math/big: panic during recursive division of very large numbers

[katiehockman]

A number of math/big.Int methods (Div, Exp, DivMod, Quo, Rem, QuoRem, Mod, ModInverse, ModSqrt, Jacobi, and GCD) can panic when provided crafted large inputs. For the panic to happen, the divisor or modulo argument must be larger than 3168 bits (on 32-bit architectures) or 6336 bits (on 64-bit architectures). Multiple math/big.Rat methods are similarly affected.

crypto/rsa.VerifyPSS, crypto/rsa.VerifyPKCS1v15, and crypto/dsa.Verify may panic when provided crafted public keys and signatures. crypto/ecdsa and crypto/elliptic operations may only be affected if custom CurveParams with unusually large field sizes (several times larger than the largest supported curve, P-521) are in use. Using crypto/x509.Verify on a crafted X.509 certificate chain can lead to a panic, even if the certificates don’t chain to a trusted root. The chain can be delivered via a crypto/tls connection to a client, or to a server that accepts and verifies client certificates. net/http clients can be made to crash by an HTTPS server, while net/http servers that accept client certificates will recover the panic and are unaffected.

Moreover, an application might crash invoking crypto/x509.(*CertificateRequest).CheckSignature on an X.509 certificate request or during a golang.org/x/crypto/otr conversation. Parsing a golang.org/x/crypto/openpgp Entity or verifying a signature may crash. Finally, a golang.org/x/crypto/ssh client can panic due to a malformed host key, while a server could panic if either PublicKeyCallback accepts a malformed public key, or if IsUserAuthority accepts a certificate with a malformed public key.

Thanks to the Go Ethereum team and the OSS-Fuzz project for reporting this. Thanks to Rémy Oudompheng and Robert Griesemer for their help developing and validating the fix.

This issue is CVE-2020-28362.

[katiehockman]

Fixed by 1e1fa59

[aaronbee]

Is go1.13 affected? I noticed that the fix touches code that was added in commit 194ae32 which is not present in go1.13.

[cristaloleg]

Looks like 1.13 isn't supported anymore, only 1.14 and 1.15 for today https://golang.org/doc/devel/release.html#policy

[aaronbee]

I understand that. I'm just asking if someone knows if go1.13 is affected. It is useful to know if programs built with go1.13 currently out in the wild are susceptible. I can check myself when the test is published, but I was hoping to get an earlier answer.

[p-rog]

It does look like this vulnerability was introduced by the recursive division algorithm implementation, which appears since go1.14beta. This issue could be limited to golang versions > 1.14, but it would be good to confirm that somehow.

[katiehockman]

I can confirm that the issue does not impact versions prior to 1.14. However, please note the official policy:

Each major Go release is supported until there are two newer major releases.

This means that right now we support only Go 1.15 and Go 1.14, and that Go 1.13 is unsupported and it won't receive backports for security fixes.

We typically don't research or document the impact of vulnerabilities on unsupported versions of Go, since that is outside the scope of responsibility for the team.

[gopherbot]

Change https://golang.org/cl/277959 mentions this issue: math/big: add test for recursive division panic