Hard problems in secure communication
If you take a survey of interesting initiatives to create more secure communication, a pattern starts to emerge: it seems that any serious attempt to build a system for secure message communication eventually comes up against the following list of seven hard problems.
- Public key problem: Public key validation is very difficult for users to manage, but without it you cannot have confidentiality.
- Availability problem: People want to smoothly switch devices, and restore their data if they lose a device, but this is very difficult to do securely.
- Update problem: Almost universally, software updates are done in ways that invite attacks and device compromises.
- Meta-data problem: Existing protocols are vulnerable to meta-data analysis, even though meta-data is often much more sensitive than content.
- Asynchronous problem: For encrypted communication, you must currently choose between forward secrecy or the ability to communicate asynchronously.
- Group problem: In practice, people work in groups, but public key cryptography doesn’t.
- Resource problem: There are no open protocols to allow users to securely share a resource.
These problems appear to be present regardless of which architectural approach you take (centralized authority, distributed peer-to-peer, or federated servers).
It is possible to safely ignore many of these problems if you don’t particularly care about usability or matching the features that users have grown accustomed to with contemporary methods of online communication. But if you do care about usability and features, then you are stuck with finding solutions to these problems.
In our work, LEAP has tried to directly face down these seven problems. In some cases, we have come up with solid solutions. In other cases, we are moving forward with temporary stop-gap measures and investigating long term solutions. In two cases, we have no current plan for addressing the problems.
Public keys is very difficult for users to manage, but without it you cannot have confidentiality.
If proper key management is a precondition for secure communication, but it is too difficult for most users, what hope do we have?
The problem of public keys breaks down into five discrete issues:
- Key discovery is the process of obtaining the public key for a particular user identifier. Currently, there is no commonly accepted standard for mapping an identifier to a public key. For OpenPGP, many people use keyservers for this (although the keyserver infrastructure was not designed to be used in this way). A related problem is how a client can discover public key information for all the contacts in their addressbook for phonebook without revealing this information to a third party.
- Key validation is the process ensuring that a public key really does map to a particular user identifier. This is also called the “binding problem” in computer science. Traditional methods of key validation have recently become discredited.
- Key availability is the assurance that the user will have access, whenever needed, to their keys and the keys of other users. Almost every attempt to solve the key validation problem turns into a key availability problem, because once you have validated a public key, you need to make sure that this validation is available to the user on all the possible devices they might want to send or receive messages on.
- Key revocation is the process of ensuring that people do not use an old public key that has been superseded by a new one.
Of these problems, key validation is the most difficult and most central to proper key management. The two traditional methods of key validation are either the X.509 Certificate Authority (CA) system or the decentralized “Web of Trust” (WoT). Recently, these schemes have come under intense criticism. Repeated security lapses at many of the Certificate Authorities have revealed serious flaws in the CA system. On the other hand, in an age where we better understand the power of social network analysis and the sensitivity of the social graph, the exposure of metadata by a “Web of Trust” is no longer acceptable from a security standpoint.
An alternative method of key validation is called TOFU for Trust On First Use. With TOFU, a public key is assumed to be the right key the first time it is used. TOFU can work well for long term associations and for people who are not being targeted for attack, but its security relies on the security of the discovery transport and the application’s ability to retain a memory of discovered keys.
TOFU can break down in many real-world situations where a user might need to generate new keys or securely communicate with a new contact. TOFU is widely used for protocols like SSH, where the user receives confirmation of key continuity each time they connect to a server. There is no such confirmation with asynchronous messaging protocols, making TOFU much less appropriate in these situations.
Other strategies for addressing parts of the key management problem include:
- Inline Keys: Many projects plan to facilitate discovery by simply including the user’s public key in every outgoing message (as an attachment, in a footer, or in a header).
- DNS: Key distributed via DNSSEC, where a service provider adds a DNS entry for each user containing the user’s public key or fingerprint. This places all the trust in the DNS owner.
- Network perspective: Validation by key endorsement (third party signatures), with audits performed via network perspective.
- Introductions: Discovery and validation of keys through acquaintance introduction.
- Mobile: Although too lengthy to manually transcribe, an app on a mobile device can be used to easily exchange keys in person (for example, via a QR code or bluetooth connection).
