1# Copyright (C) 2002-2006 Python Software Foundation
2# Author: Ben Gertzfield, Barry Warsaw
3# Contact: email-sig@python.org
4
5"""Header encoding and decoding functionality."""
6
7__all__ = [
8    'Header',
9    'decode_header',
10    'make_header',
11    ]
12
13import re
14import binascii
15
16import email.quoprimime
17import email.base64mime
18
19from email.errors import HeaderParseError
20from email.charset import Charset
21
22NL = '\n'
23SPACE = ' '
24USPACE = u' '
25SPACE8 = ' ' * 8
26UEMPTYSTRING = u''
27
28MAXLINELEN = 76
29
30USASCII = Charset('us-ascii')
31UTF8 = Charset('utf-8')
32
33# Match encoded-word strings in the form =?charset?q?Hello_World?=
34ecre = re.compile(r'''
35  =\?                   # literal =?
36  (?P<charset>[^?]*?)   # non-greedy up to the next ? is the charset
37  \?                    # literal ?
38  (?P<encoding>[qb])    # either a "q" or a "b", case insensitive
39  \?                    # literal ?
40  (?P<encoded>.*?)      # non-greedy up to the next ?= is the encoded string
41  \?=                   # literal ?=
42  (?=[ \t]|$)           # whitespace or the end of the string
43  ''', re.VERBOSE | re.IGNORECASE | re.MULTILINE)
44
45# Field name regexp, including trailing colon, but not separating whitespace,
46# according to RFC 2822.  Character range is from tilde to exclamation mark.
47# For use with .match()
48fcre = re.compile(r'[\041-\176]+:$')
49
50# Find a header embedded in a putative header value.  Used to check for
51# header injection attack.
52_embeded_header = re.compile(r'\n[^ \t]+:')
53
54
55
56# Helpers
57_max_append = email.quoprimime._max_append
58
59
60
61def decode_header(header):
62    """Decode a message header value without converting charset.
63
64    Returns a list of (decoded_string, charset) pairs containing each of the
65    decoded parts of the header.  Charset is None for non-encoded parts of the
66    header, otherwise a lower-case string containing the name of the character
67    set specified in the encoded string.
68
69    An email.errors.HeaderParseError may be raised when certain decoding error
70    occurs (e.g. a base64 decoding exception).
71    """
72    # If no encoding, just return the header
73    header = str(header)
74    if not ecre.search(header):
75        return [(header, None)]
76    decoded = []
77    dec = ''
78    for line in header.splitlines():
79        # This line might not have an encoding in it
80        if not ecre.search(line):
81            decoded.append((line, None))
82            continue
83        parts = ecre.split(line)
84        while parts:
85            unenc = parts.pop(0).strip()
86            if unenc:
87                # Should we continue a long line?
88                if decoded and decoded[-1][1] is None:
89                    decoded[-1] = (decoded[-1][0] + SPACE + unenc, None)
90                else:
91                    decoded.append((unenc, None))
92            if parts:
93                charset, encoding = [s.lower() for s in parts[0:2]]
94                encoded = parts[2]
95                dec = None
96                if encoding == 'q':
97                    dec = email.quoprimime.header_decode(encoded)
98                elif encoding == 'b':
99                    paderr = len(encoded) % 4   # Postel's law: add missing padding
100                    if paderr:
101                        encoded += '==='[:4 - paderr]
102                    try:
103                        dec = email.base64mime.decode(encoded)
104                    except binascii.Error:
105                        # Turn this into a higher level exception.  BAW: Right
106                        # now we throw the lower level exception away but
107                        # when/if we get exception chaining, we'll preserve it.
108                        raise HeaderParseError
109                if dec is None:
110                    dec = encoded
111
112                if decoded and decoded[-1][1] == charset:
113                    decoded[-1] = (decoded[-1][0] + dec, decoded[-1][1])
114                else:
115                    decoded.append((dec, charset))
116            del parts[0:3]
117    return decoded
118
119
120
121def make_header(decoded_seq, maxlinelen=None, header_name=None,
122                continuation_ws=' '):
123    """Create a Header from a sequence of pairs as returned by decode_header()
124
125    decode_header() takes a header value string and returns a sequence of
126    pairs of the format (decoded_string, charset) where charset is the string
127    name of the character set.
