src/udev/udevd.c

   1 /*
   2  * Copyright (C) 2004-2012 Kay Sievers <kay@vrfy.org>
   3  * Copyright (C) 2004 Chris Friesen <chris_friesen@sympatico.ca>
   4  * Copyright (C) 2009 Canonical Ltd.
   5  * Copyright (C) 2009 Scott James Remnant <scott@netsplit.com>
   6  *
   7  * This program is free software: you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License as published by
   9  * the Free Software Foundation, either version 2 of the License, or
  10  * (at your option) any later version.
  11  *
  12  * This program is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  * GNU General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  19  */
  20
  21 #include <stddef.h>
  22 #include <signal.h>
  23 #include <unistd.h>
  24 #include <errno.h>
  25 #include <stdio.h>
  26 #include <stdlib.h>
  27 #include <stdbool.h>
  28 #include <string.h>
  29 #include <ctype.h>
  30 #include <fcntl.h>
  31 #include <time.h>
  32 #include <getopt.h>
  33 #include <dirent.h>
  34 #include <sys/file.h>
  35 #include <sys/time.h>
  36 #include <sys/prctl.h>
  37 #include <sys/socket.h>
  38 #include <sys/un.h>
  39 #include <sys/signalfd.h>
  40 #include <sys/epoll.h>
  41 #include <sys/mount.h>
  42 #include <sys/poll.h>
  43 #include <sys/wait.h>
  44 #include <sys/stat.h>
  45 #include <sys/ioctl.h>
  46 #include <sys/inotify.h>
  47 #include <sys/utsname.h>
  48
  49 #include "udev.h"
  50 #include "udev-util.h"
  51 #include "sd-daemon.h"
  52 #include "cgroup-util.h"
  53 #include "dev-setup.h"
  54 #include "fileio.h"
  55
  56 static bool debug;
  57
  58 void udev_main_log(struct udev *udev, int priority,
  59                    const char *file, int line, const char *fn,
  60                    const char *format, va_list args) {
  61         log_metav(priority, file, line, fn, format, args);
  62 }
  63
  64 static struct udev_rules *rules;
  65 static struct udev_ctrl *udev_ctrl;
  66 static struct udev_monitor *monitor;
  67 static int worker_watch[2] = { -1, -1 };
  68 static int fd_signal = -1;
  69 static int fd_ep = -1;
  70 static int fd_inotify = -1;
  71 static bool stop_exec_queue;
  72 static bool reload;
  73 static int children;
  74 static int children_max;
  75 static int exec_delay;
  76 static usec_t event_timeout_usec = 30 * USEC_PER_SEC;
  77 static sigset_t sigmask_orig;
  78 static UDEV_LIST(event_list);
  79 static UDEV_LIST(worker_list);
  80 static char *udev_cgroup;
  81 static bool udev_exit;
  82
  83 enum event_state {
  84         EVENT_UNDEF,
  85         EVENT_QUEUED,
  86         EVENT_RUNNING,
  87 };
  88
  89 struct event {
  90         struct udev_list_node node;
  91         struct udev *udev;
  92         struct udev_device *dev;
  93         enum event_state state;
  94         int exitcode;
  95         unsigned long long int delaying_seqnum;
  96         unsigned long long int seqnum;
  97         const char *devpath;
  98         size_t devpath_len;
  99         const char *devpath_old;
 100         dev_t devnum;
 101         int ifindex;
 102         bool is_block;
 103 #ifdef HAVE_FIRMWARE
 104         bool nodelay;
 105 #endif
 106 };
 107
 108 static inline struct event *node_to_event(struct udev_list_node *node) {
 109         return container_of(node, struct event, node);
 110 }
 111
 112 static void event_queue_cleanup(struct udev *udev, enum event_state type);
 113
 114 enum worker_state {
 115         WORKER_UNDEF,
 116         WORKER_RUNNING,
 117         WORKER_IDLE,
 118         WORKER_KILLED,
 119 };
 120
 121 struct worker {
 122         struct udev_list_node node;
 123         struct udev *udev;
 124         int refcount;
 125         pid_t pid;
 126         struct udev_monitor *monitor;
 127         enum worker_state state;
 128         struct event *event;
 129         usec_t event_start_usec;
 130 };
 131
 132 /* passed from worker to main process */
 133 struct worker_message {
 134         pid_t pid;
 135         int exitcode;
 136 };
 137
 138 static inline struct worker *node_to_worker(struct udev_list_node *node) {
 139         return container_of(node, struct worker, node);
 140 }
 141
 142 static void event_queue_delete(struct event *event) {
 143         udev_list_node_remove(&event->node);
 144         udev_device_unref(event->dev);
 145         free(event);
 146 }
 147
 148 static struct worker *worker_ref(struct worker *worker) {
 149         worker->refcount++;
 150         return worker;
 151 }
 152
 153 static void worker_cleanup(struct worker *worker) {
 154         udev_list_node_remove(&worker->node);
 155         udev_monitor_unref(worker->monitor);
 156         children--;
 157         free(worker);
 158 }
 159
 160 static void worker_unref(struct worker *worker) {
 161         worker->refcount--;
 162         if (worker->refcount > 0)
 163                 return;
 164         log_debug("worker [%u] cleaned up", worker->pid);
 165         worker_cleanup(worker);
 166 }
 167
 168 static void worker_list_cleanup(struct udev *udev) {
 169         struct udev_list_node *loop, *tmp;
 170
 171         udev_list_node_foreach_safe(loop, tmp, &worker_list) {
 172                 struct worker *worker = node_to_worker(loop);
 173
 174                 worker_cleanup(worker);
 175         }
 176 }
 177
 178 static void worker_new(struct event *event) {
 179         struct udev *udev = event->udev;
 180         struct worker *worker;
 181         struct udev_monitor *worker_monitor;
 182         pid_t pid;
 183
 184         /* listen for new events */
 185         worker_monitor = udev_monitor_new_from_netlink(udev, NULL);
 186         if (worker_monitor == NULL)
 187                 return;
 188         /* allow the main daemon netlink address to send devices to the worker */
 189         udev_monitor_allow_unicast_sender(worker_monitor, monitor);
 190         udev_monitor_enable_receiving(worker_monitor);
 191
 192         worker = new0(struct worker, 1);
 193         if (worker == NULL) {
 194                 udev_monitor_unref(worker_monitor);
 195                 return;
 196         }
 197         /* worker + event reference */
 198         worker->refcount = 2;
 199         worker->udev = udev;
 200
 201         pid = fork();
 202         switch (pid) {
 203         case 0: {
 204                 struct udev_device *dev = NULL;
 205                 int fd_monitor;
 206                 struct epoll_event ep_signal, ep_monitor;
 207                 sigset_t mask;
 208                 int rc = EXIT_SUCCESS;
 209
 210                 /* take initial device from queue */
 211                 dev = event->dev;
 212                 event->dev = NULL;
 213
 214                 free(worker);
 215                 worker_list_cleanup(udev);
 216                 event_queue_cleanup(udev, EVENT_UNDEF);
 217                 udev_monitor_unref(monitor);
 218                 udev_ctrl_unref(udev_ctrl);
 219                 close(fd_signal);
 220                 close(fd_ep);
 221                 close(worker_watch[READ_END]);
 222
 223                 sigfillset(&mask);
 224                 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
 225                 if (fd_signal < 0) {
 226                         log_error("error creating signalfd %m");
 227                         rc = 2;
 228                         goto out;
 229                 }
 230
 231                 fd_ep = epoll_create1(EPOLL_CLOEXEC);
 232                 if (fd_ep < 0) {
 233                         log_error("error creating epoll fd: %m");
 234                         rc = 3;
 235                         goto out;
 236                 }
 237
 238                 memzero(&ep_signal, sizeof(struct epoll_event));
 239                 ep_signal.events = EPOLLIN;
 240                 ep_signal.data.fd = fd_signal;
 241
 242                 fd_monitor = udev_monitor_get_fd(worker_monitor);
 243                 memzero(&ep_monitor, sizeof(struct epoll_event));
 244                 ep_monitor.events = EPOLLIN;
 245                 ep_monitor.data.fd = fd_monitor;
 246
 247                 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
 248                     epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_monitor, &ep_monitor) < 0) {
 249                         log_error("fail to add fds to epoll: %m");
 250                         rc = 4;
 251                         goto out;
 252                 }
 253
 254                 /* request TERM signal if parent exits */
 255                 prctl(PR_SET_PDEATHSIG, SIGTERM);
 256
 257                 /* reset OOM score, we only protect the main daemon */
 258                 write_string_file("/proc/self/oom_score_adj", "0");
 259
 260                 for (;;) {
 261                         struct udev_event *udev_event;
 262                         struct worker_message msg;
 263                         int fd_lock = -1;
 264                         int err = 0;
 265
 266                         log_debug("seq %llu running", udev_device_get_seqnum(dev));
 267                         udev_event = udev_event_new(dev);
 268                         if (udev_event == NULL) {
 269                                 rc = 5;
 270                                 goto out;
 271                         }
 272
 273                         /* needed for SIGCHLD/SIGTERM in spawn() */
 274                         udev_event->fd_signal = fd_signal;
 275
 276                         if (exec_delay > 0)
 277                                 udev_event->exec_delay = exec_delay;
 278
 279                         /*
 280                          * Take a shared lock on the device node; this establishes
 281                          * a concept of device "ownership" to serialize device
 282                          * access. External processes holding an exclusive lock will
 283                          * cause udev to skip the event handling; in the case udev
 284                          * acquired the lock, the external process can block until
 285                          * udev has finished its event handling.
 286                          */
 287                         if (!streq_ptr(udev_device_get_action(dev), "remove") &&
 288                             streq_ptr("block", udev_device_get_subsystem(dev)) &&
 289                             !startswith(udev_device_get_sysname(dev), "dm-") &&
 290                             !startswith(udev_device_get_sysname(dev), "md")) {
 291                                 struct udev_device *d = dev;
 292
 293                                 if (streq_ptr("partition", udev_device_get_devtype(d)))
 294                                         d = udev_device_get_parent(d);
 295
 296                                 if (d) {
 297                                         fd_lock = open(udev_device_get_devnode(d), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
 298                                         if (fd_lock >= 0 && flock(fd_lock, LOCK_SH|LOCK_NB) < 0) {
 299                                                 log_debug("Unable to flock(%s), skipping event handling: %m", udev_device_get_devnode(d));
 300                                                 err = -EWOULDBLOCK;
 301                                                 fd_lock = safe_close(fd_lock);
 302                                                 goto skip;
 303                                         }
 304                                 }
 305                         }
 306
 307                         /* apply rules, create node, symlinks */
 308                         udev_event_execute_rules(udev_event, event_timeout_usec, rules, &sigmask_orig);
 309
 310                         udev_event_execute_run(udev_event, event_timeout_usec, &sigmask_orig);
 311
 312                         /* apply/restore inotify watch */
 313                         if (udev_event->inotify_watch) {
 314                                 udev_watch_begin(udev, dev);
 315                                 udev_device_update_db(dev);
 316                         }
 317
 318                         safe_close(fd_lock);
 319
 320                         /* send processed event back to libudev listeners */
 321                         udev_monitor_send_device(worker_monitor, NULL, dev);
 322
 323 skip:
 324                         /* send udevd the result of the event execution */
 325                         memzero(&msg, sizeof(struct worker_message));
 326                         msg.exitcode = err;
 327                         msg.pid = getpid();
 328                         send(worker_watch[WRITE_END], &msg, sizeof(struct worker_message), 0);
 329
 330                         log_debug("seq %llu processed with %i", udev_device_get_seqnum(dev), err);
 331
 332                         udev_device_unref(dev);
 333                         dev = NULL;
 334
 335                         if (udev_event->sigterm) {
 336                                 udev_event_unref(udev_event);
 337                                 goto out;
 338                         }
 339
 340                         udev_event_unref(udev_event);
 341
 342                         /* wait for more device messages from main udevd, or term signal */
 343                         while (dev == NULL) {
 344                                 struct epoll_event ev[4];
 345                                 int fdcount;
 346                                 int i;
 347
 348                                 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), -1);
 349                                 if (fdcount < 0) {
 350                                         if (errno == EINTR)
 351                                                 continue;
 352                                         log_error("failed to poll: %m");
 353                                         goto out;
 354                                 }
 355
 356                                 for (i = 0; i < fdcount; i++) {
 357                                         if (ev[i].data.fd == fd_monitor && ev[i].events & EPOLLIN) {
 358                                                 dev = udev_monitor_receive_device(worker_monitor);
 359                                                 break;
 360                                         } else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN) {
 361                                                 struct signalfd_siginfo fdsi;
 362                                                 ssize_t size;
 363
 364                                                 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
 365                                                 if (size != sizeof(struct signalfd_siginfo))
 366                                                         continue;
 367                                                 switch (fdsi.ssi_signo) {
 368                                                 case SIGTERM:
 369                                                         goto out;
 370                                                 }
 371                                         }
 372                                 }
 373                         }
 374                 }
 375 out:
 376                 udev_device_unref(dev);
 377                 safe_close(fd_signal);
 378                 safe_close(fd_ep);
 379                 close(fd_inotify);
 380                 close(worker_watch[WRITE_END]);
 381                 udev_rules_unref(rules);
 382                 udev_builtin_exit(udev);
 383                 udev_monitor_unref(worker_monitor);
 384                 udev_unref(udev);
 385                 log_close();
 386                 exit(rc);
 387         }
 388         case -1:
 389                 udev_monitor_unref(worker_monitor);
 390                 event->state = EVENT_QUEUED;
 391                 free(worker);
 392                 log_error("fork of child failed: %m");
 393                 break;
 394         default:
 395                 /* close monitor, but keep address around */
 396                 udev_monitor_disconnect(worker_monitor);
