%%% Copyright 2009 Andrew Thompson <andrew@hijacked.us>. All rights reserved.
%%%
%%% Redistribution and use in source and binary forms, with or without
%%% modification, are permitted provided that the following conditions are met:
%%%
%%% 1. Redistributions of source code must retain the above copyright notice,
%%% this list of conditions and the following disclaimer.
%%% 2. Redistributions in binary form must reproduce the above copyright
%%% notice, this list of conditions and the following disclaimer in the
%%% documentation and/or other materials provided with the distribution.
%%%
%%% THIS SOFTWARE IS PROVIDED BY THE FREEBSD PROJECT ``AS IS'' AND ANY EXPRESS OR
%%% IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
%%% MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
%%% EVENT SHALL THE FREEBSD PROJECT OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
%%% INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
%%% (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
%%% LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
%%% ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
%%% (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
%%% SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
%% @doc A module for decoding/encoding MIME 1.0 email.
%% The encoder and decoder operate on the same data structure, which is as follows:
%% A 5-tuple with the following elements: `{Type, SubType, Headers, Parameters, Body}'.
%%
%% `Type' and `SubType' are the MIME type of the email, examples are `text/plain' or
%% `multipart/alternative'. The decoder splits these into 2 fields so you can filter by
%% the main type or by the subtype.
%%
%% `Headers' consists of a list of key/value pairs of binary values eg.
%% `{<<"From">>, <<"Andrew Thompson <andrew@hijacked.us>">>}'. There is no parsing of
%% the header aside from un-wrapping the lines and splitting the header name from the
%% header value.
%%
%% `Parameters' is a list of 3 key/value tuples. The 3 keys are `<<"content-type-params">>',
%% `<<"dispisition">>' and `<<"disposition-params">>'.
%% `content-type-params' is a key/value list of parameters on the content-type header, this
%% usually consists of things like charset and the format parameters. `disposition' indicates
%% how the data wants to be displayed, this is usually 'inline'. `disposition-params' is a list of
%% disposition information, eg. the filename this section should be saved as, the modification
%% date the file should be saved with, etc.
%%
%% Finally, `Body' can be one of several different types, depending on the structure of the email.
%% For a simple email, the body will usually be a binary consisting of the message body, In the
%% case of a multipart email, its a list of these 5-tuple MIME structures. The third possibility,
%% in the case of a message/rfc822 attachment, body can be a single 5-tuple MIME structure.
%%
%% You should see the relevant RFCs (2045, 2046, 2047, etc.) for more information.
%%
%% Note that parts of this module (e.g., `decode/2') use the
%% <a href="https://hex.pm/packages/iconv"><tt>iconv</tt></a> library for string conversion,
%% which you will need to explicitly list as a dependency.
-module(mimemail).
-ifdef(TEST).
-include_lib("eunit/include/eunit.hrl").
-export([rfc2047_utf8_encode/1]).
-endif.
-export([encode/1, encode/2, decode/2, decode/1, get_header_value/2, get_header_value/3, parse_headers/1]).
-export([encode_quoted_printable/1, decode_quoted_printable/1]).
-export_type([
mimetuple/0,
mime_type/0,
mime_subtype/0,
headers/0,
parameters/0,
options/0,
dkim_options/0
]).
-include_lib("kernel/include/logger.hrl").
-define(LOGGER_META, #{domain => [gen_smtp]}).
-define(DEFAULT_MIME_VERSION, <<"1.0">>).
-define(DEFAULT_OPTIONS, [
% default encoding is utf-8 if we can find the iconv module
{encoding, get_default_encoding()},
% should we decode any base64/quoted printable attachments?
{decode_attachments, true},
% should we assume default mime version
{allow_missing_version, true},
% default mime version
{default_mime_version, ?DEFAULT_MIME_VERSION}
]).
% `<<"text">>'
-type mime_type() :: binary().
% `<<"plain">>'
-type mime_subtype() :: binary().
% `[{<<"Content-Type">>, <<"text/plain">>}]'
-type headers() :: [{binary(), binary()}].
-type parameters() ::
%% <<"7bit">> | <<"base64">> | <<"quoted-printable">> etc
#{
transfer_encoding => binary(),
%% [{<<"charset">>, <<"utf-8">>} | {<<"boundary">>, binary()} | {<<"name">>, binary()} etc...]
content_type_params => [{binary(), binary()}],
%% <<"inline">> | <<"attachment">> etc...
disposition => binary(),
%% [{<<"filename">>, binary()}, ]
disposition_params => [{binary(), binary()}]
}.
-type mimetuple() :: {
mime_type(),
mime_subtype(),
headers(),
parameters(),
Body :: binary() | mimetuple() | [mimetuple()]
}.
-type dkim_priv_key() ::
{pem_plain, binary()}
| {pem_encrypted, Key :: binary(), Passwd :: string()}.
-type dkim_options() :: [
{h, [binary()]}
| {d, binary()}
| {s, binary()}
| {t, now | calendar:datetime()}
| {x, calendar:datetime()}
| {c, {simple | relaxed, simple | relaxed}}
| {a, 'rsa-sha256' | 'ed25519-sha256'}
| {private_key, dkim_priv_key()}
].
-type options() :: [
{encoding, binary()}
| {decode_attachment, boolean()}
| {dkim, dkim_options()}
| {allow_missing_version, boolean()}
| {default_mime_version, binary()}
].
-spec decode(Email :: binary()) -> mimetuple().
%% @doc Decode a MIME email from a binary.
decode(All) ->
{Headers, Body} = parse_headers(All),
decode(Headers, Body, ?DEFAULT_OPTIONS).
-spec decode(Email :: binary(), Options :: options()) -> mimetuple().
%% @doc Decode with custom options
decode(All, Options) when is_binary(All), is_list(Options) ->
{Headers, Body} = parse_headers(All),
decode(Headers, Body, Options).
decode(OrigHeaders, Body, Options) ->
?LOG_DEBUG("headers: ~p", [OrigHeaders], ?LOGGER_META),
Encoding = proplists:get_value(encoding, Options, none),
%FixedHeaders = fix_headers(Headers),
Headers = decode_headers(OrigHeaders, [], Encoding),
case parse_with_comments(get_header_value(<<"MIME-Version">>, Headers)) of
undefined ->
AllowMissingVersion = proplists:get_value(allow_missing_version, Options, false),
case parse_content_type(get_header_value(<<"Content-Type">>, Headers)) of
{<<"multipart">>, _SubType, _Parameters} when AllowMissingVersion ->
MimeVersion = proplists:get_value(default_mime_version, Options, ?DEFAULT_MIME_VERSION),
decode_component(Headers, Body, MimeVersion, Options);
{<<"multipart">>, _SubType, _Parameters} ->
erlang:error(non_mime_multipart);
{Type, SubType, CTParameters} ->
NewBody = decode_body(
get_header_value(<<"Content-Transfer-Encoding">>, Headers),
Body,
proplists:get_value(<<"charset">>, CTParameters),
Encoding
),
{Disposition, DispositionParams} =
case parse_content_disposition(get_header_value(<<"Content-Disposition">>, Headers)) of
undefined ->
{<<"inline">>, []};
Disp ->
Disp
end,
Parameters = #{
content_type_params => CTParameters,
disposition => Disposition,
disposition_params => DispositionParams
},
{Type, SubType, Headers, Parameters, NewBody};
undefined ->
Parameters = #{
content_type_params => [{<<"charset">>, <<"us-ascii">>}],
disposition => <<"inline">>,
disposition_params => []
},
{<<"text">>, <<"plain">>, Headers, Parameters,
decode_body(get_header_value(<<"Content-Transfer-Encoding">>, Headers), Body)}
end;
Other ->
decode_component(Headers, Body, Other, Options)
end.
-spec encode(MimeMail :: mimetuple()) -> binary().
encode(MimeMail) ->
encode(MimeMail, []).
%% @doc Encode a MIME tuple to a binary.
encode({Type, Subtype, Headers, ContentTypeParams, Parts}, Options) ->
{FixedParams, FixedHeaders} = ensure_content_headers(Type, Subtype, ContentTypeParams, Headers, Parts, true),
CheckedHeaders = check_headers(FixedHeaders),
EncodedBody = binstr:join(
encode_component(Type, Subtype, CheckedHeaders, FixedParams, Parts),
"\r\n"
),
EncodedHeaders = encode_headers(CheckedHeaders),
SignedHeaders =
case proplists:get_value(dkim, Options) of
undefined -> EncodedHeaders;
DKIM -> dkim_sign_email(EncodedHeaders, EncodedBody, DKIM)
end,
list_to_binary([
binstr:join(SignedHeaders, "\r\n"),
"\r\n\r\n",
EncodedBody
]);
encode(_, _) ->
?LOG_DEBUG("Not a mime-decoded DATA", ?LOGGER_META),
erlang:error(non_mime).
decode_headers(Headers, _, none) ->
Headers;
decode_headers([], Acc, _Charset) ->
lists:reverse(Acc);
decode_headers([{Key, Value} | Headers], Acc, Charset) ->
decode_headers(Headers, [{Key, decode_header(Value, Charset)} | Acc], Charset).
decode_header(Value, Charset) ->
RTokens = tokenize_header(Value, []),
Tokens = lists:reverse(RTokens),
Decoded =
try
decode_header_tokens_strict(Tokens, Charset)
catch
Type:Reason:Stacktrace ->
case decode_header_tokens_permissive(Tokens, Charset, []) of
{ok, Dec} ->
Dec;
error ->
% re-throw original error
erlang:raise(Type, Reason, Stacktrace)
end
end,
iolist_to_binary(Decoded).
-type hdr_token() :: binary() | {Encoding :: binary(), Data :: binary()}.
-spec tokenize_header(binary(), [hdr_token()]) -> [hdr_token()].
tokenize_header(<<>>, Acc) ->
Acc;
tokenize_header(Value, Acc) ->
%% maybe replace "?([^\s]+)\\?" with "?([^\s]*)\\?"?
%% see msg lvuvmm593b8s7pqqfhu7cdtqd4g4najh
%% Subject: =?utf-8?Q??=
%% =?utf-8?Q?=D0=9F=D0=BE=D0=B4=D1=82=D0=B2=D0=B5=D1=80=D0=B4=D0=B8=D1=82=D0=B5=20?=
%% =?utf-8?Q?=D1=80=D0=B5=D0=B3=D0=B8=D1=81=D1=82=D1=80=D0=B0=D1=86=D0=B8=D1=8E=20?=
%% =?utf-8?Q?=D0=B2=20Moy-Rebenok.ru?=
case re:run(Value, "=\\?([-A-Za-z0-9_]+)\\?([qQbB])\\?([^\s]+)\\?=", [ungreedy]) of
nomatch ->
[Value | Acc];
{match, [{AllStart, AllLen}, {EncodingStart, EncodingLen}, {TypeStart, _}, {DataStart, DataLen}]} ->
%% RFC 2047 #2 (encoded-word)
Encoding = binstr:substr(Value, EncodingStart + 1, EncodingLen),
Type = binstr:to_lower(binstr:substr(Value, TypeStart + 1, 1)),
Data = binstr:substr(Value, DataStart + 1, DataLen),
EncodedData =
case Type of
<<"q">> ->
%% RFC 2047 #5. (3)
decode_quoted_printable(binary:replace(Data, <<"_">>, <<"=20">>, [global]));
<<"b">> ->
decode_base64(binary:replace(Data, <<"_">>, <<" ">>, [global]))
end,
Offset =
case
re:run(
binstr:substr(Value, AllStart + AllLen + 1),
"^([\s\t\n\r]+)=\\?[-A-Za-z0-9_]+\\?[^\s]\\?[^\s]+\\?=",
[ungreedy]
)
of
nomatch ->
% no 2047 block immediately following
1;
{match, [{_, _}, {_, WhiteSpaceLen}]} ->
1 + WhiteSpaceLen
end,
NewAcc =
case binstr:substr(Value, 1, AllStart) of
<<>> -> [{fix_encoding(Encoding), EncodedData} | Acc];
Other -> [{fix_encoding(Encoding), EncodedData}, Other | Acc]
end,
tokenize_header(binstr:substr(Value, AllStart + AllLen + Offset), NewAcc)
end.
decode_header_tokens_strict([], _) ->
[];
decode_header_tokens_strict([{Encoding, Data} | Tokens], Charset) ->
{ok, S} = convert(Charset, Encoding, Data),
[S | decode_header_tokens_strict(Tokens, Charset)];
decode_header_tokens_strict([Data | Tokens], Charset) ->
[Data | decode_header_tokens_strict(Tokens, Charset)].
%% this decoder can handle folded not-by-RFC UTF headers, when somebody split
%% multibyte string not by characters, but by bytes. It first join folded
%% string and only then decode it with iconv.
decode_header_tokens_permissive([], _, [Result]) when is_binary(Result) ->
{ok, Result};
decode_header_tokens_permissive([], _, Stack) ->
case lists:all(fun erlang:is_binary/1, Stack) of
true -> {ok, lists:reverse(Stack)};
false -> error
end;
decode_header_tokens_permissive([{Enc, Data} | Tokens], Charset, [{Enc, PrevData} | Stack]) ->
NewData = iolist_to_binary([PrevData, Data]),
{ok, S} = convert(Charset, Enc, NewData),
decode_header_tokens_permissive(Tokens, Charset, [S | Stack]);
decode_header_tokens_permissive([NextToken | _] = Tokens, Charset, [{_, _} | Stack]) when
is_binary(NextToken) orelse is_tuple(NextToken)
->
%% practically very rare case "=?utf-8?Q?BROKEN?=\r\n\t=?windows-1251?Q?maybe-broken?="
%% or "=?utf-8?Q?BROKEN?= raw-ascii-string"
%% drop broken value from stack
decode_header_tokens_permissive(Tokens, Charset, Stack);
decode_header_tokens_permissive([Data | Tokens], Charset, Stack) ->
decode_header_tokens_permissive(Tokens, Charset, [Data | Stack]).
%% x-binaryenc is not a real encoding and is not used for text, so let it pass through
convert(_To, <<"x-binaryenc">>, Data) ->
{ok, Data};
convert(To, From, Data) ->
Result = iconv:convert(From, To, Data),
{ok, Result}.
decode_component(Headers, Body, MimeVsn = <<"1.0", _/binary>>, Options) ->
case parse_content_disposition(get_header_value(<<"Content-Disposition">>, Headers)) of
{Disposition, DispositionParams} ->
ok;
% defaults
_ ->
Disposition = <<"inline">>,
DispositionParams = []
end,
case parse_content_type(get_header_value(<<"Content-Type">>, Headers)) of
{<<"multipart">>, SubType, Parameters} ->
case proplists:get_value(<<"boundary">>, Parameters) of
undefined ->
erlang:error(no_boundary);
Boundary ->
?LOG_DEBUG(
"this is a multipart email of type: ~s and boundary ~s", [SubType, Boundary], ?LOGGER_META
),
Parameters2 = #{
content_type_params => Parameters,
disposition => Disposition,
disposition_params => DispositionParams
},
{<<"multipart">>, SubType, Headers, Parameters2,
split_body_by_boundary(Body, list_to_binary(["--", Boundary]), MimeVsn, Options)}
end;
{<<"message">>, <<"rfc822">>, Parameters} ->
{NewHeaders, NewBody} = parse_headers(Body),
Parameters2 = #{
content_type_params => Parameters,
disposition => Disposition,
disposition_params => DispositionParams
},
{<<"message">>, <<"rfc822">>, Headers, Parameters2, decode(NewHeaders, NewBody, Options)};
{Type, SubType, Parameters} ->
?LOG_DEBUG("body is ~s/~s", [Type, SubType], ?LOGGER_META),
Parameters2 = #{
content_type_params => Parameters,
disposition => Disposition,
disposition_params => DispositionParams
},
{Type, SubType, Headers, Parameters2,
decode_body(
get_header_value(<<"Content-Transfer-Encoding">>, Headers),
Body,
proplists:get_value(<<"charset">>, Parameters),
proplists:get_value(encoding, Options, none)
)};
% defaults
undefined ->
Type = <<"text">>,
SubType = <<"plain">>,
Parameters = #{
content_type_params => [{<<"charset">>, <<"us-ascii">>}],
disposition => Disposition,
disposition_params => DispositionParams
},
{Type, SubType, Headers, Parameters,
decode_body(get_header_value(<<"Content-Transfer-Encoding">>, Headers), Body)}
end;
decode_component(_Headers, _Body, Other, _Options) ->
erlang:error({mime_version, Other}).
-spec get_header_value(Needle :: binary(), Headers :: [{binary(), binary()}], Default :: any()) -> binary() | any().
%% @doc Do a case-insensitive header lookup to return that header's value, or the specified default.
get_header_value(Needle, Headers, Default) ->
?LOG_DEBUG("Headers: ~p", [Headers], ?LOGGER_META),
NeedleLower = binstr:to_lower(Needle),
F =
fun({Header, _Value}) ->
binstr:to_lower(Header) =:= NeedleLower
end,
case lists:search(F, Headers) of
% TODO if there's duplicate headers, should we use the first or the last?
{value, {_Header, Value}} ->
Value;
false ->
Default
end.
-spec get_header_value(Needle :: binary(), Headers :: [{binary(), binary()}]) -> binary() | 'undefined'.
%% @doc Do a case-insensitive header lookup to return the header's value, or `undefined'.
get_header_value(Needle, Headers) ->
get_header_value(Needle, Headers, undefined).
-spec parse_with_comments
(Value :: binary()) -> binary() | no_return();
(Value :: atom()) -> atom().
parse_with_comments(Value) when is_binary(Value) ->
parse_with_comments(Value, [], 0, false);
parse_with_comments(Value) ->
Value.
