1 \documentclass[10pt]{article}
2 \usepackage[ansinew]{inputenc}
3 \usepackage{makeidx,amsmath,amssymb,exscale,multicol,epsfig,graphics,verbatim,ulem}
4 \usepackage{epsfig,geometry,url,listings, subcaption}
5 \usepackage{boxedminipage}
6 \usepackage[T1]{fontenc}%required
8 \geometry{headsep=3ex,hscale=0.9}
10 \hypersetup{pdftitle={GNUnet C Tutorial},
12 pdfauthor={Christian Grothoff <christian@grothoff.org>},
13 pdfkeywords={p2p,search,gnunet,tutorial}
14 %,pdfpagemode={FullScreen}
23 literate={*}{{\char42}}1
27 \newcommand{\exercise}[1]{\noindent\begin{boxedminipage}{\textwidth}{\bf Exercise:} #1 \end{boxedminipage}}
32 \large {A Tutorial for GNUnet 0.9.x (C version)}
34 Christian Grothoff $\qquad$ Bart Polot $\qquad$ Matthias Wachs
38 This tutorials explains how to install GNUnet on a GNU/Linux system ond gives an introduction how
39 GNUnet can be used to develop a Peer-to-Peer application. Detailed installation instructions for
40 various operating systems and a detailed list of all dependencies can found on our website at
41 \url{https://gnunet.org/installation}.
43 \textbf{Please read this tutorial carefully since every single step is important and do not hesitate to contact the GNUnet team if you have any questions or problems! Check here how to contact the GNUnet team:
44 \url{https://gnunet.org/contact_information}}
47 \section{Installing GNUnet}
48 First of all you have to install a current version of GNUnet. You can download a
49 tarball of a stable version from GNU FTP mirrors or obtain the latest development
50 version from our Subversion repository.
52 Most of the time you should prefer to download the stable version since with the
53 latest development version things can be broken, functionality can be changed or tests
54 can fail. You should only use the development version if you know that you require a
55 certain feature or a certain issue has been fixed since the last release.
57 \subsection{Obtaining a stable version}
58 You can download the latest stable version of GNUnet from GNU FTP mirrors:
60 \url{ftp://ftp.gnu.org/gnu/gnunet/gnunet-0.9.5a.tar.gz}
62 You should also download the signature file and verify the integrity of the tarball.
64 \url{ftp://ftp.gnu.org/gnu/gnunet/gnunet-0.9.5a.tar.gz.sig}
66 To verify the signature you should first import the GPG key used to sign the tarball
68 $ gpg --keyserver keys.gnupg.net --recv-keys 48426C7E
70 And use this key to verify the tarball's signature
72 $ gpg --verify gnunet-0.9.5a.tar.gz.sig gnunet-0.9.5a.tar.gz
74 After successfully verifying the integrity you can extract the tarball using
76 $ tar xvzf gnunet-0.9.5a.tar.gz
77 $ mv gnunet-0.9.5a gnunet # we will use the directory "gnunet" in the reminder of this document
81 \subsection{Installing Build Tool Chain and Dependencies}
82 To successfully compile GNUnet you need the tools to build GNUnet and the required dependencies.
83 Please have a look at \url{https://gnunet.org/dependencies} for a list of required dependencies
84 and \url{https://gnunet.org/generic_installation} for specific instructions for your operating system.
86 Please check the notes at the end of the configure process about required dependencies.
88 For GNUNet bootstrapping support and the http(s) plugin you should install \texttt{libcurl}.
89 For the filesharing service you should install at least one of the datastore backends \texttt{mysql},
90 \texttt{sqlite} or \texttt{postgresql}.
92 \subsection{Obtaining the latest version from Subversion}
93 The latest development version can obtained from our Subversion (\textit{svn}) repository. To obtain
94 the code you need Subversion installed and checkout the repository using:
95 \lstset{language=bash}
97 $ svn checkout https://gnunet.org/svn/gnunet
99 After cloning the repository you have to execute
100 \lstset{language=bash}
106 The remainder of this tutorial assumes that you have SVN HEAD checked out.
108 \subsection{Compiling and Installing GNUnet}
110 First, you need to install the latest {\tt
111 libgnupgerror}\footnote{\url{ftp://ftp.gnupg.org/gcrypt/libgpg-error/libgpg-error-1.11.tar.bz2}}
112 and {\tt libgcrypt} version from Git. The current GNUnet code uses
113 ECC functions not available in any released version of libgcrypt.
115 \lstset{language=bash}
117 $ git clone git://git.gnupg.org/libgcrypt.git
120 $ ./configure ; $ make install
124 Assuming all dependencies are installed, the following commands will compile and install GNUnet in your
125 home directory. You can specify the directory where GNUnet will be installed by changing the \lstinline|--prefix| value when calling \lstinline|./configure|. If you do not specifiy a prefix, GNUnet is installed in the directory \lstinline|/usr/local|. When developing new applications you may want to enable
126 verbose logging by adding \lstinline|--enable-logging=verbose|:
128 \lstset{language=bash}
130 $ ./configure --prefix=$HOME --enable-logging
135 After installing GNUnet you have to set the \lstinline|GNUNET_PREFIX| environmental variable used by GNUnet to detect it's installation directory and add your GNUnet installation to your path environmental variable.
136 This configuration is only valid for the current shell session, so you should add \lstinline|export GNUNET_PREFIX=$HOME| to your \lstinline|.bash_rc| or \lstinline|.profile| to be sure the environment variable is always set. In addition you have to create the \lstinline|.gnunet| directory in your home directory where GNUnet stores it's data and an empty GNUnet configuration file:
138 \lstset{language=bash}
140 $ export GNUNET_PREFIX=$HOME
141 $ export PATH=$PATH:$GNUNET_PREFIX/bin
142 $ echo export GNUNET_PREFIX=$HOME >> ~/.bashrc
143 $ echo export PATH=$GNUNET_PREFIX/bin:$PATH >> ~/.bashrc
145 $ touch ~/.gnunet/gnunet.conf
149 \subsection{Common Issues - Check your GNUnet installation}
150 You should check your installation to ensure that installing GNUnet was successful up to this point. You should be able to access GNUnet's binaries and run GNUnet's self check.
154 should return \lstinline|$GNUNET_PREFIX/bin/gnunet-arm|. It should be located in your GNUnet installation and the output should not be empty. If you see an output like:
159 check your {\tt PATH} variable to ensure GNUnet's {\tt bin} directory is included.
