Almost done... have to fix send

and then of course test it

there will be no bugs, right?

src/fd_ws.erl

@@ -4,20 +4,58 @@
 -module(fd_ws).
 
 -export_type([
+    opcode/0,
+    frame/0,
+    ws_msg/0
 ]).
 
 -export([
     handshake/1,
-    response_token/1,
     recv/2,
     send/2,
-    pong/1,
+    pong/1, pong/2
 ]).
 
 -include("http.hrl").
 
 -type request()  :: #request{}.
 -type response() :: #response{}.
+-type tcp_error() :: closed
+                   | {timeout, RestData :: binary() | erlang:iovec()}
+                   | inet:posix().
+
+-define(MAX_PAYLOAD_SIZE, (1 bsl 63)).
+
+%% Frames
+%% https://datatracker.ietf.org/doc/html/rfc6455#section-5.2
+
+-type opcode() :: continuation
+                | text
+                | binary
+                | close
+                | ping
+                | pong.
+
+-record(frame,
+        {fin            = none :: none | boolean(),
+         rsv            = none :: none | <<_:3>>,
+         opcode         = none :: none | opcode(),
+         mask           = none :: none | boolean(),
+         payload_length = none :: none | non_neg_integer(),
+         masking_key    = none :: none | <<>> | <<_:32>>,
+         payload        = none :: none | binary()}).
+
+-type frame() :: #frame{}.
+
+
+%% porcelain messages
+
+-type ws_msg() :: {text,   Payload :: iodata()}
+                | {binary, Payload :: iodata()}
+                | {close,  Payload :: iodata()}
+                | {ping,   Payload :: iodata()}
+                | {pong,   Payload :: iodata()}.
+
 
 
 -spec handshake(Req) -> Result
@@ -49,7 +87,23 @@
 % the retarded web date, rendering the response, sending it over the socket,
 % etc.
 %
-% ClientExtensions only joins the <<"sec-websocket-extensions">> fields with ", "
+% The returned ClientExtensions is the result of joining the
+% <<"sec-websocket-extensions">> fields with ", "
+%
+% quoth section 9.1: https://datatracker.ietf.org/doc/html/rfc6455#section-9.1
+%
+% > Note that like other HTTP header fields, this header field MAY be
+% > split or combined across multiple lines.  Ergo, the following are
+% > equivalent:
+% >
+% >       Sec-WebSocket-Extensions: foo
+% >       Sec-WebSocket-Extensions: bar; baz=2
+% >
+% > is exactly equivalent to
+% >
+% >       Sec-WebSocket-Extensions: foo, bar; baz=2
+%
+% Nobody actually uses extensions, so how you choose to parse this is on you.
 
 handshake(R = #request{method = get, headers = Hs}) ->
     %% downcase the headers because have to match on them
@@ -145,15 +199,6 @@ unfuck_protocol_string([], PartsRev) ->
     when Headers    :: [{Key, Val}],
          Key        :: binary(),
          Val        :: binary().
-
-client_extensions(DowncaseHeaders) ->
-    unfuck_extensions_string(DowncaseHeaders, []).
-
-
--spec unfuck_extensions_string(KVPairs, Acc) -> Unfucked
-    when KVPairs  :: [{Key :: binary(), Val :: binary()}],
-         Acc      :: Unfucked,
-         Unfucked :: binary().
 % @private
 % quoth section 9.1: https://datatracker.ietf.org/doc/html/rfc6455#section-9.1
 %
@@ -172,6 +217,11 @@ client_extensions(DowncaseHeaders) ->
 % matches <<"sec-websocket-extensions">>, then csv its value to the thing
 % @end
 
+
+client_extensions(DowncaseHeaders) ->
+    unfuck_extensions_string(DowncaseHeaders, []).
+
+
 unfuck_extensions_string([{<<"sec-websocket-extensions">>, Part} | Rest], Acc) ->
     unfuck_extensions_string(Rest, [Part | Acc]);
 unfuck_extensions_string([_ | Rest], Acc) ->
@@ -286,37 +336,250 @@ response_token(ChallengeToken) when is_binary(ChallengeToken) ->
 
