Merge pull request #4185 from endocode/djalal-sandbox-first-protection-v1

core:sandbox: Add new ProtectKernelTunables=, ProtectControlGroups=, ProtectSystem=strict and fixes

8 files changed, 811 insertions, 195 deletions

diff --git a/src/core/dbus-execute.c b/src/core/dbus-execute.c
index 7e33a2d201..eec4500c8c 100644
--- a/src/core/dbus-execute.c
+++ b/src/core/dbus-execute.c
@@ -707,6 +707,8 @@ const sd_bus_vtable bus_exec_vtable[] = {
         SD_BUS_PROPERTY("MountFlags", "t", bus_property_get_ulong, offsetof(ExecContext, mount_flags), SD_BUS_VTABLE_PROPERTY_CONST),
         SD_BUS_PROPERTY("PrivateTmp", "b", bus_property_get_bool, offsetof(ExecContext, private_tmp), SD_BUS_VTABLE_PROPERTY_CONST),
         SD_BUS_PROPERTY("PrivateDevices", "b", bus_property_get_bool, offsetof(ExecContext, private_devices), SD_BUS_VTABLE_PROPERTY_CONST),
+        SD_BUS_PROPERTY("ProtectKernelTunables", "b", bus_property_get_bool, offsetof(ExecContext, protect_kernel_tunables), SD_BUS_VTABLE_PROPERTY_CONST),
+        SD_BUS_PROPERTY("ProtectControlGroups", "b", bus_property_get_bool, offsetof(ExecContext, protect_control_groups), SD_BUS_VTABLE_PROPERTY_CONST),
         SD_BUS_PROPERTY("PrivateNetwork", "b", bus_property_get_bool, offsetof(ExecContext, private_network), SD_BUS_VTABLE_PROPERTY_CONST),
         SD_BUS_PROPERTY("PrivateUsers", "b", bus_property_get_bool, offsetof(ExecContext, private_users), SD_BUS_VTABLE_PROPERTY_CONST),
         SD_BUS_PROPERTY("ProtectHome", "s", bus_property_get_protect_home, offsetof(ExecContext, protect_home), SD_BUS_VTABLE_PROPERTY_CONST),
@@ -1072,7 +1074,8 @@ int bus_exec_context_set_transient_property(
                               "IgnoreSIGPIPE", "TTYVHangup", "TTYReset",
                               "PrivateTmp", "PrivateDevices", "PrivateNetwork", "PrivateUsers",
                               "NoNewPrivileges", "SyslogLevelPrefix", "MemoryDenyWriteExecute",
-                              "RestrictRealtime", "DynamicUser", "RemoveIPC")) {
+                              "RestrictRealtime", "DynamicUser", "RemoveIPC", "ProtectKernelTunables",
+                              "ProtectControlGroups")) {
                 int b;
 
                 r = sd_bus_message_read(message, "b", &b);
@@ -1106,6 +1109,10 @@ int bus_exec_context_set_transient_property(
                                 c->dynamic_user = b;
                         else if (streq(name, "RemoveIPC"))
                                 c->remove_ipc = b;
+                        else if (streq(name, "ProtectKernelTunables"))
+                                c->protect_kernel_tunables = b;
+                        else if (streq(name, "ProtectControlGroups"))
+                                c->protect_control_groups = b;
 
                         unit_write_drop_in_private_format(u, mode, name, "%s=%s", name, yes_no(b));
                 }
diff --git a/src/core/execute.c b/src/core/execute.c
index 2026137721..3da7ef3be6 100644
--- a/src/core/execute.c
+++ b/src/core/execute.c
@@ -837,6 +837,8 @@ static int null_conv(
         return PAM_CONV_ERR;
 }
 
+#endif
+
 static int setup_pam(
                 const char *name,
                 const char *user,
@@ -845,6 +847,8 @@ static int setup_pam(
                 char ***env,
                 int fds[], unsigned n_fds) {
 
+#ifdef HAVE_PAM
+
         static const struct pam_conv conv = {
                 .conv = null_conv,
                 .appdata_ptr = NULL
@@ -1038,8 +1042,10 @@ fail:
         closelog();
 
         return r;
-}
+#else
+        return 0;
 #endif
+}
 
 static void rename_process_from_path(const char *path) {
         char process_name[11];
@@ -1273,6 +1279,10 @@ static int apply_memory_deny_write_execute(const Unit* u, const ExecContext *c)
         if (!seccomp)
                 return -ENOMEM;
 
+        r = seccomp_add_secondary_archs(seccomp);
+        if (r < 0)
+                goto finish;
+
         r = seccomp_rule_add(
                         seccomp,
                         SCMP_ACT_ERRNO(EPERM),
@@ -1322,6 +1332,10 @@ static int apply_restrict_realtime(const Unit* u, const ExecContext *c) {
         if (!seccomp)
                 return -ENOMEM;
 
+        r = seccomp_add_secondary_archs(seccomp);
+        if (r < 0)
+                goto finish;
+
         /* Determine the highest policy constant we want to allow */
         for (i = 0; i < ELEMENTSOF(permitted_policies); i++)
                 if (permitted_policies[i] > max_policy)
@@ -1375,12 +1389,121 @@ finish:
         return r;
 }
 