-spec parse_with_comments(Value :: binary(), Acc :: list(), Depth :: non_neg_integer(), Quotes :: boolean()) ->
binary() | no_return().
parse_with_comments(<<>>, _Acc, _Depth, Quotes) when Quotes ->
erlang:error(unterminated_quotes);
parse_with_comments(<<>>, _Acc, Depth, _Quotes) when Depth > 0 ->
erlang:error(unterminated_comment);
parse_with_comments(<<>>, Acc, _Depth, _Quotes) ->
binstr:strip(list_to_binary(lists:reverse(Acc)));
parse_with_comments(<<$\\, H, Tail/binary>>, Acc, Depth, Quotes) when Depth > 0, H > 32, H < 127 ->
parse_with_comments(Tail, Acc, Depth, Quotes);
parse_with_comments(<<$\\, Tail/binary>>, Acc, Depth, Quotes) when Depth > 0 ->
parse_with_comments(Tail, Acc, Depth, Quotes);
parse_with_comments(<<$\\, H, Tail/binary>>, Acc, Depth, Quotes) when H > 32, H < 127 ->
parse_with_comments(Tail, [H | Acc], Depth, Quotes);
parse_with_comments(<<$\\, Tail/binary>>, Acc, Depth, Quotes) ->
parse_with_comments(Tail, [$\\ | Acc], Depth, Quotes);
parse_with_comments(<<$(, Tail/binary>>, Acc, Depth, Quotes) when not Quotes ->
parse_with_comments(Tail, Acc, Depth + 1, Quotes);
parse_with_comments(<<$), Tail/binary>>, Acc, Depth, Quotes) when Depth > 0, not Quotes ->
parse_with_comments(Tail, Acc, Depth - 1, Quotes);
parse_with_comments(<<_, Tail/binary>>, Acc, Depth, Quotes) when Depth > 0 ->
parse_with_comments(Tail, Acc, Depth, Quotes);
%"
parse_with_comments(<<$", T/binary>>, Acc, Depth, true) ->
parse_with_comments(T, Acc, Depth, false);
%"
parse_with_comments(<<$", T/binary>>, Acc, Depth, false) ->
parse_with_comments(T, Acc, Depth, true);
parse_with_comments(<<H, Tail/binary>>, Acc, Depth, Quotes) ->
parse_with_comments(Tail, [H | Acc], Depth, Quotes).
-spec parse_content_type
(Value :: 'undefined') -> 'undefined';
(Value :: binary()) -> {binary(), binary(), [{binary(), binary()}]}.
parse_content_type(undefined) ->
undefined;
parse_content_type(String) ->
try parse_content_disposition(String) of
{RawType, Parameters} ->
case binstr:strchr(RawType, $/) of
Index when Index < 2 ->
throw(bad_content_type);
Index ->
Type = binstr:substr(RawType, 1, Index - 1),
SubType = binstr:substr(RawType, Index + 1),
{binstr:to_lower(Type), binstr:to_lower(SubType), Parameters}
end
catch
bad_disposition ->
throw(bad_content_type)
end.
-spec parse_content_disposition
(Value :: 'undefined') -> 'undefined';
(String :: binary()) -> {binary(), [{binary(), binary()}]}.
parse_content_disposition(undefined) ->
undefined;
parse_content_disposition(String) ->
[Disposition | Parameters] = binstr:split(parse_with_comments(String), <<";">>),
F =
fun(X) ->
Y = binstr:strip(binstr:strip(X), both, $\t),
case binstr:strchr(Y, $=) of
Index when Index < 2 ->
throw(bad_disposition);
Index ->
Key = binstr:substr(Y, 1, Index - 1),
Value = binstr:substr(Y, Index + 1),
{binstr:to_lower(Key), Value}
end
end,
Params = lists:map(F, Parameters),
{binstr:to_lower(Disposition), Params}.
split_body_by_boundary(Body, Boundary, MimeVsn, Options) ->
% find the indices of the first and last boundary
case {binstr:strpos(Body, Boundary), binstr:strpos(Body, list_to_binary([Boundary, "--"]))} of
{0, _} ->
erlang:error(missing_boundary);
{_, 0} ->
erlang:error(missing_last_boundary);
{Start, End} ->
NewBody = binstr:substr(Body, Start + byte_size(Boundary), End - Start),
% from now on, we can be sure that each boundary is preceded by a CRLF
Parts = split_body_by_boundary_(NewBody, list_to_binary(["\r\n", Boundary]), [], Options),
[
decode_component(Headers, Body2, MimeVsn, Options)
|| {Headers, Body2} <- [V || {_, Body3} = V <- Parts, byte_size(Body3) =/= 0]
]
end.
split_body_by_boundary_(<<>>, _Boundary, Acc, _Options) ->
lists:reverse(Acc);
split_body_by_boundary_(Body, Boundary, Acc, Options) ->
% trim the incomplete first line
TrimmedBody = binstr:substr(Body, binstr:strpos(Body, "\r\n") + 2),
case binstr:strpos(TrimmedBody, Boundary) of
0 ->
lists:reverse([{[], TrimmedBody} | Acc]);
Index ->
{ParsedHdrs, BodyRest} = parse_headers(binstr:substr(TrimmedBody, 1, Index - 1)),
DecodedHdrs = decode_headers(ParsedHdrs, [], proplists:get_value(encoding, Options, none)),
split_body_by_boundary_(
binstr:substr(TrimmedBody, Index + byte_size(Boundary)),
Boundary,
[{DecodedHdrs, BodyRest} | Acc],
Options
)
end.
-spec parse_headers(Body :: binary()) -> {[{binary(), binary()}], binary()}.
%% @doc Parse the headers off of a message and return a list of headers and the trailing body.
parse_headers(Body) ->
case binstr:strpos(Body, "\r\n") of
0 ->
{[], Body};
1 ->
{[], binstr:substr(Body, 3)};
Index ->
parse_headers(binstr:substr(Body, Index + 2), binstr:substr(Body, 1, Index - 1), [])
end.
parse_headers(Body, <<H, Tail/binary>>, []) when H =:= $\s; H =:= $\t ->
% folded headers
{[], list_to_binary([H, Tail, "\r\n", Body])};
parse_headers(Body, <<H, T/binary>>, Headers) when H =:= $\s; H =:= $\t ->
% folded headers
[{FieldName, OldFieldValue} | OtherHeaders] = Headers,
FieldValue = list_to_binary([OldFieldValue, T]),
?LOG_DEBUG("~p = ~p", [FieldName, FieldValue], ?LOGGER_META),
case binstr:strpos(Body, "\r\n") of
0 ->
{lists:reverse([{FieldName, FieldValue} | OtherHeaders]), Body};
1 ->
{lists:reverse([{FieldName, FieldValue} | OtherHeaders]), binstr:substr(Body, 3)};
Index2 ->
parse_headers(binstr:substr(Body, Index2 + 2), binstr:substr(Body, 1, Index2 - 1), [
{FieldName, FieldValue} | OtherHeaders
])
end;
parse_headers(Body, Line, Headers) ->
?LOG_DEBUG("line: ~p", [Line], ?LOGGER_META),
case binstr:strchr(Line, $:) of
0 ->
{lists:reverse(Headers), list_to_binary([Line, "\r\n", Body])};
Index ->
FieldName = binstr:substr(Line, 1, Index - 1),
F = fun(X) -> X > 32 andalso X < 127 end,
case binstr:all(F, FieldName) of
true ->
F2 = fun(X) -> (X > 31 andalso X < 127) orelse X == 9 end,
FValue = binstr:strip(binstr:substr(Line, Index + 1)),
FieldValue =
case binstr:all(F2, FValue) of
true ->
FValue;
_ ->
% I couldn't figure out how to use a pure binary comprehension here :(
list_to_binary([filter_non_ascii(C) || <<C:8>> <= FValue])
end,
case binstr:strpos(Body, "\r\n") of
0 ->
{lists:reverse([{FieldName, FieldValue} | Headers]), Body};
1 ->
{lists:reverse([{FieldName, FieldValue} | Headers]), binstr:substr(Body, 3)};
Index2 ->
parse_headers(binstr:substr(Body, Index2 + 2), binstr:substr(Body, 1, Index2 - 1), [
{FieldName, FieldValue} | Headers
])
end;
false ->
{lists:reverse(Headers), list_to_binary([Line, "\r\n", Body])}
end
end.
filter_non_ascii(C) when (C > 31 andalso C < 127); C == 9 ->
<<C>>;
filter_non_ascii(_C) ->
<<"?">>.
decode_body(Type, Body, _InEncoding, none) ->
decode_body(Type, <<<<X/integer>> || <<X>> <= Body, X < 128>>);
decode_body(Type, Body, undefined, _OutEncoding) ->
decode_body(Type, <<<<X/integer>> || <<X>> <= Body, X < 128>>);
decode_body(Type, Body, <<"x-binaryenc">>, _OutEncoding) ->
% Not IANA and does not represent text, so we pass it through
decode_body(Type, Body);
decode_body(Type, Body, InEncoding, OutEncoding) ->
NewBody = decode_body(Type, Body),
InEncodingFixed = fix_encoding(InEncoding),
{ok, ConvertedBody} = convert(OutEncoding, InEncodingFixed, NewBody),
ConvertedBody.
-spec decode_body(Type :: binary() | 'undefined', Body :: binary()) -> binary().
decode_body(undefined, Body) ->
Body;
decode_body(Type, Body) ->
case binstr:to_lower(Type) of
<<"quoted-printable">> ->
decode_quoted_printable(Body);
<<"base64">> ->
decode_base64(Body);
_Other ->
Body
end.
decode_base64(Body) ->
base64:mime_decode(Body).
decode_quoted_printable(Body) ->
decode_quoted_printable(Body, false, <<>>, <<>>).
%% End of Body
decode_quoted_printable(<<>>, _HasSoftEOL, _WSPs, Acc) ->
Acc;
%% CRLF after Soft Linebreak
decode_quoted_printable(<<$\r, $\n, More/binary>>, true, _WSPs, Acc) ->
decode_quoted_printable(More, false, <<>>, Acc);
%% Space or Tab after Soft Linebreak
decode_quoted_printable(<<C, More/binary>>, true, _WSPs, Acc) when C =:= $\s; C =:= $\t ->
decode_quoted_printable(More, true, <<>>, Acc);
%% Other character after Soft Linebreak
decode_quoted_printable(_Body, true, _WSPs, _Acc) ->
throw(badchar);
%% CRLF
decode_quoted_printable(<<$\r, $\n, More/binary>>, false, _WSPs, Acc) ->
decode_quoted_printable(More, false, <<>>, <<Acc/binary, $\r, $\n>>);
%% Space or Tab
decode_quoted_printable(<<C, More/binary>>, false, WSPs, Acc) when C =:= $\s; C =:= $\t ->
decode_quoted_printable(More, false, <<WSPs/binary, C>>, Acc);
%% Encoded char
decode_quoted_printable(<<$=, C1, C2, More/binary>>, false, WSPs, Acc) when
C1 >= $0 andalso C1 =< $9 orelse C1 >= $A andalso C1 =< $F orelse C1 >= $a andalso C1 =< $f,
C2 >= $0 andalso C2 =< $9 orelse C2 >= $A andalso C2 =< $F orelse C2 >= $a andalso C2 =< $f
->
decode_quoted_printable(More, false, <<>>, <<Acc/binary, WSPs/binary, (unhex(C1)):4, (unhex(C2)):4>>);
%% Soft Linebreak
decode_quoted_printable(<<$=, More/binary>>, false, WSPs, Acc) ->
decode_quoted_printable(More, true, <<>>, <<Acc/binary, WSPs/binary>>);
%% Plain character
decode_quoted_printable(<<C, More/binary>>, false, WSPs, Acc) ->
decode_quoted_printable(More, false, <<>>, <<Acc/binary, WSPs/binary, C>>).
check_headers(Headers) ->
Checked = [<<"MIME-Version">>, <<"Date">>, <<"From">>, <<"Message-ID">>, <<"References">>, <<"Subject">>],
check_headers(Checked, lists:reverse(Headers)).
check_headers([], Headers) ->
lists:reverse(Headers);
check_headers([Header | Tail], Headers) ->
case get_header_value(Header, Headers) of
undefined when Header == <<"MIME-Version">> ->
check_headers(Tail, [{<<"MIME-Version">>, <<"1.0">>} | Headers]);
undefined when Header == <<"Date">> ->
check_headers(Tail, [{<<"Date">>, list_to_binary(smtp_util:rfc5322_timestamp())} | Headers]);
undefined when Header == <<"From">> ->
erlang:error(missing_from);
undefined when Header == <<"Message-ID">> ->
check_headers(Tail, [{<<"Message-ID">>, list_to_binary(smtp_util:generate_message_id())} | Headers]);
undefined when Header == <<"References">> ->
case get_header_value(<<"In-Reply-To">>, Headers) of
undefined ->
% ok, whatever
check_headers(Tail, Headers);
ReplyID ->
check_headers(Tail, [{<<"References">>, ReplyID} | Headers])
end;
References when Header == <<"References">> ->
% check if the in-reply-to header, if present, is in references
case get_header_value(<<"In-Reply-To">>, Headers) of
undefined ->
% ok, whatever
check_headers(Tail, Headers);
ReplyID ->
case binstr:strpos(binstr:to_lower(References), binstr:to_lower(ReplyID)) of
0 ->
% okay, tack on the reply-to to the end of References
check_headers(Tail, [
{<<"References">>, list_to_binary([References, " ", ReplyID])}
| proplists:delete(<<"References">>, Headers)
]);
_Index ->
% nothing to do
check_headers(Tail, Headers)
end
end;
_ ->
check_headers(Tail, Headers)
end.
ensure_content_headers(Type, SubType, Parameters, Headers, Body, Toplevel) ->
CheckHeaders = [<<"Content-Type">>, <<"Content-Disposition">>, <<"Content-Transfer-Encoding">>],
CheckHeadersValues = [{Name, get_header_value(Name, Headers)} || Name <- CheckHeaders],
ensure_content_headers(CheckHeadersValues, Type, SubType, Parameters, lists:reverse(Headers), Body, Toplevel).
ensure_content_headers([], _, _, Parameters, Headers, _, _) ->
{Parameters, lists:reverse(Headers)};
ensure_content_headers(
[{<<"Content-Type">>, undefined} | Tail], Type, SubType, Parameters, Headers, Body, Toplevel
) when
(Type == <<"text">> andalso SubType =/= <<"plain">>) orelse Type =/= <<"text">>
->
%% no content-type header, and its not text/plain
CT = io_lib:format("~s/~s", [Type, SubType]),
CTP =
case Type of
<<"multipart">> ->
Boundary =
case proplists:get_value(<<"boundary">>, maps:get(content_type_params, Parameters, [])) of
undefined ->
list_to_binary(smtp_util:generate_message_boundary());
B ->
B
end,
[
{<<"boundary">>, Boundary}
| proplists:delete(<<"boundary">>, maps:get(content_type_params, Parameters, []))
];
<<"text">> ->
Charset =
case proplists:get_value(<<"charset">>, maps:get(content_type_params, Parameters, [])) of
undefined ->
guess_charset(Body);
C ->
C
end,
[
{<<"charset">>, Charset}
| proplists:delete(<<"charset">>, maps:get(content_type_params, Parameters, []))
];
_ ->
maps:get(content_type_params, Parameters, [])
end,
%%CTP = proplists:get_value(<<"content-type-params">>, Parameters, [guess_charset(Body)]),
CTH = binstr:join([CT | encode_parameters(CTP)], ";"),
NewParameters = Parameters#{content_type_params => CTP},
ensure_content_headers(Tail, Type, SubType, NewParameters, [{<<"Content-Type">>, CTH} | Headers], Body, Toplevel);
ensure_content_headers(
[{<<"Content-Type">>, undefined} | Tail],
<<"text">> = Type,
<<"plain">> = SubType,
Parameters,
Headers,
Body,
Toplevel
) ->
%% no content-type header and its text/plain
Charset =
case proplists:get_value(<<"charset">>, maps:get(content_type_params, Parameters, [])) of
undefined ->
guess_charset(Body);
C ->
binstr:to_lower(C)
end,
case Charset of
<<"us-ascii">> ->
% the default
ensure_content_headers(Tail, Type, SubType, Parameters, Headers, Body, Toplevel);
_ ->
CTP = [
{<<"charset">>, Charset}
| proplists:delete(<<"charset">>, maps:get(content_type_params, Parameters, []))
],
CTH = binstr:join([<<"text/plain">> | encode_parameters(CTP)], ";"),
NewParameters = Parameters#{content_type_params => CTP},
ensure_content_headers(
Tail, Type, SubType, NewParameters, [{<<"Content-Type">>, CTH} | Headers], Body, Toplevel
)
end;
ensure_content_headers(
[{<<"Content-Transfer-Encoding">>, undefined} | Tail], Type, SubType, Parameters, Headers, Body, Toplevel
) when
Type =/= <<"multipart">>
->
Enc =
case maps:get(transfer_encoding, Parameters, undefined) of
undefined ->
guess_best_encoding(Body);
Value ->
Value
end,
case Enc of
<<"7bit">> ->
ensure_content_headers(Tail, Type, SubType, Parameters, Headers, Body, Toplevel);
_ ->
ensure_content_headers(
Tail, Type, SubType, Parameters, [{<<"Content-Transfer-Encoding">>, Enc} | Headers], Body, Toplevel
)
end;
ensure_content_headers(
[{<<"Content-Disposition">>, undefined} | Tail], Type, SubType, Parameters, Headers, Body, false = Toplevel
) ->
CD = maps:get(disposition, Parameters, <<"inline">>),
CDP = maps:get(disposition_params, Parameters, []),
CDH = binstr:join([CD | encode_parameters(CDP)], ";"),
ensure_content_headers(
Tail, Type, SubType, Parameters, [{<<"Content-Disposition">>, CDH} | Headers], Body, Toplevel
);
ensure_content_headers([_ | Tail], Type, SubType, Parameters, Headers, Body, Toplevel) ->
ensure_content_headers(Tail, Type, SubType, Parameters, Headers, Body, Toplevel).
guess_charset(Body) ->
case binstr:all(fun(X) -> X < 128 end, Body) of
true -> <<"us-ascii">>;
false -> <<"utf-8">>
end.
guess_best_encoding(Body) ->
case valid_7bit(Body) of
true ->
<<"7bit">>;
false ->
choose_transformation(Body)
end.
choose_transformation(<<Chunk:200/binary, _, _/binary>>) ->
%% Optimization - only analyze 1st 200 bytes
choose_transformation(Chunk);
choose_transformation(Body) ->
{Readable, Encoded} = partition_count_bytes(
fun(C) ->
C >= 16#20 andalso C =< 16#7E orelse C =:= $\r orelse C =:= $\n
end,
Body
),
%based on the % of printable characters, choose an encoding
if
% same as 100 * Readable / (Readable + Encoded) >= 80, but avoiding division
Readable >= 4 * Encoded ->
%% >80% printable characters
<<"quoted-printable">>;
true ->
%% =<80% printable characters
<<"base64">>
end.
