Move from dynamic to static dependencies

Previously, dependencies were passed in with a struct of pointers but
they will now be resolved in the linker. This removes a soure of
potential runtime programming errors.

Where DiceOps were previously passed, now there is just a context
parameter to replace the previous context parameter of DiceOps.

We're also able to merge some testing and fuzzing boilerplate code as we
no longer need to have case specific code for preparing the dynamic
dependencies.

Change-Id: I9623222b1178ea4a8dd2112a6c84993ddc26afda
Reviewed-on: https://pigweed-review.googlesource.com/c/open-dice/+/54260
Reviewed-by: Andrew Scull <ascull@google.com>
Reviewed-by: Darren Krahn <dkrahn@google.com>
Commit-Queue: Andrew Scull <ascull@google.com>
Pigweed-Auto-Submit: Andrew Scull <ascull@google.com>
39 files changed
tree: fe9a5bfdc8df7d4f8d3a1d5d9ad7ed6a6b6f1a2e
  1. build_overrides/
  2. docs/
  3. images/
  4. include/
  5. src/
  6. third_party/
  7. toolchains/
  8. .clang-format
  9. .gitignore
  10. .gitmodules
  11. .gn
  12. activate.sh
  13. bootstrap.sh
  14. BUILD.gn
  15. BUILDCONFIG.gn
  16. generate_test_values.py
  17. LICENSE
  18. navbar.md
  19. README.md
  20. run_fuzzer.sh
README.md

Open Profile for DICE

This repository contains the specification for the Open Profile for DICE along with production-quality code. This profile is a specialization of the Hardware Requirements for a Device Identifier Composition Engine and DICE Layering Architecture specifications published by the Trusted Computing Group (TCG). For readers already familiar with those specs, notable distinctives of this profile include:

  • Separate CDIs for attestation and sealing use cases
  • Categorized inputs, including values related to verified boot
  • Certified UDS values
  • X.509 or CBOR certificates

Mailing List

You can find us (and join us!) at https://groups.google.com/g/open-profile-for-dice. We're happy to answer questions and discuss proposed changes or features.

Specification

The specification can be found here. It is versioned using a major.minor scheme. Compatibility is maintained across minor versions but not necessarily across major versions.

Code

Production quality, portable C code is included. The main code is in dice.h and dice.c. Cryptographic and certificate generation operations are injected via a set of callbacks. Multiple implementations of these operations are provided, all equally acceptable. Integrators should choose just one of these, or write their own.

Tests are included for all code and the build files in this repository can be used to build and run these tests.

Disclaimer: This is not an officially supported Google product.

Thirdparty Dependencies

Different implementations use different third party libraries. The third_party directory contains build files and git submodules for each of these. The bootstrap script will automatically initialize all submodules.

Building and Running Tests

$ source bootstrap.sh
$ ninja -C out

The easiest way, and currently the only supported way, to build and run tests is from a Pigweed environment on Linux. Pigweed does support other host platforms so it shouldn't be too hard to get this running on Windows for example, but we use Linux.

There are two scripts to help set this up:

  • bootstrap.sh will initialize submodules, bootstrap a Pigweed environment, and generate build files. This can take some time and may download on the order of 1GB of dependencies so the normal workflow is to just do this once.

  • activate.sh quickly reactivates an environment that has been previously bootstrapped.

These scripts must be sourced into the current session: source activate.sh.

In the environment, from the base directory of the dice-profile checkout, run ninja -C out to build everything and run all tests. You can also run pw watch which will build, run tests, and continue to watch for changes.

This will build and run tests on the host using the clang toolchain. Pigweed makes it easy to configure other targets and toolchains. See toolchains/BUILD.gn and the Pigweed documentation.

Porting

The code is designed to be portable and should work with a variety of modern toolchains and in a variety of environments. The main code in dice.h and dice.c is C99; it uses uint8_t, size_t, and memcpy from the C standard library. The various ops implementations are as portable as their dependencies (often not C99 but still very portable). Notably, this code uses designated initializers for readability. This is a feature available in C since C99 but missing from C++ until C++20 where it appears in a stricter form.

Style

The Google C++ Style Guide is used. A .clang-format file is provided for convenience.

Incorporating

To incorporate the code into another project, there are a few options:

  • Copy only the necessary code. For example:

    1. Take the main code as is: include/dice/dice.h, src/dice.c

    2. Choose an implementation for crypto and certificate generation or choose to write your own. If you choose the boringssl implementation, for example, take include/dice/utils.h, include/dice/boringssl_ops.h, src/utils.c, and src/boringssl_ops.c. Taking a look at the library targets in BUILD.gn may be helpful.

  • Add this repository as a git submodule and integrate into the project build, optionally using the gn library targets provided.

  • Integrate into a project already using Pigweed using the gn build files provided.

Size Reports

The build reports code size using Bloaty McBloatface via the pw_bloat Pigweed module. There are two reports generated:

  • Library sizes - This report includes just the library code in this repository. It shows the baseline DICE code with no ops selected, and it shows the delta introduced by choosing various ops implementations. This report does not include the size of the third party dependencies.

  • Executable sizes - This report includes sizes for the library code in this repository plus all dependencies linked into a simple main function which makes a single DICE call with all-zero input. It shows the baseline DICE code with no ops (and therefore no dependencies other than libc), and it shows the delta introduced by choosing various ops implementations. This report does include the size of the third party dependencies. Note that rows specialized from ‘Boringssl Ops’ use that as a baseline for sizing.

The reports will be in the build output, but you can also find the reports in .txt files in the build output. For example, cat out/host_optimized/gen/*.txt | less will display all reports.

Thread Safety

This code does not itself use mutable global variables, or any other type of shared data structure so there is no thread-safety concerns. However, additional care is needed to ensure dependencies are configured to be thread-safe. For example, the current boringssl configuration defines OPENSSL_NO_THREADS_CORRUPT_MEMORY_AND_LEAK_SECRETS_IF_THREADED, and that would need to be changed before running in a threaded environment.

Clearing Sensitive Data

This code makes a reasonable effort to clear memory holding sensitive data. This may help with a broader strategy to clear sensitive data but it is not sufficient on its own. Here are a few things to consider.

  • The caller of this code is responsible for buffers they own (of course).
  • The ops implementations need to clear any copies they make of sensitive data. Both boringssl and mbedtls attempt to zeroize but this may need additional care to integrate correctly. For example, boringssl skips optimization prevention when OPENSSL_NO_ASM is defined (and it is currently defined).
  • Sensitive data may remain in cache.
  • Sensitive data may have been swapped out.
  • Sensitive data may be included in a crash dump.