Significantly speed up creating backups with isal via zlib-fast

isal is a drop in replacement for zlib with the
cavet that the compression level mappings are different.
zlib-fast is a tiny piece of middleware to convert
the standard zlib compression levels to isal compression
levels to allow for drop-in replacement

https://github.com/bdraco/zlib-fast/releases/tag/v0.1.0
https://github.com/pycompression/python-isal

Compression for backups is ~5x faster than the baseline

https://github.com/powturbo/TurboBench/issues/43

.devcontainer/devcontainer.json

@@ -7,34 +7,32 @@
   "appPort": ["9123:8123", "7357:4357"],
   "postCreateCommand": "bash devcontainer_bootstrap",
   "runArgs": ["-e", "GIT_EDITOR=code --wait", "--privileged"],
-  "extensions": [
-    "ms-python.python",
-    "ms-python.vscode-pylance",
-    "visualstudioexptteam.vscodeintellicode",
-    "esbenp.prettier-vscode"
-  ],
-  "mounts": ["type=volume,target=/var/lib/docker"],
-  "settings": {
-    "terminal.integrated.profiles.linux": {
-      "zsh": {
-        "path": "/usr/bin/zsh"
+  "customizations": {
+    "vscode": {
+      "extensions": [
+        "ms-python.python",
+        "ms-python.pylint",
+        "ms-python.vscode-pylance",
+        "visualstudioexptteam.vscodeintellicode",
+        "esbenp.prettier-vscode"
+      ],
+      "settings": {
+        "terminal.integrated.profiles.linux": {
+          "zsh": {
+            "path": "/usr/bin/zsh"
+          }
+        },
+        "terminal.integrated.defaultProfile.linux": "zsh",
+        "editor.formatOnPaste": false,
+        "editor.formatOnSave": true,
+        "editor.formatOnType": true,
+        "files.trimTrailingWhitespace": true,
+        "python.pythonPath": "/usr/local/bin/python3",
+        "python.formatting.provider": "black",
+        "python.formatting.blackArgs": ["--target-version", "py312"],
+        "python.formatting.blackPath": "/usr/local/bin/black"
       }
-    },
-    "terminal.integrated.defaultProfile.linux": "zsh",
-    "editor.formatOnPaste": false,
-    "editor.formatOnSave": true,
-    "editor.formatOnType": true,
-    "files.trimTrailingWhitespace": true,
-    "python.pythonPath": "/usr/local/bin/python3",
-    "python.linting.pylintEnabled": true,
-    "python.linting.enabled": true,
-    "python.formatting.provider": "black",
-    "python.formatting.blackArgs": ["--target-version", "py310"],
-    "python.formatting.blackPath": "/usr/local/bin/black",
-    "python.linting.banditPath": "/usr/local/bin/bandit",
-    "python.linting.flake8Path": "/usr/local/bin/flake8",
-    "python.linting.mypyPath": "/usr/local/bin/mypy",
-    "python.linting.pylintPath": "/usr/local/bin/pylint",
-    "python.linting.pydocstylePath": "/usr/local/bin/pydocstyle"
-  }
+    }
+  },
+  "mounts": ["type=volume,target=/var/lib/docker"]
 }

.github/workflows/builder.yml

@@ -33,7 +33,7 @@ on:
       - setup.py
 
 env:
-  DEFAULT_PYTHON: "3.11"
+  DEFAULT_PYTHON: "3.12"
   BUILD_NAME: supervisor
   BUILD_TYPE: supervisor
 
@@ -53,7 +53,7 @@ jobs:
       requirements: ${{ steps.requirements.outputs.changed }}
     steps:
       - name: Checkout the repository
-        uses: actions/checkout@v4.0.0
+        uses: actions/checkout@v4.1.1
         with:
           fetch-depth: 0
 
@@ -70,13 +70,13 @@ jobs:
       - name: Get changed files
         id: changed_files
         if: steps.version.outputs.publish == 'false'
-        uses: jitterbit/get-changed-files@v1
+        uses: masesgroup/retrieve-changed-files@v3.0.0
 
       - name: Check if requirements files changed
         id: requirements
         run: |
-          if [[ "${{ steps.changed_files.outputs.all }}" =~ (requirements.txt|build.json) ]]; then
-            echo "::set-output name=changed::true"
+          if [[ "${{ steps.changed_files.outputs.all }}" =~ (requirements.txt|build.yaml) ]]; then
+            echo "changed=true" >> "$GITHUB_OUTPUT"
           fi
 
   build:
@@ -92,7 +92,7 @@ jobs:
         arch: ${{ fromJson(needs.init.outputs.architectures) }}
     steps:
       - name: Checkout the repository
-        uses: actions/checkout@v4.0.0
+        uses: actions/checkout@v4.1.1
         with:
           fetch-depth: 0
 
@@ -106,13 +106,13 @@ jobs:
 
       - name: Build wheels
         if: needs.init.outputs.requirements == 'true'
-        uses: home-assistant/wheels@2023.04.0
+        uses: home-assistant/wheels@2024.01.0
         with:
-          abi: cp311
+          abi: cp312
           tag: musllinux_1_2
           arch: ${{ matrix.arch }}
           wheels-key: ${{ secrets.WHEELS_KEY }}
-          apk: "libffi-dev;openssl-dev"
+          apk: "libffi-dev;openssl-dev;yaml-dev"
           skip-binary: aiohttp
           env-file: true
           requirements: "requirements.txt"
@@ -125,20 +125,20 @@ jobs:
 
       - name: Set up Python ${{ env.DEFAULT_PYTHON }}
         if: needs.init.outputs.publish == 'true'
-        uses: actions/setup-python@v4.7.0
+        uses: actions/setup-python@v5.0.0
         with:
           python-version: ${{ env.DEFAULT_PYTHON }}
 
       - name: Install Cosign
         if: needs.init.outputs.publish == 'true'
-        uses: sigstore/cosign-installer@v3.1.2
+        uses: sigstore/cosign-installer@v3.3.0
         with:
           cosign-release: "v2.0.2"
 
       - name: Install dirhash and calc hash
         if: needs.init.outputs.publish == 'true'
         run: |
-          pip3 install dirhash
+          pip3 install setuptools dirhash
           dir_hash="$(dirhash "${{ github.workspace }}/supervisor" -a sha256 --match "*.py")"
           echo "${dir_hash}" > rootfs/supervisor.sha256
 
@@ -149,7 +149,7 @@ jobs:
 
       - name: Login to GitHub Container Registry
         if: needs.init.outputs.publish == 'true'
-        uses: docker/login-action@v2.2.0
+        uses: docker/login-action@v3.0.0
         with:
           registry: ghcr.io
           username: ${{ github.repository_owner }}
@@ -160,7 +160,7 @@ jobs:
         run: echo "BUILD_ARGS=--test" >> $GITHUB_ENV
 
       - name: Build supervisor
-        uses: home-assistant/builder@2023.08.0
+        uses: home-assistant/builder@2024.01.0
         with:
           args: |
             $BUILD_ARGS \
@@ -178,7 +178,7 @@ jobs:
     steps:
       - name: Checkout the repository
         if: needs.init.outputs.publish == 'true'
-        uses: actions/checkout@v4.0.0
+        uses: actions/checkout@v4.1.1
 
       - name: Initialize git
         if: needs.init.outputs.publish == 'true'
@@ -203,11 +203,11 @@ jobs:
     timeout-minutes: 60
     steps:
       - name: Checkout the repository
-        uses: actions/checkout@v4.0.0
+        uses: actions/checkout@v4.1.1
 
       - name: Build the Supervisor
         if: needs.init.outputs.publish != 'true'
-        uses: home-assistant/builder@2023.08.0
+        uses: home-assistant/builder@2024.01.0
         with:
           args: |
             --test \
@@ -324,7 +324,7 @@ jobs:
           if [ "$(echo $test | jq -r '.result')" != "ok" ]; then
             exit 1
           fi
-          echo "::set-output name=slug::$(echo $test | jq -r '.data.slug')"
+          echo "slug=$(echo $test | jq -r '.data.slug')" >> "$GITHUB_OUTPUT"
 
       - name: Uninstall SSH add-on
         run: |

.github/workflows/ci.yaml

@@ -8,8 +8,8 @@ on:
   pull_request: ~
 
 env:
-  DEFAULT_PYTHON: "3.11"
-  PRE_COMMIT_HOME: ~/.cache/pre-commit
+  DEFAULT_PYTHON: "3.12"
+  PRE_COMMIT_CACHE: ~/.cache/pre-commit
 
 concurrency:
   group: "${{ github.workflow }}-${{ github.ref }}"
@@ -25,15 +25,15 @@ jobs:
     name: Prepare Python dependencies
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v4.0.0
+        uses: actions/checkout@v4.1.1
       - name: Set up Python
         id: python
-        uses: actions/setup-python@v4.7.0
+        uses: actions/setup-python@v5.0.0
         with:
           python-version: ${{ env.DEFAULT_PYTHON }}
       - name: Restore Python virtual environment
         id: cache-venv
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
           path: venv
           key: |
@@ -47,9 +47,10 @@ jobs:
           pip install -r requirements.txt -r requirements_tests.txt
       - name: Restore pre-commit environment from cache
         id: cache-precommit
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
-          path: ${{ env.PRE_COMMIT_HOME }}
+          path: ${{ env.PRE_COMMIT_CACHE }}
+          lookup-only: true
           key: |
             ${{ runner.os }}-pre-commit-${{ hashFiles('.pre-commit-config.yaml') }}
           restore-keys: |
@@ -66,15 +67,15 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v4.0.0
+        uses: actions/checkout@v4.1.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
-        uses: actions/setup-python@v4.7.0
+        uses: actions/setup-python@v5.0.0
         id: python
         with:
           python-version: ${{ needs.prepare.outputs.python-version }}
       - name: Restore Python virtual environment
         id: cache-venv
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
           path: venv
           key: |
@@ -87,7 +88,7 @@ jobs:
       - name: Run black
         run: |
           . venv/bin/activate
-          black --target-version py38 --check supervisor tests setup.py
+          black --target-version py312 --check supervisor tests setup.py
 
   lint-dockerfile:
     name: Check Dockerfile
@@ -95,7 +96,7 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v4.0.0
+        uses: actions/checkout@v4.1.1
       - name: Register hadolint problem matcher
         run: |
           echo "::add-matcher::.github/workflows/matchers/hadolint.json"
@@ -110,15 +111,15 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v4.0.0
+        uses: actions/checkout@v4.1.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
-        uses: actions/setup-python@v4.7.0
+        uses: actions/setup-python@v5.0.0
         id: python
         with:
           python-version: ${{ needs.prepare.outputs.python-version }}
       - name: Restore Python virtual environment
         id: cache-venv
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
           path: venv
           key: |
@@ -130,9 +131,9 @@ jobs:
           exit 1
       - name: Restore pre-commit environment from cache
         id: cache-precommit
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
-          path: ${{ env.PRE_COMMIT_HOME }}
+          path: ${{ env.PRE_COMMIT_CACHE }}
           key: |
             ${{ runner.os }}-pre-commit-${{ hashFiles('.pre-commit-config.yaml') }}
       - name: Fail job if cache restore failed
@@ -154,15 +155,15 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v4.0.0
+        uses: actions/checkout@v4.1.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
-        uses: actions/setup-python@v4.7.0
+        uses: actions/setup-python@v5.0.0
         id: python
         with:
           python-version: ${{ needs.prepare.outputs.python-version }}
       - name: Restore Python virtual environment
         id: cache-venv
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
           path: venv
           key: |
@@ -186,15 +187,15 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v4.0.0
+        uses: actions/checkout@v4.1.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
-        uses: actions/setup-python@v4.7.0
+        uses: actions/setup-python@v5.0.0
         id: python
         with:
           python-version: ${{ needs.prepare.outputs.python-version }}
       - name: Restore Python virtual environment
         id: cache-venv
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
           path: venv
           key: |
@@ -206,9 +207,9 @@ jobs:
           exit 1
       - name: Restore pre-commit environment from cache
         id: cache-precommit
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
-          path: ${{ env.PRE_COMMIT_HOME }}
+          path: ${{ env.PRE_COMMIT_CACHE }}
           key: |
             ${{ runner.os }}-pre-commit-${{ hashFiles('.pre-commit-config.yaml') }}
       - name: Fail job if cache restore failed
@@ -227,15 +228,15 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v4.0.0
+        uses: actions/checkout@v4.1.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
-        uses: actions/setup-python@v4.7.0
+        uses: actions/setup-python@v5.0.0
         id: python
         with:
           python-version: ${{ needs.prepare.outputs.python-version }}
       - name: Restore Python virtual environment
         id: cache-venv
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
           path: venv
           key: |
@@ -247,9 +248,9 @@ jobs:
           exit 1
       - name: Restore pre-commit environment from cache
         id: cache-precommit
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
-          path: ${{ env.PRE_COMMIT_HOME }}
+          path: ${{ env.PRE_COMMIT_CACHE }}
           key: |
             ${{ runner.os }}-pre-commit-${{ hashFiles('.pre-commit-config.yaml') }}
       - name: Fail job if cache restore failed
@@ -271,15 +272,15 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v4.0.0
+        uses: actions/checkout@v4.1.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
-        uses: actions/setup-python@v4.7.0
+        uses: actions/setup-python@v5.0.0
         id: python
         with:
           python-version: ${{ needs.prepare.outputs.python-version }}
       - name: Restore Python virtual environment
         id: cache-venv
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
           path: venv
           key: |
@@ -303,15 +304,15 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v4.0.0
+        uses: actions/checkout@v4.1.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
-        uses: actions/setup-python@v4.7.0
+        uses: actions/setup-python@v5.0.0
         id: python
         with:
           python-version: ${{ needs.prepare.outputs.python-version }}
       - name: Restore Python virtual environment
         id: cache-venv
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
           path: venv
           key: |
@@ -323,9 +324,9 @@ jobs:
           exit 1
       - name: Restore pre-commit environment from cache
         id: cache-precommit
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
-          path: ${{ env.PRE_COMMIT_HOME }}
+          path: ${{ env.PRE_COMMIT_CACHE }}
           key: |
             ${{ runner.os }}-pre-commit-${{ hashFiles('.pre-commit-config.yaml') }}
       - name: Fail job if cache restore failed
@@ -344,19 +345,19 @@ jobs:
     name: Run tests Python ${{ needs.prepare.outputs.python-version }}
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v4.0.0
+        uses: actions/checkout@v4.1.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
-        uses: actions/setup-python@v4.7.0
+        uses: actions/setup-python@v5.0.0
         id: python
         with:
           python-version: ${{ needs.prepare.outputs.python-version }}
       - name: Install Cosign
-        uses: sigstore/cosign-installer@v3.1.2
+        uses: sigstore/cosign-installer@v3.3.0
         with:
           cosign-release: "v2.0.2"
       - name: Restore Python virtual environment
         id: cache-venv
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
           path: venv
           key: |
@@ -391,7 +392,7 @@ jobs:
             -o console_output_style=count \
             tests
       - name: Upload coverage artifact
-        uses: actions/upload-artifact@v3.1.2
+        uses: actions/upload-artifact@v4.0.0
         with:
           name: coverage-${{ matrix.python-version }}
           path: .coverage
@@ -402,15 +403,15 @@ jobs:
     needs: ["pytest", "prepare"]
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v4.0.0
+        uses: actions/checkout@v4.1.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
-        uses: actions/setup-python@v4.7.0
+        uses: actions/setup-python@v5.0.0
         id: python
         with:
           python-version: ${{ needs.prepare.outputs.python-version }}
       - name: Restore Python virtual environment
         id: cache-venv
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
           path: venv
           key: |
@@ -421,7 +422,7 @@ jobs:
           echo "Failed to restore Python virtual environment from cache"
           exit 1
       - name: Download all coverage artifacts
-        uses: actions/download-artifact@v3
+        uses: actions/download-artifact@v4.1.1
       - name: Combine coverage results
         run: |
           . venv/bin/activate

.github/workflows/lock.yml

@@ -9,7 +9,7 @@ jobs:
   lock:
     runs-on: ubuntu-latest
     steps:
-      - uses: dessant/lock-threads@v4.0.1
+      - uses: dessant/lock-threads@v5.0.1
         with:
           github-token: ${{ github.token }}
           issue-inactive-days: "30"

.github/workflows/release-drafter.yml

@@ -11,7 +11,7 @@ jobs:
     name: Release Drafter
     steps:
       - name: Checkout the repository
-        uses: actions/checkout@v4.0.0
+        uses: actions/checkout@v4.1.1
         with:
           fetch-depth: 0
 
@@ -33,10 +33,10 @@ jobs:
 
           echo Current version:    $latest
           echo New target version: $datepre.$newpost
-          echo "::set-output name=version::$datepre.$newpost"
+          echo "version=$datepre.$newpost" >> "$GITHUB_OUTPUT"
 
       - name: Run Release Drafter
-        uses: release-drafter/release-drafter@v5.24.0
+        uses: release-drafter/release-drafter@v5.25.0
         with:
           tag: ${{ steps.version.outputs.version }}
           name: ${{ steps.version.outputs.version }}

.github/workflows/sentry.yaml

@@ -10,9 +10,9 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v4.0.0
+        uses: actions/checkout@v4.1.1
       - name: Sentry Release
-        uses: getsentry/action-release@v1.4.1
+        uses: getsentry/action-release@v1.6.0
         env:
           SENTRY_AUTH_TOKEN: ${{ secrets.SENTRY_AUTH_TOKEN }}
           SENTRY_ORG: ${{ secrets.SENTRY_ORG }}

.github/workflows/stale.yml

@@ -9,7 +9,7 @@ jobs:
   stale:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/stale@v8.0.0
+      - uses: actions/stale@v9.0.0
         with:
           repo-token: ${{ secrets.GITHUB_TOKEN }}
           days-before-stale: 30

.hadolint.yaml

@@ -3,4 +3,5 @@ ignored:
   - DL3006
   - DL3013
   - DL3018
+  - DL3042
   - SC2155

.pre-commit-config.yaml

@@ -1,16 +1,16 @@
 repos:
   - repo: https://github.com/psf/black
-    rev: 23.1.0
+    rev: 23.12.1
     hooks:
       - id: black
         args:
           - --safe
           - --quiet
           - --target-version
-          - py310
+          - py312
         files: ^((supervisor|tests)/.+)?[^/]+\.py$
   - repo: https://github.com/PyCQA/flake8
-    rev: 6.0.0
+    rev: 7.0.0
     hooks:
       - id: flake8
         additional_dependencies:
@@ -18,17 +18,17 @@ repos:
           - pydocstyle==6.3.0
         files: ^(supervisor|script|tests)/.+\.py$
   - repo: https://github.com/pre-commit/pre-commit-hooks
-    rev: v4.3.0
+    rev: v4.5.0
     hooks:
       - id: check-executables-have-shebangs
         stages: [manual]
       - id: check-json
   - repo: https://github.com/PyCQA/isort
-    rev: 5.12.0
+    rev: 5.13.2
     hooks:
       - id: isort
   - repo: https://github.com/asottile/pyupgrade
-    rev: v3.4.0
+    rev: v3.15.0
     hooks:
       - id: pyupgrade
-        args: [--py310-plus]
+        args: [--py312-plus]

Dockerfile

@@ -15,6 +15,7 @@ WORKDIR /usr/src
 RUN \
     set -x \
     && apk add --no-cache \
+        findutils \
         eudev \
         eudev-libs \
         git \
@@ -22,6 +23,7 @@ RUN \
         libpulse \
         musl \
         openssl \
+        yaml \
     \
     && curl -Lso /usr/bin/cosign "https://github.com/home-assistant/cosign/releases/download/${COSIGN_VERSION}/cosign_${BUILD_ARCH}" \
     && chmod a+x /usr/bin/cosign
@@ -30,15 +32,14 @@ RUN \
 COPY requirements.txt .
 RUN \
     export MAKEFLAGS="-j$(nproc)" \
-    && pip3 install --no-cache-dir --no-index --only-binary=:all: --find-links \
-        "https://wheels.home-assistant.io/musllinux/" \
+    && pip3 install --only-binary=:all: \
         -r ./requirements.txt \
     && rm -f requirements.txt
 
 # Install Home Assistant Supervisor
 COPY . supervisor
 RUN \
-    pip3 install --no-cache-dir -e ./supervisor \
+    pip3 install -e ./supervisor \
     && python3 -m compileall ./supervisor/supervisor
 
 

build.yaml

@@ -1,10 +1,10 @@
 image: ghcr.io/home-assistant/{arch}-hassio-supervisor
 build_from:
-  aarch64: ghcr.io/home-assistant/aarch64-base-python:3.11-alpine3.16
-  armhf: ghcr.io/home-assistant/armhf-base-python:3.11-alpine3.16
-  armv7: ghcr.io/home-assistant/armv7-base-python:3.11-alpine3.16
-  amd64: ghcr.io/home-assistant/amd64-base-python:3.11-alpine3.16
-  i386: ghcr.io/home-assistant/i386-base-python:3.11-alpine3.16
+  aarch64: ghcr.io/home-assistant/aarch64-base-python:3.12-alpine3.18
+  armhf: ghcr.io/home-assistant/armhf-base-python:3.12-alpine3.18
+  armv7: ghcr.io/home-assistant/armv7-base-python:3.12-alpine3.18
+  amd64: ghcr.io/home-assistant/amd64-base-python:3.12-alpine3.18
+  i386: ghcr.io/home-assistant/i386-base-python:3.12-alpine3.18
 codenotary:
   signer: notary@home-assistant.io
   base_image: notary@home-assistant.io

pylintrc

@@ -1,45 +0,0 @@
-[MASTER]
-reports=no
-jobs=2
-
-good-names=id,i,j,k,ex,Run,_,fp,T,os
-
-extension-pkg-whitelist=
-  ciso8601
-
-# Reasons disabled:
-# format - handled by black
-# locally-disabled - it spams too much
-# duplicate-code - unavoidable
-# cyclic-import - doesn't test if both import on load
-# abstract-class-not-used - is flaky, should not show up but does
-# unused-argument - generic callbacks and setup methods create a lot of warnings
-# too-many-* - are not enforced for the sake of readability
-# too-few-* - same as too-many-*
-# abstract-method - with intro of async there are always methods missing
-disable=
-  format,
-  abstract-method,
-  cyclic-import,
-  duplicate-code,
-  locally-disabled,
-  no-else-return,
-  not-context-manager,
-  too-few-public-methods,
-  too-many-arguments,
-  too-many-branches,
-  too-many-instance-attributes,
-  too-many-lines,
-  too-many-locals,
-  too-many-public-methods,
-  too-many-return-statements,
-  too-many-statements,
-  unused-argument,
-  consider-using-with
-
-[EXCEPTIONS]
-overgeneral-exceptions=builtins.Exception
-
-
-[TYPECHECK]
-ignored-modules = distutils

pyproject.toml

@@ -0,0 +1,112 @@
+[build-system]
+requires = ["setuptools~=68.0.0", "wheel~=0.40.0"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "Supervisor"
+dynamic = ["version", "dependencies"]
+license = { text = "Apache-2.0" }
+description = "Open-source private cloud os for Home-Assistant based on HassOS"
+readme = "README.md"
+authors = [
+    { name = "The Home Assistant Authors", email = "hello@home-assistant.io" },
+]
+keywords = ["docker", "home-assistant", "api"]
+requires-python = ">=3.12.0"
+
+[project.urls]
+"Homepage" = "https://www.home-assistant.io/"
+"Source Code" = "https://github.com/home-assistant/supervisor"
+"Bug Reports" = "https://github.com/home-assistant/supervisor/issues"
+"Docs: Dev" = "https://developers.home-assistant.io/"
+"Discord" = "https://www.home-assistant.io/join-chat/"
+"Forum" = "https://community.home-assistant.io/"
+
+[tool.setuptools]
+platforms = ["any"]
+zip-safe = false
+include-package-data = true
+
+[tool.setuptools.packages.find]
+include = ["supervisor*"]
+
+[tool.pylint.MAIN]
+py-version = "3.11"
+# Use a conservative default here; 2 should speed up most setups and not hurt
+# any too bad. Override on command line as appropriate.
+jobs = 2
+persistent = false
+extension-pkg-allow-list = ["ciso8601"]
+
+[tool.pylint.BASIC]
+class-const-naming-style = "any"
+good-names = ["id", "i", "j", "k", "ex", "Run", "_", "fp", "T", "os"]
+
+[tool.pylint."MESSAGES CONTROL"]
+# Reasons disabled:
+# format - handled by black
+# abstract-method - with intro of async there are always methods missing
+# cyclic-import - doesn't test if both import on load
+# duplicate-code - unavoidable
+# locally-disabled - it spams too much
+# too-many-* - are not enforced for the sake of readability
+# too-few-* - same as too-many-*
+# unused-argument - generic callbacks and setup methods create a lot of warnings
+disable = [
+    "format",
+    "abstract-method",
+    "cyclic-import",
+    "duplicate-code",
+    "locally-disabled",
+    "no-else-return",
+    "not-context-manager",
+    "too-few-public-methods",
+    "too-many-arguments",
+    "too-many-branches",
+    "too-many-instance-attributes",
+    "too-many-lines",
+    "too-many-locals",
+    "too-many-public-methods",
+    "too-many-return-statements",
+    "too-many-statements",
+    "unused-argument",
+    "consider-using-with",
+]
+
+[tool.pylint.REPORTS]
+score = false
+
+[tool.pylint.TYPECHECK]
+ignored-modules = ["distutils"]
+
+[tool.pylint.FORMAT]
+expected-line-ending-format = "LF"
+
+[tool.pylint.EXCEPTIONS]
+overgeneral-exceptions = ["builtins.BaseException", "builtins.Exception"]
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+norecursedirs = [".git"]
+log_format = "%(asctime)s.%(msecs)03d %(levelname)-8s %(threadName)s %(name)s:%(filename)s:%(lineno)s %(message)s"
+log_date_format = "%Y-%m-%d %H:%M:%S"
+asyncio_mode = "auto"
+filterwarnings = [
+    "error",
+    "ignore:pkg_resources is deprecated as an API:DeprecationWarning:dirhash",
+    "ignore::pytest.PytestUnraisableExceptionWarning",
+]
+
+[tool.isort]
+multi_line_output = 3
+include_trailing_comma = true
+force_grid_wrap = 0
+line_length = 88
+indent = "    "
+force_sort_within_sections = true
+sections = ["FUTURE", "STDLIB", "THIRDPARTY", "FIRSTPARTY", "LOCALFOLDER"]
+default_section = "THIRDPARTY"
+forced_separate = "tests"
+combine_as_imports = true
+use_parentheses = true
+known_first_party = ["supervisor", "tests"]

pytest.ini

@@ -1,2 +0,0 @@
-[pytest]
-asyncio_mode = auto

requirements.txt

@@ -1,26 +1,30 @@
-aiodns==3.0.0
-aiohttp==3.8.5
+aiodns==3.1.1
+aiohttp==3.9.1
+aiohttp-fast-url-dispatcher==0.3.0
 async_timeout==4.0.3
 atomicwrites-homeassistant==1.4.1
-attrs==23.1.0
-awesomeversion==23.8.0
-brotli==1.0.9
-ciso8601==2.3.0
-colorlog==6.7.0
+attrs==23.2.0
+awesomeversion==23.11.0
+brotli==1.1.0
+ciso8601==2.3.1
+colorlog==6.8.0
 cpe==1.2.1
-cryptography==41.0.3
-debugpy==1.6.7
-deepmerge==1.1.0
+cryptography==41.0.7
+debugpy==1.8.0
+deepmerge==1.1.1
 dirhash==0.2.1
-docker==6.1.3
+docker==7.0.0
 faust-cchardet==2.1.19
-gitpython==3.1.34
-jinja2==3.1.2
+gitpython==3.1.41
+jinja2==3.1.3
+orjson==3.9.10
 pulsectl==23.5.2
 pyudev==0.24.1
-ruamel.yaml==0.17.21
-securetar==2023.3.0
-sentry-sdk==1.30.0
-voluptuous==0.13.1
-dbus-fast==1.94.1
-typing_extensions==4.7.1
+PyYAML==6.0.1
+securetar==2023.12.0
+sentry-sdk==1.39.2
+setuptools==69.0.3
+voluptuous==0.14.1
+dbus-fast==2.21.0
+typing_extensions==4.9.0
+zlib-fast==0.1.0

requirements_tests.txt

@@ -1,16 +1,16 @@
-black==23.7.0
-coverage==7.3.0
+black==23.12.1
+coverage==7.4.0
 flake8-docstrings==1.7.0
-flake8==6.1.0
-pre-commit==3.4.0
+flake8==7.0.0
+pre-commit==3.6.0
 pydocstyle==6.3.0
-pylint==2.17.5
-pytest-aiohttp==1.0.4
-pytest-asyncio==0.18.3
+pylint==3.0.3
+pytest-aiohttp==1.0.5
+pytest-asyncio==0.23.3
 pytest-cov==4.1.0
-pytest-timeout==2.1.0
-pytest==7.4.1
-pyupgrade==3.10.1
-time-machine==2.12.0
-typing_extensions==4.7.1
-urllib3==2.0.4
+pytest-timeout==2.2.0
+pytest==7.4.4
+pyupgrade==3.15.0
+time-machine==2.13.0
+typing_extensions==4.9.0
+urllib3==2.1.0

rootfs/etc/services.d/watchdog/run

@@ -15,7 +15,7 @@ do
     if [[ "${supervisor_state}" = "running"  ]]; then
 
         # Check API
-        if bashio::supervisor.ping; then
+        if bashio::supervisor.ping > /dev/null; then
             failed_count=0
         else
             bashio::log.warning "Maybe found an issue on API healthy"

setup.cfg

@@ -1,17 +1,3 @@
-[isort]
-multi_line_output = 3
-include_trailing_comma=True
-force_grid_wrap=0
-line_length=88
-indent = "    "
-force_sort_within_sections = true
-sections = FUTURE,STDLIB,THIRDPARTY,FIRSTPARTY,LOCALFOLDER
-default_section = THIRDPARTY
-forced_separate = tests
-combine_as_imports = true
-use_parentheses = true
-known_first_party = supervisor,tests
-
 [flake8]
 exclude = .venv,.git,.tox,docs,venv,bin,lib,deps,build
 doctests = True

setup.py

@@ -1,60 +1,27 @@
 """Home Assistant Supervisor setup."""
+from pathlib import Path
+import re
+
 from setuptools import setup
 
-from supervisor.const import SUPERVISOR_VERSION
+RE_SUPERVISOR_VERSION = re.compile(r"^SUPERVISOR_VERSION =\s*(.+)$")
+
+SUPERVISOR_DIR = Path(__file__).parent
+REQUIREMENTS_FILE = SUPERVISOR_DIR / "requirements.txt"
+CONST_FILE = SUPERVISOR_DIR / "supervisor/const.py"
+
+REQUIREMENTS = REQUIREMENTS_FILE.read_text(encoding="utf-8")
+CONSTANTS = CONST_FILE.read_text(encoding="utf-8")
+
+
+def _get_supervisor_version():
+    for line in CONSTANTS.split("/n"):
+        if match := RE_SUPERVISOR_VERSION.match(line):
+            return match.group(1)
+    return "99.9.9dev"
+
 
 setup(
-    name="Supervisor",
-    version=SUPERVISOR_VERSION,
-    license="BSD License",
-    author="The Home Assistant Authors",
-    author_email="hello@home-assistant.io",
-    url="https://home-assistant.io/",
-    description=("Open-source private cloud os for Home-Assistant" " based on HassOS"),
-    long_description=(
-        "A maintainless private cloud operator system that"
-        "setup a Home-Assistant instance. Based on HassOS"
-    ),
-    classifiers=[
-        "Intended Audience :: End Users/Desktop",
-        "Intended Audience :: Developers",
-        "License :: OSI Approved :: Apache Software License",
-        "Operating System :: OS Independent",
-        "Topic :: Home Automation",
-        "Topic :: Software Development :: Libraries :: Python Modules",
-        "Topic :: Scientific/Engineering :: Atmospheric Science",
-        "Development Status :: 5 - Production/Stable",
-        "Intended Audience :: Developers",
-        "Programming Language :: Python :: 3.8",
-    ],
-    keywords=["docker", "home-assistant", "api"],
-    zip_safe=False,
-    platforms="any",
-    packages=[
-        "supervisor.addons",
-        "supervisor.api",
-        "supervisor.backups",
-        "supervisor.dbus.network",
-        "supervisor.dbus.network.setting",
-        "supervisor.dbus",
-        "supervisor.discovery.services",
-        "supervisor.discovery",
-        "supervisor.docker",
-        "supervisor.homeassistant",
-        "supervisor.host",
-        "supervisor.jobs",
-        "supervisor.misc",
-        "supervisor.plugins",
-        "supervisor.resolution.checks",
-        "supervisor.resolution.evaluations",
-        "supervisor.resolution.fixups",
-        "supervisor.resolution",
-        "supervisor.security",
-        "supervisor.services.modules",
-        "supervisor.services",
-        "supervisor.store",
-        "supervisor.utils",
-        "supervisor",
-    ],
-    include_package_data=True,
+    version=_get_supervisor_version(),
+    dependencies=REQUIREMENTS.split("/n"),
 )

supervisor/__main__.py

@@ -5,7 +5,13 @@ import logging
 from pathlib import Path
 import sys
 
-from supervisor import bootstrap
+import zlib_fast
+
+# Enable fast zlib before importing supervisor
+zlib_fast.enable()
+
+from supervisor import bootstrap  # noqa: E402
+from supervisor.utils.logging import activate_log_queue_handler  # noqa: E402
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
 