161 GNUnet provides tests for all of it's subcomponents. Run
165 to execute tests for all components. {\tt make check} traverses all subdirectories in {\tt src}.
166 For every subdirectory you should get a message like this:
169 make[2]: Entering directory `/home/mwachs/gnunet/contrib'
170 PASS: test_gnunet_prefix
176 If you see a message like this:
179 Mar 12 16:57:56-642482 resolver-api-19449 ERROR Must specify `HOSTNAME' for `resolver' in configuration!
180 Mar 12 16:57:56-642573 test_program-19449 ERROR Assertion failed at resolver_api.c:204.
181 /bin/bash: line 5: 19449 Aborted (core dumped) ${dir}$tst
184 double check your {\tt GNUNET\_PREFIX} environmental variable and double check the steps performed in ~\ref{sub:install}
186 \section{Background: GNUnet Architecture}
187 GNUnet is organized in layers and services. Each service is composed of a
188 main service implementation and a client library for other programs to use
189 the service's functionality, described by an API. This approach is shown in
190 figure~\ref{fig:service}. Some services provide an additional command line
191 tool to enable the user to interact with the service.
193 Very often it is other GNUnet services that will use these APIs to build the
194 higher layers of GNUnet on top of the lower ones. Each layer expands or extends
195 the functionality of the service below (for instance, to build a mesh on top of
196 a DHT). See figure ~\ref{fig:interaction} for an illustration of this approach.
201 \begin{subfigure}[b]{0.3\textwidth}
203 \includegraphics[width=\textwidth]{figs/Service.pdf}
204 \caption{Service with API and network protocol}
208 \begin{subfigure}[b]{0.3\textwidth}
210 \includegraphics[width=\textwidth]{figs/System.pdf}
211 \caption{Service interaction}
212 \label{fig:interaction}
215 \caption{GNUnet's layered system architecture}
218 The main service implementation runs as a standalone process in the operating
219 system and the client code runs as part of the client program, so crashes of a
220 client do not affect the service process or other clients. The service and the
221 clients communicate via a message protocol to be defined and implemented by
224 \section{First Steps with GNUnet}
226 \subsection{Configure your peer}
227 First of all we need to configure your peer. Each peer is started with a configuration containing settings for GNUnet itself and it's services. This configuration is based on the default configuration shipped with GNUnet and can be modified. The default configuration is located in the {\tt \$GNUNET\_PREFIX/share/gnunet/config.d} directory. When starting a peer, you can specify a customized configuration using the the {\tt$-c$} command line switch when starting the ARM service and all other services. When using a modified configuration the default values are loaded and only values specified in the configuration file will replace the default values.
229 Since we want to start additional peers later, we need
230 some modifications from the default configuration. We need to create a separate service home and a file containing our modifications for this peer:
236 Now add the following lines to peer1.conf to use this directory. For simplified usage we want to prevent
237 the peer to connect to the GNUnet network since this could lead to confusing output. This modifications will replace the default settings:
240 $ SERVICEHOME = ~/gnunet1/ # Use this directory to store GNUnet data
242 $ SERVERS = # prevent bootstrapping
245 \subsection{Start a peer}
246 Each GNUnet instance (called peer) has an identity (\textit{peer ID}) based on a
247 cryptographic public private key pair. The peer ID is the printable hash of the
248 public key. So before starting the peer, you may want to just generate the peer's private
249 key using the command
250 \lstset{language=bash}
252 $ gnunet-peerinfo -c ~/peer1.conf -s
254 You should see an output containing the peer ID similar to:
255 \lstset{language=bash}
257 I am peer `0PA02UVRKQTS2C .. JL5Q78F6H0B1ACPV1CJI59MEQUMQCC5G'.
260 GNUnet services are controlled by a master service the so called \textit{Automatic Restart Manager} (ARM).
261 ARM starts, stops and even restarts services automatically or on demand when a client connects.
262 You interact with the ARM service using the \lstinline|gnunet-arm| tool.
263 GNUnet can then be started with \lstinline|gnunet-arm -s| and stopped with
264 \lstinline|gnunet-arm -e|. An additional service not automatically started
265 can be started using \lstinline|gnunet-arm -i <service name>| and stopped
266 using \lstinline|gnunet-arm -k <servicename>|.
268 \subsection{Monitor a peer}
269 In this section, we will monitor the behaviour of our peer's DHT service with respect to a
270 specific key. First we will start GNUnet and then start the DHT service and use the DHT monitor tool
271 to monitor the PUT and GET commands we issue ussing the \lstinline|gnunet-dht-put| and
272 \lstinline|gnunet-dht-get| command. Using the ``monitor'' line given below, you can observe the behavior of
273 your own peer's DHT with respect to the specified KEY:
275 \lstset{language=bash}
277 $ gnunet-arm -c ~/peer1.conf -s # start gnunet with all default services
278 $ gnunet-arm -c ~/peer1.conf -i dht # start DHT service
279 $ cd ~/gnunet/src/dht;
280 $ ./gnunet-dht-monitor -c ~/peer1.conf -k KEY
282 Now open a separate terminal and change again to the \lstinline|gnunet/src/dht| directory:
284 $ cd ~/gnunet/src/dht
285 $ ./gnunet-dht-put -c ~/peer1.conf -k KEY -d VALUE # put VALUE under KEY in the DHT
286 $ ./gnunet/src/dht/gnunet-dht-get ~/peer1.conf -k KEY # get key KEY from the DHT
287 $ gnunet-statistics -c ~/peer1.conf # print statistics about current GNUnet state
288 $ gnunet-statistics -c ~/peer1.conf -s dht # print statistics about DHT service
291 \subsection{Starting Two Peers by Hand}
292 \subsubsection{Setup a second peer}
293 We will now start a second peer on your machine.
294 For the second peer, you will need to manually create a modified
295 configuration file to avoid conflicts with ports and directories.
296 A peers configuration file is by default located in {\tt ~/.gnunet/gnunet.conf}.
297 This file is typically very short or even empty as only the differences to the
298 defaults need to be specified. The defaults are located in
299 many files in the {\tt \$GNUNET\_PREFIX/share/gnunet/config.d} directory.
