kernel/tools/testing/selftests/landlock, branch master

selftests/landlock: Add filesystem access benchmark

2026-02-10T15:46:42Z

fs_bench benchmarks the performance of Landlock's path walk by exercising it in a scenario that amplifies Landlock's overhead: * Create a large number of nested directories * Enforce a Landlock policy in which a rule is associated with each of these subdirectories * Benchmark openat() applied to the deepest directory, forcing Landlock to walk the entire path. Signed-off-by: Günther Noack Link: https://lore.kernel.org/r/20260206151154.97915-3-gnoack3000@gmail.com [mic: Fix missing mode with O_CREAT, improve text consistency, sort includes] Signed-off-by: Mickaël Salaün

selftests/landlock: Add LANDLOCK_RESTRICT_SELF_TSYNC tests

2026-02-06T16:54:39Z

Exercise various scenarios where Landlock domains are enforced across all of a processes' threads. Test coverage for security/landlock is 91.6% of 2130 lines according to LLVM 21. Cc: Andrew G. Morgan Cc: John Johansen Cc: Paul Moore Signed-off-by: Günther Noack Link: https://lore.kernel.org/r/20251127115136.3064948-3-gnoack@google.com [mic: Fix subject, use EXPECT_EQ(close()), make helpers static, add test coverage] Signed-off-by: Mickaël Salaün

landlock: Multithreading support for landlock_restrict_self()

2026-02-06T16:54:37Z

Introduce the LANDLOCK_RESTRICT_SELF_TSYNC flag. With this flag, a given Landlock ruleset is applied to all threads of the calling process, instead of only the current one. Without this flag, multithreaded userspace programs currently resort to using the nptl(7)/libpsx hack for multithreaded policy enforcement, which is also used by libcap and for setuid(2). Using this userspace-based scheme, the threads of a process enforce the same Landlock policy, but the resulting Landlock domains are still separate. The domains being separate causes multiple problems: * When using Landlock's "scoped" access rights, the domain identity is used to determine whether an operation is permitted. As a result, when using LANLDOCK_SCOPE_SIGNAL, signaling between sibling threads stops working. This is a problem for programming languages and frameworks which are inherently multithreaded (e.g. Go). * In audit logging, the domains of separate threads in a process will get logged with different domain IDs, even when they are based on the same ruleset FD, which might confuse users. Cc: Andrew G. Morgan Cc: John Johansen Cc: Paul Moore Suggested-by: Jann Horn Signed-off-by: Günther Noack Link: https://lore.kernel.org/r/20251127115136.3064948-2-gnoack@google.com [mic: Fix restrict_self_flags test, clean up Makefile, allign comments, reduce local variable scope, add missing includes] Closes: https://github.com/landlock-lsm/linux/issues/2 Signed-off-by: Mickaël Salaün

selftests/landlock: Properly close a file descriptor

2026-01-09T10:52:47Z

Add a missing close(srv_fd) call, and use EXPECT_EQ() to check the result. Signed-off-by: Günther Noack Fixes: f83d51a5bdfe ("selftests/landlock: Check IOCTL restrictions for named UNIX domain sockets") Link: https://lore.kernel.org/r/20260101134102.25938-2-gnoack3000@gmail.com [mic: Use EXPECT_EQ() and update commit message] Signed-off-by: Mickaël Salaün

selftests/landlock: Use scoped_base_variants.h for ptrace_test

2025-12-29T15:19:39Z

ptrace_test.c currently contains a duplicated version of the scoped_domains fixture variants. This patch removes that and make it use the shared scoped_base_variants.h instead, like in scoped_abstract_unix_test and scoped_signal_test. This required renaming the hierarchy fixture to scoped_domains, but the test is otherwise the same. Cc: Tahera Fahimi Signed-off-by: Tingmao Wang Link: https://lore.kernel.org/r/48148f0134f95f819a25277486a875a6fd88ecf9.1766885035.git.m@maowtm.org Signed-off-by: Mickaël Salaün

selftests/landlock: Fix missing semicolon

2025-12-29T15:19:38Z

Add missing semicolon after EXPECT_EQ(0, close(stream_server_child)) in the scoped_vs_unscoped test. I suspect currently it's just not executing the close statement after the line, but this causes no observable difference. Fixes: fefcf0f7cf47 ("selftests/landlock: Test abstract UNIX socket scoping") Cc: Tahera Fahimi Signed-off-by: Tingmao Wang Link: https://lore.kernel.org/r/d9e968e4cd4ecc9bf487593d7b7220bffbb3b5f5.1766885035.git.m@maowtm.org Signed-off-by: Mickaël Salaün

selftests/landlock: Fix typo in fs_test

2025-12-29T15:19:38Z

Signed-off-by: Tingmao Wang Link: https://lore.kernel.org/r/62d18e06eeb26f62bc49d24c4467b3793c5ba32b.1766885035.git.m@maowtm.org Signed-off-by: Mickaël Salaün

selftests/landlock: NULL-terminate unix pathname addresses

2025-12-26T19:38:58Z

The size of Unix pathname addresses is computed in selftests using offsetof(struct sockaddr_un, sun_path) + strlen(xxx). It should have been that +1, which makes addresses passed to the libc and kernel non-NULL-terminated. unix_mkname_bsd() fixes that in Linux so there is no harm, but just using sizeof(the address struct) should improve readability. Signed-off-by: Matthieu Buffet Reviewed-by: Günther Noack Link: https://lore.kernel.org/r/20251202215141.689986-1-matthieu@buffet.re Signed-off-by: Mickaël Salaün

selftests/landlock: Remove invalid unix socket bind()

2025-12-26T19:38:58Z

Remove bind() call on a client socket that doesn't make sense. Since strlen(cli_un.sun_path) returns a random value depending on stack garbage, that many uninitialized bytes are read from the stack as an unix socket address. This creates random test failures due to the bind address being invalid or already in use if the same stack value comes up twice. Fixes: f83d51a5bdfe ("selftests/landlock: Check IOCTL restrictions for named UNIX domain sockets") Signed-off-by: Matthieu Buffet Reviewed-by: Günther Noack Link: https://lore.kernel.org/r/20251201003631.190817-1-matthieu@buffet.re Signed-off-by: Mickaël Salaün

selftests/landlock: Add missing connect(minimal AF_UNSPEC) test

2025-12-26T19:38:57Z

connect_variant(unspec_any0) is called twice. Both calls end up in connect_variant_addrlen() with an address length of get_addrlen(minimal=false). However, the connect() syscall and its variants (e.g. iouring/compat) accept much shorter addresses of 4 bytes and that behaviour was not tested. Replace one of these calls with one using a minimal address length (just a bare sa_family=AF_UNSPEC field with no actual address). Also add a call using a truncated address for good measure. Signed-off-by: Matthieu Buffet Link: https://lore.kernel.org/r/20251027190726.626244-3-matthieu@buffet.re Signed-off-by: Mickaël Salaün