X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=src%2Fcore%2Fmanager.c;h=ea8887a92ec90074670519d6105db2199070219e;hp=72ce2f25956fef35c1b1a482142d3c5355f1bc72;hb=a9244623f785f504f799407b0228dea9655e24cb;hpb=68b29a9fca915c83b9192790ec61189430cd5de6 diff --git a/src/core/manager.c b/src/core/manager.c index 72ce2f259..ea8887a92 100644 --- a/src/core/manager.c +++ b/src/core/manager.c @@ -22,9 +22,7 @@ #include #include #include -#include #include -#include #include #include #include @@ -36,14 +34,15 @@ #include #include #include +#include #ifdef HAVE_AUDIT #include #endif -#include "systemd/sd-daemon.h" -#include "systemd/sd-id128.h" -#include "systemd/sd-messages.h" +#include "sd-daemon.h" +#include "sd-id128.h" +#include "sd-messages.h" #include "manager.h" #include "transaction.h" @@ -54,87 +53,193 @@ #include "util.h" #include "mkdir.h" #include "ratelimit.h" -#include "cgroup.h" +#include "locale-setup.h" #include "mount-setup.h" #include "unit-name.h" -#include "dbus-unit.h" -#include "dbus-job.h" #include "missing.h" #include "path-lookup.h" #include "special.h" -#include "bus-errors.h" #include "exit-status.h" #include "virt.h" #include "watchdog.h" #include "cgroup-util.h" #include "path-util.h" #include "audit-fd.h" - -/* As soon as 16 units are in our GC queue, make sure to run a gc sweep */ -#define GC_QUEUE_ENTRIES_MAX 16 +#include "boot-timestamps.h" +#include "env-util.h" +#include "bus-errors.h" +#include "bus-error.h" +#include "bus-util.h" +#include "dbus.h" +#include "dbus-unit.h" +#include "dbus-job.h" +#include "dbus-manager.h" +#include "bus-kernel.h" /* As soon as 5s passed since a unit was added to our GC queue, make sure to run a gc sweep */ #define GC_QUEUE_USEC_MAX (10*USEC_PER_SEC) +/* Initial delay and the interval for printing status messages about running jobs */ +#define JOBS_IN_PROGRESS_WAIT_SEC 5 +#define JOBS_IN_PROGRESS_PERIOD_SEC 1 +#define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3 + /* Where clients shall send notification messages to */ #define NOTIFY_SOCKET "@/org/freedesktop/systemd1/notify" -static int manager_setup_notify(Manager *m) { - union { - struct sockaddr sa; - struct sockaddr_un un; - } sa; - struct epoll_event ev; - int one = 1; +#define TIME_T_MAX (time_t)((1UL << ((sizeof(time_t) << 3) - 1)) - 1) + +static int manager_dispatch_notify_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); +static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata); +static int manager_dispatch_run_queue(sd_event_source *source, void *userdata); +static int manager_watch_jobs_in_progress(Manager *m) { assert(m); - m->notify_watch.type = WATCH_NOTIFY; - m->notify_watch.fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0); - if (m->notify_watch.fd < 0) { - log_error("Failed to allocate notification socket: %m"); - return -errno; - } + if (m->jobs_in_progress_event_source) + return 0; - zero(sa); - sa.sa.sa_family = AF_UNIX; + return sd_event_add_monotonic(m->event, JOBS_IN_PROGRESS_WAIT_SEC, 0, manager_dispatch_jobs_in_progress, m, &m->jobs_in_progress_event_source); +} - if (getpid() != 1 || detect_container(NULL) > 0) - snprintf(sa.un.sun_path, sizeof(sa.un.sun_path), NOTIFY_SOCKET "/%llu", random_ull()); - else - strncpy(sa.un.sun_path, NOTIFY_SOCKET, sizeof(sa.un.sun_path)); +#define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED_ON)-1) + sizeof(ANSI_HIGHLIGHT_RED_ON)-1 + 2*(sizeof(ANSI_HIGHLIGHT_OFF)-1)) - sa.un.sun_path[0] = 0; +static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) { + char *p = buffer; - if (bind(m->notify_watch.fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + 1 + strlen(sa.un.sun_path+1)) < 0) { - log_error("bind() failed: %m"); - return -errno; + assert(buflen >= CYLON_BUFFER_EXTRA + width + 1); + assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */ + + if (pos > 1) { + if (pos > 2) + p = mempset(p, ' ', pos-2); + p = stpcpy(p, ANSI_RED_ON); + *p++ = '*'; } - if (setsockopt(m->notify_watch.fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one)) < 0) { - log_error("SO_PASSCRED failed: %m"); - return -errno; + if (pos > 0 && pos <= width) { + p = stpcpy(p, ANSI_HIGHLIGHT_RED_ON); + *p++ = '*'; } - zero(ev); - ev.events = EPOLLIN; - ev.data.ptr = &m->notify_watch; + p = stpcpy(p, ANSI_HIGHLIGHT_OFF); + + if (pos < width) { + p = stpcpy(p, ANSI_RED_ON); + *p++ = '*'; + if (pos < width-1) + p = mempset(p, ' ', width-1-pos); + strcpy(p, ANSI_HIGHLIGHT_OFF); + } +} + +static void manager_print_jobs_in_progress(Manager *m) { + _cleanup_free_ char *job_of_n = NULL; + Iterator i; + Job *j; + unsigned counter = 0, print_nr; + char cylon[6 + CYLON_BUFFER_EXTRA + 1]; + unsigned cylon_pos; + + assert(m); + + print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs; + + HASHMAP_FOREACH(j, m->jobs, i) + if (j->state == JOB_RUNNING && counter++ == print_nr) + break; + + /* m->n_running_jobs must be consistent with the contents of m->jobs, + * so the above loop must have succeeded in finding j. */ + assert(counter == print_nr + 1); + assert(j); + + cylon_pos = m->jobs_in_progress_iteration % 14; + if (cylon_pos >= 8) + cylon_pos = 14 - cylon_pos; + draw_cylon(cylon, sizeof(cylon), 6, cylon_pos); + + if (m->n_running_jobs > 1) + if (asprintf(&job_of_n, "(%u of %u) ", counter, m->n_running_jobs) < 0) + job_of_n = NULL; + + manager_status_printf(m, true, cylon, "%sA %s job is running for %s", + strempty(job_of_n), job_type_to_string(j->type), unit_description(j->unit)); + + m->jobs_in_progress_iteration++; +} + +static int manager_watch_idle_pipe(Manager *m) { + int r; + + assert(m); + + if (m->idle_pipe_event_source) + return 0; + + if (m->idle_pipe[2] < 0) + return 0; + + r = sd_event_add_io(m->event, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m, &m->idle_pipe_event_source); + if (r < 0) { + log_error("Failed to watch idle pipe: %s", strerror(-r)); + return r; + } + + return 0; +} - if (epoll_ctl(m->epoll_fd, EPOLL_CTL_ADD, m->notify_watch.fd, &ev) < 0) +static void manager_close_idle_pipe(Manager *m) { + assert(m); + + close_pipe(m->idle_pipe); + close_pipe(m->idle_pipe + 2); +} + +static int manager_setup_time_change(Manager *m) { + int r; + + /* We only care for the cancellation event, hence we set the + * timeout to the latest possible value. */ + struct itimerspec its = { + .it_value.tv_sec = TIME_T_MAX, + }; + + assert(m); + assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX)); + + /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever + * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */ + + m->time_change_fd = timerfd_create(CLOCK_REALTIME, TFD_NONBLOCK|TFD_CLOEXEC); + if (m->time_change_fd < 0) { + log_error("Failed to create timerfd: %m"); return -errno; + } - sa.un.sun_path[0] = '@'; - m->notify_socket = strdup(sa.un.sun_path); - if (!m->notify_socket) - return -ENOMEM; + if (timerfd_settime(m->time_change_fd, TFD_TIMER_ABSTIME|TFD_TIMER_CANCEL_ON_SET, &its, NULL) < 0) { + log_debug("Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m"); + close_nointr_nofail(m->time_change_fd); + m->time_change_fd = -1; + return 0; + } - log_debug("Using notification socket %s", m->notify_socket); + r = sd_event_add_io(m->event, m->time_change_fd, EPOLLIN, manager_dispatch_time_change_fd, m, &m->time_change_event_source); + if (r < 0) { + log_error("Failed to create time change event source: %s", strerror(-r)); + return r; + } + + log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd."); return 0; } static int enable_special_signals(Manager *m) { - int fd; + _cleanup_close_ int fd = -1; assert(m); @@ -152,25 +257,23 @@ static int enable_special_signals(Manager *m) { } else { /* Enable that we get SIGWINCH on kbrequest */ if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0) - log_warning("Failed to enable kbrequest handling: %s", strerror(errno)); - - close_nointr_nofail(fd); + log_warning("Failed to enable kbrequest handling: %m"); } return 0; } static int manager_setup_signals(Manager *m) { + struct sigaction sa = { + .sa_handler = SIG_DFL, + .sa_flags = SA_NOCLDSTOP|SA_RESTART, + }; sigset_t mask; - struct epoll_event ev; - struct sigaction sa; + int r; assert(m); /* We are not interested in SIGSTOP and friends. */ - zero(sa); - sa.sa_handler = SIG_DFL; - sa.sa_flags = SA_NOCLDSTOP|SA_RESTART; assert_se(sigaction(SIGCHLD, &sa, NULL) == 0); assert_se(sigemptyset(&mask) == 0); @@ -207,16 +310,21 @@ static int manager_setup_signals(Manager *m) { -1); assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); - m->signal_watch.type = WATCH_SIGNAL; - if ((m->signal_watch.fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC)) < 0) + m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC); + if (m->signal_fd < 0) return -errno; - zero(ev); - ev.events = EPOLLIN; - ev.data.ptr = &m->signal_watch; + r = sd_event_add_io(m->event, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m, &m->signal_event_source); + if (r < 0) + return r; - if (epoll_ctl(m->epoll_fd, EPOLL_CTL_ADD, m->signal_watch.fd, &ev) < 0) - return -errno; + /* Process signals a bit earlier than the rest of things, but + * later that 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, -5); + if (r < 0) + return r; if (m->running_as == SYSTEMD_SYSTEM) return enable_special_signals(m); @@ -224,24 +332,38 @@ static int manager_setup_signals(Manager *m) { return 0; } -static void manager_strip_environment(Manager *m) { +static int manager_default_environment(Manager *m) { assert(m); - /* Remove variables from the inherited set that are part of - * the container interface: - * http://www.freedesktop.org/wiki/Software/systemd/ContainerInterface */ - strv_remove_prefix(m->environment, "container="); - strv_remove_prefix(m->environment, "container_"); + if (m->running_as == SYSTEMD_SYSTEM) { + /* The system manager always starts with a clean + * environment for its children. It does not import + * the kernel or the parents exported variables. + * + * The initial passed environ is untouched to keep + * /proc/self/environ valid; it is used for tagging + * the init process inside containers. */ + m->environment = strv_new("PATH=" DEFAULT_PATH, + NULL); + + /* Import locale variables LC_*= from configuration */ + locale_setup(&m->environment); + } else + /* The user manager passes its own environment + * along to its children. */ + m->environment = strv_copy(environ); + + if (!m->environment) + return -ENOMEM; - /* Remove variables from the inherited set that are part of - * the initrd interface: - * http://www.freedesktop.org/wiki/Software/systemd/InitrdInterface */ - strv_remove_prefix(m->environment, "RD_"); + strv_sort(m->environment); + + return 0; } int manager_new(SystemdRunningAs running_as, Manager **_m) { Manager *m; - int r = -ENOMEM; + int r; assert(_m); assert(running_as >= 0); @@ -251,60 +373,81 @@ int manager_new(SystemdRunningAs running_as, Manager **_m) { if (!m) return -ENOMEM; - dual_timestamp_get(&m->userspace_timestamp); +#ifdef ENABLE_EFI + if (detect_container(NULL) <= 0) + boot_timestamps(&m->userspace_timestamp, &m->firmware_timestamp, &m->loader_timestamp); +#endif m->running_as = running_as; - m->name_data_slot = m->conn_data_slot = m->subscribed_data_slot = -1; m->exit_code = _MANAGER_EXIT_CODE_INVALID; - m->pin_cgroupfs_fd = -1; - m->idle_pipe[0] = m->idle_pipe[1] = -1; - m->signal_watch.fd = m->mount_watch.fd = m->udev_watch.fd = m->epoll_fd = m->dev_autofs_fd = m->swap_watch.fd = -1; + 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 = -1; m->current_job_id = 1; /* start as id #1, so that we can leave #0 around as "null-like" value */ - m->environment = strv_copy(environ); - if (!m->environment) + r = manager_default_environment(m); + if (r < 0) goto fail; - manager_strip_environment(m); + r = hashmap_ensure_allocated(&m->units, string_hash_func, string_compare_func); + if (r < 0) + goto fail; - if (running_as == SYSTEMD_SYSTEM) { - m->default_controllers = strv_new("cpu", NULL); - if (!m->default_controllers) - goto fail; - } + r = hashmap_ensure_allocated(&m->jobs, trivial_hash_func, trivial_compare_func); + if (r < 0) + goto fail; - if (!(m->units = hashmap_new(string_hash_func, string_compare_func))) + r = hashmap_ensure_allocated(&m->cgroup_unit, string_hash_func, string_compare_func); + if (r < 0) goto fail; - if (!(m->jobs = hashmap_new(trivial_hash_func, trivial_compare_func))) + r = hashmap_ensure_allocated(&m->watch_pids, trivial_hash_func, trivial_compare_func); + if (r < 0) goto fail; - if (!(m->watch_pids = hashmap_new(trivial_hash_func, trivial_compare_func))) + r = hashmap_ensure_allocated(&m->watch_bus, string_hash_func, string_compare_func); + if (r < 0) goto fail; - if (!(m->cgroup_bondings = hashmap_new(string_hash_func, string_compare_func))) + r = sd_event_default(&m->event); + if (r < 0) goto fail; - if (!(m->watch_bus = hashmap_new(string_hash_func, string_compare_func))) + r = sd_event_add_defer(m->event, manager_dispatch_run_queue, m, &m->run_queue_event_source); + if (r < 0) goto fail; - if ((m->epoll_fd = epoll_create1(EPOLL_CLOEXEC)) < 0) + r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE); + if (r < 0) goto fail; - if ((r = manager_setup_signals(m)) < 0) + r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF); + if (r < 0) goto fail; - if ((r = manager_setup_cgroup(m)) < 0) + r = manager_setup_signals(m); + if (r < 0) goto fail; - if ((r = manager_setup_notify(m)) < 0) + r = manager_setup_cgroup(m); + if (r < 0) goto fail; - /* Try to connect to the busses, if possible. */ - if ((r = bus_init(m, running_as != SYSTEMD_SYSTEM)) < 0) + r = manager_setup_time_change(m); + if (r < 0) goto fail; + m->udev = udev_new(); + if (!m->udev) { + r = -ENOMEM; + goto fail; + } + + /* Note that we set up neither kdbus, nor the notify fd + * here. We do that after deserialization, since they might + * have gotten serialized across the reexec. */ + m->taint_usr = dir_is_empty("/usr") > 0; *_m = m; @@ -315,6 +458,122 @@ fail: return r; } +static int manager_setup_notify(Manager *m) { + union { + struct sockaddr sa; + struct sockaddr_un un; + } sa = { + .sa.sa_family = AF_UNIX, + }; + int one = 1, r; + + if (m->notify_fd < 0) { + _cleanup_close_ int fd = -1; + + /* First free all secondary fields */ + free(m->notify_socket); + m->notify_socket = NULL; + m->notify_event_source = sd_event_source_unref(m->notify_event_source); + + fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0); + if (fd < 0) { + log_error("Failed to allocate notification socket: %m"); + return -errno; + } + + if (getpid() != 1 || detect_container(NULL) > 0) + snprintf(sa.un.sun_path, sizeof(sa.un.sun_path), NOTIFY_SOCKET "/%" PRIx64, random_u64()); + else + strncpy(sa.un.sun_path, NOTIFY_SOCKET, sizeof(sa.un.sun_path)); + sa.un.sun_path[0] = 0; + + r = bind(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + 1 + strlen(sa.un.sun_path+1)); + if (r < 0) { + log_error("bind() failed: %m"); + return -errno; + } + + r = setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one)); + if (r < 0) { + log_error("SO_PASSCRED failed: %m"); + return -errno; + } + + sa.un.sun_path[0] = '@'; + m->notify_socket = strdup(sa.un.sun_path); + if (!m->notify_socket) + return log_oom(); + + m->notify_fd = fd; + fd = -1; + + log_debug("Using notification socket %s", m->notify_socket); + } + + if (!m->notify_event_source) { + r = sd_event_add_io(m->event, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m, &m->notify_event_source); + if (r < 0) { + log_error("Failed to allocate notify event source: %s", strerror(-r)); + return -errno; + } + + /* Process signals 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, -7); + if (r < 0) { + log_error("Failed to set priority of notify event source: %s", strerror(-r)); + return r; + } + } + + return 0; +} + +static int manager_setup_kdbus(Manager *m) { + _cleanup_free_ char *p = NULL; + + assert(m); + +#ifdef ENABLE_KDBUS + if (m->kdbus_fd >= 0) + return 0; + + /* If there's already a bus address set, don't set up kdbus */ + if (m->running_as == SYSTEMD_USER && getenv("DBUS_SESSION_BUS_ADDRESS")) + return 0; + + m->kdbus_fd = bus_kernel_create_bus(m->running_as == SYSTEMD_SYSTEM ? "system" : "user", m->running_as == SYSTEMD_SYSTEM, &p); + if (m->kdbus_fd < 0) { + log_debug("Failed to set up kdbus: %s", strerror(-m->kdbus_fd)); + return m->kdbus_fd; + } + + log_debug("Successfully set up kdbus on %s", p); + + /* Create the namespace directory here, so that the contents + * of that directory is not visible to non-root users. This is + * necessary to ensure that users cannot get access to busses + * of virtualized users when no UID namespacing is used. */ + mkdir_p_label("/dev/kdbus/ns", 0700); +#endif + + return 0; +} + +static int manager_connect_bus(Manager *m, bool reexecuting) { + bool try_bus_connect; + + assert(m); + + try_bus_connect = + m->kdbus_fd >= 0 || + reexecuting || + (m->running_as == SYSTEMD_USER && getenv("DBUS_SESSION_BUS_ADDRESS")); + + /* Try to connect to the busses, if possible. */ + return bus_init(m, try_bus_connect); +} + static unsigned manager_dispatch_cleanup_queue(Manager *m) { Unit *u; unsigned n = 0; @@ -398,12 +657,7 @@ static unsigned manager_dispatch_gc_queue(Manager *m) { assert(m); - if ((m->n_in_gc_queue < GC_QUEUE_ENTRIES_MAX) && - (m->gc_queue_timestamp <= 0 || - (m->gc_queue_timestamp + GC_QUEUE_USEC_MAX) > now(CLOCK_MONOTONIC))) - return 0; - - log_debug("Running GC..."); + /* log_debug("Running GC..."); */ m->gc_marker += _GC_OFFSET_MAX; if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX) @@ -416,21 +670,20 @@ static unsigned manager_dispatch_gc_queue(Manager *m) { unit_gc_sweep(u, gc_marker); - LIST_REMOVE(Unit, gc_queue, m->gc_queue, u); + LIST_REMOVE(gc_queue, m->gc_queue, u); u->in_gc_queue = false; n++; if (u->gc_marker == gc_marker + GC_OFFSET_BAD || u->gc_marker == gc_marker + GC_OFFSET_UNSURE) { - log_debug("Collecting %s", u->id); + log_debug_unit(u->id, "Collecting %s", u->id); u->gc_marker = gc_marker + GC_OFFSET_BAD; unit_add_to_cleanup_queue(u); } } m->n_in_gc_queue = 0; - m->gc_queue_timestamp = 0; return n; } @@ -454,6 +707,9 @@ static void manager_clear_jobs_and_units(Manager *m) { assert(hashmap_isempty(m->jobs)); assert(hashmap_isempty(m->units)); + + m->n_on_console = 0; + m->n_running_jobs = 0; } void manager_free(Manager *m) { @@ -481,31 +737,44 @@ void manager_free(Manager *m) { hashmap_free(m->watch_pids); hashmap_free(m->watch_bus); - if (m->epoll_fd >= 0) - close_nointr_nofail(m->epoll_fd); - if (m->signal_watch.fd >= 0) - close_nointr_nofail(m->signal_watch.fd); - if (m->notify_watch.fd >= 0) - close_nointr_nofail(m->notify_watch.fd); + sd_event_source_unref(m->signal_event_source); + sd_event_source_unref(m->notify_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->idle_pipe_event_source); + sd_event_source_unref(m->run_queue_event_source); + + if (m->signal_fd >= 0) + close_nointr_nofail(m->signal_fd); + if (m->notify_fd >= 0) + close_nointr_nofail(m->notify_fd); + if (m->time_change_fd >= 0) + close_nointr_nofail(m->time_change_fd); + if (m->kdbus_fd >= 0) + close_nointr_nofail(m->kdbus_fd); + + manager_close_idle_pipe(m); + + udev_unref(m->udev); + sd_event_unref(m->event); free(m->notify_socket); lookup_paths_free(&m->lookup_paths); strv_free(m->environment); - strv_free(m->default_controllers); - - hashmap_free(m->cgroup_bondings); + hashmap_free(m->cgroup_unit); set_free_free(m->unit_path_cache); - close_pipe(m->idle_pipe); - free(m->switch_root); free(m->switch_root_init); for (i = 0; i < RLIMIT_NLIMITS; i++) free(m->rlimit[i]); + assert(hashmap_isempty(m->units_requiring_mounts_for)); + hashmap_free(m->units_requiring_mounts_for); + free(m); } @@ -518,15 +787,17 @@ int manager_enumerate(Manager *m) { /* Let's ask every type to load all units from disk/kernel * that it might know */ for (c = 0; c < _UNIT_TYPE_MAX; c++) - if (unit_vtable[c]->enumerate) - if ((q = unit_vtable[c]->enumerate(m)) < 0) + if (unit_vtable[c]->enumerate) { + q = unit_vtable[c]->enumerate(m); + if (q < 0) r = q; + } manager_dispatch_load_queue(m); return r; } -int manager_coldplug(Manager *m) { +static int manager_coldplug(Manager *m) { int r = 0, q; Iterator i; Unit *u; @@ -550,14 +821,15 @@ int manager_coldplug(Manager *m) { static void manager_build_unit_path_cache(Manager *m) { char **i; - DIR *d = NULL; + _cleanup_free_ DIR *d = NULL; int r; assert(m); set_free_free(m->unit_path_cache); - if (!(m->unit_path_cache = set_new(string_hash_func, string_compare_func))) { + m->unit_path_cache = set_new(string_hash_func, string_compare_func); + if (!m->unit_path_cache) { log_error("Failed to allocate unit path cache."); return; } @@ -568,8 +840,10 @@ static void manager_build_unit_path_cache(Manager *m) { STRV_FOREACH(i, m->lookup_paths.unit_path) { struct dirent *de; - if (!(d = opendir(*i))) { - log_error("Failed to open directory: %m"); + d = opendir(*i); + if (!d) { + if (errno != ENOENT) + log_error("Failed to open directory %s: %m", *i); continue; } @@ -585,10 +859,9 @@ static void manager_build_unit_path_cache(Manager *m) { goto fail; } - if ((r = set_put(m->unit_path_cache, p)) < 0) { - free(p); + r = set_consume(m->unit_path_cache, p); + if (r < 0) goto fail; - } } closedir(d); @@ -602,9 +875,29 @@ fail: set_free_free(m->unit_path_cache); m->unit_path_cache = NULL; +} - if (d) - closedir(d); + +static int manager_distribute_fds(Manager *m, FDSet *fds) { + Unit *u; + Iterator i; + int r; + + assert(m); + + HASHMAP_FOREACH(u, m->units, i) { + + if (fdset_size(fds) <= 0) + break; + + if (UNIT_VTABLE(u)->distribute_fds) { + r = UNIT_VTABLE(u)->distribute_fds(u, fds); + if (r < 0) + return r; + } + } + + return 0; } int manager_startup(Manager *m, FILE *serialization, FDSet *fds) { @@ -612,7 +905,9 @@ int manager_startup(Manager *m, FILE *serialization, FDSet *fds) { assert(m); + dual_timestamp_get(&m->generators_start_timestamp); manager_run_generators(m); + dual_timestamp_get(&m->generators_finish_timestamp); r = lookup_paths_init( &m->lookup_paths, m->running_as, true, @@ -631,7 +926,9 @@ int manager_startup(Manager *m, FILE *serialization, FDSet *fds) { m->n_reloading ++; /* First, enumerate what we can from all config files */ + dual_timestamp_get(&m->units_load_start_timestamp); r = manager_enumerate(m); + dual_timestamp_get(&m->units_load_finish_timestamp); /* Second, deserialize if there is something to deserialize */ if (serialization) { @@ -640,6 +937,25 @@ int manager_startup(Manager *m, FILE *serialization, FDSet *fds) { r = q; } + /* Any fds left? Find some unit which wants them. This is + * useful to allow container managers to pass some file + * descriptors to us pre-initialized. This enables + * socket-based activation of entire containers. */ + if (fdset_size(fds) > 0) { + q = manager_distribute_fds(m, fds); + if (q < 0) + r = q; + } + + /* We might have deserialized the notify fd, but if we didn't + * then let's create the bus now */ + manager_setup_notify(m); + + /* We might have deserialized the kdbus control fd, but if we + * didn't, then let's create the bus now. */ + manager_setup_kdbus(m); + manager_connect_bus(m, !!serialization); + /* Third, fire things up! */ q = manager_coldplug(m); if (q < 0) @@ -648,12 +964,17 @@ int manager_startup(Manager *m, FILE *serialization, FDSet *fds) { if (serialization) { assert(m->n_reloading > 0); m->n_reloading --; + + /* Let's wait for the UnitNew/JobNew messages being + * sent, before we notify that the reload is + * finished */ + m->send_reloading_done = true; } return r; } -int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, bool override, DBusError *e, Job **_ret) { +int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, bool override, sd_bus_error *e, Job **_ret) { int r; Transaction *tr; @@ -663,20 +984,22 @@ int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, bool ove assert(mode < _JOB_MODE_MAX); if (mode == JOB_ISOLATE && type != JOB_START) { - dbus_set_error(e, BUS_ERROR_INVALID_JOB_MODE, "Isolate is only valid for start."); + sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start."); return -EINVAL; } if (mode == JOB_ISOLATE && !unit->allow_isolate) { - dbus_set_error(e, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated."); + sd_bus_error_setf(e, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated."); return -EPERM; } - log_debug("Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode)); + log_debug_unit(unit->id, + "Trying to enqueue job %s/%s/%s", unit->id, + job_type_to_string(type), job_mode_to_string(mode)); job_type_collapse(&type, unit); - tr = transaction_new(); + tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY); if (!tr) return -ENOMEM; @@ -696,7 +1019,9 @@ int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, bool ove if (r < 0) goto tr_abort; - log_debug("Enqueued job %s/%s as %u", unit->id, job_type_to_string(type), (unsigned) tr->anchor_job->id); + log_debug_unit(unit->id, + "Enqueued job %s/%s as %u", unit->id, + job_type_to_string(type), (unsigned) tr->anchor_job->id); if (_ret) *_ret = tr->anchor_job; @@ -710,7 +1035,7 @@ tr_abort: return r; } -int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, bool override, DBusError *e, Job **_ret) { +int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, bool override, sd_bus_error *e, Job **_ret) { Unit *unit; int r; @@ -765,7 +1090,13 @@ unsigned manager_dispatch_load_queue(Manager *m) { return n; } -int manager_load_unit_prepare(Manager *m, const char *name, const char *path, DBusError *e, Unit **_ret) { +int manager_load_unit_prepare( + Manager *m, + const char *name, + const char *path, + sd_bus_error *e, + Unit **_ret) { + Unit *ret; UnitType t; int r; @@ -776,20 +1107,16 @@ int manager_load_unit_prepare(Manager *m, const char *name, const char *path, DB /* This will prepare the unit for loading, but not actually * load anything from disk. */ - if (path && !is_path(path)) { - dbus_set_error(e, BUS_ERROR_INVALID_PATH, "Path %s is not absolute.", path); - return -EINVAL; - } + if (path && !is_path(path)) + return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path); if (!name) - name = path_get_file_name(path); + name = basename(path); t = unit_name_to_type(name); - if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, false)) { - dbus_set_error(e, BUS_ERROR_INVALID_NAME, "Unit name %s is not valid.", name); - return -EINVAL; - } + if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, TEMPLATE_INVALID)) + return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name); ret = manager_get_unit(m, name); if (ret) { @@ -809,7 +1136,8 @@ int manager_load_unit_prepare(Manager *m, const char *name, const char *path, DB } } - if ((r = unit_add_name(ret, name)) < 0) { + r = unit_add_name(ret, name); + if (r < 0) { unit_free(ret); return r; } @@ -824,7 +1152,13 @@ int manager_load_unit_prepare(Manager *m, const char *name, const char *path, DB return 0; } -int manager_load_unit(Manager *m, const char *name, const char *path, DBusError *e, Unit **_ret) { +int manager_load_unit( + Manager *m, + const char *name, + const char *path, + sd_bus_error *e, + Unit **_ret) { + int r; assert(m); @@ -878,28 +1212,30 @@ void manager_clear_jobs(Manager *m) { job_finish_and_invalidate(j, JOB_CANCELED, false); } -unsigned manager_dispatch_run_queue(Manager *m) { +static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) { + Manager *m = userdata; Job *j; - unsigned n = 0; - if (m->dispatching_run_queue) - return 0; - - m->dispatching_run_queue = true; + assert(source); + assert(m); while ((j = m->run_queue)) { assert(j->installed); assert(j->in_run_queue); job_run_and_invalidate(j); - n++; } - m->dispatching_run_queue = false; - return n; + if (m->n_running_jobs > 0) + manager_watch_jobs_in_progress(m); + + if (m->n_on_console > 0) + manager_watch_idle_pipe(m); + + return 1; } -unsigned manager_dispatch_dbus_queue(Manager *m) { +static unsigned manager_dispatch_dbus_queue(Manager *m) { Job *j; Unit *u; unsigned n = 0; @@ -926,39 +1262,56 @@ unsigned manager_dispatch_dbus_queue(Manager *m) { } m->dispatching_dbus_queue = false; + + if (m->send_reloading_done) { + m->send_reloading_done = false; + + bus_manager_send_reloading(m, false); + } + + if (m->queued_message) + bus_send_queued_message(m); + return n; } -static int manager_process_notify_fd(Manager *m) { +static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { + Manager *m = userdata; ssize_t n; assert(m); + assert(m->notify_fd == fd); + + if (revents != EPOLLIN) { + log_warning("Got unexpected poll event for notify fd."); + return 0; + } for (;;) { char buf[4096]; - struct msghdr msghdr; - struct iovec iovec; - struct ucred *ucred; + struct iovec iovec = { + .iov_base = buf, + .iov_len = sizeof(buf)-1, + }; + union { struct cmsghdr cmsghdr; uint8_t buf[CMSG_SPACE(sizeof(struct ucred))]; - } control; + } control = {}; + + struct msghdr msghdr = { + .msg_iov = &iovec, + .msg_iovlen = 1, + .msg_control = &control, + .msg_controllen = sizeof(control), + }; + struct ucred *ucred; Unit *u; - char **tags; + _cleanup_strv_free_ char **tags = NULL; - zero(iovec); - iovec.iov_base = buf; - iovec.iov_len = sizeof(buf)-1; - - zero(control); - zero(msghdr); - msghdr.msg_iov = &iovec; - msghdr.msg_iovlen = 1; - msghdr.msg_control = &control; - msghdr.msg_controllen = sizeof(control); - - if ((n = recvmsg(m->notify_watch.fd, &msghdr, MSG_DONTWAIT)) <= 0) { - if (n >= 0) + n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT); + if (n <= 0) { + if (n == 0) return -EIO; if (errno == EAGAIN || errno == EINTR) @@ -977,23 +1330,25 @@ static int manager_process_notify_fd(Manager *m) { ucred = (struct ucred*) CMSG_DATA(&control.cmsghdr); - if (!(u = hashmap_get(m->watch_pids, LONG_TO_PTR(ucred->pid)))) - if (!(u = cgroup_unit_by_pid(m, ucred->pid))) { + u = hashmap_get(m->watch_pids, LONG_TO_PTR(ucred->pid)); + if (!u) { + u = manager_get_unit_by_pid(m, ucred->pid); + if (!u) { log_warning("Cannot find unit for notify message of PID %lu.", (unsigned long) ucred->pid); continue; } + } assert((size_t) n < sizeof(buf)); buf[n] = 0; - if (!(tags = strv_split(buf, "\n\r"))) - return -ENOMEM; + tags = strv_split(buf, "\n\r"); + if (!tags) + return log_oom(); - log_debug("Got notification message for unit %s", u->id); + log_debug_unit(u->id, "Got notification message for unit %s", u->id); if (UNIT_VTABLE(u)->notify_message) UNIT_VTABLE(u)->notify_message(u, ucred->pid, tags); - - strv_free(tags); } return 0; @@ -1003,11 +1358,8 @@ static int manager_dispatch_sigchld(Manager *m) { assert(m); for (;;) { - siginfo_t si; + siginfo_t si = {}; Unit *u; - int r; - - zero(si); /* First we call waitd() for a PID and do not reap the * zombie. That way we can still access /proc/$PID for @@ -1027,23 +1379,16 @@ static int manager_dispatch_sigchld(Manager *m) { break; if (si.si_code == CLD_EXITED || si.si_code == CLD_KILLED || si.si_code == CLD_DUMPED) { - char *name = NULL; + _cleanup_free_ char *name = NULL; get_process_comm(si.si_pid, &name); log_debug("Got SIGCHLD for process %lu (%s)", (unsigned long) si.si_pid, strna(name)); - free(name); } - /* Let's flush any message the dying child might still - * have queued for us. This ensures that the process - * still exists in /proc so that we can figure out - * which cgroup and hence unit it belongs to. */ - if ((r = manager_process_notify_fd(m)) < 0) - return r; - /* And now figure out the unit this belongs to */ - if (!(u = hashmap_get(m->watch_pids, LONG_TO_PTR(si.si_pid)))) - u = cgroup_unit_by_pid(m, si.si_pid); + u = hashmap_get(m->watch_pids, LONG_TO_PTR(si.si_pid)); + if (!u) + u = manager_get_unit_by_pid(m, si.si_pid); /* And now, we actually reap the zombie. */ if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) { @@ -1067,7 +1412,8 @@ static int manager_dispatch_sigchld(Manager *m) { if (!u) continue; - log_debug("Child %lu belongs to %s", (long unsigned) si.si_pid, u->id); + log_debug_unit(u->id, + "Child %lu belongs to %s", (long unsigned) si.si_pid, u->id); hashmap_remove(m->watch_pids, LONG_TO_PTR(si.si_pid)); UNIT_VTABLE(u)->sigchld_event(u, si.si_pid, si.si_code, si.si_status); @@ -1077,30 +1423,35 @@ static int manager_dispatch_sigchld(Manager *m) { } static int manager_start_target(Manager *m, const char *name, JobMode mode) { + _cleanup_bus_error_free_ sd_bus_error error = SD_BUS_ERROR_NULL; int r; - DBusError error; - - dbus_error_init(&error); - - log_debug("Activating special unit %s", name); - if ((r = manager_add_job_by_name(m, JOB_START, name, mode, true, &error, NULL)) < 0) - log_error("Failed to enqueue %s job: %s", name, bus_error(&error, r)); + log_debug_unit(name, "Activating special unit %s", name); - dbus_error_free(&error); + r = manager_add_job_by_name(m, JOB_START, name, mode, true, &error, NULL); + if (r < 0) + log_error_unit(name, "Failed to enqueue %s job: %s", name, bus_error_message(&error, r)); return r; } -static int manager_process_signal_fd(Manager *m) { +static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { + Manager *m = userdata; ssize_t n; struct signalfd_siginfo sfsi; bool sigchld = false; assert(m); + assert(m->signal_fd == fd); + + if (revents != EPOLLIN) { + log_warning("Got unexpected events from signal file descriptor."); + return 0; + } for (;;) { - if ((n = read(m->signal_watch.fd, &sfsi, sizeof(sfsi))) != sizeof(sfsi)) { + n = read(m->signal_fd, &sfsi, sizeof(sfsi)); + if (n != sizeof(sfsi)) { if (n >= 0) return -EIO; @@ -1112,16 +1463,22 @@ static int manager_process_signal_fd(Manager *m) { } if (sfsi.ssi_pid > 0) { - char *p = NULL; + _cleanup_free_ char *p = NULL; get_process_comm(sfsi.ssi_pid, &p); - log_debug("Received SIG%s from PID %lu (%s).", - signal_to_string(sfsi.ssi_signo), - (unsigned long) sfsi.ssi_pid, strna(p)); - free(p); + log_full(sfsi.ssi_signo == SIGCHLD || + (sfsi.ssi_signo == SIGTERM && m->running_as == SYSTEMD_USER) + ? LOG_DEBUG : LOG_INFO, + "Received SIG%s from PID %lu (%s).", + signal_to_string(sfsi.ssi_signo), + (unsigned long) sfsi.ssi_pid, strna(p)); } else - log_debug("Received SIG%s.", signal_to_string(sfsi.ssi_signo)); + log_full(sfsi.ssi_signo == SIGCHLD || + (sfsi.ssi_signo == SIGTERM && m->running_as == SYSTEMD_USER) + ? LOG_DEBUG : LOG_INFO, + "Received SIG%s.", + signal_to_string(sfsi.ssi_signo)); switch (sfsi.ssi_signo) { @@ -1141,7 +1498,7 @@ static int manager_process_signal_fd(Manager *m) { case SIGINT: if (m->running_as == SYSTEMD_SYSTEM) { - manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE); + manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY); break; } @@ -1186,11 +1543,12 @@ static int manager_process_signal_fd(Manager *m) { } case SIGUSR2: { - FILE *f; - char *dump = NULL; + _cleanup_free_ char *dump = NULL; + _cleanup_fclose_ FILE *f = NULL; size_t size; - if (!(f = open_memstream(&dump, &size))) { + f = open_memstream(&dump, &size); + if (!f) { log_warning("Failed to allocate memory stream."); break; } @@ -1199,16 +1557,11 @@ static int manager_process_signal_fd(Manager *m) { manager_dump_jobs(m, f, "\t"); if (ferror(f)) { - fclose(f); - free(dump); log_warning("Failed to write status stream"); break; } - fclose(f); log_dump(LOG_INFO, dump); - free(dump); - break; } @@ -1310,103 +1663,59 @@ static int manager_process_signal_fd(Manager *m) { } if (sigchld) - return manager_dispatch_sigchld(m); + manager_dispatch_sigchld(m); return 0; } -static int process_event(Manager *m, struct epoll_event *ev) { - int r; - Watch *w; +static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { + Manager *m = userdata; + Iterator i; + Unit *u; assert(m); - assert(ev); - - assert_se(w = ev->data.ptr); - - if (w->type == WATCH_INVALID) - return 0; - - switch (w->type) { - - case WATCH_SIGNAL: - - /* An incoming signal? */ - if (ev->events != EPOLLIN) - return -EINVAL; - - if ((r = manager_process_signal_fd(m)) < 0) - return r; - - break; - - case WATCH_NOTIFY: - - /* An incoming daemon notification event? */ - if (ev->events != EPOLLIN) - return -EINVAL; - - if ((r = manager_process_notify_fd(m)) < 0) - return r; - - break; + assert(m->time_change_fd == fd); - case WATCH_FD: + log_struct(LOG_INFO, + MESSAGE_ID(SD_MESSAGE_TIME_CHANGE), + "MESSAGE=Time has been changed", + NULL); - /* Some fd event, to be dispatched to the units */ - UNIT_VTABLE(w->data.unit)->fd_event(w->data.unit, w->fd, ev->events, w); - break; + /* Restart the watch */ + m->time_change_event_source = sd_event_source_unref(m->time_change_event_source); - case WATCH_UNIT_TIMER: - case WATCH_JOB_TIMER: { - uint64_t v; - ssize_t k; + close_nointr_nofail(m->time_change_fd); + m->time_change_fd = -1; - /* Some timer event, to be dispatched to the units */ - k = read(w->fd, &v, sizeof(v)); - if (k != sizeof(v)) { + manager_setup_time_change(m); - if (k < 0 && (errno == EINTR || errno == EAGAIN)) - break; + HASHMAP_FOREACH(u, m->units, i) + if (UNIT_VTABLE(u)->time_change) + UNIT_VTABLE(u)->time_change(u); - return k < 0 ? -errno : -EIO; - } + return 0; +} - if (w->type == WATCH_UNIT_TIMER) - UNIT_VTABLE(w->data.unit)->timer_event(w->data.unit, v, w); - else - job_timer_event(w->data.job, v, w); - break; - } +static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { + Manager *m = userdata; - case WATCH_MOUNT: - /* Some mount table change, intended for the mount subsystem */ - mount_fd_event(m, ev->events); - break; + assert(m); + assert(m->idle_pipe[2] == fd); - case WATCH_SWAP: - /* Some swap table change, intended for the swap subsystem */ - swap_fd_event(m, ev->events); - break; + m->no_console_output = m->n_on_console > 0; - case WATCH_UDEV: - /* Some notification from udev, intended for the device subsystem */ - device_fd_event(m, ev->events); - break; + m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source); + manager_close_idle_pipe(m); - case WATCH_DBUS_WATCH: - bus_watch_event(m, w, ev->events); - break; + return 0; +} - case WATCH_DBUS_TIMEOUT: - bus_timeout_event(m, w, ev->events); - break; +static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) { + Manager *m = userdata; - default: - log_error("event type=%i", w->type); - assert_not_reached("Unknown epoll event type."); - } + assert(m); + manager_print_jobs_in_progress(m); return 0; } @@ -1425,15 +1734,13 @@ int manager_loop(Manager *m) { manager_check_finished(m); /* There might still be some zombies hanging around from - * before we were exec()'ed. Leat's reap them */ + * before we were exec()'ed. Let's reap them. */ r = manager_dispatch_sigchld(m); if (r < 0) return r; while (m->exit_code == MANAGER_RUNNING) { - struct epoll_event event; - int n; - int wait_msec = -1; + usec_t wait_usec; if (m->runtime_watchdog > 0 && m->running_as == SYSTEMD_SYSTEM) watchdog_ping(); @@ -1448,54 +1755,38 @@ int manager_loop(Manager *m) { if (manager_dispatch_load_queue(m) > 0) continue; - if (manager_dispatch_run_queue(m) > 0) - continue; - - if (bus_dispatch(m) > 0) + if (manager_dispatch_gc_queue(m) > 0) continue; if (manager_dispatch_cleanup_queue(m) > 0) continue; - if (manager_dispatch_gc_queue(m) > 0) + if (manager_dispatch_cgroup_queue(m) > 0) continue; if (manager_dispatch_dbus_queue(m) > 0) continue; - if (swap_dispatch_reload(m) > 0) - continue; - /* Sleep for half the watchdog time */ if (m->runtime_watchdog > 0 && m->running_as == SYSTEMD_SYSTEM) { - wait_msec = (int) (m->runtime_watchdog / 2 / USEC_PER_MSEC); - if (wait_msec <= 0) - wait_msec = 1; + wait_usec = m->runtime_watchdog / 2; + if (wait_usec <= 0) + wait_usec = 1; } else - wait_msec = -1; + wait_usec = (usec_t) -1; - n = epoll_wait(m->epoll_fd, &event, 1, wait_msec); - if (n < 0) { - - if (errno == EINTR) - continue; - - return -errno; - } else if (n == 0) - continue; - - assert(n == 1); - - r = process_event(m, &event); - if (r < 0) + r = sd_event_run(m->event, wait_usec); + if (r < 0) { + log_error("Failed to run event loop: %s", strerror(-r)); return r; + } } return m->exit_code; } -int manager_load_unit_from_dbus_path(Manager *m, const char *s, DBusError *e, Unit **_u) { - char *n; +int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) { + _cleanup_free_ char *n = NULL; Unit *u; int r; @@ -1503,16 +1794,11 @@ int manager_load_unit_from_dbus_path(Manager *m, const char *s, DBusError *e, Un assert(s); assert(_u); - if (!startswith(s, "/org/freedesktop/systemd1/unit/")) - return -EINVAL; - - n = bus_path_unescape(s+31); - if (!n) - return -ENOMEM; + r = unit_name_from_dbus_path(s, &n); + if (r < 0) + return r; r = manager_load_unit(m, n, NULL, e, &u); - free(n); - if (r < 0) return r; @@ -1522,21 +1808,25 @@ int manager_load_unit_from_dbus_path(Manager *m, const char *s, DBusError *e, Un } int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) { - Job *j; + const char *p; unsigned id; + Job *j; int r; assert(m); assert(s); assert(_j); - if (!startswith(s, "/org/freedesktop/systemd1/job/")) + p = startswith(s, "/org/freedesktop/systemd1/job/"); + if (!p) return -EINVAL; - if ((r = safe_atou(s + 30, &id)) < 0) + r = safe_atou(p, &id); + if (r < 0) return r; - if (!(j = manager_get_job(m, id))) + j = manager_get_job(m, id); + if (!j) return -ENOENT; *_j = j; @@ -1565,8 +1855,10 @@ void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) { if (u->type != UNIT_SERVICE) return; - if (!(p = unit_name_to_prefix_and_instance(u->id))) { - log_error("Failed to allocate unit name for audit message: %s", strerror(ENOMEM)); + p = unit_name_to_prefix_and_instance(u->id); + if (!p) { + log_error_unit(u->id, + "Failed to allocate unit name for audit message: %s", strerror(ENOMEM)); return; } @@ -1598,6 +1890,9 @@ void manager_send_unit_plymouth(Manager *m, Unit *u) { if (m->running_as != SYSTEMD_SYSTEM) return; + if (detect_container(NULL) > 0) + return; + if (u->type != UNIT_SERVICE && u->type != UNIT_MOUNT && u->type != UNIT_SWAP) @@ -1605,7 +1900,8 @@ void manager_send_unit_plymouth(Manager *m, Unit *u) { /* We set SOCK_NONBLOCK here so that we rather drop the * message then wait for plymouth */ - if ((fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0)) < 0) { + fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0); + if (fd < 0) { log_error("socket() failed: %m"); return; } @@ -1663,33 +1959,16 @@ void manager_dispatch_bus_name_owner_changed( assert(m); assert(name); - if (!(u = hashmap_get(m->watch_bus, name))) + u = hashmap_get(m->watch_bus, name); + if (!u) return; UNIT_VTABLE(u)->bus_name_owner_change(u, name, old_owner, new_owner); } -void manager_dispatch_bus_query_pid_done( - Manager *m, - const char *name, - pid_t pid) { - - Unit *u; - - assert(m); - assert(name); - assert(pid >= 1); - - if (!(u = hashmap_get(m->watch_bus, name))) - return; - - UNIT_VTABLE(u)->bus_query_pid_done(u, name, pid); -} - int manager_open_serialization(Manager *m, FILE **_f) { - char *path = NULL; - mode_t saved_umask; - int fd; + _cleanup_free_ char *path = NULL; + int fd = -1; FILE *f; assert(_f); @@ -1702,32 +1981,32 @@ int manager_open_serialization(Manager *m, FILE **_f) { if (!path) return -ENOMEM; - saved_umask = umask(0077); - fd = mkostemp(path, O_RDWR|O_CLOEXEC); - umask(saved_umask); + RUN_WITH_UMASK(0077) { + fd = mkostemp(path, O_RDWR|O_CLOEXEC); + } - if (fd < 0) { - free(path); + if (fd < 0) return -errno; - } unlink(path); - log_debug("Serializing state to %s", path); - free(path); - if (!