128
129    This function takes one of those sequence of pairs and returns a Header
130    instance.  Optional maxlinelen, header_name, and continuation_ws are as in
131    the Header constructor.
132    """
133    h = Header(maxlinelen=maxlinelen, header_name=header_name,
134               continuation_ws=continuation_ws)
135    for s, charset in decoded_seq:
136        # None means us-ascii but we can simply pass it on to h.append()
137        if charset is not None and not isinstance(charset, Charset):
138            charset = Charset(charset)
139        h.append(s, charset)
140    return h
141
142
143
144class Header:
145    def __init__(self, s=None, charset=None,
146                 maxlinelen=None, header_name=None,
147                 continuation_ws=' ', errors='strict'):
148        """Create a MIME-compliant header that can contain many character sets.
149
150        Optional s is the initial header value.  If None, the initial header
151        value is not set.  You can later append to the header with .append()
152        method calls.  s may be a byte string or a Unicode string, but see the
153        .append() documentation for semantics.
154
155        Optional charset serves two purposes: it has the same meaning as the
156        charset argument to the .append() method.  It also sets the default
157        character set for all subsequent .append() calls that omit the charset
158        argument.  If charset is not provided in the constructor, the us-ascii
159        charset is used both as s's initial charset and as the default for
160        subsequent .append() calls.
161
162        The maximum line length can be specified explicit via maxlinelen.  For
163        splitting the first line to a shorter value (to account for the field
164        header which isn't included in s, e.g. `Subject') pass in the name of
165        the field in header_name.  The default maxlinelen is 76.
166
167        continuation_ws must be RFC 2822 compliant folding whitespace (usually
168        either a space or a hard tab) which will be prepended to continuation
169        lines.
170
171        errors is passed through to the .append() call.
172        """
173        if charset is None:
174            charset = USASCII
175        if not isinstance(charset, Charset):
176            charset = Charset(charset)
177        self._charset = charset
178        self._continuation_ws = continuation_ws
179        cws_expanded_len = len(continuation_ws.replace('\t', SPACE8))
180        # BAW: I believe `chunks' and `maxlinelen' should be non-public.
181        self._chunks = []
182        if s is not None:
183            self.append(s, charset, errors)
184        if maxlinelen is None:
185            maxlinelen = MAXLINELEN
186        if header_name is None:
187            # We don't know anything about the field header so the first line
188            # is the same length as subsequent lines.
189            self._firstlinelen = maxlinelen
190        else:
191            # The first line should be shorter to take into account the field
192            # header.  Also subtract off 2 extra for the colon and space.
193            self._firstlinelen = maxlinelen - len(header_name) - 2
194        # Second and subsequent lines should subtract off the length in
195        # columns of the continuation whitespace prefix.
196        self._maxlinelen = maxlinelen - cws_expanded_len
197
198    def __str__(self):
199        """A synonym for self.encode()."""
200        return self.encode()
201
202    def __unicode__(self):
203        """Helper for the built-in unicode function."""
204        uchunks = []
205        lastcs = None
206        for s, charset in self._chunks:
207            # We must preserve spaces between encoded and non-encoded word
208            # boundaries, which means for us we need to add a space when we go
209            # from a charset to None/us-ascii, or from None/us-ascii to a
210            # charset.  Only do this for the second and subsequent chunks.
211            nextcs = charset
212            if uchunks:
213                if lastcs not in (None, 'us-ascii'):
214                    if nextcs in (None, 'us-ascii'):
215                        uchunks.append(USPACE)
216                        nextcs = None
217                elif nextcs not in (None, 'us-ascii'):
218                    uchunks.append(USPACE)
219            lastcs = nextcs
220            uchunks.append(unicode(s, str(charset)))
221        return UEMPTYSTRING.join(uchunks)
222
223    # Rich comparison operators for equality only.  BAW: does it make sense to
224    # have or explicitly disable <, <=, >, >= operators?
