udevd.c revision 185abfc3d6b4e8f804a3f7216cd8b0459593af87
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering/*
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering * Copyright (C) 2004-2012 Kay Sievers <kay@vrfy.org>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering * Copyright (C) 2004 Chris Friesen <chris_friesen@sympatico.ca>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering * Copyright (C) 2009 Canonical Ltd.
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering * Copyright (C) 2009 Scott James Remnant <scott@netsplit.com>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering *
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering * This program is free software: you can redistribute it and/or modify
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering * it under the terms of the GNU General Public License as published by
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering * the Free Software Foundation, either version 2 of the License, or
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering * (at your option) any later version.
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering *
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering * This program is distributed in the hope that it will be useful,
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering * but WITHOUT ANY WARRANTY; without even the implied warranty of
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering * GNU General Public License for more details.
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering *
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering * You should have received a copy of the GNU General Public License
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering * along with this program. If not, see <http://www.gnu.org/licenses/>.
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering */
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include <stddef.h>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include <signal.h>
4ad7f2761da661853dcc29d542efb4727abb1101Nick Owens#include <unistd.h>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include <errno.h>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include <stdio.h>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include <stdlib.h>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include <stdbool.h>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include <string.h>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include <fcntl.h>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include <getopt.h>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include <sys/file.h>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include <sys/time.h>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include <sys/prctl.h>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include <sys/socket.h>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include <sys/signalfd.h>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include <sys/epoll.h>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include <sys/mount.h>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include <sys/wait.h>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include <sys/stat.h>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include <sys/ioctl.h>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include <sys/inotify.h>
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include "sd-daemon.h"
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include "rtnl-util.h"
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include "cgroup-util.h"
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include "process-util.h"
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include "dev-setup.h"
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include "fileio.h"
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include "selinux-util.h"
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include "udev.h"
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include "udev-util.h"
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include "formats-util.h"
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering#include "hashmap.h"
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poetteringstatic bool arg_debug = false;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poetteringstatic int arg_daemonize = false;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poetteringstatic int arg_resolve_names = 1;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poetteringstatic unsigned arg_children_max;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poetteringstatic int arg_exec_delay;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poetteringstatic usec_t arg_event_timeout_usec = 180 * USEC_PER_SEC;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poetteringstatic usec_t arg_event_timeout_warn_usec = 180 * USEC_PER_SEC / 3;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poetteringtypedef struct Manager {
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering struct udev *udev;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering Hashmap *workers;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering struct udev_list_node events;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering char *cgroup;
7e8e0422aeb16f2a09a40546c61df753d10029b6Lennart Poettering pid_t pid; /* the process that originally allocated the manager object */
7e8e0422aeb16f2a09a40546c61df753d10029b6Lennart Poettering sigset_t sigmask_orig;
7e8e0422aeb16f2a09a40546c61df753d10029b6Lennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering struct udev_rules *rules;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering struct udev_list properties;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
7e8e0422aeb16f2a09a40546c61df753d10029b6Lennart Poettering struct udev_monitor *monitor;
7e8e0422aeb16f2a09a40546c61df753d10029b6Lennart Poettering struct udev_ctrl *ctrl;
7e8e0422aeb16f2a09a40546c61df753d10029b6Lennart Poettering struct udev_ctrl_connection *ctrl_conn_blocking;
7e8e0422aeb16f2a09a40546c61df753d10029b6Lennart Poettering
7e8e0422aeb16f2a09a40546c61df753d10029b6Lennart Poettering int fd_ep;
7e8e0422aeb16f2a09a40546c61df753d10029b6Lennart Poettering int fd_ctrl;
7e8e0422aeb16f2a09a40546c61df753d10029b6Lennart Poettering int fd_uevent;
7e8e0422aeb16f2a09a40546c61df753d10029b6Lennart Poettering int fd_signal;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering int fd_inotify;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering int fd_worker;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering int worker_watch[2];
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering bool stop_exec_queue:1;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering bool reload:1;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering bool exit:1;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering} Manager;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poetteringenum event_state {
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering EVENT_UNDEF,
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering EVENT_QUEUED,
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering EVENT_RUNNING,
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering};
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poetteringstruct event {
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering struct udev_list_node node;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering Manager *manager;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering struct udev *udev;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering struct udev_device *dev;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering struct udev_device *dev_kernel;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering struct worker *worker;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering enum event_state state;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering unsigned long long int delaying_seqnum;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering unsigned long long int seqnum;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering const char *devpath;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering size_t devpath_len;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering const char *devpath_old;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering dev_t devnum;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering int ifindex;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering bool is_block;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering usec_t start_usec;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering bool warned;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering};
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poetteringstatic inline struct event *node_to_event(struct udev_list_node *node) {
34b9656f0b2890743eee6a746ef08d817abfd5e9Lennart Poettering return container_of(node, struct event, node);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering}
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poetteringstatic void event_queue_cleanup(Manager *manager, enum event_state type);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poetteringenum worker_state {
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering WORKER_UNDEF,
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering WORKER_RUNNING,
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering WORKER_IDLE,
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering WORKER_KILLED,
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering};
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poetteringstruct worker {
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering Manager *manager;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering struct udev_list_node node;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering int refcount;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering pid_t pid;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering struct udev_monitor *monitor;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering enum worker_state state;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering struct event *event;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering};
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
34b9656f0b2890743eee6a746ef08d817abfd5e9Lennart Poettering/* passed from worker to main process */
34b9656f0b2890743eee6a746ef08d817abfd5e9Lennart Poetteringstruct worker_message {
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering};
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poetteringstatic void event_free(struct event *event) {
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering int r;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering if (!event)
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering return;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering udev_list_node_remove(&event->node);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering udev_device_unref(event->dev);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering udev_device_unref(event->dev_kernel);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering if (event->worker)
