+    }
+
+    /// <summary>阶段结束：收尾、交出列块，返回块数。</summary>
+    public async Task<int> EndPhaseAsync(TestPhaseEndParams end, CancellationToken ct)
+    {
+        if (!_phases.TryRemove(end.StreamId, out PhaseState? phase))
+        {
+            return 0;
+        }
+
+        // 最后一批包（单向数据流）、最后几个回显的回复还在途中
+        await Task.Delay(TimeSpan.FromMilliseconds(phase.Kind == "udp_train" ? 200 : 300), BlockEnv.Time, ct);
+        await phase.Running.StopAsync();
+        return phase.Running.Block.Flush(FlushMode.All, end.SentCount);
+    }
+
+    public async Task AbortAsync(TestId test)
+    {
+        foreach ((uint stream, PhaseState phase) in _phases.ToList())
+        {
+            if (phase.Test == test && _phases.TryRemove(stream, out _))
+            {
+                await phase.Running.StopAsync();
+            }
+        }
+    }
+
+    public async Task StopAllAsync()
+    {
+        List<RunningBlock> blocks;
+        lock (_gate)
+        {
+            blocks = [.. _lineBlocks.Values];
+            _lineBlocks.Clear();
+        }
+
+        blocks.AddRange(_phases.Values.Select(p => p.Running));
+        _phases.Clear();
+        foreach (RunningBlock running in blocks)
+        {
+            await running.StopAsync();
+        }
+    }
+
+    private sealed record PhaseState(TestId Test, string Kind, RunningBlock Running);
+}

backend/probe/src/MoShiro.Probe/Link/AgentDirectory.cs

@@ -0,0 +1,115 @@
+using System.Text.Json;
+using Fleet.Core;
+using Fleet.Hosting;
+using MoShiro.Protocol;
+
+namespace MoShiro.Probe;
+
+/// <summary>
+/// Probe 要连接的 Agent 列表：由面板作为期望状态（<c>probe_agents</c>）直接下发；面板连不上时用上一次保存的 <c>agents.json</c>。
+/// 列表变化时用 <see cref="ConfigStore"/> 写回 <c>agents.json</c>（这是配置，不是测量数据；只在变化时写，保护路由器的闪存）。
+/// </summary>
+public sealed class AgentDirectory
+{
+    public const string FileName = "agents.json";
+
+    private readonly object _gate = new();
+    private readonly ConfigStore _store;
+    private readonly INodeLog _log;
+    private readonly TrustAnchor _trust;
+    private List<KnownAgent> _agents = [];
+    private long _version;
+
+    public AgentDirectory(ConfigStore store, string configDirectory, INodeLog log, TrustAnchor? trust = null)
+    {
+        _trust = trust ?? TrustAnchor.None;
+        _store = store;
+        _log = log;
+        string cache = Path.Combine(configDirectory, FileName);
+        if (File.Exists(cache))
+        {
+            try
+            {
+                AgentsCache saved = NodeConfig.Load(cache, ProbeConfigContext.Default.AgentsCache);
+                _agents = saved.Agents;
+                _version = saved.Version;
+            }
+            catch (Exception ex) when (ex is IOException or InvalidDataException or JsonException or InvalidOperationException)
+            {
+                _log.Warn("agents.json is unreadable, waiting for the panel", ("err", ex.Message));
+            }
+        }
+    }
+
+    /// <summary>列表变化（调和连接）。</summary>
+    public event Action? Changed;
+
+    public long Version
+    {
+        get
+        {
+            lock (_gate)
+            {
+                return _version;
+            }
+        }
+    }
+
+    public IReadOnlyList<KnownAgent> Agents
+    {
+        get
+        {
+            lock (_gate)
+            {
+                return [.. _agents];
+            }
+        }
+    }
+
+    /// <summary>应用面板下发的列表。版本不比当前新的忽略（乱序到达的旧推送）。</summary>
+    public void Apply(ProbeAgentList result)
+    {
+        var next = new List<KnownAgent>();
+        _trust.Update(result.TrustChain);
+        foreach (AgentEndpoint agent in result.Agents)
+        {
+            if (!string.IsNullOrWhiteSpace(agent.Host))
+            {
+                next.Add(new KnownAgent { AgentId = agent.AgentId.Value, Host = agent.Host, TcpPort = agent.TcpPort, PublicKey = agent.PublicKey });
+            }
+
+            if (!string.IsNullOrWhiteSpace(agent.Ipv6Host))
+            {
+                next.Add(new KnownAgent { AgentId = agent.AgentId.Value, Host = agent.Ipv6Host, TcpPort = agent.TcpPort, PublicKey = agent.PublicKey, IpVersion = IpVersions.V6 });
+            }
+        }
+        lock (_gate)
+        {
+            if (result.Version < _version || (result.Version == _version && Same(next, _agents)))
+            {
+                return;
+            }
+
+            _version = result.Version;
+            _agents = next;
+        }
+
+        try
+        {
+            _store.Write(FileName, JsonSerializer.SerializeToUtf8Bytes(new AgentsCache { Version = result.Version, Agents = next }, ProbeConfigContext.Default.AgentsCache));
+        }
+        catch (Exception ex) when (ex is IOException or UnauthorizedAccessException)
+        {
+            _log.Warn("could not save agents.json", ("err", ex.Message));
+        }
+
+        _log.Info("agent list updated", ("version", result.Version), ("agents", next.Count));
+        Changed?.Invoke();
+    }
+
+    private static bool Same(List<KnownAgent> a, List<KnownAgent> b) =>
+        a.Count == b.Count && a.Zip(b).All(p => Equal(p.First, p.Second));
+
+    public static bool Equal(KnownAgent a, KnownAgent b) =>
+        a.AgentId == b.AgentId && a.IpVersion == b.IpVersion && a.Host == b.Host && a.TcpPort == b.TcpPort && a.PublicKey == b.PublicKey;
+}

backend/probe/src/MoShiro.Probe/Link/Support.cs

@@ -4,11 +4,11 @@ using Fleet.Link;
 using MoShiro.Measurement;
 using MoShiro.Protocol;
 