301 To configure the second peer, use the files {\tt
302 \$GNUNET\_PREFIX/share/gnunet/config.d} as a template for your main
305 \lstset{language=bash}
307 $ cat $GNUNET_PREFIX/share/gnunet/config.d/*.conf > peer2.conf
309 Now you have to edit {\tt peer2.conf} and change:
312 \item{\texttt{SERVICEHOME} under \texttt{PATHS}}
313 \item{Every value for ``\texttt{PORT}'' (add 10000) in any section (if \texttt{PORT} is enabled, may be disabled using "\#") }
314 \item{Every value for ``\texttt{UNIXPATH}'' in any section (e.g. by adding a "-p2" suffix)}
316 to a fresh, unique value. Make sure that the \texttt{PORT} numbers stay
317 below 65536. From now on, whenever you interact with the second
318 peer, you need to specify {\tt -c peer2.conf} as an additional
319 command line argument.
321 Now, generate the 2nd peer's private key:
323 \lstset{language=bash}
325 $ gnunet-peerinfo -s -c peer2.conf
329 This may take a while, generate entropy using your keyboard or mouse
330 as needed. Also, make sure the output is different from the {\tt
331 gnunet-peerinfo} output for the first peer (otherwise you made an
332 error in the configuration).
334 \subsubsection{Start the second peer and connect the peers}
335 Then, you can start a second peer using:
336 \lstset{language=bash}
338 $ gnunet-arm -c peer2.conf -s
339 $ gnunet-arm -c peer2.conf -i dht
340 $ ~/gnunet/src/dht/gnunet-dht-put -c peer2.conf -k KEY -d VALUE
341 $ ~/gnunet/src/dht/gnunet-dht-get -c peer2.conf -k KEY
343 If you want the two peers to connect, you have multiple options:
346 \item UDP neighbour discovery (automatic)
347 \item Setup a bootstrap server
348 \item Connect manually
350 To setup peer 1 as bootstrapping server change the configuration of the first one to be a hostlist server by adding the following lines to \texttt{peer1.conf} to enable bootstrapping server:
356 Then change {\tt peer2.conf} and replace the ``\texttt{SERVERS}'' line in the ``\texttt{[hostlist]}'' section with
357 ``\texttt{http://localhost:8080/}''. Restart both peers using:
359 $ gnunet-arm -c peer1.conf -e # stop first peer
360 $ gnunet-arm -c peer1.conf -s # start first peer
361 $ gnunet-arm -c peer2.conf -s # start second peer
364 Note that if you start your peers without changing these settings, they
365 will use the ``global'' hostlist servers of the GNUnet P2P network and
366 likely connect to those peers. At that point, debugging might become
367 tricky as you're going to be connected to many more peers and would
368 likely observe traffic and behaviors that are not explicitly controlled
371 \subsubsection{How to connect manually}
372 If you want to use the \texttt{peerinfo} tool to connect your peers, you should:
375 \item{Remove {\tt hostlist} from {\tt DEFAULTSERVICES} (to not connect to the global GNUnet)}
376 \item{Start both peers running {\tt gnunet-arm -c peer1.conf -s} and {\tt gnunet-arm -c peer2.conf -s}}
377 \item{Get \texttt{HELLO} message of the first peer running {\tt gnunet-peerinfo -c peer1.conf -g}}
378 \item{Give the output to the second peer by running {\tt gnunet-peerinfo -c peer2.conf -p '<output>'}}
381 Check that they are connected using {\tt gnunet-core -c peer1.conf}, which should give you the other peer's
384 $ gnunet-core -c peer1.conf
385 Peer `9TVUCS8P5A7ILLBGO6JSTSSN2B44H3D2MUIFJMLKAITC0I22UVFBFP1H8NRK2IA35VKAK16LLO0MFS7TAQ9M1KNBJ4NGCHP3JPVULDG'
388 \subsection{Starting Peers Using the Testbed Service}
390 GNUnet's testbed service is used for testing scenarios where a number of peers
391 are to be started. The testbed can manage peers on a single host or on multiple
392 hosts in a distributed fashion. On a single affordable computer, it should be
393 possible to run around tens of peers without drastically increasing the load on the
396 The testbed service can be access through its API
397 \texttt{include/gnunet\_testbed\_service.h}. The API provides many routines for
398 managing a group of peers. It also provides a helper function
399 \texttt{GNUNET\_TESTBED\_test\_run()} to quickly setup a minimalistic testing
400 environment on a single host.
402 This function takes a configuration file which will be used as a template
403 configuration for the peers. The testbed takes care of modifying relevant
404 options in the peers' configuration such as SERVICEHOME, PORT, UNIXPATH to
405 unique values so that peers run without running into conflicts. It also checks
406 and assigns the ports in configurations only if they are free.
408 Additionally, the testbed service also reads its options from the same
409 configuration file. Various available options and details about them can be
410 found in the testbed default configuration file \texttt{src/testbed/testbed.conf}.
412 With the testbed API, a sample test case can be structured as follows:
413 \lstinputlisting[language=C]{testbed_test.c}
414 The source code for the above listing can be found at
415 \url{https://gnunet.org/svn/gnunet/doc/testbed_test.c}. After installing GNUnet, the above source code can be compiled as:
416 \lstset{language=bash}
418 $ export CPPFLAGS="-I/path/to/gnunet/headers"
419 $ export LDFLAGS="-L/path/to/gnunet/libraries"
420 $ gcc -o testbed-test -lgnunettestbed -lgnunetdht -lgnunetutil testbed_test.c
422 The \texttt{CPPFLAGS} and \texttt{LDFLAGS} are necessary if GNUnet is installed
423 into a different directory other than \texttt{/usr/local}.
425 All of testbed API's peer management functions treat management actions as
426 operations and return operation handles. It is expected that the operations
427 begin immediately, but they may get delayed (to balance out load on the system).
428 The program using the API then has to take care of marking the operation as
429 ``done'' so that its associated resources can be freed immediately and other
430 waiting operations can be executed. Operations will be canceled if they are
431 marked as ``done'' before their completion.
433 An operation is treated as completed when it succeeds or fails. Completion of
434 an operation is either conveyed as events through \textit{controller event
435 callback} or through respective operation completion callbacks. In functions
436 which support completion notification through both controller event callback and
437 operation completion callback, first the controller event callback will be
438 called. If the operation is not marked as done in that callback or if the
439 callback is given as NULL when creating the operation, the operation completion
440 callback will be called. The API documentation shows which event are to be
441 expected in the controller event notifications. It also documents any
442 exceptional behaviour.
