diff options
Diffstat (limited to 'src/grp-udev/systemd-udevd')
| -rw-r--r-- | src/grp-udev/systemd-udevd/Makefile | 40 | ||||
| -rw-r--r-- | src/grp-udev/systemd-udevd/systemd-udevd.service.in | 30 | ||||
| -rw-r--r-- | src/grp-udev/systemd-udevd/systemd-udevd.service.xml | 188 | ||||
| -rw-r--r-- | src/grp-udev/systemd-udevd/udev.conf | 3 | ||||
| -rw-r--r-- | src/grp-udev/systemd-udevd/udev.conf.xml | 94 | ||||
| -rw-r--r-- | src/grp-udev/systemd-udevd/udevd.c | 1757 |
6 files changed, 2112 insertions, 0 deletions
diff --git a/src/grp-udev/systemd-udevd/Makefile b/src/grp-udev/systemd-udevd/Makefile new file mode 100644 index 0000000000..fce845e78e --- /dev/null +++ b/src/grp-udev/systemd-udevd/Makefile @@ -0,0 +1,40 @@ +# -*- Mode: makefile; indent-tabs-mode: t -*- +# +# This file is part of systemd. +# +# Copyright 2010-2012 Lennart Poettering +# Copyright 2010-2012 Kay Sievers +# Copyright 2013 Zbigniew Jędrzejewski-Szmek +# Copyright 2013 David Strauss +# Copyright 2016 Luke Shumaker +# +# systemd is free software; you can redistribute it and/or modify it +# under the terms of the GNU Lesser General Public License as published by +# the Free Software Foundation; either version 2.1 of the License, or +# (at your option) any later version. +# +# systemd is distributed in the hope that it will be useful, but +# WITHOUT ANY WARRANTY; without even the implied warranty of +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +# Lesser General Public License for more details. +# +# You should have received a copy of the GNU Lesser General Public License +# along with systemd; If not, see <http://www.gnu.org/licenses/>. +include $(dir $(lastword $(MAKEFILE_LIST)))/../../../config.mk +include $(topsrcdir)/build-aux/Makefile.head.mk + +udevconfdir = $(sysconfdir)/udev +dist_udevconf_DATA = \ + src/udev/udev.conf + +rootlibexec_PROGRAMS += \ + systemd-udevd + +systemd_udevd_SOURCES = \ + src/udev/udevd.c + +systemd_udevd_LDADD = \ + libudev-core.la \ + libsystemd-basic.la + +include $(topsrcdir)/build-aux/Makefile.tail.mk diff --git a/src/grp-udev/systemd-udevd/systemd-udevd.service.in b/src/grp-udev/systemd-udevd/systemd-udevd.service.in new file mode 100644 index 0000000000..46d637883b --- /dev/null +++ b/src/grp-udev/systemd-udevd/systemd-udevd.service.in @@ -0,0 +1,30 @@ +# This file is part of systemd. +# +# systemd is free software; you can redistribute it and/or modify it +# under the terms of the GNU Lesser General Public License as published by +# the Free Software Foundation; either version 2.1 of the License, or +# (at your option) any later version. + +[Unit] +Description=udev Kernel Device Manager +Documentation=man:systemd-udevd.service(8) man:udev(7) +DefaultDependencies=no +Wants=systemd-udevd-control.socket systemd-udevd-kernel.socket +After=systemd-udevd-control.socket systemd-udevd-kernel.socket systemd-sysusers.service +Before=sysinit.target +ConditionPathIsReadWrite=/sys + +[Service] +Type=notify +OOMScoreAdjust=-1000 +Sockets=systemd-udevd-control.socket systemd-udevd-kernel.socket +Restart=always +RestartSec=0 +ExecStart=@rootlibexecdir@/systemd-udevd +KillMode=mixed +WatchdogSec=3min +TasksMax=infinity +MountFlags=slave +MemoryDenyWriteExecute=yes +RestrictRealtime=yes +RestrictAddressFamilies=AF_UNIX AF_NETLINK AF_INET AF_INET6 diff --git a/src/grp-udev/systemd-udevd/systemd-udevd.service.xml b/src/grp-udev/systemd-udevd/systemd-udevd.service.xml new file mode 100644 index 0000000000..243fd06471 --- /dev/null +++ b/src/grp-udev/systemd-udevd/systemd-udevd.service.xml @@ -0,0 +1,188 @@ +<?xml version='1.0'?> <!--*- Mode: nxml; nxml-child-indent: 2; indent-tabs-mode: nil -*--> +<!DOCTYPE refentry PUBLIC "-//OASIS//DTD DocBook XML V4.2//EN" + "http://www.oasis-open.org/docbook/xml/4.2/docbookx.dtd"> + +<refentry id="systemd-udevd.service" + xmlns:xi="http://www.w3.org/2001/XInclude"> + + <refentryinfo> + <title>systemd-udevd.service</title> + <productname>systemd</productname> + <authorgroup> + <author> + <contrib>Developer</contrib> + <firstname>Kay</firstname> + <surname>Sievers</surname> + <email>kay@vrfy.org</email> + </author> + </authorgroup> + </refentryinfo> + + <refmeta> + <refentrytitle>systemd-udevd.service</refentrytitle> + <manvolnum>8</manvolnum> + </refmeta> + + <refnamediv> + <refname>systemd-udevd.service</refname> + <refname>systemd-udevd-control.socket</refname> + <refname>systemd-udevd-kernel.socket</refname> + <refname>systemd-udevd</refname> + <refpurpose>Device event managing