1 # Copyright (C) 2020 all contributors <meta@public-inbox.org>
2 # License: AGPL-3.0+ <https://www.gnu.org/licenses/agpl-3.0.txt>
4 # Lazy MIME parser, it still slurps the full message but keeps short
5 # lifetimes. Unlike Email::MIME, it doesn't pre-split multipart
6 # messages or do any up-front parsing of headers besides splitting
7 # the header string from the body.
9 # Contains ideas and code from Email::Simple and Email::MIME
10 # (Perl Artistic License, GPL-1+)
12 # This aims to replace Email::MIME for our purposes, similar API
13 # but internal field names are differ if they're not 100%-compatible.
15 # Includes some proposed fixes for Email::MIME:
16 # - header-less sub parts - https://github.com/rjbs/Email-MIME/issues/14
17 # - "0" as boundary - https://github.com/rjbs/Email-MIME/issues/63
20 # bdy => scalar ref for body (may be undef),
21 # hdr => scalar ref for header,
22 # crlf => "\n" or "\r\n" (scalar, not a ref),
24 # # filled in during ->each_part
25 # ct => hash ref returned by parse_content_type
27 package PublicInbox::Eml;
31 use Encode qw(find_encoding decode encode); # stdlib
32 use Text::Wrap qw(wrap); # stdlib, we need Perl 5.6+ for $huge
33 use MIME::Base64 3.05; # Perl 5.10.0 / 5.9.2
34 use MIME::QuotedPrint 3.05; # ditto
36 my $MIME_Header = find_encoding('MIME-Header');
38 use PublicInbox::EmlContentFoo qw(parse_content_type parse_content_disposition);
39 $PublicInbox::EmlContentFoo::STRICT_PARAMS = 0;
41 our $MAXPARTS = 1000; # same as SpamAssassin
42 our $MAXDEPTH = 20; # seems enough, Perl sucks, here
43 our $MAXBOUNDLEN = 2048; # same as postfix
45 my %MIME_ENC = (qp => \&enc_qp, base64 => \&encode_base64);
46 my %MIME_DEC = (qp => \&dec_qp, base64 => \&decode_base64);
47 $MIME_ENC{quotedprint} = $MIME_ENC{'quoted-printable'} = $MIME_ENC{qp};
48 $MIME_DEC{quotedprint} = $MIME_DEC{'quoted-printable'} = $MIME_DEC{qp};
49 $MIME_ENC{$_} = \&identity_codec for qw(7bit 8bit binary);
51 my %DECODE_ADDRESS = map { $_ => 1 } qw(From To Cc Sender Reply-To);
54 'Content-Description' => 1,
55 'Content-Type' => 1, # not correct, but needed, oh well
57 our %STR_TYPE = (text => 1);
58 our %STR_SUBTYPE = (plain => 1, html => 1);
63 # Do not normalize $k with lc/uc; instead strive to keep
64 # capitalization in our codebase consistent.
65 $re_memo{$k} ||= qr/^\Q$k\E:[ \t]*([^\n]*\r?\n # 1st line
67 (?:[^:\n]*?[ \t]+[^\n]*\r?\n)*)
71 # compatible with our uses of Email::MIME
73 my $ref = ref($_[1]) ? $_[1] : \(my $cpy = $_[1]);
74 # substr() can modify the first arg in-place and to avoid
75 # memcpy/memmove on a potentially large scalar. It does need
76 # to make a copy for $hdr, though. Idea stolen from Email::Simple.
