What is WebAssembly (Wasm)?

As described on https://webassembly.org/

WebAssembly (abbreviated Wasm) is a binary instruction format for a stack-based virtual machine. Wasm is designed as a portable target for compilation of high-level languages like C/C++/Rust, enabling deployment on the web for client and server applications.

Overview

OPA is able to compile Rego policies into executable Wasm modules that can be evaluated with different inputs and external data. This is not running the OPA server in Wasm, nor is this just cross-compiled Golang code. The compiled Wasm module is a planned evaluation path for the source policy and query.

Current Status

The core language is supported fully but there are a number of built-in functions that are not, and probably won’t be natively supported in Wasm (e.g., http.send). Built-in functions that are not natively supported can be implemented in the host environment (e.g., JavaScript).

Compiling Policies

You can compile Rego policies into Wasm modules using the opa build subcommand.

For example, the opa build command below compiles the example.rego file into a Wasm module and packages it into an OPA bundle. The wasm target requires at least one entrypoint rule (specified by -e).

  1. opa build -t wasm -e example/allow example.rego

The output of a Wasm module built this way contain the result of evaluating the entrypoint rule. For example:

  1. [
  2. {
  3. "result": <value of data.example.allow>
  4. }
  5. ]

The output of policy evaluation is a set of variable assignments. The variable assignments specify values that satisfy the expressions in the policy query (i.e., if the variables in the query are replaced with the values from the assignments, all of the expressions in the query would be defined and not false.)

When policies are compiled into Wasm, the user provides the path of the policy decision that should be exposed by the Wasm module. The policy decision is assigned to a variable named result. The policy decision can be ANY JSON value (boolean, string, object, etc.) but there will be at-most-one assignment. This means that callers should first check if the set of variable assignments is empty (indicating an undefined policy decision) otherwise they should select the "result" key out of the variable assignment set.

For more information on opa build run opa build --help.

Advanced Compiling Options

You can also compile Rego policies into Wasm modules from Go using the lower-level rego API that produces raw Wasm executables and the higher-level compile API that produces OPA bundle files. The compile API is recommended.

Using Compiled Policies

JavaScript SDK

There is a JavaScript SDK available that simplifies the process of loading and evaluating compiled policies. If you want to evaluate Rego policies inside JavaScript we recommend you use the JavaScript SDK.

See https://github.com/open-policy-agent/npm-opa-wasm for more details.

There is an example NodeJS application located here.

From Scratch

If you want to integrate Wasm compiled policies into a language or runtime that does not have SDK support, read this section.

Instantiating the Wasm Module

Before you can evaluate Wasm compiled policies you need to instantiate the Wasm module produced by the compilation process described earlier on this page.

To load the compiled Wasm module refer the documentation for the Wasm runtime that you are using. At a high-level you must provide a memory buffer and a set of import functions. The memory buffer is a contiguous, mutable byte-array that allows you to pass data to the policy and receive output from the policy. The import functions are dependencies of the compiled policies.

ABI Versions

Wasm modules built using OPA 0.27.0 onwards contain a global variable named opa_wasm_abi_version that has a constant i32 value indicating the ABI version this module requires. Described below you find ABI versions 1.x.

There’s another i32 constant exported, opa_wasm_abi_minor_version, used to track backwards-compatible changes.

Using tools like wasm-objdump (wasm-objdump -x policy.wasm), the ABI version can be found here:

  1. Global[3]:
  2. - global[0] i32 mutable=1 - init i32=121904
  3. - global[1] i32 mutable=0 <opa_wasm_abi_version> - init i32=1
  4. - global[2] i32 mutable=0 <opa_wasm_abi_minor_version> - init i32=0
  5. Export[19]:
  6. [...]
  7. - global[1] -> "opa_wasm_abi_version"
  8. - global[2] -> "opa_wasm_abi_minor_version"

Note the i32=1 of global[1], exported by the name of opa_wasm_abi_version.

Version notes
ABINotes
1.0Start of ABI versioning.
1.1Adds export memory.
1.2Adds exported function opa_eval.

Exports

The primary exported functions for interacting with policy modules are listed below. In the ABI column, you can find the ABI version with which the export was introduced.