225    def __eq__(self, other):
226        # other may be a Header or a string.  Both are fine so coerce
227        # ourselves to a string, swap the args and do another comparison.
228        return other == self.encode()
229
230    def __ne__(self, other):
231        return not self == other
232
233    def append(self, s, charset=None, errors='strict'):
234        """Append a string to the MIME header.
235
236        Optional charset, if given, should be a Charset instance or the name
237        of a character set (which will be converted to a Charset instance).  A
238        value of None (the default) means that the charset given in the
239        constructor is used.
240
241        s may be a byte string or a Unicode string.  If it is a byte string
242        (i.e. isinstance(s, str) is true), then charset is the encoding of
243        that byte string, and a UnicodeError will be raised if the string
244        cannot be decoded with that charset.  If s is a Unicode string, then
245        charset is a hint specifying the character set of the characters in
246        the string.  In this case, when producing an RFC 2822 compliant header
247        using RFC 2047 rules, the Unicode string will be encoded using the
248        following charsets in order: us-ascii, the charset hint, utf-8.  The
249        first character set not to provoke a UnicodeError is used.
250
251        Optional `errors' is passed as the third argument to any unicode() or
252        ustr.encode() call.
253        """
254        if charset is None:
255            charset = self._charset
256        elif not isinstance(charset, Charset):
257            charset = Charset(charset)
258        # If the charset is our faux 8bit charset, leave the string unchanged
259        if charset != '8bit':
260            # We need to test that the string can be converted to unicode and
261            # back to a byte string, given the input and output codecs of the
262            # charset.
263            if isinstance(s, str):
264                # Possibly raise UnicodeError if the byte string can't be
265                # converted to a unicode with the input codec of the charset.
266                incodec = charset.input_codec or 'us-ascii'
267                ustr = unicode(s, incodec, errors)
268                # Now make sure that the unicode could be converted back to a
269                # byte string with the output codec, which may be different
270                # than the iput coded.  Still, use the original byte string.
271                outcodec = charset.output_codec or 'us-ascii'
272                ustr.encode(outcodec, errors)
273            elif isinstance(s, unicode):
274                # Now we have to be sure the unicode string can be converted
275                # to a byte string with a reasonable output codec.  We want to
276                # use the byte string in the chunk.
277                for charset in USASCII, charset, UTF8:
278                    try:
279                        outcodec = charset.output_codec or 'us-ascii'
280                        s = s.encode(outcodec, errors)
281                        break
282                    except UnicodeError:
283                        pass
284                else:
285                    assert False, 'utf-8 conversion failed'
286        self._chunks.append((s, charset))
287
288    def _split(self, s, charset, maxlinelen, splitchars):
289        # Split up a header safely for use with encode_chunks.
290        splittable = charset.to_splittable(s)
291        encoded = charset.from_splittable(splittable, True)
292        elen = charset.encoded_header_len(encoded)
293        # If the line's encoded length first, just return it
294        if elen <= maxlinelen:
295            return [(encoded, charset)]
296        # If we have undetermined raw 8bit characters sitting in a byte
297        # string, we really don't know what the right thing to do is.  We
298        # can't really split it because it might be multibyte data which we
299        # could break if we split it between pairs.  The least harm seems to
300        # be to not split the header at all, but that means they could go out
301        # longer than maxlinelen.
302        if charset == '8bit':
303            return [(s, charset)]
304        # BAW: I'm not sure what the right test here is.  What we're trying to
305        # do is be faithful to RFC 2822's recommendation that ($2.2.3):
306        #
307        # "Note: Though structured field bodies are defined in such a way that
308        #  folding can take place between many of the lexical tokens (and even
309        #  within some of the lexical tokens), folding SHOULD be limited to
310        #  placing the CRLF at higher-level syntactic breaks."
311        #
312        # For now, I can only imagine doing this when the charset is us-ascii,
313        # although it's possible that other charsets may also benefit from the
314        # higher-level syntactic breaks.
