Define:This builds on the server-dictated ordered transcript hashes currently mentioned in [[SenderKeys]].
A message m is "fully-acked" (from the POV of a given member u) iff, for all recipients r, u has accepted a message by r whose sender-parent is >= m's seqnum.* "recipients" possibly includes u, but certainly excludes m's sender* "accepted" means delivered locally, i.e. received, then decrypted-verified including parent hash checks; all previous messages must already be accepted= Background ==
Consistency checks are triggered as followsTo summarise:we assume the central server reliably delivers messages to everyone, including the original sender, in the same order. (Discussion of potential failures and recovery will be covered elsewhere.)
* when a member u wants to partEach message has an implicit server-sequence-number ("seqnum"), they send a receive-parent ("farewell" message m:** everyone should explicit-ack this message ASAP*** this message should contain the nextrecv-parent" or "parent") and a sender-key, to be used after u leaves, encrypted to everyone except u. sequence-number (Hopefully this addresses the concern Joe brought up."own-seqnum")*** could probably do something similar for group keys** when this message . Semantics of these is fully-acked, u gains consistency for all previous messages up to m, and may leave*** other messages should be probably be discarded, u won't have a chance to verify their consistencycovered elsewhere.** TBD: need to think about simultaneous parts
* when Once a member u accepts a non-explicit-ack message m at time t** if u did not send m, and they have not acked m by t+MAX_GRACE, they should send an explicit-ack** if m is not fully-acked (received from their POV) by t+the server (2*MAX_RTTincluding own messages sent)+(k*MAX_GRACE) (k slightly > 1) then issue , a local UI warning. Cancel "transcript-hash" may be calculated for it, that commits to that message and all previous messages in the warning if/when fullserver-ack is reached laterdictated order.This consists of:
* MAX_RTT should be based on all messages seen by the transportsender when they sent m, namely:* MAX_GRACE should be based on expected user communication rate* "recv-parent" p and messages with seqnum earlier than p** all messages sent by the same sender (messages with own-seqnum earlier than m)* ''as well as'' all messages inserted by the server, into the server-ordering between p and m
This guarantees that When a warning shows up if we don't reach consistency within the timeout defined abovenew message is sent, ensuring timelinessthe transcript-hash of its recv-parent is included with it.
In terms of overhead, effectively a user will send a message at least every MAX_GRACE time period, whilst the session has other people talking. When there is a lull in the conversation, there should be no further messages.== Definitions ==
IA message m is '''fully-acked''' (from the POV of a given member ''u'') iff, for all recipients ''r'', ''u'' has accepted a message by ''r'' where seqnum(parent(r)) ≥ seqnum(m confident ). * ''recipients'' possibly includes u, but certainly excludes m's sender* ''accepted'' means delivered locally, i.e. received, then decrypted-verified including parent hash checks; all causally-previous messages must already be accepted** we can tweak assume a server-dictated ordering, so accepting a message at seqnum i means we have already accepted all messages j ≤ i. Once ''m'' is fully-acked, ''u'' knows that everyone else has seen ''m'' and all messages before it. We'll call the parameters process of gaining that knowledge, "reaching consistency for ''m''" (given context, ''m'' might be implicit). We don't have another mechanism for it, so servers henceforth we'll use "reach consistency" interchangeably with "reach full-ack", though strictly the former is a security property and the latter is a mechanism for achieving it. An '''explicit-ack''' is a message with no user content, that is sent solely for the purpose of declaring (via recv-parent) what the sender has seen. We'll make use of these to ensure consistency is reached in a timely manner, even if no-one wants to send a normal message (that contains user-level contents, and which is an '''implicit-ack'''). == Consistency == We consider two cases: (a) reaching consistency for arbitrary messages during the course of a conversation, and (b) reaching consistency when a user ''u'' leaves. Case (b) may be viewed as a special instance of case (a) plus the additional premise that ''u'' must reach consistency