App-Level Patching

Experimental Feature

App-level patching is an experimental feature that requires setting the COPA_EXPERIMENTAL=1 environment variable to enable. This feature is under active development, and future releases may introduce breaking changes. Feedback is welcome!

Copa supports patching application-level dependencies, such as Python packages, in addition to operating system packages. This feature allows you to update vulnerable libraries and packages in various programming language ecosystems.

Overview

App-level patching works by scanning and updating application dependencies found in your container images. Unlike OS-level patching which updates system packages, app-level patching focuses on:

• Python packages (pip is the supported package manager)

Please note that app-level patching requires scanner results that identify vulnerabilities in application libraries.

Package Type Filtering

Copa supports filtering between different types of packages using the --pkg-types flag:

# Patch only OS packages (default)
copa patch -i $IMAGE --pkg-types os ...

# Patch only library/app-level packages
copa patch -i $IMAGE --pkg-types library ...

# Patch both OS and library packages
copa patch -i $IMAGE --pkg-types os,library ...

Package Type Options

• os: Operating system packages (APT, YUM, APK, etc.). This is the default behavior if no --pkg-types flag is specified.
• library: Application-level packages (Python, Node.js, etc.)
• os,library: Both types

This filtering is particularly useful when you want to:

• Apply different patch policies to OS vs. application dependencies
• Separate OS security updates from application updates
• Reduce the scope of changes in production environments

Patch Level Control

Copa allows you to control how aggressively application-level packages are updated based on their versioning. This is particularly useful for managing compatibility and stability in your applications. The --library-patch-level flag determines the maximum version bump allowed for library updates:

# Only apply patch-level updates (e.g., 2.6.0 → 2.6.1)
copa patch -i $IMAGE --pkg-types library --library-patch-level patch ...

# Allow minor version updates (e.g., 2.6.0 → 2.7.0, prefer 2.6.1)
copa patch -i $IMAGE --pkg-types library --library-patch-level minor ...

# Allow major version updates (e.g., 2.6.0 → 3.0.0, prefer 2.6.1)
copa patch -i $IMAGE --pkg-types library --library-patch-level major ...

Please note that the --library-patch-level flag requires the --pkg-types library option to be set. Default behavior is patch level if not specified.

Patch Level Behavior

The patch level determines the maximum version bump allowed for library updates:

patch Level (Recommended)

• Allows: 2.6.0 → 2.6.1, 2.6.2, etc.
• Blocks: 2.6.0 → 2.7.0 or 3.0.0
• Use case: Conservative updates, minimal risk of breaking changes

minor Level

• Allows: 2.6.0 → 2.6.1 (preferred) or 2.7.0
• Blocks: 2.6.0 → 3.0.0
• Preference: If both 2.6.1 and 2.7.0 are available, it will choose 2.6.1 (patch) over 2.7.0 (minor)
• Use case: Moderate updates, some new features acceptable

major Level

• Allows: Any version update
• Preference: When comma-separated versions are available, prefers Patch > Minor > Major for compatibility. When no comma-separated versions exist, picks the highest version to fix all CVEs.
• Example: If both 2.6.1 and 2.7.0 are available, it will choose 2.6.1 for better compatibility
• Use case: Aggressive updates, all fixes applied regardless of compatibility risk

warning

Please note that copa does not guarantee compatibility with all versions. The patch level only controls the maximum version bump allowed. Always test your application after patching.

Special Package Handling

Some packages have special handling due to their nature:

• certifi: Always updated to the latest version regardless of patch level setting

Usage Examples

To use app-level patching, you need to have a scanner result file that contains vulnerabilities for application-level packages. Below is an example of how to scan a Python image and then apply patches using Copa.

export COPA_EXPERIMENTAL=1
export IMAGE=python:3.11.0

# Scan for Python package vulnerabilities
trivy image --vuln-type os,library --ignore-unfixed -f json -o python-scan.json $IMAGE

Basic Usage

# Apply patch-level Python package updates only
copa patch \
    -i $IMAGE \
    -r python-scan.json \
    --pkg-types os,library \
    --library-patch-level patch

Ignoring Errors

Sometimes, certain packages may not be compatible with the patching process or may not have available updates. In such cases, you can use the --ignore-errors flag to allow Copa to continue patching other packages even if some fail. This is useful in environments where you want to apply as many updates as possible without failing the entire patching process.

# Apply patch-level Python package updates, ignoring errors
copa patch \
    -i $IMAGE \
    -r python-scan.json \
    --pkg-types os,library \
    --library-patch-level major \
    --ignore-errors

Limitations

Due to nature of app-level patching, it may not be as comprehensive as OS-level patching. Some known limitations are:

Python

Dependency Resolution

Copa does not perform dependency resolution for application-level packages. It applies updates based on the scanner results without checking for compatibility with other packages in the environment. This means that while Copa can update vulnerable packages, it may not resolve all dependency conflicts that arise from those updates.

For example, if a package has a strict version requirement that conflicts with the updated version, you may encounter errors like:

#8 8.971 ERROR: Cannot install azure-cli and paramiko==3.4.0 because these package versions have conflicting dependencies.
#8 8.971
#8 8.971 The conflict is caused by:
#8 8.971     The user requested paramiko==3.4.0
#8 8.971     azure-cli-core 2.40.0 depends on paramiko<3.0.0 and >=2.0.8
#8 8.971
#8 8.971 To fix this you could try to:
#8 8.971 1. loosen the range of package versions you've specified
#8 8.971 2. remove package versions to allow pip attempt to solve the dependency conflict

Python Version Compatibility

Copa does not check whether the updated Python packages are compatible with the Python version in the image. For example, if you update a package that requires Python 3.12 to a version that is not compatible with Python 3.11, you may encounter runtime or dependency resolution errors.

Testing and Validation

Due to the nature of app-level patching, it is highly recommended to thoroughly test your application after applying updates. Copa does not perform any automated testing or validation of the patched application, so you should ensure that your application functions correctly with the updated dependencies.

Non Existent Versions

Trivy provides vulnerability data for Python dependencies using GitHub Security Advisories (GHSA). However, it does not check whether the patched version exists in the Python Package Index (PyPI). For example, GHSA-3749-ghw9-m3mg contains a vulnerability for torch package, but the patched version 2.0.2.7.1-rc1 does not exist in PyPI at the time of this writing.

Virtual Environment and Package Manager Support

Currently, only Python packages managed by pip are supported. We have not evaluated or implemented support for virtual environments, or other Python package managers like conda or poetry and others. This might break compatibility with applications that use these package managers.

Replacing PyPI

Copa does not support replacing the default Python Package Index (PyPI) with a custom index or mirror at this time. This means that all package updates are fetched from the official PyPI repository, which may not be suitable for all environments, especially those with strict network policies or private package registries.