Re: [RFC PATCH 2/2] livepatch: Clear relocation targets on a module removal

[Josh Poimboeuf <jpoimboe@redhat.com>]

On Thu, Sep 05, 2019 at 03:52:59PM +0200, Petr Mladek wrote:
> On Thu 2019-09-05 08:15:02, Josh Poimboeuf wrote:
> > On Thu, Sep 05, 2019 at 08:08:32AM -0500, Josh Poimboeuf wrote:
> > > On Thu, Sep 05, 2019 at 01:09:55PM +0200, Petr Mladek wrote:
> > > > > I don't have a number, but it's very common to patch a function which
> > > > > uses jump labels or alternatives.
> > > > 
> > > > Really? My impression is that both alternatives and jump_labels
> > > > are used in hot paths. I would expect them mostly in core code
> > > > that is always loaded.
> > > > 
> > > > Alternatives are often used in assembly that we are not able
> > > > to livepatch anyway.
> > > > 
> > > > Or are they spread widely via some macros or inlined functions?
> > > 
> > > Jump labels are used everywhere.  Looking at vmlinux.o in my kernel:
> > > 
> > >   Relocation section [19621] '.rela__jump_table' for section [19620] '__jump_table' at offset 0x197873c8 contains 11913 entries:
> > > 
> > > Each jump label entry has 3 entries, so 11913/3 = 3971 jump labels.
> > > 
> > > $ readelf -s vmlinux.o |grep FUNC |wc -l
> > > 46902
> > > 
> > > 3971/46902 = ~8.5%
> > > 
> > > ~8.5% of functions use jump labels.
> > 
> > Obviously some functions may use more than one jump label so this isn't
> > exactly bulletproof math.  But it gives a rough idea of how widespread
> > they are.
> 
> It looks scary. I just wonder why we have never met this problem during
> last few years.

Who knows what can happen when you disable jump label patching.
Sometimes it may be harmless.  A panic is probably the worst case.
There may be other fail modes which are harder to detect.

> My only guess is that most of these functions are either in core
> kernel or in code that we do not livepatch.

This is definitely not the case.  We recently introduced jump label
checking in kpatch-build, and it complains a lot.  The workaround is to
replace such uses with static_key_enabled().

-- 
Josh