techintro.html 54.55 KiB
<html>
<head>
<title>Introducing I2P - a scalable framework for anonymous communication</title>
<style>
p { font-size: 10; text-align: left; font-family: sans-serif }
h1 { font-size: 12; font-family: sans-serif }
h2 { font-size: 10; font-family: sans-serif }
h3 { font-size: 10; font-family: sans-serif }
blockquote { font-size: 10; font-family: monospace, sans-serif }
pre { font-size: 10; font-family: sans-serif }
.title { font-size: 14; font-family: sans-serif }
.subtitle { font-size: 12; font-family: sans-serif }
</style>
</head>
<body>
<center>
<b class="title">Introducing I2P</b><br />
<span class="subtitle">a scalable framework for anonymous communication</span><br />
<i style="font-size: 8">$Id: techintro.html,v 1.6 2005/10/04 20:11:25 jrandom Exp $</i>
<br />
<br />
<table border="0" width="50%">
<tr><td valign="top" align="left">
<pre>
* <a href="#intro">Introduction</a>
* <a href="#op">Operation</a>
* <a href="#op.overview">Overview</a>
* <a href="#op.tunnels">Tunnels</a>
* <a href="#op.netdb">Network Database</a>
* <a href="#op.transport">Transport protocols</a>
* <a href="#op.crypto">Cryptography</a>
</pre>
</td>
<td valign="top" align="left">
<pre>
* <a href="#future">Future</a>
* <a href="#future.restricted">Restricted routes</a>
* <a href="#future.variablelatency">Variable latency</a>
* <a href="#future.open">Open questions</a>
</pre>
</td>
<td valign="top" align="left">
<pre>
* <a href="#similar">Similar systems</a>
* <a href="#similar.tor">Tor</a>
* <a href="#similar.freenet">Freenet</a>
* <a href="#app">Appendix A: Application layer</a>
</pre>
</td>
</tr></table>
</center>
<hr />
<h1 id="intro">Introduction</h1>
<p>
I2P is a scalable, self organizing, resilient message based anonymous network layer,
upon which any number of different anonymity or security conscious applications
can operate. Each of these applications may make their own anonymity, latency, and
throughput tradeoffs without worrying about the proper implementation of a free
route mixnet, allowing them to blend their activity with the larger anonymity set of
users already running on top of I2P. Applications available already provide the full
range of typical Internet activities - anonymous web browsing, anonymous web hosting,
anonymous blogging (with <a href="#app.syndie">Syndie</a>), anonymous chat (via IRC or
Jabber), anonymous swarming file transfers (with <a href="#app.i2pbt">i2p-bt</a> and
<a href="#app.azneti2p">Azureus</a>), anonymous file sharing (with
<a href="#app.i2phex">I2Phex</a>), anonymous email (with <a href="#app.i2pmail">I2Pmail</a>
and <a href="#app.i2pmail">susimail</a>), anonymous newsgroups, as well as severalother applications under development. Unlike web sites hosted within content
distribution networks like <a href="#similar.freenet">Freenet</a> or
<a href="http://www.ovmj.org/GNUnet/">GNUnet</a>, the services hosted on I2P are fully
interactive - there are traditional web-style search engines, bulletin boards, blogs
you can comment on, database driven sites, and bridges to query static systems like
Freenet without needing to install it locally.
</p>
<p>
With all of these anonymity enabled applications, I2P takes on the role of the message
oriented middleware - applications say that they want to send some data to a cryptographic
identifier (a "destination") and I2P takes care of making sure it gets there securely
and anonymously. I2P also bundles a simple <a href="#app.streaming">streaming</a> library
to allow I2P's anonymous best-effort messages to transfer as reliable, in-order streams,
transparently offering a TCP based congestion control algorithm tuned for the high
bandwidth delay product of the network. While there have been several simple SOCKS
proxies available to tie existing applications into the network, their value has been
limited as nearly every application routinely exposes what in an anonymity context is
sensitive information. The only safe way to go is to fully audit an application to
ensure proper operation, and to assist in that we provide a series of APIs in various
languages which can be used to make the most out of the network.
</p>
<!-- commented out because "The details [...] are " *NOT* " given later" -->
<!--
<p>
The scope of I2P's anonymity protections varies upon the applications running on
top of them, as well as the choices that each user makes. The aim is to provide
the options necessary so that a sufficient level of anonymity can be achieved while
exposing the functionality that people facing up to state level adversaries require.
At the same time, those facing less powerful adversaries are able to improve their
throughput and latency while reducing the resources required to provide the necessary
level of cover. The details of the techniques available for facing adversaries who
are internal or external, passive or active, local, national, or global, are given
later.
</p>
-->
<p>
I2P is not a research project - academic, commercial, or governmental, but is instead
an engineering effort aimed at doing whatever is necessary to provide a sufficient
level of anonymity to those who need it. It has been in active development since
early 2003 with one full time developer and a dedicated group of part time contributors
from all over the world. All of the work done on I2P is open source and
freely available on the <a href="http://www.i2p.net/">website</a>, with the majority
of the code released outright into the public domain but making use of a few
cryptographic routines under BSD-style licenses. The people working on I2P do not
control what people release client applications under, and there are several GPL'ed
applications available (<a href="#app.i2ptunnel">I2PTunnel</a>,
<a href="#app.i2pmail">susimail</a>, <a href="#app.azneti2p">Azureus</a>,
<a href="#app.i2phex">I2Phex</a>). <a href="http://www.i2p.net/halloffame">Funding</a>
for I2P comes entirely from donations, and does not receive any tax breaks in any
jurisdiction, as many of the developers are themselves anonymous.
</p>
<h1 id="op">Operation</h1>
<h2 id="op.overview">Overview</h2>
<p>
To understand I2P's operation, it is essential to understand a few key concepts.
