Re: [edk2-devel] [patch 1/3] EmbeddedPkg: Unload image on EFI_SECURITY_VIOLATION

["Laszlo Ersek" <lersek@redhat.com>]

On 09/04/19 10:25, Dandan Bi wrote:
> For the LoadImage() boot service, with EFI_SECURITY_VIOLATION retval,
> the Image was loaded and an ImageHandle was created with a valid
> EFI_LOADED_IMAGE_PROTOCOL, but the image can not be started right now.
> This follows UEFI Spec.
> 
> But if the caller of LoadImage() doesn't have the option to defer
> the execution of an image, we can not treat EFI_SECURITY_VIOLATION
> like any other LoadImage() error, we should unload image for the
> EFI_SECURITY_VIOLATION to avoid resource leak.
> 
> This patch is to do error handling for EFI_SECURITY_VIOLATION explicitly
> for the callers in EmbeddedPkg which don't have the policy to defer the
> execution of the image.
> 
> Cc: Leif Lindholm <leif.lindholm@linaro.org>
> Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
> Cc: Laszlo Ersek <lersek@redhat.com>
> REF: https://bugzilla.tianocore.org/show_bug.cgi?id=1992
> Signed-off-by: Dandan Bi <dandan.bi@intel.com>
> ---
>  .../AndroidFastboot/Arm/BootAndroidBootImg.c         |  9 +++++++++
>  .../Library/AndroidBootImgLib/AndroidBootImgLib.c    | 12 ++++++++++++
>  2 files changed, 21 insertions(+)
> 
> diff --git a/EmbeddedPkg/Application/AndroidFastboot/Arm/BootAndroidBootImg.c b/EmbeddedPkg/Application/AndroidFastboot/Arm/BootAndroidBootImg.c
> index 591afbe7cc..9fa28e3390 100644
> --- a/EmbeddedPkg/Application/AndroidFastboot/Arm/BootAndroidBootImg.c
> +++ b/EmbeddedPkg/Application/AndroidFastboot/Arm/BootAndroidBootImg.c
> @@ -71,10 +71,19 @@ StartEfiApplication (
>  
>    // Load the image from the device path with Boot Services function
>    Status = gBS->LoadImage (TRUE, ParentImageHandle, DevicePath, NULL, 0,
>                    &ImageHandle);
>    if (EFI_ERROR (Status)) {
> +    //
> +    // With EFI_SECURITY_VIOLATION retval, the Image was loaded and an ImageHandle was created
> +    // with a valid EFI_LOADED_IMAGE_PROTOCOL, but the image can not be started right now.
> +    // If the caller doesn't have the option to defer the execution of an image, we should
> +    // unload image for the EFI_SECURITY_VIOLATION to avoid resource leak.
> +    //
> +    if (Status == EFI_SECURITY_VIOLATION) {
> +      gBS->UnloadImage (ImageHandle);
> +    }
>      return Status;
>    }
>  
>    // Passed LoadOptions to the EFI Application
>    if (LoadOptionsSize != 0) {

The return status exists here originally, so this change is strictly an
improvement.

But:

> diff --git a/EmbeddedPkg/Library/AndroidBootImgLib/AndroidBootImgLib.c b/EmbeddedPkg/Library/AndroidBootImgLib/AndroidBootImgLib.c
> index d9e7aa7d2b..2e9e74db1d 100644
> --- a/EmbeddedPkg/Library/AndroidBootImgLib/AndroidBootImgLib.c
> +++ b/EmbeddedPkg/Library/AndroidBootImgLib/AndroidBootImgLib.c
> @@ -439,10 +439,22 @@ AndroidBootImgBoot (
>                                             + KernelSize;
>  
>    Status = gBS->LoadImage (TRUE, gImageHandle,
>                             (EFI_DEVICE_PATH *)&KernelDevicePath,
>                             (VOID*)(UINTN)Kernel, KernelSize, &ImageHandle);
> +  if (EFI_ERROR (Status)) {
> +    //
> +    // With EFI_SECURITY_VIOLATION retval, the Image was loaded and an ImageHandle was created
> +    // with a valid EFI_LOADED_IMAGE_PROTOCOL, but the image can not be started right now.
> +    // If the caller doesn't have the option to defer the execution of an image, we should
> +    // unload image for the EFI_SECURITY_VIOLATION to avoid resource leak.
> +    //
> +    if (Status == EFI_SECURITY_VIOLATION) {
> +      gBS->UnloadImage (ImageHandle);
> +    }
> +    return Status;
> +  }
>  
>    // Set kernel arguments
>    Status = gBS->HandleProtocol (ImageHandle, &gEfiLoadedImageProtocolGuid,
>                                  (VOID **) &ImageInfo);
>    ImageInfo->LoadOptions = NewKernelArg;
> 

The return status is introduced afresh here. And, at this point in the
function, we have allocated at least "NewKernelArg". (And probably more
resources.)

If we take an early return here, then we're going to leak NewKernelArg.

Hm... Hmm. Well, even without this patch, if the loaded and started
image returns, then the AndroidBootImgBoot() function leaks
"NewKernelArg" anyway.

Sigh. I guess the error handling in this function should be cleaned up
first, to solve the present issue cleanly. On the other hand, I'm not
sure how much this function is used at all. And I guess the patch
doesn't make things worse than they are.

Acked-by: Laszlo Ersek <lersek@redhat.com>