Box64 Box86 Dev General

How to create a wrapping

In a previous post, we talked about the mechanism behind library wrapping. In it, I said box86/box64 used some manually written files to correctly call functions. However, there is a big question: how are these files written?

To follow this article, you will need at least a basic understanding of function signatures.

  1. The basics
  2. Registering a new library
  3. Adding simple functions
  4. Adding complex functions
  5. Addendum: a small type to letter table

The basics

First, a reminder: what is library wrapping (in box86/box64)?

Library wrapping can refer to multiple things, but here it refers to how box86/box64 interacts with native libraries.

These native libraries have a list of functions (exported symbols) that the end user (the program) can call. However, to call these functions, box86/box64 needs to know what the arguments are so it can put the correct arguments at the correct place (i.e., is that argument coming from the stack, a register…?). However, this is all done automagically behind the scene, and the wrapper writer doesn’t need to know these details.

The first step: registering a new wrapper

So, how do we write a wrapper? First, we need to tell box that the wrapper exists. This must be done once at multiple places. Therefore, this step can be skipped if you just want to add some functions.

First, in the src/library_list.h file, there is a list of all wrapped libraries. To add one, you need to add a line containing:

GO("", my_library)

where is the library file name in the file system, and my_library is how you name the auxiliary file (more precisely, it is how you name it in this auxiliary file).

Next, you need to tell the compiler toolchain about the new auxiliary file. In the CMakeLists.txt file, there is a list of filenames put in the WRAPPED variable (the block starts with set(WRAPPED, continues with the list and ends with a )). You need to add a new line in this variable containing your auxiliary file.

Congratulations, you have registered a new library wrapping! Yes, it really boils down to two lines.

The last step is to create the missing files: just paste the following templates:

#define _GNU_SOURCE         /* See feature_test_macros(7) */
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <dlfcn.h>

#include "wrappedlibs.h"

#include "wrapper.h"
#include "bridge.h"
#include "librarian/library_private.h"
#include "x64emu.h"

#include "generated/wrappedmy_librarydefs.h"

const char* my_libraryName = "";
#define LIBNAME my_library

#include "generated/wrappedmy_librarytypes.h"

#include "wrappercallback.h"

// Insert code here

#include "wrappedlib_init.h"

in the wrappedmy_library.c file, and

#if !(defined(GO) && defined(GOM) && defined(GO2) && defined(DATA))
#error Meh....

in the wrappedmy_library_private.h file. Correct the macros in the C file, and you’re done!

You should now be able to compile box86/box64, and running it with a binary that uses the library should fail with symbol not found errors.

Sometimes, some more library initialization code is required. For arbitrary code, you can put it in the CUSTOM_INIT macro. However, there are other helper macros: for example, if the library depends on loading other libraries, you can define the NEEDED_LIBS macro to the comma-separated list of other libraries to include.

Similarly, the CUSTOM_FINI macro is used when finishing the library.

These macros must be defined in the C file before including the wrappedlib_init.h file. In the generic CUSTOM_INIT macro, the library_t structure which references the current library is called lib.

The final step (easy): adding simple functions

Now, open the header file (wrappedmy_library_private.h). This file should have three non-empty lines (#if ..., #error Meh...., #endif).

You are now ready to add a new function.

First, look at its signature. I will use the example of memcpy:

void* memcpy( void *dest, const void *src, size_t count );

Notice that there are no callbacks or a variable number of arguments in this signature (no ...). Notice that the function also requires no realignment of the arguments. This is important, and we will come back to functions with those issues later.

First, you can prepare a line in the header file:

GO(memcpy, )

Notice that since this is a simple function, I used the GO macro. If the library declares this symbol as “weak” (which it usually doesn’t), you need to use the GOW macro.

The second argument should contain the function signature. Here is how:

  1. The first letter corresponds to the return type
  2. The second letter corresponds to the calling convention
  3. The remaining letters correspond to the types of the arguments in the correct order, or v if there is none

The calling convention is almost always F (for the System V calling convention).

If the function returns a void, the first letter should be v. If the function returns an int, the first letter should be i, and so on. You can look at the top of src/wrapped/generated/wrapper.h to have a list of all the common types. See also the end for a table of more types.

For example, the return type of memcpy is a pointer, as are the first two arguments, which corresponds to the letter p; the last argument, size_t, is a typedef of unsigned long which is the letter L.

Therefore, we can complete the line:

GO(memcpy, pFppL)

which is exactly what you will find if you search through wrappedlibc_private.h.

That’s it! Doing so with a few functions then compiling, you may be greeted with errors from the wrapper generator script. This means you have incorrectly written something. Read the error, correct it, and try again. (Note that the script is called automatically when compiling. Calling it manually is both non-trivial and unnecessary.)

Hopefully, all functions of the library you want to wrap are such functions. Otherwise, you will need to…

The final step (hard): adding redirected functions

For functions with a callback, or with structures with differing sizes, you will need to write C code. If you don’t know how, please just open a pull request; this is not a beginner-friendly project.

Let’s take an example of a harder function: printf. Its signature is the following:

int printf( const char *format, ... );

First, notice the use of the ...: this function takes a variable number of arguments, the exact number depending on the format string.

As with the easy way, we can add a new line in the wrappedmy_library_private.h file:

GOM(printf, iFEpV)

Notice the use of the GOM macro: this means that instead of calling the native function directly, box86/box64 will instead call its own wrapper function my_printf. For weak functions, there is an equivalent GOWM.

Side note: the my_ prefix is configurable. To change it, define the ALTMY macro to whatever you want in the C file before including the wrappedlib_init.h file. You can look at SDL2 for an example.

Next, notice the E as its first argument: my_printf will receive the active x86emu_t*/x64emu_t* as its first argument. This is a special argument letter, reserved only for functions with redirection and as the first argument only. The generator will enforce this and output an error otherwise.

The letter V as the final argument is the variadic argument.

Next, we need to define the my_printf function somewhere.

It is done in the wrappedmy_library.c file (usually before includeing wrappedlib_init.h).

You need to EXPORT the function my_printf with the correct arguments:

EXPORT int my_printf(x64emu_t *emu, void* fmt, void* b) {

Next, you need to do whatever you need to do to align the stack:

    myStackAlign(emu, (const char*)fmt, b, emu->scratch, R_EAX, 1);

And finally, call the function:

    return vprintf((const char*)fmt, VARARGS);

And you’re done!

Of course, the “whatever you need to do” may take a lot more than two lines. If you want some good-ish examples, just take a look at how SDL2 is done.

Can it be automated?

Maybe. There are a few efforts toward making an LLVM frontend to parse headers then generate the _private.h file automatically, but AFAIK none of them are ready yet, all of them having issues with integer sizes, callbacks, correctness of the output…

Furthermore, (one of) the issue(s) with all the currently available generators I know of is that once they successfully output a file, you will still need to come back to check whether they are correct, and/or you will still need to write code for the more complex functions.

Addendum: a table of types

C typeLetter
x64emu_tE (*)
unsigned charC
unsigned shortW
long longI
unsigned long longU
unsigned longL
long doubleD or K (see libm)
void*p (rarely, P (*))
Be careful of function pointers
...V (*) or N or M
float complexx
double complexX
long double complexY or y (see libm)
Small table of frequently-seen types and their corresponding wrapped letter

(*) Usually, having this letter means you need to add a wrapper function (my_...).

Leave a Reply

Your email address will not be published. Required fields are marked *