Skip to content

Latest commit

 

History

History
99 lines (85 loc) · 5.81 KB

attack-tree.md

File metadata and controls

99 lines (85 loc) · 5.81 KB

Attack Tree Analysis for pandas-dev/pandas

Objective: Exfiltrate Data OR Execute Arbitrary Code

Attack Tree Visualization

                                     +-------------------------------------------------+
                                     |  Exfiltrate Data OR Execute Arbitrary Code      |
                                     +-------------------------------------------------+
                                                     |
         +-------------------------------------------------------------------------+
         |                                                |                        |
+---------------------+                      +-------------------------+    +-------------------------+
| Data Exfiltration   |                      |  Arbitrary Code Execution|    | Denial of Service (DoS) |
+---------------------+                      +-------------------------+    +-------------------------+
         |                                                |                        |
+--------+--------+                            +--------+--------+        +--------+
|  1.  |  2.  |                            |  5.  |  6.  |        |  8.  |
|Input |Pickle|                            |Pickle|Eval/ |        |Memory|
|Manip.|Deser.|                            |Deser.|Exec  |        |Exh.  |
+--------+--------+                            +--------+--------+        +--------+
    |        |                                    |        |                |
    |        |[CRITICAL]                           |[CRITICAL]        |                |[HIGH RISK]
    |        |                                    |        |                |
 [HIGH RISK] |                                    |        |[HIGH RISK]      |
    |                                             |        |                |

Attack Tree Path: 1. Input Manipulation (Data Poisoning - Data Exfiltration)

  • Description: The attacker crafts malicious input data (e.g., CSV, Excel, JSON) that, when processed by pandas, triggers unintended behavior leading to data exfiltration. This exploits vulnerabilities in pandas' parsing logic or uses carefully constructed data to cause information disclosure.
  • Likelihood: Medium
  • Impact: High
  • Effort: Medium
  • Skill Level: Intermediate
  • Detection Difficulty: Medium
  • Examples:
    • A CSV file with crafted strings causing a buffer overflow or format string vulnerability (less likely in Python).
    • A large/nested JSON file triggering excessive memory allocation, potentially revealing other data.
  • Mitigations:
    • Thorough input validation and sanitization.
    • Limit input size.
    • Use schema validation.
    • Whitelist allowed characters and patterns.

Attack Tree Path: 2. Pickle Deserialization (Data Exfiltration) [CRITICAL]

  • Description: The application uses pandas.read_pickle() on untrusted input. The attacker provides a malicious pickle file that, when deserialized, leaks sensitive information.
  • Likelihood: High (if read_pickle() is used on untrusted input)
  • Impact: High
  • Effort: Low
  • Skill Level: Intermediate
  • Detection Difficulty: Easy (the use of read_pickle() on untrusted input is a red flag)
  • Example: A pickle file with a class that opens a network connection and sends data to the attacker.
  • Mitigations:
    • Never use read_pickle() with untrusted data.
    • Use safer serialization formats (JSON, CSV, etc.).

Attack Tree Path: 5. Pickle Deserialization (Code Execution) [CRITICAL]

  • Description: The application uses pandas.read_pickle() on untrusted input. The attacker provides a malicious pickle file that, when deserialized, executes arbitrary Python code.
  • Likelihood: Very High (if read_pickle() is used on untrusted input)
  • Impact: Very High
  • Effort: Low
  • Skill Level: Intermediate
  • Detection Difficulty: Easy (the use of read_pickle() on untrusted input is a red flag)
  • Example: A pickle file with a class that executes a system command.
  • Mitigations:
    • Never use read_pickle() with untrusted data.
    • Use safer serialization formats.

Attack Tree Path: 6. eval/exec within pandas Functions (Arbitrary Code Execution)

  • Description: Some pandas functions (e.g., DataFrame.query(), DataFrame.eval()) use eval or exec. If user-controlled input is passed directly to these functions without sanitization, it leads to code execution.
  • Likelihood: Medium
  • Impact: Very High
  • Effort: Low to Medium
  • Skill Level: Intermediate
  • Detection Difficulty: Medium
  • Example: User input to DataFrame.query() like ""index > 0; import os; os.system('malicious_command')"".
  • Mitigations:
    • Avoid passing user input directly to eval/exec functions.
    • Use parameterized queries or safe templating.

Attack Tree Path: 8. Memory Exhaustion (DoS)

  • Description: The attacker provides very large datasets or triggers operations that create large intermediate DataFrames, causing the application to run out of memory and crash.
  • Likelihood: Medium to High
  • Impact: Medium
  • Effort: Low
  • Skill Level: Novice
  • Detection Difficulty: Easy
  • Example: A CSV file with millions of rows and columns.
  • Mitigations:
    • Limit input size.
    • Monitor memory usage and set alerts.
    • Implement resource quotas.