daemon</refpurpose> + </refnamediv> + + <refsynopsisdiv> + <para><filename>systemd-udevd.service</filename></para> + <para><filename>systemd-udevd-control.socket</filename></para> + <para><filename>systemd-udevd-kernel.socket</filename></para> + + <cmdsynopsis> + <command>/usr/lib/systemd/systemd-udevd</command> + <arg><option>--daemon</option></arg> + <arg><option>--debug</option></arg> + <arg><option>--children-max=</option></arg> + <arg><option>--exec-delay=</option></arg> + <arg><option>--event-timeout=</option></arg> + <arg><option>--resolve-names=early|late|never</option></arg> + <arg><option>--version</option></arg> + <arg><option>--help</option></arg> + </cmdsynopsis> + + </refsynopsisdiv> + + <refsect1><title>Description</title> + <para><command>systemd-udevd</command> listens to kernel uevents. + For every event, systemd-udevd executes matching instructions + specified in udev rules. See <citerefentry> + <refentrytitle>udev</refentrytitle><manvolnum>7</manvolnum> + </citerefentry>.</para> + + <para>The behavior of the daemon can be configured using + <citerefentry><refentrytitle>udev.conf</refentrytitle><manvolnum>5</manvolnum></citerefentry>, + its command line options, environment variables, and on the kernel + command line, or changed dynamically with <command>udevadm + control</command>. + </para> + </refsect1> + + <refsect1><title>Options</title> + <variablelist> + <varlistentry> + <term><option>--daemon</option></term> + <listitem> + <para>Detach and run in the background.</para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>--debug</option></term> + <listitem> + <para>Print debug messages to standard error.</para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>--children-max=</option></term> + <listitem> + <para>Limit the number of events executed in parallel.</para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>--exec-delay=</option></term> + <listitem> + <para>Delay the execution of <varname>RUN</varname> + instructions by the given number of seconds. This option + might be useful when debugging system crashes during + coldplug caused by loading non-working kernel + modules.</para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>--event-timeout=</option></term> + <listitem> + <para>Set the number of seconds to wait for events to finish. After + this time, the event will be terminated. The default is 180 seconds.</para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>--resolve-names=</option></term> + <listitem> + <para>Specify when systemd-udevd should resolve names of users and groups. + When set to <option>early</option> (the default), names will be + resolved when the rules are parsed. When set to + <option>late</option>, names will be resolved for every event. + When set to <option>never</option>, names will never be resolved + and all devices will be owned by root.</para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>--help</option></term> + + <xi:include href="standard-options.xml" xpointer="help-text" /> + </varlistentry> + <xi:include href="standard-options.xml" xpointer="version" /> + </variablelist> + </refsect1> + + <refsect1><title>Kernel command line</title> + <variablelist class='kernel-commandline-options'> + <para>Parameters starting with "rd." will be read when + <command>systemd-udevd</command> is used in an initrd.</para> + <varlistentry> + <term><varname>udev.log-priority=</varname></term> + <term><varname>rd.udev.log-priority=</varname></term> + <listitem> + <para>Set the log level.</para> + </listitem> + </varlistentry> + <varlistentry> + <term><varname>udev.children-max=</varname></term> + <term><varname>rd.udev.children-max=</varname></term> + <listitem> + <para>Limit the number of events executed in parallel.</para> + </listitem> + </varlistentry> + <varlistentry> + <term><varname>udev.exec-delay=</varname></term> + <term><varname>rd.udev.exec-delay=</varname></term> + <listitem> + <para>Delay the execution of <varname>RUN</varname> instructions by the given + number of seconds. This option might be useful when + debugging system crashes during coldplug caused by loading + non-working kernel modules.</para> + </listitem> + </varlistentry> + <varlistentry> + <term><varname>udev.event-timeout=</varname></term> + <term><varname>rd.udev.event-timeout=</varname></term> + <listitem> + <para>Wait for events to finish up to the given number + of seconds. This option might be useful if events are + terminated due to kernel drivers taking too long to initialize.</para> + </listitem> + </varlistentry> + <varlistentry> + <term><varname>net.ifnames=</varname></term> + <listitem> + <para>Network interfaces are renamed to give them predictable names + when possible. It is enabled by default; specifying 0 disables it.</para> + </listitem> + </varlistentry> + </variablelist> + <!-- when adding entries here, consider also adding them + in kernel-command-line.xml --> + </refsect1> + + <refsect1> + <title>See Also</title> + <para> + <citerefentry><refentrytitle>udev.conf</refentrytitle><manvolnum>5</manvolnum></citerefentry>, + <citerefentry><refentrytitle>udev</refentrytitle><manvolnum>7</manvolnum></citerefentry>, + <citerefentry><refentrytitle>udevadm</refentrytitle><manvolnum>8</manvolnum></citerefentry> + </para> + </refsect1> +</refentry> diff --git a/src/grp-udev/systemd-udevd/udev.conf b/src/grp-udev/systemd-udevd/udev.conf new file mode 100644 index 0000000000..47d1433002 --- /dev/null +++ b/src/grp-udev/systemd-udevd/udev.conf @@ -0,0 +1,3 @@ +# see udev.conf(5) for details + +#udev_log="info" diff --git a/src/grp-udev/systemd-udevd/udev.conf.xml b/src/grp-udev/systemd-udevd/udev.conf.xml new file mode 100644 index 0000000000..e104e53f5d --- /dev/null +++ b/src/grp-udev/systemd-udevd/udev.conf.xml @@ -0,0 +1,94 @@ +<?xml version='1.0'?> +<!DOCTYPE refentry PUBLIC "-//OASIS//DTD DocBook XML V4.2//EN" + "http://www.oasis-open.org/docbook/xml/4.2/docbookx.dtd"> + +<!-- + This file is part of systemd. + + Copyright 2014 Zbigniew Jędrzejewski-Szmek + + systemd is free software; you can redistribute it and/or modify it + under the terms of the GNU Lesser General Public License as published by + the Free Software Foundation; either version 2.1 of the License, or + (at your option) any later version. + + systemd is distributed in the hope that it will be useful, but + WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public License + along with systemd; If not, see <http://www.gnu.org/licenses/>. +--> + +<refentry id="udev.conf" + xmlns:xi="http://www.w3.org/2001/XInclude"> + + <refentryinfo> + <title>udev.conf</title> + <productname>systemd</productname> + <authorgroup> + <author> + <contrib>Developer</contrib> + <firstname>Kay</firstname> + <surname>Sievers</surname> + <email>kay@vrfy.org</email> + </author> + </authorgroup> + </refentryinfo> + + <refmeta> + <refentrytitle>udev.conf</refentrytitle> + <manvolnum>5</manvolnum> + </refmeta> + + <refnamediv> + <refname>udev.conf</refname> + <refpurpose>Configuration for device event managing daemon</refpurpose> + </refnamediv> + + <refsynopsisdiv> + <para><filename>/etc/udev/udev.conf</filename></para> + </refsynopsisdiv> + + <refsect1> + <title>Description</title> + + <para> + <citerefentry><refentrytitle>systemd-udevd</refentrytitle><manvolnum>8</manvolnum></citerefentry> + expects its main configuration file at + <filename>/etc/udev/udev.conf</filename>. It consists of a set + of variables allowing the user to override default udev + values. All empty lines or lines beginning with '#' are + ignored. The following variables can be set: + </para> + + <variablelist> + <varlistentry> + <term><varname>udev_log</varname></term> + + <listitem> + <para>The log level. Valid values are the numerical + syslog priorities or their textual representations: + <option>err</option>, <option>info</option> and + <option>debug</option>.</para> + </listitem> + </varlistentry> + </variablelist> + + <para> + In addition, <filename>systemd-udevd</filename> can be configured + by command line options and the kernel command line (see + <citerefentry><refentrytitle>systemd-udevd</refentrytitle><manvolnum>8</manvolnum></citerefentry>). + </para> + </refsect1> + + <refsect1> + <title>See Also</title> + <para> + <citerefentry><refentrytitle>systemd-udevd</refentrytitle><manvolnum>8</manvolnum></citerefentry>, + <citerefentry><refentrytitle>udev</refentrytitle><manvolnum>7</manvolnum></citerefentry>, + <citerefentry><refentrytitle>udevadm</refentrytitle><manvolnum>8</manvolnum></citerefentry> + </para> + </refsect1> +</refentry> diff --git a/src/grp-udev/systemd-udevd/udevd.c b/src/grp-udev/systemd-udevd/udevd.c new file mode 100644 index 0000000000..ce50235c1b --- /dev/null +++ b/src/grp-udev/systemd-udevd/udevd.c @@ -0,0 +1,1757 @@ +/* + * Copyright (C) 2004-2012 Kay Sievers <kay@vrfy.org> + * Copyright (C) 2004 Chris Friesen <chris_friesen@sympatico.ca> + * Copyright (C) 2009 Canonical Ltd. + * Copyright (C) 2009 Scott James Remnant <scott@netsplit.com> + * + * This program is free software: you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation, either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program. If not, see <http://www.gnu.org/licenses/>. + */ + +#include <errno.h> +#include <fcntl.h> +#include <getopt.h> +#include <signal.h> +#include <stdbool.h> +#include <stddef.