freeipa/doc/guide/guide.org
Jan Cholasta ab9d1e75fc logging: do not reference loggers in arguments and attributes
Remove logger arguments in all functions and logger attributes in all
objects, with the exception of API object logger, which is now deprecated.
Replace affected logger calls with module-level logger calls.

Reviewed-By: Martin Basti <mbasti@redhat.com>
2017-07-14 15:55:59 +02:00

998 lines
52 KiB
Org Mode

#+OPTIONS: ^:{}
#+EMAIL: abokovoy@redhat.com
#+AUTHOR: Alexander Bokovoy
#+STYLE: <style type="text/css">
#+STYLE: pre {
#+STYLE: border: 1pt solid #000000;
#+STYLE: background-color: #404040;
#+STYLE: color: white;
#+STYLE: }
#+STYLE: .src {width: 940px;}
#+STYLE: dt {width: 400px; margin 25px auto;}
#+STYLE: dd {width: 940px;}
#+STYLE: p {text-align:justify;}
#+STYLE: body {width: 960px;
#+STYLE: margin: 0 auto;
#+STYLE: }
#+STYLE: div#content {margin: 0 10px 0 10px;
#+STYLE: display: inline;
#+STYLE: float: left;
#+STYLE: width: 940px;
#+STYLE: overflow: hidden;}
#+STYLE: </style>
Extending FreeIPA
* Introduction
FreeIPA is an integrated security information management solution. There is a common
framework written in Python to command LDAP server provided by a 389-ds project, certificate
services of a Dogtag project, and a MIT Kerberos server, as well as configuring various other
services typically used to maintain integrity of an enterprise environment, like DNS and
time management (NTP). The framework is written in Python, runs at a server side, and
provides access via command line tools or web-based user interface.
As core parts of the framework are implemented as pluggable modules, it is possible to
extend FreeIPA on multiple levels. This document attempts to present general ideas and
ways to make use of most of extensibility points in FreeIPA.
For information management solutions extensibility could mean multiple things. Information
objects that are managed could be extended themselves or new objects could be added. New
operations on existing objects might become needed or certain aspects of an object should
be hidden in a specific environment. All these tasks may require quite different approaches
to implement.
Following chapters will cover high-level design of FreeIPA and dive into details of its core
framework. Knowledge of Python programming language basics is required. Understanding
LDAP concepts is desirable, though it is not required for simple
extensions as FreeIPA attempts to provide sufficient mapping of LDAP concepts onto less
complex structures and Python objects, lowering a barrier to fine tune FreeIPA for
the specific use cases.
* High level design
FreeIPA core is written in Python programming language. The data is stored in LDAP
database, and client-server paradigm is used for managing it. A FreeIPA server instance
runs its own LDAP database, provided by 389-ds project (formerly Fedora Directory
Server). A single instance of LDAP database corresponds to the single FreeIPA
domain. Access to all information stored in the database is provided via FreeIPA server
core which is run as a simple WSGI application which uses XML-RPC and JSON to exchange
requests with its own clients.
Multiple replicas of the FreeIPA instance can be created on different servers, they are
managed with the help of replication mechanisms of 389-ds directory server.
As LDAP database is used for data storage, LDAP's Access Control Model is used to provide
privilege separation and Kerberos tickets are used to pass-through assertion of
authenticity. As Kerberos server is using the same LDAP database instance, use of Kerberos
tickets allows to perform operations against the database on the server if a client is
capable to forward such tickets via communication channels selected for the operation.
When FreeIPA client connects to FreeIPA server, a Kerberos ticket is forwarded
to the server and operations against LDAP database are performed under identity
authenticated when the ticket was issued. As LDAP database also uses Kerberos to establish
identity of a client, Access Control Information attributes can be used to limit what
entries could be accessed and what operations could be performed.
The approach allows to delegate operations from a FreeIPA client to the FreeIPA server
and in general gives FreeIPA server ability to interact with any Kerberos-aware service on
behalf of the client. It also allows to keep FreeIPA client side implementation relatively
light-weight: all it needs to do is to be able to forward Kerberos ticket, process XML-RPC or
JSON, and present resulting responses to the user.
Besides run-time core, FreeIPA includes few configuration tools. These tools
are split between server and client. Server-side tools are used when an instance of
FreeIPA server is set up and configured, while client-side tools are used to configure client
systems. While the server tools are used to configure LDAP database, put proper schema
definitions in use, create Kerberos domain, Certificate Authority and configure all
corresponding services, client side is more limited to configure PAM/NSS modules to work
against FreeIPA server, and make sure that appropriate information about the client host
is recorded in FreeIPA databases.
* Core plug-in framework
FreeIPA core defines few fundamentals. These are managed objects, their properties, and
methods to apply actions to the objects. Methods, in turn, are commands that are
associated with a specific object. Additionally, there are commands that do not have
directly associated objects and may perform actions over few of those. Objects are stored
using data store represented by a back end, and one of most useful back ends is LDAP store
back end.
Altogether, set of =Object=, =Method=, =Command=, and =Backend= instances
represent application programming interface, API, of FreeIPA core framework.
In Python programming language object oriented support is implemented using a fairly
simple concept that allows to modify instances in place, extending or removing their
properties and methods. While this concept is highly useful, in security-oriented
frameworks ability to lock down and trace origins of changes is also important. FreeIPA core
attempts to implement locking down feature by artificially making instances of foundation
classes read-only after their initialization has happened. If an attempt to modify object
happens after it was locked down, an exception is thrown. There are many classes
following this pattern.