- Append-only log: There is a proposal to modify Certificate Transparency to handle user accounts, where audits are performed against append-only logs.
- Biometric feedback: In the one case of voice communication, you can use recognition of the other person’s voice as a means to validate the key (when used in combination with a Short Authentication String). This is how ZRTP works.
For LEAP, we have developed a unique federated system called Nicknym that automatically discovers and validates public keys allowing the user to take advantage of public key cryptography without knowing anything about keys or signatures. Nicknym uses a combination of TOFU, provider endorsement, and network perspective. There is also a new proposal very similar to Nicknym called Nyms which we hope to also support. Nyms adds the ability of users from non-participating service providers to register their keys.
People want to smoothly switch devices, and restore their data if they lose a device, but this very difficult to do securely.
Users today demand the ability to access their data on multiple devices and to have piece of mind that their data will not be lost forever if they lose a device.
At LEAP, we have worked to solve the availability problem with a system we call Soledad (for Synchronization of Locally Encrypted Documents Among Devices). Soledad gives the client application an encrypted, synchronized, searchable document database. All data is client encrypted, both when it is stored on the local device and synced with the cloud. This is very powerful, as it allow the client developer to take advantage of a rich document database without needing to worry about how it is backed up or synchronized.
As far as we know, there is nothing else like it, either in the free software or commercial world. However, there are several projects in a similar problem space:
- Mylar, for client-encrypting web application data.
- Crypton, for client-encrypting web application data.
- Firefox Sync
Soledad tries to solve the problem of general data availability, but other initiatives have tried to tackle the more narrow problem of availability of private keys and discovered public keys. These initiatives include:
- Whiteout key sync
- Nilcat, experimental code for cloud storage of keys
- Ben Laurie’s proposed protocol for storing secrets in the cloud
- Phillip Hallam-Baker’s thoughts along similar lines
Almost universally, software updates are done in ways that invite attacks and device compromises.
The sad state of update security is especially troublesome because update attacks can now be purchased off the shelf by repressive regimes. The problem of software update is particular bad on desktop platforms. In the case of mobile and HTML5 apps, the vulnerabilities are not as dire, but the issues are also harder to fix.
To address the update problem, LEAP is adopting a unique update system called Thandy from the Tor project. Thandy is complex to manage, but is very effective at preventing known update attacks.
Thandy, and the related TUF, are designed to address the many security vulnerabilities in existing software update systems. In one example, other update systems suffer from an inability of the client to confirm that they have the most up-to-date copy, thus opening a huge vulnerability where the attacker simply waits for a security upgrade, prevents the upgrade, and launches an attack exploiting the vulnerability that should have just been fixed. Thandy/TUF provides a unique mechanism for distributing and verifying updates so that no client device will install the wrong update or miss an update without knowing it.
Related to the update problem is the backdoor problem: how do you know that an update does not have a backdoor added by the software developers themselves? Probably the best approach is that taken by Gitian, which provides a “deterministic build process to allow multiple builders to create identical binaries”. We hope to adopt Gitian in the future.
Existing protocols are vulnerable to meta-data analysis, even though meta-data is often much more sensitive than content.
As a short term measure, we are integrating opportunistic encrypted transport (TLS) for email and chat messages when relayed among servers. There are two important aspects to this:
- Relaying servers need a solid way to discover and validate the keys of one another. For this, we are initially using DNSSEC/DANE.
- An attacker must not be able to downgrade the encrypted transport back to cleartext. For this, we are modifying software to ensure that encrypted transport cannot later be downgraded.
This approach is potentially effective against external network observers, but does not protect the meta-data from the service providers themselves. Also, it does not, by itself, protect against more advanced attacks involving timing and traffic analysis.
In the medium term, LEAP plans to support direct delivery from client to server via Tor for service providers that support this. These anonymously delivered messages would be kept in a separate folder and only displayed to the user if the message signatures are valid. There is a lot of open debate as to the efficacy of using a low-latency onion routing network like Tor for something that might be better suited to a high latency mixing network. For now, Tor is useful because it exists and has a lot of traffic already. For a great discussion comparing mix networks and onion routing, see Tom Ritter’s blog post on the topic.
