chiark / gitweb /
core: general cgroup rework
[elogind.git] / src / core / main.c
index 1326b94f689814ad6edeaec253cb29d74804cbc5..3c6fccf527d4adf7f0875b1a75cebce61182efac 100644 (file)
@@ -40,7 +40,7 @@
 #include "fdset.h"
 #include "special.h"
 #include "conf-parser.h"
-#include "bus-errors.h"
+#include "dbus-common.h"
 #include "missing.h"
 #include "label.h"
 #include "build.h"
 #include "switch-root.h"
 #include "capability.h"
 #include "killall.h"
+#include "env-util.h"
+#include "hwclock.h"
+#include "sd-daemon.h"
+#include "sd-messages.h"
 
 #include "mount-setup.h"
 #include "loopback-setup.h"
+#ifdef HAVE_KMOD
 #include "kmod-setup.h"
+#endif
 #include "hostname-setup.h"
 #include "machine-id-setup.h"
 #include "locale-setup.h"
-#include "hwclock.h"
 #include "selinux-setup.h"
 #include "ima-setup.h"
+#include "fileio.h"
+#include "smack-setup.h"
 
 static enum {
         ACTION_RUN,
@@ -73,7 +80,7 @@ static enum {
 } arg_action = ACTION_RUN;
 
 static char *arg_default_unit = NULL;
-static ManagerRunningAs arg_running_as = _MANAGER_RUNNING_AS_INVALID;
+static SystemdRunningAs arg_running_as = _SYSTEMD_RUNNING_AS_INVALID;
 
 static bool arg_dump_core = true;
 static bool arg_crash_shell = false;
@@ -81,12 +88,12 @@ static int arg_crash_chvt = -1;
 static bool arg_confirm_spawn = false;
 static bool arg_show_status = true;
 static bool arg_switched_root = false;
-static char **arg_default_controllers = NULL;
 static char ***arg_join_controllers = NULL;
 static ExecOutput arg_default_std_output = EXEC_OUTPUT_JOURNAL;
 static ExecOutput arg_default_std_error = EXEC_OUTPUT_INHERIT;
 static usec_t arg_runtime_watchdog = 0;
 static usec_t arg_shutdown_watchdog = 10 * USEC_PER_MINUTE;
+static char **arg_default_environment = NULL;
 static struct rlimit *arg_default_rlimit[RLIMIT_NLIMITS] = {};
 static uint64_t arg_capability_bounding_set_drop = 0;
 static nsec_t arg_timer_slack_nsec = (nsec_t) -1;
@@ -98,37 +105,40 @@ static void nop_handler(int sig) {
 
 _noreturn_ static void crash(int sig) {
 
-        if (!arg_dump_core)
+        if (getpid() != 1)
+                /* Pass this on immediately, if this is not PID 1 */
+                raise(sig);
+        else if (!arg_dump_core)
                 log_error("Caught <%s>, not dumping core.", signal_to_string(sig));
         else {
-                struct sigaction sa;
+                struct sigaction sa = {
+                        .sa_handler = nop_handler,
+                        .sa_flags = SA_NOCLDSTOP|SA_RESTART,
+                };
                 pid_t pid;
 
                 /* We want to wait for the core process, hence let's enable SIGCHLD */
-                zero(sa);
-                sa.sa_handler = nop_handler;
-                sa.sa_flags = SA_NOCLDSTOP|SA_RESTART;
-                assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
+                sigaction(SIGCHLD, &sa, NULL);
 
-                if ((pid = fork()) < 0)
+                pid = fork();
+                if (pid < 0)
                         log_error("Caught <%s>, cannot fork for core dump: %s", signal_to_string(sig), strerror(errno));
 
                 else if (pid == 0) {
-                        struct rlimit rl;
+                        struct rlimit rl = {};
 
                         /* Enable default signal handler for core dump */
                         zero(sa);
                         sa.sa_handler = SIG_DFL;
-                        assert_se(sigaction(sig, &sa, NULL) == 0);
+                        sigaction(sig, &sa, NULL);
 
                         /* Don't limit the core dump size */
-                        zero(rl);
                         rl.rlim_cur = RLIM_INFINITY;
                         rl.rlim_max = RLIM_INFINITY;
                         setrlimit(RLIMIT_CORE, &rl);
 
                         /* Just to be sure... */
-                        assert_se(chdir("/") == 0);
+                        chdir("/");
 
                         /* Raise the signal again */
                         raise(sig);
@@ -141,7 +151,8 @@ _noreturn_ static void crash(int sig) {
                         int r;
 
                         /* Order things nicely. */
-                        if ((r = wait_for_terminate(pid, &status)) < 0)
+                        r = wait_for_terminate(pid, &status);
+                        if (r < 0)
                                 log_error("Caught <%s>, waitpid() failed: %s", signal_to_string(sig), strerror(-r));
                         else if (status.si_code != CLD_DUMPED)
                                 log_error("Caught <%s>, core dump failed.", signal_to_string(sig));
@@ -154,26 +165,26 @@ _noreturn_ static void crash(int sig) {
                 chvt(arg_crash_chvt);
 
         if (arg_crash_shell) {
-                struct sigaction sa;
+                struct sigaction sa = {
+                        .sa_handler = SIG_IGN,
+                        .sa_flags = SA_NOCLDSTOP|SA_NOCLDWAIT|SA_RESTART,
+                };
                 pid_t pid;
 
                 log_info("Executing crash shell in 10s...");
                 sleep(10);
 
                 /* Let the kernel reap children for us */
-                zero(sa);
-                sa.sa_handler = SIG_IGN;
-                sa.sa_flags = SA_NOCLDSTOP|SA_NOCLDWAIT|SA_RESTART;
                 assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
 
                 pid = fork();
                 if (pid < 0)
-                        log_error("Failed to fork off crash shell: %s", strerror(errno));
+                        log_error("Failed to fork off crash shell: %m");
                 else if (pid == 0) {
                         make_console_stdio();
                         execl("/bin/sh", "/bin/sh", NULL);
 
