Mitigations to implement on AWS to attempt to prevent log4j exploitation.
This is likely the best option to implement as Amazon will be able to keep the rule up-to-date.
Enable AWS-AWSManagedRulesKnownBadInputsRuleSet where Log4JRCE is available in the latest version.
You can find example Terraform to create the rule with just Log4RCE enabled here: aws_waf/rule.tf
If you'd like more control over filtering, suggest using Lambda@Edge and in particular origin requests with body enabled.
The faas_code/index.js attempts to capture bad inputs, it can also find base64 and test it (which systems may parse headers automatically before passing to log4j where it could be exploited).
The main engine of the code is using the following regex: /(?:\$|\%24)(?:\{|\%7b)[^\w]*?j[^\w]*?n[^\w]*?d[^\w]*?i[^\w]*?(?:\:|\%3a)/im;
See cloudfront/lambda@edge/ for example Terraform of implementing the faas_code/index.js as a Lambda@Edge origin request.
See cloudfront/functions/ for example Terraform of implementing the faas_code/index.js as a CloudFront Function viewer request.
Note: CloudFront Functions don't have an option to inspect the body (for example, where parameters from POST requests would end up).