Click to find:
Inclusions of log4j2 in compiled code |
Calls to log4j2 in compiled code |
Calls to log4j2 in source code |
|---|---|---|
| Sanity check for env mitigations |
CVE-2021-44228 poses a serious threat to a wide range of Java-based applications. The important questions a developer may ask in this context are:
Does the released code include log4j2? Which version of the library is included there? Answering these questions may not be immediate due to two factors:
-
Transitive dependencies: while
log4j2may not be in the direct dependency list of the project, it may be used indirectly by some other dependency. -
The code of this library may not appear directly as a separate file (i.e.,
log4j2-core-2.xx.0.jar), but rather be bundled in some other code jar file.
JFrog is releasing a tool to help resolve this problem: scan_log4j_versions. The tool looks for the class code of JndiManager and JndiLookup classes (regardless of containing .jar file names and content of pom.xml files), and attempts to fingerprint the versions of the objects to report whether the included version of log4j2 is vulnerable. Both Python and Java implementations are included.
The question is relevant for the cases where the developer would like to verify if the calls to log4j2 in the codebase may pass potentially attacker-controlled data. While the safest way to fix the vulnerability, as discussed in the advisories, is to apply the appropriate patches and global flags, controlling for and verifying the potential impact under assumption of unpatched log4j2 may be valuable in many situations. In order to address this problem JFrog is releasing two scripts:
scan_log4j2_calls_src.py, which locates calls to log4j2 logging functions (info, log, error etc.) with non-constant arguments in .java source files and reports the findings on the level of source file and linescan_log4j2_calls_jar.py, which locates the calls to logging functions in compiled .jars, and reports the findings as class name and method names in which each call appears.
Due to the high risk associated with the vulnerability, developers relying on mitigations may want to double check that the environment was indeed configured correctly (which Java runtime actually runs the application? Were environment and command line flags set correctly?). In order to simplify this sanity check, Jfrog releases a simple tool env_verify.jar which is intended to run in the same environment as a production application and validate it.
The tool requires Python 3, without additional dependencies.
python scan_log4j_versions.py root-folder
The tool will scan root_folder recursively for .jar and .war files; in each located file the tool looks for a *log4j/core/net/JndiManager.class and *log4j/core/lookup/JndiLookup.class (recursively in each .jar file). If at least one of the classes is found, the tool attempts to fingerprint its version (including some variations found in patches and backport patches) in order to report whether the code is vulnerable.
To reiterate, the results depend on the code of the classes rather than file names and the metadata. Files where both JndiManager and JndiLookup classes are not present (and hence are not vulnerable to CVE-2021-44228), like log4j-1.x.xx.jar, or log4j-api-2.xx.x.jar, do not appear in the results. Otherwise, vulnerability status and estimated version/patch status are displayed. When the versions of the two classes follow a pattern not accounted for, inconsistent is reported; this result should be investigated further.
| Vulnerable | Mitigated | Fixed |
|---|---|---|
2.0, 2.1 .. 2.14 |
2.15 |
2.12.2, 2.16, 2.17 ,JndiLookup removed |
Compiled jar can be downloaded from here or compiled from source.
The tool requires java runtime, without additional dependencies.
java -jar scan_jndimanager_versions.jar root-folder
The operation and displayed results are equivalent to the Python version.
The tool requires python 3 and the following 3rd party libraries: jawa, tqdm, easyargs, colorama
pip install -r requirements.txt
The default use case:
python scan_log4j_calls_jar.py root-folder
will recursively scan all .jar files in root-folder, for each printing out locations (class name and method name) of calls to info/warn/error/log/debug /trace/fatal methods of log4j2.Logger.
The tool may be configured for additional use cases using the following command line flags.
| Flag | Default value | Use |
|---|---|---|
--class_regex |
org/apache/logging/log4j/Logger | Regular expression for required class name |
--method_regex |
1 | Regular expression for required method name |
--quickmatch_string |
log4j | Pre-condition for file analysis: .jar files not containing the specified string will be ignored |
--class_existence |
Not set | When not set, look for calls to class::method as specified by regexes. When set, --method_regex is ignored, and the tool will look for existence of classes specified by --class_regex in the jar. |
--no_quickmatch |
Not set | When set, the value of --quickmatch_string is ignored and all jar files are analyzed |
For example,
python scan_log4j_calls_jar.py --class_regex ".*JndiManager$" --class_existence --no_quickmatch root-folder
Will scan all .jar files (even if they do have no mentions of log4j2) for the existence of a class ending with JndiManager.
Typical results output looks like this:
The tool requires python 3 and the following 3rd party libraries: javalang, tqdm, easyargs, colorama
pip install -r requirements.txt
The default use case:
python scan_log4j_calls_src.py root-folder
will recursively scan all .java files in root-folder, for each printing out the locations (file name and corresponding code lines) of calls to log4j2 logging methods.
The tool may be configured for additional use cases using the following command line flags:
| Flag | Default value | Use |
|---|---|---|
--class_regex |
org/apache/logging/log4j/Logger | Regular expression for required class name |
--method_regex |
1 | Regular expression for required method name |
Typical output looks like this:
Compiled jar can be downloaded from here or compiled from source, and does not require additional dependencies.
The intended use is running the tool in the same setting precisely as the production application. For example, for the original launch line in the start-up script:
eval "\"${JAVA_CMD}\" ${VMARG_LIST} application ${CLASSNAME} ${ARGS[@]}" &>/dev/null &We add the following to the script:
eval "\"${JAVA_CMD}\" ${VMARG_LIST} -jar env_verify.jar" > /tmp/env_verifyAnd read the result after the start-up script completes:
cd scan_log4j_versions
gradle build
cp build/libs/scan_log4j_versions.jar ..
cd env_verify
gradle build
cp build/libs/env_verify.jar ..