[
https://issues.apache.org/jira/browse/POOL-372?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17653703#comment-17653703
]
Phil Steitz commented on POOL-372:
----------------------------------
I am not a security expert, so I can't comment on the assertion in (b) above,
but at least as of 2.10, (a) does not apply. I like the idea of replacing the
impl using Flight recorder though, if we can figure out a way to do it. The
reason that this exists is to allow pooled objects to hold onto the stack trace
leading to their creation. What is not obvious to me about how to use Flight
Recorder is knowing what to capture in advance.
> CallStackUtils mishandles security manager check part 2
> -------------------------------------------------------
>
> Key: POOL-372
> URL: https://issues.apache.org/jira/browse/POOL-372
> Project: Commons Pool
> Issue Type: Bug
> Reporter: Volker Kleinschmidt
> Priority: Major
>
> This ticket is for (b).
> CallStackUtils determines at initialization time whether it is allowed to
> create a security manager, then sticks that info into a static variable and
> never checks it again, relying on this check to later try to create a
> SecurityManager for a SecurityManagerCallStack. This is doubly wrong:
> a) If the code is running in a privileged context at init time, it determines
> that it can create a security manager, and then later naively assumes that
> henceforth all code is privileged and also can create a security manager. Of
> course this is not true, otherwise one would not need a security manager in
> the first place! This info can never be kept in a static variable, it's
> extremely context-dependent. So this leads to AccessControlException from
> invoking newCallStack() if abandoned object logging is enabled.
> b) The permission to create a security manager must never be granted to any
> code, unless that code has AllPermission in the first place, i.e. is already
> fully privileged. This is because this permission allows circumventing the
> security manager completely (simply create one that lets all checks pass).
> Therefore even just checking whether you're allowed to create a secmgr is
> naive - if a secmgr is installed at all you should assume that you're NOT
> privileged enough to do this, simply because for sure some code that calls
> your code will not be privileged enough.
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