First, I2P makes a strict separation between the software participating
in the network (a "router") and the anonymous endpoints ("destinations") associated
with individual applications. The fact that someone is running I2P is not usually
a secret. What is hidden is information on what the user is doing, if anything at
all, as well as what router a particular destination is connected to. End users
will typically have several local destinations on their router - for instance, one
proxying in to IRC servers, another supporting the user's anonymous webserver ("eepsite"),
another for an I2Phex instance, another for torrents, etc.
</p>
<p>
Another critical concept to understand is the "tunnel" - a directed path through
an explicitly selected set of routers, making use of layered encryption so that
the messages sent in the tunnel's "gateway" appear entirely random at each hop
along the path until it reaches the tunnel's "endpoint". These unidirectional
tunnels can be seen as either "inbound" tunnels or "outbound" tunnels, referring
to whether they are bringing messages to the tunnel's creator or away from them,
respectively. The gateway of an inbound tunnel can receive messages from any
peer and will forward them down through the tunnel until it reaches the (anonymous)
endpoint (the creator). On the other hand, the gateway of an outbound tunnel is
the tunnel's creator, and messages sent through that tunnel are encoded so that
when they reach the outbound tunnel's endpoint, that router has the instructions
necessary to forward the message on to the appropriate location.
</p>
<p>
A third critical concept to understand is I2P's "network database" (or "netDb")
- a pair of algorithms used to share network metadata. The two types of metadata
carried are "routerInfo" and "leaseSets" - the routerInfo gives routers the data
necessary for contacting a particular router (their public keys, transport
addresses, etc), while the leaseSet gives routers the information necessary for
contacting a particular destination. Within each leaseSet, there are any number
of "leases", each of which specifies the gateway for one of that destination's
inbound tunnels as well as when that tunnel will expire. The leaseSet also
contains a pair of public keys which can be used for layered garlic encryption.
</p>
<p>
I2P's operation can be understood by putting those three concepts together:
</p>
<p><img src="net.png"></p>
<p>
When Alice wants to send a message to Bob, she first does a lookup in the
netDb to find Bob's leaseSet, giving her his current inbound tunnel gateways
(3 and 4). She then picks one of her outbound tunnels and sends the message
down it with instructions for the outbound tunnel's endpoint to forward the
message on to one of Bob's inbound tunnel gateways. When the outbound
tunnel endpoint receives those instructions, it forwards the message as
requested, and when Bob's inbound tunnel gateway receives it, it is
forwarded down the tunnel to Bob's router. If Alice wants Bob to be able
to reply to the message, she needs to transmit her own destination explicitly
as part of the message itself (taken care of transparently in the
<a href="#app.streaming">streaming</a> library). Alice may also cut down on
the response time by bundling her most recent leaseSet with the message so
that Bob doesn't need to do a netDb lookup for it when he wants to reply, but this
is optional.
</p>
<p>
While the tunnels themselves have layered encryption to prevent unauthorized
disclosure to peers inside the network (as the transport layer itself does to
prevent unauthorized disclosure to peers outside the network), it is necessary
to add an additional end to end layer of encryption to hide the message from the
outbound tunnel endpoint and the inbound tunnel gateway. This
"<a href="#op.garlic">garlic encryption</a>" lets Alice's router wrap up multiple
messages into a single "garlic message", encrypted to a particular public key
so that intermediary peers cannot determine either how many messages are within
the garlic, what those messages say, or where those individual cloves are
destined. For typical end to end communication between Alice and Bob, the
garlic will be encrypted to the public key published in Bob's leaseSet,
allowing the message to be encrypted without giving out the public key to Bob's
own router.
</p>
<p>
Another important fact to keep in mind is that I2P is entirely message based
and that some messages may be lost along the way. Applications using I2P
can use the message oriented interfaces and take care of their own congestioncontrol and reliability needs, but most would be best served by reusing the
provided <a href="#app.streaming">streaming</a> library to view I2P as a streams
based network.
</p>
<h2 id="op.tunnels">Tunnels</h2>
<p>
Both inbound and outbound tunnels work along similar principles - the tunnel
gateway accumulates a number of tunnel messages, eventually preprocessing them
into something for tunnel delivery. Next, the gateway encrypts that preprocessed
data and forwards it to the first hop. That peer and subsequent tunnel
participants add on a layer of encryption after verifying that it isn't a
duplicate before forward it on to the next peer. Eventually, the
message arrives at the endpoint where the messages are split out again and
forwarded on as requested. The difference arises in what
the tunnel's creator does - for inbound tunnels, the creator is the endpoint
and they simply decrypt all of the layers added, while for outbound tunnels,
the creator is the gateway and they pre-decrypt all of the layers so that after
all of the layers of per-hop encryption are added, the message arrives in the
clear at the tunnel endpoint.
</p>
<p>
The choice of specific peers to pass on messages as well as their particular
ordering is important to understanding both I2P's anonymity and performance
characteristics. While the network database (below) has its own criteria for
picking what peers to query and store entries on, tunnels may use any peers in
the network in any order (and even any number of times) in a single tunnel. If
perfect latency and capacity data were globally known, selection and ordering
would be driven by the particular needs of the client in tandem with their threat
model. Unfortunately, latency and capacity data is not trivial to gather
anonymously, and depending upon untrusted peers to provide this information has
its own serious anonymity implications.
</p>
<p>
From an anonymity perspective, the simplest technique would be to pick peers
randomly from the entire network, order them randomly, and use those peers
in that order for all eternity. From a performance perspective, the simplest
technique would be to pick the fastest peers with the necessary spare capacity,
spreading the load across different peers to handle transparent failover, and
to rebuild the tunnel whenever capacity information changes. While the former
is both brittle and inefficient, the later requires inaccessible information
and offers insufficient anonymity. I2P is instead working on offering a range
of peer selection strategies, coupled with anonymity aware measurement code to
organize the peers by their profiles.
