Re: [xmpp] comments on draft-miller-xmpp-e2e-06

[Matt Miller <[e-mail]>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA512

On 3/7/14, 5:28 AM, Carl Wallace wrote:
> 
> 
> On 3/7/14, 6:23 AM, "Matt Miller" <[e-mail]> wrote:
> 
> Thank you very much for the feedback!
> 
> I'll incorporate your suggestions as soon as I can.
> 
> On 3/6/14, 3:12 PM, Carl Wallace wrote:
>>>> Section 1 - This may be in the reqs draft, but seems worth
>>>> stating here that some end points for a given recipient may
>>>> share keys, some may use different keys, some may have no
>>>> keys and some may not support encryption or signature
>>>> verification at all.
>>>> 
> 
> We did discuss reviving the e2e-reqs draft in the WG session on 
> Tuesday, but it would be important to include this.
> 
>> It¹s also worth noting the sender may have multiple end points
>> with the same properties.  If the same SMK were used in replies,
>> it creates kind of a odd scenario where both parties could be
>> providing the SMK for the asking.
> 

Good point.

> 
>>>> Section 3 - What should the sender do if some end points for
>>>> a recipient indicate lack of support for encryption?  Are
>>>> error responses helpful independent of confirmation that some
>>>> end points support encryption?  Is the intent for messages to
>>>> only be sent to end points that advertise support?
> 
> There is little (read: nothing) that can be said on what existing 
> "legacy" end-points ought to do.  I tried to say something on what 
> existing clients already do, to the best of my knowledge.  Many
> (most) drop anything they don't understand, so getting errors could
> very well be impossible unless the applications were specifically
> updated to "anti-support" this specification.  It's also possible
> I'm missing the point here, too (-:
> 
> The intention is not for the sending end-point to only send to
> other supporting end-points.  Indeed, I want this to be compatible
> with [CARBONS] and [MUC] where the possible.  To that end, most
> clients try to do something if there is a body, and this document
> could suggest that sending end-points include an appropriate
> generic "this message is protected" body.
> 
>> I¹ve not yet looked, but I had wondered if it would be possible
>> to define the signature format so a non-supporting client could
>> still display the text (similar to cleartext email vs opaque).
>> This probably brings a can of canonicalization worms though.
> 

There is an XSF XEP that covers this using a different technology, and
calls it opportunistic signatures.

My experience has been that opportunistic signatures opens a very
large can of worms, which is why I avoided it in my draft.  There are
additional considerations for <iq/>s, since the expectation is for
there to be a single child element (except for type='error', where
there is also an <error/> in the parent's namespace).

We could decide to remove signatures from this draft almost entirely,
and leave it as someone else's exercise to deal with in the general
case.  In this scenario, I envision keeping some form of signature is
to aid in authenticating PFS key exchanges, but then it can be
specific to the keyreq then.

> 
>>>> 
>>>> - Why not include an end point's key in its info results and
>>>> enable return of the encrypted SMK (or even CEK) with the
>>>> encrypted payload (to save a round trip).  Given a single
>>>> message could trigger a number of requests for the SMK, the
>>>> option to include the encrypted CEK in the payload may be
>>>> worthwhile.  Including the encrypted CEK would help with
>>>> storage too.
> 
> The encrypted CEK is included in each message with -06.
> 
>> I made this error a couple of times.  What I meant was encrypted
>> CEK for a key already held by the end point, to avoid to the need
>> for a round trip to get the SMK, to remove the online requirement
>> for the sender at recipient reading time and to simplify future
>> use of stored encrypted messages.  The CEK could also be
>> encrypted under the SMK to bootstrap unknown end points.
> 
> 
> Including an encrypted SMK is a possibility, although I start to
> worry about the size of the overall stanza.  If it's absolutely
> known that the message will be delivered to a single end-point it
> shouldn't get too large, but I am assuming that servers will start
> multiplexing chat messages (a la [CARBONS]) which means the
> benefits of including the encrypted SMK are unrealized if only
> there's only a single instance of the encrypted SMK; including
> multiple SMKs can be very problematic (e.g., many server
> deployments reject very large stanzas).
> 
>> Do you know what the threshold is?

