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 "rtnl-util.h"
  52 #include "sd-daemon.h"
  53 #include "cgroup-util.h"
  54 #include "dev-setup.h"
  55 #include "fileio.h"
  56
  57 static struct udev_rules *rules;
  58 static struct udev_ctrl *udev_ctrl;
  59 static struct udev_monitor *monitor;
  60 static int worker_watch[2] = { -1, -1 };
  61 static int fd_signal = -1;
  62 static int fd_ep = -1;
  63 static int fd_inotify = -1;
  64 static bool stop_exec_queue;
  65 static bool reload;
  66 static int children;
  67 static bool arg_debug = false;
  68 static int arg_daemonize = false;
  69 static int arg_resolve_names = 1;
  70 static int arg_children_max;
  71 static int arg_exec_delay;
  72 static usec_t arg_event_timeout_usec = 180 * USEC_PER_SEC;
  73 static usec_t arg_event_timeout_warn_usec = 180 * USEC_PER_SEC / 3;
  74 static sigset_t sigmask_orig;
  75 static UDEV_LIST(event_list);
  76 static UDEV_LIST(worker_list);
  77 static char *udev_cgroup;
  78 static struct udev_list properties_list;
  79 static bool udev_exit;
  80
  81 enum event_state {
  82         EVENT_UNDEF,
  83         EVENT_QUEUED,
  84         EVENT_RUNNING,
  85 };
  86
  87 struct event {
  88         struct udev_list_node node;
  89         struct udev *udev;
  90         struct udev_device *dev;
  91         enum event_state state;
  92         int exitcode;
  93         unsigned long long int delaying_seqnum;
  94         unsigned long long int seqnum;
  95         const char *devpath;
  96         size_t devpath_len;
  97         const char *devpath_old;
  98         dev_t devnum;
  99         int ifindex;
 100         bool is_block;
 101 };
 102
 103 static inline struct event *node_to_event(struct udev_list_node *node) {
 104         return container_of(node, struct event, node);
 105 }
 106
 107 static void event_queue_cleanup(struct udev *udev, enum event_state type);
 108
 109 enum worker_state {
 110         WORKER_UNDEF,
 111         WORKER_RUNNING,
 112         WORKER_IDLE,
 113         WORKER_KILLED,
 114 };
 115
 116 struct worker {
 117         struct udev_list_node node;
 118         struct udev *udev;
 119         int refcount;
 120         pid_t pid;
 121         struct udev_monitor *monitor;
 122         enum worker_state state;
 123         struct event *event;
 124         usec_t event_start_usec;
 125         bool event_warned;
 126 };
 127
 128 /* passed from worker to main process */
 129 struct worker_message {
 130         pid_t pid;
 131         int exitcode;
 132 };
 133
 134 static inline struct worker *node_to_worker(struct udev_list_node *node) {
 135         return container_of(node, struct worker, node);
 136 }
 137
 138 static void event_queue_delete(struct event *event) {
 139         udev_list_node_remove(&event->node);
 140         udev_device_unref(event->dev);
 141         free(event);
 142 }
 143
 144 static struct worker *worker_ref(struct worker *worker) {
 145         worker->refcount++;
 146         return worker;
 147 }
 148
 149 static void worker_cleanup(struct worker *worker) {
 150         udev_list_node_remove(&worker->node);
 151         udev_monitor_unref(worker->monitor);
 152         children--;
 153         free(worker);
 154 }
 155
 156 static void worker_unref(struct worker *worker) {
 157         worker->refcount--;
 158         if (worker->refcount > 0)
 159                 return;
 160         log_debug("worker [%u] cleaned up", worker->pid);
 161         worker_cleanup(worker);
 162 }
 163
 164 static void worker_list_cleanup(struct udev *udev) {
 165         struct udev_list_node *loop, *tmp;
 166
 167         udev_list_node_foreach_safe(loop, tmp, &worker_list) {
 168                 struct worker *worker = node_to_worker(loop);
 169
 170                 worker_cleanup(worker);
 171         }
 172 }
 173
 174 static void worker_new(struct event *event) {
 175         struct udev *udev = event->udev;
 176         struct worker *worker;
 177         struct udev_monitor *worker_monitor;
 178         pid_t pid;
 179
 180         /* listen for new events */
 181         worker_monitor = udev_monitor_new_from_netlink(udev, NULL);
 182         if (worker_monitor == NULL)
 183                 return;
 184         /* allow the main daemon netlink address to send devices to the worker */
 185         udev_monitor_allow_unicast_sender(worker_monitor, monitor);
 186         udev_monitor_enable_receiving(worker_monitor);
 187
 188         worker = new0(struct worker, 1);
 189         if (worker == NULL) {
 190                 udev_monitor_unref(worker_monitor);
 191                 return;
 192         }
 193         /* worker + event reference */
 194         worker->refcount = 2;
 195         worker->udev = udev;
 196
 197         pid = fork();
 198         switch (pid) {
 199         case 0: {
 200                 struct udev_device *dev = NULL;
 201                 int fd_monitor;
 202                 _cleanup_rtnl_unref_ sd_rtnl *rtnl = NULL;
 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 (arg_exec_delay > 0)
 274                                 udev_event->exec_delay = arg_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                         /* needed for renaming netifs */
 305                         udev_event->rtnl = rtnl;
 306
 307                         /* apply rules, create node, symlinks */
 308                         udev_event_execute_rules(udev_event,
 309                                                  arg_event_timeout_usec, arg_event_timeout_warn_usec,
 310                                                  &properties_list,
 311                                                  rules,
 312                                                  &sigmask_orig);
 313
 314                         udev_event_execute_run(udev_event,
 315                                                arg_event_timeout_usec, arg_event_timeout_warn_usec,
 316                                                &sigmask_orig);
 317
 318                         /* in case rtnl was initialized */
 319                         rtnl = sd_rtnl_ref(udev_event->rtnl);
 320
 321                         /* apply/restore inotify watch */
 322                         if (udev_event->inotify_watch) {
 323                                 udev_watch_begin(udev, dev);
 324                                 udev_device_update_db(dev);
 325                         }
 326
 327                         safe_close(fd_lock);
 328
 329                         /* send processed event back to libudev listeners */
 330                         udev_monitor_send_device(worker_monitor, NULL, dev);
 331
 332 skip:
 333                         /* send udevd the result of the event execution */
 334                         memzero(&msg, sizeof(struct worker_message));
 335                         msg.exitcode = err;
 336                         msg.pid = getpid();
 337                         send(worker_watch[WRITE_END], &msg, sizeof(struct worker_message), 0);
 338
 339                         log_debug("seq %llu processed with %i", udev_device_get_seqnum(dev), err);
 340
 341                         udev_device_unref(dev);
 342                         dev = NULL;
 343
 344                         if (udev_event->sigterm) {
 345                                 udev_event_unref(udev_event);
 346                                 goto out;
 347                         }
 348
 349                         udev_event_unref(udev_event);
 350