</p>
<p>
As a base, I2P is constantly profiling the peers with which it interacts with
by measuring their indirect behavior - for instance, when a peer responds to
a netDb lookup in 1.3 seconds, that round trip latency is recorded in the
profiles for all of the routers involved in the two tunnels (inbound and
outbound) through which the request and response passed, as well as the queried
peer's profile. Direction measurement, such as transport layer latency or
congestion, is not used as part of the profile, as it can be manipulated and
associated with the measuring router, exposing them to trivial attacks. While
gathering these profiles, a series of calculations are run on each to summarize
its performance - its latency, capacity to handle lots of activity, whether they
are currently overloaded, and how well integrated into the network they seem to
be. These calculations are then compared for active peers to organize the routers
into four tiers - fast and high capacity, high capacity, not failing, and failing.
The thresholds for those tiers are determined dynamically, and while they
currently use fairly simple algorithms, alternatives exist.
</p>
<p>
Using this profile data, the simplest reasonable peer selection strategy is to
pick peers randomly from the top tier (fast and high capacity), and this iscurrently deployed for client tunnels. Exploratory tunnels (used for netDb
and tunnel management) pick peers randomly from the not failing tier (which
includes routers in 'better' tiers as well), allowing the peer to sample
routers more widely, in effect optimizing the peer selection through randomized
hill climbing. These strategies alone do however leak information regarding the
peers in the router's tip tier through predecessor and netDb harvesting attacks.
In turn, several alternatives exist which, while not balancing the load as evenly,
will address the attacks mounted by particular classes of adversaries.
</p>
<p>
By picking a random key and ordering the peers according to their XOR distance
from it, the information leaked is reduced in predecessor and harvesting attacks
according to the peers' failure rate and the tier's churn. Another simple strategy
for dealing with netDb harvesting attacks is to simply fix the inbound tunnel
gateway(s) yet randomize the peers further on in the tunnels. To deal with
predecessor attacks for adversaries which the client contacts, the outbound tunnel
endpoints would also remain fixed. The selection of which peer to fix on the most
exposed point would of course need to have a limit to the duration, as all peers
fail eventually, so it could either be reactively adjusted or proactively avoided
to mimic a measured mean time between failures of other routers. These two strategies
can in turn be combined, using a fixed exposed peer and an XOR based ordering within
the tunnels themselves. A more rigid strategy would fix the exact peers and ordering
of a potential tunnel, only using individual peers if all of them agree to participate
in the same way each time. This varies from the XOR based ordering in that the
predecessor and successor of each peer is always the same, while the XOR only makes
sure their order doesn't change.
</p>
<p>
As mentioned before, I2P currently (release 0.6.1.1) includes the tiered random
strategy above, but the others are planned for the 0.6.2 release. A more detailed
discussion of the mechanics involved in tunnel operation, management, and peer
selection can be found in the
<a href="http://dev.i2p.net/cgi-bin/cvsweb.cgi/i2p/router/doc/tunnel-alt.html?rev=HEAD">tunnel spec</a>.
</p>
<h2 id="op.netdb">Network Database</h2>
<p>
As mentioned earlier, I2P's netDb works to share the network's metadata. Two
algorithms are used to accomplish this - primarily, a small set of routers are
designated as "floodfill peers", while the rest of the routers participate in
the <a href="http://en.wikipedia.org/wiki/Kademlia">Kademlia </a> derived
distributed hash table for redundancy. To integrate the two algorithms, each
router always uses the Kademlia style store and fetch, but acts as if the
floodfill peers are 'closest' to the key in question. Additionally, when a
peer publishes a key into the netDb, after a brief delay they query another
random floodfill peer, asking them for the key, and if that peer does not have
it, they move on and republish the key again. Behind the scenes, when one of
the floodfill peers receives a new valid key, they republish it to the other
floodfill peers who then cache it locally.
</p>
<p>
Each piece of data in the netDb is self authenticating - signed by the
appropriate party and verified by anyone who uses or stores it. In addition,
the data has liveliness information within it, allowing irrelevant entries to be
dropped, newer entries to replace older ones, and, for the paranoid, protection
against certain classes of attack. This is also why I2P bundles the necessary
code for maintaining the correct time, occasionally querying some SNTP servers
(the <a href="http://www.pool.ntp.org/">pool.ntp.org</a> round robin by default)
and detecting skew between routers at the transport layer.
</p>
<p>
The routerInfo structure itself contains all of the information that one router
needs to know to securely send messages to another router. This includes their
identity (made up of a 2048bit ElGamal public key, a 1024bit DSA public key, and
a certificate), the transport addresses which they can be reached on, such asan IP address and port, when the structure was published, and a set of arbitrary
uninterpreted text options. In addition, there is a signature against all of
that data as generated by the included DSA public key. The key for this routerInfo
structure in the netDb is the SHA256 hash of the router's identity. The options
published are often filled with information helpful in debugging I2P's operation,
but when I2P reaches the 1.0 release, the options will be disabled and kept blank.
</p>
<p>
The leaseSet structure is similar, in that it includes the I2P destination
(comprised of a 2048bit ElGamal public key, a 1024bit DSA public key, and a
certificate), a list of "leases", and a pair of public keys for garlic encrypting
messages to the destination. Each of the leases specify one of the destination's
inbound tunnel gateways by including the SHA256 of the gateway's identity, a 4
byte tunnel id on that gateway, and when that tunnel will expire. The key for
the leaseSet in the netDb is the SHA256 of the destination itself.
</p>
<p>
As the router currently automatically bundles the leaseSet for the sender inside
a garlic message to the recipient, the leaseSet for destinations which will not
receive unsolicited messages do not need to be published in the netDb at all. If
the destination itself is sensitive, the leaseSet could instead be transmitted
through other means without ever going into the netDb.
