Erlang Run-Time System Application (ERTS)

User's Guide

Version 11.0

Chapters

13 Distribution Protocol

This description is far from complete. It will be updated if the protocol is updated. However, the protocols, both from Erlang nodes to the Erlang Port Mapper Daemon (EPMD) and between Erlang nodes are stable since many years.

The distribution protocol can be divided into four parts:

  • Low-level socket connection (1)

  • Handshake, interchange node name, and authenticate (2)

  • Authentication (done by net_kernel(3)) (3)

  • Connected (4)

A node fetches the port number of another node through the EPMD (at the other host) to initiate a connection request.

For each host, where a distributed Erlang node is running, also an EPMD is to be running. The EPMD can be started explicitly or automatically as a result of the Erlang node startup.

By default the EPMD listens on port 4369.

(3) and (4) above are performed at the same level but the net_kernel disconnects the other node if it communicates using an invalid cookie (after 1 second).

The integers in all multibyte fields are in big-endian order.

Warning

The Erlang Distribution protocol is not by itself secure and does not aim to be so. In order to get secure distribution the distributed nodes should be configured to use distribution over tls. See the Using SSL for Erlang Distribution User's Guide for details on how to setup a secure distributed node.

13.1  EPMD Protocol

The requests served by the EPMD are summarized in the following figure.

IMAGE MISSING

Figure 13.1:   Summary of EPMD Requests

Each request *_REQ is preceded by a 2 byte length field. Thus, the overall request format is as follows:

2 n
Length Request

Table 13.1:   Request Format

Register a Node in EPMD

When a distributed node is started it registers itself in the EPMD. The message ALIVE2_REQ described below is sent from the node to the EPMD. The response from the EPMD is ALIVE2_X_RESP (or ALIVE2_RESP).

1 2 1 1 2 2 2 Nlen 2 Elen
120 PortNo NodeType Protocol HighestVersion LowestVersion Nlen NodeName Elen Extra

Table 13.2:   ALIVE2_REQ (120)

PortNo

The port number on which the node accept connection requests.

NodeType

77 = normal Erlang node, 72 = hidden node (C-node), ...

Protocol

0 = TCP/IPv4, ...

HighestVersion

The highest distribution protocol version this node can handle. The value in OTP 23 and later is 6. Older nodes only support version 5.

LowestVersion

The lowest distribution version that this node can handle. Should be 5 to support connections to nodes older than OTP 23.

Nlen

The length (in bytes) of field NodeName.

NodeName

The node name as an UTF-8 encoded string of Nlen bytes.

Elen

The length of field Extra.

Extra

Extra field of Elen bytes.

The connection created to the EPMD must be kept as long as the node is a distributed node. When the connection is closed, the node is automatically unregistered from the EPMD.

The response message is either ALIVE2_X_RESP or ALIVE2_RESP depending on distribution version. If both the node and EPMD support distribution version 6 then the response is ALIVE2_X_RESP otherwise it is the older ALIVE2_RESP:

1 1 4
118 Result Creation

Table 13.3:   ALIVE2_X_RESP (118) with 32 bit creation

1 1 2
121 Result Creation

Table 13.4:   ALIVE2_RESP (121) with 16-bit creation

Result = 0 -> ok, result > 0 -> error.

Unregister a Node from EPMD

A node unregisters itself from the EPMD by closing the TCP connection to EPMD established when the node was registered.

Get the Distribution Port of Another Node

When one node wants to connect to another node it starts with a PORT_PLEASE2_REQ request to the EPMD on the host where the node resides to get the distribution port that the node listens to.

1 N
122 NodeName

Table 13.5:   PORT_PLEASE2_REQ (122)

where N = Length - 1.

1 1
119 Result

Table 13.6:   PORT2_RESP (119) Response Indicating Error, Result > 0

or

1 1 2 1 1 2 2 2 Nlen 2 Elen
119 Result PortNo NodeType Protocol HighestVersion LowestVersion Nlen NodeName Elen >Extra

Table 13.7:   PORT2_RESP, Result = 0

If Result > 0, the packet only consists of [119, Result].

The EPMD closes the socket when it has sent the information.

Get All Registered Names from EPMD

This request is used through the Erlang function net_adm:names/1,2. A TCP connection is opened to the EPMD and this request is sent.

