Guide to Hacking Tor (As of 8 October 2003, this was all accurate. If you're reading this in the distant future, stuff may have changed.) 0. Intro and required reading Onion Routing is still very much in development stages. This document aims to get you started in the right direction if you want to understand the code, add features, fix bugs, etc. Read the README file first, so you can get familiar with the basics of installing and running an onion router. Then, skim some of the introductory materials in tor-design.pdf, tor-spec.txt, and the Tor FAQ to learn more about how the Tor protocol is supposed to work. This document will assume you know about Cells, Circuits, Streams, Connections, Onion Routers, and Onion Proxies. 1. Code organization 1.1. The modules The code is divided into two directories: ./src/common and ./src/or. The "common" directory contains general purpose utility functions not specific to onion routing. The "or" directory implements all onion-routing and onion-proxy specific functionality. Files in ./src/common: aes.[ch] -- Implements the AES cipher (with 128-bit keys and blocks), and a counter-mode stream cipher on top of AES. This code is taken from the main Rijndael distribution. (We include this because many people are running older versions of OpenSSL without AES support.) crypto.[ch] -- Wrapper functions to present a consistent interface to public-key and symmetric cryptography operations from OpenSSL. fakepoll.[ch] -- Used on systems that don't have a poll() system call; reimplements() poll using the select() system call. log.[ch] -- Tor's logging subsystem. test.h -- Macros used by unit tests. torint.h -- Provides missing [u]int*_t types for environments that don't have stdint.h. tortls.[ch] -- Wrapper functions to present a consistent interface to TLS, SSL, and X.509 functions from OpenSSL. util.[ch] -- Miscellaneous portability and convenience functions. Files in ./src/or: [General-purpose modules] or.h -- Common header file: include everything, define everything. buffers.c -- Implements a generic buffer interface. Buffers are fairly opaque string holders that can read to or flush from: memory, file descriptors, or TLS connections. Also implements parsing functions to read HTTP and SOCKS commands from buffers. tree.h -- A splay tree implementation by Niels Provos. Used by dns.c for dns caching at exits, and by connection_edge.c for dns caching at clients. config.c -- Code to parse and validate the configuration file. [Background processing modules] cpuworker.c -- Implements a farm of 'CPU worker' processes to perform CPU-intensive tasks in the background, so as not interrupt the onion router. (OR only) dns.c -- Implements a farm of 'DNS worker' processes to perform DNS lookups for onion routers and cache the results. [This needs to be done in the background because of the lack of a good, ubiquitous asynchronous DNS implementation.] (OR only) [Directory-related functionality.] directory.c -- Code to send and fetch directories and router descriptors via HTTP. Directories use dirserv.c to generate the results; clients use routers.c to parse them. dirserv.c -- Code to manage directory contents and generate directories. [Directory server only] routers.c -- Code to parse directories and router descriptors; and to generate a router descriptor corresponding to this OR's capabilities. Also presents some high-level interfaces for managing an OR or OP's view of the directory. [Circuit-related modules.] circuit.c -- Code to create circuits, manage circuits, and route relay cells along circuits. onion.c -- Code to generate and respond to "onion skins". [Core protocol implementation.] connection.c -- Code used in common by all connection types. See 1.2. below for more general information about connections. connection_edge.c -- Code used only by edge connections. command.c -- Code to handle specific cell types. connection_or.c -- Code to implement cell-speaking connections. [Toplevel modules.] main.c -- Toplevel module. Initializes keys, handles signals, multiplexes between connections, implements main loop, and drives scheduled events. tor_main.c -- Stub module containing a main() function. Allows unit test binary to link against main.c [Unit tests] test.c -- Contains unit tests for many pieces of the lower level Tor modules. 1.2. All about connections All sockets in Tor are handled as different types of nonblocking 'connections'. (What the Tor spec calls a "Connection", the code refers to as a "Cell-speaking" or "OR" connection.) Connections are implemented by the connection_t struct, defined in or.h. Not every kind of connection uses all the fields in connection_t; see the comments in or.h and the assertions in assert_connection_ok() for more information. Every connection has a type and a state. Connections never change their type, but can go through many state changes in their lifetime. The connection types break down as follows: [Cell-speaking connections] CONN_TYPE_OR -- A bidirectional TLS connection transmitting a sequence of cells. May be from an OR to an OR, or from an OP to an OR. [Edge connections] CONN_TYPE_EXIT -- A TCP connection from an onion router to a Stream's destination. [OR only] CONN_TYPE_AP -- A SOCKS proxy connection from the end user application to the onion proxy. [OP only] [Listeners] CONN_TYPE_OR_LISTENER [OR only] CONN_TYPE_AP_LISTENER [OP only] CONN_TYPE_DIR_LISTENER [Directory server only] -- Bound network sockets, waiting for incoming connections. [Internal] CONN_TYPE_DNSWORKER -- Connection from the main process to a DNS worker process. [OR only] CONN_TYPE_CPUWORKER -- Connection from the main process to a CPU worker process. [OR only] Connection states are documented in or.h. Every connection has two associated input and output buffers. Listeners don't use them. For non-listener connections, incoming data is appended to conn->inbuf, and outgoing data is taken from the front of conn->outbuf. Connections differ primarily in the functions called to fill and drain these buffers. 1.3. All about circuits. A circuit_t structure fills two roles. First, a circuit_t links two connections together: either an edge connection and an OR connection, or two OR connections. (When joined to an OR connection, a circuit_t affects only cells sent to a particular circID on that connection. When joined to an edge connection, a circuit_t affects all data.) Second, a circuit_t holds the cipher keys and state for sending data along a given circuit. At the OP, it has a sequence of ciphers, each of which is shared with a single OR along the circuit. Separate ciphers are used for data going "forward" (away from the OP) and "backward" (towards the OP). At the OR, a circuit has only two stream ciphers: one for data going forward, and one for data going backward. 1.4. Asynchronous IO and the main loop. Tor uses the poll(2) system call (or it wraps select(2) to act like poll, if poll is not available) to handle nonblocking (asynchronous) IO. If you're not familiar with nonblocking IO, check out the links at the end of this document. All asynchronous logic is handled in main.c. The functions 'connection_add', 'connection_set_poll_socket', and 'connection_remove' manage an array of connection_t*, and keep in synch with the array of struct pollfd required by poll(2). (This array of connection_t* is accessible via get_connection_array, but users should generally call one of the 'connection_get_by_*' functions in connection.c to look up individual connections.) To trap read and write events, connections call the functions 'connection_{is|stop|start}_{reading|writing}'. If you want to completely reset the events you're watching for, use 'connection_watch_events'. Every time poll() finishes, main.c calls conn_read and conn_write on every connection. These functions dispatch events that have something to read to connection_handle_read, and events that have something to write to connection_handle_write, respectively. When connections need to be closed, they can respond in two ways. Most simply, they can make connection_handle_* return an error (-1), which will make conn_{read|write} close them. But if it's not convenient to return -1 (for example, processing one connection causes you to realize that a second one should close), then you can also mark a connection to close by setting conn->marked_for_close. Marked connections will be closed at the end of the current iteration of the main loop. The main loop handles several other operations: First, it checks whether any signals have been received that require a response (HUP, KILL, USR1, CHLD). Second, it calls prepare_for_poll to handle recurring tasks and compute the necessary poll timeout. These recurring tasks include periodically fetching the directory, timing out unused circuits, incrementing flow control windows and re-enabling connections that were blocking for more bandwidth, and maintaining statistics. A word about TLS: Using TLS on OR connections complicates matters in two ways. First, a TLS stream has its own read buffer independent of the connection's read buffer. (TLS needs to read an entire frame from the network before it can decrypt any data. Thus, trying to read 1 byte from TLS can require that several KB be read from the network and decrypted. The extra data is stored in TLS's decrypt buffer.) Because the data hasn't been read by tor (it's still inside the TLS), this means that sometimes a connection "has stuff to read" even when poll() didn't return POLLIN. The tor_tls_get_pending_bytes function is used in main.c to detect TLS objects with non-empty internal buffers. Second, the TLS stream's events do not correspond directly to network events: sometimes, before a TLS stream can read, the network must be ready to write -- or vice versa. 1.5. How data flows (An illustration.) Suppose an OR receives 256 bytes along an OR connection. These 256 bytes turn out to be a data relay cell, which gets decrypted and delivered to an edge connection. Here we give a possible call sequence for the delivery of this data. (This may be outdated quickly.) do_main_loop -- Calls poll(2), receives a POLLIN event on a struct pollfd, then calls: conn_read -- Looks up the corresponding connection_t, and calls: connection_handle_read -- Calls: connection_read_to_buf -- Notices that it has an OR connection so: read_to_buf_tls -- Pulls data from the TLS stream onto conn->inbuf. connection_process_inbuf -- Notices that it has an OR connection so: connection_or_process_inbuf -- Checks whether conn is open, and calls: connection_process_cell_from_inbuf -- Notices it has enough data for a cell, then calls: connection_fetch_from_buf -- Pulls the cell from the buffer. cell_unpack -- Decodes the raw cell into a cell_t command_process_cell -- Notices it is a relay cell, so calls: command_process_relay_cell -- Looks up the circuit for the cell, makes sure the circuit is live, then passes the cell to: circuit_deliver_relay_cell -- Passes the cell to each of: relay_crypt -- Strips