Re: [PATCH 05/15] x86: Implement function_nocfi

[Joao Moreira]

> Why not? In particular, I'd really like somebody to answer the question
> "why not just store a cookie before each address-taken or
> external-linkage function?".
FWIIW, this was done before (at least twice): First with grsecurity/PaX 
RAP (https://grsecurity.net/rap_faq) then with kCFI 
(https://www.blackhat.com/docs/asia-17/materials/asia-17-Moreira-Drop-The-Rop-Fine-Grained-Control-Flow-Integrity-For-The-Linux-Kernel-wp.pdf, 
https://github.com/kcfi/kcfi - which is no longer maintained).

At the time I worked on kCFI someone raised a concern regarding this 
cookie-based design being mutually exclusive to execute-only memories 
(XOM), what, if XOM is really relevant to someone, should be a valid 
concern.

Since design is being questioned, an x86/CET-specific third design for 
CFI was recently discussed here: 
https://www.openwall.com/lists/kernel-hardening/2021/02/11/1 -- I assume 
that, arch-dependency considered, this should be easier to integrate 
when compared to clang-cfi. Also, given that it is based on CET, this 
also has the benefit of constraining mispeculations (which is a nice 
side-effect).

Tks, Joao