| Message ID | 20210930085932.1.I8043d61cc238e0168e2f4ca5f4783223434aa587@changeid |
|---|---|
| State | Accepted |
| Commit | 8244a3bc27b3efd057da154b8d7e414670d5044f |
| Headers | show |
| Series |
"drm/prime: Fix use after free in mmap with drm_gem_ttm_mmap"
( rev:
1
)
in
DRI devel |
diff --git a/drivers/gpu/drm/drm_prime.c b/drivers/gpu/drm/drm_prime.c index 2a54f86856af..e1854fd24bb0 100644 --- a/drivers/gpu/drm/drm_prime.c +++ b/drivers/gpu/drm/drm_prime.c @@ -719,11 +719,13 @@ int drm_gem_prime_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma) if (obj->funcs && obj->funcs->mmap) { vma->vm_ops = obj->funcs->vm_ops; + drm_gem_object_get(obj); ret = obj->funcs->mmap(obj, vma); - if (ret) + if (ret) { + drm_gem_object_put(obj); return ret; + } vma->vm_private_data = obj; - drm_gem_object_get(obj); return 0; }
Hi Am 30.09.21 um 01:00 schrieb Anand K Mistry: > drm_gem_ttm_mmap() drops a reference to the gem object on success. If > the gem object's refcount == 1 on entry to drm_gem_prime_mmap(), that > drop will free the gem object, and the subsequent drm_gem_object_get() > will be a UAF. Fix by grabbing a reference before calling the mmap > helper. > > This issue was forseen when the reference dropping was adding in > commit 9786b65bc61ac ("drm/ttm: fix mmap refcounting"): > "For that to work properly the drm_gem_object_get() call in > drm_gem_ttm_mmap() must be moved so it happens before calling > obj->funcs->mmap(), otherwise the gem refcount would go down > to zero." > > Signed-off-by: Anand K Mistry <amistry@google.com> Acked-by: Thomas Zimmermann <tzimmermann@suse.de> This looks fine to me, but it affects many drivers. Let's maybe wait a bit if more reviews come it. Best regards Thomas > --- > > drivers/gpu/drm/drm_prime.c | 6 ++++-- > 1 file changed, 4 insertions(+), 2 deletions(-) > > diff --git a/drivers/gpu/drm/drm_prime.c b/drivers/gpu/drm/drm_prime.c > index 2a54f86856af..e1854fd24bb0 100644 > --- a/drivers/gpu/drm/drm_prime.c > +++ b/drivers/gpu/drm/drm_prime.c > @@ -719,11 +719,13 @@ int drm_gem_prime_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma) > if (obj->funcs && obj->funcs->mmap) { > vma->vm_ops = obj->funcs->vm_ops; > > + drm_gem_object_get(obj); > ret = obj->funcs->mmap(obj, vma); > - if (ret) > + if (ret) { > + drm_gem_object_put(obj); > return ret; > + } > vma->vm_private_data = obj; > - drm_gem_object_get(obj); > return 0; > } > > -- Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Felix Imendörffer
Any movement on merging this patch? Not urgent on my end (we have this patch in our tree), but I think other amd users might run into this UAF. On Thu, 30 Sept 2021 at 17:21, Thomas Zimmermann <tzimmermann@suse.de> wrote: > > Hi > > Am 30.09.21 um 01:00 schrieb Anand K Mistry: > > drm_gem_ttm_mmap() drops a reference to the gem object on success. If > > the gem object's refcount == 1 on entry to drm_gem_prime_mmap(), that > > drop will free the gem object, and the subsequent drm_gem_object_get() > > will be a UAF. Fix by grabbing a reference before calling the mmap > > helper. > > > > This issue was forseen when the reference dropping was adding in > > commit 9786b65bc61ac ("drm/ttm: fix mmap refcounting"): > > "For that to work properly the drm_gem_object_get() call in > > drm_gem_ttm_mmap() must be moved so it happens before calling > > obj->funcs->mmap(), otherwise the gem refcount would go down > > to zero." > > > > Signed-off-by: Anand K Mistry <amistry@google.com> > > Acked-by: Thomas Zimmermann <tzimmermann@suse.de> > > This looks fine to me, but it affects many drivers. Let's maybe wait a > bit if more reviews come it. > > Best regards > Thomas > > > --- > > > > drivers/gpu/drm/drm_prime.c | 6 ++++-- > > 1 file changed, 4 insertions(+), 2 deletions(-) > > > > diff --git a/drivers/gpu/drm/drm_prime.c b/drivers/gpu/drm/drm_prime.c > > index 2a54f86856af..e1854fd24bb0 100644 > > --- a/drivers/gpu/drm/drm_prime.c > > +++ b/drivers/gpu/drm/drm_prime.c > > @@ -719,11 +719,13 @@ int drm_gem_prime_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma) > > if (obj->funcs && obj->funcs->mmap) { > > vma->vm_ops = obj->funcs->vm_ops; > > > > + drm_gem_object_get(obj); > > ret = obj->funcs->mmap(obj, vma); > > - if (ret) > > + if (ret) { > > + drm_gem_object_put(obj); > > return ret; > > + } > > vma->vm_private_data = obj; > > - drm_gem_object_get(obj); > > return 0; > > } > > > > > > -- > Thomas Zimmermann > Graphics Driver Developer > SUSE Software Solutions Germany GmbH > Maxfeldstr. 5, 90409 Nürnberg, Germany > (HRB 36809, AG Nürnberg) > Geschäftsführer: Felix Imendörffer
Hi Am 03.11.21 um 01:12 schrieb Anand K. Mistry: > Any movement on merging this patch? Not urgent on my end (we have this > patch in our tree), but I think other amd users might run into this > UAF. Thanks for reminding. I've merged your patch into drm-misc-fixes. Best regards Thomas > > On Thu, 30 Sept 2021 at 17:21, Thomas Zimmermann <tzimmermann@suse.de> wrote: >> >> Hi >> >> Am 30.09.21 um 01:00 schrieb Anand K Mistry: >>> drm_gem_ttm_mmap() drops a reference to the gem object on success. If >>> the gem object's refcount == 1 on entry to drm_gem_prime_mmap(), that >>> drop will free the gem object, and the subsequent drm_gem_object_get() >>> will be a UAF. Fix by grabbing a reference before calling the mmap >>> helper. >>> >>> This issue was forseen when the reference dropping was adding in >>> commit 9786b65bc61ac ("drm/ttm: fix mmap refcounting"): >>> "For that to work properly the drm_gem_object_get() call in >>> drm_gem_ttm_mmap() must be moved so it happens before calling >>> obj->funcs->mmap(), otherwise the gem refcount would go down >>> to zero." >>> >>> Signed-off-by: Anand K Mistry <amistry@google.com> >> >> Acked-by: Thomas Zimmermann <tzimmermann@suse.de> >> >> This looks fine to me, but it affects many drivers. Let's maybe wait a >> bit if more reviews come it. >> >> Best regards >> Thomas >> >>> --- >>> >>> drivers/gpu/drm/drm_prime.c | 6 ++++-- >>> 1 file changed, 4 insertions(+), 2 deletions(-) >>> >>> diff --git a/drivers/gpu/drm/drm_prime.c b/drivers/gpu/drm/drm_prime.c >>> index 2a54f86856af..e1854fd24bb0 100644 >>> --- a/drivers/gpu/drm/drm_prime.c >>> +++ b/drivers/gpu/drm/drm_prime.c >>> @@ -719,11 +719,13 @@ int drm_gem_prime_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma) >>> if (obj->funcs && obj->funcs->mmap) { >>> vma->vm_ops = obj->funcs->vm_ops; >>> >>> + drm_gem_object_get(obj); >>> ret = obj->funcs->mmap(obj, vma); >>> - if (ret) >>> + if (ret) { >>> + drm_gem_object_put(obj); >>> return ret; >>> + } >>> vma->vm_private_data = obj; >>> - drm_gem_object_get(obj); >>> return 0; >>> } >>> >>> >> >> -- >> Thomas Zimmermann >> Graphics Driver Developer >> SUSE Software Solutions Germany GmbH >> Maxfeldstr. 5, 90409 Nürnberg, Germany >> (HRB 36809, AG Nürnberg) >> Geschäftsführer: Felix Imendörffer > -- Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Ivo Totev
drm_gem_ttm_mmap() drops a reference to the gem object on success. If the gem object's refcount == 1 on entry to drm_gem_prime_mmap(), that drop will free the gem object, and the subsequent drm_gem_object_get() will be a UAF. Fix by grabbing a reference before calling the mmap helper. This issue was forseen when the reference dropping was adding in commit 9786b65bc61ac ("drm/ttm: fix mmap refcounting"): "For that to work properly the drm_gem_object_get() call in drm_gem_ttm_mmap() must be moved so it happens before calling obj->funcs->mmap(), otherwise the gem refcount would go down to zero." Signed-off-by: Anand K Mistry <amistry@google.com> --- drivers/gpu/drm/drm_prime.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)