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <sys/epoll.h> +#include <sys/file.h> +#include <sys/inotify.h> +#include <sys/ioctl.h> +#include <sys/mount.h> +#include <sys/prctl.h> +#include <sys/signalfd.h> +#include <sys/socket.h> +#include <sys/stat.h> +#include <sys/time.h> +#include <sys/wait.h> +#include <unistd.h> + +#include <systemd/sd-daemon.h> +#include <systemd/sd-event.h> + +#include "sd-netlink/netlink-util.h" +#include "systemd-basic/alloc-util.h" +#include "systemd-basic/cgroup-util.h" +#include "systemd-basic/cpu-set-util.h" +#include "systemd-basic/fd-util.h" +#include "systemd-basic/fileio.h" +#include "systemd-basic/formats-util.h" +#include "systemd-basic/fs-util.h" +#include "systemd-basic/hashmap.h" +#include "systemd-basic/io-util.h" +#include "systemd-basic/parse-util.h" +#include "systemd-basic/proc-cmdline.h" +#include "systemd-basic/process-util.h" +#include "systemd-basic/selinux-util.h" +#include "systemd-basic/signal-util.h" +#include "systemd-basic/socket-util.h" +#include "systemd-basic/string-util.h" +#include "systemd-basic/terminal-util.h" +#include "systemd-basic/user-util.h" +#include "systemd-shared/dev-setup.h" +#include "systemd-shared/udev-util.h" +#include "udev.h" + +static bool arg_debug = false; +static int arg_daemonize = false; +static int arg_resolve_names = 1; +static unsigned arg_children_max; +static int arg_exec_delay; +static usec_t arg_event_timeout_usec = 180 * USEC_PER_SEC; +static usec_t arg_event_timeout_warn_usec = 180 * USEC_PER_SEC / 3; + +typedef struct Manager { + struct udev *udev; + sd_event *event; + Hashmap *workers; + struct udev_list_node events; + const char *cgroup; + pid_t pid; /* the process that originally allocated the manager object */ + + struct udev_rules *rules; + struct udev_list properties; + + struct udev_monitor *monitor; + struct udev_ctrl *ctrl; + struct udev_ctrl_connection *ctrl_conn_blocking; + int fd_inotify; + int worker_watch[2]; + + sd_event_source *ctrl_event; + sd_event_source *uevent_event; + sd_event_source *inotify_event; + + usec_t last_usec; + + bool stop_exec_queue:1; + bool exit:1; +} Manager; + +enum event_state { + EVENT_UNDEF, + EVENT_QUEUED, + EVENT_RUNNING, +}; + +struct event { + struct udev_list_node node; + Manager *manager; + struct udev *udev; + struct udev_device *dev; + struct udev_device *dev_kernel; + struct worker *worker; + enum event_state state; + unsigned long long int delaying_seqnum; + unsigned long long int seqnum; + const char *devpath; + size_t devpath_len; + const char *devpath_old; + dev_t devnum; + int ifindex; + bool is_block; + sd_event_source *timeout_warning; + sd_event_source *timeout; +}; + +static inline struct event *node_to_event(struct udev_list_node *node) { + return container_of(node, struct event, node); +} + +static void event_queue_cleanup(Manager *manager, enum event_state type); + +enum worker_state { + WORKER_UNDEF, + WORKER_RUNNING, + WORKER_IDLE, + WORKER_KILLED, +}; + +struct worker { + Manager *manager; + struct udev_list_node node; + int refcount; + pid_t pid; + struct udev_monitor *monitor; + enum worker_state state; + struct event *event; +}; + +/* passed from worker to main process */ +struct worker_message { +}; + +static void event_free(struct event *event) { + int r; + + if (!event) + return; + + udev_list_node_remove(&event->node); + udev_device_unref(event->dev); + udev_device_unref(event->dev_kernel); + + sd_event_source_unref(event->timeout_warning); + sd_event_source_unref(event->timeout); + + if (event->worker) + event->worker->event = NULL; + + assert(event->manager); + + if (udev_list_node_is_empty(&event->manager->events)) { + /* only clean up the queue from the process that created it */ + if (event->manager->pid == getpid()) { + r = unlink("/run/udev/queue"); + if (r < 0) + log_warning_errno(errno, "could not unlink /run/udev/queue: %m"); + } + } + + free(event); +} + +static void worker_free(struct worker *worker) { + if (!worker) + return; + + assert(worker->manager); + + hashmap_remove(worker->manager->workers, PID_TO_PTR(worker->pid)); + udev_monitor_unref(worker->monitor); + event_free(worker->event); + + free(worker); +} + +static void manager_workers_free(Manager *manager) { + struct