core dump with cp, dlclose, dlopen

[ahoward]

all sources at

        http://www-fd.fsl.noaa.gov/~ahoward/shared/



all-


i am having some problems which SEEM to be related to the choice of using
'cp' or 'install' as an argument to libtool in install mode - though that
seems unbelievable to me (this IS my first project using libtool so i'm
relatively clueless)


general :

        i'm builing a shared lib full of configuration information,
        sort of a dynamic header file - the header file only grows,
        entries are never deleted - thus the -avoid-version flags... the
        whole things works great EXCEPT that my daemon emulator core dumps
        on dlclose when a lib is updated using cp, but NOT when using
        'install'!  the emulator does simply this :

                dlopens the lib
                retrieves data structure (just character data)

                loops, printing it's contents and
                waits for input to be typed

                IFF refresh is typed (emulating SIGHUP) it
                        dlcloses the lib
                        re-dlopens the lib
                        re-retrieves data structure

                re-enters loop.


        note that the lib ALWAYS remains open (i am note pointing into
        no-where) and that the core dump occurs on the dlclose complaining
        of

                REFRESHING LIB...
        84      if(lib_so) dlclose(lib_so);
        (gdb)

        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
        CORE DUMP HERE IFF LIB IS UPDATED AT RUNTIME USING CP BUT NOT
        INSTALL!!!

        code looks like:

        if refreshing...

                if(lib_so) dlclose(lib_so);  <- core dump!!!!!!
                err stuff...

                lib_so = dlopen(lib, RTLD_LAZY);
                err stuf...

                get_table = (table_ptr (*)(void))dlsym(lib_so, "get_table");
                err stuff...

                t = (*get_table)();
                err stuf ...


Program received signal SIGSEGV, Segmentation fault.
0x40017774 in __do_global_dtors_aux () at eval.c:41
41      eval.c: No such file or directory.
        in eval.c


files :
        table.h         ->      interface to lib data structure
        table.c         ->      const data structure
                                (const char * const * const *)
        driver.c        ->      above daemon emulator


build :

        lib:

        tableinstall: libtable.la
         $(LIBTOOL) $(INSTALL) $(BLDDIR)/libtable.la $(LIBDIR)/libtable.la

        libtable.la: table.lo
         @mkdir -p $(LIBDIR)
         $(LIBTOOL) $(GCC) $(GCCFLAGS) $(LINKFLAGS) -o
         $(BLDDIR)/libtable.la $(BLDDIR)/table.lo -rpath $(LIBDIR)

        table.lo: $(INCDIR)/table.h
         @mkdir -p $(BLDDIR)
         $(LIBTOOL) $(GCC) $(GCCFLAGS) -o $(BLDDIR)/table.lo -I $(INCDIR)
         -c $(SRCDIR)/table.c


        driver:

        $(BINDIR)/d: $(BLDDIR)/driver.o
        @mkdir -p $(BINDIR)
        $(GCC) $(GCCFLAGS) -o $(BINDIR)/d $(BLDDIR)/driver.o $(RPATHS)
        $(ASMFLAGS) -ldl
        @ printf "\n setenv LD_LIBRARY_PATH $(LIBDIR)\n"

        $(BLDDIR)/driver.o: $(SRCDIR)/driver.c  $(INCDIR)/table.h
        $(GCC) $(GCCFLAGS) -o $(BLDDIR)/driver.o -c -I $(INCDIR)/
        $(SRCDIR)/driver.c



relevant flags:

        LINKFLAGS = -module -avoid-version
        GASFLAGS  = -Wa,-R

        INSTALL   = cp                  OR              !!!
        INSTALL   = /usr/local/bin/install              !!!



so.  iff i run the driver i can always unload/load the table at will, even
if it has been rebuilt and installed but NOT if the INSTALL var is 'cp',
in which case i core dump.


the question:

any ideas anyone?  more generally if a lib plans to dlopen/close a lib
multiple times during execution is there any problems with updating the
actual lib backing store at the same time?  if not what is the best way to
accomplish  this?


-ara


ps.  source at http://www-fd.fsl.noaa.gov/~ahoward/shared/