Allow quotes in URI

IE6 apparently sends such requests... Reported by Michael Carter.
2 files changed
tree: 83654cd3eef91449a05f03496e74f4e24d713c30
  1. .gitignore
  2. http_parser.h
  3. http_parser.rl
  5. Makefile
  7. test.c

HTTP Parser

This is a parser for HTTP messages written in C. It parses both requests and responses. The parser is designed to be used in performance HTTP applications. It does not make any allocations, it does not buffer data, and it can be interrupted at anytime. It only requires about 128 bytes of data per message stream (in a web server that is per connection).


  • No dependencies
  • Parses both requests and responses.
  • Handles keep-alive streams.
  • Decodes chunked encoding.
  • Extracts the following data from a message
    • header fields and values
    • content-length
    • request method
    • response status code
    • transfer-encoding
    • http version
    • request path, query string, fragment
    • message body
  • Defends against buffer overflow attacks.


One http_parser object is used per TCP connection. Initialize the struct using http_parser_init() and set the callbacks. That might look something like this:

http_parser *parser = malloc(sizeof(http_parser));
http_parser_init(parser, HTTP_REQUEST);
parser->on_path = my_path_callback;
parser->on_header_field = my_header_field_callback;
parser->data = my_socket;

When data is received on the socket execute the parser and check for errors.

size_t len = 80*1024;
char buf[len];
ssize_t recved;

recved = read(fd, buf, len);
if (recved != 0) // handle error

http_parser_execute(parser, buf, recved);

if (http_parser_has_error(parser)) {
  // handle error. usually just close the connection

Scalar valued message information such as status_code, method, and the HTTP version are stored in the parser structure. This data is only temporarlly stored in http_parser and gets reset on each new message. If this information is needed later, copy it out of the structure during the headers_complete callback.

The parser decodes the transfer-encoding for both requests and responses transparently. That is, a chunked encoding is decoded before being sent to the on_body callback.

It does not decode the content-encoding (gzip). Not all HTTP applications need to inspect the body. Decoding gzip is non-neglagable amount of processing (and requires making allocations). HTTP proxies using this parser, for example, would not want such a feature.


During the http_parser_execute() call, the callbacks set in http_parser will be executed. The parser maintains state and never looks behind, so buffering the data is not necessary. If you need to save certain data for later usage, you can do that from the callbacks.

There are two types of callbacks:

  • notification typedef int (*http_cb) (http_parser*); Callbacks: on_message_begin, on_headers_complete, on_message_complete.
  • data typedef int (*http_data_cb) (http_parser*, const char *at, size_t length); Callbacks: (requests only) on_path, on_query_string, on_uri, on_fragment, (common) on_header_field, on_header_value, on_body;

In case you parse HTTP message in chunks (i.e. read() request line from socket, parse, read half headers, parse, etc) your data callbacks may be called more than once. Http-parser guarantees that data pointer is only valid for the lifetime of callback. You can also read() into a heap allocated buffer to avoid copying memory around if this fits your application.

Reading headers may be a tricky task if you read/parse headers partially. Basically, you need to remember whether last header callback was field or value and apply following logic:

/* on_header_field and on_header_value shortened to on_h_*
 ------------------------ ------------ --------------------------------------------
| State (prev. callback) | Callback   | Description/action                         |
 ------------------------ ------------ --------------------------------------------
| nothing (first call)   | on_h_field | Allocate new buffer and copy callback data |
|                        |            | into it                                    |
 ------------------------ ------------ --------------------------------------------
| value                  | on_h_field | New header started.                        |
|                        |            | Copy current name,value buffers to headers |
|                        |            | list and allocate new buffer for new name  |
 ------------------------ ------------ --------------------------------------------
| field                  | on_h_field | Previous name continues. Reallocate name   |
|                        |            | buffer and append callback data to it      |
 ------------------------ ------------ --------------------------------------------
| field                  | on_h_value | Value for current header started. Allocate |
|                        |            | new buffer and copy callback data to it    |
 ------------------------ ------------ --------------------------------------------
| value                  | on_h_value | Value continues. Reallocate value buffer   |
|                        |            | and append callback data to it             |
 ------------------------ ------------ --------------------------------------------

See examples of reading in headers:


The source repo is at github.