path: root/doc/localcgi.md

<!-- MANON
% cha-localcgi(5) | Local CGI support in Chawan
MANOFF -->

# Local CGI support in Chawan

Chawan supports the invocation of CGI scripts locally. This feature can be
used in the following way:

* All local CGI scripts must be placed in a directory specified in
  `external.cgi-dir`. Multiple directories can be specified in an array too,
  and directories specified first have higher precedence.
* Then, a CGI script in one of these directories can be executed by visiting
  the URL `cgi-bin:script-name`. $PATH_INFO and $QUERY_STRING are set as
  normal, i.e. `cgi-bin:script-name/abcd?defgh=ijkl` will set $PATH_INFO to
  `/abcd`, and $QUERY_STRING to `defgh=ijkl`.

Further notes on processing CGI paths:

* The URL must be opaque, so you must not add a double slash after the scheme.
  e.g. `cgi-bin://script-name` will NOT work, only `cgi-bin:script-name`.
* Paths beginning with `/cgi-bin/` or `/$LIB/` are stripped of this segment
  automatically. So e.g. `cgi-bin:/cgi-bin/script-name` becomes
  `cgi-bin:script-name`.
* If `external.w3m-cgi-compat` is true, file: URLs are converted to cgi-bin:
  URLs if the path name starts with `/cgi-bin/`, `/$LIB/`, or the path of
  a local CGI script. Note: this is unsafe, please do not use it unless you
  must.
* Absolute paths are accepted as e.g. `cgi-bin:/path/to/cgi/dir/script-name`.
  Note however, that this only works if `/path/to/cgi/dir` has already been
  specified as a CGI directory in `external.cgi-dir`.

Note that this is different from w3m's cgi-bin functionality, in that we
use a custom scheme for local CGI instead of interpreting all requests to
a designated path as a CGI request. (This incompatibility is bridged over when
`external.w3m-cgi-compat` is true.)

## Headers

Local CGI scripts may send some headers that Chawan will interpret
specially (and thus will not pass forward to e.g. the fetch API, etc):

* `Status`: interpreted as the HTTP status code.
* `Cha-Control`: special header, see below.

Note that these headers MUST be sent before any regular headers. Headers
received after a regular header or a `Cha-Control: ControlDone` header will be
treated as regular headers.

The `Cha-Control` header's value is parsed as follows:

```
Cha-Control-Value = Command *Parameter
Command = ALPHA *ALPHA
Parameter = SPACE *CHAR
```

In other words, it is `Command [Param1] [Param2] ...`.

Currently available commands are:

* `Connected`: Takes no parameters. Must be the first reported header;
  it means that connection to the server has been successfully established,
  but no data has been received yet. When any other header is sent first,
  Chawan will act as if a `Cha-Control: Connected` header had been implicitly
  sent before that.
* `ConnectionError`: Must be the first reported header. Parameter 1 is the
  error code, see below. If any following parameters are given, they are
  concatenated to form a custom error message.  
  Note: short but descriptive error messages are preferred, messages that
  do not fit on the screen are currently truncated. (TODO fix this somehow :P)
* `ControlDone`: Signals that no more special headers will be sent; this
  means that `Cha-Control` and `Status` headers sent after this must be
  interpreted as regular headers (and thus e.g. will be available for JS
  code calling the script using the fetch API).  
  WARNING: this header must be sent before any non-hardcoded headers that
  take external input. For example, an HTTP client would have to send
  `Cha-Control: ControlDone` before returning the retrieved headers.

List of public error codes:

* `1 internal error`: An internal error prevented the script from retrieving
  the requested resource. CGI scripts can also use this to signal that they
  have no information on what went wrong.
* `2 invalid method`: The client requested data using a method not supported
  by this protocol.
* `3 invalid URL`: The request URL could not be interpreted as a valid URL
  for this format.
* `4 file not found`: No file was found at the requested address, and thus
  the request is meaningless. Note: this should only be used by protocols
  that do not rely on a client-server architecture, e.g. local file access,
  local databases, or peer-to-peer file retrieval mechanisms. A server
  responding with "no file found" is NOT a connection error, and is better
  represented as a response with a 404 status code.
* `5 failed to resolve host`: The hostname could not be resolved.
* `6 failed to resolve proxy`: The proxy could not be resolved.
* `7 connection refused`: The server refused to establish a connection.
* `8 proxy refused to connect`: The proxy refused to establish a connection.

## Environment variables

Chawan sets the following environment variables:

