The correctness of dlopen-ing a PIE executable

The correctness of dlopen-ing a PIE executable

[Fengkai Sun]

Hi list,

I'm very interested in the idea of dynamically loading a PIE
executable using dlopen, and I found a patch at:
https://sourceware.org/bugzilla/show_bug.cgi?id=11754

However, as the comment(
https://sourceware.org/bugzilla/show_bug.cgi?id=11754#c13) below suggests,
this is a bad idea. So glibc explicitly forbids it in dl-load.c.

But in this question:
https://stackoverflow.com/questions/64659713/why-does-dynamically-loading-of-pies-no-longer-work-in-glibc,
the author shows that:

* recompile the executable using -fPIC, which eliminates relocation type
R_X86_64_COPY, will fix the bug in
https://sourceware.org/bugzilla/show_bug.cgi?id=11754#c15
* constructors, relocations, and thread local variables seem to work OK

I know that the reasons the author above listed are probably not
exhaustive, so I want to make it clear that is there any irrecoverable
problem that prevents dlopen to load PIE executable correctly?(suppose we
force the executable enable -fPIC )
If so, how will a modified loader manage to solve that, as mentioned in
https://sourceware.org/bugzilla/show_bug.cgi?id=11754#c16  ?

Thank you very much.


Best,
Fengkai