Mishandling multi-line ISO encoded subject headers

[Nick Smith]

Hi,

MS 4.53.8, Postfix 2.2.10

I have a big problem right now with MailScanner apparently mangling some
(but not all) multi-line folded Subject headers - typically those containing
ISO encoded subjects for multi-byte character sets. Consider these two
examples:

Subject: =?iso-2022-jp?B?GyRCJCIkIiQiJCIkIiQiJCIkIiQiJCIkIiQiJCIbKEIgYWFhYWFhYQ==?=
 =?iso-2022-jp?B?YSAbJEIkIiQiJCIkIiQiJCIkIiQiJCIkIhsoQg==?=

Subject: =?iso-2022-jp?B?GyRCJCIkIiQiJCIkIiQiJCIkIiQiJCIkIiQiGyhCIGFhYWFhYWFhIA==?=
 =?iso-2022-jp?B?GyRCJCIkIiQiJCIkIiQiJCIkIiQiJCIbKEI=?=

The first one is an ISO-2022-JP encoded representation of 13 Japanese
double-byte "a" followed by a single space, 8 lower case ASCII "a", another
single space and finally 10 more Japanese double-byte "a".

The second is identical, except one of the 13 double-byte "a" has been
removed so there are only 12

Given that the first character of the second line in each example is a
space, they ought to be treated as a single header per RFC822's folded
header specification

The weird part is that the first example works, shows up unchanged in the
user's mailbox, and the Subject: header looks exactly as it did when it was
sent while the shorter second one gets broken. What shows up in the user's
mailbox (and the headers) is a decoded version of just the first line -
which looks like this:

Subject: ^[$B$"$"$"$"$"$"$"$"$"$"$"$"^[(B aaaaaaaa

MailScanner running with Sendmail does not seem to experience this problem,
and neither does Postfix running without MailScanner so it looks to be
something to do with MailScanner's Postfix-specific code

I did notice whilst digging in MailScanner's Postfix.pm that maybe the
complete handling of folded Subject: headers is not implemented - for
example lines 449-452:

if ($recdata =~ /^Subject:\s*(\S.*)?$/i) {
  $message->{subject} = $1;
  next;
}

This doesn't seem to handle the case where Subject: is on more than one
line, and will result in $message->{subject} containing only the first line
of a folded Subject:

Clearly this is not the whole story and an incomplete $message->{subject} is
not enough to kill it every time because otherwise it would never work with
a folded subject at all - as I said previously, the longer example above
does work OK as do many other folded subject headers.

Unfortunately, this is causing quite a big problem - it would be great if
somebody could suggest a fix. Failing a proper fix, is there any way to
modify the existing header_check:

/^Received:/                    HOLD

...so that it will exclude messages with (for example) /^Subject:
=*iso-2022-jp/ - that way I could maybe have these messages bypass MS for
the time being

Thanks

Nick