a layer of encryption from the cell and notices that the cell is for local delivery. connection_edge_process_relay_cell -- extracts the cell's relay command, and makes sure the edge connection is open. Since it has a DATA cell and an open connection, calls: circuit_consider_sending_sendme -- check if the total number of cells received by all streams on this circuit is enough that we should send back an acknowledgement (requesting that more cells be sent to any stream). connection_write_to_buf -- To place the data on the outgoing buffer of the correct edge connection, by calling: connection_start_writing -- To tell the main poll loop about the pending data. write_to_buf -- To actually place the outgoing data on the edge connection. connection_consider_sending_sendme -- if the outbuf waiting to flush to the exit connection is not too full, check if the total number of cells received on this stream is enough that we should send back an acknowledgement (requesting that more cells be sent to this stream). In a subsequent iteration, main notices that the edge connection is ready for writing: do_main_loop -- Calls poll(2), receives a POLLOUT event on a struct pollfd, then calls: conn_write -- Looks up the corresponding connection_t, and calls: connection_handle_write -- This isn't a TLS connection, so calls: flush_buf -- Delivers data from the edge connection's outbuf to the network. connection_wants_to_flush -- Reports that all data has been flushed. connection_finished_flushing -- Notices the connection is an exit, and calls: connection_edge_finished_flushing -- The connection is open, so it calls: connection_stop_writing -- Tells the main poll loop that this connection has no more data to write. connection_consider_sending_sendme -- now that the outbuf is empty, check again if the total number of cells received on this stream is enough that we should send back an acknowledgement (requesting that more cells be sent to this stream). 1.6. Routers, descriptors, and directories All Tor processes need to keep track of a list of onion routers, for several reasons: - OPs need to establish connections and circuits to ORs. - ORs need to establish connections to other ORs. - OPs and ORs need to fetch directories from a directory server. - ORs need to upload their descriptors to directory servers. - Directory servers need to know which ORs are allowed onto the network, what the descriptors are for those ORs, and which of those ORs are currently live. Thus, every Tor process keeps track of a list of all the ORs it knows in a static variable 'directory' in the routers.c module. This variable contains a routerinfo_t object for each known OR. On startup, the directory is initialized to a list of known directory servers (via router_get_list_from_file()). Later, the directory is updated via router_get_dir_from_string(). (OPs and ORs retrieve fresh directories from directory servers; directory servers generate their own.) Every OR must periodically regenerate a router descriptor for itself. The descriptor and the corresponding routerinfo_t are stored in the 'desc_routerinfo' and 'descriptor' static variables in routers.c. Additionally, a directory server keeps track of a list of the router descriptors it knows in a separate list in dirserv.c. It uses this list, checking which OR connections are open, to build directories. 1.7. Data model [XXX] 1.8. Flow control [XXX] 2. Coding conventions 2.1. Details Use tor_malloc, tor_strdup, and tor_gettimeofday instead of their generic equivalents. (They always succeed or exit.) Use INLINE instead of 'inline', so that we work properly on windows. 2.2. Calling and naming conventions Whenever possible, functions should return -1 on error and and 0 on success. For multi-word identifiers, use lowercase words combined with underscores. (e.g., "multi_word_identifier"). Use ALL_CAPS for macros and constants. Typenames should end with "_t". Function names should be prefixed with a module name or object name. (In general, code to manipulate an object should be a module with the same name as the object, so it's hard to tell which convention is used.) Functions that do things should have imperative-verb names (e.g. buffer_clear, buffer_resize); functions that return booleans should have predicate names (e.g. buffer_is_empty, buffer_needs_resizing). 2.3. What To Optimize Don't optimize anything if it's not in the critical path. Right now, the critical path seems to be AES, logging, and the network itself. Feel free to do your own profiling to determine otherwise. 2.4. Log conventions Log convention: use only these four log severities. ERR is if something fatal just happened. WARN if something bad happened, but we're still running. The bad thing is either a bug in the code, an attack or buggy protocol/implementation of the remote peer, etc. The operator should examine the bad thing and try to correct it. (No error or warning messages should be expected during normal OR or OP operation. I expect most people to run on -l warn eventually. If a library function is currently called such that failure always means ERR, then the library function should log WARN and let the caller log ERR.) INFO means something happened (maybe bad, maybe ok), but there's nothing you need to (or can) do about it. DEBUG is for everything louder than INFO. [XXX Proposed convention: every messages of severity INFO or higher should either (A) be intelligible to end-users who don't know the Tor source; or (B) somehow inform the