%% https://tools.ietf.org/html/rfc2045#section-2.7:
%% * ASCII codes from 1 to 127
%% * \r and \n are only allowed as `\r\n' pair, but not standalone (bare)
%% * No lines over 998 chars
%%
%% Unfortunately, any string that ends with `\n` matches the regexp, so, we need some pre-checks
valid_7bit(<<"\n">>) ->
false;
valid_7bit(<<"\r">>) ->
false;
valid_7bit(<<>>) ->
true;
valid_7bit(<<_>>) ->
true;
valid_7bit(Body) ->
Size = byte_size(Body),
case binary:at(Body, Size - 1) =:= $\n andalso binary:at(Body, Size - 2) =/= $\r of
true ->
%% last element is \n, but the one before the last is not \r
false;
false ->
%% So: (all except `\r` and `\n` in 1-127 range) OR (`\r\n`)
case re:run(Body, "^([\x01-\x09\x0b-\x0c\x0e-\x7f]|(\r\n))*$", [{capture, none}]) of
match -> not has_lines_over_998(Body);
nomatch -> false
end
end.
%% @doc If `Body' has at least one line (ending with `\r\n') that is longer than 998 chars
has_lines_over_998(Body) ->
Pattern = binary:compile_pattern(<<"\r\n">>),
has_lines_over_998(Body, binary:match(Body, Pattern), 0, Pattern).
has_lines_over_998(Bin, nomatch, Offset, _) ->
%% Last line is over 998?
(byte_size(Bin) - Offset) >= 998;
has_lines_over_998(_Bin, {FoundAt, 2}, Offset, _Patern) when (FoundAt - Offset) >= 998 ->
true;
has_lines_over_998(Bin, {FoundAt, 2}, _, Pattern) ->
NewOffset = FoundAt + 2,
Len = byte_size(Bin) - NewOffset,
has_lines_over_998(
Bin, binary:match(Bin, Pattern, [{scope, {NewOffset, Len}}]), NewOffset, Pattern
).
encode_parameters([[]]) ->
[];
encode_parameters(Parameters) ->
[encode_parameter(Parameter) || Parameter <- Parameters].
encode_parameter({X, Y}) ->
YEnc = rfc2047_utf8_encode(Y, byte_size(X) + 3, <<"\t">>),
case escape_tspecial(YEnc, false, <<>>) of
{true, Special} -> [X, $=, $", Special, $"];
false -> [X, $=, YEnc]
end.
% See also: http://www.ietf.org/rfc/rfc2045.txt section 5.1
escape_tspecial(<<>>, false, _Acc) ->
false;
escape_tspecial(<<>>, IsSpecial, Acc) ->
{IsSpecial, Acc};
escape_tspecial(<<C, Rest/binary>>, _IsSpecial, Acc) when C =:= $" ->
escape_tspecial(Rest, true, <<Acc/binary, $\\, $">>);
escape_tspecial(<<C, Rest/binary>>, _IsSpecial, Acc) when C =:= $\\ ->
escape_tspecial(Rest, true, <<Acc/binary, $\\, $\\>>);
escape_tspecial(<<C, Rest/binary>>, _IsSpecial, Acc) when
C =:= $(;
C =:= $);
C =:= $<;
C =:= $>;
C =:= $@;
C =:= $,;
C =:= $;;
C =:= $:;
C =:= $/;
C =:= $[;
C =:= $];
C =:= $?;
C =:= $=;
C =:= $\s
->
escape_tspecial(Rest, true, <<Acc/binary, C>>);
escape_tspecial(<<C, Rest/binary>>, IsSpecial, Acc) ->
escape_tspecial(Rest, IsSpecial, <<Acc/binary, C>>).
encode_headers([]) ->
[];
encode_headers([{Key, Value} | T] = _Headers) ->
EncodedHeader = maybe_encode_folded_header(Key, list_to_binary([Key, ": ", encode_header_value(Key, Value)])),
[EncodedHeader | encode_headers(T)].
maybe_encode_folded_header(H, Hdr) when
H =:= <<"To">>;
H =:= <<"Cc">>;
H =:= <<"Bcc">>;
H =:= <<"Reply-To">>;
H =:= <<"From">>
->
Hdr;
maybe_encode_folded_header(_H, Hdr) ->
encode_folded_header(Hdr, <<>>).
encode_folded_header(Rest, Acc) ->
case binstr:split(Rest, <<$;>>, 2) of
[_] ->
<<Acc/binary, Rest/binary>>;
[Before, After] ->
NewPart =
case After of
<<$\t, _Rest/binary>> ->
<<Before/binary, ";\r\n">>;
_ ->
<<Before/binary, ";\r\n\t">>
end,
encode_folded_header(After, <<Acc/binary, NewPart/binary>>)
end.
encode_header_value(H, Value) when
H =:= <<"To">>;
H =:= <<"Cc">>;
H =:= <<"Bcc">>;
H =:= <<"Reply-To">>;
H =:= <<"From">>
->
{ok, Addresses} = smtp_util:parse_rfc5322_addresses(Value),
{Names, Emails} = lists:unzip(Addresses),
NewNames = lists:map(
fun
(undefined) ->
undefined;
(Name) ->
%% `Name' contains codepoints, but we need bytes
rfc2047_utf8_encode(unicode:characters_to_binary(Name))
end,
Names
),
smtp_util:combine_rfc822_addresses(lists:zip(NewNames, Emails));
encode_header_value(H, Value) when H =:= <<"Content-Type">>; H =:= <<"Content-Disposition">> ->
% Parameters are already encoded.
Value;
encode_header_value(_, Value) ->
rfc2047_utf8_encode(Value).
encode_component(_Type, _SubType, _Headers, Params, Body) when is_list(Body) ->
% is this a multipart component?
Boundary = proplists:get_value(<<"boundary">>, maps:get(content_type_params, Params)),
% blank line before start of component
[<<>>] ++
lists:flatmap(
fun(Part) ->
% start with the boundary
[list_to_binary([<<"--">>, Boundary])] ++
encode_component_part(Part)
end,
Body
% final boundary (with /--$/)
) ++ [list_to_binary([<<"--">>, Boundary, <<"--">>])] ++
% blank line at the end of the multipart component
[<<>>];
encode_component(_Type, _SubType, Headers, _Params, Body) ->
% or an inline component?
%encode_component_part({Type, SubType, Headers, Params, Body})
encode_body(
get_header_value(<<"Content-Transfer-Encoding">>, Headers),
[Body]
).
encode_component_part({<<"multipart">>, SubType, Headers, PartParams, Body}) ->
{FixedParams, FixedHeaders} = ensure_content_headers(<<"multipart">>, SubType, PartParams, Headers, Body, false),
encode_headers(FixedHeaders) ++
encode_component(<<"multipart">>, SubType, FixedHeaders, FixedParams, Body);
encode_component_part({Type, SubType, Headers, PartParams, Body}) ->
PartData =
case Body of
{_, _, _, _, _} -> encode_component_part(Body);
String -> [String]
end,
{_FixedParams, FixedHeaders} = ensure_content_headers(Type, SubType, PartParams, Headers, Body, false),
encode_headers(FixedHeaders) ++ [<<>>] ++
encode_body(
get_header_value(<<"Content-Transfer-Encoding">>, FixedHeaders),
PartData
);
encode_component_part(Part) ->
?LOG_DEBUG("encode_component_part couldn't match Part to: ~p", [Part], ?LOGGER_META),
[].
encode_body(undefined, Body) ->
Body;
encode_body(Type, Body) ->
case binstr:to_lower(Type) of
<<"quoted-printable">> ->
[InnerBody] = Body,
encode_quoted_printable(InnerBody);
<<"base64">> ->
[InnerBody] = Body,
wrap_to_76(base64:encode(InnerBody));
_ ->
Body
end.
wrap_to_76(String) ->
[wrap_to_76(String, [])].
wrap_to_76(<<>>, Acc) ->
list_to_binary(lists:reverse(Acc));
wrap_to_76(<<Head:76/binary, Tail/binary>>, Acc) ->
wrap_to_76(Tail, [<<"\r\n">>, Head | Acc]);
wrap_to_76(Head, Acc) ->
list_to_binary(lists:reverse([<<"\r\n">>, Head | Acc])).
encode_quoted_printable(Body) ->
[encode_quoted_printable(Body, <<>>, 0, false, <<>>, 0)].
% End of body (this should only happen if the body was empty to begin with)
encode_quoted_printable(<<>>, Acc, _LineLen, _HasWSP, WordAcc, _WordLen) ->
<<Acc/binary, WordAcc/binary>>;
% CRLF
encode_quoted_printable(<<$\r, $\n, More/binary>>, Acc, _LineLen, _HasWSP, WordAcc, _WordLen) ->
encode_quoted_printable(More, <<Acc/binary, WordAcc/binary, $\r, $\n>>, 0, false, <<>>, 0);
% WSP in last position
encode_quoted_printable(<<C>>, Acc, LineLen, _HasWSP, WordAcc, WordLen) when C =:= $\s; C =:= $\t ->
Enc = encode_quoted_printable_char(C, true),
case LineLen + WordLen + 3 > 76 of
true ->
% line would become too long -> soft-break before WSP
<<Acc/binary, WordAcc/binary, $=, $\r, $\n, Enc/binary>>;
false ->
% character fits on current line
<<Acc/binary, WordAcc/binary, Enc/binary>>
end;
% WSP before CRLF
encode_quoted_printable(<<C, $\r, $\n, More/binary>>, Acc, LineLen, _HasWSP, WordAcc, WordLen) when
C =:= $\s; C =:= $\t
->
Enc = encode_quoted_printable_char(C, true),
case LineLen + WordLen + 3 > 76 of
true ->
% line would become too long -> soft-break before WSP
encode_quoted_printable(
More, <<Acc/binary, WordAcc/binary, $=, $\r, $\n, Enc/binary, $\r, $\n>>, 0, false, <<>>, 0
);
false ->
% character fits on current line
encode_quoted_printable(More, <<Acc/binary, WordAcc/binary, Enc/binary, $\r, $\n>>, 0, false, <<>>, 0)
end;
% Character elsewhere
encode_quoted_printable(<<C, More/binary>>, Acc, LineLen, HasWSP, WordAcc, WordLen) ->
Enc = encode_quoted_printable_char(C, false),
EncLen = byte_size(Enc),
% mind the 75 here, we need the 76th place for the soft linebreak
case LineLen + WordLen + EncLen > 75 of
true when C =:= $\s; C =:= $\t ->
% line would become too long, current char is WSP -> soft-break here (remember we have a WSP)
encode_quoted_printable(
More, <<Acc/binary, WordAcc/binary, $=, $\r, $\n, Enc/binary>>, EncLen, true, <<>>, 0
);
true when HasWSP, WordLen + EncLen =< 75 ->
% line would become too long, we have an earlier WSP and word plus encoded character will fit on a new line -> soft-break at earlier WSP
encode_quoted_printable(
More, <<Acc/binary, $=, $\r, $\n, WordAcc/binary, Enc/binary>>, WordLen + EncLen, false, <<>>, 0
);
true ->
% line would become too long, we have no earlier WSP or word plus encoded character will not fit on a new line -> soft break here
encode_quoted_printable(
More, <<Acc/binary, WordAcc/binary, $=, $\r, $\n, Enc/binary>>, EncLen, false, <<>>, 0
);
false when C =:= $\s; C =:= $\t ->
% WSP character fits on line -> move word and WSP to Acc (remember we have a WSP)
encode_quoted_printable(
More, <<Acc/binary, WordAcc/binary, Enc/binary>>, LineLen + WordLen + EncLen, true, <<>>, 0
);
false ->
% non-WSP character fits on line -> add character to word
encode_quoted_printable(More, Acc, LineLen, HasWSP, <<WordAcc/binary, Enc/binary>>, WordLen + EncLen)
end.
encode_quoted_printable_char(C, true) ->
<<$=, (hex(C div 16#10)), (hex(C rem 16#10))>>;
encode_quoted_printable_char($\s, false) ->
<<$\s>>;
encode_quoted_printable_char($\t, false) ->
<<$\t>>;
encode_quoted_printable_char($=, _Force) ->
<<$=, $3, $D>>;
encode_quoted_printable_char(C, _Force) when C =< 16#20; C >= 16#7F ->
encode_quoted_printable_char(C, true);
encode_quoted_printable_char(C, false) ->
<<C>>.
get_default_encoding() ->
<<"utf-8//IGNORE">>.
% convert some common invalid character names into the correct ones
fix_encoding(Encoding) when Encoding == <<"utf8">>; Encoding == <<"UTF8">> ->
<<"UTF-8">>;
fix_encoding(Encoding) ->
Encoding.
% Characters allowed to appear unencoded (RFC 2047 Sections 4.2 and 5):
% * lowercase ASCII letters
% * uppercase ASCII letters
% * decimal digits
% * "!"
% * "*"
% * "+"
% * "-"
% * "/"
% SPACE is not really an allowed letter, but since it encodes to "_"
% and thereby a single byte, we list it as allowed here
-define(is_rfc2047_q_allowed(C),
(C =:= $\s orelse (C >= $a andalso C =< $z) orelse (C >= $A andalso C =< $Z) orelse
(C >= $0 andalso C =< $9) orelse C =:= $! orelse C =:= $* orelse C =:= $+ orelse
C =:= $- orelse C =:= $/)
).
%% @doc Encode a binary or list according to RFC 2047. Input is
%% assumed to be in UTF-8 encoding bytes; not codepoints.
rfc2047_utf8_encode(Value) ->
rfc2047_utf8_encode(Value, 0, <<" ">>).
rfc2047_utf8_encode(Value, PrefixLen, LineIndent) when is_binary(Value) ->
case is_ascii_printable(Value) of
true ->
% don't encode if all characters are printable ASCII
Value;
false ->
{Readable, Encoded} = partition_count_bytes(fun(C) -> ?is_rfc2047_q_allowed(C) end, Value),
Enc =
if
Readable >= Encoded ->
% most characters would be readable in Q-Encoding,
% so we use it
q;
true ->
% most characters would have to be encoded in Q-Encoding,
% so we use B-Encoding instead
b
end,
rfc2047_utf8_encode(Enc, Value, <<>>, PrefixLen, LineIndent)
end;
rfc2047_utf8_encode(Value, PrefixLen, LineIndent) ->
rfc2047_utf8_encode(list_to_binary(Value), PrefixLen, LineIndent).
rfc2047_utf8_encode(_Enc, <<>>, Acc, _PrefixLen, _LineIndent) ->
Acc;
rfc2047_utf8_encode(b, More, Acc, PrefixLen, LineIndent) ->
% B-Encoding
% An encoded word must not be longer than 75 bytes,
% including the leading "=?", charset name, "?B?" and
% the trailing "?=". Since the charset name is fixed to
% "UTF-8", 63 remain for encoded text. Using Base64,
% a maximum of 45 raw bytes can be encoded in 63 bytes.
rfc2047_utf8_encode(b, More, Acc, <<>>, byte_size(LineIndent), LineIndent, 46 - PrefixLen);
rfc2047_utf8_encode(q, More, Acc, PrefixLen, LineIndent) ->
% Q-Encoding
% An encoded word must not be longer than 75 bytes,
% including the leading "=?", charset name, "=?UTF-8?Q?" and
% the trailing "?=". Since the charset name is fixed to
% "UTF-8", 63 remain for encoded text. Using Quoted-Printable,
% between 21 and 63 raw bytes can be encoded in 63 bytes.
rfc2047_utf8_encode(q, More, Acc, <<>>, byte_size(LineIndent), LineIndent, 63 - PrefixLen).
rfc2047_utf8_encode(Enc, <<>>, Acc, WordAcc, _PrefixLen, LineIndent, _Left) ->
rfc2047_append_word(Acc, WordAcc, Enc, LineIndent);
rfc2047_utf8_encode(Enc, All = <<C/utf8, More/binary>>, Acc, WordAcc, PrefixLen, LineIndent, Left) ->
% convert codepoint back to UTF-8 encoded bytes
Bytes = <<C/utf8>>,
Size = byte_size(Bytes),
Reqd =
case Enc of
q when not ?is_rfc2047_q_allowed(C) ->
3 * Size;
q ->
Size;
b ->
Size
end,
case Left >= Reqd of
true ->
rfc2047_utf8_encode(Enc, More, Acc, <<WordAcc/binary, Bytes/binary>>, PrefixLen, LineIndent, Left - Reqd);
false ->
rfc2047_utf8_encode(Enc, All, rfc2047_append_word(Acc, WordAcc, Enc, LineIndent), PrefixLen, LineIndent)
end.
rfc2047_append_word(Acc, <<>>, _Enc, _LineIndent) ->
% empty word
Acc;
rfc2047_append_word(<<>>, Word, Enc, _LineIndent) ->
% first word in Acc
rfc2047_encode_word(Word, Enc);
rfc2047_append_word(Acc, Word, Enc, LineIndent) ->
% subsequent word in Acc
<<Acc/binary, $\r, $\n, LineIndent/binary, (rfc2047_encode_word(Word, Enc))/binary>>.
rfc2047_encode_word(Word, q) ->
<<"=?UTF-8?Q?", (rfc2047_q_encode(Word))/binary, "?=">>;
rfc2047_encode_word(Word, b) ->
<<"=?UTF-8?B?", (base64:encode(Word))/binary, "?=">>.
rfc2047_q_encode(<<>>) ->
<<>>;
rfc2047_q_encode(<<$\s, More/binary>>) ->
% SPACE -> _
<<$_, (rfc2047_q_encode(More))/binary>>;
rfc2047_q_encode(<<C, More/binary>>) when ?is_rfc2047_q_allowed(C) ->
% character which needs no encoding
<<C, (rfc2047_q_encode(More))/binary>>;
rfc2047_q_encode(<<N1:4, N2:4, More/binary>>) ->
% characters which need encoding -> =XY
<<$=, (hex(N1)), (hex(N2)), (rfc2047_q_encode(More))/binary>>.
is_ascii_printable(<<>>) ->
'true';
is_ascii_printable(<<H, T/binary>>) when H >= 32 andalso H =< 126 ->
is_ascii_printable(T);
is_ascii_printable(_) ->
'false'.
hex(N) when N >= 10 -> N + $A - 10;
hex(N) -> N + $0.
unhex(C) when C >= $a -> C - $a + 10;
unhex(C) when C >= $A -> C - $A + 10;
unhex(C) -> C - $0.
partition_count_bytes(Fun, Bin) ->
partition_count_bytes(Fun, Bin, {0, 0}).
partition_count_bytes(_Fun, <<>>, PartitionCounts) ->
PartitionCounts;
partition_count_bytes(Fun, <<C, More/binary>>, {Trues, Falses}) ->
NewPartitionCounts =
case Fun(C) of
true -> {Trues + 1, Falses};
false -> {Trues, Falses + 1}
end,
partition_count_bytes(Fun, More, NewPartitionCounts).
