On Thu, Jun 04, 2026 at 07:58:47PM +0000, [email protected] wrote: > Thank you for your contribution! Sashiko AI review found 1 potential issue(s) > to consider: > - [High] The ioctl returns 0 on `args->handle == args->new_handle` without > validating if the handle actually exists in the IDR. > -- > > commit c59eb6aaceabfd08a4835d7c2c0133017afcb072 > Author: Simona Vetter <[email protected]> > > drm/gem: Try to fix change_handle ioctl, attempt 4 > > This commit attempts to fix a race condition in the change_handle ioctl by > reworking the two-stage idr replacement logic and renaming variables for > clarity. It also unifies the object lookup and idr_replace steps and disables > the ioctl pending testing. > > > diff --git a/drivers/gpu/drm/drm_gem.c b/drivers/gpu/drm/drm_gem.c > > index ac751a42db281..55caad80a3503 100644 > > --- a/drivers/gpu/drm/drm_gem.c > > +++ b/drivers/gpu/drm/drm_gem.c > [ ... ] > > @@ -1028,52 +1028,36 @@ int drm_gem_change_handle_ioctl(struct drm_device > > *dev, void *data, > > /* idr_alloc() limitation. */ > > if (args->new_handle > INT_MAX) > > return -EINVAL; > > - handle = args->new_handle; > > - > > - obj = drm_gem_object_lookup(file_priv, args->handle); > > - if (!obj) > > - return -ENOENT; > > + new_handle = args->new_handle; > > > > - if (args->handle == handle) { > > - ret = 0; > > - goto out; > > - } > > + if (args->handle == new_handle) > > + return 0; > > [Severity: High] > Does this new early return skip handle validation? > > Previously, drm_gem_object_lookup() was called before this check, returning > -ENOENT if the given handle was invalid. In the new code, if a user passes > a non-existent handle but sets new_handle to the exact same value, the ioctl > will return 0 (success) instead of -ENOENT. > > Should the handle existence still be verified before returning success here?
Since the igt for this ioctl still hasn't landed I don't know, either kinda makes some or little sense really since why exactly is userspace asking for this. We could also reject this case as a userspace bug with -EALREADY or something like that. Either way this needs to be nailed down with an igt, which really, really, really needs to land this time before we re-enable this all me thinks. -Sima -- Simona Vetter Software Engineer http://blog.ffwll.ch
