3 Using SSL application API

To see relevant version information for ssl, call ssl:versions/0 .

To see all supported cipher suites, call ssl:cipher_suites(all, 'tlsv1.3') . The available cipher suites for a connection depend on the TLS version and pre TLS-1.3 also on the certificate. To see the default cipher suite list change all to default. Note that TLS 1.3 and previous versions does not have any cipher suites in common, for listing cipher suites for a specific version use ssl:cipher_suites(exclusive, 'tlsv1.3') . Specific cipher suites that you want your connection to use can also be specified. Default is to use the strongest available.

3.1 Setting up Connections

This section shows a small example of how to set up client/server connections using the Erlang shell. The returned value of the sslsocket is abbreviated with [...] as it can be fairly large and is opaque.

Minimal Example

Note

The minimal setup is not the most secure setup of TLS/DTLS.

To set up client/server connections:

Step 1: Start the server side:

1 server> ssl:start().
ok

Step 2: Create a TLS listen socket: (To run DTLS add the option {protocol, dtls})

2 server> {ok, ListenSocket} =
ssl:listen(9999, [{certfile, "cert.pem"}, {keyfile, "key.pem"},{reuseaddr, true}]).
{ok,{sslsocket, [...]}}

Step 3: Do a transport accept on the TLS listen socket:

3 server> {ok, TLSTransportSocket} = ssl:transport_accept(ListenSocket).
{ok,{sslsocket, [...]}}

Step 4: Start the client side:

1 client> ssl:start().
ok

To run DTLS add the option {protocol, dtls} to third argument.

2 client> {ok, Socket} = ssl:connect("localhost", 9999,  [], infinity).
{ok,{sslsocket, [...]}}

Step 5: Do the TLS handshake:

4 server> {ok, Socket} = ssl:handshake(TLSTransportSocket).
{ok,{sslsocket, [...]}}

Step 6: Send a message over TLS:

5 server> ssl:send(Socket, "foo").
ok

Step 7: Flush the shell message queue to see that the message was sent on the server side:

3 client> flush().
Shell got {ssl,{sslsocket,[...]},"foo"}
ok

Upgrade Example - TLS only

Note

To upgrade a TCP/IP connection to a TLS connection, the client and server must agree to do so. The agreement can be accomplished by using a protocol, for example, the one used by HTTP specified in RFC 2817.

To upgrade to a TLS connection:

Step 1: Start the server side:

1 server> ssl:start().
ok

Step 2: Create a normal TCP listen socket:

2 server> {ok, ListenSocket} = gen_tcp:listen(9999, [{reuseaddr, true}]).
{ok, #Port<0.475>}

Step 3: Accept client connection:

3 server> {ok, Socket} = gen_tcp:accept(ListenSocket).
{ok, #Port<0.476>}

Step 4: Start the client side:

1 client> ssl:start().
ok

2 client> {ok, Socket} = gen_tcp:connect("localhost", 9999,  [], infinity).

Step 5: Ensure active is set to false before trying to upgrade a connection to a TLS connection, otherwise TLS handshake messages can be delivered to the wrong process:

4 server> inet:setopts(Socket, [{active, false}]).
ok

Step 6: Do the TLS handshake:

5 server> {ok, TLSSocket} = ssl:handshake(Socket, [{cacertfile, "cacerts.pem"},
{certfile, "cert.pem"}, {keyfile, "key.pem"}]).
{ok,{sslsocket,[...]}}

Step 7: Upgrade to a TLS connection. The client and server must agree upon the upgrade. The server must call ssl:handshake/2 before the client calls ssl:connect/3.

3 client>{ok, TLSSocket} = ssl:connect(Socket, [{cacertfile, "cacerts.pem"},
{certfile, "cert.pem"}, {keyfile, "key.pem"}], infinity).
{ok,{sslsocket,[...]}}

Step 8: Send a message over TLS:

4 client> ssl:send(TLSSocket, "foo").
ok

Step 9: Set active true on the TLS socket:

4 server> ssl:setopts(TLSSocket, [{active, true}]).
ok

Step 10: Flush the shell message queue to see that the message was sent on the client side:

5 server> flush().
Shell got {ssl,{sslsocket,[...]},"foo"}
ok

3.2 Customizing cipher suites

Fetch default cipher suite list for a TLS/DTLS version. Change default to all to get all possible cipher suites.