@@ -38,6 +44,8 @@ if __name__ == "__main__":
     executor = ThreadPoolExecutor(thread_name_prefix="SyncWorker")
     loop.set_default_executor(executor)
 
+    activate_log_queue_handler()
+
     _LOGGER.info("Initializing Supervisor setup")
     coresys = loop.run_until_complete(bootstrap.initialize_coresys())
     loop.set_debug(coresys.config.debug)

supervisor/addons/__init__.py

@@ -1,477 +1 @@
 """Init file for Supervisor add-ons."""
-import asyncio
-from collections.abc import Awaitable
-from contextlib import suppress
-import logging
-import tarfile
-from typing import Union
-
-from ..const import AddonBoot, AddonStartup, AddonState
-from ..coresys import CoreSys, CoreSysAttributes
-from ..exceptions import (
-    AddonConfigurationError,
-    AddonsError,
-    AddonsJobError,
-    AddonsNotSupportedError,
-    CoreDNSError,
-    DockerAPIError,
-    DockerError,
-    DockerNotFound,
-    HomeAssistantAPIError,
-    HostAppArmorError,
-)
-from ..jobs.decorator import Job, JobCondition
-from ..resolution.const import ContextType, IssueType, SuggestionType
-from ..store.addon import AddonStore
-from ..utils import check_exception_chain
-from ..utils.sentry import capture_exception
-from .addon import Addon
-from .const import ADDON_UPDATE_CONDITIONS
-from .data import AddonsData
-
-_LOGGER: logging.Logger = logging.getLogger(__name__)
-
-AnyAddon = Union[Addon, AddonStore]
-
-
-class AddonManager(CoreSysAttributes):
-    """Manage add-ons inside Supervisor."""
-
-    def __init__(self, coresys: CoreSys):
-        """Initialize Docker base wrapper."""
-        self.coresys: CoreSys = coresys
-        self.data: AddonsData = AddonsData(coresys)
-        self.local: dict[str, Addon] = {}
-        self.store: dict[str, AddonStore] = {}
-
-    @property
-    def all(self) -> list[AnyAddon]:
-        """Return a list of all add-ons."""
-        addons: dict[str, AnyAddon] = {**self.store, **self.local}
-        return list(addons.values())
-
-    @property
-    def installed(self) -> list[Addon]:
-        """Return a list of all installed add-ons."""
-        return list(self.local.values())
-
-    def get(self, addon_slug: str, local_only: bool = False) -> AnyAddon | None:
-        """Return an add-on from slug.
-
-        Prio:
-          1 - Local
-          2 - Store
-        """
-        if addon_slug in self.local:
-            return self.local[addon_slug]
-        if not local_only:
-            return self.store.get(addon_slug)
-        return None
-
-    def from_token(self, token: str) -> Addon | None:
-        """Return an add-on from Supervisor token."""
-        for addon in self.installed:
-            if token == addon.supervisor_token:
-                return addon
-        return None
-
-    async def load(self) -> None:
-        """Start up add-on management."""
-        tasks = []
-        for slug in self.data.system:
-            addon = self.local[slug] = Addon(self.coresys, slug)
-            tasks.append(self.sys_create_task(addon.load()))
-
-        # Run initial tasks
-        _LOGGER.info("Found %d installed add-ons", len(tasks))
-        if tasks:
-            await asyncio.wait(tasks)
-
-        # Sync DNS
-        await self.sync_dns()
-
-    async def boot(self, stage: AddonStartup) -> None:
-        """Boot add-ons with mode auto."""
-        tasks: list[Addon] = []
-        for addon in self.installed:
-            if addon.boot != AddonBoot.AUTO or addon.startup != stage:
-                continue
-            tasks.append(addon)
-
-        # Evaluate add-ons which need to be started
-        _LOGGER.info("Phase '%s' starting %d add-ons", stage, len(tasks))
-        if not tasks:
-            return
-
-        # Start Add-ons sequential
-        # avoid issue on slow IO
-        # Config.wait_boot is deprecated. Until addons update with healthchecks,
-        # add a sleep task for it to keep the same minimum amount of wait time
-        wait_boot: list[Awaitable[None]] = [asyncio.sleep(self.sys_config.wait_boot)]
-        for addon in tasks:
-            try:
-                if start_task := await addon.start():
-                    wait_boot.append(start_task)
-            except AddonsError as err:
-                # Check if there is an system/user issue
-                if check_exception_chain(
-                    err, (DockerAPIError, DockerNotFound, AddonConfigurationError)
-                ):
-                    addon.boot = AddonBoot.MANUAL
-                    addon.save_persist()
-            except Exception as err:  # pylint: disable=broad-except
-                capture_exception(err)
-            else:
-                continue
-
-            _LOGGER.warning("Can't start Add-on %s", addon.slug)
-
-        # Ignore exceptions from waiting for addon startup, addon errors handled elsewhere
-        await asyncio.gather(*wait_boot, return_exceptions=True)
-
-    async def shutdown(self, stage: AddonStartup) -> None:
-        """Shutdown addons."""
-        tasks: list[Addon] = []
-        for addon in self.installed:
-            if addon.state != AddonState.STARTED or addon.startup != stage:
-                continue
-            tasks.append(addon)
-
-        # Evaluate add-ons which need to be stopped
-        _LOGGER.info("Phase '%s' stopping %d add-ons", stage, len(tasks))
-        if not tasks:
-            return
-
-        # Stop Add-ons sequential
-        # avoid issue on slow IO
-        for addon in tasks:
-            try:
-                await addon.stop()
-            except Exception as err:  # pylint: disable=broad-except
-                _LOGGER.warning("Can't stop Add-on %s: %s", addon.slug, err)
-                capture_exception(err)
-
-    @Job(
-        name="addon_manager_install",
-        conditions=ADDON_UPDATE_CONDITIONS,
-        on_condition=AddonsJobError,
-    )
-    async def install(self, slug: str) -> None:
-        """Install an add-on."""
-        self.sys_jobs.current.reference = slug
-
-        if slug in self.local:
-            raise AddonsError(f"Add-on {slug} is already installed", _LOGGER.warning)
-        store = self.store.get(slug)
-
-        if not store:
-            raise AddonsError(f"Add-on {slug} does not exist", _LOGGER.error)
-
-        store.validate_availability()
-
-        self.data.install(store)
-        addon = Addon(self.coresys, slug)
-        await addon.load()
-
-        if not addon.path_data.is_dir():
-            _LOGGER.info(
-                "Creating Home Assistant add-on data folder %s", addon.path_data
-            )
-            addon.path_data.mkdir()
-
-        # Setup/Fix AppArmor profile
-        await addon.install_apparmor()
-
-        try:
-            await addon.instance.install(store.version, store.image, arch=addon.arch)
-        except DockerError as err:
-            self.data.uninstall(addon)
-            raise AddonsError() from err
-
-        self.local[slug] = addon
-
-        # Reload ingress tokens
-        if addon.with_ingress:
-            await self.sys_ingress.reload()
-
-        _LOGGER.info("Add-on '%s' successfully installed", slug)
-
-    async def uninstall(self, slug: str) -> None:
-        """Remove an add-on."""
-        if slug not in self.local:
-            _LOGGER.warning("Add-on %s is not installed", slug)
-            return
-        addon = self.local[slug]
-
-        try:
-            await addon.instance.remove()
-        except DockerError as err:
-            raise AddonsError() from err
-
-        addon.state = AddonState.UNKNOWN
-
-        await addon.unload()
-
-        # Cleanup audio settings
-        if addon.path_pulse.exists():
-            with suppress(OSError):
-                addon.path_pulse.unlink()
-
-        # Cleanup AppArmor profile
-        with suppress(HostAppArmorError):
-            await addon.uninstall_apparmor()
-
-        # Cleanup Ingress panel from sidebar
-        if addon.ingress_panel:
-            addon.ingress_panel = False
-            with suppress(HomeAssistantAPIError):
-                await self.sys_ingress.update_hass_panel(addon)
-
-        # Cleanup Ingress dynamic port assignment
-        if addon.with_ingress:
-            self.sys_create_task(self.sys_ingress.reload())
-            self.sys_ingress.del_dynamic_port(slug)
-
-        # Cleanup discovery data
-        for message in self.sys_discovery.list_messages:
-            if message.addon != addon.slug:
-                continue
-            self.sys_discovery.remove(message)
-
-        # Cleanup services data
-        for service in self.sys_services.list_services:
-            if addon.slug not in service.active:
-                continue
-            service.del_service_data(addon)
-
-        self.data.uninstall(addon)
-        self.local.pop(slug)
-
-        _LOGGER.info("Add-on '%s' successfully removed", slug)
-
-    @Job(
-        name="addon_manager_update",
-        conditions=ADDON_UPDATE_CONDITIONS,
-        on_condition=AddonsJobError,
-    )
-    async def update(
-        self, slug: str, backup: bool | None = False
-    ) -> Awaitable[None] | None:
-        """Update add-on.
-
-        Returns a coroutine that completes when addon has state 'started' (see addon.start)
-        if addon is started after update. Else nothing is returned.
-        """
-        self.sys_jobs.current.reference = slug
-
-        if slug not in self.local:
-            raise AddonsError(f"Add-on {slug} is not installed", _LOGGER.error)
-        addon = self.local[slug]
-
-        if addon.is_detached:
-            raise AddonsError(
-                f"Add-on {slug} is not available inside store", _LOGGER.error
-            )
-        store = self.store[slug]
-
-        if addon.version == store.version:
-            raise AddonsError(f"No update available for add-on {slug}", _LOGGER.warning)
-
-        # Check if available, Maybe something have changed
-        store.validate_availability()
-
-        if backup:
-            await self.sys_backups.do_backup_partial(
-                name=f"addon_{addon.slug}_{addon.version}",
-                homeassistant=False,
-                addons=[addon.slug],
-            )
-
-        # Update instance
-        last_state: AddonState = addon.state
-        old_image = addon.image
-        try:
-            await addon.instance.update(store.version, store.image)
-        except DockerError as err:
-            raise AddonsError() from err
-
-        _LOGGER.info("Add-on '%s' successfully updated", slug)
-        self.data.update(store)
-
-        # Cleanup
-        with suppress(DockerError):
-            await addon.instance.cleanup(old_image=old_image)
-
-        # Setup/Fix AppArmor profile
-        await addon.install_apparmor()
-
-        # restore state
-        return (
-            await addon.start()
-            if last_state in [AddonState.STARTED, AddonState.STARTUP]
-            else None
-        )
-
-    @Job(
-        name="addon_manager_rebuild",
-        conditions=[
-            JobCondition.FREE_SPACE,
-            JobCondition.INTERNET_HOST,
-            JobCondition.HEALTHY,
-        ],
-        on_condition=AddonsJobError,
-    )
-    async def rebuild(self, slug: str) -> Awaitable[None] | None:
-        """Perform a rebuild of local build add-on.
-
-        Returns a coroutine that completes when addon has state 'started' (see addon.start)
-        if addon is started after rebuild. Else nothing is returned.
-        """
-        self.sys_jobs.current.reference = slug
-
-        if slug not in self.local:
-            raise AddonsError(f"Add-on {slug} is not installed", _LOGGER.error)
-        addon = self.local[slug]
-
-        if addon.is_detached:
-            raise AddonsError(
-                f"Add-on {slug} is not available inside store", _LOGGER.error
-            )
-        store = self.store[slug]
-
-        # Check if a rebuild is possible now
-        if addon.version != store.version:
-            raise AddonsError(
-                "Version changed, use Update instead Rebuild", _LOGGER.error
-            )
-        if not addon.need_build:
-            raise AddonsNotSupportedError(
-                "Can't rebuild a image based add-on", _LOGGER.error
-            )
-
-        # remove docker container but not addon config
-        last_state: AddonState = addon.state
-        try:
-            await addon.instance.remove()
-            await addon.instance.install(addon.version)
-        except DockerError as err:
-            raise AddonsError() from err
-
-        self.data.update(store)
-        _LOGGER.info("Add-on '%s' successfully rebuilt", slug)
-
-        # restore state
-        return (
-            await addon.start()
-            if last_state in [AddonState.STARTED, AddonState.STARTUP]
-            else None
-        )
-
-    @Job(
-        name="addon_manager_restore",
-        conditions=[
-            JobCondition.FREE_SPACE,
-            JobCondition.INTERNET_HOST,
-            JobCondition.HEALTHY,
-        ],
-        on_condition=AddonsJobError,
-    )
-    async def restore(
-        self, slug: str, tar_file: tarfile.TarFile
-    ) -> Awaitable[None] | None:
-        """Restore state of an add-on.
-
-        Returns a coroutine that completes when addon has state 'started' (see addon.start)
-        if addon is started after restore. Else nothing is returned.
-        """
-        self.sys_jobs.current.reference = slug
-
-        if slug not in self.local:
-            _LOGGER.debug("Add-on %s is not local available for restore", slug)
-            addon = Addon(self.coresys, slug)
-        else:
-            _LOGGER.debug("Add-on %s is local available for restore", slug)
-            addon = self.local[slug]
-
-        wait_for_start = await addon.restore(tar_file)
-
-        # Check if new
-        if slug not in self.local:
-            _LOGGER.info("Detect new Add-on after restore %s", slug)
-            self.local[slug] = addon
-
-        # Update ingress
-        if addon.with_ingress:
-            await self.sys_ingress.reload()
-            with suppress(HomeAssistantAPIError):
-                await self.sys_ingress.update_hass_panel(addon)
-
-        return wait_for_start
-
-    @Job(
-        name="addon_manager_repair",
-        conditions=[JobCondition.FREE_SPACE, JobCondition.INTERNET_HOST],
-    )
-    async def repair(self) -> None:
-        """Repair local add-ons."""
-        needs_repair: list[Addon] = []
-
-        # Evaluate Add-ons to repair
-        for addon in self.installed:
-            if await addon.instance.exists():
-                continue
-            needs_repair.append(addon)
-
-        _LOGGER.info("Found %d add-ons to repair", len(needs_repair))
-        if not needs_repair:
-            return
-
-        for addon in needs_repair:
-            _LOGGER.info("Repairing for add-on: %s", addon.slug)
-            with suppress(DockerError, KeyError):
-                # Need pull a image again
-                if not addon.need_build:
-                    await addon.instance.install(addon.version, addon.image)
-                    continue
-
-                # Need local lookup
-                if addon.need_build and not addon.is_detached:
-                    store = self.store[addon.slug]
-                    # If this add-on is available for rebuild
-                    if addon.version == store.version:
-                        await addon.instance.install(addon.version, addon.image)
-                        continue
-
-            _LOGGER.error("Can't repair %s", addon.slug)
-            with suppress(AddonsError):
-                await self.uninstall(addon.slug)
-
-    async def sync_dns(self) -> None:
-        """Sync add-ons DNS names."""
-        # Update hosts
-        add_host_coros: list[Awaitable[None]] = []
-        for addon in self.installed:
-            try:
-                if not await addon.instance.is_running():
-                    continue
-            except DockerError as err:
-                _LOGGER.warning("Add-on %s is corrupt: %s", addon.slug, err)
-                self.sys_resolution.create_issue(
-                    IssueType.CORRUPT_DOCKER,
-                    ContextType.ADDON,
-                    reference=addon.slug,
-                    suggestions=[SuggestionType.EXECUTE_REPAIR],
-                )
-                capture_exception(err)
-            else:
-                add_host_coros.append(
-                    self.sys_plugins.dns.add_host(
-                        ipv4=addon.ip_address, names=[addon.hostname], write=False
-                    )
-                )
-
-        await asyncio.gather(*add_host_coros)
-
-        # Write hosts files
-        with suppress(CoreDNSError):
-            await self.sys_plugins.dns.write_hosts()

supervisor/addons/addon.py

@@ -3,6 +3,7 @@ import asyncio
 from collections.abc import Awaitable
 from contextlib import suppress
 from copy import deepcopy
+import errno
 from ipaddress import IPv4Address
 import logging
 from pathlib import Path, PurePath
@@ -64,12 +65,15 @@ from ..exceptions import (
     AddonsNotSupportedError,
     ConfigurationFileError,
     DockerError,
+    HomeAssistantAPIError,
     HostAppArmorError,
 )
 from ..hardware.data import Device
 from ..homeassistant.const import WSEvent, WSType
 from ..jobs.const import JobExecutionLimit
 from ..jobs.decorator import Job
+from ..resolution.const import UnhealthyReason
+from ..store.addon import AddonStore
 from ..utils import check_port
 from ..utils.apparmor import adjust_profile
 from ..utils.json import read_json_file, write_json_file
@@ -80,6 +84,7 @@ from .const import (
     WATCHDOG_THROTTLE_MAX_CALLS,
     WATCHDOG_THROTTLE_PERIOD,
     AddonBackupMode,
+    MappingType,
 )
 from .model import AddonModel, Data
 from .options import AddonOptions
@@ -181,6 +186,7 @@ class Addon(AddonModel):
             )
         )
 
+        await self._check_ingress_port()
         with suppress(DockerError):
             await self.instance.attach(version=self.version)
 
@@ -199,6 +205,11 @@ class Addon(AddonModel):
         """Return add-on data from store."""
         return self.sys_store.data.addons.get(self.slug, self.data)
 
+    @property
+    def addon_store(self) -> AddonStore | None:
+        """Return store representation of addon."""
+        return self.sys_addons.store.get(self.slug)
+
     @property
     def persist(self) -> Data:
         """Return add-on data/config."""
@@ -387,7 +398,7 @@ class Addon(AddonModel):
 
         port = self.data[ATTR_INGRESS_PORT]
         if port == 0:
-            return self.sys_ingress.get_dynamic_port(self.slug)
+            raise RuntimeError(f"No port set for add-on {self.slug}")
         return port
 
     @property
@@ -453,6 +464,21 @@ class Addon(AddonModel):
         """Return add-on data path external for Docker."""
         return PurePath(self.sys_config.path_extern_addons_data, self.slug)
 
+    @property
+    def addon_config_used(self) -> bool:
+        """Add-on is using its public config folder."""
+        return MappingType.ADDON_CONFIG in self.map_volumes
+
+    @property
+    def path_config(self) -> Path:
+        """Return add-on config path inside Supervisor."""
+        return Path(self.sys_config.path_addon_configs, self.slug)
+
+    @property
+    def path_extern_config(self) -> PurePath:
+        """Return add-on config path external for Docker."""
+        return PurePath(self.sys_config.path_extern_addon_configs, self.slug)
+
     @property
     def path_options(self) -> Path:
         """Return path to add-on options."""
@@ -516,7 +542,7 @@ class Addon(AddonModel):
 
         # TCP monitoring
         if s_prefix == "tcp":
-            return await self.sys_run_in_executor(check_port, self.ip_address, port)
+            return await check_port(self.ip_address, port)
 
         # lookup the correct protocol from config
         if t_proto:
@@ -532,7 +558,7 @@ class Addon(AddonModel):
             ) as req:
                 if req.status < 300:
                     return True
-        except (asyncio.TimeoutError, aiohttp.ClientError):
+        except (TimeoutError, aiohttp.ClientError):
             pass
 
         return False
@@ -559,6 +585,11 @@ class Addon(AddonModel):
 
         raise AddonConfigurationError()
 
+    @Job(
+        name="addon_unload",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
     async def unload(self) -> None:
         """Unload add-on and remove data."""
         if self._startup_task:
@@ -570,11 +601,185 @@ class Addon(AddonModel):
         for listener in self._listeners:
             self.sys_bus.remove_listener(listener)
 
-        if not self.path_data.is_dir():
+        if self.path_data.is_dir():
+            _LOGGER.info("Removing add-on data folder %s", self.path_data)
+            await remove_data(self.path_data)
+
+    async def _check_ingress_port(self):
+        """Assign a ingress port if dynamic port selection is used."""
+        if not self.with_ingress:
             return
 
-        _LOGGER.info("Removing add-on data folder %s", self.path_data)
-        await remove_data(self.path_data)
+        if self.data[ATTR_INGRESS_PORT] == 0:
+            self.data[ATTR_INGRESS_PORT] = await self.sys_ingress.get_dynamic_port(
+                self.slug
+            )
+
+    @Job(
+        name="addon_install",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
+    async def install(self) -> None:
+        """Install and setup this addon."""
+        self.sys_addons.data.install(self.addon_store)
+        await self.load()
+
+        if not self.path_data.is_dir():
+            _LOGGER.info(
+                "Creating Home Assistant add-on data folder %s", self.path_data
+            )
+            self.path_data.mkdir()
+
+        # Setup/Fix AppArmor profile
+        await self.install_apparmor()
+
+        # Install image
+        try:
+            await self.instance.install(
+                self.latest_version, self.addon_store.image, arch=self.arch
+            )
+        except DockerError as err:
+            self.sys_addons.data.uninstall(self)
+            raise AddonsError() from err
+
+        # Add to addon manager
+        self.sys_addons.local[self.slug] = self
+
+        # Reload ingress tokens
+        if self.with_ingress:
+            await self.sys_ingress.reload()
+
+    @Job(
+        name="addon_uninstall",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
+    async def uninstall(self) -> None:
+        """Uninstall and cleanup this addon."""
+        try:
+            await self.instance.remove()
+        except DockerError as err:
+            raise AddonsError() from err
+
+        self.state = AddonState.UNKNOWN
+
+        await self.unload()
+
+        # Cleanup audio settings
+        if self.path_pulse.exists():
+            with suppress(OSError):
+                self.path_pulse.unlink()
+
+        # Cleanup AppArmor profile
+        with suppress(HostAppArmorError):
+            await self.uninstall_apparmor()
+
+        # Cleanup Ingress panel from sidebar
+        if self.ingress_panel:
+            self.ingress_panel = False
+            with suppress(HomeAssistantAPIError):
+                await self.sys_ingress.update_hass_panel(self)
+
+        # Cleanup Ingress dynamic port assignment
+        if self.with_ingress:
+            self.sys_create_task(self.sys_ingress.reload())
+            self.sys_ingress.del_dynamic_port(self.slug)
+
+        # Cleanup discovery data
+        for message in self.sys_discovery.list_messages:
+            if message.addon != self.slug:
+                continue
+            self.sys_discovery.remove(message)
+
+        # Cleanup services data
+        for service in self.sys_services.list_services:
+            if self.slug not in service.active:
+                continue
+            service.del_service_data(self)
+
+        # Remove from addon manager
+        self.sys_addons.data.uninstall(self)
+        self.sys_addons.local.pop(self.slug)
+
+    @Job(
+        name="addon_update",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
+    async def update(self) -> asyncio.Task | None:
+        """Update this addon to latest version.
+
+        Returns a Task that completes when addon has state 'started' (see start)
+        if it was running. Else nothing is returned.
+        """
+        old_image = self.image
+        # Cache data to prevent races with other updates to global
+        store = self.addon_store.clone()
+
+        try:
+            await self.instance.update(store.version, store.image, arch=self.arch)
+        except DockerError as err:
+            raise AddonsError() from err
+
+        # Stop the addon if running
+        if (last_state := self.state) in {AddonState.STARTED, AddonState.STARTUP}:
+            await self.stop()
+
+        try:
+            _LOGGER.info("Add-on '%s' successfully updated", self.slug)
+            self.sys_addons.data.update(store)
+            await self._check_ingress_port()
+
+            # Cleanup
+            with suppress(DockerError):
+                await self.instance.cleanup(
+                    old_image=old_image, image=store.image, version=store.version
+                )
+
+            # Setup/Fix AppArmor profile
+            await self.install_apparmor()
+
+        finally:
+            # restore state. Return Task for caller if no exception
+            out = (
+                await self.start()
+                if last_state in {AddonState.STARTED, AddonState.STARTUP}
+                else None
+            )
+        return out
+
+    @Job(
+        name="addon_rebuild",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
+    async def rebuild(self) -> asyncio.Task | None:
+        """Rebuild this addons container and image.
+
+        Returns a Task that completes when addon has state 'started' (see start)
+        if it was running. Else nothing is returned.
+        """
+        last_state: AddonState = self.state
+        try:
+            # remove docker container but not addon config
+            try:
+                await self.instance.remove()
+                await self.instance.install(self.version)
+            except DockerError as err:
+                raise AddonsError() from err
+
+            self.sys_addons.data.update(self.addon_store)
+            _LOGGER.info("Add-on '%s' successfully rebuilt", self.slug)
+
+        finally:
+            # restore state
+            out = (
+                await self.start()
+                if last_state in [AddonState.STARTED, AddonState.STARTUP]
+                else None
+            )
+        return out
 
     def write_pulse(self) -> None:
         """Write asound config to file and return True on success."""
@@ -590,6 +795,8 @@ class Addon(AddonModel):
         try:
             self.path_pulse.write_text(pulse_config, encoding="utf-8")
         except OSError as err:
+            if err.errno == errno.EBADMSG:
+                self.sys_resolution.unhealthy = UnhealthyReason.OSERROR_BAD_MESSAGE
             _LOGGER.error(
                 "Add-on %s can't write pulse/client.config: %s", self.slug, err
             )
@@ -660,9 +867,9 @@ class Addon(AddonModel):
         try:
             self._startup_task = self.sys_create_task(self._startup_event.wait())
             await asyncio.wait_for(self._startup_task, STARTUP_TIMEOUT)
-        except asyncio.TimeoutError:
+        except TimeoutError:
             _LOGGER.warning(
-                "Timeout while waiting for addon %s to start, took more then %s seconds",
+                "Timeout while waiting for addon %s to start, took more than %s seconds",
                 self.name,
                 STARTUP_TIMEOUT,
             )
@@ -671,16 +878,21 @@ class Addon(AddonModel):
         finally:
             self._startup_task = None
 
-    async def start(self) -> Awaitable[None]:
+    @Job(
+        name="addon_start",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
+    async def start(self) -> asyncio.Task:
         """Set options and start add-on.
 
-        Returns a coroutine that completes when addon has state 'started'.
+        Returns a Task that completes when addon has state 'started'.
         For addons with a healthcheck, that is when they become healthy or unhealthy.
         Addons without a healthcheck have state 'started' immediately.
         """
         if await self.instance.is_running():
             _LOGGER.warning("%s is already running!", self.slug)
-            return self._wait_for_startup()
+            return self.sys_create_task(self._wait_for_startup())
 
         # Access Token
         self.persist[ATTR_ACCESS_TOKEN] = secrets.token_hex(56)
@@ -693,6 +905,18 @@ class Addon(AddonModel):
         if self.with_audio:
             self.write_pulse()
 
+        def _check_addon_config_dir():
+            if self.path_config.is_dir():
+                return
+
+            _LOGGER.info(
+                "Creating Home Assistant add-on config folder %s", self.path_config
+            )
+            self.path_config.mkdir()
+
+        if self.addon_config_used:
+            await self.sys_run_in_executor(_check_addon_config_dir)
+
         # Start Add-on
         self._startup_event.clear()
         try:
@@ -701,8 +925,13 @@ class Addon(AddonModel):
             self.state = AddonState.ERROR
             raise AddonsError() from err
 
-        return self._wait_for_startup()
+        return self.sys_create_task(self._wait_for_startup())
 
+    @Job(
+        name="addon_stop",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
     async def stop(self) -> None:
         """Stop add-on."""
         self._manual_stop = True
@@ -712,10 +941,15 @@ class Addon(AddonModel):
             self.state = AddonState.ERROR
             raise AddonsError() from err
 
-    async def restart(self) -> Awaitable[None]:
+    @Job(
+        name="addon_restart",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
+    async def restart(self) -> asyncio.Task:
         """Restart add-on.
 
-        Returns a coroutine that completes when addon has state 'started' (see start).
+        Returns a Task that completes when addon has state 'started' (see start).
         """
         with suppress(AddonsError):
             await self.stop()
@@ -742,6 +976,11 @@ class Addon(AddonModel):
         except DockerError as err:
             raise AddonsError() from err
 
+    @Job(
+        name="addon_write_stdin",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
     async def write_stdin(self, data) -> None:
         """Write data to add-on stdin."""
         if not self.with_stdin:
@@ -771,14 +1010,59 @@ class Addon(AddonModel):
                 _LOGGER.error,
             ) from err
 
-    async def backup(self, tar_file: tarfile.TarFile) -> Awaitable[None] | None:
+    @Job(
+        name="addon_begin_backup",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
+    async def begin_backup(self) -> bool:
+        """Execute pre commands or stop addon if necessary.
+
+        Returns value of `is_running`. Caller should not call `end_backup` if return is false.
+        """
+        if not await self.is_running():
+            return False
+
+        if self.backup_mode == AddonBackupMode.COLD:
+            _LOGGER.info("Shutdown add-on %s for cold backup", self.slug)
+            await self.stop()
+
+        elif self.backup_pre is not None:
+            await self._backup_command(self.backup_pre)
+
+        return True
+
+    @Job(
+        name="addon_end_backup",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
+    async def end_backup(self) -> asyncio.Task | None:
+        """Execute post commands or restart addon if necessary.
+
+        Returns a Task that completes when addon has state 'started' (see start)
+        for cold backup. Else nothing is returned.
+        """
+        if self.backup_mode is AddonBackupMode.COLD:
+            _LOGGER.info("Starting add-on %s again", self.slug)
+            return await self.start()
+
+        if self.backup_post is not None:
+            await self._backup_command(self.backup_post)
+        return None
+
+    @Job(
+        name="addon_backup",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
+    async def backup(self, tar_file: tarfile.TarFile) -> asyncio.Task | None:
         """Backup state of an add-on.
 
-        Returns a coroutine that completes when addon has state 'started' (see start)
+        Returns a Task that completes when addon has state 'started' (see start)
         for cold backup. Else nothing is returned.
         """
         wait_for_start: Awaitable[None] | None = None
-        is_running = await self.is_running()
 
         with TemporaryDirectory(dir=self.sys_config.path_tmp) as temp:
             temp_path = Path(temp)
@@ -830,19 +1114,16 @@ class Addon(AddonModel):
                         arcname="data",
                     )
 
-            if (
-                is_running
-                and self.backup_mode == AddonBackupMode.HOT
-                and self.backup_pre is not None
-            ):
-                await self._backup_command(self.backup_pre)
-            elif is_running and self.backup_mode == AddonBackupMode.COLD:
-                _LOGGER.info("Shutdown add-on %s for cold backup", self.slug)
-                try:
-                    await self.instance.stop()
-                except DockerError as err:
-                    raise AddonsError() from err
+                    # Backup config
+                    if self.addon_config_used:
+                        atomic_contents_add(
+                            backup,
+                            self.path_config,
+                            excludes=self.backup_exclude,
+                            arcname="config",
+                        )
 
+            is_running = await self.begin_backup()
             try:
                 _LOGGER.info("Building backup for add-on %s", self.slug)
                 await self.sys_run_in_executor(_write_tarfile)
@@ -851,23 +1132,21 @@ class Addon(AddonModel):
                     f"Can't write tarfile {tar_file}: {err}", _LOGGER.error
                 ) from err
             finally:
-                if (
-                    is_running
-                    and self.backup_mode == AddonBackupMode.HOT
-                    and self.backup_post is not None
-                ):
-                    await self._backup_command(self.backup_post)
-                elif is_running and self.backup_mode is AddonBackupMode.COLD:
-                    _LOGGER.info("Starting add-on %s again", self.slug)
-                    wait_for_start = await self.start()
+                if is_running:
+                    wait_for_start = await self.end_backup()
 
         _LOGGER.info("Finish backup for addon %s", self.slug)
         return wait_for_start
 
-    async def restore(self, tar_file: tarfile.TarFile) -> Awaitable[None] | None:
+    @Job(
+        name="addon_restore",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
+    async def restore(self, tar_file: tarfile.TarFile) -> asyncio.Task | None:
         """Restore state of an add-on.
 
