Segmentation fault in EllipticCurveKey about zerotierone HOT 9 CLOSED

This looks like a memory management issue related to object life cycles. I'm going to try to duplicate and do a bit of analysis in valgrind.

Hmm... good point about debug vs. release builds. Maybe I should keep all builds debug for now.

from zerotierone.

I was unable to exactly duplicate this one, but I did fix a race condition bug in 0.4.1. See if this crops up again. I'm going to continue testing and trying to duplicate this one myself.

from zerotierone.

By the way, can you describe the platform? Linux version, distro, etc.? That might be helpful.

from zerotierone.

Same platform/distro as #13. Still getting the same backtrace.

I ran valgrind:

==18354== ERROR SUMMARY: 55772 errors from 182 contexts (suppressed: 2 from 2)

Do you have a valgrind suppression file? If so, can you commit it? There is a lot of stuff coming from ecdsa_do_sign for example. If you can generate a suppression file for all of the "known" errors you get, I can send you the list of whatever remains.

from zerotierone.

No, but I should make one. I ran valgrind and saw a ton of errors in the libcrypto bignum library... which kind of worries me slightly. libcrypto is very mature, so I'm assuming it's okay. :)

from zerotierone.

OpenSSL warnings are normal: http://openssl.6102.n7.nabble.com/valgrind-warnings-for-0-9-8b-td6228.html

Apparently they use some uninitialized memory as one of their sources of randomness. I'll have to make a valgrind exclude file for pretty much all the libcrypto stuff.

from zerotierone.

The OpenSSL thing is very irritating... looks like every uninitialized value warning can be traced to the fact that libcrypto pulls some of its random entropy from uninitialized memory. If I had this to write over again, I'd use libcrypto++. Might convert it some day, but it's not high priority.

I did simplify some code in EllipticCurveKey and added code to throw std::out_of_range on range errors. But I can't duplicate your exact crash either on Mac or on my Linux physical box or my Linux VM. I committed it. No version.h change since it's such a tiny thing.

from zerotierone.

A huge valgrind cleanup effort has been committed: version 0.4.2. Try it now and post the error if it's still there. (I also fixed a possible unsafe op, so it may be gone now.)

from zerotierone.

Segfault is gone. #13 Still persists though.

from zerotierone.