1>  Default = ssl:cipher_suites(default, 'tlsv1.2').
    [#{cipher => aes_256_gcm,key_exchange => ecdhe_ecdsa,
    mac => aead,prf => sha384}, ....]

In OTP 20 it is desirable to remove all cipher suites that uses rsa key exchange (removed from default in 21)

2> NoRSA =
    ssl:filter_cipher_suites(Default,
                            [{key_exchange, fun(rsa) -> false;
			                       (_) -> true end}]).
    [...]

Pick just a few suites

 3> Suites =
    ssl:filter_cipher_suites(Default,
                            [{key_exchange, fun(ecdh_ecdsa) -> true;
			                       (_) -> false end},
                             {cipher, fun(aes_128_cbc) ->true;
			                  (_) ->false end}]).
    [#{cipher => aes_128_cbc,key_exchange => ecdh_ecdsa,
     mac => sha256,prf => sha256},
     #{cipher => aes_128_cbc,key_exchange => ecdh_ecdsa,mac => sha,
     prf => default_prf}]

Make some particular suites the most preferred, or least preferred by changing prepend to append.

 4>ssl:prepend_cipher_suites(Suites, Default).
  [#{cipher => aes_128_cbc,key_exchange => ecdh_ecdsa,
     mac => sha256,prf => sha256},
   #{cipher => aes_128_cbc,key_exchange => ecdh_ecdsa,mac => sha,
     prf => default_prf},
   #{cipher => aes_256_cbc,key_exchange => ecdhe_ecdsa,
     mac => sha384,prf => sha384}, ...]

3.3 Using an Engine Stored Key

Erlang ssl application is able to use private keys provided by OpenSSL engines using the following mechanism:

1> ssl:start().
ok

Load a crypto engine, should be done once per engine used. For example dynamically load the engine called MyEngine:

2> {ok, EngineRef} =
crypto:engine_load(<<"dynamic">>,
                   [{<<"SO_PATH">>, "/tmp/user/engines/MyEngine"},<<"LOAD">>],[]).
{ok,#Ref<0.2399045421.3028942852.173962>}

Create a map with the engine information and the algorithm used by the engine:

3> PrivKey =
 #{algorithm => rsa,
   engine => EngineRef,
   key_id => "id of the private key in Engine"}.

Use the map in the ssl key option:

4> {ok, SSLSocket} =
ssl:connect("localhost", 9999,
            [{cacertfile, "cacerts.pem"},
             {certfile, "cert.pem"},
             {key, PrivKey}], infinity).

3.4 Session Reuse pre TLS 1.3

Clients can request to reuse a session established by a previous full handshake between that client and server by sending the id of the session in the initial handshake message. The server may or may not agree to reuse it. If agreed the server will send back the id and if not it will send a new id. The ssl application has several options for handling session reuse.

On the client side the ssl application will save session data to try to automate session reuse on behalf of the client processes on the Erlang node. Note that only verified sessions will be saved for security reasons, that is session resumption relies on the certificate validation to have been run in the original handshake. To minimize memory consumption only unique sessions will be saved unless the special save value is specified for the following option {reuse_sessions, boolean() | save} in which case a full handhake will be performed and that specific session will have been saved before the handshake returns. The session id and even an opaque binary containing the session data can be retrieved using ssl:connection_information/1 function. A saved session (guaranteed by the save option) can be explicitly reused using {reuse_session, SessionId}. Also it is possible for the client to reuse a session that is not saved by the ssl application using {reuse_session, {SessionId, SessionData}}.

Note

When using explicit session reuse, it is up to the client to make sure that the session being reused is for the correct server and has been verified.

Here follows a client side example, divide into several steps for readability.

