Bug #14882: OpenSSL::PKey::*.new does not resume after spurious signals

• Status: Third Party's Issue
• Assignee: rhenium (Kazuki Yamaguchi)
• Target version: -
• Backport: 2.3: UNKNOWN, 2.4: UNKNOWN, 2.5: UNKNOWN

[normalperson (Eric Wong)]

I noticed some OpenSSL::PKey::*.new failures while working on [Bug #14867].

MJIT will fork processes and when they die, SIGCHLD will trigger on most OSes.
I work around this problem in ext/openssl (and possibly other exts) by
avoiding signal_enqueue if no Ruby-level trap(:CHLD) handler is defined:

(in signal.c::sighandler):

        /* avoid spurious wakeup in main thread iff nobody uses trap(:CHLD) */
        if (vm && ACCESS_ONCE(VALUE, vm->trap_list.cmd[sig])) {
            signal_enque(sig);
        }

However, other signals may be received that do not intend to permanently
stop the operation.

Attached is a script which demonstrates the problem.

When performing normal interruptible operations (e.g. IO#read),
Ruby automatically retries on EINTR, so I think OpenSSL::PKey::*.new
should do the same instead of failing completely.


I do not use proprietary messaging systems, so please forward to
the other bug tracker if you don't want to use this one.  Thanks.

[rhenium (Kazuki Yamaguchi)]

• Status changed from Open to Third Party's Issue

Thank you for the report. I made a GitHub PR that fixes the bug at the GitHub repo (add .patch to the URL for "git format-patch"-style diff):

https://github.com/ruby/openssl/pull/205