path: root/doc/magnet.txt

{{{
#!rst
==============
a power-magnet
==============

------------------
Module: mod_magnet
------------------



.. contents:: Table of Contents

Requirements
============

:Version: lighttpd 1.4.12 or higher
:Packages: lua >= 5.1

Overview
========

mod_magnet is a module to control the request handling in lighty. 

.. note::

  Keep in mind that the magnet is executed in the core of lighty. EVERY long-running operation is blocking 
  ALL connections in the server. You are warned. For time-consuming or blocking scripts use mod_fastcgi and friends.

For performance reasons mod_magnet caches the compiled script. For each script-run the script itself is checked for 
freshness and recompile if neccesary.


Installation
============

mod_magnet needs a lighty which is compiled with the lua-support ( --with-lua). Lua 5.1 or higher are required by
the module. Use "--with-lua=lua5.1" to install on Debian and friends. ::

  server.modules = ( ..., "mod_magnet", ... )

Options
=======

mod_magnet can attract a request in several stages in the request-handling. 

* either at the same level as mod_rewrite, before any parsing of the URL is done
* or at a later stage, when the doc-root is known and the physical-path is already setup

It depends on the purpose of the script which stage you want to intercept. Usually you want to use
the 2nd stage where the physical-path which relates to your request is known. At this level you
can run checks against lighty.env["physical.path"].

::

  magnet.attract-raw-url-to = ( ... )
  magnet.attract-physical-path-to = ( ... )

You can define multiple scripts when separated by a semicolon. The scripts are executed in the specified 
order. If one of them a returning a status-code, the following scripts will not be executed.

Tables
======

Most of the interaction between between mod_magnet and lighty is done through tables. Tables in lua are hashes (Perl), dictionaries (Java), arrays (PHP), ...

Request-Environment
-------------------

Lighttpd has its internal variables which are exported as read/write to the magnet. 

If "http://example.org/search.php?q=lighty" is requested this results in a request like ::

  GET /search.php?q=lighty HTTP/1.1
  Host: example.org

When you are using ``attract-raw-url-to`` you can access the following variables:

* parts of the request-line

 * lighty.env["request.uri"] = "/search.php?q=lighty"

* HTTP request-headers

  * lighty.request["Host"] = "example.org"

Later in the request-handling, the URL is splitted, cleaned up and turned into a physical path name:

* parts of the URI

 * lighty.env["uri.path"] = "/search.php"
 * lighty.env["uri.path-raw"] = "/search.php"
 * lighty.env["uri.scheme"] = "http"
 * lighty.env["uri.authority"] = "example.org"
 * lighty.env["uri.query"] = "q=lighty"

* filenames, pathnames

 * lighty.env["physical.path"] = "/my-docroot/search.php"
 * lighty.env["physical.rel-path"] = "/search.php"
 * lighty.env["physical.doc-root"] = "/my-docroot"

All of them are readable, not all of the are writable (or don't have an effect if you write to them). 

As a start, you might want to use those variables for writing: ::

  -- 1. simple rewriting is done via the request.uri
  lighty.env["request.uri"] = ... 
  return lighty.RESTART_REQUEST

  -- 2. changing the physical-path
  lighty.env["physical.path"] = ...

  -- 3. changing the query-string
  lighty.env["uri.query"] = ...

Response Headers
----------------

If you want to set a response header for your request, you can add a field to the lighty.header[] table: ::

  lighty.header["Content-Type"] = "text/html"

Sending Content
===============

You can generate your own content and send it out to the clients. ::

  lighty.content = { "<pre>", { filename = "/etc/passwd" }, "</pre>" }
  lighty.header["Content-Type"] = "text/html"

  return 200

The lighty.content[] table is executed when the script is finished. The elements of the array are processed left to right and the elements can either be a string or a table. Strings are included AS IS into the output of the request.

* Strings

  * are included as is

* Tables

  * filename = "<absolute-path>" is required
  * offset = <number> [default: 0]
  * length = <number> [default: size of the file - offset]

Internally lighty will use the sendfile() call to send out the static files at full speed.

