Merge pull request #3191 from poettering/cgroups-agent-dgram

core: use an AF_UNIX/SOCK_DGRAM socket for cgroup agent notification

26 files changed, 302 insertions, 161 deletions

diff --git a/src/basic/log.c b/src/basic/log.c
index d89e6f7274..3ea643b6e6 100644
--- a/src/basic/log.c
+++ b/src/basic/log.c
@@ -165,7 +165,7 @@ static int log_open_syslog(void) {
                 goto fail;
         }
 
-        if (connect(syslog_fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path)) < 0) {
+        if (connect(syslog_fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
                 safe_close(syslog_fd);
 
                 /* Some legacy syslog systems still use stream
@@ -177,7 +177,7 @@ static int log_open_syslog(void) {
                         goto fail;
                 }
 
-                if (connect(syslog_fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path)) < 0) {
+                if (connect(syslog_fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
                         r = -errno;
                         goto fail;
                 }
@@ -215,7 +215,7 @@ static int log_open_journal(void) {
                 goto fail;
         }
 
-        if (connect(journal_fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path)) < 0) {
+        if (connect(journal_fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
                 r = -errno;
                 goto fail;
         }
diff --git a/src/basic/socket-util.h b/src/basic/socket-util.h
index d17a2f35f8..daa4b24a37 100644
--- a/src/basic/socket-util.h
+++ b/src/basic/socket-util.h
@@ -137,3 +137,14 @@ ssize_t next_datagram_size_fd(int fd);
 
 #define CMSG_FOREACH(cmsg, mh)                                          \
         for ((cmsg) = CMSG_FIRSTHDR(mh); (cmsg); (cmsg) = CMSG_NXTHDR((mh), (cmsg)))
+
+/* Covers only file system and abstract AF_UNIX socket addresses, but not unnamed socket addresses. */
+#define SOCKADDR_UN_LEN(sa)                                             \
+        ({                                                              \
+                const struct sockaddr_un *_sa = &(sa);                  \
+                assert(_sa->sun_family == AF_UNIX);                     \
+                offsetof(struct sockaddr_un, sun_path) +                \
+                        (_sa->sun_path[0] == 0 ?                        \
+                         1 + strnlen(_sa->sun_path+1, sizeof(_sa->sun_path)-1) : \
+                         strnlen(_sa->sun_path, sizeof(_sa->sun_path))); \
+        })
diff --git a/src/cgroups-agent/cgroups-agent.c b/src/cgroups-agent/cgroups-agent.c
index aadfba0707..d7c722ac3d 100644
--- a/src/cgroups-agent/cgroups-agent.c
+++ b/src/cgroups-agent/cgroups-agent.c
@@ -18,15 +18,22 @@
 ***/
 
 #include <stdlib.h>
+#include <sys/socket.h>
 
-#include "sd-bus.h"
-
-#include "bus-util.h"
+#include "fd-util.h"
 #include "log.h"
+#include "socket-util.h"
 
 int main(int argc, char *argv[]) {
-        _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
-        int r;
+
+        static const union sockaddr_union sa = {
+                .un.sun_family = AF_UNIX,
+                .un.sun_path = "/run/systemd/cgroups-agent",
+        };
+
+        _cleanup_close_ int fd = -1;
+        ssize_t n;
+        size_t l;
 
         if (argc != 2) {
                 log_error("Incorrect number of arguments.");
@@ -37,27 +44,22 @@ int main(int argc, char *argv[]) {
         log_parse_environment();
         log_open();
 
-        /* We send this event to the private D-Bus socket and then the
-         * system instance will forward this to the system bus. We do
-         * this to avoid an activation loop when we start dbus when we
-         * are called when the dbus service is shut down. */
-
-        r = bus_connect_system_systemd(&bus);
-        if (r < 0) {
-                /* If we couldn't connect we assume this was triggered
-                 * while systemd got restarted/transitioned from
-                 * initrd to the system, so let's ignore this */
-                log_debug_errno(r, "Failed to get D-Bus connection: %m");
+        fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0);
+        if (fd < 0) {
+                log_debug_errno(errno, "Failed to allocate socket: %m");
+                return EXIT_FAILURE;
+        }
+
+        l = strlen(argv[1]);
+
+        n = sendto(fd, argv[1], l, 0, &sa.sa, SOCKADDR_UN_LEN(sa.un));
+        if (n < 0) {
+                log_debug_errno(errno, "Failed to send cgroups agent message: %m");
                 return EXIT_FAILURE;
         }
 
-        r = sd_bus_emit_signal(bus,
-                               "/org/freedesktop/systemd1/agent",
-                               "org.freedesktop.systemd1.Agent",
-                               "Released",
-                               "s", argv[1]);
-        if (r < 0) {
-                log_debug_errno(r, "Failed to send signal message on private connection: %m");
+        if ((size_t) n != l) {
+                log_debug("Datagram size mismatch");
                 return EXIT_FAILURE;
         }
 
diff --git a/src/core/cgroup.c b/src/core/cgroup.c
index 25cc6962f9..1a94b188cb 100644
--- a/src/core/cgroup.c
+++ b/src/core/cgroup.c
@@ -1312,7 +1312,9 @@ int manager_setup_cgroup(Manager *m) {
                         if (r < 0)
                                 return log_error_errno(r, "Failed to watch control group inotify object: %m");
 