 397                 worker->monitor = worker_monitor;
 398                 worker->pid = pid;
 399                 worker->state = WORKER_RUNNING;
 400                 worker->event_start_usec = now(CLOCK_MONOTONIC);
 401                 worker->event = event;
 402                 event->state = EVENT_RUNNING;
 403                 udev_list_node_append(&worker->node, &worker_list);
 404                 children++;
 405                 log_debug("seq %llu forked new worker [%u]", udev_device_get_seqnum(event->dev), pid);
 406                 break;
 407         }
 408 }
 409
 410 static void event_run(struct event *event) {
 411         struct udev_list_node *loop;
 412
 413         udev_list_node_foreach(loop, &worker_list) {
 414                 struct worker *worker = node_to_worker(loop);
 415                 ssize_t count;
 416
 417                 if (worker->state != WORKER_IDLE)
 418                         continue;
 419
 420                 count = udev_monitor_send_device(monitor, worker->monitor, event->dev);
 421                 if (count < 0) {
 422                         log_error("worker [%u] did not accept message %zi (%m), kill it", worker->pid, count);
 423                         kill(worker->pid, SIGKILL);
 424                         worker->state = WORKER_KILLED;
 425                         continue;
 426                 }
 427                 worker_ref(worker);
 428                 worker->event = event;
 429                 worker->state = WORKER_RUNNING;
 430                 worker->event_start_usec = now(CLOCK_MONOTONIC);
 431                 event->state = EVENT_RUNNING;
 432                 return;
 433         }
 434
 435         if (children >= children_max) {
 436                 if (children_max > 1)
 437                         log_debug("maximum number (%i) of children reached", children);
 438                 return;
 439         }
 440
 441         /* start new worker and pass initial device */
 442         worker_new(event);
 443 }
 444
 445 static int event_queue_insert(struct udev_device *dev) {
 446         struct event *event;
 447
 448         event = new0(struct event, 1);
 449         if (event == NULL)
 450                 return -1;
 451
 452         event->udev = udev_device_get_udev(dev);
 453         event->dev = dev;
 454         event->seqnum = udev_device_get_seqnum(dev);
 455         event->devpath = udev_device_get_devpath(dev);
 456         event->devpath_len = strlen(event->devpath);
 457         event->devpath_old = udev_device_get_devpath_old(dev);
 458         event->devnum = udev_device_get_devnum(dev);
 459         event->is_block = streq("block", udev_device_get_subsystem(dev));
 460         event->ifindex = udev_device_get_ifindex(dev);
 461 #ifdef HAVE_FIRMWARE
 462         if (streq(udev_device_get_subsystem(dev), "firmware"))
 463                 event->nodelay = true;
 464 #endif
 465
 466         log_debug("seq %llu queued, '%s' '%s'", udev_device_get_seqnum(dev),
 467              udev_device_get_action(dev), udev_device_get_subsystem(dev));
 468
 469         event->state = EVENT_QUEUED;
 470         udev_list_node_append(&event->node, &event_list);
 471         return 0;
 472 }
 473
 474 static void worker_kill(struct udev *udev) {
 475         struct udev_list_node *loop;
 476
 477         udev_list_node_foreach(loop, &worker_list) {
 478                 struct worker *worker = node_to_worker(loop);
 479
 480                 if (worker->state == WORKER_KILLED)
 481                         continue;
 482
 483                 worker->state = WORKER_KILLED;
 484                 kill(worker->pid, SIGTERM);
 485         }
 486 }
 487
 488 /* lookup event for identical, parent, child device */
 489 static bool is_devpath_busy(struct event *event) {
 490         struct udev_list_node *loop;
 491         size_t common;
 492
 493         /* check if queue contains events we depend on */
 494         udev_list_node_foreach(loop, &event_list) {
 495                 struct event *loop_event = node_to_event(loop);
 496
 497                 /* we already found a later event, earlier can not block us, no need to check again */
 498                 if (loop_event->seqnum < event->delaying_seqnum)
 499                         continue;
 500
 501                 /* event we checked earlier still exists, no need to check again */
 502                 if (loop_event->seqnum == event->delaying_seqnum)
 503                         return true;
 504
 505                 /* found ourself, no later event can block us */
 506                 if (loop_event->seqnum >= event->seqnum)
 507                         break;
 508
 509                 /* check major/minor */
 510                 if (major(event->devnum) != 0 && event->devnum == loop_event->devnum && event->is_block == loop_event->is_block)
 511                         return true;
 512
 513                 /* check network device ifindex */
 514                 if (event->ifindex != 0 && event->ifindex == loop_event->ifindex)
 515                         return true;
 516
 517                 /* check our old name */
 518                 if (event->devpath_old != NULL && streq(loop_event->devpath, event->devpath_old)) {
 519                         event->delaying_seqnum = loop_event->seqnum;
 520                         return true;
 521                 }
 522
 523                 /* compare devpath */
 524                 common = MIN(loop_event->devpath_len, event->devpath_len);
 525
 526                 /* one devpath is contained in the other? */
 527                 if (memcmp(loop_event->devpath, event->devpath, common) != 0)
 528                         continue;
 529
 530                 /* identical device event found */
 531                 if (loop_event->devpath_len == event->devpath_len) {
 532                         /* devices names might have changed/swapped in the meantime */
 533                         if (major(event->devnum) != 0 && (event->devnum != loop_event->devnum || event->is_block != loop_event->is_block))
 534                                 continue;
 535                         if (event->ifindex != 0 && event->ifindex != loop_event->ifindex)
 536                                 continue;
 537                         event->delaying_seqnum = loop_event->seqnum;
 538                         return true;
 539                 }
 540
 541 #ifdef HAVE_FIRMWARE
 542                 /* allow to bypass the dependency tracking */
 543                 if (event->nodelay)
 544                         continue;
 545 #endif
 546
 547                 /* parent device event found */
 548                 if (event->devpath[common] == '/') {
 549                         event->delaying_seqnum = loop_event->seqnum;
 550                         return true;
 551                 }
 552
 553                 /* child device event found */
 554                 if (loop_event->devpath[common] == '/') {
 555                         event->delaying_seqnum = loop_event->seqnum;
 556                         return true;
 557                 }
 558
 559                 /* no matching device */
 560                 continue;
 561         }
 562
 563         return false;
 564 }
 565
 566 static void event_queue_start(struct udev *udev) {
 567         struct udev_list_node *loop;
 568
 569         udev_list_node_foreach(loop, &event_list) {
 570                 struct event *event = node_to_event(loop);
 571
 572                 if (event->state != EVENT_QUEUED)
 573                         continue;
 574
 575                 /* do not start event if parent or child event is still running */
 576                 if (is_devpath_busy(event))
 577                         continue;
 578
 579                 event_run(event);
 580         }
 581 }
 582
 583 static void event_queue_cleanup(struct udev *udev, enum event_state match_type) {
 584         struct udev_list_node *loop, *tmp;
 585
 586         udev_list_node_foreach_safe(loop, tmp, &event_list) {
 587                 struct event *event = node_to_event(loop);
 588
 589                 if (match_type != EVENT_UNDEF && match_type != event->state)
 590                         continue;
 591
 592                 event_queue_delete(event);
 593         }
 594 }
 595
 596 static void worker_returned(int fd_worker) {
 597         for (;;) {
 598                 struct worker_message msg;
 599                 ssize_t size;
 600                 struct udev_list_node *loop;
 601
 602                 size = recv(fd_worker, &msg, sizeof(struct worker_message), MSG_DONTWAIT);
 603                 if (size != sizeof(struct worker_message))
 604                         break;
 605
 606                 /* lookup worker who sent the signal */
 607                 udev_list_node_foreach(loop, &worker_list) {
 608                         struct worker *worker = node_to_worker(loop);
 609
 610                         if (worker->pid != msg.pid)
 611                                 continue;
 612
 613                         /* worker returned */
 614                         if (worker->event) {
 615                                 worker->event->exitcode = msg.exitcode;
 616                                 event_queue_delete(worker->event);
 617                                 worker->event = NULL;
 618                         }
 619                         if (worker->state != WORKER_KILLED)
 620                                 worker->state = WORKER_IDLE;
 621                         worker_unref(worker);
 622                         break;
 623                 }
 624         }
 625 }
 626
 627 /* receive the udevd message from userspace */
 628 static struct udev_ctrl_connection *handle_ctrl_msg(struct udev_ctrl *uctrl) {
 629         struct udev *udev = udev_ctrl_get_udev(uctrl);
 630         struct udev_ctrl_connection *ctrl_conn;
 631         struct udev_ctrl_msg *ctrl_msg = NULL;
 632         const char *str;
 633         int i;
 634
 635         ctrl_conn = udev_ctrl_get_connection(uctrl);
 636         if (ctrl_conn == NULL)
 637                 goto out;
 638
 639         ctrl_msg = udev_ctrl_receive_msg(ctrl_conn);
 640         if (ctrl_msg == NULL)
 641                 goto out;
 642
 643         i = udev_ctrl_get_set_log_level(ctrl_msg);
 644         if (i >= 0) {
 645                 log_debug("udevd message (SET_LOG_LEVEL) received, log_priority=%i", i);
 646                 log_set_max_level(i);
 647                 udev_set_log_priority(udev, i);
 648                 worker_kill(udev);
 649         }
 650
 651         if (udev_ctrl_get_stop_exec_queue(ctrl_msg) > 0) {
 652                 log_debug("udevd message (STOP_EXEC_QUEUE) received");
 653                 stop_exec_queue = true;
 654         }
 655
 656         if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) {
 657                 log_debug("udevd message (START_EXEC_QUEUE) received");
 658                 stop_exec_queue = false;
 659         }
 660
 661         if (udev_ctrl_get_reload(ctrl_msg) > 0) {
 662                 log_debug("udevd message (RELOAD) received");
 663                 reload = true;
 664         }
 665
 666         str = udev_ctrl_get_set_env(ctrl_msg);
 667         if (str != NULL) {
 668                 char *key;
 669
 670                 key = strdup(str);
 671                 if (key != NULL) {
 672                         char *val;
 673
 674                         val = strchr(key, '=');
 675                         if (val != NULL) {
 676                                 val[0] = '\0';
 677                                 val = &val[1];
 678                                 if (val[0] == '\0') {
 679                                         log_debug("udevd message (ENV) received, unset '%s'", key);
 680                                         udev_add_property(udev, key, NULL);
 681                                 } else {
 682                                         log_debug("udevd message (ENV) received, set '%s=%s'", key, val);
 683                                         udev_add_property(udev, key, val);
 684                                 }
 685                         } else {
 686                                 log_error("wrong key format '%s'", key);
 687                         }
 688                         free(key);
 689                 }
 690                 worker_kill(udev);
 691         }
 692
 693         i = udev_ctrl_get_set_children_max(ctrl_msg);
 694         if (i >= 0) {
 695                 log_debug("udevd message (SET_MAX_CHILDREN) received, children_max=%i", i);
 696                 children_max = i;
 697         }
 698
 699         if (udev_ctrl_get_ping(ctrl_msg) > 0)
 700                 log_debug("udevd message (SYNC) received");
 701
 702         if (udev_ctrl_get_exit(ctrl_msg) > 0) {
 703                 log_debug("udevd message (EXIT) received");
 704                 udev_exit = true;
 705                 /* keep reference to block the client until we exit */
 706                 udev_ctrl_connection_ref(ctrl_conn);
 707         }
 708 out:
 709         udev_ctrl_msg_unref(ctrl_msg);
 710         return udev_ctrl_connection_unref(ctrl_conn);
 711 }
 712
 713 static int synthesize_change(struct udev_device *dev) {
 714         char filename[UTIL_PATH_SIZE];
 715         int r;
 716
 717         if (streq_ptr("block", udev_device_get_subsystem(dev)) &&
 718             streq_ptr("disk", udev_device_get_devtype(dev)) &&
 719             !startswith(udev_device_get_sysname(dev), "dm-")) {
 720                 bool part_table_read = false;
 721                 bool has_partitions = false;
 722                 int fd;
 723                 struct udev *udev = udev_device_get_udev(dev);
 724                 _cleanup_udev_enumerate_unref_ struct udev_enumerate *e = NULL;
 725                 struct udev_list_entry *item;
 726
 727                 /*
 728                  * Try to re-read the partition table. This only succeeds if
 729                  * none of the devices is busy. The kernel returns 0 if no
 730                  * partition table is found, and we will not get an event for
 731                  * the disk.
 732                  */
 733                 fd = open(udev_device_get_devnode(dev), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
 734                 if (fd >= 0) {
 735                         r = flock(fd, LOCK_EX|LOCK_NB);
 736                         if (r >= 0)
 737                                 r = ioctl(fd, BLKRRPART, 0);
 738
 739                         close(fd);
 740                         if (r >= 0)
 741                                 part_table_read = true;
 742                 }
 743
 744                 /* search for partitions */
 745                 e = udev_enumerate_new(udev);
 746                 if (!e)
 747                         return -ENOMEM;
 748
 749                 r = udev_enumerate_add_match_parent(e, dev);
 750                 if (r < 0)
 751                         return r;
 752
 753                 r = udev_enumerate_add_match_subsystem(e, "block");
 754                 if (r < 0)
 755                         return r;
 756
 757                 r = udev_enumerate_scan_devices(e);
 758                 if (r < 0)
 759                         return r;
 760
 761                 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
 762                         _cleanup_udev_device_unref_ struct udev_device *d = NULL;
 763
 764                         d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
 765                         if (!d)
 766                                 continue;
 767
 768                         if (!streq_ptr("partition", udev_device_get_devtype(d)))
 769                                 continue;
 770
 771                         has_partitions = true;
 772                         break;
 773                 }
 774
 775                 /*
 776                  * We have partitions and re-read the table, the kernel already sent
 777                  * out a "change" event for the disk, and "remove/add" for all
 778                  * partitions.
 779                  */
 780                 if (part_table_read && has_partitions)
 781                         return 0;
 782
 783                 /*
 784                  * We have partitions but re-reading the partition table did not
 785                  * work, synthesize "change" for the disk and all partitions.
 786                  */
 787                 log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
 788                 strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
 789                 write_string_file(filename, "change");
 790
 791                 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
 792                         _cleanup_udev_device_unref_ struct udev_device *d = NULL;
 793
 794                         d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
 795                         if (!d)
 796                                 continue;
 797
 798                         if (!streq_ptr("partition", udev_device_get_devtype(d)))
 799                                 continue;
 800
 801                         log_debug("device %s closed, synthesising partition '%s' 'change'",
 802                                   udev_device_get_devnode(dev), udev_device_get_devnode(d));
 803                         strscpyl(filename, sizeof(filename), udev_device_get_syspath(d), "/uevent", NULL);
 804                         write_string_file(filename, "change");
 805                 }
 806
 807                 return 0;
 808         }
 809
 810         log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
 811         strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
 812         write_string_file(filename, "change");
 813
 814         return 0;
 815 }
 816
 817 static int handle_inotify(struct udev *udev) {
 818         int nbytes, pos;
 819         char *buf;
 820         struct inotify_event *ev;
 821         int r;
 822
 823         r = ioctl(fd_inotify, FIONREAD, &nbytes);
 824         if (r < 0 || nbytes <= 0)
 825                 return -errno;
 826
 827         buf = malloc(nbytes);
 828         if (!buf) {
 829                 log_error("error getting buffer for inotify");
 830                 return -ENOMEM;
 831         }
 832
 833         nbytes = read(fd_inotify, buf, nbytes);
 834
 835         for (pos = 0; pos < nbytes; pos += sizeof(struct inotify_event) + ev->len) {
 836                 struct udev_device *dev;
 837
 838                 ev = (struct inotify_event *)(buf + pos);
 839                 dev = udev_watch_lookup(udev, ev->wd);
 840                 if (!dev)
 841                         continue;
 842
 843                 log_debug("inotify event: %x for %s", ev->mask, udev_device_get_devnode(dev));
 844                 if (ev->mask & IN_CLOSE_WRITE)
 845                         synthesize_change(dev);
 846                 else if (ev->mask & IN_IGNORED)
 847                         udev_watch_end(udev, dev);
 848
 849                 udev_device_unref(dev);
 850         }
 851
 852         free(buf);
 853         return 0;
 854 }
 855
 856 static void handle_signal(struct udev *udev, int signo) {
 857         switch (signo) {
 858         case SIGINT:
 859         case SIGTERM:
 860                 udev_exit = true;
 861                 break;
 862         case SIGCHLD:
 863                 for (;;) {
 864                         pid_t pid;
 865                         int status;
 866                         struct udev_list_node *loop, *tmp;
 867
 868                         pid = waitpid(-1, &status, WNOHANG);
 869                         if (pid <= 0)
 870                                 break;
 871
 872                         udev_list_node_foreach_safe(loop, tmp, &worker_list) {
 873                                 struct worker *worker = node_to_worker(loop);
 874
 875                                 if (worker->pid != pid)
 876                                         continue;
 877                                 log_debug("worker [%u] exit", pid);
 878
 879                                 if (WIFEXITED(status)) {
 880                                         if (WEXITSTATUS(status) != 0)
 881                                                 log_error("worker [%u] exit with return code %i",
 882                                                           pid, WEXITSTATUS(status));
 883                                 } else if (WIFSIGNALED(status)) {
 884                                         log_error("worker [%u] terminated by signal %i (%s)",
 885                                                   pid, WTERMSIG(status), strsignal(WTERMSIG(status)));
 886                                 } else if (WIFSTOPPED(status)) {
 887                                         log_error("worker [%u] stopped", pid);
 888                                 } else if (WIFCONTINUED(status)) {
 889                                         log_error("worker [%u] continued", pid);
 890                                 } else {
 891                                         log_error("worker [%u] exit with status 0x%04x", pid, status);
 892                                 }
 893
 894                                 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 895                                         if (worker->event) {
 896                                                 log_error("worker [%u] failed while handling '%s'",
 897                                                           pid, worker->event->devpath);
 898                                                 worker->event->exitcode = -32;
 899                                                 event_queue_delete(worker->event);
 900
 901                                                 /* drop reference taken for state 'running' */
 902                                                 worker_unref(worker);
 903                                         }
 904                                 }
 905                                 worker_unref(worker);
 906                                 break;
 907                         }
 908                 }
 909                 break;
 910         case SIGHUP:
 911                 reload = true;
 912                 break;
 913         }
 914 }
 915
 916 static int systemd_fds(struct udev *udev, int *rctrl, int *rnetlink) {
 917         int ctrl = -1, netlink = -1;
 918         int fd, n;
 919
 920         n = sd_listen_fds(true);
 921         if (n <= 0)
 922                 return -1;
 923
 924         for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
 925                 if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1)) {
 926                         if (ctrl >= 0)
 927                                 return -1;
 928                         ctrl = fd;
 929                         continue;
 930                 }
 931
 932                 if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1)) {
 933                         if (netlink >= 0)
 934                                 return -1;
 935                         netlink = fd;
 936                         continue;
 937                 }
 938
 939                 return -1;
 940         }
 941
 942         if (ctrl < 0 || netlink < 0)
 943                 return -1;
 944
 945         log_debug("ctrl=%i netlink=%i", ctrl, netlink);
 946         *rctrl = ctrl;
 947         *rnetlink = netlink;
 948         return 0;
 949 }
 950
 951 /*
 952  * read the kernel commandline, in case we need to get into debug mode
 953  *   udev.log-priority=<level>              syslog priority
 954  *   udev.children-max=<number of workers>  events are fully serialized if set to 1
 955  *   udev.exec-delay=<number of seconds>    delay execution of every executed program
 956  */
 957 static void kernel_cmdline_options(struct udev *udev) {
 958         _cleanup_free_ char *line = NULL;
 959         char *w, *state;
 960         size_t l;
 961         int r;
 962
 963         r = proc_cmdline(&line);
 964         if (r < 0)
 965                 log_warning("Failed to read /proc/cmdline, ignoring: %s", strerror(-r));
 966         if (r <= 0)
 967                 return;
 968
 969         FOREACH_WORD_QUOTED(w, l, line, state) {
 970                 char *s, *opt;
 971
 972                 s = strndup(w, l);
 973                 if (!s)
 974                         break;
 975
 976                 /* accept the same options for the initrd, prefixed with "rd." */
 977                 if (in_initrd() && startswith(s, "rd."))