For example, =ipalib.frontend.Command= class is derived from =ipalib.frontend.HasParam= class
that derives from =ipalib.plugable.Plugin= class which, in turn, is derived from
=ipalib.base.ReadOnly= class.
As result, every command has typed parameters and can dynamically be added to the
framework. At the same time, one cannot modify the properties of the command accidentally
once it is instantiated. This protects from modifications and enforces true nature of the
commands: they cannot have state that is carried over across multiple calls to the same
command unless the state is changing globally the whole environment around.
Environment also holds information about the context of execution. The /context/ is
important part of the FreeIPA framework as it also defines which methods of
the command instance are called in order to perform action. /Context/ in itself is defined
by the /environment/ which gives means to catch and store certain information about execution.
As with commands themselves, once instantiated, environment cannot be changed.
By default, for primary FreeIPA use, there are three major contexts defined: server,
client, and installer/updates.
- /server context/ :: plugins are registered and communicate with clients via XML-RPC and JSON
listeners. They validate any arguments and options defined and then execute whatever
action they supposed to perform
- /client context/ :: plugins are used to validate any arguments and options they take and
then forward the request to the FreeIPA server.
- /installer context/, /updates context/ :: plugins specific to installation and update
are loaded and registered. This context can be used to extend possible operations
during set up of FreeIPA server.
A user may define any context they want. FreeIPA names server context as '~server~'. When
using the ~ipa~ command line tool the context is '~cli~'. Server installation tools, in
particular, '~ipa-ldap-updater~', use special '~updates~' context to load specialized
plugins useful during update of the installed FreeIPA server.
Because these utilities use the same framework they will do the same validation, set default
values, and perform other basic actions in all contexts. This can help to save a
round-trip when testing for invalid data. However, for client-server communication, the
server is always authoritative and can re-define what the client has sent.
** Name space
FreeIPA has one special type of read-only objects: =NameSpace=. =NameSpace= class gives an
ordered, immutable mapping object whose values can also be accessed as attributes. A
=NameSpace= instance is constructed from iterable providing its members, which are simply
arbitrary objects with =name= attribute. This attribute must conform to two following
rules:
- Its value must be unique among the members of the name space
- Its value must pass the =check_name()= function =ipalib.base= module.
=check_name()= function encodes a simple rule of a lower-case Python identifier that
neither starts nor ends with an underscore. Actual regular expression that codifies this
rule is =NAME_REGEX= within =ipalib.constants= module.
Once name space is created, it locks itself down and becomes read-only. It means that
while original objects accessed through the name space might change, the references to
them via name space will stay intact. They cannot be removed or changed to point to other
objects.
The name spaces are used widely in FreeIPA core framework. As mentioned earlier, API
includes set of objects, commands, and methods. Objects include properties that are
defined before lock-down. At object's lock-down parameters are placed into a name space
and that locks them down so that no parameter specification can change. Command's
parameters and options also locked down and cannot change once command instance is
instantiated.
** Parameters
=Param= class is used to define attributes, arguments, or options throughout FreeIPA core
framework. The =Param= base class is not used directly but rather sub-classed to define
properties like passwords or specific data types like =Str= or =Int=.
Instances of classes inherited from =Param= base class give uniform access to the
properties required to command line interface, Web UI, and internally to FreeIPA
code. Following properties are most important:
- /name/ :: name of the parameter used internally to address the parameter in Python
code. The /name/ could include special characters to designate a =Param= spec.
- /cli_name/ :: optional name of the parameter to use in command line
interface. FreeIPA's CLI sets a mechanism to automatically translate
from a command line option name to a parameter's /name/ if /cli_name/
is specified.
- /label/ :: A short phrase describing the parameter. It is used on the CLI when
interactively prompting for the values, and as a label for the form inputs
in the Web UI. The /label/ should start with an initial capital letter.
- /doc/ :: A long description of the parameter. It is used by the CLI when displaying the
help information for a command, and as an extra instruction for the form input
on the Web UI. By default the /doc/ is the same as the /label/ but can be
overridden when a =Param= instance is created. As with /label/, /doc/ should
start with an initial capital letter and additionally should not end with any
punctuation.
- /required/ :: If set to =True=, means this parameter is required to supply. All
parameters are required by default and that means that /required/
property should only be specified when parameter *is not required*.
- /multivalue/ :: if set to =True=, means this parameter can accept a Python's tuple of
values. By default all parameters are *single-valued*.
When parameter /name/ has any of ~?~, ~*~, or ~+~ characters, it is treated as parameter
spec and is used to specify whether parameter is required, and should it be
multivalued. Following syntax is used:
| Spec | Name | Required | Multivalue |
|--------+-------+----------+------------|
| 'var' | 'var' | True | False |
| 'var?' | 'var' | False | False |
| 'var*' | 'var' | False | True |
| 'var+' | 'var' | True | True |
Access to the value stored by the =Param= class is given through a callable interface:
#+BEGIN_SRC python
age = Int('age', label='Age', default=100)
print age(10)
#+END_SRC
Following parameter classes are defined and used throughout FreeIPA framework:
- /Bool/ :: boolean parameters that are stored in Python's ~bool~ type, therefore, they
return either ~True~ or ~False~ value. However, they accept ~1~, ~True~
(Python boolean), or Unicode strings '~1~', '~true~' and '~TRUE~' as truth value, and ~0~, ~False~ (Python boolean), or Unicode strings '~0~', '~false~', and '~FALSE~' as false.
