]> git.ipfire.org Git - thirdparty/kernel/stable.git/commitdiff
selftests/namespaces: add file handle selftests
authorChristian Brauner <brauner@kernel.org>
Fri, 12 Sep 2025 11:52:56 +0000 (13:52 +0200)
committerChristian Brauner <brauner@kernel.org>
Fri, 19 Sep 2025 12:26:17 +0000 (14:26 +0200)
Add a bunch of selftests for namespace file handles.

Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
tools/testing/selftests/namespaces/.gitignore
tools/testing/selftests/namespaces/Makefile
tools/testing/selftests/namespaces/file_handle_test.c [new file with mode: 0644]

index c1e8d634dd21b1960c0d3c43bbf3ff84cc73fe06..7639dbf58bbf873b39a3fd74fd8f7b6d9c92b020 100644 (file)
@@ -1 +1,2 @@
 nsid_test
+file_handle_test
index 9280c703533e305015ad98e2b553f7b9651b24b8..f6c117ce2c2b8095c1660e0ab2317736871414dd 100644 (file)
@@ -1,7 +1,7 @@
 # SPDX-License-Identifier: GPL-2.0-only
 CFLAGS += -Wall -O0 -g $(KHDR_INCLUDES) $(TOOLS_INCLUDES)
 
-TEST_GEN_PROGS := nsid_test
+TEST_GEN_PROGS := nsid_test file_handle_test
 
 include ../lib.mk
 
diff --git a/tools/testing/selftests/namespaces/file_handle_test.c b/tools/testing/selftests/namespaces/file_handle_test.c
new file mode 100644 (file)
index 0000000..f1bc577
--- /dev/null
@@ -0,0 +1,1429 @@
+// SPDX-License-Identifier: GPL-2.0
+#define _GNU_SOURCE
+#include <errno.h>
+#include <fcntl.h>
+#include <grp.h>
+#include <limits.h>
+#include <sched.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/mount.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <unistd.h>
+#include <linux/unistd.h>
+#include "../kselftest_harness.h"
+
+#ifndef FD_NSFS_ROOT
+#define FD_NSFS_ROOT -10003 /* Root of the nsfs filesystem */
+#endif
+
+TEST(nsfs_net_handle)
+{
+       struct file_handle *handle;
+       int mount_id;
+       int ret;
+       int fd;
+       int ns_fd;
+       struct stat st1, st2;
+
+       /* Drop to unprivileged uid/gid */
+       ASSERT_EQ(setresgid(65534, 65534, 65534), 0); /* nogroup */
+       ASSERT_EQ(setresuid(65534, 65534, 65534), 0); /* nobody */
+
+       handle = malloc(sizeof(*handle) + MAX_HANDLE_SZ);
+       ASSERT_NE(handle, NULL);
+
+       /* Open a namespace file descriptor */
+       ns_fd = open("/proc/self/ns/net", O_RDONLY);
+       ASSERT_GE(ns_fd, 0);
+
+       /* Get handle for the namespace */
+       handle->handle_bytes = MAX_HANDLE_SZ;
+       ret = name_to_handle_at(ns_fd, "", handle, &mount_id, AT_EMPTY_PATH);
+       if (ret < 0 && errno == EOPNOTSUPP) {
+               SKIP(free(handle); close(ns_fd);
+                    return, "nsfs doesn't support file handles");
+       }
+       ASSERT_EQ(ret, 0);
+       ASSERT_GT(handle->handle_bytes, 0);
+
+       /* Try to open using FD_NSFS_ROOT as unprivileged user */
+       fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
+       if (fd < 0 && (errno == EINVAL || errno == EOPNOTSUPP)) {
+               SKIP(free(handle); close(ns_fd);
+                    return,
+                          "open_by_handle_at with FD_NSFS_ROOT not supported");
+       }
+       if (fd < 0 && errno == EPERM) {
+               SKIP(free(handle); close(ns_fd);
+                    return,
+                          "Permission denied for unprivileged user (expected)");
+       }
+       ASSERT_GE(fd, 0);
+
+       /* Verify we opened the correct namespace */
+       ASSERT_EQ(fstat(ns_fd, &st1), 0);
+       ASSERT_EQ(fstat(fd, &st2), 0);
+       ASSERT_EQ(st1.st_ino, st2.st_ino);
+       ASSERT_EQ(st1.st_dev, st2.st_dev);
+
+       close(fd);
+       close(ns_fd);
+       free(handle);
+}
+
+TEST(nsfs_uts_handle)
+{
+       struct file_handle *handle;
+       int mount_id;
+       int ret;
+       int fd;
+       int ns_fd;
+       struct stat st1, st2;
+
+       /* Drop to unprivileged uid/gid */
+       ASSERT_EQ(setresgid(65534, 65534, 65534), 0); /* nogroup */
+       ASSERT_EQ(setresuid(65534, 65534, 65534), 0); /* nobody */
+
+       handle = malloc(sizeof(*handle) + MAX_HANDLE_SZ);
+       ASSERT_NE(handle, NULL);
+
+       /* Open UTS namespace file descriptor */
+       ns_fd = open("/proc/self/ns/uts", O_RDONLY);
+       ASSERT_GE(ns_fd, 0);
+
+       /* Get handle for the namespace */
+       handle->handle_bytes = MAX_HANDLE_SZ;
+       ret = name_to_handle_at(ns_fd, "", handle, &mount_id, AT_EMPTY_PATH);
+       if (ret < 0 && errno == EOPNOTSUPP) {
+               SKIP(free(handle); close(ns_fd);
+                    return, "nsfs doesn't support file handles");
+       }
+       ASSERT_EQ(ret, 0);
+       ASSERT_GT(handle->handle_bytes, 0);
+
+       /* Try to open using FD_NSFS_ROOT */
+       fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
+       if (fd < 0 && (errno == EINVAL || errno == EOPNOTSUPP)) {
+               SKIP(free(handle); close(ns_fd);
+                    return,
+                          "open_by_handle_at with FD_NSFS_ROOT not supported");
+       }
+       ASSERT_GE(fd, 0);
+
+       /* Verify we opened the correct namespace */
+       ASSERT_EQ(fstat(ns_fd, &st1), 0);
+       ASSERT_EQ(fstat(fd, &st2), 0);
+       ASSERT_EQ(st1.st_ino, st2.st_ino);
+       ASSERT_EQ(st1.st_dev, st2.st_dev);
+
+       close(fd);
+       close(ns_fd);
+       free(handle);
+}
+
+TEST(nsfs_ipc_handle)
+{
+       struct file_handle *handle;
+       int mount_id;
+       int ret;
+       int fd;
+       int ns_fd;
+       struct stat st1, st2;
+
+       /* Drop to unprivileged uid/gid */
+       ASSERT_EQ(setresgid(65534, 65534, 65534), 0); /* nogroup */
+       ASSERT_EQ(setresuid(65534, 65534, 65534), 0); /* nobody */
+
+       handle = malloc(sizeof(*handle) + MAX_HANDLE_SZ);
+       ASSERT_NE(handle, NULL);
+
+       /* Open IPC namespace file descriptor */
+       ns_fd = open("/proc/self/ns/ipc", O_RDONLY);
+       ASSERT_GE(ns_fd, 0);
+
+       /* Get handle for the namespace */
+       handle->handle_bytes = MAX_HANDLE_SZ;
+       ret = name_to_handle_at(ns_fd, "", handle, &mount_id, AT_EMPTY_PATH);
+       if (ret < 0 && errno == EOPNOTSUPP) {
+               SKIP(free(handle); close(ns_fd);
+                    return, "nsfs doesn't support file handles");
+       }
+       ASSERT_EQ(ret, 0);
