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