444 Once the peers are started, test cases often need to connect some of the peers'
445 services. Normally, opening a connect to a peer's service requires the peer's
446 configuration. While using testbed, the testbed automatically generates
447 per-peer configuration. Accessing those configurations directly through file
448 system is discouraged as their locations are dynamically created and will be
449 different among various runs of testbed. To make access to these configurations
450 easy, testbed API provides the function
451 \texttt{GNUNET\_TESTBED\_service\_connect()}. This function fetches the
452 configuration of a given peer and calls the \textit{Connect Adapter}.
453 In the example code, it is the \texttt{dht\_ca}. A connect adapter is expected
454 to open the connection to the needed service by using the provided configuration
455 and return the created service connection handle. Successful connection to the
456 needed service is signaled through \texttt{service\_connect\_comp\_cb}.
458 A dual to connect adapter is the \textit{Disconnect Adapter}. This callback is
459 called after the connect adapter has been called when the operation from
460 \texttt{GNUNET\_TESTBED\_service\_connect()} is marked as ``done''. It has to
461 disconnect from the service with the provided service handle (\texttt{op\_result}).
463 \exercise{Find out how many peers you can run on your system.}
465 \exercise{Find out how to create connections from within {\tt run} and create a
466 2D torus topology. Then use the DHT API to store and retrieve values in the
469 \section{Developing Applications}
470 \subsection{gnunet-ext}
471 To develop a new peer-to-peer application or to extend GNUnet we provide
472 a template build system for writing GNUnet extensions in C. It can be
475 \lstset{language=bash}
477 $ svn checkout https://gnunet.org/svn/gnunet-ext/
480 $ ./configure --prefix=$HOME --with-gnunet=$GNUNET_PREFIX
487 The GNUnet ext template includes examples and a working buildsystem for a new GNUnet service.
488 A common GNUnet service consists of the following parts which will be discussed in detail in the
489 remainder of this document. The functionality of a GNUnet service is implemented in:
493 \item the GNUnet service (\lstinline|gnunet-ext/src/ext/gnunet-service-ext.c|)
494 \item the client API (\lstinline|gnunet-ext/src/ext/ext_api.c|)
495 \item the client application using the service API (\lstinline|gnunet-ext/src/ext/gnunet-ext.c|)
500 The interfaces for these entities are defined in:
503 \item client API interface (\lstinline|gnunet-ext/src/ext/ext.h|)
504 \item the service interface (\lstinline|gnunet-ext/src/include/gnunet_service_SERVICE.h|)
505 \item the P2P protocol (\lstinline|gnunet-ext/src/include/gnunet_protocols_ext.h|)
509 In addition the \texttt{ext} systems provides:
512 \item a test testing the API (\lstinline|gnunet-ext/src/ext/test_ext_api.c|)
513 \item a configuration template for the service (\lstinline|gnunet-ext/src/ext/ext.conf.in|)
517 \subsection{Adapting the Template}
519 The first step for writing any extension with a new service is to
520 ensure that the {\tt ext.conf.in} file contains entries for the
521 \texttt{UNIXPATH}, \texttt{PORT} and \texttt{BINARY} for the service in a section named after
524 If you want to adapt the template rename the {\tt ext.conf.in} to match your
525 services name, you have to modify the \texttt{AC\_OUTPUT} section in {\tt configure.ac}
526 in the \texttt{gnunet-ext} root.
528 \section{Writing a Client Application}
530 When writing any client application (for example, a command-line
531 tool), the basic structure is to start with the {\tt
532 GNUNET\_PROGRAM\_run} function. This function will parse
533 command-line options, setup the scheduler and then invoke the {\tt
534 run} function (with the remaining non-option arguments) and a handle
535 to the parsed configuration (and the configuration file name that was
536 used, which is typically not needed):
540 #include <gnunet/platform.h>
541 #include <gnunet/gnunet_util_lib.h>
549 const struct GNUNET_CONFIGURATION_Handle *cfg)
556 main (int argc, char *const *argv)
558 static const struct GNUNET_GETOPT_CommandLineOption options[] = {
559 GNUNET_GETOPT_OPTION_END
562 GNUNET_PROGRAM_run (argc,
565 gettext_noop ("binary description text"),
566 options, &run, NULL)) ? ret : 1;
570 \subsection{Handling command-line options}
572 Options can then be added easily by adding global variables and
573 expanding the {\tt options} array. For example, the following would
574 add a string-option and a binary flag (defaulting to {\tt NULL} and
575 {\tt GNUNET\_NO} respectively):
578 static char *string_option;
582 static const struct GNUNET_GETOPT_CommandLineOption options[] = {
583 {'s', "name", "SOMESTRING",
584 gettext_noop ("text describing the string_option NAME"), 1,
585 &GNUNET_GETOPT_set_string, &string_option},
587 gettext_noop ("text describing the flag option"), 0,
588 &GNUNET_GETOPT_set_one, &a_flag},
589 GNUNET_GETOPT_OPTION_END
594 Issues such as displaying some helpful text describing options using
595 the {\tt --help} argument and error handling are taken care of when
596 using this approach. Other {\tt GNUNET\_GETOPT\_}-functions can be used
597 to obtain integer value options, increment counters, etc. You can
598 even write custom option parsers for special circumstances not covered
599 by the available handlers.
601 Inside the {\tt run} method, the program would perform the
602 application-specific logic, which typically involves initializing and
603 using some client library to interact with the service. The client
604 library is supposed to implement the IPC whereas the service provides
605 more persistent P2P functions.
607 \exercise{Add a few command-line options and print them inside
608 of {\tt run}. What happens if the user gives invalid arguments?}
610 \subsection{Writing a Client Library}
612 The first and most important step in writing a client library is to
613 decide on an API for the library. Typical API calls include
614 connecting to the service, performing application-specific requests
615 and cleaning up. Many examples for such service APIs can be found
616 in the {\tt gnunet/src/include/gnunet\_*\_service.h} files.
618 Then, a client-service protocol needs to be designed. This typically
619 involves defining various message formats in a header that will be
620 included by both the service and the client library (but is otherwise
621 not shared and hence located within the service's directory and not
622 installed by {\tt make install}). Each message must start with a {\tt
623 struct GNUNET\_MessageHeader} and must be shorter than 64k. By
624 convention, all fields in IPC (and P2P) messages must be in big-endian
625 format (and thus should be read using {\tt ntohl} and similar
626 functions and written using {\tt htonl} and similar functions).