-                        log_error("execl() failed: %s", strerror(errno));
+                        log_error("execl() failed: %m");
                         _exit(1);
                 }
 
@@ -185,12 +196,10 @@ _noreturn_ static void crash(int sig) {
 }
 
 static void install_crash_handler(void) {
-        struct sigaction sa;
-
-        zero(sa);
-
-        sa.sa_handler = crash;
-        sa.sa_flags = SA_NODEFER;
+        struct sigaction sa = {
+                .sa_handler = crash,
+                .sa_flags = SA_NODEFER,
+        };
 
         sigaction_many(&sa, SIGNALS_CRASH_HANDLER, -1);
 }
@@ -339,7 +348,8 @@ static int parse_proc_cmdline_word(const char *word) {
                 else
                         arg_default_std_error = r;
         } else if (startswith(word, "systemd.setenv=")) {
-                char *cenv, *eq;
+                _cleanup_free_ char *cenv = NULL;
+                char *eq;
                 int r;
 
                 cenv = strdup(word + 15);
@@ -348,43 +358,63 @@ static int parse_proc_cmdline_word(const char *word) {
 
                 eq = strchr(cenv, '=');
                 if (!eq) {
-                        r = unsetenv(cenv);
-                        if (r < 0)
-                                log_warning("unsetenv failed %s. Ignoring.", strerror(errno));
+                        if (!env_name_is_valid(cenv))
+                                log_warning("Environment variable name '%s' is not valid. Ignoring.", cenv);
+                        else  {
+                                r = unsetenv(cenv);
+                                if (r < 0)
+                                        log_warning("Unsetting environment variable '%s' failed, ignoring: %m", cenv);
+                        }
                 } else {
-                        *eq = 0;
-                        r = setenv(cenv, eq + 1, 1);
-                        if (r < 0)
-                                log_warning("setenv failed %s. Ignoring.", strerror(errno));
+                        if (!env_assignment_is_valid(cenv))
+                                log_warning("Environment variable assignment '%s' is not valid. Ignoring.", cenv);
+                        else {
+                                *eq = 0;
+                                r = setenv(cenv, eq + 1, 1);
+                                if (r < 0)
+                                        log_warning("Setting environment variable '%s=%s' failed, ignoring: %m", cenv, eq + 1);
+                        }
                 }
-                free(cenv);
 
         } else if (startswith(word, "systemd.") ||
                    (in_initrd() && startswith(word, "rd.systemd."))) {
 
-                log_warning("Unknown kernel switch %s. Ignoring.", word);
-
-                log_info("Supported kernel switches:\n"
-                         "systemd.unit=UNIT                        Default unit to start\n"
-                         "rd.systemd.unit=UNIT                     Default unit to start when run in initrd\n"
-                         "systemd.dump_core=0|1                    Dump core on crash\n"
-                         "systemd.crash_shell=0|1                  Run shell on crash\n"
-                         "systemd.crash_chvt=N                     Change to VT #N on crash\n"
-                         "systemd.confirm_spawn=0|1                Confirm every process spawn\n"
-                         "systemd.show_status=0|1                  Show status updates on the console during bootup\n"
-                         "systemd.log_target=console|kmsg|journal|journal-or-kmsg|syslog|syslog-or-kmsg|null\n"
-                         "                                         Log target\n"
-                         "systemd.log_level=LEVEL                  Log level\n"
-                         "systemd.log_color=0|1                    Highlight important log messages\n"
-                         "systemd.log_location=0|1                 Include code location in log messages\n"
-                         "systemd.default_standard_output=null|tty|syslog|syslog+console|kmsg|kmsg+console|journal|journal+console\n"
-                         "                                         Set default log output for services\n"
-                         "systemd.default_standard_error=null|tty|syslog|syslog+console|kmsg|kmsg+console|journal|journal+console\n"
-                         "                                         Set default log error output for services\n"
-                         "systemd.setenv=ASSIGNMENT                Set an environment variable for all spawned processes\n");
+                const char *c;
+
+                /* Ignore systemd.journald.xyz and friends */
+                c = word;
+                if (startswith(c, "rd."))
+                        c += 3;
+                if (startswith(c, "systemd."))
+                        c += 8;
+                if (c[strcspn(c, ".=")] != '.')  {
+
+                        log_warning("Unknown kernel switch %s. Ignoring.", word);
+
+                        log_info("Supported kernel switches:\n"
+                                 "systemd.unit=UNIT                        Default unit to start\n"
+                                 "rd.systemd.unit=UNIT                     Default unit to start when run in initrd\n"
+                                 "systemd.dump_core=0|1                    Dump core on crash\n"
+                                 "systemd.crash_shell=0|1                  Run shell on crash\n"
+                                 "systemd.crash_chvt=N                     Change to VT #N on crash\n"
+                                 "systemd.confirm_spawn=0|1                Confirm every process spawn\n"
+                                 "systemd.show_status=0|1                  Show status updates on the console during bootup\n"
+                                 "systemd.log_target=console|kmsg|journal|journal-or-kmsg|syslog|syslog-or-kmsg|null\n"
+                                 "                                         Log target\n"
+                                 "systemd.log_level=LEVEL                  Log level\n"
+                                 "systemd.log_color=0|1                    Highlight important log messages\n"
+                                 "systemd.log_location=0|1                 Include code location in log messages\n"
+                                 "systemd.default_standard_output=null|tty|syslog|syslog+console|kmsg|kmsg+console|journal|journal+console\n"
+                                 "                                         Set default log output for services\n"
+                                 "systemd.default_standard_error=null|tty|syslog|syslog+console|kmsg|kmsg+console|journal|journal+console\n"
+                                 "                                         Set default log error output for services\n"
+                                 "systemd.setenv=ASSIGNMENT                Set an environment variable for all spawned processes\n");
+                }
 
         } else if (streq(word, "quiet"))
                 arg_show_status = false;
+        else if (streq(word, "debug"))
+                log_set_max_level(LOG_DEBUG);
         else if (!in_initrd()) {
                 unsigned i;
 
@@ -397,87 +427,47 @@ static int parse_proc_cmdline_word(const char *word) {
         return 0;
 }
 
