Re: [PATCH v2] drm/radeon/radeon_kms: Fix a NULL pointer dereference in radeon_driver_open_kms()

Am 30.11.21 um 16:33 schrieb Zhou Qingyang:

In radeon_driver_open_kms(), radeon_vm_bo_add() is assigned to vm->ib_bo_va and passes and used in radeon_vm_bo_set_addr(). In radeon_vm_bo_set_addr(), there is a dereference of vm->ib_bo_va, which could lead to a NULL pointer dereference on failure of radeon_vm_bo_add().

Fix this bug by adding a check of vm->ib_bo_va.

This bug was found by a static analyzer. The analysis employs differential checking to identify inconsistent security operations (e.g., checks or kfrees) between two code paths and confirms that the inconsistent operations are not recovered in the current function or the callers, so they constitute bugs.

Note that, as a bug found by static analysis, it can be a false positive or hard to trigger. Multiple researchers have cross-reviewed the bug.

Builds with CONFIG_DRM_RADEON=m show no new warnings, and our static analyzer no longer warns about this code.

Fixes: cc9e67e3d700 ("drm/radeon: fix VM IB handling") Signed-off-by: Zhou Qingyang zhou1615@umn.edu

---

Changes in v2:

• Improve the error handling into goto style

drivers/gpu/drm/radeon/radeon_kms.c | 24 ++++++++++++++---------- 1 file changed, 14 insertions(+), 10 deletions(-)

diff --git a/drivers/gpu/drm/radeon/radeon_kms.c b/drivers/gpu/drm/radeon/radeon_kms.c index 482fb0ae6cb5..e49a9d160e52 100644 --- a/drivers/gpu/drm/radeon/radeon_kms.c +++ b/drivers/gpu/drm/radeon/radeon_kms.c @@ -649,6 +649,8 @@ int radeon_driver_open_kms(struct drm_device *dev, struct drm_file *file_priv) { struct radeon_device *rdev = dev->dev_private; int r;

• struct radeon_fpriv *fpriv;
• struct radeon_vm *vm;

Please keep variables like "i" or "r" declared last.

file_priv->driver_priv = NULL;

@@ -660,8 +662,6 @@ int radeon_driver_open_kms(struct drm_device *dev, struct drm_file *file_priv)

/* new gpu have virtual address space support */ if (rdev->family >= CHIP_CAYMAN) {

• struct radeon_fpriv *fpriv;
  

• struct radeon_vm *vm;
  
  

  fpriv = kzalloc(sizeof(*fpriv), GFP_KERNEL); if (unlikely(!fpriv)) {

@@ -673,34 +673,38 @@ int radeon_driver_open_kms(struct drm_device *dev, struct drm_file *file_priv) vm = &fpriv->vm; r = radeon_vm_init(rdev, vm); if (r) {

• 		kfree(fpriv);
  

• 		goto out_suspend;
  

• 		goto out_fpriv;
  }
  
  r = radeon_bo_reserve(rdev->ring_tmp_bo.bo, false);
  if (r) {
  

• 		radeon_vm_fini(rdev, vm);
  

• 		kfree(fpriv);
  

• 		goto out_suspend;
  

• 		goto out_vm_fini;
  }
  
  /* map the ib pool buffer read only into
   * virtual address space */
  vm->ib_bo_va = radeon_vm_bo_add(rdev, vm,
  				rdev->ring_tmp_bo.bo);
  

• 	if (!vm->ib_bo_va) {
  

• 		r = -ENOMEM;
  

• 		goto out_vm_fini;
  

• 	}
  

• r = radeon_vm_bo_set_addr(rdev, vm->ib_bo_va,
  			  RADEON_VA_IB_OFFSET,
  			  RADEON_VM_PAGE_READABLE |
  			  RADEON_VM_PAGE_SNOOPED);
  if (r) {
  

• 		radeon_vm_fini(rdev, vm);
  

• 		kfree(fpriv);
  

• 		goto out_suspend;
  

• 		goto out_vm_fini;
  }
  

  } file_priv->driver_priv = fpriv; }

That here won't work.

+out_vm_fini:

• radeon_vm_fini(rdev, vm);

+out_fpriv:

• kfree(fpriv);

You are finishing the VM and freeing up the memory in the good case now as well.

Christian.

out_suspend: pm_runtime_mark_last_busy(dev->dev); pm_runtime_put_autosuspend(dev->dev);