Port of http_parser to llparse.

Let's face it, http_parser is practically unmaintainable. Even introduction of a single new method results in a significant code churn.

This project aims to:

• Make it maintainable
• Verifiable
• Improving benchmarks where possible

More details in Fedor Indutny's talk at JSConf EU 2019

Over time, different approaches for improving http_parser's code base were tried. However, all of them failed due to resulting significant performance degradation.

This project is a port of http_parser to TypeScript. llparse is used to generate the output C source file, which could be compiled and linked with the embedder's program (like Node.js).

So far llhttp outperforms http_parser:

llhttp is faster by approximately 156%.

llhttp project has about 1400 lines of TypeScript code describing the parser itself and around 450 lines of C code and headers providing the helper methods. The whole http_parser is implemented in approximately 2500 lines of C, and 436 lines of headers.

All optimizations and multi-character matching in llhttp are generated automatically, and thus doesn't add any extra maintenance cost. On the contrary, most of http_parser's code is hand-optimized and unrolled. Instead describing "how" it should parse the HTTP requests/responses, a maintainer should implement the new features in http_parser cautiously, considering possible performance degradation and manually optimizing the new code.

The state machine graph is encoded explicitly in llhttp. The llparse automatically checks the graph for absence of loops and correct reporting of the input ranges (spans) like header names and values. In the future, additional checks could be performed to get even stricter verification of the llhttp.

#include "llhttp.h"

llhttp_t parser;
llhttp_settings_t settings;

/* Initialize user callbacks and settings */
llhttp_settings_init(&settings);

/* Set user callback */
settings.on_message_complete = handle_on_message_complete;

/* Initialize the parser in HTTP_BOTH mode, meaning that it will select between
 * HTTP_REQUEST and HTTP_RESPONSE parsing automatically while reading the first
 * input.
 */
llhttp_init(&parser, HTTP_BOTH, &settings);

/* Parse request! */
const char* request = "GET / HTTP/1.1\r\n\r\n";
int request_len = strlen(request);

enum llhttp_errno err = llhttp_execute(&parser, request, request_len);
if (err == HPE_OK) {
  /* Successfully parsed! */
} else {
  fprintf(stderr, "Parse error: %s %s\n", llhttp_errno_name(err),
          parser.reason);
}

For more information on API usage, please refer to src/native/api.h.

The settings object contains a list of callbacks that the parser will invoke.

The following callbacks can return 0 (proceed normally), -1 (error) or HPE_PAUSED (pause the parser):

• on_message_begin: Invoked when a new request/response starts.
• on_message_complete: Invoked when a request/response has been completedly parsed.
• on_url_complete: Invoked after the URL has been parsed.
• on_method_complete: Invoked after the HTTP method has been parsed.
• on_version_complete: Invoked after the HTTP version has been parsed.
• on_status_complete: Invoked after the status code has been parsed.
• on_header_field_complete: Invoked after a header name has been parsed.
• on_header_value_complete: Invoked after a header value has been parsed.
• on_chunk_header: Invoked after a new chunk is started. The current chunk length is stored in parser->content_length.
• on_chunk_extension_name_complete: Invoked after a chunk extension name is started.
• on_chunk_extension_value_complete: Invoked after a chunk extension value is started.
• on_chunk_complete: Invoked after a new chunk is received.
• on_reset: Invoked after on_message_complete and before on_message_begin when a new message is received on the same parser. This is not invoked for the first message of the parser.

The following callbacks can return 0 (proceed normally), -1 (error) or HPE_USER (error from the callback):

• on_url: Invoked when another character of the URL is received.
• on_status: Invoked when another character of the status is received.
• on_method: Invoked when another character of the method is received. When parser is created with HTTP_BOTH and the input is a response, this also invoked for the sequence HTTP/ of the first message.
• on_version: Invoked when another character of the version is received.
• on_header_field: Invoked when another character of a header name is received.
• on_header_value: Invoked when another character of a header value is received.
• on_chunk_extension_name: Invoked when another character of a chunk extension name is received.
• on_chunk_extension_value: Invoked when another character of a extension value is received.

The callback on_headers_complete, invoked when headers are completed, can return:

• 0: Proceed normally.
• 1: Assume that request/response has no body, and proceed to parsing the next message.
• 2: Assume absence of body (as above) and make llhttp_execute() return HPE_PAUSED_UPGRADE.
• -1: Error
• HPE_PAUSED: Pause the parser.

Initialize the parser with specific type and user settings.

Returns the type of the parser.

Returns the major version of the HTTP protocol of the current request/response.

Returns the minor version of the HTTP protocol of the current request/response.

Returns the method of the current request.

Returns the method of the current response.

Returns 1 if request includes the Connection: upgrade header.

Reset an already initialized parser back to the start state, preserving the existing parser type, callback settings, user data, and lenient flags.

Initialize the settings object.

Parse full or partial request/response, invoking user callbacks along the way.

If any of llhttp_data_cb returns errno not equal to HPE_OK - the parsing interrupts, and such errno is returned from llhttp_execute(). If HPE_PAUSED was used as a errno, the execution can be resumed with llhttp_resume() call.

In a special case of CONNECT/Upgrade request/response HPE_PAUSED_UPGRADE is returned after fully parsing the request/response. If the user wishes to continue parsing, they need to invoke llhttp_resume_after_upgrade().

if this function ever returns a non-pause type error, it will continue to return the same error upon each successive call up until llhttp_init() is called.

