Re: llvm & RTTI over shared libraries

From: <jbo_at_insane.engineer> wrote on
Date: Thu, 14 Apr 2022 16:36:24 +0000 :
(I've line-split the text.)

> I'm in the middle of moving to FreeBSD as my primary development platform (desktop wise).
> As such, I am currently building various software tools I've written over the years on
> FreeBSD for the first time. Most of those were developed on either Linux+GCC or on
> Windows+Mingw (MinGW -> GCC).
> 
> Today I found myself debugging a piece of software which runs fine on FreeBSD when
> compiled with gcc11 but not so much when compiling with clang14.
> I managed to track down the problem but I lack the deeper understanding to resolve
> this properly - so here we are.
> 
> The software in question is written in C++20 and consisting of:
>   - An interface library (just a bunch of header files).
>   - A main executable.
>   - A bunch of plugins which the executable loads via dlopen().
> 
> The interface headers provide several types. Lets call them A, B, C and D. where B,
> C and D inherit from A.
> The plugins use std::dynamic_pointer_cast() to cast an std::shared_ptr<A> (received
> via the plugin interface) to the derived classes such as std::shared_ptr<B>.
> This is where the trouble begins.
> 
> If everything (the main executable and the plugins) are compiled using gcc11, everything
> works "as I expect it".
> However, when compiling everything with clang14, the main executable is able to load the
> plugins successfully but those std::dynamic_pointer_cast() calls within the plugins
> always return nullptr.
> 
> After some research I seem to understand that the way that RTTI is handled over shared
> library boundaries is different between GCC and LLVM.
> This is where my understanding starts to get less solid.
> 
> I read the manual page of dlopen(3). It would seem like the flag RTLD_GLOBAL would be
> potentially interesting to me: "Symbols from this shared object [...] of needed objects
> will be available for re-solving undefined references from all other shared objects."
> The software (which "works as intended" when compiled with GCC) was so far only calling
> dlopen(..., RTLD_LAZY).
> I'm not even sure whether this applies to my situation. My gut feeling tells me that I'm
> heading down the wrong direction here. After all, the main executable is able to load
> the plugins and to call the plugin's function which receives an std::shared_ptr<A>
> asparameter just fine, also when compiled with LLVM.
> Is the problem I'm experiencing related to the way that the plugin (shared library) is
> loaded or the way that the symbols are being exported?
> In the current state, the plugins do not explicitly export any symbols.
> 
> Here's a heavily simplified version of my scenario:

The simplified example was not designed to compile and test.
So I made guesses and made my own. The .cpp files have
comments on the compile/link commands used and there are
examples of c++ and g++11 compile/link/run sequences
after the source code. The code is not well commented. Nor
does it deal with error handling or the like. But it is
fairly short overall.

# more base_plugin.h
#include  <memory>

// For its own libbase_plugin.so file, load time bound, no dlopen used for it:

struct base
{
        virtual ~base();
};
struct base_plugin
{
        virtual std::shared_ptr<base> create_data_instance()             = 0;
        virtual void                  action(std::shared_ptr<base> data) = 0;
        virtual ~base_plugin();
};

extern "C" // for each derived plugin .so file:
{
        using plugin_instance_creator= base_plugin* (*)();
        const char plugin_instance_creator_name[] = "create_plugin_instance"; // Lookup via dlsym.

        using plugin_instance_destroyer= void (*)(base_plugin*);
        const char plugin_instance_destroyer_name[] = "destroy_plugin_instance"; // Lookup via dlsym.
};

# more base_plugin.cpp
// c++   -std=c++20 -O0 -g -fPIC -lc++    -olibbase_plugin.so -shared base_plugin.cpp
// g++11 -std=c++20 -O0 -g -fPIC -lstdc++ -olibbase_plugin.so -shared base_plugin.cpp

