Re: [Patch 0/3] Add more checker for Tianocompress and Ueficompress

[Laszlo Ersek <lersek@redhat.com>]

On 10/19/18 08:40, Gao, Liming wrote:
> Laszlo:
>   I try to answer your question. I also include the BZ submitter
>   brent.holtsclaw@intel.com. Holtsclaw, please add your comments if my
>   info is not enough.
>
> Thanks
> Liming
>> -----Original Message-----
>> From: Laszlo Ersek [mailto:lersek@redhat.com]
>> Sent: Friday, October 19, 2018 12:01 AM
>> To: Gao, Liming <liming.gao@intel.com>; Zeng, Star
>> <star.zeng@intel.com>; edk2-devel@lists.01.org; Ard Biesheuvel
>> <ard.biesheuvel@linaro.org>
>> Subject: Re: [edk2] [Patch 0/3] Add more checker for Tianocompress
>> and Ueficompress
>>
>> On 10/18/18 15:36, Gao, Liming wrote:
>>> Laszlo and Star:
>>>   Thank your notes. I will add CVE number in patch subject although
>>>   it will make subject long than 80 characters.
>>
>> I agree the subject will be overlong, but I also think that including
>> the CVE numbers is important enough for that.
>>
>>> Here is my proposed patch subject: CVE-2017-5731..5735 MdePkg: Add
>>> more checker in UefiDecompressLib to access the valid buffer only.
>>
>> I suggest (based on tradition) that we keep the normal subject at the
>> front, and then we append the CVE numbers at the end. Also, we should
>> spell out all those CVE identifiers individually, if the same patch
>> solves them all. It should be possible to search the subject line for
>> any one of these CVE numbers in separation, using the official CVE
>> number format.
>>
>
> So, your proposal is like:  MdePkg: Add more checker in
> UefiDecompressLib to access the valid buffer only CVE-2017-5731
> CVE-2017-5732 CVE-2017-5733 CVE-2017-5734 CVE-2017-5735

Yes:

  MdePkg: Add more checker in UefiDecompressLib to access the valid buffer only (CVE-2017-5731 CVE-2017-5732 CVE-2017-5733 CVE-2017-5734 CVE-2017-5735)

It looks terrible, but the real subject is still readable to the left,
and subjects with searchable CVE numbers take priority (in my opinion
anyway).

Actually: I wonder why we needed five different CVEs, if they can all be
fixed with a small, single patch.

More precisely: looking at the patch in more detail, I see that the
patch fixes multiple functions / separate buffer overflows. Is it
possible to associate each CVE with a specific, small code change in the
patch? Because if it is possible, then I think we should split the patch
*per CVE*. The subjects would go:

- MdePkg/UefiDecompressLib: fix potential buffer overflow (CVE-2017-5731)
- MdePkg/UefiDecompressLib: fix potential buffer overflow (CVE-2017-5732)
- MdePkg/UefiDecompressLib: fix potential buffer overflow (CVE-2017-5733)
- MdePkg/UefiDecompressLib: fix potential buffer overflow (CVE-2017-5734)
- MdePkg/UefiDecompressLib: fix potential buffer overflow (CVE-2017-5735)

(71 characters, in each subject)

If such separation is technically possible, then I think it would be an
improvement; minimally for documentation purposes.

>>>   In PEI phase, the recovery image is from the external device. If
>>>   the recovery image has the corrupt EFI compression section, they
>>>   will be handled by EFI Decompression PPI.
>>
>> In the PEI phase, if the recovery image is crafted, it could cause a
>> buffer overflow during decompression. However, if the recovery image
>> is crafted, it might as well decompress cleanly, and once it is
>> dispatched, do "bad things". Do the decompression and the dispatch
>> occur at different privilege levels?
>>
>
> This patch focuses on the wrong decompression data that cause the
> decompression failure or hang.  The data content can be signed and
> verified.
>
>>> In DXE phase, UEFI option ROM is the third party code. If it is EFI
>>> compression option ROM, EFI decompression protocol will be used to
>>> decode its data. I don't think SMM uses EFI decompression protocol.
>>> UefiDecompressionLib is used as EFI compression PPI/Protocol. It
>>> matches PI EFI compression section instead of GUID section. So, it
>>> has no GUID extraction PPI/Protocol.
>>
>> In the DXE phase, if the option ROM is crafted, it could cause a
>> buffer overflow when it is decompressed. But, again, how is that
>> different from when a crafted oprom decompresses cleanly, and then
>> does "bad things" when it is dispatched?
>>
>> Here (in the DXE phase), I can imagine two answers myself:
>>
>> (1) Decompression occurs before Secure Boot validation, but dispatch
>> occurs only after. Therefore a crafted UEFI image could cause
>> problems via decompression even if it would fail SB verification
>> later.
>>
>> (2) Decompression of UEFI option ROMs occurs before PlatformBDS locks
>> down SMRAM and lockboxes. However, the execution of UEFI option ROMs
>> is deferred until after the lockdown.
>>
>> Do these scenarios apply? Because, if they do, I agree the issue
>> qualifies as privilege escalation.
>>
>
> Yes. Decompression happen early. After decompression, PE image will be
> verified.

Got it now. Thanks!
Laszlo