-        Returns a coroutine that completes when addon has state 'started' (see start)
+        Returns a Task that completes when addon has state 'started' (see start)
         if addon is started after restore. Else nothing is returned.
         """
         wait_for_start: Awaitable[None] | None = None
@@ -876,7 +1155,11 @@ class Addon(AddonModel):
             def _extract_tarfile():
                 """Extract tar backup."""
                 with tar_file as backup:
-                    backup.extractall(path=Path(temp), members=secure_path(backup))
+                    backup.extractall(
+                        path=Path(temp),
+                        members=secure_path(backup),
+                        filter="fully_trusted",
+                    )
 
             try:
                 await self.sys_run_in_executor(_extract_tarfile)
@@ -916,64 +1199,79 @@ class Addon(AddonModel):
 
             # Stop it first if its running
             if await self.instance.is_running():
-                with suppress(DockerError):
-                    await self.instance.stop()
+                await self.stop()
 
-            # Check version / restore image
-            version = data[ATTR_VERSION]
-            if not await self.instance.exists():
-                _LOGGER.info("Restore/Install of image for addon %s", self.slug)
-
-                image_file = Path(temp, "image.tar")
-                if image_file.is_file():
-                    with suppress(DockerError):
-                        await self.instance.import_image(image_file)
-                else:
-                    with suppress(DockerError):
-                        await self.instance.install(version, restore_image)
-                        await self.instance.cleanup()
-            elif self.instance.version != version or self.legacy:
-                _LOGGER.info("Restore/Update of image for addon %s", self.slug)
-                with suppress(DockerError):
-                    await self.instance.update(version, restore_image)
-
-            # Restore data
-            def _restore_data():
-                """Restore data."""
-                temp_data = Path(temp, "data")
-                if temp_data.is_dir():
-                    shutil.copytree(temp_data, self.path_data, symlinks=True)
-                else:
-                    self.path_data.mkdir()
-
-            _LOGGER.info("Restoring data for addon %s", self.slug)
-            if self.path_data.is_dir():
-                await remove_data(self.path_data)
             try:
-                await self.sys_run_in_executor(_restore_data)
-            except shutil.Error as err:
-                raise AddonsError(
-                    f"Can't restore origin data: {err}", _LOGGER.error
-                ) from err
+                # Check version / restore image
+                version = data[ATTR_VERSION]
+                if not await self.instance.exists():
+                    _LOGGER.info("Restore/Install of image for addon %s", self.slug)
+
+                    image_file = Path(temp, "image.tar")
+                    if image_file.is_file():
+                        with suppress(DockerError):
+                            await self.instance.import_image(image_file)
+                    else:
+                        with suppress(DockerError):
+                            await self.instance.install(
+                                version, restore_image, self.arch
+                            )
+                            await self.instance.cleanup()
+                elif self.instance.version != version or self.legacy:
+                    _LOGGER.info("Restore/Update of image for addon %s", self.slug)
+                    with suppress(DockerError):
+                        await self.instance.update(version, restore_image, self.arch)
+                self._check_ingress_port()
+
+                # Restore data and config
+                def _restore_data():
+                    """Restore data and config."""
+                    temp_data = Path(temp, "data")
+                    if temp_data.is_dir():
+                        shutil.copytree(temp_data, self.path_data, symlinks=True)
+                    else:
+                        self.path_data.mkdir()
+
+                    temp_config = Path(temp, "config")
+                    if temp_config.is_dir():
+                        shutil.copytree(temp_config, self.path_config, symlinks=True)
+                    elif self.addon_config_used:
+                        self.path_config.mkdir()
+
+                _LOGGER.info("Restoring data and config for addon %s", self.slug)
+                if self.path_data.is_dir():
+                    await remove_data(self.path_data)
+                if self.path_config.is_dir():
+                    await remove_data(self.path_config)
 
-            # Restore AppArmor
-            profile_file = Path(temp, "apparmor.txt")
-            if profile_file.exists():
                 try:
-                    await self.sys_host.apparmor.load_profile(self.slug, profile_file)
-                except HostAppArmorError as err:
-                    _LOGGER.error(
-                        "Can't restore AppArmor profile for add-on %s", self.slug
-                    )
-                    raise AddonsError() from err
+                    await self.sys_run_in_executor(_restore_data)
+                except shutil.Error as err:
+                    raise AddonsError(
+                        f"Can't restore origin data: {err}", _LOGGER.error
+                    ) from err
 
-            # Is add-on loaded
-            if not self.loaded:
-                await self.load()
+                # Restore AppArmor
+                profile_file = Path(temp, "apparmor.txt")
+                if profile_file.exists():
+                    try:
+                        await self.sys_host.apparmor.load_profile(
+                            self.slug, profile_file
+                        )
+                    except HostAppArmorError as err:
+                        _LOGGER.error(
+                            "Can't restore AppArmor profile for add-on %s", self.slug
+                        )
+                        raise AddonsError() from err
 
-            # Run add-on
-            if data[ATTR_STATE] == AddonState.STARTED:
-                wait_for_start = await self.start()
+                # Is add-on loaded
+                if not self.loaded:
+                    await self.load()
+
+            finally:
+                # Run add-on
+                if data[ATTR_STATE] == AddonState.STARTED:
+                    wait_for_start = await self.start()
 
         _LOGGER.info("Finished restore for add-on %s", self.slug)
         return wait_for_start

supervisor/addons/configuration.py

@@ -0,0 +1,11 @@
+"""Confgiuration Objects for Addon Config."""
+
+from dataclasses import dataclass
+
+
+@dataclass(slots=True)
+class FolderMapping:
+    """Represent folder mapping configuration."""
+
+    path: str | None
+    read_only: bool

supervisor/addons/const.py

@@ -12,8 +12,25 @@ class AddonBackupMode(StrEnum):
     COLD = "cold"
 
 
+class MappingType(StrEnum):
+    """Mapping type of an Add-on Folder."""
+
+    DATA = "data"
+    CONFIG = "config"
+    SSL = "ssl"
+    ADDONS = "addons"
+    BACKUP = "backup"
+    SHARE = "share"
+    MEDIA = "media"
+    HOMEASSISTANT_CONFIG = "homeassistant_config"
+    ALL_ADDON_CONFIGS = "all_addon_configs"
+    ADDON_CONFIG = "addon_config"
+
+
 ATTR_BACKUP = "backup"
 ATTR_CODENOTARY = "codenotary"
+ATTR_READ_ONLY = "read_only"
+ATTR_PATH = "path"
 WATCHDOG_RETRY_SECONDS = 10
 WATCHDOG_MAX_ATTEMPTS = 5
 WATCHDOG_THROTTLE_PERIOD = timedelta(minutes=30)

supervisor/addons/manager.py

@@ -0,0 +1,374 @@
+"""Supervisor add-on manager."""
+import asyncio
+from collections.abc import Awaitable
+from contextlib import suppress
+import logging
+import tarfile
+from typing import Union
+
+from ..const import AddonBoot, AddonStartup, AddonState
+from ..coresys import CoreSys, CoreSysAttributes
+from ..exceptions import (
+    AddonConfigurationError,
+    AddonsError,
+    AddonsJobError,
+    AddonsNotSupportedError,
+    CoreDNSError,
+    DockerAPIError,
+    DockerError,
+    DockerNotFound,
+    HassioError,
+    HomeAssistantAPIError,
+)
+from ..jobs.decorator import Job, JobCondition
+from ..resolution.const import ContextType, IssueType, SuggestionType
+from ..store.addon import AddonStore
+from ..utils import check_exception_chain
+from ..utils.sentry import capture_exception
+from .addon import Addon
+from .const import ADDON_UPDATE_CONDITIONS
+from .data import AddonsData
+
+_LOGGER: logging.Logger = logging.getLogger(__name__)
+
+AnyAddon = Union[Addon, AddonStore]
+
+
+class AddonManager(CoreSysAttributes):
+    """Manage add-ons inside Supervisor."""
+
+    def __init__(self, coresys: CoreSys):
+        """Initialize Docker base wrapper."""
+        self.coresys: CoreSys = coresys
+        self.data: AddonsData = AddonsData(coresys)
+        self.local: dict[str, Addon] = {}
+        self.store: dict[str, AddonStore] = {}
+
+    @property
+    def all(self) -> list[AnyAddon]:
+        """Return a list of all add-ons."""
+        addons: dict[str, AnyAddon] = {**self.store, **self.local}
+        return list(addons.values())
+
+    @property
+    def installed(self) -> list[Addon]:
+        """Return a list of all installed add-ons."""
+        return list(self.local.values())
+
+    def get(self, addon_slug: str, local_only: bool = False) -> AnyAddon | None:
+        """Return an add-on from slug.
+
+        Prio:
+          1 - Local
+          2 - Store
+        """
+        if addon_slug in self.local:
+            return self.local[addon_slug]
+        if not local_only:
+            return self.store.get(addon_slug)
+        return None
+
+    def from_token(self, token: str) -> Addon | None:
+        """Return an add-on from Supervisor token."""
+        for addon in self.installed:
+            if token == addon.supervisor_token:
+                return addon
+        return None
+
+    async def load(self) -> None:
+        """Start up add-on management."""
+        tasks = []
+        for slug in self.data.system:
+            addon = self.local[slug] = Addon(self.coresys, slug)
+            tasks.append(self.sys_create_task(addon.load()))
+
+        # Run initial tasks
+        _LOGGER.info("Found %d installed add-ons", len(tasks))
+        if tasks:
+            await asyncio.wait(tasks)
+
+        # Sync DNS
+        await self.sync_dns()
+
+    async def boot(self, stage: AddonStartup) -> None:
+        """Boot add-ons with mode auto."""
+        tasks: list[Addon] = []
+        for addon in self.installed:
+            if addon.boot != AddonBoot.AUTO or addon.startup != stage:
+                continue
+            tasks.append(addon)
+
+        # Evaluate add-ons which need to be started
+        _LOGGER.info("Phase '%s' starting %d add-ons", stage, len(tasks))
+        if not tasks:
+            return
+
+        # Start Add-ons sequential
+        # avoid issue on slow IO
+        # Config.wait_boot is deprecated. Until addons update with healthchecks,
+        # add a sleep task for it to keep the same minimum amount of wait time
+        wait_boot: list[Awaitable[None]] = [asyncio.sleep(self.sys_config.wait_boot)]
+        for addon in tasks:
+            try:
+                if start_task := await addon.start():
+                    wait_boot.append(start_task)
+            except AddonsError as err:
+                # Check if there is an system/user issue
+                if check_exception_chain(
+                    err, (DockerAPIError, DockerNotFound, AddonConfigurationError)
+                ):
+                    addon.boot = AddonBoot.MANUAL
+                    addon.save_persist()
+            except HassioError:
+                pass  # These are already handled
+            else:
+                continue
+
+            _LOGGER.warning("Can't start Add-on %s", addon.slug)
+
+        # Ignore exceptions from waiting for addon startup, addon errors handled elsewhere
+        await asyncio.gather(*wait_boot, return_exceptions=True)
+
+    async def shutdown(self, stage: AddonStartup) -> None:
+        """Shutdown addons."""
+        tasks: list[Addon] = []
+        for addon in self.installed:
+            if addon.state != AddonState.STARTED or addon.startup != stage:
+                continue
+            tasks.append(addon)
+
+        # Evaluate add-ons which need to be stopped
+        _LOGGER.info("Phase '%s' stopping %d add-ons", stage, len(tasks))
+        if not tasks:
+            return
+
+        # Stop Add-ons sequential
+        # avoid issue on slow IO
+        for addon in tasks:
+            try:
+                await addon.stop()
+            except Exception as err:  # pylint: disable=broad-except
+                _LOGGER.warning("Can't stop Add-on %s: %s", addon.slug, err)
+                capture_exception(err)
+
+    @Job(
+        name="addon_manager_install",
+        conditions=ADDON_UPDATE_CONDITIONS,
+        on_condition=AddonsJobError,
+    )
+    async def install(self, slug: str) -> None:
+        """Install an add-on."""
+        self.sys_jobs.current.reference = slug
+
+        if slug in self.local:
+            raise AddonsError(f"Add-on {slug} is already installed", _LOGGER.warning)
+        store = self.store.get(slug)
+
+        if not store:
+            raise AddonsError(f"Add-on {slug} does not exist", _LOGGER.error)
+
+        store.validate_availability()
+
+        await Addon(self.coresys, slug).install()
+
+        _LOGGER.info("Add-on '%s' successfully installed", slug)
+
+    async def uninstall(self, slug: str) -> None:
+        """Remove an add-on."""
+        if slug not in self.local:
+            _LOGGER.warning("Add-on %s is not installed", slug)
+            return
+
+        await self.local[slug].uninstall()
+
+        _LOGGER.info("Add-on '%s' successfully removed", slug)
+
+    @Job(
+        name="addon_manager_update",
+        conditions=ADDON_UPDATE_CONDITIONS,
+        on_condition=AddonsJobError,
+    )
+    async def update(
+        self, slug: str, backup: bool | None = False
+    ) -> asyncio.Task | None:
+        """Update add-on.
+
+        Returns a Task that completes when addon has state 'started' (see addon.start)
+        if addon is started after update. Else nothing is returned.
+        """
+        self.sys_jobs.current.reference = slug
+
+        if slug not in self.local:
+            raise AddonsError(f"Add-on {slug} is not installed", _LOGGER.error)
+        addon = self.local[slug]
+
+        if addon.is_detached:
+            raise AddonsError(
+                f"Add-on {slug} is not available inside store", _LOGGER.error
+            )
+        store = self.store[slug]
+
+        if addon.version == store.version:
+            raise AddonsError(f"No update available for add-on {slug}", _LOGGER.warning)
+
+        # Check if available, Maybe something have changed
+        store.validate_availability()
+
+        if backup:
+            await self.sys_backups.do_backup_partial(
+                name=f"addon_{addon.slug}_{addon.version}",
+                homeassistant=False,
+                addons=[addon.slug],
+            )
+
+        return await addon.update()
+
+    @Job(
+        name="addon_manager_rebuild",
+        conditions=[
+            JobCondition.FREE_SPACE,
+            JobCondition.INTERNET_HOST,
+            JobCondition.HEALTHY,
+        ],
+        on_condition=AddonsJobError,
+    )
+    async def rebuild(self, slug: str) -> asyncio.Task | None:
+        """Perform a rebuild of local build add-on.
+
+        Returns a Task that completes when addon has state 'started' (see addon.start)
+        if addon is started after rebuild. Else nothing is returned.
+        """
+        self.sys_jobs.current.reference = slug
+
+        if slug not in self.local:
+            raise AddonsError(f"Add-on {slug} is not installed", _LOGGER.error)
+        addon = self.local[slug]
+
+        if addon.is_detached:
+            raise AddonsError(
+                f"Add-on {slug} is not available inside store", _LOGGER.error
+            )
+        store = self.store[slug]
+
+        # Check if a rebuild is possible now
+        if addon.version != store.version:
+            raise AddonsError(
+                "Version changed, use Update instead Rebuild", _LOGGER.error
+            )
+        if not addon.need_build:
+            raise AddonsNotSupportedError(
+                "Can't rebuild a image based add-on", _LOGGER.error
+            )
+
+        return await addon.rebuild()
+
+    @Job(
+        name="addon_manager_restore",
+        conditions=[
+            JobCondition.FREE_SPACE,
+            JobCondition.INTERNET_HOST,
+            JobCondition.HEALTHY,
+        ],
+        on_condition=AddonsJobError,
+    )
+    async def restore(
+        self, slug: str, tar_file: tarfile.TarFile
+    ) -> asyncio.Task | None:
+        """Restore state of an add-on.
+
+        Returns a Task that completes when addon has state 'started' (see addon.start)
+        if addon is started after restore. Else nothing is returned.
+        """
+        self.sys_jobs.current.reference = slug
+
+        if slug not in self.local:
+            _LOGGER.debug("Add-on %s is not local available for restore", slug)
+            addon = Addon(self.coresys, slug)
+            had_ingress = False
+        else:
+            _LOGGER.debug("Add-on %s is local available for restore", slug)
+            addon = self.local[slug]
+            had_ingress = addon.ingress_panel
+
+        wait_for_start = await addon.restore(tar_file)
+
+        # Check if new
+        if slug not in self.local:
+            _LOGGER.info("Detect new Add-on after restore %s", slug)
+            self.local[slug] = addon
+
+        # Update ingress
+        if had_ingress != addon.ingress_panel:
+            await self.sys_ingress.reload()
+            with suppress(HomeAssistantAPIError):
+                await self.sys_ingress.update_hass_panel(addon)
+
+        return wait_for_start
+
+    @Job(
+        name="addon_manager_repair",
+        conditions=[JobCondition.FREE_SPACE, JobCondition.INTERNET_HOST],
+    )
+    async def repair(self) -> None:
+        """Repair local add-ons."""
+        needs_repair: list[Addon] = []
+
+        # Evaluate Add-ons to repair
+        for addon in self.installed:
+            if await addon.instance.exists():
+                continue
+            needs_repair.append(addon)
+
+        _LOGGER.info("Found %d add-ons to repair", len(needs_repair))
+        if not needs_repair:
+            return
+
+        for addon in needs_repair:
+            _LOGGER.info("Repairing for add-on: %s", addon.slug)
+            with suppress(DockerError, KeyError):
+                # Need pull a image again
+                if not addon.need_build:
+                    await addon.instance.install(addon.version, addon.image)
+                    continue
+
+                # Need local lookup
+                if addon.need_build and not addon.is_detached:
+                    store = self.store[addon.slug]
+                    # If this add-on is available for rebuild
+                    if addon.version == store.version:
+                        await addon.instance.install(addon.version, addon.image)
+                        continue
+
+            _LOGGER.error("Can't repair %s", addon.slug)
+            with suppress(AddonsError):
+                await self.uninstall(addon.slug)
+
+    async def sync_dns(self) -> None:
+        """Sync add-ons DNS names."""
+        # Update hosts
+        add_host_coros: list[Awaitable[None]] = []
+        for addon in self.installed:
+            try:
+                if not await addon.instance.is_running():
+                    continue
+            except DockerError as err:
+                _LOGGER.warning("Add-on %s is corrupt: %s", addon.slug, err)
+                self.sys_resolution.create_issue(
+                    IssueType.CORRUPT_DOCKER,
+                    ContextType.ADDON,
+                    reference=addon.slug,
+                    suggestions=[SuggestionType.EXECUTE_REPAIR],
+                )
+                capture_exception(err)
+            else:
+                add_host_coros.append(
+                    self.sys_plugins.dns.add_host(
+                        ipv4=addon.ip_address, names=[addon.hostname], write=False
+                    )
+                )
+
+        await asyncio.gather(*add_host_coros)
+
+        # Write hosts files
+        with suppress(CoreDNSError):
+            await self.sys_plugins.dns.write_hosts()

supervisor/addons/model.py

@@ -1,7 +1,7 @@
 """Init file for Supervisor add-ons."""
 from abc import ABC, abstractmethod
 from collections import defaultdict
-from collections.abc import Awaitable, Callable
+from collections.abc import Callable
 from contextlib import suppress
 import logging
 from pathlib import Path
@@ -65,6 +65,7 @@ from ..const import (
     ATTR_TIMEOUT,
     ATTR_TMPFS,
     ATTR_TRANSLATIONS,
+    ATTR_TYPE,
     ATTR_UART,
     ATTR_UDEV,
     ATTR_URL,
@@ -85,9 +86,18 @@ from ..docker.const import Capabilities
 from ..exceptions import AddonsNotSupportedError
 from ..jobs.const import JOB_GROUP_ADDON
 from ..jobs.job_group import JobGroup
-from .const import ATTR_BACKUP, ATTR_CODENOTARY, AddonBackupMode
+from ..utils import version_is_new_enough
+from .configuration import FolderMapping
+from .const import (
+    ATTR_BACKUP,
+    ATTR_CODENOTARY,
+    ATTR_PATH,
+    ATTR_READ_ONLY,
+    AddonBackupMode,
+    MappingType,
+)
 from .options import AddonOptions, UiOptions
-from .validate import RE_SERVICE, RE_VOLUME
+from .validate import RE_SERVICE
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
 
@@ -537,14 +547,13 @@ class AddonModel(JobGroup, ABC):
         return ATTR_IMAGE not in self.data
 
     @property
-    def map_volumes(self) -> dict[str, bool]:
-        """Return a dict of {volume: read-only} from add-on."""
+    def map_volumes(self) -> dict[MappingType, FolderMapping]:
+        """Return a dict of {MappingType: FolderMapping} from add-on."""
         volumes = {}
         for volume in self.data[ATTR_MAP]:
-            result = RE_VOLUME.match(volume)
-            if not result:
-                continue
-            volumes[result.group(1)] = result.group(2) != "rw"
+            volumes[MappingType(volume[ATTR_TYPE])] = FolderMapping(
+                volume.get(ATTR_PATH), volume[ATTR_READ_ONLY]
+            )
 
         return volumes
 
@@ -645,7 +654,9 @@ class AddonModel(JobGroup, ABC):
         # Home Assistant
         version: AwesomeVersion | None = config.get(ATTR_HOMEASSISTANT)
         with suppress(AwesomeVersionException, TypeError):
-            if self.sys_homeassistant.version < version:
+            if version and not version_is_new_enough(
+                self.sys_homeassistant.version, version
+            ):
                 raise AddonsNotSupportedError(
                     f"Add-on {self.slug} not supported on this system, requires Home Assistant version {version} or greater",
                     logger,
@@ -669,19 +680,3 @@ class AddonModel(JobGroup, ABC):
 
         # local build
         return f"{config[ATTR_REPOSITORY]}/{self.sys_arch.default}-addon-{config[ATTR_SLUG]}"
-
-    def install(self) -> Awaitable[None]:
-        """Install this add-on."""
-        return self.sys_addons.install(self.slug)
-
-    def uninstall(self) -> Awaitable[None]:
-        """Uninstall this add-on."""
-        return self.sys_addons.uninstall(self.slug)
-
-    def update(self, backup: bool | None = False) -> Awaitable[Awaitable[None] | None]:
-        """Update this add-on."""
-        return self.sys_addons.update(self.slug, backup=backup)
-
-    def rebuild(self) -> Awaitable[Awaitable[None] | None]:
-        """Rebuild this add-on."""
-        return self.sys_addons.rebuild(self.slug)

supervisor/addons/validate.py

@@ -81,6 +81,7 @@ from ..const import (
     ATTR_TIMEOUT,
     ATTR_TMPFS,
     ATTR_TRANSLATIONS,
+    ATTR_TYPE,
     ATTR_UART,
     ATTR_UDEV,
     ATTR_URL,
@@ -109,12 +110,22 @@ from ..validate import (
     uuid_match,
     version_tag,
 )
-from .const import ATTR_BACKUP, ATTR_CODENOTARY, RE_SLUG, AddonBackupMode
+from .const import (
+    ATTR_BACKUP,
+    ATTR_CODENOTARY,
+    ATTR_PATH,
+    ATTR_READ_ONLY,
+    RE_SLUG,
+    AddonBackupMode,
+    MappingType,
+)
 from .options import RE_SCHEMA_ELEMENT
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
 
-RE_VOLUME = re.compile(r"^(config|ssl|addons|backup|share|media)(?::(rw|ro))?$")
+RE_VOLUME = re.compile(
+    r"^(data|config|ssl|addons|backup|share|media|homeassistant_config|all_addon_configs|addon_config)(?::(rw|ro))?$"
+)
 RE_SERVICE = re.compile(r"^(?P<service>mqtt|mysql):(?P<rights>provide|want|need)$")
 
 
@@ -143,6 +154,7 @@ RE_MACHINE = re.compile(
     r"|raspberrypi3"
     r"|raspberrypi4-64"
     r"|raspberrypi4"
+    r"|raspberrypi5-64"
     r"|yellow"
     r"|green"
     r"|tinker"
@@ -260,6 +272,48 @@ def _migrate_addon_config(protocol=False):
                     name,
                 )
 
+        # 2023-11 "map" entries can also be dict to allow path configuration
+        volumes = []
+        for entry in config.get(ATTR_MAP, []):
+            if isinstance(entry, dict):
+                volumes.append(entry)
+            if isinstance(entry, str):
+                result = RE_VOLUME.match(entry)
+                if not result:
+                    continue
+                volumes.append(
+                    {
+                        ATTR_TYPE: result.group(1),
+                        ATTR_READ_ONLY: result.group(2) != "rw",
+                    }
+                )
+
+        if volumes:
+            config[ATTR_MAP] = volumes
+
+        # 2023-10 "config" became "homeassistant" so /config can be used for addon's public config
+        if any(volume[ATTR_TYPE] == MappingType.CONFIG for volume in volumes):
+            if any(
+                volume
+                and volume[ATTR_TYPE]
+                in {MappingType.ADDON_CONFIG, MappingType.HOMEASSISTANT_CONFIG}
+                for volume in volumes
+            ):
+                _LOGGER.warning(
+                    "Add-on config using incompatible map options, '%s' and '%s' are ignored if '%s' is included. Please report this to the maintainer of %s",
+                    MappingType.ADDON_CONFIG,
+                    MappingType.HOMEASSISTANT_CONFIG,
+                    MappingType.CONFIG,
+                    name,
+                )
+            else:
+                _LOGGER.debug(
+                    "Add-on config using deprecated map option '%s' instead of '%s'. Please report this to the maintainer of %s",
+                    MappingType.CONFIG,
+                    MappingType.HOMEASSISTANT_CONFIG,
+                    name,
+                )
+
         return config
 
     return _migrate
@@ -308,7 +362,15 @@ _SCHEMA_ADDON_CONFIG = vol.Schema(
         vol.Optional(ATTR_DEVICES): [str],
         vol.Optional(ATTR_UDEV, default=False): vol.Boolean(),
         vol.Optional(ATTR_TMPFS, default=False): vol.Boolean(),
-        vol.Optional(ATTR_MAP, default=list): [vol.Match(RE_VOLUME)],
+        vol.Optional(ATTR_MAP, default=list): [
+            vol.Schema(
+                {
+                    vol.Required(ATTR_TYPE): vol.Coerce(MappingType),
+                    vol.Optional(ATTR_READ_ONLY, default=True): bool,
+                    vol.Optional(ATTR_PATH): str,
+                }
+            )
+        ],
         vol.Optional(ATTR_ENVIRONMENT): {vol.Match(r"\w*"): str},
         vol.Optional(ATTR_PRIVILEGED): [vol.Coerce(Capabilities)],
         vol.Optional(ATTR_APPARMOR, default=True): vol.Boolean(),

supervisor/api/__init__.py

@@ -5,6 +5,7 @@ from pathlib import Path
 from typing import Any
 
 from aiohttp import web
+from aiohttp_fast_url_dispatcher import FastUrlDispatcher, attach_fast_url_dispatcher
 
 from ..const import AddonState
 from ..coresys import CoreSys, CoreSysAttributes
@@ -64,9 +65,10 @@ class RestAPI(CoreSysAttributes):
                 "max_field_size": MAX_LINE_SIZE,
             },
         )
+        attach_fast_url_dispatcher(self.webapp, FastUrlDispatcher())
 
         # service stuff
-        self._runner: web.AppRunner = web.AppRunner(self.webapp)
+        self._runner: web.AppRunner = web.AppRunner(self.webapp, shutdown_timeout=5)
         self._site: web.TCPSite | None = None
 
     async def load(self) -> None:
@@ -186,6 +188,8 @@ class RestAPI(CoreSysAttributes):
         # Boards endpoints
         self.webapp.add_routes(
             [
+                web.get("/os/boards/green", api_os.boards_green_info),
+                web.post("/os/boards/green", api_os.boards_green_options),
                 web.get("/os/boards/yellow", api_os.boards_yellow_info),
                 web.post("/os/boards/yellow", api_os.boards_yellow_options),
                 web.get("/os/boards/{board}", api_os.boards_other_info),
@@ -485,6 +489,8 @@ class RestAPI(CoreSysAttributes):
                 web.get("/backups/info", api_backups.info),
                 web.post("/backups/options", api_backups.options),
                 web.post("/backups/reload", api_backups.reload),
+                web.post("/backups/freeze", api_backups.freeze),
+                web.post("/backups/thaw", api_backups.thaw),
                 web.post("/backups/new/full", api_backups.backup_full),
                 web.post("/backups/new/partial", api_backups.backup_partial),
                 web.post("/backups/new/upload", api_backups.upload),
@@ -667,9 +673,7 @@ class RestAPI(CoreSysAttributes):
     async def start(self) -> None:
         """Run RESTful API webserver."""
         await self._runner.setup()
-        self._site = web.TCPSite(
-            self._runner, host="0.0.0.0", port=80, shutdown_timeout=5
-        )
+        self._site = web.TCPSite(self._runner, host="0.0.0.0", port=80)
 
         try:
             await self._site.start()

supervisor/api/addons.py

@@ -8,8 +8,8 @@ from aiohttp import web
 import voluptuous as vol
 from voluptuous.humanize import humanize_error
 
-from ..addons import AnyAddon
 from ..addons.addon import Addon
+from ..addons.manager import AnyAddon
 from ..addons.utils import rating_security
 from ..const import (
     ATTR_ADDONS,
@@ -388,7 +388,7 @@ class APIAddons(CoreSysAttributes):
     def uninstall(self, request: web.Request) -> Awaitable[None]:
         """Uninstall add-on."""
         addon = self._extract_addon(request)
-        return asyncio.shield(addon.uninstall())
+        return asyncio.shield(self.sys_addons.uninstall(addon.slug))
 
     @api_process
     async def start(self, request: web.Request) -> None:
@@ -414,7 +414,7 @@ class APIAddons(CoreSysAttributes):
     async def rebuild(self, request: web.Request) -> None:
         """Rebuild local build add-on."""
         addon = self._extract_addon(request)
-        if start_task := await asyncio.shield(addon.rebuild()):
+        if start_task := await asyncio.shield(self.sys_addons.rebuild(addon.slug)):
             await start_task
 
     @api_process_raw(CONTENT_TYPE_BINARY)

supervisor/api/auth.py

@@ -11,6 +11,7 @@ from ..addons.addon import Addon
 from ..const import ATTR_PASSWORD, ATTR_USERNAME, REQUEST_FROM
 from ..coresys import CoreSysAttributes
 from ..exceptions import APIForbidden
+from ..utils.json import json_loads
 from .const import CONTENT_TYPE_JSON, CONTENT_TYPE_URL
 from .utils import api_process, api_validate
 
@@ -67,7 +68,7 @@ class APIAuth(CoreSysAttributes):
 
         # Json
         if request.headers.get(CONTENT_TYPE) == CONTENT_TYPE_JSON:
-            data = await request.json()
+            data = await request.json(loads=json_loads)
             return await self._process_dict(request, addon, data)
 
         # URL encoded

supervisor/api/backups.py

@@ -1,5 +1,6 @@
 """Backups RESTful API."""
 import asyncio
+import errno
 import logging
 from pathlib import Path
 import re
@@ -20,6 +21,7 @@ from ..const import (
     ATTR_DAYS_UNTIL_STALE,
     ATTR_FOLDERS,
     ATTR_HOMEASSISTANT,
+    ATTR_HOMEASSISTANT_EXCLUDE_DATABASE,
     ATTR_LOCATON,
     ATTR_NAME,
     ATTR_PASSWORD,
@@ -28,12 +30,14 @@ from ..const import (
     ATTR_SIZE,
     ATTR_SLUG,
     ATTR_SUPERVISOR_VERSION,
+    ATTR_TIMEOUT,
     ATTR_TYPE,
     ATTR_VERSION,
 )
 from ..coresys import CoreSysAttributes
 from ..exceptions import APIError
 from ..mounts.const import MountUsage
+from ..resolution.const import UnhealthyReason
 from .const import CONTENT_TYPE_TAR
 from .utils import api_process, api_validate
 
@@ -63,6 +67,7 @@ SCHEMA_BACKUP_FULL = vol.Schema(
         vol.Optional(ATTR_PASSWORD): vol.Maybe(str),
         vol.Optional(ATTR_COMPRESSED): vol.Maybe(vol.Boolean()),
         vol.Optional(ATTR_LOCATON): vol.Maybe(str),
+        vol.Optional(ATTR_HOMEASSISTANT_EXCLUDE_DATABASE): vol.Boolean(),
     }
 )
 
@@ -80,6 +85,12 @@ SCHEMA_OPTIONS = vol.Schema(
     }
 )
 
+SCHEMA_FREEZE = vol.Schema(
+    {
+        vol.Optional(ATTR_TIMEOUT): vol.All(int, vol.Range(min=1)),
+    }
+)
+
 
 class APIBackups(CoreSysAttributes):
     """Handle RESTful API for backups functions."""
@@ -142,7 +153,7 @@ class APIBackups(CoreSysAttributes):
         self.sys_backups.save_data()
 
     @api_process
-    async def reload(self, request):
+    async def reload(self, _):
         """Reload backup list."""
         await asyncio.shield(self.sys_backups.reload())
         return True
@@ -177,6 +188,7 @@ class APIBackups(CoreSysAttributes):
             ATTR_ADDONS: data_addons,
             ATTR_REPOSITORIES: backup.repositories,
             ATTR_FOLDERS: backup.folders,
+            ATTR_HOMEASSISTANT_EXCLUDE_DATABASE: backup.homeassistant_exclude_database,
         }
 
     def _location_to_mount(self, body: dict[str, Any]) -> dict[str, Any]:
@@ -233,6 +245,17 @@ class APIBackups(CoreSysAttributes):
 
         return await asyncio.shield(self.sys_backups.do_restore_partial(backup, **body))
 