%% @doc DKIM sign an email
%% DKIM sign functions
%% RFC 6376
%% `h' - list of headers to sign (lowercased binary)
%% `c' - {Headers, Body} canonicalization type. Only {simple, simple} and
%% {relaxed, simple} supported for now.
%% be located in "foo.bar._domainkey.example.com" (see RFC-6376 #3.6.2.1).
%% `t' - signature timestamp: 'now' or UTC {Date, Time}
%% `x' - signature expiration time: UTC {Date, Time}
%% `a` - signing algorithm (default: `rsa-sha256`):
%% `private_key' - private key, to sign emails. May be of 2 types: encrypted and
%% plain in PEM format:
%% RSA
%% `{pem_plain, KeyBinary}' - generated by <code>openssl genrsa -out out-file.pem 1024</code>
%% `{pem_encrypted, KeyBinary, Password}' - generated by, eg
%% <code>openssl genrsa -des3 -out out-file.pem 1024</code>
%% RFC8463
%% Ed25519 - Erlang/OTP 24.1+ only!
%% `{pem_plain, KeyBinary}' - generated by <code>openssl genpkey -algorithm ed25519 -out out-file.pem</code>
%% `{pem_encrypted, KeyBinary, Password}' - generated by, eg
%% <code>openssl genpkey -des3 -algorithm ed25519 -out out-file.pem</code>
%% 3rd paramerter is password to decrypt the key.
-spec dkim_sign_email([binary()], binary(), dkim_options()) -> [binary()].
dkim_sign_email(Headers, Body, Opts) ->
HeadersToSign = proplists:get_value(h, Opts, [<<"from">>, <<"to">>, <<"subject">>, <<"date">>]),
SDID = proplists:get_value(d, Opts),
Selector = proplists:get_value(s, Opts),
%% BodyLength = proplists:get_value(l, Opts),
OptionalTags = lists:foldl(
fun(Key, Acc) ->
case proplists:get_value(Key, Opts) of
undefined -> Acc;
Value -> [{Key, Value} | Acc]
end
end,
[],
[t, x]
),
{HdrsCanT, BodyCanT} = Can = proplists:get_value(c, Opts, {relaxed, simple}),
Algorithm = proplists:get_value(a, Opts, 'rsa-sha256'),
PrivateKey = proplists:get_value(private_key, Opts),
%% hash body
CanBody = dkim_canonicalize_body(Body, BodyCanT),
BodyHash = dkim_hash_body(CanBody),
%% {b, <<>>},
Tags = [
{v, 1},
{a, Algorithm},
{bh, BodyHash},
{c, Can},
{d, SDID},
{h, HeadersToSign},
{s, Selector}
| OptionalTags
],
%% hash headers
Headers1 = dkim_filter_headers(Headers, HeadersToSign),
CanHeaders = dkim_canonicalize_headers(Headers1, HdrsCanT),
[DkimHeaderNoB] = dkim_canonicalize_headers([dkim_make_header([{b, undefined} | Tags])], HdrsCanT),
DataHash = dkim_hash_data(CanHeaders, DkimHeaderNoB),
%% sign
Signature = dkim_sign(DataHash, Algorithm, PrivateKey),
DkimHeader = dkim_make_header([{b, Signature} | Tags]),
[DkimHeader | Headers].
dkim_filter_headers(Headers, HeadersToSign) ->
KeyedHeaders = [
begin
[Name, _] = binary:split(Hdr, <<":">>),
{binstr:strip(binstr:to_lower(Name)), Hdr}
end
|| Hdr <- Headers
],
WithUndef = [get_header_value(binstr:to_lower(Name), KeyedHeaders) || Name <- HeadersToSign],
[Hdr || Hdr <- WithUndef, Hdr =/= undefined].
dkim_canonicalize_headers(Headers, simple) ->
Headers;
dkim_canonicalize_headers(Headers, relaxed) ->
dkim_canonic_hdrs_relaxed(Headers).
dkim_canonic_hdrs_relaxed([Hdr | Rest]) ->
[Name, Value] = binary:split(Hdr, <<":">>),
LowStripName = binstr:to_lower(binstr:strip(Name)),
UnfoldedHdrValue = binary:replace(Value, <<"\r\n">>, <<>>, [global]),
SingleWSValue = re:replace(UnfoldedHdrValue, "[\t ]+", " ", [global, {return, binary}]),
StrippedWithName = <<LowStripName/binary, ":", (binstr:strip(SingleWSValue))/binary>>,
[StrippedWithName | dkim_canonic_hdrs_relaxed(Rest)];
dkim_canonic_hdrs_relaxed([]) ->
[].
dkim_canonicalize_body(<<>>, simple) ->
<<"\r\n">>;
dkim_canonicalize_body(Body, simple) ->
re:replace(Body, "(\r\n)*$", "\r\n", [{return, binary}]);
dkim_canonicalize_body(_Body, relaxed) ->
throw({not_supported, dkim_body_relaxed}).
dkim_hash_body(CanonicBody) ->
crypto:hash(sha256, CanonicBody).
%% crypto:sha256(CanonicBody).
%% RFC 5.5 & 3.7
dkim_hash_data(CanonicHeaders, DkimHeader) ->
JoinedHeaders = <<<<Hdr/binary, "\r\n">> || Hdr <- CanonicHeaders>>,
crypto:hash(sha256, <<JoinedHeaders/binary, DkimHeader/binary>>).
%% TODO: Remove once we require Erlang/OTP 24.1+
%% Related Erlang/OTP bug: https://github.com/erlang/otp/pull/5157
ed25519_supported() ->
{ok, PublicKeyAppVersionString} = application:get_key(public_key, vsn),
PublicKeyAppVersionList =
lists:map(fun erlang:list_to_integer/1, string:tokens(PublicKeyAppVersionString, ".")),
PublicKeyAppVersionList >= [1, 11, 2].
dkim_get_algorithm_digest(Algorithm) ->
case Algorithm of
'rsa-sha256' ->
sha256;
'ed25519-sha256' ->
case ed25519_supported() of
true ->
none;
false ->
throw("DKIM with Ed25519 requires Erlang/OTP 24.1+")
end
end.
dkim_sign(DataHash, Algorithm, {pem_plain, PrivBin}) ->
[PrivEntry] = public_key:pem_decode(PrivBin),
Digest = dkim_get_algorithm_digest(Algorithm),
Key = public_key:pem_entry_decode(PrivEntry),
public_key:sign({digest, DataHash}, Digest, Key);
dkim_sign(DataHash, Algorithm, {pem_encrypted, EncPrivBin, Passwd}) ->
[EncPrivEntry] = public_key:pem_decode(EncPrivBin),
Digest = dkim_get_algorithm_digest(Algorithm),
Key = public_key:pem_entry_decode(EncPrivEntry, Passwd),
public_key:sign({digest, DataHash}, Digest, Key).
dkim_make_header(Tags) ->
%so {b, ...} became last tag
RevTags = lists:reverse(Tags),
EncodedTags = binstr:join([dkim_encode_tag(K, V) || {K, V} <- RevTags], <<"; ">>),
binstr:join(encode_headers([{<<"DKIM-Signature">>, EncodedTags}]), <<"\r\n">>).
%% RFC #3.5
dkim_encode_tag(v, 1) ->
%% version
<<"v=1">>;
dkim_encode_tag(a, Algorithm) ->
%% algorithm
<<"a=", (atom_to_binary(Algorithm, utf8))/binary>>;
dkim_encode_tag(b, undefined) ->
%% signature (when hashing with no digest)
<<"b=">>;
dkim_encode_tag(b, V) ->
%% signature
B64Sign = base64:encode(V),
<<"b=", B64Sign/binary>>;
dkim_encode_tag(bh, V) ->
%% body hash
B64Sign = base64:encode(V),
<<"bh=", B64Sign/binary>>;
% 'relaxed' for body not supported yet
dkim_encode_tag(c, {Hdrs, simple}) ->
%% canonicalization type
<<"c=", (atom_to_binary(Hdrs, utf8))/binary, "/simple">>;
dkim_encode_tag(d, Domain) ->
%% SDID (domain)
<<"d=", Domain/binary>>;
dkim_encode_tag(h, Hdrs) ->
%% headers fields (case-insensitive, ":" separated)
Joined = binstr:join([binstr:to_lower(H) || H <- Hdrs], <<":">>),
<<"h=", Joined/binary>>;
dkim_encode_tag(i, V) ->
%% AUID
QPValue = dkim_qp_tag_value(V),
<<"i=", QPValue/binary>>;
dkim_encode_tag(l, IntVal) ->
%% body length count
BinVal = list_to_binary(integer_to_list(IntVal)),
<<"l=", (BinVal)/binary>>;
dkim_encode_tag(q, [<<"dns/txt">>]) ->
%% query methods (':' separated)
<<"q=dns/txt">>;
dkim_encode_tag(s, Selector) ->
%% selector
<<"s=", Selector/binary>>;
dkim_encode_tag(t, now) ->
dkim_encode_tag(t, calendar:universal_time());
dkim_encode_tag(t, DateTime) ->
%% timestamp
BinTs = datetime_to_bin_timestamp(DateTime),
<<"t=", BinTs/binary>>;
dkim_encode_tag(x, DateTime) ->
%% signature expiration
BinTs = datetime_to_bin_timestamp(DateTime),
<<"x=", BinTs/binary>>;
%% dkim_encode_tag(z, Hdrs) ->
%% %% copied header fields
%% Joined = dkim_qp_tag_value(binstr:join([(H) || H <- Hdrs], <<"|">>)),
%% <<"z=", Joined/binary>>;
dkim_encode_tag(K, V) when is_binary(K), is_binary(V) ->
<<K/binary, V/binary>>.
dkim_qp_tag_value(Value) ->
%% XXX: this not fully satisfy #2.11
[QPValue] = encode_quoted_printable(Value),
binary:replace(QPValue, <<";">>, <<"=3B">>).
datetime_to_bin_timestamp(DateTime) ->
% calendar:datetime_to_gregorian_seconds({{1970,1,1}, {0,0,0}})
EpochStart = 62167219200,
UnixTimestamp = calendar:datetime_to_gregorian_seconds(DateTime) - EpochStart,
list_to_binary(integer_to_list(UnixTimestamp)).
%% /DKIM
-ifdef(TEST).
parse_with_comments_test_() ->
[
{"bleh", fun() ->
?assertEqual(<<"1.0">>, parse_with_comments(<<"1.0">>)),
?assertEqual(<<"1.0">>, parse_with_comments(<<"1.0 (produced by MetaSend Vx.x)">>)),
?assertEqual(<<"1.0">>, parse_with_comments(<<"(produced by MetaSend Vx.x) 1.0">>)),
?assertEqual(<<"1.0">>, parse_with_comments(<<"1.(produced by MetaSend Vx.x)0">>))
end},
{"comments that parse as empty", fun() ->
?assertEqual(<<>>, parse_with_comments(<<"(comment (nested (deeply)) (and (oh no!) again))">>)),
?assertEqual(<<>>, parse_with_comments(<<"(\\)\\\\)">>)),
?assertEqual(<<>>, parse_with_comments(<<"(by way of Whatever <redir@my.org>) (generated by Eudora)">>))
end},
{"some more", fun() ->
?assertEqual(
<<":sysmail@ group. org, Muhammed. Ali @Vegas.WBA">>,
parse_with_comments(<<"\":sysmail\"@ group. org, Muhammed.(the greatest) Ali @(the)Vegas.WBA">>)
),
?assertEqual(
<<"Pete <pete@silly.test>">>,
parse_with_comments(<<"Pete(A wonderful \\) chap) <pete(his account)@silly.test(his host)>">>)
)
end},
{"non list values", fun() ->
?assertEqual(undefined, parse_with_comments(undefined)),
?assertEqual(17, parse_with_comments(17))
end},
{"Parens within quotes ignored", fun() ->
?assertEqual(<<"Height (from xkcd).eml">>, parse_with_comments(<<"\"Height (from xkcd).eml\"">>)),
?assertEqual(<<"Height (from xkcd).eml">>, parse_with_comments(<<"\"Height \(from xkcd\).eml\"">>))
end},
{"Escaped quotes are handled correctly", fun() ->
?assertEqual(<<"Hello \"world\"">>, parse_with_comments(<<"Hello \\\"world\\\"">>)),
?assertEqual(
<<"<boss@nil.test>, Giant; \"Big\" Box <sysservices@example.net>">>,
parse_with_comments(<<"<boss@nil.test>, \"Giant; \\\"Big\\\" Box\" <sysservices@example.net>">>)
)
end},
{"backslash not part of a quoted pair", fun() ->
?assertEqual(<<"AC \\ DC">>, parse_with_comments(<<"AC \\ DC">>)),
?assertEqual(<<"AC DC">>, parse_with_comments(<<"AC ( \\ ) DC">>))
end},
{"Unterminated quotes or comments", fun() ->
?assertError(unterminated_quotes, parse_with_comments(<<"\"Hello there ">>)),
?assertError(unterminated_quotes, parse_with_comments(<<"\"Hello there \\\"">>)),
?assertError(unterminated_comment, parse_with_comments(<<"(Hello there ">>)),
?assertError(unterminated_comment, parse_with_comments(<<"(Hello there \\\)">>))
end}
].
parse_content_type_test_() ->
[
{"parsing content types", fun() ->
?assertEqual(
{<<"text">>, <<"plain">>, [{<<"charset">>, <<"us-ascii">>}]},
parse_content_type(<<"text/plain; charset=us-ascii (Plain text)">>)
),
?assertEqual(
{<<"text">>, <<"plain">>, [{<<"charset">>, <<"us-ascii">>}]},
parse_content_type(<<"text/plain; charset=\"us-ascii\"">>)
),
?assertEqual(
{<<"text">>, <<"plain">>, [{<<"charset">>, <<"us-ascii">>}]},
parse_content_type(<<"Text/Plain; Charset=\"us-ascii\"">>)
),
?assertEqual(
{<<"multipart">>, <<"mixed">>, [{<<"boundary">>, <<"----_=_NextPart_001_01C9DCAE.1F2CB390">>}]},
parse_content_type(<<"multipart/mixed; boundary=\"----_=_NextPart_001_01C9DCAE.1F2CB390\"">>)
)
end},
{"parsing content type with a tab in it", fun() ->
?assertEqual(
{<<"text">>, <<"plain">>, [{<<"charset">>, <<"us-ascii">>}]},
parse_content_type(<<"text/plain;\tcharset=us-ascii">>)
),
?assertEqual(
{<<"text">>, <<"plain">>, [{<<"charset">>, <<"us-ascii">>}, {<<"foo">>, <<"bar">>}]},
parse_content_type(<<"text/plain;\tcharset=us-ascii;\tfoo=bar">>)
)
end},
{"invalid content types", fun() ->
?assertThrow(bad_content_type, parse_content_type(<<"text\\plain; charset=us-ascii">>)),
?assertThrow(bad_content_type, parse_content_type(<<"text/plain; charset us-ascii">>))
end}
].
parse_content_disposition_test_() ->
[
{"parsing valid dispositions", fun() ->
?assertEqual({<<"inline">>, []}, parse_content_disposition(<<"inline">>)),
?assertEqual({<<"inline">>, []}, parse_content_disposition(<<"inline;">>)),
?assertEqual(
{<<"attachment">>, [
{<<"filename">>, <<"genome.jpeg">>},
{<<"modification-date">>, <<"Wed, 12 Feb 1997 16:29:51 -0500">>}
]},
parse_content_disposition(
<<"attachment; filename=genome.jpeg;modification-date=\"Wed, 12 Feb 1997 16:29:51 -0500\";">>
)
),
?assertEqual(
{<<"text/plain">>, [{<<"charset">>, <<"us-ascii">>}]},
parse_content_disposition(<<"text/plain; charset=us-ascii (Plain text)">>)
)
end},
{"invalid dispositions", fun() ->
?assertThrow(bad_disposition, parse_content_disposition(<<"inline; =bar">>)),
?assertThrow(bad_disposition, parse_content_disposition(<<"inline; bar">>))
end}
].
various_parsing_test_() ->
[
{"split_body_by_boundary test", fun() ->
?assertEqual(
[{[], <<"foo bar baz">>}], split_body_by_boundary_(<<"stuff\r\nfoo bar baz">>, <<"--bleh">>, [], [])
),
?assertEqual(
[{[], <<"foo\r\n">>}, {[], <<>>}, {[], <<>>}, {[], <<"bar baz">>}],
split_body_by_boundary_(
<<"stuff\r\nfoo\r\n--bleh\r\n--bleh\r\n--bleh-- stuff\r\nbar baz">>, <<"--bleh">>, [], []
)
),
%?assertEqual([{[], []}, {[], []}, {[], "bar baz"}], split_body_by_boundary_("\r\n--bleh\r\n--bleh\r\n", "--bleh", [], [])),
%?assertMatch([{"text", "plain", [], _,"foo\r\n"}], split_body_by_boundary("stuff\r\nfoo\r\n--bleh\r\n--bleh\r\n--bleh-- stuff\r\nbar baz", "--bleh", "1.0", []))
?assertEqual({[], <<"foo: bar\r\n">>}, parse_headers(<<"\r\nfoo: bar\r\n">>)),
?assertEqual({[{<<"foo">>, <<"barbaz">>}], <<>>}, parse_headers(<<"foo: bar\r\n baz\r\n">>)),
?assertEqual({[], <<" foo bar baz\r\nbam">>}, parse_headers(<<"\sfoo bar baz\r\nbam">>)),
ok
end},
{"Headers with non-ASCII characters", fun() ->
?assertEqual({[{<<"foo">>, <<"bar ?? baz">>}], <<>>}, parse_headers(<<"foo: bar ø baz\r\n"/utf8>>)),
?assertEqual({[], <<"bär: bar baz\r\n"/utf8>>}, parse_headers(<<"bär: bar baz\r\n"/utf8>>))
end},
{"Headers with tab characters", fun() ->
?assertEqual({[{<<"foo">>, <<"bar baz">>}], <<>>}, parse_headers(<<"foo: bar baz\r\n">>))
end}
].