Status Codes
============

You might have seen it already in other examples: In case you are handling the request completly in the magnet you
can return your own status-codes. Examples are: Redirected, Input Validation, ... ::

  if (lighty.env["uri.scheme"] == "http") then
    lighty.header["Location"] = "https://" .. lighty.env["uri.authority"] .. lighty.env["request.uri"]
    return 302
  end

You every number above and equal to 100 is taken as final status code and finishes the request. No other modules are 
executed after this return.

A special return-code is lighty.RESTART_REQUEST (currently equal to 99) which is usually used in combination with 
changing the request.uri in a rewrite. It restarts the splitting of the request-uri again.

If you return nothing (or nil) the request-handling just continues.

Debugging
=========

To easy debugging we overloaded the print()-function in lua and redirect the output of print() to the error-log. ::

  print("Host: " .. lighty.request["Host"])
  print("Request-URI: " .. lighty.env["request.uri"])


Examples
========

Sending text-files as HTML
--------------------------

This is a bit simplistic, but it illustrates the idea: Take a text-file and cover it in a <pre> tag.

Config-file ::

  magnet.attract-physical-path-to = server.docroot + "/readme.lua"

readme.lua ::

  lighty.content = { "<pre>", { filename = "/README" }, "</pre>" }
  lighty.header["Content-Type"] = "text/html"
  
  return 200

Maintainance pages
------------------

Your side might be on maintainance from time to time. Instead of shutting down the server confusing all
users, you can just send a maintainance page.

Config-file ::

  magnet.attract-physical-path-to = server.docroot + "/maintainance.lua"

maintainance.lua ::

  require "lfs"

  if (nil == lfs.attributes(lighty.env["physical.doc-root"] .. "/maintainance.html")) then
    lighty.content = ( lighty.env["physical.doc-root"] .. "/maintainance.html" )

    lighty.header["Content-Type"] = "text/html"

    return 200
  end

mod_flv_streaming
-----------------

Config-file ::

  magnet.attract-physical-path-to = server.docroot + "/flv-streaming.lua"

flv-streaming.lua::

  if (lighty.env["uri.query"]) then
    -- split the query-string
    get = {}
    for k, v in string.gmatch(lighty.env["uri.query"], "(%w+)=(%w+)") do
      get[k] = v
    end

    if (get["start"]) then
      -- missing: check if start is numeric and positive

      -- send te FLV header + a seek into the file
      lighty.content = { "FLV\x1\x1\0\0\0\x9\0\0\0\x9", 
         { filename = lighty.env["physical.path"], offset = get["start"] } }
      lighty.header["Content-Type"] = "video/x-flv"

      return 200
    end
  end

  
selecting a random file from a directory
----------------------------------------

Say, you want to send a random file (ad-content) from a directory. 

To simplify the code and to improve the performance we define:

* all images have the same format (e.g. image/png)
* all images use increasing numbers starting from 1
* a special index-file names the highest number

Config ::

  server.modules += ( "mod_magnet" )
  magnet.attract-physical-path-to = "random.lua"

random.lua ::

  dir = lighty.env["physical.path"]

  f = assert(io.open(dir .. "/index", "r"))
  maxndx = f:read("*all")
  f:close()

  ndx = math.random(maxndx)

  lighty.content = { { filename = dir .. "/" .. ndx }}
  lighty.header["Content-Type"] = "image/png"

  return 200

denying illegal character sequences in the URL
----------------------------------------------

Instead of implementing mod_security, you might just want to apply filters on the content
and deny special sequences that look like SQL injection. 

A common injection is using UNION to extend a query with another SELECT query.

::

  if (string.find(lighty.env["request.uri"], "UNION%s")) then
    return 400
  end

Traffic Quotas
--------------

If you only allow your virtual hosts a certain amount for traffic each month and want to 
disable them if the traffic is reached, perhaps this helps: ::

  host_blacklist = { ["www.example.org"] = 0 }

  if (host_blacklist[lighty.request["Host"]]) then
    return 404
  end

Just add the hosts you want to blacklist into the blacklist table in the shown way.

