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

[Google]

On Thu, 17 Aug 2023 11:13:03 -0400
Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:

> On Thu, 17 Aug 2023 12:59:30 +0200
> Francis Laniel <flaniel@xxxxxxxxxxxxxxxxxxx> wrote:
> 
> > > Why can't you just pass in the address like sysfs does?  
> > 
> > To get the addresses from /proc/kallsyms, you need to either have CAP_SYSLOG 
> > or even CAP_SYS_ADMIN.
> > But to call perf_event_open(), you only need CAP_PERFMON.
> > This way, by giving only the name you can trace function with less privileges 
> > (i.e. without CAP_SYS_ADMIN).
> > Please correct me if I am wrong as I am not an expert in knowing the minimal 
> > set of capabilities you need to trace.
> 
> I wonder if we should add an option to put in the non-relocated address?
> One that can be acquired by debuginfo in the vmlinux. I'm assuming that the
> kernel has access to the added offset (I haven't looked). If it does, then
> we could allow users to just add something like "+@0xffffffffdeadbeef" and
> add the relocation offset to get to the mapped address of the function.

That's why perf probe uses the offset from '_text'. Normal KASLR will just
moves all symbols. (Finer one will move all symbols randomely)
This should not need to access /proc/kallsyms but vmlinux or SystemMap.

Thank you,

> 
> This would allow those without kallsym privileges to pass in kernel address
> for tracing.
> 
> -- Steve


-- 
Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>