Re: sagemath denial of service with abort() in gmp: overflow in mpz type

[Jeffrey Walton <noloader () gmail com>]

On Tue, Sep 6, 2022 at 7:52 AM Jeremy Stanley <fungi () yuggoth org> wrote:
> On 2022-09-06 08:47:58 +0300 (+0300), Georgi Guninski wrote:
> [...]
> > sagemath gives access to the python interpreter, so code execution
> > is trivial.
> [...]
>
> I'm not familiar with sagemath, but is it intended to protect
> against such cases? Note that even if all it does is pass
> expressions into CPython's eval(), it's pretty much impossible to
> guard against misuse without completely sandboxing the underlying
> processes. Denial of service scenarios are really the least of
> worries in that case. Many articles have been written over the years
> about this, though one of the more recent and thorough ones is:
> https://netsec.expert/posts/breaking-python3-eval-protections/

One of the problems with GMP is, it will crash instead of returning
failure. The problem becomes more acute if the program using GMP is
handling sensitive information, like a private key or passphrase. The
sensitive material can be written to a dump file and can be sent to an
error reporting service. So there's a DoS in the app, and a possible
egress of sensitive information outside the app's security boundary.

It is not clear to me whether that is happening in this instance, though.

Stepping back a bit, the GMP library is setting a policy for an
application. That is, GMP is setting the policy of "crash instead of
fail." An application should set their policy, not libraries. Apps are
in the position to determine  strategy they need, not libraries.
Libraries know nothing about an application's security requirements or
strategy.

Jeff