Whisper: Wraps any Go io.ReadWriter in a secure tunnel using Ed25519/X25519 | Hacker News Reader

Whisper: Wraps any Go io.ReadWriter in a secure tunnel using Ed25519/X25519

github.com

125 pointsbubblehack3r25 comments

There is no description of the protocol or of its security goals, so I am making some guesses based on a cursory look at the source and what I imagine this might be for.

A single symmetric key is derived for both directions, and there is no checking of nonces, so as far as I can tell any message can be dropped, reordered, or replayed in both directions. (Including replaying message from A to B as if they were from B to A.) This is a bit like using ECB and likely to lead to fun application-specific attacks like [0].

This is very much rolling your own crypto, in a dangerous way. I am on the record as being "against" the "don't roll your own crypto" refrain [1], but mostly because it doesn't work: it should discourage people from publishing hand-rolled protocols such as this, but instead people think it means "don't roll your own primitives" and accept any use of "Ed25519/X25519" as probably secure.

Please read about the Noise framework [2] to get an idea of how much nuance there is to this, and consider using a Go implementation of it [3] instead.

P.S. This kind of issue is also why I maintain that NaCl is not a high-level scheme [4]: this could have used NaCl and have the exact same issues. libsodium has a couple slightly higher-level APIs that could have helped, secretstream [5] and kx [6], but again please use Noise.

[0] https://cryptopals.com/sets/2/challenges/13

[1] https://securitycryptographywhatever.buzzsprout.com/1822302/...

[2] https://noiseprotocol.org/noise.html

[3] https://github.com/flynn/noise

[4] https://words.filippo.io/dispatches/nacl-api/

[5] https://libsodium.gitbook.io/doc/secret-key_cryptography/sec...

[6] https://libsodium.gitbook.io/doc/key_exchange

I mean this looks like standard ECIES to me. I think it could even count as a Noise protocol, as in my understanding that protocol family subsumes most of the classic Diffie Hellman based key exchanges. Reading the source the author seems to use a long-lived ECDSA key pair to sign and exchange an ephemeral ECDH key pair, then derives a symmetric AES key from that and uses that for AES-GCM. Not sure how you would do message reordering here as GCM takes care of authenticating your data and keys are not reused. If you want replay protection it's probably sufficient to either remember used keys or add a timestamp. It's a well-done example of writing such a simple protocol, I don't think the author plans to replace TLS with it and he even mentions that in the README.

Each Write encrypts a separate "record" (in the parlance of TLS and Noise). Each of those records can be arbitrarily dropped, replayed, or reflected.

Here's an example, if you do

    Write("Hello")
    Write(" the password is ")
    Write("password")

then without a key I can make you or your peer read "Hello the password is Hello" (or "HelloHelloHello" or any composition of those messages).

No Noise protocol would allow that.

Oh well, I did not realize he uses it like that.

Can you express this with Noise framework tokens? I don't think you can. Noise is fiddlier than it looks! It's not just an ordering of DH exchanges; it's transcript hashes, cipher state tracking (and reinitializing), key derivation, the whole 9. The temptation (at least for me) is to just skip to the table of handshakes and skim, but the actual protocol framework is in Section 5, where they define precisely what each of those tokens really entails.

(To say nothing of: this uses signatures, and Noise does not).

Gotta say that I generally like the Noise framework (or rather the protocols that result), but it is one of the most impenetrable specifications I've ever read

I don't remember what it is specifically about it, I just remember the document being a pain to read; a bit like the original Paxos paper in that regard.

I don't know about that. As an implementor, it is probably one of the easiest-to-follow specs I've ever worked from. You can pretty much code it from the top of the spec to the bottom; when you get to the handshake patterns section and realize that each is just a different ordering of things in an array or whatnot, it's pretty slick.

Probably not, though Noise mentions that you could replace DH operations with signatures.

Yeah, I don't mean to be coy. You can't end up with this protocol using Noise. :)

ECIES is quite fine though for most asynchronous applications, where having a signing key also makes sense as you often want to publish long-lived, signed data and build a trust chain (e.g. generate and sign session keys from a master key). I built several real-world systems based on that (e.g. [1]) and they all made it through the audits fine. I was exploring Noise-based protocols but I find it's best to rely on primitives that are supported by the Web Crypto API.

1: https://github.com/kiebitz-oss/

ECIES is a hybrid encryption construction; Noise is a protocol. They're two different levels of abstraction. This thing we're commenting on has a protocol; it's just an accidental one, which is usually not what you want. WebCrypto doesn't provide a protocol framework, just a bunch of primitives.

> If you want replay protection it's probably sufficient to either remember used keys or add a timestamp.

The replay Filippo is talking about is in the context of a single key. The remote side simply reads the nonce || message off the wire and aead.Opens it, without regard to whether or not that nonce has been used before to decrypt a previous message.