1
110

Table 13.8:   NAMES_REQ (110)

The response for a NAMES_REQ is as follows:

4  
EPMDPortNo NodeInfo*

Table 13.9:   NAMES_RESP

NodeInfo is a string written for each active node. When all NodeInfo has been written the connection is closed by the EPMD.

NodeInfo is, as expressed in Erlang:

io:format("name ~ts at port ~p~n", [NodeName, Port]).

Dump All Data from EPMD

This request is not really used, it is to be regarded as a debug feature.

1
100

Table 13.10:   DUMP_REQ

The response for a DUMP_REQ is as follows:

4  
EPMDPortNo NodeInfo*

Table 13.11:   DUMP_RESP

NodeInfo is a string written for each node kept in the EPMD. When all NodeInfo has been written the connection is closed by the EPMD.

NodeInfo is, as expressed in Erlang:

io:format("active name     ~ts at port ~p, fd = ~p~n",
          [NodeName, Port, Fd]).

or

io:format("old/unused name ~ts at port ~p, fd = ~p ~n",
          [NodeName, Port, Fd]).

Kill EPMD

This request kills the running EPMD. It is almost never used.

1
107

Table 13.12:   KILL_REQ

The response for a KILL_REQ is as follows:

2
OKString

Table 13.13:   KILL_RESP

where OKString is "OK".

STOP_REQ (Not Used)

1 n
115 NodeName

Table 13.14:   STOP_REQ

where n = Length - 1.

The current implementation of Erlang does not care if the connection to the EPMD is broken.

The response for a STOP_REQ is as follows:

7
OKString

Table 13.15:   STOP_RESP

where OKString is "STOPPED".

A negative response can look as follows:

7
NOKString

Table 13.16:   STOP_NOTOK_RESP

where NOKString is "NOEXIST".

13.2  Distribution Handshake

This section describes the distribution handshake protocol used between nodes to establishing a connection. The protocol was introduced in Erlang/OTP R6 and has remained unchanged until OTP 23. The changes made in OTP 23 were designed to be compatible with the older protocol version. That is an old node can still connect toward a new node and vice versa.

General

The TCP/IP distribution uses a handshake that expects a connection-based protocol, that is, the protocol does not include any authentication after the handshake procedure.

This is not entirely safe, as it is vulnerable against takeover attacks, but it is a tradeoff between fair safety and performance.

The cookies are never sent in cleartext and the handshake procedure expects the client (called A) to be the first one to prove that it can generate a sufficient digest. The digest is generated with the MD5 message digest algorithm and the challenges are expected to be random numbers.

Definitions

A challenge is a 32-bit integer in big-endian order. Below the function gen_challenge() returns a random 32-bit integer used as a challenge.

A digest is a (16 bytes) MD5 hash of the challenge (as text) concatenated with the cookie (as text). Below, the function gen_digest(Challenge, Cookie) generates a digest as described above.

An out_cookie is the cookie used in outgoing communication to a certain node, so that A's out_cookie for B is to correspond with B's in_cookie for A and conversely. A's out_cookie for B and A's in_cookie for B need not be the same. Below the function out_cookie(Node) returns the current node's out_cookie for Node.

An in_cookie is the cookie expected to be used by another node when communicating with us, so that A's in_cookie for B corresponds with B's out_cookie for A. Below the function in_cookie(Node) returns the current node's in_cookie for Node.

The cookies are text strings that can be viewed as passwords.

Every message in the handshake starts with a 16-bit big-endian integer, which contains the message length (not counting the two initial bytes). In Erlang this corresponds to option {packet, 2} in gen_tcp(3). Notice that after the handshake, the distribution switches to 4 byte packet headers.

The Handshake in Detail

Imagine two nodes, A that initiates the handshake and B that accepts the connection.

1) connect/accept

A connects to B through TCP/IP and B accepts the connection.

2) send_name/receive_name

A sends an initial identification to B, which receives the message. The message can have two different formats which looks as follows (the packet headers are removed):

1 2 4 Nlen
'n' Version=5 Flags Name

Table 13.17:   Old send_name ('n') for protocol version 5

1 8 4 2 Nlen
'N' Flags Creation Nlen Name

Table 13.18:   New send_name ('N') for protocol version 6

The old send_name format is sent from nodes only supporting version 5 or to nodes that might only support version 5. The Version is a 16-bit big endian integer and must always have the value 5, even if node A supports version 6. Flags are the capability flags of node A in 32-bit big endian. The flag bit DFLAG_HANDSHAKE_23 should be set if node A supports version 6. Name is the full node name of A, as a string of bytes (the packet length denotes how long it is).

