[patch] Fix for bz14333 -- race between atexit() and exit()

[Paul Pluzhnikov <ppluzhnikov at google dot com>]

Greetings,

This patch is an update to previous patch by Anoop:
https://sourceware.org/ml/libc-alpha/2008-09/msg00013.html

See discussion in the above message.

Changes from the previous patch: I extended the lifetime of the lock
from __new_exitfn to its callers, since they *also* modify global
state.

Tested: "make check" on Linux/x86_64.
Also ran the test from
https://sourceware.org/bugzilla/show_bug.cgi?id=14333#c5 a few hundred
times.


2017-07-10  Paul Pluzhnikov  <ppluzhnikov@google.com>
                    Anoop V Chakkalakkal <anoop.vijayan@in.ibm.com>

        [BZ #14333]
        * stdlib/cxa_atexit.c (__internal_atexit): Use __exit_funcs_lock.
        (__new_exitfn): Fail registration when we finished at_exit processing.
        * stdlib/on_exit.c (__on_exit): Likewise.
       * stdlib/exit.c (__exit_funcs_done): New variable.
        (__run_exit_handlers): Use __exit_funcs_lock.
        * stdlib/exit.h (__exit_funcs_lock, __exit_funcs_done): New
declarations.

-- 
Paul Pluzhnikov

Attachment: glibc-bz14333-20170709.txt
Description: Text document