(f = fdopen(fd, "w+"))) + f = fdopen(fd, "w+"); + if (!f) { + close_nointr_nofail(fd); return -errno; + } *_f = f; return 0; } -int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool serialize_jobs) { +int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) { Iterator i; Unit *u; const char *t; + char **e; int r; assert(m); @@ -1742,15 +2021,56 @@ int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool serialize_jobs) { fprintf(f, "n-failed-jobs=%u\n", m->n_failed_jobs); dual_timestamp_serialize(f, "firmware-timestamp", &m->firmware_timestamp); - dual_timestamp_serialize(f, "kernel-timestamp", &m->kernel_timestamp); dual_timestamp_serialize(f, "loader-timestamp", &m->loader_timestamp); + dual_timestamp_serialize(f, "kernel-timestamp", &m->kernel_timestamp); dual_timestamp_serialize(f, "initrd-timestamp", &m->initrd_timestamp); if (!in_initrd()) { dual_timestamp_serialize(f, "userspace-timestamp", &m->userspace_timestamp); dual_timestamp_serialize(f, "finish-timestamp", &m->finish_timestamp); + dual_timestamp_serialize(f, "security-start-timestamp", &m->security_start_timestamp); + dual_timestamp_serialize(f, "security-finish-timestamp", &m->security_finish_timestamp); + dual_timestamp_serialize(f, "generators-start-timestamp", &m->generators_start_timestamp); + dual_timestamp_serialize(f, "generators-finish-timestamp", &m->generators_finish_timestamp); + dual_timestamp_serialize(f, "units-load-start-timestamp", &m->units_load_start_timestamp); + dual_timestamp_serialize(f, "units-load-finish-timestamp", &m->units_load_finish_timestamp); } + if (!switching_root) { + STRV_FOREACH(e, m->environment) { + _cleanup_free_ char *ce; + + ce = cescape(*e); + if (!ce) + return -ENOMEM; + + fprintf(f, "env=%s\n", *e); + } + } + + if (m->notify_fd >= 0) { + int copy; + + copy = fdset_put_dup(fds, m->notify_fd); + if (copy < 0) + return copy; + + fprintf(f, "notify-fd=%i\n", copy); + fprintf(f, "notify-socket=%s\n", m->notify_socket); + } + + if (m->kdbus_fd >= 0) { + int copy; + + copy = fdset_put_dup(fds, m->kdbus_fd); + if (copy < 0) + return copy; + + fprintf(f, "kdbus-fd=%i\n", copy); + } + + bus_serialize(m, f); + fputc('\n', f); HASHMAP_FOREACH_KEY(u, t, m->units, i) { @@ -1764,7 +2084,8 @@ int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool serialize_jobs) { fputs(u->id, f); fputc('\n', f); - if ((r = unit_serialize(u, f, fds, serialize_jobs)) < 0) { + r = unit_serialize(u, f, fds, !switching_root); + if (r < 0) { m->n_reloading --; return r; } @@ -1818,6 +2139,7 @@ int manager_deserialize(Manager *m, FILE *f, FDSet *fds) { log_debug("Failed to parse current job id value %s", l+15); else m->current_job_id = MAX(m->current_job_id, id); + } else if (startswith(l, "n-installed-jobs=")) { uint32_t n; @@ -1825,6 +2147,7 @@ int manager_deserialize(Manager *m, FILE *f, FDSet *fds) { log_debug("Failed to parse installed jobs counter %s", l+17); else m->n_installed_jobs += n; + } else if (startswith(l, "n-failed-jobs=")) { uint32_t n; @@ -1832,13 +2155,16 @@ int manager_deserialize(Manager *m, FILE *f, FDSet *fds) { log_debug("Failed to parse failed jobs counter %s", l+14); else m->n_failed_jobs += n; + } else if (startswith(l, "taint-usr=")) { int b; - if ((b = parse_boolean(l+10)) < 0) + b = parse_boolean(l+10); + if (b < 0) log_debug("Failed to parse taint /usr flag %s", l+10); else m->taint_usr = m->taint_usr || b; + } else if (startswith(l, "firmware-timestamp=")) dual_timestamp_deserialize(l+19, &m->firmware_timestamp); else if (startswith(l, "loader-timestamp=")) @@ -1851,7 +2177,76 @@ int manager_deserialize(Manager *m, FILE *f, FDSet *fds) { dual_timestamp_deserialize(l+20, &m->userspace_timestamp); else if (startswith(l, "finish-timestamp=")) dual_timestamp_deserialize(l+17, &m->finish_timestamp); - else + else if (startswith(l, "security-start-timestamp=")) + dual_timestamp_deserialize(l+25, &m->security_start_timestamp); + else if (startswith(l, "security-finish-timestamp=")) + dual_timestamp_deserialize(l+26, &m->security_finish_timestamp); + else if (startswith(l, "generators-start-timestamp=")) + dual_timestamp_deserialize(l+27, &m->generators_start_timestamp); + else if (startswith(l, "generators-finish-timestamp=")) + dual_timestamp_deserialize(l+28, &m->generators_finish_timestamp); + else if (startswith(l, "units-load-start-timestamp=")) + dual_timestamp_deserialize(l+27, &m->units_load_start_timestamp); + else if (startswith(l, "units-load-finish-timestamp=")) + dual_timestamp_deserialize(l+28, &m->units_load_finish_timestamp); + else if (startswith(l, "env=")) { + _cleanup_free_ char *uce = NULL; + char **e; + + uce = cunescape(l+4); + if (!uce) { + r = -ENOMEM; + goto finish; + } + + e = strv_env_set(m->environment, uce); + if (!e) { + r = -ENOMEM; + goto finish; + } + + strv_free(m->environment); + m->environment = e; + + } else if (startswith(l, "notify-fd=")) { + int fd; + + if (safe_atoi(l + 10, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd)) + log_debug("Failed to parse notify fd: %s", l + 10); + else { + if (m->notify_fd >= 0) { + m->notify_event_source = sd_event_source_unref(m->notify_event_source); + close_nointr_nofail(m->notify_fd); + } + + m->notify_fd = fdset_remove(fds, fd); + } + + } else if (startswith(l, "notify-socket=")) { + char *n; + + n = strdup(l+14); + if (!n) { + r = -ENOMEM; + goto finish; + } + + free(m->notify_socket); + m->notify_socket = n; + + } else if (startswith(l, "kdbus-fd=")) { + int fd; + + if (safe_atoi(l + 9, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd)) + log_debug("Failed to parse kdbus fd: %s", l + 9); + else { + if (m->kdbus_fd >= 0) + close_nointr_nofail(m->kdbus_fd); + + m->kdbus_fd = fdset_remove(fds, fd); + } + + } else if (bus_deserialize_item(m, l) == 0) log_debug("Unknown serialization item '%s'", l); } @@ -1871,18 +2266,18 @@ int manager_deserialize(Manager *m, FILE *f, FDSet *fds) { char_array_0(name); - if ((r = manager_load_unit(m, strstrip(name), NULL, NULL, &u)) < 0) + r = manager_load_unit(m, strstrip(name), NULL, NULL, &u); + if (r < 0) goto finish; - if ((r = unit_deserialize(u, f, fds)) < 0) + r = unit_deserialize(u, f, fds); + if (r < 0) goto finish; } finish: - if (ferror(f)) { + if (ferror(f)) r = -EIO; - goto finish; - } assert(m->n_reloading > 0); m->n_reloading --; @@ -1892,8 +2287,8 @@ finish: int manager_reload(Manager *m) { int r, q; - FILE *f; - FDSet *fds; + _cleanup_fclose_ FILE *f = NULL; + _cleanup_fdset_free_ FDSet *fds = NULL; assert(m); @@ -1902,24 +2297,23 @@ int manager_reload(Manager *m) { return r; m->n_reloading ++; + bus_manager_send_reloading(m, true); fds = fdset_new(); if (!fds) { m->n_reloading --; - r = -ENOMEM; - goto finish; + return -ENOMEM; } - r = manager_serialize(m, f, fds, true); + r = manager_serialize(m, f, fds, false); if (r < 0) { m->n_reloading --; - goto finish; + return r; } if (fseeko(f, 0, SEEK_SET) < 0) { m->n_reloading --; - r = -errno; - goto finish; + return -errno; } /* From here on there is no way back. */ @@ -1961,17 +2355,12 @@ int manager_reload(Manager *m) { assert(m->n_reloading > 0); m->n_reloading--; -finish: - if (f) - fclose(f); - - if (fds) - fdset_free(fds); + m->send_reloading_done = true; return r; } -bool manager_is_booting_or_shutting_down(Manager *m) { +static bool manager_is_booting_or_shutting_down(Manager *m) { Unit *u; assert(m); @@ -1988,6 +2377,12 @@ bool manager_is_booting_or_shutting_down(Manager *m) { return false; } +bool manager_is_reloading_or_reexecuting(Manager *m) { + assert(m); + + return m->n_reloading != 0; +} + void manager_reset_failed(Manager *m) { Unit *u; Iterator i; @@ -1998,17 +2393,18 @@ void manager_reset_failed(Manager *m) { unit_reset_failed(u); } -bool manager_unit_pending_inactive(Manager *m, const char *name) { +bool manager_unit_inactive_or_pending(Manager *m, const char *name) { Unit *u; assert(m); assert(name); /* Returns true if the unit is inactive or going down */ - if (!