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