627 Unique message types must be defined for each message struct in the
628 {\tt gnunet\_protocols.h} header (or an extension-specific include
631 \subsubsection{Connecting to the Service}
633 Before a client library can implement the application-specific protocol
634 with the service, a connection must be created:
638 struct GNUNET_CLIENT_Connection *client;
639 client = GNUNET_CLIENT_connect ("service-name", cfg);
642 As a result a {\tt GNUNET\_CLIENT\_Connection} handle is returned
643 which has to used in later API calls related to this service.
644 The complete client API can be found in {\tt gnunet\_client\_lib.h}
646 \subsubsection{GNUnet Messages}
648 In GNUnet, messages are always sent beginning with a {\tt struct GNUNET\_MessageHeader}
649 in big endian format. This header defines the size and the type of the
650 message, the payload follows after this header.
654 struct GNUNET_MessageHeader
658 * The length of the struct (in bytes, including the length field itself),
659 * in big-endian format.
661 uint16_t size GNUNET_PACKED;
664 * The type of the message (GNUNET_MESSAGE_TYPE_XXXX), in big-endian format.
666 uint16_t type GNUNET_PACKED;
671 Existing message types are defined in {\tt gnunet\_protocols.h}\\
672 A common way to create a message is:
676 struct GNUNET_MessageHeader *msg =
677 GNUNET_malloc(payload_size + sizeof(struct GNUNET_MessageHeader));
678 msg->size = htons(payload_size + sizeof(struct GNUNET_MessageHeader));
679 msg->type = htons(GNUNET_MY_MESSAGE_TYPE);
680 memcpy(&msg[1], &payload, payload_size);
684 \exercise{Define a message struct that includes a 32-bit
685 unsigned integer in addition to the standard GNUnet MessageHeader.
686 Add a C struct and define a fresh protocol number for your message.}
689 \subsubsection{Sending Requests to the Service}
691 Any client-service protocol must start with the client sending the
692 first message to the service, since services are only notified about
693 (new) clients upon receiving a the first message.
695 Clients can transmit messages to the service using the
696 {\tt GNUNET\_CLIENT\_notify\_transmit\_ready} API:
700 transmit_cb (void *cls, size_t size, void *buf)
703 if (NULL == buf) { handle_error(); return 0; }
704 GNUNET_assert (size >= msg_size);
705 memcpy (buf, my_msg, msg_size);
711 th = GNUNET_CLIENT_notify_transmit_ready (client,
719 The client-service protocoll calls {\tt GNUNET\_CLIENT\_notify\_transmit\_ready}
720 to be notified when the client is ready to send data to the service.
721 Besides other arguments, you have to pass the client returned
722 from the {\tt connect} call, the message size and the callback function to
723 call when the client is ready to send.
725 Only a single transmission request can be queued per client at the
726 same time using this API. The handle {\tt th} can be used to cancel
727 the request if necessary (for example, during shutdown).
729 When {\tt transmit\_cb} is called the message is copied in the buffer provided and
730 the number of bytes copied into the buffer is returned. {\tt transmit\_cb}
731 could also return 0 if for some reason no message
732 could be constructed; this is not an error and the connection to the
733 service will persist in this case.
735 \exercise{Define a helper function to transmit a 32-bit
736 unsigned integer (as payload) to a service using some given client
740 \subsubsection{Receiving Replies from the Service}
742 Clients can receive messages from the service using the
743 {\tt GNUNET\_CLIENT\_receive} API:
748 * Function called with messages from stats service.
751 * @param msg message received, NULL on timeout or fatal error
754 receive_message (void *cls, const struct GNUNET_MessageHeader *msg)
756 struct MyArg *arg = cls;
762 GNUNET_CLIENT_receive (client,
769 It should be noted that this receive call only receives a single
770 message. To receive additional messages, {\tt
771 GNUNET\_CLIENT\_receive} must be called again.
773 \exercise{Expand your helper function to receive a
774 response message (for example, containing just the GNUnet MessageHeader
775 without any payload). Upon receiving the service's response, you should
776 call a callback provided to your helper function's API. You'll need to
777 define a new 'struct' to hold your local context (``closure'').}
780 \subsection{Writing a user interface}
782 Given a client library, all it takes to access a service now is to
783 combine calls to the client library with parsing command-line
786 \exercise{Call your client API from your {\tt run} method
787 in your client application to send a request to the service.
788 For example, send a 32-bit integer value based on a number given
789 at the command-line to the service.}
793 \section{Writing a Service}
795 Before you can test the client you've written so far, you'll need to also
796 implement the corresponding service.
799 \subsection{Code Placement}
801 New services are placed in their own subdirectory under {\tt gnunet/src}.
802 This subdirectory should contain the API implementation file {\tt SERVICE\_api.c},
803 the description of the client-service protocol {\tt SERVICE.h} and P2P protocol
804 {\tt SERVICE\_protocol.h}, the implementation of the service itself
805 {\tt gnunet-service-SERVICE.h} and several files for tests, including test code
806 and configuration files.
808 \subsection{Starting a Service}
810 The key API definitions for starting services are:
813 typedef void (*GNUNET_SERVICE_Main) (void *cls,
814 struct GNUNET_SERVER_Handle *server,
815 const struct GNUNET_CONFIGURATION_Handle *cfg);
816 int GNUNET_SERVICE_run (int argc,
818 const char *serviceName,
819 enum GNUNET_SERVICE_Options opt,
820 GNUNET_SERVICE_Main task,
824 Here is a starting point for your main function for your service:
828 static void my_main (void *cls,
829 struct GNUNET_SERVER_Handle *server,
830 const struct GNUNET_CONFIGURATION_Handle *cfg)
835 int main (int argc, char *const*argv)
838 GNUNET_SERVICE_run (argc, argv, "my",
839 GNUNET_SERVICE_OPTION_NONE,
846 \exercise{Write a stub service that processes no messages at all
847 in your code. Create a default configuration for it, integrate it
848 with the build system and start the service from {\tt
849 gnunet-service-arm} using {\tt gnunet-arm -i NAME}.}
852 \subsection{Receiving Requests from Clients}
854 Inside of the {\tt my\_main} method, a service typically registers for
855 the various message types from clients that it supports by providing
856 a handler function, the message type itself and possibly a fixed
857 message size (or 0 for variable-size messages):
862 handle_set (void *cls,
863 struct GNUNET_SERVER_Client *client,
864 const struct GNUNET_MessageHeader *message)
866 GNUNET_SERVER_receive_done (client, GNUNET_OK);
869 handle_get (void *cls,
870 struct GNUNET_SERVER_Client *client,
871 const struct GNUNET_MessageHeader *message)
873 GNUNET_SERVER_receive_done (client, GNUNET_OK);
876 static void my_main (void *cls,
877 struct GNUNET_SERVER_Handle *server,
878 const struct GNUNET_CONFIGURATION_Handle *cfg)
880 static const struct GNUNET_SERVER_MessageHandler handlers[] = {
881 {&handle_set, NULL, GNUNET_MESSAGE_TYPE_MYNAME_SET, 0},
882 {&handle_get, NULL, GNUNET_MESSAGE_TYPE_MYNAME_GET, 0},
885 GNUNET_SERVER_add_handlers (server, handlers);
886 /* do more setup work */
890 Each handler function {\bf must} eventually (possibly in some
891 asynchronous continuation) call {\tt GNUNET\_SERVER\_receive\_done}.