</p>
<p>
Bootstrapping the netDb itself is simple - once a router has at least one routerInfo
of a reachable peer, they query that router for references to other routers in the
network with the Kademlia healing algorithm. Each routerInfo reference is stored in
an individual file in the router's netDb subdirectory, allowing people to easily
share their references to bootstrap new users.
</p>
<p>
Unlike traditional DHTs, the very act of conducting a search distributes the data
as well, since rather passing Kademlia's standard IP+port pairs, references are given
to the routers that the peer should query next (namely, the SHA256 of those routers'
identities). As such, iteratively searching for a particular destination's leaseSet
or router's routerInfo will also provide you with the routerInfo of the peers along
the way. In addition, due to the time sensitivity of the data published, the information
doesn't often need to migrate between peers - since a tunnel is only valid for 10
minutes, the leaseSet can be dropped after that time has passed. To take into
account Sybil attacks on the netDb, the Kademlia routing location used for any given
key varies over time. For instance, rather than storing a routerInfo on the peers
closest to SHA256(routerInfo.identity), they are stored on the peers closest to
SHA256(routerInfo.identity + YYYYMMDD), requiring an adversary to remount the attack
again daily so as to maintain their closeness to the current routing key. As the
very fact that a router is making a lookup for a given key may expose sensitive data
(and the fact that a router is <i>publishing</i> a given key even more so), all netDb
messages are transmitted through the router's exploratory tunnels.
</p>
<p>
The netDb plays a very specific role in the I2P network, and the algorithms have
been tuned towards our needs. This also means that it hasn't been tuned to address the
needs we have yet to run into. As the network grows, the primary floodfill algorithm
will need to be refined to exploit the capacity available, or perhaps replaced with
another technique for securely distributing the network metadata.
</p>
<h2 id="op.transport">Transport protocols</h2>
<p>
Communication between routers needs to provide confidentiality and integrity
against external adversaries while authenticating that the router contacted
is the one who should receive a given message. The particulars of how routers
communicate with other routers aren't critical - three separate protocols have
been used at different points to provide those bare necessities. To accommodate
the need for high degree communication (as a number of routers will end up speaking with many others), I2P moved from a TCP based transport
to a UDP based one - "Secure Semireliable UDP", or "SSU". As described in the
<a href="http://dev.i2p.net/cgi-bin/cvsweb.cgi/i2p/router/doc/udp.html?rev=HEAD">SSU spec</a>:</p>
<blockquote>
The goal of this protocol is to provide secure, authenticated,
semireliable, and unordered message delivery, exposing only a minimal amount of
data easily discernible to third parties. It should support high degree
communication as well as TCP-friendly congestion control, and may include
PMTU detection. It should be capable of efficiently moving bulk data at rates
sufficient for home users. In addition, it should support techniques for
addressing network obstacles, like most NATs or firewalls.
</blockquote>
<h2 id="op.crypto">Cryptography</h2>
<p>
A bare minimum set of cryptographic primitives are combined together to provide I2P's
layered defenses against a variety of adversaries. At the lowest level, interrouter
communication is protected by the transport layer security - SSU
encrypts each packet with AES256/CBC with both an explicit IV and MAC (HMAC-SHA256-128)
after agreeing upon an ephemeral session key through a 2048bit Diffie-Hellman exchange,
station-to-station authentication with the other router's DSA key, plus each network
message has their own SHA256 hash for local integrity checking.
<a href="#op.tunnels">Tunnel</a> messages passed over the transports have their own
layered AES256/CBC encryption with an explicit IV and verified at the tunnel endpoint
with an additional SHA256 hash. Various other messages are passed along inside
"garlic messages", which are encrypted with ElGamal/AES+SessionTags (explained below).
</p>
<h3 id="op.garlic">Garlic messages</h3>
<p>
Garlic messages are an extension of "onion" layered encryption, allowing the contents
of a single message to contain multiple "cloves" - fully formed messages along side
their own instructions for delivery. Messages are wrapped into a garlic message whenever
the message would otherwise be passing in cleartext through a peer who should not have
access to the information - for instance, when a router wants to ask another router to
participate in a tunnel, they wrap the request inside a garlic, encrypt that garlic to
the receiving router's 2048bit ElGamal public key, and forward it through a tunnel.
Another example is when a client wants to send a message to a destination - the sender's
router will wrap up that data message (along side some other messages) into a garlic,
encrypt that garlic to the 2048bit ElGamal public key published in the recipient's
leaseSet, and forward it through the appropriate tunnels.
</p>
<p>
The "instructions" attached to each clove inside the encryption layer includes the
ability to request that the clove be forwarded locally, to a remote router, or to a
remote tunnel on a remote router. There are fields in those instructions allowing a
peer to request that the delivery be delayed until a certain time or condition has
been met, though they won't be honored until the
<a href="#future.variablelatency">nontrivial delays</a> are deployed. It is possible to
explicitly route garlic messages any number of hops without building tunnels, or even
to reroute tunnel messages by wrapping them in garlic messages and forwarding them a
number of hops prior to delivering them to the next hop in the tunnel, but those
techniques are not currently used in the existing implementation.
</p>
<h3 id="op.sessiontags">Session tags</h3>
<p>
As an unreliable, unordered, message based system, I2P uses a simple combination of
asymmetric and symmetric encryption algorithms to provide data confidentiality and
integrity to garlic messages. As a whole, the combination is referred to as
ElGamal/AES+SessionTags, but that is an excessively verbose way to describe the simple
use of 2048bit ElGamal, AES256, SHA256, and 32 byte nonces.