 978                         opt = s + 3;
 979                 else
 980                         opt = s;
 981
 982                 if (startswith(opt, "udev.log-priority=")) {
 983                         int prio;
 984
 985                         prio = util_log_priority(opt + 18);
 986                         log_set_max_level(prio);
 987                         udev_set_log_priority(udev, prio);
 988                 } else if (startswith(opt, "udev.children-max=")) {
 989                         children_max = strtoul(opt + 18, NULL, 0);
 990                 } else if (startswith(opt, "udev.exec-delay=")) {
 991                         exec_delay = strtoul(opt + 16, NULL, 0);
 992                 } else if (startswith(opt, "udev.event-timeout=")) {
 993                         event_timeout_usec = strtoul(opt + 16, NULL, 0) * USEC_PER_SEC;
 994                 }
 995
 996                 free(s);
 997         }
 998 }
 999
1000 int main(int argc, char *argv[]) {
1001         struct udev *udev;
1002         sigset_t mask;
1003         int daemonize = false;
1004         int resolve_names = 1;
1005         static const struct option options[] = {
1006                 { "daemon", no_argument, NULL, 'd' },
1007                 { "debug", no_argument, NULL, 'D' },
1008                 { "children-max", required_argument, NULL, 'c' },
1009                 { "exec-delay", required_argument, NULL, 'e' },
1010                 { "event-timeout", required_argument, NULL, 't' },
1011                 { "resolve-names", required_argument, NULL, 'N' },
1012                 { "help", no_argument, NULL, 'h' },
1013                 { "version", no_argument, NULL, 'V' },
1014                 {}
1015         };
1016         int fd_ctrl = -1;
1017         int fd_netlink = -1;
1018         int fd_worker = -1;
1019         struct epoll_event ep_ctrl, ep_inotify, ep_signal, ep_netlink, ep_worker;
1020         struct udev_ctrl_connection *ctrl_conn = NULL;
1021         int rc = 1;
1022
1023         udev = udev_new();
1024         if (udev == NULL)
1025                 goto exit;
1026
1027         log_set_target(LOG_TARGET_AUTO);
1028         log_parse_environment();
1029         log_open();
1030
1031         udev_set_log_fn(udev, udev_main_log);
1032         log_set_max_level(udev_get_log_priority(udev));
1033
1034         log_debug("version %s", VERSION);
1035         label_init("/dev");
1036
1037         for (;;) {
1038                 int option;
1039
1040                 option = getopt_long(argc, argv, "c:de:DtN:hV", options, NULL);
1041                 if (option == -1)
1042                         break;
1043
1044                 switch (option) {
1045                 case 'd':
1046                         daemonize = true;
1047                         break;
1048                 case 'c':
1049                         children_max = strtoul(optarg, NULL, 0);
1050                         break;
1051                 case 'e':
1052                         exec_delay = strtoul(optarg, NULL, 0);
1053                         break;
1054                 case 't':
1055                         event_timeout_usec = strtoul(optarg, NULL, 0) * USEC_PER_SEC;
1056                         break;
1057                 case 'D':
1058                         debug = true;
1059                         log_set_max_level(LOG_DEBUG);
1060                         udev_set_log_priority(udev, LOG_DEBUG);
1061                         break;
1062                 case 'N':
1063                         if (streq(optarg, "early")) {
1064                                 resolve_names = 1;
1065                         } else if (streq(optarg, "late")) {
1066                                 resolve_names = 0;
1067                         } else if (streq(optarg, "never")) {
1068                                 resolve_names = -1;
1069                         } else {
1070                                 fprintf(stderr, "resolve-names must be early, late or never\n");
1071                                 log_error("resolve-names must be early, late or never");
1072                                 goto exit;
1073                         }
1074                         break;
1075                 case 'h':
1076                         printf("Usage: udevd OPTIONS\n"
1077                                "  --daemon\n"
1078                                "  --debug\n"
1079                                "  --children-max=<maximum number of workers>\n"
1080                                "  --exec-delay=<seconds to wait before executing RUN=>\n"
1081                                "  --event-timeout=<seconds to wait before terminating an event>\n"
1082                                "  --resolve-names=early|late|never\n"
1083                                "  --version\n"
1084                                "  --help\n"
1085                                "\n");
1086                         goto exit;
1087                 case 'V':
1088                         printf("%s\n", VERSION);
1089                         goto exit;
1090                 default:
1091                         goto exit;
1092                 }
1093         }
1094
1095         kernel_cmdline_options(udev);
1096
1097         if (getuid() != 0) {
1098                 fprintf(stderr, "root privileges required\n");
1099                 log_error("root privileges required");
1100                 goto exit;
1101         }
1102
1103         /* set umask before creating any file/directory */
1104         chdir("/");
1105         umask(022);
1106
1107         mkdir("/run/udev", 0755);
1108
1109         dev_setup(NULL);
1110
1111         /* before opening new files, make sure std{in,out,err} fds are in a sane state */
1112         if (daemonize) {
1113                 int fd;
1114
1115                 fd = open("/dev/null", O_RDWR);
1116                 if (fd >= 0) {
1117                         if (write(STDOUT_FILENO, 0, 0) < 0)
1118                                 dup2(fd, STDOUT_FILENO);
1119                         if (write(STDERR_FILENO, 0, 0) < 0)
1120                                 dup2(fd, STDERR_FILENO);
1121                         if (fd > STDERR_FILENO)
1122                                 close(fd);
1123                 } else {
1124                         fprintf(stderr, "cannot open /dev/null\n");
1125                         log_error("cannot open /dev/null");
1126                 }
1127         }
1128
1129         if (systemd_fds(udev, &fd_ctrl, &fd_netlink) >= 0) {
1130                 /* get control and netlink socket from systemd */
1131                 udev_ctrl = udev_ctrl_new_from_fd(udev, fd_ctrl);
1132                 if (udev_ctrl == NULL) {
1133                         log_error("error taking over udev control socket");
1134                         rc = 1;
1135                         goto exit;
1136                 }
1137
1138                 monitor = udev_monitor_new_from_netlink_fd(udev, "kernel", fd_netlink);
1139                 if (monitor == NULL) {
1140                         log_error("error taking over netlink socket");
1141                         rc = 3;
1142                         goto exit;
1143                 }
1144
1145                 /* get our own cgroup, we regularly kill everything udev has left behind */
1146                 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &udev_cgroup) < 0)
1147                         udev_cgroup = NULL;
1148         } else {
1149                 /* open control and netlink socket */
1150                 udev_ctrl = udev_ctrl_new(udev);
1151                 if (udev_ctrl == NULL) {
1152                         fprintf(stderr, "error initializing udev control socket");
1153                         log_error("error initializing udev control socket");
1154                         rc = 1;
1155                         goto exit;
1156                 }
1157                 fd_ctrl = udev_ctrl_get_fd(udev_ctrl);
1158
1159                 monitor = udev_monitor_new_from_netlink(udev, "kernel");
