Loading two different versions of the same shared library - c

I'm in a situation that's quite similar to the following. There's libA.so that depending on some compile time flags exhibits slightly different behaviour (it's an external lib, and I can't modify the source). Then, I have libB.so that depends on libA.so (compiled with say -DVALUE=1), and in my executable I depend both on libB.so, as well as on libA.so, but compiled with -DVALUE=0. However, once I launch it, ld resolves all symbols with one of libA.so versions, so both my executable and libB.so are using the same functions.
Is there any way to specify that I want to load resolve undefined symbols of libB.so only using its dependencies? I've tried using -Wl,-Bgroup flag when building libB.so, but it didn't change anything. I know there's dlmopen that can load the library in a new namespace, but I'd like to have it loaded automatically at startup.
I'm attaching a set of files that reproduce the behaviour:
libA.so:
#include <stdio.h>
#define _STR(x) #x
#define STR(x) _STR(x)
#ifndef VALUE
#define VALUE default
#endif
void func2() {
printf(STR(VALUE) "\n");
}
void func() {
func2();
}
libB.so:
#include <stdio.h>
extern void func(void);
void b_func() {
func();
}
executable:
#include <stdio.h>
extern void b_func(void);
extern void func(void);
int main() {
func(); // should print "default"
b_func(); // should print "other"
}
build commands:
gcc -fPIC -shared A.c -o libA.so
gcc -fPIC -shared -DVALUE=other A.c -o libA2.so
gcc -fPIC -shared B.c -L. -lA2 -o libB.so
gcc main.c -L. -lA -lB -o main
Curiously, it all works fine on OS X.

Related

linking with wrap directive for unit testing can be used to redefine a subroutine?

I'm using Cmocka to write a unit testing suite for a shared object written in C, but I'm having some issue. Since I cannot share the source code, I have written a minimum "not-working" example to show what is the issue:
my program is composed 5 files: foo.c, foo.h bar.c bar.h main.c.
bar.* files define a bar() function, which simply returns the argument multiplied by 2
foo.* files declare a foo() function that uses the bar() function defined by bar.h
main.c contains a simple cmocka test and a __wrap_bar()
function, returning the argument multiplied by 3.
I compile the program by producing a libfootest.so object (foo+bar) and then I link this object with main.o passing the -Wl,--wrap=bar flag to the compiler. In this configuration libfootest is the module under test and main is the tester program. I expect the __wrap__bar to be called (failing the test), but the standard bar() is called(test is passed). How can I solve this problem? Below you find all the code I'm using.
bar.c:
#include "bar.h"
int bar(int val) {
return val*2;
}
bar.h:
int bar(int val);
foo.h:
#include <stdio.h>
int foo(int val);
foo.c:
#include "foo.h"
#include "bar.h"
int foo(int val) {
int ret;
ret = bar(val);
printf("RET: %d", ret);
return ret;
}
main.c:
#include <stdio.h>
//required include for CMOCKA
#include <stdarg.h>
#include <stddef.h>
#include <stdint.h>
#include <setjmp.h>
#include <cmocka.h>
//library under test
#include "foo.h"
int __wrap_bar(int val) {
return 3*val;
}
static void test_foo(void **state) {
int ret = foo(5);
assert_int_equal(ret, 10);
}
int main (int argc, char** argv) {
const struct CMUnitTest tests[] = {
cmocka_unit_test(test_foo),
};
return cmocka_run_group_tests(tests, NULL, NULL);
}
Makefile:
CMOCKA_LIB_DIR=../../cmocka-1.1.5/build/src
CXXFLAGS+=-g -Og -fPIC
CFLAGS+=-g -Og -std=c99 -fPIC
CC=gcc
CXX=g++
all: main.o ./libfootest.so
gcc -o linux-test -g -L. -L$(CMOCKA_LIB_DIR) $(filter %.o, $^) -lcmocka -lfootest -Wl,-rpath=. -Wall -Wl,--wrap=bar -Wl,-rpath=$(CMOCKA_LIB_DIR)
./libfootest.so: foo.o bar.o
$(CC) -shared -o $# -g $^ -pedantic -Wall
clean:
rm -f *.o
rm -f *.so
The problem is your build of the library. You don't create a link library as commonly done, with separated modules. Instead you link all given modules and place the resulting single module in the target library.
That's why the linker resolved the call to bar() already, and it is no longer unresolved when linking the test program.
The option --wrap works only for unresolved references between modules.
The solution is to build the library from separated modules. Use the tool ar for this:
ar r libfootest.a foo.o bar.o

Why is my code not throwing a duplicate symbol conflict

Why is my following code not throwing duplicate symbol conflict?
I referred to name mangling, but that seems only when there is difference in parameters. But, here there is not difference in parameters. But, still it does not throw conflict. Why?
good.c
#include <stdio.h>
void printGood() {
printf("I am good");
}
perfect.c
#include <stdio.h>
void printGood() {
printf("I am perfect");
}
A.c
extern void printGood();
void bringGood() {
printGood();
}
B.c
extern void printGood();
void bringPerfect() {
printGood();
}
orchestrator.c
#include <stdio.h>
void bringGood();
void bringPerfect();
int main() {
printf("bringing good");
bringGood();
printf("bringing perfect");
bringPerfect();
return 1;
}
compile line:
gcc -g -c good.c
gcc -g -c perfect.c
gcc -g -c A.c
gcc -g -c B.c
gcc -g -c orchestrator.c
ar rcs libA.a perfect.o A.o
ar rcs libB.a good.o B.o
gcc -o orchestrator orchestrator.o -L. -lA -lB
Why is my following code not throwing duplicate symbol conflict?
The linker looks for undefined symbols in the libraries in the order in which they are specified in the linker line. When it finds a symbol in a library, it uses that definition and stops. It does not check whether that symbol is defined in any of the other libraries specified in the linker line.
In your case, if the linker finds a symbol in A.lib, it stops there. It does not look for the symbol in B.lib.
With your commands, the linker will find function printGood() in object perfect.o in library A. It will not use the function of the same name in good.o from library B. So you effectively link orchestrator.o, A.o, B.o and perfect.o. That's why the executable program prints I am perfect twice and not I am good.
Multiple definition errors are reported only when the object files used in the linker line contain multiple definitions.
You will see the error if you use:
gcc -o orchestrator orchestrator.o a.o b.o perfect.o good.o