worker *worker; + Iterator i; + + assert(manager); + + HASHMAP_FOREACH(worker, manager->workers, i) + worker_free(worker); + + manager->workers = hashmap_free(manager->workers); +} + +static int worker_new(struct worker **ret, Manager *manager, struct udev_monitor *worker_monitor, pid_t pid) { + _cleanup_free_ struct worker *worker = NULL; + int r; + + assert(ret); + assert(manager); + assert(worker_monitor); + assert(pid > 1); + + worker = new0(struct worker, 1); + if (!worker) + return -ENOMEM; + + worker->refcount = 1; + worker->manager = manager; + /* close monitor, but keep address around */ + udev_monitor_disconnect(worker_monitor); + worker->monitor = udev_monitor_ref(worker_monitor); + worker->pid = pid; + + r = hashmap_ensure_allocated(&manager->workers, NULL); + if (r < 0) + return r; + + r = hashmap_put(manager->workers, PID_TO_PTR(pid), worker); + if (r < 0) + return r; + + *ret = worker; + worker = NULL; + + return 0; +} + +static int on_event_timeout(sd_event_source *s, uint64_t usec, void *userdata) { + struct event *event = userdata; + + assert(event); + assert(event->worker); + + kill_and_sigcont(event->worker->pid, SIGKILL); + event->worker->state = WORKER_KILLED; + + log_error("seq %llu '%s' killed", udev_device_get_seqnum(event->dev), event->devpath); + + return 1; +} + +static int on_event_timeout_warning(sd_event_source *s, uint64_t usec, void *userdata) { + struct event *event = userdata; + + assert(event); + + log_warning("seq %llu '%s' is taking a long time", udev_device_get_seqnum(event->dev), event->devpath); + + return 1; +} + +static void worker_attach_event(struct worker *worker, struct event *event) { + sd_event *e; + uint64_t usec; + + assert(worker); + assert(worker->manager); + assert(event); + assert(!event->worker); + assert(!worker->event); + + worker->state = WORKER_RUNNING; + worker->event = event; + event->state = EVENT_RUNNING; + event->worker = worker; + + e = worker->manager->event; + + assert_se(sd_event_now(e, clock_boottime_or_monotonic(), &usec) >= 0); + + (void) sd_event_add_time(e, &event->timeout_warning, clock_boottime_or_monotonic(), + usec + arg_event_timeout_warn_usec, USEC_PER_SEC, on_event_timeout_warning, event); + + (void) sd_event_add_time(e, &event->timeout, clock_boottime_or_monotonic(), + usec + arg_event_timeout_usec, USEC_PER_SEC, on_event_timeout, event); +} + +static void manager_free(Manager *manager) { + if (!manager) + return; + + udev_builtin_exit(manager->udev); + + sd_event_source_unref(manager->ctrl_event); + sd_event_source_unref(manager->uevent_event); + sd_event_source_unref(manager->inotify_event); + + udev_unref(manager->udev); + sd_event_unref(manager->event); + 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); + + 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; + int r = 0; + + /* 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); + r = udev_monitor_enable_receiving(worker_monitor); + if (r < 0) + log_error_errno(r, "worker: could not enable receiving of device: %m"); + + pid = fork(); + switch (pid) { + case 0: { + struct udev_device *dev = NULL; + _cleanup_(sd_netlink_unrefp) sd_netlink *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; + + /* take initial device from queue */ + dev = event->dev; + event->dev = NULL; + + unsetenv("NOTIFY_SOCKET"); + + 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->worker_watch[READ_END] = safe_close(manager->worker_watch[READ_END]); + + manager->ctrl_event = sd_event_source_unref(manager->ctrl_event); + manager->uevent_event = sd_event_source_unref(manager->uevent_event); + manager->inotify_event = sd_event_source_unref(manager->inotify_event); + + manager->event = sd_event_unref(manager->event); + + 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. + Ignore error, not much we can do in that case. */ + (void) prctl(PR_SET_PDEATHSIG, SIGTERM); + + /* Reset OOM score, we only protect the main daemon. */ + write_string_file("/proc/self/oom_score_adj", "0", 0); + + for (;;) { + struct udev_event *udev_event; + int fd_lock = -1; + + assert(dev); + + log_debug("seq %llu running", udev_device_get_seqnum(dev)); + udev_event = udev_event_new(dev); + if (udev_event == NULL) { + r = -ENOMEM; + goto out; + } + + 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); + 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); + + udev_event_execute_run(udev_event, + arg_event_timeout_usec, arg_event_timeout_warn_usec); + + if (udev_event->rtnl) + /* in case rtnl was initialized */ + rtnl = sd_netlink_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; + + 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; + + 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 int on_exit_timeout(sd_event_source *s, uint64_t usec, void *userdata) { + Manager *manager = userdata; + + assert(manager); + + log_error_errno(ETIMEDOUT, "giving up waiting for workers to finish"); + + sd_event_exit(manager->event, -ETIMEDOUT); + + return 1; +} + +static void manager_exit(Manager *manager) { + uint64_t usec; + int r; + + assert(manager); + + manager->exit = true; + + sd_notify(false, + "STOPPING=1\n" + "STATUS=Starting shutdown..."); + + /* close sources of new events and discard buffered events */ + manager->ctrl_event = sd_event_source_unref(manager->ctrl_event); + manager->ctrl = udev_ctrl_unref(manager->ctrl); + + manager->inotify_event = sd_event_source_unref(manager->inotify_event); + manager->fd_inotify = safe_close(manager->fd_inotify); + + manager->uevent_event = sd_event_source_unref(manager->uevent_event); + manager->monitor = udev_monitor_unref(manager->monitor); + + /* discard queued events and kill workers */ + event_queue_cleanup(manager, EVENT_QUEUED); + manager_kill_workers(manager); + + assert_se(sd_event_now(manager->event, clock_boottime_or_monotonic(), &usec) >= 0); + + r = sd_event_add_time(manager->event, NULL, clock_boottime_or_monotonic(), + usec + 30 * USEC_PER_SEC, USEC_PER_SEC, on_exit_timeout, manager); + if (r < 0) + return; +} + +/* reload requested, HUP signal received, rules changed, builtin changed */ +static void manager_reload(Manager *manager) { + + assert(manager); + + sd_notify(false, + "RELOADING=1\n" + "STATUS=Flushing configuration..."); + + manager_kill_workers(manager); + manager->rules = udev_rules_unref(manager->rules); + udev_builtin_exit(manager->udev); + + sd_notifyf(false, + "READY=1\n" + "STATUS=Processing with %u children at max", arg_children_max); +} + +static void event_queue_start(Manager *manager) { + struct udev_list_node *loop; + usec_t usec; + + assert(manager); + + if (udev_list_node_is_empty(&manager->events) || + manager->exit || manager->stop_exec_queue) + return; + + assert_se(sd_event_now(manager->event, clock_boottime_or_monotonic(), &usec) >= 0); + /* check for changed config, every 3 seconds at most */ + if (manager->last_usec == 0 || + (usec - manager->last_usec) > 3 * USEC_PER_SEC) { + if (udev_rules_check_timestamp(manager->rules) || + udev_builtin_validate(manager->udev)) + manager_reload(manager); + + manager->last_usec = usec; + } + + udev_builtin_init(manager->udev); + + if (!manager->rules) { + manager->rules = udev_rules_new(manager->udev, arg_resolve_names); + if (!manager->rules) + return; + } + + 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; + } + + CMSG_FOREACH(cmsg, &msghdr) { + 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, PID_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); + } + + /* we have free workers, try to schedule events */ + event_queue_start(manager); + + 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); + else + /* we have fresh events, try to schedule them */ + event_queue_start(manager); + } + + 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; + event_queue_start(manager); + } + + if (udev_ctrl_get_reload(ctrl_msg) > 0) { + log_debug("udevd message (RELOAD) received"); + manager_reload(manager); + } + + 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; + + (void) sd_notifyf(false, + "READY=1\n" + "STATUS=Processing with %u children at max", arg_children_max); + } + + 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(manager); + /* 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", WRITE_STRING_FILE_CREATE); + + 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", WRITE_STRING_FILE_CREATE); + } + + 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", WRITE_STRING_FILE_CREATE); + + 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(manager); + + return 1; +} + +static int on_sighup(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) { + Manager *manager = userdata; + + assert(manager); + + manager_reload(manager); + + 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, PID_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); + } + + /* we can start new workers, try to schedule events */ + event_queue_start(manager); + + return 1; +} + +static int on_post(sd_event_source *s, void *userdata) { + Manager *manager = userdata; + int r; + + assert(manager); + + if (udev_list_node_is_empty(&manager->events)) { + /* no