+       ASSERT_GT(handle->handle_bytes, 0);
+
+       /* Try to open using FD_NSFS_ROOT */
+       fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
+       if (fd < 0 && (errno == EINVAL || errno == EOPNOTSUPP)) {
+               SKIP(free(handle); close(ns_fd);
+                    return,
+                          "open_by_handle_at with FD_NSFS_ROOT not supported");
+       }
+       ASSERT_GE(fd, 0);
+
+       /* Verify we opened the correct namespace */
+       ASSERT_EQ(fstat(ns_fd, &st1), 0);
+       ASSERT_EQ(fstat(fd, &st2), 0);
+       ASSERT_EQ(st1.st_ino, st2.st_ino);
+       ASSERT_EQ(st1.st_dev, st2.st_dev);
+
+       close(fd);
+       close(ns_fd);
+       free(handle);
+}
+
+TEST(nsfs_pid_handle)
+{
+       struct file_handle *handle;
+       int mount_id;
+       int ret;
+       int fd;
+       int ns_fd;
+       struct stat st1, st2;
+
+       /* Drop to unprivileged uid/gid */
+       ASSERT_EQ(setresgid(65534, 65534, 65534), 0); /* nogroup */
+       ASSERT_EQ(setresuid(65534, 65534, 65534), 0); /* nobody */
+
+       handle = malloc(sizeof(*handle) + MAX_HANDLE_SZ);
+       ASSERT_NE(handle, NULL);
+
+       /* Open PID namespace file descriptor */
+       ns_fd = open("/proc/self/ns/pid", O_RDONLY);
+       ASSERT_GE(ns_fd, 0);
+
+       /* Get handle for the namespace */
+       handle->handle_bytes = MAX_HANDLE_SZ;
+       ret = name_to_handle_at(ns_fd, "", handle, &mount_id, AT_EMPTY_PATH);
+       if (ret < 0 && errno == EOPNOTSUPP) {
+               SKIP(free(handle); close(ns_fd);
+                    return, "nsfs doesn't support file handles");
+       }
+       ASSERT_EQ(ret, 0);
+       ASSERT_GT(handle->handle_bytes, 0);
+
+       /* Try to open using FD_NSFS_ROOT */
+       fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
+       if (fd < 0 && (errno == EINVAL || errno == EOPNOTSUPP)) {
+               SKIP(free(handle); close(ns_fd);
+                    return,
+                          "open_by_handle_at with FD_NSFS_ROOT not supported");
+       }
+       ASSERT_GE(fd, 0);
+
+       /* Verify we opened the correct namespace */
+       ASSERT_EQ(fstat(ns_fd, &st1), 0);
+       ASSERT_EQ(fstat(fd, &st2), 0);
+       ASSERT_EQ(st1.st_ino, st2.st_ino);
+       ASSERT_EQ(st1.st_dev, st2.st_dev);
+
+       close(fd);
+       close(ns_fd);
+       free(handle);
+}
+
+TEST(nsfs_mnt_handle)
+{
+       struct file_handle *handle;
+       int mount_id;
+       int ret;
+       int fd;
+       int ns_fd;
+       struct stat st1, st2;
+
+       /* Drop to unprivileged uid/gid */
+       ASSERT_EQ(setresgid(65534, 65534, 65534), 0); /* nogroup */
+       ASSERT_EQ(setresuid(65534, 65534, 65534), 0); /* nobody */
+
+       handle = malloc(sizeof(*handle) + MAX_HANDLE_SZ);
+       ASSERT_NE(handle, NULL);
+
+       /* Open mount namespace file descriptor */
+       ns_fd = open("/proc/self/ns/mnt", O_RDONLY);
+       ASSERT_GE(ns_fd, 0);
+
+       /* Get handle for the namespace */
+       handle->handle_bytes = MAX_HANDLE_SZ;
+       ret = name_to_handle_at(ns_fd, "", handle, &mount_id, AT_EMPTY_PATH);
+       if (ret < 0 && errno == EOPNOTSUPP) {
+               SKIP(free(handle); close(ns_fd);
+                    return, "nsfs doesn't support file handles");
+       }
+       ASSERT_EQ(ret, 0);
+       ASSERT_GT(handle->handle_bytes, 0);
+
+       /* Try to open using FD_NSFS_ROOT */
+       fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
+       if (fd < 0 && (errno == EINVAL || errno == EOPNOTSUPP)) {
+               SKIP(free(handle); close(ns_fd);
+                    return,
+                          "open_by_handle_at with FD_NSFS_ROOT not supported");
+       }
+       ASSERT_GE(fd, 0);
+
+       /* Verify we opened the correct namespace */
+       ASSERT_EQ(fstat(ns_fd, &st1), 0);
+       ASSERT_EQ(fstat(fd, &st2), 0);
+       ASSERT_EQ(st1.st_ino, st2.st_ino);
+       ASSERT_EQ(st1.st_dev, st2.st_dev);
+
+       close(fd);
+       close(ns_fd);
+       free(handle);
+}
+
+TEST(nsfs_user_handle)
+{
+       struct file_handle *handle;
+       int mount_id;
+       int ret;
+       int fd;
+       int ns_fd;
+       struct stat st1, st2;
+
+       /* Drop to unprivileged uid/gid */
+       ASSERT_EQ(setresgid(65534, 65534, 65534), 0); /* nogroup */
+       ASSERT_EQ(setresuid(65534, 65534, 65534), 0); /* nobody */
+
+       handle = malloc(sizeof(*handle) + MAX_HANDLE_SZ);
+       ASSERT_NE(handle, NULL);
+
+       /* Open user namespace file descriptor */
+       ns_fd = open("/proc/self/ns/user", O_RDONLY);
+       ASSERT_GE(ns_fd, 0);
+
+       /* Get handle for the namespace */
+       handle->handle_bytes = MAX_HANDLE_SZ;
+       ret = name_to_handle_at(ns_fd, "", handle, &mount_id, AT_EMPTY_PATH);
+       if (ret < 0 && errno == EOPNOTSUPP) {
+               SKIP(free(handle); close(ns_fd);
+                    return, "nsfs doesn't support file handles");
+       }
+       ASSERT_EQ(ret, 0);
+       ASSERT_GT(handle->handle_bytes, 0);
+
+       /* Try to open using FD_NSFS_ROOT */
+       fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
+       if (fd < 0 && (errno == EINVAL || errno == EOPNOTSUPP)) {
+               SKIP(free(handle); close(ns_fd);
+                    return,
+                          "open_by_handle_at with FD_NSFS_ROOT not supported");
+       }
+       ASSERT_GE(fd, 0);
+
+       /* Verify we opened the correct namespace */
+       ASSERT_EQ(fstat(ns_fd, &st1), 0);
+       ASSERT_EQ(fstat(fd, &st2), 0);
+       ASSERT_EQ(st1.st_ino, st2.st_ino);
+       ASSERT_EQ(st1.st_dev, st2.st_dev);
+
+       close(fd);
+       close(ns_fd);
+       free(handle);
+}
+
+TEST(nsfs_cgroup_handle)
+{
+       struct file_handle *handle;
+       int mount_id;
+       int ret;
+       int fd;
+       int ns_fd;
+       struct stat st1, st2;
+
+       /* Drop to unprivileged uid/gid */
+       ASSERT_EQ(setresgid(65534, 65534, 65534), 0); /* nogroup */
+       ASSERT_EQ(setresuid(65534, 65534, 65534), 0); /* nobody */
+
+       handle = malloc(sizeof(*handle) + MAX_HANDLE_SZ);
+       ASSERT_NE(handle, NULL);
+
+       /* Open cgroup namespace file descriptor */
+       ns_fd = open("/proc/self/ns/cgroup", O_RDONLY);
