CGI for Caddy

Package cgi implements the common gateway interface (CGI) for Caddy 2, a modern, full-featured, easy-to-use web server.

It has been forked from the fantastic work of Kurt Jung who wrote that plugin for Caddy 1.

Documentation

This plugin lets you generate dynamic content on your website by means of command line scripts. To collect information about the inbound HTTP request, your script examines certain environment variables such as PATH_INFO and QUERY_STRING. Then, to return a dynamically generated web page to the client, your script simply writes content to standard output. In the case of POST requests, your script reads additional inbound content from standard input.

The advantage of CGI is that you do not need to fuss with server startup and persistence, long term memory management, sockets, and crash recovery. Your script is called when a request matches one of the patterns that you specify in your Caddyfile. As soon as your script completes its response, it terminates. This simplicity makes CGI a perfect complement to the straightforward operation and configuration of Caddy. The benefits of Caddy, including HTTPS by default, basic access authentication, and lots of middleware options extend easily to your CGI scripts.

CGI has some disadvantages. For one, Caddy needs to start a new process for each request. This can adversely impact performance and, if resources are shared between CGI applications, may require the use of some interprocess synchronization mechanism such as a file lock. Your server's responsiveness could in some circumstances be affected, such as when your web server is hit with very high demand, when your script's dependencies require a long startup, or when concurrently running scripts take a long time to respond. However, in many cases, such as using a pre-compiled CGI application like fossil or a Lua script, the impact will generally be insignificant. Another restriction of CGI is that scripts will be run with the same permissions as Caddy itself. This can sometimes be less than ideal, for example when your script needs to read or write files associated with a different owner.

Security Considerations

Serving dynamic content exposes your server to more potential threats than serving static pages. There are a number of considerations of which you should be aware when using CGI applications.

CGI scripts should be located outside of Caddy's document root. Otherwise, an inadvertent misconfiguration could result in Caddy delivering the script as an ordinary static resource. At best, this could merely confuse the site visitor. At worst, it could expose sensitive internal information that should not leave the server.

Mistrust the contents of PATH_INFO, QUERY_STRING and standard input. Most of the environment variables available to your CGI program are inherently safe because they originate with Caddy and cannot be modified by external users. This is not the case with PATH_INFO, QUERY_STRING and, in the case of POST actions, the contents of standard input. Be sure to validate and sanitize all inbound content. If you use a CGI library or framework to process your scripts, make sure you understand its limitations.

Errors

An error in a CGI application is generally handled within the application itself and reported in the headers it returns.

Application Modes

Your CGI application can be executed directly or indirectly. In the direct case, the application can be a compiled native executable or it can be a shell script that contains as its first line a shebang that identifies the interpreter to which the file's name should be passed. Caddy must have permission to execute the application. On Posix systems this will mean making sure the application's ownership and permission bits are set appropriately; on Windows, this may involve properly setting up the filename extension association.

In the indirect case, the name of the CGI script is passed to an interpreter such as lua, perl or python.

Requirements

This module needs to be installed (obviously).

Refer to the Caddy documentation on how to build Caddy with plugins/modules.

Note: If you build manually, beware to include the major version in the module path. For example xcaddy build --with=github.com/aksdb/caddy-cgi/v2.

Basic Syntax

The basic cgi directive lets you add a handler in the current caddy router location with a given script and optional arguments. The matcher is a default caddy matcher that is used to restrict the scope of this directive. The directive can be repeated any reasonable number of times. Here is the basic syntax:

cgi [matcher] exec [args...]

For example:

cgi /report /usr/local/cgi-bin/report

When a request such as https://example.com/report or https://example.com/report/weekly arrives, the cgi middleware will detect the match and invoke the script named /usr/local/cgi-bin/report. The current working directory will be the same as Caddy itself. Here, it is assumed that the script is self-contained, for example a pre-compiled CGI application or a shell script. Here is an example of a standalone script, similar to one used in the cgi plugin's test suite:

#!/bin/bash

printf "Content-type: text/plain\n\n"

printf "PATH_INFO    [%s]\n" $PATH_INFO
printf "QUERY_STRING [%s]\n" $QUERY_STRING

exit 0

The environment variables PATH_INFO and QUERY_STRING are populated and passed to the script automatically. There are a number of other standard CGI variables included that are described below. If you need to pass any special environment variables or allow any environment variables that are part of Caddy's process to pass to your script, you will need to use the advanced directive syntax described below.