- /Flag/ :: boolean parameters which always have default value. Property /default/ can be
used to set the value. Defaults to ~False~:
#+BEGIN_SRC python
verbose = Flag('verbose', default=True)
#+END_SRC
- /Int/ :: integer parameters that are stored in Python's int type. Two additional properties can be
specified when constructing =Int= parameter:
- /minvalue/ :: minimal value that this parameter accepts, defaults to =MININT=
- /maxvalue/ :: maximum value this parameter can accept, defaults to =MAXINT=
- /Decimal/ :: floating point parameters that are stored in Python's Decimal type. =Decimal= has
the same two additional properties as =Int=. Unlike =Int=, there are no
default values for the minimal and maximum boundaries.
- /Bytes/ :: a parameter to represent binary data.
- /Str/ :: parameter representing a Unicode text. Both /Bytes/ and /Str/ parameters accept
following additional properties:
- /minlength/ :: minimal length of the parameter
- /maxlength/ :: maximum length of the parameter
- /length/ :: length of the parameters
- /pattern/ :: regular expression applied to the parameter's value to check its
validness
- /pattern_errmsg/ :: an error message to show when regular expression check fails
- /IA5Str/ :: string parameter as defined by RFC 4517. It means all characters of the
string must be ASCII characters (7-bit).
- /Password/ :: parameter to store passwords in Python =unicode= type. /Password/ has one
additional property:
- /confirm/ :: boolean specifying whether password should be confirmed
when entered. The confirmation is enabled by default.
- /Enum/ :: parameter can have one of predefined values that are specified with /values/
property which is a Python's =tuple=.
For most common case of enumerable strings there are two parameters:
- /BytesEnum/ :: parameter value should be one of predefined =unicode= strings
- /StrEnum/ :: equivalent to /BytesEnum/. Originally /BytesEnum/ was stored in Python's =str= class instances but to be aligned with Python 3.0 changes both
classes moved to store as =unicode=.
When more than one value should be accepted, there is /List/ parameter that allows to
provide list of strings separated by a separator, default to ','. Also, the /List/
parameter skips spaces before the next item in the list unless property /skipspace/ is set to False:
#+BEGIN_SRC python
names = List('names', separator=',', skipspace=True)
names_list = names(u'John Doe, John Lee, Brad Moe')
# names_list is (u'John Doe', u'John Lee', u'Brad Moe')
names = List('names', separator=',', skipspace=False)
names_list = names(u'John Doe, John Lee, Brad Moe')
# names_list is (u'John Doe', u' John Lee', u' Brad Moe')
#+END_SRC
** Objects
The data manipulated by FreeIPA is represented by an Object class instances. Instance of
an Object class is a collection of properties, accepted parameters, action methods, and a
reference to where this object's data is preserved. Each object also has a reference to a
property that represents a primary key for retrieving the object.
In addition to properties and parameters, Object class instances hold their labels to use
in user interfaces. In practice, there are few differences in how labels are presented
depending on whether it is command line interface or a Web UI, but they can be ignored at
this point.
To be useful, all Object sub-classes need to override =takes_param= property. This is
where most of flexibility of FreeIPA comes from.
*** takes_param attribute
Properties of every object derived from Object class can be specified manually but FreeIPA
gives a handy mechanism to perform descriptive specification. Each =Object= class has
=Object.takes_param= attribute which defines a specification of all parameters this object
type is accepting.
Next example shows how to create new object type. We create an aquarium tank by defining
its dimensions and specifying which fish is living there.
#+BEGIN_SRC python -n -r -l '(%s)'
from ipalib import api, Object
class tank(Object):
takes_params = (
StrEnum('species*', label=u'Species', doc=u'Fish species',
values=(u'Angelfish', u'Betta', u'Cichlid', u'Firemouth')),
Decimal('height', label=u'Height', doc=u'height in mm', default='400.0'),
Decimal('width', label=u'Width', doc=u'width in mm', default='400.0'),
Decimal('depth', label=u'Depth', doc=u'Depth in mm', default='300.0')
)
api.register(tank) (ref:register)
api.finalize() (ref:finalize)
print list(api.Object.tank.params)
# ['species', 'height', 'width', 'depth']
#+END_SRC
First we define new class, =tank=, that takes four parameters. On line [[(register)]] we register the class
in FreeIPA's API instance, api. This creates =tank= object in =api.Object= name
space. Many objects can be added into the API up until =api.finalize()= is called as we do
on line [[(finalize)]].
When =api.finalize()= is called, all name spaces are locked down and all registered Python
objects in those name spaces are also finalized which in turn locks their structure down
as well.
As result, once we have finalized our API instance, every registered Object can be
accessed through =api.Object.<name>=. Our aquarium tank object now has defined =params=
attribute which is a name space holding all =Param= instances. Thus we can introspect and
see which parameters this object has.
At this point we can't do anything reasonable with our aquarium tank yet because we
haven't defined methods to handle it. In addition, our object isn't very useful as it does
not know how to store the information about aquarium's dimensions and species living in
it.
*** Object methods
Methods perform actions on the associated objects. The association of methods and objects
is done through naming convention rather than using programming language features. FreeIPA
expects methods operating on an object =<name>= to be named =<name>_<action>=:
#+BEGIN_SRC python
class tank_create(Method):
def execute(self, **options):
# create new aquarium tank
api.register(tank_create)
class tank_populate(Method):
def execute(self, **options):
# populate the aquarium tank with fish
api.register(tank_populate)
#+END_SRC
As can be seen, each method is a separate Python class. This approach allows to maintain
complexity of methods isolated from each other and from the complexity of the objects and
their storage which is probably most important aspect due to LDAP complexity overall.
