RFR: 8319332: Security properties files inclusion [v19]

[Martin Balao]

On Wed, 7 Aug 2024 20:09:09 GMT, Francisco Ferrari Bihurriet <fferrari at openjdk.org> wrote:

>> The implementation of this proposal is based on the requirements, specification and design choices described in the [JDK-8319332] ticket and its respective CSR [JDK-8319333]. What follows are implementation notes organized per functional component, with the purpose of assisting to navigate the code changes in this pull-request.
>> 
>> ## Security properties loading (overview)
>> 
>> A new static class named `SecPropLoader` (nested within `java.security.Security`) is introduced to handle the loading of all security properties. Its method `loadAll` is the first one to be called, at `java.security.Security` static class initialization. The master security properties file is then loaded by `loadMaster`. When additional security properties files are allowed (the security property `security.overridePropertiesFile` is set to `true`) and the `java.security.properties` system property is passed, the method `loadExtra` handles the extra load.
>> 
>> The master properties file is loaded in `OVERRIDE` mode, meaning that the map of properties is originally empty. Any failure occurred while loading these properties is considered fatal. The extra properties file (`java.security.properties`) may be loaded in `OVERRIDE` or `APPEND` mode. Any failure in this case is ignored. This behavior maintains compatibility with the previous implementation.
>> 
>> While the `java.security.properties` system property is documented to accept an URL type of value, filesystem path values are supported in the same way that they were prior to this enhancement. Values are then interpreted as paths and, only if that fails, are considered URLs. In the latter case, there is one more attempt after opening the stream to check if there is a local file path underneath (e.g. the URL has the form of `file:///path/to/a/local/file`). The reason for preferring paths over URLs is to support relative path file inclusion in properties files.
>> 
>> ## Loading security properties from paths (`loadFromPath` method)
>> 
>> When loading a properties file from a path, the normalized file location is stored in the static field `currentPath`. This value is the current base to resolve any relative path encountered while handling an _include_ definition. Normalized paths are also saved in the `activePaths` set to detect recursive cycles. As we move down or up in the _includes_ stack, `currentPath` and `activePaths` values are updated.
>> 
>> ## Loading security properties from URLs (`loadFromUrl` method)
>> 
>> The extra properti...
>
> Francisco Ferrari Bihurriet has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Code review and additional changes
>   
>   Throw an IllegalArgumentException exception if Security.getProperty() is
>   invoked with "include" as key, also add a check in the test case.
>   
>   Initialize java.security.Security::props in a single place, and make it
>   final.
>   
>   Make sun.security.util.PropertyExpander::expandNonStrict() throw
>   AssertionError instead of RuntimeException.
>   
>   Remove "this file" reference in java.security.
>   
>   Additionally, use java.security.Security::check() to check the
>   'getProperty.<key>' permission, as with 'setProperty.<key>'.
>   
>   Co-authored-by: Martin Balao <mbalao at redhat.com>
>   Co-authored-by: Francisco Ferrari Bihurriet <fferrari at redhat.com>

If someone is really using _include_ as a security property, this is going to be a breaking change anyways because the value for the property will be anything (i.e. not a file to be included with this new enhancement) and we will be enforcing behavior and restrictions now. Looks like the focus is on `setProperty` in this last part of the discussion but the same is true for a definition of the property in a _java.security_ file.

Trying to save backward compatibility in `setProperty` will be partial and for very unlikely scenarios. I don't think that it's a good idea to support it as a real security property because it's really a directive and seeing it as a property would add to the confusion. For example, does an override of the security property mean that previous definitions were not applied as a directive? I'm concerned about the cost of this initiative.

If we resign any intention to backport this enhancement in the future, would the IAE solution —with the addition of mentioning _include_ as one of the reserved words— work for you?

-------------

PR Comment: https://git.openjdk.org/jdk/pull/16483#issuecomment-2293830485