[LTP] [PATCH STAGING v3 14/15] fchroot12: test failfs entry with shared fs_struct

[Andrea Cervesato]

From: Andrea Cervesato <andrea.cervesato@suse.com>

Verify that unprivileged fchroot() into failfs is refused with a
shared fs_struct: a CLONE_FS sibling without no_new_privs could
execute a setuid binary with the failfs root, so entry requires
fs->users == 1, the same restriction setns() applies for the mount
namespace. The sibling bumps fs->users to 2, making the syscall fail
with EINVAL.

Signed-off-by: Andrea Cervesato <andrea.cervesato@suse.com>
---
 runtest/staging                               |  1 +
 testcases/kernel/syscalls/fchroot/.gitignore  |  1 +
 testcases/kernel/syscalls/fchroot/fchroot12.c | 85 +++++++++++++++++++++++++++
 3 files changed, 87 insertions(+)

diff --git a/runtest/staging b/runtest/staging
index 137e3f99a..1655b6813 100644
--- a/runtest/staging
+++ b/runtest/staging
@@ -11,3 +11,4 @@ fchroot08 fchroot08
 fchroot09 fchroot09
 fchroot10 fchroot10
 fchroot11 fchroot11
+fchroot12 fchroot12
diff --git a/testcases/kernel/syscalls/fchroot/.gitignore b/testcases/kernel/syscalls/fchroot/.gitignore
index 650ef2690..e8a6730ee 100644
--- a/testcases/kernel/syscalls/fchroot/.gitignore
+++ b/testcases/kernel/syscalls/fchroot/.gitignore
@@ -10,3 +10,4 @@ fchroot08
 fchroot09
 fchroot10
 fchroot11
+fchroot12
diff --git a/testcases/kernel/syscalls/fchroot/fchroot12.c b/testcases/kernel/syscalls/fchroot/fchroot12.c
new file mode 100644
index 000000000..c08144ecc
--- /dev/null
+++ b/testcases/kernel/syscalls/fchroot/fchroot12.c
@@ -0,0 +1,85 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright (C) 2026 SUSE LLC Andrea Cervesato <andrea.cervesato@suse.com>
+ */
+
+/*\
+ * Test that unprivileged :manpage:`fchroot(2)` into failfs is refused
+ * with a shared ``fs_struct``.
+ *
+ * ``no_new_privs`` is checked on the calling thread, but the root lives in
+ * the fs_struct shared with a :manpage:`clone(2)` ``CLONE_FS`` sibling. A
+ * sibling without no_new_privs could execute a setuid binary with the
+ * failfs root, so entry requires ``fs->users == 1``, the same restriction
+ * :manpage:`setns(2)` applies for the mount namespace. The sibling bumps
+ * ``fs->users`` to 2, making the syscall fail with ``EINVAL``.
+ *
+ * Root is required to drop to an unprivileged user in the forked child.
+ */
+
+#define _GNU_SOURCE
+#include <sys/wait.h>
+#include <pwd.h>
+#include <sched.h>
+#include <signal.h>
+#include <unistd.h>
+#include "tst_test.h"
+#include "lapi/fcntl.h"
+#include "lapi/prctl.h"
+#include "lapi/sched.h"
+#include "lapi/syscalls.h"
+
+static struct passwd *ltpuser;
+
+static void run(void)
+{
+	if (SAFE_FORK())
+		return;
+
+	struct tst_clone_args args = {
+		.flags = CLONE_FS,
+		.exit_signal = SIGCHLD,
+	};
+	pid_t parent = getpid();
+	pid_t sib;
+
+	SAFE_SETRESUID(ltpuser->pw_uid, ltpuser->pw_uid,
+		ltpuser->pw_uid);
+
+	sib = SAFE_CLONE(&args);
+	if (!sib) {
+		/*
+		 * Parked sibling sharing the fs_struct with the
+		 * caller. It must die with the parent, otherwise it
+		 * would outlive the test.
+		 */
+		SAFE_PRCTL(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0);
+		if (getppid() == parent)
+			pause();
+		exit(0);
+	}
+
+	SAFE_PRCTL(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
+
+	TST_EXP_FAIL(tst_syscall(__NR_fchroot, FD_FAILFS_ROOT, 0),
+		EINVAL, "fchroot() with a shared fs_struct");
+
+	SAFE_KILL(sib, SIGKILL);
+	SAFE_WAITPID(sib, NULL, 0);
+
+	exit(0);
+}
+
+static void setup(void)
+{
+	clone3_supported_by_kernel();
+
+	ltpuser = SAFE_GETPWNAM("nobody");
+}
+
+static struct tst_test test = {
+	.setup = setup,
+	.test_all = run,
+	.needs_root = 1,
+	.forks_child = 1,
+};

-- 
2.51.0