X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=src%2Fcore%2Fmanager.c;h=f618bed424fdaaf9fabcd91ae00aa85740248304;hp=ba4dab3b86c919e8f79691c543beb3036a082c59;hb=f755e3b74b94296a534033dd6ae04d9506434210;hpb=290837072b136624f1f3d941c7274e4b1b275021 diff --git a/src/core/manager.c b/src/core/manager.c index ba4dab3b8..f618bed42 100644 --- a/src/core/manager.c +++ b/src/core/manager.c @@ -22,9 +22,7 @@ #include #include #include -#include #include -#include #include #include #include @@ -76,13 +74,14 @@ #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_WAIT_USEC (5*USEC_PER_SEC) +#define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3) #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3 /* Where clients shall send notification messages to */ @@ -97,62 +96,16 @@ static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32 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_setup_notify(Manager *m) { - union { - struct sockaddr sa; - struct sockaddr_un un; - } sa = { - .sa.sa_family = AF_UNIX, - }; - int one = 1, r; - - m->notify_fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0); - if (m->notify_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 "/%llu", random_ull()); - else - strncpy(sa.un.sun_path, NOTIFY_SOCKET, sizeof(sa.un.sun_path)); - sa.un.sun_path[0] = 0; - - r = bind(m->notify_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(m->notify_fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one)); - if (r < 0) { - log_error("SO_PASSCRED failed: %m"); - return -errno; - } - - 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; - } - - sa.un.sun_path[0] = '@'; - m->notify_socket = strdup(sa.un.sun_path); - if (!m->notify_socket) - return log_oom(); - - log_debug("Using notification socket %s", m->notify_socket); - - return 0; -} - static int manager_watch_jobs_in_progress(Manager *m) { + usec_t next; + assert(m); if (m->jobs_in_progress_event_source) return 0; - 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); + next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC; + return sd_event_add_monotonic(m->event, &m->jobs_in_progress_event_source, next, 0, manager_dispatch_jobs_in_progress, m); } #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED_ON)-1) + sizeof(ANSI_HIGHLIGHT_RED_ON)-1 + 2*(sizeof(ANSI_HIGHLIGHT_OFF)-1)) @@ -186,6 +139,18 @@ static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned po } } +void manager_flip_auto_status(Manager *m, bool enable) { + assert(m); + + if (enable) { + if (m->show_status == SHOW_STATUS_AUTO) + manager_set_show_status(m, SHOW_STATUS_TEMPORARY); + } else { + if (m->show_status == SHOW_STATUS_TEMPORARY) + manager_set_show_status(m, SHOW_STATUS_AUTO); + } +} + static void manager_print_jobs_in_progress(Manager *m) { _cleanup_free_ char *job_of_n = NULL; Iterator i; @@ -193,9 +158,13 @@ static void manager_print_jobs_in_progress(Manager *m) { unsigned counter = 0, print_nr; char cylon[6 + CYLON_BUFFER_EXTRA + 1]; unsigned cylon_pos; + char time[FORMAT_TIMESPAN_MAX], limit[FORMAT_TIMESPAN_MAX] = "no limit"; + uint64_t x; assert(m); + manager_flip_auto_status(m, true); + print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs; HASHMAP_FOREACH(j, m->jobs, i) @@ -212,14 +181,23 @@ static void manager_print_jobs_in_progress(Manager *m) { cylon_pos = 14 - cylon_pos; draw_cylon(cylon, sizeof(cylon), 6, cylon_pos); + m->jobs_in_progress_iteration++; + 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)); + format_timespan(time, sizeof(time), now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC); + if (job_get_timeout(j, &x) > 0) + format_timespan(limit, sizeof(limit), x - j->begin_usec, 1*USEC_PER_SEC); + + manager_status_printf(m, true, cylon, + "%sA %s job is running for %s (%s / %s)", + strempty(job_of_n), + job_type_to_string(j->type), + unit_description(j->unit), + time, limit); - m->jobs_in_progress_iteration++; } static