The new send_name is only sent from nodes supporting version 6 to nodes known to support version 6. Flags are the capability flags of node A in 64-bit big endian. The flag bit DFLAG_HANDSHAKE_23 must always be set. Creation is the node incarnation identifier used by node A to create its pids, ports and references. Name is the full node name of A, as a string of bytes. Nlen is the byte length of the node name in 16-bit big endian. Any extra data after the node Name must be accepted and ignored.

3) recv_status/send_status

B sends a status message to A, which indicates if the connection is allowed.

1 Slen
's' Status

Table 13.19:   The format of the status message

's' is the message tag. Status is the status code as a string (not null terminated). The following status codes are defined:

ok

The handshake will continue.

ok_simultaneous

The handshake will continue, but A is informed that B has another ongoing connection attempt that will be shut down (simultaneous connect where A's name is greater than B's name, compared literally).

nok

The handshake will not continue, as B already has an ongoing handshake, which it itself has initiated (simultaneous connect where B's name is greater than A's).

not_allowed

The connection is disallowed for some (unspecified) security reason.

alive

A connection to the node is already active, which either means that node A is confused or that the TCP connection breakdown of a previous node with this name has not yet reached node B. See step 3B below.

named:

The handshake willl continue, but A requested a dynamic node name by setting flag DFLAG_NAME_ME. The dynamic node name of A is supplied at the end of the status message from B.

1 Slen=6 2 Nlen
's' Status='named:' Nlen Name

Table 13.20:   The format of the 'named:' status message

Name is the full dynamic node name of A, as a string of bytes. Nlen is the byte length of the node name in 16-bit big endian. Any extra data after the node Name must be accepted and ignored.

3B) send_status/recv_status

If status was alive, node A answers with another status message containing either true, which means that the connection is to continue (the old connection from this node is broken), or false, which means that the connection is to be closed (the connection attempt was a mistake.

4) recv_challenge/send_challenge

If the status was ok or ok_simultaneous, the handshake continues with B sending A another message, the challenge. The challenge contains the same type of information as the "name" message initially sent from A to B, plus a 32-bit challenge. The challenge message can have two different formats:

1 2 4 4 Nlen
'n' Version=5 Flags Challenge Name

Table 13.21:   The old challenge message format (version 5)

1 8 4 4 2 Nlen
'N' Flags Challenge Creation Nlen Name

Table 13.22:   The new challenge message format (version 6)

The old challenge message is sent from old B nodes (supporting only version 5) or if node A had not capability flag DFLAG_HANDSHAKE_23 set. The Version is a 16-bit big endian integer and must always have the value 5.

The new challenge message is sent from new B nodes if node A had capability flag DFLAG_HANDSHAKE_23 set. Any extra data after the node Name must be accepted and ignored.

Challenge is a 32-bit big-endian integer. The other fields are node B's flags, creation and full node name, similar to the send_name message.

4B) send_complement/recv_complement

The complement message, from A to B, is only sent if node A initially sent an old name message and received back a new challenge message from node B. It contains complementary information missing in the initial old name message from node A.

1 4 4
'c' FlagsHigh Creation

Table 13.23:   The complement message

FlagsHigh are the high capability flags (bit 33-64) of node A as a 32-bit big endian integer. Creation is the incarnation identifier of node A.

5) send_challenge_reply/recv_challenge_reply

Now A has generated a digest and its own challenge. Those are sent together in a package to B:

1 4 16
'r' Challenge Digest

Table 13.24:   The challenge_reply message

Challenge is A's challenge for B to handle. Digest is the MD5 digest that A constructed from the challenge B sent in the previous step.

6) recv_challenge_ack/send_challenge_ack

B checks that the digest received from A is correct and generates a digest from the challenge received from A. The digest is then sent to A. The message is as follows:

1 16
'a' Digest

Table 13.25:   The challenge_ack message

Digest is the digest calculated by B for A's challenge.

