FWIW, some comments, much belatedly!



Frank Hecker wrote:

http://www.hecker.org/mozilla/ca-certificate-faq/policy-details/#risks


Which risks and threats are you concerned about in
connection with certificates and CAs?

As noted previously, we are concerned with the risks
borne by typical users arising from the various
certificate-enabled operations for which they might
use Mozilla, namely

  * browsing to SSL-enabled web sites or connecting
    to other SSL-enabled servers (e.g., for IMAP, SMTP, etc.)
  * sending and/or receiving signed and/or encrypted
    S/MIME email
  * downloading and installing and/or executing digitally
    signed code

For SSL-enabled browsing the most commonly-discussed
risks are associated with scenarios in which a typical
user believes they are communicating with an SSL-enabled
web site (or other server) associated with a certain
entity (e.g., their bank, a shopping site. their ISP's
SSL-enabled mail server, etc.), but in reality they are
communicating with another web site or server, operated
by an attacker. The major risk here is that an attacker
operating the second site will obtain sensitive
information relating to the user, whether entered by
the user (e.g., where an attacker is "phishing" for
passwords) or transmitted by the original site (e.g.,
where a "man in the middle" attacker is eavesdropping
on the user's connection).


Having read this, I can only say it is summarily
brief, but not inaccurate.  To concentrate on the
phishing exposure is valid as it is the only actual
case where empirically the security model is being
breached with frequency.  To ignore other threats
like straight eavesdropping or deceptions at the CA
point seems valid given the shortage of evidence
*and* the lack of space.  But, see the next two
points.

Also, it might be worth clarifying that phishing is
a cert problem and is outside the CA's hands (because
it in general bypasses the SSL security system completely).

The onus is on Mozilla to deal with phishing, in the
first instance, I don't think there is anything that a
given CA could do to deal with it, unless it is bang
on browser people to start thinking about it.  Until
browsers can show the difference between cert usage
and non-cert usage "in the user's face" then phishing
will roll on past the cert.  And until it becomes
routine to use users to perceive certs in a branded
fashion for all transactions, CAs are cut out of the
loop.

Mind you, if we *could* blame the CAs for phishing,
we'd feel a lot better, and maybe it would help in
the long run ;-)


There are also at least two other risks associated with
SSL-enabled web browsing and other SSL-enabled operations:

First, it may be that the user is actually connected to
the "correct" site or server, but the user is led to
believe that he or she is not properly connected to that
site; the risk here is that the user may not take a
desired action based on this false conclusion.


I don't understand where this comes from, is this
an actual attack model with real live results going
on?  I.e., empirically valid?  If so, where? (as I've
not heard of it, and feel ignorant.)

Alternatively, if it is only a theoretical attack,
then it sits more properly in a full list of such
attacks, and we need to consider all of them fairly,
without discrimination amongst our pets and favourites.

Or, thirdly, does this imply that some CAs are sending
out less-that-perfect certs that are fooling Mozilla
into acting incorrectly?  If so, I'd say that:  a bug
in a cert layout is considerd a risk, including where
one CA and one browser share the bug and thus attempt
to create a defact standard.


> Finally,
the user may encounter some software malfunction (e.g.,
a browser crash) due to some aspect of SSL and certificates;
again, this would possibly prevent the user from taking
some desired action.


Ditto.


For S/MIME-enabled email the most commonly-discussed risks
are associated with scenarios in which the user believes
that they are communicating with the owner of a given
email account (whether that owner be a person or entity),
but in fact the email messages in question are being
received from or by another (possibly malicious) person
or entity.


I suggest you adjust this to include mention that
there are two modes here.  Something like:

  ...but in fact the email messages in question are being
  read by another (possibly malicious) person, whether
  done by inserting in the middle and playing both ends
  (MITM) or by simply pretending to be the other person
  (spoofing).

(I'm not sure of this, myself, it just feels better
to mention the two modes.  Also, the definitions of
MITM and spoofing, etc, are not actually seriously
agreed amongst security people, so you might be more
comfortable leaving this out.  E.g., whether phishing
is MITM is a matter for the semanticians.)


> As a result the user runs the risk of exposing
sensitive information (e.g., sent in an encrypted email
read by an attacker) and/or taking inappropriate actions
based on information the user believes to be trustworthy


This equally applies to all of them, rather than just
email.  At least, to browsing, less so to code signing.


(e.g., a "phishing" site's URL contained in a signed
email sent by an attacker).


If this has actually happened - there has been signed
phishing attacks - then fine.  If it has never happened,
I'd leave it out.  There's enough confusion in attack
models as it is.


For downloadable signed code the most commonly-discussed
risks are associated with scenarios in the user believes


in WHICH the user ?


that the code in question originates from a particular
person or entity known to the user (e.g., a commercial
software vendor, or an open source developer), but in
fact the code originates from some other source, e.g.,
an attacker developing "trojan" programs, or a company
distributing "spyware" or "adware". As a result the user
might execute code that harms the user's system or exposes
sensitive information.


It might be helpful to mention that code signing
is a relatively new feature (and thus the attacks
mentioned are more envisaged as fears than validated
as risks).


As with SSL-enabled web browsing and S/MIME email, there
are also risks associated with scenarios in which the
user falsely concludes that the downloaded code is
associated with the wrong person or entity, as well as
risks associated with browser malfunctions due to
problems related to code signing and certificates.


Note that many of the risks and threat scenarios discussed above could result from factors unrelated to CAs and certificates. To take but one example, it is possible that an attacker might exploit a non-SSL vulnerability in Mozilla to cause the browser's status bar and/or URL input field to display an incorrect site name when connecting to the attacker's web site.


OK.

iang
_______________________________________________
mozilla-crypto mailing list
[EMAIL PROTECTED]
http://mail.mozilla.org/listinfo/mozilla-crypto

Reply via email to