HALF ECHO LEARN / REFERENCE COLLECTION

v262.2

src/udev/udev-manager.c

57,208 bytesGit mode 100644Download original fileGPL-2.0-or-later

Git blob: 00335b3bfb23c4cb24bb21f04ca008acb74e2e9d

Lines 1001–1542 of 1,542

        if (r < 0)
                return log_warning_errno(r, "Failed to verify type of event storage socket (%i): %m", *fd);
        if (r == 0)
                return log_warning_errno(SYNTHETIC_ERRNO(EINVAL), "Received invalid event storage socket (%i).", *fd);
 
        _cleanup_(sd_device_monitor_unrefp) sd_device_monitor *storage = NULL;
        r = device_monitor_new_full(&storage, MONITOR_GROUP_NONE, *fd);
        if (r < 0)
                return log_warning_errno(r, "Failed to initialize event storage: %m");
        TAKE_FD(*fd);
 
        r = device_monitor_allow_unicast_sender(storage, storage);
        if (r < 0)
                return log_warning_errno(r, "Failed to set trusted sender for event storage: %m");
 
        uint64_t n = 0;
        for (;;) {
                r = fd_wait_for_event(sd_device_monitor_get_fd(storage), POLLIN, 0);
                if (r == -EINTR)
                        continue;
                if (r < 0)
                        return log_warning_errno(r, "Failed to wait for event from event storage: %m");
                if (r == 0)
                        break;
 
                _cleanup_(sd_device_unrefp) sd_device *dev = NULL;
                r = sd_device_monitor_receive(storage, &dev);
                if (r < 0) {
                        log_warning_errno(r, "Failed to receive device from event storage, ignoring: %m");
                        continue;
                }
 
                r = event_queue_insert(manager, dev);
                if (r < 0) {
                        log_device_warning_errno(dev, r, "Failed to insert device into event queue, ignoring: %m");
                        continue;
                }
 
                n++;
        }
 
        if (n > 0)
                (void) manager_create_queue_file(manager);
 
        log_debug("Deserialized %"PRIu64" events.", n);
        return 0;
}
 
static int on_uevent(sd_device_monitor *monitor, sd_device *dev, void *userdata) {
        Manager *manager = ASSERT_PTR(userdata);
        int r;
 
        DEVICE_TRACE_POINT(kernel_uevent_received, dev);
 
        (void) manager_create_queue_file(manager);
 
        device_ensure_usec_initialized(dev, NULL);
 
        r = event_queue_insert(manager, dev);
        if (r < 0) {
                log_device_error_errno(dev, r, "Failed to insert device into event queue: %m");
                return 1;
        }
 
        (void) manager_requeue_locked_events_by_device(manager, dev);
        return 1;
}
 
static int manager_init_device_monitor(Manager *manager, int fd) {
        int r;
 
        assert(manager);
 
        /* This takes passed file descriptor on success. */
 
        if (fd >= 0) {
                if (manager->monitor)
                        return log_warning_errno(SYNTHETIC_ERRNO(EALREADY), "Received multiple netlink socket (%i), ignoring.", fd);
 
                r = sd_is_socket(fd, AF_NETLINK, SOCK_RAW, /* listening= */ -1);
                if (r < 0)
                        return log_warning_errno(r, "Failed to verify socket type of %i, ignoring: %m", fd);
                if (r == 0)
                        return log_warning_errno(SYNTHETIC_ERRNO(EINVAL), "Received invalid netlink socket (%i), ignoring.", fd);
        } else {
                if (manager->monitor)
                        return 0;
        }
 
        r = device_monitor_new_full(&manager->monitor, MONITOR_GROUP_KERNEL, fd);
        if (r < 0)
                return log_error_errno(r, "Failed to initialize device monitor: %m");
 
        return 0;
}
 
static int manager_start_device_monitor(Manager *manager) {
        int r;
 
        assert(manager);
 
        r = manager_init_device_monitor(manager, -EBADF);
        if (r < 0)
                return r;
 