-                        r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_IDLE - 5);
+                        /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also
+                         * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */
+                        r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-5);
                         if (r < 0)
                                 return log_error_errno(r, "Failed to set priority of inotify event source: %m");
 
@@ -1458,6 +1460,8 @@ int manager_notify_cgroup_empty(Manager *m, const char *cgroup) {
         assert(m);
         assert(cgroup);
 
+        log_debug("Got cgroup empty notification for: %s", cgroup);
+
         u = manager_get_unit_by_cgroup(m, cgroup);
         if (!u)
                 return 0;
diff --git a/src/core/dbus.c b/src/core/dbus.c
index 263955d874..3422a02d68 100644
--- a/src/core/dbus.c
+++ b/src/core/dbus.c
@@ -71,28 +71,42 @@ int bus_send_queued_message(Manager *m) {
         return 0;
 }
 
+int bus_forward_agent_released(Manager *m, const char *path) {
+        int r;
+
+        assert(m);
+        assert(path);
+
+        if (!MANAGER_IS_SYSTEM(m))
+                return 0;
+
+        if (!m->system_bus)
+                return 0;
+
+        /* If we are running a system instance we forward the agent message on the system bus, so that the user
+         * instances get notified about this, too */
+
+        r = sd_bus_emit_signal(m->system_bus,
+                               "/org/freedesktop/systemd1/agent",
+                               "org.freedesktop.systemd1.Agent",
+                               "Released",
+                               "s", path);
+        if (r < 0)
+                return log_warning_errno(r, "Failed to propagate agent release message: %m");
+
+        return 1;
+}
+
 static int signal_agent_released(sd_bus_message *message, void *userdata, sd_bus_error *error) {
         _cleanup_(sd_bus_creds_unrefp) sd_bus_creds *creds = NULL;
-        const char *cgroup, *me;
         Manager *m = userdata;
+        const char *cgroup;
         uid_t sender_uid;
-        sd_bus *bus;
         int r;
 
         assert(message);
         assert(m);
 
-        /* ignore recursive events sent by us on the system/user bus */
-        bus = sd_bus_message_get_bus(message);
-        if (!sd_bus_is_server(bus)) {
-                r = sd_bus_get_unique_name(bus, &me);
-                if (r < 0)
-                        return r;
-
-                if (streq_ptr(sd_bus_message_get_sender(message), me))
-                        return 0;
-        }
-
         /* only accept org.freedesktop.systemd1.Agent from UID=0 */
         r = sd_bus_query_sender_creds(message, SD_BUS_CREDS_EUID, &creds);
         if (r < 0)
@@ -110,16 +124,6 @@ static int signal_agent_released(sd_bus_message *message, void *userdata, sd_bus
         }
 
         manager_notify_cgroup_empty(m, cgroup);
-
-        /* if running as system-instance, forward under our name */
-        if (MANAGER_IS_SYSTEM(m) && m->system_bus) {
-                r = sd_bus_message_rewind(message, 1);
-                if (r >= 0)
-                        r = sd_bus_send(m->system_bus, message, NULL);
-                if (r < 0)
-                        log_warning_errno(r, "Failed to forward Released message: %m");
-        }
-
         return 0;
 }
 
@@ -690,25 +694,6 @@ static int bus_on_connection(sd_event_source *s, int fd, uint32_t revents, void
                 return 0;
         }
 
-        if (MANAGER_IS_SYSTEM(m)) {
-                /* When we run as system instance we get the Released
-                 * signal via a direct connection */
-
-                r = sd_bus_add_match(
-                                bus,
-                                NULL,
-                                "type='signal',"
-                                "interface='org.freedesktop.systemd1.Agent',"
-                                "member='Released',"
-                                "path='/org/freedesktop/systemd1/agent'",
-                                signal_agent_released, m);
-
-                if (r < 0) {
-                        log_warning_errno(r, "Failed to register Released match on new connection bus: %m");
-                        return 0;
-                }
-        }
-
         r = bus_setup_disconnected_match(m, bus);
         if (r < 0)
                 return 0;
@@ -906,8 +891,8 @@ static int bus_setup_system(Manager *m, sd_bus *bus) {
         assert(m);
         assert(bus);
 
-        /* On kdbus or if we are a user instance we get the Released message via the system bus */
-        if (MANAGER_IS_USER(m) || m->kdbus_fd >= 0) {
+        /* if we are a user instance we get the Released message via the system bus */
+        if (MANAGER_IS_USER(m)) {
                 r = sd_bus_add_match(
                                 bus,
                                 NULL,
@@ -990,7 +975,7 @@ static int bus_init_private(Manager *m) {
                         return 0;
 
