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