Check &self/&mut self

[jethrogb]

There's a lot of *mut pointer usage in the MbedTLS C API but perhaps some of those usages should be *const, especially with regards to callbacks.

[jack-fortanix]

I checked over every instance of &mut self and *mut pointers within mbedtls crate. Sadly I don't see anything to be done - the operations are either actually mutating the object, or they are not but mbedtls requires a non-const pointer for no particular reason. Eg pk_write_key_xxx functions are logically const (and I checked and they actually do not modify the key argument), but still take it by non-const pointer. Ditto ecdh_calc_secret, x509_crt_verify, ... ISTR seeing a bug in mbedtls issue tracker about this, basically they are worried that adding const to these declarations will break some application so are waiting for Mbedtls 3.0

We could start assuming these functions are actually const and then avoid having to take &mut self by instead casting self.inner as *const _ as *mut _, at the risk of nasty breakage if mbedtls instead changes to actually start mutating one of these arguments that they are accepting by mutable pointer. They seem to care about application compatibility so hopefully they wouldn't do that - but nothing in the docs for any of these functions states that the argument is not modified, so maybe as far as they are concerned all bets are off.

[jethrogb]

We could start assuming these functions are actually const and then avoid having to take &mut self by instead casting self.inner as *const _ as *mut _,

Yes that's what I was thinking

[jack-fortanix]

Sounds good to me. Some of these will require some digging to figure out if it is actually safe or not (for instance it appears ecdh_calc_secret may actually modify the context). In Mbed-TLS/mbedtls#1485 (comment) I suggested to upstream that they document the situation better.