Linking to an executable on OSX

On Windows it is possible to dynamically link to an executable with exported symbols. For example following code:
// main.c
void __declspec(dllexport) interface_function() {}
int main() {}
// ext.c
void interface_function();
void extension_function() {
interface_function();
}
With
cl.exe main.c
cl.exe ext.c /LD /link main.lib
would produce an executable main.exe, a static library main.lib for implicit linking, and a dynamic library ext.dll.
Similar behavior can be achieved in OSX with shared libraries:
// main.c
void interface_function() {}
int main() {}
// ext.c
void interface_function();
void extension_function() {
interface_function();
}
With
gcc main.c -o main
gcc ext.c -bundle -bundle_loader main -o ext.bundle
it is virtually equivalent to the Windows setup.
But for dynamiclib:
> gcc ext.c -dynamiclib -o ext.dylib
and shared:
> gcc ext.c -shared -o ext.so
I cannot get them to work because of undefined symbols on one hand and unable to load an executable with -l flag on the other.
I can let them resolve undefined symbols in runtime with -undefined dynamic_lookup. But this is not a sustainable way because all the link errors are now happening in run-time instead.
Is there a way to provide the list of symbols to dynamically load from an executable when linking as -shared and -dynamiclib?
Yes this is possible, but then you'll want to create a bundle rather than a shared library (see this answer for more detail).
If you have a main application like so:
#include <stdio.h>
#include <dlfcn.h>
int func(void)
{
return 42;
}
int main(void)
{
void *dl = dlopen("plugin.so", RTLD_LOCAL);
if(!dl) return -1;
int (*derp)(void) = dlsym(dl, "derp");
if(!derp) return -1;
printf("derp(): %i\n", derp());
return 0;
}
clang -o main main.c -Wall -Wl,-export_dynamic
Then you can compile bundles against it like so:
int func(void);
int derp(void)
{
return -func();
}
clang -o plugin.so plugin.c -Wall -bundle -bundle_loader ./main

Referencing global symbols from shared library loaded with dlopen

I have a shared library which I want to access symbols from the main program. For example:
main.c
#include <stdio.h>
void bar(void) { puts("bar"); }
extern void foo(void);
int main(void) {
foo();
return 0;
}
foo.c
#include <stdio.h>
extern void bar(void);
void foo(void) {
puts("foo");
bar();
}
I compile and run like:
gcc -c -fpic foo.c
gcc -shared -o libfoo.so foo.o
gcc -L$(pwd) -o test main.c -lfoo
./test
And I get the output I expect:
foo
bar
However, I must use dlopen() and dlsym() because I want to have control over when the library is loaded. The changed files are:
main.c
#include <stdio.h>
#include <dlfcn.h>
void bar(void) { puts("bar"); }
int main(void) {
void *handle = dlopen("./libfoo.so", RTLD_LAZY);
void (*foo)(void) = (void(*)(void))dlsym(handle,"foo");
foo();
return 0;
}
foo.c
#include <stdio.h>
#include <dlfcn.h>
extern void bar(void);
void foo(void) {
puts("foo");
bar();
}
I instead compile and run with:
gcc -c -fpic foo.c
gcc -shared -o libfoo.so foo.o
gcc -o test main.c -ldl
./test
However, this time I get the output
foo
./test: symbol lookup error: ./libfoo.so: undefined symbol: bar
How can I reference symbols in the main program from libfoo?
You have to add the -rdynamic option when linking test:
gcc -o test main.c -ldl -rdynamic
From here:
-rdynamic
Pass the flag -export-dynamic to the ELF linker, on targets that support it. This instructs the linker to add all symbols, not only used ones, to the dynamic symbol table. This option is needed for some uses of dlopen or to allow obtaining backtraces from within a program.

Why doesn't __attribute__((constructor)) work in a static library?

In the following example, the program should print "foo called\n":
// foo.c
#include <stdio.h>
__attribute__((constructor)) void foo()
{
printf("foo called\n");
}
// main.c
int main()
{
return 0;
}
If the program is compiled like this, it works:
gcc -o test main.c foo.c
However, if foo.c is compiled into a static library, the program prints nothing.
gcc -c main.c
gcc -c foo.c
as rcs foo.a foo.o
gcc -o test foo.a main.o
Why does this happen?
The linker does not include the code in foo.a in the final program because nothing in main.o references it. If main.c is rewritten as follows, the program will work:
//main.c
void foo();
int main()
{
void (*f)() = foo;
return 0;
}
Also, when compiling with a static library, the order of the arguments to gcc (or the linker) is significant: the library must come after the objects that reference it.
gcc -o test main.o foo.a
As it was stated, unreferenced symbols from archive does not make it to the output binary, because linker discards them by default.
To override this behaviour when linking with static library, --whole-archive/--no-whole-archive options for the linker may be used, like this:
gcc -c main.c
gcc -c foo.c
ar rcs foo.a foo.o
gcc -o test -Wl,--whole-archive foo.a -Wl,--no-whole-archive main.o
This may lead to bloated binary, because all symbols from foo.a will be included by the linker to the output, but sometimes it is justified.

Resources