+    @api_process
+    async def freeze(self, request):
+        """Initiate manual freeze for external backup."""
+        body = await api_validate(SCHEMA_FREEZE, request)
+        await asyncio.shield(self.sys_backups.freeze_all(**body))
+
+    @api_process
+    async def thaw(self, request):
+        """Begin thaw after manual freeze."""
+        await self.sys_backups.thaw_all()
+
     @api_process
     async def remove(self, request):
         """Remove a backup."""
@@ -267,6 +290,8 @@ class APIBackups(CoreSysAttributes):
                         backup.write(chunk)
 
             except OSError as err:
+                if err.errno == errno.EBADMSG:
+                    self.sys_resolution.unhealthy = UnhealthyReason.OSERROR_BAD_MESSAGE
                 _LOGGER.error("Can't write new backup file: %s", err)
                 return False
 

supervisor/api/const.py

@@ -23,7 +23,6 @@ ATTR_CONNECTION_BUS = "connection_bus"
 ATTR_DATA_DISK = "data_disk"
 ATTR_DEVICE = "device"
 ATTR_DEV_PATH = "dev_path"
-ATTR_DISK_LED = "disk_led"
 ATTR_DISKS = "disks"
 ATTR_DRIVES = "drives"
 ATTR_DT_SYNCHRONIZED = "dt_synchronized"
@@ -31,7 +30,6 @@ ATTR_DT_UTC = "dt_utc"
 ATTR_EJECTABLE = "ejectable"
 ATTR_FALLBACK = "fallback"
 ATTR_FILESYSTEMS = "filesystems"
-ATTR_HEARTBEAT_LED = "heartbeat_led"
 ATTR_IDENTIFIERS = "identifiers"
 ATTR_JOBS = "jobs"
 ATTR_LLMNR = "llmnr"
@@ -41,7 +39,6 @@ ATTR_MODEL = "model"
 ATTR_MOUNTS = "mounts"
 ATTR_MOUNT_POINTS = "mount_points"
 ATTR_PANEL_PATH = "panel_path"
-ATTR_POWER_LED = "power_led"
 ATTR_REMOVABLE = "removable"
 ATTR_REVISION = "revision"
 ATTR_SEAT = "seat"
@@ -49,6 +46,7 @@ ATTR_SIGNED = "signed"
 ATTR_STARTUP_TIME = "startup_time"
 ATTR_SUBSYSTEM = "subsystem"
 ATTR_SYSFS = "sysfs"
+ATTR_SYSTEM_HEALTH_LED = "system_health_led"
 ATTR_TIME_DETECTED = "time_detected"
 ATTR_UPDATE_TYPE = "update_type"
 ATTR_USE_NTP = "use_ntp"

supervisor/api/homeassistant.py

@@ -12,6 +12,7 @@ from ..const import (
     ATTR_AUDIO_INPUT,
     ATTR_AUDIO_OUTPUT,
     ATTR_BACKUP,
+    ATTR_BACKUPS_EXCLUDE_DATABASE,
     ATTR_BLK_READ,
     ATTR_BLK_WRITE,
     ATTR_BOOT,
@@ -51,6 +52,7 @@ SCHEMA_OPTIONS = vol.Schema(
         vol.Optional(ATTR_REFRESH_TOKEN): vol.Maybe(str),
         vol.Optional(ATTR_AUDIO_OUTPUT): vol.Maybe(str),
         vol.Optional(ATTR_AUDIO_INPUT): vol.Maybe(str),
+        vol.Optional(ATTR_BACKUPS_EXCLUDE_DATABASE): vol.Boolean(),
     }
 )
 
@@ -82,6 +84,7 @@ class APIHomeAssistant(CoreSysAttributes):
             ATTR_WATCHDOG: self.sys_homeassistant.watchdog,
             ATTR_AUDIO_INPUT: self.sys_homeassistant.audio_input,
             ATTR_AUDIO_OUTPUT: self.sys_homeassistant.audio_output,
+            ATTR_BACKUPS_EXCLUDE_DATABASE: self.sys_homeassistant.backups_exclude_database,
         }
 
     @api_process
@@ -113,6 +116,11 @@ class APIHomeAssistant(CoreSysAttributes):
         if ATTR_AUDIO_OUTPUT in body:
             self.sys_homeassistant.audio_output = body[ATTR_AUDIO_OUTPUT]
 
+        if ATTR_BACKUPS_EXCLUDE_DATABASE in body:
+            self.sys_homeassistant.backups_exclude_database = body[
+                ATTR_BACKUPS_EXCLUDE_DATABASE
+            ]
+
         self.sys_homeassistant.save_data()
 
     @api_process

supervisor/api/ingress.py

@@ -48,6 +48,29 @@ SCHEMA_INGRESS_CREATE_SESSION_DATA = vol.Schema(
 )
 
 
+# from https://github.com/aio-libs/aiohttp/blob/8ae650bee4add9f131d49b96a0a150311ea58cd1/aiohttp/helpers.py#L1059C1-L1079C1
+def must_be_empty_body(method: str, code: int) -> bool:
+    """Check if a request must return an empty body."""
+    return (
+        status_code_must_be_empty_body(code)
+        or method_must_be_empty_body(method)
+        or (200 <= code < 300 and method.upper() == hdrs.METH_CONNECT)
+    )
+
+
+def method_must_be_empty_body(method: str) -> bool:
+    """Check if a method must return an empty body."""
+    # https://datatracker.ietf.org/doc/html/rfc9112#section-6.3-2.1
+    # https://datatracker.ietf.org/doc/html/rfc9112#section-6.3-2.2
+    return method.upper() == hdrs.METH_HEAD
+
+
+def status_code_must_be_empty_body(code: int) -> bool:
+    """Check if a status code must return an empty body."""
+    # https://datatracker.ietf.org/doc/html/rfc9112#section-6.3-2.1
+    return code in {204, 304} or 100 <= code < 200
+
+
 class APIIngress(CoreSysAttributes):
     """Ingress view to handle add-on webui routing."""
 
@@ -225,10 +248,18 @@ class APIIngress(CoreSysAttributes):
             skip_auto_headers={hdrs.CONTENT_TYPE},
         ) as result:
             headers = _response_header(result)
-
+            # Avoid parsing content_type in simple cases for better performance
+            if maybe_content_type := result.headers.get(hdrs.CONTENT_TYPE):
+                content_type = (maybe_content_type.partition(";"))[0].strip()
+            else:
+                content_type = result.content_type
             # Simple request
             if (
-                hdrs.CONTENT_LENGTH in result.headers
+                # empty body responses should not be streamed,
+                # otherwise aiohttp < 3.9.0 may generate
+                # an invalid "0\r\n\r\n" chunk instead of an empty response.
+                must_be_empty_body(request.method, result.status)
+                or hdrs.CONTENT_LENGTH in result.headers
                 and int(result.headers.get(hdrs.CONTENT_LENGTH, 0)) < 4_194_000
             ):
                 # Return Response
@@ -236,13 +267,13 @@ class APIIngress(CoreSysAttributes):
                 return web.Response(
                     headers=headers,
                     status=result.status,
-                    content_type=result.content_type,
+                    content_type=content_type,
                     body=body,
                 )
 
             # Stream response
             response = web.StreamResponse(status=result.status, headers=headers)
-            response.content_type = result.content_type
+            response.content_type = content_type
 
             try:
                 await response.prepare(request)
@@ -282,8 +313,10 @@ def _init_header(
 
     if session_data is not None:
         headers[HEADER_REMOTE_USER_ID] = session_data.user.id
-        headers[HEADER_REMOTE_USER_NAME] = session_data.user.username
-        headers[HEADER_REMOTE_USER_DISPLAY_NAME] = session_data.user.display_name
+        if session_data.user.username is not None:
+            headers[HEADER_REMOTE_USER_NAME] = session_data.user.username
+        if session_data.user.display_name is not None:
+            headers[HEADER_REMOTE_USER_DISPLAY_NAME] = session_data.user.display_name
 
     # filter flags
     for name, value in request.headers.items():

supervisor/api/middleware/security.py

@@ -19,6 +19,7 @@ from ...const import (
     CoreState,
 )
 from ...coresys import CoreSys, CoreSysAttributes
+from ...utils import version_is_new_enough
 from ..utils import api_return_error, excract_supervisor_token
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
@@ -273,9 +274,8 @@ class SecurityMiddleware(CoreSysAttributes):
     @middleware
     async def core_proxy(self, request: Request, handler: RequestHandler) -> Response:
         """Validate user from Core API proxy."""
-        if (
-            request[REQUEST_FROM] != self.sys_homeassistant
-            or self.sys_homeassistant.version >= _CORE_VERSION
+        if request[REQUEST_FROM] != self.sys_homeassistant or version_is_new_enough(
+            self.sys_homeassistant.version, _CORE_VERSION
         ):
             return await handler(request)
 

supervisor/api/os.py

@@ -8,11 +8,15 @@ from aiohttp import web
 import voluptuous as vol
 
 from ..const import (
+    ATTR_ACTIVITY_LED,
     ATTR_BOARD,
     ATTR_BOOT,
     ATTR_DEVICES,
+    ATTR_DISK_LED,
+    ATTR_HEARTBEAT_LED,
     ATTR_ID,
     ATTR_NAME,
+    ATTR_POWER_LED,
     ATTR_SERIAL,
     ATTR_SIZE,
     ATTR_UPDATE_AVAILABLE,
@@ -27,21 +31,19 @@ from .const import (
     ATTR_DATA_DISK,
     ATTR_DEV_PATH,
     ATTR_DEVICE,
-    ATTR_DISK_LED,
     ATTR_DISKS,
-    ATTR_HEARTBEAT_LED,
     ATTR_MODEL,
-    ATTR_POWER_LED,
+    ATTR_SYSTEM_HEALTH_LED,
     ATTR_VENDOR,
 )
 from .utils import api_process, api_validate
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
 
+# pylint: disable=no-value-for-parameter
 SCHEMA_VERSION = vol.Schema({vol.Optional(ATTR_VERSION): version_tag})
 SCHEMA_DISK = vol.Schema({vol.Required(ATTR_DEVICE): str})
 
-# pylint: disable=no-value-for-parameter
 SCHEMA_YELLOW_OPTIONS = vol.Schema(
     {
         vol.Optional(ATTR_DISK_LED): vol.Boolean(),
@@ -49,6 +51,14 @@ SCHEMA_YELLOW_OPTIONS = vol.Schema(
         vol.Optional(ATTR_POWER_LED): vol.Boolean(),
     }
 )
+SCHEMA_GREEN_OPTIONS = vol.Schema(
+    {
+        vol.Optional(ATTR_ACTIVITY_LED): vol.Boolean(),
+        vol.Optional(ATTR_POWER_LED): vol.Boolean(),
+        vol.Optional(ATTR_SYSTEM_HEALTH_LED): vol.Boolean(),
+    }
+)
+# pylint: enable=no-value-for-parameter
 
 
 class APIOS(CoreSysAttributes):
@@ -105,6 +115,31 @@ class APIOS(CoreSysAttributes):
             ],
         }
 
+    @api_process
+    async def boards_green_info(self, request: web.Request) -> dict[str, Any]:
+        """Get green board settings."""
+        return {
+            ATTR_ACTIVITY_LED: self.sys_dbus.agent.board.green.activity_led,
+            ATTR_POWER_LED: self.sys_dbus.agent.board.green.power_led,
+            ATTR_SYSTEM_HEALTH_LED: self.sys_dbus.agent.board.green.user_led,
+        }
+
+    @api_process
+    async def boards_green_options(self, request: web.Request) -> None:
+        """Update green board settings."""
+        body = await api_validate(SCHEMA_GREEN_OPTIONS, request)
+
+        if ATTR_ACTIVITY_LED in body:
+            self.sys_dbus.agent.board.green.activity_led = body[ATTR_ACTIVITY_LED]
+
+        if ATTR_POWER_LED in body:
+            self.sys_dbus.agent.board.green.power_led = body[ATTR_POWER_LED]
+
+        if ATTR_SYSTEM_HEALTH_LED in body:
+            self.sys_dbus.agent.board.green.user_led = body[ATTR_SYSTEM_HEALTH_LED]
+
+        self.sys_dbus.agent.board.green.save_data()
+
     @api_process
     async def boards_yellow_info(self, request: web.Request) -> dict[str, Any]:
         """Get yellow board settings."""
@@ -128,6 +163,7 @@ class APIOS(CoreSysAttributes):
         if ATTR_POWER_LED in body:
             self.sys_dbus.agent.board.yellow.power_led = body[ATTR_POWER_LED]
 
+        self.sys_dbus.agent.board.yellow.save_data()
         self.sys_resolution.create_issue(
             IssueType.REBOOT_REQUIRED,
             ContextType.SYSTEM,

supervisor/api/proxy.py

@@ -6,7 +6,10 @@ import logging
 import aiohttp
 from aiohttp import web
 from aiohttp.client_exceptions import ClientConnectorError
+from aiohttp.client_ws import ClientWebSocketResponse
 from aiohttp.hdrs import AUTHORIZATION, CONTENT_TYPE
+from aiohttp.http import WSMessage
+from aiohttp.http_websocket import WSMsgType
 from aiohttp.web_exceptions import HTTPBadGateway, HTTPUnauthorized
 
 from ..coresys import CoreSysAttributes
@@ -74,7 +77,7 @@ class APIProxy(CoreSysAttributes):
             _LOGGER.error("Error on API for request %s", path)
         except aiohttp.ClientError as err:
             _LOGGER.error("Client error on API %s request %s", path, err)
-        except asyncio.TimeoutError:
+        except TimeoutError:
             _LOGGER.error("Client timeout error on API request %s", path)
 
         raise HTTPBadGateway()
@@ -114,7 +117,7 @@ class APIProxy(CoreSysAttributes):
                 body=data, status=client.status, content_type=client.content_type
             )
 
-    async def _websocket_client(self):
+    async def _websocket_client(self) -> ClientWebSocketResponse:
         """Initialize a WebSocket API connection."""
         url = f"{self.sys_homeassistant.api_url}/api/websocket"
 
@@ -167,6 +170,25 @@ class APIProxy(CoreSysAttributes):
 
         raise APIError()
 
+    async def _proxy_message(
+        self,
+        read_task: asyncio.Task,
+        target: web.WebSocketResponse | ClientWebSocketResponse,
+    ) -> None:
+        """Proxy a message from client to server or vice versa."""
+        if read_task.exception():
+            raise read_task.exception()
+
+        msg: WSMessage = read_task.result()
+        if msg.type == WSMsgType.TEXT:
+            return await target.send_str(msg.data)
+        if msg.type == WSMsgType.BINARY:
+            return await target.send_bytes(msg.data)
+
+        raise TypeError(
+            f"Cannot proxy websocket message of unsupported type: {msg.type}"
+        )
+
     async def websocket(self, request: web.Request):
         """Initialize a WebSocket API connection."""
         if not await self.sys_homeassistant.api.check_api_state():
@@ -214,13 +236,13 @@ class APIProxy(CoreSysAttributes):
 
         _LOGGER.info("Home Assistant WebSocket API request running")
         try:
-            client_read = None
-            server_read = None
+            client_read: asyncio.Task | None = None
+            server_read: asyncio.Task | None = None
             while not server.closed and not client.closed:
                 if not client_read:
-                    client_read = self.sys_create_task(client.receive_str())
+                    client_read = self.sys_create_task(client.receive())
                 if not server_read:
-                    server_read = self.sys_create_task(server.receive_str())
+                    server_read = self.sys_create_task(server.receive())
 
                 # wait until data need to be processed
                 await asyncio.wait(
@@ -229,14 +251,12 @@ class APIProxy(CoreSysAttributes):
 
                 # server
                 if server_read.done() and not client.closed:
-                    server_read.exception()
-                    await client.send_str(server_read.result())
+                    await self._proxy_message(server_read, client)
                     server_read = None
 
                 # client
                 if client_read.done() and not server.closed:
-                    client_read.exception()
-                    await server.send_str(client_read.result())
+                    await self._proxy_message(client_read, server)
                     client_read = None
 
         except asyncio.CancelledError:
@@ -246,9 +266,9 @@ class APIProxy(CoreSysAttributes):
             _LOGGER.info("Home Assistant WebSocket API error: %s", err)
 
         finally:
-            if client_read:
+            if client_read and not client_read.done():
                 client_read.cancel()
-            if server_read:
+            if server_read and not server_read.done():
                 server_read.cancel()
 
             # close connections

supervisor/api/store.py

@@ -6,7 +6,7 @@ from typing import Any
 from aiohttp import web
 import voluptuous as vol
 
-from ..addons import AnyAddon
+from ..addons.manager import AnyAddon
 from ..addons.utils import rating_security
 from ..api.const import ATTR_SIGNED
 from ..api.utils import api_process, api_process_raw, api_validate
@@ -186,18 +186,20 @@ class APIStore(CoreSysAttributes):
         }
 
     @api_process
-    async def addons_list(self, request: web.Request) -> list[dict[str, Any]]:
+    async def addons_list(self, request: web.Request) -> dict[str, Any]:
         """Return all store add-ons."""
-        return [
-            self._generate_addon_information(self.sys_addons.store[addon])
-            for addon in self.sys_addons.store
-        ]
+        return {
+            ATTR_ADDONS: [
+                self._generate_addon_information(self.sys_addons.store[addon])
+                for addon in self.sys_addons.store
+            ]
+        }
 
     @api_process
     def addons_addon_install(self, request: web.Request) -> Awaitable[None]:
         """Install add-on."""
         addon = self._extract_addon(request)
-        return asyncio.shield(addon.install())
+        return asyncio.shield(self.sys_addons.install(addon.slug))
 
     @api_process
     async def addons_addon_update(self, request: web.Request) -> None:
@@ -209,7 +211,7 @@ class APIStore(CoreSysAttributes):
         body = await api_validate(SCHEMA_UPDATE, request)
 
         if start_task := await asyncio.shield(
-            addon.update(backup=body.get(ATTR_BACKUP))
+            self.sys_addons.update(addon.slug, backup=body.get(ATTR_BACKUP))
         ):
             await start_task
 

supervisor/api/utils.py

@@ -22,7 +22,7 @@ from ..const import (
 from ..coresys import CoreSys
 from ..exceptions import APIError, APIForbidden, DockerAPIError, HassioError
 from ..utils import check_exception_chain, get_message_from_exception_chain
-from ..utils.json import JSONEncoder
+from ..utils.json import json_dumps, json_loads as json_loads_util
 from ..utils.log_format import format_message
 from .const import CONTENT_TYPE_BINARY
 
@@ -48,7 +48,7 @@ def json_loads(data: Any) -> dict[str, Any]:
     if not data:
         return {}
     try:
-        return json.loads(data)
+        return json_loads_util(data)
     except json.JSONDecodeError as err:
         raise APIError("Invalid json") from err
 
@@ -130,7 +130,7 @@ def api_return_error(
             JSON_MESSAGE: message or "Unknown error, see supervisor",
         },
         status=400,
-        dumps=lambda x: json.dumps(x, cls=JSONEncoder),
+        dumps=json_dumps,
     )
 
 
@@ -138,7 +138,7 @@ def api_return_ok(data: dict[str, Any] | None = None) -> web.Response:
     """Return an API ok answer."""
     return web.json_response(
         {JSON_RESULT: RESULT_OK, JSON_DATA: data or {}},
-        dumps=lambda x: json.dumps(x, cls=JSONEncoder),
+        dumps=json_dumps,
     )
 
 

supervisor/arch.py

@@ -28,6 +28,7 @@ class CpuArch(CoreSysAttributes):
         """Initialize CPU Architecture handler."""
         self.coresys = coresys
         self._supported_arch: list[str] = []
+        self._supported_set: set[str] = set()
         self._default_arch: str
 
     @property
@@ -70,9 +71,11 @@ class CpuArch(CoreSysAttributes):
         if native_support not in self._supported_arch:
             self._supported_arch.append(native_support)
 
+        self._supported_set = set(self._supported_arch)
+
     def is_supported(self, arch_list: list[str]) -> bool:
         """Return True if there is a supported arch by this platform."""
-        return not set(self.supported).isdisjoint(set(arch_list))
+        return not self._supported_set.isdisjoint(arch_list)
 
     def match(self, arch_list: list[str]) -> str:
         """Return best match for this CPU/Platform."""

supervisor/backups/backup.py

@@ -1,4 +1,5 @@
 """Representation of a backup file."""
+import asyncio
 from base64 import b64decode, b64encode
 from collections.abc import Awaitable
 from datetime import timedelta
@@ -18,13 +19,14 @@ from securetar import SecureTarFile, atomic_contents_add, secure_path
 import voluptuous as vol
 from voluptuous.humanize import humanize_error
 
-from ..addons import Addon
+from ..addons.manager import Addon
 from ..const import (
     ATTR_ADDONS,
     ATTR_COMPRESSED,
     ATTR_CRYPTO,
     ATTR_DATE,
     ATTR_DOCKER,
+    ATTR_EXCLUDE_DATABASE,
     ATTR_FOLDERS,
     ATTR_HOMEASSISTANT,
     ATTR_NAME,
@@ -129,7 +131,14 @@ class Backup(CoreSysAttributes):
         """Return backup Home Assistant version."""
         if self.homeassistant is None:
             return None
-        return self._data[ATTR_HOMEASSISTANT][ATTR_VERSION]
+        return self.homeassistant[ATTR_VERSION]
+
+    @property
+    def homeassistant_exclude_database(self) -> bool:
+        """Return whether database was excluded from Home Assistant backup."""
+        if self.homeassistant is None:
+            return None
+        return self.homeassistant[ATTR_EXCLUDE_DATABASE]
 
     @property
     def homeassistant(self):
@@ -306,7 +315,11 @@ class Backup(CoreSysAttributes):
         def _extract_backup():
             """Extract a backup."""
             with tarfile.open(self.tarfile, "r:") as tar:
-                tar.extractall(path=self._tmp.name, members=secure_path(tar))
+                tar.extractall(
+                    path=self._tmp.name,
+                    members=secure_path(tar),
+                    filter="fully_trusted",
+                )
 
         await self.sys_run_in_executor(_extract_backup)
 
@@ -340,14 +353,14 @@ class Backup(CoreSysAttributes):
         finally:
             self._tmp.cleanup()
 
-    async def store_addons(self, addon_list: list[str]) -> list[Awaitable[None]]:
+    async def store_addons(self, addon_list: list[str]) -> list[asyncio.Task]:
         """Add a list of add-ons into backup.
 
-        For each addon that needs to be started after backup, returns a task which
+        For each addon that needs to be started after backup, returns a Task which
         completes when that addon has state 'started' (see addon.start).
         """
 
-        async def _addon_save(addon: Addon) -> Awaitable[None] | None:
+        async def _addon_save(addon: Addon) -> asyncio.Task | None:
             """Task to store an add-on into backup."""
             tar_name = f"{addon.slug}.tar{'.gz' if self.compressed else ''}"
             addon_file = SecureTarFile(
@@ -379,7 +392,7 @@ class Backup(CoreSysAttributes):
 
         # Save Add-ons sequential
         # avoid issue on slow IO
-        start_tasks: list[Awaitable[None]] = []
+        start_tasks: list[asyncio.Task] = []
         for addon in addon_list:
             try:
                 if start_task := await _addon_save(addon):
@@ -389,10 +402,12 @@ class Backup(CoreSysAttributes):
 
         return start_tasks
 
-    async def restore_addons(self, addon_list: list[str]) -> list[Awaitable[None]]:
+    async def restore_addons(
+        self, addon_list: list[str]
+    ) -> tuple[bool, list[asyncio.Task]]:
         """Restore a list add-on from backup."""
 
-        async def _addon_restore(addon_slug: str) -> Awaitable[None] | None:
+        async def _addon_restore(addon_slug: str) -> tuple[bool, asyncio.Task | None]:
             """Task to restore an add-on into backup."""
             tar_name = f"{addon_slug}.tar{'.gz' if self.compressed else ''}"
             addon_file = SecureTarFile(
@@ -406,25 +421,31 @@ class Backup(CoreSysAttributes):
             # If exists inside backup
             if not addon_file.path.exists():
                 _LOGGER.error("Can't find backup %s", addon_slug)
-                return
+                return (False, None)
 
             # Perform a restore
             try:
-                return await self.sys_addons.restore(addon_slug, addon_file)
+                return (True, await self.sys_addons.restore(addon_slug, addon_file))
             except AddonsError:
                 _LOGGER.error("Can't restore backup %s", addon_slug)
+                return (False, None)
 
         # Save Add-ons sequential
         # avoid issue on slow IO
-        start_tasks: list[Awaitable[None]] = []
+        start_tasks: list[asyncio.Task] = []
+        success = True
         for slug in addon_list:
             try:
-                if start_task := await _addon_restore(slug):
-                    start_tasks.append(start_task)
+                addon_success, start_task = await _addon_restore(slug)
             except Exception as err:  # pylint: disable=broad-except
                 _LOGGER.warning("Can't restore Add-on %s: %s", slug, err)
+                success = False
+            else:
+                success = success and addon_success
+                if start_task:
+                    start_tasks.append(start_task)
 
-        return start_tasks
+        return (success, start_tasks)
 
     async def store_folders(self, folder_list: list[str]):
         """Backup Supervisor data into backup."""
@@ -474,10 +495,11 @@ class Backup(CoreSysAttributes):
                     f"Can't backup folder {folder}: {str(err)}", _LOGGER.error
                 ) from err
 
-    async def restore_folders(self, folder_list: list[str]):
+    async def restore_folders(self, folder_list: list[str]) -> bool:
         """Backup Supervisor data into backup."""
+        success = True
 
-        async def _folder_restore(name: str) -> None:
+        async def _folder_restore(name: str) -> bool:
             """Intenal function to restore a folder."""
             slug_name = name.replace("/", "_")
             tar_name = Path(
@@ -488,14 +510,26 @@ class Backup(CoreSysAttributes):
             # Check if exists inside backup
             if not tar_name.exists():
                 _LOGGER.warning("Can't find restore folder %s", name)
-                return
+                return False
+
+            # Unmount any mounts within folder
+            bind_mounts = [
+                bound.bind_mount
+                for bound in self.sys_mounts.bound_mounts
+                if bound.bind_mount.local_where
+                and bound.bind_mount.local_where.is_relative_to(origin_dir)
+            ]
+            if bind_mounts:
+                await asyncio.gather(
+                    *[bind_mount.unmount() for bind_mount in bind_mounts]
+                )
 
             # Clean old stuff
             if origin_dir.is_dir():
                 await remove_folder(origin_dir, content_only=True)
 
             # Perform a restore
-            def _restore() -> None:
+            def _restore() -> bool:
                 try:
                     _LOGGER.info("Restore folder %s", name)
                     with SecureTarFile(
@@ -505,24 +539,39 @@ class Backup(CoreSysAttributes):
                         gzip=self.compressed,
                         bufsize=BUF_SIZE,
                     ) as tar_file:
-                        tar_file.extractall(path=origin_dir, members=tar_file)
+                        tar_file.extractall(
+                            path=origin_dir, members=tar_file, filter="fully_trusted"
+                        )
                     _LOGGER.info("Restore folder %s done", name)
                 except (tarfile.TarError, OSError) as err:
                     _LOGGER.warning("Can't restore folder %s: %s", name, err)
+                    return False
+                return True
 
-            await self.sys_run_in_executor(_restore)
+            try:
+                return await self.sys_run_in_executor(_restore)
+            finally:
+                if bind_mounts:
+                    await asyncio.gather(
+                        *[bind_mount.mount() for bind_mount in bind_mounts]
+                    )
 
         # Restore folder sequential
         # avoid issue on slow IO
         for folder in folder_list:
             try:
-                await _folder_restore(folder)
+                success = success and await _folder_restore(folder)
             except Exception as err:  # pylint: disable=broad-except
                 _LOGGER.warning("Can't restore folder %s: %s", folder, err)
+                success = False
+        return success
 
-    async def store_homeassistant(self):
-        """Backup Home Assitant Core configuration folder."""
-        self._data[ATTR_HOMEASSISTANT] = {ATTR_VERSION: self.sys_homeassistant.version}
+    async def store_homeassistant(self, exclude_database: bool = False):
+        """Backup Home Assistant Core configuration folder."""
+        self._data[ATTR_HOMEASSISTANT] = {
+            ATTR_VERSION: self.sys_homeassistant.version,
+            ATTR_EXCLUDE_DATABASE: exclude_database,
+        }
 
         # Backup Home Assistant Core config directory
         tar_name = Path(
@@ -532,13 +581,13 @@ class Backup(CoreSysAttributes):
             tar_name, "w", key=self._key, gzip=self.compressed, bufsize=BUF_SIZE
         )
 
-        await self.sys_homeassistant.backup(homeassistant_file)
+        await self.sys_homeassistant.backup(homeassistant_file, exclude_database)
 
         # Store size
         self.homeassistant[ATTR_SIZE] = homeassistant_file.size
 
     async def restore_homeassistant(self) -> Awaitable[None]:
-        """Restore Home Assitant Core configuration folder."""
+        """Restore Home Assistant Core configuration folder."""
         await self.sys_homeassistant.core.stop()
 
         # Restore Home Assistant Core config directory
@@ -549,7 +598,9 @@ class Backup(CoreSysAttributes):
             tar_name, "r", key=self._key, gzip=self.compressed, bufsize=BUF_SIZE
         )
 
-        await self.sys_homeassistant.restore(homeassistant_file)
+        await self.sys_homeassistant.restore(
+            homeassistant_file, self.homeassistant_exclude_database
+        )
 
         # Generate restore task
         async def _core_update():
@@ -572,12 +623,12 @@ class Backup(CoreSysAttributes):
         """Store repository list into backup."""
         self.repositories = self.sys_store.repository_urls
 
-    async def restore_repositories(self, replace: bool = False):
+    def restore_repositories(self, replace: bool = False) -> Awaitable[None]:
         """Restore repositories from backup.
 
         Return a coroutine.
         """
-        await self.sys_store.update_repositories(
+        return self.sys_store.update_repositories(
             self.repositories, add_with_errors=True, replace=replace
         )
 

supervisor/backups/const.py

@@ -2,6 +2,7 @@
 from enum import StrEnum
 
 BUF_SIZE = 2**20 * 4  # 4MB
+DEFAULT_FREEZE_TIMEOUT = 600
 
 
 class BackupType(StrEnum):

supervisor/backups/manager.py

@@ -3,6 +3,7 @@ from __future__ import annotations
 
 import asyncio
 from collections.abc import Awaitable, Iterable
+import errno
 import logging
 from pathlib import Path
 
@@ -13,44 +14,35 @@ from ..const import (
     FOLDER_HOMEASSISTANT,
     CoreState,
 )
-from ..coresys import CoreSysAttributes
 from ..dbus.const import UnitActiveState
-from ..exceptions import AddonsError
-from ..jobs.decorator import Job, JobCondition
+from ..exceptions import AddonsError, BackupError, BackupInvalidError, BackupJobError
+from ..jobs.const import JOB_GROUP_BACKUP_MANAGER, JobCondition, JobExecutionLimit
+from ..jobs.decorator import Job
+from ..jobs.job_group import JobGroup
 from ..mounts.mount import Mount
+from ..resolution.const import UnhealthyReason
 from ..utils.common import FileConfiguration
 from ..utils.dt import utcnow
 from ..utils.sentinel import DEFAULT
 from ..utils.sentry import capture_exception
 from .backup import Backup
-from .const import BackupJobStage, BackupType, RestoreJobStage
+from .const import DEFAULT_FREEZE_TIMEOUT, BackupJobStage, BackupType, RestoreJobStage
 from .utils import create_slug
 from .validate import ALL_FOLDERS, SCHEMA_BACKUPS_CONFIG
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
 
 
-def _list_backup_files(path: Path) -> Iterable[Path]:
-    """Return iterable of backup files, suppress and log OSError for network mounts."""
-    try:
-        # is_dir does a stat syscall which raises if the mount is down
-        if path.is_dir():
-            return path.glob("*.tar")
-    except OSError as err:
-        _LOGGER.error("Could not list backups from %s: %s", path.as_posix(), err)
-
-    return []
-
-
-class BackupManager(FileConfiguration, CoreSysAttributes):
+class BackupManager(FileConfiguration, JobGroup):
     """Manage backups."""
 
     def __init__(self, coresys):
         """Initialize a backup manager."""
         super().__init__(FILE_HASSIO_BACKUPS, SCHEMA_BACKUPS_CONFIG)
-        self.coresys = coresys
+        super(FileConfiguration, self).__init__(coresys, JOB_GROUP_BACKUP_MANAGER)
         self._backups: dict[str, Backup] = {}
-        self.lock = asyncio.Lock()
+        self._thaw_task: Awaitable[None] | None = None
+        self._thaw_event: asyncio.Event = asyncio.Event()
 
     @property
     def list_backups(self) -> set[Backup]:
@@ -92,22 +84,44 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
 
     def _change_stage(
         self,
-        backup: Backup,
         stage: BackupJobStage | RestoreJobStage,
-        restore: bool = False,
+        backup: Backup | None = None,
     ):
         """Change the stage of the current job during backup/restore.
 
