AWS C Common

Core c99 package for AWS SDK for C. Includes cross-platform primitives, configuration, data structures, and error handling.

License

This library is licensed under the Apache 2.0 License.

Usage

Building

aws-c-common uses CMake for setting up build environments. This library has no non-kernel dependencies so the build is quite simple.

For example:

git clone git@github.com:awslabs/aws-c-common.git aws-c-common
mkdir aws-c-common-build
cd aws-c-common-build
cmake ../aws-c-common
make -j 12
make test
sudo make install

Keep in mind that CMake supports multiple build systems, so for each platform you can pass your own build system as the -G option. For example:

cmake -GNinja ../aws-c-common
ninja build
ninja test
sudo ninja install

Or on windows,

cmake -G "Visual Studio 14 2015 Win64" ../aws-c-common
msbuild.exe ALL_BUILD.vcproj

CMake Options

• -DCMAKE_CLANG_TIDY=/path/to/clang-tidy (or just clang-tidy or clang-tidy-7.0 if it is in your PATH) - Runs clang-tidy as part of your build.
• -DENABLE_SANITIZERS=ON - Enables gcc/clang sanitizers, by default this adds -fsanitizer=address,undefined to the compile flags for projects that call aws_add_sanitizers.
• -DENABLE_FUZZ_TESTS=ON - Includes fuzz tests in the unit test suite. Off by default, because fuzz tests can take a long time. Set -DFUZZ_TESTS_MAX_TIME=N to determine how long to run each fuzz test (default 60s).
• -DCMAKE_INSTALL_PREFIX=/path/to/install - Standard way of installing to a user defined path. If specified when configuring aws-c-common, ensure the same prefix is specified when configuring other aws-c-* SDKs.
• -DAWS_STATIC_MSVC_RUNTIME_LIBRARY=ON - Windows-only. Turn ON to use the statically-linked MSVC runtime lib, instead of the DLL.

API style and conventions

Every API has a specific set of styles and conventions. We'll outline them here. These conventions are followed in every library in the AWS C SDK ecosystem.

Error handling

Every function that returns an int type, returns AWS_OP_SUCCESS ( 0 ) or AWS_OP_ERR (-1) on failure. To retrieve the error code, use the function aws_last_error(). Each error code also has a corresponding error string that can be accessed via the aws_error_str() function.

In addition, you can install both a global and a thread local error handler by using the aws_set_global_error_handler_fn() and aws_set_thread_local_error_handler_fn() functions.

All error functions are in the include/aws/common/error.h header file.

Naming

Any function that allocates and initializes an object will be suffixed with new (e.g. aws_myobj_new()). Similarly, these objects will always have a corresponding function with a destroy suffix. The new functions will return the allocated object on success and NULL on failure. To respond to the error, call aws_last_error(). If several new or destroy functions are available, the variants should be named like new_x or destroy_x (e.g. aws_myobj_new_copy() or aws_myobj_destroy_secure()).

Any function that initializes an existing object will be suffixed with init (e.g. aws_myobj_init(). These objects will have a corresponding clean_up function if necessary. In these cases, you are responsible for making the decisions for how your object is allocated. The init functions return AWS_OP_SUCCESS ( 0 ) or AWS_OP_ERR (-1) on failure. If several init or clean_up functions are available, they should be named like init_x or clean_up_x (e.g. aws_myobj_init_static() or aws_myobj_clean_up_secure()).

Contributing

If you are contributing to this code-base, first off, THANK YOU!. There are a few things to keep in mind to minimize the pull request turn around time.

Coding "guidelines"

These "guidelines" are followed in every library in the AWS C SDK ecosystem.

Memory Management

• All APIs that need to be able to allocate memory, must take an instance of aws_allocator and use that. No malloc() or free() calls should be made directly.
• If an API does not allocate the memory, it does not free it. All allocations and deallocations should take place at the same level. For example, if a user allocates memory, the user is responsible for freeing it. There will inevitably be a few exceptions to this rule, but they will need significant justification to make it through the code-review.
• All functions that allocate memory must raise an AWS_ERROR_OOM error code upon allocation failures. If it is a new() function it should return NULL. If it is an init() function, it should return AWS_OP_ERR.

Threading

• Occasionally a thread is necessary. In those cases, prefer for memory not to be shared between threads. If memory must cross a thread barrier it should be a complete ownership hand-off. Bias towards, "if I need a mutex, I'm doing it wrong".
• Do not sleep or block .... ever .... under any circumstances, in non-test-code.
• Do not expose blocking APIs.

Error Handling

• For APIs returning an int error code. The only acceptable return types are AWS_OP_SUCCESS and AWS_OP_ERR. Before returning control to the caller, if you have an error to raise, use the aws_raise_error() function.
• For APIs returning an allocated instance of an object, return the memory on success, and NULL on failure. Before returning control to the caller, if you have an error to raise, use the aws_raise_error() function.

Log Subjects & Error Codes

The logging & error handling infrastructure is designed to support multiple libraries. For this to work, AWS maintained libraries have pre-slotted log subjects & error codes for each library. The currently allocated ranges are:

Each library should begin its error and log subject values at the beginning of its range and follow in sequence (don't skip codes). Upon adding an AWS maintained library, a new enum range must be approved and added to the above table.