 351                         /* wait for more device messages from main udevd, or term signal */
 352                         while (dev == NULL) {
 353                                 struct epoll_event ev[4];
 354                                 int fdcount;
 355                                 int i;
 356
 357                                 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), -1);
 358                                 if (fdcount < 0) {
 359                                         if (errno == EINTR)
 360                                                 continue;
 361                                         log_error("failed to poll: %m");
 362                                         goto out;
 363                                 }
 364
 365                                 for (i = 0; i < fdcount; i++) {
 366                                         if (ev[i].data.fd == fd_monitor && ev[i].events & EPOLLIN) {
 367                                                 dev = udev_monitor_receive_device(worker_monitor);
 368                                                 break;
 369                                         } else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN) {
 370                                                 struct signalfd_siginfo fdsi;
 371                                                 ssize_t size;
 372
 373                                                 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
 374                                                 if (size != sizeof(struct signalfd_siginfo))
 375                                                         continue;
 376                                                 switch (fdsi.ssi_signo) {
 377                                                 case SIGTERM:
 378                                                         goto out;
 379                                                 }
 380                                         }
 381                                 }
 382                         }
 383                 }
 384 out:
 385                 udev_device_unref(dev);
 386                 safe_close(fd_signal);
 387                 safe_close(fd_ep);
 388                 close(fd_inotify);
 389                 close(worker_watch[WRITE_END]);
 390                 udev_rules_unref(rules);
 391                 udev_builtin_exit(udev);
 392                 udev_monitor_unref(worker_monitor);
 393                 udev_unref(udev);
 394                 log_close();
 395                 exit(rc);
 396         }
 397         case -1:
 398                 udev_monitor_unref(worker_monitor);
 399                 event->state = EVENT_QUEUED;
 400                 free(worker);
 401                 log_error("fork of child failed: %m");
 402                 break;
 403         default:
 404                 /* close monitor, but keep address around */
 405                 udev_monitor_disconnect(worker_monitor);
 406                 worker->monitor = worker_monitor;
 407                 worker->pid = pid;
 408                 worker->state = WORKER_RUNNING;
 409                 worker->event_start_usec = now(CLOCK_MONOTONIC);
 410                 worker->event_warned = false;
 411                 worker->event = event;
 412                 event->state = EVENT_RUNNING;
 413                 udev_list_node_append(&worker->node, &worker_list);
 414                 children++;
 415                 log_debug("seq %llu forked new worker [%u]", udev_device_get_seqnum(event->dev), pid);
 416                 break;
 417         }
 418 }
 419
 420 static void event_run(struct event *event) {
 421         struct udev_list_node *loop;
 422
 423         udev_list_node_foreach(loop, &worker_list) {
 424                 struct worker *worker = node_to_worker(loop);
 425                 ssize_t count;
 426
 427                 if (worker->state != WORKER_IDLE)
 428                         continue;
 429
 430                 count = udev_monitor_send_device(monitor, worker->monitor, event->dev);
 431                 if (count < 0) {
 432                         log_error("worker [%u] did not accept message %zi (%m), kill it", worker->pid, count);
 433                         kill(worker->pid, SIGKILL);
 434                         worker->state = WORKER_KILLED;
 435                         continue;
 436                 }
 437                 worker_ref(worker);
 438                 worker->event = event;
 439                 worker->state = WORKER_RUNNING;
 440                 worker->event_start_usec = now(CLOCK_MONOTONIC);
 441                 worker->event_warned = false;
 442                 event->state = EVENT_RUNNING;
 443                 return;
 444         }
 445
 446         if (children >= arg_children_max) {
 447                 if (arg_children_max > 1)
 448                         log_debug("maximum number (%i) of children reached", children);
 449                 return;
 450         }
 451
 452         /* start new worker and pass initial device */
 453         worker_new(event);
 454 }
 455
 456 static int event_queue_insert(struct udev_device *dev) {
 457         struct event *event;
 458
 459         event = new0(struct event, 1);
 460         if (event == NULL)
 461                 return -1;
 462
 463         event->udev = udev_device_get_udev(dev);
 464         event->dev = dev;
 465         event->seqnum = udev_device_get_seqnum(dev);
 466         event->devpath = udev_device_get_devpath(dev);
 467         event->devpath_len = strlen(event->devpath);
 468         event->devpath_old = udev_device_get_devpath_old(dev);
 469         event->devnum = udev_device_get_devnum(dev);
 470         event->is_block = streq("block", udev_device_get_subsystem(dev));
 471         event->ifindex = udev_device_get_ifindex(dev);
 472
 473         log_debug("seq %llu queued, '%s' '%s'", udev_device_get_seqnum(dev),
 474              udev_device_get_action(dev), udev_device_get_subsystem(dev));
 475
 476         event->state = EVENT_QUEUED;
 477         udev_list_node_append(&event->node, &event_list);
 478         return 0;
 479 }
 480
 481 static void worker_kill(struct udev *udev) {
 482         struct udev_list_node *loop;
 483
 484         udev_list_node_foreach(loop, &worker_list) {
 485                 struct worker *worker = node_to_worker(loop);
 486
 487                 if (worker->state == WORKER_KILLED)
 488                         continue;
 489
 490                 worker->state = WORKER_KILLED;
 491                 kill(worker->pid, SIGTERM);
 492         }
 493 }
 494
 495 /* lookup event for identical, parent, child device */
 496 static bool is_devpath_busy(struct event *event) {
 497         struct udev_list_node *loop;
 498         size_t common;
 499
 500         /* check if queue contains events we depend on */
 501         udev_list_node_foreach(loop, &event_list) {
 502                 struct event *loop_event = node_to_event(loop);
 503
 504                 /* we already found a later event, earlier can not block us, no need to check again */
 505                 if (loop_event->seqnum < event->delaying_seqnum)
 506                         continue;
 507
 508                 /* event we checked earlier still exists, no need to check again */
 509                 if (loop_event->seqnum == event->delaying_seqnum)
 510                         return true;
 511
 512                 /* found ourself, no later event can block us */
 513                 if (loop_event->seqnum >= event->seqnum)
 514                         break;
