1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
/*! # A review of TLS Implementation Vulnerabilities

An important part of engineering involves studying and learning from the mistakes of the past.
It would be tremendously unfortunate to spend effort re-discovering and re-fixing the same
vulnerabilities that were discovered in the past.

## Memory safety

Being written entirely in the safe-subset of Rust immediately offers us freedom from the entire
class of memory safety vulnerabilities.  There are too many to exhaustively list, and there will
certainly be more in the future.

Examples:

- Heartbleed [CVE-2014-0160](https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2014-0160) (OpenSSL)
- Memory corruption in ASN.1 decoder [CVE-2016-2108](https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2016-2108) (OpenSSL)
- Buffer overflow in read_server_hello [CVE-2014-3466](https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2014-3466) (GnuTLS)

## `goto fail`

This is the name of a vulnerability in Apple Secure Transport [CVE-2014-1266](https://nvd.nist.gov/vuln/detail/CVE-2014-1266).
This boiled down to the following code, which validates the server's signature on the key exchange:

```c
    if ((err = SSLHashSHA1.update(&hashCtx, &serverRandom)) != 0)
        goto fail;
    if ((err = SSLHashSHA1.update(&hashCtx, &signedParams)) != 0)
        goto fail;
>       goto fail;
    if ((err = SSLHashSHA1.final(&hashCtx, &hashOut)) != 0)
        goto fail;
```

The marked line was duplicated, likely accidentally during a merge.  This meant
the remaining part of the function (including the actual signature validation)
was unconditionally skipped.

Ultimately the one countermeasure to this type of bug is basic testing: that a
valid signature returns success, and that an invalid one does not.  rustls
has such testing, but this is really table stakes for security code.

Further than this, though, we could consider that the *lack* of an error from
this function is a poor indicator that the signature was valid.  rustls, instead,
has zero-size and non-copyable types that indicate a particular signature validation
has been performed.  These types can be thought of as *capabilities* originated only
by designated signature verification functions -- such functions can then be a focus
of manual code review.  Like capabilities, values of these types are otherwise unforgeable,
and are communicable only by Rust's move semantics.

Values of these types are threaded through the protocol state machine, leading to terminal
states that look like:

```ignore
struct ExpectTraffic {
   (...)
    _cert_verified: verify::ServerCertVerified,
    _sig_verified: verify::HandshakeSignatureValid,
    _fin_verified: verify::FinishedMessageVerified,
}
```

Since this state requires a value of these types, it will be a compile-time error to
reach that state without performing the requisite security-critical operations.

This approach is not infallible, but it has zero runtime cost.

## State machine attacks: EarlyCCS and SMACK/SKIP/FREAK

EarlyCCS [CVE-2014-0224](https://nvd.nist.gov/vuln/detail/CVE-2014-0224) was a vulnerability in OpenSSL
found in 2014.  The TLS `ChangeCipherSpec` message would be processed at inappropriate times, leading
to data being encrypted with the wrong keys (specifically, keys which were not secret).  This resulted
from OpenSSL taking a *reactive* strategy to incoming messages ("when I get a message X, I should do Y")
which allows it to diverge from the proper state machine under attacker control.

[SMACK](https://mitls.org/pages/attacks/SMACK) is a similar suite of vulnerabilities found in JSSE,
CyaSSL, OpenSSL, Mono and axTLS.  "SKIP-TLS" demonstrated that some implementations allowed handshake
messages (and in one case, the entire handshake!) to be skipped leading to breaks in security.  "FREAK"
found that some implementations incorrectly allowed export-only state transitions (ie, transitions that
were only valid when an export ciphersuite was in use).

rustls represents its protocol state machine carefully to avoid these defects.  We model the handshake,
CCS and application data subprotocols in the same single state machine.  Each state in this machine is
represented with a single struct, and transitions are modelled as functions that consume the current state
plus one TLS message[^1] and return a struct representing the next state.  These functions fully validate
the message type before further operations.

A sample sequence for a full TLSv1.2 handshake by a client looks like:

- `hs::ExpectServerHello` (nb. ClientHello is logically sent before this state); transition to `tls12::ExpectCertificate`
- `tls12::ExpectCertificate`; transition to `tls12::ExpectServerKX`
- `tls12::ExpectServerKX`; transition to `tls12::ExpectServerDoneOrCertReq`
- `tls12::ExpectServerDoneOrCertReq`; delegates to `tls12::ExpectCertificateRequest` or `tls12::ExpectServerDone` depending on incoming message.
  - `tls12::ExpectServerDone`; transition to `tls12::ExpectCCS`
- `tls12::ExpectCCS`; transition to `tls12::ExpectFinished`
- `tls12::ExpectFinished`; transition to `tls12::ExpectTraffic`
- `tls12::ExpectTraffic`; terminal state; transitions to `tls12::ExpectTraffic`

In the future we plan to formally prove that all possible transitions modelled in this system of types
are correct with respect to the standard(s).  At the moment we rely merely on exhaustive testing.

[^1]: a logical TLS message: post-decryption, post-fragmentation.


*/