ietf-mailsig
[Top] [All Lists]

Re: Most recent sender.

2005-01-15 21:38:11

David Woodhouse wrote:

An invalid signature _can_ be assumed to mean forgery in the case of a
signature on a single transit through the mail system -- i.e. a
signature on the RFC2821 sender address like SES, or the bizarre 'most
recent RFC2822 sender' that seems to have been settled upon here. That
assumption is only going to be invalid in the case of an MTA mangling
the message in transit _without_ mailing lists being involved. It's not
going to be common that such mangling happens after the sending site
generates the signature and before the receiving site checks it. The
only failure mode I can really imagine is the case of an intermediate
forwarding site adding virus-checker adverts to the mail. But forwarding
sites tend not to do that in my experience; I think that's negligible.
They do in my experience; ieee.org for example. What's more, adding headers is behavior that is permitted by RFC 2822, so we have to assume it's going to happen. But it need not break the signature, if the specific headers that are signed are identified, unless an MTA happens to modify one of them. But that happens, too, for example when SpamAssassin adds [SPAM] to some messages (including some false positives), and this can be overcome by copying the headers and signing that.

If that's all I can do with it, I won't bother implementing it because
it'll give me no benefit apart from a small reduction in my
SpamAssassin-related CPU load. It doesn't let me discard anything I
wouldn't already have discarded -- since if I _don't_ like the apparent
sender I'd not have accepted the message anyway, regardless of whether
it's signed or not.
Some domains will be interested because it can guarantee a zero false positive rate for signed messages from trusted domains (business partners and customers, for example). It also allows one to use SpamAssassin more aggressively with less risk of falsely tagging a customer's message as spam.

-Jim


<Prev in Thread] Current Thread [Next in Thread>