Re: [RFC PATCH] /dev/mem: Disable /dev/mem under TDX guest

[Nikolay Borisov <nik.borisov@suse.com>]

On 18.03.25 г. 14:23 ч., Kirill A. Shutemov wrote:
> On Tue, Mar 18, 2025 at 01:36:04PM +0200, Nikolay Borisov wrote:
>> If a piece of memory is read from /dev/mem that falls outside of the
>> System Ram region i.e bios data region the kernel creates a shared
>> mapping via xlate_dev_mem_ptr() (this behavior was introduced by
>> 9aa6ea69852c ("x86/tdx: Make pages shared in ioremap()"). This results
>> in a region having both a shared and a private mapping.
>>
>> Subsequent accesses to this region via the private mapping induce a
>> SEPT violation and a crash of the VMM.
> 
> Crash of VMM or TD termination? If VMM crashes in this case, it has to be
> fixed.

Went back through the bug reports and it seems this causes a SEPT 
violation inside the guest, which crashes, and is then re-created by 
GCP. So it would seem this causes an SEPT violation, rather than a VMM 
crash, my bad for mixing up the symptoms.

> 
>> In this particular case the
>> scenario was a userspace process reading something from the bios data
>> area at address 0x497 which creates a shared mapping, and a followup
>> reboot accessing __va(0x472) which access pfn 0 via the private mapping
>> causing mayhem.
> 
> I think it should lead to unrecoverable EPT-violation, but not VMM crash.

<nod> You are correct.

> 
>> Fix this by simply forbidding access to /dev/mem when running as an TDX
>> guest.
> 
> I think we need to think wider. What about applying a subset of LOCKDOWN_*
> in all coco guests by default. Many of them are relevant for the guest security.

How do you envision this to work, by introducing another 
CONFIG_LOCK_DOWN_KERNEL_FORCE_COCO or some such ? Will it be opt-in or 
mandatory?

Should we decide to follow the lockdown route this means the owner of 
the coco guest will have the ability to disable it and a misbehaving 
userspace process will still be able to induce an EPT violation.


>