         Must be called from an existing backup/restore job.
         """
+        job_name = self.sys_jobs.current.name
+        if "restore" in job_name:
+            action = "Restore"
+        elif "freeze" in job_name:
+            action = "Freeze"
+        elif "thaw" in job_name:
+            action = "Thaw"
+        else:
+            action = "Backup"
+
         _LOGGER.info(
-            "%s %s starting stage %s",
-            "Restore" if restore else "Backup",
-            backup.slug,
+            "%s %sstarting stage %s",
+            action,
+            f"{backup.slug} " if backup else "",
             stage,
         )
         self.sys_jobs.current.stage = stage
 
+    def _list_backup_files(self, path: Path) -> Iterable[Path]:
+        """Return iterable of backup files, suppress and log OSError for network mounts."""
+        try:
+            # is_dir does a stat syscall which raises if the mount is down
+            if path.is_dir():
+                return path.glob("*.tar")
+        except OSError as err:
+            if err.errno == errno.EBADMSG and path == self.sys_config.path_backup:
+                self.sys_resolution.unhealthy = UnhealthyReason.OSERROR_BAD_MESSAGE
+            _LOGGER.error("Could not list backups from %s: %s", path.as_posix(), err)
+
+        return []
+
     def _create_backup(
         self,
         name: str,
@@ -131,9 +145,9 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
         # Add backup ID to job
         self.sys_jobs.current.reference = backup.slug
 
-        self._change_stage(backup, BackupJobStage.ADDON_REPOSITORIES)
+        self._change_stage(BackupJobStage.ADDON_REPOSITORIES, backup)
         backup.store_repositories()
-        self._change_stage(backup, BackupJobStage.DOCKER_CONFIG)
+        self._change_stage(BackupJobStage.DOCKER_CONFIG, backup)
         backup.store_dockerconfig()
 
         return backup
@@ -158,7 +172,7 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
         tasks = [
             self.sys_create_task(_load_backup(tar_file))
             for path in self.backup_locations
-            for tar_file in _list_backup_files(path)
+            for tar_file in self._list_backup_files(path)
         ]
 
         _LOGGER.info("Found %d backup files", len(tasks))
@@ -173,6 +187,11 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
             _LOGGER.info("Removed backup file %s", backup.slug)
 
         except OSError as err:
+            if (
+                err.errno == errno.EBADMSG
+                and backup.tarfile.parent == self.sys_config.path_backup
+            ):
+                self.sys_resolution.unhealthy = UnhealthyReason.OSERROR_BAD_MESSAGE
             _LOGGER.error("Can't remove backup %s: %s", backup.slug, err)
             return False
 
@@ -197,6 +216,8 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
             backup.tarfile.rename(tar_origin)
 
         except OSError as err:
+            if err.errno == errno.EBADMSG:
+                self.sys_resolution.unhealthy = UnhealthyReason.OSERROR_BAD_MESSAGE
             _LOGGER.error("Can't move backup file to storage: %s", err)
             return None
 
@@ -215,6 +236,7 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
         addon_list: list[Addon],
         folder_list: list[str],
         homeassistant: bool,
+        homeassistant_exclude_database: bool | None,
     ) -> Backup | None:
         """Create a backup.
 
@@ -228,20 +250,24 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
             async with backup:
                 # Backup add-ons
                 if addon_list:
-                    self._change_stage(backup, BackupJobStage.ADDONS)
+                    self._change_stage(BackupJobStage.ADDONS, backup)
                     addon_start_tasks = await backup.store_addons(addon_list)
 
                 # HomeAssistant Folder is for v1
                 if homeassistant:
-                    self._change_stage(backup, BackupJobStage.HOME_ASSISTANT)
-                    await backup.store_homeassistant()
+                    self._change_stage(BackupJobStage.HOME_ASSISTANT, backup)
+                    await backup.store_homeassistant(
+                        self.sys_homeassistant.backups_exclude_database
+                        if homeassistant_exclude_database is None
+                        else homeassistant_exclude_database
+                    )
 
                 # Backup folders
                 if folder_list:
-                    self._change_stage(backup, BackupJobStage.FOLDERS)
+                    self._change_stage(BackupJobStage.FOLDERS, backup)
                     await backup.store_folders(folder_list)
 
-                self._change_stage(backup, BackupJobStage.FINISHING_FILE)
+                self._change_stage(BackupJobStage.FINISHING_FILE, backup)
 
         except Exception as err:  # pylint: disable=broad-except
             _LOGGER.exception("Backup %s error", backup.slug)
@@ -251,7 +277,7 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
             self._backups[backup.slug] = backup
 
             if addon_start_tasks:
-                self._change_stage(backup, BackupJobStage.AWAIT_ADDON_RESTARTS)
+                self._change_stage(BackupJobStage.AWAIT_ADDON_RESTARTS, backup)
                 # Ignore exceptions from waiting for addon startup, addon errors handled elsewhere
                 await asyncio.gather(*addon_start_tasks, return_exceptions=True)
 
@@ -261,7 +287,9 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
 
     @Job(
         name="backup_manager_full_backup",
-        conditions=[JobCondition.FREE_SPACE, JobCondition.RUNNING],
+        conditions=[JobCondition.RUNNING],
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=BackupJobError,
     )
     async def do_backup_full(
         self,
@@ -269,28 +297,35 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
         password: str | None = None,
         compressed: bool = True,
         location: Mount | type[DEFAULT] | None = DEFAULT,
+        homeassistant_exclude_database: bool | None = None,
     ) -> Backup | None:
         """Create a full backup."""
-        if self.lock.locked():
-            _LOGGER.error("A backup/restore process is already running")
-            return None
+        if self._get_base_path(location) == self.sys_config.path_backup:
+            await Job.check_conditions(
+                self, {JobCondition.FREE_SPACE}, "BackupManager.do_backup_full"
+            )
 
         backup = self._create_backup(
             name, BackupType.FULL, password, compressed, location
         )
 
         _LOGGER.info("Creating new full backup with slug %s", backup.slug)
-        async with self.lock:
-            backup = await self._do_backup(
-                backup, self.sys_addons.installed, ALL_FOLDERS, True
-            )
-            if backup:
-                _LOGGER.info("Creating full backup with slug %s completed", backup.slug)
-            return backup
+        backup = await self._do_backup(
+            backup,
+            self.sys_addons.installed,
+            ALL_FOLDERS,
+            True,
+            homeassistant_exclude_database,
+        )
+        if backup:
+            _LOGGER.info("Creating full backup with slug %s completed", backup.slug)
+        return backup
 
     @Job(
         name="backup_manager_partial_backup",
-        conditions=[JobCondition.FREE_SPACE, JobCondition.RUNNING],
+        conditions=[JobCondition.RUNNING],
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=BackupJobError,
     )
     async def do_backup_partial(
         self,
@@ -301,11 +336,13 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
         homeassistant: bool = False,
         compressed: bool = True,
         location: Mount | type[DEFAULT] | None = DEFAULT,
+        homeassistant_exclude_database: bool | None = None,
     ) -> Backup | None:
         """Create a partial backup."""
-        if self.lock.locked():
-            _LOGGER.error("A backup/restore process is already running")
-            return None
+        if self._get_base_path(location) == self.sys_config.path_backup:
+            await Job.check_conditions(
+                self, {JobCondition.FREE_SPACE}, "BackupManager.do_backup_partial"
+            )
 
         addons = addons or []
         folders = folders or []
@@ -323,21 +360,20 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
         )
 
         _LOGGER.info("Creating new partial backup with slug %s", backup.slug)
-        async with self.lock:
-            addon_list = []
-            for addon_slug in addons:
-                addon = self.sys_addons.get(addon_slug)
-                if addon and addon.is_installed:
-                    addon_list.append(addon)
-                    continue
-                _LOGGER.warning("Add-on %s not found/installed", addon_slug)
+        addon_list = []
+        for addon_slug in addons:
+            addon = self.sys_addons.get(addon_slug)
+            if addon and addon.is_installed:
+                addon_list.append(addon)
+                continue
+            _LOGGER.warning("Add-on %s not found/installed", addon_slug)
 
-            backup = await self._do_backup(backup, addon_list, folders, homeassistant)
-            if backup:
-                _LOGGER.info(
-                    "Creating partial backup with slug %s completed", backup.slug
-                )
-            return backup
+        backup = await self._do_backup(
+            backup, addon_list, folders, homeassistant, homeassistant_exclude_database
+        )
+        if backup:
+            _LOGGER.info("Creating partial backup with slug %s completed", backup.slug)
+        return backup
 
     async def _do_restore(
         self,
@@ -352,31 +388,28 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
         Must be called from an existing restore job.
         """
         addon_start_tasks: list[Awaitable[None]] | None = None
+        success = True
 
         try:
             task_hass: asyncio.Task | None = None
             async with backup:
                 # Restore docker config
-                self._change_stage(backup, RestoreJobStage.DOCKER_CONFIG, restore=True)
+                self._change_stage(RestoreJobStage.DOCKER_CONFIG, backup)
                 backup.restore_dockerconfig(replace)
 
                 # Process folders
                 if folder_list:
-                    self._change_stage(backup, RestoreJobStage.FOLDERS, restore=True)
-                    await backup.restore_folders(folder_list)
+                    self._change_stage(RestoreJobStage.FOLDERS, backup)
+                    success = await backup.restore_folders(folder_list)
 
                 # Process Home-Assistant
                 if homeassistant:
-                    self._change_stage(
-                        backup, RestoreJobStage.HOME_ASSISTANT, restore=True
-                    )
+                    self._change_stage(RestoreJobStage.HOME_ASSISTANT, backup)
                     task_hass = await backup.restore_homeassistant()
 
                 # Delete delta add-ons
                 if replace:
-                    self._change_stage(
-                        backup, RestoreJobStage.REMOVE_DELTA_ADDONS, restore=True
-                    )
+                    self._change_stage(RestoreJobStage.REMOVE_DELTA_ADDONS, backup)
                     for addon in self.sys_addons.installed:
                         if addon.slug in backup.addon_list:
                             continue
@@ -384,47 +417,49 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
                         # Remove Add-on because it's not a part of the new env
                         # Do it sequential avoid issue on slow IO
                         try:
-                            await addon.uninstall()
+                            await self.sys_addons.uninstall(addon.slug)
                         except AddonsError:
                             _LOGGER.warning("Can't uninstall Add-on %s", addon.slug)
+                            success = False
 
                 if addon_list:
-                    self._change_stage(
-                        backup, RestoreJobStage.ADDON_REPOSITORIES, restore=True
-                    )
+                    self._change_stage(RestoreJobStage.ADDON_REPOSITORIES, backup)
                     await backup.restore_repositories(replace)
 
-                    self._change_stage(backup, RestoreJobStage.ADDONS, restore=True)
-                    addon_start_tasks = await backup.restore_addons(addon_list)
+                    self._change_stage(RestoreJobStage.ADDONS, backup)
+                    restore_success, addon_start_tasks = await backup.restore_addons(
+                        addon_list
+                    )
+                    success = success and restore_success
 
                 # Wait for Home Assistant Core update/downgrade
                 if task_hass:
                     self._change_stage(
-                        backup,
-                        RestoreJobStage.AWAIT_HOME_ASSISTANT_RESTART,
-                        restore=True,
+                        RestoreJobStage.AWAIT_HOME_ASSISTANT_RESTART, backup
                     )
                     await task_hass
-
+        except BackupError:
+            raise
         except Exception as err:  # pylint: disable=broad-except
             _LOGGER.exception("Restore %s error", backup.slug)
             capture_exception(err)
-            return False
+            raise BackupError(
+                f"Restore {backup.slug} error, check logs for details"
+            ) from err
         else:
             if addon_start_tasks:
-                self._change_stage(
-                    backup, RestoreJobStage.AWAIT_ADDON_RESTARTS, restore=True
-                )
-                # Ignore exceptions from waiting for addon startup, addon errors handled elsewhere
-                await asyncio.gather(*addon_start_tasks, return_exceptions=True)
+                self._change_stage(RestoreJobStage.AWAIT_ADDON_RESTARTS, backup)
+                # Failure to resume addons post restore is still a restore failure
+                if any(
+                    await asyncio.gather(*addon_start_tasks, return_exceptions=True)
+                ):
+                    return False
 
-            return True
+            return success
         finally:
             # Leave Home Assistant alone if it wasn't part of the restore
             if homeassistant:
-                self._change_stage(
-                    backup, RestoreJobStage.CHECK_HOME_ASSISTANT, restore=True
-                )
+                self._change_stage(RestoreJobStage.CHECK_HOME_ASSISTANT, backup)
 
                 # Do we need start Home Assistant Core?
                 if not await self.sys_homeassistant.core.is_running():
@@ -444,6 +479,8 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
             JobCondition.INTERNET_SYSTEM,
             JobCondition.RUNNING,
         ],
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=BackupJobError,
     )
     async def do_restore_full(
         self, backup: Backup, password: str | None = None
@@ -452,42 +489,39 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
         # Add backup ID to job
         self.sys_jobs.current.reference = backup.slug
 
-        if self.lock.locked():
-            _LOGGER.error("A backup/restore process is already running")
-            return False
-
         if backup.sys_type != BackupType.FULL:
-            _LOGGER.error("%s is only a partial backup!", backup.slug)
-            return False
+            raise BackupInvalidError(
+                f"{backup.slug} is only a partial backup!", _LOGGER.error
+            )
 
         if backup.protected and not backup.set_password(password):
-            _LOGGER.error("Invalid password for backup %s", backup.slug)
-            return False
+            raise BackupInvalidError(
+                f"Invalid password for backup {backup.slug}", _LOGGER.error
+            )
 
         if backup.supervisor_version > self.sys_supervisor.version:
-            _LOGGER.error(
-                "Backup was made on supervisor version %s, can't restore on %s. Must update supervisor first.",
-                backup.supervisor_version,
-                self.sys_supervisor.version,
+            raise BackupInvalidError(
+                f"Backup was made on supervisor version {backup.supervisor_version}, "
+                f"can't restore on {self.sys_supervisor.version}. Must update supervisor first.",
+                _LOGGER.error,
             )
-            return False
 
         _LOGGER.info("Full-Restore %s start", backup.slug)
-        async with self.lock:
-            self.sys_core.state = CoreState.FREEZE
+        self.sys_core.state = CoreState.FREEZE
 
+        try:
             # Stop Home-Assistant / Add-ons
             await self.sys_core.shutdown()
 
             success = await self._do_restore(
                 backup, backup.addon_list, backup.folders, True, True
             )
-
+        finally:
             self.sys_core.state = CoreState.RUNNING
 
-            if success:
-                _LOGGER.info("Full-Restore %s done", backup.slug)
-            return success
+        if success:
+            _LOGGER.info("Full-Restore %s done", backup.slug)
+        return success
 
     @Job(
         name="backup_manager_partial_restore",
@@ -498,6 +532,8 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
             JobCondition.INTERNET_SYSTEM,
             JobCondition.RUNNING,
         ],
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=BackupJobError,
     )
     async def do_restore_partial(
         self,
@@ -511,10 +547,6 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
         # Add backup ID to job
         self.sys_jobs.current.reference = backup.slug
 
-        if self.lock.locked():
-            _LOGGER.error("A backup/restore process is already running")
-            return False
-
         addon_list = addons or []
         folder_list = folders or []
 
@@ -524,31 +556,118 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
             homeassistant = True
 
         if backup.protected and not backup.set_password(password):
-            _LOGGER.error("Invalid password for backup %s", backup.slug)
-            return False
+            raise BackupInvalidError(
+                f"Invalid password for backup {backup.slug}", _LOGGER.error
+            )
 
         if backup.homeassistant is None and homeassistant:
-            _LOGGER.error("No Home Assistant Core data inside the backup")
-            return False
+            raise BackupInvalidError(
+                "No Home Assistant Core data inside the backup", _LOGGER.error
+            )
 
         if backup.supervisor_version > self.sys_supervisor.version:
-            _LOGGER.error(
-                "Backup was made on supervisor version %s, can't restore on %s. Must update supervisor first.",
-                backup.supervisor_version,
-                self.sys_supervisor.version,
+            raise BackupInvalidError(
+                f"Backup was made on supervisor version {backup.supervisor_version}, "
+                f"can't restore on {self.sys_supervisor.version}. Must update supervisor first.",
+                _LOGGER.error,
             )
-            return False
 
         _LOGGER.info("Partial-Restore %s start", backup.slug)
-        async with self.lock:
-            self.sys_core.state = CoreState.FREEZE
+        self.sys_core.state = CoreState.FREEZE
 
+        try:
             success = await self._do_restore(
                 backup, addon_list, folder_list, homeassistant, False
             )
-
+        finally:
             self.sys_core.state = CoreState.RUNNING
 
-            if success:
-                _LOGGER.info("Partial-Restore %s done", backup.slug)
-            return success
+        if success:
+            _LOGGER.info("Partial-Restore %s done", backup.slug)
+        return success
+
+    @Job(
+        name="backup_manager_freeze_all",
+        conditions=[JobCondition.RUNNING],
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=BackupJobError,
+    )
+    async def freeze_all(self, timeout: float = DEFAULT_FREEZE_TIMEOUT) -> None:
+        """Freeze system to prepare for an external backup such as an image snapshot."""
+        self.sys_core.state = CoreState.FREEZE
+
+        # Determine running addons
+        installed = self.sys_addons.installed.copy()
+        is_running: list[bool] = await asyncio.gather(
+            *[addon.is_running() for addon in installed]
+        )
+        running_addons = [
+            installed[ind] for ind in range(len(installed)) if is_running[ind]
+        ]
+
+        # Create thaw task first to ensure we eventually undo freezes even if the below fails
+        self._thaw_task = asyncio.shield(
+            self.sys_create_task(self._thaw_all(running_addons, timeout))
+        )
+
+        # Tell Home Assistant to freeze for a backup
+        self._change_stage(BackupJobStage.HOME_ASSISTANT)
+        await self.sys_homeassistant.begin_backup()
+
+        # Run all pre-backup tasks for addons
+        self._change_stage(BackupJobStage.ADDONS)
+        await asyncio.gather(*[addon.begin_backup() for addon in running_addons])
+
+    @Job(
+        name="backup_manager_thaw_all",
+        conditions=[JobCondition.FROZEN],
+        on_condition=BackupJobError,
+    )
+    async def _thaw_all(
+        self, running_addons: list[Addon], timeout: float = DEFAULT_FREEZE_TIMEOUT
+    ) -> None:
+        """Thaw system after user signal or timeout."""
+        try:
+            try:
+                await asyncio.wait_for(self._thaw_event.wait(), timeout)
+            except TimeoutError:
+                _LOGGER.warning(
+                    "Timeout waiting for signal to thaw after manual freeze, beginning thaw now"
+                )
+
+            self._change_stage(BackupJobStage.HOME_ASSISTANT)
+            await self.sys_homeassistant.end_backup()
+
+            self._change_stage(BackupJobStage.ADDONS)
+            addon_start_tasks: list[asyncio.Task] = [
+                task
+                for task in await asyncio.gather(
+                    *[addon.end_backup() for addon in running_addons]
+                )
+                if task
+            ]
+        finally:
+            self.sys_core.state = CoreState.RUNNING
+            self._thaw_event.clear()
+            self._thaw_task = None
+
+        if addon_start_tasks:
+            self._change_stage(BackupJobStage.AWAIT_ADDON_RESTARTS)
+            await asyncio.gather(*addon_start_tasks, return_exceptions=True)
+
+    @Job(
+        name="backup_manager_signal_thaw",
+        conditions=[JobCondition.FROZEN],
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=BackupJobError,
+        internal=True,
+    )
+    async def thaw_all(self) -> None:
+        """Signal thaw task to begin unfreezing the system."""
+        if not self._thaw_task:
+            raise BackupError(
+                "Freeze was not initiated by freeze API, cannot thaw this way"
+            )
+
+        self._thaw_event.set()
+        await self._thaw_task

supervisor/backups/validate.py

@@ -14,6 +14,7 @@ from ..const import (
     ATTR_DATE,
     ATTR_DAYS_UNTIL_STALE,
     ATTR_DOCKER,
+    ATTR_EXCLUDE_DATABASE,
     ATTR_FOLDERS,
     ATTR_HOMEASSISTANT,
     ATTR_NAME,
@@ -103,6 +104,9 @@ SCHEMA_BACKUP = vol.Schema(
                     {
                         vol.Required(ATTR_VERSION): version_tag,
                         vol.Optional(ATTR_SIZE, default=0): vol.Coerce(float),
+                        vol.Optional(
+                            ATTR_EXCLUDE_DATABASE, default=False
+                        ): vol.Boolean(),
                     },
                     extra=vol.REMOVE_EXTRA,
                 )

supervisor/bootstrap.py

@@ -6,7 +6,7 @@ import signal
 
 from colorlog import ColoredFormatter
 
-from .addons import AddonManager
+from .addons.manager import AddonManager
 from .api import RestAPI
 from .arch import CpuArch
 from .auth import Auth
@@ -221,6 +221,14 @@ def initialize_system(coresys: CoreSys) -> None:
         )
         config.path_emergency.mkdir()
 
+    # Addon Configs folder
+    if not config.path_addon_configs.is_dir():
+        _LOGGER.debug(
+            "Creating Supervisor add-on configs folder at '%s'",
+            config.path_addon_configs,
+        )
+        config.path_addon_configs.mkdir()
+
 
 def migrate_system_env(coresys: CoreSys) -> None:
     """Cleanup some stuff after update."""

supervisor/config.py

@@ -1,5 +1,5 @@
 """Bootstrap Supervisor."""
-from datetime import datetime
+from datetime import UTC, datetime
 import logging
 import os
 from pathlib import Path, PurePath
@@ -48,8 +48,9 @@ MEDIA_DATA = PurePath("media")
 MOUNTS_FOLDER = PurePath("mounts")
 MOUNTS_CREDENTIALS = PurePath(".mounts_credentials")
 EMERGENCY_DATA = PurePath("emergency")
+ADDON_CONFIGS = PurePath("addon_configs")
 
-DEFAULT_BOOT_TIME = datetime.utcfromtimestamp(0).isoformat()
+DEFAULT_BOOT_TIME = datetime.fromtimestamp(0, UTC).isoformat()
 
 # We filter out UTC because it's the system default fallback
 # Core also not respect the cotnainer timezone and reset timezones
@@ -163,7 +164,7 @@ class CoreConfig(FileConfiguration):
 
         boot_time = parse_datetime(boot_str)
         if not boot_time:
-            return datetime.utcfromtimestamp(1)
+            return datetime.fromtimestamp(1, UTC)
         return boot_time
 
     @last_boot.setter
@@ -231,6 +232,16 @@ class CoreConfig(FileConfiguration):
         """Return root add-on data folder external for Docker."""
         return PurePath(self.path_extern_supervisor, ADDONS_DATA)
 
+    @property
+    def path_addon_configs(self) -> Path:
+        """Return root Add-on configs folder."""
+        return self.path_supervisor / ADDON_CONFIGS
+
+    @property
+    def path_extern_addon_configs(self) -> PurePath:
+        """Return root Add-on configs folder external for Docker."""
+        return PurePath(self.path_extern_supervisor, ADDON_CONFIGS)
+
     @property
     def path_audio(self) -> Path:
         """Return root audio data folder."""

supervisor/const.py

@@ -4,6 +4,7 @@ from enum import StrEnum
 from ipaddress import ip_network
 from pathlib import Path
 from sys import version_info as systemversion
+from typing import Self
 
 from aiohttp import __version__ as aiohttpversion
 
@@ -19,6 +20,7 @@ SUPERVISOR_DATA = Path("/data")
 FILE_HASSIO_ADDONS = Path(SUPERVISOR_DATA, "addons.json")
 FILE_HASSIO_AUTH = Path(SUPERVISOR_DATA, "auth.json")
 FILE_HASSIO_BACKUPS = Path(SUPERVISOR_DATA, "backups.json")
+FILE_HASSIO_BOARD = Path(SUPERVISOR_DATA, "board.json")
 FILE_HASSIO_CONFIG = Path(SUPERVISOR_DATA, "config.json")
 FILE_HASSIO_DISCOVERY = Path(SUPERVISOR_DATA, "discovery.json")
 FILE_HASSIO_DOCKER = Path(SUPERVISOR_DATA, "docker.json")
@@ -88,6 +90,7 @@ REQUEST_FROM = "HASSIO_FROM"
 ATTR_ACCESS_TOKEN = "access_token"
 ATTR_ACCESSPOINTS = "accesspoints"
 ATTR_ACTIVE = "active"
+ATTR_ACTIVITY_LED = "activity_led"
 ATTR_ADDON = "addon"
 ATTR_ADDONS = "addons"
 ATTR_ADDONS_CUSTOM_LIST = "addons_custom_list"
@@ -113,6 +116,7 @@ ATTR_BACKUP_EXCLUDE = "backup_exclude"
 ATTR_BACKUP_POST = "backup_post"
 ATTR_BACKUP_PRE = "backup_pre"
 ATTR_BACKUPS = "backups"
+ATTR_BACKUPS_EXCLUDE_DATABASE = "backups_exclude_database"
 ATTR_BLK_READ = "blk_read"
 ATTR_BLK_WRITE = "blk_write"
 ATTR_BOARD = "board"
@@ -152,9 +156,11 @@ ATTR_DIAGNOSTICS = "diagnostics"
 ATTR_DISCOVERY = "discovery"
 ATTR_DISK = "disk"
 ATTR_DISK_FREE = "disk_free"
+ATTR_DISK_LED = "disk_led"
 ATTR_DISK_LIFE_TIME = "disk_life_time"
 ATTR_DISK_TOTAL = "disk_total"
 ATTR_DISK_USED = "disk_used"
+ATTR_DISPLAYNAME = "displayname"
 ATTR_DNS = "dns"
 ATTR_DOCKER = "docker"
 ATTR_DOCKER_API = "docker_api"
@@ -164,6 +170,7 @@ ATTR_ENABLE = "enable"
 ATTR_ENABLED = "enabled"
 ATTR_ENVIRONMENT = "environment"
 ATTR_EVENT = "event"
+ATTR_EXCLUDE_DATABASE = "exclude_database"
 ATTR_FEATURES = "features"
 ATTR_FILENAME = "filename"
 ATTR_FLAGS = "flags"
@@ -177,7 +184,9 @@ ATTR_HASSIO_API = "hassio_api"
 ATTR_HASSIO_ROLE = "hassio_role"
 ATTR_HASSOS = "hassos"
 ATTR_HEALTHY = "healthy"
+ATTR_HEARTBEAT_LED = "heartbeat_led"
 ATTR_HOMEASSISTANT = "homeassistant"
+ATTR_HOMEASSISTANT_EXCLUDE_DATABASE = "homeassistant_exclude_database"
 ATTR_HOMEASSISTANT_API = "homeassistant_api"
 ATTR_HOST = "host"
 ATTR_HOST_DBUS = "host_dbus"
@@ -252,6 +261,7 @@ ATTR_PLUGINS = "plugins"
 ATTR_PORT = "port"
 ATTR_PORTS = "ports"
 ATTR_PORTS_DESCRIPTION = "ports_description"
+ATTR_POWER_LED = "power_led"
 ATTR_PREFIX = "prefix"
 ATTR_PRIMARY = "primary"
 ATTR_PRIORITY = "priority"
@@ -315,6 +325,7 @@ ATTR_UPDATE_KEY = "update_key"
 ATTR_URL = "url"
 ATTR_USB = "usb"
 ATTR_USER = "user"
+ATTR_USER_LED = "user_led"
 ATTR_USERNAME = "username"
 ATTR_UUID = "uuid"
 ATTR_VALID = "valid"
@@ -334,14 +345,6 @@ PROVIDE_SERVICE = "provide"
 NEED_SERVICE = "need"
 WANT_SERVICE = "want"
 
-
-MAP_CONFIG = "config"
-MAP_SSL = "ssl"
-MAP_ADDONS = "addons"
-MAP_BACKUP = "backup"
-MAP_SHARE = "share"
-MAP_MEDIA = "media"
-
 ARCH_ARMHF = "armhf"
 ARCH_ARMV7 = "armv7"
 ARCH_AARCH64 = "aarch64"
@@ -476,8 +479,25 @@ class IngressSessionDataUser:
     """Format of an IngressSessionDataUser object."""
 
     id: str
-    display_name: str
-    username: str
+    display_name: str | None = None
+    username: str | None = None
+
+    def to_dict(self) -> dict[str, str | None]:
+        """Get dictionary representation."""
+        return {
+            ATTR_ID: self.id,
+            ATTR_DISPLAYNAME: self.display_name,
+            ATTR_USERNAME: self.username,
+        }
+
+    @classmethod
+    def from_dict(cls, data: dict[str, str | None]) -> Self:
+        """Return object from dictionary representation."""
+        return cls(
+            id=data[ATTR_ID],
+            display_name=data.get(ATTR_DISPLAYNAME),
+            username=data.get(ATTR_USERNAME),
+        )
 
 
 @dataclass
@@ -486,6 +506,15 @@ class IngressSessionData:
 
     user: IngressSessionDataUser
 
+    def to_dict(self) -> dict[str, dict[str, str | None]]:
+        """Get dictionary representation."""
+        return {ATTR_USER: self.user.to_dict()}
+
+    @classmethod
+    def from_dict(cls, data: dict[str, dict[str, str | None]]) -> Self:
+        """Return object from dictionary representation."""
+        return cls(user=IngressSessionDataUser.from_dict(data[ATTR_USER]))
+
 
 STARTING_STATES = [
     CoreState.INITIALIZE,

supervisor/core.py

@@ -28,7 +28,7 @@ from .homeassistant.core import LANDINGPAGE
 from .resolution.const import ContextType, IssueType, SuggestionType, UnhealthyReason
 from .utils.dt import utcnow
 from .utils.sentry import capture_exception
-from .utils.whoami import retrieve_whoami
+from .utils.whoami import WhoamiData, retrieve_whoami
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
 
@@ -250,7 +250,7 @@ class Core(CoreSysAttributes):
                 except HomeAssistantError as err:
                     capture_exception(err)
             else:
-                _LOGGER.info("Skiping start of Home Assistant")
+                _LOGGER.info("Skipping start of Home Assistant")
 
             # Core is not running
             if self.sys_homeassistant.core.error_state:
@@ -309,7 +309,7 @@ class Core(CoreSysAttributes):
                         )
                     ]
                 )
-        except asyncio.TimeoutError:
+        except TimeoutError:
             _LOGGER.warning("Stage 1: Force Shutdown!")
 