        (void) sd_device_monitor_set_description(manager->monitor, "manager");
 
        r = sd_device_monitor_attach_event(manager->monitor, manager->event);
        if (r < 0)
                return log_error_errno(r, "Failed to attach event to device monitor: %m");
 
        r = sd_device_monitor_start(manager->monitor, on_uevent, manager);
        if (r < 0)
                return log_error_errno(r, "Failed to start device monitor: %m");
 
        r = sd_event_source_set_priority(sd_device_monitor_get_event_source(manager->monitor), EVENT_PRIORITY_DEVICE_MONITOR);
        if (r < 0)
                return log_error_errno(r, "Failed to set priority to device monitor: %m");
 
        return 0;
}
 
static int on_worker_notify(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
        Manager *manager = ASSERT_PTR(userdata);
        int r;
 
        assert(fd >= 0);
 
        _cleanup_(pidref_done) PidRef sender = PIDREF_NULL;
        _cleanup_strv_free_ char **l = NULL;
        r = notify_recv_strv(fd, &l, /* ret_ucred= */ NULL, &sender);
        if (r == -EAGAIN)
                return 0;
        if (r < 0)
                return r;
 
        /* lookup worker who sent the signal */
        Worker *worker = hashmap_get(manager->workers, &sender);
        if (!worker) {
                log_warning("Received notify datagram of unknown process ["PID_FMT"], ignoring.", sender.pid);
                return 0;
        }
 
        if (strv_contains(l, "INOTIFY_WATCH_ADD=1")) {
                assert(worker->event);
 
                r = manager_add_watch(manager, worker->event->dev);
                if (ERRNO_IS_NEG_DEVICE_ABSENT(r))
                        r = 0;
                if (r < 0)
                        log_device_warning_errno(worker->event->dev, r, "Failed to add inotify watch, ignoring: %m");
 
                /* Send the result back to the worker process. */
                r = pidref_sigqueue(&sender, SIGUSR1, r);
                if (r < 0) {
                        log_device_warning_errno(worker->event->dev, r,
                                                 "Failed to send signal to worker process ["PID_FMT"], killing the worker process: %m",
                                                 sender.pid);
 
                        (void) pidref_kill(&sender, SIGTERM);
                        worker->state = WORKER_KILLED;
                }
                return 0;
        }
 
        if (strv_contains(l, "INOTIFY_WATCH_REMOVE=1")) {
                assert(worker->event);
 
                r = manager_remove_watch(manager, worker->event->dev);
                if (r < 0)
                        log_device_warning_errno(worker->event->dev, r, "Failed to remove inotify watch, ignoring: %m");
 
                /* Send the result back to the worker process. */
                r = pidref_sigqueue(&sender, SIGUSR1, r);
                if (r < 0) {
                        log_device_warning_errno(worker->event->dev, r,
                                                 "Failed to send signal to worker process ["PID_FMT"], killing the worker process: %m",
                                                 sender.pid);
 
                        (void) pidref_kill(&sender, SIGTERM);
                        worker->state = WORKER_KILLED;
                }
                return 0;
        }
 
        _cleanup_(event_enter_processedp) Event *event = worker_detach_event(worker);
 
        if (strv_contains(l, "TRY_AGAIN=1")) {
                /* Worker cannot lock the device. */
                r = event_enter_locked(event, strv_find_startswith(l, "WHOLE_DISK="));
                if (r < 0) {
                        (void) device_add_errno(event->dev, r);
                        (void) device_broadcast_on_error(event->dev, manager->monitor);
                } else
                        TAKE_PTR(event);
        }
 
        return 0;
}
 
static int manager_start_worker_notify(Manager *manager) {
        int r;
 
        assert(manager);
        assert(manager->event);
 
        r = notify_socket_prepare(
                        manager->event,
                        EVENT_PRIORITY_WORKER_NOTIFY,
                        on_worker_notify,
                        manager,
                        &manager->worker_notify_socket_path);
        if (r < 0)
                return log_error_errno(r, "Failed to prepare worker notification socket: %m");
 
