[Intel-gfx] [PATCH] drm/i915/gvt: fix double-free bug in split_2MB_gtt_entry.

[Zheng Hacker]

Here I introduced is_error to judge if the ppgtt_invalidate_spt is
called from good case or not. Only free spt in good case, leave it to
handle for the error path of caller.

Zheng Hacker <hackerzheng666 at gmail.com> 于2022年9月16日周五 14:39写道：
>
> From 8d95c1399e3ff345500a575e21254a73b0c89144 Mon Sep 17 00:00:00 2001
> From: xmzyshypnc <1002992920 at qq.com>
> Date: Fri, 16 Sep 2022 14:37:48 +0800
> Subject: [PATCH] drm/i915/gvt: fix double-free bug in split_2MB_gtt_entry
>
> There is a double-free security bug in split_2MB_gtt_entry.
>
> Here is a calling chain :
> ppgtt_populate_spt->ppgtt_populate_shadow_entry->split_2MB_gtt_entry.
> If intel_gvt_dma_map_guest_page failed, it will call
> ppgtt_invalidate_spt, which will finally call ppgtt_free_spt and
> kfree(spt). But the caller does not notice that, and it will call
> ppgtt_free_spt again in error path.
>
> Fix this by only freeing spt in ppgtt_invalidate_spt in good case.
>
> Signed-off-by: xmzyshypnc <1002992920 at qq.com>
> ---
>  drivers/gpu/drm/i915/gvt/gtt.c | 16 +++++++++-------
>  1 file changed, 9 insertions(+), 7 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/gvt/gtt.c b/drivers/gpu/drm/i915/gvt/gtt.c
> index 9f14fded8c0c..31d2a8d56384 100644
> --- a/drivers/gpu/drm/i915/gvt/gtt.c
> +++ b/drivers/gpu/drm/i915/gvt/gtt.c
> @@ -959,7 +959,7 @@ static inline int ppgtt_put_spt(struct
> intel_vgpu_ppgtt_spt *spt)
>   return atomic_dec_return(&spt->refcount);
>  }
>
> -static int ppgtt_invalidate_spt(struct intel_vgpu_ppgtt_spt *spt);
> +static int ppgtt_invalidate_spt(struct intel_vgpu_ppgtt_spt *sptm,
> int is_error);
>
>  static int ppgtt_invalidate_spt_by_shadow_entry(struct intel_vgpu *vgpu,
>   struct intel_gvt_gtt_entry *e)
> @@ -995,7 +995,7 @@ static int
> ppgtt_invalidate_spt_by_shadow_entry(struct intel_vgpu *vgpu,
>   ops->get_pfn(e));
>   return -ENXIO;
>   }
> - return ppgtt_invalidate_spt(s);
> + return ppgtt_invalidate_spt(s, 0);
>  }
>
>  static inline void ppgtt_invalidate_pte(struct intel_vgpu_ppgtt_spt *spt,
> @@ -1016,7 +1016,7 @@ static inline void ppgtt_invalidate_pte(struct
> intel_vgpu_ppgtt_spt *spt,
>   intel_gvt_dma_unmap_guest_page(vgpu, pfn << PAGE_SHIFT);
>  }
>
> -static int ppgtt_invalidate_spt(struct intel_vgpu_ppgtt_spt *spt)
> +static int ppgtt_invalidate_spt(struct intel_vgpu_ppgtt_spt *spt, int is_error)
>  {
>   struct intel_vgpu *vgpu = spt->vgpu;
>   struct intel_gvt_gtt_entry e;
> @@ -1059,9 +1059,11 @@ static int ppgtt_invalidate_spt(struct
> intel_vgpu_ppgtt_spt *spt)
>   }
>   }
>
> - trace_spt_change(spt->vgpu->id, "release", spt,
> + if (!is_error) {
> + trace_spt_change(spt->vgpu->id, "release", spt,
>   spt->guest_page.gfn, spt->shadow_page.type);
> - ppgtt_free_spt(spt);
> + ppgtt_free_spt(spt);
> + }
>   return 0;
>  fail:
>   gvt_vgpu_err("fail: shadow page %p shadow entry 0x%llx type %d\n",
> @@ -1215,7 +1217,7 @@ static int split_2MB_gtt_entry(struct intel_vgpu *vgpu,
>   ret = intel_gvt_dma_map_guest_page(vgpu, start_gfn + sub_index,
>      PAGE_SIZE, &dma_addr);
>   if (ret) {
> - ret = ppgtt_invalidate_spt(spt);
> + ret = ppgtt_invalidate_spt(spt, 1);
>   return ret;
>   }
>   sub_se.val64 = se->val64;
> @@ -1393,7 +1395,7 @@ static int
> ppgtt_handle_guest_entry_removal(struct intel_vgpu_ppgtt_spt *spt,
>   ret = -ENXIO;
>   goto fail;
>   }
> - ret = ppgtt_invalidate_spt(s);
> + ret = ppgtt_invalidate_spt(s, 0);
>   if (ret)
>   goto fail;
>   } else {
> --
> 2.25.1
>
> Zheng Hacker <hackerzheng666 at gmail.com> 于2022年9月8日周四 19:59写道：
> >
> > Hi Greg,
> >
> > I got it, Greg.
> >
> > Mid-Autumn Festival is coming and I will have a couple of days off.
> > I'll see what I can do after holiday :)
> >
> > Regards,
> >
> > Zheng Wang
> >
> > 在 2022年9月8日星期四，Greg KH <gregkh at linuxfoundation.org> 写道：
> >>
> >> On Thu, Sep 08, 2022 at 05:09:40PM +0800, Zheng Hacker wrote:
> >> > Hi Zhenyu,
> >> >
> >> > This issue has been open for a few days. Could you plz write a patch
> >> > for that :) I'm not familiar with the logical code here.
> >>
> >> As this is only able to be hit in a theoretical system, it isn't that
> >> high of a priority, if any priority at all.  Why not try to write a
> >> patch for it yourself to help resolve the issue faster?
> >>
> >> thanks,
> >>
> >> greg k-h