Re: malloc, vfork, and OpenJDK

[Florian Weimer <fweimer at redhat dot com>]

* Adhemerval Zanella:

>> And OpenJDK recently switched to posix_spawn + jspawnhelper for its
>> default process launch method, so that bug is going to go away
>> eventually.  But Java binaries tend to stick around for a long time.
>
> So the idea is to focus on making malloc work correctly with vfork,
> so old OpenJDK which uses vfork for process spawning does not issue
> undefined behaviour? Or the remaining bugs are not really affecting
> OpenJDK and the idea is start to define a strict environment where
> malloc after vfork would be safe?

I think this is an unintended, but very relevant constraint on the
future evolution of malloc.  This scenario has to keep working, even
though it's undefined behavior.

> It looks like even the *option* to use posix_spawn is a recent addition 
> to openjdk [2]. So the question I have is how effective would to focus on 
> backport the option to use posix_spawn on OpenJDK.

OpenJDK backports are unrealistic.  People will use some download of the
free/gratuit version of Oracle JDK and not upgrade to the commercial,
still-maintained version.  So even if we can get the fix into OpenJDK 8,
it's not likely to be available to users in practice.

That's why I think we can't break this usage scenario.

> We can focus on extending posix_spawn to add the missing features to remove 
> the requirement of jspawnhelper (which is orthogonal to the issue).

Yes, it's independent.

> [1] https://zatrazz.github.io/More-About-Lauching-Processes/

Nice, thanks for writing this up.

Florian