                 strcpy(sa.un.sun_path, "/run/systemd/private");
-                salen = offsetof(union sockaddr_union, un.sun_path) + strlen("/run/systemd/private");
+                salen = SOCKADDR_UN_LEN(sa.un);
         } else {
                 size_t left = sizeof(sa.un.sun_path);
                 char *p = sa.un.sun_path;
diff --git a/src/core/dbus.h b/src/core/dbus.h
index e16a84fbb8..6baaffbd75 100644
--- a/src/core/dbus.h
+++ b/src/core/dbus.h
@@ -40,3 +40,5 @@ int bus_verify_manage_units_async(Manager *m, sd_bus_message *call, sd_bus_error
 int bus_verify_manage_unit_files_async(Manager *m, sd_bus_message *call, sd_bus_error *error);
 int bus_verify_reload_daemon_async(Manager *m, sd_bus_message *call, sd_bus_error *error);
 int bus_verify_set_environment_async(Manager *m, sd_bus_message *call, sd_bus_error *error);
+
+int bus_forward_agent_released(Manager *m, const char *path);
diff --git a/src/core/execute.c b/src/core/execute.c
index ac2ac39892..5eb3f13695 100644
--- a/src/core/execute.c
+++ b/src/core/execute.c
@@ -271,7 +271,7 @@ static int connect_journal_socket(int fd, uid_t uid, gid_t gid) {
                 }
         }
 
-        r = connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
+        r = connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
         if (r < 0)
                 r = -errno;
 
diff --git a/src/core/manager.c b/src/core/manager.c
index bd00c224f4..e192cd475d 100644
--- a/src/core/manager.c
+++ b/src/core/manager.c
@@ -87,6 +87,7 @@
 #include "watchdog.h"
 
 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
+#define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
 
 /* Initial delay and the interval for printing status messages about running jobs */
 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
@@ -94,6 +95,7 @@
 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
 
 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
+static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
@@ -484,11 +486,11 @@ static int manager_setup_signals(Manager *m) {
 
         (void) sd_event_source_set_description(m->signal_event_source, "manager-signal");
 
-        /* Process signals a bit earlier than the rest of things, but
-         * later than notify_fd processing, so that the notify
-         * processing can still figure out to which process/service a
-         * message belongs, before we reap the process. */
-        r = sd_event_source_set_priority(m->signal_event_source, SD_EVENT_PRIORITY_NORMAL-5);
+        /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
+         * notify processing can still figure out to which process/service a message belongs, before we reap the
+         * process. Also, process this before handling cgroup notifications, so that we always collect child exit
+         * status information before detecting that there's no process in a cgroup. */
+        r = sd_event_source_set_priority(m->signal_event_source, SD_EVENT_PRIORITY_NORMAL-6);
         if (r < 0)
                 return r;
 
@@ -581,12 +583,12 @@ int manager_new(UnitFileScope scope, bool test_run, Manager **_m) {
 
         m->idle_pipe[0] = m->idle_pipe[1] = m->idle_pipe[2] = m->idle_pipe[3] = -1;
 
-        m->pin_cgroupfs_fd = m->notify_fd = m->signal_fd = m->time_change_fd =
-                m->dev_autofs_fd = m->private_listen_fd = m->kdbus_fd = m->cgroup_inotify_fd = -1;
+        m->pin_cgroupfs_fd = m->notify_fd = m->cgroups_agent_fd = m->signal_fd = m->time_change_fd =
+                m->dev_autofs_fd = m->private_listen_fd = m->kdbus_fd = m->cgroup_inotify_fd =
+                m->ask_password_inotify_fd = -1;
 
         m->current_job_id = 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
 
-        m->ask_password_inotify_fd = -1;
         m->have_ask_password = -EINVAL; /* we don't know */
         m->first_boot = -1;
 
@@ -703,7 +705,7 @@ static int manager_setup_notify(Manager *m) {
                 (void) unlink(m->notify_socket);
 
                 strncpy(sa.un.sun_path, m->notify_socket, sizeof(sa.un.sun_path)-1);
-                r = bind(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
+                r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
                 if (r < 0)
                         return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
 
@@ -722,8 +724,8 @@ static int manager_setup_notify(Manager *m) {
                 if (r < 0)
                         return log_error_errno(r, "Failed to allocate notify event source: %m");
 
-                /* Process signals a bit earlier than SIGCHLD, so that we can
-                 * still identify to which service an exit message belongs */
+                /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
+                 * service an exit message belongs. */
                 r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-7);
                 if (r < 0)
                         return log_error_errno(r, "Failed to set priority of notify event source: %m");
@@ -734,6 +736,79 @@ static int manager_setup_notify(Manager *m) {
         return 0;
 }
 
+static int manager_setup_cgroups_agent(Manager *m) {
+
+        static const union sockaddr_union sa = {
+                .un.sun_family = AF_UNIX,
+                .un.sun_path = "/run/systemd/cgroups-agent",
+        };
+        int r;
+
+        /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
+         * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
+         * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
+         * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
+         * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
+         * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
+         * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
+         * we thus won't lose messages.
+         *
+         * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
+         * to it. The system instance hence listens on this special socket, but the user instances listen on the system
+         * bus for these messages. */
+
+        if (m->test_run)
+                return 0;
+
+        if (!MANAGER_IS_SYSTEM(m))
+                return 0;
+
+        if (cg_unified() > 0) /* We don't need this anymore on the unified hierarchy */
+                return 0;
+
+        if (m->cgroups_agent_fd < 0) {
+                _cleanup_close_ int fd = -1;
+
+                /* First free all secondary fields */
+                m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
+
+                fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
+                if (fd < 0)
+                        return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m");
+
+                fd_inc_rcvbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE);
+
+                (void) unlink(sa.un.sun_path);
+
+                /* Only allow root to connect to this socket */
+                RUN_WITH_UMASK(0077)
+                        r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
+                if (r < 0)
+                        return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
+
+                m->cgroups_agent_fd = fd;
+                fd = -1;
+        }
+
+        if (!m->cgroups_agent_event_source) {
+                r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m);
+                if (r < 0)
+                        return log_error_errno(r, "Failed to allocate cgroups agent event source: %m");
+
+                /* Process cgroups notifications early, but after having processed service notification messages or
+                 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
+                 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
+                 * cgroup inotify for the unified cgroup stuff. */
+                r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-5);
+                if (r < 0)
+                        return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m");
+
+                (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent");
+        }
+
+        return 0;
+}
+
 static int manager_setup_kdbus(Manager *m) {
         _cleanup_free_ char *p = NULL;
 
