Version 2.0.0

Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>

.cargo/config.toml

@@ -0,0 +1,5 @@
+[alias]
+xtask = "run --package xtask --"
+
+[env]
+CARGO_WORKSPACE_DIR = { value = "", relative = true }

.github/actions/preload-img-cache/action.yml

@@ -15,39 +15,34 @@ runs:
       with:
         key: ${{ inputs.cache-key-prefix }}${{ inputs.device }}
         # Make sure any changes to path are also reflected in ci.yml setup
-        path: tests/files/${{ inputs.device }}-sparse.tar
+        path: e2e/files/${{ inputs.device }}-sparse.tar
 
     - if: ${{ steps.cache-img.outputs.cache-hit }}
       name: Extracting image from sparse archive
       shell: sh
-      run: |
-        tar -C tests/files -xf tests/files/${{ inputs.device }}-sparse.tar
+      working-directory: e2e/files
+      run: tar -xf ${{ inputs.device }}-sparse.tar
 
-    - if: ${{ ! steps.cache-img.outputs.cache-hit }}
-      uses: awalsh128/cache-apt-pkgs-action@v1
+    - name: Restore e2e executable
+      if: ${{ ! steps.cache-img.outputs.cache-hit }}
+      uses: actions/cache/restore@v3
       with:
-        packages: python3-lz4 python3-protobuf
-
-    - if: ${{ ! steps.cache-img.outputs.cache-hit }}
-      uses: awalsh128/cache-apt-pkgs-action@v1
-      with:
-        packages: python3-strictyaml
+        key: e2e-${{ github.sha }}-${{ runner.os }}
+        fail-on-cache-miss: true
+        path: |
+          target/release/e2e
+          target/release/e2e.exe
 
     - name: Downloading device image for ${{ inputs.device }}
       if: ${{ ! steps.cache-img.outputs.cache-hit }}
       shell: sh
-      run: |
-        ./tests/tests.py \
-          download \
-          --stripped \
-          --no-magisk \
-          --device \
-          ${{ inputs.device }}
+      working-directory: e2e
+      run: ../target/release/e2e download --stripped -d ${{ inputs.device }}
 
     - if: ${{ ! steps.cache-img.outputs.cache-hit }}
       name: Creating sparse archive from image
       shell: sh
+      working-directory: e2e/files
       run: |
-        cd tests/files
         tar --sparse -cf ${{ inputs.device }}-sparse.tar \
           ${{ inputs.device }}/*.stripped

.github/actions/preload-magisk-cache/action.yml

@@ -12,23 +12,20 @@ runs:
       with:
         key: ${{ inputs.cache-key }}
         # Make sure any changes to path are also reflected in ci.yml setup
-        path: tests/files/magisk
+        path: e2e/files/magisk
 
-    - if: ${{ ! steps.cache-magisk.outputs.cache-hit }}
-      uses: awalsh128/cache-apt-pkgs-action@v1
+    - name: Restore e2e executable
+      if: ${{ ! steps.cache-magisk.outputs.cache-hit }}
+      uses: actions/cache/restore@v3
       with:
-        packages: python3-lz4 python3-protobuf
-
-    - if: ${{ ! steps.cache-magisk.outputs.cache-hit }}
-      uses: awalsh128/cache-apt-pkgs-action@v1
-      with:
-        packages: python3-strictyaml
+        key: e2e-${{ github.sha }}-${{ runner.os }}
+        fail-on-cache-miss: true
+        path: |
+          target/release/e2e
+          target/release/e2e.exe
 
     - name: Downloading Magisk
       if: ${{ ! steps.cache-magisk.outputs.cache-hit }}
       shell: sh
-      run: |
-        ./tests/tests.py \
-          download \
-          --magisk \
-          --no-devices
+      working-directory: e2e
+      run: ../target/release/e2e download --magisk

.github/actions/preload-tox-cache/action.yml

@@ -1,30 +0,0 @@
-name: Preload tox cache
-inputs:
-  cache-key-prefix:
-    description: 'Tox cache-key prefix'
-    required: true
-  python-version:
-    description: 'Python version'
-    required: true
-
-runs:
-  using: "composite"
-  steps:
-    - uses: actions/cache@v3
-      with:
-        key: ${{ inputs.cache-key-prefix }}${{ inputs.python-version }}
-        restore-keys: |
-          tox-
-        # Make sure any changes to path are also reflected in ci.yml setup
-        path: |
-          .tox/
-          ~/.cache/pip
-
-    - uses: awalsh128/cache-apt-pkgs-action@v1
-      with:
-        packages: tox
-
-    - uses: actions/setup-python@v4
-      with:
-        python-version: |
-          ${{ fromJson('{ "py39": "3.9", "py310": "3.10", "py311": "3.11" }')[inputs.python-version] }}

.github/workflows/ci.yml

@@ -11,9 +11,86 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
+  build:
+    runs-on: ${{ matrix.os }}
+    env:
+      CARGO_TERM_COLOR: always
+      RUSTFLAGS: -C strip=symbols
+    strategy:
+      fail-fast: false
+      matrix:
+        os:
+          - ubuntu-latest
+          - windows-latest
+          - macos-latest
+    steps:
+      - name: Check out repository
+        uses: actions/checkout@v3
+        with:
+          # For git describe
+          fetch-depth: 0
+
+      - name: Get version
+        id: get_version
+        shell: bash
+        run: |
+          echo -n 'version=' >> "${GITHUB_OUTPUT}"
+          git describe --always \
+              | sed -E "s/^v//g;s/([^-]*-g)/r\1/;s/-/./g" \
+              >> "${GITHUB_OUTPUT}"
+
+      - name: Get Rust LLVM target triple
+        id: get_target
+        shell: bash
+        env:
+          RUSTC_BOOTSTRAP: '1'
+        run: |
+          echo -n 'name=' >> "${GITHUB_OUTPUT}"
+          rustc -Z unstable-options --print target-spec-json \
+              | jq -r '."llvm-target"' \
+              >> "${GITHUB_OUTPUT}"
+
+      - name: Cache Rust dependencies
+        uses: Swatinem/rust-cache@v2
+
+      - name: Clippy
+        run: cargo clippy --release --workspace --features static
+
+      - name: Build
+        run: cargo build --release --workspace --features static
+
+      - name: Tests
+        run: cargo test --release --workspace --features static
+
+      - name: Archive documentation
+        uses: actions/upload-artifact@v3
+        with:
+          name: avbroot-${{ steps.get_version.outputs.version }}-${{ steps.get_target.outputs.name }}
+          path: |
+            LICENSE
+            README.md
+
+      # This is separate so we can have a flat directory structure.
+      - name: Archive executable
+        uses: actions/upload-artifact@v3
+        with:
+          name: avbroot-${{ steps.get_version.outputs.version }}-${{ steps.get_target.outputs.name }}
+          path: |
+            target/release/avbroot
+            target/release/avbroot.exe
+
+      - name: Cache e2e executable
+        uses: actions/cache@v3
+        with:
+          key: e2e-${{ github.sha }}-${{ runner.os }}
+          path: |
+            target/release/e2e
+            target/release/e2e.exe
+
   setup:
     name: Prepare workflow data
     runs-on: ubuntu-latest
+    needs: build
     timeout-minutes: 2
     outputs:
       config-path: ${{ steps.load-config.outputs.config-path }}
@@ -21,54 +98,36 @@ jobs:
       magisk-key: ${{ steps.cache-keys.outputs.magisk-key }}
       img-key-prefix: ${{ steps.cache-keys.outputs.img-key-prefix }}
       img-hit: ${{ steps.get-img-cache.outputs.cache-matched-key }}
-      tox-key-prefix: ${{ steps.cache-keys.outputs.tox-key-prefix }}
-      tox-hit: ${{ steps.get-tox-cache.outputs.cache-matched-key }}
     steps:
       - uses: actions/checkout@v3
-        with:
-          submodules: true
 
-      - uses: awalsh128/cache-apt-pkgs-action@v1
+      - name: Restore e2e executable
+        uses: actions/cache/restore@v3
         with:
-          packages: python3-strictyaml
+          key: e2e-${{ github.sha }}-${{ runner.os }}
+          fail-on-cache-miss: true
+          path: |
+            target/release/e2e
+            target/release/e2e.exe
 
       - name: Loading test config
         id: load-config
-        shell: python
+        working-directory: e2e
         run: |
-          import json
-          import os
-          import sys
-
-          sys.path.append(os.environ['GITHUB_WORKSPACE'])
-          import tests.config
-
-          config_data = tests.config.load_config()
-          devices = [d.data for d in config_data['device']]
-
-          with open(os.environ['GITHUB_OUTPUT'], 'a') as f:
-              f.write(f'config-path={tests.config.CONFIG_PATH}\n')
-              f.write(f"device-list={json.dumps(devices)}\n")
+          echo 'config-path=e2e/e2e.toml' >> "${GITHUB_OUTPUT}"
+          echo -n 'device-list=' >> "${GITHUB_OUTPUT}"
+          ../target/release/e2e list \
+            | jq -cnR '[inputs | select(length > 0)]' \
+            >> "${GITHUB_OUTPUT}"
 
       - name: Generating cache keys
         id: cache-keys
         run: |
           {
-            echo "tox-key-prefix=tox-${{ hashFiles('tox.ini') }}-"; \
             echo "img-key-prefix=img-${{ hashFiles(steps.load-config.outputs.config-path) }}-"; \
             echo "magisk-key=magisk-${{ hashFiles(steps.load-config.outputs.config-path) }}";
           } >> $GITHUB_OUTPUT
 
-      - name: Checking for cached tox environments
-        id: get-tox-cache
-        uses: actions/cache/restore@v3
-        with:
-          key: ${{ steps.cache-keys.outputs.tox-key-prefix }}
-          lookup-only: true
-          path: |
-            .tox/
-            ~/.cache/pip
-
       - name: Checking for cached device images
         id: get-img-cache
         uses: actions/cache/restore@v3
@@ -76,7 +135,7 @@ jobs:
           key: ${{ steps.cache-keys.outputs.img-key-prefix }}
           lookup-only: true
           path: |
-            tests/files/${{ fromJSON(steps.load-config.outputs.device-list)[0] }}-sparse.tar
+            e2e/files/${{ fromJSON(steps.load-config.outputs.device-list)[0] }}-sparse.tar
 
       - name: Checking for cached magisk apk
         id: get-magisk-cache
@@ -84,7 +143,7 @@ jobs:
         with:
           key: ${{ steps.cache-keys.outputs.magisk-key }}
           lookup-only: true
-          path: tests/files/magisk
+          path: e2e/files/magisk
 
       - name: Preloading Magisk cache
         if: ${{ ! steps.get-magisk-cache.outputs.cache-hit }}
@@ -106,8 +165,6 @@ jobs:
         device: ${{ fromJSON(needs.setup.outputs.device-list) }}
     steps:
       - uses: actions/checkout@v3
-        with:
-          submodules: true
 
       - name: Preloading image cache
         uses: ./.github/actions/preload-img-cache
@@ -115,45 +172,24 @@ jobs:
           cache-key-prefix: ${{ needs.setup.outputs.img-key-prefix }}
           device: ${{ matrix.device }}
 
-  preload-tox:
-    name: Preload tox environments
-    runs-on: ubuntu-latest
-    needs: setup
-    timeout-minutes: 5
-    # Assume that preloading always succesfully cached all tox environments before.
-    # If for some reason only some got cached, on the first run, the cache will not be preloaded
-    # which will result in some being downloaded multiple times when running the tests.
-    if: ${{ ! needs.setup.outputs.tox-hit }}
-    strategy:
-      matrix:
-        python: [py39, py310, py311]
-    steps:
-      - uses: actions/checkout@v3
-
-      - name: Preloading tox cache
-        uses: ./.github/actions/preload-tox-cache
-        with:
-          cache-key-prefix: ${{ needs.setup.outputs.tox-key-prefix }}
-          python-version: ${{ matrix.python }}
-
-      - name: Generating tox environment
-        run: tox -e ${{ matrix.python }} --notest
-
   tests:
-    name: Run test for ${{ matrix.device }} with ${{ matrix.python }}
+    name: Run test for ${{ matrix.device }} on ${{ matrix.os }}
     runs-on: ubuntu-latest
-    needs: [setup, preload-img, preload-tox]
+    needs:
+      - setup
+      - preload-img
     timeout-minutes: 10
     # Continue on skipped but not on failures or cancels
     if: ${{ always() && ! failure() && ! cancelled() }}
     strategy:
       matrix:
         device: ${{ fromJSON(needs.setup.outputs.device-list) }}
-        python: [py39, py310, py311]
+        os:
+          - ubuntu-latest
+          - windows-latest
+          - macos-latest
     steps:
       - uses: actions/checkout@v3
-        with:
-          submodules: true
 
       - name: Restoring Magisk cache
         uses: ./.github/actions/preload-magisk-cache
@@ -166,12 +202,16 @@ jobs:
           cache-key-prefix: ${{ needs.setup.outputs.img-key-prefix }}
           device: ${{ matrix.device }}
 
-      - name: Restoring tox cache
-        uses: ./.github/actions/preload-tox-cache
+      - name: Restore e2e executable
+        uses: actions/cache/restore@v3
         with:
-          cache-key-prefix: ${{ needs.setup.outputs.tox-key-prefix }}
-          python-version: ${{ matrix.python }}
+          key: e2e-${{ github.sha }}-${{ runner.os }}
+          fail-on-cache-miss: true
+          path: |
+            target/release/e2e
+            target/release/e2e.exe
 
       # Finally run tests
-      - name: Run test for ${{ matrix.device }} with ${{ matrix.python }}
-        run: tox -e ${{ matrix.python }} -- --stripped -d ${{ matrix.device }}
+      - name: Run test for ${{ matrix.device }}
+        working-directory: e2e
+        run: ../target/release/e2e test --stripped -d ${{ matrix.device }}

.github/workflows/deny.yml

@@ -0,0 +1,16 @@
+---
+on:
+  push:
+    branches:
+      - master
+  pull_request:
+jobs:
+  check:
+    name: cargo-deny
+    runs-on: ubuntu-latest
+    steps:
+      - name: Check out repository
+        uses: actions/checkout@v3
+
+      - name: Run cargo-deny
+        uses: EmbarkStudios/cargo-deny-action@v1

.github/workflows/modules.yml

@@ -5,22 +5,27 @@ on:
       - master
   pull_request:
 jobs:
-  build-app:
+  build:
     name: Build modules
     runs-on: ubuntu-latest
     steps:
       - name: Check out repository
         uses: actions/checkout@v3
         with:
+          # For git describe
           fetch-depth: 0
 
       - name: Get version
         id: get_version
         shell: bash
-        run: echo "version=r$(git rev-list --count HEAD).$(git rev-parse --short HEAD)" >> "${GITHUB_OUTPUT}"
+        run: |
+          echo -n 'version=' >> "${GITHUB_OUTPUT}"
+          git describe --always \
+              | sed -E "s/^v//g;s/([^-]*-g)/r\1/;s/-/./g" \
+              >> "${GITHUB_OUTPUT}"
 
-      - name: Build and test
-        run: ./modules/build.py
+      - name: Build modules
+        run: cargo xtask modules -a
 
       - name: Archive artifacts
         uses: actions/upload-artifact@v3

.github/workflows/release.yml

@@ -0,0 +1,37 @@
+---
+on:
+  push:
+    # Uncomment to test against a branch
+    #branches:
+    #  - ci
+    tags:
+      - 'v*'
+jobs:
+  create_release:
+    name: Create Github release
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write
+    steps:
+      - name: Get version from tag
+        id: get_version
+        run: |
+          if [[ "${GITHUB_REF}" == refs/tags/* ]]; then
+              version=${GITHUB_REF#refs/tags/v}
+          else
+              version=0.0.0.${GITHUB_REF#refs/heads/}
+          fi
+          echo "version=${version}" >> "${GITHUB_OUTPUT}"
+
+      - name: Check out repository
+        uses: actions/checkout@v3
+
+      - name: Create release
+        uses: softprops/action-gh-release@v1
+        with:
+          token: ${{ secrets.GITHUB_TOKEN }}
+          tag_name: v${{ steps.get_version.outputs.version }}
+          name: Version ${{ steps.get_version.outputs.version }}
+          body_path: RELEASE.md
+          draft: true
+          prerelease: false

.gitignore

@@ -1,5 +1,5 @@
-# Caches
-__pycache__/
+# Build directories
+/target/
 
 # Secrets
 *.pem
@@ -11,4 +11,6 @@ __pycache__/
 *.img
 *.zip
 *.patched
-.tox
+
+# We do want the test images
+!/avbroot/tests/data/*.img

.gitmodules

@@ -1,9 +0,0 @@
-[submodule "external/avb"]
-	path = external/avb
-	url = https://android.googlesource.com/platform/external/avb/
-[submodule "external/update_engine"]
-	path = external/update_engine
-	url = https://android.googlesource.com/platform/system/update_engine
-[submodule "external/build"]
-	path = external/build
-	url = https://android.googlesource.com/platform/build

CHANGELOG.md

@@ -0,0 +1,15 @@
+<!--
+    When adding new changelog entries, use [Issue #0] to link to issues and
+    [PR #0 @user] to link to pull requests. Then run:
+
+        cargo xtask update-changelog
+
+    to update the actual links at the bottom of the file.
+-->
+
+### Version 2.0.0
+
+* Initial Rust release. The old Python implementation can be found in the `python` branch. ([PR #130 @chenxiaolong])
+
+<!-- Do not manually edit the lines below. Use `cargo xtask update-changelog` to regenerate. -->
+[PR #130 @chenxiaolong]: https://github.com/chenxiaolong/avbroot/pull/130

Cargo.toml

@@ -0,0 +1,10 @@
+[workspace]
+default-members = ["avbroot"]
+members = ["avbroot", "e2e", "xtask"]
+resolver = "2"
+
+[workspace.package]
+version = "2.0.0"
+license = "GPL-3.0-only"
+edition = "2021"
+repository = "https://github.com/chenxiaolong/avbroot"

README.extra.md

@@ -0,0 +1,83 @@
+# avbroot extra
+
+avbroot includes several feature-complete parsers for various things, like boot images. Some of these are exposed as extra subcommands. They aren't needed for normal OTA patching, but may be useful in other scenarios.
+
+Note that while avbroot maintains a stable command line interface for the patching-related subcommands, these extra subcommands do not have backwards compatibility guarantees.
+
+## `avbroot avb`
+
+### Showing vbmeta header and footer information
+
+```bash
+avbroot avb dump -i <image>
+```
+
+This subcommand shows all of the vbmeta header and footer fields. `vbmeta` partition images will only have a header, while partitions with actual data (eg. boot images) will have both a header and a footer.
+
+### Verifying AVB hashes and signatures
+
+```bash
+avbroot avb verify -i <root vbmeta image> -p <public key>
+```
+
+This subcommand verifies the vbmeta header signature and the hashes for all vbmeta descriptors (including hashtree descriptors). If the vbmeta image has a chain descriptor for another partition, that partition image will be verified as well (recursively). All partitions are expected to be in the same directory as the vbmeta image being verified.
+
+If `-p` is omitted, the signatures and hashes are checked only for validity, not that they are trusted.
+
+## `avbroot boot`
+
+### Unpacking a boot image
+
+```bash
+avbroot boot unpack -i <input boot image>
+```
+
+This subcommand unpacks all of the components of the boot image into the current directory by default (see `--help`). The header fields are saved to `header.toml` and each blob section is saved to a separate file. Each blob is written to disk as-is, without decompression.
+
+### Packing a boot image
+
+```bash
+avbroot boot pack -o <output boot image>
+```
+
+This subcommand packs a new boot image from the individual components in the current directory by default (see `--help`). The default input filenames are the same as the output filenames for the `unpack` subcommand.
+
+### Repacking a boot image
+
+```bash
+avbroot boot repack -i <input boot image> -o <output boot image>
+```
+
+This subcommand repacks a boot image without writing the individual components to disk first. This is useful for roundtrip testing of avbroot's boot image parser. The output should be identical to the input, minus any footers, like the AVB footer.
+
+### Showing information about a boot image
+
+```bash
+avbroot boot info -i <input boot image>
+```
+
+All of the `boot` subcommands show the boot image information. This specific subcommand just does it without performing any other operation. To show avbroot's internal representation of the information, pass in `-d`.
+
+## `avbroot ramdisk`
+
+### Dumping a cpio archive
+
+```bash
+avbroot ramdisk dump -i <cpio archive>
+```
+
+This subcommand dumps all information about a cpio archive to stdout. This includes the compression format, all header fields (including the trailer entry), and all data. If an entry's data can be decoded as UTF-8, then it is printed out as text. Otherwise, the binary data is printed out `\x##`-encoded for non-ASCII bytes. The escape-encoded data is truncated to 512 bytes by default to avoid outputting too much data, but this behavior can be disabled with `--no-truncate`.
+
+### Repacking a cpio archive
+
+```bash
+avbroot ramdisk repack -i <input cpio archive> -o <output cpio archive>
+```
+
+This subcommand repacks a cpio archive, including recompression if needed. This is useful for roundtrip testing of avbroot's cpio parser and compression handling. The uncompressed output should be identical to the uncompressed input, except:
+
+* files are sorted by name
+* inodes are reassigned, starting from 300000
+* there is no excess padding at the end of the file
+
+The compressed output may differ from what other tools produce due to differences in compression levels and header metadata. avbroot avoids specifying header information where possible (eg. gzip timestamp) for reproducibility.

README.md

@@ -4,6 +4,8 @@ avbroot is a program for patching Android A/B-style OTA images for root access w
 
 Having a good understanding of how AVB and A/B OTAs work is recommended prior to using avbroot. At the very least, please make sure the [warnings and caveats](#warnings-and-caveats) are well-understood to avoid the risk of hard bricking.
 
+**NOTE:** avbroot 2.0 has been rewritten in Rust and no longer relies on any AOSP code. The CLI is fully backwards compatible, but the old Python implementation can be found in the `python` branch if needed.
+
 ## Patches
 
 avbroot applies two patches to the boot images:
@@ -44,94 +46,36 @@ The boot-related components are signed with an AVB key and OTA-related component
 1. Generate the AVB and OTA signing keys:
 
     ```bash
-    openssl genrsa 4096 | openssl pkcs8 -topk8 -scrypt -out avb.key
-    openssl genrsa 4096 | openssl pkcs8 -topk8 -scrypt -out ota.key
+    avbroot key generate-key -o avb.key
+    avbroot key generate-key -o ota.key
     ```
 
 2. Convert the public key portion of the AVB signing key to the AVB public key metadata format. This is the format that the bootloader requires when setting the custom root of trust.
 
     ```bash
-    python /path/to/avbroot/external/avb/avbtool.py extract_public_key --key avb.key --output avb_pkmd.bin
+    avbroot key extract-avb -k avb.key -o avb_pkmd.bin
     ```
 
 3. Generate a self-signed certificate for the OTA signing key. This is used by recovery for verifying OTA updates.
 
     ```bash
-    openssl req -new -x509 -sha256 -key ota.key -out ota.crt -days 10000 -subj '/CN=OTA/'
+    avbroot key generate-cert -k ota.key -o ota.crt
     ```
 
-## Installing dependencies
-
-avbroot depends on the `openssl` command line tool and the `lz4` and `protobuf` Python libraries. Also, Python 3.9 or newer is required.
-
-### Linux
-
-On Linux, the dependencies can be installed from the distro's package manager:
-
-| Distro     | Command                                                    |
-|------------|------------------------------------------------------------|
-| Alpine     | `sudo apk add openssl py3-lz4 py3-protobuf`                |
-| Arch Linux | `sudo pacman -S openssl python-lz4 python-protobuf`        |
-| Debian     | `sudo apt install openssl python3-lz4 python3-protobuf`    |
-| Fedora     | `sudo dnf install openssl python3-lz4 python3-protobuf`    |
-| OpenSUSE   | `sudo zypper install openssl python3-lz4 python3-protobuf` |
-| Ubuntu     | (Same as Debian)                                           |
-
-### Windows
-
-Installing openssl and python from the [Scoop package manager](https://scoop.sh/) is suggested.
-
-```powershell
-scoop install openssl python
-```
-
-Installing from other sources should work as well, but it might be necessary to manually add `openssl`'s installation directory to the `PATH` environment variable.
-
-To install the Python dependencies:
-
-1. Create a virtual environment (replacing `<directory>` with the path where it should be created):
-
-    ```powershell
-    python -m venv <directory>
-    ```
-
-2. Activate the virtual environment. This must be done in every new terminal session before running avbroot.
-
-    ```powershell
-    . <directory>\Scripts\Activate.ps1
-    ```
-
-3. Install the dependencies.
-
-    ```powershell
-    pip install -r requirements.txt
-    ```
+The commands above are provided for convenience. avbroot is compatible with any standard PKCS8-encoded 4096-bit RSA private key and X509 certificate (eg. like those generated by openssl).
 
 ## Usage
 
 1. Make sure the caveats listed above are understood. It is possible to hard brick by doing the wrong thing!
 
-2. Clone this git repo recursively, as there are several AOSP repositories included as submodules in the `external/` directory.
+2. Download the latest version from the [releases page](https://github.com/chenxiaolong/avbroot/releases). To verify the digital signature, see the [verifying digital signatures](#verifying-digital-signatures) section.
+
+3. Follow the steps to [generate signing keys](#generating-keys).
+
+4. Patch the full OTA ZIP.
 
     ```bash
-    git clone --recursive https://github.com/chenxiaolong/avbroot.git
-    ```
-
-    If the repo is already cloned, run the following command instead to fetch the submodules:
-
-    ```bash
-    git submodule update --init --recursive
-    ```
-
-3. Follow the steps to [install dependencies](#installing-dependencies).
-
-4. Follow the steps to [generate signing keys](#generating-keys).
-
-5. Patch the full OTA ZIP.
-
-    ```bash
-    python avbroot.py \
-        patch \
+    avbroot ota patch \
         --input /path/to/ota.zip \
         --privkey-avb /path/to/avb.key \
         --privkey-ota /path/to/ota.key \
@@ -145,18 +89,17 @@ To install the Python dependencies:
 
     If you prefer to use an existing boot image patched by the Magisk app or you want to use KernelSU, see the [advanced usage section](#advanced-usage).
 
-6. **[Initial setup only]** Unlock the bootloader. This will trigger a data wipe.
+5. **[Initial setup only]** Unlock the bootloader. This will trigger a data wipe.
 
-7. **[Initial setup only]** Extract the patched images from the patched OTA.
+6. **[Initial setup only]** Extract the patched images from the patched OTA.
 
     ```bash
-    python avbroot.py \
-        extract \
+    avbroot ota extract \
         --input /path/to/ota.zip.patched \
         --directory extracted
     ```
 
-8. **[Initial setup only]** Flash the patched images and the AVB public key metadata. This sets up the custom root of trust. Future updates are done by simply sideloading patched OTA zips.
+7. **[Initial setup only]** Flash the patched images and the AVB public key metadata. This sets up the custom root of trust. Future updates are done by simply sideloading patched OTA zips.
 
     ```bash
     # Flash the boot images that were extracted
@@ -172,13 +115,13 @@ To install the Python dependencies:
     fastboot flash avb_custom_key /path/to/avb_pkmd.bin
     ```
 
-9. **[Initial setup only]** Run `dmesg | grep libfs_avb` as root to verify that AVB is working properly. A message similar to the following is expected:
+8. **[Initial setup only]** Run `dmesg | grep libfs_avb` as root to verify that AVB is working properly. A message similar to the following is expected:
 
     ```bash
     init: [libfs_avb]Returning avb_handle with status: Success
     ```
 
-10. **[Initial setup only]** Lock the bootloader. This will trigger a data wipe again. **Do not uncheck `OEM unlocking`!**
+9. **[Initial setup only]** Lock the bootloader. This will trigger a data wipe again. **Do not uncheck `OEM unlocking`!**
 
     **WARNING**: If you are flashing CalyxOS, the setup wizard will [automatically turn off the `OEM unlocking` switch](https://github.com/CalyxOS/platform_packages_apps_SetupWizard/blob/7d2df25cedcbff83ddb608e628f9d97b38259c26/src/org/lineageos/setupwizard/SetupWizardApp.java#L135-L140). Make sure to manually reenable it again from Android's developer settings. Consider using [avbroot's `oemunlockonboot` Magisk module](#oemunlockonboot-enable-oem-unlocking-on-every-boot) to automatically ensure OEM unlocking is enabled on every boot.
 
@@ -186,7 +129,7 @@ To install the Python dependencies:
 
 To update Android or Magisk:
 
-1. Follow step 5 in [the previous section](#usage) to patch the new OTA (or an existing OTA with a newer Magisk APK).
+1. Follow step 4 in [the previous section](#usage) to patch the new OTA (or an existing OTA with a newer Magisk APK).
 
 2. Reboot to recovery mode. If stuck at a `No command` screen, press the volume up button once while holding down the power button.
 
@@ -196,16 +139,14 @@ To update Android or Magisk:
 
 ## avbroot Magisk modules
 
-avbroot's Magisk modules can be built by running:
+avbroot's Magisk modules can be found on the [releases page](https://github.com/chenxiaolong/avbroot/releases) or they can be built locally by running:
 
 ```bash
-python modules/build.py
+cargo xtask modules -a
 ```
 
 This requires Java and the Android SDK to be installed. The `ANDROID_HOME` environment variable should be set to the Android SDK path.
 
-Alternatively, prebuilt modules can be downloaded [from GitHub Actions](https://github.com/chenxiaolong/avbroot/actions/workflows/modules.yml?query=branch%3Amaster). Select the latest workflow run and then download `avbroot-modules-<version>` at the bottom of the page. Note that GitHub only allows downloading the file when logged in.
-
 ### `clearotacerts`: Blocking A/B OTA Updates
 
 Unpatched OTA updates are already blocked in recovery because the original OTA certificate has been replaced with the custom certificate. To disable automatic OTAs while booted into Android, turn off `Automatic system updates` in Android's Developer Options.
@@ -227,8 +168,7 @@ Magisk versions 25211 and newer require a writable partition for storing custom
 1. Extract the boot image from the original/unpatched OTA:
 
     ```bash
-    python avbroot.py \
-        extract \
+    avbroot ota extract \
         --input /path/to/ota.zip \
         --directory . \
         --boot-only
@@ -245,8 +185,7 @@ Magisk versions 25211 and newer require a writable partition for storing custom
     Alternatively, avbroot can print out what Magisk detected by running:
 
     ```bash
-    python avbroot.py \
-        magisk-info \
+    avbroot ota magisk-info \
         --image magisk_patched-*.img
     ```
 
@@ -256,6 +195,55 @@ Magisk versions 25211 and newer require a writable partition for storing custom
 
 If it's not possible to run the Magisk app on the target device (eg. device is currently unbootable), patch and flash the OTA once using `--ignore-magisk-warnings`, follow these steps, and then repatch and reflash the OTA with `--magisk-preinit-device <name>`.
 
+## Verifying OTAs
+
+To verify all signatures and hashes related to the OTA installation and AVB boot process, run:
+
+```bash
+avbroot ota verify \
+    --input /path/to/ota.zip \
+    --cert-ota /path/to/ota.crt \
+    --public-key-avb /path/to/avb_pkmd.bin
+```
+
+If the `--cert-ota` and `--public-key-avb` options are omitted, then the signatures are only checked for validity, not that they are trusted.
+
+## Tab completion
+
+Since avbroot has tons of command line options, it may be useful to set up tab completions for the shell. These configs can be generated from avbroot itself.
+
+#### bash
+
+Add to `~/.bashrc`:
+
+```bash
+eval "$(avbroot completion -s bash)"
+```
+
+#### zsh
+
+Add to `~/.zshrc`:
+
+```bash
+eval "$(avbroot completion -s zsh)"
+```
+
+#### fish
+
+Add to `~/.config/fish/config.fish`:
+
+```bash
+avbroot completion -s fish | source
+```
+
+#### PowerShell
+
+Add to PowerShell's `profile.ps1` startup script:
+
+```powershell
+Invoke-Expression (& avbroot completion -s powershell)
+```
+
 ## Advanced Usage
 
 ### Using a prepatched boot image
@@ -295,18 +283,18 @@ avbroot prompts for the private key passphrases interactively by default. To run
 * Supply the passphrases via files:
 
     ```bash
-    avbroot patch \
-        --passphrase-avb-file /path/to/avb.passphrase \
-        --passphrase-ota-file /path/to/ota.passphrase \
+    avbroot ota patch \
+        --pass-avb-file /path/to/avb.passphrase \
+        --pass-ota-file /path/to/ota.passphrase \
         <...>
     ```
 
     On Unix-like systems, the "files" can be pipes. With shells that support process substituion (bash, zsh, etc.), the passphrase can be queried from a command (eg. querying a password manager).
 
     ```bash
-    avbroot patch \
-        --passphrase-avb-file <(command to query AVB passphrase) \
-        --passphrase-ota-file <(command to query OTA passphrase) \
+    avbroot ota patch \
+        --pass-avb-file <(command to query AVB passphrase) \
+        --pass-ota-file <(command to query OTA passphrase) \
         <...>
     ```
 
@@ -316,41 +304,58 @@ avbroot prompts for the private key passphrases interactively by default. To run
     export PASSPHRASE_AVB="the AVB passphrase"
     export PASSPHRASE_OTA="the OTA passphrase"
 
-    avbroot patch \
-        --passphrase-avb-env-var PASSPHRASE_AVB \
-        --passphrase-ota-env-var PASSPHRASE_OTA \
+    avbroot ota patch \
+        --pass-avb-env-var PASSPHRASE_AVB \
+        --pass-ota-env-var PASSPHRASE_OTA \
         <...>
     ```
 
-* Use unencrypted private keys. This is not recommended, but can be done by:
-
-    ```bash
-    openssl pkcs8 -in avb.key -topk8 -nocrypt -out avb.unencrypted.key
-    openssl pkcs8 -in ota.key -topk8 -nocrypt -out ota.unencrypted.key
-    ```
+* Use unencrypted private keys. This is strongly discouraged.
 
 ### Extracting the entire OTA
 
 To extract all images contained within the OTA's `payload.bin`, run:
 
 ```bash
-python avbroot.py \
-    extract \
+avbroot ota extract \
     --input /path/to/ota.zip \
     --directory extracted \
     --all
 ```
 
-## Implementation Details
+## Building from source
 
-* avbroot relies on AOSP's avbtool and OTA utilities. These are collections of applications that aren't meant to be used as libraries, but avbroot shoehorns them in anyway. These tools are not called via CLI because avbroot requires more control over the operations being performed than what is provided via the CLI interfaces. This "integration" is incredibly hacky and will likely require changes whenever the submodules are updated to point to newer AOSP commits.
+Make sure the [Rust toolchain](https://www.rust-lang.org/) is installed. Then run:
 
-* AVB has two methods of handling signature verification:
+```bash
+cargo build --release
+```
 
-    * An image can have an unsigned vbmeta footer, which causes the image's hash to be embedded in the (signed) root `vbmeta` image via vbmeta hash descriptors.
-    * An image can have a signed vbmeta footer, which causes a public key for verification to be embedded in the root `vbmeta` image via vbmeta chainload descriptors. This is meant for out-of-band updates where signed images can be updated without also updating the root `vbmeta` image.
+The output binary is written to `target/release/avbroot`.
 
-    avbroot preserves whether an image uses a chainload or hash descriptor. If a boot image was previously signed, then it will be signed with the AVB key during patching. This preserves the state of the AVB rollback indices, which makes it possible to flip between the original and patched images without a factory reset while debugging avbroot (with the bootloader unlocked).
+Debug builds work too, but they will run significantly slower (in the sha256 computations) due to compiler optimizations being turned off.
+
+By default, the build links to the system's bzip2 and liblzma libraries, which are the only external libraries avbroot depends on. To compile and statically link these two libraries, pass in `--features static`.
+
+## Verifying digital signatures
+
+First, save the public key to a file listing the keys to be trusted.
+
+```bash
+echo 'avbroot ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIDOe6/tBnO7xZhAWXRj3ApUYgn+XZ0wnQiXM8B7tPgv4' > avbroot_trusted_keys
+```
+
+Then, verify the signature of the zip file using the list of trusted keys.
+
+```bash
+ssh-keygen -Y verify -f avbroot_trusted_keys -I avbroot -n file -s <file>.zip.sig < <file>.zip
+```
+
+If the file is successfully verified, the output will be:
+
+```
+Good "file" signature for avbroot with ED25519 key SHA256:Ct0HoRyrFLrnF9W+A/BKEiJmwx7yWkgaW/JvghKrboA
+```
 
 ## Contributing
 

RELEASE.md

@@ -0,0 +1,7 @@
+The changelog can be found at: [`CHANGELOG.md`](./CHANGELOG.md).
+
+---
+
+See [`README.md`](./README.md) for information on how to use avbroot.
+
+The downloads are digitally signed. Please consider [verifying the digital signatures](./README.md#verifying-digital-signatures) of the binaries (or building from source) since avbroot is an application with access to your OTA/AVB signing keys.

avbroot.py

@@ -1,6 +0,0 @@
-#!/usr/bin/env python3
-
-from avbroot import main
-
-if __name__ == '__main__':
-    main.main()

avbroot/Cargo.toml

@@ -0,0 +1,76 @@
+[package]
+name = "avbroot"
+version.workspace = true
+license.workspace = true
+edition.workspace = true
+repository.workspace = true
+publish = false
+
+# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+
+[dependencies]
+anyhow = "1.0.75"
+base64 = "0.21.3"
+byteorder = "1.4.3"
+clap = { version = "4.4.1", features = ["derive"] }
+clap_complete = "4.4.0"
+cms = { version = "0.2.2", features = ["std"] }
+const-oid = "0.9.5"
+ctrlc = "3.4.0"
+flate2 = "1.0.27"
+hex = "0.4.3"
+lz4_flex = "0.11.1"
+memchr = "2.6.0"
+num-bigint-dig = "0.8.4"
+num-traits = "0.2.16"
+phf = { version = "0.11.2", features = ["macros"] }
+pkcs8 = { version = "0.10.2", features = ["encryption", "pem"] }
+quick-protobuf = "0.8.1"
+rand = "0.8.5"
+rayon = "1.7.0"
+regex = { version = "1.9.4", default-features = false, features = ["perf", "std"] }
+# We use ring instead of sha2 for sha256 digest computation of large files
+# because sha2 is significantly slower on older x86_64 CPUs without the SHA-NI
+# instructions. sha2 is still used for signing purposes.
+# https://github.com/RustCrypto/hashes/issues/327
+ring = "0.16.20"
+rpassword = "7.2.0"
+rsa = { version = "0.9.2", features = ["sha1", "sha2"] }
+serde = { version = "1.0.188", features = ["derive"] }
+sha1 = "0.10.5"
+sha2 = "0.10.7"
+tempfile = "3.8.0"
+thiserror = "1.0.47"
+toml_edit = { version = "0.19.14", features = ["serde"] }
+topological-sort = "0.2.2"
+x509-cert = { version = "0.2.4", features = ["builder"] }
+xz2 = "0.1.7"
+
+# There's an upstream bug that causes an infinite loop in the write::BzDecoder
+# destructor if the decoder is fed invalid data. While this never happens during
+# normal operation, it is possible to run into this by running `ota extract`
+# against a `--stripped` OTA file.
+# https://github.com/alexcrichton/bzip2-rs/pull/99
+[dependencies.bzip2]
+git = "https://github.com/jongiddy/bzip2-rs"
+rev = "2aefcb4d3634de1df226c73d93f758d65228bb8c"
+
+# https://github.com/zip-rs/zip/pull/383
+[dependencies.zip]
+git = "https://github.com/chenxiaolong/zip"
+rev = "989101f9384b9e94e36e6e9e0f51908fdf98bde6"
+default-features = false
+features = ["deflate"]
+
+[target.'cfg(unix)'.dependencies]
+rustix = { version = "0.38.9", default-features = false, features = ["process"] }
+
+[build-dependencies]
+# Disable the clap feature since it pulls in an ancient version of clap.
+pb-rs = { version = "0.10.0", default-features = false }
+
+[dev-dependencies]
+assert_matches = "1.5.0"
+
+[features]
+static = ["bzip2/static", "xz2/static"]

avbroot/__init__.py

@@ -1,13 +0,0 @@
-import os
-import sys
-
-external_dir = os.path.join(os.path.realpath(os.path.dirname(__file__)),
-                            '..', 'external')
-
-# OTA utilities (loaded first because there are multiple common.py files and
-# this is the one we need to import)
-sys.path.append(os.path.join(external_dir, 'build', 'tools', 'releasetools'))
-# avbtool
-sys.path.append(os.path.join(external_dir, 'avb'))
-# Payload protobuf
-sys.path.append(os.path.join(external_dir, 'update_engine', 'scripts'))

avbroot/boot.py

@@ -1,480 +0,0 @@
-import hashlib
-import io
-import lzma
-import re
-import shutil
-import zipfile
-
-import avbtool
-
-from . import openssl
-from . import util
-from . import vbmeta
-from .formats import bootimage
-from .formats import compression
-from .formats import cpio
-
-
-def _load_ramdisk(ramdisk):
-    with (
-        io.BytesIO(ramdisk) as f_raw,
-        compression.CompressedFile(f_raw, 'rb') as f,
-    ):
-        return cpio.load(f.fp), f.format
-
-
-def _save_ramdisk(entries, format):
-    with io.BytesIO() as f_raw:
-        with compression.CompressedFile(f_raw, 'wb', format=format) as f:
-            cpio.save(f.fp, entries)
-
-        return f_raw.getvalue()
-
-
-class BootImagePatch:
-    def __call__(self, image_file):
-        with open(image_file, 'r+b') as f:
-            boot_image = bootimage.load_autodetect(f)
-
-            boot_image = self.patch(image_file, boot_image)
-
-            f.seek(0)
-            f.truncate(0)
-
-            boot_image.generate(f)
-
-    def patch(self, image_file, boot_image):
-        raise NotImplementedError()
-
-
-class MagiskRootPatch(BootImagePatch):
-    '''
-    Root the boot image with Magisk.
-    '''
-
-    # - Half-open intervals.
-    # - Versions <25102 are not supported because they're missing commit
-    #   1f8c063dc64806c4f7320ed66c785ff7bc116383, which would leave devices
-    #   that use Android 13 GKIs unable to boot into recovery
-    # - Versions 25207 through 25210 are not supported because they used the
-    #   RULESDEVICE config option, which stored the writable block device as an
-    #   rdev major/minor pair, which was not consistent across reboots and was
-    #   replaced by PREINITDEVICE
-    VERS_SUPPORTED = (
-        util.Range(25102, 25207),
-        util.Range(25211, 26200),
-    )
-    VER_PREINIT_DEVICE = util.Range(25211, VERS_SUPPORTED[-1].end)
-    VER_RANDOM_SEED = util.Range(25211, VERS_SUPPORTED[-1].end)
-
-    def __init__(self, magisk_apk, preinit_device, random_seed):
-        self.magisk_apk = magisk_apk
-        self.version = self._get_version()
-
-        self.preinit_device = preinit_device
-
-        if random_seed is None:
-            # Use a hardcoded random seed by default to ensure byte-for-byte
-            # reproducibility
-            self.random_seed = 0xfedcba9876543210
-        else:
-            self.random_seed = random_seed
-
-    def _get_version(self):
-        with zipfile.ZipFile(self.magisk_apk, 'r') as z:
-            with z.open('assets/util_functions.sh', 'r') as f:
-                for line in f:
-                    if line.startswith(b'MAGISK_VER_CODE='):
-                        return int(line[16:].strip())
-
-        raise Exception('Failed to determine Magisk version from: '
-                        f'{self.magisk_apk}')
-
-    def validate(self):
-        if not any(self.version in s for s in self.VERS_SUPPORTED):
-            supported = '; '.join(str(s) for s in self.VERS_SUPPORTED)
-            raise ValueError(f'Unsupported Magisk version {self.version} '
-                             f'(supported: {supported})')
-
-        if self.preinit_device is None and \
-                self.version in self.VER_PREINIT_DEVICE:
-            raise ValueError(f'Magisk version {self.version} '
-                             f'({self.VER_PREINIT_DEVICE}) requires a preinit '
-                             f'device to be specified')
-
-    def patch(self, image_file, boot_image):
-        with zipfile.ZipFile(self.magisk_apk, 'r') as zip:
-            return self._patch(image_file, boot_image, zip)
-
-    def _patch(self, image_file, boot_image, zip):
-        if len(boot_image.ramdisks) > 1:
-            raise Exception('Boot image is not expected to have '
-                            f'{len(boot_image.ramdisks)} ramdisks')
-
-        # Magisk saves the original SHA1 digest in its config file
-        with open(image_file, 'rb') as f:
-            hasher = util.hash_file(f, hashlib.sha1())
-
-        # Load the existing ramdisk if it exists. If it doesn't, we have to
-        # generate one from scratch
-        if boot_image.ramdisks:
-            entries, ramdisk_format = _load_ramdisk(boot_image.ramdisks[0])
-        else:
-            entries, ramdisk_format = [], compression.Format.LZ4_LEGACY
-
-        old_entries = entries.copy()
-
-        # Create magisk directory structure
-        for path, perms in (
-            (b'overlay.d', 0o750),
-            (b'overlay.d/sbin', 0o750),
-        ):
-            entries.append(cpio.CpioEntryNew.new_directory(path, perms=perms))
-
-        # Delete the original init
-        if boot_image.ramdisks:
-            entries = [e for e in entries if e.name != b'init']
-
-        # Add magiskinit
-        with zip.open('lib/arm64-v8a/libmagiskinit.so', 'r') as f:
-            entries.append(cpio.CpioEntryNew.new_file(
-                b'init', perms=0o750, data=f.read()))
-
-        # Add xz-compressed magisk32 and magisk64
-        xz_files = {
-            'lib/armeabi-v7a/libmagisk32.so': b'magisk32.xz',
-            'lib/arm64-v8a/libmagisk64.so': b'magisk64.xz',
-        }
-
-        # Add stub apk, which only exists after the Magisk commit:
-        # ad0e6511e11ebec65aa9b5b916e1397342850319
-        if 'assets/stub.apk' in zip.namelist():
-            xz_files['assets/stub.apk'] = b'stub.xz'
-
-        for source, target in xz_files.items():
-            with (
-                zip.open(source, 'r') as f_in,
-                io.BytesIO() as f_out_raw,
-            ):
-                with lzma.open(f_out_raw, 'wb', preset=9,
-                               check=lzma.CHECK_CRC32) as f_out:
-                    shutil.copyfileobj(f_in, f_out)
-
-                entries.append(cpio.CpioEntryNew.new_file(
-                    b'overlay.d/sbin/' + target, perms=0o644,
-                    data=f_out_raw.getvalue()))
-
-        # Create magisk .backup directory structure
-        self._apply_magisk_backup(old_entries, entries)
-
-        # Create magisk config
-        magisk_config = \
-            b'KEEPVERITY=true\n' \
-            b'KEEPFORCEENCRYPT=true\n' \
-            b'PATCHVBMETAFLAG=false\n' \
-            b'RECOVERYMODE=false\n'
-
-        if self.version in self.VER_PREINIT_DEVICE:
-            magisk_config += b'PREINITDEVICE=%s\n' % \
-                self.preinit_device.encode('ascii')
-
-        magisk_config += b'SHA1=%s\n' % hasher.hexdigest().encode('ascii')
-
-        if self.version in self.VER_RANDOM_SEED:
-            magisk_config += b'RANDOMSEED=0x%x\n' % self.random_seed
-
-        entries.append(cpio.CpioEntryNew.new_file(
-            b'.backup/.magisk', perms=0o000, data=magisk_config))
-
-        # Repack ramdisk
-        new_ramdisk = _save_ramdisk(entries, ramdisk_format)
-        if boot_image.ramdisks:
-            boot_image.ramdisks[0] = new_ramdisk
-        else:
-            boot_image.ramdisks.append(new_ramdisk)
-
-        return boot_image
-
-    @staticmethod
-    def _apply_magisk_backup(old_entries, new_entries):
-        '''
-        Compare old and new ramdisk entry lists, creating the Magisk `.backup/`
-        directory structure. `.backup/.rmlist` will contain a sorted list of
-        NULL-terminated strings, listing which files were newly added or
-        changed. The old entries for changed files will be added to the new
-        entries as `.backup/<path>`.
-
-        Both lists and entries within the lists may be mutated.
-        '''
-
-        old_by_name = {e.name: e for e in old_entries}
-        new_by_name = {e.name: e for e in new_entries}
-
-        added = new_by_name.keys() - old_by_name.keys()
-        deleted = old_by_name.keys() - new_by_name.keys()
-        changed = set(n for n in old_by_name.keys() & new_by_name.keys()
-                      if old_by_name[n].content != new_by_name[n].content)
-
-        new_entries.append(cpio.CpioEntryNew.new_directory(
-            b'.backup', perms=0o000))
-
-        for name in deleted | changed:
-            entry = old_by_name[name]
-            entry.name = b'.backup/' + entry.name
-            new_entries.append(entry)
-
-        rmlist_data = b''.join(n + b'\0' for n in sorted(added))
-        new_entries.append(cpio.CpioEntryNew.new_file(
-            b'.backup/.rmlist', perms=0o000, data=rmlist_data))
-
-
-class OtaCertPatch(BootImagePatch):
-    '''
-    Replace the OTA certificates in the vendor_boot image with the custom OTA
-    signing certificate.
-    '''
-
-    OTACERTS_PATH = b'system/etc/security/otacerts.zip'
-
-    def __init__(self, cert_ota):
-        self.cert_ota = cert_ota
-
-    def patch(self, image_file, boot_image):
-        found_otacerts = False
-
-        # Check each ramdisk
-        for i, ramdisk in enumerate(boot_image.ramdisks):
-            entries, ramdisk_format = _load_ramdisk(ramdisk)
-
-            # Fail hard if otacerts does not exist. We don't want to lock the
-            # user out of future updates if the OTA certificate mechanism has
-            # changed.
-            otacerts = next((e for e in entries if e.name ==
-                             self.OTACERTS_PATH), None)
-            if otacerts:
-                found_otacerts = True
-            else:
-                continue
-
-            # Create new otacerts archive. The old certs are ignored since
-            # flashing a stock OTA will render the device unbootable.
-            with io.BytesIO() as f_zip:
-                with zipfile.ZipFile(f_zip, 'w') as z:
-                    # Use zeroed-out metadata to ensure the archive is bit for
-                    # bit reproducible across runs.
-                    info = zipfile.ZipInfo('ota.x509.pem')
-                    # Mark entry as created on Unix for reproducibility
-                    info.create_system = 3
-                    with (
-                        z.open(info, 'w') as f_out,
-                        open(self.cert_ota, 'rb') as f_in,
-                    ):
-                        shutil.copyfileobj(f_in, f_out)
-
-                otacerts.content = f_zip.getvalue()
-
-            # Repack ramdisk
-            boot_image.ramdisks[i] = _save_ramdisk(entries, ramdisk_format)
-
-        if not found_otacerts:
-            raise Exception(f'{self.OTACERTS_PATH} not found in ramdisk')
-
-        return boot_image
-
-
-class PrepatchedImage(BootImagePatch):
-    '''
-    Replace the boot image with a prepatched boot image if it is compatible.
-
-    An image is compatible if all the non-size-related header fields are
-    identical and the set of included sections (eg. kernel, dtb) are the same.
-    The only exception is the number of ramdisk sections, which is allowed to
-    be higher than the original image.
-    '''
-
-    MIN_LEVEL = 0
-    MAX_LEVEL = 2
-
-    VERSION_REGEX = re.compile(
-        b'Linux version (\d+\.\d+).\d+-(android\d+)-(\d+)-')
-
-    def __init__(self, prepatched, fatal_level, warning_fn):
-        self.prepatched = prepatched
-        self.fatal_level = fatal_level
-        self.warning_fn = warning_fn
-
-    def patch(self, image_file, boot_image):
-        with open(self.prepatched, 'r+b') as f:
-            prepatched_image = bootimage.load_autodetect(f)
-
-        old_header = boot_image.to_dict()
-        new_header = prepatched_image.to_dict()
-
-        # Level 0: Warnings that don't affect booting
-        # Level 1: Warnings that may affect booting
-        # Level 2: Warnings that are very likely to affect booting
-        issues = [[], [], []]
-
-        for k in new_header.keys() - old_header.keys():
-            issues[2].append(f'{k} header field was added')
-
-        for k in old_header.keys() - new_header.keys():
-            issues[2].append(f'{k} header field was removed')
-
-        for k in old_header.keys() & new_header.keys():
-            if old_header[k] != new_header[k]:
-                if k in ('id', 'os_version'):
-                    level = 0
-                elif k in ('cmdline', 'extra_cmdline'):
-                    level = 1
-                else:
-                    level = 2
-
-                issues[level].append(f'{k} header field was changed: '
-                                     f'{old_header[k]} -> {new_header[k]}')
-
-        for attr in 'kernel', 'second', 'recovery_dtbo', 'dtb', 'bootconfig':
-            original_val = getattr(boot_image, attr)
-            prepatched_val = getattr(prepatched_image, attr)
-
-            if original_val is None and prepatched_val is not None:
-                issues[1].append(f'{attr} section was added')
-            elif original_val is not None and prepatched_val is None:
-                issues[2].append(f'{attr} section was removed')
-
-        if len(prepatched_image.ramdisks) < len(boot_image.ramdisks):
-            issues[2].append('Number of ramdisk sections decreased: '
-                             f'{len(boot_image.ramdisks)} -> '
-                             f'{len(prepatched_image.ramdisks)}')
-
-        if boot_image.kernel is not None:
-            old_kmi = self._get_kmi_version(boot_image)
-            new_kmi = self._get_kmi_version(prepatched_image)
-
-            if old_kmi != new_kmi:
-                issues[2].append('Kernel module interface version changed: '
-                                 f'{old_kmi} -> {new_kmi}')
-
-        warnings = [e for i in range(self.MIN_LEVEL,
-                                     min(self.MAX_LEVEL + 1, self.fatal_level))
-                    for e in issues[i]]
-        errors = [e for i in range(max(self.MIN_LEVEL, self.fatal_level),
-                                   self.MAX_LEVEL + 1)
-                  for e in issues[i]]
-
-        if warnings:
-            self.warning_fn('The prepatched boot image may not be compatible '
-                            'with the original:\n' +
-                            '\n'.join(f'- {w}' for w in warnings))
-
-        if errors:
-            raise ValueError('The prepatched boot image is not compatible '
-                             'with the original:\n' +
-                             '\n'.join(f'- {e}' for e in errors))
-
-        return prepatched_image
-
-    @classmethod
-    def _get_kmi_version(cls, boot_image):
-        try:
-            with (
-                io.BytesIO(boot_image.kernel) as f_raw,
-                compression.CompressedFile(f_raw, 'rb') as f,
-            ):
-                decompressed = f.fp.read()
-        except ValueError:
-            decompressed = boot_image.kernel
-
-        m = cls.VERSION_REGEX.search(decompressed)
-        if not m:
-            return None
-
-        return b'-'.join(m.groups()).decode('ascii')
-
-
-def patch_boot(avb, input_path, output_path, key, passphrase,
-               only_if_previously_signed, patch_funcs):
-    '''
-    Call each function in patch_funcs against a boot image with vbmeta stripped
-    out and then resign the image using the provided private key.
-    '''
-
-    image = avbtool.ImageHandler(input_path, read_only=True)
-    footer, header, descriptors, image_size = avb._parse_image(image)
-
-    have_key_old = not not header.public_key_size
-    if not have_key_old and only_if_previously_signed:
-        key = None
-
-    have_key_new = not not key
-
-    if have_key_old != have_key_new:
-        raise Exception('Key presence does not match: %s (old) != %s (new)' %
-                        (have_key_old, have_key_new))
-
-    hash = None
-    new_descriptors = []
-
-    for d in descriptors:
-        if isinstance(d, avbtool.AvbHashDescriptor):
-            if hash is not None:
-                raise Exception('Expected only one hash descriptor')
-            hash = d
-        else:
-            new_descriptors.append(d)
-
-    if hash is None:
-        raise Exception('No hash descriptor found')
-
-    algorithm_name = avbtool.lookup_algorithm_by_type(header.algorithm_type)[0]
-
-    # Pixel 7's init_boot image is originally signed by a 2048-bit RSA key, but
-    # avbroot expects RSA 4096 keys
-    if algorithm_name == 'SHA256_RSA2048':
-        algorithm_name = 'SHA256_RSA4096'
-
-    with util.open_output_file(output_path) as f:
-        shutil.copyfile(input_path, f.name)
-
-        # Strip the vbmeta footer from the boot image
-        avb.erase_footer(f.name, False)
-
-        # Invoke the patching functions
-        for patch_func in patch_funcs:
-            patch_func(f.name)
-
-        # Sign the new boot image
-        with (
-            vbmeta.smuggle_descriptors(),
-            openssl.inject_passphrase(passphrase),
-        ):
-            avb.add_hash_footer(
-                image_filename=f.name,
-                partition_size=image_size,
-                dynamic_partition_size=False,
-                partition_name=hash.partition_name,
-                hash_algorithm=hash.hash_algorithm,
-                salt=hash.salt.hex(),
-                chain_partitions=None,
-                algorithm_name=algorithm_name,
-                key_path=key,
-                public_key_metadata_path=None,
-                rollback_index=header.rollback_index,
-                flags=header.flags,
-                rollback_index_location=header.rollback_index_location,
-                props=None,
-                props_from_file=None,
-                kernel_cmdlines=new_descriptors,
-                setup_rootfs_from_kernel=None,
-                include_descriptors_from_image=None,
-                calc_max_image_size=False,
-                signing_helper=None,
-                signing_helper_with_files=None,
-                release_string=header.release_string,
-                append_to_release_string=None,
-                output_vbmeta_image=None,
-                do_not_append_vbmeta_image=False,
-                print_required_libavb_version=False,
-                use_persistent_digest=False,
-                do_not_use_ab=False,
-            )

avbroot/build.rs

@@ -0,0 +1,43 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::{env, ffi::OsStr, fs, io, path::Path};
+
+use pb_rs::{types::FileDescriptor, ConfigBuilder};
+
+fn main() {
+    let out_dir = Path::new(&env::var("OUT_DIR").unwrap()).join("protobuf");
+    let in_dir = Path::new(&env::var("CARGO_MANIFEST_DIR").unwrap()).join("protobuf");
+
+    println!("cargo:rerun-if-changed={}", in_dir.to_str().unwrap());
+
+    let mut protos = Vec::new();
+
+    for entry in fs::read_dir(&in_dir).unwrap() {
+        let path = entry.unwrap().path();
+        if path.extension() == Some(OsStr::new("proto")) {
+            println!("cargo:rerun-if-changed={}", path.to_str().unwrap());
+            protos.push(path);
+        }
+    }
+
+    match fs::remove_dir_all(&out_dir) {
+        Err(e) if e.kind() == io::ErrorKind::NotFound => {}
+        r => r.unwrap(),
+    }
+
+    fs::create_dir_all(&out_dir).unwrap();
+
+    let config = ConfigBuilder::new(&protos, None, Some(&out_dir), &[in_dir])
+        .unwrap()
+        .dont_use_cow(true)
+        // We're using this as a means to force quick-protobuf to use BTreeMap
+        // instead of HashMap so that the serialized messages are reproducible.
+        // https://github.com/tafia/quick-protobuf/issues/251
+        .nostd(true)
+        .build();
+
+    FileDescriptor::run(&config).unwrap();
+}