+       if (ns_fd < 0) {
+               SKIP(free(handle); return, "cgroup namespace not available");
+       }
+
+       /* Get handle for the namespace */
+       handle->handle_bytes = MAX_HANDLE_SZ;
+       ret = name_to_handle_at(ns_fd, "", handle, &mount_id, AT_EMPTY_PATH);
+       if (ret < 0 && errno == EOPNOTSUPP) {
+               SKIP(free(handle); close(ns_fd);
+                    return, "nsfs doesn't support file handles");
+       }
+       ASSERT_EQ(ret, 0);
+       ASSERT_GT(handle->handle_bytes, 0);
+
+       /* Try to open using FD_NSFS_ROOT */
+       fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
+       if (fd < 0 && (errno == EINVAL || errno == EOPNOTSUPP)) {
+               SKIP(free(handle); close(ns_fd);
+                    return,
+                          "open_by_handle_at with FD_NSFS_ROOT not supported");
+       }
+       ASSERT_GE(fd, 0);
+
+       /* Verify we opened the correct namespace */
+       ASSERT_EQ(fstat(ns_fd, &st1), 0);
+       ASSERT_EQ(fstat(fd, &st2), 0);
+       ASSERT_EQ(st1.st_ino, st2.st_ino);
+       ASSERT_EQ(st1.st_dev, st2.st_dev);
+
+       close(fd);
+       close(ns_fd);
+       free(handle);
+}
+
+TEST(nsfs_time_handle)
+{
+       struct file_handle *handle;
+       int mount_id;
+       int ret;
+       int fd;
+       int ns_fd;
+       struct stat st1, st2;
+
+       /* Drop to unprivileged uid/gid */
+       ASSERT_EQ(setresgid(65534, 65534, 65534), 0); /* nogroup */
+       ASSERT_EQ(setresuid(65534, 65534, 65534), 0); /* nobody */
+
+       handle = malloc(sizeof(*handle) + MAX_HANDLE_SZ);
+       ASSERT_NE(handle, NULL);
+
+       /* Open time namespace file descriptor */
+       ns_fd = open("/proc/self/ns/time", O_RDONLY);
+       if (ns_fd < 0) {
+               SKIP(free(handle); return, "time namespace not available");
+       }
+
+       /* Get handle for the namespace */
+       handle->handle_bytes = MAX_HANDLE_SZ;
+       ret = name_to_handle_at(ns_fd, "", handle, &mount_id, AT_EMPTY_PATH);
+       if (ret < 0 && errno == EOPNOTSUPP) {
+               SKIP(free(handle); close(ns_fd);
+                    return, "nsfs doesn't support file handles");
+       }
+       ASSERT_EQ(ret, 0);
+       ASSERT_GT(handle->handle_bytes, 0);
+
+       /* Try to open using FD_NSFS_ROOT */
+       fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
+       if (fd < 0 && (errno == EINVAL || errno == EOPNOTSUPP)) {
+               SKIP(free(handle); close(ns_fd);
+                    return,
+                          "open_by_handle_at with FD_NSFS_ROOT not supported");
+       }
+       ASSERT_GE(fd, 0);
+
+       /* Verify we opened the correct namespace */
+       ASSERT_EQ(fstat(ns_fd, &st1), 0);
+       ASSERT_EQ(fstat(fd, &st2), 0);
+       ASSERT_EQ(st1.st_ino, st2.st_ino);
+       ASSERT_EQ(st1.st_dev, st2.st_dev);
+
+       close(fd);
+       close(ns_fd);
+       free(handle);
+}
+
+TEST(nsfs_user_net_namespace_isolation)
+{
+       struct file_handle *handle;
+       int mount_id;
+       int ret;
+       int fd;
+       int ns_fd;
+       pid_t pid;
+       int status;
+       int pipefd[2];
+       char result;
+
+       handle = malloc(sizeof(*handle) + MAX_HANDLE_SZ);
+       ASSERT_NE(handle, NULL);
+
+       /* Create pipe for communication */
+       ASSERT_EQ(pipe(pipefd), 0);
+
+       /* Get handle for current network namespace */
+       ns_fd = open("/proc/self/ns/net", O_RDONLY);
+       ASSERT_GE(ns_fd, 0);
+
+       handle->handle_bytes = MAX_HANDLE_SZ;
+       ret = name_to_handle_at(ns_fd, "", handle, &mount_id, AT_EMPTY_PATH);
+       if (ret < 0 && errno == EOPNOTSUPP) {
+               SKIP(free(handle); close(ns_fd); close(pipefd[0]);
+                    close(pipefd[1]);
+                    return, "nsfs doesn't support file handles");
+       }
+       ASSERT_EQ(ret, 0);
+       close(ns_fd);
+
+       pid = fork();
+       ASSERT_GE(pid, 0);
+
+       if (pid == 0) {
+               /* Child process */
+               close(pipefd[0]);
+
+               /* First create new user namespace to drop privileges */
+               ret = unshare(CLONE_NEWUSER);
+               if (ret < 0) {
+                       write(pipefd[1], "U",
+                             1); /* Unable to create user namespace */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Write uid/gid mappings to maintain some capabilities */
+               int uid_map_fd = open("/proc/self/uid_map", O_WRONLY);
+               int gid_map_fd = open("/proc/self/gid_map", O_WRONLY);
+               int setgroups_fd = open("/proc/self/setgroups", O_WRONLY);
+
+               if (uid_map_fd < 0 || gid_map_fd < 0 || setgroups_fd < 0) {
+                       write(pipefd[1], "M", 1); /* Unable to set mappings */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Disable setgroups to allow gid mapping */
+               write(setgroups_fd, "deny", 4);
+               close(setgroups_fd);
+
+               /* Map current uid/gid to root in the new namespace */
+               char mapping[64];
+               snprintf(mapping, sizeof(mapping), "0 %d 1", getuid());
+               write(uid_map_fd, mapping, strlen(mapping));
+               close(uid_map_fd);
+
+               snprintf(mapping, sizeof(mapping), "0 %d 1", getgid());
+               write(gid_map_fd, mapping, strlen(mapping));
+               close(gid_map_fd);
+
+               /* Now create new network namespace */
+               ret = unshare(CLONE_NEWNET);
+               if (ret < 0) {
+                       write(pipefd[1], "N",
+                             1); /* Unable to create network namespace */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Try to open parent's network namespace handle from new user+net namespace */
+               fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
+
+               if (fd >= 0) {
+                       /* Should NOT succeed - we're in a different user namespace */
+                       write(pipefd[1], "S", 1); /* Unexpected success */
+                       close(fd);
+               } else if (errno == ESTALE) {
+                       /* Expected: Stale file handle */
+                       write(pipefd[1], "P", 1);
+               } else {
+                       /* Other error */
+                       write(pipefd[1], "F", 1);
+               }
+
+               close(pipefd[1]);
+               exit(0);
+       }
+
+       /* Parent process */
+       close(pipefd[1]);
+       ASSERT_EQ(read(pipefd[0], &result, 1), 1);
+
+       waitpid(pid, &status, 0);
+       ASSERT_TRUE(WIFEXITED(status));
+       ASSERT_EQ(WEXITSTATUS(status), 0);
+
+       if (result == 'U') {
+               SKIP(free(handle); close(pipefd[0]);
+                    return, "Cannot create new user namespace");
+       }
+       if (result == 'M') {
+               SKIP(free(handle); close(pipefd[0]);
+                    return, "Cannot set uid/gid mappings");
+       }
+       if (result == 'N') {
+               SKIP(free(handle); close(pipefd[0]);
+                    return, "Cannot create new network namespace");
+       }
+
+       /* Should fail with permission denied since we're in a different user namespace */
+       ASSERT_EQ(result, 'P');
+
+       close(pipefd[0]);
+       free(handle);
+}
+
+TEST(nsfs_user_uts_namespace_isolation)
+{
+       struct file_handle *handle;
+       int mount_id;
+       int ret;
+       int fd;
+       int ns_fd;
+       pid_t pid;
+       int status;
+       int pipefd[2];
+       char result;
+
+       handle = malloc(sizeof(*handle) + MAX_HANDLE_SZ);
+       ASSERT_NE(handle, NULL);
+
+       /* Create pipe for communication */
+       ASSERT_EQ(pipe(pipefd), 0);
+
+       /* Get handle for current UTS namespace */
+       ns_fd = open("/proc/self/ns/uts", O_RDONLY);
+       ASSERT_GE(ns_fd, 0);
+
+       handle->handle_bytes = MAX_HANDLE_SZ;
+       ret = name_to_handle_at(ns_fd, "", handle, &mount_id, AT_EMPTY_PATH);
+       if (ret < 0 && errno == EOPNOTSUPP) {
+               SKIP(free(handle); close(ns_fd); close(pipefd[0]);
+                    close(pipefd[1]);
+                    return, "nsfs doesn't support file handles");
+       }
+       ASSERT_EQ(ret, 0);
+       close(ns_fd);
+
+       pid = fork();
+       ASSERT_GE(pid, 0);
+
+       if (pid == 0) {
+               /* Child process */
+               close(pipefd[0]);
+
+               /* First create new user namespace to drop privileges */
+               ret = unshare(CLONE_NEWUSER);
+               if (ret < 0) {
+                       write(pipefd[1], "U",
+                             1); /* Unable to create user namespace */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Write uid/gid mappings to maintain some capabilities */
+               int uid_map_fd = open("/proc/self/uid_map", O_WRONLY);
+               int gid_map_fd = open("/proc/self/gid_map", O_WRONLY);
+               int setgroups_fd = open("/proc/self/setgroups", O_WRONLY);
+
+               if (uid_map_fd < 0 || gid_map_fd < 0 || setgroups_fd < 0) {
+                       write(pipefd[1], "M", 1); /* Unable to set mappings */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Disable setgroups to allow gid mapping */
+               write(setgroups_fd, "deny", 4);
+               close(setgroups_fd);
+
+               /* Map current uid/gid to root in the new namespace */
+               char mapping[64];
+               snprintf(mapping, sizeof(mapping), "0 %d 1", getuid());
+               write(uid_map_fd, mapping, strlen(mapping));
+               close(uid_map_fd);
+
+               snprintf(mapping, sizeof(mapping), "0 %d 1", getgid());
+               write(gid_map_fd, mapping, strlen(mapping));
+               close(gid_map_fd);
+
+               /* Now create new UTS namespace */
+               ret = unshare(CLONE_NEWUTS);
+               if (ret < 0) {
+                       write(pipefd[1], "N",
+                             1); /* Unable to create UTS namespace */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Try to open parent's UTS namespace handle from new user+uts namespace */
+               fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
+
+               if (fd >= 0) {
+                       /* Should NOT succeed - we're in a different user namespace */
+                       write(pipefd[1], "S", 1); /* Unexpected success */
+                       close(fd);
+               } else if (errno == ESTALE) {
+                       /* Expected: Stale file handle */
+                       write(pipefd[1], "P", 1);
+               } else {
+                       /* Other error */
+                       write(pipefd[1], "F", 1);
+               }
+
+               close(pipefd[1]);
+               exit(0);
+       }
+
+       /* Parent process */
+       close(pipefd[1]);
+       ASSERT_EQ(read(pipefd[0], &result, 1), 1);
+
+       waitpid(pid, &status, 0);
+       ASSERT_TRUE(WIFEXITED(status));
+       ASSERT_EQ(WEXITSTATUS(status), 0);
+
+       if (result == 'U') {
+               SKIP(free(handle); close(pipefd[0]);
+                    return, "Cannot create new user namespace");
+       }
+       if (result == 'M') {
+               SKIP(free(handle); close(pipefd[0]);
+                    return, "Cannot set uid/gid mappings");
+       }
+       if (result == 'N') {
+               SKIP(free(handle); close(pipefd[0]);
+                    return, "Cannot create new UTS namespace");
+       }
+
+       /* Should fail with ESTALE since we're in a different user namespace */
+       ASSERT_EQ(result, 'P');
+
+       close(pipefd[0]);
+       free(handle);
+}
+
+TEST(nsfs_user_ipc_namespace_isolation)
+{
+       struct file_handle *handle;
+       int mount_id;
+       int ret;
+       int fd;
+       int ns_fd;
+       pid_t pid;
+       int status;
+       int pipefd[2];
+       char result;
+
+       handle = malloc(sizeof(*handle) + MAX_HANDLE_SZ);
+       ASSERT_NE(handle, NULL);
+
+       /* Create pipe for communication */
+       ASSERT_EQ(pipe(pipefd), 0);
+
+       /* Get handle for current IPC namespace */
+       ns_fd = open("/proc/self/ns/ipc", O_RDONLY);
+       ASSERT_GE(ns_fd, 0);
+
+       handle->handle_bytes = MAX_HANDLE_SZ;