Beware that in Caddy v2 it is (currently) not possible to separate the path left of the matcher from the full URL. Therefore if you require your CGI program to know the SCRIPT_NAME, make sure to pass that explicitly:

cgi /script.cgi* /path/to/my/script someargument {
  script_name /script.cgi
}

Advanced Syntax

In order to specify custom environment variables, pass along one or more environment variables known to Caddy, or specify more than one match pattern for a given rule, you will need to use the advanced directive syntax. That looks like this:

cgi [matcher] exec [args...] {
    script_name subpath
    dir working_directory
    env key1=val1 [key2=val2...]
    pass_env key1 [key2...]
    pass_all_env
    buffer_limit <size>
    unbuffered_output
    inspect
}

For example,

cgi /sample/report* /usr/local/bin/reportscript.sh {
    script_name /sample/report
    env DB=/usr/local/share/app/app.db SECRET=/usr/local/share/app/secret CGI_LOCAL=
    pass_env HOME UID
}

The script_name subdirective helps the cgi module to separate the path to the script from the (virtual) path afterwards (which shall be passed to the script).

env can be used to define a list of key=value environment variable pairs that shall be passed to the script. pass_env can be used to define a list of environment variables of the Caddy process that shall be passed to the script.

If your CGI application runs properly at the command line but fails to run from Caddy it is possible that certain environment variables may be missing. For example, the ruby gem loader evidently requires the HOME environment variable to be set; you can do this with the subdirective pass_env HOME. Another class of problematic applications require the COMPUTERNAME variable.

The pass_all_env subdirective instructs Caddy to pass each environment variable it knows about to the CGI excutable. This addresses a common frustration that is caused when an executable requires an environment variable and fails without a descriptive error message when the variable cannot be found. These applications often run fine from the command prompt but fail when invoked with CGI. The risk with this subdirective is that a lot of server information is shared with the CGI executable. Use this subdirective only with CGI applications that you trust not to leak this information.

buffer_limit is used when a http request has Transfer-Endcoding: chunked. The Go CGI Handler refused to handle these kinds of requests, see golang/go#5613. In order to work around this the chunked request is buffered by caddy and sent to the CGI application as a whole with the correct CONTENT_LENGTH set. The buffer_limit setting marks a threshold between buffering in memory and using a temporary file. Every request body smaller than the buffer_limit is buffered in-memory. It accepts all formats supported by go-humanize. Default: 4MiB.
(An example of this is git push if the objects to push are larger than the http.postBuffer)

With the unbuffered_output subdirective it is possible to instruct the CGI handler to flush output from the CGI script as soon as possible. By default, the output is buffered into chunks before it is being written to optimize the network usage and allow to determine the Content-Length. When unbuffered, bytes will be written as soon as possible. This will also force the response to be written in chunked encoding.

Troubleshooting

If you run into unexpected results with the CGI plugin, you are able to examine the environment in which your CGI application runs. To enter inspection mode, add the subdirective inspect to your CGI configuration block. This is a development option that should not be used in production. When in inspection mode, the plugin will respond to matching requests with a page that displays variables of interest. In particular, it will show the replacement value of {match} and the environment variables to which your CGI application has access.

For example, consider this example CGI block:

cgi /wapp/*.tcl /usr/local/bin/wapptclsh /home/quixote/projects{path} {
    script_name /wapp
    pass_env HOME LANG
    env DB=/usr/local/share/app/app.db SECRET=/usr/local/share/app/secret
    inspect
}

When you request a matching URL, for example,

    https://example.com/wapp/hello.tcl

the Caddy server will deliver a text page similar to the following. The CGI application (in this case, wapptclsh) will not be called.

CGI for Caddy inspection page

Executable .................... /usr/local/bin/wapptclsh
  Arg 1 ....................... /home/quixote/projects/hello.tcl
Root .......................... /
Dir ........................... /home/quixote/www
Environment
  DB .......................... /usr/local/share/app/app.db
  PATH_INFO ...................
  REMOTE_USER .................
  SCRIPT_EXEC ................. /usr/local/bin/wapptclsh /home/quixote/projects/hello.tcl
  SCRIPT_FILENAME ............. /usr/local/bin/wapptclsh
  SCRIPT_NAME ................. /wapp/hello
  SECRET ...................... /usr/local/share/app/secret
Inherited environment
  HOME ........................ /home/quixote
  LANG ........................ en_US.UTF-8
Placeholders
  {path} ...................... /hello
  {root} ...................... /
  {http.request.host} ......... example.com
  {http.request.host} ......... GET
  {http.request.host} ......... /wapp/hello.tcl

This information can be used to diagnose problems with how a CGI application is called.

To return to operation mode, remove or comment out the inspect subdirective.

Environment Variable Example

In this example, the Caddyfile looks like this:

{
    http_port 8080
    order cgi before respond
}

192.168.1.2:8080
cgi /show* /usr/local/cgi-bin/report/gen {
    script_name /show
}

Note that a request for /show gets mapped to a script named /usr/local/cgi-bin/report/gen. There is no need for any element of the script name to match any element of the match pattern.