892 Only after this call additional messages from the same client may
893 be processed. This way, the service can throttle processing messages
894 from the same client. By passing {\tt GNUNET\_SYSERR}, the service
895 can close the connection to the client, indicating an error.
897 Services must check that client requests are well-formed and must not
898 crash on protocol violations by the clients. Similarly, client
899 libraries must check replies from servers and should gracefully report
900 errors via their API.
903 \exercise{Change the service to ``handle'' the message from your
904 client (for now, by printing a message). What happens if you
905 forget to call {\tt GNUNET\_SERVER\_receive\_done}?}
908 \subsection{Responding to Clients}
910 Servers can send messages to clients using the
911 {\tt GNUNET\_SERVER\_notify\_transmit\_ready} API:
916 transmit_cb (void *cls, size_t size, void *buf)
919 if (NULL == buf) { handle_error(); return 0; }
920 GNUNET_assert (size >= msg_size);
921 memcpy (buf, my_msg, msg_size);
927 struct GNUNET_SERVER_TransmitHandle *th;
928 th = GNUNET_SERVER_notify_transmit_ready (client,
935 Only a single transmission request can be queued per client
936 at the same time using this API.
937 Additional APIs for sending messages to clients can be found
938 in the {\tt gnunet\_server\_lib.h} header.
941 \exercise{Change the service respond to the request from your
942 client. Make sure you handle malformed messages in both directions.}
945 \section{Interacting directly with other Peers using the CORE Service}
947 One of the most important services in GNUnet is the \texttt{CORE} service
948 managing connections between peers and handling encryption between peers.
950 One of the first things any service that extends the P2P protocol typically does
951 is connect to the \texttt{CORE} service using:
955 #include <gnunet/gnunet_core_service.h>
957 struct GNUNET_CORE_Handle *
958 GNUNET_CORE_connect (const struct GNUNET_CONFIGURATION_Handle *cfg,
960 GNUNET_CORE_StartupCallback init,
961 GNUNET_CORE_ConnectEventHandler connects,
962 GNUNET_CORE_DisconnectEventHandler disconnects,
963 GNUNET_CORE_MessageCallback inbound_notify,
964 int inbound_hdr_only,
965 GNUNET_CORE_MessageCallback outbound_notify,
966 int outbound_hdr_only,
967 const struct GNUNET_CORE_MessageHandler *handlers);
970 \subsection{New P2P connections}
972 Before any traffic with a different peer can be exchanged, the peer must be
973 known to the service. This is notified by the \texttt{CORE} {\tt connects} callback,
974 which communicates the identity of the new peer to the service:
980 const struct GNUNET_PeerIdentity * peer)
982 /* Save identity for later use */
983 /* Optional: start sending messages to peer */
987 \exercise{Create a service that connects to the \texttt{CORE}. Then
988 start (and connect) two peers and print a message once your connect
989 callback is invoked.}
991 \subsection{Receiving P2P Messages}
993 To receive messages from \texttt{CORE}, services register a set of handlers
994 (parameter {\tt *handlers} in the \lstinline|GNUNET_CORE_connect| call that are called by \texttt{CORE}
995 when a suitable message arrives.
1000 callback_function_for_type_one(void *cls,
1001 const struct GNUNET_PeerIdentity *peer,
1002 const struct GNUNET_MessageHeader *message)
1005 return GNUNET_OK; /* or GNUNET_SYSERR to close the connection */
1009 * Functions to handle messages from core
1011 static struct GNUNET_CORE_MessageHandler core_handlers[] = {
1012 {&callback_function_for_type_one, GNUNET_MESSAGE_TYPE_MYSERVICE_TYPE_ONE, 0},
1018 \exercise{Start one peer with a new service that has a message
1019 handler and start a second peer that only has your ``old'' service
1020 without message handlers. Which ``connect'' handlers are invoked when
1021 the two peers are connected? Why?}
1024 \subsection{Sending P2P Messages}
1026 In response to events (connect, disconnect, inbound messages,
1027 timing, etc.) services can then use this API to transmit messages:
1032 (*GNUNET_CONNECTION_TransmitReadyNotify) (void *cls,
1036 /* Fill "*buf" with up to "size" bytes, must start with GNUNET_MessageHeader */
1037 return n; /* Total size of the message put in "*buf" */
1040 struct GNUNET_CORE_TransmitHandle *
1041 GNUNET_CORE_notify_transmit_ready (struct GNUNET_CORE_Handle *handle,
1042 int cork, uint32_t priority,
1043 struct GNUNET_TIME_Relative maxdelay,
1044 const struct GNUNET_PeerIdentity *target,
1046 GNUNET_CONNECTION_TransmitReadyNotify notify,
1050 \exercise{Write a service that upon connect sends messages as
1051 fast as possible to the other peer (the other peer should run a
1052 service that ``processes'' those messages). How fast is the
1053 transmission? Count using the STATISTICS service on both ends. Are
1054 messages lost? How can you transmit messages faster? What happens if
1055 you stop the peer that is receiving your messages?}
1058 \subsection{End of P2P connections}
1060 If a message handler returns {\tt GNUNET\_SYSERR}, the remote peer shuts down or
1061 there is an unrecoverable network disconnection, CORE notifies the service that
1062 the peer disconnected. After this notification no more messages will be received
1063 from the peer and the service is no longer allowed to send messages to the peer.