</p>
<p>The first time a router wants to encrypt a garlic message to another router, they encrypt
the keying material for an AES256 session key with ElGamal and append the AES256/CBC
encrypted payload after that encrypted ElGamal block. In addition to the encrypted
payload, the AES encrypted section contains the payload length, the SHA256 hash of the
unencrypted payload, as well as a number of "session tags" - random 32 byte nonces. The
next time the sender wants to encrypt a garlic message to another router, rather than
ElGamal encrypt a new session key they simply pick one of the previously delivered session
tags and AES encrypt the payload like before, using the session key used with that
session tag, prepended with the session tag itself. When a router receives a garlic encrypted
message, they check the first 32 bytes to see if it matches an available session tag - if
it does, they simply AES decrypt the message, but if it does not, they ElGamal decrypt the
first block.
</p>
<p>
Each session tag can be used only once so as to prevent internal adversaries from unnecessarily
correlating different messages as being between the same routers. The sender of an
ElGamal/AES+SessionTag encrypted message chooses when and how many tags to deliver,
prestocking the recipient with enough tags to cover a volley of messages. Garlic messages
may detect the successful tag delivery by bundling a small additional message as a clove (a
"delivery status message") - when the garlic message arrives at the intended recipient and
is decrypted successfully, this small delivery status message is one of the cloves exposed and
has instructions for the recipient to send the clove back to the original sender (through an
inbound tunnel, of course). When the original sender receives this delivery status message,
they know that the session tags bundled in the garlic message were successfully delivered.
</p>
<p>
Session tags themselves have a very short lifetime, after which they are discarded
if not used. In addition, the quantity stored for each key is limited, as are the
number of keys themselves - if too many arrive, either new or old messages may be
dropped. The sender keeps track whether messages using session tags are getting
through, and if there isn't sufficient communication it may drop the ones previously
assumed to be properly delivered, reverting back to the full expensive ElGamal
encryption.
</p>
<p>
One alternative is to transmit only a single session tag, and from that, seed a
deterministic PRNG for determining what tags to use or expect. By keeping this
PRNG roughly synchronized between the sender and recipient (the recipient precomputes a
window of the next e.g. 50 tags), the overhead of periodically bundling a large number
of tags is removed, allowing more options in the space/time tradeoff, and perhaps
reducing the number of ElGamal encryptions necessary. However, it would depend
upon the strength of the PRNG to provide the necessary cover against internal
adversaries, though perhaps by limiting the amount of times each PRNG is used, any
weaknesses can be minimized. At the moment, there are no immediate plans to move
towards these synchronized PRNGs.
</p>
<h1 id="future">Future</h1>
<p>
While I2P is currently functional and sufficient for many scenarios, there are
several areas which require further improvement to meet the needs of those
facing more powerful adversaries as well as substantial user experience optimization.
</p>
<h2 id="future.restricted">Restricted route operation</h2>
<p>
I2P is an overlay network designed to be run on top of a functional packet switched
network, exploiting the end to end principle to offer anonymity and security.
While the Internet no longer fully embraces the end to end principle, I2P does require a
substantial portion of the network to be reachable - there may be a number of peers
along the edges running using restricted routes, but I2P does not include an
appropriate routing algorithm for the degenerate case where most peers are
unreachable. It would, however work on top of a network employing such an
algorithm.
</p>
<p>
Restricted route operation, where there are limits to what peers are
reachable directly, has several different functional and anonymity
implications, dependent upon how the restricted routes are handled. At the most
basic level, restricted routes exist when a peer is behind a NAT or firewall which
does not allow inbound connections. This was largely addressed in I2P 0.6.0.6 by
integrating distributed hole punching into the transport layer, allowing people
behind most NATs and firewalls to receive unsolicited connections without any
configuration. However, this does not limit the exposure of the peer's IP address to
routers inside the network, as they can simply get introduced to the peer through
the published introducer.
</p>
<p>
Beyond the functional handling of restricted routes, there are two levels of
restricted operation that can be used to limit the exposure of one's IP address -
using router-specific tunnels for communication, and offering 'client routers'. For
the former, routers can either build a new pool of tunnels or reuse their exploratory
pool, publishing the inbound gateways to some of them as part of their routerInfo in
place of their transport addresses. When a peer wants to get in touch with them,
they see those tunnel gateways in the netDb and simply send the relevant message to
them through one of the published tunnels. If the peer behind the restricted route
wants to reply, it may do so either directly (if they are willing to expose their IP
to the peer) or indirectly through their outbound tunnels. When the routers that the
peer has direct connections to want to reach it (to forward tunnel messages, for
instance), they simply prioritize their direct connection over the published tunnel
gateway. The concept of 'client routers' simply extends the restricted route by not
publishing any router addresses. Such a router would not even need to publish their
routerInfo in the netDb, merely providing their self signed routerInfo to the peers
that it contacts (necessary to pass the router's public keys). Both levels of
restricted route operation are planned for I2P 2.0.
</p>
<p>
There are tradeoffs for those behind restricted routes, as they would likely
participate in other people's tunnels less frequently, and the routers which
they are connected to would be able to infer traffic patterns that would not
otherwise be exposed. On the other hand, if the cost of that exposure is less
than the cost of an IP being made available, it may be worthwhile. This, of course,
assumes that the peers that the router behind a restricted route contacts are not
hostile - either the network is large enough that the probability of using a hostile
peer to get connected is small enough, or trusted (and perhaps temporary) peers are
used instead.
</p>
<h2 id="future.variablelatency">Variable latency</h2>
<p>
Even though the bulk of I2P's initial efforts have been on low latency communication,
it was designed with variable latency services in mind from the beginning. At the
most basic level, applications running on top of I2P can offer the anonymity of
medium and high latency communication while still blending their traffic patterns
in with low latency traffic. Internally though, I2P can offer its own medium and
high latency communication through the garlic encryption - specifying that the
message should be sent after a certain delay, at a certain time, after a certain
number of messages have passed, or another mix strategy. With the layered encryption,
only the router that the clove exposed the delay request would know that the message
requires high latency, allowing the traffic to blend in further with the low latency
traffic. Once the transmission precondition is met, the router holding on to the
clove (which itself would likely be a garlic message) simply forwards it as
requested - to a router, to a tunnel, or, most likely, to a remote client destination.
