mount-util: fix fd_is_mount_point() when both the parent and directory are network fs

The second call to name_to_handle_at_loop() didn't check for the specific
errors that can happen when the parent dir is mounted by nfs and instead of
falling back like it's done for the child dir, fd_is_mount_point() failed in
this case.

.editorconfig

@@ -18,6 +18,7 @@ charset = utf-8
 [*.{c,h}]
 indent_style = space
 indent_size = 8
+max_line_length = 109
 
 [*.sh]
 indent_style = space
@@ -30,3 +31,4 @@ indent_size = 8
 [man/*.xml]
 indent_size = 2
 indent_style = space
+max_line_length = 109

hwdb.d/60-evdev.hwdb

@@ -191,10 +191,10 @@ evdev:name:Elan Touchpad:dmi:*:svnASUSTeKCOMPUTERINC.:pnUX305UA:*
 
 # Asus UX362FA
 evdev:name:ELAN1401:00 04F3:30DC Touchpad:dmi:*:svnASUSTeKCOMPUTERINC.:pnZenBookUX362FA_UX362FA:*
- EVDEV_ABS_00=:::16
+ EVDEV_ABS_00=:::8
- EVDEV_ABS_01=:::16
+ EVDEV_ABS_01=:::8
- EVDEV_ABS_35=:::16
+ EVDEV_ABS_35=:::8
- EVDEV_ABS_36=:::16
+ EVDEV_ABS_36=:::8
 
 #########################################
 # Bangho

src/basic/mountpoint-util.c

@@ -157,6 +157,19 @@ static bool filename_possibly_with_slash_suffix(const char *s) {
         return filename_is_valid(copied);
 }
 
+static bool is_name_to_handle_at_fatal_error(int err) {
+        /* name_to_handle_at() can return "acceptable" errors that are due to the context. For
+         * example the kernel does not support name_to_handle_at() at all (ENOSYS), or the syscall
+         * was blocked (EACCES/EPERM; maybe through seccomp, because we are running inside of a
+         * container), or the mount point is not triggered yet (EOVERFLOW, think nfs4), or some
+         * general name_to_handle_at() flakiness (EINVAL). However other errors are not supposed to
+         * happen and therefore are considered fatal ones. */
+
+        assert(err < 0);
+
+        return !IN_SET(err, -EOPNOTSUPP, -ENOSYS, -EACCES, -EPERM, -EOVERFLOW, -EINVAL);
+}
+
 int fd_is_mount_point(int fd, const char *filename, int flags) {
         _cleanup_free_ struct file_handle *h = NULL, *h_parent = NULL;
         int mount_id = -1, mount_id_parent = -1;
@@ -206,39 +219,40 @@ int fd_is_mount_point(int fd, const char *filename, int flags) {
                 return false; /* symlinks are never mount points */
 
         r = name_to_handle_at_loop(fd, filename, &h, &mount_id, flags);
-        if (IN_SET(r, -ENOSYS, -EACCES, -EPERM, -EOVERFLOW, -EINVAL))
+        if (r < 0) {
-                /* This kernel does not support name_to_handle_at() at all (ENOSYS), or the syscall was blocked
+                if (is_name_to_handle_at_fatal_error(r))
-                 * (EACCES/EPERM; maybe through seccomp, because we are running inside of a container?), or the mount
-                 * point is not triggered yet (EOVERFLOW, think nfs4), or some general name_to_handle_at() flakiness
-                 * (EINVAL): fall back to simpler logic. */
-                goto fallback_fdinfo;
-        else if (r == -EOPNOTSUPP)
-                /* This kernel or file system does not support name_to_handle_at(), hence let's see if the upper fs
-                 * supports it (in which case it is a mount point), otherwise fall back to the traditional stat()
-                 * logic */
-                nosupp = true;
-        else if (r < 0)
                         return r;
+                if (r != -EOPNOTSUPP)
+                        goto fallback_fdinfo;
+
+                /* This kernel or file system does not support name_to_handle_at(), hence let's see
+                 * if the upper fs supports it (in which case it is a mount point), otherwise fall
+                 * back to the traditional stat() logic */
+                nosupp = true;
+        }
 
         r = name_to_handle_at_loop(fd, "", &h_parent, &mount_id_parent, AT_EMPTY_PATH);
-        if (r == -EOPNOTSUPP) {
+        if (r < 0) {
-                if (nosupp)
+                if (is_name_to_handle_at_fatal_error(r))
-                        /* Neither parent nor child do name_to_handle_at()?  We have no choice but to fall back. */
-                        goto fallback_fdinfo;
-                else
-                        /* The parent can't do name_to_handle_at() but the directory we are interested in can?  If so,
-                         * it must be a mount point. */
-                        return 1;
-        } else if (r < 0)
                         return r;
+                if (r != -EOPNOTSUPP)
+                        goto fallback_fdinfo;
+                if (nosupp)
+                        /* Both the parent and the directory can't do name_to_handle_at() */
+                        goto fallback_fdinfo;
 
