Re: Oops caused by race between livepatch and ftrace

[Jiri Kosina]

On Tue, 21 May 2019, Steven Rostedt wrote:

> > Hm.  I suppose using ftrace_lock might be less risky since that lock 
> > is only used internally by ftrace (up until now).  But I think it 
> > would also make less sense because the text_mutex is supposed to 
> > protect code patching.  And presumably ftrace_lock is supposed to be 
> > ftrace-specific.
> > 
> > Here's the latest patch, still using text_mutex.  I added some lockdep
> > assertions to ensure the permissions toggling functions are always
> > called with text_mutex.  It's running through 0-day right now.  I can
> > try to run it through various tests with CONFIG_LOCKDEP.
> 
> Yeah, text_mutex probably does make more sense. ftrace_mutex was around
> before text_mutex as ftrace was the first one to do the runtime
> patching (after boot has finished). It wasn't until we introduced
> text_poke that we decided to create the text_mutex locking as well.
> 
> > 
> > 
> > From: Josh Poimboeuf <jpoimboe@xxxxxxxxxx>
> > Subject: [PATCH] livepatch: Fix ftrace module text permissions race
> 
> Thanks,
> 
> I'll try to find some time to test this as well.

Steve, Jessica, any final word on this?

Thanks,

-- 
Jiri Kosina
SUSE Labs