-static int config_parse_level2(
-                const char *filename,
-                unsigned line,
-                const char *section,
-                const char *lvalue,
-                int ltype,
-                const char *rvalue,
-                void *data,
-                void *userdata) {
-
-        assert(filename);
-        assert(lvalue);
-        assert(rvalue);
-
-        log_set_max_level_from_string(rvalue);
-        return 0;
-}
-
-static int config_parse_target(
-                const char *filename,
-                unsigned line,
-                const char *section,
-                const char *lvalue,
-                int ltype,
-                const char *rvalue,
-                void *data,
-                void *userdata) {
-
-        assert(filename);
-        assert(lvalue);
-        assert(rvalue);
-
-        log_set_target_from_string(rvalue);
-        return 0;
-}
-
-static int config_parse_color(
-                const char *filename,
-                unsigned line,
-                const char *section,
-                const char *lvalue,
-                int ltype,
-                const char *rvalue,
-                void *data,
-                void *userdata) {
-
-        assert(filename);
-        assert(lvalue);
-        assert(rvalue);
-
-        log_show_color_from_string(rvalue);
-        return 0;
-}
+#define DEFINE_SETTER(name, func, descr)                              \
+        static int name(const char *unit,                             \
+                        const char *filename,                         \
+                        unsigned line,                                \
+                        const char *section,                          \
+                        const char *lvalue,                           \
+                        int ltype,                                    \
+                        const char *rvalue,                           \
+                        void *data,                                   \
+                        void *userdata) {                             \
+                                                                      \
+                int r;                                                \
+                                                                      \
+                assert(filename);                                     \
+                assert(lvalue);                                       \
+                assert(rvalue);                                       \
+                                                                      \
+                r = func(rvalue);                                     \
+                if (r < 0)                                            \
+                        log_syntax(unit, LOG_ERR, filename, line, -r, \
+                                   "Invalid " descr "'%s': %s",       \
+                                   rvalue, strerror(-r));             \
+                                                                      \
+                return 0;                                             \
+        }
 
-static int config_parse_location(
-                const char *filename,
-                unsigned line,
-                const char *section,
-                const char *lvalue,
-                int ltype,
-                const char *rvalue,
-                void *data,
-                void *userdata) {
+DEFINE_SETTER(config_parse_level2, log_set_max_level_from_string, "log level")
+DEFINE_SETTER(config_parse_target, log_set_target_from_string, "target")
+DEFINE_SETTER(config_parse_color, log_show_color_from_string, "color" )
+DEFINE_SETTER(config_parse_location, log_show_location_from_string, "location")
 
-        assert(filename);
-        assert(lvalue);
-        assert(rvalue);
 
-        log_show_location_from_string(rvalue);
-        return 0;
-}
-
-static int config_parse_cpu_affinity2(
-                const char *filename,
-                unsigned line,
-                const char *section,
-                const char *lvalue,
-                int ltype,
-                const char *rvalue,
-                void *data,
-                void *userdata) {
+static int config_parse_cpu_affinity2(const char *unit,
+                                      const char *filename,
+                                      unsigned line,
+                                      const char *section,
+                                      const char *lvalue,
+                                      int ltype,
+                                      const char *rvalue,
+                                      void *data,
+                                      void *userdata) {
 
         char *w;
         size_t l;
@@ -495,17 +485,18 @@ static int config_parse_cpu_affinity2(
                 unsigned cpu;
 
                 if (!(t = strndup(w, l)))
-                        return -ENOMEM;
+                        return log_oom();
 
                 r = safe_atou(t, &cpu);
                 free(t);
 
                 if (!c)
                         if (!(c = cpu_set_malloc(&ncpus)))
-                                return -ENOMEM;
+                                return log_oom();
 
                 if (r < 0 || cpu >= ncpus) {
-                        log_error("[%s:%u] Failed to parse CPU affinity: %s", filename, line, rvalue);
+                        log_syntax(unit, LOG_ERR, filename, line, -r,
+                                   "Failed to parse CPU affinity '%s'", rvalue);
                         CPU_FREE(c);
                         return -EBADMSG;
                 }
@@ -515,7 +506,7 @@ static int config_parse_cpu_affinity2(
 
         if (c) {
                 if (sched_setaffinity(0, CPU_ALLOC_SIZE(ncpus), c) < 0)
-                        log_warning("Failed to set CPU affinity: %m");
+                        log_warning_unit(unit, "Failed to set CPU affinity: %m");
 
                 CPU_FREE(c);
         }
@@ -536,22 +527,19 @@ static void strv_free_free(char ***l) {
 }
 
 static void free_join_controllers(void) {
-        if (!arg_join_controllers)
-                return;
-
         strv_free_free(arg_join_controllers);
         arg_join_controllers = NULL;
 }
 
-static int config_parse_join_controllers(
-                const char *filename,
-                unsigned line,
-                const char *section,
-                const char *lvalue,
-                int ltype,
-                const char *rvalue,
-                void *data,
-                void *userdata) {
+static int config_parse_join_controllers(const char *unit,
+                                         const char *filename,
+                                         unsigned line,
+                                         const char *section,
+                                         const char *lvalue,
+                                         int ltype,
+                                         const char *rvalue,
+                                         void *data,
+                                         void *userdata) {
 
         unsigned n = 0;
         char *state, *w;
@@ -568,7 +556,7 @@ static int config_parse_join_controllers(
 
                 s = strndup(w, length);
                 if (!s)
-                        return -ENOMEM;
+                        return log_oom();
 
                 l = strv_split(s, ",");
                 free(s);
@@ -584,7 +572,7 @@ static int config_parse_join_controllers(
                         arg_join_controllers = new(char**, 2);
                         if (!arg_join_controllers) {
                                 strv_free(l);
-                                return -ENOMEM;
+                                return log_oom();
                         }
 
                         arg_join_controllers[0] = l;
@@ -598,7 +586,7 @@ static int config_parse_join_controllers(
                         t = new0(char**, n+2);
                         if (!t) {
                                 strv_free(l);
-                                return -ENOMEM;
+                                return log_oom();
                         }
 
