[jira] [Commented] (MESOS-6216) LibeventSSLSocketImpl::create is not safe to call concurrently with os::getenv

["Vinod Kone (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MESOS-6216?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15550550#comment-15550550 ] 

Vinod Kone commented on MESOS-6216:
-----------------------------------

What's the ETA for this to land on master and backported to 1.0.x?

> LibeventSSLSocketImpl::create is not safe to call concurrently with os::getenv
> ------------------------------------------------------------------------------
>
>                 Key: MESOS-6216
>                 URL: https://issues.apache.org/jira/browse/MESOS-6216
>             Project: Mesos
>          Issue Type: Bug
>          Components: security
>            Reporter: Benjamin Bannier
>            Assignee: Benjamin Bannier
>              Labels: mesosphere
>         Attachments: build.log
>
>
> {{LibeventSSLSocketImpl::create}} is called whenever a potentially ssl-enabled socket is created. It in turn calls {{openssl::initialize}} which calls a function {{reinitialize}} using {{os::setenv}}. Here {{os::setenv}} is used to set up SSL-related libprocess environment variables {{LIBPROCESS_SSL_*}}.
> Since {{os::setenv}} is not thread-safe just like the {{::setenv}} it wraps, any calling of functions like {{os::getenv}} (or via {{os::environment}}) concurrently with the first invocation of {{LibeventSSLSocketImpl::create}} performs unsynchronized r/w access to the same data structure in the runtime.
> We usually perform most setup of the environment before we start the libprocess runtime with {{process::initialize}} from a {{main}} function, see e.g., {{src/slave/main.cpp}} or {{src/master/main.cpp}} and others. It appears that we should move the setup of libprocess' SSL environment variables to a similar spot.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)