The linking between objects and their methods goes further. All parameters defined for an
object, may be used as arguments of the methods without explicit declaration. This means
=api.Method.tank_populate= will accept ~species~ argument.
*** Methods with storage back ends
In order to store the information, =Object= class instances require a back end. FreeIPA
defines several back ends but the ones that could store data are derived of
=ipalib.CrudBackend=. CRUD, or /Create/, /Retrieve/, /Update/, and /Delete/, are basic
operations that could be performed with corresponding objects. =ipalib.crud.CrudBackend=
is an abstract class, it only defines functions that should be overridden in classes that
actually implement the back end operations.
As back end is not used directly, FreeIPA defines methods that could use back end and
operate on object's defined by certain criteria. Each method is defined as a separate
Python class. As CRUD acronym suggests, there are four base operations:
=ipalib.crud.Create=, =ipalib.crud.Retrieve=, =ipalib.crud.Update=,
=ipalib.crud.Delete=. In addition, method =ipalib.crud.Search= allows to retrieve all
entries that match a given search criteria.
When objects are defined and the back end is known, methods can be used to manipulate
information stored by the back end. Most of useful operations combine some of CRUD base
operations to perform their tasks.
In order to support flexible way to extend methods, FreeIPA gives special treatment for
the LDAP back end. Methods using LDAP back end hide complexity of handling LDAP queries and
allow to register user-provided functions that are called before or after method. This
mechanism is defined by ipalib.plugins.baseldap.CallbackInterface and used by LDAP-aware
CRUD classes, =LDAPCreate=, =LDAPRetrieve=, =LDAPUpdate=, =LDAPDelete=, and an analogue to
=ipalib.crud.Search=, =LDAPSearch=. There are also classes that define methods to operate
on reverse relationships between objects in LDAP to allow addition or removal of
membership information both in forward and reverse directions: =LDAPAddMember=,
=LDAPModMember=, =LDAPRemoveMember=, =LDAPAddReverseMember=, =LDAPModReverseMember=, =LDAPRemoveReverseMember=.
Most of CRUD classes are based on a =LDAPQuery= class which generalizes concept of
querying a record addressed with a primary key and supports JSON marshalling of the
queried attributes and their values.
Base LDAP operation classes implement everything needed to create typical methods to
work with self-contained objects stored in LDAP.
*** LDAPObject class
A large class of objects is LDAPObject. LDAPObject instances represent entries stored in
FreeIPA LDAP database instance. They are referenced by their distinguished name, DN, and
able to represent complex relationships between entries in LDAP like direct and indirect
membership.
Any class derived from LDAPObject needs to re-define few properties so that base class can
properly function for the specific object that is defined by the class. Below are commonly
redefined properties:
- /container_dn/ :: DN of the container for this object entries in LDAP. This one
usually comes from the environment associated with the API and by default is populated
from the =DEFAULT_CONFIG= of =ipalibs.constants=. For example, all accounts are
stored under =cn=accounts=, with users are under =cn=users,cn=accounts= and groups
are under =cn=groups,cn=accounts=. In case of a new object added, it
is reasonable to select its container coordinated to default configuration.
- /object_class/ :: list of LDAP object classes associated with the object
- /search_attributes/ :: list of attributes that will be used for search
- /default_attributes/ :: list of attributes that are always returned by searches
- /uuid_attribute/ :: an attribute that defines uniqueness of the entry
- /attribute_members/ :: a dict defining relations between other objects and this
one. Key is the name of attribute and value is a list of objects this attribute may
refer to. For example, =host= object defines that =memberof= attribute of a
host may refer to a =hostgroup=, =netgroup=, =role=, =hbacrule=, or =sudorule=
object. In other words, it means that =host= could be a member of any of those
objects.
- /reverse_members/ :: a dict defining reverse relations between this object and other
objects. Key is the name of attribute and value is the name of an object that refers
to this object with the attribute. For example, =role= object defines that =member=
attribute of a =privilege= refers to a =role= object.
- /password_attributes/ :: list of pairs defining an attribute in LDAP and a property of
a Python dictionary representing the LDAP object attributes that will be set
accordingly if such attribute exists in the LDAP entry. As passwords have restricted
access, often one needs only to know that there is a password set on the entry to
perform additional processing.
- /relationships/ :: a dict defining existing relationship criteria associated with the
object. These are used in Web UI to allow filtering of objects by the criteria. The
value is defined as a tuple of an UI label and two prefixes: inclusive and exclusive
that are prepended to the attribute parameter when options are generated by the
framework. LDAPObject defines few default criteria: /member/, /memberof/, /memberindirect/, /memberofindirect/, and objects can redefine or append more. Due
to regularity of the design of LDAP objects, default criteria already makes it
possible to apply searches almost uniformly: one can ask for membership of a user in
a group, as well as for a membership of a role in a privilege without explicitly
defining those relationships.
These properties define how translation would go from Python side to and from an LDAP
backend.
As an example, let's see how role is defined. This is fully functioning plugin that
provides operations on roles:
#+INCLUDE "role.py.txt" src python -n
* Extending existing object
As said earlier, until API instance is finalized, objects, methods, and commands can be
added, removed, or modified freely. This allows to extend existing objects. Before API is
finalized, we cannot address objects through the unified interface as =api.Object.foo=,
but for almost all cases an object named =foo= is defined in a plugin
=ipalib.plugins.foo=.