1160                 if (monitor == NULL) {
1161                         fprintf(stderr, "error initializing netlink socket\n");
1162                         log_error("error initializing netlink socket");
1163                         rc = 3;
1164                         goto exit;
1165                 }
1166                 fd_netlink = udev_monitor_get_fd(monitor);
1167         }
1168
1169         if (udev_monitor_enable_receiving(monitor) < 0) {
1170                 fprintf(stderr, "error binding netlink socket\n");
1171                 log_error("error binding netlink socket");
1172                 rc = 3;
1173                 goto exit;
1174         }
1175
1176         if (udev_ctrl_enable_receiving(udev_ctrl) < 0) {
1177                 fprintf(stderr, "error binding udev control socket\n");
1178                 log_error("error binding udev control socket");
1179                 rc = 1;
1180                 goto exit;
1181         }
1182
1183         udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024);
1184
1185         if (daemonize) {
1186                 pid_t pid;
1187
1188                 pid = fork();
1189                 switch (pid) {
1190                 case 0:
1191                         break;
1192                 case -1:
1193                         log_error("fork of daemon failed: %m");
1194                         rc = 4;
1195                         goto exit;
1196                 default:
1197                         rc = EXIT_SUCCESS;
1198                         goto exit_daemonize;
1199                 }
1200
1201                 setsid();
1202
1203                 write_string_file("/proc/self/oom_score_adj", "-1000");
1204         } else {
1205                 sd_notify(1, "READY=1");
1206         }
1207
1208         print_kmsg("starting version " VERSION "\n");
1209
1210         if (!debug) {
1211                 int fd;
1212
1213                 fd = open("/dev/null", O_RDWR);
1214                 if (fd >= 0) {
1215                         dup2(fd, STDIN_FILENO);
1216                         dup2(fd, STDOUT_FILENO);
1217                         dup2(fd, STDERR_FILENO);
1218                         close(fd);
1219                 }
1220         }
1221
1222         fd_inotify = udev_watch_init(udev);
1223         if (fd_inotify < 0) {
1224                 fprintf(stderr, "error initializing inotify\n");
1225                 log_error("error initializing inotify");
1226                 rc = 4;
1227                 goto exit;
1228         }
1229         udev_watch_restore(udev);
1230
1231         /* block and listen to all signals on signalfd */
1232         sigfillset(&mask);
1233         sigprocmask(SIG_SETMASK, &mask, &sigmask_orig);
1234         fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1235         if (fd_signal < 0) {
1236                 fprintf(stderr, "error creating signalfd\n");
1237                 log_error("error creating signalfd");
1238                 rc = 5;
1239                 goto exit;
1240         }
1241
1242         /* unnamed socket from workers to the main daemon */
1243         if (socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, worker_watch) < 0) {
1244                 fprintf(stderr, "error creating socketpair\n");
1245                 log_error("error creating socketpair");
1246                 rc = 6;
1247                 goto exit;
1248         }
1249         fd_worker = worker_watch[READ_END];
1250
1251         udev_builtin_init(udev);
1252
1253         rules = udev_rules_new(udev, resolve_names);
1254         if (rules == NULL) {
1255                 log_error("error reading rules");
1256                 goto exit;
1257         }
1258
1259         memzero(&ep_ctrl, sizeof(struct epoll_event));
1260         ep_ctrl.events = EPOLLIN;
1261         ep_ctrl.data.fd = fd_ctrl;
1262
1263         memzero(&ep_inotify, sizeof(struct epoll_event));
1264         ep_inotify.events = EPOLLIN;
1265         ep_inotify.data.fd = fd_inotify;
1266
1267         memzero(&ep_signal, sizeof(struct epoll_event));
1268         ep_signal.events = EPOLLIN;
1269         ep_signal.data.fd = fd_signal;
1270
1271         memzero(&ep_netlink, sizeof(struct epoll_event));
1272         ep_netlink.events = EPOLLIN;
1273         ep_netlink.data.fd = fd_netlink;
1274
1275         memzero(&ep_worker, sizeof(struct epoll_event));
1276         ep_worker.events = EPOLLIN;
1277         ep_worker.data.fd = fd_worker;
1278
1279         fd_ep = epoll_create1(EPOLL_CLOEXEC);
1280         if (fd_ep < 0) {
1281                 log_error("error creating epoll fd: %m");
1282                 goto exit;
1283         }
1284         if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_ctrl, &ep_ctrl) < 0 ||
1285             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_inotify, &ep_inotify) < 0 ||
1286             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
1287             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_netlink, &ep_netlink) < 0 ||
1288             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_worker, &ep_worker) < 0) {
1289                 log_error("fail to add fds to epoll: %m");
1290                 goto exit;
1291         }
1292
1293         if (children_max <= 0) {
1294                 cpu_set_t cpu_set;
1295
1296                 children_max = 8;
1297
1298                 if (sched_getaffinity(0, sizeof (cpu_set), &cpu_set) == 0) {
1299                         children_max +=  CPU_COUNT(&cpu_set) * 2;
1300                 }
1301         }
1302         log_debug("set children_max to %u", children_max);
1303
1304         rc = udev_rules_apply_static_dev_perms(rules);
1305         if (rc < 0)
1306                 log_error("failed to apply permissions on static device nodes - %s", strerror(-rc));
1307
1308         udev_list_node_init(&event_list);
1309         udev_list_node_init(&worker_list);
1310
1311         for (;;) {
1312                 static usec_t last_usec;
1313                 struct epoll_event ev[8];
1314                 int fdcount;
1315                 int timeout;
1316                 bool is_worker, is_signal, is_inotify, is_netlink, is_ctrl;
1317                 int i;
1318
1319                 if (udev_exit) {
1320                         /* close sources of new events and discard buffered events */
1321                         if (fd_ctrl >= 0) {
1322                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_ctrl, NULL);
1323                                 fd_ctrl = -1;
1324                         }
1325                         if (monitor != NULL) {
1326                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_netlink, NULL);
1327                                 udev_monitor_unref(monitor);
1328                                 monitor = NULL;
1329                         }
1330                         if (fd_inotify >= 0) {
1331                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_inotify, NULL);
1332                                 close(fd_inotify);
1333                                 fd_inotify = -1;
1334                         }
1335
1336                         /* discard queued events and kill workers */
1337                         event_queue_cleanup(udev, EVENT_QUEUED);
1338                         worker_kill(udev);
1339
1340                         /* exit after all has cleaned up */
1341                         if (udev_list_node_is_empty(&event_list) && children == 0)
1342                                 break;
1343
1344                         /* timeout at exit for workers to finish */
1345                         timeout = 30 * MSEC_PER_SEC;
1346                 } else if (udev_list_node_is_empty(&event_list) && children == 0) {
1347                         /* we are idle */
1348                         timeout = -1;
1349
1350                         /* cleanup possible left-over processes in our cgroup */
1351                         if (udev_cgroup)
1352                                 cg_kill(SYSTEMD_CGROUP_CONTROLLER, udev_cgroup, SIGKILL, false, true, NULL);