        return 0;
}
 
static int on_sigterm(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
        Manager *manager = ASSERT_PTR(userdata);
 
        manager_exit(manager);
 
        return 1;
}
 
static int on_sighup(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
        Manager *manager = ASSERT_PTR(userdata);
 
        manager_reload(manager, /* force= */ true);
 
        return 1;
}
 
int manager_create_queue_file(Manager *manager) {
        int r;
 
        assert(manager);
 
        if (manager->queue_file_created)
                return 0;
 
        r = touch("/run/udev/queue");
        if (r < 0)
                return log_warning_errno(r, "Failed to touch /run/udev/queue: %m");
 
        manager->queue_file_created = true;
        return 0;
}
 
static int manager_unlink_queue_file(Manager *manager) {
        assert(manager);
 
        if (manager->events)
                return 0; /* There are queued events. */
 
        if (!set_isempty(manager->synthesize_change_child_event_sources))
                return 0; /* There are child processes that should trigger synthetic events. */
 
        if (!set_isempty(manager->synthesized_events))
                return 0; /* We have triggered synthesized change events. */
 
        /* There are no queued events. Let's remove /run/udev/queue and clean up the idle processes. */
        if (unlink("/run/udev/queue") < 0) {
                if (errno != ENOENT)
                        return log_warning_errno(errno, "Failed to unlink /run/udev/queue: %m");
        } else
                log_debug("No events are queued, removed /run/udev/queue.");
 
        manager->queue_file_created = false;
        return 0;
}
 
static int on_post_exit(Manager *manager) {
        assert(manager);
        assert(manager->exit);
 
        LIST_FOREACH(event, event, manager->events)
                if (event->state == EVENT_RUNNING)
                        return 0; /* There still exist events being processed. */
 
        (void) manager_unlink_queue_file(manager);
 
        if (!hashmap_isempty(manager->workers))
                return 0; /* There still exist running workers. */
 
        (void) manager_serialize_events(manager);
 
        udev_watch_dump();
        return sd_event_exit(manager->event, 0);
}
 
static int on_post(sd_event_source *s, void *userdata) {
        Manager *manager = ASSERT_PTR(userdata);
 
        if (manager->exit)
                return on_post_exit(manager);
 
        if (manager->events) {
                (void) manager_requeue_locked_events(manager);
 
                /* Try to process pending events if idle workers exist. Why is this necessary?
                 * When a worker finished an event and became idle, even if there was a pending event,
                 * the corresponding device might have been locked and the processing of the event
                 * delayed for a while, preventing the worker from processing the event immediately.
                 * Now, the device may be unlocked. Let's try again! */
                (void) event_queue_start(manager);
                return 0;
        }
 
        /* Check whether there are uevents not received yet. If so, we will soon receive them. Do not remove
         * the queue file and do not kill workers. */
        if (sd_event_source_get_pending(sd_device_monitor_get_event_source(manager->monitor)) != 0)
                return 0;
 
        /* We are currently synthesizing change uevents, hence new uevents may come soon. */
        if (!set_isempty(manager->synthesize_change_child_event_sources))
                return 0;
 
        (void) manager_unlink_queue_file(manager);
        (void) manager_reset_kill_workers_timer(manager);
 
        if (!hashmap_isempty(manager->workers))
                return 0; /* There still exist idle workers. */
 
        if (manager->workers_cgroup)
                /* cleanup possible left-over processes in the workers cgroup */
                if (cg_kill_kernel_sigkill(manager->workers_cgroup, /* ret_n_pids_killed= */ NULL) == -EOPNOTSUPP)
                        (void) cg_kill(manager->workers_cgroup, SIGKILL, CGROUP_IGNORE_SELF, /* killed_pids= */ NULL, /* log_kill= */ NULL, /* userdata= */ NULL);
 
        return 0;
}
 
static int manager_setup_signal(
                Manager *manager,
                sd_event *event,
                int signal,
                sd_event_signal_handler_t handler,
                int64_t priority,
                const char *description) {
 