 
 
--type opcode() :: continuation
-                | text
-                | binary
-                | close
-                | ping
-                | pong.
-
--record(frame,
-        {fin            = none :: none | boolean(),
-         rsv            = none :: none | <<_:3>>,
-         opcode         = none :: none | opcode(),
-         mask           = none :: none | boolean(),
-         payload_length = none :: none | non_neg_integer(),
-         masking_key    = none :: none | <<_:256>>,
-         payload        = none :: none | binary()}).
-
--type frame() :: #frame{}.
-
-
-
 -spec recv(Socket, Received) -> Result
     when Socket    :: gen_tcp:socket(),
          Received  :: binary(),
-         Result   :: {ok, binary()}
+         Result    :: {ok, Message, Frames, Remainder}
-                   | {error, Reason}
+                    | {error, Reason},
+         Message   :: ws_msg(),
+         Frames    :: [frame()],
+         Remainder :: binary(),
          Reason    :: any().
+% @doc
+% Equivalent to recv(Socket, Received, [])
 
 recv(Sock, Recv) ->
-    recv(#frame{}, Sock, Recv).
+    recv(Sock, Recv, []).
 
 
+
+-spec recv(Socket, Received, Frames) -> Result
+    when Socket    :: gen_tcp:socket(),
+         Received  :: binary(),
+         Frames    :: [frame()],
+         Result    :: {ok, Message, NewFrames, Remainder}
+                    | {error, Reason},
+         Message   :: ws_msg(),
+         NewFrames :: Frames,
+         Remainder :: binary(),
+         Reason    :: any().
+% @doc
+% Equivalent to recv(Socket, Received, [])
+
+recv(Sock, Received, Frames) ->
+    case maybe_pop_msg(Frames) of
+        {ok, Message, NewFrames} ->
+            {ok, Message, NewFrames, Received};
+        incomplete ->
+            case recv_frame(#frame{}, Sock, Received) of
+                {ok, Frame, NewReceived} ->
+                    NewFrames = [Frame | Frames],
+                    recv(Sock, NewReceived, NewFrames);
+                Error ->
+                    Error
+            end;
+        Error ->
+            Error
+    end.
+
+
+-spec maybe_pop_msg(Frames) -> Result
+    when Frames    :: [frame()],
+         Result    :: {ok, Message, NewFrames}
+                    | incomplete
+                    | {error, Reason},
+         Message   :: ws_msg(),
+         NewFrames :: Frames,
+         Reason    :: any().
+% @private
+% try to parse the stack of frames into a single message
+%
+% ignores RSV bits
+% @end
+
+maybe_pop_msg([]) ->
+    incomplete;
+% case 1: control frames
+% note that maybe_control_msg checks that the fin bit is true
+%
+% meaning if the client sends a malicious control frame with fin=false, that
+% error will be caught in maybe_control_msg
+maybe_pop_msg([Frame = #frame{opcode = ControlOpcode} | Frames])
+        when (ControlOpcode =:= close)
+             orelse (ControlOpcode =:= ping)
+             orelse (ControlOpcode =:= pong) ->
+    case maybe_control_msg(Frame) of
+        {ok, Msg} -> {ok, Msg, Frames};
+        Error     -> Error
+    end;
+% case 2: messages
+% finished message in a single frame, just pull here
+maybe_pop_msg([Frame = #frame{fin         = true,
+                              opcode      = DataOpcode,
+                              mask        = Mask,
+                              masking_key = Key,
+                              payload     = Payload}
+              | Rest])
+        when DataOpcode =:= text; DataOpcode =:= binary ->
+    case maybe_unmask(Frame, Mask, Key, Payload) of
+        {ok, Unmasked} ->