Function SignatureDescriptionABI
int32 eval(ctxaddr)Evaluates the loaded policy with the provided evaluation context. The return value is reserved for future use.1.0
value_addr builtins(void)Returns the address of a mapping of built-in function names to numeric identifiers that are required by the policy.1.0
value_addr entrypoints(void)Returns the address of a mapping of entrypoints to numeric identifiers that can be selected when evaluating the policy.1.0
ctx_addr opa_eval_ctx_new(void)Returns the address of a newly allocated evaluation context.1.0
void opa_eval_ctx_set_input(ctx_addr, value_addr)Set the input value to use during evaluation. This must be called before each eval() call. If the input value is not set before evaluation, references to the input document result produce no results (i.e., they are undefined.)1.0
void opa_eval_ctx_set_data(ctx_addr, value_addr)Set the data value to use during evalutaion. This should be called before each eval() call. If the data value is not set before evalutaion, references to base data documents produce no results (i.e., they are undefined.)1.0
void opa_eval_ctx_set_entrypoint(ctx_addr, entrypoint_id)Set the entrypoint to evaluate. By default, entrypoint with id 0 is evaluated.1.0
value_addr opa_eval_ctx_get_result(ctx_addr)Get the result set produced by the evaluation process.1.0
addr opa_malloc(int32 size)Allocates size bytes in the shared memory and returns the starting address.1.0
void opa_free(addr)Free a pointer. Calls opa_abort on error.1.0
value_addr opa_json_parse(str_addr, size)Parses the JSON serialized value starting at str_addr of size bytes and returns the address of the parsed value. The parsed value may refer to a null, boolean, number, string, array, or object value.1.0
value_addr opa_value_parse(str_addr, size)The same as opa_json_parse except Rego set literals are supported.1.0
str_addr opa_json_dump(value_addr)Dumps the value referred to by value_addr to a null-terminated JSON serialized string and returns the address of the start of the string. Rego sets are serialized as JSON arrays. Non-string Rego object keys are serialized as strings.1.0
str_addr opa_value_dump(value_addr)The same as opa_json_dump except Rego sets are serialized using the literal syntax and non-string Rego object keys are not serialized as strings.1.0
void opa_heap_ptr_set(addr)Set the heap pointer for the next evaluation.1.0
addr opa_heap_ptr_get(void)Get the current heap pointer.1.0
int32 opa_value_add_path(base_value_addr, path_value_addr, value_addr)Add the value at the value_addr into the object referenced by base_value_addr at the given path. The path_value_addr must point to an array value with string keys (eg: [“a”, “b”, “c”]). Existing values will be updated. On success the value at value_addr is no longer owned by the caller, it will be freed with the base value. The path must be freed by the caller after use (see opa_free). If an error occurs the base value will remain unchanged. Example: base object {“a”: {“b”: 123}}, path [“a”, “x”, “y”], and value {“foo”: “bar”} will yield {“a”: {“b”: 123, “x”: {“y”: {“foo”: “bar”}}}}. Returns an error code (see below).1.0
int32 opa_value_remove_path(base_value_addr, path_value_addr)Remove the value from the object referenced by base_value_addr at the given path. Values removed will be freed. The path must be freed by the caller after use (see opa_free). The path_value_addr must point to an array value with string keys (eg: [“a”, “b”, “c”]). Returns an error code (see below).1.0
str_addr opa_eval( addr, entrypoint_id int32, data value_addr, input str_addr, input_len int32, heap_ptr addr, format int32)One-off policy evaluation method. Its arguments are everything needed to evaluate: entrypoint, address of data in memory, address and length of input JSON string in memory, heap address to use, and the output format (0 is JSON, 1 is “value”, i.e. serialized Rego values). The first argument is reserved for future use and must be 0. Returns the address to the serialised result value.1.2

The addresses passed and returned by the policy modules are 32-bit integer offsets into the shared memory region. The value_addr parameters and return values refer to OPA value data structures: null, boolean, number, string, array, object, and set.

Error codes:

OPA Wasm Error codes are int32 values defined as:

ValueNameDescription
0OPA_ERR_OKNo error.
1OPA_ERR_INTERNALUnrecoverable internal error.
2OPA_ERR_INVALID_TYPEInvalid value type was encountered.
3OPA_ERR_INVALID_PATHInvalid object path reference.