This method should be called when the other side has no further bytes to send (e.g. shutdown of readable side of the TCP connection.)

Requests without Content-Length and other messages might require treating all incoming bytes as the part of the body, up to the last byte of the connection.

This method will invoke on_message_complete() callback if the request was terminated safely. Otherwise a error code would be returned.

Returns 1 if the incoming message is parsed until the last byte, and has to be completed by calling llhttp_finish() on EOF.

Returns 1 if there might be any other messages following the last that was successfully parsed.

Make further calls of llhttp_execute() return HPE_PAUSED and set appropriate error reason.

Do not call this from user callbacks! User callbacks must return HPE_PAUSED if pausing is required.

Might be called to resume the execution after the pause in user's callback.

See llhttp_execute() above for details.

Call this only if llhttp_execute() returns HPE_PAUSED.

Might be called to resume the execution after the pause in user's callback. See llhttp_execute() above for details.

Call this only if llhttp_execute() returns HPE_PAUSED_UPGRADE

Returns the latest error.

Returns the verbal explanation of the latest returned error.

User callback should set error reason when returning the error. See llhttp_set_error_reason() for details.

Assign verbal description to the returned error. Must be called in user callbacks right before returning the errno.

HPE_USER error code might be useful in user callbacks.

Returns the pointer to the last parsed byte before the returned error. The pointer is relative to the data argument of llhttp_execute().

This method might be useful for counting the number of parsed bytes.

Returns textual name of error code.

Returns textual name of HTTP method.

Returns textual name of HTTP status.

Enables/disables lenient header value parsing (disabled by default). Lenient parsing disables header value token checks, extending llhttp's protocol support to highly non-compliant clients/server.

No HPE_INVALID_HEADER_TOKEN will be raised for incorrect header values when lenient parsing is "on".

USE AT YOUR OWN RISK!

Enables/disables lenient handling of conflicting Transfer-Encoding and Content-Length headers (disabled by default).

Normally llhttp would error when Transfer-Encoding is present in conjunction with Content-Length.

This error is important to prevent HTTP request smuggling, but may be less desirable for small number of cases involving legacy servers.

USE AT YOUR OWN RISK!

Enables/disables lenient handling of Connection: close and HTTP/1.0 requests responses.

Normally llhttp would error on (in strict mode) or discard (in loose mode) the HTTP request/response after the request/response with Connection: close and Content-Length.

This is important to prevent cache poisoning attacks, but might interact badly with outdated and insecure clients.

With this flag the extra request/response will be parsed normally.

USE AT YOUR OWN RISK!

Enables/disables lenient handling of Transfer-Encoding header.

Normally llhttp would error when a Transfer-Encoding has chunked value and another value after it (either in a single header or in multiple headers whose value are internally joined using , ).

This is mandated by the spec to reliably determine request body size and thus avoid request smuggling.

With this flag the extra value will be parsed normally.

USE AT YOUR OWN RISK!

Make sure you have Node.js, npm and npx installed. Then under project directory run:

npm install
make

• Lua: MunifTanjim/llhttp.lua
• Python: pallas/pyllhttp
• Ruby: metabahn/llhttp
• Rust: JackLiar/rust-llhttp

If you want to use this library in a CMake project as a shared library, you can use the snippet below.

FetchContent_Declare(llhttp
  URL "https://github.com/nodejs/llhttp/archive/refs/tags/release/v8.1.0.tar.gz")

FetchContent_MakeAvailable(llhttp)

# Link with the llhttp_shared target
target_link_libraries(${EXAMPLE_PROJECT_NAME} ${PROJECT_LIBRARIES} llhttp_shared ${PROJECT_NAME})

If you want to use this library in a CMake project as a static library, you can set some cache variables first.

FetchContent_Declare(llhttp
  URL "https://github.com/nodejs/llhttp/archive/refs/tags/release/v8.1.0.tar.gz")

set(BUILD_SHARED_LIBS OFF CACHE INTERNAL "")
set(BUILD_STATIC_LIBS ON CACHE INTERNAL "")
FetchContent_MakeAvailable(llhttp)

# Link with the llhttp_static target
target_link_libraries(${EXAMPLE_PROJECT_NAME} ${PROJECT_LIBRARIES} llhttp_static ${PROJECT_NAME})

Note that using the git repo directly (e.g., via a git repo url and tag) will not work with FetchContent_Declare because CMakeLists.txt requires string replacements (e.g., _RELEASE_) before it will build.

• choco install git
• choco install node
• choco install llvm (or install the C Clang tools for Windows optional package from the Visual Studio 2019 installer)
• choco install make (or if you have MinGW, it comes bundled)

1. Ensure that Clang and make are in your system path.
2. Using Git Bash, clone the repo to your preferred location.
3. Cd into the cloned directory and run npm install
4. Run make
5. Your repo/build directory should now have libllhttp.a and libllhttp.so static and dynamic libraries.
6. When building your executable, you can link to these libraries. Make sure to set the build folder as an include path when building so you can reference the declarations in repo/build/llhttp.h.

Assuming you have an executable main.cpp in your current working directory, you would run: clang -Os -g3 -Wall -Wextra -Wno-unused-parameter -I/path/to/llhttp/build main.cpp /path/to/llhttp/build/libllhttp.a -o main.exe.

If you are getting unresolved external symbol linker errors you are likely attempting to build llhttp.c without linking it with object files from api.c and http.c.

This software is licensed under the MIT License.

Copyright Fedor Indutny, 2018.

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.