-define(IMAGE_MD5, <<110, 130, 37, 247, 39, 149, 224, 61, 114, 198, 227, 138, 113, 4, 198, 60>>).
parse_example_mails_test_() ->
Getmail = fun(File) ->
{ok, Email} = file:read_file(string:concat("test/fixtures/", File)),
%Email = binary_to_list(Bin),
decode(Email)
end,
[
{"parse a plain text email", fun() ->
Decoded = Getmail("Plain-text-only.eml"),
?assertEqual(5, tuple_size(Decoded)),
{Type, SubType, _Headers, _Properties, Body} = Decoded,
?assertEqual({<<"text">>, <<"plain">>}, {Type, SubType}),
?assertEqual(<<"This message contains only plain text.\r\n">>, Body)
end},
{"parse a Python smtplib plain text email", fun() ->
Decoded = Getmail("python-smtp-lib.eml"),
?assertEqual(5, tuple_size(Decoded)),
{Type, SubType, _Headers, _Properties, Body} = Decoded,
?assertEqual({<<"text">>, <<"plain">>}, {Type, SubType}),
?assertEqual(<<"Hello world Python.\r\n">>, Body)
end},
{"parse a plain text email with no content type", fun() ->
Decoded = Getmail("Plain-text-only-no-content-type.eml"),
?assertEqual(5, tuple_size(Decoded)),
{Type, SubType, _Headers, _Properties, Body} = Decoded,
?assertEqual({<<"text">>, <<"plain">>}, {Type, SubType}),
?assertEqual(<<"This message contains only plain text.\r\n">>, Body)
end},
{"parse a plain text email with no MIME header", fun() ->
{Type, SubType, _Headers, _Properties, Body} =
Getmail("Plain-text-only-no-MIME.eml"),
?assertEqual({<<"text">>, <<"plain">>}, {Type, SubType}),
?assertEqual(<<"This message contains only plain text.\r\n">>, Body)
end},
{"parse an email that says it is multipart but contains no boundaries", fun() ->
?assertError(missing_boundary, Getmail("Plain-text-only-with-boundary-header.eml"))
end},
{"parse a multipart email with no MIME header", fun() ->
% We now insert a default Mime for missing Mime headers
% ?assertError(non_mime_multipart, Getmail("rich-text-no-MIME.eml"))
?assertMatch(
{<<"multipart">>, <<"alternative">>, _, _, [
{<<"text">>, <<"plain">>, _, _, _}, {<<"text">>, <<"html">>, _, _, _}
]},
Getmail("rich-text-no-MIME.eml")
)
end},
{"rich text", fun() ->
%% pardon my naming here. apparently 'rich text' in mac mail
%% means 'html'.
Decoded = Getmail("rich-text.eml"),
?assertEqual(5, tuple_size(Decoded)),
{Type, SubType, _Headers, _Properties, Body} = Decoded,
?assertEqual({<<"multipart">>, <<"alternative">>}, {Type, SubType}),
?assertEqual(2, length(Body)),
[Plain, Html] = Body,
?assertEqual({5, 5}, {tuple_size(Plain), tuple_size(Html)}),
?assertMatch({<<"text">>, <<"plain">>, _, _, <<"This message contains rich text.">>}, Plain),
?assertMatch(
{<<"text">>, <<"html">>, _, _,
<<"<html><body style=\"word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; \"><b>This </b><i>message </i><span class=\"Apple-style-span\" style=\"text-decoration: underline;\">contains </span>rich text.</body></html>">>},
Html
)
end},
{"rich text no boundary", fun() ->
?assertError(no_boundary, Getmail("rich-text-no-boundary.eml"))
end},
{"rich text missing first boundary", fun() ->
% TODO - should we handle this more elegantly?
Decoded = Getmail("rich-text-missing-first-boundary.eml"),
?assertEqual(5, tuple_size(Decoded)),
{Type, SubType, _Headers, _Properties, Body} = Decoded,
?assertEqual({<<"multipart">>, <<"alternative">>}, {Type, SubType}),
?assertEqual(1, length(Body)),
[Html] = Body,
?assertEqual(5, tuple_size(Html)),
?assertMatch(
{<<"text">>, <<"html">>, _, _,
<<"<html><body style=\"word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; \"><b>This </b><i>message </i><span class=\"Apple-style-span\" style=\"text-decoration: underline;\">contains </span>rich text.</body></html>">>},
Html
)
end},
{"rich text missing last boundary", fun() ->
?assertError(missing_last_boundary, Getmail("rich-text-missing-last-boundary.eml"))
end},
{"rich text wrong last boundary", fun() ->
?assertError(missing_last_boundary, Getmail("rich-text-broken-last-boundary.eml"))
end},
{"rich text missing text content type", fun() ->
%% pardon my naming here. apparently 'rich text' in mac mail
%% means 'html'.
Decoded = Getmail("rich-text-no-text-contenttype.eml"),
?assertEqual(5, tuple_size(Decoded)),
{Type, SubType, _Headers, _Properties, Body} = Decoded,
?assertEqual({<<"multipart">>, <<"alternative">>}, {Type, SubType}),
?assertEqual(2, length(Body)),
[Plain, Html] = Body,
?assertEqual({5, 5}, {tuple_size(Plain), tuple_size(Html)}),
?assertMatch({<<"text">>, <<"plain">>, _, _, <<"This message contains rich text.">>}, Plain),
?assertMatch(
{<<"text">>, <<"html">>, _, _,
<<"<html><body style=\"word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; \"><b>This </b><i>message </i><span class=\"Apple-style-span\" style=\"text-decoration: underline;\">contains </span>rich text.</body></html>">>},
Html
)
end},
{"text attachment only", fun() ->
Decoded = Getmail("text-attachment-only.eml"),
?assertEqual(5, tuple_size(Decoded)),
{Type, SubType, _Headers, _Properties, Body} = Decoded,
?assertEqual({<<"multipart">>, <<"mixed">>}, {Type, SubType}),
?assertEqual(1, length(Body)),
Rich =
<<"{\\rtf1\\ansi\\ansicpg1252\\cocoartf949\\cocoasubrtf460\r\n{\\fonttbl\\f0\\fswiss\\fcharset0 Helvetica;}\r\n{\\colortbl;\\red255\\green255\\blue255;}\r\n\\margl1440\\margr1440\\vieww9000\\viewh8400\\viewkind0\r\n\\pard\\tx720\\tx1440\\tx2160\\tx2880\\tx3600\\tx4320\\tx5040\\tx5760\\tx6480\\tx7200\\tx7920\\tx8640\\ql\\qnatural\\pardirnatural\r\n\r\n\\f0\\fs24 \\cf0 This is a basic rtf file.}">>,
?assertMatch([{<<"text">>, <<"rtf">>, _, _, Rich}], Body)
end},
{"image attachment only", fun() ->
Decoded = Getmail("image-attachment-only.eml"),
?assertEqual(5, tuple_size(Decoded)),
{Type, SubType, _Headers, _Properties, Body} = Decoded,
?assertEqual({<<"multipart">>, <<"mixed">>}, {Type, SubType}),
?assertEqual(1, length(Body)),
?assertMatch([{<<"image">>, <<"jpeg">>, _, _, _}], Body),
[H | _] = Body,
[{<<"image">>, <<"jpeg">>, _, Parameters, _Image}] = Body,
?assertEqual(?IMAGE_MD5, erlang:md5(element(5, H))),
?assertEqual(<<"inline">>, maps:get(disposition, Parameters)),
?assertEqual(
<<"chili-pepper.jpg">>, proplists:get_value(<<"filename">>, maps:get(disposition_params, Parameters))
),
?assertEqual(
<<"chili-pepper.jpg">>, proplists:get_value(<<"name">>, maps:get(content_type_params, Parameters))
)
end},
{"message attachment only", fun() ->
Decoded = Getmail("message-as-attachment.eml"),
?assertMatch({<<"multipart">>, <<"mixed">>, _, _, _}, Decoded),
[Body] = element(5, Decoded),
?assertMatch({<<"message">>, <<"rfc822">>, _, _, _}, Body),
Subbody = element(5, Body),
?assertMatch({<<"text">>, <<"plain">>, _, _, _}, Subbody),
?assertEqual(<<"This message contains only plain text.\r\n">>, element(5, Subbody))
end},
{"message, image, and rtf attachments.", fun() ->
Decoded = Getmail("message-image-text-attachments.eml"),
?assertMatch({<<"multipart">>, <<"mixed">>, _, _, _}, Decoded),
?assertEqual(3, length(element(5, Decoded))),
[Message, Rtf, Image] = element(5, Decoded),
?assertMatch({<<"message">>, <<"rfc822">>, _, _, _}, Message),
Submessage = element(5, Message),
?assertMatch({<<"text">>, <<"plain">>, _, _, <<"This message contains only plain text.\r\n">>}, Submessage),
?assertMatch({<<"text">>, <<"rtf">>, _, _, _}, Rtf),
?assertEqual(
<<"{\\rtf1\\ansi\\ansicpg1252\\cocoartf949\\cocoasubrtf460\r\n{\\fonttbl\\f0\\fswiss\\fcharset0 Helvetica;}\r\n{\\colortbl;\\red255\\green255\\blue255;}\r\n\\margl1440\\margr1440\\vieww9000\\viewh8400\\viewkind0\r\n\\pard\\tx720\\tx1440\\tx2160\\tx2880\\tx3600\\tx4320\\tx5040\\tx5760\\tx6480\\tx7200\\tx7920\\tx8640\\ql\\qnatural\\pardirnatural\r\n\r\n\\f0\\fs24 \\cf0 This is a basic rtf file.}">>,
element(5, Rtf)
),
?assertMatch({<<"image">>, <<"jpeg">>, _, _, _}, Image),
?assertEqual(?IMAGE_MD5, erlang:md5(element(5, Image)))
end},
{"alternative text/html with calendar attachment.", fun() ->
Decoded = Getmail("message-text-html-attachment.eml"),
?assertMatch(
{<<"multipart">>, <<"mixed">>, _, _, [
{<<"multipart">>, <<"alternative">>, _, _, [
{<<"text">>, <<"plain">>, _, _, _},
{<<"text">>, <<"html">>, _, _, _}
]},
{<<"text">>, <<"calendar">>, _, _, _}
]},
Decoded
)
end},
{"Outlook 2007 with leading tabs in quoted-printable.", fun() ->
Decoded = Getmail("outlook-2007.eml"),
?assertMatch({<<"multipart">>, <<"alternative">>, _, _, _}, Decoded)
end},
{"The gamut", fun() ->
% multipart/alternative
% text/plain
% multipart/mixed
% text/html
% message/rf822
% multipart/mixed
% message/rfc822
% text/plain
% text/html
% message/rtc822
% text/plain
% text/html
% image/jpeg
% text/html
% text/rtf
% text/html
Decoded = Getmail("the-gamut.eml"),
?assertMatch({<<"multipart">>, <<"alternative">>, _, _, _}, Decoded),
?assertEqual(2, length(element(5, Decoded))),
[Toptext, Topmultipart] = element(5, Decoded),
?assertMatch({<<"text">>, <<"plain">>, _, _, _}, Toptext),
?assertEqual(
<<"This is rich text.\r\n\r\nThe list is html.\r\n\r\nAttchments:\r\nan email containing an attachment of an email.\r\nan email of only plain text.\r\nan image\r\nan rtf file.\r\n">>,
element(5, Toptext)
),
?assertEqual(9, length(element(5, Topmultipart))),
[Html, Messagewithin, Brhtml, _Message, Brhtml, Image, Brhtml, Rtf, Brhtml] = element(5, Topmultipart),
?assertMatch({<<"text">>, <<"html">>, _, _, _}, Html),
?assertEqual(
<<"<html><body style=\"word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; \"><b>This</b> is <i>rich</i> text.<div><br></div><div>The list is html.</div><div><br></div><div>Attchments:</div><div><ul class=\"MailOutline\"><li>an email containing an attachment of an email.</li><li>an email of only plain text.</li><li>an image</li><li>an rtf file.</li></ul></div><div></div></body></html>">>,
element(5, Html)
),
?assertMatch({<<"message">>, <<"rfc822">>, _, _, _}, Messagewithin),
%?assertEqual(1, length(element(5, Messagewithin))),
?assertMatch(
{<<"multipart">>, <<"mixed">>, _, _, [
{<<"message">>, <<"rfc822">>, _, _,
{<<"text">>, <<"plain">>, _, _, <<"This message contains only plain text.\r\n">>}}
]},
element(5, Messagewithin)
),
?assertMatch({<<"image">>, <<"jpeg">>, _, _, _}, Image),
?assertEqual(?IMAGE_MD5, erlang:md5(element(5, Image))),
?assertMatch({<<"text">>, <<"rtf">>, _, _, _}, Rtf),
?assertEqual(
<<"{\\rtf1\\ansi\\ansicpg1252\\cocoartf949\\cocoasubrtf460\r\n{\\fonttbl\\f0\\fswiss\\fcharset0 Helvetica;}\r\n{\\colortbl;\\red255\\green255\\blue255;}\r\n\\margl1440\\margr1440\\vieww9000\\viewh8400\\viewkind0\r\n\\pard\\tx720\\tx1440\\tx2160\\tx2880\\tx3600\\tx4320\\tx5040\\tx5760\\tx6480\\tx7200\\tx7920\\tx8640\\ql\\qnatural\\pardirnatural\r\n\r\n\\f0\\fs24 \\cf0 This is a basic rtf file.}">>,
element(5, Rtf)
)
end},
{"Plain text and 2 identical attachments", fun() ->
Decoded = Getmail("plain-text-and-two-identical-attachments.eml"),
?assertMatch({<<"multipart">>, <<"mixed">>, _, _, _}, Decoded),
?assertEqual(3, length(element(5, Decoded))),
[Plain, Attach1, Attach2] = element(5, Decoded),
?assertEqual(Attach1, Attach2),
?assertMatch({<<"text">>, <<"plain">>, _, _, _}, Plain),
?assertEqual(<<"This message contains only plain text.\r\n">>, element(5, Plain))
end},
{"no \\r\\n before first boundary", fun() ->
{ok, Bin} = file:read_file("test/fixtures/html.eml"),
Decoded = decode(Bin),
?assertEqual(2, length(element(5, Decoded)))
end},
{"permissive malformed folded multibyte header decoder", fun() ->
{_, _, Headers, _, Body} = Getmail("malformed-folded-multibyte-header.eml"),
?assertEqual(<<"Hello world\n">>, Body),
Subject =
<<78, 79, 68, 51, 50, 32, 83, 109, 97, 114, 116, 32, 83, 101, 99, 117, 114, 105, 116, 121, 32, 45, 32,
208, 177, 208, 181, 209, 129, 208, 191, 208, 187, 208, 176, 209, 130, 208, 189, 208, 176, 209, 143,
32, 208, 187, 208, 184, 209, 134, 208, 181, 208, 189, 208, 183, 208, 184, 209, 143>>,
?assertEqual(Subject, proplists:get_value(<<"Subject">>, Headers))
end},
{"decode headers of multipart messages", fun() ->
{<<"multipart">>, _, _, _, [Inline, Attachment]} = Getmail("utf-attachment-name.eml"),
{<<"text">>, _, _, _, InlineBody} = Inline,
{<<"text">>, _, _, ContentHeaders, _AttachmentBody} = Attachment,
ContentTypeName = proplists:get_value(
<<"name">>,
maps:get(
content_type_params, ContentHeaders
)
),
DispositionName = proplists:get_value(
<<"filename">>,
maps:get(
disposition_params, ContentHeaders
)
),
?assertEqual(<<"Hello\r\n">>, InlineBody),
?assert(ContentTypeName == DispositionName),
% Take the filename as a literal, to prevent character set issues with Erlang
% In utf-8 the filename is:"тестовый файл.txt"
Filename =
<<209, 130, 208, 181, 209, 129, 209, 130, 208, 190, 208, 178, 209, 139, 208, 185, 32, 209, 132, 208,
176, 208, 185, 208, 187, 46, 116, 120, 116>>,
?assertEqual(Filename, ContentTypeName),
?assertEqual(Filename, DispositionName)
end},
{"testcase1", fun() ->
Multipart = <<"multipart">>,
Alternative = <<"alternative">>,
Related = <<"related">>,
Mixed = <<"mixed">>,
Text = <<"text">>,
Html = <<"html">>,
Plain = <<"plain">>,
Message = <<"message">>,
Ref822 = <<"rfc822">>,
Image = <<"image">>,
Jpeg = <<"jpeg">>,
%Imagemd5 = <<69,175,198,78,52,72,6,233,147,22,50,137,128,180,169,50>>,
Imagemd5 = <<179, 151, 42, 139, 78, 14, 182, 78, 24, 160, 123, 221, 217, 14, 141, 5>>,
Decoded = Getmail("testcase1"),
?assertMatch({Multipart, Mixed, _, _, [_, _]}, Decoded),
[Multi1, Message1] = element(5, Decoded),
?assertMatch({Multipart, Alternative, _, _, [_, _]}, Multi1),
[Plain1, Html1] = element(5, Multi1),
?assertMatch({Text, Plain, _, _, _}, Plain1),
?assertMatch({Text, Html, _, _, _}, Html1),
?assertMatch({Message, Ref822, _, _, _}, Message1),
Multi2 = element(5, Message1),
?assertMatch({Multipart, Alternative, _, _, [_, _]}, Multi2),
[Plain2, Related1] = element(5, Multi2),
?assertMatch({Text, Plain, _, _, _}, Plain2),
?assertMatch({Multipart, Related, _, _, [_, _]}, Related1),
[Html2, Image1] = element(5, Related1),
?assertMatch({Text, Html, _, _, _}, Html2),
?assertMatch({Image, Jpeg, _, _, _}, Image1),
Resimage = erlang:md5(element(5, Image1)),
?assertEqual(Imagemd5, Resimage)
end},
{"testcase2", fun() ->
Multipart = <<"multipart">>,
Alternative = <<"alternative">>,
Mixed = <<"mixed">>,
Text = <<"text">>,
Html = <<"html">>,
Plain = <<"plain">>,
Message = <<"message">>,
Ref822 = <<"rfc822">>,
Application = <<"application">>,
Octetstream = <<"octet-stream">>,
Decoded = Getmail("testcase2"),
?assertMatch({Multipart, Mixed, _, _, [_, _, _]}, Decoded),
[Plain1, Stream1, Message1] = element(5, Decoded),
?assertMatch({Text, Plain, _, _, _}, Plain1),
?assertMatch({Application, Octetstream, _, _, _}, Stream1),
?assertMatch({Message, Ref822, _, _, _}, Message1),
Multi1 = element(5, Message1),
?assertMatch({Multipart, Alternative, _, _, [_, _]}, Multi1),
[Plain2, Html1] = element(5, Multi1),
?assertMatch({Text, Plain, _, _, _}, Plain2),
?assertMatch({Text, Html, _, _, _}, Html1)
end}
].
