On Mon, May 09, 2022 at 10:00:41PM +0200, Javier Martinez Canillas wrote:

Hello Thomas,

On 5/9/22 20:32, Thomas Zimmermann wrote:

quoted

Hi

Am 09.05.22 um 18:33 schrieb Javier Martinez Canillas:

quoted

On 5/9/22 17:51, Andrzej Hajda wrote:

[snip]

quoted

quoted

quoted

quoted

+

Regarding drm:
What about drm_fb_helper_fini? It calls also framebuffer_release and is
called often from _remove paths (checked intel/radeon/nouveau). I guess
it should be fixed as well. Do you plan to fix it?

I think you are correct. Maybe we need something like the following?

diff --git a/drivers/gpu/drm/drm_fb_helper.c b/drivers/gpu/drm/drm_fb_helper.c
index d265a73313c9..b09598f7af28 100644
--- a/drivers/gpu/drm/drm_fb_helper.c
+++ b/drivers/gpu/drm/drm_fb_helper.c

@@ -631,7 +631,6 @@ void drm_fb_helper_fini(struct drm_fb_helper *fb_helper)
          if (info) {
                  if (info->cmap.len)
                          fb_dealloc_cmap(&info->cmap);
-               framebuffer_release(info);

After reviewing that code,  drm_fb_helper_fini() appears to be called 
from .fb_destroy (see drm_fbdev_release).  The code is hard to follow 
though.  If there another way of releasing the framebuffer here?

Andrzej mentioned intel/radeon/nouveau as example, I only looked at i915
and the call chain is the following as far as I can tell:

struct pci_driver i915_pci_driver = {
...
        .remove = i915_pci_remove,
...
};


i915_driver_remove
  intel_modeset_driver_remove_noirq
    intel_fbdev_fini
      intel_fbdev_destroy
        drm_fb_helper_fini
          framebuffer_release
              
So my underdestanding is that if a program has the emulated fbdev device
opened and the i915 module is removed, then a use-after-free would be
triggered on drm_fbdev_fb_destroy() once the program closes the device:

drm_fbdev_fb_destroy
  drm_fbdev_release(info->par); <-- info was already freed on .remove

Yeah the old drivers that haven't switched over to the drm_client based
fbdev emulations are all kinds of wrong and release it too early.

Note that they don't use the provided ->remove hook, since that would
result in double-cleanup like you point out. Those old drivers work more
like all the other fbdev drivers where all the cleanup is done in
->remove, and if it's a real hotunplug you just die in fire :-/

Switching them all over to the drm_client based fbdev helpers and
unexporting these old (dangerous!) functions would be really neat. But
it's also a loooooot of work, and generally those big drivers don't get
hotunplugged.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch