bofObjectRunAsyncThread() crashes with older GLIBC versions

[fermat-prime]

Calling bofObjectRunAsyncThread() on a system with an older GLIBC versions (prior to 2.34) will cause a segmentation fault due to a NULL pointer dereference. This is because pthread functionality wasn't integrated into libc.so.6 in those GLIBC versions so the pthread_create and pthread_detach function pointers are NULL.

This should be reproducible by modifying the integration-with-c example to use bofObjectRunAsyncThread (demo.patch) and executing it on a system with an older GLIBC. I also have an example fix that patches bof_launcher.zig to load the functions from libpthread.so.0 (fix.patch).

demo.zip

test@test:~$ unzip demo.zip
Archive:  demo.zip
  inflating: demo.sh                 
  inflating: demo.patch              
  inflating: fix.patch
  
# run an example that produces a crash 
test@test:~$ docker run --rm -it --entrypoint bash -v`pwd`:/data oraclelinux:8.9 /data/demo.sh

# run the example with the example fix applied
test@test:~$ docker run --rm -it --entrypoint bash -v`pwd`:/data oraclelinux:8.9 /data/demo.sh 1

I have tested this patch on GLIBC 2.17, 2.28, 2.34, and 2.39 and the patch worked in all cases. However, I have only been able to test x86_64 and not architectures such as arm or aarch64.

I don't know if introducing a reliance on libpthread.so.0 (or the stub provided with modern GLIBC versions) is how you wanted to address this until issue #1 is solved so I didn't want to preemptively make a pull request. Let me know if you would like me to make one though.

Thanks for making and maintaining this project, I've found it to be quite interesting and useful!

[michal-z]

Thanks for the report.

I think that your patch is a good temporary solution until #1 is solved. Please make a PR and @mzet- will test it on ARM.