decode_quoted_printable_test_() ->
[
{"bleh", fun() ->
?assertEqual(<<"!">>, decode_quoted_printable(<<"=21">>)),
?assertEqual(<<"!!">>, decode_quoted_printable(<<"=21=21">>)),
?assertEqual(<<"=:=">>, decode_quoted_printable(<<"=3D:=3D">>)),
?assertEqual(
<<"Thequickbrownfoxjumpedoverthelazydog.">>,
decode_quoted_printable(<<"Thequickbrownfoxjumpedoverthelazydog.">>)
)
end},
{"lowercase bleh", fun() ->
?assertEqual(<<"=:=">>, decode_quoted_printable(<<"=3d:=3d">>))
end},
{"input with spaces", fun() ->
?assertEqual(
<<"The quick brown fox jumped over the lazy dog.">>,
decode_quoted_printable(<<"The quick brown fox jumped over the lazy dog.">>)
)
end},
{"input with tabs", fun() ->
?assertEqual(
<<"The\tquick brown fox jumped over\tthe lazy dog.">>,
decode_quoted_printable(<<"The\tquick brown fox jumped over\tthe lazy dog.">>)
)
end},
{"input with trailing spaces", fun() ->
?assertEqual(
<<"The quick brown fox jumped over the lazy dog.">>,
decode_quoted_printable(<<"The quick brown fox jumped over the lazy dog. ">>)
)
end},
{"input with non-strippable trailing whitespace", fun() ->
?assertEqual(
<<"The quick brown fox jumped over the lazy dog. ">>,
decode_quoted_printable(<<"The quick brown fox jumped over the lazy dog. =20">>)
),
?assertEqual(
<<"The quick brown fox jumped over the lazy dog. \t">>,
decode_quoted_printable(<<"The quick brown fox jumped over the lazy dog. =09">>)
),
?assertEqual(
<<"The quick brown fox jumped over the lazy dog.\t \t \t \t ">>,
decode_quoted_printable(<<"The quick brown fox jumped over the lazy dog.\t \t \t =09=20">>)
),
?assertEqual(
<<"The quick brown fox jumped over the lazy dog.\t \t \t \t ">>,
decode_quoted_printable(
<<"The quick brown fox jumped over the lazy dog.\t \t \t =09=20\t \t">>
)
)
end},
{"input with trailing tabs", fun() ->
?assertEqual(
<<"The quick brown fox jumped over the lazy dog.">>,
decode_quoted_printable(<<"The quick brown fox jumped over the lazy dog.\t\t\t\t\t">>)
)
end},
{"soft new line", fun() ->
?assertEqual(
<<"The quick brown fox jumped over the lazy dog. ">>,
decode_quoted_printable(<<"The quick brown fox jumped over the lazy dog. =">>)
)
end},
{"soft new line with trailing whitespace", fun() ->
?assertEqual(
<<"The quick brown fox jumped over the lazy dog. ">>,
decode_quoted_printable(<<"The quick brown fox jumped over the lazy dog. = ">>)
)
end},
{"multiline stuff", fun() ->
?assertEqual(
<<"Now's the time for all folk to come to the aid of their country.">>,
decode_quoted_printable(
<<"Now's the time =\r\nfor all folk to come=\r\n to the aid of their country.">>
)
),
?assertEqual(
<<"Now's the time\r\nfor all folk to come\r\n to the aid of their country.">>,
decode_quoted_printable(<<"Now's the time\r\nfor all folk to come\r\n to the aid of their country.">>)
),
?assertEqual(<<"hello world">>, decode_quoted_printable(<<"hello world">>)),
?assertEqual(<<"hello\r\n\r\nworld">>, decode_quoted_printable(<<"hello\r\n\r\nworld">>))
end},
{"invalid input", fun() ->
?assertThrow(badchar, decode_quoted_printable(<<"=21=G1">>)),
?assertThrow(badchar, decode_quoted_printable(<<"=21=D1 = g ">>))
end},
%% TODO zotonic's iconv throws eilseq here.
% {"out of range characters should be stripped",
% fun() ->
% % character 150 is en-dash in windows 1252
% ?assertEqual(<<"Foo bar"/utf8>>, decode_body(<<"quoted-printable">>, <<"Foo ", 150, " bar">>, "US-ASCII", "UTF-8//IGNORE"))
% end
% },
{"out of range character in alternate charset should be converted", fun() ->
% character 150 is en-dash in windows 1252
?assertEqual(
<<"Foo ", 226, 128, 147, " bar">>,
decode_body(<<"quoted-printable">>, <<"Foo ", 150, " bar">>, "Windows-1252", "UTF-8//IGNORE")
)
end},
{"out of range character in alternate charset with no destination encoding should be stripped", fun() ->
% character 150 is en-dash in windows 1252
?assertEqual(
<<"Foo bar">>, decode_body(<<"quoted-printable">>, <<"Foo ", 150, " bar">>, "Windows-1252", none)
)
end},
{"out of range character in alternate charset with no source encoding should be stripped", fun() ->
% character 150 is en-dash in windows 1252
?assertEqual(
<<"Foo bar">>, decode_body(<<"quoted-printable">>, <<"Foo ", 150, " bar">>, undefined, "UTF-8")
)
end},
{"almost correct chatsets should work, eg. 'UTF8' instead of 'UTF-8'", fun() ->
% character 150 is en-dash in windows 1252
?assertEqual(<<"Foo bar">>, decode_body(<<"quoted-printable">>, <<"Foo bar">>, <<"UTF8">>, "UTF-8")),
?assertEqual(<<"Foo bar">>, decode_body(<<"quoted-printable">>, <<"Foo bar">>, <<"utf8">>, "UTF-8"))
end}
].
valid_smtp_mime_7bit_test() ->
?assert(valid_7bit(<<>>)),
?assert(valid_7bit(<<"abcdefghijklmnopqrstuvwxyz0123456789">>)),
?assert(valid_7bit(<<"abc\r\ndef">>)),
AllValidRange =
(lists:seq(1, $\n - 1) ++
lists:seq($\n + 1, $\r - 1) ++
lists:seq($\r + 1, 127)),
?assert(valid_7bit(list_to_binary(AllValidRange))),
?assertNot(valid_7bit(<<"\n">>)),
?assertNot(valid_7bit(<<"\r">>)),
?assertNot(valid_7bit(<<"abc\ndef">>)),
?assertNot(valid_7bit(<<"abc\rdef">>)),
?assertNot(valid_7bit(<<"abc\n\rdef">>)),
?assertNot(valid_7bit(<<128, 200, 255>>)),
?assertNot(valid_7bit(<<0, 0, 0>>)),
?assertNot(valid_7bit(<<"hello", 128, 0, 200>>)),
%% Long lines
Line800 = binary:copy(<<$a>>, 800),
?assertNot(has_lines_over_998(Line800)),
Many800Lines = list_to_binary(lists:join("\r\n", lists:duplicate(10, Line800))),
?assertNot(has_lines_over_998(Many800Lines)),
Line1000 = binary:copy(<<$a>>, 1000),
?assert(has_lines_over_998(Line1000)),
Many1000Lines = list_to_binary(lists:join("\r\n", lists:duplicate(10, Line1000))),
?assert(has_lines_over_998(Many1000Lines)),
?assert(has_lines_over_998(<<Line800/binary, "\r\n", Line1000/binary>>)).
encode_quoted_printable_test_() ->
[
{"bleh", fun() ->
?assertEqual([<<"!">>], encode_quoted_printable(<<"!">>)),
?assertEqual([<<"!!">>], encode_quoted_printable(<<"!!">>)),
?assertEqual([<<"=3D:=3D">>], encode_quoted_printable(<<"=:=">>)),
?assertEqual(
[<<"Thequickbrownfoxjumpedoverthelazydog.">>],
encode_quoted_printable(<<"Thequickbrownfoxjumpedoverthelazydog.">>)
)
end},
{"input with spaces", fun() ->
?assertEqual(
[<<"The quick brown fox jumped over the lazy dog.">>],
encode_quoted_printable(<<"The quick brown fox jumped over the lazy dog.">>)
)
end},
{"input with tabs", fun() ->
?assertEqual(
[<<"The\tquick brown fox jumped over\tthe lazy dog.">>],
encode_quoted_printable(<<"The\tquick brown fox jumped over\tthe lazy dog.">>)
)
end},
{"input with trailing spaces", fun() ->
?assertEqual(
[<<"The quick brown fox jumped over the lazy dog. =20\r\n">>],
encode_quoted_printable(<<"The quick brown fox jumped over the lazy dog. \r\n">>)
),
?assertEqual(
[<<"The quick brown fox jumped over the lazy dog. =20">>],
encode_quoted_printable(<<"The quick brown fox jumped over the lazy dog. ">>)
)
end},
{"input with trailing tabs", fun() ->
?assertEqual(
[<<"The quick brown fox jumped over the lazy dog. =09\r\n">>],
encode_quoted_printable(<<"The quick brown fox jumped over the lazy dog. \r\n">>)
),
?assertEqual(
[<<"The quick brown fox jumped over the lazy dog. =09">>],
encode_quoted_printable(<<"The quick brown fox jumped over the lazy dog. ">>)
)
end},
{"input with non-ascii characters", fun() ->
?assertEqual(
[<<"There's some n=F8n-=E1scii st=FCff in here\r\n">>],
encode_quoted_printable(<<"There's some n", 248, "n-", 225, "scii st", 252, "ff in here\r\n">>)
)
end},
{"input with invisible non-ascii characters", fun() ->
?assertEqual(
[<<"There's some stuff=C2=A0in=C2=A0here\r\n">>],
encode_quoted_printable(<<"There's some stuff in here\r\n"/utf8>>)
)
end},
{"add soft newlines", fun() ->
?assertEqual(
[
<<"The quick brown fox jumped over the lazy dog. The quick brown fox jumped =\r\nover the lazy dog.">>
],
encode_quoted_printable(
<<"The quick brown fox jumped over the lazy dog. The quick brown fox jumped over the lazy dog.">>
)
),
?assertEqual(
[
<<"The_quick_brown_fox_jumped_over_the_lazy_dog._The_quick_brown_fox_jumped_ov=\r\ner_the_lazy_dog.">>
],
encode_quoted_printable(
<<"The_quick_brown_fox_jumped_over_the_lazy_dog._The_quick_brown_fox_jumped_over_the_lazy_dog.">>
)
),
?assertEqual(
[
<<"The_quick_brown_fox_jumped_over_the_lazy_dog._The_quick_brown_fox_jumped_o=\r\n=3Dver_the_lazy_dog.">>
],
encode_quoted_printable(
<<"The_quick_brown_fox_jumped_over_the_lazy_dog._The_quick_brown_fox_jumped_o=ver_the_lazy_dog.">>
)
),
?assertEqual(
[
<<"The_quick_brown_fox_jumped_over_the_lazy_dog._The_quick_brown_fox_jumped_=\r\n=3Dover_the_lazy_dog.">>
],
encode_quoted_printable(
<<"The_quick_brown_fox_jumped_over_the_lazy_dog._The_quick_brown_fox_jumped_=over_the_lazy_dog.">>
)
),
?assertEqual(
[
<<"The_quick_brown_fox_jumped_over_the_lazy_dog._The_quick_brown_fox_jumped_o =\r\nver_the_lazy_dog.">>
],
encode_quoted_printable(
<<"The_quick_brown_fox_jumped_over_the_lazy_dog._The_quick_brown_fox_jumped_o ver_the_lazy_dog.">>
)
)
end},
{"soft newline edge cases", fun() ->
?assertEqual(
[
<<
"123456789 123456789 123456789 123456789 123456789 123456789 123456789 12345=\r\n"
"=20"
>>
],
encode_quoted_printable(
<<"123456789 123456789 123456789 123456789 123456789 123456789 123456789 12345 ">>
)
),
?assertEqual(
[
<<
"123456789 123456789 123456789 123456789 123456789 123456789 123456789 12345=\r\n"
"=20\r\n"
>>
],
encode_quoted_printable(
<<"123456789 123456789 123456789 123456789 123456789 123456789 123456789 12345 \r\n">>
)
),
?assertEqual(
[
<<
"123456789 123456789 123456789 123456789 123456789 123456789 123456789 12345=\r\n"
"=09"
>>
],
encode_quoted_printable(
<<"123456789 123456789 123456789 123456789 123456789 123456789 123456789 12345 ">>
)
),
?assertEqual(
[
<<
"123456789 123456789 123456789 123456789 123456789 123456789 123456789 12345=\r\n"
"=09\r\n"
>>
],
encode_quoted_printable(
<<"123456789 123456789 123456789 123456789 123456789 123456789 123456789 12345 \r\n">>
)
),
?assertEqual(
[
<<
"123456789 123456789 123456789 123456789 123456789 123456789 123456789 =\r\n"
"12345=3D"
>>
],
encode_quoted_printable(
<<"123456789 123456789 123456789 123456789 123456789 123456789 123456789 12345=">>
)
),
?assertEqual(
[
<<
" 23456789012345678901234567890123456789012345678901234567890123456789012345=\r\n"
"=20"
>>
],
encode_quoted_printable(
<<" 23456789012345678901234567890123456789012345678901234567890123456789012345 ">>
)
),
?assertEqual(
[
<<
" =\r\n"
"234567890123456789012345678901234567890123456789012345678901234567890123456"
>>
],
encode_quoted_printable(
<<" 234567890123456789012345678901234567890123456789012345678901234567890123456">>
)
),
?assertEqual(
[
<<
" 23456789012345678901234567890123456789012345678901234567890123456789012345=\r\n"
"=3D"
>>
],
encode_quoted_printable(
<<" 23456789012345678901234567890123456789012345678901234567890123456789012345=">>
)
)
end}
].
encode_parameter_test_() ->
[
{"Token", fun() ->
?assertEqual(
[[<<"a">>, $=, <<"abcdefghijklmnopqrstuvwxyz$%&*#!">>]],
encode_parameters([{<<"a">>, <<"abcdefghijklmnopqrstuvwxyz$%&*#!">>}])
)
end},
{"TSpecial", fun() ->
Special = " ()<>@,;:/[]?=",
[
?assertEqual([[<<"a">>, $=, $", <<C>>, $"]], encode_parameters([{<<"a">>, <<C>>}]))
|| C <- Special
],
?assertEqual([[<<"a">>, $=, $", <<$\\, $">>, $"]], encode_parameters([{<<"a">>, <<$">>}])),
?assertEqual([[<<"a">>, $=, $", <<$\\, $\\>>, $"]], encode_parameters([{<<"a">>, <<$\\>>}]))
end}
].
rfc2047_decode_test_() ->
[
{"Simple tests", fun() ->
?assertEqual(
<<"Keith Moore <moore@cs.utk.edu>"/utf8>>,
decode_header(<<"=?US-ASCII?Q?Keith_Moore?= <moore@cs.utk.edu>">>, "utf-8")
),
?assertEqual(
<<"Keld Jørn Simonsen <keld@dkuug.dk>"/utf8>>,
decode_header(<<"=?ISO-8859-1?Q?Keld_J=F8rn_Simonsen?= <keld@dkuug.dk>">>, "utf-8")
),
?assertEqual(
<<"Olle Järnefors <ojarnef@admin.kth.se>"/utf8>>,
decode_header(<<"=?ISO-8859-1?Q?Olle_J=E4rnefors?= <ojarnef@admin.kth.se>">>, "utf-8")
),
?assertEqual(
<<"André Pirard <PIRARD@vm1.ulg.ac.be>"/utf8>>,
decode_header(<<"=?ISO-8859-1?Q?Andr=E9?= Pirard <PIRARD@vm1.ulg.ac.be>">>, "utf-8")
)
end},
{"encoded words separated by whitespace should have whitespace removed", fun() ->
?assertEqual(
<<"If you can read this you understand the example.">>,
decode_header(
<<"=?ISO-8859-1?B?SWYgeW91IGNhbiByZWFkIHRoaXMgeW8=?= =?ISO-8859-2?B?dSB1bmRlcnN0YW5kIHRoZSBleGFtcGxlLg==?=">>,
"utf-8"
)
),
?assertEqual(<<"ab">>, decode_header(<<"=?ISO-8859-1?Q?a?= =?ISO-8859-1?Q?b?=">>, "utf-8")),
?assertEqual(<<"ab">>, decode_header(<<"=?ISO-8859-1?Q?a?= =?ISO-8859-1?Q?b?=">>, "utf-8")),
?assertEqual(
<<"ab">>,
decode_header(
<<
"=?ISO-8859-1?Q?a?=\n"
" =?ISO-8859-1?Q?b?="
>>,
"utf-8"
)
)
end},
{"underscores expand to spaces", fun() ->
?assertEqual(<<"a b">>, decode_header(<<"=?ISO-8859-1?Q?a_b?=">>, "utf-8")),
?assertEqual(<<"a b">>, decode_header(<<"=?ISO-8859-1?Q?a?= =?ISO-8859-2?Q?_b?=">>, "utf-8"))
end},
{"edgecases", fun() ->
?assertEqual(
<<"this is some text">>, decode_header(<<"=?iso-8859-1?q?this=20is=20some=20text?=">>, "utf-8")
),
?assertEqual(
<<"=?iso-8859-1?q?this is some text?=">>,
decode_header(<<"=?iso-8859-1?q?this is some text?=">>, "utf-8")
)
end},
{"invalid character sequence handling", fun() ->
?assertException(
throw,
eilseq,
decode_header(<<"=?us-ascii?B?dGhpcyBjb250YWlucyBhIGNvcHlyaWdodCCpIHN5bWJvbA==?=">>, "utf-8")
),
%?assertEqual(<<"this contains a copyright symbol"/utf8>>, decode_header(<<"=?us-ascii?B?dGhpcyBjb250YWlucyBhIGNvcHlyaWdodCCpIHN5bWJvbA==?=">>, "utf-8//IGNORE")),
?assertEqual(
<<"this contains a copyright © symbol"/utf8>>,
decode_header(<<"=?iso-8859-1?B?dGhpcyBjb250YWlucyBhIGNvcHlyaWdodCCpIHN5bWJvbA==?=">>, "utf-8//IGNORE")
)
end},
{"multiple unicode email addresses", fun() ->
?assertEqual(
<<"Jacek Złydach <jacek.zlydach@erlang-solutions.com>, chak de planet óóóó <jz@erlang-solutions.com>, Jacek Złydach <jacek.zlydach@erlang-solutions.com>, chak de planet óóóó <jz@erlang-solutions.com>"/utf8>>,
decode_header(
<<"=?UTF-8?B?SmFjZWsgWsWCeWRhY2g=?= <jacek.zlydach@erlang-solutions.com>, =?UTF-8?B?Y2hhayBkZSBwbGFuZXQgw7PDs8Ozw7M=?= <jz@erlang-solutions.com>, =?UTF-8?B?SmFjZWsgWsWCeWRhY2g=?= <jacek.zlydach@erlang-solutions.com>, =?UTF-8?B?Y2hhayBkZSBwbGFuZXQgw7PDs8Ozw7M=?= <jz@erlang-solutions.com>">>,
"utf-8"
)
)
end},
{"decode something I encoded myself", fun() ->
A = <<"Jacek Złydach <jacek.zlydach@erlang-solutions.com>"/utf8>>,
?assertEqual(A, decode_header(rfc2047_utf8_encode(A), "utf-8"))
end}
].