</p>
<p>
There are a substantial number of ways to exploit this capacity for high latency
comm in I2P, but for the moment, doing so has been scheduled for the I2P 3.0 release.
In the meantime, those requiring the anonymity that high latency comm can offer should
look towards the application layer to provide it.
</p>
<h2 id="future.open">Open questions</h2>
<pre>
How to get rid of the timing constraint?
Can we deal with the sessionTags more efficiently?
What, if any, batching/mixing strategies should be made available on the tunnels?
What other tunnel peer selection and ordering strategies should be available?
</pre>
<h1 id="similar">Similar systems</h1>
<p>
I2P's architecture builds on the concepts of message oriented middleware, the topology
of DHTs, the anonymity and cryptography of free route mixnets, and the adaptability of
packet switched networking. The value comes not from novel concepts of algorithms
though, but from careful engineering combining the research results of existing
systems and papers. While there are a few similar efforts worth reviewing, both for
technical and functional comparisons, two in particular are pulled out here - Tor
and Freenet.
</p>
<h2 id="similar.tor">Tor</h2>
<p><i><a href="http://tor.eff.org/">website</a></i></p>
<p>
At first glance, Tor and I2P have many functional and anonymity related similarities.
While I2P's development began before we were aware of the early stage efforts on Tor,
many of the lessons of the original onion routing and ZKS efforts were integrated into
I2P's design. Rather than building an essentially trusted, centralized system with
directory servers, I2P has a self organizing network database with each peer taking on
the responsibility of profiling other routers to determine how best to exploit available
resources. Another key difference is that while both I2P and Tor use layered and
ordered paths (tunnels and circuits/streams), I2P is fundamentally a packet switched
network, while Tor is fundamentally a circuit switched one, allowing I2P to
transparently route around congestion or other network failures, operate redundant
pathways, and load balance the data across available resources. While Tor offers
the useful outproxy functionality by offering integrated outproxy discovery and
selection, I2P leaves such application layer decisions up to applications running on
top of I2P - in fact, I2P has even externalized the TCP-like streaming library itself
to the application layer, allowing developers to experiment with different strategies,
exploiting their domain specific knowledge to offer better performance.
</p>
<p>
From an anonymity perspective, there is much similarity when the core networks are
compared. However, there are a few key differences. When dealing with an internal
adversary or most external adversaries, I2P's simplex tunnels expose half as much
traffic data than would be exposed with Tor's duplex circuits by simply looking at
the flows themselves - an HTTP request and response would follow the same path in
Tor, while in I2P the packets making up the request would go out through one or
more outbound tunnels and the packets making up the response would come back through
one or more different inbound tunnels. While I2P's peer selection and ordering
strategies should sufficiently address predecessor attacks, I2P can trivially
mimic Tor's non-redundant duplex tunnels by simply building an inbound and
outbound tunnel along the same routers.</p>
<p>
Another anonymity issue comes up in Tor's use of telescopic tunnel creation, as
simple packet counting and timing measurements as the cells in a circuit pass
through an adversary's node exposes statistical information regarding where the
adversary is within the circuit. I2P's use of exploratory tunnels for delivering
and receiving the tunnel creation requests and responses effectively spreads the
messages randomly across the network, so that each of the peers who forwards the
individual tunnel creation messages only see the peer they transmit to or receive
from, and thanks to the garlic encryption, they are not aware of whether the message
is part of a tunnel creation process or not. The participant positional information
is useful to an adversary for mounting predecessor, intersection, and traffic
confirmation attacks.
</p>
<p>
Tor's support for a second tier of "onion proxies" does offer a nontrivial degreeof anonymity while requiring a low cost of entry, while I2P will not offer this
topology until <a href="#future.restricted">2.0</a>.
</p>
<p>
On the whole, Tor and I2P complement each other in their focus - Tor works towards
offering high speed anonymous Internet outproxying, while I2P works towards offering
a decentralized resilient network in itself. In theory, both can be used to achieve
both purposes, but given limited development resources, they both have their
strengths and weaknesses. The I2P developers have considered the steps necessary to
modify Tor to take advantage of I2P's design, but concerns of Tor's viability under
resource scarcity suggest that I2P's packet switching architecture will be able to
exploit scarce resources more effectively.
</p>
<h2 id="similar.freenet">Freenet</h2>
<p><i><a href="http://www.freenetproject.org/">website</a></i></p>
<p>
Freenet played a large part in the initial stages of I2P's design - giving proof to
the viability of a vibrant pseudonymous community completely contained within the
network, demonstrating that the dangers inherent in outproxies could be avoided.
The first seed of I2P began as a replacement communication layer for Freenet,
attempting to factor out the complexities of a scalable, anonymous and secure point
to point communication from the complexities of a censorship resistant distributed
data store. Over time however, some of the anonymity and scalability issues
inherent in Freenet's algorithms made it clear that I2P's focus should stay strictly
on providing a generic anonymous communication layer, rather than as a component of
Freenet. Over the years, the Freenet developers have come to see the weaknesses
in the older design, prompting them to suggest that they will require a "premix"
layer to offer substantial anonymity. In other words, Freenet needs to run on top
of a mixnet such as I2P or Tor, with "client nodes" requesting and publishing data
through the mixnet to the "server nodes" which then fetch and store the data according
to Freenet's heuristic distributed data storage algorithms.
</p>
<p>
Freenet's functionality is very complementary to I2P's, as Freenet natively provides
many of the tools for operating medium and high latency systems, while I2P natively
provides the low latency mix network suitable for offering adequate anonymity. The
logic of separating the mixnet from the censorship resistant distributed data store
still seems self evident from an engineering, anonymity, security, and resource
allocation perspective, so hopefully the Freenet team will pursue efforts in that
direction, if not simply reusing (or helping to improve, as necessary) existing
mixnets like I2P or Tor.