#include "base_plugin.h"

base::~base()               {}
base_plugin::~base_plugin() {}

# more main_using_plugin.cpp 
// c++   -std=c++20 -O0 -g -fPIC -lc++    -L. -lbase_plugin -Wl,-rpath=. \
//                                       -omain_using_plugin main_using_plugin.cpp
// g++11 -std=c++20 -O0 -g -fPIC -lstdc++ -L. -lbase_plugin -Wl,-rpath=. \
//       -Wl,-rpath=/usr/local/lib/gcc11 -omain_using_plugin main_using_plugin.cpp

#include "base_plugin.h"
#include <dlfcn.h>

int main()
{
        auto dl= dlopen("./libsharedlib_plugin.so",RTLD_LAZY); // hardcoded .so path for the example

        union { void* as_voidptr; plugin_instance_creator as_plugin_instance_creator; } creator_plugin_func;
        creator_plugin_func.as_voidptr= dlsym(dl,plugin_instance_creator_name);

        union { void* as_voidptr; plugin_instance_destroyer as_plugin_instance_destroyer; } destroyer_plugin_func;
        destroyer_plugin_func.as_voidptr= dlsym(dl,plugin_instance_destroyer_name);

        auto plugin= (creator_plugin_func.as_plugin_instance_creator)();

        { // Local scope for data
                std::shared_ptr<base> data{plugin->create_data_instance()};
                plugin->action(data);
        } // Presume for the example that nothing requires the plugin after here.

        (destroyer_plugin_func.as_plugin_instance_destroyer)(plugin);
        destroyer_plugin_func.as_voidptr= nullptr;

        dlclose(dl);
}

NOTE: So, other than the dlopen, the above has no direct tie to
the specific dynamically loaded plugin. The base_plugin is in a
.so but is load-time bound instead of using dlopen. That .so
would be used by all the plugins found via dllopen. (I only
made one example.)

As for the .so used via dlopen/dlsym/dlclose . . .

# more sharedlib_plugin.h
#include "base_plugin.h"

// For its own libsharedlib_plugin.so file, where dlopen is used to find it:

struct sharedlib : base { int v; };
struct sharedlib_plugin : base_plugin
{
        std::shared_ptr<base> create_data_instance()             override;
        void                  action(std::shared_ptr<base> base) override;
};

# more sharedlib_plugin.cpp
// c++   -std=c++20 -O0 -g -fPIC -lc++    -olibsharedlib_plugin.so -shared sharedlib_plugin.cpp
// g++11 -std=c++20 -O0 -g -fPIC -lstdc++ -olibsharedlib_plugin.so -shared sharedlib_plugin.cpp

#include "sharedlib_plugin.h"
#include <iostream>

std::shared_ptr<base> sharedlib_plugin::create_data_instance()
{
        std::cout << "create_data_instance in use from dlopen'd .so\n";
        return std::static_pointer_cast<base>(std::make_shared<sharedlib>());
}

void sharedlib_plugin::action(std::shared_ptr<base> b)
{
        std::cout << "action in use from dlopen'd .so class\n";
        auto separate_share = std::dynamic_pointer_cast<sharedlib>(b);
        if (separate_share->v || 1 < separate_share.use_count())
                std::cout << "separate_share is not nullptr (would crash otherwise)\n";
}

extern "C" base_plugin* create_plugin_instance()
{
        std::cout << "create_plugin_instance in use from dlopen'd .so\n";
        return new sharedlib_plugin();
}

extern "C" void destroy_plugin_instance(const base_plugin* plugin)
{
        std::cout << "destroy_plugin_instance in use from dlopen'd .so\n";
        delete plugin;
}

# c++ -std=c++20 -O0 -g -fPIC -lc++ -olibbase_plugin.so -shared base_plugin.cpp
# c++ -std=c++20 -O0 -g -fPIC -lc++ -L. -lbase_plugin -Wl,-rpath=. \
      -omain_using_plugin main_using_plugin.cpp
# c++ -std=c++20 -O0 -g -fPIC -lc++ -olibsharedlib_plugin.so -shared sharedlib_plugin.cpp
# ./main_using_plugin
create_plugin_instance in use from dlopen'd .so
create_data_instance in use from dlopen'd .so
action in use from dlopen'd .so class
separate_share is not nullptr (would crash otherwise)
destroy_plugin_instance in use from dlopen'd .so