+static int apply_protect_sysctl(Unit *u, const ExecContext *c) {
+        scmp_filter_ctx *seccomp;
+        int r;
+
+        assert(c);
+
+        /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
+         * let's protect even those systems where this is left on in the kernel. */
+
+        if (skip_seccomp_unavailable(u, "ProtectKernelTunables="))
+                return 0;
+
+        seccomp = seccomp_init(SCMP_ACT_ALLOW);
+        if (!seccomp)
+                return -ENOMEM;
+
+        r = seccomp_add_secondary_archs(seccomp);
+        if (r < 0)
+                goto finish;
+
+        r = seccomp_rule_add(
+                        seccomp,
+                        SCMP_ACT_ERRNO(EPERM),
+                        SCMP_SYS(_sysctl),
+                        0);
+        if (r < 0)
+                goto finish;
+
+        r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
+        if (r < 0)
+                goto finish;
+
+        r = seccomp_load(seccomp);
+
+finish:
+        seccomp_release(seccomp);
+        return r;
+}
+
+static int apply_private_devices(Unit *u, const ExecContext *c) {
+        const SystemCallFilterSet *set;
+        scmp_filter_ctx *seccomp;
+        const char *sys;
+        bool syscalls_found = false;
+        int r;
+
+        assert(c);
+
+        /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
+
+        if (skip_seccomp_unavailable(u, "PrivateDevices="))
+                return 0;
+
+        seccomp = seccomp_init(SCMP_ACT_ALLOW);
+        if (!seccomp)
+                return -ENOMEM;
+
+        r = seccomp_add_secondary_archs(seccomp);
+        if (r < 0)
+                goto finish;
+
+        for (set = syscall_filter_sets; set->set_name; set++)
+                if (streq(set->set_name, "@raw-io")) {
+                        syscalls_found = true;
+                        break;
+                }
+
+        /* We should never fail here */
+        if (!syscalls_found) {
+                r = -EOPNOTSUPP;
+                goto finish;
+        }
+
+        NULSTR_FOREACH(sys, set->value) {
+                int id;
+                bool add = true;
+
+#ifndef __NR_s390_pci_mmio_read
+                if (streq(sys, "s390_pci_mmio_read"))
+                        add = false;
+#endif
+#ifndef __NR_s390_pci_mmio_write
+                if (streq(sys, "s390_pci_mmio_write"))
+                        add = false;
+#endif
+
+                if (!add)
+                        continue;
+
+                id = seccomp_syscall_resolve_name(sys);
+
+                r = seccomp_rule_add(
+                                seccomp,
+                                SCMP_ACT_ERRNO(EPERM),
+                                id, 0);
+                if (r < 0)
+                        goto finish;
+        }
+
+        r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
+        if (r < 0)
+                goto finish;
+
+        r = seccomp_load(seccomp);
+
+finish:
+        seccomp_release(seccomp);
+        return r;
+}
+
 #endif
 
 static void do_idle_pipe_dance(int idle_pipe[4]) {
         assert(idle_pipe);
 
-
         idle_pipe[1] = safe_close(idle_pipe[1]);
         idle_pipe[2] = safe_close(idle_pipe[2]);
 
@@ -1581,7 +1704,9 @@ static bool exec_needs_mount_namespace(
 
         if (context->private_devices ||
             context->protect_system != PROTECT_SYSTEM_NO ||
-            context->protect_home != PROTECT_HOME_NO)
+            context->protect_home != PROTECT_HOME_NO ||
+            context->protect_kernel_tunables ||
+            context->protect_control_groups)
                 return true;
 
         return false;
@@ -1740,6 +1865,111 @@ static int setup_private_users(uid_t uid, gid_t gid) {
         return 0;
 }
 
+static int setup_runtime_directory(
+                const ExecContext *context,
+                const ExecParameters *params,
+                uid_t uid,
+                gid_t gid) {
+
+        char **rt;
+        int r;
+
+        assert(context);
+        assert(params);
+
+        STRV_FOREACH(rt, context->runtime_directory) {
+                _cleanup_free_ char *p;
+
+                p = strjoin(params->runtime_prefix, "/", *rt, NULL);
+                if (!p)
+                        return -ENOMEM;
+
+                r = mkdir_p_label(p, context->runtime_directory_mode);
+                if (r < 0)
+                        return r;
+
+                r = chmod_and_chown(p, context->runtime_directory_mode, uid, gid);
+                if (r < 0)
+                        return r;
+        }
+
+        return 0;
+}
+
+static int setup_smack(
+                const ExecContext *context,
+                const ExecCommand *command) {
+
+#ifdef HAVE_SMACK
+        int r;
+
+        assert(context);
+        assert(command);
+
+        if (!mac_smack_use())
+                return 0;
+
+        if (context->smack_process_label) {
+                r = mac_smack_apply_pid(0, context->smack_process_label);
+                if (r < 0)
+                        return r;
+        }
+#ifdef SMACK_DEFAULT_PROCESS_LABEL
+        else {
+                _cleanup_free_ char *exec_label = NULL;
+
+                r = mac_smack_read(command->path, SMACK_ATTR_EXEC, &exec_label);
+                if (r < 0 && r != -ENODATA && r != -EOPNOTSUPP)
+                        return r;
+
+                r = mac_smack_apply_pid(0, exec_label ? : SMACK_DEFAULT_PROCESS_LABEL);
+                if (r < 0)
+                        return r;
+        }
+#endif
+#endif
+
+        return 0;
+}
+
+static int compile_read_write_paths(
+                const ExecContext *context,
+                const ExecParameters *params,
+                char ***ret) {
+
+        _cleanup_strv_free_ char **l = NULL;
+        char **rt;
+
+        /* Compile the list of writable paths. This is the combination of the explicitly configured paths, plus all
+         * runtime directories. */
+
+        if (strv_isempty(context->read_write_paths) &&
+            strv_isempty(context->runtime_directory)) {
+                *ret = NULL; /* NOP if neither is set */
+                return 0;
+        }
+
+        l = strv_copy(context->read_write_paths);
+        if (!l)
+                return -ENOMEM;
+
+        STRV_FOREACH(rt, context->runtime_directory) {
+                char *s;
+
+                s = strjoin(params->runtime_prefix, "/", *rt, NULL);
+                if (!s)
+                        return -ENOMEM;
+
+                if (strv_consume(&l, s) < 0)
+                        return -ENOMEM;
+        }
+
+        *ret = l;
+        l = NULL;
+
+        return 0;
+}
+
 static void append_socket_pair(int *array, unsigned *n, int pair[2]) {
         assert(array);
         assert(n);
@@ -1796,6 +2026,37 @@ static int close_remaining_fds(
         return close_all_fds(dont_close, n_dont_close);
 }
 
+static bool context_has_address_families(const ExecContext *c) {
+        assert(c);
+
+        return c->address_families_whitelist ||
+                !set_isempty(c->address_families);
+}
+
+static bool context_has_syscall_filters(const ExecContext *c) {
+        assert(c);
+
+        return c->syscall_whitelist ||
+                !set_isempty(c->syscall_filter) ||
+                !set_isempty(c->syscall_archs);
+}
+
+static bool context_has_no_new_privileges(const ExecContext *c) {
+        assert(c);
+
+        if (c->no_new_privileges)
+                return true;
+
+        if (have_effective_cap(CAP_SYS_ADMIN)) /* if we are privileged, we don't need NNP */
+                return false;
+
+        return context_has_address_families(c) || /* we need NNP if we have any form of seccomp and are unprivileged */
+                c->memory_deny_write_execute ||
+                c->restrict_realtime ||
+                c->protect_kernel_tunables ||
+                context_has_syscall_filters(c);
+}
+
 static int send_user_lookup(
                 Unit *unit,
                 int user_lookup_fd,
@@ -1940,22 +2201,14 @@ static int exec_child(
         } else {
                 if (context->user) {
                         username = context->user;
-                        r = get_user_creds(&username, &uid, &gid, &home, &shell);
+                        r = get_user_creds_clean(&username, &uid, &gid, &home, &shell);
                         if (r < 0) {
                                 *exit_status = EXIT_USER;
                                 return r;
                         }
 