7) check

A checks the digest from B and the connection is up.

Semigraphic View

A (initiator)                                      B (acceptor)

TCP connect ------------------------------------>
                                                   TCP accept

send_name -------------------------------------->
                                                   recv_name

  <---------------------------------------------- send_status
recv_status
(if status was 'alive'
 send_status - - - - - - - - - - - - - - - - - ->
                                                   recv_status)

                          (ChB)                      ChB = gen_challenge()
  <---------------------------------------------- send_challenge
recv_challenge

(if old send_name and new recv_challenge
 send_complement - - - - - - - - - - - - - - - ->
                                                   recv_complement)

ChA = gen_challenge(),
OCA = out_cookie(B),
DiA = gen_digest(ChB, OCA)
                          (ChA, DiA)
send_challenge_reply --------------------------->
                                                   recv_challenge_reply
                                                   ICB = in_cookie(A),
                                                   check:
                                                   DiA == gen_digest (ChB, ICB)?
                                                   - if OK:
                                                    OCB = out_cookie(A),
                                                    DiB = gen_digest (ChA, OCB)
                          (DiB)
  <----------------------------------------------- send_challenge_ack
recv_challenge_ack                                  DONE
ICA = in_cookie(B),                                - else:
check:                                              CLOSE
DiB == gen_digest(ChA, ICA)?
- if OK:
 DONE
- else:
 CLOSE

Distribution Flags

The following capability flags are defined:

-define(DFLAG_PUBLISHED,16#1).

The node is to be published and part of the global namespace.

-define(DFLAG_ATOM_CACHE,16#2).

The node implements an atom cache (obsolete).

-define(DFLAG_EXTENDED_REFERENCES,16#4).

The node implements extended (3 × 32 bits) references. This is required today. If not present, the connection is refused.

-define(DFLAG_DIST_MONITOR,16#8).

The node implements distributed process monitoring.

-define(DFLAG_FUN_TAGS,16#10).

The node uses separate tag for funs (lambdas) in the distribution protocol.

-define(DFLAG_DIST_MONITOR_NAME,16#20).

The node implements distributed named process monitoring.

-define(DFLAG_HIDDEN_ATOM_CACHE,16#40).

The (hidden) node implements atom cache (obsolete).

-define(DFLAG_NEW_FUN_TAGS,16#80).

The node understands the NEW_FUN_EXT tag.

-define(DFLAG_EXTENDED_PIDS_PORTS,16#100).

The node can handle extended pids and ports. This is required today. If not present, the connection is refused.

-define(DFLAG_EXPORT_PTR_TAG,16#200).

The node understands the EXPORT_EXT tag.

-define(DFLAG_BIT_BINARIES,16#400).

The node understands the BIT_BINARY_EXT tag.

-define(DFLAG_NEW_FLOATS,16#800).

The node understands the NEW_FLOAT_EXT tag.

-define(DFLAG_UNICODE_IO,16#1000).
-define(DFLAG_DIST_HDR_ATOM_CACHE,16#2000).

The node implements atom cache in distribution header.

-define(DFLAG_SMALL_ATOM_TAGS, 16#4000).

The node understands the SMALL_ATOM_EXT tag.

-define(DFLAG_UTF8_ATOMS, 16#10000).

The node understands UTF-8 atoms encoded with ATOM_UTF8_EXT and SMALL ATOM_UTF8_EXT.

-define(DFLAG_MAP_TAG, 16#20000).

The node understands the map tag MAP_EXT.

-define(DFLAG_BIG_CREATION, 16#40000).

The node understands big node creation tags NEW_PID_EXT, NEW_PORT_EXT and NEWER_REFERENCE_EXT.

-define(DFLAG_SEND_SENDER, 16#80000).

Use the SEND_SENDER control message instead of the SEND control message and use the SEND_SENDER_TT control message instead of the SEND_TT control message.

-define(DFLAG_BIG_SEQTRACE_LABELS, 16#100000).

The node understands any term as the seqtrace label.

-define(DFLAG_EXIT_PAYLOAD, 16#400000).

Use the PAYLOAD_EXIT, PAYLOAD_EXIT_TT, PAYLOAD_EXIT2, PAYLOAD_EXIT2_TT and PAYLOAD_MONITOR_P_EXIT control messages instead of the non-PAYLOAD variants.

