Executive brief
PaddlePaddle is a popular open-source deep learning framework used to build and train machine learning models. A command injection vulnerability in its convert_shape_compare component allowed attackers to execute arbitrary operating system commands. This could lead to complete system compromise, data theft, or malware installation on any machine using an affected version of PaddlePaddle.
Technical details
The vulnerability is a command injection (CWE-78) in the convert_shape_compare function within PaddlePaddle's dynamic-to-static (Dy2St) conversion module. The flaw stems from improper input sanitization when constructing or executing system commands. An attacker with the ability to supply crafted input to the Dy2St conversion process could inject shell metacharacters to break out of the intended command context and execute arbitrary commands with the privileges of the process running PaddlePaddle. The vulnerability affects all versions prior to 2.6.0 and was patched in commit 5ed9478fdef96a06eeec9093f9e768c97b094af3 by removing the vulnerable function from the public API.
Affected products
- PaddlePaddle Paddle before 2.6.0
Timeline
- 2024-01-03: disclosed: Published in GitHub Advisory Database
- 2023-12-18: patched: Patch released in version 2.6.0