-                        /* Don't set $HOME or $SHELL if they are are not particularly enlightening anyway. */
-                        if (isempty(home) || path_equal(home, "/"))
-                                home = NULL;
-
-                        if (isempty(shell) || PATH_IN_SET(shell,
-                                                          "/bin/nologin",
-                                                          "/sbin/nologin",
-                                                          "/usr/bin/nologin",
-                                                          "/usr/sbin/nologin"))
-                                shell = NULL;
+                        /* Note that we don't set $HOME or $SHELL if they are are not particularly enlightening anyway
+                         * (i.e. are "/" or "/bin/nologin"). */
                 }
 
                 if (context->group) {
@@ -2108,28 +2361,10 @@ static int exec_child(
         }
 
         if (!strv_isempty(context->runtime_directory) && params->runtime_prefix) {
-                char **rt;
-
-                STRV_FOREACH(rt, context->runtime_directory) {
-                        _cleanup_free_ char *p;
-
-                        p = strjoin(params->runtime_prefix, "/", *rt, NULL);
-                        if (!p) {
-                                *exit_status = EXIT_RUNTIME_DIRECTORY;
-                                return -ENOMEM;
-                        }
-
-                        r = mkdir_p_label(p, context->runtime_directory_mode);
-                        if (r < 0) {
-                                *exit_status = EXIT_RUNTIME_DIRECTORY;
-                                return r;
-                        }
-
-                        r = chmod_and_chown(p, context->runtime_directory_mode, uid, gid);
-                        if (r < 0) {
-                                *exit_status = EXIT_RUNTIME_DIRECTORY;
-                                return r;
-                        }
+                r = setup_runtime_directory(context, params, uid, gid);
+                if (r < 0) {
+                        *exit_status = EXIT_RUNTIME_DIRECTORY;
+                        return r;
                 }
         }
 
@@ -2168,41 +2403,15 @@ static int exec_child(
         }
         accum_env = strv_env_clean(accum_env);
 
-        umask(context->umask);
+        (void) umask(context->umask);
 
         if ((params->flags & EXEC_APPLY_PERMISSIONS) && !command->privileged) {
-                r = enforce_groups(context, username, gid);
+                r = setup_smack(context, command);
                 if (r < 0) {
-                        *exit_status = EXIT_GROUP;
+                        *exit_status = EXIT_SMACK_PROCESS_LABEL;
                         return r;
                 }
-#ifdef HAVE_SMACK
-                if (context->smack_process_label) {
-                        r = mac_smack_apply_pid(0, context->smack_process_label);
-                        if (r < 0) {
-                                *exit_status = EXIT_SMACK_PROCESS_LABEL;
-                                return r;
-                        }
-                }
-#ifdef SMACK_DEFAULT_PROCESS_LABEL
-                else {
-                        _cleanup_free_ char *exec_label = NULL;
 
-                        r = mac_smack_read(command->path, SMACK_ATTR_EXEC, &exec_label);
-                        if (r < 0 && r != -ENODATA && r != -EOPNOTSUPP) {
-                                *exit_status = EXIT_SMACK_PROCESS_LABEL;
-                                return r;
-                        }
-
-                        r = mac_smack_apply_pid(0, exec_label ? : SMACK_DEFAULT_PROCESS_LABEL);
-                        if (r < 0) {
-                                *exit_status = EXIT_SMACK_PROCESS_LABEL;
-                                return r;
-                        }
-                }
-#endif
-#endif
-#ifdef HAVE_PAM
                 if (context->pam_name && username) {
                         r = setup_pam(context->pam_name, username, uid, context->tty_path, &accum_env, fds, n_fds);
                         if (r < 0) {
@@ -2210,7 +2419,6 @@ static int exec_child(
                                 return r;
                         }
                 }
-#endif
         }
 
         if (context->private_network && runtime && runtime->netns_storage_socket[0] >= 0) {
@@ -2222,8 +2430,8 @@ static int exec_child(
         }
 
         needs_mount_namespace = exec_needs_mount_namespace(context, params, runtime);
-
         if (needs_mount_namespace) {
+                _cleanup_free_ char **rw = NULL;
                 char *tmp = NULL, *var = NULL;
 
                 /* The runtime struct only contains the parent
@@ -2239,14 +2447,22 @@ static int exec_child(
                                 var = strjoina(runtime->var_tmp_dir, "/tmp");
                 }
 
+                r = compile_read_write_paths(context, params, &rw);
+                if (r < 0) {
+                        *exit_status = EXIT_NAMESPACE;
+                        return r;
+                }
+
                 r = setup_namespace(
                                 (params->flags & EXEC_APPLY_CHROOT) ? context->root_directory : NULL,
-                                context->read_write_paths,
+                                rw,
                                 context->read_only_paths,
                                 context->inaccessible_paths,
                                 tmp,
                                 var,
                                 context->private_devices,
+                                context->protect_kernel_tunables,
+                                context->protect_control_groups,
                                 context->protect_home,
                                 context->protect_system,
                                 context->mount_flags);
@@ -2264,6 +2480,14 @@ static int exec_child(
                 }
         }
 
+        if ((params->flags & EXEC_APPLY_PERMISSIONS) && !command->privileged) {
+                r = enforce_groups(context, username, gid);
+                if (r < 0) {
+                        *exit_status = EXIT_GROUP;
+                        return r;
+                }
+        }
+
         if (context->working_directory_home)
                 wd = home;
         else if (context->working_directory)
@@ -2335,11 +2559,6 @@ static int exec_child(
 
         if ((params->flags & EXEC_APPLY_PERMISSIONS) && !command->privileged) {
 
-                bool use_address_families = context->address_families_whitelist ||
-                        !set_isempty(context->address_families);
-                bool use_syscall_filter = context->syscall_whitelist ||
-                        !set_isempty(context->syscall_filter) ||
-                        !set_isempty(context->syscall_archs);
                 int secure_bits = context->secure_bits;
 
                 for (i = 0; i < _RLIMIT_MAX; i++) {
@@ -2416,15 +2635,14 @@ static int exec_child(
                                 return -errno;
                         }
 
-                if (context->no_new_privileges ||
-                    (!have_effective_cap(CAP_SYS_ADMIN) && (use_address_families || context->memory_deny_write_execute || context->restrict_realtime || use_syscall_filter)))
+                if (context_has_no_new_privileges(context))
                         if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
                                 *exit_status = EXIT_NO_NEW_PRIVILEGES;
                                 return -errno;
                         }
 