78 # We also prefer index() on common LFLF emails since it's faster
79 # and re scan can bump RSS by length($$ref) on big strings
80 if (index($$ref, "\r\n") < 0 && (my $pos = index($$ref, "\n\n")) >= 0) {
82 my $hdr = substr($$ref, 0, $pos + 2, ''); # sv_chop on $$ref
83 chop($hdr); # lower SvCUR
84 bless { hdr => \$hdr, crlf => "\n", bdy => $ref }, __PACKAGE__;
85 } elsif ($$ref =~ /\r?\n(\r?\n)/s) {
86 my $hdr = substr($$ref, 0, $+[0], ''); # sv_chop on $$ref
87 substr($hdr, -(length($1))) = ''; # lower SvCUR
88 bless { hdr => \$hdr, crlf => $1, bdy => $ref }, __PACKAGE__;
89 } elsif ($$ref =~ /^[a-z0-9-]+[ \t]*:/ims && $$ref =~ /(\r?\n)\z/s) {
91 bless { hdr => \($$ref), crlf => $1 }, __PACKAGE__;
92 } else { # nothing useful
94 bless { hdr => \$hdr, crlf => "\n" }, __PACKAGE__;
99 my (undef, $ref) = @_;
100 # special case for messages like <85k5su9k59.fsf_-_@lola.goethe.zz>
101 $$ref =~ /\A(\r?\n)/s or goto &new;
102 my $hdr = substr($$ref, 0, $+[0], ''); # sv_chop on $$ref
103 bless { hdr => \$hdr, crlf => $1, bdy => $ref }, __PACKAGE__;
106 # same output as Email::Simple::Header::header_raw, but we extract
107 # headers on-demand instead of parsing them into a list which
108 # requires O(n) lookups anyways
110 my $re = re_memo($_[1]);
111 my @v = (${ $_[0]->{hdr} } =~ /$re/g);
113 # for compatibility w/ Email::Simple::Header,
118 wantarray ? @v : $v[0];
121 # pick the first Content-Type header to match Email::MIME behavior.
122 # It's usually the right one based on historical archives.
124 # PublicInbox::EmlContentFoo::content_type:
125 $_[0]->{ct} //= parse_content_type(header($_[0], 'Content-Type'));
128 # returns a queue of sub-parts iff it's worth descending into
129 # TODO: descend into message/rfc822 parts (Email::MIME didn't)
130 sub mp_descend ($$) {
131 my ($self, $nr) = @_; # or $once for top-level
132 my $bnd = ct($self)->{attributes}->{boundary} // return; # single-part
133 return if $bnd eq '' || length($bnd) >= $MAXBOUNDLEN;
134 $bnd = quotemeta($bnd);
136 # "multipart" messages can exist w/o a body
137 my $bdy = ($nr ? delete($self->{bdy}) : \(body_raw($self))) or return;
139 # Cut at the the first epilogue, not subsequent ones.
140 # *sigh* just the regexp match alone seems to bump RSS by
141 # length($$bdy) on a ~30M string:
142 my $epilogue_missing;
143 if ($$bdy =~ /(?:\r?\n)?^--$bnd--[ \t]*\r?$/sm) {
144 substr($$bdy, $-[0]) = '';
146 $epilogue_missing = 1;
149 # *Sigh* split() doesn't work in-place and return CoW strings
150 # because Perl wants to "\0"-terminate strings. So split()
151 # again bumps RSS by length($$bdy)
153 # Quiet warning for "Complex regular subexpression recursion limit"
154 # in case we get many empty parts, it's harmless in this case
155 no warnings 'regexp';
156 my ($pre, @parts) = split(/(?:\r?\n)?(?:^--$bnd[ \t]*\r?\n)+/ms,
158 # + 3 since we don't want the last part
159 # processed to include any other excluded
160 # parts ($nr starts at 1, and I suck at math)
161 $MAXPARTS + 3 - $nr);
163 if (@parts) { # the usual path if we got this far:
164 undef $bdy; # release memory ASAP if $nr > 0
166 # compatibility with Email::MIME
167 $parts[-1] =~ s/\n\r?\n\z/\n/s if $epilogue_missing;
169 @parts = grep /[^ \t\r\n]/s, @parts; # ignore empty parts
171 # Keep "From: someone..." from preamble in old,
172 # buggy versions of git-send-email, otherwise drop it