</p>
<p>
It is worth mentioning that there has recently been discussion and work by the
Freenet developers on a "globally scalable darknet" using restricted routes between
peers of various trust. While insufficient information has been made publicly
available regarding how such a system would operate for a full review, from what
has been said the anonymity and scalability claims seem highly dubious. In
particular, the appropriateness for use in hostile regimes against state level
adversaries has been tremendously overstated, and any analysis on the implications
of resource scarcity upon the scalability of the network has seemingly been avoided.
Specifically, while publishing the "anonymous" topology in the darknet does not
necessarily immediately expose all identities, it is equivalent to publishing an
organizational chart for a covert group, which can in turn be used by an adversary
along side existing knowledge of their target to narrow down or identify different
participants. In addition, by using only peers that are locally connected, the
network's mixnet layer is vulnerable to a class of
<a href="http://www.im.pwr.wroc.pl/~klonowsk/LocalViewAttack.ps">local view attacks</a>.
Further questions regarding susceptibility to traffic analysis, trust, and other topics
do exist, but a more in depth review of this "globally scalable darknet" will have
to wait until the Freenet team makes more information available.
</p>
<h1 id="app">Appendix A: Application layer</h1>
<p>
I2P itself doesn't really do much - it simply sends messages to remote destinations
and receives messages targeting local destinations - most of the interesting work
goes on at the layers above it. By itself, I2P could be seen as an anonymous and
secure IP layer, and the bundled <a href="#app.streaming">streaming library</a> as
an implementation of an anonymous and secure TCP layer on top of it. Beyond that,
<a href="#app.i2ptunnel">I2PTunnel</a> exposes a generic TCP proxying system for
either getting into or out of the I2P network, plus a variety of network
applications provide further functionality for end users.
</p>
<h2 id="app.streaming">Streaming library</h2>
<p>
The streaming library has grown organically for I2P - first mihi implemented the
"mini streaming library" as part of I2PTunnel, which was limited to a window
size of 1 message (requiring an ACK before sending the next one), and then it was
refactored out into a generic streaming interface (mirroring TCP sockets) and the
full streaming implementation was deployed with a sliding window protocol and
optimizations to take into account the high bandwidth x delay product. Individual
streams may adjust the maximum packet size and other options, though the default
of 4KB compressed seems a reasonable tradeoff between the bandwidth costs of
retransmitting lost messages and the latency of multiple messages.
</p>
<p>
In addition, in consideration of the relatively high cost of subsequent messages,
the streaming library's protocol for scheduling and delivering messages has been optimized to
allow individual messages passed to contain as much information as is available.
For instance, a small HTTP transaction proxied through the streaming library can
be completed in a single round trip - the first message bundles a SYN, FIN, and
the small payload (an HTTP request typically fits) and the reply bundles the SYN,
FIN, ACK, and the small payload (many HTTP responses fit). While an additional
ACK must be transmitted to tell the HTTP server that the SYN/FIN/ACK has been
received, the local HTTP proxy can deliver the full response to the browser
immediately.
</p>
<p>
On the whole, however, the streaming library bears much resemblance to an
abstraction of TCP, with its sliding windows, congestion control algorithms
(both slow start and congestion avoidance), and general packet behavior (ACK,
SYN, FIN, RST, rto calculation, etc).
</p>
<h2 id="app.naming">Naming library and addressbook</h2>
<p><i>Developed by: mihi, Ragnarok</i></p>
<p>
Naming within I2P has been an oft-debated topic since the very beginning with
advocates across the spectrum of possibilities. However, given I2P's inherent
demand for secure communication and decentralized operation, the traditional
DNS-style naming system is clearly out, as are "majority rules" voting systems.
Instead, I2P ships with a generic naming library and a base implementation
designed to work off a local name to destination mapping, as well as an optional
add-on application called the "addressbook". The addressbook is a web-of-trust
driven secure, distributed, and human readable naming system, sacrificing only
the call for all human readable names to be globally unique by mandating only
local uniqueness. While all messages in I2P are cryptographically addressed
by their destination, different people can have local addressbook entries for
"Alice" which refer to different destinations. People can still discover new
names by importing published addressbooks of peers specified in their web of trust,
by adding in the entries provided through a third party, or (if some people organize
a series of published addressbooks using a first come first serve registration
system) people can choose to treat these addressbooks as name servers, emulating
traditional DNS.
</p>
<p>
I2P does not promote the use of DNS-like services though, as the damage doneby hijacking a site can be tremendous - and insecure destinations have no
value. DNSsec itself still falls back on registrars and certificate authorities,
while with I2P, requests sent to a destination cannot be intercepted or the reply
spoofed, as they are encrypted to the destination's public keys, and a destination
itself is just a pair of public keys and a certificate. DNS-style systems on the
other hand allow any of the name servers on the lookup path to mount simple denial
of service and spoofing attacks. Adding on a certificate authenticating the
responses as signed by some centralized certificate authority would address many of
the hostile nameserver issues but would leave open replay attacks as well as
hostile certificate authority attacks.
</p>
<p>
Voting style naming is dangerous as well, especially given the effectiveness of
Sybil attacks in anonymous systems - the attacker can simply create an arbitrarily
high number of peers and "vote" with each to take over a given name. Proof-of-work
methods can be used to make identity non-free, but as the network grows the load
required to contact everyone to conduct online voting is implausible, or if the
full network is not queried, different sets of answers may be reachable.
</p>
<p>
As with the Internet however, I2P is keeping the design and operation of a
naming system out of the (IP-like) communication layer. The bundled naming library
includes a simple service provider interface which alternate naming systems can
plug into, allowing end users to drive what sort of naming tradeoffs they prefer.
