src/udev/udevd.c

   1 /*
   2  * Copyright (C) 2004-2012 Kay Sievers <kay@vrfy.org>
   3  * Copyright (C) 2004 Chris Friesen <chris_friesen@sympatico.ca>
   4  * Copyright (C) 2009 Canonical Ltd.
   5  * Copyright (C) 2009 Scott James Remnant <scott@netsplit.com>
   6  *
   7  * This program is free software: you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License as published by
   9  * the Free Software Foundation, either version 2 of the License, or
  10  * (at your option) any later version.
  11  *
  12  * This program is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  * GNU General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  19  */
  20
  21 #include <stddef.h>
  22 #include <signal.h>
  23 #include <unistd.h>
  24 #include <errno.h>
  25 #include <stdio.h>
  26 #include <stdlib.h>
  27 #include <stdbool.h>
  28 #include <string.h>
  29 #include <ctype.h>
  30 #include <fcntl.h>
  31 #include <time.h>
  32 #include <getopt.h>
  33 #include <dirent.h>
  34 #include <sys/file.h>
  35 #include <sys/time.h>
  36 #include <sys/prctl.h>
  37 #include <sys/socket.h>
  38 #include <sys/un.h>
  39 #include <sys/signalfd.h>
  40 #include <sys/epoll.h>
  41 #include <sys/mount.h>
  42 #include <sys/poll.h>
  43 #include <sys/wait.h>
  44 #include <sys/stat.h>
  45 #include <sys/ioctl.h>
  46 #include <sys/inotify.h>
  47 #include <sys/utsname.h>
  48
  49 #include "udev.h"
  50 #include "udev-util.h"
  51 #include "sd-daemon.h"
  52 #include "cgroup-util.h"
  53 #include "dev-setup.h"
  54 #include "fileio.h"
  55
  56 static bool debug;
  57
  58 void udev_main_log(struct udev *udev, int priority,
  59                    const char *file, int line, const char *fn,
  60                    const char *format, va_list args)
  61 {
  62         log_metav(priority, file, line, fn, format, args);
  63 }
  64
  65 static struct udev_rules *rules;
  66 static struct udev_ctrl *udev_ctrl;
  67 static struct udev_monitor *monitor;
  68 static int worker_watch[2] = { -1, -1 };
  69 static int fd_signal = -1;
  70 static int fd_ep = -1;
  71 static int fd_inotify = -1;
  72 static bool stop_exec_queue;
  73 static bool reload;
  74 static int children;
  75 static int children_max;
  76 static int exec_delay;
  77 static int event_timeout = 30;
  78 static sigset_t sigmask_orig;
  79 static UDEV_LIST(event_list);
  80 static UDEV_LIST(worker_list);
  81 static char *udev_cgroup;
  82 static bool udev_exit;
  83
  84 enum event_state {
  85         EVENT_UNDEF,
  86         EVENT_QUEUED,
  87         EVENT_RUNNING,
  88 };
  89
  90 struct event {
  91         struct udev_list_node node;
  92         struct udev *udev;
  93         struct udev_device *dev;
  94         enum event_state state;
  95         int exitcode;
  96         unsigned long long int delaying_seqnum;
  97         unsigned long long int seqnum;
  98         const char *devpath;
  99         size_t devpath_len;
 100         const char *devpath_old;
 101         dev_t devnum;
 102         int ifindex;
 103         bool is_block;
 104 #ifdef HAVE_FIRMWARE
 105         bool nodelay;
 106 #endif
 107 };
 108
 109 static inline struct event *node_to_event(struct udev_list_node *node)
 110 {
 111         return container_of(node, struct event, node);
 112 }
 113
 114 static void event_queue_cleanup(struct udev *udev, enum event_state type);
 115
 116 enum worker_state {
 117         WORKER_UNDEF,
 118         WORKER_RUNNING,
 119         WORKER_IDLE,
 120         WORKER_KILLED,
 121 };
 122
 123 struct worker {
 124         struct udev_list_node node;
 125         struct udev *udev;
 126         int refcount;
 127         pid_t pid;
 128         struct udev_monitor *monitor;
 129         enum worker_state state;
 130         struct event *event;
 131         usec_t event_start_usec;
 132 };
 133
 134 /* passed from worker to main process */
 135 struct worker_message {
 136         pid_t pid;
 137         int exitcode;
 138 };
 139
 140 static inline struct worker *node_to_worker(struct udev_list_node *node)
 141 {
 142         return container_of(node, struct worker, node);
 143 }
 144
 145 static void event_queue_delete(struct event *event)
 146 {
 147         udev_list_node_remove(&event->node);
 148         udev_device_unref(event->dev);
 149         free(event);
 150 }
 151
 152 static struct worker *worker_ref(struct worker *worker)
 153 {
 154         worker->refcount++;
 155         return worker;
 156 }
 157
 158 static void worker_cleanup(struct worker *worker)
 159 {
 160         udev_list_node_remove(&worker->node);
 161         udev_monitor_unref(worker->monitor);
 162         children--;
 163         free(worker);
 164 }
 165
 166 static void worker_unref(struct worker *worker)
 167 {
 168         worker->refcount--;
 169         if (worker->refcount > 0)
 170                 return;
 171         log_debug("worker [%u] cleaned up", worker->pid);
 172         worker_cleanup(worker);
 173 }
 174
 175 static void worker_list_cleanup(struct udev *udev)
 176 {
 177         struct udev_list_node *loop, *tmp;
 178
 179         udev_list_node_foreach_safe(loop, tmp, &worker_list) {
 180                 struct worker *worker = node_to_worker(loop);
 181
 182                 worker_cleanup(worker);
 183         }
 184 }
 185
 186 static void worker_new(struct event *event)
 187 {
 188         struct udev *udev = event->udev;
 189         struct worker *worker;
 190         struct udev_monitor *worker_monitor;
 191         pid_t pid;
 192
 193         /* listen for new events */
 194         worker_monitor = udev_monitor_new_from_netlink(udev, NULL);
 195         if (worker_monitor == NULL)
 196                 return;
 197         /* allow the main daemon netlink address to send devices to the worker */
 198         udev_monitor_allow_unicast_sender(worker_monitor, monitor);
 199         udev_monitor_enable_receiving(worker_monitor);
 200
 201         worker = new0(struct worker, 1);
 202         if (worker == NULL) {
 203                 udev_monitor_unref(worker_monitor);
 204                 return;
 205         }
 206         /* worker + event reference */
 207         worker->refcount = 2;
 208         worker->udev = udev;
 209
 210         pid = fork();
 211         switch (pid) {
 212         case 0: {
 213                 struct udev_device *dev = NULL;
 214                 int fd_monitor;
 215                 struct epoll_event ep_signal, ep_monitor;
 216                 sigset_t mask;
 217                 int rc = EXIT_SUCCESS;
 218
 219                 /* take initial device from queue */
 220                 dev = event->dev;
 221                 event->dev = NULL;
 222
 223                 free(worker);
 224                 worker_list_cleanup(udev);
 225                 event_queue_cleanup(udev, EVENT_UNDEF);
 226                 udev_monitor_unref(monitor);
 227                 udev_ctrl_unref(udev_ctrl);
 228                 close(fd_signal);
 229                 close(fd_ep);
 230                 close(worker_watch[READ_END]);
 231
 232                 sigfillset(&mask);
 233                 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
 234                 if (fd_signal < 0) {
 235                         log_error("error creating signalfd %m");
 236                         rc = 2;
 237                         goto out;
 238                 }
 239
 240                 fd_ep = epoll_create1(EPOLL_CLOEXEC);
 241                 if (fd_ep < 0) {
 242                         log_error("error creating epoll fd: %m");
 243                         rc = 3;
 244                         goto out;
 245                 }
 246
 247                 memzero(&ep_signal, sizeof(struct epoll_event));
 248                 ep_signal.events = EPOLLIN;
 249                 ep_signal.data.fd = fd_signal;
 250
 251                 fd_monitor = udev_monitor_get_fd(worker_monitor);
 252                 memzero(&ep_monitor, sizeof(struct epoll_event));
 253                 ep_monitor.events = EPOLLIN;
 254                 ep_monitor.data.fd = fd_monitor;
 255