avbroot/formats/bootimage.py

@@ -1,771 +0,0 @@
-import collections
-import os
-import struct
-import typing
-
-from . import padding
-from .. import util
-
-
-BOOT_MAGIC = b'ANDROID!'
-BOOT_NAME_SIZE = 16
-BOOT_ARGS_SIZE = 512
-BOOT_EXTRA_ARGS_SIZE = 1024
-
-VENDOR_BOOT_MAGIC = b'VNDRBOOT'
-VENDOR_BOOT_ARGS_SIZE = 2048
-VENDOR_BOOT_NAME_SIZE = 16
-
-VENDOR_RAMDISK_TYPE_NONE = 0
-VENDOR_RAMDISK_TYPE_PLATFORM = 1
-VENDOR_RAMDISK_TYPE_RECOVERY = 2
-VENDOR_RAMDISK_TYPE_DLKM = 3
-VENDOR_RAMDISK_NAME_SIZE = 32
-VENDOR_RAMDISK_TABLE_ENTRY_BOARD_ID_SIZE = 16
-
-PAGE_SIZE = 4096
-
-BOOT_IMG_HDR_V0 = struct.Struct(
-    '<'
-    f'{len(BOOT_MAGIC)}s'  # magic
-    'I'  # kernel_size
-    'I'  # kernel_addr
-    'I'  # ramdisk_size
-    'I'  # ramdisk_addr
-    'I'  # second_size
-    'I'  # second_addr
-    'I'  # tags_addr
-    'I'  # page_size
-    'I'  # header_version
-    'I'  # os_version
-    f'{BOOT_NAME_SIZE}s'  # name
-    f'{BOOT_ARGS_SIZE}s'  # cmdline
-    f'{8 * 4}s'  # id (uint32_t[8])
-    f'{BOOT_EXTRA_ARGS_SIZE}s'  # extra_cmdline
-)
-
-BOOT_IMG_HDR_V1_EXTRA = struct.Struct(
-    '<'
-    'I'  # recovery_dtbo_size
-    'Q'  # recovery_dtbo_offset
-    'I'  # header_size
-)
-
-BOOT_IMG_HDR_V2_EXTRA = struct.Struct(
-    '<'
-    'I'  # dtb_size
-    'Q'  # dtb_addr
-)
-
-BOOT_IMG_HDR_V3 = struct.Struct(
-    '<'
-    f'{len(BOOT_MAGIC)}s'  # magic
-    'I'  # kernel_size
-    'I'  # ramdisk_size
-    'I'  # os_version
-    'I'  # header_size
-    '16s'  # reserved (uint32_t[4])
-    'I'  # header_version
-    f'{BOOT_ARGS_SIZE + BOOT_EXTRA_ARGS_SIZE}s'  # cmdline
-)
-
-VENDOR_BOOT_IMG_HDR_V3 = struct.Struct(
-    '<'
-    f'{len(VENDOR_BOOT_MAGIC)}s'  # magic
-    'I'  # header_version
-    'I'  # page_size
-    'I'  # kernel_addr
-    'I'  # ramdisk_addr
-    'I'  # vendor_ramdisk_size
-    f'{VENDOR_BOOT_ARGS_SIZE}s'  # cmdline
-    'I'  # tags_addr
-    f'{VENDOR_BOOT_NAME_SIZE}s'  # name
-    'I'  # header_size
-    'I'  # dtb_size
-    'Q'  # dtb_addr
-)
-
-BOOT_IMG_HDR_V4_EXTRA = struct.Struct(
-    '<'
-    'I'  # signature_size
-)
-
-VENDOR_BOOT_IMG_HDR_V4_EXTRA = struct.Struct(
-    '<'
-    'I'  # vendor_ramdisk_table_size
-    'I'  # vendor_ramdisk_table_entry_num
-    'I'  # vendor_ramdisk_table_entry_size
-    'I'  # bootconfig_size
-)
-
-VENDOR_RAMDISK_TABLE_ENTRY_V4 = struct.Struct(
-    '<'
-    'I'  # ramdisk_size
-    'I'  # ramdisk_offset
-    'I'  # ramdisk_type
-    f'{VENDOR_RAMDISK_NAME_SIZE}s'  # ramdisk_name
-    f'{VENDOR_RAMDISK_TABLE_ENTRY_BOARD_ID_SIZE * 4}s'  # board_id (uint32_t[])
-)
-
-
-class WrongFormat(ValueError):
-    pass
-
-
-class BootImage:
-    def __init__(
-        self,
-        f: typing.Optional[typing.BinaryIO] = None,
-        data: typing.Optional[dict[str, typing.Any]] = None,
-    ) -> None:
-        assert (f is None) != (data is None)
-
-        self.kernel: typing.Optional[bytes] = None
-        self.ramdisks: list[bytes] = []
-        self.second: typing.Optional[bytes] = None
-        self.recovery_dtbo: typing.Optional[bytes] = None
-        self.dtb: typing.Optional[bytes] = None
-        self.bootconfig: typing.Optional[bytes] = None
-
-        if f:
-            self._from_file(f)
-        else:
-            self._from_dict(data)
-
-    def _from_file(self, f: typing.BinaryIO) -> None:
-        raise NotImplementedError()
-
-    def generate(self, f: typing.BinaryIO) -> None:
-        raise NotImplementedError()
-
-    def _from_dict(self, data: dict[str, typing.Any]) -> None:
-        raise NotImplementedError()
-
-    def to_dict(self) -> None:
-        raise NotImplementedError()
-
-
-class _BootImageV0Through2(BootImage):
-    def _from_file(self, f: typing.BinaryIO) -> None:
-        # Common fields for v0 through v2
-        magic, kernel_size, kernel_addr, ramdisk_size, ramdisk_addr, \
-            second_size, second_addr, tags_addr, page_size, header_version, \
-            os_version, name, cmdline, id, extra_cmdline = \
-            BOOT_IMG_HDR_V0.unpack(util.read_exact(f, BOOT_IMG_HDR_V0.size))
-
-        if magic != BOOT_MAGIC:
-            raise WrongFormat(f'Unknown magic: {magic}')
-        elif header_version not in (0, 1, 2):
-            raise WrongFormat(f'Unknown header version: {header_version}')
-
-        self.kernel_addr = kernel_addr
-        self.ramdisk_addr = ramdisk_addr
-        self.second_addr = second_addr
-        self.tags_addr = tags_addr
-        self.page_size = page_size
-        self.header_version = header_version
-        self.os_version = os_version
-        self.name = name.rstrip(b'\0')
-        self.cmdline = cmdline.rstrip(b'\0')
-        self.id = id
-        self.extra_cmdline = extra_cmdline.rstrip(b'\0')
-
-        # Parse v1 fields
-        if header_version >= 1:
-            recovery_dtbo_size, recovery_dtbo_offset, header_size = \
-                BOOT_IMG_HDR_V1_EXTRA.unpack(
-                    util.read_exact(f, BOOT_IMG_HDR_V1_EXTRA.size))
-
-            self.recovery_dtbo_offset = recovery_dtbo_offset
-
-        # Parse v2 fields
-        if header_version == 2:
-            dtb_size, dtb_addr = BOOT_IMG_HDR_V2_EXTRA.unpack(
-                util.read_exact(f, BOOT_IMG_HDR_V2_EXTRA.size))
-
-            self.dtb_addr = dtb_addr
-
-        if header_version >= 1 and f.tell() != header_size:
-            raise ValueError(f'Invalid header size: {header_size}')
-
-        padding.read_skip(f, page_size)
-
-        if kernel_size > 0:
-            self.kernel = util.read_exact(f, kernel_size)
-            padding.read_skip(f, page_size)
-
-        if ramdisk_size > 0:
-            self.ramdisks.append(util.read_exact(f, ramdisk_size))
-            padding.read_skip(f, page_size)
-
-        if second_size > 0:
-            self.second = util.read_exact(f, second_size)
-            padding.read_skip(f, page_size)
-
-        if header_version >= 1 and recovery_dtbo_size > 0:
-            self.recovery_dtbo = util.read_exact(f, recovery_dtbo_size)
-            padding.read_skip(f, page_size)
-
-        if header_version == 2 and dtb_size > 0:
-            self.dtb = util.read_exact(f, dtb_size)
-            padding.read_skip(f, page_size)
-
-    def generate(self, f: typing.BinaryIO) -> None:
-        if len(self.ramdisks) > 1:
-            raise ValueError('Only one ramdisk is supported')
-        elif self.bootconfig is not None:
-            raise ValueError('Boot config is not supported')
-        elif self.header_version < 1 and self.recovery_dtbo is not None:
-            raise ValueError('Recovery dtbo/acpio is not supported')
-        elif self.header_version < 2 and self.dtb is not None:
-            raise ValueError('Device tree is not supported')
-
-        f.write(BOOT_IMG_HDR_V0.pack(
-            BOOT_MAGIC,
-            len(self.kernel) if self.kernel else 0,
-            self.kernel_addr,
-            len(self.ramdisks[0]) if self.ramdisks else 0,
-            self.ramdisk_addr,
-            len(self.second) if self.second else 0,
-            self.second_addr,
-            self.tags_addr,
-            self.page_size,
-            self.header_version,
-            self.os_version,
-            self.name,
-            self.cmdline,
-            self.id,
-            self.extra_cmdline,
-        ))
-
-        if self.header_version >= 1:
-            header_size = BOOT_IMG_HDR_V0.size
-            if self.header_version >= 1:
-                header_size += BOOT_IMG_HDR_V1_EXTRA.size
-            if self.header_version == 2:
-                header_size += BOOT_IMG_HDR_V2_EXTRA.size
-
-            f.write(BOOT_IMG_HDR_V1_EXTRA.pack(
-                len(self.recovery_dtbo) if self.recovery_dtbo else 0,
-                self.recovery_dtbo_offset,
-                header_size,
-            ))
-
-        if self.header_version == 2:
-            f.write(BOOT_IMG_HDR_V2_EXTRA.pack(
-                len(self.dtb) if self.dtb else 0,
-                self.dtb_addr,
-            ))
-
-        padding.write(f, self.page_size)
-
-        if self.kernel:
-            f.write(self.kernel)
-            padding.write(f, self.page_size)
-
-        if self.ramdisks:
-            f.write(self.ramdisks[0])
-            padding.write(f, self.page_size)
-
-        if self.second:
-            f.write(self.second)
-            padding.write(f, self.page_size)
-
-        if self.header_version >= 1 and self.recovery_dtbo:
-            f.write(self.recovery_dtbo)
-            padding.write(f, self.page_size)
-
-        if self.header_version == 2 and self.dtb:
-            f.write(self.dtb)
-            padding.write(f, self.page_size)
-
-    def __str__(self) -> str:
-        kernel_size = len(self.kernel) if self.kernel else 0
-        ramdisk_size = len(self.ramdisks[0]) if self.ramdisks else 0
-        second_size = len(self.second) if self.second else 0
-
-        result = \
-            f'Boot image v{self.header_version} header:\n' \
-            f'- Kernel size:          {kernel_size}\n' \
-            f'- Kernel address:       0x{self.kernel_addr:x}\n' \
-            f'- Ramdisk size:         {ramdisk_size}\n' \
-            f'- Ramdisk address:      0x{self.ramdisk_addr:x}\n' \
-            f'- Second stage size:    {second_size}\n' \
-            f'- Second stage address: 0x{self.second_addr:x}\n' \
-            f'- Kernel tags address:  0x{self.tags_addr:x}\n' \
-            f'- Page size:            {self.page_size}\n' \
-            f'- OS version:           0x{self.os_version:x}\n' \
-            f'- Name:                 {self.name!r}\n' \
-            f'- Kernel cmdline:       {self.cmdline!r}\n' \
-            f'- ID:                   {self.id.hex()}\n' \
-            f'- Extra kernel cmdline: {self.extra_cmdline!r}\n'
-
-        if self.header_version >= 1:
-            recovery_dtbo_size = len(self.recovery_dtbo) \
-                if self.recovery_dtbo else 0
-
-            result += \
-                f'- Recovery dtbo size:   {recovery_dtbo_size}\n' \
-                f'- Recovery dtbo offset: {self.recovery_dtbo_offset}\n'
-
-        if self.header_version == 2:
-            dtb_size = len(self.dtb) if self.dtb else 0
-
-            result += \
-                f'- Device tree size:     {dtb_size}\n' \
-                f'- Device tree address:  {self.dtb_addr}\n'
-
-        return result
-
-    def _from_dict(self, data: dict[str, typing.Any]) -> None:
-        type = data.get('type')
-        header_version = data.get('header_version')
-
-        if type != 'android':
-            raise WrongFormat(f'Unknown type: {type}')
-        elif header_version not in (0, 1, 2):
-            raise WrongFormat(f'Unknown header version: {header_version}')
-
-        self.header_version = header_version
-        self.kernel_addr = data['kernel_address']
-        self.ramdisk_addr = data['ramdisk_address']
-        self.second_addr = data['second_address']
-        self.tags_addr = data['tags_address']
-        self.page_size = data['page_size']
-        self.os_version = data['os_version']
-        self.name = data['name']
-        self.cmdline = data['cmdline']
-        self.id = data['id']
-        self.extra_cmdline = data['extra_cmdline']
-
-        if header_version >= 1:
-            self.recovery_dtbo_offset = data['recovery_dtbo_offset']
-
-        if self.header_version == 2:
-            self.dtb_addr = data['dtb_address']
-
-    def to_dict(self) -> dict[str, typing.Any]:
-        result = {
-            'type': 'android',
-            'header_version': self.header_version,
-            'kernel_address': self.kernel_addr,
-            'ramdisk_address': self.ramdisk_addr,
-            'second_address': self.second_addr,
-            'tags_address': self.tags_addr,
-            'page_size': self.page_size,
-            'os_version': self.os_version,
-            'name': self.name,
-            'cmdline': self.cmdline,
-            'id': self.id,
-            'extra_cmdline': self.extra_cmdline,
-        }
-
-        if self.header_version >= 1:
-            result['recovery_dtbo_offset'] = self.recovery_dtbo_offset
-
-        if self.header_version == 2:
-            result['dtb_address'] = self.dtb_addr
-
-        return result
-
-
-class _BootImageV3Through4(BootImage):
-    def _from_file(self, f: typing.BinaryIO) -> None:
-        # Common fields for both v3 and v4
-        magic, kernel_size, ramdisk_size, os_version, header_size, reserved, \
-            header_version, cmdline = BOOT_IMG_HDR_V3.unpack(
-                util.read_exact(f, BOOT_IMG_HDR_V3.size))
-
-        if magic != BOOT_MAGIC:
-            raise WrongFormat(f'Unknown magic: {magic}')
-        elif header_version not in (3, 4):
-            raise WrongFormat(f'Unknown header version: {header_version}')
-
-        # Parse v4 fields
-        if header_version == 4:
-            signature_size, = BOOT_IMG_HDR_V4_EXTRA.unpack(
-                util.read_exact(f, BOOT_IMG_HDR_V4_EXTRA.size))
-
-        if f.tell() != header_size:
-            raise ValueError(f'Invalid header size: {header_size}')
-
-        self.header_version = header_version
-        self.os_version = os_version
-        self.reserved = reserved
-        self.cmdline = cmdline.rstrip(b'\0')
-
-        padding.read_skip(f, PAGE_SIZE)
-
-        if kernel_size > 0:
-            self.kernel = util.read_exact(f, kernel_size)
-            padding.read_skip(f, PAGE_SIZE)
-
-        if ramdisk_size > 0:
-            self.ramdisks.append(util.read_exact(f, ramdisk_size))
-            padding.read_skip(f, PAGE_SIZE)
-
-        if header_version == 4:
-            # Don't preserve the signature. It is only used for VTS tests and
-            # is not relevant for booting
-            f.seek(signature_size, os.SEEK_CUR)
-            padding.read_skip(f, PAGE_SIZE)
-
-    def generate(self, f: typing.BinaryIO) -> None:
-        if len(self.ramdisks) > 1:
-            raise ValueError('Only one ramdisk is supported')
-        elif self.second is not None:
-            raise ValueError('Second stage bootloader is not supported')
-        elif self.recovery_dtbo is not None:
-            raise ValueError('Recovery dtbo/acpio is not supported')
-        elif self.dtb is not None:
-            raise ValueError('Device tree is not supported')
-        elif self.bootconfig is not None:
-            raise ValueError('Boot config is not supported')
-
-        f.write(BOOT_IMG_HDR_V3.pack(
-            BOOT_MAGIC,
-            len(self.kernel) if self.kernel else 0,
-            len(self.ramdisks[0]) if self.ramdisks else 0,
-            self.os_version,
-            BOOT_IMG_HDR_V3.size + (BOOT_IMG_HDR_V4_EXTRA.size
-                                    if self.header_version == 4 else 0),
-            self.reserved,
-            self.header_version,
-            self.cmdline,
-        ))
-
-        if self.header_version == 4:
-            f.write(BOOT_IMG_HDR_V4_EXTRA.pack(
-                # We don't care about the VTS signature
-                0
-            ))
-
-        padding.write(f, PAGE_SIZE)
-
-        if self.kernel:
-            f.write(self.kernel)
-            padding.write(f, PAGE_SIZE)
-
-        if self.ramdisks:
-            f.write(self.ramdisks[0])
-            padding.write(f, PAGE_SIZE)
-
-    def __str__(self) -> str:
-        kernel_size = len(self.kernel) if self.kernel else 0
-        ramdisk_size = len(self.ramdisks[0]) if self.ramdisks else 0
-
-        return \
-            f'Boot image v{self.header_version} header:\n' \
-            f'- Kernel size:    {kernel_size}\n' \
-            f'- Ramdisk size:   {ramdisk_size}\n' \
-            f'- OS version:     0x{self.os_version:x}\n' \
-            f'- Reserved:       {self.reserved.hex()}\n' \
-            f'- Kernel cmdline: {self.cmdline!r}\n'
-
-    def _from_dict(self, data: dict[str, typing.Any]) -> None:
-        type = data.get('type')
-        header_version = data.get('header_version')
-
-        if type != 'android':
-            raise WrongFormat(f'Unknown type: {type}')
-        elif header_version not in (3, 4):
-            raise WrongFormat(f'Unknown header version: {header_version}')
-
-        self.header_version = header_version
-        self.os_version = data['os_version']
-        self.reserved = data['reserved']
-        self.cmdline = data['cmdline']
-
-    def to_dict(self) -> dict[str, typing.Any]:
-        return {
-            'type': 'android',
-            'header_version': self.header_version,
-            'os_version': self.os_version,
-            'reserved': self.reserved,
-            'cmdline': self.cmdline,
-        }
-
-
-_RamdiskMeta = collections.namedtuple(
-    '_RamdiskMeta', ['type', 'name', 'board_id'])
-
-
-class _VendorBootImageV3Through4(BootImage):
-    def _from_file(self, f: typing.BinaryIO) -> None:
-        # Common fields for both v3 and v4
-        magic, header_version, page_size, kernel_addr, ramdisk_addr, \
-            vendor_ramdisk_size, cmdline, tags_addr, name, header_size, \
-            dtb_size, dtb_addr = VENDOR_BOOT_IMG_HDR_V3.unpack(
-                util.read_exact(f, VENDOR_BOOT_IMG_HDR_V3.size))
-
-        if magic != VENDOR_BOOT_MAGIC:
-            raise WrongFormat(f'Unknown magic: {magic}')
-        elif header_version not in (3, 4):
-            raise WrongFormat(f'Unknown header version: {header_version}')
-
-        # Parse v4 fields
-        if header_version == 4:
-            vendor_ramdisk_table_size, vendor_ramdisk_table_entry_num, \
-                vendor_ramdisk_table_entry_size, bootconfig_size = \
-                VENDOR_BOOT_IMG_HDR_V4_EXTRA.unpack(
-                    util.read_exact(f, VENDOR_BOOT_IMG_HDR_V4_EXTRA.size))
-
-            if vendor_ramdisk_table_entry_size != \
-                    VENDOR_RAMDISK_TABLE_ENTRY_V4.size:
-                raise ValueError('Invalid ramdisk table entry size: '
-                                 f'{vendor_ramdisk_table_entry_size}')
-            elif vendor_ramdisk_table_size != vendor_ramdisk_table_entry_num \
-                    * vendor_ramdisk_table_entry_size:
-                raise ValueError('Invalid ramdisk table size: '
-                                 f'{vendor_ramdisk_table_size}')
-
-        if f.tell() != header_size:
-            raise ValueError(f'Invalid header size: {header_size}')
-
-        self.page_size = page_size
-        self.header_version = header_version
-        self.kernel_addr = kernel_addr
-        self.ramdisk_addr = ramdisk_addr
-        self.cmdline = cmdline.rstrip(b'\0')
-        self.tags_addr = tags_addr
-        self.name = name.rstrip(b'\0')
-        self.dtb_addr = dtb_addr
-
-        padding.read_skip(f, page_size)
-
-        vendor_ramdisk_offset = f.tell()
-
-        if header_version == 3:
-            # v3 has one big ramdisk
-            self.ramdisks.append(util.read_exact(f, vendor_ramdisk_size))
-        else:
-            # v4 has multiple ramdisks, processed later
-            f.seek(vendor_ramdisk_size, os.SEEK_CUR)
-
-        padding.read_skip(f, page_size)
-
-        if dtb_size > 0:
-            self.dtb = util.read_exact(f, dtb_size)
-            padding.read_skip(f, page_size)
-
-        if header_version == 4:
-            self.ramdisks_meta = []
-
-            total_ramdisk_size = 0
-
-            for _ in range(0, vendor_ramdisk_table_entry_num):
-                ramdisk_size, ramdisk_offset, ramdisk_type, ramdisk_name, \
-                    board_id = VENDOR_RAMDISK_TABLE_ENTRY_V4.unpack(
-                        util.read_exact(f, VENDOR_RAMDISK_TABLE_ENTRY_V4.size))
-
-                table_offset = f.tell()
-                f.seek(vendor_ramdisk_offset + ramdisk_offset)
-
-                self.ramdisks.append(util.read_exact(f, ramdisk_size))
-                self.ramdisks_meta.append(_RamdiskMeta(
-                    ramdisk_type,
-                    ramdisk_name.rstrip(b'\0'),
-                    board_id,
-                ))
-
-                f.seek(table_offset)
-
-                total_ramdisk_size += ramdisk_size
-
-            if total_ramdisk_size != vendor_ramdisk_size:
-                raise ValueError('Invalid vendor ramdisk size: '
-                                 f'{vendor_ramdisk_size}')
-
-            padding.read_skip(f, page_size)
-
-            if bootconfig_size > 0:
-                self.bootconfig = util.read_exact(f, bootconfig_size)
-                padding.read_skip(f, page_size)
-
-    def generate(self, f: typing.BinaryIO) -> None:
-        if self.header_version == 3:
-            if len(self.ramdisks) > 1:
-                raise ValueError('Only one ramdisk is supported')
-            elif self.bootconfig is not None:
-                raise ValueError('Boot config is not supported')
-        else:
-            if len(self.ramdisks) != len(self.ramdisks_meta):
-                raise ValueError('Mismatched ramdisk and ramdisk_meta')
-
-        if self.second is not None:
-            raise ValueError('Second stage bootloader is not supported')
-        elif self.recovery_dtbo is not None:
-            raise ValueError('Recovery dtbo/acpio is not supported')
-
-        vendor_ramdisk_size = sum(len(r) for r in self.ramdisks)
-
-        f.write(VENDOR_BOOT_IMG_HDR_V3.pack(
-            VENDOR_BOOT_MAGIC,
-            self.header_version,
-            self.page_size,
-            self.kernel_addr,
-            self.ramdisk_addr,
-            vendor_ramdisk_size,
-            self.cmdline,
-            self.tags_addr,
-            self.name,
-            VENDOR_BOOT_IMG_HDR_V3.size + (
-                VENDOR_BOOT_IMG_HDR_V4_EXTRA.size
-                if self.header_version == 4 else 0),
-            len(self.dtb) if self.dtb else 0,
-            self.dtb_addr,
-        ))
-
-        if self.header_version == 4:
-            f.write(VENDOR_BOOT_IMG_HDR_V4_EXTRA.pack(
-                len(self.ramdisks) * VENDOR_RAMDISK_TABLE_ENTRY_V4.size,
-                len(self.ramdisks),
-                VENDOR_RAMDISK_TABLE_ENTRY_V4.size,
-                len(self.bootconfig) if self.bootconfig else 0,
-            ))
-
-        padding.write(f, self.page_size)
-
-        for ramdisk in self.ramdisks:
-            f.write(ramdisk)
-
-        padding.write(f, self.page_size)
-
-        if self.dtb:
-            f.write(self.dtb)
-            padding.write(f, self.page_size)
-
-        if self.header_version == 4:
-            ramdisk_offset = 0
-
-            for ramdisk, meta in zip(self.ramdisks, self.ramdisks_meta):
-                f.write(VENDOR_RAMDISK_TABLE_ENTRY_V4.pack(
-                    len(ramdisk),
-                    ramdisk_offset,
-                    meta.type,
-                    meta.name,
-                    meta.board_id,
-                ))
-
-                ramdisk_offset += len(ramdisk)
-
-            padding.write(f, self.page_size)
-
-            if self.bootconfig:
-                f.write(self.bootconfig)
-                padding.write(f, self.page_size)
-
-    def __str__(self) -> str:
-        dtb_size = len(self.dtb) if self.dtb else 0
-
-        result = \
-            f'Vendor boot image v{self.header_version} header:\n' \
-            f'- Page size:           {self.page_size}\n' \
-            f'- Kernel address:      0x{self.kernel_addr:x}\n'
-
-        if self.header_version == 3:
-            ramdisk_size = len(self.ramdisks[0]) if self.ramdisks else 0
-            result += f'- Ramdisk size:        {ramdisk_size}\n'
-
-        result += \
-            f'- Ramdisk address:     0x{self.ramdisk_addr:x}\n' \
-            f'- Kernel cmdline:      {self.cmdline!r}\n' \
-            f'- Kernel tags address: 0x{self.tags_addr:x}\n' \
-            f'- Name:                {self.name!r}\n' \
-            f'- Device tree size:    {dtb_size}\n' \
-            f'- Device tree address: {self.dtb_addr}\n'
-
-        if self.header_version == 4:
-            for ramdisk, meta in zip(self.ramdisks, self.ramdisks_meta):
-                result += \
-                    '- Ramdisk:\n' \
-                    f'  - Size:     {len(ramdisk)}\n' \
-                    f'  - Type:     {meta.type}\n' \
-                    f'  - Name:     {meta.name}\n' \
-                    f'  - Board ID: {meta.board_id.hex()}\n'
-
-            bootconfig_size = len(self.bootconfig) if self.bootconfig else 0
-
-            result += f'- Bootconfig size:     {bootconfig_size}\n'
-
-        return result
-
-    def _from_dict(self, data: dict[str, typing.Any]) -> None:
-        type = data.get('type')
-        header_version = data.get('header_version')
-
-        if type != 'vendor':
-            raise WrongFormat(f'Unknown type: {type}')
-        elif header_version not in (3, 4):
-            raise WrongFormat(f'Unknown header version: {header_version}')
-
-        self.header_version = header_version
-        self.page_size = data['page_size']
-        self.kernel_addr = data['kernel_address']
-        self.ramdisk_addr = data['ramdisk_address']
-        self.cmdline = data['cmdline']
-        self.tags_addr = data['tags_address']
-        self.name = data['name']
-        self.dtb_addr = data['dtb_address']
-
-        if header_version == 4:
-            self.ramdisks_meta = []
-            for meta in data['ramdisk_meta']:
-                self.ramdisks_meta.append(_RamdiskMeta(
-                    meta['type'],
-                    meta['name'],
-                    meta['board_id'],
-                ))
-
-    def to_dict(self) -> dict[str, typing.Any]:
-        result = {
-            'type': 'vendor',
-            'header_version': self.header_version,
-            'page_size': self.page_size,
-            'kernel_address': self.kernel_addr,
-            'ramdisk_address': self.ramdisk_addr,
-            'cmdline': self.cmdline,
-            'tags_address': self.tags_addr,
-            'name': self.name,
-            'dtb_address': self.dtb_addr,
-        }
-
-        if self.header_version == 4:
-            result['ramdisk_meta'] = []
-            for meta in self.ramdisks_meta:
-                result['ramdisk_meta'].append({
-                    'type': meta.type,
-                    'name': meta.name,
-                    'board_id': meta.board_id,
-                })
-
-        return result
-
-
-def load_autodetect(f: typing.BinaryIO) -> BootImage:
-    for cls in (
-        _BootImageV0Through2,
-        _BootImageV3Through4,
-        _VendorBootImageV3Through4,
-    ):
-        try:
-            f.seek(0)
-            return cls(f=f)
-        except WrongFormat:
-            continue
-
-    raise ValueError('Unknown boot image format')
-
-
-def create_from_dict(data: dict) -> BootImage:
-    for cls in (
-        _BootImageV0Through2,
-        _BootImageV3Through4,
-        _VendorBootImageV3Through4,
-    ):
-        try:
-            return cls(data=data)
-        except WrongFormat:
-            continue
-
-    raise ValueError('Unknown boot image format')

avbroot/formats/compression.py

@@ -1,187 +0,0 @@
-import enum
-import gzip
-import typing
-
-import lz4.block
-
-from .. import util
-
-
-GZIP_MAGIC = b'\x1f\x8b'
-
-
-class Lz4Legacy:
-    MAGIC = b'\x02\x21\x4c\x18'
-    MAX_BLOCK_SIZE = 8 * 1024 * 1024
-
-    def __init__(self, fp: typing.BinaryIO,
-                 mode: typing.Literal['rb', 'wb'] = 'rb'):
-        if mode not in ('rb', 'wb'):
-            raise ValueError(f'Invalid mode: {mode}')
-
-        self.fp = fp
-        self.mode = mode
-
-        if mode == 'rb':
-            magic = util.read_exact(self.fp, len(self.MAGIC))
-            if magic != self.MAGIC:
-                raise ValueError(f'Invalid magic: {magic!r}')
-
-            self.rblock = b''
-            self.rblock_offset = 0
-        else:
-            self.fp.write(self.MAGIC)
-
-            self.wblock = bytearray()
-
-        self.file_offset = 0
-
-    def __enter__(self) -> 'Lz4Legacy':
-        return self
-
-    def __exit__(self, *exc_args) -> None:
-        self.close()
-
-    def _read_block(self) -> None:
-        if self.rblock_offset < len(self.rblock):
-            # Haven't finished reading block yet
-            return
-
-        size_raw = self.fp.read(4)
-        if not size_raw or size_raw == self.MAGIC:
-            self.rblock = b''
-            self.rblock_offset = 0
-            return
-        elif len(size_raw) != 4:
-            raise EOFError('Failed to read block size')
-
-        size_compressed = int.from_bytes(size_raw, 'little')
-
-        compressed = util.read_exact(self.fp, size_compressed)
-        self.rblock = lz4.block.decompress(compressed, self.MAX_BLOCK_SIZE)
-        self.rblock_offset = 0
-
-    def _write_block(self, force=False) -> None:
-        if not force and len(self.wblock) < self.MAX_BLOCK_SIZE:
-            # Block not fully filled yet
-            return
-
-        compressed = lz4.block.compress(
-            self.wblock,
-            mode='high_compression',
-            compression=12,
-            store_size=False,
-        )
-
-        self.fp.write(len(compressed).to_bytes(4, 'little'))
-        self.fp.write(compressed)
-
-        self.wblock.clear()
-
-    def read(self, size=None) -> bytes:
-        assert self.mode == 'rb'
-
-        result = bytearray()
-
-        while size is None or size > 0:
-            self._read_block()
-
-            to_read = len(self.rblock) - self.rblock_offset
-            if to_read == 0:
-                # EOF
-                break
-            elif size is not None:
-                to_read = min(to_read, size)
-
-            result.extend(self.rblock[self.rblock_offset:
-                                      self.rblock_offset + to_read])
-
-            self.rblock_offset += to_read
-            self.file_offset += to_read
-
-            if size is not None:
-                size -= to_read
-
-        return result
-
-    def write(self, data: bytes) -> int:
-        assert self.mode == 'wb'
-
-        offset = 0
-
-        while offset < len(data):
-            self._write_block()
-
-            to_write = min(
-                self.MAX_BLOCK_SIZE - len(self.wblock),
-                len(data) - offset,
-            )
-
-            self.wblock.extend(data[offset:offset + to_write])
-
-            self.file_offset += to_write
-            offset += to_write
-
-        return len(data)
-
-    def flush(self) -> None:
-        assert self.mode == 'wb'
-
-        self._write_block(force=True)
-
-    def close(self) -> None:
-        try:
-            if self.mode == 'wb':
-                self.flush()
-        finally:
-            self.mode = 'closed'
-
-    def tell(self) -> int:
-        return self.file_offset
-
-
-Format = enum.Enum('Format', ['GZIP', 'LZ4_LEGACY'])
-
-
-_MAGIC_TO_FORMAT = {
-    GZIP_MAGIC: Format.GZIP,
-    Lz4Legacy.MAGIC: Format.LZ4_LEGACY,
-}
-_MAGIC_MAX_SIZE = max(len(m) for m in _MAGIC_TO_FORMAT)
-
-
-class CompressedFile:
-    def __init__(
-        self,
-        fp: typing.BinaryIO,
-        mode: typing.Literal['rb', 'wb'] = 'rb',
-        format: typing.Optional[Format] = None,
-        raw_if_unknown = False,
-    ):
-        if mode == 'rb' and not format:
-            magic = fp.read(_MAGIC_MAX_SIZE)
-            fp.seek(0)
-
-            for m, f in _MAGIC_TO_FORMAT.items():
-                if magic.startswith(m):
-                    format = f
-                    break
-
-        if format == Format.GZIP:
-            format_fp = gzip.GzipFile(fileobj=fp, mode=mode, mtime=0)
-        elif format == Format.LZ4_LEGACY:
-            format_fp = Lz4Legacy(fp, mode)
-        elif raw_if_unknown:
-            format_fp = fp
-        else:
-            raise ValueError('Unknown compression format')
-
-        self.fp = format_fp
-        self.format = format
-
-    def __enter__(self):
-        self.fp.__enter__()
-        return self
-
-    def __exit__(self, *exc_args):
-        self.fp.__exit__(*exc_args)

avbroot/formats/cpio.py

@@ -1,282 +0,0 @@
-# This is a miniature implementation of cpio, originally written for
-# DualBootPatcher, supporting only enough of the file format for messing with
-# boot image ramdisks. Only the "new format" for cpio entries are supported.
-
-import stat
-import typing
-
-from . import padding
-from .. import util
-
-MAGIC_NEW = b'070701'      # new format
-MAGIC_NEW_CRC = b'070702'  # new format w/crc
-
-# Constants from cpio.h
-
-# A header with a filename "TRAILER!!!" indicates the end of the archive.
-CPIO_TRAILER = b'TRAILER!!!'
-
-C_ISCTG = 0o0110000
-
-IO_BLOCK_SIZE = 512
-
-
-def _read_int(f: typing.BinaryIO) -> int:
-    return int(util.read_exact(f, 8), 16)
-
-
-def _write_int(f: typing.BinaryIO, value: int) -> int:
-    if value < 0 or value > 0xffffffff:
-        raise ValueError(f'{value} out of range for 32-bit integer')
-
-    return f.write(b'%08x' % value)
-
-
-class CpioEntryNew:
-    # c_magic     - "070701" for "new" portable format
-    #               "070702" for CRC format
-    # c_ino
-    # c_mode
-    # c_uid
-    # c_gid
-    # c_nlink
-    # c_mtime
-    # c_filesize  - must be 0 for FIFOs and directories
-    # c_dev_maj
-    # c_dev_min
-    # c_rdev_maj  - only valid for chr and blk special files
-    # c_rdev_min  - only valid for chr and blk special files
-    # c_namesize  - count includes terminating NUL in pathname
-    # c_chksum    - 0 for "new" portable format; for CRC format
-    #               the sum of all the bytes in the file
-
-    @staticmethod
-    def new_trailer() -> 'CpioEntryNew':
-        entry = CpioEntryNew()
-        entry.nlink = 1  # Must be 1 for crc format
-        entry.name = CPIO_TRAILER
-
-        return entry
-
-    @staticmethod
-    def new_symlink(link_target: bytes, name: bytes) -> 'CpioEntryNew':
-        if not link_target:
-            raise ValueError('Symlink target is empty')
-        elif not name:
-            raise ValueError('Symlink name is empty')
-
-        entry = CpioEntryNew()
-        entry.mode = stat.S_IFLNK | 0o777
-        entry.nlink = 1
-        entry.name = name
-        entry.content = link_target
-
-        return entry
-
-    @staticmethod
-    def new_directory(name: bytes, perms: int = 0o755) -> 'CpioEntryNew':
-        if not name:
-            raise ValueError('Directory name is empty')
-
-        entry = CpioEntryNew()
-        entry.mode = stat.S_IFDIR | stat.S_IMODE(perms)
-        entry.nlink = 1
-        entry.name = name
-
-        return entry
-
-    @staticmethod
-    def new_file(name: bytes, perms: int = 0o644,
-                 data: bytes = b'') -> 'CpioEntryNew':
-        if not name:
-            raise ValueError('File name is empty')
-
-        entry = CpioEntryNew()
-        entry.mode = stat.S_IFREG | stat.S_IMODE(perms)
-        entry.nlink = 1
-        entry.name = name
-        entry.content = data
-
-        return entry
-
-    def __init__(self, f: typing.Optional[typing.BinaryIO] = None) -> None:
-        super(CpioEntryNew, self).__init__()
-
-        if f is None:
-            self.magic = MAGIC_NEW
-            self.ino = 0
-            self.mode = 0
-            self.uid = 0
-            self.gid = 0
-            self.nlink = 0
-            self.mtime = 0
-            self.filesize = 0
-            self.dev_maj = 0
-            self.dev_min = 0
-            self.rdev_maj = 0
-            self.rdev_min = 0
-            self.namesize = 0
-            self.chksum = 0
-
-            self._name = b''
-            self._content = b''
-        else:
-            self.magic = util.read_exact(f, 6)
-            if self.magic != MAGIC_NEW and self.magic != MAGIC_NEW_CRC:
-                raise Exception(f'Unknown magic: {self.magic!r}')
-
-            self.ino = _read_int(f)
-            self.mode = _read_int(f)
-            self.uid = _read_int(f)
-            self.gid = _read_int(f)
-            self.nlink = _read_int(f)
-            self.mtime = _read_int(f)
-            self.filesize = _read_int(f)
-            self.dev_maj = _read_int(f)
-            self.dev_min = _read_int(f)
-            self.rdev_maj = _read_int(f)
-            self.rdev_min = _read_int(f)
-            self.namesize = _read_int(f)
-            self.chksum = _read_int(f)
-
-            # Filename
-            self._name = util.read_exact(f, self.namesize - 1)
-            # Discard NULL terminator
-            util.read_exact(f, 1)
-            padding.read_skip(f, 4)
-
-            # File contents
-            self._content = util.read_exact(f, self.filesize)
-            padding.read_skip(f, 4)
-
-    def write(self, f: typing.BinaryIO):
-        if len(self.magic) != 6:
-            raise ValueError(f'Magic is not 6 bytes: {self.magic!r}')
-
-        f.write(self.magic)
-
-        _write_int(f, self.ino)
-        _write_int(f, self.mode)
-        _write_int(f, self.uid)
-        _write_int(f, self.gid)
-        _write_int(f, self.nlink)
-        _write_int(f, self.mtime)
-        _write_int(f, self.filesize)
-        _write_int(f, self.dev_maj)
-        _write_int(f, self.dev_min)
-        _write_int(f, self.rdev_maj)
-        _write_int(f, self.rdev_min)
-        _write_int(f, self.namesize)
-        _write_int(f, self.chksum)
-
-        # Filename
-        f.write(self._name)
-        f.write(b'\x00')
-        padding.write(f, 4)
-
-        # File contents
-        f.write(self._content)
-        padding.write(f, 4)
-
-    @property
-    def name(self) -> bytes:
-        return self._name
-
-    @name.setter
-    def name(self, value: bytes):
-        self._name = value
-        self.namesize = len(value) + 1
-
-    @property
-    def content(self) -> bytes:
-        return self._content
-
-    @content.setter
-    def content(self, value: bytes):
-        self._content = value
-        self.filesize = len(value)
-
-    def __str__(self) -> str:
-        filetype = stat.S_IFMT(self.mode)
-
-        if stat.S_ISDIR(self.mode):
-            ftypestr = 'directory'
-        elif stat.S_ISLNK(self.mode):
-            ftypestr = 'symbolic link'
-        elif stat.S_ISREG(self.mode):
-            ftypestr = 'regular file'
-        elif stat.S_ISFIFO(self.mode):
-            ftypestr = 'pipe'
-        elif stat.S_ISCHR(self.mode):
-            ftypestr = 'character device'
-        elif stat.S_ISBLK(self.mode):
-            ftypestr = 'block device'
-        elif stat.S_ISSOCK(self.mode):
-            ftypestr = 'socket'
-        elif filetype == C_ISCTG:
-            ftypestr = 'reserved'
-        else:
-            ftypestr = 'unknown (%o)' % filetype
-
-        return \
-            f'Filename:        {self.name!r}\n' \
-            f'Filetype:        {ftypestr}\n' \
-            f'Magic:           {self.magic!r}\n' \
-            f'Inode:           {self.ino}\n' \
-            f'Mode:            {self.mode:o}\n' \
-            f'Permissions:     {self.mode - filetype:o}\n' \
-            f'UID:             {self.uid}\n' \
-            f'GID:             {self.gid}\n' \
-            f'Links:           {self.nlink}\n' \
-            f'Modified:        {self.mtime}\n' \
-            f'File size:       {self.filesize}\n' \
-            f'Device:          {self.dev_maj:x},{self.dev_min:x}\n' \
-            f'Device ID:       {self.rdev_maj:x},{self.rdev_min:x}\n' \
-            f'Filename length: {self.namesize}\n' \
-            f'Checksum:        {self.chksum:x}\n'
-
-
-def load(f: typing.BinaryIO, include_trailer: bool = False,
-         reassign_inodes: bool = True) -> list[CpioEntryNew]:
-    entries = []
-
-    while True:
-        entry = CpioEntryNew(f)
-
-        if stat.S_IFMT(entry.mode) != stat.S_IFDIR and entry.nlink > 1:
-            raise ValueError(f'Hard links are not supported: {entry.name!r}')
-
-        # Inodes are reassigned on save
-        if reassign_inodes:
-            entry.ino = 0
-
-        if entry.name == CPIO_TRAILER:
-            if include_trailer:
-                entries.append(entry)
-            break
-
-        entries.append(entry)
-
-    return entries
-
-
-def save(f: typing.BinaryIO, entries: list[CpioEntryNew], sort=True,
-         pad_to_block_size=False):
-    inode = 300000
-
-    if sort:
-        entries = sorted(entries, key=lambda e: e.name)
-
-    for entry in entries:
-        entry.ino = inode
-        inode += 1
-
-        entry.write(f)
-
-    trailer = CpioEntryNew.new_trailer()
-    trailer.ino = inode
-    trailer.write(f)
-
-    # Pad until end of block
-    if pad_to_block_size:
-        padding.write(f, IO_BLOCK_SIZE)

avbroot/formats/padding.py

@@ -1,47 +0,0 @@
-import os
-import typing
-
-
-def _is_power_of_2(n: int) -> bool:
-    if hasattr(n, 'bit_count'):
-        return n.bit_count() == 1
-    else:
-        return bin(n).count('1') == 1
-
-
-def calc(offset: int, page_size: int) -> int:
-    '''
-    Calculate the amount of padding that needs to be added to align the
-    specified offset to a page boundary. The page size must be a power of 2.
-    '''
-
-    if not _is_power_of_2(page_size):
-        raise ValueError(f'{page_size} is not a power of 2')
-
-    return (page_size - (offset & (page_size - 1))) & (page_size - 1)
-
-
-def read_skip(f: typing.BinaryIO, page_size: int) -> int:
-    '''
-    Seek file to the next page boundary if it is not already at a page
-    boundary. If the file does not support seeking, then data is read and
-    discarded.
-    '''
-
-    padding = calc(f.tell(), page_size)
-
-    if hasattr(f, 'seek'):
-        f.seek(padding, os.SEEK_CUR)
-    else:
-        f.read(padding)
-
-    return padding
-
-
-def write(f: typing.BinaryIO, page_size: int) -> int:
-    '''
-    Write null bytes to pad the file to the next page boundary if it is not
-    already at a page boundary.
-    '''
-
-    return f.write(calc(f.tell(), page_size) * b'\x00')

avbroot/main.py

@@ -1,731 +0,0 @@
-import argparse
-import concurrent.futures
-import contextlib
-import copy
-import dataclasses
-import graphlib
-import io
-import os
-import shutil
-import struct
-import tempfile
-import time
-import typing
-import unittest.mock
-import zipfile
-
-import avbtool
-
-from . import boot
-from . import openssl
-from . import ota
-from . import util
-from . import vbmeta
-from .formats import bootimage
-from .formats import compression
-from .formats import cpio
-
-
-PATH_METADATA = 'META-INF/com/android/metadata'
-PATH_METADATA_PB = f'{PATH_METADATA}.pb'
-PATH_OTACERT = 'META-INF/com/android/otacert'
-PATH_PAYLOAD = 'payload.bin'
-PATH_PROPERTIES = 'payload_properties.txt'
-
-PARTITION_PRIORITIES = {
-    # The kernel is always in boot
-    '@gki_kernel': ('boot',),
-    # Devices launching with Android 13 use a GKI init_boot ramdisk
-    '@gki_ramdisk': ('init_boot', 'boot'),
-    # OnePlus devices have a recovery image
-    '@otacerts': ('recovery', 'vendor_boot', 'boot'),
-}
-
-
-@dataclasses.dataclass
-class PatchContext:
-    replace_images: dict[str, os.PathLike[str]]
-    boot_partition: str
-    root_patch: typing.Optional[boot.BootImagePatch]
-    clear_vbmeta_flags: bool
-    privkey_avb: os.PathLike[str]
-    passphrase_avb: str
-    privkey_ota: os.PathLike[str]
-    passphrase_ota: str
-    cert_ota: os.PathLike[str]
-
-
-def print_status(*args, **kwargs):
-    print('\x1b[1m*****', *args, '*****\x1b[0m', **kwargs)
-
-
-def print_warning(*args, **kwargs):
-    print('\x1b[1;31m*****', '[WARNING]', *args, '*****\x1b[0m', **kwargs)
-
-
-def get_partitions_by_type(manifest):
-    all_partitions = set(p.partition_name for p in manifest.partitions)
-    by_type = {}
-
-    for t, candidates in PARTITION_PRIORITIES.items():
-        partition = next((p for p in candidates if p in all_partitions), None)
-        if partition is None:
-            raise ValueError(f'Cannot find partition of type: {t}')
-
-        by_type[t] = partition
-
-    for partition in all_partitions:
-        if 'vbmeta' in partition:
-            by_type[f'@vbmeta:{partition}'] = partition
-
-    return by_type
-
-
-def get_required_images(manifest, boot_partition, with_root):
-    all_partitions = set(p.partition_name for p in manifest.partitions)
-    by_type = get_partitions_by_type(manifest)
-    images = {k: v for k, v in by_type.items()
-              if k == '@otacerts' or k.startswith('@vbmeta:')}
-
-    if with_root:
-        if boot_partition in by_type:
-            images['@rootpatch'] = by_type[boot_partition]
-        elif boot_partition in all_partitions:
-            images['@rootpatch'] = boot_partition
-        else:
-            raise ValueError(f'Boot partition not found: {boot_partition}')
-
-    return images
-
-
-def get_vbmeta_patch_order(avb, image_paths, vbmeta_images):
-    dep_graph = vbmeta.get_vbmeta_deps(
-        avb, {n: image_paths[n] for n in vbmeta_images})
-
-    # Only keep dependencies among the subset of images we're working with
-    dep_graph = {n: {d for d in deps if d in image_paths}
-                 for n, deps in dep_graph.items() if n in image_paths}
-
-    # Avoid patching vbmeta images that don't need changes
-    while True:
-        unneeded_vbmeta = set(n for n, d in dep_graph.items()
-                              if n in vbmeta_images and not d)
-        if not unneeded_vbmeta:
-            break
-
-        dep_graph = {n: {d for d in deps if d not in unneeded_vbmeta}
-                     for n, deps in dep_graph.items()
-                     if n not in unneeded_vbmeta}
-
-    full_order = graphlib.TopologicalSorter(dep_graph).static_order()
-    order = [n for n in full_order if n in vbmeta_images]
-
-    return dep_graph, order
-
-
-def patch_ota_payload(f_in, open_more_f_in, f_out, file_size,
-                      context: PatchContext):
-    with tempfile.TemporaryDirectory() as temp_dir:
-        extract_dir = os.path.join(temp_dir, 'extract')
-        patch_dir = os.path.join(temp_dir, 'patch')
-        payload_dir = os.path.join(temp_dir, 'payload')
-        os.mkdir(extract_dir)
-        os.mkdir(patch_dir)
-        os.mkdir(payload_dir)
-
-        version, manifest, blob_offset = ota.parse_payload(f_in)
-        all_partitions = set(p.partition_name for p in manifest.partitions)
-        image_paths = {}
-
-        # Use user-provided partition images if provided. This may be a larger
-        # set than what's needed for our patches.
-        for name, path in context.replace_images.items():
-            if name not in all_partitions:
-                raise ValueError(
-                    f'Cannot replace non-existent partition: {name}')
-
-            image_paths[name] = path
-
-        # Extract remaining required partition images from the original payload.
-        required_images = get_required_images(manifest, context.boot_partition,
-                                              context.root_patch is not None)
-        vbmeta_images = set(p for n, p in required_images.items()
-                            if n.startswith('@vbmeta:'))
-
-        to_extract = required_images.values() - image_paths.keys()
-        for name in to_extract:
-            image_paths[name] = os.path.join(extract_dir, f'{name}.img')
-
-        if to_extract:
-            print_status('Extracting', ', '.join(sorted(to_extract)),
-                         'from the payload')
-            ota.extract_images(open_more_f_in, manifest, blob_offset,
-                               extract_dir, to_extract)
-
-        image_patches = {}
-        if context.root_patch is not None:
-            image_patches.setdefault(required_images['@rootpatch'], []).append(
-                context.root_patch)
-        image_patches.setdefault(required_images['@otacerts'], []).append(
-            boot.OtaCertPatch(context.cert_ota))
-
-        avb = avbtool.Avb()
-
-        print_status('Patching', ', '.join(sorted(image_patches)))
-        with concurrent.futures.ThreadPoolExecutor(
-                max_workers=len(image_patches)) as executor:
-            def apply_patches(image, patches):
-                patched_path = os.path.join(patch_dir, f'{image}.img')
-
-                boot.patch_boot(
-                    avb,
-                    image_paths[image],
-                    patched_path,
-                    context.privkey_avb,
-                    context.passphrase_avb,
-                    True,
-                    patches,
-                )
-
-                image_paths[image] = patched_path
-
-            futures = [executor.submit(apply_patches, i, p)
-                       for i, p in image_patches.items()]
-
-            for future in concurrent.futures.as_completed(futures):
-                future.result()
-
-        vbmeta_deps, vbmeta_order = \
-            get_vbmeta_patch_order(avb, image_paths, vbmeta_images)
-        print_status('Building', ', '.join(vbmeta_order))
-
-        for image in vbmeta_order:
-            patched_path = os.path.join(patch_dir, f'{image}.img')
-
-            vbmeta.patch_vbmeta_image(
-                avb,
-                {n: p for n, p in image_paths.items()
-                 if n in vbmeta_deps[image]},
-                image_paths[image],
-                patched_path,
-                context.privkey_avb,
-                context.passphrase_avb,
-                manifest.block_size,
-                context.clear_vbmeta_flags,
-            )
-
-            image_paths[image] = patched_path
-
-        # Don't replace untouched vbmeta images
-        for image in vbmeta_images - set(vbmeta_order):
-            del image_paths[image]
-
-        print_status('Updating OTA payload to reference new',
-                     ', '.join(sorted(image_paths)))
-        return ota.patch_payload(
-            f_in,
-            f_out,
-            version,
-            manifest,
-            blob_offset,
-            payload_dir,
-            image_paths,
-            file_size,
-            context.privkey_ota,
-            context.passphrase_ota,
-        )
-
-
-def strip_bad_extra_fields(extra):
-    offset = 0
-    new_extra = bytearray()
-
-    while offset < len(extra):
-        record_sig, record_len = \
-            struct.unpack('<HH', extra[offset:offset + 4])
-
-        next_offset = offset + 4 + record_len
-
-        # 0xd935: ALIGNMENT_ZIP_EXTRA_DATA_FIELD_HEADER_ID
-        # 0x0001: zip64 size (zipfile will write a new record)
-        if record_sig not in (0x0001, 0xd935):
-            new_extra.extend(extra[offset:next_offset])
-
-        offset = next_offset
-
-    return new_extra
-
-
-@contextlib.contextmanager
-def fix_streaming_local_header_sizes():
-    '''
-    Older Python versions don't set the local header's two 32-bit size fields to
-    0xffffffff when writing a zip64 entry to an unseekable file. This function
-    monkey patches zipfile's local file header serialization to manually fix
-    this issue.
-    '''
-
-    orig = zipfile.ZipInfo.FileHeader
-
-    def wrapper(*args, **kwargs):
-        blob = orig(*args, **kwargs)
-        zip64 = kwargs.get('zip64')
-        if zip64 is None:
-            zip64 = args[0].file_size > zipfile.ZIP64_LIMIT or \
-                args[0].compress_size > zipfile.ZIP64_LIMIT
-
-        fields = list(struct.unpack_from(zipfile.structFileHeader, blob))
-        if fields[3] & (1 << 3) and zip64:
-            fields[8] = 0xffffffff
-            fields[9] = 0xffffffff
-
-            return struct.pack(zipfile.structFileHeader, *fields) + \
-                blob[zipfile.sizeFileHeader:]
-        else:
-            return blob
-
-    with unittest.mock.patch('zipfile.ZipInfo.FileHeader', wrapper):
-        yield
-
-
-def patch_ota_zip(f_zip_in, f_zip_out, context: PatchContext):
-    with (
-        zipfile.ZipFile(f_zip_in, 'r') as z_in,
-        zipfile.ZipFile(f_zip_out, 'w') as z_out,
-    ):
-        infolist = z_in.infolist()
-        missing = {
-            PATH_METADATA,
-            PATH_METADATA_PB,
-            PATH_OTACERT,
-            PATH_PAYLOAD,
-            PATH_PROPERTIES,
-        }
-        i_payload = -1
-        i_properties = -1
-
-        for i, info in enumerate(infolist):
-            if info.filename in missing:
-                missing.remove(info.filename)
-
-            if info.filename == PATH_PAYLOAD:
-                i_payload = i
-            elif info.filename == PATH_PROPERTIES:
-                i_properties = i
-
-            if not missing and i_payload >= 0 and i_properties >= 0:
-                break
-
-        if missing:
-            raise Exception(f'Missing files in zip: {missing}')
-
-        # Ensure payload is processed before properties
-        if i_payload > i_properties:
-            infolist[i_payload], infolist[i_properties] = \
-                infolist[i_properties], infolist[i_payload]
-
-        properties = None
-        metadata_info = None
-        metadata_pb_info = None
-        metadata_pb_raw = None
-
-        for info in infolist:
-            out_info = copy.copy(info)
-            out_info.extra = strip_bad_extra_fields(out_info.extra)
-
-            # Ignore because the plain-text legacy metadata file is regenerated
-            # from the new metadata
-            if info.filename == PATH_METADATA:
-                metadata_info = out_info
-                continue
-
-            # The existing metadata is needed to generate a new signed zip
-            elif info.filename == PATH_METADATA_PB:
-                metadata_pb_info = out_info
-
-                with z_in.open(info, 'r') as f_in:
-                    metadata_pb_raw = f_in.read()
-
-                continue
-
-            # Use the user's OTA certificate
-            elif info.filename == PATH_OTACERT:
-                print_status('Replacing', info.filename)
-
-                with (
-                    open(context.cert_ota, 'rb') as f_cert,
-                    z_out.open(out_info, 'w') as f_out,
-                ):
-                    shutil.copyfileobj(f_cert, f_out)
-
-                continue
-
-            # Copy other files, patching if needed
-            with (
-                z_in.open(info, 'r') as f_in,
-                z_out.open(out_info, 'w') as f_out,
-            ):
-                if info.filename == PATH_PAYLOAD:
-                    print_status('Patching', info.filename)
-
-                    if info.compress_type != zipfile.ZIP_STORED:
-                        raise Exception(
-                            f'{info.filename} is not stored uncompressed')
-
-                    properties = patch_ota_payload(
-                        f_in,
-                        lambda: z_in.open(info, 'r'),
-                        f_out,
-                        info.file_size,
-                        context,
-                    )
-
-                elif info.filename == PATH_PROPERTIES:
-                    print_status('Patching', info.filename)
-
-                    if info.compress_type != zipfile.ZIP_STORED:
-                        raise Exception(
-                            f'{info.filename} is not stored uncompressed')
-
-                    f_out.write(properties)
-
-                else:
-                    print_status('Copying', info.filename)
-
-                    shutil.copyfileobj(f_in, f_out)
-
-        print_status('Generating', PATH_METADATA, 'and', PATH_METADATA_PB)
-        metadata = ota.add_metadata(
-            z_out,
-            metadata_info,
-            metadata_pb_info,
-            metadata_pb_raw,
-        )
-
-        # Signing process needs to capture the zip central directory
-        f_zip_out.start_capture()
-
-        return metadata
-
-
-def patch_subcommand(args):
-    output = args.output
-    if output is None:
-        output = args.input + '.patched'
-
-    if args.rootless:
-        root_patch = None
-    elif args.magisk is not None:
-        root_patch = boot.MagiskRootPatch(
-            args.magisk, args.magisk_preinit_device, args.magisk_random_seed)
-
-        try:
-            root_patch.validate()
-        except ValueError as e:
-            if args.ignore_magisk_warnings:
-                print_warning(e)
-            else:
-                raise e
-    else:
-        root_patch = boot.PrepatchedImage(
-            args.prepatched,
-            args.ignore_prepatched_compat + 1,
-            print_warning,
-        )
-
-    # Get passphrases for keys
-    passphrase_avb = openssl.prompt_passphrase(
-        args.privkey_avb,
-        args.passphrase_avb_env_var,
-        args.passphrase_avb_file,
-    )
-    passphrase_ota = openssl.prompt_passphrase(
-        args.privkey_ota,
-        args.passphrase_ota_env_var,
-        args.passphrase_ota_file,
-    )
-
-    # Ensure that the certificate matches the private key
-    if not openssl.cert_matches_key(args.cert_ota, args.privkey_ota,
-                                    passphrase_ota):
-        raise Exception('OTA certificate does not match private key')
-
-    start = time.perf_counter_ns()
-
-    with util.open_output_file(output) as temp_raw:
-        with (
-            ota.open_signing_wrapper(temp_raw, args.privkey_ota,
-                                     passphrase_ota, args.cert_ota) as temp,
-            ota.match_android_zip64_limit(),
-            fix_streaming_local_header_sizes(),
-        ):
-            context = PatchContext(
-                replace_images=args.replace or {},
-                boot_partition=args.boot_partition,
-                root_patch=root_patch,
-                clear_vbmeta_flags=args.clear_vbmeta_flags,
-                privkey_avb=args.privkey_avb,
-                passphrase_avb=passphrase_avb,
-                privkey_ota=args.privkey_ota,
-                passphrase_ota=passphrase_ota,
-                cert_ota=args.cert_ota,
-            )
-
-            metadata = patch_ota_zip(args.input, temp, context)
-
-        # We do a lot of low-level hackery. Reopen and verify offsets
-        print_status('Verifying metadata offsets')
-        with zipfile.ZipFile(temp_raw, 'r') as z:
-            ota.verify_metadata(z, metadata)
-
-    # Excluding the time it takes for the user to type in the passwords
-    elapsed = time.perf_counter_ns() - start
-    print_status(f'Completed after {elapsed / 1_000_000_000:.1f}s')
-
-
-def extract_subcommand(args):
-    with zipfile.ZipFile(args.input, 'r') as z:
-        info = z.getinfo(PATH_PAYLOAD)
-
-        with z.open(info, 'r') as f:
-            _, manifest, blob_offset = ota.parse_payload(f)
-
-        if args.all:
-            unique_images = set(p.partition_name
-                                for p in manifest.partitions)
-        else:
-            images = get_required_images(manifest, args.boot_partition, True)
-            if args.boot_only:
-                unique_images = {images['@rootpatch']}
-            else:
-                unique_images = set(images.values())
-
-        print_status('Extracting', ', '.join(sorted(unique_images)),
-                     'from the payload')
-        os.makedirs(args.directory, exist_ok=True)
-
-        # Extract in parallel. There's is no actual I/O parallelism due to
-        # zipfile's internal locks, but this is still significantly faster than
-        # doing it single threaded. The extraction process is mostly CPU board
-        # due to decompression.
-        ota.extract_images(lambda: z.open(info, 'r'),
-                           manifest, blob_offset, args.directory,
-                           unique_images)
-
-
-def magisk_info_subcommand(args):
-    with open(args.image, 'rb') as f:
-        img = bootimage.load_autodetect(f)
-
-    if not img.ramdisks:
-        raise ValueError('Boot image does not have a ramdisk')
-
-    with (
-        io.BytesIO(img.ramdisks[0]) as f_raw,
-        compression.CompressedFile(f_raw, 'rb', raw_if_unknown=True) as f,
-    ):
-        entries = cpio.load(f.fp)
-        config = next((e for e in entries if e.name == b'.backup/.magisk'),
-                      None)
-        if config is None:
-            raise ValueError('Not a Magisk-patched boot image')
-
-        print(config.content.decode('ascii'), end='')
-
-
-def uint64_arg(arg):
-    value = int(arg)
-    if value < 0 or value >= 2 ** 64:
-        raise ValueError('Out of range for unsigned 64-bit integer')
-
-    return value
-
-
-class KeyValuePairAction(argparse.Action):
-    def __init__(self, option_strings, dest, nargs=None, **kwargs):
-        if nargs != 2:
-            raise ValueError('nargs must be 2')
-
-        super().__init__(option_strings, dest, nargs=nargs, **kwargs)
-
-    def __call__(self, parser, namespace, values, option_string=None):
-        data = getattr(namespace, self.dest, None)
-        if data is None:
-            data = {}
-
-        data[values[0]] = values[1]
-        setattr(namespace, self.dest, data)
-
-
-def parse_args(argv=None):
-    parser = argparse.ArgumentParser()
-    subparsers = parser.add_subparsers(
-        dest='subcommand',
-        required=True,
-        help='Subcommands',
-    )
-
-    patch = subparsers.add_parser(
-        'patch',
-        help='Patch a full OTA zip',
-    )
-
-    patch.add_argument(
-        '--input',
-        required=True,
-        help='Path to original raw payload or OTA zip',
-    )
-    patch.add_argument(
-        '--output',
-        help='Path to new raw payload or OTA zip',
-    )
-    patch.add_argument(
-        '--privkey-avb',
-        required=True,
-        help='Private key for signing root vbmeta image',
-    )
-    patch.add_argument(
-        '--privkey-ota',
-        required=True,
-        help='Private key for signing OTA payload',
-    )
-    patch.add_argument(
-        '--cert-ota',
-        required=True,
-        help='Certificate for OTA payload signing key',
-    )
-
-    for arg in ('AVB', 'OTA'):
-        group = patch.add_mutually_exclusive_group()
-        group.add_argument(
-            f'--passphrase-{arg.lower()}-env-var',
-            help=f'Environment variable containing {arg} private key passphrase',
-        )
-        group.add_argument(
-            f'--passphrase-{arg.lower()}-file',
-            help=f'File containing {arg} private key passphrase',
-        )
-
-    patch.add_argument(
-        '--replace',
-        nargs=2,
-        action=KeyValuePairAction,
-        help='Use partition image from a file instead of the original payload',
-    )
-
-    boot_group = patch.add_mutually_exclusive_group(required=True)
-    boot_group.add_argument(
-        '--magisk',
-        help='Path to Magisk APK',
-    )
-    boot_group.add_argument(
-        '--prepatched',
-        help='Path to prepatched boot image',
-    )
-    boot_group.add_argument(
-        '--rootless',
-        action='store_true',
-        help='Skip applying root patch',
-    )
-
-    patch.add_argument(
-        '--magisk-preinit-device',
-        help='Magisk preinit device',
-    )
-    patch.add_argument(
-        '--magisk-random-seed',
-        type=uint64_arg,
-        help='Magisk random seed',
-    )
-    patch.add_argument(
-        '--ignore-magisk-warnings',
-        action='store_true',
-        help='Ignore Magisk compatibility/version warnings',
-    )
-    patch.add_argument(
-        '--ignore-prepatched-compat',
-        default=0,
-        action='count',
-        help='Ignore compatibility issues with prepatched boot images',
-    )
-
-    patch.add_argument(
-        '--clear-vbmeta-flags',
-        action='store_true',
-        help='Forcibly clear vbmeta flags if they disable AVB',
-    )
-
-    extract = subparsers.add_parser(
-        'extract',
-        help='Extract patched images from a patched OTA zip',
-    )
-
-    extract.add_argument(
-        '--input',
-        required=True,
-        help='Path to patched OTA zip',
-    )
-    extract.add_argument(
-        '--directory',
-        default='.',
-        help='Output directory for extracted images',
-    )
-    extract_group = extract.add_mutually_exclusive_group()
-    extract_group.add_argument(
-        '--all',
-        action='store_true',
-        help='Extract all images from the payload',
-    )
-    extract_group.add_argument(
-        '--boot-only',
-        action='store_true',
-        help='Extract only the boot image',
-    )
-
-    for subcmd in (patch, extract):
-        subcmd.add_argument(
-            '--boot-partition',
-            default='@gki_ramdisk',
-            help='Boot partition name',
-        )
-
-    magisk_info = subparsers.add_parser(
-        'magisk-info',
-        help='Print Magisk config from a patched boot image',
-    )
-    magisk_info.add_argument(
-        '--image',
-        required=True,
-        help='Patch to Magisk-patched boot image',
-    )
-
-    args = parser.parse_args(args=argv)
-
-    if args.subcommand == 'patch':
-        if args.magisk is None:
-            if args.magisk_preinit_device:
-                parser.error('--magisk-preinit-device requires --magisk')
-            elif args.magisk_random_seed:
-                parser.error('--magisk-random-seed requires --magisk')
-            elif args.ignore_magisk_warnings:
-                parser.error('--ignore-magisk-warnings requires --magisk')
-        elif args.prepatched is None:
-            if args.ignore_prepatched_compat:
-                parser.error('--ignore-prepatched-compat requires --prepatched')
-
-    return args
-
-
-def main(argv=None):
-    args = parse_args(argv=argv)
-
-    util.load_umask_unsafe()
-
-    if args.subcommand == 'patch':
-        patch_subcommand(args)
-    elif args.subcommand == 'extract':
-        extract_subcommand(args)
-    elif args.subcommand == 'magisk-info':
-        magisk_info_subcommand(args)
-    else:
-        raise NotImplementedError()

avbroot/openssl.py

@@ -1,222 +0,0 @@
-import binascii
-import contextlib
-import getpass
-import os
-import random
-import string
-import subprocess
-import unittest.mock
-
-# This module calls the openssl binary because AOSP's avbtool.py already does
-# that and the operations are simple enough to not require pulling in a
-# library.
-
-
-@contextlib.contextmanager
-def _passphrase_fd(passphrase):
-    '''
-    If the specified passphrase is not None, yield the readable end of a pipe
-    that produces the passphrase encoded as UTF-8, followed by a newline. The
-    read end of the pipe is marked as inheritable. Both ends of the pipe are
-    closed after leaving the context.
-    '''
-
-    assert os.name != 'nt'
-
-    if passphrase is None:
-        yield None
-        return
-
-    # For simplicity, we don't write to the pipe on a thread, so pick a maximum
-    # length that doesn't exceed any OS's pipe buffer size, while still being
-    # usable for just about every use case.
-    if len(passphrase) >= 4096:
-        raise ValueError('Passphrase is too long')
-
-    pipe_r, pipe_w = os.pipe()
-    write_closed = False
-
-    try:
-        os.set_inheritable(pipe_r, True)
-
-        os.write(pipe_w, passphrase.encode('UTF-8'))
-        os.write(pipe_w, b'\n')
-        os.close(pipe_w)
-        write_closed = True
-
-        yield pipe_r
-    finally:
-        os.close(pipe_r)
-        if not write_closed:
-            os.close(pipe_w)
-
-
-class _PopenPassphraseWrapper:
-    '''
-    Wrapper around subprocess.Popen() that adds arguments for passing in the
-    private key passphrase via a pipe on non-Windows systems. On Windows,
-    openssl does not support reading from pipes, so the passphrase is passed in
-    via an environment variable.
-    '''
-
-    def __init__(self, passphrase):
-        self.orig_popen = subprocess.Popen
-        self.passphrase = passphrase
-
-    def __call__(self, cmd, *args, **kwargs):
-        if self.passphrase is not None and cmd and \
-                os.path.basename(cmd[0]) == 'openssl':
-            if os.name == 'nt':
-                # On Windows, opensssl does not support reading the passphrase
-                # from a file descriptor. An environment variable is the next
-                # best way to handle this.
-                if 'env' not in kwargs:
-                    kwargs['env'] = dict(os.environ)
-
-                env_var = ''.join(random.choices(string.ascii_letters, k=64))
-                kwargs['env'][env_var] = self.passphrase
-
-                new_cmd = [*cmd, '-passin', f'env:{env_var}']
-
-                return self.orig_popen(new_cmd, *args, **kwargs)
-            else:
-                with _passphrase_fd(self.passphrase) as fd:
-                    kwargs['close_fds'] = False
-
-                    new_cmd = [*cmd, '-passin', f'fd:{fd}']
-
-                    return self.orig_popen(new_cmd, *args, **kwargs)
-
-                # The pipe is closed at this point in this process, but the
-                # child already inherited the fd and the passphrase is sitting
-                # the pipe buffer.
-        else:
-            return self.orig_popen(cmd, *args, **kwargs)
-
-
-def inject_passphrase(passphrase):
-    '''
-    While this context is active, patch subprocess calls to openssl so that
-    the passphrase is specified via an injected -passin argument, if it is not
-    None. The passphrase is passed to the command via a pipe file descriptor
-    (non-Windows) or an environment variable (Windows).
-    '''
-
-    return unittest.mock.patch(
-        'subprocess.Popen', side_effect=_PopenPassphraseWrapper(passphrase))
-
-
-def _guess_format(path):
-    '''
-    Simple heuristic to determine the encoding of a key. This is needed because
-    openssl 1.1 doesn't support autodetection.
-    '''
-
-    with open(path, 'rb') as f:
-        for line in f:
-            if line.startswith(b'-----BEGIN '):
-                return 'PEM'
-
-    return 'DER'
-
-
-def _get_modulus(path, passphrase, is_x509):
-    '''
-    Get the RSA modulus of the given file, which can be a private key or
-    certificate.
-    '''
-
-    with inject_passphrase(passphrase):
-        output = subprocess.check_output([
-            'openssl',
-            'x509' if is_x509 else 'rsa',
-            '-in', path,
-            '-inform', _guess_format(path),
-            '-noout',
-            '-modulus',
-        ])
-
-    prefix, delim, suffix = output.strip().partition(b'=')
-    if not delim or prefix != b'Modulus':
-        raise Exception(f'Unexpected modulus output: {repr(output)}')
-
-    return binascii.unhexlify(suffix)
-
-
-def max_signature_size(pkey, passphrase):
-    '''
-    Get the maximum size of a signature signed by the specified RSA key. This
-    is equal to the modulus size.
-    '''
-
-    return len(_get_modulus(pkey, passphrase, False))
-
-
-def sign_data(pkey, passphrase, data):
-    '''
-    Sign <data> with <pkey>.
-    '''
-
-    with inject_passphrase(passphrase):
-        return subprocess.check_output(
-            [
-                'openssl', 'pkeyutl',
-                '-sign',
-                '-inkey', pkey,
-                '-keyform', _guess_format(pkey),
-                '-pkeyopt', 'digest:sha256',
-            ],
-            input=data,
-        )
-
-
-def cert_matches_key(cert, pkey, passphrase):
-    '''
-    Check that the x509 certificate matches the RSA private key.
-    '''
-
-    return _get_modulus(cert, None, True) \
-        == _get_modulus(pkey, passphrase, False)
-
-
-def _is_encrypted(pkey):
-    '''
-    Check if a private key is encrypted.
-    '''
-
-    with open(pkey, 'rb') as f:
-        for line in f:
-            if b'-----BEGIN ENCRYPTED PRIVATE KEY-----' == line.strip():
-                return True
-
-    return False
-
-
-def prompt_passphrase(pkey, passphrase_env_var=None, passphrase_file=None):
-    '''
-    If the private key is encrypted:
-
-    * try to read from the specified passphrase file (first line with trailing
-      line endings stripped)
-    * try to read from the passphrase environment variable
-    * prompt for the passphrase interactively
-
-    There is no fallback behavior.
-    '''
-
-    if not _is_encrypted(pkey):
-        return None
-
-    if passphrase_file is not None:
-        with open(passphrase_file, 'r') as f:
-            passphrase = f.readline().rstrip('\r\n')
-    elif passphrase_env_var is not None:
-        passphrase = os.environ[passphrase_env_var]
-    else:
-        passphrase = getpass.getpass(f'Passphrase for {pkey}: ')
-
-    # Verify that it is correct
-    with inject_passphrase(passphrase):
-        subprocess.check_output(['openssl', 'pkey', '-in', pkey, '-noout'])
-
-    return passphrase

avbroot/ota.py

@@ -1,817 +0,0 @@
-import base64
-import binascii
-import bz2
-import collections
-import concurrent.futures
-import contextlib
-import hashlib
-import io
-import lzma
-import os
-import struct
-import sys
-import subprocess
-import threading
-import unittest.mock
-import zipfile
-
-# Silence undesired warning
-orig_argv0 = sys.argv[0]
-sys.argv[0] = os.path.basename(sys.argv[0]).removesuffix('.py')
-import ota_utils
-sys.argv[0] = orig_argv0
-
-import ota_metadata_pb2
-import update_metadata_pb2
-
-from . import openssl
-from . import util
-
-
-OTA_MAGIC = b'CrAU'
-
-
-def parse_payload(f):
-    '''
-    Parse payload header from a file-like object. After this function returns,
-    the file position is set to the beginning of the blob section.
-    '''
-
-    f.seek(0)
-
-    # Validate header
-    magic = f.read(4)
-    if magic != OTA_MAGIC:
-        raise Exception(f'Invalid magic: {magic}')
-
-    version, = struct.unpack('!Q', f.read(8))
-    if version != 2:
-        raise Exception(f'Unsupported version: {version}')
-
-    manifest_size, = struct.unpack('!Q', f.read(8))
-    metadata_signature_size, = struct.unpack('!I', f.read(4))
-
-    # Read manifest
-    manifest_raw = f.read(manifest_size)
-    manifest = update_metadata_pb2.DeltaArchiveManifest()
-    manifest.ParseFromString(manifest_raw)
-
-    if any(p.HasField('old_partition_info') for p in manifest.partitions):
-        raise Exception('File is a delta OTA, not a full OTA')
-
-    # Skip manifest signatures
-    f.seek(metadata_signature_size, os.SEEK_CUR)
-
-    return (version, manifest, f.tell())
-
-
-def _extract_image(f_payload, f_out, block_size, blob_offset, partition,
-                   cancel_signal):
-    '''
-    Extract the partition image from <f_payload> to <f_out> by processing the
-    manifests list of install operations.
-    '''
-
-    Type = update_metadata_pb2.InstallOperation.Type
-
-    for op in partition.operations:
-        for extent in op.dst_extents:
-            if cancel_signal.is_set():
-                raise Exception('Interrupted')
-
-            f_payload.seek(blob_offset + op.data_offset)
-            f_out.seek(extent.start_block * block_size)
-            h_data = hashlib.sha256()
-
-            if op.type == Type.REPLACE:
-                util.copyfileobj_n(f_payload, f_out, op.data_length,
-                                   hasher=h_data)
-            elif op.type == Type.REPLACE_BZ:
-                decompressor = bz2.BZ2Decompressor()
-                util.decompress_n(decompressor, f_payload, f_out,
-                                  op.data_length, hasher=h_data)
-            elif op.type == Type.REPLACE_XZ:
-                decompressor = lzma.LZMADecompressor()
-                util.decompress_n(decompressor, f_payload, f_out,
-                                  op.data_length, hasher=h_data)
-            elif op.type == Type.ZERO or op.type == Type.DISCARD:
-                util.zero_n(f_out, extent.num_blocks * block_size)
-            else:
-                raise Exception(f'Unsupported operation: {op.type}')
-
-            if h_data.digest() != op.data_sha256_hash and op.type != Type.ZERO:
-                raise Exception('Expected hash %s, but got %s' %
-                                (h_data.hexdigest(),
-                                 binascii.hexlify(op.data_sha256_hash)))
-
-
-def extract_images(f, manifest, blob_offset, output_dir, partition_names):
-    '''
-    Extract the specified partition images from the payload into <output_dir>.
-
-    If <f> is callable, then it should produce a new file object each time it
-    is called. This allows extracting images in parallel.
-    '''
-
-    remaining = set(partition_names)
-    max_workers = len(remaining)
-    cancel_signal = threading.Event()
-    futures = []
-
-    if not callable(f):
-        f_orig = f
-
-        @contextlib.contextmanager
-        def dummy():
-            yield f_orig
-
-        f = dummy
-        max_workers = 1
-
-    def extract(p):
-        output_path = os.path.join(output_dir, p.partition_name + '.img')
-
-        with (
-            f() as f_in,
-            open(output_path, 'wb') as f_out,
-        ):
-            _extract_image(f_in, f_out, manifest.block_size, blob_offset, p,
-                           cancel_signal)
-
-    with concurrent.futures.ThreadPoolExecutor(
-            max_workers=max_workers) as executor:
-        try:
-            for p in manifest.partitions:
-                if p.partition_name not in remaining:
-                    continue
-
-                remaining.remove(p.partition_name)
-
-                futures.append(executor.submit(extract, p))
-
-            for future in concurrent.futures.as_completed(futures):
-                future.result()
-        except BaseException:
-            cancel_signal.set()
-            raise
-
-    if remaining:
-        raise Exception(f'Images not found: {remaining}')
-
-
-def _compress_image(partition, block_size, input_path, output_path):
-    '''
-    XZ-compress the image at <input_path> to <output_path> and update the
-    partition metadata with the appropriate checksums and install operations
-    metadata.
-
-    The size in the (sole) install operation is set correctly, but the offset
-    must be manually updated. It is initially set to the maximum uint64 value.
-    '''
-
-    h_uncompressed = hashlib.sha256()
-    h_compressed = hashlib.sha256()
-    size_uncompressed = 0
-    size_compressed = 0
-    # AOSP's payload_consumer does not support CRC during decompression
-    compressor = lzma.LZMACompressor(check=lzma.CHECK_NONE)
-    buf = bytearray(16384)
-    buf_view = memoryview(buf)
-
-    with (
-        open(input_path, 'rb', buffering=0) as f_in,
-        open(output_path, 'wb') as f_out,
-    ):
-        while n := f_in.readinto(buf_view):
-            h_uncompressed.update(buf_view[:n])
-            size_uncompressed += n
-
-            xz_data = compressor.compress(buf_view[:n])
-            h_compressed.update(xz_data)
-            size_compressed += len(xz_data)
-            f_out.write(xz_data)
-
-        xz_data = compressor.flush()
-        h_compressed.update(xz_data)
-        size_compressed += len(xz_data)
-        f_out.write(xz_data)
-
-    if size_uncompressed % block_size:
-        raise Exception('Size of %s (%d) is not aligned to the block size (%d)'
-                        % (partition.partition_name, size_uncompressed,
-                           block_size))
-
-    partition.new_partition_info.size = size_uncompressed
-    partition.new_partition_info.hash = h_uncompressed.digest()
-
-    extent = update_metadata_pb2.Extent()
-    extent.start_block = 0
-    extent.num_blocks = size_uncompressed // block_size
-
-    operation = update_metadata_pb2.InstallOperation()
-    operation.type = update_metadata_pb2.InstallOperation.Type.REPLACE_XZ
-    # Must be manually updated by the caller
-    operation.data_offset = 2 ** 64 - 1
-    operation.data_length = size_compressed
-    operation.dst_extents.append(extent)
-    operation.data_sha256_hash = h_compressed.digest()
-
-    partition.ClearField('operations')
-    partition.operations.append(operation)
-
-
-def _recompute_offsets(manifest, new_images):
-    '''
-    Recompute the blob offsets to account for the new images.
-
-    Returns ([(<image file>, <data offset>, <data size>)], <blob size>). If the
-    image file is None, then the data offset is relative to the blob offset of
-    the original payload. Otherwise, the data offset is an absolute offset into
-    the image file.
-    '''
-
-    # (<image file>, <data offset>, <data size>)
-    data_list = []
-    offset = 0
-
-    for p in manifest.partitions:
-        is_patched = p.partition_name in new_images
-        p_offset = 0
-
-        for op in p.operations:
-            if is_patched:
-                data_list.append((
-                    new_images[p.partition_name],
-                    p_offset,
-                    op.data_length,
-                ))
-            else:
-                data_list.append((
-                    None,
-                    op.data_offset,
-                    op.data_length,
-                ))
-
-            op.data_offset = offset
-            p_offset += op.data_length
-            offset += op.data_length
-
-    return (data_list, offset)
-
-
-def _sign_hash(hash, key, passphrase, max_sig_size):
-    '''
-    Sign <hash> with <key> and return a Signatures protobuf struct with the
-    signature padded to <max_sig_size>.
-    '''
-
-    hash_signed = openssl.sign_data(key, passphrase, hash)
-    assert len(hash_signed) <= max_sig_size
-
-    signature = update_metadata_pb2.Signatures.Signature()
-    signature.unpadded_signature_size = len(hash_signed)
-    signature.data = hash_signed + b'\0' * (max_sig_size - len(hash_signed))
-
-    signatures = update_metadata_pb2.Signatures()
-    signatures.signatures.append(signature)
-
-    return signatures
-
-
-def _serialize_protobuf(p):
-    return p.SerializeToString(deterministic=True)
-
-
-def patch_payload(f_in, f_out, version, manifest, blob_offset, temp_dir,
-                  patched, file_size, key, passphrase):
-    '''
-    Copy the payload from <f_in> to <f_out>, updating references to <patched>
-    images as they are encountered. <f_out> will be signed with <key>.
-    '''
-
-    max_sig_size = openssl.max_signature_size(key, passphrase)
-
-    # Strip out old payload signature
-    if manifest.HasField('signatures_size'):
-        trunc_file_size = blob_offset + manifest.signatures_offset
-        if trunc_file_size > file_size:
-            raise Exception('Payload signature offset is beyond EOF')
-        file_size = trunc_file_size
-
-    # Partition name -> compressed image path
-    compressed = {}
-
-    # Update the partition manifests to refer to the patched images
-    for name, path in patched.items():
-        # Find the partition in the manifest
-        partition = next((p for p in manifest.partitions
-                          if p.partition_name == name), None)
-        if partition is None:
-            raise Exception(f'Partition {name} not found in manifest')
-
-        # Compress the image and update the partition manifest accordingly
-        compressed_path = os.path.join(temp_dir, f'{name}.img')
-        _compress_image(
-            partition,
-            manifest.block_size,
-            path,
-            compressed_path,
-        )
-        compressed[name] = compressed_path
-
-    # Fill out blob offsets and compute final size
-    blob_data_list, blob_size = _recompute_offsets(manifest, compressed)
-
-    # Get the length of an dummy signature struct since the length fields are
-    # part of the data to be signed
-    dummy_sig = _sign_hash(hashlib.sha256().digest(), key, passphrase,
-                           max_sig_size)
-    dummy_sig_size = len(_serialize_protobuf(dummy_sig))
-
-    # Fill out new payload signature information
-    manifest.signatures_offset = blob_size
-    manifest.signatures_size = dummy_sig_size
-
-    # Build new manifest
-    manifest_raw_new = _serialize_protobuf(manifest)
-
-    class MultipleHasher:
-        def __init__(self, hashers):
-            self.hashers = hashers
-
-        def update(self, data):
-            for hasher in self.hashers:
-                hasher.update(data)
-
-    # Excludes signatures (hashes are for signing)
-    h_partial = hashlib.sha256()
-    # Includes signatures (hashes are for properties file)
-    h_full = hashlib.sha256()
-    # Updates both of the above
-    h_both = MultipleHasher((h_partial, h_full))
-
-    def write(hasher, data):
-        hasher.update(data)
-        f_out.write(data)
-
-    # Write header to output file
-    write(h_both, OTA_MAGIC)
-    write(h_both, struct.pack('!Q', version))
-    write(h_both, struct.pack('!Q', len(manifest_raw_new)))
-    write(h_both, struct.pack('!I', dummy_sig_size))
-
-    # Write new manifest
-    write(h_both, manifest_raw_new)
-
-    # Sign metadata (header + manifest) hash. The signature is not included in
-    # the payload hash.
-    metadata_hash = h_partial.digest()
-    metadata_sig = _sign_hash(metadata_hash, key, passphrase, max_sig_size)
-    write(h_full, _serialize_protobuf(metadata_sig))
-
-    # Write new blob
-    for image_file, data_offset, data_length in blob_data_list:
-        if image_file is None:
-            f_in.seek(blob_offset + data_offset)
-            util.copyfileobj_n(f_in, f_out, data_length, hasher=h_both)
-        else:
-            with open(image_file, 'rb') as f_image:
-                f_image.seek(data_offset)
-                util.copyfileobj_n(f_image, f_out, data_length, hasher=h_both)
-
-    # Append payload signature
-    payload_sig = _sign_hash(h_partial.digest(), key, passphrase, max_sig_size)
-    write(h_full, _serialize_protobuf(payload_sig))
-
-    # Generate properties file
-    metadata_offset = len(OTA_MAGIC) + struct.calcsize('!QQI')
-    metadata_size = metadata_offset + len(manifest_raw_new)
-    blob_size = manifest.signatures_offset + manifest.signatures_size
-    new_file_size = metadata_size + dummy_sig_size + blob_size
-
-    def b64(d): return base64.b64encode(d)
-    props = [
-        b'FILE_HASH=%s\n' % b64(h_full.digest()),
-        b'FILE_SIZE=%d\n' % new_file_size,
-        b'METADATA_HASH=%s\n' % b64(metadata_hash),
-        b'METADATA_SIZE=%d\n' % metadata_size,
-    ]
-
-    return b''.join(props)
-
-
-def _get_property_files():
-    '''
-    Return the set of property files to add to the OTA metadata files.
-    '''
-
-    return (
-        ota_utils.AbOtaPropertyFiles(),
-        ota_utils.StreamingPropertyFiles(),
-    )
-
-
-def _serialize_metadata(metadata):
-    '''
-    Generate the legacy plain-text and protobuf serializations of the given
-    metadata instance.
-    '''
-
-    legacy_metadata = ota_utils.BuildLegacyOtaMetadata(metadata)
-    legacy_metadata_str = "".join([f'{k}={v}\n' for k, v in
-                                   sorted(legacy_metadata.items())])
-    metadata_bytes = _serialize_protobuf(metadata)
-
-    return legacy_metadata_str.encode('UTF-8'), metadata_bytes
-
-
-_FileRange = collections.namedtuple(
-    '_FileRange', ('start', 'end', 'data_or_fp'))
-
-
-class _ConcatenatedFileDescriptor:
-    '''
-    A read-only seekable file descriptor that presents several file descriptors
-    or byte arrays as a single concatenated file.
-    '''
-
-    def __init__(self):
-        # List of (start, end, data_or_fp)
-        self.ranges = []
-        self.offset = 0
-
-    def _get_range(self):
-        for range in self.ranges:
-            if self.offset >= range.start and self.offset < range.end:
-                return range
-
-        return None
-
-    def _eof_offset(self):
-        return self.ranges[-1].end if self.ranges else 0
-
-    def add_file(self, fp):
-        start = self._eof_offset()
-        self.ranges.append(_FileRange(start, start + fp.tell(), fp))
-
-    def add_bytes(self, data):
-        start = self._eof_offset()
-        self.ranges.append(_FileRange(start, start + len(data), data))
-
-    def read(self, size=None):
-        buf = b''
-
-        while size is None or size > 0:
-            range = self._get_range()
-            if not range:
-                break
-
-            to_read = range.end - self.offset
-            if size is not None:
-                to_read = min(to_read, size)
-            data_offset = self.offset - range.start
-
-            if isinstance(range.data_or_fp, bytes):
-                data = range.data_or_fp[data_offset:data_offset + to_read]
-            else:
-                range.data_or_fp.seek(data_offset)
-                data = range.data_or_fp.read(to_read)
-
-            if not buf:
-                buf = data
-            else:
-                buf += data
-
-            if len(data) < to_read:
-                if range is not self.ranges[-1]:
-                    raise Exception('Unexpected EOF')
-                else:
-                    break
-
-            if size is not None:
-                size -= to_read
-
-        return buf
-
-    def seek(self, offset, whence=os.SEEK_SET):
-        if whence == os.SEEK_SET:
-            self.offset = offset
-        elif whence == os.SEEK_CUR:
-            self.offset += offset
-        elif whence == os.SEEK_END:
-            self.offset = self._eof_offset() + offset
-        else:
-            raise ValueError(f'Invalid whence: {whence}')
-
-    def tell(self):
-        return self.offset
-
-
-class _MemoryFile(io.BytesIO):
-    '''
-    Subclass of io.BytesIO where seeking can be conditionally disabled.
-    '''
-
-    def __init__(self, *args, allow_seek=True, **kwargs):
-        super().__init__(*args, **kwargs)
-        self.allow_seek = allow_seek
-
-    def seek(self, *args, **kwargs):
-        if not self.allow_seek:
-            raise AttributeError('seek is not supported')
-
-        return super().seek(*args, **kwargs)
-
-
-class _FakeZipFile:
-    '''
-    A wrapper around a ZipFile instance that allows appending new entries in
-    memory without modifying the backing file.
-
-    NOTE: The underlying ZipFile's file descriptor's position may be changed.
-    '''
-
-    def __init__(self, z):
-        self.zip = z
-
-        self.fp = _ConcatenatedFileDescriptor()
-
-        # We have a seekable underlying file descriptor to the zip, but we
-        # intentionally don't allow _TeeFileDescriptor to be seekable to
-        # guarantee that ZipFile writes sequentially.
-        self.orig_fp = self.zip.fp
-        if isinstance(self.orig_fp, _TeeFileDescriptor):
-            self.orig_fp = self.orig_fp.backing
-
-        self.fp.add_file(self.orig_fp)
-
-        self.next_offset = self.zip.start_dir
-
-        self.extra_infos = {}
-
-    def getinfo(self, name):
-        if name in self.extra_infos:
-            return self.extra_infos[name]
-        else:
-            return self.zip.getinfo(name)
-
-    def namelist(self):
-        return self.zip.namelist() + list(self.extra_infos.keys())
-
-    def add_file(self, info, data):
-        # Disable seeking to ensure that data descriptors are written, like the
-        # backing ZipFile
-        with _MemoryFile(allow_seek=False) as mem:
-            with zipfile.ZipFile(mem, 'w') as z:
-                with z.open(info, 'w') as f:
-                    f.write(data)
-
-                # Capture local file header, data, and data descriptor
-                buf_without_footer = mem.getvalue()
-                self.fp.add_bytes(buf_without_footer)
-
-                # Fix offset and add to fake entries
-                new_info = z.infolist()[-1]
-                new_info.header_offset = self.next_offset
-                self.extra_infos[new_info.filename] = new_info
-
-                self.next_offset += len(buf_without_footer)
-
-
-def add_metadata(z_out, metadata_info, metadata_pb_info, metadata_pb_raw):
-    '''
-    Add metadata files to the output OTA zip. <metadata_info> and
-    <metadata_pb_info> should be the ZipInfo instances associated with the
-    files from the original OTA zip. <metadata_pb_raw> should be the serialized
-    OTA metadata protobuf struct from the original OTA.
-
-    The zip file's backing file position MUST BE set to where the central
-    directory would start.
-    '''
-
-    metadata = ota_metadata_pb2.OtaMetadata()
-    metadata.ParseFromString(metadata_pb_raw)
-
-    metadata.property_files.clear()
-
-    props = _get_property_files()
-
-    # Create a fake zip instance that allows appending new entries in memory so
-    # that ota_utils can compute offsets for the property files
-    fake_zip = _FakeZipFile(z_out)
-
-    # Compute initial property files with reserved space as placeholders to
-    # store the self-referential metadata entries later
-    for p in props:
-        metadata.property_files[p.name] = p.Compute(fake_zip)
-
-    # Add the placeholders to the fake zip to compute final property files
-    new_metadata_raw, new_metadata_pb_raw = _serialize_metadata(metadata)
-    fake_zip.add_file(metadata_info, new_metadata_raw)
-    fake_zip.add_file(metadata_pb_info, new_metadata_pb_raw)
-
-    # Compute the final property files using the offsets of the fake entries
-    for p in props:
-        metadata.property_files[p.name] = \
-            p.Finalize(fake_zip, len(metadata.property_files[p.name]))
-
-    # Offset computation changes the file offset of the actual file. Seek back
-    # to where the next entry or central directory would go
-    fake_zip.orig_fp.seek(z_out.start_dir)
-
-    # Add the final metadata files to the real zip
-    new_metadata_raw, new_metadata_pb_raw = _serialize_metadata(metadata)
-    with z_out.open(metadata_info, 'w') as f:
-        f.write(new_metadata_raw)
-    with z_out.open(metadata_pb_info, 'w') as f:
-        f.write(new_metadata_pb_raw)
-
-    return metadata
-
-
-def verify_metadata(z, metadata):
-    '''
-    Verify that the offsets and file sizes within the metadata file properties
-    of a fully written OTA zip are correct.
-    '''
-
-    for p in _get_property_files():
-        p.Verify(z, metadata.property_files[p.name].strip())
-
-
-class _TeeFileDescriptor:
-    '''
-    A file-like instance that propagates writes to multiple streams.
-
-    start_capture() is used to pause output and divert writes to a memory
-    buffer until _finish_capture(), which can modify the buffer.
-    '''
-
-    def __init__(self, streams, file_index=None):
-        self.streams = streams
-        self.capture = None
-        self.backing = None if file_index is None else streams[file_index]
-
-    def write(self, data):
-        if self.capture:
-            self.capture.write(data)
-        else:
-            for stream in self.streams:
-                # Naive hole punching to create sparse files
-                if stream is self.backing and util.is_zero(data):
-                    stream.seek(len(data), os.SEEK_CUR)
-                else:
-                    stream.write(data)
-
-        return len(data)
-
-    def flush(self):
-        for stream in self.streams:
-            stream.flush()
-
-    def tell(self):
-        if self.backing is None:
-            # Fake non-existance
-            raise AttributeError('tell is not supported')
-
-        capture_len = self.capture.tell() if self.capture else 0
-        return self.backing.tell() + capture_len
-
-    def start_capture(self):
-        if self.capture is not None:
-            raise RuntimeError('Capture already started')
-
-        self.capture = _MemoryFile()
-
-    @contextlib.contextmanager
-    def _finish_capture(self):
-        if not self.capture:
-            raise RuntimeError('No capture started')
-
-        yield self.capture
-
-        for stream in self.streams:
-            stream.write(self.capture.getbuffer())
-
-        self.capture.close()
-        self.capture = None
-
-
-@contextlib.contextmanager
-def open_signing_wrapper(f, privkey, passphrase, cert):
-    '''
-    Create a file-like wrapper around an existing file object that performs CMS
-    signing as data is being written.
-    '''
-
-    with openssl.inject_passphrase(passphrase):
-        session_kwargs = {}
-        if os.name != 'nt':
-            # We don't want the controlling terminal to interrupt openssl on
-            # ^C or ^\. That'll cause _TeeFileDescriptor's writes to the stdin
-            # pipe to fail, and certain classes, like ZipFile, will write to
-            # the fd in their __exit__ methods. This causes a BrokenPipeError
-            # to be raised while the existing KeyboardInterrupt is being
-            # propagated up. We'll handling killing openssl ourselves.
-            session_kwargs['start_new_session'] = True
-
-        process = subprocess.Popen(
-            [
-                'openssl',
-                'cms',
-                '-sign',
-                '-binary',
-                '-outform', 'DER',
-                '-inkey', privkey,
-                '-signer', cert,
-                # Mimic signapk behavior by excluding signed attributes
-                '-noattr',
-                '-nosmimecap',
-            ],
-            stdin=subprocess.PIPE,
-            stdout=subprocess.PIPE,
-            **session_kwargs,
-        )
-
-    try:
-        wrapper = _TeeFileDescriptor((f, process.stdin), file_index=0)
-        yield wrapper
-
-        with wrapper._finish_capture() as f_buffer:
-            # Save a copy of the zip central directory
-            f_buffer.seek(0)
-            footer = f_buffer.read()
-
-            # Delete the archive comment size field
-            if len(footer) < 2:
-                raise Exception('zip central directory is too small')
-            elif footer[-2:] != b'\x00\x00':
-                raise Exception('zip has unexpected archive comment')
-
-            f_buffer.seek(-2, os.SEEK_CUR)
-            f_buffer.truncate(f_buffer.tell())
-
-        process.stdin.close()
-        signature = process.stdout.read()
-    except BaseException:
-        process.kill()
-        raise
-    finally:
-        process.wait()
-
-    if process.returncode != 0:
-        raise Exception(f'openssl exited with status: {process.returncode}')
-
-    # Double check that the EOCD magic is where it should be when there is no
-    # archive comment
-    if footer[-22:-18] != zipfile.stringEndArchive:
-        raise Exception('EOCD magic not found')
-
-    # Build a new archive comment that contains the signature
-    with io.BytesIO() as comment:
-        message = b'signed by avbroot\0'
-        comment.write(message)
-        comment.write(signature)
-
-        comment_size = comment.tell() + 6
-
-        if comment_size > 0xffff:
-            raise Exception('Archive comment with signature is too large')
-
-        comment.write(struct.pack(
-            '<HHH',
-            # Absolute value of the offset of the signature from the end of the
-            # archive comment
-            comment_size - len(message),
-            0xffff,
-            comment_size,
-        ))
-
-        # Verify that we won't be producing a duplicate EOCD magic
-        if zipfile.stringEndArchive in comment.getbuffer():
-            raise Exception('Archive comment contains EOCD magic')
-
-        # Write comment size to output file (which was removed before)
-        f.write(struct.pack('<H', comment_size))
-
-        # Write comment to output file
-        f.write(comment.getbuffer())
-
-
-@contextlib.contextmanager
-def match_android_zip64_limit():
-    '''
-    Python's ZipFile implementation uses zip64 when the size of an entry is >
-    0x7fffffff. However, Android's libarchive behavior is incorrect [1] and
-    treats the data descriptor size fields as 32-bit unless the compressed or
-    uncompressed size in the central directory is >= 0xffffffff. This causes
-    files containing entries with sizes in [2 GiB, 4 GiB - 2] to fail to flash
-    in Android's recovery environment. Work around this by changing ZipFile's
-    threshold to match Android's.
-
-    [1] https://cs.android.com/android/platform/superproject/+/android-13.0.0_r18:system/libziparchive/zip_archive.cc;l=692
-    '''
-
-    # Because Python uses > and Android uses >= 0xffffffff
-    with unittest.mock.patch('zipfile.ZIP64_LIMIT', 0xfffffffe):
-        yield

avbroot/protobuf/ota_metadata.proto

@@ -0,0 +1,115 @@
+/*
+ * Copyright (C) 2020 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+// If you change this file,
+// Please update ota_metadata_pb2.py by executing
+// protoc ota_metadata.proto --python_out
+// $ANDROID_BUILD_TOP/build/tools/releasetools
+
+syntax = "proto3";
+
+package build.tools.releasetools;
+option optimize_for = LITE_RUNTIME;
+option java_package = "android.ota";
+option java_outer_classname = "OtaPackageMetadata";
+
+// The build information of a particular partition on the device.
+message PartitionState {
+  string partition_name = 1;
+  repeated string device = 2;
+  repeated string build = 3;
+  // The version string of the partition. It's usually timestamp if present.
+  // One known exception is the boot image, who uses the kmi version, e.g.
+  // 5.4.42-android12-0
+  string version = 4;
+
+  // TODO(xunchang), revisit other necessary fields, e.g. security_patch_level.
+}
+
+// The build information on the device. The bytes of the running images are thus
+// inferred from the device state. For more information of the meaning of each
+// subfield, check
+// https://source.android.com/compatibility/android-cdd#3_2_2_build_parameters
+message DeviceState {
+  // device name. i.e. ro.product.device; if the field has multiple values, it
+  // means the ota package supports multiple devices. This usually happens when
+  // we use the same image to support multiple skus.
+  repeated string device = 1;
+  // device fingerprint. Up to R build, the value reads from
+  // ro.build.fingerprint.
+  repeated string build = 2;
+  // A value that specify a version of the android build.
+  string build_incremental = 3;
+  // The timestamp when the build is generated.
+  int64 timestamp = 4;
+  // The version of the currently-executing Android system.
+  string sdk_level = 5;
+  // A value indicating the security patch level of a build.
+  string security_patch_level = 6;
+
+  // The detailed state of each partition. For partial updates or devices with
+  // mixed build of partitions, some of the above fields may left empty. And the
+  // client will rely on the information of specific partitions to target the
+  // update.
+  repeated PartitionState partition_state = 7;
+}
+
+message ApexInfo {
+  string package_name = 1;
+  int64 version = 2;
+  bool is_compressed = 3;
+  int64 decompressed_size = 4;
+  // Used in OTA
+  int64 source_version = 5;
+}
+
+// Just a container to hold repeated apex_info, so that we can easily serialize
+// a list of apex_info to string.
+message ApexMetadata {
+  repeated ApexInfo apex_info = 1;
+}
+
+// The metadata of an OTA package. It contains the information of the package
+// and prerequisite to install the update correctly.
+message OtaMetadata {
+  enum OtaType {
+    UNKNOWN = 0;
+    AB = 1;
+    BLOCK = 2;
+    BRICK = 3;
+  };
+  OtaType type = 1;
+  // True if we need to wipe after the update.
+  bool wipe = 2;
+  // True if the timestamp of the post build is older than the pre build.
+  bool downgrade = 3;
+  // A map of name:content of property files, e.g. ota-property-files.
+  map<string, string> property_files = 4;
+
+  // The required device state in order to install the package.
+  DeviceState precondition = 5;
+  // The expected device state after the update.
+  DeviceState postcondition = 6;
+
+  // True if the ota that updates a device to support dynamic partitions, where
+  // the source build doesn't support it.
+  bool retrofit_dynamic_partitions = 7;
+  // The required size of the cache partition, only valid for non-A/B update.
+  int64 required_cache = 8;
+
+  // True iff security patch level downgrade is permitted on this OTA.
+  bool spl_downgrade = 9;
+}

avbroot/protobuf/update_metadata.proto

@@ -0,0 +1,437 @@
+//
+// Copyright (C) 2010 The Android Open Source Project
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+
+// Update file format: An update file contains all the operations needed
+// to update a system to a specific version. It can be a full payload which
+// can update from any version, or a delta payload which can only update
+// from a specific version.
+// The update format is represented by this struct pseudocode:
+// struct delta_update_file {
+//   char magic[4] = "CrAU";
+//   uint64 file_format_version;  // payload major version
+//   uint64 manifest_size;  // Size of protobuf DeltaArchiveManifest
+//
+//   // Only present if format_version >= 2:
+//   uint32 metadata_signature_size;
+//
+//   // The DeltaArchiveManifest protobuf serialized, not compressed.
+//   char manifest[manifest_size];
+//
+//   // The signature of the metadata (from the beginning of the payload up to
+//   // this location, not including the signature itself). This is a serialized
+//   // Signatures message.
+//   char metadata_signature_message[metadata_signature_size];
+//
+//   // Data blobs for files, no specific format. The specific offset
+//   // and length of each data blob is recorded in the DeltaArchiveManifest.
+//   struct {
+//     char data[];
+//   } blobs[];
+//
+//   // The signature of the entire payload, everything up to this location,
+//   // except that metadata_signature_message is skipped to simplify signing
+//   // process. These two are not signed:
+//   uint64 payload_signatures_message_size;
+//   // This is a serialized Signatures message.
+//   char payload_signatures_message[payload_signatures_message_size];
+//
+// };
+
+// The DeltaArchiveManifest protobuf is an ordered list of InstallOperation
+// objects. These objects are stored in a linear array in the
+// DeltaArchiveManifest. Each operation is applied in order by the client.
+
+// The DeltaArchiveManifest also contains the initial and final
+// checksums for the device.
+
+// The client will perform each InstallOperation in order, beginning even
+// before the entire delta file is downloaded (but after at least the
+// protobuf is downloaded). The types of operations are explained:
+// - REPLACE: Replace the dst_extents on the drive with the attached data,
+//   zero padding out to block size.
+// - REPLACE_BZ: bzip2-uncompress the attached data and write it into
+//   dst_extents on the drive, zero padding to block size.
+// - MOVE: Copy the data in src_extents to dst_extents. Extents may overlap,
+//   so it may be desirable to read all src_extents data into memory before
+//   writing it out. (deprecated)
+// - SOURCE_COPY: Copy the data in src_extents in the old partition to
+//   dst_extents in the new partition. There's no overlapping of data because
+//   the extents are in different partitions.
+// - BSDIFF: Read src_length bytes from src_extents into memory, perform
+//   bspatch with attached data, write new data to dst_extents, zero padding
+//   to block size. (deprecated)
+// - SOURCE_BSDIFF: Read the data in src_extents in the old partition, perform
+//   bspatch with the attached data and write the new data to dst_extents in the
+//   new partition.
+// - ZERO: Write zeros to the destination dst_extents.
+// - DISCARD: Discard the destination dst_extents blocks on the physical medium.
+//   the data read from those blocks is undefined.
+// - REPLACE_XZ: Replace the dst_extents with the contents of the attached
+//   xz file after decompression. The xz file should only use crc32 or no crc at
+//   all to be compatible with xz-embedded.
+// - PUFFDIFF: Read the data in src_extents in the old partition, perform
+//   puffpatch with the attached data and write the new data to dst_extents in
+//   the new partition.
+//
+// The operations allowed in the payload (supported by the client) depend on the
+// major and minor version. See InstallOperation.Type below for details.
+
+syntax = "proto2";
+
+package chromeos_update_engine;
+
+// Data is packed into blocks on disk, always starting from the beginning
+// of the block. If a file's data is too large for one block, it overflows
+// into another block, which may or may not be the following block on the
+// physical partition. An ordered list of extents is another
+// representation of an ordered list of blocks. For example, a file stored
+// in blocks 9, 10, 11, 2, 18, 12 (in that order) would be stored in
+// extents { {9, 3}, {2, 1}, {18, 1}, {12, 1} } (in that order).
+// In general, files are stored sequentially on disk, so it's more efficient
+// to use extents to encode the block lists (this is effectively
+// run-length encoding).
+// A sentinel value (kuint64max) as the start block denotes a sparse-hole
+// in a file whose block-length is specified by num_blocks.
+
+message Extent {
+  optional uint64 start_block = 1;
+  optional uint64 num_blocks = 2;
+}
+
+// Signatures: Updates may be signed by the OS vendor. The client verifies
+// an update's signature by hashing the entire download. The section of the
+// download that contains the signature is at the end of the file, so when
+// signing a file, only the part up to the signature part is signed.
+// Then, the client looks inside the download's Signatures message for a
+// Signature message that it knows how to handle. Generally, a client will
+// only know how to handle one type of signature, but an update may contain
+// many signatures to support many different types of client. Then client
+// selects a Signature message and uses that, along with a known public key,
+// to verify the download. The public key is expected to be part of the
+// client.
+
+message Signatures {
+  message Signature {
+    optional uint32 version = 1 [deprecated = true];
+    optional bytes data = 2;
+
+    // The DER encoded signature size of EC keys is nondeterministic for
+    // different input of sha256 hash. However, we need the size of the
+    // serialized signatures protobuf string to be fixed before signing;
+    // because this size is part of the content to be signed. Therefore, we
+    // always pad the signature data to the maximum possible signature size of
+    // a given key. And the payload verifier will truncate the signature to
+    // its correct size based on the value of |unpadded_signature_size|.
+    optional fixed32 unpadded_signature_size = 3;
+  }
+  repeated Signature signatures = 1;
+}
+
+message PartitionInfo {
+  optional uint64 size = 1;
+  optional bytes hash = 2;
+}
+
+message InstallOperation {
+  enum Type {
+    REPLACE = 0;     // Replace destination extents w/ attached data.
+    REPLACE_BZ = 1;  // Replace destination extents w/ attached bzipped data.
+    MOVE = 2 [deprecated = true];    // Move source extents to target extents.
+    BSDIFF = 3 [deprecated = true];  // The data is a bsdiff binary diff.
+
+    // On minor version 2 or newer, these operations are supported:
+    SOURCE_COPY = 4;    // Copy from source to target partition
+    SOURCE_BSDIFF = 5;  // Like BSDIFF, but read from source partition
+
+    // On minor version 3 or newer and on major version 2 or newer, these
+    // operations are supported:
+    REPLACE_XZ = 8;  // Replace destination extents w/ attached xz data.
+
+    // On minor version 4 or newer, these operations are supported:
+    ZERO = 6;     // Write zeros in the destination.
+    DISCARD = 7;  // Discard the destination blocks, reading as undefined.
+    BROTLI_BSDIFF = 10;  // Like SOURCE_BSDIFF, but compressed with brotli.
+
+    // On minor version 5 or newer, these operations are supported:
+    PUFFDIFF = 9;  // The data is in puffdiff format.
+
+    // On minor version 8 or newer, these operations are supported:
+    ZUCCHINI = 11;
+
+    // On minor version 9 or newer, these operations are supported:
+    LZ4DIFF_BSDIFF = 12;
+    LZ4DIFF_PUFFDIFF = 13;
+  }
+  required Type type = 1;
+
+  // Only minor version 6 or newer support 64 bits |data_offset| and
+  // |data_length|, older client will read them as uint32.
+  // The offset into the delta file (after the protobuf)
+  // where the data (if any) is stored
+  optional uint64 data_offset = 2;
+  // The length of the data in the delta file
+  optional uint64 data_length = 3;
+
+  // Ordered list of extents that are read from (if any) and written to.
+  repeated Extent src_extents = 4;
+  // Byte length of src, equal to the number of blocks in src_extents *
+  // block_size. It is used for BSDIFF and SOURCE_BSDIFF, because we need to
+  // pass that external program the number of bytes to read from the blocks we
+  // pass it.  This is not used in any other operation.
+  optional uint64 src_length = 5;
+
+  repeated Extent dst_extents = 6;
+  // Byte length of dst, equal to the number of blocks in dst_extents *
+  // block_size. Used for BSDIFF and SOURCE_BSDIFF, but not in any other
+  // operation.
+  optional uint64 dst_length = 7;
+
+  // Optional SHA 256 hash of the blob associated with this operation.
+  // This is used as a primary validation for http-based downloads and
+  // as a defense-in-depth validation for https-based downloads. If
+  // the operation doesn't refer to any blob, this field will have
+  // zero bytes.
+  optional bytes data_sha256_hash = 8;
+
+  // Indicates the SHA 256 hash of the source data referenced in src_extents at
+  // the time of applying the operation. If present, the update_engine daemon
+  // MUST read and verify the source data before applying the operation.
+  optional bytes src_sha256_hash = 9;
+}
+
+// Hints to VAB snapshot to skip writing some blocks if these blocks are
+// identical to the ones on the source image. The src & dst extents for each
+// CowMergeOperation should be contiguous, and they're a subset of an OTA
+// InstallOperation.
+// During merge time, we need to follow the pre-computed sequence to avoid
+// read after write, similar to the inplace update schema.
+message CowMergeOperation {
+  enum Type {
+    COW_COPY = 0;     // identical blocks
+    COW_XOR = 1;      // used when src/dst blocks are highly similar
+    COW_REPLACE = 2;  // Raw replace operation
+  }
+  optional Type type = 1;
+
+  optional Extent src_extent = 2;
+  optional Extent dst_extent = 3;
+  // For COW_XOR, source location might be unaligned, so this field is in range
+  // [0, block_size), representing how much should the src_extent shift toward
+  // larger block number. If this field is non-zero, then src_extent will
+  // include 1 extra block in the end, as the merge op actually references the
+  // first |src_offset| bytes of that extra block. For example, if |dst_extent|
+  // is [10, 15], |src_offset| is 500, then src_extent might look like [25, 31].
+  // Note that |src_extent| contains 1 extra block than the |dst_extent|.
+  optional uint32 src_offset = 4;
+}
+
+// Describes the update to apply to a single partition.
+message PartitionUpdate {
+  // A platform-specific name to identify the partition set being updated. For
+  // example, in Chrome OS this could be "ROOT" or "KERNEL".
+  required string partition_name = 1;
+
+  // Whether this partition carries a filesystem with post-install program that
+  // must be run to finalize the update process. See also |postinstall_path| and
+  // |filesystem_type|.
+  optional bool run_postinstall = 2;
+
+  // The path of the executable program to run during the post-install step,
+  // relative to the root of this filesystem. If not set, the default "postinst"
+  // will be used. This setting is only used when |run_postinstall| is set and
+  // true.
+  optional string postinstall_path = 3;
+
+  // The filesystem type as passed to the mount(2) syscall when mounting the new
+  // filesystem to run the post-install program. If not set, a fixed list of
+  // filesystems will be attempted. This setting is only used if
+  // |run_postinstall| is set and true.
+  optional string filesystem_type = 4;
+
+  // If present, a list of signatures of the new_partition_info.hash signed with
+  // different keys. If the update_engine daemon requires vendor-signed images
+  // and has its public key installed, one of the signatures should be valid
+  // for /postinstall to run.
+  repeated Signatures.Signature new_partition_signature = 5;
+
+  optional PartitionInfo old_partition_info = 6;
+  optional PartitionInfo new_partition_info = 7;
+
+  // The list of operations to be performed to apply this PartitionUpdate. The
+  // associated operation blobs (in operations[i].data_offset, data_length)
+  // should be stored contiguously and in the same order.
+  repeated InstallOperation operations = 8;
+
+  // Whether a failure in the postinstall step for this partition should be
+  // ignored.
+  optional bool postinstall_optional = 9;
+
+  // On minor version 6 or newer, these fields are supported:
+
+  // The extent for data covered by verity hash tree.
+  optional Extent hash_tree_data_extent = 10;
+
+  // The extent to store verity hash tree.
+  optional Extent hash_tree_extent = 11;
+
+  // The hash algorithm used in verity hash tree.
+  optional string hash_tree_algorithm = 12;
+
+  // The salt used for verity hash tree.
+  optional bytes hash_tree_salt = 13;
+
+  // The extent for data covered by FEC.
+  optional Extent fec_data_extent = 14;
+
+  // The extent to store FEC.
+  optional Extent fec_extent = 15;
+
+  // The number of FEC roots.
+  optional uint32 fec_roots = 16 [default = 2];
+
+  // Per-partition version used for downgrade detection, added
+  // as an effort to support partial updates. For most partitions,
+  // this is the build timestamp.
+  optional string version = 17;
+
+  // A sorted list of CowMergeOperation. When writing cow, we can choose to
+  // skip writing the raw bytes for these extents. During snapshot merge, the
+  // bytes will read from the source partitions instead.
+  repeated CowMergeOperation merge_operations = 18;
+
+  // Estimated size for COW image. This is used by libsnapshot
+  // as a hint. If set to 0, libsnapshot should use alternative
+  // methods for estimating size.
+  optional uint64 estimate_cow_size = 19;
+}
+
+message DynamicPartitionGroup {
+  // Name of the group.
+  required string name = 1;
+
+  // Maximum size of the group. The sum of sizes of all partitions in the group
+  // must not exceed the maximum size of the group.
+  optional uint64 size = 2;
+
+  // A list of partitions that belong to the group.
+  repeated string partition_names = 3;
+}
+
+message VABCFeatureSet {
+  optional bool threaded = 1;
+  optional bool batch_writes = 2;
+}
+
+// Metadata related to all dynamic partitions.
+message DynamicPartitionMetadata {
+  // All updatable groups present in |partitions| of this DeltaArchiveManifest.
+  // - If an updatable group is on the device but not in the manifest, it is
+  //   not updated. Hence, the group will not be resized, and partitions cannot
+  //   be added to or removed from the group.
+  // - If an updatable group is in the manifest but not on the device, the group
+  //   is added to the device.
+  repeated DynamicPartitionGroup groups = 1;
+
+  // Whether dynamic partitions have snapshots during the update. If this is
+  // set to true, the update_engine daemon creates snapshots for all dynamic
+  // partitions if possible. If this is unset, the update_engine daemon MUST
+  // NOT create snapshots for dynamic partitions.
+  optional bool snapshot_enabled = 2;
+
+  // If this is set to false, update_engine should not use VABC regardless. If
+  // this is set to true, update_engine may choose to use VABC if device
+  // supports it, but not guaranteed.
+  // VABC stands for Virtual AB Compression
+  optional bool vabc_enabled = 3;
+
+  // The compression algorithm used by VABC. Available ones are "gz", "brotli".
+  // See system/core/fs_mgr/libsnapshot/cow_writer.cpp for available options,
+  // as this parameter is ultimated forwarded to libsnapshot's CowWriter
+  optional string vabc_compression_param = 4;
+
+  // COW version used by VABC. The represents the major version in the COW
+  // header
+  optional uint32 cow_version = 5;
+
+  // A collection of knobs to tune Virtual AB Compression
+  optional VABCFeatureSet vabc_feature_set = 6;
+}
+
+// Definition has been duplicated from
+// $ANDROID_BUILD_TOP/build/tools/releasetools/ota_metadata.proto. Keep in sync.
+message ApexInfo {
+  optional string package_name = 1;
+  optional int64 version = 2;
+  optional bool is_compressed = 3;
+  optional int64 decompressed_size = 4;
+}
+
+// Definition has been duplicated from
+// $ANDROID_BUILD_TOP/build/tools/releasetools/ota_metadata.proto. Keep in sync.
+message ApexMetadata {
+  repeated ApexInfo apex_info = 1;
+}
+
+message DeltaArchiveManifest {
+  // Only present in major version = 1. List of install operations for the
+  // kernel and rootfs partitions. For major version = 2 see the |partitions|
+  // field.
+  reserved 1, 2;
+
+  // (At time of writing) usually 4096
+  optional uint32 block_size = 3 [default = 4096];
+
+  // If signatures are present, the offset into the blobs, generally
+  // tacked onto the end of the file, and the length. We use an offset
+  // rather than a bool to allow for more flexibility in future file formats.
+  // If either is absent, it means signatures aren't supported in this
+  // file.
+  optional uint64 signatures_offset = 4;
+  optional uint64 signatures_size = 5;
+
+  // Fields deprecated in major version 2.
+  reserved 6,7,8,9,10,11;
+
+  // The minor version, also referred as "delta version", of the payload.
+  // Minor version 0 is full payload, everything else is delta payload.
+  optional uint32 minor_version = 12 [default = 0];
+
+  // Only present in major version >= 2. List of partitions that will be
+  // updated, in the order they will be updated. This field replaces the
+  // |install_operations|, |kernel_install_operations| and the
+  // |{old,new}_{kernel,rootfs}_info| fields used in major version = 1. This
+  // array can have more than two partitions if needed, and they are identified
+  // by the partition name.
+  repeated PartitionUpdate partitions = 13;
+
+  // The maximum timestamp of the OS allowed to apply this payload.
+  // Can be used to prevent downgrading the OS.
+  optional int64 max_timestamp = 14;
+
+  // Metadata related to all dynamic partitions.
+  optional DynamicPartitionMetadata dynamic_partition_metadata = 15;
+
+  // If the payload only updates a subset of partitions on the device.
+  optional bool partial_update = 16;
+
+  // Information on compressed APEX to figure out how much space is required for
+  // their decompression
+  repeated ApexInfo apex_info = 17;
+
+  // Security patch level of the device, usually in the format of
+  // yyyy-mm-dd
+  optional string security_patch_level = 18;
+}