as soon as possible (because they want to leave), and that they don't see too much extra loadcare about reaching consistency for any subsequent messages that they might receive after their final "farewell" message. Case '''(a)''': when a member ''u'' accepts a non-explicit-ack message ''m'' at time ''t'': * if u did not send m, but and they have not tried acked m by t+ACK_GRACE_INTERVAL, they should send an explicit-ack* if m is not fully-acked (from their POV) by t+(2*BROADCAST_LATENCY)+ACK_GRACE_INTERVAL then issue a local UI warning. Cancel the warning if/when full-ack is reached later. Case '''(b)''': when a member ''u'' wants to model part, they send a "farewell" message ''m'': * Everyone should explicit-ack ''m'' as soon as they receive it from the server.** This ack may also bundle any cryptographic material needed for other members to agree on a key or keys that exclude ''u''. For example, in the SenderKeys scenario, this preciselywould contain the next sender-key, encrypted to everyone except ''u''.* When ''m'' is fully-acked, ''u'' reaches consistency for all previous messages up to ''m'', and may leave.* Messages that others send to ''u'' that are echoed back after ''m'', will not be acked by ''u''.** ''u'' won't have a chance to reach consistency for these messages, even if ''u'' receives them.** Others won't have a chance to check that ''u'' read them, though they may be able to check that everyone else did. But ''u'' may still read them in theory, since they were encrypted to ''u''.** In either case, communicating this meaning to the user, should be already covered by the same UI as for case (a). In the first case, ''u'' may shortcut the timeout and just issue the warning straight away, or alternatively not display those messages to the user at all.* TBD: need to think about simultaneous parts === Parameters and properties === * BROADCAST_LATENCY should be based on the transport.* ACK_GRACE_INTERVAL should be based on expected user communication rate. Both of these should be defined to cover common cases (e.g. 95th-percentile) rather than being mean values. The above checks should be applied for every single message. This guarantees that a warning shows up if we don't reach consistency within the timeout defined above, and works even if people don't manually send messages. If we set ACK_GRACE_INTERVAL high enough to match the typical interval between a user's messages, then (in theory) the overhead is only incurred once everyone stops talking and we automatically send off one extra round of acks. While the lull in activity continues, there will be no extra messages, since we don't require explicit-acks to be fully-acked themselves. (Preliminary experiments on some code I wrote shows this to be effective.)
We are still vulnerable to a "drop everything" attack, but that can't be helped unless we have unconditionally-periodic heartbeats. Not sure if we want to put these in the upcoming spec.
(The above is basically my msg-notes stuff, except assuming reliable transport, without recovery or flow control, without heartbeats, and adapted to a server-dictated ordering.)
# Timeliness== Relative ordering ==
Trevor was also worried about Ensure that messages appearing in wildlyreceived out-of-different positions in a linear order, due are highly visible to malicious server delaying messages. I didn't like the MAX_RTD idea for various reasons; here is an alternative:user.
A message m is *rejected* if its sender-parent's 'badly ordered'' if seqnum is < (parent(m's )) < seqnum - MAX_UNSYNC_COUNT(m)* - MAX_UNSYNC_COUNT may either be constant, or a linear function of the number of members, TBD.
Seqnums are re-defined to ignore *rejected* messagesMAX_UNSYNC_COUNT may either be constant, or a linear function of the number of members, TBD.
This definition These messages should hopefully be globally consistent (or else transcript consistency breaks) so it's easier to reason about, and highlighted in some way in the sender of the rejected message can detect this reliably and offer UI that is not too severe. They are still valid messages; the user should just be informed that they refer to re-sendolder context that may be surprising.
It This definition is harder to be sure that time-based rejection conditions would be globally-consistent. Of course (or else transcript consistency still breaks in that case, but ) so it is *more likely* 's easier to breakreason about, and I think reducing this chance would be betterthe warning is simpler to explain than MAX_RTD.
[[Category: mpOTR]]