                         n = 0;
@@ -612,7 +600,7 @@ static int config_parse_join_controllers(
                                         if (!c) {
                                                 strv_free(l);
                                                 strv_free_free(t);
-                                                return -ENOMEM;
+                                                return log_oom();
                                         }
 
                                         strv_free(l);
@@ -624,7 +612,7 @@ static int config_parse_join_controllers(
                                         if (!c) {
                                                 strv_free(l);
                                                 strv_free_free(t);
-                                                return -ENOMEM;
+                                                return log_oom();
                                         }
 
                                         t[n++] = c;
@@ -653,14 +641,14 @@ static int parse_config_file(void) {
                 { "Manager", "ShowStatus",            config_parse_bool,         0, &arg_show_status         },
                 { "Manager", "CrashChVT",             config_parse_int,          0, &arg_crash_chvt          },
                 { "Manager", "CPUAffinity",           config_parse_cpu_affinity2, 0, NULL                    },
-                { "Manager", "DefaultControllers",    config_parse_strv,         0, &arg_default_controllers },
                 { "Manager", "DefaultStandardOutput", config_parse_output,       0, &arg_default_std_output  },
                 { "Manager", "DefaultStandardError",  config_parse_output,       0, &arg_default_std_error   },
                 { "Manager", "JoinControllers",       config_parse_join_controllers, 0, &arg_join_controllers },
-                { "Manager", "RuntimeWatchdogSec",    config_parse_usec,         0, &arg_runtime_watchdog    },
-                { "Manager", "ShutdownWatchdogSec",   config_parse_usec,         0, &arg_shutdown_watchdog   },
+                { "Manager", "RuntimeWatchdogSec",    config_parse_sec,          0, &arg_runtime_watchdog    },
+                { "Manager", "ShutdownWatchdogSec",   config_parse_sec,          0, &arg_shutdown_watchdog   },
                 { "Manager", "CapabilityBoundingSet", config_parse_bounding_set, 0, &arg_capability_bounding_set_drop },
                 { "Manager", "TimerSlackNSec",        config_parse_nsec,         0, &arg_timer_slack_nsec    },
+                { "Manager", "DefaultEnvironment",    config_parse_environ,      0, &arg_default_environment },
                 { "Manager", "DefaultLimitCPU",       config_parse_limit,        0, &arg_default_rlimit[RLIMIT_CPU]},
                 { "Manager", "DefaultLimitFSIZE",     config_parse_limit,        0, &arg_default_rlimit[RLIMIT_FSIZE]},
                 { "Manager", "DefaultLimitDATA",      config_parse_limit,        0, &arg_default_rlimit[RLIMIT_DATA]},
@@ -680,11 +668,11 @@ static int parse_config_file(void) {
                 { NULL, NULL, NULL, 0, NULL }
         };
 
-        FILE *f;
+        _cleanup_fclose_ FILE *f;
         const char *fn;
         int r;
 
-        fn = arg_running_as == MANAGER_SYSTEM ? SYSTEM_CONFIG_FILE : USER_CONFIG_FILE;
+        fn = arg_running_as == SYSTEMD_SYSTEM ? PKGSYSCONFDIR "/system.conf" : PKGSYSCONFDIR "/user.conf";
         f = fopen(fn, "re");
         if (!f) {
                 if (errno == ENOENT)
@@ -694,17 +682,16 @@ static int parse_config_file(void) {
                 return 0;
         }
 
-        r = config_parse(fn, f, "Manager\0", config_item_table_lookup, (void*) items, false, NULL);
+        r = config_parse(NULL, fn, f, "Manager\0", config_item_table_lookup, (void*) items, false, false, NULL);
         if (r < 0)
                 log_warning("Failed to parse configuration file: %s", strerror(-r));
 
-        fclose(f);
-
         return 0;
 }
 
 static int parse_proc_cmdline(void) {
-        char *line, *w, *state;
+        _cleanup_free_ char *line = NULL;
+        char *w, *state;
         int r;
         size_t l;
 
@@ -713,31 +700,27 @@ static int parse_proc_cmdline(void) {
         if (detect_container(NULL) > 0)
                 return 0;
 
-        if ((r = read_one_line_file("/proc/cmdline", &line)) < 0) {
+        r = read_one_line_file("/proc/cmdline", &line);
+        if (r < 0) {
                 log_warning("Failed to read /proc/cmdline, ignoring: %s", strerror(-r));
                 return 0;
         }
 
         FOREACH_WORD_QUOTED(w, l, line, state) {
-                char *word;
+                _cleanup_free_ char *word;
 
-                if (!(word = strndup(w, l))) {
-                        r = -ENOMEM;
-                        goto finish;
-                }
+                word = strndup(w, l);
+                if (!word)
+                        return log_oom();
 
                 r = parse_proc_cmdline_word(word);
-                free(word);
-
-                if (r < 0)
-                        goto finish;
+                if (r < 0) {
+                        log_error("Failed on cmdline argument %s: %s", word, strerror(-r));
+                        return r;
+                }
         }
 
-        r = 0;
-
-finish:
-        free(line);
-        return r;
+        return 0;
 }
 
 static int parse_argv(int argc, char *argv[]) {
@@ -869,11 +852,11 @@ static int parse_argv(int argc, char *argv[]) {
                         break;
 
                 case ARG_SYSTEM:
-                        arg_running_as = MANAGER_SYSTEM;
+                        arg_running_as = SYSTEMD_SYSTEM;
                         break;
 
                 case ARG_USER:
-                        arg_running_as = MANAGER_USER;
+                        arg_running_as = SYSTEMD_USER;
                         break;
 
                 case ARG_TEST:
@@ -928,14 +911,18 @@ static int parse_argv(int argc, char *argv[]) {
                         int fd;
                         FILE *f;
 
-                        if ((r = safe_atoi(optarg, &fd)) < 0 || fd < 0) {
+                        r = safe_atoi(optarg, &fd);
+                        if (r < 0 || fd < 0) {
                                 log_error("Failed to parse deserialize option %s.", optarg);
-                                return r;
+                                return r < 0 ? r : -EINVAL;
                         }
 
