Re: crash on macOS with dlsym RTLD_LOCAL

["Aleix Conchillo Flaqué" <aconchillo@gmail.com>]

I did a little bit more research. The combination that actually fails
is RTLD_LAZY | RTLD_GLOBAL (what guile uses). So, you can actually
use:

- RTLD_LAZY | RTLD_GLOBAL
- RTLD_NOW | RTLD_GLOBAL
- RTLD_NOW | RTLD_LOCAL

I found the source code of dyld

   https://opensource.apple.com/source/dyld/dyld-852.2/

But I have no clue how to build this, I tried it but failed with
missing dependencies. In any case, I provided a patch so at least we
can use Guile on macOS.

Best,

Aleix

On Sat, Sep 4, 2021 at 10:48 PM Aleix Conchillo Flaqué
<aconchillo@gmail.com> wrote:
>
> Hi,
>
> I'm trying to load guile-grcypt with guile 3.0.7 and after my previous
> fix (https://git.savannah.gnu.org/cgit/guile.git/commit/?id=1f100a4f20c3a6e57922fb26fce212997e2a03cb)
> guile-gcrypt still does not load properly.
>
> TLDR; Using RTLD_LOCAL instead of RTLD_GLOBAL (default) causes issues on macOS.
>
> https://git.savannah.gnu.org/cgit/guile.git/tree/module/system/foreign-library.scm#n181
>
> In the case of guile-gcrypt this is what I'm getting:
>
> -------------------------
> scheme@(guile-user)> (use-modules (gcrypt hmac))
> dyld: lazy symbol binding failed: Symbol not found: __gcry_check_version
>   Referenced from: /usr/local/lib/libgcrypt.dylib
>   Expected in: flat namespace
>
> dyld: Symbol not found: __gcry_check_version
>   Referenced from: /usr/local/lib/libgcrypt.dylib
>   Expected in: flat namespace
> -------------------------
>
> Looking at gcrypt symbols I can see:
>
> ❯ nm -gU /usr/local/lib/libgcrypt.dylib | grep gcry_check_version
> 0000000000005954 T __gcry_check_version
> 0000000000002aa7 T _gcry_check_version
>
> Actually in libgcrypt the public functions are: gcry_check_version and
> _gcry_check_version (no extra underscore). I think extra underscores
> are automatically added when building the library but I'm not sure why
> and when and what systems. And when you call dlsym() you don't need to
> add those extra underscores.
>
> The following code (which uses RTLD_GLOBAL) works fine:
>
> ./a.out gcry_check_version
>
> -------------------------
> #include <dlfcn.h>
> #include <stdio.h>
>
> int
> main(int argc, char *argv[])
> {
>   char* (*fptr)(char *);
>   void *handle = dlopen("/usr/local/lib/libgcrypt.dylib", RTLD_LAZY |
> RTLD_GLOBAL);
>   if (handle == NULL) {
>     printf("OOOPS: %s\n", dlerror());
>   } else {
>     *(void **)(&fptr) = dlsym(handle, argv[1]);
>     if (fptr == NULL) {
>        printf("NOT FOUND: %s : %s\n", argv[1], dlerror());
>     } else {
>        printf("FOUND: %s %s\n", argv[1], (*fptr)(NULL));
>     }
>   }
>   return 0;
> }
> -------------------------
>
> But if we change to RTLD_LOCAL we get a crash like in guile's case.
>
> Sorry for being too vague but I'm not familiar with how all this
> works. All I know is that using RTLD_GLOBAL fixes the issue.
>
> Also, from dlopen man page I read:
>
> -------------------------
>     RTLD_GLOBAL  Symbols exported from this image (dynamic library or
> bundle) will be available to any images build with -flat_namespace
> option to
>                   ld(1) or to calls to dlsym() when using a special handle.
>
>      RTLD_LOCAL   Symbols exported from this image (dynamic library or
> bundle) are generally hidden and only availble to dlsym() when
> directly using
>                   the handle returned by this call to dlopen().
> -------------------------
>
> But I don't fully get what this means.
>
> Any help would be really appreciated.
>
> Thank you,
>
> Aleix