|author||Andrew Scull <firstname.lastname@example.org>||Fri Oct 15 11:25:57 2021 +0000|
|committer||CQ Bot Account <email@example.com>||Mon Oct 18 11:15:49 2021 +0000|
Stop using VLAs While the lengths were never actually variable, in the eyes of the C language, they could have been. The compiler doesn't acknowledge that the variables are actually constants but using #define instead is enough to stop it worrying about it. With this change, -Wvla is added to the build flags to make sure there aren't any truly variable length arrays and lets the library fit more easily into projects that take VLAs more seriously. Test: ninja -C out/ Change-Id: Ib41174211e1f55fe9d8f4b1e0e834e999b8a6b7d Reviewed-on: https://pigweed-review.googlesource.com/c/open-dice/+/65680 Commit-Queue: Andrew Scull <firstname.lastname@example.org> Pigweed-Auto-Submit: Andrew Scull <email@example.com> Reviewed-by: Darren Krahn <firstname.lastname@example.org>
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:
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.
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.
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.
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.
$ 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:
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.
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.
The Google C++ Style Guide is used. A
.clang-format file is provided for convenience.
To incorporate the code into another project, there are a few options:
Copy only the necessary code. For example:
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.
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.
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.
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.