-        /* The parent can do name_to_handle_at() but the directory we are interested in can't? If so, it must
+                /* The parent can't do name_to_handle_at() but the directory we are
-         * be a mount point. */
+                 * interested in can?  If so, it must be a mount point. */
+                return 1;
+        }
+
+        /* The parent can do name_to_handle_at() but the directory we are interested in can't? If
+         * so, it must be a mount point. */
         if (nosupp)
                 return 1;
 
-        /* If the file handle for the directory we are interested in and its parent are identical, we assume
+        /* If the file handle for the directory we are interested in and its parent are identical,
-         * this is the root directory, which is a mount point. */
+         * we assume this is the root directory, which is a mount point. */
 
         if (h->handle_bytes == h_parent->handle_bytes &&
             h->handle_type == h_parent->handle_type &&
@@ -338,10 +352,10 @@ int path_get_mnt_id(const char *path, int *ret) {
         }
 
         r = name_to_handle_at_loop(AT_FDCWD, path, NULL, ret, 0);
-        if (IN_SET(r, -EOPNOTSUPP, -ENOSYS, -EACCES, -EPERM, -EOVERFLOW, -EINVAL)) /* kernel/fs don't support this, or seccomp blocks access, or untriggered mount, or name_to_handle_at() is flaky */
+        if (r == 0 || is_name_to_handle_at_fatal_error(r))
-                return fd_fdinfo_mnt_id(AT_FDCWD, path, 0, ret);
-
                 return r;
+
+        return fd_fdinfo_mnt_id(AT_FDCWD, path, 0, ret);
 }
 
 bool fstype_is_network(const char *fstype) {

src/shared/seccomp-util.c

@@ -1789,6 +1789,10 @@ int seccomp_restrict_archs(Set *archs) {
         for (unsigned i = 0; seccomp_local_archs[i] != SECCOMP_LOCAL_ARCH_END; ++i) {
                 uint32_t arch = seccomp_local_archs[i];
 
+                /* See above comment, our "native" architecture is never blocked. */
+                if (arch == seccomp_arch_native())
+                        continue;
+
                 /* That architecture might have already been blocked by a previous call to seccomp_restrict_archs. */
                 if (arch == SECCOMP_LOCAL_ARCH_BLOCKED)
                         continue;

src/test/test-execute.c

@@ -713,8 +713,10 @@ static void test_exec_systemcallfilter(Manager *m) {
 
         test(m, "exec-systemcallfilter-not-failing.service", 0, CLD_EXITED);
         test(m, "exec-systemcallfilter-not-failing2.service", 0, CLD_EXITED);
+        test(m, "exec-systemcallfilter-not-failing3.service", 0, CLD_EXITED);
         test(m, "exec-systemcallfilter-failing.service", SIGSYS, CLD_KILLED);
         test(m, "exec-systemcallfilter-failing2.service", SIGSYS, CLD_KILLED);
+        test(m, "exec-systemcallfilter-failing3.service", SIGSYS, CLD_KILLED);
 
         r = find_executable("python3", NULL);
         if (r < 0) {

src/test/test-seccomp.c

@@ -890,6 +890,66 @@ static void test_load_syscall_filter_set_raw(void) {
         assert_se(wait_for_terminate_and_check("syscallrawseccomp", pid, WAIT_LOG) == EXIT_SUCCESS);
 }
 
