Detailed specification of protocol in version 00000800 ====================================================== Note: work in progress!! ====================================================== 1. DNS protocol ====================================================== Quick alphabetical index / register: 0-9 Data packet A-F Data packet I IP address L Login N Downstream fragsize (NS.topdomain A-type reply) O Options P Ping R Downstream fragsize probe S Switch upstream codec V Version W (WWW.topdomain A-type reply) Y Downstream codec check Z Upstream codec check CMC = Cache Miss Counter, increased every time it is used Version: Client sends: First byte v or V Rest encoded with base32: 4 bytes big endian protocol version CMC Server replies: 4 chars: VACK (version ok), followed by login challenge VNAK (version differs), followed by server protocol version VFUL (server has no free slots), followed by max users 4 byte value: means login challenge/server protocol version/max users 1 byte userid of the new user, or any byte if not VACK Login: Client sends: First byte l or L Rest encoded with base32: 1 byte userid 16 bytes MD5 hash of: (first 32 bytes of password) xor (8 repetitions of login challenge) CMC Server replies: LNAK means not accepted x.x.x.x-y.y.y.y-mtu-netmask means accepted (server ip, client ip, mtu, netmask bits) IP Request: (for where to try raw login or set data pipe mode) Client sends: First byte i or I 5 bits coded as Base32 char, meaning userid CMC as 3 Base32 chars Server replies BADIP if bad userid First byte I Then comes external IP address of iodined server as 4 bytes (IPv4) or 16 bytes (IPv6) Upstream codec check / bounce: Client sends: First byte z or Z Lots of data that should not be decoded Server replies: The requested domain copied raw, in the lowest-grade downstream codec available for the request type. Downstream codec check: Client sends: First byte y or Y 1 char, meaning downstream codec to use 5 bits coded as Base32 char, meaning check variant CMC as 3 Base32 chars Possibly extra data, depending on check variant Server sends: Data encoded with requested downstream codec; data content depending on check variant number. BADCODEC if requested downstream codec not available. BADLEN if check variant is not available, or problem with extra data. Downstream codec chars are same as in 'O' Option request, below. Check variants: 1: Send encoded DOWNCODECCHECK1 string as defined in encoding.h (Other variants reserved; possibly variant that sends a decoded-encoded copy of Base32-encoded extra data in the request) Switch codec: Client sends: First byte s or S 5 bits coded as Base32 char, meaning userid 5 bits coded as Base32 char, representing number of raw bits per encoded byte: 5: Base32 (a-z0-5) 6: Base64 (a-zA-Z0-9+-) 26: Base64u (a-zA-Z0-9_-) 7: Base128 (a-zA-Z0-9\274-\375) CMC as 3 Base32 chars Server sends: Name of codec if accepted. After this all upstream data packets must be encoded with the new codec. BADCODEC if not accepted. Client must then revert to previous codec BADLEN if length of query is too short Options: Client sends: First byte o or O 5 bits coded as Base32 char, meaning userid number of options (n) as decimal digit n chars, each a valid option (to be processed in order) CMC as 3 Base32 chars Server sends: Option chars in the same order as request, indicating that options were accepted by the server. BADCODEC if not accepted. Previous situation remains. BADLEN if number of options doesn't match length of query. All options affect only the requesting client. Option chars: t or T: Downstream encoding Base32, for TXT/CNAME/A/MX (default) s or S: Downstream encoding Base64, for TXT/CNAME/A/MX u or U: Downstream encoding Base64u, for TXT/CNAME/A/MX v or V: Downstream encoding Base128, for TXT/CNAME/A/MX r or R: Downstream encoding Raw, for PRIVATE/TXT/NULL (assumed for PRIVATE and NULL) If codec unsupported for request type, server will use Base32; note that server will answer any mix of request types that a client sends. Server may disregard this option; client must always use the downstream encoding type indicated in every downstream DNS packet. c or C: Downstream compression enabled (compressed before encoding) d or D: Downstream compression disabled l or L: Lazy mode, server will keep a number of requests waiting until data becomes available to send downstream or the requests time out. The timeout value for requests is controlled by the client. Applies only to data transfer; handshake is always answered immediately. i or I: Immediate (non-lazy) mode, server will answer all requests (nearly) immediately. Probe downstream fragment size: Client sends: First byte r or R Second byte userid char 2 bytes big-endian fragsize encoded as 4 bytes base32 Then follows a long random query which contents does not matter. Server sends: Requested number of bytes as a response. The first two bytes contain the requested length. The third byte is 107 (0x6B). The fourth byte is a random value, and each following byte is incremented with 107. This is checked by the client to determine corruption. BADFRAG if requested length not accepted. Set downstream fragment size: Client sends: First byte n or N Rest encoded with base32: 1 byte userid 2 bytes new downstream fragment size (big-endian) CMC Server sends: 2 bytes new downstream fragment size. After this all downstream payloads will be max (fragsize + 2) bytes long. BADFRAG if not accepted. Upstream data header: 76543 21076 54321076 54321076 5432 +!----+!----+!----!--+--!----!