The contents of /usr/local/cgi-bin/report/gen are:

#!/bin/bash

printf "Content-type: text/plain\n\n"

printf "example error message\n" > /dev/stderr

if [ "POST" = "$REQUEST_METHOD" -a -n "$CONTENT_LENGTH" ]; then
  read -n "$CONTENT_LENGTH" POST_DATA
fi

printf "AUTH_TYPE         [%s]\n" $AUTH_TYPE
printf "CONTENT_LENGTH    [%s]\n" $CONTENT_LENGTH
printf "CONTENT_TYPE      [%s]\n" $CONTENT_TYPE
printf "GATEWAY_INTERFACE [%s]\n" $GATEWAY_INTERFACE
printf "PATH_INFO         [%s]\n" $PATH_INFO
printf "PATH_TRANSLATED   [%s]\n" $PATH_TRANSLATED
printf "POST_DATA         [%s]\n" $POST_DATA
printf "QUERY_STRING      [%s]\n" $QUERY_STRING
printf "REMOTE_ADDR       [%s]\n" $REMOTE_ADDR
printf "REMOTE_HOST       [%s]\n" $REMOTE_HOST
printf "REMOTE_IDENT      [%s]\n" $REMOTE_IDENT
printf "REMOTE_USER       [%s]\n" $REMOTE_USER
printf "REQUEST_METHOD    [%s]\n" $REQUEST_METHOD
printf "SCRIPT_EXEC       [%s]\n" $SCRIPT_EXEC
printf "SCRIPT_NAME       [%s]\n" $SCRIPT_NAME
printf "SERVER_NAME       [%s]\n" $SERVER_NAME
printf "SERVER_PORT       [%s]\n" $SERVER_PORT
printf "SERVER_PROTOCOL   [%s]\n" $SERVER_PROTOCOL
printf "SERVER_SOFTWARE   [%s]\n" $SERVER_SOFTWARE

exit 0

The purpose of this script is to show how request information gets communicated to a CGI script. Note that POST data must be read from standard input. In this particular case, posted data gets stored in the variable POST_DATA. Your script may use a different method to read POST content. Secondly, the SCRIPT_EXEC variable is not a CGI standard. It is provided by this middleware and contains the entire command line, including all arguments, with which the CGI script was executed.

When a browser requests

    http://192.168.1.2:8080/show/weekly?mode=summary

the response looks like

AUTH_TYPE         []
CONTENT_LENGTH    []
CONTENT_TYPE      []
GATEWAY_INTERFACE [CGI/1.1]
PATH_INFO         [/weekly]
PATH_TRANSLATED   []
POST_DATA         []
QUERY_STRING      [mode=summary]
REMOTE_ADDR       [192.168.1.35]
REMOTE_HOST       [192.168.1.35]
REMOTE_IDENT      []
REMOTE_USER       []
REQUEST_METHOD    [GET]
SCRIPT_EXEC       [/usr/local/cgi-bin/report/gen]
SCRIPT_NAME       [/show]
SERVER_NAME       [192.168.1.2:8080]
SERVER_PORT       [8080]
SERVER_PROTOCOL   [HTTP/1.1]
SERVER_SOFTWARE   [go]

When a client makes a POST request, such as with the following command

wget -O - -q --post-data="city=San%20Francisco" http://192.168.1.2:8080/show/weekly?mode=summary

the response looks the same except for the following lines:

CONTENT_LENGTH    [20]
CONTENT_TYPE      [application/x-www-form-urlencoded]
POST_DATA         [city=San%20Francisco]
REQUEST_METHOD    [POST]

Go Source Example

This small example demonstrates how to write a CGI program in Go. The use of a bytes.Buffer makes it easy to report the content length in the CGI header.

package main

import (
    "bytes"
    "fmt"
    "os"
    "time"
)

func main() {
    var buf bytes.Buffer

    fmt.Fprintf(&buf, "Server time at %s is %s\n",
        os.Getenv("SERVER_NAME"), time.Now().Format(time.RFC1123))
    fmt.Println("Content-type: text/plain")
    fmt.Printf("Content-Length: %d\n\n", buf.Len())
    buf.WriteTo(os.Stdout)
}

When this program is compiled and installed as /usr/local/bin/servertime, the following directive in your Caddy file will make it available:

cgi /servertime /usr/local/bin/servertime

Execute dynamic script

The module is written in a way that it expects the scripts you want it to execute to actually exist. A non-existing or non-executable file is considered a setup error and will yield a HTTP 500.

If you want to make sure, only existing scripts are executed, use a more specific matcher, as explained in the Caddy docs.

Example:

@iscgi {
    path /cgi/*
    file {
        root ./app/
    }
}
cgi @iscgi ./app{path}

When calling a url like /cgi/foo/bar.pl it will check if the local file ./app/foo/bar.pl exists and only then it will proceed with calling the CGI.