avbroot/src/boot.rs

@@ -0,0 +1,788 @@
+/*
+ * SPDX-FileCopyrightText: 2022-2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::{
+    cmp::Ordering,
+    collections::HashMap,
+    fs::File,
+    io::{self, BufRead, BufReader, Cursor, Read, Seek, Write},
+    num::ParseIntError,
+    ops::Range,
+    path::{Path, PathBuf},
+    sync::{atomic::AtomicBool, Arc},
+};
+
+use regex::bytes::Regex;
+use ring::digest::Context;
+use rsa::RsaPrivateKey;
+use thiserror::Error;
+use x509_cert::Certificate;
+use xz2::{
+    stream::{Check, Stream},
+    write::XzEncoder,
+};
+use zip::{result::ZipError, write::FileOptions, CompressionMethod, ZipArchive, ZipWriter};
+
+use crate::{
+    crypto,
+    format::{
+        avb::{self, AlgorithmType, Descriptor},
+        bootimage::{self, BootImage, BootImageExt, RamdiskMeta},
+        compression::{self, CompressedFormat, CompressedReader, CompressedWriter},
+        cpio::{self, CpioEntryNew},
+    },
+    stream::{self, FromReader, HashingWriter, SectionReader, ToWriter},
+    util::EscapedString,
+};
+
+#[derive(Debug, Error)]
+pub enum Error {
+    #[error("Boot image has no vbmeta footer")]
+    NoFooter,
+    #[error("No hash descriptor found in vbmeta footer")]
+    NoHashDescriptor,
+    #[error("Found multiple hash descriptors in vbmeta footer")]
+    MultipleHashDescriptors,
+    #[error("Validation error: {0}")]
+    Validation(String),
+    #[error("Failed to parse Magisk version from line: {0:?}")]
+    ParseMagiskVersion(String, #[source] ParseIntError),
+    #[error("Failed to determine Magisk version from: {0:?}")]
+    FindMagiskVersion(PathBuf),
+    #[error("AVB error")]
+    Avb(#[from] avb::Error),
+    #[error("Boot image error")]
+    BootImage(#[from] bootimage::Error),
+    #[error("Compression error")]
+    Compression(#[from] compression::Error),
+    #[error("Crypto error")]
+    Crypto(#[from] crypto::Error),
+    #[error("CPIO error")]
+    Cpio(#[from] cpio::Error),
+    #[error("XZ stream error")]
+    XzStream(#[from] xz2::stream::Error),
+    #[error("Zip error")]
+    Zip(#[from] ZipError),
+    #[error("I/O error")]
+    IoError(#[from] io::Error),
+}
+
+type Result<T> = std::result::Result<T, Error>;
+
+fn load_ramdisk(data: &[u8]) -> Result<(Vec<CpioEntryNew>, CompressedFormat)> {
+    let raw_reader = Cursor::new(data);
+    let mut reader = CompressedReader::new(raw_reader, false)?;
+    let entries = cpio::load(&mut reader, false)?;
+
+    Ok((entries, reader.format()))
+}
+
+fn save_ramdisk(entries: &[CpioEntryNew], format: CompressedFormat) -> Result<Vec<u8>> {
+    let raw_writer = Cursor::new(vec![]);
+    let mut writer = CompressedWriter::new(raw_writer, format)?;
+    cpio::save(&mut writer, entries, false)?;
+
+    let raw_writer = writer.finish()?;
+    Ok(raw_writer.into_inner())
+}
+
+pub trait BootImagePatcher {
+    fn patch(&self, boot_image: &mut BootImage, cancel_signal: &Arc<AtomicBool>) -> Result<()>;
+}
+
+/// Root a boot image with Magisk.
+pub struct MagiskRootPatcher {
+    apk_path: PathBuf,
+    version: u32,
+    preinit_device: Option<String>,
+    random_seed: u64,
+}
+
+impl MagiskRootPatcher {
+    // - Versions <25102 are not supported because they're missing commit
+    //   1f8c063dc64806c4f7320ed66c785ff7bc116383, which would leave devices
+    //   that use Android 13 GKIs unable to boot into recovery
+    // - Versions 25207 through 25210 are not supported because they used the
+    //   RULESDEVICE config option, which stored the writable block device as an
+    //   rdev major/minor pair, which was not consistent across reboots and was
+    //   replaced by PREINITDEVICE
+    const VERS_SUPPORTED: &[Range<u32>] = &[25102..25207, 25211..26300];
+    const VER_PREINIT_DEVICE: Range<u32> =
+        25211..Self::VERS_SUPPORTED[Self::VERS_SUPPORTED.len() - 1].end;
+    const VER_RANDOM_SEED: Range<u32> = 25211..26103;
+
+    pub fn new(
+        path: &Path,
+        preinit_device: Option<&str>,
+        random_seed: Option<u64>,
+        ignore_compatibility: bool,
+        warning_fn: impl Fn(&str) + Send + 'static,
+    ) -> Result<Self> {
+        let version = Self::get_version(path)?;
+
+        if !Self::VERS_SUPPORTED.iter().any(|v| v.contains(&version)) {
+            let msg = format!(
+                "Unsupported Magisk version {} (supported: {:?})",
+                version,
+                Self::VERS_SUPPORTED,
+            );
+
+            if ignore_compatibility {
+                warning_fn(&msg);
+            } else {
+                return Err(Error::Validation(msg));
+            }
+        }
+
+        if preinit_device.is_none() && Self::VER_PREINIT_DEVICE.contains(&version) {
+            let msg = format!(
+                "Magisk version {} ({:?}) requires a preinit device to be specified",
+                version,
+                Self::VER_PREINIT_DEVICE,
+            );
+
+            if ignore_compatibility {
+                warning_fn(&msg);
+            } else {
+                return Err(Error::Validation(msg));
+            }
+        }
+
+        Ok(Self {
+            apk_path: path.to_owned(),
+            version,
+            preinit_device: preinit_device.map(|d| d.to_owned()),
+            // Use a hardcoded random seed by default to ensure byte-for-byte
+            // reproducibility.
+            random_seed: random_seed.unwrap_or(0xfedcba9876543210),
+        })
+    }
+
+    fn get_version(path: &Path) -> Result<u32> {
+        let reader = File::open(path)?;
+        let reader = BufReader::new(reader);
+        let mut zip = ZipArchive::new(reader)?;
+        let entry = zip.by_name("assets/util_functions.sh")?;
+        let mut entry = BufReader::new(entry);
+        let mut line = String::new();
+
+        loop {
+            line.clear();
+            let n = entry.read_line(&mut line)?;
+            if n == 0 {
+                return Err(Error::FindMagiskVersion(path.to_owned()));
+            }
+
+            if let Some(suffix) = line.trim_end().strip_prefix("MAGISK_VER_CODE=") {
+                let version = suffix
+                    .parse()
+                    .map_err(|e| Error::ParseMagiskVersion(suffix.to_owned(), e))?;
+                return Ok(version);
+            }
+        }
+    }
+
+    /// Compare old and new ramdisk entry lists, creating the Magisk `.backup/`
+    /// directory structure. `.backup/.rmlist` will contain a sorted list of
+    /// NULL-terminated strings, listing which files were newly added or
+    /// changed. The old entries for changed files will be added to the new
+    /// entries as `.backup/<path>`.
+    ///
+    /// Both lists and entries within the lists may be mutated.
+    fn apply_magisk_backup(old_entries: &mut [CpioEntryNew], new_entries: &mut Vec<CpioEntryNew>) {
+        cpio::sort(old_entries);
+        cpio::sort(new_entries);
+
+        let mut rm_list = vec![];
+        let mut to_back_up = vec![];
+
+        let mut old_iter = old_entries.iter().peekable();
+        let mut new_iter = new_entries.iter().peekable();
+
+        loop {
+            match (old_iter.peek(), new_iter.peek()) {
+                (Some(&old), Some(&new)) => match old.name.cmp(&new.name) {
+                    Ordering::Less => {
+                        to_back_up.push(old);
+                        old_iter.next();
+                    }
+                    Ordering::Equal => {
+                        if old.content != new.content {
+                            to_back_up.push(old);
+                        }
+                        old_iter.next();
+                        new_iter.next();
+                    }
+                    Ordering::Greater => {
+                        rm_list.extend(&new.name);
+                        rm_list.push(b'\0');
+                        new_iter.next();
+                    }
+                },
+                (Some(old), None) => {
+                    to_back_up.push(old);
+                    old_iter.next();
+                }
+                (None, Some(new)) => {
+                    rm_list.extend(&new.name);
+                    rm_list.push(b'\0');
+                    new_iter.next();
+                }
+                (None, None) => break,
+            }
+        }
+
+        // Intentially using 000 permissions to match Magisk.
+        new_entries.push(CpioEntryNew::new_directory(b".backup"));
+
+        for old_entry in to_back_up {
+            let mut new_entry = old_entry.clone();
+            new_entry.name = b".backup/".to_vec();
+            new_entry.name.extend(&old_entry.name);
+            new_entries.push(new_entry);
+        }
+
+        {
+            // Intentially using 000 permissions to match Magisk.
+            let mut entry = CpioEntryNew::new_file(b".backup/.rmlist");
+            entry.content = rm_list;
+            new_entries.push(entry);
+        }
+    }
+}
+
+impl BootImagePatcher for MagiskRootPatcher {
+    fn patch(&self, boot_image: &mut BootImage, cancel_signal: &Arc<AtomicBool>) -> Result<()> {
+        let zip_reader = File::open(&self.apk_path)?;
+        let mut zip = ZipArchive::new(BufReader::new(zip_reader))?;
+
+        // Load the first ramdisk. If it doesn't exist, we have to generate one
+        // from scratch.
+        let ramdisk = match boot_image {
+            BootImage::V0Through2(b) => Some(&b.ramdisk),
+            BootImage::V3Through4(b) => Some(&b.ramdisk),
+            BootImage::VendorV3Through4(b) => b.ramdisks.first(),
+        };
+        let (mut entries, ramdisk_format) = match ramdisk {
+            Some(r) if !r.is_empty() => load_ramdisk(r)?,
+            _ => (vec![], CompressedFormat::Lz4Legacy),
+        };
+
+        let mut old_entries = entries.clone();
+
+        // Create the Magisk directory structure.
+        for (path, perms) in [
+            (b"overlay.d".as_slice(), 0o750),
+            (b"overlay.d/sbin".as_slice(), 0o750),
+        ] {
+            let mut entry = CpioEntryNew::new_directory(path);
+            entry.mode |= perms;
+            entries.push(entry);
+        }
+
+        // Delete the original init.
+        entries.retain(|e| e.name != b"init");
+
+        // Add magiskinit.
+        {
+            let mut zip_entry = zip.by_name("lib/arm64-v8a/libmagiskinit.so")?;
+            let mut data = vec![];
+            zip_entry.read_to_end(&mut data)?;
+
+            let mut entry = CpioEntryNew::new_file(b"init");
+            entry.mode |= 0o750;
+            entry.content = data;
+            entries.push(entry);
+        }
+
+        // Add xz-compressed magisk32 and magisk64.
+        let mut xz_files = HashMap::<&str, &[u8]>::new();
+        xz_files.insert(
+            "lib/armeabi-v7a/libmagisk32.so",
+            b"overlay.d/sbin/magisk32.xz",
+        );
+        xz_files.insert(
+            "lib/arm64-v8a/libmagisk64.so",
+            b"overlay.d/sbin/magisk64.xz",
+        );
+
+        // Add stub apk, which only exists after Magisk commit
+        // ad0e6511e11ebec65aa9b5b916e1397342850319.
+        if zip.file_names().any(|n| n == "assets/stub.apk") {
+            xz_files.insert("assets/stub.apk", b"overlay.d/sbin/stub.xz");
+        }
+
+        for (source, target) in xz_files {
+            let reader = zip.by_name(source)?;
+            let raw_writer = Cursor::new(vec![]);
+            let stream = Stream::new_easy_encoder(9, Check::Crc32)?;
+            let mut writer = XzEncoder::new_stream(raw_writer, stream);
+
+            stream::copy(reader, &mut writer, cancel_signal)?;
+
+            let raw_writer = writer.finish()?;
+            let mut entry = CpioEntryNew::new_file(target);
+            entry.mode |= 0o644;
+            entry.content = raw_writer.into_inner();
+            entries.push(entry);
+        }
+
+        // Create Magisk .backup directory structure.
+        Self::apply_magisk_backup(&mut old_entries, &mut entries);
+
+        // Create Magisk config.
+        let mut magisk_config = String::new();
+        magisk_config.push_str("KEEPVERITY=true\n");
+        magisk_config.push_str("KEEPFORCEENCRYPT=true\n");
+        magisk_config.push_str("PATCHVBMETAFLAG=false\n");
+        magisk_config.push_str("RECOVERYMODE=false\n");
+
+        if Self::VER_PREINIT_DEVICE.contains(&self.version) {
+            magisk_config.push_str(&format!(
+                "PREINITDEVICE={}\n",
+                self.preinit_device.as_ref().unwrap(),
+            ));
+        }
+
+        // Magisk normally saves the original SHA1 digest in its config file. It
+        // uses this to find the original image in /data/magisk_backup_<sha1> to
+        // restore the stock boot image for uninstallation purposes. This is a
+        // feature we cannot ever use, so just use a dummy value.
+        magisk_config.push_str("SHA1=0000000000000000000000000000000000000000\n");
+
+        if Self::VER_RANDOM_SEED.contains(&self.version) {
+            magisk_config.push_str(&format!("RANDOMSEED={:#x}\n", self.random_seed));
+        }
+
+        {
+            // Intentially using 000 permissions to match Magisk.
+            let mut entry = CpioEntryNew::new_file(b".backup/.magisk");
+            entry.content = magisk_config.into_bytes();
+            entries.push(entry);
+        }
+
+        // Repack ramdisk.
+        cpio::sort(&mut entries);
+        cpio::reassign_inodes(&mut entries);
+        let new_ramdisk = save_ramdisk(&entries, ramdisk_format)?;
+
+        match boot_image {
+            BootImage::V0Through2(b) => b.ramdisk = new_ramdisk,
+            BootImage::V3Through4(b) => b.ramdisk = new_ramdisk,
+            BootImage::VendorV3Through4(b) => {
+                if b.ramdisks.is_empty() {
+                    b.ramdisks.push(new_ramdisk);
+
+                    if let Some(v4) = &mut b.v4_extra {
+                        v4.ramdisk_metas.push(RamdiskMeta {
+                            ramdisk_type: bootimage::VENDOR_RAMDISK_TYPE_NONE,
+                            ramdisk_name: String::new(),
+                            board_id: Default::default(),
+                        });
+                    }
+                } else {
+                    b.ramdisks[0] = new_ramdisk;
+                }
+            }
+        }
+
+        Ok(())
+    }
+}
+
+/// Replace the OTA certificates in the vendor_boot/recovery image with the
+/// custom OTA signing certificate.
+pub struct OtaCertPatcher {
+    cert: Certificate,
+}
+
+impl OtaCertPatcher {
+    const OTACERTS_PATH: &[u8] = b"system/etc/security/otacerts.zip";
+
+    pub fn new(cert: Certificate) -> Self {
+        Self { cert }
+    }
+
+    pub fn get_certificates(boot_image: &BootImage) -> Result<Vec<Certificate>> {
+        let mut ramdisks = vec![];
+
+        match boot_image {
+            BootImage::V0Through2(b) => ramdisks.push(&b.ramdisk),
+            BootImage::V3Through4(b) => ramdisks.push(&b.ramdisk),
+            BootImage::VendorV3Through4(b) => ramdisks.extend(b.ramdisks.iter()),
+        }
+
+        let mut certificates = vec![];
+
+        for ramdisk in ramdisks {
+            let (entries, _) = load_ramdisk(ramdisk)?;
+            let Some(entry) = entries.iter().find(|e| e.name == Self::OTACERTS_PATH) else {
+                continue;
+            };
+
+            let mut zip = ZipArchive::new(Cursor::new(&entry.content))?;
+
+            for index in 0..zip.len() {
+                let zip_entry = zip.by_index(index)?;
+                if !zip_entry.name().ends_with(".x509.pem") {
+                    continue;
+                }
+
+                let certificate = crypto::read_pem_cert(zip_entry)?;
+                certificates.push(certificate);
+            }
+        }
+
+        Ok(certificates)
+    }
+
+    fn patch_ramdisk(&self, data: &mut Vec<u8>) -> Result<bool> {
+        let (mut entries, ramdisk_format) = load_ramdisk(data)?;
+        let Some(entry) = entries.iter_mut().find(|e| e.name == Self::OTACERTS_PATH) else {
+            return Ok(false);
+        };
+
+        // Create a new otacerts archive. The old certs are ignored since
+        // flashing a stock OTA will render the device unbootable.
+        {
+            let raw_writer = Cursor::new(vec![]);
+            let mut writer = ZipWriter::new(raw_writer);
+            let options = FileOptions::default().compression_method(CompressionMethod::Stored);
+            writer.start_file("ota.x509.pem", options)?;
+
+            crypto::write_pem_cert(&mut writer, &self.cert)?;
+
+            let raw_writer = writer.finish()?;
+            entry.content = raw_writer.into_inner();
+        }
+
+        // Repack ramdisk.
+        *data = save_ramdisk(&entries, ramdisk_format)?;
+
+        Ok(true)
+    }
+}
+
+impl BootImagePatcher for OtaCertPatcher {
+    fn patch(&self, boot_image: &mut BootImage, _cancel_signal: &Arc<AtomicBool>) -> Result<()> {
+        let patched_any = match boot_image {
+            BootImage::V0Through2(b) => self.patch_ramdisk(&mut b.ramdisk)?,
+            BootImage::V3Through4(b) => self.patch_ramdisk(&mut b.ramdisk)?,
+            BootImage::VendorV3Through4(b) => {
+                let mut patched = false;
+
+                for ramdisk in &mut b.ramdisks {
+                    if self.patch_ramdisk(ramdisk)? {
+                        patched = true;
+                        break;
+                    }
+                }
+
+                patched
+            }
+        };
+
+        // Fail hard if otacerts does not exist. We don't want to lock the user
+        // out of future updates if the OTA certificate mechanism has changed.
+        if !patched_any {
+            return Err(Error::Validation(format!(
+                "No ramdisk contains {}",
+                EscapedString::new(Self::OTACERTS_PATH),
+            )));
+        }
+
+        Ok(())
+    }
+}
+
+/// Replace the boot image with a prepatched boot image if it is compatible.
+///
+/// An image is compatible if all the non-size-related header fields are
+/// identical and the set of included sections (eg. kernel, dtb) are the same.
+/// The only exception is the number of ramdisk sections, which is allowed to be
+/// higher than the original image.
+pub struct PrepatchedImagePatcher {
+    prepatched: PathBuf,
+    fatal_level: u8,
+    warning_fn: Box<dyn Fn(&str) + Send>,
+}
+
+impl PrepatchedImagePatcher {
+    const MIN_LEVEL: u8 = 0;
+    const MAX_LEVEL: u8 = 2;
+
+    // We compile without Unicode support so we have to use [0-9] instead of \d.
+    const VERSION_REGEX: &str = r"Linux version ([0-9]+\.[0-9]+).[0-9]+-(android[0-9]+)-([0-9]+)-";
+
+    pub fn new(
+        prepatched: &Path,
+        fatal_level: u8,
+        warning_fn: impl Fn(&str) + Send + 'static,
+    ) -> Self {
+        Self {
+            prepatched: prepatched.to_owned(),
+            fatal_level,
+            warning_fn: Box::new(warning_fn),
+        }
+    }
+
+    fn get_kmi_version(kernel: &[u8]) -> Result<Option<String>> {
+        let mut decompressed = vec![];
+        {
+            let raw_reader = Cursor::new(kernel);
+            let mut reader = CompressedReader::new(raw_reader, true)?;
+            reader.read_to_end(&mut decompressed)?;
+        }
+
+        let regex = Regex::new(Self::VERSION_REGEX).unwrap();
+        let Some(captures) = regex.captures(&decompressed) else {
+            return Ok(None);
+        };
+
+        let kmi_version = captures
+            .iter()
+            // Capture #0 is the entire match.
+            .skip(1)
+            .flatten()
+            .map(|c| c.as_bytes())
+            // Our regex only matches ASCII bytes.
+            .map(|c| std::str::from_utf8(c).unwrap())
+            .collect::<Vec<_>>()
+            .join("-");
+
+        Ok(Some(kmi_version))
+    }
+}
+
+impl BootImagePatcher for PrepatchedImagePatcher {
+    fn patch(&self, boot_image: &mut BootImage, _cancel_signal: &Arc<AtomicBool>) -> Result<()> {
+        let prepatched_image = {
+            let raw_reader = File::open(&self.prepatched)?;
+            BootImage::from_reader(BufReader::new(raw_reader))?
+        };
+
+        // Level 0: Warnings that don't affect booting
+        // Level 1: Warnings that may affect booting
+        // Level 2: Warnings that are very likely to affect booting
+        let mut issues = [vec![], vec![], vec![]];
+
+        macro_rules! check {
+            ($level:literal, $old:expr, $new:expr $(,)?) => {
+                let old_val = $old;
+                let new_val = $new;
+
+                if old_val != new_val {
+                    issues[$level].push(format!(
+                        "Field differs: {} ({:?}) -> {} ({:?})",
+                        stringify!($old),
+                        old_val,
+                        stringify!($new),
+                        new_val,
+                    ));
+                }
+            };
+        }
+
+        let old_kernel;
+        let new_kernel;
+
+        match (&boot_image, &prepatched_image) {
+            (BootImage::V0Through2(old), BootImage::V0Through2(new)) => {
+                check!(2, old.header_version(), new.header_version());
+                check!(2, old.kernel_addr, new.kernel_addr);
+                check!(2, old.ramdisk_addr, new.ramdisk_addr);
+                check!(2, old.second_addr, new.second_addr);
+                check!(2, old.tags_addr, new.tags_addr);
+                check!(2, old.page_size, new.page_size);
+                check!(0, old.os_version, new.os_version);
+                check!(0, &old.name, &new.name);
+                check!(1, &old.cmdline, &new.cmdline);
+                check!(0, &old.id, &new.id);
+                check!(1, &old.extra_cmdline, &new.extra_cmdline);
+                check!(2, old.kernel.is_empty(), new.kernel.is_empty());
+                check!(2, old.second.is_empty(), new.second.is_empty());
+
+                if let (Some(old_v1), Some(new_v1)) = (&old.v1_extra, &new.v1_extra) {
+                    check!(2, old_v1.recovery_dtbo_offset, new_v1.recovery_dtbo_offset);
+                    check!(
+                        2,
+                        old_v1.recovery_dtbo.is_empty(),
+                        new_v1.recovery_dtbo.is_empty(),
+                    );
+                }
+
+                if let (Some(old_v2), Some(new_v2)) = (&old.v2_extra, &new.v2_extra) {
+                    check!(2, old_v2.dtb_addr, new_v2.dtb_addr);
+                    check!(2, old_v2.dtb.is_empty(), new_v2.dtb.is_empty());
+                }
+
+                // We allow adding a ramdisk.
+                if !old.ramdisk.is_empty() || new.ramdisk.is_empty() {
+                    check!(2, old.ramdisk.is_empty(), new.ramdisk.is_empty());
+                }
+
+                old_kernel = if old.kernel.is_empty() {
+                    None
+                } else {
+                    Some(&old.kernel)
+                };
+                new_kernel = if new.kernel.is_empty() {
+                    None
+                } else {
+                    Some(&new.kernel)
+                };
+            }
+            (BootImage::V3Through4(old), BootImage::V3Through4(new)) => {
+                check!(2, old.header_version(), new.header_version());
+                check!(0, old.os_version, new.os_version);
+                check!(0, old.reserved, new.reserved);
+                check!(1, &old.cmdline, &new.cmdline);
+                check!(2, old.kernel.is_empty(), new.kernel.is_empty());
+
+                // We allow adding a ramdisk.
+                if !old.ramdisk.is_empty() || new.ramdisk.is_empty() {
+                    check!(2, old.ramdisk.is_empty(), new.ramdisk.is_empty());
+                }
+
+                old_kernel = if old.kernel.is_empty() {
+                    None
+                } else {
+                    Some(&old.kernel)
+                };
+                new_kernel = if new.kernel.is_empty() {
+                    None
+                } else {
+                    Some(&new.kernel)
+                };
+            }
+            (BootImage::VendorV3Through4(old), BootImage::VendorV3Through4(new)) => {
+                check!(2, old.page_size, new.page_size);
+                check!(2, old.kernel_addr, new.kernel_addr);
+                check!(2, old.ramdisk_addr, new.ramdisk_addr);
+                check!(1, &old.cmdline, &new.cmdline);
+                check!(2, old.tags_addr, new.tags_addr);
+                check!(0, &old.name, &new.name);
+                check!(2, old.dtb.is_empty(), new.dtb.is_empty());
+                check!(2, old.dtb_addr, new.dtb_addr);
+                check!(2, old.ramdisks.len(), new.ramdisks.len());
+
+                if let (Some(old_v4), Some(new_v4)) = (&old.v4_extra, &new.v4_extra) {
+                    check!(2, &old_v4.ramdisk_metas, &new_v4.ramdisk_metas);
+                    check!(2, &old_v4.bootconfig, &new_v4.bootconfig);
+                }
+
+                old_kernel = None;
+                new_kernel = None;
+            }
+            _ => {
+                return Err(Error::Validation(
+                    "Boot image and prepatched image are different boot image types".to_owned(),
+                ));
+            }
+        }
+
+        if let (Some(old), Some(new)) = (old_kernel, new_kernel) {
+            let old_kmi_version = Self::get_kmi_version(old)?;
+            let new_kmi_version = Self::get_kmi_version(new)?;
+
+            check!(2, old_kmi_version, new_kmi_version);
+        }
+
+        let mut warnings = vec![];
+        let mut errors = vec![];
+
+        for level in Self::MIN_LEVEL..self.fatal_level {
+            warnings.extend(&issues[level as usize]);
+        }
+        for level in self.fatal_level..=Self::MAX_LEVEL {
+            errors.extend(&issues[level as usize]);
+        }
+
+        if !warnings.is_empty() {
+            let mut msg =
+                "The prepatched boot image may not be compatible with the original:".to_owned();
+            for warning in warnings {
+                msg.push_str("\n- ");
+                msg.push_str(warning);
+            }
+
+            (self.warning_fn)(&msg);
+        }
+
+        if !errors.is_empty() {
+            let mut msg =
+                "The prepatched boot image is not compatible with the original:".to_owned();
+            for error in errors {
+                msg.push_str("\n- ");
+                msg.push_str(error);
+            }
+
+            return Err(Error::Validation(msg));
+        }
+
+        *boot_image = prepatched_image;
+
+        Ok(())
+    }
+}
+
+/// Run each patcher against the boot image with the vbmeta footer stripped off
+/// and then re-sign the image.
+pub fn patch_boot(
+    mut reader: impl Read + Seek,
+    writer: impl Write + Seek,
+    key: &RsaPrivateKey,
+    patchers: &[Box<dyn BootImagePatcher + Send>],
+    cancel_signal: &Arc<AtomicBool>,
+) -> Result<()> {
+    let (mut header, footer, image_size) = avb::load_image(&mut reader)?;
+    let Some(footer) = footer else {
+        return Err(Error::NoFooter);
+    };
+
+    let section_reader = SectionReader::new(reader, 0, footer.original_image_size)?;
+    let mut boot_image = BootImage::from_reader(section_reader)?;
+
+    for patcher in patchers {
+        patcher.patch(&mut boot_image, cancel_signal)?;
+    }
+
+    let mut descriptor_iter = header.descriptors.iter_mut().filter_map(|d| {
+        if let Descriptor::Hash(h) = d {
+            Some(h)
+        } else {
+            None
+        }
+    });
+
+    let Some(descriptor) = descriptor_iter.next() else {
+        return Err(Error::NoHashDescriptor);
+    };
+
+    // Write new boot image. We reuse the existing salt for the digest.
+    let mut context = Context::new(&ring::digest::SHA256);
+    context.update(&descriptor.salt);
+    let mut hashing_writer = HashingWriter::new(writer, context);
+    boot_image.to_writer(&mut hashing_writer)?;
+    let (mut writer, context) = hashing_writer.finish();
+
+    header.algorithm_type = AlgorithmType::Sha256Rsa4096;
+
+    descriptor.image_size = writer.stream_position()?;
+    descriptor.hash_algorithm = "sha256".to_owned();
+    descriptor.root_digest = context.finish().as_ref().to_vec();
+
+    if descriptor_iter.next().is_some() {
+        return Err(Error::MultipleHashDescriptors);
+    }
+
+    if !header.public_key.is_empty() {
+        header.sign(key)?;
+    }
+
+    avb::write_appended_image(writer, &header, &footer, image_size)?;
+
+    Ok(())
+}

avbroot/src/cli/args.rs

@@ -0,0 +1,51 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::sync::{atomic::AtomicBool, Arc};
+
+use anyhow::Result;
+use clap::{Parser, Subcommand};
+
+use crate::cli::{avb, boot, completion, key, ota, ramdisk};
+
+#[allow(clippy::large_enum_variant)]
+#[derive(Debug, Subcommand)]
+pub enum Command {
+    Avb(avb::AvbCli),
+    Boot(boot::BootCli),
+    Completion(completion::CompletionCli),
+    Key(key::KeyCli),
+    Ota(ota::OtaCli),
+    Ramdisk(ramdisk::RamdiskCli),
+    /// (Deprecated: Use `avbroot ota patch` instead.)
+    Patch(ota::PatchCli),
+    /// (Deprecated: Use `avbroot ota extract` instead.)
+    Extract(ota::ExtractCli),
+    /// (Deprecated: Use `avbroot boot magisk-info` instead.)
+    MagiskInfo(boot::MagiskInfoCli),
+}
+
+#[derive(Debug, Parser)]
+pub struct Cli {
+    #[command(subcommand)]
+    pub command: Command,
+}
+
+pub fn main(cancel_signal: &Arc<AtomicBool>) -> Result<()> {
+    let cli = Cli::parse();
+
+    match cli.command {
+        Command::Avb(c) => avb::avb_main(&c, cancel_signal),
+        Command::Boot(c) => boot::boot_main(&c),
+        Command::Completion(c) => completion::completion_main(&c),
+        Command::Key(c) => key::key_main(&c),
+        Command::Ota(c) => ota::ota_main(&c, cancel_signal),
+        Command::Ramdisk(c) => ramdisk::ramdisk_main(&c),
+        // Deprecated aliases.
+        Command::Patch(c) => ota::patch_subcommand(&c, cancel_signal),
+        Command::Extract(c) => ota::extract_subcommand(&c, cancel_signal),
+        Command::MagiskInfo(c) => boot::magisk_info_subcommand(&c),
+    }
+}

avbroot/src/cli/avb.rs

@@ -0,0 +1,237 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::{
+    collections::{HashMap, HashSet},
+    ffi::OsStr,
+    fs::{self, File},
+    io::{self, BufReader},
+    path::{Path, PathBuf},
+    str,
+    sync::{atomic::AtomicBool, Arc},
+};
+
+use anyhow::{anyhow, bail, Context, Result};
+use clap::{Parser, Subcommand};
+use rayon::prelude::{IntoParallelRefIterator, ParallelIterator};
+use rsa::RsaPublicKey;
+
+use crate::{
+    cli::{status, warning},
+    format::avb::{self, Descriptor},
+    stream::PSeekFile,
+};
+
+fn ensure_name_is_safe(name: &str) -> Result<()> {
+    if Path::new(name).file_name() != Some(OsStr::new(name)) {
+        bail!("Unsafe partition name: {name}");
+    }
+
+    Ok(())
+}
+
+/// Recursively verify an image's vbmeta header and all of the chained images.
+/// `seen` is used to prevent cycles. `descriptors` will contain all of the hash
+/// and hashtree descriptors that need to be verified.
+pub fn verify_headers(
+    directory: &Path,
+    name: &str,
+    expected_key: Option<&RsaPublicKey>,
+    seen: &mut HashSet<String>,
+    descriptors: &mut HashMap<String, Descriptor>,
+) -> Result<()> {
+    if !seen.insert(name.to_owned()) {
+        return Ok(());
+    }
+
+    ensure_name_is_safe(name)?;
+
+    let path = directory.join(format!("{name}.img"));
+    let raw_reader =
+        File::open(&path).with_context(|| anyhow!("Failed to open for reading: {path:?}"))?;
+    let (header, _, _) = avb::load_image(BufReader::new(raw_reader))
+        .with_context(|| anyhow!("Failed to load vbmeta structures: {path:?}"))?;
+
+    // Verify the header's signature.
+    let public_key = header
+        .verify()
+        .with_context(|| anyhow!("Failed to verify header signature: {path:?}"))?;
+
+    if let Some(k) = &public_key {
+        let prefix = format!("{name} has a signed vbmeta header");
+
+        if let Some(e) = expected_key {
+            if k == e {
+                status!("{prefix}");
+            } else {
+                bail!("{prefix}, but is signed by an untrusted key");
+            }
+        } else {
+            warning!("{prefix}, but parent does not list a trusted key");
+        }
+    } else {
+        status!("{name} has an unsigned vbmeta header");
+    }
+
+    for descriptor in &header.descriptors {
+        let Some(target_name) = descriptor.partition_name() else {
+            continue;
+        };
+
+        match descriptor {
+            avb::Descriptor::Hashtree(_) | avb::Descriptor::Hash(_) => {
+                if let Some(prev) = descriptors.get(target_name) {
+                    if prev != descriptor {
+                        bail!("{name} descriptor does not match previous encounter");
+                    }
+                } else {
+                    descriptors.insert(target_name.to_owned(), descriptor.clone());
+                }
+            }
+            avb::Descriptor::ChainPartition(d) => {
+                let target_key = avb::decode_public_key(&d.public_key).with_context(|| {
+                    anyhow!("Failed to decode chained public key for: {target_name}")
+                })?;
+
+                verify_headers(directory, target_name, Some(&target_key), seen, descriptors)?;
+            }
+            _ => {}
+        }
+    }
+
+    Ok(())
+}
+
+pub fn verify_descriptors(
+    directory: &Path,
+    descriptors: &HashMap<String, Descriptor>,
+    cancel_signal: &Arc<AtomicBool>,
+) -> Result<()> {
+    descriptors
+        .par_iter()
+        .map(|(name, descriptor)| {
+            let path = directory.join(format!("{name}.img"));
+            let reader = match File::open(&path).map(PSeekFile::new) {
+                Ok(f) => f,
+                // Some devices, like bluejay, have vbmeta descriptors that
+                // refer to partitions that exist on the device, but not in the
+                // OTA.
+                Err(e) if e.kind() == io::ErrorKind::NotFound => {
+                    warning!("Partition image does not exist: {path:?}");
+                    return Ok(());
+                }
+                Err(e) => {
+                    Err(e).with_context(|| format!("Failed to open for reading: {path:?}"))?
+                }
+            };
+
+            match descriptor {
+                Descriptor::Hashtree(d) => {
+                    status!("Verifying hashtree descriptor for: {name}");
+                    d.verify(
+                        || Ok(Box::new(BufReader::new(reader.clone()))),
+                        cancel_signal,
+                    )
+                    .with_context(|| anyhow!("Failed to verify hashtree descriptor for: {name}"))?;
+                }
+                Descriptor::Hash(d) => {
+                    status!("Verifying hash descriptor for: {name}");
+                    d.verify(BufReader::new(reader), cancel_signal)
+                        .with_context(|| anyhow!("Failed to verify hash descriptor for: {name}"))?;
+                }
+                _ => unreachable!("Non-verifiable descriptor: {descriptor:?}"),
+            }
+
+            Ok(())
+        })
+        .collect()
+}
+
+pub fn avb_main(cli: &AvbCli, cancel_signal: &Arc<AtomicBool>) -> Result<()> {
+    match &cli.command {
+        AvbCommand::Dump(c) => {
+            let raw_reader = File::open(&c.input)
+                .with_context(|| anyhow!("Failed to open for reading: {:?}", c.input))?;
+            let reader = BufReader::new(raw_reader);
+            let (header, footer, image_size) = avb::load_image(reader)
+                .with_context(|| anyhow!("Failed to load vbmeta structures: {:?}", c.input))?;
+
+            println!("Image size: {image_size}");
+            println!("Header: {header:#?}");
+            println!("Footer: {footer:#?}");
+        }
+        AvbCommand::Verify(c) => {
+            let public_key = if let Some(p) = &c.public_key {
+                let data = fs::read(p).with_context(|| anyhow!("Failed to read file: {p:?}"))?;
+                let key = avb::decode_public_key(&data)
+                    .with_context(|| anyhow!("Failed to decode public key: {p:?}"))?;
+
+                Some(key)
+            } else {
+                None
+            };
+
+            let directory = c.input.parent().unwrap_or_else(|| Path::new("."));
+            let name = c
+                .input
+                .file_stem()
+                .with_context(|| anyhow!("Path is not a file: {:?}", c.input))?
+                .to_str()
+                .ok_or_else(|| anyhow!("Invalid UTF-8: {:?}", c.input))?;
+
+            let mut seen = HashSet::<String>::new();
+            let mut descriptors = HashMap::<String, Descriptor>::new();
+
+            verify_headers(
+                directory,
+                name,
+                public_key.as_ref(),
+                &mut seen,
+                &mut descriptors,
+            )?;
+            verify_descriptors(directory, &descriptors, cancel_signal)?;
+
+            status!("Successfully verified all vbmeta signatures and hashes");
+        }
+    }
+
+    Ok(())
+}
+
+/// Dump AVB header and footer information.
+#[derive(Debug, Parser)]
+struct DumpCli {
+    /// Path to input image.
+    #[arg(short, long, value_name = "FILE", value_parser)]
+    input: PathBuf,
+}
+
+/// Verify vbmeta signatures.
+#[derive(Debug, Parser)]
+struct VerifyCli {
+    /// Path to input image.
+    #[arg(short, long, value_name = "FILE", value_parser)]
+    input: PathBuf,
+
+    /// Path to public key in AVB binary format.
+    ///
+    /// If this is not specified, the signatures can only be checked for
+    /// validity, not whether they are trusted.
+    #[arg(short, long, value_name = "FILE", value_parser)]
+    public_key: Option<PathBuf>,
+}
+
+#[derive(Debug, Subcommand)]
+enum AvbCommand {
+    Dump(DumpCli),
+    Verify(VerifyCli),
+}
+
+/// Show information about AVB-protected images.
+#[derive(Debug, Parser)]
+pub struct AvbCli {
+    #[command(subcommand)]
+    command: AvbCommand,
+}

avbroot/src/cli/boot.rs

@@ -0,0 +1,493 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::{
+    fs::{self, File},
+    io::{self, BufReader, BufWriter, Cursor, Write},
+    path::{Path, PathBuf},
+};
+
+use anyhow::{anyhow, bail, Context, Result};
+use clap::{Parser, Subcommand};
+
+use crate::{
+    format::{avb::Header, bootimage::BootImage, compression::CompressedReader, cpio},
+    stream::{FromReader, ToWriter},
+};
+
+fn read_image(path: &Path) -> Result<BootImage> {
+    let file = File::open(path).with_context(|| format!("Failed to open for reading: {path:?}"))?;
+    let reader = BufReader::new(file);
+    let image = BootImage::from_reader(reader)
+        .with_context(|| format!("Failed to read boot image: {path:?}"))?;
+
+    Ok(image)
+}
+
+fn write_image(path: &Path, image: &BootImage) -> Result<()> {
+    let file =
+        File::create(path).with_context(|| format!("Failed to open for writing: {path:?}"))?;
+    let mut writer = BufWriter::new(file);
+    image
+        .to_writer(&mut writer)
+        .with_context(|| format!("Failed to write boot image: {path:?}"))?;
+    writer.flush()?;
+
+    Ok(())
+}
+
+fn read_header(path: &Path) -> Result<BootImage> {
+    let data = fs::read_to_string(path)
+        .with_context(|| format!("Failed to read header TOML: {path:?}"))?;
+    let image = toml_edit::de::from_str(&data)
+        .with_context(|| format!("Failed to parse header TOML: {path:?}"))?;
+
+    Ok(image)
+}
+
+fn write_header(path: &Path, image: &BootImage) -> Result<()> {
+    let data = toml_edit::ser::to_string_pretty(image)
+        .with_context(|| format!("Failed to serialize header TOML: {path:?}"))?;
+    fs::write(path, data).with_context(|| format!("Failed to write header TOML: {path:?}"))?;
+
+    Ok(())
+}
+
+fn read_data_if_exists(path: &Path) -> Result<Option<Vec<u8>>> {
+    let data = match fs::read(path) {
+        Ok(f) => f,
+        Err(e) if e.kind() == io::ErrorKind::NotFound => return Ok(None),
+        Err(e) => Err(e).with_context(|| format!("Failed to read data: {path:?}"))?,
+    };
+
+    Ok(Some(data))
+}
+
+fn read_text_if_exists(path: &Path) -> Result<Option<String>> {
+    let data = match fs::read_to_string(path) {
+        Ok(f) => f,
+        Err(e) if e.kind() == io::ErrorKind::NotFound => return Ok(None),
+        Err(e) => Err(e).with_context(|| format!("Failed to read text: {path:?}"))?,
+    };
+
+    Ok(Some(data))
+}
+
+fn read_avb_header_if_exists(path: &Path) -> Result<Option<Header>> {
+    let file = match File::open(path) {
+        Ok(f) => f,
+        Err(e) if e.kind() == io::ErrorKind::NotFound => return Ok(None),
+        Err(e) => Err(e).with_context(|| format!("Failed to open for reading: {path:?}"))?,
+    };
+    let header = Header::from_reader(BufReader::new(file))
+        .with_context(|| anyhow!("Failed to read vbmeta header: {path:?}"))?;
+
+    Ok(Some(header))
+}
+
+fn write_data_if_not_empty(path: &Path, data: &[u8]) -> Result<()> {
+    if !data.is_empty() {
+        fs::write(path, data).with_context(|| format!("Failed to write data: {path:?}"))?;
+    }
+
+    Ok(())
+}
+
+fn write_text_if_not_empty(path: &Path, text: &str) -> Result<()> {
+    if !text.is_empty() {
+        fs::write(path, text.as_bytes())
+            .with_context(|| format!("Failed to write text: {path:?}"))?;
+    }
+
+    Ok(())
+}
+
+fn write_avb_header(path: &Path, header: &Header) -> Result<()> {
+    let file =
+        File::create(path).with_context(|| anyhow!("Failed to open for writing: {path:?}"))?;
+    header.to_writer(BufWriter::new(file))?;
+
+    Ok(())
+}
+
+fn display_info(cli: &BootCli, image: &BootImage) {
+    if !cli.quiet {
+        if cli.debug {
+            println!("{image:#?}");
+        } else {
+            println!("{image}");
+        }
+    }
+}
+
+fn unpack_subcommand(boot_cli: &BootCli, cli: &UnpackCli) -> Result<()> {
+    let image = read_image(&cli.input)?;
+    display_info(boot_cli, &image);
+
+    write_header(&cli.output_header, &image)?;
+
+    let mut kernel = None;
+    let mut second = None;
+    let mut recovery_dtbo = None;
+    let mut dtb = None;
+    let mut vts_signature = None;
+    let mut bootconfig = None;
+    let mut ramdisks = vec![];
+
+    match &image {
+        BootImage::V0Through2(b) => {
+            kernel = Some(&b.kernel);
+            second = Some(&b.second);
+            if let Some(v1) = &b.v1_extra {
+                recovery_dtbo = Some(&v1.recovery_dtbo);
+            }
+            if let Some(v2) = &b.v2_extra {
+                dtb = Some(&v2.dtb);
+            }
+            ramdisks.push(&b.ramdisk);
+        }
+        BootImage::V3Through4(b) => {
+            kernel = Some(&b.kernel);
+            if let Some(v4) = &b.v4_extra {
+                vts_signature = v4.signature.as_ref();
+            }
+            ramdisks.push(&b.ramdisk);
+        }
+        BootImage::VendorV3Through4(b) => {
+            dtb = Some(&b.dtb);
+            if let Some(v4) = &b.v4_extra {
+                bootconfig = Some(&v4.bootconfig);
+            }
+            ramdisks.extend(b.ramdisks.iter());
+        }
+    }
+
+    if let Some(data) = kernel {
+        write_data_if_not_empty(&cli.output_kernel, data)?;
+    }
+    if let Some(data) = second {
+        write_data_if_not_empty(&cli.output_second, data)?;
+    }
+    if let Some(data) = recovery_dtbo {
+        write_data_if_not_empty(&cli.output_recovery_dtbo, data)?;
+    }
+    if let Some(data) = dtb {
+        write_data_if_not_empty(&cli.output_dtb, data)?;
+    }
+    if let Some(header) = vts_signature {
+        write_avb_header(&cli.output_vts_signature, header)?;
+    }
+    if let Some(text) = bootconfig {
+        write_text_if_not_empty(&cli.output_bootconfig, text)?;
+    }
+
+    for (i, data) in ramdisks.iter().enumerate() {
+        let mut path = cli.output_ramdisk_prefix.as_os_str().to_owned();
+        path.push(i.to_string());
+
+        write_data_if_not_empty(Path::new(&path), data)?;
+    }
+
+    Ok(())
+}
+
+fn pack_subcommand(boot_cli: &BootCli, cli: &PackCli) -> Result<()> {
+    let mut image = read_header(&cli.input_header)?;
+
+    let kernel = read_data_if_exists(&cli.input_kernel)?;
+    let second = read_data_if_exists(&cli.input_second)?;
+    let recovery_dtbo = read_data_if_exists(&cli.input_recovery_dtbo)?;
+    let dtb = read_data_if_exists(&cli.input_dtb)?;
+    let vts_signature = read_avb_header_if_exists(&cli.input_vts_signature)?;
+    let bootconfig = read_text_if_exists(&cli.input_bootconfig)?;
+    let mut ramdisks = vec![];
+
+    for i in 0.. {
+        let mut path = cli.input_ramdisk_prefix.as_os_str().to_owned();
+        path.push(i.to_string());
+
+        let Some(ramdisk) = read_data_if_exists(Path::new(&path))? else {
+            break;
+        };
+
+        ramdisks.push(ramdisk);
+    }
+
+    match &mut image {
+        BootImage::V0Through2(b) => {
+            b.kernel = kernel.unwrap_or_default();
+            b.second = second.unwrap_or_default();
+            if let Some(v1) = &mut b.v1_extra {
+                v1.recovery_dtbo = recovery_dtbo.unwrap_or_default();
+            }
+            if let Some(v2) = &mut b.v2_extra {
+                v2.dtb = dtb.unwrap_or_default();
+            }
+            if ramdisks.len() > 1 {
+                bail!("Image type only supports a single ramdisk");
+            }
+            b.ramdisk = ramdisks.into_iter().next().unwrap_or_default();
+        }
+        BootImage::V3Through4(b) => {
+            b.kernel = kernel.unwrap_or_default();
+            if let Some(v4) = &mut b.v4_extra {
+                v4.signature = vts_signature;
+            }
+            if ramdisks.len() > 1 {
+                bail!("Image type only supports a single ramdisk");
+            }
+            b.ramdisk = ramdisks.into_iter().next().unwrap_or_default();
+        }
+        BootImage::VendorV3Through4(b) => {
+            b.dtb = dtb.unwrap_or_default();
+            if let Some(v4) = &mut b.v4_extra {
+                v4.bootconfig = bootconfig.unwrap_or_default();
+            }
+            b.ramdisks = ramdisks;
+        }
+    }
+
+    display_info(boot_cli, &image);
+    write_image(&cli.output, &image)?;
+
+    Ok(())
+}
+
+fn repack_subcommand(boot_cli: &BootCli, cli: &RepackCli) -> Result<()> {
+    let image = read_image(&cli.input)?;
+    display_info(boot_cli, &image);
+    write_image(&cli.output, &image)?;
+
+    Ok(())
+}
+
+fn info_subcommand(boot_cli: &BootCli, cli: &InfoCli) -> Result<()> {
+    let image = read_image(&cli.input)?;
+    display_info(boot_cli, &image);
+
+    Ok(())
+}
+
+pub fn magisk_info_subcommand(cli: &MagiskInfoCli) -> Result<()> {
+    let raw_reader = File::open(&cli.image)
+        .with_context(|| anyhow!("Failed to open for reading: {:?}", cli.image))?;
+    let boot_image = BootImage::from_reader(BufReader::new(raw_reader))
+        .with_context(|| anyhow!("Failed to load boot image: {:?}", cli.image))?;
+
+    let mut ramdisks = vec![];
+
+    match &boot_image {
+        BootImage::V0Through2(b) => {
+            if !b.ramdisk.is_empty() {
+                ramdisks.push(&b.ramdisk);
+            }
+        }
+        BootImage::V3Through4(b) => {
+            if !b.ramdisk.is_empty() {
+                ramdisks.push(&b.ramdisk);
+            }
+        }
+        BootImage::VendorV3Through4(b) => {
+            ramdisks.extend(b.ramdisks.iter());
+        }
+    }
+
+    for (i, ramdisk) in ramdisks.iter().enumerate() {
+        let reader = Cursor::new(ramdisk);
+        let reader = CompressedReader::new(reader, true)
+            .with_context(|| anyhow!("Failed to load ramdisk #{i}"))?;
+        let entries = cpio::load(reader, false)
+            .with_context(|| anyhow!("Failed to load ramdisk #{i} cpio"))?;
+
+        if let Some(e) = entries.iter().find(|e| e.name == b".backup/.magisk") {
+            io::stdout().write_all(&e.content)?;
+            return Ok(());
+        }
+    }
+
+    bail!("Not a Magisk-patched boot image");
+}
+
+pub fn boot_main(cli: &BootCli) -> Result<()> {
+    match &cli.command {
+        BootCommand::Unpack(c) => unpack_subcommand(cli, c),
+        BootCommand::Pack(c) => pack_subcommand(cli, c),
+        BootCommand::Repack(c) => repack_subcommand(cli, c),
+        BootCommand::Info(c) => info_subcommand(cli, c),
+        BootCommand::MagiskInfo(c) => magisk_info_subcommand(c),
+    }
+}
+
+/// Unpack a boot image.
+#[derive(Debug, Parser)]
+struct UnpackCli {
+    /// Path to input boot image.
+    #[arg(short, long, value_name = "FILE", value_parser)]
+    input: PathBuf,
+
+    /// Path to output header TOML.
+    #[arg(long, value_name = "FILE", value_parser, default_value = "header.toml")]
+    output_header: PathBuf,
+
+    /// Path to output kernel image.
+    #[arg(long, value_name = "FILE", value_parser, default_value = "kernel.img")]
+    output_kernel: PathBuf,
+
+    /// Path prefix for output ramdisk images.
+    #[arg(
+        long,
+        value_name = "FILE",
+        value_parser,
+        default_value = "ramdisk.img."
+    )]
+    output_ramdisk_prefix: PathBuf,
+
+    /// Path to output second stage bootloader image.
+    #[arg(long, value_name = "FILE", value_parser, default_value = "second.img")]
+    output_second: PathBuf,
+
+    /// Path to output recovery dtbo/acpio image.
+    #[arg(
+        long,
+        value_name = "FILE",
+        value_parser,
+        default_value = "recovery_dtbo.img"
+    )]
+    output_recovery_dtbo: PathBuf,
+
+    /// Path to output device tree blob image.
+    #[arg(long, value_name = "FILE", value_parser, default_value = "dtb.img")]
+    output_dtb: PathBuf,
+
+    /// Path to output VTS signature.
+    #[arg(
+        long,
+        value_name = "FILE",
+        value_parser,
+        default_value = "vts_signature.img"
+    )]
+    output_vts_signature: PathBuf,
+
+    /// Path to output bootconfig text.
+    #[arg(
+        long,
+        value_name = "FILE",
+        value_parser,
+        default_value = "bootconfig.txt"
+    )]
+    output_bootconfig: PathBuf,
+}
+
+/// Pack a boot image.
+#[derive(Debug, Parser)]
+struct PackCli {
+    /// Path to output boot image.
+    #[arg(short, long, value_name = "FILE", value_parser)]
+    output: PathBuf,
+
+    /// Path to input header TOML.
+    #[arg(long, value_name = "FILE", value_parser, default_value = "header.toml")]
+    input_header: PathBuf,
+
+    /// Path to input kernel image.
+    #[arg(long, value_name = "FILE", value_parser, default_value = "kernel.img")]
+    input_kernel: PathBuf,
+
+    /// Path prefix for input ramdisk images.
+    #[arg(
+        long,
+        value_name = "FILE",
+        value_parser,
+        default_value = "ramdisk.img."
+    )]
+    input_ramdisk_prefix: PathBuf,
+
+    /// Path to input second stage bootloader image.
+    #[arg(long, value_name = "FILE", value_parser, default_value = "second.img")]
+    input_second: PathBuf,
+
+    /// Path to input recovery dtbo/acpio image.
+    #[arg(
+        long,
+        value_name = "FILE",
+        value_parser,
+        default_value = "recovery_dtbo.img"
+    )]
+    input_recovery_dtbo: PathBuf,
+
+    /// Path to input device tree blob image.
+    #[arg(long, value_name = "FILE", value_parser, default_value = "dtb.img")]
+    input_dtb: PathBuf,
+
+    /// Path to input VTS signature.
+    #[arg(
+        long,
+        value_name = "FILE",
+        value_parser,
+        default_value = "vts_signature.img"
+    )]
+    input_vts_signature: PathBuf,
+
+    /// Path to input bootconfig text.
+    #[arg(
+        long,
+        value_name = "FILE",
+        value_parser,
+        default_value = "bootconfig.txt"
+    )]
+    input_bootconfig: PathBuf,
+}
+
+/// Repack a boot image.
+#[derive(Debug, Parser)]
+struct RepackCli {
+    /// Path to input boot image.
+    #[arg(short, long, value_name = "FILE", value_parser)]
+    input: PathBuf,
+
+    /// Path to output boot image.
+    #[arg(short, long, value_name = "FILE", value_parser)]
+    output: PathBuf,
+}
+
+/// Display boot image header information.
+#[derive(Debug, Parser)]
+struct InfoCli {
+    /// Path to input boot image.
+    #[arg(short, long, value_name = "FILE", value_parser)]
+    input: PathBuf,
+}
+
+/// Print Magisk config from a patched boot image.
+#[derive(Debug, Parser)]
+pub struct MagiskInfoCli {
+    /// Path to Magisk-patched boot image.
+    #[arg(short, long, value_name = "FILE", value_parser)]
+    pub image: PathBuf,
+}
+
+#[derive(Debug, Subcommand)]
+enum BootCommand {
+    Unpack(UnpackCli),
+    Pack(PackCli),
+    Repack(RepackCli),
+    Info(InfoCli),
+    MagiskInfo(MagiskInfoCli),
+}
+
+/// Pack or unpack boot images.
+#[derive(Debug, Parser)]
+pub struct BootCli {
+    #[command(subcommand)]
+    command: BootCommand,
+
+    /// Don't print boot image header information.
+    #[arg(short, long, global = true)]
+    quiet: bool,
+
+    /// Print boot image header information in debug format.
+    #[arg(short, long, global = true)]
+    debug: bool,
+}

avbroot/src/cli/completion.rs

@@ -0,0 +1,31 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::io;
+
+use anyhow::Result;
+use clap::{CommandFactory, Parser};
+use clap_complete::Shell;
+
+use crate::cli::args::Cli;
+
+pub fn completion_main(cli: &CompletionCli) -> Result<()> {
+    clap_complete::generate(
+        cli.shell,
+        &mut Cli::command(),
+        env!("CARGO_PKG_NAME"),
+        &mut io::stdout(),
+    );
+
+    Ok(())
+}
+
+/// Generate shell tab completion configs.
+#[derive(Debug, Parser)]
+pub struct CompletionCli {
+    /// The shell to generate completions for.
+    #[arg(short, long, value_name = "SHELL", value_parser)]
+    pub shell: Shell,
+}

avbroot/src/cli/key.rs

@@ -0,0 +1,168 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::{
+    ffi::OsString,
+    fs,
+    path::{Path, PathBuf},
+    time::Duration,
+};
+
+use anyhow::{anyhow, Context, Result};
+use clap::{Args, Parser, Subcommand};
+
+use crate::{
+    crypto::{self, PassphraseSource},
+    format::avb,
+};
+
+fn get_passphrase(group: &PassphraseGroup, key_path: &Path) -> PassphraseSource {
+    if let Some(v) = &group.pass_env_var {
+        PassphraseSource::EnvVar(v.clone())
+    } else if let Some(p) = &group.pass_file {
+        PassphraseSource::File(p.clone())
+    } else {
+        PassphraseSource::Prompt(format!("Enter passphrase for {key_path:?}: "))
+    }
+}
+
+pub fn key_main(cli: &KeyCli) -> Result<()> {
+    match &cli.command {
+        KeyCommand::GenerateKey(c) => {
+            let passphrase = get_passphrase(&c.passphrase, &c.output);
+            let private_key =
+                crypto::generate_rsa_key_pair().context("Failed to generate RSA keypair")?;
+
+            crypto::write_pem_key_file(&c.output, &private_key, &passphrase)
+                .with_context(|| anyhow!("Failed to write private key: {:?}", c.output))?;
+        }
+        KeyCommand::GenerateCert(c) => {
+            let passphrase = get_passphrase(&c.passphrase, &c.key);
+            let private_key = crypto::read_pem_key_file(&c.key, &passphrase)
+                .with_context(|| anyhow!("Failed to load key: {:?}", c.key))?;
+
+            let validity = Duration::from_secs(c.validity * 24 * 60 * 60);
+            let cert = crypto::generate_cert(&private_key, rand::random(), validity, &c.subject)
+                .context("Failed to generate certificate")?;
+
+            crypto::write_pem_cert_file(&c.output, &cert)
+                .with_context(|| anyhow!("Failed to write certificate: {:?}", c.output))?;
+        }
+        KeyCommand::ExtractAvb(c) => {
+            let public_key = if let Some(p) = &c.input.key {
+                let passphrase = get_passphrase(&c.passphrase, p);
+                let private_key = crypto::read_pem_key_file(p, &passphrase)
+                    .with_context(|| anyhow!("Failed to load key: {p:?}"))?;
+
+                private_key.to_public_key()
+            } else if let Some(p) = &c.input.cert {
+                let certificate = crypto::read_pem_cert_file(p)
+                    .with_context(|| anyhow!("Failed to load certificate: {p:?}"))?;
+
+                crypto::get_public_key(&certificate)?
+            } else {
+                unreachable!()
+            };
+
+            let encoded = avb::encode_public_key(&public_key)
+                .with_context(|| anyhow!("Failed to encode public key in AVB format"))?;
+
+            fs::write(&c.output, encoded)
+                .with_context(|| anyhow!("Failed to write public key: {:?}", c.output))?;
+        }
+    }
+
+    Ok(())
+}
+
+#[derive(Debug, Args)]
+#[group(required = true, multiple = false)]
+struct PublicKeyInputGroup {
+    /// Path to private key.
+    #[arg(short, long, value_name = "FILE", value_parser)]
+    key: Option<PathBuf>,
+
+    /// Path to certificate.
+    #[arg(short, long, value_name = "FILE", value_parser, conflicts_with_all = ["pass_env_var", "pass_file"])]
+    cert: Option<PathBuf>,
+}
+
+#[derive(Debug, Args)]
+struct PassphraseGroup {
+    /// Environment variable containing private key passphrase.
+    #[arg(long, value_name = "ENV_VAR", value_parser, group = "pass")]
+    pass_env_var: Option<OsString>,
+
+    /// File containing private key passphrase.
+    #[arg(long, value_name = "FILE", value_parser, group = "pass")]
+    pass_file: Option<PathBuf>,
+}
+
+/// Generate an 4096-bit RSA keypair.
+///
+/// The output is saved in the standard PKCS8 format.
+#[derive(Debug, Parser)]
+struct GenerateKeyCli {
+    /// Path to output private key.
+    #[arg(short, long, value_name = "FILE", value_parser)]
+    output: PathBuf,
+
+    #[command(flatten)]
+    passphrase: PassphraseGroup,
+}
+
+/// Generate a certificate.
+#[derive(Debug, Parser)]
+struct GenerateCertCli {
+    /// Path to input private key.
+    #[arg(short, long, value_name = "FILE", value_parser)]
+    key: PathBuf,
+
+    #[command(flatten)]
+    passphrase: PassphraseGroup,
+
+    /// Path to output certificate.
+    #[arg(short, long, value_name = "FILE", value_parser)]
+    output: PathBuf,
+
+    /// Certificate subject with comma-separated components.
+    #[arg(short, long, default_value = "CN=avbroot")]
+    subject: String,
+
+    /// Certificate validity in days.
+    #[arg(short, long, default_value = "10000")]
+    validity: u64,
+}
+
+/// Extract the AVB public key from a private key or certificate.
+///
+/// The public key is stored in both the private key and the certificate. Either
+/// one can be used interchangeably.
+#[derive(Debug, Parser)]
+struct ExtractAvbCli {
+    /// Path to output AVB public key.
+    #[arg(short, long, value_name = "FILE", value_parser)]
+    output: PathBuf,
+
+    #[command(flatten)]
+    input: PublicKeyInputGroup,
+
+    #[command(flatten)]
+    passphrase: PassphraseGroup,
+}
+
+#[derive(Debug, Subcommand)]
+enum KeyCommand {
+    GenerateKey(GenerateKeyCli),
+    GenerateCert(GenerateCertCli),
+    ExtractAvb(ExtractAvbCli),
+}
+
+/// Generate and convert keys.
+#[derive(Debug, Parser)]
+pub struct KeyCli {
+    #[command(subcommand)]
+    command: KeyCommand,
+}

avbroot/src/cli/mod.rs

@@ -0,0 +1,27 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+pub mod args;
+pub mod avb;
+pub mod boot;
+pub mod completion;
+pub mod key;
+pub mod ota;
+pub mod ramdisk;
+
+macro_rules! status {
+    ($($arg:tt)*) => {
+        println!("\x1b[1m[*] {}\x1b[0m", format!($($arg)*))
+    }
+}
+
+macro_rules! warning {
+    ($($arg:tt)*) => {
+        println!("\x1b[1;31m[WARNING] {}\x1b[0m", format!($($arg)+))
+    }
+}
+
+pub(crate) use status;
+pub(crate) use warning;

