[Libguestfs] nbdkit build failure (was: Re: jplesnik's nbdkit-1.35.5-4.fc39 failed to build)

[Richard W.M. Jones]

On Thu, Jul 13, 2023 at 11:03:35AM +0100, Richard W.M. Jones wrote:
> On Thu, Jul 13, 2023 at 10:29:44AM +0100, Richard W.M. Jones wrote:
> > It's a crash in Perl code, although called from the nbdkit perl plugin
> > (C bindings).  Still debugging this ...
> > 
> > Core was generated by `/home/rjones/d/nbdkit/server/nbdkit -U /tmp/nbdkittPN34a/sock -P /tmp/nbdkittPN'.
> > Program terminated with signal SIGSEGV, Segmentation fault.
> > 
> > warning: Section `.reg-xstate/150177' in core file too small.
> > #0  Perl_my_snprintf (buffer=0x7f3a1e5fc630 "p\310_\036:\177", len=32, 
> >     format=0x7f3a1ef7e322 "_<(eval %lu)")
> >     at /usr/src/debug/perl-5.38.0-499.fc39.x86_64/util.c:5248
> > Downloading source file /usr/src/debug/perl-5.38.0-499.fc39.x86_64/util.c
> > 5248            WITH_LC_NUMERIC_SET_TO_NEEDED(                                  
> > [Current thread is 1 (Thread 0x7f3a1e5fd6c0 (LWP 150177))]
> > (gdb) bt
> > #0  Perl_my_snprintf (buffer=0x7f3a1e5fc630 "p\310_\036:\177", len=32, 
> >     format=0x7f3a1ef7e322 "_<(eval %lu)")
> >     at /usr/src/debug/perl-5.38.0-499.fc39.x86_64/util.c:5248
>
> https://github.com/Perl/perl5/blob/ff7e37d5ca2e55a1e53e5b375e3cc9aedc9ce4ec/util.c#L5248
> 
> Since its hardly obvious from the code, that macro expands to the
> monster below.  I didn't bother to format it completely ...
> 
> # 5245 "util.c"
>     if (retval == -1) {
>         do {
> 	  void (*_restore_LC_NUMERIC_function)(PerlInterpreter* my_perl __attribute__((unused))) = ((void *)0);
>           do { 
>            _Bool _in_lc_numeric = ((( ((
>              _Bool
>                ) ((my_perl->Icompiling).cop_hints & 0x00000004)) || ( ((_Bool) ((my_perl->Icompiling).cop_hints & 0x00000010)) && Perl__is_in_locale_category(my_perl, (1), (
>        1
>        )))) || (((my_perl->Icurcop) && (((my_perl->Icurcop))->cop_hints + 0) & 0x00000004) || (((my_perl->Icurcop) && (((my_perl->Icurcop))->cop_hints + 0) & 0x00000010) && Perl__is_in_locale_category(my_perl, (0), (
>        1
>        ))))));
>        ((void)0);
>        if (_in_lc_numeric) {
>          if ((! (my_perl->Inumeric_underlying) && (my_perl->Inumeric_standard) < 2)) { Perl_set_numeric_underlying(my_perl); _restore_LC_NUMERIC_function = &Perl_set_numeric_standard; }
>         } else {
> 	  if ((! (my_perl->Inumeric_standard))) { Perl_set_numeric_standard(my_perl); _restore_LC_NUMERIC_function = &Perl_set_numeric_underlying; } }
>         } while (0);
> 	retval = vsnprintf(buffer, len, format, ap);;
> 	do { if (_restore_LC_NUMERIC_function) { _restore_LC_NUMERIC_function(my_perl); } ((void)0); } while (0);
>         } while (0);
> 
> ... because the fundamental problem is:
> 
> (gdb) print my_perl
> $13 = (PerlInterpreter *) 0x0
> 
> so that's not gonna work.

It turns out to be a bug in nbdkit!  A close reading of the perlembed
man page reveals that we need to set the interpreter context when
calling into Perl from a new thread.  The fix is:

https://gitlab.com/nbdkit/nbdkit/-/commit/495708e9d1b396ff450df15d5789b4f6c51f6ebc

I'll add this to Rawhide shortly.

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
nbdkit - Flexible, fast NBD server with plugins
https://gitlab.com/nbdkit/nbdkit