[LTP] [PATCH v5 10/14] fchroot09: test setns escape from failfs root

[Andrea Cervesato]

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

Verify that entering failfs with fchroot() is hard to undo: a process
inside counts as chrooted, and one way out is a pre-opened mount
namespace fd. setns() into it resets both the root and the working
directory.

Signed-off-by: Andrea Cervesato <andrea.cervesato@suse.com>
---
 runtest/syscalls                              |  1 +
 testcases/kernel/syscalls/fchroot/.gitignore  |  1 +
 testcases/kernel/syscalls/fchroot/fchroot09.c | 68 +++++++++++++++++++++++++++
 3 files changed, 70 insertions(+)

diff --git a/runtest/syscalls b/runtest/syscalls
index 24d9b1b4a..b4723fee4 100644
--- a/runtest/syscalls
+++ b/runtest/syscalls
@@ -324,6 +324,7 @@ fchroot05 fchroot05
 fchroot06 fchroot06
 fchroot07 fchroot07
 fchroot08 fchroot08
+fchroot09 fchroot09
 
 fcntl01 fcntl01
 fcntl01_64 fcntl01_64
diff --git a/testcases/kernel/syscalls/fchroot/.gitignore b/testcases/kernel/syscalls/fchroot/.gitignore
index 0a2510d8a..b577da61f 100644
--- a/testcases/kernel/syscalls/fchroot/.gitignore
+++ b/testcases/kernel/syscalls/fchroot/.gitignore
@@ -7,3 +7,4 @@ fchroot06
 fchroot07
 fchroot07_child
 fchroot08
+fchroot09
diff --git a/testcases/kernel/syscalls/fchroot/fchroot09.c b/testcases/kernel/syscalls/fchroot/fchroot09.c
new file mode 100644
index 000000000..f4f0e1100
--- /dev/null
+++ b/testcases/kernel/syscalls/fchroot/fchroot09.c
@@ -0,0 +1,68 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright (C) 2026 SUSE LLC Andrea Cervesato <andrea.cervesato@suse.com>
+ */
+
+/*\
+ * Test that :manpage:`setns(2)` escapes the failfs root.
+ *
+ * Entering failfs with :manpage:`fchroot(2)` is hard to undo: a process
+ * inside counts as chrooted, so :manpage:`chroot(2)` and :manpage:`fchroot(2)`
+ * back out require ``CAP_SYS_CHROOT``. One way out is a pre-opened mount
+ * namespace file descriptor: :manpage:`setns(2)` into it resets both the
+ * root and the working directory.
+ *
+ * Root is required because entering failfs requires ``CAP_SYS_CHROOT`` and
+ * :manpage:`setns(2)` into the mount namespace requires ``CAP_SYS_ADMIN``.
+ *
+ * The test runs in a forked child so the root of the parent process is
+ * left untouched.
+ */
+
+#define _GNU_SOURCE
+#include <fcntl.h>
+#include <sys/stat.h>
+#include "tst_test.h"
+#include "lapi/fcntl.h"
+#include "lapi/sched.h"
+#include "lapi/setns.h"
+#include "lapi/syscalls.h"
+
+static void run(void)
+{
+	if (SAFE_FORK())
+		return;
+
+	struct stat realroot, st;
+	int nsfd;
+
+	SAFE_STAT("/", &realroot);
+	nsfd = SAFE_OPEN("/proc/self/ns/mnt", O_RDONLY);
+
+	TST_EXP_PASS(tst_syscall(__NR_fchroot, FD_FAILFS_ROOT, 0),
+		"fchroot() with the FD_FAILFS_ROOT sentinel");
+
+	TST_EXP_FAIL2(open("/", O_PATH), EOPNOTSUPP,
+		"absolute lookup after entering failfs");
+
+	TST_EXP_PASS(setns(nsfd, CLONE_NEWNS),
+		"setns() back into the mount namespace");
+
+	SAFE_CLOSE(nsfd);
+
+	SAFE_STAT("/", &st);
+	TST_EXP_EXPR(st.st_dev == realroot.st_dev &&
+		st.st_ino == realroot.st_ino,
+		"root restored after setns()");
+
+	SAFE_STAT(".", &st);
+	TST_EXP_EXPR(st.st_dev == realroot.st_dev &&
+		st.st_ino == realroot.st_ino,
+		"working directory restored after setns()");
+}
+
+static struct tst_test test = {
+	.test_all = run,
+	.needs_root = 1,
+	.forks_child = 1,
+};

-- 
2.51.0