X-Git-Url: https://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=src%2Fmanager.c;h=5e627ba9c674dd14b030882aed7335f06818ac0a;hp=4dcdf2e507df4b7b0e54bbc54a3d047e6563f075;hb=33be102a214e7010949496549f4c737b0f8269a3;hpb=84e3543ef4c4758621f8a304b14642072303ef82 diff --git a/src/manager.c b/src/manager.c index 4dcdf2e50..5e627ba9c 100644 --- a/src/manager.c +++ b/src/manager.c @@ -53,6 +53,7 @@ #include "dbus-job.h" #include "missing.h" #include "path-lookup.h" +#include "special.h" /* As soon as 16 units are in our GC queue, make sure to run a gc sweep */ #define GC_QUEUE_ENTRIES_MAX 16 @@ -60,6 +61,67 @@ /* 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) +/* 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; + char *ne[2], **t; + int one = 1; + + assert(m); + + m->notify_watch.type = WATCH_NOTIFY; + if ((m->notify_watch.fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0)) < 0) { + log_error("Failed to allocate notification socket: %m"); + return -errno; + } + + zero(sa); + sa.sa.sa_family = AF_UNIX; + + if (m->running_as == MANAGER_SESSION) + snprintf(sa.un.sun_path+1, sizeof(sa.un.sun_path)-1, NOTIFY_SOCKET "/%llu", random_ull()); + else + strncpy(sa.un.sun_path+1, NOTIFY_SOCKET, sizeof(sa.un.sun_path)-1); + + if (bind(m->notify_watch.fd, &sa.sa, sizeof(sa)) < 0) { + log_error("bind() failed: %m"); + return -errno; + } + + if (setsockopt(m->notify_watch.fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one)) < 0) { + log_error("SO_PASSCRED failed: %m"); + return -errno; + } + + zero(ev); + ev.events = EPOLLIN; + ev.data.ptr = &m->notify_watch; + + if (epoll_ctl(m->epoll_fd, EPOLL_CTL_ADD, m->notify_watch.fd, &ev) < 0) + return -errno; + + if (asprintf(&ne[0], "NOTIFY_SOCKET=@%s", sa.un.sun_path+1) < 0) + return -ENOMEM; + + ne[1] = NULL; + t = strv_env_merge(2, m->environment, ne); + free(ne[0]); + + if (!t) + return -ENOMEM; + + strv_free(m->environment); + m->environment = t; + + return 0; +} + static int enable_special_signals(Manager *m) { char fd; @@ -96,14 +158,23 @@ static int manager_setup_signals(Manager *m) { assert_se(sigaction(SIGCHLD, &sa, NULL) == 0); assert_se(sigemptyset(&mask) == 0); - assert_se(sigaddset(&mask, SIGCHLD) == 0); - assert_se(sigaddset(&mask, SIGTERM) == 0); - assert_se(sigaddset(&mask, SIGHUP) == 0); - assert_se(sigaddset(&mask, SIGUSR1) == 0); - assert_se(sigaddset(&mask, SIGUSR2) == 0); - assert_se(sigaddset(&mask, SIGINT) == 0); /* Kernel sends us this on control-alt-del */ - assert_se(sigaddset(&mask, SIGWINCH) == 0); /* Kernel sends us this on kbrequest (alt-arrowup) */ - assert_se(sigaddset(&mask, SIGPWR) == 0); /* Some kernel drivers and upsd send us this on power failure */ + + sigset_add_many(&mask, + SIGCHLD, /* Child died */ + SIGTERM, /* Reexecute daemon */ + SIGHUP, /* Reload configuration */ + SIGUSR1, /* systemd/upstart: reconnect to D-Bus */ + SIGUSR2, /* systemd: dump status */ + SIGINT, /* Kernel sends us this on control-alt-del */ + SIGWINCH, /* Kernel sends us this on kbrequest (alt-arrowup) */ + SIGPWR, /* Some kernel drivers and upsd send us this on power failure */ + SIGRTMIN+0, /* systemd: start default.target */ + SIGRTMIN+1, /* systemd: start rescue.target */ + SIGRTMIN+2, /* systemd: isolate emergency.target */ + SIGRTMIN+3, /* systemd: start halt.target */ + SIGRTMIN+4, /* systemd: start poweroff.target */ + SIGRTMIN+5, /* systemd: start reboot.target */ + -1); assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); m->signal_watch.type = WATCH_SIGNAL; @@ -126,6 +197,7 @@ static int manager_setup_signals(Manager *m) { int manager_new(ManagerRunningAs running_as, bool confirm_spawn, Manager **_m) { Manager *m; int r = -ENOMEM; + char *p; assert(_m); assert(running_as >= 0); @@ -140,6 +212,7 @@ int manager_new(ManagerRunningAs running_as, bool confirm_spawn, Manager **_m) { m->confirm_spawn = confirm_spawn; m->name_data_slot = -1; m->exit_code = _MANAGER_EXIT_CODE_INVALID; + m->pin_cgroupfs_fd = -1; m->signal_watch.fd = m->mount_watch.fd = m->udev_watch.fd = m->epoll_fd = m->dev_autofs_fd = -1; m->current_job_id = 1; /* start as id #1, so that we can leave #0 around as "null-like" value */ @@ -177,11 +250,22 @@ int manager_new(ManagerRunningAs running_as, bool confirm_spawn, Manager **_m) { if ((r = manager_setup_cgroup(m)) < 0) goto fail; + if ((r = manager_setup_notify(m)) < 0) + goto fail; + /* Try to connect to the busses, if possible. */ if ((r = bus_init_system(m)) < 0 || (r = bus_init_api(m)) < 0) goto fail; + if (asprintf(&p, "%s/%s", m->cgroup_mount_point, m->cgroup_hierarchy) < 0) { + r = -ENOMEM; + goto fail; + } + + m->pin_cgroupfs_fd = open(p, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); + free(p); + *_m = m; return 0; @@ -364,15 +448,21 @@ void manager_free(Manager *m) { 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); lookup_paths_free(&m->lookup_paths); strv_free(m->environment); free(m->cgroup_controller); free(m->cgroup_hierarchy); + free(m->cgroup_mount_point); hashmap_free(m->cgroup_bondings); + if (m->pin_cgroupfs_fd >= 0) + close_nointr_nofail(m->pin_cgroupfs_fd); + free(m); } @@ -1521,12 +1611,82 @@ unsigned manager_dispatch_dbus_queue(Manager *m) { return n; } +static int manager_process_notify_fd(Manager *m) { + ssize_t n; + + assert(m); + + for (;;) { + char buf[4096]; + struct msghdr msghdr; + struct iovec iovec; + struct ucred *ucred; + union { + struct cmsghdr cmsghdr; + uint8_t buf[CMSG_SPACE(sizeof(struct ucred))]; + } control; + Unit *u; + char **tags; + + 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) + return -EIO; + + if (errno == EAGAIN) + break; + + return -errno; + } + + if (msghdr.msg_controllen < CMSG_LEN(sizeof(struct ucred)) || + control.cmsghdr.cmsg_level != SOL_SOCKET || + control.cmsghdr.cmsg_type != SCM_CREDENTIALS || + control.cmsghdr.cmsg_len != CMSG_LEN(sizeof(struct ucred))) { + log_warning("Received notify message without credentials. Ignoring."); + continue; + } + + ucred = (struct ucred*) CMSG_DATA(&control.cmsghdr); + + if (!(u = hashmap_get(m->watch_pids, UINT32_TO_PTR(ucred->pid)))) + if (!(u = cgroup_unit_by_pid(m, ucred->pid))) { + log_warning("Cannot find unit for notify message of PID %lu.", (unsigned long) ucred->pid); + continue; + } + + char_array_0(buf); + if (!(tags = strv_split(buf, "\n\r"))) + return -ENOMEM; + + log_debug("Got notification message for unit %s", u->meta.id); + + if (UNIT_VTABLE(u)->notify_message) + UNIT_VTABLE(u)->notify_message(u, tags); + + strv_free(tags); + } + + return 0; +} + static int manager_dispatch_sigchld(Manager *m) { assert(m); for (;;) { siginfo_t si; Unit *u; + int r; zero(si); @@ -1555,6 +1715,17 @@ static int manager_dispatch_sigchld(Manager *m) { 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, UINT32_TO_PTR(si.si_pid)))) + u = cgroup_unit_by_pid(m, si.si_pid); + /* And now, we actually reap the zombie. */ if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) { if (errno == EINTR) @@ -1572,21 +1743,22 @@ static int manager_dispatch_sigchld(Manager *m) { si.si_status, strna(si.si_code == CLD_EXITED ? exit_status_to_string(si.si_status) : strsignal(si.si_status))); - if (!