+static void test_native_syscalls_filtered(void) {
+        pid_t pid;
+
+        log_info("/* %s */", __func__);
+
+        if (!is_seccomp_available()) {
+                log_notice("Seccomp not available, skipping %s", __func__);
+                return;
+        }
+        if (!have_seccomp_privs()) {
+                log_notice("Not privileged, skipping %s", __func__);
+                return;
+        }
+
+        pid = fork();
+        assert_se(pid >= 0);
+
+        if (pid == 0) {
+                _cleanup_set_free_ Set *arch_s = NULL;
+                _cleanup_hashmap_free_ Hashmap *s = NULL;
+
+                /* Passing "native" or an empty set is equivalent, just do both here. */
+                assert_se(arch_s = set_new(NULL));
+                assert_se(seccomp_restrict_archs(arch_s) >= 0);
+                assert_se(set_put(arch_s, SCMP_ARCH_NATIVE) >= 0);
+                assert_se(seccomp_restrict_archs(arch_s) >= 0);
+
+                assert_se(access("/", F_OK) >= 0);
+                assert_se(poll(NULL, 0, 0) == 0);
+
+                assert_se(seccomp_load_syscall_filter_set_raw(SCMP_ACT_ALLOW, NULL, scmp_act_kill_process(), true) >= 0);
+                assert_se(access("/", F_OK) >= 0);
+                assert_se(poll(NULL, 0, 0) == 0);
+
+                assert_se(s = hashmap_new(NULL));
+#if defined __NR_access && __NR_access >= 0
+                assert_se(hashmap_put(s, UINT32_TO_PTR(__NR_access + 1), INT_TO_PTR(-1)) >= 0);
+                log_debug("has access()");
+#endif
+#if defined __NR_faccessat && __NR_faccessat >= 0
+                assert_se(hashmap_put(s, UINT32_TO_PTR(__NR_faccessat + 1), INT_TO_PTR(-1)) >= 0);
+                log_debug("has faccessat()");
+#endif
+#if defined __NR_faccessat2 && __NR_faccessat2 >= 0
+                assert_se(hashmap_put(s, UINT32_TO_PTR(__NR_faccessat2 + 1), INT_TO_PTR(-1)) >= 0);
+                log_debug("has faccessat2()");
+#endif
+
+                assert_se(!hashmap_isempty(s));
+                assert_se(seccomp_load_syscall_filter_set_raw(SCMP_ACT_ALLOW, s, SCMP_ACT_ERRNO(EUCLEAN), true) >= 0);
+
+                assert_se(access("/", F_OK) < 0);
+                assert_se(errno == EUCLEAN);
+
+                _exit(EXIT_SUCCESS);
+        }
+
+        assert_se(wait_for_terminate_and_check("nativeseccomp", pid, WAIT_LOG) == EXIT_SUCCESS);
+}
+
 static void test_lock_personality(void) {
         unsigned long current;
         pid_t pid;
@@ -1171,6 +1231,7 @@ int main(int argc, char *argv[]) {
         test_memory_deny_write_execute_shmat();
         test_restrict_archs();
         test_load_syscall_filter_set_raw();
+        test_native_syscalls_filtered();
         test_lock_personality();
         test_restrict_suid_sgid();
 

test/networkd-test.py

@@ -635,7 +635,7 @@ Name={}
 [Network]
 DHCP=ipv4
 IPv6AcceptRA=False
-DNSSECNegativeTrustAnchors=megasearch.net
+DNSSECNegativeTrustAnchors=search.example.com
 '''.format(self.iface))
 
         # create second device/dnsmasq for a .company/.lab VPN interface
@@ -681,8 +681,8 @@ DNSSECNegativeTrustAnchors=company lab
         self.assertIn(b'kettle.cantina.company: 10.241.4.4', out)
 
         # test general domains
-        out = subprocess.check_output(['resolvectl', 'query', 'megasearch.net'])
+        out = subprocess.check_output(['resolvectl', 'query', 'search.example.com'])
-        self.assertIn(b'megasearch.net: 192.168.42.1', out)
+        self.assertIn(b'search.example.com: 192.168.42.1', out)
 
         with open(self.dnsmasq_log) as f:
             general_log = f.read()
@@ -696,8 +696,8 @@ DNSSECNegativeTrustAnchors=company lab
         self.assertNotIn('.company', general_log)
 
         # general domains should not be sent to the VPN DNS
-        self.assertRegex(general_log, 'query.*megasearch.net')
+        self.assertRegex(general_log, 'query.*search.example.com')
-        self.assertNotIn('megasearch.net', vpn_log)
+        self.assertNotIn('search.example.com', vpn_log)
 
     def test_resolved_etc_hosts(self):
         '''resolved queries to /etc/hosts'''

test/test-execute/exec-systemcallfilter-failing3.service

@@ -0,0 +1,9 @@
+[Unit]
+Description=Test for SystemCallFilter
+
+[Service]
+ExecStart=/bin/sh -c '/bin/echo "This should not be seen"'
+Type=oneshot
+LimitCORE=0
+SystemCallArchitectures=native
+SystemCallFilter=~write open execve fexecve execveat exit_group close mmap munmap fstat DONOTEXIST

test/test-execute/exec-systemcallfilter-not-failing3.service

@@ -0,0 +1,8 @@
+[Unit]
+Description=Test for SystemCallFilter
+
+[Service]
+ExecStart=/bin/sh -c 'echo "Foo bar"'
+Type=oneshot
+SystemCallArchitectures=native
+SystemCallFilter=

