[edk2-devel] [RFC] MemoryProtectionLib for Dynamic Memory Guard Settings

Sat Jul 31 01:38:17 UTC 2021

Hi Sean
That is excellent topic.

I agree with you that, one of most frequent asked question in the design meeting is: Should I use protocol or library? Sigh...

I added my personal thought below. I would like other people to comment, Mike Kinney, Andrew Fish, Liming Gao, who knows the history and EDKII design philosophy very well.

==========
Besides the specification difference you mentioned in previous email,

1) *Protocol* is between 2 modules in UEFI/SMM phase. It is used to pass data or abstract function.
2) *PPI* is similar to *Protocol*. But it is between 2 modules in PEI phase.

3) *HOB* is between 2 modules in different phase, to pass the data from one phase to the other. Such as PEI passing HOB to DXE, PEI passing HOB to SMM.
4) *Variable* is also between 2 modules in different phase. In addition, the data can be *NON_VOLATILE* and always be there across boot.
5) *GUID configuration table* is also between 2 modules in different phase, but it is limited from UEFI to OS. From function perspective, GUID HOB is same as GUID Config Table. The difference is that HOB is PI/OEM Firmware scope, while the GUID Config Table is UEFI/OS interface scope.

So far, above is the EDK-I design. However, two big issues existed in EDK-I.
A) The policy protocol. Everybody hated the policy protocol in EDKII, but everybody has to use it because there is no other choice. Everyone keeps increase the policy version and adds their own stuff, which make code hard to unify.
B) The library. In EDK-I, we had crazy library design - EfiDriverLib, EfiCommonLib, EfiRuntimeLib, EfiShellLib. It looked good at first glance. But when we link them together, it fails. 2 libraries may have same function name, but implement in different way. 2 libraries may have different function name, but implement same function. 

Those issues drove the two most important feature - PCD and Library design in EDK-II.

6) *PCD* is designed to abstract the OEM configuration data.
Static PCD means OEM can determine the value at build time.
Dynamic PCD means OEM need determine the value at runtime. (The implementation of PCD is still based upon PPI, Hob, and Protocol.)

7) A generic *library* is design to share common code in one place.
In addition, EDKII *library*serves another purpose: to abstract a set of common interface (as known as Library Class). That allows the OEM can replace the library implementation (as known as Library Instance) at build time without changing any code.

==========

Usually, we expose functions in a library instead of global data. (Well, there are exceptions such as gBS, gST, gRT, gSmst, which is well known pointers in history.)
I did not see too many cases to expose global data directly.

To compare UEFI protocol or C structure/class, I feel option 1 is similar to :
typedef struct {
Public:
	UINTN Data;
} CLASS_XXX.

While, option 2 is similar to :
typedef struct {
Private:
	UINTN Data;
Public:
	UINTN GetData ();
	VOID SetData (UINTN Data);
} CLASS_XXX.

I also see the similarity with https://github.com/tianocore/edk2/blob/master/MdePkg/Include/Library/DebugLib.h, such as DebugAssertEnabled(), DebugPrintEnabled(), DebugCodeEnabled(), DebugClearMemoryEnabled(), DebugPrintLevelEnabled().
That is the reason I vote option 2.