 515
 516                 /* check major/minor */
 517                 if (major(event->devnum) != 0 && event->devnum == loop_event->devnum && event->is_block == loop_event->is_block)
 518                         return true;
 519
 520                 /* check network device ifindex */
 521                 if (event->ifindex != 0 && event->ifindex == loop_event->ifindex)
 522                         return true;
 523
 524                 /* check our old name */
 525                 if (event->devpath_old != NULL && streq(loop_event->devpath, event->devpath_old)) {
 526                         event->delaying_seqnum = loop_event->seqnum;
 527                         return true;
 528                 }
 529
 530                 /* compare devpath */
 531                 common = MIN(loop_event->devpath_len, event->devpath_len);
 532
 533                 /* one devpath is contained in the other? */
 534                 if (memcmp(loop_event->devpath, event->devpath, common) != 0)
 535                         continue;
 536
 537                 /* identical device event found */
 538                 if (loop_event->devpath_len == event->devpath_len) {
 539                         /* devices names might have changed/swapped in the meantime */
 540                         if (major(event->devnum) != 0 && (event->devnum != loop_event->devnum || event->is_block != loop_event->is_block))
 541                                 continue;
 542                         if (event->ifindex != 0 && event->ifindex != loop_event->ifindex)
 543                                 continue;
 544                         event->delaying_seqnum = loop_event->seqnum;
 545                         return true;
 546                 }
 547
 548                 /* parent device event found */
 549                 if (event->devpath[common] == '/') {
 550                         event->delaying_seqnum = loop_event->seqnum;
 551                         return true;
 552                 }
 553
 554                 /* child device event found */
 555                 if (loop_event->devpath[common] == '/') {
 556                         event->delaying_seqnum = loop_event->seqnum;
 557                         return true;
 558                 }
 559
 560                 /* no matching device */
 561                 continue;
 562         }
 563
 564         return false;
 565 }
 566
 567 static void event_queue_start(struct udev *udev) {
 568         struct udev_list_node *loop;
 569
 570         udev_list_node_foreach(loop, &event_list) {
 571                 struct event *event = node_to_event(loop);
 572
 573                 if (event->state != EVENT_QUEUED)
 574                         continue;
 575
 576                 /* do not start event if parent or child event is still running */
 577                 if (is_devpath_busy(event))
 578                         continue;
 579
 580                 event_run(event);
 581         }
 582 }
 583
 584 static void event_queue_cleanup(struct udev *udev, enum event_state match_type) {
 585         struct udev_list_node *loop, *tmp;
 586
 587         udev_list_node_foreach_safe(loop, tmp, &event_list) {
 588                 struct event *event = node_to_event(loop);
 589
 590                 if (match_type != EVENT_UNDEF && match_type != event->state)
 591                         continue;
 592
 593                 event_queue_delete(event);
 594         }
 595 }
 596
 597 static void worker_returned(int fd_worker) {
 598         for (;;) {
 599                 struct worker_message msg;
 600                 ssize_t size;
 601                 struct udev_list_node *loop;
 602
 603                 size = recv(fd_worker, &msg, sizeof(struct worker_message), MSG_DONTWAIT);
 604                 if (size != sizeof(struct worker_message))
 605                         break;
 606
 607                 /* lookup worker who sent the signal */
 608                 udev_list_node_foreach(loop, &worker_list) {
 609                         struct worker *worker = node_to_worker(loop);
 610
 611                         if (worker->pid != msg.pid)
 612                                 continue;
 613
 614                         /* worker returned */
 615                         if (worker->event) {
 616                                 worker->event->exitcode = msg.exitcode;
 617                                 event_queue_delete(worker->event);
 618                                 worker->event = NULL;
 619                         }
 620                         if (worker->state != WORKER_KILLED)
 621                                 worker->state = WORKER_IDLE;
 622                         worker_unref(worker);
 623                         break;
 624                 }
 625         }
 626 }
 627
 628 /* receive the udevd message from userspace */
 629 static struct udev_ctrl_connection *handle_ctrl_msg(struct udev_ctrl *uctrl) {
 630         struct udev *udev = udev_ctrl_get_udev(uctrl);
 631         struct udev_ctrl_connection *ctrl_conn;
 632         struct udev_ctrl_msg *ctrl_msg = NULL;
 633         const char *str;
 634         int i;
 635
 636         ctrl_conn = udev_ctrl_get_connection(uctrl);
 637         if (ctrl_conn == NULL)
 638                 goto out;
 639
 640         ctrl_msg = udev_ctrl_receive_msg(ctrl_conn);
 641         if (ctrl_msg == NULL)
 642                 goto out;
 643
 644         i = udev_ctrl_get_set_log_level(ctrl_msg);
 645         if (i >= 0) {
 646                 log_debug("udevd message (SET_LOG_LEVEL) received, log_priority=%i", i);
 647                 log_set_max_level(i);
 648                 worker_kill(udev);
 649         }
 650
 651         if (udev_ctrl_get_stop_exec_queue(ctrl_msg) > 0) {
 652                 log_debug("udevd message (STOP_EXEC_QUEUE) received");
 653                 stop_exec_queue = true;
 654         }
 655
 656         if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) {
 657                 log_debug("udevd message (START_EXEC_QUEUE) received");
 658                 stop_exec_queue = false;
 659         }
 660
 661         if (udev_ctrl_get_reload(ctrl_msg) > 0) {
 662                 log_debug("udevd message (RELOAD) received");
 663                 reload = true;
 664         }
 665
 666         str = udev_ctrl_get_set_env(ctrl_msg);
 667         if (str != NULL) {
 668                 char *key;
 669
 670                 key = strdup(str);
 671                 if (key != NULL) {
 672                         char *val;
 673
 674                         val = strchr(key, '=');
 675                         if (val != NULL) {
 676                                 val[0] = '\0';
 677                                 val = &val[1];
 678                                 if (val[0] == '\0') {
 679                                         log_debug("udevd message (ENV) received, unset '%s'", key);
 680                                         udev_list_entry_add(&properties_list, key, NULL);
 681                                 } else {
 682                                         log_debug("udevd message (ENV) received, set '%s=%s'", key, val);
 683                                         udev_list_entry_add(&properties_list, key, val);
 684                                 }
 685                         } else {
 686                                 log_error("wrong key format '%s'", key);
 687                         }
 688                         free(key);
 689                 }
 690                 worker_kill(udev);