+            Message = {DataOpcode, Unmasked},
+            {ok, Message, Rest};
+        Error ->
+            Error
+    end;
+% end of a long message
+maybe_pop_msg(Frames = [#frame{fin    = true,
+                               opcode = continuation} | _]) ->
+    maybe_long_data_msg(Frames);
+% unfinished message, say we need more
+maybe_pop_msg([#frame{fin    = false,
+                      opcode = continuation}
+               | _]) ->
+    incomplete;
+% wtf... this case should be impossible
+maybe_pop_msg([Frame | _]) ->
+    {error, {wtf_frame, Frame}}.
+
+
+
+-spec maybe_long_data_msg(Frames) -> Result
+    when Frames    :: [frame()],
+         Result    :: {ok, Message, NewFrames}
+                    | {error, Reason},
+         Message   :: ws_msg(),
+         NewFrames :: Frames,
+         Reason    :: any().
+% @private
+% assumes:
+% 1. top of stack is a finished frame
+% 2. top opcode is continuation
+% 3. the stack corresponds to a linear sequence of frames all corresponding to
+%    one message, until we get to the leading frame of the message, which must
+%    have opcode text|binary
+%
+%    the reason we can make this assumption is because anterior in the call
+%    chain is recv/3, which eagerly consumes control messages
+%
+%    meaning if we encounter a control frame in the middle here, we can assume
+%    there is some sort of bug
+%
+% TODO: I am NOT enforcing that the data message consumes the entire stack of
+% frames. Given that the context here is eager consumption, this might be a
+% point of enforcement. Need to think about this.
+% @end
+
+maybe_long_data_msg(Frames) ->
+    mldm(Frames, Frames, <<>>).
+
+
+% general case: decode the payload in this frame
+mldm(OrigFrames, [Frame | Rest], Acc) ->
+    Opcode  = Frame#frame.opcode,
+    Mask    = Frame#frame.mask,
+    Key     = Frame#frame.masking_key,
+    Payload = Frame#frame.payload,
+    case maybe_unmask(Frame, Mask, Key, Payload) of
+        {ok, Unmasked} ->
+            NewAcc = <<Unmasked/binary, Acc/binary>>,
+            case Opcode of
+                continuation -> mldm(OrigFrames, Rest, NewAcc);
+                text         -> {ok, {text, NewAcc}, Rest};
+                binary       -> {ok, {binary, NewAcc}, Rest};
+                _            -> {error, {illegal_data_frame, Frame, OrigFrames, Acc}}
+            end;
+        Error ->
+            Error
+    end;
+% out of frames
+mldm(OrigFrames, [], Acc) ->
+    {error, {no_start_frame, Acc, OrigFrames}}.
+
+
+
+-spec maybe_control_msg(Frame) -> Result
+    when Frame   :: frame(),
+         Result  :: {ok, Message}
+                  | {error, Reason},
+         Message :: ws_msg(),
+         Reason  :: any().
+% @private
+% assume the frame is a control frame, validate it, and unmask the payload
+%
+% TODO: this doesn't enforce that messages from the client HAVE to be masked,
+% which strictly speaking is part of the protocol.
+
+maybe_control_msg(F = #frame{fin            = true,
+                             opcode         = Opcode,
+                             mask           = Mask,
+                             payload_length = Len,
+                             masking_key    = Key,
+                             payload        = Payload})
+        when ((Opcode =:= close) orelse (Opcode =:= ping) orelse (Opcode =:= pong))
+             andalso (Len =< 125) ->
+    case maybe_unmask(F, Mask, Key, Payload) of
+        {ok, UnmaskedPayload} ->
+            Msg = {Opcode, UnmaskedPayload},
+            {ok, Msg};