The threshold is different for different deployments /-:

There are some without any limit, which means the system crashes if it
runs out of memory trying to deal with an extremely large stanza.
Another I'm aware of has a limit of 40KB, and will drop the connection
on anything bigger.

I don't know how common any particular limit is, though.  And worse,
there's no real way to detect prior to trying.

> 
> 
>>>> 
>>>> 
>>>> - Why is the SID required to be unique for a {sender,
>>>> recipient, SMK} tuple?  This would allow for the same SID to
>>>> be used to name every SMK used with a recipient.  Suggest
>>>> unique for each recipient (i.e., SMK's are uniquely
>>>> identified by recipient+SID for a given sender).
>>>> 
> 
> I'm not sure what you're trying to say here (it could be due to me 
> responding on the last day of an IETF meeting).  Can you provide
> an example of what your concern here is?
> 
>> Assume a sender has used two SMKs while communicating with a
>> given recipient: sender+recipient+SMK1 and sender+recipient+SMK2.
>> The current text requires the SID to be unique "for a given
>> (sender, recipient, SMK) tuple².  That allows for the same SID to
>> be used for either SMK1 or SMK2. I am suggesting the SID should
>> be unique for each recipient, from a sender¹s perspective.
> 
> 

I see your point now.

>>>> - Discuss how multiple recipients are handled
> 
> That is a very good point.  There is some text there, but it is
> fairly weak.  I was considering a form of ASCII sequence diagram of
> the general flow -- would that better help to explain things?
> 
>> That would help.
> 
> 
>>>> 
>>>> - s/optionallly/optionally
>>>> 
> 
> Fixed in working copy.
> 
>>>> - It would be nice to have a summary of new state information
>>>> the sender and recipient are required to manage (like SMKs)
>>>> and significant new operations (like authorization of key
>>>> requests where a given recipient may use a number of
>>>> different public keys)
> 
> That is a very good suggestion.  Do you think you could provide
> some text to get it started?
> 
>> Sure.  Will try to send something next week.
> 
> 
>>>> 
>>>> - Should include an example of a reply to an encrypted
>>>> message. Should/must the same SMK be used in replies?
>>>> 
> 
> That is not a bad idea.  I'll look into it.
> 
>>>> - Is there any reason pre-placed SMK values cannot be used
>>>> instead of SMK values retrieved automatically from the
>>>> sender?
>>>> 
> 
> Is there any additional concern here beyond the point a few above?
> 
>> No, just looking for another place to not use of a key already
>> know to the receiver.
> 
> 
>>>> Section 4 - How is the public key used to verify signatures 
>>>> published to the recipients?  If not included in the
>>>> payload, should be available a la the SMK.  This would be
>>>> useful, at a minimum, when the signer uses a certificate from
>>>> a PKI the recipient trusts.
>>>> 
> 
> These are good points, and it ought to be included.  The JWS
> structure does allow for inclusion of the jwk (or jku, or x5c, or
> x5t, or x5u) which would be the key used to sign the content.  I'll
> add more clarity on this point.
> 
>>>> - s/Playload/payload
>>>> 
> 
> Fixed in working copy.
> 
>>>> - Include some text indicating users MUST be able to
>>>> determine if a message was authenticated
>>>> 
> 
> That is a good point.
> 
>>>> - Should data be presented to the user when the timestamp is
>>>> not acceptable?  There are guidelines for invalid signatures,
>>>> but not for unacceptable timestamps.  Same comment on 3.3.5
>>>> too.
>>>> 
> 
> I have been hesitant on dictating what to display to users.  I
> don't want to tie the hands of implementations, and it's rather
> difficult to independently verify that a given implementation
> complies without some sort of officiating body (which we definitely
> don't want).
> 
>> I thought these UI notes were more reminders to developers to
>> make sure user¹s could determine checks had been performed.
> 

I'm sure we can come up with something that's not exactly normative
language.