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering event->worker->event = NULL;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering assert(event->manager);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering if (udev_list_node_is_empty(&event->manager->events)) {
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering /* only clean up the queue from the process that created it */
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering if (event->manager->pid == getpid()) {
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering r = unlink("/run/udev/queue");
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering if (r < 0)
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering log_warning_errno(errno, "could not unlink /run/udev/queue: %m");
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering }
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering }
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering free(event);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering}
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poetteringstatic void worker_free(struct worker *worker) {
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering if (!worker)
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering return;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering assert(worker->manager);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering hashmap_remove(worker->manager->workers, UINT_TO_PTR(worker->pid));
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering udev_monitor_unref(worker->monitor);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering event_free(worker->event);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering free(worker);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering}
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poetteringstatic void manager_workers_free(Manager *manager) {
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering struct worker *worker;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering Iterator i;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering assert(manager);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering HASHMAP_FOREACH(worker, manager->workers, i)
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering worker_free(worker);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering manager->workers = hashmap_free(manager->workers);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering}
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poetteringstatic int worker_new(struct worker **ret, Manager *manager, struct udev_monitor *worker_monitor, pid_t pid) {
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering _cleanup_free_ struct worker *worker = NULL;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering int r;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering assert(ret);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering assert(manager);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering assert(worker_monitor);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering assert(pid > 1);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering worker = new0(struct worker, 1);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering if (!worker)
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering return -ENOMEM;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering worker->refcount = 1;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering worker->manager = manager;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering /* close monitor, but keep address around */
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering udev_monitor_disconnect(worker_monitor);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering worker->monitor = udev_monitor_ref(worker_monitor);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering worker->pid = pid;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering r = hashmap_ensure_allocated(&manager->workers, NULL);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering if (r < 0)
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering return r;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering r = hashmap_put(manager->workers, UINT_TO_PTR(pid), worker);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering if (r < 0)
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering return r;
34b9656f0b2890743eee6a746ef08d817abfd5e9Lennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering *ret = worker;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering worker = NULL;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering return 0;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering}
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poetteringstatic int on_event_timeout(sd_event_source *s, uint64_t usec, void *userdata) {
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering struct event *event = userdata;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering assert(event);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering assert(event->worker);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering kill_and_sigcont(event->worker->pid, SIGKILL);
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering event->worker->state = WORKER_KILLED;
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering
faa133f3aa7a18f26563dc5d6b95898cb315c37aLennart Poettering log_error("seq %llu '%s' killed", udev_device_get_seqnum(event->dev), event->devpath);
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering return 1;
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering}
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poetteringstatic int on_event_timeout_warning(sd_event_source *s, uint64_t usec, void *userdata) {
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering struct event *event = userdata;
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering assert(event);
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering log_warning("seq %llu '%s' is taking a long time", udev_device_get_seqnum(event->dev), event->devpath);
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering return 1;
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering}
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poetteringstatic void worker_attach_event(struct worker *worker, struct event *event) {
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering assert(worker);
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering assert(event);
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering assert(!event->worker);
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering assert(!worker->event);
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering worker->state = WORKER_RUNNING;
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering worker->event = event;
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering event->state = EVENT_RUNNING;
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering event->start_usec = now(CLOCK_MONOTONIC);
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering event->warned = false;
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering event->worker = worker;
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering}
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poetteringstatic void manager_free(Manager *manager) {
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering if (!manager)
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering return;
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering udev_builtin_exit(manager->udev);
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering
b914e211f3a40f507b3cdc572838ec7f3fd5e4cfLennart Poettering udev_unref(manager->udev);
manager_workers_free(manager);
event_queue_cleanup(manager, EVENT_UNDEF);
udev_monitor_unref(manager->monitor);
udev_ctrl_unref(manager->ctrl);
udev_ctrl_connection_unref(manager->ctrl_conn_blocking);
udev_list_cleanup(&manager->properties);
udev_rules_unref(manager->rules);
free(manager->cgroup);
safe_close(manager->fd_ep);
safe_close(manager->fd_signal);
safe_close(manager->fd_inotify);
safe_close_pair(manager->worker_watch);
free(manager);
}
DEFINE_TRIVIAL_CLEANUP_FUNC(Manager*, manager_free);
static int worker_send_message(int fd) {
struct worker_message message = {};
return loop_write(fd, &message, sizeof(message), false);
}
static void worker_spawn(Manager *manager, struct event *event) {
struct udev *udev = event->udev;
_cleanup_udev_monitor_unref_ struct udev_monitor *worker_monitor = NULL;
pid_t pid;
/* listen for new events */
worker_monitor = udev_monitor_new_from_netlink(udev, NULL);
if (worker_monitor == NULL)
return;
/* allow the main daemon netlink address to send devices to the worker */
udev_monitor_allow_unicast_sender(worker_monitor, manager->monitor);
udev_monitor_enable_receiving(worker_monitor);
pid = fork();
switch (pid) {
case 0: {
struct udev_device *dev = NULL;
_cleanup_rtnl_unref_ sd_rtnl *rtnl = NULL;
int fd_monitor;
_cleanup_close_ int fd_signal = -1, fd_ep = -1;
struct epoll_event ep_signal = { .events = EPOLLIN };
struct epoll_event ep_monitor = { .events = EPOLLIN };
sigset_t mask;
int r = 0;
/* take initial device from queue */
dev = event->dev;
event->dev = NULL;
manager_workers_free(manager);
event_queue_cleanup(manager, EVENT_UNDEF);
manager->monitor = udev_monitor_unref(manager->monitor);
manager->ctrl_conn_blocking = udev_ctrl_connection_unref(manager->ctrl_conn_blocking);
manager->ctrl = udev_ctrl_unref(manager->ctrl);
manager->fd_ep = safe_close(manager->fd_ep);
manager->fd_signal = safe_close(manager->fd_signal);
manager->worker_watch[READ_END] = safe_close(manager->worker_watch[READ_END]);
sigfillset(&mask);
fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
if (fd_signal < 0) {
r = log_error_errno(errno, "error creating signalfd %m");
goto out;
}
ep_signal.data.fd = fd_signal;
fd_monitor = udev_monitor_get_fd(worker_monitor);
ep_monitor.data.fd = fd_monitor;
fd_ep = epoll_create1(EPOLL_CLOEXEC);
if (fd_ep < 0) {
r = log_error_errno(errno, "error creating epoll fd: %m");
goto out;
}
if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_monitor, &ep_monitor) < 0) {
r = log_error_errno(errno, "fail to add fds to epoll: %m");
goto out;
}
/* request TERM signal if parent exits */
prctl(PR_SET_PDEATHSIG, SIGTERM);
/* reset OOM score, we only protect the main daemon */
write_string_file("/proc/self/oom_score_adj", "0");
for (;;) {
struct udev_event *udev_event;
int fd_lock = -1;
log_debug("seq %llu running", udev_device_get_seqnum(dev));
udev_event = udev_event_new(dev);
if (udev_event == NULL) {
r = -ENOMEM;
goto out;
}
/* needed for SIGCHLD/SIGTERM in spawn() */
udev_event->fd_signal = fd_signal;
if (arg_exec_delay > 0)
udev_event->exec_delay = arg_exec_delay;
/*
* Take a shared lock on the device node; this establishes
* a concept of device "ownership" to serialize device
* access. External processes holding an exclusive lock will
* cause udev to skip the event handling; in the case udev
* acquired the lock, the external process can block until
* udev has finished its event handling.