* `SERVER_SOFTWARE="Chawan"`
* `SERVER_PROTOCOL="HTTP/1.0"`
* `SERVER_NAME="localhost"`
* `SERVER_PORT="80"`
* `REMOTE_HOST="localhost"`
* `REMOTE_ADDR="127.0.0.1"`
* `GATEWAY_INTERFACE="CGI/1.1"`
* `SCRIPT_NAME="/cgi-bin/script-name"` if called with a relative path, and
  `"/path/to/script/script-name"` if called with an absolute path.
* `SCRIPT_FILENAME="/path/to/script/script-name"`
* `QUERY_STRING=` the query string (i.e. `URL.search`). Note that this
  variable is percent-encoded.
* `PATH_INFO=` everything after the script's path name,
  e.g. for `cgi-bin:script-name/abcd/efgh` `"/abcd/efgh"`. Note that this
  variable is NOT percent-encoded.
* `REQUEST_URI="$SCRIPT_NAME/$PATH_INFO?$QUERY_STRING`
* `REQUEST_METHOD=` HTTP method used for making the request, e.g. GET or POST
* `REQUEST_HEADERS=` A newline-separated list of all headers for this request.
* `CHA_LIBEXEC_DIR=` The libexec directory Chawan was configured to use at
  compile time. See the [tools](#tools) section below for details of
  why this is useful.
* `CONTENT_TYPE=` for POST requests, the Content-Type header. Not set for
  other request types (e.g. GET).
* `CONTENT_LENGTH=` the content length, if $CONTENT_TYPE has been set.
* `ALL_PROXY=` if a proxy has been set, the proxy URL. WARNING: for security
  reasons, this MUST be respected when making external connections. If a
  CGI script does not support proxies, it must never make any external
  connections when the `ALL_PROXY` variable is set, even if this results in it
  returning an error.
* `HTTP_COOKIE=` if set, the Cookie header.
* `HTTP_REFERER=` if set, the Referer header.

For requests originating from a urimethodmap rewrite, Chawan will also set
the parsed URL's parts as environment variables. Use of these is highly
encouraged, to avoid exploits originating from double-parsing of URLs.

e.g. if example://username:password@example.org:1234/path/name.html?example
is the original URL, then:

* `MAPPED_URI_SCHEME=` the scheme of the original URL, in this case `example`.
* `MAPPED_URI_USERNAME=` the username part, in this case `username`. If no
  username was specified, the variable is set to the empty string.
* `MAPPED_URI_PASSWORD=` the password part, in this case `password`. If no
  password was specified, the variable is set to the empty string.
* `MAPPED_URI_HOST=` the host part, in this case `host.org` If no host was
  specified, the variable is set to the empty string. (An example of a URL
  with no host: `about:blank`, here `blank` is the path name.)
* `MAPPED_URI_PORT=` the port, in this case `1234`. If no port was specified,
  the variable is set to the empty string. (In this case, the CGI script is
  expected to use the default port for the scheme, if any.)
* `MAPPED_URI_PATH=` the path name, in this case `/path/name.html?example`. If
  no path was specified, the variable is set to the empty string. Note:
  the path name is percent-encoded.
* `MAPPED_URI_QUERY=` the query string, in this case `example`. Note that,
  unlike in JavaScript, no question mark is prepended to the string.  
  The query string is percent-encoded as well.

Note: the fragment part is omitted intentionally.

## Request body

If the request body is not empty, it is streamed into the program through
the standard input.

Note that this may be both an application/x-www-form-urlencoded or a
multipart/form-data request; `CONTENT_TYPE` stores information about the
request type, and in case of a multipart request, the boundary as well.

## Tools

Chawan provides certain helper binaries that may be useful for CGI
scripts. These can be portably accessed by executing
`"$CHA_LIBEXEC_DIR"/[program name]`.

Currently, the following tools are available:

* `urldec`: percent-decode strings passed on standard input.
* `urlenc`: percent-encode strings passed on standard input, taking a
  percent-encode set as the first parameter.

## Troubleshooting

Note that standard error is redirected to the browser console (by default,
M-cM-c). This makes it easy to debug a misbehaving CGI script, but may also
slow down the browser in case of excessive logging. If this is not the
desired behavior, we recommend wrapping your script into a shell script that
redirects stderr to /dev/null.

### My script is returning a "no local-CGI directory configured" error message.

Currently, the default setting includes a cgi-bin directory at
`$(which cha)/../libexec/chawan/cgi-bin`, which usually looks something like
`/usr/local/libexec/chawan/cgi-bin`. You only get the above message if you
intentionally set the cgi-dir setting to an empty array. (This will likely break
everything else too, so do not.)

To change the default local-CGI directory, use the `external.cgi-dir` option.

e.g. you could add this to your config.toml:

```toml
[external]
cgi-dir = ["~/cgi-bin", "${%CHA_LIBEXEC_DIR}/cgi-bin"]
```

and then put your script in `$HOME/cgi-bin`. Note the second element in the
array; if you don't add it, the default CGI scripts (including http, https,
etc...) will not work.

### My script is returning a "Failed to execute script" error message.

This means the `execl` call to the script failed. Make sure that your CGI
script's executable bit is set, i.e. run `chmod +x /path/to/cgi/script`.

### My script is returning an "invalid CGI path" error message.

Make sure that you did not include leading slashes. Reminder:
`cgi-bin://script-name` does not work, use `cgi-bin:script-name`.

### My script is returning a "CGI file not found" error message.

Double check that your CGI script is in the correct location. Also, make
sure that you are not accidentally calling the script with an absolute path via
`cgi-bin:/script-name` (instead of the correct `cgi-bin:script-name`).

It is also possible that `external.cgi-dir` is not really set to the directory
your script is in. Note that by default, this depends on the binary's path,
so e.g. if your binary is in `~/src/chawan/target/release/bin/cha`, but you
put your CGI script to `/usr/local/libexec/chawan/cgi-bin`, then it will
not work.

### My script is returning a "failed to set up CGI script" error message.

This means that either `pipe` or `fork` failed. Something strange is going on
with your system; we recommend exorcism. (Maybe you are running out of memory?)

<!-- MANON
## See also

**cha**(1)
MANOFF -->