        _cleanup_(sd_event_source_unrefp) sd_event_source *s = NULL;
        int r;
 
        assert(manager);
        assert(event);
 
        r = sd_event_add_signal(event, &s, signal | SD_EVENT_SIGNAL_PROCMASK, handler, manager);
        if (r < 0)
                return r;
 
        r = sd_event_source_set_priority(s, priority);
        if (r < 0)
                return r;
 
        (void) sd_event_source_set_description(s, description);
 
        r = sd_event_source_set_floating(s, true);
        if (r < 0)
                return r;
 
        return 0;
}
 
static int manager_setup_event(Manager *manager) {
        _cleanup_(sd_event_unrefp) sd_event *e = NULL;
        int r;
 
        assert(manager);
 
        r = sd_event_default(&e);
        if (r < 0)
                return log_error_errno(r, "Failed to allocate event loop: %m");
 
        r = manager_setup_signal(manager, e, SIGINT, on_sigterm, EVENT_PRIORITY_SIGTERM, "sigint-event-source");
        if (r < 0)
                return log_error_errno(r, "Failed to create SIGINT event source: %m");
 
        r = manager_setup_signal(manager, e, SIGTERM, on_sigterm, EVENT_PRIORITY_SIGTERM, "sigterm-event-source");
        if (r < 0)
                return log_error_errno(r, "Failed to create SIGTERM event source: %m");
 
        r = manager_setup_signal(manager, e, SIGHUP, on_sighup, EVENT_PRIORITY_SIGHUP, "sighup-event-source");
        if (r < 0)
                return log_error_errno(r, "Failed to create SIGHUP event source: %m");
 
        _cleanup_(sd_event_source_unrefp) sd_event_source *s = NULL;
        r = sd_event_add_post(e, &s, on_post, manager);
        if (r < 0)
                return log_error_errno(r, "Failed to create post event source: %m");
 
        r = sd_event_source_set_priority(s, EVENT_PRIORITY_POST);
        if (r < 0)
                return log_error_errno(r, "Failed to set priority to post event source: %m");
 
        (void) sd_event_source_set_description(s, "post-event-source");
 
        r = sd_event_source_set_floating(s, true);
        if (r < 0)
                return log_error_errno(r, "Failed to set floating to post event source: %m");
 
        /* Eventually, we probably want to do more here on memory pressure, for example, kill idle workers immediately */
        r = sd_event_add_memory_pressure(e, /* ret= */ NULL, /* callback= */ NULL, /* userdata= */ NULL);
        if (r < 0)
                log_full_errno(ERRNO_IS_NOT_SUPPORTED(r) || ERRNO_IS_PRIVILEGE(r) || (r == -EHOSTDOWN) ? LOG_DEBUG : LOG_WARNING, r,
                               "Failed to allocate memory pressure watch, ignoring: %m");
 
        r = sd_event_add_signal(e, /* ret= */ NULL,
                                (SIGRTMIN+18) | SD_EVENT_SIGNAL_PROCMASK, sigrtmin18_handler, /* userdata= */ NULL);
        if (r < 0)
                return log_error_errno(r, "Failed to allocate SIGRTMIN+18 event source, ignoring: %m");
 
        r = sd_event_set_watchdog(e, true);
        if (r < 0)
                return log_error_errno(r, "Failed to create watchdog event source: %m");
 
        manager->event = TAKE_PTR(e);
        return 0;
}
 
static int manager_listen_fds(Manager *manager, int *ret_varlink_fd) {
        _cleanup_strv_free_ char **names = NULL;
        int varlink_fd = -EBADF;
        int r;
 
        assert(manager);
        assert(ret_varlink_fd);
 
        int n = sd_listen_fds_with_names(/* unset_environment= */ true, &names);
        if (n < 0)
                return log_error_errno(n, "Failed to listen on fds: %m");
 
        for (int i = 0; i < n; i++) {
                int fd = SD_LISTEN_FDS_START + i;
 
