Re: [PATCH v4 4/9] drm/rockchip/rockchip_drm_gem.c: Convert to use vm_insert_range

[Russell King - ARM Linux]

On Wed, Dec 19, 2018 at 05:16:09PM +0530, Souptick Joarder wrote:
> On Wed, Dec 19, 2018 at 3:02 PM Russell King - ARM Linux
> <linux@xxxxxxxxxxxxxxx> wrote:
> >
> > On Wed, Dec 19, 2018 at 09:01:09AM +0530, Souptick Joarder wrote:
> > > On Tue, Dec 18, 2018 at 6:31 PM Russell King - ARM Linux
> > > <linux@xxxxxxxxxxxxxxx> wrote:
> > > >
> > > > On Tue, Dec 18, 2018 at 06:24:29PM +0530, Souptick Joarder wrote:
> > > > > On Tue, Dec 18, 2018 at 6:03 PM Russell King - ARM Linux
> > > > > <linux@xxxxxxxxxxxxxxx> wrote:
> > > > > >
> > > > > > On Tue, Dec 18, 2018 at 05:36:04PM +0530, Souptick Joarder wrote:
> > > > > > > On Tue, Dec 18, 2018 at 3:27 PM Russell King - ARM Linux
> > > > > > > <linux@xxxxxxxxxxxxxxx> wrote:
> > > > > > > > This looks like a change in behaviour.
> > > > > > > >
> > > > > > > > If user_count is zero, and offset is zero, then we pass into
> > > > > > > > vm_insert_range() a page_count of zero, and vm_insert_range() does
> > > > > > > > nothing and returns zero.
> > > > > > > >
> > > > > > > > However, as we can see from the above code, the original behaviour
> > > > > > > > was to return -ENXIO in that case.
> > > > > > >
> > > > > > > I think these checks are not necessary. I am not sure if we get into mmap
> > > > > > > handlers of driver with user_count = 0.
> > > > > >
> > > > > > I'm not sure either, I'm just pointing out the change of behaviour.
> > > > >
> > > > > Ok. I think feedback from Heiko might be helpful here :)
> > > > >
> > > > > >
> > > > > > > > The other thing that I'm wondering is that if (eg) count is 8 (the
> > > > > > > > object is 8 pages), offset is 2, and the user requests mapping 6
> > > > > > > > pages (user_count = 6), then we call vm_insert_range() with a
> > > > > > > > pages of rk_obj->pages + 2, and a pages_count of 6 - 2 = 4. So we
> > > > > > > > end up inserting four pages.
> > > > > > >
> > > > > > > Considering the scenario, user_count will remain 8 (user_count =
> > > > > > > vma_pages(vma) ). ? No ?
> > > > > > > Then we call vm_insert_range() with a pages of rk_obj->pages + 2, and
> > > > > > > a pages_count
> > > > > > > of 8 - 2 = 6. So we end up inserting 6 pages.
> > > > > > >
> > > > > > > Please correct me if I am wrong.
> > > > > >
> > > > > > vma_pages(vma) is the number of pages that the user requested, it is
> > > > > > the difference between vma->vm_end and vma->vm_start in pages.  As I
> > > > > > said above, "the user requests mapping 6 pages", so vma_pages() will
> > > > > > be 6, and so user_count will also be 6.  You are passing
> > > > > > user_count - offset into vm_insert_range(), which will be 6 - 2 = 4
> > > > > > in my example.  This is two pages short of what the user requested.
> > > > > >
> > > > >
> > > > > So, this should be the correct behavior.
> > > > >
> > > > >                  return vm_insert_range(vma, vma->vm_start,
> > > > > rk_obj->pages + offset,
> > > > >                                                           user_count);
> > > >
> > > > ... and by doing so, you're introducing another instance of the same
> > > > bug I pointed out in patch 2.
> > >
> > > Sorry but didn't get it ? How it will be similar to the bug pointed
> > > out in patch 2 ?
> >
> 
> Thanks for the detail explanation.
> 
> > Towards the top of this function, you have:
> >
> >         unsigned long user_count = vma_pages(vma);
> >
> > So what you are proposing does:
> >
> >         return vm_insert_range(vma, vma->vm_start, rk_obj->pages + offset,
> >                                vma_pages(vma));
> >
> > Now if we look inside vm_insert_range():
> >
> > +int vm_insert_range(struct vm_area_struct *vma, unsigned long addr,
> > +                       struct page **pages, unsigned long page_count)
> > +{
> > +       unsigned long uaddr = addr;