1064 The disconnect callback looks like the following:
1069 disconnects (void *cls,
1070 const struct GNUNET_PeerIdentity * peer)
1072 /* Remove peer's identity from known peers */
1073 /* Make sure no messages are sent to peer from now on */
1077 \exercise{Fix your service to handle peer disconnects.}
1079 \section{Using the DHT}
1080 The DHT allows to store data so other peers in the P2P network can
1081 access it and retrieve data stored by any peers in the network.
1082 This section will explain how to use the DHT. Of course, the first
1083 thing to do is to connect to the DHT service:
1086 dht_handle = GNUNET_DHT_connect (cfg, parallel_requests);
1088 The second parameter indicates how many requests in parallel to expect.
1089 It is not a hard limit, but a good approximation will make the DHT more
1092 \subsection{Storing data in the DHT}
1093 Since the DHT is a dynamic environment (peers join a leave frequently)
1094 the data that we put in the DHT does not stay there indefinitely. It is
1095 important to ``refresh'' the data periodically by simply storing it again,
1096 in order to make sure other peers can access it.
1098 The put API call offers a callback to signal that the PUT request has been
1099 sent. This does not guarantee that the data is accessible to others peers,
1100 or even that is has been stored, only that the service has requested to
1101 a neighboring peer the retransmission of the PUT request towards its final
1102 destination. Currently there is no feedback about whether or not the data
1103 has been sucessfully stored or where it has been stored. In order to improve
1104 the availablilty of the data and to compensate for possible errors, peers leaving
1105 and other unfavorable events, just make several PUT requests!
1110 message_sent_cont (void *cls, const struct GNUNET_SCHEDULER_TaskContext *tc)
1112 /* Request has left local node */
1115 struct GNUNET_DHT_PutHandle *
1116 GNUNET_DHT_put (struct GNUNET_DHT_Handle *handle,
1117 const struct GNUNET_HashCode * key,
1118 uint32_t desired_replication_level,
1119 enum GNUNET_DHT_RouteOption options, /* Route options, see next call */
1120 enum GNUNET_BLOCK_Type type, size_t size, const void *data,
1121 struct GNUNET_TIME_Absolute exp, /* When does the data expire? */
1122 struct GNUNET_TIME_Relative timeout, /* How long to try to send the request */
1123 GNUNET_DHT_PutContinuation cont,
1127 \exercise{Store a value in the DHT periodically to make sure it is available
1128 over time. You might consider using the function GNUNET\_SCHEDULER\_add\_delayed and
1129 call GNUNET\_DHT\_put from inside a helper function.}
1132 \subsection{Obtaining data from the DHT}
1133 As we saw in the previous example, the DHT works in an asynchronous mode.
1134 Each request to the DHT is executed ``in the background'' and the API
1135 calls return immediately. In order to receive results from the DHT, the
1136 API provides a callback. Once started, the request runs in the service,
1137 the service will try to get as many results as possible (filtering out
1138 duplicates) until the timeout expires or we explicitly stop the request.
1139 It is possible to give a ``forever'' timeout with
1140 {\tt GNUNET\_TIME\_UNIT\_FOREVER\_REL}.
1142 If we give a route option {\tt GNUNET\_DHT\_RO\_RECORD\_ROUTE} the callback
1143 will get a list of all the peers the data has travelled, both on the PUT
1144 path and on the GET path.
1148 get_result_iterator (void *cls, struct GNUNET_TIME_Absolute expiration,
1149 const struct GNUNET_HashCode * key,
1150 const struct GNUNET_PeerIdentity *get_path,
1151 unsigned int get_path_length,
1152 const struct GNUNET_PeerIdentity *put_path,
1153 unsigned int put_path_length,
1154 enum GNUNET_BLOCK_Type type, size_t size, const void *data)
1156 /* Do stuff with the data and/or route */
1158 GNUNET_DHT_get_stop (get_handle);
1162 GNUNET_DHT_get_start (dht_handle,
1166 GNUNET_DHT_RO_NONE, /* Route options */
1167 NULL, /* xquery: not used here */
1168 0, /* xquery size */
1169 &get_result_iterator,
1173 \exercise{Store a value in the DHT and after a while retrieve it. Show the IDs of all
1174 the peers the requests have gone through. In order to convert a peer ID to a string, use
1175 the function GNUNET\_i2s. Pay attention to the route option parameters in both calls!}
1177 \subsection{Implementing a block plugin}
1179 In order to store data in the DHT, it is necessary to provide a block
1180 plugin. The DHT uses the block plugin to ensure that only well-formed
1181 requests and replies are transmitted over the network.
1183 The block plugin should be put in a file {\tt
1184 plugin\_block\_SERVICE.c} in the service's respective directory. The
1185 mandatory functions that need to be implemented for a block plugin are
1186 described in the following sections.
1188 \subsubsection{Validating requests and replies}
1190 The evaluate function should validate a reply or a request. It returns
1191 a {\tt GNUNET\_BLOCK\_EvaluationResult}, which is an enumeration. All
1192 possible answers are in {\tt gnunet\_block\_lib.h}. The function will
1193 be called with a {\tt reply\_block} argument of {\tt NULL} for
1194 requests. Note that depending on how {\tt evaluate} is called, only
1195 some of the possible return values are valid. The specific meaning of
1196 the {\tt xquery} argument is application-specific. Applications that
1197 do not use an extended query should check that the {\tt xquery\_size}
1198 is zero. The Bloom filter is typically used to filter duplicate
1203 static enum GNUNET_BLOCK_EvaluationResult
1204 block_plugin_SERVICE_evaluate (void *cls,
1205 enum GNUNET_BLOCK_Type type,
1206 const GNUNET_HashCode * query,
1207 struct GNUNET_CONTAINER_BloomFilter **bf,
1211 const void *reply_block,
1212 size_t reply_block_size)
1214 /* Verify type, block and bloomfilter */
1218 Note that it is mandatory to detect duplicate replies in this
1219 function and return the respective status code. Duplicate
1220 detection should be done by setting the respective bits in
1221 the Bloom filter {\tt bf}. Failure to do so may cause replies
1222 to circle in the network.
1224 \subsubsection{Deriving a key from a reply}
1226 The DHT can operate more efficiently if it is possible to derive a key
1227 from the value of the corresponding block. The {\tt get\_key}
1228 function is used to obtain the key of a block --- for example, by
1229 means of hashing. If deriving the key is not possible, the function
1230 should simply return {\tt GNUNET\_SYSERR} (the DHT will still work
1231 just fine with such blocks).