                if (streq(names[i], "varlink")) {
                        varlink_fd = fd;
                        r = 0;
                } else if (streq(names[i], "systemd-udevd-kernel.socket"))
                        r = manager_init_device_monitor(manager, fd);
                else if (streq(names[i], "inotify"))
                        r = manager_init_inotify(manager, fd);
                else if (streq(names[i], "config-serialization"))
                        r = manager_deserialize_config(manager, &fd);
                else if (streq(names[i], "event-serialization"))
                        r = manager_deserialize_events(manager, &fd);
                else
                        r = log_debug_errno(SYNTHETIC_ERRNO(EINVAL),
                                            "Received unexpected fd (%s), ignoring.", names[i]);
                if (r < 0)
                        close_and_notify_warn(fd, names[i]);
        }
 
        *ret_varlink_fd = varlink_fd;
 
        return 0;
}
 
int manager_main(Manager *manager) {
        _cleanup_close_ int varlink_fd = -EBADF;
        int r;
 
        assert(manager);
 
        _cleanup_free_ char *cgroup = NULL;
        r = cg_pid_get_path(/* pid= */ 0, &cgroup);
        if (r < 0)
                log_debug_errno(r, "Failed to get cgroup, ignoring: %m");
        else if (endswith(cgroup, "/udev")) { /* If we are in a subcgroup /udev/ we assume it was delegated to us */
                log_debug("Running in delegated subcgroup '%s'.", cgroup);
 
                /* Try to create a sibling 'workers' cgroup and spawn all workers inside it, so that we can
                 * use cgroup.kill to atomically clear all workers. */
                _cleanup_free_ char *workers_cgroup = NULL;
                r = path_extract_directory(cgroup, &workers_cgroup);
                if (r < 0)
                        return log_warning_errno(r, "Failed to extract parent of cgroup '%s': %m", cgroup);
 
                if (!path_extend(&workers_cgroup, "workers"))
                        return log_oom();
 
                r = cg_create(workers_cgroup);
                if (r < 0)
                        log_warning_errno(r, "Failed to create workers cgroup '%s', ignoring: %m", workers_cgroup);
                else {
                        log_debug("Running workers in delegated subcgroup '%s'.", workers_cgroup);
                        manager->workers_cgroup = TAKE_PTR(workers_cgroup);
                }
 
        }
 
        r = manager_setup_event(manager);
        if (r < 0)
                return r;
 
        r = manager_listen_fds(manager, &varlink_fd);
        if (r < 0)
                return r;
 
        r = manager_start_varlink_server(manager, TAKE_FD(varlink_fd));
        if (r < 0)
                return r;
 
        r = manager_start_device_monitor(manager);
        if (r < 0)
                return r;
 
        r = manager_start_inotify(manager);
        if (r < 0)
                return r;
 
        r = manager_start_worker_notify(manager);
        if (r < 0)
                return r;
 
        manager->last_usec = now(CLOCK_MONOTONIC);
 
        udev_builtin_init();
 
        r = udev_rules_load(&manager->rules, manager->config.resolve_name_timing, /* extra= */ NULL);
        if (r < 0)
                return log_error_errno(r, "Failed to read udev rules: %m");
 
        r = udev_rules_apply_static_dev_perms(manager->rules);
        if (r < 0)
                log_warning_errno(r, "Failed to apply permissions on static device nodes, ignoring: %m");
 
        _unused_ _cleanup_(notify_on_cleanup) const char *notify_message =
                notify_start(NOTIFY_READY_MESSAGE, NOTIFY_STOPPING_MESSAGE);
 
        /* We will start processing events in the loop below. Before starting processing, let's remove the
         * event serialization fd from the fdstore, to avoid retrieving the serialized events again in future
         * invocations. Otherwise, the serialized events may be processed multiple times. */
        (void) notify_remove_fd_warn("event-serialization");
 
        r = sd_event_loop(manager->event);
        if (r < 0)
                return log_error_errno(r, "Event loop failed: %m");
 
        return 0;
}