         # Stage 2
@@ -326,7 +326,7 @@ class Core(CoreSysAttributes):
                         )
                     ]
                 )
-        except asyncio.TimeoutError:
+        except TimeoutError:
             _LOGGER.warning("Stage 2: Force Shutdown!")
 
         self.state = CoreState.CLOSE
@@ -363,6 +363,13 @@ class Core(CoreSysAttributes):
         self.sys_config.last_boot = self.sys_hardware.helper.last_boot
         self.sys_config.save_data()
 
+    async def _retrieve_whoami(self, with_ssl: bool) -> WhoamiData | None:
+        try:
+            return await retrieve_whoami(self.sys_websession, with_ssl)
+        except WhoamiSSLError:
+            _LOGGER.info("Whoami service SSL error")
+            return None
+
     async def _adjust_system_datetime(self):
         """Adjust system time/date on startup."""
         # If no timezone is detect or set
@@ -375,21 +382,15 @@ class Core(CoreSysAttributes):
 
         # Get Timezone data
         try:
-            data = await retrieve_whoami(self.sys_websession)
-        except WhoamiSSLError:
-            pass
+            data = await self._retrieve_whoami(True)
+
+            # SSL Date Issue & possible time drift
+            if not data:
+                data = await self._retrieve_whoami(False)
         except WhoamiError as err:
             _LOGGER.warning("Can't adjust Time/Date settings: %s", err)
             return
 
-        # SSL Date Issue & possible time drift
-        if not data:
-            try:
-                data = await retrieve_whoami(self.sys_websession, with_ssl=False)
-            except WhoamiError as err:
-                _LOGGER.error("Can't adjust Time/Date settings: %s", err)
-                return
-
         self.sys_config.timezone = self.sys_config.timezone or data.timezone
 
         # Calculate if system time is out of sync

supervisor/coresys.py

@@ -3,6 +3,7 @@ from __future__ import annotations
 
 import asyncio
 from collections.abc import Callable, Coroutine
+from contextvars import Context, copy_context
 from datetime import datetime
 from functools import partial
 import logging
@@ -17,7 +18,7 @@ from .const import ENV_SUPERVISOR_DEV, SERVER_SOFTWARE
 from .utils.dt import UTC, get_time_zone
 
 if TYPE_CHECKING:
-    from .addons import AddonManager
+    from .addons.manager import AddonManager
     from .api import RestAPI
     from .arch import CpuArch
     from .auth import Auth
@@ -99,6 +100,9 @@ class CoreSys:
             {aiohttp.hdrs.USER_AGENT: SERVER_SOFTWARE}
         )
 
+        # Task factory attributes
+        self._set_task_context: list[Callable[[Context], Context]] = []
+
     @property
     def dev(self) -> bool:
         """Return True if we run dev mode."""
@@ -520,6 +524,17 @@ class CoreSys:
         """Return now in local timezone."""
         return datetime.now(get_time_zone(self.timezone) or UTC)
 
+    def add_set_task_context_callback(
+        self, callback: Callable[[Context], Context]
+    ) -> None:
+        """Add callback used to modify context prior to creating a task.
+
+        Only used for tasks created via CoreSys.create_task. Callback can modify the provided
+        context using context.run (ex. `context.run(var.set, "new_value")`). Callback should
+        return the context to be provided to task.
+        """
+        self._set_task_context.append(callback)
+
     def run_in_executor(
         self, funct: Callable[..., T], *args: tuple[Any], **kwargs: dict[str, Any]
     ) -> Coroutine[Any, Any, T]:
@@ -531,7 +546,11 @@ class CoreSys:
 
     def create_task(self, coroutine: Coroutine) -> asyncio.Task:
         """Create an async task."""
-        return self.loop.create_task(coroutine)
+        context = copy_context()
+        for callback in self._set_task_context:
+            context = callback(context)
+
+        return self.loop.create_task(coroutine, context=context)
 
 
 class CoreSysAttributes:
@@ -706,7 +725,7 @@ class CoreSysAttributes:
     def sys_run_in_executor(
         self, funct: Callable[..., T], *args: tuple[Any], **kwargs: dict[str, Any]
     ) -> Coroutine[Any, Any, T]:
-        """Add an job to the executor pool."""
+        """Add a job to the executor pool."""
         return self.coresys.run_in_executor(funct, *args, **kwargs)
 
     def sys_create_task(self, coroutine: Coroutine) -> asyncio.Task:

supervisor/data/arch.json

@@ -5,6 +5,7 @@
   "raspberrypi3-64": ["aarch64", "armv7", "armhf"],
   "raspberrypi4": ["armv7", "armhf"],
   "raspberrypi4-64": ["aarch64", "armv7", "armhf"],
+  "raspberrypi5-64": ["aarch64", "armv7", "armhf"],
   "yellow": ["aarch64", "armv7", "armhf"],
   "green": ["aarch64", "armv7", "armhf"],
   "tinker": ["armv7", "armhf"],

supervisor/dbus/agent/__init__.py

@@ -6,7 +6,7 @@ from typing import Any
 from awesomeversion import AwesomeVersion
 from dbus_fast.aio.message_bus import MessageBus
 
-from ...exceptions import DBusError, DBusInterfaceError
+from ...exceptions import DBusError, DBusInterfaceError, DBusServiceUnkownError
 from ..const import (
     DBUS_ATTR_DIAGNOSTICS,
     DBUS_ATTR_VERSION,
@@ -99,7 +99,7 @@ class OSAgent(DBusInterfaceProxy):
             await asyncio.gather(*[dbus.connect(bus) for dbus in self.all])
         except DBusError:
             _LOGGER.warning("Can't connect to OS-Agent")
-        except DBusInterfaceError:
+        except (DBusServiceUnkownError, DBusInterfaceError):
             _LOGGER.warning(
                 "No OS-Agent support on the host. Some Host functions have been disabled."
             )

supervisor/dbus/agent/boards/__init__.py

@@ -11,7 +11,8 @@ from ...const import (
     DBUS_OBJECT_HAOS_BOARDS,
 )
 from ...interface import DBusInterfaceProxy, dbus_property
-from .const import BOARD_NAME_SUPERVISED, BOARD_NAME_YELLOW
+from .const import BOARD_NAME_GREEN, BOARD_NAME_SUPERVISED, BOARD_NAME_YELLOW
+from .green import Green
 from .interface import BoardProxy
 from .supervised import Supervised
 from .yellow import Yellow
@@ -39,6 +40,14 @@ class BoardManager(DBusInterfaceProxy):
         """Get board name."""
         return self.properties[DBUS_ATTR_BOARD]
 
+    @property
+    def green(self) -> Green:
+        """Get Green board."""
+        if self.board != BOARD_NAME_GREEN:
+            raise BoardInvalidError("Green board is not in use", _LOGGER.error)
+
+        return self._board_proxy
+
     @property
     def supervised(self) -> Supervised:
         """Get Supervised board."""
@@ -61,6 +70,8 @@ class BoardManager(DBusInterfaceProxy):
 
         if self.board == BOARD_NAME_YELLOW:
             self._board_proxy = Yellow()
+        elif self.board == BOARD_NAME_GREEN:
+            self._board_proxy = Green()
         elif self.board == BOARD_NAME_SUPERVISED:
             self._board_proxy = Supervised()
 

supervisor/dbus/agent/boards/const.py

@@ -1,4 +1,5 @@
 """Constants for boards."""
 
+BOARD_NAME_GREEN = "Green"
 BOARD_NAME_SUPERVISED = "Supervised"
 BOARD_NAME_YELLOW = "Yellow"

supervisor/dbus/agent/boards/green.py

@@ -0,0 +1,65 @@
+"""Green board management."""
+
+import asyncio
+
+from dbus_fast.aio.message_bus import MessageBus
+
+from ....const import ATTR_ACTIVITY_LED, ATTR_POWER_LED, ATTR_USER_LED
+from ...const import DBUS_ATTR_ACTIVITY_LED, DBUS_ATTR_POWER_LED, DBUS_ATTR_USER_LED
+from ...interface import dbus_property
+from .const import BOARD_NAME_GREEN
+from .interface import BoardProxy
+from .validate import SCHEMA_GREEN_BOARD
+
+
+class Green(BoardProxy):
+    """Green board manager object."""
+
+    def __init__(self) -> None:
+        """Initialize properties."""
+        super().__init__(BOARD_NAME_GREEN, SCHEMA_GREEN_BOARD)
+
+    @property
+    @dbus_property
+    def activity_led(self) -> bool:
+        """Get activity LED enabled."""
+        return self.properties[DBUS_ATTR_ACTIVITY_LED]
+
+    @activity_led.setter
+    def activity_led(self, enabled: bool) -> None:
+        """Enable/disable activity LED."""
+        self._data[ATTR_ACTIVITY_LED] = enabled
+        asyncio.create_task(self.dbus.Boards.Green.set_activity_led(enabled))
+
+    @property
+    @dbus_property
+    def power_led(self) -> bool:
+        """Get power LED enabled."""
+        return self.properties[DBUS_ATTR_POWER_LED]
+
+    @power_led.setter
+    def power_led(self, enabled: bool) -> None:
+        """Enable/disable power LED."""
+        self._data[ATTR_POWER_LED] = enabled
+        asyncio.create_task(self.dbus.Boards.Green.set_power_led(enabled))
+
+    @property
+    @dbus_property
+    def user_led(self) -> bool:
+        """Get user LED enabled."""
+        return self.properties[DBUS_ATTR_USER_LED]
+
+    @user_led.setter
+    def user_led(self, enabled: bool) -> None:
+        """Enable/disable disk LED."""
+        self._data[ATTR_USER_LED] = enabled
+        asyncio.create_task(self.dbus.Boards.Green.set_user_led(enabled))
+
+    async def connect(self, bus: MessageBus) -> None:
+        """Connect to D-Bus."""
+        await super().connect(bus)
+
+        # Set LEDs based on settings on connect
+        self.activity_led = self._data[ATTR_ACTIVITY_LED]
+        self.power_led = self._data[ATTR_POWER_LED]
+        self.user_led = self._data[ATTR_USER_LED]

supervisor/dbus/agent/boards/interface.py

@@ -1,17 +1,23 @@
 """Board dbus proxy interface."""
 
+from voluptuous import Schema
+
+from ....const import FILE_HASSIO_BOARD
+from ....utils.common import FileConfiguration
 from ...const import DBUS_IFACE_HAOS_BOARDS, DBUS_NAME_HAOS, DBUS_OBJECT_HAOS_BOARDS
 from ...interface import DBusInterfaceProxy
+from .validate import SCHEMA_BASE_BOARD
 
 
-class BoardProxy(DBusInterfaceProxy):
+class BoardProxy(FileConfiguration, DBusInterfaceProxy):
     """DBus interface proxy for os board."""
 
     bus_name: str = DBUS_NAME_HAOS
 
-    def __init__(self, name: str) -> None:
+    def __init__(self, name: str, file_schema: Schema | None = None) -> None:
         """Initialize properties."""
-        super().__init__()
+        super().__init__(FILE_HASSIO_BOARD, file_schema or SCHEMA_BASE_BOARD)
+        super(FileConfiguration, self).__init__()
 
         self._name: str = name
         self.object_path: str = f"{DBUS_OBJECT_HAOS_BOARDS}/{name}"

supervisor/dbus/agent/boards/validate.py

@@ -0,0 +1,32 @@
+"""Validation for board config."""
+
+import voluptuous as vol
+
+from ....const import (
+    ATTR_ACTIVITY_LED,
+    ATTR_DISK_LED,
+    ATTR_HEARTBEAT_LED,
+    ATTR_POWER_LED,
+    ATTR_USER_LED,
+)
+
+# pylint: disable=no-value-for-parameter
+SCHEMA_BASE_BOARD = vol.Schema({}, extra=vol.REMOVE_EXTRA)
+
+SCHEMA_GREEN_BOARD = vol.Schema(
+    {
+        vol.Optional(ATTR_ACTIVITY_LED, default=True): vol.Boolean(),
+        vol.Optional(ATTR_POWER_LED, default=True): vol.Boolean(),
+        vol.Optional(ATTR_USER_LED, default=True): vol.Boolean(),
+    },
+    extra=vol.REMOVE_EXTRA,
+)
+
+SCHEMA_YELLOW_BOARD = vol.Schema(
+    {
+        vol.Optional(ATTR_DISK_LED, default=True): vol.Boolean(),
+        vol.Optional(ATTR_HEARTBEAT_LED, default=True): vol.Boolean(),
+        vol.Optional(ATTR_POWER_LED, default=True): vol.Boolean(),
+    },
+    extra=vol.REMOVE_EXTRA,
+)

supervisor/dbus/agent/boards/yellow.py

@@ -2,10 +2,14 @@
 
 import asyncio
 
+from dbus_fast.aio.message_bus import MessageBus
+
+from ....const import ATTR_DISK_LED, ATTR_HEARTBEAT_LED, ATTR_POWER_LED
 from ...const import DBUS_ATTR_DISK_LED, DBUS_ATTR_HEARTBEAT_LED, DBUS_ATTR_POWER_LED
 from ...interface import dbus_property
 from .const import BOARD_NAME_YELLOW
 from .interface import BoardProxy
+from .validate import SCHEMA_YELLOW_BOARD
 
 
 class Yellow(BoardProxy):
@@ -13,7 +17,7 @@ class Yellow(BoardProxy):
 
     def __init__(self) -> None:
         """Initialize properties."""
-        super().__init__(BOARD_NAME_YELLOW)
+        super().__init__(BOARD_NAME_YELLOW, SCHEMA_YELLOW_BOARD)
 
     @property
     @dbus_property
@@ -24,6 +28,7 @@ class Yellow(BoardProxy):
     @heartbeat_led.setter
     def heartbeat_led(self, enabled: bool) -> None:
         """Enable/disable heartbeat LED."""
+        self._data[ATTR_HEARTBEAT_LED] = enabled
         asyncio.create_task(self.dbus.Boards.Yellow.set_heartbeat_led(enabled))
 
     @property
@@ -35,6 +40,7 @@ class Yellow(BoardProxy):
     @power_led.setter
     def power_led(self, enabled: bool) -> None:
         """Enable/disable power LED."""
+        self._data[ATTR_POWER_LED] = enabled
         asyncio.create_task(self.dbus.Boards.Yellow.set_power_led(enabled))
 
     @property
@@ -46,4 +52,14 @@ class Yellow(BoardProxy):
     @disk_led.setter
     def disk_led(self, enabled: bool) -> None:
         """Enable/disable disk LED."""
+        self._data[ATTR_DISK_LED] = enabled
         asyncio.create_task(self.dbus.Boards.Yellow.set_disk_led(enabled))
+
+    async def connect(self, bus: MessageBus) -> None:
+        """Connect to D-Bus."""
+        await super().connect(bus)
+
+        # Set LEDs based on settings on connect
+        self.disk_led = self._data[ATTR_DISK_LED]
+        self.heartbeat_led = self._data[ATTR_HEARTBEAT_LED]
+        self.power_led = self._data[ATTR_POWER_LED]

supervisor/dbus/const.py

@@ -64,6 +64,7 @@ DBUS_OBJECT_UDISKS2 = "/org/freedesktop/UDisks2/Manager"
 DBUS_ATTR_ACTIVE_ACCESSPOINT = "ActiveAccessPoint"
 DBUS_ATTR_ACTIVE_CONNECTION = "ActiveConnection"
 DBUS_ATTR_ACTIVE_CONNECTIONS = "ActiveConnections"
+DBUS_ATTR_ACTIVITY_LED = "ActivityLED"
 DBUS_ATTR_ADDRESS_DATA = "AddressData"
 DBUS_ATTR_BITRATE = "Bitrate"
 DBUS_ATTR_BOARD = "Board"
@@ -169,6 +170,7 @@ DBUS_ATTR_TIMEUSEC = "TimeUSec"
 DBUS_ATTR_TIMEZONE = "Timezone"
 DBUS_ATTR_TRANSACTION_STATISTICS = "TransactionStatistics"
 DBUS_ATTR_TYPE = "Type"
+DBUS_ATTR_USER_LED = "UserLED"
 DBUS_ATTR_USERSPACE_TIMESTAMP_MONOTONIC = "UserspaceTimestampMonotonic"
 DBUS_ATTR_UUID_UPPERCASE = "UUID"
 DBUS_ATTR_UUID = "Uuid"

supervisor/dbus/hostname.py

@@ -3,7 +3,7 @@ import logging
 
 from dbus_fast.aio.message_bus import MessageBus
 
-from ..exceptions import DBusError, DBusInterfaceError
+from ..exceptions import DBusError, DBusInterfaceError, DBusServiceUnkownError
 from .const import (
     DBUS_ATTR_CHASSIS,
     DBUS_ATTR_DEPLOYMENT,
@@ -39,7 +39,7 @@ class Hostname(DBusInterfaceProxy):
             await super().connect(bus)
         except DBusError:
             _LOGGER.warning("Can't connect to systemd-hostname")
-        except DBusInterfaceError:
+        except (DBusServiceUnkownError, DBusInterfaceError):
             _LOGGER.warning(
                 "No hostname support on the host. Hostname functions have been disabled."
             )

supervisor/dbus/logind.py

@@ -3,7 +3,7 @@ import logging
 
 from dbus_fast.aio.message_bus import MessageBus
 
-from ..exceptions import DBusError, DBusInterfaceError
+from ..exceptions import DBusError, DBusInterfaceError, DBusServiceUnkownError
 from .const import DBUS_NAME_LOGIND, DBUS_OBJECT_LOGIND
 from .interface import DBusInterface
 from .utils import dbus_connected
@@ -28,8 +28,8 @@ class Logind(DBusInterface):
             await super().connect(bus)
         except DBusError:
             _LOGGER.warning("Can't connect to systemd-logind")
-        except DBusInterfaceError:
-            _LOGGER.info("No systemd-logind support on the host.")
+        except (DBusServiceUnkownError, DBusInterfaceError):
+            _LOGGER.warning("No systemd-logind support on the host.")
 
     @dbus_connected
     async def reboot(self) -> None:

supervisor/dbus/network/__init__.py

@@ -9,6 +9,8 @@ from ...exceptions import (
     DBusError,
     DBusFatalError,
     DBusInterfaceError,
+    DBusNoReplyError,
+    DBusServiceUnkownError,
     HostNotSupportedError,
     NetworkInterfaceNotFound,
 )
@@ -143,7 +145,7 @@ class NetworkManager(DBusInterfaceProxy):
             await self.settings.connect(bus)
         except DBusError:
             _LOGGER.warning("Can't connect to Network Manager")
-        except DBusInterfaceError:
+        except (DBusServiceUnkownError, DBusInterfaceError):
             _LOGGER.warning(
                 "No Network Manager support on the host. Local network functions have been disabled."
             )
@@ -210,8 +212,22 @@ class NetworkManager(DBusInterfaceProxy):
                     # try to query it. Ignore those cases.
                     _LOGGER.debug("Can't process %s: %s", device, err)
                     continue
+                except (
+                    DBusNoReplyError,
+                    DBusServiceUnkownError,
+                ) as err:
+                    # This typically means that NetworkManager disappeared. Give up immeaditly.
+                    _LOGGER.error(
+                        "NetworkManager not responding while processing %s: %s. Giving up.",
+                        device,
+                        err,
+                    )
+                    capture_exception(err)
+                    return
                 except Exception as err:  # pylint: disable=broad-except
-                    _LOGGER.exception("Error while processing %s: %s", device, err)
+                    _LOGGER.exception(
+                        "Unkown error while processing %s: %s", device, err
+                    )
                     capture_exception(err)
                     continue
 

supervisor/dbus/network/dns.py

@@ -12,7 +12,7 @@ from ...const import (
     ATTR_PRIORITY,
     ATTR_VPN,
 )
-from ...exceptions import DBusError, DBusInterfaceError
+from ...exceptions import DBusError, DBusInterfaceError, DBusServiceUnkownError
 from ..const import (
     DBUS_ATTR_CONFIGURATION,
     DBUS_ATTR_MODE,
@@ -67,7 +67,7 @@ class NetworkManagerDNS(DBusInterfaceProxy):
             await super().connect(bus)
         except DBusError:
             _LOGGER.warning("Can't connect to DnsManager")
-        except DBusInterfaceError:
+        except (DBusServiceUnkownError, DBusInterfaceError):
             _LOGGER.warning(
                 "No DnsManager support on the host. Local DNS functions have been disabled."
             )

supervisor/dbus/network/setting/generate.py

@@ -2,7 +2,7 @@
 from __future__ import annotations
 
 import socket
-from typing import TYPE_CHECKING, Any
+from typing import TYPE_CHECKING
 from uuid import uuid4
 
 from dbus_fast import Variant
@@ -19,6 +19,7 @@ from . import (
     CONF_ATTR_PATH,
     CONF_ATTR_VLAN,
 )
+from .. import NetworkManager
 from ....host.const import InterfaceMethod, InterfaceType
 
 if TYPE_CHECKING:
@@ -26,8 +27,11 @@ if TYPE_CHECKING:
 
 
 def get_connection_from_interface(
-    interface: Interface, name: str | None = None, uuid: str | None = None
-) -> Any:
+    interface: Interface,
+    network_manager: NetworkManager,
+    name: str | None = None,
+    uuid: str | None = None,
+) -> dict[str, dict[str, Variant]]:
     """Generate message argument for network interface update."""
 
     # Generate/Update ID/name
@@ -121,9 +125,15 @@ def get_connection_from_interface(
     if interface.type == InterfaceType.ETHERNET:
         conn[CONF_ATTR_802_ETHERNET] = {ATTR_ASSIGNED_MAC: Variant("s", "preserve")}
     elif interface.type == "vlan":
+        parent = interface.vlan.interface
+        if parent in network_manager and (
+            parent_connection := network_manager.get(parent).connection
+        ):
+            parent = parent_connection.uuid
+
         conn[CONF_ATTR_VLAN] = {
             "id": Variant("u", interface.vlan.id),
-            "parent": Variant("s", interface.vlan.interface),
+            "parent": Variant("s", parent),
         }
     elif interface.type == InterfaceType.WIRELESS:
         wireless = {
@@ -138,8 +148,8 @@ def get_connection_from_interface(
             wireless["security"] = Variant("s", CONF_ATTR_802_WIRELESS_SECURITY)
             wireless_security = {}
             if interface.wifi.auth == "wep":
-                wireless_security["auth-alg"] = Variant("s", "none")
-                wireless_security["key-mgmt"] = Variant("s", "open")
+                wireless_security["auth-alg"] = Variant("s", "open")
+                wireless_security["key-mgmt"] = Variant("s", "none")
             elif interface.wifi.auth == "wpa-psk":
                 wireless_security["auth-alg"] = Variant("s", "open")
                 wireless_security["key-mgmt"] = Variant("s", "wpa-psk")

supervisor/dbus/network/settings.py

@@ -4,7 +4,7 @@ from typing import Any
 
 from dbus_fast.aio.message_bus import MessageBus
 
-from ...exceptions import DBusError, DBusInterfaceError
+from ...exceptions import DBusError, DBusInterfaceError, DBusServiceUnkownError
 from ..const import DBUS_NAME_NM, DBUS_OBJECT_SETTINGS
 from ..interface import DBusInterface
 from ..network.setting import NetworkSetting
@@ -28,7 +28,7 @@ class NetworkManagerSettings(DBusInterface):
             await super().connect(bus)
         except DBusError:
             _LOGGER.warning("Can't connect to Network Manager Settings")
-        except DBusInterfaceError:
+        except (DBusServiceUnkownError, DBusInterfaceError):
             _LOGGER.warning(
                 "No Network Manager Settings support on the host. Local network functions have been disabled."
             )

supervisor/dbus/rauc.py

@@ -4,7 +4,7 @@ from typing import Any
 
 from dbus_fast.aio.message_bus import MessageBus
 
-from ..exceptions import DBusError, DBusInterfaceError
+from ..exceptions import DBusError, DBusInterfaceError, DBusServiceUnkownError
 from ..utils.dbus import DBusSignalWrapper
 from .const import (
     DBUS_ATTR_BOOT_SLOT,
@@ -49,7 +49,7 @@ class Rauc(DBusInterfaceProxy):
             await super().connect(bus)
         except DBusError:
             _LOGGER.warning("Can't connect to rauc")
-        except DBusInterfaceError:
+        except (DBusServiceUnkownError, DBusInterfaceError):
             _LOGGER.warning("Host has no rauc support. OTA updates have been disabled.")
 
     @property

supervisor/dbus/resolved.py

@@ -5,7 +5,7 @@ import logging
 
 from dbus_fast.aio.message_bus import MessageBus
 
-from ..exceptions import DBusError, DBusInterfaceError
+from ..exceptions import DBusError, DBusInterfaceError, DBusServiceUnkownError
 from .const import (
     DBUS_ATTR_CACHE_STATISTICS,
     DBUS_ATTR_CURRENT_DNS_SERVER,
@@ -59,7 +59,7 @@ class Resolved(DBusInterfaceProxy):
             await super().connect(bus)
         except DBusError:
             _LOGGER.warning("Can't connect to systemd-resolved.")
-        except DBusInterfaceError:
+        except (DBusServiceUnkownError, DBusInterfaceError):
             _LOGGER.warning(
                 "Host has no systemd-resolved support. DNS will not work correctly."
             )

supervisor/dbus/systemd.py

@@ -10,6 +10,7 @@ from ..exceptions import (
     DBusError,
     DBusFatalError,
     DBusInterfaceError,
+    DBusServiceUnkownError,
     DBusSystemdNoSuchUnit,
 )
 from .const import (
@@ -86,7 +87,7 @@ class Systemd(DBusInterfaceProxy):
             await super().connect(bus)
         except DBusError:
             _LOGGER.warning("Can't connect to systemd")
-        except DBusInterfaceError:
+        except (DBusServiceUnkownError, DBusInterfaceError):
             _LOGGER.warning(
                 "No systemd support on the host. Host control has been disabled."
             )

supervisor/dbus/timedate.py

@@ -4,7 +4,7 @@ import logging
 
 from dbus_fast.aio.message_bus import MessageBus
 
-from ..exceptions import DBusError, DBusInterfaceError
+from ..exceptions import DBusError, DBusInterfaceError, DBusServiceUnkownError
 from ..utils.dt import utc_from_timestamp
 from .const import (
     DBUS_ATTR_NTP,
@@ -63,7 +63,7 @@ class TimeDate(DBusInterfaceProxy):
             await super().connect(bus)
         except DBusError:
             _LOGGER.warning("Can't connect to systemd-timedate")
-        except DBusInterfaceError:
+        except (DBusServiceUnkownError, DBusInterfaceError):
             _LOGGER.warning(
                 "No timedate support on the host. Time/Date functions have been disabled."
             )

supervisor/dbus/udisks2/__init__.py

@@ -6,7 +6,12 @@ from typing import Any
 from awesomeversion import AwesomeVersion
 from dbus_fast.aio import MessageBus
 
-from ...exceptions import DBusError, DBusInterfaceError, DBusObjectError
+from ...exceptions import (
+    DBusError,
+    DBusInterfaceError,
+    DBusObjectError,
+    DBusServiceUnkownError,
+)
 from ..const import (
     DBUS_ATTR_SUPPORTED_FILESYSTEMS,
     DBUS_ATTR_VERSION,
@@ -45,7 +50,7 @@ class UDisks2(DBusInterfaceProxy):
             await super().connect(bus)
         except DBusError:
             _LOGGER.warning("Can't connect to udisks2")
-        except DBusInterfaceError:
+        except (DBusServiceUnkownError, DBusInterfaceError):
             _LOGGER.warning(
                 "No udisks2 support on the host. Host control has been disabled."
             )

supervisor/dbus/udisks2/data.py

@@ -3,10 +3,9 @@
 from dataclasses import dataclass
 from inspect import get_annotations
 from pathlib import Path
-from typing import Any, TypedDict
+from typing import Any, NotRequired, TypedDict
 
 from dbus_fast import Variant
-from typing_extensions import NotRequired
 
 from .const import EncryptType, EraseMode
 

supervisor/docker/addon.py

@@ -15,15 +15,10 @@ from docker.types import Mount
 import requests
 
 from ..addons.build import AddonBuild
+from ..addons.const import MappingType
 from ..bus import EventListener
 from ..const import (
     DOCKER_CPU_RUNTIME_ALLOCATION,
-    MAP_ADDONS,
-    MAP_BACKUP,
-    MAP_CONFIG,
-    MAP_MEDIA,
-    MAP_SHARE,
-    MAP_SSL,
     SECURITY_DISABLE,
     SECURITY_PROFILE,
     SYSTEMD_JOURNAL_PERSISTENT,
@@ -329,75 +324,117 @@ class DockerAddon(DockerInterface):
         """Return mounts for container."""
         addon_mapping = self.addon.map_volumes
 
+        target_data_path = ""
+        if MappingType.DATA in addon_mapping:
+            target_data_path = addon_mapping[MappingType.DATA].path
+
         mounts = [
             MOUNT_DEV,
             Mount(
                 type=MountType.BIND,
                 source=self.addon.path_extern_data.as_posix(),
-                target="/data",
+                target=target_data_path or "/data",
                 read_only=False,
             ),
         ]
 
         # setup config mappings
-        if MAP_CONFIG in addon_mapping:
+        if MappingType.CONFIG in addon_mapping:
             mounts.append(
                 Mount(
                     type=MountType.BIND,
                     source=self.sys_config.path_extern_homeassistant.as_posix(),
-                    target="/config",
-                    read_only=addon_mapping[MAP_CONFIG],
+                    target=addon_mapping[MappingType.CONFIG].path or "/config",
+                    read_only=addon_mapping[MappingType.CONFIG].read_only,
                 )
             )
 
-        if MAP_SSL in addon_mapping:
+        else:
+            # Map addon's public config folder if not using deprecated config option
+            if self.addon.addon_config_used:
+                mounts.append(
+                    Mount(
+                        type=MountType.BIND,
+                        source=self.addon.path_extern_config.as_posix(),
+                        target=addon_mapping[MappingType.ADDON_CONFIG].path
+                        or "/config",
+                        read_only=addon_mapping[MappingType.ADDON_CONFIG].read_only,
+                    )
+                )
+
+            # Map Home Assistant config in new way
+            if MappingType.HOMEASSISTANT_CONFIG in addon_mapping:
+                mounts.append(
+                    Mount(
+                        type=MountType.BIND,
+                        source=self.sys_config.path_extern_homeassistant.as_posix(),
+                        target=addon_mapping[MappingType.HOMEASSISTANT_CONFIG].path
+                        or "/homeassistant",
+                        read_only=addon_mapping[
+                            MappingType.HOMEASSISTANT_CONFIG
+                        ].read_only,
+                    )
+                )
+
+        if MappingType.ALL_ADDON_CONFIGS in addon_mapping:
+            mounts.append(
+                Mount(
+                    type=MountType.BIND,
+                    source=self.sys_config.path_extern_addon_configs.as_posix(),
+                    target=addon_mapping[MappingType.ALL_ADDON_CONFIGS].path
+                    or "/addon_configs",
+                    read_only=addon_mapping[MappingType.ALL_ADDON_CONFIGS].read_only,
+                )
+            )
+
+        if MappingType.SSL in addon_mapping:
             mounts.append(
                 Mount(
                     type=MountType.BIND,
                     source=self.sys_config.path_extern_ssl.as_posix(),
-                    target="/ssl",
-                    read_only=addon_mapping[MAP_SSL],
+                    target=addon_mapping[MappingType.SSL].path or "/ssl",
+                    read_only=addon_mapping[MappingType.SSL].read_only,
                 )
             )
 
-        if MAP_ADDONS in addon_mapping:
+        if MappingType.ADDONS in addon_mapping:
             mounts.append(
                 Mount(
                     type=MountType.BIND,
                     source=self.sys_config.path_extern_addons_local.as_posix(),
-                    target="/addons",
-                    read_only=addon_mapping[MAP_ADDONS],
+                    target=addon_mapping[MappingType.ADDONS].path or "/addons",
+                    read_only=addon_mapping[MappingType.ADDONS].read_only,
                 )
             )
 
-        if MAP_BACKUP in addon_mapping:
+        if MappingType.BACKUP in addon_mapping:
             mounts.append(
                 Mount(
                     type=MountType.BIND,
                     source=self.sys_config.path_extern_backup.as_posix(),
-                    target="/backup",
-                    read_only=addon_mapping[MAP_BACKUP],
+                    target=addon_mapping[MappingType.BACKUP].path or "/backup",
+                    read_only=addon_mapping[MappingType.BACKUP].read_only,
                 )
             )
 
-        if MAP_SHARE in addon_mapping:
+        if MappingType.SHARE in addon_mapping:
             mounts.append(
                 Mount(
                     type=MountType.BIND,
                     source=self.sys_config.path_extern_share.as_posix(),
-                    target="/share",
-                    read_only=addon_mapping[MAP_SHARE],
+                    target=addon_mapping[MappingType.SHARE].path or "/share",
+                    read_only=addon_mapping[MappingType.SHARE].read_only,
                     propagation=PropagationMode.RSLAVE,
                 )
             )
 
-        if MAP_MEDIA in addon_mapping:
+        if MappingType.MEDIA in addon_mapping:
             mounts.append(
                 Mount(
                     type=MountType.BIND,
                     source=self.sys_config.path_extern_media.as_posix(),
-                    target="/media",
-                    read_only=addon_mapping[MAP_MEDIA],
+                    target=addon_mapping[MappingType.MEDIA].path or "/media",
+                    read_only=addon_mapping[MappingType.MEDIA].read_only,
                     propagation=PropagationMode.RSLAVE,
                 )
             )
@@ -501,24 +538,16 @@ class DockerAddon(DockerInterface):
     )
     async def run(self) -> None:
         """Run Docker image."""
-        if await self.is_running():
-            return
-
         # Security check
         if not self.addon.protected:
             _LOGGER.warning("%s running with disabled protected mode!", self.addon.name)
 
-        # Cleanup
-        await self.stop()
-
         # Don't set a hostname if no separate UTS namespace is used
         hostname = None if self.uts_mode else self.addon.hostname
 
         # Create & Run container
         try:
-            docker_container = await self.sys_run_in_executor(
-                self.sys_docker.run,
-                self.image,
+            await self._run(
                 tag=str(self.addon.version),
                 name=self.name,
                 hostname=hostname,
@@ -549,7 +578,6 @@ class DockerAddon(DockerInterface):
             )
             raise
 
-        self._meta = docker_container.attrs
         _LOGGER.info(
             "Starting Docker add-on %s with version %s", self.image, self.version
         )
@@ -575,7 +603,11 @@ class DockerAddon(DockerInterface):
         on_condition=DockerJobError,
     )
     async def update(
-        self, version: AwesomeVersion, image: str | None = None, latest: bool = False
+        self,
+        version: AwesomeVersion,
+        image: str | None = None,
+        latest: bool = False,
+        arch: CpuArch | None = None,
     ) -> None:
         """Update a docker image."""
         image = image or self.image
@@ -586,13 +618,13 @@ class DockerAddon(DockerInterface):
 