Great opportunity to pitch my recent work on a similar solution for gRPC:

https://github.com/sillystack/api/blob/main/transport/transp...

I'm not going to make any security claims as of now (it's almost certainly broken) but it uses the Noise IK handshake.

Long term I think NoiseSocket would be a good way to go https://noisesocket.org/post/1/ but perhaps in service-to-service use cases ciphersuite negotiation can be simplified.

I once heard something along the lines of "if you're writing the words RSA, you're doing it wrong." I guess that goes for ED25519 also.

If you want to secure a stream with asymmetric cryptography and don't need all the bells and whistles of TLS, what's the correct way to do it? Noise? Is there nothing simpler?

IIUC the correct answer, according to the 2009 blog post that popularized the saying you're referring to, is to just use TLS anyway. In most cases it's better to accept some extra complexity than to get crypto wrong.

Edit: Here's that classic: https://web.archive.org/web/20090606064715/http://www.matasa...

As far as language support goes, it doesn't get much easier than working with TLS, especially in Go.

As far as alternatives go, Noise Pipes[0] never took off, but feel like they'd be well suited to this task.

[0]: https://noiseprotocol.org/noise.html#noise-pipes

Really great feedback, I hope it makes it to the module author.

There has been a lot of discussion about how the crypto doesn't work. I have some quibbles about the Go.

Don't put a mutex in your io.Reader. It is assumed that, unless mentioned in the documentation, it is not safe to use anything in Go from multiple goroutines. If someone wants to make the reader synchronous for use among multiple goroutines, they can do so themselves. But it's so rare that it's unlikely anyone would want this.

read/write mutexes typically perform worse than a plain mutex. You pay a performance cost to prevent readers and writers from starving each other; if you don't care (and this code doesn't), use a Mutex. https://zephyrtronium.github.io/articles/rwmutex.html

Finally, it's fine to embed the mutex value directly in your struct if your methods take pointer receivers. &MyThing{Mutex: new(sync.Mutex)} is pretty weird to see. If the mutex were included as its value, then the zero value of MyThing is ready to use; &MyThing{} has a new mutex in it. (You can't copy the value of &MyThing either way; a correct copy requires holding the mutex. The reason to embed a *sync.Mutex instead of a sync.Mutex is so that you can copy the containing type, but that is actually unsafe. You grab a pointer to the wrong Mutex in your copy, and you also copy data protected by the mutex without holding it. So your methods MUST have a pointer receiver either way, and thus the indirection to *sync.Mutex is unnecessary.)

Interesting, though AFAIK a secure tunnel is only useful for a net.Conn, not an io.ReadWriter. What's the usecase for a "secure tunnel" over a bytes.Buffer?

btw, I noticed that the decrypt function reads a 32-bit message length and immediately allocates a slice of that size. That means an attacker can send 0xFFFFFFFF and cause you to allocate 4GiB.

You're kinda looking at the interface issue backwards. Being an io.ReadWriter is a promise that the tunnel code won't do anything other than read or write. The tunnel code shouldn't be getting local and remote addresses or setting deadlines of its own, or most of the rest of what net.Conn can do. https://pkg.go.dev/net#Conn

It is good that it doesn't take more than it needs. That bytes.Buffer happens to implement io.ReadWriter is not of consequence; for any given task you want a io.Reader or io.Writer for there may be any number of specific implementations that don't make sense to use with it, but that doesn't mean you should require more methods of the interface.

I'm speaking solely about the interface here; the many other concerns are valid.

Sure, in general you want to use the narrowest possible interface. But in this particular case, I think `net.Conn` communicates intent better than `io.ReadWriter` -- the former implies that two distinct parties are involved, whereas `io.ReadWriter` is usually for (single-user) buffers or files. Sometimes it's sensible to use a larger interface than necessary; e.g. if you have a helper function that only calls `SetDeadline` and `Read`, the argument should be a `net.Conn`, not a custom interface with just those methods.

tbh though, it's a very minor quibble and not really worth worrying about ‾\_(ツ)_/‾

I disagree with this. If the only surface area of the underlying transport that the implementation uses is Read() and Write(), then it should be an io.Reader and an io.Writer. If it wants to mess around with read/write deadlines, then you'd have to bring in net.Conn, but it doesn't, so don't.

If you're worried about someone using bytes.Buffer as their transport layer, net.Conn doesn't fix that; there is net.Pipe. (It's not buffered, though.)

Yeah this code review isn’t gonna go great.

I recently implemented a very similar service to service gRPC authentication mechanism for Go using EC25519 and the noise protocol framework.

https://github.com/sillystack/api/blob/main/transport/transp...

The code is very fresh and hasn't yet gone through a audit so please don't use it for anything where security actually matters.

Reply on news.ycombinator.com