1. Add new parameter:
#+BEGIN_SRC python -n
from ipalib.plugins.user import user
from ipalib import Str, _
user.takes_params += (
Str('foo',
cli_name='foo',
label=_('Foo'),
),
)
#+END_SRC
2. Re-define User object label to use organisation-specific terminology in Web UI:
#+BEGIN_SRC python -n
from ipalib.plugins.user import user
from ipalib import text
_ = text.GettextFactory(domain='extend-ipa')
user.label = _('Staff')
user.label_singular = _('Engineer')
#+END_SRC
Note that we re-defined locally =_= method to use different ~GettextFactory~. As
GettextFactory is supporting a single translation domain, all new translation terms need
to be placed in a separate translation domain and referred accordingly. Python rules for
scoping will keep this symbol as ~<package>._~ and as nobody imports it explicitly, it
will not interfere with the framework's provided ~text._~.
3. Assume =/dev/null= as default shell for all new users:
#+BEGIN_SRC python -n -r
from ipalib.plugins.user import user_add
def override_default_shell_cb(self, ldap, dn.
entry_attrs, attrs_list,
*keys, **options):
if 'loginshell' in entry_attrs:
default_shell = [self.api.Object.user.params['loginshell'].default]
if entry_attrs['loginshell'] == default_shell:
entry_attrs['loginshell'] = [u'/dev/null']
user_add.register_pre_callback(override_default_shell_cb)
#+END_SRC
The last example exploits a powerful feature available for every method of LDAPObject:
registered callbacks.
* Extending existing method
For objects stored in LDAP database instance all methods support adding callbacks. A
/callback/ is a user-provided function that is called at certain point of execution of a
method.
There are four types of callbacks:
- /PRE callback/ :: called before executing the method's action. Allows to modify passed
arguments, do additional validation or data transformation and
specific access control beyond what is provided by the framework.
- /POST callback/ :: called after executing the method's action. Allows to analyze results
of the action and perform additional actions or modify output.
- /EXC callback/ :: called in case execution of the method's action caused an execution
error. These callbacks provide means to recover from an erroneous execution.
- /INTERACTIVE callback/ :: called at a client context to allow a command to decide if
additional parameters should be requested from an user. This mechanism especially
useful to simplify complex interaction when there are several levels of possible
scenarios depending on what was provided at a client side.
All callback types are available to any class derived from =CallbackInterface=
class. These include all LDAP-based CRUD methods.
Callback registration methods accept a reference to callable and optionally ordering
argument =first= (~False~ by default) to allow the callback be executed before previously
registered callbacks of this type.
=CallbackInterface= class provides following class methods:
- =register_pre_callback= :: registers /PRE/ callback
- =register_post_callback= :: registers /POST/ callback
- =register_exc_callback= :: registers /EXC/ callback for purpose of recovering from
execution errors
- =register_interactive_prompt_callback= :: registers callbacks called by the client
context.
Let's look again at the last example:
#+BEGIN_SRC python -n -r
from ipalib.plugins.user import user_add
def override_default_shell_cb(self, ldap, dn.
entry_attrs, attrs_list,
*keys, **options):
if 'loginshell' in entry_attrs:
default_shell = [self.api.Object.user.params['loginshell'].default]
if entry_attrs['loginshell'] == default_shell:
entry_attrs['loginshell'] = [u'/dev/null']
user_add.register_pre_callback(override_default_shell_cb)
#+END_SRC
This extension defines a pre-processing callback that accepts number of arguments:
- /ldap/ :: reference to the back end to store and retrieve the object's data
- /dn/ :: reference to the object data in LDAP
- /entry_attrs/ :: arguments and options of the command and their values as a
dictionary. All values in /entry_attrs/ will be used for communicating
with LDAP store, thus replacing values should be done with care. For
details please see Python LDAP module documentation
- /attrs_list/ :: list of all attributes we intend to fetch from the back end
- /keys/ :: arguments of the command
- /options/ :: all other unidentified parameters passed to the method
Arguments of a post-processing callback, /POST/, are slightly different. As action is
already performed and the attributes of the entry are fetched back from the back end,
there is no need to provide =attrs_list=:
#+BEGIN_SRC python -n -r
from ipalib.plugins.user import user_add
def verify_shell_cb(self, ldap, dn. entry_attrs,
*keys, **options):
if 'loginshell' in entry_attrs:
default_shell = [self.api.Object.user.params['loginshell'].default]
if entry_attrs['loginshell'] == default_shell:
# report that default shell is assigned
user_add.register_post_callback(verify_shell_cb)
#+END_SRC
Execution error callback, /EXC/, has following signature:
#+BEGIN_SRC python -n
def user_add_error_cb(self, args, options, exc,
call_func, *call_args, **call_kwargs):
return
#+END_SRC
where arguments have following meaning:
- /args/ :: arguments of the original method
- /options/ :: options of the original method
- /exc/ :: exception object thrown by a /call_func/
- /call_func/ :: function that was called by the method and caused the error of
execution. In case of LDAP-based methods this is often =ldap.add_entry()=
or =ldap.modify_entry()=, or a similar function
- /call_args/ :: first argument passed to the /call_func/
- /call_kwargs/ :: remaining arguments of /call_func/
Finally, interactive prompt callback receives /kw/ argument which is a dictionary of all
arguments of the command.
All callbacks are supplied with a reference to the method instance, ~self~, unless the
callback itself has an attribute called '~im_self~'. As can be seen in callback examples,
self reference recursively provides access to the whole FreeIPA API structure.
This approach gives complete control of existing FreeIPA methods without
deep dive into details of LDAP programming even if the framework allows such a deep dive.
* Web UI
FreeIPA framework has two major client applications: Web UI and command line-based client
tool, ~ipa~. Web UI communicates with a FreeIPA server running WSGI application that
accepts JSON-formatted requests and translates them to calls to FreeIPA plugins.