Step 1 - Automated Session Reuse

1> ssl:start().
ok

2> {ok, C1} = ssl:connect("localhost", 9999, [{verify, verify_peer},
					      {versions, ['tlsv1.2']},
					      {cacertfile, "cacerts.pem"}]).
{ok,{sslsocket,{gen_tcp,#Port<0.7>,tls_connection,undefined}, ...}}

3> ssl:connection_information(C1, [session_id]).
{ok,[{session_id,<<95,32,43,22,35,63,249,22,26,36,106,
                   152,49,52,124,56,130,192,137,161,
                   146,145,164,232,...>>}]}

%% Reuse session if possible, note that if C2 is really fast the session
%% data might not be available for reuse.
4> {ok, C2} = ssl:connect("localhost", 9999, [{verify, verify_peer},
					      {versions, ['tlsv1.2']},
					      {cacertfile, "cacerts.pem"},
					      {reuse_sessions, true}]).
{ok,{sslsocket,{gen_tcp,#Port<0.8>,tls_connection,undefined}, ...]}}

%% C2 got same session ID as client one, session was automatically reused.
5> ssl:connection_information(C2, [session_id]).
{ok,[{session_id,<<95,32,43,22,35,63,249,22,26,36,106,
                   152,49,52,124,56,130,192,137,161,
                   146,145,164,232,...>>}]}

Step 2- Using save Option

%% We want save this particular session for reuse although it has the same basis as C1
6> {ok, C3} = ssl:connect("localhost", 9999, [{verify, verify_peer},
					      {versions, ['tlsv1.2']},
					      {cacertfile, "cacerts.pem"},
					      {reuse_sessions, save}]).
{ok,{sslsocket,{gen_tcp,#Port<0.9>,tls_connection,undefined}, ...]}}

%% A full handshake is performed and we get a new session ID
7> {ok, [{session_id, ID}]} = ssl:connection_information(C3, [session_id]).
{ok,[{session_id,<<91,84,27,151,183,39,84,90,143,141,
                   121,190,66,192,10,1,27,192,33,95,78,
                   8,34,180,...>>}]}

%% Use automatic session reuse
8> {ok, C4} = ssl:connect("localhost", 9999, [{verify, verify_peer},
					      {versions, ['tlsv1.2']},
					      {cacertfile, "cacerts.pem"},
					      {reuse_sessions, true}]).
{ok,{sslsocket,{gen_tcp,#Port<0.10>,tls_connection,
                        undefined}, ...]}}

%% The "saved" one happened to be selected, but this is not a guarantee
9> ssl:connection_information(C4, [session_id]).
{ok,[{session_id,<<91,84,27,151,183,39,84,90,143,141,
                   121,190,66,192,10,1,27,192,33,95,78,
                   8,34,180,...>>}]}

%% Make sure to reuse the "saved" session
10> {ok, C5} = ssl:connect("localhost", 9999, [{verify, verify_peer},
					       {versions, ['tlsv1.2']},
					       {cacertfile, "cacerts.pem"},
					       {reuse_session, ID}]).
{ok,{sslsocket,{gen_tcp,#Port<0.11>,tls_connection,
                        undefined}, ...]}}

11> ssl:connection_information(C5, [session_id]).
{ok,[{session_id,<<91,84,27,151,183,39,84,90,143,141,
                   121,190,66,192,10,1,27,192,33,95,78,
                   8,34,180,...>>}]}

Step 3 - Explicit Session Reuse

%% Perform a full handshake and the session will not be saved for reuse
12> {ok, C9} = ssl:connect("localhost", 9999, [{verify, verify_peer},
				               {versions, ['tlsv1.2']},
		                               {cacertfile, "cacerts.pem"},
					       {reuse_sessions, false},
					       {server_name_indication, disable}]).
{ok,{sslsocket,{gen_tcp,#Port<0.14>,tls_connection, ...}}

%% Fetch session ID and data for C9 connection
12> {ok, [{session_id, ID1}, {session_data, SessData}]} =
	ssl:connection_information(C9, [session_id, session_data]).
{ok,[{session_id,<<9,233,4,54,170,88,170,180,17,96,202,
                   85,85,99,119,47,9,68,195,50,120,52,
                   130,239,...>>},
     {session_data,<<131,104,13,100,0,7,115,101,115,115,105,
                     111,110,109,0,0,0,32,9,233,4,54,170,...>>}]}

%% Explicitly reuse the session from C9
13> {ok, C10} = ssl:connect("localhost", 9999, [{verify, verify_peer},
						{versions, ['tlsv1.2']},
						{cacertfile, "cacerts.pem"},
						{reuse_session, {ID1, SessData}}]).
{ok,{sslsocket,{gen_tcp,#Port<0.15>,tls_connection,
                        undefined}, ...}}

