`core_obj_create` and multi-threaded applications

[Thomas Bailleux]

Hello OpenSSL,

I'm currently writing a multi-threaded application using OpenSSL 3.1.
For some reason, I've decided to instantiate a `OSSL_LIB_CTX` per thread.
I'm also trying the provider API with oqs-provider, and I ran into an issue
that is caused
by the fact that `core_obj_create` is called by `OSSL_PROVIDER_load` which
itself is called by multiple threads running at the same time.

I described the bug in the following thread:
https://github.com/open-quantum-safe/oqs-provider/issues/272

Basically, the oqs-provider init function registers some custom OIDs using
the `core_obj_create` which makes a call to `OBJ_create`.

As described in the BUGS section of the `OBJ_create` man page[1] of *OpenSSL
3.0*, `OBJ_create` used to be non thread-safe.
However, in *OpenSSL 3.1*[2], it is thread safe:

These functions were not thread safe in OpenSSL 3.0 and before.
>

So I think there is a race condition in `core_obj_create`[3] function: the
function checks if the OID already exists, if so it returns OK (1),
otherwise it makes a call to `OBJ_create`. The TOCTOU is happening between
these two statements basically.

What do you suggest to mitigate this issue?

Regards,

- thomas

[1]: https://www.openssl.org/docs/man3.0/man3/OBJ_create.html
[2]: https://www.openssl.org/docs/man3.1/man3/OBJ_create.html
[3]:
https://github.com/openssl/openssl/blob/9c20f5db0feaddc4c9ea4c4b2b07e6d87d6701f1/crypto/provider_core.c#L2127-L2133

-- 

Thomas Bailleux

Cryptography software engineer

thomas at sandboxaq.com

+33 6 47 04 61 58 | Paris, FRANCE
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mta.openssl.org/pipermail/openssl-users/attachments/20231005/65bbbdfa/attachment.htm>