@@ -944,12 +1019,14 @@ Manager* manager_free(Manager *m) {
 
         sd_event_source_unref(m->signal_event_source);
         sd_event_source_unref(m->notify_event_source);
+        sd_event_source_unref(m->cgroups_agent_event_source);
         sd_event_source_unref(m->time_change_event_source);
         sd_event_source_unref(m->jobs_in_progress_event_source);
         sd_event_source_unref(m->run_queue_event_source);
 
         safe_close(m->signal_fd);
         safe_close(m->notify_fd);
+        safe_close(m->cgroups_agent_fd);
         safe_close(m->time_change_fd);
         safe_close(m->kdbus_fd);
 
@@ -1142,6 +1219,10 @@ int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
         if (q < 0 && r == 0)
                 r = q;
 
+        q = manager_setup_cgroups_agent(m);
+        if (q < 0 && r == 0)
+                r = q;
+
         /* We might have deserialized the kdbus control fd, but if we
          * didn't, then let's create the bus now. */
         manager_setup_kdbus(m);
@@ -1479,6 +1560,35 @@ static unsigned manager_dispatch_dbus_queue(Manager *m) {
         return n;
 }
 
+static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
+        Manager *m = userdata;
+        char buf[PATH_MAX+1];
+        ssize_t n;
+
+        n = recv(fd, buf, sizeof(buf), 0);
+        if (n < 0)
+                return log_error_errno(errno, "Failed to read cgroups agent message: %m");
+        if (n == 0) {
+                log_error("Got zero-length cgroups agent message, ignoring.");
+                return 0;
+        }
+        if ((size_t) n >= sizeof(buf)) {
+                log_error("Got overly long cgroups agent message, ignoring.");
+                return 0;
+        }
+
+        if (memchr(buf, 0, n)) {
+                log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
+                return 0;
+        }
+        buf[n] = 0;
+
+        manager_notify_cgroup_empty(m, buf);
+        bus_forward_agent_released(m, buf);
+
+        return 0;
+}
+
 static void manager_invoke_notify_message(Manager *m, Unit *u, pid_t pid, const char *buf, size_t n, FDSet *fds) {
         _cleanup_strv_free_ char **tags = NULL;
 
@@ -2135,11 +2245,10 @@ void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
 }
 
 void manager_send_unit_plymouth(Manager *m, Unit *u) {
-        union sockaddr_union sa = PLYMOUTH_SOCKET;
-
-        int n = 0;
+        static const union sockaddr_union sa = PLYMOUTH_SOCKET;
         _cleanup_free_ char *message = NULL;
         _cleanup_close_ int fd = -1;
+        int n = 0;
 
         /* Don't generate plymouth events if the service was already
          * started and we're just deserializing */
@@ -2165,7 +2274,7 @@ void manager_send_unit_plymouth(Manager *m, Unit *u) {
                 return;
         }
 
-        if (connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + 1 + strlen(sa.un.sun_path+1)) < 0) {
+        if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
 
                 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
                         log_error_errno(errno, "connect() failed: %m");
@@ -2265,6 +2374,16 @@ int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) {
                 fprintf(f, "notify-socket=%s\n", m->notify_socket);
         }
 
+        if (m->cgroups_agent_fd >= 0) {
+                int copy;
+
+                copy = fdset_put_dup(fds, m->cgroups_agent_fd);
+                if (copy < 0)
+                        return copy;
+
+                fprintf(f, "cgroups-agent-fd=%i\n", copy);
+        }
+
         if (m->kdbus_fd >= 0) {
                 int copy;
 
@@ -2432,6 +2551,17 @@ int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
                         free(m->notify_socket);
                         m->notify_socket = n;
 
+                } else if (startswith(l, "cgroups-agent-fd=")) {
+                        int fd;
+
+                        if (safe_atoi(l + 17, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
+                                log_debug("Failed to parse cgroups agent fd: %s", l + 10);
+                        else {
+                                m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
+                                safe_close(m->cgroups_agent_fd);
+                                m->cgroups_agent_fd = fdset_remove(fds, fd);
+                        }
+
                 } else if (startswith(l, "kdbus-fd=")) {
                         int fd;
 
@@ -2552,6 +2682,10 @@ int manager_reload(Manager *m) {
         if (q < 0 && r >= 0)
                 r = q;
 