In the long term, we plan to adopt one of the proposed schemes for securely routing meta-data. These include:
- Auto-alias-pairs: Each party auto-negotiates aliases for communicating with each other. Behind the scenes, the client then invisibly uses these aliases for subsequent communication. The advantage is that this is backward compatible with existing routing. The disadvantage is that the user’s server stores a list of their aliases. As an improvement, you could add the possibility of a third party service to maintain the alias map.
- Onion-routing-headers: A message from user A to user B is encoded so that the “to” routing information only contains the name of B’s server. When B’s server receives the message, it unwraps (unencrypts) a supplementary header that contains the actual user “B”. Like aliases, this provides no benefit if both users are on the same server. As an improvement, the message could be routed through intermediary servers.
- Third-party-dropbox: To exchange messages, user A and user B negotiate a unique “dropbox” URL for depositing messages, potentially using a third party. To send a message, user A would post the message to the “dropbox”. To receive a message, user B would regularly polls this URL to see if there are new messages.
- Mixmaster-with-signatures: Messages are bounced through a mixmaster-like set of anonymization relays and then finally delivered to the recipient’s server. The user’s client only displays the message if it is encrypted, has a valid signature, and the user has previously added the sender to a ‘allow list’ (perhaps automatically generated from the list of validated public keys).
- Tor: One scheme employed by Pond is to simply allow for direct delivery over Tor from the sender’s device to the recipient’s server. This is fairly simple, and places all the work on the existing Tor network.
In all of these cases, meta-data protected routing can make abuse prevention more difficult. For this reason, it probably makes sense to only allow once of these options once both parties have already exchanged key material, in order to prevent the user being flooded with anonymous Spam.
For encrypted communication, you must currently choose between forward secrecy or the ability to communicate asynchronously.
With the pace of growth in digital storage and decryption, forward secrecy is increasingly important. Otherwise, any encrypted communication you engage in today is likely to become cleartext communication in the near future.
In the example of email and chat, we have OpenPGP with email and OTR with chat: the former provides asynchronous capabilities, and the latter forward secrecy, but neither one supports both abilities. We need both better security for email and the ability to send/receive offline chat messages.
In the short term, we are layering forward secret transport for email and chat relay on top of traditional object encryption (OpenPGP). This approach is identical to our stop-gap approach for the meta-data problem, with the one addition that relaying servers need the ability to not simply negotiate TLS transport, but to also negotiate forward secret ciphers and to prevent a cipher downgrade.
This approach is potentially effective against external network observers, but does not achieve forward secrecy from the service providers themselves.
In the long term, we plan to work with other groups to create new encryption protocol standards that can be both asynchronous and forward secret:
The Axolotl protocol used by both Pond and TextSecure currently has the most mature approach to asynchronous forward secrecy. This could be added as an invisible upgrade to normal email encryption when the client detects that both parties support it.
In practice, people work in groups, but public key cryptography doesn’t.
We have a lot of ideas, but we don’t have any solutions yet to fix this. Essentially, the question is how to use existing public key primitives to create strong cryptographic groups, where membership and permissions are based on keys and not arbitrary server-maintained access control lists.
Most of the interesting work in this area has been done by companies working on secure file backup/sync/sharing, such as Wuala and Spideroak. Unfortunately, there are not yet any good open protocols or free software packages that can handle group cryptography.
At the moment, probably the best approach is the simple approach: a protocol where the client encrypts each message to each recipient individually, and has some mechanism to verify the transcript to ensure that all parties received the same messages.
There is some free software work on some of interesting building blocks that could be useful in building group cryptography. For example:
- Proxy re-encryption: This allows the server to re-encrypt to new recipients without gaining access to the cleartext. The SELS mailing list manager uses OpenPGP to implement a clever scheme for proxy re-encryption.
- Ring signatures: This allows any member of a group to sign, withing anyone knowing which member.
There are no open protocols to allow users to securely share a resource.
For example, when using secure chat or secure federated social networking, you need some way to link to external media, such as an image, video or file, that has the same security guarantees as the message itself. Embedding this type of resource in the messages themselves is prohibitively inefficient.
We don’t have a proposal for how to address this problem. There are a lot of great initiatives working under the banner of read-write-web, but these do not take encryption into account. In many ways, solutions to the resource problem are dependent on solutions to the the group problem.
As with the group problem, most of the progress in this area has been by people working on encrypted file sync (e.g. strategies like Lazy Revocation and Key Regression).