- From: <noah_mendelsohn@us.ibm.com>
- Date: Tue, 14 Nov 2006 12:21:18 -0500
- To: John Cowan <cowan@ccil.org>
- Cc: Vincent Quint <Vincent.Quint@inrialpes.fr>, www-tag@w3.org
John Cowan writes: > Vincent Quint scripsit: > > > Thanks to Ed, a new version of the draft finding "Passwords in the Clear" > > is available at: > > > > http://www.w3.org/2001/tag/doc/passwordsInTheClear-52-20061113.html > > http://www.w3.org/2001/tag/doc/passwordsInTheClear-52 > > This draft mentions HTTP basic authentication only obliquely; it should > make it clear that using it is an instance of passwords-in-the-clear. I wonder whether the problem isn't the failure of the finding to more "clearly" define the phrase "in the clear". In fact, I think it's a rather subtle concept, because it's relative to who you think is doing the looking. We think of http basic being "in the clear" when used over ordinary TCP connections, but less so over SSL or other "encrypted" connections. Why? I think it's because we understand that, as the finding pretty much says, messages sent using HTTP basic without SSL wind up in a lot of places where a lot of people have the tools necessary to see them in their original form, i.e. as plain text. There are log viewers, proxy caches that can be searched, etc. It's assumed that many fewer people, if any, have the tools to crack digest authentication, so in that sense passwords sent using digest are in the clear for a much smaller community, and indeed we hope only for the intended recipient. Indeed, perspectives on what's 'in the clear' may change over time. I believe there was a time when if you said that you were using spread spectrum transmission techniques the military would have presumed you were using an advanced encrypted radio channel. Now we all have 802.11 receivers that readily "crack" at least one form of spread spectrum transmission. So, I wonder whether it would be worth more carefully defining what we mean by "in the clear"? =========SUGGESTED ROUGH DRAFT OF NEW SECTION ========================== 2. When is a transmission "In The Clear"? Computer messages are encoded at a variety of levels. For example, text may be encoded using Unicode, which in turn is represented using some transmission protocol, and ultimately transmitted or stored as electrical impulses, optical brightness levels, magnetic polarizations, etc. If you have the necessary equipment and are able to decode the encodings used, you can retrieve the information in a message. This finding concerns itself with assuring that certain messages, and in particular passwords transmitted in such messages, be safely hidden from those who are not intended to receive them. Definition: Accordingly, we say that a message or a password is "in the clear" if the representation used to transmit it is likely to be decodable by those other than the intended receiver. In particular, we observe that messages transmitted using HTTP over TCP (as opposed to HTTP over an encrypted channel such as SSL or TLS) tend to be decodable from message logs, from Web caches, by using packet sniffers, and with a wide variety of other commonly available tools and software libraries. Accordingly, this finding takes the position that, unless additional means are used to protect the content, any message sent using HTTP over (unencrypted) TCP should be presumed to be "in the clear", and thus vulnerable to unintended inspection. One common means to ensure that messages or portions of messages are not in the clear is through use of encryption, by which we mean encodings that are easy to decode when certain particular information such as a key is available, but impractically difficult to decode otherwise. Encryption may be used to encode a message prior to submission for transmission, or may be achieved as a byproduct of using an encrypted channel (which typically only protects the message while it's in transit, as opposed to at the sender or receiver.) A message sent using encryption is viewed as being "not in the clear" to the extent that the encryption used is sufficiently difficult to decode by unintended receivers, I.e. that its methods a robust against decoding without a key, and that key distribution is done sufficiently carefully that unintended recipients are unlikely to obtain the necessary keys. =========END SUGGESTED ROUGH DRAFT OF NEW SECTION ========================== With that in hand, I think the admonitions to "not solicit" passwords in the clear and not "transmit passwords in the clear" take on some teeth. This definition allows us to do what I think John is asking, which is to talk a bit more about basic vs. digest authentication, and to explain the senses in which each is or isn't "in the clear", when transmitted using ordinary HTTP over TCP vs HTTP over SSL or TLS. Thanks. Noah -------------------------------------- Noah Mendelsohn IBM Corporation One Rogers Street Cambridge, MA 02142 1-617-693-4036 --------------------------------------
Received on Tuesday, 14 November 2006 20:21:39 UTC