Complex rewrites
----------------

If you want to implement caching on your document-root and only want to regenerate 
content if the requested file doesn't exist, you can attract the physical.path: ::

  magnet.attract-physical-path-to = ( server.document-root + "/rewrite.lua" )

rewrite.lua ::

  require "lfs"

  attr = lfs.attributes(lighty.env["physical.path"])

  if (not attr) then
    -- we couldn't stat() the file for some reason
    -- let the backend generate it

    lighty.env["uri.path"] = "/dispatch.fcgi"
    lighty.env["physical.rel-path"] = lighty.env["uri.path"]
    lighty.env["physical.path"] = lighty.env["physical.doc-root"] .. lighty.env["physical.rel-path"]
  fi

luafilesystem
+++++++++++++

We are requiring the lua-module 'lfs' (http://www.keplerproject.org/luafilesystem/). 

I had to compile lfs myself for lua-5.1 which required a minor patch as compat-5.1 is not needed::

  $ wget http://luaforge.net/frs/download.php/1487/luafilesystem-1.2.tar.gz
  $ wget http://www.lighttpd.net/download/luafilesystem-1.2-lua51.diff
  $ gzip -cd luafilesystem-1.2.tar.gz | tar xf -
  $ cd luafilesystem-1.2
  $ patch -ls -p1 < ../luafilesystem-1.2-lua51.diff
  $ make install

It will install lfs.so into /usr/lib/lua/5.1/ which is where lua expects the extensions on my system.

SuSE and Gentoo are known to have their own lfs packages and don't require a compile.

Usertracking
------------

... or how to store data globally in the script-context:

Each script has its own script-context. When the script is started it only contains the lua-functions
and the special lighty.* name-space. If you want to save data between script runs, you can use the global-script
context:

::

  if (nil == _G["usertrack"]) then
    _G["usertrack"] = {}
  end
  if (nil == _G["usertrack"][lighty.request["Cookie"]]) then
    _G["usertrack"][lighty.request["Cookie"]]
  else 
    _G["usertrack"][lighty.request["Cookie"]] = _G["usertrack"][lighty.request["Cookie"]] + 1
  end

  print _G["usertrack"][lighty.request["Cookie"]]

The global-context is per script. If you update the script without restarting the server, the context will still be maintained.

Counters
--------

mod_status support a global statistics page and mod_magnet allows to add and update values in the status page:

Config ::

  status.statistics-url = "/server-counters"
  magnet.attract-raw-url-to = server.docroot + "/counter.lua"

counter.lua ::

  lighty.status["core.connections"] = lighty.status["core.connections"] + 1

Result::

  core.connections: 7
  fastcgi.backend.php-foo.0.connected: 0
  fastcgi.backend.php-foo.0.died: 0
  fastcgi.backend.php-foo.0.disabled: 0
  fastcgi.backend.php-foo.0.load: 0
  fastcgi.backend.php-foo.0.overloaded: 0
  fastcgi.backend.php-foo.1.connected: 0
  fastcgi.backend.php-foo.1.died: 0
  fastcgi.backend.php-foo.1.disabled: 0
  fastcgi.backend.php-foo.1.load: 0
  fastcgi.backend.php-foo.1.overloaded: 0
  fastcgi.backend.php-foo.load: 0

Porting mod_cml scripts
-----------------------

mod_cml got replaced by mod_magnet.

A CACHE_HIT in mod_cml::
 
  output_include = { "file1", "file2" } 

  return CACHE_HIT

becomes::

  content = { { filename = "/path/to/file1" }, { filename = "/path/to/file2"} }

  return 200

while a CACHE_MISS like (CML) ::

  trigger_handler = "/index.php"

  return CACHE_MISS

becomes (magnet) ::

  lighty.env["request.uri"] = "/index.php"

  return lighty.RESTART_REQUEST

}}}