sync: xmpp pubsub, multi-user chat
[skm-ma-ws1314.git] / sec-xmpp.tex
1 \subsection{XMPP}
2 \pages{3-4}
3
4 The \term{Extensible Messaging and Presence Protocol (XMPP)} is a distributed,
5 XML-based protocol for real-time communication. Its core functionalities are
6 specified in RFCs~6120~\cite{rfc6120} and RFC~6122~\cite{rfc6121}, while protocol
7 extensions are usually defined by the XMPP community in \term{XMPP Extension
8 Proposals (XEPs)}.
9
10 \subsubsection{Addressing}
11
12 Every user account in XMPP is addressed by a globally unique identifier, called
13 the \term{Jabber ID (JID)}~\cite{rfc6122}. It has the form
14 \code{localpart@domainpart/resource}, where \code{domainpart} is the DNS name of
15 an XMPP server, and \code{localpart} is the name of a user account on that
16 server. Since a user can be logged in from multiple clients, the \code{resource}
17 part is a string chosen by the user to distinguish those clients. Only the part
18 \code{localpart@domainpart} (the \term{bare JID}) is needed to identify a user,
19 the resource is only needed for routing between client and server.
20
21 \subsubsection{Architecture}
22 \begin{wrapfigure}{r}{0.4\textwidth}
23 %\begin{figure}[htop]
24 \centering
25 \includegraphics[width=0.4\textwidth]{fig-xmpp-architecture-mock.jpg}
26 \caption{XMPP architecture}
27 \label{fig:xmpparch}
28 %\end{figure}
29 \end{wrapfigure}
30
31 The original architecture underlying XMPP strongly leans on the established
32 design of Internet Mail, and an example is depicted in Fig.~\ref{fig:xmpparch}.
33 The distributed network is formed by \term{XMPP servers} on one hand, which make
34 up the always-on backbone of the network used for routing message, and manage
35 user accounts and statuses. On the other hand, \term{XMPP clients} represent a
36 single logged-in user and are the interface for communication with other users.
37
38 Every client communicates only with the server that manages the respective user
39 account which is configured in the client, as given in the user's JID. The
40 server then routes the messages to their recipients, using the JID to determine
41 the correct server for a message to be sent to. Finally, the receiving server
42 sends the message to a client where the receiving JID is logged in. If the user
43 is not logged in at the time the message is sent, the server can store it for
44 the user and deliver it on the next login.
45
46 XMPP strongly relies on DNS Service Discovery (see Section~\ref{sec:dnssd}) to
47 determine the server being in charge of a domain. For example, the server who
48 manages the users for the domain \code{example.org} is given by the SRV record
49 \code{\_xmpp-server.\_tcp.example.org}.
50
51 \subsubsection{Communication primitives}
52
53 All communication over XMPP is based on XML. To minimize communication overhead,
54 only fragments of XML, called \term{stanzas}, are sent between hosts. A stanza
55 is always well-formed as a whole; it consist of a root element, which also
56 includes routing attributes (\code{to} and \code{from}), and its optional child
57 elements.
58
59 On top of that, living connections between hosts are represented by \term{XML
60 streams}. The client initiates a connection by sending an XML declaration
61 followed by an opening \code{<stream>} tag. The server then responds also with
62 an opening \code{<stream>} tag. The client then performs SASL authentication and
63 binds its stream to a resource for proper addressing. If this process succeeded,
64 both client and server can send an unlimited number of stanzas, until the
65 connection is closed by one side by sending an closing \code{</stream>} tag. The
66 other side then has the chance to send all outstanding stanzas and likewise
67 closes its stream. If both streams are closed, the underlying TCP connection is
68 terminated.
69
70 \todo[Example stream]
71
72 \subsubsection{Publish/Subscribe and Presence}
73
74 Typically, a user wants to chat with a more or less fixed set of other users,
75 whose JIDs she needs to know, so she needs some kind of ``address book'' that
76 remembers the JIDs for her. In XMPP, this is address book is called
77 \term{roster}, and it also shows the users' willingness to chat (``presence'').
78 In order to see their chat status (which can be one of ``online'', ``offline'',
79 and several ``away'' or ``do not disturb'' states), a user needs to subscribe to
80 the other user's status. The mechanism behind this is called
81 \term{Publish-Subscribe} and is specified in XEP-0060~\cite{xep-0060}. It can
82 be used to notify interested users about changes in personal information, and
83 implements the classic Observer pattern.
84
85 A user publishes information by creating a \term{node} on the XMPP server, which
86 acts as a handle for the data. Interested users can then query the server for
87 nodes, and request subscription to them. When the owner of the node confirms the
88 subscription request, subscribers get notified whenever the owner updates the
89 respective node.
90
91 All communication takes place between the client and the server over \code{<iq>}
92 (``information query'') stanzas.
93
94 \subsubsection{Multi-User Chats}
95
96 Besides one-to-one messaging, XMPP also allows users to create multi-user
97 chat rooms, which is specified in \cite{xep-0045}. Each chat room is given a
98 unique JID to which the users send their messages to. Each incoming message is
99 then dispatched to all users which have joined the room.
100
101 To join a room, the user sends a \code{<presence>} stanza to the room JID, where
102 the resource part of the room JID specifies the desired nick name.
103
104 \subsubsection{XMPP Serverless Messaging}\label{sec:xsm}
105 \pages{1}
106
107 To overcome the need for a central server and authentication, XMPP Serverless
108 Messaging~\cite{xep-0174} allows XMPP clients on a network to build a
109 peer-to-peer mesh network and chat directly with each other. This feature was
110 first introduced by Apple as part of their \term{Bonjour} project, and nowadays
111 it is also available in many other XMPP clients.
112
113 With XMPP Serverless Messaging, XMPP clients simply open a port on the host, and
114 then rely on mDNS and DNS-SD (see Section~\ref{sec:dns})
115 to publish instance names in the domain \code{\_presence.\_tcp.local}. For
116 example, if Juliet uses her machine (named \code{capulet}) with serverless
117 messaging, her client would publish the following four mDNS records:
118
119 \begin{itemize}
120 \item an A record \code{capulet.local}, specifying her IP address,
121 \item an SRV record \code{juliet@capulet.\_presence.\_tcp.local}, specifying
122 the port on which her XMPP client listens, and refering to
123 \code{capulet.local} as the host name
124 \item a PTR record \code{\_presence.\_tcp.local} for service discovery,
125 pointing to \code{juliet@capulet.\_presence.\_tcp.local}
126 \item and a TXT record \code{juliet@capulet.\_presence.\_tcp.local} specifying
127 more information about her (e.~g. her online status, contact data, etc.)
128 \end{itemize}
129
130 When other clients in the same network enumerate the available services by
131 querying \code{\_presence.\_tcp.local}, they notice Juliet's presence and add
132 her to the roster automatically. In that way, XMPP users can see who is
133 currently available for communication. To start a chat session, clients initiate
134 a TCP connection over the advertised ports, open their XML streams, and send
135 message or IQ stanzas like they would to an XMPP server. Presence is managed
136 over the corresponding TXT record in the mDNS. To go offline, a client
137 announces the deletion of its mDNS records.
138
139 % vim: set ft=tex et ts=2 sw=2 :
This page took 0.116552 seconds and 5 git commands to generate.