avbroot/src/cli/ramdisk.rs

@@ -0,0 +1,153 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::{
+    fs::File,
+    path::{Path, PathBuf},
+    str,
+};
+
+use anyhow::{Context, Result};
+use clap::{Parser, Subcommand};
+
+use crate::{
+    format::{
+        compression::{CompressedFormat, CompressedReader, CompressedWriter},
+        cpio::{self, CpioEntryNew},
+    },
+    util::EscapedString,
+};
+
+static CONTENT_BEGIN: &str = "----- BEGIN UTF-8 CONTENT -----";
+static CONTENT_END: &str = "----- END UTF-8 CONTENT -----";
+static CONTENT_END_NO_NEWLINE: &str = "----- END UTF-8 CONTENT (NO NEWLINE) -----";
+
+static BINARY_BEGIN: &str = "----- BEGIN BINARY CONTENT -----";
+static BINARY_END: &str = "----- END BINARY CONTENT -----";
+static BINARY_END_TRUNCATED: &str = "----- END BINARY CONTENT (TRUNCATED) -----";
+
+static NO_DATA: &str = "----- NO DATA -----";
+
+fn print_content(data: &[u8], truncate: bool) {
+    if data.is_empty() {
+        println!("{NO_DATA}");
+        return;
+    }
+
+    if !data.contains(&b'\0') {
+        if let Ok(s) = str::from_utf8(data) {
+            if !s.contains(CONTENT_BEGIN)
+                && !s.contains(CONTENT_END)
+                && !s.contains(CONTENT_END_NO_NEWLINE)
+            {
+                println!("{CONTENT_BEGIN}");
+                print!("{s}");
+                if data.last() == Some(&b'\n') {
+                    println!("{CONTENT_END}");
+                } else {
+                    println!();
+                    println!("{CONTENT_END_NO_NEWLINE}");
+                }
+
+                return;
+            }
+        }
+    }
+
+    println!("{BINARY_BEGIN}");
+
+    if data.len() > 512 && truncate {
+        println!("{}", EscapedString::new_unquoted(&data[..512]));
+        println!("{BINARY_END_TRUNCATED}");
+    } else {
+        println!("{}", EscapedString::new_unquoted(&data));
+        println!("{BINARY_END}");
+    }
+}
+
+fn load_archive(
+    path: &Path,
+    include_trailer: bool,
+) -> Result<(Vec<CpioEntryNew>, CompressedFormat)> {
+    let file = File::open(path)?;
+    let reader = CompressedReader::new(file, true)?;
+    let format = reader.format();
+    let entries = cpio::load(reader, include_trailer)?;
+
+    Ok((entries, format))
+}
+
+fn save_archive(path: &Path, entries: &[CpioEntryNew], format: CompressedFormat) -> Result<()> {
+    let file = File::create(path)?;
+    let mut writer = CompressedWriter::new(file, format)?;
+    cpio::save(&mut writer, entries, false)?;
+    writer.finish()?;
+
+    Ok(())
+}
+
+pub fn ramdisk_main(cli: &RamdiskCli) -> Result<()> {
+    match &cli.command {
+        RamdiskCommand::Dump(c) => {
+            let (entries, format) = load_archive(&c.input, true)
+                .with_context(|| format!("Failed to read cpio: {:?}", c.input))?;
+
+            println!("Compression format: {format:?}");
+            println!();
+
+            for entry in entries {
+                println!("{entry}");
+                print_content(&entry.content, !c.no_truncate);
+                println!();
+            }
+        }
+        RamdiskCommand::Repack(c) => {
+            let (entries, format) = load_archive(&c.input, false)
+                .with_context(|| format!("Failed to read cpio: {:?}", c.input))?;
+
+            save_archive(&c.output, &entries, format)
+                .with_context(|| format!("Failed to write cpio: {:?}", c.output))?;
+        }
+    }
+
+    Ok(())
+}
+
+/// Dump cpio headers and data.
+#[derive(Debug, Parser)]
+struct DumpCli {
+    /// Path to input cpio file.
+    #[arg(short, long, value_name = "FILE", value_parser)]
+    input: PathBuf,
+
+    /// Do not truncate binary file contents.
+    #[arg(long)]
+    no_truncate: bool,
+}
+
+/// Repack cpio archive.
+#[derive(Debug, Parser)]
+struct RepackCli {
+    /// Path to input cpio file.
+    #[arg(short, long, value_name = "FILE", value_parser)]
+    input: PathBuf,
+
+    /// Path to output cpio file.
+    #[arg(short, long, value_name = "FILE", value_parser)]
+    output: PathBuf,
+}
+
+#[derive(Debug, Subcommand)]
+enum RamdiskCommand {
+    Dump(DumpCli),
+    Repack(RepackCli),
+}
+
+/// Show information about ramdisk cpio archives.
+#[derive(Debug, Parser)]
+pub struct RamdiskCli {
+    #[command(subcommand)]
+    command: RamdiskCommand,
+}

avbroot/src/crypto.rs

@@ -0,0 +1,408 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::{
+    env::{self, VarError},
+    ffi::OsString,
+    fs::{self, File, OpenOptions},
+    io::{self, BufReader, BufWriter, Read, Write},
+    path::{Path, PathBuf},
+    time::Duration,
+};
+
+use cms::{
+    cert::{CertificateChoices, IssuerAndSerialNumber},
+    content_info::{CmsVersion, ContentInfo},
+    signed_data::{
+        CertificateSet, DigestAlgorithmIdentifiers, EncapsulatedContentInfo, SignatureValue,
+        SignedData, SignerIdentifier, SignerInfo, SignerInfos,
+    },
+};
+use pkcs8::{
+    pkcs5::{pbes2, scrypt},
+    DecodePrivateKey, EncodePrivateKey, EncodePublicKey, EncryptedPrivateKeyInfo, LineEnding,
+    PrivateKeyInfo,
+};
+use rand::RngCore;
+use rsa::{pkcs1v15::SigningKey, Pkcs1v15Sign, RsaPrivateKey, RsaPublicKey};
+use sha2::Sha256;
+use thiserror::Error;
+use x509_cert::{
+    builder::{Builder, CertificateBuilder, Profile},
+    der::{pem::PemLabel, referenced::OwnedToRef, Any, Decode, DecodePem, EncodePem},
+    serial_number::SerialNumber,
+    spki::{AlgorithmIdentifierOwned, SubjectPublicKeyInfoOwned},
+    time::Validity,
+    Certificate,
+};
+
+#[derive(Debug, Error)]
+pub enum Error {
+    #[error("Passphrases do not match")]
+    ConfirmPassphrase,
+    #[error("Failed to read environment variable: {0:?}")]
+    InvalidEnvVar(OsString, #[source] VarError),
+    #[error("PEM has start tag, but no end tag")]
+    PemNoEndTag,
+    #[error("Failed to load encrypted private key")]
+    LoadKeyEncrypted(#[source] pkcs8::Error),
+    #[error("Failed to load unencrypted private key")]
+    LoadKeyUnencrypted(#[source] pkcs8::Error),
+    #[error("Failed to save encrypted private key")]
+    SaveKeyEncrypted(#[source] pkcs8::Error),
+    #[error("Failed to save unencrypted private key")]
+    SaveKeyUnencrypted(#[source] pkcs8::Error),
+    #[error("X509 error")]
+    X509(#[from] x509_cert::builder::Error),
+    #[error("SPKI error")]
+    Spki(#[from] pkcs8::spki::Error),
+    #[error("DER error")]
+    Der(#[from] x509_cert::der::Error),
+    #[error("RSA error")]
+    RsaSign(#[from] rsa::Error),
+    #[error("I/O error")]
+    Io(#[from] io::Error),
+}
+
+type Result<T> = std::result::Result<T, Error>;
+
+pub enum PassphraseSource {
+    Prompt(String),
+    EnvVar(OsString),
+    File(PathBuf),
+}
+
+impl PassphraseSource {
+    pub fn acquire(&self, confirm: bool) -> Result<String> {
+        let passphrase = match self {
+            Self::Prompt(p) => {
+                let first = rpassword::prompt_password(p)?;
+
+                if confirm {
+                    let second = rpassword::prompt_password("Confirm: ")?;
+
+                    if first != second {
+                        return Err(Error::ConfirmPassphrase);
+                    }
+                }
+
+                first
+            }
+            Self::EnvVar(v) => env::var(v).map_err(|e| Error::InvalidEnvVar(v.clone(), e))?,
+            Self::File(p) => fs::read_to_string(p)?
+                .trim_end_matches(&['\r', '\n'])
+                .to_owned(),
+        };
+
+        Ok(passphrase)
+    }
+}
+
+/// Generate an 4096-bit RSA key pair.
+pub fn generate_rsa_key_pair() -> Result<RsaPrivateKey> {
+    let mut rng = rand::thread_rng();
+
+    // avbroot supports 4096-bit keys only.
+    let key = RsaPrivateKey::new(&mut rng, 4096)?;
+
+    Ok(key)
+}
+
+/// Generate a self-signed certificate.
+pub fn generate_cert(
+    key: &RsaPrivateKey,
+    serial: u64,
+    validity: Duration,
+    subject: &str,
+) -> Result<Certificate> {
+    let public_key_der = key.to_public_key().to_public_key_der()?;
+    let signing_key = SigningKey::<Sha256>::new(key.clone());
+
+    let builder = CertificateBuilder::new(
+        Profile::Root,
+        SerialNumber::from(serial),
+        Validity::from_now(validity)?,
+        subject.parse()?,
+        SubjectPublicKeyInfoOwned::from_der(public_key_der.as_bytes())?,
+        &signing_key,
+    )?;
+
+    let mut rng = rand::thread_rng();
+    let cert = builder.build_with_rng(&mut rng)?;
+
+    Ok(cert)
+}
+
+/// x509_cert/pem follow rfc7468 strictly instead of implementing a lenient
+/// parser. The PEM decoder rejects lines in the base64 section that are longer
+/// than 64 characters, excluding whitespace. We'll reformat the data to deal
+/// with this because there are certificates that do not follow the spec, like
+/// the signing cert for the Pixel 7 Pro official OTAs.
+fn reformat_pem(data: &[u8]) -> Result<Vec<u8>> {
+    let mut result = vec![];
+    let mut base64 = vec![];
+    let mut inside_base64 = false;
+
+    for mut line in data.split(|&c| c == b'\n') {
+        while !line.is_empty() && line[line.len() - 1].is_ascii_whitespace() {
+            line = &line[..line.len() - 1];
+        }
+
+        if line.is_empty() {
+            continue;
+        } else if line.starts_with(b"-----BEGIN CERTIFICATE-----") {
+            inside_base64 = true;
+        } else if line.starts_with(b"-----END CERTIFICATE-----") {
+            inside_base64 = false;
+
+            for chunk in base64.chunks(64) {
+                result.extend_from_slice(chunk);
+                result.push(b'\n');
+            }
+
+            base64.clear();
+        } else if inside_base64 {
+            base64.extend_from_slice(line);
+            continue;
+        }
+
+        result.extend_from_slice(line);
+        result.push(b'\n');
+    }
+
+    if inside_base64 {
+        return Err(Error::PemNoEndTag);
+    }
+
+    Ok(result)
+}
+
+/// Read PEM-encoded certificate from a reader.
+pub fn read_pem_cert(mut reader: impl Read) -> Result<Certificate> {
+    let mut data = vec![];
+    reader.read_to_end(&mut data)?;
+
+    let data = reformat_pem(&data)?;
+    let certificate = Certificate::from_pem(data)?;
+
+    Ok(certificate)
+}
+
+/// Write PEM-encoded certificate to a writer.
+pub fn write_pem_cert(mut writer: impl Write, cert: &Certificate) -> Result<()> {
+    let data = cert.to_pem(LineEnding::LF)?;
+
+    writer.write_all(data.as_bytes())?;
+
+    Ok(())
+}
+
+/// Read PEM-encoded certificate from a file.
+pub fn read_pem_cert_file(path: &Path) -> Result<Certificate> {
+    let file = File::open(path)?;
+    let reader = BufReader::new(file);
+
+    read_pem_cert(reader)
+}
+
+/// Write PEM-encoded certificate to a file.
+pub fn write_pem_cert_file(path: &Path, cert: &Certificate) -> Result<()> {
+    let file = File::create(path)?;
+    let writer = BufWriter::new(file);
+
+    write_pem_cert(writer, cert)
+}
+
+/// Read PEM-encoded PKCS8 private key from a reader.
+pub fn read_pem_key(mut reader: impl Read, source: &PassphraseSource) -> Result<RsaPrivateKey> {
+    let mut data = String::new();
+    reader.read_to_string(&mut data)?;
+
+    if data.contains("ENCRYPTED") {
+        let passphrase = source.acquire(false)?;
+
+        RsaPrivateKey::from_pkcs8_encrypted_pem(&data, passphrase).map_err(Error::LoadKeyEncrypted)
+    } else {
+        RsaPrivateKey::from_pkcs8_pem(&data).map_err(Error::LoadKeyUnencrypted)
+    }
+}
+
+/// Write PEM-encoded PKCS8 private key to a writer.
+pub fn write_pem_key(
+    mut writer: impl Write,
+    key: &RsaPrivateKey,
+    source: &PassphraseSource,
+) -> Result<()> {
+    let passphrase = source.acquire(true)?;
+
+    let data = if passphrase.is_empty() {
+        key.to_pkcs8_pem(LineEnding::LF)
+            .map_err(Error::SaveKeyUnencrypted)?
+    } else {
+        let mut rng = rand::thread_rng();
+
+        // Normally, we'd just use key.to_pkcs8_encrypted_pem(). However, it
+        // uses scrypt with n = 32768. This is high enough that openssl can no
+        // longer read the file and craps out with `memory limit exceeded`.
+        // Although we can read those files just fine, let's match openssl's
+        // default parameters for better compatibility.
+        //
+        // Per `man openssl-pkcs8`: -scrypt  Uses the scrypt algorithm for
+        // private key encryption using default parameters: currently N=16384,
+        // r=8 and p=1 and AES in CBC mode with a 256 bit key.
+        //
+        // https://github.com/RustCrypto/formats/issues/1205
+
+        let mut salt = [0u8; 16];
+        rng.fill_bytes(&mut salt);
+
+        let mut iv = [0u8; 16];
+        rng.fill_bytes(&mut iv);
+
+        // 14 = log_2(16384), 32 bytes = 256 bits
+        let scrypt_params = scrypt::Params::new(14, 8, 1, 32).unwrap();
+        let pbes2_params = pbes2::Parameters::scrypt_aes256cbc(scrypt_params, &salt, &iv).unwrap();
+
+        let plain_text_der = key.to_pkcs8_der().map_err(Error::SaveKeyEncrypted)?;
+        let private_key_info =
+            PrivateKeyInfo::try_from(plain_text_der.as_bytes()).map_err(Error::SaveKeyEncrypted)?;
+
+        let secret_doc = private_key_info
+            .encrypt_with_params(pbes2_params, passphrase)
+            .map_err(Error::SaveKeyEncrypted)?;
+
+        secret_doc.to_pem(EncryptedPrivateKeyInfo::PEM_LABEL, LineEnding::LF)?
+    };
+
+    writer.write_all(data.as_bytes())?;
+
+    Ok(())
+}
+
+/// Read PEM-encoded PKCS8 private key from a file.
+pub fn read_pem_key_file(path: &Path, source: &PassphraseSource) -> Result<RsaPrivateKey> {
+    let file = File::open(path)?;
+    let reader = BufReader::new(file);
+
+    read_pem_key(reader, source)
+}
+
+/// Save PEM-encoded PKCS8 private key to a file.
+pub fn write_pem_key_file(
+    path: &Path,
+    key: &RsaPrivateKey,
+    source: &PassphraseSource,
+) -> Result<()> {
+    let mut options = OpenOptions::new();
+    options.write(true);
+    options.create(true);
+    options.truncate(true);
+
+    #[cfg(unix)]
+    {
+        use std::os::unix::fs::OpenOptionsExt;
+        options.mode(0o600);
+    }
+
+    let file = options.open(path)?;
+    let writer = BufWriter::new(file);
+
+    write_pem_key(writer, key, source)
+}
+
+/// Get the RSA public key from a certificate.
+pub fn get_public_key(cert: &Certificate) -> Result<RsaPublicKey> {
+    let public_key =
+        RsaPublicKey::try_from(cert.tbs_certificate.subject_public_key_info.owned_to_ref())?;
+
+    Ok(public_key)
+}
+
+/// Check if a certificate matches a private key.
+pub fn cert_matches_key(cert: &Certificate, key: &RsaPrivateKey) -> Result<bool> {
+    let public_key = get_public_key(cert)?;
+
+    Ok(key.to_public_key() == public_key)
+}
+
+/// Parse a CMS [`SignedData`] structure from raw DER-encoded data.
+pub fn parse_cms(data: &[u8]) -> Result<SignedData> {
+    let ci = ContentInfo::from_der(data)?;
+    let sd = ci.content.decode_as::<SignedData>()?;
+
+    Ok(sd)
+}
+
+/// Get a list of all standard X509 certificates contained within a
+/// [`SignedData`] structure.
+pub fn get_cms_certs(sd: &SignedData) -> Vec<Certificate> {
+    sd.certificates.as_ref().map_or_else(Vec::new, |certs| {
+        certs
+            .0
+            .iter()
+            .filter_map(|cc| {
+                if let CertificateChoices::Certificate(c) = cc {
+                    Some(c.clone())
+                } else {
+                    None
+                }
+            })
+            .collect()
+    })
+}
+
+/// Create a CMS signature from an external digest. This implementation does not
+/// use signed attributes because AOSP recovery's otautil/verifier.cpp is not
+/// actually CMS compliant. It simply uses the CMS [`SignedData`] structure as
+/// a transport mechanism for a raw signature. Thus, we need to ensure that the
+/// signature covers nothing but the raw data.
+pub fn cms_sign_external(
+    key: &RsaPrivateKey,
+    cert: &Certificate,
+    digest: &[u8],
+) -> Result<ContentInfo> {
+    let scheme = Pkcs1v15Sign::new::<Sha256>();
+    let signature = key.sign(scheme, digest)?;
+
+    let digest_algorithm = AlgorithmIdentifierOwned {
+        oid: const_oid::db::rfc5912::ID_SHA_256,
+        parameters: None,
+    };
+
+    let signed_data = SignedData {
+        version: CmsVersion::V1,
+        digest_algorithms: DigestAlgorithmIdentifiers::try_from(vec![digest_algorithm.clone()])?,
+        encap_content_info: EncapsulatedContentInfo {
+            econtent_type: const_oid::db::rfc5911::ID_DATA,
+            econtent: None,
+        },
+        certificates: Some(CertificateSet::try_from(vec![
+            CertificateChoices::Certificate(cert.clone()),
+        ])?),
+        crls: None,
+        signer_infos: SignerInfos::try_from(vec![SignerInfo {
+            version: CmsVersion::V1,
+            sid: SignerIdentifier::IssuerAndSerialNumber(IssuerAndSerialNumber {
+                issuer: cert.tbs_certificate.issuer.clone(),
+                serial_number: cert.tbs_certificate.serial_number.clone(),
+            }),
+            digest_alg: digest_algorithm,
+            signed_attrs: None,
+            signature_algorithm: AlgorithmIdentifierOwned {
+                oid: const_oid::db::rfc5912::SHA_256_WITH_RSA_ENCRYPTION,
+                parameters: None,
+            },
+            signature: SignatureValue::new(signature)?,
+            unsigned_attrs: None,
+        }])?,
+    };
+
+    let signed_data = ContentInfo {
+        content_type: const_oid::db::rfc5911::ID_SIGNED_DATA,
+        content: Any::encode_from(&signed_data)?,
+    };
+
+    Ok(signed_data)
+}

avbroot/src/format/compression.rs

@@ -0,0 +1,207 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::io::{self, Read, Seek, Write};
+
+use byteorder::{LittleEndian, WriteBytesExt};
+use flate2::{read::GzDecoder, write::GzEncoder, Compression};
+use lz4_flex::frame::FrameDecoder;
+use thiserror::Error;
+
+static GZIP_MAGIC: &[u8; 2] = b"\x1f\x8b";
+static LZ4_LEGACY_MAGIC: &[u8; 4] = b"\x02\x21\x4c\x18";
+
+#[derive(Debug, Error)]
+pub enum Error {
+    #[error("Unknown compression format")]
+    UnknownFormat,
+    #[error("I/O error")]
+    IoError(#[from] io::Error),
+}
+
+type Result<T> = std::result::Result<T, Error>;
+
+pub struct Lz4LegacyEncoder<W: Write> {
+    writer: Option<W>,
+    buf: Vec<u8>,
+    n_filled: usize,
+}
+
+impl<W: Write> Lz4LegacyEncoder<W> {
+    pub fn new(mut writer: W) -> io::Result<Self> {
+        writer.write_all(LZ4_LEGACY_MAGIC)?;
+
+        Ok(Self {
+            writer: Some(writer),
+            // We always use the max block size.
+            buf: vec![0u8; 8 * 1024 * 1024],
+            n_filled: 0,
+        })
+    }
+
+    pub fn write_block(&mut self, force: bool) -> io::Result<()> {
+        if !force && self.n_filled < self.buf.len() {
+            // Block not fully filled yet.
+            return Ok(());
+        }
+
+        // HC is currently not supported:
+        // https://github.com/PSeitz/lz4_flex/issues/21
+        let compressed = lz4_flex::block::compress(&self.buf[..self.n_filled]);
+
+        let writer = self.writer.as_mut().unwrap();
+        writer.write_u32::<LittleEndian>(compressed.len() as u32)?;
+        writer.write_all(&compressed)?;
+
+        self.n_filled = 0;
+
+        Ok(())
+    }
+
+    pub fn finish(mut self) -> io::Result<W> {
+        self.write_block(true)?;
+        Ok(self.writer.take().unwrap())
+    }
+}
+
+impl<W: Write> Drop for Lz4LegacyEncoder<W> {
+    fn drop(&mut self) {
+        if self.writer.is_some() {
+            let _ = self.write_block(true);
+        }
+    }
+}
+
+impl<W: Write> Write for Lz4LegacyEncoder<W> {
+    fn write(&mut self, mut buf: &[u8]) -> io::Result<usize> {
+        let total = buf.len();
+
+        while !buf.is_empty() {
+            let to_write = buf.len().min(self.buf.len() - self.n_filled);
+            self.buf[self.n_filled..self.n_filled + to_write].copy_from_slice(&buf[..to_write]);
+
+            self.n_filled += to_write;
+            self.write_block(false)?;
+
+            buf = &buf[to_write..];
+        }
+
+        Ok(total)
+    }
+
+    fn flush(&mut self) -> io::Result<()> {
+        self.write_block(false)
+    }
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum CompressedFormat {
+    None,
+    Gzip,
+    Lz4Legacy,
+}
+
+pub enum CompressedReader<R: Read> {
+    None(R),
+    Gzip(GzDecoder<R>),
+    Lz4(FrameDecoder<R>),
+}
+
+impl<R: Read + Seek> CompressedReader<R> {
+    pub fn new(mut reader: R, raw_if_unknown: bool) -> Result<Self> {
+        let mut magic = [0u8; 4];
+        reader.read_exact(&mut magic)?;
+
+        reader.rewind()?;
+
+        if &magic[0..2] == GZIP_MAGIC {
+            Ok(Self::Gzip(GzDecoder::new(reader)))
+        } else if &magic == LZ4_LEGACY_MAGIC {
+            Ok(Self::Lz4(FrameDecoder::new(reader)))
+        } else if raw_if_unknown {
+            Ok(Self::None(reader))
+        } else {
+            Err(Error::UnknownFormat)
+        }
+    }
+
+    pub fn format(&self) -> CompressedFormat {
+        match self {
+            Self::None(_) => CompressedFormat::None,
+            Self::Gzip(_) => CompressedFormat::Gzip,
+            Self::Lz4(_) => CompressedFormat::Lz4Legacy,
+        }
+    }
+
+    pub fn into_inner(self) -> R {
+        match self {
+            Self::None(r) => r,
+            Self::Gzip(r) => r.into_inner(),
+            Self::Lz4(r) => r.into_inner(),
+        }
+    }
+}
+
+impl<R: Read> Read for CompressedReader<R> {
+    fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
+        match self {
+            Self::None(r) => r.read(buf),
+            Self::Gzip(r) => r.read(buf),
+            Self::Lz4(r) => r.read(buf),
+        }
+    }
+}
+
+pub enum CompressedWriter<W: Write> {
+    None(W),
+    Gzip(GzEncoder<W>),
+    Lz4Legacy(Lz4LegacyEncoder<W>),
+}
+
+impl<W: Write> CompressedWriter<W> {
+    pub fn new(writer: W, format: CompressedFormat) -> Result<Self> {
+        match format {
+            CompressedFormat::None => Ok(Self::None(writer)),
+            CompressedFormat::Gzip => {
+                Ok(Self::Gzip(GzEncoder::new(writer, Compression::default())))
+            }
+            CompressedFormat::Lz4Legacy => Ok(Self::Lz4Legacy(Lz4LegacyEncoder::new(writer)?)),
+        }
+    }
+
+    pub fn format(&self) -> CompressedFormat {
+        match self {
+            Self::None(_) => CompressedFormat::None,
+            Self::Gzip(_) => CompressedFormat::Gzip,
+            Self::Lz4Legacy(_) => CompressedFormat::Lz4Legacy,
+        }
+    }
+
+    pub fn finish(self) -> io::Result<W> {
+        match self {
+            Self::None(w) => Ok(w),
+            Self::Gzip(w) => w.finish(),
+            Self::Lz4Legacy(w) => w.finish(),
+        }
+    }
+}
+
+impl<W: Write> Write for CompressedWriter<W> {
+    fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
+        match self {
+            Self::None(w) => w.write(buf),
+            Self::Gzip(w) => w.write(buf),
+            Self::Lz4Legacy(w) => w.write(buf),
+        }
+    }
+
+    fn flush(&mut self) -> io::Result<()> {
+        match self {
+            Self::None(w) => w.flush(),
+            Self::Gzip(w) => w.flush(),
+            Self::Lz4Legacy(w) => w.flush(),
+        }
+    }
+}

avbroot/src/format/cpio.rs

@@ -0,0 +1,362 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::{
+    borrow::Cow,
+    fmt,
+    io::{self, Read, Write},
+};
+
+use num_traits::ToPrimitive;
+use thiserror::Error;
+
+use crate::{
+    format::padding,
+    stream::{CountingReader, CountingWriter, FromReader, ToWriter, WriteZerosExt},
+    util::{EscapedString, NumBytes},
+};
+
+const MAGIC_NEW: &[u8; 6] = b"070701";
+const MAGIC_NEW_CRC: &[u8; 6] = b"070702";
+
+const CPIO_TRAILER: &[u8; 10] = b"TRAILER!!!";
+
+const S_IFIFO: u32 = 0o010000;
+const S_IFCHR: u32 = 0o020000;
+const S_IFDIR: u32 = 0o040000;
+const S_IFBLK: u32 = 0o060000;
+const S_IFREG: u32 = 0o100000;
+const S_IFLNK: u32 = 0o120000;
+const S_IFSOCK: u32 = 0o140000;
+const C_ISCTG: u32 = 0o110000;
+
+const IO_BLOCK_SIZE: u64 = 512;
+
+#[derive(Debug, Error)]
+pub enum Error {
+    #[error("Unknown magic: {0:?}")]
+    UnknownMagic([u8; 6]),
+    #[error("Hard links are not supported: {0}")]
+    HardLinksNotSupported(EscapedString<Vec<u8>>),
+    #[error("{0:?} field exceeds integer bounds")]
+    IntegerTooLarge(&'static str),
+    #[error("I/O error")]
+    IoError(#[from] io::Error),
+}
+
+type Result<T> = std::result::Result<T, Error>;
+
+/// Read u32 formatted as an ASCII 8-char wide hex string.
+fn read_int(mut reader: impl Read) -> io::Result<u32> {
+    let mut buf = [0u8; 8];
+    reader.read_exact(&mut buf)?;
+
+    let mut value = 0;
+
+    for b in buf {
+        let c = b as char;
+        let digit = c.to_digit(16).ok_or_else(|| {
+            io::Error::new(
+                io::ErrorKind::InvalidData,
+                format!("{0}: Invalid hex char: {1}", EscapedString::new(&buf), c),
+            )
+        })?;
+
+        value <<= 4;
+        value |= digit;
+    }
+
+    Ok(value)
+}
+
+/// Write u32 formatted as an ASCII 8-char wide hex string.
+fn write_int(mut writer: impl Write, mut value: u32) -> io::Result<()> {
+    let mut buf = [b'0'; 8];
+    let mut index = 7;
+
+    while value != 0 {
+        buf[index] = char::from_digit(value & 0xf, 16).unwrap() as u8;
+        value >>= 4;
+        index -= 1;
+    }
+
+    writer.write_all(&buf)
+}
+
+fn file_type(mode: u32) -> u32 {
+    mode & 0o170000
+}
+
+#[derive(Clone, Default, PartialEq, Eq)]
+pub struct CpioEntryNew {
+    pub ino: u32,
+    pub mode: u32,
+    pub uid: u32,
+    pub gid: u32,
+    pub nlink: u32,
+    pub mtime: u32,
+    pub dev_maj: u32,
+    pub dev_min: u32,
+    pub rdev_maj: u32,
+    pub rdev_min: u32,
+    pub chksum: u32,
+    pub name: Vec<u8>,
+    pub content: Vec<u8>,
+}
+
+impl fmt::Debug for CpioEntryNew {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        f.debug_struct("CpioEntryNew")
+            .field("ino", &self.ino)
+            .field("mode", &self.mode)
+            .field("uid", &self.uid)
+            .field("gid", &self.gid)
+            .field("nlink", &self.nlink)
+            .field("mtime", &self.mtime)
+            .field("dev_maj", &self.dev_maj)
+            .field("dev_min", &self.dev_min)
+            .field("rdev_maj", &self.rdev_maj)
+            .field("rdev_min", &self.rdev_min)
+            .field("chksum", &self.chksum)
+            .field("name", &EscapedString::new(&self.name))
+            .field("content", &NumBytes(self.content.len()))
+            .finish()
+    }
+}
+
+impl fmt::Display for CpioEntryNew {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        let file_type_str = match file_type(self.mode) {
+            S_IFIFO => Cow::Borrowed("pipe"),
+            S_IFCHR => Cow::Borrowed("character device"),
+            S_IFDIR => Cow::Borrowed("directory"),
+            S_IFBLK => Cow::Borrowed("block device"),
+            S_IFREG => Cow::Borrowed("regular file"),
+            S_IFLNK => Cow::Borrowed("symbolic link"),
+            S_IFSOCK => Cow::Borrowed("socket"),
+            C_ISCTG => Cow::Borrowed("reserved"),
+            m => Cow::Owned(format!("unknown ({m:o})")),
+        };
+
+        writeln!(f, "Filename:  {}", EscapedString::new(&self.name))?;
+        writeln!(f, "Filetype:  {file_type_str}")?;
+        writeln!(f, "Inode:     {}", self.ino)?;
+        writeln!(f, "Mode:      {:o}", self.mode)?;
+        writeln!(f, "UID:       {}", self.uid)?;
+        writeln!(f, "GID:       {}", self.gid)?;
+        writeln!(f, "Links:     {}", self.nlink)?;
+        writeln!(f, "Modified:  {}", self.mtime)?;
+        writeln!(f, "Device:    {:x},{:x}", self.dev_maj, self.dev_min)?;
+        writeln!(f, "Device ID: {:x},{:x}", self.rdev_maj, self.rdev_min)?;
+        writeln!(f, "Checksum:  {:x}", self.chksum)?;
+        writeln!(f, "Content:   {:?}", NumBytes(self.content.len()))?;
+
+        Ok(())
+    }
+}
+
+impl CpioEntryNew {
+    pub fn new_trailer() -> Self {
+        Self {
+            // Must be 1 for CRC format.
+            nlink: 1,
+            name: CPIO_TRAILER.to_vec(),
+            ..Default::default()
+        }
+    }
+
+    pub fn new_symlink(link_target: &[u8], name: &[u8]) -> Self {
+        Self {
+            mode: S_IFLNK | 0o777,
+            nlink: 1,
+            name: name.to_owned(),
+            content: link_target.to_owned(),
+            ..Default::default()
+        }
+    }
+
+    pub fn new_directory(name: &[u8]) -> Self {
+        Self {
+            mode: S_IFDIR,
+            nlink: 1,
+            name: name.to_owned(),
+            ..Default::default()
+        }
+    }
+
+    pub fn new_file(name: &[u8]) -> Self {
+        Self {
+            mode: S_IFREG,
+            nlink: 1,
+            name: name.to_owned(),
+            ..Default::default()
+        }
+    }
+}
+
+impl<R: Read> FromReader<R> for CpioEntryNew {
+    type Error = Error;
+
+    fn from_reader(reader: R) -> Result<Self> {
+        let mut reader = CountingReader::new(reader);
+
+        let mut magic = [0u8; 6];
+        reader.read_exact(&mut magic)?;
+
+        if magic != *MAGIC_NEW && magic != *MAGIC_NEW_CRC {
+            return Err(Error::UnknownMagic(magic));
+        }
+
+        let ino = read_int(&mut reader)?;
+        let mode = read_int(&mut reader)?;
+        let uid = read_int(&mut reader)?;
+        let gid = read_int(&mut reader)?;
+        let nlink = read_int(&mut reader)?;
+        let mtime = read_int(&mut reader)?;
+        let filesize = read_int(&mut reader)?;
+        let dev_maj = read_int(&mut reader)?;
+        let dev_min = read_int(&mut reader)?;
+        let rdev_maj = read_int(&mut reader)?;
+        let rdev_min = read_int(&mut reader)?;
+        let namesize = read_int(&mut reader)?;
+        let chksum = read_int(&mut reader)?;
+
+        let mut name = vec![0u8; namesize.to_usize().unwrap()];
+        reader.read_exact(&mut name)?;
+        if name.last() != Some(&b'\0') {
+            return Err(io::Error::new(
+                io::ErrorKind::InvalidData,
+                "Filename is not NULL-terminated",
+            )
+            .into());
+        }
+        name.pop();
+        padding::read_discard(&mut reader, 4)?;
+
+        let mut content = vec![0u8; filesize.to_usize().unwrap()];
+        reader.read_exact(&mut content)?;
+        padding::read_discard(&mut reader, 4)?;
+
+        Ok(Self {
+            ino,
+            mode,
+            uid,
+            gid,
+            nlink,
+            mtime,
+            dev_maj,
+            dev_min,
+            rdev_maj,
+            rdev_min,
+            chksum,
+            name,
+            content,
+        })
+    }
+}
+
+impl<W: Write> ToWriter<W> for CpioEntryNew {
+    type Error = Error;
+
+    fn to_writer(&self, writer: W) -> Result<()> {
+        let mut writer = CountingWriter::new(writer);
+
+        let filesize = self
+            .content
+            .len()
+            .to_u32()
+            .ok_or_else(|| Error::IntegerTooLarge("filesize"))?;
+        let namesize = self
+            .name
+            .len()
+            .checked_add(1)
+            .and_then(|s| s.to_u32())
+            .ok_or_else(|| Error::IntegerTooLarge("filesize"))?;
+
+        if self.chksum == 0 {
+            writer.write_all(MAGIC_NEW)?;
+        } else {
+            writer.write_all(MAGIC_NEW_CRC)?;
+        }
+
+        write_int(&mut writer, self.ino)?;
+        write_int(&mut writer, self.mode)?;
+        write_int(&mut writer, self.uid)?;
+        write_int(&mut writer, self.gid)?;
+        write_int(&mut writer, self.nlink)?;
+        write_int(&mut writer, self.mtime)?;
+        write_int(&mut writer, filesize)?;
+        write_int(&mut writer, self.dev_maj)?;
+        write_int(&mut writer, self.dev_min)?;
+        write_int(&mut writer, self.rdev_maj)?;
+        write_int(&mut writer, self.rdev_min)?;
+        write_int(&mut writer, namesize)?;
+        write_int(&mut writer, self.chksum)?;
+
+        writer.write_all(&self.name)?;
+        writer.write_zeros_exact(1)?;
+        padding::write_zeros(&mut writer, 4)?;
+
+        writer.write_all(&self.content)?;
+        padding::write_zeros(&mut writer, 4)?;
+
+        Ok(())
+    }
+}
+
+pub fn load(mut reader: impl Read, include_trailer: bool) -> Result<Vec<CpioEntryNew>> {
+    let mut entries = vec![];
+
+    loop {
+        let entry = CpioEntryNew::from_reader(&mut reader)?;
+        if file_type(entry.mode) != S_IFDIR && entry.nlink > 1 {
+            return Err(Error::HardLinksNotSupported(EscapedString::new(entry.name)));
+        }
+
+        if entry.name == CPIO_TRAILER {
+            if include_trailer {
+                entries.push(entry);
+            }
+            break;
+        }
+
+        entries.push(entry);
+    }
+
+    Ok(entries)
+}
+
+pub fn sort(entries: &mut [CpioEntryNew]) {
+    entries.sort_by(|a, b| a.name.cmp(&b.name));
+}
+
+pub fn reassign_inodes(entries: &mut [CpioEntryNew]) {
+    let mut inode = 300000;
+
+    for entry in entries {
+        entry.ino = inode;
+        inode += 1;
+    }
+}
+
+pub fn save(writer: impl Write, entries: &[CpioEntryNew], pad_to_block_size: bool) -> Result<()> {
+    let mut writer = CountingWriter::new(writer);
+
+    for entry in entries {
+        entry.to_writer(&mut writer)?;
+    }
+
+    let mut trailer = CpioEntryNew::new_trailer();
+    // 1 higher than the highest inode if possible.
+    trailer.ino = entries.iter().map(|e| e.ino).max().map_or(0, |i| i + 1);
+    trailer.to_writer(&mut writer)?;
+
+    // Pad until the end of the block.
+    if pad_to_block_size {
+        padding::write_zeros(&mut writer, IO_BLOCK_SIZE)?;
+    }
+
+    Ok(())
+}

avbroot/src/format/mod.rs

@@ -0,0 +1,12 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+pub mod avb;
+pub mod bootimage;
+pub mod compression;
+pub mod cpio;
+pub mod ota;
+pub mod padding;
+pub mod payload;

avbroot/src/format/ota.rs

@@ -0,0 +1,656 @@
+/*
+ * SPDX-FileCopyrightText: 2022-2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::{
+    collections::BTreeMap,
+    io::{self, Cursor, Read, Seek, SeekFrom, Write},
+    iter,
+    sync::{atomic::AtomicBool, Arc},
+};
+
+use cms::signed_data::SignedData;
+use const_oid::{db::rfc5912, ObjectIdentifier};
+use memchr::memmem;
+use ring::digest::Context;
+use rsa::{Pkcs1v15Sign, RsaPrivateKey};
+use sha1::Sha1;
+use sha2::Sha256;
+use thiserror::Error;
+use x509_cert::{der::Encode, Certificate};
+use zip::{result::ZipError, write::FileOptions, CompressionMethod, ZipArchive, ZipWriter};
+
+use crate::{
+    crypto,
+    format::payload::{self, PayloadHeader},
+    protobuf::build::tools::releasetools::{mod_OtaMetadata::OtaType, OtaMetadata},
+    stream::{self, FromReader, HashingReader, HashingWriter},
+    util,
+};
+
+pub const PATH_METADATA: &str = "META-INF/com/android/metadata";
+pub const PATH_METADATA_PB: &str = "META-INF/com/android/metadata.pb";
+pub const PATH_OTACERT: &str = "META-INF/com/android/otacert";
+pub const PATH_PAYLOAD: &str = "payload.bin";
+pub const PATH_PROPERTIES: &str = "payload_properties.txt";
+
+const NAME_PAYLOAD_METADATA: &str = "payload_metadata.bin";
+
+pub const PF_NAME: &str = "ota-property-files";
+pub const PF_STREAMING_NAME: &str = "ota-streaming-property-files";
+
+const ZIP_EOCD_MAGIC: &[u8; 4] = b"PK\x05\x06";
+
+const COMMENT_MESSAGE: &[u8] = b"signed by avbroot\0";
+
+#[derive(Debug, Error)]
+pub enum Error {
+    #[error("Cannot find OTA signature footer magic")]
+    OtaMagicNotFound,
+    #[error("Cannot find EOCD magic")]
+    EocdMagicNotFound,
+    #[error("EOCD magic found in archive comment")]
+    EocdMagicInComment,
+    #[error("Zip is too small to contain EOCD")]
+    ZipTooSmall,
+    #[error("Signature offset exceeds archive comment size")]
+    SignatureOffsetTooLarge,
+    #[error("Expected exactly one CMS embedded certificate, but found {0}")]
+    NotOneCmsCertificate(usize),
+    #[error("Expected exactly one CMS SignerInfo, but found {0}")]
+    NotOneCmsSignerInfo(usize),
+    #[error("Unsupported digest algorithm: {0}")]
+    UnsupportedDigestAlgorithm(ObjectIdentifier),
+    #[error("Unsupported signature algorithm: {0}")]
+    UnsupportedSignatureAlgorithm(ObjectIdentifier),
+    #[error("Expected entry offsets {0:?}, but have {1:?}")]
+    MismatchedPropertyFiles(String, String),
+    #[error("Property files {0:?} exceed {1} byte reserved space")]
+    InsufficientReservedSpace(String, usize),
+    #[error("Invalid property file entry: {0:?}")]
+    InvalidPropertyFileEntry(String),
+    #[error("Missing entry in OTA zip: {0}")]
+    MissingZipEntry(&'static str),
+    #[error("CMS signing error")]
+    CmsSign(#[from] crypto::Error),
+    #[error("Payload error")]
+    Payload(#[from] payload::Error),
+    #[error("Protobuf error")]
+    Protobuf(#[from] quick_protobuf::Error),
+    #[error("SPKI error")]
+    Spki(#[from] pkcs8::spki::Error),
+    #[error("x509 DER error")]
+    Der(#[from] x509_cert::der::Error),
+    #[error("RSA error")]
+    Rsa(#[from] rsa::Error),
+    #[error("Zip error")]
+    Zip(#[from] ZipError),
+    #[error("I/O error")]
+    Io(#[from] io::Error),
+}
+
+type Result<T> = std::result::Result<T, Error>;
+
+/// Generate the legacy plain-text and modern protobuf serializations of the
+/// given metadata instance.
+fn serialize_metadata(metadata: &OtaMetadata) -> Result<(String, Vec<u8>)> {
+    const SEP: &str = "|";
+
+    let mut pairs = BTreeMap::<String, String>::new();
+
+    match metadata.type_pb {
+        OtaType::AB => {
+            pairs.insert("ota-type".to_owned(), "AB".to_owned());
+        }
+        OtaType::BLOCK => {
+            pairs.insert("ota-type".to_owned(), "BLOCK".to_owned());
+        }
+        _ => {}
+    }
+    if metadata.wipe {
+        pairs.insert("ota-wipe".to_owned(), "yes".to_owned());
+    }
+    if metadata.retrofit_dynamic_partitions {
+        pairs.insert(
+            "ota-retrofit-dynamic-partitions".to_owned(),
+            "yes".to_owned(),
+        );
+    }
+    if metadata.downgrade {
+        pairs.insert("ota-downgrade".to_owned(), "yes".to_owned());
+    }
+
+    pairs.insert(
+        "ota-required-cache".to_owned(),
+        metadata.required_cache.to_string(),
+    );
+
+    if let Some(p) = &metadata.postcondition {
+        pairs.insert("post-build".to_owned(), p.build.join(SEP));
+        pairs.insert(
+            "post-build-incremental".to_owned(),
+            p.build_incremental.clone(),
+        );
+        pairs.insert("post-sdk-level".to_owned(), p.sdk_level.clone());
+        pairs.insert(
+            "post-security-patch-level".to_owned(),
+            p.security_patch_level.clone(),
+        );
+        pairs.insert("post-timestamp".to_owned(), p.timestamp.to_string());
+    }
+
+    if let Some(p) = &metadata.precondition {
+        pairs.insert("pre-device".to_owned(), p.device.join(SEP));
+        if !p.build.is_empty() {
+            pairs.insert("pre-build".to_owned(), p.build.join(SEP));
+            pairs.insert(
+                "pre-build-incremental".to_owned(),
+                p.build_incremental.clone(),
+            );
+        }
+    }
+
+    if metadata.spl_downgrade {
+        pairs.insert("spl-downgrade".to_owned(), "yes".to_owned());
+    }
+
+    pairs.extend(metadata.property_files.clone());
+
+    let legacy_metadata = pairs
+        .into_iter()
+        .map(|(k, v)| format!("{k}={v}\n"))
+        .collect::<String>();
+    let modern_metadata = util::write_protobuf(metadata)?;
+
+    Ok((legacy_metadata, modern_metadata))
+}
+
+#[derive(Clone, Debug)]
+pub struct ZipEntry {
+    pub name: String,
+    pub offset: u64,
+    pub size: u64,
+}
+
+/// Parse OTA property files string.
+pub fn parse_property_files(data: &str) -> Result<Vec<ZipEntry>> {
+    let mut result = vec![];
+
+    for entry in data.trim_end().split(',') {
+        let mut pieces = entry.split(':');
+
+        let name = pieces
+            .next()
+            .map(|p| p.to_owned())
+            .ok_or_else(|| Error::InvalidPropertyFileEntry(entry.to_owned()))?;
+        let offset = pieces
+            .next()
+            .and_then(|p| p.parse::<u64>().ok())
+            .ok_or_else(|| Error::InvalidPropertyFileEntry(entry.to_owned()))?;
+        let size = pieces
+            .next()
+            .and_then(|p| p.parse::<u64>().ok())
+            .ok_or_else(|| Error::InvalidPropertyFileEntry(entry.to_owned()))?;
+
+        if pieces.next().is_some() {
+            return Err(Error::InvalidPropertyFileEntry(entry.to_owned()));
+        }
+
+        result.push(ZipEntry { name, offset, size });
+    }
+
+    Ok(result)
+}
+
+/// Compute the property files entries listing the offsets and sizes to every
+/// zip entry.
+fn compute_property_files(
+    pf_name: &str,
+    entries: &[ZipEntry],
+    max_length: Option<usize>,
+) -> Result<String> {
+    let compute = |path: &'static str| -> Result<String> {
+        let entry = entries
+            .iter()
+            .find(|e| e.name == path)
+            .ok_or_else(|| Error::MissingZipEntry(path))?;
+        let name = path.rsplit_once('/').map_or(path, |p| p.1);
+
+        Ok(format!("{name}:{}:{}", entry.offset, entry.size))
+    };
+
+    let mut tokens = vec![];
+
+    if pf_name == PF_NAME {
+        tokens.push(compute(NAME_PAYLOAD_METADATA)?);
+    }
+
+    for path in [PATH_PAYLOAD, PATH_PROPERTIES] {
+        tokens.push(compute(path)?);
+    }
+
+    for path in [
+        "apex_info.pb",
+        "care_map.pb",
+        "care_map.txt",
+        "compatibility.zip",
+    ] {
+        if let Ok(token) = compute(path) {
+            tokens.push(token);
+        }
+    }
+
+    if max_length.is_none() {
+        tokens.push(format!("metadata:{}", " ".repeat(15)));
+        tokens.push(format!("metadata.pb:{}", " ".repeat(15)));
+    } else {
+        tokens.push(compute(PATH_METADATA)?);
+        tokens.push(compute(PATH_METADATA_PB)?);
+    }
+
+    let mut joined = tokens.join(",");
+
+    if let Some(l) = max_length {
+        if joined.len() > l {
+            return Err(Error::InsufficientReservedSpace(joined, l));
+        }
+
+        let remain = l - joined.len();
+        joined.extend(iter::repeat(' ').take(remain));
+    }
+
+    Ok(joined)
+}
+
+// Add fake payload_metadata.bin entry, covering the header + header signature
+// regions of the payload.
+fn add_payload_metadata_entry(
+    entries: &mut Vec<ZipEntry>,
+    payload_metadata_size: u64,
+) -> Result<()> {
+    let payload_offset = entries
+        .iter()
+        .find(|e| e.name == PATH_PAYLOAD)
+        .ok_or_else(|| Error::MissingZipEntry(PATH_PAYLOAD))?
+        .offset;
+    entries.push(ZipEntry {
+        name: NAME_PAYLOAD_METADATA.to_owned(),
+        offset: payload_offset,
+        size: payload_metadata_size,
+    });
+
+    Ok(())
+}
+
+/// Add metadata files to the output OTA zip. `zip_entries` is the list of
+/// [`ZipEntry`] already written to `zip_writer`. `next_offset` is the current
+/// file offset (where the next zip entry's local header begins).
+/// `metadata_pb_raw` is the serialized OTA metadata protobuf message from the
+/// original OTA. `payload_metadata_size` is the size of the new payload's
+/// metadata and metadata signature regions.
+///
+/// The zip file's backing file position MUST BE set to where the central
+/// directory would start.
+pub fn add_metadata(
+    zip_entries: &[ZipEntry],
+    zip_writer: &mut ZipWriter<impl Write>,
+    next_offset: u64,
+    metadata_pb_raw: &[u8],
+    payload_metadata_size: u64,
+) -> Result<OtaMetadata> {
+    let mut metadata: OtaMetadata = util::read_protobuf(metadata_pb_raw)?;
+    let options = FileOptions::default().compression_method(CompressionMethod::Stored);
+
+    let mut zip_entries = zip_entries.to_owned();
+    add_payload_metadata_entry(&mut zip_entries, payload_metadata_size)?;
+
+    // Compute initial property files with reserved space as placeholders to
+    // store the self-referential metadata entries later.
+    metadata.property_files.clear();
+    for pf in [PF_NAME, PF_STREAMING_NAME] {
+        metadata.property_files.insert(
+            pf.to_owned(),
+            compute_property_files(pf, &zip_entries, None)?,
+        );
+    }
+
+    // Add the placeholders to a temporary zip to compute final property files.
+    let (temp_legacy_offset, temp_modern_offset) = {
+        let (legacy_raw, modern_raw) = serialize_metadata(&metadata)?;
+        let mut writer = ZipWriter::new_streaming(Cursor::new(Vec::new()));
+
+        writer.start_file_with_extra_data(PATH_METADATA, options)?;
+        let legacy_offset = writer.end_extra_data()?;
+        writer.write_all(legacy_raw.as_bytes())?;
+
+        writer.start_file_with_extra_data(PATH_METADATA_PB, options)?;
+        let modern_offset = writer.end_extra_data()?;
+        writer.write_all(&modern_raw)?;
+
+        zip_entries.push(ZipEntry {
+            name: PATH_METADATA.to_owned(),
+            offset: next_offset + legacy_offset,
+            size: legacy_raw.len() as u64,
+        });
+        zip_entries.push(ZipEntry {
+            name: PATH_METADATA_PB.to_owned(),
+            offset: next_offset + modern_offset,
+            size: modern_raw.len() as u64,
+        });
+
+        (next_offset + legacy_offset, next_offset + modern_offset)
+    };
+
+    // Compute the final property files using the offsets of the fake entries.
+    for (key, value) in &mut metadata.property_files {
+        *value = compute_property_files(key, &zip_entries, Some(value.len()))?;
+    }
+
+    // Add the final metadata files to the real zip.
+    {
+        let (legacy_raw, modern_raw) = serialize_metadata(&metadata)?;
+
+        zip_writer.start_file_with_extra_data(PATH_METADATA, options)?;
+        let legacy_offset = zip_writer.end_extra_data()?;
+        zip_writer.write_all(legacy_raw.as_bytes())?;
+
+        zip_writer.start_file_with_extra_data(PATH_METADATA_PB, options)?;
+        let modern_offset = zip_writer.end_extra_data()?;
+        zip_writer.write_all(&modern_raw)?;
+
+        assert_eq!(legacy_offset, temp_legacy_offset);
+        assert_eq!(modern_offset, temp_modern_offset);
+    }
+
+    Ok(metadata)
+}
+
+/// Verify that the zip entry offsets and sizes match the OTA metadata.
+pub fn verify_metadata(
+    reader: impl Read + Seek,
+    metadata: &OtaMetadata,
+    payload_metadata_size: u64,
+) -> Result<()> {
+    let mut zip_reader = ZipArchive::new(reader)?;
+    let mut zip_entries = vec![];
+
+    for i in 0..zip_reader.len() {
+        let entry = zip_reader.by_index(i)?;
+        zip_entries.push(ZipEntry {
+            name: entry.name().to_owned(),
+            offset: entry.data_start(),
+            size: entry.size(),
+        });
+    }
+
+    add_payload_metadata_entry(&mut zip_entries, payload_metadata_size)?;
+
+    for (key, value) in &metadata.property_files {
+        let new_value = compute_property_files(key, &zip_entries, Some(value.len()))?;
+        if *value != new_value {
+            return Err(Error::MismatchedPropertyFiles(value.clone(), new_value));
+        }
+    }
+
+    Ok(())
+}
+
+/// Parse the CMS signature from the OTA zip comment. Returns the decoded CMS
+/// [`SignedData`] structure and the length of the file (from the beginning)
+/// that's covered by the signature. This does not perform any parsing of zip
+/// data structures.
+fn parse_ota_sig(mut reader: impl Read + Seek) -> Result<(SignedData, u64)> {
+    let file_size = reader.seek(SeekFrom::End(0))?;
+
+    reader.seek(SeekFrom::Current(-6))?;
+    let mut footer = [0u8; 6];
+    reader.read_exact(&mut footer)?;
+
+    let abs_eoc_offset = u16::from_le_bytes(footer[0..2].try_into().unwrap());
+    let sig_magic = u16::from_le_bytes(footer[2..4].try_into().unwrap());
+    let comment_size = u16::from_le_bytes(footer[4..6].try_into().unwrap());
+
+    if sig_magic != 0xffff {
+        return Err(Error::OtaMagicNotFound);
+    }
+
+    // RecoverySystem.verifyPackage() always assumes a non-zip64 EOCD, so we'll
+    // do the same.
+    let eocd_size = u64::from(22 + comment_size);
+    if file_size < eocd_size {
+        return Err(Error::ZipTooSmall);
+    } else if u64::from(abs_eoc_offset) > eocd_size {
+        return Err(Error::SignatureOffsetTooLarge);
+    }
+
+    reader.seek(SeekFrom::Start(file_size - eocd_size))?;
+    let mut eocd = vec![0u8; eocd_size as usize];
+    reader.read_exact(&mut eocd)?;
+
+    let mut eocd_magic_iter = memmem::find_iter(&eocd, ZIP_EOCD_MAGIC);
+    if eocd_magic_iter.next() != Some(0) {
+        return Err(Error::EocdMagicNotFound);
+    }
+    if eocd_magic_iter.next().is_some() {
+        return Err(Error::EocdMagicInComment);
+    }
+
+    let sig_offset = eocd_size as usize - usize::from(abs_eoc_offset);
+    let sd = crypto::parse_cms(&eocd[sig_offset..eocd_size as usize - 6])?;
+    // The signature covers everything aside from the archive comment and its
+    // length field.
+    let hashed_size = file_size - 2 - u64::from(comment_size);
+
+    Ok((sd, hashed_size))
+}
+
+/// Verify an OTA zip against its embedded certificates. This function makes no
+/// assertion about whether the certificate is actually trusted. Returns the
+/// embedded certificate.
+///
+/// CMS signed attributes are intentionally not supported because AOSP recovery
+/// does not support them either. It expects the CMS [`SignedData`] structure to
+/// be used for nothing more than a raw signature transport mechanism.
+pub fn verify_ota(
+    mut reader: impl Read + Seek,
+    cancel_signal: &Arc<AtomicBool>,
+) -> Result<Certificate> {
+    let (sd, hashed_size) = parse_ota_sig(&mut reader)?;
+
+    // Make sure the certificate in the CMS structure matches the otacert zip
+    // entry.
+    let certs = crypto::get_cms_certs(&sd);
+    if certs.len() != 1 {
+        return Err(Error::NotOneCmsCertificate(certs.len()));
+    }
+
+    let cert = &certs[0];
+    let public_key = crypto::get_public_key(cert)?;
+
+    // Make sure this is a signature scheme we can handle. There's currently no
+    // Rust library to verify arbitrary CMS signatures for large files without
+    // fully reading them into memory.
+    if sd.signer_infos.0.len() != 1 {
+        return Err(Error::NotOneCmsSignerInfo(sd.signer_infos.0.len()));
+    }
+
+    let signer = sd.signer_infos.0.get(0).unwrap();
+    if signer.digest_alg.oid != rfc5912::ID_SHA_256 && signer.digest_alg.oid != rfc5912::ID_SHA_1 {
+        return Err(Error::UnsupportedDigestAlgorithm(signer.digest_alg.oid));
+    } else if signer.signature_algorithm.oid != rfc5912::RSA_ENCRYPTION
+        && signer.signature_algorithm.oid != rfc5912::SHA_256_WITH_RSA_ENCRYPTION
+    {
+        return Err(Error::UnsupportedSignatureAlgorithm(
+            signer.signature_algorithm.oid,
+        ));
+    }
+
+    // Manually hash the parts of the file covered by the signature.
+    reader.seek(SeekFrom::Start(0))?;
+
+    // We support SHA1 for verification only.
+    let (algorithm, scheme) = if signer.digest_alg.oid == rfc5912::ID_SHA_256 {
+        (&ring::digest::SHA256, Pkcs1v15Sign::new::<Sha256>())
+    } else {
+        (
+            &ring::digest::SHA1_FOR_LEGACY_USE_ONLY,
+            Pkcs1v15Sign::new::<Sha1>(),
+        )
+    };
+
+    let mut hashing_reader = HashingReader::new(reader, Context::new(algorithm));
+
+    stream::copy_n(&mut hashing_reader, io::sink(), hashed_size, cancel_signal)?;
+
+    let (_, context) = hashing_reader.finish();
+    let digest = context.finish();
+
+    // Verify the signature against the public key.
+    public_key.verify(scheme, digest.as_ref(), signer.signature.as_bytes())?;
+
+    Ok(cert.clone())
+}
+
+/// Get and parse the protobuf-encoded OTA metadata, the PEM-encoded otacert,
+/// the payload header, and the payload properties from an OTA zip.
+pub fn parse_zip_ota_info(
+    reader: impl Read + Seek,
+) -> Result<(OtaMetadata, Certificate, PayloadHeader, String)> {
+    let mut zip = ZipArchive::new(reader)?;
+
+    let metadata = {
+        let mut entry = zip.by_name(PATH_METADATA_PB)?;
+        let mut buf = vec![0u8; entry.size() as usize];
+        entry.read_exact(&mut buf)?;
+        util::read_protobuf::<OtaMetadata>(&buf)?
+    };
+
+    let certificate = {
+        let entry = zip.by_name(PATH_OTACERT)?;
+        crypto::read_pem_cert(entry)?
+    };
+
+    let header = {
+        let entry = zip.by_name(PATH_PAYLOAD)?;
+        PayloadHeader::from_reader(entry)?
+    };
+
+    let properties = {
+        let mut entry = zip.by_name(PATH_PROPERTIES)?;
+        let mut buf = String::new();
+        entry.read_to_string(&mut buf)?;
+        buf
+    };
+
+    Ok((metadata, certificate, header, properties))
+}
+
+/// A writer that produces a signapk-style signed zip file with a whole-file
+/// signature stored in the zip archive comment. The data will be left in an
+/// unusable state if [`Self::finish()`] is not called.
+pub struct SigningWriter<W: Write> {
+    inner: HashingWriter<W>,
+    // Android only supports non-zip64 EOCD.
+    queue: [u8; 22],
+    used: usize,
+}
+
+impl<W: Write> SigningWriter<W> {
+    pub fn new(inner: W) -> Self {
+        Self {
+            inner: HashingWriter::new(inner, Context::new(&ring::digest::SHA256)),
+            queue: Default::default(),
+            used: 0,
+        }
+    }
+
+    pub fn finish(mut self, key: &RsaPrivateKey, cert: &Certificate) -> Result<W> {
+        if self.used < self.queue.len() {
+            return Err(
+                io::Error::new(io::ErrorKind::InvalidData, "Too small to contain EOCD").into(),
+            );
+        } else if &self.queue[..4] != b"PK\x05\x06" {
+            return Err(io::Error::new(io::ErrorKind::InvalidData, "EOCD magic not found").into());
+        } else if &self.queue[20..22] != b"\0\0" {
+            return Err(io::Error::new(
+                io::ErrorKind::InvalidData,
+                "Archive comment is not 0 bytes",
+            )
+            .into());
+        }
+
+        // Chop off the archive comment size field and write the remaining data.
+        self.inner.write_all(&self.queue[..20])?;
+
+        let (mut raw_writer, context) = self.inner.finish();
+        let digest = context.finish();
+
+        let cms_signature = crypto::cms_sign_external(key, cert, digest.as_ref())?;
+        let cms_signature_der = cms_signature.to_der()?;
+
+        let mut comment = COMMENT_MESSAGE.to_vec();
+        comment.extend(&cms_signature_der);
+
+        let comment_size = comment.len() + 6;
+
+        // Absolute value of the offset of the signature from the end of the
+        // archive comment.
+        comment.extend((cms_signature_der.len() as u16 + 6).to_le_bytes());
+
+        // Magic value.
+        comment.extend(b"\xff\xff");
+
+        // EOCD archive comment size.
+        comment.extend(((comment_size) as u16).to_le_bytes());
+
+        if let Some(o) = memmem::find(&comment, ZIP_EOCD_MAGIC) {
+            return Err(io::Error::new(
+                io::ErrorKind::InvalidData,
+                format!("Archive comment contains EOCD magic at offset {o}"),
+            )
+            .into());
+        }
+
+        // Write EOCD comment size field, which was removed before.
+        raw_writer.write_all(&((comment_size) as u16).to_le_bytes())?;
+
+        // Finally, write the comment.
+        raw_writer.write_all(&comment)?;
+
+        Ok(raw_writer)
+    }
+}
+
+impl<W: Write> Write for SigningWriter<W> {
+    fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
+        let (front, back) = buf.split_at(buf.len().saturating_sub(self.queue.len()));
+        if front.is_empty() {
+            // Write data from the front of the queue while keeping it as full
+            // as possible.
+            let n_from_queue = (self.used + back.len()).saturating_sub(self.queue.len());
+            self.inner.write_all(&self.queue[..n_from_queue])?;
+
+            // Move unused queued bytes to the front.
+            self.queue.rotate_left(n_from_queue);
+            self.used -= n_from_queue;
+
+            // Add the remaining data to the queue.
+            self.queue[self.used..self.used + back.len()].copy_from_slice(back);
+            self.used += back.len();
+        } else {
+            // We have enough data in the back to fill the entire queue.
+            self.inner.write_all(&self.queue[..self.used])?;
+            self.inner.write_all(front)?;
+
+            self.queue.copy_from_slice(back);
+            self.used = self.queue.len();
+        }
+
+        Ok(buf.len())
+    }
+
+    fn flush(&mut self) -> io::Result<()> {
+        self.inner.flush()
+    }
+}