rfc2047_utf8_encode_test_() ->
[
{"Q-Encoding", fun() ->
?assertEqual(
<<"=?UTF-8?Q?abcdefghijklmnopqrstuvwxyz?=">>,
rfc2047_utf8_encode(q, <<"abcdefghijklmnopqrstuvwxyz">>, <<>>, 0, <<" ">>)
),
?assertEqual(
<<"=?UTF-8?Q?ABCDEFGHIJKLMNOPQRSTUVWXYZ?=">>,
rfc2047_utf8_encode(q, <<"ABCDEFGHIJKLMNOPQRSTUVWXYZ">>, <<>>, 0, <<" ">>)
),
?assertEqual(<<"=?UTF-8?Q?0123456789?=">>, rfc2047_utf8_encode(q, <<"0123456789">>, <<>>, 0, <<" ">>)),
?assertEqual(<<"=?UTF-8?Q?!*+-/?=">>, rfc2047_utf8_encode(q, <<"!*+-/">>, <<>>, 0, <<" ">>)),
?assertEqual(
<<
"=?UTF-8?Q?This_text_encodes_to_more_than_63_bytes=2E_Therefore=2C_it_shou?=\r\n"
" =?UTF-8?Q?ld_be_encoded_in_multiple_encoded_words=2E?="
>>,
rfc2047_utf8_encode(
q,
<<"This text encodes to more than 63 bytes. Therefore, it should be encoded in multiple encoded words.">>,
<<>>,
0,
<<" ">>
)
),
?assertEqual(
<<
"=?UTF-8?Q?This_text_encodes_to_more_than_63_bytes_with_offset_f?=\r\n"
"\t=?UTF-8?Q?or_a_parameter=2E_Therefore=2C_it_should_be_encoded_in_multipl?=\r\n"
"\t=?UTF-8?Q?e_encoded_words=2E?="
>>,
rfc2047_utf8_encode(
q,
<<
"This text encodes to more than 63 bytes with offset for a parameter. "
"Therefore, it should be encoded in multiple encoded words."
>>,
<<>>,
10,
<<"\t">>
)
),
?assertEqual(
<<
"=?UTF-8?Q?We_place_an_UTF8_4byte_character_over_the_breaking_point_here_?=\r\n"
" =?UTF-8?Q?=F0=9F=80=84?="
>>,
rfc2047_utf8_encode(
q,
<<"We place an UTF8 4byte character over the breaking point here ", 16#F0, 16#9F, 16#80, 16#84>>,
<<>>,
0,
<<" ">>
)
)
end},
{"B-Encoding", fun() ->
?assertEqual(
<<"=?UTF-8?B?U29tZSBzaG9ydCB0ZXh0Lg==?=">>,
rfc2047_utf8_encode(b, <<"Some short text.">>, <<>>, 0, <<" ">>)
),
?assertEqual(
<<
"=?UTF-8?B?VGhpcyB0ZXh0IGVuY29kZXMgdG8gbW9yZSB0aGFuIDYzIGJ5dGVzLiBUaGVy?=\r\n"
" =?UTF-8?B?ZWZvcmUsIGl0IHNob3VsZCBiZSBlbmNvZGVkIGluIG11bHRpcGxlIGVuY29k?=\r\n"
" =?UTF-8?B?ZWQgd29yZHMu?="
>>,
rfc2047_utf8_encode(
b,
<<"This text encodes to more than 63 bytes. Therefore, it should be encoded in multiple encoded words.">>,
<<>>,
1,
<<" ">>
)
),
?assertEqual(
<<
"=?UTF-8?B?AAECAwQFBgcICQoLDA0ODxAREhMUFRYXGBkaGxwdHh8gISIjJCUmJygpKiss?=\r\n"
" =?UTF-8?B?LS4vMDEyMzQ1Njc4OTo7PD0+P0BBQkNERUZHSElKS0xNTk9QUVJTVFVWV1hZ?=\r\n"
" =?UTF-8?B?WltcXV5fYGFiY2RlZmdoaWprbG1ub3BxcnN0dXZ3eHl6e3x9fn8=?="
>>,
rfc2047_utf8_encode(b, <<<<X>> || X <- lists:seq(0, 16#7F)>>, <<>>, 1, <<" ">>)
),
?assertEqual(
<<
"=?UTF-8?B?UGxhY2UgYW4gVVRGOCA0Ynl0ZSBjaGFyYWN0ZXIgYXQgdGhlIGJyZWFr?=\r\n"
" =?UTF-8?B?8J+AhA==?="
>>,
rfc2047_utf8_encode(
b, <<"Place an UTF8 4byte character at the break", 16#F0, 16#9F, 16#80, 16#84>>, <<>>, 1, <<" ">>
)
)
end},
{"Pick encoding", fun() ->
?assertEqual(<<"asdf">>, rfc2047_utf8_encode(<<"asdf">>)),
?assertEqual(<<"=?UTF-8?Q?x=09?=">>, rfc2047_utf8_encode(<<"x\t">>)),
?assertEqual(<<"=?UTF-8?B?CXgJ?=">>, rfc2047_utf8_encode(<<"\tx\t">>))
end}
].
encoding_test_() ->
Getmail = fun(File) ->
{ok, Email} = file:read_file(filename:join("test/fixtures/", File)),
decode(Email)
end,
[
{"Simple email", fun() ->
Email =
{<<"text">>, <<"plain">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"To">>, <<"you@example.com">>},
{<<"Subject">>, <<"This is a test">>},
{<<"Message-ID">>, <<"<abcd@example.com>">>},
{<<"MIME-Version">>, <<"1.0">>},
{<<"Date">>, <<"Sun, 01 Nov 2009 14:44:47 +0200">>}
],
#{
content_type_params =>
[{<<"charset">>, <<"US-ASCII">>}],
disposition => <<"inline">>
},
<<"This is a plain message">>},
Result =
<<"From: me@example.com\r\nTo: you@example.com\r\nSubject: This is a test\r\nMessage-ID: <abcd@example.com>\r\nMIME-Version: 1.0\r\nDate: Sun, 01 Nov 2009 14:44:47 +0200\r\n\r\nThis is a plain message">>,
?assertEqual(Result, encode(Email))
end},
{"Email with UTF-8 characters", fun() ->
Email =
{<<"text">>, <<"plain">>,
[
{<<"Subject">>, <<"Fræderik Hølljen"/utf8>>},
{<<"From">>, <<"Fræderik Hølljen <me@example.com>"/utf8>>},
{<<"To">>, <<"you@example.com">>},
{<<"Message-ID">>, <<"<abcd@example.com>">>},
{<<"MIME-Version">>, <<"1.0">>},
{<<"Date">>, <<"Sun, 01 Nov 2009 14:44:47 +0200">>}
],
#{
content_type_params =>
[{<<"charset">>, <<"US-ASCII">>}],
disposition => <<"inline">>
},
<<"This is a plain message">>},
Result =
<<"Subject: =?UTF-8?Q?Fr=C3=A6derik_H=C3=B8lljen?=\r\nFrom: =?UTF-8?Q?Fr=C3=A6derik_H=C3=B8lljen?= <me@example.com>\r\nTo: you@example.com\r\nMessage-ID: <abcd@example.com>\r\nMIME-Version: 1.0\r\nDate: Sun, 01 Nov 2009 14:44:47 +0200\r\n\r\nThis is a plain message">>,
?assertEqual(Result, encode(Email))
end},
{"Email with UTF-8 in attachment filename.", fun() ->
FileName = <<
"Čia labai ilgas el. laiško priedo pavadinimas su "/utf8,
"lietuviškomis ar kokiomis kitomis ne ascii raidėmis.pdf"/utf8
>>,
Email =
{<<"multipart">>, <<"mixed">>,
[
{<<"From">>, <<"k.petrauskas@erisata.lt">>},
{<<"Subject">>, <<"Čiobiškis"/utf8>>},
{<<"Date">>, <<"Thu, 17 Dec 2020 20:12:33 +0200">>},
{<<"Message-ID">>, <<"<47a08b7ff7d305087877361ca8eea1db@karolis.erisata.lt>">>}
],
#{
content_type_params => [
{<<"boundary">>, <<"_=boundary-123=_">>}
]
},
[
{<<"application">>, <<"pdf">>, [],
#{
content_type_params => [
{<<"name">>, FileName},
{<<"disposition">>, <<"attachment">>}
],
disposition => <<"attachment">>,
disposition_params => [{<<"filename">>, FileName}]
},
<<"data">>}
]},
Result = <<
"From: k.petrauskas@erisata.lt\r\n"
"Subject: =?UTF-8?Q?=C4=8Ciobi=C5=A1kis?=\r\n"
"Date: Thu, 17 Dec 2020 20:12:33 +0200\r\n"
"Message-ID: <47a08b7ff7d305087877361ca8eea1db@karolis.erisata.lt>\r\n"
"Content-Type: multipart/mixed;\r\n"
"\tboundary=\"_=boundary-123=_\"\r\n"
"MIME-Version: 1.0\r\n"
"\r\n"
"\r\n"
"--_=boundary-123=_\r\n"
"Content-Type: application/pdf;\r\n"
"\tname=\"=?UTF-8?Q?=C4=8Cia_labai_ilgas_el=2E_lai=C5=A1ko_priedo_pavadinima?=\r\n"
"\t=?UTF-8?Q?s_su_lietuvi=C5=A1komis_ar_kokiomis_kitomis_ne_ascii_raid?=\r\n"
"\t=?UTF-8?Q?=C4=97mis=2Epdf?=\";\r\n"
"\tdisposition=attachment\r\n"
"Content-Disposition: attachment;\r\n"
"\tfilename=\"=?UTF-8?Q?=C4=8Cia_labai_ilgas_el=2E_lai=C5=A1ko_priedo_pavadi?=\r\n"
"\t=?UTF-8?Q?nimas_su_lietuvi=C5=A1komis_ar_kokiomis_kitomis_ne_ascii_raid?=\r\n"
"\t=?UTF-8?Q?=C4=97mis=2Epdf?=\"\r\n"
"\r\n"
"data\r\n"
"--_=boundary-123=_--\r\n"
>>,
?assertEqual(Result, encode(Email))
end},
{"Email with special chars in From", fun() ->
Email =
{<<"text">>, <<"plain">>,
[
{<<"From">>, <<"\"Admin & ' ( \\\"hallo\\\" ) ; , [ ] WS\" <a@example.com>">>},
{<<"Message-ID">>, <<"<abcd@example.com>">>},
{<<"MIME-Version">>, <<"1.0">>},
{<<"Date">>, <<"Sun, 01 Nov 2009 14:44:47 +0200">>}
],
#{}, <<"This is a plain message">>},
Result =
<<"From: \"Admin & ' ( \\\"hallo\\\" ) ; , [ ] WS\" <a@example.com>\r\nMessage-ID: <abcd@example.com>\r\nMIME-Version: 1.0\r\nDate: Sun, 01 Nov 2009 14:44:47 +0200\r\n\r\nThis is a plain message">>,
?assertEqual(Result, encode(Email))
end},
{"multipart/alternative email", fun() ->
Email =
{<<"multipart">>, <<"alternative">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"To">>, <<"you@example.com">>},
{<<"Subject">>, <<"This is a test">>},
{<<"MIME-Version">>, <<"1.0">>},
{<<"Content-Type">>, <<"multipart/alternative; boundary=wtf-123234234">>}
],
#{
content_type_params =>
[{<<"boundary">>, <<"wtf-123234234">>}],
disposition => <<"inline">>,
disposition_params => []
},
[
{<<"text">>, <<"plain">>,
[
{<<"Content-Type">>, <<"text/plain;charset=US-ASCII;format=flowed">>},
{<<"Content-Transfer-Encoding">>, <<"7bit">>}
],
#{
content_type_params =>
[
{<<"charset">>, <<"US-ASCII">>},
{<<"format">>, <<"flowed">>}
],
disposition => <<"inline">>,
disposition_params => []
},
<<"This message contains rich text.">>},
{<<"text">>, <<"html">>,
[
{<<"Content-Type">>, <<"text/html;charset=US-ASCII">>},
{<<"Content-Transfer-Encoding">>, <<"7bit">>}
],
#{
content_type_params =>
[{<<"charset">>, <<"US-ASCII">>}],
disposition => <<"inline">>,
disposition_params => []
},
<<"<html><body>This message also contains HTML</body></html>">>}
]},
Result = decode(encode(Email)),
?assertMatch(
{<<"multipart">>, <<"alternative">>, _, _, [
{<<"text">>, <<"plain">>, _, _, _}, {<<"text">>, <<"html">>, _, _, _}
]},
Result
)
end},
{"multipart/alternative email with encoding", fun() ->
Email =
{<<"multipart">>, <<"alternative">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"To">>, <<"you@example.com">>},
{<<"Subject">>, <<"This is a test">>},
{<<"MIME-Version">>, <<"1.0">>},
{<<"Content-Type">>, <<"multipart/alternative; boundary=wtf-123234234">>}
],
#{
content_type_params =>
[{<<"boundary">>, <<"wtf-123234234">>}],
disposition => <<"inline">>,
disposition_params => []
},
[
{<<"text">>, <<"plain">>,
[
{<<"Content-Type">>, <<"text/plain;charset=US-ASCII;format=flowed">>},
{<<"Content-Transfer-Encoding">>, <<"quoted-printable">>}
],
#{
content_type_params =>
[
{<<"charset">>, <<"US-ASCII">>},
{<<"format">>, <<"flowed">>}
],
disposition => <<"inline">>,
disposition_params => []
},
<<"This message contains rich text.\r\n", "and is =quoted printable= encoded!">>},
{<<"text">>, <<"html">>,
[
{<<"Content-Type">>, <<"text/html;charset=US-ASCII">>},
{<<"Content-Transfer-Encoding">>, <<"base64">>}
],
#{
content_type_params =>
[{<<"charset">>, <<"US-ASCII">>}],
disposition => <<"inline">>,
disposition_params => []
},
<<"<html><body>This message also contains", "HTML and is base64",
"encoded\r\n\r\n</body></html>">>}
]},
Result = decode(encode(Email)),
?assertMatch(
{<<"multipart">>, <<"alternative">>, _, _, [
{<<"text">>, <<"plain">>, _, _,
<<"This message contains rich text.\r\n", "and is =quoted printable= encoded!">>},
{<<"text">>, <<"html">>, _, _,
<<"<html><body>This message also contains", "HTML and is base64",
"encoded\r\n\r\n</body></html>">>}
]},
Result
)
end},
{"multipart/mixed email with multipart/alternative does not add an extra empty lines", fun() ->
Email = Getmail("message-text-html-attachment.eml"),
Encoded = encode(Email),
Re = re:run(Encoded, "(?:\\r\\n){3}", [global, {capture, all, binary}]),
?assertMatch({match, [_]}, Re)
end},
{"Missing headers should be added", fun() ->
Email =
{<<"text">>, <<"plain">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"To">>, <<"you@example.com">>},
{<<"Subject">>, <<"This is a test">>}
],
#{
content_type_params =>
[{<<"charset">>, <<"US-ASCII">>}],
disposition => <<"inline">>
},
<<"This is a plain message">>},
Result = decode(encode(Email)),
?assertNot(undefined == proplists:get_value(<<"Message-ID">>, element(3, Result))),
?assertNot(undefined == proplists:get_value(<<"Date">>, element(3, Result))),
?assertEqual(undefined, proplists:get_value(<<"References">>, element(3, Result)))
end},
{"Reference header should be added in presence of In-Reply-To", fun() ->
Email =
{<<"text">>, <<"plain">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"To">>, <<"you@example.com">>},
{<<"In-Reply-To">>, <<"<abcd@example.com>">>},
{<<"Subject">>, <<"This is a test">>}
],
#{
content_type_params =>
[{<<"charset">>, <<"US-ASCII">>}],
disposition => <<"inline">>
},
<<"This is a plain message">>},
Result = decode(encode(Email)),
?assertEqual(<<"<abcd@example.com>">>, proplists:get_value(<<"References">>, element(3, Result)))
end},
{"Reference header should be appended to in presence of In-Reply-To, if appropriate", fun() ->
Email =
{<<"text">>, <<"plain">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"To">>, <<"you@example.com">>},
{<<"In-Reply-To">>, <<"<abcd@example.com>">>},
{<<"References">>, <<"<wxyz@example.com>">>},
{<<"Subject">>, <<"This is a test">>}
],
#{
content_type_params =>
[{<<"charset">>, <<"US-ASCII">>}],
disposition => <<"inline">>
},
<<"This is a plain message">>},
Result = decode(encode(Email)),
?assertEqual(
<<"<wxyz@example.com> <abcd@example.com>">>, proplists:get_value(<<"References">>, element(3, Result))
)
end},
{"Reference header should NOT be appended to in presence of In-Reply-To, if already present", fun() ->
Email =
{<<"text">>, <<"plain">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"To">>, <<"you@example.com">>},
{<<"In-Reply-To">>, <<"<abcd@example.com>">>},
{<<"References">>, <<"<wxyz@example.com> <abcd@example.com>">>},
{<<"Subject">>, <<"This is a test">>}
],
#{
content_type_params =>
[{<<"charset">>, <<"US-ASCII">>}],
disposition => <<"inline">>
},
<<"This is a plain message">>},
Result = decode(encode(Email)),
?assertEqual(
<<"<wxyz@example.com> <abcd@example.com>">>, proplists:get_value(<<"References">>, element(3, Result))
)
end},
{"Content-Transfer-Encoding header should be added if missing and appropriate", fun() ->
Email =
{<<"text">>, <<"plain">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"To">>, <<"you@example.com">>},
{<<"Subject">>, <<"This is a test">>}
],
#{}, <<"This is a plain message with some non-ascii characters øÿ\r\nso there"/utf8>>},
Encoded = encode(Email),
Result = decode(Encoded),
?assertEqual(
<<"quoted-printable">>, proplists:get_value(<<"Content-Transfer-Encoding">>, element(3, Result))
),
Email2 =
{<<"text">>, <<"plain">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"To">>, <<"you@example.com">>},
{<<"Subject">>, <<"This is a test">>}
],
#{}, <<"This is a plain message with no non-ascii characters">>},
Encoded2 = encode(Email2),
Result2 = decode(Encoded2),
?assertEqual(undefined, proplists:get_value(<<"Content-Transfer-Encoding">>, element(3, Result2))),