+        Error ->
+            Error
+    end;
+maybe_control_msg(F) ->
+    {error, {illegal_frame, F}}.
+
+
+
+-spec maybe_unmask(Frame, Mask, Key, Payload) -> Result
+    when Frame    :: frame(),
+         Mask     :: boolean(),
+         Key      :: <<>> | <<_:32>>,
+         Payload  :: binary(),
+         Result   :: {ok, Unmasked}
+                   | {error, Reason},
+         Unmasked :: binary(),
+         Reason   :: any().
+% @private
+% unmask the payload
+% @end
+
+% eliminate invalid pairs of {mask, masking_key}
+maybe_unmask(_, true,  <<Key:4/bytes>>, Payload) -> {ok, mask_unmask(Key, Payload)};
+maybe_unmask(_, false, <<>>,            Payload) -> {ok, Payload};
+maybe_unmask(F, true,  <<>>,            _)       -> {error, {illegal_frame, F}};
+maybe_unmask(F, false, <<_:4/bytes>>,   _)       -> {error, {illegal_frame, F}}.
+
+
+%% invertible
+%% see: https://datatracker.ietf.org/doc/html/rfc6455#section-5.3
+mask_unmask(Key = <<_:4/bytes>>, Payload) ->
+    mu(Key, Key, Payload, <<>>).
+
+% essentially this is a modular zipWith xor of the masking key with the payload
+mu(Key, <<KeyByte:8, KeyRest/binary>>, <<PayloadByte:8, PayloadRest/binary>>, Acc) ->
+    NewByte = KeyByte bxor PayloadByte,
+    NewAcc  = <<Acc/binary, NewByte:8>>,
+    mu(Key, KeyRest, PayloadRest, NewAcc);
+% this is the case where we need to refresh the active key
+mu(Key, <<>>, Payload, Acc) ->
+    mu(Key, Key, Payload, Acc);
+% done
+mu(_, _, <<>>, Acc) ->
+    Acc.
+
+
+
+-spec recv_frame(Parsed, Socket, Received) -> Result
+    when Parsed    :: frame(),
+         Socket    :: gen_tcp:socket(),
+         Received  :: bitstring(),
+         Result    :: {ok, frame(), Remainder}
+                    | {error, Reason},
+         Remainder :: bitstring(),
+         Reason    :: any().
+% @private
+% parse a single frame off the socket
+% @end
+
 %% frame: 1 bit
 recv_frame(Frame = #frame{fin = none}, Sock, <<FinBit:1, Rest/bits>>) ->
     NewFin =
@@ -326,41 +589,107 @@ recv_frame(Frame = #frame{fin = none}, Sock, <<FinBit:1, Rest/bits>>) ->
         end,
     NewFrame = Frame#frame{fin = NewFin},
     recv_frame(NewFrame, Sock, Rest);
-recv_frame(Frame = #frame{fin = none}, Sock, <<>>) ->
+recv_frame(Frame = #frame{fin = none}, Sock, Received = <<>>) ->
-    case inet:setopts(Sock, [{active, once}]) of
+    recv_frame_await(Frame, Sock, Received);
-        ok ->
-            receive
-                {tcp, Sock, Bin}            -> recv_frame(Frame, Sock, Bin);
-                {tcp_closed, Socket}        -> {error, tcp_closed};
-                {tcp_error, Socket, Reason} -> {error, {tcp_error, Reason}}
-            after 3000 ->
-                {error, timeout}
-            end;
-        {error, Reason} ->
-            {error, {inet, Reason}}
-    end;
 %% rsv: 3 bits
 recv_frame(Frame = #frame{rsv = none}, Sock, <<RSV:3/bits, Rest/bits>>) ->
     NewFrame = Frame#frame{rsv = RSV},
     recv_frame(NewFrame, Sock, Rest);
 recv_frame(Frame = #frame{rsv = none}, Sock, Received) ->
+    recv_frame_await(Frame, Sock, Received);
+%% opcode: 4 bits
+recv_frame(Frame = #frame{opcode = none}, Sock, <<OpcodeInt:4, Rest/bits>>) ->
+    Opcode =
+        case OpcodeInt of
+            0  -> continuation;
+            1  -> text;
+            2  -> binary;
+            8  -> close;
+            9  -> ping;
+            10 -> pong;
+            _  -> bad_opcode
+        end,
+    case Opcode of
+        bad_opcode ->
+            {error, {bad_opcode, OpcodeInt}};
+        _ ->
+            NewFrame = Frame#frame{opcode = Opcode},