Thank you
Yao Jiewen

> -----Original Message-----
> From: devel at edk2.groups.io <devel at edk2.groups.io> On Behalf Of Sean
> Sent: Saturday, July 31, 2021 2:42 AM
> To: devel at edk2.groups.io; Yao, Jiewen <jiewen.yao at intel.com>; Taylor Beebe
> <t at taylorbeebe.com>; Wang, Jian J <jian.j.wang at intel.com>
> Cc: Dong, Eric <eric.dong at intel.com>; Ni, Ray <ray.ni at intel.com>; Kumar,
> Rahul1 <rahul1.kumar at intel.com>; mikuback at linux.microsoft.com; Wu, Hao A
> <hao.a.wu at intel.com>; Bi, Dandan <dandan.bi at intel.com>;
> gaoliming at byosoft.com.cn; Dong, Guo <guo.dong at intel.com>; Ma, Maurice
> <maurice.ma at intel.com>; You, Benjamin <benjamin.you at intel.com>
> Subject: Re: [edk2-devel] [RFC] MemoryProtectionLib for Dynamic Memory
> Guard Settings
> 
> Jiewen,
> 
> **Slight rant**
> 
> I agree with libraries as an effective abstraction method.  But I think
> there needs to be a broad discussion about the order of preference for
> methods of abstraction.  Today the edk2 code base is a mix and often
> there are numerous methods abstracting the same thing which leads to
> confusion, misconfiguration, and error.
> 
> In the UEFI specification we have PPIs/Protocols/Events for functional
> abstraction.  We have variables, guided config tables, and HII for data
> abstraction.
> 
> In the PI specification we add HOBs and PCDs for data abstractions.
> 
> Finally, in EDKII we add the library class concept and leverage it
> heavily for arch, phase, and platform/behavioral abstractions.
> 
> Without clear guidance for how and when to use the above it is hard to
> keep code being developed by the larger community consistent.
> 
> **End**
> 
> I was leaning towards something closer to
> 
>  >> Option 1:
> https://github.com/TaylorBeebe/edk2/tree/memory_protection_lib_2
> 
> the HOB method and internally as we develop more code we are preferring
> HOB and data abstractions more than functional abstraction.  Data
> abstractions can be used to control functional differences as well if
> needed.  Data abstractions allow for easier validation and support
> diverse code environments.  For example standalone MM and
> payloadpkg/payload concepts.  Finally, data abstractions break the need
> for a monolithic code base.   But as you can see in option 1 it actually
> uses a library class abstraction as well because no one wants to write
> the same code over and over again to get the HOB.  The contract of the
> library is just data but it still requires library mappings.  Maybe
> these types of libraries need to be treated differently.
> 
> Anyway it would be great to hear from other members of the community
> around not just the memory protections RFC (this RFC) but around
> preferences for abstraction techniques (pro/con).  If an actual
> discussion starts it could move to design meeting.
> 
> Thanks
> Sean
> 
> 
> 
> 
> 
> 
> 
> On 7/29/2021 7:34 PM, Yao, Jiewen wrote:
> > Thanks. Code talks better.
> >
> > I prefer option 2, which is a generic way for abstraction.
> >
> > And you may enable option 1 under the cover of option 2, just create a lib
> instance to get config from Hob.
> >
> > Thank you
> > Yao Jiewen
> >
> >> -----Original Message-----
> >> From: Taylor Beebe <t at taylorbeebe.com>
> >> Sent: Friday, July 30, 2021 10:07 AM
> >> To: Yao, Jiewen <jiewen.yao at intel.com>; Wang, Jian J
> <jian.j.wang at intel.com>;
> >> devel at edk2.groups.io
> >> Cc: spbrogan at outlook.com; Dong, Eric <eric.dong at intel.com>; Ni, Ray
> >> <ray.ni at intel.com>; Kumar, Rahul1 <rahul1.kumar at intel.com>;
> >> mikuback at linux.microsoft.com; Wu, Hao A <hao.a.wu at intel.com>; Bi,
> Dandan
> >> <dandan.bi at intel.com>; gaoliming at byosoft.com.cn; Dong, Guo
> >> <guo.dong at intel.com>; Ma, Maurice <maurice.ma at intel.com>; You,
> Benjamin
> >> <benjamin.you at intel.com>
> >> Subject: Re: [RFC] MemoryProtectionLib for Dynamic Memory Guard Settings
> >>
> >> Of course - here are a couple of rough drafts:
> >>
> >> Option 1:
> https://github.com/TaylorBeebe/edk2/tree/memory_protection_lib_2
> >> Option 2:
> https://github.com/TaylorBeebe/edk2/tree/memory_protection_lib
> >>
> >> On 7/29/2021 6:57 PM, Yao, Jiewen wrote:
> >>> Hi
> >>> Sorry, I am not able to follow the discussion.
> >>>
> >>> Is there any sample or POC code to show the concept?
> >>>
> >>>> -----Original Message-----
> >>>> From: Taylor Beebe <t at taylorbeebe.com>
> >>>> Sent: Friday, July 30, 2021 9:55 AM
> >>>> To: Wang, Jian J <jian.j.wang at intel.com>; devel at edk2.groups.io