end-users that they aren't expected to understand the message (perhaps with a string like "internal error"). Option (A) is to be preferred to option (B). -NM] 3. References About Tor See http://freehaven.net/tor/ http://freehaven.net/tor/cvs/doc/tor-spec.txt http://freehaven.net/tor/cvs/doc/tor-design.tex http://freehaven.net/tor/cvs/doc/FAQ About anonymity See http://freehaven.net/anonbib/ About nonblocking IO [XXX insert references] # ====================================================================== # Old HACKING document; merge into the above, move into tor-design.tex, # or delete. # ====================================================================== The pieces. Routers. Onion routers, as far as the 'tor' program is concerned, are a bunch of data items that are loaded into the router_array when the program starts. Periodically it downloads a new set of routers from a directory server, and updates the router_array. When a new OR connection is started (see below), the relevant information is copied from the router struct to the connection struct. Connections. A connection is a long-standing tcp socket between nodes. A connection is named based on what it's connected to -- an "OR connection" has an onion router on the other end, an "OP connection" has an onion proxy on the other end, an "exit connection" has a website or other server on the other end, and an "AP connection" has an application proxy (and thus a user) on the other end. Circuits. A circuit is a path over the onion routing network. Applications can connect to one end of the circuit, and can create exit connections at the other end of the circuit. AP and exit connections have only one circuit associated with them (and thus these connection types are closed when the circuit is closed), whereas OP and OR connections multiplex many circuits at once, and stay standing even when there are no circuits running over them. Streams. Streams are specific conversations between an AP and an exit. Streams are multiplexed over circuits. Cells. Some connections, specifically OR and OP connections, speak "cells". This means that data over that connection is bundled into 256 byte packets (8 bytes of header and 248 bytes of payload). Each cell has a type, or "command", which indicates what it's for. Robustness features. [XXX no longer up to date] Bandwidth throttling. Each cell-speaking connection has a maximum bandwidth it can use, as specified in the routers.or file. Bandwidth throttling can occur on both the sender side and the receiving side. If the LinkPadding option is on, the sending side sends cells at regularly spaced intervals (e.g., a connection with a bandwidth of 25600B/s would queue a cell every 10ms). The receiving side protects against misbehaving servers that send cells more frequently, by using a simple token bucket: Each connection has a token bucket with a specified capacity. Tokens are added to the bucket each second (when the bucket is full, new tokens are discarded.) Each token represents permission to receive one byte from the network --- to receive a byte, the connection must remove a token from the bucket. Thus if the bucket is empty, that connection must wait until more tokens arrive. The number of tokens we add enforces a longterm average rate of incoming bytes, yet we still permit short-term bursts above the allowed bandwidth. Currently bucket sizes are set to ten seconds worth of traffic. The bandwidth throttling uses TCP to push back when we stop reading. We extend it with token buckets to allow more flexibility for traffic bursts. Data congestion control. Even with the above bandwidth throttling, we still need to worry about congestion, either accidental or intentional. If a lot of people make circuits into same node, and they all come out through the same connection, then that connection may become saturated (be unable to send out data cells as quickly as it wants to). An adversary can make a 'put' request through the onion routing network to a webserver he owns, and then refuse to read any of the bytes at the webserver end of the circuit. These bottlenecks can propagate back through the entire network, mucking up everything. (See the tor-spec.txt document for details of how congestion control works.) In practice, all the nodes in the circuit maintain a receive window close to maximum except the exit node, which stays around 0, periodically receiving a sendme and reading more data cells from the webserver. In this way we can use pretty much all of the available bandwidth for data, but gracefully back off when faced with multiple circuits (a new sendme arrives only after some cells have traversed the entire network), stalled network connections, or attacks. We don't need to reimplement full tcp windows, with sequence numbers, the ability to drop cells when we're full etc, because the tcp streams already guarantee in-order delivery of each cell. Rather than trying to build some sort of tcp-on-tcp scheme, we implement this minimal data congestion control; so far it's enough. Router twins. In many cases when we ask for a router with a given address and port, we really mean a router who knows a given key. Router twins are two or more routers that share the same private key. We thus give routers extra flexibility in choosing the next hop in the circuit: if some of the twins are down or slow, it can choose the more available ones. Currently the code tries for the primary router first, and if it's down, chooses the first available twin.