Segmentation fault in Perl_csighandler when receiving SIGCHLD & using threads #13515
Comments
From thierry.vignaud@gmail.comI'have a program which uses threads (but not willingly) and has a custom signal handler for SIGCHLD.. If a SIGCHLD is received & handled by a custom signal handler before the thread creation by the libraries, perl segfaults. Here is the stacktrace from gdb: The segfaults happen since glib was upgraded from glib-2.34 to 2.36 & were seen with both perl-5.16 & 5.18. I've workarounded this in some tools by setting SIG{CHLD} later. I can still make mcc (Mageia Control Center) to reliably segfault by setting SIG{CHLD} earlier (by moving "$SIG{CHLD} =..." at top of /usr/libexec/drakconf) Basically, we segfault in SIG_CHLD when webkit threads exit (webkit creates temporary ones on init). So, as far as I'm concerned there's 2 bugs: There was an old bug that got fixed in perl-5.10.0: This report looks like a variant of this bug (and is happening b/c of glib's & webkit's threads). It can be reproduced with either the attached script or one of the following one-liners: perl -MGtk2 -MGtk2::WebKit -e '$SIG{CHLD} = sub { warn }; Gtk2->init; my $wiew = Gtk2::WebKit::WebView->new; $w=Gtk2::Window->new; $w->add($view); Gtk2->main' perl -MGtk3 -MGtk3::WebKit -e '$SIG{CHLD} = sub { warn @_ }; Gtk3->init; my $wiew = Gtk3::WebKit::WebView->new; $w=Gtk3::Window->new; $w->add($view); Gtk3->main' |
|
The RT System itself - Status changed from 'new' to 'open' |
From thierry.vignaud@gmail.comThis is perl-5.18.1 as provided by Mageia and is indeed patched by vendor: |
From @tonycozOn Tue Jan 07 16:48:15 2014, tonyc wrote:
tony@mars:.../git/perl2$ gdb --args ~/perl/5.18.0-thr-debug/bin/perl -MGtk2 -MGtk2::WebKit -e '$SIG{CHLD} = sub { warn }; Gtk2->init; my $view = Gtk2::WebKit::WebView->new; $w=Gtk2::Window->new; $w->add($view); Gtk2->main' Program received signal SIGSEGV, Segmentation fault. which I suspect means we need to initialize THX here in a similar way we do on Win32: void * Tony |
From thierry.vignaud@gmail.comCare to suggest a patch that I can test? |
From @LeontOn Tue Jan 07 16:06:32 2014, thierry.vignaud@gmail.com wrote:
This sounds like csighandler is running in a thread that doesn't have any associated perl interpreter. One solution might be to block SIGCHLD (and possibly other signals) during thread creation (see also Signal::Mask or POSIX::sigprocmask), that way the signal will never be delivered to them (unless they accidentally unblock it). Leon |
From thierry.vignaud@gmail.comOn Tue Jan 07 18:12:41 2014, LeonT wrote:
As stated in my initial report, I do not have control over the threads So I would have to identify every places in the libraries that can result |
From @shlomifI am able to reproduce this problem with a perl built like so: #!/bin/sh from the Git maint5.18 branch, on Mageia Linux x86-64 4 Cauldron, and after installing the Gtk2 and Gtk2::WebKit modules from CPAN. When I run the program as my normal user everything runs fine. When I run it as root, it crashes. When I run it under gdb as root, the program does not crash. Let me know if I can be of further assistance. Regards, -- Shlomi Fish |
From @tonycozOn Tue Jan 07 18:12:41 2014, LeonT wrote:
I'll look at fixing this as part of my 81074 work. Tony |
From thierry.vignaud@gmail.comOn 8 January 2014 01:48, Tony Cook via RT <perlbug-followup@perl.org> wrote:
This is perl-5.18.1 as provided by Mageia and is indeed patched by vendor: |
|
Looks that this doesn't happen anymore as of perl-5.30.1 |
|
Closing as the bug is reported to be fixed. |
Migrated from rt.perl.org#120951 (status was 'open')
Searchable as RT120951$
The text was updated successfully, but these errors were encountered: