Cleartext Signed Message Signature Spoofing in openpgp

Impact

OpenPGP Cleartext Signed Messages are cryptographically signed messages where the signed text is readable without special tools:

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA256

This text is signed.
-----BEGIN PGP SIGNATURE-----

wnUEARMIACcFgmTkrNAJkInXCgj0fgcIFiEE1JlKzzDGQxZmmHkYidcKCPR+
BwgAAKXDAQDWGhI7tPbhB+jlKwe4+yPJ+9X8aWDUG60XFNi/w8T7ZgEAsAGd
WJrkm/H5AXGZsqyqqO6IWGF0geTCd4mWm/CsveM=
-----END PGP SIGNATURE-----

These messages typically contain a "Hash: ..." header declaring the hash algorithm used to compute the signature digest. OpenPGP.js up to v5.9.0 ignored any data preceding the "Hash: ..." texts when verifying the signature. As a result, malicious parties could add arbitrary text to a third-party Cleartext Signed Message, to lead the victim to believe that the arbitrary text was signed.

A user or application is vulnerable to said attack vector if it verifies the CleartextMessage by only checking the returned verified property, discarding the associated data information, and instead visually trusting the contents of the original message:

const cleartextMessage = `
-----BEGIN PGP SIGNED MESSAGE-----
This text is not signed but you might think it is. Hash: SHA256

This text is signed.
-----BEGIN PGP SIGNATURE-----

wnUEARMIACcFgmTkrNAJkInXCgj0fgcIFiEE1JlKzzDGQxZmmHkYidcKCPR+
BwgAAKXDAQDWGhI7tPbhB+jlKwe4+yPJ+9X8aWDUG60XFNi/w8T7ZgEAsAGd
WJrkm/H5AXGZsqyqqO6IWGF0geTCd4mWm/CsveM=
-----END PGP SIGNATURE-----
`;
const message = await openpgp.readCleartextMessage({ cleartextMessage });
const verificationResult = await verifyCleartextMessage({ message, verificationKeys });
console.log(await verificationResult.verified); // output: true
console.log(verificationResult.data); // output: 'This text is signed.'

Since verificationResult.data would always contain the actual signed data, users and apps that check this information are not vulnerable. Similarly, given a CleartextMessage object, retrieving the data using getText() or the text field returns only the contents that are considered when verifying the signature. Finally, re-armoring a CleartextMessage object (using armor() will also result in a "sanitised" version, with the extraneous text being removed. Because of this, we consider the vulnerability impact to be very limited when the CleartextMessage is processed programmatically; this is reflected in the Severity CVSS assessment, specifically in the scope's score ("Unchanged").

Patches

• v5.10.1 (current stable version) will reject messages when calling openpgp.readCleartextMessage()
• v4.10.11 (legacy version) will reject messages when calling openpgp.cleartext.readArmored()

Workarounds

Check the contents of verificationResult.data to see what data was actually signed, rather than visually trusting the contents of the armored message.

References

Similar CVE: https://sec-consult.com/vulnerability-lab/advisory/cleartext-message-spoofing-in-go-cryptography-libraries-cve-2019-11841/