*/
if (!streq_ptr(udev_device_get_action(dev), "remove") &&
streq_ptr("block", udev_device_get_subsystem(dev)) &&
!startswith(udev_device_get_sysname(dev), "dm-") &&
!startswith(udev_device_get_sysname(dev), "md")) {
struct udev_device *d = dev;
if (streq_ptr("partition", udev_device_get_devtype(d)))
d = udev_device_get_parent(d);
if (d) {
fd_lock = open(udev_device_get_devnode(d), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
if (fd_lock >= 0 && flock(fd_lock, LOCK_SH|LOCK_NB) < 0) {
log_debug_errno(errno, "Unable to flock(%s), skipping event handling: %m", udev_device_get_devnode(d));
fd_lock = safe_close(fd_lock);
r = -EAGAIN;
goto skip;
}
}
}
/* needed for renaming netifs */
udev_event->rtnl = rtnl;
/* apply rules, create node, symlinks */
udev_event_execute_rules(udev_event,
arg_event_timeout_usec, arg_event_timeout_warn_usec,
&manager->properties,
manager->rules,
&manager->sigmask_orig);
udev_event_execute_run(udev_event,
arg_event_timeout_usec, arg_event_timeout_warn_usec,
&manager->sigmask_orig);
if (udev_event->rtnl)
/* in case rtnl was initialized */
rtnl = sd_rtnl_ref(udev_event->rtnl);
/* apply/restore inotify watch */
if (udev_event->inotify_watch) {
udev_watch_begin(udev, dev);
udev_device_update_db(dev);
}
safe_close(fd_lock);
/* send processed event back to libudev listeners */
udev_monitor_send_device(worker_monitor, NULL, dev);
skip:
log_debug("seq %llu processed", udev_device_get_seqnum(dev));
/* send udevd the result of the event execution */
r = worker_send_message(manager->worker_watch[WRITE_END]);
if (r < 0)
log_error_errno(r, "failed to send result of seq %llu to main daemon: %m",
udev_device_get_seqnum(dev));
udev_device_unref(dev);
dev = NULL;
if (udev_event->sigterm) {
udev_event_unref(udev_event);
goto out;
}
udev_event_unref(udev_event);
/* wait for more device messages from main udevd, or term signal */
while (dev == NULL) {
struct epoll_event ev[4];
int fdcount;
int i;
fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), -1);
if (fdcount < 0) {
if (errno == EINTR)
continue;
r = log_error_errno(errno, "failed to poll: %m");
goto out;
}
for (i = 0; i < fdcount; i++) {
if (ev[i].data.fd == fd_monitor && ev[i].events & EPOLLIN) {
dev = udev_monitor_receive_device(worker_monitor);
break;
} else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN) {
struct signalfd_siginfo fdsi;
ssize_t size;
size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
if (size != sizeof(struct signalfd_siginfo))
continue;
switch (fdsi.ssi_signo) {
case SIGTERM:
goto out;
}
}
}
}
}
out:
udev_device_unref(dev);
manager_free(manager);
log_close();
_exit(r < 0 ? EXIT_FAILURE : EXIT_SUCCESS);
}
case -1:
event->state = EVENT_QUEUED;
log_error_errno(errno, "fork of child failed: %m");
break;
default:
{
struct worker *worker;
int r;
r = worker_new(&worker, manager, worker_monitor, pid);
if (r < 0)
return;
worker_attach_event(worker, event);
log_debug("seq %llu forked new worker ["PID_FMT"]", udev_device_get_seqnum(event->dev), pid);
break;
}
}
}
static void event_run(Manager *manager, struct event *event) {
struct worker *worker;
Iterator i;
assert(manager);
assert(event);
HASHMAP_FOREACH(worker, manager->workers, i) {
ssize_t count;
if (worker->state != WORKER_IDLE)
continue;
count = udev_monitor_send_device(manager->monitor, worker->monitor, event->dev);
if (count < 0) {
log_error_errno(errno, "worker ["PID_FMT"] did not accept message %zi (%m), kill it",
worker->pid, count);
kill(worker->pid, SIGKILL);
worker->state = WORKER_KILLED;
continue;
}
worker_attach_event(worker, event);
return;
}
if (hashmap_size(manager->workers) >= arg_children_max) {
if (arg_children_max > 1)
log_debug("maximum number (%i) of children reached", hashmap_size(manager->workers));
return;
}
/* start new worker and pass initial device */
worker_spawn(manager, event);
}
static int event_queue_insert(Manager *manager, struct udev_device *dev) {
struct event *event;
int r;
assert(manager);
assert(dev);
/* only one process can add events to the queue */
if (manager->pid == 0)
manager->pid = getpid();
assert(manager->pid == getpid());
event = new0(struct event, 1);
if (!event)
return -ENOMEM;
event->udev = udev_device_get_udev(dev);
event->manager = manager;
event->dev = dev;
event->dev_kernel = udev_device_shallow_clone(dev);
udev_device_copy_properties(event->dev_kernel, dev);
event->seqnum = udev_device_get_seqnum(dev);
event->devpath = udev_device_get_devpath(dev);
event->devpath_len = strlen(event->devpath);
event->devpath_old = udev_device_get_devpath_old(dev);
event->devnum = udev_device_get_devnum(dev);
event->is_block = streq("block", udev_device_get_subsystem(dev));
event->ifindex = udev_device_get_ifindex(dev);
log_debug("seq %llu queued, '%s' '%s'", udev_device_get_seqnum(dev),
udev_device_get_action(dev), udev_device_get_subsystem(dev));
event->state = EVENT_QUEUED;
if (udev_list_node_is_empty(&manager->events)) {
r = touch("/run/udev/queue");
if (r < 0)
log_warning_errno(r, "could not touch /run/udev/queue: %m");
}
udev_list_node_append(&event->node, &manager->events);
return 0;
}
static void manager_kill_workers(Manager *manager) {
struct worker *worker;
Iterator i;
assert(manager);
HASHMAP_FOREACH(worker, manager->workers, i) {
if (worker->state == WORKER_KILLED)
continue;
worker->state = WORKER_KILLED;
kill(worker->pid, SIGTERM);
}
}
/* lookup event for identical, parent, child device */
static bool is_devpath_busy(Manager *manager, struct event *event) {
struct udev_list_node *loop;
size_t common;
/* check if queue contains events we depend on */
udev_list_node_foreach(loop, &manager->events) {
struct event *loop_event = node_to_event(loop);
/* we already found a later event, earlier can not block us, no need to check again */
if (loop_event->seqnum < event->delaying_seqnum)
continue;
/* event we checked earlier still exists, no need to check again */
if (loop_event->seqnum == event->delaying_seqnum)
return true;
/* found ourself, no later event can block us */
if (loop_event->seqnum >= event->seqnum)
break;
/* check major/minor */
if (major(event->devnum) != 0 && event->devnum == loop_event->devnum && event->is_block == loop_event->is_block)
return true;
/* check network device ifindex */
if (event->ifindex != 0 && event->ifindex == loop_event->ifindex)
return true;
/* check our old name */
if (event->devpath_old != NULL && streq(loop_event->devpath, event->devpath_old)) {
event->delaying_seqnum = loop_event->seqnum;
return true;
}
/* compare devpath */
common = MIN(loop_event->devpath_len, event->devpath_len);
/* one devpath is contained in the other? */
if (memcmp(loop_event->devpath, event->devpath, common) != 0)
continue;
/* identical device event found */
if (loop_event->devpath_len == event->devpath_len) {
/* devices names might have changed/swapped in the meantime */
if (major(event->devnum) != 0 && (event->devnum != loop_event->devnum || event->is_block != loop_event->is_block))
continue;
if (event->ifindex != 0 && event->ifindex != loop_event->ifindex)
continue;
event->delaying_seqnum = loop_event->seqnum;
return true;
}
/* parent device event found */
if (event->devpath[common] == '/') {
event->delaying_seqnum = loop_event->seqnum;
return true;
}
/* child device event found */
if (loop_event->devpath[common] == '/') {
event->delaying_seqnum = loop_event->seqnum;
return true;
}
/* no matching device */
continue;
}
return false;
}
static void event_queue_start(Manager *manager) {
struct udev_list_node *loop;
assert(manager);
udev_list_node_foreach(loop, &manager->events) {
struct event *event = node_to_event(loop);
if (event->state != EVENT_QUEUED)
continue;
/* do not start event if parent or child event is still running */
if (is_devpath_busy(manager, event))
continue;
event_run(manager, event);
}
}
static void event_queue_cleanup(Manager *manager, enum event_state match_type) {
struct udev_list_node *loop, *tmp;
udev_list_node_foreach_safe(loop, tmp, &manager->events) {
struct event *event = node_to_event(loop);
if (match_type != EVENT_UNDEF && match_type != event->state)
continue;
event_free(event);
}
}
static int on_worker(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
Manager *manager = userdata;
assert(manager);
for (;;) {
struct worker_message msg;
struct iovec iovec = {
.iov_base = &msg,
.iov_len = sizeof(msg),
};
union {
struct cmsghdr cmsghdr;
uint8_t buf[CMSG_SPACE(sizeof(struct ucred))];
} control = {};
struct msghdr msghdr = {
.msg_iov = &iovec,
.msg_iovlen = 1,
.msg_control = &control,
.msg_controllen = sizeof(control),
};
struct cmsghdr *cmsg;
ssize_t size;
struct ucred *ucred = NULL;
struct worker *worker;
size = recvmsg(fd, &msghdr, MSG_DONTWAIT);
if (size < 0) {
if (errno == EINTR)
continue;
else if (errno == EAGAIN)
/* nothing more to read */
break;
return log_error_errno(errno, "failed to receive message: %m");
} else if (size != sizeof(struct worker_message)) {
log_warning_errno(EIO, "ignoring worker message with invalid size %zi bytes", size);
continue;
}
for (cmsg = CMSG_FIRSTHDR(&msghdr); cmsg; cmsg = CMSG_NXTHDR(&msghdr, cmsg)) {
if (cmsg->cmsg_level == SOL_SOCKET &&
cmsg->cmsg_type == SCM_CREDENTIALS &&
cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred)))
ucred = (struct ucred*) CMSG_DATA(cmsg);
}
if (!ucred || ucred->pid <= 0) {
log_warning_errno(EIO, "ignoring worker message without valid PID");
continue;
}
/* lookup worker who sent the signal */
worker = hashmap_get(manager->workers, UINT_TO_PTR(ucred->pid));
if (!worker) {
log_debug("worker ["PID_FMT"] returned, but is no longer tracked", ucred->pid);
continue;
}
if (worker->state != WORKER_KILLED)
worker->state = WORKER_IDLE;
/* worker returned */
event_free(worker->event);
}
return 1;
}
static int on_uevent(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
Manager *manager = userdata;
struct udev_device *dev;
int r;
assert(manager);
dev = udev_monitor_receive_device(manager->monitor);
if (dev) {
udev_device_ensure_usec_initialized(dev, NULL);
r = event_queue_insert(manager, dev);
if (r < 0)
udev_device_unref(dev);
}
return 1;
}
/* receive the udevd message from userspace */
static int on_ctrl_msg(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
Manager *manager = userdata;
_cleanup_udev_ctrl_connection_unref_ struct udev_ctrl_connection *ctrl_conn = NULL;
_cleanup_udev_ctrl_msg_unref_ struct udev_ctrl_msg *ctrl_msg = NULL;
const char *str;
int i;
assert(manager);