A following code in ~install/share/ui/wsgi.py~ defines FreeIPA web application:
#+INCLUDE "wsgi.py.txt" src python -n -r
At line [[(wsgi-app-bootstrap)]] we set up FreeIPA framework with server context. This means
plugins are loaded and initialized from following locations:
- ~ipalib/plugins/~ -- general FreeIPA plugins, available for all contexts
- ~ipaserver/plugins/~ -- server-specific plugins, available in '~server~' context
With =api.finalize()= call at line [[(wsgi-app-finalize)]] FreeIPA framework is locked down and all
components provided by plugins are registered at ~api~ name spaces: =api.Object=,
=api.Method=, =api.Command=, =api.Backend=.
At this point, ~api~ name spaces become usable and our WSGI entry point, defined on lines
[[(wsgi-app-start)]] to [[(wsgi-app-end)]] can access =api.Backend.session()= to generate
response for WSGI request.
Web UI itself is written in JavaScript and utilizes JQuery framework. It can be split into
three major parts:
- /communication/ :: tools defined in ~ipa.js~ to allow talking with FreeIPA server using
AJAX requests and JSON formatting
- /presentation/ :: tools in ~facet.js~, ~entity.js~, ~search.js~, ~widget.js~, ~add.js~,
and ~details.js~ to give basic building blocks of Web UI
- /objects/ :: actual implementation of Web UI for FreeIPA objects (user, group, host,
rule, and other available objects registered at =api.Object= by the server
side)
The code of these JavaScript files is loaded in ~index.html~ and kicked into work by
~webui.js~ where main navigation and document's ~onready~ event handler are defined. In
addition, ~index.html~ imports ~extension.js~ file where all extensions to Web UI can be
registered or referenced. As ~extension.js~ is loaded after all other Web UI JavaScript
files but before ~webui.js~, it can already use all tools of the Web UI.
The execution of Web UI starts with the call of =IPA.init()= function which does
following:
1. Set up AJAX asynchronous communication via POST method using JSON format.
2. Fetches meta-data about FreeIPA methods available on the server using JSON format and
makes them available as =IPA.methods=.
3. Fetches meta-data about FreeIPA objects available on the server using JSON format and
makes them available as =IPA.objects=.
4. Fetches translations of messages used in the Web UI and makes them available as
=IPA.messages=.
5. Fetches identity of the user running the Web UI, accessible as =IPA.whoami=.
6. Fetches FreeIPA environment specific for Web UI, accessible as =IPA.env=.
The communication with FreeIPA server is done using =IPA.command()= function. Commands
created with =IPA.command()= can later be executed with =execute()= method. This
separation of construction and actual execution allows to create multiple commands and
combine them together in a single request. Batch requests are created with
=IPA.batch_command()= function and command are added to them with =add_command()=
method. In addition, FreeIPA Web UI allows to run commands concurrently with
=IPA.concurrent_command()= function.
Web UI has following DOM structure:
|-----------------------+-----------------------------------+------------+-----------|
| | Container | | |
|-----------------------+-----------------------------------+------------+-----------|
| background | header | navigation | content |
| background-header | header-logo | | |
| background-navigation | header-network-activity-indicator | | |
| background-left | loggedinas | | |
| background-right | | | |
|-----------------------+-----------------------------------+------------+-----------|
~Container~ div is a top-level one, it includes background, header, navigation, content
divs. These divs and their parts can be manipulated from the JavaScript code to represent
the UI. However, FreeIPA gives an easier way to accomplish this.
** Facets
Facet is a smallest block of FreeIPA Web UI. When facet is defined, it has name, label,
link to an entity it is part of, and methods to create, show, load, and hide itself.
** Entities
Entity is addressable group of facets. FreeIPA Web UI provides a declarative way of
creating entities and defining their facets based on JavaScript's syntax. Following
example is a complete definition of a netgroup facet:
#+INCLUDE "netgroup.js" src js2-mode -n
This definition of a netgroup facet describes:
- /details facet/ :: a facet named '~identity~' and three fields, ~cn~, ~description~,
and ~nisdomainname~. In addition, ~description~ field is a text area widget. This
facet is used to display existing netgroup information.
- /association facets/ :: number of facets, linking this one with others. In case of a
netgroup, netgroups are linked to facet group ~member~ via different attributes. The
definition also adds standard association facets defined in ~entity.js~.
- /adder dialog/ :: a dialog to create a new netgroup. The dialog has two fields: ~cn~ and ~description~ where ~description~ is again a text area widget.
Similarly to FreeIPA core framework, created entity needs to be registered to the Web UI
via =IPA.register()= method.
In order to add new entity to the Web UI, one can use ~extension.js~. This file in
~/usr/share/ipa/html~ is empty and provided specifically for this purpose.