 256                 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
 257                     epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_monitor, &ep_monitor) < 0) {
 258                         log_error("fail to add fds to epoll: %m");
 259                         rc = 4;
 260                         goto out;
 261                 }
 262
 263                 /* request TERM signal if parent exits */
 264                 prctl(PR_SET_PDEATHSIG, SIGTERM);
 265
 266                 /* reset OOM score, we only protect the main daemon */
 267                 write_string_file("/proc/self/oom_score_adj", "0");
 268
 269                 for (;;) {
 270                         struct udev_event *udev_event;
 271                         struct worker_message msg;
 272                         int fd_lock = -1;
 273                         int err = 0;
 274
 275                         log_debug("seq %llu running", udev_device_get_seqnum(dev));
 276                         udev_event = udev_event_new(dev);
 277                         if (udev_event == NULL) {
 278                                 rc = 5;
 279                                 goto out;
 280                         }
 281
 282                         /* needed for SIGCHLD/SIGTERM in spawn() */
 283                         udev_event->fd_signal = fd_signal;
 284
 285                         if (exec_delay > 0)
 286                                 udev_event->exec_delay = exec_delay;
 287
 288                         /*
 289                          * Take a shared lock on the device node; this establishes
 290                          * a concept of device "ownership" to serialize device
 291                          * access. External processes holding an exclusive lock will
 292                          * cause udev to skip the event handling; in the case udev
 293                          * acquired the lock, the external process can block until
 294                          * udev has finished its event handling.
 295                          */
 296                         if (!streq_ptr(udev_device_get_action(dev), "remove") &&
 297                             streq_ptr("block", udev_device_get_subsystem(dev)) &&
 298                             !startswith(udev_device_get_sysname(dev), "dm-") &&
 299                             !startswith(udev_device_get_sysname(dev), "md")) {
 300                                 struct udev_device *d = dev;
 301
 302                                 if (streq_ptr("partition", udev_device_get_devtype(d)))
 303                                         d = udev_device_get_parent(d);
 304
 305                                 if (d) {
 306                                         fd_lock = open(udev_device_get_devnode(d), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
 307                                         if (fd_lock >= 0 && flock(fd_lock, LOCK_SH|LOCK_NB) < 0) {
 308                                                 log_debug("Unable to flock(%s), skipping event handling: %m", udev_device_get_devnode(d));
 309                                                 err = -EWOULDBLOCK;
 310                                                 fd_lock = safe_close(fd_lock);
 311                                                 goto skip;
 312                                         }
 313                                 }
 314                         }
 315
 316                         if (event_timeout != 30)
 317                                 udev_event->timeout_usec = event_timeout * USEC_PER_SEC;
 318
 319                         /* apply rules, create node, symlinks */
 320                         udev_event_execute_rules(udev_event, rules, &sigmask_orig);
 321
 322                         udev_event_execute_run(udev_event, &sigmask_orig);
 323
 324                         /* apply/restore inotify watch */
 325                         if (udev_event->inotify_watch) {
 326                                 udev_watch_begin(udev, dev);
 327                                 udev_device_update_db(dev);
 328                         }
 329
 330                         safe_close(fd_lock);
 331
 332                         /* send processed event back to libudev listeners */
 333                         udev_monitor_send_device(worker_monitor, NULL, dev);
 334
 335 skip:
 336                         /* send udevd the result of the event execution */
 337                         memzero(&msg, sizeof(struct worker_message));
 338                         msg.exitcode = err;
 339                         msg.pid = getpid();
 340                         send(worker_watch[WRITE_END], &msg, sizeof(struct worker_message), 0);
 341
 342                         log_debug("seq %llu processed with %i", udev_device_get_seqnum(dev), err);
 343
 344                         udev_device_unref(dev);
 345                         dev = NULL;
 346
 347                         if (udev_event->sigterm) {
 348                                 udev_event_unref(udev_event);
 349                                 goto out;
 350                         }
 351
 352                         udev_event_unref(udev_event);
 353
 354                         /* wait for more device messages from main udevd, or term signal */
 355                         while (dev == NULL) {
 356                                 struct epoll_event ev[4];
 357                                 int fdcount;
 358                                 int i;
 359
 360                                 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), -1);
 361                                 if (fdcount < 0) {
 362                                         if (errno == EINTR)
 363                                                 continue;
 364                                         log_error("failed to poll: %m");
 365                                         goto out;
 366                                 }
 367
 368                                 for (i = 0; i < fdcount; i++) {
 369                                         if (ev[i].data.fd == fd_monitor && ev[i].events & EPOLLIN) {
 370                                                 dev = udev_monitor_receive_device(worker_monitor);
 371                                                 break;
 372                                         } else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN) {
 373                                                 struct signalfd_siginfo fdsi;
 374                                                 ssize_t size;
 375
 376                                                 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
 377                                                 if (size != sizeof(struct signalfd_siginfo))
 378                                                         continue;
 379                                                 switch (fdsi.ssi_signo) {
 380                                                 case SIGTERM:
 381                                                         goto out;
 382                                                 }
 383                                         }
 384                                 }
 385                         }
 386                 }
 387 out:
 388                 udev_device_unref(dev);
 389                 safe_close(fd_signal);
 390                 safe_close(fd_ep);
 391                 close(fd_inotify);
 392                 close(worker_watch[WRITE_END]);
 393                 udev_rules_unref(rules);
 394                 udev_builtin_exit(udev);
 395                 udev_monitor_unref(worker_monitor);
 396                 udev_unref(udev);
 397                 log_close();
 398                 exit(rc);
 399         }
 400         case -1:
 401                 udev_monitor_unref(worker_monitor);
 402                 event->state = EVENT_QUEUED;
 403                 free(worker);
 404                 log_error("fork of child failed: %m");
 405                 break;
 406         default:
 407                 /* close monitor, but keep address around */