+        q = manager_setup_cgroups_agent(m);
+        if (q < 0 && r >= 0)
+                r = q;
+
         /* Third, fire things up! */
         manager_coldplug(m);
 
diff --git a/src/core/manager.h b/src/core/manager.h
index 17f84e6963..4bccca75cb 100644
--- a/src/core/manager.h
+++ b/src/core/manager.h
@@ -132,6 +132,9 @@ struct Manager {
         int notify_fd;
         sd_event_source *notify_event_source;
 
+        int cgroups_agent_fd;
+        sd_event_source *cgroups_agent_event_source;
+
         int signal_fd;
         sd_event_source *signal_event_source;
 
diff --git a/src/coredump/coredump.c b/src/coredump/coredump.c
index 41fc1993d5..01fdcfa909 100644
--- a/src/coredump/coredump.c
+++ b/src/coredump/coredump.c
@@ -847,7 +847,7 @@ static int send_iovec(const struct iovec iovec[], size_t n_iovec, int input_fd)
         if (fd < 0)
                 return log_error_errno(errno, "Failed to create coredump socket: %m");
 
-        if (connect(fd, &sa.sa, offsetof(union sockaddr_union, un.sun_path) + strlen(sa.un.sun_path)) < 0)
+        if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0)
                 return log_error_errno(errno, "Failed to connect to coredump service: %m");
 
         for (i = 0; i < n_iovec; i++) {
diff --git a/src/fsck/fsck.c b/src/fsck/fsck.c
index 6f56066da8..d32e1d923e 100644
--- a/src/fsck/fsck.c
+++ b/src/fsck/fsck.c
@@ -262,7 +262,7 @@ static int fsck_progress_socket(void) {
         if (fd < 0)
                 return log_warning_errno(errno, "socket(): %m");
 
-        if (connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path)) < 0) {
+        if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
                 r = log_full_errno(errno == ECONNREFUSED || errno == ENOENT ? LOG_DEBUG : LOG_WARNING,
                                    errno, "Failed to connect to progress socket %s, ignoring: %m", sa.un.sun_path);
                 safe_close(fd);
diff --git a/src/import/importd.c b/src/import/importd.c
index d2a5867a6e..956a82945c 100644
--- a/src/import/importd.c
+++ b/src/import/importd.c
@@ -677,7 +677,7 @@ static int manager_new(Manager **ret) {
         (void) mkdir_parents_label(sa.un.sun_path, 0755);
         (void) unlink(sa.un.sun_path);
 
-        if (bind(m->notify_fd, &sa.sa, offsetof(union sockaddr_union, un.sun_path) + strlen(sa.un.sun_path)) < 0)
+        if (bind(m->notify_fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0)
                 return -errno;
 
         if (setsockopt(m->notify_fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one)) < 0)
diff --git a/src/journal/journal-send.c b/src/journal/journal-send.c
index f0959b6237..5e8a3e3200 100644
--- a/src/journal/journal-send.c
+++ b/src/journal/journal-send.c
@@ -208,13 +208,13 @@ _public_ int sd_journal_sendv(const struct iovec *iov, int n) {
         struct iovec *w;
         uint64_t *l;
         int i, j = 0;
-        struct sockaddr_un sa = {
-                .sun_family = AF_UNIX,
-                .sun_path = "/run/systemd/journal/socket",
+        static const union sockaddr_union sa = {
+                .un.sun_family = AF_UNIX,
+                .un.sun_path = "/run/systemd/journal/socket",
         };
         struct msghdr mh = {
-                .msg_name = &sa,
-                .msg_namelen = offsetof(struct sockaddr_un, sun_path) + strlen(sa.sun_path),
+                .msg_name = (struct sockaddr*) &sa.sa,
+                .msg_namelen = SOCKADDR_UN_LEN(sa.un),
         };
         ssize_t k;
         bool have_syslog_identifier = false;
@@ -392,7 +392,7 @@ _public_ int sd_journal_perror(const char *message) {
 }
 
 _public_ int sd_journal_stream_fd(const char *identifier, int priority, int level_prefix) {
-        union sockaddr_union sa = {
+        static const union sockaddr_union sa = {
                 .un.sun_family = AF_UNIX,
                 .un.sun_path = "/run/systemd/journal/stdout",
         };
@@ -408,7 +408,7 @@ _public_ int sd_journal_stream_fd(const char *identifier, int priority, int leve
         if (fd < 0)
                 return -errno;
 
-        r = connect(fd, &sa.sa, offsetof(union sockaddr_union, un.sun_path) + strlen(sa.un.sun_path));
+        r = connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
         if (r < 0)
                 return -errno;
 
diff --git a/src/journal/journald-native.c b/src/journal/journald-native.c
index a445291a5e..0a1ce205c2 100644
--- a/src/journal/journald-native.c
+++ b/src/journal/journald-native.c
@@ -448,24 +448,24 @@ void server_process_native_file(
 }
 
 int server_open_native_socket(Server*s) {
+
+        static const union sockaddr_union sa = {
+                .un.sun_family = AF_UNIX,
+                .un.sun_path = "/run/systemd/journal/socket",
+        };
         static const int one = 1;
         int r;
 
         assert(s);
 
         if (s->native_fd < 0) {
-                union sockaddr_union sa = {
-                        .un.sun_family = AF_UNIX,
-                        .un.sun_path = "/run/systemd/journal/socket",
-                };
-
                 s->native_fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
                 if (s->native_fd < 0)
                         return log_error_errno(errno, "socket() failed: %m");
 
