In Tensorflow before versions 1.15.4, 2.0.3, 2.1.2, 2.2.1 and 2.3.1, changing the TensorFlows SavedModel protocol buffer and altering the name of required keys results in segfaults and data corruption while loading the model. This can cause a denial of service in products using tensorflow-serving or other inference-as-a-service installments. Fixed were added in commits f760f88b4267d981e13f4b302c437ae800445968 and fcfef195637c6e365577829c4d67681695956e7d (both going into TensorFlow 2.2.0 and 2.3.0 but not yet backported to earlier versions). However, this was not enough, as #41097 reports a different failure mode. The issue is patched in commit adf095206f25471e864a8e63a0f1caef53a0e3a6, and is released in TensorFlow versions 1.15.4, 2.0.3, 2.1.2, 2.2.1, or 2.3.1.
The product receives input or data, but it does not validate or incorrectly validates that the input has the properties that are required to process the data safely and correctly.
| Name | Vendor | Start Version | End Version |
|---|---|---|---|
| Tensorflow | * | 1.15.4 (excluding) | |
| Tensorflow | 2.0.0 (including) | 2.0.3 (excluding) | |
| Tensorflow | 2.1.0 (including) | 2.1.2 (excluding) | |
| Tensorflow | 2.2.0 (including) | 2.2.1 (excluding) | |
| Tensorflow | 2.3.0 (including) | 2.3.1 (excluding) |
Input validation is a frequently-used technique for checking potentially dangerous inputs in order to ensure that the inputs are safe for processing within the code, or when communicating with other components. When software does not validate input properly, an attacker is able to craft the input in a form that is not expected by the rest of the application. This will lead to parts of the system receiving unintended input, which may result in altered control flow, arbitrary control of a resource, or arbitrary code execution. Input validation is not the only technique for processing input, however. Other techniques attempt to transform potentially-dangerous input into something safe, such as filtering (CWE-790) - which attempts to remove dangerous inputs - or encoding/escaping (CWE-116), which attempts to ensure that the input is not misinterpreted when it is included in output to another component. Other techniques exist as well (see CWE-138 for more examples.) Input validation can be applied to:
Data can be simple or structured. Structured data can be composed of many nested layers, composed of combinations of metadata and raw data, with other simple or structured data. Many properties of raw data or metadata may need to be validated upon entry into the code, such as:
Implied or derived properties of data must often be calculated or inferred by the code itself. Errors in deriving properties may be considered a contributing factor to improper input validation.
Note that “input validation” has very different meanings to different people, or within different classification schemes. Caution must be used when referencing this CWE entry or mapping to it. For example, some weaknesses might involve inadvertently giving control to an attacker over an input when they should not be able to provide an input at all, but sometimes this is referred to as input validation. Finally, it is important to emphasize that the distinctions between input validation and output escaping are often blurred, and developers must be careful to understand the difference, including how input validation is not always sufficient to prevent vulnerabilities, especially when less stringent data types must be supported, such as free-form text. Consider a SQL injection scenario in which a person’s last name is inserted into a query. The name “O’Reilly” would likely pass the validation step since it is a common last name in the English language. However, this valid name cannot be directly inserted into the database because it contains the “’” apostrophe character, which would need to be escaped or otherwise transformed. In this case, removing the apostrophe might reduce the risk of SQL injection, but it would produce incorrect behavior because the wrong name would be recorded.