-define(DFLAG_FRAGMENTS, 16#800000).

Use fragmented distribution messages to send large messages.

-define(DFLAG_HANDSHAKE_23, 16#1000000).

The node supports the new connection setup handshake (version 6) introduced in OTP 23.

-define(DFLAG_SPAWN, (1 bsl 32)).

Set if the SPAWN_REQUEST, SPAWN_REQUEST_TT, SPAWN_REPLY, SPAWN_REPLY_TT control messages are supported.

-define(DFLAG_NAME_ME, (1 bsl 33)).

Dynamic node name. This is not a capability but rather used as a request from the connecting node to receive its node name from the accepting node as part of the handshake.

There is also function dist_util:strict_order_flags/0 returning all flags (bitwise or:ed together) corresponding to features that require strict ordering of data over distribution channels.

13.3  Protocol between Connected Nodes

Since ERTS 5.7.2 (OTP R13B) the runtime system passes a distribution flag in the handshake stage that enables the use of a distribution header on all messages passed. Messages passed between nodes have in this case the following format:

4 d n m
Length DistributionHeader ControlMessage Message

Table 13.26:   Format of Messages Passed between Nodes (as from ERTS 5.7.2 (OTP R13B))

Length

Equal to d + n + m.

DistributionHeader

Distribution header describing the atom cache and fragmented distribution messages.

ControlMessage

A tuple passed using the external format of Erlang.

Message

The message sent to another node using the '!' or the reason for a EXIT, EXIT2 or DOWN signal using the external term format.

Notice that the version number is omitted from the terms that follow a distribution header .

Nodes with an ERTS version earlier than 5.7.2 (OTP R13B) does not pass the distribution flag that enables the distribution header. Messages passed between nodes have in this case the following format:

4 1 n m
Length Type ControlMessage Message

Table 13.27:   Format of Messages Passed between Nodes (before ERTS 5.7.2 (OTP R13B))

Length

Equal to 1 + n + m.

Type

Equal to 112 (pass through).

ControlMessage

A tuple passed using the external format of Erlang.

Message

The message sent to another node using the '!' (in external format). Notice that Message is only passed in combination with a ControlMessage encoding a send ('!').

The ControlMessage is a tuple, where the first element indicates which distributed operation it encodes:

LINK

{1, FromPid, ToPid}

SEND

{2, Unused, ToPid}

Followed by Message.

Unused is kept for backward compatibility.

EXIT

{3, FromPid, ToPid, Reason}

This signal is sent when a link has been broken

UNLINK

{4, FromPid, ToPid}

NODE_LINK

{5}

REG_SEND

{6, FromPid, Unused, ToName}

Followed by Message.

Unused is kept for backward compatibility.

GROUP_LEADER

{7, FromPid, ToPid}

EXIT2

{8, FromPid, ToPid, Reason}

This signal is sent by a call to the erlang:exit/2 bif

SEND_TT

{12, Unused, ToPid, TraceToken}

Followed by Message.

Unused is kept for backward compatibility.

EXIT_TT

{13, FromPid, ToPid, TraceToken, Reason}

REG_SEND_TT

{16, FromPid, Unused, ToName, TraceToken}

Followed by Message.

Unused is kept for backward compatibility.

EXIT2_TT

{18, FromPid, ToPid, TraceToken, Reason}

MONITOR_P

{19, FromPid, ToProc, Ref}, where FromPid = monitoring process and ToProc = monitored process pid or name (atom)

DEMONITOR_P

{20, FromPid, ToProc, Ref}, where FromPid = monitoring process and ToProc = monitored process pid or name (atom)

We include FromPid just in case we want to trace this.

MONITOR_P_EXIT

{21, FromProc, ToPid, Ref, Reason}, where FromProc = monitored process pid or name (atom), ToPid = monitoring process, and Reason = exit reason for the monitored process

13.4  New Ctrlmessages for Erlang/OTP 21

SEND_SENDER

{22, FromPid, ToPid}

Followed by Message.

This control message replaces the SEND control message and will be sent when the distribution flag DFLAG_SEND_SENDER has been negotiated in the connection setup handshake.

Note

Messages encoded before the connection has been set up may still use the SEND control message. However, once a SEND_SENDER or SEND_SENDER_TT control message has been sent, no more SEND control messages will be sent in the same direction on the connection.