-                unlink(sa.un.sun_path);
+                (void) unlink(sa.un.sun_path);
 
-                r = bind(s->native_fd, &sa.sa, offsetof(union sockaddr_union, un.sun_path) + strlen(sa.un.sun_path));
+                r = bind(s->native_fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
                 if (r < 0)
                         return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
 
diff --git a/src/journal/journald-server.c b/src/journal/journald-server.c
index e14d0ad980..8f82d2a838 100644
--- a/src/journal/journald-server.c
+++ b/src/journal/journald-server.c
@@ -1696,7 +1696,7 @@ static int server_connect_notify(Server *s) {
         if (sa.un.sun_path[0] == '@')
                 sa.un.sun_path[0] = 0;
 
-        r = connect(s->notify_fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(e));
+        r = connect(s->notify_fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
         if (r < 0)
                 return log_error_errno(errno, "Failed to connect to notify socket: %m");
 
diff --git a/src/journal/journald-stream.c b/src/journal/journald-stream.c
index 59352bcb3f..4ad16ee41c 100644
--- a/src/journal/journald-stream.c
+++ b/src/journal/journald-stream.c
@@ -700,23 +700,22 @@ fail:
 }
 
 int server_open_stdout_socket(Server *s) {
+        static const union sockaddr_union sa = {
+                .un.sun_family = AF_UNIX,
+                .un.sun_path = "/run/systemd/journal/stdout",
+        };
         int r;
 
         assert(s);
 
         if (s->stdout_fd < 0) {
-                union sockaddr_union sa = {
-                        .un.sun_family = AF_UNIX,
-                        .un.sun_path = "/run/systemd/journal/stdout",
-                };
-
                 s->stdout_fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
                 if (s->stdout_fd < 0)
                         return log_error_errno(errno, "socket() failed: %m");
 
-                unlink(sa.un.sun_path);
+                (void) unlink(sa.un.sun_path);
 
-                r = bind(s->stdout_fd, &sa.sa, offsetof(union sockaddr_union, un.sun_path) + strlen(sa.un.sun_path));
+                r = bind(s->stdout_fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
                 if (r < 0)
                         return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
 
diff --git a/src/journal/journald-syslog.c b/src/journal/journald-syslog.c
index 5153fd0cce..0609b4b694 100644
--- a/src/journal/journald-syslog.c
+++ b/src/journal/journald-syslog.c
@@ -52,8 +52,7 @@ static void forward_syslog_iovec(Server *s, const struct iovec *iovec, unsigned
                 .msg_iov = (struct iovec *) iovec,
                 .msg_iovlen = n_iovec,
                 .msg_name = (struct sockaddr*) &sa.sa,
-                .msg_namelen = offsetof(union sockaddr_union, un.sun_path)
-                               + strlen("/run/systemd/journal/syslog"),
+                .msg_namelen = SOCKADDR_UN_LEN(sa.un),
         };
         struct cmsghdr *cmsg;
         union {
@@ -316,12 +315,12 @@ static void syslog_skip_date(char **buf) {
 }
 
 void server_process_syslog_message(
-        Server *s,
-        const char *buf,
-        const struct ucred *ucred,
-        const struct timeval *tv,
-        const char *label,
-        size_t label_len) {
+                Server *s,
+                const char *buf,
+                const struct ucred *ucred,
+                const struct timeval *tv,
+                const char *label,
+                size_t label_len) {
 
         char syslog_priority[sizeof("PRIORITY=") + DECIMAL_STR_MAX(int)],
              syslog_facility[sizeof("SYSLOG_FACILITY=") + DECIMAL_STR_MAX(int)];
@@ -365,14 +364,12 @@ void server_process_syslog_message(
 
         if (identifier) {
                 syslog_identifier = strjoina("SYSLOG_IDENTIFIER=", identifier);
-                if (syslog_identifier)
-                        IOVEC_SET_STRING(iovec[n++], syslog_identifier);
+                IOVEC_SET_STRING(iovec[n++], syslog_identifier);
         }
 
         if (pid) {
                 syslog_pid = strjoina("SYSLOG_PID=", pid);
-                if (syslog_pid)
-                        IOVEC_SET_STRING(iovec[n++], syslog_pid);
+                IOVEC_SET_STRING(iovec[n++], syslog_pid);
         }
 
         message = strjoina("MESSAGE=", buf);
@@ -383,24 +380,24 @@ void server_process_syslog_message(
 }
 
 int server_open_syslog_socket(Server *s) {
+
+        static const union sockaddr_union sa = {
+                .un.sun_family = AF_UNIX,
+                .un.sun_path = "/run/systemd/journal/dev-log",
+        };
         static const int one = 1;
         int r;
 
         assert(s);
 
         if (s->syslog_fd < 0) {
-                static const union sockaddr_union sa = {
-                        .un.sun_family = AF_UNIX,
-                        .un.sun_path = "/run/systemd/journal/dev-log",
-                };
-
                 s->syslog_fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
                 if (s->syslog_fd < 0)
                         return log_error_errno(errno, "socket() failed: %m");
 
-                unlink(sa.un.sun_path);
+                (void) unlink(sa.un.sun_path);
 