1353                 } else {
1354                         /* kill idle or hanging workers */
1355                         timeout = 3 * MSEC_PER_SEC;
1356                 }
1357
1358                 /* tell settle that we are busy or idle */
1359                 if (!udev_list_node_is_empty(&event_list)) {
1360                         int fd;
1361
1362                         fd = open("/run/udev/queue", O_WRONLY|O_CREAT|O_CLOEXEC|O_TRUNC|O_NOFOLLOW, 0444);
1363                         if (fd >= 0)
1364                                 close(fd);
1365                 } else {
1366                         unlink("/run/udev/queue");
1367                 }
1368
1369                 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), timeout);
1370                 if (fdcount < 0)
1371                         continue;
1372
1373                 if (fdcount == 0) {
1374                         struct udev_list_node *loop;
1375
1376                         /* timeout */
1377                         if (udev_exit) {
1378                                 log_error("timeout, giving up waiting for workers to finish");
1379                                 break;
1380                         }
1381
1382                         /* kill idle workers */
1383                         if (udev_list_node_is_empty(&event_list)) {
1384                                 log_debug("cleanup idle workers");
1385                                 worker_kill(udev);
1386                         }
1387
1388                         /* check for hanging events */
1389                         udev_list_node_foreach(loop, &worker_list) {
1390                                 struct worker *worker = node_to_worker(loop);
1391
1392                                 if (worker->state != WORKER_RUNNING)
1393                                         continue;
1394
1395                                 if ((now(CLOCK_MONOTONIC) - worker->event_start_usec) > event_timeout_usec) {
1396                                         log_error("worker [%u] %s timeout; kill it", worker->pid, worker->event->devpath);
1397                                         kill(worker->pid, SIGKILL);
1398                                         worker->state = WORKER_KILLED;
1399
1400                                         /* drop reference taken for state 'running' */
1401                                         worker_unref(worker);
1402                                         log_error("seq %llu '%s' killed", udev_device_get_seqnum(worker->event->dev), worker->event->devpath);
1403                                         worker->event->exitcode = -64;
1404                                         event_queue_delete(worker->event);
1405                                         worker->event = NULL;
1406                                 }
1407                         }
1408
1409                 }
1410
1411                 is_worker = is_signal = is_inotify = is_netlink = is_ctrl = false;
1412                 for (i = 0; i < fdcount; i++) {
1413                         if (ev[i].data.fd == fd_worker && ev[i].events & EPOLLIN)
1414                                 is_worker = true;
1415                         else if (ev[i].data.fd == fd_netlink && ev[i].events & EPOLLIN)
1416                                 is_netlink = true;
1417                         else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN)
1418                                 is_signal = true;
1419                         else if (ev[i].data.fd == fd_inotify && ev[i].events & EPOLLIN)
1420                                 is_inotify = true;
1421                         else if (ev[i].data.fd == fd_ctrl && ev[i].events & EPOLLIN)
1422                                 is_ctrl = true;
1423                 }
1424
1425                 /* check for changed config, every 3 seconds at most */
1426                 if ((now(CLOCK_MONOTONIC) - last_usec) > 3 * USEC_PER_SEC) {
1427                         if (udev_rules_check_timestamp(rules))
1428                                 reload = true;
1429                         if (udev_builtin_validate(udev))
1430                                 reload = true;
1431
1432                         last_usec = now(CLOCK_MONOTONIC);
1433                 }
1434
1435                 /* reload requested, HUP signal received, rules changed, builtin changed */
1436                 if (reload) {
1437                         worker_kill(udev);
1438                         rules = udev_rules_unref(rules);
1439                         udev_builtin_exit(udev);
1440                         reload = false;
1441                 }
1442
1443                 /* event has finished */
1444                 if (is_worker)
1445                         worker_returned(fd_worker);
1446
1447                 if (is_netlink) {
1448                         struct udev_device *dev;
1449
1450                         dev = udev_monitor_receive_device(monitor);
1451                         if (dev != NULL) {
1452                                 udev_device_set_usec_initialized(dev, now(CLOCK_MONOTONIC));
1453                                 if (event_queue_insert(dev) < 0)
1454                                         udev_device_unref(dev);
1455                         }
1456                 }
1457
1458                 /* start new events */
1459                 if (!udev_list_node_is_empty(&event_list) && !udev_exit && !stop_exec_queue) {
1460                         udev_builtin_init(udev);
1461                         if (rules == NULL)
1462                                 rules = udev_rules_new(udev, resolve_names);
1463                         if (rules != NULL)
1464                                 event_queue_start(udev);
1465                 }
1466
1467                 if (is_signal) {
1468                         struct signalfd_siginfo fdsi;
1469                         ssize_t size;
1470
1471                         size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
1472                         if (size == sizeof(struct signalfd_siginfo))
1473                                 handle_signal(udev, fdsi.ssi_signo);
1474                 }
1475
1476                 /* we are shutting down, the events below are not handled anymore */
1477                 if (udev_exit)
1478                         continue;
1479
1480                 /* device node watch */
1481                 if (is_inotify)
1482                         handle_inotify(udev);
1483
1484                 /*
1485                  * This needs to be after the inotify handling, to make sure,
1486                  * that the ping is send back after the possibly generated
1487                  * "change" events by the inotify device node watch.
1488                  *
1489                  * A single time we may receive a client connection which we need to
1490                  * keep open to block the client. It will be closed right before we
1491                  * exit.
1492                  */
1493                 if (is_ctrl)
1494                         ctrl_conn = handle_ctrl_msg(udev_ctrl);
1495         }
1496
1497         rc = EXIT_SUCCESS;
1498 exit:
1499         udev_ctrl_cleanup(udev_ctrl);
1500         unlink("/run/udev/queue");
1501 exit_daemonize:
1502         if (fd_ep >= 0)
1503                 close(fd_ep);
1504         worker_list_cleanup(udev);
1505         event_queue_cleanup(udev, EVENT_UNDEF);
1506         udev_rules_unref(rules);
1507         udev_builtin_exit(udev);
1508         if (fd_signal >= 0)
1509                 close(fd_signal);
1510         if (worker_watch[READ_END] >= 0)
1511                 close(worker_watch[READ_END]);
1512         if (worker_watch[WRITE_END] >= 0)
1513                 close(worker_watch[WRITE_END]);
1514         udev_monitor_unref(monitor);
1515         udev_ctrl_connection_unref(ctrl_conn);
1516         udev_ctrl_unref(udev_ctrl);
1517         label_finish();
1518         udev_unref(udev);
1519         log_close();
1520         return rc;
1521 }