 691         }
 692
 693         i = udev_ctrl_get_set_children_max(ctrl_msg);
 694         if (i >= 0) {
 695                 log_debug("udevd message (SET_MAX_CHILDREN) received, children_max=%i", i);
 696                 arg_children_max = i;
 697         }
 698
 699         if (udev_ctrl_get_ping(ctrl_msg) > 0)
 700                 log_debug("udevd message (SYNC) received");
 701
 702         if (udev_ctrl_get_exit(ctrl_msg) > 0) {
 703                 log_debug("udevd message (EXIT) received");
 704                 udev_exit = true;
 705                 /* keep reference to block the client until we exit */
 706                 udev_ctrl_connection_ref(ctrl_conn);
 707         }
 708 out:
 709         udev_ctrl_msg_unref(ctrl_msg);
 710         return udev_ctrl_connection_unref(ctrl_conn);
 711 }
 712
 713 static int synthesize_change(struct udev_device *dev) {
 714         char filename[UTIL_PATH_SIZE];
 715         int r;
 716
 717         if (streq_ptr("block", udev_device_get_subsystem(dev)) &&
 718             streq_ptr("disk", udev_device_get_devtype(dev)) &&
 719             !startswith(udev_device_get_sysname(dev), "dm-")) {
 720                 bool part_table_read = false;
 721                 bool has_partitions = false;
 722                 int fd;
 723                 struct udev *udev = udev_device_get_udev(dev);
 724                 _cleanup_udev_enumerate_unref_ struct udev_enumerate *e = NULL;
 725                 struct udev_list_entry *item;
 726
 727                 /*
 728                  * Try to re-read the partition table. This only succeeds if
 729                  * none of the devices is busy. The kernel returns 0 if no
 730                  * partition table is found, and we will not get an event for
 731                  * the disk.
 732                  */
 733                 fd = open(udev_device_get_devnode(dev), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
 734                 if (fd >= 0) {
 735                         r = flock(fd, LOCK_EX|LOCK_NB);
 736                         if (r >= 0)
 737                                 r = ioctl(fd, BLKRRPART, 0);
 738
 739                         close(fd);
 740                         if (r >= 0)
 741                                 part_table_read = true;
 742                 }
 743
 744                 /* search for partitions */
 745                 e = udev_enumerate_new(udev);
 746                 if (!e)
 747                         return -ENOMEM;
 748
 749                 r = udev_enumerate_add_match_parent(e, dev);
 750                 if (r < 0)
 751                         return r;
 752
 753                 r = udev_enumerate_add_match_subsystem(e, "block");
 754                 if (r < 0)
 755                         return r;
 756
 757                 r = udev_enumerate_scan_devices(e);
 758                 if (r < 0)
 759                         return r;
 760
 761                 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
 762                         _cleanup_udev_device_unref_ struct udev_device *d = NULL;
 763
 764                         d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
 765                         if (!d)
 766                                 continue;
 767
 768                         if (!streq_ptr("partition", udev_device_get_devtype(d)))
 769                                 continue;
 770
 771                         has_partitions = true;
 772                         break;
 773                 }
 774
 775                 /*
 776                  * We have partitions and re-read the table, the kernel already sent
 777                  * out a "change" event for the disk, and "remove/add" for all
 778                  * partitions.
 779                  */
 780                 if (part_table_read && has_partitions)
 781                         return 0;
 782
 783                 /*
 784                  * We have partitions but re-reading the partition table did not
 785                  * work, synthesize "change" for the disk and all partitions.
 786                  */
 787                 log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
 788                 strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
 789                 write_string_file(filename, "change");
 790
 791                 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
 792                         _cleanup_udev_device_unref_ struct udev_device *d = NULL;
 793
 794                         d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
 795                         if (!d)
 796                                 continue;
 797
 798                         if (!streq_ptr("partition", udev_device_get_devtype(d)))
 799                                 continue;
 800
 801                         log_debug("device %s closed, synthesising partition '%s' 'change'",
 802                                   udev_device_get_devnode(dev), udev_device_get_devnode(d));
 803                         strscpyl(filename, sizeof(filename), udev_device_get_syspath(d), "/uevent", NULL);
 804                         write_string_file(filename, "change");
 805                 }
 806
 807                 return 0;
 808         }
 809
 810         log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
 811         strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
 812         write_string_file(filename, "change");
 813
 814         return 0;
 815 }
 816
 817 static int handle_inotify(struct udev *udev) {
 818         int nbytes, pos;
 819         char *buf;
 820         struct inotify_event *ev;
 821         int r;
 822
 823         r = ioctl(fd_inotify, FIONREAD, &nbytes);
 824         if (r < 0 || nbytes <= 0)
 825                 return -errno;
 826
 827         buf = malloc(nbytes);
 828         if (!buf) {
 829                 log_error("error getting buffer for inotify");
 830                 return -ENOMEM;
 831         }
 832
 833         nbytes = read(fd_inotify, buf, nbytes);
 834
 835         for (pos = 0; pos < nbytes; pos += sizeof(struct inotify_event) + ev->len) {
 836                 struct udev_device *dev;
 837
 838                 ev = (struct inotify_event *)(buf + pos);
 839                 dev = udev_watch_lookup(udev, ev->wd);
 840                 if (!dev)
 841                         continue;
 842
 843                 log_debug("inotify event: %x for %s", ev->mask, udev_device_get_devnode(dev));
 844                 if (ev->mask & IN_CLOSE_WRITE)
 845                         synthesize_change(dev);
 846                 else if (ev->mask & IN_IGNORED)
 847                         udev_watch_end(udev, dev);
 848
 849                 udev_device_unref(dev);
 850         }
 851
 852         free(buf);
 853         return 0;
 854 }
 855
 856 static void handle_signal(struct udev *udev, int signo) {
 857         switch (signo) {
 858         case SIGINT:
 859         case SIGTERM:
 860                 udev_exit = true;
 861                 break;
 862         case SIGCHLD:
 863                 for (;;) {
 864                         pid_t pid;
 865                         int status;
 866                         struct udev_list_node *loop, *tmp;
 867
 868                         pid = waitpid(-1, &status, WNOHANG);
 869                         if (pid <= 0)
 870                                 break;
 871
 872                         udev_list_node_foreach_safe(loop, tmp, &worker_list) {
 873                                 struct worker *worker = node_to_worker(loop);