+       ret = name_to_handle_at(ns_fd, "", handle, &mount_id, AT_EMPTY_PATH);
+       if (ret < 0 && errno == EOPNOTSUPP) {
+               SKIP(free(handle); close(ns_fd); close(pipefd[0]);
+                    close(pipefd[1]);
+                    return, "nsfs doesn't support file handles");
+       }
+       ASSERT_EQ(ret, 0);
+       close(ns_fd);
+
+       pid = fork();
+       ASSERT_GE(pid, 0);
+
+       if (pid == 0) {
+               /* Child process */
+               close(pipefd[0]);
+
+               /* First create new user namespace to drop privileges */
+               ret = unshare(CLONE_NEWUSER);
+               if (ret < 0) {
+                       write(pipefd[1], "U",
+                             1); /* Unable to create user namespace */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Write uid/gid mappings to maintain some capabilities */
+               int uid_map_fd = open("/proc/self/uid_map", O_WRONLY);
+               int gid_map_fd = open("/proc/self/gid_map", O_WRONLY);
+               int setgroups_fd = open("/proc/self/setgroups", O_WRONLY);
+
+               if (uid_map_fd < 0 || gid_map_fd < 0 || setgroups_fd < 0) {
+                       write(pipefd[1], "M", 1); /* Unable to set mappings */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Disable setgroups to allow gid mapping */
+               write(setgroups_fd, "deny", 4);
+               close(setgroups_fd);
+
+               /* Map current uid/gid to root in the new namespace */
+               char mapping[64];
+               snprintf(mapping, sizeof(mapping), "0 %d 1", getuid());
+               write(uid_map_fd, mapping, strlen(mapping));
+               close(uid_map_fd);
+
+               snprintf(mapping, sizeof(mapping), "0 %d 1", getgid());
+               write(gid_map_fd, mapping, strlen(mapping));
+               close(gid_map_fd);
+
+               /* Now create new IPC namespace */
+               ret = unshare(CLONE_NEWIPC);
+               if (ret < 0) {
+                       write(pipefd[1], "N",
+                             1); /* Unable to create IPC namespace */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Try to open parent's IPC namespace handle from new user+ipc namespace */
+               fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
+
+               if (fd >= 0) {
+                       /* Should NOT succeed - we're in a different user namespace */
+                       write(pipefd[1], "S", 1); /* Unexpected success */
+                       close(fd);
+               } else if (errno == ESTALE) {
+                       /* Expected: Stale file handle */
+                       write(pipefd[1], "P", 1);
+               } else {
+                       /* Other error */
+                       write(pipefd[1], "F", 1);
+               }
+
+               close(pipefd[1]);
+               exit(0);
+       }
+
+       /* Parent process */
+       close(pipefd[1]);
+       ASSERT_EQ(read(pipefd[0], &result, 1), 1);
+
+       waitpid(pid, &status, 0);
+       ASSERT_TRUE(WIFEXITED(status));
+       ASSERT_EQ(WEXITSTATUS(status), 0);
+
+       if (result == 'U') {
+               SKIP(free(handle); close(pipefd[0]);
+                    return, "Cannot create new user namespace");
+       }
+       if (result == 'M') {
+               SKIP(free(handle); close(pipefd[0]);
+                    return, "Cannot set uid/gid mappings");
+       }
+       if (result == 'N') {
+               SKIP(free(handle); close(pipefd[0]);
+                    return, "Cannot create new IPC namespace");
+       }
+
+       /* Should fail with ESTALE since we're in a different user namespace */
+       ASSERT_EQ(result, 'P');
+
+       close(pipefd[0]);
+       free(handle);
+}
+
+TEST(nsfs_user_mnt_namespace_isolation)
+{
+       struct file_handle *handle;
+       int mount_id;
+       int ret;
+       int fd;
+       int ns_fd;
+       pid_t pid;
+       int status;
+       int pipefd[2];
+       char result;
+
+       handle = malloc(sizeof(*handle) + MAX_HANDLE_SZ);
+       ASSERT_NE(handle, NULL);
+
+       /* Create pipe for communication */
+       ASSERT_EQ(pipe(pipefd), 0);
+
+       /* Get handle for current mount namespace */
+       ns_fd = open("/proc/self/ns/mnt", O_RDONLY);
+       ASSERT_GE(ns_fd, 0);
+
+       handle->handle_bytes = MAX_HANDLE_SZ;
+       ret = name_to_handle_at(ns_fd, "", handle, &mount_id, AT_EMPTY_PATH);
+       if (ret < 0 && errno == EOPNOTSUPP) {
+               SKIP(free(handle); close(ns_fd); close(pipefd[0]);
+                    close(pipefd[1]);
+                    return, "nsfs doesn't support file handles");
+       }
+       ASSERT_EQ(ret, 0);
+       close(ns_fd);
+
+       pid = fork();
+       ASSERT_GE(pid, 0);
+
+       if (pid == 0) {
+               /* Child process */
+               close(pipefd[0]);
+
+               /* First create new user namespace to drop privileges */
+               ret = unshare(CLONE_NEWUSER);
+               if (ret < 0) {
+                       write(pipefd[1], "U",
+                             1); /* Unable to create user namespace */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Write uid/gid mappings to maintain some capabilities */
+               int uid_map_fd = open("/proc/self/uid_map", O_WRONLY);
+               int gid_map_fd = open("/proc/self/gid_map", O_WRONLY);
+               int setgroups_fd = open("/proc/self/setgroups", O_WRONLY);
+
+               if (uid_map_fd < 0 || gid_map_fd < 0 || setgroups_fd < 0) {
+                       write(pipefd[1], "M", 1); /* Unable to set mappings */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Disable setgroups to allow gid mapping */
+               write(setgroups_fd, "deny", 4);
+               close(setgroups_fd);
+
+               /* Map current uid/gid to root in the new namespace */
+               char mapping[64];
+               snprintf(mapping, sizeof(mapping), "0 %d 1", getuid());
+               write(uid_map_fd, mapping, strlen(mapping));
+               close(uid_map_fd);
+
+               snprintf(mapping, sizeof(mapping), "0 %d 1", getgid());
+               write(gid_map_fd, mapping, strlen(mapping));