 #ifdef HAVE_SECCOMP
-                if (use_address_families) {
+                if (context_has_address_families(context)) {
                         r = apply_address_families(unit, context);
                         if (r < 0) {
                                 *exit_status = EXIT_ADDRESS_FAMILIES;
@@ -2448,7 +2666,23 @@ static int exec_child(
                         }
                 }
 
-                if (use_syscall_filter) {
+                if (context->protect_kernel_tunables) {
+                        r = apply_protect_sysctl(unit, context);
+                        if (r < 0) {
+                                *exit_status = EXIT_SECCOMP;
+                                return r;
+                        }
+                }
+
+                if (context->private_devices) {
+                        r = apply_private_devices(unit, context);
+                        if (r < 0) {
+                                *exit_status = EXIT_SECCOMP;
+                                return r;
+                        }
+                }
+
+                if (context_has_syscall_filters(context)) {
                         r = apply_seccomp(unit, context);
                         if (r < 0) {
                                 *exit_status = EXIT_SECCOMP;
@@ -2880,6 +3114,8 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
                 "%sNonBlocking: %s\n"
                 "%sPrivateTmp: %s\n"
                 "%sPrivateDevices: %s\n"
+                "%sProtectKernelTunables: %s\n"
+                "%sProtectControlGroups: %s\n"
                 "%sPrivateNetwork: %s\n"
                 "%sPrivateUsers: %s\n"
                 "%sProtectHome: %s\n"
@@ -2893,6 +3129,8 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
                 prefix, yes_no(c->non_blocking),
                 prefix, yes_no(c->private_tmp),
                 prefix, yes_no(c->private_devices),
+                prefix, yes_no(c->protect_kernel_tunables),
+                prefix, yes_no(c->protect_control_groups),
                 prefix, yes_no(c->private_network),
                 prefix, yes_no(c->private_users),
                 prefix, protect_home_to_string(c->protect_home),
diff --git a/src/core/execute.h b/src/core/execute.h
index 6082c42aba..449180c903 100644
--- a/src/core/execute.h
+++ b/src/core/execute.h
@@ -174,6 +174,8 @@ struct ExecContext {
         bool private_users;
         ProtectSystem protect_system;
         ProtectHome protect_home;
+        bool protect_kernel_tunables;
+        bool protect_control_groups;
 
         bool no_new_privileges;
 
diff --git a/src/core/load-fragment-gperf.gperf.m4 b/src/core/load-fragment-gperf.gperf.m4
index 2e6c965aec..c49c1d6732 100644
--- a/src/core/load-fragment-gperf.gperf.m4
+++ b/src/core/load-fragment-gperf.gperf.m4
@@ -89,6 +89,8 @@ $1.ReadOnlyPaths,                config_parse_namespace_path_strv,   0,
 $1.InaccessiblePaths,            config_parse_namespace_path_strv,   0,                             offsetof($1, exec_context.inaccessible_paths)
 $1.PrivateTmp,                   config_parse_bool,                  0,                             offsetof($1, exec_context.private_tmp)
 $1.PrivateDevices,               config_parse_bool,                  0,                             offsetof($1, exec_context.private_devices)
+$1.ProtectKernelTunables,        config_parse_bool,                  0,                             offsetof($1, exec_context.protect_kernel_tunables)
+$1.ProtectControlGroups,         config_parse_bool,                  0,                             offsetof($1, exec_context.protect_control_groups)
 $1.PrivateNetwork,               config_parse_bool,                  0,                             offsetof($1, exec_context.private_network)
 $1.PrivateUsers,                 config_parse_bool,                  0,                             offsetof($1, exec_context.private_users)
 $1.ProtectSystem,                config_parse_protect_system,        0,                             offsetof($1, exec_context)
diff --git a/src/core/main.c b/src/core/main.c
index 803307c9d5..be0cb0b6d1 100644
--- a/src/core/main.c
+++ b/src/core/main.c
@@ -996,10 +996,8 @@ static int parse_argv(int argc, char *argv[]) {
 
                 case ARG_MACHINE_ID:
                         r = set_machine_id(optarg);
-                        if (r < 0) {
-                                log_error("MachineID '%s' is not valid.", optarg);
-                                return r;
-                        }
+                        if (r < 0)
+                                return log_error_errno(r, "MachineID '%s' is not valid.", optarg);
                         break;
 
                 case 'h':
diff --git a/src/core/namespace.c b/src/core/namespace.c
index 52a2505d94..43a2f4ba6e 100644
--- a/src/core/namespace.c
+++ b/src/core/namespace.c
@@ -29,6 +29,7 @@
 #include "alloc-util.h"
 #include "dev-setup.h"
 #include "fd-util.h"
+#include "fs-util.h"
 #include "loopback-setup.h"
 #include "missing.h"
 #include "mkdir.h"
@@ -53,61 +54,230 @@ typedef enum MountMode {
         PRIVATE_TMP,
         PRIVATE_VAR_TMP,
         PRIVATE_DEV,
-        READWRITE
+        READWRITE,
 } MountMode;
 
 typedef struct BindMount {
-        const char *path;
+        const char *path; /* stack memory, doesn't need to be freed explicitly */
+        char *chased; /* malloc()ed memory, needs to be freed */
         MountMode mode;
-        bool done;
-        bool ignore;
+        bool ignore; /* Ignore if path does not exist */
 } BindMount;
 
