race conditions in libcurl under valgrind

[hebrerillo]

Hello everyone!

I am executing the following libcurl example under valgrind(helgrind)

https://curl.haxx.se/libcurl/c/multithread.html

And I am getting race conditions. Because the valgrind output is huge, here I just put an example of a race condition:

==7693== Possible data race during read of size 4 at 0x54EC5E0 by thread #8
==7693== Locks held: none
==7693==    at 0x51C6128: ??? (in /lib/x86_64-linux-gnu/libcrypto.so.1.0.0)
==7693==    by 0x51C638C: ??? (in /lib/x86_64-linux-gnu/libcrypto.so.1.0.0)
==7693==    by 0x51C6F6F: ERR_remove_thread_state (in /lib/x86_64-linux-gnu/libcrypto.so.1.0.0)
==7693==    by 0x4297B4: Curl_close (in /home/romerohe/proyects/cpp/libcurl/main)
==7693==    by 0x411A40: Curl_conncache_close_all_connections (in /home/romerohe/proyects/cpp/libcurl/main)
==7693==    by 0x40BCEE: curl_multi_cleanup (in /home/romerohe/proyects/cpp/libcurl/main)
==7693==    by 0x42974B: Curl_close (in /home/romerohe/proyects/cpp/libcurl/main)
==7693==    by 0x4079C4: curl_easy_cleanup (in /home/romerohe/proyects/cpp/libcurl/main)
==7693==    by 0x40732C: pull_one_url(void*) (main.cpp:30)
==7693==    by 0x4C34E85: mythread_wrapper (hg_intercepts.c:389)
==7693==    by 0x54F56B9: start_thread (pthread_create.c:333)
==7693== 
==7693== This conflicts with a previous write of size 4 by thread #6
==7693== Locks held: none
==7693==    at 0x51C6128: ??? (in /lib/x86_64-linux-gnu/libcrypto.so.1.0.0)
==7693==    by 0x51C661C: ??? (in /lib/x86_64-linux-gnu/libcrypto.so.1.0.0)
==7693==    by 0x51C7023: ERR_get_state (in /lib/x86_64-linux-gnu/libcrypto.so.1.0.0)
==7693==    by 0x51C715D: ERR_put_error (in /lib/x86_64-linux-gnu/libcrypto.so.1.0.0)
==7693==    by 0x51F8626: PEM_read_bio (in /lib/x86_64-linux-gnu/libcrypto.so.1.0.0)
==7693==    by 0x51F69AD: PEM_X509_INFO_read_bio (in /lib/x86_64-linux-gnu/libcrypto.so.1.0.0)
==7693==    by 0x5206396: X509_load_cert_crl_file (in /lib/x86_64-linux-gnu/libcrypto.so.1.0.0)
==7693==    by 0x52064B9: ??? (in /lib/x86_64-linux-gnu/libcrypto.so.1.0.0)
==7693==  Address 0x54ec5e0 is in the BSS segment of /lib/x86_64-linux-gnu/libcrypto.so.1.0.0

But there are many more. Does anyone know why? Or maybe the problem is with valgrind?

Thanks a lot!

[jonnin]

these kinds of tools are known for false alarms, which can be mitigated by settings somewhat but if you get 0 false alarms you probably have erred the wrong way and missed issues, so best to set it so you get a few but not many. That said, there could well be issues in the library or in how you used it. These are well known libraries; do you have an example program that they provide, and does a check on their example give a ton of problems?

Race conditions can exist and never be triggered, as well. The algorithm to find them can't tell if the code ever gets into a bad state, just that it could possibly. Are you able to write an example that goes into thread snarl and locks up via any of the listed race condition problems?