avbroot/src/format/padding.rs

@@ -0,0 +1,49 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::io::{self, Read, Seek, Write};
+
+use num_traits::PrimInt;
+
+use crate::stream::{ReadDiscardExt, WriteZerosExt};
+
+/// Calculate the amount of padding that needs to be added to align the
+/// specified offset to a page boundary.
+pub fn calc<N: PrimInt>(offset: N, page_size: N) -> N {
+    let r = offset % page_size;
+    if r == N::zero() {
+        N::zero()
+    } else {
+        page_size - r
+    }
+}
+
+/// Round to the next multiple of the page size.
+pub fn round<N: PrimInt>(offset: N, page_size: N) -> Option<N> {
+    let remain = calc(offset, page_size);
+    offset.checked_add(&remain)
+}
+
+/// Read and discard data until the next multiple of the page size. [`Seek`] is
+/// only used for querying the file position.
+pub fn read_discard(mut reader: impl Read + Seek, page_size: u64) -> io::Result<u64> {
+    let pos = reader.stream_position()?;
+    let padding = calc(pos, page_size);
+
+    reader.read_discard_exact(padding)?;
+
+    Ok(padding)
+}
+
+/// Write zeros until the next multiple of the page size. [`Seek`] is only used
+/// for querying the file position.
+pub fn write_zeros(mut writer: impl Write + Seek, page_size: u64) -> io::Result<u64> {
+    let pos = writer.stream_position()?;
+    let padding = calc(pos, page_size);
+
+    writer.write_zeros_exact(padding)?;
+
+    Ok(padding)
+}

avbroot/src/format/payload.rs

@@ -0,0 +1,941 @@
+/*
+ * SPDX-FileCopyrightText: 2022-2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::{
+    collections::{HashMap, HashSet},
+    io::{self, Cursor, Read, Seek, SeekFrom, Write},
+    sync::{atomic::AtomicBool, Arc},
+};
+
+use base64::engine::general_purpose::STANDARD;
+use base64::Engine;
+use byteorder::{BigEndian, ReadBytesExt};
+use bzip2::write::BzDecoder;
+use num_traits::ToPrimitive;
+use quick_protobuf::MessageWrite;
+use rayon::prelude::{IntoParallelIterator, IntoParallelRefIterator, ParallelIterator};
+use ring::digest::{Context, Digest};
+use rsa::{traits::PublicKeyParts, Pkcs1v15Sign, RsaPrivateKey};
+use sha2::Sha256;
+use thiserror::Error;
+use x509_cert::Certificate;
+use xz2::{
+    stream::{Check, Stream},
+    write::XzDecoder,
+    write::XzEncoder,
+};
+
+use crate::{
+    crypto,
+    protobuf::chromeos_update_engine::{
+        mod_InstallOperation, mod_Signatures::Signature, DeltaArchiveManifest, Extent,
+        InstallOperation, PartitionInfo, PartitionUpdate, Signatures,
+    },
+    stream::{
+        self, CountingReader, CountingWriter, FromReader, HashingWriter, ReadDiscardExt, ReadSeek,
+        SharedCursor, WriteSeek,
+    },
+    util,
+};
+
+const OTA_MAGIC: &[u8; 4] = b"CrAU";
+const OTA_HEADER_SIZE: usize = OTA_MAGIC.len() + 8 + 8 + 4;
+
+#[derive(Debug, Error)]
+pub enum Error {
+    #[error("Unknown magic: {0:?}")]
+    UnknownMagic([u8; 4]),
+    #[error("Unsupported payload version: {0}")]
+    UnsupportedVersion(u64),
+    #[error("File is a delta OTA, not a full OTA")]
+    NotFullOta,
+    #[error("Payload contains no signatures")]
+    NoSignatures,
+    #[error("Blob offset should be {0}, but is {1}")]
+    InvalidBlobOffset(u64, u64),
+    #[error("Payload signatures offset should be {0}, but is {1}")]
+    InvalidPayloadSignaturesOffset(u64, u64),
+    #[error("Invalid payload properties line: {0:?}")]
+    InvalidPropertiesLine(String),
+    #[error("Duplicate payload property: {0:?}")]
+    DuplicateProperty(String),
+    #[error("Payload property {0:?} ({1:?}) does not match expected value {2:?}")]
+    InvalidProperty(String, String, Option<String>),
+    #[error("Unsupported partition operation: {0:?}")]
+    UnsupportedOperation(mod_InstallOperation::Type),
+    #[error("Expected sha256 {0:?}, but have {1:?}")]
+    MismatchedDigest(Option<String>, String),
+    #[error("Size of {0} ({1}) is not aligned to the block size ({2})")]
+    InvalidPartitionSize(String, u64, u32),
+    #[error("Partition not found in payload: {0}")]
+    MissingPartition(String),
+    #[error("Partitions not found in payload: {0:?}")]
+    MissingPartitions(HashSet<String>),
+    #[error("{0:?} field is missing")]
+    MissingField(&'static str),
+    #[error("{0:?} field exceeds integer bounds")]
+    IntegerTooLarge(&'static str),
+    #[error("Crypto error")]
+    Crypto(#[from] crypto::Error),
+    #[error("Protobuf error")]
+    Protobuf(#[from] quick_protobuf::Error),
+    #[error("XZ stream error")]
+    XzStream(#[from] xz2::stream::Error),
+    #[error("RSA error")]
+    Rsa(#[from] rsa::Error),
+    #[error("I/O error")]
+    Io(#[from] io::Error),
+}
+
+type Result<T> = std::result::Result<T, Error>;
+
+#[derive(Clone, Debug)]
+pub struct PayloadHeader {
+    pub version: u64,
+    pub manifest: DeltaArchiveManifest,
+    pub metadata_signature_size: u32,
+    pub blob_offset: u64,
+}
+
+impl<R: Read> FromReader<R> for PayloadHeader {
+    type Error = Error;
+
+    /// Parse the header from an OTA payload file. After this function returns,
+    /// the file position is set to the beginning of the blob section.
+    fn from_reader(reader: R) -> Result<Self> {
+        let mut reader = CountingReader::new(reader);
+
+        let mut magic = [0u8; 4];
+        reader.read_exact(&mut magic)?;
+        if magic != *OTA_MAGIC {
+            return Err(Error::UnknownMagic(magic));
+        }
+
+        let version = reader.read_u64::<BigEndian>()?;
+        if version != 2 {
+            return Err(Error::UnsupportedVersion(version));
+        }
+
+        let manifest_size = reader
+            .read_u64::<BigEndian>()?
+            .to_usize()
+            .ok_or_else(|| Error::IntegerTooLarge("manifest_size"))?;
+        let metadata_signature_size = reader.read_u32::<BigEndian>()?;
+
+        let mut manifest_raw = vec![0u8; manifest_size];
+        reader.read_exact(&mut manifest_raw)?;
+        let manifest: DeltaArchiveManifest = util::read_protobuf(&manifest_raw)?;
+
+        // Fail as soon as possible since it's impossible to support delta OTAs.
+        if manifest
+            .partitions
+            .iter()
+            .any(|p| p.old_partition_info.is_some())
+        {
+            return Err(Error::NotFullOta);
+        }
+
+        // Skip manifest signatures.
+        reader.read_discard_exact(metadata_signature_size.into())?;
+
+        Ok(Self {
+            version,
+            manifest,
+            metadata_signature_size,
+            blob_offset: reader.stream_position()?,
+        })
+    }
+}
+
+/// Sign `digest` with `key` and return a [`Signatures`] protobuf struct with
+/// the signature padded to the maximum size.
+fn sign_digest(digest: &[u8], key: &RsaPrivateKey) -> Result<Signatures> {
+    let scheme = Pkcs1v15Sign::new::<Sha256>();
+    let mut digest_signed = key.sign(scheme, digest)?;
+    assert!(
+        digest_signed.len() <= key.size(),
+        "Signature exceeds maximum size",
+    );
+
+    let unpadded_size = digest_signed.len();
+    digest_signed.resize(key.size(), 0);
+
+    let signature = Signature {
+        data: Some(digest_signed),
+        // Always fits in even a u16.
+        unpadded_signature_size: Some(unpadded_size as u32),
+    };
+
+    let signatures = Signatures {
+        signatures: vec![signature],
+    };
+
+    Ok(signatures)
+}
+
+/// Verify `digest` inside `signatures` using `cert`.
+fn verify_digest(digest: &[u8], signatures: &Signatures, cert: &Certificate) -> Result<()> {
+    let public_key = crypto::get_public_key(cert)?;
+    let mut last_error = None;
+
+    for signature in &signatures.signatures {
+        let Some(data) = &signature.data else {
+            continue;
+        };
+        let Some(size) = signature.unpadded_signature_size else {
+            continue;
+        };
+        let without_padding = &data[..size as usize];
+
+        let scheme = Pkcs1v15Sign::new::<Sha256>();
+        match public_key.verify(scheme, digest, without_padding) {
+            Ok(_) => return Ok(()),
+            Err(e) => last_error = Some(e),
+        }
+    }
+
+    Err(last_error.map_or(Error::NoSignatures, |e| e.into()))
+}
+
+fn parse_properties(data: &str) -> Result<HashMap<String, String>> {
+    let mut result = HashMap::new();
+
+    for line in data.split('\n') {
+        if line.is_empty() {
+            continue;
+        }
+
+        let Some((key, value)) = line.split_once('=') else {
+            return Err(Error::InvalidPropertiesLine(line.to_owned()));
+        };
+
+        if result.insert(key.to_owned(), value.to_owned()).is_some() {
+            return Err(Error::DuplicateProperty(key.to_owned()));
+        }
+    }
+
+    Ok(result)
+}
+
+/// Generate `payload_properties.txt` contents. The file hash and size include
+/// the signature sections, but the metadata hash and size do not.
+fn generate_properties(
+    file_hash: &[u8],
+    file_size: u64,
+    metadata_hash: &[u8],
+    metadata_size: u64,
+) -> String {
+    let mut properties = String::new();
+
+    properties.push_str("FILE_HASH=");
+    STANDARD.encode_string(file_hash, &mut properties);
+    properties.push('\n');
+
+    properties.push_str("FILE_SIZE=");
+    properties.push_str(&file_size.to_string());
+    properties.push('\n');
+
+    properties.push_str("METADATA_HASH=");
+    STANDARD.encode_string(metadata_hash, &mut properties);
+    properties.push('\n');
+
+    properties.push_str("METADATA_SIZE=");
+    properties.push_str(&metadata_size.to_string());
+    properties.push('\n');
+
+    properties
+}
+
+/// A writer for producing signed `payload.bin` files.
+pub struct PayloadWriter<W: Write> {
+    inner: W,
+    header: PayloadHeader,
+    /// Metadata (header + manifest) only, excluding the metadata signature.
+    metadata_hash: Digest,
+    metadata_size: usize,
+    /// Index of `header.manifest.partitions[]` for current entry.
+    partition_index: Option<usize>,
+    /// Index of `header.manifest.partitions[].operations[]` for current entry.
+    operation_index: Option<usize>,
+    /// Whether a next entry exists.
+    done: bool,
+    /// Number of bytes written for the current entry.
+    written: u64,
+    /// Excludes signatures (hashes are for signing).
+    h_partial: Context,
+    /// Includes signatures (hashes are for properties file).
+    h_full: Context,
+    key: RsaPrivateKey,
+}
+
+/// Write data to a writer and one or more hashers.
+macro_rules! write_hash {
+    ($writer:expr, [$($hasher:expr),+], $data:expr $(,)?) => {
+        {
+            let data = $data;
+            $(
+                $hasher.update(data);
+            )+
+            $writer.write_all(data)
+        }
+    };
+}
+
+impl<W: Write> PayloadWriter<W> {
+    /// Create a new payload writer. All information in `header` is final and
+    /// cannot be changed after this function returns since it'll already have
+    /// been committed to the writer. The [`InstallOperation::data_offset`]
+    /// fields are ignored and internally recomputed to guarantee that there are
+    /// no gaps. All partitions' install operation data is written to the blob
+    /// section in order.
+    pub fn new(mut inner: W, mut header: PayloadHeader, key: RsaPrivateKey) -> Result<Self> {
+        let mut blob_size = 0;
+
+        // The blob must contain all data in sequential order with no gaps.
+        for p in &mut header.manifest.partitions {
+            for op in &mut p.operations {
+                op.data_offset = Some(blob_size);
+
+                if let Some(length) = op.data_length {
+                    blob_size += length;
+                }
+            }
+        }
+
+        // Get the length of an dummy signature struct since the length fields
+        // are part of the data to be signed.
+        let dummy_sig = sign_digest(
+            ring::digest::digest(&ring::digest::SHA256, b"").as_ref(),
+            &key,
+        )?;
+        let dummy_sig_size = dummy_sig.get_size();
+
+        // Fill out the new payload signature information.
+        header.manifest.signatures_offset = Some(blob_size);
+        header.manifest.signatures_size = Some(dummy_sig_size as u64);
+
+        // Build new manifest.
+        let manifest_raw_new = util::write_protobuf(&header.manifest)?;
+
+        // Excludes signatures (hashes are for signing).
+        let mut h_partial = Context::new(&ring::digest::SHA256);
+        // Includes signatures (hashes are for properties file).
+        let mut h_full = Context::new(&ring::digest::SHA256);
+
+        // Write header to output file.
+        write_hash!(inner, [h_partial, h_full], OTA_MAGIC)?;
+        write_hash!(inner, [h_partial, h_full], &header.version.to_be_bytes())?;
+        write_hash!(
+            inner,
+            [h_partial, h_full],
+            &(manifest_raw_new.len() as u64).to_be_bytes(),
+        )?;
+        write_hash!(
+            inner,
+            [h_partial, h_full],
+            &(dummy_sig_size as u32).to_be_bytes()
+        )?;
+
+        // Write new manifest.
+        write_hash!(inner, [h_partial, h_full], &manifest_raw_new)?;
+
+        // Sign metadata (header + manifest) hash. The signature is not included
+        // in the payload hash.
+        let metadata_hash = h_partial.clone().finish();
+        let metadata_sig = sign_digest(metadata_hash.as_ref(), &key)?;
+        let metadata_sig_raw = util::write_protobuf(&metadata_sig)?;
+        write_hash!(inner, [h_full], &metadata_sig_raw)?;
+
+        Ok(Self {
+            inner,
+            header,
+            metadata_hash,
+            metadata_size: OTA_HEADER_SIZE + manifest_raw_new.len(),
+            partition_index: None,
+            operation_index: None,
+            done: false,
+            written: 0,
+            h_partial,
+            h_full,
+            key,
+        })
+    }
+
+    /// Finalize the payload. If this function is not called, the payload will
+    /// be left in an incomplete state. Returns the original writer, the
+    /// contents that should be written for `payload_properties.txt` and the
+    /// length of the header + manifest + manifest signature sections (for
+    /// constructing the `payload_metadata.bin` OTA metadata property files
+    /// entry).
+    pub fn finish(mut self) -> Result<(W, String, u64)> {
+        // Append payload signature.
+        let payload_partial_hash = self.h_partial.clone().finish();
+        let payload_sig = sign_digest(payload_partial_hash.as_ref(), &self.key)?;
+        let payload_sig_raw = util::write_protobuf(&payload_sig)?;
+        write_hash!(self.inner, [self.h_full], &payload_sig_raw)?;
+
+        // Everything before the blob.
+        let metadata_with_sig_size =
+            self.metadata_size as u64 + self.header.manifest.signatures_size.unwrap();
+        // Whole file, including both signatures.
+        let new_file_size = metadata_with_sig_size
+            + self.header.manifest.signatures_offset.unwrap()
+            + self.header.manifest.signatures_size.unwrap();
+
+        let full_digest = self.h_full.finish();
+
+        let properties = generate_properties(
+            full_digest.as_ref(),
+            new_file_size,
+            self.metadata_hash.as_ref(),
+            self.metadata_size as u64,
+        );
+
+        Ok((self.inner, properties, metadata_with_sig_size))
+    }
+
+    /// Prepare for writing the next source data blob corresponding to an
+    /// [`InstallOperation`]. To write all of the payload data, call this method
+    /// followed by [`Self::write()`] repeatedly until `Ok(false)` is returned
+    /// or an error occurs. This function will fail if the amount of data
+    /// written for the previous operation does not match
+    /// [`InstallOperation::data_length`].
+    pub fn begin_next_operation(&mut self) -> Result<bool> {
+        if let Some(operation) = self.operation() {
+            // ZERO/DISCARD operations will not have a length.
+            if self.written < operation.data_length.unwrap_or(0) {
+                return Err(io::Error::new(
+                    io::ErrorKind::InvalidInput,
+                    format!(
+                        "Expected {} bytes, but only wrote {} bytes",
+                        operation.data_length.unwrap(),
+                        self.written,
+                    ),
+                )
+                .into());
+            }
+        }
+
+        if let Some(pi) = &mut self.partition_index {
+            // Move to next entry.
+            loop {
+                // Try to move to next operation.
+                if let Some(oi) = &mut self.operation_index {
+                    *oi += 1;
+
+                    if *oi < self.header.manifest.partitions[*pi].operations.len() {
+                        break;
+                    } else {
+                        self.operation_index = None;
+                    }
+                }
+
+                // Try to move to next partition.
+                *pi += 1;
+
+                if *pi < self.header.manifest.partitions.len() {
+                    if !self.header.manifest.partitions[*pi].operations.is_empty() {
+                        self.operation_index = Some(0);
+                        break;
+                    }
+                } else {
+                    // No more partitions.
+                    self.partition_index = None;
+                    break;
+                }
+            }
+        } else if !self.done {
+            // Move to first entry.
+            if !self.header.manifest.partitions.is_empty()
+                && !self.header.manifest.partitions[0].operations.is_empty()
+            {
+                self.partition_index = Some(0);
+                self.operation_index = Some(0);
+            }
+        }
+
+        self.done = self.partition_index.is_none();
+        self.written = 0;
+
+        Ok(!self.done)
+    }
+
+    /// Get the partition index for the current entry. This is only valid when
+    /// [`Self::begin_next_operation()`] returns `true`.
+    pub fn partition_index(&self) -> Option<usize> {
+        self.partition_index
+    }
+
+    /// Get the install operation index for the current entry. This is only
+    /// valid when [`Self::begin_next_operation()`] returns `true`.
+    pub fn operation_index(&self) -> Option<usize> {
+        self.operation_index
+    }
+
+    /// Get the [`PartitionUpdate`] instance for the current entry. This is only
+    /// valid when [`Self::begin_next_operation()`] returns `true`.
+    pub fn partition(&self) -> Option<&PartitionUpdate> {
+        self.partition_index
+            .map(|pi| &self.header.manifest.partitions[pi])
+    }
+
+    /// Get the [`InstallOperation`] instance for the current entry. This is
+    /// only valid when [`Self::begin_next_operation()`] returns `true`.
+    pub fn operation(&self) -> Option<&InstallOperation> {
+        self.operation_index
+            .map(|oi| &self.partition().unwrap().operations[oi])
+    }
+}
+
+impl<W: Write> Write for PayloadWriter<W> {
+    /// Write data for the current partition install operation. The amount of
+    /// data written in total must match [`InstallOperation::data_length`].
+    fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
+        let pi = self.partition_index.expect("No partition selected");
+        let oi = self.operation_index.expect("No operation selected");
+        let partition = &self.header.manifest.partitions[pi];
+        let operation = &partition.operations[oi];
+
+        let to_write =
+            (operation.data_length.unwrap() - self.written).min(buf.len() as u64) as usize;
+        let n = self.inner.write(&buf[..to_write])?;
+
+        self.h_full.update(&buf[..n]);
+        self.h_partial.update(&buf[..n]);
+
+        self.written += n as u64;
+
+        Ok(n)
+    }
+
+    fn flush(&mut self) -> io::Result<()> {
+        self.inner.flush()
+    }
+}
+
+/// A writer to produce a compressed partition image suitable for directly
+/// inserting into a `payload.bin`'s blob section. The data is XZ-compressed at
+/// a low compression level, primarily to collapse zeros, and the
+/// [`PartitionUpdate`] instance is updated with the final size and hash. The
+/// entire data will be represented as a single [`InstallOperation`] and
+/// [`InstallOperation::data_offset`] will be set to `None`.
+pub struct CompressedPartitionWriter<W: Write> {
+    inner: XzEncoder<HashingWriter<CountingWriter<W>>>,
+    block_size: u32,
+    h_uncompressed: Context,
+    written: u64,
+}
+
+impl<W: Write> CompressedPartitionWriter<W> {
+    pub fn new(writer: W, block_size: u32) -> Result<Self> {
+        let counting_writer = CountingWriter::new(writer);
+        let hashing_writer =
+            HashingWriter::new(counting_writer, Context::new(&ring::digest::SHA256));
+
+        // AOSP's payload_consumer does not support CRC during decompression.
+        // Also, we intentionally pick the lowest compression level since we
+        // primarily care about squishing zeros. The non-zero portions of boot
+        // images are usually already-compressed kernels and ramdisks.
+        let stream = Stream::new_easy_encoder(0, Check::None)?;
+        let xz_writer = XzEncoder::new_stream(hashing_writer, stream);
+
+        Ok(Self {
+            inner: xz_writer,
+            block_size,
+            h_uncompressed: Context::new(&ring::digest::SHA256),
+            written: 0,
+        })
+    }
+
+    /// Finish writing and update the [`PartitionUpdate`] instance with the new
+    /// size, hash, and install operation metadata.
+    pub fn finish(mut self, partition: &mut PartitionUpdate) -> Result<W> {
+        if self.written % u64::from(self.block_size) != 0 {
+            return Err(Error::InvalidPartitionSize(
+                partition.partition_name.clone(),
+                self.written,
+                self.block_size,
+            ));
+        }
+
+        self.inner.flush()?;
+
+        let hashing_writer = self.inner.finish()?;
+        let digest_uncompressed = self.h_uncompressed.finish();
+        let (counting_writer, context_compressed) = hashing_writer.finish();
+        let digest_compressed = context_compressed.finish();
+        // XzEncoder::total_out() cannot be used for an exact byte count because
+        // XzEncoder::finish() writes data.
+        let (writer, size_compressed) = counting_writer.finish();
+
+        partition.new_partition_info = Some(PartitionInfo {
+            size: Some(self.written),
+            hash: Some(digest_uncompressed.as_ref().to_vec()),
+        });
+
+        let extent = Extent {
+            start_block: Some(0),
+            num_blocks: Some(self.written / u64::from(self.block_size)),
+        };
+
+        let operation = InstallOperation {
+            type_pb: mod_InstallOperation::Type::REPLACE_XZ,
+            // Must be manually updated by the caller.
+            data_offset: None,
+            data_length: Some(size_compressed),
+            src_extents: vec![],
+            src_length: None,
+            dst_extents: vec![extent],
+            dst_length: None,
+            data_sha256_hash: Some(digest_compressed.as_ref().to_vec()),
+            src_sha256_hash: None,
+        };
+
+        partition.operations.clear();
+        partition.operations.push(operation);
+
+        Ok(writer)
+    }
+}
+
+impl<W: Write> Write for CompressedPartitionWriter<W> {
+    fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
+        let n = self.inner.write(buf)?;
+        self.h_uncompressed.update(&buf[..n]);
+        self.written += n as u64;
+        Ok(n)
+    }
+
+    fn flush(&mut self) -> io::Result<()> {
+        self.inner.flush()
+    }
+}
+
+/// Verify the payload signatures using the specified certificate and check that
+/// the digests in `payload_properties.txt` are correct.
+pub fn verify_payload(
+    mut reader: impl Read + Seek,
+    cert: &Certificate,
+    properties_raw: &str,
+    cancel_signal: &Arc<AtomicBool>,
+) -> Result<()> {
+    let header = PayloadHeader::from_reader(&mut reader)?;
+    reader.rewind()?;
+
+    let payload_signatures_offset = header
+        .manifest
+        .signatures_offset
+        .ok_or_else(|| Error::MissingField("signatures_offset"))?;
+    let payload_signatures_size = header
+        .manifest
+        .signatures_size
+        .ok_or_else(|| Error::MissingField("signatures_size"))?;
+
+    // Excludes signatures (hashes are for signing).
+    let mut h_partial = Context::new(&ring::digest::SHA256);
+    // Includes signatures (hashes are for properties file).
+    let mut h_full = Context::new(&ring::digest::SHA256);
+
+    // Read from the beginning to the metadata signature.
+    let metadata_size = header.blob_offset - u64::from(header.metadata_signature_size);
+    stream::copy_n_inspect(
+        &mut reader,
+        io::sink(),
+        metadata_size,
+        |data| {
+            h_partial.update(data);
+            h_full.update(data);
+        },
+        cancel_signal,
+    )?;
+    let metadata_hash = h_partial.clone().finish();
+
+    // Read the metadata signatures.
+    let metadata_sigs = {
+        let mut writer = Cursor::new(Vec::new());
+
+        stream::copy_n_inspect(
+            &mut reader,
+            &mut writer,
+            header.metadata_signature_size.into(),
+            |data| h_full.update(data),
+            cancel_signal,
+        )?;
+
+        let buf = writer.into_inner();
+        util::read_protobuf::<Signatures>(&buf)?
+    };
+
+    // Check the metadata signatures.
+    verify_digest(metadata_hash.as_ref(), &metadata_sigs, cert)?;
+
+    // Check the blob offset.
+    {
+        let actual = reader.stream_position()?;
+        if header.blob_offset != actual {
+            return Err(Error::InvalidBlobOffset(header.blob_offset, actual));
+        }
+    }
+
+    // Read (and discard) all the payload blobs.
+    stream::copy_n_inspect(
+        &mut reader,
+        io::sink(),
+        payload_signatures_offset,
+        |data| {
+            h_partial.update(data);
+            h_full.update(data);
+        },
+        cancel_signal,
+    )?;
+    let payload_hash = h_partial.clone().finish();
+
+    // Check the payload signatures offset.
+    {
+        let expected = header.blob_offset + payload_signatures_offset;
+        let actual = reader.stream_position()?;
+        if expected != actual {
+            return Err(Error::InvalidPayloadSignaturesOffset(expected, actual));
+        }
+    }
+
+    // Read the payload signatures.
+    let payload_sigs = {
+        let mut writer = Cursor::new(Vec::new());
+
+        stream::copy_n_inspect(
+            &mut reader,
+            &mut writer,
+            payload_signatures_size,
+            |data| h_full.update(data),
+            cancel_signal,
+        )?;
+
+        let buf = writer.into_inner();
+        util::read_protobuf::<Signatures>(&buf)?
+    };
+
+    // Check the payload signatures.
+    verify_digest(payload_hash.as_ref(), &payload_sigs, cert)?;
+
+    // Check properties file.
+    let expected_properties_raw = generate_properties(
+        h_full.finish().as_ref(),
+        reader.stream_position()?,
+        metadata_hash.as_ref(),
+        metadata_size,
+    );
+
+    let expected_properties = parse_properties(properties_raw)?;
+    let actual_properties = parse_properties(&expected_properties_raw)?;
+
+    for (key, actual_value) in actual_properties {
+        let expected_value = expected_properties.get(&key);
+
+        if expected_value != Some(&actual_value) {
+            return Err(Error::InvalidProperty(
+                key,
+                actual_value,
+                expected_value.cloned(),
+            ));
+        }
+    }
+
+    Ok(())
+}
+
+/// Apply a partition operation from `reader` to `writer`.
+pub fn apply_operation(
+    mut reader: impl Read + Seek,
+    mut writer: impl Write + Seek,
+    block_size: u32,
+    blob_offset: u64,
+    op: &InstallOperation,
+    cancel_signal: &Arc<AtomicBool>,
+) -> Result<()> {
+    for extent in &op.dst_extents {
+        let start_block = extent
+            .start_block
+            .ok_or_else(|| Error::MissingField("start_block"))?;
+        let num_blocks = extent
+            .num_blocks
+            .ok_or_else(|| Error::MissingField("num_blocks"))?;
+
+        let out_offset = start_block
+            .checked_mul(block_size.into())
+            .ok_or_else(|| Error::IntegerTooLarge("out_offset"))?;
+        let out_data_length = num_blocks
+            .checked_mul(block_size.into())
+            .ok_or_else(|| Error::IntegerTooLarge("out_data_length"))?;
+
+        writer.seek(SeekFrom::Start(out_offset))?;
+
+        let mut hasher = Context::new(&ring::digest::SHA256);
+
+        match op.type_pb {
+            // Handle ZERO/DISCARD specially since they don't require access to
+            // the payload blob.
+            mod_InstallOperation::Type::ZERO | mod_InstallOperation::Type::DISCARD => {
+                stream::copy_n_inspect(
+                    io::repeat(0),
+                    &mut writer,
+                    out_data_length,
+                    |data| hasher.update(data),
+                    cancel_signal,
+                )?;
+            }
+            other => {
+                let data_offset = op
+                    .data_offset
+                    .ok_or_else(|| Error::MissingField("data_offset"))?;
+                let data_length = op
+                    .data_length
+                    .ok_or_else(|| Error::MissingField("data_length"))?;
+                let in_offset = blob_offset
+                    .checked_add(data_offset)
+                    .ok_or_else(|| Error::IntegerTooLarge("in_offset"))?;
+
+                reader.seek(SeekFrom::Start(in_offset))?;
+
+                match other {
+                    mod_InstallOperation::Type::REPLACE => {
+                        stream::copy_n_inspect(
+                            &mut reader,
+                            &mut writer,
+                            data_length,
+                            |data| hasher.update(data),
+                            cancel_signal,
+                        )?;
+                    }
+                    mod_InstallOperation::Type::REPLACE_BZ => {
+                        let mut decoder = BzDecoder::new(&mut writer);
+                        stream::copy_n_inspect(
+                            &mut reader,
+                            &mut decoder,
+                            data_length,
+                            |data| hasher.update(data),
+                            cancel_signal,
+                        )?;
+                        decoder.finish()?;
+                    }
+                    mod_InstallOperation::Type::REPLACE_XZ => {
+                        let mut decoder = XzDecoder::new(&mut writer);
+                        stream::copy_n_inspect(
+                            &mut reader,
+                            &mut decoder,
+                            data_length,
+                            |data| hasher.update(data),
+                            cancel_signal,
+                        )?;
+                        decoder.finish()?;
+                    }
+                    _ => return Err(Error::UnsupportedOperation(op.type_pb)),
+                }
+            }
+        }
+
+        let expected_digest = op.data_sha256_hash.as_deref();
+        let digest = hasher.finish();
+
+        if expected_digest != Some(digest.as_ref())
+            && op.type_pb != mod_InstallOperation::Type::ZERO
+        {
+            return Err(Error::MismatchedDigest(
+                expected_digest.map(hex::encode),
+                hex::encode(digest.as_ref()),
+            ));
+        }
+    }
+
+    Ok(())
+}
+
+/// Extract the specified image from the payload into memory. This is done
+/// multithreaded and uses rayon's global thread pool. `open_payload` will be
+/// called from multiple threads.
+pub fn extract_image_to_memory(
+    open_payload: impl Fn() -> io::Result<Box<dyn ReadSeek>> + Sync,
+    header: &PayloadHeader,
+    partition_name: &str,
+    cancel_signal: &Arc<AtomicBool>,
+) -> Result<SharedCursor> {
+    let partition = header
+        .manifest
+        .partitions
+        .iter()
+        .find(|p| p.partition_name == partition_name)
+        .ok_or_else(|| Error::MissingPartition(partition_name.to_owned()))?;
+    let stream = SharedCursor::default();
+
+    partition
+        .operations
+        .par_iter()
+        .map(|op| -> Result<()> {
+            let reader = open_payload()?;
+            let writer = stream.clone();
+
+            apply_operation(
+                reader,
+                writer,
+                header.manifest.block_size,
+                header.blob_offset,
+                op,
+                cancel_signal,
+            )?;
+
+            Ok(())
+        })
+        .collect::<Result<_>>()?;
+
+    Ok(stream)
+}
+
+/// Extract the specified partition images from the payload into writers. This
+/// is done multithreaded and uses rayon's global thread pool. `open_payload`
+/// and `open_output` will be called from multiple threads.
+pub fn extract_images<'a>(
+    open_payload: impl Fn() -> io::Result<Box<dyn ReadSeek>> + Sync,
+    open_output: impl Fn(&str) -> io::Result<Box<dyn WriteSeek>> + Sync,
+    header: &PayloadHeader,
+    partition_names: impl IntoIterator<Item = &'a str>,
+    cancel_signal: &Arc<AtomicBool>,
+) -> Result<()> {
+    let mut remaining = partition_names.into_iter().collect::<HashSet<_>>();
+    // We parallelize at the operation level or else one thread might get stuck
+    // processing a giant image.
+    let mut operations = vec![];
+
+    for p in &header.manifest.partitions {
+        if remaining.remove(p.partition_name.as_str()) {
+            for op in &p.operations {
+                operations.push((p.partition_name.as_str(), op));
+            }
+        }
+    }
+
+    if !remaining.is_empty() {
+        let remaining = remaining.iter().map(|&n| n.to_owned()).collect();
+        return Err(Error::MissingPartitions(remaining));
+    }
+
+    operations
+        .into_par_iter()
+        .map(|(name, op)| -> Result<()> {
+            let reader = open_payload()?;
+            let writer = open_output(name)?;
+
+            apply_operation(
+                reader,
+                writer,
+                header.manifest.block_size,
+                header.blob_offset,
+                op,
+                cancel_signal,
+            )?;
+
+            Ok(())
+        })
+        .collect()
+}

avbroot/src/lib.rs

@@ -0,0 +1,22 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+//! Since avbroot is primarily an application and not a library, the semver
+//! versioning covers the CLI only. All Rust APIs can change at any time, even
+//! in patch releases.
+//!
+//! The CLI source files use concrete types wherever possible for simplicity,
+//! while the "library"-style source files aim to be generic.
+
+// We use pb-rs' nostd mode. See build.rs.
+extern crate alloc;
+
+pub mod boot;
+pub mod cli;
+pub mod crypto;
+pub mod format;
+pub mod protobuf;
+pub mod stream;
+pub mod util;

avbroot/src/main.rs

@@ -0,0 +1,26 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::sync::{
+    atomic::{AtomicBool, Ordering},
+    Arc,
+};
+
+use anyhow::Result;
+
+fn main() -> Result<()> {
+    // Set up a cancel signal so we can properly clean up any temporary files.
+    let cancel_signal = Arc::new(AtomicBool::new(false));
+    {
+        let signal = cancel_signal.clone();
+
+        ctrlc::set_handler(move || {
+            signal.store(true, Ordering::SeqCst);
+        })
+        .expect("Failed to set signal handler");
+    }
+
+    avbroot::cli::args::main(&cancel_signal)
+}

avbroot/src/protobuf.rs

@@ -0,0 +1 @@
+include!(concat!(env!("OUT_DIR"), "/protobuf/mod.rs"));

avbroot/src/stream.rs

@@ -0,0 +1,920 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::{
+    fs::File,
+    io::{self, Cursor, Read, Seek, SeekFrom, Write},
+    sync::{
+        atomic::{AtomicBool, Ordering},
+        Arc, Mutex, RwLock,
+    },
+};
+
+use num_traits::ToPrimitive;
+use ring::digest::Context;
+
+use crate::util::{self, EscapedString};
+
+/// A trait for seekable readers. This is only needed because `dyn Read + Seek`
+/// is not a valid construct in Rust yet.
+pub trait ReadSeek: Read + Seek {}
+
+impl<R: Read + Seek> ReadSeek for R {}
+
+/// A trait for seekable writers. This is only needed because `dyn Write + Seek`
+/// is not a valid construct in Rust yet.
+pub trait WriteSeek: Write + Seek {}
+
+impl<W: Write + Seek> WriteSeek for W {}
+
+/// Common function for reading a structure from a reader.
+pub trait FromReader<R: Read>: Sized {
+    type Error;
+
+    fn from_reader(reader: R) -> Result<Self, Self::Error>;
+}
+
+/// Common function for writing a structure to a writer.
+pub trait ToWriter<W: Write>: Sized {
+    type Error;
+
+    fn to_writer(&self, writer: W) -> Result<(), Self::Error>;
+}
+
+/// Extensions for readers to read and discard data (eg. for padding).
+pub trait ReadDiscardExt {
+    fn read_discard(&mut self, size: u64) -> io::Result<u64>;
+
+    fn read_discard_exact(&mut self, size: u64) -> io::Result<()> {
+        let n = self.read_discard(size)?;
+        if n != size {
+            return Err(io::Error::new(
+                io::ErrorKind::UnexpectedEof,
+                format!("Expected to read {size} bytes, but reached EOF after {n} bytes"),
+            ));
+        }
+        Ok(())
+    }
+}
+
+impl<R: Read> ReadDiscardExt for R {
+    fn read_discard(&mut self, size: u64) -> io::Result<u64> {
+        io::copy(&mut self.take(size), &mut io::sink())
+    }
+}
+
+/// Extensions for writers to easily write zeros (eg. for padding).
+pub trait WriteZerosExt {
+    fn write_zeros(&mut self, size: u64) -> io::Result<u64>;
+
+    fn write_zeros_exact(&mut self, size: u64) -> io::Result<()> {
+        let n = self.write_zeros(size)?;
+        if n != size {
+            return Err(io::Error::new(
+                io::ErrorKind::UnexpectedEof,
+                format!("Expected to write {size} bytes, but reached EOF after {n} bytes"),
+            ));
+        }
+        Ok(())
+    }
+}
+
+impl<W: Write> WriteZerosExt for W {
+    fn write_zeros(&mut self, size: u64) -> io::Result<u64> {
+        // We don't use std::io::copy() on std::io::repeat(0) because it fails
+        // if the writer hits EOF before all data is written.
+        let mut written = 0;
+
+        while written < size {
+            let to_write = (size - written).min(util::ZEROS.len() as u64) as usize;
+            let n = self.write(&util::ZEROS[..to_write])?;
+            written += n as u64;
+
+            if n < to_write {
+                break;
+            }
+        }
+
+        Ok(written)
+    }
+}
+
+/// Extensions for readers to read strings.
+pub trait ReadStringExt {
+    /// Read exact sized string.
+    fn read_string_exact(&mut self, size: usize) -> io::Result<String>;
+
+    /// Read string with maximum size and trim trailing zeros.
+    fn read_string_padded(&mut self, max_size: usize) -> io::Result<String>;
+}
+
+impl<R: Read> ReadStringExt for R {
+    fn read_string_exact(&mut self, size: usize) -> io::Result<String> {
+        let mut buf = vec![0u8; size];
+        self.read_exact(&mut buf)?;
+
+        String::from_utf8(buf).map_err(|e| {
+            io::Error::new(
+                io::ErrorKind::InvalidData,
+                format!("Invalid UTF-8: {}: {e}", EscapedString::new(e.as_bytes())),
+            )
+        })
+    }
+
+    fn read_string_padded(&mut self, max_size: usize) -> io::Result<String> {
+        let mut buf = vec![0u8; max_size];
+        self.read_exact(&mut buf)?;
+
+        let after_last_non_zero = buf
+            .iter()
+            .rev()
+            .position(|&b| b != 0)
+            .map_or(0, |i| buf.len() - i);
+        buf.resize(after_last_non_zero, 0);
+        buf.shrink_to_fit();
+
+        String::from_utf8(buf).map_err(|e| {
+            io::Error::new(
+                io::ErrorKind::InvalidData,
+                format!("Invalid UTF-8: {}: {e}", EscapedString::new(e.as_bytes())),
+            )
+        })
+    }
+}
+
+/// Extensions for writers to write strings.
+pub trait WriteStringExt {
+    fn write_string_padded(&mut self, data: &str, max_size: usize) -> io::Result<()>;
+}
+
+impl<W: Write> WriteStringExt for W {
+    fn write_string_padded(&mut self, data: &str, max_size: usize) -> io::Result<()> {
+        if data.len() > max_size {
+            return Err(io::Error::new(
+                io::ErrorKind::InvalidInput,
+                format!("{data:?} exceeds maximum size of {max_size} bytes"),
+            ));
+        }
+
+        self.write_all(data.as_bytes())?;
+
+        let num_zeros = (max_size - data.len()) as u64;
+        self.write_zeros_exact(num_zeros)?;
+
+        Ok(())
+    }
+}
+
+/// A reader wrapper that implements [`Seek`], but only for reporting the
+/// current file position.
+pub struct CountingReader<R: Read> {
+    inner: R,
+    offset: u64,
+}
+
+impl<R: Read> CountingReader<R> {
+    pub fn new(inner: R) -> Self {
+        Self { inner, offset: 0 }
+    }
+
+    pub fn finish(self) -> (R, u64) {
+        (self.inner, self.offset)
+    }
+}
+
+impl<R: Read> Read for CountingReader<R> {
+    fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
+        let n = self.inner.read(buf)?;
+        self.offset += n as u64;
+        Ok(n)
+    }
+}
+
+impl<R: Read> Seek for CountingReader<R> {
+    fn seek(&mut self, pos: SeekFrom) -> io::Result<u64> {
+        if pos == SeekFrom::Current(0) {
+            Ok(self.offset)
+        } else {
+            Err(io::Error::new(
+                io::ErrorKind::InvalidInput,
+                "Can only report current offset",
+            ))
+        }
+    }
+}
+
+/// A writer wrapper that implements [`Seek`], but only for reporting the
+/// current file position.
+pub struct CountingWriter<W: Write> {
+    inner: W,
+    offset: u64,
+}
+
+impl<W: Write> CountingWriter<W> {
+    pub fn new(inner: W) -> Self {
+        Self { inner, offset: 0 }
+    }
+
+    pub fn finish(self) -> (W, u64) {
+        (self.inner, self.offset)
+    }
+}
+
+impl<W: Write> Write for CountingWriter<W> {
+    fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
+        let n = self.inner.write(buf)?;
+        self.offset += n as u64;
+        Ok(n)
+    }
+
+    fn flush(&mut self) -> io::Result<()> {
+        self.inner.flush()
+    }
+}
+
+impl<W: Write> Seek for CountingWriter<W> {
+    fn seek(&mut self, pos: SeekFrom) -> io::Result<u64> {
+        if pos == SeekFrom::Current(0) {
+            Ok(self.offset)
+        } else {
+            Err(io::Error::new(
+                io::ErrorKind::InvalidInput,
+                "Can only report current offset",
+            ))
+        }
+    }
+}
+
+/// A reader wrapper that hashes data as it's being read.
+pub struct HashingReader<R: Read> {
+    inner: R,
+    context: Context,
+}
+
+impl<R: Read> HashingReader<R> {
+    pub fn new(inner: R, context: Context) -> Self {
+        Self { inner, context }
+    }
+
+    pub fn finish(self) -> (R, Context) {
+        (self.inner, self.context)
+    }
+}
+
+impl<R: Read> Read for HashingReader<R> {
+    fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
+        let n = self.inner.read(buf)?;
+        self.context.update(&buf[..n]);
+        Ok(n)
+    }
+}
+
+/// A writer wrapper that hashes data as it's being written.
+pub struct HashingWriter<W: Write> {
+    inner: W,
+    context: Context,
+}
+
+impl<W: Write> HashingWriter<W> {
+    pub fn new(inner: W, context: Context) -> Self {
+        Self { inner, context }
+    }
+
+    pub fn finish(self) -> (W, Context) {
+        (self.inner, self.context)
+    }
+}
+
+impl<W: Write> Write for HashingWriter<W> {
+    fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
+        let n = self.inner.write(buf)?;
+        self.context.update(&buf[..n]);
+        Ok(n)
+    }
+
+    fn flush(&mut self) -> io::Result<()> {
+        self.inner.flush()
+    }
+}
+
+/// A reader wrapper that only allows reading a specific section of a file.
+pub struct SectionReader<R: Read + Seek> {
+    inner: R,
+    start: u64,
+    size: u64,
+    pos: u64,
+}
+
+impl<R: Read + Seek> SectionReader<R> {
+    pub fn new(mut inner: R, start: u64, size: u64) -> io::Result<Self> {
+        inner.seek(SeekFrom::Start(start))?;
+
+        Ok(Self {
+            inner,
+            start,
+            size,
+            pos: 0,
+        })
+    }
+
+    pub fn into_inner(self) -> R {
+        self.inner
+    }
+}
+
+impl<R: Read + Seek> Read for SectionReader<R> {
+    fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
+        let to_read = self.size.saturating_sub(self.pos).min(buf.len() as u64) as usize;
+        let n = self.inner.read(&mut buf[..to_read])?;
+        self.pos += n as u64;
+        Ok(n)
+    }
+}
+
+impl<R: Read + Seek> Seek for SectionReader<R> {
+    fn seek(&mut self, pos: SeekFrom) -> io::Result<u64> {
+        self.pos = match pos {
+            SeekFrom::Start(o) => o,
+            SeekFrom::End(o) => self
+                .size
+                .to_i64()
+                .and_then(|s| s.checked_add(o))
+                .and_then(|s| s.to_u64())
+                .ok_or_else(|| {
+                    io::Error::new(
+                        io::ErrorKind::InvalidInput,
+                        "Offset would be before the start of the file",
+                    )
+                })?,
+            SeekFrom::Current(o) => self
+                .pos
+                .to_i64()
+                .and_then(|s| s.checked_add(o))
+                .and_then(|s| s.to_u64())
+                .ok_or_else(|| {
+                    io::Error::new(
+                        io::ErrorKind::InvalidInput,
+                        "Offset would be before the start of the file",
+                    )
+                })?,
+        };
+
+        let raw_pos = self.inner.seek(SeekFrom::Start(self.start + self.pos))?;
+        Ok(raw_pos - self.start)
+    }
+}
+
+/// A writer wrapper that seeks instead of writing when a write buffer consists
+/// solely of zeros.
+#[derive(Debug)]
+pub struct HolePunchingWriter<W: Write + Seek> {
+    inner: W,
+}
+
+impl<W: Write + Seek> HolePunchingWriter<W> {
+    pub fn new(inner: W) -> Self {
+        Self { inner }
+    }
+
+    pub fn into_inner(self) -> W {
+        self.inner
+    }
+}
+
+impl<W: Write + Seek> Write for HolePunchingWriter<W> {
+    fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
+        if util::is_zero(buf) {
+            self.inner.seek(SeekFrom::Current(buf.len() as i64))?;
+            Ok(buf.len())
+        } else {
+            self.inner.write(buf)
+        }
+    }
+
+    fn flush(&mut self) -> io::Result<()> {
+        self.inner.flush()
+    }
+}
+
+/// A file wrapper that uses a userspace file offset. A cloned instances uses
+/// the same underlying kernel file descriptor, but a new userspace file offset.
+#[derive(Clone)]
+pub struct PSeekFile {
+    // The lock is needed because flush() takes a `&mut self`.
+    file: Arc<RwLock<File>>,
+    offset: u64,
+}
+
+impl PSeekFile {
+    pub fn new(file: File) -> Self {
+        Self {
+            file: Arc::new(RwLock::new(file)),
+            offset: 0,
+        }
+    }
+
+    pub fn set_len(&self, size: u64) -> io::Result<()> {
+        let file_locked = self.file.read().unwrap();
+        file_locked.set_len(size)
+    }
+
+    /// Read data from offset. The kernel's file position *will* be changed.
+    #[cfg(windows)]
+    fn read_at(&self, buf: &mut [u8]) -> io::Result<usize> {
+        use std::os::windows::fs::FileExt;
+        self.file.read().unwrap().seek_read(buf, self.offset)
+    }
+
+    /// Read data from offset. The kernel's file position will *not* be changed.
+    #[cfg(unix)]
+    fn read_at(&self, buf: &mut [u8]) -> io::Result<usize> {
+        use std::os::unix::fs::FileExt;
+        self.file.read().unwrap().read_at(buf, self.offset)
+    }
+
+    /// Write data to offset. The kernel's file position *will* be changed.
+    #[cfg(windows)]
+    fn write_at(&self, buf: &[u8]) -> io::Result<usize> {
+        use std::os::windows::fs::FileExt;
+        self.file.read().unwrap().seek_write(buf, self.offset)
+    }
+
+    /// Write data to offset. The kernel's file position will *not* be changed.
+    #[cfg(unix)]
+    fn write_at(&self, buf: &[u8]) -> io::Result<usize> {
+        use std::os::unix::fs::FileExt;
+        self.file.read().unwrap().write_at(buf, self.offset)
+    }
+}
+
+impl Read for PSeekFile {
+    fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
+        let n = self.read_at(buf)?;
+        self.offset += n as u64;
+        Ok(n)
+    }
+}
+
+impl Write for PSeekFile {
+    fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
+        let n = self.write_at(buf)?;
+        self.offset += n as u64;
+        Ok(n)
+    }
+
+    fn flush(&mut self) -> io::Result<()> {
+        self.file.write().unwrap().flush()
+    }
+}
+
+impl Seek for PSeekFile {
+    fn seek(&mut self, pos: SeekFrom) -> io::Result<u64> {
+        self.offset = match pos {
+            SeekFrom::Start(o) => o,
+            SeekFrom::End(o) => {
+                let file_size = self.file.read().unwrap().metadata()?.len();
+                file_size
+                    .to_i64()
+                    .and_then(|s| s.checked_add(o))
+                    .and_then(|s| s.to_u64())
+                    .ok_or_else(|| {
+                        io::Error::new(
+                            io::ErrorKind::InvalidInput,
+                            "Offset would be before the start of the file",
+                        )
+                    })?
+            }
+            SeekFrom::Current(o) => self
+                .offset
+                .to_i64()
+                .and_then(|s| s.checked_add(o))
+                .and_then(|s| s.to_u64())
+                .ok_or_else(|| {
+                    io::Error::new(
+                        io::ErrorKind::InvalidInput,
+                        "Offset would be before the start of the file",
+                    )
+                })?,
+        };
+
+        Ok(self.offset)
+    }
+}
+
+/// A small wrapper around a [`Cursor`] that allows multiple instances to share
+/// the same underlying file. All reads, writes, and seeks are single-threaded.
+/// This is useful for scenarios where data needs to be copied from multiple
+/// readers into different parts of the same [`SharedCursor`] writer and the
+/// read operation is significantly more expensive than the write operation (eg.
+/// due to decompression).
+#[derive(Clone, Default)]
+pub struct SharedCursor {
+    inner: Arc<Mutex<Cursor<Vec<u8>>>>,
+    offset: u64,
+}
+
+impl SharedCursor {
+    pub fn clone_rewind(&self) -> Self {
+        let mut new = self.clone();
+        new.offset = 0;
+        new
+    }
+}
+
+impl Read for SharedCursor {
+    fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
+        let mut inner = self.inner.lock().unwrap();
+        inner.seek(SeekFrom::Start(self.offset))?;
+
+        let n = inner.read(buf)?;
+        self.offset += n as u64;
+
+        Ok(n)
+    }
+}
+
+impl Write for SharedCursor {
+    fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
+        let mut inner = self.inner.lock().unwrap();
+        inner.seek(SeekFrom::Start(self.offset))?;
+
+        let n = inner.write(buf)?;
+        self.offset += n as u64;
+
+        Ok(n)
+    }
+
+    fn flush(&mut self) -> io::Result<()> {
+        let mut inner = self.inner.lock().unwrap();
+        inner.flush()
+    }
+}
+
+impl Seek for SharedCursor {
+    fn seek(&mut self, pos: SeekFrom) -> io::Result<u64> {
+        let mut inner = self.inner.lock().unwrap();
+        self.offset = inner.seek(pos)?;
+        Ok(self.offset)
+    }
+}
+
+/// Copy exactly `size` bytes from `reader` to `writer`, invoking `inspect`
+/// after every buffer read iteration. If either `reader` or `writer` reaches
+/// EOF before `size` bytes are copied, an error is returned. The operation is
+/// cancelled on the next loop iteration if `cancel_signal` is set to `true`.
+pub fn copy_n_inspect(
+    mut reader: impl Read,
+    mut writer: impl Write,
+    mut size: u64,
+    mut inspect: impl FnMut(&[u8]),
+    cancel_signal: &Arc<AtomicBool>,
+) -> io::Result<()> {
+    let mut buf = [0u8; 16384];
+
+    while size > 0 {
+        if cancel_signal.load(Ordering::SeqCst) {
+            return Err(io::Error::new(
+                io::ErrorKind::Interrupted,
+                "Received cancel signal",
+            ));
+        }
+
+        let to_read = size.min(buf.len() as u64) as usize;
+        reader.read_exact(&mut buf[..to_read])?;
+
+        inspect(&buf[..to_read]);
+
+        writer.write_all(&buf[..to_read])?;
+
+        size -= to_read as u64;
+    }
+
+    Ok(())
+}
+
+/// Copy exactly `size` bytes from `reader` to `writer`.
+pub fn copy_n(
+    reader: impl Read,
+    writer: impl Write,
+    size: u64,
+    cancel_signal: &Arc<AtomicBool>,
+) -> io::Result<()> {
+    copy_n_inspect(reader, writer, size, |_| {}, cancel_signal)
+}
+
+/// Copy data from `reader` to `writer` until `reader` reaches EOF. If `writer`
+/// reaches EOF before `reader` does, an error is returned. The operation is
+/// cancelled on the next loop iteration if `cancel_signal` is set to `true`.
+pub fn copy(
+    mut reader: impl Read,
+    mut writer: impl Write,
+    cancel_signal: &Arc<AtomicBool>,
+) -> io::Result<u64> {
+    let mut buf = [0u8; 16384];
+    let mut copied = 0;
+
+    loop {
+        if cancel_signal.load(Ordering::SeqCst) {
+            return Err(io::Error::new(
+                io::ErrorKind::Interrupted,
+                "Received cancel signal",
+            ));
+        }
+
+        let n = reader.read(&mut buf)?;
+        if n == 0 {
+            break;
+        }
+
+        writer.write_all(&buf[..n])?;
+
+        copied += n as u64;
+    }
+
+    Ok(copied)
+}
+
+#[cfg(test)]
+mod tests {
+    use std::{
+        io::{self, Cursor, Read, Seek, SeekFrom, Write},
+        sync::{
+            atomic::{AtomicBool, Ordering},
+            Arc,
+        },
+    };
+
+    use ring::digest::Context;
+
+    use super::{
+        CountingReader, CountingWriter, HashingReader, HashingWriter, HolePunchingWriter,
+        PSeekFile, ReadDiscardExt, ReadStringExt, SectionReader, SharedCursor, WriteStringExt,
+        WriteZerosExt,
+    };
+
+    const FOOBAR_SHA256: [u8; 32] = [
+        0xc3, 0xab, 0x8f, 0xf1, 0x37, 0x20, 0xe8, 0xad, 0x90, 0x47, 0xdd, 0x39, 0x46, 0x6b, 0x3c,
+        0x89, 0x74, 0xe5, 0x92, 0xc2, 0xfa, 0x38, 0x3d, 0x4a, 0x39, 0x60, 0x71, 0x4c, 0xae, 0xf0,
+        0xc4, 0xf2,
+    ];
+
+    #[test]
+    fn read_discard() {
+        let mut reader = Cursor::new(b"foobar");
+        reader.read_discard_exact(3).unwrap();
+
+        let mut buf = [0u8; 2];
+        reader.read_exact(&mut buf).unwrap();
+        assert_eq!(&buf, b"ba");
+
+        let n = reader.read_discard(2).unwrap();
+        assert_eq!(n, 1);
+
+        assert_eq!(reader.stream_position().unwrap(), 6);
+    }
+
+    #[test]
+    fn write_zeros() {
+        let mut writer = Cursor::new([0u8; 6]);
+
+        writer.write_zeros_exact(2).unwrap();
+        writer.write_all(b"foo").unwrap();
+
+        let n = writer.write_zeros(2).unwrap();
+        assert_eq!(n, 1);
+
+        assert_eq!(&writer.into_inner(), b"\0\0foo\0");
+    }
+
+    #[test]
+    fn read_string() {
+        let mut reader = Cursor::new(b"foo\0\0bar\0\0");
+
+        assert_eq!(reader.read_string_exact(3).unwrap(), "foo");
+        assert_eq!(reader.read_string_exact(0).unwrap(), "");
+
+        reader.rewind().unwrap();
+        assert_eq!(reader.read_string_padded(3).unwrap(), "foo");
+
+        reader.rewind().unwrap();
+        assert_eq!(reader.read_string_padded(10).unwrap(), "foo\0\0bar");
+    }
+
+    #[test]
+    fn write_string() {
+        let mut writer = Cursor::new([0xffu8; 8]);
+
+        writer.write_string_padded("foobar", 8).unwrap();
+        assert_eq!(writer.get_ref(), b"foobar\0\0");
+
+        writer.rewind().unwrap();
+        writer.write_string_padded("foobarhi", 8).unwrap();
+        assert_eq!(writer.get_ref(), b"foobarhi");
+    }
+
+    #[test]
+    fn counting_reader() {
+        let raw_reader = Cursor::new(b"foobar");
+        let mut reader = CountingReader::new(raw_reader);
+
+        let mut buf = [0u8; 6];
+        reader.read_exact(&mut buf[..0]).unwrap();
+        reader.read_exact(&mut buf[..3]).unwrap();
+        reader.read_exact(&mut buf[3..4]).unwrap();
+        reader.read_exact(&mut buf[4..6]).unwrap();
+        assert_eq!(&buf, b"foobar");
+
+        let (mut raw_reader, size) = reader.finish();
+        assert_eq!(raw_reader.stream_position().unwrap(), 6);
+        assert_eq!(size, 6);
+    }
+
+    #[test]
+    fn counting_writer() {
+        let raw_writer = Cursor::new([0u8; 6]);
+        let mut writer = CountingWriter::new(raw_writer);
+
+        writer.write_all(b"foo").unwrap();
+        writer.write_all(b"").unwrap();
+        writer.write_all(b"bar").unwrap();
+
+        let (mut raw_writer, size) = writer.finish();
+        assert_eq!(raw_writer.stream_position().unwrap(), 6);
+        assert_eq!(&raw_writer.into_inner(), b"foobar");
+        assert_eq!(size, 6);
+    }
+
+    #[test]
+    fn hashing_reader() {
+        let raw_reader = Cursor::new(b"foobar");
+        let mut reader = HashingReader::new(raw_reader, Context::new(&ring::digest::SHA256));
+
+        let mut buf = [0u8; 6];
+        reader.read_exact(&mut buf[..0]).unwrap();
+        reader.read_exact(&mut buf[..3]).unwrap();
+        reader.read_exact(&mut buf[3..4]).unwrap();
+        reader.read_exact(&mut buf[4..6]).unwrap();
+        assert_eq!(&buf, b"foobar");
+
+        let (mut raw_reader, context) = reader.finish();
+        assert_eq!(raw_reader.stream_position().unwrap(), 6);
+        assert_eq!(context.finish().as_ref(), FOOBAR_SHA256);
+    }
+
+    #[test]
+    fn hashing_writer() {
+        let raw_writer = Cursor::new([0u8; 6]);
+        let mut writer = HashingWriter::new(raw_writer, Context::new(&ring::digest::SHA256));
+
+        writer.write_all(b"").unwrap();
+        writer.write_all(b"foo").unwrap();
+        writer.write_all(b"bar").unwrap();
+
+        let (mut raw_writer, context) = writer.finish();
+        assert_eq!(raw_writer.stream_position().unwrap(), 6);
+        assert_eq!(&raw_writer.into_inner(), b"foobar");
+        assert_eq!(context.finish().as_ref(), FOOBAR_SHA256);
+    }
+
+    #[test]
+    fn section_reader() {
+        let raw_reader = Cursor::new(b"fooinnerbar");
+        let mut reader = SectionReader::new(raw_reader, 3, 5).unwrap();
+
+        let mut buf = [0u8; 5];
+        reader.read_exact(&mut buf[..0]).unwrap();
+        reader.read_exact(&mut buf[..3]).unwrap();
+        reader.read_exact(&mut buf[3..5]).unwrap();
+        assert_eq!(&buf, b"inner");
+
+        let n = reader.read_discard(1).unwrap();
+        assert_eq!(n, 0);
+
+        buf = *b"\0\0\0\0\0";
+        reader.seek(SeekFrom::Start(4)).unwrap();
+        reader.read_exact(&mut buf[..1]).unwrap();
+        assert_eq!(&buf[..1], b"r");
+
+        buf = *b"\0\0\0\0\0";
+        reader.seek(SeekFrom::End(-4)).unwrap();
+        reader.read_exact(&mut buf[..4]).unwrap();
+        assert_eq!(&buf[..4], b"nner");
+
+        buf = *b"\0\0\0\0\0";
+        reader.seek(SeekFrom::Current(-5)).unwrap();
+        reader.read_exact(&mut buf[..3]).unwrap();
+        assert_eq!(&buf[..3], b"inn");
+
+        let mut raw_reader = reader.into_inner();
+        assert_eq!(raw_reader.stream_position().unwrap(), 6);
+    }
+
+    #[test]
+    fn hole_punching_writer() {
+        let raw_writer = Cursor::new(b"foobar foobar".to_owned());
+        let mut writer = HolePunchingWriter::new(raw_writer);
+
+        writer.write_all(b"hello").unwrap();
+        writer.write_all(b"").unwrap();
+        writer.write_all(b"\0").unwrap();
+        writer.write_all(b"\0\0").unwrap();
+        writer.write_all(b"world").unwrap();
+
+        let raw_writer = writer.into_inner();
+        assert_eq!(&raw_writer.into_inner(), b"hellor fworld");
+    }
+
+    #[test]
+    fn pseek_file() {
+        let raw_file = tempfile::tempfile().unwrap();
+        let mut a = PSeekFile::new(raw_file);
+        let mut b = a.clone();
+        let mut c = b.clone();
+
+        b.write_all(b"foobar").unwrap();
+        c.write_all(b"hello").unwrap();
+        b.write_all(b"world").unwrap();
+        c.seek(SeekFrom::Start(0)).unwrap();
+        c.write_all(b"hi").unwrap();
+
+        let mut buf = [0u8; 11];
+        a.read_exact(&mut buf).unwrap();
+        assert_eq!(&buf, b"hillorworld");
+
+        let n = a.read_discard(1).unwrap();
+        assert_eq!(n, 0);
+    }
+
+    #[test]
+    fn shared_cursor() {
+        let mut a = SharedCursor::default();
+        let mut b = a.clone();
+        let mut c = b.clone();
+
+        b.write_all(b"foobar").unwrap();
+        c.write_all(b"hello").unwrap();
+        b.write_all(b"world").unwrap();
+        c.seek(SeekFrom::Start(0)).unwrap();
+        c.write_all(b"hi").unwrap();
+
+        let mut buf = [0u8; 11];
+        a.read_exact(&mut buf).unwrap();
+        assert_eq!(&buf, b"hillorworld");
+
+        let n = a.read_discard(1).unwrap();
+        assert_eq!(n, 0);
+    }
+
+    #[test]
+    fn copy() {
+        let cancel_signal = Arc::new(AtomicBool::new(false));
+        let mut reader = Cursor::new(b"foobar");
+        let mut writer = Cursor::new([0u8; 6]);
+
+        super::copy_n_inspect(&mut reader, &mut writer, 6, |_| {}, &cancel_signal).unwrap();
+        assert_eq!(writer.get_ref(), b"foobar");
+
+        // Reader early EOF.
+        reader.seek(SeekFrom::Start(3)).unwrap();
+        writer.rewind().unwrap();
+        let err =
+            super::copy_n_inspect(&mut reader, &mut writer, 6, |_| {}, &cancel_signal).unwrap_err();
+        assert_eq!(err.kind(), io::ErrorKind::UnexpectedEof);
+
+        // Writer early EOF.
+        reader.rewind().unwrap();
+        writer.seek(SeekFrom::Start(3)).unwrap();
+        let err =
+            super::copy_n_inspect(&mut reader, &mut writer, 6, |_| {}, &cancel_signal).unwrap_err();
+        assert_eq!(err.kind(), io::ErrorKind::WriteZero);
+
+        reader.rewind().unwrap();
+        writer.rewind().unwrap();
+        let n = super::copy(&mut reader, &mut writer, &cancel_signal).unwrap();
+        assert_eq!(n, 6);
+        assert_eq!(writer.get_ref(), b"foobar");
+
+        // Reader early EOF.
+        reader.seek(SeekFrom::Start(3)).unwrap();
+        writer.rewind().unwrap();
+        let n = super::copy(&mut reader, &mut writer, &cancel_signal).unwrap();
+        assert_eq!(n, 3);
+
+        // Writer early EOF.
+        reader.rewind().unwrap();
+        writer.seek(SeekFrom::Start(3)).unwrap();
+        let err = super::copy(&mut reader, &mut writer, &cancel_signal).unwrap_err();
+        assert_eq!(err.kind(), io::ErrorKind::WriteZero);
+
+        reader.rewind().unwrap();
+        writer.rewind().unwrap();
+        cancel_signal.store(true, Ordering::SeqCst);
+        let err =
+            super::copy_n_inspect(&mut reader, &mut writer, 6, |_| {}, &cancel_signal).unwrap_err();
+        assert_eq!(err.kind(), io::ErrorKind::Interrupted);
+        let err = super::copy(&mut reader, &mut writer, &cancel_signal).unwrap_err();
+        assert_eq!(err.kind(), io::ErrorKind::Interrupted);
+    }
+}