ctrl_conn = udev_ctrl_get_connection(manager->ctrl);
if (!ctrl_conn)
return 1;
ctrl_msg = udev_ctrl_receive_msg(ctrl_conn);
if (!ctrl_msg)
return 1;
i = udev_ctrl_get_set_log_level(ctrl_msg);
if (i >= 0) {
log_debug("udevd message (SET_LOG_LEVEL) received, log_priority=%i", i);
log_set_max_level(i);
manager_kill_workers(manager);
}
if (udev_ctrl_get_stop_exec_queue(ctrl_msg) > 0) {
log_debug("udevd message (STOP_EXEC_QUEUE) received");
manager->stop_exec_queue = true;
}
if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) {
log_debug("udevd message (START_EXEC_QUEUE) received");
manager->stop_exec_queue = false;
}
if (udev_ctrl_get_reload(ctrl_msg) > 0) {
log_debug("udevd message (RELOAD) received");
manager->reload = true;
}
str = udev_ctrl_get_set_env(ctrl_msg);
if (str != NULL) {
_cleanup_free_ char *key = NULL;
key = strdup(str);
if (key) {
char *val;
val = strchr(key, '=');
if (val != NULL) {
val[0] = '\0';
val = &val[1];
if (val[0] == '\0') {
log_debug("udevd message (ENV) received, unset '%s'", key);
udev_list_entry_add(&manager->properties, key, NULL);
} else {
log_debug("udevd message (ENV) received, set '%s=%s'", key, val);
udev_list_entry_add(&manager->properties, key, val);
}
} else
log_error("wrong key format '%s'", key);
}
manager_kill_workers(manager);
}
i = udev_ctrl_get_set_children_max(ctrl_msg);
if (i >= 0) {
log_debug("udevd message (SET_MAX_CHILDREN) received, children_max=%i", i);
arg_children_max = i;
}
if (udev_ctrl_get_ping(ctrl_msg) > 0)
log_debug("udevd message (SYNC) received");
if (udev_ctrl_get_exit(ctrl_msg) > 0) {
log_debug("udevd message (EXIT) received");
manager->exit = true;
/* keep reference to block the client until we exit
TODO: deal with several blocking exit requests */
manager->ctrl_conn_blocking = udev_ctrl_connection_ref(ctrl_conn);
}
return 1;
}
static int synthesize_change(struct udev_device *dev) {
char filename[UTIL_PATH_SIZE];
int r;
if (streq_ptr("block", udev_device_get_subsystem(dev)) &&
streq_ptr("disk", udev_device_get_devtype(dev)) &&
!startswith(udev_device_get_sysname(dev), "dm-")) {
bool part_table_read = false;
bool has_partitions = false;
int fd;
struct udev *udev = udev_device_get_udev(dev);
_cleanup_udev_enumerate_unref_ struct udev_enumerate *e = NULL;
struct udev_list_entry *item;
/*
* Try to re-read the partition table. This only succeeds if
* none of the devices is busy. The kernel returns 0 if no
* partition table is found, and we will not get an event for
* the disk.
*/
fd = open(udev_device_get_devnode(dev), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
if (fd >= 0) {
r = flock(fd, LOCK_EX|LOCK_NB);
if (r >= 0)
r = ioctl(fd, BLKRRPART, 0);
close(fd);
if (r >= 0)
part_table_read = true;
}
/* search for partitions */
e = udev_enumerate_new(udev);
if (!e)
return -ENOMEM;
r = udev_enumerate_add_match_parent(e, dev);
if (r < 0)
return r;
r = udev_enumerate_add_match_subsystem(e, "block");
if (r < 0)
return r;
r = udev_enumerate_scan_devices(e);
if (r < 0)
return r;
udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
_cleanup_udev_device_unref_ struct udev_device *d = NULL;
d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
if (!d)
continue;
if (!streq_ptr("partition", udev_device_get_devtype(d)))
continue;
has_partitions = true;
break;
}
/*
* We have partitions and re-read the table, the kernel already sent
* out a "change" event for the disk, and "remove/add" for all
* partitions.
*/
if (part_table_read && has_partitions)
return 0;
/*
* We have partitions but re-reading the partition table did not
* work, synthesize "change" for the disk and all partitions.
*/
log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
write_string_file(filename, "change");
udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
_cleanup_udev_device_unref_ struct udev_device *d = NULL;
d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
if (!d)
continue;
if (!streq_ptr("partition", udev_device_get_devtype(d)))
continue;
log_debug("device %s closed, synthesising partition '%s' 'change'",
udev_device_get_devnode(dev), udev_device_get_devnode(d));
strscpyl(filename, sizeof(filename), udev_device_get_syspath(d), "/uevent", NULL);
write_string_file(filename, "change");
}
return 0;
}
log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
write_string_file(filename, "change");
return 0;
}
static int on_inotify(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
Manager *manager = userdata;
union inotify_event_buffer buffer;
struct inotify_event *e;
ssize_t l;
assert(manager);
l = read(fd, &buffer, sizeof(buffer));
if (l < 0) {
if (errno == EAGAIN || errno == EINTR)
return 1;
return log_error_errno(errno, "Failed to read inotify fd: %m");
}
FOREACH_INOTIFY_EVENT(e, buffer, l) {
_cleanup_udev_device_unref_ struct udev_device *dev = NULL;
dev = udev_watch_lookup(manager->udev, e->wd);
if (!dev)
continue;
log_debug("inotify event: %x for %s", e->mask, udev_device_get_devnode(dev));
if (e->mask & IN_CLOSE_WRITE) {
synthesize_change(dev);
/* settle might be waiting on us to determine the queue
* state. If we just handled an inotify event, we might have
* generated a "change" event, but we won't have queued up
* the resultant uevent yet. Do that.
