kernel/git/next/linux-next.git/security/landlock, branch master

landlock: Clean up ruleset validation checks

2026-09-07T10:38:04+00:00

landlock_merge_ruleset() checks for a NULL ruleset after dereferencing it in lockdep_assert_held(). Move the assertion after the check so the defensive path remains effective. The mask-validation comment originated in landlock_add_fs_access_mask() to explain that its WARN_ON_ONCE() checked a caller invariant. It became self-referential when this helper and its network and scope counterparts were inlined into landlock_create_ruleset(). Restate the invariant without naming the caller. Keep both as defensive callee checks. Moving the assertion preserves the NULL check's ability to warn and return -EINVAL, while invalid masks remain warned about and masked. Reported-by: Günther Noack Closes: https://patch.msgid.link/aobYhIt3vcs2xN0b@google.com Closes: https://patch.msgid.link/aobasxUDQ8b7GYXl@google.com Link: https://patch.msgid.link/20260907103609.113325-1-mic@digikod.net Signed-off-by: Mickaël Salaün

landlock: Fix use-after-free of the source's parent directory

2026-09-07T10:01:47+00:00

current_check_refer_path() reads old_dentry->d_parent without holding a reference nor a lock on it, and then dereferences it in collect_domain_accesses() and in the audit record. A reference on a child does not pin its parent: __d_move() reassigns dentry->d_parent and drops the reference the child held on its former parent. hook_path_rename() is not affected because the rename path calls lock_rename() before the hook, so the source cannot be reparented under it. hook_path_link() has no such protection: filename_linkat() holds a reference on the source dentry but neither locks nor references its parent, so a concurrent rename(2) can reparent the source while security_path_link() runs, and the former parent can then be removed and freed while the hook walks it. A process can trigger this after entering a Landlock domain that handles at least one filesystem access right. The process can then race a linkat(2) loop against rename(2) and rmdir(2): BUG: KASAN: slab-use-after-free in collect_domain_accesses+0x278/0x290 Read of size 4 at addr ffff888160bd53f4 by task llrepro2/549 collect_domain_accesses+0x278/0x290 current_check_refer_path+0x952/0x1120 security_path_link+0x1be/0x320 filename_linkat+0x342/0x6d0 __x64_sys_linkat+0xfa/0x150 Freed by task 562: kmem_cache_free+0x139/0x4c0 i_callback+0x4b/0x80 rcu_core+0x7dc/0x10a0 Take a reference on the dentry selected as the source parent, using dget() for the common-mount-root case and dget_parent() otherwise. Release it after the hierarchy walk and synchronous audit logging. Cc: stable@vger.kernel.org Fixes: b91c3e4ea756 ("landlock: Add support for file reparenting with LANDLOCK_ACCESS_FS_REFER") Signed-off-by: Norbert Szetei Reviewed-by: Günther Noack Tested-by: Günther Noack Link: https://patch.msgid.link/E9CDD9E6-E960-4DE2-B1AC-5667D52ABB3E@doyensec.com [mic: Clarify the caller, reachability, and reference handling] Signed-off-by: Mickaël Salaün

landlock: Add tracepoints for ptrace and scope denials

2026-08-17T08:17:16+00:00

Scope and ptrace denials follow a different code path (a domain hierarchy check) than access-right denials, so they need dedicated tracepoints with type-specific TP_PROTO arguments. Complete the denial coverage with: - landlock_deny_ptrace: ptrace access denied by a domain hierarchy mismatch. - landlock_deny_scope_signal: signal delivery denied by LANDLOCK_SCOPE_SIGNAL. - landlock_deny_scope_abstract_unix_socket: abstract unix socket access denied by LANDLOCK_SCOPE_ABSTRACT_UNIX_SOCKET. TP_PROTO passes the raw kernel object (struct task_struct or struct sock) for eBPF BTF access; the comm and sun_path string fields use __print_untrusted_str() because they hold untrusted input. Unlike the deny_access events, these omit the blockers field: each maps to exactly one denial type named by the event, so the bitmask would always be zero. Like the deny_access events they carry same_exec and logged. Audit logs the task-targeted denials with generic field names (opid, ocomm), but a strongly typed trace event can use role-prefixed names (tracee_pid/tracee_comm, target_pid/target_comm) that match the mainline task-name convention (sched_process_fork's parent_comm/child_comm) and say whose name each field holds; a bare comm= would collide across events. The abstract-unix-socket event reports peer_pid instead, a tracepoint-only field with no audit counterpart. A scope or ptrace verdict compares the subject domain against the other party's domain, so each event also reports that other party's Landlock domain (tracee_domain=, target_domain=, or peer_domain=); the subject domain= alone does not let a consumer redo domain_is_scoped() or domain_ptrace(). It is reported as a scalar ID rather than a domain pointer: a domain object is immutable, but the other task can replace its credential and free the domain that credential referenced, so a stored foreign pointer could dangle before the event is consumed. The scalar ID also honors the tracepoint no-nullable-pointer rule, since the other party is frequently unsandboxed. Passing the foreign domain hierarchy object so an eBPF consumer could walk the other party's ancestry live would lengthen the RCU section on the shared denial path and needs a deferred refcount put, so it is left as a future enhancement. The relational domain-ID field (tracee_domain, target_domain, or peer_domain) is trace-only and is not added to audit records, so audit's denial format is unchanged by this series. Cc: Günther Noack Cc: Justin Suess Cc: Masami Hiramatsu Cc: Mathieu Desnoyers Cc: Steven Rostedt Cc: Tingmao Wang Link: https://patch.msgid.link/20260811094338.288094-14-mic@digikod.net Signed-off-by: Mickaël Salaün