Imports

Policy modules require the following function imports at instantiation-time:

NamespaceNameParamsResultDescription
envopa_abort(addr)voidCalled if an internal error occurs. The addr refers to a null-terminated string in the shared memory buffer.
envopa_println(addr)voidCalled to emit a message from the policy evaluation. The addr refers to a null-terminated string in the shared memory buffer.
envopa_builtin0(builtin_id, ctx)addrCalled to dispatch the built-in function identified by the builtin_id. The ctx parameter reserved for future use. The result addr must refer to a value in the shared-memory buffer. The function accepts 0 arguments.
envopa_builtin1(builtin_id, ctx, _1)addrSame as previous except the function accepts 1 argument.
envopa_builtin2(builtin_id, ctx, _1, _2)addrSame as previous except the function accepts 2 arguments.
envopa_builtin3(builtin_id, ctx, _1, _2, _3)addrSame as previous except the function accepts 3 arguments.
envopa_builtin4(builtin_id, ctx, _1, _2, _3, _4)addrSame as previous except the function accepts 4 arguments.

The policy module also requires a shared memory buffer named env.memory.

Memory Buffer

A shared memory buffer must be provided as an import for the policy module with the name env.memory. The buffer must be large enough to accommodate the input, provided data, and result of evaluation.

Built-in Functions

After instantiating the policy module, call the exported builtins function to receive a mapping of built-in functions required during evaluation. The result maps required built-in function names to the identifiers supplied to the built-in function callbacks (e.g., opa_builtin0, opa_builtin1, etc.)

For example:

  1. const memory = new WebAssembly.Memory({ initial: 5 });
  2. const policy_module = await WebAssembly.instantiate(byte_buffer, /* import object */);
  3. const addr = policy_module.instance.exports.builtins();
  4. const str_addr = policy_module.instance.exports.opa_json_dump(addr);
  5. const builtin_map = deserialize_null_terminated_JSON_string(memory, str_addr);

The built-in function mapping will contain all of the built-in functions that may be required during evaluation. For example, the following query refers to the http.send built-in function which is not included in the policy module:

  1. result := http.send({"method": "get", "url": "https://example.com/api/lookup/12345"})

If this query was compiled to Wasm the built-in map would contain a single element:

  1. {
  2. "http.send": 0
  3. }

When the evaluation runs, the opa_builtin1 callback would invoked with builtin_id set to 0.

Evaluation

Once instantiated, the policy module is ready to be evaluated. Use the opa_eval_ctx_new exported function to create an evaluation context. Use the opa_eval_ctx_set_input and opa_eval_ctx_set_data exported functions to specify the values of the input and base data documents to use during evaluation.

To evaluate, call to the exported eval function with the eval context address as the only parameter.

Input

The (optional) input document for a policy can be provided by loading a JSON string into the shared memory buffer. Use the opa_malloc exported function to allocate a buffer the size of the JSON string and copy the contents in at the returned address. After the raw string is loaded into memory you will need to call the opa_json_parse exported method to get an address to the parsed input document for use in evaluations. Set the address via the opa_eval_ctx_set_input exported functoin supplying the evaluation context address and parsed input document address.

External Data

External data can be loaded for use in evaluation. Similar to the input this is done by loading a JSON string into the shared memory buffer. Use opa_malloc and opa_json_parse followed by opa_eval_ctx_set_data to set the address on the evaluation context.

Data can be updated by using the opa_value_add_path and opa_value_remove_path and providing the same value address as the base. Similarly, use opa_malloc and opa_json_parse for the updated value and creating the path.

After loading the external data use the opa_heap_ptr_get exported method to save the current point in the heap before evaluation. After evaluation this should be reset by calling opa_heap_ptr_set to ensure that evaluation restarts back at the saved data and re-uses heap space. This is particularly important if re-evaluating many times with the same data.

Entrypoints

The compiled policy may have one or more entrypoints. If no entrypoint is set on the evaluation context the default entrypoint (0) will be evaluated. SDKs can call entrypoints() after instantiating the module to retrieve the entrypoint name to entrypoint identifier mapping. SDKs can set the entrypoint to evaluate by calling opa_eval_ctx_set_entrypoint on the evaluation context. If an invalid entrypoint identifier is passed, the eval function will invoke opa_abort.

Results

After evaluation results can be retrieved via the exported opa_eval_ctx_get_result function. Pass in the evaluation context address. The return value is an address in the shared memory buffer to the structured result. To access the JSON result use the opa_json_dump exported function to retrieve a pointer in shared memory to a null terminated JSON string.

The result of evaluation is the set variable bindings that satisfy the expressions in the query. For example, the query x = 1; y = 2; y > x would produce the following result set:

  1. [
  2. {
  3. "x": 1,
  4. "y": 2
  5. }
  6. ]

Sets are represented as JSON arrays.