Describe some easy and more difficult methods to harden SELinux
policies.
Signed-off-by: Topi Miettinen <toiwoton@xxxxxxxxx>
---
src/hardening.md | 127 +++++++++++++++++++++++++++++++
src/implementing_seaware_apps.md | 2 +-
src/reference_policy.md | 2 +-
src/toc.md | 1 +
4 files changed, 130 insertions(+), 2 deletions(-)
create mode 100644 src/hardening.md
diff --git a/src/hardening.md b/src/hardening.md
new file mode 100644
index 0000000..fa39776
--- /dev/null
+++ b/src/hardening.md
@@ -0,0 +1,127 @@
+# Hardening SELinux
+
+- [Reference Policy](#reference-policy)
+- [Custom Policy](#custom-policy)
+
+The Reference Policy sets a good basis for learning how to operate a
+system with SELinux. The policy allows the system administrators and
+users to continue using the working methods learned from time before
+familiarizing with SELinux, because of its "targeted" model and
+reasonable defaults.
+
+However, it's possible to tighten the policy considerably. The
+Reference Policy gives several options for hardening but for maximum
+control over the policy, it's possible to write custom modules or even
+replace the Reference Policy entirely.
+
+## Reference Policy
+
+The hardening suggestions are listed in the rough order of difficulty.
+
+### Booleans
+
+The Reference Policy uses booleans to control optional aspects of the
+policy. Changing the booleans is very easy way to tune the SELinux
+policy. The booleans can be also changed temporarily, without
+committing them yet to the on-disk policy, so changes are
+automatically reverted on next boot.
+
+For example, recent Firefox browsers can work with policy boolean
+`mozilla_execstack` set to `off`. This can reduce the surface to
+vulnerabilities which could use an executable stack.
+
+### Disabling Modules
+
+By default, the Reference Policy enables most policy modules to
+support any system. But when the system is known well by the
+administrator, it's possible to remove modules which aren't used.
+
+As an example, if Firefox isn't installed, module `mozilla` can be
+removed from the policy. The hardening effect comes from reducing the
+allow rules, some of which may apply to paths which aren't actively
+monitored because the corresponding application isn't installed.
+
+Care should be taken to never remove modules which actually are used,
+since this may weaken the policy instead, because the removed module
+could have contained an application more strictly than what the policy
+allows without the module.
+
+### Users and Roles
+
+In the "targeted" model, both unprivileged users and the system
+administrator (`root`) are unconfined and the TE rules are very
+relaxed. However, it's possible to change the SELinux user for these
+accounts to more confined variants.
+
+For the unprivileged users, the confined user in the Reference Policy
+is `user_u` with corresponding role `user_r`. There's also `staff_r`
+to allow transitioning to system administative roles.
+
+For the system administrator there are several options: `root` SELinux
+user, which is mostly unconfied and `sysadm_u`, which is more
+confined. The role for both is `sysadm_r`.
+
+It's also possible to divide the powers of the system administrator to
+several roles, `sysadm_r`, `auditadm_r`, `logadm_r`, `dbadm_r` and
+`secadm_r`. This can be useful when an organization wants to ensure
+that even the system administrators can be held accountable for their
+actions. The roles can be also useful for management of labor, for
+example having dedicated persons for managing databases or SELinux
+policies.
+
+It should be noted that since this isn't the default way of operating
+SELinux, the Reference Policy may need to supplemented and the
+administrators, even users, may need to be more aware of SELinux in
+order to be able to operate the system.
+
+### Network
+
+With network controls of SELinux, it's possible to enhance firewall
+rules with knowledge of SELinux types. Traditional firewall rules only
+affect the whole system by allowing certain ports and protocols but
+blocking others. With `nftables` and `iptables` it's also possible to
+make this more fine grained: certain users can access network but
+others may not. By using also SELinux controls it's possible to fine
+tune this to application level: `mozilla_t` can connect to the
+Internet but some other applications can't. SELinux packet controls
+can be also used to combine Deep Packet Inspection with SELinux TE
+rules.
+
+In an internal network of an organization, where all entities can
+agree on the same SELinux policy, using IPSec, CIPSO and CALIPSO may
+allow further policy controls. In addition to SELinux domain of the
+source application, even the SELinux domain (or at least MCS/MLS
+category/sensitivity) of the target server can be used in TE rules.
+
+## Custom Policy
+
+Further hardening can be achieved by replacing policy modules from the
+Reference Policy with custom modules. Typically the modules in the
+Reference Policy are written to allow all possible modes of operation
+for an application or its users, since the writers of the policy don't
+know the specifics of each installation. Thus the SELinux rules may be
+more relaxed than what could be optimal for a specific case. When the
+exact environment and usage patterns are known, it's possible to write
+replacement policy modules to remove excess rules and hence reduce
+attack surface.
+
+As a minimum, it should be ensured that all continously running
+services and main user applications have a dedicated policy module or
+rules, instead of running in for example `init_t`, `initrc_t` or
+`unconfined_t` types which may offer low level of protection.
+
+It's also possible to write custom SELinux policies for an entire
+system with non-trivial effort.
+
+The rules can also be analyzed with various SELinux tools, such as
+`apol`, `sedta`, `sepolicy` and many more. With the tools it may be
+possible to find hardening opportunities or errors in the policy.
+
+The downside of making the SELinux rules as tight as possible is that
+when the applications (or hardware components or network
+configuration) are updated, there's a possibility that the rules may
+also need updating because of the changes. Less generic rules are also
+less generally useful for different configurations, so the rules may
+need tuning for each installation.
+
+**[[ PREV ]](reference_policy.md)** **[[ TOP ]](#)** **[[ NEXT ]](implementing_seaware_apps.md)**
diff --git a/src/implementing_seaware_apps.md b/src/implementing_seaware_apps.md
index 74833f8..2351421 100644
--- a/src/implementing_seaware_apps.md
+++ b/src/implementing_seaware_apps.md
@@ -297,4 +297,4 @@ applied to their objects as defined by policy.
<!-- %CUTHERE% -->
---
-**[[ PREV ]](reference_policy.md)** **[[ TOP ]](#)** **[[ NEXT ]](embedded_systems.md)**
+**[[ PREV ]](hardening.md)** **[[ TOP ]](#)** **[[ NEXT ]](embedded_systems.md)**
diff --git a/src/reference_policy.md b/src/reference_policy.md
index 45e4145..ebb516f 100644
--- a/src/reference_policy.md
+++ b/src/reference_policy.md
@@ -3060,4 +3060,4 @@ section.
<!-- %CUTHERE% -->
---
-**[[ PREV ]](modular_policy_statements.md)** **[[ TOP ]](#)** **[[ NEXT ]](implementing_seaware_apps.md)**
+**[[ PREV ]](modular_policy_statements.md)** **[[ TOP ]](#)** **[[ NEXT ]](hardening.md)**
diff --git a/src/toc.md b/src/toc.md
index f5778bd..20a87f0 100644
--- a/src/toc.md
+++ b/src/toc.md
@@ -53,6 +53,7 @@
- [XEN Statements](xen_statements.md#xen-statements)
- [Modular Policy Support Statements](modular_policy_statements.md#modular-policy-support-statements)
- [The Reference Policy](reference_policy.md#the-reference-policy)
+- [Hardening SELinux](hardening.md#hardening-selinux)
- [Implementing SELinux-aware Applications](implementing_seaware_apps.md#implementing-selinux-aware-applications)
- [Embedded Systems](embedded_systems.md#embedded-systems)
- [SE for Android](seandroid.md#security-enhancements-for-android)
--
2.33.0