315        elif charset == 'us-ascii':
316            return self._split_ascii(s, charset, maxlinelen, splitchars)
317        # BAW: should we use encoded?
318        elif elen == len(s):
319            # We can split on _maxlinelen boundaries because we know that the
320            # encoding won't change the size of the string
321            splitpnt = maxlinelen
322            first = charset.from_splittable(splittable[:splitpnt], False)
323            last = charset.from_splittable(splittable[splitpnt:], False)
324        else:
325            # Binary search for split point
326            first, last = _binsplit(splittable, charset, maxlinelen)
327        # first is of the proper length so just wrap it in the appropriate
328        # chrome.  last must be recursively split.
329        fsplittable = charset.to_splittable(first)
330        fencoded = charset.from_splittable(fsplittable, True)
331        chunk = [(fencoded, charset)]
332        return chunk + self._split(last, charset, self._maxlinelen, splitchars)
333
334    def _split_ascii(self, s, charset, firstlen, splitchars):
335        chunks = _split_ascii(s, firstlen, self._maxlinelen,
336                              self._continuation_ws, splitchars)
337        return zip(chunks, [charset]*len(chunks))
338
339    def _encode_chunks(self, newchunks, maxlinelen):
340        # MIME-encode a header with many different charsets and/or encodings.
341        #
342        # Given a list of pairs (string, charset), return a MIME-encoded
343        # string suitable for use in a header field.  Each pair may have
344        # different charsets and/or encodings, and the resulting header will
345        # accurately reflect each setting.
346        #
347        # Each encoding can be email.utils.QP (quoted-printable, for
348        # ASCII-like character sets like iso-8859-1), email.utils.BASE64
349        # (Base64, for non-ASCII like character sets like KOI8-R and
350        # iso-2022-jp), or None (no encoding).
351        #
352        # Each pair will be represented on a separate line; the resulting
353        # string will be in the format:
354        #
355        # =?charset1?q?Mar=EDa_Gonz=E1lez_Alonso?=\n
356        #  =?charset2?b?SvxyZ2VuIEL2aW5n?="
357        chunks = []
358        for header, charset in newchunks:
359            if not header:
360                continue
361            if charset is None or charset.header_encoding is None:
362                s = header
363            else:
364                s = charset.header_encode(header)
365            # Don't add more folding whitespace than necessary
366            if chunks and chunks[-1].endswith(' '):
367                extra = ''
368            else:
369                extra = ' '
370            _max_append(chunks, s, maxlinelen, extra)
371        joiner = NL + self._continuation_ws
372        return joiner.join(chunks)
373
374    def encode(self, splitchars=';, '):
375        """Encode a message header into an RFC-compliant format.
376
377        There are many issues involved in converting a given string for use in
378        an email header.  Only certain character sets are readable in most
379        email clients, and as header strings can only contain a subset of
380        7-bit ASCII, care must be taken to properly convert and encode (with
381        Base64 or quoted-printable) header strings.  In addition, there is a
382        75-character length limit on any given encoded header field, so
383        line-wrapping must be performed, even with double-byte character sets.
384
385        This method will do its best to convert the string to the correct
386        character set used in email, and encode and line wrap it safely with
387        the appropriate scheme for that character set.
388
389        If the given charset is not known or an error occurs during
390        conversion, this function will return the header untouched.
391
392        Optional splitchars is a string containing characters to split long
393        ASCII lines on, in rough support of RFC 2822's `highest level
394        syntactic breaks'.  This doesn't affect RFC 2047 encoded lines.
395        """
396        newchunks = []
397        maxlinelen = self._firstlinelen
398        lastlen = 0
399        for s, charset in self._chunks:
400            # The first bit of the next chunk should be just long enough to
401            # fill the next line.  Don't forget the space separating the
402            # encoded words.