As an example, let's define an entity 'Tank' corresponding to our aquarium tank:
#+BEGIN_SRC js2-mode -n
IPA.tank = {};
IPA.tank.entity = function(spec) {
var that = IPA.entity(spec);
that.init = function(params) {
details_facet({
sections: [
{
name: 'identity',
fields: [
'species', 'height', 'width', 'depth'
]
}
]
}).
standard_association_facets().
adder_dialog({
fields: [
'species', 'height', 'width', 'depth'
]
});
};
};
IPA.register('tank', IPA.tank.entity);
#+END_SRC
* Command line tools
As an alternative to Web UI, FreeIPA server can be controlled via command-line interface
provided by the ~ipa~ utility. This utility is operating under '~client~' context and
looks even simpler than Web UI's ~wsgi.py~:
#+BEGIN_SRC python -n
import sys
from ipalib import api, cli
if __name__ == '__main__':
cli.run(api)
#+END_SRC
=cli.run()= is the central running point defined in ~ipalib/cli.py~:
#+BEGIN_SRC python -n
# <cli.py code> ....
cli_plugins = (
cli,
textui,
console,
help,
show_mappings,
)
def run(api):
error = None
try:
(_options, argv) = api.bootstrap_with_global_options(context='cli')
for klass in cli_plugins:
api.add_plugin(klass)
api.finalize()
if not 'config_loaded' in api.env and not 'help' in argv:
raise NotConfiguredError()
sys.exit(api.Backend.cli.run(argv))
except KeyboardInterrupt:
print('')
logger.info('operation aborted')
except PublicError as e:
error = e
except Exception as e:
logger.exception('%s: %s', e.__class__.__name__, str(e))
error = InternalError()
if error is not None:
assert isinstance(error, PublicError)
logger.error(error.strerror)
sys.exit(error.rval)
#+END_SRC
As with WSGI, =api= is bootstraped, though with a client context and using global options
from ~/etc/ipa/default.conf~, and command line arguments. In addition to common plugins
available in ~ipalib/plugins~, ~cli.py~ adds few command-line specific classes defined in
the module itself:
- ~cli~ :: a backend for executing from command line interface which does translation of
command line option names, basic verification of commands and fallback to show
help messages with ~help~ command, execution of the command, and translation of
the output to command-line friendly format if this is defined for the command.
- ~textui~ :: a backend to nicely format output to stdout which handles conversion from
binary to base64, prints text word-wrapped to the terminal width, formats
returned complex values so that they can be easily understood by a human
being.
#+BEGIN_EXAMPLE
>>> entry = {'name' : u'Test example', 'age' : u'100'}
>>> api.Backend.textui.print_entry(entry)
age: 100
name: Test example
#+END_EXAMPLE
- ~console~ :: starts interactive Python console with FreeIPA commands
- ~help~ :: generates help for every command and method of FreeIPA and structures it into
sections according to the registered FreeIPA objects.
#+BEGIN_EXAMPLE
>>> api.Command.help(u'user-show')
Purpose: Display information about a user.
Usage: ipa [global-options] user-show LOGIN [options]
Options:
-h, --help show this help message and exit
--rights Display the access rights of this entry (requires --all). See
ipa man page for details.
--all Retrieve and print all attributes from the server. Affects
command output.
--raw Print entries as stored on the server. Only affects output
format.
#+END_EXAMPLE
- ~show_mappings~ :: displays mappings between command's parameters and LDAP attributes:
#+BEGIN_EXAMPLE
>>> api.Command.show_mappings(command_name=u"role-find")
Parameter : LDAP attribute
========= : ==============
name : cn
desc : description
timelimit : timelimit?
sizelimit : sizelimit?
#+END_EXAMPLE
** Extending command line utility
Since ~ipa~ utility operates under client context, it loads all command plugins from
~ipalib/plugins~. A simple way to extend command line is to drop its plugin file into
~ipalib/plugins~ on the machine where ~ipa~ utility is executed. Next time ~ipa~ is
started, new plugin will be loaded together with all other plugins from ~ipalib/plugins~
and commands provided by it will be added to the =api=.
Let's add a command line plugin that allows to ping a server and measures round trip time:
#+BEGIN_SRC python -n
from ipalib import frontend
from ipalib import output
from ipalib import _, ngettext
from ipalib import api
import time
__doc__ = _("""
Local extensions to FreeIPA commands
""")
class timed_ping(frontend.Command):
__doc__ = _('Ping remote FreeIPA server and measure round-trip')
has_output = (
output.summary,
)
def run(self):
t1 = time.time()
result = self.api.Command.ping()
t2 = time.time()
summary = u"""Round-trip to the server is %f ms.
Server response is %s"""
return dict(summary=summary % ((t2-t1)*1000.0, result['summary']))
api.register(timed_ping)
#+END_SRC
When this plugin code is placed into ~ipalib/plugins/extend-cli.py~ (name of the plugin
file can be set arbitrarily), ~ipa timed-ping~ will produce following output:
#+BEGIN_EXAMPLE
$ ipa timed-ping
-----------------------------------------------------------------------------
Round-trip to the server is 286.306143 ms.
Server response is IPA server version 2.1.3GIT8a254ca. API version 2.13
-----------------------------------------------------------------------------
#+END_EXAMPLE
In this example we have created ~timed-ping~ command and overrode its =run()=
method. Effectively, this command will only work properly on the client. If the client is
also FreeIPA server (all FreeIPA servers are enrolled as FreeIPA clients), the same code
will also be loaded by the server context and will be accessible to the Web UI as well,
albeit its usefulness will be questionable as it will be measuring the round-trip to the
server from the server itself.
* File paths
Finally, it should be noted that depending on installed Python version and operating
system, paths where plugins are loaded from may differ. Usually Python extensions are
placed in ~site-packages~ Python sub-directory. In Fedora and RHEL distributions, this is
~/usr/lib/python<version>/site-packages~. Thus, full path to ~extend-cli.py~ would be
~/usr/lib/python<version>/site-packages/ipalib/plugins/extend-cli.py~.