 874
 875                                 if (worker->pid != pid)
 876                                         continue;
 877                                 log_debug("worker [%u] exit", pid);
 878
 879                                 if (WIFEXITED(status)) {
 880                                         if (WEXITSTATUS(status) != 0)
 881                                                 log_error("worker [%u] exit with return code %i",
 882                                                           pid, WEXITSTATUS(status));
 883                                 } else if (WIFSIGNALED(status)) {
 884                                         log_error("worker [%u] terminated by signal %i (%s)",
 885                                                   pid, WTERMSIG(status), strsignal(WTERMSIG(status)));
 886                                 } else if (WIFSTOPPED(status)) {
 887                                         log_error("worker [%u] stopped", pid);
 888                                 } else if (WIFCONTINUED(status)) {
 889                                         log_error("worker [%u] continued", pid);
 890                                 } else {
 891                                         log_error("worker [%u] exit with status 0x%04x", pid, status);
 892                                 }
 893
 894                                 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 895                                         if (worker->event) {
 896                                                 log_error("worker [%u] failed while handling '%s'",
 897                                                           pid, worker->event->devpath);
 898                                                 worker->event->exitcode = -32;
 899                                                 event_queue_delete(worker->event);
 900
 901                                                 /* drop reference taken for state 'running' */
 902                                                 worker_unref(worker);
 903                                         }
 904                                 }
 905                                 worker_unref(worker);
 906                                 break;
 907                         }
 908                 }
 909                 break;
 910         case SIGHUP:
 911                 reload = true;
 912                 break;
 913         }
 914 }
 915
 916 static int systemd_fds(struct udev *udev, int *rctrl, int *rnetlink) {
 917         int ctrl = -1, netlink = -1;
 918         int fd, n;
 919
 920         n = sd_listen_fds(true);
 921         if (n <= 0)
 922                 return -1;
 923
 924         for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
 925                 if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1)) {
 926                         if (ctrl >= 0)
 927                                 return -1;
 928                         ctrl = fd;
 929                         continue;
 930                 }
 931
 932                 if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1)) {
 933                         if (netlink >= 0)
 934                                 return -1;
 935                         netlink = fd;
 936                         continue;
 937                 }
 938
 939                 return -1;
 940         }
 941
 942         if (ctrl < 0 || netlink < 0)
 943                 return -1;
 944
 945         log_debug("ctrl=%i netlink=%i", ctrl, netlink);
 946         *rctrl = ctrl;
 947         *rnetlink = netlink;
 948         return 0;
 949 }
 950
 951 /*
 952  * read the kernel command line, in case we need to get into debug mode
 953  *   udev.log-priority=<level>              syslog priority
 954  *   udev.children-max=<number of workers>  events are fully serialized if set to 1
 955  *   udev.exec-delay=<number of seconds>    delay execution of every executed program
 956  */
 957 static void kernel_cmdline_options(struct udev *udev) {
 958         _cleanup_free_ char *line = NULL;
 959         const char *word, *state;
 960         size_t l;
 961         int r;
 962
 963         r = proc_cmdline(&line);
 964         if (r < 0) {
 965                 log_warning_errno(-r, "Failed to read /proc/cmdline, ignoring: %m");
 966                 return;
 967         }
 968
 969         FOREACH_WORD_QUOTED(word, l, line, state) {
 970                 char *s, *opt, *value;
 971
 972                 s = strndup(word, l);
 973                 if (!s)
 974                         break;
 975
 976                 /* accept the same options for the initrd, prefixed with "rd." */
 977                 if (in_initrd() && startswith(s, "rd."))
 978                         opt = s + 3;
 979                 else
 980                         opt = s;
 981
 982                 if ((value = startswith(opt, "udev.log-priority="))) {
 983                         int prio;
 984
 985                         prio = util_log_priority(value);
 986                         log_set_max_level(prio);
 987                 } else if ((value = startswith(opt, "udev.children-max="))) {
 988                         r = safe_atoi(value, &arg_children_max);
 989                         if (r < 0)
 990                                 log_warning("Invalid udev.children-max ignored: %s", value);
 991                 } else if ((value = startswith(opt, "udev.exec-delay="))) {
 992                         r = safe_atoi(value, &arg_exec_delay);
 993                         if (r < 0)
 994                                 log_warning("Invalid udev.exec-delay ignored: %s", value);
 995                 } else if ((value = startswith(opt, "udev.event-timeout="))) {
 996                         r = safe_atou64(value, &arg_event_timeout_usec);
 997                         if (r < 0) {
 998                                 log_warning("Invalid udev.event-timeout ignored: %s", value);
 999                                 break;
1000                         }
1001                         arg_event_timeout_usec *= USEC_PER_SEC;
1002                         arg_event_timeout_warn_usec = (arg_event_timeout_usec / 3) ? : 1;
1003                 }
1004
1005                 free(s);
1006         }
1007 }
1008
1009 static void help(void) {
1010         printf("%s [OPTIONS...]\n\n"
1011                "Manages devices.\n\n"
1012                "  --daemon\n"
1013                "  --debug\n"
1014                "  --children-max=<maximum number of workers>\n"
1015                "  --exec-delay=<seconds to wait before executing RUN=>\n"
1016                "  --event-timeout=<seconds to wait before terminating an event>\n"
1017                "  --resolve-names=early|late|never\n"
1018                "  --version\n"
1019                "  --help\n"
1020                , program_invocation_short_name);
1021 }
1022
1023 static int parse_argv(int argc, char *argv[]) {
1024         static const struct option options[] = {
1025                 { "daemon",             no_argument,            NULL, 'd' },
1026                 { "debug",              no_argument,            NULL, 'D' },
1027                 { "children-max",       required_argument,      NULL, 'c' },
1028                 { "exec-delay",         required_argument,      NULL, 'e' },
1029                 { "event-timeout",      required_argument,      NULL, 't' },
1030                 { "resolve-names",      required_argument,      NULL, 'N' },