Testing

We have a high bar for test coverage, and PRs fixing bugs or introducing new functionality need to have tests before they will be accepted. A couple of tips:

Aws Test Harness

We provide a test harness for writing unit tests. This includes an allocator that will fail your test if you have any memory leaks, as well as some ASSERT macros. To write a test:

• Create a *.c test file in the tests directory of the project.
• Implement one or more tests with the signature int test_case_name(struct aws_allocator *, void *ctx)
• Use the AWS_TEST_CASE macro to declare the test.
• Include your test in the tests/main.c file.
• Include your test in the tests/CMakeLists.txt file.

Coding Style

• No Tabs.
• Indent is 4 spaces.
• K & R style for braces.
• Space after if, before the (.
• else and else if stay on the same line as the closing brace.

Example:

if (condition) {
    do_something();
} else {
    do_something_else();
}

• Avoid C99 features in header files. For some types such as bool, uint32_t etc..., these are defined if not available for the language standard being used in aws/common/common.h, so feel free to use them.
• For C++ compatibility, don't put const members in structs.
• Avoid C++ style comments e.g. // in header files and prefer block style (/* */) for long blocks of text. C++ style comments are fine in C files.
• All public API functions need C++ guards and Windows dll semantics.
• Use Unix line endings.
• Where implementation hiding is desired for either ABI or runtime polymorphism reasons, use the void *impl pattern. v-tables should be the last member in the struct.
• For #ifdef, put a # as the first character on the line and then indent the compilation branches.

Example:

#ifdef FOO
    do_something();

#   ifdef BAR
    do_something_else();
#   endif
#endif

• For all error code names with the exception of aws-c-common, use AWS_ERROR_<lib name>_<error name>.

• All error strings should be written using correct English grammar.

• SNAKE_UPPER_CASE constants, macros, and enum members.

• snake_lower_case everything else.

• static (local file scope) variables that are not const are prefixed by s_ and lower snake case.

• Global variables not prefixed as const are prefixed by g_ and lower snake case.

• Thread local variables are prefixed as tl_ and lower snake case.

• Macros and const variables are upper snake case.

• For constants, prefer anonymous enums.

• Don't typedef structs. It breaks forward declaration ability.

• Don't typedef enums. It breaks forward declaration ability.

• typedef function definitions for use as function pointers as values and suffixed with _fn.

  Do this:

  typedef int(fn_name_fn)(void *);

  Not this:

  typedef int(*fn_name_fn)(void *);

• If a callback may be async, then always have it be async. Callbacks that are sometimes async and sometimes sync are hard to code around and lead to bugs (see this blog post). Unfortunately many callbacks in this codebase currently violate this rule, so be careful. But do not add any more.

• Every source and header file must have a copyright header (The standard AWS one for apache 2).

• Use standard include guards (e.g. #IFNDEF HEADER_NAME #define HEADER_NAME etc...).

• Include order should be: the header for the translation unit for the .c file newline header files in a directory in alphabetical order newline header files not in a directory (system and stdlib headers)

• Platform specifics should be handled in c files and partitioned by directory.

• Do not use extern inline. It's too unpredictable between compiler versions and language standards.

• Namespace all definitions in header files with aws_<libname>?_<api>_<what it does>. Lib name is not always required if a conflict is not likely and it provides better ergonomics.

• init, clean_up, new, destroy are suffixed to the function names for their object.

Example:

AWS_COMMON_API

int aws_module_init(aws_module_t module); AWS_COMMON_API void aws_module_clean_up(aws_module_t module); AWS_COMMON_API aws_module_t aws_module_new(aws_allocator_t allocator); AWS_COMMON_API void aws_module_destroy(aws_module_t *module);

• Avoid c-strings, and don't write code that depends on NULL terminators. Expose struct aws_byte_buf APIs and let the user figure it out.
• There is only one valid character encoding-- UTF-8. Try not to ever need to care about character encodings, but where you do, the working assumption should always be UTF-8 unless it's something we don't get a choice in (e.g. a protocol explicitly mandates a character set).
• If you are adding/using a compiler specific keyword, macro, or intrinsic, hide it behind a platform independent macro definition. This mainly applies to header files. Obviously, if you are writing a file that will only be built on a certain platform, you have more liberty on this.

• When checking more than one error condition, check and log each condition separately with a unique message.

  Do this:

  if (options->callback == NULL) {
      AWS_LOGF_ERROR(AWS_LS_SOME_SUBJECT, "Invalid options - callback is null");
      return aws_raise_error(AWS_ERROR_INVALID_ARGUMENT);
  }
  
  if (options->allocator == NULL) {
      AWS_LOGF_ERROR(AWS_LS_SOME_SUBJECT, "Invalid options - allocator is null");
      return aws_raise_error(AWS_ERROR_INVALID_ARGUMENT);
  }

  Not this:

  if (options->callback == NULL || options->allocator == NULL) {
      AWS_LOGF_ERROR(AWS_LS_SOME_SUBJECT, "Invalid options - something is null");
      return aws_raise_error(AWS_ERROR_INVALID_ARGUMENT);
  }

CBMC

To learn more about CBMC and proofs specifically, review the training material here.

The verification/cbmc/proofs directory contains CBMC proofs.

In order to run these proofs you will need to install CBMC and other tools by following the instructions here.