+               close(gid_map_fd);
+
+               /* Now create new mount namespace */
+               ret = unshare(CLONE_NEWNS);
+               if (ret < 0) {
+                       write(pipefd[1], "N",
+                             1); /* Unable to create mount namespace */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Try to open parent's mount namespace handle from new user+mnt namespace */
+               fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
+
+               if (fd >= 0) {
+                       /* Should NOT succeed - we're in a different user namespace */
+                       write(pipefd[1], "S", 1); /* Unexpected success */
+                       close(fd);
+               } else if (errno == ESTALE) {
+                       /* Expected: Stale file handle */
+                       write(pipefd[1], "P", 1);
+               } else {
+                       /* Other error */
+                       write(pipefd[1], "F", 1);
+               }
+
+               close(pipefd[1]);
+               exit(0);
+       }
+
+       /* Parent process */
+       close(pipefd[1]);
+       ASSERT_EQ(read(pipefd[0], &result, 1), 1);
+
+       waitpid(pid, &status, 0);
+       ASSERT_TRUE(WIFEXITED(status));
+       ASSERT_EQ(WEXITSTATUS(status), 0);
+
+       if (result == 'U') {
+               SKIP(free(handle); close(pipefd[0]);
+                    return, "Cannot create new user namespace");
+       }
+       if (result == 'M') {
+               SKIP(free(handle); close(pipefd[0]);
+                    return, "Cannot set uid/gid mappings");
+       }
+       if (result == 'N') {
+               SKIP(free(handle); close(pipefd[0]);
+                    return, "Cannot create new mount namespace");
+       }
+
+       /* Should fail with ESTALE since we're in a different user namespace */
+       ASSERT_EQ(result, 'P');
+
+       close(pipefd[0]);
+       free(handle);
+}
+
+TEST(nsfs_user_cgroup_namespace_isolation)
+{
+       struct file_handle *handle;
+       int mount_id;
+       int ret;
+       int fd;
+       int ns_fd;
+       pid_t pid;
+       int status;
+       int pipefd[2];
+       char result;
+
+       handle = malloc(sizeof(*handle) + MAX_HANDLE_SZ);
+       ASSERT_NE(handle, NULL);
+
+       /* Create pipe for communication */
+       ASSERT_EQ(pipe(pipefd), 0);
+
+       /* Get handle for current cgroup namespace */
+       ns_fd = open("/proc/self/ns/cgroup", O_RDONLY);
+       if (ns_fd < 0) {
+               SKIP(free(handle); close(pipefd[0]); close(pipefd[1]);
+                    return, "cgroup namespace not available");
+       }
+
+       handle->handle_bytes = MAX_HANDLE_SZ;
+       ret = name_to_handle_at(ns_fd, "", handle, &mount_id, AT_EMPTY_PATH);
+       if (ret < 0 && errno == EOPNOTSUPP) {
+               SKIP(free(handle); close(ns_fd); close(pipefd[0]);
+                    close(pipefd[1]);
+                    return, "nsfs doesn't support file handles");
+       }
+       ASSERT_EQ(ret, 0);
+       close(ns_fd);
+
+       pid = fork();
+       ASSERT_GE(pid, 0);
+
+       if (pid == 0) {
+               /* Child process */
+               close(pipefd[0]);
+
+               /* First create new user namespace to drop privileges */
+               ret = unshare(CLONE_NEWUSER);
+               if (ret < 0) {
+                       write(pipefd[1], "U",
+                             1); /* Unable to create user namespace */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Write uid/gid mappings to maintain some capabilities */
+               int uid_map_fd = open("/proc/self/uid_map", O_WRONLY);
+               int gid_map_fd = open("/proc/self/gid_map", O_WRONLY);
+               int setgroups_fd = open("/proc/self/setgroups", O_WRONLY);
+
+               if (uid_map_fd < 0 || gid_map_fd < 0 || setgroups_fd < 0) {
+                       write(pipefd[1], "M", 1); /* Unable to set mappings */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Disable setgroups to allow gid mapping */
+               write(setgroups_fd, "deny", 4);
+               close(setgroups_fd);
+
+               /* Map current uid/gid to root in the new namespace */
+               char mapping[64];
+               snprintf(mapping, sizeof(mapping), "0 %d 1", getuid());
+               write(uid_map_fd, mapping, strlen(mapping));
+               close(uid_map_fd);
+
+               snprintf(mapping, sizeof(mapping), "0 %d 1", getgid());
+               write(gid_map_fd, mapping, strlen(mapping));
+               close(gid_map_fd);
+
+               /* Now create new cgroup namespace */
+               ret = unshare(CLONE_NEWCGROUP);
+               if (ret < 0) {
+                       write(pipefd[1], "N",
+                             1); /* Unable to create cgroup namespace */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Try to open parent's cgroup namespace handle from new user+cgroup namespace */
+               fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
+
+               if (fd >= 0) {
+                       /* Should NOT succeed - we're in a different user namespace */
+                       write(pipefd[1], "S", 1); /* Unexpected success */
+                       close(fd);
+               } else if (errno == ESTALE) {
+                       /* Expected: Stale file handle */
+                       write(pipefd[1], "P", 1);
+               } else {
+                       /* Other error */
+                       write(pipefd[1], "F", 1);
+               }
+
+               close(pipefd[1]);
+               exit(0);
+       }
+
+       /* Parent process */
+       close(pipefd[1]);
+       ASSERT_EQ(read(pipefd[0], &result, 1), 1);
+
+       waitpid(pid, &status, 0);
+       ASSERT_TRUE(WIFEXITED(status));
+       ASSERT_EQ(WEXITSTATUS(status), 0);
+
+       if (result == 'U') {
+               SKIP(free(handle); close(pipefd[0]);
+                    return, "Cannot create new user namespace");
+       }
+       if (result == 'M') {
+               SKIP(free(handle); close(pipefd[0]);
+                    return, "Cannot set uid/gid mappings");
