core: general cgroup rework

Replace the very generic cgroup hookup with a much simpler one. With
this change only the high-level cgroup settings remain, the ability to
set arbitrary cgroup attributes is removed, so is support for adding
units to arbitrary cgroup controllers or setting arbitrary paths for
them (especially paths that are different for the various controllers).

This also introduces a new -.slice root slice, that is the parent of
system.slice and friends. This enables easy admin configuration of
root-level cgrouo properties.

This replaces DeviceDeny= by DevicePolicy=, and implicitly adds in
/dev/null, /dev/zero and friends if DeviceAllow= is used (unless this is
turned off by DevicePolicy=).

1 files changed, 39 insertions, 22 deletions

diff --git a/src/core/service.c b/src/core/service.c
index a0c648a85b..5fdbdb13a3 100644
--- a/src/core/service.c
+++ b/src/core/service.c
@@ -141,6 +141,7 @@ static void service_init(Unit *u) {
 
         exec_context_init(&s->exec_context);
         kill_context_init(&s->kill_context);
+        cgroup_context_init(&s->cgroup_context);
 
         RATELIMIT_INIT(s->start_limit, 10*USEC_PER_SEC, 5);
 
@@ -283,6 +284,7 @@ static void service_done(Unit *u) {
         free(s->status_text);
         s->status_text = NULL;
 
+        cgroup_context_done(&s->cgroup_context);
         exec_context_done(&s->exec_context, manager_is_reloading_or_reexecuting(u->manager));
         exec_command_free_array(s->exec_command, _SERVICE_EXEC_COMMAND_MAX);
         s->control_command = NULL;
@@ -1229,10 +1231,6 @@ static int service_load(Unit *u) {
                 if (r < 0)
                         return r;
 
-                r = unit_add_default_cgroups(u);
-                if (r < 0)
-                        return r;
-
 #ifdef HAVE_SYSV_COMPAT
                 r = sysv_fix_order(s);
                 if (r < 0)
@@ -1457,7 +1455,7 @@ static int service_search_main_pid(Service *s) {
 
         assert(s->main_pid <= 0);
 
-        pid = cgroup_bonding_search_main_pid_list(UNIT(s)->cgroup_bondings);
+        pid = unit_search_main_pid(UNIT(s));
         if (pid <= 0)
                 return -ENOENT;
 
@@ -1582,7 +1580,7 @@ static void service_set_state(Service *s, ServiceState state) {
         /* For the inactive states unit_notify() will trim the cgroup,
          * but for exit we have to do that ourselves... */
         if (state == SERVICE_EXITED && UNIT(s)->manager->n_reloading <= 0)
-                cgroup_bonding_trim_list(UNIT(s)->cgroup_bondings, true);
+                unit_destroy_cgroup(UNIT(s));
 
         if (old_state != state)
                 log_debug_unit(UNIT(s)->id,
@@ -1751,11 +1749,14 @@ static int service_spawn(
         unsigned n_fds = 0, n_env = 0;
         _cleanup_strv_free_ char
                 **argv = NULL, **final_env = NULL, **our_env = NULL;
+        const char *path;
 
         assert(s);
         assert(c);
         assert(_pid);
 
+        unit_realize_cgroup(UNIT(s));
+
         if (pass_fds ||
             s->exec_context.std_input == EXEC_INPUT_SOCKET ||
             s->exec_context.std_output == EXEC_OUTPUT_SOCKET ||
@@ -1811,7 +1812,7 @@ static int service_spawn(
                         goto fail;
                 }
 
-        if (s->meta.manager->running_as != SYSTEMD_SYSTEM)
+        if (UNIT(s)->manager->running_as != SYSTEMD_SYSTEM)
                 if (asprintf(our_env + n_env++, "MANAGERPID=%lu", (unsigned long) getpid()) < 0) {
                         r = -ENOMEM;
                         goto fail;
@@ -1823,6 +1824,12 @@ static int service_spawn(
                 goto fail;
         }
 