 408                 udev_monitor_disconnect(worker_monitor);
 409                 worker->monitor = worker_monitor;
 410                 worker->pid = pid;
 411                 worker->state = WORKER_RUNNING;
 412                 worker->event_start_usec = now(CLOCK_MONOTONIC);
 413                 worker->event = event;
 414                 event->state = EVENT_RUNNING;
 415                 udev_list_node_append(&worker->node, &worker_list);
 416                 children++;
 417                 log_debug("seq %llu forked new worker [%u]", udev_device_get_seqnum(event->dev), pid);
 418                 break;
 419         }
 420 }
 421
 422 static void event_run(struct event *event)
 423 {
 424         struct udev_list_node *loop;
 425
 426         udev_list_node_foreach(loop, &worker_list) {
 427                 struct worker *worker = node_to_worker(loop);
 428                 ssize_t count;
 429
 430                 if (worker->state != WORKER_IDLE)
 431                         continue;
 432
 433                 count = udev_monitor_send_device(monitor, worker->monitor, event->dev);
 434                 if (count < 0) {
 435                         log_error("worker [%u] did not accept message %zi (%m), kill it", worker->pid, count);
 436                         kill(worker->pid, SIGKILL);
 437                         worker->state = WORKER_KILLED;
 438                         continue;
 439                 }
 440                 worker_ref(worker);
 441                 worker->event = event;
 442                 worker->state = WORKER_RUNNING;
 443                 worker->event_start_usec = now(CLOCK_MONOTONIC);
 444                 event->state = EVENT_RUNNING;
 445                 return;
 446         }
 447
 448         if (children >= children_max) {
 449                 if (children_max > 1)
 450                         log_debug("maximum number (%i) of children reached", children);
 451                 return;
 452         }
 453
 454         /* start new worker and pass initial device */
 455         worker_new(event);
 456 }
 457
 458 static int event_queue_insert(struct udev_device *dev)
 459 {
 460         struct event *event;
 461
 462         event = new0(struct event, 1);
 463         if (event == NULL)
 464                 return -1;
 465
 466         event->udev = udev_device_get_udev(dev);
 467         event->dev = dev;
 468         event->seqnum = udev_device_get_seqnum(dev);
 469         event->devpath = udev_device_get_devpath(dev);
 470         event->devpath_len = strlen(event->devpath);
 471         event->devpath_old = udev_device_get_devpath_old(dev);
 472         event->devnum = udev_device_get_devnum(dev);
 473         event->is_block = streq("block", udev_device_get_subsystem(dev));
 474         event->ifindex = udev_device_get_ifindex(dev);
 475 #ifdef HAVE_FIRMWARE
 476         if (streq(udev_device_get_subsystem(dev), "firmware"))
 477                 event->nodelay = true;
 478 #endif
 479
 480         log_debug("seq %llu queued, '%s' '%s'", udev_device_get_seqnum(dev),
 481              udev_device_get_action(dev), udev_device_get_subsystem(dev));
 482
 483         event->state = EVENT_QUEUED;
 484         udev_list_node_append(&event->node, &event_list);
 485         return 0;
 486 }
 487
 488 static void worker_kill(struct udev *udev)
 489 {
 490         struct udev_list_node *loop;
 491
 492         udev_list_node_foreach(loop, &worker_list) {
 493                 struct worker *worker = node_to_worker(loop);
 494
 495                 if (worker->state == WORKER_KILLED)
 496                         continue;
 497
 498                 worker->state = WORKER_KILLED;
 499                 kill(worker->pid, SIGTERM);
 500         }
 501 }
 502
 503 /* lookup event for identical, parent, child device */
 504 static bool is_devpath_busy(struct event *event)
 505 {
 506         struct udev_list_node *loop;
 507         size_t common;
 508
 509         /* check if queue contains events we depend on */
 510         udev_list_node_foreach(loop, &event_list) {
 511                 struct event *loop_event = node_to_event(loop);
 512
 513                 /* we already found a later event, earlier can not block us, no need to check again */
 514                 if (loop_event->seqnum < event->delaying_seqnum)
 515                         continue;
 516
 517                 /* event we checked earlier still exists, no need to check again */
 518                 if (loop_event->seqnum == event->delaying_seqnum)
 519                         return true;
 520
 521                 /* found ourself, no later event can block us */
 522                 if (loop_event->seqnum >= event->seqnum)
 523                         break;
 524
 525                 /* check major/minor */
 526                 if (major(event->devnum) != 0 && event->devnum == loop_event->devnum && event->is_block == loop_event->is_block)
 527                         return true;
 528
 529                 /* check network device ifindex */
 530                 if (event->ifindex != 0 && event->ifindex == loop_event->ifindex)
 531                         return true;
 532
 533                 /* check our old name */
 534                 if (event->devpath_old != NULL && streq(loop_event->devpath, event->devpath_old)) {
 535                         event->delaying_seqnum = loop_event->seqnum;
 536                         return true;
 537                 }
 538
 539                 /* compare devpath */
 540                 common = MIN(loop_event->devpath_len, event->devpath_len);
 541
 542                 /* one devpath is contained in the other? */
 543                 if (memcmp(loop_event->devpath, event->devpath, common) != 0)
 544                         continue;
 545
 546                 /* identical device event found */
 547                 if (loop_event->devpath_len == event->devpath_len) {
 548                         /* devices names might have changed/swapped in the meantime */
 549                         if (major(event->devnum) != 0 && (event->devnum != loop_event->devnum || event->is_block != loop_event->is_block))
 550                                 continue;
 551                         if (event->ifindex != 0 && event->ifindex != loop_event->ifindex)
 552                                 continue;
 553                         event->delaying_seqnum = loop_event->seqnum;
 554                         return true;
 555                 }
 556
 557 #ifdef HAVE_FIRMWARE
 558                 /* allow to bypass the dependency tracking */
 559                 if (event->nodelay)
 560                         continue;
 561 #endif
 562
 563                 /* parent device event found */
 564                 if (event->devpath[common] == '/') {
 565                         event->delaying_seqnum = loop_event->seqnum;
 566                         return true;
 567                 }
 568
 569                 /* child device event found */
 570                 if (loop_event->devpath[common] == '/') {
 571                         event->delaying_seqnum = loop_event->seqnum;
 572                         return true;
 573                 }
 574
 575                 /* no matching device */
 576                 continue;
 577         }
 578
 579         return false;
 580 }
 581
 582 static void event_queue_start(struct udev *udev)
 583 {
 584         struct udev_list_node *loop;
 585
 586         udev_list_node_foreach(loop, &event_list) {
 587                 struct event *event = node_to_event(loop);
 588
 589                 if (event->state != EVENT_QUEUED)
 590                         continue;
 591
 592                 /* do not start event if parent or child event is still running */
 593                 if (is_devpath_busy(event))
 594                         continue;
 595
 596                 event_run(event);
 597         }
 598 }
 599
 600 static void event_queue_cleanup(struct udev *udev, enum event_state match_type)
 601 {
 602         struct udev_list_node *loop, *tmp;
 603
 604         udev_list_node_foreach_safe(loop, tmp, &event_list) {