173 # There's also a case where quoted text showed up in the
175 # <20060515162817.65F0F1BBAE@citi.umich.edu>
176 unshift(@parts, $pre) if $pre =~ /:/s;
179 # "multipart", but no boundary found, treat as single part
180 $self->{bdy} //= $bdy;
184 # $p = [ \@parts, $depth, $idx ]
185 # $idx[0] grows as $depth grows, $idx[1] == $p->[-1] == current part
186 # (callers need to be updated)
187 # \@parts is a queue which empties when we're done with a parent part
189 # same usage as PublicInbox::MsgIter::msg_iter
190 # $cb - user-supplied callback sub
191 # $arg - user-supplied arg (think pthread_create)
192 # $once - unref body scalar during iteration
194 my ($self, $cb, $arg, $once) = @_;
195 my $p = mp_descend($self, $once // 0) or
196 return $cb->([$self, 0, 0], $arg);
198 my @s; # our virtual stack
200 while ((scalar(@{$p->[0]}) || ($p = pop @s)) && ++$nr <= $MAXPARTS) {
201 ++$p->[-1]; # bump index
202 my (undef, @idx) = @$p;
203 @idx = (join('.', @idx));
204 my $depth = ($idx[0] =~ tr/././) + 1;
205 my $sub = new_sub(undef, \(shift @{$p->[0]}));
206 if ($depth < $MAXDEPTH && (my $nxt = mp_descend($sub, $nr))) {
207 push(@s, $p) if scalar @{$p->[0]};
208 $p = [ $nxt, @idx, 0 ];
209 } else { # a leaf node
210 $cb->([$sub, $depth, @idx], $arg);
216 # prevent MIME::QuotedPrint from encoding CR as =0D since it's
217 # against RFCs and breaks MUAs
218 $_[0] =~ s/\r\n/\n/sg;
219 encode_qp($_[0], "\r\n");
223 # RFC 2822 requires all lines to end in CRLF, though... :<
224 $_[0] = decode_qp($_[0]);
225 $_[0] =~ s/\n/\r\n/sg;
229 sub identity_codec { $_[0] }
231 ########### compatibility section for existing Email::MIME uses #########
234 bless { hdr => $_[0]->{hdr}, crlf => $_[0]->{crlf} }, __PACKAGE__;
239 my $parts = mp_descend($self, 0) or return ();
240 my $bnd = ct($self)->{attributes}->{boundary} // die 'BUG: no boundary';
241 my $bdy = $self->{bdy};
242 if ($$bdy =~ /\A(.*?)(?:\r?\n)?^--\Q$bnd\E[ \t]*\r?$/sm) {
243 $self->{preamble} = $1;
245 if ($$bdy =~ /^--\Q$bnd\E--[ \t]*\r?\n(.+)\z/sm) {
246 $self->{epilogue} = $1;
248 map { new_sub(undef, \$_) } @$parts;
252 my ($self, $parts) = @_;
254 # we can't fully support what Email::MIME does,
255 # just what our filter code needs:
256 my $bnd = ct($self)->{attributes}->{boundary} // die <<EOF;
257 ->parts_set not supported for single-part messages
259 my $crlf = $self->{crlf};
260 my $fin_bnd = "$crlf--$bnd--$crlf";
261 $bnd = "$crlf--$bnd$crlf";
262 ${$self->{bdy}} = join($bnd,
263 delete($self->{preamble}) // '',
264 map { $_->as_string } @$parts
267 (delete($self->{epilogue}) // '');
272 my ($self, $body) = @_;
273 my $bdy = $self->{bdy} = ref($body) ? $body : \$body;
274 if (my $cte = header_raw($self, 'Content-Transfer-Encoding')) {
275 my $enc = $MIME_ENC{lc($cte)} or croak("can't encode `$cte'");
276 $$bdy = $enc->($$bdy); # in-place
282 my ($self, $body_str) = @_;
283 my $charset = ct($self)->{attributes}->{charset} or
284 Carp::confess('body_str was given, but no charset is defined');
285 body_set($self, \(encode($charset, $body_str, Encode::FB_CROAK)));
288 sub content_type { scalar header($_[0], 'Content-Type') }
290 # we only support raw header_set
292 my ($self, $pfx, @vals) = @_;
293 my $re = re_memo($pfx);
294 my $hdr = $self->{hdr};
295 return $$hdr =~ s!$re!!g if !@vals;
297 my $len = 78 - length($pfx);
