Run getsebool -a on a fresh RHEL 10 machine and you get a wall of output: a few hundred booleans, almost all of them sitting at their defaults, most of them untouched for the life of the system. That is not a failure on your part. The policy ships with switches for workloads you will never run, daemons you have never heard of, and edge cases that only matter in one datacenter somewhere. But buried in that wall are toggles that solve real problems, and a smaller set that make your machine meaningfully harder to attack. This guide walks through the ones worth knowing.
If you have not read our earlier piece on SELinux booleans in RHEL 9, start there. It covers the mechanics: what a boolean is, how setsebool and semanage differ, why the -P flag matters, and the handful of booleans everyone meets in their first month. This article assumes that grounding and goes further into the catalogue. Nearly everything from the RHEL 9 policy carries forward into RHEL 10, so your muscle memory survives the upgrade; what changes is mostly at the edges, where retired services take their booleans with them and container tooling picks up new ones.
Before touching anything, take inventory. The command semanage boolean -l gives you more than getsebool does: current state, default state, and a one-line description of what each switch controls. Pipe it through a pager and actually read it once in your life. Then run semanage boolean -l -C to see only the booleans that have been modified locally on this machine. On a well-run system that second list is short and every entry has a story. On an inherited system it is often an archaeology dig: booleans flipped years ago to silence a denial nobody diagnosed, still open, still widening the attack surface. Auditing that list is one of the cheapest security wins available to you.
Most admins only ever flip booleans in one direction, from off to on, to make a denial go away. The policy also contains booleans that work the other way, and they deserve more use than they get.
None of these matter if the machine is in permissive mode, where denials are logged but not enforced. If that sentence gives you pause, go play level 1 on enforcing mode before going further, because booleans only bite when the policy as a whole does.
Everyone knows httpd_can_network_connect. Fewer people know its narrower siblings, and the narrow version is almost always the right call.
The container and virtualization policies carry their own boolean families, and they come up constantly on RHEL 10 hosts because that is where the workloads went.
If your containers run under Podman without root, the interaction between user namespaces and labels has its own wrinkles, and our guide to rootless Podman and SELinux walks through them properly rather than via boolean guesswork.
A boolean is a pre-packaged policy decision: the policy authors anticipated your situation and left you a switch. When no switch fits, the temptation is to flip the nearest broad one and move on. Resist that. The escalation path is: look for a narrow boolean, then consider relabelling with semanage fcontext, and only then write a small custom module with audit2allow after reading what it generated. Flipping domain_can_mmap_files or selinuxuser_execstack to silence one denial trades a scoped problem for a global one.
Readers arriving from Ubuntu or SUSE sometimes ask whether AppArmor has an equivalent mechanism; the honest answer is that the two systems carve up the problem differently, and our SELinux versus AppArmor comparison covers that ground without the flame war.
Once a quarter, run semanage boolean -l -C, read every line, and ask whether the reason still exists. Booleans are reversible, documented, and visible, which makes them the most honest kind of policy change you can make. The ones you are probably not using fall into two camps: the loosening ones you are right to ignore, and the hardening ones you have no excuse left to skip.