Email3 =
{<<"text">>, <<"plain">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"To">>, <<"you@example.com">>},
{<<"Subject">>, <<"This is a test">>}
],
#{transfer_encoding => <<"base64">>}, <<"This is a plain message with no non-ascii characters">>},
Encoded3 = encode(Email3),
Result3 = decode(Encoded3),
?assertEqual(<<"base64">>, proplists:get_value(<<"Content-Transfer-Encoding">>, element(3, Result3)))
end},
{"Content-Type header should be added if missing and appropriate", fun() ->
Email =
{<<"text">>, <<"html">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"To">>, <<"you@example.com">>},
{<<"Subject">>, <<"This is a test">>}
],
#{}, <<"This is a HTML message with some non-ascii characters øÿ\r\nso there"/utf8>>},
Encoded = encode(Email),
Result = decode(Encoded),
?assertEqual(
<<"quoted-printable">>, proplists:get_value(<<"Content-Transfer-Encoding">>, element(3, Result))
),
?assertMatch(<<"text/html;charset=utf-8">>, proplists:get_value(<<"Content-Type">>, element(3, Result))),
Email2 =
{<<"text">>, <<"html">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"To">>, <<"you@example.com">>},
{<<"Subject">>, <<"This is a test">>}
],
#{}, <<"This is a HTML message with no non-ascii characters\r\nso there">>},
Encoded2 = encode(Email2),
Result2 = decode(Encoded2),
?assertMatch(
<<"text/html;charset=us-ascii">>, proplists:get_value(<<"Content-Type">>, element(3, Result2))
),
Email3 =
{<<"text">>, <<"html">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"To">>, <<"you@example.com">>},
{<<"Subject">>, <<"This is a test">>}
],
#{}, <<"This is a text message with some invisible non-ascii characters\r\nso there"/utf8>>},
Encoded3 = encode(Email3),
Result3 = decode(Encoded3),
?assertMatch(<<"text/html;charset=utf-8">>, proplists:get_value(<<"Content-Type">>, element(3, Result3)))
end},
{"Content-Type header should be added for subparts too, if missing and appropriate", fun() ->
Email4 =
{<<"multipart">>, <<"alternative">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"To">>, <<"you@example.com">>},
{<<"Subject">>, <<"This is a test">>}
],
#{}, [
{<<"text">>, <<"plain">>, [], #{},
<<"This is a multipart message with some invisible non-ascii characters\r\nso there"/utf8>>}
]},
Encoded4 = encode(Email4),
Result4 = decode(Encoded4),
?assertMatch(
<<"text/plain;charset=utf-8">>,
proplists:get_value(<<"Content-Type">>, element(3, lists:nth(1, element(5, Result4))))
)
end},
{"Content-Type header should be not added for subparts if they're text/plain us-ascii", fun() ->
Email4 =
{<<"multipart">>, <<"alternative">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"To">>, <<"you@example.com">>},
{<<"Subject">>, <<"This is a test">>}
],
#{}, [
{<<"text">>, <<"plain">>, [], #{},
<<"This is a multipart message with no non-ascii characters\r\nso there">>}
]},
Encoded4 = encode(Email4),
Result4 = decode(Encoded4),
?assertMatch(
undefined, proplists:get_value(<<"Content-Type">>, element(3, lists:nth(1, element(5, Result4))))
)
end},
{"Content-Type header should be added for subparts if they're text/html us-ascii", fun() ->
Email4 =
{<<"multipart">>, <<"alternative">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"To">>, <<"you@example.com">>},
{<<"Subject">>, <<"This is a test">>}
],
#{}, [
{<<"text">>, <<"html">>, [], #{},
<<"This is a multipart message with no non-ascii characters\r\nso there">>}
]},
Encoded4 = encode(Email4),
Result4 = decode(Encoded4),
?assertMatch(
<<"text/html;charset=us-ascii">>,
proplists:get_value(<<"Content-Type">>, element(3, lists:nth(1, element(5, Result4))))
)
end},
{"A boundary should be generated if applicable", fun() ->
Email =
{<<"multipart">>, <<"alternative">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"To">>, <<"you@example.com">>},
{<<"Subject">>, <<"This is a test">>}
],
#{}, [
{<<"text">>, <<"plain">>, [], #{},
<<"This message contains rich text.\r\n", "and is =quoted printable= encoded!">>},
{<<"text">>, <<"html">>, [], #{},
<<"<html><body>This message also contains", "HTML and is base64",
"encoded\r\n\r\n</body></html>">>}
]},
Encoded = encode(Email),
Result = decode(Encoded),
Boundary = proplists:get_value(<<"boundary">>, maps:get(content_type_params, element(4, Result))),
?assert(is_binary(Boundary)),
% ensure we don't add the header multiple times
?assertEqual(1, length(proplists:get_all_values(<<"Content-Type">>, element(3, Result)))),
% headers should be appended, not prepended
?assertMatch({<<"From">>, _}, lists:nth(1, element(3, Result))),
ok
end}
].
roundtrip_test_() ->
[
{"roundtrip test for the gamut", fun() ->
{ok, Email} = file:read_file("test/fixtures/the-gamut.eml"),
Decoded = decode(Email),
_Encoded = encode(Decoded),
%{ok, F1} = file:open("f1", [write]),
%{ok, F2} = file:open("f2", [write]),
%file:write(F1, Email),
%file:write(F2, Encoded),
%file:close(F1),
%file:close(F2),
%?assertEqual(Email, Email),
ok
end},
{"round trip plain text only email", fun() ->
{ok, Email} = file:read_file("test/fixtures/Plain-text-only.eml"),
Decoded = decode(Email),
_Encoded = encode(Decoded),
%{ok, F1} = file:open("f1", [write]),
%{ok, F2} = file:open("f2", [write]),
%file:write(F1, Email),
%file:write(F2, Encoded),
%file:close(F1),
%file:close(F2),
%?assertEqual(Email, Email),
ok
end},
{"round trip quoted-printable email", fun() ->
{ok, Email} = file:read_file("test/fixtures/testcase1"),
Decoded = decode(Email),
_Encoded = encode(Decoded),
%{ok, F1} = file:open("f1", [write]),
%{ok, F2} = file:open("f2", [write]),
%file:write(F1, Email),
%file:write(F2, Encoded),
%file:close(F1),
%file:close(F2),
%?assertEqual(Email, Email),
ok
end}
].
dkim_canonicalization_test_() ->
%% * canonicalization from #3.4.5
Hdrs = [
<<"A : X\r\n">>,
<<"B : Y\t\r\n\tZ \r\n">>
],
Body = <<" C \r\nD \t E\r\n\r\n\r\n">>,
[
{"Simple body canonicalization", fun() ->
?assertEqual(<<" C \r\nD \t E\r\n">>, dkim_canonicalize_body(Body, simple)),
?assertEqual(<<"\r\n">>, dkim_canonicalize_body(<<>>, simple)),
?assertEqual(<<"\r\n">>, dkim_canonicalize_body(<<"\r\n\r\n\r\n">>, simple)),
?assertEqual(<<"A\r\n\r\nB\r\n">>, dkim_canonicalize_body(<<"A\r\n\r\nB\r\n\r\n">>, simple))
end},
{"Simple headers canonicalization", fun() ->
?assertEqual(
[
<<"A : X\r\n">>,
<<"B : Y\t\r\n\tZ \r\n">>
],
dkim_canonicalize_headers(Hdrs, simple)
)
end},
{"Relaxed headers canonicalization", fun() ->
% \r\n's are stripped by current impl.
?assertEqual(
[
<<"a:X">>,
<<"b:Y Z">>
],
dkim_canonicalize_headers(Hdrs, relaxed)
)
end}
].
dkim_sign_rsa_test_() ->
%% * sign using test/fixtures/dkim*.pem
{ok, PrivKey} = file:read_file("test/fixtures/dkim-rsa-private.pem"),
[
{"Sign simple", fun() ->
Email =
{<<"text">>, <<"plain">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"Subject">>, <<"Hello world!">>},
{<<"Date">>, <<"Thu, 28 Nov 2013 04:15:44 +0400">>},
{<<"Message-ID">>, <<"the-id">>},
{<<"Content-Type">>, <<"text/plain; charset=utf-8">>}
],
#{}, <<"123">>},
Options = [
{dkim, [
{s, <<"foo.bar">>},
{d, <<"example.com">>},
{c, {simple, simple}},
{t, {{2014, 2, 4}, {23, 15, 00}}},
{x, {{2114, 2, 4}, {23, 15, 00}}},
{private_key, {pem_plain, PrivKey}}
]}
],
Enc = encode(Email, Options),
%% This `Enc' value can be verified, for example, by Python script
%% https://launchpad.net/dkimpy like:
%% >>> pubkey = ''.join(open("test/fixtures/dkim-rsa-public.pem").read().splitlines()[1:-1])
%% >>> dns_mock = lambda *args: 'v=DKIM1; g=*; k=rsa; p=' + pubkey
%% >>> import dkim
%% >>> d = dkim.DKIM(mime_message) % pass `Enc' value as 1'st argument
%% >>> d.verify(dnsfunc=dns_mock)
%% True
{_, _, [{DkimHdrName, DkimHdrVal} | _], _, _} = decode(Enc),
?assertEqual(<<"DKIM-Signature">>, DkimHdrName),
?assertEqual(
<<
"t=1391555700; x=4547229300; s=foo.bar; h=from:to:subject:date; d=example.com; c=simple/simple; "
"bh=Afm/S7SaxS19en1h955RwsupTF914DQUPqYU8Nh7kpw=; a=rsa-sha256; v=1; "
"b=Mtja7WpVvtOFT8rfzOS/2fRZ492jrgsHgD5YUl5zmPQ/NEEMjVhVX0JCkfZxWpxiKe"
"qwl7nTJy3xecdg12feGT1rGC+rV0vAX8LVc+AJ4T4A50hE8L4hpJ1Tv5rt2O2t0Xu1Wx"
"yH6Cmrhhh56istjL+ba+U1EHhV7uZXGpWXGa4="
>>,
DkimHdrVal
)
end},
{"Sign relaxed headers, simple body", fun() ->
Email =
{<<"text">>, <<"plain">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"Subject">>, <<"Hello world!">>},
{<<"Date">>, <<"Thu, 28 Nov 2013 04:15:44 +0400">>},
{<<"Message-ID">>, <<"the-id-relaxed">>},
{<<"Content-Type">>, <<"text/plain; charset=utf-8">>}
],
#{}, <<"123">>},
Options = [
{dkim, [
{s, <<"foo.bar">>},
{d, <<"example.com">>},
{c, {relaxed, simple}},
{private_key, {pem_plain, PrivKey}}
]}
],
Enc = encode(Email, Options),
{_, _, [{DkimHdrName, DkimHdrVal} | _], _, _} = decode(Enc),
?assertEqual(<<"DKIM-Signature">>, DkimHdrName),
?assertEqual(
<<
"s=foo.bar; h=from:to:subject:date; d=example.com; c=relaxed/simple; "
"bh=Afm/S7SaxS19en1h955RwsupTF914DQUPqYU8Nh7kpw=; a=rsa-sha256; v=1; "
"b=dXxKq6A7m4A3AoS90feuLP+IxOyXFTPIibja52E2JCAyOsxvIGlI51xR1LvmEaelv9"
"jJTH9iGyAC7RzTKxrWV1QXayvr05bsTy3vDw7P4vfZ1gmspuP/3Icw+J8KEn+p6+CRrf"
"T97QadH42PT6XmO2v01q5nhMgNE4yQyf9DBJs="
>>,
DkimHdrVal
)
end}
].
dkim_sign_ed25519_test_() ->
case ed25519_supported() of
true ->
%% * sign using test/fixtures/dkim*.pem
{ok, PrivKey} = file:read_file("test/fixtures/dkim-ed25519-private.pem"),
[
{"Sign simple", fun() ->
Email =
{<<"text">>, <<"plain">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"Subject">>, <<"Hello world!">>},
{<<"Date">>, <<"Thu, 28 Nov 2013 04:15:44 +0400">>},
{<<"Message-ID">>, <<"the-id">>},
{<<"Content-Type">>, <<"text/plain; charset=utf-8">>}
],
#{}, <<"123">>},
Options = [
{dkim, [
{s, <<"foo.bar">>},
{d, <<"example.com">>},
{c, {simple, simple}},
{a, 'ed25519-sha256'},
{t, {{2014, 2, 4}, {23, 15, 00}}},
{x, {{2114, 2, 4}, {23, 15, 00}}},
{private_key, {pem_plain, PrivKey}}
]}
],
Enc = encode(Email, Options),
%% This `Enc' value can be verified, for example, by Python script
%% https://launchpad.net/dkimpy like:
%% >>> pubkey = ''.join(open("test/fixtures/dkim-ed25519-public.pem").read().splitlines()[1:-1])
%% >>> dns_mock = lambda *args: 'v=DKIM1; g=*; k=ed25519; p=' + pubkey
%% >>> import dkim
%% >>> d = dkim.DKIM(mime_message) % pass `Enc' value as 1'st argument
%% >>> d.verify(dnsfunc=dns_mock)
%% True
{_, _, [{DkimHdrName, DkimHdrVal} | _], _, _} = decode(Enc),
?assertEqual(<<"DKIM-Signature">>, DkimHdrName),
?assertEqual(
<<
"t=1391555700; x=4547229300; s=foo.bar; h=from:to:subject:date; d=example.com; c=simple/simple; "
"bh=Afm/S7SaxS19en1h955RwsupTF914DQUPqYU8Nh7kpw=; a=ed25519-sha256; v=1; "
"b=bFPndkFlgpFbfVKBF9HiVkQQF/3ojOQT7ycrZYp0yYe4oyItUQexlvd+Q7BviiHv/seLVBESpBjLbthbfb5HDA=="
>>,
DkimHdrVal
)
end},
{"Sign relaxed headers, simple body", fun() ->
Email =
{<<"text">>, <<"plain">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"Subject">>, <<"Hello world!">>},
{<<"Date">>, <<"Thu, 28 Nov 2013 04:15:44 +0400">>},
{<<"Message-ID">>, <<"the-id-relaxed">>},
{<<"Content-Type">>, <<"text/plain; charset=utf-8">>}
],
#{}, <<"123">>},
Options = [
{dkim, [
{s, <<"foo.bar">>},
{d, <<"example.com">>},
{c, {relaxed, simple}},
{a, 'ed25519-sha256'},
{private_key, {pem_plain, PrivKey}}
]}
],
Enc = encode(Email, Options),
{_, _, [{DkimHdrName, DkimHdrVal} | _], _, _} = decode(Enc),
?assertEqual(<<"DKIM-Signature">>, DkimHdrName),
?assertEqual(
<<
"s=foo.bar; h=from:to:subject:date; d=example.com; c=relaxed/simple; "
"bh=Afm/S7SaxS19en1h955RwsupTF914DQUPqYU8Nh7kpw=; a=ed25519-sha256; v=1; "
"b=f7wORU/qmPr4q891m5zmZMadPm9n9e596mBJHBD6tE51PAl4pHdpw9xRC1kwLGmxPTEK5SiQluPVTbDHVhVZBQ=="
>>,
DkimHdrVal
)
end}
];
false ->
[]
end.
dkim_sign_ed25519_encrypted_key_test_() ->
case ed25519_supported() of
true ->
%% * sign using test/fixtures/dkim*.pem
{ok, EncryptedPrivKey} =
file:read_file("test/fixtures/dkim-ed25519-encrypted-private.pem"),
[
{"Sign encrypted", fun() ->
Email =
{<<"text">>, <<"plain">>,
[
{<<"From">>, <<"me@example.com">>},
{<<"Subject">>, <<"Hello world!">>},
{<<"Date">>, <<"Thu, 28 Nov 2013 04:15:44 +0400">>},
{<<"Message-ID">>, <<"the-id">>},
{<<"Content-Type">>, <<"text/plain; charset=utf-8">>}
],
#{}, <<"123">>},
Options = [
{dkim, [
{s, <<"foo.bar">>},
{d, <<"example.com">>},
{c, {simple, simple}},
{a, 'ed25519-sha256'},
{t, {{2014, 2, 4}, {23, 15, 00}}},
{x, {{2114, 2, 4}, {23, 15, 00}}},
{private_key, {pem_encrypted, EncryptedPrivKey, "password"}}
]}
],
Enc = encode(Email, Options),
%% This `Enc' value can be verified, for example, by Python script
%% https://launchpad.net/dkimpy like:
%% >>> pubkey = ''.join(open("test/fixtures/dkim-ed25519-public.pem").read().splitlines()[1:-1])
%% >>> dns_mock = lambda *args: 'v=DKIM1; g=*; k=ed25519; p=' + pubkey
%% >>> import dkim
%% >>> d = dkim.DKIM(mime_message) % pass `Enc' value as 1'st argument
%% >>> d.verify(dnsfunc=dns_mock)
%% True
{_, _, [{DkimHdrName, DkimHdrVal} | _], _, _} = decode(Enc),
?assertEqual(<<"DKIM-Signature">>, DkimHdrName),
?assertEqual(
<<
"t=1391555700; x=4547229300; s=foo.bar; h=from:to:subject:date; d=example.com; c=simple/simple; "
"bh=Afm/S7SaxS19en1h955RwsupTF914DQUPqYU8Nh7kpw=; a=ed25519-sha256; v=1; "
"b=JgsuW5OmKPk188YRmxs1cLA8mrAf9FNC+s/PYK7Vat7HF4l7FglcoWWHqm0/Cg7o/V+8bP1RNwes1xDKS8/wDQ=="
>>,
DkimHdrVal
)
end}
];
false ->
[]
end.
-endif.