int manager_watch_idle_pipe(Manager *m) { @@ -233,7 +211,7 @@ static int manager_watch_idle_pipe(Manager *m) { 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); + r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m); if (r < 0) { log_error("Failed to watch idle pipe: %s", strerror(-r)); return r; @@ -277,7 +255,7 @@ static int manager_setup_time_change(Manager *m) { return 0; } - r = sd_event_add_io(m->event, m->time_change_fd, EPOLLIN, manager_dispatch_time_change_fd, m, &m->time_change_event_source); + r = sd_event_add_io(m->event, &m->time_change_event_source, m->time_change_fd, EPOLLIN, manager_dispatch_time_change_fd, m); if (r < 0) { log_error("Failed to create time change event source: %s", strerror(-r)); return r; @@ -307,7 +285,7 @@ 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)); + log_warning("Failed to enable kbrequest handling: %m"); } return 0; @@ -364,11 +342,14 @@ static int manager_setup_signals(Manager *m) { if (m->signal_fd < 0) return -errno; - r = sd_event_add_io(m->event, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m, &m->signal_event_source); + r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m); if (r < 0) return r; - /* Process signals a bit earlier than the rest of things */ + /* 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; @@ -379,6 +360,23 @@ static int manager_setup_signals(Manager *m) { return 0; } +static void manager_clean_environment(Manager *m) { + assert(m); + + /* Let's remove some environment variables that we + * need ourselves to communicate with our clients */ + strv_env_unset_many( + m->environment, + "NOTIFY_SOCKET", + "MAINPID", + "MANAGERPID", + "LISTEN_PID", + "LISTEN_FDS", + "WATCHDOG_PID", + "WATCHDOG_USEC", + NULL); +} + static int manager_default_environment(Manager *m) { assert(m); @@ -395,23 +393,24 @@ static int manager_default_environment(Manager *m) { /* Import locale variables LC_*= from configuration */ locale_setup(&m->environment); - } else + } else { /* The user manager passes its own environment * along to its children. */ m->environment = strv_copy(environ); + } if (!m->environment) return -ENOMEM; + manager_clean_environment(m); strv_sort(m->environment); return 0; } -int manager_new(SystemdRunningAs running_as, bool reexecuting, Manager **_m) { +int manager_new(SystemdRunningAs running_as, Manager **_m) { Manager *m; - int r = -ENOMEM; - bool try_bus_connect = false; + int r; assert(_m); assert(running_as >= 0); @@ -431,7 +430,7 @@ int manager_new(SystemdRunningAs running_as, bool reexecuting, Manager **_m) { m->idle_pipe[0] = m->idle_pipe[1] = m->idle_pipe[2] = m->idle_pipe[3] = -1; - m->pin_cgroupfs_fd = m->notify_fd = m->signal_fd = m->time_change_fd = m->dev_autofs_fd = m->private_listen_fd = -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 */ r = manager_default_environment(m); @@ -450,11 +449,11 @@ int manager_new(SystemdRunningAs running_as, bool reexecuting, Manager **_m) { if (r < 0) goto fail; - r = hashmap_ensure_allocated(&m->watch_pids, trivial_hash_func, trivial_compare_func); + r = hashmap_ensure_allocated(&m->watch_bus, string_hash_func, string_compare_func); if (r < 0) goto fail; - r = hashmap_ensure_allocated(&m->watch_bus, string_hash_func, string_compare_func); + r = set_ensure_allocated(&m->failed_units, trivial_hash_func, trivial_compare_func); if (r < 0) goto fail; @@ -462,7 +461,7 @@ int manager_new(SystemdRunningAs running_as, bool reexecuting, Manager **_m) { if (r < 0) goto fail; - r = sd_event_add_defer(m->event, manager_dispatch_run_queue, m, &m->run_queue_event_source); + r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m); if (r < 0) goto fail; @@ -482,25 +481,19 @@ int manager_new(SystemdRunningAs running_as, bool reexecuting, Manager **_m) { if (r < 0) goto fail; - r = manager_setup_notify(m); - if (r < 0) - goto fail; - r = manager_setup_time_change(m); if (r < 0) goto fail; - if (running_as == SYSTEMD_SYSTEM) - try_bus_connect = reexecuting; - else if (getenv("DBUS_SESSION_BUS_ADDRESS")) - try_bus_connect = true; - else - log_debug("Skipping DBus session bus connection attempt - no DBUS_SESSION_BUS_ADDRESS set..."); - - /* Try to connect to the busses, if possible. */ - r = bus_init(m, try_bus_connect); - if (r < 0) + 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; @@ -512,6 +505,121 @@ 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_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m); + 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) { +#ifdef ENABLE_KDBUS + _cleanup_free_ char *p = NULL; +#endif + +#ifdef ENABLE_KDBUS + assert(m); + + if (m->kdbus_fd >= 0) + 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. */ + if (m->running_as == SYSTEMD_SYSTEM) + mkdir_p_label("/dev/kdbus/domain", 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; @@ -672,9 +780,12 @@ void manager_free(Manager *m) { hashmap_free(m->units); hashmap_free(m->jobs); - hashmap_free(m->watch_pids); + hashmap_free(m->watch_pids1); + hashmap_free(m->watch_pids2); hashmap_free(m->watch_bus); + set_free(m->failed_units); + 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); @@ -688,9 +799,12 @@ void manager_free(Manager *m) { 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); @@ -704,7 +818,7 @@ void manager_free(Manager *m) { free(m->switch_root); free(m->switch_root_init); - for (i = 0; i < RLIMIT_NLIMITS; i++) + for (i = 0; i < _RLIMIT_MAX; i++) free(m->rlimit[i]); assert(hashmap_isempty(m->units_requiring_mounts_for)); @@ -733,7 +847,7 @@ int manager_enumerate(Manager *m) { } static int manager_coldplug(Manager *m) { - int r = 0, q; + int r = 0; Iterator i; Unit *u; char *k; @@ -742,12 +856,14 @@ static int manager_coldplug(Manager *m) { /* Then, let's set up their initial state. */ HASHMAP_FOREACH_KEY(u, k, m->units, i) { + int q; /* ignore aliases */ if (u->id != k) continue; - if ((q = unit_coldplug(u)) < 0) + q = unit_coldplug(u); + if (q < 0) r = q; } @@ -882,6 +998,16 @@ int manager_startup(Manager *m, FILE *serialization, FDSet *fds) { 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); + bus_track_coldplug(m, &m->subscribed, &m->deserialized_subscribed); + /* Third, fire things up! */ q = manager_coldplug(m); if (q < 0) @@ -1037,11 +1163,11 @@ int manager_load_unit_prepare( 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)) + 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); @@ -1201,6 +1327,26 @@ static unsigned manager_dispatch_dbus_queue(Manager *m) { return n; } +static void manager_invoke_notify_message(Manager *m, Unit *u, pid_t pid, char *buf, size_t n) { + _cleanup_strv_free_ char **tags = NULL; + + assert(m); + assert(u); + assert(buf); + assert(n > 0); + + tags = strv_split(buf, "\n\r"); + if (!tags) { + log_oom(); + return; + } + + 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, pid, tags); +} + static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { Manager *m = userdata; ssize_t n; @@ -1219,6 +1365,7 @@ static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t .iov_base = buf, .iov_len = sizeof(buf)-1, }; + bool found = false; union { struct cmsghdr cmsghdr; @@ -1233,7 +1380,6 @@ static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t }; struct ucred *ucred; Unit *u; - _cleanup_strv_free_ char **tags = NULL; n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT); if (n <= 0) { @@ -1256,37 +1402,50 @@ static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t ucred = (struct ucred*) CMSG_DATA(&control.cmsghdr); - 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; - tags = strv_split(buf, "\n\r"); - if (!tags) - return log_oom(); - log_debug_unit(u->id, "Got notification message for unit %s", u->id); + u = manager_get_unit_by_pid(m, ucred->pid); + if (u) { + manager_invoke_notify_message(m, u, ucred->pid, buf, n); + found = true; + } + + u = hashmap_get(m->watch_pids1, LONG_TO_PTR(ucred->pid)); + if (u) { + manager_invoke_notify_message(m, u, ucred->pid, buf, n); + found = true; + } + + u = hashmap_get(m->watch_pids2, LONG_TO_PTR(ucred->pid)); + if (u) { + manager_invoke_notify_message(m, u, ucred->pid, buf, n); + found = true; + } - if (UNIT_VTABLE(u)->notify_message) - UNIT_VTABLE(u)->notify_message(u, ucred->pid, tags); + if (!found) + log_warning("Cannot find unit for notify message of PID "PID_FMT".", ucred->pid); } return 0; } +static void invoke_sigchld_event(Manager *m, Unit *u, siginfo_t *si) { + assert(m); + assert(u); + assert(si); + + log_debug_unit(u->id, "Child "PID_FMT" belongs to %s", si->si_pid, u->id); + + unit_unwatch_pid(u, si->si_pid); + UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status); +} + static int manager_dispatch_sigchld(Manager *m) { assert(m); for (;;) { siginfo_t si = {}; - Unit *u; - int r; /* First we call waitd() for a PID and do not reap the * zombie. That way we can still access /proc/$PID for @@ -1307,23 +1466,30 @@ static int manager_dispatch_sigchld(Manager *m) { if (si.si_code == CLD_EXITED || si.si_code == CLD_KILLED || si.si_code == CLD_DUMPED) { _cleanup_free_ char *name = NULL; + Unit *u; get_process_comm(si.si_pid, &name); - log_debug("Got SIGCHLD for process %lu (%s)", (unsigned long) si.si_pid, strna(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. */ - r = manager_dispatch_notify_fd(m->notify_event_source, m->notify_fd, EPOLLIN, m); - if (r < 0) - return r; + log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)", + si.si_pid, strna(name), + sigchld_code_to_string(si.si_code), + si.si_status, + strna(si.si_code == CLD_EXITED + ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL) + : signal_to_string(si.si_status))); - /* And now figure out the unit this belongs to */ - u = hashmap_get(m->watch_pids, LONG_TO_PTR(si.si_pid)); - if (!u) + /* And now figure out the unit this belongs + * to, it might be multiple... */ u = manager_get_unit_by_pid(m, si.si_pid); + if (u) + invoke_sigchld_event(m, u, &si); + u = hashmap_get(m->watch_pids1, LONG_TO_PTR(si.si_pid)); + if (u) + invoke_sigchld_event(m, u, &si); + u = hashmap_get(m->watch_pids2, LONG_TO_PTR(si.si_pid)); + if (u) + invoke_sigchld_event(m, u, &si); + } /* And now, we actually reap the zombie. */ if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) { @@ -1332,26 +1498,6 @@ static int manager_dispatch_sigchld(Manager *m) { return -errno; } - - if (si.si_code != CLD_EXITED && si.si_code != CLD_KILLED && si.si_code != CLD_DUMPED) - continue; - - log_debug("Child %lu died (code=%s, status=%i/%s)", - (long unsigned) si.si_pid, - sigchld_code_to_string(si.si_code), - si.si_status, - strna(si.si_code == CLD_EXITED - ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL) - : signal_to_string(si.si_status))); - - if (!u) - continue; - - 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); } return 0; @@ -1397,17 +1543,10 @@ static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t return -errno; } - if (sfsi.ssi_pid > 0) { - 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); - } else - log_debug("Received SIG%s.", signal_to_string(sfsi.ssi_signo)); + log_received_signal(sfsi.ssi_signo == SIGCHLD || + (sfsi.ssi_signo == SIGTERM && m->running_as == SYSTEMD_USER) + ? LOG_DEBUG : LOG_INFO, + &sfsi); switch (sfsi.ssi_signo) { @@ -1490,6 +1629,11 @@ static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t break; } + if (fflush(f)) { + log_warning("Failed to flush status stream"); + break; + } + log_dump(LOG_INFO, dump); break; } @@ -1537,12 +1681,12 @@ static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t case 20: log_debug("Enabling showing of status."); - manager_set_show_status(m, true); + manager_set_show_status(m, SHOW_STATUS_YES); break; case 21: log_debug("Disabling showing of