1031                 { "help",               no_argument,            NULL, 'h' },
1032                 { "version",            no_argument,            NULL, 'V' },
1033                 {}
1034         };
1035
1036         int c;
1037
1038         assert(argc >= 0);
1039         assert(argv);
1040
1041         while ((c = getopt_long(argc, argv, "c:de:DtN:hV", options, NULL)) >= 0) {
1042                 int r;
1043
1044                 switch (c) {
1045
1046                 case 'd':
1047                         arg_daemonize = true;
1048                         break;
1049                 case 'c':
1050                         r = safe_atoi(optarg, &arg_children_max);
1051                         if (r < 0)
1052                                 log_warning("Invalid --children-max ignored: %s", optarg);
1053                         break;
1054                 case 'e':
1055                         r = safe_atoi(optarg, &arg_exec_delay);
1056                         if (r < 0)
1057                                 log_warning("Invalid --exec-delay ignored: %s", optarg);
1058                         break;
1059                 case 't':
1060                         r = safe_atou64(optarg, &arg_event_timeout_usec);
1061                         if (r < 0)
1062                                 log_warning("Invalid --event-timeout ignored: %s", optarg);
1063                         else {
1064                                 arg_event_timeout_usec *= USEC_PER_SEC;
1065                                 arg_event_timeout_warn_usec = (arg_event_timeout_usec / 3) ? : 1;
1066                         }
1067                         break;
1068                 case 'D':
1069                         arg_debug = true;
1070                         break;
1071                 case 'N':
1072                         if (streq(optarg, "early")) {
1073                                 arg_resolve_names = 1;
1074                         } else if (streq(optarg, "late")) {
1075                                 arg_resolve_names = 0;
1076                         } else if (streq(optarg, "never")) {
1077                                 arg_resolve_names = -1;
1078                         } else {
1079                                 log_error("resolve-names must be early, late or never");
1080                                 return 0;
1081                         }
1082                         break;
1083                 case 'h':
1084                         help();
1085                         return 0;
1086                 case 'V':
1087                         printf("%s\n", VERSION);
1088                         return 0;
1089                 case '?':
1090                         return -EINVAL;
1091                 default:
1092                         assert_not_reached("Unhandled option");
1093
1094                 }
1095         }
1096
1097         return 1;
1098 }
1099
1100 int main(int argc, char *argv[]) {
1101         struct udev *udev;
1102         sigset_t mask;
1103         int fd_ctrl = -1;
1104         int fd_netlink = -1;
1105         int fd_worker = -1;
1106         struct epoll_event ep_ctrl = { .events = EPOLLIN };
1107         struct epoll_event ep_inotify = { .events = EPOLLIN };
1108         struct epoll_event ep_signal = { .events = EPOLLIN };
1109         struct epoll_event ep_netlink = { .events = EPOLLIN };
1110         struct epoll_event ep_worker = { .events = EPOLLIN };
1111         struct udev_ctrl_connection *ctrl_conn = NULL;
1112         int rc = 1, r;
1113
1114         udev = udev_new();
1115         if (udev == NULL)
1116                 goto exit;
1117
1118         log_set_target(LOG_TARGET_AUTO);
1119         log_parse_environment();
1120         log_open();
1121
1122         r = parse_argv(argc, argv);
1123         if (r <= 0)
1124                 goto exit;
1125
1126         kernel_cmdline_options(udev);
1127
1128         if (arg_debug)
1129                 log_set_max_level(LOG_DEBUG);
1130
1131         if (getuid() != 0) {
1132                 log_error("root privileges required");
1133                 goto exit;
1134         }
1135
1136         r = mac_selinux_init("/dev");
1137         if (r < 0) {
1138                 log_error_errno(-r, "could not initialize labelling: %m");
1139                 goto exit;
1140         }
1141
1142         /* set umask before creating any file/directory */
1143         r = chdir("/");
1144         if (r < 0) {
1145                 log_error("could not change dir to /: %m");
1146                 goto exit;
1147         }
1148
1149         umask(022);
1150
1151         udev_list_init(udev, &properties_list, true);
1152
1153         r = mkdir("/run/udev", 0755);
1154         if (r < 0 && errno != EEXIST) {
1155                 log_error("could not create /run/udev: %m");
1156                 goto exit;
1157         }
1158
1159         dev_setup(NULL);
1160
1161         /* before opening new files, make sure std{in,out,err} fds are in a sane state */
1162         if (arg_daemonize) {
1163                 int fd;
1164
1165                 fd = open("/dev/null", O_RDWR);
1166                 if (fd >= 0) {
1167                         if (write(STDOUT_FILENO, 0, 0) < 0)
1168                                 dup2(fd, STDOUT_FILENO);
1169                         if (write(STDERR_FILENO, 0, 0) < 0)
1170                                 dup2(fd, STDERR_FILENO);
1171                         if (fd > STDERR_FILENO)
1172                                 close(fd);
1173                 } else {
1174                         log_error("cannot open /dev/null");
1175                 }
1176         }
1177
1178         if (systemd_fds(udev, &fd_ctrl, &fd_netlink) >= 0) {
1179                 /* get control and netlink socket from systemd */
1180                 udev_ctrl = udev_ctrl_new_from_fd(udev, fd_ctrl);
1181                 if (udev_ctrl == NULL) {
1182                         log_error("error taking over udev control socket");
1183                         rc = 1;
1184                         goto exit;
1185                 }
1186
1187                 monitor = udev_monitor_new_from_netlink_fd(udev, "kernel", fd_netlink);
1188                 if (monitor == NULL) {
1189                         log_error("error taking over netlink socket");
1190                         rc = 3;
1191                         goto exit;
1192                 }
1193
1194                 /* get our own cgroup, we regularly kill everything udev has left behind */
1195                 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &udev_cgroup) < 0)
1196                         udev_cgroup = NULL;
1197         } else {
1198                 /* open control and netlink socket */
1199                 udev_ctrl = udev_ctrl_new(udev);
1200                 if (udev_ctrl == NULL) {
1201                         log_error("error initializing udev control socket");
1202                         rc = 1;
1203                         goto exit;
1204                 }
1205                 fd_ctrl = udev_ctrl_get_fd(udev_ctrl);
1206
1207                 monitor = udev_monitor_new_from_netlink(udev, "kernel");
1208                 if (monitor == NULL) {
1209                         log_error("error initializing netlink socket");
1210                         rc = 3;
1211                         goto exit;
1212                 }
1213                 fd_netlink = udev_monitor_get_fd(monitor);
1214         }
1215
1216         if (udev_monitor_enable_receiving(monitor) < 0) {
1217                 log_error("error binding netlink socket");