test/units/testsuite-29.sh

@@ -4,6 +4,13 @@
 set -eux
 set -o pipefail
 
+ARGS=()
+if [[ -v ASAN_OPTIONS || -v UBSAN_OPTIONS ]]; then
+    # If we're running under sanitizers, we need to use a less restrictive
+    # profile, otherwise LSan syscall would get blocked by seccomp
+    ARGS+=(--profile=trusted)
+fi
+
 export SYSTEMD_LOG_LEVEL=debug
 mkdir -p /run/systemd/system/systemd-portabled.service.d/
 cat <<EOF >/run/systemd/system/systemd-portabled.service.d/override.conf
@@ -11,7 +18,7 @@ cat <<EOF >/run/systemd/system/systemd-portabled.service.d/override.conf
 Environment=SYSTEMD_LOG_LEVEL=debug
 EOF
 
-portablectl attach --now --runtime /usr/share/minimal_0.raw app0
+portablectl "${ARGS[@]}" attach --now --runtime /usr/share/minimal_0.raw app0
 
 systemctl is-active app0.service
 systemctl is-active app0-foo.service
@@ -21,7 +28,7 @@ systemctl is-active app0-bar.service && exit 1
 set -e
 set -o pipefail
 
-portablectl reattach --now --runtime /usr/share/minimal_1.raw app0
+portablectl "${ARGS[@]}" reattach --now --runtime /usr/share/minimal_1.raw app0
 
 systemctl is-active app0.service
 systemctl is-active app0-bar.service
@@ -42,7 +49,7 @@ portablectl list | grep -q -F "No images."
 unsquashfs -dest /tmp/minimal_0 /usr/share/minimal_0.raw
 unsquashfs -dest /tmp/minimal_1 /usr/share/minimal_1.raw
 
-portablectl attach --copy=symlink --now --runtime /tmp/minimal_0 app0
+portablectl "${ARGS[@]}" attach --copy=symlink --now --runtime /tmp/minimal_0 app0
 
 systemctl is-active app0.service
 systemctl is-active app0-foo.service
@@ -52,7 +59,7 @@ systemctl is-active app0-bar.service && exit 1
 set -e
 set -o pipefail
 
-portablectl reattach --now --enable --runtime /tmp/minimal_1 app0
+portablectl "${ARGS[@]}" reattach --now --enable --runtime /tmp/minimal_1 app0
 
 systemctl is-active app0.service
 systemctl is-active app0-bar.service
@@ -68,21 +75,21 @@ portablectl detach --now --enable --runtime /tmp/minimal_1 app0
 
 portablectl list | grep -q -F "No images."
 
-portablectl attach --now --runtime --extension /usr/share/app0.raw /usr/share/minimal_0.raw app0
+portablectl "${ARGS[@]}" attach --now --runtime --extension /usr/share/app0.raw /usr/share/minimal_0.raw app0
 
 systemctl is-active app0.service
 
-portablectl reattach --now --runtime --extension /usr/share/app0.raw /usr/share/minimal_1.raw app0
+portablectl "${ARGS[@]}" reattach --now --runtime --extension /usr/share/app0.raw /usr/share/minimal_1.raw app0
 
 systemctl is-active app0.service
 
 portablectl detach --now --runtime --extension /usr/share/app0.raw /usr/share/minimal_1.raw app0
 
-portablectl attach --now --runtime --extension /usr/share/app1.raw /usr/share/minimal_0.raw app1
+portablectl "${ARGS[@]}" attach --now --runtime --extension /usr/share/app1.raw /usr/share/minimal_0.raw app1
 
 systemctl is-active app1.service
 
-portablectl reattach --now --runtime --extension /usr/share/app1.raw /usr/share/minimal_1.raw app1
+portablectl "${ARGS[@]}" reattach --now --runtime --extension /usr/share/app1.raw /usr/share/minimal_1.raw app1
 
 systemctl is-active app1.service
 
@@ -95,7 +102,7 @@ mount /usr/share/app1.raw /tmp/app1
 mount /usr/share/minimal_0.raw /tmp/rootdir
 mount -t overlay overlay -o lowerdir=/tmp/app1:/tmp/rootdir /tmp/overlay
 
-portablectl attach --copy=symlink --now --runtime /tmp/overlay app1
+portablectl "${ARGS[@]}" attach --copy=symlink --now --runtime /tmp/overlay app1
 
 systemctl is-active app1.service