status."); - manager_set_show_status(m, false); + manager_set_show_status(m, SHOW_STATUS_NO); break; case 22: @@ -1592,7 +1736,7 @@ static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t } if (sigchld) - return manager_dispatch_sigchld(m); + manager_dispatch_sigchld(m); return 0; } @@ -1641,11 +1785,20 @@ static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) { Manager *m = userdata; + int r; + uint64_t next; assert(m); + assert(source); manager_print_jobs_in_progress(m); - return 0; + + next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_PERIOD_USEC; + r = sd_event_source_set_time(source, next); + if (r < 0) + return r; + + return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT); } int manager_loop(Manager *m) { @@ -1654,7 +1807,7 @@ int manager_loop(Manager *m) { RATELIMIT_DEFINE(rl, 1*USEC_PER_SEC, 50000); assert(m); - m->exit_code = MANAGER_RUNNING; + m->exit_code = MANAGER_OK; /* Release the path cache */ set_free_free(m->unit_path_cache); @@ -1668,7 +1821,7 @@ int manager_loop(Manager *m) { if (r < 0) return r; - while (m->exit_code == MANAGER_RUNNING) { + while (m->exit_code == MANAGER_OK) { usec_t wait_usec; if (m->runtime_watchdog > 0 && m->running_as == SYSTEMD_SYSTEM) @@ -1766,7 +1919,7 @@ int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) { void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) { #ifdef HAVE_AUDIT - char *p; + _cleanup_free_ char *p = NULL; int audit_fd; audit_fd = get_audit_fd(); @@ -1799,17 +1952,19 @@ void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) { } else log_warning("Failed to send audit message: %m"); } - - free(p); #endif } void manager_send_unit_plymouth(Manager *m, Unit *u) { - int fd = -1; - union sockaddr_union sa; + union sockaddr_union sa = { + .un.sun_family = AF_UNIX, + .un.sun_path = "\0/org/freedesktop/plymouthd", + }; + int n = 0; - char *message = NULL; + _cleanup_free_ char *message = NULL; + _cleanup_close_ int fd = -1; /* Don't generate plymouth events if the service was already * started and we're just deserializing */ @@ -1835,46 +1990,22 @@ void manager_send_unit_plymouth(Manager *m, Unit *u) { return; } - zero(sa); - sa.sa.sa_family = AF_UNIX; - strncpy(sa.un.sun_path+1, "/org/freedesktop/plymouthd", sizeof(sa.un.sun_path)-1); if (connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + 1 + strlen(sa.un.sun_path+1)) < 0) { - if (errno != EPIPE && - errno != EAGAIN && - errno != ENOENT && - errno != ECONNREFUSED && - errno != ECONNRESET && - errno != ECONNABORTED) + if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED)) log_error("connect() failed: %m"); - - goto finish; + return; } if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) { log_oom(); - goto finish; + return; } errno = 0; - if (write(fd, message, n + 1) != n + 1) { - - if (errno != EPIPE && - errno != EAGAIN && - errno != ENOENT && - errno != ECONNREFUSED && - errno != ECONNRESET && - errno != ECONNABORTED) + if (write(fd, message, n + 1) != n + 1) + if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED)) log_error("Failed to write Plymouth message: %m"); - - goto finish; - } - -finish: - if (fd >= 0) - close_nointr_nofail(fd); - - free(message); } void manager_dispatch_bus_name_owner_changed( @@ -1896,37 +2027,24 @@ void manager_dispatch_bus_name_owner_changed( } int manager_open_serialization(Manager *m, FILE **_f) { - char *path = NULL; + const char *path; int fd = -1; FILE *f; assert(_f); - if (m->running_as == SYSTEMD_SYSTEM) - asprintf(&path, "/run/systemd/dump-%lu-XXXXXX", (unsigned long) getpid()); - else - asprintf(&path, "/tmp/systemd-dump-%lu-XXXXXX", (unsigned long) getpid()); - - if (!path) - return -ENOMEM; - - RUN_WITH_UMASK(0077) { - fd = mkostemp(path, O_RDWR|O_CLOEXEC); - } - - if (fd < 0) { - free(path); + path = m->running_as == SYSTEMD_SYSTEM ? "/run/systemd" : "/tmp"; + fd = open_tmpfile(path, O_RDWR|O_CLOEXEC); + if (fd < 0) return -errno; - } - - unlink(path); log_debug("Serializing