1218                 rc = 3;
1219                 goto exit;
1220         }
1221
1222         if (udev_ctrl_enable_receiving(udev_ctrl) < 0) {
1223                 log_error("error binding udev control socket");
1224                 rc = 1;
1225                 goto exit;
1226         }
1227
1228         udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024);
1229
1230         log_info("starting version " VERSION "\n");
1231
1232         udev_builtin_init(udev);
1233
1234         rules = udev_rules_new(udev, arg_resolve_names);
1235         if (rules == NULL) {
1236                 log_error("error reading rules");
1237                 goto exit;
1238         }
1239
1240         rc = udev_rules_apply_static_dev_perms(rules);
1241         if (rc < 0)
1242                 log_error_errno(-rc, "failed to apply permissions on static device nodes - %m");
1243
1244         if (arg_daemonize) {
1245                 pid_t pid;
1246
1247                 pid = fork();
1248                 switch (pid) {
1249                 case 0:
1250                         break;
1251                 case -1:
1252                         log_error("fork of daemon failed: %m");
1253                         rc = 4;
1254                         goto exit;
1255                 default:
1256                         rc = EXIT_SUCCESS;
1257                         goto exit_daemonize;
1258                 }
1259
1260                 setsid();
1261
1262                 write_string_file("/proc/self/oom_score_adj", "-1000");
1263         } else {
1264                 sd_notify(1, "READY=1");
1265         }
1266
1267         if (arg_children_max <= 0) {
1268                 cpu_set_t cpu_set;
1269
1270                 arg_children_max = 8;
1271
1272                 if (sched_getaffinity(0, sizeof (cpu_set), &cpu_set) == 0) {
1273                         arg_children_max +=  CPU_COUNT(&cpu_set) * 2;
1274                 }
1275         }
1276         log_debug("set children_max to %u", arg_children_max);
1277
1278         udev_list_node_init(&event_list);
1279         udev_list_node_init(&worker_list);
1280
1281         fd_inotify = udev_watch_init(udev);
1282         if (fd_inotify < 0) {
1283                 log_error("error initializing inotify");
1284                 rc = 4;
1285                 goto exit;
1286         }
1287         udev_watch_restore(udev);
1288
1289         /* block and listen to all signals on signalfd */
1290         sigfillset(&mask);
1291         sigprocmask(SIG_SETMASK, &mask, &sigmask_orig);
1292         fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1293         if (fd_signal < 0) {
1294                 log_error("error creating signalfd");
1295                 rc = 5;
1296                 goto exit;
1297         }
1298
1299         /* unnamed socket from workers to the main daemon */
1300         if (socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, worker_watch) < 0) {
1301                 log_error("error creating socketpair");
1302                 rc = 6;
1303                 goto exit;
1304         }
1305         fd_worker = worker_watch[READ_END];
1306
1307         ep_ctrl.data.fd = fd_ctrl;
1308         ep_inotify.data.fd = fd_inotify;
1309         ep_signal.data.fd = fd_signal;
1310         ep_netlink.data.fd = fd_netlink;
1311         ep_worker.data.fd = fd_worker;
1312
1313         fd_ep = epoll_create1(EPOLL_CLOEXEC);
1314         if (fd_ep < 0) {
1315                 log_error("error creating epoll fd: %m");
1316                 goto exit;
1317         }
1318         if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_ctrl, &ep_ctrl) < 0 ||
1319             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_inotify, &ep_inotify) < 0 ||
1320             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
1321             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_netlink, &ep_netlink) < 0 ||
1322             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_worker, &ep_worker) < 0) {
1323                 log_error("fail to add fds to epoll: %m");
1324                 goto exit;
1325         }
1326
1327         for (;;) {
1328                 static usec_t last_usec;
1329                 struct epoll_event ev[8];
1330                 int fdcount;
1331                 int timeout;
1332                 bool is_worker, is_signal, is_inotify, is_netlink, is_ctrl;
1333                 int i;
1334
1335                 if (udev_exit) {
1336                         /* close sources of new events and discard buffered events */
1337                         if (fd_ctrl >= 0) {
1338                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_ctrl, NULL);
1339                                 fd_ctrl = -1;
1340                         }
1341                         if (monitor != NULL) {
1342                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_netlink, NULL);
1343                                 udev_monitor_unref(monitor);
1344                                 monitor = NULL;
1345                         }
1346                         if (fd_inotify >= 0) {
1347                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_inotify, NULL);
1348                                 close(fd_inotify);
1349                                 fd_inotify = -1;
1350                         }
1351
1352                         /* discard queued events and kill workers */
1353                         event_queue_cleanup(udev, EVENT_QUEUED);
1354                         worker_kill(udev);
1355
1356                         /* exit after all has cleaned up */
1357                         if (udev_list_node_is_empty(&event_list) && children == 0)
1358                                 break;
1359
1360                         /* timeout at exit for workers to finish */
1361                         timeout = 30 * MSEC_PER_SEC;
1362                 } else if (udev_list_node_is_empty(&event_list) && children == 0) {
1363                         /* we are idle */
1364                         timeout = -1;
1365
1366                         /* cleanup possible left-over processes in our cgroup */
1367                         if (udev_cgroup)
1368                                 cg_kill(SYSTEMD_CGROUP_CONTROLLER, udev_cgroup, SIGKILL, false, true, NULL);
1369                 } else {
1370                         /* kill idle or hanging workers */
1371                         timeout = 3 * MSEC_PER_SEC;
1372                 }
1373
1374                 /* tell settle that we are busy or idle */
1375                 if (!udev_list_node_is_empty(&event_list)) {
1376                         int fd;
1377
1378                         fd = open("/run/udev/queue", O_WRONLY|O_CREAT|O_CLOEXEC|O_TRUNC|O_NOFOLLOW, 0444);
1379                         if (fd >= 0)
1380                                 close(fd);
1381                 } else {
1382                         unlink("/run/udev/queue");
1383                 }
1384
1385                 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), timeout);
1386                 if (fdcount < 0)
1387                         continue;
1388
1389                 if (fdcount == 0) {
1390                         struct udev_list_node *loop;
1391
1392                         /* timeout */
1393                         if (udev_exit) {
1394                                 log_error("timeout, giving up waiting for workers to finish");
1395                                 break;
1396                         }