 605                 struct event *event = node_to_event(loop);
 606
 607                 if (match_type != EVENT_UNDEF && match_type != event->state)
 608                         continue;
 609
 610                 event_queue_delete(event);
 611         }
 612 }
 613
 614 static void worker_returned(int fd_worker)
 615 {
 616         for (;;) {
 617                 struct worker_message msg;
 618                 ssize_t size;
 619                 struct udev_list_node *loop;
 620
 621                 size = recv(fd_worker, &msg, sizeof(struct worker_message), MSG_DONTWAIT);
 622                 if (size != sizeof(struct worker_message))
 623                         break;
 624
 625                 /* lookup worker who sent the signal */
 626                 udev_list_node_foreach(loop, &worker_list) {
 627                         struct worker *worker = node_to_worker(loop);
 628
 629                         if (worker->pid != msg.pid)
 630                                 continue;
 631
 632                         /* worker returned */
 633                         if (worker->event) {
 634                                 worker->event->exitcode = msg.exitcode;
 635                                 event_queue_delete(worker->event);
 636                                 worker->event = NULL;
 637                         }
 638                         if (worker->state != WORKER_KILLED)
 639                                 worker->state = WORKER_IDLE;
 640                         worker_unref(worker);
 641                         break;
 642                 }
 643         }
 644 }
 645
 646 /* receive the udevd message from userspace */
 647 static struct udev_ctrl_connection *handle_ctrl_msg(struct udev_ctrl *uctrl)
 648 {
 649         struct udev *udev = udev_ctrl_get_udev(uctrl);
 650         struct udev_ctrl_connection *ctrl_conn;
 651         struct udev_ctrl_msg *ctrl_msg = NULL;
 652         const char *str;
 653         int i;
 654
 655         ctrl_conn = udev_ctrl_get_connection(uctrl);
 656         if (ctrl_conn == NULL)
 657                 goto out;
 658
 659         ctrl_msg = udev_ctrl_receive_msg(ctrl_conn);
 660         if (ctrl_msg == NULL)
 661                 goto out;
 662
 663         i = udev_ctrl_get_set_log_level(ctrl_msg);
 664         if (i >= 0) {
 665                 log_debug("udevd message (SET_LOG_LEVEL) received, log_priority=%i", i);
 666                 log_set_max_level(i);
 667                 udev_set_log_priority(udev, i);
 668                 worker_kill(udev);
 669         }
 670
 671         if (udev_ctrl_get_stop_exec_queue(ctrl_msg) > 0) {
 672                 log_debug("udevd message (STOP_EXEC_QUEUE) received");
 673                 stop_exec_queue = true;
 674         }
 675
 676         if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) {
 677                 log_debug("udevd message (START_EXEC_QUEUE) received");
 678                 stop_exec_queue = false;
 679         }
 680
 681         if (udev_ctrl_get_reload(ctrl_msg) > 0) {
 682                 log_debug("udevd message (RELOAD) received");
 683                 reload = true;
 684         }
 685
 686         str = udev_ctrl_get_set_env(ctrl_msg);
 687         if (str != NULL) {
 688                 char *key;
 689
 690                 key = strdup(str);
 691                 if (key != NULL) {
 692                         char *val;
 693
 694                         val = strchr(key, '=');
 695                         if (val != NULL) {
 696                                 val[0] = '\0';
 697                                 val = &val[1];
 698                                 if (val[0] == '\0') {
 699                                         log_debug("udevd message (ENV) received, unset '%s'", key);
 700                                         udev_add_property(udev, key, NULL);
 701                                 } else {
 702                                         log_debug("udevd message (ENV) received, set '%s=%s'", key, val);
 703                                         udev_add_property(udev, key, val);
 704                                 }
 705                         } else {
 706                                 log_error("wrong key format '%s'", key);
 707                         }
 708                         free(key);
 709                 }
 710                 worker_kill(udev);
 711         }
 712
 713         i = udev_ctrl_get_set_children_max(ctrl_msg);
 714         if (i >= 0) {
 715                 log_debug("udevd message (SET_MAX_CHILDREN) received, children_max=%i", i);
 716                 children_max = i;
 717         }
 718
 719         if (udev_ctrl_get_ping(ctrl_msg) > 0)
 720                 log_debug("udevd message (SYNC) received");
 721
 722         if (udev_ctrl_get_exit(ctrl_msg) > 0) {
 723                 log_debug("udevd message (EXIT) received");
 724                 udev_exit = true;
 725                 /* keep reference to block the client until we exit */
 726                 udev_ctrl_connection_ref(ctrl_conn);
 727         }
 728 out:
 729         udev_ctrl_msg_unref(ctrl_msg);
 730         return udev_ctrl_connection_unref(ctrl_conn);
 731 }
 732
 733 static int synthesize_change(struct udev_device *dev) {
 734         char filename[UTIL_PATH_SIZE];
 735         int r;
 736
 737         if (streq_ptr("block", udev_device_get_subsystem(dev)) &&
 738             streq_ptr("disk", udev_device_get_devtype(dev)) &&
 739             !startswith(udev_device_get_sysname(dev), "dm-")) {
 740                 bool part_table_read = false;
 741                 bool has_partitions = false;
 742                 int fd;
 743                 struct udev *udev = udev_device_get_udev(dev);
 744                 _cleanup_udev_enumerate_unref_ struct udev_enumerate *e = NULL;
 745                 struct udev_list_entry *item;
 746
 747                 /*
 748                  * Try to re-read the partition table. This only succeeds if
 749                  * none of the devices is busy. The kernel returns 0 if no
 750                  * partition table is found, and we will not get an event for
 751                  * the disk.
 752                  */
 753                 fd = open(udev_device_get_devnode(dev), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
 754                 if (fd >= 0) {
 755                         r = flock(fd, LOCK_EX|LOCK_NB);
 756                         if (r >= 0)
 757                                 r = ioctl(fd, BLKRRPART, 0);
 758
 759                         close(fd);
 760                         if (r >= 0)
 761                                 part_table_read = true;
 762                 }
 763
 764                 /* search for partitions */
 765                 e = udev_enumerate_new(udev);
 766                 if (!e)
 767                         return -ENOMEM;
 768
 769                 r = udev_enumerate_add_match_parent(e, dev);
 770                 if (r < 0)
 771                         return r;
 772
 773                 r = udev_enumerate_add_match_subsystem(e, "block");
 774                 if (r < 0)
 775                         return r;
 776
 777                 r = udev_enumerate_scan_devices(e);
 778                 if (r < 0)
 779                         return r;
 780
 781                 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
 782                         _cleanup_udev_device_unref_ struct udev_device *d = NULL;
 783
 784                         d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
 785                         if (!d)
 786                                 continue;
 787
 788                         if (!streq_ptr("partition", udev_device_get_devtype(d)))
 789                                 continue;
 790
 791                         has_partitions = true;
 792                         break;
 793                 }
 794
 795                 /*
 796                  * We have partitions and re-read the table, the kernel already sent
 797                  * out a "change" event for the disk, and "remove/add" for all
 798                  * partitions.
