Locate Jackson Deserialization Entry Points in Druid

Locating Jackson Deserialization Entry Points

Search Strategy

1. Find Filter Deserialization

   • Search for readValue.*DimFilter across the codebase
   • Search for objectMapper.readValue in indexing-service module
   • Search for transformSpec parsing and filter extraction
   • Look in DruidInputSource, InputSourceSampler, and SamplerResource

2. Trace the Request Path

   • The exploit targets /druid/indexer/v1/sampler endpoint
   • Find the corresponding resource handler (likely SamplerResource.java)
   • Trace from HTTP request → JSON parsing → object deserialization
   • Identify where spec.dataSchema.transformSpec.filter is deserialized

3. Identify the ObjectMapper Configuration

   • Find the ObjectMapper instance used for this deserialization
   • Check if it has custom deserializers registered
   • Locate the default deserialization behavior for filters

Expected Findings

The vulnerable flow should look similar to:

// Somewhere in SamplerResource or InputSourceSampler
String jsonInput = request.getBody();  // Raw JSON from attacker
ObjectMapper mapper = new ObjectMapper();
DataSchema schema = mapper.readValue(jsonInput, DataSchema.class);
// At this point, the empty key "" has already bypassed security
DimFilter filter = schema.getTransformSpec().getFilter();

Key Files to Examine

• druid-indexing-service/src/main/java/org/apache/druid/indexing/overlord/sampler/SamplerResource.java
• druid-indexing-service/src/main/java/org/apache/druid/indexing/overlord/sampler/InputSourceSampler.java
• druid-core/src/main/java/org/apache/druid/segment/transform/TransformSpec.java
• druid-core/src/main/java/org/apache/druid/query/filter/DimFilter.java

Validation Approach

Once the entry point is found, the fix must:

1. Validate raw JSON string before passing to ObjectMapper.readValue()
2. Reject empty keys "" at the input stream level
3. Prevent bypass of JavaScript security via malformed JSON structures