avbroot/src/util.rs

@@ -0,0 +1,118 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::fmt;
+
+use quick_protobuf::{BytesReader, MessageRead, MessageWrite, Writer};
+
+pub const ZEROS: [u8; 16384] = [0u8; 16384];
+
+/// A small wrapper to format a number as a size in bytes.
+#[derive(Clone, Copy)]
+pub struct NumBytes(pub usize);
+
+impl fmt::Debug for NumBytes {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        if self.0 == 1 {
+            write!(f, "<{:?} byte>", self.0)
+        } else {
+            write!(f, "<{:?} bytes>", self.0)
+        }
+    }
+}
+
+/// A wrapper around a byte slice to format it as ASCII with invalid bytes
+/// escaped as `\x##`.
+#[derive(Clone)]
+pub struct EscapedString<T: AsRef<[u8]>> {
+    inner: T,
+    quoted: bool,
+}
+
+impl<T: AsRef<[u8]>> EscapedString<T> {
+    pub fn new(inner: T) -> Self {
+        Self {
+            inner,
+            quoted: true,
+        }
+    }
+
+    pub fn new_unquoted(inner: T) -> Self {
+        Self {
+            inner,
+            quoted: false,
+        }
+    }
+
+    pub fn into_inner(self) -> T {
+        self.inner
+    }
+
+    pub fn get_ref(&self) -> &T {
+        &self.inner
+    }
+
+    pub fn get_mut(&mut self) -> &mut T {
+        &mut self.inner
+    }
+}
+
+impl<T: AsRef<[u8]>> fmt::Debug for EscapedString<T> {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        let escaped: String = self
+            .inner
+            .as_ref()
+            .iter()
+            .flat_map(|b| b.escape_ascii())
+            .map(char::from)
+            .collect();
+
+        if self.quoted {
+            write!(f, "\"")?;
+        }
+
+        write!(f, "{escaped}")?;
+
+        if self.quoted {
+            write!(f, "\"")?;
+        }
+
+        Ok(())
+    }
+}
+
+impl<T: AsRef<[u8]>> fmt::Display for EscapedString<T> {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        fmt::Debug::fmt(&self, f)
+    }
+}
+
+/// Check if a byte slice is all zeros.
+pub fn is_zero(mut buf: &[u8]) -> bool {
+    while !buf.is_empty() {
+        let n = buf.len().min(ZEROS.len());
+        if buf[..n] != ZEROS[..n] {
+            return false;
+        }
+
+        buf = &buf[n..];
+    }
+
+    true
+}
+
+/// Read a protobuf message with no leading size field.
+pub fn read_protobuf<'a, M: MessageRead<'a>>(data: &'a [u8]) -> quick_protobuf::Result<M> {
+    let mut reader = BytesReader::from_bytes(data);
+    M::from_reader(&mut reader, data)
+}
+
+/// Write a protobuf message with no leading size field.
+pub fn write_protobuf<M: MessageWrite>(message: &M) -> quick_protobuf::Result<Vec<u8>> {
+    let mut buf = Vec::with_capacity(message.get_size());
+    let mut writer = Writer::new(&mut buf);
+    message.write_message(&mut writer)?;
+    Ok(buf)
+}

avbroot/tests/avb.rs

@@ -0,0 +1,82 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::io::{self, Cursor, Read, Seek, SeekFrom};
+
+use assert_matches::assert_matches;
+use pkcs8::DecodePrivateKey;
+use rsa::RsaPrivateKey;
+
+use avbroot::{self, format::avb};
+
+fn get_test_key() -> RsaPrivateKey {
+    let data = include_str!(concat!(
+        env!("CARGO_WORKSPACE_DIR"),
+        "/e2e/keys/TEST_KEY_DO_NOT_USE_avb.key",
+    ));
+    let passphrase = include_str!(concat!(
+        env!("CARGO_WORKSPACE_DIR"),
+        "/e2e/keys/TEST_KEY_DO_NOT_USE_avb.passphrase",
+    ));
+
+    RsaPrivateKey::from_pkcs8_encrypted_pem(data, passphrase.trim_end()).unwrap()
+}
+
+#[test]
+fn round_trip_root_image() {
+    let data = include_bytes!(concat!(
+        env!("CARGO_MANIFEST_DIR"),
+        "/tests/data/vbmeta_root.img",
+    ));
+    let reader = Cursor::new(data);
+
+    let (mut header, footer, _) = avb::load_image(reader).unwrap();
+    assert_matches!(footer, None);
+
+    // Clear out the signature-related fields and re-sign.
+    header.hash.clear();
+    header.signature.clear();
+    header.public_key.clear();
+    let key = get_test_key();
+    header.sign(&key).unwrap();
+
+    let mut writer = Cursor::new(Vec::new());
+    avb::write_root_image(&mut writer, &header, 64).unwrap();
+    let new_data = writer.into_inner();
+
+    assert_eq!(data, new_data.as_slice());
+}
+
+#[test]
+fn round_trip_appended_image() {
+    let data = include_bytes!(concat!(
+        env!("CARGO_MANIFEST_DIR"),
+        "/tests/data/vbmeta_appended.img",
+    ));
+    let mut reader = Cursor::new(data);
+
+    let (mut header, footer, _) = avb::load_image(&mut reader).unwrap();
+    let footer = footer.unwrap();
+
+    // Clear out the signature-related fields and re-sign.
+    header.hash.clear();
+    header.signature.clear();
+    header.public_key.clear();
+    let key = get_test_key();
+    header.sign(&key).unwrap();
+
+    let mut writer = Cursor::new(Vec::new());
+
+    // Copy the partition data.
+    let image_size = reader.seek(SeekFrom::End(0)).unwrap();
+    reader.seek(SeekFrom::Start(0)).unwrap();
+    io::copy(&mut reader.take(footer.original_image_size), &mut writer).unwrap();
+
+    // Write new vbmeta structures.
+    avb::write_appended_image(&mut writer, &header, &footer, image_size).unwrap();
+    let new_data = writer.into_inner();
+
+    assert_eq!(data, new_data.as_slice());
+}

avbroot/tests/bootimage.rs

@@ -0,0 +1,125 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::io::Cursor;
+
+use avbroot::{
+    self,
+    format::bootimage::{BootImage, BootImageExt},
+    stream::{FromReader, ToWriter},
+};
+use pkcs8::DecodePrivateKey;
+use rsa::RsaPrivateKey;
+
+fn get_test_key() -> RsaPrivateKey {
+    let data = include_str!(concat!(
+        env!("CARGO_WORKSPACE_DIR"),
+        "/e2e/keys/TEST_KEY_DO_NOT_USE_avb.key",
+    ));
+    let passphrase = include_str!(concat!(
+        env!("CARGO_WORKSPACE_DIR"),
+        "/e2e/keys/TEST_KEY_DO_NOT_USE_avb.passphrase",
+    ));
+
+    RsaPrivateKey::from_pkcs8_encrypted_pem(data, passphrase.trim_end()).unwrap()
+}
+
+fn round_trip(data: &[u8], expected_version: u32) {
+    let reader = Cursor::new(data);
+    let mut image = BootImage::from_reader(reader).unwrap();
+
+    assert_eq!(image.header_version(), expected_version);
+
+    match &mut image {
+        BootImage::V3Through4(b) => {
+            let should_sign = b
+                .v4_extra
+                .as_ref()
+                .map_or(false, |v4| v4.signature.is_some());
+            let key = get_test_key();
+            let signed = b.sign(&key).unwrap();
+            assert_eq!(signed, should_sign);
+        }
+        _ => {}
+    }
+
+    let mut writer = Cursor::new(Vec::new());
+    image.to_writer(&mut writer).unwrap();
+    let new_data = writer.into_inner();
+
+    assert_eq!(data, new_data);
+}
+
+#[test]
+fn round_trip_v0() {
+    let data = include_bytes!(concat!(
+        env!("CARGO_MANIFEST_DIR"),
+        "/tests/data/boot_v0.img",
+    ));
+    round_trip(data, 0);
+}
+
+#[test]
+fn round_trip_v1() {
+    let data = include_bytes!(concat!(
+        env!("CARGO_MANIFEST_DIR"),
+        "/tests/data/boot_v1.img",
+    ));
+    round_trip(data, 1);
+}
+
+#[test]
+fn round_trip_v2() {
+    let data = include_bytes!(concat!(
+        env!("CARGO_MANIFEST_DIR"),
+        "/tests/data/boot_v2.img",
+    ));
+    round_trip(data, 2);
+}
+
+#[test]
+fn round_trip_v3() {
+    let data = include_bytes!(concat!(
+        env!("CARGO_MANIFEST_DIR"),
+        "/tests/data/boot_v3.img",
+    ));
+    round_trip(data, 3);
+}
+
+#[test]
+fn round_trip_v4() {
+    let data = include_bytes!(concat!(
+        env!("CARGO_MANIFEST_DIR"),
+        "/tests/data/boot_v4.img",
+    ));
+    round_trip(data, 4);
+}
+
+#[test]
+fn round_trip_v4_vts() {
+    let data = include_bytes!(concat!(
+        env!("CARGO_MANIFEST_DIR"),
+        "/tests/data/boot_v4_vts.img",
+    ));
+    round_trip(data, 4);
+}
+
+#[test]
+fn round_trip_vendor_v3() {
+    let data = include_bytes!(concat!(
+        env!("CARGO_MANIFEST_DIR"),
+        "/tests/data/vendor_v3.img",
+    ));
+    round_trip(data, 3);
+}
+
+#[test]
+fn round_trip_vendor_v4() {
+    let data = include_bytes!(concat!(
+        env!("CARGO_MANIFEST_DIR"),
+        "/tests/data/vendor_v4.img",
+    ));
+    round_trip(data, 4);
+}

avbroot/tests/compression.rs

@@ -0,0 +1,39 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::io::{Cursor, Read, Seek, Write};
+
+use avbroot::{
+    self,
+    format::compression::{CompressedFormat, CompressedReader, CompressedWriter},
+};
+
+fn round_trip(data: &[u8], format: CompressedFormat) {
+    let raw_writer = Cursor::new(Vec::new());
+    let mut writer = CompressedWriter::new(raw_writer, format).unwrap();
+    writer.write_all(data).unwrap();
+    let mut raw_reader = writer.finish().unwrap();
+
+    raw_reader.rewind().unwrap();
+    let mut reader = CompressedReader::new(raw_reader, false).unwrap();
+    assert_eq!(reader.format(), format);
+
+    let mut new_data = vec![];
+    reader.read_to_end(&mut new_data).unwrap();
+
+    assert_eq!(data, new_data);
+}
+
+#[test]
+fn round_trip_gzip() {
+    round_trip(b"gzip-compressed data", CompressedFormat::Gzip);
+}
+
+#[test]
+fn round_trip_lz4_legacy() {
+    // Make sure we exceed the 8MiB block boundary.
+    let data = b"Lz4Legacy".repeat(1024 * 1024);
+    round_trip(&data, CompressedFormat::Lz4Legacy);
+}

avbroot/util.py

@@ -1,221 +0,0 @@
-import contextlib
-import dataclasses
-import functools
-import os
-import tempfile
-
-
-_ZERO_BLOCK = memoryview(b'\0' * 16384)
-
-umask = None
-
-
-def load_umask_unsafe():
-    # POSIX provides no way to query the umask without changing it. Parsing
-    # /proc/self/status can work, but it's Linux only. Instead, we'll just do it
-    # once when the program is initially started.
-    global umask
-
-    if os.name != 'nt' and umask is None:
-        current_umask = os.umask(0o777)
-        os.umask(current_umask)
-
-        umask = current_umask
-
-
-@dataclasses.dataclass
-@functools.total_ordering
-class Range:
-    '''
-    Simple class to represent a half-open interval.
-    '''
-
-    start: int
-    end: int
-
-    def __repr__(self) -> str:
-        return f'[{self.start}, {self.end})'
-
-    def __str__(self) -> str:
-        return f'>={self.start}, <{self.end}'
-
-    def __lt__(self, other) -> bool:
-        return (self.start, self.end) < (other.start, other.end)
-
-    def __eq__(self, other) -> bool:
-        return (self.start, self.end) == (other.start, other.end)
-
-    def __contains__(self, item) -> bool:
-        return item >= self.start and item < self.end
-
-    def __bool__(self) -> bool:
-        return self.start < self.end
-
-    def size(self) -> int:
-        return self.end - self.start
-
-
-@contextlib.contextmanager
-def open_output_file(path):
-    '''
-    Create a temporary file in the same directory as the specified path and
-    replace it if the function succeeds. On non-Windows, the file replacement
-    is atomic. On Windows, it is not.
-    '''
-
-    directory = os.path.dirname(path)
-
-    with tempfile.NamedTemporaryFile(dir=directory, delete=False) as f:
-        try:
-            yield f
-
-            if os.name == 'nt':
-                # Windows does not allow renaming a file with handles open
-                f.close()
-
-                # Windows only supports atomic renames by calling
-                # SetFileInformationByHandle() with the FileRenameInfoEx
-                # operation and the FILE_RENAME_FLAG_REPLACE_IF_EXISTS and
-                # FILE_RENAME_FLAG_POSIX_SEMANTICS flags. This is not exposed
-                # in Python and it's not worth adding a new dependency for
-                # doing low-level win32 API calls.
-                try:
-                    os.unlink(path)
-                except FileNotFoundError:
-                    pass
-            else:
-                # NamedTemporaryFile always uses 600 permissions with no way to
-                # override it. We'll do our own umask-respecting chmod.
-                os.fchmod(f.fileno(), 0o666 & ~umask)
-
-            os.rename(f.name, path)
-        except BaseException:
-            if os.name == 'nt':
-                # Windows does not allow deleting a file with handles open
-                f.close()
-
-            os.unlink(f.name)
-            raise
-
-
-def hash_file(f, hasher, buf_size=16384):
-    '''
-    Update <hasher> when the data from <f> until EOF.
-    '''
-
-    buf = bytearray(buf_size)
-    buf_view = memoryview(buf)
-
-    while True:
-        n = f.readinto(buf_view)
-        if not n:
-            break
-
-        hasher.update(buf_view[:n])
-
-    return hasher
-
-
-def copyfileobj_n(f_in, f_out, size, buf_size=16384, hasher=None):
-    '''
-    Copy <size> bytes from <f_in> to <f_out>.
-
-    Raises IOError if EOF is reached in <f_in> before <size> bytes are read.
-    '''
-
-    buf = bytearray(buf_size)
-    buf_view = memoryview(buf)
-
-    while size:
-        to_read = min(len(buf_view), size)
-        n = f_in.readinto(buf_view[:to_read])
-        if not n:
-            break
-
-        if hasher:
-            hasher.update(buf_view[:n])
-
-        f_out.write(buf_view[:n])
-        size -= n
-
-    if size:
-        raise IOError(f'Unexpected EOF; expected {size} more bytes')
-
-
-def decompress_n(decompressor, f_in, f_out, size, buf_size=16384, hasher=None):
-    '''
-    Read <size> bytes from <f_in> and decompress them to <f_out>.
-
-    Raises IOError if EOF is reached in <f_in> before <size> bytes are read.
-    '''
-
-    buf = bytearray(buf_size)
-    buf_view = memoryview(buf)
-
-    while size:
-        to_read = min(len(buf_view), size)
-        n = f_in.readinto(buf_view[:to_read])
-        if not n:
-            break
-
-        if hasher:
-            hasher.update(buf_view[:n])
-
-        data = decompressor.decompress(buf_view[:n])
-
-        f_out.write(data)
-        size -= n
-
-    if size:
-        raise IOError(f'Unexpected EOF; expected {size} more bytes')
-    elif not decompressor.eof:
-        raise IOError('Did not reach end of compressed input')
-
-
-def zero_n(f_out, size, buf_size=16384):
-    '''
-    Write <size> zeroes to <f_out>.
-    '''
-
-    buf = bytearray(buf_size)
-    buf_view = memoryview(buf)
-
-    while size:
-        to_write = min(len(buf_view), size)
-        f_out.write(buf_view[:to_write])
-        size -= to_write
-
-
-def read_exact(f, size: int) -> bytes:
-    '''
-    Read exactly <size> bytes from <f> or raise an EOFError.
-    '''
-
-    data = f.read(size)
-    if len(data) != size:
-        raise EOFError(f'Unexpected EOF: expected {size} bytes, '
-                       f'but only read {len(data)} bytes')
-
-    if not isinstance(data, bytes):
-        # io.BytesIO returns a bytearray
-        return bytes(data)
-    else:
-        return data
-
-
-def is_zero(data):
-    '''
-    Check if all bytes in the bytes-like object are null bytes.
-    '''
-
-    view = memoryview(data)
-
-    while view:
-        n = min(len(view), len(_ZERO_BLOCK))
-
-        if view[:n] != _ZERO_BLOCK[:n]:
-            return False
-
-        view = view[n:]
-
-    return True

avbroot/vbmeta.py

@@ -1,195 +0,0 @@
-import contextlib
-import os
-import typing
-import unittest.mock
-
-import avbtool
-
-from . import openssl
-from . import util
-
-
-class SmuggledViaKernelCmdlineDescriptor:
-    def __init__(self):
-        self.kernel_cmdline = None
-
-    def encode(self):
-        return self.kernel_cmdline.encode()
-
-
-@contextlib.contextmanager
-def smuggle_descriptors():
-    '''
-    Smuggle predefined vbmeta descriptors into Avb.make_vbmeta_image via the
-    kernel_cmdlines parameter. The make_vbmeta_image function will:
-
-    * loop through kernel_cmdlines
-    * create a AvbKernelCmdlineDescriptor instance for each item
-    * assign kernel_cmdline to each descriptor instance
-    * call encode on each descriptor
-    '''
-
-    with unittest.mock.patch('avbtool.AvbKernelCmdlineDescriptor',
-                             SmuggledViaKernelCmdlineDescriptor):
-        yield
-
-
-def _get_descriptor_overrides(
-    avb: avbtool.Avb,
-    images: dict[str, os.PathLike[str]],
-) -> typing.Tuple[dict[str, bytes], dict[str, avbtool.AvbDescriptor]]:
-    '''
-    Build a set of public key (chain) and hash/hashtree descriptor overrides
-    that should be inserted in the parent vbmeta image for the given partition
-    images.
-
-    If a partition image itself is signed, then a chain descriptor will be used.
-    Otherwise, the existing hash or hashtree descriptor is used.
-    '''
-
-    # Partition name -> raw public key
-    out_public_keys = {}
-    # Partition name -> descriptor
-    out_descriptors = {}
-
-    # Construct descriptor overrides
-    for name, path in images.items():
-        image = avbtool.ImageHandler(path, read_only=True)
-        footer, header, descriptors, image_size = avb._parse_image(image)
-
-        if name in out_public_keys or name in out_descriptors:
-            raise ValueError(f'Duplicate partition name: {name}')
-
-        if header.public_key_size:
-            # vbmeta is signed; use a chain descriptor
-            blob = avb._load_vbmeta_blob(image)
-            offset = header.SIZE + \
-                header.authentication_data_block_size + \
-                header.public_key_offset
-            out_public_keys[name] = \
-                blob[offset:offset + header.public_key_size]
-        else:
-            # vbmeta is unsigned; use the existing descriptor in the footer
-            partition_descriptor = next(
-                (d for d in descriptors
-                    if (isinstance(d, avbtool.AvbHashDescriptor)
-                        or isinstance(d, avbtool.AvbHashtreeDescriptor))
-                        and d.partition_name == name),
-                None,
-            )
-            if partition_descriptor is None:
-                raise ValueError(f'{path} has no descriptor for itself')
-
-            out_descriptors[name] = partition_descriptor
-
-    return (out_public_keys, out_descriptors)
-
-
-def get_vbmeta_deps(
-    avb: avbtool.Avb,
-    vbmeta_images: dict[str, os.PathLike[str]],
-) -> dict[str, set[str]]:
-    '''
-    Return the forward and reverse dependency tree for the specified vbmeta
-    images.
-    '''
-
-    deps = {}
-
-    for name, path in vbmeta_images.items():
-        image = avbtool.ImageHandler(path, read_only=True)
-        _, _, descriptors, _ = avb._parse_image(image)
-
-        deps.setdefault(name, set())
-
-        for d in descriptors:
-            if isinstance(d, avbtool.AvbChainPartitionDescriptor) \
-                or isinstance(d, avbtool.AvbHashDescriptor) \
-                    or isinstance(d, avbtool.AvbHashtreeDescriptor):
-                deps[name].add(d.partition_name)
-                deps.setdefault(d.partition_name, set())
-
-    return deps
-
-
-def patch_vbmeta_image(
-    avb: avbtool.Avb,
-    images: dict[str, os.PathLike[str]],
-    input_path: os.PathLike[str],
-    output_path: os.PathLike[str],
-    key: os.PathLike[str],
-    passphrase: str,
-    padding_size: int,
-    clear_flags: bool,
-):
-    '''
-    Patch the vbmeta image to reference the provided images.
-    '''
-
-    # Load the original root vbmeta image
-    image = avbtool.ImageHandler(input_path, read_only=True)
-    footer, header, descriptors, image_size = avb._parse_image(image)
-
-    if header.flags != 0:
-        if clear_flags:
-            header.flags = 0
-        else:
-            raise ValueError(f'vbmeta flags disable AVB: 0x{header.flags:x}')
-
-    # Build a set of new descriptors in the same order as the original
-    # descriptors, except with the descriptors patched to reference the given
-    # images
-    override_public_keys, override_descriptors = \
-        _get_descriptor_overrides(avb, images)
-    new_descriptors = []
-
-    for d in descriptors:
-        if isinstance(d, avbtool.AvbChainPartitionDescriptor) and \
-                d.partition_name in override_public_keys:
-            d.public_key = override_public_keys.pop(d.partition_name)
-        elif (isinstance(d, avbtool.AvbHashDescriptor) or \
-                isinstance(d, avbtool.AvbHashtreeDescriptor)) and \
-                d.partition_name in override_descriptors:
-            d = override_descriptors.pop(d.partition_name)
-
-        new_descriptors.append(d)
-
-    if override_public_keys:
-        raise Exception(f'Unused public key overrides: {override_public_keys}')
-    if override_descriptors:
-        raise Exception(f'Unused descriptor overrides: {override_descriptors}')
-
-    algorithm_name = avbtool.lookup_algorithm_by_type(header.algorithm_type)[0]
-
-    # Some older Pixel devices' vbmeta images are originally signed by a
-    # 2048-bit RSA key, but avbroot expects RSA 4096 keys
-    if algorithm_name == 'SHA256_RSA2048':
-        algorithm_name = 'SHA256_RSA4096'
-
-    with util.open_output_file(output_path) as f:
-        # Smuggle in the prebuilt descriptors via kernel_cmdlines
-        with (
-            smuggle_descriptors(),
-            openssl.inject_passphrase(passphrase),
-        ):
-            avb.make_vbmeta_image(
-                output=f,
-                chain_partitions=None,
-                algorithm_name=algorithm_name,
-                key_path=key,
-                public_key_metadata_path=None,
-                rollback_index=header.rollback_index,
-                flags=header.flags,
-                rollback_index_location=header.rollback_index_location,
-                props=None,
-                props_from_file=None,
-                kernel_cmdlines=new_descriptors,
-                setup_rootfs_from_kernel=None,
-                include_descriptors_from_image=None,
-                signing_helper=None,
-                signing_helper_with_files=None,
-                release_string=header.release_string,
-                append_to_release_string=False,
-                print_required_libavb_version=False,
-                padding_size=padding_size,
-            )

deny.toml

@@ -0,0 +1,40 @@
+[advisories]
+vulnerability = "deny"
+unmaintained = "deny"
+yanked = "deny"
+notice = "deny"
+
+[licenses]
+unlicensed = "deny"
+allow = [
+    "Apache-2.0",
+    "BSD-3-Clause",
+    "ISC",
+    "MIT",
+    "OpenSSL",
+    "Unicode-DFS-2016",
+]
+copyleft = "allow"
+default = "deny"
+
+[[licenses.clarify]]
+name = "ring"
+expression = "MIT AND ISC AND OpenSSL"
+license-files = [
+    { path = "LICENSE", hash = 0xbd0eed23 },
+]
+
+[bans]
+multiple-versions = "warn"
+deny = [
+    # https://github.com/serde-rs/serde/issues/2538
+    { name = "serde_derive", version = ">=1.0.172,<1.0.184" },
+]
+
+[sources]
+unknown-registry = "deny"
+unknown-git = "deny"
+allow-git = [
+    "https://github.com/chenxiaolong/zip",
+    "https://github.com/jongiddy/bzip2-rs",
+]

e2e/Cargo.toml

@@ -0,0 +1,29 @@
+[package]
+name = "e2e"
+version.workspace = true
+license.workspace = true
+edition.workspace = true
+repository.workspace = true
+publish = false
+
+# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+
+[dependencies]
+anyhow = "1.0.75"
+avbroot = { path = "../avbroot" }
+clap = { version = "4.4.1", features = ["derive"] }
+ctrlc = "3.4.0"
+hex = { version = "0.4.3", features = ["serde"] }
+reqwest = { version = "0.11.20", features = ["stream"] }
+ring = "0.16.20"
+serde = { version = "1.0.188", features = ["derive"] }
+tempfile = "3.8.0"
+tokio = { version = "1.32.0", features = ["signal", "rt-multi-thread", "macros"] }
+tokio-stream = "0.1.14"
+toml_edit = { version = "0.19.14", features = ["serde"] }
+
+# https://github.com/zip-rs/zip/pull/383
+[dependencies.zip]
+git = "https://github.com/chenxiaolong/zip"
+rev = "989101f9384b9e94e36e6e9e0f51908fdf98bde6"
+default-features = false

e2e/README.md

@@ -0,0 +1,64 @@
+# End-to-end tests
+
+avbroot's output file is reproducible for a given input file. [`e2e.toml`](./e2e.toml) lists some OTA images with unique properties and the expected checksums before and after patching. These tests use pregenerated, hardcoded test keys for signing. **These keys should NEVER be used for any other purpose.**
+
+For each image listed in the config, the test process will:
+
+1. Download the OTA if it doesn't already exist in `./files/<device>/` (or the workdir specified by `-w`)
+2. Verify the OTA checksum
+3. Run avbroot against the OTA using `--magisk`
+4. Extract the AVB-related partitions from the patched OTA and verify their checksums
+5. Verify the patched OTA checksum
+6. Run avbroot against the OTA again using `--prepatched`
+7. Verify the patched OTA checksum again
+
+For more efficient CI testing, the tests can operate on "stripped" OTAs. A stripped OTA is identical to the full OTA, except that partitions in `payload.bin` unrelated to AVB are zeroed out. This reduces the download size and disk space requirements by a couple orders of magnitude. **A stripped OTA is NOT bootable and should never be flashed on a real device.**
+
+## Running the tests
+
+To test against the device OTA images listed in [`e2e.toml`](./e2e.toml), run:
+
+```bash
+# To test all device OTAs
+cargo run --release -- test -a
+# Or to test against specific device OTAs
+cargo run --release -- test -d cheetah -d bluejay
+```
+
+To test against stripped OTAs (smaller download, but not bootable), pass in `--stripped`.
+
+## Downloading a device image
+
+To download a full OTA image, run:
+
+```bash
+cargo run --release -- download -d <device>
+```
+
+This normally happens automatically when running the `test` subcommand. To download the stripped OTA image instead, pass in `--stripped`.
+
+If the image file does not already exist, then it will be downloaded and the checksums will be validated. If the download is interrupted, it will automatically resume when the command is rerun. If the file is already downloaded, the command is effectively a no-op unless `--revalidate` is passed in to revalidate the image checksums.
+
+## Adding a new device image
+
+To add a new device image to the testing configuration, run:
+
+```bash
+cargo run --release -- add -d <device> -u <full OTA URL> -H <expected checksum>
+```
+
+If the OS vendor does not provide a SHA-256 checksum, omit `-H` and the program will compute the checksum from the downloaded data.
+
+This process will download the full OTA, strip it, patch the full OTA, patch the stripped OTA, extract the AVB partitions, and write all of the checksums to [`e2e.toml`](./e2e.toml).
+
+The process for updating an existing device config is exactly the same as adding a new one.
+
+## Stripping a full OTA
+
+To convert a full OTA to the stripped form, run:
+
+```bash
+cargo run --release -- strip -i <input zip> -o <output zip>
+```
+
+This normally happens automatically as a part of adding a new device image.

e2e/e2e.toml

@@ -0,0 +1,116 @@
+[magisk]
+"url" = "https://github.com/topjohnwu/Magisk/releases/download/v26.0/Magisk-v26.0.apk"
+"hash" = "9e14d3d3ca1f1a2765f8ca215ebbf35ea5fd2896fb147eea581fcaa3b4e77d25"
+
+# Google Pixel 7 Pro
+# What's unique: init_boot (boot v4) + vendor_boot (vendor v4)
+[device.cheetah]
+url = "https://dl.google.com/dl/android/aosp/cheetah-ota-tq2a.230305.008.c1-6ac5ff2e.zip"
+sections = [
+    { start = 0, end = 151715 },
+    { start = 21683150, end = 23179365 },
+    { start = 2043499985, end = 2043508325 },
+    { start = 2315331822, end = 2333060126 },
+    { start = 2344084950, end = 2344090066 },
+]
+hash.original.full = "6ac5ff2e14dc16755ea4ea30e6dbe25103b889a36a465194ef943bd0d665b91c"
+hash.original.stripped = "549522015f0369a3b89385f532ab62235b47c2c39540bd0adaaf6acc81fdda94"
+hash.patched.full = "b380720852e2a1e994bcf38064f577bac68b18e799cf8166cb6a7edb8a661cb4"
+hash.patched.stripped = "560cdf4d7b25fb5bc650f96ae5a3c7593ac229a364bb39887287cfb734f6b377"
+hash.avb_images."init_boot.img" = "fac9305ce22b897fbfb193968d5346f4c70f6c18c3060bb106226f134ce5f433"
+hash.avb_images."vbmeta.img" = "0b3d719b751dd43bbec95d02b1bf57b5dae62e42a52502895892a207b773e77d"
+hash.avb_images."vbmeta_system.img" = "cf8c77dcf0a4474d49b5bdc2a44bdb3646464d5212fbe12aa5d3c5f531742f4f"
+hash.avb_images."vbmeta_vendor.img" = "660d8f61acd95a4f8ad416b4cbe126e9c039706462b4236ad723953c72ac49a8"
+hash.avb_images."vendor_boot.img" = "c788d4d8eb7926ad1bfa2a9c000343c3c73cedc580449f5270a711feb620f033"
+
+# Google Pixel 6a
+# What's unique: boot (boot v4, no ramdisk) + vendor_boot (vendor v4, 2 ramdisks)
+[device.bluejay]
+url = "https://dl.google.com/dl/android/aosp/bluejay-ota-tq2a.230305.008.e1-915f9087.zip"
+sections = [
+    { start = 0, end = 140787 },
+    { start = 1060207, end = 21612852 },
+    { start = 1886150700, end = 1886158844 },
+    { start = 2069112987, end = 2092260102 },
+    { start = 2098778558, end = 2098783674 },
+]
+hash.original.full = "915f9087b627b6961be9bb447dc63a7a1083b536753a78715e98641eaeb9c9d1"
+hash.original.stripped = "a3ee5b6e39e687665c31790118ab9f47715b0b8285ae9847dbf81307f963db14"
+hash.patched.full = "bfa7b26d90bdc889a7a199439e1564b219e5e2dbfddc1657bc1c6b73229be67e"
+hash.patched.stripped = "4e56ee4a8554f2b08ffc2f1470ad60b9b63d2b6fb469ed1f7c4b1204bbf8ad7d"
+hash.avb_images."boot.img" = "a1a705092e7034d20b83c94d78291418e13c343b1573c1b11e0fc884fc00ae62"
+hash.avb_images."vbmeta.img" = "3c123705be57ab142d2b43beef9b123eaad129df17a523b59b1d63b9122d28b0"
+hash.avb_images."vbmeta_system.img" = "285b83e4290f3257dc3678f0c3191794830bb2d72fb0969b69fc8f09d7ddff12"
+hash.avb_images."vbmeta_vendor.img" = "981f736586b91a9f4c93c4208a0d191a35ff15118c6fa505d755ee7fda8b2477"
+hash.avb_images."vendor_boot.img" = "ceffcb4fdb33aa3bb2c70621060acd5af612206b21cc413d5c2d39fee25144ab"
+
+# Google Pixel 4a 5G
+# What's unique: boot (boot v3) + vendor_boot (vendor v3)
+[device.bramble]
+url = "https://dl.google.com/dl/android/aosp/bramble-ota-tq2a.230305.008.c1-a925dd09.zip"
+sections = [
+    { start = 0, end = 140531 },
+    { start = 496187, end = 11655082 },
+    { start = 1650283561, end = 1650287993 },
+    { start = 1884739919, end = 1908081011 },
+    { start = 1910894027, end = 1910899144 },
+]
+hash.original.full = "a925dd09c8d613d46cf72677c16f4fadee18bc21734d57047c6ccf31f672507b"
+hash.original.stripped = "79322b0b417359e8f072032de676d7e5bd2715a3b3554c48ed5cc9e9a25c6866"
+hash.patched.full = "cf79cd60acd3635f5085d8bb411d4a8dc4d7e62440f62a14e7ea12f5a9b7cd8a"
+hash.patched.stripped = "b684a78fe08014f1633e74d1f380e137823ba77a47ddf428bc649364f5548b75"
+hash.avb_images."boot.img" = "2bbf2c6d2f82d454426b26ac3b4887b26ca0591b458e6cf137207ecbb8f7649d"
+hash.avb_images."vbmeta.img" = "ab3b2487671b3fc28e163898621d6c068164d335b6ddd9b94e4fde9471c95d66"
+hash.avb_images."vbmeta_system.img" = "2fcd52d7462916a8510bbb07f2f5a14200afe2de97568396fe75e04c5c283152"
+hash.avb_images."vendor_boot.img" = "e27f157c4ebf4e958165997a4b87d4de1cfc34f4dea423a153327a27b053cac7"
+
+# Google Pixel 4a
+# What's unique: boot (boot v2)
+[device.sunfish]
+url = "https://dl.google.com/dl/android/aosp/sunfish-ota-tq2a.230305.008.c1-174fd16b.zip"
+sections = [
+    { start = 0, end = 129700 },
+    { start = 476996, end = 34035261 },
+    { start = 1624751947, end = 1624756371 },
+    { start = 1823132563, end = 1823137581 },
+]
+hash.original.full = "174fd16b47ef994ea8f3cb0f3fb456df2654b0aa1f9ea6fb8e54e5c6319f2601"
+hash.original.stripped = "943ce3ae2aac8a0ccd4a7e9d4e38a9c495c39639f2621c6853be4c7a3fa0fc26"
+hash.patched.full = "cdfffa731f0aca0ab9eff1e7d7bfee8c4716054d64cf0a47e50f50da6cbdb849"
+hash.patched.stripped = "35b720058c460dd28b00f40bed4bcd1d133522d42d00ba048172b3092a7444ee"
+hash.avb_images."boot.img" = "22182f2efc7043f35d79e71abb400ce919c6b3b8419405e16469644932367ee6"
+hash.avb_images."vbmeta.img" = "4cbe171bb37515f59cd4082cc2d982ca1edfeba9db09dd62c8959dd318423888"
+hash.avb_images."vbmeta_system.img" = "7cdb590bfc1056a5a8c7606ff05e99eb344efe108296682698b5cfe83905e0cd"
+
+# OnePlus 10 Pro
+# Build NE2215_11_C.26
+# What's unique:
+# - boot (boot v4) + recovery (boot v4)
+# - boot images have VTS signature block filled with all 0s
+# - payload.bin uses ZERO blocks
+# Build info:
+# - Unofficial list of full OTAs: https://forum.xda-developers.com/t/oneplus-10-pro-rom-ota-oxygen-os-repo-of-oxygen-os-builds.4572593/
+# - The North American builds are used because they're the only ones hosted on
+#   a well known domain
+# - The build number can be found in <my_manifest>/build.prop since it's not
+#   obvious from the filename
+[device.ossi]
+url = "https://android.googleapis.com/packages/ota-api/package/4cacbe5e6a3ab6a6fade68cc40f44d0fa6a2928a.zip"
+sections = [
+    { start = 0, end = 204048 },
+    { start = 19105432, end = 34966775 },
+    { start = 2657405750, end = 2657407254 },
+    { start = 4984045446, end = 5006377281 },
+    { start = 5114504441, end = 5114507197 },
+    { start = 5138158449, end = 5138159817 },
+    { start = 5140101511, end = 5140105324 },
+]
+hash.original.full = "929f892fbd70699cf7f118a119aac1ae1b86351e1ada17715666fa4401e63472"
+hash.original.stripped = "4eabaf79b6c2b5df305e3ecdc2b9570c0dd27350b4e8d6434584000c4989ff3d"
+hash.patched.full = "8e9cf3159e57d706047325a7b774dc2dc84f0d56baae100ad54c0c723b621469"
+hash.patched.stripped = "0b5dcdfdfea742bf6662b286ba260f8572a7b0081bb1129b7274c29214cdfb49"
+hash.avb_images."boot.img" = "f5dc3b147c54589be8db00ca15257a3688424a9eb88bd9c2cec82ebf4f6bf859"
+hash.avb_images."recovery.img" = "a42c0bf4f023cd24394184a33ee113783a9c89a7cc4c0c582a5f72cc23b72309"
+hash.avb_images."vbmeta.img" = "c022cf79da301a8430af5c49704944c490707fa0306031fe3ea22c39ce4734f6"
+hash.avb_images."vbmeta_system.img" = "749616b7f04487c05e9e363ad2071a0ab3bae29d497daf1f1a7695f7c8cfa82a"
+hash.avb_images."vbmeta_vendor.img" = "a6037fce745384425fb12745b8568386b84fb57ca6f94f6e47bcf754de341ae4"

e2e/keys/TEST_KEY_DO_NOT_USE_avb.passphrase

@@ -0,0 +1 @@
+XltUCz36vqCNSzspPZxFMGXah3kLyrTXDwfmasgn6nL4CtZDw5OeeLwlmkDuV2Im

e2e/keys/TEST_KEY_DO_NOT_USE_ota.passphrase

@@ -0,0 +1 @@
+7DsqL2Sk9T609OFpeVXwnrWHRrK3iazccxEDHWDqr5zJ9tgZkONhDvXhXuCQY76o

e2e/src/cli.rs

@@ -0,0 +1,181 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::{ffi::OsString, path::PathBuf, str::FromStr};
+
+use anyhow::Result;
+use clap::{Args, Parser, Subcommand};
+
+#[derive(Debug, Args)]
+pub struct DeviceGroup {
+    /// Device config name.
+    #[arg(short, long, value_name = "NAME")]
+    pub device: Vec<String>,
+
+    /// All device configs.
+    #[arg(short, long, conflicts_with = "device")]
+    pub all: bool,
+}
+
+#[derive(Debug, Args)]
+pub struct DownloadGroup {
+    /// Revalidate hash of existing download.
+    #[arg(long)]
+    pub revalidate: bool,
+
+    /// Download the stripped OTA instead of the full OTA.
+    #[arg(long)]
+    pub stripped: bool,
+}
+
+#[derive(Debug, Args)]
+pub struct PatchGroup {
+    /// Delete patched output files on success.
+    #[arg(long)]
+    pub delete_on_success: bool,
+
+    /// Suffix for patched output files.
+    #[arg(long = "output-file-suffix", value_parser, default_value = ".patched")]
+    pub suffix: OsString,
+}
+
+#[derive(Debug, Args)]
+pub struct ConfigGroup {
+    /// Path to config file.
+    #[arg(
+        short,
+        long,
+        value_name = "FILE",
+        value_parser,
+        default_value = "e2e.toml"
+    )]
+    pub config: PathBuf,
+
+    /// Working directory for storing images.
+    #[arg(
+        short,
+        long,
+        value_name = "DIRECTORY",
+        value_parser,
+        default_value = "files"
+    )]
+    pub work_dir: PathBuf,
+}
+
+/// Convert a full OTA to stripped form.
+///
+/// A stripped OTA omits byte regions of the OTA that aren't needed for testing
+/// avbroot's patching logic (eg. the system partition image). This reduces the
+/// size of the test files by about two orders of magnitude.
+#[derive(Debug, Parser)]
+pub struct StripCli {
+    /// Path to original OTA zip.
+    #[arg(short, long, value_name = "FILE", value_parser)]
+    pub input: PathBuf,
+
+    /// Path to new stripped OTA zip.
+    #[arg(short, long, value_name = "FILE", value_parser)]
+    pub output: PathBuf,
+}
+
+#[derive(Debug, Clone)]
+pub struct Sha256Arg(pub [u8; 32]);
+
+impl FromStr for Sha256Arg {
+    type Err = hex::FromHexError;
+
+    fn from_str(s: &str) -> Result<Self, Self::Err> {
+        let mut data = [0u8; 32];
+        hex::decode_to_slice(s, &mut data)?;
+        Ok(Self(data))
+    }
+}
+
+/// Add a new OTA image to the test config.
+///
+/// This will download the OTA image, strip it, patch both images, and add the
+/// resulting metadata (eg. checksums) to the specified test config file.
+#[derive(Debug, Parser)]
+pub struct AddCli {
+    /// URL to the full OTA zip.
+    #[arg(short, long)]
+    pub url: String,
+
+    /// Device config name.
+    #[arg(short, long, value_name = "NAME")]
+    pub device: String,
+
+    /// Expected sha256 hash of the full OTA zip.
+    #[arg(short = 'H', long, value_name = "SHA256_HEX", value_parser)]
+    pub hash: Option<Sha256Arg>,
+
+    #[command(flatten)]
+    pub patch: PatchGroup,
+
+    #[command(flatten)]
+    pub config: ConfigGroup,
+
+    /// Skip verifying OTA and AVB signatures.
+    ///
+    /// OTAs for some devices (eg. ossi) ship with vbmeta partitions containing
+    /// invalid hashes. These will normally fail during validation.
+    #[arg(long)]
+    pub skip_verify: bool,
+}
+
+/// Download a device image.
+#[derive(Debug, Parser)]
+pub struct DownloadCli {
+    /// Download the Magisk APK.
+    #[arg(short, long)]
+    pub magisk: bool,
+
+    #[command(flatten)]
+    pub device: DeviceGroup,
+
+    #[command(flatten)]
+    pub download: DownloadGroup,
+
+    #[command(flatten)]
+    pub config: ConfigGroup,
+}
+
+/// Run tests.
+#[derive(Debug, Parser)]
+pub struct TestCli {
+    #[command(flatten)]
+    pub device: DeviceGroup,
+
+    #[command(flatten)]
+    pub download: DownloadGroup,
+
+    #[command(flatten)]
+    pub patch: PatchGroup,
+
+    #[command(flatten)]
+    pub config: ConfigGroup,
+}
+
+/// List devices in config file.
+#[derive(Debug, Parser)]
+pub struct ListCli {
+    #[command(flatten)]
+    pub config: ConfigGroup,
+}
+
+#[derive(Debug, Subcommand)]
+pub enum Command {
+    Strip(StripCli),
+    Add(AddCli),
+    Download(DownloadCli),
+    Test(TestCli),
+    List(ListCli),
+}
+
+#[derive(Debug, Parser)]
+pub struct Cli {
+    #[command(subcommand)]
+    pub command: Command,
+}

e2e/src/config.rs

@@ -0,0 +1,140 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::{collections::BTreeMap, fs, ops::Range, path::Path};
+
+use anyhow::{anyhow, Context, Result};
+use serde::{Deserialize, Serialize};
+use toml_edit::{
+    ser::ValueSerializer,
+    visit_mut::{self, VisitMut},
+    Array, Document, InlineTable, Item, KeyMut, Table, Value,
+};
+
+#[derive(Serialize, Deserialize)]
+pub struct Sha256Hash(
+    #[serde(
+        serialize_with = "hex::serialize",
+        deserialize_with = "hex::deserialize"
+    )]
+    pub [u8; 32],
+);
+
+#[derive(Serialize, Deserialize)]
+pub struct Magisk {
+    pub url: String,
+    pub hash: Sha256Hash,
+}
+
+#[derive(Serialize, Deserialize)]
+pub struct OtaHashes {
+    pub full: Sha256Hash,
+    pub stripped: Sha256Hash,
+}
+
+#[derive(Serialize, Deserialize)]
+pub struct ImageHashes {
+    pub original: OtaHashes,
+    pub patched: OtaHashes,
+    pub avb_images: BTreeMap<String, Sha256Hash>,
+}
+
+#[derive(Serialize, Deserialize)]
+pub struct Device {
+    pub url: String,
+    pub sections: Vec<Range<u64>>,
+    pub hash: ImageHashes,
+}
+
+#[derive(Serialize, Deserialize)]
+pub struct Config {
+    pub magisk: Magisk,
+    pub device: BTreeMap<String, Device>,
+}
+
+struct ConfigFormatter;
+
+impl VisitMut for ConfigFormatter {
+    fn visit_table_like_kv_mut(&mut self, key: KeyMut<'_>, node: &mut Item) {
+        // Convert non-array-of-tables inline tables into regular tables.
+        if let Item::Value(Value::InlineTable(t)) = node {
+            let inline_table = std::mem::replace(t, InlineTable::new());
+            *node = Item::Table(inline_table.into_table());
+        }
+
+        // But for hashes, use dotted notation until TOML 1.1, which allows
+        // newlines in inline tables, is released.
+        if key == "hash" || key == "original" || key == "patched" || key == "avb_images" {
+            if let Some(t) = node.as_table_like_mut() {
+                t.set_dotted(true);
+            }
+        }
+
+        visit_mut::visit_table_like_kv_mut(self, key, node);
+    }
+
+    fn visit_table_mut(&mut self, node: &mut Table) {
+        // Make tables implicit unless they are empty, which may be meaningful.
+        if !node.is_empty() {
+            node.set_implicit(true);
+        }
+
+        visit_mut::visit_table_mut(self, node);
+    }
+
+    fn visit_array_mut(&mut self, node: &mut Array) {
+        visit_mut::visit_array_mut(self, node);
+
+        // Put array elements on their own indented lines.
+        if node.is_empty() {
+            node.set_trailing("");
+            node.set_trailing_comma(false);
+        } else {
+            for item in node.iter_mut() {
+                item.decor_mut().set_prefix("\n    ");
+            }
+            node.set_trailing("\n");
+            node.set_trailing_comma(true);
+        }
+    }
+}
+
+/// Add a device to the config file. This leaves all comments intact, except for
+/// those contained within the existing device section if it exists.
+pub fn add_device(document: &mut Document, name: &str, device: &Device) -> Result<()> {
+    let device_table = document.entry("device").or_insert_with(|| {
+        let mut t = toml_edit::Table::new();
+        t.set_implicit(true);
+        Item::Table(t)
+    });
+    let old_table = device_table.get(name).and_then(|i| i.as_table());
+
+    let value = device.serialize(ValueSerializer::new())?;
+    let Value::InlineTable(inline_table) = value else {
+        unreachable!("Device did not serialize as an inline table");
+    };
+    let mut table = inline_table.into_table();
+
+    ConfigFormatter.visit_table_mut(&mut table);
+
+    // Keep top-level comment on the table.
+    if let Some(t) = old_table {
+        *table.decor_mut() = t.decor().clone();
+    }
+
+    device_table[name] = Item::Table(table);
+
+    Ok(())
+}
+
+pub fn load_config(path: &Path) -> Result<(Config, Document)> {
+    let contents =
+        fs::read_to_string(path).with_context(|| anyhow!("Failed to read config: {path:?}"))?;
+    let config: Config = toml_edit::de::from_str(&contents)
+        .with_context(|| anyhow!("Failed to parse config: {path:?}"))?;
+    let document: Document = contents.parse().unwrap();
+
+    Ok((config, document))
+}

