Re: a exit() bug in multithreaded program

[Dan Kegel]

SHEN Feng wrote:
> In the function exit(), the global variable __exit_funcs isn't protected
> for multithreaded program. But __exit_funcs at __new_exit_fn() in
> cxa_atexit.c is protected by a lock. The function __new_exit_fn() is
> called by atexit() and on_exit(). So __exit_funcs at exit() should also
> be protected by the lock. It’s a bug.
>
> I have had a test on this function. The test function exit_test.c and
> the result file exit_test_result_redhat.log are attached. From them we
> can concluded that the exit() may be unsafe in multithreaded program.

Hi Feng,
you didn't mention which version of glibc or the operating system;
that makes it a bit harder to help.

Assuming you're using the version shipped with Red Hat 9,
this sounds vaguely related to a problem mentioned on the NPTL mailing list:
https://listman.redhat.com/archives/phil-list/20
You might want to subscribe there; see
https://listman.redhat.com/mailman/listinfo/phil-list
- Dan