-                r = bind(s->syslog_fd, &sa.sa, offsetof(union sockaddr_union, un.sun_path) + strlen(sa.un.sun_path));
+                r = bind(s->syslog_fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
                 if (r < 0)
                         return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
 
@@ -437,6 +434,7 @@ int server_open_syslog_socket(Server *s) {
 
 void server_maybe_warn_forward_syslog_missed(Server *s) {
         usec_t n;
+
         assert(s);
 
         if (s->n_forward_syslog_missed <= 0)
diff --git a/src/libsystemd/sd-bus/sd-bus.c b/src/libsystemd/sd-bus/sd-bus.c
index 04da94e7e3..ed5f94e136 100644
--- a/src/libsystemd/sd-bus/sd-bus.c
+++ b/src/libsystemd/sd-bus/sd-bus.c
@@ -836,7 +836,7 @@ static int parse_container_unix_address(sd_bus *b, const char **p, char **guid)
 
         b->sockaddr.un.sun_family = AF_UNIX;
         strncpy(b->sockaddr.un.sun_path, "/var/run/dbus/system_bus_socket", sizeof(b->sockaddr.un.sun_path));
-        b->sockaddr_size = offsetof(struct sockaddr_un, sun_path) + strlen("/var/run/dbus/system_bus_socket");
+        b->sockaddr_size = SOCKADDR_UN_LEN(b->sockaddr.un);
 
         return 0;
 }
diff --git a/src/libsystemd/sd-daemon/sd-daemon.c b/src/libsystemd/sd-daemon/sd-daemon.c
index bd1c7f15ff..4da9dbfd63 100644
--- a/src/libsystemd/sd-daemon/sd-daemon.c
+++ b/src/libsystemd/sd-daemon/sd-daemon.c
@@ -458,9 +458,7 @@ _public_ int sd_pid_notify_with_fds(pid_t pid, int unset_environment, const char
         if (sockaddr.un.sun_path[0] == '@')
                 sockaddr.un.sun_path[0] = 0;
 
-        msghdr.msg_namelen = offsetof(struct sockaddr_un, sun_path) + strlen(e);
-        if (msghdr.msg_namelen > sizeof(struct sockaddr_un))
-                msghdr.msg_namelen = sizeof(struct sockaddr_un);
+        msghdr.msg_namelen = SOCKADDR_UN_LEN(sockaddr.un);
 
         have_pid = pid != 0 && pid != getpid();
 
diff --git a/src/login/pam_systemd.c b/src/login/pam_systemd.c
index 40e246bb06..98dc201340 100644
--- a/src/login/pam_systemd.c
+++ b/src/login/pam_systemd.c
@@ -150,7 +150,7 @@ static int get_seat_from_display(const char *display, const char **seat, uint32_
         if (fd < 0)
                 return -errno;
 
-        if (connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path)) < 0)
+        if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0)
                 return -errno;
 
         r = getpeercred(fd, &ucred);
diff --git a/src/reply-password/reply-password.c b/src/reply-password/reply-password.c
index e291758969..17eab9772e 100644
--- a/src/reply-password/reply-password.c
+++ b/src/reply-password/reply-password.c
@@ -26,14 +26,12 @@
 #include "fd-util.h"
 #include "log.h"
 #include "macro.h"
+#include "socket-util.h"
 #include "string-util.h"
 #include "util.h"
 
 static int send_on_socket(int fd, const char *socket_name, const void *packet, size_t size) {
-        union {
-                struct sockaddr sa;
-                struct sockaddr_un un;
-        } sa = {
+        union sockaddr_union sa = {
                 .un.sun_family = AF_UNIX,
         };
 
@@ -43,7 +41,7 @@ static int send_on_socket(int fd, const char *socket_name, const void *packet, s
 
         strncpy(sa.un.sun_path, socket_name, sizeof(sa.un.sun_path));
 
-        if (sendto(fd, packet, size, MSG_NOSIGNAL, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(socket_name)) < 0)
+        if (sendto(fd, packet, size, MSG_NOSIGNAL, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0)
                 return log_error_errno(errno, "Failed to send: %m");
 
         return 0;
diff --git a/src/shared/ask-password-api.c b/src/shared/ask-password-api.c
index 6805873f9e..4a4bd8d3b8 100644
--- a/src/shared/ask-password-api.c
+++ b/src/shared/ask-password-api.c
@@ -431,7 +431,7 @@ static int create_socket(char **name) {
         snprintf(sa.un.sun_path, sizeof(sa.un.sun_path)-1, "/run/systemd/ask-password/sck.%" PRIx64, random_u64());
 