+       }
+       if (result == 'N') {
+               SKIP(free(handle); close(pipefd[0]);
+                    return, "Cannot create new cgroup namespace");
+       }
+
+       /* Should fail with ESTALE since we're in a different user namespace */
+       ASSERT_EQ(result, 'P');
+
+       close(pipefd[0]);
+       free(handle);
+}
+
+TEST(nsfs_user_pid_namespace_isolation)
+{
+       struct file_handle *handle;
+       int mount_id;
+       int ret;
+       int fd;
+       int ns_fd;
+       pid_t pid;
+       int status;
+       int pipefd[2];
+       char result;
+
+       handle = malloc(sizeof(*handle) + MAX_HANDLE_SZ);
+       ASSERT_NE(handle, NULL);
+
+       /* Create pipe for communication */
+       ASSERT_EQ(pipe(pipefd), 0);
+
+       /* Get handle for current PID namespace */
+       ns_fd = open("/proc/self/ns/pid", O_RDONLY);
+       ASSERT_GE(ns_fd, 0);
+
+       handle->handle_bytes = MAX_HANDLE_SZ;
+       ret = name_to_handle_at(ns_fd, "", handle, &mount_id, AT_EMPTY_PATH);
+       if (ret < 0 && errno == EOPNOTSUPP) {
+               SKIP(free(handle); close(ns_fd); close(pipefd[0]);
+                    close(pipefd[1]);
+                    return, "nsfs doesn't support file handles");
+       }
+       ASSERT_EQ(ret, 0);
+       close(ns_fd);
+
+       pid = fork();
+       ASSERT_GE(pid, 0);
+
+       if (pid == 0) {
+               /* Child process */
+               close(pipefd[0]);
+
+               /* First create new user namespace to drop privileges */
+               ret = unshare(CLONE_NEWUSER);
+               if (ret < 0) {
+                       write(pipefd[1], "U",
+                             1); /* Unable to create user namespace */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Write uid/gid mappings to maintain some capabilities */
+               int uid_map_fd = open("/proc/self/uid_map", O_WRONLY);
+               int gid_map_fd = open("/proc/self/gid_map", O_WRONLY);
+               int setgroups_fd = open("/proc/self/setgroups", O_WRONLY);
+
+               if (uid_map_fd < 0 || gid_map_fd < 0 || setgroups_fd < 0) {
+                       write(pipefd[1], "M", 1); /* Unable to set mappings */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Disable setgroups to allow gid mapping */
+               write(setgroups_fd, "deny", 4);
+               close(setgroups_fd);
+
+               /* Map current uid/gid to root in the new namespace */
+               char mapping[64];
+               snprintf(mapping, sizeof(mapping), "0 %d 1", getuid());
+               write(uid_map_fd, mapping, strlen(mapping));
+               close(uid_map_fd);
+
+               snprintf(mapping, sizeof(mapping), "0 %d 1", getgid());
+               write(gid_map_fd, mapping, strlen(mapping));
+               close(gid_map_fd);
+
+               /* Now create new PID namespace - requires fork to take effect */
+               ret = unshare(CLONE_NEWPID);
+               if (ret < 0) {
+                       write(pipefd[1], "N",
+                             1); /* Unable to create PID namespace */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Fork again for PID namespace to take effect */
+               pid_t child_pid = fork();
+               if (child_pid < 0) {
+                       write(pipefd[1], "N",
+                             1); /* Unable to fork in PID namespace */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               if (child_pid == 0) {
+                       /* Grandchild in new PID namespace */
+                       /* Try to open parent's PID namespace handle from new user+pid namespace */
+                       fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
+
+                       if (fd >= 0) {
+                               /* Should NOT succeed - we're in a different user namespace */
+                               write(pipefd[1], "S",
+                                     1); /* Unexpected success */
+                               close(fd);
+                       } else if (errno == ESTALE) {
+                               /* Expected: Stale file handle */
+                               write(pipefd[1], "P", 1);
+                       } else {
+                               /* Other error */
+                               write(pipefd[1], "F", 1);
+                       }
+
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Wait for grandchild */
+               waitpid(child_pid, NULL, 0);
+               exit(0);
+       }
+
+       /* Parent process */
+       close(pipefd[1]);
+       ASSERT_EQ(read(pipefd[0], &result, 1), 1);
+
+       waitpid(pid, &status, 0);
+       ASSERT_TRUE(WIFEXITED(status));
+       ASSERT_EQ(WEXITSTATUS(status), 0);
+
+       if (result == 'U') {
+               SKIP(free(handle); close(pipefd[0]);
+                    return, "Cannot create new user namespace");
+       }
+       if (result == 'M') {
+               SKIP(free(handle); close(pipefd[0]);
+                    return, "Cannot set uid/gid mappings");
+       }
+       if (result == 'N') {
+               SKIP(free(handle); close(pipefd[0]);
+                    return, "Cannot create new PID namespace");
+       }
+
+       /* Should fail with ESTALE since we're in a different user namespace */
+       ASSERT_EQ(result, 'P');
+
+       close(pipefd[0]);
+       free(handle);
+}
+
+TEST(nsfs_user_time_namespace_isolation)
+{
+       struct file_handle *handle;
+       int mount_id;
+       int ret;
+       int fd;
+       int ns_fd;
+       pid_t pid;
+       int status;
+       int pipefd[2];
+       char result;
+
+       handle = malloc(sizeof(*handle) + MAX_HANDLE_SZ);
+       ASSERT_NE(handle, NULL);
+
+       /* Create pipe for communication */
+       ASSERT_EQ(pipe(pipefd), 0);
+
+       /* Get handle for current time namespace */
+       ns_fd = open("/proc/self/ns/time", O_RDONLY);
+       if (ns_fd < 0) {
+               SKIP(free(handle); close(pipefd[0]); close(pipefd[1]);
+                    return, "time namespace not available");
+       }
+
+       handle->handle_bytes = MAX_HANDLE_SZ;
+       ret = name_to_handle_at(ns_fd, "", handle, &mount_id, AT_EMPTY_PATH);