landlock: Add landlock_deny_access_fs and landlock_deny_access_net

2026-08-17T08:17:16+00:00

Add per-type tracepoints emitted from landlock_log_denial() when an access is denied: landlock_deny_access_fs for filesystem denials and landlock_deny_access_net for network denials. They use the "deny_" prefix (rather than "check_") to mark that they fire only on a denial, and they complement the check_rule events by making the denial-by-absence case explicit (when no rule matches, no check_rule event fires). Unlike the audit records, these events fire regardless of the audit configuration and the domain's log flags: the user's "disable logging" intent applies to audit records, not to kernel tracing. The logged field records whether the domain's log policy would submit the denial to audit; it is the decision computed once by landlock_log_denial() and passed to both the audit and the tracing emitter, so a stateless ftrace filter can select the audit-visible denials with logged==1. TP_PROTO passes the denying hierarchy node, not the task's current domain, so domain_id reports the specific node that blocked the access, matching audit record semantics. (check_rule instead passes the current domain, which it needs to size its per-layer array.) same_exec is also passed explicitly because it is computed from the credential bitmask and is not derivable from the hierarchy pointer alone. The denial field is named blockers to match the audit record field. The filesystem path comes from the request's audit data. Its type selects which union member holds the object, exactly as dump_common_audit_data() selects it (a path, a file's path, an ioctl op's path, or a bare dentry); reading the wrong member would dereference garbage, so every reachable type has an explicit case and an unexpected one is flagged with WARN_ONCE() instead of misread. Path-backed types resolve via d_absolute_path() (as landlock_add_rule_fs does) and the bare-dentry case via dentry_path_raw(). The inode number is read defensively. A filesystem denial can carry a negative dentry (no backing inode), for example a denied creation, so the event mirrors the guard in dump_common_audit_data() and reports inode 0 rather than dereferencing a NULL inode. The sibling fs tracepoints do not need the guard: a dentry that matches a rule during an access check, or one opened to add a rule, always has a backing inode. Landlock tracepoints are reachable by unprivileged sandboxees, so a denial on a negative dentry with the event enabled must not fault the kernel. Cc: Günther Noack Cc: Justin Suess Cc: Masami Hiramatsu Cc: Mathieu Desnoyers Cc: Steven Rostedt Cc: Tingmao Wang Link: https://patch.msgid.link/20260811094338.288094-13-mic@digikod.net Signed-off-by: Mickaël Salaün

landlock: Add tracepoints for rule checking

2026-08-17T08:17:15+00:00

Merge landlock_find_rule() into landlock_unmask_layers() so rule pointers stay inside the domain implementation while unmask checking gets the matched rule it needs for the check_rule tracepoint. landlock_unmask_layers() now takes a landlock_id and the domain instead of a rule pointer. A rename or link evaluates the same dentry against both renamed parents, so this path now looks the rule up once per parent; collapsing that back to a single lookup is left to a follow-up. Emit, via the per-type wrappers unmask_layers_fs() and unmask_layers_net(), the rights each matching rule grants at every domain layer. The events carry this as a dynamic per-layer array (up to LANDLOCK_MAX_NUM_LAYERS entries) reserved from the trace ring buffer, not the caller's stack, and rendered symbolically per layer. A WARN_ON_ONCE() in __trace_landlock_fill_layers() flags a rule whose layer levels fall outside the domain range or are unsorted, a cannot-happen case; the zero-filled slots keep the rendered output and the array bounds safe regardless. Setting allowed_parent2 to true for non-dom-check requests when get_inode_id() returns false preserves the pre-refactoring behavior: a negative dentry (no backing inode) has no matching rule, so the access is allowed at this path component. Before the refactoring, landlock_unmask_layers() with a NULL rule produced this result as a side effect; now the caller must set it explicitly. Name the trace-only check_rule fields so each printk label equals its ring-buffer field name and works directly as an ftrace filter: the request field is labelled access_request= and the per-layer array is named grants. Values audit also logs keep audit's label (domain=, ruleset=) so a single filter works across trace and audit. Cc: Günther Noack Cc: Justin Suess Cc: Masami Hiramatsu Cc: Mathieu Desnoyers Cc: Steven Rostedt Cc: Tingmao Wang Link: https://patch.msgid.link/20260811094338.288094-12-mic@digikod.net Signed-off-by: Mickaël Salaün

landlock: Add landlock_enforce_domain tracepoint

2026-08-17T08:17:15+00:00

The landlock_create_domain event records that a domain was created, once, before thread-sync. It cannot tell which threads end up enforcing it: a successful landlock_restrict_self(2) with LANDLOCK_RESTRICT_SELF_TSYNC applies the domain to the caller and every eligible sibling. Creation (the operation) and enforcement (the per-thread outcome) are distinct. Add landlock_enforce_domain(domain, complete, process_wide), emitted once per thread the domain is applied to, strictly after that thread's commit_creds(), so it fires only for a thread that is enforcing the domain, never speculatively; an aborted operation emits none. The lifecycle now reads create -> enforce* -> free. The two booleans name properties, not the implementation: - complete: marks the single event that concludes the operation. It names the outcome, the set is now enforced, not which thread finishes, which the contract leaves unspecified. - process_wide: means every eligible thread of the process is covered. It is set race-free by either establishing path, thread-sync or a single-threaded process, so complete && process_wide is the whole-process-enforced guarantee. The requesting thread and source ruleset are not repeated here: they are on create_domain (joined via domain->hierarchy->id) and on the immutable domain->hierarchy->details. Source ruleset means the ruleset_id and ruleset_version recorded on create_domain, not the ruleset object, which the caller may close before enforcement. Cc: Günther Noack Cc: Masami Hiramatsu Cc: Mathieu Desnoyers Cc: Steven Rostedt Cc: Tingmao Wang Link: https://patch.msgid.link/20260811094338.288094-11-mic@digikod.net Signed-off-by: Mickaël Salaün

landlock: Add create_domain and free_domain tracepoints

2026-08-17T08:17:14+00:00

Add a landlock_create_domain tracepoint emitted from landlock_restrict_self() after the new domain is created, so a consumer can correlate the source ruleset with the resulting domain. The flags-only path (ruleset_fd == -1) creates no domain and emits no event. Move the ruleset lock acquisition from landlock_merge_ruleset() to the caller so the lock is held across both the merge and the tracepoint emission, giving an eBPF program a consistent ruleset snapshot. Release it before the thread-sync: holding ruleset->lock across landlock_restrict_sibling_threads() would deadlock a sibling blocked on the same lock. The event therefore fires before the (rare) thread-sync failure path; when that path aborts the just-created domain, the matching free_domain event fires so the create/free pair stays balanced. Add a landlock_free_domain tracepoint that fires when a domain's hierarchy node is freed. The hierarchy node is the lifecycle boundary because it represents the domain's identity and outlives the domain's access masks, which may still be active in descendant domains. A domain freed without ever being committed to a credential was never visible to user space, so free_domain is suppressed for it. This is tracked by a new landlock_log_status value, LANDLOCK_LOG_UNCOMMITTED, which is also the zero value so a hierarchy whose initialization failed defaults to not observable. A hierarchy is born UNCOMMITTED and is promoted to LANDLOCK_LOG_PENDING (or LANDLOCK_LOG_DISABLED when logging is off) right after its create_domain event fires; a thread-sync failure does not reset it, so an aborted domain that already emitted create_domain still emits the matching free_domain. Promoting right after the event, rather than at commit_creds() time, avoids a race: on a successful thread-sync the sibling threads commit the new domain in lockstep before landlock_restrict_self() returns, so the shared domain may already have moved to LANDLOCK_LOG_RECORDED through a plain store, and a late promotion would race that store and could unbalance the domain allocation and deallocation audit records. Cc: Günther Noack Cc: Justin Suess Cc: Masami Hiramatsu Cc: Mathieu Desnoyers Cc: Steven Rostedt Cc: Tingmao Wang Link: https://patch.msgid.link/20260811094338.288094-10-mic@digikod.net Signed-off-by: Mickaël Salaün