1397
1398                         /* kill idle workers */
1399                         if (udev_list_node_is_empty(&event_list)) {
1400                                 log_debug("cleanup idle workers");
1401                                 worker_kill(udev);
1402                         }
1403
1404                         /* check for hanging events */
1405                         udev_list_node_foreach(loop, &worker_list) {
1406                                 struct worker *worker = node_to_worker(loop);
1407                                 usec_t ts;
1408
1409                                 if (worker->state != WORKER_RUNNING)
1410                                         continue;
1411
1412                                 ts = now(CLOCK_MONOTONIC);
1413
1414                                 if ((ts - worker->event_start_usec) > arg_event_timeout_warn_usec) {
1415                                         if ((ts - worker->event_start_usec) > arg_event_timeout_usec) {
1416                                                 log_error("worker [%u] %s timeout; kill it", worker->pid, worker->event->devpath);
1417                                                 kill(worker->pid, SIGKILL);
1418                                                 worker->state = WORKER_KILLED;
1419
1420                                                 /* drop reference taken for state 'running' */
1421                                                 worker_unref(worker);
1422                                                 log_error("seq %llu '%s' killed", udev_device_get_seqnum(worker->event->dev), worker->event->devpath);
1423                                                 worker->event->exitcode = -64;
1424                                                 event_queue_delete(worker->event);
1425                                                 worker->event = NULL;
1426                                         } else if (!worker->event_warned) {
1427                                                 log_warning("worker [%u] %s is taking a long time", worker->pid, worker->event->devpath);
1428                                                 worker->event_warned = true;
1429                                         }
1430                                 }
1431                         }
1432
1433                 }
1434
1435                 is_worker = is_signal = is_inotify = is_netlink = is_ctrl = false;
1436                 for (i = 0; i < fdcount; i++) {
1437                         if (ev[i].data.fd == fd_worker && ev[i].events & EPOLLIN)
1438                                 is_worker = true;
1439                         else if (ev[i].data.fd == fd_netlink && ev[i].events & EPOLLIN)
1440                                 is_netlink = true;
1441                         else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN)
1442                                 is_signal = true;
1443                         else if (ev[i].data.fd == fd_inotify && ev[i].events & EPOLLIN)
1444                                 is_inotify = true;
1445                         else if (ev[i].data.fd == fd_ctrl && ev[i].events & EPOLLIN)
1446                                 is_ctrl = true;
1447                 }
1448
1449                 /* check for changed config, every 3 seconds at most */
1450                 if ((now(CLOCK_MONOTONIC) - last_usec) > 3 * USEC_PER_SEC) {
1451                         if (udev_rules_check_timestamp(rules))
1452                                 reload = true;
1453                         if (udev_builtin_validate(udev))
1454                                 reload = true;
1455
1456                         last_usec = now(CLOCK_MONOTONIC);
1457                 }
1458
1459                 /* reload requested, HUP signal received, rules changed, builtin changed */
1460                 if (reload) {
1461                         worker_kill(udev);
1462                         rules = udev_rules_unref(rules);
1463                         udev_builtin_exit(udev);
1464                         reload = false;
1465                 }
1466
1467                 /* event has finished */
1468                 if (is_worker)
1469                         worker_returned(fd_worker);
1470
1471                 if (is_netlink) {
1472                         struct udev_device *dev;
1473
1474                         dev = udev_monitor_receive_device(monitor);
1475                         if (dev != NULL) {
1476                                 udev_device_set_usec_initialized(dev, now(CLOCK_MONOTONIC));
1477                                 if (event_queue_insert(dev) < 0)
1478                                         udev_device_unref(dev);
1479                         }
1480                 }
1481
1482                 /* start new events */
1483                 if (!udev_list_node_is_empty(&event_list) && !udev_exit && !stop_exec_queue) {
1484                         udev_builtin_init(udev);
1485                         if (rules == NULL)
1486                                 rules = udev_rules_new(udev, arg_resolve_names);
1487                         if (rules != NULL)
1488                                 event_queue_start(udev);
1489                 }
1490
1491                 if (is_signal) {
1492                         struct signalfd_siginfo fdsi;
1493                         ssize_t size;
1494
1495                         size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
1496                         if (size == sizeof(struct signalfd_siginfo))
1497                                 handle_signal(udev, fdsi.ssi_signo);
1498                 }
1499
1500                 /* we are shutting down, the events below are not handled anymore */
1501                 if (udev_exit)
1502                         continue;
1503
1504                 /* device node watch */
1505                 if (is_inotify)
1506                         handle_inotify(udev);
1507
1508                 /*
1509                  * This needs to be after the inotify handling, to make sure,
1510                  * that the ping is send back after the possibly generated
1511                  * "change" events by the inotify device node watch.
1512                  *
1513                  * A single time we may receive a client connection which we need to
1514                  * keep open to block the client. It will be closed right before we
1515                  * exit.
1516                  */
1517                 if (is_ctrl)
1518                         ctrl_conn = handle_ctrl_msg(udev_ctrl);
1519         }
1520
1521         rc = EXIT_SUCCESS;
1522 exit:
1523         udev_ctrl_cleanup(udev_ctrl);
1524         unlink("/run/udev/queue");
1525 exit_daemonize:
1526         if (fd_ep >= 0)
1527                 close(fd_ep);
1528         worker_list_cleanup(udev);
1529         event_queue_cleanup(udev, EVENT_UNDEF);
1530         udev_rules_unref(rules);
1531         udev_builtin_exit(udev);
1532         if (fd_signal >= 0)
1533                 close(fd_signal);
1534         if (worker_watch[READ_END] >= 0)
1535                 close(worker_watch[READ_END]);
1536         if (worker_watch[WRITE_END] >= 0)
1537                 close(worker_watch[WRITE_END]);
1538         udev_monitor_unref(monitor);
1539         udev_ctrl_connection_unref(ctrl_conn);
1540         udev_ctrl_unref(udev_ctrl);
1541         udev_list_cleanup(&properties_list);
1542         mac_selinux_finish();
1543         udev_unref(udev);
1544         log_close();
1545         return rc;
1546 }