+        if (is_control && UNIT(s)->cgroup_path) {
+                path = strappenda(UNIT(s)->cgroup_path, "/control");
+                cg_create(SYSTEMD_CGROUP_CONTROLLER, path);
+        } else
+                path = UNIT(s)->cgroup_path;
+
         r = exec_spawn(c,
                        argv,
                        &s->exec_context,
@@ -1832,9 +1839,8 @@ static int service_spawn(
                        apply_chroot,
                        apply_tty_stdin,
                        UNIT(s)->manager->confirm_spawn,
-                       UNIT(s)->cgroup_bondings,
-                       UNIT(s)->cgroup_attributes,
-                       is_control ? "control" : NULL,
+                       UNIT(s)->cgroup_mask,
+                       path,
                        UNIT(s)->id,
                        s->type == SERVICE_IDLE ? UNIT(s)->manager->idle_pipe : NULL,
                        &pid);
@@ -1893,7 +1899,10 @@ static int cgroup_good(Service *s) {
 
         assert(s);
 
-        r = cgroup_bonding_is_empty_list(UNIT(s)->cgroup_bondings);
+        if (!UNIT(s)->cgroup_path)
+                return 0;
+
+        r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, UNIT(s)->cgroup_path, true);
         if (r < 0)
                 return r;
 
@@ -2123,10 +2132,21 @@ fail:
         service_enter_stop(s, SERVICE_FAILURE_RESOURCES);
 }
 
+static void service_kill_control_processes(Service *s) {
+        char *p;
+
+        if (!UNIT(s)->cgroup_path)
+                return;
+
+        p = strappenda(UNIT(s)->cgroup_path, "/control");
+
+        cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, p, SIGKILL, true, true, true, NULL);
+}
+
 static void service_enter_start(Service *s) {
+        ExecCommand *c;
         pid_t pid;
         int r;
-        ExecCommand *c;
 
         assert(s);
 
@@ -2141,7 +2161,7 @@ static void service_enter_start(Service *s) {
         /* We want to ensure that nobody leaks processes from
          * START_PRE here, so let's go on a killing spree, People
          * should not spawn long running processes from START_PRE. */
-        cgroup_bonding_kill_list(UNIT(s)->cgroup_bondings, SIGKILL, true, true, NULL, "control");
+        service_kill_control_processes(s);
 
         if (s->type == SERVICE_FORKING) {
                 s->control_command_id = SERVICE_EXEC_START;
@@ -2217,11 +2237,9 @@ static void service_enter_start_pre(Service *s) {
 
         s->control_command = s->exec_command[SERVICE_EXEC_START_PRE];
         if (s->control_command) {
-
                 /* Before we start anything, let's clear up what might
                  * be left from previous runs. */
-                cgroup_bonding_kill_list(UNIT(s)->cgroup_bondings, SIGKILL,
-                                         true,true, NULL, "control");
+                service_kill_control_processes(s);
 
                 s->control_command_id = SERVICE_EXEC_START_PRE;
 
@@ -3045,7 +3063,6 @@ static void service_sigchld_event(Unit *u, pid_t pid, int code, int status) {
                 }
 
         } else if (s->control_pid == pid) {
-
                 s->control_pid = 0;
 
                 if (s->control_command) {
@@ -3066,8 +3083,7 @@ static void service_sigchld_event(Unit *u, pid_t pid, int code, int status) {
                 /* Immediately get rid of the cgroup, so that the
                  * kernel doesn't delay the cgroup empty messages for
                  * the service cgroup any longer than necessary */
-                cgroup_bonding_kill_list(UNIT(s)->cgroup_bondings, SIGKILL,
-                                         true, true, NULL, "control");
+                service_kill_control_processes(s);
 
                 if (s->control_command &&
                     s->control_command->command_next &&
@@ -3296,7 +3312,7 @@ static void service_timer_event(Unit *u, uint64_t elapsed, Watch* w) {
         }
 }
 
-static void service_cgroup_notify_event(Unit *u) {
+static void service_notify_cgroup_empty_event(Unit *u) {
         Service *s = SERVICE(u);
 
         assert(u);
@@ -3823,8 +3839,9 @@ const UnitVTable service_vtable = {
                 "Service\0"
                 "Install\0",
 
+        .private_section = "Service",
         .exec_context_offset = offsetof(Service, exec_context),
-        .exec_section = "Service",
+        .cgroup_context_offset = offsetof(Service, cgroup_context),
 
         .init = service_init,
         .done = service_done,
@@ -3857,7 +3874,7 @@ const UnitVTable service_vtable = {
 
         .reset_failed = service_reset_failed,
 
-        .cgroup_notify_empty = service_cgroup_notify_event,
+        .notify_cgroup_empty = service_notify_cgroup_empty_event,
         .notify_message = service_notify_message,
 
         .bus_name_owner_change = service_bus_name_owner_change,