On Thu, 3 Apr 2025, Takashi Yano via Cygwin wrote:
> It seems that raw_write() is called before returning from
> pthread::atforkchild() in fork::child().
>
> Moving _my_tls.fixup_after_fork() before atforkchild() seems
> to solve the issue.
>
> What about:
> ld_preload ();
> fixup_hooks_after_fork ();
> ch.rd_proc_pipe = ch.wr_proc_pipe = NULL;
> ?
>
> Should any of these be before pthread::atforkchild() asl well?
>
> I'm now performing a long-run test with the patch:
>
> diff --git a/winsup/cygwin/fork.cc b/winsup/cygwin/fork.cc
> index 0742ab363..28daf2682 100644
> --- a/winsup/cygwin/fork.cc
> +++ b/winsup/cygwin/fork.cc
> @@ -187,7 +187,6 @@ frok::child (volatile char * volatile here)
>
> ForceCloseHandle1 (fork_info->forker_finished, forker_finished);
>
> - pthread::atforkchild ();
> cygbench ("fork-child");
> ld_preload ();
> fixup_hooks_after_fork ();
> @@ -196,6 +195,7 @@ frok::child (volatile char * volatile here)
> rd_proc_pipe that would be an invalid handle. In the case of
> wr_proc_pipe it would be == my_wr_proc_pipe. Both would be bad. */
> ch.rd_proc_pipe = ch.wr_proc_pipe = NULL;
> + pthread::atforkchild ();
> CloseHandle (hParent);
> hParent = NULL;
> cygwin_finished_initializing = true;
pthread::atforkchild () calls user callbacks, right (pthread_atfork(3))?
Perhaps it should do so after cygwin_finished_initializing is set to
true, in case any functions that the callback might call would be
expecting that?
My encounter with pthread_atfork(3) in the past was an issue in glibc due
to some code that thought it'd be a good idea to mess with atfork handlers
inside an atfork handler. I can't remember now whether it was registering
a new one, or dlclose-ing a shared library that then caused handlers
registered in that library to be unregistered. I think the code in
question (in opensc) stopped doing that before glibc got fixed not to
screw up in that scenario.
--
Problem reports: https://cygwin.com/problems.html
FAQ: https://cygwin.com/faq/
Documentation: https://cygwin.com/docs.html
Unsubscribe info: https://cygwin.com/ml/#unsubscribe-simple