-                        if (!(f = fdopen(fd, "r"))) {
+                        fd_cloexec(fd, true);
+
+                        f = fdopen(fd, "r");
+                        if (!f) {
                                 log_error("Failed to open serialization fd: %m");
-                                return r;
+                                return -errno;
                         }
 
                         if (serialization)
@@ -1017,8 +1004,10 @@ static int parse_argv(int argc, char *argv[]) {
                  * instead. */
 
                 for (a = argv; a < argv + argc; a++)
-                        if ((r = parse_proc_cmdline_word(*a)) < 0)
+                        if ((r = parse_proc_cmdline_word(*a)) < 0) {
+                                log_error("Failed on cmdline argument %s: %s", *a, strerror(-r));
                                 return r;
+                        }
         }
 
         return 0;
@@ -1052,13 +1041,12 @@ static int help(void) {
 
 static int version(void) {
         puts(PACKAGE_STRING);
-        puts(DISTRIBUTION);
         puts(SYSTEMD_FEATURES);
 
         return 0;
 }
 
-static int prepare_reexecute(Manager *m, FILE **_f, FDSet **_fds, bool serialize_jobs) {
+static int prepare_reexecute(Manager *m, FILE **_f, FDSet **_fds, bool switching_root) {
         FILE *f = NULL;
         FDSet *fds = NULL;
         int r;
@@ -1083,7 +1071,7 @@ static int prepare_reexecute(Manager *m, FILE **_f, FDSet **_fds, bool serialize
                 goto fail;
         }
 
-        r = manager_serialize(m, f, fds, serialize_jobs);
+        r = manager_serialize(m, f, fds, switching_root);
         if (r < 0) {
                 log_error("Failed to serialize state: %s", strerror(-r));
                 goto fail;
@@ -1120,6 +1108,42 @@ fail:
         return r;
 }
 
+static int bump_rlimit_nofile(struct rlimit *saved_rlimit) {
+        struct rlimit nl;
+        int r;
+
+        assert(saved_rlimit);
+
+        /* Save the original RLIMIT_NOFILE so that we can reset it
+         * later when transitioning from the initrd to the main
+         * systemd or suchlike. */
+        if (getrlimit(RLIMIT_NOFILE, saved_rlimit) < 0) {
+                log_error("Reading RLIMIT_NOFILE failed: %m");
+                return -errno;
+        }
+
+        /* Make sure forked processes get the default kernel setting */
+        if (!arg_default_rlimit[RLIMIT_NOFILE]) {
+                struct rlimit *rl;
+
+                rl = newdup(struct rlimit, saved_rlimit, 1);
+                if (!rl)
+                        return log_oom();
+
+                arg_default_rlimit[RLIMIT_NOFILE] = rl;
+        }
+
+        /* Bump up the resource limit for ourselves substantially */
+        nl.rlim_cur = nl.rlim_max = 64*1024;
+        r = setrlimit_closest(RLIMIT_NOFILE, &nl);
+        if (r < 0) {
+                log_error("Setting RLIMIT_NOFILE failed: %s", strerror(-r));
+                return r;
+        }
+
+        return 0;
+}
+
 static struct dual_timestamp* parse_initrd_timestamp(struct dual_timestamp *t) {
         const char *e;
         unsigned long long a, b;
@@ -1186,6 +1210,28 @@ static void test_cgroups(void) {
         sleep(10);
 }
 
+static int initialize_join_controllers(void) {
+        /* By default, mount "cpu" + "cpuacct" together, and "net_cls"
+         * + "net_prio". We'd like to add "cpuset" to the mix, but
+         * "cpuset" does't really work for groups with no initialized
+         * attributes. */
+
+        arg_join_controllers = new(char**, 3);
+        if (!arg_join_controllers)
+                return -ENOMEM;
+
+        arg_join_controllers[0] = strv_new("cpu", "cpuacct", NULL);
+        arg_join_controllers[1] = strv_new("net_cls", "net_prio", NULL);
+        arg_join_controllers[2] = NULL;
+
+        if (!arg_join_controllers[0] || !arg_join_controllers[1]) {
+                free_join_controllers();
+                return -ENOMEM;
+        }
+
+        return 0;
+}
+
 int main(int argc, char *argv[]) {
         Manager *m = NULL;
         int r, retval = EXIT_FAILURE;
@@ -1195,6 +1241,8 @@ int main(int argc, char *argv[]) {
         bool reexecute = false;
         const char *shutdown_verb = NULL;
         dual_timestamp initrd_timestamp = { 0ULL, 0ULL };
+        dual_timestamp userspace_timestamp = { 0ULL, 0ULL };
+        dual_timestamp kernel_timestamp = { 0ULL, 0ULL };
         static char systemd[] = "systemd";
         bool skip_setup = false;
         int j;
@@ -1202,6 +1250,7 @@ int main(int argc, char *argv[]) {
         bool arm_reboot_watchdog = false;
         bool queue_default_job = false;
         char *switch_root_dir = NULL, *switch_root_init = NULL;
+        static struct rlimit saved_rlimit_nofile = { 0, 0 };
 
 #ifdef HAVE_SYSV_COMPAT
         if (getpid() != 1 && strstr(program_invocation_short_name, "init")) {
@@ -1215,22 +1264,19 @@ int main(int argc, char *argv[]) {
         }
 #endif
 
+        dual_timestamp_from_monotonic(&kernel_timestamp, 0);
+        dual_timestamp_get(&userspace_timestamp);
+
         /* Determine if this is a reexecution or normal bootup. We do
          * the full command line parsing much later, so let's just
          * have a quick peek here. */
-        for (j = 1; j < argc; j++)
-                if (streq(argv[j], "--deserialize")) {
-                        skip_setup = true;
-                        break;
-                }
+        if (strv_find(argv+1, "--deserialize"))
+                skip_setup = true;
 
         /* If we have switched root, do all the special setup
          * things */
-        for (j = 1; j < argc; j++)
-                if (streq(argv[j], "--switched-root")) {
-                        skip_setup = false;
-                        break;
-                }
+        if (strv_find(argv+1, "--switched-root"))
+                skip_setup = false;
 
