[CRIU] [PATCH] Punch holes in input files when restoring anonymous non-shared memory if --auto-dedup is enabled.

[Paweł Stradomski]

Ah, this one was merged already. Nevermind then. I'll still send a fix
by falling back to readonly first and work on using userns_call later.
pon., 30 lip 2018 o 18:18 Paweł Stradomski <pstradomski at google.com> napisał(a):
>
> pt., 27 lip 2018 o 04:00 Andrei Vagin <avagin at virtuozzo.com> wrote:
> > Applied, thanks! Here is one in-line comment
> >
> > > --- a/criu/mem.c
> > > +++ b/criu/mem.c
> > > @@ -1271,7 +1271,11 @@ static int prepare_vma_ios(struct pstree_item *t, struct task_restore_args *ta)
> > >  {
> > >       struct cr_img *pages;
> > >
> > > -     pages = open_image(CR_FD_PAGES, O_RSTR, rsti(t)->pages_img_id);
> > > +     /* if auto-dedup is on we need RDWR mode to be able to punch holes
> > > +      * in the input files (in restorer.c)
> > > +      */
> > > +     pages = open_image(CR_FD_PAGES, opts.auto_dedup ? O_RDWR : O_RSTR,
> > > +                             rsti(t)->pages_img_id);
> >
> > This doesn't work for userns tests:
> >
> > CRIU opens an image from a target userns and fails to open it for read-write:
> > -rw-r--r-- 1 root root 106496 Jul 27 04:54 test/dump/zdtm/static/env00/43/1/pages-1.img
> >
> > Probably, we need to use userns_call() to open images in this case.
>
> I've added fallback to opening read-only if read-write fails.
> userns_call seems like a good idea, but given that open_image not only
> opens file but also allocates a few objects, we'd need to break it
> apart first. I'd like to do that in a separate commit.
>
> --
> Paweł Stradomski



-- 
Paweł Stradomski