-namespace MoShiro.Agent;
+namespace MoShiro.Probe;
 
 /// <summary>
-/// 估计监听方单调时钟 − 拨号方单调时钟：每次 <c>moshiro.clock</c> 往返取中点，在最近 8 次里取往返最短的一次
-/// （往返越短，对称性假设的误差越小）。拨号方的列块头里带着这个偏移，Backend 据此对齐两端的时间戳。
+/// 估计 Agent 单调时钟 − Probe 单调时钟：每次 <c>moshiro.clock</c> 往返取中点，在最近 8 次里取往返最短的一次
+/// （往返越短，对称性假设的误差越小）。Probe 的列块头里带着这个偏移，Backend 据此对齐两端的时间戳。
 /// </summary>
 public sealed class ClockEstimator
 {
@@ -21,11 +21,11 @@ public sealed class ClockEstimator
 
     public bool HasEstimate { get; private set; }
 
-    public void Add(long sendUs, long listenerUs, long receiveUs)
+    public void Add(long sendUs, long agentUs, long receiveUs)
     {
         lock (_gate)
         {
-            _samples.Enqueue((receiveUs - sendUs, listenerUs - ((sendUs + receiveUs) / 2)));
+            _samples.Enqueue((receiveUs - sendUs, agentUs - ((sendUs + receiveUs) / 2)));
             while (_samples.Count > Window)
             {
                 _samples.Dequeue();
@@ -38,22 +38,22 @@ public sealed class ClockEstimator
 }
 
 /// <summary>
-/// 把拨号方一侧的行交给监听方：补全块头 → 编码 → 放进会话的发件箱（通道 17，至少一次，有界）。
+/// 把 Probe 一侧的行交给这个 Agent：补全块头 → 编码 → 放进会话的发件箱（通道 17，至少一次，有界）。
 /// 发件箱的内容在连接断开重连之后重发；发件箱满了丢弃最旧的。
 /// </summary>
 public sealed class FragmentShipper : IFragmentSink
 {
     private readonly Outbox _outbox;
-    private readonly NodeId _listener;
-    private readonly NodeId _dialer;
+    private readonly NodeId _agent;
+    private readonly ProbeId _probe;
     private readonly ClockEstimator _clock;
     private readonly TimeProvider _time;
 
-    public FragmentShipper(Outbox outbox, NodeId listener, NodeId dialer, ClockEstimator clock, TimeProvider time)
+    public FragmentShipper(Outbox outbox, NodeId agent, ProbeId probe, ClockEstimator clock, TimeProvider time)
     {
         _outbox = outbox;
-        _listener = listener;
-        _dialer = dialer;
+        _agent = agent;
+        _probe = probe;
         _clock = clock;
         _time = time;
     }
@@ -67,12 +67,12 @@ public sealed class FragmentShipper : IFragmentSink
             TaskId = scope.Task,
             TestId = scope.Test,
             Phase = scope.Phase,
-            AgentId = _listener,
-            DialerId = _dialer,
+            AgentId = _agent,
+            ProbeId = _probe,
             Port = port,
             StartUtcMs = startUtcMs,
             DurationMs = Math.Max(1, durationMs),
-            Clock = new ChunkClock { DialerOffsetUs = _clock.OffsetUs },
+            Clock = new ChunkClock { ProbeOffsetUs = _clock.OffsetUs },
         };
         if (parameters is not null)
         {

backend/probe/src/MoShiro.Probe/MoShiro.Probe.csproj

@@ -0,0 +1,26 @@
+<Project>
+  <Import Project="Sdk.props" Sdk="Microsoft.NET.Sdk" />
+  <Import Project="Sdk.props" Sdk="StuDev.AotAnywhere"
+          Condition="'$(UseAotAnywhere)' == 'true'" />
+  <PropertyGroup>
+    <OutputType>Exe</OutputType>
+    <AssemblyName>moshiro-probe</AssemblyName>
+    <RootNamespace>MoShiro.Probe</RootNamespace>
+    <IsAotCompatible>true</IsAotCompatible>
+    <PublishAot>true</PublishAot>
+    <PublishTrimmed>true</PublishTrimmed>
+    <TrimMode>full</TrimMode>
+    <StripSymbols>true</StripSymbols>
+    <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
+    <MetadataUpdaterSupport>false</MetadataUpdaterSupport>
+  </PropertyGroup>
+  <ItemGroup>
+    <InternalsVisibleTo Include="MoShiro.Backend.Tests" />
+    <ProjectReference Include="..\..\..\fleet\src\Fleet.Link\Fleet.Link.csproj" />
+    <ProjectReference Include="..\..\..\fleet\src\Fleet.Hosting\Fleet.Hosting.csproj" />
+    <ProjectReference Include="..\..\..\shared\MoShiro.Measurement\MoShiro.Measurement.csproj" />
+  </ItemGroup>
+  <Import Project="Sdk.targets" Sdk="Microsoft.NET.Sdk" />
+  <Import Project="Sdk.targets" Sdk="StuDev.AotAnywhere"
+          Condition="'$(UseAotAnywhere)' == 'true'" />
+</Project>