 799                  */
 800                 if (part_table_read && has_partitions)
 801                         return 0;
 802
 803                 /*
 804                  * We have partitions but re-reading the partition table did not
 805                  * work, synthesize "change" for the disk and all partitions.
 806                  */
 807                 log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
 808                 strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
 809                 write_string_file(filename, "change");
 810
 811                 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
 812                         _cleanup_udev_device_unref_ struct udev_device *d = NULL;
 813
 814                         d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
 815                         if (!d)
 816                                 continue;
 817
 818                         if (!streq_ptr("partition", udev_device_get_devtype(d)))
 819                                 continue;
 820
 821                         log_debug("device %s closed, synthesising partition '%s' 'change'",
 822                                   udev_device_get_devnode(dev), udev_device_get_devnode(d));
 823                         strscpyl(filename, sizeof(filename), udev_device_get_syspath(d), "/uevent", NULL);
 824                         write_string_file(filename, "change");
 825                 }
 826
 827                 return 0;
 828         }
 829
 830         log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
 831         strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
 832         write_string_file(filename, "change");
 833
 834         return 0;
 835 }
 836
 837 static int handle_inotify(struct udev *udev)
 838 {
 839         int nbytes, pos;
 840         char *buf;
 841         struct inotify_event *ev;
 842         int r;
 843
 844         r = ioctl(fd_inotify, FIONREAD, &nbytes);
 845         if (r < 0 || nbytes <= 0)
 846                 return -errno;
 847
 848         buf = malloc(nbytes);
 849         if (!buf) {
 850                 log_error("error getting buffer for inotify");
 851                 return -ENOMEM;
 852         }
 853
 854         nbytes = read(fd_inotify, buf, nbytes);
 855
 856         for (pos = 0; pos < nbytes; pos += sizeof(struct inotify_event) + ev->len) {
 857                 struct udev_device *dev;
 858
 859                 ev = (struct inotify_event *)(buf + pos);
 860                 dev = udev_watch_lookup(udev, ev->wd);
 861                 if (!dev)
 862                         continue;
 863
 864                 log_debug("inotify event: %x for %s", ev->mask, udev_device_get_devnode(dev));
 865                 if (ev->mask & IN_CLOSE_WRITE)
 866                         synthesize_change(dev);
 867                 else if (ev->mask & IN_IGNORED)
 868                         udev_watch_end(udev, dev);
 869
 870                 udev_device_unref(dev);
 871         }
 872
 873         free(buf);
 874         return 0;
 875 }
 876
 877 static void handle_signal(struct udev *udev, int signo)
 878 {
 879         switch (signo) {
 880         case SIGINT:
 881         case SIGTERM:
 882                 udev_exit = true;
 883                 break;
 884         case SIGCHLD:
 885                 for (;;) {
 886                         pid_t pid;
 887                         int status;
 888                         struct udev_list_node *loop, *tmp;
 889
 890                         pid = waitpid(-1, &status, WNOHANG);
 891                         if (pid <= 0)
 892                                 break;
 893
 894                         udev_list_node_foreach_safe(loop, tmp, &worker_list) {
 895                                 struct worker *worker = node_to_worker(loop);
 896
 897                                 if (worker->pid != pid)
 898                                         continue;
 899                                 log_debug("worker [%u] exit", pid);
 900
 901                                 if (WIFEXITED(status)) {
 902                                         if (WEXITSTATUS(status) != 0)
 903                                                 log_error("worker [%u] exit with return code %i",
 904                                                           pid, WEXITSTATUS(status));
 905                                 } else if (WIFSIGNALED(status)) {
 906                                         log_error("worker [%u] terminated by signal %i (%s)",
 907                                                   pid, WTERMSIG(status), strsignal(WTERMSIG(status)));
 908                                 } else if (WIFSTOPPED(status)) {
 909                                         log_error("worker [%u] stopped", pid);
 910                                 } else if (WIFCONTINUED(status)) {
 911                                         log_error("worker [%u] continued", pid);
 912                                 } else {
 913                                         log_error("worker [%u] exit with status 0x%04x", pid, status);
 914                                 }
 915
 916                                 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 917                                         if (worker->event) {
 918                                                 log_error("worker [%u] failed while handling '%s'",
 919                                                           pid, worker->event->devpath);
 920                                                 worker->event->exitcode = -32;
 921                                                 event_queue_delete(worker->event);
 922
 923                                                 /* drop reference taken for state 'running' */
 924                                                 worker_unref(worker);
 925                                         }
 926                                 }
 927                                 worker_unref(worker);
 928                                 break;
 929                         }
 930                 }
 931                 break;
 932         case SIGHUP:
 933                 reload = true;
 934                 break;
 935         }
 936 }
 937
 938 static int systemd_fds(struct udev *udev, int *rctrl, int *rnetlink)
 939 {
 940         int ctrl = -1, netlink = -1;
 941         int fd, n;
 942
 943         n = sd_listen_fds(true);
 944         if (n <= 0)
 945                 return -1;
 946
 947         for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
 948                 if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1)) {
 949                         if (ctrl >= 0)
 950                                 return -1;
 951                         ctrl = fd;
 952                         continue;
 953                 }
 954
 955                 if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1)) {
 956                         if (netlink >= 0)
 957                                 return -1;
 958                         netlink = fd;
 959                         continue;
 960                 }
 961
 962                 return -1;
 963         }
 964
 965         if (ctrl < 0 || netlink < 0)
 966                 return -1;
 967
 968         log_debug("ctrl=%i netlink=%i", ctrl, netlink);
 969         *rctrl = ctrl;
 970         *rnetlink = netlink;
 971         return 0;
 972 }
 973
 974 /*
 975  * read the kernel commandline, in case we need to get into debug mode
 976  *   udev.log-priority=<level>              syslog priority
 977  *   udev.children-max=<number of workers>  events are fully serialized if set to 1
 978  *   udev.exec-delay=<number of seconds>    delay execution of every executed program
 979  */
 980 static void kernel_cmdline_options(struct udev *udev)
 981 {
 982         _cleanup_free_ char *line = NULL;
 983         char *w, *state;
 984         size_t l;
 985         int r;
 986
 987         r = proc_cmdline(&line);
 988         if (r < 0)
 989                 log_warning("Failed to read /proc/cmdline, ignoring: %s", strerror(-r));
 990         if (r <= 0)
 991                 return;
 992
 993         FOREACH_WORD_QUOTED(w, l, line, state) {
 994                 char *s, *opt;
 995
 996                 s = strndup(w, l);
 997                 if (!s)
 998                         break;
 999
1000                 /* accept the same options for the initrd, prefixed with "rd." */
1001                 if (in_initrd() && startswith(s, "rd."))