> 
>>>> - May want to distinguish between successful authentication
>>>> vs successful verification
>>>> 
> 
> Noted.
> 
>>>> Section 5 - Should describe how the entity requesting a key
>>>> gets the SID
>>>> 
> 
> I thought I had said that in here, but I'll clarify it further.
> 
>>>> - Should make it more clear that a new CEK is generated in
>>>> order to encrypt the SMK, then the CEK is encrypted under the
>>>> public key.
>>>> 
> 
> You have this reversed: the SMK is used to encrypt the CEK, and
> the recipient's public key is used to encrypt the SMK.  It is
> completely possible that I was not consistent here.
> 
>> I thought when the SMK is communicated to the end point, it is
>> also encrypted under a content encryption key (referred to as a
>> CMK in the text), that is then encrypted using the end point¹s
>> public key.  If that is correct there are four keys at work: the
>> CEK for the stanza (CEK1), the SMK to encrypt the CEK, the CEK to
>> encrypt the SMK (CEK2) and the recipient¹s public key to encrypt
>> CEK2.
> 

You are absolutely correct; sorry for my own confuzzlement!

> 
>>>> Section 9 - The requirement for the sender to be online would
>>>> seem like sufficient grounds for including encrypted CEKs in
>>>> the payload.
>>>> 
> 
> Encrypted CEKs are in the payload in -06.
> 
>> I tried to clarify in a reply comment above.  This applies to the
>> next few comments too.  Read ³encrypted CEKs² are ³CEKs encrypted
>> using a key already possessed by the recipient².  I think I was
>> consistent in misusing the term ³encrypted CEKs² in my comments
>> in this way, and introduced the CEK term in my comments where it
>> does not appear in the draft:-)
> 
> 
> 
>>>> Section 11 - Offline storage may be another reason to
>>>> include encrypted CEKs in a payload.  Perhaps the solution
>>>> should be to encrypt for each of a recipient's known keys
>>>> plus an SMK that could be used by end points using a
>>>> different, not-yet-known-to-the-sender key.
>>>> 
> 
> You mean encrypted SMKs here.
> 
> One idea that had been discussed outside the meeting was to try
> and use [PEP] as a way to do this, but we didn't spend a lot of
> time thinking about the mechanics and implications here.
> 
> 
>>>> - Need to address authorization for SMK release.
>>>> 
> 
> I have pretty much hand-waved over this part.  One idea that came
> to light this week was to provide fingerprints for the recipients'
> public keys, somehow.  Suggestions for improvements are welcome.
> 
>>>> - Include some guidance on SMK lifetime.
> 
> There is *some* guidance in section 11.2, but I admit it is rather 
> vague.  Could you provide more specific suggestions?
> 
>> As above, will try to send some text next week.

- -- 
- - m&m

Matt Miller < [e-mail] >
Cisco Systems, Inc.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG/MacGPG2 v2.0.22 (Darwin)
Comment: GPGTools - https://gpgtools.org
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/

iQEcBAEBCgAGBQJTKLPLAAoJEDWi+S0W7cO1L3IH/igYVs7+S4bJACRXkzVqUDqr
AUiIg/Ply3x04Cf06O0oCmRmfdxe9GhNv3CRbs07tETihZY31PV0Nez+BkHkVpN2
jcTE2ACDYbChGZAb55F8ThBNHWuDZfKJuRN0F+Elfckl9GVkCuwDiGj9TPLXvnbS
1KUdoFr0OQSwF8qaLMPdX3N/1/7srTmM84bkHL0RYpT4sihlzbWwYydr0/NKYSqy
L1Zn33Yw8RQpKjrtqOw5I7CrhN/qTytTD6srZNnqR/W68+MmDfb1y4RYfqnYX8Qy
NPuDdk/hME42hSkUoJqSGEKez0iClmSTakNserA3vjl6cn3cRUYZWRzoKBk43vI=
=36MU
-----END PGP SIGNATURE-----