Developer doc

From FusionForge Wiki
Revision as of 14:35, 26 March 2010 by Olenz (talk | contribs) (Merged Security page)

Jump to: navigation, search

A few things to know about how the code works. Far from complete.


i18n in FusionForge is done via the standard Gettext library, with no particular quirks in FusionForge. This makes it a bit unwieldy to use custom/local translations or strings. Lolando has a local branch with code to generate a local translation package that can override the official ones. Need to finish it and commit it to trunk.

Database access

Database queries go through the db_query_params() method (db_query() is being deprecated to help get rid of a whole class of potential SQL injection bugs). This is a wrapper around the PostgreSQL database access methods, which passes the variable parts of a query as separate parameters, removing the need for careful escaping and unescaping. To get the full benefits of that, it is important that the query itself be immutable, and all variable parts need to go into separate parameters. For instance, a query counting the groups with a given word in their name or their description should read:

$res = db_query_params ('SELECT count(*) FROM groups WHERE group_name LIKE $1 OR description LIKE $2',
                        array ($word, $word)) ;

Thus, even if $word comes from a malicious user query, it can't do any harm in the database.

Note that this prevents usage of WHERE foo IN (...) constructs if the number of elements in the set is not constant. Fortunately, we can use an alternative way, with the WHERE foo = ANY($1), with the values built with the db_string_array_to_any_clause() or db_int_array_to_any_clause() methods:

$values = array (1, 2, 5, 8) ;
$res = db_query_params ('SELECT foo FROM bar WHERE col = ANY($1)',
                        array (db_int_array_to_any_clause($values))) ;

URLs and links

As described in FusionForge/Suggestions/URL relocation, URLs to pages in the forges should always be generated by the util_make_url() function. This allows to keep the URL scheme in a single point, so that individual pages don't have to know or care whether the forge runs in its own virtualhost, or on SSL, or in a subset of the URL space within a vhost, and so on.

util_make_link() can be used to generate links rather than just URLs, with extra parameters to add attributes to the <a ...> element in the generated HTML. A use case is to add a class for CSS styling.


The authentication uses an MD5 password stored in the database by default, but a hook allows to override that with a plugin.

The permission model is RBAC, role-based access control. 'Users' are members of any number of 'groups'. Each membership of a user in a group has a 'role', possibly shared by several users in a group. Each role is specific to a group (no cross-group sharing currently), and it has a set of 'role settings' which are permission bits for the members of the role on the tools of the group.

There is a global admin "role" given to all users member of the admin project, there is no premission control on this user that can add/remove all permissions he wants

There is a specific admin role given to the project admin, each project creator is by default project admin, he can manage roles for other users, give or remove admin role, design new roles, add/remove users of the group he is admin.

System security

This describes the concepts of FusionForge that shall ensure security of the underlying server. It describes

  • what the different components of FusionForge should do or not do,
  • what exceptions currently exist,
  • and what technical measures are taken to ensure this.


Web server

  • should not write to the filesystem directly
  • if any write access is required, only write to /var/lib/gforge
  • should only read and write the database
  • Technical measures:
    • web server runs as user www-data
    • user www-data has usually no write access anywhere
    • only where required, write access can be granted (explicitly?) by the admin
  • Exceptions: ftp upload, some plugins


  • should write only in /var/lib/gforge
  • Technical measures:
    • cronjob runs as user root
    • most jobs restrict permissions to user gforge
    • user gforge can only write to /var/lib/gforge
  • Exceptions:
    • pluginman:
      • creates links in /opt/gforge/www/plugins
      • creates links in /etc/gforge/plugins
    • configman:
      • modifies /etc/gforge/local.inc



  • contains source code of FF
  • server admin should not have to modify anything here (to simplify updates)
  • modified only by FF developers
  • Exceptions:
    • Links in /opt/gforge/www/plugins are set up by pluginman
    • Some plugins require modifications


  • contains configuration files of FF
  • modified by server admin
  • FF code should not modify anything here (to simplify updates)
  • Exceptions:
    • pluginman creates links here
    • configman modifies local.inc
    • setup autogenerates files here (e.g. httpd.conf)
  • Technical measures: no write permissions for www-data or gforge


  • contains files required by the current state of FF
  • modifed by cronjobs
  • Exceptions:
    • ftp upload
    • mediawiki upload
    • Other plugins?
  • Technical measures:
    • write permissions for user gforge
    • no write permissions for user www-data

Potential problems

  • Root cronjob is potentially dangerous
    • Possible solution: run cronjob as user gforge, use sudo when root access is required
    • More control over root jobs by admin
  • Configman writes central config file that contains paths etc.!


Configuration was previously accessed through a mess of global variables. This is being deprecated in favour of a (hopefully) clean API:

  • forge_get_config(name [, section]): get value of variable "name" in section "section" (defaults to "core");
  • forge_define_config_item(name, section, default): define a new configuration item with given name/section and default value;
  • forge_read_config_file(file): read a *.ini file and inject its contents into the configuration.

See also Development environment.