EFIDrill - IDA plugin for UEFI firmware vulnerability hunting base on data flow analysis

Supported versions of IDA: Well tested on IDA pro 8.2 and 8.3

Supported Platforms: Windows, Linux and OS X

Plugin installation

Linux / OS X

1. git clone https://github.com/cc-crack/efidrill.git
2. ln -s <EFIDRILL_PATH>/efidrill.py <IDA_DIR>/plugins
3. ln -s <EFIDRILL_PATH>/efidrill <IDA_DIR>/plugins

Windows

1. git clone https://github.com/cc-crack/efidrill.git
2. mklink <IDA_DIR>\plugins\efidrill.py <EFIDRILL_PATH>\efidrill.py
3. mklink <IDA_DIR>\plugins\efidrill <EFIDRILL_PATH>\efidrill

Configuration

Configurations of EFIDrill are in config.json, you can modify it to fit your needs.

It locates in the same directory as the plugin.

Template

We provide a template for you to modify:

{
  "deep_size": 15,
  "max_function_deep": 5,
  "workdir": "~/efidrill_workspace/",
  "logging_path": "efi_digging.log",
  "just_data_taint": 1,
  "debug_flag": 0,
  "to_ui": 1,
  "res_filename": "res.json"
}

Configuration items

• deep_size: The maximum number of instructions to be analyzed in a function.
• max_function_deep: The maximum recursive depth of functions to be analyzed.
• workdir: The directory where the analysis results and artifacts are stored.
• logging_path: The path of the log file.
• just_data_taint: Whether to perform only data taint analysis.
• debug_flag: Whether to enable debug mode.
• to_ui: Whether to output the analysis results to the UI of IDA Pro.
• res_filename: The name of the analysis result file.

Directory structure

|   check_engine.py #Loaded with all plug-ins and initializes function classes
|   config.json.template
|   config.py
|   debug.py
|   function_support.py
|   function_var.py
|   logging.py
|   plugin_mgr.py
|   rd_analysis.py  #Main program
|   result.py
|   test.py
|   user_define_work.py
|   __init__.py
|
+---arch
|       x64.py
|
+---function_type   #processing classes for different functions
|       child_smi.py
|       find_entry_point.py
|       function_ana_loader.py
|       function_type.py    #SMRAM external input data is initialized here
|       sub_function.py
|       sw_smi.py
|
+---ida_work
|       ida_support.py  #Encapsulation of the IDA library
|
+---mdis    #library that reaching definition analysis and encapsulates miasm
|       mdis_engine.py  #Generate use def engine
|       mdis_support.py #Encapsulate files processed by IR
|       ReachingDefinitionscfg.py
|       ReachingDefinitionsvar.py
|
\---plugin  #All analysis plug-ins
        base_plugin.py
        callout.py
        datasize.py
        global_var.py
        indirect_call.py
        memcopy.py
        smmvaild.py
        struct_build.py
        toctou.py

Important structures

• Function_type.interesting_op_list: Store all the content of interest def_var (toctou only has a -1 element description, the current variable does not analyze toctou).
• Function_type.all_uses, Function_type.all_defs: Store all use def {address:define}.
• Function_type.current_ins_addr: The address of the current analysis.
• Function_type.ircfg: Current IR.
• Function_type.mmap_ir_address: Mapping of IR addresses and addresses: (block_index,assm_index), address.
• variable: {miasm_struct,address}.
• Variables in miasm: {miasm_struct,(block_index,assm_index)}.
• miasm structure: Stores various OP, ID (registers), int, MEM.

Implement a custom plugin

1. Inherit the Base_Plugin class in the base_plugin.py file.
2. The add_interesting_memory_map_list function is called when a new def is added to the list of interesting variables, and use_list is the use_list of interest when calling the def, is_alias indicates whether this variable is an alias for another variable.
3. vulnerability_find is executed at each address, use_list is all the variables of interest used by that address, and def_list is all variables defined by that address.

   Note that since one def var corresponds to multiple use vars, variable assignment requires a list mapping of def var to use list.

Reaching definition(RD) problem

Note that the toctou and struct plugins do not have RD issues, so there is a path explosion (his stack depth does not exceed the maximum limit), and the other plugins have RD issues, so there is no path explosion problem.

Important functions

• mmap_ir_to_address: map intermediate representation (IR) addresses to actual addresses in a binary file.
• check_interesting_variables: Find the variables of interest (0x40e,readsavesatate, commbuffer), different functions have different variables of interest.
• ana_ins_addr: Analyze each assembly instruction.
• add_interesting_memory_map_list: Add new define to the interesting_op_list.
• vulnerability_find: Call all plug-ins.
• fix_use: Find the far jump use(def is not in the current function), the address of the far jump block ir is (-1,-1), and the actual address is -1.
• get_def: generate def.

TODO list

• UEFI Firmware Library and Automated Analysis
• Cross Architecture(ARM,MIPS,...)
• Variable Value Prediction Improve
• Writing a Fuzz Module

Can Do List

• datasize handling stack aliasing issues
• Incomplete judgment of heuristic expression jump

Incomplete hypothesis at present

• The safest assumption for SMRAM out-of-bounds access: a function takes rcx as external input and we assume that it is called, whether it is actual or not (possibly false positives).
• The external data address assumes that as long as a data comes from outside, we do not consider the memory location and SMM location in our PC (possibly false positives).

Running test

cd ./test

python3 ./generate.py for extracting all modules from the raw BIOS bin files.

python3 ./run.py for running Efidrill on all modules.

Check result

If everything is done, you can find the result in your workspace folder.

Credit

Xuxiang Yang, Security researcher @ Security Lab GIC Lenovo

Qingzhe Jiang, Security researcher, Manager, @ Security Lab GIC Lenovo

Weixiao Ji, Security researcher @ Security Lab GIC Lenovo

Thanks

Zhaoxing Sun helps to do some tests.

Xiaomin Li helps to create the project logo.

Ref

BH Asia 2024 arsenal:

https://www.blackhat.com/asia-24/arsenal/schedule/#efidrill-automated-hunting-uefi-firmware-vulnerability-through-data-flow-analysis-36764