(u = hashmap_remove(m->watch_pids, UINT32_TO_PTR(si.si_pid)))) + if (!u) continue; log_debug("Child %llu belongs to %s", (long long unsigned) si.si_pid, u->meta.id); + hashmap_remove(m->watch_pids, UINT32_TO_PTR(si.si_pid)); UNIT_VTABLE(u)->sigchld_event(u, si.si_pid, si.si_code, si.si_status); } return 0; } -static int manager_start_target(Manager *m, const char *name) { +static int manager_start_target(Manager *m, const char *name, JobMode mode) { int r; - if ((r = manager_add_job_by_name(m, JOB_START, name, JOB_REPLACE, true, NULL)) < 0) + if ((r = manager_add_job_by_name(m, JOB_START, name, mode, true, NULL)) < 0) log_error("Failed to enqueue %s job: %s", name, strerror(-r)); return r; @@ -1629,12 +1801,12 @@ static int manager_process_signal_fd(Manager *m) { case SIGINT: if (m->running_as == MANAGER_INIT) { - manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET); + manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE); break; } /* Run the exit target if there is one, if not, just exit. */ - if (manager_start_target(m, SPECIAL_EXIT_SERVICE) < 0) { + if (manager_start_target(m, SPECIAL_EXIT_SERVICE, JOB_REPLACE) < 0) { m->exit_code = MANAGER_EXIT; return 0; } @@ -1643,14 +1815,14 @@ static int manager_process_signal_fd(Manager *m) { case SIGWINCH: if (m->running_as == MANAGER_INIT) - manager_start_target(m, SPECIAL_KBREQUEST_TARGET); + manager_start_target(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE); /* This is a nop on non-init */ break; case SIGPWR: if (m->running_as == MANAGER_INIT) - manager_start_target(m, SPECIAL_SIGPWR_TARGET); + manager_start_target(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE); /* This is a nop on non-init */ break; @@ -1668,7 +1840,7 @@ static int manager_process_signal_fd(Manager *m) { if (!u || !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u))) { log_info("Loading D-Bus service..."); - manager_start_target(m, SPECIAL_DBUS_SERVICE); + manager_start_target(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE); } break; @@ -1705,9 +1877,26 @@ static int manager_process_signal_fd(Manager *m) { m->exit_code = MANAGER_RELOAD; break; - default: + default: { + static const char * const table[] = { + [0] = SPECIAL_DEFAULT_TARGET, + [1] = SPECIAL_RESCUE_TARGET, + [2] = SPECIAL_EMERGENCY_SERVICE, + [3] = SPECIAL_HALT_TARGET, + [4] = SPECIAL_POWEROFF_TARGET, + [5] = SPECIAL_REBOOT_TARGET + }; + + if ((int) sfsi.ssi_signo >= SIGRTMIN+0 && + (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(table)) { + manager_start_target(m, table[sfsi.ssi_signo - SIGRTMIN], + sfsi.ssi_signo == 2 ? JOB_ISOLATE : JOB_REPLACE); + break; + } + log_info("Got unhandled signal <%s>.", strsignal(sfsi.ssi_signo)); } + } } if (sigchld) @@ -1738,6 +1927,17 @@ static int process_event(Manager *m, struct epoll_event *ev) { 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; + case WATCH_FD: /* Some fd event, to be dispatched to the units */