+            recv_frame(NewFrame, Sock, Rest)
+    end;
+recv_frame(Frame = #frame{opcode = none}, Sock, Received) ->
+    recv_frame_await(Frame, Sock, Received);
+%% mask: 1 bit
+recv_frame(Frame = #frame{mask = none}, Sock, <<MaskBit:1, Rest/bits>>) ->
+    NewMask =
+        case MaskBit of
+            0 -> false;
+            1 -> true
+        end,
+    NewFrame = Frame#frame{mask = NewMask},
+    recv_frame(NewFrame, Sock, Rest);
+recv_frame(Frame = #frame{mask = none}, Sock, Received = <<>>) ->
+    recv_frame_await(Frame, Sock, Received);
+%% payload length: variable (yay)
+% first case: short length 0..125
+recv_frame(Frame = #frame{payload_length = none}, Sock, <<Len:7, Rest/bits>>) when Len =< 125 ->
+    NewFrame = Frame#frame{payload_length = Len},
+    recv_frame(NewFrame, Sock, Rest);
+% second case: 126 -> 2 bytes to follow
+recv_frame(Frame = #frame{payload_length = none}, Sock, <<126:7, Len:16, Rest/bits>>) ->
+    NewFrame = Frame#frame{payload_length = Len},
+    recv_frame(NewFrame, Sock, Rest);
+% third case: 127 -> 8 bytes to follow
+% bytes must start with a 0 bit
+recv_frame(_Frame = #frame{payload_length = none}, _Sock, <<127:7, 1:1, _/bits>>) ->
+    {error, {illegal_frame, "payload length >= 1 bsl 63"}};
+% 127, next is a legal length, continue
+recv_frame(Frame = #frame{payload_length = none}, Sock, <<127:7, Len:64, Rest/bits>>) ->
+    NewFrame = Frame#frame{payload_length = Len},
+    recv_frame(NewFrame, Sock, Rest);
+% otherwise wait
+recv_frame(Frame = #frame{payload_length = none}, Sock, Received) ->
+    recv_frame_await(Frame, Sock, Received);
+%% masking key: 0 or 4 bits
+% not expecting a masking key, fill in that field here
+recv_frame(Frame = #frame{mask = false, masking_key = none}, Sock, Received) ->
+    NewFrame = Frame#frame{masking_key = <<>>},
+    recv_frame(NewFrame, Sock, Received);
+% expecting one
+recv_frame(Frame = #frame{mask = true, masking_key = none}, Sock, <<Key:4/bytes, Rest/bits>>) ->
+    NewFrame = Frame#frame{masking_key = Key},
+    recv_frame(NewFrame, Sock, Rest);
+% not found
+recv_frame(Frame = #frame{mask = true, masking_key = none}, Sock, Received) ->
+    recv_frame_await(Frame, Sock, Received);
+%% payload
+recv_frame(Frame = #frame{payload_length = Len, payload = none}, Sock, Received) when is_integer(Len) ->
+    case Received of
+        % we have enough bytes
+        <<Payload:Len/bytes, Rest/bits>> ->
+            FinalFrame = Frame#frame{payload = Payload},
+            {ok, FinalFrame, Rest};
+        % we do not have enough bytes
+        _ ->
+            recv_frame_await(Frame, Sock, Received)
+    end.
+
+
+
+%% factoring this out into a function to reduce repetition
+recv_frame_await(Frame, Sock, Received) ->
     case inet:setopts(Sock, [{active, once}]) of
         ok ->
             receive
                 {tcp, Sock, Bin}          -> recv_frame(Frame, Sock, <<Received/bits, Bin/binary>>);
-                {tcp_closed, Socket}        -> {error, tcp_closed};
+                {tcp_closed, Sock}        -> {error, tcp_closed};
-                {tcp_error, Socket, Reason} -> {error, {tcp_error, Reason}}
+                {tcp_error, Sock, Reason} -> {error, {tcp_error, Reason}}
             after 3000 ->
                 {error, timeout}
             end;
         {error, Reason} ->
             {error, {inet, Reason}}
-    end;
+    end.
-%% opcode
+
-recv_frame(Frame = #frame{opcode = none}, Sock, <<OpcodeInt:4, Rest/bits>>) ->
-    if
-        OpcodeInt =:= 
-    end;
 