1002                         opt = s + 3;
1003                 else
1004                         opt = s;
1005
1006                 if (startswith(opt, "udev.log-priority=")) {
1007                         int prio;
1008
1009                         prio = util_log_priority(opt + 18);
1010                         log_set_max_level(prio);
1011                         udev_set_log_priority(udev, prio);
1012                 } else if (startswith(opt, "udev.children-max=")) {
1013                         children_max = strtoul(opt + 18, NULL, 0);
1014                 } else if (startswith(opt, "udev.exec-delay=")) {
1015                         exec_delay = strtoul(opt + 16, NULL, 0);
1016                 } else if (startswith(opt, "udev.event-timeout=")) {
1017                         event_timeout = strtoul(opt + 16, NULL, 0);
1018                 }
1019
1020                 free(s);
1021         }
1022 }
1023
1024 int main(int argc, char *argv[])
1025 {
1026         struct udev *udev;
1027         sigset_t mask;
1028         int daemonize = false;
1029         int resolve_names = 1;
1030         static const struct option options[] = {
1031                 { "daemon", no_argument, NULL, 'd' },
1032                 { "debug", no_argument, NULL, 'D' },
1033                 { "children-max", required_argument, NULL, 'c' },
1034                 { "exec-delay", required_argument, NULL, 'e' },
1035                 { "event-timeout", required_argument, NULL, 't' },
1036                 { "resolve-names", required_argument, NULL, 'N' },
1037                 { "help", no_argument, NULL, 'h' },
1038                 { "version", no_argument, NULL, 'V' },
1039                 {}
1040         };
1041         int fd_ctrl = -1;
1042         int fd_netlink = -1;
1043         int fd_worker = -1;
1044         struct epoll_event ep_ctrl, ep_inotify, ep_signal, ep_netlink, ep_worker;
1045         struct udev_ctrl_connection *ctrl_conn = NULL;
1046         int rc = 1;
1047
1048         udev = udev_new();
1049         if (udev == NULL)
1050                 goto exit;
1051
1052         log_set_target(LOG_TARGET_AUTO);
1053         log_parse_environment();
1054         log_open();
1055
1056         udev_set_log_fn(udev, udev_main_log);
1057         log_set_max_level(udev_get_log_priority(udev));
1058
1059         log_debug("version %s", VERSION);
1060         label_init("/dev");
1061
1062         for (;;) {
1063                 int option;
1064
1065                 option = getopt_long(argc, argv, "c:de:DtN:hV", options, NULL);
1066                 if (option == -1)
1067                         break;
1068
1069                 switch (option) {
1070                 case 'd':
1071                         daemonize = true;
1072                         break;
1073                 case 'c':
1074                         children_max = strtoul(optarg, NULL, 0);
1075                         break;
1076                 case 'e':
1077                         exec_delay = strtoul(optarg, NULL, 0);
1078                         break;
1079                 case 't':
1080                         event_timeout = strtoul(optarg, NULL, 0);
1081                         break;
1082                 case 'D':
1083                         debug = true;
1084                         log_set_max_level(LOG_DEBUG);
1085                         udev_set_log_priority(udev, LOG_DEBUG);
1086                         break;
1087                 case 'N':
1088                         if (streq(optarg, "early")) {
1089                                 resolve_names = 1;
1090                         } else if (streq(optarg, "late")) {
1091                                 resolve_names = 0;
1092                         } else if (streq(optarg, "never")) {
1093                                 resolve_names = -1;
1094                         } else {
1095                                 fprintf(stderr, "resolve-names must be early, late or never\n");
1096                                 log_error("resolve-names must be early, late or never");
1097                                 goto exit;
1098                         }
1099                         break;
1100                 case 'h':
1101                         printf("Usage: udevd OPTIONS\n"
1102                                "  --daemon\n"
1103                                "  --debug\n"
1104                                "  --children-max=<maximum number of workers>\n"
1105                                "  --exec-delay=<seconds to wait before executing RUN=>\n"
1106                                "  --resolve-names=early|late|never\n"
1107                                "  --version\n"
1108                                "  --help\n"
1109                                "\n");
1110                         goto exit;
1111                 case 'V':
1112                         printf("%s\n", VERSION);
1113                         goto exit;
1114                 default:
1115                         goto exit;
1116                 }
1117         }
1118
1119         kernel_cmdline_options(udev);
1120
1121         if (getuid() != 0) {
1122                 fprintf(stderr, "root privileges required\n");
1123                 log_error("root privileges required");
1124                 goto exit;
1125         }
1126
1127         /* set umask before creating any file/directory */
1128         chdir("/");
1129         umask(022);
1130
1131         mkdir("/run/udev", 0755);
1132
1133         dev_setup(NULL);
1134
1135         /* before opening new files, make sure std{in,out,err} fds are in a sane state */
1136         if (daemonize) {
1137                 int fd;
1138
1139                 fd = open("/dev/null", O_RDWR);
1140                 if (fd >= 0) {
1141                         if (write(STDOUT_FILENO, 0, 0) < 0)
1142                                 dup2(fd, STDOUT_FILENO);
1143                         if (write(STDERR_FILENO, 0, 0) < 0)
1144                                 dup2(fd, STDERR_FILENO);
1145                         if (fd > STDERR_FILENO)
1146                                 close(fd);
1147                 } else {
1148                         fprintf(stderr, "cannot open /dev/null\n");
1149                         log_error("cannot open /dev/null");
1150                 }
1151         }
1152
1153         if (systemd_fds(udev, &fd_ctrl, &fd_netlink) >= 0) {
1154                 /* get control and netlink socket from systemd */
1155                 udev_ctrl = udev_ctrl_new_from_fd(udev, fd_ctrl);
1156                 if (udev_ctrl == NULL) {
1157                         log_error("error taking over udev control socket");
1158                         rc = 1;
1159                         goto exit;
1160                 }
1161
1162                 monitor = udev_monitor_new_from_netlink_fd(udev, "kernel", fd_netlink);
1163                 if (monitor == NULL) {
1164                         log_error("error taking over netlink socket");
1165                         rc = 3;
1166                         goto exit;
1167                 }
1168
1169                 /* get our own cgroup, we regularly kill everything udev has left behind */
1170                 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &udev_cgroup) < 0)
1171                         udev_cgroup = NULL;
1172         } else {
1173                 /* open control and netlink socket */
1174                 udev_ctrl = udev_ctrl_new(udev);
1175                 if (udev_ctrl == NULL) {
1176                         fprintf(stderr, "error initializing udev control socket");
1177                         log_error("error initializing udev control socket");
1178                         rc = 1;
1179                         goto exit;
1180                 }
1181                 fd_ctrl = udev_ctrl_get_fd(udev_ctrl);
1182
1183                 monitor = udev_monitor_new_from_netlink(udev, "kernel");
1184                 if (monitor == NULL) {
1185                         fprintf(stderr, "error initializing netlink socket\n");
1186                         log_error("error initializing netlink socket");