landlock: Add landlock_add_rule_fs and landlock_add_rule_net tracepoints

2026-08-17T08:17:14+00:00

Add tracepoints for Landlock rule addition, landlock_add_rule_fs for filesystem rules and landlock_add_rule_net for network rules, so trace consumers can correlate filesystem objects and network ports with their rulesets. Both are emitted under the ruleset lock (asserted in TP_fast_assign) so an eBPF program reads the ruleset, including the rule just inserted, in a consistent snapshot. Add a version field to struct landlock_ruleset, gated on CONFIG_TRACEPOINTS like the id field and incremented under the ruleset lock on each successful landlock_add_rule(2), including when it only extends an existing rule's access rights. It fills the existing 4-byte hole after usage, so the struct does not grow. Pairing the ruleset ID with the version lets a later restrict_self event record the exact ruleset revision merged into a domain. Resolve the filesystem rule's absolute path with d_absolute_path() rather than the d_path() audit uses: d_absolute_path() produces namespace-independent paths that do not depend on the tracer's chroot state, making trace output deterministic regardless of mount namespace configuration. Distinguish the error cases as "" (-ENAMETOOLONG) and "" (anonymous files or detached mounts). Also add __trace_print_untrusted_str(), a static inline helper in the header guarded by CREATE_TRACE_POINTS: it escapes separators, quotes, backslashes, and non-printable bytes via string_escape_mem() so an untrusted string (the path here, process names in later denial events) cannot inject field separators or control characters into the ftrace text output. Cc: Christian Brauner Cc: Günther Noack Cc: Justin Suess Cc: Masami Hiramatsu Cc: Mathieu Desnoyers Cc: Steven Rostedt Cc: Tingmao Wang Link: https://patch.msgid.link/20260811094338.288094-9-mic@digikod.net Signed-off-by: Mickaël Salaün

landlock: Add create_ruleset and free_ruleset tracepoints

2026-08-17T08:17:13+00:00

Add the first Landlock tracepoints, for ruleset lifecycle: landlock_create_ruleset fires from the landlock_create_ruleset() syscall handler, and landlock_free_ruleset fires in free_ruleset() before the ruleset is freed. These tracepoints, and the ones added by the following commits, share a common design. Rather than one polymorphic event distinguished by a status field (as audit uses a shared record type with a "status=" field), each lifecycle transition and denial type gets its own event with a type-safe TP_PROTO, giving precise ftrace filtering by event name and type-safe eBPF access. TP_PROTO passes the object pointer and the fields are read from it in TP_fast_assign, so an eBPF program reads the full object state (rules, access masks, hierarchy) via BTF from a single pointer rather than from the flattened TP_STRUCT__entry fields. The whole cost is paid only when a tracer is attached; the static branch is not taken otherwise. Trace fields carry the bare access-right and scope names (read_file), reusing the audit name tables; audit prepends the category (fs.read_file), which the trace event name already conveys. The trace header's DOC comment documents the consistency and locking guarantees these events share. create_ruleset needs no lock because the ruleset is not yet shared (its file descriptor is not yet installed). The deallocation events use the "free_" prefix, not "drop_", because they fire when the object is actually freed. Add trace.c, built for CONFIG_TRACEPOINTS, which defines CREATE_TRACE_POINTS, and extend CONFIG_SECURITY_LANDLOCK_LOG to also be selected by CONFIG_TRACEPOINTS so the common log framework is available to a tracepoints-only build. Add an id field to struct landlock_ruleset, gated on CONFIG_TRACEPOINTS and assigned from landlock_get_id_range() at creation. Only the tracepoints consume it (audit identifies domains, not rulesets), so it does not exist in an audit-only build. The Landlock ID is a stable u64 that names the ruleset across the trace stream and uses the same scheme as audit, so a ruleset can be correlated between trace and audit records. Cc: Günther Noack Cc: Justin Suess Cc: Masami Hiramatsu Cc: Mathieu Desnoyers Cc: Steven Rostedt Cc: Tingmao Wang Link: https://patch.msgid.link/20260811094338.288094-8-mic@digikod.net Signed-off-by: Mickaël Salaün

landlock: Consolidate access-right and scope names in a shared header

2026-08-17T08:17:13+00:00

Audit formats denial records with per-right name strings. A following commit adds trace events that print the same access and scope masks with __print_flags() and need the same names, but a trace event header cannot include Landlock-internal headers, so the names cannot be shared from the logging unit. Define the filesystem, network, and scope names once, as the _LANDLOCK_ACCESS_FS_NAMES, _LANDLOCK_ACCESS_NET_NAMES, and _LANDLOCK_SCOPE_NAMES lists in the public Landlock header. Each entry is a _LANDLOCK_NAME_ENTRY() the consumer expands: audit maps it to a "[bit] = name" array slot for an O(1) lookup, the trace events map it to a __print_flags() { mask, name } pair. The bit value comes only from the LANDLOCK_* UAPI constant each entry references, so every bit-to-name mapping has a single source and does not depend on entry order. The shared names are unprefixed; blocker_prefix() prepends the fs./net./scope. category for audit records, so the scope names move from inline literals to the shared table too. Audit records are unchanged. No functional change. Cc: Günther Noack Cc: Tingmao Wang Link: https://patch.msgid.link/20260811094338.288094-7-mic@digikod.net Signed-off-by: Mickaël Salaün