         /* If we get started via the /sbin/init symlink then we are
            called 'init'. After a subsequent reexecution we are then
@@ -1243,14 +1289,19 @@ int main(int argc, char *argv[]) {
         saved_argc = argc;
 
         log_show_color(isatty(STDERR_FILENO) > 0);
-        log_show_location(false);
-        log_set_max_level(LOG_INFO);
 
-        if (getpid() == 1) {
+        if (getpid() == 1 && detect_container(NULL) <= 0) {
+
+                /* Running outside of a container as PID 1 */
+                arg_running_as = SYSTEMD_SYSTEM;
+                make_null_stdio();
+                log_set_target(LOG_TARGET_KMSG);
+                log_open();
+
                 if (in_initrd()) {
                         char *rd_timestamp = NULL;
 
-                        dual_timestamp_get(&initrd_timestamp);
+                        initrd_timestamp = userspace_timestamp;
                         asprintf(&rd_timestamp, "%llu %llu",
                                  (unsigned long long) initrd_timestamp.realtime,
                                  (unsigned long long) initrd_timestamp.monotonic);
@@ -1260,56 +1311,95 @@ int main(int argc, char *argv[]) {
                         }
                 }
 
-                arg_running_as = MANAGER_SYSTEM;
-                log_set_target(detect_container(NULL) > 0 ? LOG_TARGET_JOURNAL : LOG_TARGET_JOURNAL_OR_KMSG);
-
                 if (!skip_setup) {
+                        mount_setup_early();
                         if (selinux_setup(&loaded_policy) < 0)
                                 goto finish;
                         if (ima_setup() < 0)
                                 goto finish;
+                        if (smack_setup() < 0)
+                                goto finish;
                 }
 
-                log_open();
-
                 if (label_init(NULL) < 0)
                         goto finish;
 
-                if (!skip_setup)
+                if (!skip_setup) {
                         if (hwclock_is_localtime() > 0) {
                                 int min;
 
-                                r = hwclock_apply_localtime_delta(&min);
+                                /* The first-time call to settimeofday() does a time warp in the kernel */
+                                r = hwclock_set_timezone(&min);
                                 if (r < 0)
                                         log_error("Failed to apply local time delta, ignoring: %s", strerror(-r));
                                 else
                                         log_info("RTC configured in localtime, applying delta of %i minutes to system time.", min);
+                        } else if (!in_initrd()) {
+                                /*
+                                 * Do dummy first-time call to seal the kernel's time warp magic
+                                 *
+                                 * Do not call this this from inside the initrd. The initrd might not
+                                 * carry /etc/adjtime with LOCAL, but the real system could be set up
+                                 * that way. In such case, we need to delay the time-warp or the sealing
+                                 * until we reach the real system.
+                                 */
+                                hwclock_reset_timezone();
+
+                                /* Tell the kernel our time zone */
+                                r = hwclock_set_timezone(NULL);
+                                if (r < 0)
+                                        log_error("Failed to set the kernel's time zone, ignoring: %s", strerror(-r));
                         }
+                }
+
+                /* Set the default for later on, but don't actually
+                 * open the logs like this for now. Note that if we
+                 * are transitioning from the initrd there might still
+                 * be journal fd open, and we shouldn't attempt
+                 * opening that before we parsed /proc/cmdline which
+                 * might redirect output elsewhere. */
+                log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
+
+        } else if (getpid() == 1) {
+                /* Running inside a container, as PID 1 */
+                arg_running_as = SYSTEMD_SYSTEM;
+                log_set_target(LOG_TARGET_CONSOLE);
+                log_open();
+
+                /* For the later on, see above... */
+                log_set_target(LOG_TARGET_JOURNAL);
+
+                /* clear the kernel timestamp,
+                 * because we are in a container */
+                kernel_timestamp.monotonic = 0ULL;
+                kernel_timestamp.realtime = 0ULL;
 
         } else {
-                arg_running_as = MANAGER_USER;
+                /* Running as user instance */
+                arg_running_as = SYSTEMD_USER;
                 log_set_target(LOG_TARGET_AUTO);
                 log_open();
+
+                /* clear the kernel timestamp,
+                 * because we are not PID 1 */
+                kernel_timestamp.monotonic = 0ULL;
+                kernel_timestamp.realtime = 0ULL;
         }
 
         /* Initialize default unit */
-        if (set_default_unit(SPECIAL_DEFAULT_TARGET) < 0)
-                goto finish;
-
-        /* By default, mount "cpu" and "cpuacct" together */
-        arg_join_controllers = new(char**, 2);
-        if (!arg_join_controllers)
+        r = set_default_unit(SPECIAL_DEFAULT_TARGET);
+        if (r < 0) {
+                log_error("Failed to set default unit %s: %s", SPECIAL_DEFAULT_TARGET, strerror(-r));
                 goto finish;
+        }
 
-        arg_join_controllers[0] = strv_new("cpu", "cpuacct", NULL);
-        arg_join_controllers[1] = NULL;
-
-        if (!arg_join_controllers[0])
+        r = initialize_join_controllers();
+        if (r < 0)
                 goto finish;
 
         /* Mount /proc, /sys and friends, so that /proc/cmdline and
          * /proc/$PID/fd is available. */
-        if (geteuid() == 0 && !getenv("SYSTEMD_SKIP_API_MOUNTS")) {
+        if (getpid() == 1) {
                 r = mount_setup(loaded_policy);
                 if (r < 0)
                         goto finish;
@@ -1318,13 +1408,12 @@ int main(int argc, char *argv[]) {
         /* Reset all signal handlers. */
         assert_se(reset_all_signal_handlers() == 0);
 
-        /* If we are init, we can block sigkill. Yay. */
         ignore_signals(SIGNALS_IGNORE, -1);
 
         if (parse_config_file() < 0)
                 goto finish;
 
-        if (arg_running_as == MANAGER_SYSTEM)
+        if (arg_running_as == SYSTEMD_SYSTEM)
                 if (parse_proc_cmdline() < 0)
                         goto finish;
 
