GNUnet
GNUnet is a for decentralized, peer-to-peer and an official package. The framework offers , , , communication over many transports (such as , , , , and ) and various basic peer-to-peer algorithms for , and .
GNUnet’s basic network topology is that of a . GNUnet includes a distributed hash table (DHT) which is a variant of Kademlia that can still efficiently route in . GNUnet offers a “F2F topology” option for restricting connections to only the users’ trusted friends. The users’ friends’ own friends (and so on) can then indirectly exchange files with the users’ computer, never using its IP address directly.
GNUnet uses (not approved by , although an application has been made). GNUnet URIs consist of two major parts: the module and the module specific identifier. A GNUnet URI is of form gnunet://module/identifier where module is the module name and identifier is a module specific string.
The primary codebase is written in , but with there is an effort to produce an API for developing extensions in . GNUnet is part of the . It has gained interest to the after the revelations.
GNUnet consists of several subsystems, of which essential ones are Transport and Core subsystems. Transport subsystem provides insecure link-layer communications, while Core provides peer discovery and encryption. On top of the core subsystem various applications are built.
GNUnet includes various P2P applications in the main distribution of the framework, including filesharing, chat and VPN; additionally, a few external projects (such as ) are also extending the GNUnet infrastructure.
GNUnet is unrelated to the older P2P protocol. Gnutella is not an official GNU project while GNUnet is.
Contents
Transport
Originally, GNUnet used for underlying transport. Now GNUnet transport subsystem provides multiple options, such as TCP and SMTP.
The communication port, officially registered at , is 2086 ( + ).
Trust system
GNUnet provides trust system based on excess-based economic model. The idea of employing economic system is taken from network.
GNUnet network has no trusted entities so it is impossible to maintain global reputation. Instead, each peer maintains its own trust for each of its local links.
When resources, such as bandwidth and CPU time, are in excess, peer provides them to all requesting neighbors without reducing trust or otherwise charging them. When a node is under stress it drops requests from its neighbor nodes having lower internal trust value. However, when peer has less resources than enough to fulfill everyone’s requests, it denies requests of those neighbors that it trusts less and charges others by reducing their trust.
File sharing
The primary application at this point is anonymous, -resistant file-sharing, allowing users to anonymously publish or retrieve information of all kinds. GNUnet protocol that provides anonymity is called GAP (GNUnet anonymity protocol). GNUnet uses to automatically annotate shared files with .
File encoding
Files shared with GNUnet are ECRS (An Encoding for Censorship-Resistant Sharing) coded.
All content is represented as GBlocks. Each GBlock contains 1024 bytes. There are several types of GBlocks, each of them serves a particular purpose. Any GBlock <math>B</math> is uniquely identified by its RIPEMD-160 hash <math>H(B)</math>.
DBlocks store actual file contents and nothing else. File is split at 1024 byte boundaries and resulting chunks are stored in DBlocks. DBlocks are linked together into Merkle tree by means of IBlocks that store DBlock identifiers.
Blocks are encrypted with a symmetric key derived from <math>H(B)</math> when they are stored in the network.
Queries and replies
GNUnet Anonymity Protocol consists of queries and replies. Depending on load of the forwarding node, messages are forwarded to zero or more nodes.
Queries are used to search for content and request data blocks.
Query contains resource identifier, reply address, priority and TTL (Time-to-Live).
Resource identifier of datum <math>Q</math> is a triple-hash <math>H(H(H(Q)))</math>. Peer that replies to query provides <math>H(H(Q))</math>to prove that it indeed has the requested resource without providing <math>H(Q)</math> to intermediate nodes, so intermediate nodes can’t decrypt <math>Q</math>.