</p>
<h2 id="app.syndie">Syndie</h2>
<p>
Syndie is a safe, anonymous blogging / content publication / content aggregation system.
It lets you create information, share it with others, and read posts from those you're
interested in, all while taking into consideration your needs for security and anonymity.
Rather than building its own content distribution network, Syndie is designed to run on
top of existing networks, syndicating content through eepsites, Tor hidden services,
Freenet freesites, normal websites, usenet newgroups, email lists, RSS feeds, etc. Data
published with Syndie is done so as to offer pseudonymous authentication to anyone
reading or archiving it.
</p>
<h2 id="app.i2ptunnel">I2PTunnel</h2>
<p><i>Developed by: mihi</i></p>
<p>
I2PTunnel is probably I2P's most popular and versatile client application, allowing
generic proxying both into and out of the I2P network. I2PTunnel can be viewed as
four separate proxying applications - a "client" which receives inbound TCP connections
and forwards them to a given I2P destination, an "httpclient" (aka "eepproxy") which
acts like an HTTP proxy and forwards the requests to the appropriate I2P destination
(after querying the naming service if necessary), a "server" which receives inbound I2P
streaming connections on a destination and forwards them to a given TCP host+port,
and an "httpserver" which extends the "server" by parsing the HTTP request and
responses to allow safer operation. There is an additional "socksclient" application,
but its use is not encouraged for reasons previously mentioned.
</p>
<p>
I2P itself is not an outproxy network - the anonymity and security concerns inherent
in a mix net which forwards data into and out of the mix have kept I2P's design focused
on providing an anonymous network which capable of meeting the user's needs without
requiring external resources. However, the I2PTunnel "httpclient" application offers
a hook for outproxying - if the hostname requested doesn't end in ".i2p", it picks a
random destination from a user-provided set of outproxies and forwards the request to
them. These destinations are simply I2PTunnel "server" instances run by volunteers
who have explicitly chosen to run outproxies - no one is an outproxy by default, and
running an outproxy doesn't automatically tell other people to proxy through you.
While outproxies do have inherent weaknesses, they offer a simple proof of concept for
using I2P and provide some functionality under a threat model which may be sufficient
for some users.</p>
<p>
I2PTunnel enables most of the applications in use. An "httpserver" pointing at a
webserver lets anyone run their own anonymous website (or "eepsite") - a webserver
is bundled with I2P for this purpose, but any webserver can be used. Anyone may
run a "client" pointing at one of the anonymously hosted IRC servers, each of which
are running a "server" pointing at their local IRCd and communicating between IRCds
over their own "client" tunnels. End users also have "client" tunnels pointing at
<a href="#app.i2pmail">I2Pmail's</a> POP3 and SMTP destinations (which in turn are
simply "server" instances pointing at POP3 and SMTP servers), as well as "client"
tunnels pointing at I2P's CVS server, allowing anonymous development. At times people have
even run "client" proxies to access the "server" instances pointing at an NNTP server.
</p>
<h2 id="app.i2pbt">i2p-bt</h2>
<p><i>Developed by: duck, et al</i></p>
<p>
i2p-bt is a port of the mainline python BitTorrent client to run both the tracker and
peer communication over I2P. Tracker requests are forwarded through the eepproxy to
eepsites specified in the torrent file while tracker responses refer to peers by their
destination explicitly, allowing i2p-bt to open up a
<a href="#app.streaming">streaming lib</a> connection to query them for blocks.
</p>
<p>
In addition to i2p-bt, a port of bytemonsoon has been made to I2P, making a few
modifications as necessary to strip any anonymity-compromising information from the
application and to take into consideration the fact that IPs cannot be used for
identifying peers.
</p>
<h2 id="app.azneti2p">Azureus/azneti2p</h2>
<p><i>Developed by: parg, et al</i></p>
<p>
The developers of the <a href="http://azureus.sf.net/">Azureus</a> BitTorrent client
have created an "azneti2p" plugin, allowing Azureus users to participate in anonymous
swarms over I2P, or simply to access anonymously hosted trackers while contacting
each peer directly. In addition, Azureus' built in tracker lets people run their
own anonymous trackers without running bytemonsoon (which has substantial prerequisites)
or i2p-bt's tracker. The plugin is currently (July 2005) fully functional, but is in early
beta and has a fairly complicated configuration process, though it is hopefully going
to be streamlined further.
</p>
<h2 id="app.i2phex">I2Phex</h2>
<p><i>Developed by: sirup</i></p>
<p>
I2Phex is a fairly direct port of the Phex gnutella filesharing client to run
entirely on top of I2P. While it has disabled some of Phex's functionality,
such as integration with gnutella webcaches, the basic file sharing and chatting
system is fully functional.
</p>
<h2 id="app.i2pmail">I2Pmail/susimail</h2>
<p><i>Developed by: postman, susi23, mastiejaner</i></p>
<p>
I2Pmail is more a service than an application - postman offers both internal and
external email with POP3 and SMTP service through I2PTunnel instances accessing a
series of components developed with mastiejaner, allowing people to use their
preferred mail clients to send and receive mail pseudonymously. However, as most
mail clients expose substantial identifying information, I2P bundles susi23's
web based susimail client which has been built specifically with I2P's anonymity
needs in mind. The I2Pmail/mail.i2p service offers transparent virus filtering as
well as denial of service prevention with hashcash augmented quotas.
In addition, each user has control of their batching strategy prior to deliverythrough the mail.i2p outproxies, which are separate from the mail.i2p SMTP and
POP3 servers - both the outproxies and inproxies communicate with the mail.i2p
SMTP and POP3 servers through I2P itself, so compromising those non-anonymous
locations does not give access to the mail accounts or activity patterns of the
user. At the moment the developers work on a decentralized mailsystem, called
"v2mail". More information can be found on the eepsite
<a href="http://hq.postman.i2p/">hq.postman.i2p</a>.
</p>
</body>
</html>