Re: [PATCH] Kernel Lockdown: Add an option to allow raw MSR access even, in confidentiality mode.

From: Matt Parnell <mparnell@gmail.com>
To: Kees Cook <keescook@chromium.org>
Cc: linux-security-module@vger.kernel.org, dhowells@redhat.com,
	matthew.garrett@nebula.com
Subject: Re: [PATCH] Kernel Lockdown: Add an option to allow raw MSR access even, in confidentiality mode.
Date: Mon, 2 Dec 2019 12:29:05 -0600	[thread overview]
Message-ID: <96625b06-3bba-6831-7127-22b690fccf1b@gmail.com> (raw)
In-Reply-To: <f988a531-a7f8-cef6-d3b4-6fbd89f5351f@gmail.com>

After doing some research it appears that for Intel chips, only a single
register needs to be writeable. I'm not sure about AMD etc.

intel-undervolt/blob/master/config.h:

    #define MSR_ADDR_TEMPERATURE 0x1a2
    #define MSR_ADDR_UNITS 0x606
    #define MSR_ADDR_VOLTAGE 0x150

Perhaps add an MSR whitelist to allow writing, if
LOCK_DOWN_KERNEL_FORCE_CONFIDENTIALITY=Y and
CONFIG_SECURITY_LOCKDOWN_LSM_EARLY=Y?

CONFIG_SECURITY_LOCKDOWN_LSM_EARLY is likely what prevents Apparmor or
some other LSM policy manager allow this behavior...

as an option at build time would be more sensible?

On 12/1/19 2:53 PM, Matt Parnell wrote:
> That is, I was intending to use lockdown from boot, which isn't
> changeable after the fact if I'm not mistaken. How possible is granular
> control of what is and is not locked down?
>
> On 11/30/19 1:09 PM, Matt Parnell wrote:
>> I can see how using a policy would be beneficial; I only did this
>> because as I understood it, policy wouldn't be able to change these
>> particular settings since anything attempting to do so would be from
>> userspace.
>>
>> On 11/30/19 12:36 PM, Kees Cook wrote:
>>> On Sat, Nov 30, 2019 at 12:49:48AM -0600, Matt Parnell wrote:
>>>> From 452b8460e464422d268659a8abb93353a182f8c8 Mon Sep 17 00:00:00 2001
>>>> From: Matt Parnell <mparnell@gmail.com>
>>>> Date: Sat, 30 Nov 2019 00:44:09 -0600
>>>> Subject: [PATCH] Kernel Lockdown: Add an option to allow raw MSR access even
>>>>  in confidentiality mode.
>>>>
>>>> For Intel CPUs, some of the MDS mitigations utilize the new "flush" MSR, and
>>>> while this isn't something normally used in userspace, it does cause false
>>>> positives for the "Forshadow" vulnerability.
>>>>
>>>> Additionally, Intel CPUs use MSRs for voltage and frequency controls,
>>>> which in
>>>> many cases is useful for undervolting to avoid excess heat.
>>>>
>>>> Signed-off-by: Matt Parnell <mparnell@gmail.com>
>>> I would expect this to just be implemented via LSM policy, not ifdefs
>>> and Kconfig?
>>>
>>> -Kees
>>>
>>>> ---
>>>>  arch/x86/kernel/msr.c     |  5 ++++-
>>>>  security/lockdown/Kconfig | 12 ++++++++++++
>>>>  2 files changed, 16 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/arch/x86/kernel/msr.c b/arch/x86/kernel/msr.c
>>>> index 1547be359d7f..4adce59455c3 100644
>>>> --- a/arch/x86/kernel/msr.c
>>>> +++ b/arch/x86/kernel/msr.c
>>>> @@ -80,10 +80,11 @@ static ssize_t msr_write(struct file *file, const
>>>> char __user *buf,
>>>>      int err = 0;
>>>>      ssize_t bytes = 0;
>>>>  
>>>> +#if defined(LOCK_DOWN_DENY_RAW_MSR)
>>>>      err = security_locked_down(LOCKDOWN_MSR);
>>>>      if (err)
>>>>          return err;
>>>> -
>>>> +#endif
>>>>      if (count % 8)
>>>>          return -EINVAL;    /* Invalid chunk size */
>>>>  
>>>> @@ -135,9 +136,11 @@ static long msr_ioctl(struct file *file, unsigned
>>>> int ioc, unsigned long arg)
>>>>              err = -EFAULT;
>>>>              break;
>>>>          }
>>>> +#if defined(LOCK_DOWN_DENY_RAW_MSR)
>>>>          err = security_locked_down(LOCKDOWN_MSR);
>>>>          if (err)
>>>>              break;
>>>> +#endif
>>>>          err = wrmsr_safe_regs_on_cpu(cpu, regs);
>>>>          if (err)
>>>>              break;
>>>> diff --git a/security/lockdown/Kconfig b/security/lockdown/Kconfig
>>>> index e84ddf484010..f4fe72c4bf8f 100644
>>>> --- a/security/lockdown/Kconfig
>>>> +++ b/security/lockdown/Kconfig
>>>> @@ -44,4 +44,16 @@ config LOCK_DOWN_KERNEL_FORCE_CONFIDENTIALITY
>>>>       code to read confidential material held inside the kernel are
>>>>       disabled.
>>>>  
>>>> +config LOCK_DOWN_DENY_RAW_MSR
>>>> +    bool "Lock down and deny raw MSR access"
>>>> +    depends on LOCK_DOWN_KERNEL_FORCE_CONFIDENTIALITY
>>>> +    default y
>>>> +    help
>>>> +      Some Intel based systems require raw MSR access to use the flush
>>>> +      MSR for MDS mitigation confirmation. Raw access can also be used
>>>> +      to undervolt many Intel CPUs.
>>>> +
>>>> +      Say Y to prevent access or N to allow raw MSR access for such
>>>> +      cases.
>>>> +
>>>>  endchoice
>>>> -- 
>>>> 2.24.0
>>>>
>>>>
>>>