Runtime.exec : vfork() concerns and a fix proposal

David Lloyd david.lloyd at redhat.com
Wed Sep 12 15:16:39 UTC 2018


On Wed, Sep 12, 2018 at 9:58 AM Roger Riggs <roger.riggs at oracle.com> wrote:
>
> Hi David,
>
> How does your proposal differ from the current posix_spawn and
> jspawnhelper MODE_POSIX_SPAWN
> implementation available on Solaris and AIX?

Reading through the code, it ends up being pretty much identical
AFAICT; if I understand correctly, Linux would be using vfork (or its
equivalent) internally in this case.

> This area is sensitive enough that it would be prudent to implement it
> as a new mode in ProcessImpl/ProcessImpl; retaining the existing options.
> Changing the default could be a build option and would require extensive
> testing and a long stability period.

Seems worthwhile though, given vfork's now-10-year-old obsolescence.
It looks like Linux is the only platform still using vfork for
ProcessImpl in OpenJDK.

-- 
- DML


More information about the core-libs-dev mailing list