         # Update docker image
         await self.install(
-            version, image=image, latest=latest, need_build=self.addon.latest_need_build
+            version,
+            image=image,
+            latest=latest,
+            arch=arch,
+            need_build=self.addon.latest_need_build,
         )
 
-        # Stop container & cleanup
-        with suppress(DockerError):
-            await self.stop()
-
     @Job(
         name="docker_addon_install",
         limit=JobExecutionLimit.GROUP_ONCE,

supervisor/docker/audio.py

@@ -92,16 +92,7 @@ class DockerAudio(DockerInterface, CoreSysAttributes):
     )
     async def run(self) -> None:
         """Run Docker image."""
-        if await self.is_running():
-            return
-
-        # Cleanup
-        await self.stop()
-
-        # Create & Run container
-        docker_container = await self.sys_run_in_executor(
-            self.sys_docker.run,
-            self.image,
+        await self._run(
             tag=str(self.sys_plugins.audio.version),
             init=False,
             ipv4=self.sys_docker.network.audio,
@@ -118,8 +109,6 @@ class DockerAudio(DockerInterface, CoreSysAttributes):
             },
             mounts=self.mounts,
         )
-
-        self._meta = docker_container.attrs
         _LOGGER.info(
             "Starting Audio %s with version %s - %s",
             self.image,

supervisor/docker/cli.py

@@ -33,16 +33,7 @@ class DockerCli(DockerInterface, CoreSysAttributes):
     )
     async def run(self) -> None:
         """Run Docker image."""
-        if await self.is_running():
-            return
-
-        # Cleanup
-        await self.stop()
-
-        # Create & Run container
-        docker_container = await self.sys_run_in_executor(
-            self.sys_docker.run,
-            self.image,
+        await self._run(
             entrypoint=["/init"],
             tag=str(self.sys_plugins.cli.version),
             init=False,
@@ -60,8 +51,6 @@ class DockerCli(DockerInterface, CoreSysAttributes):
                 ENV_TOKEN: self.sys_plugins.cli.supervisor_token,
             },
         )
-
-        self._meta = docker_container.attrs
         _LOGGER.info(
             "Starting CLI %s with version %s - %s",
             self.image,

supervisor/docker/dns.py

@@ -35,16 +35,7 @@ class DockerDNS(DockerInterface, CoreSysAttributes):
     )
     async def run(self) -> None:
         """Run Docker image."""
-        if await self.is_running():
-            return
-
-        # Cleanup
-        await self.stop()
-
-        # Create & Run container
-        docker_container = await self.sys_run_in_executor(
-            self.sys_docker.run,
-            self.image,
+        await self._run(
             tag=str(self.sys_plugins.dns.version),
             init=False,
             dns=False,
@@ -65,8 +56,6 @@ class DockerDNS(DockerInterface, CoreSysAttributes):
             ],
             oom_score_adj=-300,
         )
-
-        self._meta = docker_container.attrs
         _LOGGER.info(
             "Starting DNS %s with version %s - %s",
             self.image,

supervisor/docker/homeassistant.py

@@ -53,9 +53,10 @@ class DockerHomeAssistant(DockerInterface):
     @property
     def timeout(self) -> int:
         """Return timeout for Docker actions."""
-        # Synchronized homeassistant's S6_SERVICES_GRACETIME
-        # to avoid killing Home Assistant Core
-        return 220 + 20
+        # Synchronized with the homeassistant core container's S6_SERVICES_GRACETIME
+        # to avoid killing Home Assistant Core, see
+        # https://github.com/home-assistant/core/tree/dev/Dockerfile
+        return 240 + 20
 
     @property
     def ip_address(self) -> IPv4Address:
@@ -152,16 +153,7 @@ class DockerHomeAssistant(DockerInterface):
     )
     async def run(self) -> None:
         """Run Docker image."""
-        if await self.is_running():
-            return
-
-        # Cleanup
-        await self.stop()
-
-        # Create & Run container
-        docker_container = await self.sys_run_in_executor(
-            self.sys_docker.run,
-            self.image,
+        await self._run(
             tag=(self.sys_homeassistant.version),
             name=self.name,
             hostname=self.name,
@@ -186,8 +178,6 @@ class DockerHomeAssistant(DockerInterface):
             tmpfs={"/tmp": ""},
             oom_score_adj=-300,
         )
-
-        self._meta = docker_container.attrs
         _LOGGER.info(
             "Starting Home Assistant %s with version %s", self.image, self.version
         )

supervisor/docker/interface.py

@@ -1,7 +1,6 @@
 """Interface class for Supervisor Docker object."""
 from __future__ import annotations
 
-import asyncio
 from collections import defaultdict
 from collections.abc import Awaitable
 from contextlib import suppress
@@ -92,7 +91,6 @@ class DockerInterface(JobGroup):
         )
         self.coresys: CoreSys = coresys
         self._meta: dict[str, Any] | None = None
-        self.lock: asyncio.Lock = asyncio.Lock()
 
     @property
     def timeout(self) -> int:
@@ -153,7 +151,7 @@ class DockerInterface(JobGroup):
     @property
     def in_progress(self) -> bool:
         """Return True if a task is in progress."""
-        return self.lock.locked()
+        return self.active_job
 
     @property
     def restart_policy(self) -> RestartPolicy | None:
@@ -379,6 +377,27 @@ class DockerInterface(JobGroup):
         """Run Docker image."""
         raise NotImplementedError()
 
+    async def _run(self, **kwargs) -> None:
+        """Run Docker image with retry inf necessary."""
+        if await self.is_running():
+            return
+
+        # Cleanup
+        await self.stop()
+
+        # Create & Run container
+        try:
+            docker_container = await self.sys_run_in_executor(
+                self.sys_docker.run, self.image, **kwargs
+            )
+        except DockerNotFound as err:
+            # If image is missing, capture the exception as this shouldn't happen
+            capture_exception(err)
+            raise
+
+        # Store metadata
+        self._meta = docker_container.attrs
+
     @Job(
         name="docker_interface_stop",
         limit=JobExecutionLimit.GROUP_ONCE,
@@ -451,12 +470,17 @@ class DockerInterface(JobGroup):
         return b""
 
     @Job(name="docker_interface_cleanup", limit=JobExecutionLimit.GROUP_WAIT)
-    def cleanup(self, old_image: str | None = None) -> Awaitable[None]:
+    def cleanup(
+        self,
+        old_image: str | None = None,
+        image: str | None = None,
+        version: AwesomeVersion | None = None,
+    ) -> Awaitable[None]:
         """Check if old version exists and cleanup."""
         return self.sys_run_in_executor(
             self.sys_docker.cleanup_old_images,
-            self.image,
-            self.version,
+            image or self.image,
+            version or self.version,
             {old_image} if old_image else None,
         )
 

supervisor/docker/multicast.py

@@ -38,16 +38,7 @@ class DockerMulticast(DockerInterface, CoreSysAttributes):
     )
     async def run(self) -> None:
         """Run Docker image."""
-        if await self.is_running():
-            return
-
-        # Cleanup
-        await self.stop()
-
-        # Create & Run container
-        docker_container = await self.sys_run_in_executor(
-            self.sys_docker.run,
-            self.image,
+        await self._run(
             tag=str(self.sys_plugins.multicast.version),
             init=False,
             name=self.name,
@@ -59,8 +50,6 @@ class DockerMulticast(DockerInterface, CoreSysAttributes):
             extra_hosts={"supervisor": self.sys_docker.network.supervisor},
             environment={ENV_TIME: self.sys_timezone},
         )
-
-        self._meta = docker_container.attrs
         _LOGGER.info(
             "Starting Multicast %s with version %s - Host", self.image, self.version
         )

supervisor/docker/observer.py

@@ -35,16 +35,7 @@ class DockerObserver(DockerInterface, CoreSysAttributes):
     )
     async def run(self) -> None:
         """Run Docker image."""
-        if await self.is_running():
-            return
-
-        # Cleanup
-        await self.stop()
-
-        # Create & Run container
-        docker_container = await self.sys_run_in_executor(
-            self.sys_docker.run,
-            self.image,
+        await self._run(
             tag=str(self.sys_plugins.observer.version),
             init=False,
             ipv4=self.sys_docker.network.observer,
@@ -63,8 +54,6 @@ class DockerObserver(DockerInterface, CoreSysAttributes):
             ports={"80/tcp": 4357},
             oom_score_adj=-300,
         )
-
-        self._meta = docker_container.attrs
         _LOGGER.info(
             "Starting Observer %s with version %s - %s",
             self.image,

supervisor/exceptions.py

@@ -67,6 +67,10 @@ class HomeAssistantCrashError(HomeAssistantError):
     """Error on crash of a Home Assistant startup."""
 
 
+class HomeAssistantStartupTimeout(HomeAssistantCrashError):
+    """Timeout waiting for Home Assistant successful startup."""
+
+
 class HomeAssistantAPIError(HomeAssistantError):
     """Home Assistant API exception."""
 
@@ -331,6 +335,10 @@ class DBusNotConnectedError(HostNotSupportedError):
     """D-Bus is not connected and call a method."""
 
 
+class DBusServiceUnkownError(HassioNotSupportedError):
+    """D-Bus service was not available."""
+
+
 class DBusInterfaceError(HassioNotSupportedError):
     """D-Bus interface not connected."""
 
@@ -359,6 +367,10 @@ class DBusTimeoutError(DBusError):
     """D-Bus call timed out."""
 
 
+class DBusNoReplyError(DBusError):
+    """D-Bus remote didn't reply/disconnected."""
+
+
 class DBusFatalError(DBusError):
     """D-Bus call going wrong.
 
@@ -581,6 +593,14 @@ class HomeAssistantBackupError(BackupError, HomeAssistantError):
     """Raise if an error during Home Assistant Core backup is happening."""
 
 
+class BackupInvalidError(BackupError):
+    """Raise if backup or password provided is invalid."""
+
+
+class BackupJobError(BackupError, JobException):
+    """Raise on Backup job error."""
+
+
 # Security
 
 

supervisor/hardware/helper.py

@@ -1,5 +1,5 @@
 """Read hardware info from system."""
-from datetime import datetime
+from datetime import UTC, datetime
 import logging
 from pathlib import Path
 import re
@@ -55,7 +55,7 @@ class HwHelper(CoreSysAttributes):
             _LOGGER.error("Can't found last boot time!")
             return None
 
-        return datetime.utcfromtimestamp(int(found.group(1)))
+        return datetime.fromtimestamp(int(found.group(1)), UTC)
 
     def hide_virtual_device(self, udev_device: pyudev.Device) -> bool:
         """Small helper to hide not needed Devices."""

supervisor/homeassistant/api.py

@@ -13,7 +13,7 @@ from ..coresys import CoreSys, CoreSysAttributes
 from ..exceptions import HomeAssistantAPIError, HomeAssistantAuthError
 from ..jobs.const import JobExecutionLimit
 from ..jobs.decorator import Job
-from ..utils import check_port
+from ..utils import check_port, version_is_new_enough
 from .const import LANDINGPAGE
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
@@ -107,7 +107,7 @@ class HomeAssistantAPI(CoreSysAttributes):
                         continue
                     yield resp
                     return
-            except (asyncio.TimeoutError, aiohttp.ClientError) as err:
+            except (TimeoutError, aiohttp.ClientError) as err:
                 _LOGGER.error("Error on call %s: %s", url, err)
                 break
 
@@ -130,34 +130,41 @@ class HomeAssistantAPI(CoreSysAttributes):
         """Return Home Assistant core state."""
         return await self._get_json("api/core/state")
 
-    async def check_api_state(self) -> bool:
-        """Return True if Home Assistant up and running."""
+    async def get_api_state(self) -> str | None:
+        """Return state of Home Assistant Core or None."""
         # Skip check on landingpage
         if (
             self.sys_homeassistant.version is None
             or self.sys_homeassistant.version == LANDINGPAGE
         ):
-            return False
+            return None
 
         # Check if port is up
-        if not await self.sys_run_in_executor(
-            check_port,
+        if not await check_port(
             self.sys_homeassistant.ip_address,
             self.sys_homeassistant.api_port,
         ):
-            return False
+            return None
 
         # Check if API is up
         with suppress(HomeAssistantAPIError):
             # get_core_state is available since 2023.8.0 and preferred
             # since it is significantly faster than get_config because
             # it does not require serializing the entire config
-            if self.sys_homeassistant.version >= GET_CORE_STATE_MIN_VERSION:
+            if version_is_new_enough(
+                self.sys_homeassistant.version, GET_CORE_STATE_MIN_VERSION
+            ):
                 data = await self.get_core_state()
             else:
                 data = await self.get_config()
             # Older versions of home assistant does not expose the state
-            if data and data.get("state", "RUNNING") == "RUNNING":
-                return True
+            if data:
+                return data.get("state", "RUNNING")
 
+        return None
+
+    async def check_api_state(self) -> bool:
+        """Return Home Assistant Core state if up."""
+        if state := await self.get_api_state():
+            return state == "RUNNING"
         return False

supervisor/homeassistant/core.py

@@ -2,12 +2,14 @@
 import asyncio
 from collections.abc import Awaitable
 from contextlib import suppress
+from dataclasses import dataclass
+from datetime import datetime, timedelta
 import logging
 import re
 import secrets
 import shutil
+from typing import Final
 
-import attr
 from awesomeversion import AwesomeVersion
 
 from ..const import ATTR_HOMEASSISTANT, BusEvent
@@ -21,6 +23,7 @@ from ..exceptions import (
     HomeAssistantCrashError,
     HomeAssistantError,
     HomeAssistantJobError,
+    HomeAssistantStartupTimeout,
     HomeAssistantUpdateError,
     JobException,
 )
@@ -40,15 +43,20 @@ from .const import (
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
 
+SECONDS_BETWEEN_API_CHECKS: Final[int] = 5
+# Core Stage 1 and some wiggle room
+STARTUP_API_RESPONSE_TIMEOUT: Final[timedelta] = timedelta(minutes=3)
+# All stages plus event start timeout and some wiggle rooom
+STARTUP_API_CHECK_RUNNING_TIMEOUT: Final[timedelta] = timedelta(minutes=15)
 RE_YAML_ERROR = re.compile(r"homeassistant\.util\.yaml")
 
 
-@attr.s(frozen=True)
+@dataclass
 class ConfigResult:
     """Return object from config check."""
 
-    valid = attr.ib()
-    log = attr.ib()
+    valid: bool
+    log: str
 
 
 class HomeAssistantCore(JobGroup):
@@ -58,7 +66,6 @@ class HomeAssistantCore(JobGroup):
         """Initialize Home Assistant object."""
         super().__init__(coresys, JOB_GROUP_HOME_ASSISTANT_CORE)
         self.instance: DockerHomeAssistant = DockerHomeAssistant(coresys)
-        self.lock: asyncio.Lock = asyncio.Lock()
         self._error_state: bool = False
 
     @property
@@ -122,7 +129,7 @@ class HomeAssistantCore(JobGroup):
         while True:
             if not self.sys_updater.image_homeassistant:
                 _LOGGER.warning(
-                    "Found no information about Home Assistant. Retry in 30sec"
+                    "Found no information about Home Assistant. Retrying in 30sec"
                 )
                 await asyncio.sleep(30)
                 await self.sys_updater.reload()
@@ -138,7 +145,7 @@ class HomeAssistantCore(JobGroup):
             except Exception as err:  # pylint: disable=broad-except
                 capture_exception(err)
 
-            _LOGGER.warning("Fails install landingpage, retry after 30sec")
+            _LOGGER.warning("Failed to install landingpage, retrying after 30sec")
             await asyncio.sleep(30)
 
         self.sys_homeassistant.version = LANDINGPAGE
@@ -170,7 +177,7 @@ class HomeAssistantCore(JobGroup):
                 except Exception as err:  # pylint: disable=broad-except
                     capture_exception(err)
 
-            _LOGGER.warning("Error on Home Assistant installation. Retry in 30sec")
+            _LOGGER.warning("Error on Home Assistant installation. Retrying in 30sec")
             await asyncio.sleep(30)
 
         _LOGGER.info("Home Assistant docker now installed")
@@ -402,7 +409,7 @@ class HomeAssistantCore(JobGroup):
     @property
     def in_progress(self) -> bool:
         """Return True if a task is in progress."""
-        return self.instance.in_progress or self.lock.locked()
+        return self.instance.in_progress or self.active_job
 
     async def check_config(self) -> ConfigResult:
         """Run Home Assistant config check."""
@@ -436,21 +443,38 @@ class HomeAssistantCore(JobGroup):
             return
         _LOGGER.info("Wait until Home Assistant is ready")
 
-        while True:
-            await asyncio.sleep(5)
+        deadline = datetime.now() + STARTUP_API_RESPONSE_TIMEOUT
+        last_state = None
+        while not (timeout := datetime.now() >= deadline):
+            await asyncio.sleep(SECONDS_BETWEEN_API_CHECKS)
 
             # 1: Check if Container is is_running
             if not await self.instance.is_running():
                 _LOGGER.error("Home Assistant has crashed!")
                 break
 
-            # 2: Check if API response
-            if await self.sys_homeassistant.api.check_api_state():
-                _LOGGER.info("Detect a running Home Assistant instance")
-                self._error_state = False
-                return
+            # 2: Check API response
+            if state := await self.sys_homeassistant.api.get_api_state():
+                if last_state is None:
+                    # API initially available, move deadline up and check API
+                    # state to be running now
+                    deadline = datetime.now() + STARTUP_API_CHECK_RUNNING_TIMEOUT
+
+                if last_state != state:
+                    _LOGGER.info("Home Assistant Core state changed to %s", state)
+                    last_state = state
+
+                if state == "RUNNING":
+                    _LOGGER.info("Detect a running Home Assistant instance")
+                    self._error_state = False
+                    return
 
         self._error_state = True
+        if timeout:
+            raise HomeAssistantStartupTimeout(
+                "No Home Assistant Core response, assuming a fatal startup error",
+                _LOGGER.error,
+            )
         raise HomeAssistantCrashError()
 
     @Job(

supervisor/homeassistant/module.py

@@ -1,6 +1,7 @@
 """Home Assistant control object."""
 import asyncio
 from datetime import timedelta
+import errno
 from ipaddress import IPv4Address
 import logging
 from pathlib import Path, PurePath
@@ -18,6 +19,7 @@ from ..const import (
     ATTR_ACCESS_TOKEN,
     ATTR_AUDIO_INPUT,
     ATTR_AUDIO_OUTPUT,
+    ATTR_BACKUPS_EXCLUDE_DATABASE,
     ATTR_BOOT,
     ATTR_IMAGE,
     ATTR_PORT,
@@ -41,6 +43,7 @@ from ..exceptions import (
 from ..hardware.const import PolicyGroup
 from ..hardware.data import Device
 from ..jobs.decorator import Job, JobExecutionLimit
+from ..resolution.const import UnhealthyReason
 from ..utils import remove_folder
 from ..utils.common import FileConfiguration
 from ..utils.json import read_json_file, write_json_file
@@ -62,6 +65,10 @@ HOMEASSISTANT_BACKUP_EXCLUDE = [
     "*.log.*",
     "OZW_Log.txt",
 ]
+HOMEASSISTANT_BACKUP_EXCLUDE_DATABASE = [
+    "home-assistant_v?.db",
+    "home-assistant_v?.db-wal",
+]
 
 
 class HomeAssistant(FileConfiguration, CoreSysAttributes):
@@ -258,6 +265,16 @@ class HomeAssistant(FileConfiguration, CoreSysAttributes):
         except (AwesomeVersionException, TypeError):
             return False
 
+    @property
+    def backups_exclude_database(self) -> bool:
+        """Exclude database from core backups by default."""
+        return self._data[ATTR_BACKUPS_EXCLUDE_DATABASE]
+
+    @backups_exclude_database.setter
+    def backups_exclude_database(self, value: bool) -> None:
+        """Set whether backups should exclude database by default."""
+        self._data[ATTR_BACKUPS_EXCLUDE_DATABASE] = value
+
     async def load(self) -> None:
         """Prepare Home Assistant object."""
         await asyncio.wait(
@@ -285,6 +302,8 @@ class HomeAssistant(FileConfiguration, CoreSysAttributes):
         try:
             self.path_pulse.write_text(pulse_config, encoding="utf-8")
         except OSError as err:
+            if err.errno == errno.EBADMSG:
+                self.sys_resolution.unhealthy = UnhealthyReason.OSERROR_BAD_MESSAGE
             _LOGGER.error("Home Assistant can't write pulse/client.config: %s", err)
         else:
             _LOGGER.info("Update pulse/client.config: %s", self.path_pulse)
@@ -306,64 +325,82 @@ class HomeAssistant(FileConfiguration, CoreSysAttributes):
 
         self.sys_homeassistant.websocket.send_message({ATTR_TYPE: "usb/scan"})
 
-    @Job(name="home_assistant_module_backup")
-    async def backup(self, tar_file: tarfile.TarFile) -> None:
-        """Backup Home Assistant Core config/ directory."""
-
-        # Let Home Assistant Core know we are about to backup
+    @Job(name="home_assistant_module_begin_backup")
+    async def begin_backup(self) -> None:
+        """Inform Home Assistant a backup is beginning."""
         try:
             await self.websocket.async_send_command({ATTR_TYPE: WSType.BACKUP_START})
-
         except HomeAssistantWSError:
             _LOGGER.warning(
                 "Preparing backup of Home Assistant Core failed. Check HA Core logs."
             )
 
-        with TemporaryDirectory(dir=self.sys_config.path_tmp) as temp:
-            temp_path = Path(temp)
+    @Job(name="home_assistant_module_end_backup")
+    async def end_backup(self) -> None:
+        """Inform Home Assistant the backup is ending."""
+        try:
+            await self.websocket.async_send_command({ATTR_TYPE: WSType.BACKUP_END})
+        except HomeAssistantWSError:
+            _LOGGER.warning(
+                "Error during Home Assistant Core backup. Check HA Core logs."
+            )
 
-            # Store local configs/state
-            try:
-                write_json_file(temp_path.joinpath("homeassistant.json"), self._data)
-            except ConfigurationFileError as err:
-                raise HomeAssistantError(
-                    f"Can't save meta for Home Assistant Core: {err!s}", _LOGGER.error
-                ) from err
+    @Job(name="home_assistant_module_backup")
+    async def backup(
+        self, tar_file: tarfile.TarFile, exclude_database: bool = False
+    ) -> None:
+        """Backup Home Assistant Core config/ directory."""
+        await self.begin_backup()
+        try:
+            with TemporaryDirectory(dir=self.sys_config.path_tmp) as temp:
+                temp_path = Path(temp)
 
-            # Backup data config folder
-            def _write_tarfile():
-                with tar_file as backup:
-                    # Backup metadata
-                    backup.add(temp, arcname=".")
-
-                    # Backup data
-                    atomic_contents_add(
-                        backup,
-                        self.sys_config.path_homeassistant,
-                        excludes=HOMEASSISTANT_BACKUP_EXCLUDE,
-                        arcname="data",
-                    )
-
-            try:
-                _LOGGER.info("Backing up Home Assistant Core config folder")
-                await self.sys_run_in_executor(_write_tarfile)
-                _LOGGER.info("Backup Home Assistant Core config folder done")
-            except (tarfile.TarError, OSError) as err:
-                raise HomeAssistantBackupError(
-                    f"Can't backup Home Assistant Core config folder: {str(err)}",
-                    _LOGGER.error,
-                ) from err
-            finally:
+                # Store local configs/state
                 try:
-                    await self.sys_homeassistant.websocket.async_send_command(
-                        {ATTR_TYPE: WSType.BACKUP_END}
-                    )
-                except HomeAssistantWSError:
-                    _LOGGER.warning(
-                        "Error during Home Assistant Core backup. Check HA Core logs."
+                    write_json_file(
+                        temp_path.joinpath("homeassistant.json"), self._data
                     )
+                except ConfigurationFileError as err:
+                    raise HomeAssistantError(
+                        f"Can't save meta for Home Assistant Core: {err!s}",
+                        _LOGGER.error,
+                    ) from err
 
-    async def restore(self, tar_file: tarfile.TarFile) -> None:
+                # Backup data config folder
+                def _write_tarfile():
+                    with tar_file as backup:
+                        # Backup metadata
+                        backup.add(temp, arcname=".")
+
+                        # Set excludes
+                        excludes = HOMEASSISTANT_BACKUP_EXCLUDE.copy()
+                        if exclude_database:
+                            excludes += HOMEASSISTANT_BACKUP_EXCLUDE_DATABASE
+
+                        # Backup data
+                        atomic_contents_add(
+                            backup,
+                            self.sys_config.path_homeassistant,
+                            excludes=excludes,
+                            arcname="data",
+                        )
+
+                try:
+                    _LOGGER.info("Backing up Home Assistant Core config folder")
+                    await self.sys_run_in_executor(_write_tarfile)
+                    _LOGGER.info("Backup Home Assistant Core config folder done")
+                except (tarfile.TarError, OSError) as err:
+                    raise HomeAssistantBackupError(
+                        f"Can't backup Home Assistant Core config folder: {str(err)}",
+                        _LOGGER.error,
+                    ) from err
+        finally:
+            await self.end_backup()
+
+    @Job(name="home_assistant_module_restore")
+    async def restore(
+        self, tar_file: tarfile.TarFile, exclude_database: bool = False
+    ) -> None:
         """Restore Home Assistant Core config/ directory."""
         with TemporaryDirectory(dir=self.sys_config.path_tmp) as temp:
             temp_path = Path(temp)
@@ -374,7 +411,11 @@ class HomeAssistant(FileConfiguration, CoreSysAttributes):
             def _extract_tarfile():
                 """Extract tar backup."""
                 with tar_file as backup:
-                    backup.extractall(path=temp_path, members=secure_path(backup))
+                    backup.extractall(
+                        path=temp_path,
+                        members=secure_path(backup),
+                        filter="fully_trusted",
+                    )
 
             try:
                 await self.sys_run_in_executor(_extract_tarfile)
@@ -391,11 +432,22 @@ class HomeAssistant(FileConfiguration, CoreSysAttributes):
             def _restore_data():
                 """Restore data."""
                 shutil.copytree(
-                    temp_data, self.sys_config.path_homeassistant, symlinks=True
+                    temp_data,
+                    self.sys_config.path_homeassistant,
+                    symlinks=True,
+                    dirs_exist_ok=True,
                 )
 
             _LOGGER.info("Restore Home Assistant Core config folder")
-            await remove_folder(self.sys_config.path_homeassistant)
+            excludes = (
+                HOMEASSISTANT_BACKUP_EXCLUDE_DATABASE if exclude_database else None
+            )
+            await remove_folder(
+                self.sys_config.path_homeassistant,
+                content_only=True,
+                excludes=excludes,
+                tmp_dir=self.sys_config.path_tmp,
+            )
             try:
                 await self.sys_run_in_executor(_restore_data)
             except shutil.Error as err:
@@ -433,7 +485,8 @@ class HomeAssistant(FileConfiguration, CoreSysAttributes):
                 ATTR_REFRESH_TOKEN,
                 ATTR_WATCHDOG,
             ):
-                self._data[attr] = data[attr]
+                if attr in data:
+                    self._data[attr] = data[attr]
 
     @Job(
         name="home_assistant_get_users",
@@ -447,9 +500,13 @@ class HomeAssistant(FileConfiguration, CoreSysAttributes):
             {ATTR_TYPE: "config/auth/list"}
         )
 
-        return [
-            IngressSessionDataUser(
-                id=data["id"], username=data["username"], display_name=data["name"]
-            )
-            for data in list_of_users
-        ]
+        if list_of_users:
+            return [
+                IngressSessionDataUser(
+                    id=data["id"],
+                    username=data.get("username"),
+                    display_name=data.get("name"),
+                )
+                for data in list_of_users
+            ]
+        return []

supervisor/homeassistant/validate.py

@@ -7,6 +7,7 @@ from ..const import (
     ATTR_ACCESS_TOKEN,
     ATTR_AUDIO_INPUT,
     ATTR_AUDIO_OUTPUT,
+    ATTR_BACKUPS_EXCLUDE_DATABASE,
     ATTR_BOOT,
     ATTR_IMAGE,
     ATTR_PORT,
@@ -32,6 +33,7 @@ SCHEMA_HASS_CONFIG = vol.Schema(
         vol.Optional(ATTR_WATCHDOG, default=True): vol.Boolean(),
         vol.Optional(ATTR_AUDIO_OUTPUT, default=None): vol.Maybe(str),
         vol.Optional(ATTR_AUDIO_INPUT, default=None): vol.Maybe(str),
+        vol.Optional(ATTR_BACKUPS_EXCLUDE_DATABASE, default=False): vol.Boolean(),
     },
     extra=vol.REMOVE_EXTRA,
 )

supervisor/homeassistant/websocket.py

@@ -154,7 +154,7 @@ class WSClient:
     @classmethod
     async def connect_with_auth(
         cls, session: aiohttp.ClientSession, loop, url: str, token: str
-    ) -> "WSClient":
+    ) -> WSClient:
         """Create an authenticated websocket client."""
         try:
             client = await session.ws_connect(url, ssl=False)

supervisor/host/apparmor.py

@@ -1,6 +1,7 @@
 """AppArmor control for host."""
 from __future__ import annotations
 
+import errno
 import logging
 from pathlib import Path
 import shutil
@@ -9,7 +10,7 @@ from awesomeversion import AwesomeVersion
 
 from ..coresys import CoreSys, CoreSysAttributes
 from ..exceptions import DBusError, HostAppArmorError
-from ..resolution.const import UnsupportedReason
+from ..resolution.const import UnhealthyReason, UnsupportedReason
 from ..utils.apparmor import validate_profile
 from .const import HostFeature
 
@@ -80,6 +81,8 @@ class AppArmorControl(CoreSysAttributes):
         try:
             await self.sys_run_in_executor(shutil.copyfile, profile_file, dest_profile)
         except OSError as err:
+            if err.errno == errno.EBADMSG:
+                self.sys_resolution.unhealthy = UnhealthyReason.OSERROR_BAD_MESSAGE
             raise HostAppArmorError(
                 f"Can't copy {profile_file}: {err}", _LOGGER.error
             ) from err
@@ -103,6 +106,8 @@ class AppArmorControl(CoreSysAttributes):
         try:
             await self.sys_run_in_executor(profile_file.unlink)
         except OSError as err:
+            if err.errno == errno.EBADMSG:
+                self.sys_resolution.unhealthy = UnhealthyReason.OSERROR_BAD_MESSAGE
             raise HostAppArmorError(
                 f"Can't remove profile: {err}", _LOGGER.error
             ) from err
@@ -117,6 +122,8 @@ class AppArmorControl(CoreSysAttributes):
         try:
             await self.sys_run_in_executor(shutil.copy, profile_file, backup_file)
         except OSError as err:
+            if err.errno == errno.EBADMSG:
+                self.sys_resolution.unhealthy = UnhealthyReason.OSERROR_BAD_MESSAGE
             raise HostAppArmorError(
                 f"Can't backup profile {profile_name}: {err}", _LOGGER.error
             ) from err

supervisor/host/manager.py

@@ -175,4 +175,4 @@ class HostManager(CoreSysAttributes):
     async def _hardware_events(self, device: Device) -> None:
         """Process hardware requests."""
         if self.sys_hardware.policy.is_match_cgroup(PolicyGroup.AUDIO, device):
-            await self.sound.update()
+            await self.sound.update(reload_pulse=True)

supervisor/host/network.py

@@ -189,6 +189,7 @@ class NetworkManager(CoreSysAttributes):
             _LOGGER.debug("Updating existing configuration for %s", interface.name)
             settings = get_connection_from_interface(
                 interface,
+                self.sys_dbus.network,
                 name=inet.settings.connection.id,
                 uuid=inet.settings.connection.uuid,
             )
@@ -217,7 +218,7 @@ class NetworkManager(CoreSysAttributes):
         # Create new configuration and activate interface
         elif inet and interface.enabled:
             _LOGGER.debug("Create new configuration for %s", interface.name)
-            settings = get_connection_from_interface(interface)
+            settings = get_connection_from_interface(interface, self.sys_dbus.network)
 
             try:
                 settings, con = await self.sys_dbus.network.add_and_activate_connection(
@@ -244,7 +245,7 @@ class NetworkManager(CoreSysAttributes):
 
         # Create new interface (like vlan)
         elif not inet:
-            settings = get_connection_from_interface(interface)
+            settings = get_connection_from_interface(interface, self.sys_dbus.network)
 
             try:
                 await self.sys_dbus.network.settings.add_connection(settings)

supervisor/host/sound.py

@@ -15,6 +15,9 @@ _LOGGER: logging.Logger = logging.getLogger(__name__)
 
 PULSE_NAME = "supervisor"
 