+----+ |0UUUU|UDCMC| Seq ID | Dn ACK |ACFL| +-----+-----+--------+--------+----+ Downstream data header: |=> only if ping (P) flag set | 0 1 2 3 4 5 6 +--------+--------+76543210+--------+--------+--------+--------+ | Seq ID | Up ACK |00IPACFL|Dn Wsize|Up Wsize|DnWstart|UpWstart| +--------+--------+--------+--------+--------+--------+--------+ UUUU = Userid L = Last fragment flag A = ACK flag F = First fragment flag C = Compression enabled for downstream packet P = ping flag: extra header present I = responded to immediately (for RTT calculation) - downstream only UDCMC = Upstream Data CMC char (base36 [a-z0-9]) Up/Dn Wsize/Wstart = upstream/downstream window size/window start Seq ID Upstream data packet starts with 1 byte ASCII hex coded user byte; then 1 char data-CMC; then 4 bytes Base32 encoded header; then comes the payload data, encoded with the chosen upstream codec. Downstream data starts with 3 byte header, followed by data, which may be compressed. If Ping flag is set, another 4 bytes are appended to the header, containing upstream and downstream window sizes and window start sequence IDs. The response does not need to contain data. If the server has no data to send, the response will always include the ping header and the ping flag will be set. In NULL and PRIVATE responses, downstream data is always raw. In all other response types, downstream data is encoded (see Options above). Encoding type is indicated by 1 prefix char (before the data header): TXT: End result is always DNS-chopped (series of len-prefixed strings <=255 bytes) t or T: Base32 encoded before chop, decoded after un-chop s or S: Base64 encoded before chop, decoded after un-chop u or U: Base64u encoded before chop, decoded after un-chop v or V: Base128 encoded before chop, decoded after un-chop r or R: Raw no encoding, only DNS-chop SRV/MX/CNAME/A: h or H: Hostname encoded with Base32 i or I: Hostname encoded with Base64 j or J: Hostname encoded with Base64u k or K: Hostname encoded with Base128 SRV and MX may reply with multiple hostnames, each encoded separately. Each has a 10-multiple priority, and encoding/decoding is done in strictly increasing priority sequence 10, 20, 30, etc. without gaps. Note that some DNS relays will shuffle the answer records in the response. Ping: Client sends: First byte p or P Second byte CMC Rest encoded with Base32: 0 1 2...8 9-10 +--------+--------+---+76543210+---+ |0000UUUU|Dn SeqID|...|000WTANR|CMC| +--------+--------+---+--------+---+ 4 bits Userid 1 byte Downstream seq ID ACK 1 byte window size (upstream) 1 byte window size (downstream) 1 byte window start (upstream) 1 byte window start (downstream) 2 bytes big-endian server timeout in ms 2 bytes big-endian downstream fragment ACK timeout in ms 1 byte flags: W = update window frag timeout T = update server timeout A = is ACKing downstream frag N = is NACKing downstream frag (unused) R = response must contain ping header (data optional) 2 bytes CMC The server responses to Ping and Data packets are compatible, and are described above (refer to downstream data header). If R (respond) bit is set, the server responds immediately with a ping header. The server must also adjust its window sizes to those provided by the ping. If the T but is set, the server sets the user's DNS timeout to the value spec- ified by the packet. In lazy mode, unless the R flag is set, the server will hold the ping until it times out or more data becomes available to send. "Lazy-mode" operation ===================== Client-server DNS traffic sequence has been reordered to provide increased (interactive) performance and greatly reduced latency. Idea taken from Lucas Nussbaum's slides (24th IFIP International Security Conference, 2009) at http://www.loria.fr/~lnussbau/tuns.html. Current implementation is original to iodine, no code or documentation from any other project was consulted during development. Server: In lazy mode, except where otherwise specified, responses are sent using the oldest pending query held in the server's buffer (QMEM). The server responds to a stored pending query when the query times out, an upstream ACK is pending (for that user), or the server has an excess of pending queries (more than the user's downstream window size). Upstream data fragments are ACK'd immediately to keep data flowing. Upstream pings are answered immediately only when the Respond flag is set (see ping header), in which case the response is to the same DNS query as the ping. Immediate responses (<10ms old) to either ping or data requests are marked and used to calculate the round-trip-time for the connection. Client: The client keeps track of all queries it sends, and maintains a minimum of pending queries to fill the server buffer. Downstream data is always ACK'd immediately with a new request (either a ping or data if available). The client sends excess requests (ie. already has enough pending queries) for ACKs or for new data. ====================================================== 2. Raw UDP protocol ====================================================== This protocol does not implement data windowing and does not guarantee data delivery, however it is likely faster due to the fact that data is not split into fragments for sending. Full packets are compressed and sent when they arrive on the tun device, and are processed immediately on the other side. All Raw UDP protcol messages start with a 3 byte header: 0x10d19e This is not the start of a valid DNS message so it is easy to identify. The fourth byte contains the command (C) and the user id (U). 7654 3210 +----+----+ |CCCC|UUUU| +----+----+ Login message (command = 1): The header is followed by a MD5 hash with the same password as in the DNS login. The client starts the raw mode by sending this message, and uses the login challenge +1, and the server responds using the login challenge -1. After the login message has been exchanged, both the server and the client switch to raw udp mode for the rest of the connection. Data message (command = 2): After the header comes the payload data, which is always compressed. Ping message (command = 3): Sent from client to server and back to keep session open. Has no payload.