14> ssl:connection_information(C10, [session_id]).
{ok,[{session_id,<<9,233,4,54,170,88,170,180,17,96,202,
                   85,85,99,119,47,9,68,195,50,120,52,
                   130,239,...>>}]}

Step 4 - Not Possible to Reuse Explicit Session by ID Only

%% Try to reuse the session from C9 using only the id
15> {ok, E} = ssl:connect("localhost", 9999, [{verify, verify_peer},
				              {versions, ['tlsv1.2']},
				              {cacertfile, "cacerts.pem"},
					      {reuse_session, ID1}]).
{ok,{sslsocket,{gen_tcp,#Port<0.18>,tls_connection,
                        undefined}, ...}}

%% This will fail (as it is not saved for reuse)
%% and a full handshake will be performed, we get a new id.
16>  ssl:connection_information(E, [session_id]).
{ok,[{session_id,<<87,46,43,126,175,68,160,153,37,29,
                   196,240,65,160,254,88,65,224,18,63,
                   18,17,174,39,...>>}]}

On the server side the the {reuse_sessions, boolean()} option determines if the server will save session data and allow session reuse or not. This can be further customized by the option {reuse_session, fun()} that may introduce a local policy for session reuse.

3.5 Session Tickets and Session Resumption in TLS 1.3

TLS 1.3 introduces a new secure way of resuming sessions by using session tickets. A session ticket is an opaque data structure that is sent in the pre_shared_key extension of a ClientHello, when a client attempts to resume a session with keying material from a previous successful handshake.

Session tickets can be stateful or stateless. A stateful session ticket is a database reference (session ticket store) and used with stateful servers, while a stateless ticket is a self-encrypted and self-authenticated data structure with cryptographic keying material and state data, enabling session resumption with stateless servers.

The choice between stateful or stateless depends on the server requirements as the session tickets are opaque for the clients. Generally, stateful tickets are smaller and the server can guarantee that tickets are only used once. Stateless tickets contain additional data, require less storage on the server side, but they offer different guarantees against anti-replay. See also Anti-Replay Protection in TLS 1.3

Session tickets are sent by servers on newly established TLS connections. The number of tickets sent and their lifetime are configurable by application variables. See also SSL's configuration.

Session tickets are protected by application traffic keys, and in stateless tickets, the opaque data structure itself is self-encrypted.

An example with automatic and manual session resumption:

Step 1 (server): Start the server:

      {ok, _} = application:ensure_all_started(ssl).
      LOpts = [{certfile, "cert.pem"},
               {keyfile, "key.pem"},
               {versions, ['tlsv1.2','tlsv1.3']},
               {session_tickets, stateless}].
      {ok, LSock} = ssl:listen(8001, LOpts).
      {ok, CSock} = ssl:transport_accept(LSock).

Step 2 (client): Start the client and connect to server:

      {ok, _} = application:ensure_all_started(ssl).
      COpts = [{cacertfile, "cert.pem"},
               {versions, ['tlsv1.2','tlsv1.3']},
               {log_level, debug},
               {session_tickets, auto}].
      ssl:connect("localhost", 8001, COpts).

Step 3 (server): Start the TLS handshake:

      ssl:handshake(CSock).

A connection is established using a full handshake. Below is a summary of the exchanged messages:

      >>> TLS 1.3 Handshake, ClientHello ...
      <<< TLS 1.3 Handshake, ServerHello ...
      <<< Handshake, EncryptedExtensions ...
      <<< Handshake, Certificate ...
      <<< Handshake, CertificateVerify ...
      <<< Handshake, Finished ...
      >>> Handshake, Finished ...
      <<< Post-Handshake, NewSessionTicket ...

At this point the client has stored the received session tickets and ready to use them when establishing new connections to the same server.

Step 4 (server): Accept a new connection on the server:

      {ok, CSock2} = ssl:transport_accept(LSock).

Step 5 (client): Make a new connection:

      ssl:connect("localhost", 8001, COpts).

Step 6 (server): Start the handshake:

      ssl:handshake(CSock2).

