Re: Re: [PATCHv3 2/3] kernel: add support for live patching

[Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>]

(2014/11/26 4:29), Jiri Kosina wrote:
> On Tue, 25 Nov 2014, Steven Rostedt wrote:
> 
>> It is not guaranteed from ftrace's stand point. What happens if we have
>> a kprobe handler that modifies it for someplace else? Changing the ip
>> address may not be a kpatch/kGraft privilege only.
> 
> This brings me back to the RFC patch I sent back then in october ... what 
> we really want to do is to at least warn about situations when we are 
> going to redirect code flow (through IPMODIFY) for function that has a 
> kprobe installed anywhere inside it.

Actually in my plan, normal kprobes/kretprobes don't set IPMODIFY
flag because it don't change the IP. Instead, you can even use
debugfs/kprobes/list to check whether the function is probed or not.
Or, I think we can provide atomic-conflict checking interface which
will iterate probes under locking kprobe list.

> Otherwise the probe will silently 
> vanish (there is no way how to migrate it to the new function 
> automatically), which might be very confusing for uses (cosider systemtap, 
> for example).

Yeah, I think we can add --force option(or sysctl) to patch functions
just ignoring probed or not. (for emergency vulnerability fixes)

Thank you,

> 
> I'll resurect my patch if noone beats me doing it. It should go in 
> together with the live patching framework I believe.
> 
> Thanks,
> 


-- 
Masami HIRAMATSU
Software Platform Research Dept. Linux Technology Research Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: masami.hiramatsu.pt@hitachi.com