lkml.org 
[lkml]   [2015]   [Feb]   [10]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [RFC PATCH 0/9] livepatch: consistency model
On Tue, Feb 10, 2015 at 09:59:58AM -0600, Josh Poimboeuf wrote:
> On Tue, Feb 10, 2015 at 08:16:59PM +0900, Masami Hiramatsu wrote:
> > (2015/02/10 2:31), Josh Poimboeuf wrote:
> > > This patch set implements a livepatch consistency model, targeted for 3.21.
> > > Now that we have a solid livepatch code base, this is the biggest remaining
> > > missing piece.
> > >
> > > This code stems from the design proposal made by Vojtech [1] in November. It
> > > makes live patching safer in general. Specifically, it allows you to apply
> > > patches which change function prototypes. It also lays the groundwork for
> > > future code changes which will enable data and data semantic changes.
> >
> > Interesting, How would you do that?
>
> As Vojtech described in the earlier thread from November, there are
> different approaches for changing data:
>
> 1. TRANSFORM_WORLD: stop the world, transform everything, resume
>
> 2. TRANSFORM_ON_ACCESS: transform data structures when you access them
>
> I would add a third category (which is what we've been doing with
> kpatch):
>
> 3. TRANSFORM_ON_CREATE: create new data structures created after a certain point
> are the "v2" versions

Sorry, bad wording, I meant to say:

3. TRANSFORM_ON_CREATE: create new versions of the data structures when
you create them

If that still doesn't make sense, hopefully the below explanation
clarifies what I mean :-)

>
> I think approach 1 seems very tricky, if not impossible in many cases,
> even if you're using stop_machine(). Right now we're focusing on
> enabling approaches 2 and 3, since they seem more practical, don't
> require stop_machine(), and are generally easier to get right.
>
> With kpatch we've been using approach 3, with a lot of success. Here's
> how I would do it with livepatch:
>
> As a prerequisite, we need shadow variables, which is a way to add
> virtual fields to existing structs at runtime. For an example, see:
>
> https://github.com/dynup/kpatch/blob/master/test/integration/shadow-newpid.patch
>
> In that example, I added "newpid" to task_struct. If it's only
> something like locking semantics that are changing, you can just add a
> "v2" field to the struct to specify that it's the 2nd version of the
> struct.
>
> When converting a patch to be used for livepatch, the patch author must
> carefully look for data struct versioning changes. It doesn't matter if
> there's a new field, or if the semantics of using that data has changed.
> Either way, the patch author must define a new version of the struct.
>
> If a struct has changed, all patched functions need to be able to deal
> with struct v1 or struct v2. This is true for those functions which
> access the structs as well as the functions which create them.
>
> For example, a function which accesses the struct might change to:
>
> if (klp_shadow_has_field(struct, "v2"))
> /* access struct the new way */
> else
> /* access struct the old way */
>
> A function which creates the struct might change to:
>
> struct foo *struct_create()
> {
> /* kmalloc and init struct here */
>
> if (klp_patching_complete())
> /* add v2 shadow fields */
> }
>
>
> The klp_patching_complete() call is needed to prevent v1 functions from
> accessing v2 data. The creation/transformation of v2 structs shouldn't
> occur until after the patching process is complete, and all tasks are
> converged to the new universe.
>
> > > disadvantages vs kpatch:
> > > - no system-wide switch point (not really a functional limitation, just forces
> > > the patch author to be more careful. but that's probably a good thing anyway)
> >
> > OK, we must check carefully that the old function and new function can be co-exist.
>
> Agreed, and this requires the patch author to look carefully for data
> version changes, as described above. Which they should be doing
> regardless.
>
> > > My biggest concerns and questions related to this patch set are:
> > >
> > > 1) To safely examine the task stacks, the transition code locks each task's rq
> > > struct, which requires using the scheduler's internal rq locking functions.
> > > It seems to work well, but I'm not sure if there's a cleaner way to safely
> > > do stack checking without stop_machine().
> >
> > We'd better ask scheduler people.
>
> Agreed, I will.
>
> > > 2) As mentioned above, kthreads which are always sleeping on a patched function
> > > will never transition to the new universe. This is really a minor issue
> > > (less than 1% of patches). It's not necessarily something that needs to be
> > > resolved with this patch set, but it would be good to have some discussion
> > > about it regardless.
> > >
> > > To overcome this issue, I have 1/2 an idea: we could add some stack checking
> > > code to the ftrace handler itself to transition the kthread to the new
> > > universe after it re-enters the function it was originally sleeping on, if
> > > the stack doesn't already have have any other to-be-patched functions.
> > > Combined with the klp_transition_work_fn()'s periodic stack checking of
> > > sleeping tasks, that would handle most of the cases (except when trying to
> > > patch the high-level thread_fn itself).
> >
> > It makes sense to me. (I just did similar thing)
> >
> > >
> > > But then how do you make the kthread wake up? As far as I can tell,
> > > wake_up_process() doesn't seem to work on a kthread (unless I messed up my
> > > testing somehow). What does kGraft do in this case?
> >
> > Hmm, at a glance, the code itself can work on kthread too...
> > Maybe you can also send you testing patch too.
>
> Yeah, I probably messed it up. I'll try it again :-)
>
> --
> Josh

--
Josh


\
 
 \ /
  Last update: 2015-02-10 18:41    [W:0.070 / U:5.452 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site