         RUN_WITH_UMASK(0177) {
-                if (bind(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path)) < 0)
+                if (bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0)
                         return -errno;
         }
 
diff --git a/src/socket-proxy/socket-proxyd.c b/src/socket-proxy/socket-proxyd.c
index 1157a0c72e..52b4db8875 100644
--- a/src/socket-proxy/socket-proxyd.c
+++ b/src/socket-proxy/socket-proxyd.c
@@ -400,28 +400,19 @@ static int resolve_remote(Connection *c) {
 
         union sockaddr_union sa = {};
         const char *node, *service;
-        socklen_t salen;
         int r;
 
         if (path_is_absolute(arg_remote_host)) {
                 sa.un.sun_family = AF_UNIX;
-                strncpy(sa.un.sun_path, arg_remote_host, sizeof(sa.un.sun_path)-1);
-                sa.un.sun_path[sizeof(sa.un.sun_path)-1] = 0;
-
-                salen = offsetof(union sockaddr_union, un.sun_path) + strlen(sa.un.sun_path);
-
-                return connection_start(c, &sa.sa, salen);
+                strncpy(sa.un.sun_path, arg_remote_host, sizeof(sa.un.sun_path));
+                return connection_start(c, &sa.sa, SOCKADDR_UN_LEN(sa.un));
         }
 
         if (arg_remote_host[0] == '@') {
                 sa.un.sun_family = AF_UNIX;
                 sa.un.sun_path[0] = 0;
-                strncpy(sa.un.sun_path+1, arg_remote_host+1, sizeof(sa.un.sun_path)-2);
-                sa.un.sun_path[sizeof(sa.un.sun_path)-1] = 0;
-
-                salen = offsetof(union sockaddr_union, un.sun_path) + 1 + strlen(sa.un.sun_path + 1);
-
-                return connection_start(c, &sa.sa, salen);
+                strncpy(sa.un.sun_path+1, arg_remote_host+1, sizeof(sa.un.sun_path)-1);
+                return connection_start(c, &sa.sa, SOCKADDR_UN_LEN(sa.un));
         }
 
         service = strrchr(arg_remote_host, ':');
diff --git a/src/test/test-socket-util.c b/src/test/test-socket-util.c
index 33ff3755bc..9e01f3afd4 100644
--- a/src/test/test-socket-util.c
+++ b/src/test/test-socket-util.c
@@ -343,6 +343,21 @@ static void test_sockaddr_equal(void) {
         assert_se(!sockaddr_equal(&b, &c));
 }
 
+static void test_sockaddr_un_len(void) {
+        static const struct sockaddr_un fs = {
+                .sun_family = AF_UNIX,
+                .sun_path = "/foo/bar/waldo",
+        };
+
+        static const struct sockaddr_un abstract = {
+                .sun_family = AF_UNIX,
+                .sun_path = "\0foobar",
+        };
+
+        assert_se(SOCKADDR_UN_LEN(fs) == offsetof(struct sockaddr_un, sun_path) + strlen(fs.sun_path));
+        assert_se(SOCKADDR_UN_LEN(abstract) == offsetof(struct sockaddr_un, sun_path) + 1 + strlen(abstract.sun_path + 1));
+}
+
 int main(int argc, char *argv[]) {
 
         log_set_max_level(LOG_DEBUG);
@@ -363,5 +378,7 @@ int main(int argc, char *argv[]) {
 
         test_sockaddr_equal();
 
+        test_sockaddr_un_len();
+
         return 0;
 }
diff --git a/src/tty-ask-password-agent/tty-ask-password-agent.c b/src/tty-ask-password-agent/tty-ask-password-agent.c
index c7ded451a2..ee879c7b89 100644
--- a/src/tty-ask-password-agent/tty-ask-password-agent.c
+++ b/src/tty-ask-password-agent/tty-ask-password-agent.c
@@ -65,8 +65,8 @@ static int ask_password_plymouth(
                 const char *flag_file,
                 char ***ret) {
 
+        static const union sockaddr_union sa = PLYMOUTH_SOCKET;
         _cleanup_close_ int fd = -1, notify = -1;
-        union sockaddr_union sa = PLYMOUTH_SOCKET;
         _cleanup_free_ char *packet = NULL;
         ssize_t k;
         int r, n;
@@ -94,7 +94,7 @@ static int ask_password_plymouth(
         if (fd < 0)
                 return -errno;
 
-        r = connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + 1 + strlen(sa.un.sun_path+1));
+        r = connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
         if (r < 0)
                 return -errno;
 
@@ -269,8 +269,7 @@ static int send_passwords(const char *socket_name, char **passwords) {
 
         strncpy(sa.un.sun_path, socket_name, sizeof(sa.un.sun_path));
 
-        r = sendto(socket_fd, packet, packet_length, MSG_NOSIGNAL, &sa.sa,
-                   offsetof(struct sockaddr_un, sun_path) + strlen(socket_name));
+        r = sendto(socket_fd, packet, packet_length, MSG_NOSIGNAL, &sa.sa, SOCKADDR_UN_LEN(sa.un));
         if (r < 0)
                 r = log_debug_errno(errno, "sendto(): %m");
 
diff --git a/src/udev/udev-ctrl.c b/src/udev/udev-ctrl.c
index 962de22f43..f68a09d7a8 100644
--- a/src/udev/udev-ctrl.c
+++ b/src/udev/udev-ctrl.c
@@ -105,7 +105,7 @@ struct udev_ctrl *udev_ctrl_new_from_fd(struct udev *udev, int fd) {
 
         uctrl->saddr.un.sun_family = AF_LOCAL;
         strscpy(uctrl->saddr.un.sun_path, sizeof(uctrl->saddr.un.sun_path), "/run/udev/control");
-        uctrl->addrlen = offsetof(struct sockaddr_un, sun_path) + strlen(uctrl->saddr.un.sun_path);
+        uctrl->addrlen = SOCKADDR_UN_LEN(uctrl->saddr.un);
         return uctrl;
 }