*/
on_uevent(NULL, -1, 0, manager);
} else if (e->mask & IN_IGNORED)
udev_watch_end(manager->udev, dev);
}
return 1;
}
static int on_sigterm(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
Manager *manager = userdata;
assert(manager);
manager->exit = true;
return 1;
}
static int on_sighup(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
Manager *manager = userdata;
assert(manager);
manager->reload = true;
return 1;
}
static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
Manager *manager = userdata;
assert(manager);
for (;;) {
pid_t pid;
int status;
struct worker *worker;
pid = waitpid(-1, &status, WNOHANG);
if (pid <= 0)
break;
worker = hashmap_get(manager->workers, UINT_TO_PTR(pid));
if (!worker) {
log_warning("worker ["PID_FMT"] is unknown, ignoring", pid);
continue;
}
if (WIFEXITED(status)) {
if (WEXITSTATUS(status) == 0)
log_debug("worker ["PID_FMT"] exited", pid);
else
log_warning("worker ["PID_FMT"] exited with return code %i", pid, WEXITSTATUS(status));
} else if (WIFSIGNALED(status)) {
log_warning("worker ["PID_FMT"] terminated by signal %i (%s)", pid, WTERMSIG(status), strsignal(WTERMSIG(status)));
} else if (WIFSTOPPED(status)) {
log_info("worker ["PID_FMT"] stopped", pid);
continue;
} else if (WIFCONTINUED(status)) {
log_info("worker ["PID_FMT"] continued", pid);
continue;
} else
log_warning("worker ["PID_FMT"] exit with status 0x%04x", pid, status);
if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
if (worker->event) {
log_error("worker ["PID_FMT"] failed while handling '%s'", pid, worker->event->devpath);
/* delete state from disk */
udev_device_delete_db(worker->event->dev);
udev_device_tag_index(worker->event->dev, NULL, false);
/* forward kernel event without amending it */
udev_monitor_send_device(manager->monitor, NULL, worker->event->dev_kernel);
}
}
worker_free(worker);
}
return 1;
}
static int systemd_fds(int *rctrl, int *rnetlink) {
int ctrl = -1, netlink = -1;
int fd, n;
n = sd_listen_fds(true);
if (n <= 0)
return -1;
for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1)) {
if (ctrl >= 0)
return -1;
ctrl = fd;
continue;
}
if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1)) {
if (netlink >= 0)
return -1;
netlink = fd;
continue;
}
return -1;
}
if (ctrl < 0 || netlink < 0)
return -1;
log_debug("ctrl=%i netlink=%i", ctrl, netlink);
*rctrl = ctrl;
*rnetlink = netlink;
return 0;
}
/*
* read the kernel command line, in case we need to get into debug mode
* udev.log-priority=<level> syslog priority
* udev.children-max=<number of workers> events are fully serialized if set to 1
* udev.exec-delay=<number of seconds> delay execution of every executed program
* udev.event-timeout=<number of seconds> seconds to wait before terminating an event
*/
static int parse_proc_cmdline_item(const char *key, const char *value) {
int r;
assert(key);
if (!value)
return 0;
if (startswith(key, "rd."))
key += strlen("rd.");
if (startswith(key, "udev."))
key += strlen("udev.");
else
return 0;
if (streq(key, "log-priority")) {
int prio;
prio = util_log_priority(value);
log_set_max_level(prio);
} else if (streq(key, "children-max")) {
r = safe_atou(value, &arg_children_max);
if (r < 0)
log_warning("invalid udev.children-max ignored: %s", value);
} else if (streq(key, "exec-delay")) {
r = safe_atoi(value, &arg_exec_delay);
if (r < 0)
log_warning("invalid udev.exec-delay ignored: %s", value);
} else if (streq(key, "event-timeout")) {
r = safe_atou64(value, &arg_event_timeout_usec);
if (r < 0)
log_warning("invalid udev.event-timeout ignored: %s", value);
else {
arg_event_timeout_usec *= USEC_PER_SEC;
arg_event_timeout_warn_usec = (arg_event_timeout_usec / 3) ? : 1;
}
}
return 0;
}
static void help(void) {
printf("%s [OPTIONS...]\n\n"
"Manages devices.\n\n"
" -h --help Print this message\n"
" --version Print version of the program\n"
" --daemon Detach and run in the background\n"
" --debug Enable debug output\n"
" --children-max=INT Set maximum number of workers\n"
" --exec-delay=SECONDS Seconds to wait before executing RUN=\n"
" --event-timeout=SECONDS Seconds to wait before terminating an event\n"
" --resolve-names=early|late|never\n"
" When to resolve users and groups\n"
, program_invocation_short_name);
}
static int parse_argv(int argc, char *argv[]) {
static const struct option options[] = {
{ "daemon", no_argument, NULL, 'd' },
{ "debug", no_argument, NULL, 'D' },
{ "children-max", required_argument, NULL, 'c' },
{ "exec-delay", required_argument, NULL, 'e' },
{ "event-timeout", required_argument, NULL, 't' },
{ "resolve-names", required_argument, NULL, 'N' },
{ "help", no_argument, NULL, 'h' },
{ "version", no_argument, NULL, 'V' },
{}
};
int c;
assert(argc >= 0);
assert(argv);
while ((c = getopt_long(argc, argv, "c:de:DtN:hV", options, NULL)) >= 0) {
int r;
switch (c) {
case 'd':
arg_daemonize = true;
break;
case 'c':
r = safe_atou(optarg, &arg_children_max);
if (r < 0)
log_warning("Invalid --children-max ignored: %s", optarg);
break;
case 'e':
r = safe_atoi(optarg, &arg_exec_delay);
if (r < 0)
log_warning("Invalid --exec-delay ignored: %s", optarg);
break;
case 't':
r = safe_atou64(optarg, &arg_event_timeout_usec);
if (r < 0)
log_warning("Invalid --event-timeout ignored: %s", optarg);
else {
arg_event_timeout_usec *= USEC_PER_SEC;
arg_event_timeout_warn_usec = (arg_event_timeout_usec / 3) ? : 1;
}
break;
case 'D':
arg_debug = true;
break;
case 'N':
if (streq(optarg, "early")) {
arg_resolve_names = 1;
} else if (streq(optarg, "late")) {
arg_resolve_names = 0;
} else if (streq(optarg, "never")) {
arg_resolve_names = -1;
} else {
log_error("resolve-names must be early, late or never");
return 0;
}
break;
