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