1236 block_plugin_SERVICE_get_key (void *cls, enum GNUNET_BLOCK_Type type,
1237 const void *block, size_t block_size,
1238 GNUNET_HashCode * key)
1240 /* Store the key in the key argument, return GNUNET_OK on success. */
1244 \subsubsection{Initialization of the plugin}
1246 The plugin is realized as a shared C library. The library must export
1247 an initialization function which should initialize the plugin. The
1248 initialization function specifies what block types the plugin cares
1249 about and returns a struct with the functions that are to be used for
1250 validation and obtaining keys (the ones just defined above).
1255 libgnunet_plugin_block_SERVICE_init (void *cls)
1257 static enum GNUNET_BLOCK_Type types[] =
1259 GNUNET_BLOCK_TYPE_SERVICE_BLOCKYPE, /* list of blocks we care about, from gnunet_block_lib.h */
1260 GNUNET_BLOCK_TYPE_ANY /* end of list */
1262 struct GNUNET_BLOCK_PluginFunctions *api;
1264 api = GNUNET_malloc (sizeof (struct GNUNET_BLOCK_PluginFunctions));
1265 api->evaluate = &block_plugin_SERICE_evaluate;
1266 api->get_key = &block_plugin_SERVICE_get_key;
1272 \subsubsection{Shutdown of the plugin}
1274 Following GNUnet's general plugin API concept, the plugin must
1275 export a second function for cleaning up. It usually does very
1281 libgnunet_plugin_block_SERVICE_done (void *cls)
1283 struct GNUNET_TRANSPORT_PluginFunctions *api = cls;
1291 \subsubsection{Integration of the plugin with the build system}
1293 In order to compile the plugin, the {\tt Makefile.am} file for the
1294 service should contain a rule similar to this:
1296 \lstset{language=make}
1298 plugin_LTLIBRARIES = \
1299 libgnunet_plugin_block_SERVICE.la
1300 libgnunet_plugin_block_SERVICE_la_SOURCES = \
1301 plugin_block_SERVICE.c
1302 libgnunet_plugin_block_SERVICE_la_LIBADD = \
1303 $(top_builddir)/src/hello/libgnunethello.la \
1304 $(top_builddir)/src/block/libgnunetblock.la \
1305 $(top_builddir)/src/util/libgnunetutil.la
1306 libgnunet_plugin_block_SERVICE_la_LDFLAGS = \
1307 $(GN_PLUGIN_LDFLAGS)
1308 libgnunet_plugin_block_SERVICE_la_DEPENDENCIES = \
1309 $(top_builddir)/src/block/libgnunetblock.la
1314 \exercise{Write a block plugin that accepts all queries
1315 and all replies but prints information about queries and replies
1316 when the respective validation hooks are called.}
1320 \subsection{Monitoring the DHT}
1321 It is possible to monitor the functioning of the local DHT service. When monitoring
1322 the DHT, the service will alert the monitoring program of any events,
1323 both started locally or received for routing from another peer. The are three different
1324 types of events possible: a GET request, a PUT request or a response (a reply to
1327 Since the different events have different associated data, the API gets 3
1328 different callbacks (one for each message type) and optional type and key parameters,
1329 to allow for filtering of messages. When an event happens, the appropiate callback
1330 is called with all the information about the event.
1334 get_callback (void *cls,
1335 enum GNUNET_DHT_RouteOption options,
1336 enum GNUNET_BLOCK_Type type,
1338 uint32_t desired_replication_level,
1339 unsigned int path_length,
1340 const struct GNUNET_PeerIdentity *path,
1341 const struct GNUNET_HashCode * key)
1346 get_resp_callback (void *cls,
1347 enum GNUNET_BLOCK_Type type,
1348 const struct GNUNET_PeerIdentity *get_path,
1349 unsigned int get_path_length,
1350 const struct GNUNET_PeerIdentity *put_path,
1351 unsigned int put_path_length,
1352 struct GNUNET_TIME_Absolute exp,
1353 const struct GNUNET_HashCode * key,
1360 put_callback (void *cls,
1361 enum GNUNET_DHT_RouteOption options,
1362 enum GNUNET_BLOCK_Type type,
1364 uint32_t desired_replication_level,
1365 unsigned int path_length,
1366 const struct GNUNET_PeerIdentity *path,
1367 struct GNUNET_TIME_Absolute exp,
1368 const struct GNUNET_HashCode * key,
1374 monitor_handle = GNUNET_DHT_monitor_start (dht_handle,
1375 block_type, /* GNUNET_BLOCK_TYPE_ANY for all */
1376 key, /* NULL for all */
1384 \section{Debugging with {\tt gnunet-arm}}
1386 Even if services are managed by {\tt gnunet-arm}, you can start them with
1387 {\tt gdb} or {\tt valgrind}. For example, you could add the following lines
1388 to your configuration file to start the DHT service in a {\tt gdb} session in a
1393 PREFIX=xterm -e gdb --args
1396 Alternatively, you can stop a service that was started via ARM and run it manually:
1398 \lstset{language=bash}
1401 $ gdb --args gnunet-service-dht -L DEBUG
1402 $ valgrind gnunet-service-dht -L DEBUG
1406 Assuming other services are well-written, they will automatically re-integrate the
1407 restarted service with the peer.
1409 GNUnet provides a powerful logging mechanism providing log levels \texttt{ERROR},
1410 \texttt{WARNING}, \texttt{INFO} and \texttt{DEBUG}. The current log level is
1411 configured using the \lstinline|$GNUNET_FORCE_LOG| environmental variable.
1412 The \texttt{DEBUG} level is only available if \lstinline|--enable-logging=verbose| was used when
1413 running \texttt{configure}. More details about logging can be found under
1414 \url{https://gnunet.org/logging}.
1416 You should also probably enable the creation of core files, by setting
1417 {\tt ulimit}, and echo'ing 1 into {\tt /proc/sys/kernel/core\_uses\_pid}.
1418 Then you can investigate the core dumps with {\tt gdb}, which is often
1419 the fastest method to find simple errors.
1421 \exercise{Add a memory leak to your service and obtain a trace
1422 pointing to the leak using {\tt valgrind} while running the service
1423 from {\tt gnunet-service-arm}.}