+typedef struct TargetMount {
+        const char *path;
+        MountMode mode;
+        bool ignore; /* Ignore if path does not exist */
+} TargetMount;
+
+/*
+ * The following Protect tables are to protect paths and mark some of them
+ * READONLY, in case a path is covered by an option from another table, then
+ * it is marked READWRITE in the current one, and the more restrictive mode is
+ * applied from that other table. This way all options can be combined in a
+ * safe and comprehensible way for users.
+ */
+
+/* ProtectKernelTunables= option and the related filesystem APIs */
+static const TargetMount protect_kernel_tunables_table[] = {
+        { "/proc/sys",                  READONLY,       false },
+        { "/proc/sysrq-trigger",        READONLY,       true  },
+        { "/proc/latency_stats",        READONLY,       true  },
+        { "/proc/mtrr",                 READONLY,       true  },
+        { "/proc/apm",                  READONLY,       true  },
+        { "/proc/acpi",                 READONLY,       true  },
+        { "/proc/timer_stats",          READONLY,       true  },
+        { "/proc/asound",               READONLY,       true  },
+        { "/proc/bus",                  READONLY,       true  },
+        { "/proc/fs",                   READONLY,       true  },
+        { "/proc/irq",                  READONLY,       true  },
+        { "/sys",                       READONLY,       false },
+        { "/sys/kernel/debug",          READONLY,       true  },
+        { "/sys/kernel/tracing",        READONLY,       true  },
+        { "/sys/fs/cgroup",             READWRITE,      false }, /* READONLY is set by ProtectControlGroups= option */
+};
+
+/*
+ * ProtectHome=read-only table, protect $HOME and $XDG_RUNTIME_DIR and rest of
+ * system should be protected by ProtectSystem=
+ */
+static const TargetMount protect_home_read_only_table[] = {
+        { "/home",      READONLY,       true },
+        { "/run/user",  READONLY,       true },
+        { "/root",      READONLY,       true },
+};
+
+/* ProtectHome=yes table */
+static const TargetMount protect_home_yes_table[] = {
+        { "/home",      INACCESSIBLE,   true },
+        { "/run/user",  INACCESSIBLE,   true },
+        { "/root",      INACCESSIBLE,   true },
+};
+
+/* ProtectSystem=yes table */
+static const TargetMount protect_system_yes_table[] = {
+        { "/usr",       READONLY,       false },
+        { "/boot",      READONLY,       true  },
+        { "/efi",       READONLY,       true  },
+};
+
+/* ProtectSystem=full includes ProtectSystem=yes */
+static const TargetMount protect_system_full_table[] = {
+        { "/usr",       READONLY,       false },
+        { "/boot",      READONLY,       true  },
+        { "/efi",       READONLY,       true  },
+        { "/etc",       READONLY,       false },
+};
+
+/*
+ * ProtectSystem=strict table. In this strict mode, we mount everything
+ * read-only, except for /proc, /dev, /sys which are the kernel API VFS,
+ * which are left writable, but PrivateDevices= + ProtectKernelTunables=
+ * protect those, and these options should be fully orthogonal.
+ * (And of course /home and friends are also left writable, as ProtectHome=
+ * shall manage those, orthogonally).
+ */
+static const TargetMount protect_system_strict_table[] = {
+        { "/",          READONLY,       false },
+        { "/proc",      READWRITE,      false },      /* ProtectKernelTunables= */
+        { "/sys",       READWRITE,      false },      /* ProtectKernelTunables= */
+        { "/dev",       READWRITE,      false },      /* PrivateDevices= */
+        { "/home",      READWRITE,      true  },      /* ProtectHome= */
+        { "/run/user",  READWRITE,      true  },      /* ProtectHome= */
+        { "/root",      READWRITE,      true  },      /* ProtectHome= */
+};
+
+static void set_bind_mount(BindMount **p, const char *path, MountMode mode, bool ignore) {
+        (*p)->path = path;
+        (*p)->mode = mode;
+        (*p)->ignore = ignore;
+}
+
 static int append_mounts(BindMount **p, char **strv, MountMode mode) {
         char **i;
 
         assert(p);
 
         STRV_FOREACH(i, strv) {
+                bool ignore = false;
 
-                (*p)->ignore = false;
-                (*p)->done = false;
-
-                if ((mode == INACCESSIBLE || mode == READONLY || mode == READWRITE) && (*i)[0] == '-') {
-                        (*p)->ignore = true;
+                if (IN_SET(mode, INACCESSIBLE, READONLY, READWRITE) && startswith(*i, "-")) {
                         (*i)++;
+                        ignore = true;
                 }
 
                 if (!path_is_absolute(*i))
                         return -EINVAL;
 
-                (*p)->path = *i;
-                (*p)->mode = mode;
+                set_bind_mount(p, *i, mode, ignore);
                 (*p)++;
         }
 
         return 0;
 }
 
-static int mount_path_compare(const void *a, const void *b) {
-        const BindMount *p = a, *q = b;
-        int d;
+static int append_target_mounts(BindMount **p, const char *root_directory, const TargetMount *mounts, const size_t size) {
+        unsigned i;
 
-        d = path_compare(p->path, q->path);
+        assert(p);
+        assert(mounts);
+
+        for (i = 0; i < size; i++) {
+                /*
+                 * Here we assume that the ignore field is set during
+                 * declaration we do not support "-" at the beginning.
+                 */
+                const TargetMount *m = &mounts[i];
+                const char *path = prefix_roota(root_directory, m->path);
+
+                if (!path_is_absolute(path))
+                        return -EINVAL;
+
+                set_bind_mount(p, path, m->mode, m->ignore);
+                (*p)++;
+        }
+
+        return 0;
+}
+
+static int append_protect_kernel_tunables(BindMount **p, const char *root_directory) {
+        assert(p);
+
+        return append_target_mounts(p, root_directory, protect_kernel_tunables_table,
+                                    ELEMENTSOF(protect_kernel_tunables_table));
+}
 
-        if (d == 0) {
-                /* If the paths are equal, check the mode */
-                if (p->mode < q->mode)
-                        return -1;
+static int append_protect_home(BindMount **p, const char *root_directory, ProtectHome protect_home) {
+        int r = 0;
 
-                if (p->mode > q->mode)
-                        return 1;
+        assert(p);
 
+        if (protect_home == PROTECT_HOME_NO)
                 return 0;
+
+        switch (protect_home) {
+        case PROTECT_HOME_READ_ONLY:
+                r = append_target_mounts(p, root_directory, protect_home_read_only_table,
+                                         ELEMENTSOF(protect_home_read_only_table));
+                break;
+        case PROTECT_HOME_YES:
+                r = append_target_mounts(p, root_directory, protect_home_yes_table,
+                                         ELEMENTSOF(protect_home_yes_table));
+                break;
+        default:
+                r = -EINVAL;
+                break;
         }
 
+        return r;
+}
+
+static int append_protect_system(BindMount **p, const char *root_directory, ProtectSystem protect_system) {
+        int r = 0;
+
+        assert(p);
+
+        if (protect_system == PROTECT_SYSTEM_NO)
+                return 0;
+
+        switch (protect_system) {
+        case PROTECT_SYSTEM_STRICT:
+                r = append_target_mounts(p, root_directory, protect_system_strict_table,
+                                         ELEMENTSOF(protect_system_strict_table));
+                break;
+        case PROTECT_SYSTEM_YES:
+                r = append_target_mounts(p, root_directory, protect_system_yes_table,
+                                         ELEMENTSOF(protect_system_yes_table));
+                break;
+        case PROTECT_SYSTEM_FULL:
+                r = append_target_mounts(p, root_directory, protect_system_full_table,
+                                         ELEMENTSOF(protect_system_full_table));
+                break;
+        default:
+                r = -EINVAL;
+                break;
+        }
+
+        return r;
+}
+
+static int mount_path_compare(const void *a, const void *b) {
+        const BindMount *p = a, *q = b;
+        int d;
+
         /* If the paths are not equal, then order prefixes first */
-        return d;
+        d = path_compare(p->path, q->path);
+        if (d != 0)
+                return d;
+
+        /* If the paths are equal, check the mode */
+        if (p->mode < q->mode)
+                return -1;
+
+        if (p->mode > q->mode)
+                return 1;
+
+        return 0;
 }
 
 static void drop_duplicates(BindMount *m, unsigned *n) {
@@ -116,16 +286,110 @@ static void drop_duplicates(BindMount *m, unsigned *n) {
         assert(m);
         assert(n);
 