+       if (ret < 0 && errno == EOPNOTSUPP) {
+               SKIP(free(handle); close(ns_fd); close(pipefd[0]);
+                    close(pipefd[1]);
+                    return, "nsfs doesn't support file handles");
+       }
+       ASSERT_EQ(ret, 0);
+       close(ns_fd);
+
+       pid = fork();
+       ASSERT_GE(pid, 0);
+
+       if (pid == 0) {
+               /* Child process */
+               close(pipefd[0]);
+
+               /* First create new user namespace to drop privileges */
+               ret = unshare(CLONE_NEWUSER);
+               if (ret < 0) {
+                       write(pipefd[1], "U",
+                             1); /* Unable to create user namespace */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Write uid/gid mappings to maintain some capabilities */
+               int uid_map_fd = open("/proc/self/uid_map", O_WRONLY);
+               int gid_map_fd = open("/proc/self/gid_map", O_WRONLY);
+               int setgroups_fd = open("/proc/self/setgroups", O_WRONLY);
+
+               if (uid_map_fd < 0 || gid_map_fd < 0 || setgroups_fd < 0) {
+                       write(pipefd[1], "M", 1); /* Unable to set mappings */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Disable setgroups to allow gid mapping */
+               write(setgroups_fd, "deny", 4);
+               close(setgroups_fd);
+
+               /* Map current uid/gid to root in the new namespace */
+               char mapping[64];
+               snprintf(mapping, sizeof(mapping), "0 %d 1", getuid());
+               write(uid_map_fd, mapping, strlen(mapping));
+               close(uid_map_fd);
+
+               snprintf(mapping, sizeof(mapping), "0 %d 1", getgid());
+               write(gid_map_fd, mapping, strlen(mapping));
+               close(gid_map_fd);
+
+               /* Now create new time namespace - requires fork to take effect */
+               ret = unshare(CLONE_NEWTIME);
+               if (ret < 0) {
+                       write(pipefd[1], "N",
+                             1); /* Unable to create time namespace */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Fork again for time namespace to take effect */
+               pid_t child_pid = fork();
+               if (child_pid < 0) {
+                       write(pipefd[1], "N",
+                             1); /* Unable to fork in time namespace */
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               if (child_pid == 0) {
+                       /* Grandchild in new time namespace */
+                       /* Try to open parent's time namespace handle from new user+time namespace */
+                       fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
+
+                       if (fd >= 0) {
+                               /* Should NOT succeed - we're in a different user namespace */
+                               write(pipefd[1], "S",
+                                     1); /* Unexpected success */
+                               close(fd);
+                       } else if (errno == ESTALE) {
+                               /* Expected: Stale file handle */
+                               write(pipefd[1], "P", 1);
+                       } else {
+                               /* Other error */
+                               write(pipefd[1], "F", 1);
+                       }
+
+                       close(pipefd[1]);
+                       exit(0);
+               }
+
+               /* Wait for grandchild */
+               waitpid(child_pid, NULL, 0);
+               exit(0);
+       }
+
+       /* Parent process */
+       close(pipefd[1]);
+       ASSERT_EQ(read(pipefd[0], &result, 1), 1);
+
+       waitpid(pid, &status, 0);
+       ASSERT_TRUE(WIFEXITED(status));
+       ASSERT_EQ(WEXITSTATUS(status), 0);
+
+       if (result == 'U') {
+               SKIP(free(handle); close(pipefd[0]);
+                    return, "Cannot create new user namespace");
+       }
+       if (result == 'M') {
+               SKIP(free(handle); close(pipefd[0]);
+                    return, "Cannot set uid/gid mappings");
+       }
+       if (result == 'N') {
+               SKIP(free(handle); close(pipefd[0]);
+                    return, "Cannot create new time namespace");
+       }
+
+       /* Should fail with ESTALE since we're in a different user namespace */
+       ASSERT_EQ(result, 'P');
+
+       close(pipefd[0]);
+       free(handle);
+}
+
+TEST(nsfs_open_flags)
+{
+       struct file_handle *handle;
+       int mount_id;
+       int ret;
+       int fd;
+       int ns_fd;
+
+       handle = malloc(sizeof(*handle) + MAX_HANDLE_SZ);
+       ASSERT_NE(handle, NULL);
+
+       /* Open a namespace file descriptor */
+       ns_fd = open("/proc/self/ns/net", O_RDONLY);
+       ASSERT_GE(ns_fd, 0);
+
+       /* Get handle for the namespace */
+       handle->handle_bytes = MAX_HANDLE_SZ;
+       ret = name_to_handle_at(ns_fd, "", handle, &mount_id, AT_EMPTY_PATH);
+       if (ret < 0 && errno == EOPNOTSUPP) {
+               SKIP(free(handle); close(ns_fd);
+                    return, "nsfs doesn't support file handles");
+       }
+       ASSERT_EQ(ret, 0);
+       ASSERT_GT(handle->handle_bytes, 0);
+
+       /* Test invalid flags that should fail */
+       fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_WRONLY);
+       ASSERT_LT(fd, 0);
+       ASSERT_EQ(errno, EPERM);
+
+       fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDWR);
+       ASSERT_LT(fd, 0);
+       ASSERT_EQ(errno, EPERM);
+
+       fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_TRUNC);
+       ASSERT_LT(fd, 0);
+       ASSERT_EQ(errno, EPERM);
+
+       fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_DIRECT);
+       ASSERT_LT(fd, 0);
+       ASSERT_EQ(errno, EINVAL);
+
+       fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_TMPFILE);
+       ASSERT_LT(fd, 0);
+       ASSERT_EQ(errno, EINVAL);
+
+       fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_DIRECTORY);
+       ASSERT_LT(fd, 0);
+       ASSERT_EQ(errno, ENOTDIR);
+
+       close(ns_fd);
+       free(handle);
+}
+
+TEST_HARNESS_MAIN