On recent Fedora distribution, following paths are used:
|--------------------+---------------------------+------------------------------------------------------------|
| Plugins | Python module prefix | File path |
|--------------------+---------------------------+------------------------------------------------------------|
| common | ipalib/plugins | /usr/lib/python2.7/site-packages/ipalib/plugins |
| server | ipaserver/plugins | /usr/lib/python2.7/site-packages/ipaserver/plugins |
| installer, updates | ipaserver/install/plugins | /usr/lib/python2.7/site-packages/ipaserver/install/plugins |
|--------------------+---------------------------+------------------------------------------------------------|
Next table explains use of contexts in FreeIPA applications:
|---------+------------------+-------------------------+----------------------------------------|
| Context | Application | Plugins | Description |
|---------+------------------+-------------------------+----------------------------------------|
| server | wsgi.py | common, server | Main FreeIPA server, server context |
| cli | ipa | common | Command line interface, client context |
| updates | ipa-ldap-updater | common, server, updates | LDAP schema updater |
|---------+------------------+-------------------------+----------------------------------------|
* Platform portability
Originally FreeIPA was created utilizing packages available in Fedora and RHEL
distributions. During configuration stages multiple system services need to be stopped
and started again, scheduled to start after reboot and re-configured. In addition, when
operating system utilizing security measures to harden the server setup, appropriate
activities need to be done as well for preserving proper security contexts. As
configuration details, service names, security features and management tools differ
substantially between various GNU/Linux distributions and other operating systems, porting
FreeIPA project's code to other environment has proven to be problematic.
When Fedora project has decided to migrate to systemd for services management, FreeIPA
packages for Fedora needed to be updated as well, at the same time preserving support for
older SystemV initialization scheme used in older releases. This prompted to develop a
'platformization' support allowing to abstract services management between different
platforms.
FreeIPA 2.1.3 includes first cut of platformization work to support Fedora 16 distribution
based on systemd. At the same time, there is an effort to port FreeIPA client side code to
Ubuntu distributions.
Platform portability in FreeIPA means centralization of code to manage system-provided
services, authentication setup, and means to manage security context and host names. It is
going to be extended in future to cover other areas as well, both client- and server-side.
The code that implements platform-specific adaptation is placed under
~ipaplatform~. As of FreeIPA 4.4.2, there are two major "platforms" supported:
- /rhel/ :: Red Hat Enterprise Linux 7-based distributions utilizing Systemd
such as CentOS 7 and Scientific Linux 7.
- /fedora/ :: Fedora distribution version 23 above are supported by this platform
module. It is based on ~systemd~ system management tool and utilizes
common code in ~ipaplatform/base/services.py~. ~fedora~ contains
only differentiation required to cover Fedora 23-specific implementation
of systemd use, depending on changes to Dogtag, Tomcat6, and 389-ds
packages.
Each platform-specific adaptation should provide few basic building blocks:
*** AuthConfig class and tasks module
=ipaplatform.tasks= module implements system-independent interface to configure system
resources. In Red Hat systems some of these tasks are done with authconfig(8) utility.
=AuthConfig= class is nothing more than a tool to gather configuration options and execute
their processing. These options then converted by an actual implementation to series of a
system calls to appropriate utilities performing real configuration.
From FreeIPA code perspective, the system configuration should be done with
use of ~ipaplatform.tasks.tasks~:
#+BEGIN_SRC python -n
from ipaplatform.tasks import tasks
tasks.set_nisdomain('nisdomain.example')
#+END_SRC
The actual implementation can differ. ~redhat~ platform module builds up arguments to
authconfig(8) tool and on =execute()= method runs it with those arguments. Other systems
will need to have processing based on their respective tools.
*** PlatformService class
=PlatformService= class abstracts out an external process running on the system which is
possible to administer: start, stop, check its status, schedule for automatic startup,
etc.
Services are used thoroughly through FreeIPA server and client install tools. There are
several services that are used especially often and they are selected to be accessible via
Python properties of =ipaplatform.services.knownservices= instance.
To facilitate more expressive way of working with often used services, ipaplatform.services
module provides a shortcut to access them by name via
ipaplatform.services.knownservices.<service>. A typical code change looks like this:
#+BEGIN_EXAMPLE
import ipaplatform.services.knownservices
....
- service.restart("dirsrv")
- service.restart("krb5kdc")
- service.restart("httpd")
+ ipaplatform.services.knownservices.dirsrv.restart()
+ ipaplatform.services.knownservices.krb5kdc.restart()
+ ipaplatform.services.knownservices.httpd.restart()
#+END_EXAMPLE
Besides expression change this also makes more explicit to platform providers access to
what services they have to implement. Service names are defined in
ipaplatform.platform.base.wellknownservices and represent definitive names to access these
services from FreeIPA code. Of course, platform provider should remap those names to
platform-specific ones -- for ipaplatform.redhat provider mapping is identity.
Porting to a new platform may be hard as can be witnessed by this example:
https://www.redhat.com/archives/freeipa-devel/2011-September/msg00408.html
If there is doubt, always consult existing providers. ~redhat/services.py~ is canonical -- it
represents the code which was used throughout FreeIPA v2 development.
*** Enabling new platform provider
When support for new platform is implemented and appropriate provider is placed to
~ipaplatform/platform/~, it is time to enable its use by the FreeIPA. Since FreeIPA is
supposed to be rolled out uniformly on multiple clients and servers, best approach is to
build and distribute software packages using platform-provided package management tools.
With this in mind, platform code selection in FreeIPA is static and run at package
production time. In order to select proper platform provider, one needs to pass
~--with-ipaplatform~ argument to FreeIPA's configure process:
#+BEGIN_EXAMPLE
./configure --with-ipaplatform=fedora
#+END_EXAMPLE