1187                         rc = 3;
1188                         goto exit;
1189                 }
1190                 fd_netlink = udev_monitor_get_fd(monitor);
1191         }
1192
1193         if (udev_monitor_enable_receiving(monitor) < 0) {
1194                 fprintf(stderr, "error binding netlink socket\n");
1195                 log_error("error binding netlink socket");
1196                 rc = 3;
1197                 goto exit;
1198         }
1199
1200         if (udev_ctrl_enable_receiving(udev_ctrl) < 0) {
1201                 fprintf(stderr, "error binding udev control socket\n");
1202                 log_error("error binding udev control socket");
1203                 rc = 1;
1204                 goto exit;
1205         }
1206
1207         udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024);
1208
1209         if (daemonize) {
1210                 pid_t pid;
1211
1212                 pid = fork();
1213                 switch (pid) {
1214                 case 0:
1215                         break;
1216                 case -1:
1217                         log_error("fork of daemon failed: %m");
1218                         rc = 4;
1219                         goto exit;
1220                 default:
1221                         rc = EXIT_SUCCESS;
1222                         goto exit_daemonize;
1223                 }
1224
1225                 setsid();
1226
1227                 write_string_file("/proc/self/oom_score_adj", "-1000");
1228         } else {
1229                 sd_notify(1, "READY=1");
1230         }
1231
1232         print_kmsg("starting version " VERSION "\n");
1233
1234         if (!debug) {
1235                 int fd;
1236
1237                 fd = open("/dev/null", O_RDWR);
1238                 if (fd >= 0) {
1239                         dup2(fd, STDIN_FILENO);
1240                         dup2(fd, STDOUT_FILENO);
1241                         dup2(fd, STDERR_FILENO);
1242                         close(fd);
1243                 }
1244         }
1245
1246         fd_inotify = udev_watch_init(udev);
1247         if (fd_inotify < 0) {
1248                 fprintf(stderr, "error initializing inotify\n");
1249                 log_error("error initializing inotify");
1250                 rc = 4;
1251                 goto exit;
1252         }
1253         udev_watch_restore(udev);
1254
1255         /* block and listen to all signals on signalfd */
1256         sigfillset(&mask);
1257         sigprocmask(SIG_SETMASK, &mask, &sigmask_orig);
1258         fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1259         if (fd_signal < 0) {
1260                 fprintf(stderr, "error creating signalfd\n");
1261                 log_error("error creating signalfd");
1262                 rc = 5;
1263                 goto exit;
1264         }
1265
1266         /* unnamed socket from workers to the main daemon */
1267         if (socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, worker_watch) < 0) {
1268                 fprintf(stderr, "error creating socketpair\n");
1269                 log_error("error creating socketpair");
1270                 rc = 6;
1271                 goto exit;
1272         }
1273         fd_worker = worker_watch[READ_END];
1274
1275         udev_builtin_init(udev);
1276
1277         rules = udev_rules_new(udev, resolve_names);
1278         if (rules == NULL) {
1279                 log_error("error reading rules");
1280                 goto exit;
1281         }
1282
1283         memzero(&ep_ctrl, sizeof(struct epoll_event));
1284         ep_ctrl.events = EPOLLIN;
1285         ep_ctrl.data.fd = fd_ctrl;
1286
1287         memzero(&ep_inotify, sizeof(struct epoll_event));
1288         ep_inotify.events = EPOLLIN;
1289         ep_inotify.data.fd = fd_inotify;
1290
1291         memzero(&ep_signal, sizeof(struct epoll_event));
1292         ep_signal.events = EPOLLIN;
1293         ep_signal.data.fd = fd_signal;
1294
1295         memzero(&ep_netlink, sizeof(struct epoll_event));
1296         ep_netlink.events = EPOLLIN;
1297         ep_netlink.data.fd = fd_netlink;
1298
1299         memzero(&ep_worker, sizeof(struct epoll_event));
1300         ep_worker.events = EPOLLIN;
1301         ep_worker.data.fd = fd_worker;
1302
1303         fd_ep = epoll_create1(EPOLL_CLOEXEC);
1304         if (fd_ep < 0) {
1305                 log_error("error creating epoll fd: %m");
1306                 goto exit;
1307         }
1308         if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_ctrl, &ep_ctrl) < 0 ||
1309             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_inotify, &ep_inotify) < 0 ||
1310             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
1311             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_netlink, &ep_netlink) < 0 ||
1312             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_worker, &ep_worker) < 0) {
1313                 log_error("fail to add fds to epoll: %m");
1314                 goto exit;
1315         }
1316
1317         if (children_max <= 0) {
1318                 cpu_set_t cpu_set;
1319
1320                 children_max = 8;
1321
1322                 if (sched_getaffinity(0, sizeof (cpu_set), &cpu_set) == 0) {
1323                         children_max +=  CPU_COUNT(&cpu_set) * 2;
1324                 }
1325         }
1326         log_debug("set children_max to %u", children_max);
1327
1328         rc = udev_rules_apply_static_dev_perms(rules);
1329         if (rc < 0)
1330                 log_error("failed to apply permissions on static device nodes - %s", strerror(-rc));
1331
1332         udev_list_node_init(&event_list);
1333         udev_list_node_init(&worker_list);
1334
1335         for (;;) {
1336                 static usec_t last_usec;
1337                 struct epoll_event ev[8];
1338                 int fdcount;
1339                 int timeout;
1340                 bool is_worker, is_signal, is_inotify, is_netlink, is_ctrl;
1341                 int i;
1342
1343                 if (udev_exit) {
1344                         /* close sources of new events and discard buffered events */
1345                         if (fd_ctrl >= 0) {
1346                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_ctrl, NULL);
1347                                 fd_ctrl = -1;
1348                         }
1349                         if (monitor != NULL) {
1350                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_netlink, NULL);
1351                                 udev_monitor_unref(monitor);
1352                                 monitor = NULL;
1353                         }
1354                         if (fd_inotify >= 0) {
1355                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_inotify, NULL);
1356                                 close(fd_inotify);
1357                                 fd_inotify = -1;
1358                         }
1359
1360                         /* discard queued events and kill workers */
1361                         event_queue_cleanup(udev, EVENT_QUEUED);
1362                         worker_kill(udev);
1363
1364                         /* exit after all has cleaned up */
1365                         if (udev_list_node_is_empty(&event_list) && children == 0)
1366                                 break;
1367
1368                         /* timeout at exit for workers to finish */
1369                         timeout = 30 * MSEC_PER_SEC;
1370                 } else if (udev_list_node_is_empty(&event_list) && children == 0) {
1371                         /* we are idle */
1372                         timeout = -1;
1373
1374                         /* cleanup possible left-over processes in our cgroup */
1375                         if (udev_cgroup)
1376                                 cg_kill(SYSTEMD_CGROUP_CONTROLLER, udev_cgroup, SIGKILL, false, true, NULL);
1377                 } else {
1378                         /* kill idle or hanging workers */
1379                         timeout = 3 * MSEC_PER_SEC;
1380                 }