case 'h':
help();
return 0;
case 'V':
printf("%s\n", VERSION);
return 0;
case '?':
return -EINVAL;
default:
assert_not_reached("Unhandled option");
}
}
return 1;
}
static int manager_new(Manager **ret) {
_cleanup_(manager_freep) Manager *manager = NULL;
struct epoll_event ep_ctrl = { .events = EPOLLIN };
struct epoll_event ep_inotify = { .events = EPOLLIN };
struct epoll_event ep_signal = { .events = EPOLLIN };
struct epoll_event ep_netlink = { .events = EPOLLIN };
struct epoll_event ep_worker = { .events = EPOLLIN };
sigset_t mask;
int r, one = 1;
assert(ret);
manager = new0(Manager, 1);
if (!manager)
return log_oom();
manager->fd_ep = -1;
manager->fd_ctrl = -1;
manager->fd_uevent = -1;
manager->fd_signal = -1;
manager->fd_inotify = -1;
manager->worker_watch[WRITE_END] = -1;
manager->worker_watch[READ_END] = -1;
manager->udev = udev_new();
if (!manager->udev)
return log_error_errno(errno, "could not allocate udev context: %m");
udev_builtin_init(manager->udev);
manager->rules = udev_rules_new(manager->udev, arg_resolve_names);
if (!manager->rules)
return log_error_errno(ENOMEM, "error reading rules");
udev_list_node_init(&manager->events);
udev_list_init(manager->udev, &manager->properties, true);
r = systemd_fds(&manager->fd_ctrl, &manager->fd_uevent);
if (r >= 0) {
/* get control and netlink socket from systemd */
manager->ctrl = udev_ctrl_new_from_fd(manager->udev, manager->fd_ctrl);
if (!manager->ctrl)
return log_error_errno(EINVAL, "error taking over udev control socket");
manager->monitor = udev_monitor_new_from_netlink_fd(manager->udev, "kernel", manager->fd_uevent);
if (!manager->monitor)
return log_error_errno(EINVAL, "error taking over netlink socket");
/* get our own cgroup, we regularly kill everything udev has left behind */
r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &manager->cgroup);
if (r < 0)
log_warning_errno(r, "failed to get cgroup: %m");
} else {
/* open control and netlink socket */
manager->ctrl = udev_ctrl_new(manager->udev);
if (!manager->ctrl)
return log_error_errno(EINVAL, "error initializing udev control socket");
manager->fd_ctrl = udev_ctrl_get_fd(manager->ctrl);
manager->monitor = udev_monitor_new_from_netlink(manager->udev, "kernel");
if (!manager->monitor)
return log_error_errno(EINVAL, "error initializing netlink socket");
manager->fd_uevent = udev_monitor_get_fd(manager->monitor);
(void) udev_monitor_set_receive_buffer_size(manager->monitor, 128 * 1024 * 1024);
}
r = udev_monitor_enable_receiving(manager->monitor);
if (r < 0)
return log_error_errno(EINVAL, "error binding netlink socket");
r = udev_ctrl_enable_receiving(manager->ctrl);
if (r < 0)
return log_error_errno(EINVAL, "error binding udev control socket");
/* unnamed socket from workers to the main daemon */
r = socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, manager->worker_watch);
if (r < 0)
return log_error_errno(errno, "error creating socketpair: %m");
manager->fd_worker = manager->worker_watch[READ_END];
r = setsockopt(manager->fd_worker, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
if (r < 0)
return log_error_errno(errno, "could not enable SO_PASSCRED: %m");
manager->fd_inotify = udev_watch_init(manager->udev);
if (manager->fd_inotify < 0)
return log_error_errno(ENOMEM, "error initializing inotify");
udev_watch_restore(manager->udev);
/* block and listen to all signals on signalfd */
sigfillset(&mask);
sigprocmask(SIG_SETMASK, &mask, &manager->sigmask_orig);
manager->fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
if (manager->fd_signal < 0)
return log_error_errno(errno, "error creating signalfd");
ep_ctrl.data.fd = manager->fd_ctrl;
ep_inotify.data.fd = manager->fd_inotify;
ep_signal.data.fd = manager->fd_signal;
ep_netlink.data.fd = manager->fd_uevent;
ep_worker.data.fd = manager->fd_worker;
manager->fd_ep = epoll_create1(EPOLL_CLOEXEC);
if (manager->fd_ep < 0)
return log_error_errno(errno, "error creating epoll fd: %m");
if (epoll_ctl(manager->fd_ep, EPOLL_CTL_ADD, manager->fd_ctrl, &ep_ctrl) < 0 ||
epoll_ctl(manager->fd_ep, EPOLL_CTL_ADD, manager->fd_inotify, &ep_inotify) < 0 ||
epoll_ctl(manager->fd_ep, EPOLL_CTL_ADD, manager->fd_signal, &ep_signal) < 0 ||
epoll_ctl(manager->fd_ep, EPOLL_CTL_ADD, manager->fd_uevent, &ep_netlink) < 0 ||
epoll_ctl(manager->fd_ep, EPOLL_CTL_ADD, manager->fd_worker, &ep_worker) < 0)
return log_error_errno(errno, "fail to add fds to epoll: %m");
*ret = manager;
manager = NULL;
return 1;
}
int main(int argc, char *argv[]) {
_cleanup_(manager_freep) Manager *manager = NULL;
int r;
log_set_target(LOG_TARGET_AUTO);
log_parse_environment();
log_open();
r = parse_argv(argc, argv);
if (r <= 0)
goto exit;
r = parse_proc_cmdline(parse_proc_cmdline_item);
if (r < 0)
log_warning_errno(r, "failed to parse kernel command line, ignoring: %m");
if (arg_debug)
log_set_max_level(LOG_DEBUG);
if (getuid() != 0) {
r = log_error_errno(EPERM, "root privileges required");
goto exit;
}
if (arg_children_max == 0) {
cpu_set_t cpu_set;
arg_children_max = 8;
if (sched_getaffinity(0, sizeof (cpu_set), &cpu_set) == 0) {
arg_children_max += CPU_COUNT(&cpu_set) * 2;
}
log_debug("set children_max to %u", arg_children_max);
}
/* before opening new files, make sure std{in,out,err} fds are in a sane state */
if (arg_daemonize) {
int fd;
fd = open("/dev/null", O_RDWR);
if (fd < 0)
log_error("cannot open /dev/null");
else {
if (write(STDOUT_FILENO, 0, 0) < 0)
dup2(fd, STDOUT_FILENO);
if (write(STDERR_FILENO, 0, 0) < 0)
dup2(fd, STDERR_FILENO);
if (fd > STDERR_FILENO)
close(fd);
}
}
/* set umask before creating any file/directory */
r = chdir("/");
if (r < 0) {
r = log_error_errno(errno, "could not change dir to /: %m");