+        /* Drops duplicate entries. Expects that the array is properly ordered already. */
+
         for (f = m, t = m, previous = NULL; f < m+*n; f++) {
 
-                /* The first one wins */
-                if (previous && path_equal(f->path, previous->path))
+                /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare()
+                 * above. */
+                if (previous && path_equal(f->path, previous->path)) {
+                        log_debug("%s is duplicate.", f->path);
                         continue;
+                }
 
                 *t = *f;
-
                 previous = t;
+                t++;
+        }
+
+        *n = t - m;
+}
+
+static void drop_inaccessible(BindMount *m, unsigned *n) {
+        BindMount *f, *t;
+        const char *clear = NULL;
+
+        assert(m);
+        assert(n);
+
+        /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly
+         * ordered already. */
 
+        for (f = m, t = m; f < m+*n; f++) {
+
+                /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop
+                 * it, as inaccessible paths really should drop the entire subtree. */
+                if (clear && path_startswith(f->path, clear)) {
+                        log_debug("%s is masked by %s.", f->path, clear);
+                        continue;
+                }
+
+                clear = f->mode == INACCESSIBLE ? f->path : NULL;
+
+                *t = *f;
+                t++;
+        }
+
+        *n = t - m;
+}
+
+static void drop_nop(BindMount *m, unsigned *n) {
+        BindMount *f, *t;
+
+        assert(m);
+        assert(n);
+
+        /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the
+         * list is ordered by prefixes. */
+
+        for (f = m, t = m; f < m+*n; f++) {
+
+                /* Only suppress such subtrees for READONLY and READWRITE entries */
+                if (IN_SET(f->mode, READONLY, READWRITE)) {
+                        BindMount *p;
+                        bool found = false;
+
+                        /* Now let's find the first parent of the entry we are looking at. */
+                        for (p = t-1; p >= m; p--) {
+                                if (path_startswith(f->path, p->path)) {
+                                        found = true;
+                                        break;
+                                }
+                        }
+
+                        /* We found it, let's see if it's the same mode, if so, we can drop this entry */
+                        if (found && p->mode == f->mode) {
+                                log_debug("%s is redundant by %s", f->path, p->path);
+                                continue;
+                        }
+                }
+
+                *t = *f;
+                t++;
+        }
+
+        *n = t - m;
+}
+
+static void drop_outside_root(const char *root_directory, BindMount *m, unsigned *n) {
+        BindMount *f, *t;
+
+        assert(m);
+        assert(n);
+
+        if (!root_directory)
+                return;
+
+        /* Drops all mounts that are outside of the root directory. */
+
+        for (f = m, t = m; f < m+*n; f++) {
+
+                if (!path_startswith(f->path, root_directory)) {
+                        log_debug("%s is outside of root directory.", f->path);
+                        continue;
+                }
+
+                *t = *f;
                 t++;
         }
 
@@ -278,24 +542,23 @@ static int apply_mount(
 
         const char *what;
         int r;
-        struct stat target;
 
         assert(m);
 
+        log_debug("Applying namespace mount on %s", m->path);
+
         switch (m->mode) {
 
-        case INACCESSIBLE:
+        case INACCESSIBLE: {
+                struct stat target;
 
                 /* First, get rid of everything that is below if there
                  * is anything... Then, overmount it with an
                  * inaccessible path. */
-                umount_recursive(m->path, 0);
+                (void) umount_recursive(m->path, 0);
 
-                if (lstat(m->path, &target) < 0) {
-                        if (m->ignore && errno == ENOENT)
-                                return 0;
-                        return -errno;
-                }
+                if (lstat(m->path, &target) < 0)
+                        return log_debug_errno(errno, "Failed to lstat() %s to determine what to mount over it: %m", m->path);
 
                 what = mode_to_inaccessible_node(target.st_mode);
                 if (!what) {
@@ -303,11 +566,20 @@ static int apply_mount(
                         return -ELOOP;
                 }
                 break;
+        }
+
         case READONLY:
         case READWRITE:
-                /* Nothing to mount here, we just later toggle the
-                 * MS_RDONLY bit for the mount point */
-                return 0;
+
+                r = path_is_mount_point(m->path, 0);
+                if (r < 0)
+                        return log_debug_errno(r, "Failed to determine whether %s is already a mount point: %m", m->path);
+                if (r > 0) /* Nothing to do here, it is already a mount. We just later toggle the MS_RDONLY bit for the mount point if needed. */
+                        return 0;
+
+                /* This isn't a mount point yet, let's make it one. */
+                what = m->path;
+                break;
 
         case PRIVATE_TMP:
                 what = tmp_dir;
@@ -326,38 +598,104 @@ static int apply_mount(
 
         assert(what);
 
-        r = mount(what, m->path, NULL, MS_BIND|MS_REC, NULL);
-        if (r >= 0) {
-                log_debug("Successfully mounted %s to %s", what, m->path);
-                return r;
-        } else {
-                if (m->ignore && errno == ENOENT)
-                        return 0;
+        if (mount(what, m->path, NULL, MS_BIND|MS_REC, NULL) < 0)
                 return log_debug_errno(errno, "Failed to mount %s to %s: %m", what, m->path);
-        }
+
+        log_debug("Successfully mounted %s to %s", what, m->path);
+        return 0;
 }
 
