]> git.ipfire.org Git - thirdparty/kernel/linux.git/commitdiff
selftests/landlock: Test LANDLOCK_ACCESS_FS_RESOLVE_UNIX
authorGünther Noack <gnoack3000@gmail.com>
Fri, 27 Mar 2026 16:48:33 +0000 (17:48 +0100)
committerMickaël Salaün <mic@digikod.net>
Tue, 7 Apr 2026 16:51:09 +0000 (18:51 +0200)
* Extract common helpers from an existing IOCTL test that
  also uses pathname unix(7) sockets.
* These tests use the common scoped domains fixture which is also used
  in other Landlock scoping tests and which was used in Tingmao Wang's
  earlier patch set in [1].

These tests exercise the cross product of the following scenarios:

* Stream connect(), Datagram connect(), Datagram sendmsg() and
  Seqpacket connect().
* Child-to-parent and parent-to-child communication
* The Landlock policy configuration as listed in the scoped_domains
  fixture.
  * In the default variant, Landlock domains are only placed where
    prescribed in the fixture.
  * In the "ALL_DOMAINS" variant, Landlock domains are also placed in
    the places where the fixture says to omit them, but with a
    LANDLOCK_RULE_PATH_BENEATH that allows connection.

Cc: Justin Suess <utilityemal77@gmail.com>
Cc: Tingmao Wang <m@maowtm.org>
Cc: Mickaël Salaün <mic@digikod.net>
Link[1]: https://lore.kernel.org/all/53b9883648225d5a08e82d2636ab0b4fda003bc9.1767115163.git.m@maowtm.org/
Signed-off-by: Günther Noack <gnoack3000@gmail.com>
Link: https://lore.kernel.org/r/20260327164838.38231-9-gnoack3000@gmail.com
Signed-off-by: Mickaël Salaün <mic@digikod.net>
tools/testing/selftests/landlock/fs_test.c

index 9fdd3b8f7b11c4cb48b25197389fe32b29ed8951..f8cfd31335e1417d7329670c24fd416ec83fd4eb 100644 (file)
@@ -4358,30 +4358,66 @@ TEST_F_FORK(layout1, named_pipe_ioctl)
        ASSERT_EQ(child_pid, waitpid(child_pid, NULL, 0));
 }
 
-/* For named UNIX domain sockets, no IOCTL restrictions apply. */
-TEST_F_FORK(layout1, named_unix_domain_socket_ioctl)
+/*
+ * set_up_named_unix_server - Create a pathname unix socket
+ *
+ * If the socket type is not SOCK_DGRAM, also invoke listen(2).
+ *
+ * Return: The listening FD - it is the caller responsibility to close it.
+ */
+static int set_up_named_unix_server(struct __test_metadata *const _metadata,
+                                   int type, const char *const path)
 {
-       const char *const path = file1_s1d1;
-       int srv_fd, cli_fd, ruleset_fd;
-       struct sockaddr_un srv_un = {
+       int fd;
+       struct sockaddr_un addr = {
                .sun_family = AF_UNIX,
        };
-       struct sockaddr_un cli_un = {
+
+       fd = socket(AF_UNIX, type, 0);
+       ASSERT_LE(0, fd);
+
+       ASSERT_LT(strlen(path), sizeof(addr.sun_path));
+       strncpy(addr.sun_path, path, sizeof(addr.sun_path) - 1);
+
+       ASSERT_EQ(0, bind(fd, (struct sockaddr *)&addr, sizeof(addr)));
+
+       if (type != SOCK_DGRAM)
+               ASSERT_EQ(0, listen(fd, 10 /* qlen */));
+       return fd;
+}
+
+/*
+ * test_connect_named_unix - connect to the given named UNIX socket
+ *
+ * Return: The errno from connect(), or 0
+ */
+static int test_connect_named_unix(struct __test_metadata *const _metadata,
+                                  int fd, const char *const path)
+{
+       struct sockaddr_un addr = {
                .sun_family = AF_UNIX,
        };
+
+       ASSERT_LT(strlen(path), sizeof(addr.sun_path));
+       strncpy(addr.sun_path, path, sizeof(addr.sun_path) - 1);
+
+       if (connect(fd, (struct sockaddr *)&addr, sizeof(addr)) == -1)
+               return errno;
+       return 0;
+}
+
+/* For named UNIX domain sockets, no IOCTL restrictions apply. */
+TEST_F_FORK(layout1, named_unix_domain_socket_ioctl)
+{
+       const char *const path = file1_s1d1;
+       int srv_fd, cli_fd, ruleset_fd;
        const struct landlock_ruleset_attr attr = {
                .handled_access_fs = LANDLOCK_ACCESS_FS_IOCTL_DEV,
        };
 
        /* Sets up a server */
        ASSERT_EQ(0, unlink(path));
-       srv_fd = socket(AF_UNIX, SOCK_STREAM, 0);
-       ASSERT_LE(0, srv_fd);
-
-       strncpy(srv_un.sun_path, path, sizeof(srv_un.sun_path));
-       ASSERT_EQ(0, bind(srv_fd, (struct sockaddr *)&srv_un, sizeof(srv_un)));
-
-       ASSERT_EQ(0, listen(srv_fd, 10 /* qlen */));
+       srv_fd = set_up_named_unix_server(_metadata, SOCK_STREAM, path);
 
        /* Enables Landlock. */
        ruleset_fd = landlock_create_ruleset(&attr, sizeof(attr), 0);
@@ -4393,9 +4429,7 @@ TEST_F_FORK(layout1, named_unix_domain_socket_ioctl)
        cli_fd = socket(AF_UNIX, SOCK_STREAM, 0);
        ASSERT_LE(0, cli_fd);
 
-       strncpy(cli_un.sun_path, path, sizeof(cli_un.sun_path));
-       ASSERT_EQ(0,
-                 connect(cli_fd, (struct sockaddr *)&cli_un, sizeof(cli_un)));
+       ASSERT_EQ(0, test_connect_named_unix(_metadata, cli_fd, path));
 
        /* FIONREAD and other IOCTLs should not be forbidden. */
        EXPECT_EQ(0, test_fionread_ioctl(cli_fd));
@@ -4570,6 +4604,330 @@ TEST_F_FORK(ioctl, handle_file_access_file)
        ASSERT_EQ(0, close(file_fd));
 }
 
+/*
+ * test_sendto_named_unix - sendto to the given named UNIX socket
+ *
+ * sendto() is equivalent to sendmsg() in this respect.
+ *
+ * Return: The errno from sendto(), or 0
+ */
+static int test_sendto_named_unix(struct __test_metadata *const _metadata,
+                                 int fd, const char *const path)
+{
+       static const char buf[] = "dummy";
+       struct sockaddr_un addr = {
+               .sun_family = AF_UNIX,
+       };
+
+       ASSERT_LT(strlen(path), sizeof(addr.sun_path));
+       strncpy(addr.sun_path, path, sizeof(addr.sun_path) - 1);
+
+       if (sendto(fd, buf, sizeof(buf), 0, (struct sockaddr *)&addr,
+                  sizeof(addr)) == -1)
+               return errno;
+       return 0;
+}
+
+/* clang-format off */
+FIXTURE(scoped_domains) {};
+/* clang-format on */
+
+#include "scoped_base_variants.h"
+
+FIXTURE_SETUP(scoped_domains)
+{
+       drop_caps(_metadata);
+};
+
+FIXTURE_TEARDOWN(scoped_domains)
+{
+}
+
+static void enforce_fs_resolve_unix(struct __test_metadata *const _metadata,
+                                   const struct rule rules[])
+{
+       if (rules) {
+               int fd = create_ruleset(_metadata,
+                                       LANDLOCK_ACCESS_FS_RESOLVE_UNIX, rules);
+               enforce_ruleset(_metadata, fd);
+               EXPECT_EQ(0, close(fd));
+       } else {
+               struct landlock_ruleset_attr attr = {
+                       .handled_access_fs = LANDLOCK_ACCESS_FS_RESOLVE_UNIX,
+               };
+               drop_access_rights(_metadata, &attr);
+       }
+}
+
+/*
+ * Flags for test_connect_to_parent and test_connect_to_child:
+ *
+ * USE_SENDTO: Use sendto() instead of connect() (for SOCK_DGRAM only)
+ * ENFORCE_ALL: Enforce a Landlock domain even when the variant says
+ *   we shouldn't.  We enforce a domain where the path is allow-listed,
+ *   and expect the behavior to be the same as if none was used.
+ */
+#define USE_SENDTO (1 << 0)
+#define ENFORCE_ALL (1 << 1)
+
+static void test_connect_to_parent(struct __test_metadata *const _metadata,
+                                  const FIXTURE_VARIANT(scoped_domains) *
+                                          variant,
+                                  int sock_type, int flags)
+{
+       const char *const path = "sock";
+       const struct rule rules[] = {
+               {
+                       .path = ".",
+                       .access = LANDLOCK_ACCESS_FS_RESOLVE_UNIX,
+               },
+               {},
+       };
+       int cli_fd, srv_fd, res, status;
+       pid_t child_pid;
+       int readiness_pipe[2];
+       char buf[1];
+
+       if (variant->domain_both)
+               enforce_fs_resolve_unix(_metadata, NULL);
+       else if (flags & ENFORCE_ALL)
+               enforce_fs_resolve_unix(_metadata, rules);
+
+       unlink(path);
+       ASSERT_EQ(0, pipe2(readiness_pipe, O_CLOEXEC));
+
+       child_pid = fork();
+       ASSERT_LE(0, child_pid);
+
+       if (child_pid == 0) {
+               if (variant->domain_child)
+                       enforce_fs_resolve_unix(_metadata, NULL);
+               else if (flags & ENFORCE_ALL)
+                       enforce_fs_resolve_unix(_metadata, rules);
+
+               /* Wait for server to be available. */
+               EXPECT_EQ(0, close(readiness_pipe[1]));
+               EXPECT_EQ(1, read(readiness_pipe[0], &buf, 1));
+               EXPECT_EQ(0, close(readiness_pipe[0]));
+
+               /* Talk to server. */
+               cli_fd = socket(AF_UNIX, sock_type, 0);
+               ASSERT_LE(0, cli_fd);
+
+               if (flags & USE_SENDTO)
+                       res = test_sendto_named_unix(_metadata, cli_fd, path);
+               else
+                       res = test_connect_named_unix(_metadata, cli_fd, path);
+
+               EXPECT_EQ(variant->domain_child ? EACCES : 0, res);
+
+               /* Clean up. */
+               EXPECT_EQ(0, close(cli_fd));
+
+               _exit(_metadata->exit_code);
+               return;
+       }
+
+       if (variant->domain_parent)
+               enforce_fs_resolve_unix(_metadata, NULL);
+       else if (flags & ENFORCE_ALL)
+               enforce_fs_resolve_unix(_metadata, rules);
+
+       srv_fd = set_up_named_unix_server(_metadata, sock_type, path);
+
+       /* Tell the child that it can connect. */
+       EXPECT_EQ(0, close(readiness_pipe[0]));
+       EXPECT_EQ(sizeof(buf), write(readiness_pipe[1], buf, sizeof(buf)));
+       EXPECT_EQ(0, close(readiness_pipe[1]));
+
+       /* Wait for child. */
+       ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
+       EXPECT_EQ(1, WIFEXITED(status));
+       EXPECT_EQ(EXIT_SUCCESS, WEXITSTATUS(status));
+
+       /* Clean up. */
+       EXPECT_EQ(0, close(srv_fd));
+       EXPECT_EQ(0, unlink(path));
+}
+
+static void test_connect_to_child(struct __test_metadata *const _metadata,
+                                 const FIXTURE_VARIANT(scoped_domains) *
+                                         variant,
+                                 int sock_type, int flags)
+{
+       const char *const path = "sock";
+       const struct rule rules[] = {
+               {
+                       .path = ".",
+                       .access = LANDLOCK_ACCESS_FS_RESOLVE_UNIX,