pending events */ + if (!hashmap_isempty(manager->workers)) { + /* there are idle workers */ + log_debug("cleanup idle workers"); + manager_kill_workers(manager); + } else { + /* we are idle */ + if (manager->exit) { + r = sd_event_exit(manager->event, 0); + if (r < 0) + return r; + } else if (manager->cgroup) + /* cleanup possible left-over processes in our cgroup */ + cg_kill(SYSTEMD_CGROUP_CONTROLLER, manager->cgroup, SIGKILL, CGROUP_IGNORE_SELF, NULL, NULL, NULL); + } + } + + return 1; +} + +static int listen_fds(int *rctrl, int *rnetlink) { + _cleanup_udev_unref_ struct udev *udev = NULL; + int ctrl_fd = -1, netlink_fd = -1; + int fd, n, r; + + assert(rctrl); + assert(rnetlink); + + n = sd_listen_fds(true); + if (n < 0) + return n; + + 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_fd >= 0) + return -EINVAL; + ctrl_fd = fd; + continue; + } + + if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1)) { + if (netlink_fd >= 0) + return -EINVAL; + netlink_fd = fd; + continue; + } + + return -EINVAL; + } + + if (ctrl_fd < 0) { + _cleanup_udev_ctrl_unref_ struct udev_ctrl *ctrl = NULL; + + udev = udev_new(); + if (!udev) + return -ENOMEM; + + ctrl = udev_ctrl_new(udev); + if (!ctrl) + return log_error_errno(EINVAL, "error initializing udev control socket"); + + r = udev_ctrl_enable_receiving(ctrl); + if (r < 0) + return log_error_errno(EINVAL, "error binding udev control socket"); + + fd = udev_ctrl_get_fd(ctrl); + if (fd < 0) + return log_error_errno(EIO, "could not get ctrl fd"); + + ctrl_fd = fcntl(fd, F_DUPFD_CLOEXEC, 3); + if (ctrl_fd < 0) + return log_error_errno(errno, "could not dup ctrl fd: %m"); + } + + if (netlink_fd < 0) { + _cleanup_udev_monitor_unref_ struct udev_monitor *monitor = NULL; + + if (!udev) { + udev = udev_new(); + if (!udev) + return -ENOMEM; + } + + monitor = udev_monitor_new_from_netlink(udev, "kernel"); + if (!monitor) + return log_error_errno(EINVAL, "error initializing netlink socket"); + + (void) udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024); + + r = udev_monitor_enable_receiving(monitor); + if (r < 0) + return log_error_errno(EINVAL, "error binding netlink socket"); + + fd = udev_monitor_get_fd(monitor); + if (fd < 0) + return log_error_errno(netlink_fd, "could not get uevent fd: %m"); + + netlink_fd = fcntl(fd, F_DUPFD_CLOEXEC, 3); + if (ctrl_fd < 0) + return log_error_errno(errno, "could not dup netlink fd: %m"); + } + + *rctrl = ctrl_fd; + *rnetlink = netlink_fd; + + 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, void *data) { + int r = 0; + + assert(key); + + if (!value) + return 0; + + if (streq(key, "udev.log-priority") && value) { + r = util_log_priority(value); + if (r >= 0) + log_set_max_level(r); + } else if (streq(key, "udev.event-timeout") && value) { + r = safe_atou64(value, &arg_event_timeout_usec); + if (r >= 0) { + arg_event_timeout_usec *= USEC_PER_SEC; + arg_event_timeout_warn_usec = (arg_event_timeout_usec / 3) ? : 1; + } + } else if (streq(key, "udev.children-max") && value) + r = safe_atou(value, &arg_children_max); + else if (streq(key, "udev.exec-delay") && value) + r = safe_atoi(value, &arg_exec_delay); + else if (startswith(key, "udev.")) + log_warning("Unknown udev kernel command line option \"%s\"", key); + + if (r < 0) + log_warning_errno(r, "Failed to parse \"%s=%s\", ignoring: %m", key, value); + 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:Dt:N: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, int fd_ctrl, int fd_uevent, const char *cgroup) { + _cleanup_(manager_freep) Manager *manager = NULL; + int r, fd_worker, one = 1; + + assert(ret); + assert(fd_ctrl >= 0); + assert(fd_uevent >= 0); + + manager = new0(Manager, 1); + if (!manager) + return log_oom(); + + 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); + + manager->cgroup = cgroup; + + manager->ctrl = udev_ctrl_new_from_fd(manager->udev, 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", fd_uevent); + if (!manager->monitor) + return log_error_errno(EINVAL, "error taking over netlink 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"); + + fd_worker = manager->worker_watch[READ_END]; + + r = setsockopt(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 */ + assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGTERM, SIGINT, SIGHUP, SIGCHLD, -1) >= 0); + + r = sd_event_default(&manager->event); + if (r < 0) + return log_error_errno(r, "could not allocate event loop: %m"); + + r = sd_event_add_signal(manager->event, NULL, SIGINT, on_sigterm, manager); + if (r < 0) + return log_error_errno(r, "error creating sigint event source: %m"); + + r = sd_event_add_signal(manager->event, NULL, SIGTERM, on_sigterm, manager); + if (r < 0) + return log_error_errno(r, "error creating sigterm event source: %m"); + + r = sd_event_add_signal(manager->event, NULL, SIGHUP, on_sighup, manager); + if (r < 0) + return log_error_errno(r, "error creating sighup event source: %m"); + + r = sd_event_add_signal(manager->event, NULL, SIGCHLD, on_sigchld, manager); + if (r < 0) + return log_error_errno(r, "error creating sigchld event source: %m"); + + r = sd_event_set_watchdog(manager->event, true); + if (r < 0) + return log_error_errno(r, "error creating watchdog event source: %m"); + + r = sd_event_add_io(manager->event, &manager->ctrl_event, fd_ctrl, EPOLLIN, on_ctrl_msg, manager); + if (r < 0) + return log_error_errno(r, "error creating ctrl event source: %m"); + + /* This needs to be after the inotify and uevent handling, to make sure + * that the ping is send back after fully processing the pending uevents + * (including the synthetic ones we may create due to inotify events). + */ + r = sd_event_source_set_priority(manager->ctrl_event, SD_EVENT_PRIORITY_IDLE); + if (r < 0) + return log_error_errno(r, "cold not set IDLE event priority for ctrl event source: %m"); + + r = sd_event_add_io(manager->event, &manager->inotify_event, manager->fd_inotify, EPOLLIN, on_inotify, manager); + if (r < 0) + return log_error_errno(r, "error creating inotify event source: %m"); + + r = sd_event_add_io(manager->event, &manager->uevent_event, fd_uevent, EPOLLIN, on_uevent, manager); + if (r < 0) + return log_error_errno(r, "error creating uevent event source: %m"); + + r = sd_event_add_io(manager->event, NULL, fd_worker, EPOLLIN, on_worker, manager); + if (r < 0) + return log_error_errno(r, "error creating worker event source: %m"); + + r = sd_event_add_post(manager->event, NULL, on_post, manager); + if (r < 0) + return log_error_errno(r, "error creating post event source: %m"); + + *ret = manager; + manager = NULL; + + return 0; +} + +static int run(int fd_ctrl, int fd_uevent, const char *cgroup) { + _cleanup_(manager_freep) Manager *manager = NULL; + int r; + + r = manager_new(&manager, fd_ctrl, fd_uevent, cgroup); + if (r < 0) { + r = log_error_errno(r, "failed to allocate manager object: %m"); + goto exit; + } + + 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"); + + (void) sd_notifyf(false, + "READY=1\n" + "STATUS=Processing with %u children at max", arg_children_max); + + r = sd_event_loop(manager->event); + if (r < 0) { + log_error_errno(r, "event loop failed: %m"); + goto exit; + } + + sd_event_get_exit_code(manager->event, &r); + +exit: + sd_notify(false, + "STOPPING=1\n" + "STATUS=Shutting down..."); + if (manager) + udev_ctrl_cleanup(manager->ctrl); + return r; +} + +int main(int argc, char *argv[]) { + _cleanup_free_ char *cgroup = NULL; + int fd_ctrl = -1, fd_uevent = -1; + 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, NULL, true); + if (r < 0) + log_warning_errno(r, "failed to parse kernel command line, ignoring: %m"); + + if (arg_debug) { + log_set_target(LOG_TARGET_CONSOLE); + 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); + } + + /* 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(); + 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); + + if (getppid() == 1) { + /* get our own cgroup, we regularly kill everything udev has left behind + we only do this on systemd systems, and only if we are directly spawned + by PID1. otherwise we are not guaranteed to have a dedicated cgroup */ + r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &cgroup); + if (r < 0) { + if (r == -ENOENT || r == -ENOMEDIUM) + log_debug_errno(r, "did not find dedicated cgroup: %m"); + else + log_warning_errno(r, "failed to get cgroup: %m"); + } + } + + r = listen_fds(&fd_ctrl, &fd_uevent); + if (r < 0) { + r = log_error_errno(r, "could not listen on fds: %m"); + goto exit; + } + + if (arg_daemonize) { + pid_t pid; + + log_info("starting version " VERSION); + + /* connect /dev/null to stdin, stdout, stderr */ + if (log_get_max_level() < LOG_DEBUG) { + r = make_null_stdio(); + if (r < 0) + log_warning_errno(r, "Failed to redirect standard streams to /dev/null: %m"); + } + + + + 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", 0); + } + + r = run(fd_ctrl, fd_uevent, cgroup); + +exit: + mac_selinux_finish(); + log_close(); + return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS; +} |