> > +       int ret = 0, i;
> > +
> > +       if (page_count > vma_pages(vma))
> > +               return -ENXIO;
> > +
> > +       for (i = 0; i < page_count; i++) {
> > +               ret = vm_insert_page(vma, uaddr, pages[i]);
> > +               if (ret < 0)
> > +                       return ret;
> > +               uaddr += PAGE_SIZE;
> > +       }
> >
> > So, page_count _is_ vma_pages(vma).  So this code does these operations:
> >
> >         if (vma_pages(vma) > vma_pages(vma))
> >                 return -ENXIO;
> >
> > This will always be false.  I've already stated in my reply to patch 2
> > in paragraph 3 about the uselessness of this test.
> 
> Agree, this will be always false for this particular/ similar instances.
> But there are places [3/9], [6/9], [9/9] where page_count is already set
> and it might be good to just cross check page_count > vma_pages(vma).
> 
> This was discussed during review of v3 [1/9].
> https://patchwork.kernel.org/patch/10716601/
> 
> We can discuss again and if not needed it can be removed in v5.
> 
> >
> >         for (i = 0; i < vma_pages(vma); i++) {
> >                 ret = vm_insert_page(vma, uaddr, pages[i]);
> >
> > So the loop will iterate over the number of pages that the user requested.
> >
> > Now, taking another example.  The object is again 8 pages long, so
> > indexes 0 through 7 in its page array are valid.  The user requests
> > 8 pages at offset 2 into the object.  Also as already stated in
> > paragraph 3 of my reply to patch 2.
> >
> > vma_pages(vma) is 8.  offset = 2.
> >
> > So we end up _inside_ vm_insert_range() with:
> >
> >         if (8 > 8)
> >                 return -ENXIO;
> >
> > As stated, always false.
> >
> >         for (i = 0; i < 8; i++) {
> >                 ret = vm_insert_page(vma, vaddr, rk_obj->pages[2 + i]);
> >
> > Which means we iterate over rk_obj->pages indicies from 2 through 9
> > inclusive.
> >
> > Since only 0 through 7 are valid, we have walked off the end of the
> > array, and attempted to map an invalid struct page pointer - we could
> > be lucky, and it could point at some struct page (potentially causing
> > us to map some sensitive page - maybe containing your bank details or
> > root password... Or it could oops the kernel.
> 
> Consider the 2nd example.
> The object is again 8 pages long, so indexes 0 through 7 in
> its page array are valid.  The user requests 8 pages at offset 2
> into the object.
> 
> The original code look like -
> 
>              unsigned long user_count = vma_pages(vma); // 8
>              unsigned long end = user_count + offset // 8 + 2 = 10
>               ...
>               for (i = offset (2) ; i < end ( 10) ; i++) {
>                   ret = vm_insert_page(vma, uaddr, rk_obj->pages[i]);
>                   if (ret)
>                      return ret;
>                   uaddr += PAGE_SIZE;
>               }
> 
> we iterate over rk_obj->pages indices from 2 through 9.
> Does it indicates the actual code have a bug when *offset != 0*.

Please look at _all_ of the original code.

Just like in your patch 2, you removed the tests that protect against
this overflow:

-       unsigned int i, count = obj->size >> PAGE_SHIFT;
        unsigned long user_count = vma_pages(vma);
-       unsigned long uaddr = vma->vm_start;
        unsigned long offset = vma->vm_pgoff;
-       unsigned long end = user_count + offset;
-       int ret;
-
-       if (user_count == 0)
-               return -ENXIO;
-       if (end > count)
-               return -ENXIO;

'count' will be 8.  'end' will be 10.  The existing code would have
therefore returned -ENXIO.

This is what I'm pointing out in my reviewed of your patches - they
remove necessary tests and, by doing so, introduce these array
overflows.

-- 
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 12.1Mbps down 622kbps up
According to speedtest.net: 11.9Mbps down 500kbps up