[ 
https://issues.apache.org/jira/browse/NIFI-13414?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17856017#comment-17856017
 ] 

David Handermann commented on NIFI-13414:
-----------------------------------------

Thanks for the question [~ferdei].

As Joe highlighted, implying a level of protection can be more dangerous 
because it leads to a false sense of security.

Without going into all the details, any kind of encryption implementation needs 
to consider the threat model involved, particular in terms of potential points 
of attack. As mentioned, anything that is capable of reading files that the 
NiFi user owns is capable of reading property values encrypted with a local key.

Given that the implementation spans NiFi, NiFi Registry, and MiNiFi, these 
changes would apply across the board.



> Remove Property Protection Modules and Encrypt Config Tools
> -----------------------------------------------------------
>
>                 Key: NIFI-13414
>                 URL: https://issues.apache.org/jira/browse/NIFI-13414
>             Project: Apache NiFi
>          Issue Type: Improvement
>            Reporter: David Handermann
>            Assignee: David Handermann
>            Priority: Major
>
> NiFi and NiFi Registry have supported several strategies for encrypting and 
> decrypting application properties located in {{nifi.properties}} apart from 
> protection of sensitive values in the flow configuration. The initial 
> implementation supported property encryption using AES-GCM with key located 
> in {{bootstrap.conf}}. Subsequent implementations provided integration with 
> external secret management services. Supporting each of these implementations 
> requires a large number of third-party libraries, and does not provide a 
> public method for extensible implementation. Issues with both the security 
> and maintainability of these existing approaches necessitates their 
> deprecation and removal from the current main branch.
> The local AES-GCM implementation does not provide sufficient security from a 
> holistic perspective of the installation. Although values in 
> {{nifi.properties}} can be encrypted, the encryption key must be stored in 
> plaintext in {{bootstrap.conf}}, and both of these files are located in the 
> {{conf}} directory. Anyone with access to read the filesystem as the 
> operating system user can put these configurations together to read the 
> values in {{nifi.properties}}.
> The service-based implementations provide externalization using property 
> value references or encrypted values that require interaction with the 
> service for reading. This approach is beneficial, but it maintaining separate 
> implementations for each service provider, and it also requires configuring 
> access credentials in supplementary bootstrap configuration files. These 
> service-based implementations have large dependency trees, the contents of 
> each is stored in the {{properties}} directory under the {{lib}} directory. 
> Incorporating copies of service provider libraries for all supported 
> implementations adds significant weight to the standard distribution, and 
> makes it more difficult to maintain, given the lack of dependency isolation.
> The existing {{nifi-property-protection-api}} and provided implementations do 
> not support a maintainable pattern for integration application property 
> security. The {{nifi-toolkit-encrypt-config}} module also contains a 
> significant amount of code required to run out-of-band for encrypting 
> application properties. The {{encrypt-config}} command is packaged apart from 
> the standard NiFi distribution, making it less useful for common deployment 
> scenarios.
> Taking these issues together, existing property protection modules for 
> {{nifi.properties}} should be removed from the main branch. This will provide 
> a streamlined distribution in the short term, and also provide a better 
> foundation for consideration more robust approaches that are not subject to 
> the same types of security concerns.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

Reply via email to