 
 -spec send(Socket, Payload) -> Result
@@ -371,7 +700,13 @@ recv_frame(Frame = #frame{opcode = none}, Sock, <<OpcodeInt:4, Rest/bits>>) ->
          Reason   :: closed | {timeout, RestData} | inet:posix(),
          RestData :: binary() | erlang:iovec().
 % @doc
+% FIXME: this should be sending a message, not an arbitrary payload
+%
 % send binary data over Socket. handles frame nonsense
+%
+% types the payload as bytes
+%
+% max payload size is 2^63 - 1 bytes
 % @end
 
 send(Socket, Payload) ->
@@ -380,7 +715,7 @@ send(Socket, Payload) ->
     send_frame(Socket, Frame).
 
 
-payload_to_frame(Payload) when byte_size(Payload) < (1 bsl 64) ->
+payload_to_frame(Payload) when byte_size(Payload) < ?MAX_PAYLOAD_SIZE ->
     #frame{fin            = true,
            opcode         = binary,
            mask           = false,
@@ -395,8 +730,7 @@ payload_to_frame(Payload) when byte_size(Payload) < (1 bsl 64) ->
          Frame    :: frame(),
          Result   :: ok
                    | {error, Reason},
-         Reason   :: closed | {timeout, RestData} | inet:posix(),
+         Reason   :: tcp_error().
-         RestData :: binary() | erlang:iovec().
 % @private
 % send a frame on the socket
 % @end
@@ -412,6 +746,14 @@ send_frame(Sock, Frame) ->
          Binary :: binary().
 % @private
 % render a frame
+%
+% TODO: this doesn't check/do masking
+%
+% This is a non-issue as long as this is only used for rendering messages sent
+% from server to client (unmasked per protocol). However, for debugging
+% purposes, a user of this library might want to test how frames render with
+% masking. This functionality is not currently supported, but is a planned
+% addition in the future.
 % @end
 
 render_frame(#frame{fin            = Fin,
@@ -466,7 +808,7 @@ render_payload_length(Len) when 0 =< Len, Len =< 125 ->
     <<Len:7>>;
 render_payload_length(Len) when 126 =< Len, Len =< 2#1111_1111_1111_1111 ->
     <<126:7, Len:16>>;
-render_payload_length(Len) when (1 bsl 16) =< Len, Len < (1 bsl 63) ->
+render_payload_length(Len) when (1 bsl 16) =< Len, Len < ?MAX_PAYLOAD_SIZE ->
     <<127:7, Len:64>>.
 
 
@@ -479,8 +821,21 @@ render_payload_length(Len) when (1 bsl 16) =< Len, Len < (1 bsl 63) ->
          RestData :: binary() | erlang:iovec().
 
 pong(Sock) ->
+    pong(Sock, <<>>).
+
+
+
+-spec pong(Socket, Payload) ->  Result
+    when Socket   :: gen_tcp:socket(),
+         Payload  :: binary(),
+         Result   :: ok
+                   | {error, Reason},
+         Reason   :: closed | {timeout, RestData} | inet:posix(),
+         RestData :: binary() | erlang:iovec().
+
+pong(Sock, Payload) when is_binary(Payload), byte_size(Payload) < ?MAX_PAYLOAD_SIZE ->
     Frame = #frame{fin            = true,
                    opcode         = pong,
-                   payload_length = 0,
+                   payload_length = byte_size(Payload),
-                   payload        = <<>>},
+                   payload        = Payload},
     send_frame(Sock, Frame).