> >>>> Cc: spbrogan at outlook.com; Dong, Eric <eric.dong at intel.com>; Ni, Ray
> >>>> <ray.ni at intel.com>; Kumar, Rahul1 <rahul1.kumar at intel.com>;
> >>>> mikuback at linux.microsoft.com; Wu, Hao A <hao.a.wu at intel.com>; Bi,
> >> Dandan
> >>>> <dandan.bi at intel.com>; gaoliming at byosoft.com.cn; Dong, Guo
> >>>> <guo.dong at intel.com>; Ma, Maurice <maurice.ma at intel.com>; You,
> >> Benjamin
> >>>> <benjamin.you at intel.com>; Yao, Jiewen <jiewen.yao at intel.com>
> >>>> Subject: Re: [RFC] MemoryProtectionLib for Dynamic Memory Guard
> Settings
> >>>>
> >>>> Thanks for your feedback, Jian.
> >>>>
> >>>> In option 2, a most basic implementation would returning the current
> >>>> FixedAtBuild PCDs assuming they are kept. If they aren't, the library
> >>>> implementer could simply hard-code the return value for each memory
> >>>> protection setting.
> >>>>
> >>>> In option 1, the HOB would be published in pre-mem and I'm not an expert
> >>>> on exploiting the pre-mem environment. Jiewen may have more to say on
> >> this.
> >>>>
> >>>> -Taylor
> >>>>
> >>>> On 7/28/2021 7:18 PM, Wang, Jian J wrote:
> >>>>> Thanks for the RFC. I'm not object to this idea. The only concern from me
> >>>>> is the potential security holes introduced by the changes. According to
> your
> >>>>> description, it allows 3rd party software to violate memory protection
> >> policy.
> >>>>> I'd like to see more explanations on how to avoid it to be exploited.
> >>>>>
> >>>>> +Jiewen, what's current process to evaluate the security threat?
> >>>>>
> >>>>> Regards,
> >>>>> Jian
> >>>>>
> >>>>>> -----Original Message-----
> >>>>>> From: Taylor Beebe <t at taylorbeebe.com>
> >>>>>> Sent: Friday, July 23, 2021 8:33 AM
> >>>>>> To: devel at edk2.groups.io
> >>>>>> Cc: spbrogan at outlook.com; Dong, Eric <eric.dong at intel.com>; Ni, Ray
> >>>>>> <ray.ni at intel.com>; Kumar, Rahul1 <Rahul1.Kumar at intel.com>;
> >>>>>> mikuback at linux.microsoft.com; Wang, Jian J <jian.j.wang at intel.com>;
> >> Wu,
> >>>>>> Hao A <hao.a.wu at intel.com>; Bi, Dandan <dandan.bi at intel.com>;
> >>>>>> gaoliming at byosoft.com.cn; Dong, Guo <guo.dong at intel.com>; Ma,
> >>>> Maurice
> >>>>>> <maurice.ma at intel.com>; You, Benjamin <benjamin.you at intel.com>
> >>>>>> Subject: [RFC] MemoryProtectionLib for Dynamic Memory Guard
> Settings
> >>>>>>
> >>>>>> Current memory protection settings rely on FixedAtBuild PCD values
> >>>>>> (minus PcdSetNxForStack). Because of this, the memory protection
> >>>>>> configuration interface is fixed in nature. Cases arise in which memory
> >>>>>> protections might need to be adjusted between boots (if platform design
> >>>>>> allows) to avoid disabling a system. For example, platforms might
> choose
> >>>>>> to allow the user to control their protection policies such as allow
> >>>>>> execution of critical 3rd party software that might violate memory
> >>>>>> protections.
> >>>>>>
> >>>>>> This RFC seeks your feedback regarding introducing an interface that
> >>>>>> allows dynamic configuration of memory protection settings.
> >>>>>>
> >>>>>> I would like to propose two options:
> >>>>>> 1. Describing the memory protection setting configuration in a HOB that
> >>>>>> is produced by the platform.
> >>>>>> 2. Introducing a library class (e.g. MemoryProtectionLib) that allows
> >>>>>> abstraction of the memory protection setting configuration data source.
> >>>>>>
> >>>>>> In addition, I would like to know if the memory protection FixedAtBuild
> >>>>>> PCDs currently in MdeModulePkg can be removed so we can move the
> >>>>>> configuration interface entirely to an option above.
> >>>>>>
> >>>>>> In any case, I would like the settings to be visible to environments
> >>>>>> such as Standalone MM where dynamic PCDs are not accessible.
> >>>>>>
> >>>>>> I am seeking your feedback on this proposal in preparation for sending
> >>>>>> an edk2 patch series.
> >>>>>>
> >>>>>> --
> >>>>>> Taylor Beebe
> >>>>>> Software Engineer @ Microsoft
> >>>>
> >>>> --
> >>>> Taylor Beebe
> >>>> Software Engineer @ Microsoft
> >>
> >> --
> >> Taylor Beebe
> >> Software Engineer @ Microsoft
> >
> >
> >
> >
> >
> 
> 
> 
> 

-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#78462): https://edk2.groups.io/g/devel/message/78462
Mute This Topic: https://groups.io/mt/84392478/1813853
Group Owner: devel+owner at edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [edk2-devel-archive at redhat.com]
-=-=-=-=-=-=-=-=-=-=-=-