+PULSE_ALSA_MODULE = "module-alsa-card"
+PULSE_UDEV_MODULE = "module-udev-detect"
+
 
 class StreamType(StrEnum):
     """INPUT/OUTPUT type of source."""
@@ -235,9 +238,9 @@ class SoundControl(CoreSysAttributes):
     @Job(
         name="sound_control_update",
         limit=JobExecutionLimit.THROTTLE_WAIT,
-        throttle_period=timedelta(seconds=10),
+        throttle_period=timedelta(seconds=2),
     )
-    async def update(self):
+    async def update(self, reload_pulse: bool = False):
         """Update properties over dbus."""
         _LOGGER.info("Updating PulseAudio information")
 
@@ -348,11 +351,32 @@ class SoundControl(CoreSysAttributes):
                     f"Error while processing pulse update: {err}", _LOGGER.error
                 ) from err
             except PulseError as err:
-                _LOGGER.debug("Can't update PulseAudio data: %s", err)
+                _LOGGER.warning("Can't update PulseAudio data: %s", err)
 
             return data
 
+        def _reload_pulse_modules():
+            try:
+                with Pulse(PULSE_NAME) as pulse:
+                    modules = pulse.module_list()
+                    for alsa_module in filter(
+                        lambda x: x.name == PULSE_ALSA_MODULE, modules
+                    ):
+                        pulse.module_unload(alsa_module.index)
+                    udev_module = next(
+                        filter(lambda x: x.name == PULSE_UDEV_MODULE, modules)
+                    )
+                    pulse.module_unload(udev_module.index)
+                    # And now reload
+                    pulse.module_load(PULSE_UDEV_MODULE)
+            except StopIteration:
+                _LOGGER.warning("Can't reload PulseAudio modules.")
+            except PulseError as err:
+                _LOGGER.warning("Can't reload PulseAudio modules: %s", err)
+
         # Update data from pulse server
+        if reload_pulse:
+            await self.sys_run_in_executor(_reload_pulse_modules)
         data: PulseData = await self.sys_run_in_executor(_get_pulse_data)
         self._applications = data.applications
         self._cards = data.cards

supervisor/ingress.py

@@ -38,7 +38,9 @@ class Ingress(FileConfiguration, CoreSysAttributes):
 
     def get_session_data(self, session_id: str) -> IngressSessionData | None:
         """Return complementary data of current session or None."""
-        return self.sessions_data.get(session_id)
+        if data := self.sessions_data.get(session_id):
+            return IngressSessionData.from_dict(data)
+        return None
 
     @property
     def sessions(self) -> dict[str, float]:
@@ -46,7 +48,7 @@ class Ingress(FileConfiguration, CoreSysAttributes):
         return self._data[ATTR_SESSION]
 
     @property
-    def sessions_data(self) -> dict[str, IngressSessionData]:
+    def sessions_data(self) -> dict[str, dict[str, str | None]]:
         """Return sessions_data."""
         return self._data[ATTR_SESSION_DATA]
 
@@ -86,7 +88,7 @@ class Ingress(FileConfiguration, CoreSysAttributes):
         now = utcnow()
 
         sessions = {}
-        sessions_data: dict[str, IngressSessionData] = {}
+        sessions_data: dict[str, dict[str, str | None]] = {}
         for session, valid in self.sessions.items():
             # check if timestamp valid, to avoid crash on malformed timestamp
             try:
@@ -100,7 +102,8 @@ class Ingress(FileConfiguration, CoreSysAttributes):
 
             # Is valid
             sessions[session] = valid
-            sessions_data[session] = self.get_session_data(session)
+            if session_data := self.sessions_data.get(session):
+                sessions_data[session] = session_data
 
         # Write back
         self.sessions.clear()
@@ -123,7 +126,7 @@ class Ingress(FileConfiguration, CoreSysAttributes):
 
         self.sessions[session] = valid.timestamp()
         if data is not None:
-            self.sessions_data[session] = data
+            self.sessions_data[session] = data.to_dict()
 
         return session
 
@@ -151,7 +154,7 @@ class Ingress(FileConfiguration, CoreSysAttributes):
 
         return True
 
-    def get_dynamic_port(self, addon_slug: str) -> int:
+    async def get_dynamic_port(self, addon_slug: str) -> int:
         """Get/Create a dynamic port from range."""
         if addon_slug in self.ports:
             return self.ports[addon_slug]
@@ -160,7 +163,7 @@ class Ingress(FileConfiguration, CoreSysAttributes):
         while (
             port is None
             or port in self.ports.values()
-            or check_port(self.sys_docker.network.gateway, port)
+            or await check_port(self.sys_docker.network.gateway, port)
         ):
             port = random.randint(62000, 65500)
 

supervisor/jobs/__init__.py

@@ -1,7 +1,7 @@
 """Supervisor job manager."""
 from collections.abc import Callable
 from contextlib import contextmanager
-from contextvars import ContextVar, Token
+from contextvars import Context, ContextVar, Token
 import logging
 from typing import Any
 from uuid import UUID, uuid4
@@ -27,6 +27,12 @@ _CURRENT_JOB: ContextVar[UUID] = ContextVar("current_job")
 _LOGGER: logging.Logger = logging.getLogger(__name__)
 
 
+def _remove_current_job(context: Context) -> Context:
+    """Remove the current job from the context."""
+    context.run(_CURRENT_JOB.set, None)
+    return context
+
+
 def _invalid_if_done(instance: "SupervisorJob", *_) -> None:
     """Validate that job is not done."""
     if instance.done:
@@ -80,7 +86,7 @@ class SupervisorJob:
         }
 
     @contextmanager
-    def start(self, *, on_done: Callable[["SupervisorJob"], None] | None = None):
+    def start(self):
         """Start the job in the current task.
 
         This can only be called if the parent ID matches the job running in the current task.
@@ -101,8 +107,6 @@ class SupervisorJob:
             self.done = True
             if token:
                 _CURRENT_JOB.reset(token)
-            if on_done:
-                on_done(self)
 
 
 class JobManager(FileConfiguration, CoreSysAttributes):
@@ -114,6 +118,9 @@ class JobManager(FileConfiguration, CoreSysAttributes):
         self.coresys: CoreSys = coresys
         self._jobs: dict[str, SupervisorJob] = {}
 
+        # Ensure tasks created via CoreSys.create_task do not have a parent
+        self.coresys.add_set_task_context_callback(_remove_current_job)
+
     @property
     def jobs(self) -> list[SupervisorJob]:
         """Return a list of current jobs."""
@@ -183,7 +190,7 @@ class JobManager(FileConfiguration, CoreSysAttributes):
         if job.uuid not in self._jobs:
             raise JobNotFound(f"Could not find job {job.name}", _LOGGER.error)
 
-        if not job.done:
+        if job.done is False:
             _LOGGER.warning("Removing incomplete job %s from job manager", job.name)
 
         del self._jobs[job.uuid]

supervisor/jobs/const.py

@@ -9,6 +9,7 @@ FILE_CONFIG_JOBS = Path(SUPERVISOR_DATA, "jobs.json")
 ATTR_IGNORE_CONDITIONS = "ignore_conditions"
 
 JOB_GROUP_ADDON = "addon_{slug}"
+JOB_GROUP_BACKUP_MANAGER = "backup_manager"
 JOB_GROUP_DOCKER_INTERFACE = "container_{name}"
 JOB_GROUP_HOME_ASSISTANT_CORE = "home_assistant_core"
 
@@ -18,6 +19,7 @@ class JobCondition(StrEnum):
 
     AUTO_UPDATE = "auto_update"
     FREE_SPACE = "free_space"
+    FROZEN = "frozen"
     HAOS = "haos"
     HEALTHY = "healthy"
     HOST_NETWORK = "host_network"

supervisor/jobs/decorator.py

@@ -174,6 +174,14 @@ class Job(CoreSysAttributes):
             return obj
         return None
 
+    def _handle_job_condition_exception(self, err: JobConditionException) -> None:
+        """Handle a job condition failure."""
+        error_msg = str(err)
+        if self.on_condition is None:
+            _LOGGER.info(error_msg)
+            return
+        raise self.on_condition(error_msg, _LOGGER.warning) from None
+
     def __call__(self, method):
         """Call the wrapper logic."""
         self._method = method
@@ -193,99 +201,125 @@ class Job(CoreSysAttributes):
                 internal=self._internal,
             )
 
-            # Handle condition
-            if self.conditions:
-                try:
-                    await self._check_conditions()
-                except JobConditionException as err:
-                    error_msg = str(err)
-                    if self.on_condition is None:
-                        _LOGGER.info(error_msg)
+            try:
+                # Handle condition
+                if self.conditions:
+                    try:
+                        await Job.check_conditions(
+                            self, set(self.conditions), self._method.__qualname__
+                        )
+                    except JobConditionException as err:
+                        return self._handle_job_condition_exception(err)
+
+                # Handle exection limits
+                if self.limit in (
+                    JobExecutionLimit.SINGLE_WAIT,
+                    JobExecutionLimit.ONCE,
+                ):
+                    await self._acquire_exection_limit()
+                elif self.limit in (
+                    JobExecutionLimit.GROUP_ONCE,
+                    JobExecutionLimit.GROUP_WAIT,
+                ):
+                    try:
+                        await obj.acquire(
+                            job, self.limit == JobExecutionLimit.GROUP_WAIT
+                        )
+                    except JobGroupExecutionLimitExceeded as err:
+                        if self.on_condition:
+                            raise self.on_condition(str(err)) from err
+                        raise err
+                elif self.limit in (
+                    JobExecutionLimit.THROTTLE,
+                    JobExecutionLimit.GROUP_THROTTLE,
+                ):
+                    time_since_last_call = datetime.now() - self.last_call(group_name)
+                    if time_since_last_call < self.throttle_period(group_name):
                         return
-                    raise self.on_condition(error_msg, _LOGGER.warning) from None
-
-            # Handle exection limits
-            if self.limit in (JobExecutionLimit.SINGLE_WAIT, JobExecutionLimit.ONCE):
-                await self._acquire_exection_limit()
-            elif self.limit in (
-                JobExecutionLimit.GROUP_ONCE,
-                JobExecutionLimit.GROUP_WAIT,
-            ):
-                try:
-                    await obj.acquire(job, self.limit == JobExecutionLimit.GROUP_WAIT)
-                except JobGroupExecutionLimitExceeded as err:
-                    if self.on_condition:
-                        raise self.on_condition(str(err)) from err
-                    raise err
-            elif self.limit in (
-                JobExecutionLimit.THROTTLE,
-                JobExecutionLimit.GROUP_THROTTLE,
-            ):
-                time_since_last_call = datetime.now() - self.last_call(group_name)
-                if time_since_last_call < self.throttle_period(group_name):
-                    return
-            elif self.limit in (
-                JobExecutionLimit.THROTTLE_WAIT,
-                JobExecutionLimit.GROUP_THROTTLE_WAIT,
-            ):
-                await self._acquire_exection_limit()
-                time_since_last_call = datetime.now() - self.last_call(group_name)
-                if time_since_last_call < self.throttle_period(group_name):
-                    self._release_exception_limits()
-                    return
-            elif self.limit in (
-                JobExecutionLimit.THROTTLE_RATE_LIMIT,
-                JobExecutionLimit.GROUP_THROTTLE_RATE_LIMIT,
-            ):
-                # Only reprocess array when necessary (at limit)
-                if len(self.rate_limited_calls(group_name)) >= self.throttle_max_calls:
-                    self.set_rate_limited_calls(
-                        [
-                            call
-                            for call in self.rate_limited_calls(group_name)
-                            if call > datetime.now() - self.throttle_period(group_name)
-                        ],
-                        group_name,
-                    )
-
-                if len(self.rate_limited_calls(group_name)) >= self.throttle_max_calls:
-                    on_condition = (
-                        JobException if self.on_condition is None else self.on_condition
-                    )
-                    raise on_condition(
-                        f"Rate limit exceeded, more then {self.throttle_max_calls} calls in {self.throttle_period(group_name)}",
-                    )
-
-            # Execute Job
-            with job.start(on_done=self.sys_jobs.remove_job if self.cleanup else None):
-                try:
-                    self.set_last_call(datetime.now(), group_name)
-                    if self.rate_limited_calls(group_name) is not None:
-                        self.add_rate_limited_call(
-                            self.last_call(group_name), group_name
+                elif self.limit in (
+                    JobExecutionLimit.THROTTLE_WAIT,
+                    JobExecutionLimit.GROUP_THROTTLE_WAIT,
+                ):
+                    await self._acquire_exection_limit()
+                    time_since_last_call = datetime.now() - self.last_call(group_name)
+                    if time_since_last_call < self.throttle_period(group_name):
+                        self._release_exception_limits()
+                        return
+                elif self.limit in (
+                    JobExecutionLimit.THROTTLE_RATE_LIMIT,
+                    JobExecutionLimit.GROUP_THROTTLE_RATE_LIMIT,
+                ):
+                    # Only reprocess array when necessary (at limit)
+                    if (
+                        len(self.rate_limited_calls(group_name))
+                        >= self.throttle_max_calls
+                    ):
+                        self.set_rate_limited_calls(
+                            [
+                                call
+                                for call in self.rate_limited_calls(group_name)
+                                if call
+                                > datetime.now() - self.throttle_period(group_name)
+                            ],
+                            group_name,
                         )
 
-                    return await self._method(obj, *args, **kwargs)
-                except HassioError as err:
-                    raise err
-                except Exception as err:
-                    _LOGGER.exception("Unhandled exception: %s", err)
-                    capture_exception(err)
-                    raise JobException() from err
-                finally:
-                    self._release_exception_limits()
-                    if self.limit in (
-                        JobExecutionLimit.GROUP_ONCE,
-                        JobExecutionLimit.GROUP_WAIT,
+                    if (
+                        len(self.rate_limited_calls(group_name))
+                        >= self.throttle_max_calls
                     ):
-                        obj.release()
+                        on_condition = (
+                            JobException
+                            if self.on_condition is None
+                            else self.on_condition
+                        )
+                        raise on_condition(
+                            f"Rate limit exceeded, more than {self.throttle_max_calls} calls in {self.throttle_period(group_name)}",
+                        )
+
+                # Execute Job
+                with job.start():
+                    try:
+                        self.set_last_call(datetime.now(), group_name)
+                        if self.rate_limited_calls(group_name) is not None:
+                            self.add_rate_limited_call(
+                                self.last_call(group_name), group_name
+                            )
+
+                        return await self._method(obj, *args, **kwargs)
+
+                    # If a method has a conditional JobCondition, they must check it in the method
+                    # These should be handled like normal JobConditions as much as possible
+                    except JobConditionException as err:
+                        return self._handle_job_condition_exception(err)
+                    except HassioError as err:
+                        raise err
+                    except Exception as err:
+                        _LOGGER.exception("Unhandled exception: %s", err)
+                        capture_exception(err)
+                        raise JobException() from err
+                    finally:
+                        self._release_exception_limits()
+                        if self.limit in (
+                            JobExecutionLimit.GROUP_ONCE,
+                            JobExecutionLimit.GROUP_WAIT,
+                        ):
+                            obj.release()
+
+            # Jobs that weren't started are always cleaned up. Also clean up done jobs if required
+            finally:
+                if job.done is None or self.cleanup:
+                    self.sys_jobs.remove_job(job)
 
         return wrapper
 
-    async def _check_conditions(self):
+    @staticmethod
+    async def check_conditions(
+        coresys: CoreSysAttributes, conditions: set[JobCondition], method_name: str
+    ):
         """Check conditions."""
-        used_conditions = set(self.conditions) - set(self.sys_jobs.ignore_conditions)
-        ignored_conditions = set(self.conditions) & set(self.sys_jobs.ignore_conditions)
+        used_conditions = set(conditions) - set(coresys.sys_jobs.ignore_conditions)
+        ignored_conditions = set(conditions) & set(coresys.sys_jobs.ignore_conditions)
 
         # Check if somethings is ignored
         if ignored_conditions:
@@ -294,85 +328,97 @@ class Job(CoreSysAttributes):
                 ignored_conditions,
             )
 
-        if JobCondition.HEALTHY in used_conditions and not self.sys_core.healthy:
+        if JobCondition.HEALTHY in used_conditions and not coresys.sys_core.healthy:
             raise JobConditionException(
-                f"'{self._method.__qualname__}' blocked from execution, system is not healthy - {', '.join(self.sys_resolution.unhealthy)}"
+                f"'{method_name}' blocked from execution, system is not healthy - {', '.join(coresys.sys_resolution.unhealthy)}"
             )
 
         if (
             JobCondition.RUNNING in used_conditions
-            and self.sys_core.state != CoreState.RUNNING
+            and coresys.sys_core.state != CoreState.RUNNING
         ):
             raise JobConditionException(
-                f"'{self._method.__qualname__}' blocked from execution, system is not running - {self.sys_core.state!s}"
+                f"'{method_name}' blocked from execution, system is not running - {coresys.sys_core.state!s}"
+            )
+
+        if (
+            JobCondition.FROZEN in used_conditions
+            and coresys.sys_core.state != CoreState.FREEZE
+        ):
+            raise JobConditionException(
+                f"'{method_name}' blocked from execution, system is not frozen - {coresys.sys_core.state!s}"
             )
 
         if (
             JobCondition.FREE_SPACE in used_conditions
-            and self.sys_host.info.free_space < MINIMUM_FREE_SPACE_THRESHOLD
+            and coresys.sys_host.info.free_space < MINIMUM_FREE_SPACE_THRESHOLD
         ):
-            self.sys_resolution.create_issue(IssueType.FREE_SPACE, ContextType.SYSTEM)
+            coresys.sys_resolution.create_issue(
+                IssueType.FREE_SPACE, ContextType.SYSTEM
+            )
             raise JobConditionException(
-                f"'{self._method.__qualname__}' blocked from execution, not enough free space ({self.sys_host.info.free_space}GB) left on the device"
+                f"'{method_name}' blocked from execution, not enough free space ({coresys.sys_host.info.free_space}GB) left on the device"
             )
 
         if JobCondition.INTERNET_SYSTEM in used_conditions:
-            await self.sys_supervisor.check_connectivity()
-            if not self.sys_supervisor.connectivity:
+            await coresys.sys_supervisor.check_connectivity()
+            if not coresys.sys_supervisor.connectivity:
                 raise JobConditionException(
-                    f"'{self._method.__qualname__}' blocked from execution, no supervisor internet connection"
+                    f"'{method_name}' blocked from execution, no supervisor internet connection"
                 )
 
         if JobCondition.INTERNET_HOST in used_conditions:
-            await self.sys_host.network.check_connectivity()
+            await coresys.sys_host.network.check_connectivity()
             if (
-                self.sys_host.network.connectivity is not None
-                and not self.sys_host.network.connectivity
+                coresys.sys_host.network.connectivity is not None
+                and not coresys.sys_host.network.connectivity
             ):
                 raise JobConditionException(
-                    f"'{self._method.__qualname__}' blocked from execution, no host internet connection"
+                    f"'{method_name}' blocked from execution, no host internet connection"
                 )
 
-        if JobCondition.HAOS in used_conditions and not self.sys_os.available:
+        if JobCondition.HAOS in used_conditions and not coresys.sys_os.available:
             raise JobConditionException(
-                f"'{self._method.__qualname__}' blocked from execution, no Home Assistant OS available"
+                f"'{method_name}' blocked from execution, no Home Assistant OS available"
             )
 
         if (
             JobCondition.OS_AGENT in used_conditions
-            and HostFeature.OS_AGENT not in self.sys_host.features
+            and HostFeature.OS_AGENT not in coresys.sys_host.features
         ):
             raise JobConditionException(
-                f"'{self._method.__qualname__}' blocked from execution, no Home Assistant OS-Agent available"
+                f"'{method_name}' blocked from execution, no Home Assistant OS-Agent available"
             )
 
         if (
             JobCondition.HOST_NETWORK in used_conditions
-            and not self.sys_dbus.network.is_connected
+            and not coresys.sys_dbus.network.is_connected
         ):
             raise JobConditionException(
-                f"'{self._method.__qualname__}' blocked from execution, host Network Manager not available"
+                f"'{method_name}' blocked from execution, host Network Manager not available"
             )
 
         if (
             JobCondition.AUTO_UPDATE in used_conditions
-            and not self.sys_updater.auto_update
+            and not coresys.sys_updater.auto_update
         ):
             raise JobConditionException(
-                f"'{self._method.__qualname__}' blocked from execution, supervisor auto updates disabled"
+                f"'{method_name}' blocked from execution, supervisor auto updates disabled"
             )
 
         if (
             JobCondition.SUPERVISOR_UPDATED in used_conditions
-            and self.sys_supervisor.need_update
+            and coresys.sys_supervisor.need_update
         ):
             raise JobConditionException(
-                f"'{self._method.__qualname__}' blocked from execution, supervisor needs to be updated first"
+                f"'{method_name}' blocked from execution, supervisor needs to be updated first"
             )
 
         if JobCondition.PLUGINS_UPDATED in used_conditions and (
             out_of_date := [
-                plugin for plugin in self.sys_plugins.all_plugins if plugin.need_update
+                plugin
+                for plugin in coresys.sys_plugins.all_plugins
+                if plugin.need_update
             ]
         ):
             errors = await asyncio.gather(
@@ -383,15 +429,15 @@ class Job(CoreSysAttributes):
                 out_of_date[i].slug for i in range(len(errors)) if errors[i] is not None
             ]:
                 raise JobConditionException(
-                    f"'{self._method.__qualname__}' blocked from execution, was unable to update plugin(s) {', '.join(update_failures)} and all plugins must be up to date first"
+                    f"'{method_name}' blocked from execution, was unable to update plugin(s) {', '.join(update_failures)} and all plugins must be up to date first"
                 )
 
         if (
             JobCondition.MOUNT_AVAILABLE in used_conditions
-            and HostFeature.MOUNT not in self.sys_host.features
+            and HostFeature.MOUNT not in coresys.sys_host.features
         ):
             raise JobConditionException(
-                f"'{self._method.__qualname__}' blocked from execution, mounting not supported on system"
+                f"'{method_name}' blocked from execution, mounting not supported on system"
             )
 
     async def _acquire_exection_limit(self) -> None:

supervisor/misc/scheduler.py

@@ -115,5 +115,5 @@ class Scheduler(CoreSysAttributes):
         try:
             async with async_timeout.timeout(timeout):
                 await asyncio.wait(running)
-        except asyncio.TimeoutError:
+        except TimeoutError:
             _LOGGER.error("Timeout while waiting for jobs shutdown")

supervisor/misc/tasks.py

@@ -7,6 +7,7 @@ from ..addons.const import ADDON_UPDATE_CONDITIONS
 from ..const import AddonState
 from ..coresys import CoreSysAttributes
 from ..exceptions import AddonsError, HomeAssistantError, ObserverError
+from ..homeassistant.const import LANDINGPAGE
 from ..jobs.decorator import Job, JobCondition
 from ..plugins.const import PLUGIN_UPDATE_CONDITIONS
 from ..utils.sentry import capture_exception
@@ -14,6 +15,8 @@ from ..utils.sentry import capture_exception
 _LOGGER: logging.Logger = logging.getLogger(__name__)
 
 HASS_WATCHDOG_API = "HASS_WATCHDOG_API"
+HASS_WATCHDOG_REANIMATE_FAILURES = "HASS_WATCHDOG_REANIMATE_FAILURES"
+HASS_WATCHDOG_MAX_REANIMATE_ATTEMPTS = 5
 
 RUN_UPDATE_SUPERVISOR = 29100
 RUN_UPDATE_ADDONS = 57600
@@ -102,7 +105,7 @@ class Tasks(CoreSysAttributes):
             # avoid issue on slow IO
             _LOGGER.info("Add-on auto update process %s", addon.slug)
             try:
-                if start_task := await addon.update(backup=True):
+                if start_task := await self.sys_addons.update(addon.slug, backup=True):
                     start_tasks.append(start_task)
             except AddonsError:
                 _LOGGER.error("Can't auto update Add-on %s", addon.slug)
@@ -142,6 +145,9 @@ class Tasks(CoreSysAttributes):
         if self.sys_homeassistant.error_state:
             # Home Assistant is in an error state, this is handled by the rollback feature
             return
+        if self.sys_homeassistant.version == LANDINGPAGE:
+            # Skip watchdog for landingpage
+            return
         if not await self.sys_homeassistant.core.is_running():
             # The home assistant container is not running
             return
@@ -150,6 +156,18 @@ class Tasks(CoreSysAttributes):
             return
         if await self.sys_homeassistant.api.check_api_state():
             # Home Assistant is running properly
+            self._cache[HASS_WATCHDOG_REANIMATE_FAILURES] = 0
+            return
+
+        # Give up after 5 reanimation failures in a row. Supervisor cannot fix this issue.
+        reanimate_fails = self._cache.get(HASS_WATCHDOG_REANIMATE_FAILURES, 0)
+        if reanimate_fails >= HASS_WATCHDOG_MAX_REANIMATE_ATTEMPTS:
+            if reanimate_fails == HASS_WATCHDOG_MAX_REANIMATE_ATTEMPTS:
+                _LOGGER.critical(
+                    "Watchdog cannot reanimate Home Assistant, failed all %s attempts.",
+                    reanimate_fails,
+                )
+                self._cache[HASS_WATCHDOG_REANIMATE_FAILURES] += 1
             return
 
         # Init cache data
@@ -167,7 +185,11 @@ class Tasks(CoreSysAttributes):
             await self.sys_homeassistant.core.restart()
         except HomeAssistantError as err:
             _LOGGER.error("Home Assistant watchdog reanimation failed!")
-            capture_exception(err)
+            if reanimate_fails == 0:
+                capture_exception(err)
+            self._cache[HASS_WATCHDOG_REANIMATE_FAILURES] = reanimate_fails + 1
+        else:
+            self._cache[HASS_WATCHDOG_REANIMATE_FAILURES] = 0
         finally:
             self._cache[HASS_WATCHDOG_API] = 0
 

supervisor/mounts/mount.py

@@ -170,6 +170,16 @@ class Mount(CoreSysAttributes, ABC):
         elif self.state != UnitActiveState.ACTIVE:
             await self.reload()
 
+    async def update_state(self) -> None:
+        """Update mount unit state."""
+        try:
+            self._state = await self.unit.get_active_state()
+        except DBusError as err:
+            capture_exception(err)
+            raise MountError(
+                f"Could not get active state of mount due to: {err!s}"
+            ) from err
+
     async def update(self) -> None:
         """Update info about mount from dbus."""
         try:
@@ -182,13 +192,7 @@ class Mount(CoreSysAttributes, ABC):
             capture_exception(err)
             raise MountError(f"Could not get mount unit due to: {err!s}") from err
 
-        try:
-            self._state = await self.unit.get_active_state()
-        except DBusError as err:
-            capture_exception(err)
-            raise MountError(
-                f"Could not get active state of mount due to: {err!s}"
-            ) from err
+        await self.update_state()
 
         # If active, dismiss corresponding failed mount issue if found
         if (
@@ -197,6 +201,20 @@ class Mount(CoreSysAttributes, ABC):
         ):
             self.sys_resolution.dismiss_issue(self.failed_issue)
 
+    async def _update_state_await(self, expected_states: list[UnitActiveState]) -> None:
+        """Update state info about mount from dbus. Wait up to 30 seconds for the state to appear."""
+        for i in range(5):
+            await self.update_state()
+            if self.state in expected_states:
+                return
+            await asyncio.sleep(i**2)
+
+        _LOGGER.warning(
+            "Mount %s still in state %s after waiting for 30 seconods to complete",
+            self.name,
+            str(self.state).lower(),
+        )
+
     async def _update_await_activating(self):
         """Update info about mount from dbus. If 'activating' wait up to 30 seconds."""
         await self.update()
@@ -269,10 +287,15 @@ class Mount(CoreSysAttributes, ABC):
         await self.update()
 
         try:
+            if self.state != UnitActiveState.FAILED:
+                await self.sys_dbus.systemd.stop_unit(self.unit_name, StopUnitMode.FAIL)
+
+            await self._update_state_await(
+                [UnitActiveState.INACTIVE, UnitActiveState.FAILED]
+            )
+
             if self.state == UnitActiveState.FAILED:
                 await self.sys_dbus.systemd.reset_failed_unit(self.unit_name)
-            else:
-                await self.sys_dbus.systemd.stop_unit(self.unit_name, StopUnitMode.FAIL)
         except DBusSystemdNoSuchUnit:
             _LOGGER.info("Mount %s is not mounted, skipping unmount", self.name)
         except DBusError as err:

supervisor/mounts/validate.py

@@ -1,9 +1,8 @@
 """Validation for mount manager."""
 
 import re
-from typing import TypedDict
+from typing import NotRequired, TypedDict
 
-from typing_extensions import NotRequired
 import voluptuous as vol
 
 from ..const import (
@@ -27,7 +26,7 @@ from .const import (
     MountUsage,
 )
 
-RE_MOUNT_NAME = re.compile(r"^\w+$")
+RE_MOUNT_NAME = re.compile(r"^[A-Za-z0-9_]+$")
 RE_PATH_PART = re.compile(r"^[^\\\/]+")
 RE_MOUNT_OPTION = re.compile(r"^[^,=]+$")
 

supervisor/os/manager.py

@@ -1,6 +1,6 @@
 """OS support on supervisor."""
-import asyncio
 from collections.abc import Awaitable
+import errno
 import logging
 from pathlib import Path
 
@@ -14,6 +14,7 @@ from ..dbus.rauc import RaucState
 from ..exceptions import DBusError, HassOSJobError, HassOSUpdateError
 from ..jobs.const import JobCondition, JobExecutionLimit
 from ..jobs.decorator import Job
+from ..resolution.const import UnhealthyReason
 from .data_disk import DataDisk
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
@@ -114,13 +115,15 @@ class OSManager(CoreSysAttributes):
 
             _LOGGER.info("Completed download of OTA update file %s", raucb)
 
-        except (aiohttp.ClientError, asyncio.TimeoutError) as err:
+        except (aiohttp.ClientError, TimeoutError) as err:
             self.sys_supervisor.connectivity = False
             raise HassOSUpdateError(
                 f"Can't fetch OTA update from {url}: {err!s}", _LOGGER.error
             ) from err
 
         except OSError as err:
+            if err.errno == errno.EBADMSG:
+                self.sys_resolution.unhealthy = UnhealthyReason.OSERROR_BAD_MESSAGE
             raise HassOSUpdateError(
                 f"Can't write OTA file: {err!s}", _LOGGER.error
             ) from err

supervisor/plugins/audio.py

@@ -4,6 +4,7 @@ Code: https://github.com/home-assistant/plugin-audio
 """
 import asyncio
 from contextlib import suppress
+import errno
 import logging
 from pathlib import Path, PurePath
 import shutil
@@ -25,6 +26,7 @@ from ..exceptions import (
 )
 from ..jobs.const import JobExecutionLimit
 from ..jobs.decorator import Job
+from ..resolution.const import UnhealthyReason
 from ..utils.json import write_json_file
 from ..utils.sentry import capture_exception
 from .base import PluginBase
@@ -83,6 +85,9 @@ class PluginAudio(PluginBase):
                 PULSE_CLIENT_TMPL.read_text(encoding="utf-8")
             )
         except OSError as err:
+            if err.errno == errno.EBADMSG:
+                self.sys_resolution.unhealthy = UnhealthyReason.OSERROR_BAD_MESSAGE
+
             _LOGGER.error("Can't read pulse-client.tmpl: %s", err)
 
         await super().load()
@@ -93,6 +98,8 @@ class PluginAudio(PluginBase):
             try:
                 shutil.copy(ASOUND_TMPL, asound)
             except OSError as err:
+                if err.errno == errno.EBADMSG:
+                    self.sys_resolution.unhealthy = UnhealthyReason.OSERROR_BAD_MESSAGE
                 _LOGGER.error("Can't create default asound: %s", err)
 
     async def install(self) -> None:

supervisor/plugins/base.py

@@ -105,11 +105,7 @@ class PluginBase(ABC, FileConfiguration, CoreSysAttributes):
         if not (event.name == self.instance.name):
             return
 
-        if event.state in [
-            ContainerState.FAILED,
-            ContainerState.STOPPED,
-            ContainerState.UNHEALTHY,
-        ]:
+        if event.state in {ContainerState.FAILED, ContainerState.UNHEALTHY}:
             await self._restart_after_problem(event.state)
 
     async def _restart_after_problem(self, state: ContainerState):
@@ -123,10 +119,7 @@ class PluginBase(ABC, FileConfiguration, CoreSysAttributes):
                     state,
                 )
                 try:
-                    if state == ContainerState.STOPPED and attempts == 0:
-                        await self.start()
-                    else:
-                        await self.rebuild()
+                    await self.rebuild()
                 except PluginError as err:
                     attempts = attempts + 1
                     _LOGGER.error("Watchdog restart of %s plugin failed!", self.slug)