The second connection is a session resumption using keying material from the previous handshake:

      >>> TLS 1.3 Handshake, ClientHello ...
      <<< TLS 1.3 Handshake, ServerHello ...
      <<< Handshake, EncryptedExtensions ...
      <<< Handshake, Finished ...
      >>> Handshake, Finished ...
      <<< Post-Handshake, NewSessionTicket ...

Manual handling of session tickets is also supported. In manual mode, it is the responsibility of the client to handle received session tickets.

Step 7 (server): Accept a new connection on the server:

      {ok, CSock3} = ssl:transport_accept(LSock).

Step 8 (client): Make a new connection to server:

      {ok, _} = application:ensure_all_started(ssl).
      COpts2 = [{cacertfile, "cert.pem"},
                {versions, ['tlsv1.2','tlsv1.3']},
                {log_level, debug},
                {session_tickets, manual}].
      ssl:connect("localhost", 8001, COpts).

Step 9 (server): Start the handshake:

      ssl:handshake(CSock3).

After the handshake is performed, the user process receives messages with the tickets sent by the server.

Step 10 (client): Receive a new session ticket:

      Ticket = receive {ssl, session_ticket, {_, TicketData}} -> TicketData end.

Step 11 (server): Accept a new connection on the server:

      {ok, CSock4} = ssl:transport_accept(LSock).

Step 12 (client): Initiate a new connection to the server with the session ticket received in Step 10:

      {ok, _} = application:ensure_all_started(ssl).
      COpts2 = [{cacertfile, "cert.pem"},
                {versions, ['tlsv1.2','tlsv1.3']},
                {log_level, debug},
                {session_tickets, manual},
                {use_ticket, [Ticket]}].
      ssl:connect("localhost", 8001, COpts).

Step 13 (server): Start the handshake:

      ssl:handshake(CSock3).

3.6 Early Data in TLS 1.3

TLS 1.3 allows clients to send data on the first flight if the endpoints have a shared crypographic secret (pre-shared key). This means that clients can send early data if they have a valid session ticket received in a previous successful handshake. For more information about session resumption see Session Tickets and Session Resumption in TLS 1.3.

The security properties of Early Data are weaker than other kinds of TLS data. This data is not forward secret, and it is vulnerable to replay attacks. For available mitigation strategies see Anti-Replay Protection in TLS 1.3.

In normal operation, clients will not know which, if any, of the available mitigation strategies servers actually implement, and hence must only send early data which they deem safe to be replayed. For example, idempotent HTTP operations, such as HEAD and GET, can usually be regarded as safe but even they can be exploited by a large number of replays causing resource limit exhaustion and other similar problems.

An example of sending early data with automatic and manual session ticket handling:

Warning

The Early Data feature is experimental in this version of OTP.

Server (with NSS key logging)

    early_data_server() ->
        application:load(ssl),
        {ok, _} = application:ensure_all_started(ssl),
        Port = 11029,
        LOpts = [{certfile, ?SERVER_CERT},
             {keyfile, ?SERVER_KEY},
             {reuseaddr, true},
             {versions, ['tlsv1.2','tlsv1.3']},
             {session_tickets, stateless},
             {early_data, enabled},
             {keep_secrets, true} %% Enable NSS key log (debug option)
            ],
        {ok, LSock} = ssl:listen(Port, LOpts),
        %% Accept first connection
        {ok, CSock0} = ssl:transport_accept(LSock),
        {ok, _} = ssl:handshake(CSock0),
        %% Accept second connection
        {ok, CSock1} = ssl:transport_accept(LSock),
        {ok, Sock} = ssl:handshake(CSock1),
        Sock.

Exporting the secrets (optional)

    {ok, [{keylog, KeylogItems}]} = ssl:connection_information(Sock, [keylog]).
    file:write_file("key.log", [[KeylogItem,$\n] || KeylogItem <- KeylogItems]).