state to %s", path); - free(path); f = fdopen(fd, "w+"); - if (!f) + if (!f) { + close_nointr_nofail(fd); return -errno; + } *_f = f; @@ -1972,12 +2090,35 @@ int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) { _cleanup_free_ char *ce; ce = cescape(*e); - if (ce) - fprintf(f, "env=%s\n", *e); + if (!ce) + return -ENOMEM; + + fprintf(f, "env=%s\n", *e); } } - bus_serialize(m, f); + 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_track_serialize(m->subscribed, f); fputc('\n', f); @@ -2067,7 +2208,8 @@ int manager_deserialize(Manager *m, FILE *f, FDSet *fds) { } 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; @@ -2114,7 +2256,46 @@ int manager_deserialize(Manager *m, FILE *f, FDSet *fds) { strv_free(m->environment); m->environment = e; - } else if (bus_deserialize_item(m, l) == 0) + + } 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_track_deserialize_item(&m->deserialized_subscribed, l) == 0) log_debug("Unknown serialization item '%s'", l); } @@ -2144,10 +2325,8 @@ int manager_deserialize(Manager *m, FILE *f, FDSet *fds) { } finish: - if (ferror(f)) { + if (ferror(f)) r = -EIO; - goto finish; - } assert(m->n_reloading > 0); m->n_reloading --; @@ -2217,6 +2396,11 @@ int manager_reload(Manager *m) { fclose(f); f = NULL; + /* Re-register notify_fd as event source */ + q = manager_setup_notify(m); + if (q < 0) + r = q; + /* Third, fire things up! */ q = manager_coldplug(m); if (q < 0) @@ -2287,11 +2471,17 @@ void manager_check_finished(Manager *m) { 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); + + if (m->jobs_in_progress_event_source) { + sd_event_source_set_time(m->jobs_in_progress_event_source, + now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC); + } + return; } + manager_flip_auto_status(m, false); + /* Notify Type=idle units that we are done now */ m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source); manager_close_idle_pipe(m); @@ -2324,9 +2514,9 @@ void manager_check_finished(Manager *m) { if (!log_on_console()) log_struct(LOG_INFO, MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED), - "KERNEL_USEC=%llu", (unsigned long long) kernel_usec, - "INITRD_USEC=%llu", (unsigned long long) initrd_usec, - "USERSPACE_USEC=%llu", (unsigned long long) userspace_usec, + "KERNEL_USEC="USEC_FMT, kernel_usec, + "INITRD_USEC="USEC_FMT, initrd_usec, + "USERSPACE_USEC="USEC_FMT, userspace_usec, "MESSAGE=Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.", format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC), format_timespan(initrd, sizeof(initrd), initrd_usec, USEC_PER_MSEC), @@ -2340,8 +2530,8 @@ void manager_check_finished(Manager *m) { if (!log_on_console()) log_struct(LOG_INFO, MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED), - "KERNEL_USEC=%llu", (unsigned long long) kernel_usec, - "USERSPACE_USEC=%llu", (unsigned long long) userspace_usec, + "KERNEL_USEC="USEC_FMT, kernel_usec, + "USERSPACE_USEC="USEC_FMT, userspace_usec, "MESSAGE=Startup finished in %s (kernel) + %s (userspace) = %s.", format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC), format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC), @@ -2355,7 +2545,7 @@ void manager_check_finished(Manager *m) { if (!log_on_console()) log_struct(LOG_INFO, MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED), - "USERSPACE_USEC=%llu", (unsigned long long) userspace_usec, + "USERSPACE_USEC="USEC_FMT, userspace_usec, "MESSAGE=Startup finished in %s.", format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC), NULL); @@ -2380,11 +2570,29 @@ static int create_generator_dir(Manager *m, char **generator, const char *name) return 0; if (m->running_as == SYSTEMD_SYSTEM && getpid() == 1) { + /* systemd --system, not running --test */ p = strappend("/run/systemd/", name); if (!p) return log_oom(); + r = mkdir_p_label(p, 0755); + if (r < 0) { + log_error("Failed to create generator directory %s: %s", + p, strerror(-r)); + free(p); + return r; + } + } else if (m->running_as == SYSTEMD_USER) { + const char *s = NULL; + + s = getenv("XDG_RUNTIME_DIR"); + if (!s) + return -EINVAL; + p = strjoin(s, "/systemd/", name, NULL); + if (!p) + return log_oom(); + r = mkdir_p_label(p, 0755); if (r < 0) { log_error("Failed to create generator directory %s: %s", @@ -2393,6 +2601,8 @@ static int create_generator_dir(Manager *m, char **generator, const char *name) return r; } } else { + /* systemd --system --test */ + p = strjoin("/tmp/systemd-", name, ".XXXXXX", NULL); if (!p) return log_oom(); @@ -2462,7 +2672,7 @@ void manager_run_generators(Manager *m) { argv[4] = NULL; RUN_WITH_UMASK(0022) - execute_directory(generator_path, d, (char**) argv); + execute_directory(generator_path, d, DEFAULT_TIMEOUT_USEC, (char**) argv); finish: trim_generator_dir(m, &m->generator_unit_path); @@ -2521,7 +2731,10 @@ int manager_environment_add(Manager *m, char **minus, char **plus) { if (b != l) strv_free(b); - m->environment = strv_sort(l); + m->environment = l; + manager_clean_environment(m); + strv_sort(m->environment); + return 0; } @@ -2530,7 +2743,7 @@ int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) { assert(m); - for (i = 0; i < RLIMIT_NLIMITS; i++) { + for (i = 0; i < _RLIMIT_MAX; i++) { if (!default_rlimit[i]) continue; @@ -2567,15 +2780,16 @@ void manager_recheck_journal(Manager *m) { log_open(); } -void manager_set_show_status(Manager *m, bool b) { +void manager_set_show_status(Manager *m, ShowStatus mode) { assert(m); + assert(IN_SET(mode, SHOW_STATUS_AUTO, SHOW_STATUS_NO, SHOW_STATUS_YES, SHOW_STATUS_TEMPORARY)); if (m->running_as != SYSTEMD_SYSTEM) return; - m->show_status = b; + m->show_status = mode; - if (b) + if (mode > 0) touch("/run/systemd/show-status"); else unlink("/run/systemd/show-status"); @@ -2590,7 +2804,7 @@ static bool manager_get_show_status(Manager *m) { if (m->no_console_output) return false; - if (m->show_status) + if (m->show_status > 0) return true; /* If Plymouth is running make sure we show the status, so @@ -2652,3 +2866,53 @@ Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path) { return hashmap_get(m->units_requiring_mounts_for, streq(p, "/") ? "" : p); } + +const char *manager_get_runtime_prefix(Manager *m) { + assert(m); + + return m->running_as == SYSTEMD_SYSTEM ? + "/run" : + getenv("XDG_RUNTIME_DIR"); +} + +ManagerState manager_state(Manager *m) { + Unit *u; + + assert(m); + + /* Did we ever finish booting? If not then we are still starting up */ + if (!dual_timestamp_is_set(&m->finish_timestamp)) + return MANAGER_STARTING; + + /* Is the special shutdown target queued? If so, we are in shutdown state */ + u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET); + if (u && u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_TRY_RESTART, JOB_RELOAD_OR_START)) + return MANAGER_STOPPING; + + /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */ + u = manager_get_unit(m, SPECIAL_RESCUE_TARGET); + if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) || + (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_TRY_RESTART, JOB_RELOAD_OR_START)))) + return MANAGER_MAINTENANCE; + + u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET); + if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) || + (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_TRY_RESTART, JOB_RELOAD_OR_START)))) + return MANAGER_MAINTENANCE; + + /* Are there any failed units? If so, we are in degraded mode */ + if (set_size(m->failed_units) > 0) + return MANAGER_DEGRADED; + + return MANAGER_RUNNING; +} + +static const char *const manager_state_table[_MANAGER_STATE_MAX] = { + [MANAGER_STARTING] = "starting", + [MANAGER_RUNNING] = "running", + [MANAGER_DEGRADED] = "degraded", + [MANAGER_MAINTENANCE] = "maintenance", + [MANAGER_STOPPING] = "stopping", +}; + +DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);