SEND_SENDER_TT

{23, FromPid, ToPid, TraceToken}

Followed by Message.

This control message replaces the SEND_TT control message and will be sent when the distribution flag DFLAG_SEND_SENDER has been negotiated in the connection setup handshake.

Note

Messages encoded before the connection has been set up may still use the SEND_TT control message. However, once a SEND_SENDER or SEND_SENDER_TT control message has been sent, no more SEND_TT control messages will be sent in the same direction on the connection.

13.5  New Ctrlmessages for Erlang/OTP 22

Note

Messages encoded before the connection has been set up may still use the non-PAYLOAD variant. However, once a PAYLOAD control message has been sent, no more non-PAYLOAD control messages will be sent in the same direction on the connection.

PAYLOAD_EXIT

{24, FromPid, ToPid}

Followed by Reason.

This control message replaces the EXIT control message and will be sent when the distribution flag DFLAG_EXIT_PAYLOAD has been negotiated in the connection setup handshake.

PAYLOAD_EXIT_TT

{25, FromPid, ToPid}

Followed by Reason.

This control message replaces the EXIT_TT control message and will be sent when the distribution flag DFLAG_EXIT_PAYLOAD has been negotiated in the connection setup handshake.

PAYLOAD_EXIT2

{26, FromPid, ToPid}

Followed by Reason.

This control message replaces the EXIT2 control message and will be sent when the distribution flag DFLAG_EXIT_PAYLOAD has been negotiated in the connection setup handshake.

PAYLOAD_EXIT2_TT

{27, FromPid, ToPid}

Followed by Reason.

This control message replaces the EXIT2_TT control message and will be sent when the distribution flag DFLAG_EXIT_PAYLOAD has been negotiated in the connection setup handshake.

PAYLOAD_MONITOR_P_EXIT

{28, FromPid, ToPid, Ref}

Followed by Reason.

This control message replaces the MONITOR_P_EXIT control message and will be sent when the distribution flag DFLAG_EXIT_PAYLOAD has been negotiated in the connection setup handshake.

13.6  New Ctrlmessages for Erlang/OTP 23

SPAWN_REQUEST

{29, ReqId, From, GroupLeader, {Module, Function, Arity}, OptList}

Followed by ArgList.

This signal is sent by the spawn_request() BIF.

ReqId :: reference()

Request identifier. Also used as monitor reference in case the monitor option has been passed.

From :: pid()

Process identifier of the process making the request. That is, the parent process to be.

GroupLeader :: pid()

Process identifier of the group leader of the newly created process.

{Module :: atom(), Function :: atom(), Arity :: integer() >= 0}

Entry point for the the new process.

OptList :: [term()]

A proper list of spawn options to use when spawning.

ArgList :: [term()]

A proper list of arguments to use in the call to the entry point.

Only supported when the DFLAG_SPAWN distribution flag has been passed.

SPAWN_REQUEST_TT

{30, ReqId, From, GroupLeader, {Module, Function, Arity}, OptList, Token}

Followed by ArgList.

Same as SPAWN_REQUEST, but also with a sequential trace Token.

Only supported when the DFLAG_SPAWN distribution flag has been passed.

SPAWN_REPLY

{31, ReqId, To, Flags, Result}

This signal is sent as a reply to a process previously sending a SPAWN_REQUEST signal.

ReqId :: reference()

Request identifier. Also used as monitor reference in case the monitor option has been passed.

To :: pid()

Process identifier of the process making the spawn request.

Flags :: integer() >= 0

A bit flag field of bit flags bitwise or:ed together. Currently the following flags are defined:

1

A link between To and Result was set up on the node where Result resides.

2

A monitor from To to Result was set up on the node where Result resides.

Result :: pid() | atom()

Result of the operation. If Result is a process identifier, the operation succeeded and the process identifier is the identifier of the newly created process. If Result is an atom, the operation failed and the atom identifies failure reason.

Only supported when the DFLAG_SPAWN distribution flag has been passed.

SPAWN_REPLY_TT

{32, ReqId, To, Flags, Result, Token}

Same as SPAWN_REPLY, but also with a sequential trace Token.

Only supported when the DFLAG_SPAWN distribution flag has been passed.