(u = manager_get_unit(m, name))) + u = manager_get_unit(m, name); + if (!u) return true; - return unit_pending_inactive(u); + return unit_inactive_or_pending(u); } void manager_check_finished(Manager *m) { @@ -2017,11 +2413,18 @@ void manager_check_finished(Manager *m) { assert(m); - if (hashmap_size(m->jobs) > 0) + if (m->n_running_jobs == 0) + m->jobs_in_progress_event_source = sd_event_source_unref(m->jobs_in_progress_event_source); + + if (hashmap_size(m->jobs) > 0) { + if (m->jobs_in_progress_event_source) + sd_event_source_set_time(m->jobs_in_progress_event_source, JOBS_IN_PROGRESS_PERIOD_SEC); return; + } /* Notify Type=idle units that we are done now */ - close_pipe(m->idle_pipe); + m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source); + manager_close_idle_pipe(m); /* Turn off confirm spawn now */ m->confirm_spawn = false; @@ -2055,10 +2458,10 @@ void manager_check_finished(Manager *m) { "INITRD_USEC=%llu", (unsigned long long) initrd_usec, "USERSPACE_USEC=%llu", (unsigned long long) userspace_usec, "MESSAGE=Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.", - format_timespan(kernel, sizeof(kernel), kernel_usec), - format_timespan(initrd, sizeof(initrd), initrd_usec), - format_timespan(userspace, sizeof(userspace), userspace_usec), - format_timespan(sum, sizeof(sum), total_usec), + format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC), + format_timespan(initrd, sizeof(initrd), initrd_usec, USEC_PER_MSEC), + format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC), + format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC), NULL); } else { kernel_usec = m->userspace_timestamp.monotonic - m->kernel_timestamp.monotonic; @@ -2070,9 +2473,9 @@ void manager_check_finished(Manager *m) { "KERNEL_USEC=%llu", (unsigned long long) kernel_usec, "USERSPACE_USEC=%llu", (unsigned long long) userspace_usec, "MESSAGE=Startup finished in %s (kernel) + %s (userspace) = %s.", - format_timespan(kernel, sizeof(kernel), kernel_usec), - format_timespan(userspace, sizeof(userspace), userspace_usec), - format_timespan(sum, sizeof(sum), total_usec), + format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC), + format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC), + format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC), NULL); } } else { @@ -2084,15 +2487,15 @@ void manager_check_finished(Manager *m) { MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED), "USERSPACE_USEC=%llu", (unsigned long long) userspace_usec, "MESSAGE=Startup finished in %s.", - format_timespan(sum, sizeof(sum), total_usec), + format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC), NULL); } - bus_broadcast_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec); + bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec); sd_notifyf(false, "READY=1\nSTATUS=Startup finished in %s.", - format_timespan(sum, sizeof(sum), total_usec)); + format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)); } static int create_generator_dir(Manager *m, char **generator, const char *name) { @@ -2114,7 +2517,8 @@ static int create_generator_dir(Manager *m, char **generator, const char *name) r = mkdir_p_label(p, 0755); if (r < 0) { - log_error("Failed to create generator directory: %s", strerror(-r)); + log_error("Failed to create generator directory %s: %s", + p, strerror(-r)); free(p); return r; } @@ -2124,8 +2528,9 @@ static int create_generator_dir(Manager *m, char **generator, const char *name) return log_oom(); if (!mkdtemp(p)) { + log_error("Failed to create generator directory %s: %m", + p); free(p); - log_error("Failed to create generator directory: %m"); return -errno; } } @@ -2150,10 +2555,9 @@ static void trim_generator_dir(Manager *m, char **generator) { } void manager_run_generators(Manager *m) { - DIR *d = NULL; + _cleanup_closedir_ DIR *d = NULL; const char *generator_path; const char *argv[5]; - mode_t u; int r; assert(m); @@ -2164,7 +2568,8 @@ void manager_run_generators(Manager *m) { if (errno == ENOENT) return; - log_error("Failed to enumerate generator directory: %m"); + log_error("Failed to enumerate generator directory %s: %m", + generator_path); return; } @@ -2186,17 +2591,13 @@ void manager_run_generators(Manager *m) { argv[3] = m->generator_unit_path_late; argv[4] = NULL; - u = umask(0022); - execute_directory(generator_path, d, (char**) argv); - umask(u); + RUN_WITH_UMASK(0022) + execute_directory(generator_path, d, (char**) argv); +finish: trim_generator_dir(m, &m->generator_unit_path); trim_generator_dir(m, &m->generator_unit_path_early); trim_generator_dir(m, &m->generator_unit_path_late); - -finish: - if (d) - closedir(d); } static void remove_generator_dir(Manager *m, char **generator) { @@ -2221,20 +2622,36 @@ void manager_undo_generators(Manager *m) { remove_generator_dir(m, &m->generator_unit_path_late); } -int manager_set_default_controllers(Manager *m, char **controllers) { - char **l; - +int manager_environment_add(Manager *m, char **minus, char **plus) { + char **a = NULL, **b = NULL, **l; assert(m); - l = strv_copy(controllers); - if (!l) - return -ENOMEM; + l = m->environment; + + if (!strv_isempty(minus)) { + a = strv_env_delete(l, 1, minus); + if (!a) + return -ENOMEM; + + l = a; + } - strv_free(m->default_controllers); - m->default_controllers = l; + if (!strv_isempty(plus)) { + b = strv_env_merge(2, l, plus); + if (!b) + return -ENOMEM; + + l = b; + } - cg_shorten_controllers(m->default_controllers); + if (m->environment != l) + strv_free(m->environment); + if (a != l) + strv_free(a); + if (b != l) + strv_free(b); + m->environment = strv_sort(l); return 0; } @@ -2294,12 +2711,15 @@ void manager_set_show_status(Manager *m, bool b) { unlink("/run/systemd/show-status"); } -bool manager_get_show_status(Manager *m) { +static bool manager_get_show_status(Manager *m) { assert(m); if (m->running_as != SYSTEMD_SYSTEM) return false; + if (m->no_console_output) + return false; + if (m->show_status) return true; @@ -2309,9 +2729,56 @@ bool manager_get_show_status(Manager *m) { return plymouth_running(); } -void watch_init(Watch *w) { - assert(w); +void manager_status_printf(Manager *m, bool ephemeral, const char *status, const char *format, ...) { + va_list ap; + + if (!manager_get_show_status(m)) + return; + + /* XXX We should totally drop the check for ephemeral here + * and thus effectively make 'Type=idle' pointless. */ + if (ephemeral && m->n_on_console > 0) + return; + + if (!manager_is_booting_or_shutting_down(m)) + return; + + va_start(ap, format); + status_vprintf(status, true, ephemeral, format, ap); + va_end(ap); +} + +int manager_get_unit_by_path(Manager *m, const char *path, const char *suffix, Unit **_found) { + _cleanup_free_ char *p = NULL; + Unit *found; + + assert(m); + assert(path); + assert(suffix); + assert(_found); + + p = unit_name_from_path(path, suffix); + if (!p) + return -ENOMEM; + + found = manager_get_unit(m, p); + if (!found) { + *_found = NULL; + return 0; + } + + *_found = found; + return 1; +} + +Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path) { + char p[strlen(path)+1]; + + assert(m); + assert(path); + + strcpy(p, path); + path_kill_slashes(p); - w->type = WATCH_INVALID; - w->fd = -1; + return hashmap_get(m->units_requiring_mounts_for, streq(p, "/") ? "" : p); }