Re: [RFC PATCH] selinux: add SELinux hooks for lockdown integrity and confidentiality

[Stephen Smalley <sds@tycho.nsa.gov>]

On 10/31/19 5:59 AM, Paul Moore wrote:
> On Wed, Oct 30, 2019 at 9:16 AM Stephen Smalley <sds@tycho.nsa.gov> wrote:
>> Add SELinux access control hooks for lockdown integrity and
>> confidentiality. This effectively mimics the current implementation of
>> lockdown (caveat noted below). If lockdown is enabled alongside SELinux,
>> then the lockdown access control will take precedence over the SELinux
>> lockdown implementation.
>>
>> Note that this SELinux implementation allows the integrity and
>> confidentiality reasons to be controlled independently from one another.
>> Thus, in an SELinux policy, one could allow integrity operations while
>> blocking confidentiality operations.
>>
>> (original patch authored by an intern who wishes to remain anonymous;
>> I am signing off on his behalf)
>>
>> Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
>> ---
>>   security/selinux/hooks.c            | 22 ++++++++++++++++++++++
>>   security/selinux/include/classmap.h |  2 ++
>>   2 files changed, 24 insertions(+)
>>
>> diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c
>> index 36e531b91df2..6722c6b4ae74 100644
>> --- a/security/selinux/hooks.c
>> +++ b/security/selinux/hooks.c
>> @@ -91,6 +91,7 @@
>>   #include <uapi/linux/mount.h>
>>   #include <linux/fsnotify.h>
>>   #include <linux/fanotify.h>
>> +#include <linux/security.h>
>>
>>   #include "avc.h"
>>   #include "objsec.h"
>> @@ -6799,6 +6800,25 @@ static void selinux_bpf_prog_free(struct bpf_prog_aux *aux)
>>   }
>>   #endif
>>
>> +static int selinux_lockdown(enum lockdown_reason what)
>> +{
>> +       u32 sid = current_sid();
>> +
>> +       if (what <= LOCKDOWN_INTEGRITY_MAX)
>> +               return avc_has_perm(&selinux_state,
>> +                               sid, sid,
>> +                               SECCLASS_LOCKDOWN, LOCKDOWN__INTEGRITY, NULL);
>> +       else if (what <= LOCKDOWN_CONFIDENTIALITY_MAX)
>> +               return avc_has_perm(&selinux_state,
>> +                               sid, sid,
>> +                               SECCLASS_LOCKDOWN, LOCKDOWN__CONFIDENTIALITY,
>> +                               NULL);
>> +
>> +       /* invalid reason */
>> +       pr_warn("SELinux: invalid lockdown reason\n");
>> +       return -EPERM;
>> +}
> 
> I don't have any objections to adding a hook to control access to the
> lockdown functionality (I think it's a good idea), but I am a little
> nervous about the granularity of the control.  Sticking with just an
> integrity and a confidentiality permission seems okay, but I worry
> about adding additional permissions until we have a better idea of how
> the lockdown functionality is adopted by developers and we see how the
> lockdown_reason evolves.

Ok, so let's discuss what if anything else is needed for a final non-RFC 
version of this patch.

One thing that I wondered about was whether we ought to include the 
reason information in the audit record as supplemental data via a new 
LSM_AUDIT_DATA_* type for help in policy debugging / development. 
However, the lockdown_reasons[] string array is presently private to the 
lockdown module so we would have to export that or replicate it for 
creating a string representation of the reason if we were to do so. 
That would expose the reasons in terms of audit data but not as a basis 
for the permission check. Note that the lockdown module logs these 
reason strings via pr_notice() when it denies access, so it appears that 
exposing the strings as part of audit data would not introduce any extra 
kernel stable ABI guarantees?

I also wasn't sure about the pr_warn() above.  If we reach it, it is 
effectively a kernel bug. We could mirror what the lockdown module does 
in lockdown_is_locked_down(), i.e. use WARN() instead.  Of course, the 
SELinux hook won't even be reached in this case if the lockdown module 
is enabled, but the lockdown module could be disabled so I guess we need 
to check it too.

If we take the lockdown class with just integrity and confidentiality 
permissions now and later introduce finer granularity, we'll presumably 
need a policy capability to select whether the coarse-grained or 
fine-grained permissions are used.