Client (automatic ticket handling):

    early_data_auto() ->
        %% First handshake 1-RTT - get session tickets
	application:load(ssl),
	{ok, _} = application:ensure_all_started(ssl),
	Port = 11029,
	Data = <<"HEAD / HTTP/1.1\r\nHost: \r\nConnection: close\r\n">>,
	COpts0 = [{cacertfile, ?CA_CERT},
	          {versions, ['tlsv1.2', 'tlsv1.3']},
	          {session_tickets, auto}],
        {ok, Sock0} = ssl:connect("localhost", Port, COpts0),

        %% Wait for session tickets
	timer:sleep(500),
	%% Close socket if server cannot handle multiple connections e.g. openssl s_server
	ssl:close(Sock0),

        %% Second handshake 0-RTT
	COpts1 = [{cacertfile, ?CA_CERT},
	          {versions, ['tlsv1.2', 'tlsv1.3']},
		  {session_tickets, auto},
		  {early_data, Data}],
        {ok, Sock} = ssl:connect("localhost", Port, COpts1),
	Sock.

Client (manual ticket handling):

    early_data_manual() ->
        %% First handshake 1-RTT - get session tickets
	application:load(ssl),
	{ok, _} = application:ensure_all_started(ssl),
	Port = 11029,
	Data = <<"HEAD / HTTP/1.1\r\nHost: \r\nConnection: close\r\n">>,
	COpts0 = [{cacertfile, ?CA_CERT},
	          {versions, ['tlsv1.2', 'tlsv1.3']},
	          {session_tickets, manual}],
        {ok, Sock0} = ssl:connect("localhost", Port, COpts0),

        %% Wait for session tickets
	Ticket =
	    receive
	        {ssl, session_ticket, Ticket0} ->
		    Ticket0
            end,

       %% Close socket if server cannot handle multiple connections
       %% e.g. openssl s_server
       ssl:close(Sock0),

       %% Second handshake 0-RTT
       COpts1 = [{cacertfile, ?CA_CERT},
                 {versions, ['tlsv1.2', 'tlsv1.3']},
		 {session_tickets, manual},
		 {use_ticket, [Ticket]},
		 {early_data, Data}],
       {ok, Sock} = ssl:connect("localhost", Port, COpts1),
       Sock.

3.7 Anti-Replay Protection in TLS 1.3

The TLS 1.3 protocol does not provide inherent protection for replay of 0-RTT data but describes mechanisms that SHOULD be implemented by compliant server implementations. The implementation of TLS 1.3 in the SSL application employs all standard methods to prevent potential threats.

Single-use tickets

This mechanism is available with stateful session tickets. Session tickets can only be used once, subsequent use of the same ticket results in a full handshake. Stateful servers enforce this rule by maintaining a database of outstanding valid tickets.

Client Hello Recording

This mechanism is available with stateless session tickets. The server records a unique value derived from the ClientHello (PSK binder) in a given time window. The ticket's age is verified by using both the "obsfuscated_ticket_age" and an additional timestamp encrypted in the ticket data. As the used datastore allows false positives, apparent replays will be answered by doing a full 1-RTT handshake.

Freshness Checks

This mechanism is available with the stateless session tickets. As the ticket data has an embedded timestamp, the server can determine if a ClientHello was sent reasonably recently and accept the 0-RTT handshake, otherwise if falls back to a full 1-RTT handshake. This mechanism is tightly coupled with the previous one, it prevents storing an unlimited number of ClientHellos.

The current implementation uses a pair of Bloom filters to implement the last two mechanisms. Bloom filters are fast, memory-efficient, probabilistic data structures that can tell if an element may be in a set or if it is definitely not in the set.

If the option anti_replay is defined in the server, a pair of Bloom filters (current and old) are used to record incoming ClientHello messages (it is the unique binder value that is actually stored). The current Bloom filter is used for WindowSize seconds to store new elements. At the end of the time window the Bloom filters are rotated (the current Bloom filter becomes the old and an empty Bloom filter is set as current.

The Anti-Replay protection feature in stateless servers executes in the following steps when a new ClientHello is received:

Reported ticket age (obfuscated ticket age) shall be less than ticket lifetime.
Actual ticket age shall be less than the ticket lifetime (stateless session tickets contain the servers timestamp when the ticket was issued).
Ticket shall be used within specified time window (freshness checks).
If all above checks passed both current and old Bloom filters are checked to detect if binder was already seen. Being a probabilistic data structure, false positives can occur and they trigger a full handshake.
If the binder is not seen, the binder is validated. If the binder is valid, the server proceeds with the 0-RTT handshake.