e2e/src/download.rs

@@ -0,0 +1,456 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+use std::{
+    cmp,
+    collections::{HashMap, VecDeque},
+    fs::{self, OpenOptions},
+    io::{self, Seek, SeekFrom, Write},
+    ops::Range,
+    path::{Path, PathBuf},
+    time::{Duration, Instant},
+};
+
+use anyhow::{anyhow, bail, Context, Result};
+use avbroot::stream::PSeekFile;
+use serde::{Deserialize, Serialize};
+use tokio::{
+    runtime::Runtime,
+    signal::ctrl_c,
+    sync::{mpsc, oneshot},
+    task::{self, JoinSet},
+};
+use tokio_stream::StreamExt;
+
+/// Minimum download chunk size per task.
+const MIN_CHUNK_SIZE: u64 = 1024 * 1024;
+
+pub trait ProgressDisplay {
+    fn progress(&mut self, current: u64, total: u64);
+
+    fn error(&mut self, msg: &str);
+
+    fn finish(&mut self);
+}
+
+pub struct BasicProgressDisplay {
+    current: u64,
+    total: u64,
+    interval: Duration,
+    last_render: Instant,
+    avg: VecDeque<(Instant, u64)>,
+}
+
+// Speed is a simple moving average over 5 seconds.
+static AVG_INTERVAL: Duration = Duration::from_millis(100);
+static AVG_WINDOW_SIZE: usize = 5000 / AVG_INTERVAL.as_millis() as usize;
+
+impl BasicProgressDisplay {
+    pub fn new(interval: Duration) -> Self {
+        Self {
+            current: 0,
+            total: 0,
+            interval,
+            last_render: Instant::now() - interval,
+            avg: VecDeque::new(),
+        }
+    }
+
+    fn clear_line(&self) {
+        eprint!("\x1b[2K\r");
+    }
+}
+
+impl ProgressDisplay for BasicProgressDisplay {
+    fn progress(&mut self, current: u64, total: u64) {
+        self.current = current;
+        self.total = total;
+
+        let now = Instant::now();
+
+        if self.avg.is_empty() || (now - self.avg.back().unwrap().0) > AVG_INTERVAL {
+            if self.avg.len() == AVG_WINDOW_SIZE {
+                self.avg.pop_front();
+            }
+
+            self.avg.push_back((now, current));
+        }
+
+        if now - self.last_render > self.interval {
+            let current_mib = current as f64 / 1024.0 / 1024.0;
+            let total_mib = total as f64 / 1024.0 / 1024.0;
+
+            let front = self.avg.front().unwrap();
+            let back = self.avg.back().unwrap();
+
+            let avg_window_mib = (back.1 - front.1) as f64 / 1024.0 / 1024.0;
+            let avg_window_duration = back.0 - front.0;
+
+            let speed_mib_s = if avg_window_duration.is_zero() {
+                0.0
+            } else {
+                avg_window_mib / avg_window_duration.as_secs_f64()
+            };
+
+            self.clear_line();
+            eprint!("{current_mib:.1} / {total_mib:.1} MiB ({speed_mib_s:.1} MiB/s)");
+
+            self.last_render = now;
+        }
+    }
+
+    fn error(&mut self, msg: &str) {
+        self.clear_line();
+        eprintln!("{msg}");
+    }
+
+    fn finish(&mut self) {
+        self.clear_line();
+    }
+}
+
+#[derive(Debug)]
+struct ProgressMessage {
+    task_id: u64,
+    bytes: u64,
+    // Controller replies with new ending offset
+    resp: oneshot::Sender<u64>,
+}
+
+/// Download a contiguous byte range. The number of bytes downloaded per loop
+/// iteration will be sent to the specified channel via a `ProgressMessage`. The
+/// receiver of the message must reply with the new ending offset for this
+/// download via the oneshot channel in the `resp` field. An appropriate error
+/// will be returned if the full range (subject to modification) cannot be fully
+/// downloaded (eg. premature EOF is an error).
+async fn download_range(
+    task_id: u64,
+    url: &str,
+    mut file: PSeekFile,
+    initial_range: Range<u64>,
+    channel: mpsc::Sender<ProgressMessage>,
+) -> Result<()> {
+    assert!(initial_range.start < initial_range.end);
+
+    let client = reqwest::ClientBuilder::new().build()?;
+
+    let response = client
+        .get(url)
+        .header(
+            reqwest::header::RANGE,
+            format!("bytes={}-{}", initial_range.start, initial_range.end - 1),
+        )
+        .send()
+        .await
+        .and_then(|r| r.error_for_status())
+        .with_context(|| anyhow!("Failed to start download for range: {initial_range:?}"))?;
+    let mut stream = response.bytes_stream();
+    let mut range = initial_range.clone();
+
+    while range.start < range.end {
+        let data = if let Some(x) = stream.next().await {
+            x?
+        } else {
+            return Err(anyhow!("Unexpected EOF from server"));
+        };
+
+        // This may overlap with another task's write when a range split occurs,
+        // but the same data will be written anyway, so it's not a huge deal.
+        task::block_in_place(|| {
+            file.seek(SeekFrom::Start(range.start))?;
+            file.write_all(&data)
+        })
+        .with_context(|| {
+            format!(
+                "Failed to write {} bytes to output file at offset {}",
+                data.len(),
+                range.start,
+            )
+        })?;
+
+        let consumed = cmp::min(range.end - range.start, data.len() as u64);
+        range.start += consumed;
+
+        // Report progress to the controller.
+        let (tx, rx) = oneshot::channel();
+        let msg = ProgressMessage {
+            task_id,
+            bytes: consumed,
+            resp: tx,
+        };
+        channel.send(msg).await?;
+
+        // Get new ending offset from controller.
+        let new_end = rx.await?;
+        if new_end != range.end {
+            debug_assert!(new_end <= range.end);
+            range.end = new_end;
+        }
+    }
+
+    Ok(())
+}
+
+/// Create download task for a byte range. This just calls [`download_range()`]
+/// and returns a tuple containing the task ID and the result.
+async fn download_task(
+    task_id: u64,
+    url: String,
+    file: PSeekFile,
+    initial_range: Range<u64>,
+    channel: mpsc::Sender<ProgressMessage>,
+) -> (u64, Result<()>) {
+    (
+        task_id,
+        download_range(task_id, &url, file, initial_range, channel).await,
+    )
+}
+
+/// Send a HEAD request to get the value of the Content-Length header.
+async fn get_content_length(url: &str) -> Result<u64> {
+    let response = reqwest::Client::new()
+        .head(url)
+        .send()
+        .await
+        .and_then(|r| r.error_for_status())
+        .context("Failed to send HEAD request to get Content-Length")?;
+
+    response
+        .headers()
+        .get("content-length")
+        .and_then(|h| h.to_str().ok())
+        .and_then(|h| h.parse().ok())
+        .ok_or_else(|| anyhow!("HEAD request did not return a valid Content-Length"))
+}
+
+/// Download a set of file chunks in parallel. Only unrecoverable errors are
+/// returned as an Err. Normal/expected errors and download progress info are
+/// reported via `display`. Returns the remaining ranges that need to be
+/// downloaded.
+async fn download_ranges(
+    url: &str,
+    output: &Path,
+    initial_ranges: Option<&[Range<u64>]>,
+    display: &mut dyn ProgressDisplay,
+    max_tasks: usize,
+    max_errors: u8,
+) -> Result<Vec<Range<u64>>> {
+    let file_size = get_content_length(url).await?;
+
+    // Open for writing, but without truncation.
+    let file = task::block_in_place(|| {
+        OpenOptions::new()
+            .write(true)
+            .create(true)
+            .open(output)
+            .map(PSeekFile::new)
+            .with_context(|| anyhow!("Failed to open for writing: {output:?}"))
+    })?;
+
+    task::block_in_place(|| file.set_len(file_size))
+        .with_context(|| anyhow!("Failed to set file size: {output:?}"))?;
+
+    // Queue of ranges that need to be downloaded.
+    let mut remaining = VecDeque::from(match initial_ranges {
+        Some(r) => r.to_vec(),
+        #[allow(clippy::single_range_in_vec_init)]
+        None => vec![0..file_size],
+    });
+    // Ranges that have failed.
+    let mut failed = Vec::<Range<u64>>::new();
+    // Ranges for currently running tasks.
+    let mut task_ranges = HashMap::<u64, Range<u64>>::new();
+
+    // Overall progress.
+    let mut progress = file_size - remaining.iter().map(|r| r.end - r.start).sum::<u64>();
+    display.progress(progress, file_size);
+
+    let mut tasks = JoinSet::new();
+    let mut next_task_id = 0;
+    let mut error_count = 0u8;
+    // Progress messages from tasks.
+    let (tx, mut rx) = mpsc::channel(max_tasks);
+
+    loop {
+        // Spawn new tasks.
+        while tasks.len() < max_tasks {
+            if remaining.is_empty() && !tasks.is_empty() {
+                // No more ranges to download. Split another task's range.
+                let (_, old_range) = task_ranges
+                    .iter_mut()
+                    .max_by_key(|(_, r)| r.end - r.start)
+                    .unwrap();
+                let size = old_range.end - old_range.start;
+
+                if size >= MIN_CHUNK_SIZE {
+                    let new_range = old_range.start + size / 2..old_range.end;
+                    old_range.end = new_range.start;
+                    remaining.push_back(new_range);
+                }
+            }
+
+            if let Some(task_range) = remaining.pop_front() {
+                tasks.spawn(download_task(
+                    next_task_id,
+                    url.to_owned(),
+                    file.clone(),
+                    task_range.clone(),
+                    tx.clone(),
+                ));
+
+                task_ranges.insert(next_task_id, task_range);
+                next_task_id += 1;
+            } else {
+                // No pending ranges and no running tasks can be split.
+                break;
+            }
+        }
+
+        tokio::select! {
+            // Interrupted by user.
+            c = ctrl_c() => {
+                c?;
+                break;
+            }
+
+            // Received progress notification.
+            msg = rx.recv() => {
+                let msg = msg.unwrap();
+
+                progress += msg.bytes;
+                display.progress(progress, file_size);
+
+                let task_range = task_ranges.get_mut(&msg.task_id).unwrap();
+                task_range.start += msg.bytes;
+
+                msg.resp.send(task_range.end).unwrap();
+            }
+
+            // Received completion message.
+            r = tasks.join_next() => {
+                match r {
+                    // All tasks exited.
+                    None => {
+                        break;
+                    },
+
+                    // Download task panicked.
+                    Some(Err(e)) => {
+                        return Err(e).context("Unexpected panic in download task");
+                    }
+
+                    // Task completed successfully.
+                    Some(Ok((task_id, Ok(_)))) => {
+                        task_ranges.remove(&task_id).unwrap();
+                    }
+
+                    // Task failed.
+                    Some(Ok((task_id, Err(e)))) => {
+                        display.error(&format!("[Task#{task_id}] {e}"));
+                        error_count += 1;
+
+                        let range = task_ranges.remove(&task_id).unwrap();
+
+                        if error_count < max_errors {
+                            remaining.push_back(range);
+                        } else {
+                            failed.push(range);
+                        }
+                    }
+                }
+            }
+        }
+    }
+
+    display.finish();
+
+    failed.extend(remaining.into_iter());
+    failed.extend(task_ranges.into_values());
+
+    Ok(failed)
+}
+
+#[derive(Serialize, Deserialize)]
+struct State {
+    ranges: Vec<Range<u64>>,
+}
+
+fn read_state(path: &Path) -> Result<Option<State>> {
+    let data = match fs::read_to_string(path) {
+        Ok(f) => f,
+        Err(e) if e.kind() == io::ErrorKind::NotFound => return Ok(None),
+        Err(e) => Err(e).with_context(|| anyhow!("Failed to read download state: {path:?}"))?,
+    };
+
+    let state = toml_edit::de::from_str(&data)
+        .with_context(|| anyhow!("Failed to parse download state: {path:?}"))?;
+
+    Ok(Some(state))
+}
+
+fn write_state(path: &Path, state: &State) -> Result<()> {
+    let data = toml_edit::ser::to_string(state).unwrap();
+
+    fs::write(path, data).with_context(|| anyhow!("Failed to write download state: {path:?}"))?;
+
+    Ok(())
+}
+
+fn delete_if_exists(path: &Path) -> Result<()> {
+    if let Err(e) = fs::remove_file(path) {
+        if e.kind() != io::ErrorKind::NotFound {
+            return Err(e).context(format!("Failed to delete file: {path:?}"));
+        }
+    }
+
+    Ok(())
+}
+
+pub fn state_path(path: &Path) -> PathBuf {
+    let mut s = path.as_os_str().to_owned();
+    s.push(".state");
+    PathBuf::from(s)
+}
+
+/// Download `url` to `output` with parallel threads.
+///
+/// If `initial_ranges` is specified, only those sections of the file will be
+/// downloaded. The empty regions are left untouched (i.e. filled with zeroes).
+/// A `.state` file is written if the download is interrupted. If the state
+/// file exists when this function is called, `initial_ranges` is ignored and
+/// the ranges from the state file are used to resume the download.
+pub fn download(
+    url: &str,
+    output: &Path,
+    initial_ranges: Option<&[Range<u64>]>,
+    display: &mut dyn ProgressDisplay,
+    max_tasks: usize,
+    max_errors: u8,
+) -> Result<()> {
+    let state_path = state_path(output);
+    let ranges = match read_state(&state_path)? {
+        Some(r) => Some(r.ranges),
+        None => initial_ranges.map(|r| r.to_vec()),
+    };
+
+    let runtime = Runtime::new()?;
+    let remaining = runtime.block_on(download_ranges(
+        url,
+        output,
+        ranges.as_deref(),
+        display,
+        max_tasks,
+        max_errors,
+    ))?;
+
+    if remaining.is_empty() {
+        delete_if_exists(&state_path)?;
+    } else {
+        write_state(&state_path, &State { ranges: remaining })?;
+        bail!("Download was interrupted");
+    }
+
+    Ok(())
+}

e2e/src/main.rs

@@ -0,0 +1,794 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+ * SPDX-FileCopyrightText: 2023 Pascal Roeleven
+ * SPDX-License-Identifier: GPL-3.0-only
+ */
+
+mod cli;
+mod config;
+mod download;
+
+use std::{
+    collections::{BTreeMap, BTreeSet, HashSet},
+    ffi::{OsStr, OsString},
+    fs::{self, File},
+    io::{self, BufReader, BufWriter, Seek, SeekFrom},
+    ops::Range,
+    path::{Path, PathBuf},
+    sync::{
+        atomic::{AtomicBool, Ordering},
+        Arc,
+    },
+    time::Duration,
+};
+
+use anyhow::{anyhow, bail, Context, Result};
+use avbroot::{
+    cli::ota::{ExtractCli, PatchCli, VerifyCli},
+    format::{ota, payload::PayloadHeader},
+    stream::{self, FromReader, HashingReader, PSeekFile, SectionReader},
+};
+use clap::Parser;
+use tempfile::TempDir;
+use zip::ZipArchive;
+
+use crate::{
+    cli::{AddCli, Cli, Command, DeviceGroup, DownloadCli, ListCli, StripCli, TestCli},
+    config::{Config, Device, ImageHashes, OtaHashes, Sha256Hash},
+};
+
+const DOWNLOAD_TASKS: usize = 4;
+const DOWNLOAD_RETRIES: u8 = 3;
+const DOWNLOAD_PROGRESS_INTERVAL: Duration = Duration::from_millis(50);
+
+/// Sort and merge overlapping intervals.
+fn merge_overlapping(sections: &[Range<u64>]) -> Vec<Range<u64>> {
+    let mut sections = sections.to_vec();
+    sections.sort_by_key(|r| (r.start, r.end));
+
+    let mut result = Vec::<Range<u64>>::new();
+
+    for section in sections {
+        if let Some(last) = result.last_mut() {
+            if section.start <= last.end {
+                last.end = section.end;
+                continue;
+            }
+        }
+
+        result.push(section);
+    }
+
+    result
+}
+
+/// Convert an exclusion list into an inclusion list in the range [start, end).
+fn exclusion_to_inclusion(holes: &[Range<u64>], file_range: Range<u64>) -> Result<Vec<Range<u64>>> {
+    let exclusions = merge_overlapping(holes);
+
+    if let (Some(first), Some(last)) = (exclusions.first(), exclusions.last()) {
+        if first.start < file_range.start || last.end > file_range.end {
+            bail!("Sections are outside of the range {file_range:?}");
+        }
+    }
+
+    let flattened = exclusions.iter().flat_map(|p| [p.start, p.end]);
+    let points = [file_range.start]
+        .into_iter()
+        .chain(flattened)
+        .chain([file_range.end])
+        .collect::<Vec<_>>();
+
+    Ok(points.chunks_exact(2).map(|c| c[0]..c[1]).collect())
+}
+
+/// Convert a full OTA to a stripped OTA with all non-AVB-related partitions
+/// removed from the payload. No headers are updated, so the output file will
+/// have invalid hashes and signatures.
+///
+/// Returns the list of file sections and the sha256 digest.
+fn strip_image(
+    input: &Path,
+    output: &Path,
+    cancel_signal: &Arc<AtomicBool>,
+) -> Result<(Vec<Range<u64>>, [u8; 32])> {
+    println!("Stripping {input:?} to {output:?}");
+
+    let mut raw_reader = File::open(input)
+        .map(PSeekFile::new)
+        .with_context(|| anyhow!("Failed to open for reading: {input:?}"))?;
+    let mut zip_reader = ZipArchive::new(BufReader::new(raw_reader.clone()))
+        .with_context(|| anyhow!("Failed to read zip: {input:?}"))?;
+    let payload_entry = zip_reader
+        .by_name(ota::PATH_PAYLOAD)
+        .with_context(|| anyhow!("Failed to open zip entry: {:?}", ota::PATH_PAYLOAD))?;
+    let payload_offset = payload_entry.data_start();
+    let payload_size = payload_entry.size();
+
+    // Open the payload data directly.
+    let mut payload_reader = SectionReader::new(
+        BufReader::new(raw_reader.clone()),
+        payload_offset,
+        payload_size,
+    )?;
+
+    let header = PayloadHeader::from_reader(&mut payload_reader)
+        .with_context(|| anyhow!("Failed to load OTA payload header"))?;
+
+    let required_images =
+        avbroot::cli::ota::get_required_images(&header.manifest, "@gki_ramdisk", true)?
+            .into_values()
+            .collect::<HashSet<_>>();
+    let mut data_holes = vec![];
+
+    use avbroot::protobuf::chromeos_update_engine::mod_InstallOperation::Type;
+
+    for p in &header.manifest.partitions {
+        if !required_images.contains(&p.partition_name) {
+            for op in &p.operations {
+                match op.type_pb {
+                    Type::ZERO | Type::DISCARD => continue,
+                    _ => {
+                        let start = payload_offset
+                            + header.blob_offset
+                            + op.data_offset.expect("Missing data_offset");
+                        let end = start + op.data_length.expect("Missing data_length");
+
+                        data_holes.push(start..end);
+                    }
+                }
+            }
+        }
+    }
+
+    // Keep all sections outside of the partitions skipped.
+    let file_size = raw_reader.seek(SeekFrom::End(0))?;
+    let sections_to_keep = exclusion_to_inclusion(&data_holes, 0..file_size)?;
+
+    let mut context = ring::digest::Context::new(&ring::digest::SHA256);
+    let raw_writer =
+        File::create(output).with_context(|| anyhow!("Failed to open for writing: {output:?}"))?;
+    raw_writer
+        .set_len(file_size)
+        .with_context(|| anyhow!("Failed to set file size: {output:?}"))?;
+    let mut buf_writer = BufWriter::new(raw_writer);
+    let mut buf_reader = BufReader::new(raw_reader);
+
+    buf_reader.rewind()?;
+
+    for section in &sections_to_keep {
+        let offset = buf_reader.stream_position()?;
+
+        // Hash holes as zeros.
+        if offset != section.start {
+            stream::copy_n_inspect(
+                io::repeat(0),
+                io::sink(),
+                section.start - offset,
+                |data| context.update(data),
+                cancel_signal,
+            )?;
+
+            buf_reader.seek(SeekFrom::Start(section.start))?;
+            buf_writer.seek(SeekFrom::Start(section.start))?;
+        }
+
+        stream::copy_n_inspect(
+            &mut buf_reader,
+            &mut buf_writer,
+            section.end - section.start,
+            |data| context.update(data),
+            cancel_signal,
+        )?;
+    }
+
+    // There can't be a hole at the end of a zip, so nothing left to hash.
+
+    let digest = context.finish();
+    Ok((sections_to_keep, digest.as_ref().try_into().unwrap()))
+}
+
+fn url_filename(url: &str) -> Result<&str> {
+    url.rsplit_once('/')
+        .map(|(_, name)| name)
+        .ok_or_else(|| anyhow!("Failed to determine filename from URL: {url}"))
+}
+
+fn hash_file(path: &Path, cancel_signal: &Arc<AtomicBool>) -> Result<[u8; 32]> {
+    println!("Calculating hash of {path:?}");
+
+    let raw_reader =
+        File::open(path).with_context(|| anyhow!("Failed to open for reading: {path:?}"))?;
+    let buf_reader = BufReader::new(raw_reader);
+    let context = ring::digest::Context::new(&ring::digest::SHA256);
+    let mut hashing_reader = HashingReader::new(buf_reader, context);
+
+    stream::copy(&mut hashing_reader, io::sink(), cancel_signal)?;
+
+    let (_, context) = hashing_reader.finish();
+    let digest = context.finish();
+
+    Ok(digest.as_ref().try_into().unwrap())
+}
+
+fn verify_hash(path: &Path, sha256: &[u8; 32], cancel_signal: &Arc<AtomicBool>) -> Result<()> {
+    let digest = hash_file(path, cancel_signal)?;
+
+    if sha256 != digest.as_ref() {
+        bail!(
+            "Expected sha256 {}, but have {}: {path:?}",
+            hex::encode(sha256),
+            hex::encode(digest),
+        );
+    }
+
+    Ok(())
+}
+
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+enum Validate {
+    Always,
+    IfNew,
+    Never,
+}
+
+fn download_file(
+    path: &Path,
+    url: &str,
+    sha256: &[u8; 32],
+    sections: Option<&[Range<u64>]>,
+    path_is_dir: bool,
+    validate: Validate,
+    cancel_signal: &Arc<AtomicBool>,
+) -> Result<PathBuf> {
+    let path = if path_is_dir {
+        path.join(url_filename(url)?)
+    } else {
+        path.to_owned()
+    };
+
+    if let Some(parent) = path.parent() {
+        fs::create_dir_all(parent)
+            .with_context(|| anyhow!("Failed to create directory: {parent:?}"))?;
+    }
+
+    let mut do_validate = validate != Validate::Never;
+
+    if path.exists() && !download::state_path(&path).exists() {
+        if validate == Validate::IfNew {
+            do_validate = false;
+        }
+    } else {
+        println!("Downloading {url} to {path:?}");
+
+        let mut display = download::BasicProgressDisplay::new(DOWNLOAD_PROGRESS_INTERVAL);
+
+        download::download(
+            url,
+            &path,
+            sections,
+            &mut display,
+            DOWNLOAD_TASKS,
+            DOWNLOAD_RETRIES,
+        )?;
+    }
+
+    if do_validate {
+        verify_hash(&path, sha256, cancel_signal)?;
+    }
+
+    Ok(path)
+}
+
+fn download_magisk(
+    config: &Config,
+    work_dir: &Path,
+    revalidate: bool,
+    cancel_signal: &Arc<AtomicBool>,
+) -> Result<PathBuf> {
+    download_file(
+        &work_dir.join("magisk"),
+        &config.magisk.url,
+        &config.magisk.hash.0,
+        None,
+        true,
+        if revalidate {
+            Validate::Always
+        } else {
+            Validate::IfNew
+        },
+        cancel_signal,
+    )
+}
+
+fn download_image(
+    config: &Config,
+    device: &str,
+    work_dir: &Path,
+    stripped: bool,
+    revalidate: bool,
+    cancel_signal: &Arc<AtomicBool>,
+) -> Result<PathBuf> {
+    let info = &config.device[device];
+    let mut path = work_dir.join(device);
+    path.push(url_filename(&info.url)?);
+    let mut sha256 = &info.hash.original.full.0;
+    let mut sections = None;
+
+    if stripped {
+        path.as_mut_os_string().push(".stripped");
+        sha256 = &info.hash.original.stripped.0;
+        sections = Some(info.sections.as_slice());
+    }
+
+    download_file(
+        &path,
+        &info.url,
+        sha256,
+        sections,
+        false,
+        if revalidate {
+            Validate::Always
+        } else {
+            Validate::IfNew
+        },
+        cancel_signal,
+    )
+}
+
+#[rustfmt::skip]
+fn test_keys() -> Result<(TempDir, Vec<OsString>, Vec<OsString>)> {
+    let avb_key = include_bytes!(concat!(
+        env!("CARGO_MANIFEST_DIR"),
+        "/keys/TEST_KEY_DO_NOT_USE_avb.key",
+    ));
+    let avb_pass = include_bytes!(concat!(
+        env!("CARGO_MANIFEST_DIR"),
+        "/keys/TEST_KEY_DO_NOT_USE_avb.passphrase",
+    ));
+    let avb_pkmd = include_bytes!(concat!(
+        env!("CARGO_MANIFEST_DIR"),
+        "/keys/TEST_KEY_DO_NOT_USE_avb_pkmd.bin",
+    ));
+    let ota_key = include_bytes!(concat!(
+        env!("CARGO_MANIFEST_DIR"),
+        "/keys/TEST_KEY_DO_NOT_USE_ota.key",
+    ));
+    let ota_pass = include_bytes!(concat!(
+        env!("CARGO_MANIFEST_DIR"),
+        "/keys/TEST_KEY_DO_NOT_USE_ota.passphrase",
+    ));
+    let ota_cert = include_bytes!(concat!(
+        env!("CARGO_MANIFEST_DIR"),
+        "/keys/TEST_KEY_DO_NOT_USE_ota.crt",
+    ));
+
+    let temp_dir = TempDir::new().context("Failed to create temporary directory for test keys")?;
+    let mut patch_args = Vec::<OsString>::new();
+    let mut verify_args = Vec::<OsString>::new();
+
+    for (name, data, patch_arg, verify_arg) in [
+        ("avb.key", &avb_key[..], Some("--key-avb"), None),
+        ("avb.pass", &avb_pass[..], Some("--pass-avb-file"), None),
+        ("avb.pkmd", &avb_pkmd[..], None, Some("--public-key-avb")),
+        ("ota.key", &ota_key[..], Some("--key-ota"), None),
+        ("ota.pass", &ota_pass[..], Some("--pass-ota-file"), None),
+        ("ota.crt", &ota_cert[..], Some("--cert-ota"), Some("--cert-ota")),
+    ] {
+        let path = temp_dir.path().join(name);
+        fs::write(&path, data).with_context(|| anyhow!("Failed to write test key: {path:?}"))?;
+
+        if let Some(arg) = patch_arg {
+            patch_args.push(arg.into());
+            patch_args.push(path.as_os_str().to_owned());
+        }
+        if let Some(arg) = verify_arg {
+            verify_args.push(arg.into());
+            verify_args.push(path.as_os_str().to_owned());
+        }
+    }
+
+    Ok((temp_dir, patch_args, verify_args))
+}
+
+fn patch_image(
+    input_file: &Path,
+    output_file: &Path,
+    extra_args: &[OsString],
+    cancel_signal: &Arc<AtomicBool>,
+) -> Result<()> {
+    println!("Patching {input_file:?}");
+
+    let (_temp_key_dir, key_args, _) = test_keys()?;
+
+    // We're intentionally using the CLI interface.
+    let mut args: Vec<OsString> = vec![
+        "patch".into(),
+        "--input".into(),
+        input_file.as_os_str().into(),
+        "--output".into(),
+        output_file.as_os_str().into(),
+    ];
+    args.extend(key_args);
+    args.extend_from_slice(extra_args);
+
+    if args.contains(&OsStr::new("--magisk").into()) {
+        // This doesn't need to be correct. The test outputs aren't meant to
+        // be booted on real devices.
+        args.push("--magisk-preinit-device".into());
+        args.push("metadata".into());
+    }
+
+    let cli = PatchCli::try_parse_from(args)?;
+    avbroot::cli::ota::patch_subcommand(&cli, cancel_signal)?;
+
+    Ok(())
+}
+
+fn extract_image(
+    input_file: &Path,
+    output_dir: &Path,
+    cancel_signal: &Arc<AtomicBool>,
+) -> Result<()> {
+    println!("Extracting AVB partitions from {input_file:?}");
+
+    let cli = ExtractCli::try_parse_from([
+        OsStr::new("extract"),
+        OsStr::new("--input"),
+        input_file.as_os_str(),
+        OsStr::new("--directory"),
+        output_dir.as_os_str(),
+    ])?;
+    avbroot::cli::ota::extract_subcommand(&cli, cancel_signal)?;
+
+    Ok(())
+}
+
+fn verify_image(input_file: &Path, cancel_signal: &Arc<AtomicBool>) -> Result<()> {
+    println!("Verifying signatures in {input_file:?}");
+
+    let (_temp_key_dir, _, key_args) = test_keys()?;
+
+    let mut args: Vec<OsString> = vec![
+        "verify".into(),
+        "--input".into(),
+        input_file.as_os_str().into(),
+    ];
+    args.extend(key_args);
+
+    let cli = VerifyCli::try_parse_from(args)?;
+    avbroot::cli::ota::verify_subcommand(&cli, cancel_signal)?;
+
+    Ok(())
+}
+
+fn get_magisk_partition(path: &Path) -> Result<String> {
+    let raw_reader =
+        File::open(path).with_context(|| anyhow!("Failed to open for reading: {path:?}"))?;
+    let mut zip = ZipArchive::new(BufReader::new(raw_reader))
+        .with_context(|| anyhow!("Failed to read zip: {path:?}"))?;
+    let payload_entry = zip
+        .by_name(ota::PATH_PAYLOAD)
+        .with_context(|| anyhow!("Failed to open zip entry: {:?}", ota::PATH_PAYLOAD))?;
+    let payload_offset = payload_entry.data_start();
+    let payload_size = payload_entry.size();
+
+    drop(payload_entry);
+    let buf_reader = zip.into_inner();
+
+    // Open the payload data directly.
+    let mut payload_reader = SectionReader::new(buf_reader, payload_offset, payload_size)?;
+
+    let header = PayloadHeader::from_reader(&mut payload_reader)
+        .with_context(|| anyhow!("Failed to load OTA payload header"))?;
+    let images = avbroot::cli::ota::get_partitions_by_type(&header.manifest)?;
+
+    Ok(images["@gki_ramdisk"].clone())
+}
+
+fn filter_devices<'a>(config: &'a Config, cli: &'a DeviceGroup) -> Result<BTreeSet<&'a str>> {
+    let mut devices = config
+        .device
+        .keys()
+        .map(|d| d.as_str())
+        .collect::<BTreeSet<_>>();
+
+    if !cli.all {
+        let invalid = cli
+            .device
+            .iter()
+            .filter(|d| !devices.contains(d.as_str()))
+            .collect::<BTreeSet<_>>();
+        if !invalid.is_empty() {
+            bail!("Invalid devices: {invalid:?}");
+        }
+
+        devices = cli.device.iter().map(|d| d.as_str()).collect();
+    }
+
+    Ok(devices)
+}
+
+fn strip_subcommand(cli: &StripCli, cancel_signal: &Arc<AtomicBool>) -> Result<()> {
+    let (sections, sha256) = strip_image(&cli.input, &cli.output, cancel_signal)?;
+
+    println!("Preserved sections:");
+    for section in sections {
+        println!("- {section:?}");
+    }
+
+    println!("SHA256: {}", hex::encode(sha256));
+
+    Ok(())
+}
+
+fn add_subcommand(cli: &AddCli, cancel_signal: &Arc<AtomicBool>) -> Result<()> {
+    let (config, mut document) = config::load_config(&cli.config.config)?;
+
+    let image_dir = cli.config.work_dir.join(&cli.device);
+
+    let full_ota = image_dir.join(url_filename(&cli.url)?);
+    let mut full_ota_patched = full_ota.clone();
+    full_ota_patched.as_mut_os_string().push(&cli.patch.suffix);
+    let mut stripped_ota = full_ota.clone();
+    stripped_ota.as_mut_os_string().push(".stripped");
+    let mut stripped_ota_patched = stripped_ota.clone();
+    stripped_ota_patched
+        .as_mut_os_string()
+        .push(&cli.patch.suffix);
+
+    let full_ota_hash = cli.hash.as_ref().map(|h| h.0);
+
+    download_file(
+        &full_ota,
+        &cli.url,
+        &full_ota_hash.unwrap_or_default(),
+        None,
+        false,
+        if full_ota_hash.is_some() {
+            Validate::Always
+        } else {
+            Validate::Never
+        },
+        cancel_signal,
+    )?;
+
+    // Calculate the hash ourselves if one wasn't provided.
+    let full_ota_hash = match full_ota_hash {
+        Some(h) => h,
+        None => hash_file(&full_ota, cancel_signal)?,
+    };
+
+    let magisk_file = download_magisk(&config, &cli.config.work_dir, true, cancel_signal)?;
+    let magisk_args = [OsString::from("--magisk"), magisk_file.into_os_string()];
+
+    // Patch the full image.
+    patch_image(&full_ota, &full_ota_patched, &magisk_args, cancel_signal)?;
+    let full_ota_patched_hash = hash_file(&full_ota_patched, cancel_signal)?;
+
+    // Check that the patched full image looks good.
+    if cli.skip_verify {
+        println!("OTA and AVB signature validation skipped");
+    } else {
+        verify_image(&full_ota_patched, cancel_signal)?;
+    }
+
+    // Strip the full image.
+    let (sections, stripped_ota_hash) = strip_image(&full_ota, &stripped_ota, cancel_signal)?;
+
+    // Patch the stripped image. This doesn't fail zip's CRC checks because the
+    // `ota patch` commands reads the payload directly from the raw backing
+    // file.
+    patch_image(
+        &stripped_ota,
+        &stripped_ota_patched,
+        &magisk_args,
+        cancel_signal,
+    )?;
+    let stripped_ota_patched_hash = hash_file(&stripped_ota_patched, cancel_signal)?;
+
+    // Hash all of the AVB-related partition images so that `e2e test` can fail
+    // fast if something goes wrong.
+    let mut avb_images = BTreeMap::<String, Sha256Hash>::new();
+
+    {
+        let temp_dir = TempDir::new().context("Failed to create temp directory")?;
+        extract_image(&full_ota_patched, temp_dir.path(), cancel_signal)?;
+
+        for entry in fs::read_dir(temp_dir.path())? {
+            let entry = entry?;
+            let hash = hash_file(&entry.path(), cancel_signal)?;
+
+            avb_images.insert(entry.file_name().into_string().unwrap(), Sha256Hash(hash));
+        }
+    }
+
+    println!("Adding {} to config file", cli.device);
+
+    let device = Device {
+        url: cli.url.clone(),
+        sections,
+        hash: ImageHashes {
+            original: OtaHashes {
+                full: Sha256Hash(full_ota_hash),
+                stripped: Sha256Hash(stripped_ota_hash),
+            },
+            patched: OtaHashes {
+                full: Sha256Hash(full_ota_patched_hash),
+                stripped: Sha256Hash(stripped_ota_patched_hash),
+            },
+            avb_images,
+        },
+    };
+
+    config::add_device(&mut document, &cli.device, &device)?;
+
+    let config_serialized = document.to_string();
+    fs::write(&cli.config.config, config_serialized)
+        .with_context(|| anyhow!("Failed to write config: {:?}", cli.config.config))?;
+
+    if cli.patch.delete_on_success {
+        for path in [full_ota_patched, stripped_ota_patched] {
+            fs::remove_file(&path).with_context(|| anyhow!("Failed to delete file: {path:?}"))?;
+        }
+    }
+
+    Ok(())
+}
+
+fn download_subcommand(cli: &DownloadCli, cancel_signal: &Arc<AtomicBool>) -> Result<()> {
+    let (config, _) = config::load_config(&cli.config.config)?;
+    let devices = filter_devices(&config, &cli.device)?;
+
+    if !cli.magisk && devices.is_empty() {
+        bail!("No downloads selected");
+    }
+
+    if cli.magisk {
+        download_magisk(
+            &config,
+            &cli.config.work_dir,
+            cli.download.revalidate,
+            cancel_signal,
+        )?;
+    }
+
+    for device in devices {
+        download_image(
+            &config,
+            device,
+            &cli.config.work_dir,
+            cli.download.stripped,
+            cli.download.revalidate,
+            cancel_signal,
+        )?;
+    }
+
+    Ok(())
+}
+
+fn test_subcommand(cli: &TestCli, cancel_signal: &Arc<AtomicBool>) -> Result<()> {
+    let (config, _) = config::load_config(&cli.config.config)?;
+    let devices = filter_devices(&config, &cli.device)?;
+
+    if devices.is_empty() {
+        bail!("No devices selected");
+    }
+
+    let magisk_file = download_magisk(
+        &config,
+        &cli.config.work_dir,
+        cli.download.revalidate,
+        cancel_signal,
+    )?;
+    let magisk_args = [OsString::from("--magisk"), magisk_file.into_os_string()];
+
+    for device in devices {
+        let info = &config.device[device];
+
+        let image_file = download_image(
+            &config,
+            device,
+            &cli.config.work_dir,
+            cli.download.stripped,
+            cli.download.revalidate,
+            cancel_signal,
+        )?;
+
+        let mut patched_file = image_file.clone();
+        patched_file.as_mut_os_string().push(&cli.patch.suffix);
+
+        let patched_hash = if cli.download.stripped {
+            &info.hash.patched.stripped.0
+        } else {
+            &info.hash.patched.full.0
+        };
+
+        patch_image(&image_file, &patched_file, &magisk_args, cancel_signal)?;
+
+        let temp_dir = TempDir::new().context("Failed to create temp directory")?;
+
+        // Check partitions first so we fail fast if the issue is with AVB.
+        extract_image(&patched_file, temp_dir.path(), cancel_signal)?;
+
+        let mut expected = info.hash.avb_images.keys().collect::<BTreeSet<_>>();
+
+        for entry in fs::read_dir(temp_dir.path())? {
+            let entry = entry?;
+            let name = entry.file_name().into_string().unwrap();
+            let hash = info
+                .hash
+                .avb_images
+                .get(&name)
+                .ok_or_else(|| anyhow!("Missing AVB image hash for {name}"))?;
+
+            verify_hash(&entry.path(), &hash.0, cancel_signal)?;
+            expected.remove(&name);
+        }
+
+        if !expected.is_empty() {
+            bail!("Missing AVB images: {expected:?}");
+        }
+
+        // Then, validate the hash of everything.
+        verify_hash(&patched_file, patched_hash, cancel_signal)?;
+
+        // Patch again, but this time, use the previously patched boot image
+        // instead of applying the Magisk patch.
+        let magisk_partition = get_magisk_partition(&patched_file)?;
+        let prepatched_args = [
+            OsStr::new("--prepatched").to_owned(),
+            temp_dir
+                .path()
+                .join(format!("{magisk_partition}.img"))
+                .into_os_string(),
+        ];
+
+        fs::remove_file(&patched_file)
+            .with_context(|| anyhow!("Failed to delete file: {patched_file:?}"))?;
+
+        patch_image(&image_file, &patched_file, &prepatched_args, cancel_signal)?;
+
+        verify_hash(&patched_file, patched_hash, cancel_signal)?;
+
+        if cli.patch.delete_on_success {
+            fs::remove_file(&patched_file)
+                .with_context(|| anyhow!("Failed to delete file: {patched_file:?}"))?;
+        }
+    }
+
+    Ok(())
+}
+
+fn list_subcommand(cli: &ListCli) -> Result<()> {
+    let (config, _) = config::load_config(&cli.config.config)?;
+
+    for device in config.device.keys() {
+        println!("{device}");
+    }
+
+    Ok(())
+}
+
+fn main() -> Result<()> {
+    // Set up a cancel signal so we can properly clean up any temporary files.
+    let cancel_signal = Arc::new(AtomicBool::new(false));
+    {
+        let signal = cancel_signal.clone();
+
+        ctrlc::set_handler(move || {
+            signal.store(true, Ordering::SeqCst);
+        })
+        .expect("Failed to set signal handler");
+    }
+
+    let cli = Cli::parse();
+
+    match cli.command {
+        Command::Strip(c) => strip_subcommand(&c, &cancel_signal),
+        Command::Add(c) => add_subcommand(&c, &cancel_signal),
+        Command::Download(c) => download_subcommand(&c, &cancel_signal),
+        Command::Test(c) => test_subcommand(&c, &cancel_signal),
+        Command::List(c) => list_subcommand(&c),
+    }
+}

extra/README.md

@@ -1,62 +0,0 @@
-# avbroot extra
-
-This directory contains some extra scripts that aren't required for avbroot's operation, but may be useful for troubleshooting.
-
-## `bootimagetool`
-
-This is a frontend to the [`avbroot/formats/bootimage.py`](../avbroot/formats/bootimage.py) library for working with boot images.
-
-### Unpacking a boot image
-
-```bash
-python bootimagetool.py unpack <input boot image>
-```
-
-This subcommand unpacks all of the components of the boot image into the current directory by default (see `--help`). The header fields are saved to `header.json` and each blob section is saved to a separate file. Each blob is written to disk as-is, without decompression.
-
-### Packing a boot image
-
-```bash
-python bootimagetool.py pack <output boot image>
-```
-
-This subcommand packs a new boot image from the individual components in the current directory by default (see `--help`). The default input filenames are the same as the output filenames for the `unpack` subcommand.
-
-### Repacking a boot image
-
-```bash
-python bootimagetool.py repack <input boot image> <output boot image>
-```
-
-This subcommand repacks a boot image without writing the individual components to disk first. This is useful for roundtrip testing of avbroot's boot image parser. The output should be identical to the input, minus any footers, like the AVB footer. The only exception is the VTS signature for v4 boot images, which is always stripped out.
-
-## `cpiotool`
-
-This is a frontend to the [`avbroot/formats/compression.py`](../avbroot/formats/compression.py) and [`avbroot/formats/cpio.py`](../avbroot/formats/cpio.py) libraries. It is useful for inspecting compressed and uncompressed cpio archives.
-
-### Dumping a cpio archive
-
-```bash
-python cpiotool.py dump <cpio archive>
-```
-
-This subcommand dumps all information about a cpio archive to stdout. This includes the compression format, all header fields (including the trailer entry), and all data. If an entry's data can be decoded as UTF-8, then it is printed out as text. Otherwise, the binary data is printed out base64-encoded. The base64-encoded data is truncated to 5 lines by default to avoid outputting too much data, but this behavior can be disabled with `--no-truncate`.
-
-### Repacking a cpio archive
-
-```bash
-python cpiotool.py repack <input cpio archive> <output cpio archive>
-```
-
-This subcommand repacks a cpio archive, including recompression if needed. This is useful for roundtrip testing of avbroot's cpio parser and compression handling. The uncompressed output should be identical to the uncompressed input, except:
-
-* files are sorted by name
-* inodes are reassigned, starting from 300000
-* there is no excess padding at the end of the file
-
-The compressed output may differ from what other tools produce because:
-
-* LZ4 legacy chunks are packed to exactly 8 MiB, except for the last chunk, which may be smaller.
-* LZ4 legacy uses the high compression mode with a compression level of 12.
-* The GZIP header has the modification timestamp set to 0 (Unix epoch time).
-* GZIP uses a compression level of 9.

extra/bootimagetool.py

@@ -1,168 +0,0 @@
-#!/usr/bin/env python3
-
-import argparse
-import itertools
-import json
-import os
-import sys
-
-sys.path.append(os.path.join(sys.path[0], '..'))
-from avbroot.formats import bootimage
-
-
-class BytesDecoder(json.JSONDecoder):
-    def __init__(self):
-        super().__init__(object_hook=self.from_dict)
-
-    @staticmethod
-    def from_dict(d):
-        # This is insufficient for arbitrary data, but we're not dealing with
-        # arbitrary data
-        if 'type' in d:
-            if d['type'] == 'UTF-8':
-                return d['data'].encode('UTF-8')
-            elif d['type'] == 'hex':
-                return bytes.fromhex(d['data'])
-
-        return d
-
-
-class BytesEncoder(json.JSONEncoder):
-    def default(self, obj):
-        if isinstance(obj, bytes):
-            if b'\0' not in obj:
-                try:
-                    return {
-                        'type': 'UTF-8',
-                        'data': obj.decode('UTF-8'),
-                    }
-                except UnicodeDecodeError:
-                    pass
-
-            return {
-                'type': 'hex',
-                'data': obj.hex(),
-            }
-
-        return super().default(obj)
-
-
-def read_or_none(path):
-    try:
-        with open(path, 'rb') as f:
-            return f.read()
-    except FileNotFoundError:
-        return None
-
-
-def write_if_not_none(path, data):
-    if data is not None:
-        with open(path, 'wb') as f:
-            f.write(data)
-
-
-def parse_args():
-    parser_kwargs = {'formatter_class': argparse.ArgumentDefaultsHelpFormatter}
-
-    parser = argparse.ArgumentParser(**parser_kwargs)
-    subparsers = parser.add_subparsers(dest='subcommand', required=True,
-                                       help='Subcommands')
-
-    base = argparse.ArgumentParser(add_help=False)
-    base.add_argument('-q', '--quiet', action='store_true',
-                      help='Do not print header information')
-
-    pack = subparsers.add_parser('pack', help='Pack a boot image',
-                                 parents=[base], **parser_kwargs)
-    unpack = subparsers.add_parser('unpack', help='Unpack a boot image',
-                                   parents=[base], **parser_kwargs)
-    repack = subparsers.add_parser('repack', help='Repack a boot image',
-                                   parents=[base], **parser_kwargs)
-
-    for p in (pack, unpack):
-        prefix = '--input-' if p == pack else '--output-'
-
-        p.add_argument('boot_image', help='Path to boot image')
-
-        p.add_argument(prefix + 'header', default='header.json',
-                       help='Path to header JSON')
-        p.add_argument(prefix + 'kernel', default='kernel.img',
-                       help='Path to kernel')
-        p.add_argument(prefix + 'ramdisk-prefix', default='ramdisk.img.',
-                       help='Path prefix for ramdisk')
-        p.add_argument(prefix + 'second', default='second.img',
-                       help='Path to second stage bootloader')
-        p.add_argument(prefix + 'recovery-dtbo', default='recovery_dtbo.img',
-                       help='Path to recovery dtbo/acpio')
-        p.add_argument(prefix + 'dtb', default='dtb.img',
-                       help='Path to device tree blob')
-        p.add_argument(prefix + 'bootconfig', default='bootconfig.txt',
-                       help='Path to bootconfig')
-
-    repack.add_argument('input', help='Path to input boot image')
-    repack.add_argument('output', help='Path to output boot image')
-
-    return parser.parse_args()
-
-
-def main():
-    args = parse_args()
-
-    if args.subcommand == 'pack':
-        with open(args.input_header, 'r') as f:
-            data = json.load(f, cls=BytesDecoder)
-
-        img = bootimage.create_from_dict(data)
-
-        img.kernel = read_or_none(args.input_kernel)
-        img.second = read_or_none(args.input_second)
-        img.recovery_dtbo = read_or_none(args.input_recovery_dtbo)
-        img.dtb = read_or_none(args.input_dtb)
-        img.bootconfig = read_or_none(args.input_bootconfig)
-
-        for i in itertools.count():
-            ramdisk = read_or_none(f'{args.input_ramdisk_prefix}{i}')
-            if ramdisk is None:
-                break
-
-            img.ramdisks.append(ramdisk)
-
-        if not args.quiet:
-            print(img)
-
-        with open(args.boot_image, 'wb') as f:
-            img.generate(f)
-
-    elif args.subcommand == 'unpack':
-        with open(args.boot_image, 'rb') as f:
-            img = bootimage.load_autodetect(f)
-            if not args.quiet:
-                print(img)
-
-        with open(args.output_header, 'w') as f:
-            json.dump(img.to_dict(), f, indent=4, cls=BytesEncoder)
-
-        write_if_not_none(args.output_kernel, img.kernel)
-        write_if_not_none(args.output_second, img.second)
-        write_if_not_none(args.output_recovery_dtbo, img.recovery_dtbo)
-        write_if_not_none(args.output_dtb, img.dtb)
-        write_if_not_none(args.output_bootconfig, img.bootconfig)
-
-        for i, ramdisk in enumerate(img.ramdisks):
-            write_if_not_none(f'{args.output_ramdisk_prefix}{i}', ramdisk)
-
-    elif args.subcommand == 'repack':
-        with open(args.input, 'rb') as f:
-            img = bootimage.load_autodetect(f)
-            if not args.quiet:
-                print(img)
-
-        with open(args.output, 'wb') as f:
-            img.generate(f)
-
-    else:
-        raise NotImplementedError()
-
-
-if __name__ == '__main__':
-    main()

extra/cpiotool.py

@@ -1,120 +0,0 @@
-#!/usr/bin/env python3
-
-import argparse
-import base64
-import os
-import sys
-
-sys.path.append(os.path.join(sys.path[0], '..'))
-from avbroot.formats import compression
-from avbroot.formats import cpio
-
-
-CONTENT_BEGIN = '----- BEGIN UTF-8 CONTENT -----'
-CONTENT_END = '----- END UTF-8 CONTENT -----'
-CONTENT_END_NO_NEWLINE = '----- END UTF-8 CONTENT (NO NEWLINE) -----'
-
-BASE64_BEGIN = '----- BEGIN BASE64 CONTENT -----'
-BASE64_END = '----- END BASE64 CONTENT -----'
-BASE64_END_TRUNCATED = '----- END BASE64 CONTENT (TRUNCATED) -----'
-
-NO_DATA = '----- NO DATA -----'
-
-
-def print_content(data, truncate=False):
-    if not data:
-        print(NO_DATA)
-        return
-
-    if b'\0' not in data:
-        try:
-            data_str = data.decode('UTF-8')
-
-            if CONTENT_BEGIN not in data_str \
-                    and CONTENT_END not in data_str \
-                    and CONTENT_END_NO_NEWLINE not in data_str:
-                print(CONTENT_BEGIN)
-                print(data_str, end='')
-                if data_str[-1] != '\n':
-                    print()
-                    print(CONTENT_END_NO_NEWLINE)
-                else:
-                    print(CONTENT_END)
-
-            return
-        except UnicodeDecodeError:
-            pass
-
-    data_base64 = base64.b64encode(data).decode('ascii')
-
-    print(BASE64_BEGIN)
-    for i, offset in enumerate(range(0, len(data_base64), 76)):
-        if truncate and i == 5:
-            print(BASE64_END_TRUNCATED)
-            return
-
-        print(data_base64[offset:offset + 76])
-    print(BASE64_END)
-
-
-def parse_args():
-    parser = argparse.ArgumentParser()
-    subparsers = parser.add_subparsers(dest='subcommand', required=True,
-                                       help='Subcommands')
-
-    dump = subparsers.add_parser('dump', help='Dump cpio headers and data')
-    repack = subparsers.add_parser('repack', help='Repack cpio archive')
-
-    dump.add_argument('--no-truncate', action='store_true',
-                      help='Do not truncate binary file contents')
-
-    for p in (dump, repack):
-        p.add_argument('input', help='Path to input cpio file')
-
-    repack.add_argument('output', help='Path to output cpio file')
-
-    return parser.parse_args()
-
-
-def load_archive(path, **cpio_kwargs):
-    with open(path, 'rb') as f_raw:
-        with compression.CompressedFile(f_raw, 'rb', raw_if_unknown=True) as f:
-            return cpio.load(f.fp, **cpio_kwargs), f.format
-
-
-def save_archive(path, entries, format):
-    with open(path, 'wb') as f_raw:
-        with compression.CompressedFile(f_raw, 'wb', format=format,
-                                        raw_if_unknown=True) as f:
-            cpio.save(f.fp, entries)
-
-
-def main():
-    args = parse_args()
-
-    if args.subcommand == 'dump':
-        entries, format = load_archive(
-            args.input,
-            # We want to show the headers exactly as they are
-            include_trailer=True,
-            reassign_inodes=False,
-        )
-
-        print('Compression format:', format)
-        print()
-
-        for entry in entries:
-            print(entry)
-            print_content(entry.content, truncate=not args.no_truncate)
-            print()
-
-    elif args.subcommand == 'repack':
-        entries, format = load_archive(args.input)
-        save_archive(args.output, entries, format)
-
-    else:
-        raise NotImplementedError()
-
-
-if __name__ == '__main__':
-    main()

modules/build.py

@@ -1,166 +0,0 @@
-#!/usr/bin/env python3
-
-import argparse
-import io
-import os
-import re
-import shutil
-import subprocess
-import sys
-import tempfile
-import zipfile
-
-
-def natsort_key(text, regex=re.compile(r'(\d+)')):
-    return [int(s) if s.isdigit() else s for s in regex.split(text)]
-
-
-def newest_child_by_name(directory):
-    children = os.listdir(directory)
-    if not children:
-        raise ValueError(f'{directory} has no children')
-
-    child = sorted(children, key=natsort_key)[-1]
-    return os.path.join(directory, child)
-
-
-def build_empty_zip():
-    stream = io.BytesIO()
-
-    with zipfile.ZipFile(stream, 'w'):
-        pass
-
-    return stream.getvalue()
-
-
-def build_dex(sources):
-    if 'ANDROID_HOME' not in os.environ:
-        raise ValueError('ANDROID_HOME must be set to the Android SDK path')
-
-    sdk = os.environ['ANDROID_HOME']
-    build_tools = newest_child_by_name(os.path.join(sdk, 'build-tools'))
-    platform = newest_child_by_name(os.path.join(sdk, 'platforms'))
-    d8 = os.path.join(build_tools, 'd8')
-    if os.name == 'nt':
-        d8 += '.bat'
-    android_jar = os.path.join(platform, 'android.jar')
-
-    with tempfile.TemporaryDirectory() as temp_dir:
-        subprocess.check_call([
-            'javac',
-            '-source', '1.8',
-            '-target', '1.8',
-            '-cp', android_jar,
-            '-d', temp_dir,
-            *sources,
-        ])
-
-        class_files = []
-        for root, _, files in os.walk(temp_dir):
-            for f in files:
-                if f.endswith('.class'):
-                    class_files.append(os.path.join(root, f))
-
-        subprocess.check_call([
-            d8,
-            '--output', temp_dir,
-            *class_files,
-        ])
-
-        with open(os.path.join(temp_dir, 'classes.dex'), 'rb') as f:
-            return f.read()
-
-
-def parse_props(raw_prop):
-    result = {}
-
-    for line in raw_prop.decode('UTF-8').splitlines():
-        k, delim, v = line.partition('=')
-        if not delim:
-            raise ValueError(f'Malformed line: {repr(line)}')
-
-        result[k.strip()] = v.strip()
-
-    return result
-
-
-def build_module(dist_dir, common_dir, module_dir, extra_files):
-    with open(os.path.join(module_dir, 'module.prop'), 'rb') as f:
-        module_prop_raw = f.read()
-        module_prop = parse_props(module_prop_raw)
-
-    name = module_prop['name']
-    version = module_prop['version'].removeprefix('v')
-    zip_path = os.path.join(dist_dir, f'{name}-{version}.zip')
-
-    with zipfile.ZipFile(zip_path, 'w') as z:
-        file_map = {
-            'META-INF/com/google/android/update-binary': {
-                'file': os.path.join(common_dir, 'update-binary'),
-            },
-            'META-INF/com/google/android/updater-script': {
-                'file': os.path.join(common_dir, 'updater-script'),
-            },
-            'module.prop': {
-                'data': module_prop_raw,
-            },
-            **extra_files,
-        }
-
-        for name, source in sorted(file_map.items()):
-            # Build our own ZipInfo to ensure archive is reproducible
-            info = zipfile.ZipInfo(name)
-            with z.open(info, 'w') as f_out:
-                if 'data' in source:
-                    f_out.write(source['data'])
-                else:
-                    with open(source['file'], 'rb') as f_in:
-                        shutil.copyfileobj(f_in, f_out)
-
-    return zip_path
-
-
-def parse_args():
-    parser = argparse.ArgumentParser()
-    parser.add_argument('module', nargs='*',
-                        default=('clearotacerts', 'oemunlockonboot'),
-                        help='Module to build')
-
-    return parser.parse_args()
-
-
-def main():
-    args = parse_args()
-
-    dist_dir = os.path.join(sys.path[0], 'dist')
-    os.makedirs(dist_dir, exist_ok=True)
-
-    common_dir = os.path.join(sys.path[0], 'common')
-
-    for module in args.module:
-        module_dir = os.path.join(sys.path[0], module)
-
-        if module == 'clearotacerts':
-            extra_files = {
-                'system/etc/security/otacerts.zip': {
-                    'data': build_empty_zip(),
-                },
-            }
-        elif module == 'oemunlockonboot':
-            extra_files = {
-                'classes.dex': {
-                    'data': build_dex([os.path.join(module_dir, 'Main.java')]),
-                },
-                'service.sh': {
-                    'file': os.path.join(module_dir, 'service.sh'),
-                },
-            }
-        else:
-            raise ValueError(f'Invalid module: {module}')
-
-        module_zip = build_module(dist_dir, common_dir, module_dir, extra_files)
-        print('Built module', module_zip)
-
-
-if __name__ == '__main__':
-    main()

modules/clearotacerts/module.prop

@@ -1,6 +1,6 @@
 id=com.chiller3.avbroot.clearotacerts
 name=clearotacerts
-version=v1.0
-versionCode=1
+version=v2.0.0
+versionCode=131072
 author=chenxiaolong
 description=Block A/B OTAs by clearing verification certificates

modules/oemunlockonboot/module.prop

@@ -1,6 +1,6 @@
 id=com.chiller3.avbroot.oemunlockonboot
 name=oemunlockonboot
-version=v1.0
-versionCode=1
+version=v2.0.0
+versionCode=131072
 author=chenxiaolong
 description=Enable OEM unlocking on every boot

requirements.txt

@@ -1,3 +0,0 @@
-lz4
-# The pregenerated AOSP Python source is for version 3
-protobuf<4

tests/README.md

@@ -1,84 +0,0 @@
-# avbroot tests
-
-avbroot's output files are reproducible for the given input files. [`tests.yaml`](./tests.yaml) lists some OTA images with unique properties and the expected checksums before and after patching. These tests use pregenerated, hardcoded test keys for signing. **These keys should NEVER be used for any other purpose.**
-
-For each image listed in the config, the test process will:
-
-1. Download the OTA if it doesn't already exist in `tests/files/<device>/` (or the workdir specified by `-w`)
-2. Verify the OTA checksum
-3. Run avbroot against the OTA using `--magisk`
-4. Extract the AVB-related partitions from the patched OTA and verify their checksums
-5. Verify the patched OTA checksum
-6. Run avbroot against the OTA again using `--prepatched`
-7. Verify the patched OTA checksum again
-
-For more efficient CI testing, the tests can operate on "stripped" OTAs. A stripped OTA is identical to the full OTA, except that partitions in `payload.bin` unrelated to AVB are zeroed out. This reduces the download size and disk space requirements by a couple orders of magnitude. **A stripped OTA is NOT bootable and should never be flashed on a real device.**
-
-## Running the tests
-
-To test against the device OTA images listed in [`tests.yaml`](./tests.yaml), run:
-
-```bash
-# To test all device OTAs
-python tests/tests.py test -a
-# Or to test against specific device OTAs
-python tests/tests.py test -d cheetah -d bluejay
-```
-
-To test against stripped OTAs (smaller download, but not bootable), pass in `--stripped`.
-
-## Running the tests in a container
-
-The tests can also be run inside a podman container for easy testing on various Linux distros. To do so, run:
-
-```bash
-python tests/tests_containerized.py
-```
-
-This will build all of the images defined in [`distros/Containerfile.<distro>`](./distros/) and run the tests inside new container instances concurrently. By default, the number of concurrent jobs is set to the number of CPUs. This can be changed with `-j <num>`.
-
-To only run tests against a specific set of distro images, use `-d <distro>`, which can be specified multiple times. All arguments after a `--` argument are passed to `tests.py` directly.
-
-For example, to test patching the `cheetah` OTA against the Fedora and Arch images, run:
-
-```bash
-python tests/tests_containerized.py -d fedora -d arch -- -d cheetah
-```
-
-## Downloading a device image
-
-To download a full OTA image, run:
-
-```bash
-python tests/tests.py download -d <device>
-```
-
-This normally happens automatically when running [`tests.py`](./tests.py). To download the stripped OTA image instead, pass in `--stripped`.
-
-If the image file does not already exist, then it will be downloaded and the checksums will be validated. If the download is interrupted, it will automatically resume when the command is rerun. If the file is already downloaded, the command is effectively a no-op unless `--revalidate` is passed in to revalidate the image checksums.
-
-## Adding a new device image
-
-To add a new device image to the testing configuration, run:
-
-```bash
-python tests/tests.py add -d <device> -u <full OTA URL> -H <expected checksum>
-```
-
-If the OS vendor does not provide a SHA-256 checksum, omit `-H` and the script will compute the checksum from the downloaded data.
-
-This process will download the full OTA, strip it, patch the full OTA, patch the stripped OTA, extract the AVB partitions, and write all of the checksums to [`tests.yaml`](./tests.yaml).
-
-The process for updating an existing device config is exactly the same as adding a new one.
-
-(Note: Due to how the strictyaml library handles comments, this might cause some comments in the config file to be removed. They'll need to be added back manually.)
-
-## Stripping a full OTA
-
-To convert a full OTA to the stripped form, run:
-
-```bash
-python tests/tests.py strip -i <input zip> -o <output zip>
-```
-
-This normally happens automatically as a part of adding a new device image.

tests/config.py

@@ -1,44 +0,0 @@
-import os
-
-from strictyaml import load, Int, Map, MapPattern, Regex, Seq, Str
-
-
-CONFIG_PATH = os.path.join(
-    os.path.realpath(os.path.dirname(__file__)), 'tests.yaml')
-
-SHA256_HEX = Regex('[0-9a-fA-F]{64}')
-
-SCHEMA = Map({
-    'magisk': Map({
-        'url': Str(),
-        'hash': SHA256_HEX,
-    }),
-    'device': MapPattern(Str(), Map({
-        'url': Str(),
-        'sections': Seq(Map({
-            'start': Int(),
-            'end': Int(),
-        })),
-        'hash': Map({
-            'original': Map({
-                'full': SHA256_HEX,
-                'stripped': SHA256_HEX,
-            }),
-            'patched': Map({
-                'full': SHA256_HEX,
-                'stripped': SHA256_HEX,
-            }),
-            'avb_images': MapPattern(Str(), SHA256_HEX),
-        }),
-    })),
-})
-
-
-def load_config():
-    with open(CONFIG_PATH, 'r') as f:
-        return load(f.read(), SCHEMA)
-
-
-def save_config(data):
-    with open(CONFIG_PATH, 'w') as f:
-        f.write(data.as_yaml())

tests/distros/Containerfile.alpine

@@ -1,5 +0,0 @@
-FROM docker.io/library/alpine:3.18
-
-RUN apk add --no-cache openssl py3-lz4 py3-pip py3-protobuf
-# Alpine does not have strictyaml in its repos
-RUN pip install strictyaml

tests/distros/Containerfile.arch

@@ -1,6 +0,0 @@
-FROM docker.io/archlinux/archlinux:latest
-
-RUN pacman --noconfirm -Syu --needed openssl python-lz4 python-pip python-protobuf \
-    && yes | pacman -Scc
-# Arch does not have strictyaml in its repos
-RUN pip install --break-system-packages strictyaml

tests/distros/Containerfile.arch-wine

@@ -1,55 +0,0 @@
-FROM docker.io/archlinux/archlinux:latest
-
-RUN pacman --noconfirm -Syu --needed aria2 xorg-server-xvfb \
-    && yes | pacman -Scc
-
-# Yep, we're installing Windows msys2 packages directly on top of Arch
-
-ARG KEYRING_VERSION=1~20230316-1
-ARG KEYRING_SHA256=eb672d0d41b2e857f97f0d9de6d3325113c992c5c1a9aab4bd468341e5471572
-
-RUN aria2c https://mirror.msys2.org/msys/x86_64/msys2-keyring-${KEYRING_VERSION}-any.pkg.tar.zst \
-        --checksum SHA-256=${KEYRING_SHA256} \
-    && pacman --noconfirm -U msys2-keyring-${KEYRING_VERSION}-any.pkg.tar.zst \
-    && rm msys2-keyring-${KEYRING_VERSION}-any.pkg.tar.zst
-
-COPY wine/pacman.additional.conf /tmp/
-RUN cat /tmp/pacman.additional.conf >> /etc/pacman.conf \
-    && pacman-key --init \
-    && pacman-key --populate \
-    && rm /tmp/pacman.additional.conf
-
-RUN pacman --noconfirm -Sy \
-        mingw-w64-x86_64-ca-certificates \
-        mingw-w64-x86_64-openssl \
-        mingw-w64-x86_64-python \
-        mingw-w64-x86_64-python-lz4 \
-        mingw-w64-x86_64-python-pip \
-        mingw-w64-x86_64-python-protobuf \
-        mingw-w64-x86_64-zlib \
-        wine \
-        wine-mono \
-    && yes | pacman -Scc
-
-# Since binfmt_misc can't work in an unprivileged container, all post-install
-# .exe executions failed, including the ones for setting up the trusted CA
-# certificates. Run the post install script using the Linux version of p11-kit
-# so we don't have to create a wine prefix (see below).
-RUN cat /var/lib/pacman/local/mingw-w64-x86_64-ca-certificates-*/install \
-    | sed 's,[^ ]*/\([^ ]\+\).exe,\1,g; $a post_install' \
-    | bash -s
-
-COPY --chmod=755 wine/python3.sh /usr/local/bin/python3
-
-# msys2 does not have strictyaml in its repos
-run WINEDLLOVERRIDES=winemenubuilder.exe=d \
-    WINEPREFIX=/tmp/wine \
-    python3 -m pip install strictyaml \
-    && rm -rf /tmp/wine
-
-# We don't precreate a wine prefix because openssl fails to gather entropy when
-# using a persisted wine prefix for some reason. It doesn't seem to actually be
-# caused by the lack of entropy.
-
-ENTRYPOINT ["xvfb-run"]
-CMD ["bash"]

tests/distros/Containerfile.debian

@@ -1,5 +0,0 @@
-FROM docker.io/library/debian:11
-
-RUN apt-get -y update \
-    && apt-get -y install openssl python3-lz4 python3-protobuf python3-strictyaml \
-    && find /var/cache/apt/archives /var/lib/apt/lists -mindepth 1 -delete

tests/distros/Containerfile.fedora

@@ -1,4 +0,0 @@
-FROM registry.fedoraproject.org/fedora-toolbox:38
-
-RUN dnf install -y openssl python3-lz4 python3-protobuf python3-strictyaml \
-    && find /var/cache/dnf -mindepth 1 -delete