Re: [RFC PATCH v1 1/1] tracing/kprobe: Add multi-probe support for 'perf_kprobe' PMU

[Kees Cook]

On Mon, Aug 21, 2023 at 10:45:50AM -0400, Steven Rostedt wrote:
> On Mon, 21 Aug 2023 19:01:52 +0900
> Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx> wrote:
> 
> > > kprobe BPF program has access to pt_regs, so it can read ip of the
> > > attached function. Can we do the same with regular kprobe (no bpf)?  
> > 
> > Yes, it can. So I think it is OK to expand CAP_PERFMON to access kallsyms.
> > But this means CAP_PERMON itself is not safe in some case.
> 
> What are the privileges that CAP_PERFMON gives. I can see why Kees told me
> to avoid capabilities when looking at what has access to tracefs. Because
> it becomes very difficult to know what the privileges you are giving when
> you give out a capability. I just stick to normal ACL (file permissions)
> and everything is much easier and simpler to know what has access to what.

At the very least, having a fd-based "handle" for access work. But yeah,
capabilities get ugly quickly.

Anyway... what does CAP_PERFMON have access to right now? If it is
allowed to read arbitrary kernel memory, then resolving symbols is fine.
If it doesn't, then no, it shouldn't: it becomes a oracle for probing
symbol locations.

-- 
Kees Cook