Reply address is the major difference compared to Freenet protocol. While in Freenet reply always propagates back using the same path as the query, in GNUnet the path may be shorter. Peer receiving a query may drop it, forward it without rewriting reply address or indirect it by replacing reply address with its own address. By indirecting queries peer provides for its own queries, while by forwarding them peer avoids being a link in reply propagation and preserves its bandwidth. This feature allows the user to trade anonymity for efficiency. User can specify an anonymity level for each publish, search and download operation. An anonymity level of zero can be used to select non-anonymous file-sharing. GNUnet’s DHT infrastructure is only used if non-anonymous file-sharing is specified. The anonymity level determines how much cover traffic a peer must have to hide the user’s own actions.
Priority specifies how much of its trust user wants to spend in case of resource shortage.
TTL is used to prevent queries from staying in the network for too long.
File sharing URIs
The fs module identifier consists of either chk, sks, ksk or loc followed by a slash and a category specific value. Most URIs contain hashes, which are encoded in .
- chk identifies files, typically: <code><nowiki>gnunet://fs/chk/[file hash].[query hash].[file size in bytes]</nowiki></code>
- File hash is the hash of the plaintext file, which allows to decrypt it once it is downloaded. Query hash is the hash of topmost GBlock which allows to download the whole tree of GBlocks that contain encrypted file. File size is required to determine the shape of the tree.
- sks identifies files within namespaces, typically: <code><nowiki>gnunet://fs/sks/NAMESPACE/IDENTIFIER</nowiki></code>
- ksk identifies search queries, typically: <code><nowiki>gnunet://fs/ksk/KEYWORD[+KEYWORD]*</nowiki></code>
- loc identifies a datum on a specific machine, typically: <code><nowiki>gnunet://fs/loc/PEER/QUERY.TYPE.KEY.SIZE</nowiki></code>
Examples
A type of GNUnet filesharing URI pointing to a specific copy of license text: <code style=”word-wrap:break-word”> gnunet://fs/chk/9E4MDN4VULE8KJG6U1C8FKH5HA8C5CHSJTILRTTPGK8MJ6VHORERHE68JU8Q0FDTOH1DGLUJ3NLE99N0ML0N9PIBAGKG7MNPBTT6UKG.1I823C58O3LKS24LLI9KB384LH82LGF9GUQRJHACCUINSCQH36SI4NF88CMAET3T3BHI93D4S0M5CC6MVDL1K8GFKVBN69Q6T307U6O.17992</code>
Another type of GNUnet filesharing URI, pointing to the search results of a search with keyword “gpl”: <code style=”word-wrap:break-word”> gnunet://fs/ksk/gpl </code>
GNU Name System
GNUnet includes an implementation of the GNU Name System (GNS), a decentralized and censorship-resistant replacement for . In GNS, each user manages his own master which is mapped into the DNS namespace under the .gnu . Users can delegate subdomains to zones managed by other users. Lookups of records defined by other users are performed using GNUnet’s DHT.
Protocol translation
GNUnet can tunnel IP traffic over the peer-to-peer network. If necessary, GNUnet can perform IPv4-IPv6 in the process. GNUnet provides a to proxy DNS requests and map addresses to the desired address family as necessary. This way, GNUnet offers a possible technology to facilitate . Furthermore, in combination with GNS, GNUnet’s protocol translation system can be used to access hidden services — IP-based services that run locally at some peer in the network and which can only be accessed by resolving a GNS name.
Social API
published in early September 2013 a thesis to present the design of a service for the GNUnet peer-to-peer framework that offers scalability, extensibility, and end-to-end encrypted communication. The scalability property is achieved through multicast message delivery, while extensibility is made possible by using (Protocol for SYnchronous Conferencing), which provides an extensible (Remote Procedure Call) syntax that can evolve over time without having to upgrade the software on all nodes in the network. Another key feature provided by the PSYC layer are stateful multicast channels, which are used to store e.g. user profiles. End-to-end encrypted communication is provided by the mesh service of GNUnet, upon which the multicast channels are built. Pseudonymous users and social places in the system have cryptographical identities — identified by their public key — these are mapped to human memorable names using GNS (), where each pseudonym has a zone pointing to its places.
That is the required building block for turning the GNUnet framework into a fully platform.
Chat
A trivial chat module has been implemented more as a test than as a serious application, but it is usable.