-static int make_read_only(BindMount *m) {
-        int r;
+static int make_read_only(BindMount *m, char **blacklist) {
+        int r = 0;
 
         assert(m);
 
         if (IN_SET(m->mode, INACCESSIBLE, READONLY))
-                r = bind_remount_recursive(m->path, true);
-        else if (IN_SET(m->mode, READWRITE, PRIVATE_TMP, PRIVATE_VAR_TMP, PRIVATE_DEV)) {
-                r = bind_remount_recursive(m->path, false);
-                if (r == 0 && m->mode == PRIVATE_DEV) /* can be readonly but the submounts can't*/
-                        if (mount(NULL, m->path, NULL, MS_REMOUNT|DEV_MOUNT_OPTIONS|MS_RDONLY, NULL) < 0)
-                                r = -errno;
+                r = bind_remount_recursive(m->path, true, blacklist);
+        else if (m->mode == PRIVATE_DEV) { /* Can be readonly but the submounts can't*/
+                if (mount(NULL, m->path, NULL, MS_REMOUNT|DEV_MOUNT_OPTIONS|MS_RDONLY, NULL) < 0)
+                        r = -errno;
         } else
-                r = 0;
-
-        if (m->ignore && r == -ENOENT)
                 return 0;
 
+        /* Not that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked read-only
+         * already stays this way. This improves compatibility with container managers, where we won't attempt to undo
+         * read-only mounts already applied. */
+
         return r;
 }
 
+static int chase_all_symlinks(const char *root_directory, BindMount *m, unsigned *n) {
+        BindMount *f, *t;
+        int r;
+
+        assert(m);
+        assert(n);
+
+        /* Since mount() will always follow symlinks and we need to take the different root directory into account we
+         * chase the symlinks on our own first. This call wil do so for all entries and remove all entries where we
+         * can't resolve the path, and which have been marked for such removal. */
+
+        for (f = m, t = m; f < m+*n; f++) {
+
+                r = chase_symlinks(f->path, root_directory, &f->chased);
+                if (r == -ENOENT && f->ignore) /* Doesn't exist? Then remove it! */
+                        continue;
+                if (r < 0)
+                        return log_debug_errno(r, "Failed to chase symlinks for %s: %m", f->path);
+
+                if (path_equal(f->path, f->chased))
+                        f->chased = mfree(f->chased);
+                else {
+                        log_debug("Chased %s → %s", f->path, f->chased);
+                        f->path = f->chased;
+                }
+
+                *t = *f;
+                t++;
+        }
+
+        *n = t - m;
+        return 0;
+}
+
+static unsigned namespace_calculate_mounts(
+                char** read_write_paths,
+                char** read_only_paths,
+                char** inaccessible_paths,
+                const char* tmp_dir,
+                const char* var_tmp_dir,
+                bool private_dev,
+                bool protect_sysctl,
+                bool protect_cgroups,
+                ProtectHome protect_home,
+                ProtectSystem protect_system) {
+
+        unsigned protect_home_cnt;
+        unsigned protect_system_cnt =
+                (protect_system == PROTECT_SYSTEM_STRICT ?
+                 ELEMENTSOF(protect_system_strict_table) :
+                 ((protect_system == PROTECT_SYSTEM_FULL) ?
+                  ELEMENTSOF(protect_system_full_table) :
+                  ((protect_system == PROTECT_SYSTEM_YES) ?
+                   ELEMENTSOF(protect_system_yes_table) : 0)));
+
+        protect_home_cnt =
+                (protect_home == PROTECT_HOME_YES ?
+                 ELEMENTSOF(protect_home_yes_table) :
+                 ((protect_home == PROTECT_HOME_READ_ONLY) ?
+                  ELEMENTSOF(protect_home_read_only_table) : 0));
+
+        return !!tmp_dir + !!var_tmp_dir +
+                strv_length(read_write_paths) +
+                strv_length(read_only_paths) +
+                strv_length(inaccessible_paths) +
+                private_dev +
+                (protect_sysctl ? ELEMENTSOF(protect_kernel_tunables_table) : 0) +
+                (protect_cgroups ? 1 : 0) +
+                protect_home_cnt + protect_system_cnt;
+}
+
 int setup_namespace(
                 const char* root_directory,
                 char** read_write_paths,
@@ -366,28 +704,31 @@ int setup_namespace(
                 const char* tmp_dir,
                 const char* var_tmp_dir,
                 bool private_dev,
+                bool protect_sysctl,
+                bool protect_cgroups,
                 ProtectHome protect_home,
                 ProtectSystem protect_system,
                 unsigned long mount_flags) {
 
         BindMount *m, *mounts = NULL;
+        bool make_slave = false;
         unsigned n;
         int r = 0;
 
         if (mount_flags == 0)
                 mount_flags = MS_SHARED;
 
-        if (unshare(CLONE_NEWNS) < 0)
-                return -errno;
+        n = namespace_calculate_mounts(read_write_paths,
+                                       read_only_paths,
+                                       inaccessible_paths,
+                                       tmp_dir, var_tmp_dir,
+                                       private_dev, protect_sysctl,
+                                       protect_cgroups, protect_home,
+                                       protect_system);
 
-        n = !!tmp_dir + !!var_tmp_dir +
-                strv_length(read_write_paths) +
-                strv_length(read_only_paths) +
-                strv_length(inaccessible_paths) +
-                private_dev +
-                (protect_home != PROTECT_HOME_NO ? 3 : 0) +
-                (protect_system != PROTECT_SYSTEM_NO ? 2 : 0) +
-                (protect_system == PROTECT_SYSTEM_FULL ? 1 : 0);
+        /* Set mount slave mode */
+        if (root_directory || n > 0)
+                make_slave = true;
 
         if (n > 0) {
                 m = mounts = (BindMount *) alloca0(n * sizeof(BindMount));
@@ -421,94 +762,112 @@ int setup_namespace(
                         m++;
                 }
 
-                if (protect_home != PROTECT_HOME_NO) {
-                        const char *home_dir, *run_user_dir, *root_dir;
+                if (protect_sysctl)
+                        append_protect_kernel_tunables(&m, root_directory);
 
-                        home_dir = prefix_roota(root_directory, "/home");
-                        home_dir = strjoina("-", home_dir);
-                        run_user_dir = prefix_roota(root_directory, "/run/user");
-                        run_user_dir = strjoina("-", run_user_dir);
-                        root_dir = prefix_roota(root_directory, "/root");
-                        root_dir = strjoina("-", root_dir);
-
-                        r = append_mounts(&m, STRV_MAKE(home_dir, run_user_dir, root_dir),
-                                protect_home == PROTECT_HOME_READ_ONLY ? READONLY : INACCESSIBLE);
-                        if (r < 0)
-                                return r;
+                if (protect_cgroups) {
+                        m->path = prefix_roota(root_directory, "/sys/fs/cgroup");
+                        m->mode = READONLY;
+                        m++;
                 }
 
