Re: [PATCH] MdeModulePkg/DxeCore: base code protection on permission attributes

[Ard Biesheuvel <ard.biesheuvel@linaro.org>]

On 17 March 2017 at 12:07, Laszlo Ersek <lersek@redhat.com> wrote:
> On 02/26/17 15:00, Ard Biesheuvel wrote:
>> On 25 February 2017 at 04:04, Yao, Jiewen <jiewen.yao@intel.com> wrote:
>>> Hi Ard
>>> I agree with you on this enhancement.
>>>
>>> I prefer to adding the description as comment in the code, so that people
>>> can get clear picture when he/she reads the code.
>>>
>>> //
>>> // Instead of assuming that a PE/COFF section of type EFI_IMAGE_SCN_CNT_CODE
>>> // can always be mapped read-only, classify a section as a code section only
>>> // if it has the executable attribute set and the writable attribute
>>> cleared.
>>> //
>>> // This adheres more closely to the PE/COFF spec, and avoids issues with
>>> // Linux OS loaders that consists of a single read/write/execute section.
>>> //
>>> if ((Section[Index].Characteristics & (EFI_IMAGE_SCN_MEM_WRITE |
>>> EFI_IMAGE_SCN_MEM_EXECUTE)) == EFI_IMAGE_SCN_MEM_EXECUTE) {
>>>
>>> With comment update, reviewed-by: Jiewen.yao@intel.com
>>>
>>
>> Thanks Jiewen
>>
>> Pushed as a2ed40c02bf2
>
> Is it possible that (recent?) Linux EFI stubs (aarch64) don't pass the
> above check? I got a report from a colleague:
>
> !!!!!!!!  ProtectUefiImageCommon - CodeSegmentCount is 0  !!!!!!!!
> EFI stub: Booting Linux Kernel...
> EFI stub: Using DTB from configuration table
> EFI stub: Exiting boot services and installing virtual address map...
>
> I tried to reproduce it with "4.5.0-15.el7.aarch64", and with
> "4.8.7-300.fc25.aarch64", and I'm not seeing the message with either.
>
> I asked him about the exact kernel version (no answer yet, his workday
> hasn't started yet).
>

Well, the warning may be a bit loud, but this is expected behavior. I
expect to be able to queue the linux/arm64 changes that split the
PE/COFF sections and align them to 4 KB for v4.12, and so current
kernels all consists of a single rwx section, which does not qualify
as a code section.

> Any idea how I could validate the section headers of a (decompressed)
> kernel image? I tried "aarch64-linux-gnu-objdump --section-headers", but
> it doesn't recognize the image.
>

It's a PE/COFF image not an ELF image.