Vulnerabilities > CVE-2021-37665 - Improper Input Validation vulnerability in Google Tensorflow
Summary
TensorFlow is an end-to-end open source platform for machine learning. In affected versions due to incomplete validation in MKL implementation of requantization, an attacker can trigger undefined behavior via binding a reference to a null pointer or can access data outside the bounds of heap allocated arrays. The [implementation](https://github.com/tensorflow/tensorflow/blob/460e000de3a83278fb00b61a16d161b1964f15f4/tensorflow/core/kernels/mkl/mkl_requantization_range_per_channel_op.cc) does not validate the dimensions of the `input` tensor. A similar issue occurs in `MklRequantizePerChannelOp`. The [implementation](https://github.com/tensorflow/tensorflow/blob/460e000de3a83278fb00b61a16d161b1964f15f4/tensorflow/core/kernels/mkl/mkl_requantize_per_channel_op.cc) does not perform full validation for all the input arguments. We have patched the issue in GitHub commit 9e62869465573cb2d9b5053f1fa02a81fce21d69 and in the Github commit 203214568f5bc237603dbab6e1fd389f1572f5c9. The fix will be included in TensorFlow 2.6.0. We will also cherrypick this commit on TensorFlow 2.5.1, TensorFlow 2.4.3, and TensorFlow 2.3.4, as these are also affected and still in supported range.
Vulnerable Configurations
Part | Description | Count |
---|---|---|
Application | 39 |
Common Weakness Enumeration (CWE)
Common Attack Pattern Enumeration and Classification (CAPEC)
- Buffer Overflow via Environment Variables This attack pattern involves causing a buffer overflow through manipulation of environment variables. Once the attacker finds that they can modify an environment variable, they may try to overflow associated buffers. This attack leverages implicit trust often placed in environment variables.
- Server Side Include (SSI) Injection An attacker can use Server Side Include (SSI) Injection to send code to a web application that then gets executed by the web server. Doing so enables the attacker to achieve similar results to Cross Site Scripting, viz., arbitrary code execution and information disclosure, albeit on a more limited scale, since the SSI directives are nowhere near as powerful as a full-fledged scripting language. Nonetheless, the attacker can conveniently gain access to sensitive files, such as password files, and execute shell commands.
- Cross Zone Scripting An attacker is able to cause a victim to load content into their web-browser that bypasses security zone controls and gain access to increased privileges to execute scripting code or other web objects such as unsigned ActiveX controls or applets. This is a privilege elevation attack targeted at zone-based web-browser security. In a zone-based model, pages belong to one of a set of zones corresponding to the level of privilege assigned to that page. Pages in an untrusted zone would have a lesser level of access to the system and/or be restricted in the types of executable content it was allowed to invoke. In a cross-zone scripting attack, a page that should be assigned to a less privileged zone is granted the privileges of a more trusted zone. This can be accomplished by exploiting bugs in the browser, exploiting incorrect configuration in the zone controls, through a cross-site scripting attack that causes the attackers' content to be treated as coming from a more trusted page, or by leveraging some piece of system functionality that is accessible from both the trusted and less trusted zone. This attack differs from "Restful Privilege Escalation" in that the latter correlates to the inadequate securing of RESTful access methods (such as HTTP DELETE) on the server, while cross-zone scripting attacks the concept of security zones as implemented by a browser.
- Cross Site Scripting through Log Files An attacker may leverage a system weakness where logs are susceptible to log injection to insert scripts into the system's logs. If these logs are later viewed by an administrator through a thin administrative interface and the log data is not properly HTML encoded before being written to the page, the attackers' scripts stored in the log will be executed in the administrative interface with potentially serious consequences. This attack pattern is really a combination of two other attack patterns: log injection and stored cross site scripting.
- Command Line Execution through SQL Injection An attacker uses standard SQL injection methods to inject data into the command line for execution. This could be done directly through misuse of directives such as MSSQL_xp_cmdshell or indirectly through injection of data into the database that would be interpreted as shell commands. Sometime later, an unscrupulous backend application (or could be part of the functionality of the same application) fetches the injected data stored in the database and uses this data as command line arguments without performing proper validation. The malicious data escapes that data plane by spawning new commands to be executed on the host.