[CXF-7287] Incorrect JSON decoding in JsonMapObjectReaderWriter when commas are used in value

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 3.1.10
• Fix Version/s: 3.1.11, 3.2.0
• Component/s: JAX-RS, JAX-RS Security

Environment

OSX, Tomcat8, Java8

• Estimated Complexity: Moderate
• Assignee: Sergey Beryozkin
• Reporter: Boris Capitanu
• Created: 17/Mar/17 01:42
• Updated: 10/Apr/17 19:04
• Resolved: 20/Mar/17 15:21

Description

Background:
Discovered when using JWT tokens. In org.apache.cxf.rs.security.jose.jws.JwsJwtCompactConsumer there is a line of code that reads:

JwtClaims theClaims = new JwtClaims(getReader().fromJson(getDecodedJwsPayload()));

If the method "getDecodedJwsPayload()" returns a JSON string like:

"\"sub\":\"admin\",\"roles\":\"admin,user\",\"iss\":\"auth0\""

then the "fromJson" method returns an incorrect Map.

The Bug
The problem occurs in org.apache.cxf.jaxrs.json.basic.JsonMapObjectReaderWriter (defined in cxf-rt-rs-json-basic-3.1.10.jar) in the readJsonObjectAsSettable method. Specifically, line 188 reads:

int commaIndex = getCommaIndex(json, sepIndex + j);

which will cause the wrong comma index to be found for the JSON key "roles" in my example above (apparently commaIndex is used to find the "next json key" position in the json string).

I'm also thinking that it's possible that line 166 could be a problem as well, if a JSON value would ever contain an (escaped) quote character.

This JSON parsing seems fragile... I wonder why a "standard" JSON library wasn't used (perhaps just to not add an additional dependency?)

Proper parsing of commas in JSON values is important for JWT purposes as the constructor of org.apache.cxf.rs.security.jose.jaxrs.JwtTokenSecurityContext expects role claims to be comma-separated.