299 # folding differs from Email::Simple::Header,
300 # we favor tabs for visibility (and space savings :P)
301 if (length($_) >= $len && (/\n[^ \t]/s || !/\n/s)) {
302 local $Text::Wrap::columns = $len;
303 local $Text::Wrap::huge = 'overflow';
304 $pfx . wrap('', "\t", $_) . $self->{crlf};
306 $pfx . $_ . $self->{crlf};
309 $$hdr =~ s!$re!shift(@vals) // ''!ge; # replace current headers, first
310 $$hdr .= join('', @vals); # append any leftovers not replaced
311 # wantarray ? @_[2..$#_] : $_[2]; # Email::Simple::Header compat
312 undef; # we don't care for the return value
315 # note: we only call this method on Subject
317 my ($self, $name, @vals) = @_;
319 next unless /[^\x20-\x7e]/;
320 utf8::encode($_); # to octets
321 # 39: int((75 - length("Subject: =?UTF-8?B?".'?=') ) / 4) * 3;
322 s/(.{1,39})/'=?UTF-8?B?'.encode_base64($1, '').'?='/ges;
324 header_set($self, $name, @vals);
327 sub mhdr_decode ($) { eval { $MIME_Header->decode($_[0]) } // $_[0] }
330 my $dis = header_raw($_[0], 'Content-Disposition');
331 my $attrs = parse_content_disposition($dis)->{attributes};
332 my $fn = $attrs->{filename};
333 $fn = ct($_[0])->{attributes}->{name} if !defined($fn) || $fn eq '';
334 (defined($fn) && $fn =~ /=\?/) ? mhdr_decode($fn) : $fn;
337 sub xs_addr_str { # helper for ->header / ->header_str
338 for (@_) { # array from header_raw()
340 my @g = parse_email_groups($_); # [ foo => [ E::A::X, ... ]
341 for (my $i = 0; $i < @g; $i += 2) {
342 if (defined($g[$i]) && $g[$i] =~ /=\?/) {
343 $g[$i] = mhdr_decode($g[$i]);
345 my $addrs = $g[$i + 1];
346 for my $eax (@$addrs) {
347 for my $m (qw(phrase comment)) {
349 $eax->$m(mhdr_decode($v)) if
354 $_ = format_email_groups(@g);
359 require Email::Address::XS;
360 Email::Address::XS->import(qw(parse_email_groups format_email_groups));
363 # fallback to just decoding everything, because parsing
364 # email addresses correctly w/o C/XS is slow
365 %DECODE_FULL = (%DECODE_FULL, %DECODE_ADDRESS);
366 %DECODE_ADDRESS = ();
369 *header = \&header_str;
371 my ($self, $name) = @_;
372 my @v = header_raw($self, $name);
373 if ($DECODE_ADDRESS{$name}) {
375 } elsif ($DECODE_FULL{$name}) {
377 $_ = mhdr_decode($_) if /=\?/;
380 wantarray ? @v : $v[0];
383 sub body_raw { ${$_[0]->{bdy} // \''}; }
386 my $raw = body_raw($_[0]);
387 my $cte = header_raw($_[0], 'Content-Transfer-Encoding') or return $raw;
388 ($cte) = ($cte =~ /([a-zA-Z0-9\-]+)/) or return $raw; # For S/MIME, etc
389 my $dec = $MIME_DEC{lc($cte)} or return $raw;
396 my $charset = $ct->{attributes}->{charset};
398 if ($STR_TYPE{$ct->{type}} && $STR_SUBTYPE{$ct->{subtype}}) {
401 Carp::confess("can't get body as a string for ",
402 join("\n\t", header_raw($self, 'Content-Type')));
404 decode($charset, body($self), Encode::FB_CROAK);
409 my $ret = ${ $self->{hdr} };
410 return $ret unless defined($self->{bdy});
411 $ret .= $self->{crlf};
412 $ret .= ${$self->{bdy}};
415 # Unlike Email::MIME::charset_set, this only changes the parsed
416 # representation of charset used for search indexing and HTML display.
417 # This does NOT affect what ->as_string returns.
419 ct($_[0])->{attributes}->{charset} = $_[1];
422 sub crlf { $_[0]->{crlf} // "\n" }
424 sub willneed { re_memo($_) for @_ }
426 willneed(qw(From To Cc Date Subject Content-Type In-Reply-To References
427 Message-ID X-Alt-Message-ID));