For reference:

# ldd main_using_plugin
main_using_plugin:
        libc++.so.1 => /lib/libc++.so.1 (0x819d0000)
        libcxxrt.so.1 => /lib/libcxxrt.so.1 (0x82735000)
        libbase_plugin.so => ./libbase_plugin.so (0x8328d000)
        libm.so.5 => /lib/libm.so.5 (0x83c47000)
        libgcc_s.so.1 => /lib/libgcc_s.so.1 (0x85861000)
        libc.so.7 => /lib/libc.so.7 (0x848f9000)

# ldd ./libsharedlib_plugin.so
./libsharedlib_plugin.so:
        libc++.so.1 => /lib/libc++.so.1 (0x3b69aeeb6000)
        libcxxrt.so.1 => /lib/libcxxrt.so.1 (0x3b69af6f2000)
        libm.so.5 => /lib/libm.so.5 (0x3b69afd1f000)
        libgcc_s.so.1 => /lib/libgcc_s.so.1 (0x3b69b0303000)
        libc.so.7 => /lib/libc.so.7 (0x3b69aafdb000)


As for g++11 use . . .

Testing with g++11 does involve additional/adjusted command line
options:
    -Wl,-rpath=/usr/local/lib/gcc11/ ( for main_using_plugin.cpp )
    -lstdc++ (for all 3 .cpp files)

(FreeBSD's libgcc_s.so.1 does not cover everything needed for
all architectures for g++11's code generation. I was working
in a context where using /usr/local/lib/gcc11//libgcc_s.so.1
was important.)


# g++11 -std=c++20 -O0 -g -fPIC -lstdc++ -olibbase_plugin.so -shared base_plugin.cpp
# g++11 -std=c++20 -O0 -g -fPIC -lstdc++ -L. -lbase_plugin -Wl,-rpath=. \
      -Wl,-rpath=/usr/local/lib/gcc11 -omain_using_plugin main_using_plugin.cpp
# g++11 -std=c++20 -O0 -g -fPIC -lstdc++ -olibsharedlib_plugin.so -shared sharedlib_plugin.cpp
# ./main_using_plugin
create_plugin_instance in use from dlopen'd .so
create_data_instance in use from dlopen'd .so
action in use from dlopen'd .so class
separate_share is not nullptr (would crash otherwise)
destroy_plugin_instance in use from dlopen'd .so

For reference:

# ldd main_using_plugin
main_using_plugin:
        libstdc++.so.6 => /usr/local/lib/gcc11//libstdc++.so.6 (0x83a00000)
        libbase_plugin.so => ./libbase_plugin.so (0x8213d000)
        libm.so.5 => /lib/libm.so.5 (0x82207000)
        libgcc_s.so.1 => /usr/local/lib/gcc11//libgcc_s.so.1 (0x82c66000)
        libc.so.7 => /lib/libc.so.7 (0x849c4000)

# ldd ./libsharedlib_plugin.so
./libsharedlib_plugin.so:
        libstdc++.so.6 => /usr/local/lib/gcc11/libstdc++.so.6 (0x1c2a7b800000)
        libm.so.5 => /lib/libm.so.5 (0x1c2a7bb1c000)
        libgcc_s.so.1 => /lib/libgcc_s.so.1 (0x1c2a7c416000)
        libc.so.7 => /lib/libc.so.7 (0x1c2a780e8000)


Overall:

Looks to me like both the system clang/llvm and g++11 contexts
are working. (The platform context was aarch64 main [so: 14],
in case it matters.)

===
Mark Millard
marklmi at yahoo.com