core: use `landlock` crate by aeryz · Pull Request #305 · multikernel/sandlock

aeryz · 2026-10-20T00:03:41Z

Instead of writing our own landlock wrapper, we now use the landlock crate. This will ease up handling of the landlock ops on multiple kernel versions.

Fixes #286

congwang-mk · 2026-10-20T02:29:06Z

-        ))
-    })?;
+    let status = ruleset.restrict_self()?;
+    if status.ruleset != RulesetStatus::FullyEnforced {


This check is load-bearing, not a sanity check: any try_compat failure under HardRequirement makes the crate state Dummy permanently, after which add_rule and restrict_self return Ok without syscalls (NotEnforced). The ignored let _ = add_rule(...) calls rely on this guard to fail closed. Please add a comment saying so, ideally a test too.

Good catch! But tbf, API-wise, it's a bit nasty to first return an error, then return Ok for the consequent calls. Let's still have this as a sanity-check but I'd say let's not ignore the errors on add_rule. Right now, we already check whether the path exists - and successfully opened - by checking the result of path_beneath_rule. After that, we should not ignore the result of add_rule since it'd be catched in the FullyEnforced check anyways.

I left this as is impl-wise. Added a comment on the behaviour and a test that covers it.

congwang-mk · 2026-10-20T02:29:06Z

-    } else {
-        LANDLOCK_ACCESS_NET_BIND_TCP | LANDLOCK_ACCESS_NET_CONNECT_TCP
-    };
+    let mut mask = AccessNet::from_all(abi);


from_all(abi) silently widens with crate bumps. If a future AccessNet gains e.g. UDP, it becomes handled with no rules, denying all UDP. Prefer explicit BindTcp | ConnectTcp, or add exact-mask tests at ABI::V9 (same for AccessFs::from_all in compute_fs_mask).

Done. Both places are now explicit.

congwang-mk · 2026-10-20T02:29:06Z

    /// Minimum Landlock ABI version the host kernel must support for
    /// this protection to be available.
-    pub const fn min_abi(self) -> u32 {
+    pub const fn min_abi(self) -> landlock::ABI {


Exposing landlock::ABI (here, ProtectionStatus::resolve, ProtectionUnavailable fields, compute_*_mask return types) makes every landlock crate bump a sandlock-core breaking change, and the CLI/FFI/tests now cast back to u32. Suggest keeping u32 publicly and converting inside confine_inner/active_protections.

Done, now only converting to ABI when we have to to use the landlock crate.

congwang-mk · 2026-10-20T02:29:06Z

-            SandlockError::Runtime(crate::error::SandboxRuntimeError::Confinement(e))
-        })?;
+
+        ruleset = ruleset.add_rule(


Kernel-side add_rule failures used to report add path rule for {path} / add net rule for port {port}; now the bare RulesetError loses the path/port. Please re-wrap with that context (also the net rules below).

Done by adding the add_net_rule and add_path_beneath_rule helpers.

congwang-mk · 2026-10-20T02:29:06Z

-            ConfinementError::Landlock(format!("restrict_self: {}", e)),
-        ))
-    })?;
+    let status = ruleset.restrict_self()?;


Nit: the crate sets PR_SET_NO_NEW_PRIVS by default here, but every caller already does. .no_new_privs(false) would keep that responsibility in one place.

Done, now using no_new_privs(false) and explicitly mentioning it in the public function warning the callers.

congwang-mk · 2026-10-20T02:29:06Z

            if supervised && shadowed_files().any(|f| f.starts_with(path)) {
-                add_read_rules_around_shadowed(&ruleset_fd, path).map_err(|e| {
+                ruleset = add_read_rules_around_shadowed(ruleset, path).map_err(|e| {
                    SandlockError::Runtime(crate::error::SandboxRuntimeError::Confinement(e))


Three error-conversion paths now: From<RulesetError> for SandlockError, cast_confinement, and this inline closure. Pick one.

Fixed by only using From<ConfinementError> for SandlockError. It helps with not manually wrapping SandlockError::Runtime all the time.

Instead of writing our own landlock wrapper, we now use `landlock` crate. This will ease up handling of the landlock ops on multiple kernel versions. Signed-off-by: kernelnomad <kernelnomad@proton.me>

congwang-mk reviewed Oct 20, 2026

View reviewed changes

aeryz force-pushed the 286-use-landlock-crate branch 2 times, most recently from 7fb1aad to 7194982 Compare October 25, 2026 11:11

core: use landlock crate

643cb09

Instead of writing our own landlock wrapper, we now use `landlock` crate. This will ease up handling of the landlock ops on multiple kernel versions. Signed-off-by: kernelnomad <kernelnomad@proton.me>

aeryz force-pushed the 286-use-landlock-crate branch from 7194982 to 643cb09 Compare October 25, 2026 11:18

congwang-mk merged commit fc92a56 into multikernel:main Oct 25, 2026
24 checks passed

congwang-mk mentioned this pull request Oct 25, 2026

core: tidy the landlock crate conversion #326

Merged

Conversation

aeryz commented Oct 20, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

aeryz Oct 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

aeryz Oct 25, 2026 •

edited

Loading