@@ -1333,12 +1422,20 @@ int main(int argc, char *argv[]) {
         if (parse_argv(argc, argv) < 0)
                 goto finish;
 
-        if (arg_action == ACTION_TEST && geteuid() == 0) {
+        if (arg_action == ACTION_TEST &&
+            geteuid() == 0) {
                 log_error("Don't run test mode as root.");
                 goto finish;
         }
 
-        if (arg_running_as == MANAGER_SYSTEM &&
+        if (arg_running_as == SYSTEMD_USER &&
+            arg_action == ACTION_RUN &&
+            sd_booted() <= 0) {
+                log_error("Trying to run as user instance, but the system has not been booted with systemd.");
+                goto finish;
+        }
+
+        if (arg_running_as == SYSTEMD_SYSTEM &&
             arg_action == ACTION_RUN &&
             running_in_chroot() > 0) {
                 log_error("Cannot be run in a chroot() environment.");
@@ -1366,16 +1463,15 @@ int main(int argc, char *argv[]) {
         log_close();
 
         /* Remember open file descriptors for later deserialization */
-        if (serialization) {
-                r = fdset_new_fill(&fds);
-                if (r < 0) {
-                        log_error("Failed to allocate fd set: %s", strerror(-r));
-                        goto finish;
-                }
+        r = fdset_new_fill(&fds);
+        if (r < 0) {
+                log_error("Failed to allocate fd set: %s", strerror(-r));
+                goto finish;
+        } else
+                fdset_cloexec(fds, true);
 
+        if (serialization)
                 assert_se(fdset_remove(fds, fileno(serialization)) >= 0);
-        } else
-                close_all_fds(NULL, 0);
 
         /* Set up PATH unless it is already set */
         setenv("PATH",
@@ -1384,9 +1480,9 @@ int main(int argc, char *argv[]) {
 #else
                "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin",
 #endif
-               arg_running_as == MANAGER_SYSTEM);
+               arg_running_as == SYSTEMD_SYSTEM);
 
-        if (arg_running_as == MANAGER_SYSTEM) {
+        if (arg_running_as == SYSTEMD_SYSTEM) {
                 /* Parse the data passed to us. We leave this
                  * variables set, but the manager later on will not
                  * pass them on to our children. */
@@ -1410,6 +1506,12 @@ int main(int argc, char *argv[]) {
                 unsetenv("USER");
                 unsetenv("LOGNAME");
 
+                /* We suppress the socket activation env vars, as
+                 * we'll try to match *any* open fd to units if
+                 * possible. */
+                unsetenv("LISTEN_FDS");
+                unsetenv("LISTEN_PID");
+
                 /* All other variables are left as is, so that clients
                  * can still read them via /proc/1/environ */
         }
@@ -1417,7 +1519,7 @@ int main(int argc, char *argv[]) {
         /* Move out of the way, so that we won't block unmounts */
         assert_se(chdir("/")  == 0);
 
-        if (arg_running_as == MANAGER_SYSTEM) {
+        if (arg_running_as == SYSTEMD_SYSTEM) {
                 /* Become a session leader if we aren't one yet. */
                 setsid();
 
@@ -1430,29 +1532,26 @@ int main(int argc, char *argv[]) {
 
         /* Reset the console, but only if this is really init and we
          * are freshly booted */
-        if (arg_running_as == MANAGER_SYSTEM && arg_action == ACTION_RUN) {
+        if (arg_running_as == SYSTEMD_SYSTEM && arg_action == ACTION_RUN)
                 console_setup(getpid() == 1 && !skip_setup);
-                make_null_stdio();
-        }
 
         /* Open the logging devices, if possible and necessary */
         log_open();
 
         /* Make sure we leave a core dump without panicing the
          * kernel. */
-        if (getpid() == 1)
+        if (getpid() == 1) {
                 install_crash_handler();
 
-        if (geteuid() == 0 && !getenv("SYSTEMD_SKIP_API_MOUNTS")) {
                 r = mount_cgroup_controllers(arg_join_controllers);
                 if (r < 0)
                         goto finish;
         }
 
-        if (arg_running_as == MANAGER_SYSTEM) {
+        if (arg_running_as == SYSTEMD_SYSTEM) {
                 const char *virtualization = NULL;
 
-                log_info(PACKAGE_STRING " running in system mode. (" SYSTEMD_FEATURES "; " DISTRIBUTION ")");
+                log_info(PACKAGE_STRING " running in system mode. (" SYSTEMD_FEATURES ")");
 
                 detect_virtualization(&virtualization);
                 if (virtualization)
@@ -1462,15 +1561,17 @@ int main(int argc, char *argv[]) {
                         log_info("Running in initial RAM disk.");
 
         } else
-                log_debug(PACKAGE_STRING " running in user mode. (" SYSTEMD_FEATURES "; " DISTRIBUTION ")");
+                log_debug(PACKAGE_STRING " running in user mode. (" SYSTEMD_FEATURES ")");
 
-        if (arg_running_as == MANAGER_SYSTEM && !skip_setup) {
+        if (arg_running_as == SYSTEMD_SYSTEM && !skip_setup) {
                 locale_setup();
 
                 if (arg_show_status || plymouth_running())
                         status_welcome();
 
+#ifdef HAVE_KMOD
                 kmod_setup();
+#endif
                 hostname_setup();
                 machine_id_setup();
                 loopback_setup();
@@ -1480,7 +1581,7 @@ int main(int argc, char *argv[]) {
                 test_cgroups();
         }
 
-        if (arg_running_as == MANAGER_SYSTEM && arg_runtime_watchdog > 0)
+        if (arg_running_as == SYSTEMD_SYSTEM && arg_runtime_watchdog > 0)
                 watchdog_set_timeout(&arg_runtime_watchdog);
 
         if (arg_timer_slack_nsec != (nsec_t) -1)
@@ -1500,15 +1601,18 @@ int main(int argc, char *argv[]) {
                 }
         }
 