403            targetlen = maxlinelen - lastlen - 1
404            if targetlen < charset.encoded_header_len(''):
405                # Stick it on the next line
406                targetlen = maxlinelen
407            newchunks += self._split(s, charset, targetlen, splitchars)
408            lastchunk, lastcharset = newchunks[-1]
409            lastlen = lastcharset.encoded_header_len(lastchunk)
410        value = self._encode_chunks(newchunks, maxlinelen)
411        if _embeded_header.search(value):
412            raise HeaderParseError("header value appears to contain "
413                "an embedded header: {!r}".format(value))
414        return value
415
416
417
418def _split_ascii(s, firstlen, restlen, continuation_ws, splitchars):
419    lines = []
420    maxlen = firstlen
421    for line in s.splitlines():
422        # Ignore any leading whitespace (i.e. continuation whitespace) already
423        # on the line, since we'll be adding our own.
424        line = line.lstrip()
425        if len(line) < maxlen:
426            lines.append(line)
427            maxlen = restlen
428            continue
429        # Attempt to split the line at the highest-level syntactic break
430        # possible.  Note that we don't have a lot of smarts about field
431        # syntax; we just try to break on semi-colons, then commas, then
432        # whitespace.
433        for ch in splitchars:
434            if ch in line:
435                break
436        else:
437            # There's nothing useful to split the line on, not even spaces, so
438            # just append this line unchanged
439            lines.append(line)
440            maxlen = restlen
441            continue
442        # Now split the line on the character plus trailing whitespace
443        cre = re.compile(r'%s\s*' % ch)
444        if ch in ';,':
445            eol = ch
446        else:
447            eol = ''
448        joiner = eol + ' '
449        joinlen = len(joiner)
450        wslen = len(continuation_ws.replace('\t', SPACE8))
451        this = []
452        linelen = 0
453        for part in cre.split(line):
454            curlen = linelen + max(0, len(this)-1) * joinlen
455            partlen = len(part)
456            onfirstline = not lines
457            # We don't want to split after the field name, if we're on the
458            # first line and the field name is present in the header string.
459            if ch == ' ' and onfirstline and \
460                   len(this) == 1 and fcre.match(this[0]):
461                this.append(part)
462                linelen += partlen
463            elif curlen + partlen > maxlen:
464                if this:
465                    lines.append(joiner.join(this) + eol)
466                # If this part is longer than maxlen and we aren't already
467                # splitting on whitespace, try to recursively split this line
468                # on whitespace.
469                if partlen > maxlen and ch != ' ':
470                    subl = _split_ascii(part, maxlen, restlen,
471                                        continuation_ws, ' ')
472                    lines.extend(subl[:-1])
473                    this = [subl[-1]]
474                else:
475                    this = [part]
476                linelen = wslen + len(this[-1])
477                maxlen = restlen
478            else:
479                this.append(part)
480                linelen += partlen
481        # Put any left over parts on a line by themselves
482        if this:
483            lines.append(joiner.join(this))
484    return lines
485
486
487
488def _binsplit(splittable, charset, maxlinelen):
489    i = 0
490    j = len(splittable)
491    while i < j:
492        # Invariants:
493        # 1. splittable[:k] fits for all k <= i (note that we *assume*,
494        #    at the start, that splittable[:0] fits).
495        # 2. splittable[:k] does not fit for any k > j (at the start,
496        #    this means we shouldn't look at any k > len(splittable)).
497        # 3. We don't know about splittable[:k] for k in i+1..j.
498        # 4. We want to set i to the largest k that fits, with i <= k <= j.
499        #
500        m = (i+j+1) >> 1  # ceiling((i+j)/2); i < m <= j
501        chunk = charset.from_splittable(splittable[:m], True)
502        chunklen = charset.encoded_header_len(chunk)
503        if chunklen <= maxlinelen:
504            # m is acceptable, so is a new lower bound.
505            i = m
506        else:
507            # m is not acceptable, so final i must be < m.
508            j = m - 1
509    # i == j.  Invariant #1 implies that splittable[:i] fits, and
510    # invariant #2 implies that splittable[:i+1] does not fit, so i
511    # is what we're looking for.
512    first = charset.from_splittable(splittable[:i], False)
513    last  = charset.from_splittable(splittable[i:], False)
514    return first, last
515