goto exit;
}
umask(022);
r = mac_selinux_init("/dev");
if (r < 0) {
log_error_errno(r, "could not initialize labelling: %m");
goto exit;
}
r = mkdir("/run/udev", 0755);
if (r < 0 && errno != EEXIST) {
r = log_error_errno(errno, "could not create /run/udev: %m");
goto exit;
}
dev_setup(NULL, UID_INVALID, GID_INVALID);
r = manager_new(&manager);
if (r < 0)
goto exit;
log_info("starting version " VERSION);
r = udev_rules_apply_static_dev_perms(manager->rules);
if (r < 0)
log_error_errno(r, "failed to apply permissions on static device nodes: %m");
if (arg_daemonize) {
pid_t pid;
pid = fork();
switch (pid) {
case 0:
break;
case -1:
r = log_error_errno(errno, "fork of daemon failed: %m");
goto exit;
default:
mac_selinux_finish();
log_close();
_exit(EXIT_SUCCESS);
}
setsid();
write_string_file("/proc/self/oom_score_adj", "-1000");
} else
sd_notify(1, "READY=1");
for (;;) {
static usec_t last_usec;
struct epoll_event ev[8];
int fdcount;
int timeout;
bool is_worker, is_signal, is_inotify, is_uevent, is_ctrl;
int i;
if (manager->exit) {
/* close sources of new events and discard buffered events */
if (manager->fd_ctrl >= 0) {
epoll_ctl(manager->fd_ep, EPOLL_CTL_DEL, manager->fd_ctrl, NULL);
manager->fd_ctrl = safe_close(manager->fd_ctrl);
}
if (manager->monitor) {
epoll_ctl(manager->fd_ep, EPOLL_CTL_DEL, manager->fd_uevent, NULL);
manager->monitor = udev_monitor_unref(manager->monitor);
}
if (manager->fd_inotify >= 0) {
epoll_ctl(manager->fd_ep, EPOLL_CTL_DEL, manager->fd_inotify, NULL);
manager->fd_inotify = safe_close(manager->fd_inotify);
}
/* discard queued events and kill workers */
event_queue_cleanup(manager, EVENT_QUEUED);
manager_kill_workers(manager);
/* exit after all has cleaned up */
if (udev_list_node_is_empty(&manager->events) && hashmap_isempty(manager->workers))
break;
/* timeout at exit for workers to finish */
timeout = 30 * MSEC_PER_SEC;
} else if (udev_list_node_is_empty(&manager->events) && hashmap_isempty(manager->workers)) {
/* we are idle */
timeout = -1;
/* cleanup possible left-over processes in our cgroup */
if (manager->cgroup)
cg_kill(SYSTEMD_CGROUP_CONTROLLER, manager->cgroup, SIGKILL, false, true, NULL);
} else {
/* kill idle or hanging workers */
timeout = 3 * MSEC_PER_SEC;
}
fdcount = epoll_wait(manager->fd_ep, ev, ELEMENTSOF(ev), timeout);
if (fdcount < 0)
continue;
if (fdcount == 0) {
struct worker *worker;
Iterator j;
/* timeout */
if (manager->exit) {
log_error("timeout, giving up waiting for workers to finish");
break;
}
/* kill idle workers */
if (udev_list_node_is_empty(&manager->events)) {
log_debug("cleanup idle workers");
manager_kill_workers(manager);
}
/* check for hanging events */
HASHMAP_FOREACH(worker, manager->workers, j) {
struct event *event = worker->event;
usec_t ts;
if (worker->state != WORKER_RUNNING)
continue;
assert(event);
ts = now(CLOCK_MONOTONIC);
if ((ts - event->start_usec) > arg_event_timeout_warn_usec) {
if ((ts - event->start_usec) > arg_event_timeout_usec)
on_event_timeout(NULL, 0, event);
else if (!event->warned) {
on_event_timeout_warning(NULL, 0, event);
event->warned = true;
}
}
}
}
is_worker = is_signal = is_inotify = is_uevent = is_ctrl = false;
for (i = 0; i < fdcount; i++) {
if (ev[i].data.fd == manager->fd_worker && ev[i].events & EPOLLIN)
is_worker = true;
else if (ev[i].data.fd == manager->fd_uevent && ev[i].events & EPOLLIN)
is_uevent = true;
else if (ev[i].data.fd == manager->fd_signal && ev[i].events & EPOLLIN)
is_signal = true;
else if (ev[i].data.fd == manager->fd_inotify && ev[i].events & EPOLLIN)
is_inotify = true;
else if (ev[i].data.fd == manager->fd_ctrl && ev[i].events & EPOLLIN)
is_ctrl = true;
}
/* check for changed config, every 3 seconds at most */
if ((now(CLOCK_MONOTONIC) - last_usec) > 3 * USEC_PER_SEC) {
if (udev_rules_check_timestamp(manager->rules))
manager->reload = true;
if (udev_builtin_validate(manager->udev))
manager->reload = true;
last_usec = now(CLOCK_MONOTONIC);
}
/* reload requested, HUP signal received, rules changed, builtin changed */
if (manager->reload) {
manager_kill_workers(manager);
manager->rules = udev_rules_unref(manager->rules);
udev_builtin_exit(manager->udev);
manager->reload = false;
}
/* event has finished */
if (is_worker)
on_worker(NULL, manager->fd_worker, 0, manager);
/* uevent from kernel */
if (is_uevent)
on_uevent(NULL, manager->fd_uevent, 0, manager);
/* start new events */
if (!udev_list_node_is_empty(&manager->events) && !manager->exit && !manager->stop_exec_queue) {
udev_builtin_init(manager->udev);
if (!manager->rules)
manager->rules = udev_rules_new(manager->udev, arg_resolve_names);
if (manager->rules)
event_queue_start(manager);
}
if (is_signal) {
struct signalfd_siginfo fdsi;
ssize_t size;
size = read(manager->fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
if (size == sizeof(struct signalfd_siginfo)) {
switch (fdsi.ssi_signo) {
case SIGINT:
case SIGTERM:
on_sigterm(NULL, &fdsi, manager);
break;
case SIGHUP:
on_sighup(NULL, &fdsi, manager);
break;
case SIGCHLD:
on_sigchld(NULL, &fdsi, manager);
break;
}
}
}
/* we are shutting down, the events below are not handled anymore */
if (manager->exit)
continue;
/* device node watch */
if (is_inotify)
on_inotify(NULL, manager->fd_inotify, 0, manager);
/*
* This needs to be after the inotify handling, to make sure,
* that the ping is send back after the possibly generated
* "change" events by the inotify device node watch.
*/
if (is_ctrl)
on_ctrl_msg(NULL, manager->fd_ctrl, 0, manager);
}
exit:
if (manager)
udev_ctrl_cleanup(manager->ctrl);
mac_selinux_finish();
log_close();
return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
}