-        if (arg_running_as == MANAGER_USER) {
+        if (arg_running_as == SYSTEMD_USER) {
                 /* Become reaper of our children */
-                r = prctl(PR_SET_CHILD_SUBREAPER, 1);
-                if (r < 0)
-                        log_error("Failed to prctl(PR_SET_CHILD_SUBREAPER): %s", strerror(-r));
-                if (r == -EINVAL)
-                        log_error("Perhaps the kernel version is too old (< 3.4?)");
+                if (prctl(PR_SET_CHILD_SUBREAPER, 1) < 0) {
+                        log_warning("Failed to make us a subreaper: %m");
+                        if (errno == EINVAL)
+                                log_info("Perhaps the kernel version is too old (< 3.4?)");
+                }
         }
 
+        if (arg_running_as == SYSTEMD_SYSTEM)
+                bump_rlimit_nofile(&saved_rlimit_nofile);
+
         r = manager_new(arg_running_as, &m);
         if (r < 0) {
                 log_error("Failed to allocate manager object: %s", strerror(-r));
@@ -1520,14 +1624,14 @@ int main(int argc, char *argv[]) {
         m->default_std_error = arg_default_std_error;
         m->runtime_watchdog = arg_runtime_watchdog;
         m->shutdown_watchdog = arg_shutdown_watchdog;
+        m->userspace_timestamp = userspace_timestamp;
+        m->kernel_timestamp = kernel_timestamp;
+        m->initrd_timestamp = initrd_timestamp;
 
         manager_set_default_rlimits(m, arg_default_rlimit);
 
-        if (dual_timestamp_is_set(&initrd_timestamp))
-                m->initrd_timestamp = initrd_timestamp;
-
-        if (arg_default_controllers)
-                manager_set_default_controllers(m, arg_default_controllers);
+        if (arg_default_environment)
+                manager_set_default_environment(m, arg_default_environment);
 
         manager_set_show_status(m, arg_show_status);
 
@@ -1542,10 +1646,7 @@ int main(int argc, char *argv[]) {
 
         /* This will close all file descriptors that were opened, but
          * not claimed by any unit. */
-        if (fds) {
-                fdset_free(fds);
-                fds = NULL;
-        }
+        fdset_free(fds);
 
         if (serialization) {
                 fclose(serialization);
@@ -1594,18 +1695,29 @@ int main(int argc, char *argv[]) {
                         manager_dump_units(m, stdout, "\t");
                 }
 
-                r = manager_add_job(m, JOB_START, target, JOB_REPLACE, false, &error, &default_unit_job);
-                if (r < 0) {
-                        log_error("Failed to start default target: %s", bus_error(&error, r));
+                r = manager_add_job(m, JOB_START, target, JOB_ISOLATE, false, &error, &default_unit_job);
+                if (r == -EPERM) {
+                        log_debug("Default target could not be isolated, starting instead: %s", bus_error(&error, r));
+                        dbus_error_free(&error);
+
+                        r = manager_add_job(m, JOB_START, target, JOB_REPLACE, false, &error, &default_unit_job);
+                        if (r < 0) {
+                                log_error("Failed to start default target: %s", bus_error(&error, r));
+                                dbus_error_free(&error);
+                                goto finish;
+                        }
+                } else if (r < 0) {
+                        log_error("Failed to isolate default target: %s", bus_error(&error, r));
                         dbus_error_free(&error);
                         goto finish;
                 }
+
                 m->default_unit_job_id = default_unit_job->id;
 
                 after_startup = now(CLOCK_MONOTONIC);
                 log_full(arg_action == ACTION_TEST ? LOG_INFO : LOG_DEBUG,
                          "Loaded units and determined initial transaction in %s.",
-                          format_timespan(timespan, sizeof(timespan), after_startup - before_startup));
+                         format_timespan(timespan, sizeof(timespan), after_startup - before_startup, 0));
 
                 if (arg_action == ACTION_TEST) {
                         printf("-> By jobs:\n");
@@ -1638,7 +1750,7 @@ int main(int argc, char *argv[]) {
 
                 case MANAGER_REEXECUTE:
 
-                        if (prepare_reexecute(m, &serialization, &fds, true) < 0)
+                        if (prepare_reexecute(m, &serialization, &fds, false) < 0)
                                 goto finish;
 
                         reexecute = true;
@@ -1652,7 +1764,7 @@ int main(int argc, char *argv[]) {
                         m->switch_root = m->switch_root_init = NULL;
 
                         if (!switch_root_init)
-                                if (prepare_reexecute(m, &serialization, &fds, false) < 0)
+                                if (prepare_reexecute(m, &serialization, &fds, true) < 0)
                                         goto finish;
 
                         reexecute = true;
@@ -1687,10 +1799,9 @@ finish:
                 manager_free(m);
 
         for (j = 0; j < RLIMIT_NLIMITS; j++)
-                free (arg_default_rlimit[j]);
+                free(arg_default_rlimit[j]);
 
         free(arg_default_unit);
-        strv_free(arg_default_controllers);
         free_join_controllers();
 
         dbus_shutdown();
@@ -1705,6 +1816,12 @@ finish:
                  * rebooted while we do that */
                 watchdog_close(true);
 
+                /* Reset the RLIMIT_NOFILE to the kernel default, so
+                 * that the new systemd can pass the kernel default to
+                 * its child processes */
+                if (saved_rlimit_nofile.rlim_cur > 0)
+                        setrlimit(RLIMIT_NOFILE, &saved_rlimit_nofile);
+
                 if (switch_root_dir) {
                         /* Kill all remaining processes from the
                          * initrd, but don't wait for them, so that we
@@ -1739,7 +1856,7 @@ finish:
                         args[i++] = SYSTEMD_BINARY_PATH;
                         if (switch_root_dir)
                                 args[i++] = "--switched-root";
-                        args[i++] = arg_running_as == MANAGER_SYSTEM ? "--system" : "--user";
+                        args[i++] = arg_running_as == SYSTEMD_SYSTEM ? "--system" : "--user";
                         args[i++] = "--deserialize";
                         args[i++] = sfd;
                         args[i++] = NULL;