+               },
+               {},
+       };
+       int readiness_pipe[2];
+       int shutdown_pipe[2];
+       int cli_fd, srv_fd, res, status;
+       pid_t child_pid;
+       char buf[1];
+
+       if (variant->domain_both)
+               enforce_fs_resolve_unix(_metadata, NULL);
+       else if (flags & ENFORCE_ALL)
+               enforce_fs_resolve_unix(_metadata, rules);
+
+       unlink(path);
+       ASSERT_EQ(0, pipe2(readiness_pipe, O_CLOEXEC));
+       ASSERT_EQ(0, pipe2(shutdown_pipe, O_CLOEXEC));
+
+       child_pid = fork();
+       ASSERT_LE(0, child_pid);
+
+       if (child_pid == 0) {
+               if (variant->domain_child)
+                       enforce_fs_resolve_unix(_metadata, NULL);
+               else if (flags & ENFORCE_ALL)
+                       enforce_fs_resolve_unix(_metadata, rules);
+
+               srv_fd = set_up_named_unix_server(_metadata, sock_type, path);
+
+               /* Tell the parent that it can connect. */
+               EXPECT_EQ(0, close(readiness_pipe[0]));
+               EXPECT_EQ(sizeof(buf),
+                         write(readiness_pipe[1], buf, sizeof(buf)));
+               EXPECT_EQ(0, close(readiness_pipe[1]));
+
+               /* Wait until it is time to shut down. */
+               EXPECT_EQ(0, close(shutdown_pipe[1]));
+               EXPECT_EQ(1, read(shutdown_pipe[0], &buf, 1));
+               EXPECT_EQ(0, close(shutdown_pipe[0]));
+
+               /* Cleanup */
+               EXPECT_EQ(0, close(srv_fd));
+               EXPECT_EQ(0, unlink(path));
+
+               _exit(_metadata->exit_code);
+               return;
+       }
+
+       if (variant->domain_parent)
+               enforce_fs_resolve_unix(_metadata, NULL);
+       else if (flags & ENFORCE_ALL)
+               enforce_fs_resolve_unix(_metadata, rules);
+
+       /* Wait for server to be available. */
+       EXPECT_EQ(0, close(readiness_pipe[1]));
+       EXPECT_EQ(1, read(readiness_pipe[0], &buf, 1));
+       EXPECT_EQ(0, close(readiness_pipe[0]));
+
+       /* Talk to server. */
+       cli_fd = socket(AF_UNIX, sock_type, 0);
+       ASSERT_LE(0, cli_fd);
+
+       if (flags & USE_SENDTO)
+               res = test_sendto_named_unix(_metadata, cli_fd, path);
+       else
+               res = test_connect_named_unix(_metadata, cli_fd, path);
+
+       EXPECT_EQ(variant->domain_parent ? EACCES : 0, res);
+
+       /* Clean up. */
+       EXPECT_EQ(0, close(cli_fd));
+
+       /* Tell the server to shut down. */
+       EXPECT_EQ(0, close(shutdown_pipe[0]));
+       EXPECT_EQ(sizeof(buf), write(shutdown_pipe[1], buf, sizeof(buf)));
+       EXPECT_EQ(0, close(shutdown_pipe[1]));
+
+       /* Wait for child. */
+       ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
+       EXPECT_EQ(1, WIFEXITED(status));
+       EXPECT_EQ(EXIT_SUCCESS, WEXITSTATUS(status));
+}
+
+TEST_F(scoped_domains, unix_stream_connect_to_parent)
+{
+       test_connect_to_parent(_metadata, variant, SOCK_STREAM, 0);
+}
+
+TEST_F(scoped_domains, unix_dgram_connect_to_parent)
+{
+       test_connect_to_parent(_metadata, variant, SOCK_DGRAM, 0);
+}
+
+TEST_F(scoped_domains, unix_dgram_sendmsg_to_parent)
+{
+       test_connect_to_parent(_metadata, variant, SOCK_DGRAM, USE_SENDTO);
+}
+
+TEST_F(scoped_domains, unix_seqpacket_connect_to_parent)
+{
+       test_connect_to_parent(_metadata, variant, SOCK_SEQPACKET, 0);
+}
+
+TEST_F(scoped_domains, unix_stream_connect_to_parent_full)
+{
+       test_connect_to_parent(_metadata, variant, SOCK_STREAM, ENFORCE_ALL);
+}
+
+TEST_F(scoped_domains, unix_dgram_connect_to_parent_full)
+{
+       test_connect_to_parent(_metadata, variant, SOCK_DGRAM, ENFORCE_ALL);
+}
+
+TEST_F(scoped_domains, unix_dgram_sendmsg_to_parent_full)
+{
+       test_connect_to_parent(_metadata, variant, SOCK_DGRAM,
+                              USE_SENDTO | ENFORCE_ALL);
+}
+
+TEST_F(scoped_domains, unix_seqpacket_connect_to_parent_full)
+{
+       test_connect_to_parent(_metadata, variant, SOCK_SEQPACKET, ENFORCE_ALL);
+}
+
+TEST_F(scoped_domains, unix_stream_connect_to_child)
+{
+       test_connect_to_child(_metadata, variant, SOCK_STREAM, 0);
+}
+
+TEST_F(scoped_domains, unix_dgram_connect_to_child)
+{
+       test_connect_to_child(_metadata, variant, SOCK_DGRAM, 0);
+}
+
+TEST_F(scoped_domains, unix_dgram_sendmsg_to_child)
+{
+       test_connect_to_child(_metadata, variant, SOCK_DGRAM, USE_SENDTO);
+}
+
+TEST_F(scoped_domains, unix_seqpacket_connect_to_child)
+{
+       test_connect_to_child(_metadata, variant, SOCK_SEQPACKET, 0);
+}
+
+TEST_F(scoped_domains, unix_stream_connect_to_child_full)
+{
+       test_connect_to_child(_metadata, variant, SOCK_STREAM, ENFORCE_ALL);
+}
+
+TEST_F(scoped_domains, unix_dgram_connect_to_child_full)
+{
+       test_connect_to_child(_metadata, variant, SOCK_DGRAM, ENFORCE_ALL);
+}
+
+TEST_F(scoped_domains, unix_dgram_sendmsg_to_child_full)
+{
+       test_connect_to_child(_metadata, variant, SOCK_DGRAM,
+                             USE_SENDTO | ENFORCE_ALL);
+}
+
+TEST_F(scoped_domains, unix_seqpacket_connect_to_child_full)
+{
+       test_connect_to_child(_metadata, variant, SOCK_SEQPACKET, ENFORCE_ALL);
+}
+
+#undef USE_SENDTO
+#undef ENFORCE_ALL
+
 /* clang-format off */
 FIXTURE(layout1_bind) {};
 /* clang-format on */