-                if (protect_system != PROTECT_SYSTEM_NO) {
-                        const char *usr_dir, *boot_dir, *etc_dir;
-
-                        usr_dir = prefix_roota(root_directory, "/usr");
-                        boot_dir = prefix_roota(root_directory, "/boot");
-                        boot_dir = strjoina("-", boot_dir);
-                        etc_dir = prefix_roota(root_directory, "/etc");
+                r = append_protect_home(&m, root_directory, protect_home);
+                if (r < 0)
+                        return r;
 
-                        r = append_mounts(&m, protect_system == PROTECT_SYSTEM_FULL
-                                ? STRV_MAKE(usr_dir, boot_dir, etc_dir)
-                                : STRV_MAKE(usr_dir, boot_dir), READONLY);
-                        if (r < 0)
-                                return r;
-                }
+                r = append_protect_system(&m, root_directory, protect_system);
+                if (r < 0)
+                        return r;
 
                 assert(mounts + n == m);
 
+                /* Resolve symlinks manually first, as mount() will always follow them relative to the host's
+                 * root. Moreover we want to suppress duplicates based on the resolved paths. This of course is a bit
+                 * racy. */
+                r = chase_all_symlinks(root_directory, mounts, &n);
+                if (r < 0)
+                        goto finish;
+
                 qsort(mounts, n, sizeof(BindMount), mount_path_compare);
+
                 drop_duplicates(mounts, &n);
+                drop_outside_root(root_directory, mounts, &n);
+                drop_inaccessible(mounts, &n);
+                drop_nop(mounts, &n);
+        }
+
+        if (unshare(CLONE_NEWNS) < 0) {
+                r = -errno;
+                goto finish;
         }
 
-        if (n > 0 || root_directory) {
+        if (make_slave) {
                 /* Remount / as SLAVE so that nothing now mounted in the namespace
                    shows up in the parent */
-                if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0)
-                        return -errno;
+                if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0) {
+                        r = -errno;
+                        goto finish;
+                }
         }
 
         if (root_directory) {
-                /* Turn directory into bind mount */
-                if (mount(root_directory, root_directory, NULL, MS_BIND|MS_REC, NULL) < 0)
-                        return -errno;
+                /* Turn directory into bind mount, if it isn't one yet */
+                r = path_is_mount_point(root_directory, AT_SYMLINK_FOLLOW);
+                if (r < 0)
+                        goto finish;
+                if (r == 0) {
+                        if (mount(root_directory, root_directory, NULL, MS_BIND|MS_REC, NULL) < 0) {
+                                r = -errno;
+                                goto finish;
+                        }
+                }
         }
 
         if (n > 0) {
+                char **blacklist;
+                unsigned j;
+
+                /* First round, add in all special mounts we need */
                 for (m = mounts; m < mounts + n; ++m) {
                         r = apply_mount(m, tmp_dir, var_tmp_dir);
                         if (r < 0)
-                                goto fail;
+                                goto finish;
                 }
 
+                /* Create a blacklist we can pass to bind_mount_recursive() */
+                blacklist = newa(char*, n+1);
+                for (j = 0; j < n; j++)
+                        blacklist[j] = (char*) mounts[j].path;
+                blacklist[j] = NULL;
+
+                /* Second round, flip the ro bits if necessary. */
                 for (m = mounts; m < mounts + n; ++m) {
-                        r = make_read_only(m);
+                        r = make_read_only(m, blacklist);
                         if (r < 0)
-                                goto fail;
+                                goto finish;
                 }
         }
 
         if (root_directory) {
                 /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
                 r = mount_move_root(root_directory);
-
-                /* at this point, we cannot rollback */
                 if (r < 0)
-                        return r;
+                        goto finish;
         }
 
         /* Remount / as the desired mode. Not that this will not
          * reestablish propagation from our side to the host, since
          * what's disconnected is disconnected. */
-        if (mount(NULL, "/", NULL, mount_flags | MS_REC, NULL) < 0)
-                /* at this point, we cannot rollback */
-                return -errno;
+        if (mount(NULL, "/", NULL, mount_flags | MS_REC, NULL) < 0) {
+                r = -errno;
+                goto finish;
+        }
 
-        return 0;
+        r = 0;
 
-fail:
-        if (n > 0) {
-                for (m = mounts; m < mounts + n; ++m)
-                        if (m->done)
-                                (void) umount2(m->path, MNT_DETACH);
-        }
+finish:
+        for (m = mounts; m < mounts + n; m++)
+                free(m->chased);
 
         return r;
 }
@@ -658,6 +1017,7 @@ static const char *const protect_system_table[_PROTECT_SYSTEM_MAX] = {
         [PROTECT_SYSTEM_NO] = "no",
         [PROTECT_SYSTEM_YES] = "yes",
         [PROTECT_SYSTEM_FULL] = "full",
+        [PROTECT_SYSTEM_STRICT] = "strict",
 };
 
 DEFINE_STRING_TABLE_LOOKUP(protect_system, ProtectSystem);
diff --git a/src/core/namespace.h b/src/core/namespace.h
index 1aedf5f208..6505bcc499 100644
--- a/src/core/namespace.h
+++ b/src/core/namespace.h
@@ -35,6 +35,7 @@ typedef enum ProtectSystem {
         PROTECT_SYSTEM_NO,
         PROTECT_SYSTEM_YES,
         PROTECT_SYSTEM_FULL,
+        PROTECT_SYSTEM_STRICT,
         _PROTECT_SYSTEM_MAX,
         _PROTECT_SYSTEM_INVALID = -1
 } ProtectSystem;
@@ -46,6 +47,8 @@ int setup_namespace(const char *chroot,
                     const char *tmp_dir,
                     const char *var_tmp_dir,
                     bool private_dev,
+                    bool protect_sysctl,
+                    bool protect_cgroups,
                     ProtectHome protect_home,
                     ProtectSystem protect_system,
                     unsigned long mount_flags);
diff --git a/src/core/unit.c b/src/core/unit.c
index de22f657c6..5d284a359d 100644
--- a/src/core/unit.c
+++ b/src/core/unit.c
@@ -3377,8 +3377,14 @@ int unit_patch_contexts(Unit *u) {
                                         return -ENOMEM;
                         }
 
+                        /* If the dynamic user option is on, let's make sure that the unit can't leave its UID/GID
+                         * around in the file system or on IPC objects. Hence enforce a strict sandbox. */
+
                         ec->private_tmp = true;
                         ec->remove_ipc = true;
+                        ec->protect_system = PROTECT_SYSTEM_STRICT;
+                        if (ec->protect_home == PROTECT_HOME_NO)
+                                ec->protect_home = PROTECT_HOME_READ_ONLY;
                 }
         }