1381
1382                 /* tell settle that we are busy or idle */
1383                 if (!udev_list_node_is_empty(&event_list)) {
1384                         int fd;
1385
1386                         fd = open("/run/udev/queue", O_WRONLY|O_CREAT|O_CLOEXEC|O_TRUNC|O_NOFOLLOW, 0444);
1387                         if (fd >= 0)
1388                                 close(fd);
1389                 } else {
1390                         unlink("/run/udev/queue");
1391                 }
1392
1393                 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), timeout);
1394                 if (fdcount < 0)
1395                         continue;
1396
1397                 if (fdcount == 0) {
1398                         struct udev_list_node *loop;
1399
1400                         /* timeout */
1401                         if (udev_exit) {
1402                                 log_error("timeout, giving up waiting for workers to finish");
1403                                 break;
1404                         }
1405
1406                         /* kill idle workers */
1407                         if (udev_list_node_is_empty(&event_list)) {
1408                                 log_debug("cleanup idle workers");
1409                                 worker_kill(udev);
1410                         }
1411
1412                         /* check for hanging events */
1413                         udev_list_node_foreach(loop, &worker_list) {
1414                                 struct worker *worker = node_to_worker(loop);
1415
1416                                 if (worker->state != WORKER_RUNNING)
1417                                         continue;
1418
1419                                 if ((now(CLOCK_MONOTONIC) - worker->event_start_usec) > event_timeout * USEC_PER_SEC) {
1420                                         log_error("worker [%u] %s timeout; kill it", worker->pid,
1421                                             worker->event ? worker->event->devpath : "<idle>");
1422                                         kill(worker->pid, SIGKILL);
1423                                         worker->state = WORKER_KILLED;
1424
1425                                         /* drop reference taken for state 'running' */
1426                                         worker_unref(worker);
1427                                         if (worker->event) {
1428                                                 log_error("seq %llu '%s' killed", udev_device_get_seqnum(worker->event->dev), worker->event->devpath);
1429                                                 worker->event->exitcode = -64;
1430                                                 event_queue_delete(worker->event);
1431                                                 worker->event = NULL;
1432                                         }
1433                                 }
1434                         }
1435
1436                 }
1437
1438                 is_worker = is_signal = is_inotify = is_netlink = is_ctrl = false;
1439                 for (i = 0; i < fdcount; i++) {
1440                         if (ev[i].data.fd == fd_worker && ev[i].events & EPOLLIN)
1441                                 is_worker = true;
1442                         else if (ev[i].data.fd == fd_netlink && ev[i].events & EPOLLIN)
1443                                 is_netlink = true;
1444                         else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN)
1445                                 is_signal = true;
1446                         else if (ev[i].data.fd == fd_inotify && ev[i].events & EPOLLIN)
1447                                 is_inotify = true;
1448                         else if (ev[i].data.fd == fd_ctrl && ev[i].events & EPOLLIN)
1449                                 is_ctrl = true;
1450                 }
1451
1452                 /* check for changed config, every 3 seconds at most */
1453                 if ((now(CLOCK_MONOTONIC) - last_usec) > 3 * USEC_PER_SEC) {
1454                         if (udev_rules_check_timestamp(rules))
1455                                 reload = true;
1456                         if (udev_builtin_validate(udev))
1457                                 reload = true;
1458
1459                         last_usec = now(CLOCK_MONOTONIC);
1460                 }
1461
1462                 /* reload requested, HUP signal received, rules changed, builtin changed */
1463                 if (reload) {
1464                         worker_kill(udev);
1465                         rules = udev_rules_unref(rules);
1466                         udev_builtin_exit(udev);
1467                         reload = false;
1468                 }
1469
1470                 /* event has finished */
1471                 if (is_worker)
1472                         worker_returned(fd_worker);
1473
1474                 if (is_netlink) {
1475                         struct udev_device *dev;
1476
1477                         dev = udev_monitor_receive_device(monitor);
1478                         if (dev != NULL) {
1479                                 udev_device_set_usec_initialized(dev, now(CLOCK_MONOTONIC));
1480                                 if (event_queue_insert(dev) < 0)
1481                                         udev_device_unref(dev);
1482                         }
1483                 }
1484
1485                 /* start new events */
1486                 if (!udev_list_node_is_empty(&event_list) && !udev_exit && !stop_exec_queue) {
1487                         udev_builtin_init(udev);
1488                         if (rules == NULL)
1489                                 rules = udev_rules_new(udev, resolve_names);
1490                         if (rules != NULL)
1491                                 event_queue_start(udev);
1492                 }
1493
1494                 if (is_signal) {
1495                         struct signalfd_siginfo fdsi;
1496                         ssize_t size;
1497
1498                         size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
1499                         if (size == sizeof(struct signalfd_siginfo))
1500                                 handle_signal(udev, fdsi.ssi_signo);
1501                 }
1502
1503                 /* we are shutting down, the events below are not handled anymore */
1504                 if (udev_exit)
1505                         continue;
1506
1507                 /* device node watch */
1508                 if (is_inotify)
1509                         handle_inotify(udev);
1510
1511                 /*
1512                  * This needs to be after the inotify handling, to make sure,
1513                  * that the ping is send back after the possibly generated
1514                  * "change" events by the inotify device node watch.
1515                  *
1516                  * A single time we may receive a client connection which we need to
1517                  * keep open to block the client. It will be closed right before we
1518                  * exit.
1519                  */
1520                 if (is_ctrl)
1521                         ctrl_conn = handle_ctrl_msg(udev_ctrl);
1522         }
1523
1524         rc = EXIT_SUCCESS;
1525 exit:
1526         udev_ctrl_cleanup(udev_ctrl);
1527         unlink("/run/udev/queue");
1528 exit_daemonize:
1529         if (fd_ep >= 0)
1530                 close(fd_ep);
1531         worker_list_cleanup(udev);
1532         event_queue_cleanup(udev, EVENT_UNDEF);
1533         udev_rules_unref(rules);
1534         udev_builtin_exit(udev);
1535         if (fd_signal >= 0)
1536                 close(fd_signal);
1537         if (worker_watch[READ_END] >= 0)
1538                 close(worker_watch[READ_END]);
1539         if (worker_watch[WRITE_END] >= 0)
1540                 close(worker_watch[WRITE_END]);
1541         udev_monitor_unref(monitor);
1542         udev_ctrl_connection_unref(ctrl_conn);
1543         udev_ctrl_unref(udev_ctrl);
1544         label_finish();
1545         udev_unref(udev);
1546         log_close();
1547         return rc;
1548 }